2 This file is part of systemd.
4 Copyright 2012 Lennart Poettering
6 systemd is free software; you can redistribute it and/or modify it
7 under the terms of the GNU Lesser General Public License as published by
8 the Free Software Foundation; either version 2.1 of the License, or
9 (at your option) any later version.
11 systemd is distributed in the hope that it will be useful, but
12 WITHOUT ANY WARRANTY; without even the implied warranty of
13 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14 Lesser General Public License for more details.
16 You should have received a copy of the GNU Lesser General Public License
17 along with systemd; If not, see <http://www.gnu.org/licenses/>.
22 #include <sys/prctl.h>
23 #include <sys/xattr.h>
28 #include <elfutils/libdwfl.h>
31 #include "sd-daemon.h"
32 #include "sd-journal.h"
34 #include "sd-messages.h"
37 #include "alloc-util.h"
38 #include "capability-util.h"
39 #include "cgroup-util.h"
41 #include "conf-parser.h"
43 #include "coredump-vacuum.h"
44 #include "dirent-util.h"
50 #include "journal-importer.h"
55 #include "parse-util.h"
56 #include "process-util.h"
57 #include "signal-util.h"
58 #include "socket-util.h"
60 #include "stacktrace.h"
61 #include "string-table.h"
62 #include "string-util.h"
64 #include "user-util.h"
67 /* The maximum size up to which we process coredumps */
68 #define PROCESS_SIZE_MAX ((uint64_t) (2LLU*1024LLU*1024LLU*1024LLU))
70 /* The maximum size up to which we leave the coredump around on disk */
71 #define EXTERNAL_SIZE_MAX PROCESS_SIZE_MAX
73 /* The maximum size up to which we store the coredump in the journal */
74 #define JOURNAL_SIZE_MAX ((size_t) (767LU*1024LU*1024LU))
76 /* Make sure to not make this larger than the maximum journal entry
77 * size. See DATA_SIZE_MAX in journald-native.c. */
78 assert_cc(JOURNAL_SIZE_MAX
<= DATA_SIZE_MAX
);
81 /* We use this as array indexes for a couple of special fields we use for
82 * naming coredump files, and attaching xattrs, and for indexing argv[].
84 * Our pattern for man:systectl(1) kernel.core_pattern is such that the
85 * kernel passes fields until CONTEXT_RLIMIT as arguments in argv[]. After
86 * that it gets complicated: the kernel passes "comm" as one or more fields
87 * starting at index CONTEXT_COMM (in other words, full "comm" is under index
88 * CONTEXT_COMM when it does not contain spaces, which is the common
89 * case). This mapping is not reversible, so we prefer to retrieve "comm"
90 * from /proc. We only fall back to argv[CONTEXT_COMM...] when that fails.
92 * In the internal context[] array, fields before CONTEXT_COMM are the
93 * strings from argv[], so they should not be freed. The strings at indices
94 * CONTEXT_COMM and higher are allocated by us and should be freed at the
108 typedef enum CoredumpStorage
{
109 COREDUMP_STORAGE_NONE
,
110 COREDUMP_STORAGE_EXTERNAL
,
111 COREDUMP_STORAGE_JOURNAL
,
112 _COREDUMP_STORAGE_MAX
,
113 _COREDUMP_STORAGE_INVALID
= -1
116 static const char* const coredump_storage_table
[_COREDUMP_STORAGE_MAX
] = {
117 [COREDUMP_STORAGE_NONE
] = "none",
118 [COREDUMP_STORAGE_EXTERNAL
] = "external",
119 [COREDUMP_STORAGE_JOURNAL
] = "journal",
122 DEFINE_PRIVATE_STRING_TABLE_LOOKUP(coredump_storage
, CoredumpStorage
);
123 static DEFINE_CONFIG_PARSE_ENUM(config_parse_coredump_storage
, coredump_storage
, CoredumpStorage
, "Failed to parse storage setting");
125 static CoredumpStorage arg_storage
= COREDUMP_STORAGE_EXTERNAL
;
126 static bool arg_compress
= true;
127 static uint64_t arg_process_size_max
= PROCESS_SIZE_MAX
;
128 static uint64_t arg_external_size_max
= EXTERNAL_SIZE_MAX
;
129 static uint64_t arg_journal_size_max
= JOURNAL_SIZE_MAX
;
130 static uint64_t arg_keep_free
= (uint64_t) -1;
131 static uint64_t arg_max_use
= (uint64_t) -1;
133 static int parse_config(void) {
134 static const ConfigTableItem items
[] = {
135 { "Coredump", "Storage", config_parse_coredump_storage
, 0, &arg_storage
},
136 { "Coredump", "Compress", config_parse_bool
, 0, &arg_compress
},
137 { "Coredump", "ProcessSizeMax", config_parse_iec_uint64
, 0, &arg_process_size_max
},
138 { "Coredump", "ExternalSizeMax", config_parse_iec_uint64
, 0, &arg_external_size_max
},
139 { "Coredump", "JournalSizeMax", config_parse_iec_size
, 0, &arg_journal_size_max
},
140 { "Coredump", "KeepFree", config_parse_iec_uint64
, 0, &arg_keep_free
},
141 { "Coredump", "MaxUse", config_parse_iec_uint64
, 0, &arg_max_use
},
145 return config_parse_many_nulstr(PKGSYSCONFDIR
"/coredump.conf",
146 CONF_PATHS_NULSTR("systemd/coredump.conf.d"),
148 config_item_table_lookup
, items
,
152 static inline uint64_t storage_size_max(void) {
153 return arg_storage
== COREDUMP_STORAGE_EXTERNAL
? arg_external_size_max
: arg_journal_size_max
;
156 static int fix_acl(int fd
, uid_t uid
) {
159 _cleanup_(acl_freep
) acl_t acl
= NULL
;
161 acl_permset_t permset
;
166 if (uid
<= SYSTEM_UID_MAX
)
169 /* Make sure normal users can read (but not write or delete)
170 * their own coredumps */
172 acl
= acl_get_fd(fd
);
174 return log_error_errno(errno
, "Failed to get ACL: %m");
176 if (acl_create_entry(&acl
, &entry
) < 0 ||
177 acl_set_tag_type(entry
, ACL_USER
) < 0 ||
178 acl_set_qualifier(entry
, &uid
) < 0)
179 return log_error_errno(errno
, "Failed to patch ACL: %m");
181 if (acl_get_permset(entry
, &permset
) < 0 ||
182 acl_add_perm(permset
, ACL_READ
) < 0)
183 return log_warning_errno(errno
, "Failed to patch ACL: %m");
185 r
= calc_acl_mask_if_needed(&acl
);
187 return log_warning_errno(r
, "Failed to patch ACL: %m");
189 if (acl_set_fd(fd
, acl
) < 0)
190 return log_error_errno(errno
, "Failed to apply ACL: %m");
196 static int fix_xattr(int fd
, const char *context
[_CONTEXT_MAX
]) {
198 static const char * const xattrs
[_CONTEXT_MAX
] = {
199 [CONTEXT_PID
] = "user.coredump.pid",
200 [CONTEXT_UID
] = "user.coredump.uid",
201 [CONTEXT_GID
] = "user.coredump.gid",
202 [CONTEXT_SIGNAL
] = "user.coredump.signal",
203 [CONTEXT_TIMESTAMP
] = "user.coredump.timestamp",
204 [CONTEXT_RLIMIT
] = "user.coredump.rlimit",
205 [CONTEXT_COMM
] = "user.coredump.comm",
206 [CONTEXT_EXE
] = "user.coredump.exe",
214 /* Attach some metadata to coredumps via extended
215 * attributes. Just because we can. */
217 for (i
= 0; i
< _CONTEXT_MAX
; i
++) {
220 if (isempty(context
[i
]) || !xattrs
[i
])
223 k
= fsetxattr(fd
, xattrs
[i
], context
[i
], strlen(context
[i
]), XATTR_CREATE
);
231 #define filename_escape(s) xescape((s), "./ ")
233 static inline const char *coredump_tmpfile_name(const char *s
) {
234 return s
? s
: "(unnamed temporary file)";
237 static int fix_permissions(
239 const char *filename
,
241 const char *context
[_CONTEXT_MAX
],
250 /* Ignore errors on these */
251 (void) fchmod(fd
, 0640);
252 (void) fix_acl(fd
, uid
);
253 (void) fix_xattr(fd
, context
);
256 return log_error_errno(errno
, "Failed to sync coredump %s: %m", coredump_tmpfile_name(filename
));
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
],
327 uint64_t *ret_size
) {
329 _cleanup_free_
char *fn
= NULL
, *tmp
= NULL
;
330 _cleanup_close_
int fd
= -1;
331 uint64_t rlimit
, max_size
;
337 assert(ret_filename
);
342 r
= parse_uid(context
[CONTEXT_UID
], &uid
);
344 return log_error_errno(r
, "Failed to parse UID: %m");
346 r
= safe_atou64(context
[CONTEXT_RLIMIT
], &rlimit
);
348 return log_error_errno(r
, "Failed to parse resource limit: %s", context
[CONTEXT_RLIMIT
]);
349 if (rlimit
< page_size()) {
350 /* Is coredumping disabled? Then don't bother saving/processing the coredump.
351 * Anything below PAGE_SIZE cannot give a readable coredump (the kernel uses
352 * ELF_EXEC_PAGESIZE which is not easily accessible, but is usually the same as PAGE_SIZE. */
353 log_info("Resource limits disable core dumping for process %s (%s).",
354 context
[CONTEXT_PID
], context
[CONTEXT_COMM
]);
358 /* Never store more than the process configured, or than we actually shall keep or process */
359 max_size
= MIN(rlimit
, MAX(arg_process_size_max
, storage_size_max()));
361 r
= make_filename(context
, &fn
);
363 return log_error_errno(r
, "Failed to determine coredump file name: %m");
365 mkdir_p_label("/var/lib/systemd/coredump", 0755);
367 fd
= open_tmpfile_linkable(fn
, O_RDWR
|O_CLOEXEC
, &tmp
);
369 return log_error_errno(fd
, "Failed to create temporary file for coredump %s: %m", fn
);
371 r
= copy_bytes(input_fd
, fd
, max_size
, 0);
373 log_error_errno(r
, "Cannot store coredump of %s (%s): %m", context
[CONTEXT_PID
], context
[CONTEXT_COMM
]);
377 LOG_MESSAGE("Core file was truncated to %zu bytes.", max_size
),
378 "SIZE_LIMIT=%zu", max_size
,
379 "MESSAGE_ID=" SD_MESSAGE_TRUNCATED_CORE_STR
,
382 if (fstat(fd
, &st
) < 0) {
383 log_error_errno(errno
, "Failed to fstat core file %s: %m", coredump_tmpfile_name(tmp
));
387 if (lseek(fd
, 0, SEEK_SET
) == (off_t
) -1) {
388 log_error_errno(errno
, "Failed to seek on %s: %m", coredump_tmpfile_name(tmp
));
392 #if defined(HAVE_XZ) || defined(HAVE_LZ4)
393 /* If we will remove the coredump anyway, do not compress. */
394 if (arg_compress
&& !maybe_remove_external_coredump(NULL
, st
.st_size
)) {
396 _cleanup_free_
char *fn_compressed
= NULL
, *tmp_compressed
= NULL
;
397 _cleanup_close_
int fd_compressed
= -1;
399 fn_compressed
= strappend(fn
, COMPRESSED_EXT
);
400 if (!fn_compressed
) {
405 fd_compressed
= open_tmpfile_linkable(fn_compressed
, O_RDWR
|O_CLOEXEC
, &tmp_compressed
);
406 if (fd_compressed
< 0) {
407 log_error_errno(fd_compressed
, "Failed to create temporary file for coredump %s: %m", fn_compressed
);
411 r
= compress_stream(fd
, fd_compressed
, -1);
413 log_error_errno(r
, "Failed to compress %s: %m", coredump_tmpfile_name(tmp_compressed
));
414 goto fail_compressed
;
417 r
= fix_permissions(fd_compressed
, tmp_compressed
, fn_compressed
, context
, uid
);
419 goto fail_compressed
;
421 /* OK, this worked, we can get rid of the uncompressed version now */
425 *ret_filename
= fn_compressed
; /* compressed */
426 *ret_node_fd
= fd_compressed
; /* compressed */
427 *ret_data_fd
= fd
; /* uncompressed */
428 *ret_size
= (uint64_t) st
.st_size
; /* uncompressed */
430 fn_compressed
= NULL
;
431 fd
= fd_compressed
= -1;
437 (void) unlink(tmp_compressed
);
443 r
= fix_permissions(fd
, tmp
, fn
, context
, uid
);
450 *ret_size
= (uint64_t) st
.st_size
;
463 static int allocate_journal_field(int fd
, size_t size
, char **ret
, size_t *ret_size
) {
464 _cleanup_free_
char *field
= NULL
;
471 if (lseek(fd
, 0, SEEK_SET
) == (off_t
) -1)
472 return log_warning_errno(errno
, "Failed to seek: %m");
474 field
= malloc(9 + size
);
476 log_warning("Failed to allocate memory for coredump, coredump will not be stored.");
480 memcpy(field
, "COREDUMP=", 9);
482 n
= read(fd
, field
+ 9, size
);
484 return log_error_errno((int) n
, "Failed to read core data: %m");
485 if ((size_t) n
< size
) {
486 log_error("Core data too short.");
491 *ret_size
= size
+ 9;
498 /* Joins /proc/[pid]/fd/ and /proc/[pid]/fdinfo/ into the following lines:
512 static int compose_open_fds(pid_t pid
, char **open_fds
) {
513 _cleanup_closedir_
DIR *proc_fd_dir
= NULL
;
514 _cleanup_close_
int proc_fdinfo_fd
= -1;
515 _cleanup_free_
char *buffer
= NULL
;
516 _cleanup_fclose_
FILE *stream
= NULL
;
517 const char *fddelim
= "", *path
;
518 struct dirent
*dent
= NULL
;
523 assert(open_fds
!= NULL
);
525 path
= procfs_file_alloca(pid
, "fd");
526 proc_fd_dir
= opendir(path
);
530 proc_fdinfo_fd
= openat(dirfd(proc_fd_dir
), "../fdinfo", O_DIRECTORY
|O_NOFOLLOW
|O_CLOEXEC
|O_PATH
);
531 if (proc_fdinfo_fd
< 0)
534 stream
= open_memstream(&buffer
, &size
);
538 FOREACH_DIRENT(dent
, proc_fd_dir
, return -errno
) {
539 _cleanup_fclose_
FILE *fdinfo
= NULL
;
540 _cleanup_free_
char *fdname
= NULL
;
544 r
= readlinkat_malloc(dirfd(proc_fd_dir
), dent
->d_name
, &fdname
);
548 fprintf(stream
, "%s%s:%s\n", fddelim
, dent
->d_name
, fdname
);
551 /* Use the directory entry from /proc/[pid]/fd with /proc/[pid]/fdinfo */
552 fd
= openat(proc_fdinfo_fd
, dent
->d_name
, O_NOFOLLOW
|O_CLOEXEC
|O_RDONLY
);
556 fdinfo
= fdopen(fd
, "re");
557 if (fdinfo
== NULL
) {
562 FOREACH_LINE(line
, fdinfo
, break) {
564 if (!endswith(line
, "\n"))
570 stream
= safe_fclose(stream
);
581 static int get_process_ns(pid_t pid
, const char *namespace, ino_t
*ns
) {
584 _cleanup_close_
int proc_ns_dir_fd
;
586 p
= procfs_file_alloca(pid
, "ns");
588 proc_ns_dir_fd
= open(p
, O_DIRECTORY
| O_CLOEXEC
| O_RDONLY
);
589 if (proc_ns_dir_fd
< 0)
592 if (fstatat(proc_ns_dir_fd
, namespace, &stbuf
, /* flags */0) < 0)
599 static int get_mount_namespace_leader(pid_t pid
, pid_t
*container_pid
) {
600 pid_t cpid
= pid
, ppid
= 0;
604 r
= get_process_ns(pid
, "mnt", &proc_mntns
);
611 r
= get_process_ppid(cpid
, &ppid
);
615 r
= get_process_ns(ppid
, "mnt", &parent_mntns
);
619 if (proc_mntns
!= parent_mntns
)
628 *container_pid
= ppid
;
632 /* Returns 1 if the parent was found.
633 * Returns 0 if there is not a process we can call the pid's
634 * container parent (the pid's process isn't 'containerized').
635 * Returns a negative number on errors.
637 static int get_process_container_parent_cmdline(pid_t pid
, char** cmdline
) {
640 const char *proc_root_path
;
641 struct stat root_stat
, proc_root_stat
;
643 /* To compare inodes of / and /proc/[pid]/root */
644 if (stat("/", &root_stat
) < 0)
647 proc_root_path
= procfs_file_alloca(pid
, "root");
648 if (stat(proc_root_path
, &proc_root_stat
) < 0)
651 /* The process uses system root. */
652 if (proc_root_stat
.st_ino
== root_stat
.st_ino
) {
657 r
= get_mount_namespace_leader(pid
, &container_pid
);
661 r
= get_process_cmdline(container_pid
, 0, false, cmdline
);
668 static int change_uid_gid(const char *context
[]) {
673 r
= parse_uid(context
[CONTEXT_UID
], &uid
);
677 if (uid
<= SYSTEM_UID_MAX
) {
678 const char *user
= "systemd-coredump";
680 r
= get_user_creds(&user
, &uid
, &gid
, NULL
, NULL
);
682 log_warning_errno(r
, "Cannot resolve %s user. Proceeding to dump core as root: %m", user
);
686 r
= parse_gid(context
[CONTEXT_GID
], &gid
);
691 return drop_privileges(uid
, gid
, 0);
694 static int submit_coredump(
695 const char *context
[_CONTEXT_MAX
],
697 size_t n_iovec_allocated
,
701 _cleanup_close_
int coredump_fd
= -1, coredump_node_fd
= -1;
702 _cleanup_free_
char *core_message
= NULL
, *filename
= NULL
, *coredump_data
= NULL
;
703 uint64_t coredump_size
= UINT64_MAX
;
708 assert(n_iovec_allocated
>= n_iovec
+ 3);
709 assert(input_fd
>= 0);
711 /* Vacuum before we write anything again */
712 (void) coredump_vacuum(-1, arg_keep_free
, arg_max_use
);
714 /* Always stream the coredump to disk, if that's possible */
715 r
= save_external_coredump(context
, input_fd
, &filename
, &coredump_node_fd
, &coredump_fd
, &coredump_size
);
717 /* Skip whole core dumping part */
720 /* If we don't want to keep the coredump on disk, remove it now, as later on we will lack the privileges for
721 * it. However, we keep the fd to it, so that we can still process it and log it. */
722 r
= maybe_remove_external_coredump(filename
, coredump_size
);
726 const char *coredump_filename
;
728 coredump_filename
= strjoina("COREDUMP_FILENAME=", filename
);
729 IOVEC_SET_STRING(iovec
[n_iovec
++], coredump_filename
);
730 } else if (arg_storage
== COREDUMP_STORAGE_EXTERNAL
)
731 log_info("The core will not be stored: size %"PRIu64
" is greater than %"PRIu64
" (the configured maximum)",
732 coredump_size
, arg_external_size_max
);
734 /* Vacuum again, but exclude the coredump we just created */
735 (void) coredump_vacuum(coredump_node_fd
>= 0 ? coredump_node_fd
: coredump_fd
, arg_keep_free
, arg_max_use
);
737 /* Now, let's drop privileges to become the user who owns the segfaulted process and allocate the coredump
738 * memory under the user's uid. This also ensures that the credentials journald will see are the ones of the
739 * coredumping user, thus making sure the user gets access to the core dump. Let's also get rid of all
740 * capabilities, if we run as root, we won't need them anymore. */
741 r
= change_uid_gid(context
);
743 return log_error_errno(r
, "Failed to drop privileges: %m");
746 /* Try to get a strack trace if we can */
747 if (coredump_size
<= arg_process_size_max
) {
748 _cleanup_free_
char *stacktrace
= NULL
;
750 r
= coredump_make_stack_trace(coredump_fd
, context
[CONTEXT_EXE
], &stacktrace
);
752 core_message
= strjoin("MESSAGE=Process ", context
[CONTEXT_PID
],
753 " (", context
[CONTEXT_COMM
], ") of user ",
754 context
[CONTEXT_UID
], " dumped core.\n\n",
756 else if (r
== -EINVAL
)
757 log_warning("Failed to generate stack trace: %s", dwfl_errmsg(dwfl_errno()));
759 log_warning_errno(r
, "Failed to generate stack trace: %m");
761 log_debug("Not generating stack trace: core size %"PRIu64
" is greater than %"PRIu64
" (the configured maximum)",
762 coredump_size
, arg_process_size_max
);
767 core_message
= strjoin("MESSAGE=Process ", context
[CONTEXT_PID
], " (",
768 context
[CONTEXT_COMM
], ") of user ",
769 context
[CONTEXT_UID
], " dumped core.");
771 IOVEC_SET_STRING(iovec
[n_iovec
++], core_message
);
773 /* Optionally store the entire coredump in the journal */
774 if (arg_storage
== COREDUMP_STORAGE_JOURNAL
) {
775 if (coredump_size
<= arg_journal_size_max
) {
778 /* Store the coredump itself in the journal */
780 r
= allocate_journal_field(coredump_fd
, (size_t) coredump_size
, &coredump_data
, &sz
);
782 iovec
[n_iovec
].iov_base
= coredump_data
;
783 iovec
[n_iovec
].iov_len
= sz
;
786 log_warning_errno(r
, "Failed to attach the core to the journal entry: %m");
788 log_info("The core will not be stored: size %"PRIu64
" is greater than %"PRIu64
" (the configured maximum)",
789 coredump_size
, arg_journal_size_max
);
792 assert(n_iovec
<= n_iovec_allocated
);
794 r
= sd_journal_sendv(iovec
, n_iovec
);
796 return log_error_errno(r
, "Failed to log coredump: %m");
801 static void map_context_fields(const struct iovec
*iovec
, const char* context
[]) {
803 static const char * const context_field_names
[_CONTEXT_MAX
] = {
804 [CONTEXT_PID
] = "COREDUMP_PID=",
805 [CONTEXT_UID
] = "COREDUMP_UID=",
806 [CONTEXT_GID
] = "COREDUMP_GID=",
807 [CONTEXT_SIGNAL
] = "COREDUMP_SIGNAL=",
808 [CONTEXT_TIMESTAMP
] = "COREDUMP_TIMESTAMP=",
809 [CONTEXT_COMM
] = "COREDUMP_COMM=",
810 [CONTEXT_EXE
] = "COREDUMP_EXE=",
811 [CONTEXT_RLIMIT
] = "COREDUMP_RLIMIT=",
819 for (i
= 0; i
< _CONTEXT_MAX
; i
++) {
822 l
= strlen(context_field_names
[i
]);
823 if (iovec
->iov_len
< l
)
826 if (memcmp(iovec
->iov_base
, context_field_names
[i
], l
) != 0)
829 /* Note that these strings are NUL terminated, because we made sure that a trailing NUL byte is in the
830 * buffer, though not included in the iov_len count. (see below) */
831 context
[i
] = (char*) iovec
->iov_base
+ l
;
836 static int process_socket(int fd
) {
837 _cleanup_close_
int coredump_fd
= -1;
838 struct iovec
*iovec
= NULL
;
839 size_t n_iovec
= 0, n_allocated
= 0, i
, k
;
840 const char *context
[_CONTEXT_MAX
] = {};
845 log_set_target(LOG_TARGET_AUTO
);
846 log_parse_environment();
849 log_debug("Processing coredump received on stdin...");
853 struct cmsghdr cmsghdr
;
854 uint8_t buf
[CMSG_SPACE(sizeof(int))];
857 .msg_control
= &control
,
858 .msg_controllen
= sizeof(control
),
864 if (!GREEDY_REALLOC(iovec
, n_allocated
, n_iovec
+ 3)) {
869 l
= next_datagram_size_fd(fd
);
871 r
= log_error_errno(l
, "Failed to determine datagram size to read: %m");
877 iovec
[n_iovec
].iov_len
= l
;
878 iovec
[n_iovec
].iov_base
= malloc(l
+ 1);
879 if (!iovec
[n_iovec
].iov_base
) {
884 mh
.msg_iov
= iovec
+ n_iovec
;
886 n
= recvmsg(fd
, &mh
, MSG_NOSIGNAL
|MSG_CMSG_CLOEXEC
);
888 free(iovec
[n_iovec
].iov_base
);
889 r
= log_error_errno(errno
, "Failed to receive datagram: %m");
894 struct cmsghdr
*cmsg
, *found
= NULL
;
895 /* The final zero-length datagram carries the file descriptor and tells us that we're done. */
897 free(iovec
[n_iovec
].iov_base
);
899 CMSG_FOREACH(cmsg
, &mh
) {
900 if (cmsg
->cmsg_level
== SOL_SOCKET
&&
901 cmsg
->cmsg_type
== SCM_RIGHTS
&&
902 cmsg
->cmsg_len
== CMSG_LEN(sizeof(int))) {
909 log_error("Coredump file descriptor missing.");
914 assert(coredump_fd
< 0);
915 coredump_fd
= *(int*) CMSG_DATA(found
);
919 /* Add trailing NUL byte, in case these are strings */
920 ((char*) iovec
[n_iovec
].iov_base
)[n
] = 0;
921 iovec
[n_iovec
].iov_len
= (size_t) n
;
924 map_context_fields(iovec
+ n_iovec
, context
);
928 if (!GREEDY_REALLOC(iovec
, n_allocated
, n_iovec
+ 3)) {
933 /* Make sure we got all data we really need */
934 assert(context
[CONTEXT_PID
]);
935 assert(context
[CONTEXT_UID
]);
936 assert(context
[CONTEXT_GID
]);
937 assert(context
[CONTEXT_SIGNAL
]);
938 assert(context
[CONTEXT_TIMESTAMP
]);
939 assert(context
[CONTEXT_RLIMIT
]);
940 assert(context
[CONTEXT_COMM
]);
941 assert(coredump_fd
>= 0);
943 /* Small quirk: the journal fields contain the timestamp padded with six zeroes, so that the kernel-supplied 1s
944 * granularity timestamps becomes 1µs granularity, i.e. the granularity systemd usually operates in. Since we
945 * are reconstructing the original kernel context, we chop this off again, here. */
946 k
= strlen(context
[CONTEXT_TIMESTAMP
]);
948 context
[CONTEXT_TIMESTAMP
] = strndupa(context
[CONTEXT_TIMESTAMP
], k
- 6);
950 r
= submit_coredump(context
, iovec
, n_allocated
, n_iovec
, coredump_fd
);
953 for (i
= 0; i
< n_iovec
; i
++)
954 free(iovec
[i
].iov_base
);
960 static int send_iovec(const struct iovec iovec
[], size_t n_iovec
, int input_fd
) {
962 static const union sockaddr_union sa
= {
963 .un
.sun_family
= AF_UNIX
,
964 .un
.sun_path
= "/run/systemd/coredump",
966 _cleanup_close_
int fd
= -1;
970 assert(iovec
|| n_iovec
<= 0);
971 assert(input_fd
>= 0);
973 fd
= socket(AF_UNIX
, SOCK_SEQPACKET
|SOCK_CLOEXEC
, 0);
975 return log_error_errno(errno
, "Failed to create coredump socket: %m");
977 if (connect(fd
, &sa
.sa
, SOCKADDR_UN_LEN(sa
.un
)) < 0)
978 return log_error_errno(errno
, "Failed to connect to coredump service: %m");
980 for (i
= 0; i
< n_iovec
; i
++) {
982 .msg_iov
= (struct iovec
*) iovec
+ i
,
985 struct iovec copy
[2];
988 if (sendmsg(fd
, &mh
, MSG_NOSIGNAL
) >= 0)
991 if (errno
== EMSGSIZE
&& mh
.msg_iov
[0].iov_len
> 0) {
992 /* This field didn't fit? That's a pity. Given that this is just metadata,
993 * let's truncate the field at half, and try again. We append three dots, in
994 * order to show that this is truncated. */
996 if (mh
.msg_iov
!= copy
) {
997 /* We don't want to modify the caller's iovec, hence let's create our
998 * own array, consisting of two new iovecs, where the first is a
999 * (truncated) copy of what we want to send, and the second one
1000 * contains the trailing dots. */
1002 copy
[1] = (struct iovec
) {
1003 .iov_base
= (char[]) { '.', '.', '.' },
1011 copy
[0].iov_len
/= 2; /* halve it, and try again */
1015 return log_error_errno(errno
, "Failed to send coredump datagram: %m");
1019 r
= send_one_fd(fd
, input_fd
, 0);
1021 return log_error_errno(r
, "Failed to send coredump fd: %m");
1026 static int process_special_crash(const char *context
[], int input_fd
) {
1027 _cleanup_close_
int coredump_fd
= -1, coredump_node_fd
= -1;
1028 _cleanup_free_
char *filename
= NULL
;
1029 uint64_t coredump_size
;
1033 assert(input_fd
>= 0);
1035 /* If we are pid1 or journald, we cut things short, don't write to the journal, but still create a coredump. */
1037 if (arg_storage
!= COREDUMP_STORAGE_NONE
)
1038 arg_storage
= COREDUMP_STORAGE_EXTERNAL
;
1040 r
= save_external_coredump(context
, input_fd
, &filename
, &coredump_node_fd
, &coredump_fd
, &coredump_size
);
1044 r
= maybe_remove_external_coredump(filename
, coredump_size
);
1048 log_notice("Detected coredump of the journal daemon or PID 1, diverted to %s.", filename
);
1053 static char* set_iovec_field(struct iovec iovec
[27], size_t *n_iovec
, const char *field
, const char *value
) {
1056 x
= strappend(field
, value
);
1058 IOVEC_SET_STRING(iovec
[(*n_iovec
)++], x
);
1062 static char* set_iovec_field_free(struct iovec iovec
[27], size_t *n_iovec
, const char *field
, char *value
) {
1065 x
= set_iovec_field(iovec
, n_iovec
, field
, value
);
1070 static int gather_pid_metadata_and_process_special_crash(
1071 char* context
[_CONTEXT_MAX
],
1072 char **comm_fallback
,
1073 struct iovec
*iovec
, size_t *n_iovec
) {
1075 /* We need 26 empty slots in iovec!
1077 * Note that if we fail on oom later on, we do not roll-back changes to the iovec structure. (It remains valid,
1078 * with the first n_iovec fields initialized.) */
1086 r
= parse_pid(context
[CONTEXT_PID
], &pid
);
1088 return log_error_errno(r
, "Failed to parse PID \"%s\": %m", context
[CONTEXT_PID
]);
1090 r
= get_process_comm(pid
, &context
[CONTEXT_COMM
]);
1092 log_warning_errno(r
, "Failed to get COMM, falling back to the command line: %m");
1093 context
[CONTEXT_COMM
] = strv_join(comm_fallback
, " ");
1094 if (!context
[CONTEXT_COMM
])
1098 r
= get_process_exe(pid
, &context
[CONTEXT_EXE
]);
1100 log_warning_errno(r
, "Failed to get EXE, ignoring: %m");
1102 if (cg_pid_get_unit(pid
, &t
) >= 0) {
1104 /* If this is PID 1 disable coredump collection, we'll unlikely be able to process it later on. */
1105 if (streq(t
, SPECIAL_INIT_SCOPE
)) {
1106 log_notice("Due to PID 1 having crashed coredump collection will now be turned off.");
1107 (void) write_string_file("/proc/sys/kernel/core_pattern", "|/bin/false", 0);
1110 /* Let's avoid dead-locks when processing journald and init crashes, as socket activation and logging
1111 * are unlikely to work then. */
1112 if (STR_IN_SET(t
, SPECIAL_JOURNALD_SERVICE
, SPECIAL_INIT_SCOPE
)) {
1114 r
= process_special_crash((const char**) context
, STDIN_FILENO
);
1118 return 1; /* > 0 means: we have already processed it, because it's a special crash */
1121 set_iovec_field_free(iovec
, n_iovec
, "COREDUMP_UNIT=", t
);
1124 /* OK, now we know it's not the journal, hence we can make use of it now. */
1125 log_set_target(LOG_TARGET_JOURNAL_OR_KMSG
);
1128 if (cg_pid_get_user_unit(pid
, &t
) >= 0)
1129 set_iovec_field_free(iovec
, n_iovec
, "COREDUMP_USER_UNIT=", t
);
1131 /* The next few are mandatory */
1132 if (!set_iovec_field(iovec
, n_iovec
, "COREDUMP_PID=", context
[CONTEXT_PID
]))
1135 if (!set_iovec_field(iovec
, n_iovec
, "COREDUMP_UID=", context
[CONTEXT_UID
]))
1138 if (!set_iovec_field(iovec
, n_iovec
, "COREDUMP_GID=", context
[CONTEXT_GID
]))
1141 if (!set_iovec_field(iovec
, n_iovec
, "COREDUMP_SIGNAL=", context
[CONTEXT_SIGNAL
]))
1144 if (!set_iovec_field(iovec
, n_iovec
, "COREDUMP_RLIMIT=", context
[CONTEXT_RLIMIT
]))
1147 if (!set_iovec_field(iovec
, n_iovec
, "COREDUMP_COMM=", context
[CONTEXT_COMM
]))
1150 if (context
[CONTEXT_EXE
] &&
1151 !set_iovec_field(iovec
, n_iovec
, "COREDUMP_EXE=", context
[CONTEXT_EXE
]))
1154 if (sd_pid_get_session(pid
, &t
) >= 0)
1155 set_iovec_field_free(iovec
, n_iovec
, "COREDUMP_SESSION=", t
);
1157 if (sd_pid_get_owner_uid(pid
, &owner_uid
) >= 0) {
1158 r
= asprintf(&t
, "COREDUMP_OWNER_UID=" UID_FMT
, owner_uid
);
1160 IOVEC_SET_STRING(iovec
[(*n_iovec
)++], t
);
1163 if (sd_pid_get_slice(pid
, &t
) >= 0)
1164 set_iovec_field_free(iovec
, n_iovec
, "COREDUMP_SLICE=", t
);
1166 if (get_process_cmdline(pid
, 0, false, &t
) >= 0)
1167 set_iovec_field_free(iovec
, n_iovec
, "COREDUMP_CMDLINE=", t
);
1169 if (cg_pid_get_path_shifted(pid
, NULL
, &t
) >= 0)
1170 set_iovec_field_free(iovec
, n_iovec
, "COREDUMP_CGROUP=", t
);
1172 if (compose_open_fds(pid
, &t
) >= 0)
1173 set_iovec_field_free(iovec
, n_iovec
, "COREDUMP_OPEN_FDS=", t
);
1175 p
= procfs_file_alloca(pid
, "status");
1176 if (read_full_file(p
, &t
, NULL
) >= 0)
1177 set_iovec_field_free(iovec
, n_iovec
, "COREDUMP_PROC_STATUS=", t
);
1179 p
= procfs_file_alloca(pid
, "maps");
1180 if (read_full_file(p
, &t
, NULL
) >= 0)
1181 set_iovec_field_free(iovec
, n_iovec
, "COREDUMP_PROC_MAPS=", t
);
1183 p
= procfs_file_alloca(pid
, "limits");
1184 if (read_full_file(p
, &t
, NULL
) >= 0)
1185 set_iovec_field_free(iovec
, n_iovec
, "COREDUMP_PROC_LIMITS=", t
);
1187 p
= procfs_file_alloca(pid
, "cgroup");
1188 if (read_full_file(p
, &t
, NULL
) >=0)
1189 set_iovec_field_free(iovec
, n_iovec
, "COREDUMP_PROC_CGROUP=", t
);
1191 p
= procfs_file_alloca(pid
, "mountinfo");
1192 if (read_full_file(p
, &t
, NULL
) >=0)
1193 set_iovec_field_free(iovec
, n_iovec
, "COREDUMP_PROC_MOUNTINFO=", t
);
1195 if (get_process_cwd(pid
, &t
) >= 0)
1196 set_iovec_field_free(iovec
, n_iovec
, "COREDUMP_CWD=", t
);
1198 if (get_process_root(pid
, &t
) >= 0) {
1199 bool proc_self_root_is_slash
;
1201 proc_self_root_is_slash
= strcmp(t
, "/") == 0;
1203 set_iovec_field_free(iovec
, n_iovec
, "COREDUMP_ROOT=", t
);
1205 /* If the process' root is "/", then there is a chance it has
1206 * mounted own root and hence being containerized. */
1207 if (proc_self_root_is_slash
&& get_process_container_parent_cmdline(pid
, &t
) > 0)
1208 set_iovec_field_free(iovec
, n_iovec
, "COREDUMP_CONTAINER_CMDLINE=", t
);
1211 if (get_process_environ(pid
, &t
) >= 0)
1212 set_iovec_field_free(iovec
, n_iovec
, "COREDUMP_ENVIRON=", t
);
1214 t
= strjoin("COREDUMP_TIMESTAMP=", context
[CONTEXT_TIMESTAMP
], "000000", NULL
);
1216 IOVEC_SET_STRING(iovec
[(*n_iovec
)++], t
);
1218 if (safe_atoi(context
[CONTEXT_SIGNAL
], &signo
) >= 0 && SIGNAL_VALID(signo
))
1219 set_iovec_field(iovec
, n_iovec
, "COREDUMP_SIGNAL_NAME=SIG", signal_to_string(signo
));
1221 return 0; /* == 0 means: we successfully acquired all metadata */
1224 static int process_kernel(int argc
, char* argv
[]) {
1226 char* context
[_CONTEXT_MAX
] = {};
1227 struct iovec iovec
[28];
1228 size_t i
, n_iovec
, n_to_free
= 0;
1231 log_debug("Processing coredump received from the kernel...");
1233 if (argc
< CONTEXT_COMM
+ 1) {
1234 log_error("Not enough arguments passed by the kernel (%i, expected %i).", argc
- 1, CONTEXT_COMM
+ 1 - 1);
1238 context
[CONTEXT_PID
] = argv
[1 + CONTEXT_PID
];
1239 context
[CONTEXT_UID
] = argv
[1 + CONTEXT_UID
];
1240 context
[CONTEXT_GID
] = argv
[1 + CONTEXT_GID
];
1241 context
[CONTEXT_SIGNAL
] = argv
[1 + CONTEXT_SIGNAL
];
1242 context
[CONTEXT_TIMESTAMP
] = argv
[1 + CONTEXT_TIMESTAMP
];
1243 context
[CONTEXT_RLIMIT
] = argv
[1 + CONTEXT_RLIMIT
];
1245 r
= gather_pid_metadata_and_process_special_crash(context
, argv
+ 1 + CONTEXT_COMM
, iovec
, &n_to_free
);
1247 /* Error, or a a special crash, which has already been processed. */
1250 n_iovec
= n_to_free
;
1252 IOVEC_SET_STRING(iovec
[n_iovec
++], "MESSAGE_ID=" SD_MESSAGE_COREDUMP_STR
);
1254 assert_cc(2 == LOG_CRIT
);
1255 IOVEC_SET_STRING(iovec
[n_iovec
++], "PRIORITY=2");
1257 assert(n_iovec
<= ELEMENTSOF(iovec
));
1259 r
= send_iovec(iovec
, n_iovec
, STDIN_FILENO
);
1262 for (i
= 0; i
< n_to_free
; i
++)
1263 free(iovec
[i
].iov_base
);
1265 /* Those fields are allocated by gather_pid_metadata_and_process_special_crash */
1266 free(context
[CONTEXT_COMM
]);
1267 free(context
[CONTEXT_EXE
]);
1272 static int process_backtrace(int argc
, char *argv
[]) {
1273 char *context
[_CONTEXT_MAX
] = {};
1274 _cleanup_free_
char *message
= NULL
;
1275 _cleanup_free_
struct iovec
*iovec
= NULL
;
1276 size_t n_iovec
, n_allocated
, n_to_free
= 0, i
;
1278 JournalImporter importer
= {
1282 log_debug("Processing backtrace on stdin...");
1284 if (argc
< CONTEXT_COMM
+ 1) {
1285 log_error("Not enough arguments passed (%i, expected %i).", argc
- 1, CONTEXT_COMM
+ 1 - 1);
1289 context
[CONTEXT_PID
] = argv
[2 + CONTEXT_PID
];
1290 context
[CONTEXT_UID
] = argv
[2 + CONTEXT_UID
];
1291 context
[CONTEXT_GID
] = argv
[2 + CONTEXT_GID
];
1292 context
[CONTEXT_SIGNAL
] = argv
[2 + CONTEXT_SIGNAL
];
1293 context
[CONTEXT_TIMESTAMP
] = argv
[2 + CONTEXT_TIMESTAMP
];
1294 context
[CONTEXT_RLIMIT
] = argv
[2 + CONTEXT_RLIMIT
];
1296 n_allocated
= 33; /* 25 metadata, 2 static, +unknown input, rounded up */
1297 iovec
= new(struct iovec
, n_allocated
);
1301 r
= gather_pid_metadata_and_process_special_crash(context
, argv
+ 2 + CONTEXT_COMM
, iovec
, &n_to_free
);
1305 /* This was a special crash, and has already been processed. */
1309 n_iovec
= n_to_free
;
1312 r
= journal_importer_process_data(&importer
);
1314 log_error_errno(r
, "Failed to parse journal entry on stdin: %m");
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_SET_STRING(iovec
[n_iovec
++], message
);
1337 for (i
= 0; i
< importer
.iovw
.count
; i
++)
1338 iovec
[n_iovec
++] = importer
.iovw
.iovec
[i
];
1341 IOVEC_SET_STRING(iovec
[n_iovec
++], "MESSAGE_ID=" SD_MESSAGE_BACKTRACE_STR
);
1342 assert_cc(2 == LOG_CRIT
);
1343 IOVEC_SET_STRING(iovec
[n_iovec
++], "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_and_process_special_crash */
1356 free(context
[CONTEXT_COMM
]);
1357 free(context
[CONTEXT_EXE
]);
1362 int main(int argc
, char *argv
[]) {
1365 /* First, log to a safe place, since we don't know what crashed and it might
1366 * be journald which we'd rather not log to then. */
1368 log_set_target(LOG_TARGET_KMSG
);
1371 /* Make sure we never enter a loop */
1372 (void) prctl(PR_SET_DUMPABLE
, 0);
1374 /* Ignore all parse errors */
1375 (void) parse_config();
1377 log_debug("Selected storage '%s'.", coredump_storage_to_string(arg_storage
));
1378 log_debug("Selected compression %s.", yes_no(arg_compress
));
1380 r
= sd_listen_fds(false);
1382 log_error_errno(r
, "Failed to determine number of file descriptor: %m");
1386 /* If we got an fd passed, we are running in coredumpd mode. Otherwise we
1387 * are invoked from the kernel as coredump handler. */
1389 if (streq_ptr(argv
[1], "--backtrace"))
1390 r
= process_backtrace(argc
, argv
);
1392 r
= process_kernel(argc
, argv
);
1394 r
= process_socket(SD_LISTEN_FDS_START
);
1396 log_error("Received unexpected number of file descriptors.");
1401 return r
< 0 ? EXIT_FAILURE
: EXIT_SUCCESS
;