]> git.ipfire.org Git - thirdparty/systemd.git/blob - src/journal/journal-file.c
projects / thirdparty / systemd.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Merge pull request #4199 from dvdhrm/hwdb-order
[thirdparty/systemd.git] / src / journal / journal-file.c
1 /***
2 This file is part of systemd.
3
4 Copyright 2011 Lennart Poettering
5
6 systemd is free software; you can redistribute it and/or modify it
7 under the terms of the GNU Lesser General Public License as published by
8 the Free Software Foundation; either version 2.1 of the License, or
9 (at your option) any later version.
10
11 systemd is distributed in the hope that it will be useful, but
12 WITHOUT ANY WARRANTY; without even the implied warranty of
13 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14 Lesser General Public License for more details.
15
16 You should have received a copy of the GNU Lesser General Public License
17 along with systemd; If not, see <http://www.gnu.org/licenses/>.
18 ***/
19
20 #include <errno.h>
21 #include <fcntl.h>
22 #include <linux/fs.h>
23 #include <pthread.h>
24 #include <stddef.h>
25 #include <sys/mman.h>
26 #include <sys/statvfs.h>
27 #include <sys/uio.h>
28 #include <unistd.h>
29
30 #include "alloc-util.h"
31 #include "btrfs-util.h"
32 #include "chattr-util.h"
33 #include "compress.h"
34 #include "fd-util.h"
35 #include "journal-authenticate.h"
36 #include "journal-def.h"
37 #include "journal-file.h"
38 #include "lookup3.h"
39 #include "parse-util.h"
40 #include "path-util.h"
41 #include "random-util.h"
42 #include "sd-event.h"
43 #include "set.h"
44 #include "string-util.h"
45 #include "xattr-util.h"
46
47 #define DEFAULT_DATA_HASH_TABLE_SIZE (2047ULL*sizeof(HashItem))
48 #define DEFAULT_FIELD_HASH_TABLE_SIZE (333ULL*sizeof(HashItem))
49
50 #define COMPRESSION_SIZE_THRESHOLD (512ULL)
51
52 /* This is the minimum journal file size */
53 #define JOURNAL_FILE_SIZE_MIN (512ULL*1024ULL) /* 512 KiB */
54
55 /* These are the lower and upper bounds if we deduce the max_use value
56 * from the file system size */
57 #define DEFAULT_MAX_USE_LOWER (1ULL*1024ULL*1024ULL) /* 1 MiB */
58 #define DEFAULT_MAX_USE_UPPER (4ULL*1024ULL*1024ULL*1024ULL) /* 4 GiB */
59
60 /* This is the default minimal use limit, how much we'll use even if keep_free suggests otherwise. */
61 #define DEFAULT_MIN_USE (1ULL*1024ULL*1024ULL) /* 1 MiB */
62
63 /* This is the upper bound if we deduce max_size from max_use */
64 #define DEFAULT_MAX_SIZE_UPPER (128ULL*1024ULL*1024ULL) /* 128 MiB */
65
66 /* This is the upper bound if we deduce the keep_free value from the
67 * file system size */
68 #define DEFAULT_KEEP_FREE_UPPER (4ULL*1024ULL*1024ULL*1024ULL) /* 4 GiB */
69
70 /* This is the keep_free value when we can't determine the system
71 * size */
72 #define DEFAULT_KEEP_FREE (1024ULL*1024ULL) /* 1 MB */
73
74 /* This is the default maximum number of journal files to keep around. */
75 #define DEFAULT_N_MAX_FILES (100)
76
77 /* n_data was the first entry we added after the initial file format design */
78 #define HEADER_SIZE_MIN ALIGN64(offsetof(Header, n_data))
79
80 /* How many entries to keep in the entry array chain cache at max */
81 #define CHAIN_CACHE_MAX 20
82
83 /* How much to increase the journal file size at once each time we allocate something new. */
84 #define FILE_SIZE_INCREASE (8ULL*1024ULL*1024ULL) /* 8MB */
85
86 /* Reread fstat() of the file for detecting deletions at least this often */
87 #define LAST_STAT_REFRESH_USEC (5*USEC_PER_SEC)
88
89 /* The mmap context to use for the header we pick as one above the last defined typed */
90 #define CONTEXT_HEADER _OBJECT_TYPE_MAX
91
92 /* This may be called from a separate thread to prevent blocking the caller for the duration of fsync().
93 * As a result we use atomic operations on f->offline_state for inter-thread communications with
94 * journal_file_set_offline() and journal_file_set_online(). */
95 static void journal_file_set_offline_internal(JournalFile *f) {
96 assert(f);
97 assert(f->fd >= 0);
98 assert(f->header);
99
100 for (;;) {
101 switch (f->offline_state) {
102 case OFFLINE_CANCEL:
103 if (!__sync_bool_compare_and_swap(&f->offline_state, OFFLINE_CANCEL, OFFLINE_DONE))
104 continue;
105 return;
106
107 case OFFLINE_AGAIN_FROM_SYNCING:
108 if (!__sync_bool_compare_and_swap(&f->offline_state, OFFLINE_AGAIN_FROM_SYNCING, OFFLINE_SYNCING))
109 continue;
110 break;
111
112 case OFFLINE_AGAIN_FROM_OFFLINING:
113 if (!__sync_bool_compare_and_swap(&f->offline_state, OFFLINE_AGAIN_FROM_OFFLINING, OFFLINE_SYNCING))
114 continue;
115 break;
116
117 case OFFLINE_SYNCING:
118 (void) fsync(f->fd);
119
120 if (!__sync_bool_compare_and_swap(&f->offline_state, OFFLINE_SYNCING, OFFLINE_OFFLINING))
121 continue;
122
123 f->header->state = f->archive ? STATE_ARCHIVED : STATE_OFFLINE;
124 (void) fsync(f->fd);
125 break;
126
127 case OFFLINE_OFFLINING:
128 if (!__sync_bool_compare_and_swap(&f->offline_state, OFFLINE_OFFLINING, OFFLINE_DONE))
129 continue;
130 /* fall through */
131
132 case OFFLINE_DONE:
133 return;
134
135 case OFFLINE_JOINED:
136 log_debug("OFFLINE_JOINED unexpected offline state for journal_file_set_offline_internal()");
137 return;
138 }
139 }
140 }
141
142 static void * journal_file_set_offline_thread(void *arg) {
143 JournalFile *f = arg;
144
145 journal_file_set_offline_internal(f);
146
147 return NULL;
148 }
149
150 static int journal_file_set_offline_thread_join(JournalFile *f) {
151 int r;
152
153 assert(f);
154
155 if (f->offline_state == OFFLINE_JOINED)
156 return 0;
157
158 r = pthread_join(f->offline_thread, NULL);
159 if (r)
160 return -r;
161
162 f->offline_state = OFFLINE_JOINED;
163
164 if (mmap_cache_got_sigbus(f->mmap, f->fd))
165 return -EIO;
166
167 return 0;
168 }
169
170 /* Trigger a restart if the offline thread is mid-flight in a restartable state. */
171 static bool journal_file_set_offline_try_restart(JournalFile *f) {
172 for (;;) {
173 switch (f->offline_state) {
174 case OFFLINE_AGAIN_FROM_SYNCING:
175 case OFFLINE_AGAIN_FROM_OFFLINING:
176 return true;
177
178 case OFFLINE_CANCEL:
179 if (!__sync_bool_compare_and_swap(&f->offline_state, OFFLINE_CANCEL, OFFLINE_AGAIN_FROM_SYNCING))
180 continue;
181 return true;
182
183 case OFFLINE_SYNCING:
184 if (!__sync_bool_compare_and_swap(&f->offline_state, OFFLINE_SYNCING, OFFLINE_AGAIN_FROM_SYNCING))
185 continue;
186 return true;
187
188 case OFFLINE_OFFLINING:
189 if (!__sync_bool_compare_and_swap(&f->offline_state, OFFLINE_OFFLINING, OFFLINE_AGAIN_FROM_OFFLINING))
190 continue;
191 return true;
192
193 default:
194 return false;
195 }
196 }
197 }
198
199 /* Sets a journal offline.
200 *
201 * If wait is false then an offline is dispatched in a separate thread for a
202 * subsequent journal_file_set_offline() or journal_file_set_online() of the
203 * same journal to synchronize with.
204 *
205 * If wait is true, then either an existing offline thread will be restarted
206 * and joined, or if none exists the offline is simply performed in this
207 * context without involving another thread.
208 */
209 int journal_file_set_offline(JournalFile *f, bool wait) {
210 bool restarted;
211 int r;
212
213 assert(f);
214
215 if (!f->writable)
216 return -EPERM;
217
218 if (!(f->fd >= 0 && f->header))
219 return -EINVAL;
220
221 /* An offlining journal is implicitly online and may modify f->header->state,
222 * we must also join any potentially lingering offline thread when not online. */
223 if (!journal_file_is_offlining(f) && f->header->state != STATE_ONLINE)
224 return journal_file_set_offline_thread_join(f);
225
226 /* Restart an in-flight offline thread and wait if needed, or join a lingering done one. */
227 restarted = journal_file_set_offline_try_restart(f);
228 if ((restarted && wait) || !restarted) {
229 r = journal_file_set_offline_thread_join(f);
230 if (r < 0)
231 return r;
232 }
233
234 if (restarted)
235 return 0;
236
237 /* Initiate a new offline. */
238 f->offline_state = OFFLINE_SYNCING;
239
240 if (wait) /* Without using a thread if waiting. */
241 journal_file_set_offline_internal(f);
242 else {
243 r = pthread_create(&f->offline_thread, NULL, journal_file_set_offline_thread, f);
244 if (r > 0) {
245 f->offline_state = OFFLINE_JOINED;
246 return -r;
247 }
248 }
249
250 return 0;
251 }
252
253 static int journal_file_set_online(JournalFile *f) {
254 bool joined = false;
255
256 assert(f);
257
258 if (!f->writable)
259 return -EPERM;
260
261 if (!(f->fd >= 0 && f->header))
262 return -EINVAL;
263
264 while (!joined) {
265 switch (f->offline_state) {
266 case OFFLINE_JOINED:
267 /* No offline thread, no need to wait. */
268 joined = true;
269 break;
270
271 case OFFLINE_SYNCING:
272 if (!__sync_bool_compare_and_swap(&f->offline_state, OFFLINE_SYNCING, OFFLINE_CANCEL))
273 continue;
274 /* Canceled syncing prior to offlining, no need to wait. */
275 break;
276
277 case OFFLINE_AGAIN_FROM_SYNCING:
278 if (!__sync_bool_compare_and_swap(&f->offline_state, OFFLINE_AGAIN_FROM_SYNCING, OFFLINE_CANCEL))
279 continue;
280 /* Canceled restart from syncing, no need to wait. */
281 break;
282
283 case OFFLINE_AGAIN_FROM_OFFLINING:
284 if (!__sync_bool_compare_and_swap(&f->offline_state, OFFLINE_AGAIN_FROM_OFFLINING, OFFLINE_CANCEL))
285 continue;
286 /* Canceled restart from offlining, must wait for offlining to complete however. */
287
288 /* fall through to wait */
289 default: {
290 int r;
291
292 r = journal_file_set_offline_thread_join(f);
293 if (r < 0)
294 return r;
295
296 joined = true;
297 break;
298 }
299 }
300 }
301
302 if (mmap_cache_got_sigbus(f->mmap, f->fd))
303 return -EIO;
304
305 switch (f->header->state) {
306 case STATE_ONLINE:
307 return 0;
308
309 case STATE_OFFLINE:
310 f->header->state = STATE_ONLINE;
311 (void) fsync(f->fd);
312 return 0;
313
314 default:
315 return -EINVAL;
316 }
317 }
318
319 bool journal_file_is_offlining(JournalFile *f) {
320 assert(f);
321
322 __sync_synchronize();
323
324 if (f->offline_state == OFFLINE_DONE ||
325 f->offline_state == OFFLINE_JOINED)
326 return false;
327
328 return true;
329 }
330
331 JournalFile* journal_file_close(JournalFile *f) {
332 assert(f);
333
334 #ifdef HAVE_GCRYPT
335 /* Write the final tag */
336 if (f->seal && f->writable) {
337 int r;
338
339 r = journal_file_append_tag(f);
340 if (r < 0)
341 log_error_errno(r, "Failed to append tag when closing journal: %m");
342 }
343 #endif
344
345 if (f->post_change_timer) {
346 int enabled;
347
348 if (sd_event_source_get_enabled(f->post_change_timer, &enabled) >= 0)
349 if (enabled == SD_EVENT_ONESHOT)
350 journal_file_post_change(f);
351
352 (void) sd_event_source_set_enabled(f->post_change_timer, SD_EVENT_OFF);
353 sd_event_source_unref(f->post_change_timer);
354 }
355
356 journal_file_set_offline(f, true);
357
358 if (f->mmap && f->fd >= 0)
359 mmap_cache_close_fd(f->mmap, f->fd);
360
361 if (f->fd >= 0 && f->defrag_on_close) {
362
363 /* Be friendly to btrfs: turn COW back on again now,
364 * and defragment the file. We won't write to the file
365 * ever again, hence remove all fragmentation, and
366 * reenable all the good bits COW usually provides
367 * (such as data checksumming). */
368
369 (void) chattr_fd(f->fd, 0, FS_NOCOW_FL);
370 (void) btrfs_defrag_fd(f->fd);
371 }
372
373 if (f->close_fd)
374 safe_close(f->fd);
375 free(f->path);
376
377 mmap_cache_unref(f->mmap);
378
379 ordered_hashmap_free_free(f->chain_cache);
380
381 #if defined(HAVE_XZ) || defined(HAVE_LZ4)
382 free(f->compress_buffer);
383 #endif
384
385 #ifdef HAVE_GCRYPT
386 if (f->fss_file)
387 munmap(f->fss_file, PAGE_ALIGN(f->fss_file_size));
388 else
389 free(f->fsprg_state);
390
391 free(f->fsprg_seed);
392
393 if (f->hmac)
394 gcry_md_close(f->hmac);
395 #endif
396
397 free(f);
398 return NULL;
399 }
400
401 void journal_file_close_set(Set *s) {
402 JournalFile *f;
403
404 assert(s);
405
406 while ((f = set_steal_first(s)))
407 (void) journal_file_close(f);
408 }
409
410 static int journal_file_init_header(JournalFile *f, JournalFile *template) {
411 Header h = {};
412 ssize_t k;
413 int r;
414
415 assert(f);
416
417 memcpy(h.signature, HEADER_SIGNATURE, 8);
418 h.header_size = htole64(ALIGN64(sizeof(h)));
419
420 h.incompatible_flags |= htole32(
421 f->compress_xz * HEADER_INCOMPATIBLE_COMPRESSED_XZ |
422 f->compress_lz4 * HEADER_INCOMPATIBLE_COMPRESSED_LZ4);
423
424 h.compatible_flags = htole32(
425 f->seal * HEADER_COMPATIBLE_SEALED);
426
427 r = sd_id128_randomize(&h.file_id);
428 if (r < 0)
429 return r;
430
431 if (template) {
432 h.seqnum_id = template->header->seqnum_id;
433 h.tail_entry_seqnum = template->header->tail_entry_seqnum;
434 } else
435 h.seqnum_id = h.file_id;
436
437 k = pwrite(f->fd, &h, sizeof(h), 0);
438 if (k < 0)
439 return -errno;
440
441 if (k != sizeof(h))
442 return -EIO;
443
444 return 0;
445 }
446
447 static int fsync_directory_of_file(int fd) {
448 _cleanup_free_ char *path = NULL, *dn = NULL;
449 _cleanup_close_ int dfd = -1;
450 struct stat st;
451 int r;
452
453 if (fstat(fd, &st) < 0)
454 return -errno;
455
456 if (!S_ISREG(st.st_mode))
457 return -EBADFD;
458
459 r = fd_get_path(fd, &path);
460 if (r < 0)
461 return r;
462
463 if (!path_is_absolute(path))
464 return -EINVAL;
465
466 dn = dirname_malloc(path);
467 if (!dn)
468 return -ENOMEM;
469
470 dfd = open(dn, O_RDONLY|O_CLOEXEC|O_DIRECTORY);
471 if (dfd < 0)
472 return -errno;
473
474 if (fsync(dfd) < 0)
475 return -errno;
476
477 return 0;
478 }
479
480 static int journal_file_refresh_header(JournalFile *f) {
481 sd_id128_t boot_id;
482 int r;
483
484 assert(f);
485 assert(f->header);
486
487 r = sd_id128_get_machine(&f->header->machine_id);
488 if (r < 0)
489 return r;
490
491 r = sd_id128_get_boot(&boot_id);
492 if (r < 0)
493 return r;
494
495 if (sd_id128_equal(boot_id, f->header->boot_id))
496 f->tail_entry_monotonic_valid = true;
497
498 f->header->boot_id = boot_id;
499
500 r = journal_file_set_online(f);
501
502 /* Sync the online state to disk */
503 (void) fsync(f->fd);
504
505 /* We likely just created a new file, also sync the directory this file is located in. */
506 (void) fsync_directory_of_file(f->fd);
507
508 return r;
509 }
510
511 static int journal_file_verify_header(JournalFile *f) {
512 uint32_t flags;
513
514 assert(f);
515 assert(f->header);
516
517 if (memcmp(f->header->signature, HEADER_SIGNATURE, 8))
518 return -EBADMSG;
519
520 /* In both read and write mode we refuse to open files with
521 * incompatible flags we don't know */
522 flags = le32toh(f->header->incompatible_flags);
523 if (flags & ~HEADER_INCOMPATIBLE_SUPPORTED) {
524 if (flags & ~HEADER_INCOMPATIBLE_ANY)
525 log_debug("Journal file %s has unknown incompatible flags %"PRIx32,
526 f->path, flags & ~HEADER_INCOMPATIBLE_ANY);
527 flags = (flags & HEADER_INCOMPATIBLE_ANY) & ~HEADER_INCOMPATIBLE_SUPPORTED;
528 if (flags)
529 log_debug("Journal file %s uses incompatible flags %"PRIx32
530 " disabled at compilation time.", f->path, flags);
531 return -EPROTONOSUPPORT;
532 }
533
534 /* When open for writing we refuse to open files with
535 * compatible flags, too */
536 flags = le32toh(f->header->compatible_flags);
537 if (f->writable && (flags & ~HEADER_COMPATIBLE_SUPPORTED)) {
538 if (flags & ~HEADER_COMPATIBLE_ANY)
539 log_debug("Journal file %s has unknown compatible flags %"PRIx32,
540 f->path, flags & ~HEADER_COMPATIBLE_ANY);
541 flags = (flags & HEADER_COMPATIBLE_ANY) & ~HEADER_COMPATIBLE_SUPPORTED;
542 if (flags)
543 log_debug("Journal file %s uses compatible flags %"PRIx32
544 " disabled at compilation time.", f->path, flags);
545 return -EPROTONOSUPPORT;
546 }
547
548 if (f->header->state >= _STATE_MAX)
549 return -EBADMSG;
550
551 /* The first addition was n_data, so check that we are at least this large */
552 if (le64toh(f->header->header_size) < HEADER_SIZE_MIN)
553 return -EBADMSG;
554
555 if (JOURNAL_HEADER_SEALED(f->header) && !JOURNAL_HEADER_CONTAINS(f->header, n_entry_arrays))
556 return -EBADMSG;
557
558 if ((le64toh(f->header->header_size) + le64toh(f->header->arena_size)) > (uint64_t) f->last_stat.st_size)
559 return -ENODATA;
560
561 if (le64toh(f->header->tail_object_offset) > (le64toh(f->header->header_size) + le64toh(f->header->arena_size)))
562 return -ENODATA;
563
564 if (!VALID64(le64toh(f->header->data_hash_table_offset)) ||
565 !VALID64(le64toh(f->header->field_hash_table_offset)) ||
566 !VALID64(le64toh(f->header->tail_object_offset)) ||
567 !VALID64(le64toh(f->header->entry_array_offset)))
568 return -ENODATA;
569
570 if (f->writable) {
571 uint8_t state;
572 sd_id128_t machine_id;
573 int r;
574
575 r = sd_id128_get_machine(&machine_id);
576 if (r < 0)
577 return r;
578
579 if (!sd_id128_equal(machine_id, f->header->machine_id))
580 return -EHOSTDOWN;
581
582 state = f->header->state;
583
584 if (state == STATE_ONLINE) {
585 log_debug("Journal file %s is already online. Assuming unclean closing.", f->path);
586 return -EBUSY;
587 } else if (state == STATE_ARCHIVED)
588 return -ESHUTDOWN;
589 else if (state != STATE_OFFLINE) {
590 log_debug("Journal file %s has unknown state %i.", f->path, state);
591 return -EBUSY;
592 }
593 }
594
595 f->compress_xz = JOURNAL_HEADER_COMPRESSED_XZ(f->header);
596 f->compress_lz4 = JOURNAL_HEADER_COMPRESSED_LZ4(f->header);
597
598 f->seal = JOURNAL_HEADER_SEALED(f->header);
599
600 return 0;
601 }
602
603 static int journal_file_fstat(JournalFile *f) {
604 assert(f);
605 assert(f->fd >= 0);
606
607 if (fstat(f->fd, &f->last_stat) < 0)
608 return -errno;
609
610 f->last_stat_usec = now(CLOCK_MONOTONIC);
611
612 /* Refuse appending to files that are already deleted */
613 if (f->last_stat.st_nlink <= 0)
614 return -EIDRM;
615
616 return 0;
617 }
618
619 static int journal_file_allocate(JournalFile *f, uint64_t offset, uint64_t size) {
620 uint64_t old_size, new_size;
621 int r;
622
623 assert(f);
624 assert(f->header);
625
626 /* We assume that this file is not sparse, and we know that
627 * for sure, since we always call posix_fallocate()
628 * ourselves */
629
630 if (mmap_cache_got_sigbus(f->mmap, f->fd))
631 return -EIO;
632
633 old_size =
634 le64toh(f->header->header_size) +
635 le64toh(f->header->arena_size);
636
637 new_size = PAGE_ALIGN(offset + size);
638 if (new_size < le64toh(f->header->header_size))
639 new_size = le64toh(f->header->header_size);
640
641 if (new_size <= old_size) {
642
643 /* We already pre-allocated enough space, but before
644 * we write to it, let's check with fstat() if the
645 * file got deleted, in order make sure we don't throw
646 * away the data immediately. Don't check fstat() for
647 * all writes though, but only once ever 10s. */
648
649 if (f->last_stat_usec + LAST_STAT_REFRESH_USEC > now(CLOCK_MONOTONIC))
650 return 0;
651
652 return journal_file_fstat(f);
653 }
654
655 /* Allocate more space. */
656
657 if (f->metrics.max_size > 0 && new_size > f->metrics.max_size)
658 return -E2BIG;
659
660 if (new_size > f->metrics.min_size && f->metrics.keep_free > 0) {
661 struct statvfs svfs;
662
663 if (fstatvfs(f->fd, &svfs) >= 0) {
664 uint64_t available;
665
666 available = LESS_BY((uint64_t) svfs.f_bfree * (uint64_t) svfs.f_bsize, f->metrics.keep_free);
667
668 if (new_size - old_size > available)
669 return -E2BIG;
670 }
671 }
672
673 /* Increase by larger blocks at once */
674 new_size = ((new_size+FILE_SIZE_INCREASE-1) / FILE_SIZE_INCREASE) * FILE_SIZE_INCREASE;
675 if (f->metrics.max_size > 0 && new_size > f->metrics.max_size)
676 new_size = f->metrics.max_size;
677
678 /* Note that the glibc fallocate() fallback is very
679 inefficient, hence we try to minimize the allocation area
680 as we can. */
681 r = posix_fallocate(f->fd, old_size, new_size - old_size);
682 if (r != 0)
683 return -r;
684
685 f->header->arena_size = htole64(new_size - le64toh(f->header->header_size));
686
687 return journal_file_fstat(f);
688 }
689
690 static unsigned type_to_context(ObjectType type) {
691 /* One context for each type, plus one catch-all for the rest */
692 assert_cc(_OBJECT_TYPE_MAX <= MMAP_CACHE_MAX_CONTEXTS);
693 assert_cc(CONTEXT_HEADER < MMAP_CACHE_MAX_CONTEXTS);
694 return type > OBJECT_UNUSED && type < _OBJECT_TYPE_MAX ? type : 0;
695 }
696
697 static int journal_file_move_to(JournalFile *f, ObjectType type, bool keep_always, uint64_t offset, uint64_t size, void **ret) {
698 int r;
699
700 assert(f);
701 assert(ret);
702
703 if (size <= 0)
704 return -EINVAL;
705
706 /* Avoid SIGBUS on invalid accesses */
707 if (offset + size > (uint64_t) f->last_stat.st_size) {
708 /* Hmm, out of range? Let's refresh the fstat() data
709 * first, before we trust that check. */
710
711 r = journal_file_fstat(f);
712 if (r < 0)
713 return r;
714
715 if (offset + size > (uint64_t) f->last_stat.st_size)
716 return -EADDRNOTAVAIL;
717 }
718
719 return mmap_cache_get(f->mmap, f->fd, f->prot, type_to_context(type), keep_always, offset, size, &f->last_stat, ret);
720 }
721
722 static uint64_t minimum_header_size(Object *o) {
723
724 static const uint64_t table[] = {
725 [OBJECT_DATA] = sizeof(DataObject),
726 [OBJECT_FIELD] = sizeof(FieldObject),
727 [OBJECT_ENTRY] = sizeof(EntryObject),
728 [OBJECT_DATA_HASH_TABLE] = sizeof(HashTableObject),
729 [OBJECT_FIELD_HASH_TABLE] = sizeof(HashTableObject),
730 [OBJECT_ENTRY_ARRAY] = sizeof(EntryArrayObject),
731 [OBJECT_TAG] = sizeof(TagObject),
732 };
733
734 if (o->object.type >= ELEMENTSOF(table) || table[o->object.type] <= 0)
735 return sizeof(ObjectHeader);
736
737 return table[o->object.type];
738 }
739
740 int journal_file_move_to_object(JournalFile *f, ObjectType type, uint64_t offset, Object **ret) {
741 int r;
742 void *t;
743 Object *o;
744 uint64_t s;
745
746 assert(f);
747 assert(ret);
748
749 /* Objects may only be located at multiple of 64 bit */
750 if (!VALID64(offset))
751 return -EBADMSG;
752
753 /* Object may not be located in the file header */
754 if (offset < le64toh(f->header->header_size))
755 return -EBADMSG;
756
757 r = journal_file_move_to(f, type, false, offset, sizeof(ObjectHeader), &t);
758 if (r < 0)
759 return r;
760
761 o = (Object*) t;
762 s = le64toh(o->object.size);
763
764 if (s < sizeof(ObjectHeader))
765 return -EBADMSG;
766
767 if (o->object.type <= OBJECT_UNUSED)
768 return -EBADMSG;
769
770 if (s < minimum_header_size(o))
771 return -EBADMSG;
772
773 if (type > OBJECT_UNUSED && o->object.type != type)
774 return -EBADMSG;
775
776 if (s > sizeof(ObjectHeader)) {
777 r = journal_file_move_to(f, type, false, offset, s, &t);
778 if (r < 0)
779 return r;
780
781 o = (Object*) t;
782 }
783
784 *ret = o;
785 return 0;
786 }
787
788 static uint64_t journal_file_entry_seqnum(JournalFile *f, uint64_t *seqnum) {
789 uint64_t r;
790
791 assert(f);
792 assert(f->header);
793
794 r = le64toh(f->header->tail_entry_seqnum) + 1;
795
796 if (seqnum) {
797 /* If an external seqnum counter was passed, we update
798 * both the local and the external one, and set it to
799 * the maximum of both */
800
801 if (*seqnum + 1 > r)
802 r = *seqnum + 1;
803
804 *seqnum = r;
805 }
806
807 f->header->tail_entry_seqnum = htole64(r);
808
809 if (f->header->head_entry_seqnum == 0)
810 f->header->head_entry_seqnum = htole64(r);
811
812 return r;
813 }
814
815 int journal_file_append_object(JournalFile *f, ObjectType type, uint64_t size, Object **ret, uint64_t *offset) {
816 int r;
817 uint64_t p;
818 Object *tail, *o;
819 void *t;
820
821 assert(f);
822 assert(f->header);
823 assert(type > OBJECT_UNUSED && type < _OBJECT_TYPE_MAX);
824 assert(size >= sizeof(ObjectHeader));
825 assert(offset);
826 assert(ret);
827
828 r = journal_file_set_online(f);
829 if (r < 0)
830 return r;
831
832 p = le64toh(f->header->tail_object_offset);
833 if (p == 0)
834 p = le64toh(f->header->header_size);
835 else {
836 r = journal_file_move_to_object(f, OBJECT_UNUSED, p, &tail);
837 if (r < 0)
838 return r;
839
840 p += ALIGN64(le64toh(tail->object.size));
841 }
842
843 r = journal_file_allocate(f, p, size);
844 if (r < 0)
845 return r;
846
847 r = journal_file_move_to(f, type, false, p, size, &t);
848 if (r < 0)
849 return r;
850
851 o = (Object*) t;
852
853 zero(o->object);
854 o->object.type = type;
855 o->object.size = htole64(size);
856
857 f->header->tail_object_offset = htole64(p);
858 f->header->n_objects = htole64(le64toh(f->header->n_objects) + 1);
859
860 *ret = o;
861 *offset = p;
862
863 return 0;
864 }
865
866 static int journal_file_setup_data_hash_table(JournalFile *f) {
867 uint64_t s, p;
868 Object *o;
869 int r;
870
871 assert(f);
872 assert(f->header);
873
874 /* We estimate that we need 1 hash table entry per 768 bytes
875 of journal file and we want to make sure we never get
876 beyond 75% fill level. Calculate the hash table size for
877 the maximum file size based on these metrics. */
878
879 s = (f->metrics.max_size * 4 / 768 / 3) * sizeof(HashItem);
880 if (s < DEFAULT_DATA_HASH_TABLE_SIZE)
881 s = DEFAULT_DATA_HASH_TABLE_SIZE;
882
883 log_debug("Reserving %"PRIu64" entries in hash table.", s / sizeof(HashItem));
884
885 r = journal_file_append_object(f,
886 OBJECT_DATA_HASH_TABLE,
887 offsetof(Object, hash_table.items) + s,
888 &o, &p);
889 if (r < 0)
890 return r;
891
892 memzero(o->hash_table.items, s);
893
894 f->header->data_hash_table_offset = htole64(p + offsetof(Object, hash_table.items));
895 f->header->data_hash_table_size = htole64(s);
896
897 return 0;
898 }
899
900 static int journal_file_setup_field_hash_table(JournalFile *f) {
901 uint64_t s, p;
902 Object *o;
903 int r;
904
905 assert(f);
906 assert(f->header);
907
908 /* We use a fixed size hash table for the fields as this
909 * number should grow very slowly only */
910
911 s = DEFAULT_FIELD_HASH_TABLE_SIZE;
912 r = journal_file_append_object(f,
913 OBJECT_FIELD_HASH_TABLE,
914 offsetof(Object, hash_table.items) + s,
915 &o, &p);
916 if (r < 0)
917 return r;
918
919 memzero(o->hash_table.items, s);
920
921 f->header->field_hash_table_offset = htole64(p + offsetof(Object, hash_table.items));
922 f->header->field_hash_table_size = htole64(s);
923
924 return 0;
925 }
926
927 int journal_file_map_data_hash_table(JournalFile *f) {
928 uint64_t s, p;
929 void *t;
930 int r;
931
932 assert(f);
933 assert(f->header);
934
935 if (f->data_hash_table)
936 return 0;
937
938 p = le64toh(f->header->data_hash_table_offset);
939 s = le64toh(f->header->data_hash_table_size);
940
941 r = journal_file_move_to(f,
942 OBJECT_DATA_HASH_TABLE,
943 true,
944 p, s,
945 &t);
946 if (r < 0)
947 return r;
948
949 f->data_hash_table = t;
950 return 0;
951 }
952
953 int journal_file_map_field_hash_table(JournalFile *f) {
954 uint64_t s, p;
955 void *t;
956 int r;
957
958 assert(f);
959 assert(f->header);
960
961 if (f->field_hash_table)
962 return 0;
963
964 p = le64toh(f->header->field_hash_table_offset);
965 s = le64toh(f->header->field_hash_table_size);
966
967 r = journal_file_move_to(f,
968 OBJECT_FIELD_HASH_TABLE,
969 true,
970 p, s,
971 &t);
972 if (r < 0)
973 return r;
974
975 f->field_hash_table = t;
976 return 0;
977 }
978
979 static int journal_file_link_field(
980 JournalFile *f,
981 Object *o,
982 uint64_t offset,
983 uint64_t hash) {
984
985 uint64_t p, h, m;
986 int r;
987
988 assert(f);
989 assert(f->header);
990 assert(f->field_hash_table);
991 assert(o);
992 assert(offset > 0);
993
994 if (o->object.type != OBJECT_FIELD)
995 return -EINVAL;
996
997 m = le64toh(f->header->field_hash_table_size) / sizeof(HashItem);
998 if (m <= 0)
999 return -EBADMSG;
1000
1001 /* This might alter the window we are looking at */
1002 o->field.next_hash_offset = o->field.head_data_offset = 0;
1003
1004 h = hash % m;
1005 p = le64toh(f->field_hash_table[h].tail_hash_offset);
1006 if (p == 0)
1007 f->field_hash_table[h].head_hash_offset = htole64(offset);
1008 else {
1009 r = journal_file_move_to_object(f, OBJECT_FIELD, p, &o);
1010 if (r < 0)
1011 return r;
1012
1013 o->field.next_hash_offset = htole64(offset);
1014 }
1015
1016 f->field_hash_table[h].tail_hash_offset = htole64(offset);
1017
1018 if (JOURNAL_HEADER_CONTAINS(f->header, n_fields))
1019 f->header->n_fields = htole64(le64toh(f->header->n_fields) + 1);
1020
1021 return 0;
1022 }
1023
1024 static int journal_file_link_data(
1025 JournalFile *f,
1026 Object *o,
1027 uint64_t offset,
1028 uint64_t hash) {
1029
1030 uint64_t p, h, m;
1031 int r;
1032
1033 assert(f);
1034 assert(f->header);
1035 assert(f->data_hash_table);
1036 assert(o);
1037 assert(offset > 0);
1038
1039 if (o->object.type != OBJECT_DATA)
1040 return -EINVAL;
1041
1042 m = le64toh(f->header->data_hash_table_size) / sizeof(HashItem);
1043 if (m <= 0)
1044 return -EBADMSG;
1045
1046 /* This might alter the window we are looking at */
1047 o->data.next_hash_offset = o->data.next_field_offset = 0;
1048 o->data.entry_offset = o->data.entry_array_offset = 0;
1049 o->data.n_entries = 0;
1050
1051 h = hash % m;
1052 p = le64toh(f->data_hash_table[h].tail_hash_offset);
1053 if (p == 0)
1054 /* Only entry in the hash table is easy */
1055 f->data_hash_table[h].head_hash_offset = htole64(offset);
1056 else {
1057 /* Move back to the previous data object, to patch in
1058 * pointer */
1059
1060 r = journal_file_move_to_object(f, OBJECT_DATA, p, &o);
1061 if (r < 0)
1062 return r;
1063
1064 o->data.next_hash_offset = htole64(offset);
1065 }
1066
1067 f->data_hash_table[h].tail_hash_offset = htole64(offset);
1068
1069 if (JOURNAL_HEADER_CONTAINS(f->header, n_data))
1070 f->header->n_data = htole64(le64toh(f->header->n_data) + 1);
1071
1072 return 0;
1073 }
1074
1075 int journal_file_find_field_object_with_hash(
1076 JournalFile *f,
1077 const void *field, uint64_t size, uint64_t hash,
1078 Object **ret, uint64_t *offset) {
1079
1080 uint64_t p, osize, h, m;
1081 int r;
1082
1083 assert(f);
1084 assert(f->header);
1085 assert(field && size > 0);
1086
1087 /* If the field hash table is empty, we can't find anything */
1088 if (le64toh(f->header->field_hash_table_size) <= 0)
1089 return 0;
1090
1091 /* Map the field hash table, if it isn't mapped yet. */
1092 r = journal_file_map_field_hash_table(f);
1093 if (r < 0)
1094 return r;
1095
1096 osize = offsetof(Object, field.payload) + size;
1097
1098 m = le64toh(f->header->field_hash_table_size) / sizeof(HashItem);
1099 if (m <= 0)
1100 return -EBADMSG;
1101
1102 h = hash % m;
1103 p = le64toh(f->field_hash_table[h].head_hash_offset);
1104
1105 while (p > 0) {
1106 Object *o;
1107
1108 r = journal_file_move_to_object(f, OBJECT_FIELD, p, &o);
1109 if (r < 0)
1110 return r;
1111
1112 if (le64toh(o->field.hash) == hash &&
1113 le64toh(o->object.size) == osize &&
1114 memcmp(o->field.payload, field, size) == 0) {
1115
1116 if (ret)
1117 *ret = o;
1118 if (offset)
1119 *offset = p;
1120
1121 return 1;
1122 }
1123
1124 p = le64toh(o->field.next_hash_offset);
1125 }
1126
1127 return 0;
1128 }
1129
1130 int journal_file_find_field_object(
1131 JournalFile *f,
1132 const void *field, uint64_t size,
1133 Object **ret, uint64_t *offset) {
1134
1135 uint64_t hash;
1136
1137 assert(f);
1138 assert(field && size > 0);
1139
1140 hash = hash64(field, size);
1141
1142 return journal_file_find_field_object_with_hash(f,
1143 field, size, hash,
1144 ret, offset);
1145 }
1146
1147 int journal_file_find_data_object_with_hash(
1148 JournalFile *f,
1149 const void *data, uint64_t size, uint64_t hash,
1150 Object **ret, uint64_t *offset) {
1151
1152 uint64_t p, osize, h, m;
1153 int r;
1154
1155 assert(f);
1156 assert(f->header);
1157 assert(data || size == 0);
1158
1159 /* If there's no data hash table, then there's no entry. */
1160 if (le64toh(f->header->data_hash_table_size) <= 0)
1161 return 0;
1162
1163 /* Map the data hash table, if it isn't mapped yet. */
1164 r = journal_file_map_data_hash_table(f);
1165 if (r < 0)
1166 return r;
1167
1168 osize = offsetof(Object, data.payload) + size;
1169
1170 m = le64toh(f->header->data_hash_table_size) / sizeof(HashItem);
1171 if (m <= 0)
1172 return -EBADMSG;
1173
1174 h = hash % m;
1175 p = le64toh(f->data_hash_table[h].head_hash_offset);
1176
1177 while (p > 0) {
1178 Object *o;
1179
1180 r = journal_file_move_to_object(f, OBJECT_DATA, p, &o);
1181 if (r < 0)
1182 return r;
1183
1184 if (le64toh(o->data.hash) != hash)
1185 goto next;
1186
1187 if (o->object.flags & OBJECT_COMPRESSION_MASK) {
1188 #if defined(HAVE_XZ) || defined(HAVE_LZ4)
1189 uint64_t l;
1190 size_t rsize = 0;
1191
1192 l = le64toh(o->object.size);
1193 if (l <= offsetof(Object, data.payload))
1194 return -EBADMSG;
1195
1196 l -= offsetof(Object, data.payload);
1197
1198 r = decompress_blob(o->object.flags & OBJECT_COMPRESSION_MASK,
1199 o->data.payload, l, &f->compress_buffer, &f->compress_buffer_size, &rsize, 0);
1200 if (r < 0)
1201 return r;
1202
1203 if (rsize == size &&
1204 memcmp(f->compress_buffer, data, size) == 0) {
1205
1206 if (ret)
1207 *ret = o;
1208
1209 if (offset)
1210 *offset = p;
1211
1212 return 1;
1213 }
1214 #else
1215 return -EPROTONOSUPPORT;
1216 #endif
1217 } else if (le64toh(o->object.size) == osize &&
1218 memcmp(o->data.payload, data, size) == 0) {
1219
1220 if (ret)
1221 *ret = o;
1222
1223 if (offset)
1224 *offset = p;
1225
1226 return 1;
1227 }
1228
1229 next:
1230 p = le64toh(o->data.next_hash_offset);
1231 }
1232
1233 return 0;
1234 }
1235
1236 int journal_file_find_data_object(
1237 JournalFile *f,
1238 const void *data, uint64_t size,
1239 Object **ret, uint64_t *offset) {
1240
1241 uint64_t hash;
1242
1243 assert(f);
1244 assert(data || size == 0);
1245
1246 hash = hash64(data, size);
1247
1248 return journal_file_find_data_object_with_hash(f,
1249 data, size, hash,
1250 ret, offset);
1251 }
1252
1253 static int journal_file_append_field(
1254 JournalFile *f,
1255 const void *field, uint64_t size,
1256 Object **ret, uint64_t *offset) {
1257
1258 uint64_t hash, p;
1259 uint64_t osize;
1260 Object *o;
1261 int r;
1262
1263 assert(f);
1264 assert(field && size > 0);
1265
1266 hash = hash64(field, size);
1267
1268 r = journal_file_find_field_object_with_hash(f, field, size, hash, &o, &p);
1269 if (r < 0)
1270 return r;
1271 else if (r > 0) {
1272
1273 if (ret)
1274 *ret = o;
1275
1276 if (offset)
1277 *offset = p;
1278
1279 return 0;
1280 }
1281
1282 osize = offsetof(Object, field.payload) + size;
1283 r = journal_file_append_object(f, OBJECT_FIELD, osize, &o, &p);
1284 if (r < 0)
1285 return r;
1286
1287 o->field.hash = htole64(hash);
1288 memcpy(o->field.payload, field, size);
1289
1290 r = journal_file_link_field(f, o, p, hash);
1291 if (r < 0)
1292 return r;
1293
1294 /* The linking might have altered the window, so let's
1295 * refresh our pointer */
1296 r = journal_file_move_to_object(f, OBJECT_FIELD, p, &o);
1297 if (r < 0)
1298 return r;
1299
1300 #ifdef HAVE_GCRYPT
1301 r = journal_file_hmac_put_object(f, OBJECT_FIELD, o, p);
1302 if (r < 0)
1303 return r;
1304 #endif
1305
1306 if (ret)
1307 *ret = o;
1308
1309 if (offset)
1310 *offset = p;
1311
1312 return 0;
1313 }
1314
1315 static int journal_file_append_data(
1316 JournalFile *f,
1317 const void *data, uint64_t size,
1318 Object **ret, uint64_t *offset) {
1319
1320 uint64_t hash, p;
1321 uint64_t osize;
1322 Object *o;
1323 int r, compression = 0;
1324 const void *eq;
1325
1326 assert(f);
1327 assert(data || size == 0);
1328
1329 hash = hash64(data, size);
1330
1331 r = journal_file_find_data_object_with_hash(f, data, size, hash, &o, &p);
1332 if (r < 0)
1333 return r;
1334 if (r > 0) {
1335
1336 if (ret)
1337 *ret = o;
1338
1339 if (offset)
1340 *offset = p;
1341
1342 return 0;
1343 }
1344
1345 osize = offsetof(Object, data.payload) + size;
1346 r = journal_file_append_object(f, OBJECT_DATA, osize, &o, &p);
1347 if (r < 0)
1348 return r;
1349
1350 o->data.hash = htole64(hash);
1351
1352 #if defined(HAVE_XZ) || defined(HAVE_LZ4)
1353 if (JOURNAL_FILE_COMPRESS(f) && size >= COMPRESSION_SIZE_THRESHOLD) {
1354 size_t rsize = 0;
1355
1356 compression = compress_blob(data, size, o->data.payload, size - 1, &rsize);
1357
1358 if (compression >= 0) {
1359 o->object.size = htole64(offsetof(Object, data.payload) + rsize);
1360 o->object.flags |= compression;
1361
1362 log_debug("Compressed data object %"PRIu64" -> %zu using %s",
1363 size, rsize, object_compressed_to_string(compression));
1364 } else
1365 /* Compression didn't work, we don't really care why, let's continue without compression */
1366 compression = 0;
1367 }
1368 #endif
1369
1370 if (compression == 0)
1371 memcpy_safe(o->data.payload, data, size);
1372
1373 r = journal_file_link_data(f, o, p, hash);
1374 if (r < 0)
1375 return r;
1376
1377 #ifdef HAVE_GCRYPT
1378 r = journal_file_hmac_put_object(f, OBJECT_DATA, o, p);
1379 if (r < 0)
1380 return r;
1381 #endif
1382
1383 /* The linking might have altered the window, so let's
1384 * refresh our pointer */
1385 r = journal_file_move_to_object(f, OBJECT_DATA, p, &o);
1386 if (r < 0)
1387 return r;
1388
1389 if (!data)
1390 eq = NULL;
1391 else
1392 eq = memchr(data, '=', size);
1393 if (eq && eq > data) {
1394 Object *fo = NULL;
1395 uint64_t fp;
1396
1397 /* Create field object ... */
1398 r = journal_file_append_field(f, data, (uint8_t*) eq - (uint8_t*) data, &fo, &fp);
1399 if (r < 0)
1400 return r;
1401
1402 /* ... and link it in. */
1403 o->data.next_field_offset = fo->field.head_data_offset;
1404 fo->field.head_data_offset = le64toh(p);
1405 }
1406
1407 if (ret)
1408 *ret = o;
1409
1410 if (offset)
1411 *offset = p;
1412
1413 return 0;
1414 }
1415
1416 uint64_t journal_file_entry_n_items(Object *o) {
1417 assert(o);
1418
1419 if (o->object.type != OBJECT_ENTRY)
1420 return 0;
1421
1422 return (le64toh(o->object.size) - offsetof(Object, entry.items)) / sizeof(EntryItem);
1423 }
1424
1425 uint64_t journal_file_entry_array_n_items(Object *o) {
1426 assert(o);
1427
1428 if (o->object.type != OBJECT_ENTRY_ARRAY)
1429 return 0;
1430
1431 return (le64toh(o->object.size) - offsetof(Object, entry_array.items)) / sizeof(uint64_t);
1432 }
1433
1434 uint64_t journal_file_hash_table_n_items(Object *o) {
1435 assert(o);
1436
1437 if (o->object.type != OBJECT_DATA_HASH_TABLE &&
1438 o->object.type != OBJECT_FIELD_HASH_TABLE)
1439 return 0;
1440
1441 return (le64toh(o->object.size) - offsetof(Object, hash_table.items)) / sizeof(HashItem);
1442 }
1443
1444 static int link_entry_into_array(JournalFile *f,
1445 le64_t *first,
1446 le64_t *idx,
1447 uint64_t p) {
1448 int r;
1449 uint64_t n = 0, ap = 0, q, i, a, hidx;
1450 Object *o;
1451
1452 assert(f);
1453 assert(f->header);
1454 assert(first);
1455 assert(idx);
1456 assert(p > 0);
1457
1458 a = le64toh(*first);
1459 i = hidx = le64toh(*idx);
1460 while (a > 0) {
1461
1462 r = journal_file_move_to_object(f, OBJECT_ENTRY_ARRAY, a, &o);
1463 if (r < 0)
1464 return r;
1465
1466 n = journal_file_entry_array_n_items(o);
1467 if (i < n) {
1468 o->entry_array.items[i] = htole64(p);
1469 *idx = htole64(hidx + 1);
1470 return 0;
1471 }
1472
1473 i -= n;
1474 ap = a;
1475 a = le64toh(o->entry_array.next_entry_array_offset);
1476 }
1477
1478 if (hidx > n)
1479 n = (hidx+1) * 2;
1480 else
1481 n = n * 2;
1482
1483 if (n < 4)
1484 n = 4;
1485
1486 r = journal_file_append_object(f, OBJECT_ENTRY_ARRAY,
1487 offsetof(Object, entry_array.items) + n * sizeof(uint64_t),
1488 &o, &q);
1489 if (r < 0)
1490 return r;
1491
1492 #ifdef HAVE_GCRYPT
1493 r = journal_file_hmac_put_object(f, OBJECT_ENTRY_ARRAY, o, q);
1494 if (r < 0)
1495 return r;
1496 #endif
1497
1498 o->entry_array.items[i] = htole64(p);
1499
1500 if (ap == 0)
1501 *first = htole64(q);
1502 else {
1503 r = journal_file_move_to_object(f, OBJECT_ENTRY_ARRAY, ap, &o);
1504 if (r < 0)
1505 return r;
1506
1507 o->entry_array.next_entry_array_offset = htole64(q);
1508 }
1509
1510 if (JOURNAL_HEADER_CONTAINS(f->header, n_entry_arrays))
1511 f->header->n_entry_arrays = htole64(le64toh(f->header->n_entry_arrays) + 1);
1512
1513 *idx = htole64(hidx + 1);
1514
1515 return 0;
1516 }
1517
1518 static int link_entry_into_array_plus_one(JournalFile *f,
1519 le64_t *extra,
1520 le64_t *first,
1521 le64_t *idx,
1522 uint64_t p) {
1523
1524 int r;
1525
1526 assert(f);
1527 assert(extra);
1528 assert(first);
1529 assert(idx);
1530 assert(p > 0);
1531
1532 if (*idx == 0)
1533 *extra = htole64(p);
1534 else {
1535 le64_t i;
1536
1537 i = htole64(le64toh(*idx) - 1);
1538 r = link_entry_into_array(f, first, &i, p);
1539 if (r < 0)
1540 return r;
1541 }
1542
1543 *idx = htole64(le64toh(*idx) + 1);
1544 return 0;
1545 }
1546
1547 static int journal_file_link_entry_item(JournalFile *f, Object *o, uint64_t offset, uint64_t i) {
1548 uint64_t p;
1549 int r;
1550 assert(f);
1551 assert(o);
1552 assert(offset > 0);
1553
1554 p = le64toh(o->entry.items[i].object_offset);
1555 if (p == 0)
1556 return -EINVAL;
1557
1558 r = journal_file_move_to_object(f, OBJECT_DATA, p, &o);
1559 if (r < 0)
1560 return r;
1561
1562 return link_entry_into_array_plus_one(f,
1563 &o->data.entry_offset,
1564 &o->data.entry_array_offset,
1565 &o->data.n_entries,
1566 offset);
1567 }
1568
1569 static int journal_file_link_entry(JournalFile *f, Object *o, uint64_t offset) {
1570 uint64_t n, i;
1571 int r;
1572
1573 assert(f);
1574 assert(f->header);
1575 assert(o);
1576 assert(offset > 0);
1577
1578 if (o->object.type != OBJECT_ENTRY)
1579 return -EINVAL;
1580
1581 __sync_synchronize();
1582
1583 /* Link up the entry itself */
1584 r = link_entry_into_array(f,
1585 &f->header->entry_array_offset,
1586 &f->header->n_entries,
1587 offset);
1588 if (r < 0)
1589 return r;
1590
1591 /* log_debug("=> %s seqnr=%"PRIu64" n_entries=%"PRIu64, f->path, o->entry.seqnum, f->header->n_entries); */
1592
1593 if (f->header->head_entry_realtime == 0)
1594 f->header->head_entry_realtime = o->entry.realtime;
1595
1596 f->header->tail_entry_realtime = o->entry.realtime;
1597 f->header->tail_entry_monotonic = o->entry.monotonic;
1598
1599 f->tail_entry_monotonic_valid = true;
1600
1601 /* Link up the items */
1602 n = journal_file_entry_n_items(o);
1603 for (i = 0; i < n; i++) {
1604 r = journal_file_link_entry_item(f, o, offset, i);
1605 if (r < 0)
1606 return r;
1607 }
1608
1609 return 0;
1610 }
1611
1612 static int journal_file_append_entry_internal(
1613 JournalFile *f,
1614 const dual_timestamp *ts,
1615 uint64_t xor_hash,
1616 const EntryItem items[], unsigned n_items,
1617 uint64_t *seqnum,
1618 Object **ret, uint64_t *offset) {
1619 uint64_t np;
1620 uint64_t osize;
1621 Object *o;
1622 int r;
1623
1624 assert(f);
1625 assert(f->header);
1626 assert(items || n_items == 0);
1627 assert(ts);
1628
1629 osize = offsetof(Object, entry.items) + (n_items * sizeof(EntryItem));
1630
1631 r = journal_file_append_object(f, OBJECT_ENTRY, osize, &o, &np);
1632 if (r < 0)
1633 return r;
1634
1635 o->entry.seqnum = htole64(journal_file_entry_seqnum(f, seqnum));
1636 memcpy_safe(o->entry.items, items, n_items * sizeof(EntryItem));
1637 o->entry.realtime = htole64(ts->realtime);
1638 o->entry.monotonic = htole64(ts->monotonic);
1639 o->entry.xor_hash = htole64(xor_hash);
1640 o->entry.boot_id = f->header->boot_id;
1641
1642 #ifdef HAVE_GCRYPT
1643 r = journal_file_hmac_put_object(f, OBJECT_ENTRY, o, np);
1644 if (r < 0)
1645 return r;
1646 #endif
1647
1648 r = journal_file_link_entry(f, o, np);
1649 if (r < 0)
1650 return r;
1651
1652 if (ret)
1653 *ret = o;
1654
1655 if (offset)
1656 *offset = np;
1657
1658 return 0;
1659 }
1660
1661 void journal_file_post_change(JournalFile *f) {
1662 assert(f);
1663
1664 /* inotify() does not receive IN_MODIFY events from file
1665 * accesses done via mmap(). After each access we hence
1666 * trigger IN_MODIFY by truncating the journal file to its
1667 * current size which triggers IN_MODIFY. */
1668
1669 __sync_synchronize();
1670
1671 if (ftruncate(f->fd, f->last_stat.st_size) < 0)
1672 log_debug_errno(errno, "Failed to truncate file to its own size: %m");
1673 }
1674
1675 static int post_change_thunk(sd_event_source *timer, uint64_t usec, void *userdata) {
1676 assert(userdata);
1677
1678 journal_file_post_change(userdata);
1679
1680 return 1;
1681 }
1682
1683 static void schedule_post_change(JournalFile *f) {
1684 sd_event_source *timer;
1685 int enabled, r;
1686 uint64_t now;
1687
1688 assert(f);
1689 assert(f->post_change_timer);
1690
1691 timer = f->post_change_timer;
1692
1693 r = sd_event_source_get_enabled(timer, &enabled);
1694 if (r < 0) {
1695 log_debug_errno(r, "Failed to get ftruncate timer state: %m");
1696 goto fail;
1697 }
1698
1699 if (enabled == SD_EVENT_ONESHOT)
1700 return;
1701
1702 r = sd_event_now(sd_event_source_get_event(timer), CLOCK_MONOTONIC, &now);
1703 if (r < 0) {
1704 log_debug_errno(r, "Failed to get clock's now for scheduling ftruncate: %m");
1705 goto fail;
1706 }
1707
1708 r = sd_event_source_set_time(timer, now+f->post_change_timer_period);
1709 if (r < 0) {
1710 log_debug_errno(r, "Failed to set time for scheduling ftruncate: %m");
1711 goto fail;
1712 }
1713
1714 r = sd_event_source_set_enabled(timer, SD_EVENT_ONESHOT);
1715 if (r < 0) {
1716 log_debug_errno(r, "Failed to enable scheduled ftruncate: %m");
1717 goto fail;
1718 }
1719
1720 return;
1721
1722 fail:
1723 /* On failure, let's simply post the change immediately. */
1724 journal_file_post_change(f);
1725 }
1726
1727 /* Enable coalesced change posting in a timer on the provided sd_event instance */
1728 int journal_file_enable_post_change_timer(JournalFile *f, sd_event *e, usec_t t) {
1729 _cleanup_(sd_event_source_unrefp) sd_event_source *timer = NULL;
1730 int r;
1731
1732 assert(f);
1733 assert_return(!f->post_change_timer, -EINVAL);
1734 assert(e);
1735 assert(t);
1736
1737 r = sd_event_add_time(e, &timer, CLOCK_MONOTONIC, 0, 0, post_change_thunk, f);
1738 if (r < 0)
1739 return r;
1740
1741 r = sd_event_source_set_enabled(timer, SD_EVENT_OFF);
1742 if (r < 0)
1743 return r;
1744
1745 f->post_change_timer = timer;
1746 timer = NULL;
1747 f->post_change_timer_period = t;
1748
1749 return r;
1750 }
1751
1752 static int entry_item_cmp(const void *_a, const void *_b) {
1753 const EntryItem *a = _a, *b = _b;
1754
1755 if (le64toh(a->object_offset) < le64toh(b->object_offset))
1756 return -1;
1757 if (le64toh(a->object_offset) > le64toh(b->object_offset))
1758 return 1;
1759 return 0;
1760 }
1761
1762 int journal_file_append_entry(JournalFile *f, const dual_timestamp *ts, const struct iovec iovec[], unsigned n_iovec, uint64_t *seqnum, Object **ret, uint64_t *offset) {
1763 unsigned i;
1764 EntryItem *items;
1765 int r;
1766 uint64_t xor_hash = 0;
1767 struct dual_timestamp _ts;
1768
1769 assert(f);
1770 assert(f->header);
1771 assert(iovec || n_iovec == 0);
1772
1773 if (!ts) {
1774 dual_timestamp_get(&_ts);
1775 ts = &_ts;
1776 }
1777
1778 #ifdef HAVE_GCRYPT
1779 r = journal_file_maybe_append_tag(f, ts->realtime);
1780 if (r < 0)
1781 return r;
1782 #endif
1783
1784 /* alloca() can't take 0, hence let's allocate at least one */
1785 items = alloca(sizeof(EntryItem) * MAX(1u, n_iovec));
1786
1787 for (i = 0; i < n_iovec; i++) {
1788 uint64_t p;
1789 Object *o;
1790
1791 r = journal_file_append_data(f, iovec[i].iov_base, iovec[i].iov_len, &o, &p);
1792 if (r < 0)
1793 return r;
1794
1795 xor_hash ^= le64toh(o->data.hash);
1796 items[i].object_offset = htole64(p);
1797 items[i].hash = o->data.hash;
1798 }
1799
1800 /* Order by the position on disk, in order to improve seek
1801 * times for rotating media. */
1802 qsort_safe(items, n_iovec, sizeof(EntryItem), entry_item_cmp);
1803
1804 r = journal_file_append_entry_internal(f, ts, xor_hash, items, n_iovec, seqnum, ret, offset);
1805
1806 /* If the memory mapping triggered a SIGBUS then we return an
1807 * IO error and ignore the error code passed down to us, since
1808 * it is very likely just an effect of a nullified replacement
1809 * mapping page */
1810
1811 if (mmap_cache_got_sigbus(f->mmap, f->fd))
1812 r = -EIO;
1813
1814 if (f->post_change_timer)
1815 schedule_post_change(f);
1816 else
1817 journal_file_post_change(f);
1818
1819 return r;
1820 }
1821
1822 typedef struct ChainCacheItem {
1823 uint64_t first; /* the array at the beginning of the chain */
1824 uint64_t array; /* the cached array */
1825 uint64_t begin; /* the first item in the cached array */
1826 uint64_t total; /* the total number of items in all arrays before this one in the chain */
1827 uint64_t last_index; /* the last index we looked at, to optimize locality when bisecting */
1828 } ChainCacheItem;
1829
1830 static void chain_cache_put(
1831 OrderedHashmap *h,
1832 ChainCacheItem *ci,
1833 uint64_t first,
1834 uint64_t array,
1835 uint64_t begin,
1836 uint64_t total,
1837 uint64_t last_index) {
1838
1839 if (!ci) {
1840 /* If the chain item to cache for this chain is the
1841 * first one it's not worth caching anything */
1842 if (array == first)
1843 return;
1844
1845 if (ordered_hashmap_size(h) >= CHAIN_CACHE_MAX) {
1846 ci = ordered_hashmap_steal_first(h);
1847 assert(ci);
1848 } else {
1849 ci = new(ChainCacheItem, 1);
1850 if (!ci)
1851 return;
1852 }
1853
1854 ci->first = first;
1855
1856 if (ordered_hashmap_put(h, &ci->first, ci) < 0) {
1857 free(ci);
1858 return;
1859 }
1860 } else
1861 assert(ci->first == first);
1862
1863 ci->array = array;
1864 ci->begin = begin;
1865 ci->total = total;
1866 ci->last_index = last_index;
1867 }
1868
1869 static int generic_array_get(
1870 JournalFile *f,
1871 uint64_t first,
1872 uint64_t i,
1873 Object **ret, uint64_t *offset) {
1874
1875 Object *o;
1876 uint64_t p = 0, a, t = 0;
1877 int r;
1878 ChainCacheItem *ci;
1879
1880 assert(f);
1881
1882 a = first;
1883
1884 /* Try the chain cache first */
1885 ci = ordered_hashmap_get(f->chain_cache, &first);
1886 if (ci && i > ci->total) {
1887 a = ci->array;
1888 i -= ci->total;
1889 t = ci->total;
1890 }
1891
1892 while (a > 0) {
1893 uint64_t k;
1894
1895 r = journal_file_move_to_object(f, OBJECT_ENTRY_ARRAY, a, &o);
1896 if (r < 0)
1897 return r;
1898
1899 k = journal_file_entry_array_n_items(o);
1900 if (i < k) {
1901 p = le64toh(o->entry_array.items[i]);
1902 goto found;
1903 }
1904
1905 i -= k;
1906 t += k;
1907 a = le64toh(o->entry_array.next_entry_array_offset);
1908 }
1909
1910 return 0;
1911
1912 found:
1913 /* Let's cache this item for the next invocation */
1914 chain_cache_put(f->chain_cache, ci, first, a, le64toh(o->entry_array.items[0]), t, i);
1915
1916 r = journal_file_move_to_object(f, OBJECT_ENTRY, p, &o);
1917 if (r < 0)
1918 return r;
1919
1920 if (ret)
1921 *ret = o;
1922
1923 if (offset)
1924 *offset = p;
1925
1926 return 1;
1927 }
1928
1929 static int generic_array_get_plus_one(
1930 JournalFile *f,
1931 uint64_t extra,
1932 uint64_t first,
1933 uint64_t i,
1934 Object **ret, uint64_t *offset) {
1935
1936 Object *o;
1937
1938 assert(f);
1939
1940 if (i == 0) {
1941 int r;
1942
1943 r = journal_file_move_to_object(f, OBJECT_ENTRY, extra, &o);
1944 if (r < 0)
1945 return r;
1946
1947 if (ret)
1948 *ret = o;
1949
1950 if (offset)
1951 *offset = extra;
1952
1953 return 1;
1954 }
1955
1956 return generic_array_get(f, first, i-1, ret, offset);
1957 }
1958
1959 enum {
1960 TEST_FOUND,
1961 TEST_LEFT,
1962 TEST_RIGHT
1963 };
1964
1965 static int generic_array_bisect(
1966 JournalFile *f,
1967 uint64_t first,
1968 uint64_t n,
1969 uint64_t needle,
1970 int (*test_object)(JournalFile *f, uint64_t p, uint64_t needle),
1971 direction_t direction,
1972 Object **ret,
1973 uint64_t *offset,
1974 uint64_t *idx) {
1975
1976 uint64_t a, p, t = 0, i = 0, last_p = 0, last_index = (uint64_t) -1;
1977 bool subtract_one = false;
1978 Object *o, *array = NULL;
1979 int r;
1980 ChainCacheItem *ci;
1981
1982 assert(f);
1983 assert(test_object);
1984
1985 /* Start with the first array in the chain */
1986 a = first;
1987
1988 ci = ordered_hashmap_get(f->chain_cache, &first);
1989 if (ci && n > ci->total) {
1990 /* Ah, we have iterated this bisection array chain
1991 * previously! Let's see if we can skip ahead in the
1992 * chain, as far as the last time. But we can't jump
1993 * backwards in the chain, so let's check that
1994 * first. */
1995
1996 r = test_object(f, ci->begin, needle);
1997 if (r < 0)
1998 return r;
1999
2000 if (r == TEST_LEFT) {
2001 /* OK, what we are looking for is right of the
2002 * begin of this EntryArray, so let's jump
2003 * straight to previously cached array in the
2004 * chain */
2005
2006 a = ci->array;
2007 n -= ci->total;
2008 t = ci->total;
2009 last_index = ci->last_index;
2010 }
2011 }
2012
2013 while (a > 0) {
2014 uint64_t left, right, k, lp;
2015
2016 r = journal_file_move_to_object(f, OBJECT_ENTRY_ARRAY, a, &array);
2017 if (r < 0)
2018 return r;
2019
2020 k = journal_file_entry_array_n_items(array);
2021 right = MIN(k, n);
2022 if (right <= 0)
2023 return 0;
2024
2025 i = right - 1;
2026 lp = p = le64toh(array->entry_array.items[i]);
2027 if (p <= 0)
2028 r = -EBADMSG;
2029 else
2030 r = test_object(f, p, needle);
2031 if (r == -EBADMSG) {
2032 log_debug_errno(r, "Encountered invalid entry while bisecting, cutting algorithm short. (1)");
2033 n = i;
2034 continue;
2035 }
2036 if (r < 0)
2037 return r;
2038
2039 if (r == TEST_FOUND)
2040 r = direction == DIRECTION_DOWN ? TEST_RIGHT : TEST_LEFT;
2041
2042 if (r == TEST_RIGHT) {
2043 left = 0;
2044 right -= 1;
2045
2046 if (last_index != (uint64_t) -1) {
2047 assert(last_index <= right);
2048
2049 /* If we cached the last index we
2050 * looked at, let's try to not to jump
2051 * too wildly around and see if we can
2052 * limit the range to look at early to
2053 * the immediate neighbors of the last
2054 * index we looked at. */
2055
2056 if (last_index > 0) {
2057 uint64_t x = last_index - 1;
2058
2059 p = le64toh(array->entry_array.items[x]);
2060 if (p <= 0)
2061 return -EBADMSG;
2062
2063 r = test_object(f, p, needle);
2064 if (r < 0)
2065 return r;
2066
2067 if (r == TEST_FOUND)
2068 r = direction == DIRECTION_DOWN ? TEST_RIGHT : TEST_LEFT;
2069
2070 if (r == TEST_RIGHT)
2071 right = x;
2072 else
2073 left = x + 1;
2074 }
2075
2076 if (last_index < right) {
2077 uint64_t y = last_index + 1;
2078
2079 p = le64toh(array->entry_array.items[y]);
2080 if (p <= 0)
2081 return -EBADMSG;
2082
2083 r = test_object(f, p, needle);
2084 if (r < 0)
2085 return r;
2086
2087 if (r == TEST_FOUND)
2088 r = direction == DIRECTION_DOWN ? TEST_RIGHT : TEST_LEFT;
2089
2090 if (r == TEST_RIGHT)
2091 right = y;
2092 else
2093 left = y + 1;
2094 }
2095 }
2096
2097 for (;;) {
2098 if (left == right) {
2099 if (direction == DIRECTION_UP)
2100 subtract_one = true;
2101
2102 i = left;
2103 goto found;
2104 }
2105
2106 assert(left < right);
2107 i = (left + right) / 2;
2108
2109 p = le64toh(array->entry_array.items[i]);
2110 if (p <= 0)
2111 r = -EBADMSG;
2112 else
2113 r = test_object(f, p, needle);
2114 if (r == -EBADMSG) {
2115 log_debug_errno(r, "Encountered invalid entry while bisecting, cutting algorithm short. (2)");
2116 right = n = i;
2117 continue;
2118 }
2119 if (r < 0)
2120 return r;
2121
2122 if (r == TEST_FOUND)
2123 r = direction == DIRECTION_DOWN ? TEST_RIGHT : TEST_LEFT;
2124
2125 if (r == TEST_RIGHT)
2126 right = i;
2127 else
2128 left = i + 1;
2129 }
2130 }
2131
2132 if (k >= n) {
2133 if (direction == DIRECTION_UP) {
2134 i = n;
2135 subtract_one = true;
2136 goto found;
2137 }
2138
2139 return 0;
2140 }
2141
2142 last_p = lp;
2143
2144 n -= k;
2145 t += k;
2146 last_index = (uint64_t) -1;
2147 a = le64toh(array->entry_array.next_entry_array_offset);
2148 }
2149
2150 return 0;
2151
2152 found:
2153 if (subtract_one && t == 0 && i == 0)
2154 return 0;
2155
2156 /* Let's cache this item for the next invocation */
2157 chain_cache_put(f->chain_cache, ci, first, a, le64toh(array->entry_array.items[0]), t, subtract_one ? (i > 0 ? i-1 : (uint64_t) -1) : i);
2158
2159 if (subtract_one && i == 0)
2160 p = last_p;
2161 else if (subtract_one)
2162 p = le64toh(array->entry_array.items[i-1]);
2163 else
2164 p = le64toh(array->entry_array.items[i]);
2165
2166 r = journal_file_move_to_object(f, OBJECT_ENTRY, p, &o);
2167 if (r < 0)
2168 return r;
2169
2170 if (ret)
2171 *ret = o;
2172
2173 if (offset)
2174 *offset = p;
2175
2176 if (idx)
2177 *idx = t + i + (subtract_one ? -1 : 0);
2178
2179 return 1;
2180 }
2181
2182 static int generic_array_bisect_plus_one(
2183 JournalFile *f,
2184 uint64_t extra,
2185 uint64_t first,
2186 uint64_t n,
2187 uint64_t needle,
2188 int (*test_object)(JournalFile *f, uint64_t p, uint64_t needle),
2189 direction_t direction,
2190 Object **ret,
2191 uint64_t *offset,
2192 uint64_t *idx) {
2193
2194 int r;
2195 bool step_back = false;
2196 Object *o;
2197
2198 assert(f);
2199 assert(test_object);
2200
2201 if (n <= 0)
2202 return 0;
2203
2204 /* This bisects the array in object 'first', but first checks
2205 * an extra */
2206 r = test_object(f, extra, needle);
2207 if (r < 0)
2208 return r;
2209
2210 if (r == TEST_FOUND)
2211 r = direction == DIRECTION_DOWN ? TEST_RIGHT : TEST_LEFT;
2212
2213 /* if we are looking with DIRECTION_UP then we need to first
2214 see if in the actual array there is a matching entry, and
2215 return the last one of that. But if there isn't any we need
2216 to return this one. Hence remember this, and return it
2217 below. */
2218 if (r == TEST_LEFT)
2219 step_back = direction == DIRECTION_UP;
2220
2221 if (r == TEST_RIGHT) {
2222 if (direction == DIRECTION_DOWN)
2223 goto found;
2224 else
2225 return 0;
2226 }
2227
2228 r = generic_array_bisect(f, first, n-1, needle, test_object, direction, ret, offset, idx);
2229
2230 if (r == 0 && step_back)
2231 goto found;
2232
2233 if (r > 0 && idx)
2234 (*idx)++;
2235
2236 return r;
2237
2238 found:
2239 r = journal_file_move_to_object(f, OBJECT_ENTRY, extra, &o);
2240 if (r < 0)
2241 return r;
2242
2243 if (ret)
2244 *ret = o;
2245
2246 if (offset)
2247 *offset = extra;
2248
2249 if (idx)
2250 *idx = 0;
2251
2252 return 1;
2253 }
2254
2255 _pure_ static int test_object_offset(JournalFile *f, uint64_t p, uint64_t needle) {
2256 assert(f);
2257 assert(p > 0);
2258
2259 if (p == needle)
2260 return TEST_FOUND;
2261 else if (p < needle)
2262 return TEST_LEFT;
2263 else
2264 return TEST_RIGHT;
2265 }
2266
2267 static int test_object_seqnum(JournalFile *f, uint64_t p, uint64_t needle) {
2268 Object *o;
2269 int r;
2270
2271 assert(f);
2272 assert(p > 0);
2273
2274 r = journal_file_move_to_object(f, OBJECT_ENTRY, p, &o);
2275 if (r < 0)
2276 return r;
2277
2278 if (le64toh(o->entry.seqnum) == needle)
2279 return TEST_FOUND;
2280 else if (le64toh(o->entry.seqnum) < needle)
2281 return TEST_LEFT;
2282 else
2283 return TEST_RIGHT;
2284 }
2285
2286 int journal_file_move_to_entry_by_seqnum(
2287 JournalFile *f,
2288 uint64_t seqnum,
2289 direction_t direction,
2290 Object **ret,
2291 uint64_t *offset) {
2292 assert(f);
2293 assert(f->header);
2294
2295 return generic_array_bisect(f,
2296 le64toh(f->header->entry_array_offset),
2297 le64toh(f->header->n_entries),
2298 seqnum,
2299 test_object_seqnum,
2300 direction,
2301 ret, offset, NULL);
2302 }
2303
2304 static int test_object_realtime(JournalFile *f, uint64_t p, uint64_t needle) {
2305 Object *o;
2306 int r;
2307
2308 assert(f);
2309 assert(p > 0);
2310
2311 r = journal_file_move_to_object(f, OBJECT_ENTRY, p, &o);
2312 if (r < 0)
2313 return r;
2314
2315 if (le64toh(o->entry.realtime) == needle)
2316 return TEST_FOUND;
2317 else if (le64toh(o->entry.realtime) < needle)
2318 return TEST_LEFT;
2319 else
2320 return TEST_RIGHT;
2321 }
2322
2323 int journal_file_move_to_entry_by_realtime(
2324 JournalFile *f,
2325 uint64_t realtime,
2326 direction_t direction,
2327 Object **ret,
2328 uint64_t *offset) {
2329 assert(f);
2330 assert(f->header);
2331
2332 return generic_array_bisect(f,
2333 le64toh(f->header->entry_array_offset),
2334 le64toh(f->header->n_entries),
2335 realtime,
2336 test_object_realtime,
2337 direction,
2338 ret, offset, NULL);
2339 }
2340
2341 static int test_object_monotonic(JournalFile *f, uint64_t p, uint64_t needle) {
2342 Object *o;
2343 int r;
2344
2345 assert(f);
2346 assert(p > 0);
2347
2348 r = journal_file_move_to_object(f, OBJECT_ENTRY, p, &o);
2349 if (r < 0)
2350 return r;
2351
2352 if (le64toh(o->entry.monotonic) == needle)
2353 return TEST_FOUND;
2354 else if (le64toh(o->entry.monotonic) < needle)
2355 return TEST_LEFT;
2356 else
2357 return TEST_RIGHT;
2358 }
2359
2360 static int find_data_object_by_boot_id(
2361 JournalFile *f,
2362 sd_id128_t boot_id,
2363 Object **o,
2364 uint64_t *b) {
2365
2366 char t[sizeof("_BOOT_ID=")-1 + 32 + 1] = "_BOOT_ID=";
2367
2368 sd_id128_to_string(boot_id, t + 9);
2369 return journal_file_find_data_object(f, t, sizeof(t) - 1, o, b);
2370 }
2371
2372 int journal_file_move_to_entry_by_monotonic(
2373 JournalFile *f,
2374 sd_id128_t boot_id,
2375 uint64_t monotonic,
2376 direction_t direction,
2377 Object **ret,
2378 uint64_t *offset) {
2379
2380 Object *o;
2381 int r;
2382
2383 assert(f);
2384
2385 r = find_data_object_by_boot_id(f, boot_id, &o, NULL);
2386 if (r < 0)
2387 return r;
2388 if (r == 0)
2389 return -ENOENT;
2390
2391 return generic_array_bisect_plus_one(f,
2392 le64toh(o->data.entry_offset),
2393 le64toh(o->data.entry_array_offset),
2394 le64toh(o->data.n_entries),
2395 monotonic,
2396 test_object_monotonic,
2397 direction,
2398 ret, offset, NULL);
2399 }
2400
2401 void journal_file_reset_location(JournalFile *f) {
2402 f->location_type = LOCATION_HEAD;
2403 f->current_offset = 0;
2404 f->current_seqnum = 0;
2405 f->current_realtime = 0;
2406 f->current_monotonic = 0;
2407 zero(f->current_boot_id);
2408 f->current_xor_hash = 0;
2409 }
2410
2411 void journal_file_save_location(JournalFile *f, Object *o, uint64_t offset) {
2412 f->location_type = LOCATION_SEEK;
2413 f->current_offset = offset;
2414 f->current_seqnum = le64toh(o->entry.seqnum);
2415 f->current_realtime = le64toh(o->entry.realtime);
2416 f->current_monotonic = le64toh(o->entry.monotonic);
2417 f->current_boot_id = o->entry.boot_id;
2418 f->current_xor_hash = le64toh(o->entry.xor_hash);
2419 }
2420
2421 int journal_file_compare_locations(JournalFile *af, JournalFile *bf) {
2422 assert(af);
2423 assert(af->header);
2424 assert(bf);
2425 assert(bf->header);
2426 assert(af->location_type == LOCATION_SEEK);
2427 assert(bf->location_type == LOCATION_SEEK);
2428
2429 /* If contents and timestamps match, these entries are
2430 * identical, even if the seqnum does not match */
2431 if (sd_id128_equal(af->current_boot_id, bf->current_boot_id) &&
2432 af->current_monotonic == bf->current_monotonic &&
2433 af->current_realtime == bf->current_realtime &&
2434 af->current_xor_hash == bf->current_xor_hash)
2435 return 0;
2436
2437 if (sd_id128_equal(af->header->seqnum_id, bf->header->seqnum_id)) {
2438
2439 /* If this is from the same seqnum source, compare
2440 * seqnums */
2441 if (af->current_seqnum < bf->current_seqnum)
2442 return -1;
2443 if (af->current_seqnum > bf->current_seqnum)
2444 return 1;
2445
2446 /* Wow! This is weird, different data but the same
2447 * seqnums? Something is borked, but let's make the
2448 * best of it and compare by time. */
2449 }
2450
2451 if (sd_id128_equal(af->current_boot_id, bf->current_boot_id)) {
2452
2453 /* If the boot id matches, compare monotonic time */
2454 if (af->current_monotonic < bf->current_monotonic)
2455 return -1;
2456 if (af->current_monotonic > bf->current_monotonic)
2457 return 1;
2458 }
2459
2460 /* Otherwise, compare UTC time */
2461 if (af->current_realtime < bf->current_realtime)
2462 return -1;
2463 if (af->current_realtime > bf->current_realtime)
2464 return 1;
2465
2466 /* Finally, compare by contents */
2467 if (af->current_xor_hash < bf->current_xor_hash)
2468 return -1;
2469 if (af->current_xor_hash > bf->current_xor_hash)
2470 return 1;
2471
2472 return 0;
2473 }
2474
2475 int journal_file_next_entry(
2476 JournalFile *f,
2477 uint64_t p,
2478 direction_t direction,
2479 Object **ret, uint64_t *offset) {
2480
2481 uint64_t i, n, ofs;
2482 int r;
2483
2484 assert(f);
2485 assert(f->header);
2486
2487 n = le64toh(f->header->n_entries);
2488 if (n <= 0)
2489 return 0;
2490
2491 if (p == 0)
2492 i = direction == DIRECTION_DOWN ? 0 : n - 1;
2493 else {
2494 r = generic_array_bisect(f,
2495 le64toh(f->header->entry_array_offset),
2496 le64toh(f->header->n_entries),
2497 p,
2498 test_object_offset,
2499 DIRECTION_DOWN,
2500 NULL, NULL,
2501 &i);
2502 if (r <= 0)
2503 return r;
2504
2505 if (direction == DIRECTION_DOWN) {
2506 if (i >= n - 1)
2507 return 0;
2508
2509 i++;
2510 } else {
2511 if (i <= 0)
2512 return 0;
2513
2514 i--;
2515 }
2516 }
2517
2518 /* And jump to it */
2519 r = generic_array_get(f,
2520 le64toh(f->header->entry_array_offset),
2521 i,
2522 ret, &ofs);
2523 if (r == -EBADMSG && direction == DIRECTION_DOWN) {
2524 /* Special case: when we iterate throught the journal file linearly, and hit an entry we can't read,
2525 * consider this the end of the journal file. */
2526 log_debug_errno(r, "Encountered entry we can't read while iterating through journal file. Considering this the end of the file.");
2527 return 0;
2528 }
2529 if (r <= 0)
2530 return r;
2531
2532 if (p > 0 &&
2533 (direction == DIRECTION_DOWN ? ofs <= p : ofs >= p)) {
2534 log_debug("%s: entry array corrupted at entry %" PRIu64, f->path, i);
2535 return -EBADMSG;
2536 }
2537
2538 if (offset)
2539 *offset = ofs;
2540
2541 return 1;
2542 }
2543
2544 int journal_file_next_entry_for_data(
2545 JournalFile *f,
2546 Object *o, uint64_t p,
2547 uint64_t data_offset,
2548 direction_t direction,
2549 Object **ret, uint64_t *offset) {
2550
2551 uint64_t n, i;
2552 int r;
2553 Object *d;
2554
2555 assert(f);
2556 assert(p > 0 || !o);
2557
2558 r = journal_file_move_to_object(f, OBJECT_DATA, data_offset, &d);
2559 if (r < 0)
2560 return r;
2561
2562 n = le64toh(d->data.n_entries);
2563 if (n <= 0)
2564 return n;
2565
2566 if (!o)
2567 i = direction == DIRECTION_DOWN ? 0 : n - 1;
2568 else {
2569 if (o->object.type != OBJECT_ENTRY)
2570 return -EINVAL;
2571
2572 r = generic_array_bisect_plus_one(f,
2573 le64toh(d->data.entry_offset),
2574 le64toh(d->data.entry_array_offset),
2575 le64toh(d->data.n_entries),
2576 p,
2577 test_object_offset,
2578 DIRECTION_DOWN,
2579 NULL, NULL,
2580 &i);
2581
2582 if (r <= 0)
2583 return r;
2584
2585 if (direction == DIRECTION_DOWN) {
2586 if (i >= n - 1)
2587 return 0;
2588
2589 i++;
2590 } else {
2591 if (i <= 0)
2592 return 0;
2593
2594 i--;
2595 }
2596
2597 }
2598
2599 return generic_array_get_plus_one(f,
2600 le64toh(d->data.entry_offset),
2601 le64toh(d->data.entry_array_offset),
2602 i,
2603 ret, offset);
2604 }
2605
2606 int journal_file_move_to_entry_by_offset_for_data(
2607 JournalFile *f,
2608 uint64_t data_offset,
2609 uint64_t p,
2610 direction_t direction,
2611 Object **ret, uint64_t *offset) {
2612
2613 int r;
2614 Object *d;
2615
2616 assert(f);
2617
2618 r = journal_file_move_to_object(f, OBJECT_DATA, data_offset, &d);
2619 if (r < 0)
2620 return r;
2621
2622 return generic_array_bisect_plus_one(f,
2623 le64toh(d->data.entry_offset),
2624 le64toh(d->data.entry_array_offset),
2625 le64toh(d->data.n_entries),
2626 p,
2627 test_object_offset,
2628 direction,
2629 ret, offset, NULL);
2630 }
2631
2632 int journal_file_move_to_entry_by_monotonic_for_data(
2633 JournalFile *f,
2634 uint64_t data_offset,
2635 sd_id128_t boot_id,
2636 uint64_t monotonic,
2637 direction_t direction,
2638 Object **ret, uint64_t *offset) {
2639
2640 Object *o, *d;
2641 int r;
2642 uint64_t b, z;
2643
2644 assert(f);
2645
2646 /* First, seek by time */
2647 r = find_data_object_by_boot_id(f, boot_id, &o, &b);
2648 if (r < 0)
2649 return r;
2650 if (r == 0)
2651 return -ENOENT;
2652
2653 r = generic_array_bisect_plus_one(f,
2654 le64toh(o->data.entry_offset),
2655 le64toh(o->data.entry_array_offset),
2656 le64toh(o->data.n_entries),
2657 monotonic,
2658 test_object_monotonic,
2659 direction,
2660 NULL, &z, NULL);
2661 if (r <= 0)
2662 return r;
2663
2664 /* And now, continue seeking until we find an entry that
2665 * exists in both bisection arrays */
2666
2667 for (;;) {
2668 Object *qo;
2669 uint64_t p, q;
2670
2671 r = journal_file_move_to_object(f, OBJECT_DATA, data_offset, &d);
2672 if (r < 0)
2673 return r;
2674
2675 r = generic_array_bisect_plus_one(f,
2676 le64toh(d->data.entry_offset),
2677 le64toh(d->data.entry_array_offset),
2678 le64toh(d->data.n_entries),
2679 z,
2680 test_object_offset,
2681 direction,
2682 NULL, &p, NULL);
2683 if (r <= 0)
2684 return r;
2685
2686 r = journal_file_move_to_object(f, OBJECT_DATA, b, &o);
2687 if (r < 0)
2688 return r;
2689
2690 r = generic_array_bisect_plus_one(f,
2691 le64toh(o->data.entry_offset),
2692 le64toh(o->data.entry_array_offset),
2693 le64toh(o->data.n_entries),
2694 p,
2695 test_object_offset,
2696 direction,
2697 &qo, &q, NULL);
2698
2699 if (r <= 0)
2700 return r;
2701
2702 if (p == q) {
2703 if (ret)
2704 *ret = qo;
2705 if (offset)
2706 *offset = q;
2707
2708 return 1;
2709 }
2710
2711 z = q;
2712 }
2713 }
2714
2715 int journal_file_move_to_entry_by_seqnum_for_data(
2716 JournalFile *f,
2717 uint64_t data_offset,
2718 uint64_t seqnum,
2719 direction_t direction,
2720 Object **ret, uint64_t *offset) {
2721
2722 Object *d;
2723 int r;
2724
2725 assert(f);
2726
2727 r = journal_file_move_to_object(f, OBJECT_DATA, data_offset, &d);
2728 if (r < 0)
2729 return r;
2730
2731 return generic_array_bisect_plus_one(f,
2732 le64toh(d->data.entry_offset),
2733 le64toh(d->data.entry_array_offset),
2734 le64toh(d->data.n_entries),
2735 seqnum,
2736 test_object_seqnum,
2737 direction,
2738 ret, offset, NULL);
2739 }
2740
2741 int journal_file_move_to_entry_by_realtime_for_data(
2742 JournalFile *f,
2743 uint64_t data_offset,
2744 uint64_t realtime,
2745 direction_t direction,
2746 Object **ret, uint64_t *offset) {
2747
2748 Object *d;
2749 int r;
2750
2751 assert(f);
2752
2753 r = journal_file_move_to_object(f, OBJECT_DATA, data_offset, &d);
2754 if (r < 0)
2755 return r;
2756
2757 return generic_array_bisect_plus_one(f,
2758 le64toh(d->data.entry_offset),
2759 le64toh(d->data.entry_array_offset),
2760 le64toh(d->data.n_entries),
2761 realtime,
2762 test_object_realtime,
2763 direction,
2764 ret, offset, NULL);
2765 }
2766
2767 void journal_file_dump(JournalFile *f) {
2768 Object *o;
2769 int r;
2770 uint64_t p;
2771
2772 assert(f);
2773 assert(f->header);
2774
2775 journal_file_print_header(f);
2776
2777 p = le64toh(f->header->header_size);
2778 while (p != 0) {
2779 r = journal_file_move_to_object(f, OBJECT_UNUSED, p, &o);
2780 if (r < 0)
2781 goto fail;
2782
2783 switch (o->object.type) {
2784
2785 case OBJECT_UNUSED:
2786 printf("Type: OBJECT_UNUSED\n");
2787 break;
2788
2789 case OBJECT_DATA:
2790 printf("Type: OBJECT_DATA\n");
2791 break;
2792
2793 case OBJECT_FIELD:
2794 printf("Type: OBJECT_FIELD\n");
2795 break;
2796
2797 case OBJECT_ENTRY:
2798 printf("Type: OBJECT_ENTRY seqnum=%"PRIu64" monotonic=%"PRIu64" realtime=%"PRIu64"\n",
2799 le64toh(o->entry.seqnum),
2800 le64toh(o->entry.monotonic),
2801 le64toh(o->entry.realtime));
2802 break;
2803
2804 case OBJECT_FIELD_HASH_TABLE:
2805 printf("Type: OBJECT_FIELD_HASH_TABLE\n");
2806 break;
2807
2808 case OBJECT_DATA_HASH_TABLE:
2809 printf("Type: OBJECT_DATA_HASH_TABLE\n");
2810 break;
2811
2812 case OBJECT_ENTRY_ARRAY:
2813 printf("Type: OBJECT_ENTRY_ARRAY\n");
2814 break;
2815
2816 case OBJECT_TAG:
2817 printf("Type: OBJECT_TAG seqnum=%"PRIu64" epoch=%"PRIu64"\n",
2818 le64toh(o->tag.seqnum),
2819 le64toh(o->tag.epoch));
2820 break;
2821
2822 default:
2823 printf("Type: unknown (%i)\n", o->object.type);
2824 break;
2825 }
2826
2827 if (o->object.flags & OBJECT_COMPRESSION_MASK)
2828 printf("Flags: %s\n",
2829 object_compressed_to_string(o->object.flags & OBJECT_COMPRESSION_MASK));
2830
2831 if (p == le64toh(f->header->tail_object_offset))
2832 p = 0;
2833 else
2834 p = p + ALIGN64(le64toh(o->object.size));
2835 }
2836
2837 return;
2838 fail:
2839 log_error("File corrupt");
2840 }
2841
2842 static const char* format_timestamp_safe(char *buf, size_t l, usec_t t) {
2843 const char *x;
2844
2845 x = format_timestamp(buf, l, t);
2846 if (x)
2847 return x;
2848 return " --- ";
2849 }
2850
2851 void journal_file_print_header(JournalFile *f) {
2852 char a[33], b[33], c[33], d[33];
2853 char x[FORMAT_TIMESTAMP_MAX], y[FORMAT_TIMESTAMP_MAX], z[FORMAT_TIMESTAMP_MAX];
2854 struct stat st;
2855 char bytes[FORMAT_BYTES_MAX];
2856
2857 assert(f);
2858 assert(f->header);
2859
2860 printf("File Path: %s\n"
2861 "File ID: %s\n"
2862 "Machine ID: %s\n"
2863 "Boot ID: %s\n"
2864 "Sequential Number ID: %s\n"
2865 "State: %s\n"
2866 "Compatible Flags:%s%s\n"
2867 "Incompatible Flags:%s%s%s\n"
2868 "Header size: %"PRIu64"\n"
2869 "Arena size: %"PRIu64"\n"
2870 "Data Hash Table Size: %"PRIu64"\n"
2871 "Field Hash Table Size: %"PRIu64"\n"
2872 "Rotate Suggested: %s\n"
2873 "Head Sequential Number: %"PRIu64" (%"PRIx64")\n"
2874 "Tail Sequential Number: %"PRIu64" (%"PRIx64")\n"
2875 "Head Realtime Timestamp: %s (%"PRIx64")\n"
2876 "Tail Realtime Timestamp: %s (%"PRIx64")\n"
2877 "Tail Monotonic Timestamp: %s (%"PRIx64")\n"
2878 "Objects: %"PRIu64"\n"
2879 "Entry Objects: %"PRIu64"\n",
2880 f->path,
2881 sd_id128_to_string(f->header->file_id, a),
2882 sd_id128_to_string(f->header->machine_id, b),
2883 sd_id128_to_string(f->header->boot_id, c),
2884 sd_id128_to_string(f->header->seqnum_id, d),
2885 f->header->state == STATE_OFFLINE ? "OFFLINE" :
2886 f->header->state == STATE_ONLINE ? "ONLINE" :
2887 f->header->state == STATE_ARCHIVED ? "ARCHIVED" : "UNKNOWN",
2888 JOURNAL_HEADER_SEALED(f->header) ? " SEALED" : "",
2889 (le32toh(f->header->compatible_flags) & ~HEADER_COMPATIBLE_ANY) ? " ???" : "",
2890 JOURNAL_HEADER_COMPRESSED_XZ(f->header) ? " COMPRESSED-XZ" : "",
2891 JOURNAL_HEADER_COMPRESSED_LZ4(f->header) ? " COMPRESSED-LZ4" : "",
2892 (le32toh(f->header->incompatible_flags) & ~HEADER_INCOMPATIBLE_ANY) ? " ???" : "",
2893 le64toh(f->header->header_size),
2894 le64toh(f->header->arena_size),
2895 le64toh(f->header->data_hash_table_size) / sizeof(HashItem),
2896 le64toh(f->header->field_hash_table_size) / sizeof(HashItem),
2897 yes_no(journal_file_rotate_suggested(f, 0)),
2898 le64toh(f->header->head_entry_seqnum), le64toh(f->header->head_entry_seqnum),
2899 le64toh(f->header->tail_entry_seqnum), le64toh(f->header->tail_entry_seqnum),
2900 format_timestamp_safe(x, sizeof(x), le64toh(f->header->head_entry_realtime)), le64toh(f->header->head_entry_realtime),
2901 format_timestamp_safe(y, sizeof(y), le64toh(f->header->tail_entry_realtime)), le64toh(f->header->tail_entry_realtime),
2902 format_timespan(z, sizeof(z), le64toh(f->header->tail_entry_monotonic), USEC_PER_MSEC), le64toh(f->header->tail_entry_monotonic),
2903 le64toh(f->header->n_objects),
2904 le64toh(f->header->n_entries));
2905
2906 if (JOURNAL_HEADER_CONTAINS(f->header, n_data))
2907 printf("Data Objects: %"PRIu64"\n"
2908 "Data Hash Table Fill: %.1f%%\n",
2909 le64toh(f->header->n_data),
2910 100.0 * (double) le64toh(f->header->n_data) / ((double) (le64toh(f->header->data_hash_table_size) / sizeof(HashItem))));
2911
2912 if (JOURNAL_HEADER_CONTAINS(f->header, n_fields))
2913 printf("Field Objects: %"PRIu64"\n"
2914 "Field Hash Table Fill: %.1f%%\n",
2915 le64toh(f->header->n_fields),
2916 100.0 * (double) le64toh(f->header->n_fields) / ((double) (le64toh(f->header->field_hash_table_size) / sizeof(HashItem))));
2917
2918 if (JOURNAL_HEADER_CONTAINS(f->header, n_tags))
2919 printf("Tag Objects: %"PRIu64"\n",
2920 le64toh(f->header->n_tags));
2921 if (JOURNAL_HEADER_CONTAINS(f->header, n_entry_arrays))
2922 printf("Entry Array Objects: %"PRIu64"\n",
2923 le64toh(f->header->n_entry_arrays));
2924
2925 if (fstat(f->fd, &st) >= 0)
2926 printf("Disk usage: %s\n", format_bytes(bytes, sizeof(bytes), (uint64_t) st.st_blocks * 512ULL));
2927 }
2928
2929 static int journal_file_warn_btrfs(JournalFile *f) {
2930 unsigned attrs;
2931 int r;
2932
2933 assert(f);
2934
2935 /* Before we write anything, check if the COW logic is turned
2936 * off on btrfs. Given our write pattern that is quite
2937 * unfriendly to COW file systems this should greatly improve
2938 * performance on COW file systems, such as btrfs, at the
2939 * expense of data integrity features (which shouldn't be too
2940 * bad, given that we do our own checksumming). */
2941
2942 r = btrfs_is_filesystem(f->fd);
2943 if (r < 0)
2944 return log_warning_errno(r, "Failed to determine if journal is on btrfs: %m");
2945 if (!r)
2946 return 0;
2947
2948 r = read_attr_fd(f->fd, &attrs);
2949 if (r < 0)
2950 return log_warning_errno(r, "Failed to read file attributes: %m");
2951
2952 if (attrs & FS_NOCOW_FL) {
2953 log_debug("Detected btrfs file system with copy-on-write disabled, all is good.");
2954 return 0;
2955 }
2956
2957 log_notice("Creating journal file %s on a btrfs file system, and copy-on-write is enabled. "
2958 "This is likely to slow down journal access substantially, please consider turning "
2959 "off the copy-on-write file attribute on the journal directory, using chattr +C.", f->path);
2960
2961 return 1;
2962 }
2963
2964 int journal_file_open(
2965 int fd,
2966 const char *fname,
2967 int flags,
2968 mode_t mode,
2969 bool compress,
2970 bool seal,
2971 JournalMetrics *metrics,
2972 MMapCache *mmap_cache,
2973 Set *deferred_closes,
2974 JournalFile *template,
2975 JournalFile **ret) {
2976
2977 bool newly_created = false;
2978 JournalFile *f;
2979 void *h;
2980 int r;
2981
2982 assert(ret);
2983 assert(fd >= 0 || fname);
2984
2985 if ((flags & O_ACCMODE) != O_RDONLY &&
2986 (flags & O_ACCMODE) != O_RDWR)
2987 return -EINVAL;
2988
2989 if (fname) {
2990 if (!endswith(fname, ".journal") &&
2991 !endswith(fname, ".journal~"))
2992 return -EINVAL;
2993 }
2994
2995 f = new0(JournalFile, 1);
2996 if (!f)
2997 return -ENOMEM;
2998
2999 f->fd = fd;
3000 f->mode = mode;
3001
3002 f->flags = flags;
3003 f->prot = prot_from_flags(flags);
3004 f->writable = (flags & O_ACCMODE) != O_RDONLY;
3005 #if defined(HAVE_LZ4)
3006 f->compress_lz4 = compress;
3007 #elif defined(HAVE_XZ)
3008 f->compress_xz = compress;
3009 #endif
3010 #ifdef HAVE_GCRYPT
3011 f->seal = seal;
3012 #endif
3013
3014 if (mmap_cache)
3015 f->mmap = mmap_cache_ref(mmap_cache);
3016 else {
3017 f->mmap = mmap_cache_new();
3018 if (!f->mmap) {
3019 r = -ENOMEM;
3020 goto fail;
3021 }
3022 }
3023
3024 if (fname)
3025 f->path = strdup(fname);
3026 else /* If we don't know the path, fill in something explanatory and vaguely useful */
3027 asprintf(&f->path, "/proc/self/%i", fd);
3028 if (!f->path) {
3029 r = -ENOMEM;
3030 goto fail;
3031 }
3032
3033 f->chain_cache = ordered_hashmap_new(&uint64_hash_ops);
3034 if (!f->chain_cache) {
3035 r = -ENOMEM;
3036 goto fail;
3037 }
3038
3039 if (f->fd < 0) {
3040 f->fd = open(f->path, f->flags|O_CLOEXEC, f->mode);
3041 if (f->fd < 0) {
3042 r = -errno;
3043 goto fail;
3044 }
3045
3046 /* fds we opened here by us should also be closed by us. */
3047 f->close_fd = true;
3048 }
3049
3050 r = journal_file_fstat(f);
3051 if (r < 0)
3052 goto fail;
3053
3054 if (f->last_stat.st_size == 0 && f->writable) {
3055
3056 (void) journal_file_warn_btrfs(f);
3057
3058 /* Let's attach the creation time to the journal file,
3059 * so that the vacuuming code knows the age of this
3060 * file even if the file might end up corrupted one
3061 * day... Ideally we'd just use the creation time many
3062 * file systems maintain for each file, but there is
3063 * currently no usable API to query this, hence let's
3064 * emulate this via extended attributes. If extended
3065 * attributes are not supported we'll just skip this,
3066 * and rely solely on mtime/atime/ctime of the file. */
3067
3068 fd_setcrtime(f->fd, 0);
3069
3070 #ifdef HAVE_GCRYPT
3071 /* Try to load the FSPRG state, and if we can't, then
3072 * just don't do sealing */
3073 if (f->seal) {
3074 r = journal_file_fss_load(f);
3075 if (r < 0)
3076 f->seal = false;
3077 }
3078 #endif
3079
3080 r = journal_file_init_header(f, template);
3081 if (r < 0)
3082 goto fail;
3083
3084 r = journal_file_fstat(f);
3085 if (r < 0)
3086 goto fail;
3087
3088 newly_created = true;
3089 }
3090
3091 if (f->last_stat.st_size < (off_t) HEADER_SIZE_MIN) {
3092 r = -ENODATA;
3093 goto fail;
3094 }
3095
3096 r = mmap_cache_get(f->mmap, f->fd, f->prot, CONTEXT_HEADER, true, 0, PAGE_ALIGN(sizeof(Header)), &f->last_stat, &h);
3097 if (r < 0)
3098 goto fail;
3099
3100 f->header = h;
3101
3102 if (!newly_created) {
3103 if (deferred_closes)
3104 journal_file_close_set(deferred_closes);
3105
3106 r = journal_file_verify_header(f);
3107 if (r < 0)
3108 goto fail;
3109 }
3110
3111 #ifdef HAVE_GCRYPT
3112 if (!newly_created && f->writable) {
3113 r = journal_file_fss_load(f);
3114 if (r < 0)
3115 goto fail;
3116 }
3117 #endif
3118
3119 if (f->writable) {
3120 if (metrics) {
3121 journal_default_metrics(metrics, f->fd);
3122 f->metrics = *metrics;
3123 } else if (template)
3124 f->metrics = template->metrics;
3125
3126 r = journal_file_refresh_header(f);
3127 if (r < 0)
3128 goto fail;
3129 }
3130
3131 #ifdef HAVE_GCRYPT
3132 r = journal_file_hmac_setup(f);
3133 if (r < 0)
3134 goto fail;
3135 #endif
3136
3137 if (newly_created) {
3138 r = journal_file_setup_field_hash_table(f);
3139 if (r < 0)
3140 goto fail;
3141
3142 r = journal_file_setup_data_hash_table(f);
3143 if (r < 0)
3144 goto fail;
3145
3146 #ifdef HAVE_GCRYPT
3147 r = journal_file_append_first_tag(f);
3148 if (r < 0)
3149 goto fail;
3150 #endif
3151 }
3152
3153 if (mmap_cache_got_sigbus(f->mmap, f->fd)) {
3154 r = -EIO;
3155 goto fail;
3156 }
3157
3158 if (template && template->post_change_timer) {
3159 r = journal_file_enable_post_change_timer(
3160 f,
3161 sd_event_source_get_event(template->post_change_timer),
3162 template->post_change_timer_period);
3163
3164 if (r < 0)
3165 goto fail;
3166 }
3167
3168 /* The file is opened now successfully, thus we take possession of any passed in fd. */
3169 f->close_fd = true;
3170
3171 *ret = f;
3172 return 0;
3173
3174 fail:
3175 if (f->fd >= 0 && mmap_cache_got_sigbus(f->mmap, f->fd))
3176 r = -EIO;
3177
3178 (void) journal_file_close(f);
3179
3180 return r;
3181 }
3182
3183 int journal_file_rotate(JournalFile **f, bool compress, bool seal, Set *deferred_closes) {
3184 _cleanup_free_ char *p = NULL;
3185 size_t l;
3186 JournalFile *old_file, *new_file = NULL;
3187 int r;
3188
3189 assert(f);
3190 assert(*f);
3191
3192 old_file = *f;
3193
3194 if (!old_file->writable)
3195 return -EINVAL;
3196
3197 /* Is this a journal file that was passed to us as fd? If so, we synthesized a path name for it, and we refuse
3198 * rotation, since we don't know the actual path, and couldn't rename the file hence.*/
3199 if (path_startswith(old_file->path, "/proc/self/fd"))
3200 return -EINVAL;
3201
3202 if (!endswith(old_file->path, ".journal"))
3203 return -EINVAL;
3204
3205 l = strlen(old_file->path);
3206 r = asprintf(&p, "%.*s@" SD_ID128_FORMAT_STR "-%016"PRIx64"-%016"PRIx64".journal",
3207 (int) l - 8, old_file->path,
3208 SD_ID128_FORMAT_VAL(old_file->header->seqnum_id),
3209 le64toh((*f)->header->head_entry_seqnum),
3210 le64toh((*f)->header->head_entry_realtime));
3211 if (r < 0)
3212 return -ENOMEM;
3213
3214 /* Try to rename the file to the archived version. If the file
3215 * already was deleted, we'll get ENOENT, let's ignore that
3216 * case. */
3217 r = rename(old_file->path, p);
3218 if (r < 0 && errno != ENOENT)
3219 return -errno;
3220
3221 /* Sync the rename to disk */
3222 (void) fsync_directory_of_file(old_file->fd);
3223
3224 /* Set as archive so offlining commits w/state=STATE_ARCHIVED.
3225 * Previously we would set old_file->header->state to STATE_ARCHIVED directly here,
3226 * but journal_file_set_offline() short-circuits when state != STATE_ONLINE, which
3227 * would result in the rotated journal never getting fsync() called before closing.
3228 * Now we simply queue the archive state by setting an archive bit, leaving the state
3229 * as STATE_ONLINE so proper offlining occurs. */
3230 old_file->archive = true;
3231
3232 /* Currently, btrfs is not very good with out write patterns
3233 * and fragments heavily. Let's defrag our journal files when
3234 * we archive them */
3235 old_file->defrag_on_close = true;
3236
3237 r = journal_file_open(-1, old_file->path, old_file->flags, old_file->mode, compress, seal, NULL, old_file->mmap, deferred_closes, old_file, &new_file);
3238
3239 if (deferred_closes &&
3240 set_put(deferred_closes, old_file) >= 0)
3241 (void) journal_file_set_offline(old_file, false);
3242 else
3243 (void) journal_file_close(old_file);
3244
3245 *f = new_file;
3246 return r;
3247 }
3248
3249 int journal_file_open_reliably(
3250 const char *fname,
3251 int flags,
3252 mode_t mode,
3253 bool compress,
3254 bool seal,
3255 JournalMetrics *metrics,
3256 MMapCache *mmap_cache,
3257 Set *deferred_closes,
3258 JournalFile *template,
3259 JournalFile **ret) {
3260
3261 int r;
3262 size_t l;
3263 _cleanup_free_ char *p = NULL;
3264
3265 r = journal_file_open(-1, fname, flags, mode, compress, seal, metrics, mmap_cache, deferred_closes, template, ret);
3266 if (!IN_SET(r,
3267 -EBADMSG, /* corrupted */
3268 -ENODATA, /* truncated */
3269 -EHOSTDOWN, /* other machine */
3270 -EPROTONOSUPPORT, /* incompatible feature */
3271 -EBUSY, /* unclean shutdown */
3272 -ESHUTDOWN, /* already archived */
3273 -EIO, /* IO error, including SIGBUS on mmap */
3274 -EIDRM /* File has been deleted */))
3275 return r;
3276
3277 if ((flags & O_ACCMODE) == O_RDONLY)
3278 return r;
3279
3280 if (!(flags & O_CREAT))
3281 return r;
3282
3283 if (!endswith(fname, ".journal"))
3284 return r;
3285
3286 /* The file is corrupted. Rotate it away and try it again (but only once) */
3287
3288 l = strlen(fname);
3289 if (asprintf(&p, "%.*s@%016"PRIx64 "-%016"PRIx64 ".journal~",
3290 (int) l - 8, fname,
3291 now(CLOCK_REALTIME),
3292 random_u64()) < 0)
3293 return -ENOMEM;
3294
3295 if (rename(fname, p) < 0)
3296 return -errno;
3297
3298 /* btrfs doesn't cope well with our write pattern and
3299 * fragments heavily. Let's defrag all files we rotate */
3300
3301 (void) chattr_path(p, 0, FS_NOCOW_FL);
3302 (void) btrfs_defrag(p);
3303
3304 log_warning_errno(r, "File %s corrupted or uncleanly shut down, renaming and replacing.", fname);
3305
3306 return journal_file_open(-1, fname, flags, mode, compress, seal, metrics, mmap_cache, deferred_closes, template, ret);
3307 }
3308
3309 int journal_file_copy_entry(JournalFile *from, JournalFile *to, Object *o, uint64_t p, uint64_t *seqnum, Object **ret, uint64_t *offset) {
3310 uint64_t i, n;
3311 uint64_t q, xor_hash = 0;
3312 int r;
3313 EntryItem *items;
3314 dual_timestamp ts;
3315
3316 assert(from);
3317 assert(to);
3318 assert(o);
3319 assert(p);
3320
3321 if (!to->writable)
3322 return -EPERM;
3323
3324 ts.monotonic = le64toh(o->entry.monotonic);
3325 ts.realtime = le64toh(o->entry.realtime);
3326
3327 n = journal_file_entry_n_items(o);
3328 /* alloca() can't take 0, hence let's allocate at least one */
3329 items = alloca(sizeof(EntryItem) * MAX(1u, n));
3330
3331 for (i = 0; i < n; i++) {
3332 uint64_t l, h;
3333 le64_t le_hash;
3334 size_t t;
3335 void *data;
3336 Object *u;
3337
3338 q = le64toh(o->entry.items[i].object_offset);
3339 le_hash = o->entry.items[i].hash;
3340
3341 r = journal_file_move_to_object(from, OBJECT_DATA, q, &o);
3342 if (r < 0)
3343 return r;
3344
3345 if (le_hash != o->data.hash)
3346 return -EBADMSG;
3347
3348 l = le64toh(o->object.size) - offsetof(Object, data.payload);
3349 t = (size_t) l;
3350
3351 /* We hit the limit on 32bit machines */
3352 if ((uint64_t) t != l)
3353 return -E2BIG;
3354
3355 if (o->object.flags & OBJECT_COMPRESSION_MASK) {
3356 #if defined(HAVE_XZ) || defined(HAVE_LZ4)
3357 size_t rsize = 0;
3358
3359 r = decompress_blob(o->object.flags & OBJECT_COMPRESSION_MASK,
3360 o->data.payload, l, &from->compress_buffer, &from->compress_buffer_size, &rsize, 0);
3361 if (r < 0)
3362 return r;
3363
3364 data = from->compress_buffer;
3365 l = rsize;
3366 #else
3367 return -EPROTONOSUPPORT;
3368 #endif
3369 } else
3370 data = o->data.payload;
3371
3372 r = journal_file_append_data(to, data, l, &u, &h);
3373 if (r < 0)
3374 return r;
3375
3376 xor_hash ^= le64toh(u->data.hash);
3377 items[i].object_offset = htole64(h);
3378 items[i].hash = u->data.hash;
3379
3380 r = journal_file_move_to_object(from, OBJECT_ENTRY, p, &o);
3381 if (r < 0)
3382 return r;
3383 }
3384
3385 r = journal_file_append_entry_internal(to, &ts, xor_hash, items, n, seqnum, ret, offset);
3386
3387 if (mmap_cache_got_sigbus(to->mmap, to->fd))
3388 return -EIO;
3389
3390 return r;
3391 }
3392
3393 void journal_reset_metrics(JournalMetrics *m) {
3394 assert(m);
3395
3396 /* Set everything to "pick automatic values". */
3397
3398 *m = (JournalMetrics) {
3399 .min_use = (uint64_t) -1,
3400 .max_use = (uint64_t) -1,
3401 .min_size = (uint64_t) -1,
3402 .max_size = (uint64_t) -1,
3403 .keep_free = (uint64_t) -1,
3404 .n_max_files = (uint64_t) -1,
3405 };
3406 }
3407
3408 void journal_default_metrics(JournalMetrics *m, int fd) {
3409 char a[FORMAT_BYTES_MAX], b[FORMAT_BYTES_MAX], c[FORMAT_BYTES_MAX], d[FORMAT_BYTES_MAX], e[FORMAT_BYTES_MAX];
3410 struct statvfs ss;
3411 uint64_t fs_size;
3412
3413 assert(m);
3414 assert(fd >= 0);
3415
3416 if (fstatvfs(fd, &ss) >= 0)
3417 fs_size = ss.f_frsize * ss.f_blocks;
3418 else {
3419 log_debug_errno(errno, "Failed to detremine disk size: %m");
3420 fs_size = 0;
3421 }
3422
3423 if (m->max_use == (uint64_t) -1) {
3424
3425 if (fs_size > 0) {
3426 m->max_use = PAGE_ALIGN(fs_size / 10); /* 10% of file system size */
3427
3428 if (m->max_use > DEFAULT_MAX_USE_UPPER)
3429 m->max_use = DEFAULT_MAX_USE_UPPER;
3430
3431 if (m->max_use < DEFAULT_MAX_USE_LOWER)
3432 m->max_use = DEFAULT_MAX_USE_LOWER;
3433 } else
3434 m->max_use = DEFAULT_MAX_USE_LOWER;
3435 } else {
3436 m->max_use = PAGE_ALIGN(m->max_use);
3437
3438 if (m->max_use != 0 && m->max_use < JOURNAL_FILE_SIZE_MIN*2)
3439 m->max_use = JOURNAL_FILE_SIZE_MIN*2;
3440 }
3441
3442 if (m->min_use == (uint64_t) -1)
3443 m->min_use = DEFAULT_MIN_USE;
3444
3445 if (m->min_use > m->max_use)
3446 m->min_use = m->max_use;
3447
3448 if (m->max_size == (uint64_t) -1) {
3449 m->max_size = PAGE_ALIGN(m->max_use / 8); /* 8 chunks */
3450
3451 if (m->max_size > DEFAULT_MAX_SIZE_UPPER)
3452 m->max_size = DEFAULT_MAX_SIZE_UPPER;
3453 } else
3454 m->max_size = PAGE_ALIGN(m->max_size);
3455
3456 if (m->max_size != 0) {
3457 if (m->max_size < JOURNAL_FILE_SIZE_MIN)
3458 m->max_size = JOURNAL_FILE_SIZE_MIN;
3459
3460 if (m->max_use != 0 && m->max_size*2 > m->max_use)
3461 m->max_use = m->max_size*2;
3462 }
3463
3464 if (m->min_size == (uint64_t) -1)
3465 m->min_size = JOURNAL_FILE_SIZE_MIN;
3466 else {
3467 m->min_size = PAGE_ALIGN(m->min_size);
3468
3469 if (m->min_size < JOURNAL_FILE_SIZE_MIN)
3470 m->min_size = JOURNAL_FILE_SIZE_MIN;
3471
3472 if (m->max_size != 0 && m->min_size > m->max_size)
3473 m->max_size = m->min_size;
3474 }
3475
3476 if (m->keep_free == (uint64_t) -1) {
3477
3478 if (fs_size > 0) {
3479 m->keep_free = PAGE_ALIGN(fs_size * 3 / 20); /* 15% of file system size */
3480
3481 if (m->keep_free > DEFAULT_KEEP_FREE_UPPER)
3482 m->keep_free = DEFAULT_KEEP_FREE_UPPER;
3483
3484 } else
3485 m->keep_free = DEFAULT_KEEP_FREE;
3486 }
3487
3488 if (m->n_max_files == (uint64_t) -1)
3489 m->n_max_files = DEFAULT_N_MAX_FILES;
3490
3491 log_debug("Fixed min_use=%s max_use=%s max_size=%s min_size=%s keep_free=%s n_max_files=%" PRIu64,
3492 format_bytes(a, sizeof(a), m->min_use),
3493 format_bytes(b, sizeof(b), m->max_use),
3494 format_bytes(c, sizeof(c), m->max_size),
3495 format_bytes(d, sizeof(d), m->min_size),
3496 format_bytes(e, sizeof(e), m->keep_free),
3497 m->n_max_files);
3498 }
3499
3500 int journal_file_get_cutoff_realtime_usec(JournalFile *f, usec_t *from, usec_t *to) {
3501 assert(f);
3502 assert(f->header);
3503 assert(from || to);
3504
3505 if (from) {
3506 if (f->header->head_entry_realtime == 0)
3507 return -ENOENT;
3508
3509 *from = le64toh(f->header->head_entry_realtime);
3510 }
3511
3512 if (to) {
3513 if (f->header->tail_entry_realtime == 0)
3514 return -ENOENT;
3515
3516 *to = le64toh(f->header->tail_entry_realtime);
3517 }
3518
3519 return 1;
3520 }
3521
3522 int journal_file_get_cutoff_monotonic_usec(JournalFile *f, sd_id128_t boot_id, usec_t *from, usec_t *to) {
3523 Object *o;
3524 uint64_t p;
3525 int r;
3526
3527 assert(f);
3528 assert(from || to);
3529
3530 r = find_data_object_by_boot_id(f, boot_id, &o, &p);
3531 if (r <= 0)
3532 return r;
3533
3534 if (le64toh(o->data.n_entries) <= 0)
3535 return 0;
3536
3537 if (from) {
3538 r = journal_file_move_to_object(f, OBJECT_ENTRY, le64toh(o->data.entry_offset), &o);
3539 if (r < 0)
3540 return r;
3541
3542 *from = le64toh(o->entry.monotonic);
3543 }
3544
3545 if (to) {
3546 r = journal_file_move_to_object(f, OBJECT_DATA, p, &o);
3547 if (r < 0)
3548 return r;
3549
3550 r = generic_array_get_plus_one(f,
3551 le64toh(o->data.entry_offset),
3552 le64toh(o->data.entry_array_offset),
3553 le64toh(o->data.n_entries)-1,
3554 &o, NULL);
3555 if (r <= 0)
3556 return r;
3557
3558 *to = le64toh(o->entry.monotonic);
3559 }
3560
3561 return 1;
3562 }
3563
3564 bool journal_file_rotate_suggested(JournalFile *f, usec_t max_file_usec) {
3565 assert(f);
3566 assert(f->header);
3567
3568 /* If we gained new header fields we gained new features,
3569 * hence suggest a rotation */
3570 if (le64toh(f->header->header_size) < sizeof(Header)) {
3571 log_debug("%s uses an outdated header, suggesting rotation.", f->path);
3572 return true;
3573 }
3574
3575 /* Let's check if the hash tables grew over a certain fill
3576 * level (75%, borrowing this value from Java's hash table
3577 * implementation), and if so suggest a rotation. To calculate
3578 * the fill level we need the n_data field, which only exists
3579 * in newer versions. */
3580
3581 if (JOURNAL_HEADER_CONTAINS(f->header, n_data))
3582 if (le64toh(f->header->n_data) * 4ULL > (le64toh(f->header->data_hash_table_size) / sizeof(HashItem)) * 3ULL) {
3583 log_debug("Data hash table of %s has a fill level at %.1f (%"PRIu64" of %"PRIu64" items, %llu file size, %"PRIu64" bytes per hash table item), suggesting rotation.",
3584 f->path,
3585 100.0 * (double) le64toh(f->header->n_data) / ((double) (le64toh(f->header->data_hash_table_size) / sizeof(HashItem))),
3586 le64toh(f->header->n_data),
3587 le64toh(f->header->data_hash_table_size) / sizeof(HashItem),
3588 (unsigned long long) f->last_stat.st_size,
3589 f->last_stat.st_size / le64toh(f->header->n_data));
3590 return true;
3591 }
3592
3593 if (JOURNAL_HEADER_CONTAINS(f->header, n_fields))
3594 if (le64toh(f->header->n_fields) * 4ULL > (le64toh(f->header->field_hash_table_size) / sizeof(HashItem)) * 3ULL) {
3595 log_debug("Field hash table of %s has a fill level at %.1f (%"PRIu64" of %"PRIu64" items), suggesting rotation.",
3596 f->path,
3597 100.0 * (double) le64toh(f->header->n_fields) / ((double) (le64toh(f->header->field_hash_table_size) / sizeof(HashItem))),
3598 le64toh(f->header->n_fields),
3599 le64toh(f->header->field_hash_table_size) / sizeof(HashItem));
3600 return true;
3601 }
3602
3603 /* Are the data objects properly indexed by field objects? */
3604 if (JOURNAL_HEADER_CONTAINS(f->header, n_data) &&
3605 JOURNAL_HEADER_CONTAINS(f->header, n_fields) &&
3606 le64toh(f->header->n_data) > 0 &&
3607 le64toh(f->header->n_fields) == 0)
3608 return true;
3609
3610 if (max_file_usec > 0) {
3611 usec_t t, h;
3612
3613 h = le64toh(f->header->head_entry_realtime);
3614 t = now(CLOCK_REALTIME);
3615
3616 if (h > 0 && t > h + max_file_usec)
3617 return true;
3618 }
3619
3620 return false;
3621 }
Unnamed repository; edit this file 'description' to name the repository.