]> 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
journal: split out check for properly ordered arrays into its own function
[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 log_debug("Attempt to move to object at non-64bit boundary: %" PRIu64, offset);
752 return -EBADMSG;
753 }
754
755 /* Object may not be located in the file header */
756 if (offset < le64toh(f->header->header_size)) {
757 log_debug("Attempt to move to object located in file header: %" PRIu64, offset);
758 return -EBADMSG;
759 }
760
761 r = journal_file_move_to(f, type, false, offset, sizeof(ObjectHeader), &t);
762 if (r < 0)
763 return r;
764
765 o = (Object*) t;
766 s = le64toh(o->object.size);
767
768 if (s < sizeof(ObjectHeader)) {
769 log_debug("Attempt to move to overly short object: %" PRIu64, offset);
770 return -EBADMSG;
771 }
772
773 if (o->object.type <= OBJECT_UNUSED) {
774 log_debug("Attempt to move to object with invalid type: %" PRIu64, offset);
775 return -EBADMSG;
776 }
777
778 if (s < minimum_header_size(o)) {
779 log_debug("Attempt to move to truncated object: %" PRIu64, offset);
780 return -EBADMSG;
781 }
782
783 if (type > OBJECT_UNUSED && o->object.type != type) {
784 log_debug("Attempt to move to object of unexpected type: %" PRIu64, offset);
785 return -EBADMSG;
786 }
787
788 if (s > sizeof(ObjectHeader)) {
789 r = journal_file_move_to(f, type, false, offset, s, &t);
790 if (r < 0)
791 return r;
792
793 o = (Object*) t;
794 }
795
796 *ret = o;
797 return 0;
798 }
799
800 static uint64_t journal_file_entry_seqnum(JournalFile *f, uint64_t *seqnum) {
801 uint64_t r;
802
803 assert(f);
804 assert(f->header);
805
806 r = le64toh(f->header->tail_entry_seqnum) + 1;
807
808 if (seqnum) {
809 /* If an external seqnum counter was passed, we update
810 * both the local and the external one, and set it to
811 * the maximum of both */
812
813 if (*seqnum + 1 > r)
814 r = *seqnum + 1;
815
816 *seqnum = r;
817 }
818
819 f->header->tail_entry_seqnum = htole64(r);
820
821 if (f->header->head_entry_seqnum == 0)
822 f->header->head_entry_seqnum = htole64(r);
823
824 return r;
825 }
826
827 int journal_file_append_object(JournalFile *f, ObjectType type, uint64_t size, Object **ret, uint64_t *offset) {
828 int r;
829 uint64_t p;
830 Object *tail, *o;
831 void *t;
832
833 assert(f);
834 assert(f->header);
835 assert(type > OBJECT_UNUSED && type < _OBJECT_TYPE_MAX);
836 assert(size >= sizeof(ObjectHeader));
837 assert(offset);
838 assert(ret);
839
840 r = journal_file_set_online(f);
841 if (r < 0)
842 return r;
843
844 p = le64toh(f->header->tail_object_offset);
845 if (p == 0)
846 p = le64toh(f->header->header_size);
847 else {
848 r = journal_file_move_to_object(f, OBJECT_UNUSED, p, &tail);
849 if (r < 0)
850 return r;
851
852 p += ALIGN64(le64toh(tail->object.size));
853 }
854
855 r = journal_file_allocate(f, p, size);
856 if (r < 0)
857 return r;
858
859 r = journal_file_move_to(f, type, false, p, size, &t);
860 if (r < 0)
861 return r;
862
863 o = (Object*) t;
864
865 zero(o->object);
866 o->object.type = type;
867 o->object.size = htole64(size);
868
869 f->header->tail_object_offset = htole64(p);
870 f->header->n_objects = htole64(le64toh(f->header->n_objects) + 1);
871
872 *ret = o;
873 *offset = p;
874
875 return 0;
876 }
877
878 static int journal_file_setup_data_hash_table(JournalFile *f) {
879 uint64_t s, p;
880 Object *o;
881 int r;
882
883 assert(f);
884 assert(f->header);
885
886 /* We estimate that we need 1 hash table entry per 768 bytes
887 of journal file and we want to make sure we never get
888 beyond 75% fill level. Calculate the hash table size for
889 the maximum file size based on these metrics. */
890
891 s = (f->metrics.max_size * 4 / 768 / 3) * sizeof(HashItem);
892 if (s < DEFAULT_DATA_HASH_TABLE_SIZE)
893 s = DEFAULT_DATA_HASH_TABLE_SIZE;
894
895 log_debug("Reserving %"PRIu64" entries in hash table.", s / sizeof(HashItem));
896
897 r = journal_file_append_object(f,
898 OBJECT_DATA_HASH_TABLE,
899 offsetof(Object, hash_table.items) + s,
900 &o, &p);
901 if (r < 0)
902 return r;
903
904 memzero(o->hash_table.items, s);
905
906 f->header->data_hash_table_offset = htole64(p + offsetof(Object, hash_table.items));
907 f->header->data_hash_table_size = htole64(s);
908
909 return 0;
910 }
911
912 static int journal_file_setup_field_hash_table(JournalFile *f) {
913 uint64_t s, p;
914 Object *o;
915 int r;
916
917 assert(f);
918 assert(f->header);
919
920 /* We use a fixed size hash table for the fields as this
921 * number should grow very slowly only */
922
923 s = DEFAULT_FIELD_HASH_TABLE_SIZE;
924 r = journal_file_append_object(f,
925 OBJECT_FIELD_HASH_TABLE,
926 offsetof(Object, hash_table.items) + s,
927 &o, &p);
928 if (r < 0)
929 return r;
930
931 memzero(o->hash_table.items, s);
932
933 f->header->field_hash_table_offset = htole64(p + offsetof(Object, hash_table.items));
934 f->header->field_hash_table_size = htole64(s);
935
936 return 0;
937 }
938
939 int journal_file_map_data_hash_table(JournalFile *f) {
940 uint64_t s, p;
941 void *t;
942 int r;
943
944 assert(f);
945 assert(f->header);
946
947 if (f->data_hash_table)
948 return 0;
949
950 p = le64toh(f->header->data_hash_table_offset);
951 s = le64toh(f->header->data_hash_table_size);
952
953 r = journal_file_move_to(f,
954 OBJECT_DATA_HASH_TABLE,
955 true,
956 p, s,
957 &t);
958 if (r < 0)
959 return r;
960
961 f->data_hash_table = t;
962 return 0;
963 }
964
965 int journal_file_map_field_hash_table(JournalFile *f) {
966 uint64_t s, p;
967 void *t;
968 int r;
969
970 assert(f);
971 assert(f->header);
972
973 if (f->field_hash_table)
974 return 0;
975
976 p = le64toh(f->header->field_hash_table_offset);
977 s = le64toh(f->header->field_hash_table_size);
978
979 r = journal_file_move_to(f,
980 OBJECT_FIELD_HASH_TABLE,
981 true,
982 p, s,
983 &t);
984 if (r < 0)
985 return r;
986
987 f->field_hash_table = t;
988 return 0;
989 }
990
991 static int journal_file_link_field(
992 JournalFile *f,
993 Object *o,
994 uint64_t offset,
995 uint64_t hash) {
996
997 uint64_t p, h, m;
998 int r;
999
1000 assert(f);
1001 assert(f->header);
1002 assert(f->field_hash_table);
1003 assert(o);
1004 assert(offset > 0);
1005
1006 if (o->object.type != OBJECT_FIELD)
1007 return -EINVAL;
1008
1009 m = le64toh(f->header->field_hash_table_size) / sizeof(HashItem);
1010 if (m <= 0)
1011 return -EBADMSG;
1012
1013 /* This might alter the window we are looking at */
1014 o->field.next_hash_offset = o->field.head_data_offset = 0;
1015
1016 h = hash % m;
1017 p = le64toh(f->field_hash_table[h].tail_hash_offset);
1018 if (p == 0)
1019 f->field_hash_table[h].head_hash_offset = htole64(offset);
1020 else {
1021 r = journal_file_move_to_object(f, OBJECT_FIELD, p, &o);
1022 if (r < 0)
1023 return r;
1024
1025 o->field.next_hash_offset = htole64(offset);
1026 }
1027
1028 f->field_hash_table[h].tail_hash_offset = htole64(offset);
1029
1030 if (JOURNAL_HEADER_CONTAINS(f->header, n_fields))
1031 f->header->n_fields = htole64(le64toh(f->header->n_fields) + 1);
1032
1033 return 0;
1034 }
1035
1036 static int journal_file_link_data(
1037 JournalFile *f,
1038 Object *o,
1039 uint64_t offset,
1040 uint64_t hash) {
1041
1042 uint64_t p, h, m;
1043 int r;
1044
1045 assert(f);
1046 assert(f->header);
1047 assert(f->data_hash_table);
1048 assert(o);
1049 assert(offset > 0);
1050
1051 if (o->object.type != OBJECT_DATA)
1052 return -EINVAL;
1053
1054 m = le64toh(f->header->data_hash_table_size) / sizeof(HashItem);
1055 if (m <= 0)
1056 return -EBADMSG;
1057
1058 /* This might alter the window we are looking at */
1059 o->data.next_hash_offset = o->data.next_field_offset = 0;
1060 o->data.entry_offset = o->data.entry_array_offset = 0;
1061 o->data.n_entries = 0;
1062
1063 h = hash % m;
1064 p = le64toh(f->data_hash_table[h].tail_hash_offset);
1065 if (p == 0)
1066 /* Only entry in the hash table is easy */
1067 f->data_hash_table[h].head_hash_offset = htole64(offset);
1068 else {
1069 /* Move back to the previous data object, to patch in
1070 * pointer */
1071
1072 r = journal_file_move_to_object(f, OBJECT_DATA, p, &o);
1073 if (r < 0)
1074 return r;
1075
1076 o->data.next_hash_offset = htole64(offset);
1077 }
1078
1079 f->data_hash_table[h].tail_hash_offset = htole64(offset);
1080
1081 if (JOURNAL_HEADER_CONTAINS(f->header, n_data))
1082 f->header->n_data = htole64(le64toh(f->header->n_data) + 1);
1083
1084 return 0;
1085 }
1086
1087 int journal_file_find_field_object_with_hash(
1088 JournalFile *f,
1089 const void *field, uint64_t size, uint64_t hash,
1090 Object **ret, uint64_t *offset) {
1091
1092 uint64_t p, osize, h, m;
1093 int r;
1094
1095 assert(f);
1096 assert(f->header);
1097 assert(field && size > 0);
1098
1099 /* If the field hash table is empty, we can't find anything */
1100 if (le64toh(f->header->field_hash_table_size) <= 0)
1101 return 0;
1102
1103 /* Map the field hash table, if it isn't mapped yet. */
1104 r = journal_file_map_field_hash_table(f);
1105 if (r < 0)
1106 return r;
1107
1108 osize = offsetof(Object, field.payload) + size;
1109
1110 m = le64toh(f->header->field_hash_table_size) / sizeof(HashItem);
1111 if (m <= 0)
1112 return -EBADMSG;
1113
1114 h = hash % m;
1115 p = le64toh(f->field_hash_table[h].head_hash_offset);
1116
1117 while (p > 0) {
1118 Object *o;
1119
1120 r = journal_file_move_to_object(f, OBJECT_FIELD, p, &o);
1121 if (r < 0)
1122 return r;
1123
1124 if (le64toh(o->field.hash) == hash &&
1125 le64toh(o->object.size) == osize &&
1126 memcmp(o->field.payload, field, size) == 0) {
1127
1128 if (ret)
1129 *ret = o;
1130 if (offset)
1131 *offset = p;
1132
1133 return 1;
1134 }
1135
1136 p = le64toh(o->field.next_hash_offset);
1137 }
1138
1139 return 0;
1140 }
1141
1142 int journal_file_find_field_object(
1143 JournalFile *f,
1144 const void *field, uint64_t size,
1145 Object **ret, uint64_t *offset) {
1146
1147 uint64_t hash;
1148
1149 assert(f);
1150 assert(field && size > 0);
1151
1152 hash = hash64(field, size);
1153
1154 return journal_file_find_field_object_with_hash(f,
1155 field, size, hash,
1156 ret, offset);
1157 }
1158
1159 int journal_file_find_data_object_with_hash(
1160 JournalFile *f,
1161 const void *data, uint64_t size, uint64_t hash,
1162 Object **ret, uint64_t *offset) {
1163
1164 uint64_t p, osize, h, m;
1165 int r;
1166
1167 assert(f);
1168 assert(f->header);
1169 assert(data || size == 0);
1170
1171 /* If there's no data hash table, then there's no entry. */
1172 if (le64toh(f->header->data_hash_table_size) <= 0)
1173 return 0;
1174
1175 /* Map the data hash table, if it isn't mapped yet. */
1176 r = journal_file_map_data_hash_table(f);
1177 if (r < 0)
1178 return r;
1179
1180 osize = offsetof(Object, data.payload) + size;
1181
1182 m = le64toh(f->header->data_hash_table_size) / sizeof(HashItem);
1183 if (m <= 0)
1184 return -EBADMSG;
1185
1186 h = hash % m;
1187 p = le64toh(f->data_hash_table[h].head_hash_offset);
1188
1189 while (p > 0) {
1190 Object *o;
1191
1192 r = journal_file_move_to_object(f, OBJECT_DATA, p, &o);
1193 if (r < 0)
1194 return r;
1195
1196 if (le64toh(o->data.hash) != hash)
1197 goto next;
1198
1199 if (o->object.flags & OBJECT_COMPRESSION_MASK) {
1200 #if defined(HAVE_XZ) || defined(HAVE_LZ4)
1201 uint64_t l;
1202 size_t rsize = 0;
1203
1204 l = le64toh(o->object.size);
1205 if (l <= offsetof(Object, data.payload))
1206 return -EBADMSG;
1207
1208 l -= offsetof(Object, data.payload);
1209
1210 r = decompress_blob(o->object.flags & OBJECT_COMPRESSION_MASK,
1211 o->data.payload, l, &f->compress_buffer, &f->compress_buffer_size, &rsize, 0);
1212 if (r < 0)
1213 return r;
1214
1215 if (rsize == size &&
1216 memcmp(f->compress_buffer, data, size) == 0) {
1217
1218 if (ret)
1219 *ret = o;
1220
1221 if (offset)
1222 *offset = p;
1223
1224 return 1;
1225 }
1226 #else
1227 return -EPROTONOSUPPORT;
1228 #endif
1229 } else if (le64toh(o->object.size) == osize &&
1230 memcmp(o->data.payload, data, size) == 0) {
1231
1232 if (ret)
1233 *ret = o;
1234
1235 if (offset)
1236 *offset = p;
1237
1238 return 1;
1239 }
1240
1241 next:
1242 p = le64toh(o->data.next_hash_offset);
1243 }
1244
1245 return 0;
1246 }
1247
1248 int journal_file_find_data_object(
1249 JournalFile *f,
1250 const void *data, uint64_t size,
1251 Object **ret, uint64_t *offset) {
1252
1253 uint64_t hash;
1254
1255 assert(f);
1256 assert(data || size == 0);
1257
1258 hash = hash64(data, size);
1259
1260 return journal_file_find_data_object_with_hash(f,
1261 data, size, hash,
1262 ret, offset);
1263 }
1264
1265 static int journal_file_append_field(
1266 JournalFile *f,
1267 const void *field, uint64_t size,
1268 Object **ret, uint64_t *offset) {
1269
1270 uint64_t hash, p;
1271 uint64_t osize;
1272 Object *o;
1273 int r;
1274
1275 assert(f);
1276 assert(field && size > 0);
1277
1278 hash = hash64(field, size);
1279
1280 r = journal_file_find_field_object_with_hash(f, field, size, hash, &o, &p);
1281 if (r < 0)
1282 return r;
1283 else if (r > 0) {
1284
1285 if (ret)
1286 *ret = o;
1287
1288 if (offset)
1289 *offset = p;
1290
1291 return 0;
1292 }
1293
1294 osize = offsetof(Object, field.payload) + size;
1295 r = journal_file_append_object(f, OBJECT_FIELD, osize, &o, &p);
1296 if (r < 0)
1297 return r;
1298
1299 o->field.hash = htole64(hash);
1300 memcpy(o->field.payload, field, size);
1301
1302 r = journal_file_link_field(f, o, p, hash);
1303 if (r < 0)
1304 return r;
1305
1306 /* The linking might have altered the window, so let's
1307 * refresh our pointer */
1308 r = journal_file_move_to_object(f, OBJECT_FIELD, p, &o);
1309 if (r < 0)
1310 return r;
1311
1312 #ifdef HAVE_GCRYPT
1313 r = journal_file_hmac_put_object(f, OBJECT_FIELD, o, p);
1314 if (r < 0)
1315 return r;
1316 #endif
1317
1318 if (ret)
1319 *ret = o;
1320
1321 if (offset)
1322 *offset = p;
1323
1324 return 0;
1325 }
1326
1327 static int journal_file_append_data(
1328 JournalFile *f,
1329 const void *data, uint64_t size,
1330 Object **ret, uint64_t *offset) {
1331
1332 uint64_t hash, p;
1333 uint64_t osize;
1334 Object *o;
1335 int r, compression = 0;
1336 const void *eq;
1337
1338 assert(f);
1339 assert(data || size == 0);
1340
1341 hash = hash64(data, size);
1342
1343 r = journal_file_find_data_object_with_hash(f, data, size, hash, &o, &p);
1344 if (r < 0)
1345 return r;
1346 if (r > 0) {
1347
1348 if (ret)
1349 *ret = o;
1350
1351 if (offset)
1352 *offset = p;
1353
1354 return 0;
1355 }
1356
1357 osize = offsetof(Object, data.payload) + size;
1358 r = journal_file_append_object(f, OBJECT_DATA, osize, &o, &p);
1359 if (r < 0)
1360 return r;
1361
1362 o->data.hash = htole64(hash);
1363
1364 #if defined(HAVE_XZ) || defined(HAVE_LZ4)
1365 if (JOURNAL_FILE_COMPRESS(f) && size >= COMPRESSION_SIZE_THRESHOLD) {
1366 size_t rsize = 0;
1367
1368 compression = compress_blob(data, size, o->data.payload, size - 1, &rsize);
1369
1370 if (compression >= 0) {
1371 o->object.size = htole64(offsetof(Object, data.payload) + rsize);
1372 o->object.flags |= compression;
1373
1374 log_debug("Compressed data object %"PRIu64" -> %zu using %s",
1375 size, rsize, object_compressed_to_string(compression));
1376 } else
1377 /* Compression didn't work, we don't really care why, let's continue without compression */
1378 compression = 0;
1379 }
1380 #endif
1381
1382 if (compression == 0)
1383 memcpy_safe(o->data.payload, data, size);
1384
1385 r = journal_file_link_data(f, o, p, hash);
1386 if (r < 0)
1387 return r;
1388
1389 #ifdef HAVE_GCRYPT
1390 r = journal_file_hmac_put_object(f, OBJECT_DATA, o, p);
1391 if (r < 0)
1392 return r;
1393 #endif
1394
1395 /* The linking might have altered the window, so let's
1396 * refresh our pointer */
1397 r = journal_file_move_to_object(f, OBJECT_DATA, p, &o);
1398 if (r < 0)
1399 return r;
1400
1401 if (!data)
1402 eq = NULL;
1403 else
1404 eq = memchr(data, '=', size);
1405 if (eq && eq > data) {
1406 Object *fo = NULL;
1407 uint64_t fp;
1408
1409 /* Create field object ... */
1410 r = journal_file_append_field(f, data, (uint8_t*) eq - (uint8_t*) data, &fo, &fp);
1411 if (r < 0)
1412 return r;
1413
1414 /* ... and link it in. */
1415 o->data.next_field_offset = fo->field.head_data_offset;
1416 fo->field.head_data_offset = le64toh(p);
1417 }
1418
1419 if (ret)
1420 *ret = o;
1421
1422 if (offset)
1423 *offset = p;
1424
1425 return 0;
1426 }
1427
1428 uint64_t journal_file_entry_n_items(Object *o) {
1429 assert(o);
1430
1431 if (o->object.type != OBJECT_ENTRY)
1432 return 0;
1433
1434 return (le64toh(o->object.size) - offsetof(Object, entry.items)) / sizeof(EntryItem);
1435 }
1436
1437 uint64_t journal_file_entry_array_n_items(Object *o) {
1438 assert(o);
1439
1440 if (o->object.type != OBJECT_ENTRY_ARRAY)
1441 return 0;
1442
1443 return (le64toh(o->object.size) - offsetof(Object, entry_array.items)) / sizeof(uint64_t);
1444 }
1445
1446 uint64_t journal_file_hash_table_n_items(Object *o) {
1447 assert(o);
1448
1449 if (o->object.type != OBJECT_DATA_HASH_TABLE &&
1450 o->object.type != OBJECT_FIELD_HASH_TABLE)
1451 return 0;
1452
1453 return (le64toh(o->object.size) - offsetof(Object, hash_table.items)) / sizeof(HashItem);
1454 }
1455
1456 static int link_entry_into_array(JournalFile *f,
1457 le64_t *first,
1458 le64_t *idx,
1459 uint64_t p) {
1460 int r;
1461 uint64_t n = 0, ap = 0, q, i, a, hidx;
1462 Object *o;
1463
1464 assert(f);
1465 assert(f->header);
1466 assert(first);
1467 assert(idx);
1468 assert(p > 0);
1469
1470 a = le64toh(*first);
1471 i = hidx = le64toh(*idx);
1472 while (a > 0) {
1473
1474 r = journal_file_move_to_object(f, OBJECT_ENTRY_ARRAY, a, &o);
1475 if (r < 0)
1476 return r;
1477
1478 n = journal_file_entry_array_n_items(o);
1479 if (i < n) {
1480 o->entry_array.items[i] = htole64(p);
1481 *idx = htole64(hidx + 1);
1482 return 0;
1483 }
1484
1485 i -= n;
1486 ap = a;
1487 a = le64toh(o->entry_array.next_entry_array_offset);
1488 }
1489
1490 if (hidx > n)
1491 n = (hidx+1) * 2;
1492 else
1493 n = n * 2;
1494
1495 if (n < 4)
1496 n = 4;
1497
1498 r = journal_file_append_object(f, OBJECT_ENTRY_ARRAY,
1499 offsetof(Object, entry_array.items) + n * sizeof(uint64_t),
1500 &o, &q);
1501 if (r < 0)
1502 return r;
1503
1504 #ifdef HAVE_GCRYPT
1505 r = journal_file_hmac_put_object(f, OBJECT_ENTRY_ARRAY, o, q);
1506 if (r < 0)
1507 return r;
1508 #endif
1509
1510 o->entry_array.items[i] = htole64(p);
1511
1512 if (ap == 0)
1513 *first = htole64(q);
1514 else {
1515 r = journal_file_move_to_object(f, OBJECT_ENTRY_ARRAY, ap, &o);
1516 if (r < 0)
1517 return r;
1518
1519 o->entry_array.next_entry_array_offset = htole64(q);
1520 }
1521
1522 if (JOURNAL_HEADER_CONTAINS(f->header, n_entry_arrays))
1523 f->header->n_entry_arrays = htole64(le64toh(f->header->n_entry_arrays) + 1);
1524
1525 *idx = htole64(hidx + 1);
1526
1527 return 0;
1528 }
1529
1530 static int link_entry_into_array_plus_one(JournalFile *f,
1531 le64_t *extra,
1532 le64_t *first,
1533 le64_t *idx,
1534 uint64_t p) {
1535
1536 int r;
1537
1538 assert(f);
1539 assert(extra);
1540 assert(first);
1541 assert(idx);
1542 assert(p > 0);
1543
1544 if (*idx == 0)
1545 *extra = htole64(p);
1546 else {
1547 le64_t i;
1548
1549 i = htole64(le64toh(*idx) - 1);
1550 r = link_entry_into_array(f, first, &i, p);
1551 if (r < 0)
1552 return r;
1553 }
1554
1555 *idx = htole64(le64toh(*idx) + 1);
1556 return 0;
1557 }
1558
1559 static int journal_file_link_entry_item(JournalFile *f, Object *o, uint64_t offset, uint64_t i) {
1560 uint64_t p;
1561 int r;
1562 assert(f);
1563 assert(o);
1564 assert(offset > 0);
1565
1566 p = le64toh(o->entry.items[i].object_offset);
1567 if (p == 0)
1568 return -EINVAL;
1569
1570 r = journal_file_move_to_object(f, OBJECT_DATA, p, &o);
1571 if (r < 0)
1572 return r;
1573
1574 return link_entry_into_array_plus_one(f,
1575 &o->data.entry_offset,
1576 &o->data.entry_array_offset,
1577 &o->data.n_entries,
1578 offset);
1579 }
1580
1581 static int journal_file_link_entry(JournalFile *f, Object *o, uint64_t offset) {
1582 uint64_t n, i;
1583 int r;
1584
1585 assert(f);
1586 assert(f->header);
1587 assert(o);
1588 assert(offset > 0);
1589
1590 if (o->object.type != OBJECT_ENTRY)
1591 return -EINVAL;
1592
1593 __sync_synchronize();
1594
1595 /* Link up the entry itself */
1596 r = link_entry_into_array(f,
1597 &f->header->entry_array_offset,
1598 &f->header->n_entries,
1599 offset);
1600 if (r < 0)
1601 return r;
1602
1603 /* log_debug("=> %s seqnr=%"PRIu64" n_entries=%"PRIu64, f->path, o->entry.seqnum, f->header->n_entries); */
1604
1605 if (f->header->head_entry_realtime == 0)
1606 f->header->head_entry_realtime = o->entry.realtime;
1607
1608 f->header->tail_entry_realtime = o->entry.realtime;
1609 f->header->tail_entry_monotonic = o->entry.monotonic;
1610
1611 f->tail_entry_monotonic_valid = true;
1612
1613 /* Link up the items */
1614 n = journal_file_entry_n_items(o);
1615 for (i = 0; i < n; i++) {
1616 r = journal_file_link_entry_item(f, o, offset, i);
1617 if (r < 0)
1618 return r;
1619 }
1620
1621 return 0;
1622 }
1623
1624 static int journal_file_append_entry_internal(
1625 JournalFile *f,
1626 const dual_timestamp *ts,
1627 uint64_t xor_hash,
1628 const EntryItem items[], unsigned n_items,
1629 uint64_t *seqnum,
1630 Object **ret, uint64_t *offset) {
1631 uint64_t np;
1632 uint64_t osize;
1633 Object *o;
1634 int r;
1635
1636 assert(f);
1637 assert(f->header);
1638 assert(items || n_items == 0);
1639 assert(ts);
1640
1641 osize = offsetof(Object, entry.items) + (n_items * sizeof(EntryItem));
1642
1643 r = journal_file_append_object(f, OBJECT_ENTRY, osize, &o, &np);
1644 if (r < 0)
1645 return r;
1646
1647 o->entry.seqnum = htole64(journal_file_entry_seqnum(f, seqnum));
1648 memcpy_safe(o->entry.items, items, n_items * sizeof(EntryItem));
1649 o->entry.realtime = htole64(ts->realtime);
1650 o->entry.monotonic = htole64(ts->monotonic);
1651 o->entry.xor_hash = htole64(xor_hash);
1652 o->entry.boot_id = f->header->boot_id;
1653
1654 #ifdef HAVE_GCRYPT
1655 r = journal_file_hmac_put_object(f, OBJECT_ENTRY, o, np);
1656 if (r < 0)
1657 return r;
1658 #endif
1659
1660 r = journal_file_link_entry(f, o, np);
1661 if (r < 0)
1662 return r;
1663
1664 if (ret)
1665 *ret = o;
1666
1667 if (offset)
1668 *offset = np;
1669
1670 return 0;
1671 }
1672
1673 void journal_file_post_change(JournalFile *f) {
1674 assert(f);
1675
1676 /* inotify() does not receive IN_MODIFY events from file
1677 * accesses done via mmap(). After each access we hence
1678 * trigger IN_MODIFY by truncating the journal file to its
1679 * current size which triggers IN_MODIFY. */
1680
1681 __sync_synchronize();
1682
1683 if (ftruncate(f->fd, f->last_stat.st_size) < 0)
1684 log_debug_errno(errno, "Failed to truncate file to its own size: %m");
1685 }
1686
1687 static int post_change_thunk(sd_event_source *timer, uint64_t usec, void *userdata) {
1688 assert(userdata);
1689
1690 journal_file_post_change(userdata);
1691
1692 return 1;
1693 }
1694
1695 static void schedule_post_change(JournalFile *f) {
1696 sd_event_source *timer;
1697 int enabled, r;
1698 uint64_t now;
1699
1700 assert(f);
1701 assert(f->post_change_timer);
1702
1703 timer = f->post_change_timer;
1704
1705 r = sd_event_source_get_enabled(timer, &enabled);
1706 if (r < 0) {
1707 log_debug_errno(r, "Failed to get ftruncate timer state: %m");
1708 goto fail;
1709 }
1710
1711 if (enabled == SD_EVENT_ONESHOT)
1712 return;
1713
1714 r = sd_event_now(sd_event_source_get_event(timer), CLOCK_MONOTONIC, &now);
1715 if (r < 0) {
1716 log_debug_errno(r, "Failed to get clock's now for scheduling ftruncate: %m");
1717 goto fail;
1718 }
1719
1720 r = sd_event_source_set_time(timer, now+f->post_change_timer_period);
1721 if (r < 0) {
1722 log_debug_errno(r, "Failed to set time for scheduling ftruncate: %m");
1723 goto fail;
1724 }
1725
1726 r = sd_event_source_set_enabled(timer, SD_EVENT_ONESHOT);
1727 if (r < 0) {
1728 log_debug_errno(r, "Failed to enable scheduled ftruncate: %m");
1729 goto fail;
1730 }
1731
1732 return;
1733
1734 fail:
1735 /* On failure, let's simply post the change immediately. */
1736 journal_file_post_change(f);
1737 }
1738
1739 /* Enable coalesced change posting in a timer on the provided sd_event instance */
1740 int journal_file_enable_post_change_timer(JournalFile *f, sd_event *e, usec_t t) {
1741 _cleanup_(sd_event_source_unrefp) sd_event_source *timer = NULL;
1742 int r;
1743
1744 assert(f);
1745 assert_return(!f->post_change_timer, -EINVAL);
1746 assert(e);
1747 assert(t);
1748
1749 r = sd_event_add_time(e, &timer, CLOCK_MONOTONIC, 0, 0, post_change_thunk, f);
1750 if (r < 0)
1751 return r;
1752
1753 r = sd_event_source_set_enabled(timer, SD_EVENT_OFF);
1754 if (r < 0)
1755 return r;
1756
1757 f->post_change_timer = timer;
1758 timer = NULL;
1759 f->post_change_timer_period = t;
1760
1761 return r;
1762 }
1763
1764 static int entry_item_cmp(const void *_a, const void *_b) {
1765 const EntryItem *a = _a, *b = _b;
1766
1767 if (le64toh(a->object_offset) < le64toh(b->object_offset))
1768 return -1;
1769 if (le64toh(a->object_offset) > le64toh(b->object_offset))
1770 return 1;
1771 return 0;
1772 }
1773
1774 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) {
1775 unsigned i;
1776 EntryItem *items;
1777 int r;
1778 uint64_t xor_hash = 0;
1779 struct dual_timestamp _ts;
1780
1781 assert(f);
1782 assert(f->header);
1783 assert(iovec || n_iovec == 0);
1784
1785 if (!ts) {
1786 dual_timestamp_get(&_ts);
1787 ts = &_ts;
1788 }
1789
1790 #ifdef HAVE_GCRYPT
1791 r = journal_file_maybe_append_tag(f, ts->realtime);
1792 if (r < 0)
1793 return r;
1794 #endif
1795
1796 /* alloca() can't take 0, hence let's allocate at least one */
1797 items = alloca(sizeof(EntryItem) * MAX(1u, n_iovec));
1798
1799 for (i = 0; i < n_iovec; i++) {
1800 uint64_t p;
1801 Object *o;
1802
1803 r = journal_file_append_data(f, iovec[i].iov_base, iovec[i].iov_len, &o, &p);
1804 if (r < 0)
1805 return r;
1806
1807 xor_hash ^= le64toh(o->data.hash);
1808 items[i].object_offset = htole64(p);
1809 items[i].hash = o->data.hash;
1810 }
1811
1812 /* Order by the position on disk, in order to improve seek
1813 * times for rotating media. */
1814 qsort_safe(items, n_iovec, sizeof(EntryItem), entry_item_cmp);
1815
1816 r = journal_file_append_entry_internal(f, ts, xor_hash, items, n_iovec, seqnum, ret, offset);
1817
1818 /* If the memory mapping triggered a SIGBUS then we return an
1819 * IO error and ignore the error code passed down to us, since
1820 * it is very likely just an effect of a nullified replacement
1821 * mapping page */
1822
1823 if (mmap_cache_got_sigbus(f->mmap, f->fd))
1824 r = -EIO;
1825
1826 if (f->post_change_timer)
1827 schedule_post_change(f);
1828 else
1829 journal_file_post_change(f);
1830
1831 return r;
1832 }
1833
1834 typedef struct ChainCacheItem {
1835 uint64_t first; /* the array at the beginning of the chain */
1836 uint64_t array; /* the cached array */
1837 uint64_t begin; /* the first item in the cached array */
1838 uint64_t total; /* the total number of items in all arrays before this one in the chain */
1839 uint64_t last_index; /* the last index we looked at, to optimize locality when bisecting */
1840 } ChainCacheItem;
1841
1842 static void chain_cache_put(
1843 OrderedHashmap *h,
1844 ChainCacheItem *ci,
1845 uint64_t first,
1846 uint64_t array,
1847 uint64_t begin,
1848 uint64_t total,
1849 uint64_t last_index) {
1850
1851 if (!ci) {
1852 /* If the chain item to cache for this chain is the
1853 * first one it's not worth caching anything */
1854 if (array == first)
1855 return;
1856
1857 if (ordered_hashmap_size(h) >= CHAIN_CACHE_MAX) {
1858 ci = ordered_hashmap_steal_first(h);
1859 assert(ci);
1860 } else {
1861 ci = new(ChainCacheItem, 1);
1862 if (!ci)
1863 return;
1864 }
1865
1866 ci->first = first;
1867
1868 if (ordered_hashmap_put(h, &ci->first, ci) < 0) {
1869 free(ci);
1870 return;
1871 }
1872 } else
1873 assert(ci->first == first);
1874
1875 ci->array = array;
1876 ci->begin = begin;
1877 ci->total = total;
1878 ci->last_index = last_index;
1879 }
1880
1881 static int generic_array_get(
1882 JournalFile *f,
1883 uint64_t first,
1884 uint64_t i,
1885 Object **ret, uint64_t *offset) {
1886
1887 Object *o;
1888 uint64_t p = 0, a, t = 0;
1889 int r;
1890 ChainCacheItem *ci;
1891
1892 assert(f);
1893
1894 a = first;
1895
1896 /* Try the chain cache first */
1897 ci = ordered_hashmap_get(f->chain_cache, &first);
1898 if (ci && i > ci->total) {
1899 a = ci->array;
1900 i -= ci->total;
1901 t = ci->total;
1902 }
1903
1904 while (a > 0) {
1905 uint64_t k;
1906
1907 r = journal_file_move_to_object(f, OBJECT_ENTRY_ARRAY, a, &o);
1908 if (r < 0)
1909 return r;
1910
1911 k = journal_file_entry_array_n_items(o);
1912 if (i < k) {
1913 p = le64toh(o->entry_array.items[i]);
1914 goto found;
1915 }
1916
1917 i -= k;
1918 t += k;
1919 a = le64toh(o->entry_array.next_entry_array_offset);
1920 }
1921
1922 return 0;
1923
1924 found:
1925 /* Let's cache this item for the next invocation */
1926 chain_cache_put(f->chain_cache, ci, first, a, le64toh(o->entry_array.items[0]), t, i);
1927
1928 r = journal_file_move_to_object(f, OBJECT_ENTRY, p, &o);
1929 if (r < 0)
1930 return r;
1931
1932 if (ret)
1933 *ret = o;
1934
1935 if (offset)
1936 *offset = p;
1937
1938 return 1;
1939 }
1940
1941 static int generic_array_get_plus_one(
1942 JournalFile *f,
1943 uint64_t extra,
1944 uint64_t first,
1945 uint64_t i,
1946 Object **ret, uint64_t *offset) {
1947
1948 Object *o;
1949
1950 assert(f);
1951
1952 if (i == 0) {
1953 int r;
1954
1955 r = journal_file_move_to_object(f, OBJECT_ENTRY, extra, &o);
1956 if (r < 0)
1957 return r;
1958
1959 if (ret)
1960 *ret = o;
1961
1962 if (offset)
1963 *offset = extra;
1964
1965 return 1;
1966 }
1967
1968 return generic_array_get(f, first, i-1, ret, offset);
1969 }
1970
1971 enum {
1972 TEST_FOUND,
1973 TEST_LEFT,
1974 TEST_RIGHT
1975 };
1976
1977 static int generic_array_bisect(
1978 JournalFile *f,
1979 uint64_t first,
1980 uint64_t n,
1981 uint64_t needle,
1982 int (*test_object)(JournalFile *f, uint64_t p, uint64_t needle),
1983 direction_t direction,
1984 Object **ret,
1985 uint64_t *offset,
1986 uint64_t *idx) {
1987
1988 uint64_t a, p, t = 0, i = 0, last_p = 0, last_index = (uint64_t) -1;
1989 bool subtract_one = false;
1990 Object *o, *array = NULL;
1991 int r;
1992 ChainCacheItem *ci;
1993
1994 assert(f);
1995 assert(test_object);
1996
1997 /* Start with the first array in the chain */
1998 a = first;
1999
2000 ci = ordered_hashmap_get(f->chain_cache, &first);
2001 if (ci && n > ci->total) {
2002 /* Ah, we have iterated this bisection array chain
2003 * previously! Let's see if we can skip ahead in the
2004 * chain, as far as the last time. But we can't jump
2005 * backwards in the chain, so let's check that
2006 * first. */
2007
2008 r = test_object(f, ci->begin, needle);
2009 if (r < 0)
2010 return r;
2011
2012 if (r == TEST_LEFT) {
2013 /* OK, what we are looking for is right of the
2014 * begin of this EntryArray, so let's jump
2015 * straight to previously cached array in the
2016 * chain */
2017
2018 a = ci->array;
2019 n -= ci->total;
2020 t = ci->total;
2021 last_index = ci->last_index;
2022 }
2023 }
2024
2025 while (a > 0) {
2026 uint64_t left, right, k, lp;
2027
2028 r = journal_file_move_to_object(f, OBJECT_ENTRY_ARRAY, a, &array);
2029 if (r < 0)
2030 return r;
2031
2032 k = journal_file_entry_array_n_items(array);
2033 right = MIN(k, n);
2034 if (right <= 0)
2035 return 0;
2036
2037 i = right - 1;
2038 lp = p = le64toh(array->entry_array.items[i]);
2039 if (p <= 0)
2040 r = -EBADMSG;
2041 else
2042 r = test_object(f, p, needle);
2043 if (r == -EBADMSG) {
2044 log_debug_errno(r, "Encountered invalid entry while bisecting, cutting algorithm short. (1)");
2045 n = i;
2046 continue;
2047 }
2048 if (r < 0)
2049 return r;
2050
2051 if (r == TEST_FOUND)
2052 r = direction == DIRECTION_DOWN ? TEST_RIGHT : TEST_LEFT;
2053
2054 if (r == TEST_RIGHT) {
2055 left = 0;
2056 right -= 1;
2057
2058 if (last_index != (uint64_t) -1) {
2059 assert(last_index <= right);
2060
2061 /* If we cached the last index we
2062 * looked at, let's try to not to jump
2063 * too wildly around and see if we can
2064 * limit the range to look at early to
2065 * the immediate neighbors of the last
2066 * index we looked at. */
2067
2068 if (last_index > 0) {
2069 uint64_t x = last_index - 1;
2070
2071 p = le64toh(array->entry_array.items[x]);
2072 if (p <= 0)
2073 return -EBADMSG;
2074
2075 r = test_object(f, p, needle);
2076 if (r < 0)
2077 return r;
2078
2079 if (r == TEST_FOUND)
2080 r = direction == DIRECTION_DOWN ? TEST_RIGHT : TEST_LEFT;
2081
2082 if (r == TEST_RIGHT)
2083 right = x;
2084 else
2085 left = x + 1;
2086 }
2087
2088 if (last_index < right) {
2089 uint64_t y = last_index + 1;
2090
2091 p = le64toh(array->entry_array.items[y]);
2092 if (p <= 0)
2093 return -EBADMSG;
2094
2095 r = test_object(f, p, needle);
2096 if (r < 0)
2097 return r;
2098
2099 if (r == TEST_FOUND)
2100 r = direction == DIRECTION_DOWN ? TEST_RIGHT : TEST_LEFT;
2101
2102 if (r == TEST_RIGHT)
2103 right = y;
2104 else
2105 left = y + 1;
2106 }
2107 }
2108
2109 for (;;) {
2110 if (left == right) {
2111 if (direction == DIRECTION_UP)
2112 subtract_one = true;
2113
2114 i = left;
2115 goto found;
2116 }
2117
2118 assert(left < right);
2119 i = (left + right) / 2;
2120
2121 p = le64toh(array->entry_array.items[i]);
2122 if (p <= 0)
2123 r = -EBADMSG;
2124 else
2125 r = test_object(f, p, needle);
2126 if (r == -EBADMSG) {
2127 log_debug_errno(r, "Encountered invalid entry while bisecting, cutting algorithm short. (2)");
2128 right = n = i;
2129 continue;
2130 }
2131 if (r < 0)
2132 return r;
2133
2134 if (r == TEST_FOUND)
2135 r = direction == DIRECTION_DOWN ? TEST_RIGHT : TEST_LEFT;
2136
2137 if (r == TEST_RIGHT)
2138 right = i;
2139 else
2140 left = i + 1;
2141 }
2142 }
2143
2144 if (k >= n) {
2145 if (direction == DIRECTION_UP) {
2146 i = n;
2147 subtract_one = true;
2148 goto found;
2149 }
2150
2151 return 0;
2152 }
2153
2154 last_p = lp;
2155
2156 n -= k;
2157 t += k;
2158 last_index = (uint64_t) -1;
2159 a = le64toh(array->entry_array.next_entry_array_offset);
2160 }
2161
2162 return 0;
2163
2164 found:
2165 if (subtract_one && t == 0 && i == 0)
2166 return 0;
2167
2168 /* Let's cache this item for the next invocation */
2169 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);
2170
2171 if (subtract_one && i == 0)
2172 p = last_p;
2173 else if (subtract_one)
2174 p = le64toh(array->entry_array.items[i-1]);
2175 else
2176 p = le64toh(array->entry_array.items[i]);
2177
2178 r = journal_file_move_to_object(f, OBJECT_ENTRY, p, &o);
2179 if (r < 0)
2180 return r;
2181
2182 if (ret)
2183 *ret = o;
2184
2185 if (offset)
2186 *offset = p;
2187
2188 if (idx)
2189 *idx = t + i + (subtract_one ? -1 : 0);
2190
2191 return 1;
2192 }
2193
2194 static int generic_array_bisect_plus_one(
2195 JournalFile *f,
2196 uint64_t extra,
2197 uint64_t first,
2198 uint64_t n,
2199 uint64_t needle,
2200 int (*test_object)(JournalFile *f, uint64_t p, uint64_t needle),
2201 direction_t direction,
2202 Object **ret,
2203 uint64_t *offset,
2204 uint64_t *idx) {
2205
2206 int r;
2207 bool step_back = false;
2208 Object *o;
2209
2210 assert(f);
2211 assert(test_object);
2212
2213 if (n <= 0)
2214 return 0;
2215
2216 /* This bisects the array in object 'first', but first checks
2217 * an extra */
2218 r = test_object(f, extra, needle);
2219 if (r < 0)
2220 return r;
2221
2222 if (r == TEST_FOUND)
2223 r = direction == DIRECTION_DOWN ? TEST_RIGHT : TEST_LEFT;
2224
2225 /* if we are looking with DIRECTION_UP then we need to first
2226 see if in the actual array there is a matching entry, and
2227 return the last one of that. But if there isn't any we need
2228 to return this one. Hence remember this, and return it
2229 below. */
2230 if (r == TEST_LEFT)
2231 step_back = direction == DIRECTION_UP;
2232
2233 if (r == TEST_RIGHT) {
2234 if (direction == DIRECTION_DOWN)
2235 goto found;
2236 else
2237 return 0;
2238 }
2239
2240 r = generic_array_bisect(f, first, n-1, needle, test_object, direction, ret, offset, idx);
2241
2242 if (r == 0 && step_back)
2243 goto found;
2244
2245 if (r > 0 && idx)
2246 (*idx)++;
2247
2248 return r;
2249
2250 found:
2251 r = journal_file_move_to_object(f, OBJECT_ENTRY, extra, &o);
2252 if (r < 0)
2253 return r;
2254
2255 if (ret)
2256 *ret = o;
2257
2258 if (offset)
2259 *offset = extra;
2260
2261 if (idx)
2262 *idx = 0;
2263
2264 return 1;
2265 }
2266
2267 _pure_ static int test_object_offset(JournalFile *f, uint64_t p, uint64_t needle) {
2268 assert(f);
2269 assert(p > 0);
2270
2271 if (p == needle)
2272 return TEST_FOUND;
2273 else if (p < needle)
2274 return TEST_LEFT;
2275 else
2276 return TEST_RIGHT;
2277 }
2278
2279 static int test_object_seqnum(JournalFile *f, uint64_t p, uint64_t needle) {
2280 Object *o;
2281 int r;
2282
2283 assert(f);
2284 assert(p > 0);
2285
2286 r = journal_file_move_to_object(f, OBJECT_ENTRY, p, &o);
2287 if (r < 0)
2288 return r;
2289
2290 if (le64toh(o->entry.seqnum) == needle)
2291 return TEST_FOUND;
2292 else if (le64toh(o->entry.seqnum) < needle)
2293 return TEST_LEFT;
2294 else
2295 return TEST_RIGHT;
2296 }
2297
2298 int journal_file_move_to_entry_by_seqnum(
2299 JournalFile *f,
2300 uint64_t seqnum,
2301 direction_t direction,
2302 Object **ret,
2303 uint64_t *offset) {
2304 assert(f);
2305 assert(f->header);
2306
2307 return generic_array_bisect(f,
2308 le64toh(f->header->entry_array_offset),
2309 le64toh(f->header->n_entries),
2310 seqnum,
2311 test_object_seqnum,
2312 direction,
2313 ret, offset, NULL);
2314 }
2315
2316 static int test_object_realtime(JournalFile *f, uint64_t p, uint64_t needle) {
2317 Object *o;
2318 int r;
2319
2320 assert(f);
2321 assert(p > 0);
2322
2323 r = journal_file_move_to_object(f, OBJECT_ENTRY, p, &o);
2324 if (r < 0)
2325 return r;
2326
2327 if (le64toh(o->entry.realtime) == needle)
2328 return TEST_FOUND;
2329 else if (le64toh(o->entry.realtime) < needle)
2330 return TEST_LEFT;
2331 else
2332 return TEST_RIGHT;
2333 }
2334
2335 int journal_file_move_to_entry_by_realtime(
2336 JournalFile *f,
2337 uint64_t realtime,
2338 direction_t direction,
2339 Object **ret,
2340 uint64_t *offset) {
2341 assert(f);
2342 assert(f->header);
2343
2344 return generic_array_bisect(f,
2345 le64toh(f->header->entry_array_offset),
2346 le64toh(f->header->n_entries),
2347 realtime,
2348 test_object_realtime,
2349 direction,
2350 ret, offset, NULL);
2351 }
2352
2353 static int test_object_monotonic(JournalFile *f, uint64_t p, uint64_t needle) {
2354 Object *o;
2355 int r;
2356
2357 assert(f);
2358 assert(p > 0);
2359
2360 r = journal_file_move_to_object(f, OBJECT_ENTRY, p, &o);
2361 if (r < 0)
2362 return r;
2363
2364 if (le64toh(o->entry.monotonic) == needle)
2365 return TEST_FOUND;
2366 else if (le64toh(o->entry.monotonic) < needle)
2367 return TEST_LEFT;
2368 else
2369 return TEST_RIGHT;
2370 }
2371
2372 static int find_data_object_by_boot_id(
2373 JournalFile *f,
2374 sd_id128_t boot_id,
2375 Object **o,
2376 uint64_t *b) {
2377
2378 char t[sizeof("_BOOT_ID=")-1 + 32 + 1] = "_BOOT_ID=";
2379
2380 sd_id128_to_string(boot_id, t + 9);
2381 return journal_file_find_data_object(f, t, sizeof(t) - 1, o, b);
2382 }
2383
2384 int journal_file_move_to_entry_by_monotonic(
2385 JournalFile *f,
2386 sd_id128_t boot_id,
2387 uint64_t monotonic,
2388 direction_t direction,
2389 Object **ret,
2390 uint64_t *offset) {
2391
2392 Object *o;
2393 int r;
2394
2395 assert(f);
2396
2397 r = find_data_object_by_boot_id(f, boot_id, &o, NULL);
2398 if (r < 0)
2399 return r;
2400 if (r == 0)
2401 return -ENOENT;
2402
2403 return generic_array_bisect_plus_one(f,
2404 le64toh(o->data.entry_offset),
2405 le64toh(o->data.entry_array_offset),
2406 le64toh(o->data.n_entries),
2407 monotonic,
2408 test_object_monotonic,
2409 direction,
2410 ret, offset, NULL);
2411 }
2412
2413 void journal_file_reset_location(JournalFile *f) {
2414 f->location_type = LOCATION_HEAD;
2415 f->current_offset = 0;
2416 f->current_seqnum = 0;
2417 f->current_realtime = 0;
2418 f->current_monotonic = 0;
2419 zero(f->current_boot_id);
2420 f->current_xor_hash = 0;
2421 }
2422
2423 void journal_file_save_location(JournalFile *f, Object *o, uint64_t offset) {
2424 f->location_type = LOCATION_SEEK;
2425 f->current_offset = offset;
2426 f->current_seqnum = le64toh(o->entry.seqnum);
2427 f->current_realtime = le64toh(o->entry.realtime);
2428 f->current_monotonic = le64toh(o->entry.monotonic);
2429 f->current_boot_id = o->entry.boot_id;
2430 f->current_xor_hash = le64toh(o->entry.xor_hash);
2431 }
2432
2433 int journal_file_compare_locations(JournalFile *af, JournalFile *bf) {
2434 assert(af);
2435 assert(af->header);
2436 assert(bf);
2437 assert(bf->header);
2438 assert(af->location_type == LOCATION_SEEK);
2439 assert(bf->location_type == LOCATION_SEEK);
2440
2441 /* If contents and timestamps match, these entries are
2442 * identical, even if the seqnum does not match */
2443 if (sd_id128_equal(af->current_boot_id, bf->current_boot_id) &&
2444 af->current_monotonic == bf->current_monotonic &&
2445 af->current_realtime == bf->current_realtime &&
2446 af->current_xor_hash == bf->current_xor_hash)
2447 return 0;
2448
2449 if (sd_id128_equal(af->header->seqnum_id, bf->header->seqnum_id)) {
2450
2451 /* If this is from the same seqnum source, compare
2452 * seqnums */
2453 if (af->current_seqnum < bf->current_seqnum)
2454 return -1;
2455 if (af->current_seqnum > bf->current_seqnum)
2456 return 1;
2457
2458 /* Wow! This is weird, different data but the same
2459 * seqnums? Something is borked, but let's make the
2460 * best of it and compare by time. */
2461 }
2462
2463 if (sd_id128_equal(af->current_boot_id, bf->current_boot_id)) {
2464
2465 /* If the boot id matches, compare monotonic time */
2466 if (af->current_monotonic < bf->current_monotonic)
2467 return -1;
2468 if (af->current_monotonic > bf->current_monotonic)
2469 return 1;
2470 }
2471
2472 /* Otherwise, compare UTC time */
2473 if (af->current_realtime < bf->current_realtime)
2474 return -1;
2475 if (af->current_realtime > bf->current_realtime)
2476 return 1;
2477
2478 /* Finally, compare by contents */
2479 if (af->current_xor_hash < bf->current_xor_hash)
2480 return -1;
2481 if (af->current_xor_hash > bf->current_xor_hash)
2482 return 1;
2483
2484 return 0;
2485 }
2486
2487 static int bump_array_index(uint64_t *i, direction_t direction, uint64_t n) {
2488
2489 /* Increase or decrease the specified index, in the right direction. */
2490
2491 if (direction == DIRECTION_DOWN) {
2492 if (*i >= n - 1)
2493 return 0;
2494
2495 (*i) ++;
2496 } else {
2497 if (*i <= 0)
2498 return 0;
2499
2500 (*i) --;
2501 }
2502
2503 return 1;
2504 }
2505
2506 static bool check_properly_ordered(uint64_t new_offset, uint64_t old_offset, direction_t direction) {
2507
2508 /* Consider it an error if any of the two offsets is uninitialized */
2509 if (old_offset == 0 || new_offset == 0)
2510 return false;
2511
2512 /* If we go down, the new offset must be larger than the old one. */
2513 return direction == DIRECTION_DOWN ?
2514 new_offset > old_offset :
2515 new_offset < old_offset;
2516 }
2517
2518 int journal_file_next_entry(
2519 JournalFile *f,
2520 uint64_t p,
2521 direction_t direction,
2522 Object **ret, uint64_t *offset) {
2523
2524 uint64_t i, n, ofs;
2525 int r;
2526
2527 assert(f);
2528 assert(f->header);
2529
2530 n = le64toh(f->header->n_entries);
2531 if (n <= 0)
2532 return 0;
2533
2534 if (p == 0)
2535 i = direction == DIRECTION_DOWN ? 0 : n - 1;
2536 else {
2537 r = generic_array_bisect(f,
2538 le64toh(f->header->entry_array_offset),
2539 le64toh(f->header->n_entries),
2540 p,
2541 test_object_offset,
2542 DIRECTION_DOWN,
2543 NULL, NULL,
2544 &i);
2545 if (r <= 0)
2546 return r;
2547
2548 r = bump_array_index(&i, direction, n);
2549 if (r <= 0)
2550 return r;
2551 }
2552
2553 /* And jump to it */
2554 r = generic_array_get(f,
2555 le64toh(f->header->entry_array_offset),
2556 i,
2557 ret, &ofs);
2558 if (r == -EBADMSG && direction == DIRECTION_DOWN) {
2559 /* Special case: when we iterate throught the journal file linearly, and hit an entry we can't read,
2560 * consider this the end of the journal file. */
2561 log_debug_errno(r, "Encountered entry we can't read while iterating through journal file. Considering this the end of the file.");
2562 return 0;
2563 }
2564 if (r <= 0)
2565 return r;
2566
2567 /* Ensure our array is properly ordered. */
2568 if (p > 0 && !check_properly_ordered(ofs, p, direction)) {
2569 log_debug("%s: entry array not properly ordered at entry %" PRIu64, f->path, i);
2570 return -EBADMSG;
2571 }
2572
2573 if (offset)
2574 *offset = ofs;
2575
2576 return 1;
2577 }
2578
2579 int journal_file_next_entry_for_data(
2580 JournalFile *f,
2581 Object *o, uint64_t p,
2582 uint64_t data_offset,
2583 direction_t direction,
2584 Object **ret, uint64_t *offset) {
2585
2586 uint64_t n, i;
2587 int r;
2588 Object *d;
2589
2590 assert(f);
2591 assert(p > 0 || !o);
2592
2593 r = journal_file_move_to_object(f, OBJECT_DATA, data_offset, &d);
2594 if (r < 0)
2595 return r;
2596
2597 n = le64toh(d->data.n_entries);
2598 if (n <= 0)
2599 return n;
2600
2601 if (!o)
2602 i = direction == DIRECTION_DOWN ? 0 : n - 1;
2603 else {
2604 if (o->object.type != OBJECT_ENTRY)
2605 return -EINVAL;
2606
2607 r = generic_array_bisect_plus_one(f,
2608 le64toh(d->data.entry_offset),
2609 le64toh(d->data.entry_array_offset),
2610 le64toh(d->data.n_entries),
2611 p,
2612 test_object_offset,
2613 DIRECTION_DOWN,
2614 NULL, NULL,
2615 &i);
2616
2617 if (r <= 0)
2618 return r;
2619
2620 r = bump_array_index(&i, direction, n);
2621 if (r <= 0)
2622 return r;
2623 }
2624
2625 return generic_array_get_plus_one(f,
2626 le64toh(d->data.entry_offset),
2627 le64toh(d->data.entry_array_offset),
2628 i,
2629 ret, offset);
2630 }
2631
2632 int journal_file_move_to_entry_by_offset_for_data(
2633 JournalFile *f,
2634 uint64_t data_offset,
2635 uint64_t p,
2636 direction_t direction,
2637 Object **ret, uint64_t *offset) {
2638
2639 int r;
2640 Object *d;
2641
2642 assert(f);
2643
2644 r = journal_file_move_to_object(f, OBJECT_DATA, data_offset, &d);
2645 if (r < 0)
2646 return r;
2647
2648 return generic_array_bisect_plus_one(f,
2649 le64toh(d->data.entry_offset),
2650 le64toh(d->data.entry_array_offset),
2651 le64toh(d->data.n_entries),
2652 p,
2653 test_object_offset,
2654 direction,
2655 ret, offset, NULL);
2656 }
2657
2658 int journal_file_move_to_entry_by_monotonic_for_data(
2659 JournalFile *f,
2660 uint64_t data_offset,
2661 sd_id128_t boot_id,
2662 uint64_t monotonic,
2663 direction_t direction,
2664 Object **ret, uint64_t *offset) {
2665
2666 Object *o, *d;
2667 int r;
2668 uint64_t b, z;
2669
2670 assert(f);
2671
2672 /* First, seek by time */
2673 r = find_data_object_by_boot_id(f, boot_id, &o, &b);
2674 if (r < 0)
2675 return r;
2676 if (r == 0)
2677 return -ENOENT;
2678
2679 r = generic_array_bisect_plus_one(f,
2680 le64toh(o->data.entry_offset),
2681 le64toh(o->data.entry_array_offset),
2682 le64toh(o->data.n_entries),
2683 monotonic,
2684 test_object_monotonic,
2685 direction,
2686 NULL, &z, NULL);
2687 if (r <= 0)
2688 return r;
2689
2690 /* And now, continue seeking until we find an entry that
2691 * exists in both bisection arrays */
2692
2693 for (;;) {
2694 Object *qo;
2695 uint64_t p, q;
2696
2697 r = journal_file_move_to_object(f, OBJECT_DATA, data_offset, &d);
2698 if (r < 0)
2699 return r;
2700
2701 r = generic_array_bisect_plus_one(f,
2702 le64toh(d->data.entry_offset),
2703 le64toh(d->data.entry_array_offset),
2704 le64toh(d->data.n_entries),
2705 z,
2706 test_object_offset,
2707 direction,
2708 NULL, &p, NULL);
2709 if (r <= 0)
2710 return r;
2711
2712 r = journal_file_move_to_object(f, OBJECT_DATA, b, &o);
2713 if (r < 0)
2714 return r;
2715
2716 r = generic_array_bisect_plus_one(f,
2717 le64toh(o->data.entry_offset),
2718 le64toh(o->data.entry_array_offset),
2719 le64toh(o->data.n_entries),
2720 p,
2721 test_object_offset,
2722 direction,
2723 &qo, &q, NULL);
2724
2725 if (r <= 0)
2726 return r;
2727
2728 if (p == q) {
2729 if (ret)
2730 *ret = qo;
2731 if (offset)
2732 *offset = q;
2733
2734 return 1;
2735 }
2736
2737 z = q;
2738 }
2739 }
2740
2741 int journal_file_move_to_entry_by_seqnum_for_data(
2742 JournalFile *f,
2743 uint64_t data_offset,
2744 uint64_t seqnum,
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 seqnum,
2762 test_object_seqnum,
2763 direction,
2764 ret, offset, NULL);
2765 }
2766
2767 int journal_file_move_to_entry_by_realtime_for_data(
2768 JournalFile *f,
2769 uint64_t data_offset,
2770 uint64_t realtime,
2771 direction_t direction,
2772 Object **ret, uint64_t *offset) {
2773
2774 Object *d;
2775 int r;
2776
2777 assert(f);
2778
2779 r = journal_file_move_to_object(f, OBJECT_DATA, data_offset, &d);
2780 if (r < 0)
2781 return r;
2782
2783 return generic_array_bisect_plus_one(f,
2784 le64toh(d->data.entry_offset),
2785 le64toh(d->data.entry_array_offset),
2786 le64toh(d->data.n_entries),
2787 realtime,
2788 test_object_realtime,
2789 direction,
2790 ret, offset, NULL);
2791 }
2792
2793 void journal_file_dump(JournalFile *f) {
2794 Object *o;
2795 int r;
2796 uint64_t p;
2797
2798 assert(f);
2799 assert(f->header);
2800
2801 journal_file_print_header(f);
2802
2803 p = le64toh(f->header->header_size);
2804 while (p != 0) {
2805 r = journal_file_move_to_object(f, OBJECT_UNUSED, p, &o);
2806 if (r < 0)
2807 goto fail;
2808
2809 switch (o->object.type) {
2810
2811 case OBJECT_UNUSED:
2812 printf("Type: OBJECT_UNUSED\n");
2813 break;
2814
2815 case OBJECT_DATA:
2816 printf("Type: OBJECT_DATA\n");
2817 break;
2818
2819 case OBJECT_FIELD:
2820 printf("Type: OBJECT_FIELD\n");
2821 break;
2822
2823 case OBJECT_ENTRY:
2824 printf("Type: OBJECT_ENTRY seqnum=%"PRIu64" monotonic=%"PRIu64" realtime=%"PRIu64"\n",
2825 le64toh(o->entry.seqnum),
2826 le64toh(o->entry.monotonic),
2827 le64toh(o->entry.realtime));
2828 break;
2829
2830 case OBJECT_FIELD_HASH_TABLE:
2831 printf("Type: OBJECT_FIELD_HASH_TABLE\n");
2832 break;
2833
2834 case OBJECT_DATA_HASH_TABLE:
2835 printf("Type: OBJECT_DATA_HASH_TABLE\n");
2836 break;
2837
2838 case OBJECT_ENTRY_ARRAY:
2839 printf("Type: OBJECT_ENTRY_ARRAY\n");
2840 break;
2841
2842 case OBJECT_TAG:
2843 printf("Type: OBJECT_TAG seqnum=%"PRIu64" epoch=%"PRIu64"\n",
2844 le64toh(o->tag.seqnum),
2845 le64toh(o->tag.epoch));
2846 break;
2847
2848 default:
2849 printf("Type: unknown (%i)\n", o->object.type);
2850 break;
2851 }
2852
2853 if (o->object.flags & OBJECT_COMPRESSION_MASK)
2854 printf("Flags: %s\n",
2855 object_compressed_to_string(o->object.flags & OBJECT_COMPRESSION_MASK));
2856
2857 if (p == le64toh(f->header->tail_object_offset))
2858 p = 0;
2859 else
2860 p = p + ALIGN64(le64toh(o->object.size));
2861 }
2862
2863 return;
2864 fail:
2865 log_error("File corrupt");
2866 }
2867
2868 static const char* format_timestamp_safe(char *buf, size_t l, usec_t t) {
2869 const char *x;
2870
2871 x = format_timestamp(buf, l, t);
2872 if (x)
2873 return x;
2874 return " --- ";
2875 }
2876
2877 void journal_file_print_header(JournalFile *f) {
2878 char a[33], b[33], c[33], d[33];
2879 char x[FORMAT_TIMESTAMP_MAX], y[FORMAT_TIMESTAMP_MAX], z[FORMAT_TIMESTAMP_MAX];
2880 struct stat st;
2881 char bytes[FORMAT_BYTES_MAX];
2882
2883 assert(f);
2884 assert(f->header);
2885
2886 printf("File Path: %s\n"
2887 "File ID: %s\n"
2888 "Machine ID: %s\n"
2889 "Boot ID: %s\n"
2890 "Sequential Number ID: %s\n"
2891 "State: %s\n"
2892 "Compatible Flags:%s%s\n"
2893 "Incompatible Flags:%s%s%s\n"
2894 "Header size: %"PRIu64"\n"
2895 "Arena size: %"PRIu64"\n"
2896 "Data Hash Table Size: %"PRIu64"\n"
2897 "Field Hash Table Size: %"PRIu64"\n"
2898 "Rotate Suggested: %s\n"
2899 "Head Sequential Number: %"PRIu64" (%"PRIx64")\n"
2900 "Tail Sequential Number: %"PRIu64" (%"PRIx64")\n"
2901 "Head Realtime Timestamp: %s (%"PRIx64")\n"
2902 "Tail Realtime Timestamp: %s (%"PRIx64")\n"
2903 "Tail Monotonic Timestamp: %s (%"PRIx64")\n"
2904 "Objects: %"PRIu64"\n"
2905 "Entry Objects: %"PRIu64"\n",
2906 f->path,
2907 sd_id128_to_string(f->header->file_id, a),
2908 sd_id128_to_string(f->header->machine_id, b),
2909 sd_id128_to_string(f->header->boot_id, c),
2910 sd_id128_to_string(f->header->seqnum_id, d),
2911 f->header->state == STATE_OFFLINE ? "OFFLINE" :
2912 f->header->state == STATE_ONLINE ? "ONLINE" :
2913 f->header->state == STATE_ARCHIVED ? "ARCHIVED" : "UNKNOWN",
2914 JOURNAL_HEADER_SEALED(f->header) ? " SEALED" : "",
2915 (le32toh(f->header->compatible_flags) & ~HEADER_COMPATIBLE_ANY) ? " ???" : "",
2916 JOURNAL_HEADER_COMPRESSED_XZ(f->header) ? " COMPRESSED-XZ" : "",
2917 JOURNAL_HEADER_COMPRESSED_LZ4(f->header) ? " COMPRESSED-LZ4" : "",
2918 (le32toh(f->header->incompatible_flags) & ~HEADER_INCOMPATIBLE_ANY) ? " ???" : "",
2919 le64toh(f->header->header_size),
2920 le64toh(f->header->arena_size),
2921 le64toh(f->header->data_hash_table_size) / sizeof(HashItem),
2922 le64toh(f->header->field_hash_table_size) / sizeof(HashItem),
2923 yes_no(journal_file_rotate_suggested(f, 0)),
2924 le64toh(f->header->head_entry_seqnum), le64toh(f->header->head_entry_seqnum),
2925 le64toh(f->header->tail_entry_seqnum), le64toh(f->header->tail_entry_seqnum),
2926 format_timestamp_safe(x, sizeof(x), le64toh(f->header->head_entry_realtime)), le64toh(f->header->head_entry_realtime),
2927 format_timestamp_safe(y, sizeof(y), le64toh(f->header->tail_entry_realtime)), le64toh(f->header->tail_entry_realtime),
2928 format_timespan(z, sizeof(z), le64toh(f->header->tail_entry_monotonic), USEC_PER_MSEC), le64toh(f->header->tail_entry_monotonic),
2929 le64toh(f->header->n_objects),
2930 le64toh(f->header->n_entries));
2931
2932 if (JOURNAL_HEADER_CONTAINS(f->header, n_data))
2933 printf("Data Objects: %"PRIu64"\n"
2934 "Data Hash Table Fill: %.1f%%\n",
2935 le64toh(f->header->n_data),
2936 100.0 * (double) le64toh(f->header->n_data) / ((double) (le64toh(f->header->data_hash_table_size) / sizeof(HashItem))));
2937
2938 if (JOURNAL_HEADER_CONTAINS(f->header, n_fields))
2939 printf("Field Objects: %"PRIu64"\n"
2940 "Field Hash Table Fill: %.1f%%\n",
2941 le64toh(f->header->n_fields),
2942 100.0 * (double) le64toh(f->header->n_fields) / ((double) (le64toh(f->header->field_hash_table_size) / sizeof(HashItem))));
2943
2944 if (JOURNAL_HEADER_CONTAINS(f->header, n_tags))
2945 printf("Tag Objects: %"PRIu64"\n",
2946 le64toh(f->header->n_tags));
2947 if (JOURNAL_HEADER_CONTAINS(f->header, n_entry_arrays))
2948 printf("Entry Array Objects: %"PRIu64"\n",
2949 le64toh(f->header->n_entry_arrays));
2950
2951 if (fstat(f->fd, &st) >= 0)
2952 printf("Disk usage: %s\n", format_bytes(bytes, sizeof(bytes), (uint64_t) st.st_blocks * 512ULL));
2953 }
2954
2955 static int journal_file_warn_btrfs(JournalFile *f) {
2956 unsigned attrs;
2957 int r;
2958
2959 assert(f);
2960
2961 /* Before we write anything, check if the COW logic is turned
2962 * off on btrfs. Given our write pattern that is quite
2963 * unfriendly to COW file systems this should greatly improve
2964 * performance on COW file systems, such as btrfs, at the
2965 * expense of data integrity features (which shouldn't be too
2966 * bad, given that we do our own checksumming). */
2967
2968 r = btrfs_is_filesystem(f->fd);
2969 if (r < 0)
2970 return log_warning_errno(r, "Failed to determine if journal is on btrfs: %m");
2971 if (!r)
2972 return 0;
2973
2974 r = read_attr_fd(f->fd, &attrs);
2975 if (r < 0)
2976 return log_warning_errno(r, "Failed to read file attributes: %m");
2977
2978 if (attrs & FS_NOCOW_FL) {
2979 log_debug("Detected btrfs file system with copy-on-write disabled, all is good.");
2980 return 0;
2981 }
2982
2983 log_notice("Creating journal file %s on a btrfs file system, and copy-on-write is enabled. "
2984 "This is likely to slow down journal access substantially, please consider turning "
2985 "off the copy-on-write file attribute on the journal directory, using chattr +C.", f->path);
2986
2987 return 1;
2988 }
2989
2990 int journal_file_open(
2991 int fd,
2992 const char *fname,
2993 int flags,
2994 mode_t mode,
2995 bool compress,
2996 bool seal,
2997 JournalMetrics *metrics,
2998 MMapCache *mmap_cache,
2999 Set *deferred_closes,
3000 JournalFile *template,
3001 JournalFile **ret) {
3002
3003 bool newly_created = false;
3004 JournalFile *f;
3005 void *h;
3006 int r;
3007
3008 assert(ret);
3009 assert(fd >= 0 || fname);
3010
3011 if ((flags & O_ACCMODE) != O_RDONLY &&
3012 (flags & O_ACCMODE) != O_RDWR)
3013 return -EINVAL;
3014
3015 if (fname) {
3016 if (!endswith(fname, ".journal") &&
3017 !endswith(fname, ".journal~"))
3018 return -EINVAL;
3019 }
3020
3021 f = new0(JournalFile, 1);
3022 if (!f)
3023 return -ENOMEM;
3024
3025 f->fd = fd;
3026 f->mode = mode;
3027
3028 f->flags = flags;
3029 f->prot = prot_from_flags(flags);
3030 f->writable = (flags & O_ACCMODE) != O_RDONLY;
3031 #if defined(HAVE_LZ4)
3032 f->compress_lz4 = compress;
3033 #elif defined(HAVE_XZ)
3034 f->compress_xz = compress;
3035 #endif
3036 #ifdef HAVE_GCRYPT
3037 f->seal = seal;
3038 #endif
3039
3040 if (mmap_cache)
3041 f->mmap = mmap_cache_ref(mmap_cache);
3042 else {
3043 f->mmap = mmap_cache_new();
3044 if (!f->mmap) {
3045 r = -ENOMEM;
3046 goto fail;
3047 }
3048 }
3049
3050 if (fname)
3051 f->path = strdup(fname);
3052 else /* If we don't know the path, fill in something explanatory and vaguely useful */
3053 asprintf(&f->path, "/proc/self/%i", fd);
3054 if (!f->path) {
3055 r = -ENOMEM;
3056 goto fail;
3057 }
3058
3059 f->chain_cache = ordered_hashmap_new(&uint64_hash_ops);
3060 if (!f->chain_cache) {
3061 r = -ENOMEM;
3062 goto fail;
3063 }
3064
3065 if (f->fd < 0) {
3066 f->fd = open(f->path, f->flags|O_CLOEXEC, f->mode);
3067 if (f->fd < 0) {
3068 r = -errno;
3069 goto fail;
3070 }
3071
3072 /* fds we opened here by us should also be closed by us. */
3073 f->close_fd = true;
3074 }
3075
3076 r = journal_file_fstat(f);
3077 if (r < 0)
3078 goto fail;
3079
3080 if (f->last_stat.st_size == 0 && f->writable) {
3081
3082 (void) journal_file_warn_btrfs(f);
3083
3084 /* Let's attach the creation time to the journal file,
3085 * so that the vacuuming code knows the age of this
3086 * file even if the file might end up corrupted one
3087 * day... Ideally we'd just use the creation time many
3088 * file systems maintain for each file, but there is
3089 * currently no usable API to query this, hence let's
3090 * emulate this via extended attributes. If extended
3091 * attributes are not supported we'll just skip this,
3092 * and rely solely on mtime/atime/ctime of the file. */
3093
3094 fd_setcrtime(f->fd, 0);
3095
3096 #ifdef HAVE_GCRYPT
3097 /* Try to load the FSPRG state, and if we can't, then
3098 * just don't do sealing */
3099 if (f->seal) {
3100 r = journal_file_fss_load(f);
3101 if (r < 0)
3102 f->seal = false;
3103 }
3104 #endif
3105
3106 r = journal_file_init_header(f, template);
3107 if (r < 0)
3108 goto fail;
3109
3110 r = journal_file_fstat(f);
3111 if (r < 0)
3112 goto fail;
3113
3114 newly_created = true;
3115 }
3116
3117 if (f->last_stat.st_size < (off_t) HEADER_SIZE_MIN) {
3118 r = -ENODATA;
3119 goto fail;
3120 }
3121
3122 r = mmap_cache_get(f->mmap, f->fd, f->prot, CONTEXT_HEADER, true, 0, PAGE_ALIGN(sizeof(Header)), &f->last_stat, &h);
3123 if (r < 0)
3124 goto fail;
3125
3126 f->header = h;
3127
3128 if (!newly_created) {
3129 if (deferred_closes)
3130 journal_file_close_set(deferred_closes);
3131
3132 r = journal_file_verify_header(f);
3133 if (r < 0)
3134 goto fail;
3135 }
3136
3137 #ifdef HAVE_GCRYPT
3138 if (!newly_created && f->writable) {
3139 r = journal_file_fss_load(f);
3140 if (r < 0)
3141 goto fail;
3142 }
3143 #endif
3144
3145 if (f->writable) {
3146 if (metrics) {
3147 journal_default_metrics(metrics, f->fd);
3148 f->metrics = *metrics;
3149 } else if (template)
3150 f->metrics = template->metrics;
3151
3152 r = journal_file_refresh_header(f);
3153 if (r < 0)
3154 goto fail;
3155 }
3156
3157 #ifdef HAVE_GCRYPT
3158 r = journal_file_hmac_setup(f);
3159 if (r < 0)
3160 goto fail;
3161 #endif
3162
3163 if (newly_created) {
3164 r = journal_file_setup_field_hash_table(f);
3165 if (r < 0)
3166 goto fail;
3167
3168 r = journal_file_setup_data_hash_table(f);
3169 if (r < 0)
3170 goto fail;
3171
3172 #ifdef HAVE_GCRYPT
3173 r = journal_file_append_first_tag(f);
3174 if (r < 0)
3175 goto fail;
3176 #endif
3177 }
3178
3179 if (mmap_cache_got_sigbus(f->mmap, f->fd)) {
3180 r = -EIO;
3181 goto fail;
3182 }
3183
3184 if (template && template->post_change_timer) {
3185 r = journal_file_enable_post_change_timer(
3186 f,
3187 sd_event_source_get_event(template->post_change_timer),
3188 template->post_change_timer_period);
3189
3190 if (r < 0)
3191 goto fail;
3192 }
3193
3194 /* The file is opened now successfully, thus we take possession of any passed in fd. */
3195 f->close_fd = true;
3196
3197 *ret = f;
3198 return 0;
3199
3200 fail:
3201 if (f->fd >= 0 && mmap_cache_got_sigbus(f->mmap, f->fd))
3202 r = -EIO;
3203
3204 (void) journal_file_close(f);
3205
3206 return r;
3207 }
3208
3209 int journal_file_rotate(JournalFile **f, bool compress, bool seal, Set *deferred_closes) {
3210 _cleanup_free_ char *p = NULL;
3211 size_t l;
3212 JournalFile *old_file, *new_file = NULL;
3213 int r;
3214
3215 assert(f);
3216 assert(*f);
3217
3218 old_file = *f;
3219
3220 if (!old_file->writable)
3221 return -EINVAL;
3222
3223 /* Is this a journal file that was passed to us as fd? If so, we synthesized a path name for it, and we refuse
3224 * rotation, since we don't know the actual path, and couldn't rename the file hence.*/
3225 if (path_startswith(old_file->path, "/proc/self/fd"))
3226 return -EINVAL;
3227
3228 if (!endswith(old_file->path, ".journal"))
3229 return -EINVAL;
3230
3231 l = strlen(old_file->path);
3232 r = asprintf(&p, "%.*s@" SD_ID128_FORMAT_STR "-%016"PRIx64"-%016"PRIx64".journal",
3233 (int) l - 8, old_file->path,
3234 SD_ID128_FORMAT_VAL(old_file->header->seqnum_id),
3235 le64toh((*f)->header->head_entry_seqnum),
3236 le64toh((*f)->header->head_entry_realtime));
3237 if (r < 0)
3238 return -ENOMEM;
3239
3240 /* Try to rename the file to the archived version. If the file
3241 * already was deleted, we'll get ENOENT, let's ignore that
3242 * case. */
3243 r = rename(old_file->path, p);
3244 if (r < 0 && errno != ENOENT)
3245 return -errno;
3246
3247 /* Sync the rename to disk */
3248 (void) fsync_directory_of_file(old_file->fd);
3249
3250 /* Set as archive so offlining commits w/state=STATE_ARCHIVED.
3251 * Previously we would set old_file->header->state to STATE_ARCHIVED directly here,
3252 * but journal_file_set_offline() short-circuits when state != STATE_ONLINE, which
3253 * would result in the rotated journal never getting fsync() called before closing.
3254 * Now we simply queue the archive state by setting an archive bit, leaving the state
3255 * as STATE_ONLINE so proper offlining occurs. */
3256 old_file->archive = true;
3257
3258 /* Currently, btrfs is not very good with out write patterns
3259 * and fragments heavily. Let's defrag our journal files when
3260 * we archive them */
3261 old_file->defrag_on_close = true;
3262
3263 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);
3264
3265 if (deferred_closes &&
3266 set_put(deferred_closes, old_file) >= 0)
3267 (void) journal_file_set_offline(old_file, false);
3268 else
3269 (void) journal_file_close(old_file);
3270
3271 *f = new_file;
3272 return r;
3273 }
3274
3275 int journal_file_open_reliably(
3276 const char *fname,
3277 int flags,
3278 mode_t mode,
3279 bool compress,
3280 bool seal,
3281 JournalMetrics *metrics,
3282 MMapCache *mmap_cache,
3283 Set *deferred_closes,
3284 JournalFile *template,
3285 JournalFile **ret) {
3286
3287 int r;
3288 size_t l;
3289 _cleanup_free_ char *p = NULL;
3290
3291 r = journal_file_open(-1, fname, flags, mode, compress, seal, metrics, mmap_cache, deferred_closes, template, ret);
3292 if (!IN_SET(r,
3293 -EBADMSG, /* corrupted */
3294 -ENODATA, /* truncated */
3295 -EHOSTDOWN, /* other machine */
3296 -EPROTONOSUPPORT, /* incompatible feature */
3297 -EBUSY, /* unclean shutdown */
3298 -ESHUTDOWN, /* already archived */
3299 -EIO, /* IO error, including SIGBUS on mmap */
3300 -EIDRM /* File has been deleted */))
3301 return r;
3302
3303 if ((flags & O_ACCMODE) == O_RDONLY)
3304 return r;
3305
3306 if (!(flags & O_CREAT))
3307 return r;
3308
3309 if (!endswith(fname, ".journal"))
3310 return r;
3311
3312 /* The file is corrupted. Rotate it away and try it again (but only once) */
3313
3314 l = strlen(fname);
3315 if (asprintf(&p, "%.*s@%016"PRIx64 "-%016"PRIx64 ".journal~",
3316 (int) l - 8, fname,
3317 now(CLOCK_REALTIME),
3318 random_u64()) < 0)
3319 return -ENOMEM;
3320
3321 if (rename(fname, p) < 0)
3322 return -errno;
3323
3324 /* btrfs doesn't cope well with our write pattern and
3325 * fragments heavily. Let's defrag all files we rotate */
3326
3327 (void) chattr_path(p, 0, FS_NOCOW_FL);
3328 (void) btrfs_defrag(p);
3329
3330 log_warning_errno(r, "File %s corrupted or uncleanly shut down, renaming and replacing.", fname);
3331
3332 return journal_file_open(-1, fname, flags, mode, compress, seal, metrics, mmap_cache, deferred_closes, template, ret);
3333 }
3334
3335 int journal_file_copy_entry(JournalFile *from, JournalFile *to, Object *o, uint64_t p, uint64_t *seqnum, Object **ret, uint64_t *offset) {
3336 uint64_t i, n;
3337 uint64_t q, xor_hash = 0;
3338 int r;
3339 EntryItem *items;
3340 dual_timestamp ts;
3341
3342 assert(from);
3343 assert(to);
3344 assert(o);
3345 assert(p);
3346
3347 if (!to->writable)
3348 return -EPERM;
3349
3350 ts.monotonic = le64toh(o->entry.monotonic);
3351 ts.realtime = le64toh(o->entry.realtime);
3352
3353 n = journal_file_entry_n_items(o);
3354 /* alloca() can't take 0, hence let's allocate at least one */
3355 items = alloca(sizeof(EntryItem) * MAX(1u, n));
3356
3357 for (i = 0; i < n; i++) {
3358 uint64_t l, h;
3359 le64_t le_hash;
3360 size_t t;
3361 void *data;
3362 Object *u;
3363
3364 q = le64toh(o->entry.items[i].object_offset);
3365 le_hash = o->entry.items[i].hash;
3366
3367 r = journal_file_move_to_object(from, OBJECT_DATA, q, &o);
3368 if (r < 0)
3369 return r;
3370
3371 if (le_hash != o->data.hash)
3372 return -EBADMSG;
3373
3374 l = le64toh(o->object.size) - offsetof(Object, data.payload);
3375 t = (size_t) l;
3376
3377 /* We hit the limit on 32bit machines */
3378 if ((uint64_t) t != l)
3379 return -E2BIG;
3380
3381 if (o->object.flags & OBJECT_COMPRESSION_MASK) {
3382 #if defined(HAVE_XZ) || defined(HAVE_LZ4)
3383 size_t rsize = 0;
3384
3385 r = decompress_blob(o->object.flags & OBJECT_COMPRESSION_MASK,
3386 o->data.payload, l, &from->compress_buffer, &from->compress_buffer_size, &rsize, 0);
3387 if (r < 0)
3388 return r;
3389
3390 data = from->compress_buffer;
3391 l = rsize;
3392 #else
3393 return -EPROTONOSUPPORT;
3394 #endif
3395 } else
3396 data = o->data.payload;
3397
3398 r = journal_file_append_data(to, data, l, &u, &h);
3399 if (r < 0)
3400 return r;
3401
3402 xor_hash ^= le64toh(u->data.hash);
3403 items[i].object_offset = htole64(h);
3404 items[i].hash = u->data.hash;
3405
3406 r = journal_file_move_to_object(from, OBJECT_ENTRY, p, &o);
3407 if (r < 0)
3408 return r;
3409 }
3410
3411 r = journal_file_append_entry_internal(to, &ts, xor_hash, items, n, seqnum, ret, offset);
3412
3413 if (mmap_cache_got_sigbus(to->mmap, to->fd))
3414 return -EIO;
3415
3416 return r;
3417 }
3418
3419 void journal_reset_metrics(JournalMetrics *m) {
3420 assert(m);
3421
3422 /* Set everything to "pick automatic values". */
3423
3424 *m = (JournalMetrics) {
3425 .min_use = (uint64_t) -1,
3426 .max_use = (uint64_t) -1,
3427 .min_size = (uint64_t) -1,
3428 .max_size = (uint64_t) -1,
3429 .keep_free = (uint64_t) -1,
3430 .n_max_files = (uint64_t) -1,
3431 };
3432 }
3433
3434 void journal_default_metrics(JournalMetrics *m, int fd) {
3435 char a[FORMAT_BYTES_MAX], b[FORMAT_BYTES_MAX], c[FORMAT_BYTES_MAX], d[FORMAT_BYTES_MAX], e[FORMAT_BYTES_MAX];
3436 struct statvfs ss;
3437 uint64_t fs_size;
3438
3439 assert(m);
3440 assert(fd >= 0);
3441
3442 if (fstatvfs(fd, &ss) >= 0)
3443 fs_size = ss.f_frsize * ss.f_blocks;
3444 else {
3445 log_debug_errno(errno, "Failed to detremine disk size: %m");
3446 fs_size = 0;
3447 }
3448
3449 if (m->max_use == (uint64_t) -1) {
3450
3451 if (fs_size > 0) {
3452 m->max_use = PAGE_ALIGN(fs_size / 10); /* 10% of file system size */
3453
3454 if (m->max_use > DEFAULT_MAX_USE_UPPER)
3455 m->max_use = DEFAULT_MAX_USE_UPPER;
3456
3457 if (m->max_use < DEFAULT_MAX_USE_LOWER)
3458 m->max_use = DEFAULT_MAX_USE_LOWER;
3459 } else
3460 m->max_use = DEFAULT_MAX_USE_LOWER;
3461 } else {
3462 m->max_use = PAGE_ALIGN(m->max_use);
3463
3464 if (m->max_use != 0 && m->max_use < JOURNAL_FILE_SIZE_MIN*2)
3465 m->max_use = JOURNAL_FILE_SIZE_MIN*2;
3466 }
3467
3468 if (m->min_use == (uint64_t) -1)
3469 m->min_use = DEFAULT_MIN_USE;
3470
3471 if (m->min_use > m->max_use)
3472 m->min_use = m->max_use;
3473
3474 if (m->max_size == (uint64_t) -1) {
3475 m->max_size = PAGE_ALIGN(m->max_use / 8); /* 8 chunks */
3476
3477 if (m->max_size > DEFAULT_MAX_SIZE_UPPER)
3478 m->max_size = DEFAULT_MAX_SIZE_UPPER;
3479 } else
3480 m->max_size = PAGE_ALIGN(m->max_size);
3481
3482 if (m->max_size != 0) {
3483 if (m->max_size < JOURNAL_FILE_SIZE_MIN)
3484 m->max_size = JOURNAL_FILE_SIZE_MIN;
3485
3486 if (m->max_use != 0 && m->max_size*2 > m->max_use)
3487 m->max_use = m->max_size*2;
3488 }
3489
3490 if (m->min_size == (uint64_t) -1)
3491 m->min_size = JOURNAL_FILE_SIZE_MIN;
3492 else {
3493 m->min_size = PAGE_ALIGN(m->min_size);
3494
3495 if (m->min_size < JOURNAL_FILE_SIZE_MIN)
3496 m->min_size = JOURNAL_FILE_SIZE_MIN;
3497
3498 if (m->max_size != 0 && m->min_size > m->max_size)
3499 m->max_size = m->min_size;
3500 }
3501
3502 if (m->keep_free == (uint64_t) -1) {
3503
3504 if (fs_size > 0) {
3505 m->keep_free = PAGE_ALIGN(fs_size * 3 / 20); /* 15% of file system size */
3506
3507 if (m->keep_free > DEFAULT_KEEP_FREE_UPPER)
3508 m->keep_free = DEFAULT_KEEP_FREE_UPPER;
3509
3510 } else
3511 m->keep_free = DEFAULT_KEEP_FREE;
3512 }
3513
3514 if (m->n_max_files == (uint64_t) -1)
3515 m->n_max_files = DEFAULT_N_MAX_FILES;
3516
3517 log_debug("Fixed min_use=%s max_use=%s max_size=%s min_size=%s keep_free=%s n_max_files=%" PRIu64,
3518 format_bytes(a, sizeof(a), m->min_use),
3519 format_bytes(b, sizeof(b), m->max_use),
3520 format_bytes(c, sizeof(c), m->max_size),
3521 format_bytes(d, sizeof(d), m->min_size),
3522 format_bytes(e, sizeof(e), m->keep_free),
3523 m->n_max_files);
3524 }
3525
3526 int journal_file_get_cutoff_realtime_usec(JournalFile *f, usec_t *from, usec_t *to) {
3527 assert(f);
3528 assert(f->header);
3529 assert(from || to);
3530
3531 if (from) {
3532 if (f->header->head_entry_realtime == 0)
3533 return -ENOENT;
3534
3535 *from = le64toh(f->header->head_entry_realtime);
3536 }
3537
3538 if (to) {
3539 if (f->header->tail_entry_realtime == 0)
3540 return -ENOENT;
3541
3542 *to = le64toh(f->header->tail_entry_realtime);
3543 }
3544
3545 return 1;
3546 }
3547
3548 int journal_file_get_cutoff_monotonic_usec(JournalFile *f, sd_id128_t boot_id, usec_t *from, usec_t *to) {
3549 Object *o;
3550 uint64_t p;
3551 int r;
3552
3553 assert(f);
3554 assert(from || to);
3555
3556 r = find_data_object_by_boot_id(f, boot_id, &o, &p);
3557 if (r <= 0)
3558 return r;
3559
3560 if (le64toh(o->data.n_entries) <= 0)
3561 return 0;
3562
3563 if (from) {
3564 r = journal_file_move_to_object(f, OBJECT_ENTRY, le64toh(o->data.entry_offset), &o);
3565 if (r < 0)
3566 return r;
3567
3568 *from = le64toh(o->entry.monotonic);
3569 }
3570
3571 if (to) {
3572 r = journal_file_move_to_object(f, OBJECT_DATA, p, &o);
3573 if (r < 0)
3574 return r;
3575
3576 r = generic_array_get_plus_one(f,
3577 le64toh(o->data.entry_offset),
3578 le64toh(o->data.entry_array_offset),
3579 le64toh(o->data.n_entries)-1,
3580 &o, NULL);
3581 if (r <= 0)
3582 return r;
3583
3584 *to = le64toh(o->entry.monotonic);
3585 }
3586
3587 return 1;
3588 }
3589
3590 bool journal_file_rotate_suggested(JournalFile *f, usec_t max_file_usec) {
3591 assert(f);
3592 assert(f->header);
3593
3594 /* If we gained new header fields we gained new features,
3595 * hence suggest a rotation */
3596 if (le64toh(f->header->header_size) < sizeof(Header)) {
3597 log_debug("%s uses an outdated header, suggesting rotation.", f->path);
3598 return true;
3599 }
3600
3601 /* Let's check if the hash tables grew over a certain fill
3602 * level (75%, borrowing this value from Java's hash table
3603 * implementation), and if so suggest a rotation. To calculate
3604 * the fill level we need the n_data field, which only exists
3605 * in newer versions. */
3606
3607 if (JOURNAL_HEADER_CONTAINS(f->header, n_data))
3608 if (le64toh(f->header->n_data) * 4ULL > (le64toh(f->header->data_hash_table_size) / sizeof(HashItem)) * 3ULL) {
3609 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.",
3610 f->path,
3611 100.0 * (double) le64toh(f->header->n_data) / ((double) (le64toh(f->header->data_hash_table_size) / sizeof(HashItem))),
3612 le64toh(f->header->n_data),
3613 le64toh(f->header->data_hash_table_size) / sizeof(HashItem),
3614 (unsigned long long) f->last_stat.st_size,
3615 f->last_stat.st_size / le64toh(f->header->n_data));
3616 return true;
3617 }
3618
3619 if (JOURNAL_HEADER_CONTAINS(f->header, n_fields))
3620 if (le64toh(f->header->n_fields) * 4ULL > (le64toh(f->header->field_hash_table_size) / sizeof(HashItem)) * 3ULL) {
3621 log_debug("Field hash table of %s has a fill level at %.1f (%"PRIu64" of %"PRIu64" items), suggesting rotation.",
3622 f->path,
3623 100.0 * (double) le64toh(f->header->n_fields) / ((double) (le64toh(f->header->field_hash_table_size) / sizeof(HashItem))),
3624 le64toh(f->header->n_fields),
3625 le64toh(f->header->field_hash_table_size) / sizeof(HashItem));
3626 return true;
3627 }
3628
3629 /* Are the data objects properly indexed by field objects? */
3630 if (JOURNAL_HEADER_CONTAINS(f->header, n_data) &&
3631 JOURNAL_HEADER_CONTAINS(f->header, n_fields) &&
3632 le64toh(f->header->n_data) > 0 &&
3633 le64toh(f->header->n_fields) == 0)
3634 return true;
3635
3636 if (max_file_usec > 0) {
3637 usec_t t, h;
3638
3639 h = le64toh(f->header->head_entry_realtime);
3640 t = now(CLOCK_REALTIME);
3641
3642 if (h > 0 && t > h + max_file_usec)
3643 return true;
3644 }
3645
3646 return false;
3647 }
Unnamed repository; edit this file 'description' to name the repository.