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