]> git.ipfire.org Git - thirdparty/systemd.git/blame - src/coredump/coredump.c
update NEWS contributors list again
[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"
0c773903 37#include "missing.h"
4f5dd394 38#include "mkdir.h"
6bedfcbb 39#include "parse-util.h"
0b452006 40#include "process-util.h"
d14bcb4e 41#include "signal-util.h"
3c171f0b 42#include "socket-util.h"
4f5dd394
LP
43#include "special.h"
44#include "stacktrace.h"
8b43440b 45#include "string-table.h"
07630cea 46#include "string-util.h"
4f5dd394 47#include "strv.h"
b1d4f8e1 48#include "user-util.h"
4f5dd394 49#include "util.h"
34727273 50
34c10968 51/* The maximum size up to which we process coredumps */
59f448cf 52#define PROCESS_SIZE_MAX ((uint64_t) (2LLU*1024LLU*1024LLU*1024LLU))
34c10968 53
bdfd7b2c 54/* The maximum size up to which we leave the coredump around on disk */
34c10968
LP
55#define EXTERNAL_SIZE_MAX PROCESS_SIZE_MAX
56
bdfd7b2c 57/* The maximum size up to which we store the coredump in the journal */
34c10968 58#define JOURNAL_SIZE_MAX ((size_t) (767LU*1024LU*1024LU))
f5e04665 59
c4aa09b0 60/* Make sure to not make this larger than the maximum journal entry
874bc134
ZJS
61 * size. See DATA_SIZE_MAX in journald-native.c. */
62assert_cc(JOURNAL_SIZE_MAX <= DATA_SIZE_MAX);
f5e04665
LP
63
64enum {
ea5cc2a8
ZJS
65 /* We use this as array indexes for a couple of special fields we use for
66 * naming coredump files, and attaching xattrs, and for indexing argv[].
67
68 * Our pattern for man:systectl(1) kernel.core_pattern is such that the
69 * kernel passes fields until CONTEXT_RLIMIT as arguments in argv[]. After
70 * that it gets complicated: the kernel passes "comm" as one or more fields
71 * starting at index CONTEXT_COMM (in other words, full "comm" is under index
72 * CONTEXT_COMM when it does not contain spaces, which is the common
73 * case). This mapping is not reversible, so we prefer to retrieve "comm"
74 * from /proc. We only fall back to argv[CONTEXT_COMM...] when that fails.
75 *
76 * In the internal context[] array, fields before CONTEXT_COMM are the
77 * strings from argv[], so they should not be freed. The strings at indices
78 * CONTEXT_COMM and higher are allocated by us and should be freed at the
79 * end.
80 */
3c171f0b
LP
81 CONTEXT_PID,
82 CONTEXT_UID,
83 CONTEXT_GID,
84 CONTEXT_SIGNAL,
85 CONTEXT_TIMESTAMP,
bdfd7b2c 86 CONTEXT_RLIMIT,
f45b8015 87 CONTEXT_HOSTNAME,
3c171f0b
LP
88 CONTEXT_COMM,
89 CONTEXT_EXE,
92e92d71 90 CONTEXT_UNIT,
3c171f0b 91 _CONTEXT_MAX
f5e04665
LP
92};
93
34c10968
LP
94typedef enum CoredumpStorage {
95 COREDUMP_STORAGE_NONE,
96 COREDUMP_STORAGE_EXTERNAL,
97 COREDUMP_STORAGE_JOURNAL,
34c10968
LP
98 _COREDUMP_STORAGE_MAX,
99 _COREDUMP_STORAGE_INVALID = -1
100} CoredumpStorage;
101
34c10968
LP
102static const char* const coredump_storage_table[_COREDUMP_STORAGE_MAX] = {
103 [COREDUMP_STORAGE_NONE] = "none",
104 [COREDUMP_STORAGE_EXTERNAL] = "external",
105 [COREDUMP_STORAGE_JOURNAL] = "journal",
34c10968
LP
106};
107
108DEFINE_PRIVATE_STRING_TABLE_LOOKUP(coredump_storage, CoredumpStorage);
8c9571d0 109static DEFINE_CONFIG_PARSE_ENUM(config_parse_coredump_storage, coredump_storage, CoredumpStorage, "Failed to parse storage setting");
34727273
ZJS
110
111static CoredumpStorage arg_storage = COREDUMP_STORAGE_EXTERNAL;
8c9571d0 112static bool arg_compress = true;
59f448cf
LP
113static uint64_t arg_process_size_max = PROCESS_SIZE_MAX;
114static uint64_t arg_external_size_max = EXTERNAL_SIZE_MAX;
6e2b4a69 115static uint64_t arg_journal_size_max = JOURNAL_SIZE_MAX;
59f448cf
LP
116static uint64_t arg_keep_free = (uint64_t) -1;
117static uint64_t arg_max_use = (uint64_t) -1;
34c10968
LP
118
119static int parse_config(void) {
34c10968 120 static const ConfigTableItem items[] = {
8c9571d0
LP
121 { "Coredump", "Storage", config_parse_coredump_storage, 0, &arg_storage },
122 { "Coredump", "Compress", config_parse_bool, 0, &arg_compress },
59f448cf
LP
123 { "Coredump", "ProcessSizeMax", config_parse_iec_uint64, 0, &arg_process_size_max },
124 { "Coredump", "ExternalSizeMax", config_parse_iec_uint64, 0, &arg_external_size_max },
8c9571d0 125 { "Coredump", "JournalSizeMax", config_parse_iec_size, 0, &arg_journal_size_max },
59f448cf
LP
126 { "Coredump", "KeepFree", config_parse_iec_uint64, 0, &arg_keep_free },
127 { "Coredump", "MaxUse", config_parse_iec_uint64, 0, &arg_max_use },
34c10968
LP
128 {}
129 };
130
43688c49 131 return config_parse_many_nulstr(PKGSYSCONFDIR "/coredump.conf",
da412854
YW
132 CONF_PATHS_NULSTR("systemd/coredump.conf.d"),
133 "Coredump\0",
134 config_item_table_lookup, items,
bcde742e 135 CONFIG_PARSE_WARN, NULL);
34c10968
LP
136}
137
73a99163 138static inline uint64_t storage_size_max(void) {
ee0449fd
ZJS
139 if (arg_storage == COREDUMP_STORAGE_EXTERNAL)
140 return arg_external_size_max;
141 if (arg_storage == COREDUMP_STORAGE_JOURNAL)
142 return arg_journal_size_max;
143 assert(arg_storage == COREDUMP_STORAGE_NONE);
144 return 0;
73a99163
ZJS
145}
146
34c10968
LP
147static int fix_acl(int fd, uid_t uid) {
148
349cc4a5 149#if HAVE_ACL
34c10968
LP
150 _cleanup_(acl_freep) acl_t acl = NULL;
151 acl_entry_t entry;
152 acl_permset_t permset;
709f6e46 153 int r;
34c10968 154
b59233e6
LP
155 assert(fd >= 0);
156
05fd2156 157 if (uid_is_system(uid) || uid_is_dynamic(uid) || uid == UID_NOBODY)
34c10968
LP
158 return 0;
159
160 /* Make sure normal users can read (but not write or delete)
161 * their own coredumps */
162
163 acl = acl_get_fd(fd);
4a62c710
MS
164 if (!acl)
165 return log_error_errno(errno, "Failed to get ACL: %m");
34c10968
LP
166
167 if (acl_create_entry(&acl, &entry) < 0 ||
168 acl_set_tag_type(entry, ACL_USER) < 0 ||
d710aaf7
ZJS
169 acl_set_qualifier(entry, &uid) < 0)
170 return log_error_errno(errno, "Failed to patch ACL: %m");
34c10968
LP
171
172 if (acl_get_permset(entry, &permset) < 0 ||
709f6e46
MS
173 acl_add_perm(permset, ACL_READ) < 0)
174 return log_warning_errno(errno, "Failed to patch ACL: %m");
175
176 r = calc_acl_mask_if_needed(&acl);
177 if (r < 0)
178 return log_warning_errno(r, "Failed to patch ACL: %m");
34c10968 179
4a62c710
MS
180 if (acl_set_fd(fd, acl) < 0)
181 return log_error_errno(errno, "Failed to apply ACL: %m");
34c10968
LP
182#endif
183
184 return 0;
185}
186
3c171f0b 187static int fix_xattr(int fd, const char *context[_CONTEXT_MAX]) {
0cd77f97 188
3c171f0b
LP
189 static const char * const xattrs[_CONTEXT_MAX] = {
190 [CONTEXT_PID] = "user.coredump.pid",
191 [CONTEXT_UID] = "user.coredump.uid",
192 [CONTEXT_GID] = "user.coredump.gid",
193 [CONTEXT_SIGNAL] = "user.coredump.signal",
194 [CONTEXT_TIMESTAMP] = "user.coredump.timestamp",
6d337300 195 [CONTEXT_RLIMIT] = "user.coredump.rlimit",
f45b8015 196 [CONTEXT_HOSTNAME] = "user.coredump.hostname",
3c171f0b
LP
197 [CONTEXT_COMM] = "user.coredump.comm",
198 [CONTEXT_EXE] = "user.coredump.exe",
0cd77f97
LP
199 };
200
34c10968 201 int r = 0;
0cd77f97 202 unsigned i;
34c10968 203
b59233e6
LP
204 assert(fd >= 0);
205
1eef15b1 206 /* Attach some metadata to coredumps via extended
34c10968
LP
207 * attributes. Just because we can. */
208
3c171f0b 209 for (i = 0; i < _CONTEXT_MAX; i++) {
1eef15b1
ZJS
210 int k;
211
3c171f0b 212 if (isempty(context[i]) || !xattrs[i])
0cd77f97 213 continue;
34c10968 214
3c171f0b 215 k = fsetxattr(fd, xattrs[i], context[i], strlen(context[i]), XATTR_CREATE);
1eef15b1 216 if (k < 0 && r == 0)
34c10968 217 r = -errno;
0cd77f97 218 }
34c10968
LP
219
220 return r;
221}
222
b0b21dce 223#define filename_escape(s) xescape((s), "./ ")
34c10968 224
0c773903
EV
225static inline const char *coredump_tmpfile_name(const char *s) {
226 return s ? s : "(unnamed temporary file)";
227}
228
b59233e6
LP
229static int fix_permissions(
230 int fd,
231 const char *filename,
232 const char *target,
3c171f0b 233 const char *context[_CONTEXT_MAX],
b59233e6
LP
234 uid_t uid) {
235
03532f0a
LP
236 int r;
237
b59233e6 238 assert(fd >= 0);
b59233e6 239 assert(target);
3c171f0b 240 assert(context);
cfd652ed
ZJS
241
242 /* Ignore errors on these */
3c171f0b
LP
243 (void) fchmod(fd, 0640);
244 (void) fix_acl(fd, uid);
245 (void) fix_xattr(fd, context);
cfd652ed 246
4a62c710 247 if (fsync(fd) < 0)
0c773903 248 return log_error_errno(errno, "Failed to sync coredump %s: %m", coredump_tmpfile_name(filename));
cfd652ed 249
8ac2f74f
LP
250 (void) fsync_directory_of_file(fd);
251
03532f0a
LP
252 r = link_tmpfile(fd, filename, target);
253 if (r < 0)
254 return log_error_errno(r, "Failed to move coredump %s into place: %m", target);
cfd652ed
ZJS
255
256 return 0;
257}
258
59f448cf 259static int maybe_remove_external_coredump(const char *filename, uint64_t size) {
cfd652ed 260
b59233e6 261 /* Returns 1 if might remove, 0 if will not remove, < 0 on error. */
cfd652ed 262
fc6cec86 263 if (arg_storage == COREDUMP_STORAGE_EXTERNAL &&
cfd652ed
ZJS
264 size <= arg_external_size_max)
265 return 0;
266
267 if (!filename)
268 return 1;
269
4a62c710
MS
270 if (unlink(filename) < 0 && errno != ENOENT)
271 return log_error_errno(errno, "Failed to unlink %s: %m", filename);
cfd652ed
ZJS
272
273 return 1;
274}
275
3c171f0b 276static int make_filename(const char *context[_CONTEXT_MAX], char **ret) {
b59233e6 277 _cleanup_free_ char *c = NULL, *u = NULL, *p = NULL, *t = NULL;
a7f7d1bd 278 sd_id128_t boot = {};
34c10968
LP
279 int r;
280
3c171f0b 281 assert(context);
34c10968 282
3c171f0b 283 c = filename_escape(context[CONTEXT_COMM]);
34c10968 284 if (!c)
b59233e6 285 return -ENOMEM;
34c10968 286
3c171f0b 287 u = filename_escape(context[CONTEXT_UID]);
0dc5d23c 288 if (!u)
b59233e6 289 return -ENOMEM;
34c10968
LP
290
291 r = sd_id128_get_boot(&boot);
b59233e6 292 if (r < 0)
34c10968 293 return r;
34c10968 294
3c171f0b 295 p = filename_escape(context[CONTEXT_PID]);
b59233e6
LP
296 if (!p)
297 return -ENOMEM;
298
3c171f0b 299 t = filename_escape(context[CONTEXT_TIMESTAMP]);
b59233e6
LP
300 if (!t)
301 return -ENOMEM;
302
303 if (asprintf(ret,
0dc5d23c 304 "/var/lib/systemd/coredump/core.%s.%s." SD_ID128_FORMAT_STR ".%s.%s000000",
34c10968 305 c,
0dc5d23c 306 u,
34c10968
LP
307 SD_ID128_FORMAT_VAL(boot),
308 p,
b59233e6
LP
309 t) < 0)
310 return -ENOMEM;
311
312 return 0;
313}
314
315static int save_external_coredump(
3c171f0b
LP
316 const char *context[_CONTEXT_MAX],
317 int input_fd,
b59233e6 318 char **ret_filename,
5f3e0a74
HW
319 int *ret_node_fd,
320 int *ret_data_fd,
0cd4e913 321 uint64_t *ret_size,
cc4419ed 322 bool *ret_truncated) {
b59233e6
LP
323
324 _cleanup_free_ char *fn = NULL, *tmp = NULL;
325 _cleanup_close_ int fd = -1;
ee0449fd 326 uint64_t rlimit, process_limit, max_size;
b59233e6 327 struct stat st;
3c171f0b 328 uid_t uid;
b59233e6
LP
329 int r;
330
3c171f0b 331 assert(context);
b59233e6 332 assert(ret_filename);
5f3e0a74
HW
333 assert(ret_node_fd);
334 assert(ret_data_fd);
b59233e6
LP
335 assert(ret_size);
336
3c171f0b
LP
337 r = parse_uid(context[CONTEXT_UID], &uid);
338 if (r < 0)
339 return log_error_errno(r, "Failed to parse UID: %m");
340
bdfd7b2c
LP
341 r = safe_atou64(context[CONTEXT_RLIMIT], &rlimit);
342 if (r < 0)
343 return log_error_errno(r, "Failed to parse resource limit: %s", context[CONTEXT_RLIMIT]);
6998b540
ZJS
344 if (rlimit < page_size()) {
345 /* Is coredumping disabled? Then don't bother saving/processing the coredump.
346 * Anything below PAGE_SIZE cannot give a readable coredump (the kernel uses
347 * ELF_EXEC_PAGESIZE which is not easily accessible, but is usually the same as PAGE_SIZE. */
73a99163
ZJS
348 log_info("Resource limits disable core dumping for process %s (%s).",
349 context[CONTEXT_PID], context[CONTEXT_COMM]);
bdfd7b2c
LP
350 return -EBADSLT;
351 }
352
ee0449fd
ZJS
353 process_limit = MAX(arg_process_size_max, storage_size_max());
354 if (process_limit == 0) {
355 log_debug("Limits for coredump processing and storage are both 0, not dumping core.");
356 return -EBADSLT;
357 }
358
bdfd7b2c 359 /* Never store more than the process configured, or than we actually shall keep or process */
ee0449fd 360 max_size = MIN(rlimit, process_limit);
bdfd7b2c 361
3c171f0b 362 r = make_filename(context, &fn);
23bbb0de
MS
363 if (r < 0)
364 return log_error_errno(r, "Failed to determine coredump file name: %m");
34c10968 365
d2e54fae 366 mkdir_p_label("/var/lib/systemd/coredump", 0755);
803a3464 367
03532f0a 368 fd = open_tmpfile_linkable(fn, O_RDWR|O_CLOEXEC, &tmp);
4a62c710 369 if (fd < 0)
03532f0a 370 return log_error_errno(fd, "Failed to create temporary file for coredump %s: %m", fn);
803a3464 371
1c876927 372 r = copy_bytes(input_fd, fd, max_size, 0);
73a99163
ZJS
373 if (r < 0) {
374 log_error_errno(r, "Cannot store coredump of %s (%s): %m", context[CONTEXT_PID], context[CONTEXT_COMM]);
93240d3a 375 goto fail;
0cd4e913 376 }
cc4419ed
ZJS
377 *ret_truncated = r == 1;
378 if (*ret_truncated)
73a99163
ZJS
379 log_struct(LOG_INFO,
380 LOG_MESSAGE("Core file was truncated to %zu bytes.", max_size),
381 "SIZE_LIMIT=%zu", max_size,
a1230ff9 382 "MESSAGE_ID=" SD_MESSAGE_TRUNCATED_CORE_STR);
803a3464 383
34c10968 384 if (fstat(fd, &st) < 0) {
73a99163 385 log_error_errno(errno, "Failed to fstat core file %s: %m", coredump_tmpfile_name(tmp));
34c10968
LP
386 goto fail;
387 }
388
7849c2ac 389 if (lseek(fd, 0, SEEK_SET) == (off_t) -1) {
0c773903 390 log_error_errno(errno, "Failed to seek on %s: %m", coredump_tmpfile_name(tmp));
b59233e6 391 goto fail;
7849c2ac
TA
392 }
393
349cc4a5 394#if HAVE_XZ || HAVE_LZ4
cfd652ed 395 /* If we will remove the coredump anyway, do not compress. */
6e9ef603 396 if (arg_compress && !maybe_remove_external_coredump(NULL, st.st_size)) {
cfd652ed 397
b59233e6
LP
398 _cleanup_free_ char *fn_compressed = NULL, *tmp_compressed = NULL;
399 _cleanup_close_ int fd_compressed = -1;
cfd652ed 400
d89c8fdf 401 fn_compressed = strappend(fn, COMPRESSED_EXT);
b59233e6 402 if (!fn_compressed) {
d89c8fdf 403 log_oom();
cfd652ed
ZJS
404 goto uncompressed;
405 }
406
03532f0a
LP
407 fd_compressed = open_tmpfile_linkable(fn_compressed, O_RDWR|O_CLOEXEC, &tmp_compressed);
408 if (fd_compressed < 0) {
409 log_error_errno(fd_compressed, "Failed to create temporary file for coredump %s: %m", fn_compressed);
b59233e6 410 goto uncompressed;
03532f0a 411 }
cfd652ed 412
d89c8fdf 413 r = compress_stream(fd, fd_compressed, -1);
b59233e6 414 if (r < 0) {
0c773903 415 log_error_errno(r, "Failed to compress %s: %m", coredump_tmpfile_name(tmp_compressed));
b59233e6
LP
416 goto fail_compressed;
417 }
418
3c171f0b 419 r = fix_permissions(fd_compressed, tmp_compressed, fn_compressed, context, uid);
cfd652ed 420 if (r < 0)
b59233e6
LP
421 goto fail_compressed;
422
423 /* OK, this worked, we can get rid of the uncompressed version now */
0c773903
EV
424 if (tmp)
425 unlink_noerrno(tmp);
cfd652ed 426
1cc6c93a
YW
427 *ret_filename = TAKE_PTR(fn_compressed); /* compressed */
428 *ret_node_fd = TAKE_FD(fd_compressed); /* compressed */
429 *ret_data_fd = TAKE_FD(fd); /* uncompressed */
59f448cf 430 *ret_size = (uint64_t) st.st_size; /* uncompressed */
cfd652ed 431
cfd652ed
ZJS
432 return 0;
433
b59233e6 434 fail_compressed:
0c773903
EV
435 if (tmp_compressed)
436 (void) unlink(tmp_compressed);
34c10968 437 }
cfd652ed
ZJS
438
439uncompressed:
3b1a55e1 440#endif
5f3e0a74 441
3c171f0b 442 r = fix_permissions(fd, tmp, fn, context, uid);
cfd652ed
ZJS
443 if (r < 0)
444 goto fail;
34c10968 445
1cc6c93a
YW
446 *ret_filename = TAKE_PTR(fn);
447 *ret_data_fd = TAKE_FD(fd);
5f3e0a74 448 *ret_node_fd = -1;
59f448cf 449 *ret_size = (uint64_t) st.st_size;
34c10968 450
34c10968
LP
451 return 0;
452
453fail:
0c773903
EV
454 if (tmp)
455 (void) unlink(tmp);
34c10968
LP
456 return r;
457}
458
459static int allocate_journal_field(int fd, size_t size, char **ret, size_t *ret_size) {
460 _cleanup_free_ char *field = NULL;
461 ssize_t n;
462
8d4e028f 463 assert(fd >= 0);
34c10968
LP
464 assert(ret);
465 assert(ret_size);
466
4a62c710
MS
467 if (lseek(fd, 0, SEEK_SET) == (off_t) -1)
468 return log_warning_errno(errno, "Failed to seek: %m");
803a3464 469
34c10968
LP
470 field = malloc(9 + size);
471 if (!field) {
cfd652ed 472 log_warning("Failed to allocate memory for coredump, coredump will not be stored.");
34c10968
LP
473 return -ENOMEM;
474 }
475
476 memcpy(field, "COREDUMP=", 9);
477
478 n = read(fd, field + 9, size);
23bbb0de
MS
479 if (n < 0)
480 return log_error_errno((int) n, "Failed to read core data: %m");
34c10968
LP
481 if ((size_t) n < size) {
482 log_error("Core data too short.");
483 return -EIO;
484 }
485
1cc6c93a 486 *ret = TAKE_PTR(field);
34c10968
LP
487 *ret_size = size + 9;
488
34c10968
LP
489 return 0;
490}
803a3464 491
3f132692
JF
492/* Joins /proc/[pid]/fd/ and /proc/[pid]/fdinfo/ into the following lines:
493 * 0:/dev/pts/23
494 * pos: 0
495 * flags: 0100002
496 *
497 * 1:/dev/pts/23
498 * pos: 0
499 * flags: 0100002
500 *
501 * 2:/dev/pts/23
502 * pos: 0
503 * flags: 0100002
504 * EOF
505 */
506static int compose_open_fds(pid_t pid, char **open_fds) {
4d84bc2f
LP
507 _cleanup_closedir_ DIR *proc_fd_dir = NULL;
508 _cleanup_close_ int proc_fdinfo_fd = -1;
509 _cleanup_free_ char *buffer = NULL;
3f132692 510 _cleanup_fclose_ FILE *stream = NULL;
59059b4a 511 const char *fddelim = "", *path;
3f132692 512 struct dirent *dent = NULL;
4d84bc2f 513 size_t size = 0;
3f132692
JF
514 int r = 0;
515
516 assert(pid >= 0);
517 assert(open_fds != NULL);
518
59059b4a 519 path = procfs_file_alloca(pid, "fd");
3f132692 520 proc_fd_dir = opendir(path);
59059b4a
ZJS
521 if (!proc_fd_dir)
522 return -errno;
3f132692 523
4d84bc2f 524 proc_fdinfo_fd = openat(dirfd(proc_fd_dir), "../fdinfo", O_DIRECTORY|O_NOFOLLOW|O_CLOEXEC|O_PATH);
59059b4a
ZJS
525 if (proc_fdinfo_fd < 0)
526 return -errno;
3f132692 527
4d84bc2f 528 stream = open_memstream(&buffer, &size);
3f132692
JF
529 if (!stream)
530 return -ENOMEM;
531
0d536673
LP
532 (void) __fsetlocking(stream, FSETLOCKING_BYCALLER);
533
4d84bc2f 534 FOREACH_DIRENT(dent, proc_fd_dir, return -errno) {
3f132692 535 _cleanup_fclose_ FILE *fdinfo = NULL;
4d84bc2f 536 _cleanup_free_ char *fdname = NULL;
59059b4a 537 char line[LINE_MAX];
4d84bc2f 538 int fd;
3f132692 539
59059b4a 540 r = readlinkat_malloc(dirfd(proc_fd_dir), dent->d_name, &fdname);
3f132692
JF
541 if (r < 0)
542 return r;
543
544 fprintf(stream, "%s%s:%s\n", fddelim, dent->d_name, fdname);
545 fddelim = "\n";
546
547 /* Use the directory entry from /proc/[pid]/fd with /proc/[pid]/fdinfo */
59059b4a
ZJS
548 fd = openat(proc_fdinfo_fd, dent->d_name, O_NOFOLLOW|O_CLOEXEC|O_RDONLY);
549 if (fd < 0)
3f132692
JF
550 continue;
551
59059b4a 552 fdinfo = fdopen(fd, "re");
234519ae 553 if (!fdinfo) {
0d536673 554 safe_close(fd);
3f132692 555 continue;
59059b4a 556 }
3f132692 557
4d84bc2f 558 FOREACH_LINE(line, fdinfo, break) {
0d536673 559 fputs(line, stream);
4d84bc2f
LP
560 if (!endswith(line, "\n"))
561 fputc('\n', stream);
562 }
3f132692
JF
563 }
564
4d84bc2f 565 errno = 0;
74ca738f 566 stream = safe_fclose(stream);
4d84bc2f 567
b3267152 568 if (errno > 0)
4d84bc2f
LP
569 return -errno;
570
ae2a15bc 571 *open_fds = TAKE_PTR(buffer);
4d84bc2f 572
3f132692
JF
573 return 0;
574}
575
7ed03ce6
JF
576static int get_process_ns(pid_t pid, const char *namespace, ino_t *ns) {
577 const char *p;
578 struct stat stbuf;
579 _cleanup_close_ int proc_ns_dir_fd;
580
581 p = procfs_file_alloca(pid, "ns");
582
583 proc_ns_dir_fd = open(p, O_DIRECTORY | O_CLOEXEC | O_RDONLY);
584 if (proc_ns_dir_fd < 0)
585 return -errno;
586
587 if (fstatat(proc_ns_dir_fd, namespace, &stbuf, /* flags */0) < 0)
588 return -errno;
589
590 *ns = stbuf.st_ino;
591 return 0;
592}
593
594static int get_mount_namespace_leader(pid_t pid, pid_t *container_pid) {
595 pid_t cpid = pid, ppid = 0;
596 ino_t proc_mntns;
597 int r = 0;
598
599 r = get_process_ns(pid, "mnt", &proc_mntns);
600 if (r < 0)
601 return r;
602
aa7530d6 603 for (;;) {
7ed03ce6
JF
604 ino_t parent_mntns;
605
606 r = get_process_ppid(cpid, &ppid);
607 if (r < 0)
608 return r;
609
610 r = get_process_ns(ppid, "mnt", &parent_mntns);
611 if (r < 0)
612 return r;
613
614 if (proc_mntns != parent_mntns)
615 break;
616
617 if (ppid == 1)
618 return -ENOENT;
619
620 cpid = ppid;
621 }
622
623 *container_pid = ppid;
624 return 0;
625}
626
627/* Returns 1 if the parent was found.
628 * Returns 0 if there is not a process we can call the pid's
629 * container parent (the pid's process isn't 'containerized').
630 * Returns a negative number on errors.
631 */
632static int get_process_container_parent_cmdline(pid_t pid, char** cmdline) {
633 int r = 0;
634 pid_t container_pid;
635 const char *proc_root_path;
636 struct stat root_stat, proc_root_stat;
637
638 /* To compare inodes of / and /proc/[pid]/root */
639 if (stat("/", &root_stat) < 0)
640 return -errno;
641
642 proc_root_path = procfs_file_alloca(pid, "root");
643 if (stat(proc_root_path, &proc_root_stat) < 0)
644 return -errno;
645
646 /* The process uses system root. */
647 if (proc_root_stat.st_ino == root_stat.st_ino) {
648 *cmdline = NULL;
649 return 0;
650 }
651
652 r = get_mount_namespace_leader(pid, &container_pid);
653 if (r < 0)
654 return r;
655
d3cba4ea
EV
656 r = get_process_cmdline(container_pid, 0, false, cmdline);
657 if (r < 0)
658 return r;
659
660 return 1;
7ed03ce6
JF
661}
662
3c171f0b
LP
663static int change_uid_gid(const char *context[]) {
664 uid_t uid;
665 gid_t gid;
666 int r;
34c10968 667
3c171f0b
LP
668 r = parse_uid(context[CONTEXT_UID], &uid);
669 if (r < 0)
670 return r;
8c8549db 671
888e378d
LP
672 if (uid <= SYSTEM_UID_MAX) {
673 const char *user = "systemd-coredump";
674
675 r = get_user_creds(&user, &uid, &gid, NULL, NULL);
676 if (r < 0) {
677 log_warning_errno(r, "Cannot resolve %s user. Proceeding to dump core as root: %m", user);
678 uid = gid = 0;
679 }
680 } else {
681 r = parse_gid(context[CONTEXT_GID], &gid);
682 if (r < 0)
683 return r;
684 }
3c171f0b
LP
685
686 return drop_privileges(uid, gid, 0);
687}
8c8549db 688
b8cda92d
LP
689static bool is_journald_crash(const char *context[_CONTEXT_MAX]) {
690 assert(context);
691
692 return streq_ptr(context[CONTEXT_UNIT], SPECIAL_JOURNALD_SERVICE);
693}
694
695static bool is_pid1_crash(const char *context[_CONTEXT_MAX]) {
696 assert(context);
697
698 return streq_ptr(context[CONTEXT_UNIT], SPECIAL_INIT_SCOPE) ||
699 streq_ptr(context[CONTEXT_PID], "1");
700}
701
0cd4e913
ZJS
702#define SUBMIT_COREDUMP_FIELDS 4
703
3c171f0b
LP
704static int submit_coredump(
705 const char *context[_CONTEXT_MAX],
706 struct iovec *iovec,
707 size_t n_iovec_allocated,
708 size_t n_iovec,
709 int input_fd) {
34c10968 710
5f3e0a74 711 _cleanup_close_ int coredump_fd = -1, coredump_node_fd = -1;
3c171f0b 712 _cleanup_free_ char *core_message = NULL, *filename = NULL, *coredump_data = NULL;
a5ca3649 713 uint64_t coredump_size = UINT64_MAX;
92e92d71 714 bool truncated = false, journald_crash;
3c171f0b 715 int r;
f5e04665 716
3c171f0b
LP
717 assert(context);
718 assert(iovec);
0cd4e913 719 assert(n_iovec_allocated >= n_iovec + SUBMIT_COREDUMP_FIELDS);
3c171f0b 720 assert(input_fd >= 0);
f5e04665 721
b8cda92d 722 journald_crash = is_journald_crash(context);
92e92d71 723
3c171f0b
LP
724 /* Vacuum before we write anything again */
725 (void) coredump_vacuum(-1, arg_keep_free, arg_max_use);
803a3464 726
3c171f0b 727 /* Always stream the coredump to disk, if that's possible */
0cd4e913
ZJS
728 r = save_external_coredump(context, input_fd,
729 &filename, &coredump_node_fd, &coredump_fd, &coredump_size, &truncated);
3c171f0b
LP
730 if (r < 0)
731 /* Skip whole core dumping part */
732 goto log;
733
734 /* If we don't want to keep the coredump on disk, remove it now, as later on we will lack the privileges for
735 * it. However, we keep the fd to it, so that we can still process it and log it. */
736 r = maybe_remove_external_coredump(filename, coredump_size);
737 if (r < 0)
738 return r;
739 if (r == 0) {
740 const char *coredump_filename;
741
742 coredump_filename = strjoina("COREDUMP_FILENAME=", filename);
e6a7ec4b 743 iovec[n_iovec++] = IOVEC_MAKE_STRING(coredump_filename);
6e9ef603 744 } else if (arg_storage == COREDUMP_STORAGE_EXTERNAL)
5206a724 745 log_info("The core will not be stored: size %"PRIu64" is greater than %"PRIu64" (the configured maximum)",
6e9ef603 746 coredump_size, arg_external_size_max);
f5e04665 747
3c171f0b
LP
748 /* Vacuum again, but exclude the coredump we just created */
749 (void) coredump_vacuum(coredump_node_fd >= 0 ? coredump_node_fd : coredump_fd, arg_keep_free, arg_max_use);
8c9571d0 750
3c171f0b
LP
751 /* Now, let's drop privileges to become the user who owns the segfaulted process and allocate the coredump
752 * memory under the user's uid. This also ensures that the credentials journald will see are the ones of the
753 * coredumping user, thus making sure the user gets access to the core dump. Let's also get rid of all
754 * capabilities, if we run as root, we won't need them anymore. */
755 r = change_uid_gid(context);
756 if (r < 0)
757 return log_error_errno(r, "Failed to drop privileges: %m");
34c10968 758
349cc4a5 759#if HAVE_ELFUTILS
3c171f0b
LP
760 /* Try to get a strack trace if we can */
761 if (coredump_size <= arg_process_size_max) {
762 _cleanup_free_ char *stacktrace = NULL;
763
764 r = coredump_make_stack_trace(coredump_fd, context[CONTEXT_EXE], &stacktrace);
765 if (r >= 0)
605405c6
ZJS
766 core_message = strjoin("MESSAGE=Process ", context[CONTEXT_PID],
767 " (", context[CONTEXT_COMM], ") of user ",
92e92d71
ZJS
768 context[CONTEXT_UID], " dumped core.",
769 journald_crash ? "\nCoredump diverted to " : "",
770 journald_crash ? filename : "",
771 "\n\n", stacktrace);
3c171f0b
LP
772 else if (r == -EINVAL)
773 log_warning("Failed to generate stack trace: %s", dwfl_errmsg(dwfl_errno()));
774 else
775 log_warning_errno(r, "Failed to generate stack trace: %m");
6e9ef603 776 } else
5206a724 777 log_debug("Not generating stack trace: core size %"PRIu64" is greater than %"PRIu64" (the configured maximum)",
6e9ef603 778 coredump_size, arg_process_size_max);
803a3464 779
3c171f0b
LP
780 if (!core_message)
781#endif
782log:
92e92d71
ZJS
783 core_message = strjoin("MESSAGE=Process ", context[CONTEXT_PID],
784 " (", context[CONTEXT_COMM], ") of user ",
785 context[CONTEXT_UID], " dumped core.",
786 journald_crash ? "\nCoredump diverted to " : NULL,
787 journald_crash ? filename : NULL);
788 if (!core_message)
789 return log_oom();
790
791 if (journald_crash) {
792 /* We cannot log to the journal, so just print the MESSAGE.
793 * The target was set previously to something safe. */
ae2173d6 794 log_dispatch(LOG_ERR, 0, core_message);
92e92d71
ZJS
795 return 0;
796 }
797
e6a7ec4b 798 iovec[n_iovec++] = IOVEC_MAKE_STRING(core_message);
3c171f0b 799
0cd4e913 800 if (truncated)
e6a7ec4b 801 iovec[n_iovec++] = IOVEC_MAKE_STRING("COREDUMP_TRUNCATED=1");
0cd4e913 802
3c171f0b 803 /* Optionally store the entire coredump in the journal */
6e9ef603
ZJS
804 if (arg_storage == COREDUMP_STORAGE_JOURNAL) {
805 if (coredump_size <= arg_journal_size_max) {
806 size_t sz = 0;
807
808 /* Store the coredump itself in the journal */
809
810 r = allocate_journal_field(coredump_fd, (size_t) coredump_size, &coredump_data, &sz);
e6a7ec4b
LP
811 if (r >= 0)
812 iovec[n_iovec++] = IOVEC_MAKE(coredump_data, sz);
813 else
6e9ef603
ZJS
814 log_warning_errno(r, "Failed to attach the core to the journal entry: %m");
815 } else
5206a724 816 log_info("The core will not be stored: size %"PRIu64" is greater than %"PRIu64" (the configured maximum)",
6e9ef603 817 coredump_size, arg_journal_size_max);
f5e04665
LP
818 }
819
3c171f0b
LP
820 assert(n_iovec <= n_iovec_allocated);
821
822 r = sd_journal_sendv(iovec, n_iovec);
823 if (r < 0)
824 return log_error_errno(r, "Failed to log coredump: %m");
825
826 return 0;
827}
828
ea5cc2a8 829static void map_context_fields(const struct iovec *iovec, const char* context[]) {
3c171f0b 830
92e92d71 831 static const char * const context_field_names[] = {
3c171f0b
LP
832 [CONTEXT_PID] = "COREDUMP_PID=",
833 [CONTEXT_UID] = "COREDUMP_UID=",
834 [CONTEXT_GID] = "COREDUMP_GID=",
835 [CONTEXT_SIGNAL] = "COREDUMP_SIGNAL=",
836 [CONTEXT_TIMESTAMP] = "COREDUMP_TIMESTAMP=",
92e92d71 837 [CONTEXT_RLIMIT] = "COREDUMP_RLIMIT=",
f45b8015 838 [CONTEXT_HOSTNAME] = "COREDUMP_HOSTNAME=",
3c171f0b
LP
839 [CONTEXT_COMM] = "COREDUMP_COMM=",
840 [CONTEXT_EXE] = "COREDUMP_EXE=",
841 };
842
843 unsigned i;
844
845 assert(iovec);
846 assert(context);
847
92e92d71 848 for (i = 0; i < ELEMENTSOF(context_field_names); i++) {
d27b725a 849 char *p;
3c171f0b 850
92e92d71
ZJS
851 if (!context_field_names[i])
852 continue;
853
d27b725a
LP
854 p = memory_startswith(iovec->iov_base, iovec->iov_len, context_field_names[i]);
855 if (!p)
3c171f0b
LP
856 continue;
857
858 /* Note that these strings are NUL terminated, because we made sure that a trailing NUL byte is in the
859 * buffer, though not included in the iov_len count. (see below) */
d27b725a 860 context[i] = p;
3c171f0b
LP
861 break;
862 }
863}
864
865static int process_socket(int fd) {
866 _cleanup_close_ int coredump_fd = -1;
867 struct iovec *iovec = NULL;
80002f66 868 size_t n_iovec = 0, n_allocated = 0, i, k;
3c171f0b
LP
869 const char *context[_CONTEXT_MAX] = {};
870 int r;
871
872 assert(fd >= 0);
873
874 log_set_target(LOG_TARGET_AUTO);
875 log_parse_environment();
876 log_open();
877
988e89ee
ZJS
878 log_debug("Processing coredump received on stdin...");
879
3c171f0b
LP
880 for (;;) {
881 union {
882 struct cmsghdr cmsghdr;
883 uint8_t buf[CMSG_SPACE(sizeof(int))];
884 } control = {};
885 struct msghdr mh = {
886 .msg_control = &control,
887 .msg_controllen = sizeof(control),
888 .msg_iovlen = 1,
889 };
890 ssize_t n;
fe1ef0f8 891 ssize_t l;
3c171f0b 892
0cd4e913 893 if (!GREEDY_REALLOC(iovec, n_allocated, n_iovec + SUBMIT_COREDUMP_FIELDS)) {
3c171f0b
LP
894 r = log_oom();
895 goto finish;
896 }
897
fe1ef0f8
EV
898 l = next_datagram_size_fd(fd);
899 if (l < 0) {
900 r = log_error_errno(l, "Failed to determine datagram size to read: %m");
3c171f0b
LP
901 goto finish;
902 }
903
904 assert(l >= 0);
905
906 iovec[n_iovec].iov_len = l;
907 iovec[n_iovec].iov_base = malloc(l + 1);
3c171f0b
LP
908 if (!iovec[n_iovec].iov_base) {
909 r = log_oom();
910 goto finish;
911 }
912
913 mh.msg_iov = iovec + n_iovec;
914
915 n = recvmsg(fd, &mh, MSG_NOSIGNAL|MSG_CMSG_CLOEXEC);
916 if (n < 0) {
917 free(iovec[n_iovec].iov_base);
918 r = log_error_errno(errno, "Failed to receive datagram: %m");
919 goto finish;
920 }
921
922 if (n == 0) {
923 struct cmsghdr *cmsg, *found = NULL;
924 /* The final zero-length datagram carries the file descriptor and tells us that we're done. */
925
926 free(iovec[n_iovec].iov_base);
927
928 CMSG_FOREACH(cmsg, &mh) {
929 if (cmsg->cmsg_level == SOL_SOCKET &&
930 cmsg->cmsg_type == SCM_RIGHTS &&
931 cmsg->cmsg_len == CMSG_LEN(sizeof(int))) {
932 assert(!found);
933 found = cmsg;
934 }
935 }
936
937 if (!found) {
938 log_error("Coredump file descriptor missing.");
939 r = -EBADMSG;
940 goto finish;
941 }
942
943 assert(coredump_fd < 0);
944 coredump_fd = *(int*) CMSG_DATA(found);
945 break;
946 }
947
948 /* Add trailing NUL byte, in case these are strings */
949 ((char*) iovec[n_iovec].iov_base)[n] = 0;
950 iovec[n_iovec].iov_len = (size_t) n;
951
952 cmsg_close_all(&mh);
953 map_context_fields(iovec + n_iovec, context);
954 n_iovec++;
955 }
956
0cd4e913 957 if (!GREEDY_REALLOC(iovec, n_allocated, n_iovec + SUBMIT_COREDUMP_FIELDS)) {
3c171f0b 958 r = log_oom();
34c10968
LP
959 goto finish;
960 }
961
61233823 962 /* Make sure we got all data we really need */
3c171f0b
LP
963 assert(context[CONTEXT_PID]);
964 assert(context[CONTEXT_UID]);
965 assert(context[CONTEXT_GID]);
966 assert(context[CONTEXT_SIGNAL]);
967 assert(context[CONTEXT_TIMESTAMP]);
bdfd7b2c 968 assert(context[CONTEXT_RLIMIT]);
f45b8015 969 assert(context[CONTEXT_HOSTNAME]);
3c171f0b
LP
970 assert(context[CONTEXT_COMM]);
971 assert(coredump_fd >= 0);
972
80002f66
LP
973 /* Small quirk: the journal fields contain the timestamp padded with six zeroes, so that the kernel-supplied 1s
974 * granularity timestamps becomes 1µs granularity, i.e. the granularity systemd usually operates in. Since we
975 * are reconstructing the original kernel context, we chop this off again, here. */
976 k = strlen(context[CONTEXT_TIMESTAMP]);
977 if (k > 6)
978 context[CONTEXT_TIMESTAMP] = strndupa(context[CONTEXT_TIMESTAMP], k - 6);
979
5b45a160 980 r = submit_coredump(context, iovec, n_allocated, n_iovec, coredump_fd);
3c171f0b
LP
981
982finish:
983 for (i = 0; i < n_iovec; i++)
984 free(iovec[i].iov_base);
985 free(iovec);
986
987 return r;
988}
989
990static int send_iovec(const struct iovec iovec[], size_t n_iovec, int input_fd) {
991
992 static const union sockaddr_union sa = {
993 .un.sun_family = AF_UNIX,
994 .un.sun_path = "/run/systemd/coredump",
995 };
996 _cleanup_close_ int fd = -1;
997 size_t i;
998 int r;
999
1000 assert(iovec || n_iovec <= 0);
1001 assert(input_fd >= 0);
1002
1003 fd = socket(AF_UNIX, SOCK_SEQPACKET|SOCK_CLOEXEC, 0);
1004 if (fd < 0)
1005 return log_error_errno(errno, "Failed to create coredump socket: %m");
1006
fc2fffe7 1007 if (connect(fd, &sa.sa, SOCKADDR_UN_LEN(sa.un)) < 0)
3c171f0b
LP
1008 return log_error_errno(errno, "Failed to connect to coredump service: %m");
1009
1010 for (i = 0; i < n_iovec; i++) {
fec603eb
LP
1011 struct msghdr mh = {
1012 .msg_iov = (struct iovec*) iovec + i,
1013 .msg_iovlen = 1,
1014 };
1015 struct iovec copy[2];
1016
1017 for (;;) {
1018 if (sendmsg(fd, &mh, MSG_NOSIGNAL) >= 0)
1019 break;
1020
1021 if (errno == EMSGSIZE && mh.msg_iov[0].iov_len > 0) {
1022 /* This field didn't fit? That's a pity. Given that this is just metadata,
1023 * let's truncate the field at half, and try again. We append three dots, in
1024 * order to show that this is truncated. */
1025
1026 if (mh.msg_iov != copy) {
1027 /* We don't want to modify the caller's iovec, hence let's create our
1028 * own array, consisting of two new iovecs, where the first is a
1029 * (truncated) copy of what we want to send, and the second one
1030 * contains the trailing dots. */
1031 copy[0] = iovec[i];
1032 copy[1] = (struct iovec) {
1033 .iov_base = (char[]) { '.', '.', '.' },
1034 .iov_len = 3,
1035 };
1036
1037 mh.msg_iov = copy;
1038 mh.msg_iovlen = 2;
1039 }
1040
1041 copy[0].iov_len /= 2; /* halve it, and try again */
1042 continue;
1043 }
3c171f0b 1044
3c171f0b 1045 return log_error_errno(errno, "Failed to send coredump datagram: %m");
fec603eb 1046 }
1eef15b1
ZJS
1047 }
1048
3c171f0b
LP
1049 r = send_one_fd(fd, input_fd, 0);
1050 if (r < 0)
1051 return log_error_errno(r, "Failed to send coredump fd: %m");
1eef15b1 1052
3c171f0b
LP
1053 return 0;
1054}
1eef15b1 1055
f45b8015 1056static char* set_iovec_field(struct iovec *iovec, size_t *n_iovec, const char *field, const char *value) {
9aa82023 1057 char *x;
3c171f0b 1058
9aa82023
ZJS
1059 x = strappend(field, value);
1060 if (x)
e6a7ec4b 1061 iovec[(*n_iovec)++] = IOVEC_MAKE_STRING(x);
9aa82023
ZJS
1062 return x;
1063}
3c171f0b 1064
f45b8015 1065static char* set_iovec_field_free(struct iovec *iovec, size_t *n_iovec, const char *field, char *value) {
9aa82023
ZJS
1066 char *x;
1067
1068 x = set_iovec_field(iovec, n_iovec, field, value);
1069 free(value);
1070 return x;
1071}
1072
92e92d71 1073static int gather_pid_metadata(
ea5cc2a8 1074 char* context[_CONTEXT_MAX],
9aa82023 1075 char **comm_fallback,
5b45a160 1076 struct iovec *iovec, size_t *n_iovec) {
d14bcb4e 1077
f45b8015 1078 /* We need 27 empty slots in iovec!
d14bcb4e
LP
1079 *
1080 * Note that if we fail on oom later on, we do not roll-back changes to the iovec structure. (It remains valid,
1081 * with the first n_iovec fields initialized.) */
3c171f0b 1082
3c171f0b 1083 uid_t owner_uid;
3c171f0b
LP
1084 pid_t pid;
1085 char *t;
9aa82023 1086 const char *p;
d14bcb4e 1087 int r, signo;
3c171f0b 1088
9aa82023 1089 r = parse_pid(context[CONTEXT_PID], &pid);
3c171f0b 1090 if (r < 0)
988e89ee 1091 return log_error_errno(r, "Failed to parse PID \"%s\": %m", context[CONTEXT_PID]);
3c171f0b 1092
ea5cc2a8 1093 r = get_process_comm(pid, &context[CONTEXT_COMM]);
3c171f0b
LP
1094 if (r < 0) {
1095 log_warning_errno(r, "Failed to get COMM, falling back to the command line: %m");
ea5cc2a8
ZJS
1096 context[CONTEXT_COMM] = strv_join(comm_fallback, " ");
1097 if (!context[CONTEXT_COMM])
3c171f0b
LP
1098 return log_oom();
1099 }
1100
ea5cc2a8 1101 r = get_process_exe(pid, &context[CONTEXT_EXE]);
3c171f0b
LP
1102 if (r < 0)
1103 log_warning_errno(r, "Failed to get EXE, ignoring: %m");
1104
92e92d71 1105 if (cg_pid_get_unit(pid, &context[CONTEXT_UNIT]) >= 0) {
b8cda92d 1106 if (!is_journald_crash((const char**) context)) {
92e92d71
ZJS
1107 /* OK, now we know it's not the journal, hence we can make use of it now. */
1108 log_set_target(LOG_TARGET_JOURNAL_OR_KMSG);
1109 log_open();
1110 }
3c171f0b 1111
c8091d92 1112 /* If this is PID 1 disable coredump collection, we'll unlikely be able to process it later on. */
b8cda92d 1113 if (is_pid1_crash((const char**) context)) {
c8091d92 1114 log_notice("Due to PID 1 having crashed coredump collection will now be turned off.");
e557b1a6 1115 disable_coredumps();
c8091d92
LP
1116 }
1117
92e92d71 1118 set_iovec_field(iovec, n_iovec, "COREDUMP_UNIT=", context[CONTEXT_UNIT]);
8c8549db 1119 }
803a3464 1120
9aa82023
ZJS
1121 if (cg_pid_get_user_unit(pid, &t) >= 0)
1122 set_iovec_field_free(iovec, n_iovec, "COREDUMP_USER_UNIT=", t);
3c171f0b 1123
9aa82023
ZJS
1124 /* The next few are mandatory */
1125 if (!set_iovec_field(iovec, n_iovec, "COREDUMP_PID=", context[CONTEXT_PID]))
1126 return log_oom();
3c171f0b 1127
9aa82023
ZJS
1128 if (!set_iovec_field(iovec, n_iovec, "COREDUMP_UID=", context[CONTEXT_UID]))
1129 return log_oom();
f5e04665 1130
9aa82023
ZJS
1131 if (!set_iovec_field(iovec, n_iovec, "COREDUMP_GID=", context[CONTEXT_GID]))
1132 return log_oom();
f5e04665 1133
9aa82023
ZJS
1134 if (!set_iovec_field(iovec, n_iovec, "COREDUMP_SIGNAL=", context[CONTEXT_SIGNAL]))
1135 return log_oom();
f5e04665 1136
9aa82023
ZJS
1137 if (!set_iovec_field(iovec, n_iovec, "COREDUMP_RLIMIT=", context[CONTEXT_RLIMIT]))
1138 return log_oom();
f5e04665 1139
f45b8015
JF
1140 if (!set_iovec_field(iovec, n_iovec, "COREDUMP_HOSTNAME=", context[CONTEXT_HOSTNAME]))
1141 return log_oom();
1142
ea5cc2a8 1143 if (!set_iovec_field(iovec, n_iovec, "COREDUMP_COMM=", context[CONTEXT_COMM]))
9aa82023 1144 return log_oom();
bdfd7b2c 1145
ea5cc2a8
ZJS
1146 if (context[CONTEXT_EXE] &&
1147 !set_iovec_field(iovec, n_iovec, "COREDUMP_EXE=", context[CONTEXT_EXE]))
9aa82023 1148 return log_oom();
f5e04665 1149
9aa82023
ZJS
1150 if (sd_pid_get_session(pid, &t) >= 0)
1151 set_iovec_field_free(iovec, n_iovec, "COREDUMP_SESSION=", t);
f5e04665 1152
a035f819 1153 if (sd_pid_get_owner_uid(pid, &owner_uid) >= 0) {
9aa82023 1154 r = asprintf(&t, "COREDUMP_OWNER_UID=" UID_FMT, owner_uid);
7de80bfe 1155 if (r > 0)
e6a7ec4b 1156 iovec[(*n_iovec)++] = IOVEC_MAKE_STRING(t);
f5e04665
LP
1157 }
1158
9aa82023
ZJS
1159 if (sd_pid_get_slice(pid, &t) >= 0)
1160 set_iovec_field_free(iovec, n_iovec, "COREDUMP_SLICE=", t);
f5e04665 1161
9aa82023
ZJS
1162 if (get_process_cmdline(pid, 0, false, &t) >= 0)
1163 set_iovec_field_free(iovec, n_iovec, "COREDUMP_CMDLINE=", t);
a035f819 1164
9aa82023
ZJS
1165 if (cg_pid_get_path_shifted(pid, NULL, &t) >= 0)
1166 set_iovec_field_free(iovec, n_iovec, "COREDUMP_CGROUP=", t);
a035f819 1167
9aa82023
ZJS
1168 if (compose_open_fds(pid, &t) >= 0)
1169 set_iovec_field_free(iovec, n_iovec, "COREDUMP_OPEN_FDS=", t);
3f132692
JF
1170
1171 p = procfs_file_alloca(pid, "status");
9aa82023
ZJS
1172 if (read_full_file(p, &t, NULL) >= 0)
1173 set_iovec_field_free(iovec, n_iovec, "COREDUMP_PROC_STATUS=", t);
3f132692
JF
1174
1175 p = procfs_file_alloca(pid, "maps");
9aa82023
ZJS
1176 if (read_full_file(p, &t, NULL) >= 0)
1177 set_iovec_field_free(iovec, n_iovec, "COREDUMP_PROC_MAPS=", t);
3f132692
JF
1178
1179 p = procfs_file_alloca(pid, "limits");
9aa82023
ZJS
1180 if (read_full_file(p, &t, NULL) >= 0)
1181 set_iovec_field_free(iovec, n_iovec, "COREDUMP_PROC_LIMITS=", t);
3f132692
JF
1182
1183 p = procfs_file_alloca(pid, "cgroup");
9aa82023
ZJS
1184 if (read_full_file(p, &t, NULL) >=0)
1185 set_iovec_field_free(iovec, n_iovec, "COREDUMP_PROC_CGROUP=", t);
3f132692 1186
d7032b1f 1187 p = procfs_file_alloca(pid, "mountinfo");
9aa82023
ZJS
1188 if (read_full_file(p, &t, NULL) >=0)
1189 set_iovec_field_free(iovec, n_iovec, "COREDUMP_PROC_MOUNTINFO=", t);
d7032b1f 1190
9aa82023
ZJS
1191 if (get_process_cwd(pid, &t) >= 0)
1192 set_iovec_field_free(iovec, n_iovec, "COREDUMP_CWD=", t);
3f132692
JF
1193
1194 if (get_process_root(pid, &t) >= 0) {
9aa82023
ZJS
1195 bool proc_self_root_is_slash;
1196
1197 proc_self_root_is_slash = strcmp(t, "/") == 0;
3f132692 1198
9aa82023 1199 set_iovec_field_free(iovec, n_iovec, "COREDUMP_ROOT=", t);
7ed03ce6
JF
1200
1201 /* If the process' root is "/", then there is a chance it has
1202 * mounted own root and hence being containerized. */
9aa82023
ZJS
1203 if (proc_self_root_is_slash && get_process_container_parent_cmdline(pid, &t) > 0)
1204 set_iovec_field_free(iovec, n_iovec, "COREDUMP_CONTAINER_CMDLINE=", t);
3f132692
JF
1205 }
1206
9aa82023
ZJS
1207 if (get_process_environ(pid, &t) >= 0)
1208 set_iovec_field_free(iovec, n_iovec, "COREDUMP_ENVIRON=", t);
1209
4600a396 1210 t = strjoin("COREDUMP_TIMESTAMP=", context[CONTEXT_TIMESTAMP], "000000");
9aa82023 1211 if (t)
e6a7ec4b 1212 iovec[(*n_iovec)++] = IOVEC_MAKE_STRING(t);
9aa82023 1213
d14bcb4e
LP
1214 if (safe_atoi(context[CONTEXT_SIGNAL], &signo) >= 0 && SIGNAL_VALID(signo))
1215 set_iovec_field(iovec, n_iovec, "COREDUMP_SIGNAL_NAME=SIG", signal_to_string(signo));
1216
92e92d71 1217 return 0; /* we successfully acquired all metadata */
9aa82023 1218}
3f132692 1219
9aa82023
ZJS
1220static int process_kernel(int argc, char* argv[]) {
1221
ea5cc2a8 1222 char* context[_CONTEXT_MAX] = {};
f45b8015 1223 struct iovec iovec[29 + SUBMIT_COREDUMP_FIELDS];
988e89ee 1224 size_t i, n_iovec, n_to_free = 0;
9aa82023
ZJS
1225 int r;
1226
988e89ee
ZJS
1227 log_debug("Processing coredump received from the kernel...");
1228
9aa82023 1229 if (argc < CONTEXT_COMM + 1) {
988e89ee 1230 log_error("Not enough arguments passed by the kernel (%i, expected %i).", argc - 1, CONTEXT_COMM + 1 - 1);
9aa82023 1231 return -EINVAL;
3f132692
JF
1232 }
1233
ea5cc2a8
ZJS
1234 context[CONTEXT_PID] = argv[1 + CONTEXT_PID];
1235 context[CONTEXT_UID] = argv[1 + CONTEXT_UID];
1236 context[CONTEXT_GID] = argv[1 + CONTEXT_GID];
1237 context[CONTEXT_SIGNAL] = argv[1 + CONTEXT_SIGNAL];
1238 context[CONTEXT_TIMESTAMP] = argv[1 + CONTEXT_TIMESTAMP];
1239 context[CONTEXT_RLIMIT] = argv[1 + CONTEXT_RLIMIT];
f45b8015 1240 context[CONTEXT_HOSTNAME] = argv[1 + CONTEXT_HOSTNAME];
9aa82023 1241
92e92d71
ZJS
1242 r = gather_pid_metadata(context, argv + 1 + CONTEXT_COMM, iovec, &n_to_free);
1243 if (r < 0)
86562420 1244 goto finish;
86562420 1245
988e89ee 1246 n_iovec = n_to_free;
f5e04665 1247
e6a7ec4b 1248 iovec[n_iovec++] = IOVEC_MAKE_STRING("MESSAGE_ID=" SD_MESSAGE_COREDUMP_STR);
4850d39a
LP
1249
1250 assert_cc(2 == LOG_CRIT);
e6a7ec4b 1251 iovec[n_iovec++] = IOVEC_MAKE_STRING("PRIORITY=2");
0dc5d23c 1252
3c171f0b 1253 assert(n_iovec <= ELEMENTSOF(iovec));
34c10968 1254
b8cda92d 1255 if (is_journald_crash((const char**) context) || is_pid1_crash((const char**) context))
92e92d71
ZJS
1256 r = submit_coredump((const char**) context,
1257 iovec, ELEMENTSOF(iovec), n_iovec,
1258 STDIN_FILENO);
1259 else
1260 r = send_iovec(iovec, n_iovec, STDIN_FILENO);
9aa82023
ZJS
1261
1262 finish:
1263 for (i = 0; i < n_to_free; i++)
1264 free(iovec[i].iov_base);
1265
92e92d71 1266 /* Those fields are allocated by gather_pid_metadata */
ea5cc2a8
ZJS
1267 free(context[CONTEXT_COMM]);
1268 free(context[CONTEXT_EXE]);
92e92d71 1269 free(context[CONTEXT_UNIT]);
ea5cc2a8 1270
9aa82023 1271 return r;
3c171f0b 1272}
34c10968 1273
988e89ee 1274static int process_backtrace(int argc, char *argv[]) {
ea5cc2a8
ZJS
1275 char *context[_CONTEXT_MAX] = {};
1276 _cleanup_free_ char *message = NULL;
5b45a160
ZJS
1277 _cleanup_free_ struct iovec *iovec = NULL;
1278 size_t n_iovec, n_allocated, n_to_free = 0, i;
988e89ee 1279 int r;
5b45a160
ZJS
1280 JournalImporter importer = {
1281 .fd = STDIN_FILENO,
1282 };
988e89ee
ZJS
1283
1284 log_debug("Processing backtrace on stdin...");
1285
1286 if (argc < CONTEXT_COMM + 1) {
1287 log_error("Not enough arguments passed (%i, expected %i).", argc - 1, CONTEXT_COMM + 1 - 1);
1288 return -EINVAL;
1289 }
1290
ea5cc2a8
ZJS
1291 context[CONTEXT_PID] = argv[2 + CONTEXT_PID];
1292 context[CONTEXT_UID] = argv[2 + CONTEXT_UID];
1293 context[CONTEXT_GID] = argv[2 + CONTEXT_GID];
1294 context[CONTEXT_SIGNAL] = argv[2 + CONTEXT_SIGNAL];
1295 context[CONTEXT_TIMESTAMP] = argv[2 + CONTEXT_TIMESTAMP];
1296 context[CONTEXT_RLIMIT] = argv[2 + CONTEXT_RLIMIT];
f45b8015 1297 context[CONTEXT_HOSTNAME] = argv[2 + CONTEXT_HOSTNAME];
988e89ee 1298
f45b8015
JF
1299 n_allocated = 34 + COREDUMP_STORAGE_EXTERNAL;
1300 /* 26 metadata, 2 static, +unknown input, 4 storage, rounded up */
5b45a160
ZJS
1301 iovec = new(struct iovec, n_allocated);
1302 if (!iovec)
1303 return log_oom();
1304
92e92d71 1305 r = gather_pid_metadata(context, argv + 2 + CONTEXT_COMM, iovec, &n_to_free);
988e89ee
ZJS
1306 if (r < 0)
1307 goto finish;
86562420
LP
1308 if (r > 0) {
1309 /* This was a special crash, and has already been processed. */
1310 r = 0;
1311 goto finish;
1312 }
5b45a160 1313 n_iovec = n_to_free;
988e89ee 1314
86562420 1315 for (;;) {
5b45a160
ZJS
1316 r = journal_importer_process_data(&importer);
1317 if (r < 0) {
1318 log_error_errno(r, "Failed to parse journal entry on stdin: %m");
1319 goto finish;
1320 }
d74dc4f2
ZJS
1321 if (r == 1 || /* complete entry */
1322 journal_importer_eof(&importer)) /* end of data */
5b45a160 1323 break;
988e89ee 1324 }
988e89ee 1325
5b45a160
ZJS
1326 if (!GREEDY_REALLOC(iovec, n_allocated, n_iovec + importer.iovw.count + 2))
1327 return log_oom();
1328
1329 if (journal_importer_eof(&importer)) {
1330 log_warning("Did not receive a full journal entry on stdin, ignoring message sent by reporter");
988e89ee 1331
ea5cc2a8
ZJS
1332 message = strjoin("MESSAGE=Process ", context[CONTEXT_PID],
1333 " (", context[CONTEXT_COMM], ")"
5b45a160
ZJS
1334 " of user ", context[CONTEXT_UID],
1335 " failed with ", context[CONTEXT_SIGNAL]);
1336 if (!message) {
1337 r = log_oom();
1338 goto finish;
1339 }
e6a7ec4b 1340 iovec[n_iovec++] = IOVEC_MAKE_STRING(message);
5b45a160
ZJS
1341 } else {
1342 for (i = 0; i < importer.iovw.count; i++)
1343 iovec[n_iovec++] = importer.iovw.iovec[i];
1344 }
988e89ee 1345
e6a7ec4b 1346 iovec[n_iovec++] = IOVEC_MAKE_STRING("MESSAGE_ID=" SD_MESSAGE_BACKTRACE_STR);
988e89ee 1347 assert_cc(2 == LOG_CRIT);
e6a7ec4b 1348 iovec[n_iovec++] = IOVEC_MAKE_STRING("PRIORITY=2");
988e89ee 1349
5b45a160 1350 assert(n_iovec <= n_allocated);
988e89ee
ZJS
1351
1352 r = sd_journal_sendv(iovec, n_iovec);
1353 if (r < 0)
1354 log_error_errno(r, "Failed to log backtrace: %m");
1355
1356 finish:
1357 for (i = 0; i < n_to_free; i++)
1358 free(iovec[i].iov_base);
1359
92e92d71 1360 /* Those fields are allocated by gather_pid_metadata */
ea5cc2a8
ZJS
1361 free(context[CONTEXT_COMM]);
1362 free(context[CONTEXT_EXE]);
92e92d71 1363 free(context[CONTEXT_UNIT]);
ea5cc2a8 1364
988e89ee
ZJS
1365 return r;
1366}
1367
3c171f0b
LP
1368int main(int argc, char *argv[]) {
1369 int r;
fee80f69 1370
9aa82023
ZJS
1371 /* First, log to a safe place, since we don't know what crashed and it might
1372 * be journald which we'd rather not log to then. */
8d4e028f 1373
3c171f0b
LP
1374 log_set_target(LOG_TARGET_KMSG);
1375 log_open();
8d4e028f 1376
3c171f0b
LP
1377 /* Make sure we never enter a loop */
1378 (void) prctl(PR_SET_DUMPABLE, 0);
8d4e028f 1379
3c171f0b
LP
1380 /* Ignore all parse errors */
1381 (void) parse_config();
fee80f69 1382
3c171f0b
LP
1383 log_debug("Selected storage '%s'.", coredump_storage_to_string(arg_storage));
1384 log_debug("Selected compression %s.", yes_no(arg_compress));
fee80f69 1385
3c171f0b
LP
1386 r = sd_listen_fds(false);
1387 if (r < 0) {
1388 log_error_errno(r, "Failed to determine number of file descriptor: %m");
1389 goto finish;
fee80f69
LP
1390 }
1391
9aa82023
ZJS
1392 /* If we got an fd passed, we are running in coredumpd mode. Otherwise we
1393 * are invoked from the kernel as coredump handler. */
988e89ee
ZJS
1394 if (r == 0) {
1395 if (streq_ptr(argv[1], "--backtrace"))
1396 r = process_backtrace(argc, argv);
1397 else
1398 r = process_kernel(argc, argv);
1399 } else if (r == 1)
3c171f0b
LP
1400 r = process_socket(SD_LISTEN_FDS_START);
1401 else {
1402 log_error("Received unexpected number of file descriptors.");
1403 r = -EINVAL;
1404 }
f5e04665
LP
1405
1406finish:
f5e04665
LP
1407 return r < 0 ? EXIT_FAILURE : EXIT_SUCCESS;
1408}