1 /* SPDX-License-Identifier: LGPL-2.1+ */
4 #include <selinux/selinux.h>
8 #include <sys/signalfd.h>
9 #include <sys/statvfs.h>
10 #include <linux/sockios.h>
12 #include "sd-daemon.h"
13 #include "sd-journal.h"
14 #include "sd-messages.h"
17 #include "alloc-util.h"
18 #include "audit-util.h"
19 #include "cgroup-util.h"
20 #include "conf-parser.h"
21 #include "dirent-util.h"
22 #include "extract-word.h"
25 #include "format-util.h"
28 #include "hostname-util.h"
29 #include "id128-util.h"
31 #include "journal-authenticate.h"
32 #include "journal-file.h"
33 #include "journal-internal.h"
34 #include "journal-vacuum.h"
35 #include "journald-audit.h"
36 #include "journald-context.h"
37 #include "journald-kmsg.h"
38 #include "journald-native.h"
39 #include "journald-rate-limit.h"
40 #include "journald-server.h"
41 #include "journald-stream.h"
42 #include "journald-syslog.h"
44 #include "missing_audit.h"
46 #include "parse-util.h"
47 #include "path-util.h"
48 #include "proc-cmdline.h"
49 #include "process-util.h"
51 #include "selinux-util.h"
52 #include "signal-util.h"
53 #include "socket-util.h"
54 #include "stdio-util.h"
55 #include "string-table.h"
56 #include "string-util.h"
57 #include "syslog-util.h"
58 #include "user-record.h"
59 #include "user-util.h"
61 #define USER_JOURNALS_MAX 1024
63 #define DEFAULT_SYNC_INTERVAL_USEC (5*USEC_PER_MINUTE)
64 #define DEFAULT_RATE_LIMIT_INTERVAL (30*USEC_PER_SEC)
65 #define DEFAULT_RATE_LIMIT_BURST 10000
66 #define DEFAULT_MAX_FILE_USEC USEC_PER_MONTH
68 #define RECHECK_SPACE_USEC (30*USEC_PER_SEC)
70 #define NOTIFY_SNDBUF_SIZE (8*1024*1024)
72 /* The period to insert between posting changes for coalescing */
73 #define POST_CHANGE_TIMER_INTERVAL_USEC (250*USEC_PER_MSEC)
75 /* Pick a good default that is likely to fit into AF_UNIX and AF_INET SOCK_DGRAM datagrams, and even leaves some room
76 * for a bit of additional metadata. */
77 #define DEFAULT_LINE_MAX (48*1024)
79 #define DEFERRED_CLOSES_MAX (4096)
81 #define IDLE_TIMEOUT_USEC (30*USEC_PER_SEC)
83 static int determine_path_usage(
89 _cleanup_closedir_
DIR *d
= NULL
;
100 return log_full_errno(errno
== ENOENT
? LOG_DEBUG
: LOG_ERR
,
101 errno
, "Failed to open %s: %m", path
);
103 if (fstatvfs(dirfd(d
), &ss
) < 0)
104 return log_error_errno(errno
, "Failed to fstatvfs(%s): %m", path
);
106 *ret_free
= ss
.f_bsize
* ss
.f_bavail
;
108 FOREACH_DIRENT_ALL(de
, d
, break) {
111 if (!endswith(de
->d_name
, ".journal") &&
112 !endswith(de
->d_name
, ".journal~"))
115 if (fstatat(dirfd(d
), de
->d_name
, &st
, AT_SYMLINK_NOFOLLOW
) < 0) {
116 log_debug_errno(errno
, "Failed to stat %s/%s, ignoring: %m", path
, de
->d_name
);
120 if (!S_ISREG(st
.st_mode
))
123 *ret_used
+= (uint64_t) st
.st_blocks
* 512UL;
129 static void cache_space_invalidate(JournalStorageSpace
*space
) {
133 static int cache_space_refresh(Server
*s
, JournalStorage
*storage
) {
134 JournalStorageSpace
*space
;
135 JournalMetrics
*metrics
;
136 uint64_t vfs_used
, vfs_avail
, avail
;
142 metrics
= &storage
->metrics
;
143 space
= &storage
->space
;
145 ts
= now(CLOCK_MONOTONIC
);
147 if (space
->timestamp
!= 0 && space
->timestamp
+ RECHECK_SPACE_USEC
> ts
)
150 r
= determine_path_usage(s
, storage
->path
, &vfs_used
, &vfs_avail
);
154 space
->vfs_used
= vfs_used
;
155 space
->vfs_available
= vfs_avail
;
157 avail
= LESS_BY(vfs_avail
, metrics
->keep_free
);
159 space
->limit
= MIN(MAX(vfs_used
+ avail
, metrics
->min_use
), metrics
->max_use
);
160 space
->available
= LESS_BY(space
->limit
, vfs_used
);
161 space
->timestamp
= ts
;
165 static void patch_min_use(JournalStorage
*storage
) {
168 /* Let's bump the min_use limit to the current usage on disk. We do
169 * this when starting up and first opening the journal files. This way
170 * sudden spikes in disk usage will not cause journald to vacuum files
171 * without bounds. Note that this means that only a restart of journald
172 * will make it reset this value. */
174 storage
->metrics
.min_use
= MAX(storage
->metrics
.min_use
, storage
->space
.vfs_used
);
177 static JournalStorage
* server_current_storage(Server
*s
) {
180 return s
->system_journal
? &s
->system_storage
: &s
->runtime_storage
;
183 static int determine_space(Server
*s
, uint64_t *available
, uint64_t *limit
) {
189 js
= server_current_storage(s
);
191 r
= cache_space_refresh(s
, js
);
194 *available
= js
->space
.available
;
196 *limit
= js
->space
.limit
;
201 void server_space_usage_message(Server
*s
, JournalStorage
*storage
) {
202 char fb1
[FORMAT_BYTES_MAX
], fb2
[FORMAT_BYTES_MAX
], fb3
[FORMAT_BYTES_MAX
],
203 fb4
[FORMAT_BYTES_MAX
], fb5
[FORMAT_BYTES_MAX
], fb6
[FORMAT_BYTES_MAX
];
204 JournalMetrics
*metrics
;
209 storage
= server_current_storage(s
);
211 if (cache_space_refresh(s
, storage
) < 0)
214 metrics
= &storage
->metrics
;
215 format_bytes(fb1
, sizeof(fb1
), storage
->space
.vfs_used
);
216 format_bytes(fb2
, sizeof(fb2
), metrics
->max_use
);
217 format_bytes(fb3
, sizeof(fb3
), metrics
->keep_free
);
218 format_bytes(fb4
, sizeof(fb4
), storage
->space
.vfs_available
);
219 format_bytes(fb5
, sizeof(fb5
), storage
->space
.limit
);
220 format_bytes(fb6
, sizeof(fb6
), storage
->space
.available
);
222 server_driver_message(s
, 0,
223 "MESSAGE_ID=" SD_MESSAGE_JOURNAL_USAGE_STR
,
224 LOG_MESSAGE("%s (%s) is %s, max %s, %s free.",
225 storage
->name
, storage
->path
, fb1
, fb5
, fb6
),
226 "JOURNAL_NAME=%s", storage
->name
,
227 "JOURNAL_PATH=%s", storage
->path
,
228 "CURRENT_USE=%"PRIu64
, storage
->space
.vfs_used
,
229 "CURRENT_USE_PRETTY=%s", fb1
,
230 "MAX_USE=%"PRIu64
, metrics
->max_use
,
231 "MAX_USE_PRETTY=%s", fb2
,
232 "DISK_KEEP_FREE=%"PRIu64
, metrics
->keep_free
,
233 "DISK_KEEP_FREE_PRETTY=%s", fb3
,
234 "DISK_AVAILABLE=%"PRIu64
, storage
->space
.vfs_available
,
235 "DISK_AVAILABLE_PRETTY=%s", fb4
,
236 "LIMIT=%"PRIu64
, storage
->space
.limit
,
237 "LIMIT_PRETTY=%s", fb5
,
238 "AVAILABLE=%"PRIu64
, storage
->space
.available
,
239 "AVAILABLE_PRETTY=%s", fb6
,
243 static bool uid_for_system_journal(uid_t uid
) {
245 /* Returns true if the specified UID shall get its data stored in the system journal*/
247 return uid_is_system(uid
) || uid_is_dynamic(uid
) || uid
== UID_NOBODY
;
250 static void server_add_acls(JournalFile
*f
, uid_t uid
) {
256 if (uid_for_system_journal(uid
))
259 r
= fd_add_uid_acl_permission(f
->fd
, uid
, ACL_READ
);
261 log_warning_errno(r
, "Failed to set ACL on %s, ignoring: %m", f
->path
);
265 static int open_journal(
271 JournalMetrics
*metrics
,
274 _cleanup_(journal_file_closep
) JournalFile
*f
= NULL
;
282 r
= journal_file_open_reliably(fname
, flags
, 0640, s
->compress
.enabled
, s
->compress
.threshold_bytes
,
283 seal
, metrics
, s
->mmap
, s
->deferred_closes
, NULL
, &f
);
285 r
= journal_file_open(-1, fname
, flags
, 0640, s
->compress
.enabled
, s
->compress
.threshold_bytes
, seal
,
286 metrics
, s
->mmap
, s
->deferred_closes
, NULL
, &f
);
291 r
= journal_file_enable_post_change_timer(f
, s
->event
, POST_CHANGE_TIMER_INTERVAL_USEC
);
299 static bool flushed_flag_is_set(Server
*s
) {
304 /* We don't support the "flushing" concept for namespace instances, we assume them to always have
309 fn
= strjoina(s
->runtime_directory
, "/flushed");
310 return access(fn
, F_OK
) >= 0;
313 static int system_journal_open(Server
*s
, bool flush_requested
, bool relinquish_requested
) {
317 if (!s
->system_journal
&&
318 IN_SET(s
->storage
, STORAGE_PERSISTENT
, STORAGE_AUTO
) &&
319 (flush_requested
|| flushed_flag_is_set(s
)) &&
320 !relinquish_requested
) {
322 /* If in auto mode: first try to create the machine path, but not the prefix.
324 * If in persistent mode: create /var/log/journal and the machine path */
326 if (s
->storage
== STORAGE_PERSISTENT
)
327 (void) mkdir_parents(s
->system_storage
.path
, 0755);
329 (void) mkdir(s
->system_storage
.path
, 0755);
331 fn
= strjoina(s
->system_storage
.path
, "/system.journal");
332 r
= open_journal(s
, true, fn
, O_RDWR
|O_CREAT
, s
->seal
, &s
->system_storage
.metrics
, &s
->system_journal
);
334 server_add_acls(s
->system_journal
, 0);
335 (void) cache_space_refresh(s
, &s
->system_storage
);
336 patch_min_use(&s
->system_storage
);
338 if (!IN_SET(r
, -ENOENT
, -EROFS
))
339 log_warning_errno(r
, "Failed to open system journal: %m");
344 /* If the runtime journal is open, and we're post-flush, we're recovering from a failed
345 * system journal rotate (ENOSPC) for which the runtime journal was reopened.
347 * Perform an implicit flush to var, leaving the runtime journal closed, now that the system
350 if (!flush_requested
)
351 (void) server_flush_to_var(s
, true);
354 if (!s
->runtime_journal
&&
355 (s
->storage
!= STORAGE_NONE
)) {
357 fn
= strjoina(s
->runtime_storage
.path
, "/system.journal");
359 if (s
->system_journal
&& !relinquish_requested
) {
361 /* Try to open the runtime journal, but only
362 * if it already exists, so that we can flush
363 * it into the system journal */
365 r
= open_journal(s
, false, fn
, O_RDWR
, false, &s
->runtime_storage
.metrics
, &s
->runtime_journal
);
368 log_warning_errno(r
, "Failed to open runtime journal: %m");
375 /* OK, we really need the runtime journal, so create it if necessary. */
377 (void) mkdir_parents(s
->runtime_storage
.path
, 0755);
378 (void) mkdir(s
->runtime_storage
.path
, 0750);
380 r
= open_journal(s
, true, fn
, O_RDWR
|O_CREAT
, false, &s
->runtime_storage
.metrics
, &s
->runtime_journal
);
382 return log_error_errno(r
, "Failed to open runtime journal: %m");
385 if (s
->runtime_journal
) {
386 server_add_acls(s
->runtime_journal
, 0);
387 (void) cache_space_refresh(s
, &s
->runtime_storage
);
388 patch_min_use(&s
->runtime_storage
);
395 static JournalFile
* find_journal(Server
*s
, uid_t uid
) {
396 _cleanup_free_
char *p
= NULL
;
402 /* A rotate that fails to create the new journal (ENOSPC) leaves the rotated journal as NULL. Unless
403 * we revisit opening, even after space is made available we'll continue to return NULL indefinitely.
405 * system_journal_open() is a noop if the journals are already open, so we can just call it here to
406 * recover from failed rotates (or anything else that's left the journals as NULL).
408 * Fixes https://github.com/systemd/systemd/issues/3968 */
409 (void) system_journal_open(s
, false, false);
411 /* We split up user logs only on /var, not on /run. If the runtime file is open, we write to it
412 * exclusively, in order to guarantee proper order as soon as we flush /run to /var and close the
415 if (s
->runtime_journal
)
416 return s
->runtime_journal
;
418 if (uid_for_system_journal(uid
))
419 return s
->system_journal
;
421 f
= ordered_hashmap_get(s
->user_journals
, UID_TO_PTR(uid
));
425 if (asprintf(&p
, "%s/user-" UID_FMT
".journal", s
->system_storage
.path
, uid
) < 0) {
427 return s
->system_journal
;
430 /* Too many open? Then let's close one (or more) */
431 while (ordered_hashmap_size(s
->user_journals
) >= USER_JOURNALS_MAX
) {
432 assert_se(f
= ordered_hashmap_steal_first(s
->user_journals
));
433 (void) journal_file_close(f
);
436 r
= open_journal(s
, true, p
, O_RDWR
|O_CREAT
, s
->seal
, &s
->system_storage
.metrics
, &f
);
438 return s
->system_journal
;
440 r
= ordered_hashmap_put(s
->user_journals
, UID_TO_PTR(uid
), f
);
442 (void) journal_file_close(f
);
443 return s
->system_journal
;
446 server_add_acls(f
, uid
);
450 static int do_rotate(
463 r
= journal_file_rotate(f
, s
->compress
.enabled
, s
->compress
.threshold_bytes
, seal
, s
->deferred_closes
);
466 return log_error_errno(r
, "Failed to rotate %s: %m", (*f
)->path
);
468 return log_error_errno(r
, "Failed to create new %s journal: %m", name
);
471 server_add_acls(*f
, uid
);
475 static void server_process_deferred_closes(Server
*s
) {
478 /* Perform any deferred closes which aren't still offlining. */
479 SET_FOREACH(f
, s
->deferred_closes
) {
480 if (journal_file_is_offlining(f
))
483 (void) set_remove(s
->deferred_closes
, f
);
484 (void) journal_file_close(f
);
488 static void server_vacuum_deferred_closes(Server
*s
) {
491 /* Make some room in the deferred closes list, so that it doesn't grow without bounds */
492 if (set_size(s
->deferred_closes
) < DEFERRED_CLOSES_MAX
)
495 /* Let's first remove all journal files that might already have completed closing */
496 server_process_deferred_closes(s
);
498 /* And now, let's close some more until we reach the limit again. */
499 while (set_size(s
->deferred_closes
) >= DEFERRED_CLOSES_MAX
) {
502 assert_se(f
= set_steal_first(s
->deferred_closes
));
503 journal_file_close(f
);
507 static int vacuum_offline_user_journals(Server
*s
) {
508 _cleanup_closedir_
DIR *d
= NULL
;
513 d
= opendir(s
->system_storage
.path
);
518 return log_error_errno(errno
, "Failed to open %s: %m", s
->system_storage
.path
);
522 _cleanup_free_
char *u
= NULL
, *full
= NULL
;
523 _cleanup_close_
int fd
= -1;
530 de
= readdir_no_dot(d
);
533 log_warning_errno(errno
, "Failed to enumerate %s, ignoring: %m", s
->system_storage
.path
);
538 a
= startswith(de
->d_name
, "user-");
541 b
= endswith(de
->d_name
, ".journal");
549 r
= parse_uid(u
, &uid
);
551 log_debug_errno(r
, "Failed to parse UID from file name '%s', ignoring: %m", de
->d_name
);
555 /* Already rotated in the above loop? i.e. is it an open user journal? */
556 if (ordered_hashmap_contains(s
->user_journals
, UID_TO_PTR(uid
)))
559 full
= path_join(s
->system_storage
.path
, de
->d_name
);
563 fd
= openat(dirfd(d
), de
->d_name
, O_RDWR
|O_CLOEXEC
|O_NOCTTY
|O_NOFOLLOW
|O_NONBLOCK
);
565 log_full_errno(IN_SET(errno
, ELOOP
, ENOENT
) ? LOG_DEBUG
: LOG_WARNING
, errno
,
566 "Failed to open journal file '%s' for rotation: %m", full
);
570 /* Make some room in the set of deferred close()s */
571 server_vacuum_deferred_closes(s
);
573 /* Open the file briefly, so that we can archive it */
574 r
= journal_file_open(fd
,
579 s
->compress
.threshold_bytes
,
581 &s
->system_storage
.metrics
,
587 log_warning_errno(r
, "Failed to read journal file %s for rotation, trying to move it out of the way: %m", full
);
589 r
= journal_file_dispose(dirfd(d
), de
->d_name
);
591 log_warning_errno(r
, "Failed to move %s out of the way, ignoring: %m", full
);
593 log_debug("Successfully moved %s out of the way.", full
);
598 TAKE_FD(fd
); /* Donated to journal_file_open() */
600 r
= journal_file_archive(f
);
602 log_debug_errno(r
, "Failed to archive journal file '%s', ignoring: %m", full
);
604 f
= journal_initiate_close(f
, s
->deferred_closes
);
610 void server_rotate(Server
*s
) {
615 log_debug("Rotating...");
617 /* First, rotate the system journal (either in its runtime flavour or in its runtime flavour) */
618 (void) do_rotate(s
, &s
->runtime_journal
, "runtime", false, 0);
619 (void) do_rotate(s
, &s
->system_journal
, "system", s
->seal
, 0);
621 /* Then, rotate all user journals we have open (keeping them open) */
622 ORDERED_HASHMAP_FOREACH_KEY(f
, k
, s
->user_journals
) {
623 r
= do_rotate(s
, &f
, "user", s
->seal
, PTR_TO_UID(k
));
625 ordered_hashmap_replace(s
->user_journals
, k
, f
);
627 /* Old file has been closed and deallocated */
628 ordered_hashmap_remove(s
->user_journals
, k
);
631 /* Finally, also rotate all user journals we currently do not have open. (But do so only if we
632 * actually have access to /var, i.e. are not in the log-to-runtime-journal mode). */
633 if (!s
->runtime_journal
)
634 (void) vacuum_offline_user_journals(s
);
636 server_process_deferred_closes(s
);
639 void server_sync(Server
*s
) {
643 if (s
->system_journal
) {
644 r
= journal_file_set_offline(s
->system_journal
, false);
646 log_warning_errno(r
, "Failed to sync system journal, ignoring: %m");
649 ORDERED_HASHMAP_FOREACH(f
, s
->user_journals
) {
650 r
= journal_file_set_offline(f
, false);
652 log_warning_errno(r
, "Failed to sync user journal, ignoring: %m");
655 if (s
->sync_event_source
) {
656 r
= sd_event_source_set_enabled(s
->sync_event_source
, SD_EVENT_OFF
);
658 log_error_errno(r
, "Failed to disable sync timer source: %m");
661 s
->sync_scheduled
= false;
664 static void do_vacuum(Server
*s
, JournalStorage
*storage
, bool verbose
) {
671 (void) cache_space_refresh(s
, storage
);
674 server_space_usage_message(s
, storage
);
676 r
= journal_directory_vacuum(storage
->path
, storage
->space
.limit
,
677 storage
->metrics
.n_max_files
, s
->max_retention_usec
,
678 &s
->oldest_file_usec
, verbose
);
679 if (r
< 0 && r
!= -ENOENT
)
680 log_warning_errno(r
, "Failed to vacuum %s, ignoring: %m", storage
->path
);
682 cache_space_invalidate(&storage
->space
);
685 int server_vacuum(Server
*s
, bool verbose
) {
688 log_debug("Vacuuming...");
690 s
->oldest_file_usec
= 0;
692 if (s
->system_journal
)
693 do_vacuum(s
, &s
->system_storage
, verbose
);
694 if (s
->runtime_journal
)
695 do_vacuum(s
, &s
->runtime_storage
, verbose
);
700 static void server_cache_machine_id(Server
*s
) {
706 r
= sd_id128_get_machine(&id
);
710 sd_id128_to_string(id
, stpcpy(s
->machine_id_field
, "_MACHINE_ID="));
713 static void server_cache_boot_id(Server
*s
) {
719 r
= sd_id128_get_boot(&id
);
723 sd_id128_to_string(id
, stpcpy(s
->boot_id_field
, "_BOOT_ID="));
726 static void server_cache_hostname(Server
*s
) {
727 _cleanup_free_
char *t
= NULL
;
732 t
= gethostname_malloc();
736 x
= strjoin("_HOSTNAME=", t
);
740 free_and_replace(s
->hostname_field
, x
);
743 static bool shall_try_append_again(JournalFile
*f
, int r
) {
746 case -E2BIG
: /* Hit configured limit */
747 case -EFBIG
: /* Hit fs limit */
748 case -EDQUOT
: /* Quota limit hit */
749 case -ENOSPC
: /* Disk full */
750 log_debug("%s: Allocation limit reached, rotating.", f
->path
);
753 case -EIO
: /* I/O error of some kind (mmap) */
754 log_warning("%s: IO error, rotating.", f
->path
);
757 case -EHOSTDOWN
: /* Other machine */
758 log_info("%s: Journal file from other machine, rotating.", f
->path
);
761 case -EBUSY
: /* Unclean shutdown */
762 log_info("%s: Unclean shutdown, rotating.", f
->path
);
765 case -EPROTONOSUPPORT
: /* Unsupported feature */
766 log_info("%s: Unsupported feature, rotating.", f
->path
);
769 case -EBADMSG
: /* Corrupted */
770 case -ENODATA
: /* Truncated */
771 case -ESHUTDOWN
: /* Already archived */
772 log_warning("%s: Journal file corrupted, rotating.", f
->path
);
775 case -EIDRM
: /* Journal file has been deleted */
776 log_warning("%s: Journal file has been deleted, rotating.", f
->path
);
779 case -ETXTBSY
: /* Journal file is from the future */
780 log_warning("%s: Journal file is from the future, rotating.", f
->path
);
784 log_warning("%s: underlying file system does not support memory mapping or another required file system feature.", f
->path
);
792 static void write_to_journal(Server
*s
, uid_t uid
, struct iovec
*iovec
, size_t n
, int priority
) {
793 bool vacuumed
= false, rotate
= false;
794 struct dual_timestamp ts
;
802 /* Get the closest, linearized time we have for this log event from the event loop. (Note that we do not use
803 * the source time, and not even the time the event was originally seen, but instead simply the time we started
804 * processing it, as we want strictly linear ordering in what we write out.) */
805 assert_se(sd_event_now(s
->event
, CLOCK_REALTIME
, &ts
.realtime
) >= 0);
806 assert_se(sd_event_now(s
->event
, CLOCK_MONOTONIC
, &ts
.monotonic
) >= 0);
808 if (ts
.realtime
< s
->last_realtime_clock
) {
809 /* When the time jumps backwards, let's immediately rotate. Of course, this should not happen during
810 * regular operation. However, when it does happen, then we should make sure that we start fresh files
811 * to ensure that the entries in the journal files are strictly ordered by time, in order to ensure
812 * bisection works correctly. */
814 log_debug("Time jumped backwards, rotating.");
818 f
= find_journal(s
, uid
);
822 if (journal_file_rotate_suggested(f
, s
->max_file_usec
)) {
823 log_debug("%s: Journal header limits reached or header out-of-date, rotating.", f
->path
);
830 server_vacuum(s
, false);
833 f
= find_journal(s
, uid
);
838 s
->last_realtime_clock
= ts
.realtime
;
840 r
= journal_file_append_entry(f
, &ts
, NULL
, iovec
, n
, &s
->seqnum
, NULL
, NULL
);
842 server_schedule_sync(s
, priority
);
846 if (vacuumed
|| !shall_try_append_again(f
, r
)) {
847 log_error_errno(r
, "Failed to write entry (%zu items, %zu bytes), ignoring: %m", n
, IOVEC_TOTAL_SIZE(iovec
, n
));
852 server_vacuum(s
, false);
854 f
= find_journal(s
, uid
);
858 log_debug("Retrying write.");
859 r
= journal_file_append_entry(f
, &ts
, NULL
, iovec
, n
, &s
->seqnum
, NULL
, NULL
);
861 log_error_errno(r
, "Failed to write entry (%zu items, %zu bytes) despite vacuuming, ignoring: %m", n
, IOVEC_TOTAL_SIZE(iovec
, n
));
863 server_schedule_sync(s
, priority
);
866 #define IOVEC_ADD_NUMERIC_FIELD(iovec, n, value, type, isset, format, field) \
867 if (isset(value)) { \
869 k = newa(char, STRLEN(field "=") + DECIMAL_STR_MAX(type) + 1); \
870 sprintf(k, field "=" format, value); \
871 iovec[n++] = IOVEC_MAKE_STRING(k); \
874 #define IOVEC_ADD_STRING_FIELD(iovec, n, value, field) \
875 if (!isempty(value)) { \
877 k = strjoina(field "=", value); \
878 iovec[n++] = IOVEC_MAKE_STRING(k); \
881 #define IOVEC_ADD_ID128_FIELD(iovec, n, value, field) \
882 if (!sd_id128_is_null(value)) { \
884 k = newa(char, STRLEN(field "=") + SD_ID128_STRING_MAX); \
885 sd_id128_to_string(value, stpcpy(k, field "=")); \
886 iovec[n++] = IOVEC_MAKE_STRING(k); \
889 #define IOVEC_ADD_SIZED_FIELD(iovec, n, value, value_size, field) \
890 if (value_size > 0) { \
892 k = newa(char, STRLEN(field "=") + value_size + 1); \
893 *((char*) mempcpy(stpcpy(k, field "="), value, value_size)) = 0; \
894 iovec[n++] = IOVEC_MAKE_STRING(k); \
897 static void dispatch_message_real(
899 struct iovec
*iovec
, size_t n
, size_t m
,
900 const ClientContext
*c
,
901 const struct timeval
*tv
,
905 char source_time
[sizeof("_SOURCE_REALTIME_TIMESTAMP=") + DECIMAL_STR_MAX(usec_t
)];
906 _cleanup_free_
char *cmdline1
= NULL
, *cmdline2
= NULL
;
914 N_IOVEC_META_FIELDS
+
915 (pid_is_valid(object_pid
) ? N_IOVEC_OBJECT_FIELDS
: 0) +
916 client_context_extra_fields_n_iovec(c
) <= m
);
919 IOVEC_ADD_NUMERIC_FIELD(iovec
, n
, c
->pid
, pid_t
, pid_is_valid
, PID_FMT
, "_PID");
920 IOVEC_ADD_NUMERIC_FIELD(iovec
, n
, c
->uid
, uid_t
, uid_is_valid
, UID_FMT
, "_UID");
921 IOVEC_ADD_NUMERIC_FIELD(iovec
, n
, c
->gid
, gid_t
, gid_is_valid
, GID_FMT
, "_GID");
923 IOVEC_ADD_STRING_FIELD(iovec
, n
, c
->comm
, "_COMM"); /* At most TASK_COMM_LENGTH (16 bytes) */
924 IOVEC_ADD_STRING_FIELD(iovec
, n
, c
->exe
, "_EXE"); /* A path, so at most PATH_MAX (4096 bytes) */
927 /* At most _SC_ARG_MAX (2MB usually), which is too much to put on stack.
928 * Let's use a heap allocation for this one. */
929 cmdline1
= set_iovec_string_field(iovec
, &n
, "_CMDLINE=", c
->cmdline
);
931 IOVEC_ADD_STRING_FIELD(iovec
, n
, c
->capeff
, "_CAP_EFFECTIVE"); /* Read from /proc/.../status */
932 IOVEC_ADD_SIZED_FIELD(iovec
, n
, c
->label
, c
->label_size
, "_SELINUX_CONTEXT");
933 IOVEC_ADD_NUMERIC_FIELD(iovec
, n
, c
->auditid
, uint32_t, audit_session_is_valid
, "%" PRIu32
, "_AUDIT_SESSION");
934 IOVEC_ADD_NUMERIC_FIELD(iovec
, n
, c
->loginuid
, uid_t
, uid_is_valid
, UID_FMT
, "_AUDIT_LOGINUID");
936 IOVEC_ADD_STRING_FIELD(iovec
, n
, c
->cgroup
, "_SYSTEMD_CGROUP"); /* A path */
937 IOVEC_ADD_STRING_FIELD(iovec
, n
, c
->session
, "_SYSTEMD_SESSION");
938 IOVEC_ADD_NUMERIC_FIELD(iovec
, n
, c
->owner_uid
, uid_t
, uid_is_valid
, UID_FMT
, "_SYSTEMD_OWNER_UID");
939 IOVEC_ADD_STRING_FIELD(iovec
, n
, c
->unit
, "_SYSTEMD_UNIT"); /* Unit names are bounded by UNIT_NAME_MAX */
940 IOVEC_ADD_STRING_FIELD(iovec
, n
, c
->user_unit
, "_SYSTEMD_USER_UNIT");
941 IOVEC_ADD_STRING_FIELD(iovec
, n
, c
->slice
, "_SYSTEMD_SLICE");
942 IOVEC_ADD_STRING_FIELD(iovec
, n
, c
->user_slice
, "_SYSTEMD_USER_SLICE");
944 IOVEC_ADD_ID128_FIELD(iovec
, n
, c
->invocation_id
, "_SYSTEMD_INVOCATION_ID");
946 if (c
->extra_fields_n_iovec
> 0) {
947 memcpy(iovec
+ n
, c
->extra_fields_iovec
, c
->extra_fields_n_iovec
* sizeof(struct iovec
));
948 n
+= c
->extra_fields_n_iovec
;
954 if (pid_is_valid(object_pid
) && client_context_get(s
, object_pid
, NULL
, NULL
, 0, NULL
, &o
) >= 0) {
956 IOVEC_ADD_NUMERIC_FIELD(iovec
, n
, o
->pid
, pid_t
, pid_is_valid
, PID_FMT
, "OBJECT_PID");
957 IOVEC_ADD_NUMERIC_FIELD(iovec
, n
, o
->uid
, uid_t
, uid_is_valid
, UID_FMT
, "OBJECT_UID");
958 IOVEC_ADD_NUMERIC_FIELD(iovec
, n
, o
->gid
, gid_t
, gid_is_valid
, GID_FMT
, "OBJECT_GID");
960 /* See above for size limits, only ->cmdline may be large, so use a heap allocation for it. */
961 IOVEC_ADD_STRING_FIELD(iovec
, n
, o
->comm
, "OBJECT_COMM");
962 IOVEC_ADD_STRING_FIELD(iovec
, n
, o
->exe
, "OBJECT_EXE");
964 cmdline2
= set_iovec_string_field(iovec
, &n
, "OBJECT_CMDLINE=", o
->cmdline
);
966 IOVEC_ADD_STRING_FIELD(iovec
, n
, o
->capeff
, "OBJECT_CAP_EFFECTIVE");
967 IOVEC_ADD_SIZED_FIELD(iovec
, n
, o
->label
, o
->label_size
, "OBJECT_SELINUX_CONTEXT");
968 IOVEC_ADD_NUMERIC_FIELD(iovec
, n
, o
->auditid
, uint32_t, audit_session_is_valid
, "%" PRIu32
, "OBJECT_AUDIT_SESSION");
969 IOVEC_ADD_NUMERIC_FIELD(iovec
, n
, o
->loginuid
, uid_t
, uid_is_valid
, UID_FMT
, "OBJECT_AUDIT_LOGINUID");
971 IOVEC_ADD_STRING_FIELD(iovec
, n
, o
->cgroup
, "OBJECT_SYSTEMD_CGROUP");
972 IOVEC_ADD_STRING_FIELD(iovec
, n
, o
->session
, "OBJECT_SYSTEMD_SESSION");
973 IOVEC_ADD_NUMERIC_FIELD(iovec
, n
, o
->owner_uid
, uid_t
, uid_is_valid
, UID_FMT
, "OBJECT_SYSTEMD_OWNER_UID");
974 IOVEC_ADD_STRING_FIELD(iovec
, n
, o
->unit
, "OBJECT_SYSTEMD_UNIT");
975 IOVEC_ADD_STRING_FIELD(iovec
, n
, o
->user_unit
, "OBJECT_SYSTEMD_USER_UNIT");
976 IOVEC_ADD_STRING_FIELD(iovec
, n
, o
->slice
, "OBJECT_SYSTEMD_SLICE");
977 IOVEC_ADD_STRING_FIELD(iovec
, n
, o
->user_slice
, "OBJECT_SYSTEMD_USER_SLICE");
979 IOVEC_ADD_ID128_FIELD(iovec
, n
, o
->invocation_id
, "OBJECT_SYSTEMD_INVOCATION_ID=");
985 sprintf(source_time
, "_SOURCE_REALTIME_TIMESTAMP=" USEC_FMT
, timeval_load(tv
));
986 iovec
[n
++] = IOVEC_MAKE_STRING(source_time
);
989 /* Note that strictly speaking storing the boot id here is
990 * redundant since the entry includes this in-line
991 * anyway. However, we need this indexed, too. */
992 if (!isempty(s
->boot_id_field
))
993 iovec
[n
++] = IOVEC_MAKE_STRING(s
->boot_id_field
);
995 if (!isempty(s
->machine_id_field
))
996 iovec
[n
++] = IOVEC_MAKE_STRING(s
->machine_id_field
);
998 if (!isempty(s
->hostname_field
))
999 iovec
[n
++] = IOVEC_MAKE_STRING(s
->hostname_field
);
1001 if (!isempty(s
->namespace_field
))
1002 iovec
[n
++] = IOVEC_MAKE_STRING(s
->namespace_field
);
1006 if (s
->split_mode
== SPLIT_UID
&& c
&& uid_is_valid(c
->uid
))
1007 /* Split up strictly by (non-root) UID */
1008 journal_uid
= c
->uid
;
1009 else if (s
->split_mode
== SPLIT_LOGIN
&& c
&& c
->uid
> 0 && uid_is_valid(c
->owner_uid
))
1010 /* Split up by login UIDs. We do this only if the
1011 * realuid is not root, in order not to accidentally
1012 * leak privileged information to the user that is
1013 * logged by a privileged process that is part of an
1014 * unprivileged session. */
1015 journal_uid
= c
->owner_uid
;
1019 write_to_journal(s
, journal_uid
, iovec
, n
, priority
);
1022 void server_driver_message(Server
*s
, pid_t object_pid
, const char *message_id
, const char *format
, ...) {
1024 struct iovec
*iovec
;
1032 m
= N_IOVEC_META_FIELDS
+ 5 + N_IOVEC_PAYLOAD_FIELDS
+ client_context_extra_fields_n_iovec(s
->my_context
) + N_IOVEC_OBJECT_FIELDS
;
1033 iovec
= newa(struct iovec
, m
);
1035 assert_cc(3 == LOG_FAC(LOG_DAEMON
));
1036 iovec
[n
++] = IOVEC_MAKE_STRING("SYSLOG_FACILITY=3");
1037 iovec
[n
++] = IOVEC_MAKE_STRING("SYSLOG_IDENTIFIER=systemd-journald");
1039 iovec
[n
++] = IOVEC_MAKE_STRING("_TRANSPORT=driver");
1040 assert_cc(6 == LOG_INFO
);
1041 iovec
[n
++] = IOVEC_MAKE_STRING("PRIORITY=6");
1044 iovec
[n
++] = IOVEC_MAKE_STRING(message_id
);
1047 va_start(ap
, format
);
1048 r
= log_format_iovec(iovec
, m
, &n
, false, 0, format
, ap
);
1049 /* Error handling below */
1053 dispatch_message_real(s
, iovec
, n
, m
, s
->my_context
, NULL
, LOG_INFO
, object_pid
);
1056 free(iovec
[k
++].iov_base
);
1059 /* We failed to format the message. Emit a warning instead. */
1062 xsprintf(buf
, "MESSAGE=Entry printing failed: %s", strerror_safe(r
));
1065 iovec
[n
++] = IOVEC_MAKE_STRING("PRIORITY=4");
1066 iovec
[n
++] = IOVEC_MAKE_STRING(buf
);
1067 dispatch_message_real(s
, iovec
, n
, m
, s
->my_context
, NULL
, LOG_INFO
, object_pid
);
1071 void server_dispatch_message(
1073 struct iovec
*iovec
, size_t n
, size_t m
,
1075 const struct timeval
*tv
,
1079 uint64_t available
= 0;
1083 assert(iovec
|| n
== 0);
1088 if (LOG_PRI(priority
) > s
->max_level_store
)
1091 /* Stop early in case the information will not be stored
1093 if (s
->storage
== STORAGE_NONE
)
1097 (void) determine_space(s
, &available
, NULL
);
1099 rl
= journal_ratelimit_test(s
->ratelimit
, c
->unit
, c
->log_ratelimit_interval
, c
->log_ratelimit_burst
, priority
& LOG_PRIMASK
, available
);
1103 /* Write a suppression message if we suppressed something */
1105 server_driver_message(s
, c
->pid
,
1106 "MESSAGE_ID=" SD_MESSAGE_JOURNAL_DROPPED_STR
,
1107 LOG_MESSAGE("Suppressed %i messages from %s", rl
- 1, c
->unit
),
1108 "N_DROPPED=%i", rl
- 1,
1112 dispatch_message_real(s
, iovec
, n
, m
, c
, tv
, priority
, object_pid
);
1115 int server_flush_to_var(Server
*s
, bool require_flag_file
) {
1116 char ts
[FORMAT_TIMESPAN_MAX
];
1117 sd_journal
*j
= NULL
;
1125 if (!IN_SET(s
->storage
, STORAGE_AUTO
, STORAGE_PERSISTENT
))
1128 if (s
->namespace) /* Flushing concept does not exist for namespace instances */
1131 if (!s
->runtime_journal
) /* Nothing to flush? */
1134 if (require_flag_file
&& !flushed_flag_is_set(s
))
1137 (void) system_journal_open(s
, true, false);
1139 if (!s
->system_journal
)
1142 log_debug("Flushing to %s...", s
->system_storage
.path
);
1144 start
= now(CLOCK_MONOTONIC
);
1146 r
= sd_journal_open(&j
, SD_JOURNAL_RUNTIME_ONLY
);
1148 return log_error_errno(r
, "Failed to read runtime journal: %m");
1150 sd_journal_set_data_threshold(j
, 0);
1152 SD_JOURNAL_FOREACH(j
) {
1156 f
= j
->current_file
;
1157 assert(f
&& f
->current_offset
> 0);
1161 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, f
->current_offset
, &o
);
1163 log_error_errno(r
, "Can't read entry: %m");
1167 r
= journal_file_copy_entry(f
, s
->system_journal
, o
, f
->current_offset
);
1171 if (!shall_try_append_again(s
->system_journal
, r
)) {
1172 log_error_errno(r
, "Can't write entry: %m");
1177 server_vacuum(s
, false);
1179 if (!s
->system_journal
) {
1180 log_notice("Didn't flush runtime journal since rotation of system journal wasn't successful.");
1185 log_debug("Retrying write.");
1186 r
= journal_file_copy_entry(f
, s
->system_journal
, o
, f
->current_offset
);
1188 log_error_errno(r
, "Can't write entry: %m");
1196 if (s
->system_journal
)
1197 journal_file_post_change(s
->system_journal
);
1199 s
->runtime_journal
= journal_file_close(s
->runtime_journal
);
1202 (void) rm_rf(s
->runtime_storage
.path
, REMOVE_ROOT
);
1204 sd_journal_close(j
);
1206 server_driver_message(s
, 0, NULL
,
1207 LOG_MESSAGE("Time spent on flushing to %s is %s for %u entries.",
1208 s
->system_storage
.path
,
1209 format_timespan(ts
, sizeof(ts
), now(CLOCK_MONOTONIC
) - start
, 0),
1213 fn
= strjoina(s
->runtime_directory
, "/flushed");
1216 log_warning_errno(k
, "Failed to touch %s, ignoring: %m", fn
);
1218 server_refresh_idle_timer(s
);
1222 static int server_relinquish_var(Server
*s
) {
1226 if (s
->storage
== STORAGE_NONE
)
1229 if (s
->namespace) /* Concept does not exist for namespaced instances */
1232 if (s
->runtime_journal
&& !s
->system_journal
)
1235 log_debug("Relinquishing %s...", s
->system_storage
.path
);
1237 (void) system_journal_open(s
, false, true);
1239 s
->system_journal
= journal_file_close(s
->system_journal
);
1240 ordered_hashmap_clear_with_destructor(s
->user_journals
, journal_file_close
);
1241 set_clear_with_destructor(s
->deferred_closes
, journal_file_close
);
1243 fn
= strjoina(s
->runtime_directory
, "/flushed");
1244 if (unlink(fn
) < 0 && errno
!= ENOENT
)
1245 log_warning_errno(errno
, "Failed to unlink %s, ignoring: %m", fn
);
1247 server_refresh_idle_timer(s
);
1251 int server_process_datagram(
1252 sd_event_source
*es
,
1257 Server
*s
= userdata
;
1258 struct ucred
*ucred
= NULL
;
1259 struct timeval
*tv
= NULL
;
1260 struct cmsghdr
*cmsg
;
1262 size_t label_len
= 0, m
;
1265 int *fds
= NULL
, v
= 0;
1268 /* We use NAME_MAX space for the SELinux label here. The kernel currently enforces no limit, but
1269 * according to suggestions from the SELinux people this will change and it will probably be
1270 * identical to NAME_MAX. For now we use that, but this should be updated one day when the final
1271 * limit is known. */
1272 CMSG_BUFFER_TYPE(CMSG_SPACE(sizeof(struct ucred
)) +
1273 CMSG_SPACE(sizeof(struct timeval
)) +
1274 CMSG_SPACE(sizeof(int)) + /* fd */
1275 CMSG_SPACE(NAME_MAX
) /* selinux label */) control
;
1277 union sockaddr_union sa
= {};
1279 struct msghdr msghdr
= {
1282 .msg_control
= &control
,
1283 .msg_controllen
= sizeof(control
),
1285 .msg_namelen
= sizeof(sa
),
1289 assert(fd
== s
->native_fd
|| fd
== s
->syslog_fd
|| fd
== s
->audit_fd
);
1291 if (revents
!= EPOLLIN
)
1292 return log_error_errno(SYNTHETIC_ERRNO(EIO
),
1293 "Got invalid event from epoll for datagram fd: %" PRIx32
,
1296 /* Try to get the right size, if we can. (Not all sockets support SIOCINQ, hence we just try, but don't rely on
1298 (void) ioctl(fd
, SIOCINQ
, &v
);
1300 /* Fix it up, if it is too small. We use the same fixed value as auditd here. Awful! */
1301 m
= PAGE_ALIGN(MAX3((size_t) v
+ 1,
1303 ALIGN(sizeof(struct nlmsghdr
)) + ALIGN((size_t) MAX_AUDIT_MESSAGE_LENGTH
)) + 1);
1305 if (!GREEDY_REALLOC(s
->buffer
, s
->buffer_size
, m
))
1308 iovec
= IOVEC_MAKE(s
->buffer
, s
->buffer_size
- 1); /* Leave room for trailing NUL we add later */
1310 n
= recvmsg_safe(fd
, &msghdr
, MSG_DONTWAIT
|MSG_CMSG_CLOEXEC
);
1311 if (IN_SET(n
, -EINTR
, -EAGAIN
))
1314 log_warning("Got message with truncated control data (too many fds sent?), ignoring.");
1318 return log_error_errno(n
, "recvmsg() failed: %m");
1320 CMSG_FOREACH(cmsg
, &msghdr
)
1321 if (cmsg
->cmsg_level
== SOL_SOCKET
&&
1322 cmsg
->cmsg_type
== SCM_CREDENTIALS
&&
1323 cmsg
->cmsg_len
== CMSG_LEN(sizeof(struct ucred
))) {
1325 ucred
= (struct ucred
*) CMSG_DATA(cmsg
);
1326 } else if (cmsg
->cmsg_level
== SOL_SOCKET
&&
1327 cmsg
->cmsg_type
== SCM_SECURITY
) {
1329 label
= (char*) CMSG_DATA(cmsg
);
1330 label_len
= cmsg
->cmsg_len
- CMSG_LEN(0);
1331 } else if (cmsg
->cmsg_level
== SOL_SOCKET
&&
1332 cmsg
->cmsg_type
== SO_TIMESTAMP
&&
1333 cmsg
->cmsg_len
== CMSG_LEN(sizeof(struct timeval
))) {
1335 tv
= (struct timeval
*) CMSG_DATA(cmsg
);
1336 } else if (cmsg
->cmsg_level
== SOL_SOCKET
&&
1337 cmsg
->cmsg_type
== SCM_RIGHTS
) {
1339 fds
= (int*) CMSG_DATA(cmsg
);
1340 n_fds
= (cmsg
->cmsg_len
- CMSG_LEN(0)) / sizeof(int);
1343 /* And a trailing NUL, just in case */
1346 if (fd
== s
->syslog_fd
) {
1347 if (n
> 0 && n_fds
== 0)
1348 server_process_syslog_message(s
, s
->buffer
, n
, ucred
, tv
, label
, label_len
);
1350 log_warning("Got file descriptors via syslog socket. Ignoring.");
1352 } else if (fd
== s
->native_fd
) {
1353 if (n
> 0 && n_fds
== 0)
1354 server_process_native_message(s
, s
->buffer
, n
, ucred
, tv
, label
, label_len
);
1355 else if (n
== 0 && n_fds
== 1)
1356 server_process_native_file(s
, fds
[0], ucred
, tv
, label
, label_len
);
1358 log_warning("Got too many file descriptors via native socket. Ignoring.");
1361 assert(fd
== s
->audit_fd
);
1363 if (n
> 0 && n_fds
== 0)
1364 server_process_audit_message(s
, s
->buffer
, n
, ucred
, &sa
, msghdr
.msg_namelen
);
1366 log_warning("Got file descriptors via audit socket. Ignoring.");
1369 close_many(fds
, n_fds
);
1371 server_refresh_idle_timer(s
);
1375 static void server_full_flush(Server
*s
) {
1378 (void) server_flush_to_var(s
, false);
1380 server_vacuum(s
, false);
1382 server_space_usage_message(s
, NULL
);
1384 server_refresh_idle_timer(s
);
1387 static int dispatch_sigusr1(sd_event_source
*es
, const struct signalfd_siginfo
*si
, void *userdata
) {
1388 Server
*s
= userdata
;
1393 log_error("Received SIGUSR1 signal from PID " PID_FMT
", but flushing runtime journals not supported for namespaced instances.", si
->ssi_pid
);
1397 log_info("Received SIGUSR1 signal from PID " PID_FMT
", as request to flush runtime journal.", si
->ssi_pid
);
1398 server_full_flush(s
);
1403 static void server_full_rotate(Server
*s
) {
1410 server_vacuum(s
, true);
1412 if (s
->system_journal
)
1413 patch_min_use(&s
->system_storage
);
1414 if (s
->runtime_journal
)
1415 patch_min_use(&s
->runtime_storage
);
1417 /* Let clients know when the most recent rotation happened. */
1418 fn
= strjoina(s
->runtime_directory
, "/rotated");
1419 r
= write_timestamp_file_atomic(fn
, now(CLOCK_MONOTONIC
));
1421 log_warning_errno(r
, "Failed to write %s, ignoring: %m", fn
);
1424 static int dispatch_sigusr2(sd_event_source
*es
, const struct signalfd_siginfo
*si
, void *userdata
) {
1425 Server
*s
= userdata
;
1429 log_info("Received SIGUSR2 signal from PID " PID_FMT
", as request to rotate journal.", si
->ssi_pid
);
1430 server_full_rotate(s
);
1435 static int dispatch_sigterm(sd_event_source
*es
, const struct signalfd_siginfo
*si
, void *userdata
) {
1436 Server
*s
= userdata
;
1440 log_received_signal(LOG_INFO
, si
);
1442 sd_event_exit(s
->event
, 0);
1446 static void server_full_sync(Server
*s
) {
1454 /* Let clients know when the most recent sync happened. */
1455 fn
= strjoina(s
->runtime_directory
, "/synced");
1456 r
= write_timestamp_file_atomic(fn
, now(CLOCK_MONOTONIC
));
1458 log_warning_errno(r
, "Failed to write %s, ignoring: %m", fn
);
1463 static int dispatch_sigrtmin1(sd_event_source
*es
, const struct signalfd_siginfo
*si
, void *userdata
) {
1464 Server
*s
= userdata
;
1468 log_debug("Received SIGRTMIN1 signal from PID " PID_FMT
", as request to sync.", si
->ssi_pid
);
1469 server_full_sync(s
);
1474 static int setup_signals(Server
*s
) {
1479 assert_se(sigprocmask_many(SIG_SETMASK
, NULL
, SIGINT
, SIGTERM
, SIGUSR1
, SIGUSR2
, SIGRTMIN
+1, -1) >= 0);
1481 r
= sd_event_add_signal(s
->event
, &s
->sigusr1_event_source
, SIGUSR1
, dispatch_sigusr1
, s
);
1485 r
= sd_event_add_signal(s
->event
, &s
->sigusr2_event_source
, SIGUSR2
, dispatch_sigusr2
, s
);
1489 r
= sd_event_add_signal(s
->event
, &s
->sigterm_event_source
, SIGTERM
, dispatch_sigterm
, s
);
1493 /* Let's process SIGTERM late, so that we flush all queued messages to disk before we exit */
1494 r
= sd_event_source_set_priority(s
->sigterm_event_source
, SD_EVENT_PRIORITY_NORMAL
+20);
1498 /* When journald is invoked on the terminal (when debugging), it's useful if C-c is handled
1499 * equivalent to SIGTERM. */
1500 r
= sd_event_add_signal(s
->event
, &s
->sigint_event_source
, SIGINT
, dispatch_sigterm
, s
);
1504 r
= sd_event_source_set_priority(s
->sigint_event_source
, SD_EVENT_PRIORITY_NORMAL
+20);
1508 /* SIGRTMIN+1 causes an immediate sync. We process this very late, so that everything else queued at
1509 * this point is really written to disk. Clients can watch /run/systemd/journal/synced with inotify
1510 * until its mtime changes to see when a sync happened. */
1511 r
= sd_event_add_signal(s
->event
, &s
->sigrtmin1_event_source
, SIGRTMIN
+1, dispatch_sigrtmin1
, s
);
1515 r
= sd_event_source_set_priority(s
->sigrtmin1_event_source
, SD_EVENT_PRIORITY_NORMAL
+15);
1522 static int parse_proc_cmdline_item(const char *key
, const char *value
, void *data
) {
1528 if (proc_cmdline_key_streq(key
, "systemd.journald.forward_to_syslog")) {
1530 r
= value
? parse_boolean(value
) : true;
1532 log_warning("Failed to parse forward to syslog switch \"%s\". Ignoring.", value
);
1534 s
->forward_to_syslog
= r
;
1536 } else if (proc_cmdline_key_streq(key
, "systemd.journald.forward_to_kmsg")) {
1538 r
= value
? parse_boolean(value
) : true;
1540 log_warning("Failed to parse forward to kmsg switch \"%s\". Ignoring.", value
);
1542 s
->forward_to_kmsg
= r
;
1544 } else if (proc_cmdline_key_streq(key
, "systemd.journald.forward_to_console")) {
1546 r
= value
? parse_boolean(value
) : true;
1548 log_warning("Failed to parse forward to console switch \"%s\". Ignoring.", value
);
1550 s
->forward_to_console
= r
;
1552 } else if (proc_cmdline_key_streq(key
, "systemd.journald.forward_to_wall")) {
1554 r
= value
? parse_boolean(value
) : true;
1556 log_warning("Failed to parse forward to wall switch \"%s\". Ignoring.", value
);
1558 s
->forward_to_wall
= r
;
1560 } else if (proc_cmdline_key_streq(key
, "systemd.journald.max_level_console")) {
1562 if (proc_cmdline_value_missing(key
, value
))
1565 r
= log_level_from_string(value
);
1567 log_warning("Failed to parse max level console value \"%s\". Ignoring.", value
);
1569 s
->max_level_console
= r
;
1571 } else if (proc_cmdline_key_streq(key
, "systemd.journald.max_level_store")) {
1573 if (proc_cmdline_value_missing(key
, value
))
1576 r
= log_level_from_string(value
);
1578 log_warning("Failed to parse max level store value \"%s\". Ignoring.", value
);
1580 s
->max_level_store
= r
;
1582 } else if (proc_cmdline_key_streq(key
, "systemd.journald.max_level_syslog")) {
1584 if (proc_cmdline_value_missing(key
, value
))
1587 r
= log_level_from_string(value
);
1589 log_warning("Failed to parse max level syslog value \"%s\". Ignoring.", value
);
1591 s
->max_level_syslog
= r
;
1593 } else if (proc_cmdline_key_streq(key
, "systemd.journald.max_level_kmsg")) {
1595 if (proc_cmdline_value_missing(key
, value
))
1598 r
= log_level_from_string(value
);
1600 log_warning("Failed to parse max level kmsg value \"%s\". Ignoring.", value
);
1602 s
->max_level_kmsg
= r
;
1604 } else if (proc_cmdline_key_streq(key
, "systemd.journald.max_level_wall")) {
1606 if (proc_cmdline_value_missing(key
, value
))
1609 r
= log_level_from_string(value
);
1611 log_warning("Failed to parse max level wall value \"%s\". Ignoring.", value
);
1613 s
->max_level_wall
= r
;
1615 } else if (startswith(key
, "systemd.journald"))
1616 log_warning("Unknown journald kernel command line option \"%s\". Ignoring.", key
);
1618 /* do not warn about state here, since probably systemd already did */
1622 static int server_parse_config_file(Server
*s
) {
1628 const char *namespaced
;
1630 /* If we are running in namespace mode, load the namespace specific configuration file, and nothing else */
1631 namespaced
= strjoina(PKGSYSCONFDIR
"/journald@", s
->namespace, ".conf");
1633 r
= config_parse(NULL
,
1636 config_item_perf_lookup
, journald_gperf_lookup
,
1637 CONFIG_PARSE_WARN
, s
,
1645 return config_parse_many_nulstr(
1646 PKGSYSCONFDIR
"/journald.conf",
1647 CONF_PATHS_NULSTR("systemd/journald.conf.d"),
1649 config_item_perf_lookup
, journald_gperf_lookup
,
1650 CONFIG_PARSE_WARN
, s
, NULL
);
1653 static int server_dispatch_sync(sd_event_source
*es
, usec_t t
, void *userdata
) {
1654 Server
*s
= userdata
;
1662 int server_schedule_sync(Server
*s
, int priority
) {
1667 if (priority
<= LOG_CRIT
) {
1668 /* Immediately sync to disk when this is of priority CRIT, ALERT, EMERG */
1673 if (s
->sync_scheduled
)
1676 if (s
->sync_interval_usec
> 0) {
1678 if (!s
->sync_event_source
) {
1679 r
= sd_event_add_time_relative(
1681 &s
->sync_event_source
,
1683 s
->sync_interval_usec
, 0,
1684 server_dispatch_sync
, s
);
1688 r
= sd_event_source_set_priority(s
->sync_event_source
, SD_EVENT_PRIORITY_IMPORTANT
);
1690 r
= sd_event_source_set_time_relative(s
->sync_event_source
, s
->sync_interval_usec
);
1694 r
= sd_event_source_set_enabled(s
->sync_event_source
, SD_EVENT_ONESHOT
);
1699 s
->sync_scheduled
= true;
1705 static int dispatch_hostname_change(sd_event_source
*es
, int fd
, uint32_t revents
, void *userdata
) {
1706 Server
*s
= userdata
;
1710 server_cache_hostname(s
);
1714 static int server_open_hostname(Server
*s
) {
1719 s
->hostname_fd
= open("/proc/sys/kernel/hostname",
1720 O_RDONLY
|O_CLOEXEC
|O_NONBLOCK
|O_NOCTTY
);
1721 if (s
->hostname_fd
< 0)
1722 return log_error_errno(errno
, "Failed to open /proc/sys/kernel/hostname: %m");
1724 r
= sd_event_add_io(s
->event
, &s
->hostname_event_source
, s
->hostname_fd
, 0, dispatch_hostname_change
, s
);
1726 /* kernels prior to 3.2 don't support polling this file. Ignore
1729 log_warning_errno(r
, "Failed to register hostname fd in event loop, ignoring: %m");
1730 s
->hostname_fd
= safe_close(s
->hostname_fd
);
1734 return log_error_errno(r
, "Failed to register hostname fd in event loop: %m");
1737 r
= sd_event_source_set_priority(s
->hostname_event_source
, SD_EVENT_PRIORITY_IMPORTANT
-10);
1739 return log_error_errno(r
, "Failed to adjust priority of hostname event source: %m");
1744 static int dispatch_notify_event(sd_event_source
*es
, int fd
, uint32_t revents
, void *userdata
) {
1745 Server
*s
= userdata
;
1749 assert(s
->notify_event_source
== es
);
1750 assert(s
->notify_fd
== fd
);
1752 /* The $NOTIFY_SOCKET is writable again, now send exactly one
1753 * message on it. Either it's the watchdog event, the initial
1754 * READY=1 event or an stdout stream event. If there's nothing
1755 * to write anymore, turn our event source off. The next time
1756 * there's something to send it will be turned on again. */
1758 if (!s
->sent_notify_ready
) {
1759 static const char p
[] =
1761 "STATUS=Processing requests...";
1764 l
= send(s
->notify_fd
, p
, strlen(p
), MSG_DONTWAIT
);
1766 if (errno
== EAGAIN
)
1769 return log_error_errno(errno
, "Failed to send READY=1 notification message: %m");
1772 s
->sent_notify_ready
= true;
1773 log_debug("Sent READY=1 notification.");
1775 } else if (s
->send_watchdog
) {
1777 static const char p
[] =
1782 l
= send(s
->notify_fd
, p
, strlen(p
), MSG_DONTWAIT
);
1784 if (errno
== EAGAIN
)
1787 return log_error_errno(errno
, "Failed to send WATCHDOG=1 notification message: %m");
1790 s
->send_watchdog
= false;
1791 log_debug("Sent WATCHDOG=1 notification.");
1793 } else if (s
->stdout_streams_notify_queue
)
1794 /* Dispatch one stream notification event */
1795 stdout_stream_send_notify(s
->stdout_streams_notify_queue
);
1797 /* Leave us enabled if there's still more to do. */
1798 if (s
->send_watchdog
|| s
->stdout_streams_notify_queue
)
1801 /* There was nothing to do anymore, let's turn ourselves off. */
1802 r
= sd_event_source_set_enabled(es
, SD_EVENT_OFF
);
1804 return log_error_errno(r
, "Failed to turn off notify event source: %m");
1809 static int dispatch_watchdog(sd_event_source
*es
, uint64_t usec
, void *userdata
) {
1810 Server
*s
= userdata
;
1815 s
->send_watchdog
= true;
1817 r
= sd_event_source_set_enabled(s
->notify_event_source
, SD_EVENT_ON
);
1819 log_warning_errno(r
, "Failed to turn on notify event source: %m");
1821 r
= sd_event_source_set_time(s
->watchdog_event_source
, usec
+ s
->watchdog_usec
/ 2);
1823 return log_error_errno(r
, "Failed to restart watchdog event source: %m");
1825 r
= sd_event_source_set_enabled(s
->watchdog_event_source
, SD_EVENT_ON
);
1827 return log_error_errno(r
, "Failed to enable watchdog event source: %m");
1832 static int server_connect_notify(Server
*s
) {
1833 union sockaddr_union sa
;
1839 assert(s
->notify_fd
< 0);
1840 assert(!s
->notify_event_source
);
1843 * So here's the problem: we'd like to send notification messages to PID 1, but we cannot do that via
1844 * sd_notify(), since that's synchronous, and we might end up blocking on it. Specifically: given
1845 * that PID 1 might block on dbus-daemon during IPC, and dbus-daemon is logging to us, and might
1846 * hence block on us, we might end up in a deadlock if we block on sending PID 1 notification
1847 * messages — by generating a full blocking circle. To avoid this, let's create a non-blocking
1848 * socket, and connect it to the notification socket, and then wait for POLLOUT before we send
1849 * anything. This should efficiently avoid any deadlocks, as we'll never block on PID 1, hence PID 1
1850 * can safely block on dbus-daemon which can safely block on us again.
1852 * Don't think that this issue is real? It is, see: https://github.com/systemd/systemd/issues/1505
1855 e
= getenv("NOTIFY_SOCKET");
1859 r
= sockaddr_un_set_path(&sa
.un
, e
);
1861 return log_error_errno(r
, "NOTIFY_SOCKET set to invalid value '%s': %m", e
);
1864 s
->notify_fd
= socket(AF_UNIX
, SOCK_DGRAM
|SOCK_CLOEXEC
|SOCK_NONBLOCK
, 0);
1865 if (s
->notify_fd
< 0)
1866 return log_error_errno(errno
, "Failed to create notify socket: %m");
1868 (void) fd_inc_sndbuf(s
->notify_fd
, NOTIFY_SNDBUF_SIZE
);
1870 r
= connect(s
->notify_fd
, &sa
.sa
, sa_len
);
1872 return log_error_errno(errno
, "Failed to connect to notify socket: %m");
1874 r
= sd_event_add_io(s
->event
, &s
->notify_event_source
, s
->notify_fd
, EPOLLOUT
, dispatch_notify_event
, s
);
1876 return log_error_errno(r
, "Failed to watch notification socket: %m");
1878 if (sd_watchdog_enabled(false, &s
->watchdog_usec
) > 0) {
1879 s
->send_watchdog
= true;
1881 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
);
1883 return log_error_errno(r
, "Failed to add watchdog time event: %m");
1886 /* This should fire pretty soon, which we'll use to send the READY=1 event. */
1891 static int synchronize_second_half(sd_event_source
*event_source
, void *userdata
) {
1892 Varlink
*link
= userdata
;
1897 assert_se(s
= varlink_get_userdata(link
));
1899 /* This is the "second half" of the Synchronize() varlink method. This function is called as deferred
1900 * event source at a low priority to ensure the synchronization completes after all queued log
1901 * messages are processed. */
1902 server_full_sync(s
);
1904 /* Let's get rid of the event source now, by marking it as non-floating again. It then has no ref
1905 * anymore and is immediately destroyed after we return from this function, i.e. from this event
1906 * source handler at the end. */
1907 r
= sd_event_source_set_floating(event_source
, false);
1909 return log_error_errno(r
, "Failed to mark event source as non-floating: %m");
1911 return varlink_reply(link
, NULL
);
1914 static void synchronize_destroy(void *userdata
) {
1915 varlink_unref(userdata
);
1918 static int vl_method_synchronize(Varlink
*link
, JsonVariant
*parameters
, VarlinkMethodFlags flags
, void *userdata
) {
1919 _cleanup_(sd_event_source_unrefp
) sd_event_source
*event_source
= NULL
;
1920 Server
*s
= userdata
;
1926 if (json_variant_elements(parameters
) > 0)
1927 return varlink_error_invalid_parameter(link
, parameters
);
1929 log_info("Received client request to rotate journal.");
1931 /* We don't do the main work now, but instead enqueue a deferred event loop job which will do
1932 * it. That job is scheduled at low priority, so that we return from this method call only after all
1933 * queued but not processed log messages are written to disk, so that this method call returning can
1934 * be used as nice synchronization point. */
1935 r
= sd_event_add_defer(s
->event
, &event_source
, synchronize_second_half
, link
);
1937 return log_error_errno(r
, "Failed to allocate defer event source: %m");
1939 r
= sd_event_source_set_destroy_callback(event_source
, synchronize_destroy
);
1941 return log_error_errno(r
, "Failed to set event source destroy callback: %m");
1943 varlink_ref(link
); /* The varlink object is now left to the destroy callback to unref */
1945 r
= sd_event_source_set_priority(event_source
, SD_EVENT_PRIORITY_NORMAL
+15);
1947 return log_error_errno(r
, "Failed to set defer event source priority: %m");
1949 /* Give up ownership of this event source. It will now be destroyed along with event loop itself,
1950 * unless it destroys itself earlier. */
1951 r
= sd_event_source_set_floating(event_source
, true);
1953 return log_error_errno(r
, "Failed to mark event source as floating: %m");
1955 (void) sd_event_source_set_description(event_source
, "deferred-sync");
1960 static int vl_method_rotate(Varlink
*link
, JsonVariant
*parameters
, VarlinkMethodFlags flags
, void *userdata
) {
1961 Server
*s
= userdata
;
1966 if (json_variant_elements(parameters
) > 0)
1967 return varlink_error_invalid_parameter(link
, parameters
);
1969 log_info("Received client request to rotate journal.");
1970 server_full_rotate(s
);
1972 return varlink_reply(link
, NULL
);
1975 static int vl_method_flush_to_var(Varlink
*link
, JsonVariant
*parameters
, VarlinkMethodFlags flags
, void *userdata
) {
1976 Server
*s
= userdata
;
1981 if (json_variant_elements(parameters
) > 0)
1982 return varlink_error_invalid_parameter(link
, parameters
);
1984 return varlink_error(link
, "io.systemd.Journal.NotSupportedByNamespaces", NULL
);
1986 log_info("Received client request to flush runtime journal.");
1987 server_full_flush(s
);
1989 return varlink_reply(link
, NULL
);
1992 static int vl_method_relinquish_var(Varlink
*link
, JsonVariant
*parameters
, VarlinkMethodFlags flags
, void *userdata
) {
1993 Server
*s
= userdata
;
1998 if (json_variant_elements(parameters
) > 0)
1999 return varlink_error_invalid_parameter(link
, parameters
);
2001 return varlink_error(link
, "io.systemd.Journal.NotSupportedByNamespaces", NULL
);
2003 log_info("Received client request to relinquish %s access.", s
->system_storage
.path
);
2004 server_relinquish_var(s
);
2006 return varlink_reply(link
, NULL
);
2009 static int vl_connect(VarlinkServer
*server
, Varlink
*link
, void *userdata
) {
2010 Server
*s
= userdata
;
2016 (void) server_start_or_stop_idle_timer(s
); /* maybe we are no longer idle */
2021 static void vl_disconnect(VarlinkServer
*server
, Varlink
*link
, void *userdata
) {
2022 Server
*s
= userdata
;
2028 (void) server_start_or_stop_idle_timer(s
); /* maybe we are idle now */
2031 static int server_open_varlink(Server
*s
, const char *socket
, int fd
) {
2036 r
= varlink_server_new(&s
->varlink_server
, VARLINK_SERVER_ROOT_ONLY
);
2040 varlink_server_set_userdata(s
->varlink_server
, s
);
2042 r
= varlink_server_bind_method_many(
2044 "io.systemd.Journal.Synchronize", vl_method_synchronize
,
2045 "io.systemd.Journal.Rotate", vl_method_rotate
,
2046 "io.systemd.Journal.FlushToVar", vl_method_flush_to_var
,
2047 "io.systemd.Journal.RelinquishVar", vl_method_relinquish_var
);
2051 r
= varlink_server_bind_connect(s
->varlink_server
, vl_connect
);
2055 r
= varlink_server_bind_disconnect(s
->varlink_server
, vl_disconnect
);
2060 r
= varlink_server_listen_address(s
->varlink_server
, socket
, 0600);
2062 r
= varlink_server_listen_fd(s
->varlink_server
, fd
);
2066 r
= varlink_server_attach_event(s
->varlink_server
, s
->event
, SD_EVENT_PRIORITY_NORMAL
);
2073 static bool server_is_idle(Server
*s
) {
2076 /* The server for the main namespace is never idle */
2080 /* If a retention maximum is set larger than the idle time we need to be running to enforce it, hence
2081 * turn off the idle logic. */
2082 if (s
->max_retention_usec
> IDLE_TIMEOUT_USEC
)
2085 /* We aren't idle if we have a varlink client */
2086 if (varlink_server_current_connections(s
->varlink_server
) > 0)
2089 /* If we have stdout streams we aren't idle */
2090 if (s
->n_stdout_streams
> 0)
2096 static int server_idle_handler(sd_event_source
*source
, uint64_t usec
, void *userdata
) {
2097 Server
*s
= userdata
;
2102 log_debug("Server is idle, exiting.");
2103 sd_event_exit(s
->event
, 0);
2107 int server_start_or_stop_idle_timer(Server
*s
) {
2108 _cleanup_(sd_event_source_unrefp
) sd_event_source
*source
= NULL
;
2113 if (!server_is_idle(s
)) {
2114 s
->idle_event_source
= sd_event_source_disable_unref(s
->idle_event_source
);
2118 if (s
->idle_event_source
)
2121 r
= sd_event_add_time_relative(s
->event
, &source
, CLOCK_MONOTONIC
, IDLE_TIMEOUT_USEC
, 0, server_idle_handler
, s
);
2123 return log_error_errno(r
, "Failed to allocate idle timer: %m");
2125 r
= sd_event_source_set_priority(source
, SD_EVENT_PRIORITY_IDLE
);
2127 return log_error_errno(r
, "Failed to set idle timer priority: %m");
2129 (void) sd_event_source_set_description(source
, "idle-timer");
2131 s
->idle_event_source
= TAKE_PTR(source
);
2135 int server_refresh_idle_timer(Server
*s
) {
2140 if (!s
->idle_event_source
)
2143 r
= sd_event_source_set_time_relative(s
->idle_event_source
, IDLE_TIMEOUT_USEC
);
2145 return log_error_errno(r
, "Failed to refresh idle timer: %m");
2150 static int set_namespace(Server
*s
, const char *namespace) {
2156 if (!log_namespace_name_valid(namespace))
2157 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Specified namespace name not valid, refusing: %s", namespace);
2159 s
->namespace = strdup(namespace);
2163 s
->namespace_field
= strjoin("_NAMESPACE=", namespace);
2164 if (!s
->namespace_field
)
2170 int server_init(Server
*s
, const char *namespace) {
2171 const char *native_socket
, *syslog_socket
, *stdout_socket
, *varlink_socket
, *e
;
2172 _cleanup_fdset_free_ FDSet
*fds
= NULL
;
2173 int n
, r
, fd
, varlink_fd
= -1;
2187 .compress
.enabled
= true,
2188 .compress
.threshold_bytes
= (uint64_t) -1,
2193 .watchdog_usec
= USEC_INFINITY
,
2195 .sync_interval_usec
= DEFAULT_SYNC_INTERVAL_USEC
,
2196 .sync_scheduled
= false,
2198 .ratelimit_interval
= DEFAULT_RATE_LIMIT_INTERVAL
,
2199 .ratelimit_burst
= DEFAULT_RATE_LIMIT_BURST
,
2201 .forward_to_wall
= true,
2203 .max_file_usec
= DEFAULT_MAX_FILE_USEC
,
2205 .max_level_store
= LOG_DEBUG
,
2206 .max_level_syslog
= LOG_DEBUG
,
2207 .max_level_kmsg
= LOG_NOTICE
,
2208 .max_level_console
= LOG_INFO
,
2209 .max_level_wall
= LOG_EMERG
,
2211 .line_max
= DEFAULT_LINE_MAX
,
2213 .runtime_storage
.name
= "Runtime Journal",
2214 .system_storage
.name
= "System Journal",
2217 r
= set_namespace(s
, namespace);
2221 /* By default, only read from /dev/kmsg if are the main namespace */
2222 s
->read_kmsg
= !s
->namespace;
2223 s
->storage
= s
->namespace ? STORAGE_PERSISTENT
: STORAGE_AUTO
;
2225 journal_reset_metrics(&s
->system_storage
.metrics
);
2226 journal_reset_metrics(&s
->runtime_storage
.metrics
);
2228 server_parse_config_file(s
);
2230 if (!s
->namespace) {
2231 /* Parse kernel command line, but only if we are not a namespace instance */
2232 r
= proc_cmdline_parse(parse_proc_cmdline_item
, s
, PROC_CMDLINE_STRIP_RD_PREFIX
);
2234 log_warning_errno(r
, "Failed to parse kernel command line, ignoring: %m");
2237 if (!!s
->ratelimit_interval
!= !!s
->ratelimit_burst
) { /* One set to 0 and the other not? */
2238 log_debug("Setting both rate limit interval and burst from "USEC_FMT
",%u to 0,0",
2239 s
->ratelimit_interval
, s
->ratelimit_burst
);
2240 s
->ratelimit_interval
= s
->ratelimit_burst
= 0;
2243 e
= getenv("RUNTIME_DIRECTORY");
2245 s
->runtime_directory
= strdup(e
);
2246 else if (s
->namespace)
2247 s
->runtime_directory
= strjoin("/run/systemd/journal.", s
->namespace);
2249 s
->runtime_directory
= strdup("/run/systemd/journal");
2250 if (!s
->runtime_directory
)
2253 (void) mkdir_p(s
->runtime_directory
, 0755);
2255 s
->user_journals
= ordered_hashmap_new(NULL
);
2256 if (!s
->user_journals
)
2259 s
->mmap
= mmap_cache_new();
2263 s
->deferred_closes
= set_new(NULL
);
2264 if (!s
->deferred_closes
)
2267 r
= sd_event_default(&s
->event
);
2269 return log_error_errno(r
, "Failed to create event loop: %m");
2271 n
= sd_listen_fds(true);
2273 return log_error_errno(n
, "Failed to read listening file descriptors from environment: %m");
2275 native_socket
= strjoina(s
->runtime_directory
, "/socket");
2276 stdout_socket
= strjoina(s
->runtime_directory
, "/stdout");
2277 syslog_socket
= strjoina(s
->runtime_directory
, "/dev-log");
2278 varlink_socket
= strjoina(s
->runtime_directory
, "/io.systemd.journal");
2280 for (fd
= SD_LISTEN_FDS_START
; fd
< SD_LISTEN_FDS_START
+ n
; fd
++) {
2282 if (sd_is_socket_unix(fd
, SOCK_DGRAM
, -1, native_socket
, 0) > 0) {
2284 if (s
->native_fd
>= 0)
2285 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
2286 "Too many native sockets passed.");
2290 } else if (sd_is_socket_unix(fd
, SOCK_STREAM
, 1, stdout_socket
, 0) > 0) {
2292 if (s
->stdout_fd
>= 0)
2293 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
2294 "Too many stdout sockets passed.");
2298 } else if (sd_is_socket_unix(fd
, SOCK_DGRAM
, -1, syslog_socket
, 0) > 0) {
2300 if (s
->syslog_fd
>= 0)
2301 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
2302 "Too many /dev/log sockets passed.");
2306 } else if (sd_is_socket_unix(fd
, SOCK_STREAM
, 1, varlink_socket
, 0) > 0) {
2308 if (varlink_fd
>= 0)
2309 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
2310 "Too many varlink sockets passed.");
2313 } else if (sd_is_socket(fd
, AF_NETLINK
, SOCK_RAW
, -1) > 0) {
2315 if (s
->audit_fd
>= 0)
2316 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
2317 "Too many audit sockets passed.");
2329 r
= fdset_put(fds
, fd
);
2335 /* Try to restore streams, but don't bother if this fails */
2336 (void) server_restore_streams(s
, fds
);
2338 if (fdset_size(fds
) > 0) {
2339 log_warning("%u unknown file descriptors passed, closing.", fdset_size(fds
));
2340 fds
= fdset_free(fds
);
2343 no_sockets
= s
->native_fd
< 0 && s
->stdout_fd
< 0 && s
->syslog_fd
< 0 && s
->audit_fd
< 0 && varlink_fd
< 0;
2345 /* always open stdout, syslog, native, and kmsg sockets */
2347 /* systemd-journald.socket: /run/systemd/journal/stdout */
2348 r
= server_open_stdout_socket(s
, stdout_socket
);
2352 /* systemd-journald-dev-log.socket: /run/systemd/journal/dev-log */
2353 r
= server_open_syslog_socket(s
, syslog_socket
);
2357 /* systemd-journald.socket: /run/systemd/journal/socket */
2358 r
= server_open_native_socket(s
, native_socket
);
2363 r
= server_open_dev_kmsg(s
);
2367 /* Unless we got *some* sockets and not audit, open audit socket */
2368 if (s
->audit_fd
>= 0 || no_sockets
) {
2369 r
= server_open_audit(s
);
2374 r
= server_open_varlink(s
, varlink_socket
, varlink_fd
);
2378 r
= server_open_kernel_seqnum(s
);
2382 r
= server_open_hostname(s
);
2386 r
= setup_signals(s
);
2390 s
->ratelimit
= journal_ratelimit_new();
2394 r
= cg_get_root_path(&s
->cgroup_root
);
2396 return log_error_errno(r
, "Failed to acquire cgroup root path: %m");
2398 server_cache_hostname(s
);
2399 server_cache_boot_id(s
);
2400 server_cache_machine_id(s
);
2403 s
->runtime_storage
.path
= strjoin("/run/log/journal/", SERVER_MACHINE_ID(s
), ".", s
->namespace);
2405 s
->runtime_storage
.path
= strjoin("/run/log/journal/", SERVER_MACHINE_ID(s
));
2406 if (!s
->runtime_storage
.path
)
2409 e
= getenv("LOGS_DIRECTORY");
2411 s
->system_storage
.path
= strdup(e
);
2412 else if (s
->namespace)
2413 s
->system_storage
.path
= strjoin("/var/log/journal/", SERVER_MACHINE_ID(s
), ".", s
->namespace);
2415 s
->system_storage
.path
= strjoin("/var/log/journal/", SERVER_MACHINE_ID(s
));
2416 if (!s
->system_storage
.path
)
2419 (void) server_connect_notify(s
);
2421 (void) client_context_acquire_default(s
);
2423 r
= system_journal_open(s
, false, false);
2427 server_start_or_stop_idle_timer(s
);
2431 void server_maybe_append_tags(Server
*s
) {
2436 n
= now(CLOCK_REALTIME
);
2438 if (s
->system_journal
)
2439 journal_file_maybe_append_tag(s
->system_journal
, n
);
2441 ORDERED_HASHMAP_FOREACH(f
, s
->user_journals
)
2442 journal_file_maybe_append_tag(f
, n
);
2446 void server_done(Server
*s
) {
2450 free(s
->namespace_field
);
2452 set_free_with_destructor(s
->deferred_closes
, journal_file_close
);
2454 while (s
->stdout_streams
)
2455 stdout_stream_free(s
->stdout_streams
);
2457 client_context_flush_all(s
);
2459 (void) journal_file_close(s
->system_journal
);
2460 (void) journal_file_close(s
->runtime_journal
);
2462 ordered_hashmap_free_with_destructor(s
->user_journals
, journal_file_close
);
2464 varlink_server_unref(s
->varlink_server
);
2466 sd_event_source_unref(s
->syslog_event_source
);
2467 sd_event_source_unref(s
->native_event_source
);
2468 sd_event_source_unref(s
->stdout_event_source
);
2469 sd_event_source_unref(s
->dev_kmsg_event_source
);
2470 sd_event_source_unref(s
->audit_event_source
);
2471 sd_event_source_unref(s
->sync_event_source
);
2472 sd_event_source_unref(s
->sigusr1_event_source
);
2473 sd_event_source_unref(s
->sigusr2_event_source
);
2474 sd_event_source_unref(s
->sigterm_event_source
);
2475 sd_event_source_unref(s
->sigint_event_source
);
2476 sd_event_source_unref(s
->sigrtmin1_event_source
);
2477 sd_event_source_unref(s
->hostname_event_source
);
2478 sd_event_source_unref(s
->notify_event_source
);
2479 sd_event_source_unref(s
->watchdog_event_source
);
2480 sd_event_source_unref(s
->idle_event_source
);
2481 sd_event_unref(s
->event
);
2483 safe_close(s
->syslog_fd
);
2484 safe_close(s
->native_fd
);
2485 safe_close(s
->stdout_fd
);
2486 safe_close(s
->dev_kmsg_fd
);
2487 safe_close(s
->audit_fd
);
2488 safe_close(s
->hostname_fd
);
2489 safe_close(s
->notify_fd
);
2492 journal_ratelimit_free(s
->ratelimit
);
2494 if (s
->kernel_seqnum
)
2495 munmap(s
->kernel_seqnum
, sizeof(uint64_t));
2499 free(s
->cgroup_root
);
2500 free(s
->hostname_field
);
2501 free(s
->runtime_storage
.path
);
2502 free(s
->system_storage
.path
);
2503 free(s
->runtime_directory
);
2505 mmap_cache_unref(s
->mmap
);
2508 static const char* const storage_table
[_STORAGE_MAX
] = {
2509 [STORAGE_AUTO
] = "auto",
2510 [STORAGE_VOLATILE
] = "volatile",
2511 [STORAGE_PERSISTENT
] = "persistent",
2512 [STORAGE_NONE
] = "none"
2515 DEFINE_STRING_TABLE_LOOKUP(storage
, Storage
);
2516 DEFINE_CONFIG_PARSE_ENUM(config_parse_storage
, storage
, Storage
, "Failed to parse storage setting");
2518 static const char* const split_mode_table
[_SPLIT_MAX
] = {
2519 [SPLIT_LOGIN
] = "login",
2520 [SPLIT_UID
] = "uid",
2521 [SPLIT_NONE
] = "none",
2524 DEFINE_STRING_TABLE_LOOKUP(split_mode
, SplitMode
);
2525 DEFINE_CONFIG_PARSE_ENUM(config_parse_split_mode
, split_mode
, SplitMode
, "Failed to parse split mode setting");
2527 int config_parse_line_max(
2529 const char *filename
,
2531 const char *section
,
2532 unsigned section_line
,
2547 if (isempty(rvalue
))
2548 /* Empty assignment means default */
2549 *sz
= DEFAULT_LINE_MAX
;
2553 r
= parse_size(rvalue
, 1024, &v
);
2555 log_syntax(unit
, LOG_WARNING
, filename
, line
, r
, "Failed to parse LineMax= value, ignoring: %s", rvalue
);
2560 /* Why specify 79 here as minimum line length? Simply, because the most common traditional
2561 * terminal size is 80ch, and it might make sense to break one character before the natural
2562 * line break would occur on that. */
2563 log_syntax(unit
, LOG_WARNING
, filename
, line
, 0, "LineMax= too small, clamping to 79: %s", rvalue
);
2565 } else if (v
> (uint64_t) (SSIZE_MAX
-1)) {
2566 /* So, why specify SSIZE_MAX-1 here? Because that's one below the largest size value read()
2567 * can return, and we need one extra byte for the trailing NUL byte. Of course IRL such large
2568 * memory allocations will fail anyway, hence this limit is mostly theoretical anyway, as we'll
2569 * fail much earlier anyway. */
2570 log_syntax(unit
, LOG_WARNING
, filename
, line
, 0, "LineMax= too large, clamping to %" PRIu64
": %s", (uint64_t) (SSIZE_MAX
-1), rvalue
);
2579 int config_parse_compress(
2581 const char *filename
,
2583 const char *section
,
2584 unsigned section_line
,
2591 JournalCompressOptions
* compress
= data
;
2594 if (isempty(rvalue
)) {
2595 compress
->enabled
= true;
2596 compress
->threshold_bytes
= (uint64_t) -1;
2597 } else if (streq(rvalue
, "1")) {
2598 log_syntax(unit
, LOG_WARNING
, filename
, line
, 0,
2599 "Compress= ambiguously specified as 1, enabling compression with default threshold");
2600 compress
->enabled
= true;
2601 } else if (streq(rvalue
, "0")) {
2602 log_syntax(unit
, LOG_WARNING
, filename
, line
, 0,
2603 "Compress= ambiguously specified as 0, disabling compression");
2604 compress
->enabled
= false;
2606 r
= parse_boolean(rvalue
);
2608 r
= parse_size(rvalue
, 1024, &compress
->threshold_bytes
);
2610 log_syntax(unit
, LOG_WARNING
, filename
, line
, r
,
2611 "Failed to parse Compress= value, ignoring: %s", rvalue
);
2613 compress
->enabled
= true;
2615 compress
->enabled
= r
;