]> git.ipfire.org Git - thirdparty/systemd.git/blob - src/journal/journald-server.c
projects / thirdparty / systemd.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Merge pull request #30480 from keszybz/kernel-install-more-paths
[thirdparty/systemd.git] / src / journal / journald-server.c
1 /* SPDX-License-Identifier: LGPL-2.1-or-later */
2
3 #if HAVE_SELINUX
4 #include <selinux/selinux.h>
5 #endif
6 #include <sys/ioctl.h>
7 #include <sys/mman.h>
8 #include <sys/signalfd.h>
9 #include <sys/statvfs.h>
10 #include <linux/sockios.h>
11
12 #include "sd-daemon.h"
13 #include "sd-journal.h"
14 #include "sd-messages.h"
15
16 #include "acl-util.h"
17 #include "alloc-util.h"
18 #include "audit-util.h"
19 #include "cgroup-util.h"
20 #include "conf-parser.h"
21 #include "creds-util.h"
22 #include "dirent-util.h"
23 #include "extract-word.h"
24 #include "fd-util.h"
25 #include "fileio.h"
26 #include "format-util.h"
27 #include "fs-util.h"
28 #include "hashmap.h"
29 #include "hostname-util.h"
30 #include "id128-util.h"
31 #include "initrd-util.h"
32 #include "iovec-util.h"
33 #include "journal-authenticate.h"
34 #include "journal-file-util.h"
35 #include "journal-internal.h"
36 #include "journal-vacuum.h"
37 #include "journald-audit.h"
38 #include "journald-context.h"
39 #include "journald-kmsg.h"
40 #include "journald-native.h"
41 #include "journald-rate-limit.h"
42 #include "journald-server.h"
43 #include "journald-socket.h"
44 #include "journald-stream.h"
45 #include "journald-syslog.h"
46 #include "log.h"
47 #include "memory-util.h"
48 #include "missing_audit.h"
49 #include "mkdir.h"
50 #include "parse-util.h"
51 #include "path-util.h"
52 #include "proc-cmdline.h"
53 #include "process-util.h"
54 #include "rm-rf.h"
55 #include "selinux-util.h"
56 #include "signal-util.h"
57 #include "socket-netlink.h"
58 #include "socket-util.h"
59 #include "stdio-util.h"
60 #include "string-table.h"
61 #include "string-util.h"
62 #include "syslog-util.h"
63 #include "uid-classification.h"
64 #include "user-util.h"
65 #include "varlink-io.systemd.Journal.h"
66
67 #define USER_JOURNALS_MAX 1024
68
69 #define DEFAULT_SYNC_INTERVAL_USEC (5*USEC_PER_MINUTE)
70 #define DEFAULT_RATE_LIMIT_INTERVAL (30*USEC_PER_SEC)
71 #define DEFAULT_RATE_LIMIT_BURST 10000
72 #define DEFAULT_MAX_FILE_USEC USEC_PER_MONTH
73
74 #define DEFAULT_KMSG_OWN_INTERVAL (5 * USEC_PER_SEC)
75 #define DEFAULT_KMSG_OWN_BURST 50
76
77 #define RECHECK_SPACE_USEC (30*USEC_PER_SEC)
78
79 #define NOTIFY_SNDBUF_SIZE (8*1024*1024)
80
81 /* The period to insert between posting changes for coalescing */
82 #define POST_CHANGE_TIMER_INTERVAL_USEC (250*USEC_PER_MSEC)
83
84 /* Pick a good default that is likely to fit into AF_UNIX and AF_INET SOCK_DGRAM datagrams, and even leaves some room
85 * for a bit of additional metadata. */
86 #define DEFAULT_LINE_MAX (48*1024)
87
88 #define DEFERRED_CLOSES_MAX (4096)
89
90 #define IDLE_TIMEOUT_USEC (30*USEC_PER_SEC)
91
92 #define FAILED_TO_WRITE_ENTRY_RATELIMIT ((const RateLimit) { .interval = 1 * USEC_PER_SEC, .burst = 1 })
93
94 static int server_determine_path_usage(
95 Server *s,
96 const char *path,
97 uint64_t *ret_used,
98 uint64_t *ret_free) {
99
100 _cleanup_closedir_ DIR *d = NULL;
101 struct statvfs ss;
102
103 assert(s);
104 assert(path);
105 assert(ret_used);
106 assert(ret_free);
107
108 d = opendir(path);
109 if (!d)
110 return log_ratelimit_full_errno(errno == ENOENT ? LOG_DEBUG : LOG_ERR,
111 errno, JOURNAL_LOG_RATELIMIT, "Failed to open %s: %m", path);
112
113 if (fstatvfs(dirfd(d), &ss) < 0)
114 return log_ratelimit_error_errno(errno, JOURNAL_LOG_RATELIMIT,
115 "Failed to fstatvfs(%s): %m", path);
116
117 *ret_free = ss.f_bsize * ss.f_bavail;
118 *ret_used = 0;
119 FOREACH_DIRENT_ALL(de, d, break) {
120 struct stat st;
121
122 if (!endswith(de->d_name, ".journal") &&
123 !endswith(de->d_name, ".journal~"))
124 continue;
125
126 if (fstatat(dirfd(d), de->d_name, &st, AT_SYMLINK_NOFOLLOW) < 0) {
127 log_debug_errno(errno, "Failed to stat %s/%s, ignoring: %m", path, de->d_name);
128 continue;
129 }
130
131 if (!S_ISREG(st.st_mode))
132 continue;
133
134 *ret_used += (uint64_t) st.st_blocks * 512UL;
135 }
136
137 return 0;
138 }
139
140 static void cache_space_invalidate(JournalStorageSpace *space) {
141 zero(*space);
142 }
143
144 static int cache_space_refresh(Server *s, JournalStorage *storage) {
145 JournalStorageSpace *space;
146 JournalMetrics *metrics;
147 uint64_t vfs_used, vfs_avail, avail;
148 usec_t ts;
149 int r;
150
151 assert(s);
152
153 metrics = &storage->metrics;
154 space = &storage->space;
155
156 ts = now(CLOCK_MONOTONIC);
157
158 if (space->timestamp != 0 && usec_add(space->timestamp, RECHECK_SPACE_USEC) > ts)
159 return 0;
160
161 r = server_determine_path_usage(s, storage->path, &vfs_used, &vfs_avail);
162 if (r < 0)
163 return r;
164
165 space->vfs_used = vfs_used;
166 space->vfs_available = vfs_avail;
167
168 avail = LESS_BY(vfs_avail, metrics->keep_free);
169
170 space->limit = CLAMP(vfs_used + avail, metrics->min_use, metrics->max_use);
171 space->available = LESS_BY(space->limit, vfs_used);
172 space->timestamp = ts;
173 return 1;
174 }
175
176 static void patch_min_use(JournalStorage *storage) {
177 assert(storage);
178
179 /* Let's bump the min_use limit to the current usage on disk. We do
180 * this when starting up and first opening the journal files. This way
181 * sudden spikes in disk usage will not cause journald to vacuum files
182 * without bounds. Note that this means that only a restart of journald
183 * will make it reset this value. */
184
185 storage->metrics.min_use = MAX(storage->metrics.min_use, storage->space.vfs_used);
186 }
187
188 static JournalStorage* server_current_storage(Server *s) {
189 assert(s);
190
191 return s->system_journal ? &s->system_storage : &s->runtime_storage;
192 }
193
194 static int server_determine_space(Server *s, uint64_t *available, uint64_t *limit) {
195 JournalStorage *js;
196 int r;
197
198 assert(s);
199
200 js = server_current_storage(s);
201
202 r = cache_space_refresh(s, js);
203 if (r >= 0) {
204 if (available)
205 *available = js->space.available;
206 if (limit)
207 *limit = js->space.limit;
208 }
209 return r;
210 }
211
212 void server_space_usage_message(Server *s, JournalStorage *storage) {
213 assert(s);
214
215 if (!storage)
216 storage = server_current_storage(s);
217
218 if (cache_space_refresh(s, storage) < 0)
219 return;
220
221 const JournalMetrics *metrics = &storage->metrics;
222
223 server_driver_message(s, 0,
224 "MESSAGE_ID=" SD_MESSAGE_JOURNAL_USAGE_STR,
225 LOG_MESSAGE("%s (%s) is %s, max %s, %s free.",
226 storage->name, storage->path,
227 FORMAT_BYTES(storage->space.vfs_used),
228 FORMAT_BYTES(storage->space.limit),
229 FORMAT_BYTES(storage->space.available)),
230 "JOURNAL_NAME=%s", storage->name,
231 "JOURNAL_PATH=%s", storage->path,
232 "CURRENT_USE=%"PRIu64, storage->space.vfs_used,
233 "CURRENT_USE_PRETTY=%s", FORMAT_BYTES(storage->space.vfs_used),
234 "MAX_USE=%"PRIu64, metrics->max_use,
235 "MAX_USE_PRETTY=%s", FORMAT_BYTES(metrics->max_use),
236 "DISK_KEEP_FREE=%"PRIu64, metrics->keep_free,
237 "DISK_KEEP_FREE_PRETTY=%s", FORMAT_BYTES(metrics->keep_free),
238 "DISK_AVAILABLE=%"PRIu64, storage->space.vfs_available,
239 "DISK_AVAILABLE_PRETTY=%s", FORMAT_BYTES(storage->space.vfs_available),
240 "LIMIT=%"PRIu64, storage->space.limit,
241 "LIMIT_PRETTY=%s", FORMAT_BYTES(storage->space.limit),
242 "AVAILABLE=%"PRIu64, storage->space.available,
243 "AVAILABLE_PRETTY=%s", FORMAT_BYTES(storage->space.available),
244 NULL);
245 }
246
247 static void server_add_acls(JournalFile *f, uid_t uid) {
248 assert(f);
249
250 #if HAVE_ACL
251 int r;
252
253 if (uid_for_system_journal(uid))
254 return;
255
256 r = fd_add_uid_acl_permission(f->fd, uid, ACL_READ);
257 if (r < 0)
258 log_ratelimit_warning_errno(r, JOURNAL_LOG_RATELIMIT,
259 "Failed to set ACL on %s, ignoring: %m", f->path);
260 #endif
261 }
262
263 static int server_open_journal(
264 Server *s,
265 bool reliably,
266 const char *fname,
267 int open_flags,
268 bool seal,
269 JournalMetrics *metrics,
270 JournalFile **ret) {
271
272 _cleanup_(journal_file_offline_closep) JournalFile *f = NULL;
273 JournalFileFlags file_flags;
274 int r;
275
276 assert(s);
277 assert(fname);
278 assert(ret);
279
280 file_flags =
281 (s->compress.enabled ? JOURNAL_COMPRESS : 0) |
282 (seal ? JOURNAL_SEAL : 0) |
283 JOURNAL_STRICT_ORDER;
284
285 set_clear_with_destructor(s->deferred_closes, journal_file_offline_close);
286
287 if (reliably)
288 r = journal_file_open_reliably(
289 fname,
290 open_flags,
291 file_flags,
292 0640,
293 s->compress.threshold_bytes,
294 metrics,
295 s->mmap,
296 &f);
297 else
298 r = journal_file_open(
299 /* fd= */ -EBADF,
300 fname,
301 open_flags,
302 file_flags,
303 0640,
304 s->compress.threshold_bytes,
305 metrics,
306 s->mmap,
307 /* template= */ NULL,
308 &f);
309 if (r < 0)
310 return r;
311
312 r = journal_file_enable_post_change_timer(f, s->event, POST_CHANGE_TIMER_INTERVAL_USEC);
313 if (r < 0)
314 return r;
315
316 *ret = TAKE_PTR(f);
317 return r;
318 }
319
320 static bool server_flushed_flag_is_set(Server *s) {
321 const char *fn;
322
323 assert(s);
324
325 /* We don't support the "flushing" concept for namespace instances, we assume them to always have
326 * access to /var */
327 if (s->namespace)
328 return true;
329
330 fn = strjoina(s->runtime_directory, "/flushed");
331 return access(fn, F_OK) >= 0;
332 }
333
334 static int server_system_journal_open(
335 Server *s,
336 bool flush_requested,
337 bool relinquish_requested) {
338
339 const char *fn;
340 int r = 0;
341
342 if (!s->system_journal &&
343 IN_SET(s->storage, STORAGE_PERSISTENT, STORAGE_AUTO) &&
344 (flush_requested || server_flushed_flag_is_set(s)) &&
345 !relinquish_requested) {
346
347 /* If in auto mode: first try to create the machine path, but not the prefix.
348 *
349 * If in persistent mode: create /var/log/journal and the machine path */
350
351 if (s->storage == STORAGE_PERSISTENT)
352 (void) mkdir_parents(s->system_storage.path, 0755);
353
354 (void) mkdir(s->system_storage.path, 0755);
355
356 fn = strjoina(s->system_storage.path, "/system.journal");
357 r = server_open_journal(
358 s,
359 /* reliably= */ true,
360 fn,
361 O_RDWR|O_CREAT,
362 s->seal,
363 &s->system_storage.metrics,
364 &s->system_journal);
365 if (r >= 0) {
366 server_add_acls(s->system_journal, 0);
367 (void) cache_space_refresh(s, &s->system_storage);
368 patch_min_use(&s->system_storage);
369 } else {
370 if (!IN_SET(r, -ENOENT, -EROFS))
371 log_ratelimit_warning_errno(r, JOURNAL_LOG_RATELIMIT,
372 "Failed to open system journal: %m");
373
374 r = 0;
375 }
376
377 /* If the runtime journal is open, and we're post-flush, we're recovering from a failed
378 * system journal rotate (ENOSPC) for which the runtime journal was reopened.
379 *
380 * Perform an implicit flush to var, leaving the runtime journal closed, now that the system
381 * journal is back.
382 */
383 if (!flush_requested)
384 (void) server_flush_to_var(s, true);
385 }
386
387 if (!s->runtime_journal &&
388 (s->storage != STORAGE_NONE)) {
389
390 fn = strjoina(s->runtime_storage.path, "/system.journal");
391
392 if (!s->system_journal || relinquish_requested) {
393
394 /* OK, we really need the runtime journal, so create it if necessary. */
395
396 (void) mkdir_parents(s->runtime_storage.path, 0755);
397 (void) mkdir(s->runtime_storage.path, 0750);
398
399 r = server_open_journal(
400 s,
401 /* reliably= */ true,
402 fn,
403 O_RDWR|O_CREAT,
404 /* seal= */ false,
405 &s->runtime_storage.metrics,
406 &s->runtime_journal);
407 if (r < 0)
408 return log_ratelimit_warning_errno(r, JOURNAL_LOG_RATELIMIT,
409 "Failed to open runtime journal: %m");
410
411 } else if (!server_flushed_flag_is_set(s)) {
412 /* Try to open the runtime journal, but only if it already exists, so that we can
413 * flush it into the system journal */
414
415 r = server_open_journal(
416 s,
417 /* reliably= */ false,
418 fn,
419 O_RDWR,
420 /* seal= */ false,
421 &s->runtime_storage.metrics,
422 &s->runtime_journal);
423 if (r < 0) {
424 if (r != -ENOENT)
425 log_ratelimit_warning_errno(r, JOURNAL_LOG_RATELIMIT,
426 "Failed to open runtime journal: %m");
427
428 r = 0;
429 }
430 }
431
432 if (s->runtime_journal) {
433 server_add_acls(s->runtime_journal, 0);
434 (void) cache_space_refresh(s, &s->runtime_storage);
435 patch_min_use(&s->runtime_storage);
436 }
437 }
438
439 return r;
440 }
441
442 static int server_find_user_journal(Server *s, uid_t uid, JournalFile **ret) {
443 _cleanup_(journal_file_offline_closep) JournalFile *f = NULL;
444 _cleanup_free_ char *p = NULL;
445 int r;
446
447 assert(!uid_for_system_journal(uid));
448
449 f = ordered_hashmap_get(s->user_journals, UID_TO_PTR(uid));
450 if (f)
451 goto found;
452
453 if (asprintf(&p, "%s/user-" UID_FMT ".journal", s->system_storage.path, uid) < 0)
454 return log_oom();
455
456 /* Too many open? Then let's close one (or more) */
457 while (ordered_hashmap_size(s->user_journals) >= USER_JOURNALS_MAX) {
458 JournalFile *first;
459
460 assert_se(first = ordered_hashmap_steal_first(s->user_journals));
461 (void) journal_file_offline_close(first);
462 }
463
464 r = server_open_journal(
465 s,
466 /* reliably= */ true,
467 p,
468 O_RDWR|O_CREAT,
469 s->seal,
470 &s->system_storage.metrics,
471 &f);
472 if (r < 0)
473 return r;
474
475 r = ordered_hashmap_put(s->user_journals, UID_TO_PTR(uid), f);
476 if (r < 0)
477 return r;
478
479 server_add_acls(f, uid);
480
481 found:
482 *ret = TAKE_PTR(f);
483 return 0;
484 }
485
486 static JournalFile* server_find_journal(Server *s, uid_t uid) {
487 int r;
488
489 assert(s);
490
491 /* A rotate that fails to create the new journal (ENOSPC) leaves the rotated journal as NULL. Unless
492 * we revisit opening, even after space is made available we'll continue to return NULL indefinitely.
493 *
494 * system_journal_open() is a noop if the journals are already open, so we can just call it here to
495 * recover from failed rotates (or anything else that's left the journals as NULL).
496 *
497 * Fixes https://github.com/systemd/systemd/issues/3968 */
498 (void) server_system_journal_open(s, /* flush_requested= */ false, /* relinquish_requested= */ false);
499
500 /* We split up user logs only on /var, not on /run. If the runtime file is open, we write to it
501 * exclusively, in order to guarantee proper order as soon as we flush /run to /var and close the
502 * runtime file. */
503
504 if (s->runtime_journal)
505 return s->runtime_journal;
506
507 /* If we are not in persistent mode, then we need return NULL immediately rather than opening a
508 * persistent journal of any sort.
509 *
510 * Fixes https://github.com/systemd/systemd/issues/20390 */
511 if (!IN_SET(s->storage, STORAGE_AUTO, STORAGE_PERSISTENT))
512 return NULL;
513
514 if (!uid_for_system_journal(uid)) {
515 JournalFile *f = NULL;
516
517 r = server_find_user_journal(s, uid, &f);
518 if (r >= 0)
519 return ASSERT_PTR(f);
520
521 log_warning_errno(r, "Failed to open user journal file, falling back to system journal: %m");
522 }
523
524 return s->system_journal;
525 }
526
527 static int server_do_rotate(
528 Server *s,
529 JournalFile **f,
530 const char* name,
531 bool seal,
532 uint32_t uid) {
533
534 JournalFileFlags file_flags;
535 int r;
536
537 assert(s);
538
539 if (!*f)
540 return -EINVAL;
541
542 file_flags =
543 (s->compress.enabled ? JOURNAL_COMPRESS : 0)|
544 (seal ? JOURNAL_SEAL : 0) |
545 JOURNAL_STRICT_ORDER;
546
547 r = journal_file_rotate(f, s->mmap, file_flags, s->compress.threshold_bytes, s->deferred_closes);
548 if (r < 0) {
549 if (*f)
550 return log_ratelimit_error_errno(r, JOURNAL_LOG_RATELIMIT,
551 "Failed to rotate %s: %m", (*f)->path);
552 else
553 return log_ratelimit_error_errno(r, JOURNAL_LOG_RATELIMIT,
554 "Failed to create new %s journal: %m", name);
555 }
556
557 server_add_acls(*f, uid);
558 return r;
559 }
560
561 static void server_process_deferred_closes(Server *s) {
562 JournalFile *f;
563
564 /* Perform any deferred closes which aren't still offlining. */
565 SET_FOREACH(f, s->deferred_closes) {
566 if (journal_file_is_offlining(f))
567 continue;
568
569 (void) set_remove(s->deferred_closes, f);
570 (void) journal_file_offline_close(f);
571 }
572 }
573
574 static void server_vacuum_deferred_closes(Server *s) {
575 assert(s);
576
577 /* Make some room in the deferred closes list, so that it doesn't grow without bounds */
578 if (set_size(s->deferred_closes) < DEFERRED_CLOSES_MAX)
579 return;
580
581 /* Let's first remove all journal files that might already have completed closing */
582 server_process_deferred_closes(s);
583
584 /* And now, let's close some more until we reach the limit again. */
585 while (set_size(s->deferred_closes) >= DEFERRED_CLOSES_MAX) {
586 JournalFile *f;
587
588 assert_se(f = set_steal_first(s->deferred_closes));
589 journal_file_offline_close(f);
590 }
591 }
592
593 static int server_archive_offline_user_journals(Server *s) {
594 _cleanup_closedir_ DIR *d = NULL;
595 int r;
596
597 assert(s);
598
599 d = opendir(s->system_storage.path);
600 if (!d) {
601 if (errno == ENOENT)
602 return 0;
603
604 return log_ratelimit_error_errno(errno, JOURNAL_LOG_RATELIMIT,
605 "Failed to open %s: %m", s->system_storage.path);
606 }
607
608 for (;;) {
609 _cleanup_free_ char *full = NULL;
610 _cleanup_close_ int fd = -EBADF;
611 struct dirent *de;
612 JournalFile *f;
613 uid_t uid;
614
615 errno = 0;
616 de = readdir_no_dot(d);
617 if (!de) {
618 if (errno != 0)
619 log_ratelimit_warning_errno(errno, JOURNAL_LOG_RATELIMIT,
620 "Failed to enumerate %s, ignoring: %m",
621 s->system_storage.path);
622 break;
623 }
624
625 r = journal_file_parse_uid_from_filename(de->d_name, &uid);
626 if (r < 0) {
627 /* Don't warn if the file is not an online or offline user journal. */
628 if (r != -EREMOTE)
629 log_warning_errno(r, "Failed to parse UID from file name '%s', ignoring: %m", de->d_name);
630 continue;
631 }
632
633 /* Already rotated in the above loop? i.e. is it an open user journal? */
634 if (ordered_hashmap_contains(s->user_journals, UID_TO_PTR(uid)))
635 continue;
636
637 full = path_join(s->system_storage.path, de->d_name);
638 if (!full)
639 return log_oom();
640
641 fd = openat(dirfd(d), de->d_name, O_RDWR|O_CLOEXEC|O_NOCTTY|O_NOFOLLOW|O_NONBLOCK);
642 if (fd < 0) {
643 log_ratelimit_full_errno(IN_SET(errno, ELOOP, ENOENT) ? LOG_DEBUG : LOG_WARNING,
644 errno, JOURNAL_LOG_RATELIMIT,
645 "Failed to open journal file '%s' for rotation: %m", full);
646 continue;
647 }
648
649 /* Make some room in the set of deferred close()s */
650 server_vacuum_deferred_closes(s);
651
652 /* Open the file briefly, so that we can archive it */
653 r = journal_file_open(
654 fd,
655 full,
656 O_RDWR,
657 (s->compress.enabled ? JOURNAL_COMPRESS : 0) |
658 (s->seal ? JOURNAL_SEAL : 0), /* strict order does not matter here */
659 0640,
660 s->compress.threshold_bytes,
661 &s->system_storage.metrics,
662 s->mmap,
663 /* template= */ NULL,
664 &f);
665 if (r < 0) {
666 log_ratelimit_warning_errno(r, JOURNAL_LOG_RATELIMIT,
667 "Failed to read journal file %s for rotation, trying to move it out of the way: %m",
668 full);
669
670 r = journal_file_dispose(dirfd(d), de->d_name);
671 if (r < 0)
672 log_ratelimit_warning_errno(r, JOURNAL_LOG_RATELIMIT,
673 "Failed to move %s out of the way, ignoring: %m",
674 full);
675 else
676 log_debug("Successfully moved %s out of the way.", full);
677
678 continue;
679 }
680
681 TAKE_FD(fd); /* Donated to journal_file_open() */
682
683 journal_file_write_final_tag(f);
684 r = journal_file_archive(f, NULL);
685 if (r < 0)
686 log_debug_errno(r, "Failed to archive journal file '%s', ignoring: %m", full);
687
688 journal_file_initiate_close(TAKE_PTR(f), s->deferred_closes);
689 }
690
691 return 0;
692 }
693
694 void server_rotate(Server *s) {
695 JournalFile *f;
696 void *k;
697 int r;
698
699 log_debug("Rotating...");
700
701 /* First, rotate the system journal (either in its runtime flavour or in its runtime flavour) */
702 (void) server_do_rotate(s, &s->runtime_journal, "runtime", /* seal= */ false, /* uid= */ 0);
703 (void) server_do_rotate(s, &s->system_journal, "system", s->seal, /* uid= */ 0);
704
705 /* Then, rotate all user journals we have open (keeping them open) */
706 ORDERED_HASHMAP_FOREACH_KEY(f, k, s->user_journals) {
707 r = server_do_rotate(s, &f, "user", s->seal, PTR_TO_UID(k));
708 if (r >= 0)
709 ordered_hashmap_replace(s->user_journals, k, f);
710 else if (!f)
711 /* Old file has been closed and deallocated */
712 ordered_hashmap_remove(s->user_journals, k);
713 }
714
715 /* Finally, also rotate all user journals we currently do not have open. (But do so only if we
716 * actually have access to /var, i.e. are not in the log-to-runtime-journal mode). */
717 if (!s->runtime_journal)
718 (void) server_archive_offline_user_journals(s);
719
720 server_process_deferred_closes(s);
721 }
722
723 void server_sync(Server *s) {
724 JournalFile *f;
725 int r;
726
727 if (s->system_journal) {
728 r = journal_file_set_offline(s->system_journal, false);
729 if (r < 0)
730 log_ratelimit_warning_errno(r, JOURNAL_LOG_RATELIMIT,
731 "Failed to sync system journal, ignoring: %m");
732 }
733
734 ORDERED_HASHMAP_FOREACH(f, s->user_journals) {
735 r = journal_file_set_offline(f, false);
736 if (r < 0)
737 log_ratelimit_warning_errno(r, JOURNAL_LOG_RATELIMIT,
738 "Failed to sync user journal, ignoring: %m");
739 }
740
741 if (s->sync_event_source) {
742 r = sd_event_source_set_enabled(s->sync_event_source, SD_EVENT_OFF);
743 if (r < 0)
744 log_ratelimit_error_errno(r, JOURNAL_LOG_RATELIMIT,
745 "Failed to disable sync timer source: %m");
746 }
747
748 s->sync_scheduled = false;
749 }
750
751 static void server_do_vacuum(Server *s, JournalStorage *storage, bool verbose) {
752
753 int r;
754
755 assert(s);
756 assert(storage);
757
758 (void) cache_space_refresh(s, storage);
759
760 if (verbose)
761 server_space_usage_message(s, storage);
762
763 r = journal_directory_vacuum(storage->path, storage->space.limit,
764 storage->metrics.n_max_files, s->max_retention_usec,
765 &s->oldest_file_usec, verbose);
766 if (r < 0 && r != -ENOENT)
767 log_ratelimit_warning_errno(r, JOURNAL_LOG_RATELIMIT,
768 "Failed to vacuum %s, ignoring: %m", storage->path);
769
770 cache_space_invalidate(&storage->space);
771 }
772
773 void server_vacuum(Server *s, bool verbose) {
774 assert(s);
775
776 log_debug("Vacuuming...");
777
778 s->oldest_file_usec = 0;
779
780 if (s->system_journal)
781 server_do_vacuum(s, &s->system_storage, verbose);
782 if (s->runtime_journal)
783 server_do_vacuum(s, &s->runtime_storage, verbose);
784 }
785
786 static void server_cache_machine_id(Server *s) {
787 sd_id128_t id;
788 int r;
789
790 assert(s);
791
792 r = sd_id128_get_machine(&id);
793 if (r < 0)
794 return;
795
796 sd_id128_to_string(id, stpcpy(s->machine_id_field, "_MACHINE_ID="));
797 }
798
799 static void server_cache_boot_id(Server *s) {
800 sd_id128_t id;
801 int r;
802
803 assert(s);
804
805 r = sd_id128_get_boot(&id);
806 if (r < 0)
807 return;
808
809 sd_id128_to_string(id, stpcpy(s->boot_id_field, "_BOOT_ID="));
810 }
811
812 static void server_cache_hostname(Server *s) {
813 _cleanup_free_ char *t = NULL;
814 char *x;
815
816 assert(s);
817
818 t = gethostname_malloc();
819 if (!t)
820 return;
821
822 x = strjoin("_HOSTNAME=", t);
823 if (!x)
824 return;
825
826 free_and_replace(s->hostname_field, x);
827 }
828
829 static bool shall_try_append_again(JournalFile *f, int r) {
830 switch (r) {
831
832 case -E2BIG: /* Hit configured limit */
833 case -EFBIG: /* Hit fs limit */
834 case -EDQUOT: /* Quota limit hit */
835 case -ENOSPC: /* Disk full */
836 log_debug_errno(r, "%s: Allocation limit reached, rotating.", f->path);
837 return true;
838
839 case -EROFS: /* Read-only file system */
840 /* When appending an entry fails if shall_try_append_again returns true, the journal is
841 * rotated. If the FS is read-only, rotation will fail and s->system_journal will be set to
842 * NULL. After that, when find_journal will try to open the journal since s->system_journal
843 * will be NULL, it will open the runtime journal. */
844 log_ratelimit_warning_errno(r, JOURNAL_LOG_RATELIMIT, "%s: Read-only file system, rotating.", f->path);
845 return true;
846
847 case -EIO: /* I/O error of some kind (mmap) */
848 log_ratelimit_warning_errno(r, JOURNAL_LOG_RATELIMIT, "%s: IO error, rotating.", f->path);
849 return true;
850
851 case -EHOSTDOWN: /* Other machine */
852 log_ratelimit_info_errno(r, JOURNAL_LOG_RATELIMIT, "%s: Journal file from other machine, rotating.", f->path);
853 return true;
854
855 case -EBUSY: /* Unclean shutdown */
856 log_ratelimit_info_errno(r, JOURNAL_LOG_RATELIMIT, "%s: Unclean shutdown, rotating.", f->path);
857 return true;
858
859 case -EPROTONOSUPPORT: /* Unsupported feature */
860 log_ratelimit_info_errno(r, JOURNAL_LOG_RATELIMIT, "%s: Unsupported feature, rotating.", f->path);
861 return true;
862
863 case -EBADMSG: /* Corrupted */
864 case -ENODATA: /* Truncated */
865 case -ESHUTDOWN: /* Already archived */
866 case -EADDRNOTAVAIL: /* Referenced object offset out of bounds */
867 log_ratelimit_info_errno(r, JOURNAL_LOG_RATELIMIT, "%s: Journal file corrupted, rotating.", f->path);
868 return true;
869
870 case -EIDRM: /* Journal file has been deleted */
871 log_ratelimit_info_errno(r, JOURNAL_LOG_RATELIMIT, "%s: Journal file has been deleted, rotating.", f->path);
872 return true;
873
874 case -EREMCHG: /* Wallclock time (CLOCK_REALTIME) jumped backwards relative to last journal entry */
875 log_ratelimit_info_errno(r, JOURNAL_LOG_RATELIMIT, "%s: Realtime clock jumped backwards relative to last journal entry, rotating.", f->path);
876 return true;
877
878 case -ENOTNAM: /* Monotonic time (CLOCK_MONOTONIC) jumped backwards relative to last journal entry with the same boot ID */
879 log_ratelimit_info_errno(
880 r,
881 JOURNAL_LOG_RATELIMIT,
882 "%s: Monotonic clock jumped backwards relative to last journal entry with the same boot ID, rotating.",
883 f->path);
884 return true;
885
886 case -EILSEQ: /* seqnum ID last used in the file doesn't match the one we'd passed when writing an entry to it */
887 log_ratelimit_info_errno(r, JOURNAL_LOG_RATELIMIT, "%s: Journal file uses a different sequence number ID, rotating.", f->path);
888 return true;
889
890 case -EAFNOSUPPORT:
891 log_ratelimit_error_errno(r, JOURNAL_LOG_RATELIMIT, "%s: Underlying file system does not support memory mapping or another required file system feature.", f->path);
892 return false;
893
894 default:
895 log_ratelimit_error_errno(r, JOURNAL_LOG_RATELIMIT, "%s: Unexpected error while writing to journal file: %m", f->path);
896 return false;
897 }
898 }
899
900 static void server_write_to_journal(
901 Server *s,
902 uid_t uid,
903 const struct iovec *iovec,
904 size_t n,
905 int priority) {
906
907 bool vacuumed = false, rotate = false;
908 struct dual_timestamp ts;
909 JournalFile *f;
910 int r;
911
912 assert(s);
913 assert(iovec);
914 assert(n > 0);
915
916 /* Get the closest, linearized time we have for this log event from the event loop. (Note that we do not use
917 * the source time, and not even the time the event was originally seen, but instead simply the time we started
918 * processing it, as we want strictly linear ordering in what we write out.) */
919 assert_se(sd_event_now(s->event, CLOCK_REALTIME, &ts.realtime) >= 0);
920 assert_se(sd_event_now(s->event, CLOCK_MONOTONIC, &ts.monotonic) >= 0);
921
922 if (ts.realtime < s->last_realtime_clock) {
923 /* When the time jumps backwards, let's immediately rotate. Of course, this should not happen during
924 * regular operation. However, when it does happen, then we should make sure that we start fresh files
925 * to ensure that the entries in the journal files are strictly ordered by time, in order to ensure
926 * bisection works correctly. */
927
928 log_ratelimit_info(JOURNAL_LOG_RATELIMIT, "Time jumped backwards, rotating.");
929 rotate = true;
930 } else {
931
932 f = server_find_journal(s, uid);
933 if (!f)
934 return;
935
936 if (journal_file_rotate_suggested(f, s->max_file_usec, LOG_DEBUG)) {
937 log_debug("%s: Journal header limits reached or header out-of-date, rotating.",
938 f->path);
939 rotate = true;
940 }
941 }
942
943 if (rotate) {
944 server_rotate(s);
945 server_vacuum(s, false);
946 vacuumed = true;
947
948 f = server_find_journal(s, uid);
949 if (!f)
950 return;
951 }
952
953 s->last_realtime_clock = ts.realtime;
954
955 r = journal_file_append_entry(
956 f,
957 &ts,
958 /* boot_id= */ NULL,
959 iovec, n,
960 &s->seqnum->seqnum,
961 &s->seqnum->id,
962 /* ret_object= */ NULL,
963 /* ret_offset= */ NULL);
964 if (r >= 0) {
965 server_schedule_sync(s, priority);
966 return;
967 }
968
969 log_debug_errno(r, "Failed to write entry to %s (%zu items, %zu bytes): %m", f->path, n, iovec_total_size(iovec, n));
970
971 if (!shall_try_append_again(f, r))
972 return;
973 if (vacuumed) {
974 log_ratelimit_warning_errno(r, JOURNAL_LOG_RATELIMIT,
975 "Suppressing rotation, as we already rotated immediately before write attempt. Giving up.");
976 return;
977 }
978
979 server_rotate(s);
980 server_vacuum(s, false);
981
982 f = server_find_journal(s, uid);
983 if (!f)
984 return;
985
986 log_debug_errno(r, "Retrying write.");
987 r = journal_file_append_entry(
988 f,
989 &ts,
990 /* boot_id= */ NULL,
991 iovec, n,
992 &s->seqnum->seqnum,
993 &s->seqnum->id,
994 /* ret_object= */ NULL,
995 /* ret_offset= */ NULL);
996 if (r < 0)
997 log_ratelimit_error_errno(r, FAILED_TO_WRITE_ENTRY_RATELIMIT,
998 "Failed to write entry to %s (%zu items, %zu bytes) despite vacuuming, ignoring: %m",
999 f->path, n, iovec_total_size(iovec, n));
1000 else
1001 server_schedule_sync(s, priority);
1002 }
1003
1004 #define IOVEC_ADD_NUMERIC_FIELD(iovec, n, value, type, isset, format, field) \
1005 if (isset(value)) { \
1006 char *k; \
1007 k = newa(char, STRLEN(field "=") + DECIMAL_STR_MAX(type) + 1); \
1008 sprintf(k, field "=" format, value); \
1009 iovec[n++] = IOVEC_MAKE_STRING(k); \
1010 }
1011
1012 #define IOVEC_ADD_STRING_FIELD(iovec, n, value, field) \
1013 if (!isempty(value)) { \
1014 char *k; \
1015 k = strjoina(field "=", value); \
1016 iovec[n++] = IOVEC_MAKE_STRING(k); \
1017 }
1018
1019 #define IOVEC_ADD_ID128_FIELD(iovec, n, value, field) \
1020 if (!sd_id128_is_null(value)) { \
1021 char *k; \
1022 k = newa(char, STRLEN(field "=") + SD_ID128_STRING_MAX); \
1023 sd_id128_to_string(value, stpcpy(k, field "=")); \
1024 iovec[n++] = IOVEC_MAKE_STRING(k); \
1025 }
1026
1027 #define IOVEC_ADD_SIZED_FIELD(iovec, n, value, value_size, field) \
1028 if (value_size > 0) { \
1029 char *k; \
1030 k = newa(char, STRLEN(field "=") + value_size + 1); \
1031 *((char*) mempcpy(stpcpy(k, field "="), value, value_size)) = 0; \
1032 iovec[n++] = IOVEC_MAKE_STRING(k); \
1033 } \
1034
1035 static void server_dispatch_message_real(
1036 Server *s,
1037 struct iovec *iovec, size_t n, size_t m,
1038 const ClientContext *c,
1039 const struct timeval *tv,
1040 int priority,
1041 pid_t object_pid) {
1042
1043 char source_time[sizeof("_SOURCE_REALTIME_TIMESTAMP=") + DECIMAL_STR_MAX(usec_t)];
1044 _unused_ _cleanup_free_ char *cmdline1 = NULL, *cmdline2 = NULL;
1045 uid_t journal_uid;
1046 ClientContext *o;
1047
1048 assert(s);
1049 assert(iovec);
1050 assert(n > 0);
1051 assert(n +
1052 N_IOVEC_META_FIELDS +
1053 (pid_is_valid(object_pid) ? N_IOVEC_OBJECT_FIELDS : 0) +
1054 client_context_extra_fields_n_iovec(c) <= m);
1055
1056 if (c) {
1057 IOVEC_ADD_NUMERIC_FIELD(iovec, n, c->pid, pid_t, pid_is_valid, PID_FMT, "_PID");
1058 IOVEC_ADD_NUMERIC_FIELD(iovec, n, c->uid, uid_t, uid_is_valid, UID_FMT, "_UID");
1059 IOVEC_ADD_NUMERIC_FIELD(iovec, n, c->gid, gid_t, gid_is_valid, GID_FMT, "_GID");
1060
1061 IOVEC_ADD_STRING_FIELD(iovec, n, c->comm, "_COMM"); /* At most TASK_COMM_LENGTH (16 bytes) */
1062 IOVEC_ADD_STRING_FIELD(iovec, n, c->exe, "_EXE"); /* A path, so at most PATH_MAX (4096 bytes) */
1063
1064 if (c->cmdline)
1065 /* At most _SC_ARG_MAX (2MB usually), which is too much to put on stack.
1066 * Let's use a heap allocation for this one. */
1067 cmdline1 = set_iovec_string_field(iovec, &n, "_CMDLINE=", c->cmdline);
1068
1069 IOVEC_ADD_STRING_FIELD(iovec, n, c->capeff, "_CAP_EFFECTIVE"); /* Read from /proc/.../status */
1070 IOVEC_ADD_SIZED_FIELD(iovec, n, c->label, c->label_size, "_SELINUX_CONTEXT");
1071 IOVEC_ADD_NUMERIC_FIELD(iovec, n, c->auditid, uint32_t, audit_session_is_valid, "%" PRIu32, "_AUDIT_SESSION");
1072 IOVEC_ADD_NUMERIC_FIELD(iovec, n, c->loginuid, uid_t, uid_is_valid, UID_FMT, "_AUDIT_LOGINUID");
1073
1074 IOVEC_ADD_STRING_FIELD(iovec, n, c->cgroup, "_SYSTEMD_CGROUP"); /* A path */
1075 IOVEC_ADD_STRING_FIELD(iovec, n, c->session, "_SYSTEMD_SESSION");
1076 IOVEC_ADD_NUMERIC_FIELD(iovec, n, c->owner_uid, uid_t, uid_is_valid, UID_FMT, "_SYSTEMD_OWNER_UID");
1077 IOVEC_ADD_STRING_FIELD(iovec, n, c->unit, "_SYSTEMD_UNIT"); /* Unit names are bounded by UNIT_NAME_MAX */
1078 IOVEC_ADD_STRING_FIELD(iovec, n, c->user_unit, "_SYSTEMD_USER_UNIT");
1079 IOVEC_ADD_STRING_FIELD(iovec, n, c->slice, "_SYSTEMD_SLICE");
1080 IOVEC_ADD_STRING_FIELD(iovec, n, c->user_slice, "_SYSTEMD_USER_SLICE");
1081
1082 IOVEC_ADD_ID128_FIELD(iovec, n, c->invocation_id, "_SYSTEMD_INVOCATION_ID");
1083
1084 if (c->extra_fields_n_iovec > 0) {
1085 memcpy(iovec + n, c->extra_fields_iovec, c->extra_fields_n_iovec * sizeof(struct iovec));
1086 n += c->extra_fields_n_iovec;
1087 }
1088 }
1089
1090 assert(n <= m);
1091
1092 if (pid_is_valid(object_pid) && client_context_get(s, object_pid, NULL, NULL, 0, NULL, &o) >= 0) {
1093
1094 IOVEC_ADD_NUMERIC_FIELD(iovec, n, o->pid, pid_t, pid_is_valid, PID_FMT, "OBJECT_PID");
1095 IOVEC_ADD_NUMERIC_FIELD(iovec, n, o->uid, uid_t, uid_is_valid, UID_FMT, "OBJECT_UID");
1096 IOVEC_ADD_NUMERIC_FIELD(iovec, n, o->gid, gid_t, gid_is_valid, GID_FMT, "OBJECT_GID");
1097
1098 /* See above for size limits, only ->cmdline may be large, so use a heap allocation for it. */
1099 IOVEC_ADD_STRING_FIELD(iovec, n, o->comm, "OBJECT_COMM");
1100 IOVEC_ADD_STRING_FIELD(iovec, n, o->exe, "OBJECT_EXE");
1101 if (o->cmdline)
1102 cmdline2 = set_iovec_string_field(iovec, &n, "OBJECT_CMDLINE=", o->cmdline);
1103
1104 IOVEC_ADD_STRING_FIELD(iovec, n, o->capeff, "OBJECT_CAP_EFFECTIVE");
1105 IOVEC_ADD_SIZED_FIELD(iovec, n, o->label, o->label_size, "OBJECT_SELINUX_CONTEXT");
1106 IOVEC_ADD_NUMERIC_FIELD(iovec, n, o->auditid, uint32_t, audit_session_is_valid, "%" PRIu32, "OBJECT_AUDIT_SESSION");
1107 IOVEC_ADD_NUMERIC_FIELD(iovec, n, o->loginuid, uid_t, uid_is_valid, UID_FMT, "OBJECT_AUDIT_LOGINUID");
1108
1109 IOVEC_ADD_STRING_FIELD(iovec, n, o->cgroup, "OBJECT_SYSTEMD_CGROUP");
1110 IOVEC_ADD_STRING_FIELD(iovec, n, o->session, "OBJECT_SYSTEMD_SESSION");
1111 IOVEC_ADD_NUMERIC_FIELD(iovec, n, o->owner_uid, uid_t, uid_is_valid, UID_FMT, "OBJECT_SYSTEMD_OWNER_UID");
1112 IOVEC_ADD_STRING_FIELD(iovec, n, o->unit, "OBJECT_SYSTEMD_UNIT");
1113 IOVEC_ADD_STRING_FIELD(iovec, n, o->user_unit, "OBJECT_SYSTEMD_USER_UNIT");
1114 IOVEC_ADD_STRING_FIELD(iovec, n, o->slice, "OBJECT_SYSTEMD_SLICE");
1115 IOVEC_ADD_STRING_FIELD(iovec, n, o->user_slice, "OBJECT_SYSTEMD_USER_SLICE");
1116
1117 IOVEC_ADD_ID128_FIELD(iovec, n, o->invocation_id, "OBJECT_SYSTEMD_INVOCATION_ID=");
1118 }
1119
1120 assert(n <= m);
1121
1122 if (tv) {
1123 sprintf(source_time, "_SOURCE_REALTIME_TIMESTAMP=" USEC_FMT, timeval_load(tv));
1124 iovec[n++] = IOVEC_MAKE_STRING(source_time);
1125 }
1126
1127 /* Note that strictly speaking storing the boot id here is
1128 * redundant since the entry includes this in-line
1129 * anyway. However, we need this indexed, too. */
1130 if (!isempty(s->boot_id_field))
1131 iovec[n++] = IOVEC_MAKE_STRING(s->boot_id_field);
1132
1133 if (!isempty(s->machine_id_field))
1134 iovec[n++] = IOVEC_MAKE_STRING(s->machine_id_field);
1135
1136 if (!isempty(s->hostname_field))
1137 iovec[n++] = IOVEC_MAKE_STRING(s->hostname_field);
1138
1139 if (!isempty(s->namespace_field))
1140 iovec[n++] = IOVEC_MAKE_STRING(s->namespace_field);
1141
1142 iovec[n++] = in_initrd() ? IOVEC_MAKE_STRING("_RUNTIME_SCOPE=initrd") : IOVEC_MAKE_STRING("_RUNTIME_SCOPE=system");
1143 assert(n <= m);
1144
1145 if (s->split_mode == SPLIT_UID && c && uid_is_valid(c->uid))
1146 /* Split up strictly by (non-root) UID */
1147 journal_uid = c->uid;
1148 else if (s->split_mode == SPLIT_LOGIN && c && c->uid > 0 && uid_is_valid(c->owner_uid))
1149 /* Split up by login UIDs. We do this only if the
1150 * realuid is not root, in order not to accidentally
1151 * leak privileged information to the user that is
1152 * logged by a privileged process that is part of an
1153 * unprivileged session. */
1154 journal_uid = c->owner_uid;
1155 else
1156 journal_uid = 0;
1157
1158 (void) server_forward_socket(s, iovec, n, priority);
1159
1160 server_write_to_journal(s, journal_uid, iovec, n, priority);
1161 }
1162
1163 void server_driver_message(Server *s, pid_t object_pid, const char *message_id, const char *format, ...) {
1164
1165 struct iovec *iovec;
1166 size_t n = 0, k, m;
1167 va_list ap;
1168 int r;
1169
1170 assert(s);
1171 assert(format);
1172
1173 m = N_IOVEC_META_FIELDS + 5 + N_IOVEC_PAYLOAD_FIELDS + client_context_extra_fields_n_iovec(s->my_context) + N_IOVEC_OBJECT_FIELDS;
1174 iovec = newa(struct iovec, m);
1175
1176 assert_cc(3 == LOG_FAC(LOG_DAEMON));
1177 iovec[n++] = IOVEC_MAKE_STRING("SYSLOG_FACILITY=3");
1178 iovec[n++] = IOVEC_MAKE_STRING("SYSLOG_IDENTIFIER=systemd-journald");
1179
1180 iovec[n++] = IOVEC_MAKE_STRING("_TRANSPORT=driver");
1181 assert_cc(6 == LOG_INFO);
1182 iovec[n++] = IOVEC_MAKE_STRING("PRIORITY=6");
1183
1184 if (message_id)
1185 iovec[n++] = IOVEC_MAKE_STRING(message_id);
1186 k = n;
1187
1188 va_start(ap, format);
1189 r = log_format_iovec(iovec, m, &n, false, 0, format, ap);
1190 /* Error handling below */
1191 va_end(ap);
1192
1193 if (r >= 0)
1194 server_dispatch_message_real(s, iovec, n, m, s->my_context, /* tv= */ NULL, LOG_INFO, object_pid);
1195
1196 while (k < n)
1197 free(iovec[k++].iov_base);
1198
1199 if (r < 0) {
1200 /* We failed to format the message. Emit a warning instead. */
1201 char buf[LINE_MAX];
1202
1203 errno = -r;
1204 xsprintf(buf, "MESSAGE=Entry printing failed: %m");
1205
1206 n = 3;
1207 iovec[n++] = IOVEC_MAKE_STRING("PRIORITY=4");
1208 iovec[n++] = IOVEC_MAKE_STRING(buf);
1209 server_dispatch_message_real(s, iovec, n, m, s->my_context, /* tv= */ NULL, LOG_INFO, object_pid);
1210 }
1211 }
1212
1213 void server_dispatch_message(
1214 Server *s,
1215 struct iovec *iovec, size_t n, size_t m,
1216 ClientContext *c,
1217 const struct timeval *tv,
1218 int priority,
1219 pid_t object_pid) {
1220
1221 uint64_t available = 0;
1222 int rl;
1223
1224 assert(s);
1225 assert(iovec || n == 0);
1226
1227 if (n == 0)
1228 return;
1229
1230 if (LOG_PRI(priority) > s->max_level_store)
1231 return;
1232
1233 /* Stop early in case the information will not be stored
1234 * in a journal. */
1235 if (s->storage == STORAGE_NONE)
1236 return;
1237
1238 if (c && c->unit) {
1239 (void) server_determine_space(s, &available, /* limit= */ NULL);
1240
1241 rl = journal_ratelimit_test(s->ratelimit, c->unit, c->log_ratelimit_interval, c->log_ratelimit_burst, priority & LOG_PRIMASK, available);
1242 if (rl == 0)
1243 return;
1244
1245 /* Write a suppression message if we suppressed something */
1246 if (rl > 1)
1247 server_driver_message(s, c->pid,
1248 "MESSAGE_ID=" SD_MESSAGE_JOURNAL_DROPPED_STR,
1249 LOG_MESSAGE("Suppressed %i messages from %s", rl - 1, c->unit),
1250 "N_DROPPED=%i", rl - 1,
1251 NULL);
1252 }
1253
1254 server_dispatch_message_real(s, iovec, n, m, c, tv, priority, object_pid);
1255 }
1256
1257 int server_flush_to_var(Server *s, bool require_flag_file) {
1258 sd_journal *j = NULL;
1259 const char *fn;
1260 unsigned n = 0;
1261 usec_t start;
1262 int r, k;
1263
1264 assert(s);
1265
1266 if (!IN_SET(s->storage, STORAGE_AUTO, STORAGE_PERSISTENT))
1267 return 0;
1268
1269 if (s->namespace) /* Flushing concept does not exist for namespace instances */
1270 return 0;
1271
1272 if (!s->runtime_journal) /* Nothing to flush? */
1273 return 0;
1274
1275 if (require_flag_file && !server_flushed_flag_is_set(s))
1276 return 0;
1277
1278 (void) server_system_journal_open(s, /* flush_requested=*/ true, /* relinquish_requested= */ false);
1279
1280 if (!s->system_journal)
1281 return 0;
1282
1283 /* Reset current seqnum data to avoid unnecessary rotation when switching to system journal.
1284 * See issue #30092. */
1285 zero(*s->seqnum);
1286
1287 log_debug("Flushing to %s...", s->system_storage.path);
1288
1289 start = now(CLOCK_MONOTONIC);
1290
1291 r = sd_journal_open(&j, SD_JOURNAL_RUNTIME_ONLY);
1292 if (r < 0)
1293 return log_ratelimit_error_errno(r, JOURNAL_LOG_RATELIMIT,
1294 "Failed to read runtime journal: %m");
1295
1296 sd_journal_set_data_threshold(j, 0);
1297
1298 SD_JOURNAL_FOREACH(j) {
1299 Object *o = NULL;
1300 JournalFile *f;
1301
1302 f = j->current_file;
1303 assert(f && f->current_offset > 0);
1304
1305 n++;
1306
1307 r = journal_file_move_to_object(f, OBJECT_ENTRY, f->current_offset, &o);
1308 if (r < 0) {
1309 log_ratelimit_error_errno(r, JOURNAL_LOG_RATELIMIT, "Can't read entry: %m");
1310 goto finish;
1311 }
1312
1313 r = journal_file_copy_entry(
1314 f,
1315 s->system_journal,
1316 o,
1317 f->current_offset,
1318 &s->seqnum->seqnum,
1319 &s->seqnum->id);
1320 if (r >= 0)
1321 continue;
1322
1323 if (!shall_try_append_again(s->system_journal, r)) {
1324 log_ratelimit_error_errno(r, JOURNAL_LOG_RATELIMIT, "Can't write entry: %m");
1325 goto finish;
1326 }
1327
1328 log_ratelimit_info(JOURNAL_LOG_RATELIMIT, "Rotating system journal.");
1329
1330 server_rotate(s);
1331 server_vacuum(s, false);
1332
1333 if (!s->system_journal) {
1334 log_ratelimit_notice(JOURNAL_LOG_RATELIMIT,
1335 "Didn't flush runtime journal since rotation of system journal wasn't successful.");
1336 r = -EIO;
1337 goto finish;
1338 }
1339
1340 log_debug("Retrying write.");
1341 r = journal_file_copy_entry(
1342 f,
1343 s->system_journal,
1344 o,
1345 f->current_offset,
1346 &s->seqnum->seqnum,
1347 &s->seqnum->id);
1348 if (r < 0) {
1349 log_ratelimit_error_errno(r, JOURNAL_LOG_RATELIMIT, "Can't write entry: %m");
1350 goto finish;
1351 }
1352 }
1353
1354 r = 0;
1355
1356 finish:
1357 if (s->system_journal)
1358 journal_file_post_change(s->system_journal);
1359
1360 /* Save parent directories of runtime journals before closing runtime journals. */
1361 _cleanup_strv_free_ char **dirs = NULL;
1362 (void) journal_get_directories(j, &dirs);
1363
1364 /* First, close all runtime journals opened in the above. */
1365 sd_journal_close(j);
1366
1367 /* Offline and close the 'main' runtime journal file. */
1368 s->runtime_journal = journal_file_offline_close(s->runtime_journal);
1369
1370 /* Remove the runtime directory if the all entries are successfully flushed to /var/. */
1371 if (r >= 0) {
1372 (void) rm_rf(s->runtime_storage.path, REMOVE_ROOT);
1373
1374 /* The initrd may have a different machine ID from the host's one. Typically, that happens
1375 * when our tests running on qemu, as the host's initrd is picked as is without updating
1376 * the machine ID in the initrd with the one used in the image. Even in such the case, the
1377 * runtime journals in the subdirectory named with the initrd's machine ID are flushed to
1378 * the persistent journal. To make not the runtime journal flushed multiple times, let's
1379 * also remove the runtime directories. */
1380 STRV_FOREACH(p, dirs)
1381 (void) rm_rf(*p, REMOVE_ROOT);
1382 }
1383
1384 server_driver_message(s, 0, NULL,
1385 LOG_MESSAGE("Time spent on flushing to %s is %s for %u entries.",
1386 s->system_storage.path,
1387 FORMAT_TIMESPAN(usec_sub_unsigned(now(CLOCK_MONOTONIC), start), 0),
1388 n),
1389 NULL);
1390
1391 fn = strjoina(s->runtime_directory, "/flushed");
1392 k = touch(fn);
1393 if (k < 0)
1394 log_ratelimit_warning_errno(k, JOURNAL_LOG_RATELIMIT,
1395 "Failed to touch %s, ignoring: %m", fn);
1396
1397 server_refresh_idle_timer(s);
1398 return r;
1399 }
1400
1401 static int server_relinquish_var(Server *s) {
1402 const char *fn;
1403 assert(s);
1404
1405 if (s->storage == STORAGE_NONE)
1406 return 0;
1407
1408 if (s->namespace) /* Concept does not exist for namespaced instances */
1409 return -EOPNOTSUPP;
1410
1411 if (s->runtime_journal && !s->system_journal)
1412 return 0;
1413
1414 log_debug("Relinquishing %s...", s->system_storage.path);
1415
1416 (void) server_system_journal_open(s, /* flush_requested */ false, /* relinquish_requested=*/ true);
1417
1418 s->system_journal = journal_file_offline_close(s->system_journal);
1419 ordered_hashmap_clear_with_destructor(s->user_journals, journal_file_offline_close);
1420 set_clear_with_destructor(s->deferred_closes, journal_file_offline_close);
1421
1422 fn = strjoina(s->runtime_directory, "/flushed");
1423 if (unlink(fn) < 0 && errno != ENOENT)
1424 log_ratelimit_warning_errno(errno, JOURNAL_LOG_RATELIMIT,
1425 "Failed to unlink %s, ignoring: %m", fn);
1426
1427 server_refresh_idle_timer(s);
1428 return 0;
1429 }
1430
1431 int server_process_datagram(
1432 sd_event_source *es,
1433 int fd,
1434 uint32_t revents,
1435 void *userdata) {
1436
1437 size_t label_len = 0, m;
1438 Server *s = ASSERT_PTR(userdata);
1439 struct ucred *ucred = NULL;
1440 struct timeval tv_buf, *tv = NULL;
1441 struct cmsghdr *cmsg;
1442 char *label = NULL;
1443 struct iovec iovec;
1444 ssize_t n;
1445 int *fds = NULL, v = 0;
1446 size_t n_fds = 0;
1447
1448 /* We use NAME_MAX space for the SELinux label here. The kernel currently enforces no limit, but
1449 * according to suggestions from the SELinux people this will change and it will probably be
1450 * identical to NAME_MAX. For now we use that, but this should be updated one day when the final
1451 * limit is known.
1452 *
1453 * Here, we need to explicitly initialize the buffer with zero, as glibc has a bug in
1454 * __convert_scm_timestamps(), which assumes the buffer is initialized. See #20741. */
1455 CMSG_BUFFER_TYPE(CMSG_SPACE(sizeof(struct ucred)) +
1456 CMSG_SPACE_TIMEVAL +
1457 CMSG_SPACE(sizeof(int)) + /* fd */
1458 CMSG_SPACE(NAME_MAX) /* selinux label */) control = {};
1459
1460 union sockaddr_union sa = {};
1461
1462 struct msghdr msghdr = {
1463 .msg_iov = &iovec,
1464 .msg_iovlen = 1,
1465 .msg_control = &control,
1466 .msg_controllen = sizeof(control),
1467 .msg_name = &sa,
1468 .msg_namelen = sizeof(sa),
1469 };
1470
1471 assert(fd == s->native_fd || fd == s->syslog_fd || fd == s->audit_fd);
1472
1473 if (revents != EPOLLIN)
1474 return log_error_errno(SYNTHETIC_ERRNO(EIO),
1475 "Got invalid event from epoll for datagram fd: %" PRIx32,
1476 revents);
1477
1478 /* Try to get the right size, if we can. (Not all sockets support SIOCINQ, hence we just try, but don't rely on
1479 * it.) */
1480 (void) ioctl(fd, SIOCINQ, &v);
1481
1482 /* Fix it up, if it is too small. We use the same fixed value as auditd here. Awful! */
1483 m = PAGE_ALIGN(MAX3((size_t) v + 1,
1484 (size_t) LINE_MAX,
1485 ALIGN(sizeof(struct nlmsghdr)) + ALIGN((size_t) MAX_AUDIT_MESSAGE_LENGTH)) + 1);
1486
1487 if (!GREEDY_REALLOC(s->buffer, m))
1488 return log_oom();
1489
1490 iovec = IOVEC_MAKE(s->buffer, MALLOC_ELEMENTSOF(s->buffer) - 1); /* Leave room for trailing NUL we add later */
1491
1492 n = recvmsg_safe(fd, &msghdr, MSG_DONTWAIT|MSG_CMSG_CLOEXEC);
1493 if (n < 0) {
1494 if (ERRNO_IS_TRANSIENT(n))
1495 return 0;
1496 if (n == -EXFULL) {
1497 log_ratelimit_warning(JOURNAL_LOG_RATELIMIT,
1498 "Got message with truncated control data (too many fds sent?), ignoring.");
1499 return 0;
1500 }
1501 return log_ratelimit_error_errno(n, JOURNAL_LOG_RATELIMIT, "recvmsg() failed: %m");
1502 }
1503
1504 CMSG_FOREACH(cmsg, &msghdr)
1505 if (cmsg->cmsg_level == SOL_SOCKET &&
1506 cmsg->cmsg_type == SCM_CREDENTIALS &&
1507 cmsg->cmsg_len == CMSG_LEN(sizeof(struct ucred))) {
1508 assert(!ucred);
1509 ucred = CMSG_TYPED_DATA(cmsg, struct ucred);
1510 } else if (cmsg->cmsg_level == SOL_SOCKET &&
1511 cmsg->cmsg_type == SCM_SECURITY) {
1512 assert(!label);
1513 label = CMSG_TYPED_DATA(cmsg, char);
1514 label_len = cmsg->cmsg_len - CMSG_LEN(0);
1515 } else if (cmsg->cmsg_level == SOL_SOCKET &&
1516 cmsg->cmsg_type == SCM_TIMESTAMP &&
1517 cmsg->cmsg_len == CMSG_LEN(sizeof(struct timeval))) {
1518 assert(!tv);
1519 tv = memcpy(&tv_buf, CMSG_DATA(cmsg), sizeof(struct timeval));
1520 } else if (cmsg->cmsg_level == SOL_SOCKET &&
1521 cmsg->cmsg_type == SCM_RIGHTS) {
1522 assert(!fds);
1523 fds = CMSG_TYPED_DATA(cmsg, int);
1524 n_fds = (cmsg->cmsg_len - CMSG_LEN(0)) / sizeof(int);
1525 }
1526
1527 /* And a trailing NUL, just in case */
1528 s->buffer[n] = 0;
1529
1530 if (fd == s->syslog_fd) {
1531 if (n > 0 && n_fds == 0)
1532 server_process_syslog_message(s, s->buffer, n, ucred, tv, label, label_len);
1533 else if (n_fds > 0)
1534 log_ratelimit_warning(JOURNAL_LOG_RATELIMIT,
1535 "Got file descriptors via syslog socket. Ignoring.");
1536
1537 } else if (fd == s->native_fd) {
1538 if (n > 0 && n_fds == 0)
1539 server_process_native_message(s, s->buffer, n, ucred, tv, label, label_len);
1540 else if (n == 0 && n_fds == 1)
1541 (void) server_process_native_file(s, fds[0], ucred, tv, label, label_len);
1542 else if (n_fds > 0)
1543 log_ratelimit_warning(JOURNAL_LOG_RATELIMIT,
1544 "Got too many file descriptors via native socket. Ignoring.");
1545
1546 } else {
1547 assert(fd == s->audit_fd);
1548
1549 if (n > 0 && n_fds == 0)
1550 server_process_audit_message(s, s->buffer, n, ucred, &sa, msghdr.msg_namelen);
1551 else if (n_fds > 0)
1552 log_ratelimit_warning(JOURNAL_LOG_RATELIMIT,
1553 "Got file descriptors via audit socket. Ignoring.");
1554 }
1555
1556 close_many(fds, n_fds);
1557
1558 server_refresh_idle_timer(s);
1559 return 0;
1560 }
1561
1562 static void server_full_flush(Server *s) {
1563 assert(s);
1564
1565 (void) server_flush_to_var(s, false);
1566 server_sync(s);
1567 server_vacuum(s, false);
1568
1569 server_space_usage_message(s, NULL);
1570
1571 server_refresh_idle_timer(s);
1572 }
1573
1574 static int dispatch_sigusr1(sd_event_source *es, const struct signalfd_siginfo *si, void *userdata) {
1575 Server *s = ASSERT_PTR(userdata);
1576
1577 if (s->namespace) {
1578 log_error("Received SIGUSR1 signal from PID %u, but flushing runtime journals not supported for namespaced instances.", si->ssi_pid);
1579 return 0;
1580 }
1581
1582 log_info("Received SIGUSR1 signal from PID %u, as request to flush runtime journal.", si->ssi_pid);
1583 server_full_flush(s);
1584
1585 return 0;
1586 }
1587
1588 static void server_full_rotate(Server *s) {
1589 const char *fn;
1590 int r;
1591
1592 assert(s);
1593
1594 server_rotate(s);
1595 server_vacuum(s, true);
1596
1597 if (s->system_journal)
1598 patch_min_use(&s->system_storage);
1599 if (s->runtime_journal)
1600 patch_min_use(&s->runtime_storage);
1601
1602 /* Let clients know when the most recent rotation happened. */
1603 fn = strjoina(s->runtime_directory, "/rotated");
1604 r = write_timestamp_file_atomic(fn, now(CLOCK_MONOTONIC));
1605 if (r < 0)
1606 log_ratelimit_warning_errno(r, JOURNAL_LOG_RATELIMIT,
1607 "Failed to write %s, ignoring: %m", fn);
1608 }
1609
1610 static int dispatch_sigusr2(sd_event_source *es, const struct signalfd_siginfo *si, void *userdata) {
1611 Server *s = ASSERT_PTR(userdata);
1612
1613 log_info("Received SIGUSR2 signal from PID %u, as request to rotate journal, rotating.", si->ssi_pid);
1614 server_full_rotate(s);
1615
1616 return 0;
1617 }
1618
1619 static int dispatch_sigterm(sd_event_source *es, const struct signalfd_siginfo *si, void *userdata) {
1620 _cleanup_(sd_event_source_disable_unrefp) sd_event_source *news = NULL;
1621 Server *s = ASSERT_PTR(userdata);
1622 int r;
1623
1624 log_received_signal(LOG_INFO, si);
1625
1626 (void) sd_event_source_set_enabled(es, SD_EVENT_OFF); /* Make sure this handler is called at most once */
1627
1628 /* So on one hand we want to ensure that SIGTERMs are definitely handled in appropriate, bounded
1629 * time. On the other hand we want that everything pending is first comprehensively processed and
1630 * written to disk. These goals are incompatible, hence we try to find a middle ground: we'll process
1631 * SIGTERM with high priority, but from the handler (this one right here) we'll install two new event
1632 * sources: one low priority idle one that will issue the exit once everything else is processed (and
1633 * which is hopefully the regular, clean codepath); and one high priority timer that acts as safety
1634 * net: if our idle handler isn't run within 10s, we'll exit anyway.
1635 *
1636 * TLDR: we'll exit either when everything is processed, or after 10s max, depending on what happens
1637 * first.
1638 *
1639 * Note that exiting before the idle event is hit doesn't typically mean that we lose any data, as
1640 * messages will remain queued in the sockets they came in from, and thus can be processed when we
1641 * start up next – unless we are going down for the final system shutdown, in which case everything
1642 * is lost. */
1643
1644 r = sd_event_add_defer(s->event, &news, NULL, NULL); /* NULL handler means → exit when triggered */
1645 if (r < 0) {
1646 log_error_errno(r, "Failed to allocate exit idle event handler: %m");
1647 goto fail;
1648 }
1649
1650 (void) sd_event_source_set_description(news, "exit-idle");
1651
1652 /* Run everything relevant before this. */
1653 r = sd_event_source_set_priority(news, SD_EVENT_PRIORITY_NORMAL+20);
1654 if (r < 0) {
1655 log_error_errno(r, "Failed to adjust priority of exit idle event handler: %m");
1656 goto fail;
1657 }
1658
1659 /* Give up ownership, so that this event source is freed automatically when the event loop is freed. */
1660 r = sd_event_source_set_floating(news, true);
1661 if (r < 0) {
1662 log_error_errno(r, "Failed to make exit idle event handler floating: %m");
1663 goto fail;
1664 }
1665
1666 news = sd_event_source_unref(news);
1667
1668 r = sd_event_add_time_relative(s->event, &news, CLOCK_MONOTONIC, 10 * USEC_PER_SEC, 0, NULL, NULL);
1669 if (r < 0) {
1670 log_error_errno(r, "Failed to allocate exit timeout event handler: %m");
1671 goto fail;
1672 }
1673
1674 (void) sd_event_source_set_description(news, "exit-timeout");
1675
1676 r = sd_event_source_set_priority(news, SD_EVENT_PRIORITY_IMPORTANT-20); /* This is a safety net, with highest priority */
1677 if (r < 0) {
1678 log_error_errno(r, "Failed to adjust priority of exit timeout event handler: %m");
1679 goto fail;
1680 }
1681
1682 r = sd_event_source_set_floating(news, true);
1683 if (r < 0) {
1684 log_error_errno(r, "Failed to make exit timeout event handler floating: %m");
1685 goto fail;
1686 }
1687
1688 news = sd_event_source_unref(news);
1689
1690 log_debug("Exit event sources are now pending.");
1691 return 0;
1692
1693 fail:
1694 sd_event_exit(s->event, 0);
1695 return 0;
1696 }
1697
1698 static void server_full_sync(Server *s) {
1699 const char *fn;
1700 int r;
1701
1702 assert(s);
1703
1704 server_sync(s);
1705
1706 /* Let clients know when the most recent sync happened. */
1707 fn = strjoina(s->runtime_directory, "/synced");
1708 r = write_timestamp_file_atomic(fn, now(CLOCK_MONOTONIC));
1709 if (r < 0)
1710 log_ratelimit_warning_errno(r, JOURNAL_LOG_RATELIMIT,
1711 "Failed to write %s, ignoring: %m", fn);
1712
1713 return;
1714 }
1715
1716 static int dispatch_sigrtmin1(sd_event_source *es, const struct signalfd_siginfo *si, void *userdata) {
1717 Server *s = ASSERT_PTR(userdata);
1718
1719 log_debug("Received SIGRTMIN1 signal from PID %u, as request to sync.", si->ssi_pid);
1720 server_full_sync(s);
1721
1722 return 0;
1723 }
1724
1725 static int server_setup_signals(Server *s) {
1726 int r;
1727
1728 assert(s);
1729
1730 assert_se(sigprocmask_many(SIG_SETMASK, NULL, SIGINT, SIGTERM, SIGUSR1, SIGUSR2, SIGRTMIN+1, SIGRTMIN+18) >= 0);
1731
1732 r = sd_event_add_signal(s->event, &s->sigusr1_event_source, SIGUSR1, dispatch_sigusr1, s);
1733 if (r < 0)
1734 return r;
1735
1736 r = sd_event_add_signal(s->event, &s->sigusr2_event_source, SIGUSR2, dispatch_sigusr2, s);
1737 if (r < 0)
1738 return r;
1739
1740 r = sd_event_add_signal(s->event, &s->sigterm_event_source, SIGTERM, dispatch_sigterm, s);
1741 if (r < 0)
1742 return r;
1743
1744 /* Let's process SIGTERM early, so that we definitely react to it */
1745 r = sd_event_source_set_priority(s->sigterm_event_source, SD_EVENT_PRIORITY_IMPORTANT-10);
1746 if (r < 0)
1747 return r;
1748
1749 /* When journald is invoked on the terminal (when debugging), it's useful if C-c is handled
1750 * equivalent to SIGTERM. */
1751 r = sd_event_add_signal(s->event, &s->sigint_event_source, SIGINT, dispatch_sigterm, s);
1752 if (r < 0)
1753 return r;
1754
1755 r = sd_event_source_set_priority(s->sigint_event_source, SD_EVENT_PRIORITY_IMPORTANT-10);
1756 if (r < 0)
1757 return r;
1758
1759 /* SIGRTMIN+1 causes an immediate sync. We process this very late, so that everything else queued at
1760 * this point is really written to disk. Clients can watch /run/systemd/journal/synced with inotify
1761 * until its mtime changes to see when a sync happened. */
1762 r = sd_event_add_signal(s->event, &s->sigrtmin1_event_source, SIGRTMIN+1, dispatch_sigrtmin1, s);
1763 if (r < 0)
1764 return r;
1765
1766 r = sd_event_source_set_priority(s->sigrtmin1_event_source, SD_EVENT_PRIORITY_NORMAL+15);
1767 if (r < 0)
1768 return r;
1769
1770 r = sd_event_add_signal(s->event, NULL, SIGRTMIN+18, sigrtmin18_handler, &s->sigrtmin18_info);
1771 if (r < 0)
1772 return r;
1773
1774 return 0;
1775 }
1776
1777 static int parse_proc_cmdline_item(const char *key, const char *value, void *data) {
1778 Server *s = ASSERT_PTR(data);
1779 int r;
1780
1781 if (proc_cmdline_key_streq(key, "systemd.journald.forward_to_syslog")) {
1782
1783 r = value ? parse_boolean(value) : true;
1784 if (r < 0)
1785 log_warning("Failed to parse forward to syslog switch \"%s\". Ignoring.", value);
1786 else
1787 s->forward_to_syslog = r;
1788
1789 } else if (proc_cmdline_key_streq(key, "systemd.journald.forward_to_kmsg")) {
1790
1791 r = value ? parse_boolean(value) : true;
1792 if (r < 0)
1793 log_warning("Failed to parse forward to kmsg switch \"%s\". Ignoring.", value);
1794 else
1795 s->forward_to_kmsg = r;
1796
1797 } else if (proc_cmdline_key_streq(key, "systemd.journald.forward_to_console")) {
1798
1799 r = value ? parse_boolean(value) : true;
1800 if (r < 0)
1801 log_warning("Failed to parse forward to console switch \"%s\". Ignoring.", value);
1802 else
1803 s->forward_to_console = r;
1804
1805 } else if (proc_cmdline_key_streq(key, "systemd.journald.forward_to_wall")) {
1806
1807 r = value ? parse_boolean(value) : true;
1808 if (r < 0)
1809 log_warning("Failed to parse forward to wall switch \"%s\". Ignoring.", value);
1810 else
1811 s->forward_to_wall = r;
1812
1813 } else if (proc_cmdline_key_streq(key, "systemd.journald.max_level_console")) {
1814
1815 if (proc_cmdline_value_missing(key, value))
1816 return 0;
1817
1818 r = log_level_from_string(value);
1819 if (r < 0)
1820 log_warning("Failed to parse max level console value \"%s\". Ignoring.", value);
1821 else
1822 s->max_level_console = r;
1823
1824 } else if (proc_cmdline_key_streq(key, "systemd.journald.max_level_store")) {
1825
1826 if (proc_cmdline_value_missing(key, value))
1827 return 0;
1828
1829 r = log_level_from_string(value);
1830 if (r < 0)
1831 log_warning("Failed to parse max level store value \"%s\". Ignoring.", value);
1832 else
1833 s->max_level_store = r;
1834
1835 } else if (proc_cmdline_key_streq(key, "systemd.journald.max_level_syslog")) {
1836
1837 if (proc_cmdline_value_missing(key, value))
1838 return 0;
1839
1840 r = log_level_from_string(value);
1841 if (r < 0)
1842 log_warning("Failed to parse max level syslog value \"%s\". Ignoring.", value);
1843 else
1844 s->max_level_syslog = r;
1845
1846 } else if (proc_cmdline_key_streq(key, "systemd.journald.max_level_kmsg")) {
1847
1848 if (proc_cmdline_value_missing(key, value))
1849 return 0;
1850
1851 r = log_level_from_string(value);
1852 if (r < 0)
1853 log_warning("Failed to parse max level kmsg value \"%s\". Ignoring.", value);
1854 else
1855 s->max_level_kmsg = r;
1856
1857 } else if (proc_cmdline_key_streq(key, "systemd.journald.max_level_wall")) {
1858
1859 if (proc_cmdline_value_missing(key, value))
1860 return 0;
1861
1862 r = log_level_from_string(value);
1863 if (r < 0)
1864 log_warning("Failed to parse max level wall value \"%s\". Ignoring.", value);
1865 else
1866 s->max_level_wall = r;
1867
1868 } else if (proc_cmdline_key_streq(key, "systemd.journald.max_level_socket")) {
1869
1870 if (proc_cmdline_value_missing(key, value))
1871 return 0;
1872
1873 r = log_level_from_string(value);
1874 if (r < 0)
1875 log_warning("Failed to parse max level socket value \"%s\". Ignoring.", value);
1876 else
1877 s->max_level_socket = r;
1878
1879 } else if (startswith(key, "systemd.journald"))
1880 log_warning("Unknown journald kernel command line option \"%s\". Ignoring.", key);
1881
1882 /* do not warn about state here, since probably systemd already did */
1883 return 0;
1884 }
1885
1886 static int server_parse_config_file(Server *s) {
1887 const char *conf_file;
1888
1889 assert(s);
1890
1891 if (s->namespace)
1892 conf_file = strjoina("systemd/journald@", s->namespace, ".conf");
1893 else
1894 conf_file = "systemd/journald.conf";
1895
1896 return config_parse_standard_file_with_dropins(
1897 conf_file,
1898 "Journal\0",
1899 config_item_perf_lookup, journald_gperf_lookup,
1900 CONFIG_PARSE_WARN,
1901 /* userdata= */ s);
1902 }
1903
1904 static int server_dispatch_sync(sd_event_source *es, usec_t t, void *userdata) {
1905 Server *s = ASSERT_PTR(userdata);
1906
1907 server_sync(s);
1908 return 0;
1909 }
1910
1911 int server_schedule_sync(Server *s, int priority) {
1912 int r;
1913
1914 assert(s);
1915
1916 if (priority <= LOG_CRIT) {
1917 /* Immediately sync to disk when this is of priority CRIT, ALERT, EMERG */
1918 server_sync(s);
1919 return 0;
1920 }
1921
1922 if (!s->event || sd_event_get_state(s->event) == SD_EVENT_FINISHED) {
1923 /* Shutting down the server? Let's sync immediately. */
1924 server_sync(s);
1925 return 0;
1926 }
1927
1928 if (s->sync_scheduled)
1929 return 0;
1930
1931 if (s->sync_interval_usec > 0) {
1932
1933 if (!s->sync_event_source) {
1934 r = sd_event_add_time_relative(
1935 s->event,
1936 &s->sync_event_source,
1937 CLOCK_MONOTONIC,
1938 s->sync_interval_usec, 0,
1939 server_dispatch_sync, s);
1940 if (r < 0)
1941 return r;
1942
1943 r = sd_event_source_set_priority(s->sync_event_source, SD_EVENT_PRIORITY_IMPORTANT);
1944 } else {
1945 r = sd_event_source_set_time_relative(s->sync_event_source, s->sync_interval_usec);
1946 if (r < 0)
1947 return r;
1948
1949 r = sd_event_source_set_enabled(s->sync_event_source, SD_EVENT_ONESHOT);
1950 }
1951 if (r < 0)
1952 return r;
1953
1954 s->sync_scheduled = true;
1955 }
1956
1957 return 0;
1958 }
1959
1960 static int dispatch_hostname_change(sd_event_source *es, int fd, uint32_t revents, void *userdata) {
1961 Server *s = ASSERT_PTR(userdata);
1962
1963 server_cache_hostname(s);
1964 return 0;
1965 }
1966
1967 static int server_open_hostname(Server *s) {
1968 int r;
1969
1970 assert(s);
1971
1972 s->hostname_fd = open("/proc/sys/kernel/hostname",
1973 O_RDONLY|O_CLOEXEC|O_NONBLOCK|O_NOCTTY);
1974 if (s->hostname_fd < 0)
1975 return log_error_errno(errno, "Failed to open /proc/sys/kernel/hostname: %m");
1976
1977 r = sd_event_add_io(s->event, &s->hostname_event_source, s->hostname_fd, 0, dispatch_hostname_change, s);
1978 if (r < 0) {
1979 /* kernels prior to 3.2 don't support polling this file. Ignore
1980 * the failure. */
1981 if (r == -EPERM) {
1982 log_warning_errno(r, "Failed to register hostname fd in event loop, ignoring: %m");
1983 s->hostname_fd = safe_close(s->hostname_fd);
1984 return 0;
1985 }
1986
1987 return log_error_errno(r, "Failed to register hostname fd in event loop: %m");
1988 }
1989
1990 r = sd_event_source_set_priority(s->hostname_event_source, SD_EVENT_PRIORITY_IMPORTANT-10);
1991 if (r < 0)
1992 return log_error_errno(r, "Failed to adjust priority of hostname event source: %m");
1993
1994 return 0;
1995 }
1996
1997 static int dispatch_notify_event(sd_event_source *es, int fd, uint32_t revents, void *userdata) {
1998 Server *s = ASSERT_PTR(userdata);
1999 int r;
2000
2001 assert(s->notify_event_source == es);
2002 assert(s->notify_fd == fd);
2003
2004 /* The $NOTIFY_SOCKET is writable again, now send exactly one
2005 * message on it. Either it's the watchdog event, the initial
2006 * READY=1 event or an stdout stream event. If there's nothing
2007 * to write anymore, turn our event source off. The next time
2008 * there's something to send it will be turned on again. */
2009
2010 if (!s->sent_notify_ready) {
2011 static const char p[] = "READY=1\n"
2012 "STATUS=Processing requests...";
2013
2014 if (send(s->notify_fd, p, strlen(p), MSG_DONTWAIT) < 0) {
2015 if (errno == EAGAIN)
2016 return 0;
2017
2018 return log_error_errno(errno, "Failed to send READY=1 notification message: %m");
2019 }
2020
2021 s->sent_notify_ready = true;
2022 log_debug("Sent READY=1 notification.");
2023
2024 } else if (s->send_watchdog) {
2025 static const char p[] = "WATCHDOG=1";
2026
2027 if (send(s->notify_fd, p, strlen(p), MSG_DONTWAIT) < 0) {
2028 if (errno == EAGAIN)
2029 return 0;
2030
2031 return log_error_errno(errno, "Failed to send WATCHDOG=1 notification message: %m");
2032 }
2033
2034 s->send_watchdog = false;
2035 log_debug("Sent WATCHDOG=1 notification.");
2036
2037 } else if (s->stdout_streams_notify_queue)
2038 /* Dispatch one stream notification event */
2039 stdout_stream_send_notify(s->stdout_streams_notify_queue);
2040
2041 /* Leave us enabled if there's still more to do. */
2042 if (s->send_watchdog || s->stdout_streams_notify_queue)
2043 return 0;
2044
2045 /* There was nothing to do anymore, let's turn ourselves off. */
2046 r = sd_event_source_set_enabled(es, SD_EVENT_OFF);
2047 if (r < 0)
2048 return log_error_errno(r, "Failed to turn off notify event source: %m");
2049
2050 return 0;
2051 }
2052
2053 static int dispatch_watchdog(sd_event_source *es, uint64_t usec, void *userdata) {
2054 Server *s = ASSERT_PTR(userdata);
2055 int r;
2056
2057 s->send_watchdog = true;
2058
2059 r = sd_event_source_set_enabled(s->notify_event_source, SD_EVENT_ON);
2060 if (r < 0)
2061 log_warning_errno(r, "Failed to turn on notify event source: %m");
2062
2063 r = sd_event_source_set_time(s->watchdog_event_source, usec + s->watchdog_usec / 2);
2064 if (r < 0)
2065 return log_error_errno(r, "Failed to restart watchdog event source: %m");
2066
2067 r = sd_event_source_set_enabled(s->watchdog_event_source, SD_EVENT_ON);
2068 if (r < 0)
2069 return log_error_errno(r, "Failed to enable watchdog event source: %m");
2070
2071 return 0;
2072 }
2073
2074 static int server_connect_notify(Server *s) {
2075 union sockaddr_union sa;
2076 socklen_t sa_len;
2077 const char *e;
2078 int r;
2079
2080 assert(s);
2081 assert(s->notify_fd < 0);
2082 assert(!s->notify_event_source);
2083
2084 /*
2085 * So here's the problem: we'd like to send notification messages to PID 1, but we cannot do that via
2086 * sd_notify(), since that's synchronous, and we might end up blocking on it. Specifically: given
2087 * that PID 1 might block on dbus-daemon during IPC, and dbus-daemon is logging to us, and might
2088 * hence block on us, we might end up in a deadlock if we block on sending PID 1 notification
2089 * messages — by generating a full blocking circle. To avoid this, let's create a non-blocking
2090 * socket, and connect it to the notification socket, and then wait for POLLOUT before we send
2091 * anything. This should efficiently avoid any deadlocks, as we'll never block on PID 1, hence PID 1
2092 * can safely block on dbus-daemon which can safely block on us again.
2093 *
2094 * Don't think that this issue is real? It is, see: https://github.com/systemd/systemd/issues/1505
2095 */
2096
2097 e = getenv("NOTIFY_SOCKET");
2098 if (!e)
2099 return 0;
2100
2101 r = sockaddr_un_set_path(&sa.un, e);
2102 if (r < 0)
2103 return log_error_errno(r, "NOTIFY_SOCKET set to invalid value '%s': %m", e);
2104 sa_len = r;
2105
2106 s->notify_fd = socket(AF_UNIX, SOCK_DGRAM|SOCK_CLOEXEC|SOCK_NONBLOCK, 0);
2107 if (s->notify_fd < 0)
2108 return log_error_errno(errno, "Failed to create notify socket: %m");
2109
2110 (void) fd_inc_sndbuf(s->notify_fd, NOTIFY_SNDBUF_SIZE);
2111
2112 r = connect(s->notify_fd, &sa.sa, sa_len);
2113 if (r < 0)
2114 return log_error_errno(errno, "Failed to connect to notify socket: %m");
2115
2116 r = sd_event_add_io(s->event, &s->notify_event_source, s->notify_fd, EPOLLOUT, dispatch_notify_event, s);
2117 if (r < 0)
2118 return log_error_errno(r, "Failed to watch notification socket: %m");
2119
2120 if (sd_watchdog_enabled(false, &s->watchdog_usec) > 0) {
2121 s->send_watchdog = true;
2122
2123 r = sd_event_add_time_relative(s->event, &s->watchdog_event_source, CLOCK_MONOTONIC, s->watchdog_usec/2, s->watchdog_usec/4, dispatch_watchdog, s);
2124 if (r < 0)
2125 return log_error_errno(r, "Failed to add watchdog time event: %m");
2126 }
2127
2128 /* This should fire pretty soon, which we'll use to send the READY=1 event. */
2129
2130 return 0;
2131 }
2132
2133 static int synchronize_second_half(sd_event_source *event_source, void *userdata) {
2134 Varlink *link = ASSERT_PTR(userdata);
2135 Server *s;
2136 int r;
2137
2138 assert_se(s = varlink_get_userdata(link));
2139
2140 /* This is the "second half" of the Synchronize() varlink method. This function is called as deferred
2141 * event source at a low priority to ensure the synchronization completes after all queued log
2142 * messages are processed. */
2143 server_full_sync(s);
2144
2145 /* Let's get rid of the event source now, by marking it as non-floating again. It then has no ref
2146 * anymore and is immediately destroyed after we return from this function, i.e. from this event
2147 * source handler at the end. */
2148 r = sd_event_source_set_floating(event_source, false);
2149 if (r < 0)
2150 return log_error_errno(r, "Failed to mark event source as non-floating: %m");
2151
2152 return varlink_reply(link, NULL);
2153 }
2154
2155 static void synchronize_destroy(void *userdata) {
2156 varlink_unref(userdata);
2157 }
2158
2159 static int vl_method_synchronize(Varlink *link, JsonVariant *parameters, VarlinkMethodFlags flags, void *userdata) {
2160 _cleanup_(sd_event_source_unrefp) sd_event_source *event_source = NULL;
2161 Server *s = ASSERT_PTR(userdata);
2162 int r;
2163
2164 assert(link);
2165
2166 if (json_variant_elements(parameters) > 0)
2167 return varlink_error_invalid_parameter(link, parameters);
2168
2169 log_info("Received client request to sync journal.");
2170
2171 /* We don't do the main work now, but instead enqueue a deferred event loop job which will do
2172 * it. That job is scheduled at low priority, so that we return from this method call only after all
2173 * queued but not processed log messages are written to disk, so that this method call returning can
2174 * be used as nice synchronization point. */
2175 r = sd_event_add_defer(s->event, &event_source, synchronize_second_half, link);
2176 if (r < 0)
2177 return log_error_errno(r, "Failed to allocate defer event source: %m");
2178
2179 r = sd_event_source_set_destroy_callback(event_source, synchronize_destroy);
2180 if (r < 0)
2181 return log_error_errno(r, "Failed to set event source destroy callback: %m");
2182
2183 varlink_ref(link); /* The varlink object is now left to the destroy callback to unref */
2184
2185 r = sd_event_source_set_priority(event_source, SD_EVENT_PRIORITY_NORMAL+15);
2186 if (r < 0)
2187 return log_error_errno(r, "Failed to set defer event source priority: %m");
2188
2189 /* Give up ownership of this event source. It will now be destroyed along with event loop itself,
2190 * unless it destroys itself earlier. */
2191 r = sd_event_source_set_floating(event_source, true);
2192 if (r < 0)
2193 return log_error_errno(r, "Failed to mark event source as floating: %m");
2194
2195 (void) sd_event_source_set_description(event_source, "deferred-sync");
2196
2197 return 0;
2198 }
2199
2200 static int vl_method_rotate(Varlink *link, JsonVariant *parameters, VarlinkMethodFlags flags, void *userdata) {
2201 Server *s = ASSERT_PTR(userdata);
2202
2203 assert(link);
2204
2205 if (json_variant_elements(parameters) > 0)
2206 return varlink_error_invalid_parameter(link, parameters);
2207
2208 log_info("Received client request to rotate journal, rotating.");
2209 server_full_rotate(s);
2210
2211 return varlink_reply(link, NULL);
2212 }
2213
2214 static int vl_method_flush_to_var(Varlink *link, JsonVariant *parameters, VarlinkMethodFlags flags, void *userdata) {
2215 Server *s = ASSERT_PTR(userdata);
2216
2217 assert(link);
2218
2219 if (json_variant_elements(parameters) > 0)
2220 return varlink_error_invalid_parameter(link, parameters);
2221 if (s->namespace)
2222 return varlink_error(link, "io.systemd.Journal.NotSupportedByNamespaces", NULL);
2223
2224 log_info("Received client request to flush runtime journal.");
2225 server_full_flush(s);
2226
2227 return varlink_reply(link, NULL);
2228 }
2229
2230 static int vl_method_relinquish_var(Varlink *link, JsonVariant *parameters, VarlinkMethodFlags flags, void *userdata) {
2231 Server *s = ASSERT_PTR(userdata);
2232
2233 assert(link);
2234
2235 if (json_variant_elements(parameters) > 0)
2236 return varlink_error_invalid_parameter(link, parameters);
2237 if (s->namespace)
2238 return varlink_error(link, "io.systemd.Journal.NotSupportedByNamespaces", NULL);
2239
2240 log_info("Received client request to relinquish %s access.", s->system_storage.path);
2241 server_relinquish_var(s);
2242
2243 return varlink_reply(link, NULL);
2244 }
2245
2246 static int vl_connect(VarlinkServer *server, Varlink *link, void *userdata) {
2247 Server *s = ASSERT_PTR(userdata);
2248
2249 assert(server);
2250 assert(link);
2251
2252 (void) server_start_or_stop_idle_timer(s); /* maybe we are no longer idle */
2253
2254 return 0;
2255 }
2256
2257 static void vl_disconnect(VarlinkServer *server, Varlink *link, void *userdata) {
2258 Server *s = ASSERT_PTR(userdata);
2259
2260 assert(server);
2261 assert(link);
2262
2263 (void) server_start_or_stop_idle_timer(s); /* maybe we are idle now */
2264 }
2265
2266 static int server_open_varlink(Server *s, const char *socket, int fd) {
2267 int r;
2268
2269 assert(s);
2270
2271 r = varlink_server_new(&s->varlink_server, VARLINK_SERVER_ROOT_ONLY|VARLINK_SERVER_INHERIT_USERDATA);
2272 if (r < 0)
2273 return r;
2274
2275 varlink_server_set_userdata(s->varlink_server, s);
2276
2277 r = varlink_server_add_interface(s->varlink_server, &vl_interface_io_systemd_Journal);
2278 if (r < 0)
2279 return log_error_errno(r, "Failed to add Journal interface to varlink server: %m");
2280
2281 r = varlink_server_bind_method_many(
2282 s->varlink_server,
2283 "io.systemd.Journal.Synchronize", vl_method_synchronize,
2284 "io.systemd.Journal.Rotate", vl_method_rotate,
2285 "io.systemd.Journal.FlushToVar", vl_method_flush_to_var,
2286 "io.systemd.Journal.RelinquishVar", vl_method_relinquish_var);
2287 if (r < 0)
2288 return r;
2289
2290 r = varlink_server_bind_connect(s->varlink_server, vl_connect);
2291 if (r < 0)
2292 return r;
2293
2294 r = varlink_server_bind_disconnect(s->varlink_server, vl_disconnect);
2295 if (r < 0)
2296 return r;
2297
2298 if (fd < 0)
2299 r = varlink_server_listen_address(s->varlink_server, socket, 0600);
2300 else
2301 r = varlink_server_listen_fd(s->varlink_server, fd);
2302 if (r < 0)
2303 return r;
2304
2305 r = varlink_server_attach_event(s->varlink_server, s->event, SD_EVENT_PRIORITY_NORMAL);
2306 if (r < 0)
2307 return r;
2308
2309 return 0;
2310 }
2311
2312 int server_map_seqnum_file(
2313 Server *s,
2314 const char *fname,
2315 size_t size,
2316 void **ret) {
2317
2318 _cleanup_free_ char *fn = NULL;
2319 _cleanup_close_ int fd = -EBADF;
2320 uint64_t *p;
2321 int r;
2322
2323 assert(s);
2324 assert(fname);
2325 assert(size > 0);
2326 assert(ret);
2327
2328 fn = path_join(s->runtime_directory, fname);
2329 if (!fn)
2330 return -ENOMEM;
2331
2332 fd = open(fn, O_RDWR|O_CREAT|O_CLOEXEC|O_NOCTTY|O_NOFOLLOW, 0644);
2333 if (fd < 0)
2334 return -errno;
2335
2336 r = posix_fallocate_loop(fd, 0, size);
2337 if (r < 0)
2338 return r;
2339
2340 p = mmap(NULL, size, PROT_READ|PROT_WRITE, MAP_SHARED, fd, 0);
2341 if (p == MAP_FAILED)
2342 return -errno;
2343
2344 *ret = p;
2345 return 0;
2346 }
2347
2348 void server_unmap_seqnum_file(void *p, size_t size) {
2349 assert(size > 0);
2350
2351 if (!p)
2352 return;
2353
2354 assert_se(munmap(p, size) >= 0);
2355 }
2356
2357 static bool server_is_idle(Server *s) {
2358 assert(s);
2359
2360 /* The server for the main namespace is never idle */
2361 if (!s->namespace)
2362 return false;
2363
2364 /* If a retention maximum is set larger than the idle time we need to be running to enforce it, hence
2365 * turn off the idle logic. */
2366 if (s->max_retention_usec > IDLE_TIMEOUT_USEC)
2367 return false;
2368
2369 /* We aren't idle if we have a varlink client */
2370 if (varlink_server_current_connections(s->varlink_server) > 0)
2371 return false;
2372
2373 /* If we have stdout streams we aren't idle */
2374 if (s->n_stdout_streams > 0)
2375 return false;
2376
2377 return true;
2378 }
2379
2380 static int server_idle_handler(sd_event_source *source, uint64_t usec, void *userdata) {
2381 Server *s = ASSERT_PTR(userdata);
2382
2383 assert(source);
2384
2385 log_debug("Server is idle, exiting.");
2386 sd_event_exit(s->event, 0);
2387 return 0;
2388 }
2389
2390 int server_start_or_stop_idle_timer(Server *s) {
2391 _cleanup_(sd_event_source_unrefp) sd_event_source *source = NULL;
2392 int r;
2393
2394 assert(s);
2395
2396 if (!server_is_idle(s)) {
2397 s->idle_event_source = sd_event_source_disable_unref(s->idle_event_source);
2398 return 0;
2399 }
2400
2401 if (s->idle_event_source)
2402 return 1;
2403
2404 r = sd_event_add_time_relative(s->event, &source, CLOCK_MONOTONIC, IDLE_TIMEOUT_USEC, 0, server_idle_handler, s);
2405 if (r < 0)
2406 return log_error_errno(r, "Failed to allocate idle timer: %m");
2407
2408 r = sd_event_source_set_priority(source, SD_EVENT_PRIORITY_IDLE);
2409 if (r < 0)
2410 return log_error_errno(r, "Failed to set idle timer priority: %m");
2411
2412 (void) sd_event_source_set_description(source, "idle-timer");
2413
2414 s->idle_event_source = TAKE_PTR(source);
2415 return 1;
2416 }
2417
2418 int server_refresh_idle_timer(Server *s) {
2419 int r;
2420
2421 assert(s);
2422
2423 if (!s->idle_event_source)
2424 return 0;
2425
2426 r = sd_event_source_set_time_relative(s->idle_event_source, IDLE_TIMEOUT_USEC);
2427 if (r < 0)
2428 return log_error_errno(r, "Failed to refresh idle timer: %m");
2429
2430 return 1;
2431 }
2432
2433 static int server_set_namespace(Server *s, const char *namespace) {
2434 assert(s);
2435
2436 if (!namespace)
2437 return 0;
2438
2439 if (!log_namespace_name_valid(namespace))
2440 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "Specified namespace name not valid, refusing: %s", namespace);
2441
2442 s->namespace = strdup(namespace);
2443 if (!s->namespace)
2444 return log_oom();
2445
2446 s->namespace_field = strjoin("_NAMESPACE=", namespace);
2447 if (!s->namespace_field)
2448 return log_oom();
2449
2450 return 1;
2451 }
2452
2453 static int server_memory_pressure(sd_event_source *es, void *userdata) {
2454 Server *s = ASSERT_PTR(userdata);
2455
2456 log_info("Under memory pressure, flushing caches.");
2457
2458 /* Flushed the cached info we might have about client processes */
2459 client_context_flush_regular(s);
2460
2461 /* Let's also close all user files (but keep the system/runtime one open) */
2462 for (;;) {
2463 JournalFile *first = ordered_hashmap_steal_first(s->user_journals);
2464
2465 if (!first)
2466 break;
2467
2468 (void) journal_file_offline_close(first);
2469 }
2470
2471 sd_event_trim_memory();
2472
2473 return 0;
2474 }
2475
2476 static int server_setup_memory_pressure(Server *s) {
2477 int r;
2478
2479 assert(s);
2480
2481 r = sd_event_add_memory_pressure(s->event, NULL, server_memory_pressure, s);
2482 if (r < 0)
2483 log_full_errno(ERRNO_IS_NOT_SUPPORTED(r) || ERRNO_IS_PRIVILEGE(r) || (r == -EHOSTDOWN) ? LOG_DEBUG : LOG_NOTICE, r,
2484 "Failed to install memory pressure event source, ignoring: %m");
2485
2486 return 0;
2487 }
2488
2489 static void server_load_credentials(Server *s) {
2490 _cleanup_free_ void *data = NULL;
2491 int r;
2492
2493 assert(s);
2494
2495 /* if we already have a forward address from config don't load the credential */
2496 if (s->forward_to_socket.sockaddr.sa.sa_family != AF_UNSPEC)
2497 return log_debug("Socket forward address already set not loading journal.forward_to_socket");
2498
2499 r = read_credential("journal.forward_to_socket", &data, NULL);
2500 if (r < 0)
2501 return (void) log_debug_errno(r, "Failed to read credential journal.forward_to_socket, ignoring: %m");
2502
2503 r = socket_address_parse(&s->forward_to_socket, data);
2504 if (r < 0)
2505 log_debug_errno(r, "Failed to parse credential journal.forward_to_socket, ignoring: %m");
2506 }
2507
2508 int server_new(Server **ret) {
2509 _cleanup_(server_freep) Server *s = NULL;
2510
2511 assert(ret);
2512
2513 s = new(Server, 1);
2514 if (!s)
2515 return -ENOMEM;
2516
2517 *s = (Server) {
2518 .syslog_fd = -EBADF,
2519 .native_fd = -EBADF,
2520 .stdout_fd = -EBADF,
2521 .dev_kmsg_fd = -EBADF,
2522 .audit_fd = -EBADF,
2523 .hostname_fd = -EBADF,
2524 .notify_fd = -EBADF,
2525 .forward_socket_fd = -EBADF,
2526
2527 .compress.enabled = true,
2528 .compress.threshold_bytes = UINT64_MAX,
2529 .seal = true,
2530
2531 .set_audit = true,
2532
2533 .watchdog_usec = USEC_INFINITY,
2534
2535 .sync_interval_usec = DEFAULT_SYNC_INTERVAL_USEC,
2536 .sync_scheduled = false,
2537
2538 .ratelimit_interval = DEFAULT_RATE_LIMIT_INTERVAL,
2539 .ratelimit_burst = DEFAULT_RATE_LIMIT_BURST,
2540
2541 .forward_to_wall = true,
2542 .forward_to_socket = { .sockaddr.sa.sa_family = AF_UNSPEC },
2543
2544 .max_file_usec = DEFAULT_MAX_FILE_USEC,
2545
2546 .max_level_store = LOG_DEBUG,
2547 .max_level_syslog = LOG_DEBUG,
2548 .max_level_kmsg = LOG_NOTICE,
2549 .max_level_console = LOG_INFO,
2550 .max_level_wall = LOG_EMERG,
2551 .max_level_socket = LOG_DEBUG,
2552
2553 .line_max = DEFAULT_LINE_MAX,
2554
2555 .runtime_storage.name = "Runtime Journal",
2556 .system_storage.name = "System Journal",
2557
2558 .kmsg_own_ratelimit = {
2559 .interval = DEFAULT_KMSG_OWN_INTERVAL,
2560 .burst = DEFAULT_KMSG_OWN_BURST,
2561 },
2562
2563 .sigrtmin18_info.memory_pressure_handler = server_memory_pressure,
2564 .sigrtmin18_info.memory_pressure_userdata = s,
2565 };
2566
2567 *ret = TAKE_PTR(s);
2568 return 0;
2569 }
2570
2571 int server_init(Server *s, const char *namespace) {
2572 const char *native_socket, *syslog_socket, *stdout_socket, *varlink_socket, *e;
2573 _cleanup_fdset_free_ FDSet *fds = NULL;
2574 int n, r, varlink_fd = -EBADF;
2575 bool no_sockets;
2576
2577 assert(s);
2578
2579 r = server_set_namespace(s, namespace);
2580 if (r < 0)
2581 return r;
2582
2583 /* By default, only read from /dev/kmsg if are the main namespace */
2584 s->read_kmsg = !s->namespace;
2585 s->storage = s->namespace ? STORAGE_PERSISTENT : STORAGE_AUTO;
2586
2587 journal_reset_metrics(&s->system_storage.metrics);
2588 journal_reset_metrics(&s->runtime_storage.metrics);
2589
2590 server_parse_config_file(s);
2591
2592 server_load_credentials(s);
2593
2594 if (!s->namespace) {
2595 /* Parse kernel command line, but only if we are not a namespace instance */
2596 r = proc_cmdline_parse(parse_proc_cmdline_item, s, PROC_CMDLINE_STRIP_RD_PREFIX);
2597 if (r < 0)
2598 log_warning_errno(r, "Failed to parse kernel command line, ignoring: %m");
2599 }
2600
2601 if (!!s->ratelimit_interval != !!s->ratelimit_burst) { /* One set to 0 and the other not? */
2602 log_debug("Setting both rate limit interval and burst from "USEC_FMT",%u to 0,0",
2603 s->ratelimit_interval, s->ratelimit_burst);
2604 s->ratelimit_interval = s->ratelimit_burst = 0;
2605 }
2606
2607 e = getenv("RUNTIME_DIRECTORY");
2608 if (e)
2609 s->runtime_directory = strdup(e);
2610 else if (s->namespace)
2611 s->runtime_directory = strjoin("/run/systemd/journal.", s->namespace);
2612 else
2613 s->runtime_directory = strdup("/run/systemd/journal");
2614 if (!s->runtime_directory)
2615 return log_oom();
2616
2617 (void) mkdir_p(s->runtime_directory, 0755);
2618
2619 s->user_journals = ordered_hashmap_new(NULL);
2620 if (!s->user_journals)
2621 return log_oom();
2622
2623 s->mmap = mmap_cache_new();
2624 if (!s->mmap)
2625 return log_oom();
2626
2627 s->deferred_closes = set_new(NULL);
2628 if (!s->deferred_closes)
2629 return log_oom();
2630
2631 r = sd_event_default(&s->event);
2632 if (r < 0)
2633 return log_error_errno(r, "Failed to create event loop: %m");
2634
2635 n = sd_listen_fds(true);
2636 if (n < 0)
2637 return log_error_errno(n, "Failed to read listening file descriptors from environment: %m");
2638
2639 native_socket = strjoina(s->runtime_directory, "/socket");
2640 stdout_socket = strjoina(s->runtime_directory, "/stdout");
2641 syslog_socket = strjoina(s->runtime_directory, "/dev-log");
2642 varlink_socket = strjoina(s->runtime_directory, "/io.systemd.journal");
2643
2644 for (int fd = SD_LISTEN_FDS_START; fd < SD_LISTEN_FDS_START + n; fd++)
2645
2646 if (sd_is_socket_unix(fd, SOCK_DGRAM, -1, native_socket, 0) > 0) {
2647
2648 if (s->native_fd >= 0)
2649 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
2650 "Too many native sockets passed.");
2651
2652 s->native_fd = fd;
2653
2654 } else if (sd_is_socket_unix(fd, SOCK_STREAM, 1, stdout_socket, 0) > 0) {
2655
2656 if (s->stdout_fd >= 0)
2657 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
2658 "Too many stdout sockets passed.");
2659
2660 s->stdout_fd = fd;
2661
2662 } else if (sd_is_socket_unix(fd, SOCK_DGRAM, -1, syslog_socket, 0) > 0) {
2663
2664 if (s->syslog_fd >= 0)
2665 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
2666 "Too many /dev/log sockets passed.");
2667
2668 s->syslog_fd = fd;
2669
2670 } else if (sd_is_socket_unix(fd, SOCK_STREAM, 1, varlink_socket, 0) > 0) {
2671
2672 if (varlink_fd >= 0)
2673 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
2674 "Too many varlink sockets passed.");
2675
2676 varlink_fd = fd;
2677 } else if (sd_is_socket(fd, AF_NETLINK, SOCK_RAW, -1) > 0) {
2678
2679 if (s->audit_fd >= 0)
2680 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
2681 "Too many audit sockets passed.");
2682
2683 s->audit_fd = fd;
2684
2685 } else {
2686
2687 if (!fds) {
2688 fds = fdset_new();
2689 if (!fds)
2690 return log_oom();
2691 }
2692
2693 r = fdset_put(fds, fd);
2694 if (r < 0)
2695 return log_oom();
2696 }
2697
2698 /* Try to restore streams, but don't bother if this fails */
2699 (void) server_restore_streams(s, fds);
2700
2701 if (!fdset_isempty(fds)) {
2702 log_warning("%u unknown file descriptors passed, closing.", fdset_size(fds));
2703 fds = fdset_free(fds);
2704 }
2705
2706 no_sockets = s->native_fd < 0 && s->stdout_fd < 0 && s->syslog_fd < 0 && s->audit_fd < 0 && varlink_fd < 0;
2707
2708 /* always open stdout, syslog, native, and kmsg sockets */
2709
2710 /* systemd-journald.socket: /run/systemd/journal/stdout */
2711 r = server_open_stdout_socket(s, stdout_socket);
2712 if (r < 0)
2713 return r;
2714
2715 /* systemd-journald-dev-log.socket: /run/systemd/journal/dev-log */
2716 r = server_open_syslog_socket(s, syslog_socket);
2717 if (r < 0)
2718 return r;
2719
2720 /* systemd-journald.socket: /run/systemd/journal/socket */
2721 r = server_open_native_socket(s, native_socket);
2722 if (r < 0)
2723 return r;
2724
2725 /* /dev/kmsg */
2726 r = server_open_dev_kmsg(s);
2727 if (r < 0)
2728 return r;
2729
2730 /* Unless we got *some* sockets and not audit, open audit socket */
2731 if (s->audit_fd >= 0 || no_sockets) {
2732 log_info("Collecting audit messages is enabled.");
2733
2734 r = server_open_audit(s);
2735 if (r < 0)
2736 return r;
2737 } else
2738 log_info("Collecting audit messages is disabled.");
2739
2740 r = server_open_varlink(s, varlink_socket, varlink_fd);
2741 if (r < 0)
2742 return r;
2743
2744 r = server_map_seqnum_file(s, "seqnum", sizeof(SeqnumData), (void**) &s->seqnum);
2745 if (r < 0)
2746 return log_error_errno(r, "Failed to map main seqnum file: %m");
2747
2748 r = server_open_kernel_seqnum(s);
2749 if (r < 0)
2750 return r;
2751
2752 r = server_open_hostname(s);
2753 if (r < 0)
2754 return r;
2755
2756 r = server_setup_signals(s);
2757 if (r < 0)
2758 return r;
2759
2760 r = server_setup_memory_pressure(s);
2761 if (r < 0)
2762 return r;
2763
2764 s->ratelimit = journal_ratelimit_new();
2765 if (!s->ratelimit)
2766 return log_oom();
2767
2768 r = cg_get_root_path(&s->cgroup_root);
2769 if (r < 0)
2770 return log_error_errno(r, "Failed to acquire cgroup root path: %m");
2771
2772 server_cache_hostname(s);
2773 server_cache_boot_id(s);
2774 server_cache_machine_id(s);
2775
2776 if (s->namespace)
2777 s->runtime_storage.path = strjoin("/run/log/journal/", SERVER_MACHINE_ID(s), ".", s->namespace);
2778 else
2779 s->runtime_storage.path = strjoin("/run/log/journal/", SERVER_MACHINE_ID(s));
2780 if (!s->runtime_storage.path)
2781 return log_oom();
2782
2783 e = getenv("LOGS_DIRECTORY");
2784 if (e)
2785 s->system_storage.path = strdup(e);
2786 else if (s->namespace)
2787 s->system_storage.path = strjoin("/var/log/journal/", SERVER_MACHINE_ID(s), ".", s->namespace);
2788 else
2789 s->system_storage.path = strjoin("/var/log/journal/", SERVER_MACHINE_ID(s));
2790 if (!s->system_storage.path)
2791 return log_oom();
2792
2793 (void) server_connect_notify(s);
2794
2795 (void) client_context_acquire_default(s);
2796
2797 r = server_system_journal_open(s, /* flush_requested= */ false, /* relinquish_requested= */ false);
2798 if (r < 0)
2799 return r;
2800
2801 server_start_or_stop_idle_timer(s);
2802
2803 return 0;
2804 }
2805
2806 void server_maybe_append_tags(Server *s) {
2807 #if HAVE_GCRYPT
2808 JournalFile *f;
2809 usec_t n;
2810
2811 n = now(CLOCK_REALTIME);
2812
2813 if (s->system_journal)
2814 journal_file_maybe_append_tag(s->system_journal, n);
2815
2816 ORDERED_HASHMAP_FOREACH(f, s->user_journals)
2817 journal_file_maybe_append_tag(f, n);
2818 #endif
2819 }
2820
2821 Server* server_free(Server *s) {
2822 if (!s)
2823 return NULL;
2824
2825 free(s->namespace);
2826 free(s->namespace_field);
2827
2828 set_free_with_destructor(s->deferred_closes, journal_file_offline_close);
2829
2830 while (s->stdout_streams)
2831 stdout_stream_free(s->stdout_streams);
2832
2833 client_context_flush_all(s);
2834
2835 (void) journal_file_offline_close(s->system_journal);
2836 (void) journal_file_offline_close(s->runtime_journal);
2837
2838 ordered_hashmap_free_with_destructor(s->user_journals, journal_file_offline_close);
2839
2840 varlink_server_unref(s->varlink_server);
2841
2842 sd_event_source_unref(s->syslog_event_source);
2843 sd_event_source_unref(s->native_event_source);
2844 sd_event_source_unref(s->stdout_event_source);
2845 sd_event_source_unref(s->dev_kmsg_event_source);
2846 sd_event_source_unref(s->audit_event_source);
2847 sd_event_source_unref(s->sync_event_source);
2848 sd_event_source_unref(s->sigusr1_event_source);
2849 sd_event_source_unref(s->sigusr2_event_source);
2850 sd_event_source_unref(s->sigterm_event_source);
2851 sd_event_source_unref(s->sigint_event_source);
2852 sd_event_source_unref(s->sigrtmin1_event_source);
2853 sd_event_source_unref(s->hostname_event_source);
2854 sd_event_source_unref(s->notify_event_source);
2855 sd_event_source_unref(s->watchdog_event_source);
2856 sd_event_source_unref(s->idle_event_source);
2857 sd_event_unref(s->event);
2858
2859 safe_close(s->syslog_fd);
2860 safe_close(s->native_fd);
2861 safe_close(s->stdout_fd);
2862 safe_close(s->dev_kmsg_fd);
2863 safe_close(s->audit_fd);
2864 safe_close(s->hostname_fd);
2865 safe_close(s->notify_fd);
2866 safe_close(s->forward_socket_fd);
2867
2868 if (s->ratelimit)
2869 journal_ratelimit_free(s->ratelimit);
2870
2871 server_unmap_seqnum_file(s->seqnum, sizeof(*s->seqnum));
2872 server_unmap_seqnum_file(s->kernel_seqnum, sizeof(*s->kernel_seqnum));
2873
2874 free(s->buffer);
2875 free(s->tty_path);
2876 free(s->cgroup_root);
2877 free(s->hostname_field);
2878 free(s->runtime_storage.path);
2879 free(s->system_storage.path);
2880 free(s->runtime_directory);
2881
2882 mmap_cache_unref(s->mmap);
2883
2884 return mfree(s);
2885 }
2886
2887 static const char* const storage_table[_STORAGE_MAX] = {
2888 [STORAGE_AUTO] = "auto",
2889 [STORAGE_VOLATILE] = "volatile",
2890 [STORAGE_PERSISTENT] = "persistent",
2891 [STORAGE_NONE] = "none"
2892 };
2893
2894 DEFINE_STRING_TABLE_LOOKUP(storage, Storage);
2895 DEFINE_CONFIG_PARSE_ENUM(config_parse_storage, storage, Storage, "Failed to parse storage setting");
2896
2897 static const char* const split_mode_table[_SPLIT_MAX] = {
2898 [SPLIT_LOGIN] = "login",
2899 [SPLIT_UID] = "uid",
2900 [SPLIT_NONE] = "none",
2901 };
2902
2903 DEFINE_STRING_TABLE_LOOKUP(split_mode, SplitMode);
2904 DEFINE_CONFIG_PARSE_ENUM(config_parse_split_mode, split_mode, SplitMode, "Failed to parse split mode setting");
2905
2906 int config_parse_line_max(
2907 const char* unit,
2908 const char *filename,
2909 unsigned line,
2910 const char *section,
2911 unsigned section_line,
2912 const char *lvalue,
2913 int ltype,
2914 const char *rvalue,
2915 void *data,
2916 void *userdata) {
2917
2918 size_t *sz = ASSERT_PTR(data);
2919 int r;
2920
2921 assert(filename);
2922 assert(lvalue);
2923 assert(rvalue);
2924
2925 if (isempty(rvalue))
2926 /* Empty assignment means default */
2927 *sz = DEFAULT_LINE_MAX;
2928 else {
2929 uint64_t v;
2930
2931 r = parse_size(rvalue, 1024, &v);
2932 if (r < 0) {
2933 log_syntax(unit, LOG_WARNING, filename, line, r, "Failed to parse LineMax= value, ignoring: %s", rvalue);
2934 return 0;
2935 }
2936
2937 if (v < 79) {
2938 /* Why specify 79 here as minimum line length? Simply, because the most common traditional
2939 * terminal size is 80ch, and it might make sense to break one character before the natural
2940 * line break would occur on that. */
2941 log_syntax(unit, LOG_WARNING, filename, line, 0, "LineMax= too small, clamping to 79: %s", rvalue);
2942 *sz = 79;
2943 } else if (v > (uint64_t) (SSIZE_MAX-1)) {
2944 /* So, why specify SSIZE_MAX-1 here? Because that's one below the largest size value read()
2945 * can return, and we need one extra byte for the trailing NUL byte. Of course IRL such large
2946 * memory allocations will fail anyway, hence this limit is mostly theoretical anyway, as we'll
2947 * fail much earlier anyway. */
2948 log_syntax(unit, LOG_WARNING, filename, line, 0, "LineMax= too large, clamping to %" PRIu64 ": %s", (uint64_t) (SSIZE_MAX-1), rvalue);
2949 *sz = SSIZE_MAX-1;
2950 } else
2951 *sz = (size_t) v;
2952 }
2953
2954 return 0;
2955 }
2956
2957 int config_parse_compress(
2958 const char* unit,
2959 const char *filename,
2960 unsigned line,
2961 const char *section,
2962 unsigned section_line,
2963 const char *lvalue,
2964 int ltype,
2965 const char *rvalue,
2966 void *data,
2967 void *userdata) {
2968
2969 JournalCompressOptions* compress = ASSERT_PTR(data);
2970 int r;
2971
2972 assert(unit);
2973 assert(filename);
2974 assert(rvalue);
2975
2976 if (isempty(rvalue)) {
2977 compress->enabled = true;
2978 compress->threshold_bytes = UINT64_MAX;
2979 } else if (streq(rvalue, "1")) {
2980 log_syntax(unit, LOG_WARNING, filename, line, 0,
2981 "Compress= ambiguously specified as 1, enabling compression with default threshold");
2982 compress->enabled = true;
2983 } else if (streq(rvalue, "0")) {
2984 log_syntax(unit, LOG_WARNING, filename, line, 0,
2985 "Compress= ambiguously specified as 0, disabling compression");
2986 compress->enabled = false;
2987 } else {
2988 r = parse_boolean(rvalue);
2989 if (r < 0) {
2990 r = parse_size(rvalue, 1024, &compress->threshold_bytes);
2991 if (r < 0)
2992 log_syntax(unit, LOG_WARNING, filename, line, r,
2993 "Failed to parse Compress= value, ignoring: %s", rvalue);
2994 else
2995 compress->enabled = true;
2996 } else
2997 compress->enabled = r;
2998 }
2999
3000 return 0;
3001 }
3002
3003 int config_parse_forward_to_socket(
3004 const char* unit,
3005 const char *filename,
3006 unsigned line,
3007 const char *section,
3008 unsigned section_line,
3009 const char *lvalue,
3010 int ltype,
3011 const char *rvalue,
3012 void *data,
3013 void *userdata) {
3014
3015 SocketAddress* addr = ASSERT_PTR(data);
3016 int r;
3017
3018 assert(unit);
3019 assert(filename);
3020 assert(rvalue);
3021
3022 if (isempty(rvalue))
3023 *addr = (SocketAddress) { .sockaddr.sa.sa_family = AF_UNSPEC };
3024 else {
3025 r = socket_address_parse(addr, rvalue);
3026 if (r < 0)
3027 log_syntax(unit, LOG_WARNING, filename, line, r,
3028 "Failed to parse ForwardToSocket= value, ignoring: %s", rvalue);
3029 }
3030
3031 return 0;
3032 }
Unnamed repository; edit this file 'description' to name the repository.