summaryrefslogtreecommitdiffstats
path: root/daemon/daemon.c
diff options
context:
space:
mode:
Diffstat (limited to 'daemon/daemon.c')
-rw-r--r--daemon/daemon.c523
1 files changed, 0 insertions, 523 deletions
diff --git a/daemon/daemon.c b/daemon/daemon.c
deleted file mode 100644
index 433fa0373..000000000
--- a/daemon/daemon.c
+++ /dev/null
@@ -1,523 +0,0 @@
-// SPDX-License-Identifier: GPL-3.0-or-later
-
-#include "common.h"
-#include <sched.h>
-
-char pidfile[FILENAME_MAX + 1] = "";
-char claiming_directory[FILENAME_MAX + 1];
-char netdata_exe_path[FILENAME_MAX + 1];
-char netdata_exe_file[FILENAME_MAX + 1];
-
-void get_netdata_execution_path(void) {
- int ret;
- size_t exepath_size = 0;
- struct passwd *passwd = NULL;
- char *user = NULL;
-
- passwd = getpwuid(getuid());
- user = (passwd && passwd->pw_name) ? passwd->pw_name : "";
-
- exepath_size = sizeof(netdata_exe_file) - 1;
- ret = uv_exepath(netdata_exe_file, &exepath_size);
- if (0 != ret) {
- netdata_log_error("uv_exepath(\"%s\", %u) (user: %s) failed (%s).", netdata_exe_file, (unsigned)exepath_size, user,
- uv_strerror(ret));
- fatal("Cannot start netdata without getting execution path.");
- }
-
- netdata_exe_file[exepath_size] = '\0';
-
- // macOS's dirname(3) does not modify passed string
- char *tmpdir = strdupz(netdata_exe_file);
- strcpy(netdata_exe_path, dirname(tmpdir));
- freez(tmpdir);
-}
-
-static void fix_directory_file_permissions(const char *dirname, uid_t uid, gid_t gid, bool recursive)
-{
- char filename[FILENAME_MAX + 1];
-
- DIR *dir = opendir(dirname);
- if (!dir)
- return;
-
- struct dirent *de = NULL;
-
- while ((de = readdir(dir))) {
- if (de->d_type == DT_DIR && (!strcmp(de->d_name, ".") || !strcmp(de->d_name, "..")))
- continue;
-
- (void) snprintfz(filename, FILENAME_MAX, "%s/%s", dirname, de->d_name);
- if (de->d_type == DT_REG || recursive) {
- if (chown(filename, uid, gid) == -1)
- netdata_log_error("Cannot chown %s '%s' to %u:%u", de->d_type == DT_DIR ? "directory" : "file", filename, (unsigned int)uid, (unsigned int)gid);
- }
-
- if (de->d_type == DT_DIR && recursive)
- fix_directory_file_permissions(filename, uid, gid, recursive);
- }
-
- closedir(dir);
-}
-
-void change_dir_ownership(const char *dir, uid_t uid, gid_t gid, bool recursive)
-{
- if (chown(dir, uid, gid) == -1)
- netdata_log_error("Cannot chown directory '%s' to %u:%u", dir, (unsigned int)uid, (unsigned int)gid);
-
- fix_directory_file_permissions(dir, uid, gid, recursive);
-}
-
-void clean_directory(char *dirname)
-{
- DIR *dir = opendir(dirname);
- if(!dir) return;
-
- int dir_fd = dirfd(dir);
- struct dirent *de = NULL;
-
- while((de = readdir(dir)))
- if(de->d_type == DT_REG)
- if (unlinkat(dir_fd, de->d_name, 0))
- netdata_log_error("Cannot delete %s/%s", dirname, de->d_name);
-
- closedir(dir);
-}
-
-void prepare_required_directories(uid_t uid, gid_t gid) {
- change_dir_ownership(netdata_configured_cache_dir, uid, gid, true);
- change_dir_ownership(netdata_configured_varlib_dir, uid, gid, false);
- change_dir_ownership(netdata_configured_lock_dir, uid, gid, false);
- change_dir_ownership(netdata_configured_log_dir, uid, gid, false);
- change_dir_ownership(claiming_directory, uid, gid, false);
-
- char filename[FILENAME_MAX + 1];
- snprintfz(filename, FILENAME_MAX, "%s/registry", netdata_configured_varlib_dir);
- change_dir_ownership(filename, uid, gid, false);
-
- clean_directory(netdata_configured_lock_dir);
-}
-
-int become_user(const char *username, int pid_fd) {
- int am_i_root = (getuid() == 0)?1:0;
-
- struct passwd *pw = getpwnam(username);
- if(!pw) {
- netdata_log_error("User %s is not present.", username);
- return -1;
- }
-
- uid_t uid = pw->pw_uid;
- gid_t gid = pw->pw_gid;
-
- prepare_required_directories(uid, gid);
-
- if(pidfile[0]) {
- if(chown(pidfile, uid, gid) == -1)
- netdata_log_error("Cannot chown '%s' to %u:%u", pidfile, (unsigned int)uid, (unsigned int)gid);
- }
-
- int ngroups = (int)sysconf(_SC_NGROUPS_MAX);
- gid_t *supplementary_groups = NULL;
- if(ngroups > 0) {
- supplementary_groups = mallocz(sizeof(gid_t) * ngroups);
-#ifdef __APPLE__
- if(getgrouplist(username, gid, (int *)supplementary_groups, &ngroups) == -1) {
-#else
- if(getgrouplist(username, gid, supplementary_groups, &ngroups) == -1) {
-#endif /* __APPLE__ */
- if(am_i_root)
- netdata_log_error("Cannot get supplementary groups of user '%s'.", username);
-
- ngroups = 0;
- }
- }
-
- nd_log_chown_log_files(uid, gid);
- chown_open_file(STDOUT_FILENO, uid, gid);
- chown_open_file(STDERR_FILENO, uid, gid);
- chown_open_file(pid_fd, uid, gid);
-
- if(supplementary_groups && ngroups > 0) {
- if(setgroups((size_t)ngroups, supplementary_groups) == -1) {
- if(am_i_root)
- netdata_log_error("Cannot set supplementary groups for user '%s'", username);
- }
- ngroups = 0;
- }
-
- if(supplementary_groups)
- freez(supplementary_groups);
-
-#ifdef __APPLE__
- if(setregid(gid, gid) != 0) {
-#else
- if(setresgid(gid, gid, gid) != 0) {
-#endif /* __APPLE__ */
- netdata_log_error("Cannot switch to user's %s group (gid: %u).", username, gid);
- return -1;
- }
-
-#ifdef __APPLE__
- if(setreuid(uid, uid) != 0) {
-#else
- if(setresuid(uid, uid, uid) != 0) {
-#endif /* __APPLE__ */
- netdata_log_error("Cannot switch to user %s (uid: %u).", username, uid);
- return -1;
- }
-
- if(setgid(gid) != 0) {
- netdata_log_error("Cannot switch to user's %s group (gid: %u).", username, gid);
- return -1;
- }
- if(setegid(gid) != 0) {
- netdata_log_error("Cannot effectively switch to user's %s group (gid: %u).", username, gid);
- return -1;
- }
- if(setuid(uid) != 0) {
- netdata_log_error("Cannot switch to user %s (uid: %u).", username, uid);
- return -1;
- }
- if(seteuid(uid) != 0) {
- netdata_log_error("Cannot effectively switch to user %s (uid: %u).", username, uid);
- return -1;
- }
-
- return(0);
-}
-
-#ifndef OOM_SCORE_ADJ_MAX
-#define OOM_SCORE_ADJ_MAX (1000)
-#endif
-#ifndef OOM_SCORE_ADJ_MIN
-#define OOM_SCORE_ADJ_MIN (-1000)
-#endif
-
-static void oom_score_adj(void) {
- char buf[30 + 1];
- long long int old_score, wanted_score = 0, final_score = 0;
-
- // read the existing score
- if(read_single_signed_number_file("/proc/self/oom_score_adj", &old_score)) {
- netdata_log_error("Out-Of-Memory (OOM) score setting is not supported on this system.");
- return;
- }
-
- if (old_score != 0) {
- wanted_score = old_score;
- analytics_report_oom_score(old_score);
- }
-
- // check the environment
- char *s = getenv("OOMScoreAdjust");
- if(!s || !*s) {
- snprintfz(buf, sizeof(buf) - 1, "%d", (int)wanted_score);
- s = buf;
- }
-
- // check netdata.conf configuration
- s = config_get(CONFIG_SECTION_GLOBAL, "OOM score", s);
- if(s && *s && (isdigit(*s) || *s == '-' || *s == '+'))
- wanted_score = atoll(s);
- else if(s && !strcmp(s, "keep")) {
- netdata_log_info("Out-Of-Memory (OOM) kept as-is (running with %d)", (int) old_score);
- return;
- }
- else {
- netdata_log_info("Out-Of-Memory (OOM) score not changed due to non-numeric setting: '%s' (running with %d)", s, (int)old_score);
- return;
- }
-
- if(wanted_score < OOM_SCORE_ADJ_MIN) {
- netdata_log_error("Wanted Out-Of-Memory (OOM) score %d is too small. Using %d", (int)wanted_score, (int)OOM_SCORE_ADJ_MIN);
- wanted_score = OOM_SCORE_ADJ_MIN;
- }
-
- if(wanted_score > OOM_SCORE_ADJ_MAX) {
- netdata_log_error("Wanted Out-Of-Memory (OOM) score %d is too big. Using %d", (int)wanted_score, (int)OOM_SCORE_ADJ_MAX);
- wanted_score = OOM_SCORE_ADJ_MAX;
- }
-
- if(old_score == wanted_score) {
- netdata_log_info("Out-Of-Memory (OOM) score is already set to the wanted value %d", (int)old_score);
- return;
- }
-
- int written = 0;
- int fd = open("/proc/self/oom_score_adj", O_WRONLY);
- if(fd != -1) {
- snprintfz(buf, sizeof(buf) - 1, "%d", (int)wanted_score);
- ssize_t len = strlen(buf);
- if(len > 0 && write(fd, buf, (size_t)len) == len) written = 1;
- close(fd);
-
- if(written) {
- if(read_single_signed_number_file("/proc/self/oom_score_adj", &final_score))
- netdata_log_error("Adjusted my Out-Of-Memory (OOM) score to %d, but cannot verify it.", (int)wanted_score);
- else if(final_score == wanted_score)
- netdata_log_info("Adjusted my Out-Of-Memory (OOM) score from %d to %d.", (int)old_score, (int)final_score);
- else
- netdata_log_error("Adjusted my Out-Of-Memory (OOM) score from %d to %d, but it has been set to %d.", (int)old_score, (int)wanted_score, (int)final_score);
- analytics_report_oom_score(final_score);
- }
- else
- netdata_log_error("Failed to adjust my Out-Of-Memory (OOM) score to %d. Running with %d. (systemd systems may change it via netdata.service)", (int)wanted_score, (int)old_score);
- }
- else
- netdata_log_error("Failed to adjust my Out-Of-Memory (OOM) score. Cannot open /proc/self/oom_score_adj for writing.");
-}
-
-static void process_nice_level(void) {
-#ifdef HAVE_NICE
- int nice_level = (int)config_get_number(CONFIG_SECTION_GLOBAL, "process nice level", 19);
- if(nice(nice_level) == -1)
- netdata_log_error("Cannot set netdata CPU nice level to %d.", nice_level);
- else
- netdata_log_debug(D_SYSTEM, "Set netdata nice level to %d.", nice_level);
-#endif // HAVE_NICE
-};
-
-#define SCHED_FLAG_NONE 0x00
-#define SCHED_FLAG_PRIORITY_CONFIGURABLE 0x01 // the priority is user configurable
-#define SCHED_FLAG_KEEP_AS_IS 0x04 // do not attempt to set policy, priority or nice()
-#define SCHED_FLAG_USE_NICE 0x08 // use nice() after setting this policy
-
-struct sched_def {
- char *name;
- int policy;
- int priority;
- uint8_t flags;
-} scheduler_defaults[] = {
-
- // the order of array members is important!
- // the first defined is the default used by netdata
-
- // the available members are important too!
- // these are all the possible scheduling policies supported by netdata
-
-#ifdef SCHED_BATCH
- { "batch", SCHED_BATCH, 0, SCHED_FLAG_USE_NICE },
-#endif
-
-#ifdef SCHED_OTHER
- { "other", SCHED_OTHER, 0, SCHED_FLAG_USE_NICE },
- { "nice", SCHED_OTHER, 0, SCHED_FLAG_USE_NICE },
-#endif
-
-#ifdef SCHED_IDLE
- { "idle", SCHED_IDLE, 0, SCHED_FLAG_NONE },
-#endif
-
-#ifdef SCHED_RR
- { "rr", SCHED_RR, 0, SCHED_FLAG_PRIORITY_CONFIGURABLE },
-#endif
-
-#ifdef SCHED_FIFO
- { "fifo", SCHED_FIFO, 0, SCHED_FLAG_PRIORITY_CONFIGURABLE },
-#endif
-
- // do not change the scheduling priority
- { "keep", 0, 0, SCHED_FLAG_KEEP_AS_IS },
- { "none", 0, 0, SCHED_FLAG_KEEP_AS_IS },
-
- // array termination
- { NULL, 0, 0, 0 }
-};
-
-
-#ifdef HAVE_SCHED_GETSCHEDULER
-static void sched_getscheduler_report(void) {
- int sched = sched_getscheduler(0);
- if(sched == -1) {
- netdata_log_error("Cannot get my current process scheduling policy.");
- return;
- }
- else {
- int i;
- for(i = 0 ; scheduler_defaults[i].name ; i++) {
- if(scheduler_defaults[i].policy == sched) {
- if(scheduler_defaults[i].flags & SCHED_FLAG_PRIORITY_CONFIGURABLE) {
- struct sched_param param;
- if(sched_getparam(0, &param) == -1) {
- netdata_log_error("Cannot get the process scheduling priority for my policy '%s'", scheduler_defaults[i].name);
- return;
- }
- else {
- netdata_log_info("Running with process scheduling policy '%s', priority %d", scheduler_defaults[i].name, param.sched_priority);
- }
- }
- else if(scheduler_defaults[i].flags & SCHED_FLAG_USE_NICE) {
- #ifdef HAVE_GETPRIORITY
- int n = getpriority(PRIO_PROCESS, 0);
- netdata_log_info("Running with process scheduling policy '%s', nice level %d", scheduler_defaults[i].name, n);
- #else // !HAVE_GETPRIORITY
- netdata_log_info("Running with process scheduling policy '%s'", scheduler_defaults[i].name);
- #endif // !HAVE_GETPRIORITY
- }
- else {
- netdata_log_info("Running with process scheduling policy '%s'", scheduler_defaults[i].name);
- }
-
- return;
- }
- }
- }
-}
-#endif /* HAVE_SCHED_GETSCHEDULER */
-
-#ifdef HAVE_SCHED_SETSCHEDULER
-
-static void sched_setscheduler_set(void) {
-
- if(scheduler_defaults[0].name) {
- const char *name = scheduler_defaults[0].name;
- int policy = scheduler_defaults[0].policy, priority = scheduler_defaults[0].priority;
- uint8_t flags = scheduler_defaults[0].flags;
- int found = 0;
-
- // read the configuration
- name = config_get(CONFIG_SECTION_GLOBAL, "process scheduling policy", name);
- int i;
- for(i = 0 ; scheduler_defaults[i].name ; i++) {
- if(!strcmp(name, scheduler_defaults[i].name)) {
- found = 1;
- policy = scheduler_defaults[i].policy;
- priority = scheduler_defaults[i].priority;
- flags = scheduler_defaults[i].flags;
-
- if(flags & SCHED_FLAG_KEEP_AS_IS)
- goto report;
-
- if(flags & SCHED_FLAG_PRIORITY_CONFIGURABLE)
- priority = (int)config_get_number(CONFIG_SECTION_GLOBAL, "process scheduling priority", priority);
-
-#ifdef HAVE_SCHED_GET_PRIORITY_MIN
- errno = 0;
- if(priority < sched_get_priority_min(policy)) {
- netdata_log_error("scheduler %s (%d) priority %d is below the minimum %d. Using the minimum.", name, policy, priority, sched_get_priority_min(policy));
- priority = sched_get_priority_min(policy);
- }
-#endif
-#ifdef HAVE_SCHED_GET_PRIORITY_MAX
- errno = 0;
- if(priority > sched_get_priority_max(policy)) {
- netdata_log_error("scheduler %s (%d) priority %d is above the maximum %d. Using the maximum.", name, policy, priority, sched_get_priority_max(policy));
- priority = sched_get_priority_max(policy);
- }
-#endif
- break;
- }
- }
-
- if(!found) {
- netdata_log_error("Unknown scheduling policy '%s' - falling back to nice", name);
- goto fallback;
- }
-
- const struct sched_param param = {
- .sched_priority = priority
- };
-
- errno = 0;
- i = sched_setscheduler(0, policy, &param);
- if(i != 0) {
- netdata_log_error("Cannot adjust netdata scheduling policy to %s (%d), with priority %d. Falling back to nice.",
- name,
- policy,
- priority);
- }
- else {
- netdata_log_info("Adjusted netdata scheduling policy to %s (%d), with priority %d.", name, policy, priority);
- if(!(flags & SCHED_FLAG_USE_NICE))
- goto report;
- }
- }
-
-fallback:
- process_nice_level();
-
-report:
- sched_getscheduler_report();
-}
-#else /* HAVE_SCHED_SETSCHEDULER */
-static void sched_setscheduler_set(void) {
- process_nice_level();
-}
-#endif /* HAVE_SCHED_SETSCHEDULER */
-
-int become_daemon(int dont_fork, const char *user)
-{
- if(!dont_fork) {
- int i = fork();
- if(i == -1) {
- perror("cannot fork");
- exit(1);
- }
- if(i != 0) {
- exit(0); // the parent
- }
-
- // become session leader
- if (setsid() < 0) {
- perror("Cannot become session leader.");
- exit(2);
- }
-
- // fork() again
- i = fork();
- if(i == -1) {
- perror("cannot fork");
- exit(1);
- }
- if(i != 0) {
- exit(0); // the parent
- }
- }
-
- // generate our pid file
- int pidfd = -1;
- if(pidfile[0]) {
- pidfd = open(pidfile, O_WRONLY | O_CREAT, 0644);
- if(pidfd >= 0) {
- if(ftruncate(pidfd, 0) != 0)
- netdata_log_error("Cannot truncate pidfile '%s'.", pidfile);
-
- char b[100];
- sprintf(b, "%d\n", getpid());
- ssize_t i = write(pidfd, b, strlen(b));
- if(i <= 0)
- netdata_log_error("Cannot write pidfile '%s'.", pidfile);
- }
- else
- netdata_log_error("Failed to open pidfile '%s'.", pidfile);
- }
-
- // Set new file permissions
- umask(0007);
-
- // adjust my Out-Of-Memory score
- oom_score_adj();
-
- // never become a problem
- sched_setscheduler_set();
-
- // Set claiming directory based on user config directory with correct ownership
- snprintfz(claiming_directory, FILENAME_MAX, "%s/cloud.d", netdata_configured_varlib_dir);
-
- if(user && *user) {
- if(become_user(user, pidfd) != 0) {
- netdata_log_error("Cannot become user '%s'. Continuing as we are.", user);
- }
- else
- netdata_log_debug(D_SYSTEM, "Successfully became user '%s'.", user);
- }
- else {
- prepare_required_directories(getuid(), getgid());
- }
-
- if(pidfd != -1)
- close(pidfd);
-
- return(0);
-}