summaryrefslogtreecommitdiffstats
path: root/src/coredump
diff options
context:
space:
mode:
authorDaniel Baumann <daniel.baumann@progress-linux.org>2024-04-10 20:49:52 +0000
committerDaniel Baumann <daniel.baumann@progress-linux.org>2024-04-10 20:49:52 +0000
commit55944e5e40b1be2afc4855d8d2baf4b73d1876b5 (patch)
tree33f869f55a1b149e9b7c2b7e201867ca5dd52992 /src/coredump
parentInitial commit. (diff)
downloadsystemd-55944e5e40b1be2afc4855d8d2baf4b73d1876b5.tar.xz
systemd-55944e5e40b1be2afc4855d8d2baf4b73d1876b5.zip
Adding upstream version 255.4.upstream/255.4
Signed-off-by: Daniel Baumann <daniel.baumann@progress-linux.org>
Diffstat (limited to 'src/coredump')
-rw-r--r--src/coredump/coredump-vacuum.c244
-rw-r--r--src/coredump/coredump-vacuum.h7
-rw-r--r--src/coredump/coredump.c1718
-rw-r--r--src/coredump/coredump.conf27
-rw-r--r--src/coredump/coredumpctl.c1418
-rw-r--r--src/coredump/meson.build48
-rw-r--r--src/coredump/test-coredump-vacuum.c15
7 files changed, 3477 insertions, 0 deletions
diff --git a/src/coredump/coredump-vacuum.c b/src/coredump/coredump-vacuum.c
new file mode 100644
index 0000000..7e0c98c
--- /dev/null
+++ b/src/coredump/coredump-vacuum.c
@@ -0,0 +1,244 @@
+/* SPDX-License-Identifier: LGPL-2.1-or-later */
+
+#include <sys/statvfs.h>
+#include <sys/stat.h>
+#include <sys/types.h>
+#include <unistd.h>
+
+#include "alloc-util.h"
+#include "coredump-vacuum.h"
+#include "dirent-util.h"
+#include "fd-util.h"
+#include "fs-util.h"
+#include "hashmap.h"
+#include "macro.h"
+#include "memory-util.h"
+#include "stat-util.h"
+#include "string-util.h"
+#include "time-util.h"
+#include "user-util.h"
+
+#define DEFAULT_MAX_USE_LOWER (uint64_t) (1ULL*1024ULL*1024ULL) /* 1 MiB */
+#define DEFAULT_MAX_USE_UPPER (uint64_t) (4ULL*1024ULL*1024ULL*1024ULL) /* 4 GiB */
+#define DEFAULT_KEEP_FREE_UPPER (uint64_t) (4ULL*1024ULL*1024ULL*1024ULL) /* 4 GiB */
+#define DEFAULT_KEEP_FREE (uint64_t) (1024ULL*1024ULL) /* 1 MB */
+
+typedef struct VacuumCandidate {
+ unsigned n_files;
+ char *oldest_file;
+ usec_t oldest_mtime;
+} VacuumCandidate;
+
+static VacuumCandidate* vacuum_candidate_free(VacuumCandidate *c) {
+ if (!c)
+ return NULL;
+
+ free(c->oldest_file);
+ return mfree(c);
+}
+DEFINE_TRIVIAL_CLEANUP_FUNC(VacuumCandidate*, vacuum_candidate_free);
+
+static Hashmap* vacuum_candidate_hashmap_free(Hashmap *h) {
+ return hashmap_free_with_destructor(h, vacuum_candidate_free);
+}
+
+DEFINE_TRIVIAL_CLEANUP_FUNC(Hashmap*, vacuum_candidate_hashmap_free);
+
+static int uid_from_file_name(const char *filename, uid_t *uid) {
+ const char *p, *e, *u;
+
+ p = startswith(filename, "core.");
+ if (!p)
+ return -EINVAL;
+
+ /* Skip the comm field */
+ p = strchr(p, '.');
+ if (!p)
+ return -EINVAL;
+ p++;
+
+ /* Find end up UID */
+ e = strchr(p, '.');
+ if (!e)
+ return -EINVAL;
+
+ u = strndupa_safe(p, e - p);
+ return parse_uid(u, uid);
+}
+
+static bool vacuum_necessary(int fd, uint64_t sum, uint64_t keep_free, uint64_t max_use) {
+ uint64_t fs_size = 0, fs_free = UINT64_MAX;
+ struct statvfs sv;
+
+ assert(fd >= 0);
+
+ if (fstatvfs(fd, &sv) >= 0) {
+ fs_size = sv.f_frsize * sv.f_blocks;
+ fs_free = sv.f_frsize * sv.f_bfree;
+ }
+
+ if (max_use == UINT64_MAX) {
+
+ if (fs_size > 0) {
+ max_use = PAGE_ALIGN(fs_size / 10); /* 10% */
+
+ if (max_use > DEFAULT_MAX_USE_UPPER)
+ max_use = DEFAULT_MAX_USE_UPPER;
+
+ if (max_use < DEFAULT_MAX_USE_LOWER)
+ max_use = DEFAULT_MAX_USE_LOWER;
+ } else
+ max_use = DEFAULT_MAX_USE_LOWER;
+ } else
+ max_use = PAGE_ALIGN(max_use);
+
+ if (max_use > 0 && sum > max_use)
+ return true;
+
+ if (keep_free == UINT64_MAX) {
+
+ if (fs_size > 0) {
+ keep_free = PAGE_ALIGN((fs_size * 3) / 20); /* 15% */
+
+ if (keep_free > DEFAULT_KEEP_FREE_UPPER)
+ keep_free = DEFAULT_KEEP_FREE_UPPER;
+ } else
+ keep_free = DEFAULT_KEEP_FREE;
+ } else
+ keep_free = PAGE_ALIGN(keep_free);
+
+ if (keep_free > 0 && fs_free < keep_free)
+ return true;
+
+ return false;
+}
+
+int coredump_vacuum(int exclude_fd, uint64_t keep_free, uint64_t max_use) {
+ _cleanup_closedir_ DIR *d = NULL;
+ struct stat exclude_st;
+ int r;
+
+ if (keep_free == 0 && max_use == 0)
+ return 0;
+
+ if (exclude_fd >= 0) {
+ if (fstat(exclude_fd, &exclude_st) < 0)
+ return log_error_errno(errno, "Failed to fstat(): %m");
+ }
+
+ /* This algorithm will keep deleting the oldest file of the
+ * user with the most coredumps until we are back in the size
+ * limits. Note that vacuuming for journal files is different,
+ * because we rely on rate-limiting of the messages there,
+ * to avoid being flooded. */
+
+ d = opendir("/var/lib/systemd/coredump");
+ if (!d) {
+ if (errno == ENOENT)
+ return 0;
+
+ return log_error_errno(errno, "Can't open coredump directory: %m");
+ }
+
+ for (;;) {
+ _cleanup_(vacuum_candidate_hashmap_freep) Hashmap *h = NULL;
+ VacuumCandidate *worst = NULL;
+ uint64_t sum = 0;
+
+ rewinddir(d);
+
+ FOREACH_DIRENT(de, d, goto fail) {
+ VacuumCandidate *c;
+ struct stat st;
+ uid_t uid;
+ usec_t t;
+
+ r = uid_from_file_name(de->d_name, &uid);
+ if (r < 0)
+ continue;
+
+ if (fstatat(dirfd(d), de->d_name, &st, AT_NO_AUTOMOUNT|AT_SYMLINK_NOFOLLOW) < 0) {
+ if (errno == ENOENT)
+ continue;
+
+ log_warning_errno(errno, "Failed to stat /var/lib/systemd/coredump/%s: %m", de->d_name);
+ continue;
+ }
+
+ if (!S_ISREG(st.st_mode))
+ continue;
+
+ if (exclude_fd >= 0 && stat_inode_same(&exclude_st, &st))
+ continue;
+
+ r = hashmap_ensure_allocated(&h, NULL);
+ if (r < 0)
+ return log_oom();
+
+ t = timespec_load(&st.st_mtim);
+
+ c = hashmap_get(h, UID_TO_PTR(uid));
+ if (c) {
+
+ if (t < c->oldest_mtime) {
+ char *n;
+
+ n = strdup(de->d_name);
+ if (!n)
+ return log_oom();
+
+ free_and_replace(c->oldest_file, n);
+ c->oldest_mtime = t;
+ }
+
+ } else {
+ _cleanup_(vacuum_candidate_freep) VacuumCandidate *n = NULL;
+
+ n = new0(VacuumCandidate, 1);
+ if (!n)
+ return log_oom();
+
+ n->oldest_file = strdup(de->d_name);
+ if (!n->oldest_file)
+ return log_oom();
+
+ n->oldest_mtime = t;
+
+ r = hashmap_put(h, UID_TO_PTR(uid), n);
+ if (r < 0)
+ return log_oom();
+
+ c = TAKE_PTR(n);
+ }
+
+ c->n_files++;
+
+ if (!worst ||
+ worst->n_files < c->n_files ||
+ (worst->n_files == c->n_files && c->oldest_mtime < worst->oldest_mtime))
+ worst = c;
+
+ sum += st.st_blocks * 512;
+ }
+
+ if (!worst)
+ break;
+
+ r = vacuum_necessary(dirfd(d), sum, keep_free, max_use);
+ if (r <= 0)
+ return r;
+
+ r = unlinkat_deallocate(dirfd(d), worst->oldest_file, 0);
+ if (r == -ENOENT)
+ continue;
+ if (r < 0)
+ return log_error_errno(r, "Failed to remove file %s: %m", worst->oldest_file);
+
+ log_info("Removed old coredump %s.", worst->oldest_file);
+ }
+
+ return 0;
+
+fail:
+ return log_error_errno(errno, "Failed to read directory: %m");
+}
diff --git a/src/coredump/coredump-vacuum.h b/src/coredump/coredump-vacuum.h
new file mode 100644
index 0000000..8ad5baf
--- /dev/null
+++ b/src/coredump/coredump-vacuum.h
@@ -0,0 +1,7 @@
+/* SPDX-License-Identifier: LGPL-2.1-or-later */
+#pragma once
+
+#include <inttypes.h>
+#include <sys/types.h>
+
+int coredump_vacuum(int exclude_fd, uint64_t keep_free, uint64_t max_use);
diff --git a/src/coredump/coredump.c b/src/coredump/coredump.c
new file mode 100644
index 0000000..32c1766
--- /dev/null
+++ b/src/coredump/coredump.c
@@ -0,0 +1,1718 @@
+/* SPDX-License-Identifier: LGPL-2.1-or-later */
+
+#include <errno.h>
+#include <stdio.h>
+#include <sys/prctl.h>
+#include <sys/statvfs.h>
+#include <sys/auxv.h>
+#include <sys/xattr.h>
+#include <unistd.h>
+
+#include "sd-daemon.h"
+#include "sd-journal.h"
+#include "sd-login.h"
+#include "sd-messages.h"
+
+#include "acl-util.h"
+#include "alloc-util.h"
+#include "bus-error.h"
+#include "capability-util.h"
+#include "cgroup-util.h"
+#include "compress.h"
+#include "conf-parser.h"
+#include "copy.h"
+#include "coredump-util.h"
+#include "coredump-vacuum.h"
+#include "dirent-util.h"
+#include "elf-util.h"
+#include "escape.h"
+#include "fd-util.h"
+#include "fileio.h"
+#include "fs-util.h"
+#include "iovec-util.h"
+#include "journal-importer.h"
+#include "journal-send.h"
+#include "log.h"
+#include "macro.h"
+#include "main-func.h"
+#include "memory-util.h"
+#include "memstream-util.h"
+#include "mkdir-label.h"
+#include "namespace-util.h"
+#include "parse-util.h"
+#include "path-util.h"
+#include "process-util.h"
+#include "signal-util.h"
+#include "socket-util.h"
+#include "special.h"
+#include "stat-util.h"
+#include "string-table.h"
+#include "string-util.h"
+#include "strv.h"
+#include "sync-util.h"
+#include "tmpfile-util.h"
+#include "uid-alloc-range.h"
+#include "user-util.h"
+
+/* The maximum size up to which we process coredumps. We use 1G on 32-bit systems, and 32G on 64-bit systems */
+#if __SIZEOF_POINTER__ == 4
+#define PROCESS_SIZE_MAX ((uint64_t) (1LLU*1024LLU*1024LLU*1024LLU))
+#elif __SIZEOF_POINTER__ == 8
+#define PROCESS_SIZE_MAX ((uint64_t) (32LLU*1024LLU*1024LLU*1024LLU))
+#else
+#error "Unexpected pointer size"
+#endif
+
+/* The maximum size up to which we leave the coredump around on disk */
+#define EXTERNAL_SIZE_MAX PROCESS_SIZE_MAX
+
+/* The maximum size up to which we store the coredump in the journal */
+#ifndef FUZZING_BUILD_MODE_UNSAFE_FOR_PRODUCTION
+#define JOURNAL_SIZE_MAX ((size_t) (767LU*1024LU*1024LU))
+#else
+/* oss-fuzz limits memory usage. */
+#define JOURNAL_SIZE_MAX ((size_t) (10LU*1024LU*1024LU))
+#endif
+
+/* When checking for available memory and setting lower limits, don't
+ * go below 4MB for writing core files to storage. */
+#define PROCESS_SIZE_MIN (4U*1024U*1024U)
+
+/* Make sure to not make this larger than the maximum journal entry
+ * size. See DATA_SIZE_MAX in journal-importer.h. */
+assert_cc(JOURNAL_SIZE_MAX <= DATA_SIZE_MAX);
+
+enum {
+ /* We use these as array indexes for our process metadata cache.
+ *
+ * The first indices of the cache stores the same metadata as the ones passed by
+ * the kernel via argv[], ie the strings array passed by the kernel according to
+ * our pattern defined in /proc/sys/kernel/core_pattern (see man:core(5)). */
+
+ META_ARGV_PID, /* %P: as seen in the initial pid namespace */
+ META_ARGV_UID, /* %u: as seen in the initial user namespace */
+ META_ARGV_GID, /* %g: as seen in the initial user namespace */
+ META_ARGV_SIGNAL, /* %s: number of signal causing dump */
+ META_ARGV_TIMESTAMP, /* %t: time of dump, expressed as seconds since the Epoch (we expand this to μs granularity) */
+ META_ARGV_RLIMIT, /* %c: core file size soft resource limit */
+ META_ARGV_HOSTNAME, /* %h: hostname */
+ _META_ARGV_MAX,
+
+ /* The following indexes are cached for a couple of special fields we use (and
+ * thereby need to be retrieved quickly) for naming coredump files, and attaching
+ * xattrs. Unlike the previous ones they are retrieved from the runtime
+ * environment. */
+
+ META_COMM = _META_ARGV_MAX,
+ _META_MANDATORY_MAX,
+
+ /* The rest are similar to the previous ones except that we won't fail if one of
+ * them is missing. */
+
+ META_EXE = _META_MANDATORY_MAX,
+ META_UNIT,
+ META_PROC_AUXV,
+ _META_MAX
+};
+
+static const char * const meta_field_names[_META_MAX] = {
+ [META_ARGV_PID] = "COREDUMP_PID=",
+ [META_ARGV_UID] = "COREDUMP_UID=",
+ [META_ARGV_GID] = "COREDUMP_GID=",
+ [META_ARGV_SIGNAL] = "COREDUMP_SIGNAL=",
+ [META_ARGV_TIMESTAMP] = "COREDUMP_TIMESTAMP=",
+ [META_ARGV_RLIMIT] = "COREDUMP_RLIMIT=",
+ [META_ARGV_HOSTNAME] = "COREDUMP_HOSTNAME=",
+ [META_COMM] = "COREDUMP_COMM=",
+ [META_EXE] = "COREDUMP_EXE=",
+ [META_UNIT] = "COREDUMP_UNIT=",
+ [META_PROC_AUXV] = "COREDUMP_PROC_AUXV=",
+};
+
+typedef struct Context {
+ const char *meta[_META_MAX];
+ size_t meta_size[_META_MAX];
+ pid_t pid;
+ uid_t uid;
+ gid_t gid;
+ bool is_pid1;
+ bool is_journald;
+} Context;
+
+typedef enum CoredumpStorage {
+ COREDUMP_STORAGE_NONE,
+ COREDUMP_STORAGE_EXTERNAL,
+ COREDUMP_STORAGE_JOURNAL,
+ _COREDUMP_STORAGE_MAX,
+ _COREDUMP_STORAGE_INVALID = -EINVAL,
+} CoredumpStorage;
+
+static const char* const coredump_storage_table[_COREDUMP_STORAGE_MAX] = {
+ [COREDUMP_STORAGE_NONE] = "none",
+ [COREDUMP_STORAGE_EXTERNAL] = "external",
+ [COREDUMP_STORAGE_JOURNAL] = "journal",
+};
+
+DEFINE_PRIVATE_STRING_TABLE_LOOKUP(coredump_storage, CoredumpStorage);
+static DEFINE_CONFIG_PARSE_ENUM(config_parse_coredump_storage, coredump_storage, CoredumpStorage, "Failed to parse storage setting");
+
+static CoredumpStorage arg_storage = COREDUMP_STORAGE_EXTERNAL;
+static bool arg_compress = true;
+static uint64_t arg_process_size_max = PROCESS_SIZE_MAX;
+static uint64_t arg_external_size_max = EXTERNAL_SIZE_MAX;
+static uint64_t arg_journal_size_max = JOURNAL_SIZE_MAX;
+static uint64_t arg_keep_free = UINT64_MAX;
+static uint64_t arg_max_use = UINT64_MAX;
+
+static int parse_config(void) {
+ static const ConfigTableItem items[] = {
+ { "Coredump", "Storage", config_parse_coredump_storage, 0, &arg_storage },
+ { "Coredump", "Compress", config_parse_bool, 0, &arg_compress },
+ { "Coredump", "ProcessSizeMax", config_parse_iec_uint64, 0, &arg_process_size_max },
+ { "Coredump", "ExternalSizeMax", config_parse_iec_uint64_infinity, 0, &arg_external_size_max },
+ { "Coredump", "JournalSizeMax", config_parse_iec_size, 0, &arg_journal_size_max },
+ { "Coredump", "KeepFree", config_parse_iec_uint64, 0, &arg_keep_free },
+ { "Coredump", "MaxUse", config_parse_iec_uint64, 0, &arg_max_use },
+ {}
+ };
+
+ int r;
+
+ r = config_parse_config_file(
+ "coredump.conf",
+ "Coredump\0",
+ config_item_table_lookup,
+ items,
+ CONFIG_PARSE_WARN,
+ /* userdata= */ NULL);
+ if (r < 0)
+ return r;
+
+ /* Let's make sure we fix up the maximum size we send to the journal here on the client side, for
+ * efficiency reasons. journald wouldn't accept anything larger anyway. */
+ if (arg_journal_size_max > JOURNAL_SIZE_MAX) {
+ log_warning("JournalSizeMax= set to larger value (%s) than journald would accept (%s), lowering automatically.",
+ FORMAT_BYTES(arg_journal_size_max), FORMAT_BYTES(JOURNAL_SIZE_MAX));
+ arg_journal_size_max = JOURNAL_SIZE_MAX;
+ }
+
+ return 0;
+}
+
+static uint64_t storage_size_max(void) {
+ if (arg_storage == COREDUMP_STORAGE_EXTERNAL)
+ return arg_external_size_max;
+ if (arg_storage == COREDUMP_STORAGE_JOURNAL)
+ return arg_journal_size_max;
+ assert(arg_storage == COREDUMP_STORAGE_NONE);
+ return 0;
+}
+
+static int fix_acl(int fd, uid_t uid, bool allow_user) {
+ assert(fd >= 0);
+ assert(uid_is_valid(uid));
+
+#if HAVE_ACL
+ int r;
+
+ /* We don't allow users to read coredumps if the uid or capabilities were changed. */
+ if (!allow_user)
+ return 0;
+
+ if (uid_is_system(uid) || uid_is_dynamic(uid) || uid == UID_NOBODY)
+ return 0;
+
+ /* Make sure normal users can read (but not write or delete) their own coredumps */
+ r = fd_add_uid_acl_permission(fd, uid, ACL_READ);
+ if (r < 0)
+ return log_error_errno(r, "Failed to adjust ACL of the coredump: %m");
+#endif
+
+ return 0;
+}
+
+static int fix_xattr(int fd, const Context *context) {
+
+ static const char * const xattrs[_META_MAX] = {
+ [META_ARGV_PID] = "user.coredump.pid",
+ [META_ARGV_UID] = "user.coredump.uid",
+ [META_ARGV_GID] = "user.coredump.gid",
+ [META_ARGV_SIGNAL] = "user.coredump.signal",
+ [META_ARGV_TIMESTAMP] = "user.coredump.timestamp",
+ [META_ARGV_RLIMIT] = "user.coredump.rlimit",
+ [META_ARGV_HOSTNAME] = "user.coredump.hostname",
+ [META_COMM] = "user.coredump.comm",
+ [META_EXE] = "user.coredump.exe",
+ };
+
+ int r = 0;
+
+ assert(fd >= 0);
+
+ /* Attach some metadata to coredumps via extended attributes. Just because we can. */
+
+ for (unsigned i = 0; i < _META_MAX; i++) {
+ int k;
+
+ if (isempty(context->meta[i]) || !xattrs[i])
+ continue;
+
+ k = RET_NERRNO(fsetxattr(fd, xattrs[i], context->meta[i], strlen(context->meta[i]), XATTR_CREATE));
+ RET_GATHER(r, k);
+ }
+
+ return r;
+}
+
+#define filename_escape(s) xescape((s), "./ ")
+
+static const char *coredump_tmpfile_name(const char *s) {
+ return s ?: "(unnamed temporary file)";
+}
+
+static int fix_permissions(
+ int fd,
+ const char *filename,
+ const char *target,
+ const Context *context,
+ bool allow_user) {
+
+ int r;
+
+ assert(fd >= 0);
+ assert(target);
+ assert(context);
+
+ /* Ignore errors on these */
+ (void) fchmod(fd, 0640);
+ (void) fix_acl(fd, context->uid, allow_user);
+ (void) fix_xattr(fd, context);
+
+ r = link_tmpfile(fd, filename, target, LINK_TMPFILE_SYNC);
+ if (r < 0)
+ return log_error_errno(r, "Failed to move coredump %s into place: %m", target);
+
+ return 0;
+}
+
+static int maybe_remove_external_coredump(const char *filename, uint64_t size) {
+
+ /* Returns 1 if might remove, 0 if will not remove, < 0 on error. */
+
+ if (arg_storage == COREDUMP_STORAGE_EXTERNAL &&
+ size <= arg_external_size_max)
+ return 0;
+
+ if (!filename)
+ return 1;
+
+ if (unlink(filename) < 0 && errno != ENOENT)
+ return log_error_errno(errno, "Failed to unlink %s: %m", filename);
+
+ return 1;
+}
+
+static int make_filename(const Context *context, char **ret) {
+ _cleanup_free_ char *c = NULL, *u = NULL, *p = NULL, *t = NULL;
+ sd_id128_t boot = {};
+ int r;
+
+ assert(context);
+
+ c = filename_escape(context->meta[META_COMM]);
+ if (!c)
+ return -ENOMEM;
+
+ u = filename_escape(context->meta[META_ARGV_UID]);
+ if (!u)
+ return -ENOMEM;
+
+ r = sd_id128_get_boot(&boot);
+ if (r < 0)
+ return r;
+
+ p = filename_escape(context->meta[META_ARGV_PID]);
+ if (!p)
+ return -ENOMEM;
+
+ t = filename_escape(context->meta[META_ARGV_TIMESTAMP]);
+ if (!t)
+ return -ENOMEM;
+
+ if (asprintf(ret,
+ "/var/lib/systemd/coredump/core.%s.%s." SD_ID128_FORMAT_STR ".%s.%s",
+ c,
+ u,
+ SD_ID128_FORMAT_VAL(boot),
+ p,
+ t) < 0)
+ return -ENOMEM;
+
+ return 0;
+}
+
+static int grant_user_access(int core_fd, const Context *context) {
+ int at_secure = -1;
+ uid_t uid = UID_INVALID, euid = UID_INVALID;
+ uid_t gid = GID_INVALID, egid = GID_INVALID;
+ int r;
+
+ assert(core_fd >= 0);
+ assert(context);
+
+ if (!context->meta[META_PROC_AUXV])
+ return log_warning_errno(SYNTHETIC_ERRNO(ENODATA), "No auxv data, not adjusting permissions.");
+
+ uint8_t elf[EI_NIDENT];
+ errno = 0;
+ if (pread(core_fd, &elf, sizeof(elf), 0) != sizeof(elf))
+ return log_warning_errno(errno_or_else(EIO),
+ "Failed to pread from coredump fd: %s", STRERROR_OR_EOF(errno));
+
+ if (elf[EI_MAG0] != ELFMAG0 ||
+ elf[EI_MAG1] != ELFMAG1 ||
+ elf[EI_MAG2] != ELFMAG2 ||
+ elf[EI_MAG3] != ELFMAG3 ||
+ elf[EI_VERSION] != EV_CURRENT)
+ return log_info_errno(SYNTHETIC_ERRNO(EUCLEAN),
+ "Core file does not have ELF header, not adjusting permissions.");
+ if (!IN_SET(elf[EI_CLASS], ELFCLASS32, ELFCLASS64) ||
+ !IN_SET(elf[EI_DATA], ELFDATA2LSB, ELFDATA2MSB))
+ return log_info_errno(SYNTHETIC_ERRNO(EUCLEAN),
+ "Core file has strange ELF class, not adjusting permissions.");
+
+ if ((elf[EI_DATA] == ELFDATA2LSB) != (__BYTE_ORDER == __LITTLE_ENDIAN))
+ return log_info_errno(SYNTHETIC_ERRNO(EUCLEAN),
+ "Core file has non-native endianness, not adjusting permissions.");
+
+ r = parse_auxv(LOG_WARNING,
+ /* elf_class= */ elf[EI_CLASS],
+ context->meta[META_PROC_AUXV],
+ context->meta_size[META_PROC_AUXV],
+ &at_secure, &uid, &euid, &gid, &egid);
+ if (r < 0)
+ return r;
+
+ /* We allow access if we got all the data and at_secure is not set and
+ * the uid/gid matches euid/egid. */
+ bool ret =
+ at_secure == 0 &&
+ uid != UID_INVALID && euid != UID_INVALID && uid == euid &&
+ gid != GID_INVALID && egid != GID_INVALID && gid == egid;
+ log_debug("Will %s access (uid="UID_FMT " euid="UID_FMT " gid="GID_FMT " egid="GID_FMT " at_secure=%s)",
+ ret ? "permit" : "restrict",
+ uid, euid, gid, egid, yes_no(at_secure));
+ return ret;
+}
+
+static int save_external_coredump(
+ const Context *context,
+ int input_fd,
+ char **ret_filename,
+ int *ret_node_fd,
+ int *ret_data_fd,
+ uint64_t *ret_size,
+ uint64_t *ret_compressed_size,
+ bool *ret_truncated) {
+
+ _cleanup_(unlink_and_freep) char *tmp = NULL;
+ _cleanup_free_ char *fn = NULL;
+ _cleanup_close_ int fd = -EBADF;
+ uint64_t rlimit, process_limit, max_size;
+ bool truncated, storage_on_tmpfs;
+ struct stat st;
+ int r;
+
+ assert(context);
+ assert(ret_filename);
+ assert(ret_node_fd);
+ assert(ret_data_fd);
+ assert(ret_size);
+ assert(ret_compressed_size);
+ assert(ret_truncated);
+
+ r = safe_atou64(context->meta[META_ARGV_RLIMIT], &rlimit);
+ if (r < 0)
+ return log_error_errno(r, "Failed to parse resource limit '%s': %m",
+ context->meta[META_ARGV_RLIMIT]);
+ if (rlimit < page_size())
+ /* Is coredumping disabled? Then don't bother saving/processing the
+ * coredump. Anything below PAGE_SIZE cannot give a readable coredump
+ * (the kernel uses ELF_EXEC_PAGESIZE which is not easily accessible, but
+ * is usually the same as PAGE_SIZE. */
+ return log_info_errno(SYNTHETIC_ERRNO(EBADSLT),
+ "Resource limits disable core dumping for process %s (%s).",
+ context->meta[META_ARGV_PID], context->meta[META_COMM]);
+
+ process_limit = MAX(arg_process_size_max, storage_size_max());
+ if (process_limit == 0)
+ return log_debug_errno(SYNTHETIC_ERRNO(EBADSLT),
+ "Limits for coredump processing and storage are both 0, not dumping core.");
+
+ /* Never store more than the process configured, or than we actually shall keep or process */
+ max_size = MIN(rlimit, process_limit);
+
+ r = make_filename(context, &fn);
+ if (r < 0)
+ return log_error_errno(r, "Failed to determine coredump file name: %m");
+
+ (void) mkdir_parents_label(fn, 0755);
+
+ fd = open_tmpfile_linkable(fn, O_RDWR|O_CLOEXEC, &tmp);
+ if (fd < 0)
+ return log_error_errno(fd, "Failed to create temporary file for coredump %s: %m", fn);
+
+ /* If storage is on tmpfs, the kernel oomd might kill us if there's MemoryMax set on
+ * the service or the slice it belongs to. This is common on low-resources systems,
+ * to avoid crashing processes to take away too many system resources.
+ * Check the cgroup settings, and set max_size to a bit less than half of the
+ * available memory left to the process.
+ * Then, attempt to write the core file uncompressed first - if the write gets
+ * interrupted, we know we won't be able to write it all, so instead compress what
+ * was written so far, delete the uncompressed truncated core, and then continue
+ * compressing from STDIN. Given the compressed core cannot be larger than the
+ * uncompressed one, and 1KB for metadata is accounted for in the calculation, we
+ * should be able to at least store the full compressed core file. */
+
+ storage_on_tmpfs = fd_is_temporary_fs(fd) > 0;
+ if (storage_on_tmpfs && arg_compress) {
+ _cleanup_(sd_bus_flush_close_unrefp) sd_bus *bus = NULL;
+ uint64_t cgroup_limit = UINT64_MAX;
+ struct statvfs sv;
+
+ /* If we can't get the cgroup limit, just ignore it, but don't fail,
+ * try anyway with the config settings. */
+ r = sd_bus_default_system(&bus);
+ if (r < 0)
+ log_info_errno(r, "Failed to connect to system bus, skipping MemoryAvailable check: %m");
+ else {
+ _cleanup_(sd_bus_error_free) sd_bus_error error = SD_BUS_ERROR_NULL;
+
+ r = sd_bus_get_property_trivial(
+ bus,
+ "org.freedesktop.systemd1",
+ "/org/freedesktop/systemd1/unit/self",
+ "org.freedesktop.systemd1.Service",
+ "MemoryAvailable",
+ &error,
+ 't', &cgroup_limit);
+ if (r < 0)
+ log_warning_errno(r,
+ "Failed to query MemoryAvailable for current unit, "
+ "falling back to static config settings: %s",
+ bus_error_message(&error, r));
+ }
+
+ max_size = MIN(cgroup_limit, max_size);
+ max_size = LESS_BY(max_size, 1024U) / 2; /* Account for 1KB metadata overhead for compressing */
+ max_size = MAX(PROCESS_SIZE_MIN, max_size); /* Impose a lower minimum */
+
+ /* tmpfs might get full quickly, so check the available space too.
+ * But don't worry about errors here, failing to access the storage
+ * location will be better logged when writing to it. */
+ if (fstatvfs(fd, &sv) >= 0)
+ max_size = MIN((uint64_t)sv.f_frsize * (uint64_t)sv.f_bfree, max_size);
+
+ log_debug("Limiting core file size to %" PRIu64 " bytes due to cgroup memory limits.", max_size);
+ }
+
+ r = copy_bytes(input_fd, fd, max_size, 0);
+ if (r < 0)
+ return log_error_errno(r, "Cannot store coredump of %s (%s): %m",
+ context->meta[META_ARGV_PID], context->meta[META_COMM]);
+ truncated = r == 1;
+
+ bool allow_user = grant_user_access(fd, context) > 0;
+
+#if HAVE_COMPRESSION
+ if (arg_compress) {
+ _cleanup_(unlink_and_freep) char *tmp_compressed = NULL;
+ _cleanup_free_ char *fn_compressed = NULL;
+ _cleanup_close_ int fd_compressed = -EBADF;
+ uint64_t uncompressed_size = 0;
+
+ if (lseek(fd, 0, SEEK_SET) < 0)
+ return log_error_errno(errno, "Failed to seek on coredump %s: %m", fn);
+
+ fn_compressed = strjoin(fn, default_compression_extension());
+ if (!fn_compressed)
+ return log_oom();
+
+ fd_compressed = open_tmpfile_linkable(fn_compressed, O_RDWR|O_CLOEXEC, &tmp_compressed);
+ if (fd_compressed < 0)
+ return log_error_errno(fd_compressed, "Failed to create temporary file for coredump %s: %m", fn_compressed);
+
+ r = compress_stream(fd, fd_compressed, max_size, &uncompressed_size);
+ if (r < 0)
+ return log_error_errno(r, "Failed to compress %s: %m", coredump_tmpfile_name(tmp_compressed));
+
+ if (truncated && storage_on_tmpfs) {
+ uint64_t partial_uncompressed_size = 0;
+
+ /* Uncompressed write was truncated and we are writing to tmpfs: delete
+ * the uncompressed core, and compress the remaining part from STDIN. */
+
+ tmp = unlink_and_free(tmp);
+ fd = safe_close(fd);
+
+ r = compress_stream(input_fd, fd_compressed, max_size, &partial_uncompressed_size);
+ if (r < 0)
+ return log_error_errno(r, "Failed to compress %s: %m", coredump_tmpfile_name(tmp_compressed));
+ uncompressed_size += partial_uncompressed_size;
+ }
+
+ r = fix_permissions(fd_compressed, tmp_compressed, fn_compressed, context, allow_user);
+ if (r < 0)
+ return r;
+
+ if (fstat(fd_compressed, &st) < 0)
+ return log_error_errno(errno,
+ "Failed to fstat core file %s: %m",
+ coredump_tmpfile_name(tmp_compressed));
+
+ *ret_filename = TAKE_PTR(fn_compressed); /* compressed */
+ *ret_node_fd = TAKE_FD(fd_compressed); /* compressed */
+ *ret_compressed_size = (uint64_t) st.st_size; /* compressed */
+ *ret_data_fd = TAKE_FD(fd);
+ *ret_size = uncompressed_size;
+ *ret_truncated = truncated;
+ tmp_compressed = mfree(tmp_compressed);
+
+ return 0;
+ }
+#endif
+
+ if (truncated)
+ log_struct(LOG_INFO,
+ LOG_MESSAGE("Core file was truncated to %"PRIu64" bytes.", max_size),
+ "SIZE_LIMIT=%"PRIu64, max_size,
+ "MESSAGE_ID=" SD_MESSAGE_TRUNCATED_CORE_STR);
+
+ r = fix_permissions(fd, tmp, fn, context, allow_user);
+ if (r < 0)
+ return log_error_errno(r, "Failed to fix permissions and finalize coredump %s into %s: %m", coredump_tmpfile_name(tmp), fn);
+
+ if (fstat(fd, &st) < 0)
+ return log_error_errno(errno, "Failed to fstat core file %s: %m", coredump_tmpfile_name(tmp));
+
+ if (lseek(fd, 0, SEEK_SET) < 0)
+ return log_error_errno(errno, "Failed to seek on coredump %s: %m", fn);
+
+ *ret_filename = TAKE_PTR(fn);
+ *ret_data_fd = TAKE_FD(fd);
+ *ret_size = (uint64_t) st.st_size;
+ *ret_truncated = truncated;
+
+ return 0;
+}
+
+static int allocate_journal_field(int fd, size_t size, char **ret, size_t *ret_size) {
+ _cleanup_free_ char *field = NULL;
+ ssize_t n;
+
+ assert(fd >= 0);
+ assert(ret);
+ assert(ret_size);
+
+ if (lseek(fd, 0, SEEK_SET) < 0)
+ return log_warning_errno(errno, "Failed to seek: %m");
+
+ field = malloc(9 + size);
+ if (!field)
+ return log_warning_errno(SYNTHETIC_ERRNO(ENOMEM),
+ "Failed to allocate memory for coredump, coredump will not be stored.");
+
+ memcpy(field, "COREDUMP=", 9);
+
+ /* NB: simple read() would fail for overly large coredumps, since read() on Linux can only deal with
+ * 0x7ffff000 bytes max. Hence call things in a loop. */
+ n = loop_read(fd, field + 9, size, /* do_poll= */ false);
+ if (n < 0)
+ return log_error_errno((int) n, "Failed to read core data: %m");
+ if ((size_t) n < size)
+ return log_error_errno(SYNTHETIC_ERRNO(EIO),
+ "Core data too short.");
+
+ *ret = TAKE_PTR(field);
+ *ret_size = size + 9;
+
+ return 0;
+}
+
+/* Joins /proc/[pid]/fd/ and /proc/[pid]/fdinfo/ into the following lines:
+ * 0:/dev/pts/23
+ * pos: 0
+ * flags: 0100002
+ *
+ * 1:/dev/pts/23
+ * pos: 0
+ * flags: 0100002
+ *
+ * 2:/dev/pts/23
+ * pos: 0
+ * flags: 0100002
+ * EOF
+ */
+static int compose_open_fds(pid_t pid, char **ret) {
+ _cleanup_(memstream_done) MemStream m = {};
+ _cleanup_closedir_ DIR *proc_fd_dir = NULL;
+ _cleanup_close_ int proc_fdinfo_fd = -EBADF;
+ const char *fddelim = "", *path;
+ FILE *stream;
+ int r;
+
+ assert(pid >= 0);
+ assert(ret);
+
+ path = procfs_file_alloca(pid, "fd");
+ proc_fd_dir = opendir(path);
+ if (!proc_fd_dir)
+ return -errno;
+
+ proc_fdinfo_fd = openat(dirfd(proc_fd_dir), "../fdinfo", O_DIRECTORY|O_NOFOLLOW|O_CLOEXEC|O_PATH);
+ if (proc_fdinfo_fd < 0)
+ return -errno;
+
+ stream = memstream_init(&m);
+ if (!stream)
+ return -ENOMEM;
+
+ FOREACH_DIRENT(de, proc_fd_dir, return -errno) {
+ _cleanup_fclose_ FILE *fdinfo = NULL;
+ _cleanup_free_ char *fdname = NULL;
+ _cleanup_close_ int fd = -EBADF;
+
+ r = readlinkat_malloc(dirfd(proc_fd_dir), de->d_name, &fdname);
+ if (r < 0)
+ return r;
+
+ fprintf(stream, "%s%s:%s\n", fddelim, de->d_name, fdname);
+ fddelim = "\n";
+
+ /* Use the directory entry from /proc/[pid]/fd with /proc/[pid]/fdinfo */
+ fd = openat(proc_fdinfo_fd, de->d_name, O_NOFOLLOW|O_CLOEXEC|O_RDONLY);
+ if (fd < 0)
+ continue;
+
+ fdinfo = take_fdopen(&fd, "r");
+ if (!fdinfo)
+ continue;
+
+ for (;;) {
+ _cleanup_free_ char *line = NULL;
+
+ r = read_line(fdinfo, LONG_LINE_MAX, &line);
+ if (r < 0)
+ return r;
+ if (r == 0)
+ break;
+
+ fputs(line, stream);
+ fputc('\n', stream);
+ }
+ }
+
+ return memstream_finalize(&m, ret, NULL);
+}
+
+/* Returns 1 if the parent was found.
+ * Returns 0 if there is not a process we can call the pid's
+ * container parent (the pid's process isn't 'containerized').
+ * Returns a negative number on errors.
+ */
+static int get_process_container_parent_cmdline(pid_t pid, char** cmdline) {
+ pid_t container_pid;
+ const char *proc_root_path;
+ struct stat root_stat, proc_root_stat;
+ int r;
+
+ /* To compare inodes of / and /proc/[pid]/root */
+ if (stat("/", &root_stat) < 0)
+ return -errno;
+
+ proc_root_path = procfs_file_alloca(pid, "root");
+ if (stat(proc_root_path, &proc_root_stat) < 0)
+ return -errno;
+
+ /* The process uses system root. */
+ if (stat_inode_same(&proc_root_stat, &root_stat)) {
+ *cmdline = NULL;
+ return 0;
+ }
+
+ r = namespace_get_leader(pid, NAMESPACE_MOUNT, &container_pid);
+ if (r < 0)
+ return r;
+
+ r = pid_get_cmdline(container_pid, SIZE_MAX, PROCESS_CMDLINE_QUOTE_POSIX, cmdline);
+ if (r < 0)
+ return r;
+
+ return 1;
+}
+
+static int change_uid_gid(const Context *context) {
+ uid_t uid = context->uid;
+ gid_t gid = context->gid;
+ int r;
+
+ if (uid_is_system(uid)) {
+ const char *user = "systemd-coredump";
+
+ r = get_user_creds(&user, &uid, &gid, NULL, NULL, 0);
+ if (r < 0) {
+ log_warning_errno(r, "Cannot resolve %s user. Proceeding to dump core as root: %m", user);
+ uid = gid = 0;
+ }
+ }
+
+ return drop_privileges(uid, gid, 0);
+}
+
+static int submit_coredump(
+ const Context *context,
+ struct iovec_wrapper *iovw,
+ int input_fd) {
+
+ _cleanup_(json_variant_unrefp) JsonVariant *json_metadata = NULL;
+ _cleanup_close_ int coredump_fd = -EBADF, coredump_node_fd = -EBADF;
+ _cleanup_free_ char *filename = NULL, *coredump_data = NULL;
+ _cleanup_free_ char *stacktrace = NULL;
+ const char *module_name;
+ uint64_t coredump_size = UINT64_MAX, coredump_compressed_size = UINT64_MAX;
+ bool truncated = false, written = false;
+ JsonVariant *module_json;
+ int r;
+
+ assert(context);
+ assert(iovw);
+ assert(input_fd >= 0);
+
+ /* Vacuum before we write anything again */
+ (void) coredump_vacuum(-1, arg_keep_free, arg_max_use);
+
+ /* Always stream the coredump to disk, if that's possible */
+ written = save_external_coredump(
+ context, input_fd,
+ &filename, &coredump_node_fd, &coredump_fd,
+ &coredump_size, &coredump_compressed_size, &truncated) >= 0;
+ if (written) {
+ /* If we could write it to disk we can now process it. */
+ /* If we don't want to keep the coredump on disk, remove it now, as later on we
+ * will lack the privileges for it. However, we keep the fd to it, so that we can
+ * still process it and log it. */
+ r = maybe_remove_external_coredump(filename, coredump_node_fd >= 0 ? coredump_compressed_size : coredump_size);
+ if (r < 0)
+ return r;
+ if (r == 0)
+ (void) iovw_put_string_field(iovw, "COREDUMP_FILENAME=", filename);
+ else if (arg_storage == COREDUMP_STORAGE_EXTERNAL)
+ log_info("The core will not be stored: size %"PRIu64" is greater than %"PRIu64" (the configured maximum)",
+ coredump_node_fd >= 0 ? coredump_compressed_size : coredump_size, arg_external_size_max);
+
+ /* Vacuum again, but exclude the coredump we just created */
+ (void) coredump_vacuum(coredump_node_fd >= 0 ? coredump_node_fd : coredump_fd, arg_keep_free, arg_max_use);
+ }
+
+ /* Now, let's drop privileges to become the user who owns the segfaulted process and allocate the
+ * coredump memory under the user's uid. This also ensures that the credentials journald will see are
+ * the ones of the coredumping user, thus making sure the user gets access to the core dump. Let's
+ * also get rid of all capabilities, if we run as root, we won't need them anymore. */
+ r = change_uid_gid(context);
+ if (r < 0)
+ return log_error_errno(r, "Failed to drop privileges: %m");
+
+ if (written) {
+ /* Try to get a stack trace if we can */
+ if (coredump_size > arg_process_size_max)
+ log_debug("Not generating stack trace: core size %"PRIu64" is greater "
+ "than %"PRIu64" (the configured maximum)",
+ coredump_size, arg_process_size_max);
+ else if (coredump_fd >= 0) {
+ bool skip = startswith(context->meta[META_COMM], "systemd-coredum"); /* COMM is 16 bytes usually */
+
+ (void) parse_elf_object(coredump_fd,
+ context->meta[META_EXE],
+ /* fork_disable_dump= */ skip, /* avoid loops */
+ &stacktrace,
+ &json_metadata);
+ }
+ }
+
+ _cleanup_free_ char *core_message = NULL;
+ core_message = strjoin(
+ "Process ", context->meta[META_ARGV_PID],
+ " (", context->meta[META_COMM],
+ ") of user ", context->meta[META_ARGV_UID],
+ written ? " dumped core." : " terminated abnormally without generating a coredump.");
+ if (!core_message)
+ return log_oom();
+
+ if (context->is_journald && filename)
+ if (!strextend(&core_message, "\nCoredump diverted to ", filename))
+ return log_oom();
+
+ if (stacktrace)
+ if (!strextend(&core_message, "\n\n", stacktrace))
+ return log_oom();
+
+ if (context->is_journald)
+ /* We might not be able to log to the journal, so let's always print the message to another
+ * log target. The target was set previously to something safe. */
+ log_dispatch(LOG_ERR, 0, core_message);
+
+ (void) iovw_put_string_field(iovw, "MESSAGE=", core_message);
+
+ if (truncated)
+ (void) iovw_put_string_field(iovw, "COREDUMP_TRUNCATED=", "1");
+
+ /* If we managed to parse any ELF metadata (build-id, ELF package meta),
+ * attach it as journal metadata. */
+ if (json_metadata) {
+ _cleanup_free_ char *formatted_json = NULL;
+
+ r = json_variant_format(json_metadata, 0, &formatted_json);
+ if (r < 0)
+ return log_error_errno(r, "Failed to format JSON package metadata: %m");
+
+ (void) iovw_put_string_field(iovw, "COREDUMP_PACKAGE_JSON=", formatted_json);
+ }
+
+ /* In the unlikely scenario that context->meta[META_EXE] is not available,
+ * let's avoid guessing the module name and skip the loop. */
+ if (context->meta[META_EXE])
+ JSON_VARIANT_OBJECT_FOREACH(module_name, module_json, json_metadata) {
+ JsonVariant *t;
+
+ /* We only add structured fields for the 'main' ELF module, and only if we can identify it. */
+ if (!path_equal_filename(module_name, context->meta[META_EXE]))
+ continue;
+
+ t = json_variant_by_key(module_json, "name");
+ if (t)
+ (void) iovw_put_string_field(iovw, "COREDUMP_PACKAGE_NAME=", json_variant_string(t));
+
+ t = json_variant_by_key(module_json, "version");
+ if (t)
+ (void) iovw_put_string_field(iovw, "COREDUMP_PACKAGE_VERSION=", json_variant_string(t));
+ }
+
+ /* Optionally store the entire coredump in the journal */
+ if (arg_storage == COREDUMP_STORAGE_JOURNAL && coredump_fd >= 0) {
+ if (coredump_size <= arg_journal_size_max) {
+ size_t sz = 0;
+
+ /* Store the coredump itself in the journal */
+
+ r = allocate_journal_field(coredump_fd, (size_t) coredump_size, &coredump_data, &sz);
+ if (r >= 0) {
+ if (iovw_put(iovw, coredump_data, sz) >= 0)
+ TAKE_PTR(coredump_data);
+ } else
+ log_warning_errno(r, "Failed to attach the core to the journal entry: %m");
+ } else
+ log_info("The core will not be stored: size %"PRIu64" is greater than %"PRIu64" (the configured maximum)",
+ coredump_size, arg_journal_size_max);
+ }
+
+ /* If journald is coredumping, we have to be careful that we don't deadlock when trying to write the
+ * coredump to the journal, so we put the journal socket in nonblocking mode before trying to write
+ * the coredump to the socket. */
+
+ if (context->is_journald) {
+ r = journal_fd_nonblock(true);
+ if (r < 0)
+ return log_error_errno(r, "Failed to make journal socket non-blocking: %m");
+ }
+
+ r = sd_journal_sendv(iovw->iovec, iovw->count);
+
+ if (context->is_journald) {
+ int k;
+
+ k = journal_fd_nonblock(false);
+ if (k < 0)
+ return log_error_errno(k, "Failed to make journal socket blocking: %m");
+ }
+
+ if (r == -EAGAIN && context->is_journald)
+ log_warning_errno(r, "Failed to log journal coredump, ignoring: %m");
+ else if (r < 0)
+ return log_error_errno(r, "Failed to log coredump: %m");
+
+ return 0;
+}
+
+static int save_context(Context *context, const struct iovec_wrapper *iovw) {
+ const char *unit;
+ int r;
+
+ assert(context);
+ assert(iovw);
+ assert(iovw->count >= _META_ARGV_MAX);
+
+ /* The context does not allocate any memory on its own */
+
+ for (size_t n = 0; n < iovw->count; n++) {
+ struct iovec *iovec = iovw->iovec + n;
+
+ for (size_t i = 0; i < ELEMENTSOF(meta_field_names); i++) {
+ /* Note that these strings are NUL terminated, because we made sure that a
+ * trailing NUL byte is in the buffer, though not included in the iov_len
+ * count (see process_socket() and gather_pid_metadata_*()) */
+ assert(((char*) iovec->iov_base)[iovec->iov_len] == 0);
+
+ const char *p = startswith(iovec->iov_base, meta_field_names[i]);
+ if (p) {
+ context->meta[i] = p;
+ context->meta_size[i] = iovec->iov_len - strlen(meta_field_names[i]);
+ break;
+ }
+ }
+ }
+
+ if (!context->meta[META_ARGV_PID])
+ return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
+ "Failed to find the PID of crashing process");
+
+ r = parse_pid(context->meta[META_ARGV_PID], &context->pid);
+ if (r < 0)
+ return log_error_errno(r, "Failed to parse PID \"%s\": %m", context->meta[META_ARGV_PID]);
+
+ r = parse_uid(context->meta[META_ARGV_UID], &context->uid);
+ if (r < 0)
+ return log_error_errno(r, "Failed to parse UID \"%s\": %m", context->meta[META_ARGV_UID]);
+
+ r = parse_gid(context->meta[META_ARGV_GID], &context->gid);
+ if (r < 0)
+ return log_error_errno(r, "Failed to parse GID \"%s\": %m", context->meta[META_ARGV_GID]);
+
+ unit = context->meta[META_UNIT];
+ context->is_pid1 = streq(context->meta[META_ARGV_PID], "1") || streq_ptr(unit, SPECIAL_INIT_SCOPE);
+ context->is_journald = streq_ptr(unit, SPECIAL_JOURNALD_SERVICE);
+
+ return 0;
+}
+
+static int process_socket(int fd) {
+ _cleanup_close_ int input_fd = -EBADF;
+ Context context = {};
+ struct iovec_wrapper iovw = {};
+ struct iovec iovec;
+ int r;
+
+ assert(fd >= 0);
+
+ log_setup();
+
+ log_debug("Processing coredump received on stdin...");
+
+ for (;;) {
+ CMSG_BUFFER_TYPE(CMSG_SPACE(sizeof(int))) control;
+ struct msghdr mh = {
+ .msg_control = &control,
+ .msg_controllen = sizeof(control),
+ .msg_iovlen = 1,
+ };
+ ssize_t n;
+ ssize_t l;
+
+ l = next_datagram_size_fd(fd);
+ if (l < 0) {
+ r = log_error_errno(l, "Failed to determine datagram size to read: %m");
+ goto finish;
+ }
+
+ iovec.iov_len = l;
+ iovec.iov_base = malloc(l + 1);
+ if (!iovec.iov_base) {
+ r = log_oom();
+ goto finish;
+ }
+
+ mh.msg_iov = &iovec;
+
+ n = recvmsg_safe(fd, &mh, MSG_CMSG_CLOEXEC);
+ if (n < 0) {
+ free(iovec.iov_base);
+ r = log_error_errno(n, "Failed to receive datagram: %m");
+ goto finish;
+ }
+
+ /* The final zero-length datagram carries the file descriptor and tells us
+ * that we're done. */
+ if (n == 0) {
+ struct cmsghdr *found;
+
+ free(iovec.iov_base);
+
+ found = cmsg_find(&mh, SOL_SOCKET, SCM_RIGHTS, CMSG_LEN(sizeof(int)));
+ if (!found) {
+ cmsg_close_all(&mh);
+ r = log_error_errno(SYNTHETIC_ERRNO(EBADMSG),
+ "Coredump file descriptor missing.");
+ goto finish;
+ }
+
+ assert(input_fd < 0);
+ input_fd = *CMSG_TYPED_DATA(found, int);
+ break;
+ } else
+ cmsg_close_all(&mh);
+
+ /* Add trailing NUL byte, in case these are strings */
+ ((char*) iovec.iov_base)[n] = 0;
+ iovec.iov_len = (size_t) n;
+
+ r = iovw_put(&iovw, iovec.iov_base, iovec.iov_len);
+ if (r < 0)
+ goto finish;
+ }
+
+ /* Make sure we got all data we really need */
+ assert(input_fd >= 0);
+
+ r = save_context(&context, &iovw);
+ if (r < 0)
+ goto finish;
+
+ /* Make sure we received at least all fields we need. */
+ for (int i = 0; i < _META_MANDATORY_MAX; i++)
+ if (!context.meta[i]) {
+ r = log_error_errno(SYNTHETIC_ERRNO(EINVAL),
+ "A mandatory argument (%i) has not been sent, aborting.",
+ i);
+ goto finish;
+ }
+
+ r = submit_coredump(&context, &iovw, input_fd);
+
+finish:
+ iovw_free_contents(&iovw, true);
+ return r;
+}
+
+static int send_iovec(const struct iovec_wrapper *iovw, int input_fd) {
+ _cleanup_close_ int fd = -EBADF;
+ int r;
+
+ assert(iovw);
+ assert(input_fd >= 0);
+
+ fd = socket(AF_UNIX, SOCK_SEQPACKET|SOCK_CLOEXEC, 0);
+ if (fd < 0)
+ return log_error_errno(errno, "Failed to create coredump socket: %m");
+
+ r = connect_unix_path(fd, AT_FDCWD, "/run/systemd/coredump");
+ if (r < 0)
+ return log_error_errno(r, "Failed to connect to coredump service: %m");
+
+ for (size_t i = 0; i < iovw->count; i++) {
+ struct msghdr mh = {
+ .msg_iov = iovw->iovec + i,
+ .msg_iovlen = 1,
+ };
+ struct iovec copy[2];
+
+ for (;;) {
+ if (sendmsg(fd, &mh, MSG_NOSIGNAL) >= 0)
+ break;
+
+ if (errno == EMSGSIZE && mh.msg_iov[0].iov_len > 0) {
+ /* This field didn't fit? That's a pity. Given that this is
+ * just metadata, let's truncate the field at half, and try
+ * again. We append three dots, in order to show that this is
+ * truncated. */
+
+ if (mh.msg_iov != copy) {
+ /* We don't want to modify the caller's iovec, hence
+ * let's create our own array, consisting of two new
+ * iovecs, where the first is a (truncated) copy of
+ * what we want to send, and the second one contains
+ * the trailing dots. */
+ copy[0] = iovw->iovec[i];
+ copy[1] = IOVEC_MAKE(((char[]){'.', '.', '.'}), 3);
+
+ mh.msg_iov = copy;
+ mh.msg_iovlen = 2;
+ }
+
+ copy[0].iov_len /= 2; /* halve it, and try again */
+ continue;
+ }
+
+ return log_error_errno(errno, "Failed to send coredump datagram: %m");
+ }
+ }
+
+ r = send_one_fd(fd, input_fd, 0);
+ if (r < 0)
+ return log_error_errno(r, "Failed to send coredump fd: %m");
+
+ return 0;
+}
+
+static int gather_pid_metadata_from_argv(
+ struct iovec_wrapper *iovw,
+ Context *context,
+ int argc, char **argv) {
+
+ _cleanup_free_ char *free_timestamp = NULL;
+ int r, signo;
+ char *t;
+
+ assert(iovw);
+ assert(context);
+
+ /* We gather all metadata that were passed via argv[] into an array of iovecs that
+ * we'll forward to the socket unit */
+
+ if (argc < _META_ARGV_MAX)
+ return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
+ "Not enough arguments passed by the kernel (%i, expected %i).",
+ argc, _META_ARGV_MAX);
+
+ for (int i = 0; i < _META_ARGV_MAX; i++) {
+
+ t = argv[i];
+
+ switch (i) {
+
+ case META_ARGV_TIMESTAMP:
+ /* The journal fields contain the timestamp padded with six
+ * zeroes, so that the kernel-supplied 1s granularity timestamps
+ * becomes 1μs granularity, i.e. the granularity systemd usually
+ * operates in. */
+ t = free_timestamp = strjoin(argv[i], "000000");
+ if (!t)
+ return log_oom();
+ break;
+
+ case META_ARGV_SIGNAL:
+ /* For signal, record its pretty name too */
+ if (safe_atoi(argv[i], &signo) >= 0 && SIGNAL_VALID(signo))
+ (void) iovw_put_string_field(iovw, "COREDUMP_SIGNAL_NAME=SIG",
+ signal_to_string(signo));
+ break;
+
+ default:
+ break;
+ }
+
+ r = iovw_put_string_field(iovw, meta_field_names[i], t);
+ if (r < 0)
+ return r;
+ }
+
+ /* Cache some of the process metadata we collected so far and that we'll need to
+ * access soon */
+ return save_context(context, iovw);
+}
+
+static int gather_pid_metadata_from_procfs(struct iovec_wrapper *iovw, Context *context) {
+ uid_t owner_uid;
+ pid_t pid;
+ char *t;
+ size_t size;
+ const char *p;
+ int r;
+
+ assert(iovw);
+ assert(context);
+
+ /* Note that if we fail on oom later on, we do not roll-back changes to the iovec
+ * structure. (It remains valid, with the first iovec fields initialized.) */
+
+ pid = context->pid;
+
+ /* The following is mandatory */
+ r = pid_get_comm(pid, &t);
+ if (r < 0)
+ return log_error_errno(r, "Failed to get COMM: %m");
+
+ r = iovw_put_string_field_free(iovw, "COREDUMP_COMM=", t);
+ if (r < 0)
+ return r;
+
+ /* The following are optional, but we use them if present. */
+ r = get_process_exe(pid, &t);
+ if (r >= 0)
+ r = iovw_put_string_field_free(iovw, "COREDUMP_EXE=", t);
+ if (r < 0)
+ log_warning_errno(r, "Failed to get EXE, ignoring: %m");
+
+ if (cg_pid_get_unit(pid, &t) >= 0)
+ (void) iovw_put_string_field_free(iovw, "COREDUMP_UNIT=", t);
+
+ if (cg_pid_get_user_unit(pid, &t) >= 0)
+ (void) iovw_put_string_field_free(iovw, "COREDUMP_USER_UNIT=", t);
+
+ if (sd_pid_get_session(pid, &t) >= 0)
+ (void) iovw_put_string_field_free(iovw, "COREDUMP_SESSION=", t);
+
+ if (sd_pid_get_owner_uid(pid, &owner_uid) >= 0) {
+ r = asprintf(&t, UID_FMT, owner_uid);
+ if (r > 0)
+ (void) iovw_put_string_field_free(iovw, "COREDUMP_OWNER_UID=", t);
+ }
+
+ if (sd_pid_get_slice(pid, &t) >= 0)
+ (void) iovw_put_string_field_free(iovw, "COREDUMP_SLICE=", t);
+
+ if (pid_get_cmdline(pid, SIZE_MAX, PROCESS_CMDLINE_QUOTE_POSIX, &t) >= 0)
+ (void) iovw_put_string_field_free(iovw, "COREDUMP_CMDLINE=", t);
+
+ if (cg_pid_get_path_shifted(pid, NULL, &t) >= 0)
+ (void) iovw_put_string_field_free(iovw, "COREDUMP_CGROUP=", t);
+
+ if (compose_open_fds(pid, &t) >= 0)
+ (void) iovw_put_string_field_free(iovw, "COREDUMP_OPEN_FDS=", t);
+
+ p = procfs_file_alloca(pid, "status");
+ if (read_full_virtual_file(p, &t, NULL) >= 0)
+ (void) iovw_put_string_field_free(iovw, "COREDUMP_PROC_STATUS=", t);
+
+ p = procfs_file_alloca(pid, "maps");
+ if (read_full_virtual_file(p, &t, NULL) >= 0)
+ (void) iovw_put_string_field_free(iovw, "COREDUMP_PROC_MAPS=", t);
+
+ p = procfs_file_alloca(pid, "limits");
+ if (read_full_virtual_file(p, &t, NULL) >= 0)
+ (void) iovw_put_string_field_free(iovw, "COREDUMP_PROC_LIMITS=", t);
+
+ p = procfs_file_alloca(pid, "cgroup");
+ if (read_full_virtual_file(p, &t, NULL) >= 0)
+ (void) iovw_put_string_field_free(iovw, "COREDUMP_PROC_CGROUP=", t);
+
+ p = procfs_file_alloca(pid, "mountinfo");
+ if (read_full_virtual_file(p, &t, NULL) >= 0)
+ (void) iovw_put_string_field_free(iovw, "COREDUMP_PROC_MOUNTINFO=", t);
+
+ /* We attach /proc/auxv here. ELF coredumps also contain a note for this (NT_AUXV), see elf(5). */
+ p = procfs_file_alloca(pid, "auxv");
+ if (read_full_virtual_file(p, &t, &size) >= 0) {
+ char *buf = malloc(strlen("COREDUMP_PROC_AUXV=") + size + 1);
+ if (buf) {
+ /* Add a dummy terminator to make save_context() happy. */
+ *((uint8_t*) mempcpy(stpcpy(buf, "COREDUMP_PROC_AUXV="), t, size)) = '\0';
+ (void) iovw_consume(iovw, buf, size + strlen("COREDUMP_PROC_AUXV="));
+ }
+
+ free(t);
+ }
+
+ if (get_process_cwd(pid, &t) >= 0)
+ (void) iovw_put_string_field_free(iovw, "COREDUMP_CWD=", t);
+
+ if (get_process_root(pid, &t) >= 0) {
+ bool proc_self_root_is_slash;
+
+ proc_self_root_is_slash = strcmp(t, "/") == 0;
+
+ (void) iovw_put_string_field_free(iovw, "COREDUMP_ROOT=", t);
+
+ /* If the process' root is "/", then there is a chance it has
+ * mounted own root and hence being containerized. */
+ if (proc_self_root_is_slash && get_process_container_parent_cmdline(pid, &t) > 0)
+ (void) iovw_put_string_field_free(iovw, "COREDUMP_CONTAINER_CMDLINE=", t);
+ }
+
+ if (get_process_environ(pid, &t) >= 0)
+ (void) iovw_put_string_field_free(iovw, "COREDUMP_ENVIRON=", t);
+
+ /* we successfully acquired all metadata */
+ return save_context(context, iovw);
+}
+
+static int send_ucred(int transport_fd, struct ucred *ucred) {
+ CMSG_BUFFER_TYPE(CMSG_SPACE(sizeof(struct ucred))) control = {};
+ struct msghdr mh = {
+ .msg_control = &control,
+ .msg_controllen = sizeof(control),
+ };
+ struct cmsghdr *cmsg;
+
+ assert(transport_fd >= 0);
+
+ cmsg = CMSG_FIRSTHDR(&mh);
+ *cmsg = (struct cmsghdr) {
+ .cmsg_level = SOL_SOCKET,
+ .cmsg_type = SCM_CREDENTIALS,
+ .cmsg_len = CMSG_LEN(sizeof(struct ucred)),
+ };
+ memcpy(CMSG_DATA(cmsg), ucred, sizeof(struct ucred));
+
+ return RET_NERRNO(sendmsg(transport_fd, &mh, MSG_NOSIGNAL));
+}
+
+static int receive_ucred(int transport_fd, struct ucred *ret_ucred) {
+ CMSG_BUFFER_TYPE(CMSG_SPACE(sizeof(struct ucred))) control = {};
+ struct msghdr mh = {
+ .msg_control = &control,
+ .msg_controllen = sizeof(control),
+ };
+ struct cmsghdr *cmsg = NULL;
+ struct ucred *ucred = NULL;
+ ssize_t n;
+
+ assert(ret_ucred);
+
+ n = recvmsg_safe(transport_fd, &mh, 0);
+ if (n < 0)
+ return n;
+
+ CMSG_FOREACH(cmsg, &mh)
+ if (cmsg->cmsg_level == SOL_SOCKET &&
+ cmsg->cmsg_type == SCM_CREDENTIALS &&
+ cmsg->cmsg_len == CMSG_LEN(sizeof(struct ucred))) {
+
+ assert(!ucred);
+ ucred = CMSG_TYPED_DATA(cmsg, struct ucred);
+ }
+
+ if (!ucred)
+ return -EIO;
+
+ *ret_ucred = *ucred;
+
+ return 0;
+}
+
+static int can_forward_coredump(pid_t pid) {
+ _cleanup_free_ char *cgroup = NULL, *path = NULL, *unit = NULL;
+ int r;
+
+ r = cg_pid_get_path(SYSTEMD_CGROUP_CONTROLLER, pid, &cgroup);
+ if (r < 0)
+ return r;
+
+ r = path_extract_directory(cgroup, &path);
+ if (r < 0)
+ return r;
+
+ r = cg_path_get_unit_path(path, &unit);
+ if (r == -ENOMEM)
+ return log_oom();
+ if (r == -ENXIO)
+ /* No valid units in this path. */
+ return false;
+ if (r < 0)
+ return r;
+
+ /* We require that this process belongs to a delegated cgroup
+ * (i.e. Delegate=yes), with CoredumpReceive=yes also. */
+ r = cg_is_delegated(unit);
+ if (r <= 0)
+ return r;
+
+ return cg_has_coredump_receive(unit);
+}
+
+static int forward_coredump_to_container(Context *context) {
+ _cleanup_close_ int pidnsfd = -EBADF, mntnsfd = -EBADF, netnsfd = -EBADF, usernsfd = -EBADF, rootfd = -EBADF;
+ _cleanup_close_pair_ int pair[2] = EBADF_PAIR;
+ pid_t pid, child;
+ struct ucred ucred = {
+ .pid = context->pid,
+ .uid = context->uid,
+ .gid = context->gid,
+ };
+ int r;
+
+ r = namespace_get_leader(context->pid, NAMESPACE_PID, &pid);
+ if (r < 0)
+ return log_debug_errno(r, "Failed to get namespace leader: %m");
+
+ r = can_forward_coredump(pid);
+ if (r < 0)
+ return log_debug_errno(r, "Failed to check if coredump can be forwarded: %m");
+ if (r == 0)
+ return log_debug_errno(SYNTHETIC_ERRNO(ENOENT),
+ "Coredump will not be forwarded because no target cgroup was found.");
+
+ r = RET_NERRNO(socketpair(AF_UNIX, SOCK_DGRAM|SOCK_CLOEXEC, 0, pair));
+ if (r < 0)
+ return log_debug_errno(r, "Failed to create socket pair: %m");
+
+ r = setsockopt_int(pair[1], SOL_SOCKET, SO_PASSCRED, true);
+ if (r < 0)
+ return log_debug_errno(r, "Failed to set SO_PASSCRED: %m");
+
+ r = namespace_open(pid, &pidnsfd, &mntnsfd, &netnsfd, &usernsfd, &rootfd);
+ if (r < 0)
+ return log_debug_errno(r, "Failed to join namespaces of PID " PID_FMT ": %m", pid);
+
+ r = namespace_fork("(sd-coredumpns)", "(sd-coredump)", NULL, 0,
+ FORK_RESET_SIGNALS|FORK_DEATHSIG_SIGTERM,
+ pidnsfd, mntnsfd, netnsfd, usernsfd, rootfd, &child);
+ if (r < 0)
+ return log_debug_errno(r, "Failed to fork into namespaces of PID " PID_FMT ": %m", pid);
+ if (r == 0) {
+ _cleanup_(iovw_free_freep) struct iovec_wrapper *iovw = NULL;
+ Context child_context = {};
+
+ pair[0] = safe_close(pair[0]);
+
+ if (laccess("/run/systemd/coredump", W_OK) < 0) {
+ log_debug_errno(errno, "Cannot find coredump socket, exiting: %m");
+ _exit(EXIT_FAILURE);
+ }
+
+ r = receive_ucred(pair[1], &ucred);
+ if (r < 0) {
+ log_debug_errno(r, "Failed to receive ucred and fd: %m");
+ _exit(EXIT_FAILURE);
+ }
+
+ iovw = iovw_new();
+ if (!iovw) {
+ log_oom();
+ _exit(EXIT_FAILURE);
+ }
+
+ (void) iovw_put_string_field(iovw, "MESSAGE_ID=", SD_MESSAGE_COREDUMP_STR);
+ (void) iovw_put_string_field(iovw, "PRIORITY=", STRINGIFY(LOG_CRIT));
+ (void) iovw_put_string_field(iovw, "COREDUMP_FORWARDED=", "1");
+
+ for (int i = 0; i < _META_ARGV_MAX; i++) {
+ int signo;
+ char buf[DECIMAL_STR_MAX(pid_t)];
+ const char *t = context->meta[i];
+
+ switch(i) {
+
+ case META_ARGV_PID:
+ xsprintf(buf, PID_FMT, ucred.pid);
+ t = buf;
+
+ break;
+
+ case META_ARGV_UID:
+ xsprintf(buf, UID_FMT, ucred.uid);
+ t = buf;
+ break;
+
+ case META_ARGV_GID:
+ xsprintf(buf, GID_FMT, ucred.gid);
+ t = buf;
+ break;
+
+ case META_ARGV_SIGNAL:
+ if (safe_atoi(t, &signo) >= 0 && SIGNAL_VALID(signo))
+ (void) iovw_put_string_field(iovw,
+ "COREDUMP_SIGNAL_NAME=SIG",
+ signal_to_string(signo));
+ break;
+
+ default:
+ break;
+ }
+
+ r = iovw_put_string_field(iovw, meta_field_names[i], t);
+ if (r < 0) {
+ log_debug_errno(r, "Failed to construct iovec: %m");
+ _exit(EXIT_FAILURE);
+ }
+ }
+
+ r = save_context(&child_context, iovw);
+ if (r < 0) {
+ log_debug_errno(r, "Failed to save context: %m");
+ _exit(EXIT_FAILURE);
+ }
+
+ r = gather_pid_metadata_from_procfs(iovw, &child_context);
+ if (r < 0) {
+ log_debug_errno(r, "Failed to gather metadata from procfs: %m");
+ _exit(EXIT_FAILURE);
+ }
+
+ r = send_iovec(iovw, STDIN_FILENO);
+ if (r < 0) {
+ log_debug_errno(r, "Failed to send iovec to coredump socket: %m");
+ _exit(EXIT_FAILURE);
+ }
+
+ _exit(EXIT_SUCCESS);
+ }
+
+ pair[1] = safe_close(pair[1]);
+
+ /* We need to translate the PID, UID, and GID of the crashing process
+ * to the container's namespaces. Do this by sending an SCM_CREDENTIALS
+ * message on a socket pair, and read the result when we join the
+ * container. The kernel will perform the translation for us. */
+ r = send_ucred(pair[0], &ucred);
+ if (r < 0)
+ return log_debug_errno(r, "Failed to send metadata to container: %m");
+
+ r = wait_for_terminate_and_check("(sd-coredumpns)", child, 0);
+ if (r < 0)
+ return log_debug_errno(r, "Failed to wait for child to terminate: %m");
+ if (r != EXIT_SUCCESS)
+ return log_debug_errno(SYNTHETIC_ERRNO(EPROTO), "Failed to process coredump in container: %m");
+
+ return 0;
+}
+
+static int process_kernel(int argc, char* argv[]) {
+ _cleanup_(iovw_free_freep) struct iovec_wrapper *iovw = NULL;
+ Context context = {};
+ int r;
+
+ /* When we're invoked by the kernel, stdout/stderr are closed which is dangerous because the fds
+ * could get reallocated. To avoid hard to debug issues, let's instead bind stdout/stderr to
+ * /dev/null. */
+ r = rearrange_stdio(STDIN_FILENO, -EBADF, -EBADF);
+ if (r < 0)
+ return log_error_errno(r, "Failed to connect stdout/stderr to /dev/null: %m");
+
+ log_debug("Processing coredump received from the kernel...");
+
+ iovw = iovw_new();
+ if (!iovw)
+ return log_oom();
+
+ /* Collect all process metadata passed by the kernel through argv[] */
+ r = gather_pid_metadata_from_argv(iovw, &context, argc - 1, argv + 1);
+ if (r < 0)
+ return r;
+
+ /* Collect the rest of the process metadata retrieved from the runtime */
+ r = gather_pid_metadata_from_procfs(iovw, &context);
+ if (r < 0)
+ return r;
+
+ if (!context.is_journald)
+ /* OK, now we know it's not the journal, hence we can make use of it now. */
+ log_set_target_and_open(LOG_TARGET_JOURNAL_OR_KMSG);
+
+ r = in_same_namespace(getpid_cached(), context.pid, NAMESPACE_PID);
+ if (r < 0)
+ log_debug_errno(r, "Failed to check pidns of crashing process, ignoring: %m");
+ if (r == 0) {
+ /* If this fails, fallback to the old behavior so that
+ * there is still some record of the crash. */
+ r = forward_coredump_to_container(&context);
+ if (r >= 0)
+ return 0;
+ }
+
+ /* If this is PID 1 disable coredump collection, we'll unlikely be able to process
+ * it later on.
+ *
+ * FIXME: maybe we should disable coredumps generation from the beginning and
+ * re-enable it only when we know it's either safe (ie we're not running OOM) or
+ * it's not pid1 ? */
+ if (context.is_pid1) {
+ log_notice("Due to PID 1 having crashed coredump collection will now be turned off.");
+ disable_coredumps();
+ }
+
+ (void) iovw_put_string_field(iovw, "MESSAGE_ID=", SD_MESSAGE_COREDUMP_STR);
+ (void) iovw_put_string_field(iovw, "PRIORITY=", STRINGIFY(LOG_CRIT));
+
+ if (context.is_journald || context.is_pid1)
+ return submit_coredump(&context, iovw, STDIN_FILENO);
+
+ return send_iovec(iovw, STDIN_FILENO);
+}
+
+static int process_backtrace(int argc, char *argv[]) {
+ _cleanup_(journal_importer_cleanup) JournalImporter importer = JOURNAL_IMPORTER_INIT(STDIN_FILENO);
+ _cleanup_(iovw_free_freep) struct iovec_wrapper *iovw = NULL;
+ Context context = {};
+ char *message;
+ int r;
+
+ log_debug("Processing backtrace on stdin...");
+
+ iovw = iovw_new();
+ if (!iovw)
+ return log_oom();
+
+ (void) iovw_put_string_field(iovw, "MESSAGE_ID=", SD_MESSAGE_BACKTRACE_STR);
+ (void) iovw_put_string_field(iovw, "PRIORITY=", STRINGIFY(LOG_CRIT));
+
+ /* Collect all process metadata from argv[] by making sure to skip the
+ * '--backtrace' option */
+ r = gather_pid_metadata_from_argv(iovw, &context, argc - 2, argv + 2);
+ if (r < 0)
+ return r;
+
+ /* Collect the rest of the process metadata retrieved from the runtime */
+ r = gather_pid_metadata_from_procfs(iovw, &context);
+ if (r < 0)
+ return r;
+
+ for (;;) {
+ r = journal_importer_process_data(&importer);
+ if (r < 0)
+ return log_error_errno(r, "Failed to parse journal entry on stdin: %m");
+ if (r == 1 || /* complete entry */
+ journal_importer_eof(&importer)) /* end of data */
+ break;
+ }
+
+ if (journal_importer_eof(&importer)) {
+ log_warning("Did not receive a full journal entry on stdin, ignoring message sent by reporter");
+
+ message = strjoina("Process ", context.meta[META_ARGV_PID],
+ " (", context.meta[META_COMM], ")"
+ " of user ", context.meta[META_ARGV_UID],
+ " failed with ", context.meta[META_ARGV_SIGNAL]);
+
+ r = iovw_put_string_field(iovw, "MESSAGE=", message);
+ if (r < 0)
+ return r;
+ } else {
+ /* The imported iovecs are not supposed to be freed by us so let's copy and merge them at the
+ * end of the array. */
+ r = iovw_append(iovw, &importer.iovw);
+ if (r < 0)
+ return r;
+ }
+
+ r = sd_journal_sendv(iovw->iovec, iovw->count);
+ if (r < 0)
+ return log_error_errno(r, "Failed to log backtrace: %m");
+
+ return 0;
+}
+
+static int run(int argc, char *argv[]) {
+ int r;
+
+ /* First, log to a safe place, since we don't know what crashed and it might
+ * be journald which we'd rather not log to then. */
+
+ log_set_target_and_open(LOG_TARGET_KMSG);
+
+ /* Make sure we never enter a loop */
+ (void) prctl(PR_SET_DUMPABLE, 0);
+
+ /* Ignore all parse errors */
+ (void) parse_config();
+
+ log_debug("Selected storage '%s'.", coredump_storage_to_string(arg_storage));
+ log_debug("Selected compression %s.", yes_no(arg_compress));
+
+ r = sd_listen_fds(false);
+ if (r < 0)
+ return log_error_errno(r, "Failed to determine the number of file descriptors: %m");
+
+ /* If we got an fd passed, we are running in coredumpd mode. Otherwise we
+ * are invoked from the kernel as coredump handler. */
+ if (r == 0) {
+ if (streq_ptr(argv[1], "--backtrace"))
+ return process_backtrace(argc, argv);
+ else
+ return process_kernel(argc, argv);
+ } else if (r == 1)
+ return process_socket(SD_LISTEN_FDS_START);
+
+ return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
+ "Received unexpected number of file descriptors.");
+}
+
+DEFINE_MAIN_FUNCTION(run);
diff --git a/src/coredump/coredump.conf b/src/coredump/coredump.conf
new file mode 100644
index 0000000..ae341e4
--- /dev/null
+++ b/src/coredump/coredump.conf
@@ -0,0 +1,27 @@
+# This file is part of systemd.
+#
+# systemd is free software; you can redistribute it and/or modify it under the
+# terms of the GNU Lesser General Public License as published by the Free
+# Software Foundation; either version 2.1 of the License, or (at your option)
+# any later version.
+#
+# Entries in this file show the compile time defaults. Local configuration
+# should be created by either modifying this file (or a copy of it placed in
+# /etc/ if the original file is shipped in /usr/), or by creating "drop-ins" in
+# the /etc/systemd/coredump.conf.d/ directory. The latter is generally
+# recommended. Defaults can be restored by simply deleting the main
+# configuration file and all drop-ins located in /etc/.
+#
+# Use 'systemd-analyze cat-config systemd/coredump.conf' to display the full config.
+#
+# See coredump.conf(5) for details.
+
+[Coredump]
+#Storage=external
+#Compress=yes
+# On 32-bit, the default is 1G instead of 32G.
+#ProcessSizeMax=32G
+#ExternalSizeMax=32G
+#JournalSizeMax=767M
+#MaxUse=
+#KeepFree=
diff --git a/src/coredump/coredumpctl.c b/src/coredump/coredumpctl.c
new file mode 100644
index 0000000..84d4531
--- /dev/null
+++ b/src/coredump/coredumpctl.c
@@ -0,0 +1,1418 @@
+/* SPDX-License-Identifier: LGPL-2.1-or-later */
+
+#include <fcntl.h>
+#include <getopt.h>
+#include <locale.h>
+#include <stdio.h>
+#include <unistd.h>
+
+#include "sd-bus.h"
+#include "sd-journal.h"
+#include "sd-messages.h"
+
+#include "alloc-util.h"
+#include "build.h"
+#include "bus-error.h"
+#include "bus-locator.h"
+#include "bus-util.h"
+#include "chase.h"
+#include "compress.h"
+#include "constants.h"
+#include "dissect-image.h"
+#include "escape.h"
+#include "fd-util.h"
+#include "format-table.h"
+#include "fs-util.h"
+#include "glob-util.h"
+#include "journal-internal.h"
+#include "journal-util.h"
+#include "log.h"
+#include "macro.h"
+#include "main-func.h"
+#include "mount-util.h"
+#include "pager.h"
+#include "parse-argument.h"
+#include "parse-util.h"
+#include "path-util.h"
+#include "pretty-print.h"
+#include "process-util.h"
+#include "rlimit-util.h"
+#include "sigbus.h"
+#include "signal-util.h"
+#include "string-util.h"
+#include "strv.h"
+#include "terminal-util.h"
+#include "tmpfile-util.h"
+#include "user-util.h"
+#include "verbs.h"
+
+#define SHORT_BUS_CALL_TIMEOUT_USEC (3 * USEC_PER_SEC)
+
+static usec_t arg_since = USEC_INFINITY, arg_until = USEC_INFINITY;
+static const char* arg_field = NULL;
+static const char *arg_debugger = NULL;
+static char **arg_debugger_args = NULL;
+static const char *arg_directory = NULL;
+static char *arg_root = NULL;
+static char *arg_image = NULL;
+static char **arg_file = NULL;
+static JsonFormatFlags arg_json_format_flags = JSON_FORMAT_OFF;
+static PagerFlags arg_pager_flags = 0;
+static int arg_legend = true;
+static size_t arg_rows_max = SIZE_MAX;
+static const char* arg_output = NULL;
+static bool arg_reverse = false;
+static bool arg_quiet = false;
+static bool arg_all = false;
+static ImagePolicy *arg_image_policy = NULL;
+
+STATIC_DESTRUCTOR_REGISTER(arg_debugger_args, strv_freep);
+STATIC_DESTRUCTOR_REGISTER(arg_file, strv_freep);
+STATIC_DESTRUCTOR_REGISTER(arg_image_policy, image_policy_freep);
+
+static int add_match(sd_journal *j, const char *match) {
+ _cleanup_free_ char *p = NULL;
+ const char* prefix, *pattern;
+ pid_t pid;
+ int r;
+
+ if (strchr(match, '='))
+ prefix = "";
+ else if (strchr(match, '/')) {
+ r = path_make_absolute_cwd(match, &p);
+ if (r < 0)
+ return log_error_errno(r, "path_make_absolute_cwd(\"%s\"): %m", match);
+
+ match = p;
+ prefix = "COREDUMP_EXE=";
+ } else if (parse_pid(match, &pid) >= 0)
+ prefix = "COREDUMP_PID=";
+ else
+ prefix = "COREDUMP_COMM=";
+
+ pattern = strjoina(prefix, match);
+ log_debug("Adding match: %s", pattern);
+ r = sd_journal_add_match(j, pattern, 0);
+ if (r < 0)
+ return log_error_errno(r, "Failed to add match \"%s\": %m", match);
+
+ return 0;
+}
+
+static int add_matches(sd_journal *j, char **matches) {
+ int r;
+
+ r = sd_journal_add_match(j, "MESSAGE_ID=" SD_MESSAGE_COREDUMP_STR, 0);
+ if (r < 0)
+ return log_error_errno(r, "Failed to add match \"%s\": %m", "MESSAGE_ID=" SD_MESSAGE_COREDUMP_STR);
+
+ r = sd_journal_add_match(j, "MESSAGE_ID=" SD_MESSAGE_BACKTRACE_STR, 0);
+ if (r < 0)
+ return log_error_errno(r, "Failed to add match \"%s\": %m", "MESSAGE_ID=" SD_MESSAGE_BACKTRACE_STR);
+
+ STRV_FOREACH(match, matches) {
+ r = add_match(j, *match);
+ if (r < 0)
+ return r;
+ }
+
+ return 0;
+}
+
+static int acquire_journal(sd_journal **ret, char **matches) {
+ _cleanup_(sd_journal_closep) sd_journal *j = NULL;
+ int r;
+
+ assert(ret);
+
+ if (arg_directory) {
+ r = sd_journal_open_directory(&j, arg_directory, 0);
+ if (r < 0)
+ return log_error_errno(r, "Failed to open journals in directory: %s: %m", arg_directory);
+ } else if (arg_root) {
+ r = sd_journal_open_directory(&j, arg_root, SD_JOURNAL_OS_ROOT);
+ if (r < 0)
+ return log_error_errno(r, "Failed to open journals in root directory: %s: %m", arg_root);
+ } else if (arg_file) {
+ r = sd_journal_open_files(&j, (const char**)arg_file, 0);
+ if (r < 0)
+ return log_error_errno(r, "Failed to open journal files: %m");
+ } else {
+ r = sd_journal_open(&j, arg_all ? 0 : SD_JOURNAL_LOCAL_ONLY);
+ if (r < 0)
+ return log_error_errno(r, "Failed to open journal: %m");
+ }
+
+ r = journal_access_check_and_warn(j, arg_quiet, true);
+ if (r < 0)
+ return r;
+
+ r = add_matches(j, matches);
+ if (r < 0)
+ return r;
+
+ if (DEBUG_LOGGING) {
+ _cleanup_free_ char *filter = NULL;
+
+ filter = journal_make_match_string(j);
+ log_debug("Journal filter: %s", filter);
+ }
+
+ *ret = TAKE_PTR(j);
+
+ return 0;
+}
+
+static int verb_help(int argc, char **argv, void *userdata) {
+ _cleanup_free_ char *link = NULL;
+ int r;
+
+ r = terminal_urlify_man("coredumpctl", "1", &link);
+ if (r < 0)
+ return log_oom();
+
+ printf("%1$s [OPTIONS...] COMMAND ...\n\n"
+ "%5$sList or retrieve coredumps from the journal.%6$s\n"
+ "\n%3$sCommands:%4$s\n"
+ " list [MATCHES...] List available coredumps (default)\n"
+ " info [MATCHES...] Show detailed information about one or more coredumps\n"
+ " dump [MATCHES...] Print first matching coredump to stdout\n"
+ " debug [MATCHES...] Start a debugger for the first matching coredump\n"
+ "\n%3$sOptions:%4$s\n"
+ " -h --help Show this help\n"
+ " --version Print version string\n"
+ " --no-pager Do not pipe output into a pager\n"
+ " --no-legend Do not print the column headers\n"
+ " --json=pretty|short|off\n"
+ " Generate JSON output\n"
+ " --debugger=DEBUGGER Use the given debugger\n"
+ " -A --debugger-arguments=ARGS Pass the given arguments to the debugger\n"
+ " -n INT Show maximum number of rows\n"
+ " -1 Show information about most recent entry only\n"
+ " -S --since=DATE Only print coredumps since the date\n"
+ " -U --until=DATE Only print coredumps until the date\n"
+ " -r --reverse Show the newest entries first\n"
+ " -F --field=FIELD List all values a certain field takes\n"
+ " -o --output=FILE Write output to FILE\n"
+ " --file=PATH Use journal file\n"
+ " -D --directory=DIR Use journal files from directory\n\n"
+ " -q --quiet Do not show info messages and privilege warning\n"
+ " --all Look at all journal files instead of local ones\n"
+ " --root=PATH Operate on an alternate filesystem root\n"
+ " --image=PATH Operate on disk image as filesystem root\n"
+ " --image-policy=POLICY Specify disk image dissection policy\n"
+ "\nSee the %2$s for details.\n",
+ program_invocation_short_name,
+ link,
+ ansi_underline(),
+ ansi_normal(),
+ ansi_highlight(),
+ ansi_normal());
+
+ return 0;
+}
+
+static int parse_argv(int argc, char *argv[]) {
+ enum {
+ ARG_VERSION = 0x100,
+ ARG_NO_PAGER,
+ ARG_NO_LEGEND,
+ ARG_JSON,
+ ARG_DEBUGGER,
+ ARG_FILE,
+ ARG_ROOT,
+ ARG_IMAGE,
+ ARG_IMAGE_POLICY,
+ ARG_ALL,
+ };
+
+ int c, r;
+
+ static const struct option options[] = {
+ { "help", no_argument, NULL, 'h' },
+ { "version" , no_argument, NULL, ARG_VERSION },
+ { "no-pager", no_argument, NULL, ARG_NO_PAGER },
+ { "no-legend", no_argument, NULL, ARG_NO_LEGEND },
+ { "debugger", required_argument, NULL, ARG_DEBUGGER },
+ { "debugger-arguments", required_argument, NULL, 'A' },
+ { "output", required_argument, NULL, 'o' },
+ { "field", required_argument, NULL, 'F' },
+ { "file", required_argument, NULL, ARG_FILE },
+ { "directory", required_argument, NULL, 'D' },
+ { "reverse", no_argument, NULL, 'r' },
+ { "since", required_argument, NULL, 'S' },
+ { "until", required_argument, NULL, 'U' },
+ { "quiet", no_argument, NULL, 'q' },
+ { "json", required_argument, NULL, ARG_JSON },
+ { "root", required_argument, NULL, ARG_ROOT },
+ { "image", required_argument, NULL, ARG_IMAGE },
+ { "image-policy", required_argument, NULL, ARG_IMAGE_POLICY },
+ { "all", no_argument, NULL, ARG_ALL },
+ {}
+ };
+
+ assert(argc >= 0);
+ assert(argv);
+
+ while ((c = getopt_long(argc, argv, "hA:o:F:1D:rS:U:qn:", options, NULL)) >= 0)
+ switch (c) {
+ case 'h':
+ return verb_help(0, NULL, NULL);
+
+ case ARG_VERSION:
+ return version();
+
+ case ARG_NO_PAGER:
+ arg_pager_flags |= PAGER_DISABLE;
+ break;
+
+ case ARG_NO_LEGEND:
+ arg_legend = false;
+ break;
+
+ case ARG_DEBUGGER:
+ arg_debugger = optarg;
+ break;
+
+ case 'A': {
+ _cleanup_strv_free_ char **l = NULL;
+ r = strv_split_full(&l, optarg, WHITESPACE, EXTRACT_UNQUOTE);
+ if (r < 0)
+ return log_error_errno(r, "Failed to parse debugger arguments '%s': %m", optarg);
+ strv_free_and_replace(arg_debugger_args, l);
+ break;
+ }
+
+ case ARG_FILE:
+ r = glob_extend(&arg_file, optarg, GLOB_NOCHECK);
+ if (r < 0)
+ return log_error_errno(r, "Failed to add paths: %m");
+ break;
+
+ case 'o':
+ if (arg_output)
+ return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
+ "Cannot set output more than once.");
+
+ arg_output = optarg;
+ break;
+
+ case 'S':
+ r = parse_timestamp(optarg, &arg_since);
+ if (r < 0)
+ return log_error_errno(r, "Failed to parse timestamp '%s': %m", optarg);
+ break;
+
+ case 'U':
+ r = parse_timestamp(optarg, &arg_until);
+ if (r < 0)
+ return log_error_errno(r, "Failed to parse timestamp '%s': %m", optarg);
+ break;
+
+ case 'F':
+ if (arg_field)
+ return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
+ "Cannot use --field/-F more than once.");
+ arg_field = optarg;
+ break;
+
+ case '1':
+ arg_rows_max = 1;
+ arg_reverse = true;
+ break;
+
+ case 'n': {
+ unsigned n;
+
+ r = safe_atou(optarg, &n);
+ if (r < 0 || n < 1)
+ return log_error_errno(r < 0 ? r : SYNTHETIC_ERRNO(EINVAL),
+ "Invalid numeric parameter to -n: %s", optarg);
+
+ arg_rows_max = n;
+ break;
+ }
+
+ case 'D':
+ arg_directory = optarg;
+ break;
+
+ case ARG_ROOT:
+ r = parse_path_argument(optarg, false, &arg_root);
+ if (r < 0)
+ return r;
+ break;
+
+ case ARG_IMAGE:
+ r = parse_path_argument(optarg, false, &arg_image);
+ if (r < 0)
+ return r;
+ break;
+
+ case ARG_IMAGE_POLICY:
+ r = parse_image_policy_argument(optarg, &arg_image_policy);
+ if (r < 0)
+ return r;
+ break;
+
+ case 'r':
+ arg_reverse = true;
+ break;
+
+ case 'q':
+ arg_quiet = true;
+ break;
+
+ case ARG_JSON:
+ r = parse_json_argument(optarg, &arg_json_format_flags);
+ if (r <= 0)
+ return r;
+
+ break;
+
+ case ARG_ALL:
+ arg_all = true;
+ break;
+
+ case '?':
+ return -EINVAL;
+
+ default:
+ assert_not_reached();
+ }
+
+ if (arg_since != USEC_INFINITY && arg_until != USEC_INFINITY &&
+ arg_since > arg_until)
+ return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
+ "--since= must be before --until=.");
+
+ if ((!!arg_directory + !!arg_image + !!arg_root) > 1)
+ return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "Please specify either --root=, --image= or -D/--directory=, the combination of these options is not supported.");
+
+ return 1;
+}
+
+static int retrieve(const void *data,
+ size_t len,
+ const char *name,
+ char **var) {
+
+ size_t ident;
+ char *v;
+
+ ident = strlen(name) + 1; /* name + "=" */
+
+ if (len < ident)
+ return 0;
+
+ if (memcmp(data, name, ident - 1) != 0)
+ return 0;
+
+ if (((const char*) data)[ident - 1] != '=')
+ return 0;
+
+ v = strndup((const char*)data + ident, len - ident);
+ if (!v)
+ return log_oom();
+
+ free_and_replace(*var, v);
+ return 1;
+}
+
+static int print_field(FILE* file, sd_journal *j) {
+ const void *d;
+ size_t l;
+
+ assert(file);
+ assert(j);
+
+ assert(arg_field);
+
+ /* A (user-specified) field may appear more than once for a given entry.
+ * We will print all of the occurrences.
+ * This is different below for fields that systemd-coredump uses,
+ * because they cannot meaningfully appear more than once.
+ */
+ SD_JOURNAL_FOREACH_DATA(j, d, l) {
+ _cleanup_free_ char *value = NULL;
+ int r;
+
+ r = retrieve(d, l, arg_field, &value);
+ if (r < 0)
+ return r;
+ if (r > 0)
+ fprintf(file, "%s\n", value);
+ }
+
+ return 0;
+}
+
+#define RETRIEVE(d, l, name, arg) \
+ { \
+ int _r = retrieve(d, l, name, &arg); \
+ if (_r < 0) \
+ return _r; \
+ if (_r > 0) \
+ continue; \
+ }
+
+static void analyze_coredump_file(
+ const char *path,
+ const char **ret_state,
+ const char **ret_color,
+ uint64_t *ret_size) {
+
+ _cleanup_close_ int fd = -EBADF;
+ struct stat st;
+ int r;
+
+ assert(path);
+ assert(ret_state);
+ assert(ret_color);
+ assert(ret_size);
+
+ fd = open(path, O_PATH|O_CLOEXEC);
+ if (fd < 0) {
+ if (errno == ENOENT) {
+ *ret_state = "missing";
+ *ret_color = ansi_grey();
+ *ret_size = UINT64_MAX;
+ return;
+ }
+
+ r = -errno;
+ } else
+ r = access_fd(fd, R_OK);
+ if (r < 0) {
+ if (ERRNO_IS_PRIVILEGE(r)) {
+ *ret_state = "inaccessible";
+ *ret_color = ansi_highlight_yellow();
+ *ret_size = UINT64_MAX;
+ return;
+ }
+ goto error;
+ }
+
+ if (fstat(fd, &st) < 0)
+ goto error;
+
+ if (!S_ISREG(st.st_mode))
+ goto error;
+
+ *ret_state = "present";
+ *ret_color = NULL;
+ *ret_size = st.st_size;
+ return;
+
+error:
+ *ret_state = "error";
+ *ret_color = ansi_highlight_red();
+ *ret_size = UINT64_MAX;
+}
+
+static int resolve_filename(const char *root, char **p) {
+ char *resolved = NULL;
+ int r;
+
+ if (!*p)
+ return 0;
+
+ r = chase(*p, root, CHASE_PREFIX_ROOT|CHASE_NONEXISTENT, &resolved, NULL);
+ if (r < 0)
+ return log_error_errno(r, "Failed to resolve \"%s%s\": %m", strempty(root), *p);
+
+ free_and_replace(*p, resolved);
+
+ /* chase() with flag CHASE_NONEXISTENT will return 0 if the file doesn't exist and 1 if it does.
+ * Return that to the caller
+ */
+ return r;
+}
+
+static int print_list(FILE* file, sd_journal *j, Table *t) {
+ _cleanup_free_ char
+ *mid = NULL, *pid = NULL, *uid = NULL, *gid = NULL,
+ *sgnl = NULL, *exe = NULL, *comm = NULL, *cmdline = NULL,
+ *filename = NULL, *truncated = NULL, *coredump = NULL;
+ const void *d;
+ size_t l;
+ usec_t ts;
+ int r, signal_as_int = 0;
+ const char *present = NULL, *color = NULL;
+ uint64_t size = UINT64_MAX;
+ bool normal_coredump;
+ uid_t uid_as_int = UID_INVALID;
+ gid_t gid_as_int = GID_INVALID;
+ pid_t pid_as_int = 0;
+
+ assert(file);
+ assert(j);
+ assert(t);
+
+ SD_JOURNAL_FOREACH_DATA(j, d, l) {
+ RETRIEVE(d, l, "MESSAGE_ID", mid);
+ RETRIEVE(d, l, "COREDUMP_PID", pid);
+ RETRIEVE(d, l, "COREDUMP_UID", uid);
+ RETRIEVE(d, l, "COREDUMP_GID", gid);
+ RETRIEVE(d, l, "COREDUMP_SIGNAL", sgnl);
+ RETRIEVE(d, l, "COREDUMP_EXE", exe);
+ RETRIEVE(d, l, "COREDUMP_COMM", comm);
+ RETRIEVE(d, l, "COREDUMP_CMDLINE", cmdline);
+ RETRIEVE(d, l, "COREDUMP_FILENAME", filename);
+ RETRIEVE(d, l, "COREDUMP_TRUNCATED", truncated);
+ RETRIEVE(d, l, "COREDUMP", coredump);
+ }
+
+ if (!pid && !uid && !gid && !sgnl && !exe && !comm && !cmdline && !filename)
+ return log_warning_errno(SYNTHETIC_ERRNO(EINVAL), "Empty coredump log entry");
+
+ (void) parse_uid(uid, &uid_as_int);
+ (void) parse_gid(gid, &gid_as_int);
+ (void) parse_pid(pid, &pid_as_int);
+ signal_as_int = signal_from_string(sgnl);
+
+ r = sd_journal_get_realtime_usec(j, &ts);
+ if (r < 0)
+ return log_error_errno(r, "Failed to get realtime timestamp: %m");
+
+ normal_coredump = streq_ptr(mid, SD_MESSAGE_COREDUMP_STR);
+
+ if (filename) {
+ r = resolve_filename(arg_root, &filename);
+ if (r < 0)
+ return r;
+
+ analyze_coredump_file(filename, &present, &color, &size);
+ } else if (coredump)
+ present = "journal";
+ else if (normal_coredump) {
+ present = "none";
+ color = ansi_grey();
+ } else
+ present = NULL;
+
+ if (STRPTR_IN_SET(present, "present", "journal") && truncated && parse_boolean(truncated) > 0)
+ present = "truncated";
+
+ r = table_add_many(
+ t,
+ TABLE_TIMESTAMP, ts,
+ TABLE_PID, pid_as_int,
+ TABLE_UID, uid_as_int,
+ TABLE_GID, gid_as_int,
+ TABLE_SIGNAL, normal_coredump ? signal_as_int : 0,
+ TABLE_STRING, present,
+ TABLE_SET_COLOR, color,
+ TABLE_STRING, exe ?: comm ?: cmdline,
+ TABLE_SIZE, size);
+ if (r < 0)
+ return table_log_add_error(r);
+
+ return 0;
+}
+
+static int print_info(FILE *file, sd_journal *j, bool need_space) {
+ _cleanup_free_ char
+ *mid = NULL, *pid = NULL, *uid = NULL, *gid = NULL,
+ *sgnl = NULL, *exe = NULL, *comm = NULL, *cmdline = NULL,
+ *unit = NULL, *user_unit = NULL, *session = NULL,
+ *boot_id = NULL, *machine_id = NULL, *hostname = NULL,
+ *slice = NULL, *cgroup = NULL, *owner_uid = NULL,
+ *message = NULL, *timestamp = NULL, *filename = NULL,
+ *truncated = NULL, *coredump = NULL,
+ *pkgmeta_name = NULL, *pkgmeta_version = NULL, *pkgmeta_json = NULL;
+ const void *d;
+ size_t l;
+ bool normal_coredump;
+ int r;
+
+ assert(file);
+ assert(j);
+
+ (void) sd_journal_set_data_threshold(j, 0);
+
+ SD_JOURNAL_FOREACH_DATA(j, d, l) {
+ RETRIEVE(d, l, "MESSAGE_ID", mid);
+ RETRIEVE(d, l, "COREDUMP_PID", pid);
+ RETRIEVE(d, l, "COREDUMP_UID", uid);
+ RETRIEVE(d, l, "COREDUMP_GID", gid);
+ RETRIEVE(d, l, "COREDUMP_SIGNAL", sgnl);
+ RETRIEVE(d, l, "COREDUMP_EXE", exe);
+ RETRIEVE(d, l, "COREDUMP_COMM", comm);
+ RETRIEVE(d, l, "COREDUMP_CMDLINE", cmdline);
+ RETRIEVE(d, l, "COREDUMP_HOSTNAME", hostname);
+ RETRIEVE(d, l, "COREDUMP_UNIT", unit);
+ RETRIEVE(d, l, "COREDUMP_USER_UNIT", user_unit);
+ RETRIEVE(d, l, "COREDUMP_SESSION", session);
+ RETRIEVE(d, l, "COREDUMP_OWNER_UID", owner_uid);
+ RETRIEVE(d, l, "COREDUMP_SLICE", slice);
+ RETRIEVE(d, l, "COREDUMP_CGROUP", cgroup);
+ RETRIEVE(d, l, "COREDUMP_TIMESTAMP", timestamp);
+ RETRIEVE(d, l, "COREDUMP_FILENAME", filename);
+ RETRIEVE(d, l, "COREDUMP_TRUNCATED", truncated);
+ RETRIEVE(d, l, "COREDUMP", coredump);
+ RETRIEVE(d, l, "COREDUMP_PACKAGE_NAME", pkgmeta_name);
+ RETRIEVE(d, l, "COREDUMP_PACKAGE_VERSION", pkgmeta_version);
+ RETRIEVE(d, l, "COREDUMP_PACKAGE_JSON", pkgmeta_json);
+ RETRIEVE(d, l, "_BOOT_ID", boot_id);
+ RETRIEVE(d, l, "_MACHINE_ID", machine_id);
+ RETRIEVE(d, l, "MESSAGE", message);
+ }
+
+ if (need_space)
+ fputs("\n", file);
+
+ normal_coredump = streq_ptr(mid, SD_MESSAGE_COREDUMP_STR);
+
+ if (comm)
+ fprintf(file,
+ " PID: %s%s%s (%s)\n",
+ ansi_highlight(), strna(pid), ansi_normal(), comm);
+ else
+ fprintf(file,
+ " PID: %s%s%s\n",
+ ansi_highlight(), strna(pid), ansi_normal());
+
+ if (uid) {
+ uid_t n;
+
+ if (parse_uid(uid, &n) >= 0) {
+ _cleanup_free_ char *u = NULL;
+
+ u = uid_to_name(n);
+ fprintf(file,
+ " UID: %s (%s)\n",
+ uid, u);
+ } else {
+ fprintf(file,
+ " UID: %s\n",
+ uid);
+ }
+ }
+
+ if (gid) {
+ gid_t n;
+
+ if (parse_gid(gid, &n) >= 0) {
+ _cleanup_free_ char *g = NULL;
+
+ g = gid_to_name(n);
+ fprintf(file,
+ " GID: %s (%s)\n",
+ gid, g);
+ } else {
+ fprintf(file,
+ " GID: %s\n",
+ gid);
+ }
+ }
+
+ if (sgnl) {
+ int sig;
+ const char *name = normal_coredump ? "Signal" : "Reason";
+
+ if (normal_coredump && safe_atoi(sgnl, &sig) >= 0)
+ fprintf(file, " %s: %s (%s)\n", name, sgnl, signal_to_string(sig));
+ else
+ fprintf(file, " %s: %s\n", name, sgnl);
+ }
+
+ if (timestamp) {
+ usec_t u;
+
+ r = safe_atou64(timestamp, &u);
+ if (r >= 0)
+ fprintf(file, " Timestamp: %s (%s)\n",
+ FORMAT_TIMESTAMP(u), FORMAT_TIMESTAMP_RELATIVE(u));
+
+ else
+ fprintf(file, " Timestamp: %s\n", timestamp);
+ }
+
+ if (cmdline)
+ fprintf(file, " Command Line: %s\n", cmdline);
+ if (exe)
+ fprintf(file, " Executable: %s%s%s\n", ansi_highlight(), exe, ansi_normal());
+ if (cgroup)
+ fprintf(file, " Control Group: %s\n", cgroup);
+ if (unit)
+ fprintf(file, " Unit: %s\n", unit);
+ if (user_unit)
+ fprintf(file, " User Unit: %s\n", user_unit);
+ if (slice)
+ fprintf(file, " Slice: %s\n", slice);
+ if (session)
+ fprintf(file, " Session: %s\n", session);
+ if (owner_uid) {
+ uid_t n;
+
+ if (parse_uid(owner_uid, &n) >= 0) {
+ _cleanup_free_ char *u = NULL;
+
+ u = uid_to_name(n);
+ fprintf(file,
+ " Owner UID: %s (%s)\n",
+ owner_uid, u);
+ } else {
+ fprintf(file,
+ " Owner UID: %s\n",
+ owner_uid);
+ }
+ }
+ if (boot_id)
+ fprintf(file, " Boot ID: %s\n", boot_id);
+ if (machine_id)
+ fprintf(file, " Machine ID: %s\n", machine_id);
+ if (hostname)
+ fprintf(file, " Hostname: %s\n", hostname);
+
+ if (filename) {
+ r = resolve_filename(arg_root, &filename);
+ if (r < 0)
+ return r;
+
+ const char *state = NULL, *color = NULL;
+ uint64_t size = UINT64_MAX;
+
+ analyze_coredump_file(filename, &state, &color, &size);
+
+ if (STRPTR_IN_SET(state, "present", "journal") && truncated && parse_boolean(truncated) > 0)
+ state = "truncated";
+
+ fprintf(file,
+ " Storage: %s%s (%s)%s\n",
+ strempty(color),
+ filename,
+ state,
+ ansi_normal());
+
+ if (size != UINT64_MAX)
+ fprintf(file, " Size on Disk: %s\n", FORMAT_BYTES(size));
+
+ } else if (coredump)
+ fprintf(file, " Storage: journal\n");
+ else
+ fprintf(file, " Storage: none\n");
+
+ if (pkgmeta_name && pkgmeta_version)
+ fprintf(file, " Package: %s/%s\n", pkgmeta_name, pkgmeta_version);
+
+ /* Print out the build-id of the 'main' ELF module, by matching the JSON key
+ * with the 'exe' field. */
+ if (exe && pkgmeta_json) {
+ _cleanup_(json_variant_unrefp) JsonVariant *v = NULL;
+
+ r = json_parse(pkgmeta_json, 0, &v, NULL, NULL);
+ if (r < 0) {
+ _cleanup_free_ char *esc = cescape(pkgmeta_json);
+ log_warning_errno(r, "json_parse on \"%s\" failed, ignoring: %m", strnull(esc));
+ } else {
+ const char *module_name;
+ JsonVariant *module_json;
+
+ JSON_VARIANT_OBJECT_FOREACH(module_name, module_json, v) {
+ JsonVariant *build_id;
+
+ /* We only print the build-id for the 'main' ELF module */
+ if (!path_equal_filename(module_name, exe))
+ continue;
+
+ build_id = json_variant_by_key(module_json, "buildId");
+ if (build_id)
+ fprintf(file, " build-id: %s\n", json_variant_string(build_id));
+
+ break;
+ }
+ }
+ }
+
+ if (message) {
+ _cleanup_free_ char *m = NULL;
+
+ m = strreplace(message, "\n", "\n ");
+
+ fprintf(file, " Message: %s\n", strstrip(m ?: message));
+ }
+
+ return 0;
+}
+
+static int focus(sd_journal *j) {
+ int r;
+
+ r = sd_journal_seek_tail(j);
+ if (r == 0)
+ r = sd_journal_previous(j);
+ if (r < 0)
+ return log_error_errno(r, "Failed to search journal: %m");
+ if (r == 0)
+ return log_error_errno(SYNTHETIC_ERRNO(ESRCH),
+ "No match found.");
+ return r;
+}
+
+static int print_entry(
+ sd_journal *j,
+ size_t n_found,
+ Table *t) {
+
+ assert(j);
+
+ if (t)
+ return print_list(stdout, j, t);
+ else if (arg_field)
+ return print_field(stdout, j);
+ else
+ return print_info(stdout, j, n_found > 0);
+}
+
+static int dump_list(int argc, char **argv, void *userdata) {
+ _cleanup_(sd_journal_closep) sd_journal *j = NULL;
+ _cleanup_(table_unrefp) Table *t = NULL;
+ size_t n_found = 0;
+ bool verb_is_info;
+ int r;
+
+ verb_is_info = argc >= 1 && streq(argv[0], "info");
+
+ r = acquire_journal(&j, argv + 1);
+ if (r < 0)
+ return r;
+
+ /* The coredumps are likely compressed, and for just listing them we don't need to decompress them,
+ * so let's pick a fairly low data threshold here */
+ (void) sd_journal_set_data_threshold(j, 4096);
+
+ if (!verb_is_info && !arg_field) {
+ t = table_new("time", "pid", "uid", "gid", "sig", "corefile", "exe", "size");
+ if (!t)
+ return log_oom();
+
+ (void) table_set_align_percent(t, TABLE_HEADER_CELL(1), 100);
+ (void) table_set_align_percent(t, TABLE_HEADER_CELL(2), 100);
+ (void) table_set_align_percent(t, TABLE_HEADER_CELL(3), 100);
+ (void) table_set_align_percent(t, TABLE_HEADER_CELL(7), 100);
+
+ table_set_ersatz_string(t, TABLE_ERSATZ_DASH);
+ } else
+ pager_open(arg_pager_flags);
+
+ /* "info" without pattern implies "-1" */
+ if ((arg_rows_max == 1 && arg_reverse) || (verb_is_info && argc == 1)) {
+ r = focus(j);
+ if (r < 0)
+ return r;
+
+ r = print_entry(j, 0, t);
+ if (r < 0)
+ return r;
+ } else {
+ if (arg_since != USEC_INFINITY && !arg_reverse)
+ r = sd_journal_seek_realtime_usec(j, arg_since);
+ else if (arg_until != USEC_INFINITY && arg_reverse)
+ r = sd_journal_seek_realtime_usec(j, arg_until);
+ else if (arg_reverse)
+ r = sd_journal_seek_tail(j);
+ else
+ r = sd_journal_seek_head(j);
+ if (r < 0)
+ return log_error_errno(r, "Failed to seek to date: %m");
+
+ for (;;) {
+ if (!arg_reverse)
+ r = sd_journal_next(j);
+ else
+ r = sd_journal_previous(j);
+ if (r < 0)
+ return log_error_errno(r, "Failed to iterate through journal: %m");
+ if (r == 0)
+ break;
+
+ if (arg_until != USEC_INFINITY && !arg_reverse) {
+ usec_t usec;
+
+ r = sd_journal_get_realtime_usec(j, &usec);
+ if (r < 0)
+ return log_error_errno(r, "Failed to determine timestamp: %m");
+ if (usec > arg_until)
+ continue;
+ }
+
+ if (arg_since != USEC_INFINITY && arg_reverse) {
+ usec_t usec;
+
+ r = sd_journal_get_realtime_usec(j, &usec);
+ if (r < 0)
+ return log_error_errno(r, "Failed to determine timestamp: %m");
+ if (usec < arg_since)
+ continue;
+ }
+
+ r = print_entry(j, n_found++, t);
+ if (r < 0)
+ return r;
+
+ if (arg_rows_max != SIZE_MAX && n_found >= arg_rows_max)
+ break;
+ }
+
+ if (!arg_field && n_found <= 0) {
+ if (!arg_quiet)
+ log_notice("No coredumps found.");
+ return -ESRCH;
+ }
+ }
+
+ if (t) {
+ r = table_print_with_pager(t, arg_json_format_flags, arg_pager_flags, arg_legend);
+ if (r < 0)
+ return r;
+ }
+
+ return 0;
+}
+
+static int save_core(sd_journal *j, FILE *file, char **path, bool *unlink_temp) {
+ const char *data;
+ _cleanup_free_ char *filename = NULL;
+ size_t len;
+ int r, fd;
+ _cleanup_close_ int fdt = -EBADF;
+ char *temp = NULL;
+
+ assert(!(file && path)); /* At most one can be specified */
+ assert(!!path == !!unlink_temp); /* Those must be specified together */
+
+ /* Look for a coredump on disk first. */
+ r = sd_journal_get_data(j, "COREDUMP_FILENAME", (const void**) &data, &len);
+ if (r == 0) {
+ _cleanup_free_ char *resolved = NULL;
+
+ r = retrieve(data, len, "COREDUMP_FILENAME", &filename);
+ if (r < 0)
+ return r;
+ assert(r > 0);
+
+ r = chase_and_access(filename, arg_root, CHASE_PREFIX_ROOT, F_OK, &resolved);
+ if (r < 0)
+ return log_error_errno(r, "Cannot access \"%s%s\": %m", strempty(arg_root), filename);
+
+ free_and_replace(filename, resolved);
+
+ if (path && !ENDSWITH_SET(filename, ".xz", ".lz4", ".zst")) {
+ *path = TAKE_PTR(filename);
+
+ return 0;
+ }
+
+ } else {
+ if (r != -ENOENT)
+ return log_error_errno(r, "Failed to retrieve COREDUMP_FILENAME field: %m");
+ /* Check that we can have a COREDUMP field. We still haven't set a high
+ * data threshold, so we'll get a few kilobytes at most.
+ */
+
+ r = sd_journal_get_data(j, "COREDUMP", (const void**) &data, &len);
+ if (r == -ENOENT)
+ return log_error_errno(r, "Coredump entry has no core attached (neither internally in the journal nor externally on disk).");
+ if (r < 0)
+ return log_error_errno(r, "Failed to retrieve COREDUMP field: %m");
+ }
+
+ if (path) {
+ const char *vt;
+
+ /* Create a temporary file to write the uncompressed core to. */
+
+ r = var_tmp_dir(&vt);
+ if (r < 0)
+ return log_error_errno(r, "Failed to acquire temporary directory path: %m");
+
+ temp = path_join(vt, "coredump-XXXXXX");
+ if (!temp)
+ return log_oom();
+
+ fdt = mkostemp_safe(temp);
+ if (fdt < 0)
+ return log_error_errno(fdt, "Failed to create temporary file: %m");
+ log_debug("Created temporary file %s", temp);
+
+ fd = fdt;
+ } else {
+ /* If neither path or file are specified, we will write to stdout. Let's now check
+ * if stdout is connected to a tty. We checked that the file exists, or that the
+ * core might be stored in the journal. In this second case, if we found the entry,
+ * in all likelihood we will be able to access the COREDUMP= field. In either case,
+ * we stop before doing any "real" work, i.e. before starting decompression or
+ * reading from the file or creating temporary files.
+ */
+ if (!file) {
+ if (on_tty())
+ return log_error_errno(SYNTHETIC_ERRNO(ENOTTY),
+ "Refusing to dump core to tty"
+ " (use shell redirection or specify --output).");
+ file = stdout;
+ }
+
+ fd = fileno(file);
+ }
+
+ if (filename) {
+#if HAVE_COMPRESSION
+ _cleanup_close_ int fdf = -EBADF;
+
+ fdf = open(filename, O_RDONLY | O_CLOEXEC);
+ if (fdf < 0) {
+ r = log_error_errno(errno, "Failed to open %s: %m", filename);
+ goto error;
+ }
+
+ r = decompress_stream(filename, fdf, fd, -1);
+ if (r < 0) {
+ log_error_errno(r, "Failed to decompress %s: %m", filename);
+ goto error;
+ }
+#else
+ r = log_error_errno(SYNTHETIC_ERRNO(EOPNOTSUPP),
+ "Cannot decompress file. Compiled without compression support.");
+ goto error;
+#endif
+ } else {
+ ssize_t sz;
+
+ /* We want full data, nothing truncated. */
+ sd_journal_set_data_threshold(j, 0);
+
+ r = sd_journal_get_data(j, "COREDUMP", (const void**) &data, &len);
+ if (r < 0)
+ return log_error_errno(r, "Failed to retrieve COREDUMP field: %m");
+
+ assert(len >= 9);
+ data += 9;
+ len -= 9;
+
+ sz = write(fd, data, len);
+ if (sz < 0) {
+ r = log_error_errno(errno, "Failed to write output: %m");
+ goto error;
+ }
+ if (sz != (ssize_t) len) {
+ log_error("Short write to output.");
+ r = -EIO;
+ goto error;
+ }
+ }
+
+ if (temp) {
+ *path = temp;
+ *unlink_temp = true;
+ }
+ return 0;
+
+error:
+ if (temp) {
+ (void) unlink(temp);
+ log_debug("Removed temporary file %s", temp);
+ }
+ return r;
+}
+
+static int dump_core(int argc, char **argv, void *userdata) {
+ _cleanup_(sd_journal_closep) sd_journal *j = NULL;
+ _cleanup_fclose_ FILE *f = NULL;
+ int r;
+
+ if (arg_field)
+ return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
+ "Option --field/-F only makes sense with list");
+
+ r = acquire_journal(&j, argv + 1);
+ if (r < 0)
+ return r;
+
+ r = focus(j);
+ if (r < 0)
+ return r;
+
+ if (arg_output) {
+ f = fopen(arg_output, "we");
+ if (!f)
+ return log_error_errno(errno, "Failed to open \"%s\" for writing: %m", arg_output);
+ }
+
+ print_info(f ? stdout : stderr, j, false);
+
+ r = save_core(j, f, NULL, NULL);
+ if (r < 0)
+ return r;
+
+ r = sd_journal_previous(j);
+ if (r > 0 && !arg_quiet)
+ log_notice("More than one entry matches, ignoring rest.");
+
+ return 0;
+}
+
+static void sigterm_handler(int signal, siginfo_t *info, void *ucontext) {
+ assert(signal == SIGTERM);
+ assert(info);
+
+ /* If the sender is not us, propagate the signal to all processes in
+ * the same process group */
+ if (pid_is_valid(info->si_pid) && info->si_pid != getpid_cached())
+ (void) kill(0, signal);
+}
+
+static int run_debug(int argc, char **argv, void *userdata) {
+ _cleanup_(sd_journal_closep) sd_journal *j = NULL;
+ _cleanup_free_ char *exe = NULL, *path = NULL;
+ _cleanup_strv_free_ char **debugger_call = NULL;
+ struct sigaction sa = {
+ .sa_sigaction = sigterm_handler,
+ .sa_flags = SA_SIGINFO,
+ };
+ bool unlink_path = false;
+ const char *data, *fork_name;
+ size_t len;
+ pid_t pid;
+ int r;
+
+ if (!arg_debugger) {
+ char *env_debugger;
+
+ env_debugger = getenv("SYSTEMD_DEBUGGER");
+ if (env_debugger)
+ arg_debugger = env_debugger;
+ else
+ arg_debugger = "gdb";
+ }
+
+ r = strv_extend(&debugger_call, arg_debugger);
+ if (r < 0)
+ return log_oom();
+
+ r = strv_extend_strv(&debugger_call, arg_debugger_args, false);
+ if (r < 0)
+ return log_oom();
+
+ if (arg_field)
+ return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
+ "Option --field/-F only makes sense with list");
+
+ r = acquire_journal(&j, argv + 1);
+ if (r < 0)
+ return r;
+
+ r = focus(j);
+ if (r < 0)
+ return r;
+
+ if (!arg_quiet) {
+ print_info(stdout, j, false);
+ fputs("\n", stdout);
+ }
+
+ r = sd_journal_get_data(j, "COREDUMP_EXE", (const void**) &data, &len);
+ if (r < 0)
+ return log_error_errno(r, "Failed to retrieve COREDUMP_EXE field: %m");
+
+ assert(len > STRLEN("COREDUMP_EXE="));
+ data += STRLEN("COREDUMP_EXE=");
+ len -= STRLEN("COREDUMP_EXE=");
+
+ exe = strndup(data, len);
+ if (!exe)
+ return log_oom();
+
+ if (endswith(exe, " (deleted)"))
+ return log_error_errno(SYNTHETIC_ERRNO(ENOENT),
+ "Binary already deleted.");
+
+ if (!path_is_absolute(exe))
+ return log_error_errno(SYNTHETIC_ERRNO(ENOENT),
+ "Binary is not an absolute path.");
+
+ r = resolve_filename(arg_root, &exe);
+ if (r < 0)
+ return r;
+
+ r = save_core(j, NULL, &path, &unlink_path);
+ if (r < 0)
+ return r;
+
+ r = strv_extend_strv(&debugger_call, STRV_MAKE(exe, "-c", path), false);
+ if (r < 0)
+ return log_oom();
+
+ if (arg_root) {
+ if (streq(arg_debugger, "gdb")) {
+ const char *sysroot_cmd;
+ sysroot_cmd = strjoina("set sysroot ", arg_root);
+
+ r = strv_extend_strv(&debugger_call, STRV_MAKE("-iex", sysroot_cmd), false);
+ if (r < 0)
+ return log_oom();
+ } else if (streq(arg_debugger, "lldb")) {
+ const char *sysroot_cmd;
+ sysroot_cmd = strjoina("platform select --sysroot ", arg_root, " host");
+
+ r = strv_extend_strv(&debugger_call, STRV_MAKE("-O", sysroot_cmd), false);
+ if (r < 0)
+ return log_oom();
+ }
+ }
+
+ /* Don't interfere with gdb and its handling of SIGINT. */
+ (void) ignore_signals(SIGINT);
+ (void) sigaction(SIGTERM, &sa, NULL);
+
+ fork_name = strjoina("(", debugger_call[0], ")");
+
+ r = safe_fork(fork_name, FORK_RESET_SIGNALS|FORK_DEATHSIG_SIGTERM|FORK_CLOSE_ALL_FDS|FORK_RLIMIT_NOFILE_SAFE|FORK_LOG|FORK_FLUSH_STDIO, &pid);
+ if (r < 0)
+ goto finish;
+ if (r == 0) {
+ execvp(debugger_call[0], debugger_call);
+ log_open();
+ log_error_errno(errno, "Failed to invoke %s: %m", debugger_call[0]);
+ _exit(EXIT_FAILURE);
+ }
+
+ r = wait_for_terminate_and_check(debugger_call[0], pid, WAIT_LOG_ABNORMAL);
+
+finish:
+ (void) default_signals(SIGINT, SIGTERM);
+
+ if (unlink_path) {
+ log_debug("Removed temporary file %s", path);
+ (void) unlink(path);
+ }
+
+ return r;
+}
+
+static int check_units_active(void) {
+ _cleanup_(sd_bus_flush_close_unrefp) sd_bus *bus = NULL;
+ _cleanup_(sd_bus_message_unrefp) sd_bus_message *m = NULL;
+ _cleanup_(sd_bus_error_free) sd_bus_error error = SD_BUS_ERROR_NULL;
+ _cleanup_(sd_bus_message_unrefp) sd_bus_message *reply = NULL;
+ int c = 0, r;
+ const char *id, *state, *substate;
+
+ if (arg_quiet)
+ return false;
+
+ r = sd_bus_default_system(&bus);
+ if (r == -ENOENT) {
+ log_debug("D-Bus is not running, skipping active unit check");
+ return 0;
+ }
+ if (r < 0)
+ return log_error_errno(r, "Failed to acquire bus: %m");
+
+ r = bus_message_new_method_call(bus, &m, bus_systemd_mgr, "ListUnitsByPatterns");
+ if (r < 0)
+ return bus_log_create_error(r);
+
+ r = sd_bus_message_append_strv(m, NULL);
+ if (r < 0)
+ return bus_log_create_error(r);
+
+ r = sd_bus_message_append_strv(m, STRV_MAKE("systemd-coredump@*.service"));
+ if (r < 0)
+ return bus_log_create_error(r);
+
+ r = sd_bus_call(bus, m, SHORT_BUS_CALL_TIMEOUT_USEC, &error, &reply);
+ if (r < 0)
+ return log_error_errno(r, "Failed to check if any systemd-coredump@.service units are running: %s",
+ bus_error_message(&error, r));
+
+ r = sd_bus_message_enter_container(reply, SD_BUS_TYPE_ARRAY, "(ssssssouso)");
+ if (r < 0)
+ return bus_log_parse_error(r);
+
+ while ((r = sd_bus_message_read(
+ reply, "(ssssssouso)",
+ &id, NULL, NULL, &state, &substate,
+ NULL, NULL, NULL, NULL, NULL)) > 0) {
+ bool found = !STR_IN_SET(state, "inactive", "dead", "failed");
+ log_debug("Unit %s is %s/%s, %scounting it.", id, state, substate, found ? "" : "not ");
+ c += found;
+ }
+ if (r < 0)
+ return bus_log_parse_error(r);
+
+ r = sd_bus_message_exit_container(reply);
+ if (r < 0)
+ return bus_log_parse_error(r);
+
+ return c;
+}
+
+static int coredumpctl_main(int argc, char *argv[]) {
+
+ static const Verb verbs[] = {
+ { "list", VERB_ANY, VERB_ANY, VERB_DEFAULT, dump_list },
+ { "info", VERB_ANY, VERB_ANY, 0, dump_list },
+ { "dump", VERB_ANY, VERB_ANY, 0, dump_core },
+ { "debug", VERB_ANY, VERB_ANY, 0, run_debug },
+ { "gdb", VERB_ANY, VERB_ANY, 0, run_debug },
+ { "help", VERB_ANY, 1, 0, verb_help },
+ {}
+ };
+
+ return dispatch_verb(argc, argv, verbs, NULL);
+}
+
+static int run(int argc, char *argv[]) {
+ _cleanup_(loop_device_unrefp) LoopDevice *loop_device = NULL;
+ _cleanup_(umount_and_freep) char *mounted_dir = NULL;
+ int r, units_active;
+
+ setlocale(LC_ALL, "");
+ log_setup();
+
+ /* The journal merging logic potentially needs a lot of fds. */
+ (void) rlimit_nofile_bump(HIGH_RLIMIT_NOFILE);
+
+ r = parse_argv(argc, argv);
+ if (r <= 0)
+ return r;
+
+ sigbus_install();
+
+ units_active = check_units_active(); /* error is treated the same as 0 */
+
+ if (arg_image) {
+ assert(!arg_root);
+
+ r = mount_image_privately_interactively(
+ arg_image,
+ arg_image_policy,
+ DISSECT_IMAGE_GENERIC_ROOT |
+ DISSECT_IMAGE_REQUIRE_ROOT |
+ DISSECT_IMAGE_RELAX_VAR_CHECK |
+ DISSECT_IMAGE_VALIDATE_OS,
+ &mounted_dir,
+ /* ret_dir_fd= */ NULL,
+ &loop_device);
+ if (r < 0)
+ return r;
+
+ arg_root = strdup(mounted_dir);
+ if (!arg_root)
+ return log_oom();
+ }
+
+ r = coredumpctl_main(argc, argv);
+
+ if (units_active > 0)
+ printf("%s-- Notice: %d systemd-coredump@.service %s, output may be incomplete.%s\n",
+ ansi_highlight_red(),
+ units_active, units_active == 1 ? "unit is running" : "units are running",
+ ansi_normal());
+
+ return r;
+}
+
+DEFINE_MAIN_FUNCTION_WITH_POSITIVE_FAILURE(run);
diff --git a/src/coredump/meson.build b/src/coredump/meson.build
new file mode 100644
index 0000000..a699746
--- /dev/null
+++ b/src/coredump/meson.build
@@ -0,0 +1,48 @@
+# SPDX-License-Identifier: LGPL-2.1-or-later
+
+systemd_coredump_sources = files(
+ 'coredump.c',
+ 'coredump-vacuum.c',
+)
+
+common_link_with = [
+ libshared,
+ libbasic_compress,
+]
+
+common_dependencies = [
+ liblz4,
+ libxz,
+ libzstd,
+ threads,
+]
+
+executables += [
+ libexec_template + {
+ 'name' : 'systemd-coredump',
+ 'conditions' : ['ENABLE_COREDUMP'],
+ 'sources' : systemd_coredump_sources,
+ 'link_with' : common_link_with,
+ 'dependencies' : common_dependencies + [libacl],
+ },
+ executable_template + {
+ 'name' : 'coredumpctl',
+ 'public' : true,
+ 'conditions' : ['ENABLE_COREDUMP'],
+ 'sources' : files('coredumpctl.c'),
+ 'link_with' : common_link_with,
+ 'dependencies' : common_dependencies,
+ },
+ test_template + {
+ 'sources' : files(
+ 'test-coredump-vacuum.c',
+ 'coredump-vacuum.c',
+ ),
+ 'type' : 'manual',
+ },
+]
+
+if conf.get('ENABLE_COREDUMP') == 1 and install_sysconfdir_samples
+ install_data('coredump.conf',
+ install_dir : pkgconfigfiledir)
+endif
diff --git a/src/coredump/test-coredump-vacuum.c b/src/coredump/test-coredump-vacuum.c
new file mode 100644
index 0000000..27f8330
--- /dev/null
+++ b/src/coredump/test-coredump-vacuum.c
@@ -0,0 +1,15 @@
+/* SPDX-License-Identifier: LGPL-2.1-or-later */
+
+#include <stdlib.h>
+
+#include "coredump-vacuum.h"
+#include "tests.h"
+
+int main(int argc, char *argv[]) {
+ test_setup_logging(LOG_DEBUG);
+
+ if (coredump_vacuum(-1, UINT64_MAX, 70 * 1024) < 0)
+ return EXIT_FAILURE;
+
+ return EXIT_SUCCESS;
+}
generated by cgit v1.2.3 (git 2.39.1) at 2024-11-27 21:53:05 +0000