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