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-journal.h"
33 #include "sd-daemon.h"
36 #include "alloc-util.h"
37 #include "capability-util.h"
38 #include "cgroup-util.h"
40 #include "conf-parser.h"
42 #include "coredump-vacuum.h"
43 #include "dirent-util.h"
49 #include "journald-native.h"
54 #include "parse-util.h"
55 #include "process-util.h"
56 #include "socket-util.h"
58 #include "stacktrace.h"
59 #include "string-table.h"
60 #include "string-util.h"
62 #include "user-util.h"
65 /* The maximum size up to which we process coredumps */
66 #define PROCESS_SIZE_MAX ((uint64_t) (2LLU*1024LLU*1024LLU*1024LLU))
68 /* The maximum size up to which we leave the coredump around on disk */
69 #define EXTERNAL_SIZE_MAX PROCESS_SIZE_MAX
71 /* The maximum size up to which we store the coredump in the journal */
72 #define JOURNAL_SIZE_MAX ((size_t) (767LU*1024LU*1024LU))
74 /* Make sure to not make this larger than the maximum journal entry
75 * size. See DATA_SIZE_MAX in journald-native.c. */
76 assert_cc(JOURNAL_SIZE_MAX
<= DATA_SIZE_MAX
);
79 /* We use this as array indexes for a couple of special fields we use for naming coredumping files, and
92 typedef enum CoredumpStorage
{
93 COREDUMP_STORAGE_NONE
,
94 COREDUMP_STORAGE_EXTERNAL
,
95 COREDUMP_STORAGE_JOURNAL
,
96 COREDUMP_STORAGE_BOTH
,
97 _COREDUMP_STORAGE_MAX
,
98 _COREDUMP_STORAGE_INVALID
= -1
101 static const char* const coredump_storage_table
[_COREDUMP_STORAGE_MAX
] = {
102 [COREDUMP_STORAGE_NONE
] = "none",
103 [COREDUMP_STORAGE_EXTERNAL
] = "external",
104 [COREDUMP_STORAGE_JOURNAL
] = "journal",
105 [COREDUMP_STORAGE_BOTH
] = "both",
108 DEFINE_PRIVATE_STRING_TABLE_LOOKUP(coredump_storage
, CoredumpStorage
);
109 static DEFINE_CONFIG_PARSE_ENUM(config_parse_coredump_storage
, coredump_storage
, CoredumpStorage
, "Failed to parse storage setting");
111 static CoredumpStorage arg_storage
= COREDUMP_STORAGE_EXTERNAL
;
112 static bool arg_compress
= true;
113 static uint64_t arg_process_size_max
= PROCESS_SIZE_MAX
;
114 static uint64_t arg_external_size_max
= EXTERNAL_SIZE_MAX
;
115 static size_t arg_journal_size_max
= JOURNAL_SIZE_MAX
;
116 static uint64_t arg_keep_free
= (uint64_t) -1;
117 static uint64_t arg_max_use
= (uint64_t) -1;
119 static int parse_config(void) {
120 static const ConfigTableItem items
[] = {
121 { "Coredump", "Storage", config_parse_coredump_storage
, 0, &arg_storage
},
122 { "Coredump", "Compress", config_parse_bool
, 0, &arg_compress
},
123 { "Coredump", "ProcessSizeMax", config_parse_iec_uint64
, 0, &arg_process_size_max
},
124 { "Coredump", "ExternalSizeMax", config_parse_iec_uint64
, 0, &arg_external_size_max
},
125 { "Coredump", "JournalSizeMax", config_parse_iec_size
, 0, &arg_journal_size_max
},
126 { "Coredump", "KeepFree", config_parse_iec_uint64
, 0, &arg_keep_free
},
127 { "Coredump", "MaxUse", config_parse_iec_uint64
, 0, &arg_max_use
},
131 return config_parse_many(PKGSYSCONFDIR
"/coredump.conf",
132 CONF_PATHS_NULSTR("systemd/coredump.conf.d"),
134 config_item_table_lookup
, items
,
138 static int fix_acl(int fd
, uid_t uid
) {
141 _cleanup_(acl_freep
) acl_t acl
= NULL
;
143 acl_permset_t permset
;
148 if (uid
<= SYSTEM_UID_MAX
)
151 /* Make sure normal users can read (but not write or delete)
152 * their own coredumps */
154 acl
= acl_get_fd(fd
);
156 return log_error_errno(errno
, "Failed to get ACL: %m");
158 if (acl_create_entry(&acl
, &entry
) < 0 ||
159 acl_set_tag_type(entry
, ACL_USER
) < 0 ||
160 acl_set_qualifier(entry
, &uid
) < 0)
161 return log_error_errno(errno
, "Failed to patch ACL: %m");
163 if (acl_get_permset(entry
, &permset
) < 0 ||
164 acl_add_perm(permset
, ACL_READ
) < 0)
165 return log_warning_errno(errno
, "Failed to patch ACL: %m");
167 r
= calc_acl_mask_if_needed(&acl
);
169 return log_warning_errno(r
, "Failed to patch ACL: %m");
171 if (acl_set_fd(fd
, acl
) < 0)
172 return log_error_errno(errno
, "Failed to apply ACL: %m");
178 static int fix_xattr(int fd
, const char *context
[_CONTEXT_MAX
]) {
180 static const char * const xattrs
[_CONTEXT_MAX
] = {
181 [CONTEXT_PID
] = "user.coredump.pid",
182 [CONTEXT_UID
] = "user.coredump.uid",
183 [CONTEXT_GID
] = "user.coredump.gid",
184 [CONTEXT_SIGNAL
] = "user.coredump.signal",
185 [CONTEXT_TIMESTAMP
] = "user.coredump.timestamp",
186 [CONTEXT_COMM
] = "user.coredump.comm",
187 [CONTEXT_EXE
] = "user.coredump.exe",
195 /* Attach some metadata to coredumps via extended
196 * attributes. Just because we can. */
198 for (i
= 0; i
< _CONTEXT_MAX
; i
++) {
201 if (isempty(context
[i
]) || !xattrs
[i
])
204 k
= fsetxattr(fd
, xattrs
[i
], context
[i
], strlen(context
[i
]), XATTR_CREATE
);
212 #define filename_escape(s) xescape((s), "./ ")
214 static inline const char *coredump_tmpfile_name(const char *s
) {
215 return s
? s
: "(unnamed temporary file)";
218 static int fix_permissions(
220 const char *filename
,
222 const char *context
[_CONTEXT_MAX
],
231 /* Ignore errors on these */
232 (void) fchmod(fd
, 0640);
233 (void) fix_acl(fd
, uid
);
234 (void) fix_xattr(fd
, context
);
237 return log_error_errno(errno
, "Failed to sync coredump %s: %m", coredump_tmpfile_name(filename
));
239 r
= link_tmpfile(fd
, filename
, target
);
241 return log_error_errno(r
, "Failed to move coredump %s into place: %m", target
);
246 static int maybe_remove_external_coredump(const char *filename
, uint64_t size
) {
248 /* Returns 1 if might remove, 0 if will not remove, < 0 on error. */
250 if (IN_SET(arg_storage
, COREDUMP_STORAGE_EXTERNAL
, COREDUMP_STORAGE_BOTH
) &&
251 size
<= arg_external_size_max
)
257 if (unlink(filename
) < 0 && errno
!= ENOENT
)
258 return log_error_errno(errno
, "Failed to unlink %s: %m", filename
);
263 static int make_filename(const char *context
[_CONTEXT_MAX
], char **ret
) {
264 _cleanup_free_
char *c
= NULL
, *u
= NULL
, *p
= NULL
, *t
= NULL
;
265 sd_id128_t boot
= {};
270 c
= filename_escape(context
[CONTEXT_COMM
]);
274 u
= filename_escape(context
[CONTEXT_UID
]);
278 r
= sd_id128_get_boot(&boot
);
282 p
= filename_escape(context
[CONTEXT_PID
]);
286 t
= filename_escape(context
[CONTEXT_TIMESTAMP
]);
291 "/var/lib/systemd/coredump/core.%s.%s." SD_ID128_FORMAT_STR
".%s.%s000000",
294 SD_ID128_FORMAT_VAL(boot
),
302 static int save_external_coredump(
303 const char *context
[_CONTEXT_MAX
],
308 uint64_t *ret_size
) {
310 _cleanup_free_
char *fn
= NULL
, *tmp
= NULL
;
311 _cleanup_close_
int fd
= -1;
312 uint64_t rlimit
, max_size
;
318 assert(ret_filename
);
323 r
= parse_uid(context
[CONTEXT_UID
], &uid
);
325 return log_error_errno(r
, "Failed to parse UID: %m");
327 r
= safe_atou64(context
[CONTEXT_RLIMIT
], &rlimit
);
329 return log_error_errno(r
, "Failed to parse resource limit: %s", context
[CONTEXT_RLIMIT
]);
331 /* Is coredumping disabled? Then don't bother saving/processing the coredump */
332 log_info("Core Dumping has been disabled for process %s (%s).", context
[CONTEXT_PID
], context
[CONTEXT_COMM
]);
336 /* Never store more than the process configured, or than we actually shall keep or process */
337 max_size
= MIN(rlimit
, MAX(arg_process_size_max
, arg_external_size_max
));
339 r
= make_filename(context
, &fn
);
341 return log_error_errno(r
, "Failed to determine coredump file name: %m");
343 mkdir_p_label("/var/lib/systemd/coredump", 0755);
345 fd
= open_tmpfile_linkable(fn
, O_RDWR
|O_CLOEXEC
, &tmp
);
347 return log_error_errno(fd
, "Failed to create temporary file for coredump %s: %m", fn
);
349 r
= copy_bytes(input_fd
, fd
, max_size
, false);
351 log_error("Coredump of %s (%s) is larger than configured processing limit, refusing.", context
[CONTEXT_PID
], context
[CONTEXT_COMM
]);
353 } else if (IN_SET(r
, -EDQUOT
, -ENOSPC
)) {
354 log_error("Not enough disk space for coredump of %s (%s), refusing.", context
[CONTEXT_PID
], context
[CONTEXT_COMM
]);
357 log_error_errno(r
, "Failed to dump coredump to file: %m");
361 if (fstat(fd
, &st
) < 0) {
362 log_error_errno(errno
, "Failed to fstat coredump %s: %m", coredump_tmpfile_name(tmp
));
366 if (lseek(fd
, 0, SEEK_SET
) == (off_t
) -1) {
367 log_error_errno(errno
, "Failed to seek on %s: %m", coredump_tmpfile_name(tmp
));
371 #if defined(HAVE_XZ) || defined(HAVE_LZ4)
372 /* If we will remove the coredump anyway, do not compress. */
373 if (maybe_remove_external_coredump(NULL
, st
.st_size
) == 0
376 _cleanup_free_
char *fn_compressed
= NULL
, *tmp_compressed
= NULL
;
377 _cleanup_close_
int fd_compressed
= -1;
379 fn_compressed
= strappend(fn
, COMPRESSED_EXT
);
380 if (!fn_compressed
) {
385 fd_compressed
= open_tmpfile_linkable(fn_compressed
, O_RDWR
|O_CLOEXEC
, &tmp_compressed
);
386 if (fd_compressed
< 0) {
387 log_error_errno(fd_compressed
, "Failed to create temporary file for coredump %s: %m", fn_compressed
);
391 r
= compress_stream(fd
, fd_compressed
, -1);
393 log_error_errno(r
, "Failed to compress %s: %m", coredump_tmpfile_name(tmp_compressed
));
394 goto fail_compressed
;
397 r
= fix_permissions(fd_compressed
, tmp_compressed
, fn_compressed
, context
, uid
);
399 goto fail_compressed
;
401 /* OK, this worked, we can get rid of the uncompressed version now */
405 *ret_filename
= fn_compressed
; /* compressed */
406 *ret_node_fd
= fd_compressed
; /* compressed */
407 *ret_data_fd
= fd
; /* uncompressed */
408 *ret_size
= (uint64_t) st
.st_size
; /* uncompressed */
410 fn_compressed
= NULL
;
411 fd
= fd_compressed
= -1;
417 (void) unlink(tmp_compressed
);
423 r
= fix_permissions(fd
, tmp
, fn
, context
, uid
);
430 *ret_size
= (uint64_t) st
.st_size
;
443 static int allocate_journal_field(int fd
, size_t size
, char **ret
, size_t *ret_size
) {
444 _cleanup_free_
char *field
= NULL
;
451 if (lseek(fd
, 0, SEEK_SET
) == (off_t
) -1)
452 return log_warning_errno(errno
, "Failed to seek: %m");
454 field
= malloc(9 + size
);
456 log_warning("Failed to allocate memory for coredump, coredump will not be stored.");
460 memcpy(field
, "COREDUMP=", 9);
462 n
= read(fd
, field
+ 9, size
);
464 return log_error_errno((int) n
, "Failed to read core data: %m");
465 if ((size_t) n
< size
) {
466 log_error("Core data too short.");
471 *ret_size
= size
+ 9;
478 /* Joins /proc/[pid]/fd/ and /proc/[pid]/fdinfo/ into the following lines:
492 static int compose_open_fds(pid_t pid
, char **open_fds
) {
493 _cleanup_closedir_
DIR *proc_fd_dir
= NULL
;
494 _cleanup_close_
int proc_fdinfo_fd
= -1;
495 _cleanup_free_
char *buffer
= NULL
;
496 _cleanup_fclose_
FILE *stream
= NULL
;
497 const char *fddelim
= "", *path
;
498 struct dirent
*dent
= NULL
;
503 assert(open_fds
!= NULL
);
505 path
= procfs_file_alloca(pid
, "fd");
506 proc_fd_dir
= opendir(path
);
510 proc_fdinfo_fd
= openat(dirfd(proc_fd_dir
), "../fdinfo", O_DIRECTORY
|O_NOFOLLOW
|O_CLOEXEC
|O_PATH
);
511 if (proc_fdinfo_fd
< 0)
514 stream
= open_memstream(&buffer
, &size
);
518 FOREACH_DIRENT(dent
, proc_fd_dir
, return -errno
) {
519 _cleanup_fclose_
FILE *fdinfo
= NULL
;
520 _cleanup_free_
char *fdname
= NULL
;
524 r
= readlinkat_malloc(dirfd(proc_fd_dir
), dent
->d_name
, &fdname
);
528 fprintf(stream
, "%s%s:%s\n", fddelim
, dent
->d_name
, fdname
);
531 /* Use the directory entry from /proc/[pid]/fd with /proc/[pid]/fdinfo */
532 fd
= openat(proc_fdinfo_fd
, dent
->d_name
, O_NOFOLLOW
|O_CLOEXEC
|O_RDONLY
);
536 fdinfo
= fdopen(fd
, "re");
537 if (fdinfo
== NULL
) {
542 FOREACH_LINE(line
, fdinfo
, break) {
544 if (!endswith(line
, "\n"))
550 stream
= safe_fclose(stream
);
561 static int change_uid_gid(const char *context
[]) {
566 r
= parse_uid(context
[CONTEXT_UID
], &uid
);
570 if (uid
<= SYSTEM_UID_MAX
) {
571 const char *user
= "systemd-coredump";
573 r
= get_user_creds(&user
, &uid
, &gid
, NULL
, NULL
);
575 log_warning_errno(r
, "Cannot resolve %s user. Proceeding to dump core as root: %m", user
);
579 r
= parse_gid(context
[CONTEXT_GID
], &gid
);
584 return drop_privileges(uid
, gid
, 0);
587 static int submit_coredump(
588 const char *context
[_CONTEXT_MAX
],
590 size_t n_iovec_allocated
,
594 _cleanup_close_
int coredump_fd
= -1, coredump_node_fd
= -1;
595 _cleanup_free_
char *core_message
= NULL
, *filename
= NULL
, *coredump_data
= NULL
;
596 uint64_t coredump_size
;
601 assert(n_iovec_allocated
>= n_iovec
+ 3);
602 assert(input_fd
>= 0);
604 /* Vacuum before we write anything again */
605 (void) coredump_vacuum(-1, arg_keep_free
, arg_max_use
);
607 /* Always stream the coredump to disk, if that's possible */
608 r
= save_external_coredump(context
, input_fd
, &filename
, &coredump_node_fd
, &coredump_fd
, &coredump_size
);
610 /* Skip whole core dumping part */
613 /* If we don't want to keep the coredump on disk, remove it now, as later on we will lack the privileges for
614 * it. However, we keep the fd to it, so that we can still process it and log it. */
615 r
= maybe_remove_external_coredump(filename
, coredump_size
);
619 const char *coredump_filename
;
621 coredump_filename
= strjoina("COREDUMP_FILENAME=", filename
);
622 IOVEC_SET_STRING(iovec
[n_iovec
++], coredump_filename
);
625 /* Vacuum again, but exclude the coredump we just created */
626 (void) coredump_vacuum(coredump_node_fd
>= 0 ? coredump_node_fd
: coredump_fd
, arg_keep_free
, arg_max_use
);
628 /* Now, let's drop privileges to become the user who owns the segfaulted process and allocate the coredump
629 * memory under the user's uid. This also ensures that the credentials journald will see are the ones of the
630 * coredumping user, thus making sure the user gets access to the core dump. Let's also get rid of all
631 * capabilities, if we run as root, we won't need them anymore. */
632 r
= change_uid_gid(context
);
634 return log_error_errno(r
, "Failed to drop privileges: %m");
637 /* Try to get a strack trace if we can */
638 if (coredump_size
<= arg_process_size_max
) {
639 _cleanup_free_
char *stacktrace
= NULL
;
641 r
= coredump_make_stack_trace(coredump_fd
, context
[CONTEXT_EXE
], &stacktrace
);
643 core_message
= strjoin("MESSAGE=Process ", context
[CONTEXT_PID
], " (", context
[CONTEXT_COMM
], ") of user ", context
[CONTEXT_UID
], " dumped core.\n\n", stacktrace
, NULL
);
644 else if (r
== -EINVAL
)
645 log_warning("Failed to generate stack trace: %s", dwfl_errmsg(dwfl_errno()));
647 log_warning_errno(r
, "Failed to generate stack trace: %m");
653 core_message
= strjoin("MESSAGE=Process ", context
[CONTEXT_PID
], " (", context
[CONTEXT_COMM
], ") of user ", context
[CONTEXT_UID
], " dumped core.", NULL
);
655 IOVEC_SET_STRING(iovec
[n_iovec
++], core_message
);
657 /* Optionally store the entire coredump in the journal */
658 if (IN_SET(arg_storage
, COREDUMP_STORAGE_JOURNAL
, COREDUMP_STORAGE_BOTH
) &&
659 coredump_size
<= arg_journal_size_max
) {
662 /* Store the coredump itself in the journal */
664 r
= allocate_journal_field(coredump_fd
, (size_t) coredump_size
, &coredump_data
, &sz
);
666 iovec
[n_iovec
].iov_base
= coredump_data
;
667 iovec
[n_iovec
].iov_len
= sz
;
672 assert(n_iovec
<= n_iovec_allocated
);
674 r
= sd_journal_sendv(iovec
, n_iovec
);
676 return log_error_errno(r
, "Failed to log coredump: %m");
681 static void map_context_fields(const struct iovec
*iovec
, const char *context
[]) {
683 static const char * const context_field_names
[_CONTEXT_MAX
] = {
684 [CONTEXT_PID
] = "COREDUMP_PID=",
685 [CONTEXT_UID
] = "COREDUMP_UID=",
686 [CONTEXT_GID
] = "COREDUMP_GID=",
687 [CONTEXT_SIGNAL
] = "COREDUMP_SIGNAL=",
688 [CONTEXT_TIMESTAMP
] = "COREDUMP_TIMESTAMP=",
689 [CONTEXT_COMM
] = "COREDUMP_COMM=",
690 [CONTEXT_EXE
] = "COREDUMP_EXE=",
691 [CONTEXT_RLIMIT
] = "COREDUMP_RLIMIT=",
699 for (i
= 0; i
< _CONTEXT_MAX
; i
++) {
702 l
= strlen(context_field_names
[i
]);
703 if (iovec
->iov_len
< l
)
706 if (memcmp(iovec
->iov_base
, context_field_names
[i
], l
) != 0)
709 /* Note that these strings are NUL terminated, because we made sure that a trailing NUL byte is in the
710 * buffer, though not included in the iov_len count. (see below) */
711 context
[i
] = (char*) iovec
->iov_base
+ l
;
716 static int process_socket(int fd
) {
717 _cleanup_close_
int coredump_fd
= -1;
718 struct iovec
*iovec
= NULL
;
719 size_t n_iovec
= 0, n_iovec_allocated
= 0, i
;
720 const char *context
[_CONTEXT_MAX
] = {};
725 log_set_target(LOG_TARGET_AUTO
);
726 log_parse_environment();
731 struct cmsghdr cmsghdr
;
732 uint8_t buf
[CMSG_SPACE(sizeof(int))];
735 .msg_control
= &control
,
736 .msg_controllen
= sizeof(control
),
742 if (!GREEDY_REALLOC(iovec
, n_iovec_allocated
, n_iovec
+ 3)) {
747 l
= next_datagram_size_fd(fd
);
749 r
= log_error_errno(l
, "Failed to determine datagram size to read: %m");
755 iovec
[n_iovec
].iov_len
= l
;
756 iovec
[n_iovec
].iov_base
= malloc(l
+ 1);
757 if (!iovec
[n_iovec
].iov_base
) {
762 mh
.msg_iov
= iovec
+ n_iovec
;
764 n
= recvmsg(fd
, &mh
, MSG_NOSIGNAL
|MSG_CMSG_CLOEXEC
);
766 free(iovec
[n_iovec
].iov_base
);
767 r
= log_error_errno(errno
, "Failed to receive datagram: %m");
772 struct cmsghdr
*cmsg
, *found
= NULL
;
773 /* The final zero-length datagram carries the file descriptor and tells us that we're done. */
775 free(iovec
[n_iovec
].iov_base
);
777 CMSG_FOREACH(cmsg
, &mh
) {
778 if (cmsg
->cmsg_level
== SOL_SOCKET
&&
779 cmsg
->cmsg_type
== SCM_RIGHTS
&&
780 cmsg
->cmsg_len
== CMSG_LEN(sizeof(int))) {
787 log_error("Coredump file descriptor missing.");
792 assert(coredump_fd
< 0);
793 coredump_fd
= *(int*) CMSG_DATA(found
);
797 /* Add trailing NUL byte, in case these are strings */
798 ((char*) iovec
[n_iovec
].iov_base
)[n
] = 0;
799 iovec
[n_iovec
].iov_len
= (size_t) n
;
802 map_context_fields(iovec
+ n_iovec
, context
);
806 if (!GREEDY_REALLOC(iovec
, n_iovec_allocated
, n_iovec
+ 3)) {
811 /* Make sure we got all data we really need */
812 assert(context
[CONTEXT_PID
]);
813 assert(context
[CONTEXT_UID
]);
814 assert(context
[CONTEXT_GID
]);
815 assert(context
[CONTEXT_SIGNAL
]);
816 assert(context
[CONTEXT_TIMESTAMP
]);
817 assert(context
[CONTEXT_RLIMIT
]);
818 assert(context
[CONTEXT_COMM
]);
819 assert(coredump_fd
>= 0);
821 r
= submit_coredump(context
, iovec
, n_iovec_allocated
, n_iovec
, coredump_fd
);
824 for (i
= 0; i
< n_iovec
; i
++)
825 free(iovec
[i
].iov_base
);
831 static int send_iovec(const struct iovec iovec
[], size_t n_iovec
, int input_fd
) {
833 static const union sockaddr_union sa
= {
834 .un
.sun_family
= AF_UNIX
,
835 .un
.sun_path
= "/run/systemd/coredump",
837 _cleanup_close_
int fd
= -1;
841 assert(iovec
|| n_iovec
<= 0);
842 assert(input_fd
>= 0);
844 fd
= socket(AF_UNIX
, SOCK_SEQPACKET
|SOCK_CLOEXEC
, 0);
846 return log_error_errno(errno
, "Failed to create coredump socket: %m");
848 if (connect(fd
, &sa
.sa
, SOCKADDR_UN_LEN(sa
.un
)) < 0)
849 return log_error_errno(errno
, "Failed to connect to coredump service: %m");
851 for (i
= 0; i
< n_iovec
; i
++) {
853 .msg_iov
= (struct iovec
*) iovec
+ i
,
856 struct iovec copy
[2];
859 if (sendmsg(fd
, &mh
, MSG_NOSIGNAL
) >= 0)
862 if (errno
== EMSGSIZE
&& mh
.msg_iov
[0].iov_len
> 0) {
863 /* This field didn't fit? That's a pity. Given that this is just metadata,
864 * let's truncate the field at half, and try again. We append three dots, in
865 * order to show that this is truncated. */
867 if (mh
.msg_iov
!= copy
) {
868 /* We don't want to modify the caller's iovec, hence let's create our
869 * own array, consisting of two new iovecs, where the first is a
870 * (truncated) copy of what we want to send, and the second one
871 * contains the trailing dots. */
873 copy
[1] = (struct iovec
) {
874 .iov_base
= (char[]) { '.', '.', '.' },
882 copy
[0].iov_len
/= 2; /* halve it, and try again */
886 return log_error_errno(errno
, "Failed to send coredump datagram: %m");
890 r
= send_one_fd(fd
, input_fd
, 0);
892 return log_error_errno(r
, "Failed to send coredump fd: %m");
897 static int process_special_crash(const char *context
[], int input_fd
) {
898 _cleanup_close_
int coredump_fd
= -1, coredump_node_fd
= -1;
899 _cleanup_free_
char *filename
= NULL
;
900 uint64_t coredump_size
;
904 assert(input_fd
>= 0);
906 /* If we are pid1 or journald, we cut things short, don't write to the journal, but still create a coredump. */
908 if (arg_storage
!= COREDUMP_STORAGE_NONE
)
909 arg_storage
= COREDUMP_STORAGE_EXTERNAL
;
911 r
= save_external_coredump(context
, input_fd
, &filename
, &coredump_node_fd
, &coredump_fd
, &coredump_size
);
915 r
= maybe_remove_external_coredump(filename
, coredump_size
);
919 log_notice("Detected coredump of the journal daemon or PID 1, diverted to %s.", filename
);
924 static int process_kernel(int argc
, char* argv
[]) {
926 /* The small core field we allocate on the stack, to keep things simple */
928 *core_pid
= NULL
, *core_uid
= NULL
, *core_gid
= NULL
, *core_signal
= NULL
,
929 *core_session
= NULL
, *core_exe
= NULL
, *core_comm
= NULL
, *core_cmdline
= NULL
,
930 *core_cgroup
= NULL
, *core_cwd
= NULL
, *core_root
= NULL
, *core_unit
= NULL
,
931 *core_user_unit
= NULL
, *core_slice
= NULL
, *core_timestamp
= NULL
, *core_rlimit
= NULL
;
933 /* The larger ones we allocate on the heap */
935 *core_owner_uid
= NULL
, *core_open_fds
= NULL
, *core_proc_status
= NULL
,
936 *core_proc_maps
= NULL
, *core_proc_limits
= NULL
, *core_proc_cgroup
= NULL
, *core_environ
= NULL
;
938 _cleanup_free_
char *exe
= NULL
, *comm
= NULL
;
939 const char *context
[_CONTEXT_MAX
];
940 struct iovec iovec
[25];
948 if (argc
< CONTEXT_COMM
+ 1) {
949 log_error("Not enough arguments passed from kernel (%i, expected %i).", argc
- 1, CONTEXT_COMM
+ 1 - 1);
953 r
= parse_pid(argv
[CONTEXT_PID
+ 1], &pid
);
955 return log_error_errno(r
, "Failed to parse PID.");
957 r
= get_process_comm(pid
, &comm
);
959 log_warning_errno(r
, "Failed to get COMM, falling back to the command line: %m");
960 comm
= strv_join(argv
+ CONTEXT_COMM
+ 1, " ");
965 r
= get_process_exe(pid
, &exe
);
967 log_warning_errno(r
, "Failed to get EXE, ignoring: %m");
969 context
[CONTEXT_PID
] = argv
[CONTEXT_PID
+ 1];
970 context
[CONTEXT_UID
] = argv
[CONTEXT_UID
+ 1];
971 context
[CONTEXT_GID
] = argv
[CONTEXT_GID
+ 1];
972 context
[CONTEXT_SIGNAL
] = argv
[CONTEXT_SIGNAL
+ 1];
973 context
[CONTEXT_TIMESTAMP
] = argv
[CONTEXT_TIMESTAMP
+ 1];
974 context
[CONTEXT_RLIMIT
] = argv
[CONTEXT_RLIMIT
+ 1];
975 context
[CONTEXT_COMM
] = comm
;
976 context
[CONTEXT_EXE
] = exe
;
978 if (cg_pid_get_unit(pid
, &t
) >= 0) {
980 /* If this is PID 1 disable coredump collection, we'll unlikely be able to process it later on. */
981 if (streq(t
, SPECIAL_INIT_SCOPE
)) {
982 log_notice("Due to PID 1 having crashed coredump collection will now be turned off.");
983 (void) write_string_file("/proc/sys/kernel/core_pattern", "|/bin/false", 0);
986 /* Let's avoid dead-locks when processing journald and init crashes, as socket activation and logging
987 * are unlikely to work then. */
988 if (STR_IN_SET(t
, SPECIAL_JOURNALD_SERVICE
, SPECIAL_INIT_SCOPE
)) {
990 return process_special_crash(context
, STDIN_FILENO
);
993 core_unit
= strjoina("COREDUMP_UNIT=", t
);
996 IOVEC_SET_STRING(iovec
[n_iovec
++], core_unit
);
999 /* OK, now we know it's not the journal, hence we can make use of it now. */
1000 log_set_target(LOG_TARGET_JOURNAL_OR_KMSG
);
1003 if (cg_pid_get_user_unit(pid
, &t
) >= 0) {
1004 core_user_unit
= strjoina("COREDUMP_USER_UNIT=", t
);
1007 IOVEC_SET_STRING(iovec
[n_iovec
++], core_user_unit
);
1010 core_pid
= strjoina("COREDUMP_PID=", context
[CONTEXT_PID
]);
1011 IOVEC_SET_STRING(iovec
[n_iovec
++], core_pid
);
1013 core_uid
= strjoina("COREDUMP_UID=", context
[CONTEXT_UID
]);
1014 IOVEC_SET_STRING(iovec
[n_iovec
++], core_uid
);
1016 core_gid
= strjoina("COREDUMP_GID=", context
[CONTEXT_GID
]);
1017 IOVEC_SET_STRING(iovec
[n_iovec
++], core_gid
);
1019 core_signal
= strjoina("COREDUMP_SIGNAL=", context
[CONTEXT_SIGNAL
]);
1020 IOVEC_SET_STRING(iovec
[n_iovec
++], core_signal
);
1022 core_rlimit
= strjoina("COREDUMP_RLIMIT=", context
[CONTEXT_RLIMIT
]);
1023 IOVEC_SET_STRING(iovec
[n_iovec
++], core_rlimit
);
1025 if (sd_pid_get_session(pid
, &t
) >= 0) {
1026 core_session
= strjoina("COREDUMP_SESSION=", t
);
1029 IOVEC_SET_STRING(iovec
[n_iovec
++], core_session
);
1032 if (sd_pid_get_owner_uid(pid
, &owner_uid
) >= 0) {
1033 r
= asprintf(&core_owner_uid
, "COREDUMP_OWNER_UID=" UID_FMT
, owner_uid
);
1035 IOVEC_SET_STRING(iovec
[n_iovec
++], core_owner_uid
);
1038 if (sd_pid_get_slice(pid
, &t
) >= 0) {
1039 core_slice
= strjoina("COREDUMP_SLICE=", t
);
1042 IOVEC_SET_STRING(iovec
[n_iovec
++], core_slice
);
1046 core_comm
= strjoina("COREDUMP_COMM=", comm
);
1047 IOVEC_SET_STRING(iovec
[n_iovec
++], core_comm
);
1051 core_exe
= strjoina("COREDUMP_EXE=", exe
);
1052 IOVEC_SET_STRING(iovec
[n_iovec
++], core_exe
);
1055 if (get_process_cmdline(pid
, 0, false, &t
) >= 0) {
1056 core_cmdline
= strjoina("COREDUMP_CMDLINE=", t
);
1059 IOVEC_SET_STRING(iovec
[n_iovec
++], core_cmdline
);
1062 if (cg_pid_get_path_shifted(pid
, NULL
, &t
) >= 0) {
1063 core_cgroup
= strjoina("COREDUMP_CGROUP=", t
);
1066 IOVEC_SET_STRING(iovec
[n_iovec
++], core_cgroup
);
1069 if (compose_open_fds(pid
, &t
) >= 0) {
1070 core_open_fds
= strappend("COREDUMP_OPEN_FDS=", t
);
1074 IOVEC_SET_STRING(iovec
[n_iovec
++], core_open_fds
);
1077 p
= procfs_file_alloca(pid
, "status");
1078 if (read_full_file(p
, &t
, NULL
) >= 0) {
1079 core_proc_status
= strappend("COREDUMP_PROC_STATUS=", t
);
1082 if (core_proc_status
)
1083 IOVEC_SET_STRING(iovec
[n_iovec
++], core_proc_status
);
1086 p
= procfs_file_alloca(pid
, "maps");
1087 if (read_full_file(p
, &t
, NULL
) >= 0) {
1088 core_proc_maps
= strappend("COREDUMP_PROC_MAPS=", t
);
1092 IOVEC_SET_STRING(iovec
[n_iovec
++], core_proc_maps
);
1095 p
= procfs_file_alloca(pid
, "limits");
1096 if (read_full_file(p
, &t
, NULL
) >= 0) {
1097 core_proc_limits
= strappend("COREDUMP_PROC_LIMITS=", t
);
1100 if (core_proc_limits
)
1101 IOVEC_SET_STRING(iovec
[n_iovec
++], core_proc_limits
);
1104 p
= procfs_file_alloca(pid
, "cgroup");
1105 if (read_full_file(p
, &t
, NULL
) >=0) {
1106 core_proc_cgroup
= strappend("COREDUMP_PROC_CGROUP=", t
);
1109 if (core_proc_cgroup
)
1110 IOVEC_SET_STRING(iovec
[n_iovec
++], core_proc_cgroup
);
1113 if (get_process_cwd(pid
, &t
) >= 0) {
1114 core_cwd
= strjoina("COREDUMP_CWD=", t
);
1117 IOVEC_SET_STRING(iovec
[n_iovec
++], core_cwd
);
1120 if (get_process_root(pid
, &t
) >= 0) {
1121 core_root
= strjoina("COREDUMP_ROOT=", t
);
1124 IOVEC_SET_STRING(iovec
[n_iovec
++], core_root
);
1127 if (get_process_environ(pid
, &t
) >= 0) {
1128 core_environ
= strappend("COREDUMP_ENVIRON=", t
);
1132 IOVEC_SET_STRING(iovec
[n_iovec
++], core_environ
);
1135 core_timestamp
= strjoina("COREDUMP_TIMESTAMP=", context
[CONTEXT_TIMESTAMP
], "000000");
1136 IOVEC_SET_STRING(iovec
[n_iovec
++], core_timestamp
);
1138 IOVEC_SET_STRING(iovec
[n_iovec
++], "MESSAGE_ID=fc2e22bc6ee647b6b90729ab34a250b1");
1140 assert_cc(2 == LOG_CRIT
);
1141 IOVEC_SET_STRING(iovec
[n_iovec
++], "PRIORITY=2");
1143 assert(n_iovec
<= ELEMENTSOF(iovec
));
1145 return send_iovec(iovec
, n_iovec
, STDIN_FILENO
);
1148 int main(int argc
, char *argv
[]) {
1151 /* First, log to a safe place, since we don't know what crashed and it might be journald which we'd rather not
1154 log_set_target(LOG_TARGET_KMSG
);
1157 /* Make sure we never enter a loop */
1158 (void) prctl(PR_SET_DUMPABLE
, 0);
1160 /* Ignore all parse errors */
1161 (void) parse_config();
1163 log_debug("Selected storage '%s'.", coredump_storage_to_string(arg_storage
));
1164 log_debug("Selected compression %s.", yes_no(arg_compress
));
1166 r
= sd_listen_fds(false);
1168 log_error_errno(r
, "Failed to determine number of file descriptor: %m");
1172 /* If we got an fd passed, we are running in coredumpd mode. Otherwise we are invoked from the kernel as
1173 * coredump handler */
1175 r
= process_kernel(argc
, argv
);
1177 r
= process_socket(SD_LISTEN_FDS_START
);
1179 log_error("Received unexpected number of file descriptors.");
1184 return r
< 0 ? EXIT_FAILURE
: EXIT_SUCCESS
;