1 /* SPDX-License-Identifier: LGPL-2.1+ */
11 #include <elfutils/libdwfl.h>
14 #include "sd-daemon.h"
15 #include "sd-journal.h"
17 #include "sd-messages.h"
20 #include "alloc-util.h"
21 #include "capability-util.h"
22 #include "cgroup-util.h"
24 #include "conf-parser.h"
26 #include "coredump-vacuum.h"
27 #include "dirent-util.h"
33 #include "journal-importer.h"
36 #include "main-func.h"
37 #include "memory-util.h"
40 #include "parse-util.h"
41 #include "process-util.h"
42 #include "signal-util.h"
43 #include "socket-util.h"
45 #include "stacktrace.h"
46 #include "string-table.h"
47 #include "string-util.h"
49 #include "tmpfile-util.h"
50 #include "user-util.h"
52 /* The maximum size up to which we process coredumps */
53 #define PROCESS_SIZE_MAX ((uint64_t) (2LLU*1024LLU*1024LLU*1024LLU))
55 /* The maximum size up to which we leave the coredump around on disk */
56 #define EXTERNAL_SIZE_MAX PROCESS_SIZE_MAX
58 /* The maximum size up to which we store the coredump in the journal */
59 #ifndef FUZZING_BUILD_MODE_UNSAFE_FOR_PRODUCTION
60 #define JOURNAL_SIZE_MAX ((size_t) (767LU*1024LU*1024LU))
62 /* oss-fuzz limits memory usage. */
63 #define JOURNAL_SIZE_MAX ((size_t) (10LU*1024LU*1024LU))
66 /* Make sure to not make this larger than the maximum journal entry
67 * size. See DATA_SIZE_MAX in journal-importer.h. */
68 assert_cc(JOURNAL_SIZE_MAX
<= DATA_SIZE_MAX
);
71 /* We use this as array indexes for a couple of special fields we use for
72 * naming coredump files, and attaching xattrs, and for indexing argv[].
74 * Our pattern for man:systectl(1) kernel.core_pattern is such that the
75 * kernel passes fields until CONTEXT_RLIMIT as arguments in argv[]. After
76 * that it gets complicated: the kernel passes "comm" as one or more fields
77 * starting at index CONTEXT_COMM (in other words, full "comm" is under index
78 * CONTEXT_COMM when it does not contain spaces, which is the common
79 * case). This mapping is not reversible, so we prefer to retrieve "comm"
80 * from /proc. We only fall back to argv[CONTEXT_COMM...] when that fails.
82 * In the internal context[] array, fields before CONTEXT_COMM are the
83 * strings from argv[], so they should not be freed. The strings at indices
84 * CONTEXT_COMM and higher are allocated by us and should be freed at the
100 typedef enum CoredumpStorage
{
101 COREDUMP_STORAGE_NONE
,
102 COREDUMP_STORAGE_EXTERNAL
,
103 COREDUMP_STORAGE_JOURNAL
,
104 _COREDUMP_STORAGE_MAX
,
105 _COREDUMP_STORAGE_INVALID
= -1
108 static const char* const coredump_storage_table
[_COREDUMP_STORAGE_MAX
] = {
109 [COREDUMP_STORAGE_NONE
] = "none",
110 [COREDUMP_STORAGE_EXTERNAL
] = "external",
111 [COREDUMP_STORAGE_JOURNAL
] = "journal",
114 DEFINE_PRIVATE_STRING_TABLE_LOOKUP(coredump_storage
, CoredumpStorage
);
115 static DEFINE_CONFIG_PARSE_ENUM(config_parse_coredump_storage
, coredump_storage
, CoredumpStorage
, "Failed to parse storage setting");
117 static CoredumpStorage arg_storage
= COREDUMP_STORAGE_EXTERNAL
;
118 static bool arg_compress
= true;
119 static uint64_t arg_process_size_max
= PROCESS_SIZE_MAX
;
120 static uint64_t arg_external_size_max
= EXTERNAL_SIZE_MAX
;
121 static uint64_t arg_journal_size_max
= JOURNAL_SIZE_MAX
;
122 static uint64_t arg_keep_free
= (uint64_t) -1;
123 static uint64_t arg_max_use
= (uint64_t) -1;
125 static int parse_config(void) {
126 static const ConfigTableItem items
[] = {
127 { "Coredump", "Storage", config_parse_coredump_storage
, 0, &arg_storage
},
128 { "Coredump", "Compress", config_parse_bool
, 0, &arg_compress
},
129 { "Coredump", "ProcessSizeMax", config_parse_iec_uint64
, 0, &arg_process_size_max
},
130 { "Coredump", "ExternalSizeMax", config_parse_iec_uint64
, 0, &arg_external_size_max
},
131 { "Coredump", "JournalSizeMax", config_parse_iec_size
, 0, &arg_journal_size_max
},
132 { "Coredump", "KeepFree", config_parse_iec_uint64
, 0, &arg_keep_free
},
133 { "Coredump", "MaxUse", config_parse_iec_uint64
, 0, &arg_max_use
},
137 return config_parse_many_nulstr(PKGSYSCONFDIR
"/coredump.conf",
138 CONF_PATHS_NULSTR("systemd/coredump.conf.d"),
140 config_item_table_lookup
, items
,
141 CONFIG_PARSE_WARN
, NULL
);
144 static uint64_t storage_size_max(void) {
145 if (arg_storage
== COREDUMP_STORAGE_EXTERNAL
)
146 return arg_external_size_max
;
147 if (arg_storage
== COREDUMP_STORAGE_JOURNAL
)
148 return arg_journal_size_max
;
149 assert(arg_storage
== COREDUMP_STORAGE_NONE
);
153 static int fix_acl(int fd
, uid_t uid
) {
156 _cleanup_(acl_freep
) acl_t acl
= NULL
;
158 acl_permset_t permset
;
163 if (uid_is_system(uid
) || uid_is_dynamic(uid
) || uid
== UID_NOBODY
)
166 /* Make sure normal users can read (but not write or delete)
167 * their own coredumps */
169 acl
= acl_get_fd(fd
);
171 return log_error_errno(errno
, "Failed to get ACL: %m");
173 if (acl_create_entry(&acl
, &entry
) < 0 ||
174 acl_set_tag_type(entry
, ACL_USER
) < 0 ||
175 acl_set_qualifier(entry
, &uid
) < 0)
176 return log_error_errno(errno
, "Failed to patch ACL: %m");
178 if (acl_get_permset(entry
, &permset
) < 0 ||
179 acl_add_perm(permset
, ACL_READ
) < 0)
180 return log_warning_errno(errno
, "Failed to patch ACL: %m");
182 r
= calc_acl_mask_if_needed(&acl
);
184 return log_warning_errno(r
, "Failed to patch ACL: %m");
186 if (acl_set_fd(fd
, acl
) < 0)
187 return log_error_errno(errno
, "Failed to apply ACL: %m");
193 static int fix_xattr(int fd
, const char *context
[_CONTEXT_MAX
]) {
195 static const char * const xattrs
[_CONTEXT_MAX
] = {
196 [CONTEXT_PID
] = "user.coredump.pid",
197 [CONTEXT_UID
] = "user.coredump.uid",
198 [CONTEXT_GID
] = "user.coredump.gid",
199 [CONTEXT_SIGNAL
] = "user.coredump.signal",
200 [CONTEXT_TIMESTAMP
] = "user.coredump.timestamp",
201 [CONTEXT_RLIMIT
] = "user.coredump.rlimit",
202 [CONTEXT_HOSTNAME
] = "user.coredump.hostname",
203 [CONTEXT_COMM
] = "user.coredump.comm",
204 [CONTEXT_EXE
] = "user.coredump.exe",
212 /* Attach some metadata to coredumps via extended
213 * attributes. Just because we can. */
215 for (i
= 0; i
< _CONTEXT_MAX
; i
++) {
218 if (isempty(context
[i
]) || !xattrs
[i
])
221 k
= fsetxattr(fd
, xattrs
[i
], context
[i
], strlen(context
[i
]), XATTR_CREATE
);
229 #define filename_escape(s) xescape((s), "./ ")
231 static const char *coredump_tmpfile_name(const char *s
) {
232 return s
? s
: "(unnamed temporary file)";
235 static int fix_permissions(
237 const char *filename
,
239 const char *context
[_CONTEXT_MAX
],
248 /* Ignore errors on these */
249 (void) fchmod(fd
, 0640);
250 (void) fix_acl(fd
, uid
);
251 (void) fix_xattr(fd
, context
);
254 return log_error_errno(errno
, "Failed to sync coredump %s: %m", coredump_tmpfile_name(filename
));
256 (void) fsync_directory_of_file(fd
);
258 r
= link_tmpfile(fd
, filename
, target
);
260 return log_error_errno(r
, "Failed to move coredump %s into place: %m", target
);
265 static int maybe_remove_external_coredump(const char *filename
, uint64_t size
) {
267 /* Returns 1 if might remove, 0 if will not remove, < 0 on error. */
269 if (arg_storage
== COREDUMP_STORAGE_EXTERNAL
&&
270 size
<= arg_external_size_max
)
276 if (unlink(filename
) < 0 && errno
!= ENOENT
)
277 return log_error_errno(errno
, "Failed to unlink %s: %m", filename
);
282 static int make_filename(const char *context
[_CONTEXT_MAX
], char **ret
) {
283 _cleanup_free_
char *c
= NULL
, *u
= NULL
, *p
= NULL
, *t
= NULL
;
284 sd_id128_t boot
= {};
289 c
= filename_escape(context
[CONTEXT_COMM
]);
293 u
= filename_escape(context
[CONTEXT_UID
]);
297 r
= sd_id128_get_boot(&boot
);
301 p
= filename_escape(context
[CONTEXT_PID
]);
305 t
= filename_escape(context
[CONTEXT_TIMESTAMP
]);
310 "/var/lib/systemd/coredump/core.%s.%s." SD_ID128_FORMAT_STR
".%s.%s000000",
313 SD_ID128_FORMAT_VAL(boot
),
321 static int save_external_coredump(
322 const char *context
[_CONTEXT_MAX
],
328 bool *ret_truncated
) {
330 _cleanup_free_
char *fn
= NULL
, *tmp
= NULL
;
331 _cleanup_close_
int fd
= -1;
332 uint64_t rlimit
, process_limit
, max_size
;
338 assert(ret_filename
);
343 r
= parse_uid(context
[CONTEXT_UID
], &uid
);
345 return log_error_errno(r
, "Failed to parse UID: %m");
347 r
= safe_atou64(context
[CONTEXT_RLIMIT
], &rlimit
);
349 return log_error_errno(r
, "Failed to parse resource limit '%s': %m", context
[CONTEXT_RLIMIT
]);
350 if (rlimit
< page_size()) {
351 /* Is coredumping disabled? Then don't bother saving/processing the coredump.
352 * Anything below PAGE_SIZE cannot give a readable coredump (the kernel uses
353 * ELF_EXEC_PAGESIZE which is not easily accessible, but is usually the same as PAGE_SIZE. */
354 return log_info_errno(SYNTHETIC_ERRNO(EBADSLT
),
355 "Resource limits disable core dumping for process %s (%s).",
356 context
[CONTEXT_PID
], context
[CONTEXT_COMM
]);
359 process_limit
= MAX(arg_process_size_max
, storage_size_max());
360 if (process_limit
== 0)
361 return log_debug_errno(SYNTHETIC_ERRNO(EBADSLT
),
362 "Limits for coredump processing and storage are both 0, not dumping core.");
364 /* Never store more than the process configured, or than we actually shall keep or process */
365 max_size
= MIN(rlimit
, process_limit
);
367 r
= make_filename(context
, &fn
);
369 return log_error_errno(r
, "Failed to determine coredump file name: %m");
371 (void) mkdir_p_label("/var/lib/systemd/coredump", 0755);
373 fd
= open_tmpfile_linkable(fn
, O_RDWR
|O_CLOEXEC
, &tmp
);
375 return log_error_errno(fd
, "Failed to create temporary file for coredump %s: %m", fn
);
377 r
= copy_bytes(input_fd
, fd
, max_size
, 0);
379 log_error_errno(r
, "Cannot store coredump of %s (%s): %m", context
[CONTEXT_PID
], context
[CONTEXT_COMM
]);
382 *ret_truncated
= r
== 1;
385 LOG_MESSAGE("Core file was truncated to %zu bytes.", max_size
),
386 "SIZE_LIMIT=%zu", max_size
,
387 "MESSAGE_ID=" SD_MESSAGE_TRUNCATED_CORE_STR
);
389 if (fstat(fd
, &st
) < 0) {
390 log_error_errno(errno
, "Failed to fstat core file %s: %m", coredump_tmpfile_name(tmp
));
394 if (lseek(fd
, 0, SEEK_SET
) == (off_t
) -1) {
395 log_error_errno(errno
, "Failed to seek on %s: %m", coredump_tmpfile_name(tmp
));
399 #if HAVE_XZ || HAVE_LZ4
400 /* If we will remove the coredump anyway, do not compress. */
401 if (arg_compress
&& !maybe_remove_external_coredump(NULL
, st
.st_size
)) {
403 _cleanup_free_
char *fn_compressed
= NULL
, *tmp_compressed
= NULL
;
404 _cleanup_close_
int fd_compressed
= -1;
406 fn_compressed
= strappend(fn
, COMPRESSED_EXT
);
407 if (!fn_compressed
) {
412 fd_compressed
= open_tmpfile_linkable(fn_compressed
, O_RDWR
|O_CLOEXEC
, &tmp_compressed
);
413 if (fd_compressed
< 0) {
414 log_error_errno(fd_compressed
, "Failed to create temporary file for coredump %s: %m", fn_compressed
);
418 r
= compress_stream(fd
, fd_compressed
, -1);
420 log_error_errno(r
, "Failed to compress %s: %m", coredump_tmpfile_name(tmp_compressed
));
421 goto fail_compressed
;
424 r
= fix_permissions(fd_compressed
, tmp_compressed
, fn_compressed
, context
, uid
);
426 goto fail_compressed
;
428 /* OK, this worked, we can get rid of the uncompressed version now */
432 *ret_filename
= TAKE_PTR(fn_compressed
); /* compressed */
433 *ret_node_fd
= TAKE_FD(fd_compressed
); /* compressed */
434 *ret_data_fd
= TAKE_FD(fd
); /* uncompressed */
435 *ret_size
= (uint64_t) st
.st_size
; /* uncompressed */
441 (void) unlink(tmp_compressed
);
447 r
= fix_permissions(fd
, tmp
, fn
, context
, uid
);
451 *ret_filename
= TAKE_PTR(fn
);
452 *ret_data_fd
= TAKE_FD(fd
);
454 *ret_size
= (uint64_t) st
.st_size
;
464 static int allocate_journal_field(int fd
, size_t size
, char **ret
, size_t *ret_size
) {
465 _cleanup_free_
char *field
= NULL
;
472 if (lseek(fd
, 0, SEEK_SET
) == (off_t
) -1)
473 return log_warning_errno(errno
, "Failed to seek: %m");
475 field
= malloc(9 + size
);
477 log_warning("Failed to allocate memory for coredump, coredump will not be stored.");
481 memcpy(field
, "COREDUMP=", 9);
483 n
= read(fd
, field
+ 9, size
);
485 return log_error_errno((int) n
, "Failed to read core data: %m");
486 if ((size_t) n
< size
)
487 return log_error_errno(SYNTHETIC_ERRNO(EIO
),
488 "Core data too short.");
490 *ret
= TAKE_PTR(field
);
491 *ret_size
= size
+ 9;
496 /* Joins /proc/[pid]/fd/ and /proc/[pid]/fdinfo/ into the following lines:
510 static int compose_open_fds(pid_t pid
, char **open_fds
) {
511 _cleanup_closedir_
DIR *proc_fd_dir
= NULL
;
512 _cleanup_close_
int proc_fdinfo_fd
= -1;
513 _cleanup_free_
char *buffer
= NULL
;
514 _cleanup_fclose_
FILE *stream
= NULL
;
515 const char *fddelim
= "", *path
;
516 struct dirent
*dent
= NULL
;
521 assert(open_fds
!= NULL
);
523 path
= procfs_file_alloca(pid
, "fd");
524 proc_fd_dir
= opendir(path
);
528 proc_fdinfo_fd
= openat(dirfd(proc_fd_dir
), "../fdinfo", O_DIRECTORY
|O_NOFOLLOW
|O_CLOEXEC
|O_PATH
);
529 if (proc_fdinfo_fd
< 0)
532 stream
= open_memstream_unlocked(&buffer
, &size
);
536 FOREACH_DIRENT(dent
, proc_fd_dir
, return -errno
) {
537 _cleanup_fclose_
FILE *fdinfo
= NULL
;
538 _cleanup_free_
char *fdname
= NULL
;
541 r
= readlinkat_malloc(dirfd(proc_fd_dir
), dent
->d_name
, &fdname
);
545 fprintf(stream
, "%s%s:%s\n", fddelim
, dent
->d_name
, fdname
);
548 /* Use the directory entry from /proc/[pid]/fd with /proc/[pid]/fdinfo */
549 fd
= openat(proc_fdinfo_fd
, dent
->d_name
, O_NOFOLLOW
|O_CLOEXEC
|O_RDONLY
);
553 fdinfo
= fdopen(fd
, "r");
560 _cleanup_free_
char *line
= NULL
;
562 r
= read_line(fdinfo
, LONG_LINE_MAX
, &line
);
574 stream
= safe_fclose(stream
);
579 *open_fds
= TAKE_PTR(buffer
);
584 static int get_process_ns(pid_t pid
, const char *namespace, ino_t
*ns
) {
587 _cleanup_close_
int proc_ns_dir_fd
;
589 p
= procfs_file_alloca(pid
, "ns");
591 proc_ns_dir_fd
= open(p
, O_DIRECTORY
| O_CLOEXEC
| O_RDONLY
);
592 if (proc_ns_dir_fd
< 0)
595 if (fstatat(proc_ns_dir_fd
, namespace, &stbuf
, /* flags */0) < 0)
602 static int get_mount_namespace_leader(pid_t pid
, pid_t
*container_pid
) {
603 pid_t cpid
= pid
, ppid
= 0;
607 r
= get_process_ns(pid
, "mnt", &proc_mntns
);
614 r
= get_process_ppid(cpid
, &ppid
);
618 r
= get_process_ns(ppid
, "mnt", &parent_mntns
);
622 if (proc_mntns
!= parent_mntns
)
631 *container_pid
= ppid
;
635 /* Returns 1 if the parent was found.
636 * Returns 0 if there is not a process we can call the pid's
637 * container parent (the pid's process isn't 'containerized').
638 * Returns a negative number on errors.
640 static int get_process_container_parent_cmdline(pid_t pid
, char** cmdline
) {
643 const char *proc_root_path
;
644 struct stat root_stat
, proc_root_stat
;
646 /* To compare inodes of / and /proc/[pid]/root */
647 if (stat("/", &root_stat
) < 0)
650 proc_root_path
= procfs_file_alloca(pid
, "root");
651 if (stat(proc_root_path
, &proc_root_stat
) < 0)
654 /* The process uses system root. */
655 if (proc_root_stat
.st_ino
== root_stat
.st_ino
) {
660 r
= get_mount_namespace_leader(pid
, &container_pid
);
664 r
= get_process_cmdline(container_pid
, SIZE_MAX
, 0, cmdline
);
671 static int change_uid_gid(const char *context
[]) {
676 r
= parse_uid(context
[CONTEXT_UID
], &uid
);
680 if (uid
<= SYSTEM_UID_MAX
) {
681 const char *user
= "systemd-coredump";
683 r
= get_user_creds(&user
, &uid
, &gid
, NULL
, NULL
, 0);
685 log_warning_errno(r
, "Cannot resolve %s user. Proceeding to dump core as root: %m", user
);
689 r
= parse_gid(context
[CONTEXT_GID
], &gid
);
694 return drop_privileges(uid
, gid
, 0);
697 static bool is_journald_crash(const char *context
[_CONTEXT_MAX
]) {
700 return streq_ptr(context
[CONTEXT_UNIT
], SPECIAL_JOURNALD_SERVICE
);
703 static bool is_pid1_crash(const char *context
[_CONTEXT_MAX
]) {
706 return streq_ptr(context
[CONTEXT_UNIT
], SPECIAL_INIT_SCOPE
) ||
707 streq_ptr(context
[CONTEXT_PID
], "1");
710 #define SUBMIT_COREDUMP_FIELDS 4
712 static int submit_coredump(
713 const char *context
[_CONTEXT_MAX
],
715 size_t n_iovec_allocated
,
719 _cleanup_close_
int coredump_fd
= -1, coredump_node_fd
= -1;
720 _cleanup_free_
char *core_message
= NULL
, *filename
= NULL
, *coredump_data
= NULL
;
721 uint64_t coredump_size
= UINT64_MAX
;
722 bool truncated
= false, journald_crash
;
727 assert(n_iovec_allocated
>= n_iovec
+ SUBMIT_COREDUMP_FIELDS
);
728 assert(input_fd
>= 0);
730 journald_crash
= is_journald_crash(context
);
732 /* Vacuum before we write anything again */
733 (void) coredump_vacuum(-1, arg_keep_free
, arg_max_use
);
735 /* Always stream the coredump to disk, if that's possible */
736 r
= save_external_coredump(context
, input_fd
,
737 &filename
, &coredump_node_fd
, &coredump_fd
, &coredump_size
, &truncated
);
739 /* Skip whole core dumping part */
742 /* If we don't want to keep the coredump on disk, remove it now, as later on we will lack the privileges for
743 * it. However, we keep the fd to it, so that we can still process it and log it. */
744 r
= maybe_remove_external_coredump(filename
, coredump_size
);
748 const char *coredump_filename
;
750 coredump_filename
= strjoina("COREDUMP_FILENAME=", filename
);
751 iovec
[n_iovec
++] = IOVEC_MAKE_STRING(coredump_filename
);
752 } else if (arg_storage
== COREDUMP_STORAGE_EXTERNAL
)
753 log_info("The core will not be stored: size %"PRIu64
" is greater than %"PRIu64
" (the configured maximum)",
754 coredump_size
, arg_external_size_max
);
756 /* Vacuum again, but exclude the coredump we just created */
757 (void) coredump_vacuum(coredump_node_fd
>= 0 ? coredump_node_fd
: coredump_fd
, arg_keep_free
, arg_max_use
);
759 /* Now, let's drop privileges to become the user who owns the segfaulted process and allocate the coredump
760 * memory under the user's uid. This also ensures that the credentials journald will see are the ones of the
761 * coredumping user, thus making sure the user gets access to the core dump. Let's also get rid of all
762 * capabilities, if we run as root, we won't need them anymore. */
763 r
= change_uid_gid(context
);
765 return log_error_errno(r
, "Failed to drop privileges: %m");
768 /* Try to get a stack trace if we can */
769 if (coredump_size
<= arg_process_size_max
) {
770 _cleanup_free_
char *stacktrace
= NULL
;
772 r
= coredump_make_stack_trace(coredump_fd
, context
[CONTEXT_EXE
], &stacktrace
);
774 core_message
= strjoin("MESSAGE=Process ", context
[CONTEXT_PID
],
775 " (", context
[CONTEXT_COMM
], ") of user ",
776 context
[CONTEXT_UID
], " dumped core.",
777 journald_crash
? "\nCoredump diverted to " : "",
778 journald_crash
? filename
: "",
780 else if (r
== -EINVAL
)
781 log_warning("Failed to generate stack trace: %s", dwfl_errmsg(dwfl_errno()));
783 log_warning_errno(r
, "Failed to generate stack trace: %m");
785 log_debug("Not generating stack trace: core size %"PRIu64
" is greater than %"PRIu64
" (the configured maximum)",
786 coredump_size
, arg_process_size_max
);
791 core_message
= strjoin("MESSAGE=Process ", context
[CONTEXT_PID
],
792 " (", context
[CONTEXT_COMM
], ") of user ",
793 context
[CONTEXT_UID
], " dumped core.",
794 journald_crash
&& filename
? "\nCoredump diverted to " : NULL
,
795 journald_crash
&& filename
? filename
: NULL
);
799 if (journald_crash
) {
800 /* We cannot log to the journal, so just print the message.
801 * The target was set previously to something safe. */
802 assert(startswith(core_message
, "MESSAGE="));
803 log_dispatch(LOG_ERR
, 0, core_message
+ strlen("MESSAGE="));
807 iovec
[n_iovec
++] = IOVEC_MAKE_STRING(core_message
);
810 iovec
[n_iovec
++] = IOVEC_MAKE_STRING("COREDUMP_TRUNCATED=1");
812 /* Optionally store the entire coredump in the journal */
813 if (arg_storage
== COREDUMP_STORAGE_JOURNAL
) {
814 if (coredump_size
<= arg_journal_size_max
) {
817 /* Store the coredump itself in the journal */
819 r
= allocate_journal_field(coredump_fd
, (size_t) coredump_size
, &coredump_data
, &sz
);
821 iovec
[n_iovec
++] = IOVEC_MAKE(coredump_data
, sz
);
823 log_warning_errno(r
, "Failed to attach the core to the journal entry: %m");
825 log_info("The core will not be stored: size %"PRIu64
" is greater than %"PRIu64
" (the configured maximum)",
826 coredump_size
, arg_journal_size_max
);
829 assert(n_iovec
<= n_iovec_allocated
);
831 r
= sd_journal_sendv(iovec
, n_iovec
);
833 return log_error_errno(r
, "Failed to log coredump: %m");
838 static void map_context_fields(const struct iovec
*iovec
, const char* context
[]) {
840 static const char * const context_field_names
[] = {
841 [CONTEXT_PID
] = "COREDUMP_PID=",
842 [CONTEXT_UID
] = "COREDUMP_UID=",
843 [CONTEXT_GID
] = "COREDUMP_GID=",
844 [CONTEXT_SIGNAL
] = "COREDUMP_SIGNAL=",
845 [CONTEXT_TIMESTAMP
] = "COREDUMP_TIMESTAMP=",
846 [CONTEXT_RLIMIT
] = "COREDUMP_RLIMIT=",
847 [CONTEXT_HOSTNAME
] = "COREDUMP_HOSTNAME=",
848 [CONTEXT_COMM
] = "COREDUMP_COMM=",
849 [CONTEXT_EXE
] = "COREDUMP_EXE=",
857 for (i
= 0; i
< ELEMENTSOF(context_field_names
); i
++) {
860 if (!context_field_names
[i
])
863 p
= memory_startswith(iovec
->iov_base
, iovec
->iov_len
, context_field_names
[i
]);
867 /* Note that these strings are NUL terminated, because we made sure that a trailing NUL byte is in the
868 * buffer, though not included in the iov_len count. (see below) */
874 static int process_socket(int fd
) {
875 _cleanup_close_
int coredump_fd
= -1;
876 struct iovec
*iovec
= NULL
;
877 size_t n_iovec
= 0, n_allocated
= 0, i
, k
;
878 const char *context
[_CONTEXT_MAX
] = {};
885 log_debug("Processing coredump received on stdin...");
889 struct cmsghdr cmsghdr
;
890 uint8_t buf
[CMSG_SPACE(sizeof(int))];
893 .msg_control
= &control
,
894 .msg_controllen
= sizeof(control
),
900 if (!GREEDY_REALLOC(iovec
, n_allocated
, n_iovec
+ SUBMIT_COREDUMP_FIELDS
)) {
905 l
= next_datagram_size_fd(fd
);
907 r
= log_error_errno(l
, "Failed to determine datagram size to read: %m");
913 iovec
[n_iovec
].iov_len
= l
;
914 iovec
[n_iovec
].iov_base
= malloc(l
+ 1);
915 if (!iovec
[n_iovec
].iov_base
) {
920 mh
.msg_iov
= iovec
+ n_iovec
;
922 n
= recvmsg(fd
, &mh
, MSG_CMSG_CLOEXEC
);
924 free(iovec
[n_iovec
].iov_base
);
925 r
= log_error_errno(errno
, "Failed to receive datagram: %m");
930 struct cmsghdr
*cmsg
, *found
= NULL
;
931 /* The final zero-length datagram carries the file descriptor and tells us that we're done. */
933 free(iovec
[n_iovec
].iov_base
);
935 CMSG_FOREACH(cmsg
, &mh
) {
936 if (cmsg
->cmsg_level
== SOL_SOCKET
&&
937 cmsg
->cmsg_type
== SCM_RIGHTS
&&
938 cmsg
->cmsg_len
== CMSG_LEN(sizeof(int))) {
945 log_error("Coredump file descriptor missing.");
950 assert(coredump_fd
< 0);
951 coredump_fd
= *(int*) CMSG_DATA(found
);
955 /* Add trailing NUL byte, in case these are strings */
956 ((char*) iovec
[n_iovec
].iov_base
)[n
] = 0;
957 iovec
[n_iovec
].iov_len
= (size_t) n
;
960 map_context_fields(iovec
+ n_iovec
, context
);
964 if (!GREEDY_REALLOC(iovec
, n_allocated
, n_iovec
+ SUBMIT_COREDUMP_FIELDS
)) {
969 /* Make sure we got all data we really need */
970 assert(context
[CONTEXT_PID
]);
971 assert(context
[CONTEXT_UID
]);
972 assert(context
[CONTEXT_GID
]);
973 assert(context
[CONTEXT_SIGNAL
]);
974 assert(context
[CONTEXT_TIMESTAMP
]);
975 assert(context
[CONTEXT_RLIMIT
]);
976 assert(context
[CONTEXT_HOSTNAME
]);
977 assert(context
[CONTEXT_COMM
]);
978 assert(coredump_fd
>= 0);
980 /* Small quirk: the journal fields contain the timestamp padded with six zeroes, so that the kernel-supplied 1s
981 * granularity timestamps becomes 1µs granularity, i.e. the granularity systemd usually operates in. Since we
982 * are reconstructing the original kernel context, we chop this off again, here. */
983 k
= strlen(context
[CONTEXT_TIMESTAMP
]);
985 context
[CONTEXT_TIMESTAMP
] = strndupa(context
[CONTEXT_TIMESTAMP
], k
- 6);
987 r
= submit_coredump(context
, iovec
, n_allocated
, n_iovec
, coredump_fd
);
990 for (i
= 0; i
< n_iovec
; i
++)
991 free(iovec
[i
].iov_base
);
997 static int send_iovec(const struct iovec iovec
[], size_t n_iovec
, int input_fd
) {
999 static const union sockaddr_union sa
= {
1000 .un
.sun_family
= AF_UNIX
,
1001 .un
.sun_path
= "/run/systemd/coredump",
1003 _cleanup_close_
int fd
= -1;
1007 assert(iovec
|| n_iovec
<= 0);
1008 assert(input_fd
>= 0);
1010 fd
= socket(AF_UNIX
, SOCK_SEQPACKET
|SOCK_CLOEXEC
, 0);
1012 return log_error_errno(errno
, "Failed to create coredump socket: %m");
1014 if (connect(fd
, &sa
.sa
, SOCKADDR_UN_LEN(sa
.un
)) < 0)
1015 return log_error_errno(errno
, "Failed to connect to coredump service: %m");
1017 for (i
= 0; i
< n_iovec
; i
++) {
1018 struct msghdr mh
= {
1019 .msg_iov
= (struct iovec
*) iovec
+ i
,
1022 struct iovec copy
[2];
1025 if (sendmsg(fd
, &mh
, MSG_NOSIGNAL
) >= 0)
1028 if (errno
== EMSGSIZE
&& mh
.msg_iov
[0].iov_len
> 0) {
1029 /* This field didn't fit? That's a pity. Given that this is just metadata,
1030 * let's truncate the field at half, and try again. We append three dots, in
1031 * order to show that this is truncated. */
1033 if (mh
.msg_iov
!= copy
) {
1034 /* We don't want to modify the caller's iovec, hence let's create our
1035 * own array, consisting of two new iovecs, where the first is a
1036 * (truncated) copy of what we want to send, and the second one
1037 * contains the trailing dots. */
1039 copy
[1] = IOVEC_MAKE(((char[]){'.', '.', '.'}), 3);
1045 copy
[0].iov_len
/= 2; /* halve it, and try again */
1049 return log_error_errno(errno
, "Failed to send coredump datagram: %m");
1053 r
= send_one_fd(fd
, input_fd
, 0);
1055 return log_error_errno(r
, "Failed to send coredump fd: %m");
1060 static char* set_iovec_field_free(struct iovec
*iovec
, size_t *n_iovec
, const char *field
, char *value
) {
1063 x
= set_iovec_string_field(iovec
, n_iovec
, field
, value
);
1068 static int gather_pid_metadata(
1069 char* context
[_CONTEXT_MAX
],
1070 char **comm_fallback
,
1071 struct iovec
*iovec
, size_t *n_iovec
) {
1073 /* We need 27 empty slots in iovec!
1075 * Note that if we fail on oom later on, we do not roll-back changes to the iovec structure. (It remains valid,
1076 * with the first n_iovec fields initialized.) */
1084 r
= parse_pid(context
[CONTEXT_PID
], &pid
);
1086 return log_error_errno(r
, "Failed to parse PID \"%s\": %m", context
[CONTEXT_PID
]);
1088 r
= get_process_comm(pid
, &context
[CONTEXT_COMM
]);
1090 log_warning_errno(r
, "Failed to get COMM, falling back to the command line: %m");
1091 context
[CONTEXT_COMM
] = strv_join(comm_fallback
, " ");
1092 if (!context
[CONTEXT_COMM
])
1096 r
= get_process_exe(pid
, &context
[CONTEXT_EXE
]);
1098 log_warning_errno(r
, "Failed to get EXE, ignoring: %m");
1100 if (cg_pid_get_unit(pid
, &context
[CONTEXT_UNIT
]) >= 0) {
1101 if (!is_journald_crash((const char**) context
)) {
1102 /* OK, now we know it's not the journal, hence we can make use of it now. */
1103 log_set_target(LOG_TARGET_JOURNAL_OR_KMSG
);
1107 /* If this is PID 1 disable coredump collection, we'll unlikely be able to process it later on. */
1108 if (is_pid1_crash((const char**) context
)) {
1109 log_notice("Due to PID 1 having crashed coredump collection will now be turned off.");
1110 disable_coredumps();
1113 set_iovec_string_field(iovec
, n_iovec
, "COREDUMP_UNIT=", context
[CONTEXT_UNIT
]);
1116 if (cg_pid_get_user_unit(pid
, &t
) >= 0)
1117 set_iovec_field_free(iovec
, n_iovec
, "COREDUMP_USER_UNIT=", t
);
1119 /* The next few are mandatory */
1120 if (!set_iovec_string_field(iovec
, n_iovec
, "COREDUMP_PID=", context
[CONTEXT_PID
]))
1123 if (!set_iovec_string_field(iovec
, n_iovec
, "COREDUMP_UID=", context
[CONTEXT_UID
]))
1126 if (!set_iovec_string_field(iovec
, n_iovec
, "COREDUMP_GID=", context
[CONTEXT_GID
]))
1129 if (!set_iovec_string_field(iovec
, n_iovec
, "COREDUMP_SIGNAL=", context
[CONTEXT_SIGNAL
]))
1132 if (!set_iovec_string_field(iovec
, n_iovec
, "COREDUMP_RLIMIT=", context
[CONTEXT_RLIMIT
]))
1135 if (!set_iovec_string_field(iovec
, n_iovec
, "COREDUMP_HOSTNAME=", context
[CONTEXT_HOSTNAME
]))
1138 if (!set_iovec_string_field(iovec
, n_iovec
, "COREDUMP_COMM=", context
[CONTEXT_COMM
]))
1141 if (context
[CONTEXT_EXE
] &&
1142 !set_iovec_string_field(iovec
, n_iovec
, "COREDUMP_EXE=", context
[CONTEXT_EXE
]))
1145 if (sd_pid_get_session(pid
, &t
) >= 0)
1146 set_iovec_field_free(iovec
, n_iovec
, "COREDUMP_SESSION=", t
);
1148 if (sd_pid_get_owner_uid(pid
, &owner_uid
) >= 0) {
1149 r
= asprintf(&t
, "COREDUMP_OWNER_UID=" UID_FMT
, owner_uid
);
1151 iovec
[(*n_iovec
)++] = IOVEC_MAKE_STRING(t
);
1154 if (sd_pid_get_slice(pid
, &t
) >= 0)
1155 set_iovec_field_free(iovec
, n_iovec
, "COREDUMP_SLICE=", t
);
1157 if (get_process_cmdline(pid
, SIZE_MAX
, 0, &t
) >= 0)
1158 set_iovec_field_free(iovec
, n_iovec
, "COREDUMP_CMDLINE=", t
);
1160 if (cg_pid_get_path_shifted(pid
, NULL
, &t
) >= 0)
1161 set_iovec_field_free(iovec
, n_iovec
, "COREDUMP_CGROUP=", t
);
1163 if (compose_open_fds(pid
, &t
) >= 0)
1164 set_iovec_field_free(iovec
, n_iovec
, "COREDUMP_OPEN_FDS=", t
);
1166 p
= procfs_file_alloca(pid
, "status");
1167 if (read_full_file(p
, &t
, NULL
) >= 0)
1168 set_iovec_field_free(iovec
, n_iovec
, "COREDUMP_PROC_STATUS=", t
);
1170 p
= procfs_file_alloca(pid
, "maps");
1171 if (read_full_file(p
, &t
, NULL
) >= 0)
1172 set_iovec_field_free(iovec
, n_iovec
, "COREDUMP_PROC_MAPS=", t
);
1174 p
= procfs_file_alloca(pid
, "limits");
1175 if (read_full_file(p
, &t
, NULL
) >= 0)
1176 set_iovec_field_free(iovec
, n_iovec
, "COREDUMP_PROC_LIMITS=", t
);
1178 p
= procfs_file_alloca(pid
, "cgroup");
1179 if (read_full_file(p
, &t
, NULL
) >=0)
1180 set_iovec_field_free(iovec
, n_iovec
, "COREDUMP_PROC_CGROUP=", t
);
1182 p
= procfs_file_alloca(pid
, "mountinfo");
1183 if (read_full_file(p
, &t
, NULL
) >=0)
1184 set_iovec_field_free(iovec
, n_iovec
, "COREDUMP_PROC_MOUNTINFO=", t
);
1186 if (get_process_cwd(pid
, &t
) >= 0)
1187 set_iovec_field_free(iovec
, n_iovec
, "COREDUMP_CWD=", t
);
1189 if (get_process_root(pid
, &t
) >= 0) {
1190 bool proc_self_root_is_slash
;
1192 proc_self_root_is_slash
= strcmp(t
, "/") == 0;
1194 set_iovec_field_free(iovec
, n_iovec
, "COREDUMP_ROOT=", t
);
1196 /* If the process' root is "/", then there is a chance it has
1197 * mounted own root and hence being containerized. */
1198 if (proc_self_root_is_slash
&& get_process_container_parent_cmdline(pid
, &t
) > 0)
1199 set_iovec_field_free(iovec
, n_iovec
, "COREDUMP_CONTAINER_CMDLINE=", t
);
1202 if (get_process_environ(pid
, &t
) >= 0)
1203 set_iovec_field_free(iovec
, n_iovec
, "COREDUMP_ENVIRON=", t
);
1205 t
= strjoin("COREDUMP_TIMESTAMP=", context
[CONTEXT_TIMESTAMP
], "000000");
1207 iovec
[(*n_iovec
)++] = IOVEC_MAKE_STRING(t
);
1209 if (safe_atoi(context
[CONTEXT_SIGNAL
], &signo
) >= 0 && SIGNAL_VALID(signo
))
1210 set_iovec_string_field(iovec
, n_iovec
, "COREDUMP_SIGNAL_NAME=SIG", signal_to_string(signo
));
1212 return 0; /* we successfully acquired all metadata */
1215 static int process_kernel(int argc
, char* argv
[]) {
1217 char* context
[_CONTEXT_MAX
] = {};
1218 struct iovec iovec
[29 + SUBMIT_COREDUMP_FIELDS
];
1219 size_t i
, n_iovec
, n_to_free
= 0;
1222 log_debug("Processing coredump received from the kernel...");
1224 if (argc
< CONTEXT_COMM
+ 1)
1225 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1226 "Not enough arguments passed by the kernel (%i, expected %i).",
1227 argc
- 1, CONTEXT_COMM
+ 1 - 1);
1229 context
[CONTEXT_PID
] = argv
[1 + CONTEXT_PID
];
1230 context
[CONTEXT_UID
] = argv
[1 + CONTEXT_UID
];
1231 context
[CONTEXT_GID
] = argv
[1 + CONTEXT_GID
];
1232 context
[CONTEXT_SIGNAL
] = argv
[1 + CONTEXT_SIGNAL
];
1233 context
[CONTEXT_TIMESTAMP
] = argv
[1 + CONTEXT_TIMESTAMP
];
1234 context
[CONTEXT_RLIMIT
] = argv
[1 + CONTEXT_RLIMIT
];
1235 context
[CONTEXT_HOSTNAME
] = argv
[1 + CONTEXT_HOSTNAME
];
1237 r
= gather_pid_metadata(context
, argv
+ 1 + CONTEXT_COMM
, iovec
, &n_to_free
);
1241 n_iovec
= n_to_free
;
1243 iovec
[n_iovec
++] = IOVEC_MAKE_STRING("MESSAGE_ID=" SD_MESSAGE_COREDUMP_STR
);
1245 assert_cc(2 == LOG_CRIT
);
1246 iovec
[n_iovec
++] = IOVEC_MAKE_STRING("PRIORITY=2");
1248 assert(n_iovec
<= ELEMENTSOF(iovec
));
1250 if (is_journald_crash((const char**) context
) || is_pid1_crash((const char**) context
))
1251 r
= submit_coredump((const char**) context
,
1252 iovec
, ELEMENTSOF(iovec
), n_iovec
,
1255 r
= send_iovec(iovec
, n_iovec
, STDIN_FILENO
);
1258 for (i
= 0; i
< n_to_free
; i
++)
1259 free(iovec
[i
].iov_base
);
1261 /* Those fields are allocated by gather_pid_metadata */
1262 free(context
[CONTEXT_COMM
]);
1263 free(context
[CONTEXT_EXE
]);
1264 free(context
[CONTEXT_UNIT
]);
1269 static int process_backtrace(int argc
, char *argv
[]) {
1270 char *context
[_CONTEXT_MAX
] = {};
1271 _cleanup_free_
char *message
= NULL
;
1272 _cleanup_free_
struct iovec
*iovec
= NULL
;
1273 size_t n_iovec
, n_allocated
, n_to_free
= 0, i
;
1275 JournalImporter importer
= {
1279 log_debug("Processing backtrace on stdin...");
1281 if (argc
< CONTEXT_COMM
+ 1)
1282 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1283 "Not enough arguments passed (%i, expected %i).",
1284 argc
- 1, CONTEXT_COMM
+ 1 - 1);
1286 context
[CONTEXT_PID
] = argv
[2 + CONTEXT_PID
];
1287 context
[CONTEXT_UID
] = argv
[2 + CONTEXT_UID
];
1288 context
[CONTEXT_GID
] = argv
[2 + CONTEXT_GID
];
1289 context
[CONTEXT_SIGNAL
] = argv
[2 + CONTEXT_SIGNAL
];
1290 context
[CONTEXT_TIMESTAMP
] = argv
[2 + CONTEXT_TIMESTAMP
];
1291 context
[CONTEXT_RLIMIT
] = argv
[2 + CONTEXT_RLIMIT
];
1292 context
[CONTEXT_HOSTNAME
] = argv
[2 + CONTEXT_HOSTNAME
];
1294 n_allocated
= 34 + COREDUMP_STORAGE_EXTERNAL
;
1295 /* 26 metadata, 2 static, +unknown input, 4 storage, rounded up */
1296 iovec
= new(struct iovec
, n_allocated
);
1300 r
= gather_pid_metadata(context
, argv
+ 2 + CONTEXT_COMM
, iovec
, &n_to_free
);
1304 /* This was a special crash, and has already been processed. */
1308 n_iovec
= n_to_free
;
1311 r
= journal_importer_process_data(&importer
);
1313 log_error_errno(r
, "Failed to parse journal entry on stdin: %m");
1316 if (r
== 1 || /* complete entry */
1317 journal_importer_eof(&importer
)) /* end of data */
1321 if (!GREEDY_REALLOC(iovec
, n_allocated
, n_iovec
+ importer
.iovw
.count
+ 2))
1324 if (journal_importer_eof(&importer
)) {
1325 log_warning("Did not receive a full journal entry on stdin, ignoring message sent by reporter");
1327 message
= strjoin("MESSAGE=Process ", context
[CONTEXT_PID
],
1328 " (", context
[CONTEXT_COMM
], ")"
1329 " of user ", context
[CONTEXT_UID
],
1330 " failed with ", context
[CONTEXT_SIGNAL
]);
1335 iovec
[n_iovec
++] = IOVEC_MAKE_STRING(message
);
1337 for (i
= 0; i
< importer
.iovw
.count
; i
++)
1338 iovec
[n_iovec
++] = importer
.iovw
.iovec
[i
];
1341 iovec
[n_iovec
++] = IOVEC_MAKE_STRING("MESSAGE_ID=" SD_MESSAGE_BACKTRACE_STR
);
1342 assert_cc(2 == LOG_CRIT
);
1343 iovec
[n_iovec
++] = IOVEC_MAKE_STRING("PRIORITY=2");
1345 assert(n_iovec
<= n_allocated
);
1347 r
= sd_journal_sendv(iovec
, n_iovec
);
1349 log_error_errno(r
, "Failed to log backtrace: %m");
1352 for (i
= 0; i
< n_to_free
; i
++)
1353 free(iovec
[i
].iov_base
);
1355 /* Those fields are allocated by gather_pid_metadata */
1356 free(context
[CONTEXT_COMM
]);
1357 free(context
[CONTEXT_EXE
]);
1358 free(context
[CONTEXT_UNIT
]);
1363 static int run(int argc
, char *argv
[]) {
1366 /* First, log to a safe place, since we don't know what crashed and it might
1367 * be journald which we'd rather not log to then. */
1369 log_set_target(LOG_TARGET_KMSG
);
1372 /* Make sure we never enter a loop */
1373 (void) prctl(PR_SET_DUMPABLE
, 0);
1375 /* Ignore all parse errors */
1376 (void) parse_config();
1378 log_debug("Selected storage '%s'.", coredump_storage_to_string(arg_storage
));
1379 log_debug("Selected compression %s.", yes_no(arg_compress
));
1381 r
= sd_listen_fds(false);
1383 return log_error_errno(r
, "Failed to determine the number of file descriptors: %m");
1385 /* If we got an fd passed, we are running in coredumpd mode. Otherwise we
1386 * are invoked from the kernel as coredump handler. */
1388 if (streq_ptr(argv
[1], "--backtrace"))
1389 return process_backtrace(argc
, argv
);
1391 return process_kernel(argc
, argv
);
1393 return process_socket(SD_LISTEN_FDS_START
);
1395 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1396 "Received unexpected number of file descriptors.");
1399 DEFINE_MAIN_FUNCTION(run
);