]> git.ipfire.org Git - thirdparty/systemd.git/blob - src/coredump/coredump.c
Merge pull request #10736 from yuwata/coredump-comment
[thirdparty/systemd.git] / src / coredump / coredump.c
1 /* SPDX-License-Identifier: LGPL-2.1+ */
2
3 #include <errno.h>
4 #include <stdio.h>
5 #include <stdio_ext.h>
6 #include <sys/prctl.h>
7 #include <sys/xattr.h>
8 #include <unistd.h>
9
10 #if HAVE_ELFUTILS
11 #include <dwarf.h>
12 #include <elfutils/libdwfl.h>
13 #endif
14
15 #include "sd-daemon.h"
16 #include "sd-journal.h"
17 #include "sd-login.h"
18 #include "sd-messages.h"
19
20 #include "acl-util.h"
21 #include "alloc-util.h"
22 #include "capability-util.h"
23 #include "cgroup-util.h"
24 #include "compress.h"
25 #include "conf-parser.h"
26 #include "copy.h"
27 #include "coredump-vacuum.h"
28 #include "dirent-util.h"
29 #include "escape.h"
30 #include "fd-util.h"
31 #include "fileio.h"
32 #include "fs-util.h"
33 #include "io-util.h"
34 #include "journal-importer.h"
35 #include "log.h"
36 #include "macro.h"
37 #include "missing.h"
38 #include "mkdir.h"
39 #include "parse-util.h"
40 #include "process-util.h"
41 #include "signal-util.h"
42 #include "socket-util.h"
43 #include "special.h"
44 #include "stacktrace.h"
45 #include "string-table.h"
46 #include "string-util.h"
47 #include "strv.h"
48 #include "user-util.h"
49 #include "util.h"
50
51 /* The maximum size up to which we process coredumps */
52 #define PROCESS_SIZE_MAX ((uint64_t) (2LLU*1024LLU*1024LLU*1024LLU))
53
54 /* The maximum size up to which we leave the coredump around on disk */
55 #define EXTERNAL_SIZE_MAX PROCESS_SIZE_MAX
56
57 /* The maximum size up to which we store the coredump in the journal */
58 #ifndef FUZZING_BUILD_MODE_UNSAFE_FOR_PRODUCTION
59 #define JOURNAL_SIZE_MAX ((size_t) (767LU*1024LU*1024LU))
60 #else
61 /* oss-fuzz limits memory usage. */
62 #define JOURNAL_SIZE_MAX ((size_t) (10LU*1024LU*1024LU))
63 #endif
64
65 /* Make sure to not make this larger than the maximum journal entry
66 * size. See DATA_SIZE_MAX in journal-importer.h. */
67 assert_cc(JOURNAL_SIZE_MAX <= DATA_SIZE_MAX);
68
69 enum {
70 /* We use this as array indexes for a couple of special fields we use for
71 * naming coredump files, and attaching xattrs, and for indexing argv[].
72
73 * Our pattern for man:systectl(1) kernel.core_pattern is such that the
74 * kernel passes fields until CONTEXT_RLIMIT as arguments in argv[]. After
75 * that it gets complicated: the kernel passes "comm" as one or more fields
76 * starting at index CONTEXT_COMM (in other words, full "comm" is under index
77 * CONTEXT_COMM when it does not contain spaces, which is the common
78 * case). This mapping is not reversible, so we prefer to retrieve "comm"
79 * from /proc. We only fall back to argv[CONTEXT_COMM...] when that fails.
80 *
81 * In the internal context[] array, fields before CONTEXT_COMM are the
82 * strings from argv[], so they should not be freed. The strings at indices
83 * CONTEXT_COMM and higher are allocated by us and should be freed at the
84 * end.
85 */
86 CONTEXT_PID,
87 CONTEXT_UID,
88 CONTEXT_GID,
89 CONTEXT_SIGNAL,
90 CONTEXT_TIMESTAMP,
91 CONTEXT_RLIMIT,
92 CONTEXT_HOSTNAME,
93 CONTEXT_COMM,
94 CONTEXT_EXE,
95 CONTEXT_UNIT,
96 _CONTEXT_MAX
97 };
98
99 typedef enum CoredumpStorage {
100 COREDUMP_STORAGE_NONE,
101 COREDUMP_STORAGE_EXTERNAL,
102 COREDUMP_STORAGE_JOURNAL,
103 _COREDUMP_STORAGE_MAX,
104 _COREDUMP_STORAGE_INVALID = -1
105 } CoredumpStorage;
106
107 static const char* const coredump_storage_table[_COREDUMP_STORAGE_MAX] = {
108 [COREDUMP_STORAGE_NONE] = "none",
109 [COREDUMP_STORAGE_EXTERNAL] = "external",
110 [COREDUMP_STORAGE_JOURNAL] = "journal",
111 };
112
113 DEFINE_PRIVATE_STRING_TABLE_LOOKUP(coredump_storage, CoredumpStorage);
114 static DEFINE_CONFIG_PARSE_ENUM(config_parse_coredump_storage, coredump_storage, CoredumpStorage, "Failed to parse storage setting");
115
116 static CoredumpStorage arg_storage = COREDUMP_STORAGE_EXTERNAL;
117 static bool arg_compress = true;
118 static uint64_t arg_process_size_max = PROCESS_SIZE_MAX;
119 static uint64_t arg_external_size_max = EXTERNAL_SIZE_MAX;
120 static uint64_t arg_journal_size_max = JOURNAL_SIZE_MAX;
121 static uint64_t arg_keep_free = (uint64_t) -1;
122 static uint64_t arg_max_use = (uint64_t) -1;
123
124 static int parse_config(void) {
125 static const ConfigTableItem items[] = {
126 { "Coredump", "Storage", config_parse_coredump_storage, 0, &arg_storage },
127 { "Coredump", "Compress", config_parse_bool, 0, &arg_compress },
128 { "Coredump", "ProcessSizeMax", config_parse_iec_uint64, 0, &arg_process_size_max },
129 { "Coredump", "ExternalSizeMax", config_parse_iec_uint64, 0, &arg_external_size_max },
130 { "Coredump", "JournalSizeMax", config_parse_iec_size, 0, &arg_journal_size_max },
131 { "Coredump", "KeepFree", config_parse_iec_uint64, 0, &arg_keep_free },
132 { "Coredump", "MaxUse", config_parse_iec_uint64, 0, &arg_max_use },
133 {}
134 };
135
136 return config_parse_many_nulstr(PKGSYSCONFDIR "/coredump.conf",
137 CONF_PATHS_NULSTR("systemd/coredump.conf.d"),
138 "Coredump\0",
139 config_item_table_lookup, items,
140 CONFIG_PARSE_WARN, NULL);
141 }
142
143 static inline uint64_t storage_size_max(void) {
144 if (arg_storage == COREDUMP_STORAGE_EXTERNAL)
145 return arg_external_size_max;
146 if (arg_storage == COREDUMP_STORAGE_JOURNAL)
147 return arg_journal_size_max;
148 assert(arg_storage == COREDUMP_STORAGE_NONE);
149 return 0;
150 }
151
152 static int fix_acl(int fd, uid_t uid) {
153
154 #if HAVE_ACL
155 _cleanup_(acl_freep) acl_t acl = NULL;
156 acl_entry_t entry;
157 acl_permset_t permset;
158 int r;
159
160 assert(fd >= 0);
161
162 if (uid_is_system(uid) || uid_is_dynamic(uid) || uid == UID_NOBODY)
163 return 0;
164
165 /* Make sure normal users can read (but not write or delete)
166 * their own coredumps */
167
168 acl = acl_get_fd(fd);
169 if (!acl)
170 return log_error_errno(errno, "Failed to get ACL: %m");
171
172 if (acl_create_entry(&acl, &entry) < 0 ||
173 acl_set_tag_type(entry, ACL_USER) < 0 ||
174 acl_set_qualifier(entry, &uid) < 0)
175 return log_error_errno(errno, "Failed to patch ACL: %m");
176
177 if (acl_get_permset(entry, &permset) < 0 ||
178 acl_add_perm(permset, ACL_READ) < 0)
179 return log_warning_errno(errno, "Failed to patch ACL: %m");
180
181 r = calc_acl_mask_if_needed(&acl);
182 if (r < 0)
183 return log_warning_errno(r, "Failed to patch ACL: %m");
184
185 if (acl_set_fd(fd, acl) < 0)
186 return log_error_errno(errno, "Failed to apply ACL: %m");
187 #endif
188
189 return 0;
190 }
191
192 static int fix_xattr(int fd, const char *context[_CONTEXT_MAX]) {
193
194 static const char * const xattrs[_CONTEXT_MAX] = {
195 [CONTEXT_PID] = "user.coredump.pid",
196 [CONTEXT_UID] = "user.coredump.uid",
197 [CONTEXT_GID] = "user.coredump.gid",
198 [CONTEXT_SIGNAL] = "user.coredump.signal",
199 [CONTEXT_TIMESTAMP] = "user.coredump.timestamp",
200 [CONTEXT_RLIMIT] = "user.coredump.rlimit",
201 [CONTEXT_HOSTNAME] = "user.coredump.hostname",
202 [CONTEXT_COMM] = "user.coredump.comm",
203 [CONTEXT_EXE] = "user.coredump.exe",
204 };
205
206 int r = 0;
207 unsigned i;
208
209 assert(fd >= 0);
210
211 /* Attach some metadata to coredumps via extended
212 * attributes. Just because we can. */
213
214 for (i = 0; i < _CONTEXT_MAX; i++) {
215 int k;
216
217 if (isempty(context[i]) || !xattrs[i])
218 continue;
219
220 k = fsetxattr(fd, xattrs[i], context[i], strlen(context[i]), XATTR_CREATE);
221 if (k < 0 && r == 0)
222 r = -errno;
223 }
224
225 return r;
226 }
227
228 #define filename_escape(s) xescape((s), "./ ")
229
230 static inline const char *coredump_tmpfile_name(const char *s) {
231 return s ? s : "(unnamed temporary file)";
232 }
233
234 static int fix_permissions(
235 int fd,
236 const char *filename,
237 const char *target,
238 const char *context[_CONTEXT_MAX],
239 uid_t uid) {
240
241 int r;
242
243 assert(fd >= 0);
244 assert(target);
245 assert(context);
246
247 /* Ignore errors on these */
248 (void) fchmod(fd, 0640);
249 (void) fix_acl(fd, uid);
250 (void) fix_xattr(fd, context);
251
252 if (fsync(fd) < 0)
253 return log_error_errno(errno, "Failed to sync coredump %s: %m", coredump_tmpfile_name(filename));
254
255 (void) fsync_directory_of_file(fd);
256
257 r = link_tmpfile(fd, filename, target);
258 if (r < 0)
259 return log_error_errno(r, "Failed to move coredump %s into place: %m", target);
260
261 return 0;
262 }
263
264 static int maybe_remove_external_coredump(const char *filename, uint64_t size) {
265
266 /* Returns 1 if might remove, 0 if will not remove, < 0 on error. */
267
268 if (arg_storage == COREDUMP_STORAGE_EXTERNAL &&
269 size <= arg_external_size_max)
270 return 0;
271
272 if (!filename)
273 return 1;
274
275 if (unlink(filename) < 0 && errno != ENOENT)
276 return log_error_errno(errno, "Failed to unlink %s: %m", filename);
277
278 return 1;
279 }
280
281 static int make_filename(const char *context[_CONTEXT_MAX], char **ret) {
282 _cleanup_free_ char *c = NULL, *u = NULL, *p = NULL, *t = NULL;
283 sd_id128_t boot = {};
284 int r;
285
286 assert(context);
287
288 c = filename_escape(context[CONTEXT_COMM]);
289 if (!c)
290 return -ENOMEM;
291
292 u = filename_escape(context[CONTEXT_UID]);
293 if (!u)
294 return -ENOMEM;
295
296 r = sd_id128_get_boot(&boot);
297 if (r < 0)
298 return r;
299
300 p = filename_escape(context[CONTEXT_PID]);
301 if (!p)
302 return -ENOMEM;
303
304 t = filename_escape(context[CONTEXT_TIMESTAMP]);
305 if (!t)
306 return -ENOMEM;
307
308 if (asprintf(ret,
309 "/var/lib/systemd/coredump/core.%s.%s." SD_ID128_FORMAT_STR ".%s.%s000000",
310 c,
311 u,
312 SD_ID128_FORMAT_VAL(boot),
313 p,
314 t) < 0)
315 return -ENOMEM;
316
317 return 0;
318 }
319
320 static int save_external_coredump(
321 const char *context[_CONTEXT_MAX],
322 int input_fd,
323 char **ret_filename,
324 int *ret_node_fd,
325 int *ret_data_fd,
326 uint64_t *ret_size,
327 bool *ret_truncated) {
328
329 _cleanup_free_ char *fn = NULL, *tmp = NULL;
330 _cleanup_close_ int fd = -1;
331 uint64_t rlimit, process_limit, max_size;
332 struct stat st;
333 uid_t uid;
334 int r;
335
336 assert(context);
337 assert(ret_filename);
338 assert(ret_node_fd);
339 assert(ret_data_fd);
340 assert(ret_size);
341
342 r = parse_uid(context[CONTEXT_UID], &uid);
343 if (r < 0)
344 return log_error_errno(r, "Failed to parse UID: %m");
345
346 r = safe_atou64(context[CONTEXT_RLIMIT], &rlimit);
347 if (r < 0)
348 return log_error_errno(r, "Failed to parse resource limit '%s': %m", 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]);
355 return -EBADSLT;
356 }
357
358 process_limit = MAX(arg_process_size_max, storage_size_max());
359 if (process_limit == 0) {
360 log_debug("Limits for coredump processing and storage are both 0, not dumping core.");
361 return -EBADSLT;
362 }
363
364 /* Never store more than the process configured, or than we actually shall keep or process */
365 max_size = MIN(rlimit, process_limit);
366
367 r = make_filename(context, &fn);
368 if (r < 0)
369 return log_error_errno(r, "Failed to determine coredump file name: %m");
370
371 mkdir_p_label("/var/lib/systemd/coredump", 0755);
372
373 fd = open_tmpfile_linkable(fn, O_RDWR|O_CLOEXEC, &tmp);
374 if (fd < 0)
375 return log_error_errno(fd, "Failed to create temporary file for coredump %s: %m", fn);
376
377 r = copy_bytes(input_fd, fd, max_size, 0);
378 if (r < 0) {
379 log_error_errno(r, "Cannot store coredump of %s (%s): %m", context[CONTEXT_PID], context[CONTEXT_COMM]);
380 goto fail;
381 }
382 *ret_truncated = r == 1;
383 if (*ret_truncated)
384 log_struct(LOG_INFO,
385 LOG_MESSAGE("Core file was truncated to %zu bytes.", max_size),
386 "SIZE_LIMIT=%zu", max_size,
387 "MESSAGE_ID=" SD_MESSAGE_TRUNCATED_CORE_STR);
388
389 if (fstat(fd, &st) < 0) {
390 log_error_errno(errno, "Failed to fstat core file %s: %m", coredump_tmpfile_name(tmp));
391 goto fail;
392 }
393
394 if (lseek(fd, 0, SEEK_SET) == (off_t) -1) {
395 log_error_errno(errno, "Failed to seek on %s: %m", coredump_tmpfile_name(tmp));
396 goto fail;
397 }
398
399 #if HAVE_XZ || HAVE_LZ4
400 /* If we will remove the coredump anyway, do not compress. */
401 if (arg_compress && !maybe_remove_external_coredump(NULL, st.st_size)) {
402
403 _cleanup_free_ char *fn_compressed = NULL, *tmp_compressed = NULL;
404 _cleanup_close_ int fd_compressed = -1;
405
406 fn_compressed = strappend(fn, COMPRESSED_EXT);
407 if (!fn_compressed) {
408 log_oom();
409 goto uncompressed;
410 }
411
412 fd_compressed = open_tmpfile_linkable(fn_compressed, O_RDWR|O_CLOEXEC, &tmp_compressed);
413 if (fd_compressed < 0) {
414 log_error_errno(fd_compressed, "Failed to create temporary file for coredump %s: %m", fn_compressed);
415 goto uncompressed;
416 }
417
418 r = compress_stream(fd, fd_compressed, -1);
419 if (r < 0) {
420 log_error_errno(r, "Failed to compress %s: %m", coredump_tmpfile_name(tmp_compressed));
421 goto fail_compressed;
422 }
423
424 r = fix_permissions(fd_compressed, tmp_compressed, fn_compressed, context, uid);
425 if (r < 0)
426 goto fail_compressed;
427
428 /* OK, this worked, we can get rid of the uncompressed version now */
429 if (tmp)
430 unlink_noerrno(tmp);
431
432 *ret_filename = TAKE_PTR(fn_compressed); /* compressed */
433 *ret_node_fd = TAKE_FD(fd_compressed); /* compressed */
434 *ret_data_fd = TAKE_FD(fd); /* uncompressed */
435 *ret_size = (uint64_t) st.st_size; /* uncompressed */
436
437 return 0;
438
439 fail_compressed:
440 if (tmp_compressed)
441 (void) unlink(tmp_compressed);
442 }
443
444 uncompressed:
445 #endif
446
447 r = fix_permissions(fd, tmp, fn, context, uid);
448 if (r < 0)
449 goto fail;
450
451 *ret_filename = TAKE_PTR(fn);
452 *ret_data_fd = TAKE_FD(fd);
453 *ret_node_fd = -1;
454 *ret_size = (uint64_t) st.st_size;
455
456 return 0;
457
458 fail:
459 if (tmp)
460 (void) unlink(tmp);
461 return r;
462 }
463
464 static int allocate_journal_field(int fd, size_t size, char **ret, size_t *ret_size) {
465 _cleanup_free_ char *field = NULL;
466 ssize_t n;
467
468 assert(fd >= 0);
469 assert(ret);
470 assert(ret_size);
471
472 if (lseek(fd, 0, SEEK_SET) == (off_t) -1)
473 return log_warning_errno(errno, "Failed to seek: %m");
474
475 field = malloc(9 + size);
476 if (!field) {
477 log_warning("Failed to allocate memory for coredump, coredump will not be stored.");
478 return -ENOMEM;
479 }
480
481 memcpy(field, "COREDUMP=", 9);
482
483 n = read(fd, field + 9, size);
484 if (n < 0)
485 return log_error_errno((int) n, "Failed to read core data: %m");
486 if ((size_t) n < size) {
487 log_error("Core data too short.");
488 return -EIO;
489 }
490
491 *ret = TAKE_PTR(field);
492 *ret_size = size + 9;
493
494 return 0;
495 }
496
497 /* Joins /proc/[pid]/fd/ and /proc/[pid]/fdinfo/ into the following lines:
498 * 0:/dev/pts/23
499 * pos: 0
500 * flags: 0100002
501 *
502 * 1:/dev/pts/23
503 * pos: 0
504 * flags: 0100002
505 *
506 * 2:/dev/pts/23
507 * pos: 0
508 * flags: 0100002
509 * EOF
510 */
511 static int compose_open_fds(pid_t pid, char **open_fds) {
512 _cleanup_closedir_ DIR *proc_fd_dir = NULL;
513 _cleanup_close_ int proc_fdinfo_fd = -1;
514 _cleanup_free_ char *buffer = NULL;
515 _cleanup_fclose_ FILE *stream = NULL;
516 const char *fddelim = "", *path;
517 struct dirent *dent = NULL;
518 size_t size = 0;
519 int r;
520
521 assert(pid >= 0);
522 assert(open_fds != NULL);
523
524 path = procfs_file_alloca(pid, "fd");
525 proc_fd_dir = opendir(path);
526 if (!proc_fd_dir)
527 return -errno;
528
529 proc_fdinfo_fd = openat(dirfd(proc_fd_dir), "../fdinfo", O_DIRECTORY|O_NOFOLLOW|O_CLOEXEC|O_PATH);
530 if (proc_fdinfo_fd < 0)
531 return -errno;
532
533 stream = open_memstream(&buffer, &size);
534 if (!stream)
535 return -ENOMEM;
536
537 (void) __fsetlocking(stream, FSETLOCKING_BYCALLER);
538
539 FOREACH_DIRENT(dent, proc_fd_dir, return -errno) {
540 _cleanup_fclose_ FILE *fdinfo = NULL;
541 _cleanup_free_ char *fdname = NULL;
542 int fd;
543
544 r = readlinkat_malloc(dirfd(proc_fd_dir), dent->d_name, &fdname);
545 if (r < 0)
546 return r;
547
548 fprintf(stream, "%s%s:%s\n", fddelim, dent->d_name, fdname);
549 fddelim = "\n";
550
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);
553 if (fd < 0)
554 continue;
555
556 fdinfo = fdopen(fd, "re");
557 if (!fdinfo) {
558 safe_close(fd);
559 continue;
560 }
561
562 for (;;) {
563 _cleanup_free_ char *line = NULL;
564
565 r = read_line(fdinfo, LONG_LINE_MAX, &line);
566 if (r < 0)
567 return r;
568 if (r == 0)
569 break;
570
571 fputs(line, stream);
572 fputc('\n', stream);
573 }
574 }
575
576 errno = 0;
577 stream = safe_fclose(stream);
578
579 if (errno > 0)
580 return -errno;
581
582 *open_fds = TAKE_PTR(buffer);
583
584 return 0;
585 }
586
587 static int get_process_ns(pid_t pid, const char *namespace, ino_t *ns) {
588 const char *p;
589 struct stat stbuf;
590 _cleanup_close_ int proc_ns_dir_fd;
591
592 p = procfs_file_alloca(pid, "ns");
593
594 proc_ns_dir_fd = open(p, O_DIRECTORY | O_CLOEXEC | O_RDONLY);
595 if (proc_ns_dir_fd < 0)
596 return -errno;
597
598 if (fstatat(proc_ns_dir_fd, namespace, &stbuf, /* flags */0) < 0)
599 return -errno;
600
601 *ns = stbuf.st_ino;
602 return 0;
603 }
604
605 static int get_mount_namespace_leader(pid_t pid, pid_t *container_pid) {
606 pid_t cpid = pid, ppid = 0;
607 ino_t proc_mntns;
608 int r = 0;
609
610 r = get_process_ns(pid, "mnt", &proc_mntns);
611 if (r < 0)
612 return r;
613
614 for (;;) {
615 ino_t parent_mntns;
616
617 r = get_process_ppid(cpid, &ppid);
618 if (r < 0)
619 return r;
620
621 r = get_process_ns(ppid, "mnt", &parent_mntns);
622 if (r < 0)
623 return r;
624
625 if (proc_mntns != parent_mntns)
626 break;
627
628 if (ppid == 1)
629 return -ENOENT;
630
631 cpid = ppid;
632 }
633
634 *container_pid = ppid;
635 return 0;
636 }
637
638 /* Returns 1 if the parent was found.
639 * Returns 0 if there is not a process we can call the pid's
640 * container parent (the pid's process isn't 'containerized').
641 * Returns a negative number on errors.
642 */
643 static int get_process_container_parent_cmdline(pid_t pid, char** cmdline) {
644 int r = 0;
645 pid_t container_pid;
646 const char *proc_root_path;
647 struct stat root_stat, proc_root_stat;
648
649 /* To compare inodes of / and /proc/[pid]/root */
650 if (stat("/", &root_stat) < 0)
651 return -errno;
652
653 proc_root_path = procfs_file_alloca(pid, "root");
654 if (stat(proc_root_path, &proc_root_stat) < 0)
655 return -errno;
656
657 /* The process uses system root. */
658 if (proc_root_stat.st_ino == root_stat.st_ino) {
659 *cmdline = NULL;
660 return 0;
661 }
662
663 r = get_mount_namespace_leader(pid, &container_pid);
664 if (r < 0)
665 return r;
666
667 r = get_process_cmdline(container_pid, 0, false, cmdline);
668 if (r < 0)
669 return r;
670
671 return 1;
672 }
673
674 static int change_uid_gid(const char *context[]) {
675 uid_t uid;
676 gid_t gid;
677 int r;
678
679 r = parse_uid(context[CONTEXT_UID], &uid);
680 if (r < 0)
681 return r;
682
683 if (uid <= SYSTEM_UID_MAX) {
684 const char *user = "systemd-coredump";
685
686 r = get_user_creds(&user, &uid, &gid, NULL, NULL, 0);
687 if (r < 0) {
688 log_warning_errno(r, "Cannot resolve %s user. Proceeding to dump core as root: %m", user);
689 uid = gid = 0;
690 }
691 } else {
692 r = parse_gid(context[CONTEXT_GID], &gid);
693 if (r < 0)
694 return r;
695 }
696
697 return drop_privileges(uid, gid, 0);
698 }
699
700 static bool is_journald_crash(const char *context[_CONTEXT_MAX]) {
701 assert(context);
702
703 return streq_ptr(context[CONTEXT_UNIT], SPECIAL_JOURNALD_SERVICE);
704 }
705
706 static bool is_pid1_crash(const char *context[_CONTEXT_MAX]) {
707 assert(context);
708
709 return streq_ptr(context[CONTEXT_UNIT], SPECIAL_INIT_SCOPE) ||
710 streq_ptr(context[CONTEXT_PID], "1");
711 }
712
713 #define SUBMIT_COREDUMP_FIELDS 4
714
715 static int submit_coredump(
716 const char *context[_CONTEXT_MAX],
717 struct iovec *iovec,
718 size_t n_iovec_allocated,
719 size_t n_iovec,
720 int input_fd) {
721
722 _cleanup_close_ int coredump_fd = -1, coredump_node_fd = -1;
723 _cleanup_free_ char *core_message = NULL, *filename = NULL, *coredump_data = NULL;
724 uint64_t coredump_size = UINT64_MAX;
725 bool truncated = false, journald_crash;
726 int r;
727
728 assert(context);
729 assert(iovec);
730 assert(n_iovec_allocated >= n_iovec + SUBMIT_COREDUMP_FIELDS);
731 assert(input_fd >= 0);
732
733 journald_crash = is_journald_crash(context);
734
735 /* Vacuum before we write anything again */
736 (void) coredump_vacuum(-1, arg_keep_free, arg_max_use);
737
738 /* Always stream the coredump to disk, if that's possible */
739 r = save_external_coredump(context, input_fd,
740 &filename, &coredump_node_fd, &coredump_fd, &coredump_size, &truncated);
741 if (r < 0)
742 /* Skip whole core dumping part */
743 goto log;
744
745 /* If we don't want to keep the coredump on disk, remove it now, as later on we will lack the privileges for
746 * it. However, we keep the fd to it, so that we can still process it and log it. */
747 r = maybe_remove_external_coredump(filename, coredump_size);
748 if (r < 0)
749 return r;
750 if (r == 0) {
751 const char *coredump_filename;
752
753 coredump_filename = strjoina("COREDUMP_FILENAME=", filename);
754 iovec[n_iovec++] = IOVEC_MAKE_STRING(coredump_filename);
755 } else if (arg_storage == COREDUMP_STORAGE_EXTERNAL)
756 log_info("The core will not be stored: size %"PRIu64" is greater than %"PRIu64" (the configured maximum)",
757 coredump_size, arg_external_size_max);
758
759 /* Vacuum again, but exclude the coredump we just created */
760 (void) coredump_vacuum(coredump_node_fd >= 0 ? coredump_node_fd : coredump_fd, arg_keep_free, arg_max_use);
761
762 /* Now, let's drop privileges to become the user who owns the segfaulted process and allocate the coredump
763 * memory under the user's uid. This also ensures that the credentials journald will see are the ones of the
764 * coredumping user, thus making sure the user gets access to the core dump. Let's also get rid of all
765 * capabilities, if we run as root, we won't need them anymore. */
766 r = change_uid_gid(context);
767 if (r < 0)
768 return log_error_errno(r, "Failed to drop privileges: %m");
769
770 #if HAVE_ELFUTILS
771 /* Try to get a strack trace if we can */
772 if (coredump_size <= arg_process_size_max) {
773 _cleanup_free_ char *stacktrace = NULL;
774
775 r = coredump_make_stack_trace(coredump_fd, context[CONTEXT_EXE], &stacktrace);
776 if (r >= 0)
777 core_message = strjoin("MESSAGE=Process ", context[CONTEXT_PID],
778 " (", context[CONTEXT_COMM], ") of user ",
779 context[CONTEXT_UID], " dumped core.",
780 journald_crash ? "\nCoredump diverted to " : "",
781 journald_crash ? filename : "",
782 "\n\n", stacktrace);
783 else if (r == -EINVAL)
784 log_warning("Failed to generate stack trace: %s", dwfl_errmsg(dwfl_errno()));
785 else
786 log_warning_errno(r, "Failed to generate stack trace: %m");
787 } else
788 log_debug("Not generating stack trace: core size %"PRIu64" is greater than %"PRIu64" (the configured maximum)",
789 coredump_size, arg_process_size_max);
790
791 if (!core_message)
792 #endif
793 log:
794 core_message = strjoin("MESSAGE=Process ", context[CONTEXT_PID],
795 " (", context[CONTEXT_COMM], ") of user ",
796 context[CONTEXT_UID], " dumped core.",
797 journald_crash ? "\nCoredump diverted to " : NULL,
798 journald_crash ? filename : NULL);
799 if (!core_message)
800 return log_oom();
801
802 if (journald_crash) {
803 /* We cannot log to the journal, so just print the MESSAGE.
804 * The target was set previously to something safe. */
805 log_dispatch(LOG_ERR, 0, core_message);
806 return 0;
807 }
808
809 iovec[n_iovec++] = IOVEC_MAKE_STRING(core_message);
810
811 if (truncated)
812 iovec[n_iovec++] = IOVEC_MAKE_STRING("COREDUMP_TRUNCATED=1");
813
814 /* Optionally store the entire coredump in the journal */
815 if (arg_storage == COREDUMP_STORAGE_JOURNAL) {
816 if (coredump_size <= arg_journal_size_max) {
817 size_t sz = 0;
818
819 /* Store the coredump itself in the journal */
820
821 r = allocate_journal_field(coredump_fd, (size_t) coredump_size, &coredump_data, &sz);
822 if (r >= 0)
823 iovec[n_iovec++] = IOVEC_MAKE(coredump_data, sz);
824 else
825 log_warning_errno(r, "Failed to attach the core to the journal entry: %m");
826 } else
827 log_info("The core will not be stored: size %"PRIu64" is greater than %"PRIu64" (the configured maximum)",
828 coredump_size, arg_journal_size_max);
829 }
830
831 assert(n_iovec <= n_iovec_allocated);
832
833 r = sd_journal_sendv(iovec, n_iovec);
834 if (r < 0)
835 return log_error_errno(r, "Failed to log coredump: %m");
836
837 return 0;
838 }
839
840 static void map_context_fields(const struct iovec *iovec, const char* context[]) {
841
842 static const char * const context_field_names[] = {
843 [CONTEXT_PID] = "COREDUMP_PID=",
844 [CONTEXT_UID] = "COREDUMP_UID=",
845 [CONTEXT_GID] = "COREDUMP_GID=",
846 [CONTEXT_SIGNAL] = "COREDUMP_SIGNAL=",
847 [CONTEXT_TIMESTAMP] = "COREDUMP_TIMESTAMP=",
848 [CONTEXT_RLIMIT] = "COREDUMP_RLIMIT=",
849 [CONTEXT_HOSTNAME] = "COREDUMP_HOSTNAME=",
850 [CONTEXT_COMM] = "COREDUMP_COMM=",
851 [CONTEXT_EXE] = "COREDUMP_EXE=",
852 };
853
854 unsigned i;
855
856 assert(iovec);
857 assert(context);
858
859 for (i = 0; i < ELEMENTSOF(context_field_names); i++) {
860 char *p;
861
862 if (!context_field_names[i])
863 continue;
864
865 p = memory_startswith(iovec->iov_base, iovec->iov_len, context_field_names[i]);
866 if (!p)
867 continue;
868
869 /* Note that these strings are NUL terminated, because we made sure that a trailing NUL byte is in the
870 * buffer, though not included in the iov_len count. (see below) */
871 context[i] = p;
872 break;
873 }
874 }
875
876 static int process_socket(int fd) {
877 _cleanup_close_ int coredump_fd = -1;
878 struct iovec *iovec = NULL;
879 size_t n_iovec = 0, n_allocated = 0, i, k;
880 const char *context[_CONTEXT_MAX] = {};
881 int r;
882
883 assert(fd >= 0);
884
885 log_set_target(LOG_TARGET_AUTO);
886 log_parse_environment();
887 log_open();
888
889 log_debug("Processing coredump received on stdin...");
890
891 for (;;) {
892 union {
893 struct cmsghdr cmsghdr;
894 uint8_t buf[CMSG_SPACE(sizeof(int))];
895 } control = {};
896 struct msghdr mh = {
897 .msg_control = &control,
898 .msg_controllen = sizeof(control),
899 .msg_iovlen = 1,
900 };
901 ssize_t n;
902 ssize_t l;
903
904 if (!GREEDY_REALLOC(iovec, n_allocated, n_iovec + SUBMIT_COREDUMP_FIELDS)) {
905 r = log_oom();
906 goto finish;
907 }
908
909 l = next_datagram_size_fd(fd);
910 if (l < 0) {
911 r = log_error_errno(l, "Failed to determine datagram size to read: %m");
912 goto finish;
913 }
914
915 assert(l >= 0);
916
917 iovec[n_iovec].iov_len = l;
918 iovec[n_iovec].iov_base = malloc(l + 1);
919 if (!iovec[n_iovec].iov_base) {
920 r = log_oom();
921 goto finish;
922 }
923
924 mh.msg_iov = iovec + n_iovec;
925
926 n = recvmsg(fd, &mh, MSG_CMSG_CLOEXEC);
927 if (n < 0) {
928 free(iovec[n_iovec].iov_base);
929 r = log_error_errno(errno, "Failed to receive datagram: %m");
930 goto finish;
931 }
932
933 if (n == 0) {
934 struct cmsghdr *cmsg, *found = NULL;
935 /* The final zero-length datagram carries the file descriptor and tells us that we're done. */
936
937 free(iovec[n_iovec].iov_base);
938
939 CMSG_FOREACH(cmsg, &mh) {
940 if (cmsg->cmsg_level == SOL_SOCKET &&
941 cmsg->cmsg_type == SCM_RIGHTS &&
942 cmsg->cmsg_len == CMSG_LEN(sizeof(int))) {
943 assert(!found);
944 found = cmsg;
945 }
946 }
947
948 if (!found) {
949 log_error("Coredump file descriptor missing.");
950 r = -EBADMSG;
951 goto finish;
952 }
953
954 assert(coredump_fd < 0);
955 coredump_fd = *(int*) CMSG_DATA(found);
956 break;
957 }
958
959 /* Add trailing NUL byte, in case these are strings */
960 ((char*) iovec[n_iovec].iov_base)[n] = 0;
961 iovec[n_iovec].iov_len = (size_t) n;
962
963 cmsg_close_all(&mh);
964 map_context_fields(iovec + n_iovec, context);
965 n_iovec++;
966 }
967
968 if (!GREEDY_REALLOC(iovec, n_allocated, n_iovec + SUBMIT_COREDUMP_FIELDS)) {
969 r = log_oom();
970 goto finish;
971 }
972
973 /* Make sure we got all data we really need */
974 assert(context[CONTEXT_PID]);
975 assert(context[CONTEXT_UID]);
976 assert(context[CONTEXT_GID]);
977 assert(context[CONTEXT_SIGNAL]);
978 assert(context[CONTEXT_TIMESTAMP]);
979 assert(context[CONTEXT_RLIMIT]);
980 assert(context[CONTEXT_HOSTNAME]);
981 assert(context[CONTEXT_COMM]);
982 assert(coredump_fd >= 0);
983
984 /* Small quirk: the journal fields contain the timestamp padded with six zeroes, so that the kernel-supplied 1s
985 * granularity timestamps becomes 1µs granularity, i.e. the granularity systemd usually operates in. Since we
986 * are reconstructing the original kernel context, we chop this off again, here. */
987 k = strlen(context[CONTEXT_TIMESTAMP]);
988 if (k > 6)
989 context[CONTEXT_TIMESTAMP] = strndupa(context[CONTEXT_TIMESTAMP], k - 6);
990
991 r = submit_coredump(context, iovec, n_allocated, n_iovec, coredump_fd);
992
993 finish:
994 for (i = 0; i < n_iovec; i++)
995 free(iovec[i].iov_base);
996 free(iovec);
997
998 return r;
999 }
1000
1001 static int send_iovec(const struct iovec iovec[], size_t n_iovec, int input_fd) {
1002
1003 static const union sockaddr_union sa = {
1004 .un.sun_family = AF_UNIX,
1005 .un.sun_path = "/run/systemd/coredump",
1006 };
1007 _cleanup_close_ int fd = -1;
1008 size_t i;
1009 int r;
1010
1011 assert(iovec || n_iovec <= 0);
1012 assert(input_fd >= 0);
1013
1014 fd = socket(AF_UNIX, SOCK_SEQPACKET|SOCK_CLOEXEC, 0);
1015 if (fd < 0)
1016 return log_error_errno(errno, "Failed to create coredump socket: %m");
1017
1018 if (connect(fd, &sa.sa, SOCKADDR_UN_LEN(sa.un)) < 0)
1019 return log_error_errno(errno, "Failed to connect to coredump service: %m");
1020
1021 for (i = 0; i < n_iovec; i++) {
1022 struct msghdr mh = {
1023 .msg_iov = (struct iovec*) iovec + i,
1024 .msg_iovlen = 1,
1025 };
1026 struct iovec copy[2];
1027
1028 for (;;) {
1029 if (sendmsg(fd, &mh, MSG_NOSIGNAL) >= 0)
1030 break;
1031
1032 if (errno == EMSGSIZE && mh.msg_iov[0].iov_len > 0) {
1033 /* This field didn't fit? That's a pity. Given that this is just metadata,
1034 * let's truncate the field at half, and try again. We append three dots, in
1035 * order to show that this is truncated. */
1036
1037 if (mh.msg_iov != copy) {
1038 /* We don't want to modify the caller's iovec, hence let's create our
1039 * own array, consisting of two new iovecs, where the first is a
1040 * (truncated) copy of what we want to send, and the second one
1041 * contains the trailing dots. */
1042 copy[0] = iovec[i];
1043 copy[1] = (struct iovec) {
1044 .iov_base = (char[]) { '.', '.', '.' },
1045 .iov_len = 3,
1046 };
1047
1048 mh.msg_iov = copy;
1049 mh.msg_iovlen = 2;
1050 }
1051
1052 copy[0].iov_len /= 2; /* halve it, and try again */
1053 continue;
1054 }
1055
1056 return log_error_errno(errno, "Failed to send coredump datagram: %m");
1057 }
1058 }
1059
1060 r = send_one_fd(fd, input_fd, 0);
1061 if (r < 0)
1062 return log_error_errno(r, "Failed to send coredump fd: %m");
1063
1064 return 0;
1065 }
1066
1067 static char* set_iovec_field(struct iovec *iovec, size_t *n_iovec, const char *field, const char *value) {
1068 char *x;
1069
1070 x = strappend(field, value);
1071 if (x)
1072 iovec[(*n_iovec)++] = IOVEC_MAKE_STRING(x);
1073 return x;
1074 }
1075
1076 static char* set_iovec_field_free(struct iovec *iovec, size_t *n_iovec, const char *field, char *value) {
1077 char *x;
1078
1079 x = set_iovec_field(iovec, n_iovec, field, value);
1080 free(value);
1081 return x;
1082 }
1083
1084 static int gather_pid_metadata(
1085 char* context[_CONTEXT_MAX],
1086 char **comm_fallback,
1087 struct iovec *iovec, size_t *n_iovec) {
1088
1089 /* We need 27 empty slots in iovec!
1090 *
1091 * Note that if we fail on oom later on, we do not roll-back changes to the iovec structure. (It remains valid,
1092 * with the first n_iovec fields initialized.) */
1093
1094 uid_t owner_uid;
1095 pid_t pid;
1096 char *t;
1097 const char *p;
1098 int r, signo;
1099
1100 r = parse_pid(context[CONTEXT_PID], &pid);
1101 if (r < 0)
1102 return log_error_errno(r, "Failed to parse PID \"%s\": %m", context[CONTEXT_PID]);
1103
1104 r = get_process_comm(pid, &context[CONTEXT_COMM]);
1105 if (r < 0) {
1106 log_warning_errno(r, "Failed to get COMM, falling back to the command line: %m");
1107 context[CONTEXT_COMM] = strv_join(comm_fallback, " ");
1108 if (!context[CONTEXT_COMM])
1109 return log_oom();
1110 }
1111
1112 r = get_process_exe(pid, &context[CONTEXT_EXE]);
1113 if (r < 0)
1114 log_warning_errno(r, "Failed to get EXE, ignoring: %m");
1115
1116 if (cg_pid_get_unit(pid, &context[CONTEXT_UNIT]) >= 0) {
1117 if (!is_journald_crash((const char**) context)) {
1118 /* OK, now we know it's not the journal, hence we can make use of it now. */
1119 log_set_target(LOG_TARGET_JOURNAL_OR_KMSG);
1120 log_open();
1121 }
1122
1123 /* If this is PID 1 disable coredump collection, we'll unlikely be able to process it later on. */
1124 if (is_pid1_crash((const char**) context)) {
1125 log_notice("Due to PID 1 having crashed coredump collection will now be turned off.");
1126 disable_coredumps();
1127 }
1128
1129 set_iovec_field(iovec, n_iovec, "COREDUMP_UNIT=", context[CONTEXT_UNIT]);
1130 }
1131
1132 if (cg_pid_get_user_unit(pid, &t) >= 0)
1133 set_iovec_field_free(iovec, n_iovec, "COREDUMP_USER_UNIT=", t);
1134
1135 /* The next few are mandatory */
1136 if (!set_iovec_field(iovec, n_iovec, "COREDUMP_PID=", context[CONTEXT_PID]))
1137 return log_oom();
1138
1139 if (!set_iovec_field(iovec, n_iovec, "COREDUMP_UID=", context[CONTEXT_UID]))
1140 return log_oom();
1141
1142 if (!set_iovec_field(iovec, n_iovec, "COREDUMP_GID=", context[CONTEXT_GID]))
1143 return log_oom();
1144
1145 if (!set_iovec_field(iovec, n_iovec, "COREDUMP_SIGNAL=", context[CONTEXT_SIGNAL]))
1146 return log_oom();
1147
1148 if (!set_iovec_field(iovec, n_iovec, "COREDUMP_RLIMIT=", context[CONTEXT_RLIMIT]))
1149 return log_oom();
1150
1151 if (!set_iovec_field(iovec, n_iovec, "COREDUMP_HOSTNAME=", context[CONTEXT_HOSTNAME]))
1152 return log_oom();
1153
1154 if (!set_iovec_field(iovec, n_iovec, "COREDUMP_COMM=", context[CONTEXT_COMM]))
1155 return log_oom();
1156
1157 if (context[CONTEXT_EXE] &&
1158 !set_iovec_field(iovec, n_iovec, "COREDUMP_EXE=", context[CONTEXT_EXE]))
1159 return log_oom();
1160
1161 if (sd_pid_get_session(pid, &t) >= 0)
1162 set_iovec_field_free(iovec, n_iovec, "COREDUMP_SESSION=", t);
1163
1164 if (sd_pid_get_owner_uid(pid, &owner_uid) >= 0) {
1165 r = asprintf(&t, "COREDUMP_OWNER_UID=" UID_FMT, owner_uid);
1166 if (r > 0)
1167 iovec[(*n_iovec)++] = IOVEC_MAKE_STRING(t);
1168 }
1169
1170 if (sd_pid_get_slice(pid, &t) >= 0)
1171 set_iovec_field_free(iovec, n_iovec, "COREDUMP_SLICE=", t);
1172
1173 if (get_process_cmdline(pid, 0, false, &t) >= 0)
1174 set_iovec_field_free(iovec, n_iovec, "COREDUMP_CMDLINE=", t);
1175
1176 if (cg_pid_get_path_shifted(pid, NULL, &t) >= 0)
1177 set_iovec_field_free(iovec, n_iovec, "COREDUMP_CGROUP=", t);
1178
1179 if (compose_open_fds(pid, &t) >= 0)
1180 set_iovec_field_free(iovec, n_iovec, "COREDUMP_OPEN_FDS=", t);
1181
1182 p = procfs_file_alloca(pid, "status");
1183 if (read_full_file(p, &t, NULL) >= 0)
1184 set_iovec_field_free(iovec, n_iovec, "COREDUMP_PROC_STATUS=", t);
1185
1186 p = procfs_file_alloca(pid, "maps");
1187 if (read_full_file(p, &t, NULL) >= 0)
1188 set_iovec_field_free(iovec, n_iovec, "COREDUMP_PROC_MAPS=", t);
1189
1190 p = procfs_file_alloca(pid, "limits");
1191 if (read_full_file(p, &t, NULL) >= 0)
1192 set_iovec_field_free(iovec, n_iovec, "COREDUMP_PROC_LIMITS=", t);
1193
1194 p = procfs_file_alloca(pid, "cgroup");
1195 if (read_full_file(p, &t, NULL) >=0)
1196 set_iovec_field_free(iovec, n_iovec, "COREDUMP_PROC_CGROUP=", t);
1197
1198 p = procfs_file_alloca(pid, "mountinfo");
1199 if (read_full_file(p, &t, NULL) >=0)
1200 set_iovec_field_free(iovec, n_iovec, "COREDUMP_PROC_MOUNTINFO=", t);
1201
1202 if (get_process_cwd(pid, &t) >= 0)
1203 set_iovec_field_free(iovec, n_iovec, "COREDUMP_CWD=", t);
1204
1205 if (get_process_root(pid, &t) >= 0) {
1206 bool proc_self_root_is_slash;
1207
1208 proc_self_root_is_slash = strcmp(t, "/") == 0;
1209
1210 set_iovec_field_free(iovec, n_iovec, "COREDUMP_ROOT=", t);
1211
1212 /* If the process' root is "/", then there is a chance it has
1213 * mounted own root and hence being containerized. */
1214 if (proc_self_root_is_slash && get_process_container_parent_cmdline(pid, &t) > 0)
1215 set_iovec_field_free(iovec, n_iovec, "COREDUMP_CONTAINER_CMDLINE=", t);
1216 }
1217
1218 if (get_process_environ(pid, &t) >= 0)
1219 set_iovec_field_free(iovec, n_iovec, "COREDUMP_ENVIRON=", t);
1220
1221 t = strjoin("COREDUMP_TIMESTAMP=", context[CONTEXT_TIMESTAMP], "000000");
1222 if (t)
1223 iovec[(*n_iovec)++] = IOVEC_MAKE_STRING(t);
1224
1225 if (safe_atoi(context[CONTEXT_SIGNAL], &signo) >= 0 && SIGNAL_VALID(signo))
1226 set_iovec_field(iovec, n_iovec, "COREDUMP_SIGNAL_NAME=SIG", signal_to_string(signo));
1227
1228 return 0; /* we successfully acquired all metadata */
1229 }
1230
1231 static int process_kernel(int argc, char* argv[]) {
1232
1233 char* context[_CONTEXT_MAX] = {};
1234 struct iovec iovec[29 + SUBMIT_COREDUMP_FIELDS];
1235 size_t i, n_iovec, n_to_free = 0;
1236 int r;
1237
1238 log_debug("Processing coredump received from the kernel...");
1239
1240 if (argc < CONTEXT_COMM + 1) {
1241 log_error("Not enough arguments passed by the kernel (%i, expected %i).", argc - 1, CONTEXT_COMM + 1 - 1);
1242 return -EINVAL;
1243 }
1244
1245 context[CONTEXT_PID] = argv[1 + CONTEXT_PID];
1246 context[CONTEXT_UID] = argv[1 + CONTEXT_UID];
1247 context[CONTEXT_GID] = argv[1 + CONTEXT_GID];
1248 context[CONTEXT_SIGNAL] = argv[1 + CONTEXT_SIGNAL];
1249 context[CONTEXT_TIMESTAMP] = argv[1 + CONTEXT_TIMESTAMP];
1250 context[CONTEXT_RLIMIT] = argv[1 + CONTEXT_RLIMIT];
1251 context[CONTEXT_HOSTNAME] = argv[1 + CONTEXT_HOSTNAME];
1252
1253 r = gather_pid_metadata(context, argv + 1 + CONTEXT_COMM, iovec, &n_to_free);
1254 if (r < 0)
1255 goto finish;
1256
1257 n_iovec = n_to_free;
1258
1259 iovec[n_iovec++] = IOVEC_MAKE_STRING("MESSAGE_ID=" SD_MESSAGE_COREDUMP_STR);
1260
1261 assert_cc(2 == LOG_CRIT);
1262 iovec[n_iovec++] = IOVEC_MAKE_STRING("PRIORITY=2");
1263
1264 assert(n_iovec <= ELEMENTSOF(iovec));
1265
1266 if (is_journald_crash((const char**) context) || is_pid1_crash((const char**) context))
1267 r = submit_coredump((const char**) context,
1268 iovec, ELEMENTSOF(iovec), n_iovec,
1269 STDIN_FILENO);
1270 else
1271 r = send_iovec(iovec, n_iovec, STDIN_FILENO);
1272
1273 finish:
1274 for (i = 0; i < n_to_free; i++)
1275 free(iovec[i].iov_base);
1276
1277 /* Those fields are allocated by gather_pid_metadata */
1278 free(context[CONTEXT_COMM]);
1279 free(context[CONTEXT_EXE]);
1280 free(context[CONTEXT_UNIT]);
1281
1282 return r;
1283 }
1284
1285 static int process_backtrace(int argc, char *argv[]) {
1286 char *context[_CONTEXT_MAX] = {};
1287 _cleanup_free_ char *message = NULL;
1288 _cleanup_free_ struct iovec *iovec = NULL;
1289 size_t n_iovec, n_allocated, n_to_free = 0, i;
1290 int r;
1291 JournalImporter importer = {
1292 .fd = STDIN_FILENO,
1293 };
1294
1295 log_debug("Processing backtrace on stdin...");
1296
1297 if (argc < CONTEXT_COMM + 1) {
1298 log_error("Not enough arguments passed (%i, expected %i).", argc - 1, CONTEXT_COMM + 1 - 1);
1299 return -EINVAL;
1300 }
1301
1302 context[CONTEXT_PID] = argv[2 + CONTEXT_PID];
1303 context[CONTEXT_UID] = argv[2 + CONTEXT_UID];
1304 context[CONTEXT_GID] = argv[2 + CONTEXT_GID];
1305 context[CONTEXT_SIGNAL] = argv[2 + CONTEXT_SIGNAL];
1306 context[CONTEXT_TIMESTAMP] = argv[2 + CONTEXT_TIMESTAMP];
1307 context[CONTEXT_RLIMIT] = argv[2 + CONTEXT_RLIMIT];
1308 context[CONTEXT_HOSTNAME] = argv[2 + CONTEXT_HOSTNAME];
1309
1310 n_allocated = 34 + COREDUMP_STORAGE_EXTERNAL;
1311 /* 26 metadata, 2 static, +unknown input, 4 storage, rounded up */
1312 iovec = new(struct iovec, n_allocated);
1313 if (!iovec)
1314 return log_oom();
1315
1316 r = gather_pid_metadata(context, argv + 2 + CONTEXT_COMM, iovec, &n_to_free);
1317 if (r < 0)
1318 goto finish;
1319 if (r > 0) {
1320 /* This was a special crash, and has already been processed. */
1321 r = 0;
1322 goto finish;
1323 }
1324 n_iovec = n_to_free;
1325
1326 for (;;) {
1327 r = journal_importer_process_data(&importer);
1328 if (r < 0) {
1329 log_error_errno(r, "Failed to parse journal entry on stdin: %m");
1330 goto finish;
1331 }
1332 if (r == 1 || /* complete entry */
1333 journal_importer_eof(&importer)) /* end of data */
1334 break;
1335 }
1336
1337 if (!GREEDY_REALLOC(iovec, n_allocated, n_iovec + importer.iovw.count + 2))
1338 return log_oom();
1339
1340 if (journal_importer_eof(&importer)) {
1341 log_warning("Did not receive a full journal entry on stdin, ignoring message sent by reporter");
1342
1343 message = strjoin("MESSAGE=Process ", context[CONTEXT_PID],
1344 " (", context[CONTEXT_COMM], ")"
1345 " of user ", context[CONTEXT_UID],
1346 " failed with ", context[CONTEXT_SIGNAL]);
1347 if (!message) {
1348 r = log_oom();
1349 goto finish;
1350 }
1351 iovec[n_iovec++] = IOVEC_MAKE_STRING(message);
1352 } else {
1353 for (i = 0; i < importer.iovw.count; i++)
1354 iovec[n_iovec++] = importer.iovw.iovec[i];
1355 }
1356
1357 iovec[n_iovec++] = IOVEC_MAKE_STRING("MESSAGE_ID=" SD_MESSAGE_BACKTRACE_STR);
1358 assert_cc(2 == LOG_CRIT);
1359 iovec[n_iovec++] = IOVEC_MAKE_STRING("PRIORITY=2");
1360
1361 assert(n_iovec <= n_allocated);
1362
1363 r = sd_journal_sendv(iovec, n_iovec);
1364 if (r < 0)
1365 log_error_errno(r, "Failed to log backtrace: %m");
1366
1367 finish:
1368 for (i = 0; i < n_to_free; i++)
1369 free(iovec[i].iov_base);
1370
1371 /* Those fields are allocated by gather_pid_metadata */
1372 free(context[CONTEXT_COMM]);
1373 free(context[CONTEXT_EXE]);
1374 free(context[CONTEXT_UNIT]);
1375
1376 return r;
1377 }
1378
1379 int main(int argc, char *argv[]) {
1380 int r;
1381
1382 /* First, log to a safe place, since we don't know what crashed and it might
1383 * be journald which we'd rather not log to then. */
1384
1385 log_set_target(LOG_TARGET_KMSG);
1386 log_open();
1387
1388 /* Make sure we never enter a loop */
1389 (void) prctl(PR_SET_DUMPABLE, 0);
1390
1391 /* Ignore all parse errors */
1392 (void) parse_config();
1393
1394 log_debug("Selected storage '%s'.", coredump_storage_to_string(arg_storage));
1395 log_debug("Selected compression %s.", yes_no(arg_compress));
1396
1397 r = sd_listen_fds(false);
1398 if (r < 0) {
1399 log_error_errno(r, "Failed to determine number of file descriptor: %m");
1400 goto finish;
1401 }
1402
1403 /* If we got an fd passed, we are running in coredumpd mode. Otherwise we
1404 * are invoked from the kernel as coredump handler. */
1405 if (r == 0) {
1406 if (streq_ptr(argv[1], "--backtrace"))
1407 r = process_backtrace(argc, argv);
1408 else
1409 r = process_kernel(argc, argv);
1410 } else if (r == 1)
1411 r = process_socket(SD_LISTEN_FDS_START);
1412 else {
1413 log_error("Received unexpected number of file descriptors.");
1414 r = -EINVAL;
1415 }
1416
1417 finish:
1418 return r < 0 ? EXIT_FAILURE : EXIT_SUCCESS;
1419 }