]> git.ipfire.org Git - thirdparty/systemd.git/blob - src/coredump/coredump.c
1c470e60da86ed87e90f05b641629801700bab66
[thirdparty/systemd.git] / src / coredump / coredump.c
1 /* SPDX-License-Identifier: LGPL-2.1+ */
2 /***
3 This file is part of systemd.
4
5 Copyright 2012 Lennart Poettering
6 ***/
7
8 #include <errno.h>
9 #include <stdio.h>
10 #include <stdio_ext.h>
11 #include <sys/prctl.h>
12 #include <sys/xattr.h>
13 #include <unistd.h>
14
15 #if HAVE_ELFUTILS
16 #include <dwarf.h>
17 #include <elfutils/libdwfl.h>
18 #endif
19
20 #include "sd-daemon.h"
21 #include "sd-journal.h"
22 #include "sd-login.h"
23 #include "sd-messages.h"
24
25 #include "acl-util.h"
26 #include "alloc-util.h"
27 #include "capability-util.h"
28 #include "cgroup-util.h"
29 #include "compress.h"
30 #include "conf-parser.h"
31 #include "copy.h"
32 #include "coredump-vacuum.h"
33 #include "dirent-util.h"
34 #include "escape.h"
35 #include "fd-util.h"
36 #include "fileio.h"
37 #include "fs-util.h"
38 #include "io-util.h"
39 #include "journal-importer.h"
40 #include "log.h"
41 #include "macro.h"
42 #include "missing.h"
43 #include "mkdir.h"
44 #include "parse-util.h"
45 #include "process-util.h"
46 #include "signal-util.h"
47 #include "socket-util.h"
48 #include "special.h"
49 #include "stacktrace.h"
50 #include "string-table.h"
51 #include "string-util.h"
52 #include "strv.h"
53 #include "user-util.h"
54 #include "util.h"
55
56 /* The maximum size up to which we process coredumps */
57 #define PROCESS_SIZE_MAX ((uint64_t) (2LLU*1024LLU*1024LLU*1024LLU))
58
59 /* The maximum size up to which we leave the coredump around on disk */
60 #define EXTERNAL_SIZE_MAX PROCESS_SIZE_MAX
61
62 /* The maximum size up to which we store the coredump in the journal */
63 #define JOURNAL_SIZE_MAX ((size_t) (767LU*1024LU*1024LU))
64
65 /* Make sure to not make this larger than the maximum journal entry
66 * size. See DATA_SIZE_MAX in journald-native.c. */
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", 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 NULL);
389
390 if (fstat(fd, &st) < 0) {
391 log_error_errno(errno, "Failed to fstat core file %s: %m", coredump_tmpfile_name(tmp));
392 goto fail;
393 }
394
395 if (lseek(fd, 0, SEEK_SET) == (off_t) -1) {
396 log_error_errno(errno, "Failed to seek on %s: %m", coredump_tmpfile_name(tmp));
397 goto fail;
398 }
399
400 #if HAVE_XZ || HAVE_LZ4
401 /* If we will remove the coredump anyway, do not compress. */
402 if (arg_compress && !maybe_remove_external_coredump(NULL, st.st_size)) {
403
404 _cleanup_free_ char *fn_compressed = NULL, *tmp_compressed = NULL;
405 _cleanup_close_ int fd_compressed = -1;
406
407 fn_compressed = strappend(fn, COMPRESSED_EXT);
408 if (!fn_compressed) {
409 log_oom();
410 goto uncompressed;
411 }
412
413 fd_compressed = open_tmpfile_linkable(fn_compressed, O_RDWR|O_CLOEXEC, &tmp_compressed);
414 if (fd_compressed < 0) {
415 log_error_errno(fd_compressed, "Failed to create temporary file for coredump %s: %m", fn_compressed);
416 goto uncompressed;
417 }
418
419 r = compress_stream(fd, fd_compressed, -1);
420 if (r < 0) {
421 log_error_errno(r, "Failed to compress %s: %m", coredump_tmpfile_name(tmp_compressed));
422 goto fail_compressed;
423 }
424
425 r = fix_permissions(fd_compressed, tmp_compressed, fn_compressed, context, uid);
426 if (r < 0)
427 goto fail_compressed;
428
429 /* OK, this worked, we can get rid of the uncompressed version now */
430 if (tmp)
431 unlink_noerrno(tmp);
432
433 *ret_filename = TAKE_PTR(fn_compressed); /* compressed */
434 *ret_node_fd = TAKE_FD(fd_compressed); /* compressed */
435 *ret_data_fd = TAKE_FD(fd); /* uncompressed */
436 *ret_size = (uint64_t) st.st_size; /* uncompressed */
437
438 return 0;
439
440 fail_compressed:
441 if (tmp_compressed)
442 (void) unlink(tmp_compressed);
443 }
444
445 uncompressed:
446 #endif
447
448 r = fix_permissions(fd, tmp, fn, context, uid);
449 if (r < 0)
450 goto fail;
451
452 *ret_filename = TAKE_PTR(fn);
453 *ret_data_fd = TAKE_FD(fd);
454 *ret_node_fd = -1;
455 *ret_size = (uint64_t) st.st_size;
456
457 return 0;
458
459 fail:
460 if (tmp)
461 (void) unlink(tmp);
462 return r;
463 }
464
465 static int allocate_journal_field(int fd, size_t size, char **ret, size_t *ret_size) {
466 _cleanup_free_ char *field = NULL;
467 ssize_t n;
468
469 assert(fd >= 0);
470 assert(ret);
471 assert(ret_size);
472
473 if (lseek(fd, 0, SEEK_SET) == (off_t) -1)
474 return log_warning_errno(errno, "Failed to seek: %m");
475
476 field = malloc(9 + size);
477 if (!field) {
478 log_warning("Failed to allocate memory for coredump, coredump will not be stored.");
479 return -ENOMEM;
480 }
481
482 memcpy(field, "COREDUMP=", 9);
483
484 n = read(fd, field + 9, size);
485 if (n < 0)
486 return log_error_errno((int) n, "Failed to read core data: %m");
487 if ((size_t) n < size) {
488 log_error("Core data too short.");
489 return -EIO;
490 }
491
492 *ret = TAKE_PTR(field);
493 *ret_size = size + 9;
494
495 return 0;
496 }
497
498 /* Joins /proc/[pid]/fd/ and /proc/[pid]/fdinfo/ into the following lines:
499 * 0:/dev/pts/23
500 * pos: 0
501 * flags: 0100002
502 *
503 * 1:/dev/pts/23
504 * pos: 0
505 * flags: 0100002
506 *
507 * 2:/dev/pts/23
508 * pos: 0
509 * flags: 0100002
510 * EOF
511 */
512 static int compose_open_fds(pid_t pid, char **open_fds) {
513 _cleanup_closedir_ DIR *proc_fd_dir = NULL;
514 _cleanup_close_ int proc_fdinfo_fd = -1;
515 _cleanup_free_ char *buffer = NULL;
516 _cleanup_fclose_ FILE *stream = NULL;
517 const char *fddelim = "", *path;
518 struct dirent *dent = NULL;
519 size_t size = 0;
520 int r = 0;
521
522 assert(pid >= 0);
523 assert(open_fds != NULL);
524
525 path = procfs_file_alloca(pid, "fd");
526 proc_fd_dir = opendir(path);
527 if (!proc_fd_dir)
528 return -errno;
529
530 proc_fdinfo_fd = openat(dirfd(proc_fd_dir), "../fdinfo", O_DIRECTORY|O_NOFOLLOW|O_CLOEXEC|O_PATH);
531 if (proc_fdinfo_fd < 0)
532 return -errno;
533
534 stream = open_memstream(&buffer, &size);
535 if (!stream)
536 return -ENOMEM;
537
538 (void) __fsetlocking(stream, FSETLOCKING_BYCALLER);
539
540 FOREACH_DIRENT(dent, proc_fd_dir, return -errno) {
541 _cleanup_fclose_ FILE *fdinfo = NULL;
542 _cleanup_free_ char *fdname = NULL;
543 char line[LINE_MAX];
544 int fd;
545
546 r = readlinkat_malloc(dirfd(proc_fd_dir), dent->d_name, &fdname);
547 if (r < 0)
548 return r;
549
550 fprintf(stream, "%s%s:%s\n", fddelim, dent->d_name, fdname);
551 fddelim = "\n";
552
553 /* Use the directory entry from /proc/[pid]/fd with /proc/[pid]/fdinfo */
554 fd = openat(proc_fdinfo_fd, dent->d_name, O_NOFOLLOW|O_CLOEXEC|O_RDONLY);
555 if (fd < 0)
556 continue;
557
558 fdinfo = fdopen(fd, "re");
559 if (!fdinfo) {
560 safe_close(fd);
561 continue;
562 }
563
564 FOREACH_LINE(line, fdinfo, break) {
565 fputs(line, stream);
566 if (!endswith(line, "\n"))
567 fputc('\n', stream);
568 }
569 }
570
571 errno = 0;
572 stream = safe_fclose(stream);
573
574 if (errno > 0)
575 return -errno;
576
577 *open_fds = TAKE_PTR(buffer);
578
579 return 0;
580 }
581
582 static int get_process_ns(pid_t pid, const char *namespace, ino_t *ns) {
583 const char *p;
584 struct stat stbuf;
585 _cleanup_close_ int proc_ns_dir_fd;
586
587 p = procfs_file_alloca(pid, "ns");
588
589 proc_ns_dir_fd = open(p, O_DIRECTORY | O_CLOEXEC | O_RDONLY);
590 if (proc_ns_dir_fd < 0)
591 return -errno;
592
593 if (fstatat(proc_ns_dir_fd, namespace, &stbuf, /* flags */0) < 0)
594 return -errno;
595
596 *ns = stbuf.st_ino;
597 return 0;
598 }
599
600 static int get_mount_namespace_leader(pid_t pid, pid_t *container_pid) {
601 pid_t cpid = pid, ppid = 0;
602 ino_t proc_mntns;
603 int r = 0;
604
605 r = get_process_ns(pid, "mnt", &proc_mntns);
606 if (r < 0)
607 return r;
608
609 for (;;) {
610 ino_t parent_mntns;
611
612 r = get_process_ppid(cpid, &ppid);
613 if (r < 0)
614 return r;
615
616 r = get_process_ns(ppid, "mnt", &parent_mntns);
617 if (r < 0)
618 return r;
619
620 if (proc_mntns != parent_mntns)
621 break;
622
623 if (ppid == 1)
624 return -ENOENT;
625
626 cpid = ppid;
627 }
628
629 *container_pid = ppid;
630 return 0;
631 }
632
633 /* Returns 1 if the parent was found.
634 * Returns 0 if there is not a process we can call the pid's
635 * container parent (the pid's process isn't 'containerized').
636 * Returns a negative number on errors.
637 */
638 static int get_process_container_parent_cmdline(pid_t pid, char** cmdline) {
639 int r = 0;
640 pid_t container_pid;
641 const char *proc_root_path;
642 struct stat root_stat, proc_root_stat;
643
644 /* To compare inodes of / and /proc/[pid]/root */
645 if (stat("/", &root_stat) < 0)
646 return -errno;
647
648 proc_root_path = procfs_file_alloca(pid, "root");
649 if (stat(proc_root_path, &proc_root_stat) < 0)
650 return -errno;
651
652 /* The process uses system root. */
653 if (proc_root_stat.st_ino == root_stat.st_ino) {
654 *cmdline = NULL;
655 return 0;
656 }
657
658 r = get_mount_namespace_leader(pid, &container_pid);
659 if (r < 0)
660 return r;
661
662 r = get_process_cmdline(container_pid, 0, false, cmdline);
663 if (r < 0)
664 return r;
665
666 return 1;
667 }
668
669 static int change_uid_gid(const char *context[]) {
670 uid_t uid;
671 gid_t gid;
672 int r;
673
674 r = parse_uid(context[CONTEXT_UID], &uid);
675 if (r < 0)
676 return r;
677
678 if (uid <= SYSTEM_UID_MAX) {
679 const char *user = "systemd-coredump";
680
681 r = get_user_creds(&user, &uid, &gid, NULL, NULL);
682 if (r < 0) {
683 log_warning_errno(r, "Cannot resolve %s user. Proceeding to dump core as root: %m", user);
684 uid = gid = 0;
685 }
686 } else {
687 r = parse_gid(context[CONTEXT_GID], &gid);
688 if (r < 0)
689 return r;
690 }
691
692 return drop_privileges(uid, gid, 0);
693 }
694
695 static bool is_journald_crash(const char *context[_CONTEXT_MAX]) {
696 assert(context);
697
698 return streq_ptr(context[CONTEXT_UNIT], SPECIAL_JOURNALD_SERVICE);
699 }
700
701 static bool is_pid1_crash(const char *context[_CONTEXT_MAX]) {
702 assert(context);
703
704 return streq_ptr(context[CONTEXT_UNIT], SPECIAL_INIT_SCOPE) ||
705 streq_ptr(context[CONTEXT_PID], "1");
706 }
707
708 #define SUBMIT_COREDUMP_FIELDS 4
709
710 static int submit_coredump(
711 const char *context[_CONTEXT_MAX],
712 struct iovec *iovec,
713 size_t n_iovec_allocated,
714 size_t n_iovec,
715 int input_fd) {
716
717 _cleanup_close_ int coredump_fd = -1, coredump_node_fd = -1;
718 _cleanup_free_ char *core_message = NULL, *filename = NULL, *coredump_data = NULL;
719 uint64_t coredump_size = UINT64_MAX;
720 bool truncated = false, journald_crash;
721 int r;
722
723 assert(context);
724 assert(iovec);
725 assert(n_iovec_allocated >= n_iovec + SUBMIT_COREDUMP_FIELDS);
726 assert(input_fd >= 0);
727
728 journald_crash = is_journald_crash(context);
729
730 /* Vacuum before we write anything again */
731 (void) coredump_vacuum(-1, arg_keep_free, arg_max_use);
732
733 /* Always stream the coredump to disk, if that's possible */
734 r = save_external_coredump(context, input_fd,
735 &filename, &coredump_node_fd, &coredump_fd, &coredump_size, &truncated);
736 if (r < 0)
737 /* Skip whole core dumping part */
738 goto log;
739
740 /* If we don't want to keep the coredump on disk, remove it now, as later on we will lack the privileges for
741 * it. However, we keep the fd to it, so that we can still process it and log it. */
742 r = maybe_remove_external_coredump(filename, coredump_size);
743 if (r < 0)
744 return r;
745 if (r == 0) {
746 const char *coredump_filename;
747
748 coredump_filename = strjoina("COREDUMP_FILENAME=", filename);
749 iovec[n_iovec++] = IOVEC_MAKE_STRING(coredump_filename);
750 } else if (arg_storage == COREDUMP_STORAGE_EXTERNAL)
751 log_info("The core will not be stored: size %"PRIu64" is greater than %"PRIu64" (the configured maximum)",
752 coredump_size, arg_external_size_max);
753
754 /* Vacuum again, but exclude the coredump we just created */
755 (void) coredump_vacuum(coredump_node_fd >= 0 ? coredump_node_fd : coredump_fd, arg_keep_free, arg_max_use);
756
757 /* Now, let's drop privileges to become the user who owns the segfaulted process and allocate the coredump
758 * memory under the user's uid. This also ensures that the credentials journald will see are the ones of the
759 * coredumping user, thus making sure the user gets access to the core dump. Let's also get rid of all
760 * capabilities, if we run as root, we won't need them anymore. */
761 r = change_uid_gid(context);
762 if (r < 0)
763 return log_error_errno(r, "Failed to drop privileges: %m");
764
765 #if HAVE_ELFUTILS
766 /* Try to get a strack trace if we can */
767 if (coredump_size <= arg_process_size_max) {
768 _cleanup_free_ char *stacktrace = NULL;
769
770 r = coredump_make_stack_trace(coredump_fd, context[CONTEXT_EXE], &stacktrace);
771 if (r >= 0)
772 core_message = strjoin("MESSAGE=Process ", context[CONTEXT_PID],
773 " (", context[CONTEXT_COMM], ") of user ",
774 context[CONTEXT_UID], " dumped core.",
775 journald_crash ? "\nCoredump diverted to " : "",
776 journald_crash ? filename : "",
777 "\n\n", stacktrace);
778 else if (r == -EINVAL)
779 log_warning("Failed to generate stack trace: %s", dwfl_errmsg(dwfl_errno()));
780 else
781 log_warning_errno(r, "Failed to generate stack trace: %m");
782 } else
783 log_debug("Not generating stack trace: core size %"PRIu64" is greater than %"PRIu64" (the configured maximum)",
784 coredump_size, arg_process_size_max);
785
786 if (!core_message)
787 #endif
788 log:
789 core_message = strjoin("MESSAGE=Process ", context[CONTEXT_PID],
790 " (", context[CONTEXT_COMM], ") of user ",
791 context[CONTEXT_UID], " dumped core.",
792 journald_crash ? "\nCoredump diverted to " : NULL,
793 journald_crash ? filename : NULL);
794 if (!core_message)
795 return log_oom();
796
797 if (journald_crash) {
798 /* We cannot log to the journal, so just print the MESSAGE.
799 * The target was set previously to something safe. */
800 log_dispatch(LOG_ERR, 0, core_message);
801 return 0;
802 }
803
804 iovec[n_iovec++] = IOVEC_MAKE_STRING(core_message);
805
806 if (truncated)
807 iovec[n_iovec++] = IOVEC_MAKE_STRING("COREDUMP_TRUNCATED=1");
808
809 /* Optionally store the entire coredump in the journal */
810 if (arg_storage == COREDUMP_STORAGE_JOURNAL) {
811 if (coredump_size <= arg_journal_size_max) {
812 size_t sz = 0;
813
814 /* Store the coredump itself in the journal */
815
816 r = allocate_journal_field(coredump_fd, (size_t) coredump_size, &coredump_data, &sz);
817 if (r >= 0)
818 iovec[n_iovec++] = IOVEC_MAKE(coredump_data, sz);
819 else
820 log_warning_errno(r, "Failed to attach the core to the journal entry: %m");
821 } else
822 log_info("The core will not be stored: size %"PRIu64" is greater than %"PRIu64" (the configured maximum)",
823 coredump_size, arg_journal_size_max);
824 }
825
826 assert(n_iovec <= n_iovec_allocated);
827
828 r = sd_journal_sendv(iovec, n_iovec);
829 if (r < 0)
830 return log_error_errno(r, "Failed to log coredump: %m");
831
832 return 0;
833 }
834
835 static void map_context_fields(const struct iovec *iovec, const char* context[]) {
836
837 static const char * const context_field_names[] = {
838 [CONTEXT_PID] = "COREDUMP_PID=",
839 [CONTEXT_UID] = "COREDUMP_UID=",
840 [CONTEXT_GID] = "COREDUMP_GID=",
841 [CONTEXT_SIGNAL] = "COREDUMP_SIGNAL=",
842 [CONTEXT_TIMESTAMP] = "COREDUMP_TIMESTAMP=",
843 [CONTEXT_RLIMIT] = "COREDUMP_RLIMIT=",
844 [CONTEXT_HOSTNAME] = "COREDUMP_HOSTNAME=",
845 [CONTEXT_COMM] = "COREDUMP_COMM=",
846 [CONTEXT_EXE] = "COREDUMP_EXE=",
847 };
848
849 unsigned i;
850
851 assert(iovec);
852 assert(context);
853
854 for (i = 0; i < ELEMENTSOF(context_field_names); i++) {
855 char *p;
856
857 if (!context_field_names[i])
858 continue;
859
860 p = memory_startswith(iovec->iov_base, iovec->iov_len, context_field_names[i]);
861 if (!p)
862 continue;
863
864 /* Note that these strings are NUL terminated, because we made sure that a trailing NUL byte is in the
865 * buffer, though not included in the iov_len count. (see below) */
866 context[i] = p;
867 break;
868 }
869 }
870
871 static int process_socket(int fd) {
872 _cleanup_close_ int coredump_fd = -1;
873 struct iovec *iovec = NULL;
874 size_t n_iovec = 0, n_allocated = 0, i, k;
875 const char *context[_CONTEXT_MAX] = {};
876 int r;
877
878 assert(fd >= 0);
879
880 log_set_target(LOG_TARGET_AUTO);
881 log_parse_environment();
882 log_open();
883
884 log_debug("Processing coredump received on stdin...");
885
886 for (;;) {
887 union {
888 struct cmsghdr cmsghdr;
889 uint8_t buf[CMSG_SPACE(sizeof(int))];
890 } control = {};
891 struct msghdr mh = {
892 .msg_control = &control,
893 .msg_controllen = sizeof(control),
894 .msg_iovlen = 1,
895 };
896 ssize_t n;
897 ssize_t l;
898
899 if (!GREEDY_REALLOC(iovec, n_allocated, n_iovec + SUBMIT_COREDUMP_FIELDS)) {
900 r = log_oom();
901 goto finish;
902 }
903
904 l = next_datagram_size_fd(fd);
905 if (l < 0) {
906 r = log_error_errno(l, "Failed to determine datagram size to read: %m");
907 goto finish;
908 }
909
910 assert(l >= 0);
911
912 iovec[n_iovec].iov_len = l;
913 iovec[n_iovec].iov_base = malloc(l + 1);
914 if (!iovec[n_iovec].iov_base) {
915 r = log_oom();
916 goto finish;
917 }
918
919 mh.msg_iov = iovec + n_iovec;
920
921 n = recvmsg(fd, &mh, MSG_NOSIGNAL|MSG_CMSG_CLOEXEC);
922 if (n < 0) {
923 free(iovec[n_iovec].iov_base);
924 r = log_error_errno(errno, "Failed to receive datagram: %m");
925 goto finish;
926 }
927
928 if (n == 0) {
929 struct cmsghdr *cmsg, *found = NULL;
930 /* The final zero-length datagram carries the file descriptor and tells us that we're done. */
931
932 free(iovec[n_iovec].iov_base);
933
934 CMSG_FOREACH(cmsg, &mh) {
935 if (cmsg->cmsg_level == SOL_SOCKET &&
936 cmsg->cmsg_type == SCM_RIGHTS &&
937 cmsg->cmsg_len == CMSG_LEN(sizeof(int))) {
938 assert(!found);
939 found = cmsg;
940 }
941 }
942
943 if (!found) {
944 log_error("Coredump file descriptor missing.");
945 r = -EBADMSG;
946 goto finish;
947 }
948
949 assert(coredump_fd < 0);
950 coredump_fd = *(int*) CMSG_DATA(found);
951 break;
952 }
953
954 /* Add trailing NUL byte, in case these are strings */
955 ((char*) iovec[n_iovec].iov_base)[n] = 0;
956 iovec[n_iovec].iov_len = (size_t) n;
957
958 cmsg_close_all(&mh);
959 map_context_fields(iovec + n_iovec, context);
960 n_iovec++;
961 }
962
963 if (!GREEDY_REALLOC(iovec, n_allocated, n_iovec + SUBMIT_COREDUMP_FIELDS)) {
964 r = log_oom();
965 goto finish;
966 }
967
968 /* Make sure we got all data we really need */
969 assert(context[CONTEXT_PID]);
970 assert(context[CONTEXT_UID]);
971 assert(context[CONTEXT_GID]);
972 assert(context[CONTEXT_SIGNAL]);
973 assert(context[CONTEXT_TIMESTAMP]);
974 assert(context[CONTEXT_RLIMIT]);
975 assert(context[CONTEXT_HOSTNAME]);
976 assert(context[CONTEXT_COMM]);
977 assert(coredump_fd >= 0);
978
979 /* Small quirk: the journal fields contain the timestamp padded with six zeroes, so that the kernel-supplied 1s
980 * granularity timestamps becomes 1µs granularity, i.e. the granularity systemd usually operates in. Since we
981 * are reconstructing the original kernel context, we chop this off again, here. */
982 k = strlen(context[CONTEXT_TIMESTAMP]);
983 if (k > 6)
984 context[CONTEXT_TIMESTAMP] = strndupa(context[CONTEXT_TIMESTAMP], k - 6);
985
986 r = submit_coredump(context, iovec, n_allocated, n_iovec, coredump_fd);
987
988 finish:
989 for (i = 0; i < n_iovec; i++)
990 free(iovec[i].iov_base);
991 free(iovec);
992
993 return r;
994 }
995
996 static int send_iovec(const struct iovec iovec[], size_t n_iovec, int input_fd) {
997
998 static const union sockaddr_union sa = {
999 .un.sun_family = AF_UNIX,
1000 .un.sun_path = "/run/systemd/coredump",
1001 };
1002 _cleanup_close_ int fd = -1;
1003 size_t i;
1004 int r;
1005
1006 assert(iovec || n_iovec <= 0);
1007 assert(input_fd >= 0);
1008
1009 fd = socket(AF_UNIX, SOCK_SEQPACKET|SOCK_CLOEXEC, 0);
1010 if (fd < 0)
1011 return log_error_errno(errno, "Failed to create coredump socket: %m");
1012
1013 if (connect(fd, &sa.sa, SOCKADDR_UN_LEN(sa.un)) < 0)
1014 return log_error_errno(errno, "Failed to connect to coredump service: %m");
1015
1016 for (i = 0; i < n_iovec; i++) {
1017 struct msghdr mh = {
1018 .msg_iov = (struct iovec*) iovec + i,
1019 .msg_iovlen = 1,
1020 };
1021 struct iovec copy[2];
1022
1023 for (;;) {
1024 if (sendmsg(fd, &mh, MSG_NOSIGNAL) >= 0)
1025 break;
1026
1027 if (errno == EMSGSIZE && mh.msg_iov[0].iov_len > 0) {
1028 /* This field didn't fit? That's a pity. Given that this is just metadata,
1029 * let's truncate the field at half, and try again. We append three dots, in
1030 * order to show that this is truncated. */
1031
1032 if (mh.msg_iov != copy) {
1033 /* We don't want to modify the caller's iovec, hence let's create our
1034 * own array, consisting of two new iovecs, where the first is a
1035 * (truncated) copy of what we want to send, and the second one
1036 * contains the trailing dots. */
1037 copy[0] = iovec[i];
1038 copy[1] = (struct iovec) {
1039 .iov_base = (char[]) { '.', '.', '.' },
1040 .iov_len = 3,
1041 };
1042
1043 mh.msg_iov = copy;
1044 mh.msg_iovlen = 2;
1045 }
1046
1047 copy[0].iov_len /= 2; /* halve it, and try again */
1048 continue;
1049 }
1050
1051 return log_error_errno(errno, "Failed to send coredump datagram: %m");
1052 }
1053 }
1054
1055 r = send_one_fd(fd, input_fd, 0);
1056 if (r < 0)
1057 return log_error_errno(r, "Failed to send coredump fd: %m");
1058
1059 return 0;
1060 }
1061
1062 static char* set_iovec_field(struct iovec *iovec, size_t *n_iovec, const char *field, const char *value) {
1063 char *x;
1064
1065 x = strappend(field, value);
1066 if (x)
1067 iovec[(*n_iovec)++] = IOVEC_MAKE_STRING(x);
1068 return x;
1069 }
1070
1071 static char* set_iovec_field_free(struct iovec *iovec, size_t *n_iovec, const char *field, char *value) {
1072 char *x;
1073
1074 x = set_iovec_field(iovec, n_iovec, field, value);
1075 free(value);
1076 return x;
1077 }
1078
1079 static int gather_pid_metadata(
1080 char* context[_CONTEXT_MAX],
1081 char **comm_fallback,
1082 struct iovec *iovec, size_t *n_iovec) {
1083
1084 /* We need 27 empty slots in iovec!
1085 *
1086 * Note that if we fail on oom later on, we do not roll-back changes to the iovec structure. (It remains valid,
1087 * with the first n_iovec fields initialized.) */
1088
1089 uid_t owner_uid;
1090 pid_t pid;
1091 char *t;
1092 const char *p;
1093 int r, signo;
1094
1095 r = parse_pid(context[CONTEXT_PID], &pid);
1096 if (r < 0)
1097 return log_error_errno(r, "Failed to parse PID \"%s\": %m", context[CONTEXT_PID]);
1098
1099 r = get_process_comm(pid, &context[CONTEXT_COMM]);
1100 if (r < 0) {
1101 log_warning_errno(r, "Failed to get COMM, falling back to the command line: %m");
1102 context[CONTEXT_COMM] = strv_join(comm_fallback, " ");
1103 if (!context[CONTEXT_COMM])
1104 return log_oom();
1105 }
1106
1107 r = get_process_exe(pid, &context[CONTEXT_EXE]);
1108 if (r < 0)
1109 log_warning_errno(r, "Failed to get EXE, ignoring: %m");
1110
1111 if (cg_pid_get_unit(pid, &context[CONTEXT_UNIT]) >= 0) {
1112 if (!is_journald_crash((const char**) context)) {
1113 /* OK, now we know it's not the journal, hence we can make use of it now. */
1114 log_set_target(LOG_TARGET_JOURNAL_OR_KMSG);
1115 log_open();
1116 }
1117
1118 /* If this is PID 1 disable coredump collection, we'll unlikely be able to process it later on. */
1119 if (is_pid1_crash((const char**) context)) {
1120 log_notice("Due to PID 1 having crashed coredump collection will now be turned off.");
1121 disable_coredumps();
1122 }
1123
1124 set_iovec_field(iovec, n_iovec, "COREDUMP_UNIT=", context[CONTEXT_UNIT]);
1125 }
1126
1127 if (cg_pid_get_user_unit(pid, &t) >= 0)
1128 set_iovec_field_free(iovec, n_iovec, "COREDUMP_USER_UNIT=", t);
1129
1130 /* The next few are mandatory */
1131 if (!set_iovec_field(iovec, n_iovec, "COREDUMP_PID=", context[CONTEXT_PID]))
1132 return log_oom();
1133
1134 if (!set_iovec_field(iovec, n_iovec, "COREDUMP_UID=", context[CONTEXT_UID]))
1135 return log_oom();
1136
1137 if (!set_iovec_field(iovec, n_iovec, "COREDUMP_GID=", context[CONTEXT_GID]))
1138 return log_oom();
1139
1140 if (!set_iovec_field(iovec, n_iovec, "COREDUMP_SIGNAL=", context[CONTEXT_SIGNAL]))
1141 return log_oom();
1142
1143 if (!set_iovec_field(iovec, n_iovec, "COREDUMP_RLIMIT=", context[CONTEXT_RLIMIT]))
1144 return log_oom();
1145
1146 if (!set_iovec_field(iovec, n_iovec, "COREDUMP_HOSTNAME=", context[CONTEXT_HOSTNAME]))
1147 return log_oom();
1148
1149 if (!set_iovec_field(iovec, n_iovec, "COREDUMP_COMM=", context[CONTEXT_COMM]))
1150 return log_oom();
1151
1152 if (context[CONTEXT_EXE] &&
1153 !set_iovec_field(iovec, n_iovec, "COREDUMP_EXE=", context[CONTEXT_EXE]))
1154 return log_oom();
1155
1156 if (sd_pid_get_session(pid, &t) >= 0)
1157 set_iovec_field_free(iovec, n_iovec, "COREDUMP_SESSION=", t);
1158
1159 if (sd_pid_get_owner_uid(pid, &owner_uid) >= 0) {
1160 r = asprintf(&t, "COREDUMP_OWNER_UID=" UID_FMT, owner_uid);
1161 if (r > 0)
1162 iovec[(*n_iovec)++] = IOVEC_MAKE_STRING(t);
1163 }
1164
1165 if (sd_pid_get_slice(pid, &t) >= 0)
1166 set_iovec_field_free(iovec, n_iovec, "COREDUMP_SLICE=", t);
1167
1168 if (get_process_cmdline(pid, 0, false, &t) >= 0)
1169 set_iovec_field_free(iovec, n_iovec, "COREDUMP_CMDLINE=", t);
1170
1171 if (cg_pid_get_path_shifted(pid, NULL, &t) >= 0)
1172 set_iovec_field_free(iovec, n_iovec, "COREDUMP_CGROUP=", t);
1173
1174 if (compose_open_fds(pid, &t) >= 0)
1175 set_iovec_field_free(iovec, n_iovec, "COREDUMP_OPEN_FDS=", t);
1176
1177 p = procfs_file_alloca(pid, "status");
1178 if (read_full_file(p, &t, NULL) >= 0)
1179 set_iovec_field_free(iovec, n_iovec, "COREDUMP_PROC_STATUS=", t);
1180
1181 p = procfs_file_alloca(pid, "maps");
1182 if (read_full_file(p, &t, NULL) >= 0)
1183 set_iovec_field_free(iovec, n_iovec, "COREDUMP_PROC_MAPS=", t);
1184
1185 p = procfs_file_alloca(pid, "limits");
1186 if (read_full_file(p, &t, NULL) >= 0)
1187 set_iovec_field_free(iovec, n_iovec, "COREDUMP_PROC_LIMITS=", t);
1188
1189 p = procfs_file_alloca(pid, "cgroup");
1190 if (read_full_file(p, &t, NULL) >=0)
1191 set_iovec_field_free(iovec, n_iovec, "COREDUMP_PROC_CGROUP=", t);
1192
1193 p = procfs_file_alloca(pid, "mountinfo");
1194 if (read_full_file(p, &t, NULL) >=0)
1195 set_iovec_field_free(iovec, n_iovec, "COREDUMP_PROC_MOUNTINFO=", t);
1196
1197 if (get_process_cwd(pid, &t) >= 0)
1198 set_iovec_field_free(iovec, n_iovec, "COREDUMP_CWD=", t);
1199
1200 if (get_process_root(pid, &t) >= 0) {
1201 bool proc_self_root_is_slash;
1202
1203 proc_self_root_is_slash = strcmp(t, "/") == 0;
1204
1205 set_iovec_field_free(iovec, n_iovec, "COREDUMP_ROOT=", t);
1206
1207 /* If the process' root is "/", then there is a chance it has
1208 * mounted own root and hence being containerized. */
1209 if (proc_self_root_is_slash && get_process_container_parent_cmdline(pid, &t) > 0)
1210 set_iovec_field_free(iovec, n_iovec, "COREDUMP_CONTAINER_CMDLINE=", t);
1211 }
1212
1213 if (get_process_environ(pid, &t) >= 0)
1214 set_iovec_field_free(iovec, n_iovec, "COREDUMP_ENVIRON=", t);
1215
1216 t = strjoin("COREDUMP_TIMESTAMP=", context[CONTEXT_TIMESTAMP], "000000");
1217 if (t)
1218 iovec[(*n_iovec)++] = IOVEC_MAKE_STRING(t);
1219
1220 if (safe_atoi(context[CONTEXT_SIGNAL], &signo) >= 0 && SIGNAL_VALID(signo))
1221 set_iovec_field(iovec, n_iovec, "COREDUMP_SIGNAL_NAME=SIG", signal_to_string(signo));
1222
1223 return 0; /* we successfully acquired all metadata */
1224 }
1225
1226 static int process_kernel(int argc, char* argv[]) {
1227
1228 char* context[_CONTEXT_MAX] = {};
1229 struct iovec iovec[29 + SUBMIT_COREDUMP_FIELDS];
1230 size_t i, n_iovec, n_to_free = 0;
1231 int r;
1232
1233 log_debug("Processing coredump received from the kernel...");
1234
1235 if (argc < CONTEXT_COMM + 1) {
1236 log_error("Not enough arguments passed by the kernel (%i, expected %i).", argc - 1, CONTEXT_COMM + 1 - 1);
1237 return -EINVAL;
1238 }
1239
1240 context[CONTEXT_PID] = argv[1 + CONTEXT_PID];
1241 context[CONTEXT_UID] = argv[1 + CONTEXT_UID];
1242 context[CONTEXT_GID] = argv[1 + CONTEXT_GID];
1243 context[CONTEXT_SIGNAL] = argv[1 + CONTEXT_SIGNAL];
1244 context[CONTEXT_TIMESTAMP] = argv[1 + CONTEXT_TIMESTAMP];
1245 context[CONTEXT_RLIMIT] = argv[1 + CONTEXT_RLIMIT];
1246 context[CONTEXT_HOSTNAME] = argv[1 + CONTEXT_HOSTNAME];
1247
1248 r = gather_pid_metadata(context, argv + 1 + CONTEXT_COMM, iovec, &n_to_free);
1249 if (r < 0)
1250 goto finish;
1251
1252 n_iovec = n_to_free;
1253
1254 iovec[n_iovec++] = IOVEC_MAKE_STRING("MESSAGE_ID=" SD_MESSAGE_COREDUMP_STR);
1255
1256 assert_cc(2 == LOG_CRIT);
1257 iovec[n_iovec++] = IOVEC_MAKE_STRING("PRIORITY=2");
1258
1259 assert(n_iovec <= ELEMENTSOF(iovec));
1260
1261 if (is_journald_crash((const char**) context) || is_pid1_crash((const char**) context))
1262 r = submit_coredump((const char**) context,
1263 iovec, ELEMENTSOF(iovec), n_iovec,
1264 STDIN_FILENO);
1265 else
1266 r = send_iovec(iovec, n_iovec, STDIN_FILENO);
1267
1268 finish:
1269 for (i = 0; i < n_to_free; i++)
1270 free(iovec[i].iov_base);
1271
1272 /* Those fields are allocated by gather_pid_metadata */
1273 free(context[CONTEXT_COMM]);
1274 free(context[CONTEXT_EXE]);
1275 free(context[CONTEXT_UNIT]);
1276
1277 return r;
1278 }
1279
1280 static int process_backtrace(int argc, char *argv[]) {
1281 char *context[_CONTEXT_MAX] = {};
1282 _cleanup_free_ char *message = NULL;
1283 _cleanup_free_ struct iovec *iovec = NULL;
1284 size_t n_iovec, n_allocated, n_to_free = 0, i;
1285 int r;
1286 JournalImporter importer = {
1287 .fd = STDIN_FILENO,
1288 };
1289
1290 log_debug("Processing backtrace on stdin...");
1291
1292 if (argc < CONTEXT_COMM + 1) {
1293 log_error("Not enough arguments passed (%i, expected %i).", argc - 1, CONTEXT_COMM + 1 - 1);
1294 return -EINVAL;
1295 }
1296
1297 context[CONTEXT_PID] = argv[2 + CONTEXT_PID];
1298 context[CONTEXT_UID] = argv[2 + CONTEXT_UID];
1299 context[CONTEXT_GID] = argv[2 + CONTEXT_GID];
1300 context[CONTEXT_SIGNAL] = argv[2 + CONTEXT_SIGNAL];
1301 context[CONTEXT_TIMESTAMP] = argv[2 + CONTEXT_TIMESTAMP];
1302 context[CONTEXT_RLIMIT] = argv[2 + CONTEXT_RLIMIT];
1303 context[CONTEXT_HOSTNAME] = argv[2 + CONTEXT_HOSTNAME];
1304
1305 n_allocated = 34 + COREDUMP_STORAGE_EXTERNAL;
1306 /* 26 metadata, 2 static, +unknown input, 4 storage, rounded up */
1307 iovec = new(struct iovec, n_allocated);
1308 if (!iovec)
1309 return log_oom();
1310
1311 r = gather_pid_metadata(context, argv + 2 + CONTEXT_COMM, iovec, &n_to_free);
1312 if (r < 0)
1313 goto finish;
1314 if (r > 0) {
1315 /* This was a special crash, and has already been processed. */
1316 r = 0;
1317 goto finish;
1318 }
1319 n_iovec = n_to_free;
1320
1321 for (;;) {
1322 r = journal_importer_process_data(&importer);
1323 if (r < 0) {
1324 log_error_errno(r, "Failed to parse journal entry on stdin: %m");
1325 goto finish;
1326 }
1327 if (r == 1 || /* complete entry */
1328 journal_importer_eof(&importer)) /* end of data */
1329 break;
1330 }
1331
1332 if (!GREEDY_REALLOC(iovec, n_allocated, n_iovec + importer.iovw.count + 2))
1333 return log_oom();
1334
1335 if (journal_importer_eof(&importer)) {
1336 log_warning("Did not receive a full journal entry on stdin, ignoring message sent by reporter");
1337
1338 message = strjoin("MESSAGE=Process ", context[CONTEXT_PID],
1339 " (", context[CONTEXT_COMM], ")"
1340 " of user ", context[CONTEXT_UID],
1341 " failed with ", context[CONTEXT_SIGNAL]);
1342 if (!message) {
1343 r = log_oom();
1344 goto finish;
1345 }
1346 iovec[n_iovec++] = IOVEC_MAKE_STRING(message);
1347 } else {
1348 for (i = 0; i < importer.iovw.count; i++)
1349 iovec[n_iovec++] = importer.iovw.iovec[i];
1350 }
1351
1352 iovec[n_iovec++] = IOVEC_MAKE_STRING("MESSAGE_ID=" SD_MESSAGE_BACKTRACE_STR);
1353 assert_cc(2 == LOG_CRIT);
1354 iovec[n_iovec++] = IOVEC_MAKE_STRING("PRIORITY=2");
1355
1356 assert(n_iovec <= n_allocated);
1357
1358 r = sd_journal_sendv(iovec, n_iovec);
1359 if (r < 0)
1360 log_error_errno(r, "Failed to log backtrace: %m");
1361
1362 finish:
1363 for (i = 0; i < n_to_free; i++)
1364 free(iovec[i].iov_base);
1365
1366 /* Those fields are allocated by gather_pid_metadata */
1367 free(context[CONTEXT_COMM]);
1368 free(context[CONTEXT_EXE]);
1369 free(context[CONTEXT_UNIT]);
1370
1371 return r;
1372 }
1373
1374 int main(int argc, char *argv[]) {
1375 int r;
1376
1377 /* First, log to a safe place, since we don't know what crashed and it might
1378 * be journald which we'd rather not log to then. */
1379
1380 log_set_target(LOG_TARGET_KMSG);
1381 log_open();
1382
1383 /* Make sure we never enter a loop */
1384 (void) prctl(PR_SET_DUMPABLE, 0);
1385
1386 /* Ignore all parse errors */
1387 (void) parse_config();
1388
1389 log_debug("Selected storage '%s'.", coredump_storage_to_string(arg_storage));
1390 log_debug("Selected compression %s.", yes_no(arg_compress));
1391
1392 r = sd_listen_fds(false);
1393 if (r < 0) {
1394 log_error_errno(r, "Failed to determine number of file descriptor: %m");
1395 goto finish;
1396 }
1397
1398 /* If we got an fd passed, we are running in coredumpd mode. Otherwise we
1399 * are invoked from the kernel as coredump handler. */
1400 if (r == 0) {
1401 if (streq_ptr(argv[1], "--backtrace"))
1402 r = process_backtrace(argc, argv);
1403 else
1404 r = process_kernel(argc, argv);
1405 } else if (r == 1)
1406 r = process_socket(SD_LISTEN_FDS_START);
1407 else {
1408 log_error("Received unexpected number of file descriptors.");
1409 r = -EINVAL;
1410 }
1411
1412 finish:
1413 return r < 0 ? EXIT_FAILURE : EXIT_SUCCESS;
1414 }