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 "journald-native.h"
55 #include "parse-util.h"
56 #include "process-util.h"
57 #include "socket-util.h"
59 #include "stacktrace.h"
60 #include "string-table.h"
61 #include "string-util.h"
63 #include "user-util.h"
66 /* The maximum size up to which we process coredumps */
67 #define PROCESS_SIZE_MAX ((uint64_t) (2LLU*1024LLU*1024LLU*1024LLU))
69 /* The maximum size up to which we leave the coredump around on disk */
70 #define EXTERNAL_SIZE_MAX PROCESS_SIZE_MAX
72 /* The maximum size up to which we store the coredump in the journal */
73 #define JOURNAL_SIZE_MAX ((size_t) (767LU*1024LU*1024LU))
75 /* Make sure to not make this larger than the maximum journal entry
76 * size. See DATA_SIZE_MAX in journald-native.c. */
77 assert_cc(JOURNAL_SIZE_MAX
<= DATA_SIZE_MAX
);
80 /* We use this as array indexes for a couple of special fields we use for naming coredumping files, and
93 typedef enum CoredumpStorage
{
94 COREDUMP_STORAGE_NONE
,
95 COREDUMP_STORAGE_EXTERNAL
,
96 COREDUMP_STORAGE_JOURNAL
,
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",
107 DEFINE_PRIVATE_STRING_TABLE_LOOKUP(coredump_storage
, CoredumpStorage
);
108 static DEFINE_CONFIG_PARSE_ENUM(config_parse_coredump_storage
, coredump_storage
, CoredumpStorage
, "Failed to parse storage setting");
110 static CoredumpStorage arg_storage
= COREDUMP_STORAGE_EXTERNAL
;
111 static bool arg_compress
= true;
112 static uint64_t arg_process_size_max
= PROCESS_SIZE_MAX
;
113 static uint64_t arg_external_size_max
= EXTERNAL_SIZE_MAX
;
114 static uint64_t arg_journal_size_max
= JOURNAL_SIZE_MAX
;
115 static uint64_t arg_keep_free
= (uint64_t) -1;
116 static uint64_t arg_max_use
= (uint64_t) -1;
118 static int parse_config(void) {
119 static const ConfigTableItem items
[] = {
120 { "Coredump", "Storage", config_parse_coredump_storage
, 0, &arg_storage
},
121 { "Coredump", "Compress", config_parse_bool
, 0, &arg_compress
},
122 { "Coredump", "ProcessSizeMax", config_parse_iec_uint64
, 0, &arg_process_size_max
},
123 { "Coredump", "ExternalSizeMax", config_parse_iec_uint64
, 0, &arg_external_size_max
},
124 { "Coredump", "JournalSizeMax", config_parse_iec_size
, 0, &arg_journal_size_max
},
125 { "Coredump", "KeepFree", config_parse_iec_uint64
, 0, &arg_keep_free
},
126 { "Coredump", "MaxUse", config_parse_iec_uint64
, 0, &arg_max_use
},
130 return config_parse_many_nulstr(PKGSYSCONFDIR
"/coredump.conf",
131 CONF_PATHS_NULSTR("systemd/coredump.conf.d"),
133 config_item_table_lookup
, items
,
137 static inline uint64_t storage_size_max(void) {
138 return arg_storage
== COREDUMP_STORAGE_EXTERNAL
? arg_external_size_max
: arg_journal_size_max
;
141 static int fix_acl(int fd
, uid_t uid
) {
144 _cleanup_(acl_freep
) acl_t acl
= NULL
;
146 acl_permset_t permset
;
151 if (uid
<= SYSTEM_UID_MAX
)
154 /* Make sure normal users can read (but not write or delete)
155 * their own coredumps */
157 acl
= acl_get_fd(fd
);
159 return log_error_errno(errno
, "Failed to get ACL: %m");
161 if (acl_create_entry(&acl
, &entry
) < 0 ||
162 acl_set_tag_type(entry
, ACL_USER
) < 0 ||
163 acl_set_qualifier(entry
, &uid
) < 0)
164 return log_error_errno(errno
, "Failed to patch ACL: %m");
166 if (acl_get_permset(entry
, &permset
) < 0 ||
167 acl_add_perm(permset
, ACL_READ
) < 0)
168 return log_warning_errno(errno
, "Failed to patch ACL: %m");
170 r
= calc_acl_mask_if_needed(&acl
);
172 return log_warning_errno(r
, "Failed to patch ACL: %m");
174 if (acl_set_fd(fd
, acl
) < 0)
175 return log_error_errno(errno
, "Failed to apply ACL: %m");
181 static int fix_xattr(int fd
, const char *context
[_CONTEXT_MAX
]) {
183 static const char * const xattrs
[_CONTEXT_MAX
] = {
184 [CONTEXT_PID
] = "user.coredump.pid",
185 [CONTEXT_UID
] = "user.coredump.uid",
186 [CONTEXT_GID
] = "user.coredump.gid",
187 [CONTEXT_SIGNAL
] = "user.coredump.signal",
188 [CONTEXT_TIMESTAMP
] = "user.coredump.timestamp",
189 [CONTEXT_COMM
] = "user.coredump.comm",
190 [CONTEXT_EXE
] = "user.coredump.exe",
198 /* Attach some metadata to coredumps via extended
199 * attributes. Just because we can. */
201 for (i
= 0; i
< _CONTEXT_MAX
; i
++) {
204 if (isempty(context
[i
]) || !xattrs
[i
])
207 k
= fsetxattr(fd
, xattrs
[i
], context
[i
], strlen(context
[i
]), XATTR_CREATE
);
215 #define filename_escape(s) xescape((s), "./ ")
217 static inline const char *coredump_tmpfile_name(const char *s
) {
218 return s
? s
: "(unnamed temporary file)";
221 static int fix_permissions(
223 const char *filename
,
225 const char *context
[_CONTEXT_MAX
],
234 /* Ignore errors on these */
235 (void) fchmod(fd
, 0640);
236 (void) fix_acl(fd
, uid
);
237 (void) fix_xattr(fd
, context
);
240 return log_error_errno(errno
, "Failed to sync coredump %s: %m", coredump_tmpfile_name(filename
));
242 r
= link_tmpfile(fd
, filename
, target
);
244 return log_error_errno(r
, "Failed to move coredump %s into place: %m", target
);
249 static int maybe_remove_external_coredump(const char *filename
, uint64_t size
) {
251 /* Returns 1 if might remove, 0 if will not remove, < 0 on error. */
253 if (arg_storage
== COREDUMP_STORAGE_EXTERNAL
&&
254 size
<= arg_external_size_max
)
260 if (unlink(filename
) < 0 && errno
!= ENOENT
)
261 return log_error_errno(errno
, "Failed to unlink %s: %m", filename
);
266 static int make_filename(const char *context
[_CONTEXT_MAX
], char **ret
) {
267 _cleanup_free_
char *c
= NULL
, *u
= NULL
, *p
= NULL
, *t
= NULL
;
268 sd_id128_t boot
= {};
273 c
= filename_escape(context
[CONTEXT_COMM
]);
277 u
= filename_escape(context
[CONTEXT_UID
]);
281 r
= sd_id128_get_boot(&boot
);
285 p
= filename_escape(context
[CONTEXT_PID
]);
289 t
= filename_escape(context
[CONTEXT_TIMESTAMP
]);
294 "/var/lib/systemd/coredump/core.%s.%s." SD_ID128_FORMAT_STR
".%s.%s000000",
297 SD_ID128_FORMAT_VAL(boot
),
305 static int save_external_coredump(
306 const char *context
[_CONTEXT_MAX
],
311 uint64_t *ret_size
) {
313 _cleanup_free_
char *fn
= NULL
, *tmp
= NULL
;
314 _cleanup_close_
int fd
= -1;
315 uint64_t rlimit
, max_size
;
321 assert(ret_filename
);
326 r
= parse_uid(context
[CONTEXT_UID
], &uid
);
328 return log_error_errno(r
, "Failed to parse UID: %m");
330 r
= safe_atou64(context
[CONTEXT_RLIMIT
], &rlimit
);
332 return log_error_errno(r
, "Failed to parse resource limit: %s", context
[CONTEXT_RLIMIT
]);
333 if (rlimit
< page_size()) {
334 /* Is coredumping disabled? Then don't bother saving/processing the coredump.
335 * Anything below PAGE_SIZE cannot give a readable coredump (the kernel uses
336 * ELF_EXEC_PAGESIZE which is not easily accessible, but is usually the same as PAGE_SIZE. */
337 log_info("Resource limits disable core dumping for process %s (%s).",
338 context
[CONTEXT_PID
], context
[CONTEXT_COMM
]);
342 /* Never store more than the process configured, or than we actually shall keep or process */
343 max_size
= MIN(rlimit
, MAX(arg_process_size_max
, storage_size_max()));
345 r
= make_filename(context
, &fn
);
347 return log_error_errno(r
, "Failed to determine coredump file name: %m");
349 mkdir_p_label("/var/lib/systemd/coredump", 0755);
351 fd
= open_tmpfile_linkable(fn
, O_RDWR
|O_CLOEXEC
, &tmp
);
353 return log_error_errno(fd
, "Failed to create temporary file for coredump %s: %m", fn
);
355 r
= copy_bytes(input_fd
, fd
, max_size
, false);
357 log_error_errno(r
, "Cannot store coredump of %s (%s): %m", context
[CONTEXT_PID
], context
[CONTEXT_COMM
]);
361 LOG_MESSAGE("Core file was truncated to %zu bytes.", max_size
),
362 "SIZE_LIMIT=%zu", max_size
,
363 LOG_MESSAGE_ID(SD_MESSAGE_TRUNCATED_CORE
),
366 if (fstat(fd
, &st
) < 0) {
367 log_error_errno(errno
, "Failed to fstat core file %s: %m", coredump_tmpfile_name(tmp
));
371 if (lseek(fd
, 0, SEEK_SET
) == (off_t
) -1) {
372 log_error_errno(errno
, "Failed to seek on %s: %m", coredump_tmpfile_name(tmp
));
376 #if defined(HAVE_XZ) || defined(HAVE_LZ4)
377 /* If we will remove the coredump anyway, do not compress. */
378 if (arg_compress
&& !maybe_remove_external_coredump(NULL
, st
.st_size
)) {
380 _cleanup_free_
char *fn_compressed
= NULL
, *tmp_compressed
= NULL
;
381 _cleanup_close_
int fd_compressed
= -1;
383 fn_compressed
= strappend(fn
, COMPRESSED_EXT
);
384 if (!fn_compressed
) {
389 fd_compressed
= open_tmpfile_linkable(fn_compressed
, O_RDWR
|O_CLOEXEC
, &tmp_compressed
);
390 if (fd_compressed
< 0) {
391 log_error_errno(fd_compressed
, "Failed to create temporary file for coredump %s: %m", fn_compressed
);
395 r
= compress_stream(fd
, fd_compressed
, -1);
397 log_error_errno(r
, "Failed to compress %s: %m", coredump_tmpfile_name(tmp_compressed
));
398 goto fail_compressed
;
401 r
= fix_permissions(fd_compressed
, tmp_compressed
, fn_compressed
, context
, uid
);
403 goto fail_compressed
;
405 /* OK, this worked, we can get rid of the uncompressed version now */
409 *ret_filename
= fn_compressed
; /* compressed */
410 *ret_node_fd
= fd_compressed
; /* compressed */
411 *ret_data_fd
= fd
; /* uncompressed */
412 *ret_size
= (uint64_t) st
.st_size
; /* uncompressed */
414 fn_compressed
= NULL
;
415 fd
= fd_compressed
= -1;
421 (void) unlink(tmp_compressed
);
427 r
= fix_permissions(fd
, tmp
, fn
, context
, uid
);
434 *ret_size
= (uint64_t) st
.st_size
;
447 static int allocate_journal_field(int fd
, size_t size
, char **ret
, size_t *ret_size
) {
448 _cleanup_free_
char *field
= NULL
;
455 if (lseek(fd
, 0, SEEK_SET
) == (off_t
) -1)
456 return log_warning_errno(errno
, "Failed to seek: %m");
458 field
= malloc(9 + size
);
460 log_warning("Failed to allocate memory for coredump, coredump will not be stored.");
464 memcpy(field
, "COREDUMP=", 9);
466 n
= read(fd
, field
+ 9, size
);
468 return log_error_errno((int) n
, "Failed to read core data: %m");
469 if ((size_t) n
< size
) {
470 log_error("Core data too short.");
475 *ret_size
= size
+ 9;
482 /* Joins /proc/[pid]/fd/ and /proc/[pid]/fdinfo/ into the following lines:
496 static int compose_open_fds(pid_t pid
, char **open_fds
) {
497 _cleanup_closedir_
DIR *proc_fd_dir
= NULL
;
498 _cleanup_close_
int proc_fdinfo_fd
= -1;
499 _cleanup_free_
char *buffer
= NULL
;
500 _cleanup_fclose_
FILE *stream
= NULL
;
501 const char *fddelim
= "", *path
;
502 struct dirent
*dent
= NULL
;
507 assert(open_fds
!= NULL
);
509 path
= procfs_file_alloca(pid
, "fd");
510 proc_fd_dir
= opendir(path
);
514 proc_fdinfo_fd
= openat(dirfd(proc_fd_dir
), "../fdinfo", O_DIRECTORY
|O_NOFOLLOW
|O_CLOEXEC
|O_PATH
);
515 if (proc_fdinfo_fd
< 0)
518 stream
= open_memstream(&buffer
, &size
);
522 FOREACH_DIRENT(dent
, proc_fd_dir
, return -errno
) {
523 _cleanup_fclose_
FILE *fdinfo
= NULL
;
524 _cleanup_free_
char *fdname
= NULL
;
528 r
= readlinkat_malloc(dirfd(proc_fd_dir
), dent
->d_name
, &fdname
);
532 fprintf(stream
, "%s%s:%s\n", fddelim
, dent
->d_name
, fdname
);
535 /* Use the directory entry from /proc/[pid]/fd with /proc/[pid]/fdinfo */
536 fd
= openat(proc_fdinfo_fd
, dent
->d_name
, O_NOFOLLOW
|O_CLOEXEC
|O_RDONLY
);
540 fdinfo
= fdopen(fd
, "re");
541 if (fdinfo
== NULL
) {
546 FOREACH_LINE(line
, fdinfo
, break) {
548 if (!endswith(line
, "\n"))
554 stream
= safe_fclose(stream
);
565 static int get_process_ns(pid_t pid
, const char *namespace, ino_t
*ns
) {
568 _cleanup_close_
int proc_ns_dir_fd
;
570 p
= procfs_file_alloca(pid
, "ns");
572 proc_ns_dir_fd
= open(p
, O_DIRECTORY
| O_CLOEXEC
| O_RDONLY
);
573 if (proc_ns_dir_fd
< 0)
576 if (fstatat(proc_ns_dir_fd
, namespace, &stbuf
, /* flags */0) < 0)
583 static int get_mount_namespace_leader(pid_t pid
, pid_t
*container_pid
) {
584 pid_t cpid
= pid
, ppid
= 0;
588 r
= get_process_ns(pid
, "mnt", &proc_mntns
);
595 r
= get_process_ppid(cpid
, &ppid
);
599 r
= get_process_ns(ppid
, "mnt", &parent_mntns
);
603 if (proc_mntns
!= parent_mntns
)
612 *container_pid
= ppid
;
616 /* Returns 1 if the parent was found.
617 * Returns 0 if there is not a process we can call the pid's
618 * container parent (the pid's process isn't 'containerized').
619 * Returns a negative number on errors.
621 static int get_process_container_parent_cmdline(pid_t pid
, char** cmdline
) {
624 const char *proc_root_path
;
625 struct stat root_stat
, proc_root_stat
;
627 /* To compare inodes of / and /proc/[pid]/root */
628 if (stat("/", &root_stat
) < 0)
631 proc_root_path
= procfs_file_alloca(pid
, "root");
632 if (stat(proc_root_path
, &proc_root_stat
) < 0)
635 /* The process uses system root. */
636 if (proc_root_stat
.st_ino
== root_stat
.st_ino
) {
641 r
= get_mount_namespace_leader(pid
, &container_pid
);
645 return get_process_cmdline(container_pid
, 0, false, cmdline
);
648 static int change_uid_gid(const char *context
[]) {
653 r
= parse_uid(context
[CONTEXT_UID
], &uid
);
657 if (uid
<= SYSTEM_UID_MAX
) {
658 const char *user
= "systemd-coredump";
660 r
= get_user_creds(&user
, &uid
, &gid
, NULL
, NULL
);
662 log_warning_errno(r
, "Cannot resolve %s user. Proceeding to dump core as root: %m", user
);
666 r
= parse_gid(context
[CONTEXT_GID
], &gid
);
671 return drop_privileges(uid
, gid
, 0);
674 static int submit_coredump(
675 const char *context
[_CONTEXT_MAX
],
677 size_t n_iovec_allocated
,
681 _cleanup_close_
int coredump_fd
= -1, coredump_node_fd
= -1;
682 _cleanup_free_
char *core_message
= NULL
, *filename
= NULL
, *coredump_data
= NULL
;
683 uint64_t coredump_size
= UINT64_MAX
;
688 assert(n_iovec_allocated
>= n_iovec
+ 3);
689 assert(input_fd
>= 0);
691 /* Vacuum before we write anything again */
692 (void) coredump_vacuum(-1, arg_keep_free
, arg_max_use
);
694 /* Always stream the coredump to disk, if that's possible */
695 r
= save_external_coredump(context
, input_fd
, &filename
, &coredump_node_fd
, &coredump_fd
, &coredump_size
);
697 /* Skip whole core dumping part */
700 /* If we don't want to keep the coredump on disk, remove it now, as later on we will lack the privileges for
701 * it. However, we keep the fd to it, so that we can still process it and log it. */
702 r
= maybe_remove_external_coredump(filename
, coredump_size
);
706 const char *coredump_filename
;
708 coredump_filename
= strjoina("COREDUMP_FILENAME=", filename
);
709 IOVEC_SET_STRING(iovec
[n_iovec
++], coredump_filename
);
710 } else if (arg_storage
== COREDUMP_STORAGE_EXTERNAL
)
711 log_info("The core will not be stored: size %"PRIu64
" is greater than %"PRIu64
" (the configured maximum)",
712 coredump_size
, arg_external_size_max
);
714 /* Vacuum again, but exclude the coredump we just created */
715 (void) coredump_vacuum(coredump_node_fd
>= 0 ? coredump_node_fd
: coredump_fd
, arg_keep_free
, arg_max_use
);
717 /* Now, let's drop privileges to become the user who owns the segfaulted process and allocate the coredump
718 * memory under the user's uid. This also ensures that the credentials journald will see are the ones of the
719 * coredumping user, thus making sure the user gets access to the core dump. Let's also get rid of all
720 * capabilities, if we run as root, we won't need them anymore. */
721 r
= change_uid_gid(context
);
723 return log_error_errno(r
, "Failed to drop privileges: %m");
726 /* Try to get a strack trace if we can */
727 if (coredump_size
<= arg_process_size_max
) {
728 _cleanup_free_
char *stacktrace
= NULL
;
730 r
= coredump_make_stack_trace(coredump_fd
, context
[CONTEXT_EXE
], &stacktrace
);
732 core_message
= strjoin("MESSAGE=Process ", context
[CONTEXT_PID
],
733 " (", context
[CONTEXT_COMM
], ") of user ",
734 context
[CONTEXT_UID
], " dumped core.\n\n",
736 else if (r
== -EINVAL
)
737 log_warning("Failed to generate stack trace: %s", dwfl_errmsg(dwfl_errno()));
739 log_warning_errno(r
, "Failed to generate stack trace: %m");
741 log_debug("Not generating stack trace: core size %"PRIu64
" is greater than %"PRIu64
" (the configured maximum)",
742 coredump_size
, arg_process_size_max
);
747 core_message
= strjoin("MESSAGE=Process ", context
[CONTEXT_PID
], " (",
748 context
[CONTEXT_COMM
], ") of user ",
749 context
[CONTEXT_UID
], " dumped core.");
751 IOVEC_SET_STRING(iovec
[n_iovec
++], core_message
);
753 /* Optionally store the entire coredump in the journal */
754 if (arg_storage
== COREDUMP_STORAGE_JOURNAL
) {
755 if (coredump_size
<= arg_journal_size_max
) {
758 /* Store the coredump itself in the journal */
760 r
= allocate_journal_field(coredump_fd
, (size_t) coredump_size
, &coredump_data
, &sz
);
762 iovec
[n_iovec
].iov_base
= coredump_data
;
763 iovec
[n_iovec
].iov_len
= sz
;
766 log_warning_errno(r
, "Failed to attach the core to the journal entry: %m");
768 log_info("The core will not be stored: size %"PRIu64
" is greater than %"PRIu64
" (the configured maximum)",
769 coredump_size
, arg_journal_size_max
);
772 assert(n_iovec
<= n_iovec_allocated
);
774 r
= sd_journal_sendv(iovec
, n_iovec
);
776 return log_error_errno(r
, "Failed to log coredump: %m");
781 static void map_context_fields(const struct iovec
*iovec
, const char *context
[]) {
783 static const char * const context_field_names
[_CONTEXT_MAX
] = {
784 [CONTEXT_PID
] = "COREDUMP_PID=",
785 [CONTEXT_UID
] = "COREDUMP_UID=",
786 [CONTEXT_GID
] = "COREDUMP_GID=",
787 [CONTEXT_SIGNAL
] = "COREDUMP_SIGNAL=",
788 [CONTEXT_TIMESTAMP
] = "COREDUMP_TIMESTAMP=",
789 [CONTEXT_COMM
] = "COREDUMP_COMM=",
790 [CONTEXT_EXE
] = "COREDUMP_EXE=",
791 [CONTEXT_RLIMIT
] = "COREDUMP_RLIMIT=",
799 for (i
= 0; i
< _CONTEXT_MAX
; i
++) {
802 l
= strlen(context_field_names
[i
]);
803 if (iovec
->iov_len
< l
)
806 if (memcmp(iovec
->iov_base
, context_field_names
[i
], l
) != 0)
809 /* Note that these strings are NUL terminated, because we made sure that a trailing NUL byte is in the
810 * buffer, though not included in the iov_len count. (see below) */
811 context
[i
] = (char*) iovec
->iov_base
+ l
;
816 static int process_socket(int fd
) {
817 _cleanup_close_
int coredump_fd
= -1;
818 struct iovec
*iovec
= NULL
;
819 size_t n_iovec
= 0, n_iovec_allocated
= 0, i
;
820 const char *context
[_CONTEXT_MAX
] = {};
825 log_set_target(LOG_TARGET_AUTO
);
826 log_parse_environment();
831 struct cmsghdr cmsghdr
;
832 uint8_t buf
[CMSG_SPACE(sizeof(int))];
835 .msg_control
= &control
,
836 .msg_controllen
= sizeof(control
),
842 if (!GREEDY_REALLOC(iovec
, n_iovec_allocated
, n_iovec
+ 3)) {
847 l
= next_datagram_size_fd(fd
);
849 r
= log_error_errno(l
, "Failed to determine datagram size to read: %m");
855 iovec
[n_iovec
].iov_len
= l
;
856 iovec
[n_iovec
].iov_base
= malloc(l
+ 1);
857 if (!iovec
[n_iovec
].iov_base
) {
862 mh
.msg_iov
= iovec
+ n_iovec
;
864 n
= recvmsg(fd
, &mh
, MSG_NOSIGNAL
|MSG_CMSG_CLOEXEC
);
866 free(iovec
[n_iovec
].iov_base
);
867 r
= log_error_errno(errno
, "Failed to receive datagram: %m");
872 struct cmsghdr
*cmsg
, *found
= NULL
;
873 /* The final zero-length datagram carries the file descriptor and tells us that we're done. */
875 free(iovec
[n_iovec
].iov_base
);
877 CMSG_FOREACH(cmsg
, &mh
) {
878 if (cmsg
->cmsg_level
== SOL_SOCKET
&&
879 cmsg
->cmsg_type
== SCM_RIGHTS
&&
880 cmsg
->cmsg_len
== CMSG_LEN(sizeof(int))) {
887 log_error("Coredump file descriptor missing.");
892 assert(coredump_fd
< 0);
893 coredump_fd
= *(int*) CMSG_DATA(found
);
897 /* Add trailing NUL byte, in case these are strings */
898 ((char*) iovec
[n_iovec
].iov_base
)[n
] = 0;
899 iovec
[n_iovec
].iov_len
= (size_t) n
;
902 map_context_fields(iovec
+ n_iovec
, context
);
906 if (!GREEDY_REALLOC(iovec
, n_iovec_allocated
, n_iovec
+ 3)) {
911 /* Make sure we got all data we really need */
912 assert(context
[CONTEXT_PID
]);
913 assert(context
[CONTEXT_UID
]);
914 assert(context
[CONTEXT_GID
]);
915 assert(context
[CONTEXT_SIGNAL
]);
916 assert(context
[CONTEXT_TIMESTAMP
]);
917 assert(context
[CONTEXT_RLIMIT
]);
918 assert(context
[CONTEXT_COMM
]);
919 assert(coredump_fd
>= 0);
921 r
= submit_coredump(context
, iovec
, n_iovec_allocated
, n_iovec
, coredump_fd
);
924 for (i
= 0; i
< n_iovec
; i
++)
925 free(iovec
[i
].iov_base
);
931 static int send_iovec(const struct iovec iovec
[], size_t n_iovec
, int input_fd
) {
933 static const union sockaddr_union sa
= {
934 .un
.sun_family
= AF_UNIX
,
935 .un
.sun_path
= "/run/systemd/coredump",
937 _cleanup_close_
int fd
= -1;
941 assert(iovec
|| n_iovec
<= 0);
942 assert(input_fd
>= 0);
944 fd
= socket(AF_UNIX
, SOCK_SEQPACKET
|SOCK_CLOEXEC
, 0);
946 return log_error_errno(errno
, "Failed to create coredump socket: %m");
948 if (connect(fd
, &sa
.sa
, SOCKADDR_UN_LEN(sa
.un
)) < 0)
949 return log_error_errno(errno
, "Failed to connect to coredump service: %m");
951 for (i
= 0; i
< n_iovec
; i
++) {
953 .msg_iov
= (struct iovec
*) iovec
+ i
,
956 struct iovec copy
[2];
959 if (sendmsg(fd
, &mh
, MSG_NOSIGNAL
) >= 0)
962 if (errno
== EMSGSIZE
&& mh
.msg_iov
[0].iov_len
> 0) {
963 /* This field didn't fit? That's a pity. Given that this is just metadata,
964 * let's truncate the field at half, and try again. We append three dots, in
965 * order to show that this is truncated. */
967 if (mh
.msg_iov
!= copy
) {
968 /* We don't want to modify the caller's iovec, hence let's create our
969 * own array, consisting of two new iovecs, where the first is a
970 * (truncated) copy of what we want to send, and the second one
971 * contains the trailing dots. */
973 copy
[1] = (struct iovec
) {
974 .iov_base
= (char[]) { '.', '.', '.' },
982 copy
[0].iov_len
/= 2; /* halve it, and try again */
986 return log_error_errno(errno
, "Failed to send coredump datagram: %m");
990 r
= send_one_fd(fd
, input_fd
, 0);
992 return log_error_errno(r
, "Failed to send coredump fd: %m");
997 static int process_special_crash(const char *context
[], int input_fd
) {
998 _cleanup_close_
int coredump_fd
= -1, coredump_node_fd
= -1;
999 _cleanup_free_
char *filename
= NULL
;
1000 uint64_t coredump_size
;
1004 assert(input_fd
>= 0);
1006 /* If we are pid1 or journald, we cut things short, don't write to the journal, but still create a coredump. */
1008 if (arg_storage
!= COREDUMP_STORAGE_NONE
)
1009 arg_storage
= COREDUMP_STORAGE_EXTERNAL
;
1011 r
= save_external_coredump(context
, input_fd
, &filename
, &coredump_node_fd
, &coredump_fd
, &coredump_size
);
1015 r
= maybe_remove_external_coredump(filename
, coredump_size
);
1019 log_notice("Detected coredump of the journal daemon or PID 1, diverted to %s.", filename
);
1024 static int process_kernel(int argc
, char* argv
[]) {
1026 /* The small core field we allocate on the stack, to keep things simple */
1028 *core_pid
= NULL
, *core_uid
= NULL
, *core_gid
= NULL
, *core_signal
= NULL
,
1029 *core_session
= NULL
, *core_exe
= NULL
, *core_comm
= NULL
, *core_cmdline
= NULL
,
1030 *core_cgroup
= NULL
, *core_cwd
= NULL
, *core_root
= NULL
, *core_unit
= NULL
,
1031 *core_user_unit
= NULL
, *core_slice
= NULL
, *core_timestamp
= NULL
, *core_rlimit
= NULL
;
1033 /* The larger ones we allocate on the heap */
1035 *core_owner_uid
= NULL
, *core_open_fds
= NULL
, *core_proc_status
= NULL
,
1036 *core_proc_maps
= NULL
, *core_proc_limits
= NULL
, *core_proc_cgroup
= NULL
, *core_environ
= NULL
,
1037 *core_proc_mountinfo
= NULL
, *core_container_cmdline
= NULL
;
1039 _cleanup_free_
char *exe
= NULL
, *comm
= NULL
;
1040 const char *context
[_CONTEXT_MAX
];
1041 bool proc_self_root_is_slash
;
1042 struct iovec iovec
[27];
1050 if (argc
< CONTEXT_COMM
+ 1) {
1051 log_error("Not enough arguments passed from kernel (%i, expected %i).", argc
- 1, CONTEXT_COMM
+ 1 - 1);
1055 r
= parse_pid(argv
[CONTEXT_PID
+ 1], &pid
);
1057 return log_error_errno(r
, "Failed to parse PID.");
1059 r
= get_process_comm(pid
, &comm
);
1061 log_warning_errno(r
, "Failed to get COMM, falling back to the command line: %m");
1062 comm
= strv_join(argv
+ CONTEXT_COMM
+ 1, " ");
1067 r
= get_process_exe(pid
, &exe
);
1069 log_warning_errno(r
, "Failed to get EXE, ignoring: %m");
1071 context
[CONTEXT_PID
] = argv
[CONTEXT_PID
+ 1];
1072 context
[CONTEXT_UID
] = argv
[CONTEXT_UID
+ 1];
1073 context
[CONTEXT_GID
] = argv
[CONTEXT_GID
+ 1];
1074 context
[CONTEXT_SIGNAL
] = argv
[CONTEXT_SIGNAL
+ 1];
1075 context
[CONTEXT_TIMESTAMP
] = argv
[CONTEXT_TIMESTAMP
+ 1];
1076 context
[CONTEXT_RLIMIT
] = argv
[CONTEXT_RLIMIT
+ 1];
1077 context
[CONTEXT_COMM
] = comm
;
1078 context
[CONTEXT_EXE
] = exe
;
1080 if (cg_pid_get_unit(pid
, &t
) >= 0) {
1082 /* If this is PID 1 disable coredump collection, we'll unlikely be able to process it later on. */
1083 if (streq(t
, SPECIAL_INIT_SCOPE
)) {
1084 log_notice("Due to PID 1 having crashed coredump collection will now be turned off.");
1085 (void) write_string_file("/proc/sys/kernel/core_pattern", "|/bin/false", 0);
1088 /* Let's avoid dead-locks when processing journald and init crashes, as socket activation and logging
1089 * are unlikely to work then. */
1090 if (STR_IN_SET(t
, SPECIAL_JOURNALD_SERVICE
, SPECIAL_INIT_SCOPE
)) {
1092 return process_special_crash(context
, STDIN_FILENO
);
1095 core_unit
= strjoina("COREDUMP_UNIT=", t
);
1098 IOVEC_SET_STRING(iovec
[n_iovec
++], core_unit
);
1101 /* OK, now we know it's not the journal, hence we can make use of it now. */
1102 log_set_target(LOG_TARGET_JOURNAL_OR_KMSG
);
1105 if (cg_pid_get_user_unit(pid
, &t
) >= 0) {
1106 core_user_unit
= strjoina("COREDUMP_USER_UNIT=", t
);
1109 IOVEC_SET_STRING(iovec
[n_iovec
++], core_user_unit
);
1112 core_pid
= strjoina("COREDUMP_PID=", context
[CONTEXT_PID
]);
1113 IOVEC_SET_STRING(iovec
[n_iovec
++], core_pid
);
1115 core_uid
= strjoina("COREDUMP_UID=", context
[CONTEXT_UID
]);
1116 IOVEC_SET_STRING(iovec
[n_iovec
++], core_uid
);
1118 core_gid
= strjoina("COREDUMP_GID=", context
[CONTEXT_GID
]);
1119 IOVEC_SET_STRING(iovec
[n_iovec
++], core_gid
);
1121 core_signal
= strjoina("COREDUMP_SIGNAL=", context
[CONTEXT_SIGNAL
]);
1122 IOVEC_SET_STRING(iovec
[n_iovec
++], core_signal
);
1124 core_rlimit
= strjoina("COREDUMP_RLIMIT=", context
[CONTEXT_RLIMIT
]);
1125 IOVEC_SET_STRING(iovec
[n_iovec
++], core_rlimit
);
1127 if (sd_pid_get_session(pid
, &t
) >= 0) {
1128 core_session
= strjoina("COREDUMP_SESSION=", t
);
1131 IOVEC_SET_STRING(iovec
[n_iovec
++], core_session
);
1134 if (sd_pid_get_owner_uid(pid
, &owner_uid
) >= 0) {
1135 r
= asprintf(&core_owner_uid
, "COREDUMP_OWNER_UID=" UID_FMT
, owner_uid
);
1137 IOVEC_SET_STRING(iovec
[n_iovec
++], core_owner_uid
);
1140 if (sd_pid_get_slice(pid
, &t
) >= 0) {
1141 core_slice
= strjoina("COREDUMP_SLICE=", t
);
1144 IOVEC_SET_STRING(iovec
[n_iovec
++], core_slice
);
1148 core_comm
= strjoina("COREDUMP_COMM=", comm
);
1149 IOVEC_SET_STRING(iovec
[n_iovec
++], core_comm
);
1153 core_exe
= strjoina("COREDUMP_EXE=", exe
);
1154 IOVEC_SET_STRING(iovec
[n_iovec
++], core_exe
);
1157 if (get_process_cmdline(pid
, 0, false, &t
) >= 0) {
1158 core_cmdline
= strjoina("COREDUMP_CMDLINE=", t
);
1161 IOVEC_SET_STRING(iovec
[n_iovec
++], core_cmdline
);
1164 if (cg_pid_get_path_shifted(pid
, NULL
, &t
) >= 0) {
1165 core_cgroup
= strjoina("COREDUMP_CGROUP=", t
);
1168 IOVEC_SET_STRING(iovec
[n_iovec
++], core_cgroup
);
1171 if (compose_open_fds(pid
, &t
) >= 0) {
1172 core_open_fds
= strappend("COREDUMP_OPEN_FDS=", t
);
1176 IOVEC_SET_STRING(iovec
[n_iovec
++], core_open_fds
);
1179 p
= procfs_file_alloca(pid
, "status");
1180 if (read_full_file(p
, &t
, NULL
) >= 0) {
1181 core_proc_status
= strappend("COREDUMP_PROC_STATUS=", t
);
1184 if (core_proc_status
)
1185 IOVEC_SET_STRING(iovec
[n_iovec
++], core_proc_status
);
1188 p
= procfs_file_alloca(pid
, "maps");
1189 if (read_full_file(p
, &t
, NULL
) >= 0) {
1190 core_proc_maps
= strappend("COREDUMP_PROC_MAPS=", t
);
1194 IOVEC_SET_STRING(iovec
[n_iovec
++], core_proc_maps
);
1197 p
= procfs_file_alloca(pid
, "limits");
1198 if (read_full_file(p
, &t
, NULL
) >= 0) {
1199 core_proc_limits
= strappend("COREDUMP_PROC_LIMITS=", t
);
1202 if (core_proc_limits
)
1203 IOVEC_SET_STRING(iovec
[n_iovec
++], core_proc_limits
);
1206 p
= procfs_file_alloca(pid
, "cgroup");
1207 if (read_full_file(p
, &t
, NULL
) >=0) {
1208 core_proc_cgroup
= strappend("COREDUMP_PROC_CGROUP=", t
);
1211 if (core_proc_cgroup
)
1212 IOVEC_SET_STRING(iovec
[n_iovec
++], core_proc_cgroup
);
1215 p
= procfs_file_alloca(pid
, "mountinfo");
1216 if (read_full_file(p
, &t
, NULL
) >=0) {
1217 core_proc_mountinfo
= strappend("COREDUMP_PROC_MOUNTINFO=", t
);
1220 if (core_proc_mountinfo
)
1221 IOVEC_SET_STRING(iovec
[n_iovec
++], core_proc_mountinfo
);
1224 if (get_process_cwd(pid
, &t
) >= 0) {
1225 core_cwd
= strjoina("COREDUMP_CWD=", t
);
1228 IOVEC_SET_STRING(iovec
[n_iovec
++], core_cwd
);
1231 if (get_process_root(pid
, &t
) >= 0) {
1232 core_root
= strjoina("COREDUMP_ROOT=", t
);
1234 IOVEC_SET_STRING(iovec
[n_iovec
++], core_root
);
1236 /* If the process' root is "/", then there is a chance it has
1237 * mounted own root and hence being containerized. */
1238 proc_self_root_is_slash
= strcmp(t
, "/") == 0;
1240 if (proc_self_root_is_slash
&& get_process_container_parent_cmdline(pid
, &t
) > 0) {
1241 core_container_cmdline
= strappend("COREDUMP_CONTAINER_CMDLINE=", t
);
1244 if (core_container_cmdline
)
1245 IOVEC_SET_STRING(iovec
[n_iovec
++], core_container_cmdline
);
1249 if (get_process_environ(pid
, &t
) >= 0) {
1250 core_environ
= strappend("COREDUMP_ENVIRON=", t
);
1254 IOVEC_SET_STRING(iovec
[n_iovec
++], core_environ
);
1257 core_timestamp
= strjoina("COREDUMP_TIMESTAMP=", context
[CONTEXT_TIMESTAMP
], "000000");
1258 IOVEC_SET_STRING(iovec
[n_iovec
++], core_timestamp
);
1260 IOVEC_SET_STRING(iovec
[n_iovec
++], "MESSAGE_ID=fc2e22bc6ee647b6b90729ab34a250b1");
1262 assert_cc(2 == LOG_CRIT
);
1263 IOVEC_SET_STRING(iovec
[n_iovec
++], "PRIORITY=2");
1265 assert(n_iovec
<= ELEMENTSOF(iovec
));
1267 return send_iovec(iovec
, n_iovec
, STDIN_FILENO
);
1270 int main(int argc
, char *argv
[]) {
1273 /* First, log to a safe place, since we don't know what crashed and it might be journald which we'd rather not
1276 log_set_target(LOG_TARGET_KMSG
);
1279 /* Make sure we never enter a loop */
1280 (void) prctl(PR_SET_DUMPABLE
, 0);
1282 /* Ignore all parse errors */
1283 (void) parse_config();
1285 log_debug("Selected storage '%s'.", coredump_storage_to_string(arg_storage
));
1286 log_debug("Selected compression %s.", yes_no(arg_compress
));
1288 r
= sd_listen_fds(false);
1290 log_error_errno(r
, "Failed to determine number of file descriptor: %m");
1294 /* If we got an fd passed, we are running in coredumpd mode. Otherwise we are invoked from the kernel as
1295 * coredump handler */
1297 r
= process_kernel(argc
, argv
);
1299 r
= process_socket(SD_LISTEN_FDS_START
);
1301 log_error("Received unexpected number of file descriptors.");
1306 return r
< 0 ? EXIT_FAILURE
: EXIT_SUCCESS
;