diff options
Diffstat (limited to '')
58 files changed, 24660 insertions, 0 deletions
diff --git a/collectors/ebpf.plugin/Makefile.am b/collectors/ebpf.plugin/Makefile.am new file mode 100644 index 0000000..2d5f92a --- /dev/null +++ b/collectors/ebpf.plugin/Makefile.am @@ -0,0 +1,42 @@ +# SPDX-License-Identifier: GPL-3.0-or-later + +AUTOMAKE_OPTIONS = subdir-objects +MAINTAINERCLEANFILES = $(srcdir)/Makefile.in + +include $(top_srcdir)/build/subst.inc +SUFFIXES = .in + +userebpfconfigdir=$(configdir)/ebpf.d + +# Explicitly install directories to avoid permission issues due to umask +install-exec-local: + $(INSTALL) -d $(DESTDIR)$(userebpfconfigdir) + +dist_noinst_DATA = \ + README.md \ + $(NULL) + +ebpfconfigdir=$(libconfigdir)/ebpf.d +dist_libconfig_DATA = \ + ebpf.d.conf \ + $(NULL) + +dist_ebpfconfig_DATA = \ + ebpf.d/ebpf_kernel_reject_list.txt \ + ebpf.d/cachestat.conf \ + ebpf.d/dcstat.conf \ + ebpf.d/disk.conf \ + ebpf.d/fd.conf \ + ebpf.d/filesystem.conf \ + ebpf.d/hardirq.conf \ + ebpf.d/mdflush.conf \ + ebpf.d/mount.conf \ + ebpf.d/network.conf \ + ebpf.d/oomkill.conf \ + ebpf.d/process.conf \ + ebpf.d/shm.conf \ + ebpf.d/softirq.conf \ + ebpf.d/sync.conf \ + ebpf.d/swap.conf \ + ebpf.d/vfs.conf \ + $(NULL) diff --git a/collectors/ebpf.plugin/README.md b/collectors/ebpf.plugin/README.md new file mode 100644 index 0000000..7762ed3 --- /dev/null +++ b/collectors/ebpf.plugin/README.md @@ -0,0 +1,970 @@ +<!-- +title: "eBPF monitoring with Netdata" +description: "Use Netdata's extended Berkeley Packet Filter (eBPF) collector to monitor kernel-level metrics about your +complex applications with per-second granularity." +custom_edit_url: https://github.com/netdata/netdata/edit/master/collectors/ebpf.plugin/README.md +sidebar_label: "eBPF" +--> + +# eBPF monitoring with Netdata + +The Netdata Agent provides many [eBPF](https://ebpf.io/what-is-ebpf/) programs to help you troubleshoot and debug how applications interact with the Linux kernel. The `ebpf.plugin` uses [tracepoints, trampoline, and2 kprobes](#how-netdata-collects-data-using-probes-and-tracepoints) to collect a wide array of high value data about the host that would otherwise be impossible to capture. + +> ❗ eBPF monitoring only works on Linux systems and with specific Linux kernels, including all kernels newer than `4.11.0`, and all kernels on CentOS 7.6 or later. For kernels older than `4.11.0`, improved support is in active development. + +This document provides comprehensive details about the `ebpf.plugin`. +For hands-on configuration and troubleshooting tips see our [tutorial on troubleshooting apps with eBPF metrics](/docs/guides/troubleshoot/monitor-debug-applications-ebpf.md). + +<figure> + <img src="https://user-images.githubusercontent.com/1153921/74746434-ad6a1e00-5222-11ea-858a-a7882617ae02.png" alt="An example of VFS charts, made possible by the eBPF collector plugin" /> + <figcaption>An example of virtual file system (VFS) charts made possible by the eBPF collector plugin.</figcaption> +</figure> + +## How Netdata collects data using probes and tracepoints + +Netdata uses the following features from the Linux kernel to run eBPF programs: + +- Tracepoints are hooks to call specific functions. Tracepoints are more stable than `kprobes` and are preferred when + both options are available. +- Trampolines are bridges between kernel functions, and BPF programs. Netdata uses them by default whenever available. +- Kprobes and return probes (`kretprobe`): Probes can insert virtually into any kernel instruction. When eBPF runs in `entry` mode, it attaches only `kprobes` for internal functions monitoring calls and some arguments every time a function is called. The user can also change configuration to use [`return`](#global-configuration-options) mode, and this will allow users to monitor return from these functions and detect possible failures. + +In each case, wherever a normal kprobe, kretprobe, or tracepoint would have run its hook function, an eBPF program is run instead, performing various collection logic before letting the kernel continue its normal control flow. + +There are more methods to trigger eBPF programs, such as uprobes, but currently are not supported. + +## Configuring ebpf.plugin + +The eBPF collector is installed and enabled by default on most new installations of the Agent. +If your Agent is v1.22 or older, you may to enable the collector yourself. + +### Enable the eBPF collector + +To enable or disable the entire eBPF collector: + +1. Navigate to the [Netdata config directory](/docs/configure/nodes.md#the-netdata-config-directory). + ```bash + cd /etc/netdata + ``` + +2. Use the [`edit-config`](/docs/configure/nodes.md#use-edit-config-to-edit-configuration-files) script to edit `netdata.conf`. + + ```bash + ./edit-config netdata.conf + ``` + +3. Enable the collector by scrolling down to the `[plugins]` section. Uncomment the line `ebpf` (not + `ebpf_process`) and set it to `yes`. + + ```conf + [plugins] + ebpf = yes + ``` + +### Configure the eBPF collector + +You can configure the eBPF collector's behavior to fine-tune which metrics you receive and [optimize performance]\(#performance opimization). + +To edit the `ebpf.d.conf`: + +1. Navigate to the [Netdata config directory](/docs/configure/nodes.md#the-netdata-config-directory). + ```bash + cd /etc/netdata + ``` +2. Use the [`edit-config`](/docs/configure/nodes.md#use-edit-config-to-edit-configuration-files) script to edit [`ebpf.d.conf`](https://github.com/netdata/netdata/blob/master/collectors/ebpf.plugin/ebpf.d.conf). + + ```bash + ./edit-config ebpf.d.conf + ``` + + You can now edit the behavior of the eBPF collector. The following sections describe each configuration option in detail. + +### `[global]` configuration options + +The `[global]` section defines settings for the whole eBPF collector. + +#### eBPF load mode + +The collector uses two different eBPF programs. These programs rely on the same functions inside the kernel, but they +monitor, process, and display different kinds of information. + +By default, this plugin uses the `entry` mode. Changing this mode can create significant overhead on your operating +system, but also offer valuable information if you are developing or debugging software. The `ebpf load mode` option +accepts the following values: + +- `entry`: This is the default mode. In this mode, the eBPF collector only monitors calls for the functions described in + the sections above, and does not show charts related to errors. +- `return`: In the `return` mode, the eBPF collector monitors the same kernel functions as `entry`, but also creates new + charts for the return of these functions, such as errors. Monitoring function returns can help in debugging software, + such as failing to close file descriptors or creating zombie processes. +- `update every`: Number of seconds used for eBPF to send data for Netdata. +- `pid table size`: Defines the maximum number of PIDs stored inside the application hash table. + +#### Integration with `apps.plugin` + +The eBPF collector also creates charts for each running application through an integration with the +[`apps.plugin`](/collectors/apps.plugin/README.md). This integration helps you understand how specific applications +interact with the Linux kernel. + +If you want to _disable_ the integration with `apps.plugin` along with the above charts, change the setting `apps` to +`no`. + +```conf +[global] + apps = yes +``` + +When the integration is enabled, eBPF collector allocates memory for each process running. The total allocated memory +has direct relationship with the kernel version. When the eBPF plugin is running on kernels newer than `4.15`, it uses +per-cpu maps to speed up the update of hash tables. This also implies storing data for the same PID for each processor +it runs. + +#### Integration with `cgroups.plugin` + +The eBPF collector also creates charts for each cgroup through an integration with the +[`cgroups.plugin`](/collectors/cgroups.plugin/README.md). This integration helps you understand how a specific cgroup +interacts with the Linux kernel. + +The integration with `cgroups.plugin` is disabled by default to avoid creating overhead on your system. If you want to +_enable_ the integration with `cgroups.plugin`, change the `cgroups` setting to `yes`. + +```conf +[global] + cgroups = yes +``` + +If you do not need to monitor specific metrics for your `cgroups`, you can enable `cgroups` inside +`ebpf.d.conf`, and then disable the plugin for a specific `thread` by following the steps in the +[Configuration](#configuring-ebpfplugin) section. + +#### Collect PID + +When one of the previous integrations is enabled, `ebpf.plugin` will use Process Identifier (`PID`) to identify the +process group for which it needs to plot data. + +There are different ways to collect PID, and you can select the way `ebpf.plugin` collects data with the following +values: + +- `real parent`: This is the default mode. Collection will aggregate data for the real parent, the thread that creates + child threads. +- `parent`: Parent and real parent are the same when a process starts, but this value can be changed during run time. +- `all`: This option will store all PIDs that run on the host. Note, this method can be expensive for the host, + because more memory needs to be allocated and parsed. + +The threads that have integration with other collectors have an internal clean up wherein they attach either a +`trampoline` or a `kprobe` to `release_task` internal function. To avoid `overload` on this function, `ebpf.plugin` +will only enable these threads integrated with other collectors when the kernel is compiled with +`CONFIG_DEBUG_INFO_BTF`, unless you enable them manually. + +#### Integration Dashboard Elements + +When an integration is enabled, your dashboard will also show the following cgroups and apps charts using low-level +Linux metrics: + +> Note: The parenthetical accompanying each bulleted item provides the chart name. + +- mem + - Number of processes killed due out of memory. (`oomkills`) +- process + - Number of processes created with `do_fork`. (`process_create`) + - Number of threads created with `do_fork` or `clone (2)`, depending on your system's kernel + version. (`thread_create`) + - Number of times that a process called `do_exit`. (`task_exit`) + - Number of times that a process called `release_task`. (`task_close`) + - Number of times that an error happened to create thread or process. (`task_error`) +- swap + - Number of calls to `swap_readpage`. (`swap_read_call`) + - Number of calls to `swap_writepage`. (`swap_write_call`) +- network + - Number of outbound connections using TCP/IPv4. (`outbound_conn_ipv4`) + - Number of outbound connections using TCP/IPv6. (`outbound_conn_ipv6`) + - Number of bytes sent. (`total_bandwidth_sent`) + - Number of bytes received. (`total_bandwidth_recv`) + - Number of calls to `tcp_sendmsg`. (`bandwidth_tcp_send`) + - Number of calls to `tcp_cleanup_rbuf`. (`bandwidth_tcp_recv`) + - Number of calls to `tcp_retransmit_skb`. (`bandwidth_tcp_retransmit`) + - Number of calls to `udp_sendmsg`. (`bandwidth_udp_send`) + - Number of calls to `udp_recvmsg`. (`bandwidth_udp_recv`) +- file access + - Number of calls to open files. (`file_open`) + - Number of calls to open files that returned errors. (`open_error`) + - Number of files closed. (`file_closed`) + - Number of calls to close files that returned errors. (`file_error_closed`) +- vfs + - Number of calls to `vfs_unlink`. (`file_deleted`) + - Number of calls to `vfs_write`. (`vfs_write_call`) + - Number of calls to write a file that returned errors. (`vfs_write_error`) + - Number of calls to `vfs_read`. (`vfs_read_call`) + - - Number of calls to read a file that returned errors. (`vfs_read_error`) + - Number of bytes written with `vfs_write`. (`vfs_write_bytes`) + - Number of bytes read with `vfs_read`. (`vfs_read_bytes`) + - Number of calls to `vfs_fsync`. (`vfs_fsync`) + - Number of calls to sync file that returned errors. (`vfs_fsync_error`) + - Number of calls to `vfs_open`. (`vfs_open`) + - Number of calls to open file that returned errors. (`vfs_open_error`) + - Number of calls to `vfs_create`. (`vfs_create`) + - Number of calls to open file that returned errors. (`vfs_create_error`) +- page cache + - Ratio of pages accessed. (`cachestat_ratio`) + - Number of modified pages ("dirty"). (`cachestat_dirties`) + - Number of accessed pages. (`cachestat_hits`) + - Number of pages brought from disk. (`cachestat_misses`) +- directory cache + - Ratio of files available in directory cache. (`dc_hit_ratio`) + - Number of files accessed. (`dc_reference`) + - Number of files accessed that were not in cache. (`dc_not_cache`) + - Number of files not found. (`dc_not_found`) +- ipc shm + - Number of calls to `shm_get`. (`shmget_call`) + - Number of calls to `shm_at`. (`shmat_call`) + - Number of calls to `shm_dt`. (`shmdt_call`) + - Number of calls to `shm_ctl`. (`shmctl_call`) + +### `[ebpf programs]` configuration options + +The eBPF collector enables and runs the following eBPF programs by default: + +- `fd` : This eBPF program creates charts that show information about calls to open files. +- `mount`: This eBPF program creates charts that show calls to syscalls mount(2) and umount(2). +- `shm`: This eBPF program creates charts that show calls to syscalls shmget(2), shmat(2), shmdt(2) and shmctl(2). +- `sync`: Monitor calls to syscalls sync(2), fsync(2), fdatasync(2), syncfs(2), msync(2), and sync_file_range(2). +- `network viewer`: This eBPF program creates charts with information about `TCP` and `UDP` functions, including the + bandwidth consumed by each. +- `vfs`: This eBPF program creates charts that show information about VFS (Virtual File System) functions. +- `process`: This eBPF program creates charts that show information about process life. When in `return` mode, it also + creates charts showing errors when these operations are executed. +- `hardirq`: This eBPF program creates charts that show information about time spent servicing individual hardware + interrupt requests (hard IRQs). +- `softirq`: This eBPF program creates charts that show information about time spent servicing individual software + interrupt requests (soft IRQs). +- `oomkill`: This eBPF program creates a chart that shows OOM kills for all applications recognized via + the `apps.plugin` integration. Note that this program will show application charts regardless of whether apps + integration is turned on or off. + +You can also enable the following eBPF programs: + +- `cachestat`: Netdata's eBPF data collector creates charts about the memory page cache. When the integration with + [`apps.plugin`](/collectors/apps.plugin/README.md) is enabled, this collector creates charts for the whole host _and_ + for each application. +- `dcstat` : This eBPF program creates charts that show information about file access using directory cache. It appends + `kprobes` for `lookup_fast()` and `d_lookup()` to identify if files are inside directory cache, outside and files are + not found. +- `disk` : This eBPF program creates charts that show information about disk latency independent of filesystem. +- `filesystem` : This eBPF program creates charts that show information about some filesystem latency. +- `swap` : This eBPF program creates charts that show information about swap access. +- `mdflush`: This eBPF program creates charts that show information about + multi-device software flushes. + +### Configuring eBPF threads + +You can configure each thread of the eBPF data collector. This allows you to overwrite global options defined in `/etc/netdata/ebpf.d.conf` and configure specific options for each thread. + +To configure an eBPF thread: + +1. Navigate to the [Netdata config directory](/docs/configure/nodes.md#the-netdata-config-directory). + ```bash + cd /etc/netdata + ``` +2. Use the [`edit-config`](/docs/configure/nodes.md#use-edit-config-to-edit-configuration-files) script to edit a thread configuration file. The following configuration files are available: + + - `network.conf`: Configuration for the [`network` thread](#network-configuration). This config file overwrites the global options and also + lets you specify which network the eBPF collector monitors. + - `process.conf`: Configuration for the [`process` thread](#sync-configuration). + - `cachestat.conf`: Configuration for the `cachestat` thread(#filesystem-configuration). + - `dcstat.conf`: Configuration for the `dcstat` thread. + - `disk.conf`: Configuration for the `disk` thread. + - `fd.conf`: Configuration for the `file descriptor` thread. + - `filesystem.conf`: Configuration for the `filesystem` thread. + - `hardirq.conf`: Configuration for the `hardirq` thread. + - `softirq.conf`: Configuration for the `softirq` thread. + - `sync.conf`: Configuration for the `sync` thread. + - `vfs.conf`: Configuration for the `vfs` thread. + + ```bash + ./edit-config FILE.conf + ``` + +### Network configuration + +The network configuration has specific options to configure which network(s) the eBPF collector monitors. These options +are divided in the following sections: + +#### `[network connections]` + +You can configure the information shown on `outbound` and `inbound` charts with the settings in this section. + +```conf +[network connections] + maximum dimensions = 500 + resolve hostname ips = no + ports = 1-1024 !145 !domain + hostnames = !example.com + ips = !127.0.0.1/8 10.0.0.0/8 172.16.0.0/12 192.168.0.0/16 fc00::/7 +``` + +When you define a `ports` setting, Netdata will collect network metrics for that specific port. For example, if you +write `ports = 19999`, Netdata will collect only connections for itself. The `hostnames` setting accepts +[simple patterns](/libnetdata/simple_pattern/README.md). The `ports`, and `ips` settings accept negation (`!`) to deny +specific values or asterisk alone to define all values. + +In the above example, Netdata will collect metrics for all ports between 1 and 443, with the exception of 53 (domain) +and 145. + +The following options are available: + +- `ports`: Define the destination ports for Netdata to monitor. +- `hostnames`: The list of hostnames that can be resolved to an IP address. +- `ips`: The IP or range of IPs that you want to monitor. You can use IPv4 or IPv6 addresses, use dashes to define a + range of IPs, or use CIDR values. By default, only data for private IP addresses is collected, but this can + be changed with the `ips` setting. + +By default, Netdata displays up to 500 dimensions on network connection charts. If there are more possible dimensions, +they will be bundled into the `other` dimension. You can increase the number of shown dimensions by changing +the `maximum dimensions` setting. + +The dimensions for the traffic charts are created using the destination IPs of the sockets by default. This can be +changed setting `resolve hostname ips = yes` and restarting Netdata, after this Netdata will create dimensions using +the `hostnames` every time that is possible to resolve IPs to their hostnames. + +#### `[service name]` + +Netdata uses the list of services in `/etc/services` to plot network connection charts. If this file does not contain +the name for a particular service you use in your infrastructure, you will need to add it to the `[service name]` +section. + +For example, Netdata's default port (`19999`) is not listed in `/etc/services`. To associate that port with the Netdata +service in network connection charts, and thus see the name of the service instead of its port, define it: + +```conf +[service name] + 19999 = Netdata +``` + +### Sync configuration + +The sync configuration has specific options to disable monitoring for syscalls. All syscalls are monitored by default. + +```conf +[syscalls] + sync = yes + msync = yes + fsync = yes + fdatasync = yes + syncfs = yes + sync_file_range = yes +``` + +### Filesystem configuration + +The filesystem configuration has specific options to disable monitoring for filesystems; by default, all filesystems are +monitored. + +```conf +[filesystem] + btrfsdist = yes + ext4dist = yes + nfsdist = yes + xfsdist = yes + zfsdist = yes +``` + +The ebpf program `nfsdist` monitors only `nfs` mount points. + +## Troubleshooting + +If the eBPF collector does not work, you can troubleshoot it by running the `ebpf.plugin` command and investigating its +output. + +```bash +cd /usr/libexec/netdata/plugins.d/ +sudo su -s /bin/bash ./ebpf.plugin +``` + +You can also use `grep` to search the Agent's `error.log` for messages related to eBPF monitoring. + +```bash +grep -i ebpf /var/log/netdata/error.log +``` + +### Confirm kernel compatibility + +The eBPF collector only works on Linux systems and with specific Linux kernels. We support all kernels more recent than +`4.11.0`, and all kernels on CentOS 7.6 or later. + +You can run our helper script to determine whether your system can support eBPF monitoring. If it returns no output, your system is ready to compile and run the eBPF collector. + +```bash +curl -sSL https://raw.githubusercontent.com/netdata/kernel-collector/master/tools/check-kernel-config.sh | sudo bash +``` + + +If you see a warning about a missing kernel +configuration (`KPROBES KPROBES_ON_FTRACE HAVE_KPROBES BPF BPF_SYSCALL BPF_JIT`), you will need to recompile your kernel +to support this configuration. The process of recompiling Linux kernels varies based on your distribution and version. +Read the documentation for your system's distribution to learn more about the specific workflow for recompiling the +kernel, ensuring that you set all the necessary + +- [Ubuntu](https://wiki.ubuntu.com/Kernel/BuildYourOwnKernel) +- [Debian](https://kernel-team.pages.debian.net/kernel-handbook/ch-common-tasks.html#s-common-official) +- [Fedora](https://fedoraproject.org/wiki/Building_a_custom_kernel) +- [CentOS](https://wiki.centos.org/HowTos/Custom_Kernel) +- [Arch Linux](https://wiki.archlinux.org/index.php/Kernel/Traditional_compilation) +- [Slackware](https://docs.slackware.com/howtos:slackware_admin:kernelbuilding) + +### Mount `debugfs` and `tracefs` + +The eBPF collector also requires both the `tracefs` and `debugfs` filesystems. Try mounting the `tracefs` and `debugfs` +filesystems using the commands below: + +```bash +sudo mount -t debugfs nodev /sys/kernel/debug +sudo mount -t tracefs nodev /sys/kernel/tracing +``` + +If they are already mounted, you will see an error. You can also configure your system's `/etc/fstab` configuration to +mount these filesystems on startup. More information can be found in +the [ftrace documentation](https://www.kernel.org/doc/Documentation/trace/ftrace.txt). + +## Charts + +The eBPF collector creates charts on different menus, like System Overview, Memory, MD arrays, Disks, Filesystem, +Mount Points, Networking Stack, systemd Services, and Applications. + +The collector stores the actual value inside of its process, but charts only show the difference between the values +collected in the previous and current seconds. + +### System overview + +Not all charts within the System Overview menu are enabled by default. Charts that rely on `kprobes` are disabled by default because they add around 100ns overhead for each function call. This is a small number from a human's perspective, but the functions are called many times and create an impact +on host. See the [configuration](#configuring-ebpfplugin) section for details about how to enable them. + +#### Processes + +Internally, the Linux kernel treats both processes and threads as `tasks`. To create a thread, the kernel offers a few +system calls: `fork(2)`, `vfork(2)`, and `clone(2)`. To generate this chart, the eBPF +collector uses the following `tracepoints` and `kprobe`: + +- `sched/sched_process_fork`: Tracepoint called after a call for `fork (2)`, `vfork (2)` and `clone (2)`. +- `sched/sched_process_exec`: Tracepoint called after a exec-family syscall. +- `kprobe/kernel_clone`: This is the main [`fork()`](https://elixir.bootlin.com/linux/v5.10/source/kernel/fork.c#L2415) + routine since kernel `5.10.0` was released. +- `kprobe/_do_fork`: Like `kernel_clone`, but this was the main function between kernels `4.2.0` and `5.9.16` +- `kprobe/do_fork`: This was the main function before kernel `4.2.0`. + +#### Process Exit + +Ending a task requires two steps. The first is a call to the internal function `do_exit`, which notifies the operating +system that the task is finishing its work. The second step is to release the kernel information with the internal +function `release_task`. The difference between the two dimensions can help you discover +[zombie processes](https://en.wikipedia.org/wiki/Zombie_process). To get the metrics, the collector uses: + +- `sched/sched_process_exit`: Tracepoint called after a task exits. +- `kprobe/release_task`: This function is called when a process exits, as the kernel still needs to remove the process + descriptor. + +#### Task error + +The functions responsible for ending tasks do not return values, so this chart contains information about failures on +process and thread creation only. + +#### Swap + +Inside the swap submenu the eBPF plugin creates the chart `swapcalls`; this chart is displaying when processes are +calling functions [`swap_readpage` and `swap_writepage`](https://hzliu123.github.io/linux-kernel/Page%20Cache%20in%20Linux%202.6.pdf), +which are functions responsible for doing IO in swap memory. To collect the exact moment that an access to swap happens, +the collector attaches `kprobes` for cited functions. + +#### Soft IRQ + +The following `tracepoints` are used to measure time usage for soft IRQs: + +- [`irq/softirq_entry`](https://www.kernel.org/doc/html/latest/core-api/tracepoint.html#c.trace_softirq_entry): Called + before softirq handler +- [`irq/softirq_exit`](https://www.kernel.org/doc/html/latest/core-api/tracepoint.html#c.trace_softirq_exit): Called when + softirq handler returns. + +#### Hard IRQ + +The following tracepoints are used to measure the latency of servicing a +hardware interrupt request (hard IRQ). + +- [`irq/irq_handler_entry`](https://www.kernel.org/doc/html/latest/core-api/tracepoint.html#c.trace_irq_handler_entry): + Called immediately before the IRQ action handler. +- [`irq/irq_handler_exit`](https://www.kernel.org/doc/html/latest/core-api/tracepoint.html#c.trace_irq_handler_exit): + Called immediately after the IRQ action handler returns. +- `irq_vectors`: These are traces from `irq_handler_entry` and + `irq_handler_exit` when an IRQ is handled. The following elements from vector + are triggered: + - `irq_vectors/local_timer_entry` + - `irq_vectors/local_timer_exit` + - `irq_vectors/reschedule_entry` + - `irq_vectors/reschedule_exit` + - `irq_vectors/call_function_entry` + - `irq_vectors/call_function_exit` + - `irq_vectors/call_function_single_entry` + - `irq_vectors/call_function_single_xit` + - `irq_vectors/irq_work_entry` + - `irq_vectors/irq_work_exit` + - `irq_vectors/error_apic_entry` + - `irq_vectors/error_apic_exit` + - `irq_vectors/thermal_apic_entry` + - `irq_vectors/thermal_apic_exit` + - `irq_vectors/threshold_apic_entry` + - `irq_vectors/threshold_apic_exit` + - `irq_vectors/deferred_error_entry` + - `irq_vectors/deferred_error_exit` + - `irq_vectors/spurious_apic_entry` + - `irq_vectors/spurious_apic_exit` + - `irq_vectors/x86_platform_ipi_entry` + - `irq_vectors/x86_platform_ipi_exit` + +#### IPC shared memory + +To monitor shared memory system call counts, Netdata attaches tracing in the following functions: + +- `shmget`: Runs when [`shmget`](https://man7.org/linux/man-pages/man2/shmget.2.html) is called. +- `shmat`: Runs when [`shmat`](https://man7.org/linux/man-pages/man2/shmat.2.html) is called. +- `shmdt`: Runs when [`shmdt`](https://man7.org/linux/man-pages/man2/shmat.2.html) is called. +- `shmctl`: Runs when [`shmctl`](https://man7.org/linux/man-pages/man2/shmctl.2.html) is called. + +### Memory + +In the memory submenu the eBPF plugin creates two submenus **page cache** and **synchronization** with the following +organization: + +- Page Cache + - Page cache ratio + - Dirty pages + - Page cache hits + - Page cache misses +- Synchronization + - File sync + - Memory map sync + - File system sync + - File range sync + +#### Page cache hits + +When the processor needs to read or write a location in main memory, it checks for a corresponding entry in the page cache. + If the entry is there, a page cache hit has occurred and the read is from the cache. + +A page cache hit is when the page cache is successfully accessed with a read operation. We do not count pages that were +added relatively recently. + +#### Dirty pages + +A "dirty page" is a page in the page cache that was modified after being created. Since non-dirty pages in the page cache + have identical copies in secondary storage (e.g. hard disk drive or solid-state drive), discarding and reusing their space + is much quicker than paging out application memory, and is often preferred over flushing the dirty pages into secondary storage + and reusing their space. + +On `cachestat_dirties` Netdata demonstrates the number of pages that were modified. This chart shows the number of calls +to the function `mark_buffer_dirty`. + +#### Page cache ratio + +When the processor needs to read or write in a specific memory address, it checks for a corresponding entry in the page cache. +If the processor hits a page cache (`page cache hit`), it reads the entry from the cache. If there is no entry (`page cache miss`), + the kernel allocates a new entry and copies data from the disk. Netdata calculates the percentage of accessed files that are cached on + memory. The ratio is calculated counting the accessed cached pages + (without counting [dirty pages](#dirty-pages) and pages added because of read misses) divided by total access without dirty pages. + +> \_\_**\_\_\_\_**<ins>Number of accessed cached pages</ins>\***\*\_\_\*\***<br/> +> Number of total accessed pages - dirty pages - missed pages + +The chart `cachestat_ratio` shows how processes are accessing page cache. In a normal scenario, we expect values around +100%, which means that the majority of the work on the machine is processed in memory. To calculate the ratio, Netdata +attaches `kprobes` for kernel functions: + +- `add_to_page_cache_lru`: Page addition. +- `mark_page_accessed`: Access to cache. +- `account_page_dirtied`: Dirty (modified) pages. +- `mark_buffer_dirty`: Writes to page cache. + +#### Page cache misses + +A page cache miss means that a page was not inside memory when the process tried to access it. This chart shows the +result of the difference for calls between functions `add_to_page_cache_lru` and `account_page_dirtied`. + +#### File sync + +This chart shows calls to synchronization methods, [`fsync(2)`](https://man7.org/linux/man-pages/man2/fdatasync.2.html) +and [`fdatasync(2)`](https://man7.org/linux/man-pages/man2/fdatasync.2.html), to transfer all modified page caches +for the files on disk devices. These calls block until the disk reports that the transfer has been completed. They flush +data for specific file descriptors. + +#### Memory map sync + +The chart shows calls to [`msync(2)`](https://man7.org/linux/man-pages/man2/msync.2.html) syscalls. This syscall flushes +changes to a file that was mapped into memory using [`mmap(2)`](https://man7.org/linux/man-pages/man2/mmap.2.html). + +#### File system sync + +This chart monitors calls demonstrating commits from filesystem caches to disk. Netdata attaches `tracing` for +[`sync(2)`](https://man7.org/linux/man-pages/man2/sync.2.html), and [`syncfs(2)`](https://man7.org/linux/man-pages/man2/sync.2.html). + +#### File range sync + +This chart shows calls to [`sync_file_range(2)`](https://man7.org/linux/man-pages/man2/sync_file_range.2.html) which +synchronizes file segments with disk. + +> Note: This is the most dangerous syscall to synchronize data, according to its manual. + +### Multiple Device (MD) arrays + +The eBPF plugin shows multi-device flushes happening in real time. This can be used to explain some spikes happening +in [disk latency](#disk) charts. + +By default, MD flush is disabled. To enable it, configure your +`/etc/netdata/ebpf.d.conf` file as: + +```conf +[global] + mdflush = yes +``` + +#### MD flush + +To collect data related to Linux multi-device (MD) flushing, the following kprobe is used: + +- `kprobe/md_flush_request`: called whenever a request for flushing multi-device data is made. + +### Disk + +The eBPF plugin also shows a chart in the Disk section when the `disk` thread is enabled. + +#### Disk Latency + +This will create the chart `disk_latency_io` for each disk on the host. The following tracepoints are used: + +- [`block/block_rq_issue`](https://www.kernel.org/doc/html/latest/core-api/tracepoint.html#c.trace_block_rq_issue): + IO request operation to a device drive. +- [`block/block_rq_complete`](https://www.kernel.org/doc/html/latest/core-api/tracepoint.html#c.trace_block_rq_complete): + IO operation completed by device. + +Disk Latency is the single most important metric to focus on when it comes to storage performance, under most circumstances. +For hard drives, an average latency somewhere between 10 to 20 ms can be considered acceptable. For SSD (Solid State Drives), +in most cases, workloads experience less than 1 ms latency numbers, but workloads should never reach higher than 3 ms. +The dimensions refer to time intervals. + +### Filesystem + +This group has charts demonstrating how applications interact with the Linux kernel to open and close file descriptors. +It also brings latency charts for several different filesystems. + +#### Latency Algorithm + +We calculate the difference between the calling and return times, spanning disk I/O, file system operations (lock, I/O), +run queue latency and all events related to the monitored action. + +#### ext4 + +To measure the latency of executing some actions in an +[ext4](https://elixir.bootlin.com/linux/latest/source/fs/ext4) filesystem, the +collector needs to attach `kprobes` and `kretprobes` for each of the following +functions: + +- `ext4_file_read_iter`: Function used to measure read latency. +- `ext4_file_write_iter`: Function used to measure write latency. +- `ext4_file_open`: Function used to measure open latency. +- `ext4_sync_file`: Function used to measure sync latency. + +#### ZFS + +To measure the latency of executing some actions in a zfs filesystem, the +collector needs to attach `kprobes` and `kretprobes` for each of the following +functions: + +- `zpl_iter_read`: Function used to measure read latency. +- `zpl_iter_write`: Function used to measure write latency. +- `zpl_open`: Function used to measure open latency. +- `zpl_fsync`: Function used to measure sync latency. + +#### XFS + +To measure the latency of executing some actions in an +[xfs](https://elixir.bootlin.com/linux/latest/source/fs/xfs) filesystem, the +collector needs to attach `kprobes` and `kretprobes` for each of the following +functions: + +- `xfs_file_read_iter`: Function used to measure read latency. +- `xfs_file_write_iter`: Function used to measure write latency. +- `xfs_file_open`: Function used to measure open latency. +- `xfs_file_fsync`: Function used to measure sync latency. + +#### NFS + +To measure the latency of executing some actions in an +[nfs](https://elixir.bootlin.com/linux/latest/source/fs/nfs) filesystem, the +collector needs to attach `kprobes` and `kretprobes` for each of the following +functions: + +- `nfs_file_read`: Function used to measure read latency. +- `nfs_file_write`: Function used to measure write latency. +- `nfs_file_open`: Functions used to measure open latency. +- `nfs4_file_open`: Functions used to measure open latency for NFS v4. +- `nfs_getattr`: Function used to measure sync latency. + +#### btrfs + +To measure the latency of executing some actions in a [btrfs](https://elixir.bootlin.com/linux/latest/source/fs/btrfs/file.c) +filesystem, the collector needs to attach `kprobes` and `kretprobes` for each of the following functions: + +> Note: We are listing two functions used to measure `read` latency, but we use either `btrfs_file_read_iter` or +> `generic_file_read_iter`, depending on kernel version. + +- `btrfs_file_read_iter`: Function used to measure read latency since kernel `5.10.0`. +- `generic_file_read_iter`: Like `btrfs_file_read_iter`, but this function was used before kernel `5.10.0`. +- `btrfs_file_write_iter`: Function used to write data. +- `btrfs_file_open`: Function used to open files. +- `btrfs_sync_file`: Function used to synchronize data to filesystem. + +#### File descriptor + +To give metrics related to `open` and `close` events, instead of attaching kprobes for each syscall used to do these +events, the collector attaches `kprobes` for the common function used for syscalls: + +- [`do_sys_open`](https://0xax.gitbooks.io/linux-insides/content/SysCall/linux-syscall-5.html): Internal function used to + open files. +- [`do_sys_openat2`](https://elixir.bootlin.com/linux/v5.6/source/fs/open.c#L1162): + Function called from `do_sys_open` since version `5.6.0`. +- [`close_fd`](https://www.mail-archive.com/linux-kernel@vger.kernel.org/msg2271761.html): Function used to close file + descriptor since kernel `5.11.0`. +- `__close_fd`: Function used to close files before version `5.11.0`. + +#### File error + +This chart shows the number of times some software tried and failed to open or close a file descriptor. + +#### VFS + +The Linux Virtual File System (VFS) is an abstraction layer on top of a +concrete filesystem like the ones listed in the parent section, e.g. `ext4`. + +In this section we list the mechanism by which we gather VFS data, and what +charts are consequently created. + +##### VFS eBPF Hooks + +To measure the latency and total quantity of executing some VFS-level +functions, ebpf.plugin needs to attach kprobes and kretprobes for each of the +following functions: + +- `vfs_write`: Function used monitoring the number of successful & failed + filesystem write calls, as well as the total number of written bytes. +- `vfs_writev`: Same function as `vfs_write` but for vector writes (i.e. a + single write operation using a group of buffers rather than 1). +- `vfs_read`: Function used for monitoring the number of successful & failed + filesystem read calls, as well as the total number of read bytes. +- `vfs_readv` Same function as `vfs_read` but for vector reads (i.e. a single + read operation using a group of buffers rather than 1). +- `vfs_unlink`: Function used for monitoring the number of successful & failed + filesystem unlink calls. +- `vfs_fsync`: Function used for monitoring the number of successful & failed + filesystem fsync calls. +- `vfs_open`: Function used for monitoring the number of successful & failed + filesystem open calls. +- `vfs_create`: Function used for monitoring the number of successful & failed + filesystem create calls. + +##### VFS Deleted objects + +This chart monitors calls to `vfs_unlink`. This function is responsible for removing objects from the file system. + +##### VFS IO + +This chart shows the number of calls to the functions `vfs_read` and `vfs_write`. + +##### VFS IO bytes + +This chart also monitors `vfs_read` and `vfs_write` but, instead of the number of calls, it shows the total amount of +bytes read and written with these functions. + +The Agent displays the number of bytes written as negative because they are moving down to disk. + +##### VFS IO errors + +The Agent counts and shows the number of instances where a running program experiences a read or write error. + +##### VFS Create + +This chart shows the number of calls to `vfs_create`. This function is responsible for creating files. + +##### VFS Synchronization + +This chart shows the number of calls to `vfs_fsync`. This function is responsible for calling `fsync(2)` or +`fdatasync(2)` on a file. You can see more details in the Synchronization section. + +##### VFS Open + +This chart shows the number of calls to `vfs_open`. This function is responsible for opening files. + +#### Directory Cache + +Metrics for directory cache are collected using kprobe for `lookup_fast`, because we are interested in the number of +times this function is accessed. On the other hand, for `d_lookup` we are not only interested in the number of times it +is accessed, but also in possible errors, so we need to attach a `kretprobe`. For this reason, the following is used: + +- [`lookup_fast`](https://lwn.net/Articles/649115/): Called to look at data inside the directory cache. +- [`d_lookup`](https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/fs/dcache.c?id=052b398a43a7de8c68c13e7fa05d6b3d16ce6801#n2223): + Called when the desired file is not inside the directory cache. + +##### Directory Cache Interpretation + +When directory cache is showing 100% that means that every accessed file was present in the directory cache. +If files are not present in the directory cache, they are either not present in the file system or the files were not +accessed before. + +### Mount Points + +The following `tracing` are used to collect `mount` & `unmount` call counts: + +- [`mount`](https://man7.org/linux/man-pages/man2/mount.2.html): mount filesystem on host. +- [`umount`](https://man7.org/linux/man-pages/man2/umount.2.html): umount filesystem on host. + +### Networking Stack + +Netdata monitors socket bandwidth attaching `tracing` for internal functions. + +#### TCP outbound connections + +This chart demonstrates calls to `tcp_v4_connection` and `tcp_v6_connection` that start connections for IPV4 and IPV6, respectively. + +#### TCP inbound connections + +This chart demonstrates TCP and UDP connections that the host receives. +To collect this information, netdata attaches a tracing to `inet_csk_accept`. + +#### TCP bandwidth functions + +This chart demonstrates calls to functions `tcp_sendmsg`, `tcp_cleanup_rbuf`, and `tcp_close`; these functions are used +to send & receive data and to close connections when `TCP` protocol is used. + +#### TCP bandwidth + +This chart demonstrates calls to functions: + +- `tcp_sendmsg`: Function responsible to send data for a specified destination. +- `tcp_cleanup_rbuf`: We use this function instead of `tcp_recvmsg`, because the last one misses `tcp_read_sock` traffic + and we would also need to add more `tracing` to get the socket and package size. +- `tcp_close`: Function responsible to close connection. + +#### TCP retransmit + +This chart demonstrates calls to function `tcp_retransmit` that is responsible for executing TCP retransmission when the +receiver did not return the packet during the expected time. + +#### UDP functions + +This chart demonstrates calls to functions `udp_sendmsg` and `udp_recvmsg`, which are responsible for sending & +receiving data for connections when the `UDP` protocol is used. + +#### UDP bandwidth + +Like the previous chart, this one also monitors `udp_sendmsg` and `udp_recvmsg`, but instead of showing the number of +calls, it monitors the number of bytes sent and received. + +### Apps + +#### OOM Killing + +These are tracepoints related to [OOM](https://en.wikipedia.org/wiki/Out_of_memory) killing processes. + +- `oom/mark_victim`: Monitors when an oomkill event happens. + +## Known issues + +### Performance opimization + +eBPF monitoring is complex and produces a large volume of metrics. We've discovered scenarios where the eBPF plugin +significantly increases kernel memory usage by several hundred MB. + +If your node is experiencing high memory usage and there is no obvious culprit to be found in the `apps.mem` chart, +consider testing for high kernel memory usage by [disabling eBPF monitoring](#configuring-ebpfplugin). Next, +[restart Netdata](/docs/configure/start-stop-restart.md) with `sudo systemctl restart netdata` to see if system memory +usage (see the `system.ram` chart) has dropped significantly. + +Beginning with `v1.31`, kernel memory usage is configurable via the [`pid table size` setting](#ebpf-load-mode) +in `ebpf.conf`. + +### SELinux + +When [SELinux](https://www.redhat.com/en/topics/linux/what-is-selinux) is enabled, it may prevent `ebpf.plugin` from +starting correctly. Check the Agent's `error.log` file for errors like the ones below: + +```bash +2020-06-14 15:32:08: ebpf.plugin ERROR : EBPF PROCESS : Cannot load program: /usr/libexec/netdata/plugins.d/pnetdata_ebpf_process.3.10.0.o (errno 13, Permission denied) +2020-06-14 15:32:19: netdata ERROR : PLUGINSD[ebpf] : read failed: end of file (errno 9, Bad file descriptor) +``` + +You can also check for errors related to `ebpf.plugin` inside `/var/log/audit/audit.log`: + +```bash +type=AVC msg=audit(1586260134.952:97): avc: denied { map_create } for pid=1387 comm="ebpf.pl" scontext=system_u:system_r:unconfined_service_t:s0 tcontext=system_u:system_r:unconfined_service_t:s0 tclass=bpf permissive=0 +type=SYSCALL msg=audit(1586260134.952:97): arch=c000003e syscall=321 success=no exit=-13 a0=0 a1=7ffe6b36f000 a2=70 a3=0 items=0 ppid=1135 pid=1387 auid=4294967295 uid=994 gid=990 euid=0 suid=0 fsuid=0 egid=990 sgid=990 fsgid=990 tty=(none) ses=4294967295 comm="ebpf_proc +ess.pl" exe="/usr/libexec/netdata/plugins.d/ebpf.plugin" subj=system_u:system_r:unconfined_service_t:s0 key=(null) +``` + +If you see similar errors, you will have to adjust SELinux's policies to enable the eBPF collector. + +#### Creation of bpf policies + +To enable `ebpf.plugin` to run on a distribution with SELinux enabled, it will be necessary to take the following +actions. + +First, stop the Netdata Agent. + +```bash +# systemctl stop netdata +``` + +Next, create a policy with the `audit.log` file you examined earlier. + +```bash +# grep ebpf.plugin /var/log/audit/audit.log | audit2allow -M netdata_ebpf +``` + +This will create two new files: `netdata_ebpf.te` and `netdata_ebpf.mod`. + +Edit the `netdata_ebpf.te` file to change the options `class` and `allow`. You should have the following at the end of +the `netdata_ebpf.te` file. + +```conf +module netdata_ebpf 1.0; +require { + type unconfined_service_t; + class bpf { map_create map_read map_write prog_load prog_run }; +} +#============= unconfined_service_t ============== +allow unconfined_service_t self:bpf { map_create map_read map_write prog_load prog_run }; +``` + +Then compile your `netdata_ebpf.te` file with the following commands to create a binary that loads the new policies: + +```bash +# checkmodule -M -m -o netdata_ebpf.mod netdata_ebpf.te +# semodule_package -o netdata_ebpf.pp -m netdata_ebpf.mod +``` + +Finally, you can load the new policy and start the Netdata agent again: + +```bash +# semodule -i netdata_ebpf.pp +# systemctl start netdata +``` + +### Linux kernel lockdown + +Beginning with [version 5.4](https://www.zdnet.com/article/linux-to-get-kernel-lockdown-feature/), the Linux kernel has +a feature called "lockdown," which may affect `ebpf.plugin` depending how the kernel was compiled. The following table +shows how the lockdown module impacts `ebpf.plugin` based on the selected options: + +| Enforcing kernel lockdown | Enable lockdown LSM early in init | Default lockdown mode | Can `ebpf.plugin` run with this? | +| :------------------------ | :-------------------------------- | :-------------------- | :------------------------------- | +| YES | NO | NO | YES | +| YES | Yes | None | YES | +| YES | Yes | Integrity | YES | +| YES | Yes | Confidentiality | NO | + +If you or your distribution compiled the kernel with the last combination, your system cannot load shared libraries +required to run `ebpf.plugin`. diff --git a/collectors/ebpf.plugin/ebpf.c b/collectors/ebpf.plugin/ebpf.c new file mode 100644 index 0000000..00b53a5 --- /dev/null +++ b/collectors/ebpf.plugin/ebpf.c @@ -0,0 +1,2249 @@ +// SPDX-License-Identifier: GPL-3.0-or-later + +#include <sys/time.h> +#include <sys/resource.h> +#include <ifaddrs.h> + +#include "ebpf.h" +#include "ebpf_socket.h" +#include "libnetdata/required_dummies.h" + +/***************************************************************** + * + * GLOBAL VARIABLES + * + *****************************************************************/ + +char *ebpf_plugin_dir = PLUGINS_DIR; +static char *ebpf_configured_log_dir = LOG_DIR; + +char *ebpf_algorithms[] = {"absolute", "incremental"}; +struct config collector_config = { .first_section = NULL, + .last_section = NULL, + .mutex = NETDATA_MUTEX_INITIALIZER, + .index = { .avl_tree = { .root = NULL, .compar = appconfig_section_compare }, + .rwlock = AVL_LOCK_INITIALIZER } }; + +int running_on_kernel = 0; +int ebpf_nprocs; +int isrh = 0; +int main_thread_id = 0; + +pthread_mutex_t lock; +pthread_mutex_t ebpf_exit_cleanup; +pthread_mutex_t collect_data_mutex; +pthread_cond_t collect_data_cond_var; + +ebpf_module_t ebpf_modules[] = { + { .thread_name = "process", .config_name = "process", .enabled = 0, .start_routine = ebpf_process_thread, + .update_every = EBPF_DEFAULT_UPDATE_EVERY, .global_charts = 1, .apps_charts = NETDATA_EBPF_APPS_FLAG_NO, + .apps_level = NETDATA_APPS_LEVEL_REAL_PARENT, .cgroup_charts = CONFIG_BOOLEAN_NO, .mode = MODE_ENTRY, .optional = 0, + .apps_routine = ebpf_process_create_apps_charts, .maps = NULL, + .pid_map_size = ND_EBPF_DEFAULT_PID_SIZE, .names = NULL, .cfg = &process_config, + .config_file = NETDATA_PROCESS_CONFIG_FILE, + .kernels = NETDATA_V3_10 | NETDATA_V4_14 | NETDATA_V4_16 | NETDATA_V4_18 | NETDATA_V5_4 | NETDATA_V5_10 | + NETDATA_V5_14, + .load = EBPF_LOAD_LEGACY, .targets = NULL, .probe_links = NULL, .objects = NULL, .thread = NULL}, + { .thread_name = "socket", .config_name = "socket", .enabled = 0, .start_routine = ebpf_socket_thread, + .update_every = EBPF_DEFAULT_UPDATE_EVERY, .global_charts = 1, .apps_charts = NETDATA_EBPF_APPS_FLAG_NO, + .apps_level = NETDATA_APPS_LEVEL_REAL_PARENT, .cgroup_charts = CONFIG_BOOLEAN_NO, .mode = MODE_ENTRY, .optional = 0, + .apps_routine = ebpf_socket_create_apps_charts, .maps = NULL, + .pid_map_size = ND_EBPF_DEFAULT_PID_SIZE, .names = NULL, .cfg = &socket_config, + .config_file = NETDATA_NETWORK_CONFIG_FILE, + .kernels = NETDATA_V3_10 | NETDATA_V4_14 | NETDATA_V4_16 | NETDATA_V4_18 | NETDATA_V5_4 | NETDATA_V5_14, + .load = EBPF_LOAD_LEGACY, .targets = socket_targets, .probe_links = NULL, .objects = NULL, .thread = NULL}, + { .thread_name = "cachestat", .config_name = "cachestat", .enabled = 0, .start_routine = ebpf_cachestat_thread, + .update_every = EBPF_DEFAULT_UPDATE_EVERY, .global_charts = 1, .apps_charts = NETDATA_EBPF_APPS_FLAG_NO, + .apps_level = NETDATA_APPS_LEVEL_REAL_PARENT, .cgroup_charts = CONFIG_BOOLEAN_NO, .mode = MODE_ENTRY, .optional = 0, + .apps_routine = ebpf_cachestat_create_apps_charts, .maps = cachestat_maps, + .pid_map_size = ND_EBPF_DEFAULT_PID_SIZE, .names = NULL, .cfg = &cachestat_config, + .config_file = NETDATA_CACHESTAT_CONFIG_FILE, + .kernels = NETDATA_V3_10 | NETDATA_V4_14 | NETDATA_V4_16 | NETDATA_V4_18| + NETDATA_V5_4 | NETDATA_V5_14 | NETDATA_V5_15 | NETDATA_V5_16, + .load = EBPF_LOAD_LEGACY, .targets = cachestat_targets, .probe_links = NULL, .objects = NULL, .thread = NULL}, + { .thread_name = "sync", .config_name = "sync", .enabled = 0, .start_routine = ebpf_sync_thread, + .update_every = EBPF_DEFAULT_UPDATE_EVERY, .global_charts = 1, .apps_charts = NETDATA_EBPF_APPS_FLAG_NO, + .apps_level = NETDATA_APPS_NOT_SET, .cgroup_charts = CONFIG_BOOLEAN_NO, .mode = MODE_ENTRY, .optional = 0, + .apps_routine = NULL, .maps = NULL, .pid_map_size = ND_EBPF_DEFAULT_PID_SIZE, .names = NULL, .cfg = &sync_config, + .config_file = NETDATA_SYNC_CONFIG_FILE, + // All syscalls have the same kernels + .kernels = NETDATA_V3_10 | NETDATA_V4_14 | NETDATA_V4_16 | NETDATA_V4_18 | NETDATA_V5_4 | NETDATA_V5_14, + .load = EBPF_LOAD_LEGACY, .targets = sync_targets, .probe_links = NULL, .objects = NULL, .thread = NULL}, + { .thread_name = "dc", .config_name = "dc", .enabled = 0, .start_routine = ebpf_dcstat_thread, + .update_every = EBPF_DEFAULT_UPDATE_EVERY, .global_charts = 1, .apps_charts = NETDATA_EBPF_APPS_FLAG_NO, + .apps_level = NETDATA_APPS_LEVEL_REAL_PARENT, .cgroup_charts = CONFIG_BOOLEAN_NO, .mode = MODE_ENTRY, .optional = 0, + .apps_routine = ebpf_dcstat_create_apps_charts, .maps = dcstat_maps, + .pid_map_size = ND_EBPF_DEFAULT_PID_SIZE, .names = NULL, .cfg = &dcstat_config, + .config_file = NETDATA_DIRECTORY_DCSTAT_CONFIG_FILE, + .kernels = NETDATA_V3_10 | NETDATA_V4_14 | NETDATA_V4_16 | NETDATA_V4_18 | NETDATA_V5_4 | NETDATA_V5_14, + .load = EBPF_LOAD_LEGACY, .targets = dc_targets, .probe_links = NULL, .objects = NULL, .thread = NULL}, + { .thread_name = "swap", .config_name = "swap", .enabled = 0, .start_routine = ebpf_swap_thread, + .update_every = EBPF_DEFAULT_UPDATE_EVERY, .global_charts = 1, .apps_charts = NETDATA_EBPF_APPS_FLAG_NO, + .apps_level = NETDATA_APPS_LEVEL_REAL_PARENT, .cgroup_charts = CONFIG_BOOLEAN_NO, .mode = MODE_ENTRY, .optional = 0, + .apps_routine = ebpf_swap_create_apps_charts, .maps = NULL, + .pid_map_size = ND_EBPF_DEFAULT_PID_SIZE, .names = NULL, .cfg = &swap_config, + .config_file = NETDATA_DIRECTORY_SWAP_CONFIG_FILE, + .kernels = NETDATA_V3_10 | NETDATA_V4_14 | NETDATA_V4_16 | NETDATA_V4_18 | NETDATA_V5_4 | NETDATA_V5_14, + .load = EBPF_LOAD_LEGACY, .targets = swap_targets, .probe_links = NULL, .objects = NULL, .thread = NULL}, + { .thread_name = "vfs", .config_name = "vfs", .enabled = 0, .start_routine = ebpf_vfs_thread, + .update_every = EBPF_DEFAULT_UPDATE_EVERY, .global_charts = 1, .apps_charts = NETDATA_EBPF_APPS_FLAG_NO, + .apps_level = NETDATA_APPS_LEVEL_REAL_PARENT, .cgroup_charts = CONFIG_BOOLEAN_NO, .mode = MODE_ENTRY, .optional = 0, + .apps_routine = ebpf_vfs_create_apps_charts, .maps = NULL, + .pid_map_size = ND_EBPF_DEFAULT_PID_SIZE, .names = NULL, .cfg = &vfs_config, + .config_file = NETDATA_DIRECTORY_VFS_CONFIG_FILE, + .kernels = NETDATA_V3_10 | NETDATA_V4_14 | NETDATA_V4_16 | NETDATA_V4_18 | NETDATA_V5_4 | NETDATA_V5_14, + .load = EBPF_LOAD_LEGACY, .targets = vfs_targets, .probe_links = NULL, .objects = NULL, .thread = NULL}, + { .thread_name = "filesystem", .config_name = "filesystem", .enabled = 0, .start_routine = ebpf_filesystem_thread, + .update_every = EBPF_DEFAULT_UPDATE_EVERY, .global_charts = 1, .apps_charts = NETDATA_EBPF_APPS_FLAG_NO, + .apps_level = NETDATA_APPS_NOT_SET, .cgroup_charts = CONFIG_BOOLEAN_NO, .mode = MODE_ENTRY, .optional = 0, + .apps_routine = NULL, .maps = NULL, .pid_map_size = ND_EBPF_DEFAULT_PID_SIZE, .names = NULL, .cfg = &fs_config, + .config_file = NETDATA_FILESYSTEM_CONFIG_FILE, + //We are setting kernels as zero, because we load eBPF programs according the kernel running. + .kernels = 0, .load = EBPF_LOAD_LEGACY, .targets = NULL, .probe_links = NULL, .objects = NULL, .thread = NULL }, + { .thread_name = "disk", .config_name = "disk", .enabled = 0, .start_routine = ebpf_disk_thread, + .update_every = EBPF_DEFAULT_UPDATE_EVERY, .global_charts = 1, .apps_charts = NETDATA_EBPF_APPS_FLAG_NO, + .apps_level = NETDATA_APPS_NOT_SET, .cgroup_charts = CONFIG_BOOLEAN_NO, .mode = MODE_ENTRY, .optional = 0, + .apps_routine = NULL, .maps = NULL, .pid_map_size = ND_EBPF_DEFAULT_PID_SIZE, .names = NULL, .cfg = &disk_config, + .config_file = NETDATA_DISK_CONFIG_FILE, + .kernels = NETDATA_V3_10 | NETDATA_V4_14 | NETDATA_V4_16 | NETDATA_V4_18 | NETDATA_V5_4 | NETDATA_V5_14, + .load = EBPF_LOAD_LEGACY, .targets = NULL, .probe_links = NULL, .objects = NULL, .thread = NULL}, + { .thread_name = "mount", .config_name = "mount", .enabled = 0, .start_routine = ebpf_mount_thread, + .update_every = EBPF_DEFAULT_UPDATE_EVERY, .global_charts = 1, .apps_charts = NETDATA_EBPF_APPS_FLAG_NO, + .apps_level = NETDATA_APPS_NOT_SET, .cgroup_charts = CONFIG_BOOLEAN_NO, .mode = MODE_ENTRY, .optional = 0, + .apps_routine = NULL, .maps = NULL, .pid_map_size = ND_EBPF_DEFAULT_PID_SIZE, .names = NULL, .cfg = &mount_config, + .config_file = NETDATA_MOUNT_CONFIG_FILE, + .kernels = NETDATA_V3_10 | NETDATA_V4_14 | NETDATA_V4_16 | NETDATA_V4_18 | NETDATA_V5_4 | NETDATA_V5_14, + .load = EBPF_LOAD_LEGACY, .targets = mount_targets, .probe_links = NULL, .objects = NULL, .thread = NULL}, + { .thread_name = "fd", .config_name = "fd", .enabled = 0, .start_routine = ebpf_fd_thread, + .update_every = EBPF_DEFAULT_UPDATE_EVERY, .global_charts = 1, .apps_charts = NETDATA_EBPF_APPS_FLAG_NO, + .apps_level = NETDATA_APPS_LEVEL_REAL_PARENT, .cgroup_charts = CONFIG_BOOLEAN_NO, .mode = MODE_ENTRY, .optional = 0, + .apps_routine = ebpf_fd_create_apps_charts, .maps = NULL, + .pid_map_size = ND_EBPF_DEFAULT_PID_SIZE, .names = NULL, .cfg = &fd_config, + .config_file = NETDATA_FD_CONFIG_FILE, + .kernels = NETDATA_V3_10 | NETDATA_V4_14 | NETDATA_V4_16 | NETDATA_V4_18 | NETDATA_V5_4 | NETDATA_V5_11 | + NETDATA_V5_14, + .load = EBPF_LOAD_LEGACY, .targets = fd_targets, .probe_links = NULL, .objects = NULL, .thread = NULL}, + { .thread_name = "hardirq", .config_name = "hardirq", .enabled = 0, .start_routine = ebpf_hardirq_thread, + .update_every = EBPF_DEFAULT_UPDATE_EVERY, .global_charts = 1, .apps_charts = NETDATA_EBPF_APPS_FLAG_NO, + .apps_level = NETDATA_APPS_NOT_SET, .cgroup_charts = CONFIG_BOOLEAN_NO, .mode = MODE_ENTRY, .optional = 0, + .apps_routine = NULL, .maps = NULL, .pid_map_size = ND_EBPF_DEFAULT_PID_SIZE, .names = NULL, .cfg = &hardirq_config, + .config_file = NETDATA_HARDIRQ_CONFIG_FILE, + .kernels = NETDATA_V3_10 | NETDATA_V4_14 | NETDATA_V4_16 | NETDATA_V4_18 | NETDATA_V5_4 | NETDATA_V5_14, + .load = EBPF_LOAD_LEGACY, .targets = NULL, .probe_links = NULL, .objects = NULL, .thread = NULL}, + { .thread_name = "softirq", .config_name = "softirq", .enabled = 0, .start_routine = ebpf_softirq_thread, + .update_every = EBPF_DEFAULT_UPDATE_EVERY, .global_charts = 1, .apps_charts = NETDATA_EBPF_APPS_FLAG_NO, + .apps_level = NETDATA_APPS_NOT_SET, .cgroup_charts = CONFIG_BOOLEAN_NO, .mode = MODE_ENTRY, .optional = 0, + .apps_routine = NULL, .maps = NULL, .pid_map_size = ND_EBPF_DEFAULT_PID_SIZE, .names = NULL, .cfg = &softirq_config, + .config_file = NETDATA_SOFTIRQ_CONFIG_FILE, + .kernels = NETDATA_V3_10 | NETDATA_V4_14 | NETDATA_V4_16 | NETDATA_V4_18 | NETDATA_V5_4 | NETDATA_V5_14, + .load = EBPF_LOAD_LEGACY, .targets = NULL, .probe_links = NULL, .objects = NULL, .thread = NULL}, + { .thread_name = "oomkill", .config_name = "oomkill", .enabled = 0, .start_routine = ebpf_oomkill_thread, + .update_every = EBPF_DEFAULT_UPDATE_EVERY, .global_charts = 1, .apps_charts = NETDATA_EBPF_APPS_FLAG_NO, + .apps_level = NETDATA_APPS_LEVEL_REAL_PARENT, .cgroup_charts = CONFIG_BOOLEAN_NO, .mode = MODE_ENTRY, .optional = 0, + .apps_routine = ebpf_oomkill_create_apps_charts, .maps = NULL, + .pid_map_size = ND_EBPF_DEFAULT_PID_SIZE, .names = NULL, .cfg = &oomkill_config, + .config_file = NETDATA_OOMKILL_CONFIG_FILE, + .kernels = NETDATA_V4_14 | NETDATA_V4_16 | NETDATA_V4_18 | NETDATA_V5_4 | NETDATA_V5_14, + .load = EBPF_LOAD_LEGACY, .targets = NULL, .probe_links = NULL, .objects = NULL, .thread = NULL}, + { .thread_name = "shm", .config_name = "shm", .enabled = 0, .start_routine = ebpf_shm_thread, + .update_every = EBPF_DEFAULT_UPDATE_EVERY, .global_charts = 1, .apps_charts = NETDATA_EBPF_APPS_FLAG_NO, + .apps_level = NETDATA_APPS_LEVEL_REAL_PARENT, .cgroup_charts = CONFIG_BOOLEAN_NO, .mode = MODE_ENTRY, .optional = 0, + .apps_routine = ebpf_shm_create_apps_charts, .maps = NULL, + .pid_map_size = ND_EBPF_DEFAULT_PID_SIZE, .names = NULL, .cfg = &shm_config, + .config_file = NETDATA_DIRECTORY_SHM_CONFIG_FILE, + .kernels = NETDATA_V3_10 | NETDATA_V4_14 | NETDATA_V4_16 | NETDATA_V4_18 | NETDATA_V5_4 | NETDATA_V5_14, + .load = EBPF_LOAD_LEGACY, .targets = shm_targets, .probe_links = NULL, .objects = NULL, .thread = NULL}, + { .thread_name = "mdflush", .config_name = "mdflush", .enabled = 0, .start_routine = ebpf_mdflush_thread, + .update_every = EBPF_DEFAULT_UPDATE_EVERY, .global_charts = 1, .apps_charts = NETDATA_EBPF_APPS_FLAG_NO, + .apps_level = NETDATA_APPS_NOT_SET, .cgroup_charts = CONFIG_BOOLEAN_NO, .mode = MODE_ENTRY, .optional = 0, + .apps_routine = NULL, .maps = NULL, .pid_map_size = ND_EBPF_DEFAULT_PID_SIZE, .names = NULL, .cfg = &mdflush_config, + .config_file = NETDATA_DIRECTORY_MDFLUSH_CONFIG_FILE, + .kernels = NETDATA_V3_10 | NETDATA_V4_14 | NETDATA_V4_16 | NETDATA_V4_18 | NETDATA_V5_4 | NETDATA_V5_14, + .load = EBPF_LOAD_LEGACY, .targets = NULL, .probe_links = NULL, .objects = NULL, .thread = NULL}, + { .thread_name = NULL, .enabled = 0, .start_routine = NULL, .update_every = EBPF_DEFAULT_UPDATE_EVERY, + .global_charts = 0, .apps_charts = NETDATA_EBPF_APPS_FLAG_NO, .apps_level = NETDATA_APPS_NOT_SET, + .cgroup_charts = CONFIG_BOOLEAN_NO, .mode = MODE_ENTRY, .optional = 0, .apps_routine = NULL, .maps = NULL, + .pid_map_size = 0, .names = NULL, .cfg = NULL, .config_name = NULL, .kernels = 0, .load = EBPF_LOAD_LEGACY, + .targets = NULL, .probe_links = NULL, .objects = NULL, .thread = NULL}, +}; + +struct netdata_static_thread ebpf_threads[] = { + { + .name = "EBPF PROCESS", + .config_section = NULL, + .config_name = NULL, + .env_name = NULL, + .enabled = 1, + .thread = NULL, + .init_routine = NULL, + .start_routine = NULL + }, + { + .name = "EBPF SOCKET", + .config_section = NULL, + .config_name = NULL, + .env_name = NULL, + .enabled = 1, + .thread = NULL, + .init_routine = NULL, + .start_routine = NULL + }, + { + .name = "EBPF CACHESTAT", + .config_section = NULL, + .config_name = NULL, + .env_name = NULL, + .enabled = 1, + .thread = NULL, + .init_routine = NULL, + .start_routine = NULL + }, + { + .name = "EBPF SYNC", + .config_section = NULL, + .config_name = NULL, + .env_name = NULL, + .enabled = 1, + .thread = NULL, + .init_routine = NULL, + .start_routine = NULL + }, + { + .name = "EBPF DCSTAT", + .config_section = NULL, + .config_name = NULL, + .env_name = NULL, + .enabled = 1, + .thread = NULL, + .init_routine = NULL, + .start_routine = NULL + }, + { + .name = "EBPF SWAP", + .config_section = NULL, + .config_name = NULL, + .env_name = NULL, + .enabled = 1, + .thread = NULL, + .init_routine = NULL, + .start_routine = NULL + }, + { + .name = "EBPF VFS", + .config_section = NULL, + .config_name = NULL, + .env_name = NULL, + .enabled = 1, + .thread = NULL, + .init_routine = NULL, + .start_routine = NULL + }, + { + .name = "EBPF FILESYSTEM", + .config_section = NULL, + .config_name = NULL, + .env_name = NULL, + .enabled = 1, + .thread = NULL, + .init_routine = NULL, + .start_routine = NULL + }, + { + .name = "EBPF DISK", + .config_section = NULL, + .config_name = NULL, + .env_name = NULL, + .enabled = 1, + .thread = NULL, + .init_routine = NULL, + .start_routine = NULL + }, + { + .name = "EBPF MOUNT", + .config_section = NULL, + .config_name = NULL, + .env_name = NULL, + .enabled = 1, + .thread = NULL, + .init_routine = NULL, + .start_routine = NULL + }, + { + .name = "EBPF FD", + .config_section = NULL, + .config_name = NULL, + .env_name = NULL, + .enabled = 1, + .thread = NULL, + .init_routine = NULL, + .start_routine = NULL + }, + { + .name = "EBPF HARDIRQ", + .config_section = NULL, + .config_name = NULL, + .env_name = NULL, + .enabled = 1, + .thread = NULL, + .init_routine = NULL, + .start_routine = NULL + }, + { + .name = "EBPF SOFTIRQ", + .config_section = NULL, + .config_name = NULL, + .env_name = NULL, + .enabled = 1, + .thread = NULL, + .init_routine = NULL, + .start_routine = NULL + }, + { + .name = "EBPF OOMKILL", + .config_section = NULL, + .config_name = NULL, + .env_name = NULL, + .enabled = 1, + .thread = NULL, + .init_routine = NULL, + .start_routine = NULL + }, + { + .name = "EBPF SHM", + .config_section = NULL, + .config_name = NULL, + .env_name = NULL, + .enabled = 1, + .thread = NULL, + .init_routine = NULL, + .start_routine = NULL + }, + { + .name = "EBPF MDFLUSH", + .config_section = NULL, + .config_name = NULL, + .env_name = NULL, + .enabled = 1, + .thread = NULL, + .init_routine = NULL, + .start_routine = NULL + }, + { + .name = NULL, + .config_section = NULL, + .config_name = NULL, + .env_name = NULL, + .enabled = 0, + .thread = NULL, + .init_routine = NULL, + .start_routine = NULL + }, +}; + +ebpf_filesystem_partitions_t localfs[] = + {{.filesystem = "ext4", + .optional_filesystem = NULL, + .family = "ext4", + .objects = NULL, + .probe_links = NULL, + .flags = NETDATA_FILESYSTEM_FLAG_NO_PARTITION, + .enabled = CONFIG_BOOLEAN_YES, + .addresses = {.function = NULL, .addr = 0}, + .kernels = NETDATA_V3_10 | NETDATA_V4_14 | NETDATA_V4_16 | NETDATA_V4_18 | NETDATA_V5_4}, + {.filesystem = "xfs", + .optional_filesystem = NULL, + .family = "xfs", + .objects = NULL, + .probe_links = NULL, + .flags = NETDATA_FILESYSTEM_FLAG_NO_PARTITION, + .enabled = CONFIG_BOOLEAN_YES, + .addresses = {.function = NULL, .addr = 0}, + .kernels = NETDATA_V3_10 | NETDATA_V4_14 | NETDATA_V4_16 | NETDATA_V4_18 | NETDATA_V5_4}, + {.filesystem = "nfs", + .optional_filesystem = "nfs4", + .family = "nfs", + .objects = NULL, + .probe_links = NULL, + .flags = NETDATA_FILESYSTEM_ATTR_CHARTS, + .enabled = CONFIG_BOOLEAN_YES, + .addresses = {.function = NULL, .addr = 0}, + .kernels = NETDATA_V3_10 | NETDATA_V4_14 | NETDATA_V4_16 | NETDATA_V4_18 | NETDATA_V5_4}, + {.filesystem = "zfs", + .optional_filesystem = NULL, + .family = "zfs", + .objects = NULL, + .probe_links = NULL, + .flags = NETDATA_FILESYSTEM_FLAG_NO_PARTITION, + .enabled = CONFIG_BOOLEAN_YES, + .addresses = {.function = NULL, .addr = 0}, + .kernels = NETDATA_V3_10 | NETDATA_V4_14 | NETDATA_V4_16 | NETDATA_V4_18 | NETDATA_V5_4}, + {.filesystem = "btrfs", + .optional_filesystem = NULL, + .family = "btrfs", + .objects = NULL, + .probe_links = NULL, + .flags = NETDATA_FILESYSTEM_FILL_ADDRESS_TABLE, + .enabled = CONFIG_BOOLEAN_YES, + .addresses = {.function = "btrfs_file_operations", .addr = 0}, + .kernels = NETDATA_V3_10 | NETDATA_V4_14 | NETDATA_V4_16 | NETDATA_V4_18 | NETDATA_V5_4 | NETDATA_V5_10}, + {.filesystem = NULL, + .optional_filesystem = NULL, + .family = NULL, + .objects = NULL, + .probe_links = NULL, + .flags = NETDATA_FILESYSTEM_FLAG_NO_PARTITION, + .enabled = CONFIG_BOOLEAN_YES, + .addresses = {.function = NULL, .addr = 0}, + .kernels = 0}}; + +ebpf_sync_syscalls_t local_syscalls[] = { + {.syscall = NETDATA_SYSCALLS_SYNC, .enabled = CONFIG_BOOLEAN_YES, .objects = NULL, .probe_links = NULL, +#ifdef LIBBPF_MAJOR_VERSION + .sync_obj = NULL +#endif + }, + {.syscall = NETDATA_SYSCALLS_SYNCFS, .enabled = CONFIG_BOOLEAN_YES, .objects = NULL, .probe_links = NULL, +#ifdef LIBBPF_MAJOR_VERSION + .sync_obj = NULL +#endif + }, + {.syscall = NETDATA_SYSCALLS_MSYNC, .enabled = CONFIG_BOOLEAN_YES, .objects = NULL, .probe_links = NULL, +#ifdef LIBBPF_MAJOR_VERSION + .sync_obj = NULL +#endif + }, + {.syscall = NETDATA_SYSCALLS_FSYNC, .enabled = CONFIG_BOOLEAN_YES, .objects = NULL, .probe_links = NULL, +#ifdef LIBBPF_MAJOR_VERSION + .sync_obj = NULL +#endif + }, + {.syscall = NETDATA_SYSCALLS_FDATASYNC, .enabled = CONFIG_BOOLEAN_YES, .objects = NULL, .probe_links = NULL, +#ifdef LIBBPF_MAJOR_VERSION + .sync_obj = NULL +#endif + }, + {.syscall = NETDATA_SYSCALLS_SYNC_FILE_RANGE, .enabled = CONFIG_BOOLEAN_YES, .objects = NULL, .probe_links = NULL, +#ifdef LIBBPF_MAJOR_VERSION + .sync_obj = NULL +#endif + }, + {.syscall = NULL, .enabled = CONFIG_BOOLEAN_NO, .objects = NULL, .probe_links = NULL, +#ifdef LIBBPF_MAJOR_VERSION + .sync_obj = NULL +#endif + } +}; + + +// Link with apps.plugin +ebpf_process_stat_t *global_process_stat = NULL; + +// Link with cgroup.plugin +netdata_ebpf_cgroup_shm_t shm_ebpf_cgroup = {NULL, NULL}; +int shm_fd_ebpf_cgroup = -1; +sem_t *shm_sem_ebpf_cgroup = SEM_FAILED; +pthread_mutex_t mutex_cgroup_shm; + +//Network viewer +ebpf_network_viewer_options_t network_viewer_opt; + +// Statistic +ebpf_plugin_stats_t plugin_statistics = {.core = 0, .legacy = 0, .running = 0, .threads = 0, .tracepoints = 0, + .probes = 0, .retprobes = 0, .trampolines = 0}; + +#ifdef LIBBPF_MAJOR_VERSION +struct btf *default_btf = NULL; +#else +void *default_btf = NULL; +#endif +char *btf_path = NULL; + +/***************************************************************** + * + * FUNCTIONS USED TO CLEAN MEMORY AND OPERATE SYSTEM FILES + * + *****************************************************************/ + +/** + * Close the collector gracefully + */ +static void ebpf_exit() +{ +#ifdef LIBBPF_MAJOR_VERSION + pthread_mutex_lock(&ebpf_exit_cleanup); + if (default_btf) { + btf__free(default_btf); + default_btf = NULL; + } + pthread_mutex_unlock(&ebpf_exit_cleanup); +#endif + + char filename[FILENAME_MAX + 1]; + ebpf_pid_file(filename, FILENAME_MAX); + if (unlink(filename)) + error("Cannot remove PID file %s", filename); + + exit(0); +} + +/** + * Unload loegacy code + * + * @param objects objects loaded from eBPF programs + * @param probe_links links from loader + */ +static void ebpf_unload_legacy_code(struct bpf_object *objects, struct bpf_link **probe_links) +{ + if (!probe_links || !objects) + return; + + struct bpf_program *prog; + size_t j = 0 ; + bpf_object__for_each_program(prog, objects) { + bpf_link__destroy(probe_links[j]); + j++; + } + freez(probe_links); + if (objects) + bpf_object__close(objects); +} + +int ebpf_exit_plugin = 0; +/** + * Close the collector gracefully + * + * @param sig is the signal number used to close the collector + */ +static void ebpf_stop_threads(int sig) +{ + UNUSED(sig); + static int only_one = 0; + + int i; + // Child thread should be closed by itself. + pthread_mutex_lock(&ebpf_exit_cleanup); + if (main_thread_id != gettid() || only_one) { + pthread_mutex_unlock(&ebpf_exit_cleanup); + return; + } + only_one = 1; + for (i = 0; ebpf_threads[i].name != NULL; i++) { + if (ebpf_threads[i].enabled != NETDATA_THREAD_EBPF_STOPPED) + netdata_thread_cancel(*ebpf_threads[i].thread); + } + pthread_mutex_unlock(&ebpf_exit_cleanup); + + ebpf_exit_plugin = 1; + usec_t max = 3 * USEC_PER_SEC, step = 100000; + while (i && max) { + max -= step; + sleep_usec(step); + i = 0; + int j; + pthread_mutex_lock(&ebpf_exit_cleanup); + for (j = 0; ebpf_threads[j].name != NULL; j++) { + if (ebpf_threads[j].enabled != NETDATA_THREAD_EBPF_STOPPED) + i++; + } + pthread_mutex_unlock(&ebpf_exit_cleanup); + } + + //Unload threads(except sync and filesystem) + pthread_mutex_lock(&ebpf_exit_cleanup); + for (i = 0; ebpf_threads[i].name != NULL; i++) { + if (ebpf_threads[i].enabled == NETDATA_THREAD_EBPF_STOPPED && i != EBPF_MODULE_FILESYSTEM_IDX && + i != EBPF_MODULE_SYNC_IDX) + ebpf_unload_legacy_code(ebpf_modules[i].objects, ebpf_modules[i].probe_links); + } + pthread_mutex_unlock(&ebpf_exit_cleanup); + + //Unload filesystem + pthread_mutex_lock(&ebpf_exit_cleanup); + if (ebpf_threads[EBPF_MODULE_FILESYSTEM_IDX].enabled == NETDATA_THREAD_EBPF_STOPPED) { + for (i = 0; localfs[i].filesystem != NULL; i++) { + ebpf_unload_legacy_code(localfs[i].objects, localfs[i].probe_links); + } + } + pthread_mutex_unlock(&ebpf_exit_cleanup); + + //Unload Sync + pthread_mutex_lock(&ebpf_exit_cleanup); + if (ebpf_threads[EBPF_MODULE_SYNC_IDX].enabled == NETDATA_THREAD_EBPF_STOPPED) { + for (i = 0; local_syscalls[i].syscall != NULL; i++) { + ebpf_unload_legacy_code(local_syscalls[i].objects, local_syscalls[i].probe_links); + } + } + pthread_mutex_unlock(&ebpf_exit_cleanup); + + ebpf_exit(); +} + +/***************************************************************** + * + * FUNCTIONS TO CREATE CHARTS + * + *****************************************************************/ + +/** + * Get a value from a structure. + * + * @param basis it is the first address of the structure + * @param offset it is the offset of the data you want to access. + * @return + */ +collected_number get_value_from_structure(char *basis, size_t offset) +{ + collected_number *value = (collected_number *)(basis + offset); + + collected_number ret = (collected_number)llabs(*value); + // this reset is necessary to avoid keep a constant value while processing is not executing a task + *value = 0; + + return ret; +} + +/** + * Write begin command on standard output + * + * @param family the chart family name + * @param name the chart name + */ +void write_begin_chart(char *family, char *name) +{ + printf("BEGIN %s.%s\n", family, name); +} + +/** + * Write END command on stdout. + */ +inline void write_end_chart() +{ + printf("END\n"); +} + +/** + * Write set command on standard output + * + * @param dim the dimension name + * @param value the value for the dimension + */ +void write_chart_dimension(char *dim, long long value) +{ + printf("SET %s = %lld\n", dim, value); +} + +/** + * Call the necessary functions to create a chart. + * + * @param name the chart name + * @param family the chart family + * @param move the pointer with the values that will be published + * @param end the number of values that will be written on standard output + * + * @return It returns a variable that maps the charts that did not have zero values. + */ +void write_count_chart(char *name, char *family, netdata_publish_syscall_t *move, uint32_t end) +{ + write_begin_chart(family, name); + + uint32_t i = 0; + while (move && i < end) { + write_chart_dimension(move->name, move->ncall); + + move = move->next; + i++; + } + + write_end_chart(); +} + +/** + * Call the necessary functions to create a chart. + * + * @param name the chart name + * @param family the chart family + * @param move the pointer with the values that will be published + * @param end the number of values that will be written on standard output + */ +void write_err_chart(char *name, char *family, netdata_publish_syscall_t *move, int end) +{ + write_begin_chart(family, name); + + int i = 0; + while (move && i < end) { + write_chart_dimension(move->name, move->nerr); + + move = move->next; + i++; + } + + write_end_chart(); +} + +/** + * Write charts + * + * Write the current information to publish the charts. + * + * @param family chart family + * @param chart chart id + * @param dim dimension name + * @param v1 value. + */ +void ebpf_one_dimension_write_charts(char *family, char *chart, char *dim, long long v1) +{ + write_begin_chart(family, chart); + + write_chart_dimension(dim, v1); + + write_end_chart(); +} + +/** + * Call the necessary functions to create a chart. + * + * @param chart the chart name + * @param family the chart family + * @param dwrite the dimension name + * @param vwrite the value for previous dimension + * @param dread the dimension name + * @param vread the value for previous dimension + * + * @return It returns a variable that maps the charts that did not have zero values. + */ +void write_io_chart(char *chart, char *family, char *dwrite, long long vwrite, char *dread, long long vread) +{ + write_begin_chart(family, chart); + + write_chart_dimension(dwrite, vwrite); + write_chart_dimension(dread, vread); + + write_end_chart(); +} + +/** + * Write chart cmd on standard output + * + * @param type chart type + * @param id chart id + * @param title chart title + * @param units units label + * @param family group name used to attach the chart on dashboard + * @param charttype chart type + * @param context chart context + * @param order chart order + * @param update_every update interval used by plugin + * @param module chart module name, this is the eBPF thread. + */ +void ebpf_write_chart_cmd(char *type, char *id, char *title, char *units, char *family, + char *charttype, char *context, int order, int update_every, char *module) +{ + printf("CHART %s.%s '' '%s' '%s' '%s' '%s' '%s' %d %d '' 'ebpf.plugin' '%s'\n", + type, + id, + title, + units, + (family)?family:"", + (context)?context:"", + (charttype)?charttype:"", + order, + update_every, + module); +} + +/** + * Write chart cmd on standard output + * + * @param type chart type + * @param id chart id + * @param title chart title + * @param units units label + * @param family group name used to attach the chart on dashboard + * @param charttype chart type + * @param context chart context + * @param order chart order + * @param update_every value to overwrite the update frequency set by the server. + */ +void ebpf_write_chart_obsolete(char *type, char *id, char *title, char *units, char *family, + char *charttype, char *context, int order, int update_every) +{ + printf("CHART %s.%s '' '%s' '%s' '%s' '%s' '%s' %d %d 'obsolete'\n", + type, + id, + title, + units, + (family)?family:"", + (context)?context:"", + (charttype)?charttype:"", + order, + update_every); +} + +/** + * Write the dimension command on standard output + * + * @param name the dimension name + * @param id the dimension id + * @param algo the dimension algorithm + */ +void ebpf_write_global_dimension(char *name, char *id, char *algorithm) +{ + printf("DIMENSION %s %s %s 1 1\n", name, id, algorithm); +} + +/** + * Call ebpf_write_global_dimension to create the dimensions for a specific chart + * + * @param ptr a pointer to a structure of the type netdata_publish_syscall_t + * @param end the number of dimensions for the structure ptr + */ +void ebpf_create_global_dimension(void *ptr, int end) +{ + netdata_publish_syscall_t *move = ptr; + + int i = 0; + while (move && i < end) { + ebpf_write_global_dimension(move->name, move->dimension, move->algorithm); + + move = move->next; + i++; + } +} + +/** + * Call write_chart_cmd to create the charts + * + * @param type chart type + * @param id chart id + * @param title chart title + * @param units axis label + * @param family group name used to attach the chart on dashboard + * @param context chart context + * @param charttype chart type + * @param order order number of the specified chart + * @param ncd a pointer to a function called to create dimensions + * @param move a pointer for a structure that has the dimensions + * @param end number of dimensions for the chart created + * @param update_every update interval used with chart. + * @param module chart module name, this is the eBPF thread. + */ +void ebpf_create_chart(char *type, + char *id, + char *title, + char *units, + char *family, + char *context, + char *charttype, + int order, + void (*ncd)(void *, int), + void *move, + int end, + int update_every, + char *module) +{ + ebpf_write_chart_cmd(type, id, title, units, family, charttype, context, order, update_every, module); + + if (ncd) { + ncd(move, end); + } +} + +/** + * Create charts on apps submenu + * + * @param id the chart id + * @param title the value displayed on vertical axis. + * @param units the value displayed on vertical axis. + * @param family Submenu that the chart will be attached on dashboard. + * @param charttype chart type + * @param order the chart order + * @param algorithm the algorithm used by dimension + * @param root structure used to create the dimensions. + * @param update_every update interval used by plugin + * @param module chart module name, this is the eBPF thread. + */ +void ebpf_create_charts_on_apps(char *id, char *title, char *units, char *family, char *charttype, int order, + char *algorithm, struct target *root, int update_every, char *module) +{ + struct target *w; + ebpf_write_chart_cmd(NETDATA_APPS_FAMILY, id, title, units, family, charttype, NULL, order, + update_every, module); + + for (w = root; w; w = w->next) { + if (unlikely(w->exposed)) + fprintf(stdout, "DIMENSION %s '' %s 1 1\n", w->name, algorithm); + } +} + +/** + * Call the necessary functions to create a name. + * + * @param family family name + * @param name chart name + * @param hist0 histogram values + * @param dimensions dimension values. + * @param end number of bins that will be sent to Netdata. + * + * @return It returns a variable that maps the charts that did not have zero values. + */ +void write_histogram_chart(char *family, char *name, const netdata_idx_t *hist, char **dimensions, uint32_t end) +{ + write_begin_chart(family, name); + + uint32_t i; + for (i = 0; i < end; i++) { + write_chart_dimension(dimensions[i], (long long) hist[i]); + } + + write_end_chart(); + + fflush(stdout); +} + +/***************************************************************** + * + * FUNCTIONS TO DEFINE OPTIONS + * + *****************************************************************/ + +/** + * Define labels used to generate charts + * + * @param is structure with information about number of calls made for a function. + * @param pio structure used to generate charts. + * @param dim a pointer for the dimensions name + * @param name a pointer for the tensor with the name of the functions. + * @param algorithm a vector with the algorithms used to make the charts + * @param end the number of elements in the previous 4 arguments. + */ +void ebpf_global_labels(netdata_syscall_stat_t *is, netdata_publish_syscall_t *pio, char **dim, + char **name, int *algorithm, int end) +{ + int i; + + netdata_syscall_stat_t *prev = NULL; + netdata_publish_syscall_t *publish_prev = NULL; + for (i = 0; i < end; i++) { + if (prev) { + prev->next = &is[i]; + } + prev = &is[i]; + + pio[i].dimension = dim[i]; + pio[i].name = name[i]; + pio[i].algorithm = strdupz(ebpf_algorithms[algorithm[i]]); + if (publish_prev) { + publish_prev->next = &pio[i]; + } + publish_prev = &pio[i]; + } +} + +/** + * Define thread mode for all ebpf program. + * + * @param lmode the mode that will be used for them. + */ +static inline void ebpf_set_thread_mode(netdata_run_mode_t lmode) +{ + int i; + for (i = 0; ebpf_modules[i].thread_name; i++) { + ebpf_modules[i].mode = lmode; + } +} + +/** + * Enable specific charts selected by user. + * + * @param em the structure that will be changed + * @param disable_apps the status about the apps charts. + * @param disable_cgroup the status about the cgroups charts. + */ +static inline void ebpf_enable_specific_chart(struct ebpf_module *em, int disable_apps, int disable_cgroup) +{ + em->enabled = CONFIG_BOOLEAN_YES; + + // oomkill stores data inside apps submenu, so it always need to have apps_enabled for plugin to create + // its chart, without this comparison eBPF.plugin will try to store invalid data when apps is disabled. + if (!disable_apps || !strcmp(em->thread_name, "oomkill")) { + em->apps_charts = NETDATA_EBPF_APPS_FLAG_YES; + } + + if (!disable_cgroup) { + em->cgroup_charts = CONFIG_BOOLEAN_YES; + } + + em->global_charts = CONFIG_BOOLEAN_YES; +} + +/** + * Enable all charts + * + * @param apps what is the current status of apps + * @param cgroups what is the current status of cgroups + */ +static inline void ebpf_enable_all_charts(int apps, int cgroups) +{ + int i; + for (i = 0; ebpf_modules[i].thread_name; i++) { + ebpf_enable_specific_chart(&ebpf_modules[i], apps, cgroups); + } +} + +/** + * Disable all Global charts + * + * Disable charts + */ +static inline void disable_all_global_charts() +{ + int i; + for (i = 0; ebpf_modules[i].thread_name; i++) { + ebpf_modules[i].enabled = 0; + ebpf_modules[i].global_charts = 0; + } +} + + +/** + * Enable the specified chart group + * + * @param idx the index of ebpf_modules that I am enabling + * @param disable_apps should I keep apps charts? + */ +static inline void ebpf_enable_chart(int idx, int disable_apps, int disable_cgroup) +{ + int i; + for (i = 0; ebpf_modules[i].thread_name; i++) { + if (i == idx) { + ebpf_enable_specific_chart(&ebpf_modules[i], disable_apps, disable_cgroup); + break; + } + } +} + +/** + * Disable APPs + * + * Disable charts for apps loading only global charts. + */ +static inline void ebpf_disable_apps() +{ + int i; + for (i = 0; ebpf_modules[i].thread_name; i++) { + ebpf_modules[i].apps_charts = NETDATA_EBPF_APPS_FLAG_NO; + } +} + +/** + * Disable Cgroups + * + * Disable charts for apps loading only global charts. + */ +static inline void ebpf_disable_cgroups() +{ + int i; + for (i = 0; ebpf_modules[i].thread_name; i++) { + ebpf_modules[i].cgroup_charts = 0; + } +} + +/** + * Update Disabled Plugins + * + * This function calls ebpf_update_stats to update statistics for collector. + * + * @param em a pointer to `struct ebpf_module` + */ +void ebpf_update_disabled_plugin_stats(ebpf_module_t *em) +{ + pthread_mutex_lock(&lock); + ebpf_update_stats(&plugin_statistics, em); + pthread_mutex_unlock(&lock); +} + +/** + * Print help on standard error for user knows how to use the collector. + */ +void ebpf_print_help() +{ + const time_t t = time(NULL); + struct tm ct; + struct tm *test = localtime_r(&t, &ct); + int year; + if (test) + year = ct.tm_year; + else + year = 0; + + fprintf(stderr, + "\n" + " Netdata ebpf.plugin %s\n" + " Copyright (C) 2016-%d Costa Tsaousis <costa@tsaousis.gr>\n" + " Released under GNU General Public License v3 or later.\n" + " All rights reserved.\n" + "\n" + " This eBPF.plugin is a data collector plugin for netdata.\n" + "\n" + " This plugin only accepts long options with one or two dashes. The available command line options are:\n" + "\n" + " SECONDS Set the data collection frequency.\n" + "\n" + " [-]-help Show this help.\n" + "\n" + " [-]-version Show software version.\n" + "\n" + " [-]-global Disable charts per application and cgroup.\n" + "\n" + " [-]-all Enable all chart groups (global, apps, and cgroup), unless -g is also given.\n" + "\n" + " [-]-cachestat Enable charts related to process run time.\n" + "\n" + " [-]-dcstat Enable charts related to directory cache.\n" + "\n" + " [-]-disk Enable charts related to disk monitoring.\n" + "\n" + " [-]-filesystem Enable chart related to filesystem run time.\n" + "\n" + " [-]-hardirq Enable chart related to hard IRQ latency.\n" + "\n" + " [-]-mdflush Enable charts related to multi-device flush.\n" + "\n" + " [-]-mount Enable charts related to mount monitoring.\n" + "\n" + " [-]-net Enable network viewer charts.\n" + "\n" + " [-]-oomkill Enable chart related to OOM kill tracking.\n" + "\n" + " [-]-process Enable charts related to process run time.\n" + "\n" + " [-]-return Run the collector in return mode.\n" + "\n" + " [-]-shm Enable chart related to shared memory tracking.\n" + "\n" + " [-]-softirq Enable chart related to soft IRQ latency.\n" + "\n" + " [-]-sync Enable chart related to sync run time.\n" + "\n" + " [-]-swap Enable chart related to swap run time.\n" + "\n" + " [-]-vfs Enable chart related to vfs run time.\n" + "\n" + " [-]-legacy Load legacy eBPF programs.\n" + "\n" + " [-]-core Use CO-RE when available(Working in progress).\n" + "\n", + VERSION, + (year >= 116) ? year + 1900 : 2020); +} + +/***************************************************************** + * + * TRACEPOINT MANAGEMENT FUNCTIONS + * + *****************************************************************/ + +/** + * Enable a tracepoint. + * + * @return 0 on success, -1 on error. + */ +int ebpf_enable_tracepoint(ebpf_tracepoint_t *tp) +{ + int test = ebpf_is_tracepoint_enabled(tp->class, tp->event); + + // err? + if (test == -1) { + return -1; + } + // disabled? + else if (test == 0) { + // enable it then. + if (ebpf_enable_tracing_values(tp->class, tp->event)) { + return -1; + } + } + + // enabled now or already was. + tp->enabled = true; + + return 0; +} + +/** + * Disable a tracepoint if it's enabled. + * + * @return 0 on success, -1 on error. + */ +int ebpf_disable_tracepoint(ebpf_tracepoint_t *tp) +{ + int test = ebpf_is_tracepoint_enabled(tp->class, tp->event); + + // err? + if (test == -1) { + return -1; + } + // enabled? + else if (test == 1) { + // disable it then. + if (ebpf_disable_tracing_values(tp->class, tp->event)) { + return -1; + } + } + + // disable now or already was. + tp->enabled = false; + + return 0; +} + +/** + * Enable multiple tracepoints on a list of tracepoints which end when the + * class is NULL. + * + * @return the number of successful enables. + */ +uint32_t ebpf_enable_tracepoints(ebpf_tracepoint_t *tps) +{ + uint32_t cnt = 0; + for (int i = 0; tps[i].class != NULL; i++) { + if (ebpf_enable_tracepoint(&tps[i]) == -1) { + infoerr("failed to enable tracepoint %s:%s", + tps[i].class, tps[i].event); + } + else { + cnt += 1; + } + } + return cnt; +} + +/***************************************************************** + * + * AUXILIARY FUNCTIONS USED DURING INITIALIZATION + * + *****************************************************************/ + +/** + * Read Local Ports + * + * Parse /proc/net/{tcp,udp} and get the ports Linux is listening. + * + * @param filename the proc file to parse. + * @param proto is the magic number associated to the protocol file we are reading. + */ +static void read_local_ports(char *filename, uint8_t proto) +{ + procfile *ff = procfile_open(filename, " \t:", PROCFILE_FLAG_DEFAULT); + if (!ff) + return; + + ff = procfile_readall(ff); + if (!ff) + return; + + size_t lines = procfile_lines(ff), l; + netdata_passive_connection_t values = {.counter = 0, .tgid = 0, .pid = 0}; + for(l = 0; l < lines ;l++) { + size_t words = procfile_linewords(ff, l); + // This is header or end of file + if (unlikely(words < 14)) + continue; + + // https://elixir.bootlin.com/linux/v5.7.8/source/include/net/tcp_states.h + // 0A = TCP_LISTEN + if (strcmp("0A", procfile_lineword(ff, l, 5))) + continue; + + // Read local port + uint16_t port = (uint16_t)strtol(procfile_lineword(ff, l, 2), NULL, 16); + update_listen_table(htons(port), proto, &values); + } + + procfile_close(ff); +} + +/** + * Read Local addresseses + * + * Read the local address from the interfaces. + */ +static void read_local_addresses() +{ + struct ifaddrs *ifaddr, *ifa; + if (getifaddrs(&ifaddr) == -1) { + error("Cannot get the local IP addresses, it is no possible to do separation between inbound and outbound connections"); + return; + } + + char *notext = { "No text representation" }; + for (ifa = ifaddr; ifa != NULL; ifa = ifa->ifa_next) { + if (ifa->ifa_addr == NULL) + continue; + + if ((ifa->ifa_addr->sa_family != AF_INET) && (ifa->ifa_addr->sa_family != AF_INET6)) + continue; + + ebpf_network_viewer_ip_list_t *w = callocz(1, sizeof(ebpf_network_viewer_ip_list_t)); + + int family = ifa->ifa_addr->sa_family; + w->ver = (uint8_t) family; + char text[INET6_ADDRSTRLEN]; + if (family == AF_INET) { + struct sockaddr_in *in = (struct sockaddr_in*) ifa->ifa_addr; + + w->first.addr32[0] = in->sin_addr.s_addr; + w->last.addr32[0] = in->sin_addr.s_addr; + + if (inet_ntop(AF_INET, w->first.addr8, text, INET_ADDRSTRLEN)) { + w->value = strdupz(text); + w->hash = simple_hash(text); + } else { + w->value = strdupz(notext); + w->hash = simple_hash(notext); + } + } else { + struct sockaddr_in6 *in6 = (struct sockaddr_in6*) ifa->ifa_addr; + + memcpy(w->first.addr8, (void *)&in6->sin6_addr, sizeof(struct in6_addr)); + memcpy(w->last.addr8, (void *)&in6->sin6_addr, sizeof(struct in6_addr)); + + if (inet_ntop(AF_INET6, w->first.addr8, text, INET_ADDRSTRLEN)) { + w->value = strdupz(text); + w->hash = simple_hash(text); + } else { + w->value = strdupz(notext); + w->hash = simple_hash(notext); + } + } + + fill_ip_list((family == AF_INET)?&network_viewer_opt.ipv4_local_ip:&network_viewer_opt.ipv6_local_ip, + w, + "selector"); + } + + freeifaddrs(ifaddr); +} + +/** + * Start Pthread Variable + * + * This function starts all pthread variables. + * + * @return It returns 0 on success and -1. + */ +int ebpf_start_pthread_variables() +{ + pthread_mutex_init(&lock, NULL); + pthread_mutex_init(&ebpf_exit_cleanup, NULL); + pthread_mutex_init(&collect_data_mutex, NULL); + + if (pthread_cond_init(&collect_data_cond_var, NULL)) { + error("Cannot start conditional variable to control Apps charts."); + return -1; + } + + return 0; +} + +/** + * Am I collecting PIDs? + * + * Test if eBPF plugin needs to collect PID information. + * + * @return It returns 1 if at least one thread needs to collect the data, or zero otherwise. + */ +static inline uint32_t ebpf_am_i_collect_pids() +{ + uint32_t ret = 0; + int i; + for (i = 0; ebpf_modules[i].thread_name; i++) { + ret |= ebpf_modules[i].cgroup_charts | (ebpf_modules[i].apps_charts & NETDATA_EBPF_APPS_FLAG_YES); + } + + return ret; +} + +/** + * Allocate the vectors used for all threads. + */ +static void ebpf_allocate_common_vectors() +{ + if (unlikely(!ebpf_am_i_collect_pids())) { + return; + } + + all_pids = callocz((size_t)pid_max, sizeof(struct pid_stat *)); + global_process_stat = callocz((size_t)ebpf_nprocs, sizeof(ebpf_process_stat_t)); +} + +/** + * Define how to load the ebpf programs + * + * @param ptr the option given by users + */ +static inline void how_to_load(char *ptr) +{ + if (!strcasecmp(ptr, EBPF_CFG_LOAD_MODE_RETURN)) + ebpf_set_thread_mode(MODE_RETURN); + else if (!strcasecmp(ptr, EBPF_CFG_LOAD_MODE_DEFAULT)) + ebpf_set_thread_mode(MODE_ENTRY); + else + error("the option %s for \"ebpf load mode\" is not a valid option.", ptr); +} + +/** + * Update interval + * + * Update default interval with value from user + * + * @param update_every value to overwrite the update frequency set by the server. + */ +static void ebpf_update_interval(int update_every) +{ + int i; + int value = (int) appconfig_get_number(&collector_config, EBPF_GLOBAL_SECTION, EBPF_CFG_UPDATE_EVERY, + update_every); + for (i = 0; ebpf_modules[i].thread_name; i++) { + ebpf_modules[i].update_every = value; + } +} + +/** + * Update PID table size + * + * Update default size with value from user + */ +static void ebpf_update_table_size() +{ + int i; + uint32_t value = (uint32_t) appconfig_get_number(&collector_config, EBPF_GLOBAL_SECTION, + EBPF_CFG_PID_SIZE, ND_EBPF_DEFAULT_PID_SIZE); + for (i = 0; ebpf_modules[i].thread_name; i++) { + ebpf_modules[i].pid_map_size = value; + } +} + +/** + * Set Load mode + * + * @param origin specify the configuration file loaded + */ +static inline void ebpf_set_load_mode(netdata_ebpf_load_mode_t load, netdata_ebpf_load_mode_t origin) +{ + int i; + for (i = 0; ebpf_modules[i].thread_name; i++) { + ebpf_modules[i].load &= ~NETDATA_EBPF_LOAD_METHODS; + ebpf_modules[i].load |= load | origin ; + } +} + +/** + * Update mode + * + * @param str value read from configuration file. + * @param origin specify the configuration file loaded + */ +static inline void epbf_update_load_mode(char *str, netdata_ebpf_load_mode_t origin) +{ + netdata_ebpf_load_mode_t load = epbf_convert_string_to_load_mode(str); + + ebpf_set_load_mode(load, origin); +} + +/** + * Read collector values + * + * @param disable_apps variable to store information related to apps. + * @param disable_cgroups variable to store information related to cgroups. + * @param update_every value to overwrite the update frequency set by the server. + * @param origin specify the configuration file loaded + */ +static void read_collector_values(int *disable_apps, int *disable_cgroups, + int update_every, netdata_ebpf_load_mode_t origin) +{ + // Read global section + char *value; + if (appconfig_exists(&collector_config, EBPF_GLOBAL_SECTION, "load")) // Backward compatibility + value = appconfig_get(&collector_config, EBPF_GLOBAL_SECTION, "load", + EBPF_CFG_LOAD_MODE_DEFAULT); + else + value = appconfig_get(&collector_config, EBPF_GLOBAL_SECTION, EBPF_CFG_LOAD_MODE, + EBPF_CFG_LOAD_MODE_DEFAULT); + + how_to_load(value); + + btf_path = appconfig_get(&collector_config, EBPF_GLOBAL_SECTION, EBPF_CFG_PROGRAM_PATH, + EBPF_DEFAULT_BTF_PATH); + +#ifdef LIBBPF_MAJOR_VERSION + default_btf = ebpf_load_btf_file(btf_path, EBPF_DEFAULT_BTF_FILE); +#endif + + value = appconfig_get(&collector_config, EBPF_GLOBAL_SECTION, EBPF_CFG_TYPE_FORMAT, EBPF_CFG_DEFAULT_PROGRAM); + + epbf_update_load_mode(value, origin); + + ebpf_update_interval(update_every); + + ebpf_update_table_size(); + + // This is kept to keep compatibility + uint32_t enabled = appconfig_get_boolean(&collector_config, EBPF_GLOBAL_SECTION, "disable apps", + CONFIG_BOOLEAN_NO); + if (!enabled) { + // Apps is a positive sentence, so we need to invert the values to disable apps. + enabled = appconfig_get_boolean(&collector_config, EBPF_GLOBAL_SECTION, EBPF_CFG_APPLICATION, + CONFIG_BOOLEAN_YES); + enabled = (enabled == CONFIG_BOOLEAN_NO)?CONFIG_BOOLEAN_YES:CONFIG_BOOLEAN_NO; + } + *disable_apps = (int)enabled; + + // Cgroup is a positive sentence, so we need to invert the values to disable apps. + // We are using the same pattern for cgroup and apps + enabled = appconfig_get_boolean(&collector_config, EBPF_GLOBAL_SECTION, EBPF_CFG_CGROUP, CONFIG_BOOLEAN_NO); + *disable_cgroups = (enabled == CONFIG_BOOLEAN_NO)?CONFIG_BOOLEAN_YES:CONFIG_BOOLEAN_NO; + + // Read ebpf programs section + enabled = appconfig_get_boolean(&collector_config, EBPF_PROGRAMS_SECTION, + ebpf_modules[EBPF_MODULE_PROCESS_IDX].config_name, CONFIG_BOOLEAN_YES); + int started = 0; + if (enabled) { + ebpf_enable_chart(EBPF_MODULE_PROCESS_IDX, *disable_apps, *disable_cgroups); + started++; + } + + // This is kept to keep compatibility + enabled = appconfig_get_boolean(&collector_config, EBPF_PROGRAMS_SECTION, "network viewer", + CONFIG_BOOLEAN_NO); + if (!enabled) + enabled = appconfig_get_boolean(&collector_config, EBPF_PROGRAMS_SECTION, + ebpf_modules[EBPF_MODULE_SOCKET_IDX].config_name, + CONFIG_BOOLEAN_NO); + + if (enabled) { + ebpf_enable_chart(EBPF_MODULE_SOCKET_IDX, *disable_apps, *disable_cgroups); + // Read network viewer section if network viewer is enabled + // This is kept here to keep backward compatibility + parse_network_viewer_section(&collector_config); + parse_service_name_section(&collector_config); + started++; + } + + // This is kept to keep compatibility + enabled = appconfig_get_boolean(&collector_config, EBPF_PROGRAMS_SECTION, "network connection monitoring", + CONFIG_BOOLEAN_NO); + if (!enabled) + enabled = appconfig_get_boolean(&collector_config, EBPF_PROGRAMS_SECTION, "network connections", + CONFIG_BOOLEAN_NO); + ebpf_modules[EBPF_MODULE_SOCKET_IDX].optional = (int)enabled; + + enabled = appconfig_get_boolean(&collector_config, EBPF_PROGRAMS_SECTION, "cachestat", + CONFIG_BOOLEAN_NO); + + if (enabled) { + ebpf_enable_chart(EBPF_MODULE_CACHESTAT_IDX, *disable_apps, *disable_cgroups); + started++; + } + + enabled = appconfig_get_boolean(&collector_config, EBPF_PROGRAMS_SECTION, "sync", + CONFIG_BOOLEAN_YES); + + if (enabled) { + ebpf_enable_chart(EBPF_MODULE_SYNC_IDX, *disable_apps, *disable_cgroups); + started++; + } + + enabled = appconfig_get_boolean(&collector_config, EBPF_PROGRAMS_SECTION, "dcstat", + CONFIG_BOOLEAN_NO); + if (enabled) { + ebpf_enable_chart(EBPF_MODULE_DCSTAT_IDX, *disable_apps, *disable_cgroups); + started++; + } + + enabled = appconfig_get_boolean(&collector_config, EBPF_PROGRAMS_SECTION, "swap", + CONFIG_BOOLEAN_NO); + if (enabled) { + ebpf_enable_chart(EBPF_MODULE_SWAP_IDX, *disable_apps, *disable_cgroups); + started++; + } + + enabled = appconfig_get_boolean(&collector_config, EBPF_PROGRAMS_SECTION, "vfs", + CONFIG_BOOLEAN_NO); + if (enabled) { + ebpf_enable_chart(EBPF_MODULE_VFS_IDX, *disable_apps, *disable_cgroups); + started++; + } + + enabled = appconfig_get_boolean(&collector_config, EBPF_PROGRAMS_SECTION, "filesystem", + CONFIG_BOOLEAN_NO); + if (enabled) { + ebpf_enable_chart(EBPF_MODULE_FILESYSTEM_IDX, *disable_apps, *disable_cgroups); + started++; + } + + enabled = appconfig_get_boolean(&collector_config, EBPF_PROGRAMS_SECTION, "disk", + CONFIG_BOOLEAN_NO); + if (enabled) { + ebpf_enable_chart(EBPF_MODULE_DISK_IDX, *disable_apps, *disable_cgroups); + started++; + } + + enabled = appconfig_get_boolean(&collector_config, EBPF_PROGRAMS_SECTION, "mount", + CONFIG_BOOLEAN_YES); + if (enabled) { + ebpf_enable_chart(EBPF_MODULE_MOUNT_IDX, *disable_apps, *disable_cgroups); + started++; + } + + enabled = appconfig_get_boolean(&collector_config, EBPF_PROGRAMS_SECTION, "fd", + CONFIG_BOOLEAN_YES); + if (enabled) { + ebpf_enable_chart(EBPF_MODULE_FD_IDX, *disable_apps, *disable_cgroups); + started++; + } + + enabled = appconfig_get_boolean(&collector_config, EBPF_PROGRAMS_SECTION, "hardirq", + CONFIG_BOOLEAN_YES); + if (enabled) { + ebpf_enable_chart(EBPF_MODULE_HARDIRQ_IDX, *disable_apps, *disable_cgroups); + started++; + } + + enabled = appconfig_get_boolean(&collector_config, EBPF_PROGRAMS_SECTION, "softirq", + CONFIG_BOOLEAN_YES); + if (enabled) { + ebpf_enable_chart(EBPF_MODULE_SOFTIRQ_IDX, *disable_apps, *disable_cgroups); + started++; + } + + enabled = appconfig_get_boolean(&collector_config, EBPF_PROGRAMS_SECTION, "oomkill", + CONFIG_BOOLEAN_YES); + if (enabled) { + ebpf_enable_chart(EBPF_MODULE_OOMKILL_IDX, *disable_apps, *disable_cgroups); + started++; + } + + enabled = appconfig_get_boolean(&collector_config, EBPF_PROGRAMS_SECTION, "shm", + CONFIG_BOOLEAN_YES); + if (enabled) { + ebpf_enable_chart(EBPF_MODULE_SHM_IDX, *disable_apps, *disable_cgroups); + started++; + } + + enabled = appconfig_get_boolean(&collector_config, EBPF_PROGRAMS_SECTION, "mdflush", + CONFIG_BOOLEAN_NO); + if (enabled) { + ebpf_enable_chart(EBPF_MODULE_MDFLUSH_IDX, *disable_apps, *disable_cgroups); + started++; + } + + if (!started){ + ebpf_enable_all_charts(*disable_apps, *disable_cgroups); + // Read network viewer section + // This is kept here to keep backward compatibility + parse_network_viewer_section(&collector_config); + parse_service_name_section(&collector_config); + } +} + +/** + * Load collector config + * + * @param path the path where the file ebpf.conf is stored. + * @param disable_apps variable to store the information about apps plugin status. + * @param disable_cgroups variable to store the information about cgroups plugin status. + * @param update_every value to overwrite the update frequency set by the server. + * + * @return 0 on success and -1 otherwise. + */ +static int load_collector_config(char *path, int *disable_apps, int *disable_cgroups, int update_every) +{ + char lpath[4096]; + netdata_ebpf_load_mode_t origin; + + snprintf(lpath, 4095, "%s/%s", path, NETDATA_EBPF_CONFIG_FILE); + if (!appconfig_load(&collector_config, lpath, 0, NULL)) { + snprintf(lpath, 4095, "%s/%s", path, NETDATA_EBPF_OLD_CONFIG_FILE); + if (!appconfig_load(&collector_config, lpath, 0, NULL)) { + return -1; + } + origin = EBPF_LOADED_FROM_STOCK; + } else + origin = EBPF_LOADED_FROM_USER; + + read_collector_values(disable_apps, disable_cgroups, update_every, origin); + + return 0; +} + +/** + * Set global variables reading environment variables + */ +void set_global_variables() +{ + // Get environment variables + ebpf_plugin_dir = getenv("NETDATA_PLUGINS_DIR"); + if (!ebpf_plugin_dir) + ebpf_plugin_dir = PLUGINS_DIR; + + ebpf_user_config_dir = getenv("NETDATA_USER_CONFIG_DIR"); + if (!ebpf_user_config_dir) + ebpf_user_config_dir = CONFIG_DIR; + + ebpf_stock_config_dir = getenv("NETDATA_STOCK_CONFIG_DIR"); + if (!ebpf_stock_config_dir) + ebpf_stock_config_dir = LIBCONFIG_DIR; + + ebpf_configured_log_dir = getenv("NETDATA_LOG_DIR"); + if (!ebpf_configured_log_dir) + ebpf_configured_log_dir = LOG_DIR; + + ebpf_nprocs = (int)sysconf(_SC_NPROCESSORS_ONLN); + if (ebpf_nprocs > NETDATA_MAX_PROCESSOR) { + ebpf_nprocs = NETDATA_MAX_PROCESSOR; + } + + isrh = get_redhat_release(); + pid_max = get_system_pid_max(); + running_on_kernel = ebpf_get_kernel_version(); +} + +/** + * Load collector config + */ +static inline void ebpf_load_thread_config() +{ + int i; + for (i = 0; ebpf_modules[i].thread_name; i++) { + ebpf_update_module(&ebpf_modules[i], default_btf, running_on_kernel, isrh); + } +} + +/** + * Parse arguments given from user. + * + * @param argc the number of arguments + * @param argv the pointer to the arguments + */ +static void ebpf_parse_args(int argc, char **argv) +{ + int disable_apps = 0; + int disable_cgroups = 1; + int freq = 0; + int option_index = 0; + uint64_t select_threads = 0; + static struct option long_options[] = { + {"process", no_argument, 0, 0 }, + {"net", no_argument, 0, 0 }, + {"cachestat", no_argument, 0, 0 }, + {"sync", no_argument, 0, 0 }, + {"dcstat", no_argument, 0, 0 }, + {"swap", no_argument, 0, 0 }, + {"vfs", no_argument, 0, 0 }, + {"filesystem", no_argument, 0, 0 }, + {"disk", no_argument, 0, 0 }, + {"mount", no_argument, 0, 0 }, + {"filedescriptor", no_argument, 0, 0 }, + {"hardirq", no_argument, 0, 0 }, + {"softirq", no_argument, 0, 0 }, + {"oomkill", no_argument, 0, 0 }, + {"shm", no_argument, 0, 0 }, + {"mdflush", no_argument, 0, 0 }, + /* INSERT NEW THREADS BEFORE THIS COMMENT TO KEEP COMPATIBILITY WITH enum ebpf_module_indexes */ + {"all", no_argument, 0, 0 }, + {"version", no_argument, 0, 0 }, + {"help", no_argument, 0, 0 }, + {"global", no_argument, 0, 0 }, + {"return", no_argument, 0, 0 }, + {"legacy", no_argument, 0, 0 }, + {"core", no_argument, 0, 0 }, + {0, 0, 0, 0} + }; + + memset(&network_viewer_opt, 0, sizeof(network_viewer_opt)); + network_viewer_opt.max_dim = NETDATA_NV_CAP_VALUE; + + if (argc > 1) { + int n = (int)str2l(argv[1]); + if (n > 0) { + freq = n; + } + } + + if (!freq) + freq = EBPF_DEFAULT_UPDATE_EVERY; + + if (load_collector_config(ebpf_user_config_dir, &disable_apps, &disable_cgroups, freq)) { + info( + "Does not have a configuration file inside `%s/ebpf.d.conf. It will try to load stock file.", + ebpf_user_config_dir); + if (load_collector_config(ebpf_stock_config_dir, &disable_apps, &disable_cgroups, freq)) { + info("Does not have a stock file. It is starting with default options."); + } + } + + ebpf_load_thread_config(); + + while (1) { + int c = getopt_long_only(argc, argv, "", long_options, &option_index); + if (c == -1) + break; + + switch (option_index) { + case EBPF_MODULE_PROCESS_IDX: { + select_threads |= 1<<EBPF_MODULE_PROCESS_IDX; +#ifdef NETDATA_INTERNAL_CHECKS + info("EBPF enabling \"PROCESS\" charts, because it was started with the option \"[-]-process\"."); +#endif + break; + } + case EBPF_MODULE_SOCKET_IDX: { + select_threads |= 1<<EBPF_MODULE_SOCKET_IDX; +#ifdef NETDATA_INTERNAL_CHECKS + info("EBPF enabling \"NET\" charts, because it was started with the option \"[-]-net\"."); +#endif + break; + } + case EBPF_MODULE_CACHESTAT_IDX: { + select_threads |= 1<<EBPF_MODULE_CACHESTAT_IDX; +#ifdef NETDATA_INTERNAL_CHECKS + info("EBPF enabling \"CACHESTAT\" charts, because it was started with the option \"[-]-cachestat\"."); +#endif + break; + } + case EBPF_MODULE_SYNC_IDX: { + select_threads |= 1<<EBPF_MODULE_SYNC_IDX; +#ifdef NETDATA_INTERNAL_CHECKS + info("EBPF enabling \"SYNC\" chart, because it was started with the option \"[-]-sync\"."); +#endif + break; + } + case EBPF_MODULE_DCSTAT_IDX: { + select_threads |= 1<<EBPF_MODULE_DCSTAT_IDX; +#ifdef NETDATA_INTERNAL_CHECKS + info("EBPF enabling \"DCSTAT\" charts, because it was started with the option \"[-]-dcstat\"."); +#endif + break; + } + case EBPF_MODULE_SWAP_IDX: { + select_threads |= 1<<EBPF_MODULE_SWAP_IDX; +#ifdef NETDATA_INTERNAL_CHECKS + info("EBPF enabling \"SWAP\" chart, because it was started with the option \"[-]-swap\"."); +#endif + break; + } + case EBPF_MODULE_VFS_IDX: { + select_threads |= 1<<EBPF_MODULE_VFS_IDX; +#ifdef NETDATA_INTERNAL_CHECKS + info("EBPF enabling \"VFS\" chart, because it was started with the option \"[-]-vfs\"."); +#endif + break; + } + case EBPF_MODULE_FILESYSTEM_IDX: { + select_threads |= 1<<EBPF_MODULE_FILESYSTEM_IDX; +#ifdef NETDATA_INTERNAL_CHECKS + info("EBPF enabling \"FILESYSTEM\" chart, because it was started with the option \"[-]-filesystem\"."); +#endif + break; + } + case EBPF_MODULE_DISK_IDX: { + select_threads |= 1<<EBPF_MODULE_DISK_IDX; +#ifdef NETDATA_INTERNAL_CHECKS + info("EBPF enabling \"DISK\" chart, because it was started with the option \"[-]-disk\"."); +#endif + break; + } + case EBPF_MODULE_MOUNT_IDX: { + select_threads |= 1<<EBPF_MODULE_MOUNT_IDX; +#ifdef NETDATA_INTERNAL_CHECKS + info("EBPF enabling \"MOUNT\" chart, because it was started with the option \"[-]-mount\"."); +#endif + break; + } + case EBPF_MODULE_FD_IDX: { + select_threads |= 1<<EBPF_MODULE_FD_IDX; +#ifdef NETDATA_INTERNAL_CHECKS + info("EBPF enabling \"FILEDESCRIPTOR\" chart, because it was started with the option \"[-]-filedescriptor\"."); +#endif + break; + } + case EBPF_MODULE_HARDIRQ_IDX: { + select_threads |= 1<<EBPF_MODULE_HARDIRQ_IDX; +#ifdef NETDATA_INTERNAL_CHECKS + info("EBPF enabling \"HARDIRQ\" chart, because it was started with the option \"[-]-hardirq\"."); +#endif + break; + } + case EBPF_MODULE_SOFTIRQ_IDX: { + select_threads |= 1<<EBPF_MODULE_SOFTIRQ_IDX; +#ifdef NETDATA_INTERNAL_CHECKS + info("EBPF enabling \"SOFTIRQ\" chart, because it was started with the option \"[-]-softirq\"."); +#endif + break; + } + case EBPF_MODULE_OOMKILL_IDX: { + select_threads |= 1<<EBPF_MODULE_OOMKILL_IDX; +#ifdef NETDATA_INTERNAL_CHECKS + info("EBPF enabling \"OOMKILL\" chart, because it was started with the option \"[-]-oomkill\"."); +#endif + break; + } + case EBPF_MODULE_SHM_IDX: { + select_threads |= 1<<EBPF_MODULE_SHM_IDX; +#ifdef NETDATA_INTERNAL_CHECKS + info("EBPF enabling \"SHM\" chart, because it was started with the option \"[-]-shm\"."); +#endif + break; + } + case EBPF_MODULE_MDFLUSH_IDX: { + select_threads |= 1<<EBPF_MODULE_MDFLUSH_IDX; +#ifdef NETDATA_INTERNAL_CHECKS + info("EBPF enabling \"MDFLUSH\" chart, because it was started with the option \"[-]-mdflush\"."); +#endif + break; + } + case EBPF_OPTION_ALL_CHARTS: { + disable_apps = 0; + disable_cgroups = 0; +#ifdef NETDATA_INTERNAL_CHECKS + info("EBPF running with all chart groups, because it was started with the option \"[-]-all\"."); +#endif + break; + } + case EBPF_OPTION_VERSION: { + printf("ebpf.plugin %s\n", VERSION); + exit(0); + } + case EBPF_OPTION_HELP: { + ebpf_print_help(); + exit(0); + } + case EBPF_OPTION_GLOBAL_CHART: { + disable_apps = 1; + disable_cgroups = 1; +#ifdef NETDATA_INTERNAL_CHECKS + info("EBPF running with global chart group, because it was started with the option \"[-]-global\"."); +#endif + break; + } + case EBPF_OPTION_RETURN_MODE: { + ebpf_set_thread_mode(MODE_RETURN); +#ifdef NETDATA_INTERNAL_CHECKS + info("EBPF running in \"RETURN\" mode, because it was started with the option \"[-]-return\"."); +#endif + break; + } + case EBPF_OPTION_LEGACY: { + ebpf_set_load_mode(EBPF_LOAD_LEGACY, EBPF_LOADED_FROM_USER); +#ifdef NETDATA_INTERNAL_CHECKS + info("EBPF running with \"LEGACY\" code, because it was started with the option \"[-]-legacy\"."); +#endif + break; + } + case EBPF_OPTION_CORE: { + ebpf_set_load_mode(EBPF_LOAD_CORE, EBPF_LOADED_FROM_USER); +#ifdef NETDATA_INTERNAL_CHECKS + info("EBPF running with \"CO-RE\" code, because it was started with the option \"[-]-core\"."); +#endif + break; + } + default: { + break; + } + } + } + + if (disable_apps || disable_cgroups) { + if (disable_apps) + ebpf_disable_apps(); + + if (disable_cgroups) + ebpf_disable_cgroups(); + } + + if (select_threads) { + disable_all_global_charts(); + uint64_t idx; + for (idx = 0; idx < EBPF_OPTION_ALL_CHARTS; idx++) { + if (select_threads & 1<<idx) + ebpf_enable_specific_chart(&ebpf_modules[idx], disable_apps, disable_cgroups); + } + } + + // Load apps_groups.conf + if (ebpf_read_apps_groups_conf( + &apps_groups_default_target, &apps_groups_root_target, ebpf_user_config_dir, "groups")) { + info("Cannot read process groups configuration file '%s/apps_groups.conf'. Will try '%s/apps_groups.conf'", + ebpf_user_config_dir, ebpf_stock_config_dir); + if (ebpf_read_apps_groups_conf( + &apps_groups_default_target, &apps_groups_root_target, ebpf_stock_config_dir, "groups")) { + error("Cannot read process groups '%s/apps_groups.conf'. There are no internal defaults. Failing.", + ebpf_stock_config_dir); + ebpf_exit(); + } + } else + info("Loaded config file '%s/apps_groups.conf'", ebpf_user_config_dir); +} + +/***************************************************************** + * + * COLLECTOR ENTRY POINT + * + *****************************************************************/ + +/** + * Update PID file + * + * Update the content of PID file + * + * @param filename is the full name of the file. + * @param pid that identifies the process + */ +static void ebpf_update_pid_file(char *filename, pid_t pid) +{ + FILE *fp = fopen(filename, "w"); + if (!fp) + return; + + fprintf(fp, "%d", pid); + fclose(fp); +} + +/** + * Get Process Name + * + * Get process name from /proc/PID/status + * + * @param pid that identifies the process + */ +static char *ebpf_get_process_name(pid_t pid) +{ + char *name = NULL; + char filename[FILENAME_MAX + 1]; + snprintfz(filename, FILENAME_MAX, "/proc/%d/status", pid); + + procfile *ff = procfile_open(filename, " \t", PROCFILE_FLAG_DEFAULT); + if(unlikely(!ff)) { + error("Cannot open %s", filename); + return name; + } + + ff = procfile_readall(ff); + if(unlikely(!ff)) + return name; + + unsigned long i, lines = procfile_lines(ff); + for(i = 0; i < lines ; i++) { + char *cmp = procfile_lineword(ff, i, 0); + if (!strcmp(cmp, "Name:")) { + name = strdupz(procfile_lineword(ff, i, 1)); + break; + } + } + + procfile_close(ff); + + return name; +} + +/** + * Read Previous PID + * + * @param filename is the full name of the file. + * + * @return It returns the PID used during previous execution on success or 0 otherwise + */ +static pid_t ebpf_read_previous_pid(char *filename) +{ + FILE *fp = fopen(filename, "r"); + if (!fp) + return 0; + + char buffer[64]; + size_t length = fread(buffer, sizeof(*buffer), 63, fp); + pid_t old_pid = 0; + if (length) { + if (length > 63) + length = 63; + + buffer[length] = '\0'; + old_pid = (pid_t)str2uint32_t(buffer); + } + fclose(fp); + + return old_pid; +} + +/** + * Kill previous process + * + * Kill previous process whether it was not closed. + * + * @param filename is the full name of the file. + * @param pid that identifies the process + */ +static void ebpf_kill_previous_process(char *filename, pid_t pid) +{ + pid_t old_pid = ebpf_read_previous_pid(filename); + if (!old_pid) + return; + + // Process is not running + char *prev_name = ebpf_get_process_name(old_pid); + if (!prev_name) + return; + + char *current_name = ebpf_get_process_name(pid); + + if (!strcmp(prev_name, current_name)) + kill(old_pid, SIGKILL); + + freez(prev_name); + freez(current_name); + + // wait few microseconds before start new plugin + sleep_usec(USEC_PER_MS * 300); +} + +/** + * PID file + * + * Write the filename for PID inside the given vector. + * + * @param filename vector where we will store the name. + * @param length number of bytes available in filename vector + */ +void ebpf_pid_file(char *filename, size_t length) +{ + snprintfz(filename, length, "%s%s/ebpf.d/ebpf.pid", netdata_configured_host_prefix, ebpf_plugin_dir); +} + +/** + * Manage PID + * + * This function kills another instance of eBPF whether it is necessary and update the file content. + * + * @param pid that identifies the process + */ +static void ebpf_manage_pid(pid_t pid) +{ + char filename[FILENAME_MAX + 1]; + ebpf_pid_file(filename, FILENAME_MAX); + + ebpf_kill_previous_process(filename, pid); + ebpf_update_pid_file(filename, pid); +} + +/** + * Set start routine + * + * Set static routine before threads to be created. + */ + static void ebpf_set_static_routine() + { + int i; + for (i = 0; ebpf_modules[i].thread_name; i++) { + ebpf_threads[i].start_routine = ebpf_modules[i].start_routine; + } + } + +/** + * Entry point + * + * @param argc the number of arguments + * @param argv the pointer to the arguments + * + * @return it returns 0 on success and another integer otherwise + */ +int main(int argc, char **argv) +{ + clocks_init(); + main_thread_id = gettid(); + + set_global_variables(); + ebpf_parse_args(argc, argv); + ebpf_manage_pid(getpid()); + + if (!has_condition_to_run(running_on_kernel)) { + error("The current collector cannot run on this kernel."); + return 2; + } + + if (!am_i_running_as_root()) { + error( + "ebpf.plugin should either run as root (now running with uid %u, euid %u) or have special capabilities..", + (unsigned int)getuid(), (unsigned int)geteuid()); + return 3; + } + + // set name + program_name = "ebpf.plugin"; + + // disable syslog + error_log_syslog = 0; + + // set errors flood protection to 100 logs per hour + error_log_errors_per_period = 100; + error_log_throttle_period = 3600; + + struct rlimit r = { RLIM_INFINITY, RLIM_INFINITY }; + if (setrlimit(RLIMIT_MEMLOCK, &r)) { + error("Setrlimit(RLIMIT_MEMLOCK)"); + return 4; + } + + signal(SIGINT, ebpf_stop_threads); + signal(SIGQUIT, ebpf_stop_threads); + signal(SIGTERM, ebpf_stop_threads); + signal(SIGPIPE, ebpf_stop_threads); + + if (ebpf_start_pthread_variables()) { + error("Cannot start mutex to control overall charts."); + ebpf_exit(); + } + + netdata_configured_host_prefix = getenv("NETDATA_HOST_PREFIX"); + if(verify_netdata_host_prefix() == -1) ebpf_exit(6); + + ebpf_allocate_common_vectors(); + +#ifdef LIBBPF_MAJOR_VERSION + libbpf_set_strict_mode(LIBBPF_STRICT_ALL); +#endif + + read_local_addresses(); + read_local_ports("/proc/net/tcp", IPPROTO_TCP); + read_local_ports("/proc/net/tcp6", IPPROTO_TCP); + read_local_ports("/proc/net/udp", IPPROTO_UDP); + read_local_ports("/proc/net/udp6", IPPROTO_UDP); + + ebpf_set_static_routine(); + + int i; + for (i = 0; ebpf_threads[i].name != NULL; i++) { + struct netdata_static_thread *st = &ebpf_threads[i]; + + ebpf_module_t *em = &ebpf_modules[i]; + em->thread = st; + // We always initialize process, because it is responsible to take care of apps integration + if (em->enabled || !i) { + st->thread = mallocz(sizeof(netdata_thread_t)); + em->thread_id = i; + st->enabled = NETDATA_THREAD_EBPF_RUNNING; + netdata_thread_create(st->thread, st->name, NETDATA_THREAD_OPTION_DEFAULT, st->start_routine, em); + } else { + st->enabled = NETDATA_THREAD_EBPF_STOPPED; + } + } + + usec_t step = EBPF_DEFAULT_UPDATE_EVERY * USEC_PER_SEC; + heartbeat_t hb; + heartbeat_init(&hb); + //Plugin will be killed when it receives a signal + while (!ebpf_exit_plugin) { + (void)heartbeat_next(&hb, step); + } + + return 0; +} diff --git a/collectors/ebpf.plugin/ebpf.d.conf b/collectors/ebpf.plugin/ebpf.d.conf new file mode 100644 index 0000000..cf5c740 --- /dev/null +++ b/collectors/ebpf.plugin/ebpf.d.conf @@ -0,0 +1,69 @@ +# +# Global options +# +# The `ebpf load mode` option accepts the following values : +# `entry` : The eBPF collector only monitors calls for the functions, and does not show charts related to errors. +# `return : In the `return` mode, the eBPF collector monitors the same kernel functions as `entry`, but also creates +# new charts for the return of these functions, such as errors. +# +# The eBPF collector also creates charts for each running application through an integration with the `apps.plugin` +# or `cgroups.plugin`. +# If you want to disable the integration with `apps.plugin` or `cgroups.plugin` along with the above charts, change the setting +# `apps` and `cgroups` to 'no'. +# +# The `update every` option defines the number of seconds used to read data from kernel and send to netdata +# +# The `pid table size` defines the maximum number of PIDs stored in the application hash tables. +# +[global] + ebpf load mode = entry + apps = yes + cgroups = no + update every = 5 + pid table size = 32768 + btf path = /sys/kernel/btf/ + +# +# eBPF Programs +# +# The eBPF collector has the following eBPF programs: +# +# `cachestat` : Make charts for kernel functions related to page cache. +# `dcstat` : Make charts for kernel functions related to directory cache. +# `disk` : Monitor I/O latencies for disks +# `fd` : This eBPF program creates charts that show information about file manipulation. +# `filesystem`: Monitor calls for functions used to manipulate specific filesystems +# `hardirq` : Monitor latency of serving hardware interrupt requests (hard IRQs). +# `mdflush` : Monitors flush counts for multi-devices. +# `mount` : Monitor calls for syscalls mount and umount +# `oomkill` : This eBPF program creates a chart that shows which process got OOM killed and when. +# `process` : This eBPF program creates charts that show information about process life. +# `shm` : Monitor calls for syscalls shmget, shmat, shmdt and shmctl. +# `socket` : This eBPF program creates charts with information about `TCP` and `UDP` functions, including the +# bandwidth consumed by each. +# `softirq` : Monitor latency of serving software interrupt requests (soft IRQs). +# `sync` : Monitor calls for syscall sync(2). +# `swap` : Monitor calls for internal swap functions. +# `vfs` : This eBPF program creates charts that show information about process VFS IO, VFS file manipulation and +# files removed. +# +# When plugin detects that system has support to BTF, it enables integration with apps.plugin. +# +[ebpf programs] + cachestat = no + dcstat = no + disk = no + fd = yes + filesystem = no + hardirq = yes + mdflush = no + mount = yes + oomkill = yes + process = yes + shm = no + socket = yes + softirq = yes + sync = yes + swap = no + vfs = yes + network connections = no diff --git a/collectors/ebpf.plugin/ebpf.d/cachestat.conf b/collectors/ebpf.plugin/ebpf.d/cachestat.conf new file mode 100644 index 0000000..52466be --- /dev/null +++ b/collectors/ebpf.plugin/ebpf.d/cachestat.conf @@ -0,0 +1,36 @@ +# The `ebpf load mode` option accepts the following values : +# `entry` : The eBPF collector only monitors calls for the functions, and does not show charts related to errors. +# `return : In the `return` mode, the eBPF collector monitors the same kernel functions as `entry`, but also creates +# new charts for the return of these functions, such as errors. +# +# The eBPF collector also creates charts for each running application through an integration with the `apps.plugin` +# or `cgroups.plugin`. +# If you want to disable the integration with `apps.plugin` or `cgroups.plugin` along with the above charts, change +# the setting `apps` and `cgroups` to 'no'. +# +# The `pid table size` defines the maximum number of PIDs stored inside the application hash table. +# +# The `ebpf type format` option accepts the following values : +# `auto` : The eBPF collector will investigate hardware and select between the two next options. +# `legacy`: The eBPF collector will load the legacy code. Note: This has a bigger overload. +# `co-re` : The eBPF collector will use latest tracing method. Note: This is not available on all platforms. +# +# The `ebpf co-re tracing` option accepts the following values: +# `trampoline`: This is the default mode used by the eBPF collector, due the small overhead added to host. +# `probe` : This is the same as legacy code. +# +# The `collect pid` option defines the PID stored inside hash tables and accepts the following options: +# `real parent`: Only stores real parent inside PID +# `parent` : Only stores parent PID. +# `all` : Stores all PIDs used by software. This is the most expensive option. +# +# Uncomment lines to define specific options for thread. +[global] +# ebpf load mode = entry +# apps = yes +# cgroups = no +# update every = 10 +# pid table size = 32768 + ebpf type format = auto + ebpf co-re tracing = trampoline + collect pid = real parent diff --git a/collectors/ebpf.plugin/ebpf.d/dcstat.conf b/collectors/ebpf.plugin/ebpf.d/dcstat.conf new file mode 100644 index 0000000..8aed8f7 --- /dev/null +++ b/collectors/ebpf.plugin/ebpf.d/dcstat.conf @@ -0,0 +1,34 @@ +# The `ebpf load mode` option accepts the following values : +# `entry` : The eBPF collector only monitors calls for the functions, and does not show charts related to errors. +# `return : In the `return` mode, the eBPF collector monitors the same kernel functions as `entry`, but also creates +# new charts for the return of these functions, such as errors. +# +# The eBPF collector also creates charts for each running application through an integration with the `apps.plugin` +# or `cgroups.plugin`. +# If you want to disable the integration with `apps.plugin` or `cgroups.plugin` along with the above charts, change +# the setting `apps` and `cgroups` to 'no'. +# +# The `ebpf type format` option accepts the following values : +# `auto` : The eBPF collector will investigate hardware and select between the two next options. +# `legacy`: The eBPF collector will load the legacy code. Note: This has a bigger overload. +# `co-re` : The eBPF collector will use latest tracing method. Note: This is not available on all platforms. +# +# The `ebpf co-re tracing` option accepts the following values: +# `trampoline`: This is the default mode used by the eBPF collector, due the small overhead added to host. +# `probe` : This is the same as legacy code. +# +# The `collect pid` option defines the PID stored inside hash tables and accepts the following options: +# `real parent`: Only stores real parent inside PID +# `parent` : Only stores parent PID. +# `all` : Stores all PIDs used by software. This is the most expensive option. +# +# Uncomment lines to define specific options for thread. +[global] +# ebpf load mode = entry +# apps = yes +# cgroups = no +# update every = 10 +# pid table size = 32768 + ebpf type format = auto + ebpf co-re tracing = trampoline + collect pid = real parent diff --git a/collectors/ebpf.plugin/ebpf.d/disk.conf b/collectors/ebpf.plugin/ebpf.d/disk.conf new file mode 100644 index 0000000..4adf88e --- /dev/null +++ b/collectors/ebpf.plugin/ebpf.d/disk.conf @@ -0,0 +1,9 @@ +# The `ebpf load mode` option accepts the following values : +# `entry` : The eBPF collector only monitors calls for the functions, and does not show charts related to errors. +# `return : In the `return` mode, the eBPF collector monitors the same kernel functions as `entry`, but also creates +# new charts for the return of these functions, such as errors. +# +#[global] +# ebpf load mode = entry +# update every = 10 + diff --git a/collectors/ebpf.plugin/ebpf.d/ebpf_kernel_reject_list.txt b/collectors/ebpf.plugin/ebpf.d/ebpf_kernel_reject_list.txt new file mode 100644 index 0000000..539bf35 --- /dev/null +++ b/collectors/ebpf.plugin/ebpf.d/ebpf_kernel_reject_list.txt @@ -0,0 +1 @@ +Ubuntu 4.18.0 diff --git a/collectors/ebpf.plugin/ebpf.d/fd.conf b/collectors/ebpf.plugin/ebpf.d/fd.conf new file mode 100644 index 0000000..8333520 --- /dev/null +++ b/collectors/ebpf.plugin/ebpf.d/fd.conf @@ -0,0 +1,21 @@ +# The `ebpf load mode` option accepts the following values : +# `entry` : The eBPF collector only monitors calls for the functions, and does not show charts related to errors. +# `return : In the `return` mode, the eBPF collector monitors the same kernel functions as `entry`, but also creates +# new charts for the return of these functions, such as errors. +# +# The eBPF collector also creates charts for each running application through an integration with the `apps.plugin` +# or `cgroups.plugin`. +# If you want to disable the integration with `apps.plugin` or `cgroups.plugin` along with the above charts, change +# the setting `apps` and `cgroups` to 'no'. +# +# The `pid table size` defines the maximum number of PIDs stored inside the hash table. +# +# Uncomment lines to define specific options for thread. +[global] +# ebpf load mode = entry +# apps = yes +# cgroups = no +# update every = 10 +# pid table size = 32768 + ebpf type format = auto + ebpf co-re tracing = trampoline diff --git a/collectors/ebpf.plugin/ebpf.d/filesystem.conf b/collectors/ebpf.plugin/ebpf.d/filesystem.conf new file mode 100644 index 0000000..c5eb01e --- /dev/null +++ b/collectors/ebpf.plugin/ebpf.d/filesystem.conf @@ -0,0 +1,20 @@ +# The `ebpf load mode` option accepts the following values : +# `entry` : The eBPF collector only monitors calls for the functions, and does not show charts related to errors. +# `return : In the `return` mode, the eBPF collector monitors the same kernel functions as `entry`, but also creates +# new charts for the return of these functions, such as errors. +# +# The eBPF collector also creates charts for each running application through an integration with the `apps plugin`. +# If you want to disable the integration with `apps.plugin` along with the above charts, change the setting `apps` to +# 'no'. +# +#[global] +# ebpf load mode = entry +# update every = 10 + +# All filesystems are named as 'NAMEdist' where NAME is the filesystem name while 'dist' is a reference for distribution. +[filesystem] + btrfsdist = yes + ext4dist = yes + nfsdist = yes + xfsdist = yes + zfsdist = yes diff --git a/collectors/ebpf.plugin/ebpf.d/hardirq.conf b/collectors/ebpf.plugin/ebpf.d/hardirq.conf new file mode 100644 index 0000000..f2bae1d --- /dev/null +++ b/collectors/ebpf.plugin/ebpf.d/hardirq.conf @@ -0,0 +1,8 @@ +# The `ebpf load mode` option accepts the following values : +# `entry` : The eBPF collector only monitors calls for the functions, and does not show charts related to errors. +# `return : In the `return` mode, the eBPF collector monitors the same kernel functions as `entry`, but also creates +# new charts for the return of these functions, such as errors. +# +#[global] +# ebpf load mode = entry +# update every = 10 diff --git a/collectors/ebpf.plugin/ebpf.d/mdflush.conf b/collectors/ebpf.plugin/ebpf.d/mdflush.conf new file mode 100644 index 0000000..e65e867 --- /dev/null +++ b/collectors/ebpf.plugin/ebpf.d/mdflush.conf @@ -0,0 +1,7 @@ +# The `ebpf load mode` option accepts the following values : +# `entry` : The eBPF collector only monitors calls for the functions, and does not show charts related to errors. +# `return : In the `return` mode, the eBPF collector monitors the same kernel functions as `entry`, but also creates +# new charts for the return of these functions, such as errors. +#[global] +# ebpf load mode = entry +# update every = 1 diff --git a/collectors/ebpf.plugin/ebpf.d/mount.conf b/collectors/ebpf.plugin/ebpf.d/mount.conf new file mode 100644 index 0000000..fdd82f2 --- /dev/null +++ b/collectors/ebpf.plugin/ebpf.d/mount.conf @@ -0,0 +1,19 @@ +# The `ebpf load mode` option accepts the following values : +# `entry` : The eBPF collector only monitors calls for the functions, and does not show charts related to errors. +# `return : In the `return` mode, the eBPF collector monitors the same kernel functions as `entry`, but also creates +# new charts for the return of these functions, such as errors. +# +# The `ebpf type format` option accepts the following values : +# `auto` : The eBPF collector will investigate hardware and select between the two next options. +# `legacy`: The eBPF collector will load the legacy code. Note: This has a bigger overload. +# `co-re` : The eBPF collector will use latest tracing method. Note: This is not available on all platforms. +# +# The `ebpf co-re tracing` option accepts the following values: +# `trampoline`: This is the default mode used by the eBPF collector, due the small overhead added to host. +# `tracepoint`: When available, the eBPF collector will use kernel tracepoint to monitor syscall. +# `probe` : This is the same as legacy code. +[global] +# ebpf load mode = entry +# update every = 1 + ebpf type format = auto + ebpf co-re tracing = trampoline diff --git a/collectors/ebpf.plugin/ebpf.d/network.conf b/collectors/ebpf.plugin/ebpf.d/network.conf new file mode 100644 index 0000000..d939d8e --- /dev/null +++ b/collectors/ebpf.plugin/ebpf.d/network.conf @@ -0,0 +1,53 @@ +# The `ebpf load mode` option accepts the following values : +# `entry` : The eBPF collector only monitors calls for the functions, and does not show charts related to errors. +# `return : In the `return` mode, the eBPF collector monitors the same kernel functions as `entry`, but also creates +# new charts for the return of these functions, such as errors. +# +# The eBPF collector also creates charts for each running application through an integration with the `apps.plugin` +# or `cgroups.plugin`. +# If you want to disable the integration with `apps.plugin` or `cgroups.plugin` along with the above charts, change +# the setting `apps` and `cgroups` to 'no'. +# +# The following options change the hash table size: +# `bandwidth table size`: Maximum number of connections monitored +# `ipv4 connection table size`: Maximum number of IPV4 connections monitored +# `ipv6 connection table size`: Maximum number of IPV6 connections monitored +# `udp connection table size`: Maximum number of UDP connections monitored +# +# The `ebpf type format` option accepts the following values : +# `auto` : The eBPF collector will investigate hardware and select between the two next options. +# `legacy`: The eBPF collector will load the legacy code. Note: This has a bigger overload. +# `co-re` : The eBPF collector will use latest tracing method. Note: This is not available on all platforms. +# +# The `ebpf co-re tracing` option accepts the following values: +# `trampoline`: This is the default mode used by the eBPF collector, due the small overhead added to host. +# `tracepoint`: When available, the eBPF collector will use kernel tracepoint to monitor syscall. +# `probe` : This is the same as legacy code. +# +[global] +# ebpf load mode = entry +# apps = yes +# cgroups = no +# update every = 10 + bandwidth table size = 16384 + ipv4 connection table size = 16384 + ipv6 connection table size = 16384 + udp connection table size = 4096 + ebpf type format = auto + ebpf co-re tracing = trampoline + +# +# Network Connection +# +# This is a feature with status WIP(Work in Progress) +# +[network connections] + maximum dimensions = 50 + resolve hostnames = no + resolve service names = no + ports = * + ips = !127.0.0.1/8 10.0.0.0/8 172.16.0.0/12 192.168.0.0/16 fc00::/7 !::1/128 + hostnames = * + +[service name] + 19999 = Netdata diff --git a/collectors/ebpf.plugin/ebpf.d/oomkill.conf b/collectors/ebpf.plugin/ebpf.d/oomkill.conf new file mode 100644 index 0000000..e65e867 --- /dev/null +++ b/collectors/ebpf.plugin/ebpf.d/oomkill.conf @@ -0,0 +1,7 @@ +# The `ebpf load mode` option accepts the following values : +# `entry` : The eBPF collector only monitors calls for the functions, and does not show charts related to errors. +# `return : In the `return` mode, the eBPF collector monitors the same kernel functions as `entry`, but also creates +# new charts for the return of these functions, such as errors. +#[global] +# ebpf load mode = entry +# update every = 1 diff --git a/collectors/ebpf.plugin/ebpf.d/process.conf b/collectors/ebpf.plugin/ebpf.d/process.conf new file mode 100644 index 0000000..1da5f84 --- /dev/null +++ b/collectors/ebpf.plugin/ebpf.d/process.conf @@ -0,0 +1,25 @@ +# The `ebpf load mode` option accepts the following values : +# `entry` : The eBPF collector only monitors calls for the functions, and does not show charts related to errors. +# `return : In the `return` mode, the eBPF collector monitors the same kernel functions as `entry`, but also creates +# new charts for the return of these functions, such as errors. +# +# The eBPF collector also creates charts for each running application through an integration with the `apps.plugin` +# or `cgroups.plugin`. +# If you want to disable the integration with `apps.plugin` or `cgroups.plugin` along with the above charts, change +# the setting `apps` and `cgroups` to 'no'. +# +# The `pid table size` defines the maximum number of PIDs stored inside the hash table. +# +# The `collect pid` option defines the PID stored inside hash tables and accepts the following options: +# `real parent`: Only stores real parent inside PID +# `parent` : Only stores parent PID. +# `all` : Stores all PIDs used by software. This is the most expensive option. +# +# Uncomment lines to define specific options for thread. +#[global] +# ebpf load mode = entry +# apps = yes +# cgroups = no +# update every = 10 +# pid table size = 32768 +# collect pid = real parent diff --git a/collectors/ebpf.plugin/ebpf.d/shm.conf b/collectors/ebpf.plugin/ebpf.d/shm.conf new file mode 100644 index 0000000..23ab96d --- /dev/null +++ b/collectors/ebpf.plugin/ebpf.d/shm.conf @@ -0,0 +1,36 @@ +# The `ebpf load mode` option accepts the following values : +# `entry` : The eBPF collector only monitors calls for the functions, and does not show charts related to errors. +# `return : In the `return` mode, the eBPF collector monitors the same kernel functions as `entry`, but also creates +# new charts for the return of these functions, such as errors. +# +# The eBPF collector also creates charts for each running application through an integration with the `apps.plugin` +# or `cgroups.plugin`. +# If you want to disable the integration with `apps.plugin` or `cgroups.plugin` along with the above charts, change +# the setting `apps` and `cgroups` to 'no'. +# +# The `ebpf type format` option accepts the following values : +# `auto` : The eBPF collector will investigate hardware and select between the two next options. +# `legacy`: The eBPF collector will load the legacy code. Note: This has a bigger overload. +# `co-re` : The eBPF collector will use latest tracing method. Note: This is not available on all platforms. +# +# The `ebpf co-re tracing` option accepts the following values: +# `trampoline`: This is the default mode used by the eBPF collector, due the small overhead added to host. +# `tracepoint`: When available, the eBPF collector will use kernel tracepoint to monitor syscall. +# `probe` : This is the same as legacy code. +# +# Uncomment lines to define specific options for thread. +[global] +# ebpf load mode = entry +# apps = yes +# cgroups = no +# update every = 10 +# pid table size = 32768 + ebpf type format = auto + ebpf co-re tracing = trampoline + +# List of monitored syscalls +[syscalls] + shmget = yes + shmat = yes + shmdt = yes + shmctl = yes diff --git a/collectors/ebpf.plugin/ebpf.d/softirq.conf b/collectors/ebpf.plugin/ebpf.d/softirq.conf new file mode 100644 index 0000000..f2bae1d --- /dev/null +++ b/collectors/ebpf.plugin/ebpf.d/softirq.conf @@ -0,0 +1,8 @@ +# The `ebpf load mode` option accepts the following values : +# `entry` : The eBPF collector only monitors calls for the functions, and does not show charts related to errors. +# `return : In the `return` mode, the eBPF collector monitors the same kernel functions as `entry`, but also creates +# new charts for the return of these functions, such as errors. +# +#[global] +# ebpf load mode = entry +# update every = 10 diff --git a/collectors/ebpf.plugin/ebpf.d/swap.conf b/collectors/ebpf.plugin/ebpf.d/swap.conf new file mode 100644 index 0000000..3986ae4 --- /dev/null +++ b/collectors/ebpf.plugin/ebpf.d/swap.conf @@ -0,0 +1,28 @@ +# The `ebpf load mode` option accepts the following values : +# `entry` : The eBPF collector only monitors calls for the functions, and does not show charts related to errors. +# `return : In the `return` mode, the eBPF collector monitors the same kernel functions as `entry`, but also creates +# new charts for the return of these functions, such as errors. +# +# The eBPF collector also creates charts for each running application through an integration with the `apps.plugin` +# or `cgroups.plugin`. +# If you want to disable the integration with `apps.plugin` or `cgroups.plugin` along with the above charts, change +# the setting `apps` and `cgroups` to 'no'. +# +# The `ebpf type format` option accepts the following values : +# `auto` : The eBPF collector will investigate hardware and select between the two next options. +# `legacy`: The eBPF collector will load the legacy code. Note: This has a bigger overload. +# `co-re` : The eBPF collector will use latest tracing method. Note: This is not available on all platforms. +# +# The `ebpf co-re tracing` option accepts the following values: +# `trampoline`: This is the default mode used by the eBPF collector, due the small overhead added to host. +# `probe` : This is the same as legacy code. +# +# Uncomment lines to define specific options for thread. +[global] +# ebpf load mode = entry +# apps = yes +# cgroups = no +# update every = 10 +# pid table size = 32768 + ebpf type format = auto + ebpf co-re tracing = trampoline diff --git a/collectors/ebpf.plugin/ebpf.d/sync.conf b/collectors/ebpf.plugin/ebpf.d/sync.conf new file mode 100644 index 0000000..ebec5d3 --- /dev/null +++ b/collectors/ebpf.plugin/ebpf.d/sync.conf @@ -0,0 +1,36 @@ +# The `ebpf load mode` option accepts the following values : +# `entry` : The eBPF collector only monitors calls for the functions, and does not show charts related to errors. +# `return : In the `return` mode, the eBPF collector monitors the same kernel functions as `entry`, but also creates +# new charts for the return of these functions, such as errors. +# +# The eBPF collector also creates charts for each running application through an integration with the `apps.plugin` +# or `cgroups.plugin`. +# If you want to disable the integration with `apps.plugin` or `cgroups.plugin` along with the above charts, change +# the setting `apps` and `cgroups` to 'no'. +# +# The `ebpf type format` option accepts the following values : +# `auto` : The eBPF collector will investigate hardware and select between the two next options. +# `legacy`: The eBPF collector will load the legacy code. Note: This has a bigger overload. +# `co-re` : The eBPF collector will use latest tracing method. Note: This is not available on all platforms. +# +# The `ebpf co-re tracing` option accepts the following values: +# `trampoline`: This is the default mode used by the eBPF collector, due the small overhead added to host. +# `tracepoint`: When available, the eBPF collector will use kernel tracepoint to monitor syscall. +# `probe` : This is the same as legacy code. +# +[global] +# ebpf load mode = entry +# apps = yes +# cgroups = no +# update every = 10 + ebpf type format = auto + ebpf co-re tracing = trampoline + +# List of monitored syscalls +[syscalls] + sync = yes + msync = yes + fsync = yes + fdatasync = yes + syncfs = yes + sync_file_range = yes diff --git a/collectors/ebpf.plugin/ebpf.d/vfs.conf b/collectors/ebpf.plugin/ebpf.d/vfs.conf new file mode 100644 index 0000000..fa5d5b4 --- /dev/null +++ b/collectors/ebpf.plugin/ebpf.d/vfs.conf @@ -0,0 +1,19 @@ +# The `ebpf load mode` option accepts the following values : +# `entry` : The eBPF collector only monitors calls for the functions, and does not show charts related to errors. +# `return : In the `return` mode, the eBPF collector monitors the same kernel functions as `entry`, but also creates +# new charts for the return of these functions, such as errors. +# +# The eBPF collector also creates charts for each running application through an integration with the `apps.plugin` +# or `cgroups.plugin`. +# If you want to disable the integration with `apps.plugin` or `cgroups.plugin` along with the above charts, change +# the setting `apps` and `cgroups` to 'no'. +# +# Uncomment lines to define specific options for thread. +[global] +# ebpf load mode = entry +# apps = yes +# cgroups = no +# update every = 10 +# pid table size = 32768 + ebpf type format = auto + ebpf co-re tracing = trampoline diff --git a/collectors/ebpf.plugin/ebpf.h b/collectors/ebpf.plugin/ebpf.h new file mode 100644 index 0000000..28b04ce --- /dev/null +++ b/collectors/ebpf.plugin/ebpf.h @@ -0,0 +1,299 @@ +// SPDX-License-Identifier: GPL-3.0-or-later + +#ifndef NETDATA_COLLECTOR_EBPF_H +#define NETDATA_COLLECTOR_EBPF_H 1 + +#ifndef __FreeBSD__ +#include <linux/perf_event.h> +#endif +#include <stdint.h> +#include <errno.h> +#include <signal.h> +#include <stdio.h> +#include <stdlib.h> +#include <string.h> +#include <unistd.h> +#include <dlfcn.h> + +#include <fcntl.h> +#include <ctype.h> +#include <dirent.h> + +// From libnetdata.h +#include "libnetdata/threads/threads.h" +#include "libnetdata/locks/locks.h" +#include "libnetdata/avl/avl.h" +#include "libnetdata/clocks/clocks.h" +#include "libnetdata/config/appconfig.h" +#include "libnetdata/ebpf/ebpf.h" +#include "libnetdata/procfile/procfile.h" +#include "collectors/cgroups.plugin/sys_fs_cgroup.h" +#include "daemon/main.h" + +#include "ebpf_apps.h" +#include "ebpf_cgroup.h" + +#define NETDATA_EBPF_OLD_CONFIG_FILE "ebpf.conf" +#define NETDATA_EBPF_CONFIG_FILE "ebpf.d.conf" + +typedef struct netdata_syscall_stat { + unsigned long bytes; // total number of bytes + uint64_t call; // total number of calls + uint64_t ecall; // number of calls that returned error + struct netdata_syscall_stat *next; // Link list +} netdata_syscall_stat_t; + +typedef uint64_t netdata_idx_t; + +typedef struct netdata_publish_syscall { + char *dimension; + char *name; + char *algorithm; + unsigned long nbyte; + unsigned long pbyte; + uint64_t ncall; + uint64_t pcall; + uint64_t nerr; + uint64_t perr; + struct netdata_publish_syscall *next; +} netdata_publish_syscall_t; + +typedef struct netdata_publish_vfs_common { + long write; + long read; + + long running; + long zombie; +} netdata_publish_vfs_common_t; + +typedef struct netdata_error_report { + char comm[16]; + __u32 pid; + + int type; + int err; +} netdata_error_report_t; + +extern ebpf_module_t ebpf_modules[]; +enum ebpf_main_index { + EBPF_MODULE_PROCESS_IDX, + EBPF_MODULE_SOCKET_IDX, + EBPF_MODULE_CACHESTAT_IDX, + EBPF_MODULE_SYNC_IDX, + EBPF_MODULE_DCSTAT_IDX, + EBPF_MODULE_SWAP_IDX, + EBPF_MODULE_VFS_IDX, + EBPF_MODULE_FILESYSTEM_IDX, + EBPF_MODULE_DISK_IDX, + EBPF_MODULE_MOUNT_IDX, + EBPF_MODULE_FD_IDX, + EBPF_MODULE_HARDIRQ_IDX, + EBPF_MODULE_SOFTIRQ_IDX, + EBPF_MODULE_OOMKILL_IDX, + EBPF_MODULE_SHM_IDX, + EBPF_MODULE_MDFLUSH_IDX, + /* THREADS MUST BE INCLUDED BEFORE THIS COMMENT */ + EBPF_OPTION_ALL_CHARTS, + EBPF_OPTION_VERSION, + EBPF_OPTION_HELP, + EBPF_OPTION_GLOBAL_CHART, + EBPF_OPTION_RETURN_MODE, + EBPF_OPTION_LEGACY, + EBPF_OPTION_CORE +}; + +typedef struct ebpf_tracepoint { + bool enabled; + char *class; + char *event; +} ebpf_tracepoint_t; + +enum ebpf_threads_status { + NETDATA_THREAD_EBPF_RUNNING, + NETDATA_THREAD_EBPF_STOPPING, + NETDATA_THREAD_EBPF_STOPPED +}; + +// Copied from musl header +#ifndef offsetof +#if __GNUC__ > 3 +#define offsetof(type, member) __builtin_offsetof(type, member) +#else +#define offsetof(type, member) ((size_t)((char *)&(((type *)0)->member) - (char *)0)) +#endif +#endif + +// Chart definitions +#define NETDATA_EBPF_FAMILY "ebpf" +#define NETDATA_EBPF_IP_FAMILY "ip" +#define NETDATA_FILESYSTEM_FAMILY "filesystem" +#define NETDATA_EBPF_MOUNT_GLOBAL_FAMILY "mount_points" +#define NETDATA_EBPF_CHART_TYPE_LINE "line" +#define NETDATA_EBPF_CHART_TYPE_STACKED "stacked" +#define NETDATA_EBPF_MEMORY_GROUP "mem" +#define NETDATA_EBPF_SYSTEM_GROUP "system" +#define NETDATA_SYSTEM_SWAP_SUBMENU "swap" +#define NETDATA_SYSTEM_CGROUP_SWAP_SUBMENU "swap (eBPF)" +#define NETDATA_SYSTEM_IPC_SHM_SUBMENU "ipc shared memory" +#define NETDATA_MONITORING_FAMILY "netdata" + +// Statistics charts +#define NETDATA_EBPF_THREADS "ebpf_threads" +#define NETDATA_EBPF_LOAD_METHOD "ebpf_load_methods" + +// Log file +#define NETDATA_DEVELOPER_LOG_FILE "developer.log" + +// Maximum number of processors monitored on perf events +#define NETDATA_MAX_PROCESSOR 512 + +// Kernel versions calculated with the formula: +// R = MAJOR*65536 + MINOR*256 + PATCH +#define NETDATA_KERNEL_V5_3 328448 +#define NETDATA_KERNEL_V4_15 265984 + +#define EBPF_SYS_CLONE_IDX 11 +#define EBPF_MAX_MAPS 32 + +#define EBPF_DEFAULT_UPDATE_EVERY 10 + +enum ebpf_algorithms_list { + NETDATA_EBPF_ABSOLUTE_IDX, + NETDATA_EBPF_INCREMENTAL_IDX +}; + +// Threads +void *ebpf_process_thread(void *ptr); +void *ebpf_socket_thread(void *ptr); + +// Common variables +extern pthread_mutex_t lock; +extern pthread_mutex_t ebpf_exit_cleanup; +extern int ebpf_nprocs; +extern int running_on_kernel; +extern int isrh; +extern char *ebpf_plugin_dir; + +extern pthread_mutex_t collect_data_mutex; +extern pthread_cond_t collect_data_cond_var; + +// Common functions +void ebpf_global_labels(netdata_syscall_stat_t *is, + netdata_publish_syscall_t *pio, + char **dim, + char **name, + int *algorithm, + int end); + +void ebpf_write_chart_cmd(char *type, + char *id, + char *title, + char *units, + char *family, + char *charttype, + char *context, + int order, + int update_every, + char *module); + +void ebpf_write_global_dimension(char *name, char *id, char *algorithm); + +void ebpf_create_global_dimension(void *ptr, int end); + +void ebpf_create_chart(char *type, + char *id, + char *title, + char *units, + char *family, + char *context, + char *charttype, + int order, + void (*ncd)(void *, int), + void *move, + int end, + int update_every, + char *module); + +void write_begin_chart(char *family, char *name); + +void write_chart_dimension(char *dim, long long value); + +void write_count_chart(char *name, char *family, netdata_publish_syscall_t *move, uint32_t end); + +void write_err_chart(char *name, char *family, netdata_publish_syscall_t *move, int end); + +void write_io_chart(char *chart, char *family, char *dwrite, long long vwrite, + char *dread, long long vread); + +void ebpf_create_charts_on_apps(char *name, + char *title, + char *units, + char *family, + char *charttype, + int order, + char *algorithm, + struct target *root, + int update_every, + char *module); + +void write_end_chart(); + +void ebpf_cleanup_publish_syscall(netdata_publish_syscall_t *nps); + +int ebpf_enable_tracepoint(ebpf_tracepoint_t *tp); +int ebpf_disable_tracepoint(ebpf_tracepoint_t *tp); +uint32_t ebpf_enable_tracepoints(ebpf_tracepoint_t *tps); + +void ebpf_pid_file(char *filename, size_t length); + +#define EBPF_PROGRAMS_SECTION "ebpf programs" + +#define EBPF_COMMON_DIMENSION_PERCENTAGE "%" +#define EBPF_COMMON_DIMENSION_CALL "calls/s" +#define EBPF_COMMON_DIMENSION_CONNECTIONS "connections/s" +#define EBPF_COMMON_DIMENSION_BITS "kilobits/s" +#define EBPF_COMMON_DIMENSION_BYTES "bytes/s" +#define EBPF_COMMON_DIMENSION_DIFFERENCE "difference" +#define EBPF_COMMON_DIMENSION_PACKETS "packets" +#define EBPF_COMMON_DIMENSION_FILES "files" +#define EBPF_COMMON_DIMENSION_MILLISECONDS "milliseconds" +#define EBPF_COMMON_DIMENSION_KILLS "kills" + +// Common variables +extern int debug_enabled; +extern struct pid_stat *root_of_pids; +extern ebpf_cgroup_target_t *ebpf_cgroup_pids; +extern char *ebpf_algorithms[]; +extern struct config collector_config; +extern ebpf_process_stat_t *global_process_stat; +extern netdata_ebpf_cgroup_shm_t shm_ebpf_cgroup; +extern int shm_fd_ebpf_cgroup; +extern sem_t *shm_sem_ebpf_cgroup; +extern pthread_mutex_t mutex_cgroup_shm; +extern size_t all_pids_count; +extern ebpf_plugin_stats_t plugin_statistics; +#ifdef LIBBPF_MAJOR_VERSION +extern struct btf *default_btf; +#else +extern void *default_btf; +#endif + +// Socket functions and variables +// Common functions +void ebpf_process_create_apps_charts(struct ebpf_module *em, void *ptr); +void ebpf_socket_create_apps_charts(struct ebpf_module *em, void *ptr); +void ebpf_cachestat_create_apps_charts(struct ebpf_module *em, void *root); +void ebpf_one_dimension_write_charts(char *family, char *chart, char *dim, long long v1); +collected_number get_value_from_structure(char *basis, size_t offset); +void ebpf_update_pid_table(ebpf_local_maps_t *pid, ebpf_module_t *em); +void ebpf_write_chart_obsolete(char *type, char *id, char *title, char *units, char *family, + char *charttype, char *context, int order, int update_every); +void write_histogram_chart(char *family, char *name, const netdata_idx_t *hist, char **dimensions, uint32_t end); +void ebpf_update_disabled_plugin_stats(ebpf_module_t *em); +extern ebpf_filesystem_partitions_t localfs[]; +extern ebpf_sync_syscalls_t local_syscalls[]; +extern int ebpf_exit_plugin; + +#define EBPF_MAX_SYNCHRONIZATION_TIME 300 + +#endif /* NETDATA_COLLECTOR_EBPF_H */ diff --git a/collectors/ebpf.plugin/ebpf_apps.c b/collectors/ebpf.plugin/ebpf_apps.c new file mode 100644 index 0000000..7519e06 --- /dev/null +++ b/collectors/ebpf.plugin/ebpf_apps.c @@ -0,0 +1,1155 @@ +// SPDX-License-Identifier: GPL-3.0-or-later + +#include "ebpf.h" +#include "ebpf_socket.h" +#include "ebpf_apps.h" + +// ---------------------------------------------------------------------------- +// internal flags +// handled in code (automatically set) + +static int proc_pid_cmdline_is_needed = 0; // 1 when we need to read /proc/cmdline + +/***************************************************************** + * + * FUNCTIONS USED TO READ HASH TABLES + * + *****************************************************************/ + +/** + * Read statistic hash table. + * + * @param ep the output structure. + * @param fd the file descriptor mapped from kernel ring. + * @param pid the index used to select the data. + * @param bpf_map_lookup_elem a pointer for the function used to read data. + * + * @return It returns 0 when the data was copied and -1 otherwise + */ +int ebpf_read_hash_table(void *ep, int fd, uint32_t pid) +{ + if (!ep) + return -1; + + if (!bpf_map_lookup_elem(fd, &pid, ep)) + return 0; + + return -1; +} + +/** + * Read socket statistic + * + * Read information from kernel ring to user ring. + * + * @param ep the table with all process stats values. + * @param fd the file descriptor mapped from kernel + * @param ef a pointer for the functions mapped from dynamic library + * @param pids the list of pids associated to a target. + * + * @return + */ +size_t read_bandwidth_statistic_using_pid_on_target(ebpf_bandwidth_t **ep, int fd, struct pid_on_target *pids) +{ + size_t count = 0; + while (pids) { + uint32_t current_pid = pids->pid; + if (!ebpf_read_hash_table(ep[current_pid], fd, current_pid)) + count++; + + pids = pids->next; + } + + return count; +} + +/** + * Read bandwidth statistic using hash table + * + * @param out the output tensor that will receive the information. + * @param fd the file descriptor that has the data + * @param bpf_map_lookup_elem a pointer for the function to read the data + * @param bpf_map_get_next_key a pointer fo the function to read the index. + */ +size_t read_bandwidth_statistic_using_hash_table(ebpf_bandwidth_t **out, int fd) +{ + size_t count = 0; + uint32_t key = 0; + uint32_t next_key = 0; + + while (bpf_map_get_next_key(fd, &key, &next_key) == 0) { + ebpf_bandwidth_t *eps = out[next_key]; + if (!eps) { + eps = callocz(1, sizeof(ebpf_process_stat_t)); + out[next_key] = eps; + } + ebpf_read_hash_table(eps, fd, next_key); + } + + return count; +} + +/***************************************************************** + * + * FUNCTIONS CALLED FROM COLLECTORS + * + *****************************************************************/ + +/** + * Am I running as Root + * + * Verify the user that is running the collector. + * + * @return It returns 1 for root and 0 otherwise. + */ +int am_i_running_as_root() +{ + uid_t uid = getuid(), euid = geteuid(); + + if (uid == 0 || euid == 0) { + return 1; + } + + return 0; +} + +/** + * Reset the target values + * + * @param root the pointer to the chain that will be reset. + * + * @return it returns the number of structures that was reset. + */ +size_t zero_all_targets(struct target *root) +{ + struct target *w; + size_t count = 0; + + for (w = root; w; w = w->next) { + count++; + + if (unlikely(w->root_pid)) { + struct pid_on_target *pid_on_target = w->root_pid; + + while (pid_on_target) { + struct pid_on_target *pid_on_target_to_free = pid_on_target; + pid_on_target = pid_on_target->next; + freez(pid_on_target_to_free); + } + + w->root_pid = NULL; + } + } + + return count; +} + +/** + * Clean the allocated structures + * + * @param agrt the pointer to be cleaned. + */ +void clean_apps_groups_target(struct target *agrt) +{ + struct target *current_target; + while (agrt) { + current_target = agrt; + agrt = current_target->target; + + freez(current_target); + } +} + +/** + * Find or create a new target + * there are targets that are just aggregated to other target (the second argument) + * + * @param id + * @param target + * @param name + * + * @return It returns the target on success and NULL otherwise + */ +struct target *get_apps_groups_target(struct target **agrt, const char *id, struct target *target, const char *name) +{ + int tdebug = 0, thidden = target ? target->hidden : 0, ends_with = 0; + const char *nid = id; + + // extract the options + while (nid[0] == '-' || nid[0] == '+' || nid[0] == '*') { + if (nid[0] == '-') + thidden = 1; + if (nid[0] == '+') + tdebug = 1; + if (nid[0] == '*') + ends_with = 1; + nid++; + } + uint32_t hash = simple_hash(id); + + // find if it already exists + struct target *w, *last = *agrt; + for (w = *agrt; w; w = w->next) { + if (w->idhash == hash && strncmp(nid, w->id, MAX_NAME) == 0) + return w; + + last = w; + } + + // find an existing target + if (unlikely(!target)) { + while (*name == '-') { + if (*name == '-') + thidden = 1; + name++; + } + + for (target = *agrt; target != NULL; target = target->next) { + if (!target->target && strcmp(name, target->name) == 0) + break; + } + } + + if (target && target->target) + fatal( + "Internal Error: request to link process '%s' to target '%s' which is linked to target '%s'", id, + target->id, target->target->id); + + w = callocz(1, sizeof(struct target)); + strncpyz(w->id, nid, MAX_NAME); + w->idhash = simple_hash(w->id); + + if (unlikely(!target)) + // copy the name + strncpyz(w->name, name, MAX_NAME); + else + // copy the id + strncpyz(w->name, nid, MAX_NAME); + + strncpyz(w->compare, nid, MAX_COMPARE_NAME); + size_t len = strlen(w->compare); + if (w->compare[len - 1] == '*') { + w->compare[len - 1] = '\0'; + w->starts_with = 1; + } + w->ends_with = ends_with; + + if (w->starts_with && w->ends_with) + proc_pid_cmdline_is_needed = 1; + + w->comparehash = simple_hash(w->compare); + w->comparelen = strlen(w->compare); + + w->hidden = thidden; +#ifdef NETDATA_INTERNAL_CHECKS + w->debug_enabled = tdebug; +#else + if (tdebug) + fprintf(stderr, "apps.plugin has been compiled without debugging\n"); +#endif + w->target = target; + + // append it, to maintain the order in apps_groups.conf + if (last) + last->next = w; + else + *agrt = w; + + return w; +} + +/** + * Read the apps_groups.conf file + * + * @param agrt a pointer to apps_group_root_target + * @param path the directory to search apps_%s.conf + * @param file the word to complement the file name. + * + * @return It returns 0 on success and -1 otherwise + */ +int ebpf_read_apps_groups_conf(struct target **agdt, struct target **agrt, const char *path, const char *file) +{ + char filename[FILENAME_MAX + 1]; + + snprintfz(filename, FILENAME_MAX, "%s/apps_%s.conf", path, file); + + // ---------------------------------------- + + procfile *ff = procfile_open_no_log(filename, " :\t", PROCFILE_FLAG_DEFAULT); + if (!ff) + return -1; + + procfile_set_quotes(ff, "'\""); + + ff = procfile_readall(ff); + if (!ff) + return -1; + + size_t line, lines = procfile_lines(ff); + + for (line = 0; line < lines; line++) { + size_t word, words = procfile_linewords(ff, line); + if (!words) + continue; + + char *name = procfile_lineword(ff, line, 0); + if (!name || !*name) + continue; + + // find a possibly existing target + struct target *w = NULL; + + // loop through all words, skipping the first one (the name) + for (word = 0; word < words; word++) { + char *s = procfile_lineword(ff, line, word); + if (!s || !*s) + continue; + if (*s == '#') + break; + + // is this the first word? skip it + if (s == name) + continue; + + // add this target + struct target *n = get_apps_groups_target(agrt, s, w, name); + if (!n) { + error("Cannot create target '%s' (line %zu, word %zu)", s, line, word); + continue; + } + + // just some optimization + // to avoid searching for a target for each process + if (!w) + w = n->target ? n->target : n; + } + } + + procfile_close(ff); + + *agdt = get_apps_groups_target(agrt, "p+!o@w#e$i^r&7*5(-i)l-o_", NULL, "other"); // match nothing + if (!*agdt) + fatal("Cannot create default target"); + + struct target *ptr = *agdt; + if (ptr->target) + *agdt = ptr->target; + + return 0; +} + +// the minimum PID of the system +// this is also the pid of the init process +#define INIT_PID 1 + +// ---------------------------------------------------------------------------- +// string lengths + +#define MAX_COMPARE_NAME 100 +#define MAX_NAME 100 +#define MAX_CMDLINE 16384 + +struct pid_stat **all_pids = NULL; // to avoid allocations, we pre-allocate the + // the entire pid space. +struct pid_stat *root_of_pids = NULL; // global list of all processes running + +size_t all_pids_count = 0; // the number of processes running + +struct target + *apps_groups_default_target = NULL, // the default target + *apps_groups_root_target = NULL, // apps_groups.conf defined + *users_root_target = NULL, // users + *groups_root_target = NULL; // user groups + +size_t apps_groups_targets_count = 0; // # of apps_groups.conf targets + +// ---------------------------------------------------------------------------- +// internal counters + +static size_t + // global_iterations_counter = 1, + calls_counter = 0, + // file_counter = 0, + // filenames_allocated_counter = 0, + // inodes_changed_counter = 0, + // links_changed_counter = 0, + targets_assignment_counter = 0; + +// ---------------------------------------------------------------------------- +// debugging + +// log each problem once per process +// log flood protection flags (log_thrown) +#define PID_LOG_IO 0x00000001 +#define PID_LOG_STATUS 0x00000002 +#define PID_LOG_CMDLINE 0x00000004 +#define PID_LOG_FDS 0x00000008 +#define PID_LOG_STAT 0x00000010 + +int debug_enabled = 0; + +#ifdef NETDATA_INTERNAL_CHECKS + +#define debug_log(fmt, args...) \ + do { \ + if (unlikely(debug_enabled)) \ + debug_log_int(fmt, ##args); \ + } while (0) + +#else + +static inline void debug_log_dummy(void) +{ +} +#define debug_log(fmt, args...) debug_log_dummy() + +#endif + +/** + * Managed log + * + * Store log information if it is necessary. + * + * @param p the pid stat structure + * @param log the log id + * @param status the return from a function. + * + * @return It returns the status value. + */ +static inline int managed_log(struct pid_stat *p, uint32_t log, int status) +{ + if (unlikely(!status)) { + // error("command failed log %u, errno %d", log, errno); + + if (unlikely(debug_enabled || errno != ENOENT)) { + if (unlikely(debug_enabled || !(p->log_thrown & log))) { + p->log_thrown |= log; + switch (log) { + case PID_LOG_IO: + error( + "Cannot process %s/proc/%d/io (command '%s')", netdata_configured_host_prefix, p->pid, + p->comm); + break; + + case PID_LOG_STATUS: + error( + "Cannot process %s/proc/%d/status (command '%s')", netdata_configured_host_prefix, p->pid, + p->comm); + break; + + case PID_LOG_CMDLINE: + error( + "Cannot process %s/proc/%d/cmdline (command '%s')", netdata_configured_host_prefix, p->pid, + p->comm); + break; + + case PID_LOG_FDS: + error( + "Cannot process entries in %s/proc/%d/fd (command '%s')", netdata_configured_host_prefix, + p->pid, p->comm); + break; + + case PID_LOG_STAT: + break; + + default: + error("unhandled error for pid %d, command '%s'", p->pid, p->comm); + break; + } + } + } + errno = 0; + } else if (unlikely(p->log_thrown & log)) { + // error("unsetting log %u on pid %d", log, p->pid); + p->log_thrown &= ~log; + } + + return status; +} + +/** + * Get PID entry + * + * Get or allocate the PID entry for the specified pid. + * + * @param pid the pid to search the data. + * + * @return It returns the pid entry structure + */ +static inline struct pid_stat *get_pid_entry(pid_t pid) +{ + if (unlikely(all_pids[pid])) + return all_pids[pid]; + + struct pid_stat *p = callocz(1, sizeof(struct pid_stat)); + + if (likely(root_of_pids)) + root_of_pids->prev = p; + + p->next = root_of_pids; + root_of_pids = p; + + p->pid = pid; + + all_pids[pid] = p; + all_pids_count++; + + return p; +} + +/** + * Assign the PID to a target. + * + * @param p the pid_stat structure to assign for a target. + */ +static inline void assign_target_to_pid(struct pid_stat *p) +{ + targets_assignment_counter++; + + uint32_t hash = simple_hash(p->comm); + size_t pclen = strlen(p->comm); + + struct target *w; + for (w = apps_groups_root_target; w; w = w->next) { + // if(debug_enabled || (p->target && p->target->debug_enabled)) debug_log_int("\t\tcomparing '%s' with '%s'", w->compare, p->comm); + + // find it - 4 cases: + // 1. the target is not a pattern + // 2. the target has the prefix + // 3. the target has the suffix + // 4. the target is something inside cmdline + + if (unlikely( + ((!w->starts_with && !w->ends_with && w->comparehash == hash && !strcmp(w->compare, p->comm)) || + (w->starts_with && !w->ends_with && !strncmp(w->compare, p->comm, w->comparelen)) || + (!w->starts_with && w->ends_with && pclen >= w->comparelen && !strcmp(w->compare, &p->comm[pclen - w->comparelen])) || + (proc_pid_cmdline_is_needed && w->starts_with && w->ends_with && p->cmdline && strstr(p->cmdline, w->compare))))) { + if (w->target) + p->target = w->target; + else + p->target = w; + + if (debug_enabled || (p->target && p->target->debug_enabled)) + debug_log_int("%s linked to target %s", p->comm, p->target->name); + + break; + } + } +} + +// ---------------------------------------------------------------------------- +// update pids from proc + +/** + * Read cmd line from /proc/PID/cmdline + * + * @param p the pid_stat_structure. + * + * @return It returns 1 on success and 0 otherwise. + */ +static inline int read_proc_pid_cmdline(struct pid_stat *p) +{ + static char cmdline[MAX_CMDLINE + 1]; + + if (unlikely(!p->cmdline_filename)) { + char filename[FILENAME_MAX + 1]; + snprintfz(filename, FILENAME_MAX, "%s/proc/%d/cmdline", netdata_configured_host_prefix, p->pid); + p->cmdline_filename = strdupz(filename); + } + + int fd = open(p->cmdline_filename, procfile_open_flags, 0666); + if (unlikely(fd == -1)) + goto cleanup; + + ssize_t i, bytes = read(fd, cmdline, MAX_CMDLINE); + close(fd); + + if (unlikely(bytes < 0)) + goto cleanup; + + cmdline[bytes] = '\0'; + for (i = 0; i < bytes; i++) { + if (unlikely(!cmdline[i])) + cmdline[i] = ' '; + } + + if (p->cmdline) + freez(p->cmdline); + p->cmdline = strdupz(cmdline); + + debug_log("Read file '%s' contents: %s", p->cmdline_filename, p->cmdline); + + return 1; + +cleanup: + // copy the command to the command line + if (p->cmdline) + freez(p->cmdline); + p->cmdline = strdupz(p->comm); + return 0; +} + +/** + * Read information from /proc/PID/stat and /proc/PID/cmdline + * Assign target to pid + * + * @param p the pid stat structure to store the data. + * @param ptr an useless argument. + */ +static inline int read_proc_pid_stat(struct pid_stat *p, void *ptr) +{ + UNUSED(ptr); + + static procfile *ff = NULL; + + if (unlikely(!p->stat_filename)) { + char filename[FILENAME_MAX + 1]; + snprintfz(filename, FILENAME_MAX, "%s/proc/%d/stat", netdata_configured_host_prefix, p->pid); + p->stat_filename = strdupz(filename); + } + + int set_quotes = (!ff) ? 1 : 0; + + struct stat statbuf; + if (stat(p->stat_filename, &statbuf)) + return 0; + + ff = procfile_reopen(ff, p->stat_filename, NULL, PROCFILE_FLAG_NO_ERROR_ON_FILE_IO); + if (unlikely(!ff)) + return 0; + + if (unlikely(set_quotes)) + procfile_set_open_close(ff, "(", ")"); + + ff = procfile_readall(ff); + if (unlikely(!ff)) + return 0; + + p->last_stat_collected_usec = p->stat_collected_usec; + p->stat_collected_usec = now_monotonic_usec(); + calls_counter++; + + char *comm = procfile_lineword(ff, 0, 1); + p->ppid = (int32_t)str2pid_t(procfile_lineword(ff, 0, 3)); + + if (strcmp(p->comm, comm) != 0) { + if (unlikely(debug_enabled)) { + if (p->comm[0]) + debug_log("\tpid %d (%s) changed name to '%s'", p->pid, p->comm, comm); + else + debug_log("\tJust added %d (%s)", p->pid, comm); + } + + strncpyz(p->comm, comm, MAX_COMPARE_NAME); + + // /proc/<pid>/cmdline + if (likely(proc_pid_cmdline_is_needed)) + managed_log(p, PID_LOG_CMDLINE, read_proc_pid_cmdline(p)); + + assign_target_to_pid(p); + } + + if (unlikely(debug_enabled || (p->target && p->target->debug_enabled))) + debug_log_int( + "READ PROC/PID/STAT: %s/proc/%d/stat, process: '%s' on target '%s' (dt=%llu)", + netdata_configured_host_prefix, p->pid, p->comm, (p->target) ? p->target->name : "UNSET", + p->stat_collected_usec - p->last_stat_collected_usec); + + return 1; +} + +/** + * Collect data for PID + * + * @param pid the current pid that we are working + * @param ptr a NULL value + * + * @return It returns 1 on success and 0 otherwise + */ +static inline int collect_data_for_pid(pid_t pid, void *ptr) +{ + if (unlikely(pid < 0 || pid > pid_max)) { + error("Invalid pid %d read (expected %d to %d). Ignoring process.", pid, 0, pid_max); + return 0; + } + + struct pid_stat *p = get_pid_entry(pid); + if (unlikely(!p || p->read)) + return 0; + p->read = 1; + + if (unlikely(!managed_log(p, PID_LOG_STAT, read_proc_pid_stat(p, ptr)))) + // there is no reason to proceed if we cannot get its status + return 0; + + // check its parent pid + if (unlikely(p->ppid < 0 || p->ppid > pid_max)) { + error("Pid %d (command '%s') states invalid parent pid %d. Using 0.", pid, p->comm, p->ppid); + p->ppid = 0; + } + + // mark it as updated + p->updated = 1; + p->keep = 0; + p->keeploops = 0; + + return 1; +} + +/** + * Fill link list of parents with children PIDs + */ +static inline void link_all_processes_to_their_parents(void) +{ + struct pid_stat *p, *pp; + + // link all children to their parents + // and update children count on parents + for (p = root_of_pids; p; p = p->next) { + // for each process found + + p->sortlist = 0; + p->parent = NULL; + + if (unlikely(!p->ppid)) { + p->parent = NULL; + continue; + } + + pp = all_pids[p->ppid]; + if (likely(pp)) { + p->parent = pp; + pp->children_count++; + + if (unlikely(debug_enabled || (p->target && p->target->debug_enabled))) + debug_log_int( + "child %d (%s, %s) on target '%s' has parent %d (%s, %s).", p->pid, p->comm, + p->updated ? "running" : "exited", (p->target) ? p->target->name : "UNSET", pp->pid, pp->comm, + pp->updated ? "running" : "exited"); + } else { + p->parent = NULL; + debug_log("pid %d %s states parent %d, but the later does not exist.", p->pid, p->comm, p->ppid); + } + } +} + +/** + * Aggregate PIDs to targets. + */ +static void apply_apps_groups_targets_inheritance(void) +{ + struct pid_stat *p = NULL; + + // children that do not have a target + // inherit their target from their parent + int found = 1, loops = 0; + while (found) { + if (unlikely(debug_enabled)) + loops++; + found = 0; + for (p = root_of_pids; p; p = p->next) { + // if this process does not have a target + // and it has a parent + // and its parent has a target + // then, set the parent's target to this process + if (unlikely(!p->target && p->parent && p->parent->target)) { + p->target = p->parent->target; + found++; + + if (debug_enabled || (p->target && p->target->debug_enabled)) + debug_log_int( + "TARGET INHERITANCE: %s is inherited by %d (%s) from its parent %d (%s).", p->target->name, + p->pid, p->comm, p->parent->pid, p->parent->comm); + } + } + } + + // find all the procs with 0 childs and merge them to their parents + // repeat, until nothing more can be done. + int sortlist = 1; + found = 1; + while (found) { + if (unlikely(debug_enabled)) + loops++; + found = 0; + + for (p = root_of_pids; p; p = p->next) { + if (unlikely(!p->sortlist && !p->children_count)) + p->sortlist = sortlist++; + + if (unlikely( + !p->children_count // if this process does not have any children + && !p->merged // and is not already merged + && p->parent // and has a parent + && p->parent->children_count // and its parent has children + // and the target of this process and its parent is the same, + // or the parent does not have a target + && (p->target == p->parent->target || !p->parent->target) && + p->ppid != INIT_PID // and its parent is not init + )) { + // mark it as merged + p->parent->children_count--; + p->merged = 1; + + // the parent inherits the child's target, if it does not have a target itself + if (unlikely(p->target && !p->parent->target)) { + p->parent->target = p->target; + + if (debug_enabled || (p->target && p->target->debug_enabled)) + debug_log_int( + "TARGET INHERITANCE: %s is inherited by %d (%s) from its child %d (%s).", p->target->name, + p->parent->pid, p->parent->comm, p->pid, p->comm); + } + + found++; + } + } + + debug_log("TARGET INHERITANCE: merged %d processes", found); + } + + // init goes always to default target + if (all_pids[INIT_PID]) + all_pids[INIT_PID]->target = apps_groups_default_target; + + // pid 0 goes always to default target + if (all_pids[0]) + all_pids[0]->target = apps_groups_default_target; + + // give a default target on all top level processes + if (unlikely(debug_enabled)) + loops++; + for (p = root_of_pids; p; p = p->next) { + // if the process is not merged itself + // then is is a top level process + if (unlikely(!p->merged && !p->target)) + p->target = apps_groups_default_target; + + // make sure all processes have a sortlist + if (unlikely(!p->sortlist)) + p->sortlist = sortlist++; + } + + if (all_pids[1]) + all_pids[1]->sortlist = sortlist++; + + // give a target to all merged child processes + found = 1; + while (found) { + if (unlikely(debug_enabled)) + loops++; + found = 0; + for (p = root_of_pids; p; p = p->next) { + if (unlikely(!p->target && p->merged && p->parent && p->parent->target)) { + p->target = p->parent->target; + found++; + + if (debug_enabled || (p->target && p->target->debug_enabled)) + debug_log_int( + "TARGET INHERITANCE: %s is inherited by %d (%s) from its parent %d (%s) at phase 2.", + p->target->name, p->pid, p->comm, p->parent->pid, p->parent->comm); + } + } + } + + debug_log("apply_apps_groups_targets_inheritance() made %d loops on the process tree", loops); +} + +/** + * Update target timestamp. + * + * @param root the targets that will be updated. + */ +static inline void post_aggregate_targets(struct target *root) +{ + struct target *w; + for (w = root; w; w = w->next) { + if (w->collected_starttime) { + if (!w->starttime || w->collected_starttime < w->starttime) { + w->starttime = w->collected_starttime; + } + } else { + w->starttime = 0; + } + } +} + +/** + * Remove PID from the link list. + * + * @param pid the PID that will be removed. + */ +static inline void del_pid_entry(pid_t pid) +{ + struct pid_stat *p = all_pids[pid]; + + if (unlikely(!p)) { + error("attempted to free pid %d that is not allocated.", pid); + return; + } + + debug_log("process %d %s exited, deleting it.", pid, p->comm); + + if (root_of_pids == p) + root_of_pids = p->next; + + if (p->next) + p->next->prev = p->prev; + if (p->prev) + p->prev->next = p->next; + + freez(p->stat_filename); + freez(p->status_filename); + freez(p->io_filename); + freez(p->cmdline_filename); + freez(p->cmdline); + freez(p); + + all_pids[pid] = NULL; + all_pids_count--; +} + +/** + * Get command string associated with a PID. + * This can only safely be used when holding the `collect_data_mutex` lock. + * + * @param pid the pid to search the data. + * @param n the maximum amount of bytes to copy into dest. + * if this is greater than the size of the command, it is clipped. + * @param dest the target memory buffer to write the command into. + * @return -1 if the PID hasn't been scraped yet, 0 otherwise. + */ +int get_pid_comm(pid_t pid, size_t n, char *dest) +{ + struct pid_stat *stat; + + stat = all_pids[pid]; + if (unlikely(stat == NULL)) { + return -1; + } + + if (unlikely(n > sizeof(stat->comm))) { + n = sizeof(stat->comm); + } + + strncpyz(dest, stat->comm, n); + return 0; +} + +/** + * Cleanup variable from other threads + * + * @param pid current pid. + */ +void cleanup_variables_from_other_threads(uint32_t pid) +{ + // Clean socket structures + if (socket_bandwidth_curr) { + freez(socket_bandwidth_curr[pid]); + socket_bandwidth_curr[pid] = NULL; + } + + // Clean cachestat structure + if (cachestat_pid) { + freez(cachestat_pid[pid]); + cachestat_pid[pid] = NULL; + } + + // Clean directory cache structure + if (dcstat_pid) { + freez(dcstat_pid[pid]); + dcstat_pid[pid] = NULL; + } + + // Clean swap structure + if (swap_pid) { + freez(swap_pid[pid]); + swap_pid[pid] = NULL; + } + + // Clean vfs structure + if (vfs_pid) { + freez(vfs_pid[pid]); + vfs_pid[pid] = NULL; + } + + // Clean fd structure + if (fd_pid) { + freez(fd_pid[pid]); + fd_pid[pid] = NULL; + } + + // Clean shm structure + if (shm_pid) { + freez(shm_pid[pid]); + shm_pid[pid] = NULL; + } +} + +/** + * Remove PIDs when they are not running more. + */ +void cleanup_exited_pids() +{ + struct pid_stat *p = NULL; + + for (p = root_of_pids; p;) { + if (!p->updated && (!p->keep || p->keeploops > 0)) { + if (unlikely(debug_enabled && (p->keep || p->keeploops))) + debug_log(" > CLEANUP cannot keep exited process %d (%s) anymore - removing it.", p->pid, p->comm); + + pid_t r = p->pid; + p = p->next; + + // Clean process structure + freez(global_process_stats[r]); + global_process_stats[r] = NULL; + + freez(current_apps_data[r]); + current_apps_data[r] = NULL; + + cleanup_variables_from_other_threads(r); + + del_pid_entry(r); + } else { + if (unlikely(p->keep)) + p->keeploops++; + p->keep = 0; + p = p->next; + } + } +} + +/** + * Read proc filesystem for the first time. + * + * @return It returns 0 on success and -1 otherwise. + */ +static inline void read_proc_filesystem() +{ + char dirname[FILENAME_MAX + 1]; + + snprintfz(dirname, FILENAME_MAX, "%s/proc", netdata_configured_host_prefix); + DIR *dir = opendir(dirname); + if (!dir) + return; + + struct dirent *de = NULL; + + while ((de = readdir(dir))) { + char *endptr = de->d_name; + + if (unlikely(de->d_type != DT_DIR || de->d_name[0] < '0' || de->d_name[0] > '9')) + continue; + + pid_t pid = (pid_t)strtoul(de->d_name, &endptr, 10); + + // make sure we read a valid number + if (unlikely(endptr == de->d_name || *endptr != '\0')) + continue; + + collect_data_for_pid(pid, NULL); + } + closedir(dir); +} + +/** + * Aggregated PID on target + * + * @param w the target output + * @param p the pid with information to update + * @param o never used + */ +static inline void aggregate_pid_on_target(struct target *w, struct pid_stat *p, struct target *o) +{ + UNUSED(o); + + if (unlikely(!p->updated)) { + // the process is not running + return; + } + + if (unlikely(!w)) { + error("pid %d %s was left without a target!", p->pid, p->comm); + return; + } + + w->processes++; + struct pid_on_target *pid_on_target = mallocz(sizeof(struct pid_on_target)); + pid_on_target->pid = p->pid; + pid_on_target->next = w->root_pid; + w->root_pid = pid_on_target; +} + +/** + * Collect data for all process + * + * Read data from hash table and store it in appropriate vectors. + * It also creates the link between targets and PIDs. + * + * @param tbl_pid_stats_fd The mapped file descriptor for the hash table. + */ +void collect_data_for_all_processes(int tbl_pid_stats_fd) +{ + if (unlikely(!all_pids)) + return; + + struct pid_stat *pids = root_of_pids; // global list of all processes running + while (pids) { + if (pids->updated_twice) { + pids->read = 0; // mark it as not read, so that collect_data_for_pid() will read it + pids->updated = 0; + pids->merged = 0; + pids->children_count = 0; + pids->parent = NULL; + } else { + if (pids->updated) + pids->updated_twice = 1; + } + + pids = pids->next; + } + + read_proc_filesystem(); + + uint32_t key; + pids = root_of_pids; // global list of all processes running + // while (bpf_map_get_next_key(tbl_pid_stats_fd, &key, &next_key) == 0) { + while (pids) { + key = pids->pid; + ebpf_process_stat_t *w = global_process_stats[key]; + if (!w) { + w = callocz(1, sizeof(ebpf_process_stat_t)); + global_process_stats[key] = w; + } + + if (bpf_map_lookup_elem(tbl_pid_stats_fd, &key, w)) { + // Clean Process structures + freez(w); + global_process_stats[key] = NULL; + + freez(current_apps_data[key]); + current_apps_data[key] = NULL; + + cleanup_variables_from_other_threads(key); + + pids = pids->next; + continue; + } + + pids = pids->next; + } + + link_all_processes_to_their_parents(); + + apply_apps_groups_targets_inheritance(); + + apps_groups_targets_count = zero_all_targets(apps_groups_root_target); + + // this has to be done, before the cleanup + // // concentrate everything on the targets + for (pids = root_of_pids; pids; pids = pids->next) + aggregate_pid_on_target(pids->target, pids, NULL); + + post_aggregate_targets(apps_groups_root_target); +} diff --git a/collectors/ebpf.plugin/ebpf_apps.h b/collectors/ebpf.plugin/ebpf_apps.h new file mode 100644 index 0000000..0bea912 --- /dev/null +++ b/collectors/ebpf.plugin/ebpf_apps.h @@ -0,0 +1,441 @@ +// SPDX-License-Identifier: GPL-3.0-or-later + +#ifndef NETDATA_EBPF_APPS_H +#define NETDATA_EBPF_APPS_H 1 + +#include "libnetdata/threads/threads.h" +#include "libnetdata/locks/locks.h" +#include "libnetdata/avl/avl.h" +#include "libnetdata/clocks/clocks.h" +#include "libnetdata/config/appconfig.h" +#include "libnetdata/ebpf/ebpf.h" + +#define NETDATA_APPS_FAMILY "apps" +#define NETDATA_APPS_FILE_GROUP "file_access" +#define NETDATA_APPS_FILE_CGROUP_GROUP "file_access (eBPF)" +#define NETDATA_APPS_PROCESS_GROUP "process (eBPF)" +#define NETDATA_APPS_NET_GROUP "net" +#define NETDATA_APPS_IPC_SHM_GROUP "ipc shm (eBPF)" + +#include "ebpf_process.h" +#include "ebpf_dcstat.h" +#include "ebpf_disk.h" +#include "ebpf_fd.h" +#include "ebpf_filesystem.h" +#include "ebpf_hardirq.h" +#include "ebpf_cachestat.h" +#include "ebpf_mdflush.h" +#include "ebpf_mount.h" +#include "ebpf_oomkill.h" +#include "ebpf_shm.h" +#include "ebpf_socket.h" +#include "ebpf_softirq.h" +#include "ebpf_sync.h" +#include "ebpf_swap.h" +#include "ebpf_vfs.h" + +#define MAX_COMPARE_NAME 100 +#define MAX_NAME 100 + +// ---------------------------------------------------------------------------- +// process_pid_stat +// +// Fields read from the kernel ring for a specific PID +// +typedef struct process_pid_stat { + uint64_t pid_tgid; // Unique identifier + uint32_t pid; // process id + + // Count number of calls done for specific function + uint32_t open_call; + uint32_t write_call; + uint32_t writev_call; + uint32_t read_call; + uint32_t readv_call; + uint32_t unlink_call; + uint32_t exit_call; + uint32_t release_call; + uint32_t fork_call; + uint32_t clone_call; + uint32_t close_call; + + // Count number of bytes written or read + uint64_t write_bytes; + uint64_t writev_bytes; + uint64_t readv_bytes; + uint64_t read_bytes; + + // Count number of errors for the specified function + uint32_t open_err; + uint32_t write_err; + uint32_t writev_err; + uint32_t read_err; + uint32_t readv_err; + uint32_t unlink_err; + uint32_t fork_err; + uint32_t clone_err; + uint32_t close_err; +} process_pid_stat_t; + +// ---------------------------------------------------------------------------- +// socket_bandwidth +// +// Fields read from the kernel ring for a specific PID +// +typedef struct socket_bandwidth { + uint64_t first; + uint64_t ct; + uint64_t sent; + uint64_t received; + unsigned char removed; +} socket_bandwidth_t; + +// ---------------------------------------------------------------------------- +// pid_stat +// +// structure to store data for each process running +// see: man proc for the description of the fields + +struct pid_fd { + int fd; + +#ifndef __FreeBSD__ + ino_t inode; + char *filename; + uint32_t link_hash; + size_t cache_iterations_counter; + size_t cache_iterations_reset; +#endif +}; + +struct target { + char compare[MAX_COMPARE_NAME + 1]; + uint32_t comparehash; + size_t comparelen; + + char id[MAX_NAME + 1]; + uint32_t idhash; + + char name[MAX_NAME + 1]; + + uid_t uid; + gid_t gid; + + // Changes made to simplify integration between apps and eBPF. + netdata_publish_cachestat_t cachestat; + netdata_publish_dcstat_t dcstat; + netdata_publish_swap_t swap; + netdata_publish_vfs_t vfs; + netdata_fd_stat_t fd; + netdata_publish_shm_t shm; + + /* These variables are not necessary for eBPF collector + kernel_uint_t minflt; + kernel_uint_t cminflt; + kernel_uint_t majflt; + kernel_uint_t cmajflt; + kernel_uint_t utime; + kernel_uint_t stime; + kernel_uint_t gtime; + kernel_uint_t cutime; + kernel_uint_t cstime; + kernel_uint_t cgtime; + kernel_uint_t num_threads; + // kernel_uint_t rss; + + kernel_uint_t status_vmsize; + kernel_uint_t status_vmrss; + kernel_uint_t status_vmshared; + kernel_uint_t status_rssfile; + kernel_uint_t status_rssshmem; + kernel_uint_t status_vmswap; + + kernel_uint_t io_logical_bytes_read; + kernel_uint_t io_logical_bytes_written; + // kernel_uint_t io_read_calls; + // kernel_uint_t io_write_calls; + kernel_uint_t io_storage_bytes_read; + kernel_uint_t io_storage_bytes_written; + // kernel_uint_t io_cancelled_write_bytes; + + int *target_fds; + int target_fds_size; + + kernel_uint_t openfiles; + kernel_uint_t openpipes; + kernel_uint_t opensockets; + kernel_uint_t openinotifies; + kernel_uint_t openeventfds; + kernel_uint_t opentimerfds; + kernel_uint_t opensignalfds; + kernel_uint_t openeventpolls; + kernel_uint_t openother; + */ + + kernel_uint_t starttime; + kernel_uint_t collected_starttime; + + /* + kernel_uint_t uptime_min; + kernel_uint_t uptime_sum; + kernel_uint_t uptime_max; + */ + + unsigned int processes; // how many processes have been merged to this + int exposed; // if set, we have sent this to netdata + int hidden; // if set, we set the hidden flag on the dimension + int debug_enabled; + int ends_with; + int starts_with; // if set, the compare string matches only the + // beginning of the command + + struct pid_on_target *root_pid; // list of aggregated pids for target debugging + + struct target *target; // the one that will be reported to netdata + struct target *next; +}; + +extern struct target *apps_groups_default_target; +extern struct target *apps_groups_root_target; +extern struct target *users_root_target; +extern struct target *groups_root_target; + +struct pid_stat { + int32_t pid; + char comm[MAX_COMPARE_NAME + 1]; + char *cmdline; + + uint32_t log_thrown; + + // char state; + int32_t ppid; + + // int32_t pgrp; + // int32_t session; + // int32_t tty_nr; + // int32_t tpgid; + // uint64_t flags; + + /* + // these are raw values collected + kernel_uint_t minflt_raw; + kernel_uint_t cminflt_raw; + kernel_uint_t majflt_raw; + kernel_uint_t cmajflt_raw; + kernel_uint_t utime_raw; + kernel_uint_t stime_raw; + kernel_uint_t gtime_raw; // guest_time + kernel_uint_t cutime_raw; + kernel_uint_t cstime_raw; + kernel_uint_t cgtime_raw; // cguest_time + + // these are rates + kernel_uint_t minflt; + kernel_uint_t cminflt; + kernel_uint_t majflt; + kernel_uint_t cmajflt; + kernel_uint_t utime; + kernel_uint_t stime; + kernel_uint_t gtime; + kernel_uint_t cutime; + kernel_uint_t cstime; + kernel_uint_t cgtime; + + // int64_t priority; + // int64_t nice; + int32_t num_threads; + // int64_t itrealvalue; + kernel_uint_t collected_starttime; + // kernel_uint_t vsize; + // kernel_uint_t rss; + // kernel_uint_t rsslim; + // kernel_uint_t starcode; + // kernel_uint_t endcode; + // kernel_uint_t startstack; + // kernel_uint_t kstkesp; + // kernel_uint_t kstkeip; + // uint64_t signal; + // uint64_t blocked; + // uint64_t sigignore; + // uint64_t sigcatch; + // uint64_t wchan; + // uint64_t nswap; + // uint64_t cnswap; + // int32_t exit_signal; + // int32_t processor; + // uint32_t rt_priority; + // uint32_t policy; + // kernel_uint_t delayacct_blkio_ticks; + + uid_t uid; + gid_t gid; + + kernel_uint_t status_vmsize; + kernel_uint_t status_vmrss; + kernel_uint_t status_vmshared; + kernel_uint_t status_rssfile; + kernel_uint_t status_rssshmem; + kernel_uint_t status_vmswap; +#ifndef __FreeBSD__ + ARL_BASE *status_arl; +#endif + + kernel_uint_t io_logical_bytes_read_raw; + kernel_uint_t io_logical_bytes_written_raw; + // kernel_uint_t io_read_calls_raw; + // kernel_uint_t io_write_calls_raw; + kernel_uint_t io_storage_bytes_read_raw; + kernel_uint_t io_storage_bytes_written_raw; + // kernel_uint_t io_cancelled_write_bytes_raw; + + kernel_uint_t io_logical_bytes_read; + kernel_uint_t io_logical_bytes_written; + // kernel_uint_t io_read_calls; + // kernel_uint_t io_write_calls; + kernel_uint_t io_storage_bytes_read; + kernel_uint_t io_storage_bytes_written; + // kernel_uint_t io_cancelled_write_bytes; + */ + + struct pid_fd *fds; // array of fds it uses + size_t fds_size; // the size of the fds array + + int children_count; // number of processes directly referencing this + unsigned char keep : 1; // 1 when we need to keep this process in memory even after it exited + int keeploops; // increases by 1 every time keep is 1 and updated 0 + unsigned char updated : 1; // 1 when the process is currently running + unsigned char updated_twice : 1; // 1 when the process was running in the previous iteration + unsigned char merged : 1; // 1 when it has been merged to its parent + unsigned char read : 1; // 1 when we have already read this process for this iteration + + int sortlist; // higher numbers = top on the process tree + + // each process gets a unique number + + struct target *target; // app_groups.conf targets + struct target *user_target; // uid based targets + struct target *group_target; // gid based targets + + usec_t stat_collected_usec; + usec_t last_stat_collected_usec; + + usec_t io_collected_usec; + usec_t last_io_collected_usec; + + kernel_uint_t uptime; + + char *fds_dirname; // the full directory name in /proc/PID/fd + + char *stat_filename; + char *status_filename; + char *io_filename; + char *cmdline_filename; + + struct pid_stat *parent; + struct pid_stat *prev; + struct pid_stat *next; +}; + +// ---------------------------------------------------------------------------- +// target +// +// target is the structure that processes are aggregated to be reported +// to netdata. +// +// - Each entry in /etc/apps_groups.conf creates a target. +// - Each user and group used by a process in the system, creates a target. +struct pid_on_target { + int32_t pid; + struct pid_on_target *next; +}; + +// ---------------------------------------------------------------------------- +// Structures used to read information from kernel ring +typedef struct ebpf_process_stat { + uint64_t pid_tgid; + uint32_t pid; + + //Counter + uint32_t exit_call; + uint32_t release_call; + uint32_t create_process; + uint32_t create_thread; + + //Counter + uint32_t task_err; + + uint8_t removeme; +} ebpf_process_stat_t; + +typedef struct ebpf_bandwidth { + uint32_t pid; + + uint64_t first; // First timestamp + uint64_t ct; // Last timestamp + uint64_t bytes_sent; // Bytes sent + uint64_t bytes_received; // Bytes received + uint64_t call_tcp_sent; // Number of times tcp_sendmsg was called + uint64_t call_tcp_received; // Number of times tcp_cleanup_rbuf was called + uint64_t retransmit; // Number of times tcp_retransmit was called + uint64_t call_udp_sent; // Number of times udp_sendmsg was called + uint64_t call_udp_received; // Number of times udp_recvmsg was called + uint64_t close; // Number of times tcp_close was called + uint64_t drop; // THIS IS NOT USED FOR WHILE, we are in groom section + uint32_t tcp_v4_connection; // Number of times tcp_v4_connection was called. + uint32_t tcp_v6_connection; // Number of times tcp_v6_connection was called. +} ebpf_bandwidth_t; + +/** + * Internal function used to write debug messages. + * + * @param fmt the format to create the message. + * @param ... the arguments to fill the format. + */ +static inline void debug_log_int(const char *fmt, ...) +{ + va_list args; + + fprintf(stderr, "apps.plugin: "); + va_start(args, fmt); + vfprintf(stderr, fmt, args); + va_end(args); + + fputc('\n', stderr); +} + +// ---------------------------------------------------------------------------- +// Exported variabled and functions +// +extern struct pid_stat **all_pids; + +int ebpf_read_apps_groups_conf(struct target **apps_groups_default_target, + struct target **apps_groups_root_target, + const char *path, + const char *file); + +void clean_apps_groups_target(struct target *apps_groups_root_target); + +size_t zero_all_targets(struct target *root); + +int am_i_running_as_root(); + +void cleanup_exited_pids(); + +int ebpf_read_hash_table(void *ep, int fd, uint32_t pid); + +int get_pid_comm(pid_t pid, size_t n, char *dest); + +size_t read_processes_statistic_using_pid_on_target(ebpf_process_stat_t **ep, + int fd, + struct pid_on_target *pids); + +size_t read_bandwidth_statistic_using_pid_on_target(ebpf_bandwidth_t **ep, int fd, struct pid_on_target *pids); + +void collect_data_for_all_processes(int tbl_pid_stats_fd); + +extern ebpf_process_stat_t **global_process_stats; +extern ebpf_process_publish_apps_t **current_apps_data; +extern netdata_publish_cachestat_t **cachestat_pid; +extern netdata_publish_dcstat_t **dcstat_pid; + +#endif /* NETDATA_EBPF_APPS_H */ diff --git a/collectors/ebpf.plugin/ebpf_cachestat.c b/collectors/ebpf.plugin/ebpf_cachestat.c new file mode 100644 index 0000000..4c41064 --- /dev/null +++ b/collectors/ebpf.plugin/ebpf_cachestat.c @@ -0,0 +1,1335 @@ +// SPDX-License-Identifier: GPL-3.0-or-later + +#include "ebpf.h" +#include "ebpf_cachestat.h" + +netdata_publish_cachestat_t **cachestat_pid; + +static char *cachestat_counter_dimension_name[NETDATA_CACHESTAT_END] = { "ratio", "dirty", "hit", + "miss" }; +static netdata_syscall_stat_t cachestat_counter_aggregated_data[NETDATA_CACHESTAT_END]; +static netdata_publish_syscall_t cachestat_counter_publish_aggregated[NETDATA_CACHESTAT_END]; + +netdata_cachestat_pid_t *cachestat_vector = NULL; + +static netdata_idx_t cachestat_hash_values[NETDATA_CACHESTAT_END]; +static netdata_idx_t *cachestat_values = NULL; + +struct netdata_static_thread cachestat_threads = {.name = "CACHESTAT KERNEL", + .config_section = NULL, + .config_name = NULL, + .env_name = NULL, + .enabled = 1, + .thread = NULL, + .init_routine = NULL, + .start_routine = NULL}; + +ebpf_local_maps_t cachestat_maps[] = {{.name = "cstat_global", .internal_input = NETDATA_CACHESTAT_END, + .user_input = 0, .type = NETDATA_EBPF_MAP_STATIC, + .map_fd = ND_EBPF_MAP_FD_NOT_INITIALIZED}, + {.name = "cstat_pid", .internal_input = ND_EBPF_DEFAULT_PID_SIZE, + .user_input = 0, + .type = NETDATA_EBPF_MAP_RESIZABLE | NETDATA_EBPF_MAP_PID, + .map_fd = ND_EBPF_MAP_FD_NOT_INITIALIZED}, + {.name = "cstat_ctrl", .internal_input = NETDATA_CONTROLLER_END, + .user_input = 0, + .type = NETDATA_EBPF_MAP_CONTROLLER, + .map_fd = ND_EBPF_MAP_FD_NOT_INITIALIZED}, + {.name = NULL, .internal_input = 0, .user_input = 0, + .type = NETDATA_EBPF_MAP_CONTROLLER, + .map_fd = ND_EBPF_MAP_FD_NOT_INITIALIZED}}; + +struct config cachestat_config = { .first_section = NULL, + .last_section = NULL, + .mutex = NETDATA_MUTEX_INITIALIZER, + .index = { .avl_tree = { .root = NULL, .compar = appconfig_section_compare }, + .rwlock = AVL_LOCK_INITIALIZER } }; + +netdata_ebpf_targets_t cachestat_targets[] = { {.name = "add_to_page_cache_lru", .mode = EBPF_LOAD_TRAMPOLINE}, + {.name = "mark_page_accessed", .mode = EBPF_LOAD_TRAMPOLINE}, + {.name = NULL, .mode = EBPF_LOAD_TRAMPOLINE}, + {.name = "mark_buffer_dirty", .mode = EBPF_LOAD_TRAMPOLINE}, + {.name = NULL, .mode = EBPF_LOAD_TRAMPOLINE}}; + +#ifdef LIBBPF_MAJOR_VERSION +#include "includes/cachestat.skel.h" // BTF code + +static struct cachestat_bpf *bpf_obj = NULL; + +/** + * Disable probe + * + * Disable all probes to use exclusively another method. + * + * @param obj is the main structure for bpf objects + */ +static void ebpf_cachestat_disable_probe(struct cachestat_bpf *obj) +{ + bpf_program__set_autoload(obj->progs.netdata_add_to_page_cache_lru_kprobe, false); + bpf_program__set_autoload(obj->progs.netdata_mark_page_accessed_kprobe, false); + bpf_program__set_autoload(obj->progs.netdata_folio_mark_dirty_kprobe, false); + bpf_program__set_autoload(obj->progs.netdata_set_page_dirty_kprobe, false); + bpf_program__set_autoload(obj->progs.netdata_account_page_dirtied_kprobe, false); + bpf_program__set_autoload(obj->progs.netdata_mark_buffer_dirty_kprobe, false); + bpf_program__set_autoload(obj->progs.netdata_release_task_kprobe, false); +} + +/* + * Disable specific probe + * + * Disable probes according the kernel version + * + * @param obj is the main structure for bpf objects + */ +static void ebpf_cachestat_disable_specific_probe(struct cachestat_bpf *obj) +{ + if (running_on_kernel >= NETDATA_EBPF_KERNEL_5_16) { + bpf_program__set_autoload(obj->progs.netdata_account_page_dirtied_kprobe, false); + bpf_program__set_autoload(obj->progs.netdata_set_page_dirty_kprobe, false); + } else if (running_on_kernel >= NETDATA_EBPF_KERNEL_5_15) { + bpf_program__set_autoload(obj->progs.netdata_folio_mark_dirty_kprobe, false); + bpf_program__set_autoload(obj->progs.netdata_account_page_dirtied_kprobe, false); + } else { + bpf_program__set_autoload(obj->progs.netdata_folio_mark_dirty_kprobe, false); + bpf_program__set_autoload(obj->progs.netdata_set_page_dirty_kprobe, false); + } +} + +/* + * Disable trampoline + * + * Disable all trampoline to use exclusively another method. + * + * @param obj is the main structure for bpf objects. + */ +static void ebpf_cachestat_disable_trampoline(struct cachestat_bpf *obj) +{ + bpf_program__set_autoload(obj->progs.netdata_add_to_page_cache_lru_fentry, false); + bpf_program__set_autoload(obj->progs.netdata_mark_page_accessed_fentry, false); + bpf_program__set_autoload(obj->progs.netdata_folio_mark_dirty_fentry, false); + bpf_program__set_autoload(obj->progs.netdata_set_page_dirty_fentry, false); + bpf_program__set_autoload(obj->progs.netdata_account_page_dirtied_fentry, false); + bpf_program__set_autoload(obj->progs.netdata_mark_buffer_dirty_fentry, false); + bpf_program__set_autoload(obj->progs.netdata_release_task_fentry, false); +} + +/* + * Disable specific trampoline + * + * Disable trampoline according to kernel version. + * + * @param obj is the main structure for bpf objects. + */ +static void ebpf_cachestat_disable_specific_trampoline(struct cachestat_bpf *obj) +{ + if (running_on_kernel >= NETDATA_EBPF_KERNEL_5_16) { + bpf_program__set_autoload(obj->progs.netdata_account_page_dirtied_fentry, false); + bpf_program__set_autoload(obj->progs.netdata_set_page_dirty_fentry, false); + } else if (running_on_kernel >= NETDATA_EBPF_KERNEL_5_15) { + bpf_program__set_autoload(obj->progs.netdata_folio_mark_dirty_fentry, false); + bpf_program__set_autoload(obj->progs.netdata_account_page_dirtied_fentry, false); + } else { + bpf_program__set_autoload(obj->progs.netdata_folio_mark_dirty_fentry, false); + bpf_program__set_autoload(obj->progs.netdata_set_page_dirty_fentry, false); + } +} + +/** + * Set trampoline target + * + * Set the targets we will monitor. + * + * @param obj is the main structure for bpf objects. + */ +static inline void netdata_set_trampoline_target(struct cachestat_bpf *obj) +{ + bpf_program__set_attach_target(obj->progs.netdata_add_to_page_cache_lru_fentry, 0, + cachestat_targets[NETDATA_KEY_CALLS_ADD_TO_PAGE_CACHE_LRU].name); + + bpf_program__set_attach_target(obj->progs.netdata_mark_page_accessed_fentry, 0, + cachestat_targets[NETDATA_KEY_CALLS_MARK_PAGE_ACCESSED].name); + + if (running_on_kernel >= NETDATA_EBPF_KERNEL_5_16) { + bpf_program__set_attach_target(obj->progs.netdata_folio_mark_dirty_fentry, 0, + cachestat_targets[NETDATA_KEY_CALLS_ACCOUNT_PAGE_DIRTIED].name); + } else if (running_on_kernel >= NETDATA_EBPF_KERNEL_5_15) { + bpf_program__set_attach_target(obj->progs.netdata_set_page_dirty_fentry, 0, + cachestat_targets[NETDATA_KEY_CALLS_ACCOUNT_PAGE_DIRTIED].name); + } else { + bpf_program__set_attach_target(obj->progs.netdata_account_page_dirtied_fentry, 0, + cachestat_targets[NETDATA_KEY_CALLS_ACCOUNT_PAGE_DIRTIED].name); + } + + bpf_program__set_attach_target(obj->progs.netdata_mark_buffer_dirty_fentry, 0, + cachestat_targets[NETDATA_KEY_CALLS_MARK_BUFFER_DIRTY].name); + + bpf_program__set_attach_target(obj->progs.netdata_release_task_fentry, 0, + EBPF_COMMON_FNCT_CLEAN_UP); +} + +/** + * Mount Attach Probe + * + * Attach probes to target + * + * @param obj is the main structure for bpf objects. + * + * @return It returns 0 on success and -1 otherwise. + */ +static int ebpf_cachestat_attach_probe(struct cachestat_bpf *obj) +{ + obj->links.netdata_add_to_page_cache_lru_kprobe = bpf_program__attach_kprobe(obj->progs.netdata_add_to_page_cache_lru_kprobe, + false, + cachestat_targets[NETDATA_KEY_CALLS_ADD_TO_PAGE_CACHE_LRU].name); + int ret = libbpf_get_error(obj->links.netdata_add_to_page_cache_lru_kprobe); + if (ret) + return -1; + + obj->links.netdata_mark_page_accessed_kprobe = bpf_program__attach_kprobe(obj->progs.netdata_mark_page_accessed_kprobe, + false, + cachestat_targets[NETDATA_KEY_CALLS_MARK_PAGE_ACCESSED].name); + ret = libbpf_get_error(obj->links.netdata_mark_page_accessed_kprobe); + if (ret) + return -1; + + if (running_on_kernel >= NETDATA_EBPF_KERNEL_5_16) { + obj->links.netdata_folio_mark_dirty_kprobe = bpf_program__attach_kprobe(obj->progs.netdata_folio_mark_dirty_kprobe, + false, + cachestat_targets[NETDATA_KEY_CALLS_ACCOUNT_PAGE_DIRTIED].name); + ret = libbpf_get_error(obj->links.netdata_folio_mark_dirty_kprobe); + } else if (running_on_kernel >= NETDATA_EBPF_KERNEL_5_15) { + obj->links.netdata_set_page_dirty_kprobe = bpf_program__attach_kprobe(obj->progs.netdata_set_page_dirty_kprobe, + false, + cachestat_targets[NETDATA_KEY_CALLS_ACCOUNT_PAGE_DIRTIED].name); + ret = libbpf_get_error(obj->links.netdata_set_page_dirty_kprobe); + } else { + obj->links.netdata_account_page_dirtied_kprobe = bpf_program__attach_kprobe(obj->progs.netdata_account_page_dirtied_kprobe, + false, + cachestat_targets[NETDATA_KEY_CALLS_ACCOUNT_PAGE_DIRTIED].name); + ret = libbpf_get_error(obj->links.netdata_account_page_dirtied_kprobe); + } + + if (ret) + return -1; + + obj->links.netdata_mark_buffer_dirty_kprobe = bpf_program__attach_kprobe(obj->progs.netdata_mark_buffer_dirty_kprobe, + false, + cachestat_targets[NETDATA_KEY_CALLS_MARK_BUFFER_DIRTY].name); + ret = libbpf_get_error(obj->links.netdata_mark_buffer_dirty_kprobe); + if (ret) + return -1; + + obj->links.netdata_release_task_kprobe = bpf_program__attach_kprobe(obj->progs.netdata_release_task_kprobe, + false, + EBPF_COMMON_FNCT_CLEAN_UP); + ret = libbpf_get_error(obj->links.netdata_release_task_kprobe); + if (ret) + return -1; + + return 0; +} + +/** + * Adjust Map Size + * + * Resize maps according input from users. + * + * @param obj is the main structure for bpf objects. + * @param em structure with configuration + */ +static void ebpf_cachestat_adjust_map_size(struct cachestat_bpf *obj, ebpf_module_t *em) +{ + ebpf_update_map_size(obj->maps.cstat_pid, &cachestat_maps[NETDATA_CACHESTAT_PID_STATS], + em, bpf_map__name(obj->maps.cstat_pid)); +} + +/** + * Set hash tables + * + * Set the values for maps according the value given by kernel. + * + * @param obj is the main structure for bpf objects. + */ +static void ebpf_cachestat_set_hash_tables(struct cachestat_bpf *obj) +{ + cachestat_maps[NETDATA_CACHESTAT_GLOBAL_STATS].map_fd = bpf_map__fd(obj->maps.cstat_global); + cachestat_maps[NETDATA_CACHESTAT_PID_STATS].map_fd = bpf_map__fd(obj->maps.cstat_pid); + cachestat_maps[NETDATA_CACHESTAT_CTRL].map_fd = bpf_map__fd(obj->maps.cstat_ctrl); +} + +/** + * Disable Release Task + * + * Disable release task when apps is not enabled. + * + * @param obj is the main structure for bpf objects. + */ +static void ebpf_cachestat_disable_release_task(struct cachestat_bpf *obj) +{ + bpf_program__set_autoload(obj->progs.netdata_release_task_kprobe, false); + bpf_program__set_autoload(obj->progs.netdata_release_task_fentry, false); +} + +/** + * Load and attach + * + * Load and attach the eBPF code in kernel. + * + * @param obj is the main structure for bpf objects. + * @param em structure with configuration + * + * @return it returns 0 on succes and -1 otherwise + */ +static inline int ebpf_cachestat_load_and_attach(struct cachestat_bpf *obj, ebpf_module_t *em) +{ + netdata_ebpf_targets_t *mt = em->targets; + netdata_ebpf_program_loaded_t test = mt[NETDATA_KEY_CALLS_ADD_TO_PAGE_CACHE_LRU].mode; + + if (test == EBPF_LOAD_TRAMPOLINE) { + ebpf_cachestat_disable_probe(obj); + ebpf_cachestat_disable_specific_trampoline(obj); + + netdata_set_trampoline_target(obj); + } else { + ebpf_cachestat_disable_trampoline(obj); + ebpf_cachestat_disable_specific_probe(obj); + } + + ebpf_cachestat_adjust_map_size(obj, em); + + if (!em->apps_charts && !em->cgroup_charts) + ebpf_cachestat_disable_release_task(obj); + + int ret = cachestat_bpf__load(obj); + if (ret) { + return ret; + } + + ret = (test == EBPF_LOAD_TRAMPOLINE) ? cachestat_bpf__attach(obj) : ebpf_cachestat_attach_probe(obj); + if (!ret) { + ebpf_cachestat_set_hash_tables(obj); + + ebpf_update_controller(cachestat_maps[NETDATA_CACHESTAT_CTRL].map_fd, em); + } + + return ret; +} +#endif +/***************************************************************** + * + * FUNCTIONS TO CLOSE THE THREAD + * + *****************************************************************/ + +/** + * Cachestat Free + * + * Cleanup variables after child threads to stop + * + * @param ptr thread data. + */ +static void ebpf_cachestat_free(ebpf_module_t *em) +{ + pthread_mutex_lock(&ebpf_exit_cleanup); + if (em->thread->enabled == NETDATA_THREAD_EBPF_RUNNING) { + em->thread->enabled = NETDATA_THREAD_EBPF_STOPPING; + pthread_mutex_unlock(&ebpf_exit_cleanup); + return; + } + pthread_mutex_unlock(&ebpf_exit_cleanup); + + ebpf_cleanup_publish_syscall(cachestat_counter_publish_aggregated); + + freez(cachestat_vector); + freez(cachestat_values); + freez(cachestat_threads.thread); + +#ifdef LIBBPF_MAJOR_VERSION + if (bpf_obj) + cachestat_bpf__destroy(bpf_obj); +#endif + pthread_mutex_lock(&ebpf_exit_cleanup); + em->thread->enabled = NETDATA_THREAD_EBPF_STOPPED; + pthread_mutex_unlock(&ebpf_exit_cleanup); +} + +/** + * Cachestat exit. + * + * Cancel child and exit. + * + * @param ptr thread data. + */ +static void ebpf_cachestat_exit(void *ptr) +{ + ebpf_module_t *em = (ebpf_module_t *)ptr; + netdata_thread_cancel(*cachestat_threads.thread); + ebpf_cachestat_free(em); +} + +/** + * Cachestat cleanup + * + * Clean up allocated addresses. + * + * @param ptr thread data. + */ +static void ebpf_cachestat_cleanup(void *ptr) +{ + ebpf_module_t *em = (ebpf_module_t *)ptr; + ebpf_cachestat_free(em); +} + +/***************************************************************** + * + * COMMON FUNCTIONS + * + *****************************************************************/ + +/** + * Update publish + * + * Update publish values before to write dimension. + * + * @param out structure that will receive data. + * @param mpa calls for mark_page_accessed during the last second. + * @param mbd calls for mark_buffer_dirty during the last second. + * @param apcl calls for add_to_page_cache_lru during the last second. + * @param apd calls for account_page_dirtied during the last second. + */ +void cachestat_update_publish(netdata_publish_cachestat_t *out, uint64_t mpa, uint64_t mbd, + uint64_t apcl, uint64_t apd) +{ + // Adapted algorithm from https://github.com/iovisor/bcc/blob/master/tools/cachestat.py#L126-L138 + NETDATA_DOUBLE total = (NETDATA_DOUBLE) (((long long)mpa) - ((long long)mbd)); + if (total < 0) + total = 0; + + NETDATA_DOUBLE misses = (NETDATA_DOUBLE) ( ((long long) apcl) - ((long long) apd) ); + if (misses < 0) + misses = 0; + + // If hits are < 0, then its possible misses are overestimate due to possibly page cache read ahead adding + // more pages than needed. In this case just assume misses as total and reset hits. + NETDATA_DOUBLE hits = total - misses; + if (hits < 0 ) { + misses = total; + hits = 0; + } + + NETDATA_DOUBLE ratio = (total > 0) ? hits/total : 1; + + out->ratio = (long long )(ratio*100); + out->hit = (long long)hits; + out->miss = (long long)misses; +} + +/** + * Save previous values + * + * Save values used this time. + * + * @param publish + */ +static void save_previous_values(netdata_publish_cachestat_t *publish) { + publish->prev.mark_page_accessed = cachestat_hash_values[NETDATA_KEY_CALLS_MARK_PAGE_ACCESSED]; + publish->prev.account_page_dirtied = cachestat_hash_values[NETDATA_KEY_CALLS_ACCOUNT_PAGE_DIRTIED]; + publish->prev.add_to_page_cache_lru = cachestat_hash_values[NETDATA_KEY_CALLS_ADD_TO_PAGE_CACHE_LRU]; + publish->prev.mark_buffer_dirty = cachestat_hash_values[NETDATA_KEY_CALLS_MARK_BUFFER_DIRTY]; +} + +/** + * Calculate statistics + * + * @param publish the structure where we will store the data. + */ +static void calculate_stats(netdata_publish_cachestat_t *publish) { + if (!publish->prev.mark_page_accessed) { + save_previous_values(publish); + return; + } + + uint64_t mpa = cachestat_hash_values[NETDATA_KEY_CALLS_MARK_PAGE_ACCESSED] - publish->prev.mark_page_accessed; + uint64_t mbd = cachestat_hash_values[NETDATA_KEY_CALLS_MARK_BUFFER_DIRTY] - publish->prev.mark_buffer_dirty; + uint64_t apcl = cachestat_hash_values[NETDATA_KEY_CALLS_ADD_TO_PAGE_CACHE_LRU] - publish->prev.add_to_page_cache_lru; + uint64_t apd = cachestat_hash_values[NETDATA_KEY_CALLS_ACCOUNT_PAGE_DIRTIED] - publish->prev.account_page_dirtied; + + save_previous_values(publish); + + // We are changing the original algorithm to have a smooth ratio. + cachestat_update_publish(publish, mpa, mbd, apcl, apd); +} + + +/***************************************************************** + * + * APPS + * + *****************************************************************/ + +/** + * Apps Accumulator + * + * Sum all values read from kernel and store in the first address. + * + * @param out the vector with read values. + */ +static void cachestat_apps_accumulator(netdata_cachestat_pid_t *out) +{ + int i, end = (running_on_kernel >= NETDATA_KERNEL_V4_15) ? ebpf_nprocs : 1; + netdata_cachestat_pid_t *total = &out[0]; + for (i = 1; i < end; i++) { + netdata_cachestat_pid_t *w = &out[i]; + total->account_page_dirtied += w->account_page_dirtied; + total->add_to_page_cache_lru += w->add_to_page_cache_lru; + total->mark_buffer_dirty += w->mark_buffer_dirty; + total->mark_page_accessed += w->mark_page_accessed; + } +} + +/** + * Save Pid values + * + * Save the current values inside the structure + * + * @param out vector used to plot charts + * @param publish vector with values read from hash tables. + */ +static inline void cachestat_save_pid_values(netdata_publish_cachestat_t *out, netdata_cachestat_pid_t *publish) +{ + if (!out->current.mark_page_accessed) { + memcpy(&out->current, &publish[0], sizeof(netdata_cachestat_pid_t)); + return; + } + + memcpy(&out->prev, &out->current, sizeof(netdata_cachestat_pid_t)); + memcpy(&out->current, &publish[0], sizeof(netdata_cachestat_pid_t)); +} + +/** + * Fill PID + * + * Fill PID structures + * + * @param current_pid pid that we are collecting data + * @param out values read from hash tables; + */ +static void cachestat_fill_pid(uint32_t current_pid, netdata_cachestat_pid_t *publish) +{ + netdata_publish_cachestat_t *curr = cachestat_pid[current_pid]; + if (!curr) { + curr = callocz(1, sizeof(netdata_publish_cachestat_t)); + cachestat_pid[current_pid] = curr; + + cachestat_save_pid_values(curr, publish); + return; + } + + cachestat_save_pid_values(curr, publish); +} + +/** + * Read APPS table + * + * Read the apps table and store data inside the structure. + */ +static void read_apps_table() +{ + netdata_cachestat_pid_t *cv = cachestat_vector; + uint32_t key; + struct pid_stat *pids = root_of_pids; + int fd = cachestat_maps[NETDATA_CACHESTAT_PID_STATS].map_fd; + size_t length = sizeof(netdata_cachestat_pid_t)*ebpf_nprocs; + while (pids) { + key = pids->pid; + + if (bpf_map_lookup_elem(fd, &key, cv)) { + pids = pids->next; + continue; + } + + cachestat_apps_accumulator(cv); + + cachestat_fill_pid(key, cv); + + // We are cleaning to avoid passing data read from one process to other. + memset(cv, 0, length); + + pids = pids->next; + } +} + +/** + * Update cgroup + * + * Update cgroup data based in + */ +static void ebpf_update_cachestat_cgroup() +{ + netdata_cachestat_pid_t *cv = cachestat_vector; + int fd = cachestat_maps[NETDATA_CACHESTAT_PID_STATS].map_fd; + size_t length = sizeof(netdata_cachestat_pid_t) * ebpf_nprocs; + + ebpf_cgroup_target_t *ect; + pthread_mutex_lock(&mutex_cgroup_shm); + for (ect = ebpf_cgroup_pids; ect; ect = ect->next) { + struct pid_on_target2 *pids; + for (pids = ect->pids; pids; pids = pids->next) { + int pid = pids->pid; + netdata_cachestat_pid_t *out = &pids->cachestat; + if (likely(cachestat_pid) && cachestat_pid[pid]) { + netdata_publish_cachestat_t *in = cachestat_pid[pid]; + + memcpy(out, &in->current, sizeof(netdata_cachestat_pid_t)); + } else { + memset(cv, 0, length); + if (bpf_map_lookup_elem(fd, &pid, cv)) { + continue; + } + + cachestat_apps_accumulator(cv); + + memcpy(out, cv, sizeof(netdata_cachestat_pid_t)); + } + } + } + pthread_mutex_unlock(&mutex_cgroup_shm); +} + +/** + * Create apps charts + * + * Call ebpf_create_chart to create the charts on apps submenu. + * + * @param em a pointer to the structure with the default values. + */ +void ebpf_cachestat_create_apps_charts(struct ebpf_module *em, void *ptr) +{ + struct target *root = ptr; + ebpf_create_charts_on_apps(NETDATA_CACHESTAT_HIT_RATIO_CHART, + "Hit ratio", + EBPF_COMMON_DIMENSION_PERCENTAGE, + NETDATA_CACHESTAT_SUBMENU, + NETDATA_EBPF_CHART_TYPE_LINE, + 20090, + ebpf_algorithms[NETDATA_EBPF_ABSOLUTE_IDX], + root, em->update_every, NETDATA_EBPF_MODULE_NAME_CACHESTAT); + + ebpf_create_charts_on_apps(NETDATA_CACHESTAT_DIRTY_CHART, + "Number of dirty pages", + EBPF_CACHESTAT_DIMENSION_PAGE, + NETDATA_CACHESTAT_SUBMENU, + NETDATA_EBPF_CHART_TYPE_STACKED, + 20091, + ebpf_algorithms[NETDATA_EBPF_INCREMENTAL_IDX], + root, em->update_every, NETDATA_EBPF_MODULE_NAME_CACHESTAT); + + ebpf_create_charts_on_apps(NETDATA_CACHESTAT_HIT_CHART, + "Number of accessed files", + EBPF_CACHESTAT_DIMENSION_HITS, + NETDATA_CACHESTAT_SUBMENU, + NETDATA_EBPF_CHART_TYPE_STACKED, + 20092, + ebpf_algorithms[NETDATA_EBPF_ABSOLUTE_IDX], + root, em->update_every, NETDATA_EBPF_MODULE_NAME_CACHESTAT); + + ebpf_create_charts_on_apps(NETDATA_CACHESTAT_MISSES_CHART, + "Files out of page cache", + EBPF_CACHESTAT_DIMENSION_MISSES, + NETDATA_CACHESTAT_SUBMENU, + NETDATA_EBPF_CHART_TYPE_STACKED, + 20093, + ebpf_algorithms[NETDATA_EBPF_ABSOLUTE_IDX], + root, em->update_every, NETDATA_EBPF_MODULE_NAME_CACHESTAT); + + em->apps_charts |= NETDATA_EBPF_APPS_FLAG_CHART_CREATED; +} + +/***************************************************************** + * + * MAIN LOOP + * + *****************************************************************/ + +/** + * Read global counter + * + * Read the table with number of calls for all functions + */ +static void read_global_table() +{ + uint32_t idx; + netdata_idx_t *val = cachestat_hash_values; + netdata_idx_t *stored = cachestat_values; + int fd = cachestat_maps[NETDATA_CACHESTAT_GLOBAL_STATS].map_fd; + + for (idx = NETDATA_KEY_CALLS_ADD_TO_PAGE_CACHE_LRU; idx < NETDATA_CACHESTAT_END; idx++) { + if (!bpf_map_lookup_elem(fd, &idx, stored)) { + int i; + int end = ebpf_nprocs; + netdata_idx_t total = 0; + for (i = 0; i < end; i++) + total += stored[i]; + + val[idx] = total; + } + } +} + +/** + * Socket read hash + * + * This is the thread callback. + * This thread is necessary, because we cannot freeze the whole plugin to read the data on very busy socket. + * + * @param ptr It is a NULL value for this thread. + * + * @return It always returns NULL. + */ +void *ebpf_cachestat_read_hash(void *ptr) +{ + netdata_thread_cleanup_push(ebpf_cachestat_cleanup, ptr); + heartbeat_t hb; + heartbeat_init(&hb); + + ebpf_module_t *em = (ebpf_module_t *)ptr; + + usec_t step = NETDATA_LATENCY_CACHESTAT_SLEEP_MS * em->update_every; + while (!ebpf_exit_plugin) { + (void)heartbeat_next(&hb, step); + + read_global_table(); + } + + netdata_thread_cleanup_pop(1); + return NULL; +} + +/** + * Send global + * + * Send global charts to Netdata + */ +static void cachestat_send_global(netdata_publish_cachestat_t *publish) +{ + calculate_stats(publish); + + netdata_publish_syscall_t *ptr = cachestat_counter_publish_aggregated; + ebpf_one_dimension_write_charts( + NETDATA_EBPF_MEMORY_GROUP, NETDATA_CACHESTAT_HIT_RATIO_CHART, ptr[NETDATA_CACHESTAT_IDX_RATIO].dimension, + publish->ratio); + + ebpf_one_dimension_write_charts( + NETDATA_EBPF_MEMORY_GROUP, NETDATA_CACHESTAT_DIRTY_CHART, ptr[NETDATA_CACHESTAT_IDX_DIRTY].dimension, + cachestat_hash_values[NETDATA_KEY_CALLS_MARK_BUFFER_DIRTY]); + + ebpf_one_dimension_write_charts( + NETDATA_EBPF_MEMORY_GROUP, NETDATA_CACHESTAT_HIT_CHART, ptr[NETDATA_CACHESTAT_IDX_HIT].dimension, publish->hit); + + ebpf_one_dimension_write_charts( + NETDATA_EBPF_MEMORY_GROUP, NETDATA_CACHESTAT_MISSES_CHART, ptr[NETDATA_CACHESTAT_IDX_MISS].dimension, + publish->miss); +} + +/** + * Cachestat sum PIDs + * + * Sum values for all PIDs associated to a group + * + * @param publish output structure. + * @param root structure with listed IPs + */ +void ebpf_cachestat_sum_pids(netdata_publish_cachestat_t *publish, struct pid_on_target *root) +{ + memcpy(&publish->prev, &publish->current,sizeof(publish->current)); + memset(&publish->current, 0, sizeof(publish->current)); + + netdata_cachestat_pid_t *dst = &publish->current; + while (root) { + int32_t pid = root->pid; + netdata_publish_cachestat_t *w = cachestat_pid[pid]; + if (w) { + netdata_cachestat_pid_t *src = &w->current; + dst->account_page_dirtied += src->account_page_dirtied; + dst->add_to_page_cache_lru += src->add_to_page_cache_lru; + dst->mark_buffer_dirty += src->mark_buffer_dirty; + dst->mark_page_accessed += src->mark_page_accessed; + } + + root = root->next; + } +} + +/** + * Send data to Netdata calling auxiliary functions. + * + * @param root the target list. +*/ +void ebpf_cache_send_apps_data(struct target *root) +{ + struct target *w; + collected_number value; + + write_begin_chart(NETDATA_APPS_FAMILY, NETDATA_CACHESTAT_HIT_RATIO_CHART); + for (w = root; w; w = w->next) { + if (unlikely(w->exposed && w->processes)) { + ebpf_cachestat_sum_pids(&w->cachestat, w->root_pid); + netdata_cachestat_pid_t *current = &w->cachestat.current; + netdata_cachestat_pid_t *prev = &w->cachestat.prev; + + uint64_t mpa = current->mark_page_accessed - prev->mark_page_accessed; + uint64_t mbd = current->mark_buffer_dirty - prev->mark_buffer_dirty; + w->cachestat.dirty = mbd; + uint64_t apcl = current->add_to_page_cache_lru - prev->add_to_page_cache_lru; + uint64_t apd = current->account_page_dirtied - prev->account_page_dirtied; + + cachestat_update_publish(&w->cachestat, mpa, mbd, apcl, apd); + value = (collected_number) w->cachestat.ratio; + // Here we are using different approach to have a chart more smooth + write_chart_dimension(w->name, value); + } + } + write_end_chart(); + + write_begin_chart(NETDATA_APPS_FAMILY, NETDATA_CACHESTAT_DIRTY_CHART); + for (w = root; w; w = w->next) { + if (unlikely(w->exposed && w->processes)) { + value = (collected_number) w->cachestat.dirty; + write_chart_dimension(w->name, value); + } + } + write_end_chart(); + + write_begin_chart(NETDATA_APPS_FAMILY, NETDATA_CACHESTAT_HIT_CHART); + for (w = root; w; w = w->next) { + if (unlikely(w->exposed && w->processes)) { + value = (collected_number) w->cachestat.hit; + write_chart_dimension(w->name, value); + } + } + write_end_chart(); + + write_begin_chart(NETDATA_APPS_FAMILY, NETDATA_CACHESTAT_MISSES_CHART); + for (w = root; w; w = w->next) { + if (unlikely(w->exposed && w->processes)) { + value = (collected_number) w->cachestat.miss; + write_chart_dimension(w->name, value); + } + } + write_end_chart(); +} + +/** + * Cachestat sum PIDs + * + * Sum values for all PIDs associated to a group + * + * @param publish output structure. + * @param root structure with listed IPs + */ +void ebpf_cachestat_sum_cgroup_pids(netdata_publish_cachestat_t *publish, struct pid_on_target2 *root) +{ + memcpy(&publish->prev, &publish->current,sizeof(publish->current)); + memset(&publish->current, 0, sizeof(publish->current)); + + netdata_cachestat_pid_t *dst = &publish->current; + while (root) { + netdata_cachestat_pid_t *src = &root->cachestat; + + dst->account_page_dirtied += src->account_page_dirtied; + dst->add_to_page_cache_lru += src->add_to_page_cache_lru; + dst->mark_buffer_dirty += src->mark_buffer_dirty; + dst->mark_page_accessed += src->mark_page_accessed; + + root = root->next; + } +} + +/** + * Calc chart values + * + * Do necessary math to plot charts. + */ +void ebpf_cachestat_calc_chart_values() +{ + ebpf_cgroup_target_t *ect; + for (ect = ebpf_cgroup_pids; ect ; ect = ect->next) { + ebpf_cachestat_sum_cgroup_pids(&ect->publish_cachestat, ect->pids); + + netdata_cachestat_pid_t *current = &ect->publish_cachestat.current; + netdata_cachestat_pid_t *prev = &ect->publish_cachestat.prev; + + uint64_t mpa = current->mark_page_accessed - prev->mark_page_accessed; + uint64_t mbd = current->mark_buffer_dirty - prev->mark_buffer_dirty; + ect->publish_cachestat.dirty = mbd; + uint64_t apcl = current->add_to_page_cache_lru - prev->add_to_page_cache_lru; + uint64_t apd = current->account_page_dirtied - prev->account_page_dirtied; + + cachestat_update_publish(&ect->publish_cachestat, mpa, mbd, apcl, apd); + } +} + +/** + * Create Systemd cachestat Charts + * + * Create charts when systemd is enabled + * + * @param update_every value to overwrite the update frequency set by the server. + **/ +static void ebpf_create_systemd_cachestat_charts(int update_every) +{ + ebpf_create_charts_on_systemd(NETDATA_CACHESTAT_HIT_RATIO_CHART, + "Hit ratio", + EBPF_COMMON_DIMENSION_PERCENTAGE, NETDATA_CACHESTAT_SUBMENU, + NETDATA_EBPF_CHART_TYPE_LINE, 21100, + ebpf_algorithms[NETDATA_EBPF_ABSOLUTE_IDX], + NETDATA_SYSTEMD_CACHESTAT_HIT_RATIO_CONTEXT, NETDATA_EBPF_MODULE_NAME_CACHESTAT, + update_every); + + ebpf_create_charts_on_systemd(NETDATA_CACHESTAT_DIRTY_CHART, + "Number of dirty pages", + EBPF_CACHESTAT_DIMENSION_PAGE, NETDATA_CACHESTAT_SUBMENU, + NETDATA_EBPF_CHART_TYPE_LINE, 21101, + ebpf_algorithms[NETDATA_EBPF_ABSOLUTE_IDX], + NETDATA_SYSTEMD_CACHESTAT_MODIFIED_CACHE_CONTEXT, NETDATA_EBPF_MODULE_NAME_CACHESTAT, + update_every); + + ebpf_create_charts_on_systemd(NETDATA_CACHESTAT_HIT_CHART, "Number of accessed files", + EBPF_CACHESTAT_DIMENSION_HITS, NETDATA_CACHESTAT_SUBMENU, + NETDATA_EBPF_CHART_TYPE_LINE, 21102, + ebpf_algorithms[NETDATA_EBPF_ABSOLUTE_IDX], + NETDATA_SYSTEMD_CACHESTAT_HIT_FILE_CONTEXT, NETDATA_EBPF_MODULE_NAME_CACHESTAT, + update_every); + + ebpf_create_charts_on_systemd(NETDATA_CACHESTAT_MISSES_CHART, "Files out of page cache", + EBPF_CACHESTAT_DIMENSION_MISSES, NETDATA_CACHESTAT_SUBMENU, + NETDATA_EBPF_CHART_TYPE_LINE, 21103, + ebpf_algorithms[NETDATA_EBPF_ABSOLUTE_IDX], + NETDATA_SYSTEMD_CACHESTAT_MISS_FILES_CONTEXT, NETDATA_EBPF_MODULE_NAME_CACHESTAT, + update_every); +} + +/** + * Send Cache Stat charts + * + * Send collected data to Netdata. + */ +static void ebpf_send_systemd_cachestat_charts() +{ + ebpf_cgroup_target_t *ect; + + write_begin_chart(NETDATA_SERVICE_FAMILY, NETDATA_CACHESTAT_HIT_RATIO_CHART); + for (ect = ebpf_cgroup_pids; ect; ect = ect->next) { + if (unlikely(ect->systemd) && unlikely(ect->updated)) { + write_chart_dimension(ect->name, (long long)ect->publish_cachestat.ratio); + } + } + write_end_chart(); + + write_begin_chart(NETDATA_SERVICE_FAMILY, NETDATA_CACHESTAT_DIRTY_CHART); + for (ect = ebpf_cgroup_pids; ect; ect = ect->next) { + if (unlikely(ect->systemd) && unlikely(ect->updated)) { + write_chart_dimension(ect->name, (long long)ect->publish_cachestat.dirty); + } + } + write_end_chart(); + + write_begin_chart(NETDATA_SERVICE_FAMILY, NETDATA_CACHESTAT_HIT_CHART); + for (ect = ebpf_cgroup_pids; ect; ect = ect->next) { + if (unlikely(ect->systemd) && unlikely(ect->updated)) { + write_chart_dimension(ect->name, (long long)ect->publish_cachestat.hit); + } + } + write_end_chart(); + + write_begin_chart(NETDATA_SERVICE_FAMILY, NETDATA_CACHESTAT_MISSES_CHART); + for (ect = ebpf_cgroup_pids; ect; ect = ect->next) { + if (unlikely(ect->systemd) && unlikely(ect->updated)) { + write_chart_dimension(ect->name, (long long)ect->publish_cachestat.miss); + } + } + write_end_chart(); +} + +/** + * Send Directory Cache charts + * + * Send collected data to Netdata. + */ +static void ebpf_send_specific_cachestat_data(char *type, netdata_publish_cachestat_t *npc) +{ + write_begin_chart(type, NETDATA_CACHESTAT_HIT_RATIO_CHART); + write_chart_dimension(cachestat_counter_publish_aggregated[NETDATA_CACHESTAT_IDX_RATIO].name, (long long)npc->ratio); + write_end_chart(); + + write_begin_chart(type, NETDATA_CACHESTAT_DIRTY_CHART); + write_chart_dimension(cachestat_counter_publish_aggregated[NETDATA_CACHESTAT_IDX_DIRTY].name, (long long)npc->dirty); + write_end_chart(); + + write_begin_chart(type, NETDATA_CACHESTAT_HIT_CHART); + write_chart_dimension(cachestat_counter_publish_aggregated[NETDATA_CACHESTAT_IDX_HIT].name, (long long)npc->hit); + write_end_chart(); + + write_begin_chart(type, NETDATA_CACHESTAT_MISSES_CHART); + write_chart_dimension(cachestat_counter_publish_aggregated[NETDATA_CACHESTAT_IDX_MISS].name, (long long)npc->miss); + write_end_chart(); +} + +/** + * Create specific cache Stat charts + * + * Create charts for cgroup/application. + * + * @param type the chart type. + * @param update_every value to overwrite the update frequency set by the server. + */ +static void ebpf_create_specific_cachestat_charts(char *type, int update_every) +{ + ebpf_create_chart(type, NETDATA_CACHESTAT_HIT_RATIO_CHART, + "Hit ratio", + EBPF_COMMON_DIMENSION_PERCENTAGE, NETDATA_CACHESTAT_CGROUP_SUBMENU, + NETDATA_CGROUP_CACHESTAT_HIT_RATIO_CONTEXT, + NETDATA_EBPF_CHART_TYPE_LINE, NETDATA_CHART_PRIO_CGROUPS_CONTAINERS + 5200, + ebpf_create_global_dimension, + cachestat_counter_publish_aggregated, 1, update_every, NETDATA_EBPF_MODULE_NAME_CACHESTAT); + + ebpf_create_chart(type, NETDATA_CACHESTAT_DIRTY_CHART, + "Number of dirty pages", + EBPF_CACHESTAT_DIMENSION_PAGE, NETDATA_CACHESTAT_CGROUP_SUBMENU, + NETDATA_CGROUP_CACHESTAT_MODIFIED_CACHE_CONTEXT, + NETDATA_EBPF_CHART_TYPE_LINE, NETDATA_CHART_PRIO_CGROUPS_CONTAINERS + 5201, + ebpf_create_global_dimension, + &cachestat_counter_publish_aggregated[NETDATA_CACHESTAT_IDX_DIRTY], 1, + update_every, NETDATA_EBPF_MODULE_NAME_CACHESTAT); + + ebpf_create_chart(type, NETDATA_CACHESTAT_HIT_CHART, + "Number of accessed files", + EBPF_CACHESTAT_DIMENSION_HITS, NETDATA_CACHESTAT_CGROUP_SUBMENU, + NETDATA_CGROUP_CACHESTAT_HIT_FILES_CONTEXT, + NETDATA_EBPF_CHART_TYPE_LINE, NETDATA_CHART_PRIO_CGROUPS_CONTAINERS + 5202, + ebpf_create_global_dimension, + &cachestat_counter_publish_aggregated[NETDATA_CACHESTAT_IDX_HIT], 1, + update_every, NETDATA_EBPF_MODULE_NAME_CACHESTAT); + + ebpf_create_chart(type, NETDATA_CACHESTAT_MISSES_CHART, + "Files out of page cache", + EBPF_CACHESTAT_DIMENSION_MISSES, NETDATA_CACHESTAT_CGROUP_SUBMENU, + NETDATA_CGROUP_CACHESTAT_MISS_FILES_CONTEXT, + NETDATA_EBPF_CHART_TYPE_LINE, NETDATA_CHART_PRIO_CGROUPS_CONTAINERS + 5203, + ebpf_create_global_dimension, + &cachestat_counter_publish_aggregated[NETDATA_CACHESTAT_IDX_MISS], 1, + update_every, NETDATA_EBPF_MODULE_NAME_CACHESTAT); +} + +/** + * Obsolete specific cache stat charts + * + * Obsolete charts for cgroup/application. + * + * @param type the chart type. + * @param update_every value to overwrite the update frequency set by the server. + */ +static void ebpf_obsolete_specific_cachestat_charts(char *type, int update_every) +{ + ebpf_write_chart_obsolete(type, NETDATA_CACHESTAT_HIT_RATIO_CHART, + "Hit ratio", + EBPF_COMMON_DIMENSION_PERCENTAGE, NETDATA_CACHESTAT_SUBMENU, + NETDATA_EBPF_CHART_TYPE_LINE, NETDATA_CGROUP_CACHESTAT_HIT_RATIO_CONTEXT, + NETDATA_CHART_PRIO_CGROUPS_CONTAINERS + 5200, update_every); + + ebpf_write_chart_obsolete(type, NETDATA_CACHESTAT_DIRTY_CHART, + "Number of dirty pages", + EBPF_CACHESTAT_DIMENSION_PAGE, NETDATA_CACHESTAT_SUBMENU, + NETDATA_EBPF_CHART_TYPE_LINE, NETDATA_CGROUP_CACHESTAT_MODIFIED_CACHE_CONTEXT, + NETDATA_CHART_PRIO_CGROUPS_CONTAINERS + 5201, update_every); + + ebpf_write_chart_obsolete(type, NETDATA_CACHESTAT_HIT_CHART, + "Number of accessed files", + EBPF_CACHESTAT_DIMENSION_HITS, NETDATA_CACHESTAT_SUBMENU, + NETDATA_EBPF_CHART_TYPE_LINE, NETDATA_CGROUP_CACHESTAT_HIT_FILES_CONTEXT, + NETDATA_CHART_PRIO_CGROUPS_CONTAINERS + 5202, update_every); + + ebpf_write_chart_obsolete(type, NETDATA_CACHESTAT_MISSES_CHART, + "Files out of page cache", + EBPF_CACHESTAT_DIMENSION_MISSES, NETDATA_CACHESTAT_SUBMENU, + NETDATA_EBPF_CHART_TYPE_LINE, NETDATA_CGROUP_CACHESTAT_MISS_FILES_CONTEXT, + NETDATA_CHART_PRIO_CGROUPS_CONTAINERS + 5203, update_every); +} + +/** + * Send data to Netdata calling auxiliary functions. + * + * @param update_every value to overwrite the update frequency set by the server. +*/ +void ebpf_cachestat_send_cgroup_data(int update_every) +{ + if (!ebpf_cgroup_pids) + return; + + pthread_mutex_lock(&mutex_cgroup_shm); + ebpf_cgroup_target_t *ect; + ebpf_cachestat_calc_chart_values(); + + int has_systemd = shm_ebpf_cgroup.header->systemd_enabled; + if (has_systemd) { + if (send_cgroup_chart) { + ebpf_create_systemd_cachestat_charts(update_every); + } + + ebpf_send_systemd_cachestat_charts(); + } + + for (ect = ebpf_cgroup_pids; ect ; ect = ect->next) { + if (ect->systemd) + continue; + + if (!(ect->flags & NETDATA_EBPF_CGROUP_HAS_CACHESTAT_CHART) && ect->updated) { + ebpf_create_specific_cachestat_charts(ect->name, update_every); + ect->flags |= NETDATA_EBPF_CGROUP_HAS_CACHESTAT_CHART; + } + + if (ect->flags & NETDATA_EBPF_CGROUP_HAS_CACHESTAT_CHART) { + if (ect->updated) { + ebpf_send_specific_cachestat_data(ect->name, &ect->publish_cachestat); + } else { + ebpf_obsolete_specific_cachestat_charts(ect->name, update_every); + ect->flags &= ~NETDATA_EBPF_CGROUP_HAS_CACHESTAT_CHART; + } + } + } + + pthread_mutex_unlock(&mutex_cgroup_shm); +} + +/** +* Main loop for this collector. +*/ +static void cachestat_collector(ebpf_module_t *em) +{ + cachestat_threads.thread = callocz(1, sizeof(netdata_thread_t)); + cachestat_threads.start_routine = ebpf_cachestat_read_hash; + + netdata_thread_create(cachestat_threads.thread, cachestat_threads.name, NETDATA_THREAD_OPTION_DEFAULT, + ebpf_cachestat_read_hash, em); + + netdata_publish_cachestat_t publish; + memset(&publish, 0, sizeof(publish)); + int cgroups = em->cgroup_charts; + int update_every = em->update_every; + heartbeat_t hb; + heartbeat_init(&hb); + usec_t step = update_every * USEC_PER_SEC; + //This will be cancelled by its parent + while (!ebpf_exit_plugin) { + (void)heartbeat_next(&hb, step); + if (ebpf_exit_plugin) + break; + + netdata_apps_integration_flags_t apps = em->apps_charts; + pthread_mutex_lock(&collect_data_mutex); + if (apps) + read_apps_table(); + + if (cgroups) + ebpf_update_cachestat_cgroup(); + + pthread_mutex_lock(&lock); + + cachestat_send_global(&publish); + + if (apps & NETDATA_EBPF_APPS_FLAG_CHART_CREATED) + ebpf_cache_send_apps_data(apps_groups_root_target); + + if (cgroups) + ebpf_cachestat_send_cgroup_data(update_every); + + pthread_mutex_unlock(&lock); + pthread_mutex_unlock(&collect_data_mutex); + } +} + +/***************************************************************** + * + * INITIALIZE THREAD + * + *****************************************************************/ + +/** + * Create global charts + * + * Call ebpf_create_chart to create the charts for the collector. + * + * @param em a pointer to `struct ebpf_module` + */ +static void ebpf_create_memory_charts(ebpf_module_t *em) +{ + ebpf_create_chart(NETDATA_EBPF_MEMORY_GROUP, NETDATA_CACHESTAT_HIT_RATIO_CHART, + "Hit ratio", + EBPF_COMMON_DIMENSION_PERCENTAGE, NETDATA_CACHESTAT_SUBMENU, + NULL, + NETDATA_EBPF_CHART_TYPE_LINE, + 21100, + ebpf_create_global_dimension, + cachestat_counter_publish_aggregated, 1, em->update_every, NETDATA_EBPF_MODULE_NAME_CACHESTAT); + + ebpf_create_chart(NETDATA_EBPF_MEMORY_GROUP, NETDATA_CACHESTAT_DIRTY_CHART, + "Number of dirty pages", + EBPF_CACHESTAT_DIMENSION_PAGE, NETDATA_CACHESTAT_SUBMENU, + NULL, + NETDATA_EBPF_CHART_TYPE_LINE, + 21101, + ebpf_create_global_dimension, + &cachestat_counter_publish_aggregated[NETDATA_CACHESTAT_IDX_DIRTY], 1, + em->update_every, NETDATA_EBPF_MODULE_NAME_CACHESTAT); + + ebpf_create_chart(NETDATA_EBPF_MEMORY_GROUP, NETDATA_CACHESTAT_HIT_CHART, + "Number of accessed files", + EBPF_CACHESTAT_DIMENSION_HITS, NETDATA_CACHESTAT_SUBMENU, + NULL, + NETDATA_EBPF_CHART_TYPE_LINE, + 21102, + ebpf_create_global_dimension, + &cachestat_counter_publish_aggregated[NETDATA_CACHESTAT_IDX_HIT], 1, + em->update_every, NETDATA_EBPF_MODULE_NAME_CACHESTAT); + + ebpf_create_chart(NETDATA_EBPF_MEMORY_GROUP, NETDATA_CACHESTAT_MISSES_CHART, + "Files out of page cache", + EBPF_CACHESTAT_DIMENSION_MISSES, NETDATA_CACHESTAT_SUBMENU, + NULL, + NETDATA_EBPF_CHART_TYPE_LINE, + 21103, + ebpf_create_global_dimension, + &cachestat_counter_publish_aggregated[NETDATA_CACHESTAT_IDX_MISS], 1, + em->update_every, NETDATA_EBPF_MODULE_NAME_CACHESTAT); + + fflush(stdout); +} + +/** + * Allocate vectors used with this thread. + * + * We are not testing the return, because callocz does this and shutdown the software + * case it was not possible to allocate. + * + * @param apps is apps enabled? + */ +static void ebpf_cachestat_allocate_global_vectors(int apps) +{ + if (apps) + cachestat_pid = callocz((size_t)pid_max, sizeof(netdata_publish_cachestat_t *)); + + cachestat_vector = callocz((size_t)ebpf_nprocs, sizeof(netdata_cachestat_pid_t)); + + cachestat_values = callocz((size_t)ebpf_nprocs, sizeof(netdata_idx_t)); + + memset(cachestat_hash_values, 0, NETDATA_CACHESTAT_END * sizeof(netdata_idx_t)); + memset(cachestat_counter_aggregated_data, 0, NETDATA_CACHESTAT_END * sizeof(netdata_syscall_stat_t)); + memset(cachestat_counter_publish_aggregated, 0, NETDATA_CACHESTAT_END * sizeof(netdata_publish_syscall_t)); +} + +/***************************************************************** + * + * MAIN THREAD + * + *****************************************************************/ + +/** + * Update Internal value + * + * Update values used during runtime. + */ +static void ebpf_cachestat_set_internal_value() +{ + static char *account_page[] = { "account_page_dirtied", "__set_page_dirty", "__folio_mark_dirty" }; + if (running_on_kernel >= NETDATA_EBPF_KERNEL_5_16) + cachestat_targets[NETDATA_KEY_CALLS_ACCOUNT_PAGE_DIRTIED].name = account_page[NETDATA_CACHESTAT_FOLIO_DIRTY]; + else if (running_on_kernel >= NETDATA_EBPF_KERNEL_5_15) + cachestat_targets[NETDATA_KEY_CALLS_ACCOUNT_PAGE_DIRTIED].name = account_page[NETDATA_CACHESTAT_SET_PAGE_DIRTY]; + else + cachestat_targets[NETDATA_KEY_CALLS_ACCOUNT_PAGE_DIRTIED].name = account_page[NETDATA_CACHESTAT_ACCOUNT_PAGE_DIRTY]; +} + +/* + * Load BPF + * + * Load BPF files. + * + * @param em the structure with configuration + */ +static int ebpf_cachestat_load_bpf(ebpf_module_t *em) +{ + int ret = 0; + ebpf_adjust_apps_cgroup(em, em->targets[NETDATA_KEY_CALLS_ADD_TO_PAGE_CACHE_LRU].mode); + if (em->load & EBPF_LOAD_LEGACY) { + em->probe_links = ebpf_load_program(ebpf_plugin_dir, em, running_on_kernel, isrh, &em->objects); + if (!em->probe_links) { + ret = -1; + } + } +#ifdef LIBBPF_MAJOR_VERSION + else { + bpf_obj = cachestat_bpf__open(); + if (!bpf_obj) + ret = -1; + else + ret = ebpf_cachestat_load_and_attach(bpf_obj, em); + } +#endif + + if (ret) + error("%s %s", EBPF_DEFAULT_ERROR_MSG, em->thread_name); + + return ret; +} + +/** + * Cachestat thread + * + * Thread used to make cachestat thread + * + * @param ptr a pointer to `struct ebpf_module` + * + * @return It always return NULL + */ +void *ebpf_cachestat_thread(void *ptr) +{ + netdata_thread_cleanup_push(ebpf_cachestat_exit, ptr); + + ebpf_module_t *em = (ebpf_module_t *)ptr; + em->maps = cachestat_maps; + + ebpf_update_pid_table(&cachestat_maps[NETDATA_CACHESTAT_PID_STATS], em); + + ebpf_cachestat_set_internal_value(); + +#ifdef LIBBPF_MAJOR_VERSION + ebpf_adjust_thread_load(em, default_btf); +#endif + if (ebpf_cachestat_load_bpf(em)) { + em->thread->enabled = NETDATA_THREAD_EBPF_STOPPED; + goto endcachestat; + } + + ebpf_cachestat_allocate_global_vectors(em->apps_charts); + + int algorithms[NETDATA_CACHESTAT_END] = { + NETDATA_EBPF_ABSOLUTE_IDX, NETDATA_EBPF_INCREMENTAL_IDX, NETDATA_EBPF_ABSOLUTE_IDX, NETDATA_EBPF_ABSOLUTE_IDX + }; + + ebpf_global_labels(cachestat_counter_aggregated_data, cachestat_counter_publish_aggregated, + cachestat_counter_dimension_name, cachestat_counter_dimension_name, + algorithms, NETDATA_CACHESTAT_END); + + pthread_mutex_lock(&lock); + ebpf_update_stats(&plugin_statistics, em); + ebpf_create_memory_charts(em); + pthread_mutex_unlock(&lock); + + cachestat_collector(em); + +endcachestat: + ebpf_update_disabled_plugin_stats(em); + + netdata_thread_cleanup_pop(1); + return NULL; +} diff --git a/collectors/ebpf.plugin/ebpf_cachestat.h b/collectors/ebpf.plugin/ebpf_cachestat.h new file mode 100644 index 0000000..07f0745 --- /dev/null +++ b/collectors/ebpf.plugin/ebpf_cachestat.h @@ -0,0 +1,90 @@ +// SPDX-License-Identifier: GPL-3.0-or-later + +#ifndef NETDATA_EBPF_CACHESTAT_H +#define NETDATA_EBPF_CACHESTAT_H 1 + +// Module name +#define NETDATA_EBPF_MODULE_NAME_CACHESTAT "cachestat" + +// charts +#define NETDATA_CACHESTAT_HIT_RATIO_CHART "cachestat_ratio" +#define NETDATA_CACHESTAT_DIRTY_CHART "cachestat_dirties" +#define NETDATA_CACHESTAT_HIT_CHART "cachestat_hits" +#define NETDATA_CACHESTAT_MISSES_CHART "cachestat_misses" + +#define NETDATA_CACHESTAT_SUBMENU "page_cache" +#define NETDATA_CACHESTAT_CGROUP_SUBMENU "page cache (eBPF)" + +#define EBPF_CACHESTAT_DIMENSION_PAGE "pages/s" +#define EBPF_CACHESTAT_DIMENSION_HITS "hits/s" +#define EBPF_CACHESTAT_DIMENSION_MISSES "misses/s" + +#define NETDATA_LATENCY_CACHESTAT_SLEEP_MS 600000ULL + +// configuration file +#define NETDATA_CACHESTAT_CONFIG_FILE "cachestat.conf" + +// Contexts +#define NETDATA_CGROUP_CACHESTAT_HIT_RATIO_CONTEXT "cgroup.cachestat_ratio" +#define NETDATA_CGROUP_CACHESTAT_MODIFIED_CACHE_CONTEXT "cgroup.cachestat_dirties" +#define NETDATA_CGROUP_CACHESTAT_HIT_FILES_CONTEXT "cgroup.cachestat_hits" +#define NETDATA_CGROUP_CACHESTAT_MISS_FILES_CONTEXT "cgroup.cachestat_misses" + +#define NETDATA_SYSTEMD_CACHESTAT_HIT_RATIO_CONTEXT "services.cachestat_ratio" +#define NETDATA_SYSTEMD_CACHESTAT_MODIFIED_CACHE_CONTEXT "services.cachestat_dirties" +#define NETDATA_SYSTEMD_CACHESTAT_HIT_FILE_CONTEXT "services.cachestat_hits" +#define NETDATA_SYSTEMD_CACHESTAT_MISS_FILES_CONTEXT "services.cachestat_misses" + +// variables +enum cachestat_counters { + NETDATA_KEY_CALLS_ADD_TO_PAGE_CACHE_LRU, + NETDATA_KEY_CALLS_MARK_PAGE_ACCESSED, + NETDATA_KEY_CALLS_ACCOUNT_PAGE_DIRTIED, + NETDATA_KEY_CALLS_MARK_BUFFER_DIRTY, + + NETDATA_CACHESTAT_END +}; + +enum cachestat_account_dirty_pages { + NETDATA_CACHESTAT_ACCOUNT_PAGE_DIRTY, + NETDATA_CACHESTAT_SET_PAGE_DIRTY, + NETDATA_CACHESTAT_FOLIO_DIRTY +}; + +enum cachestat_indexes { + NETDATA_CACHESTAT_IDX_RATIO, + NETDATA_CACHESTAT_IDX_DIRTY, + NETDATA_CACHESTAT_IDX_HIT, + NETDATA_CACHESTAT_IDX_MISS +}; + +enum cachestat_tables { + NETDATA_CACHESTAT_GLOBAL_STATS, + NETDATA_CACHESTAT_PID_STATS, + NETDATA_CACHESTAT_CTRL +}; + +typedef struct netdata_publish_cachestat_pid { + uint64_t add_to_page_cache_lru; + uint64_t mark_page_accessed; + uint64_t account_page_dirtied; + uint64_t mark_buffer_dirty; +} netdata_cachestat_pid_t; + +typedef struct netdata_publish_cachestat { + long long ratio; + long long dirty; + long long hit; + long long miss; + + netdata_cachestat_pid_t current; + netdata_cachestat_pid_t prev; +} netdata_publish_cachestat_t; + +void *ebpf_cachestat_thread(void *ptr); + +extern struct config cachestat_config; +extern netdata_ebpf_targets_t cachestat_targets[]; +extern ebpf_local_maps_t cachestat_maps[]; + +#endif // NETDATA_EBPF_CACHESTAT_H diff --git a/collectors/ebpf.plugin/ebpf_cgroup.c b/collectors/ebpf.plugin/ebpf_cgroup.c new file mode 100644 index 0000000..42c0453 --- /dev/null +++ b/collectors/ebpf.plugin/ebpf_cgroup.c @@ -0,0 +1,317 @@ +// SPDX-License-Identifier: GPL-3.0-or-later + +#include <sys/resource.h> + +#include "ebpf.h" +#include "ebpf_cgroup.h" + +ebpf_cgroup_target_t *ebpf_cgroup_pids = NULL; +int send_cgroup_chart = 0; + +// -------------------------------------------------------------------------------------------------------------------- +// Map shared memory + +/** + * Map Shared Memory locally + * + * Map the shared memory for current process + * + * @param fd file descriptor returned after shm_open was called. + * @param length length of the shared memory + * + * @return It returns a pointer to the region mapped. + */ +static inline void *ebpf_cgroup_map_shm_locally(int fd, size_t length) +{ + void *value; + + value = mmap(NULL, length, PROT_READ | PROT_WRITE, MAP_SHARED, fd, 0); + if (!value) { + error("Cannot map shared memory used between eBPF and cgroup, integration between processes won't happen"); + close(shm_fd_ebpf_cgroup); + shm_fd_ebpf_cgroup = -1; + shm_unlink(NETDATA_SHARED_MEMORY_EBPF_CGROUP_NAME); + } + + return value; +} + +/** + * Map cgroup shared memory + * + * Map cgroup shared memory from cgroup to plugin + */ +void ebpf_map_cgroup_shared_memory() +{ + static int limit_try = 0; + static time_t next_try = 0; + + if (shm_ebpf_cgroup.header || limit_try > NETDATA_EBPF_CGROUP_MAX_TRIES) + return; + + time_t curr_time = time(NULL); + if (curr_time < next_try) + return; + + limit_try++; + next_try = curr_time + NETDATA_EBPF_CGROUP_NEXT_TRY_SEC; + + shm_fd_ebpf_cgroup = shm_open(NETDATA_SHARED_MEMORY_EBPF_CGROUP_NAME, O_RDWR, 0660); + if (shm_fd_ebpf_cgroup < 0) { + if (limit_try == NETDATA_EBPF_CGROUP_MAX_TRIES) + error("Shared memory was not initialized, integration between processes won't happen."); + + return; + } + + // Map only header + shm_ebpf_cgroup.header = (netdata_ebpf_cgroup_shm_header_t *) ebpf_cgroup_map_shm_locally(shm_fd_ebpf_cgroup, + sizeof(netdata_ebpf_cgroup_shm_header_t)); + if (!shm_ebpf_cgroup.header) { + limit_try = NETDATA_EBPF_CGROUP_MAX_TRIES + 1; + return; + } + + size_t length = shm_ebpf_cgroup.header->body_length; + + munmap(shm_ebpf_cgroup.header, sizeof(netdata_ebpf_cgroup_shm_header_t)); + + shm_ebpf_cgroup.header = (netdata_ebpf_cgroup_shm_header_t *)ebpf_cgroup_map_shm_locally(shm_fd_ebpf_cgroup, length); + if (!shm_ebpf_cgroup.header) { + limit_try = NETDATA_EBPF_CGROUP_MAX_TRIES + 1; + return; + } + shm_ebpf_cgroup.body = (netdata_ebpf_cgroup_shm_body_t *) ((char *)shm_ebpf_cgroup.header + + sizeof(netdata_ebpf_cgroup_shm_header_t)); + + shm_sem_ebpf_cgroup = sem_open(NETDATA_NAMED_SEMAPHORE_EBPF_CGROUP_NAME, O_CREAT, 0660, 1); + + if (shm_sem_ebpf_cgroup == SEM_FAILED) { + error("Cannot create semaphore, integration between eBPF and cgroup won't happen"); + munmap(shm_ebpf_cgroup.header, length); + shm_ebpf_cgroup.header = NULL; + close(shm_fd_ebpf_cgroup); + shm_fd_ebpf_cgroup = -1; + shm_unlink(NETDATA_SHARED_MEMORY_EBPF_CGROUP_NAME); + } +} + +// -------------------------------------------------------------------------------------------------------------------- +// Close and Cleanup + +/** + * Clean Specific cgroup pid + * + * Clean all PIDs associated with cgroup. + * + * @param pt structure pid on target that will have your PRs removed + */ +static inline void ebpf_clean_specific_cgroup_pids(struct pid_on_target2 *pt) +{ + while (pt) { + struct pid_on_target2 *next_pid = pt->next; + + freez(pt); + pt = next_pid; + } +} + +/** + * Remove Cgroup Update Target Update List + * + * Remove from cgroup target and update the link list + */ +static void ebpf_remove_cgroup_target_update_list() +{ + ebpf_cgroup_target_t *next, *ect = ebpf_cgroup_pids; + ebpf_cgroup_target_t *prev = ebpf_cgroup_pids; + while (ect) { + next = ect->next; + if (!ect->updated) { + if (ect == ebpf_cgroup_pids) { + ebpf_cgroup_pids = next; + prev = next; + } else { + prev->next = next; + } + + ebpf_clean_specific_cgroup_pids(ect->pids); + freez(ect); + } else { + prev = ect; + } + + ect = next; + } +} + +// -------------------------------------------------------------------------------------------------------------------- +// Fill variables + +/** + * Set Target Data + * + * Set local variable values according shared memory information. + * + * @param out local output variable. + * @param ptr input from shared memory. + */ +static inline void ebpf_cgroup_set_target_data(ebpf_cgroup_target_t *out, netdata_ebpf_cgroup_shm_body_t *ptr) +{ + out->hash = ptr->hash; + snprintfz(out->name, 255, "%s", ptr->name); + out->systemd = ptr->options & CGROUP_OPTIONS_SYSTEM_SLICE_SERVICE; + out->updated = 1; +} + +/** + * Find or create + * + * Find the structure inside the link list or allocate and link when it is not present. + * + * @param ptr Input from shared memory. + * + * @return It returns a pointer for the structure associated with the input. + */ +static ebpf_cgroup_target_t * ebpf_cgroup_find_or_create(netdata_ebpf_cgroup_shm_body_t *ptr) +{ + ebpf_cgroup_target_t *ect, *prev; + for (ect = ebpf_cgroup_pids, prev = ebpf_cgroup_pids; ect; prev = ect, ect = ect->next) { + if (ect->hash == ptr->hash && !strcmp(ect->name, ptr->name)) { + ect->updated = 1; + return ect; + } + } + + ebpf_cgroup_target_t *new_ect = callocz(1, sizeof(ebpf_cgroup_target_t)); + + ebpf_cgroup_set_target_data(new_ect, ptr); + if (!ebpf_cgroup_pids) { + ebpf_cgroup_pids = new_ect; + } else { + prev->next = new_ect; + } + + return new_ect; +} + +/** + * Update pid link list + * + * Update PIDs list associated with specific cgroup. + * + * @param ect cgroup structure where pids will be stored + * @param path file with PIDs associated to cgroup. + */ +static void ebpf_update_pid_link_list(ebpf_cgroup_target_t *ect, char *path) +{ + procfile *ff = procfile_open_no_log(path, " \t:", PROCFILE_FLAG_DEFAULT); + if (!ff) + return; + + ff = procfile_readall(ff); + if (!ff) + return; + + size_t lines = procfile_lines(ff), l; + for (l = 0; l < lines ;l++) { + int pid = (int)str2l(procfile_lineword(ff, l, 0)); + if (pid) { + struct pid_on_target2 *pt, *prev; + for (pt = ect->pids, prev = ect->pids; pt; prev = pt, pt = pt->next) { + if (pt->pid == pid) + break; + } + + if (!pt) { + struct pid_on_target2 *w = callocz(1, sizeof(struct pid_on_target2)); + w->pid = pid; + if (!ect->pids) + ect->pids = w; + else + prev->next = w; + } + } + } + + procfile_close(ff); +} + +/** + * Set remove var + * + * Set variable remove. If this variable is not reset, the structure will be removed from link list. + */ +void ebpf_reset_updated_var() + { + ebpf_cgroup_target_t *ect; + for (ect = ebpf_cgroup_pids; ect; ect = ect->next) { + ect->updated = 0; + } + } + +/** + * Parse cgroup shared memory + * + * This function is responsible to copy necessary data from shared memory to local memory. + */ +void ebpf_parse_cgroup_shm_data() +{ + static int previous = 0; + if (shm_ebpf_cgroup.header) { + sem_wait(shm_sem_ebpf_cgroup); + int i, end = shm_ebpf_cgroup.header->cgroup_root_count; + + pthread_mutex_lock(&mutex_cgroup_shm); + + ebpf_remove_cgroup_target_update_list(); + + ebpf_reset_updated_var(); + + for (i = 0; i < end; i++) { + netdata_ebpf_cgroup_shm_body_t *ptr = &shm_ebpf_cgroup.body[i]; + if (ptr->enabled) { + ebpf_cgroup_target_t *ect = ebpf_cgroup_find_or_create(ptr); + ebpf_update_pid_link_list(ect, ptr->path); + } + } + send_cgroup_chart = previous != shm_ebpf_cgroup.header->cgroup_root_count; + previous = shm_ebpf_cgroup.header->cgroup_root_count; +#ifdef NETDATA_DEV_MODE + error("Updating cgroup %d (Previous: %d, Current: %d)", send_cgroup_chart, previous, shm_ebpf_cgroup.header->cgroup_root_count); +#endif + pthread_mutex_unlock(&mutex_cgroup_shm); + + sem_post(shm_sem_ebpf_cgroup); + } +} + +// -------------------------------------------------------------------------------------------------------------------- +// Create charts + +/** + * Create charts on systemd submenu + * + * @param id the chart id + * @param title the value displayed on vertical axis. + * @param units the value displayed on vertical axis. + * @param family Submenu that the chart will be attached on dashboard. + * @param charttype chart type + * @param order the chart order + * @param algorithm the algorithm used by dimension + * @param context add context for chart + * @param module chart module name, this is the eBPF thread. + * @param update_every value to overwrite the update frequency set by the server. + */ +void ebpf_create_charts_on_systemd(char *id, char *title, char *units, char *family, char *charttype, int order, + char *algorithm, char *context, char *module, int update_every) +{ + ebpf_cgroup_target_t *w; + ebpf_write_chart_cmd(NETDATA_SERVICE_FAMILY, id, title, units, family, charttype, context, + order, update_every, module); + + for (w = ebpf_cgroup_pids; w; w = w->next) { + if (unlikely(w->systemd) && unlikely(w->updated)) + fprintf(stdout, "DIMENSION %s '' %s 1 1\n", w->name, algorithm); + } +} diff --git a/collectors/ebpf.plugin/ebpf_cgroup.h b/collectors/ebpf.plugin/ebpf_cgroup.h new file mode 100644 index 0000000..19da7fc --- /dev/null +++ b/collectors/ebpf.plugin/ebpf_cgroup.h @@ -0,0 +1,69 @@ +// SPDX-License-Identifier: GPL-3.0-or-later + +#ifndef NETDATA_EBPF_CGROUP_H +#define NETDATA_EBPF_CGROUP_H 1 + +#define NETDATA_EBPF_CGROUP_MAX_TRIES 3 +#define NETDATA_EBPF_CGROUP_NEXT_TRY_SEC 30 + +#include "ebpf.h" +#include "ebpf_apps.h" + +#define NETDATA_SERVICE_FAMILY "services" + +struct pid_on_target2 { + int32_t pid; + int updated; + + netdata_publish_swap_t swap; + netdata_fd_stat_t fd; + netdata_publish_vfs_t vfs; + ebpf_process_stat_t ps; + netdata_dcstat_pid_t dc; + netdata_publish_shm_t shm; + ebpf_bandwidth_t socket; + netdata_cachestat_pid_t cachestat; + + struct pid_on_target2 *next; +}; + +enum ebpf_cgroup_flags { + NETDATA_EBPF_CGROUP_HAS_PROCESS_CHART = 1, + NETDATA_EBPF_CGROUP_HAS_SWAP_CHART = 1<<2, + NETDATA_EBPF_CGROUP_HAS_SOCKET_CHART = 1<<3, + NETDATA_EBPF_CGROUP_HAS_FD_CHART = 1<<4, + NETDATA_EBPF_CGROUP_HAS_VFS_CHART = 1<<5, + NETDATA_EBPF_CGROUP_HAS_OOMKILL_CHART = 1<<6, + NETDATA_EBPF_CGROUP_HAS_CACHESTAT_CHART = 1<<7, + NETDATA_EBPF_CGROUP_HAS_DC_CHART = 1<<8, + NETDATA_EBPF_CGROUP_HAS_SHM_CHART = 1<<9 +}; + +typedef struct ebpf_cgroup_target { + char name[256]; // title + uint32_t hash; + uint32_t flags; + uint32_t systemd; + uint32_t updated; + + netdata_publish_swap_t publish_systemd_swap; + netdata_fd_stat_t publish_systemd_fd; + netdata_publish_vfs_t publish_systemd_vfs; + ebpf_process_stat_t publish_systemd_ps; + netdata_publish_dcstat_t publish_dc; + int oomkill; + netdata_publish_shm_t publish_shm; + ebpf_socket_publish_apps_t publish_socket; + netdata_publish_cachestat_t publish_cachestat; + + struct pid_on_target2 *pids; + struct ebpf_cgroup_target *next; +} ebpf_cgroup_target_t; + +void ebpf_map_cgroup_shared_memory(); +void ebpf_parse_cgroup_shm_data(); +void ebpf_create_charts_on_systemd(char *id, char *title, char *units, char *family, char *charttype, int order, + char *algorithm, char *context, char *module, int update_every); +extern int send_cgroup_chart; + +#endif /* NETDATA_EBPF_CGROUP_H */ diff --git a/collectors/ebpf.plugin/ebpf_dcstat.c b/collectors/ebpf.plugin/ebpf_dcstat.c new file mode 100644 index 0000000..71169e1 --- /dev/null +++ b/collectors/ebpf.plugin/ebpf_dcstat.c @@ -0,0 +1,1225 @@ +// SPDX-License-Identifier: GPL-3.0-or-later + +#include "ebpf.h" +#include "ebpf_dcstat.h" + +static char *dcstat_counter_dimension_name[NETDATA_DCSTAT_IDX_END] = { "ratio", "reference", "slow", "miss" }; +static netdata_syscall_stat_t dcstat_counter_aggregated_data[NETDATA_DCSTAT_IDX_END]; +static netdata_publish_syscall_t dcstat_counter_publish_aggregated[NETDATA_DCSTAT_IDX_END]; + +netdata_dcstat_pid_t *dcstat_vector = NULL; +netdata_publish_dcstat_t **dcstat_pid = NULL; + +static netdata_idx_t dcstat_hash_values[NETDATA_DCSTAT_IDX_END]; +static netdata_idx_t *dcstat_values = NULL; + +struct config dcstat_config = { .first_section = NULL, + .last_section = NULL, + .mutex = NETDATA_MUTEX_INITIALIZER, + .index = { .avl_tree = { .root = NULL, .compar = appconfig_section_compare }, + .rwlock = AVL_LOCK_INITIALIZER } }; + +struct netdata_static_thread dcstat_threads = {"DCSTAT KERNEL", + .config_section = NULL, + .config_name = NULL, + .env_name = NULL, + .enabled = 1, + .thread = NULL, + .init_routine = NULL, + .start_routine = NULL}; + +ebpf_local_maps_t dcstat_maps[] = {{.name = "dcstat_global", .internal_input = NETDATA_DIRECTORY_CACHE_END, + .user_input = 0, .type = NETDATA_EBPF_MAP_STATIC, + .map_fd = ND_EBPF_MAP_FD_NOT_INITIALIZED}, + {.name = "dcstat_pid", .internal_input = ND_EBPF_DEFAULT_PID_SIZE, + .user_input = 0, + .type = NETDATA_EBPF_MAP_RESIZABLE | NETDATA_EBPF_MAP_PID, + .map_fd = ND_EBPF_MAP_FD_NOT_INITIALIZED}, + {.name = "dcstat_ctrl", .internal_input = NETDATA_CONTROLLER_END, + .user_input = 0, + .type = NETDATA_EBPF_MAP_CONTROLLER, + .map_fd = ND_EBPF_MAP_FD_NOT_INITIALIZED}, + {.name = NULL, .internal_input = 0, .user_input = 0, + .type = NETDATA_EBPF_MAP_CONTROLLER, + .map_fd = ND_EBPF_MAP_FD_NOT_INITIALIZED}}; + +static ebpf_specify_name_t dc_optional_name[] = { {.program_name = "netdata_lookup_fast", + .function_to_attach = "lookup_fast", + .optional = NULL, + .retprobe = CONFIG_BOOLEAN_NO}, + {.program_name = NULL}}; + +netdata_ebpf_targets_t dc_targets[] = { {.name = "lookup_fast", .mode = EBPF_LOAD_TRAMPOLINE}, + {.name = "d_lookup", .mode = EBPF_LOAD_TRAMPOLINE}, + {.name = NULL, .mode = EBPF_LOAD_TRAMPOLINE}}; + +#ifdef LIBBPF_MAJOR_VERSION +#include "includes/dc.skel.h" // BTF code + +static struct dc_bpf *bpf_obj = NULL; + +/** + * Disable probe + * + * Disable all probes to use exclusively another method. + * + * @param obj is the main structure for bpf objects + */ +static inline void ebpf_dc_disable_probes(struct dc_bpf *obj) +{ + bpf_program__set_autoload(obj->progs.netdata_lookup_fast_kprobe, false); + bpf_program__set_autoload(obj->progs.netdata_d_lookup_kretprobe, false); + bpf_program__set_autoload(obj->progs.netdata_dcstat_release_task_kprobe, false); +} + +/* + * Disable trampoline + * + * Disable all trampoline to use exclusively another method. + * + * @param obj is the main structure for bpf objects. + */ +static inline void ebpf_dc_disable_trampoline(struct dc_bpf *obj) +{ + bpf_program__set_autoload(obj->progs.netdata_lookup_fast_fentry, false); + bpf_program__set_autoload(obj->progs.netdata_d_lookup_fexit, false); + bpf_program__set_autoload(obj->progs.netdata_dcstat_release_task_fentry, false); +} + +/** + * Set trampoline target + * + * Set the targets we will monitor. + * + * @param obj is the main structure for bpf objects. + */ +static void ebpf_dc_set_trampoline_target(struct dc_bpf *obj) +{ + bpf_program__set_attach_target(obj->progs.netdata_lookup_fast_fentry, 0, + dc_targets[NETDATA_DC_TARGET_LOOKUP_FAST].name); + + bpf_program__set_attach_target(obj->progs.netdata_d_lookup_fexit, 0, + dc_targets[NETDATA_DC_TARGET_D_LOOKUP].name); + + bpf_program__set_attach_target(obj->progs.netdata_dcstat_release_task_fentry, 0, + EBPF_COMMON_FNCT_CLEAN_UP); +} + +/** + * Mount Attach Probe + * + * Attach probes to target + * + * @param obj is the main structure for bpf objects. + * + * @return It returns 0 on success and -1 otherwise. + */ +static int ebpf_dc_attach_probes(struct dc_bpf *obj) +{ + obj->links.netdata_d_lookup_kretprobe = bpf_program__attach_kprobe(obj->progs.netdata_d_lookup_kretprobe, + true, + dc_targets[NETDATA_DC_TARGET_D_LOOKUP].name); + int ret = libbpf_get_error(obj->links.netdata_d_lookup_kretprobe); + if (ret) + return -1; + + char *lookup_name = (dc_optional_name[NETDATA_DC_TARGET_LOOKUP_FAST].optional) ? + dc_optional_name[NETDATA_DC_TARGET_LOOKUP_FAST].optional : + dc_targets[NETDATA_DC_TARGET_LOOKUP_FAST].name ; + + obj->links.netdata_lookup_fast_kprobe = bpf_program__attach_kprobe(obj->progs.netdata_lookup_fast_kprobe, + false, + lookup_name); + ret = libbpf_get_error(obj->links.netdata_lookup_fast_kprobe); + if (ret) + return -1; + + obj->links.netdata_dcstat_release_task_kprobe = bpf_program__attach_kprobe(obj->progs.netdata_dcstat_release_task_kprobe, + false, + EBPF_COMMON_FNCT_CLEAN_UP); + ret = libbpf_get_error(obj->links.netdata_dcstat_release_task_kprobe); + if (ret) + return -1; + + return 0; +} + +/** + * Adjust Map Size + * + * Resize maps according input from users. + * + * @param obj is the main structure for bpf objects. + * @param em structure with configuration + */ +static void ebpf_dc_adjust_map_size(struct dc_bpf *obj, ebpf_module_t *em) +{ + ebpf_update_map_size(obj->maps.dcstat_pid, &dcstat_maps[NETDATA_DCSTAT_PID_STATS], + em, bpf_map__name(obj->maps.dcstat_pid)); +} + +/** + * Set hash tables + * + * Set the values for maps according the value given by kernel. + * + * @param obj is the main structure for bpf objects. + */ +static void ebpf_dc_set_hash_tables(struct dc_bpf *obj) +{ + dcstat_maps[NETDATA_DCSTAT_GLOBAL_STATS].map_fd = bpf_map__fd(obj->maps.dcstat_global); + dcstat_maps[NETDATA_DCSTAT_PID_STATS].map_fd = bpf_map__fd(obj->maps.dcstat_pid); + dcstat_maps[NETDATA_DCSTAT_CTRL].map_fd = bpf_map__fd(obj->maps.dcstat_ctrl); +} + +/** + * Update Load + * + * For directory cache, some distributions change the function name, and we do not have condition to use + * TRAMPOLINE like other functions. + * + * @param em structure with configuration + * + * @return When then symbols were not modified, it returns TRAMPOLINE, else it returns RETPROBE. + */ +netdata_ebpf_program_loaded_t ebpf_dc_update_load(ebpf_module_t *em) +{ + if (!strcmp(dc_optional_name[NETDATA_DC_TARGET_LOOKUP_FAST].optional, + dc_optional_name[NETDATA_DC_TARGET_LOOKUP_FAST].function_to_attach)) + return EBPF_LOAD_TRAMPOLINE; + + if (em->targets[NETDATA_DC_TARGET_LOOKUP_FAST].mode != EBPF_LOAD_RETPROBE) + info("When your kernel was compiled the symbol %s was modified, instead to use `trampoline`, the plugin will use `probes`.", + dc_optional_name[NETDATA_DC_TARGET_LOOKUP_FAST].function_to_attach); + + return EBPF_LOAD_RETPROBE; +} + +/** + * Disable Release Task + * + * Disable release task when apps is not enabled. + * + * @param obj is the main structure for bpf objects. + */ +static void ebpf_dc_disable_release_task(struct dc_bpf *obj) +{ + bpf_program__set_autoload(obj->progs.netdata_dcstat_release_task_kprobe, false); + bpf_program__set_autoload(obj->progs.netdata_dcstat_release_task_fentry, false); +} + +/** + * Load and attach + * + * Load and attach the eBPF code in kernel. + * + * @param obj is the main structure for bpf objects. + * @param em structure with configuration + * + * @return it returns 0 on succes and -1 otherwise + */ +static inline int ebpf_dc_load_and_attach(struct dc_bpf *obj, ebpf_module_t *em) +{ + netdata_ebpf_program_loaded_t test = ebpf_dc_update_load(em); + if (test == EBPF_LOAD_TRAMPOLINE) { + ebpf_dc_disable_probes(obj); + + ebpf_dc_set_trampoline_target(obj); + } else { + ebpf_dc_disable_trampoline(obj); + } + + ebpf_dc_adjust_map_size(obj, em); + + if (!em->apps_charts && !em->cgroup_charts) + ebpf_dc_disable_release_task(obj); + + int ret = dc_bpf__load(obj); + if (ret) { + return ret; + } + + ret = (test == EBPF_LOAD_TRAMPOLINE) ? dc_bpf__attach(obj) : ebpf_dc_attach_probes(obj); + if (!ret) { + ebpf_dc_set_hash_tables(obj); + + ebpf_update_controller(dcstat_maps[NETDATA_DCSTAT_CTRL].map_fd, em); + } + + return ret; +} +#endif + +/***************************************************************** + * + * COMMON FUNCTIONS + * + *****************************************************************/ + +/** + * Update publish + * + * Update publish values before to write dimension. + * + * @param out structure that will receive data. + * @param cache_access number of access to directory cache. + * @param not_found number of files not found on the file system + */ +void dcstat_update_publish(netdata_publish_dcstat_t *out, uint64_t cache_access, uint64_t not_found) +{ + NETDATA_DOUBLE successful_access = (NETDATA_DOUBLE) (((long long)cache_access) - ((long long)not_found)); + NETDATA_DOUBLE ratio = (cache_access) ? successful_access/(NETDATA_DOUBLE)cache_access : 0; + + out->ratio = (long long )(ratio*100); +} + +/***************************************************************** + * + * FUNCTIONS TO CLOSE THE THREAD + * + *****************************************************************/ + +/** + * Clean names + * + * Clean the optional names allocated during startup. + */ +void ebpf_dcstat_clean_names() +{ + size_t i = 0; + while (dc_optional_name[i].program_name) { + freez(dc_optional_name[i].optional); + i++; + } +} + +/** + * DCstat Free + * + * Cleanup variables after child threads to stop + * + * @param ptr thread data. + */ +static void ebpf_dcstat_free(ebpf_module_t *em ) +{ + pthread_mutex_lock(&ebpf_exit_cleanup); + if (em->thread->enabled == NETDATA_THREAD_EBPF_RUNNING) { + em->thread->enabled = NETDATA_THREAD_EBPF_STOPPING; + pthread_mutex_unlock(&ebpf_exit_cleanup); + return; + } + pthread_mutex_unlock(&ebpf_exit_cleanup); + + freez(dcstat_vector); + freez(dcstat_values); + freez(dcstat_threads.thread); + + ebpf_cleanup_publish_syscall(dcstat_counter_publish_aggregated); + + ebpf_dcstat_clean_names(); + +#ifdef LIBBPF_MAJOR_VERSION + if (bpf_obj) + dc_bpf__destroy(bpf_obj); +#endif + + pthread_mutex_lock(&ebpf_exit_cleanup); + em->thread->enabled = NETDATA_THREAD_EBPF_STOPPED; + pthread_mutex_unlock(&ebpf_exit_cleanup); +} + +/** + * DCstat exit + * + * Cancel child and exit. + * + * @param ptr thread data. + */ +static void ebpf_dcstat_exit(void *ptr) +{ + ebpf_module_t *em = (ebpf_module_t *)ptr; + netdata_thread_cancel(*dcstat_threads.thread); + ebpf_dcstat_free(em); +} + +/** + * Clean up the main thread. + * + * @param ptr thread data. + */ +static void ebpf_dcstat_cleanup(void *ptr) +{ + ebpf_module_t *em = (ebpf_module_t *)ptr; + ebpf_dcstat_free(em); +} + +/***************************************************************** + * + * APPS + * + *****************************************************************/ + +/** + * Create apps charts + * + * Call ebpf_create_chart to create the charts on apps submenu. + * + * @param em a pointer to the structure with the default values. + */ +void ebpf_dcstat_create_apps_charts(struct ebpf_module *em, void *ptr) +{ + struct target *root = ptr; + ebpf_create_charts_on_apps(NETDATA_DC_HIT_CHART, + "Percentage of files inside directory cache", + EBPF_COMMON_DIMENSION_PERCENTAGE, + NETDATA_DIRECTORY_CACHE_SUBMENU, + NETDATA_EBPF_CHART_TYPE_LINE, + 20100, + ebpf_algorithms[NETDATA_EBPF_ABSOLUTE_IDX], + root, em->update_every, NETDATA_EBPF_MODULE_NAME_DCSTAT); + + ebpf_create_charts_on_apps(NETDATA_DC_REFERENCE_CHART, + "Count file access", + EBPF_COMMON_DIMENSION_FILES, + NETDATA_DIRECTORY_CACHE_SUBMENU, + NETDATA_EBPF_CHART_TYPE_STACKED, + 20101, + ebpf_algorithms[NETDATA_EBPF_ABSOLUTE_IDX], + root, em->update_every, NETDATA_EBPF_MODULE_NAME_DCSTAT); + + ebpf_create_charts_on_apps(NETDATA_DC_REQUEST_NOT_CACHE_CHART, + "Files not present inside directory cache", + EBPF_COMMON_DIMENSION_FILES, + NETDATA_DIRECTORY_CACHE_SUBMENU, + NETDATA_EBPF_CHART_TYPE_STACKED, + 20102, + ebpf_algorithms[NETDATA_EBPF_ABSOLUTE_IDX], + root, em->update_every, NETDATA_EBPF_MODULE_NAME_DCSTAT); + + ebpf_create_charts_on_apps(NETDATA_DC_REQUEST_NOT_FOUND_CHART, + "Files not found", + EBPF_COMMON_DIMENSION_FILES, + NETDATA_DIRECTORY_CACHE_SUBMENU, + NETDATA_EBPF_CHART_TYPE_STACKED, + 20103, + ebpf_algorithms[NETDATA_EBPF_ABSOLUTE_IDX], + root, em->update_every, NETDATA_EBPF_MODULE_NAME_DCSTAT); + + em->apps_charts |= NETDATA_EBPF_APPS_FLAG_CHART_CREATED; +} + +/***************************************************************** + * + * MAIN LOOP + * + *****************************************************************/ + +/** + * Apps Accumulator + * + * Sum all values read from kernel and store in the first address. + * + * @param out the vector with read values. + */ +static void dcstat_apps_accumulator(netdata_dcstat_pid_t *out) +{ + int i, end = (running_on_kernel >= NETDATA_KERNEL_V4_15) ? ebpf_nprocs : 1; + netdata_dcstat_pid_t *total = &out[0]; + for (i = 1; i < end; i++) { + netdata_dcstat_pid_t *w = &out[i]; + total->cache_access += w->cache_access; + total->file_system += w->file_system; + total->not_found += w->not_found; + } +} + +/** + * Save PID values + * + * Save the current values inside the structure + * + * @param out vector used to plot charts + * @param publish vector with values read from hash tables. + */ +static inline void dcstat_save_pid_values(netdata_publish_dcstat_t *out, netdata_dcstat_pid_t *publish) +{ + memcpy(&out->curr, &publish[0], sizeof(netdata_dcstat_pid_t)); +} + +/** + * Fill PID + * + * Fill PID structures + * + * @param current_pid pid that we are collecting data + * @param out values read from hash tables; + */ +static void dcstat_fill_pid(uint32_t current_pid, netdata_dcstat_pid_t *publish) +{ + netdata_publish_dcstat_t *curr = dcstat_pid[current_pid]; + if (!curr) { + curr = callocz(1, sizeof(netdata_publish_dcstat_t)); + dcstat_pid[current_pid] = curr; + } + + dcstat_save_pid_values(curr, publish); +} + +/** + * Read APPS table + * + * Read the apps table and store data inside the structure. + */ +static void read_apps_table() +{ + netdata_dcstat_pid_t *cv = dcstat_vector; + uint32_t key; + struct pid_stat *pids = root_of_pids; + int fd = dcstat_maps[NETDATA_DCSTAT_PID_STATS].map_fd; + size_t length = sizeof(netdata_dcstat_pid_t)*ebpf_nprocs; + while (pids) { + key = pids->pid; + + if (bpf_map_lookup_elem(fd, &key, cv)) { + pids = pids->next; + continue; + } + + dcstat_apps_accumulator(cv); + + dcstat_fill_pid(key, cv); + + // We are cleaning to avoid passing data read from one process to other. + memset(cv, 0, length); + + pids = pids->next; + } +} + +/** + * Update cgroup + * + * Update cgroup data based in + */ +static void ebpf_update_dc_cgroup() +{ + netdata_dcstat_pid_t *cv = dcstat_vector; + int fd = dcstat_maps[NETDATA_DCSTAT_PID_STATS].map_fd; + size_t length = sizeof(netdata_dcstat_pid_t)*ebpf_nprocs; + + ebpf_cgroup_target_t *ect; + pthread_mutex_lock(&mutex_cgroup_shm); + for (ect = ebpf_cgroup_pids; ect; ect = ect->next) { + struct pid_on_target2 *pids; + for (pids = ect->pids; pids; pids = pids->next) { + int pid = pids->pid; + netdata_dcstat_pid_t *out = &pids->dc; + if (likely(dcstat_pid) && dcstat_pid[pid]) { + netdata_publish_dcstat_t *in = dcstat_pid[pid]; + + memcpy(out, &in->curr, sizeof(netdata_dcstat_pid_t)); + } else { + memset(cv, 0, length); + if (bpf_map_lookup_elem(fd, &pid, cv)) { + continue; + } + + dcstat_apps_accumulator(cv); + + memcpy(out, cv, sizeof(netdata_dcstat_pid_t)); + } + } + } + pthread_mutex_unlock(&mutex_cgroup_shm); +} + +/** + * Read global table + * + * Read the table with number of calls for all functions + */ +static void read_global_table() +{ + uint32_t idx; + netdata_idx_t *val = dcstat_hash_values; + netdata_idx_t *stored = dcstat_values; + int fd = dcstat_maps[NETDATA_DCSTAT_GLOBAL_STATS].map_fd; + + for (idx = NETDATA_KEY_DC_REFERENCE; idx < NETDATA_DIRECTORY_CACHE_END; idx++) { + if (!bpf_map_lookup_elem(fd, &idx, stored)) { + int i; + int end = ebpf_nprocs; + netdata_idx_t total = 0; + for (i = 0; i < end; i++) + total += stored[i]; + + val[idx] = total; + } + } +} + +/** + * DCstat read hash + * + * This is the thread callback. + * This thread is necessary, because we cannot freeze the whole plugin to read the data. + * + * @param ptr It is a NULL value for this thread. + * + * @return It always returns NULL. + */ +void *ebpf_dcstat_read_hash(void *ptr) +{ + netdata_thread_cleanup_push(ebpf_dcstat_cleanup, ptr); + heartbeat_t hb; + heartbeat_init(&hb); + + ebpf_module_t *em = (ebpf_module_t *)ptr; + + usec_t step = NETDATA_LATENCY_DCSTAT_SLEEP_MS * em->update_every; + while (!ebpf_exit_plugin) { + (void)heartbeat_next(&hb, step); + + read_global_table(); + } + + netdata_thread_cleanup_pop(1); + return NULL; +} + +/** + * Cachestat sum PIDs + * + * Sum values for all PIDs associated to a group + * + * @param publish output structure. + * @param root structure with listed IPs + */ +void ebpf_dcstat_sum_pids(netdata_publish_dcstat_t *publish, struct pid_on_target *root) +{ + memset(&publish->curr, 0, sizeof(netdata_dcstat_pid_t)); + netdata_dcstat_pid_t *dst = &publish->curr; + while (root) { + int32_t pid = root->pid; + netdata_publish_dcstat_t *w = dcstat_pid[pid]; + if (w) { + netdata_dcstat_pid_t *src = &w->curr; + dst->cache_access += src->cache_access; + dst->file_system += src->file_system; + dst->not_found += src->not_found; + } + + root = root->next; + } +} + +/** + * Send data to Netdata calling auxiliary functions. + * + * @param root the target list. +*/ +void ebpf_dcache_send_apps_data(struct target *root) +{ + struct target *w; + collected_number value; + + write_begin_chart(NETDATA_APPS_FAMILY, NETDATA_DC_HIT_CHART); + for (w = root; w; w = w->next) { + if (unlikely(w->exposed && w->processes)) { + ebpf_dcstat_sum_pids(&w->dcstat, w->root_pid); + + uint64_t cache = w->dcstat.curr.cache_access; + uint64_t not_found = w->dcstat.curr.not_found; + + dcstat_update_publish(&w->dcstat, cache, not_found); + value = (collected_number) w->dcstat.ratio; + write_chart_dimension(w->name, value); + } + } + write_end_chart(); + + write_begin_chart(NETDATA_APPS_FAMILY, NETDATA_DC_REFERENCE_CHART); + for (w = root; w; w = w->next) { + if (unlikely(w->exposed && w->processes)) { + if (w->dcstat.curr.cache_access < w->dcstat.prev.cache_access) { + w->dcstat.prev.cache_access = 0; + } + + w->dcstat.cache_access = (long long)w->dcstat.curr.cache_access - (long long)w->dcstat.prev.cache_access; + value = (collected_number) w->dcstat.cache_access; + write_chart_dimension(w->name, value); + w->dcstat.prev.cache_access = w->dcstat.curr.cache_access; + } + } + write_end_chart(); + + write_begin_chart(NETDATA_APPS_FAMILY, NETDATA_DC_REQUEST_NOT_CACHE_CHART); + for (w = root; w; w = w->next) { + if (unlikely(w->exposed && w->processes)) { + if (w->dcstat.curr.file_system < w->dcstat.prev.file_system) { + w->dcstat.prev.file_system = 0; + } + + value = (collected_number) (!w->dcstat.cache_access) ? 0 : + (long long )w->dcstat.curr.file_system - (long long)w->dcstat.prev.file_system; + write_chart_dimension(w->name, value); + w->dcstat.prev.file_system = w->dcstat.curr.file_system; + } + } + write_end_chart(); + + write_begin_chart(NETDATA_APPS_FAMILY, NETDATA_DC_REQUEST_NOT_FOUND_CHART); + for (w = root; w; w = w->next) { + if (unlikely(w->exposed && w->processes)) { + if (w->dcstat.curr.not_found < w->dcstat.prev.not_found) { + w->dcstat.prev.not_found = 0; + } + value = (collected_number) (!w->dcstat.cache_access) ? 0 : + (long long)w->dcstat.curr.not_found - (long long)w->dcstat.prev.not_found; + write_chart_dimension(w->name, value); + w->dcstat.prev.not_found = w->dcstat.curr.not_found; + } + } + write_end_chart(); +} + +/** + * Send global + * + * Send global charts to Netdata + */ +static void dcstat_send_global(netdata_publish_dcstat_t *publish) +{ + dcstat_update_publish(publish, dcstat_hash_values[NETDATA_KEY_DC_REFERENCE], + dcstat_hash_values[NETDATA_KEY_DC_MISS]); + + netdata_publish_syscall_t *ptr = dcstat_counter_publish_aggregated; + netdata_idx_t value = dcstat_hash_values[NETDATA_KEY_DC_REFERENCE]; + if (value != ptr[NETDATA_DCSTAT_IDX_REFERENCE].pcall) { + ptr[NETDATA_DCSTAT_IDX_REFERENCE].ncall = value - ptr[NETDATA_DCSTAT_IDX_REFERENCE].pcall; + ptr[NETDATA_DCSTAT_IDX_REFERENCE].pcall = value; + + value = dcstat_hash_values[NETDATA_KEY_DC_SLOW]; + ptr[NETDATA_DCSTAT_IDX_SLOW].ncall = value - ptr[NETDATA_DCSTAT_IDX_SLOW].pcall; + ptr[NETDATA_DCSTAT_IDX_SLOW].pcall = value; + + value = dcstat_hash_values[NETDATA_KEY_DC_MISS]; + ptr[NETDATA_DCSTAT_IDX_MISS].ncall = value - ptr[NETDATA_DCSTAT_IDX_MISS].pcall; + ptr[NETDATA_DCSTAT_IDX_MISS].pcall = value; + } else { + ptr[NETDATA_DCSTAT_IDX_REFERENCE].ncall = 0; + ptr[NETDATA_DCSTAT_IDX_SLOW].ncall = 0; + ptr[NETDATA_DCSTAT_IDX_MISS].ncall = 0; + } + + ebpf_one_dimension_write_charts(NETDATA_FILESYSTEM_FAMILY, NETDATA_DC_HIT_CHART, + ptr[NETDATA_DCSTAT_IDX_RATIO].dimension, publish->ratio); + + write_count_chart( + NETDATA_DC_REFERENCE_CHART, NETDATA_FILESYSTEM_FAMILY, + &dcstat_counter_publish_aggregated[NETDATA_DCSTAT_IDX_REFERENCE], 3); +} + +/** + * Create specific directory cache charts + * + * Create charts for cgroup/application. + * + * @param type the chart type. + * @param update_every value to overwrite the update frequency set by the server. + */ +static void ebpf_create_specific_dc_charts(char *type, int update_every) +{ + ebpf_create_chart(type, NETDATA_DC_HIT_CHART, "Percentage of files inside directory cache", + EBPF_COMMON_DIMENSION_PERCENTAGE, NETDATA_DIRECTORY_CACHE_SUBMENU, + NETDATA_CGROUP_DC_HIT_RATIO_CONTEXT, NETDATA_EBPF_CHART_TYPE_LINE, + NETDATA_CHART_PRIO_CGROUPS_CONTAINERS + 5700, + ebpf_create_global_dimension, + dcstat_counter_publish_aggregated, 1, update_every, NETDATA_EBPF_MODULE_NAME_DCSTAT); + + ebpf_create_chart(type, NETDATA_DC_REFERENCE_CHART, "Count file access", + EBPF_COMMON_DIMENSION_FILES, NETDATA_DIRECTORY_CACHE_SUBMENU, + NETDATA_CGROUP_DC_REFERENCE_CONTEXT, NETDATA_EBPF_CHART_TYPE_LINE, + NETDATA_CHART_PRIO_CGROUPS_CONTAINERS + 5701, + ebpf_create_global_dimension, + &dcstat_counter_publish_aggregated[NETDATA_DCSTAT_IDX_REFERENCE], 1, + update_every, NETDATA_EBPF_MODULE_NAME_DCSTAT); + + ebpf_create_chart(type, NETDATA_DC_REQUEST_NOT_CACHE_CHART, + "Files not present inside directory cache", + EBPF_COMMON_DIMENSION_FILES, NETDATA_DIRECTORY_CACHE_SUBMENU, + NETDATA_CGROUP_DC_NOT_CACHE_CONTEXT, NETDATA_EBPF_CHART_TYPE_LINE, + NETDATA_CHART_PRIO_CGROUPS_CONTAINERS + 5702, + ebpf_create_global_dimension, + &dcstat_counter_publish_aggregated[NETDATA_DCSTAT_IDX_SLOW], 1, + update_every, NETDATA_EBPF_MODULE_NAME_DCSTAT); + + ebpf_create_chart(type, NETDATA_DC_REQUEST_NOT_FOUND_CHART, + "Files not found", + EBPF_COMMON_DIMENSION_FILES, NETDATA_DIRECTORY_CACHE_SUBMENU, + NETDATA_CGROUP_DC_NOT_FOUND_CONTEXT, NETDATA_EBPF_CHART_TYPE_LINE, + NETDATA_CHART_PRIO_CGROUPS_CONTAINERS + 5703, + ebpf_create_global_dimension, + &dcstat_counter_publish_aggregated[NETDATA_DCSTAT_IDX_MISS], 1, + update_every, NETDATA_EBPF_MODULE_NAME_DCSTAT); +} + +/** + * Obsolete specific directory cache charts + * + * Obsolete charts for cgroup/application. + * + * @param type the chart type. + * @param update_every value to overwrite the update frequency set by the server. + */ +static void ebpf_obsolete_specific_dc_charts(char *type, int update_every) +{ + ebpf_write_chart_obsolete(type, NETDATA_DC_HIT_CHART, + "Percentage of files inside directory cache", + EBPF_COMMON_DIMENSION_PERCENTAGE, NETDATA_DIRECTORY_CACHE_SUBMENU, + NETDATA_EBPF_CHART_TYPE_LINE, NETDATA_CGROUP_DC_HIT_RATIO_CONTEXT, + NETDATA_CHART_PRIO_CGROUPS_CONTAINERS + 5700, update_every); + + ebpf_write_chart_obsolete(type, NETDATA_DC_REFERENCE_CHART, + "Count file access", + EBPF_COMMON_DIMENSION_FILES, NETDATA_DIRECTORY_CACHE_SUBMENU, + NETDATA_EBPF_CHART_TYPE_LINE, NETDATA_CGROUP_DC_REFERENCE_CONTEXT, + NETDATA_CHART_PRIO_CGROUPS_CONTAINERS + 5701, update_every); + + ebpf_write_chart_obsolete(type, NETDATA_DC_REQUEST_NOT_CACHE_CHART, + "Files not present inside directory cache", + EBPF_COMMON_DIMENSION_FILES, NETDATA_DIRECTORY_CACHE_SUBMENU, + NETDATA_EBPF_CHART_TYPE_LINE, NETDATA_CGROUP_DC_NOT_CACHE_CONTEXT, + NETDATA_CHART_PRIO_CGROUPS_CONTAINERS + 5702, update_every); + + ebpf_write_chart_obsolete(type, NETDATA_DC_REQUEST_NOT_FOUND_CHART, + "Files not found", + EBPF_COMMON_DIMENSION_FILES, NETDATA_DIRECTORY_CACHE_SUBMENU, + NETDATA_EBPF_CHART_TYPE_LINE, NETDATA_CGROUP_DC_NOT_FOUND_CONTEXT, + NETDATA_CHART_PRIO_CGROUPS_CONTAINERS + 5703, update_every); +} + +/** + * Cachestat sum PIDs + * + * Sum values for all PIDs associated to a group + * + * @param publish output structure. + * @param root structure with listed IPs + */ +void ebpf_dc_sum_cgroup_pids(netdata_publish_dcstat_t *publish, struct pid_on_target2 *root) +{ + memset(&publish->curr, 0, sizeof(netdata_dcstat_pid_t)); + netdata_dcstat_pid_t *dst = &publish->curr; + while (root) { + netdata_dcstat_pid_t *src = &root->dc; + + dst->cache_access += src->cache_access; + dst->file_system += src->file_system; + dst->not_found += src->not_found; + + root = root->next; + } +} + +/** + * Calc chart values + * + * Do necessary math to plot charts. + */ +void ebpf_dc_calc_chart_values() +{ + ebpf_cgroup_target_t *ect; + for (ect = ebpf_cgroup_pids; ect ; ect = ect->next) { + ebpf_dc_sum_cgroup_pids(&ect->publish_dc, ect->pids); + uint64_t cache = ect->publish_dc.curr.cache_access; + uint64_t not_found = ect->publish_dc.curr.not_found; + + dcstat_update_publish(&ect->publish_dc, cache, not_found); + + ect->publish_dc.cache_access = (long long)ect->publish_dc.curr.cache_access - + (long long)ect->publish_dc.prev.cache_access; + ect->publish_dc.prev.cache_access = ect->publish_dc.curr.cache_access; + + if (ect->publish_dc.curr.not_found < ect->publish_dc.prev.not_found) { + ect->publish_dc.prev.not_found = 0; + } + } +} + +/** + * Create Systemd directory cache Charts + * + * Create charts when systemd is enabled + * + * @param update_every value to overwrite the update frequency set by the server. + **/ +static void ebpf_create_systemd_dc_charts(int update_every) +{ + ebpf_create_charts_on_systemd(NETDATA_DC_HIT_CHART, + "Percentage of files inside directory cache", + EBPF_COMMON_DIMENSION_PERCENTAGE, + NETDATA_DIRECTORY_CACHE_SUBMENU, + NETDATA_EBPF_CHART_TYPE_LINE, + 21200, + ebpf_algorithms[NETDATA_EBPF_ABSOLUTE_IDX], + NETDATA_SYSTEMD_DC_HIT_RATIO_CONTEXT, NETDATA_EBPF_MODULE_NAME_DCSTAT, + update_every); + + ebpf_create_charts_on_systemd(NETDATA_DC_REFERENCE_CHART, + "Count file access", + EBPF_COMMON_DIMENSION_FILES, + NETDATA_DIRECTORY_CACHE_SUBMENU, + NETDATA_EBPF_CHART_TYPE_LINE, + 21201, + ebpf_algorithms[NETDATA_EBPF_ABSOLUTE_IDX], + NETDATA_SYSTEMD_DC_REFERENCE_CONTEXT, NETDATA_EBPF_MODULE_NAME_DCSTAT, + update_every); + + ebpf_create_charts_on_systemd(NETDATA_DC_REQUEST_NOT_CACHE_CHART, + "Files not present inside directory cache", + EBPF_COMMON_DIMENSION_FILES, + NETDATA_DIRECTORY_CACHE_SUBMENU, + NETDATA_EBPF_CHART_TYPE_LINE, + 21202, + ebpf_algorithms[NETDATA_EBPF_ABSOLUTE_IDX], + NETDATA_SYSTEMD_DC_NOT_CACHE_CONTEXT, NETDATA_EBPF_MODULE_NAME_DCSTAT, + update_every); + + ebpf_create_charts_on_systemd(NETDATA_DC_REQUEST_NOT_FOUND_CHART, + "Files not found", + EBPF_COMMON_DIMENSION_FILES, + NETDATA_DIRECTORY_CACHE_SUBMENU, + NETDATA_EBPF_CHART_TYPE_LINE, + 21202, + ebpf_algorithms[NETDATA_EBPF_ABSOLUTE_IDX], + NETDATA_SYSTEMD_DC_NOT_FOUND_CONTEXT, NETDATA_EBPF_MODULE_NAME_DCSTAT, + update_every); +} + +/** + * Send Directory Cache charts + * + * Send collected data to Netdata. + */ +static void ebpf_send_systemd_dc_charts() +{ + collected_number value; + ebpf_cgroup_target_t *ect; + write_begin_chart(NETDATA_SERVICE_FAMILY, NETDATA_DC_HIT_CHART); + for (ect = ebpf_cgroup_pids; ect ; ect = ect->next) { + if (unlikely(ect->systemd) && unlikely(ect->updated)) { + write_chart_dimension(ect->name, (long long) ect->publish_dc.ratio); + } + } + write_end_chart(); + + write_begin_chart(NETDATA_SERVICE_FAMILY, NETDATA_DC_REFERENCE_CHART); + for (ect = ebpf_cgroup_pids; ect ; ect = ect->next) { + if (unlikely(ect->systemd) && unlikely(ect->updated)) { + write_chart_dimension(ect->name, (long long) ect->publish_dc.cache_access); + } + } + write_end_chart(); + + write_begin_chart(NETDATA_SERVICE_FAMILY, NETDATA_DC_REQUEST_NOT_CACHE_CHART); + for (ect = ebpf_cgroup_pids; ect ; ect = ect->next) { + if (unlikely(ect->systemd) && unlikely(ect->updated)) { + value = (collected_number) (!ect->publish_dc.cache_access) ? 0 : + (long long )ect->publish_dc.curr.file_system - (long long)ect->publish_dc.prev.file_system; + ect->publish_dc.prev.file_system = ect->publish_dc.curr.file_system; + + write_chart_dimension(ect->name, (long long) value); + } + } + write_end_chart(); + + write_begin_chart(NETDATA_SERVICE_FAMILY, NETDATA_DC_REQUEST_NOT_FOUND_CHART); + for (ect = ebpf_cgroup_pids; ect ; ect = ect->next) { + if (unlikely(ect->systemd) && unlikely(ect->updated)) { + value = (collected_number) (!ect->publish_dc.cache_access) ? 0 : + (long long)ect->publish_dc.curr.not_found - (long long)ect->publish_dc.prev.not_found; + + ect->publish_dc.prev.not_found = ect->publish_dc.curr.not_found; + + write_chart_dimension(ect->name, (long long) value); + } + } + write_end_chart(); +} + +/** + * Send Directory Cache charts + * + * Send collected data to Netdata. + * + */ +static void ebpf_send_specific_dc_data(char *type, netdata_publish_dcstat_t *pdc) +{ + collected_number value; + write_begin_chart(type, NETDATA_DC_HIT_CHART); + write_chart_dimension(dcstat_counter_publish_aggregated[NETDATA_DCSTAT_IDX_RATIO].name, + (long long) pdc->ratio); + write_end_chart(); + + write_begin_chart(type, NETDATA_DC_REFERENCE_CHART); + write_chart_dimension(dcstat_counter_publish_aggregated[NETDATA_DCSTAT_IDX_REFERENCE].name, + (long long) pdc->cache_access); + write_end_chart(); + + value = (collected_number) (!pdc->cache_access) ? 0 : + (long long )pdc->curr.file_system - (long long)pdc->prev.file_system; + pdc->prev.file_system = pdc->curr.file_system; + + write_begin_chart(type, NETDATA_DC_REQUEST_NOT_CACHE_CHART); + write_chart_dimension(dcstat_counter_publish_aggregated[NETDATA_DCSTAT_IDX_SLOW].name, (long long) value); + write_end_chart(); + + value = (collected_number) (!pdc->cache_access) ? 0 : + (long long)pdc->curr.not_found - (long long)pdc->prev.not_found; + pdc->prev.not_found = pdc->curr.not_found; + + write_begin_chart(type, NETDATA_DC_REQUEST_NOT_FOUND_CHART); + write_chart_dimension(dcstat_counter_publish_aggregated[NETDATA_DCSTAT_IDX_MISS].name, (long long) value); + write_end_chart(); +} + +/** + * Send data to Netdata calling auxiliary functions. + * + * @param update_every value to overwrite the update frequency set by the server. +*/ +void ebpf_dc_send_cgroup_data(int update_every) +{ + if (!ebpf_cgroup_pids) + return; + + pthread_mutex_lock(&mutex_cgroup_shm); + ebpf_cgroup_target_t *ect; + ebpf_dc_calc_chart_values(); + + int has_systemd = shm_ebpf_cgroup.header->systemd_enabled; + if (has_systemd) { + if (send_cgroup_chart) { + ebpf_create_systemd_dc_charts(update_every); + } + + ebpf_send_systemd_dc_charts(); + } + + for (ect = ebpf_cgroup_pids; ect ; ect = ect->next) { + if (ect->systemd) + continue; + + if (!(ect->flags & NETDATA_EBPF_CGROUP_HAS_DC_CHART) && ect->updated) { + ebpf_create_specific_dc_charts(ect->name, update_every); + ect->flags |= NETDATA_EBPF_CGROUP_HAS_DC_CHART; + } + + if (ect->flags & NETDATA_EBPF_CGROUP_HAS_DC_CHART) { + if (ect->updated) { + ebpf_send_specific_dc_data(ect->name, &ect->publish_dc); + } else { + ebpf_obsolete_specific_dc_charts(ect->name, update_every); + ect->flags &= ~NETDATA_EBPF_CGROUP_HAS_DC_CHART; + } + } + } + + pthread_mutex_unlock(&mutex_cgroup_shm); +} + +/** +* Main loop for this collector. +*/ +static void dcstat_collector(ebpf_module_t *em) +{ + dcstat_threads.thread = mallocz(sizeof(netdata_thread_t)); + dcstat_threads.start_routine = ebpf_dcstat_read_hash; + + netdata_thread_create(dcstat_threads.thread, dcstat_threads.name, NETDATA_THREAD_OPTION_DEFAULT, + ebpf_dcstat_read_hash, em); + + netdata_publish_dcstat_t publish; + memset(&publish, 0, sizeof(publish)); + int cgroups = em->cgroup_charts; + int update_every = em->update_every; + heartbeat_t hb; + heartbeat_init(&hb); + usec_t step = update_every * USEC_PER_SEC; + while (!ebpf_exit_plugin) { + (void)heartbeat_next(&hb, step); + if (ebpf_exit_plugin) + break; + + netdata_apps_integration_flags_t apps = em->apps_charts; + pthread_mutex_lock(&collect_data_mutex); + if (apps) + read_apps_table(); + + if (cgroups) + ebpf_update_dc_cgroup(); + + pthread_mutex_lock(&lock); + + dcstat_send_global(&publish); + + if (apps & NETDATA_EBPF_APPS_FLAG_CHART_CREATED) + ebpf_dcache_send_apps_data(apps_groups_root_target); + + if (cgroups) + ebpf_dc_send_cgroup_data(update_every); + + pthread_mutex_unlock(&lock); + pthread_mutex_unlock(&collect_data_mutex); + } +} + +/***************************************************************** + * + * INITIALIZE THREAD + * + *****************************************************************/ + +/** + * Create filesystem charts + * + * Call ebpf_create_chart to create the charts for the collector. + * + * @param update_every value to overwrite the update frequency set by the server. + */ +static void ebpf_create_filesystem_charts(int update_every) +{ + ebpf_create_chart(NETDATA_FILESYSTEM_FAMILY, NETDATA_DC_HIT_CHART, + "Percentage of files inside directory cache", + EBPF_COMMON_DIMENSION_PERCENTAGE, NETDATA_DIRECTORY_CACHE_SUBMENU, + NULL, + NETDATA_EBPF_CHART_TYPE_LINE, + 21200, + ebpf_create_global_dimension, + dcstat_counter_publish_aggregated, 1, update_every, NETDATA_EBPF_MODULE_NAME_DCSTAT); + + ebpf_create_chart(NETDATA_FILESYSTEM_FAMILY, NETDATA_DC_REFERENCE_CHART, + "Variables used to calculate hit ratio.", + EBPF_COMMON_DIMENSION_FILES, NETDATA_DIRECTORY_CACHE_SUBMENU, + NULL, + NETDATA_EBPF_CHART_TYPE_LINE, + 21201, + ebpf_create_global_dimension, + &dcstat_counter_publish_aggregated[NETDATA_DCSTAT_IDX_REFERENCE], 3, + update_every, NETDATA_EBPF_MODULE_NAME_DCSTAT); + + fflush(stdout); +} + +/** + * Allocate vectors used with this thread. + * + * We are not testing the return, because callocz does this and shutdown the software + * case it was not possible to allocate. + * + * @param apps is apps enabled? + */ +static void ebpf_dcstat_allocate_global_vectors(int apps) +{ + if (apps) + dcstat_pid = callocz((size_t)pid_max, sizeof(netdata_publish_dcstat_t *)); + + dcstat_vector = callocz((size_t)ebpf_nprocs, sizeof(netdata_dcstat_pid_t)); + dcstat_values = callocz((size_t)ebpf_nprocs, sizeof(netdata_idx_t)); + + memset(dcstat_counter_aggregated_data, 0, NETDATA_DCSTAT_IDX_END * sizeof(netdata_syscall_stat_t)); + memset(dcstat_counter_publish_aggregated, 0, NETDATA_DCSTAT_IDX_END * sizeof(netdata_publish_syscall_t)); +} + +/***************************************************************** + * + * MAIN THREAD + * + *****************************************************************/ + +/* + * Load BPF + * + * Load BPF files. + * + * @param em the structure with configuration + */ +static int ebpf_dcstat_load_bpf(ebpf_module_t *em) +{ + int ret = 0; + ebpf_adjust_apps_cgroup(em, em->targets[NETDATA_DC_TARGET_LOOKUP_FAST].mode); + if (em->load & EBPF_LOAD_LEGACY) { + em->probe_links = ebpf_load_program(ebpf_plugin_dir, em, running_on_kernel, isrh, &em->objects); + if (!em->probe_links) { + ret = -1; + } + } +#ifdef LIBBPF_MAJOR_VERSION + else { + bpf_obj = dc_bpf__open(); + if (!bpf_obj) + ret = -1; + else + ret = ebpf_dc_load_and_attach(bpf_obj, em); + } +#endif + + if (ret) + error("%s %s", EBPF_DEFAULT_ERROR_MSG, em->thread_name); + + return ret; +} + +/** + * Directory Cache thread + * + * Thread used to make dcstat thread + * + * @param ptr a pointer to `struct ebpf_module` + * + * @return It always returns NULL + */ +void *ebpf_dcstat_thread(void *ptr) +{ + netdata_thread_cleanup_push(ebpf_dcstat_exit, ptr); + + ebpf_module_t *em = (ebpf_module_t *)ptr; + em->maps = dcstat_maps; + + ebpf_update_pid_table(&dcstat_maps[NETDATA_DCSTAT_PID_STATS], em); + + ebpf_update_names(dc_optional_name, em); + +#ifdef LIBBPF_MAJOR_VERSION + ebpf_adjust_thread_load(em, default_btf); +#endif + if (ebpf_dcstat_load_bpf(em)) { + em->thread->enabled = NETDATA_THREAD_EBPF_STOPPED; + goto enddcstat; + } + + ebpf_dcstat_allocate_global_vectors(em->apps_charts); + + int algorithms[NETDATA_DCSTAT_IDX_END] = { + NETDATA_EBPF_ABSOLUTE_IDX, NETDATA_EBPF_ABSOLUTE_IDX, NETDATA_EBPF_ABSOLUTE_IDX, + NETDATA_EBPF_ABSOLUTE_IDX + }; + + ebpf_global_labels(dcstat_counter_aggregated_data, dcstat_counter_publish_aggregated, + dcstat_counter_dimension_name, dcstat_counter_dimension_name, + algorithms, NETDATA_DCSTAT_IDX_END); + + pthread_mutex_lock(&lock); + ebpf_create_filesystem_charts(em->update_every); + ebpf_update_stats(&plugin_statistics, em); + pthread_mutex_unlock(&lock); + + dcstat_collector(em); + +enddcstat: + ebpf_update_disabled_plugin_stats(em); + + netdata_thread_cleanup_pop(1); + return NULL; +} diff --git a/collectors/ebpf.plugin/ebpf_dcstat.h b/collectors/ebpf.plugin/ebpf_dcstat.h new file mode 100644 index 0000000..d8687f9 --- /dev/null +++ b/collectors/ebpf.plugin/ebpf_dcstat.h @@ -0,0 +1,84 @@ +// SPDX-License-Identifier: GPL-3.0-or-later + +#ifndef NETDATA_EBPF_DCSTAT_H +#define NETDATA_EBPF_DCSTAT_H 1 + +// Module name +#define NETDATA_EBPF_MODULE_NAME_DCSTAT "dcstat" + +// charts +#define NETDATA_DC_HIT_CHART "dc_hit_ratio" +#define NETDATA_DC_REFERENCE_CHART "dc_reference" +#define NETDATA_DC_REQUEST_NOT_CACHE_CHART "dc_not_cache" +#define NETDATA_DC_REQUEST_NOT_FOUND_CHART "dc_not_found" + +#define NETDATA_DIRECTORY_CACHE_SUBMENU "directory cache (eBPF)" + +// configuration file +#define NETDATA_DIRECTORY_DCSTAT_CONFIG_FILE "dcstat.conf" + +// Contexts +#define NETDATA_CGROUP_DC_HIT_RATIO_CONTEXT "cgroup.dc_ratio" +#define NETDATA_CGROUP_DC_REFERENCE_CONTEXT "cgroup.dc_reference" +#define NETDATA_CGROUP_DC_NOT_CACHE_CONTEXT "cgroup.dc_not_cache" +#define NETDATA_CGROUP_DC_NOT_FOUND_CONTEXT "cgroup.dc_not_found" + +#define NETDATA_SYSTEMD_DC_HIT_RATIO_CONTEXT "services.dc_ratio" +#define NETDATA_SYSTEMD_DC_REFERENCE_CONTEXT "services.dc_reference" +#define NETDATA_SYSTEMD_DC_NOT_CACHE_CONTEXT "services.dc_not_cache" +#define NETDATA_SYSTEMD_DC_NOT_FOUND_CONTEXT "services.dc_not_found" + +#define NETDATA_LATENCY_DCSTAT_SLEEP_MS 700000ULL + +enum directory_cache_indexes { + NETDATA_DCSTAT_IDX_RATIO, + NETDATA_DCSTAT_IDX_REFERENCE, + NETDATA_DCSTAT_IDX_SLOW, + NETDATA_DCSTAT_IDX_MISS, + + // Keep this as last and don't skip numbers as it is used as element counter + NETDATA_DCSTAT_IDX_END +}; + +enum directory_cache_tables { + NETDATA_DCSTAT_GLOBAL_STATS, + NETDATA_DCSTAT_PID_STATS, + NETDATA_DCSTAT_CTRL +}; + +// variables +enum directory_cache_counters { + NETDATA_KEY_DC_REFERENCE, + NETDATA_KEY_DC_SLOW, + NETDATA_KEY_DC_MISS, + + // Keep this as last and don't skip numbers as it is used as element counter + NETDATA_DIRECTORY_CACHE_END +}; + +enum directory_cache_targets { + NETDATA_DC_TARGET_LOOKUP_FAST, + NETDATA_DC_TARGET_D_LOOKUP +}; + +typedef struct netdata_publish_dcstat_pid { + uint64_t cache_access; + uint64_t file_system; + uint64_t not_found; +} netdata_dcstat_pid_t; + +typedef struct netdata_publish_dcstat { + long long ratio; + long long cache_access; + + netdata_dcstat_pid_t curr; + netdata_dcstat_pid_t prev; +} netdata_publish_dcstat_t; + +void *ebpf_dcstat_thread(void *ptr); +void ebpf_dcstat_create_apps_charts(struct ebpf_module *em, void *ptr); +extern struct config dcstat_config; +extern netdata_ebpf_targets_t dc_targets[]; +extern ebpf_local_maps_t dcstat_maps[]; + +#endif // NETDATA_EBPF_DCSTAT_H diff --git a/collectors/ebpf.plugin/ebpf_disk.c b/collectors/ebpf.plugin/ebpf_disk.c new file mode 100644 index 0000000..a27bd81 --- /dev/null +++ b/collectors/ebpf.plugin/ebpf_disk.c @@ -0,0 +1,865 @@ +// SPDX-License-Identifier: GPL-3.0-or-later + +#include <sys/resource.h> +#include <stdlib.h> + +#include "ebpf.h" +#include "ebpf_disk.h" + +struct config disk_config = { .first_section = NULL, + .last_section = NULL, + .mutex = NETDATA_MUTEX_INITIALIZER, + .index = { .avl_tree = { .root = NULL, .compar = appconfig_section_compare }, + .rwlock = AVL_LOCK_INITIALIZER } }; + +static ebpf_local_maps_t disk_maps[] = {{.name = "tbl_disk_iocall", .internal_input = NETDATA_DISK_HISTOGRAM_LENGTH, + .user_input = 0, .type = NETDATA_EBPF_MAP_STATIC, + .map_fd = ND_EBPF_MAP_FD_NOT_INITIALIZED}, + {.name = NULL, .internal_input = 0, .user_input = 0, + .type = NETDATA_EBPF_MAP_CONTROLLER, + .map_fd = ND_EBPF_MAP_FD_NOT_INITIALIZED}}; +static avl_tree_lock disk_tree; +netdata_ebpf_disks_t *disk_list = NULL; + +char *tracepoint_block_type = { "block"} ; +char *tracepoint_block_issue = { "block_rq_issue" }; +char *tracepoint_block_rq_complete = { "block_rq_complete" }; + +static int was_block_issue_enabled = 0; +static int was_block_rq_complete_enabled = 0; + +static char **dimensions = NULL; +static netdata_syscall_stat_t disk_aggregated_data[NETDATA_EBPF_HIST_MAX_BINS]; +static netdata_publish_syscall_t disk_publish_aggregated[NETDATA_EBPF_HIST_MAX_BINS]; + +static netdata_idx_t *disk_hash_values = NULL; +static struct netdata_static_thread disk_threads = { + .name = "DISK KERNEL", + .config_section = NULL, + .config_name = NULL, + .env_name = NULL, + .enabled = 1, + .thread = NULL, + .init_routine = NULL, + .start_routine = NULL +}; + +ebpf_publish_disk_t *plot_disks = NULL; +pthread_mutex_t plot_mutex; + +/***************************************************************** + * + * FUNCTIONS TO MANIPULATE HARD DISKS + * + *****************************************************************/ + +/** + * Parse start + * + * Parse start address of disk + * + * @param w structure where data is stored + * @param filename variable used to store value + * + * @return It returns 0 on success and -1 otherwise + */ +static inline int ebpf_disk_parse_start(netdata_ebpf_disks_t *w, char *filename) +{ + char content[FILENAME_MAX + 1]; + int fd = open(filename, O_RDONLY, 0); + if (fd < 0) { + return -1; + } + + ssize_t file_length = read(fd, content, 4095); + if (file_length > 0) { + if (file_length > FILENAME_MAX) + file_length = FILENAME_MAX; + + content[file_length] = '\0'; + w->start = strtoul(content, NULL, 10); + } + close(fd); + + return 0; +} + +/** + * Parse uevent + * + * Parse uevent file + * + * @param w structure where data is stored + * @param filename variable used to store value + * + * @return It returns 0 on success and -1 otherwise + */ +static inline int ebpf_parse_uevent(netdata_ebpf_disks_t *w, char *filename) +{ + char content[FILENAME_MAX + 1]; + int fd = open(filename, O_RDONLY, 0); + if (fd < 0) { + return -1; + } + + ssize_t file_length = read(fd, content, FILENAME_MAX); + if (file_length > 0) { + if (file_length > FILENAME_MAX) + file_length = FILENAME_MAX; + + content[file_length] = '\0'; + + char *s = strstr(content, "PARTNAME=EFI"); + if (s) { + w->main->boot_partition = w; + w->flags |= NETDATA_DISK_HAS_EFI; + w->boot_chart = strdupz("disk_bootsector"); + } + } + close(fd); + + return 0; +} + +/** + * Parse Size + * + * @param w structure where data is stored + * @param filename variable used to store value + * + * @return It returns 0 on success and -1 otherwise + */ +static inline int ebpf_parse_size(netdata_ebpf_disks_t *w, char *filename) +{ + char content[FILENAME_MAX + 1]; + int fd = open(filename, O_RDONLY, 0); + if (fd < 0) { + return -1; + } + + ssize_t file_length = read(fd, content, FILENAME_MAX); + if (file_length > 0) { + if (file_length > FILENAME_MAX) + file_length = FILENAME_MAX; + + content[file_length] = '\0'; + w->end = w->start + strtoul(content, NULL, 10) -1; + } + close(fd); + + return 0; +} + +/** + * Read Disk information + * + * Read disk information from /sys/block + * + * @param w structure where data is stored + * @param name disk name + */ +static void ebpf_read_disk_info(netdata_ebpf_disks_t *w, char *name) +{ + static netdata_ebpf_disks_t *main_disk = NULL; + static uint32_t key = 0; + char *path = { "/sys/block" }; + char disk[NETDATA_DISK_NAME_LEN + 1]; + char filename[FILENAME_MAX + 1]; + snprintfz(disk, NETDATA_DISK_NAME_LEN, "%s", name); + size_t length = strlen(disk); + if (!length) { + return; + } + + length--; + size_t curr = length; + while (isdigit((int)disk[length])) { + disk[length--] = '\0'; + } + + // We are looking for partition information, if it is a device we will ignore it. + if (curr == length) { + main_disk = w; + key = MKDEV(w->major, w->minor); + w->bootsector_key = key; + return; + } + w->bootsector_key = key; + w->main = main_disk; + + snprintfz(filename, FILENAME_MAX, "%s/%s/%s/uevent", path, disk, name); + if (ebpf_parse_uevent(w, filename)) + return; + + snprintfz(filename, FILENAME_MAX, "%s/%s/%s/start", path, disk, name); + if (ebpf_disk_parse_start(w, filename)) + return; + + snprintfz(filename, FILENAME_MAX, "%s/%s/%s/size", path, disk, name); + ebpf_parse_size(w, filename); +} + +/** + * New encode dev + * + * New encode algorithm extracted from https://elixir.bootlin.com/linux/v5.10.8/source/include/linux/kdev_t.h#L39 + * + * @param major driver major number + * @param minor driver minor number + * + * @return + */ +static inline uint32_t netdata_new_encode_dev(uint32_t major, uint32_t minor) { + return (minor & 0xff) | (major << 8) | ((minor & ~0xff) << 12); +} + +/** + * Compare disks + * + * Compare major and minor values to add disks to tree. + * + * @param a pointer to netdata_ebpf_disks + * @param b pointer to netdata_ebpf_disks + * + * @return It returns 0 case the values are equal, 1 case a is bigger than b and -1 case a is smaller than b. +*/ +static int ebpf_compare_disks(void *a, void *b) +{ + netdata_ebpf_disks_t *ptr1 = a; + netdata_ebpf_disks_t *ptr2 = b; + + if (ptr1->dev > ptr2->dev) + return 1; + if (ptr1->dev < ptr2->dev) + return -1; + + return 0; +} + +/** + * Update listen table + * + * Update link list when it is necessary. + * + * @param name disk name + * @param major major disk identifier + * @param minor minor disk identifier + * @param current_time current timestamp + */ +static void update_disk_table(char *name, int major, int minor, time_t current_time) +{ + netdata_ebpf_disks_t find; + netdata_ebpf_disks_t *w; + size_t length; + + uint32_t dev = netdata_new_encode_dev(major, minor); + find.dev = dev; + netdata_ebpf_disks_t *ret = (netdata_ebpf_disks_t *) avl_search_lock(&disk_tree, (avl_t *)&find); + if (ret) { // Disk is already present + ret->flags |= NETDATA_DISK_IS_HERE; + ret->last_update = current_time; + return; + } + + netdata_ebpf_disks_t *update_next = disk_list; + if (likely(disk_list)) { + netdata_ebpf_disks_t *move = disk_list; + while (move) { + if (dev == move->dev) + return; + + update_next = move; + move = move->next; + } + + w = callocz(1, sizeof(netdata_ebpf_disks_t)); + length = strlen(name); + if (length >= NETDATA_DISK_NAME_LEN) + length = NETDATA_DISK_NAME_LEN; + + memcpy(w->family, name, length); + w->family[length] = '\0'; + w->major = major; + w->minor = minor; + w->dev = netdata_new_encode_dev(major, minor); + update_next->next = w; + } else { + disk_list = callocz(1, sizeof(netdata_ebpf_disks_t)); + length = strlen(name); + if (length >= NETDATA_DISK_NAME_LEN) + length = NETDATA_DISK_NAME_LEN; + + memcpy(disk_list->family, name, length); + disk_list->family[length] = '\0'; + disk_list->major = major; + disk_list->minor = minor; + disk_list->dev = netdata_new_encode_dev(major, minor); + + w = disk_list; + } + + ebpf_read_disk_info(w, name); + + netdata_ebpf_disks_t *check; + check = (netdata_ebpf_disks_t *) avl_insert_lock(&disk_tree, (avl_t *)w); + if (check != w) + error("Internal error, cannot insert the AVL tree."); + +#ifdef NETDATA_INTERNAL_CHECKS + info("The Latency is monitoring the hard disk %s (Major = %d, Minor = %d, Device = %u)", name, major, minor,w->dev); +#endif + + w->flags |= NETDATA_DISK_IS_HERE; +} + +/** + * Read Local Disks + * + * Parse /proc/partitions to get block disks used to measure latency. + * + * @return It returns 0 on success and -1 otherwise + */ +static int read_local_disks() +{ + char filename[FILENAME_MAX + 1]; + snprintfz(filename, FILENAME_MAX, "%s%s", netdata_configured_host_prefix, NETDATA_EBPF_PROC_PARTITIONS); + procfile *ff = procfile_open(filename, " \t:", PROCFILE_FLAG_DEFAULT); + if (!ff) + return -1; + + ff = procfile_readall(ff); + if (!ff) + return -1; + + size_t lines = procfile_lines(ff), l; + time_t current_time = now_realtime_sec(); + for(l = 2; l < lines ;l++) { + size_t words = procfile_linewords(ff, l); + // This is header or end of file + if (unlikely(words < 4)) + continue; + + int major = (int)strtol(procfile_lineword(ff, l, 0), NULL, 10); + // The main goal of this thread is to measure block devices, so any block device with major number + // smaller than 7 according /proc/devices is not "important". + if (major > 7) { + int minor = (int)strtol(procfile_lineword(ff, l, 1), NULL, 10); + update_disk_table(procfile_lineword(ff, l, 3), major, minor, current_time); + } + } + + procfile_close(ff); + + return 0; +} + +/** + * Update disks + * + * @param em main thread structure + */ +void ebpf_update_disks(ebpf_module_t *em) +{ + static time_t update_every = 0; + time_t curr = now_realtime_sec(); + if (curr < update_every) + return; + + update_every = curr + 5 * em->update_every; + + (void)read_local_disks(); +} + +/***************************************************************** + * + * FUNCTIONS TO CLOSE THE THREAD + * + *****************************************************************/ + +/** + * Disk disable tracepoints + * + * Disable tracepoints when the plugin was responsible to enable it. + */ +static void ebpf_disk_disable_tracepoints() +{ + char *default_message = { "Cannot disable the tracepoint" }; + if (!was_block_issue_enabled) { + if (ebpf_disable_tracing_values(tracepoint_block_type, tracepoint_block_issue)) + error("%s %s/%s.", default_message, tracepoint_block_type, tracepoint_block_issue); + } + + if (!was_block_rq_complete_enabled) { + if (ebpf_disable_tracing_values(tracepoint_block_type, tracepoint_block_rq_complete)) + error("%s %s/%s.", default_message, tracepoint_block_type, tracepoint_block_rq_complete); + } +} + +/** + * Cleanup plot disks + * + * Clean disk list + */ +static void ebpf_cleanup_plot_disks() +{ + ebpf_publish_disk_t *move = plot_disks, *next; + while (move) { + next = move->next; + + freez(move); + + move = next; + } +} + +/** + * Cleanup Disk List + */ +static void ebpf_cleanup_disk_list() +{ + netdata_ebpf_disks_t *move = disk_list; + while (move) { + netdata_ebpf_disks_t *next = move->next; + + freez(move->histogram.name); + freez(move->boot_chart); + freez(move); + + move = next; + } +} + +/** + * DISK Free + * + * Cleanup variables after child threads to stop + * + * @param ptr thread data. + */ +static void ebpf_disk_free(ebpf_module_t *em) +{ + pthread_mutex_lock(&ebpf_exit_cleanup); + if (em->thread->enabled == NETDATA_THREAD_EBPF_RUNNING) { + em->thread->enabled = NETDATA_THREAD_EBPF_STOPPING; + pthread_mutex_unlock(&ebpf_exit_cleanup); + return; + } + pthread_mutex_unlock(&ebpf_exit_cleanup); + + ebpf_disk_disable_tracepoints(); + + if (dimensions) + ebpf_histogram_dimension_cleanup(dimensions, NETDATA_EBPF_HIST_MAX_BINS); + + freez(disk_hash_values); + freez(disk_threads.thread); + pthread_mutex_destroy(&plot_mutex); + + ebpf_cleanup_plot_disks(); + ebpf_cleanup_disk_list(); + + pthread_mutex_lock(&ebpf_exit_cleanup); + em->thread->enabled = NETDATA_THREAD_EBPF_STOPPED; + pthread_mutex_unlock(&ebpf_exit_cleanup); +} + +/** + * Disk exit. + * + * Cancel child and exit. + * + * @param ptr thread data. + */ +static void ebpf_disk_exit(void *ptr) +{ + ebpf_module_t *em = (ebpf_module_t *)ptr; + netdata_thread_cancel(*disk_threads.thread); + ebpf_disk_free(em); +} + +/** + * Disk Cleanup + * + * Clean up allocated memory. + * + * @param ptr thread data. + */ +static void ebpf_disk_cleanup(void *ptr) +{ + ebpf_module_t *em = (ebpf_module_t *)ptr; + ebpf_disk_free(em); +} + +/***************************************************************** + * + * MAIN LOOP + * + *****************************************************************/ + +/** + * Fill Plot list + * + * @param ptr a pointer for current disk + */ +static void ebpf_fill_plot_disks(netdata_ebpf_disks_t *ptr) +{ + pthread_mutex_lock(&plot_mutex); + ebpf_publish_disk_t *w; + if (likely(plot_disks)) { + ebpf_publish_disk_t *move = plot_disks, *store = plot_disks; + while (move) { + if (move->plot == ptr) { + pthread_mutex_unlock(&plot_mutex); + return; + } + + store = move; + move = move->next; + } + + w = callocz(1, sizeof(ebpf_publish_disk_t)); + w->plot = ptr; + store->next = w; + } else { + plot_disks = callocz(1, sizeof(ebpf_publish_disk_t)); + plot_disks->plot = ptr; + } + pthread_mutex_unlock(&plot_mutex); + + ptr->flags |= NETDATA_DISK_ADDED_TO_PLOT_LIST; +} + +/** + * Read hard disk table + * + * @param table file descriptor for table + * + * Read the table with number of calls for all functions + */ +static void read_hard_disk_tables(int table) +{ + netdata_idx_t *values = disk_hash_values; + block_key_t key = {}; + block_key_t next_key = {}; + + netdata_ebpf_disks_t *ret = NULL; + + while (bpf_map_get_next_key(table, &key, &next_key) == 0) { + int test = bpf_map_lookup_elem(table, &key, values); + if (test < 0) { + key = next_key; + continue; + } + + netdata_ebpf_disks_t find; + find.dev = key.dev; + + if (likely(ret)) { + if (find.dev != ret->dev) + ret = (netdata_ebpf_disks_t *)avl_search_lock(&disk_tree, (avl_t *)&find); + } else + ret = (netdata_ebpf_disks_t *)avl_search_lock(&disk_tree, (avl_t *)&find); + + // Disk was inserted after we parse /proc/partitions + if (!ret) { + if (read_local_disks()) { + key = next_key; + continue; + } + + ret = (netdata_ebpf_disks_t *)avl_search_lock(&disk_tree, (avl_t *)&find); + if (!ret) { + // We should never reach this point, but we are adding it to keep a safe code + key = next_key; + continue; + } + } + + uint64_t total = 0; + int i; + int end = (running_on_kernel < NETDATA_KERNEL_V4_15) ? 1 : ebpf_nprocs; + for (i = 0; i < end; i++) { + total += values[i]; + } + + ret->histogram.histogram[key.bin] = total; + + if (!(ret->flags & NETDATA_DISK_ADDED_TO_PLOT_LIST)) + ebpf_fill_plot_disks(ret); + + key = next_key; + } +} + +/** + * Disk read hash + * + * This is the thread callback. + * This thread is necessary, because we cannot freeze the whole plugin to read the data on very busy socket. + * + * @param ptr It is a NULL value for this thread. + * + * @return It always returns NULL. + */ +void *ebpf_disk_read_hash(void *ptr) +{ + netdata_thread_cleanup_push(ebpf_disk_cleanup, ptr); + heartbeat_t hb; + heartbeat_init(&hb); + + ebpf_module_t *em = (ebpf_module_t *)ptr; + + usec_t step = NETDATA_LATENCY_DISK_SLEEP_MS * em->update_every; + while (!ebpf_exit_plugin) { + (void)heartbeat_next(&hb, step); + + read_hard_disk_tables(disk_maps[NETDATA_DISK_READ].map_fd); + } + + netdata_thread_cleanup_pop(1); + return NULL; +} + +/** + * Obsolete Hard Disk charts + * + * Make Hard disk charts and fill chart name + * + * @param w the structure with necessary information to create the chart + * @param update_every value to overwrite the update frequency set by the server. + */ +static void ebpf_obsolete_hd_charts(netdata_ebpf_disks_t *w, int update_every) +{ + ebpf_write_chart_obsolete(w->histogram.name, w->family, w->histogram.title, EBPF_COMMON_DIMENSION_CALL, + w->family, NETDATA_EBPF_CHART_TYPE_STACKED, "disk.latency_io", + w->histogram.order, update_every); + + w->flags = 0; +} + +/** + * Create Hard Disk charts + * + * Make Hard disk charts and fill chart name + * + * @param w the structure with necessary information to create the chart + * @param update_every value to overwrite the update frequency set by the server. + */ +static void ebpf_create_hd_charts(netdata_ebpf_disks_t *w, int update_every) +{ + int order = NETDATA_CHART_PRIO_DISK_LATENCY; + char *family = w->family; + + w->histogram.name = strdupz("disk_latency_io"); + w->histogram.title = NULL; + w->histogram.order = order; + + ebpf_create_chart(w->histogram.name, family, "Disk latency", EBPF_COMMON_DIMENSION_CALL, + family, "disk.latency_io", NETDATA_EBPF_CHART_TYPE_STACKED, order, + ebpf_create_global_dimension, disk_publish_aggregated, NETDATA_EBPF_HIST_MAX_BINS, + update_every, NETDATA_EBPF_MODULE_NAME_DISK); + order++; + + w->flags |= NETDATA_DISK_CHART_CREATED; +} + +/** + * Remove pointer from plot + * + * Remove pointer from plot list when the disk is not present. + */ +static void ebpf_remove_pointer_from_plot_disk(ebpf_module_t *em) +{ + time_t current_time = now_realtime_sec(); + time_t limit = 10 * em->update_every; + pthread_mutex_lock(&plot_mutex); + ebpf_publish_disk_t *move = plot_disks, *prev = plot_disks; + int update_every = em->update_every; + while (move) { + netdata_ebpf_disks_t *ned = move->plot; + uint32_t flags = ned->flags; + + if (!(flags & NETDATA_DISK_IS_HERE) && ((current_time - ned->last_update) > limit)) { + ebpf_obsolete_hd_charts(ned, update_every); + avl_t *ret = (avl_t *)avl_remove_lock(&disk_tree, (avl_t *)ned); + UNUSED(ret); + if (move == plot_disks) { + freez(move); + plot_disks = NULL; + break; + } else { + prev->next = move->next; + ebpf_publish_disk_t *clean = move; + move = move->next; + freez(clean); + continue; + } + } + + prev = move; + move = move->next; + } + pthread_mutex_unlock(&plot_mutex); +} + +/** + * Send Hard disk data + * + * Send hard disk information to Netdata. + * + * @param update_every value to overwrite the update frequency set by the server. + */ +static void ebpf_latency_send_hd_data(int update_every) +{ + pthread_mutex_lock(&plot_mutex); + if (!plot_disks) { + pthread_mutex_unlock(&plot_mutex); + return; + } + + ebpf_publish_disk_t *move = plot_disks; + while (move) { + netdata_ebpf_disks_t *ned = move->plot; + uint32_t flags = ned->flags; + if (!(flags & NETDATA_DISK_CHART_CREATED)) { + ebpf_create_hd_charts(ned, update_every); + } + + if ((flags & NETDATA_DISK_CHART_CREATED)) { + write_histogram_chart(ned->histogram.name, ned->family, + ned->histogram.histogram, dimensions, NETDATA_EBPF_HIST_MAX_BINS); + } + + ned->flags &= ~NETDATA_DISK_IS_HERE; + + move = move->next; + } + pthread_mutex_unlock(&plot_mutex); +} + +/** +* Main loop for this collector. +*/ +static void disk_collector(ebpf_module_t *em) +{ + disk_hash_values = callocz(ebpf_nprocs, sizeof(netdata_idx_t)); + disk_threads.thread = mallocz(sizeof(netdata_thread_t)); + disk_threads.start_routine = ebpf_disk_read_hash; + + netdata_thread_create(disk_threads.thread, disk_threads.name, NETDATA_THREAD_OPTION_DEFAULT, + ebpf_disk_read_hash, em); + + int update_every = em->update_every; + heartbeat_t hb; + heartbeat_init(&hb); + usec_t step = update_every * USEC_PER_SEC; + while (!ebpf_exit_plugin) { + (void)heartbeat_next(&hb, step); + if (ebpf_exit_plugin) + break; + + pthread_mutex_lock(&lock); + ebpf_remove_pointer_from_plot_disk(em); + ebpf_latency_send_hd_data(update_every); + + pthread_mutex_unlock(&lock); + + ebpf_update_disks(em); + } +} + +/***************************************************************** + * + * EBPF DISK THREAD + * + *****************************************************************/ + +/** + * Enable tracepoints + * + * Enable necessary tracepoints for thread. + * + * @return It returns 0 on success and -1 otherwise + */ +static int ebpf_disk_enable_tracepoints() +{ + int test = ebpf_is_tracepoint_enabled(tracepoint_block_type, tracepoint_block_issue); + if (test == -1) + return -1; + else if (!test) { + if (ebpf_enable_tracing_values(tracepoint_block_type, tracepoint_block_issue)) + return -1; + } + was_block_issue_enabled = test; + + test = ebpf_is_tracepoint_enabled(tracepoint_block_type, tracepoint_block_rq_complete); + if (test == -1) + return -1; + else if (!test) { + if (ebpf_enable_tracing_values(tracepoint_block_type, tracepoint_block_rq_complete)) + return -1; + } + was_block_rq_complete_enabled = test; + + return 0; +} + +/** + * Disk thread + * + * Thread used to generate disk charts. + * + * @param ptr a pointer to `struct ebpf_module` + * + * @return It always return NULL + */ +void *ebpf_disk_thread(void *ptr) +{ + netdata_thread_cleanup_push(ebpf_disk_exit, ptr); + + ebpf_module_t *em = (ebpf_module_t *)ptr; + em->maps = disk_maps; + + if (ebpf_disk_enable_tracepoints()) { + em->thread->enabled = NETDATA_THREAD_EBPF_STOPPED; + goto enddisk; + } + + avl_init_lock(&disk_tree, ebpf_compare_disks); + if (read_local_disks()) { + em->thread->enabled = NETDATA_THREAD_EBPF_STOPPED; + goto enddisk; + } + + if (pthread_mutex_init(&plot_mutex, NULL)) { + em->thread->enabled = NETDATA_THREAD_EBPF_STOPPED; + error("Cannot initialize local mutex"); + goto enddisk; + } + + em->probe_links = ebpf_load_program(ebpf_plugin_dir, em, running_on_kernel, isrh, &em->objects); + if (!em->probe_links) { + em->thread->enabled = NETDATA_THREAD_EBPF_STOPPED; + goto enddisk; + } + + int algorithms[NETDATA_EBPF_HIST_MAX_BINS]; + ebpf_fill_algorithms(algorithms, NETDATA_EBPF_HIST_MAX_BINS, NETDATA_EBPF_INCREMENTAL_IDX); + dimensions = ebpf_fill_histogram_dimension(NETDATA_EBPF_HIST_MAX_BINS); + + ebpf_global_labels(disk_aggregated_data, disk_publish_aggregated, dimensions, dimensions, algorithms, + NETDATA_EBPF_HIST_MAX_BINS); + + pthread_mutex_lock(&lock); + ebpf_update_stats(&plugin_statistics, em); + pthread_mutex_unlock(&lock); + + disk_collector(em); + +enddisk: + ebpf_update_disabled_plugin_stats(em); + + netdata_thread_cleanup_pop(1); + + return NULL; +} diff --git a/collectors/ebpf.plugin/ebpf_disk.h b/collectors/ebpf.plugin/ebpf_disk.h new file mode 100644 index 0000000..c14b887 --- /dev/null +++ b/collectors/ebpf.plugin/ebpf_disk.h @@ -0,0 +1,78 @@ +// SPDX-License-Identifier: GPL-3.0-or-later + +#ifndef NETDATA_EBPF_DISK_H +#define NETDATA_EBPF_DISK_H 1 + +// Module name +#define NETDATA_EBPF_MODULE_NAME_DISK "disk" + +#include "libnetdata/avl/avl.h" +#include "libnetdata/ebpf/ebpf.h" + +#define NETDATA_EBPF_PROC_PARTITIONS "/proc/partitions" + +#define NETDATA_LATENCY_DISK_SLEEP_MS 650000ULL + +// Process configuration name +#define NETDATA_DISK_CONFIG_FILE "disk.conf" + +// Decode function extracted from: https://elixir.bootlin.com/linux/v5.10.8/source/include/linux/kdev_t.h#L7 +#define MINORBITS 20 +#define MKDEV(ma,mi) (((ma) << MINORBITS) | (mi)) + +enum netdata_latency_disks_flags { + NETDATA_DISK_ADDED_TO_PLOT_LIST = 1, + NETDATA_DISK_CHART_CREATED = 2, + NETDATA_DISK_IS_HERE = 4, + NETDATA_DISK_HAS_EFI = 8 +}; + +/* + * The definition (DISK_NAME_LEN) has been a stable value since Kernel 3.0, + * I decided to bring it as internal definition, to avoid include linux/genhd.h. + */ +#define NETDATA_DISK_NAME_LEN 32 +typedef struct netdata_ebpf_disks { + // Search + avl_t avl; + uint32_t dev; + uint32_t major; + uint32_t minor; + uint32_t bootsector_key; + uint64_t start; // start sector + uint64_t end; // end sector + + // Print information + char family[NETDATA_DISK_NAME_LEN + 1]; + char *boot_chart; + + netdata_ebpf_histogram_t histogram; + + uint32_t flags; + time_t last_update; + + struct netdata_ebpf_disks *main; + struct netdata_ebpf_disks *boot_partition; + struct netdata_ebpf_disks *next; +} netdata_ebpf_disks_t; + +enum ebpf_disk_tables { + NETDATA_DISK_READ +}; + +typedef struct block_key { + uint32_t bin; + uint32_t dev; +} block_key_t; + +typedef struct netdata_ebpf_publish_disk { + netdata_ebpf_disks_t *plot; + struct netdata_ebpf_publish_disk *next; +} ebpf_publish_disk_t; + +extern struct config disk_config; + +void *ebpf_disk_thread(void *ptr); + +#endif /* NETDATA_EBPF_DISK_H */ + diff --git a/collectors/ebpf.plugin/ebpf_fd.c b/collectors/ebpf.plugin/ebpf_fd.c new file mode 100644 index 0000000..30b7f22 --- /dev/null +++ b/collectors/ebpf.plugin/ebpf_fd.c @@ -0,0 +1,1183 @@ +// SPDX-License-Identifier: GPL-3.0-or-later + +#include "ebpf.h" +#include "ebpf_fd.h" + +static char *fd_dimension_names[NETDATA_FD_SYSCALL_END] = { "open", "close" }; +static char *fd_id_names[NETDATA_FD_SYSCALL_END] = { "do_sys_open", "__close_fd" }; + +static netdata_syscall_stat_t fd_aggregated_data[NETDATA_FD_SYSCALL_END]; +static netdata_publish_syscall_t fd_publish_aggregated[NETDATA_FD_SYSCALL_END]; + +static ebpf_local_maps_t fd_maps[] = {{.name = "tbl_fd_pid", .internal_input = ND_EBPF_DEFAULT_PID_SIZE, + .user_input = 0, + .type = NETDATA_EBPF_MAP_RESIZABLE | NETDATA_EBPF_MAP_PID, + .map_fd = ND_EBPF_MAP_FD_NOT_INITIALIZED}, + {.name = "tbl_fd_global", .internal_input = NETDATA_KEY_END_VECTOR, + .user_input = 0, .type = NETDATA_EBPF_MAP_STATIC, + .map_fd = ND_EBPF_MAP_FD_NOT_INITIALIZED}, + {.name = "fd_ctrl", .internal_input = NETDATA_CONTROLLER_END, + .user_input = 0, + .type = NETDATA_EBPF_MAP_CONTROLLER, + .map_fd = ND_EBPF_MAP_FD_NOT_INITIALIZED}, + {.name = NULL, .internal_input = 0, .user_input = 0, + .type = NETDATA_EBPF_MAP_CONTROLLER, + .map_fd = ND_EBPF_MAP_FD_NOT_INITIALIZED}}; + + +struct config fd_config = { .first_section = NULL, .last_section = NULL, .mutex = NETDATA_MUTEX_INITIALIZER, + .index = {.avl_tree = { .root = NULL, .compar = appconfig_section_compare }, + .rwlock = AVL_LOCK_INITIALIZER } }; + +struct netdata_static_thread fd_thread = {"FD KERNEL", + .config_section = NULL, + .config_name = NULL, + .env_name = NULL, + .enabled = 1, + .thread = NULL, + .init_routine = NULL, + .start_routine = NULL}; + +static netdata_idx_t fd_hash_values[NETDATA_FD_COUNTER]; +static netdata_idx_t *fd_values = NULL; + +netdata_fd_stat_t *fd_vector = NULL; +netdata_fd_stat_t **fd_pid = NULL; + +netdata_ebpf_targets_t fd_targets[] = { {.name = "open", .mode = EBPF_LOAD_TRAMPOLINE}, + {.name = "close", .mode = EBPF_LOAD_TRAMPOLINE}, + {.name = NULL, .mode = EBPF_LOAD_TRAMPOLINE}}; + +#ifdef LIBBPF_MAJOR_VERSION +#include "includes/fd.skel.h" // BTF code + +static struct fd_bpf *bpf_obj = NULL; + +/** + * Disable probe + * + * Disable all probes to use exclusively another method. + * + * @param obj is the main structure for bpf objects +*/ +static inline void ebpf_fd_disable_probes(struct fd_bpf *obj) +{ + bpf_program__set_autoload(obj->progs.netdata_sys_open_kprobe, false); + bpf_program__set_autoload(obj->progs.netdata_sys_open_kretprobe, false); + bpf_program__set_autoload(obj->progs.netdata_release_task_fd_kprobe, false); + if (running_on_kernel >= NETDATA_EBPF_KERNEL_5_11) { + bpf_program__set_autoload(obj->progs.netdata___close_fd_kretprobe, false); + bpf_program__set_autoload(obj->progs.netdata___close_fd_kprobe, false); + bpf_program__set_autoload(obj->progs.netdata_close_fd_kprobe, false); + } else { + bpf_program__set_autoload(obj->progs.netdata___close_fd_kprobe, false); + bpf_program__set_autoload(obj->progs.netdata_close_fd_kretprobe, false); + bpf_program__set_autoload(obj->progs.netdata_close_fd_kprobe, false); + } +} + +/* + * Disable specific probe + * + * Disable probes according the kernel version + * + * @param obj is the main structure for bpf objects + */ +static inline void ebpf_disable_specific_probes(struct fd_bpf *obj) +{ + if (running_on_kernel >= NETDATA_EBPF_KERNEL_5_11) { + bpf_program__set_autoload(obj->progs.netdata___close_fd_kretprobe, false); + bpf_program__set_autoload(obj->progs.netdata___close_fd_kprobe, false); + } else { + bpf_program__set_autoload(obj->progs.netdata_close_fd_kretprobe, false); + bpf_program__set_autoload(obj->progs.netdata_close_fd_kprobe, false); + } +} + +/* + * Disable trampoline + * + * Disable all trampoline to use exclusively another method. + * + * @param obj is the main structure for bpf objects. + */ +static inline void ebpf_disable_trampoline(struct fd_bpf *obj) +{ + bpf_program__set_autoload(obj->progs.netdata_sys_open_fentry, false); + bpf_program__set_autoload(obj->progs.netdata_sys_open_fexit, false); + bpf_program__set_autoload(obj->progs.netdata_close_fd_fentry, false); + bpf_program__set_autoload(obj->progs.netdata_close_fd_fexit, false); + bpf_program__set_autoload(obj->progs.netdata___close_fd_fentry, false); + bpf_program__set_autoload(obj->progs.netdata___close_fd_fexit, false); + bpf_program__set_autoload(obj->progs.netdata_release_task_fd_fentry, false); +} + +/* + * Disable specific trampoline + * + * Disable trampoline according to kernel version. + * + * @param obj is the main structure for bpf objects. + */ +static inline void ebpf_disable_specific_trampoline(struct fd_bpf *obj) +{ + if (running_on_kernel >= NETDATA_EBPF_KERNEL_5_11) { + bpf_program__set_autoload(obj->progs.netdata___close_fd_fentry, false); + bpf_program__set_autoload(obj->progs.netdata___close_fd_fexit, false); + } else { + bpf_program__set_autoload(obj->progs.netdata_close_fd_fentry, false); + bpf_program__set_autoload(obj->progs.netdata_close_fd_fexit, false); + } +} + +/** + * Set trampoline target + * + * Set the targets we will monitor. + * + * @param obj is the main structure for bpf objects. + */ +static void ebpf_set_trampoline_target(struct fd_bpf *obj) +{ + bpf_program__set_attach_target(obj->progs.netdata_sys_open_fentry, 0, fd_targets[NETDATA_FD_SYSCALL_OPEN].name); + bpf_program__set_attach_target(obj->progs.netdata_sys_open_fexit, 0, fd_targets[NETDATA_FD_SYSCALL_OPEN].name); + bpf_program__set_attach_target(obj->progs.netdata_release_task_fd_fentry, 0, EBPF_COMMON_FNCT_CLEAN_UP); + + if (running_on_kernel >= NETDATA_EBPF_KERNEL_5_11) { + bpf_program__set_attach_target( + obj->progs.netdata_close_fd_fentry, 0, fd_targets[NETDATA_FD_SYSCALL_CLOSE].name); + bpf_program__set_attach_target(obj->progs.netdata_close_fd_fexit, 0, fd_targets[NETDATA_FD_SYSCALL_CLOSE].name); + } else { + bpf_program__set_attach_target( + obj->progs.netdata___close_fd_fentry, 0, fd_targets[NETDATA_FD_SYSCALL_CLOSE].name); + bpf_program__set_attach_target( + obj->progs.netdata___close_fd_fexit, 0, fd_targets[NETDATA_FD_SYSCALL_CLOSE].name); + } +} + +/** + * Mount Attach Probe + * + * Attach probes to target + * + * @param obj is the main structure for bpf objects. + * + * @return It returns 0 on success and -1 otherwise. + */ +static int ebpf_fd_attach_probe(struct fd_bpf *obj) +{ + obj->links.netdata_sys_open_kprobe = bpf_program__attach_kprobe(obj->progs.netdata_sys_open_kprobe, false, + fd_targets[NETDATA_FD_SYSCALL_OPEN].name); + int ret = libbpf_get_error(obj->links.netdata_sys_open_kprobe); + if (ret) + return -1; + + obj->links.netdata_sys_open_kretprobe = bpf_program__attach_kprobe(obj->progs.netdata_sys_open_kretprobe, true, + fd_targets[NETDATA_FD_SYSCALL_OPEN].name); + ret = libbpf_get_error(obj->links.netdata_sys_open_kretprobe); + if (ret) + return -1; + + obj->links.netdata_release_task_fd_kprobe = bpf_program__attach_kprobe(obj->progs.netdata_release_task_fd_kprobe, + false, + EBPF_COMMON_FNCT_CLEAN_UP); + ret = libbpf_get_error(obj->links.netdata_release_task_fd_kprobe); + if (ret) + return -1; + + if (running_on_kernel >= NETDATA_EBPF_KERNEL_5_11) { + obj->links.netdata_close_fd_kretprobe = bpf_program__attach_kprobe(obj->progs.netdata_close_fd_kretprobe, true, + fd_targets[NETDATA_FD_SYSCALL_CLOSE].name); + ret = libbpf_get_error(obj->links.netdata_close_fd_kretprobe); + if (ret) + return -1; + + obj->links.netdata_close_fd_kprobe = bpf_program__attach_kprobe(obj->progs.netdata_close_fd_kprobe, false, + fd_targets[NETDATA_FD_SYSCALL_CLOSE].name); + ret = libbpf_get_error(obj->links.netdata_close_fd_kprobe); + if (ret) + return -1; + } else { + obj->links.netdata___close_fd_kretprobe = bpf_program__attach_kprobe(obj->progs.netdata___close_fd_kretprobe, + true, + fd_targets[NETDATA_FD_SYSCALL_CLOSE].name); + ret = libbpf_get_error(obj->links.netdata___close_fd_kretprobe); + if (ret) + return -1; + + obj->links.netdata___close_fd_kprobe = bpf_program__attach_kprobe(obj->progs.netdata___close_fd_kprobe, + false, + fd_targets[NETDATA_FD_SYSCALL_CLOSE].name); + ret = libbpf_get_error(obj->links.netdata___close_fd_kprobe); + if (ret) + return -1; + } + + return 0; +} + +/** + * Set target values + * + * Set pointers used to laod data. + */ +static void ebpf_fd_set_target_values() +{ + static char *close_targets[] = {"close_fd", "__close_fd"}; + static char *open_targets[] = {"do_sys_openat2", "do_sys_open"}; + if (running_on_kernel >= NETDATA_EBPF_KERNEL_5_11) { + fd_targets[NETDATA_FD_SYSCALL_OPEN].name = open_targets[0]; + fd_targets[NETDATA_FD_SYSCALL_CLOSE].name = close_targets[0]; + } else { + fd_targets[NETDATA_FD_SYSCALL_OPEN].name = open_targets[1]; + fd_targets[NETDATA_FD_SYSCALL_CLOSE].name = close_targets[1]; + } +} + +/** + * Set hash tables + * + * Set the values for maps according the value given by kernel. + * + * @param obj is the main structure for bpf objects. + */ +static void ebpf_fd_set_hash_tables(struct fd_bpf *obj) +{ + fd_maps[NETDATA_FD_GLOBAL_STATS].map_fd = bpf_map__fd(obj->maps.tbl_fd_global); + fd_maps[NETDATA_FD_PID_STATS].map_fd = bpf_map__fd(obj->maps.tbl_fd_pid); + fd_maps[NETDATA_FD_CONTROLLER].map_fd = bpf_map__fd(obj->maps.fd_ctrl); +} + +/** + * Adjust Map Size + * + * Resize maps according input from users. + * + * @param obj is the main structure for bpf objects. + * @param em structure with configuration + */ +static void ebpf_fd_adjust_map_size(struct fd_bpf *obj, ebpf_module_t *em) +{ + ebpf_update_map_size(obj->maps.tbl_fd_pid, &fd_maps[NETDATA_FD_PID_STATS], + em, bpf_map__name(obj->maps.tbl_fd_pid)); +} + +/** + * Disable Release Task + * + * Disable release task when apps is not enabled. + * + * @param obj is the main structure for bpf objects. + */ +static void ebpf_fd_disable_release_task(struct fd_bpf *obj) +{ + bpf_program__set_autoload(obj->progs.netdata_release_task_fd_kprobe, false); + bpf_program__set_autoload(obj->progs.netdata_release_task_fd_fentry, false); +} + +/** + * Load and attach + * + * Load and attach the eBPF code in kernel. + * + * @param obj is the main structure for bpf objects. + * @param em structure with configuration + * + * @return it returns 0 on succes and -1 otherwise + */ +static inline int ebpf_fd_load_and_attach(struct fd_bpf *obj, ebpf_module_t *em) +{ + netdata_ebpf_targets_t *mt = em->targets; + netdata_ebpf_program_loaded_t test = mt[NETDATA_FD_SYSCALL_OPEN].mode; + + ebpf_fd_set_target_values(); + if (test == EBPF_LOAD_TRAMPOLINE) { + ebpf_fd_disable_probes(obj); + ebpf_disable_specific_trampoline(obj); + + ebpf_set_trampoline_target(obj); + // TODO: Remove this in next PR, because this specific trampoline has an error. + bpf_program__set_autoload(obj->progs.netdata_release_task_fd_fentry, false); + } else { + ebpf_disable_trampoline(obj); + ebpf_disable_specific_probes(obj); + } + + ebpf_fd_adjust_map_size(obj, em); + + if (!em->apps_charts && !em->cgroup_charts) + ebpf_fd_disable_release_task(obj); + + int ret = fd_bpf__load(obj); + if (ret) { + return ret; + } + + ret = (test == EBPF_LOAD_TRAMPOLINE) ? fd_bpf__attach(obj) : ebpf_fd_attach_probe(obj); + if (!ret) { + ebpf_fd_set_hash_tables(obj); + + ebpf_update_controller(fd_maps[NETDATA_CACHESTAT_CTRL].map_fd, em); + } + + return ret; +} +#endif + +/***************************************************************** + * + * FUNCTIONS TO CLOSE THE THREAD + * + *****************************************************************/ + +/** + * FD Free + * + * Cleanup variables after child threads to stop + * + * @param ptr thread data. + */ +static void ebpf_fd_free(ebpf_module_t *em) +{ + pthread_mutex_lock(&ebpf_exit_cleanup); + if (em->thread->enabled == NETDATA_THREAD_EBPF_RUNNING) { + em->thread->enabled = NETDATA_THREAD_EBPF_STOPPING; + pthread_mutex_unlock(&ebpf_exit_cleanup); + return; + } + pthread_mutex_unlock(&ebpf_exit_cleanup); + + ebpf_cleanup_publish_syscall(fd_publish_aggregated); + freez(fd_thread.thread); + freez(fd_values); + freez(fd_vector); + +#ifdef LIBBPF_MAJOR_VERSION + if (bpf_obj) + fd_bpf__destroy(bpf_obj); +#endif + + pthread_mutex_lock(&ebpf_exit_cleanup); + em->thread->enabled = NETDATA_THREAD_EBPF_STOPPED; + pthread_mutex_unlock(&ebpf_exit_cleanup); +} + +/** + * FD Exit + * + * Cancel child thread and exit. + * + * @param ptr thread data. + */ +static void ebpf_fd_exit(void *ptr) +{ + ebpf_module_t *em = (ebpf_module_t *)ptr; + netdata_thread_cancel(*fd_thread.thread); + ebpf_fd_free(em); +} + +/** + * Clean up the main thread. + * + * @param ptr thread data. + */ +static void ebpf_fd_cleanup(void *ptr) +{ + ebpf_module_t *em = (ebpf_module_t *)ptr; + ebpf_fd_free(em); +} + +/***************************************************************** + * + * MAIN LOOP + * + *****************************************************************/ + +/** + * Send data to Netdata calling auxiliary functions. + * + * @param em the structure with thread information + */ +static void ebpf_fd_send_data(ebpf_module_t *em) +{ + fd_publish_aggregated[NETDATA_FD_SYSCALL_OPEN].ncall = fd_hash_values[NETDATA_KEY_CALLS_DO_SYS_OPEN]; + fd_publish_aggregated[NETDATA_FD_SYSCALL_OPEN].nerr = fd_hash_values[NETDATA_KEY_ERROR_DO_SYS_OPEN]; + + fd_publish_aggregated[NETDATA_FD_SYSCALL_CLOSE].ncall = fd_hash_values[NETDATA_KEY_CALLS_CLOSE_FD]; + fd_publish_aggregated[NETDATA_FD_SYSCALL_CLOSE].nerr = fd_hash_values[NETDATA_KEY_ERROR_CLOSE_FD]; + + write_count_chart(NETDATA_FILE_OPEN_CLOSE_COUNT, NETDATA_FILESYSTEM_FAMILY, fd_publish_aggregated, + NETDATA_FD_SYSCALL_END); + + if (em->mode < MODE_ENTRY) { + write_err_chart(NETDATA_FILE_OPEN_ERR_COUNT, NETDATA_FILESYSTEM_FAMILY, + fd_publish_aggregated, NETDATA_FD_SYSCALL_END); + } +} + +/** + * Read global counter + * + * Read the table with number of calls for all functions + */ +static void read_global_table() +{ + uint32_t idx; + netdata_idx_t *val = fd_hash_values; + netdata_idx_t *stored = fd_values; + int fd = fd_maps[NETDATA_FD_GLOBAL_STATS].map_fd; + + for (idx = NETDATA_KEY_CALLS_DO_SYS_OPEN; idx < NETDATA_FD_COUNTER; idx++) { + if (!bpf_map_lookup_elem(fd, &idx, stored)) { + int i; + int end = ebpf_nprocs; + netdata_idx_t total = 0; + for (i = 0; i < end; i++) + total += stored[i]; + + val[idx] = total; + } + } +} + +/** + * File descriptor read hash + * + * This is the thread callback. + * This thread is necessary, because we cannot freeze the whole plugin to read the data. + * + * @param ptr It is a NULL value for this thread. + * + * @return It always returns NULL. + */ +void *ebpf_fd_read_hash(void *ptr) +{ + netdata_thread_cleanup_push(ebpf_fd_cleanup, ptr); + heartbeat_t hb; + heartbeat_init(&hb); + + ebpf_module_t *em = (ebpf_module_t *)ptr; + usec_t step = NETDATA_FD_SLEEP_MS * em->update_every; + while (!ebpf_exit_plugin) { + (void)heartbeat_next(&hb, step); + + read_global_table(); + } + + netdata_thread_cleanup_pop(1); + return NULL; +} + +/** + * Apps Accumulator + * + * Sum all values read from kernel and store in the first address. + * + * @param out the vector with read values. + */ +static void fd_apps_accumulator(netdata_fd_stat_t *out) +{ + int i, end = (running_on_kernel >= NETDATA_KERNEL_V4_15) ? ebpf_nprocs : 1; + netdata_fd_stat_t *total = &out[0]; + for (i = 1; i < end; i++) { + netdata_fd_stat_t *w = &out[i]; + total->open_call += w->open_call; + total->close_call += w->close_call; + total->open_err += w->open_err; + total->close_err += w->close_err; + } +} + +/** + * Fill PID + * + * Fill PID structures + * + * @param current_pid pid that we are collecting data + * @param out values read from hash tables; + */ +static void fd_fill_pid(uint32_t current_pid, netdata_fd_stat_t *publish) +{ + netdata_fd_stat_t *curr = fd_pid[current_pid]; + if (!curr) { + curr = callocz(1, sizeof(netdata_fd_stat_t)); + fd_pid[current_pid] = curr; + } + + memcpy(curr, &publish[0], sizeof(netdata_fd_stat_t)); +} + +/** + * Read APPS table + * + * Read the apps table and store data inside the structure. + */ +static void read_apps_table() +{ + netdata_fd_stat_t *fv = fd_vector; + uint32_t key; + struct pid_stat *pids = root_of_pids; + int fd = fd_maps[NETDATA_FD_PID_STATS].map_fd; + size_t length = sizeof(netdata_fd_stat_t) * ebpf_nprocs; + while (pids) { + key = pids->pid; + + if (bpf_map_lookup_elem(fd, &key, fv)) { + pids = pids->next; + continue; + } + + fd_apps_accumulator(fv); + + fd_fill_pid(key, fv); + + // We are cleaning to avoid passing data read from one process to other. + memset(fv, 0, length); + + pids = pids->next; + } +} + +/** + * Update cgroup + * + * Update cgroup data based in + */ +static void ebpf_update_fd_cgroup() +{ + ebpf_cgroup_target_t *ect ; + netdata_fd_stat_t *fv = fd_vector; + int fd = fd_maps[NETDATA_FD_PID_STATS].map_fd; + size_t length = sizeof(netdata_fd_stat_t) * ebpf_nprocs; + + pthread_mutex_lock(&mutex_cgroup_shm); + for (ect = ebpf_cgroup_pids; ect; ect = ect->next) { + struct pid_on_target2 *pids; + for (pids = ect->pids; pids; pids = pids->next) { + int pid = pids->pid; + netdata_fd_stat_t *out = &pids->fd; + if (likely(fd_pid) && fd_pid[pid]) { + netdata_fd_stat_t *in = fd_pid[pid]; + + memcpy(out, in, sizeof(netdata_fd_stat_t)); + } else { + memset(fv, 0, length); + if (!bpf_map_lookup_elem(fd, &pid, fv)) { + fd_apps_accumulator(fv); + + memcpy(out, fv, sizeof(netdata_fd_stat_t)); + } + } + } + } + pthread_mutex_unlock(&mutex_cgroup_shm); +} + +/** + * Sum PIDs + * + * Sum values for all targets. + * + * @param fd the output + * @param root list of pids + */ +static void ebpf_fd_sum_pids(netdata_fd_stat_t *fd, struct pid_on_target *root) +{ + uint32_t open_call = 0; + uint32_t close_call = 0; + uint32_t open_err = 0; + uint32_t close_err = 0; + + while (root) { + int32_t pid = root->pid; + netdata_fd_stat_t *w = fd_pid[pid]; + if (w) { + open_call += w->open_call; + close_call += w->close_call; + open_err += w->open_err; + close_err += w->close_err; + } + + root = root->next; + } + + // These conditions were added, because we are using incremental algorithm + fd->open_call = (open_call >= fd->open_call) ? open_call : fd->open_call; + fd->close_call = (close_call >= fd->close_call) ? close_call : fd->close_call; + fd->open_err = (open_err >= fd->open_err) ? open_err : fd->open_err; + fd->close_err = (close_err >= fd->close_err) ? close_err : fd->close_err; +} + +/** + * Send data to Netdata calling auxiliary functions. + * + * @param em the structure with thread information + * @param root the target list. +*/ +void ebpf_fd_send_apps_data(ebpf_module_t *em, struct target *root) +{ + struct target *w; + for (w = root; w; w = w->next) { + if (unlikely(w->exposed && w->processes)) { + ebpf_fd_sum_pids(&w->fd, w->root_pid); + } + } + + write_begin_chart(NETDATA_APPS_FAMILY, NETDATA_SYSCALL_APPS_FILE_OPEN); + for (w = root; w; w = w->next) { + if (unlikely(w->exposed && w->processes)) { + write_chart_dimension(w->name, w->fd.open_call); + } + } + write_end_chart(); + + if (em->mode < MODE_ENTRY) { + write_begin_chart(NETDATA_APPS_FAMILY, NETDATA_SYSCALL_APPS_FILE_OPEN_ERROR); + for (w = root; w; w = w->next) { + if (unlikely(w->exposed && w->processes)) { + write_chart_dimension(w->name, w->fd.open_err); + } + } + write_end_chart(); + } + + write_begin_chart(NETDATA_APPS_FAMILY, NETDATA_SYSCALL_APPS_FILE_CLOSED); + for (w = root; w; w = w->next) { + if (unlikely(w->exposed && w->processes)) { + write_chart_dimension(w->name, w->fd.close_call); + } + } + write_end_chart(); + + if (em->mode < MODE_ENTRY) { + write_begin_chart(NETDATA_APPS_FAMILY, NETDATA_SYSCALL_APPS_FILE_CLOSE_ERROR); + for (w = root; w; w = w->next) { + if (unlikely(w->exposed && w->processes)) { + write_chart_dimension(w->name, w->fd.close_err); + } + } + write_end_chart(); + } +} + +/** + * Sum PIDs + * + * Sum values for all targets. + * + * @param fd structure used to store data + * @param pids input data + */ +static void ebpf_fd_sum_cgroup_pids(netdata_fd_stat_t *fd, struct pid_on_target2 *pids) +{ + netdata_fd_stat_t accumulator; + memset(&accumulator, 0, sizeof(accumulator)); + + while (pids) { + netdata_fd_stat_t *w = &pids->fd; + + accumulator.open_err += w->open_err; + accumulator.open_call += w->open_call; + accumulator.close_call += w->close_call; + accumulator.close_err += w->close_err; + + pids = pids->next; + } + + fd->open_call = (accumulator.open_call >= fd->open_call) ? accumulator.open_call : fd->open_call; + fd->open_err = (accumulator.open_err >= fd->open_err) ? accumulator.open_err : fd->open_err; + fd->close_call = (accumulator.close_call >= fd->close_call) ? accumulator.close_call : fd->close_call; + fd->close_err = (accumulator.close_err >= fd->close_err) ? accumulator.close_err : fd->close_err; +} + +/** + * Create specific file descriptor charts + * + * Create charts for cgroup/application. + * + * @param type the chart type. + * @param em the main thread structure. + */ +static void ebpf_create_specific_fd_charts(char *type, ebpf_module_t *em) +{ + ebpf_create_chart(type, NETDATA_SYSCALL_APPS_FILE_OPEN, "Number of open files", + EBPF_COMMON_DIMENSION_CALL, NETDATA_APPS_FILE_CGROUP_GROUP, + NETDATA_CGROUP_FD_OPEN_CONTEXT, NETDATA_EBPF_CHART_TYPE_LINE, + NETDATA_CHART_PRIO_CGROUPS_CONTAINERS + 5400, + ebpf_create_global_dimension, + &fd_publish_aggregated[NETDATA_FD_SYSCALL_OPEN], + 1, em->update_every, NETDATA_EBPF_MODULE_NAME_SWAP); + + if (em->mode < MODE_ENTRY) { + ebpf_create_chart(type, NETDATA_SYSCALL_APPS_FILE_OPEN_ERROR, "Fails to open files", + EBPF_COMMON_DIMENSION_CALL, NETDATA_APPS_FILE_CGROUP_GROUP, + NETDATA_CGROUP_FD_OPEN_ERR_CONTEXT, NETDATA_EBPF_CHART_TYPE_LINE, + NETDATA_CHART_PRIO_CGROUPS_CONTAINERS + 5401, + ebpf_create_global_dimension, + &fd_publish_aggregated[NETDATA_FD_SYSCALL_OPEN], + 1, em->update_every, + NETDATA_EBPF_MODULE_NAME_SWAP); + } + + ebpf_create_chart(type, NETDATA_SYSCALL_APPS_FILE_CLOSED, "Files closed", + EBPF_COMMON_DIMENSION_CALL, NETDATA_APPS_FILE_CGROUP_GROUP, + NETDATA_CGROUP_FD_CLOSE_CONTEXT, NETDATA_EBPF_CHART_TYPE_LINE, + NETDATA_CHART_PRIO_CGROUPS_CONTAINERS + 5402, + ebpf_create_global_dimension, + &fd_publish_aggregated[NETDATA_FD_SYSCALL_CLOSE], + 1, em->update_every, NETDATA_EBPF_MODULE_NAME_SWAP); + + if (em->mode < MODE_ENTRY) { + ebpf_create_chart(type, NETDATA_SYSCALL_APPS_FILE_CLOSE_ERROR, "Fails to close files", + EBPF_COMMON_DIMENSION_CALL, NETDATA_APPS_FILE_CGROUP_GROUP, + NETDATA_CGROUP_FD_CLOSE_ERR_CONTEXT, NETDATA_EBPF_CHART_TYPE_LINE, + NETDATA_CHART_PRIO_CGROUPS_CONTAINERS + 5403, + ebpf_create_global_dimension, + &fd_publish_aggregated[NETDATA_FD_SYSCALL_CLOSE], + 1, em->update_every, + NETDATA_EBPF_MODULE_NAME_SWAP); + } +} + +/** + * Obsolete specific file descriptor charts + * + * Obsolete charts for cgroup/application. + * + * @param type the chart type. + * @param em the main thread structure. + */ +static void ebpf_obsolete_specific_fd_charts(char *type, ebpf_module_t *em) +{ + ebpf_write_chart_obsolete(type, NETDATA_SYSCALL_APPS_FILE_OPEN, "Number of open files", + EBPF_COMMON_DIMENSION_CALL, NETDATA_APPS_FILE_GROUP, + NETDATA_EBPF_CHART_TYPE_LINE, NETDATA_CGROUP_FD_OPEN_CONTEXT, + NETDATA_CHART_PRIO_CGROUPS_CONTAINERS + 5400, em->update_every); + + if (em->mode < MODE_ENTRY) { + ebpf_write_chart_obsolete(type, NETDATA_SYSCALL_APPS_FILE_OPEN_ERROR, "Fails to open files", + EBPF_COMMON_DIMENSION_CALL, NETDATA_APPS_FILE_GROUP, + NETDATA_EBPF_CHART_TYPE_LINE, NETDATA_CGROUP_FD_OPEN_ERR_CONTEXT, + NETDATA_CHART_PRIO_CGROUPS_CONTAINERS + 5401, em->update_every); + } + + ebpf_write_chart_obsolete(type, NETDATA_SYSCALL_APPS_FILE_CLOSED, "Files closed", + EBPF_COMMON_DIMENSION_CALL, NETDATA_APPS_FILE_GROUP, + NETDATA_EBPF_CHART_TYPE_LINE, NETDATA_CGROUP_FD_CLOSE_CONTEXT, + NETDATA_CHART_PRIO_CGROUPS_CONTAINERS + 5402, em->update_every); + + if (em->mode < MODE_ENTRY) { + ebpf_write_chart_obsolete(type, NETDATA_SYSCALL_APPS_FILE_CLOSE_ERROR, "Fails to close files", + EBPF_COMMON_DIMENSION_CALL, NETDATA_APPS_FILE_GROUP, + NETDATA_EBPF_CHART_TYPE_LINE, NETDATA_CGROUP_FD_CLOSE_ERR_CONTEXT, + NETDATA_CHART_PRIO_CGROUPS_CONTAINERS + 5403, em->update_every); + } +} + +/* + * Send specific file descriptor data + * + * Send data for specific cgroup/apps. + * + * @param type chart type + * @param values structure with values that will be sent to netdata + */ +static void ebpf_send_specific_fd_data(char *type, netdata_fd_stat_t *values, ebpf_module_t *em) +{ + write_begin_chart(type, NETDATA_SYSCALL_APPS_FILE_OPEN); + write_chart_dimension(fd_publish_aggregated[NETDATA_FD_SYSCALL_OPEN].name, (long long)values->open_call); + write_end_chart(); + + if (em->mode < MODE_ENTRY) { + write_begin_chart(type, NETDATA_SYSCALL_APPS_FILE_OPEN_ERROR); + write_chart_dimension(fd_publish_aggregated[NETDATA_FD_SYSCALL_OPEN].name, (long long)values->open_err); + write_end_chart(); + } + + write_begin_chart(type, NETDATA_SYSCALL_APPS_FILE_CLOSED); + write_chart_dimension(fd_publish_aggregated[NETDATA_FD_SYSCALL_CLOSE].name, (long long)values->close_call); + write_end_chart(); + + if (em->mode < MODE_ENTRY) { + write_begin_chart(type, NETDATA_SYSCALL_APPS_FILE_CLOSE_ERROR); + write_chart_dimension(fd_publish_aggregated[NETDATA_FD_SYSCALL_CLOSE].name, (long long)values->close_err); + write_end_chart(); + } +} + +/** + * Create systemd file descriptor charts + * + * Create charts when systemd is enabled + * + * @param em the main collector structure + **/ +static void ebpf_create_systemd_fd_charts(ebpf_module_t *em) +{ + ebpf_create_charts_on_systemd(NETDATA_SYSCALL_APPS_FILE_OPEN, "Number of open files", + EBPF_COMMON_DIMENSION_CALL, NETDATA_APPS_FILE_CGROUP_GROUP, + NETDATA_EBPF_CHART_TYPE_STACKED, 20061, + ebpf_algorithms[NETDATA_EBPF_INCREMENTAL_IDX], NETDATA_SYSTEMD_FD_OPEN_CONTEXT, + NETDATA_EBPF_MODULE_NAME_PROCESS, em->update_every); + + if (em->mode < MODE_ENTRY) { + ebpf_create_charts_on_systemd(NETDATA_SYSCALL_APPS_FILE_OPEN_ERROR, "Fails to open files", + EBPF_COMMON_DIMENSION_CALL, NETDATA_APPS_FILE_CGROUP_GROUP, + NETDATA_EBPF_CHART_TYPE_STACKED, 20062, + ebpf_algorithms[NETDATA_EBPF_INCREMENTAL_IDX], NETDATA_SYSTEMD_FD_OPEN_ERR_CONTEXT, + NETDATA_EBPF_MODULE_NAME_PROCESS, em->update_every); + } + + ebpf_create_charts_on_systemd(NETDATA_SYSCALL_APPS_FILE_CLOSED, "Files closed", + EBPF_COMMON_DIMENSION_CALL, NETDATA_APPS_FILE_CGROUP_GROUP, + NETDATA_EBPF_CHART_TYPE_STACKED, 20063, + ebpf_algorithms[NETDATA_EBPF_INCREMENTAL_IDX], NETDATA_SYSTEMD_FD_CLOSE_CONTEXT, + NETDATA_EBPF_MODULE_NAME_PROCESS, em->update_every); + + if (em->mode < MODE_ENTRY) { + ebpf_create_charts_on_systemd(NETDATA_SYSCALL_APPS_FILE_CLOSE_ERROR, "Fails to close files", + EBPF_COMMON_DIMENSION_CALL, NETDATA_APPS_FILE_CGROUP_GROUP, + NETDATA_EBPF_CHART_TYPE_STACKED, 20064, + ebpf_algorithms[NETDATA_EBPF_INCREMENTAL_IDX], NETDATA_SYSTEMD_FD_CLOSE_ERR_CONTEXT, + NETDATA_EBPF_MODULE_NAME_PROCESS, em->update_every); + } +} + +/** + * Send Systemd charts + * + * Send collected data to Netdata. + * + * @param em the main collector structure + */ +static void ebpf_send_systemd_fd_charts(ebpf_module_t *em) +{ + ebpf_cgroup_target_t *ect; + write_begin_chart(NETDATA_SERVICE_FAMILY, NETDATA_SYSCALL_APPS_FILE_OPEN); + for (ect = ebpf_cgroup_pids; ect ; ect = ect->next) { + if (unlikely(ect->systemd) && unlikely(ect->updated)) { + write_chart_dimension(ect->name, ect->publish_systemd_fd.open_call); + } + } + write_end_chart(); + + if (em->mode < MODE_ENTRY) { + write_begin_chart(NETDATA_SERVICE_FAMILY, NETDATA_SYSCALL_APPS_FILE_OPEN_ERROR); + for (ect = ebpf_cgroup_pids; ect ; ect = ect->next) { + if (unlikely(ect->systemd) && unlikely(ect->updated)) { + write_chart_dimension(ect->name, ect->publish_systemd_fd.open_err); + } + } + write_end_chart(); + } + + write_begin_chart(NETDATA_SERVICE_FAMILY, NETDATA_SYSCALL_APPS_FILE_CLOSED); + for (ect = ebpf_cgroup_pids; ect ; ect = ect->next) { + if (unlikely(ect->systemd) && unlikely(ect->updated)) { + write_chart_dimension(ect->name, ect->publish_systemd_fd.close_call); + } + } + write_end_chart(); + + if (em->mode < MODE_ENTRY) { + write_begin_chart(NETDATA_SERVICE_FAMILY, NETDATA_SYSCALL_APPS_FILE_CLOSE_ERROR); + for (ect = ebpf_cgroup_pids; ect ; ect = ect->next) { + if (unlikely(ect->systemd) && unlikely(ect->updated)) { + write_chart_dimension(ect->name, ect->publish_systemd_fd.close_err); + } + } + write_end_chart(); + } +} + +/** + * Send data to Netdata calling auxiliary functions. + * + * @param em the main collector structure +*/ +static void ebpf_fd_send_cgroup_data(ebpf_module_t *em) +{ + if (!ebpf_cgroup_pids) + return; + + pthread_mutex_lock(&mutex_cgroup_shm); + ebpf_cgroup_target_t *ect; + for (ect = ebpf_cgroup_pids; ect ; ect = ect->next) { + ebpf_fd_sum_cgroup_pids(&ect->publish_systemd_fd, ect->pids); + } + + int has_systemd = shm_ebpf_cgroup.header->systemd_enabled; + if (has_systemd) { + if (send_cgroup_chart) { + ebpf_create_systemd_fd_charts(em); + } + + ebpf_send_systemd_fd_charts(em); + } + + for (ect = ebpf_cgroup_pids; ect ; ect = ect->next) { + if (ect->systemd) + continue; + + if (!(ect->flags & NETDATA_EBPF_CGROUP_HAS_FD_CHART) && ect->updated) { + ebpf_create_specific_fd_charts(ect->name, em); + ect->flags |= NETDATA_EBPF_CGROUP_HAS_FD_CHART; + } + + if (ect->flags & NETDATA_EBPF_CGROUP_HAS_FD_CHART ) { + if (ect->updated) { + ebpf_send_specific_fd_data(ect->name, &ect->publish_systemd_fd, em); + } else { + ebpf_obsolete_specific_fd_charts(ect->name, em); + ect->flags &= ~NETDATA_EBPF_CGROUP_HAS_FD_CHART; + } + } + } + + pthread_mutex_unlock(&mutex_cgroup_shm); +} + +/** +* Main loop for this collector. +*/ +static void fd_collector(ebpf_module_t *em) +{ + fd_thread.thread = mallocz(sizeof(netdata_thread_t)); + fd_thread.start_routine = ebpf_fd_read_hash; + + netdata_thread_create(fd_thread.thread, fd_thread.name, NETDATA_THREAD_OPTION_DEFAULT, + ebpf_fd_read_hash, em); + + int cgroups = em->cgroup_charts; + heartbeat_t hb; + heartbeat_init(&hb); + usec_t step = em->update_every * USEC_PER_SEC; + while (!ebpf_exit_plugin) { + (void)heartbeat_next(&hb, step); + if (ebpf_exit_plugin) + break; + + netdata_apps_integration_flags_t apps = em->apps_charts; + pthread_mutex_lock(&collect_data_mutex); + if (apps) + read_apps_table(); + + if (cgroups) + ebpf_update_fd_cgroup(); + + pthread_mutex_lock(&lock); + + ebpf_fd_send_data(em); + + if (apps & NETDATA_EBPF_APPS_FLAG_CHART_CREATED) + ebpf_fd_send_apps_data(em, apps_groups_root_target); + + if (cgroups) + ebpf_fd_send_cgroup_data(em); + + pthread_mutex_unlock(&lock); + pthread_mutex_unlock(&collect_data_mutex); + } +} + +/***************************************************************** + * + * CREATE CHARTS + * + *****************************************************************/ + +/** + * Create apps charts + * + * Call ebpf_create_chart to create the charts on apps submenu. + * + * @param em a pointer to the structure with the default values. + */ +void ebpf_fd_create_apps_charts(struct ebpf_module *em, void *ptr) +{ + struct target *root = ptr; + ebpf_create_charts_on_apps(NETDATA_SYSCALL_APPS_FILE_OPEN, + "Number of open files", + EBPF_COMMON_DIMENSION_CALL, + NETDATA_APPS_FILE_GROUP, + NETDATA_EBPF_CHART_TYPE_STACKED, + 20061, + ebpf_algorithms[NETDATA_EBPF_INCREMENTAL_IDX], + root, em->update_every, NETDATA_EBPF_MODULE_NAME_PROCESS); + + if (em->mode < MODE_ENTRY) { + ebpf_create_charts_on_apps(NETDATA_SYSCALL_APPS_FILE_OPEN_ERROR, + "Fails to open files", + EBPF_COMMON_DIMENSION_CALL, + NETDATA_APPS_FILE_GROUP, + NETDATA_EBPF_CHART_TYPE_STACKED, + 20062, + ebpf_algorithms[NETDATA_EBPF_INCREMENTAL_IDX], + root, em->update_every, NETDATA_EBPF_MODULE_NAME_PROCESS); + } + + ebpf_create_charts_on_apps(NETDATA_SYSCALL_APPS_FILE_CLOSED, + "Files closed", + EBPF_COMMON_DIMENSION_CALL, + NETDATA_APPS_FILE_GROUP, + NETDATA_EBPF_CHART_TYPE_STACKED, + 20063, + ebpf_algorithms[NETDATA_EBPF_INCREMENTAL_IDX], + root, em->update_every, NETDATA_EBPF_MODULE_NAME_PROCESS); + + if (em->mode < MODE_ENTRY) { + ebpf_create_charts_on_apps(NETDATA_SYSCALL_APPS_FILE_CLOSE_ERROR, + "Fails to close files", + EBPF_COMMON_DIMENSION_CALL, + NETDATA_APPS_FILE_GROUP, + NETDATA_EBPF_CHART_TYPE_STACKED, + 20064, + ebpf_algorithms[NETDATA_EBPF_INCREMENTAL_IDX], + root, em->update_every, NETDATA_EBPF_MODULE_NAME_PROCESS); + } + + em->apps_charts |= NETDATA_EBPF_APPS_FLAG_CHART_CREATED; +} + +/** + * Create global charts + * + * Call ebpf_create_chart to create the charts for the collector. + * + * @param em a pointer to the structure with the default values. + */ +static void ebpf_create_fd_global_charts(ebpf_module_t *em) +{ + ebpf_create_chart(NETDATA_FILESYSTEM_FAMILY, + NETDATA_FILE_OPEN_CLOSE_COUNT, + "Open and close calls", + EBPF_COMMON_DIMENSION_CALL, + NETDATA_FILE_GROUP, + NULL, + NETDATA_EBPF_CHART_TYPE_LINE, + NETDATA_CHART_PRIO_EBPF_FD_CHARTS, + ebpf_create_global_dimension, + fd_publish_aggregated, + NETDATA_FD_SYSCALL_END, + em->update_every, NETDATA_EBPF_MODULE_NAME_FD); + + if (em->mode < MODE_ENTRY) { + ebpf_create_chart(NETDATA_FILESYSTEM_FAMILY, + NETDATA_FILE_OPEN_ERR_COUNT, + "Open fails", + EBPF_COMMON_DIMENSION_CALL, + NETDATA_FILE_GROUP, + NULL, + NETDATA_EBPF_CHART_TYPE_LINE, + NETDATA_CHART_PRIO_EBPF_FD_CHARTS + 1, + ebpf_create_global_dimension, + fd_publish_aggregated, + NETDATA_FD_SYSCALL_END, + em->update_every, NETDATA_EBPF_MODULE_NAME_FD); + } +} + +/***************************************************************** + * + * MAIN THREAD + * + *****************************************************************/ + +/** + * Allocate vectors used with this thread. + * + * We are not testing the return, because callocz does this and shutdown the software + * case it was not possible to allocate. + * + * @param apps is apps enabled? + */ +static void ebpf_fd_allocate_global_vectors(int apps) +{ + if (apps) + fd_pid = callocz((size_t)pid_max, sizeof(netdata_fd_stat_t *)); + + fd_vector = callocz((size_t)ebpf_nprocs, sizeof(netdata_fd_stat_t)); + + fd_values = callocz((size_t)ebpf_nprocs, sizeof(netdata_idx_t)); +} + +/* + * Load BPF + * + * Load BPF files. + * + * @param em the structure with configuration + */ +static int ebpf_fd_load_bpf(ebpf_module_t *em) +{ + int ret = 0; + ebpf_adjust_apps_cgroup(em, em->targets[NETDATA_FD_SYSCALL_OPEN].mode); + if (em->load & EBPF_LOAD_LEGACY) { + em->probe_links = ebpf_load_program(ebpf_plugin_dir, em, running_on_kernel, isrh, &em->objects); + if (!em->probe_links) { + em->enabled = CONFIG_BOOLEAN_NO; + ret = -1; + } + } +#ifdef LIBBPF_MAJOR_VERSION + else { + bpf_obj = fd_bpf__open(); + if (!bpf_obj) + ret = -1; + else + ret = ebpf_fd_load_and_attach(bpf_obj, em); + } +#endif + + if (ret) + error("%s %s", EBPF_DEFAULT_ERROR_MSG, em->thread_name); + + return ret; +} + +/** + * Directory Cache thread + * + * Thread used to make dcstat thread + * + * @param ptr a pointer to `struct ebpf_module` + * + * @return It always returns NULL + */ +void *ebpf_fd_thread(void *ptr) +{ + netdata_thread_cleanup_push(ebpf_fd_exit, ptr); + + ebpf_module_t *em = (ebpf_module_t *)ptr; + em->maps = fd_maps; + +#ifdef LIBBPF_MAJOR_VERSION + ebpf_adjust_thread_load(em, default_btf); +#endif + if (ebpf_fd_load_bpf(em)) { + em->thread->enabled = NETDATA_THREAD_EBPF_STOPPED; + goto endfd; + } + + ebpf_fd_allocate_global_vectors(em->apps_charts); + + int algorithms[NETDATA_FD_SYSCALL_END] = { + NETDATA_EBPF_INCREMENTAL_IDX, NETDATA_EBPF_INCREMENTAL_IDX + }; + + ebpf_global_labels(fd_aggregated_data, fd_publish_aggregated, fd_dimension_names, fd_id_names, + algorithms, NETDATA_FD_SYSCALL_END); + + pthread_mutex_lock(&lock); + ebpf_create_fd_global_charts(em); + ebpf_update_stats(&plugin_statistics, em); + pthread_mutex_unlock(&lock); + + fd_collector(em); + +endfd: + ebpf_update_disabled_plugin_stats(em); + + netdata_thread_cleanup_pop(1); + return NULL; +} diff --git a/collectors/ebpf.plugin/ebpf_fd.h b/collectors/ebpf.plugin/ebpf_fd.h new file mode 100644 index 0000000..914a34b --- /dev/null +++ b/collectors/ebpf.plugin/ebpf_fd.h @@ -0,0 +1,85 @@ +// SPDX-License-Identifier: GPL-3.0-or-later + +#ifndef NETDATA_EBPF_FD_H +#define NETDATA_EBPF_FD_H 1 + +// Module name +#define NETDATA_EBPF_MODULE_NAME_FD "filedescriptor" + +#define NETDATA_FD_SLEEP_MS 850000ULL + +// Menu group +#define NETDATA_FILE_GROUP "file_access" + +// Global chart name +#define NETDATA_FILE_OPEN_CLOSE_COUNT "file_descriptor" +#define NETDATA_FILE_OPEN_ERR_COUNT "file_error" + +// Charts created on Apps submenu +#define NETDATA_SYSCALL_APPS_FILE_OPEN "file_open" +#define NETDATA_SYSCALL_APPS_FILE_CLOSED "file_closed" +#define NETDATA_SYSCALL_APPS_FILE_OPEN_ERROR "file_open_error" +#define NETDATA_SYSCALL_APPS_FILE_CLOSE_ERROR "file_close_error" + +// Process configuration name +#define NETDATA_FD_CONFIG_FILE "fd.conf" + +// Contexts +#define NETDATA_CGROUP_FD_OPEN_CONTEXT "cgroup.fd_open" +#define NETDATA_CGROUP_FD_OPEN_ERR_CONTEXT "cgroup.fd_open_error" +#define NETDATA_CGROUP_FD_CLOSE_CONTEXT "cgroup.fd_close" +#define NETDATA_CGROUP_FD_CLOSE_ERR_CONTEXT "cgroup.fd_close_error" + +#define NETDATA_SYSTEMD_FD_OPEN_CONTEXT "services.fd_open" +#define NETDATA_SYSTEMD_FD_OPEN_ERR_CONTEXT "services.fd_open_error" +#define NETDATA_SYSTEMD_FD_CLOSE_CONTEXT "services.fd_close" +#define NETDATA_SYSTEMD_FD_CLOSE_ERR_CONTEXT "services.fd_close_error" + +typedef struct netdata_fd_stat { + uint64_t pid_tgid; // Unique identifier + uint32_t pid; // Process ID + + uint32_t open_call; // Open syscalls (open and openat) + uint32_t close_call; // Close syscall (close) + + // Errors + uint32_t open_err; + uint32_t close_err; +} netdata_fd_stat_t; + +enum fd_tables { + NETDATA_FD_PID_STATS, + NETDATA_FD_GLOBAL_STATS, + + // Keep this as last and don't skip numbers as it is used as element counter + NETDATA_FD_CONTROLLER +}; + +enum fd_counters { + NETDATA_KEY_CALLS_DO_SYS_OPEN, + NETDATA_KEY_ERROR_DO_SYS_OPEN, + + NETDATA_KEY_CALLS_CLOSE_FD, + NETDATA_KEY_ERROR_CLOSE_FD, + + // Keep this as last and don't skip numbers as it is used as element counter + NETDATA_FD_COUNTER +}; + +enum fd_syscalls { + NETDATA_FD_SYSCALL_OPEN, + NETDATA_FD_SYSCALL_CLOSE, + + // Do not insert nothing after this value + NETDATA_FD_SYSCALL_END +}; + + +void *ebpf_fd_thread(void *ptr); +void ebpf_fd_create_apps_charts(struct ebpf_module *em, void *ptr); +extern struct config fd_config; +extern netdata_fd_stat_t **fd_pid; +extern netdata_ebpf_targets_t fd_targets[]; + +#endif /* NETDATA_EBPF_FD_H */ + diff --git a/collectors/ebpf.plugin/ebpf_filesystem.c b/collectors/ebpf.plugin/ebpf_filesystem.c new file mode 100644 index 0000000..7dbec74 --- /dev/null +++ b/collectors/ebpf.plugin/ebpf_filesystem.c @@ -0,0 +1,638 @@ +// SPDX-License-Identifier: GPL-3.0-or-later + +#include "ebpf_filesystem.h" + +struct config fs_config = { .first_section = NULL, + .last_section = NULL, + .mutex = NETDATA_MUTEX_INITIALIZER, + .index = { .avl_tree = { .root = NULL, .compar = appconfig_section_compare }, + .rwlock = AVL_LOCK_INITIALIZER } }; + +static ebpf_local_maps_t fs_maps[] = {{.name = "tbl_ext4", .internal_input = NETDATA_KEY_CALLS_SYNC, + .user_input = 0, .type = NETDATA_EBPF_MAP_STATIC, + .map_fd = ND_EBPF_MAP_FD_NOT_INITIALIZED}, + {.name = "tbl_xfs", .internal_input = NETDATA_KEY_CALLS_SYNC, + .user_input = 0, .type = NETDATA_EBPF_MAP_STATIC, + .map_fd = ND_EBPF_MAP_FD_NOT_INITIALIZED}, + {.name = "tbl_nfs", .internal_input = NETDATA_KEY_CALLS_SYNC, + .user_input = 0, .type = NETDATA_EBPF_MAP_STATIC, + .map_fd = ND_EBPF_MAP_FD_NOT_INITIALIZED}, + {.name = "tbl_zfs", .internal_input = NETDATA_KEY_CALLS_SYNC, + .user_input = 0, .type = NETDATA_EBPF_MAP_STATIC, + .map_fd = ND_EBPF_MAP_FD_NOT_INITIALIZED}, + {.name = "tbl_btrfs", .internal_input = NETDATA_KEY_CALLS_SYNC, + .user_input = 0, .type = NETDATA_EBPF_MAP_STATIC, + .map_fd = ND_EBPF_MAP_FD_NOT_INITIALIZED}, + {.name = "tbl_ext_addr", .internal_input = 1, + .user_input = 0, .type = NETDATA_EBPF_MAP_STATIC, + .map_fd = ND_EBPF_MAP_FD_NOT_INITIALIZED}, + {.name = NULL, .internal_input = 0, .user_input = 0, + .type = NETDATA_EBPF_MAP_CONTROLLER, + .map_fd = ND_EBPF_MAP_FD_NOT_INITIALIZED}}; + +struct netdata_static_thread filesystem_threads = { + .name = "EBPF FS READ", + .config_section = NULL, + .config_name = NULL, + .env_name = NULL, + .enabled = 1, + .thread = NULL, + .init_routine = NULL, + .start_routine = NULL +}; + +static netdata_syscall_stat_t filesystem_aggregated_data[NETDATA_EBPF_HIST_MAX_BINS]; +static netdata_publish_syscall_t filesystem_publish_aggregated[NETDATA_EBPF_HIST_MAX_BINS]; + +char **dimensions = NULL; +static netdata_idx_t *filesystem_hash_values = NULL; + +/***************************************************************** + * + * COMMON FUNCTIONS + * + *****************************************************************/ + +/** + * Create Filesystem chart + * + * Create latency charts + * + * @param update_every value to overwrite the update frequency set by the server. + */ +static void ebpf_obsolete_fs_charts(int update_every) +{ + int i; + uint32_t test = NETDATA_FILESYSTEM_FLAG_CHART_CREATED | NETDATA_FILESYSTEM_REMOVE_CHARTS; + for (i = 0; localfs[i].filesystem; i++) { + ebpf_filesystem_partitions_t *efp = &localfs[i]; + uint32_t flags = efp->flags; + if ((flags & test) == test) { + flags &= ~NETDATA_FILESYSTEM_FLAG_CHART_CREATED; + + ebpf_write_chart_obsolete(NETDATA_FILESYSTEM_FAMILY, efp->hread.name, + efp->hread.title, + EBPF_COMMON_DIMENSION_CALL, efp->family_name, + NULL, NETDATA_EBPF_CHART_TYPE_STACKED, efp->hread.order, update_every); + + ebpf_write_chart_obsolete(NETDATA_FILESYSTEM_FAMILY, efp->hwrite.name, + efp->hwrite.title, + EBPF_COMMON_DIMENSION_CALL, efp->family_name, + NULL, NETDATA_EBPF_CHART_TYPE_STACKED, efp->hwrite.order, update_every); + + ebpf_write_chart_obsolete(NETDATA_FILESYSTEM_FAMILY, efp->hopen.name, efp->hopen.title, + EBPF_COMMON_DIMENSION_CALL, efp->family_name, + NULL, NETDATA_EBPF_CHART_TYPE_STACKED, efp->hopen.order, update_every); + + ebpf_write_chart_obsolete(NETDATA_FILESYSTEM_FAMILY, efp->hadditional.name, efp->hadditional.title, + EBPF_COMMON_DIMENSION_CALL, efp->family_name, + NULL, NETDATA_EBPF_CHART_TYPE_STACKED, efp->hadditional.order, + update_every); + } + efp->flags = flags; + } +} + +/** + * Create Filesystem chart + * + * Create latency charts + * + * @param update_every value to overwrite the update frequency set by the server. + */ +static void ebpf_create_fs_charts(int update_every) +{ + static int order = NETDATA_CHART_PRIO_EBPF_FILESYSTEM_CHARTS; + char chart_name[64], title[256], family[64]; + int i; + uint32_t test = NETDATA_FILESYSTEM_FLAG_CHART_CREATED|NETDATA_FILESYSTEM_REMOVE_CHARTS; + for (i = 0; localfs[i].filesystem; i++) { + ebpf_filesystem_partitions_t *efp = &localfs[i]; + uint32_t flags = efp->flags; + if (flags & NETDATA_FILESYSTEM_FLAG_HAS_PARTITION && !(flags & test)) { + snprintfz(title, 255, "%s latency for each read request.", efp->filesystem); + snprintfz(family, 63, "%s_latency", efp->family); + snprintfz(chart_name, 63, "%s_read_latency", efp->filesystem); + efp->hread.name = strdupz(chart_name); + efp->hread.title = strdupz(title); + efp->hread.order = order; + efp->family_name = strdupz(family); + + ebpf_create_chart(NETDATA_FILESYSTEM_FAMILY, efp->hread.name, + title, + EBPF_COMMON_DIMENSION_CALL, family, + NULL, NETDATA_EBPF_CHART_TYPE_STACKED, order, ebpf_create_global_dimension, + filesystem_publish_aggregated, NETDATA_EBPF_HIST_MAX_BINS, + update_every, NETDATA_EBPF_MODULE_NAME_FILESYSTEM); + order++; + + snprintfz(title, 255, "%s latency for each write request.", efp->filesystem); + snprintfz(chart_name, 63, "%s_write_latency", efp->filesystem); + efp->hwrite.name = strdupz(chart_name); + efp->hwrite.title = strdupz(title); + efp->hwrite.order = order; + ebpf_create_chart(NETDATA_FILESYSTEM_FAMILY, efp->hwrite.name, + title, + EBPF_COMMON_DIMENSION_CALL, family, + NULL, NETDATA_EBPF_CHART_TYPE_STACKED, order, ebpf_create_global_dimension, + filesystem_publish_aggregated, NETDATA_EBPF_HIST_MAX_BINS, + update_every, NETDATA_EBPF_MODULE_NAME_FILESYSTEM); + order++; + + snprintfz(title, 255, "%s latency for each open request.", efp->filesystem); + snprintfz(chart_name, 63, "%s_open_latency", efp->filesystem); + efp->hopen.name = strdupz(chart_name); + efp->hopen.title = strdupz(title); + efp->hopen.order = order; + ebpf_create_chart(NETDATA_FILESYSTEM_FAMILY, efp->hopen.name, + title, + EBPF_COMMON_DIMENSION_CALL, family, + NULL, NETDATA_EBPF_CHART_TYPE_STACKED, order, ebpf_create_global_dimension, + filesystem_publish_aggregated, NETDATA_EBPF_HIST_MAX_BINS, + update_every, NETDATA_EBPF_MODULE_NAME_FILESYSTEM); + order++; + + char *type = (efp->flags & NETDATA_FILESYSTEM_ATTR_CHARTS) ? "attribute" : "sync"; + snprintfz(title, 255, "%s latency for each %s request.", efp->filesystem, type); + snprintfz(chart_name, 63, "%s_%s_latency", efp->filesystem, type); + efp->hadditional.name = strdupz(chart_name); + efp->hadditional.title = strdupz(title); + efp->hadditional.order = order; + ebpf_create_chart(NETDATA_FILESYSTEM_FAMILY, efp->hadditional.name, title, + EBPF_COMMON_DIMENSION_CALL, family, + NULL, NETDATA_EBPF_CHART_TYPE_STACKED, order, ebpf_create_global_dimension, + filesystem_publish_aggregated, NETDATA_EBPF_HIST_MAX_BINS, + update_every, NETDATA_EBPF_MODULE_NAME_FILESYSTEM); + order++; + efp->flags |= NETDATA_FILESYSTEM_FLAG_CHART_CREATED; + } + } +} + +/** + * Initialize eBPF data + * + * @param em main thread structure. + * + * @return it returns 0 on success and -1 otherwise. + */ +int ebpf_filesystem_initialize_ebpf_data(ebpf_module_t *em) +{ + int i; + const char *saved_name = em->thread_name; + uint64_t kernels = em->kernels; + for (i = 0; localfs[i].filesystem; i++) { + ebpf_filesystem_partitions_t *efp = &localfs[i]; + if (!efp->probe_links && efp->flags & NETDATA_FILESYSTEM_LOAD_EBPF_PROGRAM) { + em->thread_name = efp->filesystem; + em->kernels = efp->kernels; + efp->probe_links = ebpf_load_program(ebpf_plugin_dir, em, running_on_kernel, isrh, &efp->objects); + if (!efp->probe_links) { + em->thread_name = saved_name; + em->kernels = kernels; + return -1; + } + efp->flags |= NETDATA_FILESYSTEM_FLAG_HAS_PARTITION; + + // Nedeed for filesystems like btrfs + if ((efp->flags & NETDATA_FILESYSTEM_FILL_ADDRESS_TABLE) && (efp->addresses.function)) { + ebpf_load_addresses(&efp->addresses, fs_maps[i + 1].map_fd); + } + } + efp->flags &= ~NETDATA_FILESYSTEM_LOAD_EBPF_PROGRAM; + } + em->thread_name = saved_name; + em->kernels = kernels; + + if (!dimensions) { + dimensions = ebpf_fill_histogram_dimension(NETDATA_EBPF_HIST_MAX_BINS); + + memset(filesystem_aggregated_data, 0 , NETDATA_EBPF_HIST_MAX_BINS * sizeof(netdata_syscall_stat_t)); + memset(filesystem_publish_aggregated, 0 , NETDATA_EBPF_HIST_MAX_BINS * sizeof(netdata_publish_syscall_t)); + + filesystem_hash_values = callocz(ebpf_nprocs, sizeof(netdata_idx_t)); + } + + return 0; +} + +/** + * Read Local partitions + * + * @return the total of partitions that will be monitored + */ +static int ebpf_read_local_partitions() +{ + char filename[FILENAME_MAX + 1]; + snprintfz(filename, FILENAME_MAX, "%s/proc/self/mountinfo", netdata_configured_host_prefix); + procfile *ff = procfile_open(filename, " \t", PROCFILE_FLAG_DEFAULT); + if(unlikely(!ff)) { + snprintfz(filename, FILENAME_MAX, "%s/proc/1/mountinfo", netdata_configured_host_prefix); + ff = procfile_open(filename, " \t", PROCFILE_FLAG_DEFAULT); + if(unlikely(!ff)) return 0; + } + + ff = procfile_readall(ff); + if(unlikely(!ff)) + return 0; + + int count = 0; + unsigned long l, i, lines = procfile_lines(ff); + for (i = 0; localfs[i].filesystem; i++) { + localfs[i].flags |= NETDATA_FILESYSTEM_REMOVE_CHARTS; + } + + for(l = 0; l < lines ; l++) { + // In "normal" situation the expected value is at column 7 + // When `shared` options is added to mount information, the filesystem is at column 8 + // Finally when we have systemd starting netdata, it will be at column 9 + unsigned long index = procfile_linewords(ff, l) - 3; + + char *fs = procfile_lineword(ff, l, index); + + for (i = 0; localfs[i].filesystem; i++) { + ebpf_filesystem_partitions_t *w = &localfs[i]; + if (w->enabled && (!strcmp(fs, w->filesystem) || + (w->optional_filesystem && !strcmp(fs, w->optional_filesystem)))) { + localfs[i].flags |= NETDATA_FILESYSTEM_LOAD_EBPF_PROGRAM; + localfs[i].flags &= ~NETDATA_FILESYSTEM_REMOVE_CHARTS; + count++; + break; + } + } + } + procfile_close(ff); + + return count; +} + +/** + * Update partition + * + * Update the partition structures before to plot + * + * @param em main thread structure + * + * @return 0 on success and -1 otherwise. + */ +static int ebpf_update_partitions(ebpf_module_t *em) +{ + static time_t update_every = 0; + time_t curr = now_realtime_sec(); + if (curr < update_every) + return 0; + + update_every = curr + 5 * em->update_every; + if (!ebpf_read_local_partitions()) { + em->optional = -1; + return -1; + } + + if (ebpf_filesystem_initialize_ebpf_data(em)) { + return -1; + } + + return 0; +} + +/***************************************************************** + * + * CLEANUP FUNCTIONS + * + *****************************************************************/ + +/* + * Cleanup eBPF data + */ +void ebpf_filesystem_cleanup_ebpf_data() +{ + int i; + for (i = 0; localfs[i].filesystem; i++) { + ebpf_filesystem_partitions_t *efp = &localfs[i]; + if (efp->probe_links) { + freez(efp->family_name); + + freez(efp->hread.name); + freez(efp->hread.title); + + freez(efp->hwrite.name); + freez(efp->hwrite.title); + + freez(efp->hopen.name); + freez(efp->hopen.title); + + freez(efp->hadditional.name); + freez(efp->hadditional.title); + } + } +} + +/** + * Filesystem Free + * + * Cleanup variables after child threads to stop + * + * @param ptr thread data. + */ +static void ebpf_filesystem_free(ebpf_module_t *em) +{ + pthread_mutex_lock(&ebpf_exit_cleanup); + if (em->thread->enabled == NETDATA_THREAD_EBPF_RUNNING) { + em->thread->enabled = NETDATA_THREAD_EBPF_STOPPING; + pthread_mutex_unlock(&ebpf_exit_cleanup); + return; + } + pthread_mutex_unlock(&ebpf_exit_cleanup); + + freez(filesystem_threads.thread); + ebpf_cleanup_publish_syscall(filesystem_publish_aggregated); + + ebpf_filesystem_cleanup_ebpf_data(); + if (dimensions) + ebpf_histogram_dimension_cleanup(dimensions, NETDATA_EBPF_HIST_MAX_BINS); + freez(filesystem_hash_values); + + pthread_mutex_lock(&ebpf_exit_cleanup); + em->thread->enabled = NETDATA_THREAD_EBPF_STOPPED; + pthread_mutex_unlock(&ebpf_exit_cleanup); +} + +/** + * Filesystem exit + * + * Cancel child thread. + * + * @param ptr thread data. + */ +static void ebpf_filesystem_exit(void *ptr) +{ + ebpf_module_t *em = (ebpf_module_t *)ptr; + netdata_thread_cancel(*filesystem_threads.thread); + ebpf_filesystem_free(em); +} + +/** + * File system cleanup + * + * Clean up allocated thread. + * + * @param ptr thread data. + */ +static void ebpf_filesystem_cleanup(void *ptr) +{ + ebpf_module_t *em = (ebpf_module_t *)ptr; + ebpf_filesystem_free(em); +} + +/***************************************************************** + * + * MAIN THREAD + * + *****************************************************************/ + +/** + * Select hist + * + * Select a histogram to store data. + * + * @param efp pointer for the structure with pointers. + * @param id histogram selector + * + * @return It returns a pointer for the histogram + */ +static inline netdata_ebpf_histogram_t *select_hist(ebpf_filesystem_partitions_t *efp, uint32_t *idx, uint32_t id) +{ + if (id < NETDATA_KEY_CALLS_READ) { + *idx = id; + return &efp->hread; + } else if (id < NETDATA_KEY_CALLS_WRITE) { + *idx = id - NETDATA_KEY_CALLS_READ; + return &efp->hwrite; + } else if (id < NETDATA_KEY_CALLS_OPEN) { + *idx = id - NETDATA_KEY_CALLS_WRITE; + return &efp->hopen; + } else if (id < NETDATA_KEY_CALLS_SYNC ){ + *idx = id - NETDATA_KEY_CALLS_OPEN; + return &efp->hadditional; + } + + return NULL; +} + +/** + * Read hard disk table + * + * @param table index for the hash table + * + * Read the table with number of calls for all functions + */ +static void read_filesystem_table(ebpf_filesystem_partitions_t *efp, int fd) +{ + netdata_idx_t *values = filesystem_hash_values; + uint32_t key; + uint32_t idx; + for (key = 0; key < NETDATA_KEY_CALLS_SYNC; key++) { + netdata_ebpf_histogram_t *w = select_hist(efp, &idx, key); + if (!w) { + continue; + } + + int test = bpf_map_lookup_elem(fd, &key, values); + if (test < 0) { + continue; + } + + uint64_t total = 0; + int i; + int end = ebpf_nprocs; + for (i = 0; i < end; i++) { + total += values[i]; + } + + if (idx >= NETDATA_EBPF_HIST_MAX_BINS) + idx = NETDATA_EBPF_HIST_MAX_BINS - 1; + w->histogram[idx] = total; + } +} + +/** + * Read hard disk table + * + * @param table index for the hash table + * + * Read the table with number of calls for all functions + */ +static void read_filesystem_tables() +{ + int i; + for (i = 0; localfs[i].filesystem; i++) { + ebpf_filesystem_partitions_t *efp = &localfs[i]; + if (efp->flags & NETDATA_FILESYSTEM_FLAG_HAS_PARTITION) { + read_filesystem_table(efp, fs_maps[i].map_fd); + } + } +} + +/** + * Socket read hash + * + * This is the thread callback. + * This thread is necessary, because we cannot freeze the whole plugin to read the data on very busy socket. + * + * @param ptr It is a NULL value for this thread. + * + * @return It always returns NULL. + */ +void *ebpf_filesystem_read_hash(void *ptr) +{ + netdata_thread_cleanup_push(ebpf_filesystem_cleanup, ptr); + ebpf_module_t *em = (ebpf_module_t *)ptr; + + heartbeat_t hb; + heartbeat_init(&hb); + usec_t step = NETDATA_FILESYSTEM_READ_SLEEP_MS * em->update_every; + int update_every = em->update_every; + while (!ebpf_exit_plugin) { + (void)heartbeat_next(&hb, step); + + (void) ebpf_update_partitions(em); + ebpf_obsolete_fs_charts(update_every); + + // No more partitions, it is not necessary to read tables + if (em->optional) + continue; + + read_filesystem_tables(); + } + + netdata_thread_cleanup_pop(1); + return NULL; +} + +/** + * Send Hard disk data + * + * Send hard disk information to Netdata. + */ +static void ebpf_histogram_send_data() +{ + uint32_t i; + uint32_t test = NETDATA_FILESYSTEM_FLAG_HAS_PARTITION | NETDATA_FILESYSTEM_REMOVE_CHARTS; + for (i = 0; localfs[i].filesystem; i++) { + ebpf_filesystem_partitions_t *efp = &localfs[i]; + if ((efp->flags & test) == NETDATA_FILESYSTEM_FLAG_HAS_PARTITION) { + write_histogram_chart(NETDATA_FILESYSTEM_FAMILY, efp->hread.name, + efp->hread.histogram, dimensions, NETDATA_EBPF_HIST_MAX_BINS); + + write_histogram_chart(NETDATA_FILESYSTEM_FAMILY, efp->hwrite.name, + efp->hwrite.histogram, dimensions, NETDATA_EBPF_HIST_MAX_BINS); + + write_histogram_chart(NETDATA_FILESYSTEM_FAMILY, efp->hopen.name, + efp->hopen.histogram, dimensions, NETDATA_EBPF_HIST_MAX_BINS); + + write_histogram_chart(NETDATA_FILESYSTEM_FAMILY, efp->hadditional.name, + efp->hadditional.histogram, dimensions, NETDATA_EBPF_HIST_MAX_BINS); + } + } +} + +/** + * Main loop for this collector. + * + * @param em main structure for this thread + */ +static void filesystem_collector(ebpf_module_t *em) +{ + filesystem_threads.thread = mallocz(sizeof(netdata_thread_t)); + filesystem_threads.start_routine = ebpf_filesystem_read_hash; + + netdata_thread_create(filesystem_threads.thread, filesystem_threads.name, + NETDATA_THREAD_OPTION_DEFAULT, ebpf_filesystem_read_hash, em); + + int update_every = em->update_every; + heartbeat_t hb; + heartbeat_init(&hb); + usec_t step = update_every * USEC_PER_SEC; + while (!ebpf_exit_plugin) { + (void)heartbeat_next(&hb, step); + if (ebpf_exit_plugin) + break; + + pthread_mutex_lock(&lock); + + ebpf_create_fs_charts(update_every); + ebpf_histogram_send_data(); + + pthread_mutex_unlock(&lock); + } +} + +/***************************************************************** + * + * ENTRY THREAD + * + *****************************************************************/ + +/** + * Update Filesystem + * + * Update file system structure using values read from configuration file. + */ +static void ebpf_update_filesystem() +{ + char dist[NETDATA_FS_MAX_DIST_NAME + 1]; + int i; + for (i = 0; localfs[i].filesystem; i++) { + snprintfz(dist, NETDATA_FS_MAX_DIST_NAME, "%sdist", localfs[i].filesystem); + + localfs[i].enabled = appconfig_get_boolean(&fs_config, NETDATA_FILESYSTEM_CONFIG_NAME, dist, + CONFIG_BOOLEAN_YES); + } +} + +/** + * Filesystem thread + * + * Thread used to generate socket charts. + * + * @param ptr a pointer to `struct ebpf_module` + * + * @return It always return NULL + */ +void *ebpf_filesystem_thread(void *ptr) +{ + netdata_thread_cleanup_push(ebpf_filesystem_exit, ptr); + + ebpf_module_t *em = (ebpf_module_t *)ptr; + em->maps = fs_maps; + ebpf_update_filesystem(); + + // Initialize optional as zero, to identify when there are not partitions to monitor + em->optional = 0; + + if (ebpf_update_partitions(em)) { + if (em->optional) + info("Netdata cannot monitor the filesystems used on this host."); + + em->thread->enabled = NETDATA_THREAD_EBPF_STOPPED; + goto endfilesystem; + } + + int algorithms[NETDATA_EBPF_HIST_MAX_BINS]; + ebpf_fill_algorithms(algorithms, NETDATA_EBPF_HIST_MAX_BINS, NETDATA_EBPF_INCREMENTAL_IDX); + ebpf_global_labels(filesystem_aggregated_data, filesystem_publish_aggregated, dimensions, dimensions, + algorithms, NETDATA_EBPF_HIST_MAX_BINS); + + pthread_mutex_lock(&lock); + ebpf_create_fs_charts(em->update_every); + ebpf_update_stats(&plugin_statistics, em); + pthread_mutex_unlock(&lock); + + filesystem_collector(em); + +endfilesystem: + ebpf_update_disabled_plugin_stats(em); + + netdata_thread_cleanup_pop(1); + return NULL; +} diff --git a/collectors/ebpf.plugin/ebpf_filesystem.h b/collectors/ebpf.plugin/ebpf_filesystem.h new file mode 100644 index 0000000..0d558df --- /dev/null +++ b/collectors/ebpf.plugin/ebpf_filesystem.h @@ -0,0 +1,49 @@ +// SPDX-License-Identifier: GPL-3.0-or-later + +#ifndef NETDATA_EBPF_FILESYSTEM_H +#define NETDATA_EBPF_FILESYSTEM_H 1 + +// Module name +#define NETDATA_EBPF_MODULE_NAME_FILESYSTEM "filesystem" + +#include "ebpf.h" + +#define NETDATA_FS_MAX_DIST_NAME 64UL + +#define NETDATA_FILESYSTEM_CONFIG_NAME "filesystem" +#define NETDATA_FILESYSTEM_READ_SLEEP_MS 600000ULL + +// Process configuration name +#define NETDATA_FILESYSTEM_CONFIG_FILE "filesystem.conf" + +typedef struct netdata_fs_hist { + uint32_t hist_id; + uint32_t bin; +} netdata_fs_hist_t; + +enum filesystem_limit { + NETDATA_KEY_CALLS_READ = 24, + NETDATA_KEY_CALLS_WRITE = 48, + NETDATA_KEY_CALLS_OPEN = 72, + NETDATA_KEY_CALLS_SYNC = 96 +}; + +enum netdata_filesystem_flags { + NETDATA_FILESYSTEM_FLAG_NO_PARTITION = 0, + NETDATA_FILESYSTEM_LOAD_EBPF_PROGRAM = 1, + NETDATA_FILESYSTEM_FLAG_HAS_PARTITION = 2, + NETDATA_FILESYSTEM_FLAG_CHART_CREATED = 4, + NETDATA_FILESYSTEM_FILL_ADDRESS_TABLE = 8, + NETDATA_FILESYSTEM_REMOVE_CHARTS = 16, + NETDATA_FILESYSTEM_ATTR_CHARTS = 32 +}; + +enum netdata_filesystem_table { + NETDATA_MAIN_FS_TABLE, + NETDATA_ADDR_FS_TABLE +}; + +void *ebpf_filesystem_thread(void *ptr); +extern struct config fs_config; + +#endif /* NETDATA_EBPF_FILESYSTEM_H */ diff --git a/collectors/ebpf.plugin/ebpf_hardirq.c b/collectors/ebpf.plugin/ebpf_hardirq.c new file mode 100644 index 0000000..b07dd24 --- /dev/null +++ b/collectors/ebpf.plugin/ebpf_hardirq.c @@ -0,0 +1,507 @@ +// SPDX-License-Identifier: GPL-3.0-or-later + +#include "ebpf.h" +#include "ebpf_hardirq.h" + +struct config hardirq_config = { .first_section = NULL, + .last_section = NULL, + .mutex = NETDATA_MUTEX_INITIALIZER, + .index = { .avl_tree = { .root = NULL, .compar = appconfig_section_compare }, + .rwlock = AVL_LOCK_INITIALIZER } }; + +#define HARDIRQ_MAP_LATENCY 0 +#define HARDIRQ_MAP_LATENCY_STATIC 1 +static ebpf_local_maps_t hardirq_maps[] = { + { + .name = "tbl_hardirq", + .internal_input = NETDATA_HARDIRQ_MAX_IRQS, + .user_input = 0, + .type = NETDATA_EBPF_MAP_STATIC, + .map_fd = ND_EBPF_MAP_FD_NOT_INITIALIZED + }, + { + .name = "tbl_hardirq_static", + .internal_input = HARDIRQ_EBPF_STATIC_END, + .user_input = 0, + .type = NETDATA_EBPF_MAP_STATIC, + .map_fd = ND_EBPF_MAP_FD_NOT_INITIALIZED + }, + /* end */ + { + .name = NULL, + .internal_input = 0, + .user_input = 0, + .type = NETDATA_EBPF_MAP_CONTROLLER, + .map_fd = ND_EBPF_MAP_FD_NOT_INITIALIZED + } +}; + +#define HARDIRQ_TP_CLASS_IRQ "irq" +#define HARDIRQ_TP_CLASS_IRQ_VECTORS "irq_vectors" +static ebpf_tracepoint_t hardirq_tracepoints[] = { + {.enabled = false, .class = HARDIRQ_TP_CLASS_IRQ, .event = "irq_handler_entry"}, + {.enabled = false, .class = HARDIRQ_TP_CLASS_IRQ, .event = "irq_handler_exit"}, + {.enabled = false, .class = HARDIRQ_TP_CLASS_IRQ_VECTORS, .event = "thermal_apic_entry"}, + {.enabled = false, .class = HARDIRQ_TP_CLASS_IRQ_VECTORS, .event = "thermal_apic_exit"}, + {.enabled = false, .class = HARDIRQ_TP_CLASS_IRQ_VECTORS, .event = "threshold_apic_entry"}, + {.enabled = false, .class = HARDIRQ_TP_CLASS_IRQ_VECTORS, .event = "threshold_apic_exit"}, + {.enabled = false, .class = HARDIRQ_TP_CLASS_IRQ_VECTORS, .event = "error_apic_entry"}, + {.enabled = false, .class = HARDIRQ_TP_CLASS_IRQ_VECTORS, .event = "error_apic_exit"}, + {.enabled = false, .class = HARDIRQ_TP_CLASS_IRQ_VECTORS, .event = "deferred_error_apic_entry"}, + {.enabled = false, .class = HARDIRQ_TP_CLASS_IRQ_VECTORS, .event = "deferred_error_apic_exit"}, + {.enabled = false, .class = HARDIRQ_TP_CLASS_IRQ_VECTORS, .event = "spurious_apic_entry"}, + {.enabled = false, .class = HARDIRQ_TP_CLASS_IRQ_VECTORS, .event = "spurious_apic_exit"}, + {.enabled = false, .class = HARDIRQ_TP_CLASS_IRQ_VECTORS, .event = "call_function_entry"}, + {.enabled = false, .class = HARDIRQ_TP_CLASS_IRQ_VECTORS, .event = "call_function_exit"}, + {.enabled = false, .class = HARDIRQ_TP_CLASS_IRQ_VECTORS, .event = "call_function_single_entry"}, + {.enabled = false, .class = HARDIRQ_TP_CLASS_IRQ_VECTORS, .event = "call_function_single_exit"}, + {.enabled = false, .class = HARDIRQ_TP_CLASS_IRQ_VECTORS, .event = "reschedule_entry"}, + {.enabled = false, .class = HARDIRQ_TP_CLASS_IRQ_VECTORS, .event = "reschedule_exit"}, + {.enabled = false, .class = HARDIRQ_TP_CLASS_IRQ_VECTORS, .event = "local_timer_entry"}, + {.enabled = false, .class = HARDIRQ_TP_CLASS_IRQ_VECTORS, .event = "local_timer_exit"}, + {.enabled = false, .class = HARDIRQ_TP_CLASS_IRQ_VECTORS, .event = "irq_work_entry"}, + {.enabled = false, .class = HARDIRQ_TP_CLASS_IRQ_VECTORS, .event = "irq_work_exit"}, + {.enabled = false, .class = HARDIRQ_TP_CLASS_IRQ_VECTORS, .event = "x86_platform_ipi_entry"}, + {.enabled = false, .class = HARDIRQ_TP_CLASS_IRQ_VECTORS, .event = "x86_platform_ipi_exit"}, + /* end */ + {.enabled = false, .class = NULL, .event = NULL} +}; + +static hardirq_static_val_t hardirq_static_vals[] = { + { + .idx = HARDIRQ_EBPF_STATIC_APIC_THERMAL, + .name = "apic_thermal", + .latency = 0 + }, + { + .idx = HARDIRQ_EBPF_STATIC_APIC_THRESHOLD, + .name = "apic_threshold", + .latency = 0 + }, + { + .idx = HARDIRQ_EBPF_STATIC_APIC_ERROR, + .name = "apic_error", + .latency = 0 + }, + { + .idx = HARDIRQ_EBPF_STATIC_APIC_DEFERRED_ERROR, + .name = "apic_deferred_error", + .latency = 0 + }, + { + .idx = HARDIRQ_EBPF_STATIC_APIC_SPURIOUS, + .name = "apic_spurious", + .latency = 0 + }, + { + .idx = HARDIRQ_EBPF_STATIC_FUNC_CALL, + .name = "func_call", + .latency = 0 + }, + { + .idx = HARDIRQ_EBPF_STATIC_FUNC_CALL_SINGLE, + .name = "func_call_single", + .latency = 0 + }, + { + .idx = HARDIRQ_EBPF_STATIC_RESCHEDULE, + .name = "reschedule", + .latency = 0 + }, + { + .idx = HARDIRQ_EBPF_STATIC_LOCAL_TIMER, + .name = "local_timer", + .latency = 0 + }, + { + .idx = HARDIRQ_EBPF_STATIC_IRQ_WORK, + .name = "irq_work", + .latency = 0 + }, + { + .idx = HARDIRQ_EBPF_STATIC_X86_PLATFORM_IPI, + .name = "x86_platform_ipi", + .latency = 0 + }, +}; + +// store for "published" data from the reader thread, which the collector +// thread will write to netdata agent. +static avl_tree_lock hardirq_pub; + +// tmp store for dynamic hard IRQ values we get from a per-CPU eBPF map. +static hardirq_ebpf_val_t *hardirq_ebpf_vals = NULL; + +// tmp store for static hard IRQ values we get from a per-CPU eBPF map. +static hardirq_ebpf_static_val_t *hardirq_ebpf_static_vals = NULL; + +static struct netdata_static_thread hardirq_threads = { + .name = "HARDIRQ KERNEL", + .config_section = NULL, + .config_name = NULL, + .env_name = NULL, + .enabled = 1, + .thread = NULL, + .init_routine = NULL, + .start_routine = NULL +}; + +/** + * Hardirq Free + * + * Cleanup variables after child threads to stop + * + * @param ptr thread data. + */ +static void ebpf_hardirq_free(ebpf_module_t *em) +{ + pthread_mutex_lock(&ebpf_exit_cleanup); + if (em->thread->enabled == NETDATA_THREAD_EBPF_RUNNING) { + em->thread->enabled = NETDATA_THREAD_EBPF_STOPPING; + pthread_mutex_unlock(&ebpf_exit_cleanup); + return; + } + pthread_mutex_unlock(&ebpf_exit_cleanup); + + freez(hardirq_threads.thread); + for (int i = 0; hardirq_tracepoints[i].class != NULL; i++) { + ebpf_disable_tracepoint(&hardirq_tracepoints[i]); + } + freez(hardirq_ebpf_vals); + freez(hardirq_ebpf_static_vals); + + em->thread->enabled = NETDATA_THREAD_EBPF_STOPPED; +} + +/** + * Hardirq Exit + * + * Cancel child and exit. + * + * @param ptr thread data. + */ +static void hardirq_exit(void *ptr) +{ + ebpf_module_t *em = (ebpf_module_t *)ptr; + netdata_thread_cancel(*hardirq_threads.thread); + ebpf_hardirq_free(em); +} + +/** + * Hardirq clean up + * + * Clean up allocated memory. + * + * @param ptr thread data. + */ +static void hardirq_cleanup(void *ptr) +{ + ebpf_module_t *em = (ebpf_module_t *)ptr; + ebpf_hardirq_free(em); +} + +/***************************************************************** + * MAIN LOOP + *****************************************************************/ + +/** + * Compare hard IRQ values. + * + * @param a `hardirq_val_t *`. + * @param b `hardirq_val_t *`. + * + * @return 0 if a==b, 1 if a>b, -1 if a<b. +*/ +static int hardirq_val_cmp(void *a, void *b) +{ + hardirq_val_t *ptr1 = a; + hardirq_val_t *ptr2 = b; + + if (ptr1->irq > ptr2->irq) { + return 1; + } + else if (ptr1->irq < ptr2->irq) { + return -1; + } + else { + return 0; + } +} + +static void hardirq_read_latency_map(int mapfd) +{ + hardirq_ebpf_key_t key = {}; + hardirq_ebpf_key_t next_key = {}; + hardirq_val_t search_v = {}; + hardirq_val_t *v = NULL; + + while (bpf_map_get_next_key(mapfd, &key, &next_key) == 0) { + // get val for this key. + int test = bpf_map_lookup_elem(mapfd, &key, hardirq_ebpf_vals); + if (unlikely(test < 0)) { + key = next_key; + continue; + } + + // is this IRQ saved yet? + // + // if not, make a new one, mark it as unsaved for now, and continue; we + // will insert it at the end after all of its values are correctly set, + // so that we can safely publish it to the collector within a single, + // short locked operation. + // + // otherwise simply continue; we will only update the latency, which + // can be republished safely without a lock. + // + // NOTE: lock isn't strictly necessary for this initial search, as only + // this thread does writing, but the AVL is using a read-write lock so + // there is no congestion. + bool v_is_new = false; + search_v.irq = key.irq; + v = (hardirq_val_t *)avl_search_lock(&hardirq_pub, (avl_t *)&search_v); + if (unlikely(v == NULL)) { + // latency/name can only be added reliably at a later time. + // when they're added, only then will we AVL insert. + v = callocz(1, sizeof(hardirq_val_t)); + v->irq = key.irq; + v->dim_exists = false; + + v_is_new = true; + } + + // note two things: + // 1. we must add up latency value for this IRQ across all CPUs. + // 2. the name is unfortunately *not* available on all CPU maps - only + // a single map contains the name, so we must find it. we only need + // to copy it though if the IRQ is new for us. + bool name_saved = false; + uint64_t total_latency = 0; + int i; + int end = (running_on_kernel < NETDATA_KERNEL_V4_15) ? 1 : ebpf_nprocs; + for (i = 0; i < end; i++) { + total_latency += hardirq_ebpf_vals[i].latency/1000; + + // copy name for new IRQs. + if (v_is_new && !name_saved && hardirq_ebpf_vals[i].name[0] != '\0') { + strncpyz( + v->name, + hardirq_ebpf_vals[i].name, + NETDATA_HARDIRQ_NAME_LEN + ); + name_saved = true; + } + } + + // can now safely publish latency for existing IRQs. + v->latency = total_latency; + + // can now safely publish new IRQ. + if (v_is_new) { + avl_t *check = avl_insert_lock(&hardirq_pub, (avl_t *)v); + if (check != (avl_t *)v) { + error("Internal error, cannot insert the AVL tree."); + } + } + + key = next_key; + } +} + +static void hardirq_read_latency_static_map(int mapfd) +{ + uint32_t i; + for (i = 0; i < HARDIRQ_EBPF_STATIC_END; i++) { + uint32_t map_i = hardirq_static_vals[i].idx; + int test = bpf_map_lookup_elem(mapfd, &map_i, hardirq_ebpf_static_vals); + if (unlikely(test < 0)) { + continue; + } + + uint64_t total_latency = 0; + int cpu_i; + int end = (running_on_kernel < NETDATA_KERNEL_V4_15) ? 1 : ebpf_nprocs; + for (cpu_i = 0; cpu_i < end; cpu_i++) { + total_latency += hardirq_ebpf_static_vals[cpu_i].latency/1000; + } + + hardirq_static_vals[i].latency = total_latency; + } +} + +/** + * Read eBPF maps for hard IRQ. + */ +static void *hardirq_reader(void *ptr) +{ + netdata_thread_cleanup_push(hardirq_cleanup, ptr); + heartbeat_t hb; + heartbeat_init(&hb); + + ebpf_module_t *em = (ebpf_module_t *)ptr; + + usec_t step = NETDATA_HARDIRQ_SLEEP_MS * em->update_every; + while (!ebpf_exit_plugin) { + (void)heartbeat_next(&hb, step); + + hardirq_read_latency_map(hardirq_maps[HARDIRQ_MAP_LATENCY].map_fd); + hardirq_read_latency_static_map(hardirq_maps[HARDIRQ_MAP_LATENCY_STATIC].map_fd); + } + + netdata_thread_cleanup_pop(1); + return NULL; +} + +static void hardirq_create_charts(int update_every) +{ + ebpf_create_chart( + NETDATA_EBPF_SYSTEM_GROUP, + "hardirq_latency", + "Hardware IRQ latency", + EBPF_COMMON_DIMENSION_MILLISECONDS, + "interrupts", + NULL, + NETDATA_EBPF_CHART_TYPE_STACKED, + NETDATA_CHART_PRIO_HARDIRQ_LATENCY, + NULL, NULL, 0, update_every, + NETDATA_EBPF_MODULE_NAME_HARDIRQ + ); + + fflush(stdout); +} + +static void hardirq_create_static_dims() +{ + uint32_t i; + for (i = 0; i < HARDIRQ_EBPF_STATIC_END; i++) { + ebpf_write_global_dimension( + hardirq_static_vals[i].name, hardirq_static_vals[i].name, + ebpf_algorithms[NETDATA_EBPF_INCREMENTAL_IDX] + ); + } +} + +// callback for avl tree traversal on `hardirq_pub`. +static int hardirq_write_dims(void *entry, void *data) +{ + UNUSED(data); + + hardirq_val_t *v = entry; + + // IRQs get dynamically added in, so add the dimension if we haven't yet. + if (!v->dim_exists) { + ebpf_write_global_dimension( + v->name, v->name, + ebpf_algorithms[NETDATA_EBPF_INCREMENTAL_IDX] + ); + v->dim_exists = true; + } + + write_chart_dimension(v->name, v->latency); + + return 1; +} + +static inline void hardirq_write_static_dims() +{ + uint32_t i; + for (i = 0; i < HARDIRQ_EBPF_STATIC_END; i++) { + write_chart_dimension( + hardirq_static_vals[i].name, + hardirq_static_vals[i].latency + ); + } +} + +/** +* Main loop for this collector. +*/ +static void hardirq_collector(ebpf_module_t *em) +{ + hardirq_ebpf_vals = callocz( + (running_on_kernel < NETDATA_KERNEL_V4_15) ? 1 : ebpf_nprocs, + sizeof(hardirq_ebpf_val_t) + ); + hardirq_ebpf_static_vals = callocz( + (running_on_kernel < NETDATA_KERNEL_V4_15) ? 1 : ebpf_nprocs, + sizeof(hardirq_ebpf_static_val_t) + ); + + avl_init_lock(&hardirq_pub, hardirq_val_cmp); + + // create reader thread. + hardirq_threads.thread = mallocz(sizeof(netdata_thread_t)); + hardirq_threads.start_routine = hardirq_reader; + netdata_thread_create( + hardirq_threads.thread, + hardirq_threads.name, + NETDATA_THREAD_OPTION_DEFAULT, + hardirq_reader, + em + ); + + // create chart and static dims. + pthread_mutex_lock(&lock); + hardirq_create_charts(em->update_every); + hardirq_create_static_dims(); + ebpf_update_stats(&plugin_statistics, em); + pthread_mutex_unlock(&lock); + + // loop and read from published data until ebpf plugin is closed. + heartbeat_t hb; + heartbeat_init(&hb); + usec_t step = em->update_every * USEC_PER_SEC; + //This will be cancelled by its parent + while (!ebpf_exit_plugin) { + (void)heartbeat_next(&hb, step); + if (ebpf_exit_plugin) + break; + + pthread_mutex_lock(&lock); + + // write dims now for all hitherto discovered IRQs. + write_begin_chart(NETDATA_EBPF_SYSTEM_GROUP, "hardirq_latency"); + avl_traverse_lock(&hardirq_pub, hardirq_write_dims, NULL); + hardirq_write_static_dims(); + write_end_chart(); + + pthread_mutex_unlock(&lock); + } +} + +/***************************************************************** + * EBPF HARDIRQ THREAD + *****************************************************************/ + +/** + * Hard IRQ latency thread. + * + * @param ptr a `ebpf_module_t *`. + * @return always NULL. + */ +void *ebpf_hardirq_thread(void *ptr) +{ + netdata_thread_cleanup_push(hardirq_exit, ptr); + + ebpf_module_t *em = (ebpf_module_t *)ptr; + em->maps = hardirq_maps; + + if (ebpf_enable_tracepoints(hardirq_tracepoints) == 0) { + em->thread->enabled = NETDATA_THREAD_EBPF_STOPPED; + goto endhardirq; + } + + em->probe_links = ebpf_load_program(ebpf_plugin_dir, em, running_on_kernel, isrh, &em->objects); + if (!em->probe_links) { + em->thread->enabled = NETDATA_THREAD_EBPF_STOPPED; + goto endhardirq; + } + + hardirq_collector(em); + +endhardirq: + ebpf_update_disabled_plugin_stats(em); + + netdata_thread_cleanup_pop(1); + + return NULL; +} diff --git a/collectors/ebpf.plugin/ebpf_hardirq.h b/collectors/ebpf.plugin/ebpf_hardirq.h new file mode 100644 index 0000000..381da57 --- /dev/null +++ b/collectors/ebpf.plugin/ebpf_hardirq.h @@ -0,0 +1,73 @@ +// SPDX-License-Identifier: GPL-3.0-or-later + +#ifndef NETDATA_EBPF_HARDIRQ_H +#define NETDATA_EBPF_HARDIRQ_H 1 + +/***************************************************************** + * copied from kernel-collectors repo, with modifications needed + * for inclusion here. + *****************************************************************/ + +#define NETDATA_HARDIRQ_NAME_LEN 32 +#define NETDATA_HARDIRQ_MAX_IRQS 1024L + +typedef struct hardirq_ebpf_key { + int irq; +} hardirq_ebpf_key_t; + +typedef struct hardirq_ebpf_val { + uint64_t latency; + uint64_t ts; + char name[NETDATA_HARDIRQ_NAME_LEN]; +} hardirq_ebpf_val_t; + +enum hardirq_ebpf_static { + HARDIRQ_EBPF_STATIC_APIC_THERMAL, + HARDIRQ_EBPF_STATIC_APIC_THRESHOLD, + HARDIRQ_EBPF_STATIC_APIC_ERROR, + HARDIRQ_EBPF_STATIC_APIC_DEFERRED_ERROR, + HARDIRQ_EBPF_STATIC_APIC_SPURIOUS, + HARDIRQ_EBPF_STATIC_FUNC_CALL, + HARDIRQ_EBPF_STATIC_FUNC_CALL_SINGLE, + HARDIRQ_EBPF_STATIC_RESCHEDULE, + HARDIRQ_EBPF_STATIC_LOCAL_TIMER, + HARDIRQ_EBPF_STATIC_IRQ_WORK, + HARDIRQ_EBPF_STATIC_X86_PLATFORM_IPI, + + HARDIRQ_EBPF_STATIC_END +}; + +typedef struct hardirq_ebpf_static_val { + uint64_t latency; + uint64_t ts; +} hardirq_ebpf_static_val_t; + +/***************************************************************** + * below this is eBPF plugin-specific code. + *****************************************************************/ + +#define NETDATA_EBPF_MODULE_NAME_HARDIRQ "hardirq" +#define NETDATA_HARDIRQ_SLEEP_MS 650000ULL +#define NETDATA_HARDIRQ_CONFIG_FILE "hardirq.conf" + +typedef struct hardirq_val { + // must be at top for simplified AVL tree usage. + // if it's not at the top, we need to use `containerof` for almost all ops. + avl_t avl; + + int irq; + bool dim_exists; // keep this after `int irq` for alignment byte savings. + uint64_t latency; + char name[NETDATA_HARDIRQ_NAME_LEN]; +} hardirq_val_t; + +typedef struct hardirq_static_val { + enum hardirq_ebpf_static idx; + char *name; + uint64_t latency; +} hardirq_static_val_t; + +extern struct config hardirq_config; +void *ebpf_hardirq_thread(void *ptr); + +#endif /* NETDATA_EBPF_HARDIRQ_H */ diff --git a/collectors/ebpf.plugin/ebpf_mdflush.c b/collectors/ebpf.plugin/ebpf_mdflush.c new file mode 100644 index 0000000..dc805da --- /dev/null +++ b/collectors/ebpf.plugin/ebpf_mdflush.c @@ -0,0 +1,334 @@ +// SPDX-License-Identifier: GPL-3.0-or-later + +#include "ebpf.h" +#include "ebpf_mdflush.h" + +struct config mdflush_config = { .first_section = NULL, + .last_section = NULL, + .mutex = NETDATA_MUTEX_INITIALIZER, + .index = { .avl_tree = { .root = NULL, .compar = appconfig_section_compare }, + .rwlock = AVL_LOCK_INITIALIZER } }; + +#define MDFLUSH_MAP_COUNT 0 +static ebpf_local_maps_t mdflush_maps[] = { + { + .name = "tbl_mdflush", + .internal_input = 1024, + .user_input = 0, + .type = NETDATA_EBPF_MAP_STATIC, + .map_fd = ND_EBPF_MAP_FD_NOT_INITIALIZED + }, + /* end */ + { + .name = NULL, + .internal_input = 0, + .user_input = 0, + .type = NETDATA_EBPF_MAP_CONTROLLER, + .map_fd = ND_EBPF_MAP_FD_NOT_INITIALIZED + } +}; + +// store for "published" data from the reader thread, which the collector +// thread will write to netdata agent. +static avl_tree_lock mdflush_pub; + +// tmp store for mdflush values we get from a per-CPU eBPF map. +static mdflush_ebpf_val_t *mdflush_ebpf_vals = NULL; + +static struct netdata_static_thread mdflush_threads = { + .name = "MDFLUSH KERNEL", + .config_section = NULL, + .config_name = NULL, + .env_name = NULL, + .enabled = 1, + .thread = NULL, + .init_routine = NULL, + .start_routine = NULL +}; + +/** + * MDflush Free + * + * Cleanup variables after child threads to stop + * + * @param ptr thread data. + */ +static void ebpf_mdflush_free(ebpf_module_t *em) +{ + pthread_mutex_lock(&ebpf_exit_cleanup); + if (em->thread->enabled == NETDATA_THREAD_EBPF_RUNNING) { + em->thread->enabled = NETDATA_THREAD_EBPF_STOPPING; + pthread_mutex_unlock(&ebpf_exit_cleanup); + return; + } + pthread_mutex_unlock(&ebpf_exit_cleanup); + + freez(mdflush_ebpf_vals); + freez(mdflush_threads.thread); + + em->thread->enabled = NETDATA_THREAD_EBPF_STOPPED; +} + +/** + * MDflush exit + * + * Cancel thread and exit. + * + * @param ptr thread data. + */ +static void mdflush_exit(void *ptr) +{ + ebpf_module_t *em = (ebpf_module_t *)ptr; + ebpf_mdflush_free(em); +} + +/** + * CLeanup + * + * Clean allocated memory. + * + * @param ptr thread data. + */ +static void mdflush_cleanup(void *ptr) +{ + ebpf_module_t *em = (ebpf_module_t *)ptr; + netdata_thread_cancel(*mdflush_threads.thread); + ebpf_mdflush_free(em); +} + +/** + * Compare mdflush values. + * + * @param a `netdata_mdflush_t *`. + * @param b `netdata_mdflush_t *`. + * + * @return 0 if a==b, 1 if a>b, -1 if a<b. +*/ +static int mdflush_val_cmp(void *a, void *b) +{ + netdata_mdflush_t *ptr1 = a; + netdata_mdflush_t *ptr2 = b; + + if (ptr1->unit > ptr2->unit) { + return 1; + } + else if (ptr1->unit < ptr2->unit) { + return -1; + } + else { + return 0; + } +} + +static void mdflush_read_count_map() +{ + int mapfd = mdflush_maps[MDFLUSH_MAP_COUNT].map_fd; + mdflush_ebpf_key_t curr_key = (uint32_t)-1; + mdflush_ebpf_key_t key = (uint32_t)-1; + netdata_mdflush_t search_v; + netdata_mdflush_t *v = NULL; + + while (bpf_map_get_next_key(mapfd, &curr_key, &key) == 0) { + curr_key = key; + + // get val for this key. + int test = bpf_map_lookup_elem(mapfd, &key, mdflush_ebpf_vals); + if (unlikely(test < 0)) { + continue; + } + + // is this record saved yet? + // + // if not, make a new one, mark it as unsaved for now, and continue; we + // will insert it at the end after all of its values are correctly set, + // so that we can safely publish it to the collector within a single, + // short locked operation. + // + // otherwise simply continue; we will only update the flush count, + // which can be republished safely without a lock. + // + // NOTE: lock isn't strictly necessary for this initial search, as only + // this thread does writing, but the AVL is using a read-write lock so + // there is no congestion. + bool v_is_new = false; + search_v.unit = key; + v = (netdata_mdflush_t *)avl_search_lock( + &mdflush_pub, + (avl_t *)&search_v + ); + if (unlikely(v == NULL)) { + // flush count can only be added reliably at a later time. + // when they're added, only then will we AVL insert. + v = callocz(1, sizeof(netdata_mdflush_t)); + v->unit = key; + sprintf(v->disk_name, "md%u", key); + v->dim_exists = false; + + v_is_new = true; + } + + // we must add up count value for this record across all CPUs. + uint64_t total_cnt = 0; + int i; + int end = (running_on_kernel < NETDATA_KERNEL_V4_15) ? 1 : ebpf_nprocs; + for (i = 0; i < end; i++) { + total_cnt += mdflush_ebpf_vals[i]; + } + + // can now safely publish count for existing records. + v->cnt = total_cnt; + + // can now safely publish new record. + if (v_is_new) { + avl_t *check = avl_insert_lock(&mdflush_pub, (avl_t *)v); + if (check != (avl_t *)v) { + error("Internal error, cannot insert the AVL tree."); + } + } + } +} + +/** + * Read eBPF maps for mdflush. + */ +static void *mdflush_reader(void *ptr) +{ + netdata_thread_cleanup_push(mdflush_cleanup, ptr); + heartbeat_t hb; + heartbeat_init(&hb); + + ebpf_module_t *em = (ebpf_module_t *)ptr; + + usec_t step = NETDATA_MDFLUSH_SLEEP_MS * em->update_every; + while (!ebpf_exit_plugin) { + (void)heartbeat_next(&hb, step); + + mdflush_read_count_map(); + } + + netdata_thread_cleanup_pop(1); + return NULL; +} + +static void mdflush_create_charts(int update_every) +{ + ebpf_create_chart( + "mdstat", + "mdstat_flush", + "MD flushes", + "flushes", + "flush (eBPF)", + "md.flush", + NETDATA_EBPF_CHART_TYPE_STACKED, + NETDATA_CHART_PRIO_MDSTAT_FLUSH, + NULL, NULL, 0, update_every, + NETDATA_EBPF_MODULE_NAME_MDFLUSH + ); + + fflush(stdout); +} + +// callback for avl tree traversal on `mdflush_pub`. +static int mdflush_write_dims(void *entry, void *data) +{ + UNUSED(data); + + netdata_mdflush_t *v = entry; + + // records get dynamically added in, so add the dim if we haven't yet. + if (!v->dim_exists) { + ebpf_write_global_dimension( + v->disk_name, v->disk_name, + ebpf_algorithms[NETDATA_EBPF_INCREMENTAL_IDX] + ); + v->dim_exists = true; + } + + write_chart_dimension(v->disk_name, v->cnt); + + return 1; +} + +/** +* Main loop for this collector. +*/ +static void mdflush_collector(ebpf_module_t *em) +{ + mdflush_ebpf_vals = callocz(ebpf_nprocs, sizeof(mdflush_ebpf_val_t)); + + avl_init_lock(&mdflush_pub, mdflush_val_cmp); + + // create reader thread. + mdflush_threads.thread = mallocz(sizeof(netdata_thread_t)); + mdflush_threads.start_routine = mdflush_reader; + netdata_thread_create( + mdflush_threads.thread, + mdflush_threads.name, + NETDATA_THREAD_OPTION_DEFAULT, + mdflush_reader, + em + ); + + // create chart and static dims. + pthread_mutex_lock(&lock); + mdflush_create_charts(em->update_every); + ebpf_update_stats(&plugin_statistics, em); + pthread_mutex_unlock(&lock); + + // loop and read from published data until ebpf plugin is closed. + heartbeat_t hb; + heartbeat_init(&hb); + usec_t step = em->update_every * USEC_PER_SEC; + while (!ebpf_exit_plugin) { + (void)heartbeat_next(&hb, step); + if (ebpf_exit_plugin) + break; + + // write dims now for all hitherto discovered devices. + write_begin_chart("mdstat", "mdstat_flush"); + avl_traverse_lock(&mdflush_pub, mdflush_write_dims, NULL); + write_end_chart(); + + pthread_mutex_unlock(&lock); + } +} + +/** + * mdflush thread. + * + * @param ptr a `ebpf_module_t *`. + * @return always NULL. + */ +void *ebpf_mdflush_thread(void *ptr) +{ + netdata_thread_cleanup_push(mdflush_exit, ptr); + + ebpf_module_t *em = (ebpf_module_t *)ptr; + em->maps = mdflush_maps; + + char *md_flush_request = ebpf_find_symbol("md_flush_request"); + if (!md_flush_request) { + em->thread->enabled = NETDATA_THREAD_EBPF_STOPPED; + error("Cannot monitor MD devices, because md is not loaded."); + } + freez(md_flush_request); + + if (em->thread->enabled == NETDATA_THREAD_EBPF_STOPPED) { + goto endmdflush; + } + + em->probe_links = ebpf_load_program(ebpf_plugin_dir, em, running_on_kernel, isrh, &em->objects); + if (!em->probe_links) { + em->enabled = NETDATA_THREAD_EBPF_STOPPED; + goto endmdflush; + } + + mdflush_collector(em); + +endmdflush: + ebpf_update_disabled_plugin_stats(em); + + netdata_thread_cleanup_pop(1); + + return NULL; +} diff --git a/collectors/ebpf.plugin/ebpf_mdflush.h b/collectors/ebpf.plugin/ebpf_mdflush.h new file mode 100644 index 0000000..b04eefd --- /dev/null +++ b/collectors/ebpf.plugin/ebpf_mdflush.h @@ -0,0 +1,42 @@ +// SPDX-License-Identifier: GPL-3.0-or-later + +#ifndef NETDATA_EBPF_MDFLUSH_H +#define NETDATA_EBPF_MDFLUSH_H 1 + +// Module name +#define NETDATA_EBPF_MODULE_NAME_MDFLUSH "mdflush" + +#define NETDATA_MDFLUSH_SLEEP_MS 850000ULL + +// charts +#define NETDATA_MDFLUSH_GLOBAL_CHART "mdflush" + +// configuration file +#define NETDATA_DIRECTORY_MDFLUSH_CONFIG_FILE "mdflush.conf" + +// copy of mdflush types from kernel-collectors repo. +typedef uint32_t mdflush_ebpf_key_t; +typedef uint64_t mdflush_ebpf_val_t; + +typedef struct netdata_mdflush { + // must be at top for simplified AVL tree usage. + // if it's not at the top, we need to use `containerof` for almost all ops. + avl_t avl; + + // key & name of device. + // the name is generated by the key, usually as `md<unit>`. + uint32_t unit; + char disk_name[32]; + + // have we defined the dimension for this device yet? + bool dim_exists; + + // incremental flush count value. + uint64_t cnt; +} netdata_mdflush_t; + +void *ebpf_mdflush_thread(void *ptr); + +extern struct config mdflush_config; + +#endif diff --git a/collectors/ebpf.plugin/ebpf_mount.c b/collectors/ebpf.plugin/ebpf_mount.c new file mode 100644 index 0000000..ec1f07a --- /dev/null +++ b/collectors/ebpf.plugin/ebpf_mount.c @@ -0,0 +1,517 @@ +// SPDX-License-Identifier: GPL-3.0-or-later + +#include "ebpf.h" +#include "ebpf_mount.h" + +static ebpf_local_maps_t mount_maps[] = {{.name = "tbl_mount", .internal_input = NETDATA_MOUNT_END, + .user_input = 0, .type = NETDATA_EBPF_MAP_STATIC, + .map_fd = ND_EBPF_MAP_FD_NOT_INITIALIZED}, + {.name = NULL, .internal_input = 0, .user_input = 0, + .type = NETDATA_EBPF_MAP_CONTROLLER, + .map_fd = ND_EBPF_MAP_FD_NOT_INITIALIZED}}; + +static char *mount_dimension_name[NETDATA_EBPF_MOUNT_SYSCALL] = { "mount", "umount" }; +static netdata_syscall_stat_t mount_aggregated_data[NETDATA_EBPF_MOUNT_SYSCALL]; +static netdata_publish_syscall_t mount_publish_aggregated[NETDATA_EBPF_MOUNT_SYSCALL]; + +struct config mount_config = { .first_section = NULL, .last_section = NULL, .mutex = NETDATA_MUTEX_INITIALIZER, + .index = {.avl_tree = { .root = NULL, .compar = appconfig_section_compare }, + .rwlock = AVL_LOCK_INITIALIZER } }; + +static netdata_idx_t *mount_values = NULL; + +static netdata_idx_t mount_hash_values[NETDATA_MOUNT_END]; + +struct netdata_static_thread mount_thread = { + .name = "MOUNT KERNEL", + .config_section = NULL, + .config_name = NULL, + .env_name = NULL, + .enabled = 1, + .thread = NULL, + .init_routine = NULL, + .start_routine = NULL +}; + +netdata_ebpf_targets_t mount_targets[] = { {.name = "mount", .mode = EBPF_LOAD_TRAMPOLINE}, + {.name = "umount", .mode = EBPF_LOAD_TRAMPOLINE}, + {.name = NULL, .mode = EBPF_LOAD_TRAMPOLINE}}; + +#ifdef LIBBPF_MAJOR_VERSION +#include "includes/mount.skel.h" // BTF code + +static struct mount_bpf *bpf_obj = NULL; + +/***************************************************************** + * + * BTF FUNCTIONS + * + *****************************************************************/ + +/* + * Disable probe + * + * Disable all probes to use exclusively another method. + * + * @param obj is the main structure for bpf objects. + */ +static inline void ebpf_mount_disable_probe(struct mount_bpf *obj) +{ + bpf_program__set_autoload(obj->progs.netdata_mount_probe, false); + bpf_program__set_autoload(obj->progs.netdata_umount_probe, false); + + bpf_program__set_autoload(obj->progs.netdata_mount_retprobe, false); + bpf_program__set_autoload(obj->progs.netdata_umount_retprobe, false); +} + +/* + * Disable tracepoint + * + * Disable all tracepoints to use exclusively another method. + * + * @param obj is the main structure for bpf objects. + */ +static inline void ebpf_mount_disable_tracepoint(struct mount_bpf *obj) +{ + bpf_program__set_autoload(obj->progs.netdata_mount_exit, false); + bpf_program__set_autoload(obj->progs.netdata_umount_exit, false); +} + +/* + * Disable trampoline + * + * Disable all trampoline to use exclusively another method. + * + * @param obj is the main structure for bpf objects. + */ +static inline void ebpf_mount_disable_trampoline(struct mount_bpf *obj) +{ + bpf_program__set_autoload(obj->progs.netdata_mount_fentry, false); + bpf_program__set_autoload(obj->progs.netdata_umount_fentry, false); + bpf_program__set_autoload(obj->progs.netdata_mount_fexit, false); + bpf_program__set_autoload(obj->progs.netdata_umount_fexit, false); +} + +/** + * Set trampoline target + * + * Set the targets we will monitor. + * + * @param obj is the main structure for bpf objects. + */ +static inline void netdata_set_trampoline_target(struct mount_bpf *obj) +{ + char syscall[NETDATA_EBPF_MAX_SYSCALL_LENGTH + 1]; + ebpf_select_host_prefix(syscall, NETDATA_EBPF_MAX_SYSCALL_LENGTH, + mount_targets[NETDATA_MOUNT_SYSCALL].name, running_on_kernel); + + bpf_program__set_attach_target(obj->progs.netdata_mount_fentry, 0, + syscall); + + bpf_program__set_attach_target(obj->progs.netdata_mount_fexit, 0, + syscall); + + ebpf_select_host_prefix(syscall, NETDATA_EBPF_MAX_SYSCALL_LENGTH, + mount_targets[NETDATA_UMOUNT_SYSCALL].name, running_on_kernel); + + bpf_program__set_attach_target(obj->progs.netdata_umount_fentry, 0, + syscall); + + bpf_program__set_attach_target(obj->progs.netdata_umount_fexit, 0, + syscall); +} + +/** + * Mount Attach Probe + * + * Attach probes to target + * + * @param obj is the main structure for bpf objects. + * + * @return It returns 0 on success and -1 otherwise. + */ +static int ebpf_mount_attach_probe(struct mount_bpf *obj) +{ + char syscall[NETDATA_EBPF_MAX_SYSCALL_LENGTH + 1]; + + ebpf_select_host_prefix(syscall, NETDATA_EBPF_MAX_SYSCALL_LENGTH, + mount_targets[NETDATA_MOUNT_SYSCALL].name, running_on_kernel); + + obj->links.netdata_mount_probe = bpf_program__attach_kprobe(obj->progs.netdata_mount_probe, + false, syscall); + int ret = (int)libbpf_get_error(obj->links.netdata_mount_probe); + if (ret) + return -1; + + obj->links.netdata_mount_retprobe = bpf_program__attach_kprobe(obj->progs.netdata_mount_retprobe, + true, syscall); + ret = (int)libbpf_get_error(obj->links.netdata_mount_retprobe); + if (ret) + return -1; + + ebpf_select_host_prefix(syscall, NETDATA_EBPF_MAX_SYSCALL_LENGTH, + mount_targets[NETDATA_UMOUNT_SYSCALL].name, running_on_kernel); + + obj->links.netdata_umount_probe = bpf_program__attach_kprobe(obj->progs.netdata_umount_probe, + false, syscall); + ret = (int)libbpf_get_error(obj->links.netdata_umount_probe); + if (ret) + return -1; + + obj->links.netdata_umount_retprobe = bpf_program__attach_kprobe(obj->progs.netdata_umount_retprobe, + true, syscall); + ret = (int)libbpf_get_error(obj->links.netdata_umount_retprobe); + if (ret) + return -1; + + return 0; +} + +/** + * Set hash tables + * + * Set the values for maps according the value given by kernel. + * + * @param obj is the main structure for bpf objects. + */ +static void ebpf_mount_set_hash_tables(struct mount_bpf *obj) +{ + mount_maps[NETDATA_KEY_MOUNT_TABLE].map_fd = bpf_map__fd(obj->maps.tbl_mount); +} + +/** + * Load and attach + * + * Load and attach the eBPF code in kernel. + * + * @param obj is the main structure for bpf objects. + * @param em structure with configuration + * + * @return it returns 0 on succes and -1 otherwise + */ +static inline int ebpf_mount_load_and_attach(struct mount_bpf *obj, ebpf_module_t *em) +{ + netdata_ebpf_targets_t *mt = em->targets; + netdata_ebpf_program_loaded_t test = mt[NETDATA_MOUNT_SYSCALL].mode; + + // We are testing only one, because all will have the same behavior + if (test == EBPF_LOAD_TRAMPOLINE ) { + ebpf_mount_disable_probe(obj); + ebpf_mount_disable_tracepoint(obj); + + netdata_set_trampoline_target(obj); + } else if (test == EBPF_LOAD_PROBE || + test == EBPF_LOAD_RETPROBE ) { + ebpf_mount_disable_tracepoint(obj); + ebpf_mount_disable_trampoline(obj); + } else { + ebpf_mount_disable_probe(obj); + ebpf_mount_disable_trampoline(obj); + } + + int ret = mount_bpf__load(obj); + if (!ret) { + if (test != EBPF_LOAD_PROBE && test != EBPF_LOAD_RETPROBE ) + ret = mount_bpf__attach(obj); + else + ret = ebpf_mount_attach_probe(obj); + + if (!ret) + ebpf_mount_set_hash_tables(obj); + } + + return ret; +} +#endif +/***************************************************************** + * + * FUNCTIONS TO CLOSE THE THREAD + * + *****************************************************************/ + +/** + * Mount Free + * + * Cleanup variables after child threads to stop + * + * @param ptr thread data. + */ +static void ebpf_mount_free(ebpf_module_t *em) +{ + pthread_mutex_lock(&ebpf_exit_cleanup); + if (em->thread->enabled == NETDATA_THREAD_EBPF_RUNNING) { + em->thread->enabled = NETDATA_THREAD_EBPF_STOPPING; + pthread_mutex_unlock(&ebpf_exit_cleanup); + return; + } + pthread_mutex_unlock(&ebpf_exit_cleanup); + + freez(mount_thread.thread); + freez(mount_values); + +#ifdef LIBBPF_MAJOR_VERSION + if (bpf_obj) + mount_bpf__destroy(bpf_obj); +#endif + + pthread_mutex_lock(&ebpf_exit_cleanup); + em->thread->enabled = NETDATA_THREAD_EBPF_STOPPED; + pthread_mutex_unlock(&ebpf_exit_cleanup); +} + +/** + * Mount Exit + * + * Cancel child thread. + * + * @param ptr thread data. + */ +static void ebpf_mount_exit(void *ptr) +{ + ebpf_module_t *em = (ebpf_module_t *)ptr; + netdata_thread_cancel(*mount_thread.thread); + ebpf_mount_free(em); +} + +/** + * Mount cleanup + * + * Clean up allocated memory. + * + * @param ptr thread data. + */ +static void ebpf_mount_cleanup(void *ptr) +{ + ebpf_module_t *em = (ebpf_module_t *)ptr; + ebpf_mount_free(em); +} + +/***************************************************************** + * + * MAIN LOOP + * + *****************************************************************/ + +/** + * Read global table + * + * Read the table with number of calls for all functions + */ +static void read_global_table() +{ + uint32_t idx; + netdata_idx_t *val = mount_hash_values; + netdata_idx_t *stored = mount_values; + int fd = mount_maps[NETDATA_KEY_MOUNT_TABLE].map_fd; + + for (idx = NETDATA_KEY_MOUNT_CALL; idx < NETDATA_MOUNT_END; idx++) { + if (!bpf_map_lookup_elem(fd, &idx, stored)) { + int i; + int end = ebpf_nprocs; + netdata_idx_t total = 0; + for (i = 0; i < end; i++) + total += stored[i]; + + val[idx] = total; + } + } +} + +/** + * Mount read hash + * + * This is the thread callback. + * This thread is necessary, because we cannot freeze the whole plugin to read the data. + * + * @param ptr It is a NULL value for this thread. + * + * @return It always returns NULL. + */ +void *ebpf_mount_read_hash(void *ptr) +{ + netdata_thread_cleanup_push(ebpf_mount_cleanup, ptr); + heartbeat_t hb; + heartbeat_init(&hb); + + ebpf_module_t *em = (ebpf_module_t *)ptr; + + usec_t step = NETDATA_LATENCY_MOUNT_SLEEP_MS * em->update_every; + //This will be cancelled by its parent + while (!ebpf_exit_plugin) { + (void)heartbeat_next(&hb, step); + + read_global_table(); + } + + netdata_thread_cleanup_pop(1); + return NULL; +} + +/** + * Send data to Netdata calling auxiliary functions. +*/ +static void ebpf_mount_send_data() +{ + int i, j; + int end = NETDATA_EBPF_MOUNT_SYSCALL; + for (i = NETDATA_KEY_MOUNT_CALL, j = NETDATA_KEY_MOUNT_ERROR; i < end; i++, j++) { + mount_publish_aggregated[i].ncall = mount_hash_values[i]; + mount_publish_aggregated[i].nerr = mount_hash_values[j]; + } + + write_count_chart(NETDATA_EBPF_MOUNT_CALLS, NETDATA_EBPF_MOUNT_GLOBAL_FAMILY, + mount_publish_aggregated, NETDATA_EBPF_MOUNT_SYSCALL); + + write_err_chart(NETDATA_EBPF_MOUNT_ERRORS, NETDATA_EBPF_MOUNT_GLOBAL_FAMILY, + mount_publish_aggregated, NETDATA_EBPF_MOUNT_SYSCALL); +} + +/** +* Main loop for this collector. +*/ +static void mount_collector(ebpf_module_t *em) +{ + mount_thread.thread = mallocz(sizeof(netdata_thread_t)); + mount_thread.start_routine = ebpf_mount_read_hash; + memset(mount_hash_values, 0, sizeof(mount_hash_values)); + + mount_values = callocz((size_t)ebpf_nprocs, sizeof(netdata_idx_t)); + + netdata_thread_create(mount_thread.thread, mount_thread.name, NETDATA_THREAD_OPTION_DEFAULT, + ebpf_mount_read_hash, em); + + heartbeat_t hb; + heartbeat_init(&hb); + usec_t step = em->update_every * USEC_PER_SEC; + while (!ebpf_exit_plugin) { + (void)heartbeat_next(&hb, step); + if (ebpf_exit_plugin) + break; + + pthread_mutex_lock(&lock); + + ebpf_mount_send_data(); + + pthread_mutex_unlock(&lock); + } +} + +/***************************************************************** + * + * INITIALIZE THREAD + * + *****************************************************************/ + +/** + * Create mount charts + * + * Call ebpf_create_chart to create the charts for the collector. + * + * @param update_every value to overwrite the update frequency set by the server. + */ +static void ebpf_create_mount_charts(int update_every) +{ + ebpf_create_chart(NETDATA_EBPF_MOUNT_GLOBAL_FAMILY, NETDATA_EBPF_MOUNT_CALLS, + "Calls to mount and umount syscalls", + EBPF_COMMON_DIMENSION_CALL, NETDATA_EBPF_MOUNT_FAMILY, + NULL, + NETDATA_EBPF_CHART_TYPE_LINE, + NETDATA_CHART_PRIO_EBPF_MOUNT_CHARTS, + ebpf_create_global_dimension, + mount_publish_aggregated, NETDATA_EBPF_MOUNT_SYSCALL, + update_every, NETDATA_EBPF_MODULE_NAME_MOUNT); + + ebpf_create_chart(NETDATA_EBPF_MOUNT_GLOBAL_FAMILY, NETDATA_EBPF_MOUNT_ERRORS, + "Errors to mount and umount file systems", + EBPF_COMMON_DIMENSION_CALL, NETDATA_EBPF_MOUNT_FAMILY, + NULL, + NETDATA_EBPF_CHART_TYPE_LINE, + NETDATA_CHART_PRIO_EBPF_MOUNT_CHARTS + 1, + ebpf_create_global_dimension, + mount_publish_aggregated, NETDATA_EBPF_MOUNT_SYSCALL, + update_every, NETDATA_EBPF_MODULE_NAME_MOUNT); + + fflush(stdout); +} + +/***************************************************************** + * + * MAIN THREAD + * + *****************************************************************/ + +/* + * Load BPF + * + * Load BPF files. + * + * @param em the structure with configuration + */ +static int ebpf_mount_load_bpf(ebpf_module_t *em) +{ + int ret = 0; + if (em->load & EBPF_LOAD_LEGACY) { + em->probe_links = ebpf_load_program(ebpf_plugin_dir, em, running_on_kernel, isrh, &em->objects); + if (!em->probe_links) { + em->enabled = CONFIG_BOOLEAN_NO; + ret = -1; + } + } +#ifdef LIBBPF_MAJOR_VERSION + else { + bpf_obj = mount_bpf__open(); + if (!bpf_obj) + ret = -1; + else + ret = ebpf_mount_load_and_attach(bpf_obj, em); + } +#endif + + if (ret) + error("%s %s", EBPF_DEFAULT_ERROR_MSG, em->thread_name); + + return ret; +} + +/** + * Mount thread + * + * Thread used to make mount thread + * + * @param ptr a pointer to `struct ebpf_module` + * + * @return It always returns NULL + */ +void *ebpf_mount_thread(void *ptr) +{ + netdata_thread_cleanup_push(ebpf_mount_exit, ptr); + + ebpf_module_t *em = (ebpf_module_t *)ptr; + em->maps = mount_maps; + +#ifdef LIBBPF_MAJOR_VERSION + ebpf_adjust_thread_load(em, default_btf); +#endif + if (ebpf_mount_load_bpf(em)) { + em->thread->enabled = NETDATA_THREAD_EBPF_STOPPED; + goto endmount; + } + + int algorithms[NETDATA_EBPF_MOUNT_SYSCALL] = { NETDATA_EBPF_INCREMENTAL_IDX, NETDATA_EBPF_INCREMENTAL_IDX }; + + ebpf_global_labels(mount_aggregated_data, mount_publish_aggregated, mount_dimension_name, mount_dimension_name, + algorithms, NETDATA_EBPF_MOUNT_SYSCALL); + + pthread_mutex_lock(&lock); + ebpf_create_mount_charts(em->update_every); + ebpf_update_stats(&plugin_statistics, em); + pthread_mutex_unlock(&lock); + + mount_collector(em); + +endmount: + ebpf_update_disabled_plugin_stats(em); + + netdata_thread_cleanup_pop(1); + return NULL; +} diff --git a/collectors/ebpf.plugin/ebpf_mount.h b/collectors/ebpf.plugin/ebpf_mount.h new file mode 100644 index 0000000..5a8d11a --- /dev/null +++ b/collectors/ebpf.plugin/ebpf_mount.h @@ -0,0 +1,44 @@ +// SPDX-License-Identifier: GPL-3.0-or-later + +#ifndef NETDATA_EBPF_MOUNT_H +#define NETDATA_EBPF_MOUNT_H 1 + +// Module name +#define NETDATA_EBPF_MODULE_NAME_MOUNT "mount" + +#define NETDATA_EBPF_MOUNT_SYSCALL 2 + +#define NETDATA_LATENCY_MOUNT_SLEEP_MS 700000ULL + +#define NETDATA_EBPF_MOUNT_CALLS "call" +#define NETDATA_EBPF_MOUNT_ERRORS "error" +#define NETDATA_EBPF_MOUNT_FAMILY "mount (eBPF)" + +// Process configuration name +#define NETDATA_MOUNT_CONFIG_FILE "mount.conf" + +enum mount_counters { + NETDATA_KEY_MOUNT_CALL, + NETDATA_KEY_UMOUNT_CALL, + NETDATA_KEY_MOUNT_ERROR, + NETDATA_KEY_UMOUNT_ERROR, + + NETDATA_MOUNT_END +}; + +enum mount_tables { + NETDATA_KEY_MOUNT_TABLE +}; + +enum netdata_mount_syscalls { + NETDATA_MOUNT_SYSCALL, + NETDATA_UMOUNT_SYSCALL, + + NETDATA_MOUNT_SYSCALLS_END +}; + +extern struct config mount_config; +void *ebpf_mount_thread(void *ptr); +extern netdata_ebpf_targets_t mount_targets[]; + +#endif /* NETDATA_EBPF_MOUNT_H */ diff --git a/collectors/ebpf.plugin/ebpf_oomkill.c b/collectors/ebpf.plugin/ebpf_oomkill.c new file mode 100644 index 0000000..d93e415 --- /dev/null +++ b/collectors/ebpf.plugin/ebpf_oomkill.c @@ -0,0 +1,402 @@ +// SPDX-License-Identifier: GPL-3.0-or-later + +#include "ebpf.h" +#include "ebpf_oomkill.h" + +struct config oomkill_config = { .first_section = NULL, + .last_section = NULL, + .mutex = NETDATA_MUTEX_INITIALIZER, + .index = { .avl_tree = { .root = NULL, .compar = appconfig_section_compare }, + .rwlock = AVL_LOCK_INITIALIZER } }; + +#define OOMKILL_MAP_KILLCNT 0 +static ebpf_local_maps_t oomkill_maps[] = { + { + .name = "tbl_oomkill", + .internal_input = NETDATA_OOMKILL_MAX_ENTRIES, + .user_input = 0, + .type = NETDATA_EBPF_MAP_STATIC, + .map_fd = ND_EBPF_MAP_FD_NOT_INITIALIZED + }, + /* end */ + { + .name = NULL, + .internal_input = 0, + .user_input = 0, + .type = NETDATA_EBPF_MAP_CONTROLLER, + .map_fd = ND_EBPF_MAP_FD_NOT_INITIALIZED + } +}; + +static ebpf_tracepoint_t oomkill_tracepoints[] = { + {.enabled = false, .class = "oom", .event = "mark_victim"}, + /* end */ + {.enabled = false, .class = NULL, .event = NULL} +}; + +static netdata_publish_syscall_t oomkill_publish_aggregated = {.name = "oomkill", .dimension = "oomkill", + .algorithm = "absolute", + .next = NULL}; + +/** + * Clean up the main thread. + * + * @param ptr thread data. + */ +static void oomkill_cleanup(void *ptr) +{ + ebpf_module_t *em = (ebpf_module_t *)ptr; + em->thread->enabled = NETDATA_THREAD_EBPF_STOPPED; +} + +static void oomkill_write_data(int32_t *keys, uint32_t total) +{ + // for each app, see if it was OOM killed. record as 1 if so otherwise 0. + struct target *w; + for (w = apps_groups_root_target; w != NULL; w = w->next) { + if (likely(w->exposed && w->processes)) { + bool was_oomkilled = false; + struct pid_on_target *pids = w->root_pid; + while (pids) { + uint32_t j; + for (j = 0; j < total; j++) { + if (pids->pid == keys[j]) { + was_oomkilled = true; + // set to 0 so we consider it "done". + keys[j] = 0; + goto write_dim; + } + } + pids = pids->next; + } + + write_dim:; + write_chart_dimension(w->name, was_oomkilled); + } + } + + // for any remaining keys for which we couldn't find a group, this could be + // for various reasons, but the primary one is that the PID has not yet + // been picked up by the process thread when parsing the proc filesystem. + // since it's been OOM killed, it will never be parsed in the future, so + // we have no choice but to dump it into `other`. + uint32_t j; + uint32_t rem_count = 0; + for (j = 0; j < total; j++) { + int32_t key = keys[j]; + if (key != 0) { + rem_count += 1; + } + } + if (rem_count > 0) { + write_chart_dimension("other", rem_count); + } +} + +/** + * Create specific OOMkill charts + * + * Create charts for cgroup/application. + * + * @param type the chart type. + * @param update_every value to overwrite the update frequency set by the server. + */ +static void ebpf_create_specific_oomkill_charts(char *type, int update_every) +{ + ebpf_create_chart(type, NETDATA_OOMKILL_CHART, "OOM kills. This chart is provided by eBPF plugin.", + EBPF_COMMON_DIMENSION_KILLS, NETDATA_EBPF_MEMORY_GROUP, + NETDATA_CGROUP_OOMKILLS_CONTEXT, NETDATA_EBPF_CHART_TYPE_LINE, + NETDATA_CHART_PRIO_CGROUPS_CONTAINERS + 5600, + ebpf_create_global_dimension, + &oomkill_publish_aggregated, 1, update_every, NETDATA_EBPF_MODULE_NAME_OOMKILL); +} + +/** + * Create Systemd OOMkill Charts + * + * Create charts when systemd is enabled + * + * @param update_every value to overwrite the update frequency set by the server. + **/ +static void ebpf_create_systemd_oomkill_charts(int update_every) +{ + ebpf_create_charts_on_systemd(NETDATA_OOMKILL_CHART, "OOM kills. This chart is provided by eBPF plugin.", + EBPF_COMMON_DIMENSION_KILLS, NETDATA_EBPF_MEMORY_GROUP, + NETDATA_EBPF_CHART_TYPE_LINE, 20191, + ebpf_algorithms[NETDATA_EBPF_INCREMENTAL_IDX], NULL, + NETDATA_EBPF_MODULE_NAME_OOMKILL, update_every); +} + +/** + * Send Systemd charts + * + * Send collected data to Netdata. + */ +static void ebpf_send_systemd_oomkill_charts() +{ + ebpf_cgroup_target_t *ect; + write_begin_chart(NETDATA_SERVICE_FAMILY, NETDATA_OOMKILL_CHART); + for (ect = ebpf_cgroup_pids; ect ; ect = ect->next) { + if (unlikely(ect->systemd) && unlikely(ect->updated)) { + write_chart_dimension(ect->name, (long long) ect->oomkill); + ect->oomkill = 0; + } + } + write_end_chart(); +} + +/* + * Send Specific OOMkill data + * + * Send data for specific cgroup/apps. + * + * @param type chart type + * @param value value for oomkill + */ +static void ebpf_send_specific_oomkill_data(char *type, int value) +{ + write_begin_chart(type, NETDATA_OOMKILL_CHART); + write_chart_dimension(oomkill_publish_aggregated.name, (long long)value); + write_end_chart(); +} + +/** + * Create specific OOMkill charts + * + * Create charts for cgroup/application. + * + * @param type the chart type. + * @param update_every value to overwrite the update frequency set by the server. + */ +static void ebpf_obsolete_specific_oomkill_charts(char *type, int update_every) +{ + ebpf_write_chart_obsolete(type, NETDATA_OOMKILL_CHART, "OOM kills. This chart is provided by eBPF plugin.", + EBPF_COMMON_DIMENSION_KILLS, NETDATA_EBPF_MEMORY_GROUP, + NETDATA_EBPF_CHART_TYPE_LINE, NETDATA_CGROUP_OOMKILLS_CONTEXT, + NETDATA_CHART_PRIO_CGROUPS_CONTAINERS + 5600, update_every); +} + +/** + * Send data to Netdata calling auxiliary functions. + * + * @param update_every value to overwrite the update frequency set by the server. +*/ +void ebpf_oomkill_send_cgroup_data(int update_every) +{ + if (!ebpf_cgroup_pids) + return; + + pthread_mutex_lock(&mutex_cgroup_shm); + ebpf_cgroup_target_t *ect; + + int has_systemd = shm_ebpf_cgroup.header->systemd_enabled; + if (has_systemd) { + if (send_cgroup_chart) { + ebpf_create_systemd_oomkill_charts(update_every); + } + ebpf_send_systemd_oomkill_charts(); + } + + for (ect = ebpf_cgroup_pids; ect ; ect = ect->next) { + if (ect->systemd) + continue; + + if (!(ect->flags & NETDATA_EBPF_CGROUP_HAS_OOMKILL_CHART) && ect->updated) { + ebpf_create_specific_oomkill_charts(ect->name, update_every); + ect->flags |= NETDATA_EBPF_CGROUP_HAS_OOMKILL_CHART; + } + + if (ect->flags & NETDATA_EBPF_CGROUP_HAS_OOMKILL_CHART && ect->updated) { + ebpf_send_specific_oomkill_data(ect->name, ect->oomkill); + } else { + ebpf_obsolete_specific_oomkill_charts(ect->name, update_every); + ect->flags &= ~NETDATA_EBPF_CGROUP_HAS_OOMKILL_CHART; + } + } + + pthread_mutex_unlock(&mutex_cgroup_shm); +} + +/** + * Read data + * + * Read OOMKILL events from table. + * + * @param keys vector where data will be stored + * + * @return It returns the number of read elements + */ +static uint32_t oomkill_read_data(int32_t *keys) +{ + // the first `i` entries of `keys` will contain the currently active PIDs + // in the eBPF map. + uint32_t i = 0; + + uint32_t curr_key = 0; + uint32_t key = 0; + int mapfd = oomkill_maps[OOMKILL_MAP_KILLCNT].map_fd; + while (bpf_map_get_next_key(mapfd, &curr_key, &key) == 0) { + curr_key = key; + + keys[i] = (int32_t)key; + i += 1; + + // delete this key now that we've recorded its existence. there's no + // race here, as the same PID will only get OOM killed once. + int test = bpf_map_delete_elem(mapfd, &key); + if (unlikely(test < 0)) { + // since there's only 1 thread doing these deletions, it should be + // impossible to get this condition. + error("key unexpectedly not available for deletion."); + } + } + + return i; +} + +/** + * Update cgroup + * + * Update cgroup data based in + * + * @param keys vector with pids that had oomkill event + * @param total number of elements in keys vector. + */ +static void ebpf_update_oomkill_cgroup(int32_t *keys, uint32_t total) +{ + ebpf_cgroup_target_t *ect; + for (ect = ebpf_cgroup_pids; ect; ect = ect->next) { + ect->oomkill = 0; + struct pid_on_target2 *pids; + for (pids = ect->pids; pids; pids = pids->next) { + uint32_t j; + int32_t pid = pids->pid; + for (j = 0; j < total; j++) { + if (pid == keys[j]) { + ect->oomkill = 1; + break; + } + } + } + } +} + +/** +* Main loop for this collector. +*/ +static void oomkill_collector(ebpf_module_t *em) +{ + int cgroups = em->cgroup_charts; + int update_every = em->update_every; + int32_t keys[NETDATA_OOMKILL_MAX_ENTRIES]; + memset(keys, 0, sizeof(keys)); + + // loop and read until ebpf plugin is closed. + heartbeat_t hb; + heartbeat_init(&hb); + usec_t step = update_every * USEC_PER_SEC; + while (!ebpf_exit_plugin) { + (void)heartbeat_next(&hb, step); + if (ebpf_exit_plugin) + break; + + pthread_mutex_lock(&collect_data_mutex); + pthread_mutex_lock(&lock); + + uint32_t count = oomkill_read_data(keys); + if (cgroups && count) + ebpf_update_oomkill_cgroup(keys, count); + + // write everything from the ebpf map. + if (cgroups) + ebpf_oomkill_send_cgroup_data(update_every); + + if (em->apps_charts & NETDATA_EBPF_APPS_FLAG_CHART_CREATED) { + write_begin_chart(NETDATA_APPS_FAMILY, NETDATA_OOMKILL_CHART); + oomkill_write_data(keys, count); + write_end_chart(); + } + + pthread_mutex_unlock(&lock); + pthread_mutex_unlock(&collect_data_mutex); + } +} + +/** + * Create apps charts + * + * Call ebpf_create_chart to create the charts on apps submenu. + * + * @param em a pointer to the structure with the default values. + */ +void ebpf_oomkill_create_apps_charts(struct ebpf_module *em, void *ptr) +{ + struct target *root = ptr; + ebpf_create_charts_on_apps(NETDATA_OOMKILL_CHART, + "OOM kills", + EBPF_COMMON_DIMENSION_KILLS, + "mem", + NETDATA_EBPF_CHART_TYPE_STACKED, + 20020, + ebpf_algorithms[NETDATA_EBPF_ABSOLUTE_IDX], + root, em->update_every, NETDATA_EBPF_MODULE_NAME_OOMKILL); + + em->apps_charts |= NETDATA_EBPF_APPS_FLAG_CHART_CREATED; +} + +/** + * OOM kill tracking thread. + * + * @param ptr a `ebpf_module_t *`. + * @return always NULL. + */ +void *ebpf_oomkill_thread(void *ptr) +{ + netdata_thread_cleanup_push(oomkill_cleanup, ptr); + + ebpf_module_t *em = (ebpf_module_t *)ptr; + em->maps = oomkill_maps; + +#define NETDATA_DEFAULT_OOM_DISABLED_MSG "Disabling OOMKILL thread, because" + if (unlikely(!all_pids || !em->apps_charts)) { + // When we are not running integration with apps, we won't fill necessary variables for this thread to run, so + // we need to disable it. + if (em->thread->enabled) + info("%s apps integration is completely disabled.", NETDATA_DEFAULT_OOM_DISABLED_MSG); + + em->thread->enabled = NETDATA_THREAD_EBPF_STOPPED; + } else if (running_on_kernel < NETDATA_EBPF_KERNEL_4_14) { + if (em->thread->enabled) + info("%s kernel does not have necessary tracepoints.", NETDATA_DEFAULT_OOM_DISABLED_MSG); + + em->thread->enabled = NETDATA_THREAD_EBPF_STOPPED; + } + + if (em->thread->enabled == NETDATA_THREAD_EBPF_STOPPED) { + goto endoomkill; + } + + if (ebpf_enable_tracepoints(oomkill_tracepoints) == 0) { + em->thread->enabled = NETDATA_THREAD_EBPF_STOPPED; + goto endoomkill; + } + + em->probe_links = ebpf_load_program(ebpf_plugin_dir, em, running_on_kernel, isrh, &em->objects); + if (!em->probe_links) { + em->thread->enabled = NETDATA_THREAD_EBPF_STOPPED; + goto endoomkill; + } + + pthread_mutex_lock(&lock); + ebpf_update_stats(&plugin_statistics, em); + pthread_mutex_unlock(&lock); + + oomkill_collector(em); + +endoomkill: + ebpf_update_disabled_plugin_stats(em); + + netdata_thread_cleanup_pop(1); + + return NULL; +} diff --git a/collectors/ebpf.plugin/ebpf_oomkill.h b/collectors/ebpf.plugin/ebpf_oomkill.h new file mode 100644 index 0000000..7860863 --- /dev/null +++ b/collectors/ebpf.plugin/ebpf_oomkill.h @@ -0,0 +1,32 @@ +// SPDX-License-Identifier: GPL-3.0-or-later + +#ifndef NETDATA_EBPF_OOMKILL_H +#define NETDATA_EBPF_OOMKILL_H 1 + +/***************************************************************** + * copied from kernel-collectors repo, with modifications needed + * for inclusion here. + *****************************************************************/ + +#define NETDATA_OOMKILL_MAX_ENTRIES 64 + +typedef uint8_t oomkill_ebpf_val_t; + +/***************************************************************** + * below this is eBPF plugin-specific code. + *****************************************************************/ + +#define NETDATA_EBPF_MODULE_NAME_OOMKILL "oomkill" +#define NETDATA_OOMKILL_SLEEP_MS 650000ULL +#define NETDATA_OOMKILL_CONFIG_FILE "oomkill.conf" + +#define NETDATA_OOMKILL_CHART "oomkills" + +// Contexts +#define NETDATA_CGROUP_OOMKILLS_CONTEXT "cgroup.oomkills" + +extern struct config oomkill_config; +void *ebpf_oomkill_thread(void *ptr); +void ebpf_oomkill_create_apps_charts(struct ebpf_module *em, void *ptr); + +#endif /* NETDATA_EBPF_OOMKILL_H */ diff --git a/collectors/ebpf.plugin/ebpf_process.c b/collectors/ebpf.plugin/ebpf_process.c new file mode 100644 index 0000000..682577d --- /dev/null +++ b/collectors/ebpf.plugin/ebpf_process.c @@ -0,0 +1,1327 @@ +// SPDX-License-Identifier: GPL-3.0-or-later + +#include <sys/resource.h> + +#include "ebpf.h" +#include "ebpf_process.h" + +/***************************************************************** + * + * GLOBAL VARIABLES + * + *****************************************************************/ + +static char *process_dimension_names[NETDATA_KEY_PUBLISH_PROCESS_END] = { "process", "task", "process", "thread" }; +static char *process_id_names[NETDATA_KEY_PUBLISH_PROCESS_END] = { "do_exit", "release_task", "_do_fork", "sys_clone" }; +static char *status[] = { "process", "zombie" }; + +static ebpf_local_maps_t process_maps[] = {{.name = "tbl_pid_stats", .internal_input = ND_EBPF_DEFAULT_PID_SIZE, + .user_input = 0, + .type = NETDATA_EBPF_MAP_RESIZABLE | NETDATA_EBPF_MAP_PID, + .map_fd = ND_EBPF_MAP_FD_NOT_INITIALIZED}, + {.name = "tbl_total_stats", .internal_input = NETDATA_KEY_END_VECTOR, + .user_input = 0, .type = NETDATA_EBPF_MAP_STATIC, + .map_fd = ND_EBPF_MAP_FD_NOT_INITIALIZED}, + {.name = "process_ctrl", .internal_input = NETDATA_CONTROLLER_END, + .user_input = 0, + .type = NETDATA_EBPF_MAP_CONTROLLER, + .map_fd = ND_EBPF_MAP_FD_NOT_INITIALIZED}, + {.name = NULL, .internal_input = 0, .user_input = 0, + .type = NETDATA_EBPF_MAP_CONTROLLER, + .map_fd = ND_EBPF_MAP_FD_NOT_INITIALIZED}}; + +char *tracepoint_sched_type = { "sched" } ; +char *tracepoint_sched_process_exit = { "sched_process_exit" }; +char *tracepoint_sched_process_exec = { "sched_process_exec" }; +char *tracepoint_sched_process_fork = { "sched_process_fork" }; +static int was_sched_process_exit_enabled = 0; +static int was_sched_process_exec_enabled = 0; +static int was_sched_process_fork_enabled = 0; + +static netdata_idx_t *process_hash_values = NULL; +static netdata_syscall_stat_t process_aggregated_data[NETDATA_KEY_PUBLISH_PROCESS_END]; +static netdata_publish_syscall_t process_publish_aggregated[NETDATA_KEY_PUBLISH_PROCESS_END]; + +ebpf_process_stat_t **global_process_stats = NULL; +ebpf_process_publish_apps_t **current_apps_data = NULL; + +int process_enabled = 0; +bool publish_internal_metrics = true; + +struct config process_config = { .first_section = NULL, + .last_section = NULL, + .mutex = NETDATA_MUTEX_INITIALIZER, + .index = { .avl_tree = { .root = NULL, .compar = appconfig_section_compare }, + .rwlock = AVL_LOCK_INITIALIZER } }; + +static char *threads_stat[NETDATA_EBPF_THREAD_STAT_END] = {"total", "running"}; +static char *load_event_stat[NETDATA_EBPF_LOAD_STAT_END] = {"legacy", "co-re"}; + +static struct netdata_static_thread cgroup_thread = { + .name = "EBPF CGROUP", + .config_section = NULL, + .config_name = NULL, + .env_name = NULL, + .enabled = 1, + .thread = NULL, + .init_routine = NULL, + .start_routine = NULL +}; + +/***************************************************************** + * + * PROCESS DATA AND SEND TO NETDATA + * + *****************************************************************/ + +/** + * Update publish structure before to send data to Netdata. + * + * @param publish the first output structure with independent dimensions + * @param pvc the second output structure with correlated dimensions + * @param input the structure with the input data. + */ +static void ebpf_update_global_publish(netdata_publish_syscall_t *publish, netdata_publish_vfs_common_t *pvc, + netdata_syscall_stat_t *input) +{ + netdata_publish_syscall_t *move = publish; + int selector = NETDATA_KEY_PUBLISH_PROCESS_EXIT; + while (move) { + move->ncall = (input->call > move->pcall) ? input->call - move->pcall : move->pcall - input->call; + move->nbyte = (input->bytes > move->pbyte) ? input->bytes - move->pbyte : move->pbyte - input->bytes; + move->nerr = (input->ecall > move->nerr) ? input->ecall - move->perr : move->perr - input->ecall; + + move->pcall = input->call; + move->pbyte = input->bytes; + move->perr = input->ecall; + + input = input->next; + move = move->next; + selector++; + } + + pvc->running = (long)publish[NETDATA_KEY_PUBLISH_PROCESS_FORK].ncall - + (long)publish[NETDATA_KEY_PUBLISH_PROCESS_CLONE].ncall; + publish[NETDATA_KEY_PUBLISH_PROCESS_RELEASE_TASK].ncall = -publish[NETDATA_KEY_PUBLISH_PROCESS_RELEASE_TASK].ncall; + pvc->zombie = (long)publish[NETDATA_KEY_PUBLISH_PROCESS_EXIT].ncall + + (long)publish[NETDATA_KEY_PUBLISH_PROCESS_RELEASE_TASK].ncall; +} + +/** + * Call the necessary functions to create a chart. + * + * @param family the chart family + * @param move the pointer with the values that will be published + */ +static void write_status_chart(char *family, netdata_publish_vfs_common_t *pvc) +{ + write_begin_chart(family, NETDATA_PROCESS_STATUS_NAME); + + write_chart_dimension(status[0], (long long)pvc->running); + write_chart_dimension(status[1], (long long)pvc->zombie); + + write_end_chart(); +} + +/** + * Send data to Netdata calling auxiliary functions. + * + * @param em the structure with thread information + */ +static void ebpf_process_send_data(ebpf_module_t *em) +{ + netdata_publish_vfs_common_t pvc; + ebpf_update_global_publish(process_publish_aggregated, &pvc, process_aggregated_data); + + write_count_chart(NETDATA_EXIT_SYSCALL, NETDATA_EBPF_SYSTEM_GROUP, + &process_publish_aggregated[NETDATA_KEY_PUBLISH_PROCESS_EXIT], 2); + write_count_chart(NETDATA_PROCESS_SYSCALL, NETDATA_EBPF_SYSTEM_GROUP, + &process_publish_aggregated[NETDATA_KEY_PUBLISH_PROCESS_FORK], 2); + + write_status_chart(NETDATA_EBPF_SYSTEM_GROUP, &pvc); + if (em->mode < MODE_ENTRY) { + write_err_chart(NETDATA_PROCESS_ERROR_NAME, NETDATA_EBPF_SYSTEM_GROUP, + &process_publish_aggregated[NETDATA_KEY_PUBLISH_PROCESS_FORK], 2); + } +} + +/** + * Sum values for pid + * + * @param root the structure with all available PIDs + * + * @param offset the address that we are reading + * + * @return it returns the sum of all PIDs + */ +long long ebpf_process_sum_values_for_pids(struct pid_on_target *root, size_t offset) +{ + long long ret = 0; + while (root) { + int32_t pid = root->pid; + ebpf_process_publish_apps_t *w = current_apps_data[pid]; + if (w) { + ret += get_value_from_structure((char *)w, offset); + } + + root = root->next; + } + + return ret; +} + +/** + * Remove process pid + * + * Remove from PID task table when task_release was called. + */ +void ebpf_process_remove_pids() +{ + struct pid_stat *pids = root_of_pids; + int pid_fd = process_maps[NETDATA_PROCESS_PID_TABLE].map_fd; + while (pids) { + uint32_t pid = pids->pid; + ebpf_process_stat_t *w = global_process_stats[pid]; + if (w) { + freez(w); + global_process_stats[pid] = NULL; + bpf_map_delete_elem(pid_fd, &pid); + } + + pids = pids->next; + } +} + +/** + * Send data to Netdata calling auxiliary functions. + * + * @param root the target list. + */ +void ebpf_process_send_apps_data(struct target *root, ebpf_module_t *em) +{ + struct target *w; + collected_number value; + + write_begin_chart(NETDATA_APPS_FAMILY, NETDATA_SYSCALL_APPS_TASK_PROCESS); + for (w = root; w; w = w->next) { + if (unlikely(w->exposed && w->processes)) { + value = ebpf_process_sum_values_for_pids(w->root_pid, offsetof(ebpf_process_publish_apps_t, create_process)); + write_chart_dimension(w->name, value); + } + } + write_end_chart(); + + write_begin_chart(NETDATA_APPS_FAMILY, NETDATA_SYSCALL_APPS_TASK_THREAD); + for (w = root; w; w = w->next) { + if (unlikely(w->exposed && w->processes)) { + value = ebpf_process_sum_values_for_pids(w->root_pid, offsetof(ebpf_process_publish_apps_t, create_thread)); + write_chart_dimension(w->name, value); + } + } + write_end_chart(); + + write_begin_chart(NETDATA_APPS_FAMILY, NETDATA_SYSCALL_APPS_TASK_EXIT); + for (w = root; w; w = w->next) { + if (unlikely(w->exposed && w->processes)) { + value = ebpf_process_sum_values_for_pids(w->root_pid, offsetof(ebpf_process_publish_apps_t, + call_do_exit)); + write_chart_dimension(w->name, value); + } + } + write_end_chart(); + + write_begin_chart(NETDATA_APPS_FAMILY, NETDATA_SYSCALL_APPS_TASK_CLOSE); + for (w = root; w; w = w->next) { + if (unlikely(w->exposed && w->processes)) { + value = ebpf_process_sum_values_for_pids(w->root_pid, offsetof(ebpf_process_publish_apps_t, + call_release_task)); + write_chart_dimension(w->name, value); + } + } + write_end_chart(); + + if (em->mode < MODE_ENTRY) { + write_begin_chart(NETDATA_APPS_FAMILY, NETDATA_SYSCALL_APPS_TASK_ERROR); + for (w = root; w; w = w->next) { + if (unlikely(w->exposed && w->processes)) { + value = ebpf_process_sum_values_for_pids(w->root_pid, offsetof(ebpf_process_publish_apps_t, + task_err)); + write_chart_dimension(w->name, value); + } + } + write_end_chart(); + } + + ebpf_process_remove_pids(); +} + +/***************************************************************** + * + * READ INFORMATION FROM KERNEL RING + * + *****************************************************************/ + +/** + * Read the hash table and store data to allocated vectors. + */ +static void read_hash_global_tables() +{ + uint64_t idx; + netdata_idx_t res[NETDATA_KEY_END_VECTOR]; + + netdata_idx_t *val = process_hash_values; + int fd = process_maps[NETDATA_PROCESS_GLOBAL_TABLE].map_fd; + for (idx = 0; idx < NETDATA_KEY_END_VECTOR; idx++) { + if (!bpf_map_lookup_elem(fd, &idx, val)) { + uint64_t total = 0; + int i; + int end = ebpf_nprocs; + for (i = 0; i < end; i++) + total += val[i]; + + res[idx] = total; + } else { + res[idx] = 0; + } + } + + process_aggregated_data[NETDATA_KEY_PUBLISH_PROCESS_EXIT].call = res[NETDATA_KEY_CALLS_DO_EXIT]; + process_aggregated_data[NETDATA_KEY_PUBLISH_PROCESS_RELEASE_TASK].call = res[NETDATA_KEY_CALLS_RELEASE_TASK]; + process_aggregated_data[NETDATA_KEY_PUBLISH_PROCESS_FORK].call = res[NETDATA_KEY_CALLS_DO_FORK]; + process_aggregated_data[NETDATA_KEY_PUBLISH_PROCESS_CLONE].call = res[NETDATA_KEY_CALLS_SYS_CLONE]; + + process_aggregated_data[NETDATA_KEY_PUBLISH_PROCESS_FORK].ecall = res[NETDATA_KEY_ERROR_DO_FORK]; + process_aggregated_data[NETDATA_KEY_PUBLISH_PROCESS_CLONE].ecall = res[NETDATA_KEY_ERROR_SYS_CLONE]; +} + +/** + * Read the hash table and store data to allocated vectors. + */ +static void ebpf_process_update_apps_data() +{ + struct pid_stat *pids = root_of_pids; + while (pids) { + uint32_t current_pid = pids->pid; + ebpf_process_stat_t *ps = global_process_stats[current_pid]; + if (!ps) { + pids = pids->next; + continue; + } + + ebpf_process_publish_apps_t *cad = current_apps_data[current_pid]; + if (!cad) { + cad = callocz(1, sizeof(ebpf_process_publish_apps_t)); + current_apps_data[current_pid] = cad; + } + + //Read data + cad->call_do_exit = ps->exit_call; + cad->call_release_task = ps->release_call; + cad->create_process = ps->create_process; + cad->create_thread = ps->create_thread; + + cad->task_err = ps->task_err; + + pids = pids->next; + } +} + +/** + * Cgroup Exit + * + * Function used with netdata_thread_clean_push + * + * @param ptr unused argument + */ +static void ebpf_cgroup_exit(void *ptr) +{ + UNUSED(ptr); +} + +/** + * Cgroup update shm + * + * This is the thread callback. + * This thread is necessary, because we cannot freeze the whole plugin to read the data from shared memory. + * + * @param ptr It is a NULL value for this thread. + * + * @return It always returns NULL. + */ +void *ebpf_cgroup_update_shm(void *ptr) +{ + netdata_thread_cleanup_push(ebpf_cgroup_exit, ptr); + heartbeat_t hb; + heartbeat_init(&hb); + + usec_t step = 3 * USEC_PER_SEC; + int counter = NETDATA_EBPF_CGROUP_UPDATE - 1; + //This will be cancelled by its parent + while (!ebpf_exit_plugin) { + (void)heartbeat_next(&hb, step); + + // We are using a small heartbeat time to wake up thread, + // but we should not update so frequently the shared memory data + if (++counter >= NETDATA_EBPF_CGROUP_UPDATE) { + counter = 0; + if (!shm_ebpf_cgroup.header) + ebpf_map_cgroup_shared_memory(); + + ebpf_parse_cgroup_shm_data(); + } + } + + netdata_thread_cleanup_pop(1); + return NULL; +} + +/** + * Update cgroup + * + * Update cgroup data based in + */ +static void ebpf_update_process_cgroup() +{ + ebpf_cgroup_target_t *ect ; + int pid_fd = process_maps[NETDATA_PROCESS_PID_TABLE].map_fd; + + pthread_mutex_lock(&mutex_cgroup_shm); + for (ect = ebpf_cgroup_pids; ect; ect = ect->next) { + struct pid_on_target2 *pids; + for (pids = ect->pids; pids; pids = pids->next) { + int pid = pids->pid; + ebpf_process_stat_t *out = &pids->ps; + if (global_process_stats[pid]) { + ebpf_process_stat_t *in = global_process_stats[pid]; + + memcpy(out, in, sizeof(ebpf_process_stat_t)); + } else { + if (bpf_map_lookup_elem(pid_fd, &pid, out)) { + memset(out, 0, sizeof(ebpf_process_stat_t)); + } + } + } + } + pthread_mutex_unlock(&mutex_cgroup_shm); +} + +/***************************************************************** + * + * FUNCTIONS TO CREATE CHARTS + * + *****************************************************************/ + +/** + * Create process status chart + * + * @param family the chart family + * @param name the chart name + * @param axis the axis label + * @param web the group name used to attach the chart on dashboard + * @param order the order number of the specified chart + * @param update_every value to overwrite the update frequency set by the server. + */ +static void ebpf_process_status_chart(char *family, char *name, char *axis, + char *web, char *algorithm, int order, int update_every) +{ + printf("CHART %s.%s '' 'Process not closed' '%s' '%s' '' line %d %d '' 'ebpf.plugin' 'process'\n", + family, + name, + axis, + web, + order, + update_every); + + printf("DIMENSION %s '' %s 1 1\n", status[0], algorithm); + printf("DIMENSION %s '' %s 1 1\n", status[1], algorithm); +} + +/** + * Create global charts + * + * Call ebpf_create_chart to create the charts for the collector. + * + * @param em a pointer to the structure with the default values. + */ +static void ebpf_create_global_charts(ebpf_module_t *em) +{ + ebpf_create_chart(NETDATA_EBPF_SYSTEM_GROUP, + NETDATA_PROCESS_SYSCALL, + "Start process", + EBPF_COMMON_DIMENSION_CALL, + NETDATA_PROCESS_GROUP, + NULL, + NETDATA_EBPF_CHART_TYPE_LINE, + 21002, + ebpf_create_global_dimension, + &process_publish_aggregated[NETDATA_KEY_PUBLISH_PROCESS_FORK], + 2, em->update_every, NETDATA_EBPF_MODULE_NAME_PROCESS); + + ebpf_create_chart(NETDATA_EBPF_SYSTEM_GROUP, + NETDATA_EXIT_SYSCALL, + "Exit process", + EBPF_COMMON_DIMENSION_CALL, + NETDATA_PROCESS_GROUP, + NULL, + NETDATA_EBPF_CHART_TYPE_LINE, + 21003, + ebpf_create_global_dimension, + &process_publish_aggregated[NETDATA_KEY_PUBLISH_PROCESS_EXIT], + 2, em->update_every, NETDATA_EBPF_MODULE_NAME_PROCESS); + + ebpf_process_status_chart(NETDATA_EBPF_SYSTEM_GROUP, + NETDATA_PROCESS_STATUS_NAME, + EBPF_COMMON_DIMENSION_DIFFERENCE, + NETDATA_PROCESS_GROUP, + ebpf_algorithms[NETDATA_EBPF_ABSOLUTE_IDX], + 21004, em->update_every); + + if (em->mode < MODE_ENTRY) { + ebpf_create_chart(NETDATA_EBPF_SYSTEM_GROUP, + NETDATA_PROCESS_ERROR_NAME, + "Fails to create process", + EBPF_COMMON_DIMENSION_CALL, + NETDATA_PROCESS_GROUP, + NULL, + NETDATA_EBPF_CHART_TYPE_LINE, + 21005, + ebpf_create_global_dimension, + &process_publish_aggregated[NETDATA_KEY_PUBLISH_PROCESS_FORK], + 2, em->update_every, NETDATA_EBPF_MODULE_NAME_PROCESS); + } +} + +/** + * Create chart for Statistic Thread + * + * Write to standard output current values for threads. + * + * @param em a pointer to the structure with the default values. + */ +static inline void ebpf_create_statistic_thread_chart(ebpf_module_t *em) +{ + ebpf_write_chart_cmd(NETDATA_MONITORING_FAMILY, + NETDATA_EBPF_THREADS, + "Threads info.", + "threads", + NETDATA_EBPF_FAMILY, + NETDATA_EBPF_CHART_TYPE_LINE, + NULL, + 140000, + em->update_every, + NETDATA_EBPF_MODULE_NAME_PROCESS); + + ebpf_write_global_dimension(threads_stat[NETDATA_EBPF_THREAD_STAT_TOTAL], + threads_stat[NETDATA_EBPF_THREAD_STAT_TOTAL], + ebpf_algorithms[NETDATA_EBPF_ABSOLUTE_IDX]); + + ebpf_write_global_dimension(threads_stat[NETDATA_EBPF_THREAD_STAT_RUNNING], + threads_stat[NETDATA_EBPF_THREAD_STAT_RUNNING], + ebpf_algorithms[NETDATA_EBPF_ABSOLUTE_IDX]); +} + +/** + * Create chart for Load Thread + * + * Write to standard output current values for load mode. + * + * @param em a pointer to the structure with the default values. + */ +static inline void ebpf_create_statistic_load_chart(ebpf_module_t *em) +{ + ebpf_write_chart_cmd(NETDATA_MONITORING_FAMILY, + NETDATA_EBPF_LOAD_METHOD, + "Load info.", + "methods", + NETDATA_EBPF_FAMILY, + NETDATA_EBPF_CHART_TYPE_LINE, + NULL, + 140001, + em->update_every, + NETDATA_EBPF_MODULE_NAME_PROCESS); + + ebpf_write_global_dimension(load_event_stat[NETDATA_EBPF_LOAD_STAT_LEGACY], + load_event_stat[NETDATA_EBPF_LOAD_STAT_LEGACY], + ebpf_algorithms[NETDATA_EBPF_ABSOLUTE_IDX]); + + ebpf_write_global_dimension(load_event_stat[NETDATA_EBPF_LOAD_STAT_CORE], + load_event_stat[NETDATA_EBPF_LOAD_STAT_CORE], + ebpf_algorithms[NETDATA_EBPF_ABSOLUTE_IDX]); +} + +/** + * Update Internal Metric variable + * + * By default eBPF.plugin sends internal metrics for netdata, but user can + * disable this. + * + * The function updates the variable used to send charts. + */ +static void update_internal_metric_variable() +{ + const char *s = getenv("NETDATA_INTERNALS_MONITORING"); + if (s && *s && strcmp(s, "NO") == 0) + publish_internal_metrics = false; +} + +/** + * Create Statistics Charts + * + * Create charts that will show statistics related to eBPF plugin. + * + * @param em a pointer to the structure with the default values. + */ +static void ebpf_create_statistic_charts(ebpf_module_t *em) +{ + update_internal_metric_variable(); + if (!publish_internal_metrics) + return; + + ebpf_create_statistic_thread_chart(em); + + ebpf_create_statistic_load_chart(em); +} + +/** + * Create process apps charts + * + * Call ebpf_create_chart to create the charts on apps submenu. + * + * @param em a pointer to the structure with the default values. + * @param ptr a pointer for the targets. + */ +void ebpf_process_create_apps_charts(struct ebpf_module *em, void *ptr) +{ + struct target *root = ptr; + ebpf_create_charts_on_apps(NETDATA_SYSCALL_APPS_TASK_PROCESS, + "Process started", + EBPF_COMMON_DIMENSION_CALL, + NETDATA_PROCESS_GROUP, + NETDATA_EBPF_CHART_TYPE_STACKED, + 20065, + ebpf_algorithms[NETDATA_EBPF_ABSOLUTE_IDX], + root, em->update_every, NETDATA_EBPF_MODULE_NAME_PROCESS); + + ebpf_create_charts_on_apps(NETDATA_SYSCALL_APPS_TASK_THREAD, + "Threads started", + EBPF_COMMON_DIMENSION_CALL, + NETDATA_PROCESS_GROUP, + NETDATA_EBPF_CHART_TYPE_STACKED, + 20066, + ebpf_algorithms[NETDATA_EBPF_ABSOLUTE_IDX], + root, em->update_every, NETDATA_EBPF_MODULE_NAME_PROCESS); + + ebpf_create_charts_on_apps(NETDATA_SYSCALL_APPS_TASK_EXIT, + "Tasks starts exit process.", + EBPF_COMMON_DIMENSION_CALL, + NETDATA_PROCESS_GROUP, + NETDATA_EBPF_CHART_TYPE_STACKED, + 20067, + ebpf_algorithms[NETDATA_EBPF_ABSOLUTE_IDX], + root, em->update_every, NETDATA_EBPF_MODULE_NAME_PROCESS); + + ebpf_create_charts_on_apps(NETDATA_SYSCALL_APPS_TASK_CLOSE, + "Tasks closed", + EBPF_COMMON_DIMENSION_CALL, + NETDATA_PROCESS_GROUP, + NETDATA_EBPF_CHART_TYPE_STACKED, + 20068, + ebpf_algorithms[NETDATA_EBPF_ABSOLUTE_IDX], + root, em->update_every, NETDATA_EBPF_MODULE_NAME_PROCESS); + + if (em->mode < MODE_ENTRY) { + ebpf_create_charts_on_apps(NETDATA_SYSCALL_APPS_TASK_ERROR, + "Errors to create process or threads.", + EBPF_COMMON_DIMENSION_CALL, + NETDATA_PROCESS_GROUP, + NETDATA_EBPF_CHART_TYPE_STACKED, + 20069, + ebpf_algorithms[NETDATA_EBPF_ABSOLUTE_IDX], + root, + em->update_every, NETDATA_EBPF_MODULE_NAME_PROCESS); + } + + em->apps_charts |= NETDATA_EBPF_APPS_FLAG_CHART_CREATED; +} + +/** + * Create apps charts + * + * Call ebpf_create_chart to create the charts on apps submenu. + * + * @param root a pointer for the targets. + */ +static void ebpf_create_apps_charts(struct target *root) +{ + if (unlikely(!all_pids)) + return; + + struct target *w; + int newly_added = 0; + + for (w = root; w; w = w->next) { + if (w->target) + continue; + + if (unlikely(w->processes && (debug_enabled || w->debug_enabled))) { + struct pid_on_target *pid_on_target; + + fprintf( + stderr, "ebpf.plugin: target '%s' has aggregated %u process%s:", w->name, w->processes, + (w->processes == 1) ? "" : "es"); + + for (pid_on_target = w->root_pid; pid_on_target; pid_on_target = pid_on_target->next) { + fprintf(stderr, " %d", pid_on_target->pid); + } + + fputc('\n', stderr); + } + + if (!w->exposed && w->processes) { + newly_added++; + w->exposed = 1; + if (debug_enabled || w->debug_enabled) + debug_log_int("%s just added - regenerating charts.", w->name); + } + } + + if (!newly_added) + return; + + int counter; + for (counter = 0; ebpf_modules[counter].thread_name; counter++) { + ebpf_module_t *current = &ebpf_modules[counter]; + if (current->enabled && current->apps_charts && current->apps_routine) + current->apps_routine(current, root); + } +} + +/***************************************************************** + * + * FUNCTIONS TO CLOSE THE THREAD + * + *****************************************************************/ + +/** + * Process disable tracepoints + * + * Disable tracepoints when the plugin was responsible to enable it. + */ +static void ebpf_process_disable_tracepoints() +{ + char *default_message = { "Cannot disable the tracepoint" }; + if (!was_sched_process_exit_enabled) { + if (ebpf_disable_tracing_values(tracepoint_sched_type, tracepoint_sched_process_exit)) + error("%s %s/%s.", default_message, tracepoint_sched_type, tracepoint_sched_process_exit); + } + + if (!was_sched_process_exec_enabled) { + if (ebpf_disable_tracing_values(tracepoint_sched_type, tracepoint_sched_process_exec)) + error("%s %s/%s.", default_message, tracepoint_sched_type, tracepoint_sched_process_exec); + } + + if (!was_sched_process_fork_enabled) { + if (ebpf_disable_tracing_values(tracepoint_sched_type, tracepoint_sched_process_fork)) + error("%s %s/%s.", default_message, tracepoint_sched_type, tracepoint_sched_process_fork); + } +} + +/** + * Process Exit + * + * Cancel child thread. + * + * @param ptr thread data. + */ +static void ebpf_process_exit(void *ptr) +{ + ebpf_module_t *em = (ebpf_module_t *)ptr; + + ebpf_cleanup_publish_syscall(process_publish_aggregated); + freez(process_hash_values); + + ebpf_process_disable_tracepoints(); + + pthread_mutex_lock(&ebpf_exit_cleanup); + em->thread->enabled = NETDATA_THREAD_EBPF_STOPPED; + pthread_mutex_unlock(&ebpf_exit_cleanup); + pthread_cancel(*cgroup_thread.thread); +} + +/***************************************************************** + * + * FUNCTIONS WITH THE MAIN LOOP + * + *****************************************************************/ + + +/** + * Sum PIDs + * + * Sum values for all targets. + * + * @param ps structure used to store data + * @param pids input data + */ +static void ebpf_process_sum_cgroup_pids(ebpf_process_stat_t *ps, struct pid_on_target2 *pids) +{ + ebpf_process_stat_t accumulator; + memset(&accumulator, 0, sizeof(accumulator)); + + while (pids) { + ebpf_process_stat_t *ps = &pids->ps; + + accumulator.exit_call += ps->exit_call; + accumulator.release_call += ps->release_call; + accumulator.create_process += ps->create_process; + accumulator.create_thread += ps->create_thread; + + accumulator.task_err += ps->task_err; + + pids = pids->next; + } + + ps->exit_call = (accumulator.exit_call >= ps->exit_call) ? accumulator.exit_call : ps->exit_call; + ps->release_call = (accumulator.release_call >= ps->release_call) ? accumulator.release_call : ps->release_call; + ps->create_process = (accumulator.create_process >= ps->create_process) ? accumulator.create_process : ps->create_process; + ps->create_thread = (accumulator.create_thread >= ps->create_thread) ? accumulator.create_thread : ps->create_thread; + + ps->task_err = (accumulator.task_err >= ps->task_err) ? accumulator.task_err : ps->task_err; +} + +/* + * Send Specific Process data + * + * Send data for specific cgroup/apps. + * + * @param type chart type + * @param values structure with values that will be sent to netdata + * @param em the structure with thread information + */ +static void ebpf_send_specific_process_data(char *type, ebpf_process_stat_t *values, ebpf_module_t *em) +{ + write_begin_chart(type, NETDATA_SYSCALL_APPS_TASK_PROCESS); + write_chart_dimension(process_publish_aggregated[NETDATA_KEY_PUBLISH_PROCESS_FORK].name, + (long long) values->create_process); + write_end_chart(); + + write_begin_chart(type, NETDATA_SYSCALL_APPS_TASK_THREAD); + write_chart_dimension(process_publish_aggregated[NETDATA_KEY_PUBLISH_PROCESS_CLONE].name, + (long long) values->create_thread); + write_end_chart(); + + write_begin_chart(type, NETDATA_SYSCALL_APPS_TASK_EXIT); + write_chart_dimension(process_publish_aggregated[NETDATA_KEY_PUBLISH_PROCESS_EXIT].name, + (long long) values->release_call); + write_end_chart(); + + write_begin_chart(type, NETDATA_SYSCALL_APPS_TASK_CLOSE); + write_chart_dimension(process_publish_aggregated[NETDATA_KEY_PUBLISH_PROCESS_RELEASE_TASK].name, + (long long) values->release_call); + write_end_chart(); + + if (em->mode < MODE_ENTRY) { + write_begin_chart(type, NETDATA_SYSCALL_APPS_TASK_ERROR); + write_chart_dimension(process_publish_aggregated[NETDATA_KEY_PUBLISH_PROCESS_EXIT].name, + (long long) values->task_err); + write_end_chart(); + } +} + +/** + * Create specific process charts + * + * Create charts for cgroup/application + * + * @param type the chart type. + * @param em the structure with thread information + */ +static void ebpf_create_specific_process_charts(char *type, ebpf_module_t *em) +{ + ebpf_create_chart(type, NETDATA_SYSCALL_APPS_TASK_PROCESS, "Process started", + EBPF_COMMON_DIMENSION_CALL, NETDATA_PROCESS_CGROUP_GROUP, + NETDATA_CGROUP_PROCESS_CREATE_CONTEXT, NETDATA_EBPF_CHART_TYPE_LINE, + NETDATA_CHART_PRIO_CGROUPS_CONTAINERS + 5000, + ebpf_create_global_dimension, &process_publish_aggregated[NETDATA_KEY_PUBLISH_PROCESS_FORK], + 1, em->update_every, NETDATA_EBPF_MODULE_NAME_PROCESS); + + ebpf_create_chart(type, NETDATA_SYSCALL_APPS_TASK_THREAD, "Threads started", + EBPF_COMMON_DIMENSION_CALL, NETDATA_PROCESS_CGROUP_GROUP, + NETDATA_CGROUP_THREAD_CREATE_CONTEXT, NETDATA_EBPF_CHART_TYPE_LINE, + NETDATA_CHART_PRIO_CGROUPS_CONTAINERS + 5001, + ebpf_create_global_dimension, + &process_publish_aggregated[NETDATA_KEY_PUBLISH_PROCESS_CLONE], + 1, em->update_every, NETDATA_EBPF_MODULE_NAME_PROCESS); + + ebpf_create_chart(type, NETDATA_SYSCALL_APPS_TASK_EXIT, "Tasks starts exit process.", + EBPF_COMMON_DIMENSION_CALL, NETDATA_PROCESS_CGROUP_GROUP, + NETDATA_CGROUP_PROCESS_EXIT_CONTEXT, + NETDATA_EBPF_CHART_TYPE_LINE, NETDATA_CHART_PRIO_CGROUPS_CONTAINERS + 5002, + ebpf_create_global_dimension, + &process_publish_aggregated[NETDATA_KEY_PUBLISH_PROCESS_EXIT], + 1, em->update_every, NETDATA_EBPF_MODULE_NAME_PROCESS); + + ebpf_create_chart(type, NETDATA_SYSCALL_APPS_TASK_CLOSE, "Tasks closed", + EBPF_COMMON_DIMENSION_CALL, NETDATA_PROCESS_CGROUP_GROUP, + NETDATA_CGROUP_PROCESS_CLOSE_CONTEXT, + NETDATA_EBPF_CHART_TYPE_LINE, NETDATA_CHART_PRIO_CGROUPS_CONTAINERS + 5003, + ebpf_create_global_dimension, + &process_publish_aggregated[NETDATA_KEY_PUBLISH_PROCESS_RELEASE_TASK], + 1, em->update_every, NETDATA_EBPF_MODULE_NAME_PROCESS); + + if (em->mode < MODE_ENTRY) { + ebpf_create_chart(type, NETDATA_SYSCALL_APPS_TASK_ERROR, "Errors to create process or threads.", + EBPF_COMMON_DIMENSION_CALL, NETDATA_PROCESS_CGROUP_GROUP, + NETDATA_CGROUP_PROCESS_ERROR_CONTEXT, + NETDATA_EBPF_CHART_TYPE_LINE, NETDATA_CHART_PRIO_CGROUPS_CONTAINERS + 5004, + ebpf_create_global_dimension, + &process_publish_aggregated[NETDATA_KEY_PUBLISH_PROCESS_EXIT], + 1, em->update_every, NETDATA_EBPF_MODULE_NAME_PROCESS); + } +} + +/** + * Obsolete specific process charts + * + * Obsolete charts for cgroup/application + * + * @param type the chart type. + * @param em the structure with thread information + */ +static void ebpf_obsolete_specific_process_charts(char *type, ebpf_module_t *em) +{ + ebpf_write_chart_obsolete(type, NETDATA_SYSCALL_APPS_TASK_PROCESS, "Process started", + EBPF_COMMON_DIMENSION_CALL, NETDATA_PROCESS_GROUP, NETDATA_EBPF_CHART_TYPE_LINE, + NETDATA_CGROUP_PROCESS_CREATE_CONTEXT, NETDATA_CHART_PRIO_CGROUPS_CONTAINERS + 5000, + em->update_every); + + ebpf_write_chart_obsolete(type, NETDATA_SYSCALL_APPS_TASK_THREAD, "Threads started", + EBPF_COMMON_DIMENSION_CALL, NETDATA_PROCESS_GROUP, NETDATA_EBPF_CHART_TYPE_LINE, + NETDATA_CGROUP_THREAD_CREATE_CONTEXT, NETDATA_CHART_PRIO_CGROUPS_CONTAINERS + 5001, + em->update_every); + + ebpf_write_chart_obsolete(type, NETDATA_SYSCALL_APPS_TASK_EXIT,"Tasks starts exit process.", + EBPF_COMMON_DIMENSION_CALL, NETDATA_PROCESS_GROUP, NETDATA_EBPF_CHART_TYPE_LINE, + NETDATA_CGROUP_PROCESS_EXIT_CONTEXT, NETDATA_CHART_PRIO_CGROUPS_CONTAINERS + 5002, + em->update_every); + + ebpf_write_chart_obsolete(type, NETDATA_SYSCALL_APPS_TASK_CLOSE,"Tasks closed", + EBPF_COMMON_DIMENSION_CALL, NETDATA_PROCESS_GROUP, NETDATA_EBPF_CHART_TYPE_LINE, + NETDATA_CGROUP_PROCESS_CLOSE_CONTEXT, NETDATA_CHART_PRIO_CGROUPS_CONTAINERS + 5003, + em->update_every); + + if (em->mode < MODE_ENTRY) { + ebpf_write_chart_obsolete(type, NETDATA_SYSCALL_APPS_TASK_ERROR,"Errors to create process or threads.", + EBPF_COMMON_DIMENSION_CALL, NETDATA_PROCESS_GROUP, NETDATA_EBPF_CHART_TYPE_LINE, + NETDATA_CGROUP_PROCESS_ERROR_CONTEXT, NETDATA_CHART_PRIO_CGROUPS_CONTAINERS + 5004, + em->update_every); + } +} + +/** + * Create Systemd process Charts + * + * Create charts when systemd is enabled + * + * @param em the structure with thread information + **/ +static void ebpf_create_systemd_process_charts(ebpf_module_t *em) +{ + ebpf_create_charts_on_systemd(NETDATA_SYSCALL_APPS_TASK_PROCESS, "Process started", + EBPF_COMMON_DIMENSION_CALL, NETDATA_APPS_PROCESS_GROUP, + NETDATA_EBPF_CHART_TYPE_STACKED, 20065, + ebpf_algorithms[NETDATA_EBPF_INCREMENTAL_IDX], NETDATA_SYSTEMD_PROCESS_CREATE_CONTEXT, + NETDATA_EBPF_MODULE_NAME_PROCESS, em->update_every); + + ebpf_create_charts_on_systemd(NETDATA_SYSCALL_APPS_TASK_THREAD, "Threads started", + EBPF_COMMON_DIMENSION_CALL, NETDATA_APPS_PROCESS_GROUP, + NETDATA_EBPF_CHART_TYPE_STACKED, 20066, + ebpf_algorithms[NETDATA_EBPF_INCREMENTAL_IDX], NETDATA_SYSTEMD_THREAD_CREATE_CONTEXT, + NETDATA_EBPF_MODULE_NAME_PROCESS, em->update_every); + + ebpf_create_charts_on_systemd(NETDATA_SYSCALL_APPS_TASK_CLOSE, "Tasks starts exit process.", + EBPF_COMMON_DIMENSION_CALL, NETDATA_APPS_PROCESS_GROUP, + NETDATA_EBPF_CHART_TYPE_STACKED, 20067, + ebpf_algorithms[NETDATA_EBPF_INCREMENTAL_IDX], NETDATA_SYSTEMD_PROCESS_EXIT_CONTEXT, + NETDATA_EBPF_MODULE_NAME_PROCESS, em->update_every); + + ebpf_create_charts_on_systemd(NETDATA_SYSCALL_APPS_TASK_EXIT, "Tasks closed", + EBPF_COMMON_DIMENSION_CALL, NETDATA_APPS_PROCESS_GROUP, + NETDATA_EBPF_CHART_TYPE_STACKED, 20068, + ebpf_algorithms[NETDATA_EBPF_INCREMENTAL_IDX], NETDATA_SYSTEMD_PROCESS_CLOSE_CONTEXT, + NETDATA_EBPF_MODULE_NAME_PROCESS, em->update_every); + + if (em->mode < MODE_ENTRY) { + ebpf_create_charts_on_systemd(NETDATA_SYSCALL_APPS_TASK_ERROR, "Errors to create process or threads.", + EBPF_COMMON_DIMENSION_CALL, NETDATA_APPS_PROCESS_GROUP, + NETDATA_EBPF_CHART_TYPE_STACKED, 20069, + ebpf_algorithms[NETDATA_EBPF_INCREMENTAL_IDX], NETDATA_SYSTEMD_PROCESS_ERROR_CONTEXT, + NETDATA_EBPF_MODULE_NAME_PROCESS, em->update_every); + } +} + +/** + * Send Systemd charts + * + * Send collected data to Netdata. + * + * @param em the structure with thread information + */ +static void ebpf_send_systemd_process_charts(ebpf_module_t *em) +{ + ebpf_cgroup_target_t *ect; + write_begin_chart(NETDATA_SERVICE_FAMILY, NETDATA_SYSCALL_APPS_TASK_PROCESS); + for (ect = ebpf_cgroup_pids; ect ; ect = ect->next) { + if (unlikely(ect->systemd) && unlikely(ect->updated)) { + write_chart_dimension(ect->name, ect->publish_systemd_ps.create_process); + } + } + write_end_chart(); + + write_begin_chart(NETDATA_SERVICE_FAMILY, NETDATA_SYSCALL_APPS_TASK_THREAD); + for (ect = ebpf_cgroup_pids; ect ; ect = ect->next) { + if (unlikely(ect->systemd) && unlikely(ect->updated)) { + write_chart_dimension(ect->name, ect->publish_systemd_ps.create_thread); + } + } + write_end_chart(); + + write_begin_chart(NETDATA_SERVICE_FAMILY, NETDATA_SYSCALL_APPS_TASK_EXIT); + for (ect = ebpf_cgroup_pids; ect ; ect = ect->next) { + if (unlikely(ect->systemd) && unlikely(ect->updated)) { + write_chart_dimension(ect->name, ect->publish_systemd_ps.exit_call); + } + } + write_end_chart(); + + write_begin_chart(NETDATA_SERVICE_FAMILY, NETDATA_SYSCALL_APPS_TASK_CLOSE); + for (ect = ebpf_cgroup_pids; ect ; ect = ect->next) { + if (unlikely(ect->systemd) && unlikely(ect->updated)) { + write_chart_dimension(ect->name, ect->publish_systemd_ps.release_call); + } + } + write_end_chart(); + + if (em->mode < MODE_ENTRY) { + write_begin_chart(NETDATA_SERVICE_FAMILY, NETDATA_SYSCALL_APPS_TASK_ERROR); + for (ect = ebpf_cgroup_pids; ect ; ect = ect->next) { + if (unlikely(ect->systemd) && unlikely(ect->updated)) { + write_chart_dimension(ect->name, ect->publish_systemd_ps.task_err); + } + } + write_end_chart(); + } +} + +/** + * Send data to Netdata calling auxiliary functions. + * + * @param em the structure with thread information +*/ +static void ebpf_process_send_cgroup_data(ebpf_module_t *em) +{ + if (!ebpf_cgroup_pids) + return; + + pthread_mutex_lock(&mutex_cgroup_shm); + ebpf_cgroup_target_t *ect; + for (ect = ebpf_cgroup_pids; ect ; ect = ect->next) { + ebpf_process_sum_cgroup_pids(&ect->publish_systemd_ps, ect->pids); + } + + int has_systemd = shm_ebpf_cgroup.header->systemd_enabled; + + if (has_systemd) { + if (send_cgroup_chart) { + ebpf_create_systemd_process_charts(em); + } + + ebpf_send_systemd_process_charts(em); + } + + for (ect = ebpf_cgroup_pids; ect ; ect = ect->next) { + if (ect->systemd) + continue; + + if (!(ect->flags & NETDATA_EBPF_CGROUP_HAS_PROCESS_CHART) && ect->updated) { + ebpf_create_specific_process_charts(ect->name, em); + ect->flags |= NETDATA_EBPF_CGROUP_HAS_PROCESS_CHART; + } + + if (ect->flags & NETDATA_EBPF_CGROUP_HAS_PROCESS_CHART) { + if (ect->updated) { + ebpf_send_specific_process_data(ect->name, &ect->publish_systemd_ps, em); + } else { + ebpf_obsolete_specific_process_charts(ect->name, em); + ect->flags &= ~NETDATA_EBPF_CGROUP_HAS_PROCESS_CHART; + } + } + } + + pthread_mutex_unlock(&mutex_cgroup_shm); +} + +/** + * Update Cgroup algorithm + * + * Change algorithm from absolute to incremental + */ +void ebpf_process_update_cgroup_algorithm() +{ + int i; + for (i = 0; i < NETDATA_KEY_PUBLISH_PROCESS_END; i++) { + netdata_publish_syscall_t *ptr = &process_publish_aggregated[i]; + freez(ptr->algorithm); + ptr->algorithm = strdupz(ebpf_algorithms[NETDATA_EBPF_INCREMENTAL_IDX]); + } +} + +/** + * Send Statistic Data + * + * Send statistic information to netdata. + */ +void ebpf_send_statistic_data() +{ + if (!publish_internal_metrics) + return; + + write_begin_chart(NETDATA_MONITORING_FAMILY, NETDATA_EBPF_THREADS); + write_chart_dimension(threads_stat[NETDATA_EBPF_THREAD_STAT_TOTAL], (long long)plugin_statistics.threads); + write_chart_dimension(threads_stat[NETDATA_EBPF_THREAD_STAT_RUNNING], (long long)plugin_statistics.running); + write_end_chart(); + + write_begin_chart(NETDATA_MONITORING_FAMILY, NETDATA_EBPF_LOAD_METHOD); + write_chart_dimension(load_event_stat[NETDATA_EBPF_LOAD_STAT_LEGACY], (long long)plugin_statistics.legacy); + write_chart_dimension(load_event_stat[NETDATA_EBPF_LOAD_STAT_CORE], (long long)plugin_statistics.core); + write_end_chart(); +} + +/** + * Main loop for this collector. + * + * @param em the structure with thread information + */ +static void process_collector(ebpf_module_t *em) +{ + // Start cgroup integration before other threads + cgroup_thread.thread = mallocz(sizeof(netdata_thread_t)); + cgroup_thread.start_routine = ebpf_cgroup_update_shm; + + netdata_thread_create(cgroup_thread.thread, cgroup_thread.name, NETDATA_THREAD_OPTION_DEFAULT, + ebpf_cgroup_update_shm, NULL); + + heartbeat_t hb; + heartbeat_init(&hb); + int publish_global = em->global_charts; + int cgroups = em->cgroup_charts; + int thread_enabled = em->enabled; + if (cgroups) + ebpf_process_update_cgroup_algorithm(); + + int update_apps_every = (int) EBPF_CFG_UPDATE_APPS_EVERY_DEFAULT; + int pid_fd = process_maps[NETDATA_PROCESS_PID_TABLE].map_fd; + int update_every = em->update_every; + int counter = update_every - 1; + int update_apps_list = update_apps_every - 1; + while (!ebpf_exit_plugin) { + usec_t dt = heartbeat_next(&hb, USEC_PER_SEC); + (void)dt; + if (ebpf_exit_plugin) + break; + + pthread_mutex_lock(&collect_data_mutex); + if (++update_apps_list == update_apps_every) { + update_apps_list = 0; + cleanup_exited_pids(); + collect_data_for_all_processes(pid_fd); + } + pthread_mutex_unlock(&collect_data_mutex); + + if (++counter == update_every) { + counter = 0; + + read_hash_global_tables(); + + netdata_apps_integration_flags_t apps_enabled = em->apps_charts; + pthread_mutex_lock(&collect_data_mutex); + + ebpf_create_apps_charts(apps_groups_root_target); + if (all_pids_count > 0) { + if (apps_enabled) { + ebpf_process_update_apps_data(); + } + + if (cgroups) { + ebpf_update_process_cgroup(); + } + } + + pthread_mutex_lock(&lock); + ebpf_send_statistic_data(); + + if (thread_enabled) { + if (publish_global) { + ebpf_process_send_data(em); + } + + if (apps_enabled & NETDATA_EBPF_APPS_FLAG_CHART_CREATED) { + ebpf_process_send_apps_data(apps_groups_root_target, em); + } + + if (cgroups) { + ebpf_process_send_cgroup_data(em); + } + } + pthread_mutex_unlock(&lock); + pthread_mutex_unlock(&collect_data_mutex); + } + + fflush(stdout); + } +} + +/***************************************************************** + * + * FUNCTIONS TO START THREAD + * + *****************************************************************/ + +/** + * Allocate vectors used with this thread. + * We are not testing the return, because callocz does this and shutdown the software + * case it was not possible to allocate. + * + * @param length is the length for the vectors used inside the collector. + */ +static void ebpf_process_allocate_global_vectors(size_t length) +{ + memset(process_aggregated_data, 0, length * sizeof(netdata_syscall_stat_t)); + memset(process_publish_aggregated, 0, length * sizeof(netdata_publish_syscall_t)); + process_hash_values = callocz(ebpf_nprocs, sizeof(netdata_idx_t)); + + global_process_stats = callocz((size_t)pid_max, sizeof(ebpf_process_stat_t *)); + current_apps_data = callocz((size_t)pid_max, sizeof(ebpf_process_publish_apps_t *)); +} + +static void change_syscalls() +{ + static char *lfork = { "do_fork" }; + process_id_names[NETDATA_KEY_PUBLISH_PROCESS_FORK] = lfork; +} + +/** + * Set local variables + * + */ +static void set_local_pointers() +{ + if (isrh >= NETDATA_MINIMUM_RH_VERSION && isrh < NETDATA_RH_8) + change_syscalls(); +} + +/***************************************************************** + * + * EBPF PROCESS THREAD + * + *****************************************************************/ + +/** + * Enable tracepoints + * + * Enable necessary tracepoints for thread. + * + * @return It returns 0 on success and -1 otherwise + */ +static int ebpf_process_enable_tracepoints() +{ + int test = ebpf_is_tracepoint_enabled(tracepoint_sched_type, tracepoint_sched_process_exit); + if (test == -1) + return -1; + else if (!test) { + if (ebpf_enable_tracing_values(tracepoint_sched_type, tracepoint_sched_process_exit)) + return -1; + } + was_sched_process_exit_enabled = test; + + test = ebpf_is_tracepoint_enabled(tracepoint_sched_type, tracepoint_sched_process_exec); + if (test == -1) + return -1; + else if (!test) { + if (ebpf_enable_tracing_values(tracepoint_sched_type, tracepoint_sched_process_exec)) + return -1; + } + was_sched_process_exec_enabled = test; + + test = ebpf_is_tracepoint_enabled(tracepoint_sched_type, tracepoint_sched_process_fork); + if (test == -1) + return -1; + else if (!test) { + if (ebpf_enable_tracing_values(tracepoint_sched_type, tracepoint_sched_process_fork)) + return -1; + } + was_sched_process_fork_enabled = test; + + return 0; +} + +/** + * Process thread + * + * Thread used to generate process charts. + * + * @param ptr a pointer to `struct ebpf_module` + * + * @return It always return NULL + */ +void *ebpf_process_thread(void *ptr) +{ + netdata_thread_cleanup_push(ebpf_process_exit, ptr); + + ebpf_module_t *em = (ebpf_module_t *)ptr; + em->maps = process_maps; + + if (ebpf_process_enable_tracepoints()) { + em->enabled = em->global_charts = em->apps_charts = em->cgroup_charts = CONFIG_BOOLEAN_NO; + } + process_enabled = em->enabled; + + pthread_mutex_lock(&lock); + ebpf_process_allocate_global_vectors(NETDATA_KEY_PUBLISH_PROCESS_END); + + ebpf_update_pid_table(&process_maps[0], em); + + set_local_pointers(); + em->probe_links = ebpf_load_program(ebpf_plugin_dir, em, running_on_kernel, isrh, &em->objects); + if (!em->probe_links) { + em->enabled = CONFIG_BOOLEAN_NO; + pthread_mutex_unlock(&lock); + goto endprocess; + } + + int algorithms[NETDATA_KEY_PUBLISH_PROCESS_END] = { + NETDATA_EBPF_ABSOLUTE_IDX, NETDATA_EBPF_ABSOLUTE_IDX, NETDATA_EBPF_ABSOLUTE_IDX, NETDATA_EBPF_ABSOLUTE_IDX + }; + + ebpf_global_labels( + process_aggregated_data, process_publish_aggregated, process_dimension_names, process_id_names, + algorithms, NETDATA_KEY_PUBLISH_PROCESS_END); + + if (process_enabled) { + ebpf_create_global_charts(em); + } + + ebpf_update_stats(&plugin_statistics, em); + ebpf_create_statistic_charts(em); + + pthread_mutex_unlock(&lock); + + process_collector(em); + +endprocess: + if (!em->enabled) + ebpf_update_disabled_plugin_stats(em); + + netdata_thread_cleanup_pop(1); + return NULL; +} diff --git a/collectors/ebpf.plugin/ebpf_process.h b/collectors/ebpf.plugin/ebpf_process.h new file mode 100644 index 0000000..43df34d --- /dev/null +++ b/collectors/ebpf.plugin/ebpf_process.h @@ -0,0 +1,107 @@ +// SPDX-License-Identifier: GPL-3.0-or-later + +#ifndef NETDATA_EBPF_PROCESS_H +#define NETDATA_EBPF_PROCESS_H 1 + +// Module name +#define NETDATA_EBPF_MODULE_NAME_PROCESS "process" + +// Groups used on Dashboard +#define NETDATA_PROCESS_GROUP "processes" +#define NETDATA_PROCESS_CGROUP_GROUP "processes (eBPF)" + +// Global chart name +#define NETDATA_EXIT_SYSCALL "exit" +#define NETDATA_PROCESS_SYSCALL "process_thread" +#define NETDATA_PROCESS_ERROR_NAME "task_error" +#define NETDATA_PROCESS_STATUS_NAME "process_status" + +// Charts created on Apps submenu +#define NETDATA_SYSCALL_APPS_TASK_PROCESS "process_create" +#define NETDATA_SYSCALL_APPS_TASK_THREAD "thread_create" +#define NETDATA_SYSCALL_APPS_TASK_EXIT "task_exit" +#define NETDATA_SYSCALL_APPS_TASK_CLOSE "task_close" +#define NETDATA_SYSCALL_APPS_TASK_ERROR "task_error" + +// Process configuration name +#define NETDATA_PROCESS_CONFIG_FILE "process.conf" + +// Contexts +#define NETDATA_CGROUP_PROCESS_CREATE_CONTEXT "cgroup.process_create" +#define NETDATA_CGROUP_THREAD_CREATE_CONTEXT "cgroup.thread_create" +#define NETDATA_CGROUP_PROCESS_CLOSE_CONTEXT "cgroup.task_close" +#define NETDATA_CGROUP_PROCESS_EXIT_CONTEXT "cgroup.task_exit" +#define NETDATA_CGROUP_PROCESS_ERROR_CONTEXT "cgroup.task_error" + +#define NETDATA_SYSTEMD_PROCESS_CREATE_CONTEXT "services.process_create" +#define NETDATA_SYSTEMD_THREAD_CREATE_CONTEXT "services.thread_create" +#define NETDATA_SYSTEMD_PROCESS_CLOSE_CONTEXT "services.task_close" +#define NETDATA_SYSTEMD_PROCESS_EXIT_CONTEXT "services.task_exit" +#define NETDATA_SYSTEMD_PROCESS_ERROR_CONTEXT "services.task_error" + +#define NETDATA_EBPF_CGROUP_UPDATE 10 + +// Statistical information +enum netdata_ebpf_thread_stats{ + NETDATA_EBPF_THREAD_STAT_TOTAL, + NETDATA_EBPF_THREAD_STAT_RUNNING, + + NETDATA_EBPF_THREAD_STAT_END +}; + +enum netdata_ebpf_load_mode_stats{ + NETDATA_EBPF_LOAD_STAT_LEGACY, + NETDATA_EBPF_LOAD_STAT_CORE, + + NETDATA_EBPF_LOAD_STAT_END +}; + +// Index from kernel +typedef enum ebpf_process_index { + NETDATA_KEY_CALLS_DO_EXIT, + + NETDATA_KEY_CALLS_RELEASE_TASK, + + NETDATA_KEY_CALLS_DO_FORK, + NETDATA_KEY_ERROR_DO_FORK, + + NETDATA_KEY_CALLS_SYS_CLONE, + NETDATA_KEY_ERROR_SYS_CLONE, + + NETDATA_KEY_END_VECTOR +} ebpf_process_index_t; + +// This enum acts as an index for publish vector. +// Do not change the enum order because we use +// different algorithms to make charts with incremental +// values (the three initial positions) and absolute values +// (the remaining charts). +typedef enum netdata_publish_process { + NETDATA_KEY_PUBLISH_PROCESS_EXIT, + NETDATA_KEY_PUBLISH_PROCESS_RELEASE_TASK, + NETDATA_KEY_PUBLISH_PROCESS_FORK, + NETDATA_KEY_PUBLISH_PROCESS_CLONE, + + NETDATA_KEY_PUBLISH_PROCESS_END +} netdata_publish_process_t; + +typedef struct ebpf_process_publish_apps { + // Number of calls during the last read + uint64_t call_do_exit; + uint64_t call_release_task; + uint64_t create_process; + uint64_t create_thread; + + // Number of errors during the last read + uint64_t task_err; +} ebpf_process_publish_apps_t; + +enum ebpf_process_tables { + NETDATA_PROCESS_PID_TABLE, + NETDATA_PROCESS_GLOBAL_TABLE, + NETDATA_PROCESS_CTRL_TABLE +}; + +extern struct config process_config; + +#endif /* NETDATA_EBPF_PROCESS_H */ diff --git a/collectors/ebpf.plugin/ebpf_shm.c b/collectors/ebpf.plugin/ebpf_shm.c new file mode 100644 index 0000000..f81287d --- /dev/null +++ b/collectors/ebpf.plugin/ebpf_shm.c @@ -0,0 +1,1137 @@ +// SPDX-License-Identifier: GPL-3.0-or-later + +#include "ebpf.h" +#include "ebpf_shm.h" + +static char *shm_dimension_name[NETDATA_SHM_END] = { "get", "at", "dt", "ctl" }; +static netdata_syscall_stat_t shm_aggregated_data[NETDATA_SHM_END]; +static netdata_publish_syscall_t shm_publish_aggregated[NETDATA_SHM_END]; + +netdata_publish_shm_t *shm_vector = NULL; + +static netdata_idx_t shm_hash_values[NETDATA_SHM_END]; +static netdata_idx_t *shm_values = NULL; + +netdata_publish_shm_t **shm_pid = NULL; + +struct config shm_config = { .first_section = NULL, + .last_section = NULL, + .mutex = NETDATA_MUTEX_INITIALIZER, + .index = { .avl_tree = { .root = NULL, .compar = appconfig_section_compare }, + .rwlock = AVL_LOCK_INITIALIZER } }; + +static ebpf_local_maps_t shm_maps[] = {{.name = "tbl_pid_shm", .internal_input = ND_EBPF_DEFAULT_PID_SIZE, + .user_input = 0, + .type = NETDATA_EBPF_MAP_RESIZABLE | NETDATA_EBPF_MAP_PID, + .map_fd = ND_EBPF_MAP_FD_NOT_INITIALIZED}, + {.name = "shm_ctrl", .internal_input = NETDATA_CONTROLLER_END, + .user_input = 0, + .type = NETDATA_EBPF_MAP_CONTROLLER, + .map_fd = ND_EBPF_MAP_FD_NOT_INITIALIZED}, + {.name = "tbl_shm", .internal_input = NETDATA_SHM_END, + .user_input = 0, + .type = NETDATA_EBPF_MAP_STATIC, + .map_fd = ND_EBPF_MAP_FD_NOT_INITIALIZED}, + {.name = NULL, .internal_input = 0, .user_input = 0}}; + +struct netdata_static_thread shm_threads = { + .name = "SHM KERNEL", + .config_section = NULL, + .config_name = NULL, + .env_name = NULL, + .enabled = 1, + .thread = NULL, + .init_routine = NULL, + .start_routine = NULL +}; + +netdata_ebpf_targets_t shm_targets[] = { {.name = "shmget", .mode = EBPF_LOAD_TRAMPOLINE}, + {.name = "shmat", .mode = EBPF_LOAD_TRAMPOLINE}, + {.name = "shmdt", .mode = EBPF_LOAD_TRAMPOLINE}, + {.name = "shmctl", .mode = EBPF_LOAD_TRAMPOLINE}, + {.name = NULL, .mode = EBPF_LOAD_TRAMPOLINE}}; + +#ifdef LIBBPF_MAJOR_VERSION +#include "includes/shm.skel.h" + +static struct shm_bpf *bpf_obj = NULL; + +/***************************************************************** + * + * BTF FUNCTIONS + * + *****************************************************************/ + +/* + * Disable tracepoint + * + * Disable all tracepoints to use exclusively another method. + * + * @param obj is the main structure for bpf objects. + */ +static void ebpf_shm_disable_tracepoint(struct shm_bpf *obj) +{ + bpf_program__set_autoload(obj->progs.netdata_syscall_shmget, false); + bpf_program__set_autoload(obj->progs.netdata_syscall_shmat, false); + bpf_program__set_autoload(obj->progs.netdata_syscall_shmdt, false); + bpf_program__set_autoload(obj->progs.netdata_syscall_shmctl, false); +} + +/* + * Disable probe + * + * Disable all probes to use exclusively another method. + * + * @param obj is the main structure for bpf objects. + */ +static void ebpf_disable_probe(struct shm_bpf *obj) +{ + bpf_program__set_autoload(obj->progs.netdata_shmget_probe, false); + bpf_program__set_autoload(obj->progs.netdata_shmat_probe, false); + bpf_program__set_autoload(obj->progs.netdata_shmdt_probe, false); + bpf_program__set_autoload(obj->progs.netdata_shmctl_probe, false); + bpf_program__set_autoload(obj->progs.netdata_shm_release_task_probe, false); +} + +/* + * Disable trampoline + * + * Disable all trampoline to use exclusively another method. + * + * @param obj is the main structure for bpf objects. + */ +static void ebpf_disable_trampoline(struct shm_bpf *obj) +{ + bpf_program__set_autoload(obj->progs.netdata_shmget_fentry, false); + bpf_program__set_autoload(obj->progs.netdata_shmat_fentry, false); + bpf_program__set_autoload(obj->progs.netdata_shmdt_fentry, false); + bpf_program__set_autoload(obj->progs.netdata_shmctl_fentry, false); + bpf_program__set_autoload(obj->progs.netdata_shm_release_task_fentry, false); +} + +/** + * Set trampoline target + * + * Set the targets we will monitor. + * + * @param obj is the main structure for bpf objects. + */ +static void ebpf_set_trampoline_target(struct shm_bpf *obj) +{ + char syscall[NETDATA_EBPF_MAX_SYSCALL_LENGTH + 1]; + ebpf_select_host_prefix(syscall, NETDATA_EBPF_MAX_SYSCALL_LENGTH, + shm_targets[NETDATA_KEY_SHMGET_CALL].name, running_on_kernel); + + bpf_program__set_attach_target(obj->progs.netdata_shmget_fentry, 0, + syscall); + + ebpf_select_host_prefix(syscall, NETDATA_EBPF_MAX_SYSCALL_LENGTH, + shm_targets[NETDATA_KEY_SHMAT_CALL].name, running_on_kernel); + bpf_program__set_attach_target(obj->progs.netdata_shmat_fentry, 0, + syscall); + + ebpf_select_host_prefix(syscall, NETDATA_EBPF_MAX_SYSCALL_LENGTH, + shm_targets[NETDATA_KEY_SHMDT_CALL].name, running_on_kernel); + bpf_program__set_attach_target(obj->progs.netdata_shmdt_fentry, 0, + syscall); + + ebpf_select_host_prefix(syscall, NETDATA_EBPF_MAX_SYSCALL_LENGTH, + shm_targets[NETDATA_KEY_SHMCTL_CALL].name, running_on_kernel); + bpf_program__set_attach_target(obj->progs.netdata_shmctl_fentry, 0, + syscall); + + bpf_program__set_attach_target(obj->progs.netdata_shm_release_task_fentry, 0, + EBPF_COMMON_FNCT_CLEAN_UP); +} + +/** + * SHM Attach Probe + * + * Attach probes to target + * + * @param obj is the main structure for bpf objects. + * + * @return It returns 0 on success and -1 otherwise. + */ +static int ebpf_shm_attach_probe(struct shm_bpf *obj) +{ + char syscall[NETDATA_EBPF_MAX_SYSCALL_LENGTH + 1]; + ebpf_select_host_prefix(syscall, NETDATA_EBPF_MAX_SYSCALL_LENGTH, + shm_targets[NETDATA_KEY_SHMGET_CALL].name, running_on_kernel); + + obj->links.netdata_shmget_probe = bpf_program__attach_kprobe(obj->progs.netdata_shmget_probe, + false, syscall); + int ret = (int)libbpf_get_error(obj->links.netdata_shmget_probe); + if (ret) + return -1; + + ebpf_select_host_prefix(syscall, NETDATA_EBPF_MAX_SYSCALL_LENGTH, + shm_targets[NETDATA_KEY_SHMAT_CALL].name, running_on_kernel); + obj->links.netdata_shmat_probe = bpf_program__attach_kprobe(obj->progs.netdata_shmat_probe, + false, syscall); + ret = (int)libbpf_get_error(obj->links.netdata_shmat_probe); + if (ret) + return -1; + + ebpf_select_host_prefix(syscall, NETDATA_EBPF_MAX_SYSCALL_LENGTH, + shm_targets[NETDATA_KEY_SHMDT_CALL].name, running_on_kernel); + obj->links.netdata_shmdt_probe = bpf_program__attach_kprobe(obj->progs.netdata_shmdt_probe, + false, syscall); + ret = (int)libbpf_get_error(obj->links.netdata_shmdt_probe); + if (ret) + return -1; + + ebpf_select_host_prefix(syscall, NETDATA_EBPF_MAX_SYSCALL_LENGTH, + shm_targets[NETDATA_KEY_SHMCTL_CALL].name, running_on_kernel); + obj->links.netdata_shmctl_probe = bpf_program__attach_kprobe(obj->progs.netdata_shmctl_probe, + false, syscall); + ret = (int)libbpf_get_error(obj->links.netdata_shmctl_probe); + if (ret) + return -1; + + obj->links.netdata_shm_release_task_probe = bpf_program__attach_kprobe(obj->progs.netdata_shm_release_task_probe, + false, EBPF_COMMON_FNCT_CLEAN_UP); + ret = (int)libbpf_get_error(obj->links.netdata_shm_release_task_probe); + if (ret) + return -1; + + + return 0; +} + +/** + * Set hash tables + * + * Set the values for maps according the value given by kernel. + */ +static void ebpf_shm_set_hash_tables(struct shm_bpf *obj) +{ + shm_maps[NETDATA_PID_SHM_TABLE].map_fd = bpf_map__fd(obj->maps.tbl_pid_shm); + shm_maps[NETDATA_SHM_CONTROLLER].map_fd = bpf_map__fd(obj->maps.shm_ctrl); + shm_maps[NETDATA_SHM_GLOBAL_TABLE].map_fd = bpf_map__fd(obj->maps.tbl_shm); +} + +/** + * Disable Release Task + * + * Disable release task when apps is not enabled. + * + * @param obj is the main structure for bpf objects. + */ +static void ebpf_shm_disable_release_task(struct shm_bpf *obj) +{ + bpf_program__set_autoload(obj->progs.netdata_shm_release_task_probe, false); + bpf_program__set_autoload(obj->progs.netdata_shm_release_task_fentry, false); +} + +/** + * Adjust Map Size + * + * Resize maps according input from users. + * + * @param obj is the main structure for bpf objects. + * @param em structure with configuration + */ +static void ebpf_shm_adjust_map_size(struct shm_bpf *obj, ebpf_module_t *em) +{ + ebpf_update_map_size(obj->maps.tbl_pid_shm, &shm_maps[NETDATA_PID_SHM_TABLE], + em, bpf_map__name(obj->maps.tbl_pid_shm)); +} + +/** + * Load and attach + * + * Load and attach the eBPF code in kernel. + * + * @param obj is the main structure for bpf objects. + * @param em structure with configuration + * + * @return it returns 0 on succes and -1 otherwise + */ +static inline int ebpf_shm_load_and_attach(struct shm_bpf *obj, ebpf_module_t *em) +{ + netdata_ebpf_targets_t *shmt = em->targets; + netdata_ebpf_program_loaded_t test = shmt[NETDATA_KEY_SHMGET_CALL].mode; + + // We are testing only one, because all will have the same behavior + if (test == EBPF_LOAD_TRAMPOLINE ) { + ebpf_shm_disable_tracepoint(obj); + ebpf_disable_probe(obj); + + ebpf_set_trampoline_target(obj); + } else if (test == EBPF_LOAD_PROBE || test == EBPF_LOAD_RETPROBE ) { + ebpf_shm_disable_tracepoint(obj); + ebpf_disable_trampoline(obj); + } else { + ebpf_disable_probe(obj); + ebpf_disable_trampoline(obj); + } + + ebpf_shm_adjust_map_size(obj, em); + if (!em->apps_charts && !em->cgroup_charts) + ebpf_shm_disable_release_task(obj); + + int ret = shm_bpf__load(obj); + if (!ret) { + if (test != EBPF_LOAD_PROBE && test != EBPF_LOAD_RETPROBE) + shm_bpf__attach(obj); + else + ret = ebpf_shm_attach_probe(obj); + + if (!ret) + ebpf_shm_set_hash_tables(obj); + } + + return ret; +} +#endif +/***************************************************************** + * FUNCTIONS TO CLOSE THE THREAD + *****************************************************************/ + +/** + * SHM Free + * + * Cleanup variables after child threads to stop + * + * @param ptr thread data. + */ +static void ebpf_shm_free(ebpf_module_t *em) +{ + pthread_mutex_lock(&ebpf_exit_cleanup); + if (em->thread->enabled == NETDATA_THREAD_EBPF_RUNNING) { + em->thread->enabled = NETDATA_THREAD_EBPF_STOPPING; + pthread_mutex_unlock(&ebpf_exit_cleanup); + return; + } + pthread_mutex_unlock(&ebpf_exit_cleanup); + + ebpf_cleanup_publish_syscall(shm_publish_aggregated); + + freez(shm_vector); + freez(shm_values); + +#ifdef LIBBPF_MAJOR_VERSION + if (bpf_obj) + shm_bpf__destroy(bpf_obj); +#endif + + em->thread->enabled = NETDATA_THREAD_EBPF_STOPPED; +} + +/** + * SHM Exit + * + * Cancel child thread. + * + * @param ptr thread data. + */ +static void ebpf_shm_exit(void *ptr) +{ + ebpf_module_t *em = (ebpf_module_t *)ptr; + netdata_thread_cancel(*shm_threads.thread); + ebpf_shm_free(em); +} + +/** + * SHM Cleanup + * + * Clean up allocated memory. + * + * @param ptr thread data. + */ +static void ebpf_shm_cleanup(void *ptr) +{ + ebpf_module_t *em = (ebpf_module_t *)ptr; + ebpf_shm_free(em); +} + +/***************************************************************** + * COLLECTOR THREAD + *****************************************************************/ + +/** + * Apps Accumulator + * + * Sum all values read from kernel and store in the first address. + * + * @param out the vector with read values. + */ +static void shm_apps_accumulator(netdata_publish_shm_t *out) +{ + int i, end = (running_on_kernel >= NETDATA_KERNEL_V4_15) ? ebpf_nprocs : 1; + netdata_publish_shm_t *total = &out[0]; + for (i = 1; i < end; i++) { + netdata_publish_shm_t *w = &out[i]; + total->get += w->get; + total->at += w->at; + total->dt += w->dt; + total->ctl += w->ctl; + } +} + +/** + * Fill PID + * + * Fill PID structures + * + * @param current_pid pid that we are collecting data + * @param out values read from hash tables; + */ +static void shm_fill_pid(uint32_t current_pid, netdata_publish_shm_t *publish) +{ + netdata_publish_shm_t *curr = shm_pid[current_pid]; + if (!curr) { + curr = callocz(1, sizeof(netdata_publish_shm_t)); + shm_pid[current_pid] = curr; + } + + memcpy(curr, publish, sizeof(netdata_publish_shm_t)); +} + +/** + * Update cgroup + * + * Update cgroup data based in + */ +static void ebpf_update_shm_cgroup() +{ + netdata_publish_shm_t *cv = shm_vector; + int fd = shm_maps[NETDATA_PID_SHM_TABLE].map_fd; + size_t length = sizeof(netdata_publish_shm_t) * ebpf_nprocs; + ebpf_cgroup_target_t *ect; + + memset(cv, 0, length); + + pthread_mutex_lock(&mutex_cgroup_shm); + for (ect = ebpf_cgroup_pids; ect; ect = ect->next) { + struct pid_on_target2 *pids; + for (pids = ect->pids; pids; pids = pids->next) { + int pid = pids->pid; + netdata_publish_shm_t *out = &pids->shm; + if (likely(shm_pid) && shm_pid[pid]) { + netdata_publish_shm_t *in = shm_pid[pid]; + + memcpy(out, in, sizeof(netdata_publish_shm_t)); + } else { + if (!bpf_map_lookup_elem(fd, &pid, cv)) { + shm_apps_accumulator(cv); + + memcpy(out, cv, sizeof(netdata_publish_shm_t)); + + // now that we've consumed the value, zero it out in the map. + memset(cv, 0, length); + bpf_map_update_elem(fd, &pid, cv, BPF_EXIST); + } + } + } + } + pthread_mutex_unlock(&mutex_cgroup_shm); +} + +/** + * Read APPS table + * + * Read the apps table and store data inside the structure. + */ +static void read_apps_table() +{ + netdata_publish_shm_t *cv = shm_vector; + uint32_t key; + struct pid_stat *pids = root_of_pids; + int fd = shm_maps[NETDATA_PID_SHM_TABLE].map_fd; + size_t length = sizeof(netdata_publish_shm_t)*ebpf_nprocs; + while (pids) { + key = pids->pid; + + if (bpf_map_lookup_elem(fd, &key, cv)) { + pids = pids->next; + continue; + } + + shm_apps_accumulator(cv); + + shm_fill_pid(key, cv); + + // now that we've consumed the value, zero it out in the map. + memset(cv, 0, length); + bpf_map_update_elem(fd, &key, cv, BPF_EXIST); + + pids = pids->next; + } +} + +/** +* Send global charts to netdata agent. +*/ +static void shm_send_global() +{ + write_begin_chart(NETDATA_EBPF_SYSTEM_GROUP, NETDATA_SHM_GLOBAL_CHART); + write_chart_dimension( + shm_publish_aggregated[NETDATA_KEY_SHMGET_CALL].dimension, + (long long) shm_hash_values[NETDATA_KEY_SHMGET_CALL] + ); + write_chart_dimension( + shm_publish_aggregated[NETDATA_KEY_SHMAT_CALL].dimension, + (long long) shm_hash_values[NETDATA_KEY_SHMAT_CALL] + ); + write_chart_dimension( + shm_publish_aggregated[NETDATA_KEY_SHMDT_CALL].dimension, + (long long) shm_hash_values[NETDATA_KEY_SHMDT_CALL] + ); + write_chart_dimension( + shm_publish_aggregated[NETDATA_KEY_SHMCTL_CALL].dimension, + (long long) shm_hash_values[NETDATA_KEY_SHMCTL_CALL] + ); + write_end_chart(); +} + +/** + * Read global counter + * + * Read the table with number of calls for all functions + */ +static void read_global_table() +{ + netdata_idx_t *stored = shm_values; + netdata_idx_t *val = shm_hash_values; + int fd = shm_maps[NETDATA_SHM_GLOBAL_TABLE].map_fd; + + uint32_t i, end = NETDATA_SHM_END; + for (i = NETDATA_KEY_SHMGET_CALL; i < end; i++) { + if (!bpf_map_lookup_elem(fd, &i, stored)) { + int j; + int last = ebpf_nprocs; + netdata_idx_t total = 0; + for (j = 0; j < last; j++) + total += stored[j]; + + val[i] = total; + } + } +} + +/** + * Shared memory reader thread. + * + * @param ptr It is a NULL value for this thread. + * @return It always returns NULL. + */ +void *ebpf_shm_read_hash(void *ptr) +{ + netdata_thread_cleanup_push(ebpf_shm_cleanup, ptr); + heartbeat_t hb; + heartbeat_init(&hb); + + ebpf_module_t *em = (ebpf_module_t *)ptr; + usec_t step = NETDATA_SHM_SLEEP_MS * em->update_every; + while (!ebpf_exit_plugin) { + (void)heartbeat_next(&hb, step); + + read_global_table(); + } + + netdata_thread_cleanup_pop(1); + return NULL; +} + +/** + * Sum values for all targets. + */ +static void ebpf_shm_sum_pids(netdata_publish_shm_t *shm, struct pid_on_target *root) +{ + while (root) { + int32_t pid = root->pid; + netdata_publish_shm_t *w = shm_pid[pid]; + if (w) { + shm->get += w->get; + shm->at += w->at; + shm->dt += w->dt; + shm->ctl += w->ctl; + + // reset for next collection. + w->get = 0; + w->at = 0; + w->dt = 0; + w->ctl = 0; + } + root = root->next; + } +} + +/** + * Send data to Netdata calling auxiliary functions. + * + * @param root the target list. +*/ +void ebpf_shm_send_apps_data(struct target *root) +{ + struct target *w; + for (w = root; w; w = w->next) { + if (unlikely(w->exposed && w->processes)) { + ebpf_shm_sum_pids(&w->shm, w->root_pid); + } + } + + write_begin_chart(NETDATA_APPS_FAMILY, NETDATA_SHMGET_CHART); + for (w = root; w; w = w->next) { + if (unlikely(w->exposed && w->processes)) { + write_chart_dimension(w->name, (long long) w->shm.get); + } + } + write_end_chart(); + + write_begin_chart(NETDATA_APPS_FAMILY, NETDATA_SHMAT_CHART); + for (w = root; w; w = w->next) { + if (unlikely(w->exposed && w->processes)) { + write_chart_dimension(w->name, (long long) w->shm.at); + } + } + write_end_chart(); + + write_begin_chart(NETDATA_APPS_FAMILY, NETDATA_SHMDT_CHART); + for (w = root; w; w = w->next) { + if (unlikely(w->exposed && w->processes)) { + write_chart_dimension(w->name, (long long) w->shm.dt); + } + } + write_end_chart(); + + write_begin_chart(NETDATA_APPS_FAMILY, NETDATA_SHMCTL_CHART); + for (w = root; w; w = w->next) { + if (unlikely(w->exposed && w->processes)) { + write_chart_dimension(w->name, (long long) w->shm.ctl); + } + } + write_end_chart(); +} + +/** + * Sum values for all targets. + */ +static void ebpf_shm_sum_cgroup_pids(netdata_publish_shm_t *shm, struct pid_on_target2 *root) +{ + netdata_publish_shm_t shmv; + memset(&shmv, 0, sizeof(shmv)); + while (root) { + netdata_publish_shm_t *w = &root->shm; + shmv.get += w->get; + shmv.at += w->at; + shmv.dt += w->dt; + shmv.ctl += w->ctl; + + root = root->next; + } + + memcpy(shm, &shmv, sizeof(shmv)); +} + +/** + * Create specific shared memory charts + * + * Create charts for cgroup/application. + * + * @param type the chart type. + * @param update_every value to overwrite the update frequency set by the server. + */ +static void ebpf_create_specific_shm_charts(char *type, int update_every) +{ + ebpf_create_chart(type, NETDATA_SHMGET_CHART, + "Calls to syscall <code>shmget(2)</code>.", + EBPF_COMMON_DIMENSION_CALL, + NETDATA_APPS_IPC_SHM_GROUP, + NETDATA_CGROUP_SHM_GET_CONTEXT, + NETDATA_EBPF_CHART_TYPE_LINE, + NETDATA_CHART_PRIO_CGROUPS_CONTAINERS + 5800, + ebpf_create_global_dimension, + &shm_publish_aggregated[NETDATA_KEY_SHMGET_CALL], + 1, + update_every, + NETDATA_EBPF_MODULE_NAME_SHM); + + ebpf_create_chart(type, NETDATA_SHMAT_CHART, + "Calls to syscall <code>shmat(2)</code>.", + EBPF_COMMON_DIMENSION_CALL, + NETDATA_APPS_IPC_SHM_GROUP, + NETDATA_CGROUP_SHM_AT_CONTEXT, + NETDATA_EBPF_CHART_TYPE_LINE, + NETDATA_CHART_PRIO_CGROUPS_CONTAINERS + 5801, + ebpf_create_global_dimension, + &shm_publish_aggregated[NETDATA_KEY_SHMAT_CALL], + 1, + update_every, + NETDATA_EBPF_MODULE_NAME_SHM); + + ebpf_create_chart(type, NETDATA_SHMDT_CHART, + "Calls to syscall <code>shmdt(2)</code>.", + EBPF_COMMON_DIMENSION_CALL, + NETDATA_APPS_IPC_SHM_GROUP, + NETDATA_CGROUP_SHM_DT_CONTEXT, + NETDATA_EBPF_CHART_TYPE_LINE, + NETDATA_CHART_PRIO_CGROUPS_CONTAINERS + 5802, + ebpf_create_global_dimension, + &shm_publish_aggregated[NETDATA_KEY_SHMDT_CALL], + 1, + update_every, + NETDATA_EBPF_MODULE_NAME_SHM); + + ebpf_create_chart(type, NETDATA_SHMCTL_CHART, + "Calls to syscall <code>shmctl(2)</code>.", + EBPF_COMMON_DIMENSION_CALL, + NETDATA_APPS_IPC_SHM_GROUP, + NETDATA_CGROUP_SHM_CTL_CONTEXT, + NETDATA_EBPF_CHART_TYPE_LINE, + NETDATA_CHART_PRIO_CGROUPS_CONTAINERS + 5803, + ebpf_create_global_dimension, + &shm_publish_aggregated[NETDATA_KEY_SHMCTL_CALL], + 1, + update_every, + NETDATA_EBPF_MODULE_NAME_SHM); +} + +/** + * Obsolete specific shared memory charts + * + * Obsolete charts for cgroup/application. + * + * @param type the chart type. + * @param update_every value to overwrite the update frequency set by the server. + */ +static void ebpf_obsolete_specific_shm_charts(char *type, int update_every) +{ + ebpf_write_chart_obsolete(type, NETDATA_SHMGET_CHART, + "Calls to syscall <code>shmget(2)</code>.", + EBPF_COMMON_DIMENSION_CALL, + NETDATA_APPS_IPC_SHM_GROUP, + NETDATA_EBPF_CHART_TYPE_LINE, NETDATA_CGROUP_SHM_GET_CONTEXT, + NETDATA_CHART_PRIO_CGROUPS_CONTAINERS + 5800, update_every); + + ebpf_write_chart_obsolete(type, NETDATA_SHMAT_CHART, + "Calls to syscall <code>shmat(2)</code>.", + EBPF_COMMON_DIMENSION_CALL, + NETDATA_APPS_IPC_SHM_GROUP, + NETDATA_EBPF_CHART_TYPE_LINE, NETDATA_CGROUP_SHM_AT_CONTEXT, + NETDATA_CHART_PRIO_CGROUPS_CONTAINERS + 5801, update_every); + + ebpf_write_chart_obsolete(type, NETDATA_SHMDT_CHART, + "Calls to syscall <code>shmdt(2)</code>.", + EBPF_COMMON_DIMENSION_CALL, + NETDATA_APPS_IPC_SHM_GROUP, + NETDATA_EBPF_CHART_TYPE_LINE, NETDATA_CGROUP_SHM_DT_CONTEXT, + NETDATA_CHART_PRIO_CGROUPS_CONTAINERS + 5802, update_every); + + ebpf_write_chart_obsolete(type, NETDATA_SHMCTL_CHART, + "Calls to syscall <code>shmctl(2)</code>.", + EBPF_COMMON_DIMENSION_CALL, + NETDATA_APPS_IPC_SHM_GROUP, + NETDATA_EBPF_CHART_TYPE_LINE, NETDATA_CGROUP_SHM_CTL_CONTEXT, + NETDATA_CHART_PRIO_CGROUPS_CONTAINERS + 5803, update_every); +} + +/** + * Create Systemd Swap Charts + * + * Create charts when systemd is enabled + * + * @param update_every value to overwrite the update frequency set by the server. + **/ +static void ebpf_create_systemd_shm_charts(int update_every) +{ + ebpf_create_charts_on_systemd(NETDATA_SHMGET_CHART, + "Calls to syscall <code>shmget(2)</code>.", + EBPF_COMMON_DIMENSION_CALL, + NETDATA_APPS_IPC_SHM_GROUP, + NETDATA_EBPF_CHART_TYPE_STACKED, + 20191, + ebpf_algorithms[NETDATA_EBPF_INCREMENTAL_IDX], + NETDATA_SYSTEMD_SHM_GET_CONTEXT, NETDATA_EBPF_MODULE_NAME_SHM, update_every); + + ebpf_create_charts_on_systemd(NETDATA_SHMAT_CHART, + "Calls to syscall <code>shmat(2)</code>.", + EBPF_COMMON_DIMENSION_CALL, + NETDATA_APPS_IPC_SHM_GROUP, + NETDATA_EBPF_CHART_TYPE_STACKED, + 20192, + ebpf_algorithms[NETDATA_EBPF_INCREMENTAL_IDX], + NETDATA_SYSTEMD_SHM_AT_CONTEXT, NETDATA_EBPF_MODULE_NAME_SHM, update_every); + + ebpf_create_charts_on_systemd(NETDATA_SHMDT_CHART, + "Calls to syscall <code>shmdt(2)</code>.", + EBPF_COMMON_DIMENSION_CALL, + NETDATA_APPS_IPC_SHM_GROUP, + NETDATA_EBPF_CHART_TYPE_STACKED, + 20193, + ebpf_algorithms[NETDATA_EBPF_INCREMENTAL_IDX], + NETDATA_SYSTEMD_SHM_DT_CONTEXT, NETDATA_EBPF_MODULE_NAME_SHM, update_every); + + ebpf_create_charts_on_systemd(NETDATA_SHMCTL_CHART, + "Calls to syscall <code>shmctl(2)</code>.", + EBPF_COMMON_DIMENSION_CALL, + NETDATA_APPS_IPC_SHM_GROUP, + NETDATA_EBPF_CHART_TYPE_STACKED, + 20193, + ebpf_algorithms[NETDATA_EBPF_INCREMENTAL_IDX], + NETDATA_SYSTEMD_SHM_CTL_CONTEXT, NETDATA_EBPF_MODULE_NAME_SHM, update_every); +} + +/** + * Send Systemd charts + * + * Send collected data to Netdata. + */ +static void ebpf_send_systemd_shm_charts() +{ + ebpf_cgroup_target_t *ect; + write_begin_chart(NETDATA_SERVICE_FAMILY, NETDATA_SHMGET_CHART); + for (ect = ebpf_cgroup_pids; ect; ect = ect->next) { + if (unlikely(ect->systemd) && unlikely(ect->updated)) { + write_chart_dimension(ect->name, (long long)ect->publish_shm.get); + } + } + write_end_chart(); + + write_begin_chart(NETDATA_SERVICE_FAMILY, NETDATA_SHMAT_CHART); + for (ect = ebpf_cgroup_pids; ect; ect = ect->next) { + if (unlikely(ect->systemd) && unlikely(ect->updated)) { + write_chart_dimension(ect->name, (long long)ect->publish_shm.at); + } + } + write_end_chart(); + + write_begin_chart(NETDATA_SERVICE_FAMILY, NETDATA_SHMDT_CHART); + for (ect = ebpf_cgroup_pids; ect; ect = ect->next) { + if (unlikely(ect->systemd) && unlikely(ect->updated)) { + write_chart_dimension(ect->name, (long long)ect->publish_shm.dt); + } + } + write_end_chart(); + + write_begin_chart(NETDATA_SERVICE_FAMILY, NETDATA_SHMCTL_CHART); + for (ect = ebpf_cgroup_pids; ect; ect = ect->next) { + if (unlikely(ect->systemd) && unlikely(ect->updated)) { + write_chart_dimension(ect->name, (long long)ect->publish_shm.ctl); + } + } + write_end_chart(); +} + +/* + * Send Specific Shared memory data + * + * Send data for specific cgroup/apps. + * + * @param type chart type + * @param values structure with values that will be sent to netdata + */ +static void ebpf_send_specific_shm_data(char *type, netdata_publish_shm_t *values) +{ + write_begin_chart(type, NETDATA_SHMGET_CHART); + write_chart_dimension(shm_publish_aggregated[NETDATA_KEY_SHMGET_CALL].name, (long long)values->get); + write_end_chart(); + + write_begin_chart(type, NETDATA_SHMAT_CHART); + write_chart_dimension(shm_publish_aggregated[NETDATA_KEY_SHMAT_CALL].name, (long long)values->at); + write_end_chart(); + + write_begin_chart(type, NETDATA_SHMDT_CHART); + write_chart_dimension(shm_publish_aggregated[NETDATA_KEY_SHMDT_CALL].name, (long long)values->dt); + write_end_chart(); + + write_begin_chart(type, NETDATA_SHMCTL_CHART); + write_chart_dimension(shm_publish_aggregated[NETDATA_KEY_SHMCTL_CALL].name, (long long)values->ctl); + write_end_chart(); +} + +/** + * Send data to Netdata calling auxiliary functions. + * + * @param update_every value to overwrite the update frequency set by the server. +*/ +void ebpf_shm_send_cgroup_data(int update_every) +{ + if (!ebpf_cgroup_pids) + return; + + pthread_mutex_lock(&mutex_cgroup_shm); + ebpf_cgroup_target_t *ect; + for (ect = ebpf_cgroup_pids; ect ; ect = ect->next) { + ebpf_shm_sum_cgroup_pids(&ect->publish_shm, ect->pids); + } + + int has_systemd = shm_ebpf_cgroup.header->systemd_enabled; + if (has_systemd) { + if (send_cgroup_chart) { + ebpf_create_systemd_shm_charts(update_every); + } + + ebpf_send_systemd_shm_charts(); + } + + for (ect = ebpf_cgroup_pids; ect ; ect = ect->next) { + if (ect->systemd) + continue; + + if (!(ect->flags & NETDATA_EBPF_CGROUP_HAS_SHM_CHART) && ect->updated) { + ebpf_create_specific_shm_charts(ect->name, update_every); + ect->flags |= NETDATA_EBPF_CGROUP_HAS_SHM_CHART; + } + + if (ect->flags & NETDATA_EBPF_CGROUP_HAS_SHM_CHART) { + if (ect->updated) { + ebpf_send_specific_shm_data(ect->name, &ect->publish_shm); + } else { + ebpf_obsolete_specific_shm_charts(ect->name, update_every); + ect->flags &= ~NETDATA_EBPF_CGROUP_HAS_SWAP_CHART; + } + } + } + + pthread_mutex_unlock(&mutex_cgroup_shm); +} + +/** +* Main loop for this collector. +*/ +static void shm_collector(ebpf_module_t *em) +{ + shm_threads.thread = mallocz(sizeof(netdata_thread_t)); + shm_threads.start_routine = ebpf_shm_read_hash; + + netdata_thread_create( + shm_threads.thread, + shm_threads.name, + NETDATA_THREAD_OPTION_DEFAULT, + ebpf_shm_read_hash, + em + ); + + int cgroups = em->cgroup_charts; + int update_every = em->update_every; + heartbeat_t hb; + heartbeat_init(&hb); + usec_t step = update_every * USEC_PER_SEC; + while (!ebpf_exit_plugin) { + (void)heartbeat_next(&hb, step); + if (ebpf_exit_plugin) + break; + + netdata_apps_integration_flags_t apps = em->apps_charts; + pthread_mutex_lock(&collect_data_mutex); + if (apps) { + read_apps_table(); + } + + if (cgroups) { + ebpf_update_shm_cgroup(); + } + + pthread_mutex_lock(&lock); + + shm_send_global(); + + if (apps & NETDATA_EBPF_APPS_FLAG_CHART_CREATED) { + ebpf_shm_send_apps_data(apps_groups_root_target); + } + + if (cgroups) { + ebpf_shm_send_cgroup_data(update_every); + } + + pthread_mutex_unlock(&lock); + pthread_mutex_unlock(&collect_data_mutex); + } +} + +/***************************************************************** + * INITIALIZE THREAD + *****************************************************************/ + +/** + * Create apps charts + * + * Call ebpf_create_chart to create the charts on apps submenu. + * + * @param em a pointer to the structure with the default values. + */ +void ebpf_shm_create_apps_charts(struct ebpf_module *em, void *ptr) +{ + struct target *root = ptr; + ebpf_create_charts_on_apps(NETDATA_SHMGET_CHART, + "Calls to syscall <code>shmget(2)</code>.", + EBPF_COMMON_DIMENSION_CALL, + NETDATA_APPS_IPC_SHM_GROUP, + NETDATA_EBPF_CHART_TYPE_STACKED, + 20191, + ebpf_algorithms[NETDATA_EBPF_INCREMENTAL_IDX], + root, em->update_every, NETDATA_EBPF_MODULE_NAME_SHM); + + ebpf_create_charts_on_apps(NETDATA_SHMAT_CHART, + "Calls to syscall <code>shmat(2)</code>.", + EBPF_COMMON_DIMENSION_CALL, + NETDATA_APPS_IPC_SHM_GROUP, + NETDATA_EBPF_CHART_TYPE_STACKED, + 20192, + ebpf_algorithms[NETDATA_EBPF_INCREMENTAL_IDX], + root, em->update_every, NETDATA_EBPF_MODULE_NAME_SHM); + + ebpf_create_charts_on_apps(NETDATA_SHMDT_CHART, + "Calls to syscall <code>shmdt(2)</code>.", + EBPF_COMMON_DIMENSION_CALL, + NETDATA_APPS_IPC_SHM_GROUP, + NETDATA_EBPF_CHART_TYPE_STACKED, + 20193, + ebpf_algorithms[NETDATA_EBPF_INCREMENTAL_IDX], + root, em->update_every, NETDATA_EBPF_MODULE_NAME_SHM); + + ebpf_create_charts_on_apps(NETDATA_SHMCTL_CHART, + "Calls to syscall <code>shmctl(2)</code>.", + EBPF_COMMON_DIMENSION_CALL, + NETDATA_APPS_IPC_SHM_GROUP, + NETDATA_EBPF_CHART_TYPE_STACKED, + 20194, + ebpf_algorithms[NETDATA_EBPF_INCREMENTAL_IDX], + root, em->update_every, NETDATA_EBPF_MODULE_NAME_SHM); + + em->apps_charts |= NETDATA_EBPF_APPS_FLAG_CHART_CREATED; +} + +/** + * Allocate vectors used with this thread. + * + * We are not testing the return, because callocz does this and shutdown the software + * case it was not possible to allocate. + * + * @param apps is apps enabled? + */ +static void ebpf_shm_allocate_global_vectors(int apps) +{ + if (apps) + shm_pid = callocz((size_t)pid_max, sizeof(netdata_publish_shm_t *)); + + shm_vector = callocz((size_t)ebpf_nprocs, sizeof(netdata_publish_shm_t)); + + shm_values = callocz((size_t)ebpf_nprocs, sizeof(netdata_idx_t)); + + memset(shm_hash_values, 0, sizeof(shm_hash_values)); +} + +/***************************************************************** + * MAIN THREAD + *****************************************************************/ + +/** + * Create global charts + * + * Call ebpf_create_chart to create the charts for the collector. + * + * @param update_every value to overwrite the update frequency set by the server. + */ +static void ebpf_create_shm_charts(int update_every) +{ + ebpf_create_chart( + NETDATA_EBPF_SYSTEM_GROUP, + NETDATA_SHM_GLOBAL_CHART, + "Calls to shared memory system calls", + EBPF_COMMON_DIMENSION_CALL, + NETDATA_SYSTEM_IPC_SHM_SUBMENU, + NULL, + NETDATA_EBPF_CHART_TYPE_LINE, + NETDATA_CHART_PRIO_SYSTEM_IPC_SHARED_MEM_CALLS, + ebpf_create_global_dimension, + shm_publish_aggregated, + NETDATA_SHM_END, + update_every, NETDATA_EBPF_MODULE_NAME_SHM + ); + + fflush(stdout); +} + +/* + * Load BPF + * + * Load BPF files. + * + * @param em the structure with configuration + */ +static int ebpf_shm_load_bpf(ebpf_module_t *em) +{ + int ret = 0; + + ebpf_adjust_apps_cgroup(em, em->targets[NETDATA_KEY_SHMGET_CALL].mode); + if (em->load & EBPF_LOAD_LEGACY) { + em->probe_links = ebpf_load_program(ebpf_plugin_dir, em, running_on_kernel, isrh, &em->objects); + if (!em->probe_links) { + em->enabled = CONFIG_BOOLEAN_NO; + ret = -1; + } + } +#ifdef LIBBPF_MAJOR_VERSION + else { + bpf_obj = shm_bpf__open(); + if (!bpf_obj) + ret = -1; + else + ret = ebpf_shm_load_and_attach(bpf_obj, em); + } +#endif + + + if (ret) + error("%s %s", EBPF_DEFAULT_ERROR_MSG, em->thread_name); + + return ret; +} + +/** + * Shared memory thread. + * + * @param ptr a pointer to `struct ebpf_module` + * @return It always return NULL + */ +void *ebpf_shm_thread(void *ptr) +{ + netdata_thread_cleanup_push(ebpf_shm_exit, ptr); + + ebpf_module_t *em = (ebpf_module_t *)ptr; + em->maps = shm_maps; + + ebpf_update_pid_table(&shm_maps[NETDATA_PID_SHM_TABLE], em); + +#ifdef LIBBPF_MAJOR_VERSION + ebpf_adjust_thread_load(em, default_btf); +#endif + if (ebpf_shm_load_bpf(em)) { + em->thread->enabled = NETDATA_THREAD_EBPF_STOPPED; + goto endshm; + } + + ebpf_shm_allocate_global_vectors(em->apps_charts); + + int algorithms[NETDATA_SHM_END] = { + NETDATA_EBPF_INCREMENTAL_IDX, + NETDATA_EBPF_INCREMENTAL_IDX, + NETDATA_EBPF_INCREMENTAL_IDX, + NETDATA_EBPF_INCREMENTAL_IDX + }; + ebpf_global_labels( + shm_aggregated_data, + shm_publish_aggregated, + shm_dimension_name, + shm_dimension_name, + algorithms, + NETDATA_SHM_END + ); + + pthread_mutex_lock(&lock); + ebpf_create_shm_charts(em->update_every); + ebpf_update_stats(&plugin_statistics, em); + pthread_mutex_unlock(&lock); + + shm_collector(em); + +endshm: + ebpf_update_disabled_plugin_stats(em); + + netdata_thread_cleanup_pop(1); + return NULL; +} diff --git a/collectors/ebpf.plugin/ebpf_shm.h b/collectors/ebpf.plugin/ebpf_shm.h new file mode 100644 index 0000000..4e06881 --- /dev/null +++ b/collectors/ebpf.plugin/ebpf_shm.h @@ -0,0 +1,63 @@ +// SPDX-License-Identifier: GPL-3.0-or-later + +#ifndef NETDATA_EBPF_SHM_H +#define NETDATA_EBPF_SHM_H 1 + +// Module name +#define NETDATA_EBPF_MODULE_NAME_SHM "shm" + +#define NETDATA_SHM_SLEEP_MS 850000ULL + +// charts +#define NETDATA_SHM_GLOBAL_CHART "shared_memory_calls" +#define NETDATA_SHMGET_CHART "shmget_call" +#define NETDATA_SHMAT_CHART "shmat_call" +#define NETDATA_SHMDT_CHART "shmdt_call" +#define NETDATA_SHMCTL_CHART "shmctl_call" + +// configuration file +#define NETDATA_DIRECTORY_SHM_CONFIG_FILE "shm.conf" + +// Contexts +#define NETDATA_CGROUP_SHM_GET_CONTEXT "cgroup.shmget" +#define NETDATA_CGROUP_SHM_AT_CONTEXT "cgroup.shmat" +#define NETDATA_CGROUP_SHM_DT_CONTEXT "cgroup.shmdt" +#define NETDATA_CGROUP_SHM_CTL_CONTEXT "cgroup.shmctl" + +#define NETDATA_SYSTEMD_SHM_GET_CONTEXT "services.shmget" +#define NETDATA_SYSTEMD_SHM_AT_CONTEXT "services.shmat" +#define NETDATA_SYSTEMD_SHM_DT_CONTEXT "services.shmdt" +#define NETDATA_SYSTEMD_SHM_CTL_CONTEXT "services.shmctl" + +typedef struct netdata_publish_shm { + uint64_t get; + uint64_t at; + uint64_t dt; + uint64_t ctl; +} netdata_publish_shm_t; + +enum shm_tables { + NETDATA_PID_SHM_TABLE, + NETDATA_SHM_CONTROLLER, + NETDATA_SHM_GLOBAL_TABLE +}; + +enum shm_counters { + NETDATA_KEY_SHMGET_CALL, + NETDATA_KEY_SHMAT_CALL, + NETDATA_KEY_SHMDT_CALL, + NETDATA_KEY_SHMCTL_CALL, + + // Keep this as last and don't skip numbers as it is used as element counter + NETDATA_SHM_END +}; + +extern netdata_publish_shm_t **shm_pid; + +void *ebpf_shm_thread(void *ptr); +void ebpf_shm_create_apps_charts(struct ebpf_module *em, void *ptr); +extern netdata_ebpf_targets_t shm_targets[]; + +extern struct config shm_config; + +#endif diff --git a/collectors/ebpf.plugin/ebpf_socket.c b/collectors/ebpf.plugin/ebpf_socket.c new file mode 100644 index 0000000..3a023e4 --- /dev/null +++ b/collectors/ebpf.plugin/ebpf_socket.c @@ -0,0 +1,3976 @@ +// SPDX-License-Identifier: GPL-3.0-or-later + +#include <sys/resource.h> + +#include "ebpf.h" +#include "ebpf_socket.h" + +/***************************************************************** + * + * GLOBAL VARIABLES + * + *****************************************************************/ + +static char *socket_dimension_names[NETDATA_MAX_SOCKET_VECTOR] = { "received", "sent", "close", + "received", "sent", "retransmitted", + "connected_V4", "connected_V6", "connected_tcp", + "connected_udp"}; +static char *socket_id_names[NETDATA_MAX_SOCKET_VECTOR] = { "tcp_cleanup_rbuf", "tcp_sendmsg", "tcp_close", + "udp_recvmsg", "udp_sendmsg", "tcp_retransmit_skb", + "tcp_connect_v4", "tcp_connect_v6", "inet_csk_accept_tcp", + "inet_csk_accept_udp" }; + +static ebpf_local_maps_t socket_maps[] = {{.name = "tbl_bandwidth", + .internal_input = NETDATA_COMPILED_CONNECTIONS_ALLOWED, + .user_input = NETDATA_MAXIMUM_CONNECTIONS_ALLOWED, + .type = NETDATA_EBPF_MAP_RESIZABLE | NETDATA_EBPF_MAP_PID, + .map_fd = ND_EBPF_MAP_FD_NOT_INITIALIZED}, + {.name = "tbl_global_sock", + .internal_input = NETDATA_SOCKET_COUNTER, + .user_input = 0, .type = NETDATA_EBPF_MAP_STATIC, + .map_fd = ND_EBPF_MAP_FD_NOT_INITIALIZED}, + {.name = "tbl_lports", + .internal_input = NETDATA_SOCKET_COUNTER, + .user_input = 0, .type = NETDATA_EBPF_MAP_STATIC, + .map_fd = ND_EBPF_MAP_FD_NOT_INITIALIZED}, + {.name = "tbl_conn_ipv4", + .internal_input = NETDATA_COMPILED_CONNECTIONS_ALLOWED, + .user_input = NETDATA_MAXIMUM_CONNECTIONS_ALLOWED, + .type = NETDATA_EBPF_MAP_STATIC, + .map_fd = ND_EBPF_MAP_FD_NOT_INITIALIZED}, + {.name = "tbl_conn_ipv6", + .internal_input = NETDATA_COMPILED_CONNECTIONS_ALLOWED, + .user_input = NETDATA_MAXIMUM_CONNECTIONS_ALLOWED, + .type = NETDATA_EBPF_MAP_STATIC, + .map_fd = ND_EBPF_MAP_FD_NOT_INITIALIZED}, + {.name = "tbl_nv_udp", + .internal_input = NETDATA_COMPILED_UDP_CONNECTIONS_ALLOWED, + .user_input = NETDATA_MAXIMUM_UDP_CONNECTIONS_ALLOWED, + .type = NETDATA_EBPF_MAP_STATIC, + .map_fd = ND_EBPF_MAP_FD_NOT_INITIALIZED}, + {.name = "socket_ctrl", .internal_input = NETDATA_CONTROLLER_END, + .user_input = 0, + .type = NETDATA_EBPF_MAP_CONTROLLER, + .map_fd = ND_EBPF_MAP_FD_NOT_INITIALIZED}, + {.name = NULL, .internal_input = 0, .user_input = 0}}; + +static netdata_idx_t *socket_hash_values = NULL; +static netdata_syscall_stat_t socket_aggregated_data[NETDATA_MAX_SOCKET_VECTOR]; +static netdata_publish_syscall_t socket_publish_aggregated[NETDATA_MAX_SOCKET_VECTOR]; + +ebpf_socket_publish_apps_t **socket_bandwidth_curr = NULL; +static ebpf_bandwidth_t *bandwidth_vector = NULL; + +pthread_mutex_t nv_mutex; +int wait_to_plot = 0; + +netdata_vector_plot_t inbound_vectors = { .plot = NULL, .next = 0, .last = 0 }; +netdata_vector_plot_t outbound_vectors = { .plot = NULL, .next = 0, .last = 0 }; +netdata_socket_t *socket_values; + +ebpf_network_viewer_port_list_t *listen_ports = NULL; + +struct config socket_config = { .first_section = NULL, + .last_section = NULL, + .mutex = NETDATA_MUTEX_INITIALIZER, + .index = { .avl_tree = { .root = NULL, .compar = appconfig_section_compare }, + .rwlock = AVL_LOCK_INITIALIZER } }; + +netdata_ebpf_targets_t socket_targets[] = { {.name = "inet_csk_accept", .mode = EBPF_LOAD_TRAMPOLINE}, + {.name = "tcp_retransmit_skb", .mode = EBPF_LOAD_TRAMPOLINE}, + {.name = "tcp_cleanup_rbuf", .mode = EBPF_LOAD_TRAMPOLINE}, + {.name = "tcp_close", .mode = EBPF_LOAD_TRAMPOLINE}, + {.name = "udp_recvmsg", .mode = EBPF_LOAD_TRAMPOLINE}, + {.name = "tcp_sendmsg", .mode = EBPF_LOAD_TRAMPOLINE}, + {.name = "udp_sendmsg", .mode = EBPF_LOAD_TRAMPOLINE}, + {.name = "tcp_v4_connect", .mode = EBPF_LOAD_TRAMPOLINE}, + {.name = "tcp_v6_connect", .mode = EBPF_LOAD_TRAMPOLINE}, + {.name = NULL, .mode = EBPF_LOAD_TRAMPOLINE}}; + +struct netdata_static_thread socket_threads = { + .name = "EBPF SOCKET READ", + .config_section = NULL, + .config_name = NULL, + .env_name = NULL, + .enabled = 1, + .thread = NULL, + .init_routine = NULL, + .start_routine = NULL +}; + +#ifdef LIBBPF_MAJOR_VERSION +#include "includes/socket.skel.h" // BTF code + +static struct socket_bpf *bpf_obj = NULL; + +/** + * Disable Probe + * + * Disable probes to use trampoline. + * + * @param obj is the main structure for bpf objects. + */ +static void ebpf_socket_disable_probes(struct socket_bpf *obj) +{ + bpf_program__set_autoload(obj->progs.netdata_inet_csk_accept_kretprobe, false); + bpf_program__set_autoload(obj->progs.netdata_tcp_v4_connect_kretprobe, false); + bpf_program__set_autoload(obj->progs.netdata_tcp_v6_connect_kretprobe, false); + bpf_program__set_autoload(obj->progs.netdata_tcp_retransmit_skb_kprobe, false); + bpf_program__set_autoload(obj->progs.netdata_tcp_cleanup_rbuf_kprobe, false); + bpf_program__set_autoload(obj->progs.netdata_tcp_close_kprobe, false); + bpf_program__set_autoload(obj->progs.netdata_udp_recvmsg_kprobe, false); + bpf_program__set_autoload(obj->progs.netdata_udp_recvmsg_kretprobe, false); + bpf_program__set_autoload(obj->progs.netdata_tcp_sendmsg_kretprobe, false); + bpf_program__set_autoload(obj->progs.netdata_tcp_sendmsg_kprobe, false); + bpf_program__set_autoload(obj->progs.netdata_udp_sendmsg_kretprobe, false); + bpf_program__set_autoload(obj->progs.netdata_udp_sendmsg_kprobe, false); + bpf_program__set_autoload(obj->progs.netdata_socket_release_task_kprobe, false); +} + +/** + * Disable Trampoline + * + * Disable trampoline to use probes. + * + * @param obj is the main structure for bpf objects. + */ +static void ebpf_socket_disable_trampoline(struct socket_bpf *obj) +{ + bpf_program__set_autoload(obj->progs.netdata_inet_csk_accept_fentry, false); + bpf_program__set_autoload(obj->progs.netdata_tcp_v4_connect_fexit, false); + bpf_program__set_autoload(obj->progs.netdata_tcp_v6_connect_fexit, false); + bpf_program__set_autoload(obj->progs.netdata_tcp_retransmit_skb_fentry, false); + bpf_program__set_autoload(obj->progs.netdata_tcp_cleanup_rbuf_fentry, false); + bpf_program__set_autoload(obj->progs.netdata_tcp_close_fentry, false); + bpf_program__set_autoload(obj->progs.netdata_udp_recvmsg_fentry, false); + bpf_program__set_autoload(obj->progs.netdata_udp_recvmsg_fexit, false); + bpf_program__set_autoload(obj->progs.netdata_tcp_sendmsg_fentry, false); + bpf_program__set_autoload(obj->progs.netdata_tcp_sendmsg_fexit, false); + bpf_program__set_autoload(obj->progs.netdata_udp_sendmsg_fentry, false); + bpf_program__set_autoload(obj->progs.netdata_udp_sendmsg_fexit, false); + bpf_program__set_autoload(obj->progs.netdata_socket_release_task_fentry, false); +} + +/** + * Set trampoline target. + * + * @param obj is the main structure for bpf objects. + */ +static void ebpf_set_trampoline_target(struct socket_bpf *obj) +{ + bpf_program__set_attach_target(obj->progs.netdata_inet_csk_accept_fentry, 0, + socket_targets[NETDATA_FCNT_INET_CSK_ACCEPT].name); + + bpf_program__set_attach_target(obj->progs.netdata_tcp_v4_connect_fexit, 0, + socket_targets[NETDATA_FCNT_TCP_V4_CONNECT].name); + + bpf_program__set_attach_target(obj->progs.netdata_tcp_v6_connect_fexit, 0, + socket_targets[NETDATA_FCNT_TCP_V6_CONNECT].name); + + bpf_program__set_attach_target(obj->progs.netdata_tcp_retransmit_skb_fentry, 0, + socket_targets[NETDATA_FCNT_TCP_RETRANSMIT].name); + + bpf_program__set_attach_target(obj->progs.netdata_tcp_cleanup_rbuf_fentry, 0, + socket_targets[NETDATA_FCNT_CLEANUP_RBUF].name); + + bpf_program__set_attach_target(obj->progs.netdata_tcp_close_fentry, 0, socket_targets[NETDATA_FCNT_TCP_CLOSE].name); + + bpf_program__set_attach_target(obj->progs.netdata_udp_recvmsg_fentry, 0, + socket_targets[NETDATA_FCNT_UDP_RECEVMSG].name); + + bpf_program__set_attach_target(obj->progs.netdata_udp_recvmsg_fexit, 0, + socket_targets[NETDATA_FCNT_UDP_RECEVMSG].name); + + bpf_program__set_attach_target(obj->progs.netdata_tcp_sendmsg_fentry, 0, + socket_targets[NETDATA_FCNT_TCP_SENDMSG].name); + + bpf_program__set_attach_target(obj->progs.netdata_tcp_sendmsg_fexit, 0, + socket_targets[NETDATA_FCNT_TCP_SENDMSG].name); + + bpf_program__set_attach_target(obj->progs.netdata_udp_sendmsg_fentry, 0, + socket_targets[NETDATA_FCNT_UDP_SENDMSG].name); + + bpf_program__set_attach_target(obj->progs.netdata_udp_sendmsg_fexit, 0, + socket_targets[NETDATA_FCNT_UDP_SENDMSG].name); + + bpf_program__set_attach_target(obj->progs.netdata_socket_release_task_fentry, 0, EBPF_COMMON_FNCT_CLEAN_UP); +} + + +/** + * Disable specific trampoline + * + * Disable specific trampoline to match user selection. + * + * @param obj is the main structure for bpf objects. + * @param sel option selected by user. + */ +static inline void ebpf_socket_disable_specific_trampoline(struct socket_bpf *obj, netdata_run_mode_t sel) +{ + if (sel == MODE_RETURN) { + bpf_program__set_autoload(obj->progs.netdata_tcp_sendmsg_fentry, false); + bpf_program__set_autoload(obj->progs.netdata_udp_sendmsg_fentry, false); + } else { + bpf_program__set_autoload(obj->progs.netdata_tcp_sendmsg_fexit, false); + bpf_program__set_autoload(obj->progs.netdata_udp_sendmsg_fexit, false); + } +} + +/** + * Disable specific probe + * + * Disable specific probe to match user selection. + * + * @param obj is the main structure for bpf objects. + * @param sel option selected by user. + */ +static inline void ebpf_socket_disable_specific_probe(struct socket_bpf *obj, netdata_run_mode_t sel) +{ + if (sel == MODE_RETURN) { + bpf_program__set_autoload(obj->progs.netdata_tcp_sendmsg_kprobe, false); + bpf_program__set_autoload(obj->progs.netdata_udp_sendmsg_kprobe, false); + } else { + bpf_program__set_autoload(obj->progs.netdata_tcp_sendmsg_kretprobe, false); + bpf_program__set_autoload(obj->progs.netdata_udp_sendmsg_kretprobe, false); + } +} + +/** + * Attach probes + * + * Attach probes to targets. + * + * @param obj is the main structure for bpf objects. + * @param sel option selected by user. + */ +static int ebpf_socket_attach_probes(struct socket_bpf *obj, netdata_run_mode_t sel) +{ + obj->links.netdata_inet_csk_accept_kretprobe = bpf_program__attach_kprobe(obj->progs.netdata_inet_csk_accept_kretprobe, + true, + socket_targets[NETDATA_FCNT_INET_CSK_ACCEPT].name); + int ret = libbpf_get_error(obj->links.netdata_inet_csk_accept_kretprobe); + if (ret) + return -1; + + obj->links.netdata_tcp_v4_connect_kretprobe = bpf_program__attach_kprobe(obj->progs.netdata_tcp_v4_connect_kretprobe, + true, + socket_targets[NETDATA_FCNT_TCP_V4_CONNECT].name); + ret = libbpf_get_error(obj->links.netdata_tcp_v4_connect_kretprobe); + if (ret) + return -1; + + obj->links.netdata_tcp_v6_connect_kretprobe = bpf_program__attach_kprobe(obj->progs.netdata_tcp_v6_connect_kretprobe, + true, + socket_targets[NETDATA_FCNT_TCP_V6_CONNECT].name); + ret = libbpf_get_error(obj->links.netdata_tcp_v6_connect_kretprobe); + if (ret) + return -1; + + obj->links.netdata_tcp_retransmit_skb_kprobe = bpf_program__attach_kprobe(obj->progs.netdata_tcp_retransmit_skb_kprobe, + false, + socket_targets[NETDATA_FCNT_TCP_RETRANSMIT].name); + ret = libbpf_get_error(obj->links.netdata_tcp_retransmit_skb_kprobe); + if (ret) + return -1; + + obj->links.netdata_tcp_cleanup_rbuf_kprobe = bpf_program__attach_kprobe(obj->progs.netdata_tcp_cleanup_rbuf_kprobe, + false, + socket_targets[NETDATA_FCNT_CLEANUP_RBUF].name); + ret = libbpf_get_error(obj->links.netdata_tcp_cleanup_rbuf_kprobe); + if (ret) + return -1; + + obj->links.netdata_tcp_close_kprobe = bpf_program__attach_kprobe(obj->progs.netdata_tcp_close_kprobe, + false, + socket_targets[NETDATA_FCNT_TCP_CLOSE].name); + ret = libbpf_get_error(obj->links.netdata_tcp_close_kprobe); + if (ret) + return -1; + + obj->links.netdata_udp_recvmsg_kprobe = bpf_program__attach_kprobe(obj->progs.netdata_udp_recvmsg_kprobe, + false, + socket_targets[NETDATA_FCNT_UDP_RECEVMSG].name); + ret = libbpf_get_error(obj->links.netdata_udp_recvmsg_kprobe); + if (ret) + return -1; + + obj->links.netdata_udp_recvmsg_kretprobe = bpf_program__attach_kprobe(obj->progs.netdata_udp_recvmsg_kretprobe, + true, + socket_targets[NETDATA_FCNT_UDP_RECEVMSG].name); + ret = libbpf_get_error(obj->links.netdata_udp_recvmsg_kretprobe); + if (ret) + return -1; + + if (sel == MODE_RETURN) { + obj->links.netdata_tcp_sendmsg_kretprobe = bpf_program__attach_kprobe(obj->progs.netdata_tcp_sendmsg_kretprobe, + true, + socket_targets[NETDATA_FCNT_TCP_SENDMSG].name); + ret = libbpf_get_error(obj->links.netdata_tcp_sendmsg_kretprobe); + if (ret) + return -1; + + obj->links.netdata_udp_sendmsg_kretprobe = bpf_program__attach_kprobe(obj->progs.netdata_udp_sendmsg_kretprobe, + true, + socket_targets[NETDATA_FCNT_UDP_SENDMSG].name); + ret = libbpf_get_error(obj->links.netdata_udp_sendmsg_kretprobe); + if (ret) + return -1; + } else { + obj->links.netdata_tcp_sendmsg_kprobe = bpf_program__attach_kprobe(obj->progs.netdata_tcp_sendmsg_kprobe, + false, + socket_targets[NETDATA_FCNT_TCP_SENDMSG].name); + ret = libbpf_get_error(obj->links.netdata_tcp_sendmsg_kprobe); + if (ret) + return -1; + + obj->links.netdata_udp_sendmsg_kprobe = bpf_program__attach_kprobe(obj->progs.netdata_udp_sendmsg_kprobe, + false, + socket_targets[NETDATA_FCNT_UDP_SENDMSG].name); + ret = libbpf_get_error(obj->links.netdata_udp_sendmsg_kprobe); + if (ret) + return -1; + } + + obj->links.netdata_socket_release_task_kprobe = bpf_program__attach_kprobe(obj->progs.netdata_socket_release_task_kprobe, + false, EBPF_COMMON_FNCT_CLEAN_UP); + ret = libbpf_get_error(obj->links.netdata_socket_release_task_kprobe); + if (ret) + return -1; + + return 0; +} + +/** + * Set hash tables + * + * Set the values for maps according the value given by kernel. + * + * @param obj is the main structure for bpf objects. + */ +static void ebpf_socket_set_hash_tables(struct socket_bpf *obj) +{ + socket_maps[NETDATA_SOCKET_TABLE_BANDWIDTH].map_fd = bpf_map__fd(obj->maps.tbl_bandwidth); + socket_maps[NETDATA_SOCKET_GLOBAL].map_fd = bpf_map__fd(obj->maps.tbl_global_sock); + socket_maps[NETDATA_SOCKET_LPORTS].map_fd = bpf_map__fd(obj->maps.tbl_lports); + socket_maps[NETDATA_SOCKET_TABLE_IPV4].map_fd = bpf_map__fd(obj->maps.tbl_conn_ipv4); + socket_maps[NETDATA_SOCKET_TABLE_IPV6].map_fd = bpf_map__fd(obj->maps.tbl_conn_ipv6); + socket_maps[NETDATA_SOCKET_TABLE_UDP].map_fd = bpf_map__fd(obj->maps.tbl_nv_udp); + socket_maps[NETDATA_SOCKET_TABLE_CTRL].map_fd = bpf_map__fd(obj->maps.socket_ctrl); +} + +/** + * Adjust Map Size + * + * Resize maps according input from users. + * + * @param obj is the main structure for bpf objects. + * @param em structure with configuration + */ +static void ebpf_socket_adjust_map_size(struct socket_bpf *obj, ebpf_module_t *em) +{ + ebpf_update_map_size(obj->maps.tbl_bandwidth, &socket_maps[NETDATA_SOCKET_TABLE_BANDWIDTH], + em, bpf_map__name(obj->maps.tbl_bandwidth)); + + ebpf_update_map_size(obj->maps.tbl_conn_ipv4, &socket_maps[NETDATA_SOCKET_TABLE_IPV4], + em, bpf_map__name(obj->maps.tbl_conn_ipv4)); + + ebpf_update_map_size(obj->maps.tbl_conn_ipv6, &socket_maps[NETDATA_SOCKET_TABLE_IPV6], + em, bpf_map__name(obj->maps.tbl_conn_ipv6)); + + ebpf_update_map_size(obj->maps.tbl_nv_udp, &socket_maps[NETDATA_SOCKET_TABLE_UDP], + em, bpf_map__name(obj->maps.tbl_nv_udp)); +} + +/** + * Load and attach + * + * Load and attach the eBPF code in kernel. + * + * @param obj is the main structure for bpf objects. + * @param em structure with configuration + * + * @return it returns 0 on succes and -1 otherwise + */ +static inline int ebpf_socket_load_and_attach(struct socket_bpf *obj, ebpf_module_t *em) +{ + netdata_ebpf_targets_t *mt = em->targets; + netdata_ebpf_program_loaded_t test = mt[NETDATA_FCNT_INET_CSK_ACCEPT].mode; + + if (test == EBPF_LOAD_TRAMPOLINE) { + ebpf_socket_disable_probes(obj); + + ebpf_set_trampoline_target(obj); + ebpf_socket_disable_specific_trampoline(obj, em->mode); + } else { // We are not using tracepoints for this thread. + ebpf_socket_disable_trampoline(obj); + + ebpf_socket_disable_specific_probe(obj, em->mode); + } + + int ret = socket_bpf__load(obj); + if (ret) { + fprintf(stderr, "failed to load BPF object: %d\n", ret); + return ret; + } + + ebpf_socket_adjust_map_size(obj, em); + + if (test == EBPF_LOAD_TRAMPOLINE) { + ret = socket_bpf__attach(obj); + } else { + ret = ebpf_socket_attach_probes(obj, em->mode); + } + + if (!ret) { + ebpf_socket_set_hash_tables(obj); + + ebpf_update_controller(socket_maps[NETDATA_SOCKET_TABLE_CTRL].map_fd, em); + } + + return ret; +} +#endif + +/***************************************************************** + * + * FUNCTIONS TO CLOSE THE THREAD + * + *****************************************************************/ + +/** + * Clean internal socket plot + * + * Clean all structures allocated with strdupz. + * + * @param ptr the pointer with addresses to clean. + */ +static inline void clean_internal_socket_plot(netdata_socket_plot_t *ptr) +{ + freez(ptr->dimension_recv); + freez(ptr->dimension_sent); + freez(ptr->resolved_name); + freez(ptr->dimension_retransmit); +} + +/** + * Clean socket plot + * + * Clean the allocated data for inbound and outbound vectors. + */ +static void clean_allocated_socket_plot() +{ + uint32_t i; + uint32_t end = inbound_vectors.last; + netdata_socket_plot_t *plot = inbound_vectors.plot; + for (i = 0; i < end; i++) { + clean_internal_socket_plot(&plot[i]); + } + + clean_internal_socket_plot(&plot[inbound_vectors.last]); + + end = outbound_vectors.last; + plot = outbound_vectors.plot; + for (i = 0; i < end; i++) { + clean_internal_socket_plot(&plot[i]); + } + clean_internal_socket_plot(&plot[outbound_vectors.last]); +} + +/** + * Clean network ports allocated during initialization. + * + * @param ptr a pointer to the link list. + */ +static void clean_network_ports(ebpf_network_viewer_port_list_t *ptr) +{ + if (unlikely(!ptr)) + return; + + while (ptr) { + ebpf_network_viewer_port_list_t *next = ptr->next; + freez(ptr->value); + freez(ptr); + ptr = next; + } +} + +/** + * Clean service names + * + * Clean the allocated link list that stores names. + * + * @param names the link list. + */ +static void clean_service_names(ebpf_network_viewer_dim_name_t *names) +{ + if (unlikely(!names)) + return; + + while (names) { + ebpf_network_viewer_dim_name_t *next = names->next; + freez(names->name); + freez(names); + names = next; + } +} + +/** + * Clean hostnames + * + * @param hostnames the hostnames to clean + */ +static void clean_hostnames(ebpf_network_viewer_hostname_list_t *hostnames) +{ + if (unlikely(!hostnames)) + return; + + while (hostnames) { + ebpf_network_viewer_hostname_list_t *next = hostnames->next; + freez(hostnames->value); + simple_pattern_free(hostnames->value_pattern); + freez(hostnames); + hostnames = next; + } +} + +/** + * Cleanup publish syscall + * + * @param nps list of structures to clean + */ +void ebpf_cleanup_publish_syscall(netdata_publish_syscall_t *nps) +{ + while (nps) { + freez(nps->algorithm); + nps = nps->next; + } +} + +/** + * Clean port Structure + * + * Clean the allocated list. + * + * @param clean the list that will be cleaned + */ +void clean_port_structure(ebpf_network_viewer_port_list_t **clean) +{ + ebpf_network_viewer_port_list_t *move = *clean; + while (move) { + ebpf_network_viewer_port_list_t *next = move->next; + freez(move->value); + freez(move); + + move = next; + } + *clean = NULL; +} + +/** + * Clean IP structure + * + * Clean the allocated list. + * + * @param clean the list that will be cleaned + */ +static void clean_ip_structure(ebpf_network_viewer_ip_list_t **clean) +{ + ebpf_network_viewer_ip_list_t *move = *clean; + while (move) { + ebpf_network_viewer_ip_list_t *next = move->next; + freez(move->value); + freez(move); + + move = next; + } + *clean = NULL; +} + +/** + * Socket Free + * + * Cleanup variables after child threads to stop + * + * @param ptr thread data. + */ +static void ebpf_socket_free(ebpf_module_t *em ) +{ + pthread_mutex_lock(&ebpf_exit_cleanup); + if (em->thread->enabled == NETDATA_THREAD_EBPF_RUNNING) { + em->thread->enabled = NETDATA_THREAD_EBPF_STOPPING; + pthread_mutex_unlock(&ebpf_exit_cleanup); + return; + } + pthread_mutex_unlock(&ebpf_exit_cleanup); + + ebpf_cleanup_publish_syscall(socket_publish_aggregated); + freez(socket_hash_values); + + freez(bandwidth_vector); + + freez(socket_values); + clean_allocated_socket_plot(); + freez(inbound_vectors.plot); + freez(outbound_vectors.plot); + + clean_port_structure(&listen_ports); + + ebpf_modules[EBPF_MODULE_SOCKET_IDX].enabled = 0; + + clean_network_ports(network_viewer_opt.included_port); + clean_network_ports(network_viewer_opt.excluded_port); + clean_service_names(network_viewer_opt.names); + clean_hostnames(network_viewer_opt.included_hostnames); + clean_hostnames(network_viewer_opt.excluded_hostnames); + + pthread_mutex_destroy(&nv_mutex); + + freez(socket_threads.thread); + +#ifdef LIBBPF_MAJOR_VERSION + if (bpf_obj) + socket_bpf__destroy(bpf_obj); +#endif + + pthread_mutex_lock(&ebpf_exit_cleanup); + em->thread->enabled = NETDATA_THREAD_EBPF_STOPPED; + pthread_mutex_unlock(&ebpf_exit_cleanup); +} + +/** + * Socket exit + * + * Clean up the main thread. + * + * @param ptr thread data. + */ +static void ebpf_socket_exit(void *ptr) +{ + ebpf_module_t *em = (ebpf_module_t *)ptr; + netdata_thread_cancel(*socket_threads.thread); + ebpf_socket_free(em); +} + +/** + * Socket cleanup + * + * Clean up allocated addresses. + * + * @param ptr thread data. + */ +void ebpf_socket_cleanup(void *ptr) +{ + ebpf_module_t *em = (ebpf_module_t *)ptr; + ebpf_socket_free(em); +} + +/***************************************************************** + * + * PROCESS DATA AND SEND TO NETDATA + * + *****************************************************************/ + +/** + * Update publish structure before to send data to Netdata. + * + * @param publish the first output structure with independent dimensions + * @param tcp structure to store IO from tcp sockets + * @param udp structure to store IO from udp sockets + * @param input the structure with the input data. + */ +static void ebpf_update_global_publish( + netdata_publish_syscall_t *publish, netdata_publish_vfs_common_t *tcp, netdata_publish_vfs_common_t *udp, + netdata_syscall_stat_t *input) +{ + netdata_publish_syscall_t *move = publish; + while (move) { + if (input->call != move->pcall) { + // This condition happens to avoid initial values with dimensions higher than normal values. + if (move->pcall) { + move->ncall = (input->call > move->pcall) ? input->call - move->pcall : move->pcall - input->call; + move->nbyte = (input->bytes > move->pbyte) ? input->bytes - move->pbyte : move->pbyte - input->bytes; + move->nerr = (input->ecall > move->nerr) ? input->ecall - move->perr : move->perr - input->ecall; + } else { + move->ncall = 0; + move->nbyte = 0; + move->nerr = 0; + } + + move->pcall = input->call; + move->pbyte = input->bytes; + move->perr = input->ecall; + } else { + move->ncall = 0; + move->nbyte = 0; + move->nerr = 0; + } + + input = input->next; + move = move->next; + } + + tcp->write = -(long)publish[0].nbyte; + tcp->read = (long)publish[1].nbyte; + + udp->write = -(long)publish[3].nbyte; + udp->read = (long)publish[4].nbyte; +} + +/** + * Update Network Viewer plot data + * + * @param plot the structure where the data will be stored + * @param sock the last update from the socket + */ +static inline void update_nv_plot_data(netdata_plot_values_t *plot, netdata_socket_t *sock) +{ + if (sock->ct > plot->last_time) { + plot->last_time = sock->ct; + plot->plot_recv_packets = sock->recv_packets; + plot->plot_sent_packets = sock->sent_packets; + plot->plot_recv_bytes = sock->recv_bytes; + plot->plot_sent_bytes = sock->sent_bytes; + plot->plot_retransmit = sock->retransmit; + } + + sock->recv_packets = 0; + sock->sent_packets = 0; + sock->recv_bytes = 0; + sock->sent_bytes = 0; + sock->retransmit = 0; +} + +/** + * Calculate Network Viewer Plot + * + * Do math with collected values before to plot data. + */ +static inline void calculate_nv_plot() +{ + uint32_t i; + uint32_t end = inbound_vectors.next; + for (i = 0; i < end; i++) { + update_nv_plot_data(&inbound_vectors.plot[i].plot, &inbound_vectors.plot[i].sock); + } + inbound_vectors.max_plot = end; + + // The 'Other' dimension is always calculated for the chart to have at least one dimension + update_nv_plot_data(&inbound_vectors.plot[inbound_vectors.last].plot, + &inbound_vectors.plot[inbound_vectors.last].sock); + + end = outbound_vectors.next; + for (i = 0; i < end; i++) { + update_nv_plot_data(&outbound_vectors.plot[i].plot, &outbound_vectors.plot[i].sock); + } + outbound_vectors.max_plot = end; + + // The 'Other' dimension is always calculated for the chart to have at least one dimension + update_nv_plot_data(&outbound_vectors.plot[outbound_vectors.last].plot, + &outbound_vectors.plot[outbound_vectors.last].sock); +} + +/** + * Network viewer send bytes + * + * @param ptr the structure with values to plot + * @param chart the chart name. + */ +static inline void ebpf_socket_nv_send_bytes(netdata_vector_plot_t *ptr, char *chart) +{ + uint32_t i; + uint32_t end = ptr->last_plot; + netdata_socket_plot_t *w = ptr->plot; + collected_number value; + + write_begin_chart(NETDATA_EBPF_FAMILY, chart); + for (i = 0; i < end; i++) { + value = ((collected_number) w[i].plot.plot_sent_bytes); + write_chart_dimension(w[i].dimension_sent, value); + value = (collected_number) w[i].plot.plot_recv_bytes; + write_chart_dimension(w[i].dimension_recv, value); + } + + i = ptr->last; + value = ((collected_number) w[i].plot.plot_sent_bytes); + write_chart_dimension(w[i].dimension_sent, value); + value = (collected_number) w[i].plot.plot_recv_bytes; + write_chart_dimension(w[i].dimension_recv, value); + write_end_chart(); +} + +/** + * Network Viewer Send packets + * + * @param ptr the structure with values to plot + * @param chart the chart name. + */ +static inline void ebpf_socket_nv_send_packets(netdata_vector_plot_t *ptr, char *chart) +{ + uint32_t i; + uint32_t end = ptr->last_plot; + netdata_socket_plot_t *w = ptr->plot; + collected_number value; + + write_begin_chart(NETDATA_EBPF_FAMILY, chart); + for (i = 0; i < end; i++) { + value = ((collected_number)w[i].plot.plot_sent_packets); + write_chart_dimension(w[i].dimension_sent, value); + value = (collected_number) w[i].plot.plot_recv_packets; + write_chart_dimension(w[i].dimension_recv, value); + } + + i = ptr->last; + value = ((collected_number)w[i].plot.plot_sent_packets); + write_chart_dimension(w[i].dimension_sent, value); + value = (collected_number)w[i].plot.plot_recv_packets; + write_chart_dimension(w[i].dimension_recv, value); + write_end_chart(); +} + +/** + * Network Viewer Send Retransmit + * + * @param ptr the structure with values to plot + * @param chart the chart name. + */ +static inline void ebpf_socket_nv_send_retransmit(netdata_vector_plot_t *ptr, char *chart) +{ + uint32_t i; + uint32_t end = ptr->last_plot; + netdata_socket_plot_t *w = ptr->plot; + collected_number value; + + write_begin_chart(NETDATA_EBPF_FAMILY, chart); + for (i = 0; i < end; i++) { + value = (collected_number) w[i].plot.plot_retransmit; + write_chart_dimension(w[i].dimension_retransmit, value); + } + + i = ptr->last; + value = (collected_number)w[i].plot.plot_retransmit; + write_chart_dimension(w[i].dimension_retransmit, value); + write_end_chart(); +} + +/** + * Send network viewer data + * + * @param ptr the pointer to plot data + */ +static void ebpf_socket_send_nv_data(netdata_vector_plot_t *ptr) +{ + if (!ptr->flags) + return; + + if (ptr == (netdata_vector_plot_t *)&outbound_vectors) { + ebpf_socket_nv_send_bytes(ptr, NETDATA_NV_OUTBOUND_BYTES); + fflush(stdout); + + ebpf_socket_nv_send_packets(ptr, NETDATA_NV_OUTBOUND_PACKETS); + fflush(stdout); + + ebpf_socket_nv_send_retransmit(ptr, NETDATA_NV_OUTBOUND_RETRANSMIT); + fflush(stdout); + } else { + ebpf_socket_nv_send_bytes(ptr, NETDATA_NV_INBOUND_BYTES); + fflush(stdout); + + ebpf_socket_nv_send_packets(ptr, NETDATA_NV_INBOUND_PACKETS); + fflush(stdout); + } +} + +/** + * Send Global Inbound connection + * + * Send number of connections read per protocol. + */ +static void ebpf_socket_send_global_inbound_conn() +{ + uint64_t udp_conn = 0; + uint64_t tcp_conn = 0; + ebpf_network_viewer_port_list_t *move = listen_ports; + while (move) { + if (move->protocol == IPPROTO_TCP) + tcp_conn += move->connections; + else + udp_conn += move->connections; + + move = move->next; + } + + write_begin_chart(NETDATA_EBPF_IP_FAMILY, NETDATA_INBOUND_CONNECTIONS); + write_chart_dimension(socket_publish_aggregated[NETDATA_IDX_INCOMING_CONNECTION_TCP].name, (long long) tcp_conn); + write_chart_dimension(socket_publish_aggregated[NETDATA_IDX_INCOMING_CONNECTION_UDP].name, (long long) udp_conn); + write_end_chart(); +} + +/** + * Send data to Netdata calling auxiliary functions. + * + * @param em the structure with thread information + */ +static void ebpf_socket_send_data(ebpf_module_t *em) +{ + netdata_publish_vfs_common_t common_tcp; + netdata_publish_vfs_common_t common_udp; + ebpf_update_global_publish(socket_publish_aggregated, &common_tcp, &common_udp, socket_aggregated_data); + + ebpf_socket_send_global_inbound_conn(); + write_count_chart(NETDATA_TCP_OUTBOUND_CONNECTIONS, NETDATA_EBPF_IP_FAMILY, + &socket_publish_aggregated[NETDATA_IDX_TCP_CONNECTION_V4], 2); + + // We read bytes from function arguments, but bandwidth is given in bits, + // so we need to multiply by 8 to convert for the final value. + write_count_chart(NETDATA_TCP_FUNCTION_COUNT, NETDATA_EBPF_IP_FAMILY, socket_publish_aggregated, 3); + write_io_chart(NETDATA_TCP_FUNCTION_BITS, NETDATA_EBPF_IP_FAMILY, socket_id_names[0], + common_tcp.read * 8/BITS_IN_A_KILOBIT, socket_id_names[1], + common_tcp.write * 8/BITS_IN_A_KILOBIT); + if (em->mode < MODE_ENTRY) { + write_err_chart(NETDATA_TCP_FUNCTION_ERROR, NETDATA_EBPF_IP_FAMILY, socket_publish_aggregated, 2); + } + write_count_chart(NETDATA_TCP_RETRANSMIT, NETDATA_EBPF_IP_FAMILY, + &socket_publish_aggregated[NETDATA_IDX_TCP_RETRANSMIT],1); + + write_count_chart(NETDATA_UDP_FUNCTION_COUNT, NETDATA_EBPF_IP_FAMILY, + &socket_publish_aggregated[NETDATA_IDX_UDP_RECVBUF],2); + write_io_chart(NETDATA_UDP_FUNCTION_BITS, NETDATA_EBPF_IP_FAMILY, + socket_id_names[3], (long long)common_udp.read * 8/BITS_IN_A_KILOBIT, + socket_id_names[4], (long long)common_udp.write * 8/BITS_IN_A_KILOBIT); + if (em->mode < MODE_ENTRY) { + write_err_chart(NETDATA_UDP_FUNCTION_ERROR, NETDATA_EBPF_IP_FAMILY, + &socket_publish_aggregated[NETDATA_UDP_START], 2); + } +} + +/** + * Sum values for pid + * + * @param root the structure with all available PIDs + * + * @param offset the address that we are reading + * + * @return it returns the sum of all PIDs + */ +long long ebpf_socket_sum_values_for_pids(struct pid_on_target *root, size_t offset) +{ + long long ret = 0; + while (root) { + int32_t pid = root->pid; + ebpf_socket_publish_apps_t *w = socket_bandwidth_curr[pid]; + if (w) { + ret += get_value_from_structure((char *)w, offset); + } + + root = root->next; + } + + return ret; +} + +/** + * Send data to Netdata calling auxiliary functions. + * + * @param em the structure with thread information + * @param root the target list. + */ +void ebpf_socket_send_apps_data(ebpf_module_t *em, struct target *root) +{ + UNUSED(em); + + struct target *w; + collected_number value; + + write_begin_chart(NETDATA_APPS_FAMILY, NETDATA_NET_APPS_CONNECTION_TCP_V4); + for (w = root; w; w = w->next) { + if (unlikely(w->exposed && w->processes)) { + value = ebpf_socket_sum_values_for_pids(w->root_pid, offsetof(ebpf_socket_publish_apps_t, + call_tcp_v4_connection)); + write_chart_dimension(w->name, value); + } + } + write_end_chart(); + + write_begin_chart(NETDATA_APPS_FAMILY, NETDATA_NET_APPS_CONNECTION_TCP_V6); + for (w = root; w; w = w->next) { + if (unlikely(w->exposed && w->processes)) { + value = ebpf_socket_sum_values_for_pids(w->root_pid, offsetof(ebpf_socket_publish_apps_t, + call_tcp_v6_connection)); + write_chart_dimension(w->name, value); + } + } + write_end_chart(); + + write_begin_chart(NETDATA_APPS_FAMILY, NETDATA_NET_APPS_BANDWIDTH_SENT); + for (w = root; w; w = w->next) { + if (unlikely(w->exposed && w->processes)) { + value = ebpf_socket_sum_values_for_pids(w->root_pid, offsetof(ebpf_socket_publish_apps_t, + bytes_sent)); + // We multiply by 0.008, because we read bytes, but we display bits + write_chart_dimension(w->name, ((value)*8)/1000); + } + } + write_end_chart(); + + write_begin_chart(NETDATA_APPS_FAMILY, NETDATA_NET_APPS_BANDWIDTH_RECV); + for (w = root; w; w = w->next) { + if (unlikely(w->exposed && w->processes)) { + value = ebpf_socket_sum_values_for_pids(w->root_pid, offsetof(ebpf_socket_publish_apps_t, + bytes_received)); + // We multiply by 0.008, because we read bytes, but we display bits + write_chart_dimension(w->name, ((value)*8)/1000); + } + } + write_end_chart(); + + write_begin_chart(NETDATA_APPS_FAMILY, NETDATA_NET_APPS_BANDWIDTH_TCP_SEND_CALLS); + for (w = root; w; w = w->next) { + if (unlikely(w->exposed && w->processes)) { + value = ebpf_socket_sum_values_for_pids(w->root_pid, offsetof(ebpf_socket_publish_apps_t, + call_tcp_sent)); + write_chart_dimension(w->name, value); + } + } + write_end_chart(); + + write_begin_chart(NETDATA_APPS_FAMILY, NETDATA_NET_APPS_BANDWIDTH_TCP_RECV_CALLS); + for (w = root; w; w = w->next) { + if (unlikely(w->exposed && w->processes)) { + value = ebpf_socket_sum_values_for_pids(w->root_pid, offsetof(ebpf_socket_publish_apps_t, + call_tcp_received)); + write_chart_dimension(w->name, value); + } + } + write_end_chart(); + + write_begin_chart(NETDATA_APPS_FAMILY, NETDATA_NET_APPS_BANDWIDTH_TCP_RETRANSMIT); + for (w = root; w; w = w->next) { + if (unlikely(w->exposed && w->processes)) { + value = ebpf_socket_sum_values_for_pids(w->root_pid, offsetof(ebpf_socket_publish_apps_t, + retransmit)); + write_chart_dimension(w->name, value); + } + } + write_end_chart(); + + write_begin_chart(NETDATA_APPS_FAMILY, NETDATA_NET_APPS_BANDWIDTH_UDP_SEND_CALLS); + for (w = root; w; w = w->next) { + if (unlikely(w->exposed && w->processes)) { + value = ebpf_socket_sum_values_for_pids(w->root_pid, offsetof(ebpf_socket_publish_apps_t, + call_udp_sent)); + write_chart_dimension(w->name, value); + } + } + write_end_chart(); + + write_begin_chart(NETDATA_APPS_FAMILY, NETDATA_NET_APPS_BANDWIDTH_UDP_RECV_CALLS); + for (w = root; w; w = w->next) { + if (unlikely(w->exposed && w->processes)) { + value = ebpf_socket_sum_values_for_pids(w->root_pid, offsetof(ebpf_socket_publish_apps_t, + call_udp_received)); + write_chart_dimension(w->name, value); + } + } + write_end_chart(); + +} + +/***************************************************************** + * + * FUNCTIONS TO CREATE CHARTS + * + *****************************************************************/ + +/** + * Create global charts + * + * Call ebpf_create_chart to create the charts for the collector. + * + * @param em a pointer to the structure with the default values. + */ +static void ebpf_create_global_charts(ebpf_module_t *em) +{ + int order = 21070; + ebpf_create_chart(NETDATA_EBPF_IP_FAMILY, + NETDATA_INBOUND_CONNECTIONS, + "Inbound connections.", + EBPF_COMMON_DIMENSION_CONNECTIONS, + NETDATA_SOCKET_KERNEL_FUNCTIONS, + NULL, + NETDATA_EBPF_CHART_TYPE_LINE, + order++, + ebpf_create_global_dimension, + &socket_publish_aggregated[NETDATA_IDX_INCOMING_CONNECTION_TCP], + 2, em->update_every, NETDATA_EBPF_MODULE_NAME_SOCKET); + + ebpf_create_chart(NETDATA_EBPF_IP_FAMILY, + NETDATA_TCP_OUTBOUND_CONNECTIONS, + "TCP outbound connections.", + EBPF_COMMON_DIMENSION_CONNECTIONS, + NETDATA_SOCKET_KERNEL_FUNCTIONS, + NULL, + NETDATA_EBPF_CHART_TYPE_LINE, + order++, + ebpf_create_global_dimension, + &socket_publish_aggregated[NETDATA_IDX_TCP_CONNECTION_V4], + 2, em->update_every, NETDATA_EBPF_MODULE_NAME_SOCKET); + + + ebpf_create_chart(NETDATA_EBPF_IP_FAMILY, + NETDATA_TCP_FUNCTION_COUNT, + "Calls to internal functions", + EBPF_COMMON_DIMENSION_CALL, + NETDATA_SOCKET_KERNEL_FUNCTIONS, + NULL, + NETDATA_EBPF_CHART_TYPE_LINE, + order++, + ebpf_create_global_dimension, + socket_publish_aggregated, + 3, em->update_every, NETDATA_EBPF_MODULE_NAME_SOCKET); + + ebpf_create_chart(NETDATA_EBPF_IP_FAMILY, NETDATA_TCP_FUNCTION_BITS, + "TCP bandwidth", EBPF_COMMON_DIMENSION_BITS, + NETDATA_SOCKET_KERNEL_FUNCTIONS, + NULL, + NETDATA_EBPF_CHART_TYPE_LINE, + order++, + ebpf_create_global_dimension, + socket_publish_aggregated, + 2, em->update_every, NETDATA_EBPF_MODULE_NAME_SOCKET); + + if (em->mode < MODE_ENTRY) { + ebpf_create_chart(NETDATA_EBPF_IP_FAMILY, + NETDATA_TCP_FUNCTION_ERROR, + "TCP errors", + EBPF_COMMON_DIMENSION_CALL, + NETDATA_SOCKET_KERNEL_FUNCTIONS, + NULL, + NETDATA_EBPF_CHART_TYPE_LINE, + order++, + ebpf_create_global_dimension, + socket_publish_aggregated, + 2, em->update_every, NETDATA_EBPF_MODULE_NAME_SOCKET); + } + + ebpf_create_chart(NETDATA_EBPF_IP_FAMILY, + NETDATA_TCP_RETRANSMIT, + "Packages retransmitted", + EBPF_COMMON_DIMENSION_CALL, + NETDATA_SOCKET_KERNEL_FUNCTIONS, + NULL, + NETDATA_EBPF_CHART_TYPE_LINE, + order++, + ebpf_create_global_dimension, + &socket_publish_aggregated[NETDATA_IDX_TCP_RETRANSMIT], + 1, em->update_every, NETDATA_EBPF_MODULE_NAME_SOCKET); + + ebpf_create_chart(NETDATA_EBPF_IP_FAMILY, + NETDATA_UDP_FUNCTION_COUNT, + "UDP calls", + EBPF_COMMON_DIMENSION_CALL, + NETDATA_SOCKET_KERNEL_FUNCTIONS, + NULL, + NETDATA_EBPF_CHART_TYPE_LINE, + order++, + ebpf_create_global_dimension, + &socket_publish_aggregated[NETDATA_IDX_UDP_RECVBUF], + 2, em->update_every, NETDATA_EBPF_MODULE_NAME_SOCKET); + + ebpf_create_chart(NETDATA_EBPF_IP_FAMILY, NETDATA_UDP_FUNCTION_BITS, + "UDP bandwidth", EBPF_COMMON_DIMENSION_BITS, + NETDATA_SOCKET_KERNEL_FUNCTIONS, + NULL, + NETDATA_EBPF_CHART_TYPE_LINE, + order++, + ebpf_create_global_dimension, + &socket_publish_aggregated[NETDATA_IDX_UDP_RECVBUF], + 2, em->update_every, NETDATA_EBPF_MODULE_NAME_SOCKET); + + if (em->mode < MODE_ENTRY) { + ebpf_create_chart(NETDATA_EBPF_IP_FAMILY, + NETDATA_UDP_FUNCTION_ERROR, + "UDP errors", + EBPF_COMMON_DIMENSION_CALL, + NETDATA_SOCKET_KERNEL_FUNCTIONS, + NULL, + NETDATA_EBPF_CHART_TYPE_LINE, + order++, + ebpf_create_global_dimension, + &socket_publish_aggregated[NETDATA_IDX_UDP_RECVBUF], + 2, em->update_every, NETDATA_EBPF_MODULE_NAME_SOCKET); + } +} + +/** + * Create apps charts + * + * Call ebpf_create_chart to create the charts on apps submenu. + * + * @param em a pointer to the structure with the default values. + * @param ptr a pointer for targets + */ +void ebpf_socket_create_apps_charts(struct ebpf_module *em, void *ptr) +{ + struct target *root = ptr; + int order = 20080; + ebpf_create_charts_on_apps(NETDATA_NET_APPS_CONNECTION_TCP_V4, + "Calls to tcp_v4_connection", EBPF_COMMON_DIMENSION_CONNECTIONS, + NETDATA_APPS_NET_GROUP, + NETDATA_EBPF_CHART_TYPE_STACKED, + order++, + ebpf_algorithms[NETDATA_EBPF_INCREMENTAL_IDX], + root, em->update_every, NETDATA_EBPF_MODULE_NAME_SOCKET); + + ebpf_create_charts_on_apps(NETDATA_NET_APPS_CONNECTION_TCP_V6, + "Calls to tcp_v6_connection", EBPF_COMMON_DIMENSION_CONNECTIONS, + NETDATA_APPS_NET_GROUP, + NETDATA_EBPF_CHART_TYPE_STACKED, + order++, + ebpf_algorithms[NETDATA_EBPF_INCREMENTAL_IDX], + root, em->update_every, NETDATA_EBPF_MODULE_NAME_SOCKET); + + ebpf_create_charts_on_apps(NETDATA_NET_APPS_BANDWIDTH_SENT, + "Bytes sent", EBPF_COMMON_DIMENSION_BITS, + NETDATA_APPS_NET_GROUP, + NETDATA_EBPF_CHART_TYPE_STACKED, + order++, + ebpf_algorithms[NETDATA_EBPF_INCREMENTAL_IDX], + root, em->update_every, NETDATA_EBPF_MODULE_NAME_SOCKET); + + ebpf_create_charts_on_apps(NETDATA_NET_APPS_BANDWIDTH_RECV, + "bytes received", EBPF_COMMON_DIMENSION_BITS, + NETDATA_APPS_NET_GROUP, + NETDATA_EBPF_CHART_TYPE_STACKED, + order++, + ebpf_algorithms[NETDATA_EBPF_INCREMENTAL_IDX], + root, em->update_every, NETDATA_EBPF_MODULE_NAME_SOCKET); + + ebpf_create_charts_on_apps(NETDATA_NET_APPS_BANDWIDTH_TCP_SEND_CALLS, + "Calls for tcp_sendmsg", + EBPF_COMMON_DIMENSION_CALL, + NETDATA_APPS_NET_GROUP, + NETDATA_EBPF_CHART_TYPE_STACKED, + order++, + ebpf_algorithms[NETDATA_EBPF_INCREMENTAL_IDX], + root, em->update_every, NETDATA_EBPF_MODULE_NAME_SOCKET); + + ebpf_create_charts_on_apps(NETDATA_NET_APPS_BANDWIDTH_TCP_RECV_CALLS, + "Calls for tcp_cleanup_rbuf", + EBPF_COMMON_DIMENSION_CALL, + NETDATA_APPS_NET_GROUP, + NETDATA_EBPF_CHART_TYPE_STACKED, + order++, + ebpf_algorithms[NETDATA_EBPF_INCREMENTAL_IDX], + root, em->update_every, NETDATA_EBPF_MODULE_NAME_SOCKET); + + ebpf_create_charts_on_apps(NETDATA_NET_APPS_BANDWIDTH_TCP_RETRANSMIT, + "Calls for tcp_retransmit", + EBPF_COMMON_DIMENSION_CALL, + NETDATA_APPS_NET_GROUP, + NETDATA_EBPF_CHART_TYPE_STACKED, + order++, + ebpf_algorithms[NETDATA_EBPF_INCREMENTAL_IDX], + root, em->update_every, NETDATA_EBPF_MODULE_NAME_SOCKET); + + ebpf_create_charts_on_apps(NETDATA_NET_APPS_BANDWIDTH_UDP_SEND_CALLS, + "Calls for udp_sendmsg", + EBPF_COMMON_DIMENSION_CALL, + NETDATA_APPS_NET_GROUP, + NETDATA_EBPF_CHART_TYPE_STACKED, + order++, + ebpf_algorithms[NETDATA_EBPF_INCREMENTAL_IDX], + root, em->update_every, NETDATA_EBPF_MODULE_NAME_SOCKET); + + ebpf_create_charts_on_apps(NETDATA_NET_APPS_BANDWIDTH_UDP_RECV_CALLS, + "Calls for udp_recvmsg", + EBPF_COMMON_DIMENSION_CALL, + NETDATA_APPS_NET_GROUP, + NETDATA_EBPF_CHART_TYPE_STACKED, + order++, + ebpf_algorithms[NETDATA_EBPF_INCREMENTAL_IDX], + root, em->update_every, NETDATA_EBPF_MODULE_NAME_SOCKET); + + em->apps_charts |= NETDATA_EBPF_APPS_FLAG_CHART_CREATED; +} + +/** + * Create network viewer chart + * + * Create common charts. + * + * @param id chart id + * @param title chart title + * @param units units label + * @param family group name used to attach the chart on dashboard + * @param order chart order + * @param update_every value to overwrite the update frequency set by the server. + * @param ptr plot structure with values. + */ +static void ebpf_socket_create_nv_chart(char *id, char *title, char *units, + char *family, int order, int update_every, netdata_vector_plot_t *ptr) +{ + ebpf_write_chart_cmd(NETDATA_EBPF_FAMILY, + id, + title, + units, + family, + NETDATA_EBPF_CHART_TYPE_STACKED, + NULL, + order, + update_every, + NETDATA_EBPF_MODULE_NAME_SOCKET); + + uint32_t i; + uint32_t end = ptr->last_plot; + netdata_socket_plot_t *w = ptr->plot; + for (i = 0; i < end; i++) { + fprintf(stdout, "DIMENSION %s '' incremental -1 1\n", w[i].dimension_sent); + fprintf(stdout, "DIMENSION %s '' incremental 1 1\n", w[i].dimension_recv); + } + + end = ptr->last; + fprintf(stdout, "DIMENSION %s '' incremental -1 1\n", w[end].dimension_sent); + fprintf(stdout, "DIMENSION %s '' incremental 1 1\n", w[end].dimension_recv); +} + +/** + * Create network viewer retransmit + * + * Create a specific chart. + * + * @param id the chart id + * @param title the chart title + * @param units the units label + * @param family the group name used to attach the chart on dashboard + * @param order the chart order + * @param update_every value to overwrite the update frequency set by the server. + * @param ptr the plot structure with values. + */ +static void ebpf_socket_create_nv_retransmit(char *id, char *title, char *units, + char *family, int order, int update_every, netdata_vector_plot_t *ptr) +{ + ebpf_write_chart_cmd(NETDATA_EBPF_FAMILY, + id, + title, + units, + family, + NETDATA_EBPF_CHART_TYPE_STACKED, + NULL, + order, + update_every, + NETDATA_EBPF_MODULE_NAME_SOCKET); + + uint32_t i; + uint32_t end = ptr->last_plot; + netdata_socket_plot_t *w = ptr->plot; + for (i = 0; i < end; i++) { + fprintf(stdout, "DIMENSION %s '' incremental 1 1\n", w[i].dimension_retransmit); + } + + end = ptr->last; + fprintf(stdout, "DIMENSION %s '' incremental 1 1\n", w[end].dimension_retransmit); +} + +/** + * Create Network Viewer charts + * + * Recreate the charts when new sockets are created. + * + * @param ptr a pointer for inbound or outbound vectors. + * @param update_every value to overwrite the update frequency set by the server. + */ +static void ebpf_socket_create_nv_charts(netdata_vector_plot_t *ptr, int update_every) +{ + // We do not have new sockets, so we do not need move forward + if (ptr->max_plot == ptr->last_plot) + return; + + ptr->last_plot = ptr->max_plot; + + if (ptr == (netdata_vector_plot_t *)&outbound_vectors) { + ebpf_socket_create_nv_chart(NETDATA_NV_OUTBOUND_BYTES, + "Outbound connections (bytes).", EBPF_COMMON_DIMENSION_BYTES, + NETDATA_NETWORK_CONNECTIONS_GROUP, + 21080, + update_every, ptr); + + ebpf_socket_create_nv_chart(NETDATA_NV_OUTBOUND_PACKETS, + "Outbound connections (packets)", + EBPF_COMMON_DIMENSION_PACKETS, + NETDATA_NETWORK_CONNECTIONS_GROUP, + 21082, + update_every, ptr); + + ebpf_socket_create_nv_retransmit(NETDATA_NV_OUTBOUND_RETRANSMIT, + "Retransmitted packets", + EBPF_COMMON_DIMENSION_CALL, + NETDATA_NETWORK_CONNECTIONS_GROUP, + 21083, + update_every, ptr); + } else { + ebpf_socket_create_nv_chart(NETDATA_NV_INBOUND_BYTES, + "Inbound connections (bytes)", EBPF_COMMON_DIMENSION_BYTES, + NETDATA_NETWORK_CONNECTIONS_GROUP, + 21084, + update_every, ptr); + + ebpf_socket_create_nv_chart(NETDATA_NV_INBOUND_PACKETS, + "Inbound connections (packets)", + EBPF_COMMON_DIMENSION_PACKETS, + NETDATA_NETWORK_CONNECTIONS_GROUP, + 21085, + update_every, ptr); + } + + ptr->flags |= NETWORK_VIEWER_CHARTS_CREATED; +} + +/***************************************************************** + * + * READ INFORMATION FROM KERNEL RING + * + *****************************************************************/ + +/** + * Is specific ip inside the range + * + * Check if the ip is inside a IP range previously defined + * + * @param cmp the IP to compare + * @param family the IP family + * + * @return It returns 1 if the IP is inside the range and 0 otherwise + */ +static int is_specific_ip_inside_range(union netdata_ip_t *cmp, int family) +{ + if (!network_viewer_opt.excluded_ips && !network_viewer_opt.included_ips) + return 1; + + uint32_t ipv4_test = ntohl(cmp->addr32[0]); + ebpf_network_viewer_ip_list_t *move = network_viewer_opt.excluded_ips; + while (move) { + if (family == AF_INET) { + if (ntohl(move->first.addr32[0]) <= ipv4_test && + ipv4_test <= ntohl(move->last.addr32[0]) ) + return 0; + } else { + if (memcmp(move->first.addr8, cmp->addr8, sizeof(union netdata_ip_t)) <= 0 && + memcmp(move->last.addr8, cmp->addr8, sizeof(union netdata_ip_t)) >= 0) { + return 0; + } + } + move = move->next; + } + + move = network_viewer_opt.included_ips; + while (move) { + if (family == AF_INET) { + if (ntohl(move->first.addr32[0]) <= ipv4_test && + ntohl(move->last.addr32[0]) >= ipv4_test) + return 1; + } else { + if (memcmp(move->first.addr8, cmp->addr8, sizeof(union netdata_ip_t)) <= 0 && + memcmp(move->last.addr8, cmp->addr8, sizeof(union netdata_ip_t)) >= 0) { + return 1; + } + } + move = move->next; + } + + return 0; +} + +/** + * Is port inside range + * + * Verify if the cmp port is inside the range [first, last]. + * This function expects only the last parameter as big endian. + * + * @param cmp the value to compare + * + * @return It returns 1 when cmp is inside and 0 otherwise. + */ +static int is_port_inside_range(uint16_t cmp) +{ + // We do not have restrictions for ports. + if (!network_viewer_opt.excluded_port && !network_viewer_opt.included_port) + return 1; + + // Test if port is excluded + ebpf_network_viewer_port_list_t *move = network_viewer_opt.excluded_port; + cmp = htons(cmp); + while (move) { + if (move->cmp_first <= cmp && cmp <= move->cmp_last) + return 0; + + move = move->next; + } + + // Test if the port is inside allowed range + move = network_viewer_opt.included_port; + while (move) { + if (move->cmp_first <= cmp && cmp <= move->cmp_last) + return 1; + + move = move->next; + } + + return 0; +} + +/** + * Hostname matches pattern + * + * @param cmp the value to compare + * + * @return It returns 1 when the value matches and zero otherwise. + */ +int hostname_matches_pattern(char *cmp) +{ + if (!network_viewer_opt.included_hostnames && !network_viewer_opt.excluded_hostnames) + return 1; + + ebpf_network_viewer_hostname_list_t *move = network_viewer_opt.excluded_hostnames; + while (move) { + if (simple_pattern_matches(move->value_pattern, cmp)) + return 0; + + move = move->next; + } + + move = network_viewer_opt.included_hostnames; + while (move) { + if (simple_pattern_matches(move->value_pattern, cmp)) + return 1; + + move = move->next; + } + + + return 0; +} + +/** + * Is socket allowed? + * + * Compare destination addresses and destination ports to define next steps + * + * @param key the socket read from kernel ring + * @param family the family used to compare IPs (AF_INET and AF_INET6) + * + * @return It returns 1 if this socket is inside the ranges and 0 otherwise. + */ +int is_socket_allowed(netdata_socket_idx_t *key, int family) +{ + if (!is_port_inside_range(key->dport)) + return 0; + + return is_specific_ip_inside_range(&key->daddr, family); +} + +/** + * Compare sockets + * + * Compare destination address and destination port. + * We do not compare source port, because it is random. + * We also do not compare source address, because inbound and outbound connections are stored in separated AVL trees. + * + * @param a pointer to netdata_socket_plot + * @param b pointer to netdata_socket_plot + * + * @return It returns 0 case the values are equal, 1 case a is bigger than b and -1 case a is smaller than b. + */ +static int compare_sockets(void *a, void *b) +{ + struct netdata_socket_plot *val1 = a; + struct netdata_socket_plot *val2 = b; + int cmp; + + // We do not need to compare val2 family, because data inside hash table is always from the same family + if (val1->family == AF_INET) { //IPV4 + if (val1->flags & NETDATA_INBOUND_DIRECTION) { + if (val1->index.sport == val2->index.sport) + cmp = 0; + else { + cmp = (val1->index.sport > val2->index.sport)?1:-1; + } + } else { + cmp = memcmp(&val1->index.dport, &val2->index.dport, sizeof(uint16_t)); + if (!cmp) { + cmp = memcmp(&val1->index.daddr.addr32[0], &val2->index.daddr.addr32[0], sizeof(uint32_t)); + } + } + } else { + if (val1->flags & NETDATA_INBOUND_DIRECTION) { + if (val1->index.sport == val2->index.sport) + cmp = 0; + else { + cmp = (val1->index.sport > val2->index.sport)?1:-1; + } + } else { + cmp = memcmp(&val1->index.dport, &val2->index.dport, sizeof(uint16_t)); + if (!cmp) { + cmp = memcmp(&val1->index.daddr.addr32, &val2->index.daddr.addr32, 4*sizeof(uint32_t)); + } + } + } + + return cmp; +} + +/** + * Build dimension name + * + * Fill dimension name vector with values given + * + * @param dimname the output vector + * @param hostname the hostname for the socket. + * @param service_name the service used to connect. + * @param proto the protocol used in this connection + * @param family is this IPV4(AF_INET) or IPV6(AF_INET6) + * + * @return it returns the size of the data copied on success and -1 otherwise. + */ +static inline int build_outbound_dimension_name(char *dimname, char *hostname, char *service_name, + char *proto, int family) +{ + return snprintf(dimname, CONFIG_MAX_NAME - 7, (family == AF_INET)?"%s:%s:%s_":"%s:%s:[%s]_", + service_name, proto, + hostname); +} + +/** + * Fill inbound dimension name + * + * Mount the dimension name with the input given + * + * @param dimname the output vector + * @param service_name the service used to connect. + * @param proto the protocol used in this connection + * + * @return it returns the size of the data copied on success and -1 otherwise. + */ +static inline int build_inbound_dimension_name(char *dimname, char *service_name, char *proto) +{ + return snprintf(dimname, CONFIG_MAX_NAME - 7, "%s:%s_", service_name, + proto); +} + +/** + * Fill Resolved Name + * + * Fill the resolved name structure with the value given. + * The hostname is the largest value possible, if it is necessary to cut some value, it must be cut. + * + * @param ptr the output vector + * @param hostname the hostname resolved or IP. + * @param length the length for the hostname. + * @param service_name the service name associated to the connection + * @param is_outbound the is this an outbound connection + */ +static inline void fill_resolved_name(netdata_socket_plot_t *ptr, char *hostname, size_t length, + char *service_name, int is_outbound) +{ + if (length < NETDATA_MAX_NETWORK_COMBINED_LENGTH) + ptr->resolved_name = strdupz(hostname); + else { + length = NETDATA_MAX_NETWORK_COMBINED_LENGTH; + ptr->resolved_name = mallocz( NETDATA_MAX_NETWORK_COMBINED_LENGTH + 1); + memcpy(ptr->resolved_name, hostname, length); + ptr->resolved_name[length] = '\0'; + } + + char dimname[CONFIG_MAX_NAME]; + int size; + char *protocol; + if (ptr->sock.protocol == IPPROTO_UDP) { + protocol = "UDP"; + } else if (ptr->sock.protocol == IPPROTO_TCP) { + protocol = "TCP"; + } else { + protocol = "ALL"; + } + + if (is_outbound) + size = build_outbound_dimension_name(dimname, hostname, service_name, protocol, ptr->family); + else + size = build_inbound_dimension_name(dimname,service_name, protocol); + + if (size > 0) { + strcpy(&dimname[size], "sent"); + dimname[size + 4] = '\0'; + ptr->dimension_sent = strdupz(dimname); + + strcpy(&dimname[size], "recv"); + ptr->dimension_recv = strdupz(dimname); + + dimname[size - 1] = '\0'; + ptr->dimension_retransmit = strdupz(dimname); + } +} + +/** + * Mount dimension names + * + * Fill the vector names after to resolve the addresses + * + * @param ptr a pointer to the structure where the values are stored. + * @param is_outbound is a outbound ptr value? + * + * @return It returns 1 if the name is valid and 0 otherwise. + */ +int fill_names(netdata_socket_plot_t *ptr, int is_outbound) +{ + char hostname[NI_MAXHOST], service_name[NI_MAXSERV]; + if (ptr->resolved) + return 1; + + int ret; + static int resolve_name = -1; + static int resolve_service = -1; + if (resolve_name == -1) + resolve_name = network_viewer_opt.hostname_resolution_enabled; + + if (resolve_service == -1) + resolve_service = network_viewer_opt.service_resolution_enabled; + + netdata_socket_idx_t *idx = &ptr->index; + + char *errname = { "Not resolved" }; + // Resolve Name + if (ptr->family == AF_INET) { //IPV4 + struct sockaddr_in myaddr; + memset(&myaddr, 0 , sizeof(myaddr)); + + myaddr.sin_family = ptr->family; + if (is_outbound) { + myaddr.sin_port = idx->dport; + myaddr.sin_addr.s_addr = idx->daddr.addr32[0]; + } else { + myaddr.sin_port = idx->sport; + myaddr.sin_addr.s_addr = idx->saddr.addr32[0]; + } + + ret = (!resolve_name)?-1:getnameinfo((struct sockaddr *)&myaddr, sizeof(myaddr), hostname, + sizeof(hostname), service_name, sizeof(service_name), NI_NAMEREQD); + + if (!ret && !resolve_service) { + snprintf(service_name, sizeof(service_name), "%u", ntohs(myaddr.sin_port)); + } + + if (ret) { + // I cannot resolve the name, I will use the IP + if (!inet_ntop(AF_INET, &myaddr.sin_addr.s_addr, hostname, NI_MAXHOST)) { + strncpy(hostname, errname, 13); + } + + snprintf(service_name, sizeof(service_name), "%u", ntohs(myaddr.sin_port)); + ret = 1; + } + } else { // IPV6 + struct sockaddr_in6 myaddr6; + memset(&myaddr6, 0 , sizeof(myaddr6)); + + myaddr6.sin6_family = AF_INET6; + if (is_outbound) { + myaddr6.sin6_port = idx->dport; + memcpy(myaddr6.sin6_addr.s6_addr, idx->daddr.addr8, sizeof(union netdata_ip_t)); + } else { + myaddr6.sin6_port = idx->sport; + memcpy(myaddr6.sin6_addr.s6_addr, idx->saddr.addr8, sizeof(union netdata_ip_t)); + } + + ret = (!resolve_name)?-1:getnameinfo((struct sockaddr *)&myaddr6, sizeof(myaddr6), hostname, + sizeof(hostname), service_name, sizeof(service_name), NI_NAMEREQD); + + if (!ret && !resolve_service) { + snprintf(service_name, sizeof(service_name), "%u", ntohs(myaddr6.sin6_port)); + } + + if (ret) { + // I cannot resolve the name, I will use the IP + if (!inet_ntop(AF_INET6, myaddr6.sin6_addr.s6_addr, hostname, NI_MAXHOST)) { + strncpy(hostname, errname, 13); + } + + snprintf(service_name, sizeof(service_name), "%u", ntohs(myaddr6.sin6_port)); + + ret = 1; + } + } + + fill_resolved_name(ptr, hostname, + strlen(hostname) + strlen(service_name)+ NETDATA_DOTS_PROTOCOL_COMBINED_LENGTH, + service_name, is_outbound); + + if (resolve_name && !ret) + ret = hostname_matches_pattern(hostname); + + ptr->resolved++; + + return ret; +} + +/** + * Fill last Network Viewer Dimension + * + * Fill the unique dimension that is always plotted. + * + * @param ptr the pointer for the last dimension + * @param is_outbound is this an inbound structure? + */ +static void fill_last_nv_dimension(netdata_socket_plot_t *ptr, int is_outbound) +{ + char hostname[NI_MAXHOST], service_name[NI_MAXSERV]; + char *other = { "other" }; + // We are also copying the NULL bytes to avoid warnings in new compilers + strncpy(hostname, other, 6); + strncpy(service_name, other, 6); + + ptr->family = AF_INET; + ptr->sock.protocol = 255; + ptr->flags = (!is_outbound)?NETDATA_INBOUND_DIRECTION:NETDATA_OUTBOUND_DIRECTION; + + fill_resolved_name(ptr, hostname, 10 + NETDATA_DOTS_PROTOCOL_COMBINED_LENGTH, service_name, is_outbound); + +#ifdef NETDATA_INTERNAL_CHECKS + info("Last %s dimension added: ID = %u, IP = OTHER, NAME = %s, DIM1 = %s, DIM2 = %s, DIM3 = %s", + (is_outbound)?"outbound":"inbound", network_viewer_opt.max_dim - 1, ptr->resolved_name, + ptr->dimension_recv, ptr->dimension_sent, ptr->dimension_retransmit); +#endif +} + +/** + * Update Socket Data + * + * Update the socket information with last collected data + * + * @param sock + * @param lvalues + */ +static inline void update_socket_data(netdata_socket_t *sock, netdata_socket_t *lvalues) +{ + sock->recv_packets += lvalues->recv_packets; + sock->sent_packets += lvalues->sent_packets; + sock->recv_bytes += lvalues->recv_bytes; + sock->sent_bytes += lvalues->sent_bytes; + sock->retransmit += lvalues->retransmit; + + if (lvalues->ct > sock->ct) + sock->ct = lvalues->ct; +} + +/** + * Store socket inside avl + * + * Store the socket values inside the avl tree. + * + * @param out the structure with information used to plot charts. + * @param lvalues Values read from socket ring. + * @param lindex the index information, the real socket. + * @param family the family associated to the socket + * @param flags the connection flags + */ +static void store_socket_inside_avl(netdata_vector_plot_t *out, netdata_socket_t *lvalues, + netdata_socket_idx_t *lindex, int family, uint32_t flags) +{ + netdata_socket_plot_t test, *ret ; + + memcpy(&test.index, lindex, sizeof(netdata_socket_idx_t)); + test.flags = flags; + + ret = (netdata_socket_plot_t *) avl_search_lock(&out->tree, (avl_t *)&test); + if (ret) { + if (lvalues->ct > ret->plot.last_time) { + update_socket_data(&ret->sock, lvalues); + } + } else { + uint32_t curr = out->next; + uint32_t last = out->last; + + netdata_socket_plot_t *w = &out->plot[curr]; + + int resolved; + if (curr == last) { + if (lvalues->ct > w->plot.last_time) { + update_socket_data(&w->sock, lvalues); + } + return; + } else { + memcpy(&w->sock, lvalues, sizeof(netdata_socket_t)); + memcpy(&w->index, lindex, sizeof(netdata_socket_idx_t)); + w->family = family; + + resolved = fill_names(w, out != (netdata_vector_plot_t *)&inbound_vectors); + } + + if (!resolved) { + freez(w->resolved_name); + freez(w->dimension_sent); + freez(w->dimension_recv); + freez(w->dimension_retransmit); + + memset(w, 0, sizeof(netdata_socket_plot_t)); + + return; + } + + w->flags = flags; + netdata_socket_plot_t *check ; + check = (netdata_socket_plot_t *) avl_insert_lock(&out->tree, (avl_t *)w); + if (check != w) + error("Internal error, cannot insert the AVL tree."); + +#ifdef NETDATA_INTERNAL_CHECKS + char iptext[INET6_ADDRSTRLEN]; + if (inet_ntop(family, &w->index.daddr.addr8, iptext, sizeof(iptext))) + info("New %s dimension added: ID = %u, IP = %s, NAME = %s, DIM1 = %s, DIM2 = %s, DIM3 = %s", + (out == &inbound_vectors)?"inbound":"outbound", curr, iptext, w->resolved_name, + w->dimension_recv, w->dimension_sent, w->dimension_retransmit); +#endif + curr++; + if (curr > last) + curr = last; + out->next = curr; + } +} + +/** + * Compare Vector to store + * + * Compare input values with local address to select table to store. + * + * @param direction store inbound and outbound direction. + * @param cmp index read from hash table. + * @param proto the protocol read. + * + * @return It returns the structure with address to compare. + */ +netdata_vector_plot_t * select_vector_to_store(uint32_t *direction, netdata_socket_idx_t *cmp, uint8_t proto) +{ + if (!listen_ports) { + *direction = NETDATA_OUTBOUND_DIRECTION; + return &outbound_vectors; + } + + ebpf_network_viewer_port_list_t *move_ports = listen_ports; + while (move_ports) { + if (move_ports->protocol == proto && move_ports->first == cmp->sport) { + *direction = NETDATA_INBOUND_DIRECTION; + return &inbound_vectors; + } + + move_ports = move_ports->next; + } + + *direction = NETDATA_OUTBOUND_DIRECTION; + return &outbound_vectors; +} + +/** + * Hash accumulator + * + * @param values the values used to calculate the data. + * @param key the key to store data. + * @param family the connection family + * @param end the values size. + */ +static void hash_accumulator(netdata_socket_t *values, netdata_socket_idx_t *key, int family, int end) +{ + uint64_t bsent = 0, brecv = 0, psent = 0, precv = 0; + uint16_t retransmit = 0; + int i; + uint8_t protocol = values[0].protocol; + uint64_t ct = values[0].ct; + for (i = 1; i < end; i++) { + netdata_socket_t *w = &values[i]; + + precv += w->recv_packets; + psent += w->sent_packets; + brecv += w->recv_bytes; + bsent += w->sent_bytes; + retransmit += w->retransmit; + + if (!protocol) + protocol = w->protocol; + + if (w->ct > ct) + ct = w->ct; + } + + values[0].recv_packets += precv; + values[0].sent_packets += psent; + values[0].recv_bytes += brecv; + values[0].sent_bytes += bsent; + values[0].retransmit += retransmit; + values[0].protocol = (!protocol)?IPPROTO_TCP:protocol; + values[0].ct = ct; + + if (is_socket_allowed(key, family)) { + uint32_t dir; + netdata_vector_plot_t *table = select_vector_to_store(&dir, key, protocol); + store_socket_inside_avl(table, &values[0], key, family, dir); + } +} + +/** + * Read socket hash table + * + * Read data from hash tables created on kernel ring. + * + * @param fd the hash table with data. + * @param family the family associated to the hash table + * + * @return it returns 0 on success and -1 otherwise. + */ +static void read_socket_hash_table(int fd, int family, int network_connection) +{ + if (wait_to_plot) + return; + + netdata_socket_idx_t key = {}; + netdata_socket_idx_t next_key = {}; + + netdata_socket_t *values = socket_values; + size_t length = ebpf_nprocs*sizeof(netdata_socket_t); + int test, end = (running_on_kernel < NETDATA_KERNEL_V4_15) ? 1 : ebpf_nprocs; + + while (bpf_map_get_next_key(fd, &key, &next_key) == 0) { + // We need to reset the values when we are working on kernel 4.15 or newer, because kernel does not create + // values for specific processor unless it is used to store data. As result of this behavior one the next socket + // can have values from the previous one. + memset(values, 0, length); + test = bpf_map_lookup_elem(fd, &key, values); + if (test < 0) { + key = next_key; + continue; + } + + if (network_connection) { + hash_accumulator(values, &key, family, end); + } + + key = next_key; + } +} + +/** + * Fill Network Viewer Port list + * + * Fill the strcture with values read from /proc or hash table. + * + * @param out the structure where we will store data. + * @param value the ports we are listen to. + * @param proto the protocol used for this connection. + * @param in the strcuture with values read form different sources. + */ +static inline void fill_nv_port_list(ebpf_network_viewer_port_list_t *out, uint16_t value, uint16_t proto, + netdata_passive_connection_t *in) +{ + out->first = value; + out->protocol = proto; + out->pid = in->pid; + out->tgid = in->tgid; + out->connections = in->counter; +} + +/** + * Update listen table + * + * Update link list when it is necessary. + * + * @param value the ports we are listen to. + * @param proto the protocol used with port connection. + * @param in the strcuture with values read form different sources. + */ +void update_listen_table(uint16_t value, uint16_t proto, netdata_passive_connection_t *in) +{ + ebpf_network_viewer_port_list_t *w; + if (likely(listen_ports)) { + ebpf_network_viewer_port_list_t *move = listen_ports, *store = listen_ports; + while (move) { + if (move->protocol == proto && move->first == value) { + move->pid = in->pid; + move->tgid = in->tgid; + move->connections = in->counter; + return; + } + + store = move; + move = move->next; + } + + w = callocz(1, sizeof(ebpf_network_viewer_port_list_t)); + store->next = w; + } else { + w = callocz(1, sizeof(ebpf_network_viewer_port_list_t)); + + listen_ports = w; + } + fill_nv_port_list(w, value, proto, in); + +#ifdef NETDATA_INTERNAL_CHECKS + info("The network viewer is monitoring inbound connections for port %u", ntohs(value)); +#endif +} + +/** + * Read listen table + * + * Read the table with all ports that we are listen on host. + */ +static void read_listen_table() +{ + netdata_passive_connection_idx_t key = {}; + netdata_passive_connection_idx_t next_key = {}; + + int fd = socket_maps[NETDATA_SOCKET_LPORTS].map_fd; + netdata_passive_connection_t value = {}; + while (bpf_map_get_next_key(fd, &key, &next_key) == 0) { + int test = bpf_map_lookup_elem(fd, &key, &value); + if (test < 0) { + key = next_key; + continue; + } + + // The correct protocol must come from kernel + update_listen_table(key.port, key.protocol, &value); + + key = next_key; + memset(&value, 0, sizeof(value)); + } + + if (next_key.port && value.pid) { + // The correct protocol must come from kernel + update_listen_table(next_key.port, next_key.protocol, &value); + } +} + +/** + * Socket read hash + * + * This is the thread callback. + * This thread is necessary, because we cannot freeze the whole plugin to read the data on very busy socket. + * + * @param ptr It is a NULL value for this thread. + * + * @return It always returns NULL. + */ +void *ebpf_socket_read_hash(void *ptr) +{ + netdata_thread_cleanup_push(ebpf_socket_cleanup, ptr); + ebpf_module_t *em = (ebpf_module_t *)ptr; + + heartbeat_t hb; + heartbeat_init(&hb); + usec_t step = NETDATA_SOCKET_READ_SLEEP_MS * em->update_every; + int fd_ipv4 = socket_maps[NETDATA_SOCKET_TABLE_IPV4].map_fd; + int fd_ipv6 = socket_maps[NETDATA_SOCKET_TABLE_IPV6].map_fd; + int network_connection = em->optional; + while (!ebpf_exit_plugin) { + (void)heartbeat_next(&hb, step); + + pthread_mutex_lock(&nv_mutex); + read_listen_table(); + read_socket_hash_table(fd_ipv4, AF_INET, network_connection); + read_socket_hash_table(fd_ipv6, AF_INET6, network_connection); + wait_to_plot = 1; + pthread_mutex_unlock(&nv_mutex); + } + + netdata_thread_cleanup_pop(1); + return NULL; +} + +/** + * Read the hash table and store data to allocated vectors. + */ +static void read_hash_global_tables() +{ + uint64_t idx; + netdata_idx_t res[NETDATA_SOCKET_COUNTER]; + + netdata_idx_t *val = socket_hash_values; + int fd = socket_maps[NETDATA_SOCKET_GLOBAL].map_fd; + for (idx = 0; idx < NETDATA_SOCKET_COUNTER; idx++) { + if (!bpf_map_lookup_elem(fd, &idx, val)) { + uint64_t total = 0; + int i; + int end = ebpf_nprocs; + for (i = 0; i < end; i++) + total += val[i]; + + res[idx] = total; + } else { + res[idx] = 0; + } + } + + socket_aggregated_data[NETDATA_IDX_TCP_SENDMSG].call = res[NETDATA_KEY_CALLS_TCP_SENDMSG]; + socket_aggregated_data[NETDATA_IDX_TCP_CLEANUP_RBUF].call = res[NETDATA_KEY_CALLS_TCP_CLEANUP_RBUF]; + socket_aggregated_data[NETDATA_IDX_TCP_CLOSE].call = res[NETDATA_KEY_CALLS_TCP_CLOSE]; + socket_aggregated_data[NETDATA_IDX_UDP_RECVBUF].call = res[NETDATA_KEY_CALLS_UDP_RECVMSG]; + socket_aggregated_data[NETDATA_IDX_UDP_SENDMSG].call = res[NETDATA_KEY_CALLS_UDP_SENDMSG]; + socket_aggregated_data[NETDATA_IDX_TCP_RETRANSMIT].call = res[NETDATA_KEY_TCP_RETRANSMIT]; + socket_aggregated_data[NETDATA_IDX_TCP_CONNECTION_V4].call = res[NETDATA_KEY_CALLS_TCP_CONNECT_IPV4]; + socket_aggregated_data[NETDATA_IDX_TCP_CONNECTION_V6].call = res[NETDATA_KEY_CALLS_TCP_CONNECT_IPV6]; + + socket_aggregated_data[NETDATA_IDX_TCP_SENDMSG].ecall = res[NETDATA_KEY_ERROR_TCP_SENDMSG]; + socket_aggregated_data[NETDATA_IDX_TCP_CLEANUP_RBUF].ecall = res[NETDATA_KEY_ERROR_TCP_CLEANUP_RBUF]; + socket_aggregated_data[NETDATA_IDX_UDP_RECVBUF].ecall = res[NETDATA_KEY_ERROR_UDP_RECVMSG]; + socket_aggregated_data[NETDATA_IDX_UDP_SENDMSG].ecall = res[NETDATA_KEY_ERROR_UDP_SENDMSG]; + socket_aggregated_data[NETDATA_IDX_TCP_CONNECTION_V4].ecall = res[NETDATA_KEY_ERROR_TCP_CONNECT_IPV4]; + socket_aggregated_data[NETDATA_IDX_TCP_CONNECTION_V6].ecall = res[NETDATA_KEY_ERROR_TCP_CONNECT_IPV6]; + + socket_aggregated_data[NETDATA_IDX_TCP_SENDMSG].bytes = res[NETDATA_KEY_BYTES_TCP_SENDMSG]; + socket_aggregated_data[NETDATA_IDX_TCP_CLEANUP_RBUF].bytes = res[NETDATA_KEY_BYTES_TCP_CLEANUP_RBUF]; + socket_aggregated_data[NETDATA_IDX_UDP_RECVBUF].bytes = res[NETDATA_KEY_BYTES_UDP_RECVMSG]; + socket_aggregated_data[NETDATA_IDX_UDP_SENDMSG].bytes = res[NETDATA_KEY_BYTES_UDP_SENDMSG]; +} + +/** + * Fill publish apps when necessary. + * + * @param current_pid the PID that I am updating + * @param eb the structure with data read from memory. + */ +void ebpf_socket_fill_publish_apps(uint32_t current_pid, ebpf_bandwidth_t *eb) +{ + ebpf_socket_publish_apps_t *curr = socket_bandwidth_curr[current_pid]; + if (!curr) { + curr = callocz(1, sizeof(ebpf_socket_publish_apps_t)); + socket_bandwidth_curr[current_pid] = curr; + } + + curr->bytes_sent = eb->bytes_sent; + curr->bytes_received = eb->bytes_received; + curr->call_tcp_sent = eb->call_tcp_sent; + curr->call_tcp_received = eb->call_tcp_received; + curr->retransmit = eb->retransmit; + curr->call_udp_sent = eb->call_udp_sent; + curr->call_udp_received = eb->call_udp_received; + curr->call_close = eb->close; + curr->call_tcp_v4_connection = eb->tcp_v4_connection; + curr->call_tcp_v6_connection = eb->tcp_v6_connection; +} + +/** + * Bandwidth accumulator. + * + * @param out the vector with the values to sum + */ +void ebpf_socket_bandwidth_accumulator(ebpf_bandwidth_t *out) +{ + int i, end = (running_on_kernel >= NETDATA_KERNEL_V4_15) ? ebpf_nprocs : 1; + ebpf_bandwidth_t *total = &out[0]; + for (i = 1; i < end; i++) { + ebpf_bandwidth_t *move = &out[i]; + total->bytes_sent += move->bytes_sent; + total->bytes_received += move->bytes_received; + total->call_tcp_sent += move->call_tcp_sent; + total->call_tcp_received += move->call_tcp_received; + total->retransmit += move->retransmit; + total->call_udp_sent += move->call_udp_sent; + total->call_udp_received += move->call_udp_received; + total->close += move->close; + total->tcp_v4_connection += move->tcp_v4_connection; + total->tcp_v6_connection += move->tcp_v6_connection; + } +} + +/** + * Update the apps data reading information from the hash table + */ +static void ebpf_socket_update_apps_data() +{ + int fd = socket_maps[NETDATA_SOCKET_TABLE_BANDWIDTH].map_fd; + ebpf_bandwidth_t *eb = bandwidth_vector; + uint32_t key; + struct pid_stat *pids = root_of_pids; + while (pids) { + key = pids->pid; + + if (bpf_map_lookup_elem(fd, &key, eb)) { + pids = pids->next; + continue; + } + + ebpf_socket_bandwidth_accumulator(eb); + + ebpf_socket_fill_publish_apps(key, eb); + + pids = pids->next; + } +} + +/** + * Update cgroup + * + * Update cgroup data based in + */ +static void ebpf_update_socket_cgroup() +{ + ebpf_cgroup_target_t *ect ; + + ebpf_bandwidth_t *eb = bandwidth_vector; + int fd = socket_maps[NETDATA_SOCKET_TABLE_BANDWIDTH].map_fd; + + pthread_mutex_lock(&mutex_cgroup_shm); + for (ect = ebpf_cgroup_pids; ect; ect = ect->next) { + struct pid_on_target2 *pids; + for (pids = ect->pids; pids; pids = pids->next) { + int pid = pids->pid; + ebpf_bandwidth_t *out = &pids->socket; + ebpf_socket_publish_apps_t *publish = &ect->publish_socket; + if (likely(socket_bandwidth_curr) && socket_bandwidth_curr[pid]) { + ebpf_socket_publish_apps_t *in = socket_bandwidth_curr[pid]; + + publish->bytes_sent = in->bytes_sent; + publish->bytes_received = in->bytes_received; + publish->call_tcp_sent = in->call_tcp_sent; + publish->call_tcp_received = in->call_tcp_received; + publish->retransmit = in->retransmit; + publish->call_udp_sent = in->call_udp_sent; + publish->call_udp_received = in->call_udp_received; + publish->call_close = in->call_close; + publish->call_tcp_v4_connection = in->call_tcp_v4_connection; + publish->call_tcp_v6_connection = in->call_tcp_v6_connection; + } else { + if (!bpf_map_lookup_elem(fd, &pid, eb)) { + ebpf_socket_bandwidth_accumulator(eb); + + memcpy(out, eb, sizeof(ebpf_bandwidth_t)); + + publish->bytes_sent = out->bytes_sent; + publish->bytes_received = out->bytes_received; + publish->call_tcp_sent = out->call_tcp_sent; + publish->call_tcp_received = out->call_tcp_received; + publish->retransmit = out->retransmit; + publish->call_udp_sent = out->call_udp_sent; + publish->call_udp_received = out->call_udp_received; + publish->call_close = out->close; + publish->call_tcp_v4_connection = out->tcp_v4_connection; + publish->call_tcp_v6_connection = out->tcp_v6_connection; + } + } + } + } + pthread_mutex_unlock(&mutex_cgroup_shm); +} + +/** + * Sum PIDs + * + * Sum values for all targets. + * + * @param fd structure used to store data + * @param pids input data + */ +static void ebpf_socket_sum_cgroup_pids(ebpf_socket_publish_apps_t *socket, struct pid_on_target2 *pids) +{ + ebpf_socket_publish_apps_t accumulator; + memset(&accumulator, 0, sizeof(accumulator)); + + while (pids) { + ebpf_bandwidth_t *w = &pids->socket; + + accumulator.bytes_received += w->bytes_received; + accumulator.bytes_sent += w->bytes_sent; + accumulator.call_tcp_received += w->call_tcp_received; + accumulator.call_tcp_sent += w->call_tcp_sent; + accumulator.retransmit += w->retransmit; + accumulator.call_udp_received += w->call_udp_received; + accumulator.call_udp_sent += w->call_udp_sent; + accumulator.call_close += w->close; + accumulator.call_tcp_v4_connection += w->tcp_v4_connection; + accumulator.call_tcp_v6_connection += w->tcp_v6_connection; + + pids = pids->next; + } + + socket->bytes_sent = (accumulator.bytes_sent >= socket->bytes_sent) ? accumulator.bytes_sent : socket->bytes_sent; + socket->bytes_received = (accumulator.bytes_received >= socket->bytes_received) ? accumulator.bytes_received : socket->bytes_received; + socket->call_tcp_sent = (accumulator.call_tcp_sent >= socket->call_tcp_sent) ? accumulator.call_tcp_sent : socket->call_tcp_sent; + socket->call_tcp_received = (accumulator.call_tcp_received >= socket->call_tcp_received) ? accumulator.call_tcp_received : socket->call_tcp_received; + socket->retransmit = (accumulator.retransmit >= socket->retransmit) ? accumulator.retransmit : socket->retransmit; + socket->call_udp_sent = (accumulator.call_udp_sent >= socket->call_udp_sent) ? accumulator.call_udp_sent : socket->call_udp_sent; + socket->call_udp_received = (accumulator.call_udp_received >= socket->call_udp_received) ? accumulator.call_udp_received : socket->call_udp_received; + socket->call_close = (accumulator.call_close >= socket->call_close) ? accumulator.call_close : socket->call_close; + socket->call_tcp_v4_connection = (accumulator.call_tcp_v4_connection >= socket->call_tcp_v4_connection) ? + accumulator.call_tcp_v4_connection : socket->call_tcp_v4_connection; + socket->call_tcp_v6_connection = (accumulator.call_tcp_v6_connection >= socket->call_tcp_v6_connection) ? + accumulator.call_tcp_v6_connection : socket->call_tcp_v6_connection; +} + +/** + * Create specific socket charts + * + * Create charts for cgroup/application. + * + * @param type the chart type. + * @param update_every value to overwrite the update frequency set by the server. + */ +static void ebpf_create_specific_socket_charts(char *type, int update_every) +{ + int order_basis = 5300; + ebpf_create_chart(type, NETDATA_NET_APPS_CONNECTION_TCP_V4, + "Calls to tcp_v4_connection", + EBPF_COMMON_DIMENSION_CONNECTIONS, NETDATA_CGROUP_NET_GROUP, + NETDATA_CGROUP_TCP_V4_CONN_CONTEXT, + NETDATA_EBPF_CHART_TYPE_LINE, + NETDATA_CHART_PRIO_CGROUPS_CONTAINERS + order_basis++, + ebpf_create_global_dimension, + &socket_publish_aggregated[NETDATA_IDX_TCP_CONNECTION_V4], 1, + update_every, NETDATA_EBPF_MODULE_NAME_SOCKET); + + ebpf_create_chart(type, NETDATA_NET_APPS_CONNECTION_TCP_V6, + "Calls to tcp_v6_connection", + EBPF_COMMON_DIMENSION_CONNECTIONS, NETDATA_CGROUP_NET_GROUP, + NETDATA_CGROUP_TCP_V6_CONN_CONTEXT, + NETDATA_EBPF_CHART_TYPE_LINE, + NETDATA_CHART_PRIO_CGROUPS_CONTAINERS + order_basis++, + ebpf_create_global_dimension, + &socket_publish_aggregated[NETDATA_IDX_TCP_CONNECTION_V6], 1, + update_every, NETDATA_EBPF_MODULE_NAME_SOCKET); + + ebpf_create_chart(type, NETDATA_NET_APPS_BANDWIDTH_RECV, + "Bytes received", + EBPF_COMMON_DIMENSION_CALL, NETDATA_CGROUP_NET_GROUP, + NETDATA_CGROUP_SOCKET_BYTES_RECV_CONTEXT, + NETDATA_EBPF_CHART_TYPE_LINE, + NETDATA_CHART_PRIO_CGROUPS_CONTAINERS + order_basis++, + ebpf_create_global_dimension, + &socket_publish_aggregated[NETDATA_IDX_TCP_CLEANUP_RBUF], 1, + update_every, NETDATA_EBPF_MODULE_NAME_SOCKET); + + ebpf_create_chart(type, NETDATA_NET_APPS_BANDWIDTH_SENT, + "Bytes sent", + EBPF_COMMON_DIMENSION_CALL, NETDATA_CGROUP_NET_GROUP, + NETDATA_CGROUP_SOCKET_BYTES_SEND_CONTEXT, + NETDATA_EBPF_CHART_TYPE_LINE, + NETDATA_CHART_PRIO_CGROUPS_CONTAINERS + order_basis++, + ebpf_create_global_dimension, + socket_publish_aggregated, 1, + update_every, NETDATA_EBPF_MODULE_NAME_SOCKET); + + ebpf_create_chart(type, NETDATA_NET_APPS_BANDWIDTH_TCP_RECV_CALLS, + "Calls to tcp_cleanup_rbuf.", + EBPF_COMMON_DIMENSION_CALL, NETDATA_CGROUP_NET_GROUP, + NETDATA_CGROUP_SOCKET_TCP_RECV_CONTEXT, + NETDATA_EBPF_CHART_TYPE_LINE, + NETDATA_CHART_PRIO_CGROUPS_CONTAINERS + order_basis++, + ebpf_create_global_dimension, + &socket_publish_aggregated[NETDATA_IDX_TCP_CLEANUP_RBUF], 1, + update_every, NETDATA_EBPF_MODULE_NAME_SOCKET); + + ebpf_create_chart(type, NETDATA_NET_APPS_BANDWIDTH_TCP_SEND_CALLS, + "Calls to tcp_sendmsg.", + EBPF_COMMON_DIMENSION_CALL, NETDATA_CGROUP_NET_GROUP, + NETDATA_CGROUP_SOCKET_TCP_SEND_CONTEXT, + NETDATA_EBPF_CHART_TYPE_LINE, + NETDATA_CHART_PRIO_CGROUPS_CONTAINERS + order_basis++, + ebpf_create_global_dimension, + socket_publish_aggregated, 1, + update_every, NETDATA_EBPF_MODULE_NAME_SOCKET); + + ebpf_create_chart(type, NETDATA_NET_APPS_BANDWIDTH_TCP_RETRANSMIT, + "Calls to tcp_retransmit.", + EBPF_COMMON_DIMENSION_CALL, NETDATA_CGROUP_NET_GROUP, + NETDATA_CGROUP_SOCKET_TCP_RETRANSMIT_CONTEXT, + NETDATA_EBPF_CHART_TYPE_LINE, + NETDATA_CHART_PRIO_CGROUPS_CONTAINERS + order_basis++, + ebpf_create_global_dimension, + &socket_publish_aggregated[NETDATA_IDX_TCP_RETRANSMIT], 1, + update_every, NETDATA_EBPF_MODULE_NAME_SOCKET); + + ebpf_create_chart(type, NETDATA_NET_APPS_BANDWIDTH_UDP_SEND_CALLS, + "Calls to udp_sendmsg", + EBPF_COMMON_DIMENSION_CALL, NETDATA_CGROUP_NET_GROUP, + NETDATA_CGROUP_SOCKET_UDP_SEND_CONTEXT, + NETDATA_EBPF_CHART_TYPE_LINE, + NETDATA_CHART_PRIO_CGROUPS_CONTAINERS + order_basis++, + ebpf_create_global_dimension, + &socket_publish_aggregated[NETDATA_IDX_UDP_SENDMSG], 1, + update_every, NETDATA_EBPF_MODULE_NAME_SOCKET); + + ebpf_create_chart(type, NETDATA_NET_APPS_BANDWIDTH_UDP_RECV_CALLS, + "Calls to udp_recvmsg", + EBPF_COMMON_DIMENSION_CALL, NETDATA_CGROUP_NET_GROUP, + NETDATA_CGROUP_SOCKET_UDP_RECV_CONTEXT, + NETDATA_EBPF_CHART_TYPE_LINE, + NETDATA_CHART_PRIO_CGROUPS_CONTAINERS + order_basis++, + ebpf_create_global_dimension, + &socket_publish_aggregated[NETDATA_IDX_UDP_RECVBUF], 1, + update_every, NETDATA_EBPF_MODULE_NAME_SOCKET); +} + +/** + * Obsolete specific socket charts + * + * Obsolete charts for cgroup/application. + * + * @param type the chart type. + * @param update_every value to overwrite the update frequency set by the server. + */ +static void ebpf_obsolete_specific_socket_charts(char *type, int update_every) +{ + int order_basis = 5300; + ebpf_write_chart_obsolete(type, NETDATA_NET_APPS_CONNECTION_TCP_V4, "Calls to tcp_v4_connection", + EBPF_COMMON_DIMENSION_CONNECTIONS, NETDATA_APPS_NET_GROUP, + NETDATA_EBPF_CHART_TYPE_LINE, NETDATA_SERVICES_SOCKET_TCP_V4_CONN_CONTEXT, + NETDATA_CHART_PRIO_CGROUPS_CONTAINERS + order_basis++, update_every); + + ebpf_write_chart_obsolete(type, NETDATA_NET_APPS_CONNECTION_TCP_V6,"Calls to tcp_v6_connection", + EBPF_COMMON_DIMENSION_CONNECTIONS, NETDATA_APPS_NET_GROUP, + NETDATA_EBPF_CHART_TYPE_LINE, NETDATA_SERVICES_SOCKET_TCP_V6_CONN_CONTEXT, + NETDATA_CHART_PRIO_CGROUPS_CONTAINERS + order_basis++, update_every); + + ebpf_write_chart_obsolete(type, NETDATA_NET_APPS_BANDWIDTH_RECV, "Bytes received", + EBPF_COMMON_DIMENSION_CALL, NETDATA_APPS_NET_GROUP, + NETDATA_EBPF_CHART_TYPE_LINE, NETDATA_SERVICES_SOCKET_BYTES_RECV_CONTEXT, + NETDATA_CHART_PRIO_CGROUPS_CONTAINERS + order_basis++, update_every); + + ebpf_write_chart_obsolete(type, NETDATA_NET_APPS_BANDWIDTH_SENT,"Bytes sent", + EBPF_COMMON_DIMENSION_CALL, NETDATA_APPS_NET_GROUP, + NETDATA_EBPF_CHART_TYPE_LINE, NETDATA_SERVICES_SOCKET_BYTES_SEND_CONTEXT, + NETDATA_CHART_PRIO_CGROUPS_CONTAINERS + order_basis++, update_every); + + ebpf_write_chart_obsolete(type, NETDATA_NET_APPS_BANDWIDTH_TCP_RECV_CALLS, "Calls to tcp_cleanup_rbuf.", + EBPF_COMMON_DIMENSION_CALL, NETDATA_APPS_NET_GROUP, + NETDATA_EBPF_CHART_TYPE_LINE, NETDATA_SERVICES_SOCKET_TCP_RECV_CONTEXT, + NETDATA_CHART_PRIO_CGROUPS_CONTAINERS + order_basis++, update_every); + + ebpf_write_chart_obsolete(type, NETDATA_NET_APPS_BANDWIDTH_TCP_SEND_CALLS, "Calls to tcp_sendmsg.", + EBPF_COMMON_DIMENSION_CALL, NETDATA_APPS_NET_GROUP, + NETDATA_EBPF_CHART_TYPE_LINE, NETDATA_SERVICES_SOCKET_TCP_SEND_CONTEXT, + NETDATA_CHART_PRIO_CGROUPS_CONTAINERS + order_basis++, update_every); + + ebpf_write_chart_obsolete(type, NETDATA_NET_APPS_BANDWIDTH_TCP_RETRANSMIT, "Calls to tcp_retransmit.", + EBPF_COMMON_DIMENSION_CALL, NETDATA_APPS_NET_GROUP, + NETDATA_EBPF_CHART_TYPE_LINE, NETDATA_SERVICES_SOCKET_TCP_RETRANSMIT_CONTEXT, + NETDATA_CHART_PRIO_CGROUPS_CONTAINERS + order_basis++, update_every); + + ebpf_write_chart_obsolete(type, NETDATA_NET_APPS_BANDWIDTH_UDP_SEND_CALLS, "Calls to udp_sendmsg", + EBPF_COMMON_DIMENSION_CALL, NETDATA_APPS_NET_GROUP, + NETDATA_EBPF_CHART_TYPE_LINE, NETDATA_SERVICES_SOCKET_UDP_SEND_CONTEXT, + NETDATA_CHART_PRIO_CGROUPS_CONTAINERS + order_basis++, update_every); + + ebpf_write_chart_obsolete(type, NETDATA_NET_APPS_BANDWIDTH_UDP_RECV_CALLS, "Calls to udp_recvmsg", + EBPF_COMMON_DIMENSION_CALL, NETDATA_APPS_NET_GROUP, NETDATA_EBPF_CHART_TYPE_LINE, + NETDATA_SERVICES_SOCKET_UDP_RECV_CONTEXT, + NETDATA_CHART_PRIO_CGROUPS_CONTAINERS + order_basis++, update_every); +} + +/* + * Send Specific Swap data + * + * Send data for specific cgroup/apps. + * + * @param type chart type + * @param values structure with values that will be sent to netdata + */ +static void ebpf_send_specific_socket_data(char *type, ebpf_socket_publish_apps_t *values) +{ + write_begin_chart(type, NETDATA_NET_APPS_CONNECTION_TCP_V4); + write_chart_dimension(socket_publish_aggregated[NETDATA_IDX_TCP_CONNECTION_V4].name, + (long long) values->call_tcp_v4_connection); + write_end_chart(); + + write_begin_chart(type, NETDATA_NET_APPS_CONNECTION_TCP_V6); + write_chart_dimension(socket_publish_aggregated[NETDATA_IDX_TCP_CONNECTION_V6].name, + (long long) values->call_tcp_v6_connection); + write_end_chart(); + + write_begin_chart(type, NETDATA_NET_APPS_BANDWIDTH_SENT); + write_chart_dimension(socket_publish_aggregated[NETDATA_IDX_TCP_SENDMSG].name, + (long long) values->bytes_sent); + write_end_chart(); + + write_begin_chart(type, NETDATA_NET_APPS_BANDWIDTH_RECV); + write_chart_dimension(socket_publish_aggregated[NETDATA_IDX_TCP_CLEANUP_RBUF].name, + (long long) values->bytes_received); + write_end_chart(); + + write_begin_chart(type, NETDATA_NET_APPS_BANDWIDTH_TCP_SEND_CALLS); + write_chart_dimension(socket_publish_aggregated[NETDATA_IDX_TCP_SENDMSG].name, + (long long) values->call_tcp_sent); + write_end_chart(); + + write_begin_chart(type, NETDATA_NET_APPS_BANDWIDTH_TCP_RECV_CALLS); + write_chart_dimension(socket_publish_aggregated[NETDATA_IDX_TCP_CLEANUP_RBUF].name, + (long long) values->call_tcp_received); + write_end_chart(); + + write_begin_chart(type, NETDATA_NET_APPS_BANDWIDTH_TCP_RETRANSMIT); + write_chart_dimension(socket_publish_aggregated[NETDATA_IDX_TCP_RETRANSMIT].name, + (long long) values->retransmit); + write_end_chart(); + + write_begin_chart(type, NETDATA_NET_APPS_BANDWIDTH_UDP_SEND_CALLS); + write_chart_dimension(socket_publish_aggregated[NETDATA_IDX_UDP_SENDMSG].name, + (long long) values->call_udp_sent); + write_end_chart(); + + write_begin_chart(type, NETDATA_NET_APPS_BANDWIDTH_UDP_RECV_CALLS); + write_chart_dimension(socket_publish_aggregated[NETDATA_IDX_UDP_RECVBUF].name, + (long long) values->call_udp_received); + write_end_chart(); +} + +/** + * Create Systemd Socket Charts + * + * Create charts when systemd is enabled + * + * @param update_every value to overwrite the update frequency set by the server. + **/ +static void ebpf_create_systemd_socket_charts(int update_every) +{ + int order = 20080; + ebpf_create_charts_on_systemd(NETDATA_NET_APPS_CONNECTION_TCP_V4, + "Calls to tcp_v4_connection", EBPF_COMMON_DIMENSION_CONNECTIONS, + NETDATA_APPS_NET_GROUP, + NETDATA_EBPF_CHART_TYPE_STACKED, + order++, + ebpf_algorithms[NETDATA_EBPF_INCREMENTAL_IDX], + NETDATA_SERVICES_SOCKET_TCP_V4_CONN_CONTEXT, NETDATA_EBPF_MODULE_NAME_SOCKET, + update_every); + + ebpf_create_charts_on_systemd(NETDATA_NET_APPS_CONNECTION_TCP_V6, + "Calls to tcp_v6_connection", EBPF_COMMON_DIMENSION_CONNECTIONS, + NETDATA_APPS_NET_GROUP, + NETDATA_EBPF_CHART_TYPE_STACKED, + order++, + ebpf_algorithms[NETDATA_EBPF_INCREMENTAL_IDX], + NETDATA_SERVICES_SOCKET_TCP_V6_CONN_CONTEXT, NETDATA_EBPF_MODULE_NAME_SOCKET, + update_every); + + ebpf_create_charts_on_systemd(NETDATA_NET_APPS_BANDWIDTH_RECV, + "Bytes received", EBPF_COMMON_DIMENSION_BITS, + NETDATA_APPS_NET_GROUP, + NETDATA_EBPF_CHART_TYPE_STACKED, + order++, + ebpf_algorithms[NETDATA_EBPF_INCREMENTAL_IDX], + NETDATA_SERVICES_SOCKET_BYTES_RECV_CONTEXT, NETDATA_EBPF_MODULE_NAME_SOCKET, + update_every); + + ebpf_create_charts_on_systemd(NETDATA_NET_APPS_BANDWIDTH_SENT, + "Bytes sent", EBPF_COMMON_DIMENSION_BITS, + NETDATA_APPS_NET_GROUP, + NETDATA_EBPF_CHART_TYPE_STACKED, + order++, + ebpf_algorithms[NETDATA_EBPF_INCREMENTAL_IDX], + NETDATA_SERVICES_SOCKET_BYTES_SEND_CONTEXT, NETDATA_EBPF_MODULE_NAME_SOCKET, + update_every); + + ebpf_create_charts_on_systemd(NETDATA_NET_APPS_BANDWIDTH_TCP_RECV_CALLS, + "Calls to tcp_cleanup_rbuf.", + EBPF_COMMON_DIMENSION_CALL, + NETDATA_APPS_NET_GROUP, + NETDATA_EBPF_CHART_TYPE_STACKED, + order++, + ebpf_algorithms[NETDATA_EBPF_INCREMENTAL_IDX], + NETDATA_SERVICES_SOCKET_TCP_RECV_CONTEXT, NETDATA_EBPF_MODULE_NAME_SOCKET, + update_every); + + ebpf_create_charts_on_systemd(NETDATA_NET_APPS_BANDWIDTH_TCP_SEND_CALLS, + "Calls to tcp_sendmsg.", + EBPF_COMMON_DIMENSION_CALL, + NETDATA_APPS_NET_GROUP, + NETDATA_EBPF_CHART_TYPE_STACKED, + order++, + ebpf_algorithms[NETDATA_EBPF_INCREMENTAL_IDX], + NETDATA_SERVICES_SOCKET_TCP_SEND_CONTEXT, NETDATA_EBPF_MODULE_NAME_SOCKET, + update_every); + + ebpf_create_charts_on_systemd(NETDATA_NET_APPS_BANDWIDTH_TCP_RETRANSMIT, + "Calls to tcp_retransmit", + EBPF_COMMON_DIMENSION_CALL, + NETDATA_APPS_NET_GROUP, + NETDATA_EBPF_CHART_TYPE_STACKED, + order++, + ebpf_algorithms[NETDATA_EBPF_INCREMENTAL_IDX], + NETDATA_SERVICES_SOCKET_TCP_RETRANSMIT_CONTEXT, NETDATA_EBPF_MODULE_NAME_SOCKET, + update_every); + + ebpf_create_charts_on_systemd(NETDATA_NET_APPS_BANDWIDTH_UDP_SEND_CALLS, + "Calls to udp_sendmsg", + EBPF_COMMON_DIMENSION_CALL, + NETDATA_APPS_NET_GROUP, + NETDATA_EBPF_CHART_TYPE_STACKED, + order++, + ebpf_algorithms[NETDATA_EBPF_INCREMENTAL_IDX], + NETDATA_SERVICES_SOCKET_UDP_SEND_CONTEXT, NETDATA_EBPF_MODULE_NAME_SOCKET, + update_every); + + ebpf_create_charts_on_systemd(NETDATA_NET_APPS_BANDWIDTH_UDP_RECV_CALLS, + "Calls to udp_recvmsg", + EBPF_COMMON_DIMENSION_CALL, + NETDATA_APPS_NET_GROUP, + NETDATA_EBPF_CHART_TYPE_STACKED, + order++, + ebpf_algorithms[NETDATA_EBPF_INCREMENTAL_IDX], + NETDATA_SERVICES_SOCKET_UDP_RECV_CONTEXT, NETDATA_EBPF_MODULE_NAME_SOCKET, + update_every); +} + +/** + * Send Systemd charts + * + * Send collected data to Netdata. + */ +static void ebpf_send_systemd_socket_charts() +{ + ebpf_cgroup_target_t *ect; + write_begin_chart(NETDATA_SERVICE_FAMILY, NETDATA_NET_APPS_CONNECTION_TCP_V4); + for (ect = ebpf_cgroup_pids; ect ; ect = ect->next) { + if (unlikely(ect->systemd) && unlikely(ect->updated)) { + write_chart_dimension(ect->name, (long long)ect->publish_socket.call_tcp_v4_connection); + } + } + write_end_chart(); + + write_begin_chart(NETDATA_SERVICE_FAMILY, NETDATA_NET_APPS_CONNECTION_TCP_V6); + for (ect = ebpf_cgroup_pids; ect ; ect = ect->next) { + if (unlikely(ect->systemd) && unlikely(ect->updated)) { + write_chart_dimension(ect->name, (long long)ect->publish_socket.call_tcp_v6_connection); + } + } + write_end_chart(); + + write_begin_chart(NETDATA_SERVICE_FAMILY, NETDATA_NET_APPS_BANDWIDTH_SENT); + for (ect = ebpf_cgroup_pids; ect ; ect = ect->next) { + if (unlikely(ect->systemd) && unlikely(ect->updated)) { + write_chart_dimension(ect->name, (long long)ect->publish_socket.bytes_sent); + } + } + write_end_chart(); + + write_begin_chart(NETDATA_SERVICE_FAMILY, NETDATA_NET_APPS_BANDWIDTH_RECV); + for (ect = ebpf_cgroup_pids; ect ; ect = ect->next) { + if (unlikely(ect->systemd) && unlikely(ect->updated)) { + write_chart_dimension(ect->name, (long long)ect->publish_socket.bytes_received); + } + } + write_end_chart(); + + write_begin_chart(NETDATA_SERVICE_FAMILY, NETDATA_NET_APPS_BANDWIDTH_TCP_SEND_CALLS); + for (ect = ebpf_cgroup_pids; ect ; ect = ect->next) { + if (unlikely(ect->systemd) && unlikely(ect->updated)) { + write_chart_dimension(ect->name, (long long)ect->publish_socket.call_tcp_sent); + } + } + write_end_chart(); + + write_begin_chart(NETDATA_SERVICE_FAMILY, NETDATA_NET_APPS_BANDWIDTH_TCP_RECV_CALLS); + for (ect = ebpf_cgroup_pids; ect ; ect = ect->next) { + if (unlikely(ect->systemd) && unlikely(ect->updated)) { + write_chart_dimension(ect->name, (long long)ect->publish_socket.call_tcp_received); + } + } + write_end_chart(); + + write_begin_chart(NETDATA_SERVICE_FAMILY, NETDATA_NET_APPS_BANDWIDTH_TCP_RETRANSMIT); + for (ect = ebpf_cgroup_pids; ect ; ect = ect->next) { + if (unlikely(ect->systemd) && unlikely(ect->updated)) { + write_chart_dimension(ect->name, (long long)ect->publish_socket.retransmit); + } + } + write_end_chart(); + + write_begin_chart(NETDATA_SERVICE_FAMILY, NETDATA_NET_APPS_BANDWIDTH_UDP_SEND_CALLS); + for (ect = ebpf_cgroup_pids; ect ; ect = ect->next) { + if (unlikely(ect->systemd) && unlikely(ect->updated)) { + write_chart_dimension(ect->name, (long long)ect->publish_socket.call_udp_sent); + } + } + write_end_chart(); + + write_begin_chart(NETDATA_SERVICE_FAMILY, NETDATA_NET_APPS_BANDWIDTH_UDP_RECV_CALLS); + for (ect = ebpf_cgroup_pids; ect ; ect = ect->next) { + if (unlikely(ect->systemd) && unlikely(ect->updated)) { + write_chart_dimension(ect->name, (long long)ect->publish_socket.call_udp_received); + } + } + write_end_chart(); +} + +/** + * Update Cgroup algorithm + * + * Change algorithm from absolute to incremental + */ +void ebpf_socket_update_cgroup_algorithm() +{ + int i; + for (i = 0; i < NETDATA_MAX_SOCKET_VECTOR; i++) { + netdata_publish_syscall_t *ptr = &socket_publish_aggregated[i]; + freez(ptr->algorithm); + ptr->algorithm = strdupz(ebpf_algorithms[NETDATA_EBPF_INCREMENTAL_IDX]); + } +} + +/** + * Send data to Netdata calling auxiliary functions. + * + * @param update_every value to overwrite the update frequency set by the server. +*/ +static void ebpf_socket_send_cgroup_data(int update_every) +{ + if (!ebpf_cgroup_pids) + return; + + pthread_mutex_lock(&mutex_cgroup_shm); + ebpf_cgroup_target_t *ect; + for (ect = ebpf_cgroup_pids; ect ; ect = ect->next) { + ebpf_socket_sum_cgroup_pids(&ect->publish_socket, ect->pids); + } + + int has_systemd = shm_ebpf_cgroup.header->systemd_enabled; + if (has_systemd) { + if (send_cgroup_chart) { + ebpf_create_systemd_socket_charts(update_every); + } + ebpf_send_systemd_socket_charts(); + } + + for (ect = ebpf_cgroup_pids; ect ; ect = ect->next) { + if (ect->systemd) + continue; + + if (!(ect->flags & NETDATA_EBPF_CGROUP_HAS_SOCKET_CHART)) { + ebpf_create_specific_socket_charts(ect->name, update_every); + ect->flags |= NETDATA_EBPF_CGROUP_HAS_SOCKET_CHART; + } + + if (ect->flags & NETDATA_EBPF_CGROUP_HAS_SOCKET_CHART && ect->updated) { + ebpf_send_specific_socket_data(ect->name, &ect->publish_socket); + } else { + ebpf_obsolete_specific_socket_charts(ect->name, update_every); + ect->flags &= ~NETDATA_EBPF_CGROUP_HAS_SOCKET_CHART; + } + } + + pthread_mutex_unlock(&mutex_cgroup_shm); +} + +/***************************************************************** + * + * FUNCTIONS WITH THE MAIN LOOP + * + *****************************************************************/ + +/** + * Main loop for this collector. + * + * @param step the number of microseconds used with heart beat + * @param em the structure with thread information + */ +static void socket_collector(usec_t step, ebpf_module_t *em) +{ + heartbeat_t hb; + heartbeat_init(&hb); + + socket_threads.thread = mallocz(sizeof(netdata_thread_t)); + socket_threads.start_routine = ebpf_socket_read_hash; + + netdata_thread_create(socket_threads.thread, socket_threads.name, + NETDATA_THREAD_OPTION_DEFAULT, ebpf_socket_read_hash, em); + + int cgroups = em->cgroup_charts; + if (cgroups) + ebpf_socket_update_cgroup_algorithm(); + + int socket_global_enabled = em->global_charts; + int network_connection = em->optional; + int update_every = em->update_every; + while (!ebpf_exit_plugin) { + (void)heartbeat_next(&hb, step); + if (ebpf_exit_plugin) + break; + + netdata_apps_integration_flags_t socket_apps_enabled = em->apps_charts; + pthread_mutex_lock(&collect_data_mutex); + if (socket_global_enabled) + read_hash_global_tables(); + + if (socket_apps_enabled) + ebpf_socket_update_apps_data(); + + if (cgroups) + ebpf_update_socket_cgroup(); + + calculate_nv_plot(); + + pthread_mutex_lock(&lock); + if (socket_global_enabled) + ebpf_socket_send_data(em); + + if (socket_apps_enabled & NETDATA_EBPF_APPS_FLAG_CHART_CREATED) + ebpf_socket_send_apps_data(em, apps_groups_root_target); + + if (cgroups) + ebpf_socket_send_cgroup_data(update_every); + + fflush(stdout); + + if (network_connection) { + // We are calling fflush many times, because when we have a lot of dimensions + // we began to have not expected outputs and Netdata closed the plugin. + pthread_mutex_lock(&nv_mutex); + ebpf_socket_create_nv_charts(&inbound_vectors, update_every); + fflush(stdout); + ebpf_socket_send_nv_data(&inbound_vectors); + + ebpf_socket_create_nv_charts(&outbound_vectors, update_every); + fflush(stdout); + ebpf_socket_send_nv_data(&outbound_vectors); + wait_to_plot = 0; + pthread_mutex_unlock(&nv_mutex); + + } + pthread_mutex_unlock(&lock); + pthread_mutex_unlock(&collect_data_mutex); + } +} + +/***************************************************************** + * + * FUNCTIONS TO START THREAD + * + *****************************************************************/ + +/** + * Allocate vectors used with this thread. + * We are not testing the return, because callocz does this and shutdown the software + * case it was not possible to allocate. + * + * @param apps is apps enabled? + */ +static void ebpf_socket_allocate_global_vectors(int apps) +{ + memset(socket_aggregated_data, 0 ,NETDATA_MAX_SOCKET_VECTOR * sizeof(netdata_syscall_stat_t)); + memset(socket_publish_aggregated, 0 ,NETDATA_MAX_SOCKET_VECTOR * sizeof(netdata_publish_syscall_t)); + socket_hash_values = callocz(ebpf_nprocs, sizeof(netdata_idx_t)); + + if (apps) + socket_bandwidth_curr = callocz((size_t)pid_max, sizeof(ebpf_socket_publish_apps_t *)); + + bandwidth_vector = callocz((size_t)ebpf_nprocs, sizeof(ebpf_bandwidth_t)); + + socket_values = callocz((size_t)ebpf_nprocs, sizeof(netdata_socket_t)); + inbound_vectors.plot = callocz(network_viewer_opt.max_dim, sizeof(netdata_socket_plot_t)); + outbound_vectors.plot = callocz(network_viewer_opt.max_dim, sizeof(netdata_socket_plot_t)); +} + +/** + * Initialize Inbound and Outbound + * + * Initialize the common outbound and inbound sockets. + */ +static void initialize_inbound_outbound() +{ + inbound_vectors.last = network_viewer_opt.max_dim - 1; + outbound_vectors.last = inbound_vectors.last; + fill_last_nv_dimension(&inbound_vectors.plot[inbound_vectors.last], 0); + fill_last_nv_dimension(&outbound_vectors.plot[outbound_vectors.last], 1); +} + +/***************************************************************** + * + * EBPF SOCKET THREAD + * + *****************************************************************/ + +/** + * Fill Port list + * + * @param out a pointer to the link list. + * @param in the structure that will be linked. + */ +static inline void fill_port_list(ebpf_network_viewer_port_list_t **out, ebpf_network_viewer_port_list_t *in) +{ + if (likely(*out)) { + ebpf_network_viewer_port_list_t *move = *out, *store = *out; + uint16_t first = ntohs(in->first); + uint16_t last = ntohs(in->last); + while (move) { + uint16_t cmp_first = ntohs(move->first); + uint16_t cmp_last = ntohs(move->last); + if (cmp_first <= first && first <= cmp_last && + cmp_first <= last && last <= cmp_last ) { + info("The range/value (%u, %u) is inside the range/value (%u, %u) already inserted, it will be ignored.", + first, last, cmp_first, cmp_last); + freez(in->value); + freez(in); + return; + } else if (first <= cmp_first && cmp_first <= last && + first <= cmp_last && cmp_last <= last) { + info("The range (%u, %u) is bigger than previous range (%u, %u) already inserted, the previous will be ignored.", + first, last, cmp_first, cmp_last); + freez(move->value); + move->value = in->value; + move->first = in->first; + move->last = in->last; + freez(in); + return; + } + + store = move; + move = move->next; + } + + store->next = in; + } else { + *out = in; + } + +#ifdef NETDATA_INTERNAL_CHECKS + info("Adding values %s( %u, %u) to %s port list used on network viewer", + in->value, ntohs(in->first), ntohs(in->last), + (*out == network_viewer_opt.included_port)?"included":"excluded"); +#endif +} + +/** + * Parse Service List + * + * @param out a pointer to store the link list + * @param service the service used to create the structure that will be linked. + */ +static void parse_service_list(void **out, char *service) +{ + ebpf_network_viewer_port_list_t **list = (ebpf_network_viewer_port_list_t **)out; + struct servent *serv = getservbyname((const char *)service, "tcp"); + if (!serv) + serv = getservbyname((const char *)service, "udp"); + + if (!serv) { + info("Cannot resolv the service '%s' with protocols TCP and UDP, it will be ignored", service); + return; + } + + ebpf_network_viewer_port_list_t *w = callocz(1, sizeof(ebpf_network_viewer_port_list_t)); + w->value = strdupz(service); + w->hash = simple_hash(service); + + w->first = w->last = (uint16_t)serv->s_port; + + fill_port_list(list, w); +} + +/** + * Netmask + * + * Copied from iprange (https://github.com/firehol/iprange/blob/master/iprange.h) + * + * @param prefix create the netmask based in the CIDR value. + * + * @return + */ +static inline in_addr_t netmask(int prefix) { + + if (prefix == 0) + return (~((in_addr_t) - 1)); + else + return (in_addr_t)(~((1 << (32 - prefix)) - 1)); + +} + +/** + * Broadcast + * + * Copied from iprange (https://github.com/firehol/iprange/blob/master/iprange.h) + * + * @param addr is the ip address + * @param prefix is the CIDR value. + * + * @return It returns the last address of the range + */ +static inline in_addr_t broadcast(in_addr_t addr, int prefix) +{ + return (addr | ~netmask(prefix)); +} + +/** + * Network + * + * Copied from iprange (https://github.com/firehol/iprange/blob/master/iprange.h) + * + * @param addr is the ip address + * @param prefix is the CIDR value. + * + * @return It returns the first address of the range. + */ +static inline in_addr_t ipv4_network(in_addr_t addr, int prefix) +{ + return (addr & netmask(prefix)); +} + +/** + * IP to network long + * + * @param dst the vector to store the result + * @param ip the source ip given by our users. + * @param domain the ip domain (IPV4 or IPV6) + * @param source the original string + * + * @return it returns 0 on success and -1 otherwise. + */ +static inline int ip2nl(uint8_t *dst, char *ip, int domain, char *source) +{ + if (inet_pton(domain, ip, dst) <= 0) { + error("The address specified (%s) is invalid ", source); + return -1; + } + + return 0; +} + +/** + * Get IPV6 Last Address + * + * @param out the address to store the last address. + * @param in the address used to do the math. + * @param prefix number of bits used to calculate the address + */ +static void get_ipv6_last_addr(union netdata_ip_t *out, union netdata_ip_t *in, uint64_t prefix) +{ + uint64_t mask,tmp; + uint64_t ret[2]; + memcpy(ret, in->addr32, sizeof(union netdata_ip_t)); + + if (prefix == 128) { + memcpy(out->addr32, in->addr32, sizeof(union netdata_ip_t)); + return; + } else if (!prefix) { + ret[0] = ret[1] = 0xFFFFFFFFFFFFFFFF; + memcpy(out->addr32, ret, sizeof(union netdata_ip_t)); + return; + } else if (prefix <= 64) { + ret[1] = 0xFFFFFFFFFFFFFFFFULL; + + tmp = be64toh(ret[0]); + if (prefix > 0) { + mask = 0xFFFFFFFFFFFFFFFFULL << (64 - prefix); + tmp |= ~mask; + } + ret[0] = htobe64(tmp); + } else { + mask = 0xFFFFFFFFFFFFFFFFULL << (128 - prefix); + tmp = be64toh(ret[1]); + tmp |= ~mask; + ret[1] = htobe64(tmp); + } + + memcpy(out->addr32, ret, sizeof(union netdata_ip_t)); +} + +/** + * Calculate ipv6 first address + * + * @param out the address to store the first address. + * @param in the address used to do the math. + * @param prefix number of bits used to calculate the address + */ +static void get_ipv6_first_addr(union netdata_ip_t *out, union netdata_ip_t *in, uint64_t prefix) +{ + uint64_t mask,tmp; + uint64_t ret[2]; + + memcpy(ret, in->addr32, sizeof(union netdata_ip_t)); + + if (prefix == 128) { + memcpy(out->addr32, in->addr32, sizeof(union netdata_ip_t)); + return; + } else if (!prefix) { + ret[0] = ret[1] = 0; + memcpy(out->addr32, ret, sizeof(union netdata_ip_t)); + return; + } else if (prefix <= 64) { + ret[1] = 0ULL; + + tmp = be64toh(ret[0]); + if (prefix > 0) { + mask = 0xFFFFFFFFFFFFFFFFULL << (64 - prefix); + tmp &= mask; + } + ret[0] = htobe64(tmp); + } else { + mask = 0xFFFFFFFFFFFFFFFFULL << (128 - prefix); + tmp = be64toh(ret[1]); + tmp &= mask; + ret[1] = htobe64(tmp); + } + + memcpy(out->addr32, ret, sizeof(union netdata_ip_t)); +} + +/** + * Is ip inside the range + * + * Check if the ip is inside a IP range + * + * @param rfirst the first ip address of the range + * @param rlast the last ip address of the range + * @param cmpfirst the first ip to compare + * @param cmplast the last ip to compare + * @param family the IP family + * + * @return It returns 1 if the IP is inside the range and 0 otherwise + */ +static int is_ip_inside_range(union netdata_ip_t *rfirst, union netdata_ip_t *rlast, + union netdata_ip_t *cmpfirst, union netdata_ip_t *cmplast, int family) +{ + if (family == AF_INET) { + if (ntohl(rfirst->addr32[0]) <= ntohl(cmpfirst->addr32[0]) && + ntohl(rlast->addr32[0]) >= ntohl(cmplast->addr32[0])) + return 1; + } else { + if (memcmp(rfirst->addr8, cmpfirst->addr8, sizeof(union netdata_ip_t)) <= 0 && + memcmp(rlast->addr8, cmplast->addr8, sizeof(union netdata_ip_t)) >= 0) { + return 1; + } + + } + return 0; +} + +/** + * Fill IP list + * + * @param out a pointer to the link list. + * @param in the structure that will be linked. + */ +void fill_ip_list(ebpf_network_viewer_ip_list_t **out, ebpf_network_viewer_ip_list_t *in, char *table) +{ +#ifndef NETDATA_INTERNAL_CHECKS + UNUSED(table); +#endif + if (likely(*out)) { + ebpf_network_viewer_ip_list_t *move = *out, *store = *out; + while (move) { + if (in->ver == move->ver && is_ip_inside_range(&move->first, &move->last, &in->first, &in->last, in->ver)) { + info("The range/value (%s) is inside the range/value (%s) already inserted, it will be ignored.", + in->value, move->value); + freez(in->value); + freez(in); + return; + } + store = move; + move = move->next; + } + + store->next = in; + } else { + *out = in; + } + +#ifdef NETDATA_INTERNAL_CHECKS + char first[512], last[512]; + if (in->ver == AF_INET) { + if (inet_ntop(AF_INET, in->first.addr8, first, INET_ADDRSTRLEN) && + inet_ntop(AF_INET, in->last.addr8, last, INET_ADDRSTRLEN)) + info("Adding values %s - %s to %s IP list \"%s\" used on network viewer", + first, last, + (*out == network_viewer_opt.included_ips)?"included":"excluded", + table); + } else { + if (inet_ntop(AF_INET6, in->first.addr8, first, INET6_ADDRSTRLEN) && + inet_ntop(AF_INET6, in->last.addr8, last, INET6_ADDRSTRLEN)) + info("Adding values %s - %s to %s IP list \"%s\" used on network viewer", + first, last, + (*out == network_viewer_opt.included_ips)?"included":"excluded", + table); + } +#endif +} + +/** + * Parse IP List + * + * Parse IP list and link it. + * + * @param out a pointer to store the link list + * @param ip the value given as parameter + */ +static void parse_ip_list(void **out, char *ip) +{ + ebpf_network_viewer_ip_list_t **list = (ebpf_network_viewer_ip_list_t **)out; + + char *ipdup = strdupz(ip); + union netdata_ip_t first = { }; + union netdata_ip_t last = { }; + char *is_ipv6; + if (*ip == '*' && *(ip+1) == '\0') { + memset(first.addr8, 0, sizeof(first.addr8)); + memset(last.addr8, 0xFF, sizeof(last.addr8)); + + is_ipv6 = ip; + + clean_ip_structure(list); + goto storethisip; + } + + char *end = ip; + // Move while I cannot find a separator + while (*end && *end != '/' && *end != '-') end++; + + // We will use only the classic IPV6 for while, but we could consider the base 85 in a near future + // https://tools.ietf.org/html/rfc1924 + is_ipv6 = strchr(ip, ':'); + + int select; + if (*end && !is_ipv6) { // IPV4 range + select = (*end == '/') ? 0 : 1; + *end++ = '\0'; + if (*end == '!') { + info("The exclusion cannot be in the second part of the range %s, it will be ignored.", ipdup); + goto cleanipdup; + } + + if (!select) { // CIDR + select = ip2nl(first.addr8, ip, AF_INET, ipdup); + if (select) + goto cleanipdup; + + select = (int) str2i(end); + if (select < NETDATA_MINIMUM_IPV4_CIDR || select > NETDATA_MAXIMUM_IPV4_CIDR) { + info("The specified CIDR %s is not valid, the IP %s will be ignored.", end, ip); + goto cleanipdup; + } + + last.addr32[0] = htonl(broadcast(ntohl(first.addr32[0]), select)); + // This was added to remove + // https://app.codacy.com/manual/netdata/netdata/pullRequest?prid=5810941&bid=19021977 + UNUSED(last.addr32[0]); + + uint32_t ipv4_test = htonl(ipv4_network(ntohl(first.addr32[0]), select)); + if (first.addr32[0] != ipv4_test) { + first.addr32[0] = ipv4_test; + struct in_addr ipv4_convert; + ipv4_convert.s_addr = ipv4_test; + char ipv4_msg[INET_ADDRSTRLEN]; + if(inet_ntop(AF_INET, &ipv4_convert, ipv4_msg, INET_ADDRSTRLEN)) + info("The network value of CIDR %s was updated for %s .", ipdup, ipv4_msg); + } + } else { // Range + select = ip2nl(first.addr8, ip, AF_INET, ipdup); + if (select) + goto cleanipdup; + + select = ip2nl(last.addr8, end, AF_INET, ipdup); + if (select) + goto cleanipdup; + } + + if (htonl(first.addr32[0]) > htonl(last.addr32[0])) { + info("The specified range %s is invalid, the second address is smallest than the first, it will be ignored.", + ipdup); + goto cleanipdup; + } + } else if (is_ipv6) { // IPV6 + if (!*end) { // Unique + select = ip2nl(first.addr8, ip, AF_INET6, ipdup); + if (select) + goto cleanipdup; + + memcpy(last.addr8, first.addr8, sizeof(first.addr8)); + } else if (*end == '-') { + *end++ = 0x00; + if (*end == '!') { + info("The exclusion cannot be in the second part of the range %s, it will be ignored.", ipdup); + goto cleanipdup; + } + + select = ip2nl(first.addr8, ip, AF_INET6, ipdup); + if (select) + goto cleanipdup; + + select = ip2nl(last.addr8, end, AF_INET6, ipdup); + if (select) + goto cleanipdup; + } else { // CIDR + *end++ = 0x00; + if (*end == '!') { + info("The exclusion cannot be in the second part of the range %s, it will be ignored.", ipdup); + goto cleanipdup; + } + + select = str2i(end); + if (select < 0 || select > 128) { + info("The CIDR %s is not valid, the address %s will be ignored.", end, ip); + goto cleanipdup; + } + + uint64_t prefix = (uint64_t)select; + select = ip2nl(first.addr8, ip, AF_INET6, ipdup); + if (select) + goto cleanipdup; + + get_ipv6_last_addr(&last, &first, prefix); + + union netdata_ip_t ipv6_test; + get_ipv6_first_addr(&ipv6_test, &first, prefix); + + if (memcmp(first.addr8, ipv6_test.addr8, sizeof(union netdata_ip_t)) != 0) { + memcpy(first.addr8, ipv6_test.addr8, sizeof(union netdata_ip_t)); + + struct in6_addr ipv6_convert; + memcpy(ipv6_convert.s6_addr, ipv6_test.addr8, sizeof(union netdata_ip_t)); + + char ipv6_msg[INET6_ADDRSTRLEN]; + if(inet_ntop(AF_INET6, &ipv6_convert, ipv6_msg, INET6_ADDRSTRLEN)) + info("The network value of CIDR %s was updated for %s .", ipdup, ipv6_msg); + } + } + + if ((be64toh(*(uint64_t *)&first.addr32[2]) > be64toh(*(uint64_t *)&last.addr32[2]) && + !memcmp(first.addr32, last.addr32, 2*sizeof(uint32_t))) || + (be64toh(*(uint64_t *)&first.addr32) > be64toh(*(uint64_t *)&last.addr32)) ) { + info("The specified range %s is invalid, the second address is smallest than the first, it will be ignored.", + ipdup); + goto cleanipdup; + } + } else { // Unique ip + select = ip2nl(first.addr8, ip, AF_INET, ipdup); + if (select) + goto cleanipdup; + + memcpy(last.addr8, first.addr8, sizeof(first.addr8)); + } + + ebpf_network_viewer_ip_list_t *store; + + storethisip: + store = callocz(1, sizeof(ebpf_network_viewer_ip_list_t)); + store->value = ipdup; + store->hash = simple_hash(ipdup); + store->ver = (uint8_t)(!is_ipv6)?AF_INET:AF_INET6; + memcpy(store->first.addr8, first.addr8, sizeof(first.addr8)); + memcpy(store->last.addr8, last.addr8, sizeof(last.addr8)); + + fill_ip_list(list, store, "socket"); + return; + +cleanipdup: + freez(ipdup); +} + +/** + * Parse IP Range + * + * Parse the IP ranges given and create Network Viewer IP Structure + * + * @param ptr is a pointer with the text to parse. + */ +static void parse_ips(char *ptr) +{ + // No value + if (unlikely(!ptr)) + return; + + while (likely(ptr)) { + // Move forward until next valid character + while (isspace(*ptr)) ptr++; + + // No valid value found + if (unlikely(!*ptr)) + return; + + // Find space that ends the list + char *end = strchr(ptr, ' '); + if (end) { + *end++ = '\0'; + } + + int neg = 0; + if (*ptr == '!') { + neg++; + ptr++; + } + + if (isascii(*ptr)) { // Parse port + parse_ip_list((!neg)?(void **)&network_viewer_opt.included_ips:(void **)&network_viewer_opt.excluded_ips, + ptr); + } + + ptr = end; + } +} + + + +/** + * Parse port list + * + * Parse an allocated port list with the range given + * + * @param out a pointer to store the link list + * @param range the informed range for the user. + */ +static void parse_port_list(void **out, char *range) +{ + int first, last; + ebpf_network_viewer_port_list_t **list = (ebpf_network_viewer_port_list_t **)out; + + char *copied = strdupz(range); + if (*range == '*' && *(range+1) == '\0') { + first = 1; + last = 65535; + + clean_port_structure(list); + goto fillenvpl; + } + + char *end = range; + //Move while I cannot find a separator + while (*end && *end != ':' && *end != '-') end++; + + //It has a range + if (likely(*end)) { + *end++ = '\0'; + if (*end == '!') { + info("The exclusion cannot be in the second part of the range, the range %s will be ignored.", copied); + freez(copied); + return; + } + last = str2i((const char *)end); + } else { + last = 0; + } + + first = str2i((const char *)range); + if (first < NETDATA_MINIMUM_PORT_VALUE || first > NETDATA_MAXIMUM_PORT_VALUE) { + info("The first port %d of the range \"%s\" is invalid and it will be ignored!", first, copied); + freez(copied); + return; + } + + if (!last) + last = first; + + if (last < NETDATA_MINIMUM_PORT_VALUE || last > NETDATA_MAXIMUM_PORT_VALUE) { + info("The second port %d of the range \"%s\" is invalid and the whole range will be ignored!", last, copied); + freez(copied); + return; + } + + if (first > last) { + info("The specified order %s is wrong, the smallest value is always the first, it will be ignored!", copied); + freez(copied); + return; + } + + ebpf_network_viewer_port_list_t *w; +fillenvpl: + w = callocz(1, sizeof(ebpf_network_viewer_port_list_t)); + w->value = copied; + w->hash = simple_hash(copied); + w->first = (uint16_t)htons((uint16_t)first); + w->last = (uint16_t)htons((uint16_t)last); + w->cmp_first = (uint16_t)first; + w->cmp_last = (uint16_t)last; + + fill_port_list(list, w); +} + +/** + * Read max dimension. + * + * Netdata plot two dimensions per connection, so it is necessary to adjust the values. + * + * @param cfg the configuration structure + */ +static void read_max_dimension(struct config *cfg) +{ + int maxdim ; + maxdim = (int) appconfig_get_number(cfg, + EBPF_NETWORK_VIEWER_SECTION, + EBPF_MAXIMUM_DIMENSIONS, + NETDATA_NV_CAP_VALUE); + if (maxdim < 0) { + error("'maximum dimensions = %d' must be a positive number, Netdata will change for default value %ld.", + maxdim, NETDATA_NV_CAP_VALUE); + maxdim = NETDATA_NV_CAP_VALUE; + } + + maxdim /= 2; + if (!maxdim) { + info("The number of dimensions is too small (%u), we are setting it to minimum 2", network_viewer_opt.max_dim); + network_viewer_opt.max_dim = 1; + return; + } + + network_viewer_opt.max_dim = (uint32_t)maxdim; +} + +/** + * Parse Port Range + * + * Parse the port ranges given and create Network Viewer Port Structure + * + * @param ptr is a pointer with the text to parse. + */ +static void parse_ports(char *ptr) +{ + // No value + if (unlikely(!ptr)) + return; + + while (likely(ptr)) { + // Move forward until next valid character + while (isspace(*ptr)) ptr++; + + // No valid value found + if (unlikely(!*ptr)) + return; + + // Find space that ends the list + char *end = strchr(ptr, ' '); + if (end) { + *end++ = '\0'; + } + + int neg = 0; + if (*ptr == '!') { + neg++; + ptr++; + } + + if (isdigit(*ptr)) { // Parse port + parse_port_list((!neg)?(void **)&network_viewer_opt.included_port:(void **)&network_viewer_opt.excluded_port, + ptr); + } else if (isalpha(*ptr)) { // Parse service + parse_service_list((!neg)?(void **)&network_viewer_opt.included_port:(void **)&network_viewer_opt.excluded_port, + ptr); + } else if (*ptr == '*') { // All + parse_port_list((!neg)?(void **)&network_viewer_opt.included_port:(void **)&network_viewer_opt.excluded_port, + ptr); + } + + ptr = end; + } +} + +/** + * Link hostname + * + * @param out is the output link list + * @param in the hostname to add to list. + */ +static void link_hostname(ebpf_network_viewer_hostname_list_t **out, ebpf_network_viewer_hostname_list_t *in) +{ + if (likely(*out)) { + ebpf_network_viewer_hostname_list_t *move = *out; + for (; move->next ; move = move->next ) { + if (move->hash == in->hash && !strcmp(move->value, in->value)) { + info("The hostname %s was already inserted, it will be ignored.", in->value); + freez(in->value); + simple_pattern_free(in->value_pattern); + freez(in); + return; + } + } + + move->next = in; + } else { + *out = in; + } +#ifdef NETDATA_INTERNAL_CHECKS + info("Adding value %s to %s hostname list used on network viewer", + in->value, + (*out == network_viewer_opt.included_hostnames)?"included":"excluded"); +#endif +} + +/** + * Link Hostnames + * + * Parse the list of hostnames to create the link list. + * This is not associated with the IP, because simple patterns like *example* cannot be resolved to IP. + * + * @param out is the output link list + * @param parse is a pointer with the text to parser. + */ +static void link_hostnames(char *parse) +{ + // No value + if (unlikely(!parse)) + return; + + while (likely(parse)) { + // Find the first valid value + while (isspace(*parse)) parse++; + + // No valid value found + if (unlikely(!*parse)) + return; + + // Find space that ends the list + char *end = strchr(parse, ' '); + if (end) { + *end++ = '\0'; + } + + int neg = 0; + if (*parse == '!') { + neg++; + parse++; + } + + ebpf_network_viewer_hostname_list_t *hostname = callocz(1 , sizeof(ebpf_network_viewer_hostname_list_t)); + hostname->value = strdupz(parse); + hostname->hash = simple_hash(parse); + hostname->value_pattern = simple_pattern_create(parse, NULL, SIMPLE_PATTERN_EXACT); + + link_hostname((!neg)?&network_viewer_opt.included_hostnames:&network_viewer_opt.excluded_hostnames, + hostname); + + parse = end; + } +} + +/** + * Parse network viewer section + * + * @param cfg the configuration structure + */ +void parse_network_viewer_section(struct config *cfg) +{ + read_max_dimension(cfg); + + network_viewer_opt.hostname_resolution_enabled = appconfig_get_boolean(cfg, + EBPF_NETWORK_VIEWER_SECTION, + EBPF_CONFIG_RESOLVE_HOSTNAME, + CONFIG_BOOLEAN_NO); + + network_viewer_opt.service_resolution_enabled = appconfig_get_boolean(cfg, + EBPF_NETWORK_VIEWER_SECTION, + EBPF_CONFIG_RESOLVE_SERVICE, + CONFIG_BOOLEAN_NO); + + char *value = appconfig_get(cfg, EBPF_NETWORK_VIEWER_SECTION, EBPF_CONFIG_PORTS, NULL); + parse_ports(value); + + if (network_viewer_opt.hostname_resolution_enabled) { + value = appconfig_get(cfg, EBPF_NETWORK_VIEWER_SECTION, EBPF_CONFIG_HOSTNAMES, NULL); + link_hostnames(value); + } else { + info("Name resolution is disabled, collector will not parser \"hostnames\" list."); + } + + value = appconfig_get(cfg, EBPF_NETWORK_VIEWER_SECTION, + "ips", "!127.0.0.1/8 10.0.0.0/8 172.16.0.0/12 192.168.0.0/16 fc00::/7 !::1/128"); + parse_ips(value); +} + +/** + * Link dimension name + * + * Link user specified names inside a link list. + * + * @param port the port number associated to the dimension name. + * @param hash the calculated hash for the dimension name. + * @param name the dimension name. + */ +static void link_dimension_name(char *port, uint32_t hash, char *value) +{ + int test = str2i(port); + if (test < NETDATA_MINIMUM_PORT_VALUE || test > NETDATA_MAXIMUM_PORT_VALUE){ + error("The dimension given (%s = %s) has an invalid value and it will be ignored.", port, value); + return; + } + + ebpf_network_viewer_dim_name_t *w; + w = callocz(1, sizeof(ebpf_network_viewer_dim_name_t)); + + w->name = strdupz(value); + w->hash = hash; + + w->port = (uint16_t) htons(test); + + ebpf_network_viewer_dim_name_t *names = network_viewer_opt.names; + if (unlikely(!names)) { + network_viewer_opt.names = w; + } else { + for (; names->next; names = names->next) { + if (names->port == w->port) { + info("Duplicated definition for a service, the name %s will be ignored. ", names->name); + freez(names->name); + names->name = w->name; + names->hash = w->hash; + freez(w); + return; + } + } + names->next = w; + } + +#ifdef NETDATA_INTERNAL_CHECKS + info("Adding values %s( %u) to dimension name list used on network viewer", w->name, htons(w->port)); +#endif +} + +/** + * Parse service Name section. + * + * This function gets the values that will be used to overwrite dimensions. + * + * @param cfg the configuration structure + */ +void parse_service_name_section(struct config *cfg) +{ + struct section *co = appconfig_get_section(cfg, EBPF_SERVICE_NAME_SECTION); + if (co) { + struct config_option *cv; + for (cv = co->values; cv ; cv = cv->next) { + link_dimension_name(cv->name, cv->hash, cv->value); + } + } + + // Always associated the default port to Netdata + ebpf_network_viewer_dim_name_t *names = network_viewer_opt.names; + if (names) { + uint16_t default_port = htons(19999); + while (names) { + if (names->port == default_port) + return; + + names = names->next; + } + } + + char *port_string = getenv("NETDATA_LISTEN_PORT"); + if (port_string) { + // if variable has an invalid value, we assume netdata is using 19999 + int default_port = str2i(port_string); + if (default_port > 0 && default_port < 65536) + link_dimension_name(port_string, simple_hash(port_string), "Netdata"); + } +} + +void parse_table_size_options(struct config *cfg) +{ + socket_maps[NETDATA_SOCKET_TABLE_BANDWIDTH].user_input = (uint32_t) appconfig_get_number(cfg, + EBPF_GLOBAL_SECTION, + EBPF_CONFIG_BANDWIDTH_SIZE, NETDATA_MAXIMUM_CONNECTIONS_ALLOWED); + + socket_maps[NETDATA_SOCKET_TABLE_IPV4].user_input = (uint32_t) appconfig_get_number(cfg, + EBPF_GLOBAL_SECTION, + EBPF_CONFIG_IPV4_SIZE, NETDATA_MAXIMUM_CONNECTIONS_ALLOWED); + + socket_maps[NETDATA_SOCKET_TABLE_IPV6].user_input = (uint32_t) appconfig_get_number(cfg, + EBPF_GLOBAL_SECTION, + EBPF_CONFIG_IPV6_SIZE, NETDATA_MAXIMUM_CONNECTIONS_ALLOWED); + + socket_maps[NETDATA_SOCKET_TABLE_UDP].user_input = (uint32_t) appconfig_get_number(cfg, + EBPF_GLOBAL_SECTION, + EBPF_CONFIG_UDP_SIZE, NETDATA_MAXIMUM_UDP_CONNECTIONS_ALLOWED); +} + +/* + * Load BPF + * + * Load BPF files. + * + * @param em the structure with configuration + */ +static int ebpf_socket_load_bpf(ebpf_module_t *em) +{ + int ret = 0; + + if (em->load & EBPF_LOAD_LEGACY) { + em->probe_links = ebpf_load_program(ebpf_plugin_dir, em, running_on_kernel, isrh, &em->objects); + if (!em->probe_links) { + ret = -1; + } + } +#ifdef LIBBPF_MAJOR_VERSION + else { + bpf_obj = socket_bpf__open(); + if (!bpf_obj) + ret = -1; + else + ret = ebpf_socket_load_and_attach(bpf_obj, em); + } +#endif + + if (ret) { + error("%s %s", EBPF_DEFAULT_ERROR_MSG, em->thread_name); + } + + return ret; +} + +/** + * Socket thread + * + * Thread used to generate socket charts. + * + * @param ptr a pointer to `struct ebpf_module` + * + * @return It always return NULL + */ +void *ebpf_socket_thread(void *ptr) +{ + netdata_thread_cleanup_push(ebpf_socket_exit, ptr); + + memset(&inbound_vectors.tree, 0, sizeof(avl_tree_lock)); + memset(&outbound_vectors.tree, 0, sizeof(avl_tree_lock)); + avl_init_lock(&inbound_vectors.tree, compare_sockets); + avl_init_lock(&outbound_vectors.tree, compare_sockets); + + ebpf_module_t *em = (ebpf_module_t *)ptr; + em->maps = socket_maps; + + parse_network_viewer_section(&socket_config); + parse_service_name_section(&socket_config); + parse_table_size_options(&socket_config); + + if (pthread_mutex_init(&nv_mutex, NULL)) { + em->thread->enabled = NETDATA_THREAD_EBPF_STOPPED; + error("Cannot initialize local mutex"); + goto endsocket; + } + + ebpf_socket_allocate_global_vectors(em->apps_charts); + initialize_inbound_outbound(); + + if (running_on_kernel < NETDATA_EBPF_KERNEL_5_0) + em->mode = MODE_ENTRY; + +#ifdef LIBBPF_MAJOR_VERSION + ebpf_adjust_thread_load(em, default_btf); +#endif + if (ebpf_socket_load_bpf(em)) { + em->enabled = CONFIG_BOOLEAN_NO; + pthread_mutex_unlock(&lock); + goto endsocket; + } + + int algorithms[NETDATA_MAX_SOCKET_VECTOR] = { + NETDATA_EBPF_ABSOLUTE_IDX, NETDATA_EBPF_ABSOLUTE_IDX, NETDATA_EBPF_ABSOLUTE_IDX, + NETDATA_EBPF_ABSOLUTE_IDX, NETDATA_EBPF_ABSOLUTE_IDX, NETDATA_EBPF_ABSOLUTE_IDX, + NETDATA_EBPF_ABSOLUTE_IDX, NETDATA_EBPF_ABSOLUTE_IDX, NETDATA_EBPF_INCREMENTAL_IDX, + NETDATA_EBPF_INCREMENTAL_IDX + }; + ebpf_global_labels( + socket_aggregated_data, socket_publish_aggregated, socket_dimension_names, socket_id_names, + algorithms, NETDATA_MAX_SOCKET_VECTOR); + + pthread_mutex_lock(&lock); + ebpf_create_global_charts(em); + + ebpf_update_stats(&plugin_statistics, em); + + pthread_mutex_unlock(&lock); + + socket_collector((usec_t)(em->update_every * USEC_PER_SEC), em); + +endsocket: + ebpf_update_disabled_plugin_stats(em); + + netdata_thread_cleanup_pop(1); + return NULL; +} diff --git a/collectors/ebpf.plugin/ebpf_socket.h b/collectors/ebpf.plugin/ebpf_socket.h new file mode 100644 index 0000000..ca6b193 --- /dev/null +++ b/collectors/ebpf.plugin/ebpf_socket.h @@ -0,0 +1,370 @@ +// SPDX-License-Identifier: GPL-3.0-or-later +#ifndef NETDATA_EBPF_SOCKET_H +#define NETDATA_EBPF_SOCKET_H 1 +#include <stdint.h> +#include "libnetdata/avl/avl.h" + +// Module name +#define NETDATA_EBPF_MODULE_NAME_SOCKET "socket" + +// Vector indexes +#define NETDATA_UDP_START 3 + +#define NETDATA_SOCKET_READ_SLEEP_MS 800000ULL + +// config file +#define NETDATA_NETWORK_CONFIG_FILE "network.conf" +#define EBPF_NETWORK_VIEWER_SECTION "network connections" +#define EBPF_SERVICE_NAME_SECTION "service name" +#define EBPF_CONFIG_RESOLVE_HOSTNAME "resolve hostnames" +#define EBPF_CONFIG_RESOLVE_SERVICE "resolve service names" +#define EBPF_CONFIG_PORTS "ports" +#define EBPF_CONFIG_HOSTNAMES "hostnames" +#define EBPF_CONFIG_BANDWIDTH_SIZE "bandwidth table size" +#define EBPF_CONFIG_IPV4_SIZE "ipv4 connection table size" +#define EBPF_CONFIG_IPV6_SIZE "ipv6 connection table size" +#define EBPF_CONFIG_UDP_SIZE "udp connection table size" +#define EBPF_MAXIMUM_DIMENSIONS "maximum dimensions" + +enum ebpf_socket_table_list { + NETDATA_SOCKET_TABLE_BANDWIDTH, + NETDATA_SOCKET_GLOBAL, + NETDATA_SOCKET_LPORTS, + NETDATA_SOCKET_TABLE_IPV4, + NETDATA_SOCKET_TABLE_IPV6, + NETDATA_SOCKET_TABLE_UDP, + NETDATA_SOCKET_TABLE_CTRL +}; + +enum ebpf_socket_publish_index { + NETDATA_IDX_TCP_SENDMSG, + NETDATA_IDX_TCP_CLEANUP_RBUF, + NETDATA_IDX_TCP_CLOSE, + NETDATA_IDX_UDP_RECVBUF, + NETDATA_IDX_UDP_SENDMSG, + NETDATA_IDX_TCP_RETRANSMIT, + NETDATA_IDX_TCP_CONNECTION_V4, + NETDATA_IDX_TCP_CONNECTION_V6, + NETDATA_IDX_INCOMING_CONNECTION_TCP, + NETDATA_IDX_INCOMING_CONNECTION_UDP, + + // Keep this as last and don't skip numbers as it is used as element counter + NETDATA_MAX_SOCKET_VECTOR +}; + +enum socket_functions { + NETDATA_FCNT_INET_CSK_ACCEPT, + NETDATA_FCNT_TCP_RETRANSMIT, + NETDATA_FCNT_CLEANUP_RBUF, + NETDATA_FCNT_TCP_CLOSE, + NETDATA_FCNT_UDP_RECEVMSG, + NETDATA_FCNT_TCP_SENDMSG, + NETDATA_FCNT_UDP_SENDMSG, + NETDATA_FCNT_TCP_V4_CONNECT, + NETDATA_FCNT_TCP_V6_CONNECT +}; + +typedef enum ebpf_socket_idx { + NETDATA_KEY_CALLS_TCP_SENDMSG, + NETDATA_KEY_ERROR_TCP_SENDMSG, + NETDATA_KEY_BYTES_TCP_SENDMSG, + + NETDATA_KEY_CALLS_TCP_CLEANUP_RBUF, + NETDATA_KEY_ERROR_TCP_CLEANUP_RBUF, + NETDATA_KEY_BYTES_TCP_CLEANUP_RBUF, + + NETDATA_KEY_CALLS_TCP_CLOSE, + + NETDATA_KEY_CALLS_UDP_RECVMSG, + NETDATA_KEY_ERROR_UDP_RECVMSG, + NETDATA_KEY_BYTES_UDP_RECVMSG, + + NETDATA_KEY_CALLS_UDP_SENDMSG, + NETDATA_KEY_ERROR_UDP_SENDMSG, + NETDATA_KEY_BYTES_UDP_SENDMSG, + + NETDATA_KEY_TCP_RETRANSMIT, + + NETDATA_KEY_CALLS_TCP_CONNECT_IPV4, + NETDATA_KEY_ERROR_TCP_CONNECT_IPV4, + + NETDATA_KEY_CALLS_TCP_CONNECT_IPV6, + NETDATA_KEY_ERROR_TCP_CONNECT_IPV6, + + // Keep this as last and don't skip numbers as it is used as element counter + NETDATA_SOCKET_COUNTER +} ebpf_socket_index_t; + +#define NETDATA_SOCKET_KERNEL_FUNCTIONS "kernel" +#define NETDATA_NETWORK_CONNECTIONS_GROUP "network connections" +#define NETDATA_CGROUP_NET_GROUP "network (eBPF)" + +// Global chart name +#define NETDATA_TCP_OUTBOUND_CONNECTIONS "tcp_outbound_conn" +#define NETDATA_INBOUND_CONNECTIONS "inbound_conn" +#define NETDATA_TCP_FUNCTION_COUNT "tcp_functions" +#define NETDATA_TCP_FUNCTION_BITS "total_tcp_bandwidth" +#define NETDATA_TCP_FUNCTION_ERROR "tcp_error" +#define NETDATA_TCP_RETRANSMIT "tcp_retransmit" +#define NETDATA_UDP_FUNCTION_COUNT "udp_functions" +#define NETDATA_UDP_FUNCTION_BITS "total_udp_bandwidth" +#define NETDATA_UDP_FUNCTION_ERROR "udp_error" + +// Charts created on Apps submenu +#define NETDATA_NET_APPS_CONNECTION_TCP_V4 "outbound_conn_v4" +#define NETDATA_NET_APPS_CONNECTION_TCP_V6 "outbound_conn_v6" +#define NETDATA_NET_APPS_BANDWIDTH_SENT "total_bandwidth_sent" +#define NETDATA_NET_APPS_BANDWIDTH_RECV "total_bandwidth_recv" +#define NETDATA_NET_APPS_BANDWIDTH_TCP_SEND_CALLS "bandwidth_tcp_send" +#define NETDATA_NET_APPS_BANDWIDTH_TCP_RECV_CALLS "bandwidth_tcp_recv" +#define NETDATA_NET_APPS_BANDWIDTH_TCP_RETRANSMIT "bandwidth_tcp_retransmit" +#define NETDATA_NET_APPS_BANDWIDTH_UDP_SEND_CALLS "bandwidth_udp_send" +#define NETDATA_NET_APPS_BANDWIDTH_UDP_RECV_CALLS "bandwidth_udp_recv" + +// Network viewer charts +#define NETDATA_NV_OUTBOUND_BYTES "outbound_bytes" +#define NETDATA_NV_OUTBOUND_PACKETS "outbound_packets" +#define NETDATA_NV_OUTBOUND_RETRANSMIT "outbound_retransmit" +#define NETDATA_NV_INBOUND_BYTES "inbound_bytes" +#define NETDATA_NV_INBOUND_PACKETS "inbound_packets" + +// Port range +#define NETDATA_MINIMUM_PORT_VALUE 1 +#define NETDATA_MAXIMUM_PORT_VALUE 65535 +#define NETDATA_COMPILED_CONNECTIONS_ALLOWED 65535U +#define NETDATA_MAXIMUM_CONNECTIONS_ALLOWED 16384U +#define NETDATA_COMPILED_UDP_CONNECTIONS_ALLOWED 8192U +#define NETDATA_MAXIMUM_UDP_CONNECTIONS_ALLOWED 4096U + +#define NETDATA_MINIMUM_IPV4_CIDR 0 +#define NETDATA_MAXIMUM_IPV4_CIDR 32 + +// Contexts +#define NETDATA_CGROUP_TCP_V4_CONN_CONTEXT "cgroup.net_conn_ipv4" +#define NETDATA_CGROUP_TCP_V6_CONN_CONTEXT "cgroup.net_conn_ipv6" +#define NETDATA_CGROUP_SOCKET_BYTES_RECV_CONTEXT "cgroup.net_bytes_recv" +#define NETDATA_CGROUP_SOCKET_BYTES_SEND_CONTEXT "cgroup.net_bytes_send" +#define NETDATA_CGROUP_SOCKET_TCP_RECV_CONTEXT "cgroup.net_tcp_recv" +#define NETDATA_CGROUP_SOCKET_TCP_SEND_CONTEXT "cgroup.net_tcp_send" +#define NETDATA_CGROUP_SOCKET_TCP_RETRANSMIT_CONTEXT "cgroup.net_retransmit" +#define NETDATA_CGROUP_SOCKET_UDP_RECV_CONTEXT "cgroup.net_udp_recv" +#define NETDATA_CGROUP_SOCKET_UDP_SEND_CONTEXT "cgroup.net_udp_send" + +#define NETDATA_SERVICES_SOCKET_TCP_V4_CONN_CONTEXT "services.net_conn_ipv4" +#define NETDATA_SERVICES_SOCKET_TCP_V6_CONN_CONTEXT "services.net_conn_ipv6" +#define NETDATA_SERVICES_SOCKET_BYTES_RECV_CONTEXT "services.net_bytes_recv" +#define NETDATA_SERVICES_SOCKET_BYTES_SEND_CONTEXT "services.net_bytes_send" +#define NETDATA_SERVICES_SOCKET_TCP_RECV_CONTEXT "services.net_tcp_recv" +#define NETDATA_SERVICES_SOCKET_TCP_SEND_CONTEXT "services.net_tcp_send" +#define NETDATA_SERVICES_SOCKET_TCP_RETRANSMIT_CONTEXT "services.net_retransmit" +#define NETDATA_SERVICES_SOCKET_UDP_RECV_CONTEXT "services.net_udp_recv" +#define NETDATA_SERVICES_SOCKET_UDP_SEND_CONTEXT "services.net_udp_send" + +typedef struct ebpf_socket_publish_apps { + // Data read + uint64_t bytes_sent; // Bytes sent + uint64_t bytes_received; // Bytes received + uint64_t call_tcp_sent; // Number of times tcp_sendmsg was called + uint64_t call_tcp_received; // Number of times tcp_cleanup_rbuf was called + uint64_t retransmit; // Number of times tcp_retransmit was called + uint64_t call_udp_sent; // Number of times udp_sendmsg was called + uint64_t call_udp_received; // Number of times udp_recvmsg was called + uint64_t call_close; // Number of times tcp_close was called + uint64_t call_tcp_v4_connection;// Number of times tcp_v4_connect was called + uint64_t call_tcp_v6_connection;// Number of times tcp_v6_connect was called +} ebpf_socket_publish_apps_t; + +typedef struct ebpf_network_viewer_dimension_names { + char *name; + uint32_t hash; + + uint16_t port; + + struct ebpf_network_viewer_dimension_names *next; +} ebpf_network_viewer_dim_name_t ; + +typedef struct ebpf_network_viewer_port_list { + char *value; + uint32_t hash; + + uint16_t first; + uint16_t last; + + uint16_t cmp_first; + uint16_t cmp_last; + + uint16_t protocol; + uint32_t pid; + uint32_t tgid; + uint64_t connections; + struct ebpf_network_viewer_port_list *next; +} ebpf_network_viewer_port_list_t; + +typedef struct netdata_passive_connection { + uint32_t tgid; + uint32_t pid; + uint64_t counter; +} netdata_passive_connection_t; + +typedef struct netdata_passive_connection_idx { + uint16_t protocol; + uint16_t port; +} netdata_passive_connection_idx_t; + +/** + * Union used to store ip addresses + */ +union netdata_ip_t { + uint8_t addr8[16]; + uint16_t addr16[8]; + uint32_t addr32[4]; + uint64_t addr64[2]; +}; + +typedef struct ebpf_network_viewer_ip_list { + char *value; // IP value + uint32_t hash; // IP hash + + uint8_t ver; // IP version + + union netdata_ip_t first; // The IP address informed + union netdata_ip_t last; // The IP address informed + + struct ebpf_network_viewer_ip_list *next; +} ebpf_network_viewer_ip_list_t; + +typedef struct ebpf_network_viewer_hostname_list { + char *value; // IP value + uint32_t hash; // IP hash + + SIMPLE_PATTERN *value_pattern; + + struct ebpf_network_viewer_hostname_list *next; +} ebpf_network_viewer_hostname_list_t; + +#define NETDATA_NV_CAP_VALUE 50L +typedef struct ebpf_network_viewer_options { + uint32_t max_dim; // Store value read from 'maximum dimensions' + + uint32_t hostname_resolution_enabled; + uint32_t service_resolution_enabled; + + ebpf_network_viewer_port_list_t *excluded_port; + ebpf_network_viewer_port_list_t *included_port; + + ebpf_network_viewer_dim_name_t *names; + + ebpf_network_viewer_ip_list_t *excluded_ips; + ebpf_network_viewer_ip_list_t *included_ips; + + ebpf_network_viewer_hostname_list_t *excluded_hostnames; + ebpf_network_viewer_hostname_list_t *included_hostnames; + + ebpf_network_viewer_ip_list_t *ipv4_local_ip; + ebpf_network_viewer_ip_list_t *ipv6_local_ip; +} ebpf_network_viewer_options_t; + +extern ebpf_network_viewer_options_t network_viewer_opt; + +/** + * Structure to store socket information + */ +typedef struct netdata_socket { + uint64_t recv_packets; + uint64_t sent_packets; + uint64_t recv_bytes; + uint64_t sent_bytes; + uint64_t first; // First timestamp + uint64_t ct; // Current timestamp + uint32_t retransmit; // It is never used with UDP + uint16_t protocol; + uint16_t reserved; +} netdata_socket_t; + +typedef struct netdata_plot_values { + // Values used in the previous iteration + uint64_t recv_packets; + uint64_t sent_packets; + uint64_t recv_bytes; + uint64_t sent_bytes; + uint32_t retransmit; + + uint64_t last_time; + + // Values used to plot + uint64_t plot_recv_packets; + uint64_t plot_sent_packets; + uint64_t plot_recv_bytes; + uint64_t plot_sent_bytes; + uint16_t plot_retransmit; +} netdata_plot_values_t; + +/** + * Index used together previous structure + */ +typedef struct netdata_socket_idx { + union netdata_ip_t saddr; + uint16_t sport; + union netdata_ip_t daddr; + uint16_t dport; +} netdata_socket_idx_t; + +// Next values were defined according getnameinfo(3) +#define NETDATA_MAX_NETWORK_COMBINED_LENGTH 1018 +#define NETDATA_DOTS_PROTOCOL_COMBINED_LENGTH 5 // :TCP: +#define NETDATA_DIM_LENGTH_WITHOUT_SERVICE_PROTOCOL 979 + +#define NETDATA_INBOUND_DIRECTION (uint32_t)1 +#define NETDATA_OUTBOUND_DIRECTION (uint32_t)2 +/** + * Allocate the maximum number of structures in the beginning, this can force the collector to use more memory + * in the long term, on the other had it is faster. + */ +typedef struct netdata_socket_plot { + // Search + avl_t avl; + netdata_socket_idx_t index; + + // Current data + netdata_socket_t sock; + + // Previous values and values used to write on chart. + netdata_plot_values_t plot; + + int family; // AF_INET or AF_INET6 + char *resolved_name; // Resolve only in the first call + unsigned char resolved; + + char *dimension_sent; + char *dimension_recv; + char *dimension_retransmit; + + uint32_t flags; +} netdata_socket_plot_t; + +#define NETWORK_VIEWER_CHARTS_CREATED (uint32_t)1 +typedef struct netdata_vector_plot { + netdata_socket_plot_t *plot; // Vector used to plot charts + + avl_tree_lock tree; // AVL tree to speed up search + uint32_t last; // The 'other' dimension, the last chart accepted. + uint32_t next; // The next position to store in the vector. + uint32_t max_plot; // Max number of elements to plot. + uint32_t last_plot; // Last element plot + + uint32_t flags; // Flags + +} netdata_vector_plot_t; + +void clean_port_structure(ebpf_network_viewer_port_list_t **clean); +extern ebpf_network_viewer_port_list_t *listen_ports; +void update_listen_table(uint16_t value, uint16_t proto, netdata_passive_connection_t *values); +void parse_network_viewer_section(struct config *cfg); +void fill_ip_list(ebpf_network_viewer_ip_list_t **out, ebpf_network_viewer_ip_list_t *in, char *table); +void parse_service_name_section(struct config *cfg); + +extern ebpf_socket_publish_apps_t **socket_bandwidth_curr; +extern struct config socket_config; +extern netdata_ebpf_targets_t socket_targets[]; + +#endif diff --git a/collectors/ebpf.plugin/ebpf_softirq.c b/collectors/ebpf.plugin/ebpf_softirq.c new file mode 100644 index 0000000..3b5d159 --- /dev/null +++ b/collectors/ebpf.plugin/ebpf_softirq.c @@ -0,0 +1,290 @@ +// SPDX-License-Identifier: GPL-3.0-or-later + +#include "ebpf.h" +#include "ebpf_softirq.h" + +struct config softirq_config = { .first_section = NULL, + .last_section = NULL, + .mutex = NETDATA_MUTEX_INITIALIZER, + .index = { .avl_tree = { .root = NULL, .compar = appconfig_section_compare }, + .rwlock = AVL_LOCK_INITIALIZER } }; + +#define SOFTIRQ_MAP_LATENCY 0 +static ebpf_local_maps_t softirq_maps[] = { + { + .name = "tbl_softirq", + .internal_input = NETDATA_SOFTIRQ_MAX_IRQS, + .user_input = 0, + .type = NETDATA_EBPF_MAP_STATIC, + .map_fd = ND_EBPF_MAP_FD_NOT_INITIALIZED + }, + /* end */ + { + .name = NULL, + .internal_input = 0, + .user_input = 0, + .type = NETDATA_EBPF_MAP_CONTROLLER, + .map_fd = ND_EBPF_MAP_FD_NOT_INITIALIZED + } +}; + +#define SOFTIRQ_TP_CLASS_IRQ "irq" +static ebpf_tracepoint_t softirq_tracepoints[] = { + {.enabled = false, .class = SOFTIRQ_TP_CLASS_IRQ, .event = "softirq_entry"}, + {.enabled = false, .class = SOFTIRQ_TP_CLASS_IRQ, .event = "softirq_exit"}, + /* end */ + {.enabled = false, .class = NULL, .event = NULL} +}; + +// these must be in the order defined by the kernel: +// https://elixir.bootlin.com/linux/v5.12.19/source/include/trace/events/irq.h#L13 +static softirq_val_t softirq_vals[] = { + {.name = "HI", .latency = 0}, + {.name = "TIMER", .latency = 0}, + {.name = "NET_TX", .latency = 0}, + {.name = "NET_RX", .latency = 0}, + {.name = "BLOCK", .latency = 0}, + {.name = "IRQ_POLL", .latency = 0}, + {.name = "TASKLET", .latency = 0}, + {.name = "SCHED", .latency = 0}, + {.name = "HRTIMER", .latency = 0}, + {.name = "RCU", .latency = 0}, +}; + +// tmp store for soft IRQ values we get from a per-CPU eBPF map. +static softirq_ebpf_val_t *softirq_ebpf_vals = NULL; + +static struct netdata_static_thread softirq_threads = { + .name = "SOFTIRQ KERNEL", + .config_section = NULL, + .config_name = NULL, + .env_name = NULL, + .enabled = 1, + .thread = NULL, + .init_routine = NULL, + .start_routine = NULL +}; + +/** + * Cachestat Free + * + * Cleanup variables after child threads to stop + * + * @param ptr thread data. + */ +static void ebpf_softirq_free(ebpf_module_t *em) +{ + pthread_mutex_lock(&ebpf_exit_cleanup); + if (em->thread->enabled == NETDATA_THREAD_EBPF_RUNNING) { + em->thread->enabled = NETDATA_THREAD_EBPF_STOPPING; + pthread_mutex_unlock(&ebpf_exit_cleanup); + return; + } + pthread_mutex_unlock(&ebpf_exit_cleanup); + + freez(softirq_threads.thread); + + for (int i = 0; softirq_tracepoints[i].class != NULL; i++) { + ebpf_disable_tracepoint(&softirq_tracepoints[i]); + } + freez(softirq_ebpf_vals); + + pthread_mutex_lock(&ebpf_exit_cleanup); + em->thread->enabled = NETDATA_MAIN_THREAD_EXITED; + pthread_mutex_unlock(&ebpf_exit_cleanup); +} + +/** + * Exit + * + * Cancel thread. + * + * @param ptr thread data. + */ +static void softirq_exit(void *ptr) +{ + ebpf_module_t *em = (ebpf_module_t *)ptr; + netdata_thread_cancel(*softirq_threads.thread); + ebpf_softirq_free(em); +} + +/** + * Cleanup + * + * Clean up allocated memory. + * + * @param ptr thread data. + */ +static void softirq_cleanup(void *ptr) +{ + ebpf_module_t *em = (ebpf_module_t *)ptr; + ebpf_softirq_free(em); +} + +/***************************************************************** + * MAIN LOOP + *****************************************************************/ + +static void softirq_read_latency_map() +{ + int fd = softirq_maps[SOFTIRQ_MAP_LATENCY].map_fd; + int i; + for (i = 0; i < NETDATA_SOFTIRQ_MAX_IRQS; i++) { + int test = bpf_map_lookup_elem(fd, &i, softirq_ebpf_vals); + if (unlikely(test < 0)) { + continue; + } + + uint64_t total_latency = 0; + int cpu_i; + int end = ebpf_nprocs; + for (cpu_i = 0; cpu_i < end; cpu_i++) { + total_latency += softirq_ebpf_vals[cpu_i].latency/1000; + } + + softirq_vals[i].latency = total_latency; + } +} + +/** + * Read eBPF maps for soft IRQ. + */ +static void *softirq_reader(void *ptr) +{ + netdata_thread_cleanup_push(softirq_exit, ptr); + heartbeat_t hb; + heartbeat_init(&hb); + + ebpf_module_t *em = (ebpf_module_t *)ptr; + + usec_t step = NETDATA_SOFTIRQ_SLEEP_MS * em->update_every; + while (!ebpf_exit_plugin) { + (void)heartbeat_next(&hb, step); + + softirq_read_latency_map(); + } + + netdata_thread_cleanup_pop(1); + return NULL; +} + +static void softirq_create_charts(int update_every) +{ + ebpf_create_chart( + NETDATA_EBPF_SYSTEM_GROUP, + "softirq_latency", + "Software IRQ latency", + EBPF_COMMON_DIMENSION_MILLISECONDS, + "softirqs", + NULL, + NETDATA_EBPF_CHART_TYPE_STACKED, + NETDATA_CHART_PRIO_SYSTEM_SOFTIRQS+1, + NULL, NULL, 0, update_every, + NETDATA_EBPF_MODULE_NAME_SOFTIRQ + ); + + fflush(stdout); +} + +static void softirq_create_dims() +{ + uint32_t i; + for (i = 0; i < NETDATA_SOFTIRQ_MAX_IRQS; i++) { + ebpf_write_global_dimension( + softirq_vals[i].name, softirq_vals[i].name, + ebpf_algorithms[NETDATA_EBPF_INCREMENTAL_IDX] + ); + } +} + +static inline void softirq_write_dims() +{ + uint32_t i; + for (i = 0; i < NETDATA_SOFTIRQ_MAX_IRQS; i++) { + write_chart_dimension(softirq_vals[i].name, softirq_vals[i].latency); + } +} + +/** +* Main loop for this collector. +*/ +static void softirq_collector(ebpf_module_t *em) +{ + softirq_ebpf_vals = callocz(ebpf_nprocs, sizeof(softirq_ebpf_val_t)); + + // create reader thread. + softirq_threads.thread = mallocz(sizeof(netdata_thread_t)); + softirq_threads.start_routine = softirq_reader; + netdata_thread_create( + softirq_threads.thread, + softirq_threads.name, + NETDATA_THREAD_OPTION_DEFAULT, + softirq_reader, + em + ); + + // create chart and static dims. + pthread_mutex_lock(&lock); + softirq_create_charts(em->update_every); + softirq_create_dims(); + ebpf_update_stats(&plugin_statistics, em); + pthread_mutex_unlock(&lock); + + // loop and read from published data until ebpf plugin is closed. + heartbeat_t hb; + heartbeat_init(&hb); + usec_t step = em->update_every * USEC_PER_SEC; + //This will be cancelled by its parent + while (!ebpf_exit_plugin) { + (void)heartbeat_next(&hb, step); + if (ebpf_exit_plugin) + break; + + pthread_mutex_lock(&lock); + + // write dims now for all hitherto discovered IRQs. + write_begin_chart(NETDATA_EBPF_SYSTEM_GROUP, "softirq_latency"); + softirq_write_dims(); + write_end_chart(); + + pthread_mutex_unlock(&lock); + } +} + +/***************************************************************** + * EBPF SOFTIRQ THREAD + *****************************************************************/ + +/** + * Soft IRQ latency thread. + * + * @param ptr a `ebpf_module_t *`. + * @return always NULL. + */ +void *ebpf_softirq_thread(void *ptr) +{ + netdata_thread_cleanup_push(softirq_cleanup, ptr); + + ebpf_module_t *em = (ebpf_module_t *)ptr; + em->maps = softirq_maps; + + if (ebpf_enable_tracepoints(softirq_tracepoints) == 0) { + em->thread->enabled = NETDATA_THREAD_EBPF_STOPPED; + goto endsoftirq; + } + + em->probe_links = ebpf_load_program(ebpf_plugin_dir, em, running_on_kernel, isrh, &em->objects); + if (!em->probe_links) { + em->thread->enabled = NETDATA_THREAD_EBPF_STOPPED; + goto endsoftirq; + } + + softirq_collector(em); + +endsoftirq: + ebpf_update_disabled_plugin_stats(em); + + netdata_thread_cleanup_pop(1); + + return NULL; +} diff --git a/collectors/ebpf.plugin/ebpf_softirq.h b/collectors/ebpf.plugin/ebpf_softirq.h new file mode 100644 index 0000000..7dcddbb --- /dev/null +++ b/collectors/ebpf.plugin/ebpf_softirq.h @@ -0,0 +1,34 @@ +// SPDX-License-Identifier: GPL-3.0-or-later + +#ifndef NETDATA_EBPF_SOFTIRQ_H +#define NETDATA_EBPF_SOFTIRQ_H 1 + +/***************************************************************** + * copied from kernel-collectors repo, with modifications needed + * for inclusion here. + *****************************************************************/ + +#define NETDATA_SOFTIRQ_MAX_IRQS 10 + +typedef struct softirq_ebpf_val { + uint64_t latency; + uint64_t ts; +} softirq_ebpf_val_t; + +/***************************************************************** + * below this is eBPF plugin-specific code. + *****************************************************************/ + +#define NETDATA_EBPF_MODULE_NAME_SOFTIRQ "softirq" +#define NETDATA_SOFTIRQ_SLEEP_MS 650000ULL +#define NETDATA_SOFTIRQ_CONFIG_FILE "softirq.conf" + +typedef struct sofirq_val { + uint64_t latency; + char *name; +} softirq_val_t; + +extern struct config softirq_config; +void *ebpf_softirq_thread(void *ptr); + +#endif /* NETDATA_EBPF_SOFTIRQ_H */ diff --git a/collectors/ebpf.plugin/ebpf_swap.c b/collectors/ebpf.plugin/ebpf_swap.c new file mode 100644 index 0000000..8199573 --- /dev/null +++ b/collectors/ebpf.plugin/ebpf_swap.c @@ -0,0 +1,915 @@ +// SPDX-License-Identifier: GPL-3.0-or-later + +#include "ebpf.h" +#include "ebpf_swap.h" + +static char *swap_dimension_name[NETDATA_SWAP_END] = { "read", "write" }; +static netdata_syscall_stat_t swap_aggregated_data[NETDATA_SWAP_END]; +static netdata_publish_syscall_t swap_publish_aggregated[NETDATA_SWAP_END]; + +netdata_publish_swap_t *swap_vector = NULL; + +static netdata_idx_t swap_hash_values[NETDATA_SWAP_END]; +static netdata_idx_t *swap_values = NULL; + +netdata_publish_swap_t **swap_pid = NULL; + +struct config swap_config = { .first_section = NULL, + .last_section = NULL, + .mutex = NETDATA_MUTEX_INITIALIZER, + .index = { .avl_tree = { .root = NULL, .compar = appconfig_section_compare }, + .rwlock = AVL_LOCK_INITIALIZER } }; + +static ebpf_local_maps_t swap_maps[] = {{.name = "tbl_pid_swap", .internal_input = ND_EBPF_DEFAULT_PID_SIZE, + .user_input = 0, + .type = NETDATA_EBPF_MAP_RESIZABLE | NETDATA_EBPF_MAP_PID, + .map_fd = ND_EBPF_MAP_FD_NOT_INITIALIZED}, + {.name = "swap_ctrl", .internal_input = NETDATA_CONTROLLER_END, + .user_input = 0, + .type = NETDATA_EBPF_MAP_CONTROLLER, + .map_fd = ND_EBPF_MAP_FD_NOT_INITIALIZED}, + {.name = "tbl_swap", .internal_input = NETDATA_SWAP_END, + .user_input = 0, + .type = NETDATA_EBPF_MAP_STATIC, + .map_fd = ND_EBPF_MAP_FD_NOT_INITIALIZED}, + {.name = NULL, .internal_input = 0, .user_input = 0}}; + +struct netdata_static_thread swap_threads = { + .name = "SWAP KERNEL", + .config_section = NULL, + .config_name = NULL, + .env_name = NULL, + .enabled = 1, + .thread = NULL, + .init_routine = NULL, + .start_routine = NULL +}; + +netdata_ebpf_targets_t swap_targets[] = { {.name = "swap_readpage", .mode = EBPF_LOAD_TRAMPOLINE}, + {.name = "swap_writepage", .mode = EBPF_LOAD_TRAMPOLINE}, + {.name = NULL, .mode = EBPF_LOAD_TRAMPOLINE}}; + +#ifdef LIBBPF_MAJOR_VERSION +#include "includes/swap.skel.h" // BTF code + +static struct swap_bpf *bpf_obj = NULL; + +/** + * Disable probe + * + * Disable all probes to use exclusively another method. + * + * @param obj is the main structure for bpf objects + */ +static void ebpf_swap_disable_probe(struct swap_bpf *obj) +{ + bpf_program__set_autoload(obj->progs.netdata_swap_readpage_probe, false); + bpf_program__set_autoload(obj->progs.netdata_swap_writepage_probe, false); +} + +/* + * Disable trampoline + * + * Disable all trampoline to use exclusively another method. + * + * @param obj is the main structure for bpf objects. + */ +static void ebpf_swap_disable_trampoline(struct swap_bpf *obj) +{ + bpf_program__set_autoload(obj->progs.netdata_swap_readpage_fentry, false); + bpf_program__set_autoload(obj->progs.netdata_swap_writepage_fentry, false); + bpf_program__set_autoload(obj->progs.netdata_release_task_fentry, false); +} + +/** + * Set trampoline target + * + * Set the targets we will monitor. + * + * @param obj is the main structure for bpf objects. + */ +static void ebpf_swap_set_trampoline_target(struct swap_bpf *obj) +{ + bpf_program__set_attach_target(obj->progs.netdata_swap_readpage_fentry, 0, + swap_targets[NETDATA_KEY_SWAP_READPAGE_CALL].name); + + bpf_program__set_attach_target(obj->progs.netdata_swap_writepage_fentry, 0, + swap_targets[NETDATA_KEY_SWAP_WRITEPAGE_CALL].name); + + bpf_program__set_attach_target(obj->progs.netdata_release_task_fentry, 0, + EBPF_COMMON_FNCT_CLEAN_UP); +} + +/** + * Mount Attach Probe + * + * Attach probes to target + * + * @param obj is the main structure for bpf objects. + * + * @return It returns 0 on success and -1 otherwise. + */ +static int ebpf_swap_attach_kprobe(struct swap_bpf *obj) +{ + obj->links.netdata_swap_readpage_probe = bpf_program__attach_kprobe(obj->progs.netdata_swap_readpage_probe, + false, + swap_targets[NETDATA_KEY_SWAP_READPAGE_CALL].name); + int ret = libbpf_get_error(obj->links.netdata_swap_readpage_probe); + if (ret) + return -1; + + obj->links.netdata_swap_writepage_probe = bpf_program__attach_kprobe(obj->progs.netdata_swap_writepage_probe, + false, + swap_targets[NETDATA_KEY_SWAP_WRITEPAGE_CALL].name); + ret = libbpf_get_error(obj->links.netdata_swap_writepage_probe); + if (ret) + return -1; + + obj->links.netdata_release_task_probe = bpf_program__attach_kprobe(obj->progs.netdata_release_task_probe, + false, + EBPF_COMMON_FNCT_CLEAN_UP); + ret = libbpf_get_error(obj->links.netdata_swap_writepage_probe); + if (ret) + return -1; + + return 0; +} + +/** + * Set hash tables + * + * Set the values for maps according the value given by kernel. + * + * @param obj is the main structure for bpf objects. + */ +static void ebpf_swap_set_hash_tables(struct swap_bpf *obj) +{ + swap_maps[NETDATA_PID_SWAP_TABLE].map_fd = bpf_map__fd(obj->maps.tbl_pid_swap); + swap_maps[NETDATA_SWAP_CONTROLLER].map_fd = bpf_map__fd(obj->maps.swap_ctrl); + swap_maps[NETDATA_SWAP_GLOBAL_TABLE].map_fd = bpf_map__fd(obj->maps.tbl_swap); +} + +/** + * Adjust Map Size + * + * Resize maps according input from users. + * + * @param obj is the main structure for bpf objects. + * @param em structure with configuration + */ +static void ebpf_swap_adjust_map_size(struct swap_bpf *obj, ebpf_module_t *em) +{ + ebpf_update_map_size(obj->maps.tbl_pid_swap, &swap_maps[NETDATA_PID_SWAP_TABLE], + em, bpf_map__name(obj->maps.tbl_pid_swap)); +} + +/** + * Disable Release Task + * + * Disable release task when apps is not enabled. + * + * @param obj is the main structure for bpf objects. + */ +static void ebpf_swap_disable_release_task(struct swap_bpf *obj) +{ + bpf_program__set_autoload(obj->progs.netdata_release_task_fentry, false); + bpf_program__set_autoload(obj->progs.netdata_release_task_probe, false); +} + +/** + * Load and attach + * + * Load and attach the eBPF code in kernel. + * + * @param obj is the main structure for bpf objects. + * @param em structure with configuration + * + * @return it returns 0 on succes and -1 otherwise + */ +static inline int ebpf_swap_load_and_attach(struct swap_bpf *obj, ebpf_module_t *em) +{ + netdata_ebpf_targets_t *mt = em->targets; + netdata_ebpf_program_loaded_t test = mt[NETDATA_KEY_SWAP_READPAGE_CALL].mode; + + if (test == EBPF_LOAD_TRAMPOLINE) { + ebpf_swap_disable_probe(obj); + + ebpf_swap_set_trampoline_target(obj); + } else { + ebpf_swap_disable_trampoline(obj); + } + + ebpf_swap_adjust_map_size(obj, em); + + if (!em->apps_charts && !em->cgroup_charts) + ebpf_swap_disable_release_task(obj); + + int ret = swap_bpf__load(obj); + if (ret) { + return ret; + } + + ret = (test == EBPF_LOAD_TRAMPOLINE) ? swap_bpf__attach(obj) : ebpf_swap_attach_kprobe(obj); + if (!ret) { + ebpf_swap_set_hash_tables(obj); + + ebpf_update_controller(swap_maps[NETDATA_SWAP_CONTROLLER].map_fd, em); + } + + return ret; +} +#endif + +/***************************************************************** + * + * FUNCTIONS TO CLOSE THE THREAD + * + *****************************************************************/ + +/** + * Cachestat Free + * + * Cleanup variables after child threads to stop + * + * @param ptr thread data. + */ +static void ebpf_swap_free(ebpf_module_t *em) +{ + pthread_mutex_lock(&ebpf_exit_cleanup); + if (em->thread->enabled == NETDATA_THREAD_EBPF_RUNNING) { + em->thread->enabled = NETDATA_THREAD_EBPF_STOPPING; + pthread_mutex_unlock(&ebpf_exit_cleanup); + return; + } + pthread_mutex_unlock(&ebpf_exit_cleanup); + + ebpf_cleanup_publish_syscall(swap_publish_aggregated); + + freez(swap_vector); + freez(swap_values); + freez(swap_threads.thread); + +#ifdef LIBBPF_MAJOR_VERSION + if (bpf_obj) + swap_bpf__destroy(bpf_obj); +#endif + pthread_mutex_lock(&ebpf_exit_cleanup); + em->thread->enabled = NETDATA_THREAD_EBPF_STOPPED; + pthread_mutex_unlock(&ebpf_exit_cleanup); +} + +/** + * Swap exit + * + * Cancel thread and exit. + * + * @param ptr thread data. + */ +static void ebpf_swap_exit(void *ptr) +{ + ebpf_module_t *em = (ebpf_module_t *)ptr; + netdata_thread_cancel(*swap_threads.thread); + ebpf_swap_free(em); +} + +/** + * Swap cleanup + * + * Clean up allocated memory. + * + * @param ptr thread data. + */ +static void ebpf_swap_cleanup(void *ptr) +{ + ebpf_module_t *em = (ebpf_module_t *)ptr; + ebpf_swap_free(em); +} + +/***************************************************************** + * + * COLLECTOR THREAD + * + *****************************************************************/ + +/** + * Apps Accumulator + * + * Sum all values read from kernel and store in the first address. + * + * @param out the vector with read values. + */ +static void swap_apps_accumulator(netdata_publish_swap_t *out) +{ + int i, end = (running_on_kernel >= NETDATA_KERNEL_V4_15) ? ebpf_nprocs : 1; + netdata_publish_swap_t *total = &out[0]; + for (i = 1; i < end; i++) { + netdata_publish_swap_t *w = &out[i]; + total->write += w->write; + total->read += w->read; + } +} + +/** + * Fill PID + * + * Fill PID structures + * + * @param current_pid pid that we are collecting data + * @param out values read from hash tables; + */ +static void swap_fill_pid(uint32_t current_pid, netdata_publish_swap_t *publish) +{ + netdata_publish_swap_t *curr = swap_pid[current_pid]; + if (!curr) { + curr = callocz(1, sizeof(netdata_publish_swap_t)); + swap_pid[current_pid] = curr; + } + + memcpy(curr, publish, sizeof(netdata_publish_swap_t)); +} + +/** + * Update cgroup + * + * Update cgroup data based in + */ +static void ebpf_update_swap_cgroup() +{ + ebpf_cgroup_target_t *ect ; + netdata_publish_swap_t *cv = swap_vector; + int fd = swap_maps[NETDATA_PID_SWAP_TABLE].map_fd; + size_t length = sizeof(netdata_publish_swap_t)*ebpf_nprocs; + pthread_mutex_lock(&mutex_cgroup_shm); + for (ect = ebpf_cgroup_pids; ect; ect = ect->next) { + struct pid_on_target2 *pids; + for (pids = ect->pids; pids; pids = pids->next) { + int pid = pids->pid; + netdata_publish_swap_t *out = &pids->swap; + if (likely(swap_pid) && swap_pid[pid]) { + netdata_publish_swap_t *in = swap_pid[pid]; + + memcpy(out, in, sizeof(netdata_publish_swap_t)); + } else { + memset(cv, 0, length); + if (!bpf_map_lookup_elem(fd, &pid, cv)) { + swap_apps_accumulator(cv); + + memcpy(out, cv, sizeof(netdata_publish_swap_t)); + } + } + } + } + pthread_mutex_unlock(&mutex_cgroup_shm); +} + +/** + * Read APPS table + * + * Read the apps table and store data inside the structure. + */ +static void read_apps_table() +{ + netdata_publish_swap_t *cv = swap_vector; + uint32_t key; + struct pid_stat *pids = root_of_pids; + int fd = swap_maps[NETDATA_PID_SWAP_TABLE].map_fd; + size_t length = sizeof(netdata_publish_swap_t)*ebpf_nprocs; + while (pids) { + key = pids->pid; + + if (bpf_map_lookup_elem(fd, &key, cv)) { + pids = pids->next; + continue; + } + + swap_apps_accumulator(cv); + + swap_fill_pid(key, cv); + + // We are cleaning to avoid passing data read from one process to other. + memset(cv, 0, length); + + pids = pids->next; + } +} + +/** +* Send global +* +* Send global charts to Netdata +*/ +static void swap_send_global() +{ + write_io_chart(NETDATA_MEM_SWAP_CHART, NETDATA_EBPF_SYSTEM_GROUP, + swap_publish_aggregated[NETDATA_KEY_SWAP_WRITEPAGE_CALL].dimension, + (long long) swap_hash_values[NETDATA_KEY_SWAP_WRITEPAGE_CALL], + swap_publish_aggregated[NETDATA_KEY_SWAP_READPAGE_CALL].dimension, + (long long) swap_hash_values[NETDATA_KEY_SWAP_READPAGE_CALL]); +} + +/** + * Read global counter + * + * Read the table with number of calls to all functions + */ +static void read_global_table() +{ + netdata_idx_t *stored = swap_values; + netdata_idx_t *val = swap_hash_values; + int fd = swap_maps[NETDATA_SWAP_GLOBAL_TABLE].map_fd; + + uint32_t i, end = NETDATA_SWAP_END; + for (i = NETDATA_KEY_SWAP_READPAGE_CALL; i < end; i++) { + if (!bpf_map_lookup_elem(fd, &i, stored)) { + int j; + int last = ebpf_nprocs; + netdata_idx_t total = 0; + for (j = 0; j < last; j++) + total += stored[j]; + + val[i] = total; + } + } +} + +/** + * Swap read hash + * + * This is the thread callback. + * + * @param ptr It is a NULL value for this thread. + * + * @return It always returns NULL. + */ +void *ebpf_swap_read_hash(void *ptr) +{ + netdata_thread_cleanup_push(ebpf_swap_cleanup, ptr); + heartbeat_t hb; + heartbeat_init(&hb); + + ebpf_module_t *em = (ebpf_module_t *)ptr; + usec_t step = NETDATA_SWAP_SLEEP_MS * em->update_every; + while (!ebpf_exit_plugin) { + (void)heartbeat_next(&hb, step); + + read_global_table(); + } + + netdata_thread_cleanup_pop(1); + return NULL; +} + +/** + * Sum PIDs + * + * Sum values for all targets. + * + * @param swap + * @param root + */ +static void ebpf_swap_sum_pids(netdata_publish_swap_t *swap, struct pid_on_target *root) +{ + uint64_t local_read = 0; + uint64_t local_write = 0; + + while (root) { + int32_t pid = root->pid; + netdata_publish_swap_t *w = swap_pid[pid]; + if (w) { + local_write += w->write; + local_read += w->read; + } + root = root->next; + } + + // These conditions were added, because we are using incremental algorithm + swap->write = (local_write >= swap->write) ? local_write : swap->write; + swap->read = (local_read >= swap->read) ? local_read : swap->read; +} + +/** + * Send data to Netdata calling auxiliary functions. + * + * @param root the target list. +*/ +void ebpf_swap_send_apps_data(struct target *root) +{ + struct target *w; + for (w = root; w; w = w->next) { + if (unlikely(w->exposed && w->processes)) { + ebpf_swap_sum_pids(&w->swap, w->root_pid); + } + } + + write_begin_chart(NETDATA_APPS_FAMILY, NETDATA_MEM_SWAP_READ_CHART); + for (w = root; w; w = w->next) { + if (unlikely(w->exposed && w->processes)) { + write_chart_dimension(w->name, (long long) w->swap.read); + } + } + write_end_chart(); + + write_begin_chart(NETDATA_APPS_FAMILY, NETDATA_MEM_SWAP_WRITE_CHART); + for (w = root; w; w = w->next) { + if (unlikely(w->exposed && w->processes)) { + write_chart_dimension(w->name, (long long) w->swap.write); + } + } + write_end_chart(); +} + +/** + * Sum PIDs + * + * Sum values for all targets. + * + * @param swap + * @param root + */ +static void ebpf_swap_sum_cgroup_pids(netdata_publish_swap_t *swap, struct pid_on_target2 *pids) +{ + uint64_t local_read = 0; + uint64_t local_write = 0; + + while (pids) { + netdata_publish_swap_t *w = &pids->swap; + local_write += w->write; + local_read += w->read; + + pids = pids->next; + } + + // These conditions were added, because we are using incremental algorithm + swap->write = (local_write >= swap->write) ? local_write : swap->write; + swap->read = (local_read >= swap->read) ? local_read : swap->read; +} + +/** + * Send Systemd charts + * + * Send collected data to Netdata. + */ +static void ebpf_send_systemd_swap_charts() +{ + ebpf_cgroup_target_t *ect; + write_begin_chart(NETDATA_SERVICE_FAMILY, NETDATA_MEM_SWAP_READ_CHART); + for (ect = ebpf_cgroup_pids; ect ; ect = ect->next) { + if (unlikely(ect->systemd) && unlikely(ect->updated)) { + write_chart_dimension(ect->name, (long long) ect->publish_systemd_swap.read); + } + } + write_end_chart(); + + write_begin_chart(NETDATA_SERVICE_FAMILY, NETDATA_MEM_SWAP_WRITE_CHART); + for (ect = ebpf_cgroup_pids; ect ; ect = ect->next) { + if (unlikely(ect->systemd) && unlikely(ect->updated)) { + write_chart_dimension(ect->name, (long long) ect->publish_systemd_swap.write); + } + } + write_end_chart(); +} + +/** + * Create specific swap charts + * + * Create charts for cgroup/application. + * + * @param type the chart type. + * @param update_every value to overwrite the update frequency set by the server. + */ +static void ebpf_create_specific_swap_charts(char *type, int update_every) +{ + ebpf_create_chart(type, NETDATA_MEM_SWAP_READ_CHART, + "Calls to function <code>swap_readpage</code>.", + EBPF_COMMON_DIMENSION_CALL, NETDATA_SYSTEM_CGROUP_SWAP_SUBMENU, + NETDATA_CGROUP_SWAP_READ_CONTEXT, NETDATA_EBPF_CHART_TYPE_LINE, + NETDATA_CHART_PRIO_CGROUPS_CONTAINERS + 5100, + ebpf_create_global_dimension, + swap_publish_aggregated, 1, update_every, NETDATA_EBPF_MODULE_NAME_SWAP); + + ebpf_create_chart(type, NETDATA_MEM_SWAP_WRITE_CHART, + "Calls to function <code>swap_writepage</code>.", + EBPF_COMMON_DIMENSION_CALL, NETDATA_SYSTEM_CGROUP_SWAP_SUBMENU, + NETDATA_CGROUP_SWAP_WRITE_CONTEXT, NETDATA_EBPF_CHART_TYPE_LINE, + NETDATA_CHART_PRIO_CGROUPS_CONTAINERS + 5101, + ebpf_create_global_dimension, + &swap_publish_aggregated[NETDATA_KEY_SWAP_WRITEPAGE_CALL], 1, + update_every, NETDATA_EBPF_MODULE_NAME_SWAP); +} + +/** + * Create specific swap charts + * + * Create charts for cgroup/application. + * + * @param type the chart type. + * @param update_every value to overwrite the update frequency set by the server. + */ +static void ebpf_obsolete_specific_swap_charts(char *type, int update_every) +{ + ebpf_write_chart_obsolete(type, NETDATA_MEM_SWAP_READ_CHART,"Calls to function <code>swap_readpage</code>.", + EBPF_COMMON_DIMENSION_CALL, NETDATA_SYSTEM_CGROUP_SWAP_SUBMENU, + NETDATA_EBPF_CHART_TYPE_LINE, NETDATA_CGROUP_SWAP_READ_CONTEXT, + NETDATA_CHART_PRIO_CGROUPS_CONTAINERS + 5100, update_every); + + ebpf_write_chart_obsolete(type, NETDATA_MEM_SWAP_WRITE_CHART, "Calls to function <code>swap_writepage</code>.", + EBPF_COMMON_DIMENSION_CALL, NETDATA_SYSTEM_CGROUP_SWAP_SUBMENU, + NETDATA_EBPF_CHART_TYPE_LINE, NETDATA_CGROUP_SWAP_WRITE_CONTEXT, + NETDATA_CHART_PRIO_CGROUPS_CONTAINERS + 5101, update_every); +} + +/* + * Send Specific Swap data + * + * Send data for specific cgroup/apps. + * + * @param type chart type + * @param values structure with values that will be sent to netdata + */ +static void ebpf_send_specific_swap_data(char *type, netdata_publish_swap_t *values) +{ + write_begin_chart(type, NETDATA_MEM_SWAP_READ_CHART); + write_chart_dimension(swap_publish_aggregated[NETDATA_KEY_SWAP_READPAGE_CALL].name, (long long) values->read); + write_end_chart(); + + write_begin_chart(type, NETDATA_MEM_SWAP_WRITE_CHART); + write_chart_dimension(swap_publish_aggregated[NETDATA_KEY_SWAP_WRITEPAGE_CALL].name, (long long) values->write); + write_end_chart(); +} + +/** + * Create Systemd Swap Charts + * + * Create charts when systemd is enabled + * + * @param update_every value to overwrite the update frequency set by the server. + **/ +static void ebpf_create_systemd_swap_charts(int update_every) +{ + ebpf_create_charts_on_systemd(NETDATA_MEM_SWAP_READ_CHART, + "Calls to <code>swap_readpage</code>.", + EBPF_COMMON_DIMENSION_CALL, NETDATA_SYSTEM_CGROUP_SWAP_SUBMENU, + NETDATA_EBPF_CHART_TYPE_STACKED, 20191, + ebpf_algorithms[NETDATA_EBPF_INCREMENTAL_IDX], NETDATA_SYSTEMD_SWAP_READ_CONTEXT, + NETDATA_EBPF_MODULE_NAME_SWAP, update_every); + + ebpf_create_charts_on_systemd(NETDATA_MEM_SWAP_WRITE_CHART, + "Calls to function <code>swap_writepage</code>.", + EBPF_COMMON_DIMENSION_CALL, NETDATA_SYSTEM_CGROUP_SWAP_SUBMENU, + NETDATA_EBPF_CHART_TYPE_STACKED, 20192, + ebpf_algorithms[NETDATA_EBPF_INCREMENTAL_IDX], NETDATA_SYSTEMD_SWAP_WRITE_CONTEXT, + NETDATA_EBPF_MODULE_NAME_SWAP, update_every); +} + +/** + * Send data to Netdata calling auxiliary functions. + * + * @param update_every value to overwrite the update frequency set by the server. +*/ +void ebpf_swap_send_cgroup_data(int update_every) +{ + if (!ebpf_cgroup_pids) + return; + + pthread_mutex_lock(&mutex_cgroup_shm); + ebpf_cgroup_target_t *ect; + for (ect = ebpf_cgroup_pids; ect ; ect = ect->next) { + ebpf_swap_sum_cgroup_pids(&ect->publish_systemd_swap, ect->pids); + } + + int has_systemd = shm_ebpf_cgroup.header->systemd_enabled; + + if (has_systemd) { + if (send_cgroup_chart) { + ebpf_create_systemd_swap_charts(update_every); + fflush(stdout); + } + ebpf_send_systemd_swap_charts(); + } + + for (ect = ebpf_cgroup_pids; ect ; ect = ect->next) { + if (ect->systemd) + continue; + + if (!(ect->flags & NETDATA_EBPF_CGROUP_HAS_SWAP_CHART) && ect->updated) { + ebpf_create_specific_swap_charts(ect->name, update_every); + ect->flags |= NETDATA_EBPF_CGROUP_HAS_SWAP_CHART; + } + + if (ect->flags & NETDATA_EBPF_CGROUP_HAS_SWAP_CHART) { + if (ect->updated) { + ebpf_send_specific_swap_data(ect->name, &ect->publish_systemd_swap); + } else { + ebpf_obsolete_specific_swap_charts(ect->name, update_every); + ect->flags &= ~NETDATA_EBPF_CGROUP_HAS_SWAP_CHART; + } + } + } + + pthread_mutex_unlock(&mutex_cgroup_shm); +} + +/** +* Main loop for this collector. +*/ +static void swap_collector(ebpf_module_t *em) +{ + swap_threads.thread = mallocz(sizeof(netdata_thread_t)); + swap_threads.start_routine = ebpf_swap_read_hash; + + netdata_thread_create(swap_threads.thread, swap_threads.name, NETDATA_THREAD_OPTION_DEFAULT, + ebpf_swap_read_hash, em); + + int cgroup = em->cgroup_charts; + int update_every = em->update_every; + heartbeat_t hb; + heartbeat_init(&hb); + usec_t step = update_every * USEC_PER_SEC; + while (!ebpf_exit_plugin) { + (void)heartbeat_next(&hb, step); + if (ebpf_exit_plugin) + break; + + netdata_apps_integration_flags_t apps = em->apps_charts; + pthread_mutex_lock(&collect_data_mutex); + if (apps) + read_apps_table(); + + if (cgroup) + ebpf_update_swap_cgroup(); + + pthread_mutex_lock(&lock); + + swap_send_global(); + + if (apps & NETDATA_EBPF_APPS_FLAG_CHART_CREATED) + ebpf_swap_send_apps_data(apps_groups_root_target); + + if (cgroup) + ebpf_swap_send_cgroup_data(update_every); + + pthread_mutex_unlock(&lock); + pthread_mutex_unlock(&collect_data_mutex); + } +} + +/***************************************************************** + * + * INITIALIZE THREAD + * + *****************************************************************/ + +/** + * Create apps charts + * + * Call ebpf_create_chart to create the charts on apps submenu. + * + * @param em a pointer to the structure with the default values. + */ +void ebpf_swap_create_apps_charts(struct ebpf_module *em, void *ptr) +{ + struct target *root = ptr; + ebpf_create_charts_on_apps(NETDATA_MEM_SWAP_READ_CHART, + "Calls to function <code>swap_readpage</code>.", + EBPF_COMMON_DIMENSION_CALL, + NETDATA_SWAP_SUBMENU, + NETDATA_EBPF_CHART_TYPE_STACKED, + 20191, + ebpf_algorithms[NETDATA_EBPF_INCREMENTAL_IDX], + root, em->update_every, NETDATA_EBPF_MODULE_NAME_SWAP); + + ebpf_create_charts_on_apps(NETDATA_MEM_SWAP_WRITE_CHART, + "Calls to function <code>swap_writepage</code>.", + EBPF_COMMON_DIMENSION_CALL, + NETDATA_SWAP_SUBMENU, + NETDATA_EBPF_CHART_TYPE_STACKED, + 20192, + ebpf_algorithms[NETDATA_EBPF_INCREMENTAL_IDX], + root, em->update_every, NETDATA_EBPF_MODULE_NAME_SWAP); + em->apps_charts |= NETDATA_EBPF_APPS_FLAG_CHART_CREATED; +} + +/** + * Allocate vectors used with this thread. + * + * We are not testing the return, because callocz does this and shutdown the software + * case it was not possible to allocate. + * + * @param apps is apps enabled? + */ +static void ebpf_swap_allocate_global_vectors(int apps) +{ + if (apps) + swap_pid = callocz((size_t)pid_max, sizeof(netdata_publish_swap_t *)); + + swap_vector = callocz((size_t)ebpf_nprocs, sizeof(netdata_publish_swap_t)); + + swap_values = callocz((size_t)ebpf_nprocs, sizeof(netdata_idx_t)); + + memset(swap_hash_values, 0, sizeof(swap_hash_values)); +} + +/***************************************************************** + * + * MAIN THREAD + * + *****************************************************************/ + +/** + * Create global charts + * + * Call ebpf_create_chart to create the charts for the collector. + * + * @param update_every value to overwrite the update frequency set by the server. + */ +static void ebpf_create_swap_charts(int update_every) +{ + ebpf_create_chart(NETDATA_EBPF_SYSTEM_GROUP, NETDATA_MEM_SWAP_CHART, + "Calls to access swap memory", + EBPF_COMMON_DIMENSION_CALL, NETDATA_SYSTEM_SWAP_SUBMENU, + NULL, + NETDATA_EBPF_CHART_TYPE_LINE, + 202, + ebpf_create_global_dimension, + swap_publish_aggregated, NETDATA_SWAP_END, + update_every, NETDATA_EBPF_MODULE_NAME_SWAP); +} + +/* + * Load BPF + * + * Load BPF files. + * + * @param em the structure with configuration + */ +static int ebpf_swap_load_bpf(ebpf_module_t *em) +{ + int ret = 0; + ebpf_adjust_apps_cgroup(em, em->targets[NETDATA_KEY_SWAP_READPAGE_CALL].mode); + if (em->load & EBPF_LOAD_LEGACY) { + em->probe_links = ebpf_load_program(ebpf_plugin_dir, em, running_on_kernel, isrh, &em->objects); + if (!em->probe_links) { + ret = -1; + } + } +#ifdef LIBBPF_MAJOR_VERSION + else { + bpf_obj = swap_bpf__open(); + if (!bpf_obj) + ret = -1; + else + ret = ebpf_swap_load_and_attach(bpf_obj, em); + } +#endif + + if (ret) + error("%s %s", EBPF_DEFAULT_ERROR_MSG, em->thread_name); + + return ret; +} + +/** + * SWAP thread + * + * Thread used to make swap thread + * + * @param ptr a pointer to `struct ebpf_module` + * + * @return It always return NULL + */ +void *ebpf_swap_thread(void *ptr) +{ + netdata_thread_cleanup_push(ebpf_swap_exit, ptr); + + ebpf_module_t *em = (ebpf_module_t *)ptr; + em->maps = swap_maps; + + ebpf_update_pid_table(&swap_maps[NETDATA_PID_SWAP_TABLE], em); + +#ifdef LIBBPF_MAJOR_VERSION + ebpf_adjust_thread_load(em, default_btf); +#endif + if (ebpf_swap_load_bpf(em)) { + em->thread->enabled = NETDATA_THREAD_EBPF_STOPPED; + goto endswap; + } + + ebpf_swap_allocate_global_vectors(em->apps_charts); + + int algorithms[NETDATA_SWAP_END] = { NETDATA_EBPF_INCREMENTAL_IDX, NETDATA_EBPF_INCREMENTAL_IDX }; + ebpf_global_labels(swap_aggregated_data, swap_publish_aggregated, swap_dimension_name, swap_dimension_name, + algorithms, NETDATA_SWAP_END); + + pthread_mutex_lock(&lock); + ebpf_create_swap_charts(em->update_every); + ebpf_update_stats(&plugin_statistics, em); + pthread_mutex_unlock(&lock); + + swap_collector(em); + +endswap: + ebpf_update_disabled_plugin_stats(em); + + netdata_thread_cleanup_pop(1); + return NULL; +} diff --git a/collectors/ebpf.plugin/ebpf_swap.h b/collectors/ebpf.plugin/ebpf_swap.h new file mode 100644 index 0000000..79182e5 --- /dev/null +++ b/collectors/ebpf.plugin/ebpf_swap.h @@ -0,0 +1,53 @@ +// SPDX-License-Identifier: GPL-3.0-or-later + +#ifndef NETDATA_EBPF_SWAP_H +#define NETDATA_EBPF_SWAP_H 1 + +// Module name +#define NETDATA_EBPF_MODULE_NAME_SWAP "swap" + +#define NETDATA_SWAP_SLEEP_MS 850000ULL + +// charts +#define NETDATA_MEM_SWAP_CHART "swapcalls" +#define NETDATA_MEM_SWAP_READ_CHART "swap_read_call" +#define NETDATA_MEM_SWAP_WRITE_CHART "swap_write_call" +#define NETDATA_SWAP_SUBMENU "swap" + +// configuration file +#define NETDATA_DIRECTORY_SWAP_CONFIG_FILE "swap.conf" + +// Contexts +#define NETDATA_CGROUP_SWAP_READ_CONTEXT "cgroup.swap_read" +#define NETDATA_CGROUP_SWAP_WRITE_CONTEXT "cgroup.swap_write" +#define NETDATA_SYSTEMD_SWAP_READ_CONTEXT "services.swap_read" +#define NETDATA_SYSTEMD_SWAP_WRITE_CONTEXT "services.swap_write" + +typedef struct netdata_publish_swap { + uint64_t read; + uint64_t write; +} netdata_publish_swap_t; + +enum swap_tables { + NETDATA_PID_SWAP_TABLE, + NETDATA_SWAP_CONTROLLER, + NETDATA_SWAP_GLOBAL_TABLE +}; + +enum swap_counters { + NETDATA_KEY_SWAP_READPAGE_CALL, + NETDATA_KEY_SWAP_WRITEPAGE_CALL, + + // Keep this as last and don't skip numbers as it is used as element counter + NETDATA_SWAP_END +}; + +extern netdata_publish_swap_t **swap_pid; + +void *ebpf_swap_thread(void *ptr); +void ebpf_swap_create_apps_charts(struct ebpf_module *em, void *ptr); + +extern struct config swap_config; +extern netdata_ebpf_targets_t swap_targets[]; + +#endif diff --git a/collectors/ebpf.plugin/ebpf_sync.c b/collectors/ebpf.plugin/ebpf_sync.c new file mode 100644 index 0000000..8404975 --- /dev/null +++ b/collectors/ebpf.plugin/ebpf_sync.c @@ -0,0 +1,608 @@ +// SPDX-License-Identifier: GPL-3.0-or-later + +#include "ebpf.h" +#include "ebpf_sync.h" + +static char *sync_counter_dimension_name[NETDATA_SYNC_IDX_END] = { "sync", "syncfs", "msync", "fsync", "fdatasync", + "sync_file_range" }; +static netdata_syscall_stat_t sync_counter_aggregated_data[NETDATA_SYNC_IDX_END]; +static netdata_publish_syscall_t sync_counter_publish_aggregated[NETDATA_SYNC_IDX_END]; + +static netdata_idx_t sync_hash_values[NETDATA_SYNC_IDX_END]; + +struct netdata_static_thread sync_threads = { + .name = "SYNC KERNEL", + .config_section = NULL, + .config_name = NULL, + .env_name = NULL, + .enabled = 1, + .thread = NULL, + .init_routine = NULL, + .start_routine = NULL +}; + +static ebpf_local_maps_t sync_maps[] = {{.name = "tbl_sync", .internal_input = NETDATA_SYNC_END, + .user_input = 0, .type = NETDATA_EBPF_MAP_STATIC, + .map_fd = ND_EBPF_MAP_FD_NOT_INITIALIZED}, + {.name = "tbl_syncfs", .internal_input = NETDATA_SYNC_END, + .user_input = 0, .type = NETDATA_EBPF_MAP_STATIC, + .map_fd = ND_EBPF_MAP_FD_NOT_INITIALIZED}, + {.name = "tbl_msync", .internal_input = NETDATA_SYNC_END, + .user_input = 0, .type = NETDATA_EBPF_MAP_STATIC, + .map_fd = ND_EBPF_MAP_FD_NOT_INITIALIZED}, + {.name = "tbl_fsync", .internal_input = NETDATA_SYNC_END, + .user_input = 0, .type = NETDATA_EBPF_MAP_STATIC, + .map_fd = ND_EBPF_MAP_FD_NOT_INITIALIZED}, + {.name = "tbl_fdatasync", .internal_input = NETDATA_SYNC_END, + .user_input = 0, .type = NETDATA_EBPF_MAP_STATIC, + .map_fd = ND_EBPF_MAP_FD_NOT_INITIALIZED}, + {.name = "tbl_syncfr", .internal_input = NETDATA_SYNC_END, + .user_input = 0, .type = NETDATA_EBPF_MAP_STATIC, + .map_fd = ND_EBPF_MAP_FD_NOT_INITIALIZED}, + {.name = NULL, .internal_input = 0, .user_input = 0, + .type = NETDATA_EBPF_MAP_CONTROLLER, + .map_fd = ND_EBPF_MAP_FD_NOT_INITIALIZED}}; + +struct config sync_config = { .first_section = NULL, + .last_section = NULL, + .mutex = NETDATA_MUTEX_INITIALIZER, + .index = { .avl_tree = { .root = NULL, .compar = appconfig_section_compare }, + .rwlock = AVL_LOCK_INITIALIZER } }; + +netdata_ebpf_targets_t sync_targets[] = { {.name = NETDATA_SYSCALLS_SYNC, .mode = EBPF_LOAD_TRAMPOLINE}, + {.name = NETDATA_SYSCALLS_SYNCFS, .mode = EBPF_LOAD_TRAMPOLINE}, + {.name = NETDATA_SYSCALLS_MSYNC, .mode = EBPF_LOAD_TRAMPOLINE}, + {.name = NETDATA_SYSCALLS_FSYNC, .mode = EBPF_LOAD_TRAMPOLINE}, + {.name = NETDATA_SYSCALLS_FDATASYNC, .mode = EBPF_LOAD_TRAMPOLINE}, + {.name = NETDATA_SYSCALLS_SYNC_FILE_RANGE, .mode = EBPF_LOAD_TRAMPOLINE}, + {.name = NULL, .mode = EBPF_LOAD_TRAMPOLINE}}; + +#ifdef LIBBPF_MAJOR_VERSION +/***************************************************************** + * + * BTF FUNCTIONS + * + *****************************************************************/ + +/** + * Disable probe + * + * Disable kprobe to use another method. + * + * @param obj is the main structure for bpf objects. + */ +static inline void ebpf_sync_disable_probe(struct sync_bpf *obj) +{ + bpf_program__set_autoload(obj->progs.netdata_sync_kprobe, false); +} + +/** + * Disable tramppoline + * + * Disable trampoline to use another method. + * + * @param obj is the main structure for bpf objects. + */ +static inline void ebpf_sync_disable_trampoline(struct sync_bpf *obj) +{ + bpf_program__set_autoload(obj->progs.netdata_sync_fentry, false); +} + +/** + * Disable tracepoint + * + * Disable tracepoints according information given. + * + * @param obj object loaded + * @param idx Which syscall will not be disabled + */ +void ebpf_sync_disable_tracepoints(struct sync_bpf *obj, sync_syscalls_index_t idx) +{ + if (idx != NETDATA_SYNC_SYNC_IDX) + bpf_program__set_autoload(obj->progs.netdata_sync_entry, false); + + if (idx != NETDATA_SYNC_SYNCFS_IDX) + bpf_program__set_autoload(obj->progs.netdata_syncfs_entry, false); + + if (idx != NETDATA_SYNC_MSYNC_IDX) + bpf_program__set_autoload(obj->progs.netdata_msync_entry, false); + + if (idx != NETDATA_SYNC_FSYNC_IDX) + bpf_program__set_autoload(obj->progs.netdata_fsync_entry, false); + + if (idx != NETDATA_SYNC_FDATASYNC_IDX) + bpf_program__set_autoload(obj->progs.netdata_fdatasync_entry, false); + + if (idx != NETDATA_SYNC_SYNC_FILE_RANGE_IDX) + bpf_program__set_autoload(obj->progs.netdata_sync_file_range_entry, false); +} + +/** + * Set hash tables + * + * Set the values for maps according the value given by kernel. + * + * @param obj is the main structure for bpf objects. + * @param idx the index for the main structure + */ +static void ebpf_sync_set_hash_tables(struct sync_bpf *obj, sync_syscalls_index_t idx) +{ + sync_maps[idx].map_fd = bpf_map__fd(obj->maps.tbl_sync); +} + +/** + * Load and attach + * + * Load and attach the eBPF code in kernel. + * + * @param obj is the main structure for bpf objects. + * @param em the structure with configuration + * @param target the syscall that we are attaching a tracer. + * @param idx the index for the main structure + * + * @return it returns 0 on succes and -1 otherwise + */ +static inline int ebpf_sync_load_and_attach(struct sync_bpf *obj, ebpf_module_t *em, char *target, + sync_syscalls_index_t idx) +{ + netdata_ebpf_targets_t *synct = em->targets; + netdata_ebpf_program_loaded_t test = synct[NETDATA_SYNC_SYNC_IDX].mode; + + if (test == EBPF_LOAD_TRAMPOLINE) { + ebpf_sync_disable_probe(obj); + ebpf_sync_disable_tracepoints(obj, NETDATA_SYNC_IDX_END); + + bpf_program__set_attach_target(obj->progs.netdata_sync_fentry, 0, + target); + } else if (test == EBPF_LOAD_PROBE || + test == EBPF_LOAD_RETPROBE) { + ebpf_sync_disable_tracepoints(obj, NETDATA_SYNC_IDX_END); + ebpf_sync_disable_trampoline(obj); + } else { + ebpf_sync_disable_probe(obj); + ebpf_sync_disable_trampoline(obj); + + ebpf_sync_disable_tracepoints(obj, idx); + } + + int ret = sync_bpf__load(obj); + if (!ret) { + if (test != EBPF_LOAD_PROBE && test != EBPF_LOAD_RETPROBE) { + ret = sync_bpf__attach(obj); + } else { + obj->links.netdata_sync_kprobe = bpf_program__attach_kprobe(obj->progs.netdata_sync_kprobe, + false, target); + ret = (int)libbpf_get_error(obj->links.netdata_sync_kprobe); + } + + if (!ret) + ebpf_sync_set_hash_tables(obj, idx); + } + + return ret; +} +#endif + +/***************************************************************** + * + * CLEANUP THREAD + * + *****************************************************************/ + +#ifdef LIBBPF_MAJOR_VERSION +/** + * Cleanup Objects + * + * Cleanup loaded objects when thread was initialized. + */ +void ebpf_sync_cleanup_objects() +{ + int i; + for (i = 0; local_syscalls[i].syscall; i++) { + ebpf_sync_syscalls_t *w = &local_syscalls[i]; + if (w->sync_obj) + sync_bpf__destroy(w->sync_obj); + } +} +#endif + +/** + * Sync Free + * + * Cleanup variables after child threads to stop + * + * @param ptr thread data. + */ +static void ebpf_sync_free(ebpf_module_t *em) +{ + pthread_mutex_lock(&ebpf_exit_cleanup); + if (em->thread->enabled == NETDATA_THREAD_EBPF_RUNNING) { + em->thread->enabled = NETDATA_THREAD_EBPF_STOPPING; + pthread_mutex_unlock(&ebpf_exit_cleanup); + return; + } + pthread_mutex_unlock(&ebpf_exit_cleanup); + +#ifdef LIBBPF_MAJOR_VERSION + ebpf_sync_cleanup_objects(); +#endif + freez(sync_threads.thread); + + pthread_mutex_lock(&ebpf_exit_cleanup); + em->thread->enabled = NETDATA_THREAD_EBPF_STOPPED; + pthread_mutex_unlock(&ebpf_exit_cleanup); +} + +/** + * Exit + * + * Clean up the main thread. + * + * @param ptr thread data. + */ +static void ebpf_sync_exit(void *ptr) +{ + ebpf_module_t *em = (ebpf_module_t *)ptr; + netdata_thread_cancel(*sync_threads.thread); + ebpf_sync_free(em); +} + +/** + * Clean up the main thread. + * + * @param ptr thread data. + */ +static void ebpf_sync_cleanup(void *ptr) +{ + ebpf_module_t *em = (ebpf_module_t *)ptr; + ebpf_sync_free(em); +} + +/***************************************************************** + * + * INITIALIZE THREAD + * + *****************************************************************/ + +/** + * Load Legacy + * + * Load legacy code. + * + * @param w is the sync output structure with pointers to objects loaded. + * @param em is structure with configuration + * + * @return 0 on success and -1 otherwise. + */ +static int ebpf_sync_load_legacy(ebpf_sync_syscalls_t *w, ebpf_module_t *em) +{ + em->thread_name = w->syscall; + if (!w->probe_links) { + w->probe_links = ebpf_load_program(ebpf_plugin_dir, em, running_on_kernel, isrh, &w->objects); + if (!w->probe_links) { + return -1; + } + } + + return 0; +} + +/* + * Initialize Syscalls + * + * Load the eBPF programs to monitor syscalls + * + * @return 0 on success and -1 otherwise. + */ +static int ebpf_sync_initialize_syscall(ebpf_module_t *em) +{ + int i; + const char *saved_name = em->thread_name; + int errors = 0; + for (i = 0; local_syscalls[i].syscall; i++) { + ebpf_sync_syscalls_t *w = &local_syscalls[i]; + if (w->enabled) { + if (em->load & EBPF_LOAD_LEGACY) { + if (ebpf_sync_load_legacy(w, em)) + errors++; + + em->thread_name = saved_name; + } +#ifdef LIBBPF_MAJOR_VERSION + else { + char syscall[NETDATA_EBPF_MAX_SYSCALL_LENGTH]; + ebpf_select_host_prefix(syscall, NETDATA_EBPF_MAX_SYSCALL_LENGTH, w->syscall, running_on_kernel); + w->sync_obj = sync_bpf__open(); + if (!w->sync_obj) { + errors++; + } else { + if (ebpf_is_function_inside_btf(default_btf, syscall)) { + if (ebpf_sync_load_and_attach(w->sync_obj, em, syscall, i)) { + errors++; + } + } else { + if (ebpf_sync_load_legacy(w, em)) + errors++; + } + em->thread_name = saved_name; + } + } +#endif + } + } + em->thread_name = saved_name; + + memset(sync_counter_aggregated_data, 0 , NETDATA_SYNC_IDX_END * sizeof(netdata_syscall_stat_t)); + memset(sync_counter_publish_aggregated, 0 , NETDATA_SYNC_IDX_END * sizeof(netdata_publish_syscall_t)); + memset(sync_hash_values, 0 , NETDATA_SYNC_IDX_END * sizeof(netdata_idx_t)); + + return (errors) ? -1 : 0; +} + +/***************************************************************** + * + * DATA THREAD + * + *****************************************************************/ + +/** + * Read global table + * + * Read the table with number of calls for all functions + */ +static void read_global_table() +{ + netdata_idx_t stored; + uint32_t idx = NETDATA_SYNC_CALL; + int i; + for (i = 0; local_syscalls[i].syscall; i++) { + if (local_syscalls[i].enabled) { + int fd = sync_maps[i].map_fd; + if (!bpf_map_lookup_elem(fd, &idx, &stored)) { + sync_hash_values[i] = stored; + } + } + } +} + +/** + * Sync read hash + * + * This is the thread callback. + * + * @param ptr It is a NULL value for this thread. + * + * @return It always returns NULL. + */ +void *ebpf_sync_read_hash(void *ptr) +{ + netdata_thread_cleanup_push(ebpf_sync_cleanup, ptr); + ebpf_module_t *em = (ebpf_module_t *)ptr; + + heartbeat_t hb; + heartbeat_init(&hb); + usec_t step = NETDATA_EBPF_SYNC_SLEEP_MS * em->update_every; + + while (!ebpf_exit_plugin) { + (void)heartbeat_next(&hb, step); + + read_global_table(); + } + + netdata_thread_cleanup_pop(1); + return NULL; +} + +/** + * Create Sync charts + * + * Create charts and dimensions according user input. + * + * @param id chart id + * @param idx the first index with data. + * @param end the last index with data. + */ +static void ebpf_send_sync_chart(char *id, + int idx, + int end) +{ + write_begin_chart(NETDATA_EBPF_MEMORY_GROUP, id); + + netdata_publish_syscall_t *move = &sync_counter_publish_aggregated[idx]; + + while (move && idx <= end) { + if (local_syscalls[idx].enabled) + write_chart_dimension(move->name, sync_hash_values[idx]); + + move = move->next; + idx++; + } + + write_end_chart(); +} + +/** + * Send data + * + * Send global charts to Netdata + */ +static void sync_send_data() +{ + if (local_syscalls[NETDATA_SYNC_FSYNC_IDX].enabled || local_syscalls[NETDATA_SYNC_FDATASYNC_IDX].enabled) { + ebpf_send_sync_chart(NETDATA_EBPF_FILE_SYNC_CHART, NETDATA_SYNC_FSYNC_IDX, NETDATA_SYNC_FDATASYNC_IDX); + } + + if (local_syscalls[NETDATA_SYNC_MSYNC_IDX].enabled) + ebpf_one_dimension_write_charts(NETDATA_EBPF_MEMORY_GROUP, NETDATA_EBPF_MSYNC_CHART, + sync_counter_publish_aggregated[NETDATA_SYNC_MSYNC_IDX].dimension, + sync_hash_values[NETDATA_SYNC_MSYNC_IDX]); + + if (local_syscalls[NETDATA_SYNC_SYNC_IDX].enabled || local_syscalls[NETDATA_SYNC_SYNCFS_IDX].enabled) { + ebpf_send_sync_chart(NETDATA_EBPF_SYNC_CHART, NETDATA_SYNC_SYNC_IDX, NETDATA_SYNC_SYNCFS_IDX); + } + + if (local_syscalls[NETDATA_SYNC_SYNC_FILE_RANGE_IDX].enabled) + ebpf_one_dimension_write_charts(NETDATA_EBPF_MEMORY_GROUP, NETDATA_EBPF_FILE_SEGMENT_CHART, + sync_counter_publish_aggregated[NETDATA_SYNC_SYNC_FILE_RANGE_IDX].dimension, + sync_hash_values[NETDATA_SYNC_SYNC_FILE_RANGE_IDX]); +} + +/** +* Main loop for this collector. +*/ +static void sync_collector(ebpf_module_t *em) +{ + sync_threads.thread = mallocz(sizeof(netdata_thread_t)); + sync_threads.start_routine = ebpf_sync_read_hash; + + netdata_thread_create(sync_threads.thread, sync_threads.name, NETDATA_THREAD_OPTION_DEFAULT, + ebpf_sync_read_hash, em); + + heartbeat_t hb; + heartbeat_init(&hb); + usec_t step = em->update_every * USEC_PER_SEC; + while (!ebpf_exit_plugin) { + (void)heartbeat_next(&hb, step); + if (ebpf_exit_plugin) + break; + + pthread_mutex_lock(&lock); + + sync_send_data(); + + pthread_mutex_unlock(&lock); + } +} + + +/***************************************************************** + * + * MAIN THREAD + * + *****************************************************************/ + +/** + * Create Sync charts + * + * Create charts and dimensions according user input. + * + * @param id chart id + * @param title chart title + * @param order order number of the specified chart + * @param idx the first index with data. + * @param end the last index with data. + * @param update_every value to overwrite the update frequency set by the server. + */ +static void ebpf_create_sync_chart(char *id, + char *title, + int order, + int idx, + int end, + int update_every) +{ + ebpf_write_chart_cmd(NETDATA_EBPF_MEMORY_GROUP, id, title, EBPF_COMMON_DIMENSION_CALL, + NETDATA_EBPF_SYNC_SUBMENU, NETDATA_EBPF_CHART_TYPE_LINE, NULL, order, + update_every, + NETDATA_EBPF_MODULE_NAME_SYNC); + + netdata_publish_syscall_t *move = &sync_counter_publish_aggregated[idx]; + + while (move && idx <= end) { + if (local_syscalls[idx].enabled) + ebpf_write_global_dimension(move->name, move->dimension, move->algorithm); + + move = move->next; + idx++; + } +} + +/** + * Create global charts + * + * Call ebpf_create_chart to create the charts for the collector. + * + * @param update_every value to overwrite the update frequency set by the server. + */ +static void ebpf_create_sync_charts(int update_every) +{ + if (local_syscalls[NETDATA_SYNC_FSYNC_IDX].enabled || local_syscalls[NETDATA_SYNC_FDATASYNC_IDX].enabled) + ebpf_create_sync_chart(NETDATA_EBPF_FILE_SYNC_CHART, + "Monitor calls for <code>fsync(2)</code> and <code>fdatasync(2)</code>.", 21300, + NETDATA_SYNC_FSYNC_IDX, NETDATA_SYNC_FDATASYNC_IDX, update_every); + + if (local_syscalls[NETDATA_SYNC_MSYNC_IDX].enabled) + ebpf_create_sync_chart(NETDATA_EBPF_MSYNC_CHART, + "Monitor calls for <code>msync(2)</code>.", 21301, + NETDATA_SYNC_MSYNC_IDX, NETDATA_SYNC_MSYNC_IDX, update_every); + + if (local_syscalls[NETDATA_SYNC_SYNC_IDX].enabled || local_syscalls[NETDATA_SYNC_SYNCFS_IDX].enabled) + ebpf_create_sync_chart(NETDATA_EBPF_SYNC_CHART, + "Monitor calls for <code>sync(2)</code> and <code>syncfs(2)</code>.", 21302, + NETDATA_SYNC_SYNC_IDX, NETDATA_SYNC_SYNCFS_IDX, update_every); + + if (local_syscalls[NETDATA_SYNC_SYNC_FILE_RANGE_IDX].enabled) + ebpf_create_sync_chart(NETDATA_EBPF_FILE_SEGMENT_CHART, + "Monitor calls for <code>sync_file_range(2)</code>.", 21303, + NETDATA_SYNC_SYNC_FILE_RANGE_IDX, NETDATA_SYNC_SYNC_FILE_RANGE_IDX, update_every); +} + +/** + * Parse Syscalls + * + * Parse syscall options available inside ebpf.d/sync.conf + */ +static void ebpf_sync_parse_syscalls() +{ + int i; + for (i = 0; local_syscalls[i].syscall; i++) { + local_syscalls[i].enabled = appconfig_get_boolean(&sync_config, NETDATA_SYNC_CONFIG_NAME, + local_syscalls[i].syscall, CONFIG_BOOLEAN_YES); + } +} + +/** + * Sync thread + * + * Thread used to make sync thread + * + * @param ptr a pointer to `struct ebpf_module` + * + * @return It always return NULL + */ +void *ebpf_sync_thread(void *ptr) +{ + netdata_thread_cleanup_push(ebpf_sync_exit, ptr); + + ebpf_module_t *em = (ebpf_module_t *)ptr; + em->maps = sync_maps; + + ebpf_sync_parse_syscalls(); + +#ifdef LIBBPF_MAJOR_VERSION + ebpf_adjust_thread_load(em, default_btf); +#endif + if (ebpf_sync_initialize_syscall(em)) { + em->thread->enabled = NETDATA_THREAD_EBPF_STOPPED; + goto endsync; + } + + int algorithms[NETDATA_SYNC_IDX_END] = { NETDATA_EBPF_INCREMENTAL_IDX, NETDATA_EBPF_INCREMENTAL_IDX, + NETDATA_EBPF_INCREMENTAL_IDX, NETDATA_EBPF_INCREMENTAL_IDX, + NETDATA_EBPF_INCREMENTAL_IDX, NETDATA_EBPF_INCREMENTAL_IDX }; + ebpf_global_labels(sync_counter_aggregated_data, sync_counter_publish_aggregated, + sync_counter_dimension_name, sync_counter_dimension_name, + algorithms, NETDATA_SYNC_IDX_END); + + pthread_mutex_lock(&lock); + ebpf_create_sync_charts(em->update_every); + ebpf_update_stats(&plugin_statistics, em); + pthread_mutex_unlock(&lock); + + sync_collector(em); + +endsync: + ebpf_update_disabled_plugin_stats(em); + + netdata_thread_cleanup_pop(1); + return NULL; +} diff --git a/collectors/ebpf.plugin/ebpf_sync.h b/collectors/ebpf.plugin/ebpf_sync.h new file mode 100644 index 0000000..cace2a1 --- /dev/null +++ b/collectors/ebpf.plugin/ebpf_sync.h @@ -0,0 +1,59 @@ +// SPDX-License-Identifier: GPL-3.0-or-later + +#ifndef NETDATA_EBPF_SYNC_H +#define NETDATA_EBPF_SYNC_H 1 + +#ifdef LIBBPF_MAJOR_VERSION +#include "includes/sync.skel.h" +#endif + +// Module name +#define NETDATA_EBPF_MODULE_NAME_SYNC "sync" + +// charts +#define NETDATA_EBPF_SYNC_CHART "sync" +#define NETDATA_EBPF_MSYNC_CHART "memory_map" +#define NETDATA_EBPF_FILE_SYNC_CHART "file_sync" +#define NETDATA_EBPF_FILE_SEGMENT_CHART "file_segment" +#define NETDATA_EBPF_SYNC_SUBMENU "synchronization (eBPF)" + +#define NETDATA_SYSCALLS_SYNC "sync" +#define NETDATA_SYSCALLS_SYNCFS "syncfs" +#define NETDATA_SYSCALLS_MSYNC "msync" +#define NETDATA_SYSCALLS_FSYNC "fsync" +#define NETDATA_SYSCALLS_FDATASYNC "fdatasync" +#define NETDATA_SYSCALLS_SYNC_FILE_RANGE "sync_file_range" + +#define NETDATA_EBPF_SYNC_SLEEP_MS 800000ULL + +// configuration file +#define NETDATA_SYNC_CONFIG_FILE "sync.conf" +#define NETDATA_SYNC_CONFIG_NAME "syscalls" + +typedef enum sync_syscalls_index { + NETDATA_SYNC_SYNC_IDX, + NETDATA_SYNC_SYNCFS_IDX, + NETDATA_SYNC_MSYNC_IDX, + NETDATA_SYNC_FSYNC_IDX, + NETDATA_SYNC_FDATASYNC_IDX, + NETDATA_SYNC_SYNC_FILE_RANGE_IDX, + + NETDATA_SYNC_IDX_END +} sync_syscalls_index_t; + +enum netdata_sync_charts { + NETDATA_SYNC_CALL, + + // Keep this as last and don't skip numbers as it is used as element counter + NETDATA_SYNC_END +}; + +enum netdata_sync_table { + NETDATA_SYNC_GLOBAL_TABLE +}; + +void *ebpf_sync_thread(void *ptr); +extern struct config sync_config; +extern netdata_ebpf_targets_t sync_targets[]; + +#endif /* NETDATA_EBPF_SYNC_H */ diff --git a/collectors/ebpf.plugin/ebpf_vfs.c b/collectors/ebpf.plugin/ebpf_vfs.c new file mode 100644 index 0000000..ad6de4a --- /dev/null +++ b/collectors/ebpf.plugin/ebpf_vfs.c @@ -0,0 +1,1983 @@ +// SPDX-License-Identifier: GPL-3.0-or-later + +#include <sys/resource.h> + +#include "ebpf.h" +#include "ebpf_vfs.h" + +static char *vfs_dimension_names[NETDATA_KEY_PUBLISH_VFS_END] = { "delete", "read", "write", + "fsync", "open", "create" }; +static char *vfs_id_names[NETDATA_KEY_PUBLISH_VFS_END] = { "vfs_unlink", "vfs_read", "vfs_write", + "vfs_fsync", "vfs_open", "vfs_create"}; + +static netdata_idx_t *vfs_hash_values = NULL; +static netdata_syscall_stat_t vfs_aggregated_data[NETDATA_KEY_PUBLISH_VFS_END]; +static netdata_publish_syscall_t vfs_publish_aggregated[NETDATA_KEY_PUBLISH_VFS_END]; +netdata_publish_vfs_t **vfs_pid = NULL; +netdata_publish_vfs_t *vfs_vector = NULL; + +static ebpf_local_maps_t vfs_maps[] = {{.name = "tbl_vfs_pid", .internal_input = ND_EBPF_DEFAULT_PID_SIZE, + .user_input = 0, .type = NETDATA_EBPF_MAP_RESIZABLE | NETDATA_EBPF_MAP_PID, + .map_fd = ND_EBPF_MAP_FD_NOT_INITIALIZED}, + {.name = "tbl_vfs_stats", .internal_input = NETDATA_VFS_COUNTER, + .user_input = 0, .type = NETDATA_EBPF_MAP_STATIC, + .map_fd = ND_EBPF_MAP_FD_NOT_INITIALIZED}, + {.name = "vfs_ctrl", .internal_input = NETDATA_CONTROLLER_END, + .user_input = 0, + .type = NETDATA_EBPF_MAP_CONTROLLER, + .map_fd = ND_EBPF_MAP_FD_NOT_INITIALIZED}, + {.name = NULL, .internal_input = 0, .user_input = 0}}; + +struct config vfs_config = { .first_section = NULL, + .last_section = NULL, + .mutex = NETDATA_MUTEX_INITIALIZER, + .index = { .avl_tree = { .root = NULL, .compar = appconfig_section_compare }, + .rwlock = AVL_LOCK_INITIALIZER } }; + +struct netdata_static_thread vfs_threads = { + .name = "VFS KERNEL", + .config_section = NULL, + .config_name = NULL, + .env_name = NULL, + .enabled = 1, + .thread = NULL, + .init_routine = NULL, + .start_routine = NULL +}; + +netdata_ebpf_targets_t vfs_targets[] = { {.name = "vfs_write", .mode = EBPF_LOAD_TRAMPOLINE}, + {.name = "vfs_writev", .mode = EBPF_LOAD_TRAMPOLINE}, + {.name = "vfs_read", .mode = EBPF_LOAD_TRAMPOLINE}, + {.name = "vfs_readv", .mode = EBPF_LOAD_TRAMPOLINE}, + {.name = "vfs_unlink", .mode = EBPF_LOAD_TRAMPOLINE}, + {.name = "vfs_fsync", .mode = EBPF_LOAD_TRAMPOLINE}, + {.name = "vfs_open", .mode = EBPF_LOAD_TRAMPOLINE}, + {.name = "vfs_create", .mode = EBPF_LOAD_TRAMPOLINE}, + {.name = "release_task", .mode = EBPF_LOAD_TRAMPOLINE}, + {.name = NULL, .mode = EBPF_LOAD_TRAMPOLINE}}; + +#ifdef LIBBPF_MAJOR_VERSION +#include "includes/vfs.skel.h" // BTF code + +static struct vfs_bpf *bpf_obj = NULL; + +/** + * Disable probe + * + * Disable all probes to use exclusively another method. + * + * @param obj is the main structure for bpf objects + */ +static void ebpf_vfs_disable_probes(struct vfs_bpf *obj) +{ + bpf_program__set_autoload(obj->progs.netdata_vfs_write_kprobe, false); + bpf_program__set_autoload(obj->progs.netdata_vfs_write_kretprobe, false); + bpf_program__set_autoload(obj->progs.netdata_vfs_writev_kprobe, false); + bpf_program__set_autoload(obj->progs.netdata_vfs_writev_kretprobe, false); + bpf_program__set_autoload(obj->progs.netdata_vfs_read_kprobe, false); + bpf_program__set_autoload(obj->progs.netdata_vfs_read_kretprobe, false); + bpf_program__set_autoload(obj->progs.netdata_vfs_readv_kprobe, false); + bpf_program__set_autoload(obj->progs.netdata_vfs_readv_kretprobe, false); + bpf_program__set_autoload(obj->progs.netdata_vfs_unlink_kprobe, false); + bpf_program__set_autoload(obj->progs.netdata_vfs_unlink_kretprobe, false); + bpf_program__set_autoload(obj->progs.netdata_vfs_fsync_kprobe, false); + bpf_program__set_autoload(obj->progs.netdata_vfs_fsync_kretprobe, false); + bpf_program__set_autoload(obj->progs.netdata_vfs_open_kprobe, false); + bpf_program__set_autoload(obj->progs.netdata_vfs_open_kretprobe, false); + bpf_program__set_autoload(obj->progs.netdata_vfs_create_kprobe, false); + bpf_program__set_autoload(obj->progs.netdata_vfs_create_kretprobe, false); + bpf_program__set_autoload(obj->progs.netdata_vfs_release_task_kprobe, false); +} + +/* + * Disable trampoline + * + * Disable all trampoline to use exclusively another method. + * + * @param obj is the main structure for bpf objects. + */ +static void ebpf_vfs_disable_trampoline(struct vfs_bpf *obj) +{ + bpf_program__set_autoload(obj->progs.netdata_vfs_write_fentry, false); + bpf_program__set_autoload(obj->progs.netdata_vfs_write_fexit, false); + bpf_program__set_autoload(obj->progs.netdata_vfs_writev_fentry, false); + bpf_program__set_autoload(obj->progs.netdata_vfs_writev_fexit, false); + bpf_program__set_autoload(obj->progs.netdata_vfs_read_fentry, false); + bpf_program__set_autoload(obj->progs.netdata_vfs_read_fexit, false); + bpf_program__set_autoload(obj->progs.netdata_vfs_readv_fentry, false); + bpf_program__set_autoload(obj->progs.netdata_vfs_readv_fexit, false); + bpf_program__set_autoload(obj->progs.netdata_vfs_unlink_fentry, false); + bpf_program__set_autoload(obj->progs.netdata_vfs_fsync_fentry, false); + bpf_program__set_autoload(obj->progs.netdata_vfs_fsync_fexit, false); + bpf_program__set_autoload(obj->progs.netdata_vfs_open_fentry, false); + bpf_program__set_autoload(obj->progs.netdata_vfs_open_fexit, false); + bpf_program__set_autoload(obj->progs.netdata_vfs_create_fentry, false); + bpf_program__set_autoload(obj->progs.netdata_vfs_release_task_fentry, false); +} + +/** + * Set trampoline target + * + * Set the targets we will monitor. + * + * @param obj is the main structure for bpf objects. + */ +static void ebpf_vfs_set_trampoline_target(struct vfs_bpf *obj) +{ + bpf_program__set_attach_target(obj->progs.netdata_vfs_write_fentry, 0, vfs_targets[NETDATA_EBPF_VFS_WRITE].name); + + bpf_program__set_attach_target(obj->progs.netdata_vfs_write_fexit, 0, vfs_targets[NETDATA_EBPF_VFS_WRITE].name); + + bpf_program__set_attach_target(obj->progs.netdata_vfs_writev_fentry, 0, vfs_targets[NETDATA_EBPF_VFS_WRITEV].name); + + bpf_program__set_attach_target(obj->progs.netdata_vfs_writev_fexit, 0, vfs_targets[NETDATA_EBPF_VFS_WRITEV].name); + + bpf_program__set_attach_target(obj->progs.netdata_vfs_read_fentry, 0, vfs_targets[NETDATA_EBPF_VFS_READ].name); + + bpf_program__set_attach_target(obj->progs.netdata_vfs_read_fexit, 0, vfs_targets[NETDATA_EBPF_VFS_READ].name); + + bpf_program__set_attach_target(obj->progs.netdata_vfs_readv_fentry, 0, vfs_targets[NETDATA_EBPF_VFS_READV].name); + + bpf_program__set_attach_target(obj->progs.netdata_vfs_readv_fexit, 0, vfs_targets[NETDATA_EBPF_VFS_READV].name); + + bpf_program__set_attach_target(obj->progs.netdata_vfs_unlink_fentry, 0, vfs_targets[NETDATA_EBPF_VFS_UNLINK].name); + + bpf_program__set_attach_target(obj->progs.netdata_vfs_fsync_fentry, 0, vfs_targets[NETDATA_EBPF_VFS_FSYNC].name); + + bpf_program__set_attach_target(obj->progs.netdata_vfs_fsync_fexit, 0, vfs_targets[NETDATA_EBPF_VFS_FSYNC].name); + + bpf_program__set_attach_target(obj->progs.netdata_vfs_open_fentry, 0, vfs_targets[NETDATA_EBPF_VFS_OPEN].name); + + bpf_program__set_attach_target(obj->progs.netdata_vfs_open_fexit, 0, vfs_targets[NETDATA_EBPF_VFS_OPEN].name); + + bpf_program__set_attach_target(obj->progs.netdata_vfs_create_fentry, 0, vfs_targets[NETDATA_EBPF_VFS_CREATE].name); + + bpf_program__set_attach_target(obj->progs.netdata_vfs_release_task_fentry, 0, EBPF_COMMON_FNCT_CLEAN_UP); +} + +/** + * Attach Probe + * + * Attach probes to target + * + * @param obj is the main structure for bpf objects. + * + * @return It returns 0 on success and -1 otherwise. + */ +static int ebpf_vfs_attach_probe(struct vfs_bpf *obj) +{ + obj->links.netdata_vfs_write_kprobe = bpf_program__attach_kprobe(obj->progs.netdata_vfs_write_kprobe, false, + vfs_targets[NETDATA_EBPF_VFS_WRITE].name); + int ret = libbpf_get_error(obj->links.netdata_vfs_write_kprobe); + if (ret) + return -1; + + obj->links.netdata_vfs_write_kretprobe = bpf_program__attach_kprobe(obj->progs.netdata_vfs_write_kretprobe, true, + vfs_targets[NETDATA_EBPF_VFS_WRITE].name); + ret = libbpf_get_error(obj->links.netdata_vfs_write_kretprobe); + if (ret) + return -1; + + obj->links.netdata_vfs_writev_kprobe = bpf_program__attach_kprobe(obj->progs.netdata_vfs_writev_kprobe, false, + vfs_targets[NETDATA_EBPF_VFS_WRITEV].name); + ret = libbpf_get_error(obj->links.netdata_vfs_writev_kprobe); + if (ret) + return -1; + + obj->links.netdata_vfs_writev_kretprobe = bpf_program__attach_kprobe(obj->progs.netdata_vfs_writev_kretprobe, true, + vfs_targets[NETDATA_EBPF_VFS_WRITEV].name); + ret = libbpf_get_error(obj->links.netdata_vfs_writev_kretprobe); + if (ret) + return -1; + + obj->links.netdata_vfs_read_kprobe = bpf_program__attach_kprobe(obj->progs.netdata_vfs_read_kprobe, false, + vfs_targets[NETDATA_EBPF_VFS_READ].name); + ret = libbpf_get_error(obj->links.netdata_vfs_read_kprobe); + if (ret) + return -1; + + obj->links.netdata_vfs_read_kretprobe = bpf_program__attach_kprobe(obj->progs.netdata_vfs_read_kretprobe, true, + vfs_targets[NETDATA_EBPF_VFS_READ].name); + ret = libbpf_get_error(obj->links.netdata_vfs_read_kretprobe); + if (ret) + return -1; + + obj->links.netdata_vfs_readv_kprobe = bpf_program__attach_kprobe(obj->progs.netdata_vfs_readv_kprobe, false, + vfs_targets[NETDATA_EBPF_VFS_READV].name); + ret = libbpf_get_error(obj->links.netdata_vfs_readv_kprobe); + if (ret) + return -1; + + obj->links.netdata_vfs_readv_kretprobe = bpf_program__attach_kprobe(obj->progs.netdata_vfs_readv_kretprobe, true, + vfs_targets[NETDATA_EBPF_VFS_READV].name); + ret = libbpf_get_error(obj->links.netdata_vfs_readv_kretprobe); + if (ret) + return -1; + + obj->links.netdata_vfs_unlink_kprobe = bpf_program__attach_kprobe(obj->progs.netdata_vfs_unlink_kprobe, false, + vfs_targets[NETDATA_EBPF_VFS_UNLINK].name); + ret = libbpf_get_error(obj->links.netdata_vfs_unlink_kprobe); + if (ret) + return -1; + + obj->links.netdata_vfs_unlink_kretprobe = bpf_program__attach_kprobe(obj->progs.netdata_vfs_unlink_kretprobe, true, + vfs_targets[NETDATA_EBPF_VFS_UNLINK].name); + ret = libbpf_get_error(obj->links.netdata_vfs_unlink_kretprobe); + if (ret) + return -1; + + obj->links.netdata_vfs_fsync_kprobe = bpf_program__attach_kprobe(obj->progs.netdata_vfs_fsync_kprobe, false, + vfs_targets[NETDATA_EBPF_VFS_FSYNC].name); + ret = libbpf_get_error(obj->links.netdata_vfs_fsync_kprobe); + if (ret) + return -1; + + obj->links.netdata_vfs_fsync_kretprobe = bpf_program__attach_kprobe(obj->progs.netdata_vfs_fsync_kretprobe, true, + vfs_targets[NETDATA_EBPF_VFS_FSYNC].name); + ret = libbpf_get_error(obj->links.netdata_vfs_fsync_kretprobe); + if (ret) + return -1; + + obj->links.netdata_vfs_open_kprobe = bpf_program__attach_kprobe(obj->progs.netdata_vfs_open_kprobe, false, + vfs_targets[NETDATA_EBPF_VFS_OPEN].name); + ret = libbpf_get_error(obj->links.netdata_vfs_open_kprobe); + if (ret) + return -1; + + obj->links.netdata_vfs_open_kretprobe = bpf_program__attach_kprobe(obj->progs.netdata_vfs_open_kretprobe, true, + vfs_targets[NETDATA_EBPF_VFS_OPEN].name); + ret = libbpf_get_error(obj->links.netdata_vfs_open_kretprobe); + if (ret) + return -1; + + obj->links.netdata_vfs_create_kprobe = bpf_program__attach_kprobe(obj->progs.netdata_vfs_create_kprobe, false, + vfs_targets[NETDATA_EBPF_VFS_CREATE].name); + ret = libbpf_get_error(obj->links.netdata_vfs_create_kprobe); + if (ret) + return -1; + + obj->links.netdata_vfs_create_kretprobe = bpf_program__attach_kprobe(obj->progs.netdata_vfs_create_kretprobe, true, + vfs_targets[NETDATA_EBPF_VFS_CREATE].name); + ret = libbpf_get_error(obj->links.netdata_vfs_create_kretprobe); + if (ret) + return -1; + + obj->links.netdata_vfs_fsync_kprobe = bpf_program__attach_kprobe(obj->progs.netdata_vfs_fsync_kprobe, false, + vfs_targets[NETDATA_EBPF_VFS_FSYNC].name); + ret = libbpf_get_error(obj->links.netdata_vfs_fsync_kprobe); + if (ret) + return -1; + + obj->links.netdata_vfs_fsync_kretprobe = bpf_program__attach_kprobe(obj->progs.netdata_vfs_fsync_kretprobe, true, + vfs_targets[NETDATA_EBPF_VFS_FSYNC].name); + ret = libbpf_get_error(obj->links.netdata_vfs_fsync_kretprobe); + if (ret) + return -1; + + obj->links.netdata_vfs_open_kprobe = bpf_program__attach_kprobe(obj->progs.netdata_vfs_open_kprobe, false, + vfs_targets[NETDATA_EBPF_VFS_OPEN].name); + ret = libbpf_get_error(obj->links.netdata_vfs_open_kprobe); + if (ret) + return -1; + + obj->links.netdata_vfs_open_kretprobe = bpf_program__attach_kprobe(obj->progs.netdata_vfs_open_kretprobe, true, + vfs_targets[NETDATA_EBPF_VFS_OPEN].name); + ret = libbpf_get_error(obj->links.netdata_vfs_open_kretprobe); + if (ret) + return -1; + + obj->links.netdata_vfs_create_kprobe = bpf_program__attach_kprobe(obj->progs.netdata_vfs_create_kprobe, false, + vfs_targets[NETDATA_EBPF_VFS_CREATE].name); + ret = libbpf_get_error(obj->links.netdata_vfs_create_kprobe); + if (ret) + return -1; + + obj->links.netdata_vfs_create_kretprobe = bpf_program__attach_kprobe(obj->progs.netdata_vfs_create_kretprobe, true, + vfs_targets[NETDATA_EBPF_VFS_CREATE].name); + ret = libbpf_get_error(obj->links.netdata_vfs_create_kretprobe); + if (ret) + return -1; + + obj->links.netdata_vfs_release_task_kprobe = bpf_program__attach_kprobe(obj->progs.netdata_vfs_release_task_fentry, + true, + EBPF_COMMON_FNCT_CLEAN_UP); + ret = libbpf_get_error(obj->links.netdata_vfs_release_task_kprobe); + if (ret) + return -1; + + return 0; +} + +/** + * Adjust Map Size + * + * Resize maps according input from users. + * + * @param obj is the main structure for bpf objects. + * @param em structure with configuration + */ +static void ebpf_vfs_adjust_map_size(struct vfs_bpf *obj, ebpf_module_t *em) +{ + ebpf_update_map_size(obj->maps.tbl_vfs_pid, &vfs_maps[NETDATA_VFS_PID], + em, bpf_map__name(obj->maps.tbl_vfs_pid)); +} + +/** + * Set hash tables + * + * Set the values for maps according the value given by kernel. + * + * @param obj is the main structure for bpf objects. + */ +static void ebpf_vfs_set_hash_tables(struct vfs_bpf *obj) +{ + vfs_maps[NETDATA_VFS_ALL].map_fd = bpf_map__fd(obj->maps.tbl_vfs_stats); + vfs_maps[NETDATA_VFS_PID].map_fd = bpf_map__fd(obj->maps.tbl_vfs_pid); + vfs_maps[NETDATA_VFS_CTRL].map_fd = bpf_map__fd(obj->maps.vfs_ctrl); +} + +/** + * Disable Release Task + * + * Disable release task when apps is not enabled. + * + * @param obj is the main structure for bpf objects. + */ +static void ebpf_vfs_disable_release_task(struct vfs_bpf *obj) +{ + bpf_program__set_autoload(obj->progs.netdata_vfs_release_task_fentry, false); + bpf_program__set_autoload(obj->progs.netdata_vfs_release_task_kprobe, false); +} + +/** + * Load and attach + * + * Load and attach the eBPF code in kernel. + * + * @param obj is the main structure for bpf objects. + * @param em structure with configuration + * + * @return it returns 0 on succes and -1 otherwise + */ +static inline int ebpf_vfs_load_and_attach(struct vfs_bpf *obj, ebpf_module_t *em) +{ + netdata_ebpf_targets_t *mt = em->targets; + netdata_ebpf_program_loaded_t test = mt[NETDATA_EBPF_VFS_WRITE].mode; + + if (test == EBPF_LOAD_TRAMPOLINE) { + ebpf_vfs_disable_probes(obj); + + ebpf_vfs_set_trampoline_target(obj); + } else { + ebpf_vfs_disable_trampoline(obj); + } + + ebpf_vfs_adjust_map_size(obj, em); + + if (!em->apps_charts && !em->cgroup_charts) + ebpf_vfs_disable_release_task(obj); + + int ret = vfs_bpf__load(obj); + if (ret) { + return ret; + } + + ret = (test == EBPF_LOAD_TRAMPOLINE) ? vfs_bpf__attach(obj) : ebpf_vfs_attach_probe(obj); + if (!ret) { + ebpf_vfs_set_hash_tables(obj); + + ebpf_update_controller(vfs_maps[NETDATA_VFS_CTRL].map_fd, em); + } + + return ret; +} +#endif + +/***************************************************************** + * + * FUNCTIONS TO CLOSE THE THREAD + * + *****************************************************************/ + +/** + * Cachestat Free + * + * Cleanup variables after child threads to stop + * + * @param ptr thread data. + */ +static void ebpf_vfs_free(ebpf_module_t *em) +{ + pthread_mutex_lock(&ebpf_exit_cleanup); + if (em->thread->enabled == NETDATA_THREAD_EBPF_RUNNING) { + em->thread->enabled = NETDATA_THREAD_EBPF_STOPPING; + pthread_mutex_unlock(&ebpf_exit_cleanup); + return; + } + pthread_mutex_unlock(&ebpf_exit_cleanup); + + freez(vfs_hash_values); + freez(vfs_vector); + freez(vfs_threads.thread); + +#ifdef LIBBPF_MAJOR_VERSION + if (bpf_obj) + vfs_bpf__destroy(bpf_obj); +#endif + + pthread_mutex_lock(&ebpf_exit_cleanup); + em->thread->enabled = NETDATA_THREAD_EBPF_STOPPED; + pthread_mutex_unlock(&ebpf_exit_cleanup); +} + +/** + * Exit + * + * Cancel thread and exit. + * + * @param ptr thread data. +**/ +static void ebpf_vfs_exit(void *ptr) +{ + ebpf_module_t *em = (ebpf_module_t *)ptr; + netdata_thread_cancel(*vfs_threads.thread); + ebpf_vfs_free(em); +} + +/** +* Clean up the main thread. +* +* @param ptr thread data. +**/ +static void ebpf_vfs_cleanup(void *ptr) +{ + ebpf_module_t *em = (ebpf_module_t *)ptr; + ebpf_vfs_free(em); +} + +/***************************************************************** + * + * FUNCTIONS WITH THE MAIN LOOP + * + *****************************************************************/ + +/** + * Send data to Netdata calling auxiliary functions. + * + * @param em the structure with thread information +*/ +static void ebpf_vfs_send_data(ebpf_module_t *em) +{ + netdata_publish_vfs_common_t pvc; + + pvc.write = (long)vfs_aggregated_data[NETDATA_KEY_PUBLISH_VFS_WRITE].bytes; + pvc.read = (long)vfs_aggregated_data[NETDATA_KEY_PUBLISH_VFS_READ].bytes; + + write_count_chart(NETDATA_VFS_FILE_CLEAN_COUNT, NETDATA_FILESYSTEM_FAMILY, + &vfs_publish_aggregated[NETDATA_KEY_PUBLISH_VFS_UNLINK], 1); + + write_count_chart(NETDATA_VFS_FILE_IO_COUNT, NETDATA_FILESYSTEM_FAMILY, + &vfs_publish_aggregated[NETDATA_KEY_PUBLISH_VFS_READ], 2); + + if (em->mode < MODE_ENTRY) { + write_err_chart(NETDATA_VFS_FILE_ERR_COUNT, NETDATA_FILESYSTEM_FAMILY, + &vfs_publish_aggregated[NETDATA_KEY_PUBLISH_VFS_READ], 2); + } + + write_io_chart(NETDATA_VFS_IO_FILE_BYTES, NETDATA_FILESYSTEM_FAMILY, vfs_id_names[NETDATA_KEY_PUBLISH_VFS_WRITE], + (long long)pvc.write, vfs_id_names[NETDATA_KEY_PUBLISH_VFS_READ], (long long)pvc.read); + + write_count_chart(NETDATA_VFS_FSYNC, NETDATA_FILESYSTEM_FAMILY, + &vfs_publish_aggregated[NETDATA_KEY_PUBLISH_VFS_FSYNC], 1); + + if (em->mode < MODE_ENTRY) { + write_err_chart(NETDATA_VFS_FSYNC_ERR, NETDATA_FILESYSTEM_FAMILY, + &vfs_publish_aggregated[NETDATA_KEY_PUBLISH_VFS_FSYNC], 1); + } + + write_count_chart(NETDATA_VFS_OPEN, NETDATA_FILESYSTEM_FAMILY, + &vfs_publish_aggregated[NETDATA_KEY_PUBLISH_VFS_OPEN], 1); + + if (em->mode < MODE_ENTRY) { + write_err_chart(NETDATA_VFS_OPEN_ERR, NETDATA_FILESYSTEM_FAMILY, + &vfs_publish_aggregated[NETDATA_KEY_PUBLISH_VFS_OPEN], 1); + } + + write_count_chart(NETDATA_VFS_CREATE, NETDATA_FILESYSTEM_FAMILY, + &vfs_publish_aggregated[NETDATA_KEY_PUBLISH_VFS_CREATE], 1); + + if (em->mode < MODE_ENTRY) { + write_err_chart( + NETDATA_VFS_CREATE_ERR, + NETDATA_FILESYSTEM_FAMILY, + &vfs_publish_aggregated[NETDATA_KEY_PUBLISH_VFS_CREATE], + 1); + } +} + +/** + * Read the hash table and store data to allocated vectors. + */ +static void read_global_table() +{ + uint64_t idx; + netdata_idx_t res[NETDATA_VFS_COUNTER]; + + netdata_idx_t *val = vfs_hash_values; + int fd = vfs_maps[NETDATA_VFS_ALL].map_fd; + for (idx = 0; idx < NETDATA_VFS_COUNTER; idx++) { + uint64_t total = 0; + if (!bpf_map_lookup_elem(fd, &idx, val)) { + int i; + int end = ebpf_nprocs; + for (i = 0; i < end; i++) + total += val[i]; + } + res[idx] = total; + } + + vfs_publish_aggregated[NETDATA_KEY_PUBLISH_VFS_UNLINK].ncall = res[NETDATA_KEY_CALLS_VFS_UNLINK]; + vfs_publish_aggregated[NETDATA_KEY_PUBLISH_VFS_READ].ncall = res[NETDATA_KEY_CALLS_VFS_READ] + + res[NETDATA_KEY_CALLS_VFS_READV]; + vfs_publish_aggregated[NETDATA_KEY_PUBLISH_VFS_WRITE].ncall = res[NETDATA_KEY_CALLS_VFS_WRITE] + + res[NETDATA_KEY_CALLS_VFS_WRITEV]; + vfs_publish_aggregated[NETDATA_KEY_PUBLISH_VFS_FSYNC].ncall = res[NETDATA_KEY_CALLS_VFS_FSYNC]; + vfs_publish_aggregated[NETDATA_KEY_PUBLISH_VFS_OPEN].ncall = res[NETDATA_KEY_CALLS_VFS_OPEN]; + vfs_publish_aggregated[NETDATA_KEY_PUBLISH_VFS_CREATE].ncall = res[NETDATA_KEY_CALLS_VFS_CREATE]; + + vfs_publish_aggregated[NETDATA_KEY_PUBLISH_VFS_UNLINK].nerr = res[NETDATA_KEY_ERROR_VFS_UNLINK]; + vfs_publish_aggregated[NETDATA_KEY_PUBLISH_VFS_READ].nerr = res[NETDATA_KEY_ERROR_VFS_READ] + + res[NETDATA_KEY_ERROR_VFS_READV]; + vfs_publish_aggregated[NETDATA_KEY_PUBLISH_VFS_WRITE].nerr = res[NETDATA_KEY_ERROR_VFS_WRITE] + + res[NETDATA_KEY_ERROR_VFS_WRITEV]; + vfs_publish_aggregated[NETDATA_KEY_PUBLISH_VFS_FSYNC].nerr = res[NETDATA_KEY_ERROR_VFS_FSYNC]; + vfs_publish_aggregated[NETDATA_KEY_PUBLISH_VFS_OPEN].nerr = res[NETDATA_KEY_ERROR_VFS_OPEN]; + vfs_publish_aggregated[NETDATA_KEY_PUBLISH_VFS_CREATE].nerr = res[NETDATA_KEY_ERROR_VFS_CREATE]; + + vfs_aggregated_data[NETDATA_KEY_PUBLISH_VFS_WRITE].bytes = (uint64_t)res[NETDATA_KEY_BYTES_VFS_WRITE] + + (uint64_t)res[NETDATA_KEY_BYTES_VFS_WRITEV]; + vfs_aggregated_data[NETDATA_KEY_PUBLISH_VFS_READ].bytes = (uint64_t)res[NETDATA_KEY_BYTES_VFS_READ] + + (uint64_t)res[NETDATA_KEY_BYTES_VFS_READV]; +} + +/** + * Sum PIDs + * + * Sum values for all targets. + * + * @param swap output structure + * @param root link list with structure to be used + */ +static void ebpf_vfs_sum_pids(netdata_publish_vfs_t *vfs, struct pid_on_target *root) +{ + netdata_publish_vfs_t accumulator; + memset(&accumulator, 0, sizeof(accumulator)); + + while (root) { + int32_t pid = root->pid; + netdata_publish_vfs_t *w = vfs_pid[pid]; + if (w) { + accumulator.write_call += w->write_call; + accumulator.writev_call += w->writev_call; + accumulator.read_call += w->read_call; + accumulator.readv_call += w->readv_call; + accumulator.unlink_call += w->unlink_call; + accumulator.fsync_call += w->fsync_call; + accumulator.open_call += w->open_call; + accumulator.create_call += w->create_call; + + accumulator.write_bytes += w->write_bytes; + accumulator.writev_bytes += w->writev_bytes; + accumulator.read_bytes += w->read_bytes; + accumulator.readv_bytes += w->readv_bytes; + + accumulator.write_err += w->write_err; + accumulator.writev_err += w->writev_err; + accumulator.read_err += w->read_err; + accumulator.readv_err += w->readv_err; + accumulator.unlink_err += w->unlink_err; + accumulator.fsync_err += w->fsync_err; + accumulator.open_err += w->open_err; + accumulator.create_err += w->create_err; + } + root = root->next; + } + + // These conditions were added, because we are using incremental algorithm + vfs->write_call = (accumulator.write_call >= vfs->write_call) ? accumulator.write_call : vfs->write_call; + vfs->writev_call = (accumulator.writev_call >= vfs->writev_call) ? accumulator.writev_call : vfs->writev_call; + vfs->read_call = (accumulator.read_call >= vfs->read_call) ? accumulator.read_call : vfs->read_call; + vfs->readv_call = (accumulator.readv_call >= vfs->readv_call) ? accumulator.readv_call : vfs->readv_call; + vfs->unlink_call = (accumulator.unlink_call >= vfs->unlink_call) ? accumulator.unlink_call : vfs->unlink_call; + vfs->fsync_call = (accumulator.fsync_call >= vfs->fsync_call) ? accumulator.fsync_call : vfs->fsync_call; + vfs->open_call = (accumulator.open_call >= vfs->open_call) ? accumulator.open_call : vfs->open_call; + vfs->create_call = (accumulator.create_call >= vfs->create_call) ? accumulator.create_call : vfs->create_call; + + vfs->write_bytes = (accumulator.write_bytes >= vfs->write_bytes) ? accumulator.write_bytes : vfs->write_bytes; + vfs->writev_bytes = (accumulator.writev_bytes >= vfs->writev_bytes) ? accumulator.writev_bytes : vfs->writev_bytes; + vfs->read_bytes = (accumulator.read_bytes >= vfs->read_bytes) ? accumulator.read_bytes : vfs->read_bytes; + vfs->readv_bytes = (accumulator.readv_bytes >= vfs->readv_bytes) ? accumulator.readv_bytes : vfs->readv_bytes; + + vfs->write_err = (accumulator.write_err >= vfs->write_err) ? accumulator.write_err : vfs->write_err; + vfs->writev_err = (accumulator.writev_err >= vfs->writev_err) ? accumulator.writev_err : vfs->writev_err; + vfs->read_err = (accumulator.read_err >= vfs->read_err) ? accumulator.read_err : vfs->read_err; + vfs->readv_err = (accumulator.readv_err >= vfs->readv_err) ? accumulator.readv_err : vfs->readv_err; + vfs->unlink_err = (accumulator.unlink_err >= vfs->unlink_err) ? accumulator.unlink_err : vfs->unlink_err; + vfs->fsync_err = (accumulator.fsync_err >= vfs->fsync_err) ? accumulator.fsync_err : vfs->fsync_err; + vfs->open_err = (accumulator.open_err >= vfs->open_err) ? accumulator.open_err : vfs->open_err; + vfs->create_err = (accumulator.create_err >= vfs->create_err) ? accumulator.create_err : vfs->create_err; +} + +/** + * Send data to Netdata calling auxiliary functions. + * + * @param em the structure with thread information + * @param root the target list. + */ +void ebpf_vfs_send_apps_data(ebpf_module_t *em, struct target *root) +{ + struct target *w; + for (w = root; w; w = w->next) { + if (unlikely(w->exposed && w->processes)) { + ebpf_vfs_sum_pids(&w->vfs, w->root_pid); + } + } + + write_begin_chart(NETDATA_APPS_FAMILY, NETDATA_SYSCALL_APPS_FILE_DELETED); + for (w = root; w; w = w->next) { + if (unlikely(w->exposed && w->processes)) { + write_chart_dimension(w->name, w->vfs.unlink_call); + } + } + write_end_chart(); + + write_begin_chart(NETDATA_APPS_FAMILY, NETDATA_SYSCALL_APPS_VFS_WRITE_CALLS); + for (w = root; w; w = w->next) { + if (unlikely(w->exposed && w->processes)) { + write_chart_dimension(w->name, w->vfs.write_call + w->vfs.writev_call); + } + } + write_end_chart(); + + if (em->mode < MODE_ENTRY) { + write_begin_chart(NETDATA_APPS_FAMILY, NETDATA_SYSCALL_APPS_VFS_WRITE_CALLS_ERROR); + for (w = root; w; w = w->next) { + if (unlikely(w->exposed && w->processes)) { + write_chart_dimension(w->name, w->vfs.write_err + w->vfs.writev_err); + } + } + write_end_chart(); + } + + write_begin_chart(NETDATA_APPS_FAMILY, NETDATA_SYSCALL_APPS_VFS_READ_CALLS); + for (w = root; w; w = w->next) { + if (unlikely(w->exposed && w->processes)) { + write_chart_dimension(w->name, w->vfs.read_call + w->vfs.readv_call); + } + } + write_end_chart(); + + if (em->mode < MODE_ENTRY) { + write_begin_chart(NETDATA_APPS_FAMILY, NETDATA_SYSCALL_APPS_VFS_READ_CALLS_ERROR); + for (w = root; w; w = w->next) { + if (unlikely(w->exposed && w->processes)) { + write_chart_dimension(w->name, w->vfs.read_err + w->vfs.readv_err); + } + } + write_end_chart(); + } + + write_begin_chart(NETDATA_APPS_FAMILY, NETDATA_SYSCALL_APPS_VFS_WRITE_BYTES); + for (w = root; w; w = w->next) { + if (unlikely(w->exposed && w->processes)) { + write_chart_dimension(w->name, w->vfs.write_bytes + w->vfs.writev_bytes); + } + } + write_end_chart(); + + write_begin_chart(NETDATA_APPS_FAMILY, NETDATA_SYSCALL_APPS_VFS_READ_BYTES); + for (w = root; w; w = w->next) { + if (unlikely(w->exposed && w->processes)) { + write_chart_dimension(w->name, w->vfs.read_bytes + w->vfs.readv_bytes); + } + } + write_end_chart(); + + write_begin_chart(NETDATA_APPS_FAMILY, NETDATA_SYSCALL_APPS_VFS_FSYNC); + for (w = root; w; w = w->next) { + if (unlikely(w->exposed && w->processes)) { + write_chart_dimension(w->name, w->vfs.fsync_call); + } + } + write_end_chart(); + + if (em->mode < MODE_ENTRY) { + write_begin_chart(NETDATA_APPS_FAMILY, NETDATA_SYSCALL_APPS_VFS_FSYNC_CALLS_ERROR); + for (w = root; w; w = w->next) { + if (unlikely(w->exposed && w->processes)) { + write_chart_dimension(w->name, w->vfs.fsync_err); + } + } + write_end_chart(); + } + + write_begin_chart(NETDATA_APPS_FAMILY, NETDATA_SYSCALL_APPS_VFS_OPEN); + for (w = root; w; w = w->next) { + if (unlikely(w->exposed && w->processes)) { + write_chart_dimension(w->name, w->vfs.open_call); + } + } + write_end_chart(); + + if (em->mode < MODE_ENTRY) { + write_begin_chart(NETDATA_APPS_FAMILY, NETDATA_SYSCALL_APPS_VFS_OPEN_CALLS_ERROR); + for (w = root; w; w = w->next) { + if (unlikely(w->exposed && w->processes)) { + write_chart_dimension(w->name, w->vfs.open_err); + } + } + write_end_chart(); + } + + write_begin_chart(NETDATA_APPS_FAMILY, NETDATA_SYSCALL_APPS_VFS_CREATE); + for (w = root; w; w = w->next) { + if (unlikely(w->exposed && w->processes)) { + write_chart_dimension(w->name, w->vfs.create_call); + } + } + write_end_chart(); + + if (em->mode < MODE_ENTRY) { + write_begin_chart(NETDATA_APPS_FAMILY, NETDATA_SYSCALL_APPS_VFS_CREATE_CALLS_ERROR); + for (w = root; w; w = w->next) { + if (unlikely(w->exposed && w->processes)) { + write_chart_dimension(w->name, w->vfs.create_err); + } + } + write_end_chart(); + } +} + +/** + * Apps Accumulator + * + * Sum all values read from kernel and store in the first address. + * + * @param out the vector with read values. + */ +static void vfs_apps_accumulator(netdata_publish_vfs_t *out) +{ + int i, end = (running_on_kernel >= NETDATA_KERNEL_V4_15) ? ebpf_nprocs : 1; + netdata_publish_vfs_t *total = &out[0]; + for (i = 1; i < end; i++) { + netdata_publish_vfs_t *w = &out[i]; + + total->write_call += w->write_call; + total->writev_call += w->writev_call; + total->read_call += w->read_call; + total->readv_call += w->readv_call; + total->unlink_call += w->unlink_call; + + total->write_bytes += w->write_bytes; + total->writev_bytes += w->writev_bytes; + total->read_bytes += w->read_bytes; + total->readv_bytes += w->readv_bytes; + + total->write_err += w->write_err; + total->writev_err += w->writev_err; + total->read_err += w->read_err; + total->readv_err += w->readv_err; + total->unlink_err += w->unlink_err; + } +} + +/** + * Fill PID + * + * Fill PID structures + * + * @param current_pid pid that we are collecting data + * @param out values read from hash tables; + */ +static void vfs_fill_pid(uint32_t current_pid, netdata_publish_vfs_t *publish) +{ + netdata_publish_vfs_t *curr = vfs_pid[current_pid]; + if (!curr) { + curr = callocz(1, sizeof(netdata_publish_vfs_t)); + vfs_pid[current_pid] = curr; + } + + memcpy(curr, &publish[0], sizeof(netdata_publish_vfs_t)); +} + +/** + * Read the hash table and store data to allocated vectors. + */ +static void ebpf_vfs_read_apps() +{ + struct pid_stat *pids = root_of_pids; + netdata_publish_vfs_t *vv = vfs_vector; + int fd = vfs_maps[NETDATA_VFS_PID].map_fd; + size_t length = sizeof(netdata_publish_vfs_t) * ebpf_nprocs; + while (pids) { + uint32_t key = pids->pid; + + if (bpf_map_lookup_elem(fd, &key, vv)) { + pids = pids->next; + continue; + } + + vfs_apps_accumulator(vv); + + vfs_fill_pid(key, vv); + + // We are cleaning to avoid passing data read from one process to other. + memset(vv, 0, length); + + pids = pids->next; + } +} + +/** + * Update cgroup + * + * Update cgroup data based in + */ +static void read_update_vfs_cgroup() +{ + ebpf_cgroup_target_t *ect ; + netdata_publish_vfs_t *vv = vfs_vector; + int fd = vfs_maps[NETDATA_VFS_PID].map_fd; + size_t length = sizeof(netdata_publish_vfs_t) * ebpf_nprocs; + + pthread_mutex_lock(&mutex_cgroup_shm); + for (ect = ebpf_cgroup_pids; ect; ect = ect->next) { + struct pid_on_target2 *pids; + for (pids = ect->pids; pids; pids = pids->next) { + int pid = pids->pid; + netdata_publish_vfs_t *out = &pids->vfs; + if (likely(vfs_pid) && vfs_pid[pid]) { + netdata_publish_vfs_t *in = vfs_pid[pid]; + + memcpy(out, in, sizeof(netdata_publish_vfs_t)); + } else { + memset(vv, 0, length); + if (!bpf_map_lookup_elem(fd, &pid, vv)) { + vfs_apps_accumulator(vv); + + memcpy(out, vv, sizeof(netdata_publish_vfs_t)); + } + } + } + } + pthread_mutex_unlock(&mutex_cgroup_shm); +} + +/** + * VFS read hash + * + * This is the thread callback. + * This thread is necessary, because we cannot freeze the whole plugin to read the data. + * + * @param ptr It is a NULL value for this thread. + * + * @return It always returns NULL. + */ +void *ebpf_vfs_read_hash(void *ptr) +{ + netdata_thread_cleanup_push(ebpf_vfs_cleanup, ptr); + heartbeat_t hb; + heartbeat_init(&hb); + + ebpf_module_t *em = (ebpf_module_t *)ptr; + + usec_t step = NETDATA_LATENCY_VFS_SLEEP_MS * em->update_every; + //This will be cancelled by its parent + while (!ebpf_exit_plugin) { + (void)heartbeat_next(&hb, step); + + read_global_table(); + } + + netdata_thread_cleanup_pop(1); + return NULL; +} + +/** + * Sum PIDs + * + * Sum values for all targets. + * + * @param vfs structure used to store data + * @param pids input data + */ +static void ebpf_vfs_sum_cgroup_pids(netdata_publish_vfs_t *vfs, struct pid_on_target2 *pids) + { + netdata_publish_vfs_t accumulator; + memset(&accumulator, 0, sizeof(accumulator)); + + while (pids) { + netdata_publish_vfs_t *w = &pids->vfs; + + accumulator.write_call += w->write_call; + accumulator.writev_call += w->writev_call; + accumulator.read_call += w->read_call; + accumulator.readv_call += w->readv_call; + accumulator.unlink_call += w->unlink_call; + accumulator.fsync_call += w->fsync_call; + accumulator.open_call += w->open_call; + accumulator.create_call += w->create_call; + + accumulator.write_bytes += w->write_bytes; + accumulator.writev_bytes += w->writev_bytes; + accumulator.read_bytes += w->read_bytes; + accumulator.readv_bytes += w->readv_bytes; + + accumulator.write_err += w->write_err; + accumulator.writev_err += w->writev_err; + accumulator.read_err += w->read_err; + accumulator.readv_err += w->readv_err; + accumulator.unlink_err += w->unlink_err; + accumulator.fsync_err += w->fsync_err; + accumulator.open_err += w->open_err; + accumulator.create_err += w->create_err; + + pids = pids->next; + } + + // These conditions were added, because we are using incremental algorithm + vfs->write_call = (accumulator.write_call >= vfs->write_call) ? accumulator.write_call : vfs->write_call; + vfs->writev_call = (accumulator.writev_call >= vfs->writev_call) ? accumulator.writev_call : vfs->writev_call; + vfs->read_call = (accumulator.read_call >= vfs->read_call) ? accumulator.read_call : vfs->read_call; + vfs->readv_call = (accumulator.readv_call >= vfs->readv_call) ? accumulator.readv_call : vfs->readv_call; + vfs->unlink_call = (accumulator.unlink_call >= vfs->unlink_call) ? accumulator.unlink_call : vfs->unlink_call; + vfs->fsync_call = (accumulator.fsync_call >= vfs->fsync_call) ? accumulator.fsync_call : vfs->fsync_call; + vfs->open_call = (accumulator.open_call >= vfs->open_call) ? accumulator.open_call : vfs->open_call; + vfs->create_call = (accumulator.create_call >= vfs->create_call) ? accumulator.create_call : vfs->create_call; + + vfs->write_bytes = (accumulator.write_bytes >= vfs->write_bytes) ? accumulator.write_bytes : vfs->write_bytes; + vfs->writev_bytes = (accumulator.writev_bytes >= vfs->writev_bytes) ? accumulator.writev_bytes : vfs->writev_bytes; + vfs->read_bytes = (accumulator.read_bytes >= vfs->read_bytes) ? accumulator.read_bytes : vfs->read_bytes; + vfs->readv_bytes = (accumulator.readv_bytes >= vfs->readv_bytes) ? accumulator.readv_bytes : vfs->readv_bytes; + + vfs->write_err = (accumulator.write_err >= vfs->write_err) ? accumulator.write_err : vfs->write_err; + vfs->writev_err = (accumulator.writev_err >= vfs->writev_err) ? accumulator.writev_err : vfs->writev_err; + vfs->read_err = (accumulator.read_err >= vfs->read_err) ? accumulator.read_err : vfs->read_err; + vfs->readv_err = (accumulator.readv_err >= vfs->readv_err) ? accumulator.readv_err : vfs->readv_err; + vfs->unlink_err = (accumulator.unlink_err >= vfs->unlink_err) ? accumulator.unlink_err : vfs->unlink_err; + vfs->fsync_err = (accumulator.fsync_err >= vfs->fsync_err) ? accumulator.fsync_err : vfs->fsync_err; + vfs->open_err = (accumulator.open_err >= vfs->open_err) ? accumulator.open_err : vfs->open_err; + vfs->create_err = (accumulator.create_err >= vfs->create_err) ? accumulator.create_err : vfs->create_err; +} + +/** + * Create specific VFS charts + * + * Create charts for cgroup/application. + * + * @param type the chart type. + * @param em the main thread structure. + */ +static void ebpf_create_specific_vfs_charts(char *type, ebpf_module_t *em) +{ + ebpf_create_chart(type, NETDATA_SYSCALL_APPS_FILE_DELETED,"Files deleted", + EBPF_COMMON_DIMENSION_CALL, NETDATA_VFS_CGROUP_GROUP, NETDATA_CGROUP_VFS_UNLINK_CONTEXT, + NETDATA_EBPF_CHART_TYPE_LINE, NETDATA_CHART_PRIO_CGROUPS_CONTAINERS + 5500, + ebpf_create_global_dimension, &vfs_publish_aggregated[NETDATA_KEY_PUBLISH_VFS_UNLINK], + 1, em->update_every, NETDATA_EBPF_MODULE_NAME_SWAP); + + ebpf_create_chart(type, NETDATA_SYSCALL_APPS_VFS_WRITE_CALLS, "Write to disk", + EBPF_COMMON_DIMENSION_CALL, NETDATA_VFS_CGROUP_GROUP, NETDATA_CGROUP_VFS_WRITE_CONTEXT, + NETDATA_EBPF_CHART_TYPE_LINE, NETDATA_CHART_PRIO_CGROUPS_CONTAINERS + 5501, + ebpf_create_global_dimension, &vfs_publish_aggregated[NETDATA_KEY_PUBLISH_VFS_WRITE], + 1, em->update_every, NETDATA_EBPF_MODULE_NAME_SWAP); + + if (em->mode < MODE_ENTRY) { + ebpf_create_chart(type, NETDATA_SYSCALL_APPS_VFS_WRITE_CALLS_ERROR, "Fails to write", + EBPF_COMMON_DIMENSION_CALL, NETDATA_VFS_CGROUP_GROUP, NETDATA_CGROUP_VFS_WRITE_ERROR_CONTEXT, + NETDATA_EBPF_CHART_TYPE_LINE, NETDATA_CHART_PRIO_CGROUPS_CONTAINERS + 5502, + ebpf_create_global_dimension, &vfs_publish_aggregated[NETDATA_KEY_PUBLISH_VFS_WRITE], + 1, em->update_every, NETDATA_EBPF_MODULE_NAME_SWAP); + } + + ebpf_create_chart(type, NETDATA_SYSCALL_APPS_VFS_READ_CALLS, "Read from disk", + EBPF_COMMON_DIMENSION_CALL, NETDATA_VFS_CGROUP_GROUP, NETDATA_CGROUP_VFS_READ_CONTEXT, + NETDATA_EBPF_CHART_TYPE_LINE, NETDATA_CHART_PRIO_CGROUPS_CONTAINERS + 5503, + ebpf_create_global_dimension, &vfs_publish_aggregated[NETDATA_KEY_PUBLISH_VFS_READ], + 1, em->update_every, NETDATA_EBPF_MODULE_NAME_SWAP); + + if (em->mode < MODE_ENTRY) { + ebpf_create_chart(type, NETDATA_SYSCALL_APPS_VFS_READ_CALLS_ERROR, "Fails to read", + EBPF_COMMON_DIMENSION_CALL, NETDATA_VFS_CGROUP_GROUP, NETDATA_CGROUP_VFS_READ_ERROR_CONTEXT, + NETDATA_EBPF_CHART_TYPE_LINE, NETDATA_CHART_PRIO_CGROUPS_CONTAINERS + 5504, + ebpf_create_global_dimension, &vfs_publish_aggregated[NETDATA_KEY_PUBLISH_VFS_READ], + 1, em->update_every, NETDATA_EBPF_MODULE_NAME_SWAP); + } + + ebpf_create_chart(type, NETDATA_SYSCALL_APPS_VFS_WRITE_BYTES, "Bytes written on disk", + EBPF_COMMON_DIMENSION_BYTES, NETDATA_VFS_CGROUP_GROUP, NETDATA_CGROUP_VFS_WRITE_BYTES_CONTEXT, + NETDATA_EBPF_CHART_TYPE_LINE, NETDATA_CHART_PRIO_CGROUPS_CONTAINERS + 5505, + ebpf_create_global_dimension, &vfs_publish_aggregated[NETDATA_KEY_PUBLISH_VFS_WRITE], + 1, em->update_every, NETDATA_EBPF_MODULE_NAME_SWAP); + + ebpf_create_chart(type, NETDATA_SYSCALL_APPS_VFS_READ_BYTES, "Bytes read from disk", + EBPF_COMMON_DIMENSION_BYTES, NETDATA_VFS_CGROUP_GROUP, NETDATA_CGROUP_VFS_READ_BYTES_CONTEXT, + NETDATA_EBPF_CHART_TYPE_LINE, NETDATA_CHART_PRIO_CGROUPS_CONTAINERS + 5506, + ebpf_create_global_dimension, &vfs_publish_aggregated[NETDATA_KEY_PUBLISH_VFS_READ], + 1, em->update_every, NETDATA_EBPF_MODULE_NAME_SWAP); + + ebpf_create_chart(type, NETDATA_SYSCALL_APPS_VFS_FSYNC, "Calls for <code>vfs_fsync</code>", + EBPF_COMMON_DIMENSION_CALL, NETDATA_VFS_CGROUP_GROUP, NETDATA_CGROUP_VFS_FSYNC_CONTEXT, + NETDATA_EBPF_CHART_TYPE_LINE, NETDATA_CHART_PRIO_CGROUPS_CONTAINERS + 5507, + ebpf_create_global_dimension, &vfs_publish_aggregated[NETDATA_KEY_PUBLISH_VFS_FSYNC], + 1, em->update_every, NETDATA_EBPF_MODULE_NAME_SWAP); + + if (em->mode < MODE_ENTRY) { + ebpf_create_chart(type, NETDATA_SYSCALL_APPS_VFS_FSYNC_CALLS_ERROR, "Sync error", + EBPF_COMMON_DIMENSION_CALL, NETDATA_VFS_CGROUP_GROUP, NETDATA_CGROUP_VFS_FSYNC_ERROR_CONTEXT, + NETDATA_EBPF_CHART_TYPE_LINE, NETDATA_CHART_PRIO_CGROUPS_CONTAINERS + 5508, + ebpf_create_global_dimension, &vfs_publish_aggregated[NETDATA_KEY_PUBLISH_VFS_FSYNC], + 1, em->update_every, NETDATA_EBPF_MODULE_NAME_SWAP); + } + + ebpf_create_chart(type, NETDATA_SYSCALL_APPS_VFS_OPEN, "Calls for <code>vfs_open</code>", + EBPF_COMMON_DIMENSION_CALL, NETDATA_VFS_CGROUP_GROUP, NETDATA_CGROUP_VFS_OPEN_CONTEXT, + NETDATA_EBPF_CHART_TYPE_LINE, NETDATA_CHART_PRIO_CGROUPS_CONTAINERS + 5509, + ebpf_create_global_dimension, &vfs_publish_aggregated[NETDATA_KEY_PUBLISH_VFS_OPEN], + 1, em->update_every, NETDATA_EBPF_MODULE_NAME_SWAP); + + if (em->mode < MODE_ENTRY) { + ebpf_create_chart(type, NETDATA_SYSCALL_APPS_VFS_OPEN_CALLS_ERROR, "Open error", + EBPF_COMMON_DIMENSION_CALL, NETDATA_VFS_CGROUP_GROUP, NETDATA_CGROUP_VFS_OPEN_ERROR_CONTEXT, + NETDATA_EBPF_CHART_TYPE_LINE, NETDATA_CHART_PRIO_CGROUPS_CONTAINERS + 5510, + ebpf_create_global_dimension, &vfs_publish_aggregated[NETDATA_KEY_PUBLISH_VFS_OPEN], + 1, em->update_every, NETDATA_EBPF_MODULE_NAME_SWAP); + } + + ebpf_create_chart(type, NETDATA_SYSCALL_APPS_VFS_CREATE, "Calls for <code>vfs_create</code>", + EBPF_COMMON_DIMENSION_CALL, NETDATA_VFS_CGROUP_GROUP, NETDATA_CGROUP_VFS_CREATE_CONTEXT, + NETDATA_EBPF_CHART_TYPE_LINE, NETDATA_CHART_PRIO_CGROUPS_CONTAINERS + 5511, + ebpf_create_global_dimension, &vfs_publish_aggregated[NETDATA_KEY_PUBLISH_VFS_CREATE], + 1, em->update_every, NETDATA_EBPF_MODULE_NAME_SWAP); + + if (em->mode < MODE_ENTRY) { + ebpf_create_chart(type, NETDATA_SYSCALL_APPS_VFS_CREATE_CALLS_ERROR, "Create error", + EBPF_COMMON_DIMENSION_CALL, NETDATA_VFS_CGROUP_GROUP, NETDATA_CGROUP_VFS_CREATE_ERROR_CONTEXT, + NETDATA_EBPF_CHART_TYPE_LINE, NETDATA_CHART_PRIO_CGROUPS_CONTAINERS + 5512, + ebpf_create_global_dimension, &vfs_publish_aggregated[NETDATA_KEY_PUBLISH_VFS_CREATE], + 1, em->update_every, NETDATA_EBPF_MODULE_NAME_SWAP); + } +} + +/** + * Obsolete specific VFS charts + * + * Obsolete charts for cgroup/application. + * + * @param type the chart type. + * @param em the main thread structure. + */ +static void ebpf_obsolete_specific_vfs_charts(char *type, ebpf_module_t *em) +{ + ebpf_write_chart_obsolete(type, NETDATA_SYSCALL_APPS_FILE_DELETED, "Files deleted", + EBPF_COMMON_DIMENSION_CALL, NETDATA_VFS_GROUP, + NETDATA_EBPF_CHART_TYPE_LINE, NETDATA_CGROUP_VFS_UNLINK_CONTEXT, + NETDATA_CHART_PRIO_CGROUPS_CONTAINERS + 5500, em->update_every); + + ebpf_write_chart_obsolete(type, NETDATA_SYSCALL_APPS_VFS_WRITE_CALLS, "Write to disk", + EBPF_COMMON_DIMENSION_CALL, NETDATA_VFS_GROUP, + NETDATA_EBPF_CHART_TYPE_LINE, NETDATA_CGROUP_VFS_WRITE_CONTEXT, + NETDATA_CHART_PRIO_CGROUPS_CONTAINERS + 5501, em->update_every); + + if (em->mode < MODE_ENTRY) { + ebpf_write_chart_obsolete(type, NETDATA_SYSCALL_APPS_VFS_WRITE_CALLS_ERROR, "Fails to write", + EBPF_COMMON_DIMENSION_CALL, NETDATA_VFS_GROUP, + NETDATA_EBPF_CHART_TYPE_LINE, NETDATA_CGROUP_VFS_WRITE_ERROR_CONTEXT, + NETDATA_CHART_PRIO_CGROUPS_CONTAINERS + 5502, em->update_every); + } + + ebpf_write_chart_obsolete(type, NETDATA_SYSCALL_APPS_VFS_READ_CALLS, "Read from disk", + EBPF_COMMON_DIMENSION_CALL, NETDATA_VFS_GROUP, + NETDATA_EBPF_CHART_TYPE_LINE, NETDATA_CGROUP_VFS_READ_CONTEXT, + NETDATA_CHART_PRIO_CGROUPS_CONTAINERS + 5503, em->update_every); + + if (em->mode < MODE_ENTRY) { + ebpf_write_chart_obsolete(type, NETDATA_SYSCALL_APPS_VFS_READ_CALLS_ERROR, "Fails to read", + EBPF_COMMON_DIMENSION_CALL, NETDATA_VFS_GROUP, + NETDATA_EBPF_CHART_TYPE_LINE, NETDATA_CGROUP_VFS_READ_ERROR_CONTEXT, + NETDATA_CHART_PRIO_CGROUPS_CONTAINERS + 5504, em->update_every); + } + + ebpf_write_chart_obsolete(type, NETDATA_SYSCALL_APPS_VFS_WRITE_BYTES, "Bytes written on disk", + EBPF_COMMON_DIMENSION_BYTES, NETDATA_VFS_GROUP, + NETDATA_EBPF_CHART_TYPE_LINE, NETDATA_CGROUP_VFS_WRITE_BYTES_CONTEXT, + NETDATA_CHART_PRIO_CGROUPS_CONTAINERS + 5505, em->update_every); + + ebpf_write_chart_obsolete(type, NETDATA_SYSCALL_APPS_VFS_READ_BYTES, "Bytes read from disk", + EBPF_COMMON_DIMENSION_BYTES, NETDATA_VFS_GROUP, + NETDATA_EBPF_CHART_TYPE_LINE, NETDATA_CGROUP_VFS_READ_BYTES_CONTEXT, + NETDATA_CHART_PRIO_CGROUPS_CONTAINERS + 5506, em->update_every); + + ebpf_write_chart_obsolete(type, NETDATA_SYSCALL_APPS_VFS_FSYNC, "Calls for <code>vfs_fsync</code>", + EBPF_COMMON_DIMENSION_CALL, NETDATA_VFS_GROUP, + NETDATA_EBPF_CHART_TYPE_LINE, NETDATA_CGROUP_VFS_FSYNC_CONTEXT, + NETDATA_CHART_PRIO_CGROUPS_CONTAINERS + 5507, em->update_every); + + if (em->mode < MODE_ENTRY) { + ebpf_write_chart_obsolete(type, NETDATA_SYSCALL_APPS_VFS_FSYNC_CALLS_ERROR, "Sync error", + EBPF_COMMON_DIMENSION_CALL, NETDATA_VFS_GROUP, + NETDATA_EBPF_CHART_TYPE_LINE, NETDATA_CGROUP_VFS_FSYNC_ERROR_CONTEXT, + NETDATA_CHART_PRIO_CGROUPS_CONTAINERS + 5508, em->update_every); + } + + ebpf_write_chart_obsolete(type, NETDATA_SYSCALL_APPS_VFS_OPEN, "Calls for <code>vfs_open</code>", + EBPF_COMMON_DIMENSION_CALL, NETDATA_VFS_GROUP, + NETDATA_EBPF_CHART_TYPE_LINE, NETDATA_CGROUP_VFS_OPEN_CONTEXT, + NETDATA_CHART_PRIO_CGROUPS_CONTAINERS + 5509, em->update_every); + + if (em->mode < MODE_ENTRY) { + ebpf_write_chart_obsolete(type, NETDATA_SYSCALL_APPS_VFS_OPEN_CALLS_ERROR, "Open error", + EBPF_COMMON_DIMENSION_CALL, NETDATA_VFS_GROUP, + NETDATA_EBPF_CHART_TYPE_LINE, NETDATA_CGROUP_VFS_OPEN_ERROR_CONTEXT, + NETDATA_CHART_PRIO_CGROUPS_CONTAINERS + 5510, em->update_every); + } + + ebpf_write_chart_obsolete(type, NETDATA_SYSCALL_APPS_VFS_CREATE, "Calls for <code>vfs_create</code>", + EBPF_COMMON_DIMENSION_CALL, NETDATA_VFS_GROUP, + NETDATA_EBPF_CHART_TYPE_LINE, NETDATA_CGROUP_VFS_CREATE_CONTEXT, + NETDATA_CHART_PRIO_CGROUPS_CONTAINERS + 5511, em->update_every); + + if (em->mode < MODE_ENTRY) { + ebpf_write_chart_obsolete(type, NETDATA_SYSCALL_APPS_VFS_CREATE_CALLS_ERROR, "Create error", + EBPF_COMMON_DIMENSION_CALL, NETDATA_VFS_GROUP, + NETDATA_EBPF_CHART_TYPE_LINE, NETDATA_CGROUP_VFS_CREATE_ERROR_CONTEXT, + NETDATA_CHART_PRIO_CGROUPS_CONTAINERS + 5512, em->update_every); + } +} + +/* + * Send specific VFS data + * + * Send data for specific cgroup/apps. + * + * @param type chart type + * @param values structure with values that will be sent to netdata + */ +static void ebpf_send_specific_vfs_data(char *type, netdata_publish_vfs_t *values, ebpf_module_t *em) +{ + write_begin_chart(type, NETDATA_SYSCALL_APPS_FILE_DELETED); + write_chart_dimension(vfs_publish_aggregated[NETDATA_KEY_PUBLISH_VFS_UNLINK].name, (long long)values->unlink_call); + write_end_chart(); + + write_begin_chart(type, NETDATA_SYSCALL_APPS_VFS_WRITE_CALLS); + write_chart_dimension(vfs_publish_aggregated[NETDATA_KEY_PUBLISH_VFS_WRITE].name, + (long long)values->write_call + (long long)values->writev_call); + write_end_chart(); + + if (em->mode < MODE_ENTRY) { + write_begin_chart(type, NETDATA_SYSCALL_APPS_VFS_WRITE_CALLS_ERROR); + write_chart_dimension(vfs_publish_aggregated[NETDATA_KEY_PUBLISH_VFS_WRITE].name, + (long long)values->write_err + (long long)values->writev_err); + write_end_chart(); + } + + write_begin_chart(type, NETDATA_SYSCALL_APPS_VFS_READ_CALLS); + write_chart_dimension(vfs_publish_aggregated[NETDATA_KEY_PUBLISH_VFS_READ].name, + (long long)values->read_call + (long long)values->readv_call); + write_end_chart(); + + if (em->mode < MODE_ENTRY) { + write_begin_chart(type, NETDATA_SYSCALL_APPS_VFS_READ_CALLS_ERROR); + write_chart_dimension(vfs_publish_aggregated[NETDATA_KEY_PUBLISH_VFS_READ].name, + (long long)values->read_err + (long long)values->readv_err); + write_end_chart(); + } + + write_begin_chart(type, NETDATA_SYSCALL_APPS_VFS_WRITE_BYTES); + write_chart_dimension(vfs_publish_aggregated[NETDATA_KEY_PUBLISH_VFS_WRITE].name, + (long long)values->write_bytes + (long long)values->writev_bytes); + write_end_chart(); + + write_begin_chart(type, NETDATA_SYSCALL_APPS_VFS_READ_BYTES); + write_chart_dimension(vfs_publish_aggregated[NETDATA_KEY_PUBLISH_VFS_READ].name, + (long long)values->read_bytes + (long long)values->readv_bytes); + write_end_chart(); + + write_begin_chart(type, NETDATA_SYSCALL_APPS_VFS_FSYNC); + write_chart_dimension(vfs_publish_aggregated[NETDATA_KEY_PUBLISH_VFS_FSYNC].name, + (long long)values->fsync_call); + write_end_chart(); + + if (em->mode < MODE_ENTRY) { + write_begin_chart(type, NETDATA_SYSCALL_APPS_VFS_FSYNC_CALLS_ERROR); + write_chart_dimension(vfs_publish_aggregated[NETDATA_KEY_PUBLISH_VFS_FSYNC].name, + (long long)values->fsync_err); + write_end_chart(); + } + + write_begin_chart(type, NETDATA_SYSCALL_APPS_VFS_OPEN); + write_chart_dimension(vfs_publish_aggregated[NETDATA_KEY_PUBLISH_VFS_OPEN].name, + (long long)values->open_call); + write_end_chart(); + + if (em->mode < MODE_ENTRY) { + write_begin_chart(type, NETDATA_SYSCALL_APPS_VFS_OPEN_CALLS_ERROR); + write_chart_dimension(vfs_publish_aggregated[NETDATA_KEY_PUBLISH_VFS_OPEN].name, + (long long)values->open_err); + write_end_chart(); + } + + write_begin_chart(type, NETDATA_SYSCALL_APPS_VFS_CREATE); + write_chart_dimension(vfs_publish_aggregated[NETDATA_KEY_PUBLISH_VFS_CREATE].name, + (long long)values->create_call); + write_end_chart(); + + if (em->mode < MODE_ENTRY) { + write_begin_chart(type, NETDATA_SYSCALL_APPS_VFS_CREATE_CALLS_ERROR); + write_chart_dimension(vfs_publish_aggregated[NETDATA_KEY_PUBLISH_VFS_CREATE].name, + (long long)values->create_err); + write_end_chart(); + } +} + +/** + * Create Systemd Socket Charts + * + * Create charts when systemd is enabled + * + * @param em the main collector structure + **/ +static void ebpf_create_systemd_vfs_charts(ebpf_module_t *em) +{ + ebpf_create_charts_on_systemd(NETDATA_SYSCALL_APPS_FILE_DELETED, "Files deleted", + EBPF_COMMON_DIMENSION_CALL, NETDATA_VFS_CGROUP_GROUP, + NETDATA_EBPF_CHART_TYPE_STACKED, 20065, + ebpf_algorithms[NETDATA_EBPF_INCREMENTAL_IDX], NETDATA_SYSTEMD_VFS_UNLINK_CONTEXT, + NETDATA_EBPF_MODULE_NAME_VFS, em->update_every); + + ebpf_create_charts_on_systemd(NETDATA_SYSCALL_APPS_VFS_WRITE_CALLS, "Write to disk", + EBPF_COMMON_DIMENSION_CALL, NETDATA_VFS_CGROUP_GROUP, + NETDATA_EBPF_CHART_TYPE_STACKED, 20066, + ebpf_algorithms[NETDATA_EBPF_INCREMENTAL_IDX], NETDATA_SYSTEMD_VFS_WRITE_CONTEXT, + NETDATA_EBPF_MODULE_NAME_VFS, em->update_every); + + if (em->mode < MODE_ENTRY) { + ebpf_create_charts_on_systemd(NETDATA_SYSCALL_APPS_VFS_WRITE_CALLS_ERROR, "Fails to write", + EBPF_COMMON_DIMENSION_CALL, NETDATA_VFS_CGROUP_GROUP, + NETDATA_EBPF_CHART_TYPE_STACKED, 20067, + ebpf_algorithms[NETDATA_EBPF_INCREMENTAL_IDX], + NETDATA_SYSTEMD_VFS_WRITE_ERROR_CONTEXT, + NETDATA_EBPF_MODULE_NAME_VFS, em->update_every); + } + + ebpf_create_charts_on_systemd(NETDATA_SYSCALL_APPS_VFS_READ_CALLS, "Read from disk", + EBPF_COMMON_DIMENSION_CALL, NETDATA_VFS_CGROUP_GROUP, + NETDATA_EBPF_CHART_TYPE_STACKED, 20068, + ebpf_algorithms[NETDATA_EBPF_INCREMENTAL_IDX], NETDATA_SYSTEMD_VFS_READ_CONTEXT, + NETDATA_EBPF_MODULE_NAME_VFS, em->update_every); + + if (em->mode < MODE_ENTRY) { + ebpf_create_charts_on_systemd(NETDATA_SYSCALL_APPS_VFS_READ_CALLS_ERROR, "Fails to read", + EBPF_COMMON_DIMENSION_CALL, NETDATA_VFS_CGROUP_GROUP, + NETDATA_EBPF_CHART_TYPE_STACKED, 20069, + ebpf_algorithms[NETDATA_EBPF_INCREMENTAL_IDX], + NETDATA_SYSTEMD_VFS_READ_ERROR_CONTEXT, + NETDATA_EBPF_MODULE_NAME_VFS, em->update_every); + } + + ebpf_create_charts_on_systemd(NETDATA_SYSCALL_APPS_VFS_WRITE_BYTES, "Bytes written on disk", + EBPF_COMMON_DIMENSION_BYTES, NETDATA_VFS_CGROUP_GROUP, + NETDATA_EBPF_CHART_TYPE_STACKED, 20070, + ebpf_algorithms[NETDATA_EBPF_INCREMENTAL_IDX], NETDATA_SYSTEMD_VFS_WRITE_BYTES_CONTEXT, + NETDATA_EBPF_MODULE_NAME_VFS, em->update_every); + + ebpf_create_charts_on_systemd(NETDATA_SYSCALL_APPS_VFS_READ_BYTES, "Bytes read from disk", + EBPF_COMMON_DIMENSION_BYTES, NETDATA_VFS_CGROUP_GROUP, + NETDATA_EBPF_CHART_TYPE_STACKED, 20071, + ebpf_algorithms[NETDATA_EBPF_INCREMENTAL_IDX], NETDATA_SYSTEMD_VFS_READ_BYTES_CONTEXT, + NETDATA_EBPF_MODULE_NAME_VFS, em->update_every); + + ebpf_create_charts_on_systemd(NETDATA_SYSCALL_APPS_VFS_FSYNC, "Calls to <code>vfs_fsync</code>", + EBPF_COMMON_DIMENSION_CALL, NETDATA_VFS_CGROUP_GROUP, + NETDATA_EBPF_CHART_TYPE_STACKED, 20072, + ebpf_algorithms[NETDATA_EBPF_INCREMENTAL_IDX], NETDATA_SYSTEMD_VFS_FSYNC_CONTEXT, + NETDATA_EBPF_MODULE_NAME_VFS, em->update_every); + + if (em->mode < MODE_ENTRY) { + ebpf_create_charts_on_systemd(NETDATA_SYSCALL_APPS_VFS_FSYNC_CALLS_ERROR, "Sync error", + EBPF_COMMON_DIMENSION_CALL, NETDATA_VFS_CGROUP_GROUP, + NETDATA_EBPF_CHART_TYPE_STACKED, 20073, + ebpf_algorithms[NETDATA_EBPF_INCREMENTAL_IDX], NETDATA_SYSTEMD_VFS_FSYNC_ERROR_CONTEXT, + NETDATA_EBPF_MODULE_NAME_VFS, em->update_every); + } + ebpf_create_charts_on_systemd(NETDATA_SYSCALL_APPS_VFS_OPEN, "Calls to <code>vfs_open</code>", + EBPF_COMMON_DIMENSION_CALL, NETDATA_VFS_CGROUP_GROUP, + NETDATA_EBPF_CHART_TYPE_STACKED, 20074, + ebpf_algorithms[NETDATA_EBPF_INCREMENTAL_IDX], NETDATA_SYSTEMD_VFS_OPEN_CONTEXT, + NETDATA_EBPF_MODULE_NAME_VFS, em->update_every); + + if (em->mode < MODE_ENTRY) { + ebpf_create_charts_on_systemd(NETDATA_SYSCALL_APPS_VFS_OPEN_CALLS_ERROR, "Open error", + EBPF_COMMON_DIMENSION_CALL, NETDATA_VFS_CGROUP_GROUP, + NETDATA_EBPF_CHART_TYPE_STACKED, 20075, + ebpf_algorithms[NETDATA_EBPF_INCREMENTAL_IDX], NETDATA_SYSTEMD_VFS_OPEN_ERROR_CONTEXT, + NETDATA_EBPF_MODULE_NAME_VFS, em->update_every); + } + + ebpf_create_charts_on_systemd(NETDATA_SYSCALL_APPS_VFS_CREATE, "Calls to <code>vfs_create</code>", + EBPF_COMMON_DIMENSION_CALL, NETDATA_VFS_CGROUP_GROUP, + NETDATA_EBPF_CHART_TYPE_STACKED, 20076, + ebpf_algorithms[NETDATA_EBPF_INCREMENTAL_IDX], NETDATA_SYSTEMD_VFS_CREATE_CONTEXT, + NETDATA_EBPF_MODULE_NAME_VFS, em->update_every); + + if (em->mode < MODE_ENTRY) { + ebpf_create_charts_on_systemd(NETDATA_SYSCALL_APPS_VFS_CREATE_CALLS_ERROR, "Create error", + EBPF_COMMON_DIMENSION_CALL, NETDATA_VFS_CGROUP_GROUP, + NETDATA_EBPF_CHART_TYPE_STACKED, 20077, + ebpf_algorithms[NETDATA_EBPF_INCREMENTAL_IDX], NETDATA_SYSTEMD_VFS_CREATE_ERROR_CONTEXT, + NETDATA_EBPF_MODULE_NAME_VFS, em->update_every); + } +} + +/** + * Send Systemd charts + * + * Send collected data to Netdata. + * + * @param em the main collector structure + */ +static void ebpf_send_systemd_vfs_charts(ebpf_module_t *em) +{ + ebpf_cgroup_target_t *ect; + write_begin_chart(NETDATA_SERVICE_FAMILY, NETDATA_SYSCALL_APPS_FILE_DELETED); + for (ect = ebpf_cgroup_pids; ect ; ect = ect->next) { + if (unlikely(ect->systemd) && unlikely(ect->updated)) { + write_chart_dimension(ect->name, ect->publish_systemd_vfs.unlink_call); + } + } + write_end_chart(); + + write_begin_chart(NETDATA_SERVICE_FAMILY, NETDATA_SYSCALL_APPS_VFS_WRITE_CALLS); + for (ect = ebpf_cgroup_pids; ect ; ect = ect->next) { + if (unlikely(ect->systemd) && unlikely(ect->updated)) { + write_chart_dimension(ect->name, ect->publish_systemd_vfs.write_call + + ect->publish_systemd_vfs.writev_call); + } + } + write_end_chart(); + + if (em->mode < MODE_ENTRY) { + write_begin_chart(NETDATA_SERVICE_FAMILY, NETDATA_SYSCALL_APPS_VFS_WRITE_CALLS_ERROR); + for (ect = ebpf_cgroup_pids; ect ; ect = ect->next) { + if (unlikely(ect->systemd) && unlikely(ect->updated)) { + write_chart_dimension(ect->name, ect->publish_systemd_vfs.write_err + + ect->publish_systemd_vfs.writev_err); + } + } + write_end_chart(); + } + + write_begin_chart(NETDATA_SERVICE_FAMILY, NETDATA_SYSCALL_APPS_VFS_READ_CALLS); + for (ect = ebpf_cgroup_pids; ect ; ect = ect->next) { + if (unlikely(ect->systemd) && unlikely(ect->updated)) { + write_chart_dimension(ect->name, ect->publish_systemd_vfs.read_call + + ect->publish_systemd_vfs.readv_call); + } + } + write_end_chart(); + + if (em->mode < MODE_ENTRY) { + write_begin_chart(NETDATA_SERVICE_FAMILY, NETDATA_SYSCALL_APPS_VFS_READ_CALLS_ERROR); + for (ect = ebpf_cgroup_pids; ect ; ect = ect->next) { + if (unlikely(ect->systemd) && unlikely(ect->updated)) { + write_chart_dimension(ect->name, ect->publish_systemd_vfs.read_err + + ect->publish_systemd_vfs.readv_err); + } + } + write_end_chart(); + } + + write_begin_chart(NETDATA_SERVICE_FAMILY, NETDATA_SYSCALL_APPS_VFS_WRITE_BYTES); + for (ect = ebpf_cgroup_pids; ect ; ect = ect->next) { + if (unlikely(ect->systemd) && unlikely(ect->updated)) { + write_chart_dimension(ect->name, ect->publish_systemd_vfs.write_bytes + + ect->publish_systemd_vfs.writev_bytes); + } + } + write_end_chart(); + + + write_begin_chart(NETDATA_SERVICE_FAMILY, NETDATA_SYSCALL_APPS_VFS_READ_BYTES); + for (ect = ebpf_cgroup_pids; ect ; ect = ect->next) { + if (unlikely(ect->systemd) && unlikely(ect->updated)) { + write_chart_dimension(ect->name, ect->publish_systemd_vfs.read_bytes + + ect->publish_systemd_vfs.readv_bytes); + } + } + write_end_chart(); + + write_begin_chart(NETDATA_SERVICE_FAMILY, NETDATA_SYSCALL_APPS_VFS_FSYNC); + for (ect = ebpf_cgroup_pids; ect ; ect = ect->next) { + if (unlikely(ect->systemd) && unlikely(ect->updated)) { + write_chart_dimension(ect->name, ect->publish_systemd_vfs.fsync_call); + } + } + write_end_chart(); + + if (em->mode < MODE_ENTRY) { + write_begin_chart(NETDATA_SERVICE_FAMILY, NETDATA_SYSCALL_APPS_VFS_FSYNC_CALLS_ERROR); + for (ect = ebpf_cgroup_pids; ect ; ect = ect->next) { + if (unlikely(ect->systemd) && unlikely(ect->updated)) { + write_chart_dimension(ect->name, ect->publish_systemd_vfs.fsync_err); + } + } + write_end_chart(); + } + + write_begin_chart(NETDATA_SERVICE_FAMILY, NETDATA_SYSCALL_APPS_VFS_OPEN); + for (ect = ebpf_cgroup_pids; ect ; ect = ect->next) { + if (unlikely(ect->systemd) && unlikely(ect->updated)) { + write_chart_dimension(ect->name, ect->publish_systemd_vfs.open_call); + } + } + write_end_chart(); + + if (em->mode < MODE_ENTRY) { + write_begin_chart(NETDATA_SERVICE_FAMILY, NETDATA_SYSCALL_APPS_VFS_OPEN_CALLS_ERROR); + for (ect = ebpf_cgroup_pids; ect ; ect = ect->next) { + if (unlikely(ect->systemd) && unlikely(ect->updated)) { + write_chart_dimension(ect->name, ect->publish_systemd_vfs.open_err); + } + } + write_end_chart(); + } + + write_begin_chart(NETDATA_SERVICE_FAMILY, NETDATA_SYSCALL_APPS_VFS_CREATE); + for (ect = ebpf_cgroup_pids; ect ; ect = ect->next) { + if (unlikely(ect->systemd) && unlikely(ect->updated)) { + write_chart_dimension(ect->name, ect->publish_systemd_vfs.create_call); + } + } + write_end_chart(); + + if (em->mode < MODE_ENTRY) { + write_begin_chart(NETDATA_SERVICE_FAMILY, NETDATA_SYSCALL_APPS_VFS_CREATE_CALLS_ERROR); + for (ect = ebpf_cgroup_pids; ect ; ect = ect->next) { + if (unlikely(ect->systemd) && unlikely(ect->updated)) { + write_chart_dimension(ect->name, ect->publish_systemd_vfs.create_err); + } + } + write_end_chart(); + } +} + +/** + * Send data to Netdata calling auxiliary functions. + * + * @param em the main collector structure +*/ +static void ebpf_vfs_send_cgroup_data(ebpf_module_t *em) +{ + if (!ebpf_cgroup_pids) + return; + + pthread_mutex_lock(&mutex_cgroup_shm); + ebpf_cgroup_target_t *ect; + for (ect = ebpf_cgroup_pids; ect ; ect = ect->next) { + ebpf_vfs_sum_cgroup_pids(&ect->publish_systemd_vfs, ect->pids); + } + + int has_systemd = shm_ebpf_cgroup.header->systemd_enabled; + if (has_systemd) { + if (send_cgroup_chart) { + ebpf_create_systemd_vfs_charts(em); + } + ebpf_send_systemd_vfs_charts(em); + } + + for (ect = ebpf_cgroup_pids; ect ; ect = ect->next) { + if (ect->systemd) + continue; + + if (!(ect->flags & NETDATA_EBPF_CGROUP_HAS_VFS_CHART) && ect->updated) { + ebpf_create_specific_vfs_charts(ect->name, em); + ect->flags |= NETDATA_EBPF_CGROUP_HAS_VFS_CHART; + } + + if (ect->flags & NETDATA_EBPF_CGROUP_HAS_VFS_CHART) { + if (ect->updated) { + ebpf_send_specific_vfs_data(ect->name, &ect->publish_systemd_vfs, em); + } else { + ebpf_obsolete_specific_vfs_charts(ect->name, em); + ect->flags &= ~NETDATA_EBPF_CGROUP_HAS_VFS_CHART; + } + } + } + + pthread_mutex_unlock(&mutex_cgroup_shm); +} + +/** + * Main loop for this collector. + * + * @param step the number of microseconds used with heart beat + * @param em the structure with thread information + */ +static void vfs_collector(ebpf_module_t *em) +{ + vfs_threads.thread = mallocz(sizeof(netdata_thread_t)); + vfs_threads.start_routine = ebpf_vfs_read_hash; + + netdata_thread_create(vfs_threads.thread, vfs_threads.name, NETDATA_THREAD_OPTION_DEFAULT, + ebpf_vfs_read_hash, em); + + int cgroups = em->cgroup_charts; + heartbeat_t hb; + heartbeat_init(&hb); + usec_t step = em->update_every * USEC_PER_SEC; + while (!ebpf_exit_plugin) { + (void)heartbeat_next(&hb, step); + if (ebpf_exit_plugin) + break; + + netdata_apps_integration_flags_t apps = em->apps_charts; + pthread_mutex_lock(&collect_data_mutex); + if (apps) + ebpf_vfs_read_apps(); + + if (cgroups) + read_update_vfs_cgroup(); + + pthread_mutex_lock(&lock); + + ebpf_vfs_send_data(em); + fflush(stdout); + + if (apps & NETDATA_EBPF_APPS_FLAG_CHART_CREATED) + ebpf_vfs_send_apps_data(em, apps_groups_root_target); + + if (cgroups) + ebpf_vfs_send_cgroup_data(em); + + pthread_mutex_unlock(&lock); + pthread_mutex_unlock(&collect_data_mutex); + } +} + +/***************************************************************** + * + * FUNCTIONS TO CREATE CHARTS + * + *****************************************************************/ + +/** + * Create IO chart + * + * @param family the chart family + * @param name the chart name + * @param axis the axis label + * @param web the group name used to attach the chart on dashboard + * @param order the order number of the specified chart + * @param algorithm the algorithm used to make the charts. + * @param update_every value to overwrite the update frequency set by the server. + */ +static void ebpf_create_io_chart(char *family, char *name, char *axis, char *web, + int order, int algorithm, int update_every) +{ + printf("CHART %s.%s '' 'Bytes written and read' '%s' '%s' '' line %d %d '' 'ebpf.plugin' 'filesystem'\n", + family, + name, + axis, + web, + order, + update_every); + + printf("DIMENSION %s %s %s 1 1\n", + vfs_id_names[NETDATA_KEY_PUBLISH_VFS_READ], + vfs_dimension_names[NETDATA_KEY_PUBLISH_VFS_READ], + ebpf_algorithms[algorithm]); + printf("DIMENSION %s %s %s -1 1\n", + vfs_id_names[NETDATA_KEY_PUBLISH_VFS_WRITE], + vfs_dimension_names[NETDATA_KEY_PUBLISH_VFS_WRITE], + ebpf_algorithms[algorithm]); +} + +/** + * Create global charts + * + * Call ebpf_create_chart to create the charts for the collector. + * + * @param em a pointer to the structure with the default values. + */ +static void ebpf_create_global_charts(ebpf_module_t *em) +{ + ebpf_create_chart(NETDATA_FILESYSTEM_FAMILY, + NETDATA_VFS_FILE_CLEAN_COUNT, + "Remove files", + EBPF_COMMON_DIMENSION_CALL, + NETDATA_VFS_GROUP, + NULL, + NETDATA_EBPF_CHART_TYPE_LINE, + NETDATA_CHART_PRIO_FILESYSTEM_VFS_CLEAN, + ebpf_create_global_dimension, + &vfs_publish_aggregated[NETDATA_KEY_PUBLISH_VFS_UNLINK], + 1, em->update_every, NETDATA_EBPF_MODULE_NAME_VFS); + + ebpf_create_chart(NETDATA_FILESYSTEM_FAMILY, + NETDATA_VFS_FILE_IO_COUNT, + "Calls to IO", + EBPF_COMMON_DIMENSION_CALL, + NETDATA_VFS_GROUP, + NULL, + NETDATA_EBPF_CHART_TYPE_LINE, + NETDATA_CHART_PRIO_FILESYSTEM_VFS_IO_COUNT, + ebpf_create_global_dimension, + &vfs_publish_aggregated[NETDATA_KEY_PUBLISH_VFS_READ], + 2, em->update_every, NETDATA_EBPF_MODULE_NAME_VFS); + + ebpf_create_io_chart(NETDATA_FILESYSTEM_FAMILY, + NETDATA_VFS_IO_FILE_BYTES, EBPF_COMMON_DIMENSION_BYTES, + NETDATA_VFS_GROUP, + NETDATA_CHART_PRIO_FILESYSTEM_VFS_IO_BYTES, + NETDATA_EBPF_INCREMENTAL_IDX, em->update_every); + + if (em->mode < MODE_ENTRY) { + ebpf_create_chart(NETDATA_FILESYSTEM_FAMILY, + NETDATA_VFS_FILE_ERR_COUNT, + "Fails to write or read", + EBPF_COMMON_DIMENSION_CALL, + NETDATA_VFS_GROUP, + NULL, + NETDATA_EBPF_CHART_TYPE_LINE, + NETDATA_CHART_PRIO_FILESYSTEM_VFS_IO_EBYTES, + ebpf_create_global_dimension, + &vfs_publish_aggregated[NETDATA_KEY_PUBLISH_VFS_READ], + 2, em->update_every, NETDATA_EBPF_MODULE_NAME_VFS); + } + + ebpf_create_chart(NETDATA_FILESYSTEM_FAMILY, + NETDATA_VFS_FSYNC, + "Calls for <code>vfs_fsync</code>", + EBPF_COMMON_DIMENSION_CALL, + NETDATA_VFS_GROUP, + NULL, + NETDATA_EBPF_CHART_TYPE_LINE, + NETDATA_CHART_PRIO_FILESYSTEM_VFS_IO_FSYNC, + ebpf_create_global_dimension, + &vfs_publish_aggregated[NETDATA_KEY_PUBLISH_VFS_FSYNC], + 1, em->update_every, NETDATA_EBPF_MODULE_NAME_VFS); + + if (em->mode < MODE_ENTRY) { + ebpf_create_chart(NETDATA_FILESYSTEM_FAMILY, + NETDATA_VFS_FSYNC_ERR, + "Fails to synchronize", + EBPF_COMMON_DIMENSION_CALL, + NETDATA_VFS_GROUP, + NULL, + NETDATA_EBPF_CHART_TYPE_LINE, + NETDATA_CHART_PRIO_FILESYSTEM_VFS_IO_EFSYNC, + ebpf_create_global_dimension, + &vfs_publish_aggregated[NETDATA_KEY_PUBLISH_VFS_FSYNC], + 1, em->update_every, NETDATA_EBPF_MODULE_NAME_VFS); + } + + ebpf_create_chart(NETDATA_FILESYSTEM_FAMILY, + NETDATA_VFS_OPEN, + "Calls for <code>vfs_open</code>", + EBPF_COMMON_DIMENSION_CALL, + NETDATA_VFS_GROUP, + NULL, + NETDATA_EBPF_CHART_TYPE_LINE, + NETDATA_CHART_PRIO_FILESYSTEM_VFS_IO_OPEN, + ebpf_create_global_dimension, + &vfs_publish_aggregated[NETDATA_KEY_PUBLISH_VFS_OPEN], + 1, em->update_every, NETDATA_EBPF_MODULE_NAME_VFS); + + if (em->mode < MODE_ENTRY) { + ebpf_create_chart(NETDATA_FILESYSTEM_FAMILY, + NETDATA_VFS_OPEN_ERR, + "Fails to open a file", + EBPF_COMMON_DIMENSION_CALL, + NETDATA_VFS_GROUP, + NULL, + NETDATA_EBPF_CHART_TYPE_LINE, + NETDATA_CHART_PRIO_FILESYSTEM_VFS_IO_EOPEN, + ebpf_create_global_dimension, + &vfs_publish_aggregated[NETDATA_KEY_PUBLISH_VFS_OPEN], + 1, em->update_every, NETDATA_EBPF_MODULE_NAME_VFS); + } + + ebpf_create_chart(NETDATA_FILESYSTEM_FAMILY, + NETDATA_VFS_CREATE, + "Calls for <code>vfs_create</code>", + EBPF_COMMON_DIMENSION_CALL, + NETDATA_VFS_GROUP, + NULL, + NETDATA_EBPF_CHART_TYPE_LINE, + NETDATA_CHART_PRIO_FILESYSTEM_VFS_IO_CREATE, + ebpf_create_global_dimension, + &vfs_publish_aggregated[NETDATA_KEY_PUBLISH_VFS_CREATE], + 1, em->update_every, NETDATA_EBPF_MODULE_NAME_VFS); + + if (em->mode < MODE_ENTRY) { + ebpf_create_chart(NETDATA_FILESYSTEM_FAMILY, + NETDATA_VFS_CREATE_ERR, + "Fails to create a file.", + EBPF_COMMON_DIMENSION_CALL, + NETDATA_VFS_GROUP, + NULL, + NETDATA_EBPF_CHART_TYPE_LINE, + NETDATA_CHART_PRIO_FILESYSTEM_VFS_IO_ECREATE, + ebpf_create_global_dimension, + &vfs_publish_aggregated[NETDATA_KEY_PUBLISH_VFS_CREATE], + 1, em->update_every, NETDATA_EBPF_MODULE_NAME_VFS); + } +} + +/** + * Create process apps charts + * + * Call ebpf_create_chart to create the charts on apps submenu. + * + * @param em a pointer to the structure with the default values. + * @param ptr a pointer for the targets. + **/ +void ebpf_vfs_create_apps_charts(struct ebpf_module *em, void *ptr) +{ + struct target *root = ptr; + + ebpf_create_charts_on_apps(NETDATA_SYSCALL_APPS_FILE_DELETED, + "Files deleted", + EBPF_COMMON_DIMENSION_CALL, + NETDATA_VFS_GROUP, + NETDATA_EBPF_CHART_TYPE_STACKED, + 20065, + ebpf_algorithms[NETDATA_EBPF_INCREMENTAL_IDX], + root, em->update_every, NETDATA_EBPF_MODULE_NAME_VFS); + + ebpf_create_charts_on_apps(NETDATA_SYSCALL_APPS_VFS_WRITE_CALLS, + "Write to disk", + EBPF_COMMON_DIMENSION_CALL, + NETDATA_VFS_GROUP, + NETDATA_EBPF_CHART_TYPE_STACKED, + 20066, + ebpf_algorithms[NETDATA_EBPF_INCREMENTAL_IDX], + root, em->update_every, NETDATA_EBPF_MODULE_NAME_VFS); + + if (em->mode < MODE_ENTRY) { + ebpf_create_charts_on_apps(NETDATA_SYSCALL_APPS_VFS_WRITE_CALLS_ERROR, + "Fails to write", + EBPF_COMMON_DIMENSION_CALL, + NETDATA_VFS_GROUP, + NETDATA_EBPF_CHART_TYPE_STACKED, + 20067, + ebpf_algorithms[NETDATA_EBPF_INCREMENTAL_IDX], + root, em->update_every, NETDATA_EBPF_MODULE_NAME_VFS); + } + + ebpf_create_charts_on_apps(NETDATA_SYSCALL_APPS_VFS_READ_CALLS, + "Read from disk", + EBPF_COMMON_DIMENSION_CALL, + NETDATA_VFS_GROUP, + NETDATA_EBPF_CHART_TYPE_STACKED, + 20068, + ebpf_algorithms[NETDATA_EBPF_INCREMENTAL_IDX], + root, em->update_every, NETDATA_EBPF_MODULE_NAME_VFS); + + if (em->mode < MODE_ENTRY) { + ebpf_create_charts_on_apps(NETDATA_SYSCALL_APPS_VFS_READ_CALLS_ERROR, + "Fails to read", + EBPF_COMMON_DIMENSION_CALL, + NETDATA_VFS_GROUP, + NETDATA_EBPF_CHART_TYPE_STACKED, + 20069, + ebpf_algorithms[NETDATA_EBPF_INCREMENTAL_IDX], + root, em->update_every, NETDATA_EBPF_MODULE_NAME_VFS); + } + + ebpf_create_charts_on_apps(NETDATA_SYSCALL_APPS_VFS_WRITE_BYTES, + "Bytes written on disk", EBPF_COMMON_DIMENSION_BYTES, + NETDATA_VFS_GROUP, + NETDATA_EBPF_CHART_TYPE_STACKED, + 20070, + ebpf_algorithms[NETDATA_EBPF_INCREMENTAL_IDX], + root, em->update_every, NETDATA_EBPF_MODULE_NAME_VFS); + + ebpf_create_charts_on_apps(NETDATA_SYSCALL_APPS_VFS_READ_BYTES, + "Bytes read from disk", EBPF_COMMON_DIMENSION_BYTES, + NETDATA_VFS_GROUP, + NETDATA_EBPF_CHART_TYPE_STACKED, + 20071, + ebpf_algorithms[NETDATA_EBPF_INCREMENTAL_IDX], + root, em->update_every, NETDATA_EBPF_MODULE_NAME_VFS); + + ebpf_create_charts_on_apps(NETDATA_SYSCALL_APPS_VFS_FSYNC, + "Calls for <code>vfs_fsync</code>", EBPF_COMMON_DIMENSION_CALL, + NETDATA_VFS_GROUP, + NETDATA_EBPF_CHART_TYPE_STACKED, + 20072, + ebpf_algorithms[NETDATA_EBPF_INCREMENTAL_IDX], + root, em->update_every, NETDATA_EBPF_MODULE_NAME_VFS); + + if (em->mode < MODE_ENTRY) { + ebpf_create_charts_on_apps(NETDATA_SYSCALL_APPS_VFS_FSYNC_CALLS_ERROR, + "Sync error", + EBPF_COMMON_DIMENSION_CALL, + NETDATA_VFS_GROUP, + NETDATA_EBPF_CHART_TYPE_STACKED, + 20073, + ebpf_algorithms[NETDATA_EBPF_INCREMENTAL_IDX], + root, em->update_every, NETDATA_EBPF_MODULE_NAME_VFS); + } + + ebpf_create_charts_on_apps(NETDATA_SYSCALL_APPS_VFS_OPEN, + "Calls for <code>vfs_open</code>", EBPF_COMMON_DIMENSION_CALL, + NETDATA_VFS_GROUP, + NETDATA_EBPF_CHART_TYPE_STACKED, + 20074, + ebpf_algorithms[NETDATA_EBPF_INCREMENTAL_IDX], + root, em->update_every, NETDATA_EBPF_MODULE_NAME_VFS); + + if (em->mode < MODE_ENTRY) { + ebpf_create_charts_on_apps(NETDATA_SYSCALL_APPS_VFS_OPEN_CALLS_ERROR, + "Open error", + EBPF_COMMON_DIMENSION_CALL, + NETDATA_VFS_GROUP, + NETDATA_EBPF_CHART_TYPE_STACKED, + 20075, + ebpf_algorithms[NETDATA_EBPF_INCREMENTAL_IDX], + root, em->update_every, NETDATA_EBPF_MODULE_NAME_VFS); + } + + ebpf_create_charts_on_apps(NETDATA_SYSCALL_APPS_VFS_CREATE, + "Calls for <code>vfs_create</code>", EBPF_COMMON_DIMENSION_CALL, + NETDATA_VFS_GROUP, + NETDATA_EBPF_CHART_TYPE_STACKED, + 20076, + ebpf_algorithms[NETDATA_EBPF_INCREMENTAL_IDX], + root, em->update_every, NETDATA_EBPF_MODULE_NAME_VFS); + + if (em->mode < MODE_ENTRY) { + ebpf_create_charts_on_apps(NETDATA_SYSCALL_APPS_VFS_CREATE_CALLS_ERROR, + "Create error", + EBPF_COMMON_DIMENSION_CALL, + NETDATA_VFS_GROUP, + NETDATA_EBPF_CHART_TYPE_STACKED, + 20077, + ebpf_algorithms[NETDATA_EBPF_INCREMENTAL_IDX], + root, em->update_every, NETDATA_EBPF_MODULE_NAME_VFS); + } + + em->apps_charts |= NETDATA_EBPF_APPS_FLAG_CHART_CREATED; +} + +/***************************************************************** + * + * FUNCTIONS TO START THREAD + * + *****************************************************************/ + +/** + * Allocate vectors used with this thread. + * We are not testing the return, because callocz does this and shutdown the software + * case it was not possible to allocate. + * + * @param apps is apps enabled? + */ +static void ebpf_vfs_allocate_global_vectors(int apps) +{ + memset(vfs_aggregated_data, 0, sizeof(vfs_aggregated_data)); + memset(vfs_publish_aggregated, 0, sizeof(vfs_publish_aggregated)); + + vfs_hash_values = callocz(ebpf_nprocs, sizeof(netdata_idx_t)); + vfs_vector = callocz(ebpf_nprocs, sizeof(netdata_publish_vfs_t)); + + if (apps) + vfs_pid = callocz((size_t)pid_max, sizeof(netdata_publish_vfs_t *)); +} + +/***************************************************************** + * + * EBPF VFS THREAD + * + *****************************************************************/ + +/* + * Load BPF + * + * Load BPF files. + * + * @param em the structure with configuration + */ +static int ebpf_vfs_load_bpf(ebpf_module_t *em) +{ + int ret = 0; + ebpf_adjust_apps_cgroup(em, em->targets[NETDATA_EBPF_VFS_WRITE].mode); + if (em->load & EBPF_LOAD_LEGACY) { + em->probe_links = ebpf_load_program(ebpf_plugin_dir, em, running_on_kernel, isrh, &em->objects); + if (!em->probe_links) { + ret = -1; + } + } +#ifdef LIBBPF_MAJOR_VERSION + else { + bpf_obj = vfs_bpf__open(); + if (!bpf_obj) + ret = -1; + else + ret = ebpf_vfs_load_and_attach(bpf_obj, em); + } +#endif + + return ret; +} + +/** + * Process thread + * + * Thread used to generate process charts. + * + * @param ptr a pointer to `struct ebpf_module` + * + * @return It always return NULL + */ +void *ebpf_vfs_thread(void *ptr) +{ + netdata_thread_cleanup_push(ebpf_vfs_exit, ptr); + + ebpf_module_t *em = (ebpf_module_t *)ptr; + em->maps = vfs_maps; + + ebpf_update_pid_table(&vfs_maps[NETDATA_VFS_PID], em); + + ebpf_vfs_allocate_global_vectors(em->apps_charts); + +#ifdef LIBBPF_MAJOR_VERSION + ebpf_adjust_thread_load(em, default_btf); +#endif + if (ebpf_vfs_load_bpf(em)) { + em->thread->enabled = NETDATA_THREAD_EBPF_STOPPED; + goto endvfs; + } + + int algorithms[NETDATA_KEY_PUBLISH_VFS_END] = { + NETDATA_EBPF_INCREMENTAL_IDX, NETDATA_EBPF_INCREMENTAL_IDX,NETDATA_EBPF_INCREMENTAL_IDX, + NETDATA_EBPF_INCREMENTAL_IDX, NETDATA_EBPF_INCREMENTAL_IDX,NETDATA_EBPF_INCREMENTAL_IDX + }; + + ebpf_global_labels(vfs_aggregated_data, vfs_publish_aggregated, vfs_dimension_names, + vfs_id_names, algorithms, NETDATA_KEY_PUBLISH_VFS_END); + + pthread_mutex_lock(&lock); + ebpf_create_global_charts(em); + ebpf_update_stats(&plugin_statistics, em); + pthread_mutex_unlock(&lock); + + vfs_collector(em); + +endvfs: + ebpf_update_disabled_plugin_stats(em); + + netdata_thread_cleanup_pop(1); + return NULL; +} diff --git a/collectors/ebpf.plugin/ebpf_vfs.h b/collectors/ebpf.plugin/ebpf_vfs.h new file mode 100644 index 0000000..2e3c7cc --- /dev/null +++ b/collectors/ebpf.plugin/ebpf_vfs.h @@ -0,0 +1,177 @@ +// SPDX-License-Identifier: GPL-3.0-or-later + +#ifndef NETDATA_EBPF_VFS_H +#define NETDATA_EBPF_VFS_H 1 + +// Module name +#define NETDATA_EBPF_MODULE_NAME_VFS "vfs" + +#define NETDATA_DIRECTORY_VFS_CONFIG_FILE "vfs.conf" + +#define NETDATA_LATENCY_VFS_SLEEP_MS 750000ULL + +// Global chart name +#define NETDATA_VFS_FILE_CLEAN_COUNT "vfs_deleted_objects" +#define NETDATA_VFS_FILE_IO_COUNT "vfs_io" +#define NETDATA_VFS_FILE_ERR_COUNT "vfs_io_error" +#define NETDATA_VFS_IO_FILE_BYTES "vfs_io_bytes" +#define NETDATA_VFS_FSYNC "vfs_fsync" +#define NETDATA_VFS_FSYNC_ERR "vfs_fsync_error" +#define NETDATA_VFS_OPEN "vfs_open" +#define NETDATA_VFS_OPEN_ERR "vfs_open_error" +#define NETDATA_VFS_CREATE "vfs_create" +#define NETDATA_VFS_CREATE_ERR "vfs_create_error" + +// Charts created on Apps submenu +#define NETDATA_SYSCALL_APPS_FILE_DELETED "file_deleted" +#define NETDATA_SYSCALL_APPS_VFS_WRITE_CALLS "vfs_write_call" +#define NETDATA_SYSCALL_APPS_VFS_READ_CALLS "vfs_read_call" +#define NETDATA_SYSCALL_APPS_VFS_WRITE_BYTES "vfs_write_bytes" +#define NETDATA_SYSCALL_APPS_VFS_READ_BYTES "vfs_read_bytes" +#define NETDATA_SYSCALL_APPS_VFS_FSYNC "vfs_fsync" +#define NETDATA_SYSCALL_APPS_VFS_OPEN "vfs_open" +#define NETDATA_SYSCALL_APPS_VFS_CREATE "vfs_create" + +#define NETDATA_SYSCALL_APPS_VFS_WRITE_CALLS_ERROR "vfs_write_error" +#define NETDATA_SYSCALL_APPS_VFS_READ_CALLS_ERROR "vfs_read_error" +#define NETDATA_SYSCALL_APPS_VFS_FSYNC_CALLS_ERROR "vfs_fsync_error" +#define NETDATA_SYSCALL_APPS_VFS_OPEN_CALLS_ERROR "vfs_open_error" +#define NETDATA_SYSCALL_APPS_VFS_CREATE_CALLS_ERROR "vfs_create_error" + +// Group used on Dashboard +#define NETDATA_VFS_GROUP "vfs" +#define NETDATA_VFS_CGROUP_GROUP "vfs (eBPF)" + +// Contexts +#define NETDATA_CGROUP_VFS_UNLINK_CONTEXT "cgroup.vfs_unlink" +#define NETDATA_CGROUP_VFS_WRITE_CONTEXT "cgroup.vfs_write" +#define NETDATA_CGROUP_VFS_WRITE_ERROR_CONTEXT "cgroup.vfs_write_error" +#define NETDATA_CGROUP_VFS_READ_CONTEXT "cgroup.vfs_read" +#define NETDATA_CGROUP_VFS_READ_ERROR_CONTEXT "cgroup.vfs_read_error" +#define NETDATA_CGROUP_VFS_WRITE_BYTES_CONTEXT "cgroup.vfs_write_bytes" +#define NETDATA_CGROUP_VFS_READ_BYTES_CONTEXT "cgroup.vfs_read_bytes" +#define NETDATA_CGROUP_VFS_CREATE_CONTEXT "cgroup.vfs_create" +#define NETDATA_CGROUP_VFS_CREATE_ERROR_CONTEXT "cgroup.vfs_create_error" +#define NETDATA_CGROUP_VFS_OPEN_CONTEXT "cgroup.vfs_open" +#define NETDATA_CGROUP_VFS_OPEN_ERROR_CONTEXT "cgroup.vfs_open_error" +#define NETDATA_CGROUP_VFS_FSYNC_CONTEXT "cgroup.vfs_fsync" +#define NETDATA_CGROUP_VFS_FSYNC_ERROR_CONTEXT "cgroup.vfs_fsync_error" + +#define NETDATA_SYSTEMD_VFS_UNLINK_CONTEXT "services.vfs_unlink" +#define NETDATA_SYSTEMD_VFS_WRITE_CONTEXT "services.vfs_write" +#define NETDATA_SYSTEMD_VFS_WRITE_ERROR_CONTEXT "services.vfs_write_error" +#define NETDATA_SYSTEMD_VFS_READ_CONTEXT "services.vfs_read" +#define NETDATA_SYSTEMD_VFS_READ_ERROR_CONTEXT "services.vfs_read_error" +#define NETDATA_SYSTEMD_VFS_WRITE_BYTES_CONTEXT "services.vfs_write_bytes" +#define NETDATA_SYSTEMD_VFS_READ_BYTES_CONTEXT "services.vfs_read_bytes" +#define NETDATA_SYSTEMD_VFS_CREATE_CONTEXT "services.vfs_create" +#define NETDATA_SYSTEMD_VFS_CREATE_ERROR_CONTEXT "services.vfs_create_error" +#define NETDATA_SYSTEMD_VFS_OPEN_CONTEXT "services.vfs_open" +#define NETDATA_SYSTEMD_VFS_OPEN_ERROR_CONTEXT "services.vfs_open_error" +#define NETDATA_SYSTEMD_VFS_FSYNC_CONTEXT "services.vfs_fsync" +#define NETDATA_SYSTEMD_VFS_FSYNC_ERROR_CONTEXT "services.vfs_fsync_error" + +typedef struct netdata_publish_vfs { + uint64_t pid_tgid; + uint32_t pid; + uint32_t pad; + + //Counter + uint32_t write_call; + uint32_t writev_call; + uint32_t read_call; + uint32_t readv_call; + uint32_t unlink_call; + uint32_t fsync_call; + uint32_t open_call; + uint32_t create_call; + + //Accumulator + uint64_t write_bytes; + uint64_t writev_bytes; + uint64_t readv_bytes; + uint64_t read_bytes; + + //Counter + uint32_t write_err; + uint32_t writev_err; + uint32_t read_err; + uint32_t readv_err; + uint32_t unlink_err; + uint32_t fsync_err; + uint32_t open_err; + uint32_t create_err; +} netdata_publish_vfs_t; + +enum netdata_publish_vfs_list { + NETDATA_KEY_PUBLISH_VFS_UNLINK, + NETDATA_KEY_PUBLISH_VFS_READ, + NETDATA_KEY_PUBLISH_VFS_WRITE, + NETDATA_KEY_PUBLISH_VFS_FSYNC, + NETDATA_KEY_PUBLISH_VFS_OPEN, + NETDATA_KEY_PUBLISH_VFS_CREATE, + + NETDATA_KEY_PUBLISH_VFS_END +}; + +enum vfs_counters { + NETDATA_KEY_CALLS_VFS_WRITE, + NETDATA_KEY_ERROR_VFS_WRITE, + NETDATA_KEY_BYTES_VFS_WRITE, + + NETDATA_KEY_CALLS_VFS_WRITEV, + NETDATA_KEY_ERROR_VFS_WRITEV, + NETDATA_KEY_BYTES_VFS_WRITEV, + + NETDATA_KEY_CALLS_VFS_READ, + NETDATA_KEY_ERROR_VFS_READ, + NETDATA_KEY_BYTES_VFS_READ, + + NETDATA_KEY_CALLS_VFS_READV, + NETDATA_KEY_ERROR_VFS_READV, + NETDATA_KEY_BYTES_VFS_READV, + + NETDATA_KEY_CALLS_VFS_UNLINK, + NETDATA_KEY_ERROR_VFS_UNLINK, + + NETDATA_KEY_CALLS_VFS_FSYNC, + NETDATA_KEY_ERROR_VFS_FSYNC, + + NETDATA_KEY_CALLS_VFS_OPEN, + NETDATA_KEY_ERROR_VFS_OPEN, + + NETDATA_KEY_CALLS_VFS_CREATE, + NETDATA_KEY_ERROR_VFS_CREATE, + + // Keep this as last and don't skip numbers as it is used as element counter + NETDATA_VFS_COUNTER +}; + +enum netdata_vfs_tables { + NETDATA_VFS_PID, + NETDATA_VFS_ALL, + NETDATA_VFS_CTRL +}; + +enum netdata_vfs_calls_name { + NETDATA_EBPF_VFS_WRITE, + NETDATA_EBPF_VFS_WRITEV, + NETDATA_EBPF_VFS_READ, + NETDATA_EBPF_VFS_READV, + NETDATA_EBPF_VFS_UNLINK, + NETDATA_EBPF_VFS_FSYNC, + NETDATA_EBPF_VFS_OPEN, + NETDATA_EBPF_VFS_CREATE, + + NETDATA_VFS_END_LIST +}; + +extern netdata_publish_vfs_t **vfs_pid; + +void *ebpf_vfs_thread(void *ptr); +void ebpf_vfs_create_apps_charts(struct ebpf_module *em, void *ptr); +extern netdata_ebpf_targets_t vfs_targets[]; + +extern struct config vfs_config; + +#endif /* NETDATA_EBPF_VFS_H */ |