]> git.ipfire.org Git - thirdparty/systemd.git/blame - src/coredump/coredump.c
Merge pull request #7411 from joergsteffens/tapechanger
[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)
9d77ca39 343 return log_error_errno(r, "Failed to parse resource limit '%s': %m", 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;
7b26ea6f 514 int r;
3f132692
JF
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;
4d84bc2f 537 int fd;
3f132692 538
59059b4a 539 r = readlinkat_malloc(dirfd(proc_fd_dir), dent->d_name, &fdname);
3f132692
JF
540 if (r < 0)
541 return r;
542
543 fprintf(stream, "%s%s:%s\n", fddelim, dent->d_name, fdname);
544 fddelim = "\n";
545
546 /* Use the directory entry from /proc/[pid]/fd with /proc/[pid]/fdinfo */
59059b4a
ZJS
547 fd = openat(proc_fdinfo_fd, dent->d_name, O_NOFOLLOW|O_CLOEXEC|O_RDONLY);
548 if (fd < 0)
3f132692
JF
549 continue;
550
59059b4a 551 fdinfo = fdopen(fd, "re");
234519ae 552 if (!fdinfo) {
0d536673 553 safe_close(fd);
3f132692 554 continue;
59059b4a 555 }
3f132692 556
7b26ea6f
LP
557 for (;;) {
558 _cleanup_free_ char *line = NULL;
559
560 r = read_line(fdinfo, LONG_LINE_MAX, &line);
561 if (r < 0)
562 return r;
563 if (r == 0)
564 break;
565
0d536673 566 fputs(line, stream);
7b26ea6f 567 fputc('\n', stream);
4d84bc2f 568 }
3f132692
JF
569 }
570
4d84bc2f 571 errno = 0;
74ca738f 572 stream = safe_fclose(stream);
4d84bc2f 573
b3267152 574 if (errno > 0)
4d84bc2f
LP
575 return -errno;
576
ae2a15bc 577 *open_fds = TAKE_PTR(buffer);
4d84bc2f 578
3f132692
JF
579 return 0;
580}
581
7ed03ce6
JF
582static int get_process_ns(pid_t pid, const char *namespace, ino_t *ns) {
583 const char *p;
584 struct stat stbuf;
585 _cleanup_close_ int proc_ns_dir_fd;
586
587 p = procfs_file_alloca(pid, "ns");
588
589 proc_ns_dir_fd = open(p, O_DIRECTORY | O_CLOEXEC | O_RDONLY);
590 if (proc_ns_dir_fd < 0)
591 return -errno;
592
593 if (fstatat(proc_ns_dir_fd, namespace, &stbuf, /* flags */0) < 0)
594 return -errno;
595
596 *ns = stbuf.st_ino;
597 return 0;
598}
599
600static int get_mount_namespace_leader(pid_t pid, pid_t *container_pid) {
601 pid_t cpid = pid, ppid = 0;
602 ino_t proc_mntns;
603 int r = 0;
604
605 r = get_process_ns(pid, "mnt", &proc_mntns);
606 if (r < 0)
607 return r;
608
aa7530d6 609 for (;;) {
7ed03ce6
JF
610 ino_t parent_mntns;
611
612 r = get_process_ppid(cpid, &ppid);
613 if (r < 0)
614 return r;
615
616 r = get_process_ns(ppid, "mnt", &parent_mntns);
617 if (r < 0)
618 return r;
619
620 if (proc_mntns != parent_mntns)
621 break;
622
623 if (ppid == 1)
624 return -ENOENT;
625
626 cpid = ppid;
627 }
628
629 *container_pid = ppid;
630 return 0;
631}
632
633/* Returns 1 if the parent was found.
634 * Returns 0 if there is not a process we can call the pid's
635 * container parent (the pid's process isn't 'containerized').
636 * Returns a negative number on errors.
637 */
638static int get_process_container_parent_cmdline(pid_t pid, char** cmdline) {
639 int r = 0;
640 pid_t container_pid;
641 const char *proc_root_path;
642 struct stat root_stat, proc_root_stat;
643
644 /* To compare inodes of / and /proc/[pid]/root */
645 if (stat("/", &root_stat) < 0)
646 return -errno;
647
648 proc_root_path = procfs_file_alloca(pid, "root");
649 if (stat(proc_root_path, &proc_root_stat) < 0)
650 return -errno;
651
652 /* The process uses system root. */
653 if (proc_root_stat.st_ino == root_stat.st_ino) {
654 *cmdline = NULL;
655 return 0;
656 }
657
658 r = get_mount_namespace_leader(pid, &container_pid);
659 if (r < 0)
660 return r;
661
d3cba4ea
EV
662 r = get_process_cmdline(container_pid, 0, false, cmdline);
663 if (r < 0)
664 return r;
665
666 return 1;
7ed03ce6
JF
667}
668
3c171f0b
LP
669static int change_uid_gid(const char *context[]) {
670 uid_t uid;
671 gid_t gid;
672 int r;
34c10968 673
3c171f0b
LP
674 r = parse_uid(context[CONTEXT_UID], &uid);
675 if (r < 0)
676 return r;
8c8549db 677
888e378d
LP
678 if (uid <= SYSTEM_UID_MAX) {
679 const char *user = "systemd-coredump";
680
fafff8f1 681 r = get_user_creds(&user, &uid, &gid, NULL, NULL, 0);
888e378d
LP
682 if (r < 0) {
683 log_warning_errno(r, "Cannot resolve %s user. Proceeding to dump core as root: %m", user);
684 uid = gid = 0;
685 }
686 } else {
687 r = parse_gid(context[CONTEXT_GID], &gid);
688 if (r < 0)
689 return r;
690 }
3c171f0b
LP
691
692 return drop_privileges(uid, gid, 0);
693}
8c8549db 694
b8cda92d
LP
695static bool is_journald_crash(const char *context[_CONTEXT_MAX]) {
696 assert(context);
697
698 return streq_ptr(context[CONTEXT_UNIT], SPECIAL_JOURNALD_SERVICE);
699}
700
701static bool is_pid1_crash(const char *context[_CONTEXT_MAX]) {
702 assert(context);
703
704 return streq_ptr(context[CONTEXT_UNIT], SPECIAL_INIT_SCOPE) ||
705 streq_ptr(context[CONTEXT_PID], "1");
706}
707
0cd4e913
ZJS
708#define SUBMIT_COREDUMP_FIELDS 4
709
3c171f0b
LP
710static int submit_coredump(
711 const char *context[_CONTEXT_MAX],
712 struct iovec *iovec,
713 size_t n_iovec_allocated,
714 size_t n_iovec,
715 int input_fd) {
34c10968 716
5f3e0a74 717 _cleanup_close_ int coredump_fd = -1, coredump_node_fd = -1;
3c171f0b 718 _cleanup_free_ char *core_message = NULL, *filename = NULL, *coredump_data = NULL;
a5ca3649 719 uint64_t coredump_size = UINT64_MAX;
92e92d71 720 bool truncated = false, journald_crash;
3c171f0b 721 int r;
f5e04665 722
3c171f0b
LP
723 assert(context);
724 assert(iovec);
0cd4e913 725 assert(n_iovec_allocated >= n_iovec + SUBMIT_COREDUMP_FIELDS);
3c171f0b 726 assert(input_fd >= 0);
f5e04665 727
b8cda92d 728 journald_crash = is_journald_crash(context);
92e92d71 729
3c171f0b
LP
730 /* Vacuum before we write anything again */
731 (void) coredump_vacuum(-1, arg_keep_free, arg_max_use);
803a3464 732
3c171f0b 733 /* Always stream the coredump to disk, if that's possible */
0cd4e913
ZJS
734 r = save_external_coredump(context, input_fd,
735 &filename, &coredump_node_fd, &coredump_fd, &coredump_size, &truncated);
3c171f0b
LP
736 if (r < 0)
737 /* Skip whole core dumping part */
738 goto log;
739
740 /* If we don't want to keep the coredump on disk, remove it now, as later on we will lack the privileges for
741 * it. However, we keep the fd to it, so that we can still process it and log it. */
742 r = maybe_remove_external_coredump(filename, coredump_size);
743 if (r < 0)
744 return r;
745 if (r == 0) {
746 const char *coredump_filename;
747
748 coredump_filename = strjoina("COREDUMP_FILENAME=", filename);
e6a7ec4b 749 iovec[n_iovec++] = IOVEC_MAKE_STRING(coredump_filename);
6e9ef603 750 } else if (arg_storage == COREDUMP_STORAGE_EXTERNAL)
5206a724 751 log_info("The core will not be stored: size %"PRIu64" is greater than %"PRIu64" (the configured maximum)",
6e9ef603 752 coredump_size, arg_external_size_max);
f5e04665 753
3c171f0b
LP
754 /* Vacuum again, but exclude the coredump we just created */
755 (void) coredump_vacuum(coredump_node_fd >= 0 ? coredump_node_fd : coredump_fd, arg_keep_free, arg_max_use);
8c9571d0 756
3c171f0b
LP
757 /* Now, let's drop privileges to become the user who owns the segfaulted process and allocate the coredump
758 * memory under the user's uid. This also ensures that the credentials journald will see are the ones of the
759 * coredumping user, thus making sure the user gets access to the core dump. Let's also get rid of all
760 * capabilities, if we run as root, we won't need them anymore. */
761 r = change_uid_gid(context);
762 if (r < 0)
763 return log_error_errno(r, "Failed to drop privileges: %m");
34c10968 764
349cc4a5 765#if HAVE_ELFUTILS
3c171f0b
LP
766 /* Try to get a strack trace if we can */
767 if (coredump_size <= arg_process_size_max) {
768 _cleanup_free_ char *stacktrace = NULL;
769
770 r = coredump_make_stack_trace(coredump_fd, context[CONTEXT_EXE], &stacktrace);
771 if (r >= 0)
605405c6
ZJS
772 core_message = strjoin("MESSAGE=Process ", context[CONTEXT_PID],
773 " (", context[CONTEXT_COMM], ") of user ",
92e92d71
ZJS
774 context[CONTEXT_UID], " dumped core.",
775 journald_crash ? "\nCoredump diverted to " : "",
776 journald_crash ? filename : "",
777 "\n\n", stacktrace);
3c171f0b
LP
778 else if (r == -EINVAL)
779 log_warning("Failed to generate stack trace: %s", dwfl_errmsg(dwfl_errno()));
780 else
781 log_warning_errno(r, "Failed to generate stack trace: %m");
6e9ef603 782 } else
5206a724 783 log_debug("Not generating stack trace: core size %"PRIu64" is greater than %"PRIu64" (the configured maximum)",
6e9ef603 784 coredump_size, arg_process_size_max);
803a3464 785
3c171f0b
LP
786 if (!core_message)
787#endif
788log:
92e92d71
ZJS
789 core_message = strjoin("MESSAGE=Process ", context[CONTEXT_PID],
790 " (", context[CONTEXT_COMM], ") of user ",
791 context[CONTEXT_UID], " dumped core.",
792 journald_crash ? "\nCoredump diverted to " : NULL,
793 journald_crash ? filename : NULL);
794 if (!core_message)
795 return log_oom();
796
797 if (journald_crash) {
798 /* We cannot log to the journal, so just print the MESSAGE.
799 * The target was set previously to something safe. */
ae2173d6 800 log_dispatch(LOG_ERR, 0, core_message);
92e92d71
ZJS
801 return 0;
802 }
803
e6a7ec4b 804 iovec[n_iovec++] = IOVEC_MAKE_STRING(core_message);
3c171f0b 805
0cd4e913 806 if (truncated)
e6a7ec4b 807 iovec[n_iovec++] = IOVEC_MAKE_STRING("COREDUMP_TRUNCATED=1");
0cd4e913 808
3c171f0b 809 /* Optionally store the entire coredump in the journal */
6e9ef603
ZJS
810 if (arg_storage == COREDUMP_STORAGE_JOURNAL) {
811 if (coredump_size <= arg_journal_size_max) {
812 size_t sz = 0;
813
814 /* Store the coredump itself in the journal */
815
816 r = allocate_journal_field(coredump_fd, (size_t) coredump_size, &coredump_data, &sz);
e6a7ec4b
LP
817 if (r >= 0)
818 iovec[n_iovec++] = IOVEC_MAKE(coredump_data, sz);
819 else
6e9ef603
ZJS
820 log_warning_errno(r, "Failed to attach the core to the journal entry: %m");
821 } else
5206a724 822 log_info("The core will not be stored: size %"PRIu64" is greater than %"PRIu64" (the configured maximum)",
6e9ef603 823 coredump_size, arg_journal_size_max);
f5e04665
LP
824 }
825
3c171f0b
LP
826 assert(n_iovec <= n_iovec_allocated);
827
828 r = sd_journal_sendv(iovec, n_iovec);
829 if (r < 0)
830 return log_error_errno(r, "Failed to log coredump: %m");
831
832 return 0;
833}
834
ea5cc2a8 835static void map_context_fields(const struct iovec *iovec, const char* context[]) {
3c171f0b 836
92e92d71 837 static const char * const context_field_names[] = {
3c171f0b
LP
838 [CONTEXT_PID] = "COREDUMP_PID=",
839 [CONTEXT_UID] = "COREDUMP_UID=",
840 [CONTEXT_GID] = "COREDUMP_GID=",
841 [CONTEXT_SIGNAL] = "COREDUMP_SIGNAL=",
842 [CONTEXT_TIMESTAMP] = "COREDUMP_TIMESTAMP=",
92e92d71 843 [CONTEXT_RLIMIT] = "COREDUMP_RLIMIT=",
f45b8015 844 [CONTEXT_HOSTNAME] = "COREDUMP_HOSTNAME=",
3c171f0b
LP
845 [CONTEXT_COMM] = "COREDUMP_COMM=",
846 [CONTEXT_EXE] = "COREDUMP_EXE=",
847 };
848
849 unsigned i;
850
851 assert(iovec);
852 assert(context);
853
92e92d71 854 for (i = 0; i < ELEMENTSOF(context_field_names); i++) {
d27b725a 855 char *p;
3c171f0b 856
92e92d71
ZJS
857 if (!context_field_names[i])
858 continue;
859
d27b725a
LP
860 p = memory_startswith(iovec->iov_base, iovec->iov_len, context_field_names[i]);
861 if (!p)
3c171f0b
LP
862 continue;
863
864 /* Note that these strings are NUL terminated, because we made sure that a trailing NUL byte is in the
865 * buffer, though not included in the iov_len count. (see below) */
d27b725a 866 context[i] = p;
3c171f0b
LP
867 break;
868 }
869}
870
871static int process_socket(int fd) {
872 _cleanup_close_ int coredump_fd = -1;
873 struct iovec *iovec = NULL;
80002f66 874 size_t n_iovec = 0, n_allocated = 0, i, k;
3c171f0b
LP
875 const char *context[_CONTEXT_MAX] = {};
876 int r;
877
878 assert(fd >= 0);
879
880 log_set_target(LOG_TARGET_AUTO);
881 log_parse_environment();
882 log_open();
883
988e89ee
ZJS
884 log_debug("Processing coredump received on stdin...");
885
3c171f0b
LP
886 for (;;) {
887 union {
888 struct cmsghdr cmsghdr;
889 uint8_t buf[CMSG_SPACE(sizeof(int))];
890 } control = {};
891 struct msghdr mh = {
892 .msg_control = &control,
893 .msg_controllen = sizeof(control),
894 .msg_iovlen = 1,
895 };
896 ssize_t n;
fe1ef0f8 897 ssize_t l;
3c171f0b 898
0cd4e913 899 if (!GREEDY_REALLOC(iovec, n_allocated, n_iovec + SUBMIT_COREDUMP_FIELDS)) {
3c171f0b
LP
900 r = log_oom();
901 goto finish;
902 }
903
fe1ef0f8
EV
904 l = next_datagram_size_fd(fd);
905 if (l < 0) {
906 r = log_error_errno(l, "Failed to determine datagram size to read: %m");
3c171f0b
LP
907 goto finish;
908 }
909
910 assert(l >= 0);
911
912 iovec[n_iovec].iov_len = l;
913 iovec[n_iovec].iov_base = malloc(l + 1);
3c171f0b
LP
914 if (!iovec[n_iovec].iov_base) {
915 r = log_oom();
916 goto finish;
917 }
918
919 mh.msg_iov = iovec + n_iovec;
920
a6887cc0 921 n = recvmsg(fd, &mh, MSG_CMSG_CLOEXEC);
3c171f0b
LP
922 if (n < 0) {
923 free(iovec[n_iovec].iov_base);
924 r = log_error_errno(errno, "Failed to receive datagram: %m");
925 goto finish;
926 }
927
928 if (n == 0) {
929 struct cmsghdr *cmsg, *found = NULL;
930 /* The final zero-length datagram carries the file descriptor and tells us that we're done. */
931
932 free(iovec[n_iovec].iov_base);
933
934 CMSG_FOREACH(cmsg, &mh) {
935 if (cmsg->cmsg_level == SOL_SOCKET &&
936 cmsg->cmsg_type == SCM_RIGHTS &&
937 cmsg->cmsg_len == CMSG_LEN(sizeof(int))) {
938 assert(!found);
939 found = cmsg;
940 }
941 }
942
943 if (!found) {
944 log_error("Coredump file descriptor missing.");
945 r = -EBADMSG;
946 goto finish;
947 }
948
949 assert(coredump_fd < 0);
950 coredump_fd = *(int*) CMSG_DATA(found);
951 break;
952 }
953
954 /* Add trailing NUL byte, in case these are strings */
955 ((char*) iovec[n_iovec].iov_base)[n] = 0;
956 iovec[n_iovec].iov_len = (size_t) n;
957
958 cmsg_close_all(&mh);
959 map_context_fields(iovec + n_iovec, context);
960 n_iovec++;
961 }
962
0cd4e913 963 if (!GREEDY_REALLOC(iovec, n_allocated, n_iovec + SUBMIT_COREDUMP_FIELDS)) {
3c171f0b 964 r = log_oom();
34c10968
LP
965 goto finish;
966 }
967
61233823 968 /* Make sure we got all data we really need */
3c171f0b
LP
969 assert(context[CONTEXT_PID]);
970 assert(context[CONTEXT_UID]);
971 assert(context[CONTEXT_GID]);
972 assert(context[CONTEXT_SIGNAL]);
973 assert(context[CONTEXT_TIMESTAMP]);
bdfd7b2c 974 assert(context[CONTEXT_RLIMIT]);
f45b8015 975 assert(context[CONTEXT_HOSTNAME]);
3c171f0b
LP
976 assert(context[CONTEXT_COMM]);
977 assert(coredump_fd >= 0);
978
80002f66
LP
979 /* Small quirk: the journal fields contain the timestamp padded with six zeroes, so that the kernel-supplied 1s
980 * granularity timestamps becomes 1µs granularity, i.e. the granularity systemd usually operates in. Since we
981 * are reconstructing the original kernel context, we chop this off again, here. */
982 k = strlen(context[CONTEXT_TIMESTAMP]);
983 if (k > 6)
984 context[CONTEXT_TIMESTAMP] = strndupa(context[CONTEXT_TIMESTAMP], k - 6);
985
5b45a160 986 r = submit_coredump(context, iovec, n_allocated, n_iovec, coredump_fd);
3c171f0b
LP
987
988finish:
989 for (i = 0; i < n_iovec; i++)
990 free(iovec[i].iov_base);
991 free(iovec);
992
993 return r;
994}
995
996static int send_iovec(const struct iovec iovec[], size_t n_iovec, int input_fd) {
997
998 static const union sockaddr_union sa = {
999 .un.sun_family = AF_UNIX,
1000 .un.sun_path = "/run/systemd/coredump",
1001 };
1002 _cleanup_close_ int fd = -1;
1003 size_t i;
1004 int r;
1005
1006 assert(iovec || n_iovec <= 0);
1007 assert(input_fd >= 0);
1008
1009 fd = socket(AF_UNIX, SOCK_SEQPACKET|SOCK_CLOEXEC, 0);
1010 if (fd < 0)
1011 return log_error_errno(errno, "Failed to create coredump socket: %m");
1012
fc2fffe7 1013 if (connect(fd, &sa.sa, SOCKADDR_UN_LEN(sa.un)) < 0)
3c171f0b
LP
1014 return log_error_errno(errno, "Failed to connect to coredump service: %m");
1015
1016 for (i = 0; i < n_iovec; i++) {
fec603eb
LP
1017 struct msghdr mh = {
1018 .msg_iov = (struct iovec*) iovec + i,
1019 .msg_iovlen = 1,
1020 };
1021 struct iovec copy[2];
1022
1023 for (;;) {
1024 if (sendmsg(fd, &mh, MSG_NOSIGNAL) >= 0)
1025 break;
1026
1027 if (errno == EMSGSIZE && mh.msg_iov[0].iov_len > 0) {
1028 /* This field didn't fit? That's a pity. Given that this is just metadata,
1029 * let's truncate the field at half, and try again. We append three dots, in
1030 * order to show that this is truncated. */
1031
1032 if (mh.msg_iov != copy) {
1033 /* We don't want to modify the caller's iovec, hence let's create our
1034 * own array, consisting of two new iovecs, where the first is a
1035 * (truncated) copy of what we want to send, and the second one
1036 * contains the trailing dots. */
1037 copy[0] = iovec[i];
1038 copy[1] = (struct iovec) {
1039 .iov_base = (char[]) { '.', '.', '.' },
1040 .iov_len = 3,
1041 };
1042
1043 mh.msg_iov = copy;
1044 mh.msg_iovlen = 2;
1045 }
1046
1047 copy[0].iov_len /= 2; /* halve it, and try again */
1048 continue;
1049 }
3c171f0b 1050
3c171f0b 1051 return log_error_errno(errno, "Failed to send coredump datagram: %m");
fec603eb 1052 }
1eef15b1
ZJS
1053 }
1054
3c171f0b
LP
1055 r = send_one_fd(fd, input_fd, 0);
1056 if (r < 0)
1057 return log_error_errno(r, "Failed to send coredump fd: %m");
1eef15b1 1058
3c171f0b
LP
1059 return 0;
1060}
1eef15b1 1061
f45b8015 1062static char* set_iovec_field(struct iovec *iovec, size_t *n_iovec, const char *field, const char *value) {
9aa82023 1063 char *x;
3c171f0b 1064
9aa82023
ZJS
1065 x = strappend(field, value);
1066 if (x)
e6a7ec4b 1067 iovec[(*n_iovec)++] = IOVEC_MAKE_STRING(x);
9aa82023
ZJS
1068 return x;
1069}
3c171f0b 1070
f45b8015 1071static char* set_iovec_field_free(struct iovec *iovec, size_t *n_iovec, const char *field, char *value) {
9aa82023
ZJS
1072 char *x;
1073
1074 x = set_iovec_field(iovec, n_iovec, field, value);
1075 free(value);
1076 return x;
1077}
1078
92e92d71 1079static int gather_pid_metadata(
ea5cc2a8 1080 char* context[_CONTEXT_MAX],
9aa82023 1081 char **comm_fallback,
5b45a160 1082 struct iovec *iovec, size_t *n_iovec) {
d14bcb4e 1083
f45b8015 1084 /* We need 27 empty slots in iovec!
d14bcb4e
LP
1085 *
1086 * Note that if we fail on oom later on, we do not roll-back changes to the iovec structure. (It remains valid,
1087 * with the first n_iovec fields initialized.) */
3c171f0b 1088
3c171f0b 1089 uid_t owner_uid;
3c171f0b
LP
1090 pid_t pid;
1091 char *t;
9aa82023 1092 const char *p;
d14bcb4e 1093 int r, signo;
3c171f0b 1094
9aa82023 1095 r = parse_pid(context[CONTEXT_PID], &pid);
3c171f0b 1096 if (r < 0)
988e89ee 1097 return log_error_errno(r, "Failed to parse PID \"%s\": %m", context[CONTEXT_PID]);
3c171f0b 1098
ea5cc2a8 1099 r = get_process_comm(pid, &context[CONTEXT_COMM]);
3c171f0b
LP
1100 if (r < 0) {
1101 log_warning_errno(r, "Failed to get COMM, falling back to the command line: %m");
ea5cc2a8
ZJS
1102 context[CONTEXT_COMM] = strv_join(comm_fallback, " ");
1103 if (!context[CONTEXT_COMM])
3c171f0b
LP
1104 return log_oom();
1105 }
1106
ea5cc2a8 1107 r = get_process_exe(pid, &context[CONTEXT_EXE]);
3c171f0b
LP
1108 if (r < 0)
1109 log_warning_errno(r, "Failed to get EXE, ignoring: %m");
1110
92e92d71 1111 if (cg_pid_get_unit(pid, &context[CONTEXT_UNIT]) >= 0) {
b8cda92d 1112 if (!is_journald_crash((const char**) context)) {
92e92d71
ZJS
1113 /* OK, now we know it's not the journal, hence we can make use of it now. */
1114 log_set_target(LOG_TARGET_JOURNAL_OR_KMSG);
1115 log_open();
1116 }
3c171f0b 1117
c8091d92 1118 /* If this is PID 1 disable coredump collection, we'll unlikely be able to process it later on. */
b8cda92d 1119 if (is_pid1_crash((const char**) context)) {
c8091d92 1120 log_notice("Due to PID 1 having crashed coredump collection will now be turned off.");
e557b1a6 1121 disable_coredumps();
c8091d92
LP
1122 }
1123
92e92d71 1124 set_iovec_field(iovec, n_iovec, "COREDUMP_UNIT=", context[CONTEXT_UNIT]);
8c8549db 1125 }
803a3464 1126
9aa82023
ZJS
1127 if (cg_pid_get_user_unit(pid, &t) >= 0)
1128 set_iovec_field_free(iovec, n_iovec, "COREDUMP_USER_UNIT=", t);
3c171f0b 1129
9aa82023
ZJS
1130 /* The next few are mandatory */
1131 if (!set_iovec_field(iovec, n_iovec, "COREDUMP_PID=", context[CONTEXT_PID]))
1132 return log_oom();
3c171f0b 1133
9aa82023
ZJS
1134 if (!set_iovec_field(iovec, n_iovec, "COREDUMP_UID=", context[CONTEXT_UID]))
1135 return log_oom();
f5e04665 1136
9aa82023
ZJS
1137 if (!set_iovec_field(iovec, n_iovec, "COREDUMP_GID=", context[CONTEXT_GID]))
1138 return log_oom();
f5e04665 1139
9aa82023
ZJS
1140 if (!set_iovec_field(iovec, n_iovec, "COREDUMP_SIGNAL=", context[CONTEXT_SIGNAL]))
1141 return log_oom();
f5e04665 1142
9aa82023
ZJS
1143 if (!set_iovec_field(iovec, n_iovec, "COREDUMP_RLIMIT=", context[CONTEXT_RLIMIT]))
1144 return log_oom();
f5e04665 1145
f45b8015
JF
1146 if (!set_iovec_field(iovec, n_iovec, "COREDUMP_HOSTNAME=", context[CONTEXT_HOSTNAME]))
1147 return log_oom();
1148
ea5cc2a8 1149 if (!set_iovec_field(iovec, n_iovec, "COREDUMP_COMM=", context[CONTEXT_COMM]))
9aa82023 1150 return log_oom();
bdfd7b2c 1151
ea5cc2a8
ZJS
1152 if (context[CONTEXT_EXE] &&
1153 !set_iovec_field(iovec, n_iovec, "COREDUMP_EXE=", context[CONTEXT_EXE]))
9aa82023 1154 return log_oom();
f5e04665 1155
9aa82023
ZJS
1156 if (sd_pid_get_session(pid, &t) >= 0)
1157 set_iovec_field_free(iovec, n_iovec, "COREDUMP_SESSION=", t);
f5e04665 1158
a035f819 1159 if (sd_pid_get_owner_uid(pid, &owner_uid) >= 0) {
9aa82023 1160 r = asprintf(&t, "COREDUMP_OWNER_UID=" UID_FMT, owner_uid);
7de80bfe 1161 if (r > 0)
e6a7ec4b 1162 iovec[(*n_iovec)++] = IOVEC_MAKE_STRING(t);
f5e04665
LP
1163 }
1164
9aa82023
ZJS
1165 if (sd_pid_get_slice(pid, &t) >= 0)
1166 set_iovec_field_free(iovec, n_iovec, "COREDUMP_SLICE=", t);
f5e04665 1167
9aa82023
ZJS
1168 if (get_process_cmdline(pid, 0, false, &t) >= 0)
1169 set_iovec_field_free(iovec, n_iovec, "COREDUMP_CMDLINE=", t);
a035f819 1170
9aa82023
ZJS
1171 if (cg_pid_get_path_shifted(pid, NULL, &t) >= 0)
1172 set_iovec_field_free(iovec, n_iovec, "COREDUMP_CGROUP=", t);
a035f819 1173
9aa82023
ZJS
1174 if (compose_open_fds(pid, &t) >= 0)
1175 set_iovec_field_free(iovec, n_iovec, "COREDUMP_OPEN_FDS=", t);
3f132692
JF
1176
1177 p = procfs_file_alloca(pid, "status");
9aa82023
ZJS
1178 if (read_full_file(p, &t, NULL) >= 0)
1179 set_iovec_field_free(iovec, n_iovec, "COREDUMP_PROC_STATUS=", t);
3f132692
JF
1180
1181 p = procfs_file_alloca(pid, "maps");
9aa82023
ZJS
1182 if (read_full_file(p, &t, NULL) >= 0)
1183 set_iovec_field_free(iovec, n_iovec, "COREDUMP_PROC_MAPS=", t);
3f132692
JF
1184
1185 p = procfs_file_alloca(pid, "limits");
9aa82023
ZJS
1186 if (read_full_file(p, &t, NULL) >= 0)
1187 set_iovec_field_free(iovec, n_iovec, "COREDUMP_PROC_LIMITS=", t);
3f132692
JF
1188
1189 p = procfs_file_alloca(pid, "cgroup");
9aa82023
ZJS
1190 if (read_full_file(p, &t, NULL) >=0)
1191 set_iovec_field_free(iovec, n_iovec, "COREDUMP_PROC_CGROUP=", t);
3f132692 1192
d7032b1f 1193 p = procfs_file_alloca(pid, "mountinfo");
9aa82023
ZJS
1194 if (read_full_file(p, &t, NULL) >=0)
1195 set_iovec_field_free(iovec, n_iovec, "COREDUMP_PROC_MOUNTINFO=", t);
d7032b1f 1196
9aa82023
ZJS
1197 if (get_process_cwd(pid, &t) >= 0)
1198 set_iovec_field_free(iovec, n_iovec, "COREDUMP_CWD=", t);
3f132692
JF
1199
1200 if (get_process_root(pid, &t) >= 0) {
9aa82023
ZJS
1201 bool proc_self_root_is_slash;
1202
1203 proc_self_root_is_slash = strcmp(t, "/") == 0;
3f132692 1204
9aa82023 1205 set_iovec_field_free(iovec, n_iovec, "COREDUMP_ROOT=", t);
7ed03ce6
JF
1206
1207 /* If the process' root is "/", then there is a chance it has
1208 * mounted own root and hence being containerized. */
9aa82023
ZJS
1209 if (proc_self_root_is_slash && get_process_container_parent_cmdline(pid, &t) > 0)
1210 set_iovec_field_free(iovec, n_iovec, "COREDUMP_CONTAINER_CMDLINE=", t);
3f132692
JF
1211 }
1212
9aa82023
ZJS
1213 if (get_process_environ(pid, &t) >= 0)
1214 set_iovec_field_free(iovec, n_iovec, "COREDUMP_ENVIRON=", t);
1215
4600a396 1216 t = strjoin("COREDUMP_TIMESTAMP=", context[CONTEXT_TIMESTAMP], "000000");
9aa82023 1217 if (t)
e6a7ec4b 1218 iovec[(*n_iovec)++] = IOVEC_MAKE_STRING(t);
9aa82023 1219
d14bcb4e
LP
1220 if (safe_atoi(context[CONTEXT_SIGNAL], &signo) >= 0 && SIGNAL_VALID(signo))
1221 set_iovec_field(iovec, n_iovec, "COREDUMP_SIGNAL_NAME=SIG", signal_to_string(signo));
1222
92e92d71 1223 return 0; /* we successfully acquired all metadata */
9aa82023 1224}
3f132692 1225
9aa82023
ZJS
1226static int process_kernel(int argc, char* argv[]) {
1227
ea5cc2a8 1228 char* context[_CONTEXT_MAX] = {};
f45b8015 1229 struct iovec iovec[29 + SUBMIT_COREDUMP_FIELDS];
988e89ee 1230 size_t i, n_iovec, n_to_free = 0;
9aa82023
ZJS
1231 int r;
1232
988e89ee
ZJS
1233 log_debug("Processing coredump received from the kernel...");
1234
9aa82023 1235 if (argc < CONTEXT_COMM + 1) {
988e89ee 1236 log_error("Not enough arguments passed by the kernel (%i, expected %i).", argc - 1, CONTEXT_COMM + 1 - 1);
9aa82023 1237 return -EINVAL;
3f132692
JF
1238 }
1239
ea5cc2a8
ZJS
1240 context[CONTEXT_PID] = argv[1 + CONTEXT_PID];
1241 context[CONTEXT_UID] = argv[1 + CONTEXT_UID];
1242 context[CONTEXT_GID] = argv[1 + CONTEXT_GID];
1243 context[CONTEXT_SIGNAL] = argv[1 + CONTEXT_SIGNAL];
1244 context[CONTEXT_TIMESTAMP] = argv[1 + CONTEXT_TIMESTAMP];
1245 context[CONTEXT_RLIMIT] = argv[1 + CONTEXT_RLIMIT];
f45b8015 1246 context[CONTEXT_HOSTNAME] = argv[1 + CONTEXT_HOSTNAME];
9aa82023 1247
92e92d71
ZJS
1248 r = gather_pid_metadata(context, argv + 1 + CONTEXT_COMM, iovec, &n_to_free);
1249 if (r < 0)
86562420 1250 goto finish;
86562420 1251
988e89ee 1252 n_iovec = n_to_free;
f5e04665 1253
e6a7ec4b 1254 iovec[n_iovec++] = IOVEC_MAKE_STRING("MESSAGE_ID=" SD_MESSAGE_COREDUMP_STR);
4850d39a
LP
1255
1256 assert_cc(2 == LOG_CRIT);
e6a7ec4b 1257 iovec[n_iovec++] = IOVEC_MAKE_STRING("PRIORITY=2");
0dc5d23c 1258
3c171f0b 1259 assert(n_iovec <= ELEMENTSOF(iovec));
34c10968 1260
b8cda92d 1261 if (is_journald_crash((const char**) context) || is_pid1_crash((const char**) context))
92e92d71
ZJS
1262 r = submit_coredump((const char**) context,
1263 iovec, ELEMENTSOF(iovec), n_iovec,
1264 STDIN_FILENO);
1265 else
1266 r = send_iovec(iovec, n_iovec, STDIN_FILENO);
9aa82023
ZJS
1267
1268 finish:
1269 for (i = 0; i < n_to_free; i++)
1270 free(iovec[i].iov_base);
1271
92e92d71 1272 /* Those fields are allocated by gather_pid_metadata */
ea5cc2a8
ZJS
1273 free(context[CONTEXT_COMM]);
1274 free(context[CONTEXT_EXE]);
92e92d71 1275 free(context[CONTEXT_UNIT]);
ea5cc2a8 1276
9aa82023 1277 return r;
3c171f0b 1278}
34c10968 1279
988e89ee 1280static int process_backtrace(int argc, char *argv[]) {
ea5cc2a8
ZJS
1281 char *context[_CONTEXT_MAX] = {};
1282 _cleanup_free_ char *message = NULL;
5b45a160
ZJS
1283 _cleanup_free_ struct iovec *iovec = NULL;
1284 size_t n_iovec, n_allocated, n_to_free = 0, i;
988e89ee 1285 int r;
5b45a160
ZJS
1286 JournalImporter importer = {
1287 .fd = STDIN_FILENO,
1288 };
988e89ee
ZJS
1289
1290 log_debug("Processing backtrace on stdin...");
1291
1292 if (argc < CONTEXT_COMM + 1) {
1293 log_error("Not enough arguments passed (%i, expected %i).", argc - 1, CONTEXT_COMM + 1 - 1);
1294 return -EINVAL;
1295 }
1296
ea5cc2a8
ZJS
1297 context[CONTEXT_PID] = argv[2 + CONTEXT_PID];
1298 context[CONTEXT_UID] = argv[2 + CONTEXT_UID];
1299 context[CONTEXT_GID] = argv[2 + CONTEXT_GID];
1300 context[CONTEXT_SIGNAL] = argv[2 + CONTEXT_SIGNAL];
1301 context[CONTEXT_TIMESTAMP] = argv[2 + CONTEXT_TIMESTAMP];
1302 context[CONTEXT_RLIMIT] = argv[2 + CONTEXT_RLIMIT];
f45b8015 1303 context[CONTEXT_HOSTNAME] = argv[2 + CONTEXT_HOSTNAME];
988e89ee 1304
f45b8015
JF
1305 n_allocated = 34 + COREDUMP_STORAGE_EXTERNAL;
1306 /* 26 metadata, 2 static, +unknown input, 4 storage, rounded up */
5b45a160
ZJS
1307 iovec = new(struct iovec, n_allocated);
1308 if (!iovec)
1309 return log_oom();
1310
92e92d71 1311 r = gather_pid_metadata(context, argv + 2 + CONTEXT_COMM, iovec, &n_to_free);
988e89ee
ZJS
1312 if (r < 0)
1313 goto finish;
86562420
LP
1314 if (r > 0) {
1315 /* This was a special crash, and has already been processed. */
1316 r = 0;
1317 goto finish;
1318 }
5b45a160 1319 n_iovec = n_to_free;
988e89ee 1320
86562420 1321 for (;;) {
5b45a160
ZJS
1322 r = journal_importer_process_data(&importer);
1323 if (r < 0) {
1324 log_error_errno(r, "Failed to parse journal entry on stdin: %m");
1325 goto finish;
1326 }
d74dc4f2
ZJS
1327 if (r == 1 || /* complete entry */
1328 journal_importer_eof(&importer)) /* end of data */
5b45a160 1329 break;
988e89ee 1330 }
988e89ee 1331
5b45a160
ZJS
1332 if (!GREEDY_REALLOC(iovec, n_allocated, n_iovec + importer.iovw.count + 2))
1333 return log_oom();
1334
1335 if (journal_importer_eof(&importer)) {
1336 log_warning("Did not receive a full journal entry on stdin, ignoring message sent by reporter");
988e89ee 1337
ea5cc2a8
ZJS
1338 message = strjoin("MESSAGE=Process ", context[CONTEXT_PID],
1339 " (", context[CONTEXT_COMM], ")"
5b45a160
ZJS
1340 " of user ", context[CONTEXT_UID],
1341 " failed with ", context[CONTEXT_SIGNAL]);
1342 if (!message) {
1343 r = log_oom();
1344 goto finish;
1345 }
e6a7ec4b 1346 iovec[n_iovec++] = IOVEC_MAKE_STRING(message);
5b45a160
ZJS
1347 } else {
1348 for (i = 0; i < importer.iovw.count; i++)
1349 iovec[n_iovec++] = importer.iovw.iovec[i];
1350 }
988e89ee 1351
e6a7ec4b 1352 iovec[n_iovec++] = IOVEC_MAKE_STRING("MESSAGE_ID=" SD_MESSAGE_BACKTRACE_STR);
988e89ee 1353 assert_cc(2 == LOG_CRIT);
e6a7ec4b 1354 iovec[n_iovec++] = IOVEC_MAKE_STRING("PRIORITY=2");
988e89ee 1355
5b45a160 1356 assert(n_iovec <= n_allocated);
988e89ee
ZJS
1357
1358 r = sd_journal_sendv(iovec, n_iovec);
1359 if (r < 0)
1360 log_error_errno(r, "Failed to log backtrace: %m");
1361
1362 finish:
1363 for (i = 0; i < n_to_free; i++)
1364 free(iovec[i].iov_base);
1365
92e92d71 1366 /* Those fields are allocated by gather_pid_metadata */
ea5cc2a8
ZJS
1367 free(context[CONTEXT_COMM]);
1368 free(context[CONTEXT_EXE]);
92e92d71 1369 free(context[CONTEXT_UNIT]);
ea5cc2a8 1370
988e89ee
ZJS
1371 return r;
1372}
1373
3c171f0b
LP
1374int main(int argc, char *argv[]) {
1375 int r;
fee80f69 1376
9aa82023
ZJS
1377 /* First, log to a safe place, since we don't know what crashed and it might
1378 * be journald which we'd rather not log to then. */
8d4e028f 1379
3c171f0b
LP
1380 log_set_target(LOG_TARGET_KMSG);
1381 log_open();
8d4e028f 1382
3c171f0b
LP
1383 /* Make sure we never enter a loop */
1384 (void) prctl(PR_SET_DUMPABLE, 0);
8d4e028f 1385
3c171f0b
LP
1386 /* Ignore all parse errors */
1387 (void) parse_config();
fee80f69 1388
3c171f0b
LP
1389 log_debug("Selected storage '%s'.", coredump_storage_to_string(arg_storage));
1390 log_debug("Selected compression %s.", yes_no(arg_compress));
fee80f69 1391
3c171f0b
LP
1392 r = sd_listen_fds(false);
1393 if (r < 0) {
1394 log_error_errno(r, "Failed to determine number of file descriptor: %m");
1395 goto finish;
fee80f69
LP
1396 }
1397
9aa82023
ZJS
1398 /* If we got an fd passed, we are running in coredumpd mode. Otherwise we
1399 * are invoked from the kernel as coredump handler. */
988e89ee
ZJS
1400 if (r == 0) {
1401 if (streq_ptr(argv[1], "--backtrace"))
1402 r = process_backtrace(argc, argv);
1403 else
1404 r = process_kernel(argc, argv);
1405 } else if (r == 1)
3c171f0b
LP
1406 r = process_socket(SD_LISTEN_FDS_START);
1407 else {
1408 log_error("Received unexpected number of file descriptors.");
1409 r = -EINVAL;
1410 }
f5e04665
LP
1411
1412finish:
f5e04665
LP
1413 return r < 0 ? EXIT_FAILURE : EXIT_SUCCESS;
1414}