]> git.ipfire.org Git - thirdparty/systemd.git/blame - src/coredump/coredump.c
tree-wide: drop 'This file is part of systemd' blurb
[thirdparty/systemd.git] / src / coredump / coredump.c
CommitLineData
53e1b683 1/* SPDX-License-Identifier: LGPL-2.1+ */
f5e04665 2/***
f5e04665 3 Copyright 2012 Lennart Poettering
f5e04665
LP
4***/
5
6#include <errno.h>
803a3464 7#include <stdio.h>
0d536673 8#include <stdio_ext.h>
803a3464 9#include <sys/prctl.h>
cacd6403 10#include <sys/xattr.h>
4f5dd394 11#include <unistd.h>
f5e04665 12
349cc4a5 13#if HAVE_ELFUTILS
3c171f0b
LP
14#include <dwarf.h>
15#include <elfutils/libdwfl.h>
4d229b31
UTL
16#endif
17
73a99163 18#include "sd-daemon.h"
f11943c5
LP
19#include "sd-journal.h"
20#include "sd-login.h"
73a99163 21#include "sd-messages.h"
4f5dd394
LP
22
23#include "acl-util.h"
b5efdb8a 24#include "alloc-util.h"
430f0182 25#include "capability-util.h"
ba1261bc 26#include "cgroup-util.h"
4f5dd394 27#include "compress.h"
34c10968
LP
28#include "conf-parser.h"
29#include "copy.h"
f11943c5 30#include "coredump-vacuum.h"
a0956174 31#include "dirent-util.h"
4f5dd394 32#include "escape.h"
3ffd4af2 33#include "fd-util.h"
4f5dd394 34#include "fileio.h"
f4f15635 35#include "fs-util.h"
afc5dbf3 36#include "io-util.h"
b18453ed 37#include "journal-importer.h"
4f5dd394
LP
38#include "log.h"
39#include "macro.h"
0c773903 40#include "missing.h"
4f5dd394 41#include "mkdir.h"
6bedfcbb 42#include "parse-util.h"
0b452006 43#include "process-util.h"
d14bcb4e 44#include "signal-util.h"
3c171f0b 45#include "socket-util.h"
4f5dd394
LP
46#include "special.h"
47#include "stacktrace.h"
8b43440b 48#include "string-table.h"
07630cea 49#include "string-util.h"
4f5dd394 50#include "strv.h"
b1d4f8e1 51#include "user-util.h"
4f5dd394 52#include "util.h"
34727273 53
34c10968 54/* The maximum size up to which we process coredumps */
59f448cf 55#define PROCESS_SIZE_MAX ((uint64_t) (2LLU*1024LLU*1024LLU*1024LLU))
34c10968 56
bdfd7b2c 57/* The maximum size up to which we leave the coredump around on disk */
34c10968
LP
58#define EXTERNAL_SIZE_MAX PROCESS_SIZE_MAX
59
bdfd7b2c 60/* The maximum size up to which we store the coredump in the journal */
34c10968 61#define JOURNAL_SIZE_MAX ((size_t) (767LU*1024LU*1024LU))
f5e04665 62
c4aa09b0 63/* Make sure to not make this larger than the maximum journal entry
874bc134
ZJS
64 * size. See DATA_SIZE_MAX in journald-native.c. */
65assert_cc(JOURNAL_SIZE_MAX <= DATA_SIZE_MAX);
f5e04665
LP
66
67enum {
ea5cc2a8
ZJS
68 /* We use this as array indexes for a couple of special fields we use for
69 * naming coredump files, and attaching xattrs, and for indexing argv[].
70
71 * Our pattern for man:systectl(1) kernel.core_pattern is such that the
72 * kernel passes fields until CONTEXT_RLIMIT as arguments in argv[]. After
73 * that it gets complicated: the kernel passes "comm" as one or more fields
74 * starting at index CONTEXT_COMM (in other words, full "comm" is under index
75 * CONTEXT_COMM when it does not contain spaces, which is the common
76 * case). This mapping is not reversible, so we prefer to retrieve "comm"
77 * from /proc. We only fall back to argv[CONTEXT_COMM...] when that fails.
78 *
79 * In the internal context[] array, fields before CONTEXT_COMM are the
80 * strings from argv[], so they should not be freed. The strings at indices
81 * CONTEXT_COMM and higher are allocated by us and should be freed at the
82 * end.
83 */
3c171f0b
LP
84 CONTEXT_PID,
85 CONTEXT_UID,
86 CONTEXT_GID,
87 CONTEXT_SIGNAL,
88 CONTEXT_TIMESTAMP,
bdfd7b2c 89 CONTEXT_RLIMIT,
f45b8015 90 CONTEXT_HOSTNAME,
3c171f0b
LP
91 CONTEXT_COMM,
92 CONTEXT_EXE,
92e92d71 93 CONTEXT_UNIT,
3c171f0b 94 _CONTEXT_MAX
f5e04665
LP
95};
96
34c10968
LP
97typedef enum CoredumpStorage {
98 COREDUMP_STORAGE_NONE,
99 COREDUMP_STORAGE_EXTERNAL,
100 COREDUMP_STORAGE_JOURNAL,
34c10968
LP
101 _COREDUMP_STORAGE_MAX,
102 _COREDUMP_STORAGE_INVALID = -1
103} CoredumpStorage;
104
34c10968
LP
105static const char* const coredump_storage_table[_COREDUMP_STORAGE_MAX] = {
106 [COREDUMP_STORAGE_NONE] = "none",
107 [COREDUMP_STORAGE_EXTERNAL] = "external",
108 [COREDUMP_STORAGE_JOURNAL] = "journal",
34c10968
LP
109};
110
111DEFINE_PRIVATE_STRING_TABLE_LOOKUP(coredump_storage, CoredumpStorage);
8c9571d0 112static DEFINE_CONFIG_PARSE_ENUM(config_parse_coredump_storage, coredump_storage, CoredumpStorage, "Failed to parse storage setting");
34727273
ZJS
113
114static CoredumpStorage arg_storage = COREDUMP_STORAGE_EXTERNAL;
8c9571d0 115static bool arg_compress = true;
59f448cf
LP
116static uint64_t arg_process_size_max = PROCESS_SIZE_MAX;
117static uint64_t arg_external_size_max = EXTERNAL_SIZE_MAX;
6e2b4a69 118static uint64_t arg_journal_size_max = JOURNAL_SIZE_MAX;
59f448cf
LP
119static uint64_t arg_keep_free = (uint64_t) -1;
120static uint64_t arg_max_use = (uint64_t) -1;
34c10968
LP
121
122static int parse_config(void) {
34c10968 123 static const ConfigTableItem items[] = {
8c9571d0
LP
124 { "Coredump", "Storage", config_parse_coredump_storage, 0, &arg_storage },
125 { "Coredump", "Compress", config_parse_bool, 0, &arg_compress },
59f448cf
LP
126 { "Coredump", "ProcessSizeMax", config_parse_iec_uint64, 0, &arg_process_size_max },
127 { "Coredump", "ExternalSizeMax", config_parse_iec_uint64, 0, &arg_external_size_max },
8c9571d0 128 { "Coredump", "JournalSizeMax", config_parse_iec_size, 0, &arg_journal_size_max },
59f448cf
LP
129 { "Coredump", "KeepFree", config_parse_iec_uint64, 0, &arg_keep_free },
130 { "Coredump", "MaxUse", config_parse_iec_uint64, 0, &arg_max_use },
34c10968
LP
131 {}
132 };
133
43688c49 134 return config_parse_many_nulstr(PKGSYSCONFDIR "/coredump.conf",
da412854
YW
135 CONF_PATHS_NULSTR("systemd/coredump.conf.d"),
136 "Coredump\0",
137 config_item_table_lookup, items,
bcde742e 138 CONFIG_PARSE_WARN, NULL);
34c10968
LP
139}
140
73a99163 141static inline uint64_t storage_size_max(void) {
ee0449fd
ZJS
142 if (arg_storage == COREDUMP_STORAGE_EXTERNAL)
143 return arg_external_size_max;
144 if (arg_storage == COREDUMP_STORAGE_JOURNAL)
145 return arg_journal_size_max;
146 assert(arg_storage == COREDUMP_STORAGE_NONE);
147 return 0;
73a99163
ZJS
148}
149
34c10968
LP
150static int fix_acl(int fd, uid_t uid) {
151
349cc4a5 152#if HAVE_ACL
34c10968
LP
153 _cleanup_(acl_freep) acl_t acl = NULL;
154 acl_entry_t entry;
155 acl_permset_t permset;
709f6e46 156 int r;
34c10968 157
b59233e6
LP
158 assert(fd >= 0);
159
05fd2156 160 if (uid_is_system(uid) || uid_is_dynamic(uid) || uid == UID_NOBODY)
34c10968
LP
161 return 0;
162
163 /* Make sure normal users can read (but not write or delete)
164 * their own coredumps */
165
166 acl = acl_get_fd(fd);
4a62c710
MS
167 if (!acl)
168 return log_error_errno(errno, "Failed to get ACL: %m");
34c10968
LP
169
170 if (acl_create_entry(&acl, &entry) < 0 ||
171 acl_set_tag_type(entry, ACL_USER) < 0 ||
d710aaf7
ZJS
172 acl_set_qualifier(entry, &uid) < 0)
173 return log_error_errno(errno, "Failed to patch ACL: %m");
34c10968
LP
174
175 if (acl_get_permset(entry, &permset) < 0 ||
709f6e46
MS
176 acl_add_perm(permset, ACL_READ) < 0)
177 return log_warning_errno(errno, "Failed to patch ACL: %m");
178
179 r = calc_acl_mask_if_needed(&acl);
180 if (r < 0)
181 return log_warning_errno(r, "Failed to patch ACL: %m");
34c10968 182
4a62c710
MS
183 if (acl_set_fd(fd, acl) < 0)
184 return log_error_errno(errno, "Failed to apply ACL: %m");
34c10968
LP
185#endif
186
187 return 0;
188}
189
3c171f0b 190static int fix_xattr(int fd, const char *context[_CONTEXT_MAX]) {
0cd77f97 191
3c171f0b
LP
192 static const char * const xattrs[_CONTEXT_MAX] = {
193 [CONTEXT_PID] = "user.coredump.pid",
194 [CONTEXT_UID] = "user.coredump.uid",
195 [CONTEXT_GID] = "user.coredump.gid",
196 [CONTEXT_SIGNAL] = "user.coredump.signal",
197 [CONTEXT_TIMESTAMP] = "user.coredump.timestamp",
6d337300 198 [CONTEXT_RLIMIT] = "user.coredump.rlimit",
f45b8015 199 [CONTEXT_HOSTNAME] = "user.coredump.hostname",
3c171f0b
LP
200 [CONTEXT_COMM] = "user.coredump.comm",
201 [CONTEXT_EXE] = "user.coredump.exe",
0cd77f97
LP
202 };
203
34c10968 204 int r = 0;
0cd77f97 205 unsigned i;
34c10968 206
b59233e6
LP
207 assert(fd >= 0);
208
1eef15b1 209 /* Attach some metadata to coredumps via extended
34c10968
LP
210 * attributes. Just because we can. */
211
3c171f0b 212 for (i = 0; i < _CONTEXT_MAX; i++) {
1eef15b1
ZJS
213 int k;
214
3c171f0b 215 if (isempty(context[i]) || !xattrs[i])
0cd77f97 216 continue;
34c10968 217
3c171f0b 218 k = fsetxattr(fd, xattrs[i], context[i], strlen(context[i]), XATTR_CREATE);
1eef15b1 219 if (k < 0 && r == 0)
34c10968 220 r = -errno;
0cd77f97 221 }
34c10968
LP
222
223 return r;
224}
225
b0b21dce 226#define filename_escape(s) xescape((s), "./ ")
34c10968 227
0c773903
EV
228static inline const char *coredump_tmpfile_name(const char *s) {
229 return s ? s : "(unnamed temporary file)";
230}
231
b59233e6
LP
232static int fix_permissions(
233 int fd,
234 const char *filename,
235 const char *target,
3c171f0b 236 const char *context[_CONTEXT_MAX],
b59233e6
LP
237 uid_t uid) {
238
03532f0a
LP
239 int r;
240
b59233e6 241 assert(fd >= 0);
b59233e6 242 assert(target);
3c171f0b 243 assert(context);
cfd652ed
ZJS
244
245 /* Ignore errors on these */
3c171f0b
LP
246 (void) fchmod(fd, 0640);
247 (void) fix_acl(fd, uid);
248 (void) fix_xattr(fd, context);
cfd652ed 249
4a62c710 250 if (fsync(fd) < 0)
0c773903 251 return log_error_errno(errno, "Failed to sync coredump %s: %m", coredump_tmpfile_name(filename));
cfd652ed 252
8ac2f74f
LP
253 (void) fsync_directory_of_file(fd);
254
03532f0a
LP
255 r = link_tmpfile(fd, filename, target);
256 if (r < 0)
257 return log_error_errno(r, "Failed to move coredump %s into place: %m", target);
cfd652ed
ZJS
258
259 return 0;
260}
261
59f448cf 262static int maybe_remove_external_coredump(const char *filename, uint64_t size) {
cfd652ed 263
b59233e6 264 /* Returns 1 if might remove, 0 if will not remove, < 0 on error. */
cfd652ed 265
fc6cec86 266 if (arg_storage == COREDUMP_STORAGE_EXTERNAL &&
cfd652ed
ZJS
267 size <= arg_external_size_max)
268 return 0;
269
270 if (!filename)
271 return 1;
272
4a62c710
MS
273 if (unlink(filename) < 0 && errno != ENOENT)
274 return log_error_errno(errno, "Failed to unlink %s: %m", filename);
cfd652ed
ZJS
275
276 return 1;
277}
278
3c171f0b 279static int make_filename(const char *context[_CONTEXT_MAX], char **ret) {
b59233e6 280 _cleanup_free_ char *c = NULL, *u = NULL, *p = NULL, *t = NULL;
a7f7d1bd 281 sd_id128_t boot = {};
34c10968
LP
282 int r;
283
3c171f0b 284 assert(context);
34c10968 285
3c171f0b 286 c = filename_escape(context[CONTEXT_COMM]);
34c10968 287 if (!c)
b59233e6 288 return -ENOMEM;
34c10968 289
3c171f0b 290 u = filename_escape(context[CONTEXT_UID]);
0dc5d23c 291 if (!u)
b59233e6 292 return -ENOMEM;
34c10968
LP
293
294 r = sd_id128_get_boot(&boot);
b59233e6 295 if (r < 0)
34c10968 296 return r;
34c10968 297
3c171f0b 298 p = filename_escape(context[CONTEXT_PID]);
b59233e6
LP
299 if (!p)
300 return -ENOMEM;
301
3c171f0b 302 t = filename_escape(context[CONTEXT_TIMESTAMP]);
b59233e6
LP
303 if (!t)
304 return -ENOMEM;
305
306 if (asprintf(ret,
0dc5d23c 307 "/var/lib/systemd/coredump/core.%s.%s." SD_ID128_FORMAT_STR ".%s.%s000000",
34c10968 308 c,
0dc5d23c 309 u,
34c10968
LP
310 SD_ID128_FORMAT_VAL(boot),
311 p,
b59233e6
LP
312 t) < 0)
313 return -ENOMEM;
314
315 return 0;
316}
317
318static int save_external_coredump(
3c171f0b
LP
319 const char *context[_CONTEXT_MAX],
320 int input_fd,
b59233e6 321 char **ret_filename,
5f3e0a74
HW
322 int *ret_node_fd,
323 int *ret_data_fd,
0cd4e913 324 uint64_t *ret_size,
cc4419ed 325 bool *ret_truncated) {
b59233e6
LP
326
327 _cleanup_free_ char *fn = NULL, *tmp = NULL;
328 _cleanup_close_ int fd = -1;
ee0449fd 329 uint64_t rlimit, process_limit, max_size;
b59233e6 330 struct stat st;
3c171f0b 331 uid_t uid;
b59233e6
LP
332 int r;
333
3c171f0b 334 assert(context);
b59233e6 335 assert(ret_filename);
5f3e0a74
HW
336 assert(ret_node_fd);
337 assert(ret_data_fd);
b59233e6
LP
338 assert(ret_size);
339
3c171f0b
LP
340 r = parse_uid(context[CONTEXT_UID], &uid);
341 if (r < 0)
342 return log_error_errno(r, "Failed to parse UID: %m");
343
bdfd7b2c
LP
344 r = safe_atou64(context[CONTEXT_RLIMIT], &rlimit);
345 if (r < 0)
346 return log_error_errno(r, "Failed to parse resource limit: %s", context[CONTEXT_RLIMIT]);
6998b540
ZJS
347 if (rlimit < page_size()) {
348 /* Is coredumping disabled? Then don't bother saving/processing the coredump.
349 * Anything below PAGE_SIZE cannot give a readable coredump (the kernel uses
350 * ELF_EXEC_PAGESIZE which is not easily accessible, but is usually the same as PAGE_SIZE. */
73a99163
ZJS
351 log_info("Resource limits disable core dumping for process %s (%s).",
352 context[CONTEXT_PID], context[CONTEXT_COMM]);
bdfd7b2c
LP
353 return -EBADSLT;
354 }
355
ee0449fd
ZJS
356 process_limit = MAX(arg_process_size_max, storage_size_max());
357 if (process_limit == 0) {
358 log_debug("Limits for coredump processing and storage are both 0, not dumping core.");
359 return -EBADSLT;
360 }
361
bdfd7b2c 362 /* Never store more than the process configured, or than we actually shall keep or process */
ee0449fd 363 max_size = MIN(rlimit, process_limit);
bdfd7b2c 364
3c171f0b 365 r = make_filename(context, &fn);
23bbb0de
MS
366 if (r < 0)
367 return log_error_errno(r, "Failed to determine coredump file name: %m");
34c10968 368
d2e54fae 369 mkdir_p_label("/var/lib/systemd/coredump", 0755);
803a3464 370
03532f0a 371 fd = open_tmpfile_linkable(fn, O_RDWR|O_CLOEXEC, &tmp);
4a62c710 372 if (fd < 0)
03532f0a 373 return log_error_errno(fd, "Failed to create temporary file for coredump %s: %m", fn);
803a3464 374
1c876927 375 r = copy_bytes(input_fd, fd, max_size, 0);
73a99163
ZJS
376 if (r < 0) {
377 log_error_errno(r, "Cannot store coredump of %s (%s): %m", context[CONTEXT_PID], context[CONTEXT_COMM]);
93240d3a 378 goto fail;
0cd4e913 379 }
cc4419ed
ZJS
380 *ret_truncated = r == 1;
381 if (*ret_truncated)
73a99163
ZJS
382 log_struct(LOG_INFO,
383 LOG_MESSAGE("Core file was truncated to %zu bytes.", max_size),
384 "SIZE_LIMIT=%zu", max_size,
a1230ff9 385 "MESSAGE_ID=" SD_MESSAGE_TRUNCATED_CORE_STR);
803a3464 386
34c10968 387 if (fstat(fd, &st) < 0) {
73a99163 388 log_error_errno(errno, "Failed to fstat core file %s: %m", coredump_tmpfile_name(tmp));
34c10968
LP
389 goto fail;
390 }
391
7849c2ac 392 if (lseek(fd, 0, SEEK_SET) == (off_t) -1) {
0c773903 393 log_error_errno(errno, "Failed to seek on %s: %m", coredump_tmpfile_name(tmp));
b59233e6 394 goto fail;
7849c2ac
TA
395 }
396
349cc4a5 397#if HAVE_XZ || HAVE_LZ4
cfd652ed 398 /* If we will remove the coredump anyway, do not compress. */
6e9ef603 399 if (arg_compress && !maybe_remove_external_coredump(NULL, st.st_size)) {
cfd652ed 400
b59233e6
LP
401 _cleanup_free_ char *fn_compressed = NULL, *tmp_compressed = NULL;
402 _cleanup_close_ int fd_compressed = -1;
cfd652ed 403
d89c8fdf 404 fn_compressed = strappend(fn, COMPRESSED_EXT);
b59233e6 405 if (!fn_compressed) {
d89c8fdf 406 log_oom();
cfd652ed
ZJS
407 goto uncompressed;
408 }
409
03532f0a
LP
410 fd_compressed = open_tmpfile_linkable(fn_compressed, O_RDWR|O_CLOEXEC, &tmp_compressed);
411 if (fd_compressed < 0) {
412 log_error_errno(fd_compressed, "Failed to create temporary file for coredump %s: %m", fn_compressed);
b59233e6 413 goto uncompressed;
03532f0a 414 }
cfd652ed 415
d89c8fdf 416 r = compress_stream(fd, fd_compressed, -1);
b59233e6 417 if (r < 0) {
0c773903 418 log_error_errno(r, "Failed to compress %s: %m", coredump_tmpfile_name(tmp_compressed));
b59233e6
LP
419 goto fail_compressed;
420 }
421
3c171f0b 422 r = fix_permissions(fd_compressed, tmp_compressed, fn_compressed, context, uid);
cfd652ed 423 if (r < 0)
b59233e6
LP
424 goto fail_compressed;
425
426 /* OK, this worked, we can get rid of the uncompressed version now */
0c773903
EV
427 if (tmp)
428 unlink_noerrno(tmp);
cfd652ed 429
1cc6c93a
YW
430 *ret_filename = TAKE_PTR(fn_compressed); /* compressed */
431 *ret_node_fd = TAKE_FD(fd_compressed); /* compressed */
432 *ret_data_fd = TAKE_FD(fd); /* uncompressed */
59f448cf 433 *ret_size = (uint64_t) st.st_size; /* uncompressed */
cfd652ed 434
cfd652ed
ZJS
435 return 0;
436
b59233e6 437 fail_compressed:
0c773903
EV
438 if (tmp_compressed)
439 (void) unlink(tmp_compressed);
34c10968 440 }
cfd652ed
ZJS
441
442uncompressed:
3b1a55e1 443#endif
5f3e0a74 444
3c171f0b 445 r = fix_permissions(fd, tmp, fn, context, uid);
cfd652ed
ZJS
446 if (r < 0)
447 goto fail;
34c10968 448
1cc6c93a
YW
449 *ret_filename = TAKE_PTR(fn);
450 *ret_data_fd = TAKE_FD(fd);
5f3e0a74 451 *ret_node_fd = -1;
59f448cf 452 *ret_size = (uint64_t) st.st_size;
34c10968 453
34c10968
LP
454 return 0;
455
456fail:
0c773903
EV
457 if (tmp)
458 (void) unlink(tmp);
34c10968
LP
459 return r;
460}
461
462static int allocate_journal_field(int fd, size_t size, char **ret, size_t *ret_size) {
463 _cleanup_free_ char *field = NULL;
464 ssize_t n;
465
8d4e028f 466 assert(fd >= 0);
34c10968
LP
467 assert(ret);
468 assert(ret_size);
469
4a62c710
MS
470 if (lseek(fd, 0, SEEK_SET) == (off_t) -1)
471 return log_warning_errno(errno, "Failed to seek: %m");
803a3464 472
34c10968
LP
473 field = malloc(9 + size);
474 if (!field) {
cfd652ed 475 log_warning("Failed to allocate memory for coredump, coredump will not be stored.");
34c10968
LP
476 return -ENOMEM;
477 }
478
479 memcpy(field, "COREDUMP=", 9);
480
481 n = read(fd, field + 9, size);
23bbb0de
MS
482 if (n < 0)
483 return log_error_errno((int) n, "Failed to read core data: %m");
34c10968
LP
484 if ((size_t) n < size) {
485 log_error("Core data too short.");
486 return -EIO;
487 }
488
1cc6c93a 489 *ret = TAKE_PTR(field);
34c10968
LP
490 *ret_size = size + 9;
491
34c10968
LP
492 return 0;
493}
803a3464 494
3f132692
JF
495/* Joins /proc/[pid]/fd/ and /proc/[pid]/fdinfo/ into the following lines:
496 * 0:/dev/pts/23
497 * pos: 0
498 * flags: 0100002
499 *
500 * 1:/dev/pts/23
501 * pos: 0
502 * flags: 0100002
503 *
504 * 2:/dev/pts/23
505 * pos: 0
506 * flags: 0100002
507 * EOF
508 */
509static int compose_open_fds(pid_t pid, char **open_fds) {
4d84bc2f
LP
510 _cleanup_closedir_ DIR *proc_fd_dir = NULL;
511 _cleanup_close_ int proc_fdinfo_fd = -1;
512 _cleanup_free_ char *buffer = NULL;
3f132692 513 _cleanup_fclose_ FILE *stream = NULL;
59059b4a 514 const char *fddelim = "", *path;
3f132692 515 struct dirent *dent = NULL;
4d84bc2f 516 size_t size = 0;
3f132692
JF
517 int r = 0;
518
519 assert(pid >= 0);
520 assert(open_fds != NULL);
521
59059b4a 522 path = procfs_file_alloca(pid, "fd");
3f132692 523 proc_fd_dir = opendir(path);
59059b4a
ZJS
524 if (!proc_fd_dir)
525 return -errno;
3f132692 526
4d84bc2f 527 proc_fdinfo_fd = openat(dirfd(proc_fd_dir), "../fdinfo", O_DIRECTORY|O_NOFOLLOW|O_CLOEXEC|O_PATH);
59059b4a
ZJS
528 if (proc_fdinfo_fd < 0)
529 return -errno;
3f132692 530
4d84bc2f 531 stream = open_memstream(&buffer, &size);
3f132692
JF
532 if (!stream)
533 return -ENOMEM;
534
0d536673
LP
535 (void) __fsetlocking(stream, FSETLOCKING_BYCALLER);
536
4d84bc2f 537 FOREACH_DIRENT(dent, proc_fd_dir, return -errno) {
3f132692 538 _cleanup_fclose_ FILE *fdinfo = NULL;
4d84bc2f 539 _cleanup_free_ char *fdname = NULL;
59059b4a 540 char line[LINE_MAX];
4d84bc2f 541 int fd;
3f132692 542
59059b4a 543 r = readlinkat_malloc(dirfd(proc_fd_dir), dent->d_name, &fdname);
3f132692
JF
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 */
59059b4a
ZJS
551 fd = openat(proc_fdinfo_fd, dent->d_name, O_NOFOLLOW|O_CLOEXEC|O_RDONLY);
552 if (fd < 0)
3f132692
JF
553 continue;
554
59059b4a 555 fdinfo = fdopen(fd, "re");
234519ae 556 if (!fdinfo) {
0d536673 557 safe_close(fd);
3f132692 558 continue;
59059b4a 559 }
3f132692 560
4d84bc2f 561 FOREACH_LINE(line, fdinfo, break) {
0d536673 562 fputs(line, stream);
4d84bc2f
LP
563 if (!endswith(line, "\n"))
564 fputc('\n', stream);
565 }
3f132692
JF
566 }
567
4d84bc2f 568 errno = 0;
74ca738f 569 stream = safe_fclose(stream);
4d84bc2f 570
b3267152 571 if (errno > 0)
4d84bc2f
LP
572 return -errno;
573
ae2a15bc 574 *open_fds = TAKE_PTR(buffer);
4d84bc2f 575
3f132692
JF
576 return 0;
577}
578
7ed03ce6
JF
579static int get_process_ns(pid_t pid, const char *namespace, ino_t *ns) {
580 const char *p;
581 struct stat stbuf;
582 _cleanup_close_ int proc_ns_dir_fd;
583
584 p = procfs_file_alloca(pid, "ns");
585
586 proc_ns_dir_fd = open(p, O_DIRECTORY | O_CLOEXEC | O_RDONLY);
587 if (proc_ns_dir_fd < 0)
588 return -errno;
589
590 if (fstatat(proc_ns_dir_fd, namespace, &stbuf, /* flags */0) < 0)
591 return -errno;
592
593 *ns = stbuf.st_ino;
594 return 0;
595}
596
597static int get_mount_namespace_leader(pid_t pid, pid_t *container_pid) {
598 pid_t cpid = pid, ppid = 0;
599 ino_t proc_mntns;
600 int r = 0;
601
602 r = get_process_ns(pid, "mnt", &proc_mntns);
603 if (r < 0)
604 return r;
605
aa7530d6 606 for (;;) {
7ed03ce6
JF
607 ino_t parent_mntns;
608
609 r = get_process_ppid(cpid, &ppid);
610 if (r < 0)
611 return r;
612
613 r = get_process_ns(ppid, "mnt", &parent_mntns);
614 if (r < 0)
615 return r;
616
617 if (proc_mntns != parent_mntns)
618 break;
619
620 if (ppid == 1)
621 return -ENOENT;
622
623 cpid = ppid;
624 }
625
626 *container_pid = ppid;
627 return 0;
628}
629
630/* Returns 1 if the parent was found.
631 * Returns 0 if there is not a process we can call the pid's
632 * container parent (the pid's process isn't 'containerized').
633 * Returns a negative number on errors.
634 */
635static int get_process_container_parent_cmdline(pid_t pid, char** cmdline) {
636 int r = 0;
637 pid_t container_pid;
638 const char *proc_root_path;
639 struct stat root_stat, proc_root_stat;
640
641 /* To compare inodes of / and /proc/[pid]/root */
642 if (stat("/", &root_stat) < 0)
643 return -errno;
644
645 proc_root_path = procfs_file_alloca(pid, "root");
646 if (stat(proc_root_path, &proc_root_stat) < 0)
647 return -errno;
648
649 /* The process uses system root. */
650 if (proc_root_stat.st_ino == root_stat.st_ino) {
651 *cmdline = NULL;
652 return 0;
653 }
654
655 r = get_mount_namespace_leader(pid, &container_pid);
656 if (r < 0)
657 return r;
658
d3cba4ea
EV
659 r = get_process_cmdline(container_pid, 0, false, cmdline);
660 if (r < 0)
661 return r;
662
663 return 1;
7ed03ce6
JF
664}
665
3c171f0b
LP
666static int change_uid_gid(const char *context[]) {
667 uid_t uid;
668 gid_t gid;
669 int r;
34c10968 670
3c171f0b
LP
671 r = parse_uid(context[CONTEXT_UID], &uid);
672 if (r < 0)
673 return r;
8c8549db 674
888e378d
LP
675 if (uid <= SYSTEM_UID_MAX) {
676 const char *user = "systemd-coredump";
677
678 r = get_user_creds(&user, &uid, &gid, NULL, NULL);
679 if (r < 0) {
680 log_warning_errno(r, "Cannot resolve %s user. Proceeding to dump core as root: %m", user);
681 uid = gid = 0;
682 }
683 } else {
684 r = parse_gid(context[CONTEXT_GID], &gid);
685 if (r < 0)
686 return r;
687 }
3c171f0b
LP
688
689 return drop_privileges(uid, gid, 0);
690}
8c8549db 691
b8cda92d
LP
692static bool is_journald_crash(const char *context[_CONTEXT_MAX]) {
693 assert(context);
694
695 return streq_ptr(context[CONTEXT_UNIT], SPECIAL_JOURNALD_SERVICE);
696}
697
698static bool is_pid1_crash(const char *context[_CONTEXT_MAX]) {
699 assert(context);
700
701 return streq_ptr(context[CONTEXT_UNIT], SPECIAL_INIT_SCOPE) ||
702 streq_ptr(context[CONTEXT_PID], "1");
703}
704
0cd4e913
ZJS
705#define SUBMIT_COREDUMP_FIELDS 4
706
3c171f0b
LP
707static int submit_coredump(
708 const char *context[_CONTEXT_MAX],
709 struct iovec *iovec,
710 size_t n_iovec_allocated,
711 size_t n_iovec,
712 int input_fd) {
34c10968 713
5f3e0a74 714 _cleanup_close_ int coredump_fd = -1, coredump_node_fd = -1;
3c171f0b 715 _cleanup_free_ char *core_message = NULL, *filename = NULL, *coredump_data = NULL;
a5ca3649 716 uint64_t coredump_size = UINT64_MAX;
92e92d71 717 bool truncated = false, journald_crash;
3c171f0b 718 int r;
f5e04665 719
3c171f0b
LP
720 assert(context);
721 assert(iovec);
0cd4e913 722 assert(n_iovec_allocated >= n_iovec + SUBMIT_COREDUMP_FIELDS);
3c171f0b 723 assert(input_fd >= 0);
f5e04665 724
b8cda92d 725 journald_crash = is_journald_crash(context);
92e92d71 726
3c171f0b
LP
727 /* Vacuum before we write anything again */
728 (void) coredump_vacuum(-1, arg_keep_free, arg_max_use);
803a3464 729
3c171f0b 730 /* Always stream the coredump to disk, if that's possible */
0cd4e913
ZJS
731 r = save_external_coredump(context, input_fd,
732 &filename, &coredump_node_fd, &coredump_fd, &coredump_size, &truncated);
3c171f0b
LP
733 if (r < 0)
734 /* Skip whole core dumping part */
735 goto log;
736
737 /* If we don't want to keep the coredump on disk, remove it now, as later on we will lack the privileges for
738 * it. However, we keep the fd to it, so that we can still process it and log it. */
739 r = maybe_remove_external_coredump(filename, coredump_size);
740 if (r < 0)
741 return r;
742 if (r == 0) {
743 const char *coredump_filename;
744
745 coredump_filename = strjoina("COREDUMP_FILENAME=", filename);
e6a7ec4b 746 iovec[n_iovec++] = IOVEC_MAKE_STRING(coredump_filename);
6e9ef603 747 } else if (arg_storage == COREDUMP_STORAGE_EXTERNAL)
5206a724 748 log_info("The core will not be stored: size %"PRIu64" is greater than %"PRIu64" (the configured maximum)",
6e9ef603 749 coredump_size, arg_external_size_max);
f5e04665 750
3c171f0b
LP
751 /* Vacuum again, but exclude the coredump we just created */
752 (void) coredump_vacuum(coredump_node_fd >= 0 ? coredump_node_fd : coredump_fd, arg_keep_free, arg_max_use);
8c9571d0 753
3c171f0b
LP
754 /* Now, let's drop privileges to become the user who owns the segfaulted process and allocate the coredump
755 * memory under the user's uid. This also ensures that the credentials journald will see are the ones of the
756 * coredumping user, thus making sure the user gets access to the core dump. Let's also get rid of all
757 * capabilities, if we run as root, we won't need them anymore. */
758 r = change_uid_gid(context);
759 if (r < 0)
760 return log_error_errno(r, "Failed to drop privileges: %m");
34c10968 761
349cc4a5 762#if HAVE_ELFUTILS
3c171f0b
LP
763 /* Try to get a strack trace if we can */
764 if (coredump_size <= arg_process_size_max) {
765 _cleanup_free_ char *stacktrace = NULL;
766
767 r = coredump_make_stack_trace(coredump_fd, context[CONTEXT_EXE], &stacktrace);
768 if (r >= 0)
605405c6
ZJS
769 core_message = strjoin("MESSAGE=Process ", context[CONTEXT_PID],
770 " (", context[CONTEXT_COMM], ") of user ",
92e92d71
ZJS
771 context[CONTEXT_UID], " dumped core.",
772 journald_crash ? "\nCoredump diverted to " : "",
773 journald_crash ? filename : "",
774 "\n\n", stacktrace);
3c171f0b
LP
775 else if (r == -EINVAL)
776 log_warning("Failed to generate stack trace: %s", dwfl_errmsg(dwfl_errno()));
777 else
778 log_warning_errno(r, "Failed to generate stack trace: %m");
6e9ef603 779 } else
5206a724 780 log_debug("Not generating stack trace: core size %"PRIu64" is greater than %"PRIu64" (the configured maximum)",
6e9ef603 781 coredump_size, arg_process_size_max);
803a3464 782
3c171f0b
LP
783 if (!core_message)
784#endif
785log:
92e92d71
ZJS
786 core_message = strjoin("MESSAGE=Process ", context[CONTEXT_PID],
787 " (", context[CONTEXT_COMM], ") of user ",
788 context[CONTEXT_UID], " dumped core.",
789 journald_crash ? "\nCoredump diverted to " : NULL,
790 journald_crash ? filename : NULL);
791 if (!core_message)
792 return log_oom();
793
794 if (journald_crash) {
795 /* We cannot log to the journal, so just print the MESSAGE.
796 * The target was set previously to something safe. */
ae2173d6 797 log_dispatch(LOG_ERR, 0, core_message);
92e92d71
ZJS
798 return 0;
799 }
800
e6a7ec4b 801 iovec[n_iovec++] = IOVEC_MAKE_STRING(core_message);
3c171f0b 802
0cd4e913 803 if (truncated)
e6a7ec4b 804 iovec[n_iovec++] = IOVEC_MAKE_STRING("COREDUMP_TRUNCATED=1");
0cd4e913 805
3c171f0b 806 /* Optionally store the entire coredump in the journal */
6e9ef603
ZJS
807 if (arg_storage == COREDUMP_STORAGE_JOURNAL) {
808 if (coredump_size <= arg_journal_size_max) {
809 size_t sz = 0;
810
811 /* Store the coredump itself in the journal */
812
813 r = allocate_journal_field(coredump_fd, (size_t) coredump_size, &coredump_data, &sz);
e6a7ec4b
LP
814 if (r >= 0)
815 iovec[n_iovec++] = IOVEC_MAKE(coredump_data, sz);
816 else
6e9ef603
ZJS
817 log_warning_errno(r, "Failed to attach the core to the journal entry: %m");
818 } else
5206a724 819 log_info("The core will not be stored: size %"PRIu64" is greater than %"PRIu64" (the configured maximum)",
6e9ef603 820 coredump_size, arg_journal_size_max);
f5e04665
LP
821 }
822
3c171f0b
LP
823 assert(n_iovec <= n_iovec_allocated);
824
825 r = sd_journal_sendv(iovec, n_iovec);
826 if (r < 0)
827 return log_error_errno(r, "Failed to log coredump: %m");
828
829 return 0;
830}
831
ea5cc2a8 832static void map_context_fields(const struct iovec *iovec, const char* context[]) {
3c171f0b 833
92e92d71 834 static const char * const context_field_names[] = {
3c171f0b
LP
835 [CONTEXT_PID] = "COREDUMP_PID=",
836 [CONTEXT_UID] = "COREDUMP_UID=",
837 [CONTEXT_GID] = "COREDUMP_GID=",
838 [CONTEXT_SIGNAL] = "COREDUMP_SIGNAL=",
839 [CONTEXT_TIMESTAMP] = "COREDUMP_TIMESTAMP=",
92e92d71 840 [CONTEXT_RLIMIT] = "COREDUMP_RLIMIT=",
f45b8015 841 [CONTEXT_HOSTNAME] = "COREDUMP_HOSTNAME=",
3c171f0b
LP
842 [CONTEXT_COMM] = "COREDUMP_COMM=",
843 [CONTEXT_EXE] = "COREDUMP_EXE=",
844 };
845
846 unsigned i;
847
848 assert(iovec);
849 assert(context);
850
92e92d71 851 for (i = 0; i < ELEMENTSOF(context_field_names); i++) {
d27b725a 852 char *p;
3c171f0b 853
92e92d71
ZJS
854 if (!context_field_names[i])
855 continue;
856
d27b725a
LP
857 p = memory_startswith(iovec->iov_base, iovec->iov_len, context_field_names[i]);
858 if (!p)
3c171f0b
LP
859 continue;
860
861 /* Note that these strings are NUL terminated, because we made sure that a trailing NUL byte is in the
862 * buffer, though not included in the iov_len count. (see below) */
d27b725a 863 context[i] = p;
3c171f0b
LP
864 break;
865 }
866}
867
868static int process_socket(int fd) {
869 _cleanup_close_ int coredump_fd = -1;
870 struct iovec *iovec = NULL;
80002f66 871 size_t n_iovec = 0, n_allocated = 0, i, k;
3c171f0b
LP
872 const char *context[_CONTEXT_MAX] = {};
873 int r;
874
875 assert(fd >= 0);
876
877 log_set_target(LOG_TARGET_AUTO);
878 log_parse_environment();
879 log_open();
880
988e89ee
ZJS
881 log_debug("Processing coredump received on stdin...");
882
3c171f0b
LP
883 for (;;) {
884 union {
885 struct cmsghdr cmsghdr;
886 uint8_t buf[CMSG_SPACE(sizeof(int))];
887 } control = {};
888 struct msghdr mh = {
889 .msg_control = &control,
890 .msg_controllen = sizeof(control),
891 .msg_iovlen = 1,
892 };
893 ssize_t n;
fe1ef0f8 894 ssize_t l;
3c171f0b 895
0cd4e913 896 if (!GREEDY_REALLOC(iovec, n_allocated, n_iovec + SUBMIT_COREDUMP_FIELDS)) {
3c171f0b
LP
897 r = log_oom();
898 goto finish;
899 }
900
fe1ef0f8
EV
901 l = next_datagram_size_fd(fd);
902 if (l < 0) {
903 r = log_error_errno(l, "Failed to determine datagram size to read: %m");
3c171f0b
LP
904 goto finish;
905 }
906
907 assert(l >= 0);
908
909 iovec[n_iovec].iov_len = l;
910 iovec[n_iovec].iov_base = malloc(l + 1);
3c171f0b
LP
911 if (!iovec[n_iovec].iov_base) {
912 r = log_oom();
913 goto finish;
914 }
915
916 mh.msg_iov = iovec + n_iovec;
917
918 n = recvmsg(fd, &mh, MSG_NOSIGNAL|MSG_CMSG_CLOEXEC);
919 if (n < 0) {
920 free(iovec[n_iovec].iov_base);
921 r = log_error_errno(errno, "Failed to receive datagram: %m");
922 goto finish;
923 }
924
925 if (n == 0) {
926 struct cmsghdr *cmsg, *found = NULL;
927 /* The final zero-length datagram carries the file descriptor and tells us that we're done. */
928
929 free(iovec[n_iovec].iov_base);
930
931 CMSG_FOREACH(cmsg, &mh) {
932 if (cmsg->cmsg_level == SOL_SOCKET &&
933 cmsg->cmsg_type == SCM_RIGHTS &&
934 cmsg->cmsg_len == CMSG_LEN(sizeof(int))) {
935 assert(!found);
936 found = cmsg;
937 }
938 }
939
940 if (!found) {
941 log_error("Coredump file descriptor missing.");
942 r = -EBADMSG;
943 goto finish;
944 }
945
946 assert(coredump_fd < 0);
947 coredump_fd = *(int*) CMSG_DATA(found);
948 break;
949 }
950
951 /* Add trailing NUL byte, in case these are strings */
952 ((char*) iovec[n_iovec].iov_base)[n] = 0;
953 iovec[n_iovec].iov_len = (size_t) n;
954
955 cmsg_close_all(&mh);
956 map_context_fields(iovec + n_iovec, context);
957 n_iovec++;
958 }
959
0cd4e913 960 if (!GREEDY_REALLOC(iovec, n_allocated, n_iovec + SUBMIT_COREDUMP_FIELDS)) {
3c171f0b 961 r = log_oom();
34c10968
LP
962 goto finish;
963 }
964
61233823 965 /* Make sure we got all data we really need */
3c171f0b
LP
966 assert(context[CONTEXT_PID]);
967 assert(context[CONTEXT_UID]);
968 assert(context[CONTEXT_GID]);
969 assert(context[CONTEXT_SIGNAL]);
970 assert(context[CONTEXT_TIMESTAMP]);
bdfd7b2c 971 assert(context[CONTEXT_RLIMIT]);
f45b8015 972 assert(context[CONTEXT_HOSTNAME]);
3c171f0b
LP
973 assert(context[CONTEXT_COMM]);
974 assert(coredump_fd >= 0);
975
80002f66
LP
976 /* Small quirk: the journal fields contain the timestamp padded with six zeroes, so that the kernel-supplied 1s
977 * granularity timestamps becomes 1µs granularity, i.e. the granularity systemd usually operates in. Since we
978 * are reconstructing the original kernel context, we chop this off again, here. */
979 k = strlen(context[CONTEXT_TIMESTAMP]);
980 if (k > 6)
981 context[CONTEXT_TIMESTAMP] = strndupa(context[CONTEXT_TIMESTAMP], k - 6);
982
5b45a160 983 r = submit_coredump(context, iovec, n_allocated, n_iovec, coredump_fd);
3c171f0b
LP
984
985finish:
986 for (i = 0; i < n_iovec; i++)
987 free(iovec[i].iov_base);
988 free(iovec);
989
990 return r;
991}
992
993static int send_iovec(const struct iovec iovec[], size_t n_iovec, int input_fd) {
994
995 static const union sockaddr_union sa = {
996 .un.sun_family = AF_UNIX,
997 .un.sun_path = "/run/systemd/coredump",
998 };
999 _cleanup_close_ int fd = -1;
1000 size_t i;
1001 int r;
1002
1003 assert(iovec || n_iovec <= 0);
1004 assert(input_fd >= 0);
1005
1006 fd = socket(AF_UNIX, SOCK_SEQPACKET|SOCK_CLOEXEC, 0);
1007 if (fd < 0)
1008 return log_error_errno(errno, "Failed to create coredump socket: %m");
1009
fc2fffe7 1010 if (connect(fd, &sa.sa, SOCKADDR_UN_LEN(sa.un)) < 0)
3c171f0b
LP
1011 return log_error_errno(errno, "Failed to connect to coredump service: %m");
1012
1013 for (i = 0; i < n_iovec; i++) {
fec603eb
LP
1014 struct msghdr mh = {
1015 .msg_iov = (struct iovec*) iovec + i,
1016 .msg_iovlen = 1,
1017 };
1018 struct iovec copy[2];
1019
1020 for (;;) {
1021 if (sendmsg(fd, &mh, MSG_NOSIGNAL) >= 0)
1022 break;
1023
1024 if (errno == EMSGSIZE && mh.msg_iov[0].iov_len > 0) {
1025 /* This field didn't fit? That's a pity. Given that this is just metadata,
1026 * let's truncate the field at half, and try again. We append three dots, in
1027 * order to show that this is truncated. */
1028
1029 if (mh.msg_iov != copy) {
1030 /* We don't want to modify the caller's iovec, hence let's create our
1031 * own array, consisting of two new iovecs, where the first is a
1032 * (truncated) copy of what we want to send, and the second one
1033 * contains the trailing dots. */
1034 copy[0] = iovec[i];
1035 copy[1] = (struct iovec) {
1036 .iov_base = (char[]) { '.', '.', '.' },
1037 .iov_len = 3,
1038 };
1039
1040 mh.msg_iov = copy;
1041 mh.msg_iovlen = 2;
1042 }
1043
1044 copy[0].iov_len /= 2; /* halve it, and try again */
1045 continue;
1046 }
3c171f0b 1047
3c171f0b 1048 return log_error_errno(errno, "Failed to send coredump datagram: %m");
fec603eb 1049 }
1eef15b1
ZJS
1050 }
1051
3c171f0b
LP
1052 r = send_one_fd(fd, input_fd, 0);
1053 if (r < 0)
1054 return log_error_errno(r, "Failed to send coredump fd: %m");
1eef15b1 1055
3c171f0b
LP
1056 return 0;
1057}
1eef15b1 1058
f45b8015 1059static char* set_iovec_field(struct iovec *iovec, size_t *n_iovec, const char *field, const char *value) {
9aa82023 1060 char *x;
3c171f0b 1061
9aa82023
ZJS
1062 x = strappend(field, value);
1063 if (x)
e6a7ec4b 1064 iovec[(*n_iovec)++] = IOVEC_MAKE_STRING(x);
9aa82023
ZJS
1065 return x;
1066}
3c171f0b 1067
f45b8015 1068static char* set_iovec_field_free(struct iovec *iovec, size_t *n_iovec, const char *field, char *value) {
9aa82023
ZJS
1069 char *x;
1070
1071 x = set_iovec_field(iovec, n_iovec, field, value);
1072 free(value);
1073 return x;
1074}
1075
92e92d71 1076static int gather_pid_metadata(
ea5cc2a8 1077 char* context[_CONTEXT_MAX],
9aa82023 1078 char **comm_fallback,
5b45a160 1079 struct iovec *iovec, size_t *n_iovec) {
d14bcb4e 1080
f45b8015 1081 /* We need 27 empty slots in iovec!
d14bcb4e
LP
1082 *
1083 * Note that if we fail on oom later on, we do not roll-back changes to the iovec structure. (It remains valid,
1084 * with the first n_iovec fields initialized.) */
3c171f0b 1085
3c171f0b 1086 uid_t owner_uid;
3c171f0b
LP
1087 pid_t pid;
1088 char *t;
9aa82023 1089 const char *p;
d14bcb4e 1090 int r, signo;
3c171f0b 1091
9aa82023 1092 r = parse_pid(context[CONTEXT_PID], &pid);
3c171f0b 1093 if (r < 0)
988e89ee 1094 return log_error_errno(r, "Failed to parse PID \"%s\": %m", context[CONTEXT_PID]);
3c171f0b 1095
ea5cc2a8 1096 r = get_process_comm(pid, &context[CONTEXT_COMM]);
3c171f0b
LP
1097 if (r < 0) {
1098 log_warning_errno(r, "Failed to get COMM, falling back to the command line: %m");
ea5cc2a8
ZJS
1099 context[CONTEXT_COMM] = strv_join(comm_fallback, " ");
1100 if (!context[CONTEXT_COMM])
3c171f0b
LP
1101 return log_oom();
1102 }
1103
ea5cc2a8 1104 r = get_process_exe(pid, &context[CONTEXT_EXE]);
3c171f0b
LP
1105 if (r < 0)
1106 log_warning_errno(r, "Failed to get EXE, ignoring: %m");
1107
92e92d71 1108 if (cg_pid_get_unit(pid, &context[CONTEXT_UNIT]) >= 0) {
b8cda92d 1109 if (!is_journald_crash((const char**) context)) {
92e92d71
ZJS
1110 /* OK, now we know it's not the journal, hence we can make use of it now. */
1111 log_set_target(LOG_TARGET_JOURNAL_OR_KMSG);
1112 log_open();
1113 }
3c171f0b 1114
c8091d92 1115 /* If this is PID 1 disable coredump collection, we'll unlikely be able to process it later on. */
b8cda92d 1116 if (is_pid1_crash((const char**) context)) {
c8091d92 1117 log_notice("Due to PID 1 having crashed coredump collection will now be turned off.");
e557b1a6 1118 disable_coredumps();
c8091d92
LP
1119 }
1120
92e92d71 1121 set_iovec_field(iovec, n_iovec, "COREDUMP_UNIT=", context[CONTEXT_UNIT]);
8c8549db 1122 }
803a3464 1123
9aa82023
ZJS
1124 if (cg_pid_get_user_unit(pid, &t) >= 0)
1125 set_iovec_field_free(iovec, n_iovec, "COREDUMP_USER_UNIT=", t);
3c171f0b 1126
9aa82023
ZJS
1127 /* The next few are mandatory */
1128 if (!set_iovec_field(iovec, n_iovec, "COREDUMP_PID=", context[CONTEXT_PID]))
1129 return log_oom();
3c171f0b 1130
9aa82023
ZJS
1131 if (!set_iovec_field(iovec, n_iovec, "COREDUMP_UID=", context[CONTEXT_UID]))
1132 return log_oom();
f5e04665 1133
9aa82023
ZJS
1134 if (!set_iovec_field(iovec, n_iovec, "COREDUMP_GID=", context[CONTEXT_GID]))
1135 return log_oom();
f5e04665 1136
9aa82023
ZJS
1137 if (!set_iovec_field(iovec, n_iovec, "COREDUMP_SIGNAL=", context[CONTEXT_SIGNAL]))
1138 return log_oom();
f5e04665 1139
9aa82023
ZJS
1140 if (!set_iovec_field(iovec, n_iovec, "COREDUMP_RLIMIT=", context[CONTEXT_RLIMIT]))
1141 return log_oom();
f5e04665 1142
f45b8015
JF
1143 if (!set_iovec_field(iovec, n_iovec, "COREDUMP_HOSTNAME=", context[CONTEXT_HOSTNAME]))
1144 return log_oom();
1145
ea5cc2a8 1146 if (!set_iovec_field(iovec, n_iovec, "COREDUMP_COMM=", context[CONTEXT_COMM]))
9aa82023 1147 return log_oom();
bdfd7b2c 1148
ea5cc2a8
ZJS
1149 if (context[CONTEXT_EXE] &&
1150 !set_iovec_field(iovec, n_iovec, "COREDUMP_EXE=", context[CONTEXT_EXE]))
9aa82023 1151 return log_oom();
f5e04665 1152
9aa82023
ZJS
1153 if (sd_pid_get_session(pid, &t) >= 0)
1154 set_iovec_field_free(iovec, n_iovec, "COREDUMP_SESSION=", t);
f5e04665 1155
a035f819 1156 if (sd_pid_get_owner_uid(pid, &owner_uid) >= 0) {
9aa82023 1157 r = asprintf(&t, "COREDUMP_OWNER_UID=" UID_FMT, owner_uid);
7de80bfe 1158 if (r > 0)
e6a7ec4b 1159 iovec[(*n_iovec)++] = IOVEC_MAKE_STRING(t);
f5e04665
LP
1160 }
1161
9aa82023
ZJS
1162 if (sd_pid_get_slice(pid, &t) >= 0)
1163 set_iovec_field_free(iovec, n_iovec, "COREDUMP_SLICE=", t);
f5e04665 1164
9aa82023
ZJS
1165 if (get_process_cmdline(pid, 0, false, &t) >= 0)
1166 set_iovec_field_free(iovec, n_iovec, "COREDUMP_CMDLINE=", t);
a035f819 1167
9aa82023
ZJS
1168 if (cg_pid_get_path_shifted(pid, NULL, &t) >= 0)
1169 set_iovec_field_free(iovec, n_iovec, "COREDUMP_CGROUP=", t);
a035f819 1170
9aa82023
ZJS
1171 if (compose_open_fds(pid, &t) >= 0)
1172 set_iovec_field_free(iovec, n_iovec, "COREDUMP_OPEN_FDS=", t);
3f132692
JF
1173
1174 p = procfs_file_alloca(pid, "status");
9aa82023
ZJS
1175 if (read_full_file(p, &t, NULL) >= 0)
1176 set_iovec_field_free(iovec, n_iovec, "COREDUMP_PROC_STATUS=", t);
3f132692
JF
1177
1178 p = procfs_file_alloca(pid, "maps");
9aa82023
ZJS
1179 if (read_full_file(p, &t, NULL) >= 0)
1180 set_iovec_field_free(iovec, n_iovec, "COREDUMP_PROC_MAPS=", t);
3f132692
JF
1181
1182 p = procfs_file_alloca(pid, "limits");
9aa82023
ZJS
1183 if (read_full_file(p, &t, NULL) >= 0)
1184 set_iovec_field_free(iovec, n_iovec, "COREDUMP_PROC_LIMITS=", t);
3f132692
JF
1185
1186 p = procfs_file_alloca(pid, "cgroup");
9aa82023
ZJS
1187 if (read_full_file(p, &t, NULL) >=0)
1188 set_iovec_field_free(iovec, n_iovec, "COREDUMP_PROC_CGROUP=", t);
3f132692 1189
d7032b1f 1190 p = procfs_file_alloca(pid, "mountinfo");
9aa82023
ZJS
1191 if (read_full_file(p, &t, NULL) >=0)
1192 set_iovec_field_free(iovec, n_iovec, "COREDUMP_PROC_MOUNTINFO=", t);
d7032b1f 1193
9aa82023
ZJS
1194 if (get_process_cwd(pid, &t) >= 0)
1195 set_iovec_field_free(iovec, n_iovec, "COREDUMP_CWD=", t);
3f132692
JF
1196
1197 if (get_process_root(pid, &t) >= 0) {
9aa82023
ZJS
1198 bool proc_self_root_is_slash;
1199
1200 proc_self_root_is_slash = strcmp(t, "/") == 0;
3f132692 1201
9aa82023 1202 set_iovec_field_free(iovec, n_iovec, "COREDUMP_ROOT=", t);
7ed03ce6
JF
1203
1204 /* If the process' root is "/", then there is a chance it has
1205 * mounted own root and hence being containerized. */
9aa82023
ZJS
1206 if (proc_self_root_is_slash && get_process_container_parent_cmdline(pid, &t) > 0)
1207 set_iovec_field_free(iovec, n_iovec, "COREDUMP_CONTAINER_CMDLINE=", t);
3f132692
JF
1208 }
1209
9aa82023
ZJS
1210 if (get_process_environ(pid, &t) >= 0)
1211 set_iovec_field_free(iovec, n_iovec, "COREDUMP_ENVIRON=", t);
1212
4600a396 1213 t = strjoin("COREDUMP_TIMESTAMP=", context[CONTEXT_TIMESTAMP], "000000");
9aa82023 1214 if (t)
e6a7ec4b 1215 iovec[(*n_iovec)++] = IOVEC_MAKE_STRING(t);
9aa82023 1216
d14bcb4e
LP
1217 if (safe_atoi(context[CONTEXT_SIGNAL], &signo) >= 0 && SIGNAL_VALID(signo))
1218 set_iovec_field(iovec, n_iovec, "COREDUMP_SIGNAL_NAME=SIG", signal_to_string(signo));
1219
92e92d71 1220 return 0; /* we successfully acquired all metadata */
9aa82023 1221}
3f132692 1222
9aa82023
ZJS
1223static int process_kernel(int argc, char* argv[]) {
1224
ea5cc2a8 1225 char* context[_CONTEXT_MAX] = {};
f45b8015 1226 struct iovec iovec[29 + SUBMIT_COREDUMP_FIELDS];
988e89ee 1227 size_t i, n_iovec, n_to_free = 0;
9aa82023
ZJS
1228 int r;
1229
988e89ee
ZJS
1230 log_debug("Processing coredump received from the kernel...");
1231
9aa82023 1232 if (argc < CONTEXT_COMM + 1) {
988e89ee 1233 log_error("Not enough arguments passed by the kernel (%i, expected %i).", argc - 1, CONTEXT_COMM + 1 - 1);
9aa82023 1234 return -EINVAL;
3f132692
JF
1235 }
1236
ea5cc2a8
ZJS
1237 context[CONTEXT_PID] = argv[1 + CONTEXT_PID];
1238 context[CONTEXT_UID] = argv[1 + CONTEXT_UID];
1239 context[CONTEXT_GID] = argv[1 + CONTEXT_GID];
1240 context[CONTEXT_SIGNAL] = argv[1 + CONTEXT_SIGNAL];
1241 context[CONTEXT_TIMESTAMP] = argv[1 + CONTEXT_TIMESTAMP];
1242 context[CONTEXT_RLIMIT] = argv[1 + CONTEXT_RLIMIT];
f45b8015 1243 context[CONTEXT_HOSTNAME] = argv[1 + CONTEXT_HOSTNAME];
9aa82023 1244
92e92d71
ZJS
1245 r = gather_pid_metadata(context, argv + 1 + CONTEXT_COMM, iovec, &n_to_free);
1246 if (r < 0)
86562420 1247 goto finish;
86562420 1248
988e89ee 1249 n_iovec = n_to_free;
f5e04665 1250
e6a7ec4b 1251 iovec[n_iovec++] = IOVEC_MAKE_STRING("MESSAGE_ID=" SD_MESSAGE_COREDUMP_STR);
4850d39a
LP
1252
1253 assert_cc(2 == LOG_CRIT);
e6a7ec4b 1254 iovec[n_iovec++] = IOVEC_MAKE_STRING("PRIORITY=2");
0dc5d23c 1255
3c171f0b 1256 assert(n_iovec <= ELEMENTSOF(iovec));
34c10968 1257
b8cda92d 1258 if (is_journald_crash((const char**) context) || is_pid1_crash((const char**) context))
92e92d71
ZJS
1259 r = submit_coredump((const char**) context,
1260 iovec, ELEMENTSOF(iovec), n_iovec,
1261 STDIN_FILENO);
1262 else
1263 r = send_iovec(iovec, n_iovec, STDIN_FILENO);
9aa82023
ZJS
1264
1265 finish:
1266 for (i = 0; i < n_to_free; i++)
1267 free(iovec[i].iov_base);
1268
92e92d71 1269 /* Those fields are allocated by gather_pid_metadata */
ea5cc2a8
ZJS
1270 free(context[CONTEXT_COMM]);
1271 free(context[CONTEXT_EXE]);
92e92d71 1272 free(context[CONTEXT_UNIT]);
ea5cc2a8 1273
9aa82023 1274 return r;
3c171f0b 1275}
34c10968 1276
988e89ee 1277static int process_backtrace(int argc, char *argv[]) {
ea5cc2a8
ZJS
1278 char *context[_CONTEXT_MAX] = {};
1279 _cleanup_free_ char *message = NULL;
5b45a160
ZJS
1280 _cleanup_free_ struct iovec *iovec = NULL;
1281 size_t n_iovec, n_allocated, n_to_free = 0, i;
988e89ee 1282 int r;
5b45a160
ZJS
1283 JournalImporter importer = {
1284 .fd = STDIN_FILENO,
1285 };
988e89ee
ZJS
1286
1287 log_debug("Processing backtrace on stdin...");
1288
1289 if (argc < CONTEXT_COMM + 1) {
1290 log_error("Not enough arguments passed (%i, expected %i).", argc - 1, CONTEXT_COMM + 1 - 1);
1291 return -EINVAL;
1292 }
1293
ea5cc2a8
ZJS
1294 context[CONTEXT_PID] = argv[2 + CONTEXT_PID];
1295 context[CONTEXT_UID] = argv[2 + CONTEXT_UID];
1296 context[CONTEXT_GID] = argv[2 + CONTEXT_GID];
1297 context[CONTEXT_SIGNAL] = argv[2 + CONTEXT_SIGNAL];
1298 context[CONTEXT_TIMESTAMP] = argv[2 + CONTEXT_TIMESTAMP];
1299 context[CONTEXT_RLIMIT] = argv[2 + CONTEXT_RLIMIT];
f45b8015 1300 context[CONTEXT_HOSTNAME] = argv[2 + CONTEXT_HOSTNAME];
988e89ee 1301
f45b8015
JF
1302 n_allocated = 34 + COREDUMP_STORAGE_EXTERNAL;
1303 /* 26 metadata, 2 static, +unknown input, 4 storage, rounded up */
5b45a160
ZJS
1304 iovec = new(struct iovec, n_allocated);
1305 if (!iovec)
1306 return log_oom();
1307
92e92d71 1308 r = gather_pid_metadata(context, argv + 2 + CONTEXT_COMM, iovec, &n_to_free);
988e89ee
ZJS
1309 if (r < 0)
1310 goto finish;
86562420
LP
1311 if (r > 0) {
1312 /* This was a special crash, and has already been processed. */
1313 r = 0;
1314 goto finish;
1315 }
5b45a160 1316 n_iovec = n_to_free;
988e89ee 1317
86562420 1318 for (;;) {
5b45a160
ZJS
1319 r = journal_importer_process_data(&importer);
1320 if (r < 0) {
1321 log_error_errno(r, "Failed to parse journal entry on stdin: %m");
1322 goto finish;
1323 }
d74dc4f2
ZJS
1324 if (r == 1 || /* complete entry */
1325 journal_importer_eof(&importer)) /* end of data */
5b45a160 1326 break;
988e89ee 1327 }
988e89ee 1328
5b45a160
ZJS
1329 if (!GREEDY_REALLOC(iovec, n_allocated, n_iovec + importer.iovw.count + 2))
1330 return log_oom();
1331
1332 if (journal_importer_eof(&importer)) {
1333 log_warning("Did not receive a full journal entry on stdin, ignoring message sent by reporter");
988e89ee 1334
ea5cc2a8
ZJS
1335 message = strjoin("MESSAGE=Process ", context[CONTEXT_PID],
1336 " (", context[CONTEXT_COMM], ")"
5b45a160
ZJS
1337 " of user ", context[CONTEXT_UID],
1338 " failed with ", context[CONTEXT_SIGNAL]);
1339 if (!message) {
1340 r = log_oom();
1341 goto finish;
1342 }
e6a7ec4b 1343 iovec[n_iovec++] = IOVEC_MAKE_STRING(message);
5b45a160
ZJS
1344 } else {
1345 for (i = 0; i < importer.iovw.count; i++)
1346 iovec[n_iovec++] = importer.iovw.iovec[i];
1347 }
988e89ee 1348
e6a7ec4b 1349 iovec[n_iovec++] = IOVEC_MAKE_STRING("MESSAGE_ID=" SD_MESSAGE_BACKTRACE_STR);
988e89ee 1350 assert_cc(2 == LOG_CRIT);
e6a7ec4b 1351 iovec[n_iovec++] = IOVEC_MAKE_STRING("PRIORITY=2");
988e89ee 1352
5b45a160 1353 assert(n_iovec <= n_allocated);
988e89ee
ZJS
1354
1355 r = sd_journal_sendv(iovec, n_iovec);
1356 if (r < 0)
1357 log_error_errno(r, "Failed to log backtrace: %m");
1358
1359 finish:
1360 for (i = 0; i < n_to_free; i++)
1361 free(iovec[i].iov_base);
1362
92e92d71 1363 /* Those fields are allocated by gather_pid_metadata */
ea5cc2a8
ZJS
1364 free(context[CONTEXT_COMM]);
1365 free(context[CONTEXT_EXE]);
92e92d71 1366 free(context[CONTEXT_UNIT]);
ea5cc2a8 1367
988e89ee
ZJS
1368 return r;
1369}
1370
3c171f0b
LP
1371int main(int argc, char *argv[]) {
1372 int r;
fee80f69 1373
9aa82023
ZJS
1374 /* First, log to a safe place, since we don't know what crashed and it might
1375 * be journald which we'd rather not log to then. */
8d4e028f 1376
3c171f0b
LP
1377 log_set_target(LOG_TARGET_KMSG);
1378 log_open();
8d4e028f 1379
3c171f0b
LP
1380 /* Make sure we never enter a loop */
1381 (void) prctl(PR_SET_DUMPABLE, 0);
8d4e028f 1382
3c171f0b
LP
1383 /* Ignore all parse errors */
1384 (void) parse_config();
fee80f69 1385
3c171f0b
LP
1386 log_debug("Selected storage '%s'.", coredump_storage_to_string(arg_storage));
1387 log_debug("Selected compression %s.", yes_no(arg_compress));
fee80f69 1388
3c171f0b
LP
1389 r = sd_listen_fds(false);
1390 if (r < 0) {
1391 log_error_errno(r, "Failed to determine number of file descriptor: %m");
1392 goto finish;
fee80f69
LP
1393 }
1394
9aa82023
ZJS
1395 /* If we got an fd passed, we are running in coredumpd mode. Otherwise we
1396 * are invoked from the kernel as coredump handler. */
988e89ee
ZJS
1397 if (r == 0) {
1398 if (streq_ptr(argv[1], "--backtrace"))
1399 r = process_backtrace(argc, argv);
1400 else
1401 r = process_kernel(argc, argv);
1402 } else if (r == 1)
3c171f0b
LP
1403 r = process_socket(SD_LISTEN_FDS_START);
1404 else {
1405 log_error("Received unexpected number of file descriptors.");
1406 r = -EINVAL;
1407 }
f5e04665
LP
1408
1409finish:
f5e04665
LP
1410 return r < 0 ? EXIT_FAILURE : EXIT_SUCCESS;
1411}