]> git.ipfire.org Git - thirdparty/systemd.git/blame - src/coredump/coredump.c
Merge pull request #11827 from keszybz/pkgconfig-variables
[thirdparty/systemd.git] / src / coredump / coredump.c
CommitLineData
53e1b683 1/* SPDX-License-Identifier: LGPL-2.1+ */
f5e04665
LP
2
3#include <errno.h>
803a3464 4#include <stdio.h>
0d536673 5#include <stdio_ext.h>
803a3464 6#include <sys/prctl.h>
cacd6403 7#include <sys/xattr.h>
4f5dd394 8#include <unistd.h>
f5e04665 9
349cc4a5 10#if HAVE_ELFUTILS
3c171f0b
LP
11#include <dwarf.h>
12#include <elfutils/libdwfl.h>
4d229b31
UTL
13#endif
14
73a99163 15#include "sd-daemon.h"
f11943c5
LP
16#include "sd-journal.h"
17#include "sd-login.h"
73a99163 18#include "sd-messages.h"
4f5dd394
LP
19
20#include "acl-util.h"
b5efdb8a 21#include "alloc-util.h"
430f0182 22#include "capability-util.h"
ba1261bc 23#include "cgroup-util.h"
4f5dd394 24#include "compress.h"
34c10968
LP
25#include "conf-parser.h"
26#include "copy.h"
f11943c5 27#include "coredump-vacuum.h"
a0956174 28#include "dirent-util.h"
4f5dd394 29#include "escape.h"
3ffd4af2 30#include "fd-util.h"
4f5dd394 31#include "fileio.h"
f4f15635 32#include "fs-util.h"
afc5dbf3 33#include "io-util.h"
b18453ed 34#include "journal-importer.h"
4f5dd394
LP
35#include "log.h"
36#include "macro.h"
5e332028 37#include "main-func.h"
0c773903 38#include "missing.h"
4f5dd394 39#include "mkdir.h"
6bedfcbb 40#include "parse-util.h"
0b452006 41#include "process-util.h"
d14bcb4e 42#include "signal-util.h"
3c171f0b 43#include "socket-util.h"
4f5dd394
LP
44#include "special.h"
45#include "stacktrace.h"
8b43440b 46#include "string-table.h"
07630cea 47#include "string-util.h"
4f5dd394 48#include "strv.h"
e4de7287 49#include "tmpfile-util.h"
b1d4f8e1 50#include "user-util.h"
4f5dd394 51#include "util.h"
34727273 52
34c10968 53/* The maximum size up to which we process coredumps */
59f448cf 54#define PROCESS_SIZE_MAX ((uint64_t) (2LLU*1024LLU*1024LLU*1024LLU))
34c10968 55
bdfd7b2c 56/* The maximum size up to which we leave the coredump around on disk */
34c10968
LP
57#define EXTERNAL_SIZE_MAX PROCESS_SIZE_MAX
58
bdfd7b2c 59/* The maximum size up to which we store the coredump in the journal */
25cad95c 60#ifndef FUZZING_BUILD_MODE_UNSAFE_FOR_PRODUCTION
34c10968 61#define JOURNAL_SIZE_MAX ((size_t) (767LU*1024LU*1024LU))
25cad95c
YW
62#else
63/* oss-fuzz limits memory usage. */
64#define JOURNAL_SIZE_MAX ((size_t) (10LU*1024LU*1024LU))
65#endif
f5e04665 66
c4aa09b0 67/* Make sure to not make this larger than the maximum journal entry
27f931d1 68 * size. See DATA_SIZE_MAX in journal-importer.h. */
874bc134 69assert_cc(JOURNAL_SIZE_MAX <= DATA_SIZE_MAX);
f5e04665
LP
70
71enum {
ea5cc2a8
ZJS
72 /* We use this as array indexes for a couple of special fields we use for
73 * naming coredump files, and attaching xattrs, and for indexing argv[].
74
75 * Our pattern for man:systectl(1) kernel.core_pattern is such that the
76 * kernel passes fields until CONTEXT_RLIMIT as arguments in argv[]. After
77 * that it gets complicated: the kernel passes "comm" as one or more fields
78 * starting at index CONTEXT_COMM (in other words, full "comm" is under index
79 * CONTEXT_COMM when it does not contain spaces, which is the common
80 * case). This mapping is not reversible, so we prefer to retrieve "comm"
81 * from /proc. We only fall back to argv[CONTEXT_COMM...] when that fails.
82 *
83 * In the internal context[] array, fields before CONTEXT_COMM are the
84 * strings from argv[], so they should not be freed. The strings at indices
85 * CONTEXT_COMM and higher are allocated by us and should be freed at the
86 * end.
87 */
3c171f0b
LP
88 CONTEXT_PID,
89 CONTEXT_UID,
90 CONTEXT_GID,
91 CONTEXT_SIGNAL,
92 CONTEXT_TIMESTAMP,
bdfd7b2c 93 CONTEXT_RLIMIT,
f45b8015 94 CONTEXT_HOSTNAME,
3c171f0b
LP
95 CONTEXT_COMM,
96 CONTEXT_EXE,
92e92d71 97 CONTEXT_UNIT,
3c171f0b 98 _CONTEXT_MAX
f5e04665
LP
99};
100
34c10968
LP
101typedef enum CoredumpStorage {
102 COREDUMP_STORAGE_NONE,
103 COREDUMP_STORAGE_EXTERNAL,
104 COREDUMP_STORAGE_JOURNAL,
34c10968
LP
105 _COREDUMP_STORAGE_MAX,
106 _COREDUMP_STORAGE_INVALID = -1
107} CoredumpStorage;
108
34c10968
LP
109static const char* const coredump_storage_table[_COREDUMP_STORAGE_MAX] = {
110 [COREDUMP_STORAGE_NONE] = "none",
111 [COREDUMP_STORAGE_EXTERNAL] = "external",
112 [COREDUMP_STORAGE_JOURNAL] = "journal",
34c10968
LP
113};
114
115DEFINE_PRIVATE_STRING_TABLE_LOOKUP(coredump_storage, CoredumpStorage);
8c9571d0 116static DEFINE_CONFIG_PARSE_ENUM(config_parse_coredump_storage, coredump_storage, CoredumpStorage, "Failed to parse storage setting");
34727273
ZJS
117
118static CoredumpStorage arg_storage = COREDUMP_STORAGE_EXTERNAL;
8c9571d0 119static bool arg_compress = true;
59f448cf
LP
120static uint64_t arg_process_size_max = PROCESS_SIZE_MAX;
121static uint64_t arg_external_size_max = EXTERNAL_SIZE_MAX;
6e2b4a69 122static uint64_t arg_journal_size_max = JOURNAL_SIZE_MAX;
59f448cf
LP
123static uint64_t arg_keep_free = (uint64_t) -1;
124static uint64_t arg_max_use = (uint64_t) -1;
34c10968
LP
125
126static int parse_config(void) {
34c10968 127 static const ConfigTableItem items[] = {
8c9571d0
LP
128 { "Coredump", "Storage", config_parse_coredump_storage, 0, &arg_storage },
129 { "Coredump", "Compress", config_parse_bool, 0, &arg_compress },
59f448cf
LP
130 { "Coredump", "ProcessSizeMax", config_parse_iec_uint64, 0, &arg_process_size_max },
131 { "Coredump", "ExternalSizeMax", config_parse_iec_uint64, 0, &arg_external_size_max },
8c9571d0 132 { "Coredump", "JournalSizeMax", config_parse_iec_size, 0, &arg_journal_size_max },
59f448cf
LP
133 { "Coredump", "KeepFree", config_parse_iec_uint64, 0, &arg_keep_free },
134 { "Coredump", "MaxUse", config_parse_iec_uint64, 0, &arg_max_use },
34c10968
LP
135 {}
136 };
137
43688c49 138 return config_parse_many_nulstr(PKGSYSCONFDIR "/coredump.conf",
da412854
YW
139 CONF_PATHS_NULSTR("systemd/coredump.conf.d"),
140 "Coredump\0",
141 config_item_table_lookup, items,
bcde742e 142 CONFIG_PARSE_WARN, NULL);
34c10968
LP
143}
144
a1e92eee 145static uint64_t storage_size_max(void) {
ee0449fd
ZJS
146 if (arg_storage == COREDUMP_STORAGE_EXTERNAL)
147 return arg_external_size_max;
148 if (arg_storage == COREDUMP_STORAGE_JOURNAL)
149 return arg_journal_size_max;
150 assert(arg_storage == COREDUMP_STORAGE_NONE);
151 return 0;
73a99163
ZJS
152}
153
34c10968
LP
154static int fix_acl(int fd, uid_t uid) {
155
349cc4a5 156#if HAVE_ACL
34c10968
LP
157 _cleanup_(acl_freep) acl_t acl = NULL;
158 acl_entry_t entry;
159 acl_permset_t permset;
709f6e46 160 int r;
34c10968 161
b59233e6
LP
162 assert(fd >= 0);
163
05fd2156 164 if (uid_is_system(uid) || uid_is_dynamic(uid) || uid == UID_NOBODY)
34c10968
LP
165 return 0;
166
167 /* Make sure normal users can read (but not write or delete)
168 * their own coredumps */
169
170 acl = acl_get_fd(fd);
4a62c710
MS
171 if (!acl)
172 return log_error_errno(errno, "Failed to get ACL: %m");
34c10968
LP
173
174 if (acl_create_entry(&acl, &entry) < 0 ||
175 acl_set_tag_type(entry, ACL_USER) < 0 ||
d710aaf7
ZJS
176 acl_set_qualifier(entry, &uid) < 0)
177 return log_error_errno(errno, "Failed to patch ACL: %m");
34c10968
LP
178
179 if (acl_get_permset(entry, &permset) < 0 ||
709f6e46
MS
180 acl_add_perm(permset, ACL_READ) < 0)
181 return log_warning_errno(errno, "Failed to patch ACL: %m");
182
183 r = calc_acl_mask_if_needed(&acl);
184 if (r < 0)
185 return log_warning_errno(r, "Failed to patch ACL: %m");
34c10968 186
4a62c710
MS
187 if (acl_set_fd(fd, acl) < 0)
188 return log_error_errno(errno, "Failed to apply ACL: %m");
34c10968
LP
189#endif
190
191 return 0;
192}
193
3c171f0b 194static int fix_xattr(int fd, const char *context[_CONTEXT_MAX]) {
0cd77f97 195
3c171f0b
LP
196 static const char * const xattrs[_CONTEXT_MAX] = {
197 [CONTEXT_PID] = "user.coredump.pid",
198 [CONTEXT_UID] = "user.coredump.uid",
199 [CONTEXT_GID] = "user.coredump.gid",
200 [CONTEXT_SIGNAL] = "user.coredump.signal",
201 [CONTEXT_TIMESTAMP] = "user.coredump.timestamp",
6d337300 202 [CONTEXT_RLIMIT] = "user.coredump.rlimit",
f45b8015 203 [CONTEXT_HOSTNAME] = "user.coredump.hostname",
3c171f0b
LP
204 [CONTEXT_COMM] = "user.coredump.comm",
205 [CONTEXT_EXE] = "user.coredump.exe",
0cd77f97
LP
206 };
207
34c10968 208 int r = 0;
0cd77f97 209 unsigned i;
34c10968 210
b59233e6
LP
211 assert(fd >= 0);
212
1eef15b1 213 /* Attach some metadata to coredumps via extended
34c10968
LP
214 * attributes. Just because we can. */
215
3c171f0b 216 for (i = 0; i < _CONTEXT_MAX; i++) {
1eef15b1
ZJS
217 int k;
218
3c171f0b 219 if (isempty(context[i]) || !xattrs[i])
0cd77f97 220 continue;
34c10968 221
3c171f0b 222 k = fsetxattr(fd, xattrs[i], context[i], strlen(context[i]), XATTR_CREATE);
1eef15b1 223 if (k < 0 && r == 0)
34c10968 224 r = -errno;
0cd77f97 225 }
34c10968
LP
226
227 return r;
228}
229
b0b21dce 230#define filename_escape(s) xescape((s), "./ ")
34c10968 231
a1e92eee 232static const char *coredump_tmpfile_name(const char *s) {
0c773903
EV
233 return s ? s : "(unnamed temporary file)";
234}
235
b59233e6
LP
236static int fix_permissions(
237 int fd,
238 const char *filename,
239 const char *target,
3c171f0b 240 const char *context[_CONTEXT_MAX],
b59233e6
LP
241 uid_t uid) {
242
03532f0a
LP
243 int r;
244
b59233e6 245 assert(fd >= 0);
b59233e6 246 assert(target);
3c171f0b 247 assert(context);
cfd652ed
ZJS
248
249 /* Ignore errors on these */
3c171f0b
LP
250 (void) fchmod(fd, 0640);
251 (void) fix_acl(fd, uid);
252 (void) fix_xattr(fd, context);
cfd652ed 253
4a62c710 254 if (fsync(fd) < 0)
0c773903 255 return log_error_errno(errno, "Failed to sync coredump %s: %m", coredump_tmpfile_name(filename));
cfd652ed 256
8ac2f74f
LP
257 (void) fsync_directory_of_file(fd);
258
03532f0a
LP
259 r = link_tmpfile(fd, filename, target);
260 if (r < 0)
261 return log_error_errno(r, "Failed to move coredump %s into place: %m", target);
cfd652ed
ZJS
262
263 return 0;
264}
265
59f448cf 266static int maybe_remove_external_coredump(const char *filename, uint64_t size) {
cfd652ed 267
b59233e6 268 /* Returns 1 if might remove, 0 if will not remove, < 0 on error. */
cfd652ed 269
fc6cec86 270 if (arg_storage == COREDUMP_STORAGE_EXTERNAL &&
cfd652ed
ZJS
271 size <= arg_external_size_max)
272 return 0;
273
274 if (!filename)
275 return 1;
276
4a62c710
MS
277 if (unlink(filename) < 0 && errno != ENOENT)
278 return log_error_errno(errno, "Failed to unlink %s: %m", filename);
cfd652ed
ZJS
279
280 return 1;
281}
282
3c171f0b 283static int make_filename(const char *context[_CONTEXT_MAX], char **ret) {
b59233e6 284 _cleanup_free_ char *c = NULL, *u = NULL, *p = NULL, *t = NULL;
a7f7d1bd 285 sd_id128_t boot = {};
34c10968
LP
286 int r;
287
3c171f0b 288 assert(context);
34c10968 289
3c171f0b 290 c = filename_escape(context[CONTEXT_COMM]);
34c10968 291 if (!c)
b59233e6 292 return -ENOMEM;
34c10968 293
3c171f0b 294 u = filename_escape(context[CONTEXT_UID]);
0dc5d23c 295 if (!u)
b59233e6 296 return -ENOMEM;
34c10968
LP
297
298 r = sd_id128_get_boot(&boot);
b59233e6 299 if (r < 0)
34c10968 300 return r;
34c10968 301
3c171f0b 302 p = filename_escape(context[CONTEXT_PID]);
b59233e6
LP
303 if (!p)
304 return -ENOMEM;
305
3c171f0b 306 t = filename_escape(context[CONTEXT_TIMESTAMP]);
b59233e6
LP
307 if (!t)
308 return -ENOMEM;
309
310 if (asprintf(ret,
0dc5d23c 311 "/var/lib/systemd/coredump/core.%s.%s." SD_ID128_FORMAT_STR ".%s.%s000000",
34c10968 312 c,
0dc5d23c 313 u,
34c10968
LP
314 SD_ID128_FORMAT_VAL(boot),
315 p,
b59233e6
LP
316 t) < 0)
317 return -ENOMEM;
318
319 return 0;
320}
321
322static int save_external_coredump(
3c171f0b
LP
323 const char *context[_CONTEXT_MAX],
324 int input_fd,
b59233e6 325 char **ret_filename,
5f3e0a74
HW
326 int *ret_node_fd,
327 int *ret_data_fd,
0cd4e913 328 uint64_t *ret_size,
cc4419ed 329 bool *ret_truncated) {
b59233e6
LP
330
331 _cleanup_free_ char *fn = NULL, *tmp = NULL;
332 _cleanup_close_ int fd = -1;
ee0449fd 333 uint64_t rlimit, process_limit, max_size;
b59233e6 334 struct stat st;
3c171f0b 335 uid_t uid;
b59233e6
LP
336 int r;
337
3c171f0b 338 assert(context);
b59233e6 339 assert(ret_filename);
5f3e0a74
HW
340 assert(ret_node_fd);
341 assert(ret_data_fd);
b59233e6
LP
342 assert(ret_size);
343
3c171f0b
LP
344 r = parse_uid(context[CONTEXT_UID], &uid);
345 if (r < 0)
346 return log_error_errno(r, "Failed to parse UID: %m");
347
bdfd7b2c
LP
348 r = safe_atou64(context[CONTEXT_RLIMIT], &rlimit);
349 if (r < 0)
9d77ca39 350 return log_error_errno(r, "Failed to parse resource limit '%s': %m", context[CONTEXT_RLIMIT]);
6998b540
ZJS
351 if (rlimit < page_size()) {
352 /* Is coredumping disabled? Then don't bother saving/processing the coredump.
353 * Anything below PAGE_SIZE cannot give a readable coredump (the kernel uses
354 * ELF_EXEC_PAGESIZE which is not easily accessible, but is usually the same as PAGE_SIZE. */
baaa35ad
ZJS
355 return log_info_errno(SYNTHETIC_ERRNO(EBADSLT),
356 "Resource limits disable core dumping for process %s (%s).",
357 context[CONTEXT_PID], context[CONTEXT_COMM]);
bdfd7b2c
LP
358 }
359
ee0449fd 360 process_limit = MAX(arg_process_size_max, storage_size_max());
baaa35ad
ZJS
361 if (process_limit == 0)
362 return log_debug_errno(SYNTHETIC_ERRNO(EBADSLT),
363 "Limits for coredump processing and storage are both 0, not dumping core.");
ee0449fd 364
bdfd7b2c 365 /* Never store more than the process configured, or than we actually shall keep or process */
ee0449fd 366 max_size = MIN(rlimit, process_limit);
bdfd7b2c 367
3c171f0b 368 r = make_filename(context, &fn);
23bbb0de
MS
369 if (r < 0)
370 return log_error_errno(r, "Failed to determine coredump file name: %m");
34c10968 371
d2e54fae 372 mkdir_p_label("/var/lib/systemd/coredump", 0755);
803a3464 373
03532f0a 374 fd = open_tmpfile_linkable(fn, O_RDWR|O_CLOEXEC, &tmp);
4a62c710 375 if (fd < 0)
03532f0a 376 return log_error_errno(fd, "Failed to create temporary file for coredump %s: %m", fn);
803a3464 377
1c876927 378 r = copy_bytes(input_fd, fd, max_size, 0);
73a99163
ZJS
379 if (r < 0) {
380 log_error_errno(r, "Cannot store coredump of %s (%s): %m", context[CONTEXT_PID], context[CONTEXT_COMM]);
93240d3a 381 goto fail;
0cd4e913 382 }
cc4419ed
ZJS
383 *ret_truncated = r == 1;
384 if (*ret_truncated)
73a99163
ZJS
385 log_struct(LOG_INFO,
386 LOG_MESSAGE("Core file was truncated to %zu bytes.", max_size),
387 "SIZE_LIMIT=%zu", max_size,
a1230ff9 388 "MESSAGE_ID=" SD_MESSAGE_TRUNCATED_CORE_STR);
803a3464 389
34c10968 390 if (fstat(fd, &st) < 0) {
73a99163 391 log_error_errno(errno, "Failed to fstat core file %s: %m", coredump_tmpfile_name(tmp));
34c10968
LP
392 goto fail;
393 }
394
7849c2ac 395 if (lseek(fd, 0, SEEK_SET) == (off_t) -1) {
0c773903 396 log_error_errno(errno, "Failed to seek on %s: %m", coredump_tmpfile_name(tmp));
b59233e6 397 goto fail;
7849c2ac
TA
398 }
399
349cc4a5 400#if HAVE_XZ || HAVE_LZ4
cfd652ed 401 /* If we will remove the coredump anyway, do not compress. */
6e9ef603 402 if (arg_compress && !maybe_remove_external_coredump(NULL, st.st_size)) {
cfd652ed 403
b59233e6
LP
404 _cleanup_free_ char *fn_compressed = NULL, *tmp_compressed = NULL;
405 _cleanup_close_ int fd_compressed = -1;
cfd652ed 406
d89c8fdf 407 fn_compressed = strappend(fn, COMPRESSED_EXT);
b59233e6 408 if (!fn_compressed) {
d89c8fdf 409 log_oom();
cfd652ed
ZJS
410 goto uncompressed;
411 }
412
03532f0a
LP
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);
b59233e6 416 goto uncompressed;
03532f0a 417 }
cfd652ed 418
d89c8fdf 419 r = compress_stream(fd, fd_compressed, -1);
b59233e6 420 if (r < 0) {
0c773903 421 log_error_errno(r, "Failed to compress %s: %m", coredump_tmpfile_name(tmp_compressed));
b59233e6
LP
422 goto fail_compressed;
423 }
424
3c171f0b 425 r = fix_permissions(fd_compressed, tmp_compressed, fn_compressed, context, uid);
cfd652ed 426 if (r < 0)
b59233e6
LP
427 goto fail_compressed;
428
429 /* OK, this worked, we can get rid of the uncompressed version now */
0c773903
EV
430 if (tmp)
431 unlink_noerrno(tmp);
cfd652ed 432
1cc6c93a
YW
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 */
59f448cf 436 *ret_size = (uint64_t) st.st_size; /* uncompressed */
cfd652ed 437
cfd652ed
ZJS
438 return 0;
439
b59233e6 440 fail_compressed:
0c773903
EV
441 if (tmp_compressed)
442 (void) unlink(tmp_compressed);
34c10968 443 }
cfd652ed
ZJS
444
445uncompressed:
3b1a55e1 446#endif
5f3e0a74 447
3c171f0b 448 r = fix_permissions(fd, tmp, fn, context, uid);
cfd652ed
ZJS
449 if (r < 0)
450 goto fail;
34c10968 451
1cc6c93a
YW
452 *ret_filename = TAKE_PTR(fn);
453 *ret_data_fd = TAKE_FD(fd);
5f3e0a74 454 *ret_node_fd = -1;
59f448cf 455 *ret_size = (uint64_t) st.st_size;
34c10968 456
34c10968
LP
457 return 0;
458
459fail:
0c773903
EV
460 if (tmp)
461 (void) unlink(tmp);
34c10968
LP
462 return r;
463}
464
465static 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
8d4e028f 469 assert(fd >= 0);
34c10968
LP
470 assert(ret);
471 assert(ret_size);
472
4a62c710
MS
473 if (lseek(fd, 0, SEEK_SET) == (off_t) -1)
474 return log_warning_errno(errno, "Failed to seek: %m");
803a3464 475
34c10968
LP
476 field = malloc(9 + size);
477 if (!field) {
cfd652ed 478 log_warning("Failed to allocate memory for coredump, coredump will not be stored.");
34c10968
LP
479 return -ENOMEM;
480 }
481
482 memcpy(field, "COREDUMP=", 9);
483
484 n = read(fd, field + 9, size);
23bbb0de
MS
485 if (n < 0)
486 return log_error_errno((int) n, "Failed to read core data: %m");
baaa35ad
ZJS
487 if ((size_t) n < size)
488 return log_error_errno(SYNTHETIC_ERRNO(EIO),
489 "Core data too short.");
34c10968 490
1cc6c93a 491 *ret = TAKE_PTR(field);
34c10968
LP
492 *ret_size = size + 9;
493
34c10968
LP
494 return 0;
495}
803a3464 496
3f132692
JF
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 */
511static int compose_open_fds(pid_t pid, char **open_fds) {
4d84bc2f
LP
512 _cleanup_closedir_ DIR *proc_fd_dir = NULL;
513 _cleanup_close_ int proc_fdinfo_fd = -1;
514 _cleanup_free_ char *buffer = NULL;
3f132692 515 _cleanup_fclose_ FILE *stream = NULL;
59059b4a 516 const char *fddelim = "", *path;
3f132692 517 struct dirent *dent = NULL;
4d84bc2f 518 size_t size = 0;
7b26ea6f 519 int r;
3f132692
JF
520
521 assert(pid >= 0);
522 assert(open_fds != NULL);
523
59059b4a 524 path = procfs_file_alloca(pid, "fd");
3f132692 525 proc_fd_dir = opendir(path);
59059b4a
ZJS
526 if (!proc_fd_dir)
527 return -errno;
3f132692 528
4d84bc2f 529 proc_fdinfo_fd = openat(dirfd(proc_fd_dir), "../fdinfo", O_DIRECTORY|O_NOFOLLOW|O_CLOEXEC|O_PATH);
59059b4a
ZJS
530 if (proc_fdinfo_fd < 0)
531 return -errno;
3f132692 532
4d84bc2f 533 stream = open_memstream(&buffer, &size);
3f132692
JF
534 if (!stream)
535 return -ENOMEM;
536
0d536673
LP
537 (void) __fsetlocking(stream, FSETLOCKING_BYCALLER);
538
4d84bc2f 539 FOREACH_DIRENT(dent, proc_fd_dir, return -errno) {
3f132692 540 _cleanup_fclose_ FILE *fdinfo = NULL;
4d84bc2f 541 _cleanup_free_ char *fdname = NULL;
4d84bc2f 542 int fd;
3f132692 543
59059b4a 544 r = readlinkat_malloc(dirfd(proc_fd_dir), dent->d_name, &fdname);
3f132692
JF
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 */
59059b4a
ZJS
552 fd = openat(proc_fdinfo_fd, dent->d_name, O_NOFOLLOW|O_CLOEXEC|O_RDONLY);
553 if (fd < 0)
3f132692
JF
554 continue;
555
e92aaed3 556 fdinfo = fdopen(fd, "r");
234519ae 557 if (!fdinfo) {
0d536673 558 safe_close(fd);
3f132692 559 continue;
59059b4a 560 }
3f132692 561
7b26ea6f
LP
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
0d536673 571 fputs(line, stream);
7b26ea6f 572 fputc('\n', stream);
4d84bc2f 573 }
3f132692
JF
574 }
575
4d84bc2f 576 errno = 0;
74ca738f 577 stream = safe_fclose(stream);
4d84bc2f 578
b3267152 579 if (errno > 0)
4d84bc2f
LP
580 return -errno;
581
ae2a15bc 582 *open_fds = TAKE_PTR(buffer);
4d84bc2f 583
3f132692
JF
584 return 0;
585}
586
7ed03ce6
JF
587static 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
605static 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
aa7530d6 614 for (;;) {
7ed03ce6
JF
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 */
643static 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
d3cba4ea
EV
667 r = get_process_cmdline(container_pid, 0, false, cmdline);
668 if (r < 0)
669 return r;
670
671 return 1;
7ed03ce6
JF
672}
673
3c171f0b
LP
674static int change_uid_gid(const char *context[]) {
675 uid_t uid;
676 gid_t gid;
677 int r;
34c10968 678
3c171f0b
LP
679 r = parse_uid(context[CONTEXT_UID], &uid);
680 if (r < 0)
681 return r;
8c8549db 682
888e378d
LP
683 if (uid <= SYSTEM_UID_MAX) {
684 const char *user = "systemd-coredump";
685
fafff8f1 686 r = get_user_creds(&user, &uid, &gid, NULL, NULL, 0);
888e378d
LP
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 }
3c171f0b
LP
696
697 return drop_privileges(uid, gid, 0);
698}
8c8549db 699
b8cda92d
LP
700static 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
706static 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
0cd4e913
ZJS
713#define SUBMIT_COREDUMP_FIELDS 4
714
3c171f0b
LP
715static 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) {
34c10968 721
5f3e0a74 722 _cleanup_close_ int coredump_fd = -1, coredump_node_fd = -1;
3c171f0b 723 _cleanup_free_ char *core_message = NULL, *filename = NULL, *coredump_data = NULL;
a5ca3649 724 uint64_t coredump_size = UINT64_MAX;
92e92d71 725 bool truncated = false, journald_crash;
3c171f0b 726 int r;
f5e04665 727
3c171f0b
LP
728 assert(context);
729 assert(iovec);
0cd4e913 730 assert(n_iovec_allocated >= n_iovec + SUBMIT_COREDUMP_FIELDS);
3c171f0b 731 assert(input_fd >= 0);
f5e04665 732
b8cda92d 733 journald_crash = is_journald_crash(context);
92e92d71 734
3c171f0b
LP
735 /* Vacuum before we write anything again */
736 (void) coredump_vacuum(-1, arg_keep_free, arg_max_use);
803a3464 737
3c171f0b 738 /* Always stream the coredump to disk, if that's possible */
0cd4e913
ZJS
739 r = save_external_coredump(context, input_fd,
740 &filename, &coredump_node_fd, &coredump_fd, &coredump_size, &truncated);
3c171f0b
LP
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);
e6a7ec4b 754 iovec[n_iovec++] = IOVEC_MAKE_STRING(coredump_filename);
6e9ef603 755 } else if (arg_storage == COREDUMP_STORAGE_EXTERNAL)
5206a724 756 log_info("The core will not be stored: size %"PRIu64" is greater than %"PRIu64" (the configured maximum)",
6e9ef603 757 coredump_size, arg_external_size_max);
f5e04665 758
3c171f0b
LP
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);
8c9571d0 761
3c171f0b
LP
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");
34c10968 769
349cc4a5 770#if HAVE_ELFUTILS
3c171f0b
LP
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)
605405c6
ZJS
777 core_message = strjoin("MESSAGE=Process ", context[CONTEXT_PID],
778 " (", context[CONTEXT_COMM], ") of user ",
92e92d71
ZJS
779 context[CONTEXT_UID], " dumped core.",
780 journald_crash ? "\nCoredump diverted to " : "",
781 journald_crash ? filename : "",
782 "\n\n", stacktrace);
3c171f0b
LP
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");
6e9ef603 787 } else
5206a724 788 log_debug("Not generating stack trace: core size %"PRIu64" is greater than %"PRIu64" (the configured maximum)",
6e9ef603 789 coredump_size, arg_process_size_max);
803a3464 790
3c171f0b
LP
791 if (!core_message)
792#endif
793log:
92e92d71
ZJS
794 core_message = strjoin("MESSAGE=Process ", context[CONTEXT_PID],
795 " (", context[CONTEXT_COMM], ") of user ",
796 context[CONTEXT_UID], " dumped core.",
f0136e09
ZJS
797 journald_crash && filename ? "\nCoredump diverted to " : NULL,
798 journald_crash && filename ? filename : NULL);
92e92d71
ZJS
799 if (!core_message)
800 return log_oom();
801
802 if (journald_crash) {
4f62556d 803 /* We cannot log to the journal, so just print the message.
92e92d71 804 * The target was set previously to something safe. */
4f62556d
ZJS
805 assert(startswith(core_message, "MESSAGE="));
806 log_dispatch(LOG_ERR, 0, core_message + strlen("MESSAGE="));
92e92d71
ZJS
807 return 0;
808 }
809
e6a7ec4b 810 iovec[n_iovec++] = IOVEC_MAKE_STRING(core_message);
3c171f0b 811
0cd4e913 812 if (truncated)
e6a7ec4b 813 iovec[n_iovec++] = IOVEC_MAKE_STRING("COREDUMP_TRUNCATED=1");
0cd4e913 814
3c171f0b 815 /* Optionally store the entire coredump in the journal */
6e9ef603
ZJS
816 if (arg_storage == COREDUMP_STORAGE_JOURNAL) {
817 if (coredump_size <= arg_journal_size_max) {
818 size_t sz = 0;
819
820 /* Store the coredump itself in the journal */
821
822 r = allocate_journal_field(coredump_fd, (size_t) coredump_size, &coredump_data, &sz);
e6a7ec4b
LP
823 if (r >= 0)
824 iovec[n_iovec++] = IOVEC_MAKE(coredump_data, sz);
825 else
6e9ef603
ZJS
826 log_warning_errno(r, "Failed to attach the core to the journal entry: %m");
827 } else
5206a724 828 log_info("The core will not be stored: size %"PRIu64" is greater than %"PRIu64" (the configured maximum)",
6e9ef603 829 coredump_size, arg_journal_size_max);
f5e04665
LP
830 }
831
3c171f0b
LP
832 assert(n_iovec <= n_iovec_allocated);
833
834 r = sd_journal_sendv(iovec, n_iovec);
835 if (r < 0)
836 return log_error_errno(r, "Failed to log coredump: %m");
837
838 return 0;
839}
840
ea5cc2a8 841static void map_context_fields(const struct iovec *iovec, const char* context[]) {
3c171f0b 842
92e92d71 843 static const char * const context_field_names[] = {
3c171f0b
LP
844 [CONTEXT_PID] = "COREDUMP_PID=",
845 [CONTEXT_UID] = "COREDUMP_UID=",
846 [CONTEXT_GID] = "COREDUMP_GID=",
847 [CONTEXT_SIGNAL] = "COREDUMP_SIGNAL=",
848 [CONTEXT_TIMESTAMP] = "COREDUMP_TIMESTAMP=",
92e92d71 849 [CONTEXT_RLIMIT] = "COREDUMP_RLIMIT=",
f45b8015 850 [CONTEXT_HOSTNAME] = "COREDUMP_HOSTNAME=",
3c171f0b
LP
851 [CONTEXT_COMM] = "COREDUMP_COMM=",
852 [CONTEXT_EXE] = "COREDUMP_EXE=",
853 };
854
855 unsigned i;
856
857 assert(iovec);
858 assert(context);
859
92e92d71 860 for (i = 0; i < ELEMENTSOF(context_field_names); i++) {
d27b725a 861 char *p;
3c171f0b 862
92e92d71
ZJS
863 if (!context_field_names[i])
864 continue;
865
d27b725a
LP
866 p = memory_startswith(iovec->iov_base, iovec->iov_len, context_field_names[i]);
867 if (!p)
3c171f0b
LP
868 continue;
869
870 /* Note that these strings are NUL terminated, because we made sure that a trailing NUL byte is in the
871 * buffer, though not included in the iov_len count. (see below) */
d27b725a 872 context[i] = p;
3c171f0b
LP
873 break;
874 }
875}
876
877static int process_socket(int fd) {
878 _cleanup_close_ int coredump_fd = -1;
879 struct iovec *iovec = NULL;
80002f66 880 size_t n_iovec = 0, n_allocated = 0, i, k;
3c171f0b
LP
881 const char *context[_CONTEXT_MAX] = {};
882 int r;
883
884 assert(fd >= 0);
885
6bf3c61c 886 log_setup_service();
3c171f0b 887
988e89ee
ZJS
888 log_debug("Processing coredump received on stdin...");
889
3c171f0b
LP
890 for (;;) {
891 union {
892 struct cmsghdr cmsghdr;
893 uint8_t buf[CMSG_SPACE(sizeof(int))];
894 } control = {};
895 struct msghdr mh = {
896 .msg_control = &control,
897 .msg_controllen = sizeof(control),
898 .msg_iovlen = 1,
899 };
900 ssize_t n;
fe1ef0f8 901 ssize_t l;
3c171f0b 902
0cd4e913 903 if (!GREEDY_REALLOC(iovec, n_allocated, n_iovec + SUBMIT_COREDUMP_FIELDS)) {
3c171f0b
LP
904 r = log_oom();
905 goto finish;
906 }
907
fe1ef0f8
EV
908 l = next_datagram_size_fd(fd);
909 if (l < 0) {
910 r = log_error_errno(l, "Failed to determine datagram size to read: %m");
3c171f0b
LP
911 goto finish;
912 }
913
914 assert(l >= 0);
915
916 iovec[n_iovec].iov_len = l;
917 iovec[n_iovec].iov_base = malloc(l + 1);
3c171f0b
LP
918 if (!iovec[n_iovec].iov_base) {
919 r = log_oom();
920 goto finish;
921 }
922
923 mh.msg_iov = iovec + n_iovec;
924
a6887cc0 925 n = recvmsg(fd, &mh, MSG_CMSG_CLOEXEC);
3c171f0b
LP
926 if (n < 0) {
927 free(iovec[n_iovec].iov_base);
928 r = log_error_errno(errno, "Failed to receive datagram: %m");
929 goto finish;
930 }
931
932 if (n == 0) {
933 struct cmsghdr *cmsg, *found = NULL;
934 /* The final zero-length datagram carries the file descriptor and tells us that we're done. */
935
936 free(iovec[n_iovec].iov_base);
937
938 CMSG_FOREACH(cmsg, &mh) {
939 if (cmsg->cmsg_level == SOL_SOCKET &&
940 cmsg->cmsg_type == SCM_RIGHTS &&
941 cmsg->cmsg_len == CMSG_LEN(sizeof(int))) {
942 assert(!found);
943 found = cmsg;
944 }
945 }
946
947 if (!found) {
948 log_error("Coredump file descriptor missing.");
949 r = -EBADMSG;
950 goto finish;
951 }
952
953 assert(coredump_fd < 0);
954 coredump_fd = *(int*) CMSG_DATA(found);
955 break;
956 }
957
958 /* Add trailing NUL byte, in case these are strings */
959 ((char*) iovec[n_iovec].iov_base)[n] = 0;
960 iovec[n_iovec].iov_len = (size_t) n;
961
962 cmsg_close_all(&mh);
963 map_context_fields(iovec + n_iovec, context);
964 n_iovec++;
965 }
966
0cd4e913 967 if (!GREEDY_REALLOC(iovec, n_allocated, n_iovec + SUBMIT_COREDUMP_FIELDS)) {
3c171f0b 968 r = log_oom();
34c10968
LP
969 goto finish;
970 }
971
61233823 972 /* Make sure we got all data we really need */
3c171f0b
LP
973 assert(context[CONTEXT_PID]);
974 assert(context[CONTEXT_UID]);
975 assert(context[CONTEXT_GID]);
976 assert(context[CONTEXT_SIGNAL]);
977 assert(context[CONTEXT_TIMESTAMP]);
bdfd7b2c 978 assert(context[CONTEXT_RLIMIT]);
f45b8015 979 assert(context[CONTEXT_HOSTNAME]);
3c171f0b
LP
980 assert(context[CONTEXT_COMM]);
981 assert(coredump_fd >= 0);
982
80002f66
LP
983 /* Small quirk: the journal fields contain the timestamp padded with six zeroes, so that the kernel-supplied 1s
984 * granularity timestamps becomes 1µs granularity, i.e. the granularity systemd usually operates in. Since we
985 * are reconstructing the original kernel context, we chop this off again, here. */
986 k = strlen(context[CONTEXT_TIMESTAMP]);
987 if (k > 6)
988 context[CONTEXT_TIMESTAMP] = strndupa(context[CONTEXT_TIMESTAMP], k - 6);
989
5b45a160 990 r = submit_coredump(context, iovec, n_allocated, n_iovec, coredump_fd);
3c171f0b
LP
991
992finish:
993 for (i = 0; i < n_iovec; i++)
994 free(iovec[i].iov_base);
995 free(iovec);
996
997 return r;
998}
999
1000static int send_iovec(const struct iovec iovec[], size_t n_iovec, int input_fd) {
1001
1002 static const union sockaddr_union sa = {
1003 .un.sun_family = AF_UNIX,
1004 .un.sun_path = "/run/systemd/coredump",
1005 };
1006 _cleanup_close_ int fd = -1;
1007 size_t i;
1008 int r;
1009
1010 assert(iovec || n_iovec <= 0);
1011 assert(input_fd >= 0);
1012
1013 fd = socket(AF_UNIX, SOCK_SEQPACKET|SOCK_CLOEXEC, 0);
1014 if (fd < 0)
1015 return log_error_errno(errno, "Failed to create coredump socket: %m");
1016
fc2fffe7 1017 if (connect(fd, &sa.sa, SOCKADDR_UN_LEN(sa.un)) < 0)
3c171f0b
LP
1018 return log_error_errno(errno, "Failed to connect to coredump service: %m");
1019
1020 for (i = 0; i < n_iovec; i++) {
fec603eb
LP
1021 struct msghdr mh = {
1022 .msg_iov = (struct iovec*) iovec + i,
1023 .msg_iovlen = 1,
1024 };
1025 struct iovec copy[2];
1026
1027 for (;;) {
1028 if (sendmsg(fd, &mh, MSG_NOSIGNAL) >= 0)
1029 break;
1030
1031 if (errno == EMSGSIZE && mh.msg_iov[0].iov_len > 0) {
1032 /* This field didn't fit? That's a pity. Given that this is just metadata,
1033 * let's truncate the field at half, and try again. We append three dots, in
1034 * order to show that this is truncated. */
1035
1036 if (mh.msg_iov != copy) {
1037 /* We don't want to modify the caller's iovec, hence let's create our
1038 * own array, consisting of two new iovecs, where the first is a
1039 * (truncated) copy of what we want to send, and the second one
1040 * contains the trailing dots. */
1041 copy[0] = iovec[i];
1042 copy[1] = (struct iovec) {
1043 .iov_base = (char[]) { '.', '.', '.' },
1044 .iov_len = 3,
1045 };
1046
1047 mh.msg_iov = copy;
1048 mh.msg_iovlen = 2;
1049 }
1050
1051 copy[0].iov_len /= 2; /* halve it, and try again */
1052 continue;
1053 }
3c171f0b 1054
3c171f0b 1055 return log_error_errno(errno, "Failed to send coredump datagram: %m");
fec603eb 1056 }
1eef15b1
ZJS
1057 }
1058
3c171f0b
LP
1059 r = send_one_fd(fd, input_fd, 0);
1060 if (r < 0)
1061 return log_error_errno(r, "Failed to send coredump fd: %m");
1eef15b1 1062
3c171f0b
LP
1063 return 0;
1064}
1eef15b1 1065
f45b8015 1066static char* set_iovec_field_free(struct iovec *iovec, size_t *n_iovec, const char *field, char *value) {
9aa82023
ZJS
1067 char *x;
1068
084eeb86 1069 x = set_iovec_string_field(iovec, n_iovec, field, value);
9aa82023
ZJS
1070 free(value);
1071 return x;
1072}
1073
92e92d71 1074static int gather_pid_metadata(
ea5cc2a8 1075 char* context[_CONTEXT_MAX],
9aa82023 1076 char **comm_fallback,
5b45a160 1077 struct iovec *iovec, size_t *n_iovec) {
d14bcb4e 1078
f45b8015 1079 /* We need 27 empty slots in iovec!
d14bcb4e
LP
1080 *
1081 * Note that if we fail on oom later on, we do not roll-back changes to the iovec structure. (It remains valid,
1082 * with the first n_iovec fields initialized.) */
3c171f0b 1083
3c171f0b 1084 uid_t owner_uid;
3c171f0b
LP
1085 pid_t pid;
1086 char *t;
9aa82023 1087 const char *p;
d14bcb4e 1088 int r, signo;
3c171f0b 1089
9aa82023 1090 r = parse_pid(context[CONTEXT_PID], &pid);
3c171f0b 1091 if (r < 0)
988e89ee 1092 return log_error_errno(r, "Failed to parse PID \"%s\": %m", context[CONTEXT_PID]);
3c171f0b 1093
ea5cc2a8 1094 r = get_process_comm(pid, &context[CONTEXT_COMM]);
3c171f0b
LP
1095 if (r < 0) {
1096 log_warning_errno(r, "Failed to get COMM, falling back to the command line: %m");
ea5cc2a8
ZJS
1097 context[CONTEXT_COMM] = strv_join(comm_fallback, " ");
1098 if (!context[CONTEXT_COMM])
3c171f0b
LP
1099 return log_oom();
1100 }
1101
ea5cc2a8 1102 r = get_process_exe(pid, &context[CONTEXT_EXE]);
3c171f0b
LP
1103 if (r < 0)
1104 log_warning_errno(r, "Failed to get EXE, ignoring: %m");
1105
92e92d71 1106 if (cg_pid_get_unit(pid, &context[CONTEXT_UNIT]) >= 0) {
b8cda92d 1107 if (!is_journald_crash((const char**) context)) {
92e92d71
ZJS
1108 /* OK, now we know it's not the journal, hence we can make use of it now. */
1109 log_set_target(LOG_TARGET_JOURNAL_OR_KMSG);
1110 log_open();
1111 }
3c171f0b 1112
c8091d92 1113 /* If this is PID 1 disable coredump collection, we'll unlikely be able to process it later on. */
b8cda92d 1114 if (is_pid1_crash((const char**) context)) {
c8091d92 1115 log_notice("Due to PID 1 having crashed coredump collection will now be turned off.");
e557b1a6 1116 disable_coredumps();
c8091d92
LP
1117 }
1118
084eeb86 1119 set_iovec_string_field(iovec, n_iovec, "COREDUMP_UNIT=", context[CONTEXT_UNIT]);
8c8549db 1120 }
803a3464 1121
9aa82023
ZJS
1122 if (cg_pid_get_user_unit(pid, &t) >= 0)
1123 set_iovec_field_free(iovec, n_iovec, "COREDUMP_USER_UNIT=", t);
3c171f0b 1124
9aa82023 1125 /* The next few are mandatory */
084eeb86 1126 if (!set_iovec_string_field(iovec, n_iovec, "COREDUMP_PID=", context[CONTEXT_PID]))
9aa82023 1127 return log_oom();
3c171f0b 1128
084eeb86 1129 if (!set_iovec_string_field(iovec, n_iovec, "COREDUMP_UID=", context[CONTEXT_UID]))
9aa82023 1130 return log_oom();
f5e04665 1131
084eeb86 1132 if (!set_iovec_string_field(iovec, n_iovec, "COREDUMP_GID=", context[CONTEXT_GID]))
9aa82023 1133 return log_oom();
f5e04665 1134
084eeb86 1135 if (!set_iovec_string_field(iovec, n_iovec, "COREDUMP_SIGNAL=", context[CONTEXT_SIGNAL]))
9aa82023 1136 return log_oom();
f5e04665 1137
084eeb86 1138 if (!set_iovec_string_field(iovec, n_iovec, "COREDUMP_RLIMIT=", context[CONTEXT_RLIMIT]))
9aa82023 1139 return log_oom();
f5e04665 1140
084eeb86 1141 if (!set_iovec_string_field(iovec, n_iovec, "COREDUMP_HOSTNAME=", context[CONTEXT_HOSTNAME]))
f45b8015
JF
1142 return log_oom();
1143
084eeb86 1144 if (!set_iovec_string_field(iovec, n_iovec, "COREDUMP_COMM=", context[CONTEXT_COMM]))
9aa82023 1145 return log_oom();
bdfd7b2c 1146
ea5cc2a8 1147 if (context[CONTEXT_EXE] &&
084eeb86 1148 !set_iovec_string_field(iovec, n_iovec, "COREDUMP_EXE=", context[CONTEXT_EXE]))
9aa82023 1149 return log_oom();
f5e04665 1150
9aa82023
ZJS
1151 if (sd_pid_get_session(pid, &t) >= 0)
1152 set_iovec_field_free(iovec, n_iovec, "COREDUMP_SESSION=", t);
f5e04665 1153
a035f819 1154 if (sd_pid_get_owner_uid(pid, &owner_uid) >= 0) {
9aa82023 1155 r = asprintf(&t, "COREDUMP_OWNER_UID=" UID_FMT, owner_uid);
7de80bfe 1156 if (r > 0)
e6a7ec4b 1157 iovec[(*n_iovec)++] = IOVEC_MAKE_STRING(t);
f5e04665
LP
1158 }
1159
9aa82023
ZJS
1160 if (sd_pid_get_slice(pid, &t) >= 0)
1161 set_iovec_field_free(iovec, n_iovec, "COREDUMP_SLICE=", t);
f5e04665 1162
9aa82023
ZJS
1163 if (get_process_cmdline(pid, 0, false, &t) >= 0)
1164 set_iovec_field_free(iovec, n_iovec, "COREDUMP_CMDLINE=", t);
a035f819 1165
9aa82023
ZJS
1166 if (cg_pid_get_path_shifted(pid, NULL, &t) >= 0)
1167 set_iovec_field_free(iovec, n_iovec, "COREDUMP_CGROUP=", t);
a035f819 1168
9aa82023
ZJS
1169 if (compose_open_fds(pid, &t) >= 0)
1170 set_iovec_field_free(iovec, n_iovec, "COREDUMP_OPEN_FDS=", t);
3f132692
JF
1171
1172 p = procfs_file_alloca(pid, "status");
9aa82023
ZJS
1173 if (read_full_file(p, &t, NULL) >= 0)
1174 set_iovec_field_free(iovec, n_iovec, "COREDUMP_PROC_STATUS=", t);
3f132692
JF
1175
1176 p = procfs_file_alloca(pid, "maps");
9aa82023
ZJS
1177 if (read_full_file(p, &t, NULL) >= 0)
1178 set_iovec_field_free(iovec, n_iovec, "COREDUMP_PROC_MAPS=", t);
3f132692
JF
1179
1180 p = procfs_file_alloca(pid, "limits");
9aa82023
ZJS
1181 if (read_full_file(p, &t, NULL) >= 0)
1182 set_iovec_field_free(iovec, n_iovec, "COREDUMP_PROC_LIMITS=", t);
3f132692
JF
1183
1184 p = procfs_file_alloca(pid, "cgroup");
9aa82023
ZJS
1185 if (read_full_file(p, &t, NULL) >=0)
1186 set_iovec_field_free(iovec, n_iovec, "COREDUMP_PROC_CGROUP=", t);
3f132692 1187
d7032b1f 1188 p = procfs_file_alloca(pid, "mountinfo");
9aa82023
ZJS
1189 if (read_full_file(p, &t, NULL) >=0)
1190 set_iovec_field_free(iovec, n_iovec, "COREDUMP_PROC_MOUNTINFO=", t);
d7032b1f 1191
9aa82023
ZJS
1192 if (get_process_cwd(pid, &t) >= 0)
1193 set_iovec_field_free(iovec, n_iovec, "COREDUMP_CWD=", t);
3f132692
JF
1194
1195 if (get_process_root(pid, &t) >= 0) {
9aa82023
ZJS
1196 bool proc_self_root_is_slash;
1197
1198 proc_self_root_is_slash = strcmp(t, "/") == 0;
3f132692 1199
9aa82023 1200 set_iovec_field_free(iovec, n_iovec, "COREDUMP_ROOT=", t);
7ed03ce6
JF
1201
1202 /* If the process' root is "/", then there is a chance it has
1203 * mounted own root and hence being containerized. */
9aa82023
ZJS
1204 if (proc_self_root_is_slash && get_process_container_parent_cmdline(pid, &t) > 0)
1205 set_iovec_field_free(iovec, n_iovec, "COREDUMP_CONTAINER_CMDLINE=", t);
3f132692
JF
1206 }
1207
9aa82023
ZJS
1208 if (get_process_environ(pid, &t) >= 0)
1209 set_iovec_field_free(iovec, n_iovec, "COREDUMP_ENVIRON=", t);
1210
4600a396 1211 t = strjoin("COREDUMP_TIMESTAMP=", context[CONTEXT_TIMESTAMP], "000000");
9aa82023 1212 if (t)
e6a7ec4b 1213 iovec[(*n_iovec)++] = IOVEC_MAKE_STRING(t);
9aa82023 1214
d14bcb4e 1215 if (safe_atoi(context[CONTEXT_SIGNAL], &signo) >= 0 && SIGNAL_VALID(signo))
084eeb86 1216 set_iovec_string_field(iovec, n_iovec, "COREDUMP_SIGNAL_NAME=SIG", signal_to_string(signo));
d14bcb4e 1217
92e92d71 1218 return 0; /* we successfully acquired all metadata */
9aa82023 1219}
3f132692 1220
9aa82023
ZJS
1221static int process_kernel(int argc, char* argv[]) {
1222
ea5cc2a8 1223 char* context[_CONTEXT_MAX] = {};
f45b8015 1224 struct iovec iovec[29 + SUBMIT_COREDUMP_FIELDS];
988e89ee 1225 size_t i, n_iovec, n_to_free = 0;
9aa82023
ZJS
1226 int r;
1227
988e89ee
ZJS
1228 log_debug("Processing coredump received from the kernel...");
1229
baaa35ad
ZJS
1230 if (argc < CONTEXT_COMM + 1)
1231 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
1232 "Not enough arguments passed by the kernel (%i, expected %i).",
1233 argc - 1, CONTEXT_COMM + 1 - 1);
3f132692 1234
ea5cc2a8
ZJS
1235 context[CONTEXT_PID] = argv[1 + CONTEXT_PID];
1236 context[CONTEXT_UID] = argv[1 + CONTEXT_UID];
1237 context[CONTEXT_GID] = argv[1 + CONTEXT_GID];
1238 context[CONTEXT_SIGNAL] = argv[1 + CONTEXT_SIGNAL];
1239 context[CONTEXT_TIMESTAMP] = argv[1 + CONTEXT_TIMESTAMP];
1240 context[CONTEXT_RLIMIT] = argv[1 + CONTEXT_RLIMIT];
f45b8015 1241 context[CONTEXT_HOSTNAME] = argv[1 + CONTEXT_HOSTNAME];
9aa82023 1242
92e92d71
ZJS
1243 r = gather_pid_metadata(context, argv + 1 + CONTEXT_COMM, iovec, &n_to_free);
1244 if (r < 0)
86562420 1245 goto finish;
86562420 1246
988e89ee 1247 n_iovec = n_to_free;
f5e04665 1248
e6a7ec4b 1249 iovec[n_iovec++] = IOVEC_MAKE_STRING("MESSAGE_ID=" SD_MESSAGE_COREDUMP_STR);
4850d39a
LP
1250
1251 assert_cc(2 == LOG_CRIT);
e6a7ec4b 1252 iovec[n_iovec++] = IOVEC_MAKE_STRING("PRIORITY=2");
0dc5d23c 1253
3c171f0b 1254 assert(n_iovec <= ELEMENTSOF(iovec));
34c10968 1255
b8cda92d 1256 if (is_journald_crash((const char**) context) || is_pid1_crash((const char**) context))
92e92d71
ZJS
1257 r = submit_coredump((const char**) context,
1258 iovec, ELEMENTSOF(iovec), n_iovec,
1259 STDIN_FILENO);
1260 else
1261 r = send_iovec(iovec, n_iovec, STDIN_FILENO);
9aa82023
ZJS
1262
1263 finish:
1264 for (i = 0; i < n_to_free; i++)
1265 free(iovec[i].iov_base);
1266
92e92d71 1267 /* Those fields are allocated by gather_pid_metadata */
ea5cc2a8
ZJS
1268 free(context[CONTEXT_COMM]);
1269 free(context[CONTEXT_EXE]);
92e92d71 1270 free(context[CONTEXT_UNIT]);
ea5cc2a8 1271
9aa82023 1272 return r;
3c171f0b 1273}
34c10968 1274
988e89ee 1275static int process_backtrace(int argc, char *argv[]) {
ea5cc2a8
ZJS
1276 char *context[_CONTEXT_MAX] = {};
1277 _cleanup_free_ char *message = NULL;
5b45a160
ZJS
1278 _cleanup_free_ struct iovec *iovec = NULL;
1279 size_t n_iovec, n_allocated, n_to_free = 0, i;
988e89ee 1280 int r;
5b45a160
ZJS
1281 JournalImporter importer = {
1282 .fd = STDIN_FILENO,
1283 };
988e89ee
ZJS
1284
1285 log_debug("Processing backtrace on stdin...");
1286
baaa35ad
ZJS
1287 if (argc < CONTEXT_COMM + 1)
1288 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
1289 "Not enough arguments passed (%i, expected %i).",
1290 argc - 1, CONTEXT_COMM + 1 - 1);
988e89ee 1291
ea5cc2a8
ZJS
1292 context[CONTEXT_PID] = argv[2 + CONTEXT_PID];
1293 context[CONTEXT_UID] = argv[2 + CONTEXT_UID];
1294 context[CONTEXT_GID] = argv[2 + CONTEXT_GID];
1295 context[CONTEXT_SIGNAL] = argv[2 + CONTEXT_SIGNAL];
1296 context[CONTEXT_TIMESTAMP] = argv[2 + CONTEXT_TIMESTAMP];
1297 context[CONTEXT_RLIMIT] = argv[2 + CONTEXT_RLIMIT];
f45b8015 1298 context[CONTEXT_HOSTNAME] = argv[2 + CONTEXT_HOSTNAME];
988e89ee 1299
f45b8015
JF
1300 n_allocated = 34 + COREDUMP_STORAGE_EXTERNAL;
1301 /* 26 metadata, 2 static, +unknown input, 4 storage, rounded up */
5b45a160
ZJS
1302 iovec = new(struct iovec, n_allocated);
1303 if (!iovec)
1304 return log_oom();
1305
92e92d71 1306 r = gather_pid_metadata(context, argv + 2 + CONTEXT_COMM, iovec, &n_to_free);
988e89ee
ZJS
1307 if (r < 0)
1308 goto finish;
86562420
LP
1309 if (r > 0) {
1310 /* This was a special crash, and has already been processed. */
1311 r = 0;
1312 goto finish;
1313 }
5b45a160 1314 n_iovec = n_to_free;
988e89ee 1315
86562420 1316 for (;;) {
5b45a160
ZJS
1317 r = journal_importer_process_data(&importer);
1318 if (r < 0) {
1319 log_error_errno(r, "Failed to parse journal entry on stdin: %m");
1320 goto finish;
1321 }
d74dc4f2
ZJS
1322 if (r == 1 || /* complete entry */
1323 journal_importer_eof(&importer)) /* end of data */
5b45a160 1324 break;
988e89ee 1325 }
988e89ee 1326
5b45a160
ZJS
1327 if (!GREEDY_REALLOC(iovec, n_allocated, n_iovec + importer.iovw.count + 2))
1328 return log_oom();
1329
1330 if (journal_importer_eof(&importer)) {
1331 log_warning("Did not receive a full journal entry on stdin, ignoring message sent by reporter");
988e89ee 1332
ea5cc2a8
ZJS
1333 message = strjoin("MESSAGE=Process ", context[CONTEXT_PID],
1334 " (", context[CONTEXT_COMM], ")"
5b45a160
ZJS
1335 " of user ", context[CONTEXT_UID],
1336 " failed with ", context[CONTEXT_SIGNAL]);
1337 if (!message) {
1338 r = log_oom();
1339 goto finish;
1340 }
e6a7ec4b 1341 iovec[n_iovec++] = IOVEC_MAKE_STRING(message);
5b45a160
ZJS
1342 } else {
1343 for (i = 0; i < importer.iovw.count; i++)
1344 iovec[n_iovec++] = importer.iovw.iovec[i];
1345 }
988e89ee 1346
e6a7ec4b 1347 iovec[n_iovec++] = IOVEC_MAKE_STRING("MESSAGE_ID=" SD_MESSAGE_BACKTRACE_STR);
988e89ee 1348 assert_cc(2 == LOG_CRIT);
e6a7ec4b 1349 iovec[n_iovec++] = IOVEC_MAKE_STRING("PRIORITY=2");
988e89ee 1350
5b45a160 1351 assert(n_iovec <= n_allocated);
988e89ee
ZJS
1352
1353 r = sd_journal_sendv(iovec, n_iovec);
1354 if (r < 0)
1355 log_error_errno(r, "Failed to log backtrace: %m");
1356
1357 finish:
1358 for (i = 0; i < n_to_free; i++)
1359 free(iovec[i].iov_base);
1360
92e92d71 1361 /* Those fields are allocated by gather_pid_metadata */
ea5cc2a8
ZJS
1362 free(context[CONTEXT_COMM]);
1363 free(context[CONTEXT_EXE]);
92e92d71 1364 free(context[CONTEXT_UNIT]);
ea5cc2a8 1365
988e89ee
ZJS
1366 return r;
1367}
1368
4515a95e 1369static int run(int argc, char *argv[]) {
3c171f0b 1370 int r;
fee80f69 1371
9aa82023
ZJS
1372 /* First, log to a safe place, since we don't know what crashed and it might
1373 * be journald which we'd rather not log to then. */
8d4e028f 1374
3c171f0b
LP
1375 log_set_target(LOG_TARGET_KMSG);
1376 log_open();
8d4e028f 1377
3c171f0b
LP
1378 /* Make sure we never enter a loop */
1379 (void) prctl(PR_SET_DUMPABLE, 0);
8d4e028f 1380
3c171f0b
LP
1381 /* Ignore all parse errors */
1382 (void) parse_config();
fee80f69 1383
3c171f0b
LP
1384 log_debug("Selected storage '%s'.", coredump_storage_to_string(arg_storage));
1385 log_debug("Selected compression %s.", yes_no(arg_compress));
fee80f69 1386
3c171f0b 1387 r = sd_listen_fds(false);
4515a95e
ZJS
1388 if (r < 0)
1389 return log_error_errno(r, "Failed to determine the number of file descriptors: %m");
fee80f69 1390
9aa82023
ZJS
1391 /* If we got an fd passed, we are running in coredumpd mode. Otherwise we
1392 * are invoked from the kernel as coredump handler. */
988e89ee
ZJS
1393 if (r == 0) {
1394 if (streq_ptr(argv[1], "--backtrace"))
4515a95e 1395 return process_backtrace(argc, argv);
988e89ee 1396 else
4515a95e 1397 return process_kernel(argc, argv);
988e89ee 1398 } else if (r == 1)
4515a95e 1399 return process_socket(SD_LISTEN_FDS_START);
f5e04665 1400
baaa35ad
ZJS
1401 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
1402 "Received unexpected number of file descriptors.");
f5e04665 1403}
4515a95e
ZJS
1404
1405DEFINE_MAIN_FUNCTION(run);