]> git.ipfire.org Git - thirdparty/systemd.git/blob - src/journal/journal-file.c
projects / thirdparty / systemd.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Merge pull request #10525 from poettering/journal-vaccum-all
[thirdparty/systemd.git] / src / journal / journal-file.c
1 /* SPDX-License-Identifier: LGPL-2.1+ */
2
3 #include <errno.h>
4 #include <fcntl.h>
5 #include <linux/fs.h>
6 #include <pthread.h>
7 #include <stddef.h>
8 #include <sys/mman.h>
9 #include <sys/statvfs.h>
10 #include <sys/uio.h>
11 #include <unistd.h>
12
13 #include "alloc-util.h"
14 #include "btrfs-util.h"
15 #include "chattr-util.h"
16 #include "compress.h"
17 #include "fd-util.h"
18 #include "fs-util.h"
19 #include "journal-authenticate.h"
20 #include "journal-def.h"
21 #include "journal-file.h"
22 #include "lookup3.h"
23 #include "parse-util.h"
24 #include "path-util.h"
25 #include "random-util.h"
26 #include "sd-event.h"
27 #include "set.h"
28 #include "stat-util.h"
29 #include "string-util.h"
30 #include "strv.h"
31 #include "xattr-util.h"
32
33 #define DEFAULT_DATA_HASH_TABLE_SIZE (2047ULL*sizeof(HashItem))
34 #define DEFAULT_FIELD_HASH_TABLE_SIZE (333ULL*sizeof(HashItem))
35
36 #define DEFAULT_COMPRESS_THRESHOLD (512ULL)
37 #define MIN_COMPRESS_THRESHOLD (8ULL)
38
39 /* This is the minimum journal file size */
40 #define JOURNAL_FILE_SIZE_MIN (512ULL*1024ULL) /* 512 KiB */
41
42 /* These are the lower and upper bounds if we deduce the max_use value
43 * from the file system size */
44 #define DEFAULT_MAX_USE_LOWER (1ULL*1024ULL*1024ULL) /* 1 MiB */
45 #define DEFAULT_MAX_USE_UPPER (4ULL*1024ULL*1024ULL*1024ULL) /* 4 GiB */
46
47 /* This is the default minimal use limit, how much we'll use even if keep_free suggests otherwise. */
48 #define DEFAULT_MIN_USE (1ULL*1024ULL*1024ULL) /* 1 MiB */
49
50 /* This is the upper bound if we deduce max_size from max_use */
51 #define DEFAULT_MAX_SIZE_UPPER (128ULL*1024ULL*1024ULL) /* 128 MiB */
52
53 /* This is the upper bound if we deduce the keep_free value from the
54 * file system size */
55 #define DEFAULT_KEEP_FREE_UPPER (4ULL*1024ULL*1024ULL*1024ULL) /* 4 GiB */
56
57 /* This is the keep_free value when we can't determine the system
58 * size */
59 #define DEFAULT_KEEP_FREE (1024ULL*1024ULL) /* 1 MB */
60
61 /* This is the default maximum number of journal files to keep around. */
62 #define DEFAULT_N_MAX_FILES (100)
63
64 /* n_data was the first entry we added after the initial file format design */
65 #define HEADER_SIZE_MIN ALIGN64(offsetof(Header, n_data))
66
67 /* How many entries to keep in the entry array chain cache at max */
68 #define CHAIN_CACHE_MAX 20
69
70 /* How much to increase the journal file size at once each time we allocate something new. */
71 #define FILE_SIZE_INCREASE (8ULL*1024ULL*1024ULL) /* 8MB */
72
73 /* Reread fstat() of the file for detecting deletions at least this often */
74 #define LAST_STAT_REFRESH_USEC (5*USEC_PER_SEC)
75
76 /* The mmap context to use for the header we pick as one above the last defined typed */
77 #define CONTEXT_HEADER _OBJECT_TYPE_MAX
78
79 #ifdef __clang__
80 # pragma GCC diagnostic ignored "-Waddress-of-packed-member"
81 #endif
82
83 /* This may be called from a separate thread to prevent blocking the caller for the duration of fsync().
84 * As a result we use atomic operations on f->offline_state for inter-thread communications with
85 * journal_file_set_offline() and journal_file_set_online(). */
86 static void journal_file_set_offline_internal(JournalFile *f) {
87 assert(f);
88 assert(f->fd >= 0);
89 assert(f->header);
90
91 for (;;) {
92 switch (f->offline_state) {
93 case OFFLINE_CANCEL:
94 if (!__sync_bool_compare_and_swap(&f->offline_state, OFFLINE_CANCEL, OFFLINE_DONE))
95 continue;
96 return;
97
98 case OFFLINE_AGAIN_FROM_SYNCING:
99 if (!__sync_bool_compare_and_swap(&f->offline_state, OFFLINE_AGAIN_FROM_SYNCING, OFFLINE_SYNCING))
100 continue;
101 break;
102
103 case OFFLINE_AGAIN_FROM_OFFLINING:
104 if (!__sync_bool_compare_and_swap(&f->offline_state, OFFLINE_AGAIN_FROM_OFFLINING, OFFLINE_SYNCING))
105 continue;
106 break;
107
108 case OFFLINE_SYNCING:
109 (void) fsync(f->fd);
110
111 if (!__sync_bool_compare_and_swap(&f->offline_state, OFFLINE_SYNCING, OFFLINE_OFFLINING))
112 continue;
113
114 f->header->state = f->archive ? STATE_ARCHIVED : STATE_OFFLINE;
115 (void) fsync(f->fd);
116 break;
117
118 case OFFLINE_OFFLINING:
119 if (!__sync_bool_compare_and_swap(&f->offline_state, OFFLINE_OFFLINING, OFFLINE_DONE))
120 continue;
121 _fallthrough_;
122 case OFFLINE_DONE:
123 return;
124
125 case OFFLINE_JOINED:
126 log_debug("OFFLINE_JOINED unexpected offline state for journal_file_set_offline_internal()");
127 return;
128 }
129 }
130 }
131
132 static void * journal_file_set_offline_thread(void *arg) {
133 JournalFile *f = arg;
134
135 (void) pthread_setname_np(pthread_self(), "journal-offline");
136
137 journal_file_set_offline_internal(f);
138
139 return NULL;
140 }
141
142 static int journal_file_set_offline_thread_join(JournalFile *f) {
143 int r;
144
145 assert(f);
146
147 if (f->offline_state == OFFLINE_JOINED)
148 return 0;
149
150 r = pthread_join(f->offline_thread, NULL);
151 if (r)
152 return -r;
153
154 f->offline_state = OFFLINE_JOINED;
155
156 if (mmap_cache_got_sigbus(f->mmap, f->cache_fd))
157 return -EIO;
158
159 return 0;
160 }
161
162 /* Trigger a restart if the offline thread is mid-flight in a restartable state. */
163 static bool journal_file_set_offline_try_restart(JournalFile *f) {
164 for (;;) {
165 switch (f->offline_state) {
166 case OFFLINE_AGAIN_FROM_SYNCING:
167 case OFFLINE_AGAIN_FROM_OFFLINING:
168 return true;
169
170 case OFFLINE_CANCEL:
171 if (!__sync_bool_compare_and_swap(&f->offline_state, OFFLINE_CANCEL, OFFLINE_AGAIN_FROM_SYNCING))
172 continue;
173 return true;
174
175 case OFFLINE_SYNCING:
176 if (!__sync_bool_compare_and_swap(&f->offline_state, OFFLINE_SYNCING, OFFLINE_AGAIN_FROM_SYNCING))
177 continue;
178 return true;
179
180 case OFFLINE_OFFLINING:
181 if (!__sync_bool_compare_and_swap(&f->offline_state, OFFLINE_OFFLINING, OFFLINE_AGAIN_FROM_OFFLINING))
182 continue;
183 return true;
184
185 default:
186 return false;
187 }
188 }
189 }
190
191 /* Sets a journal offline.
192 *
193 * If wait is false then an offline is dispatched in a separate thread for a
194 * subsequent journal_file_set_offline() or journal_file_set_online() of the
195 * same journal to synchronize with.
196 *
197 * If wait is true, then either an existing offline thread will be restarted
198 * and joined, or if none exists the offline is simply performed in this
199 * context without involving another thread.
200 */
201 int journal_file_set_offline(JournalFile *f, bool wait) {
202 bool restarted;
203 int r;
204
205 assert(f);
206
207 if (!f->writable)
208 return -EPERM;
209
210 if (f->fd < 0 || !f->header)
211 return -EINVAL;
212
213 /* An offlining journal is implicitly online and may modify f->header->state,
214 * we must also join any potentially lingering offline thread when not online. */
215 if (!journal_file_is_offlining(f) && f->header->state != STATE_ONLINE)
216 return journal_file_set_offline_thread_join(f);
217
218 /* Restart an in-flight offline thread and wait if needed, or join a lingering done one. */
219 restarted = journal_file_set_offline_try_restart(f);
220 if ((restarted && wait) || !restarted) {
221 r = journal_file_set_offline_thread_join(f);
222 if (r < 0)
223 return r;
224 }
225
226 if (restarted)
227 return 0;
228
229 /* Initiate a new offline. */
230 f->offline_state = OFFLINE_SYNCING;
231
232 if (wait) /* Without using a thread if waiting. */
233 journal_file_set_offline_internal(f);
234 else {
235 sigset_t ss, saved_ss;
236 int k;
237
238 if (sigfillset(&ss) < 0)
239 return -errno;
240
241 r = pthread_sigmask(SIG_BLOCK, &ss, &saved_ss);
242 if (r > 0)
243 return -r;
244
245 r = pthread_create(&f->offline_thread, NULL, journal_file_set_offline_thread, f);
246
247 k = pthread_sigmask(SIG_SETMASK, &saved_ss, NULL);
248 if (r > 0) {
249 f->offline_state = OFFLINE_JOINED;
250 return -r;
251 }
252 if (k > 0)
253 return -k;
254 }
255
256 return 0;
257 }
258
259 static int journal_file_set_online(JournalFile *f) {
260 bool wait = true;
261
262 assert(f);
263
264 if (!f->writable)
265 return -EPERM;
266
267 if (f->fd < 0 || !f->header)
268 return -EINVAL;
269
270 while (wait) {
271 switch (f->offline_state) {
272 case OFFLINE_JOINED:
273 /* No offline thread, no need to wait. */
274 wait = false;
275 break;
276
277 case OFFLINE_SYNCING:
278 if (!__sync_bool_compare_and_swap(&f->offline_state, OFFLINE_SYNCING, OFFLINE_CANCEL))
279 continue;
280 /* Canceled syncing prior to offlining, no need to wait. */
281 wait = false;
282 break;
283
284 case OFFLINE_AGAIN_FROM_SYNCING:
285 if (!__sync_bool_compare_and_swap(&f->offline_state, OFFLINE_AGAIN_FROM_SYNCING, OFFLINE_CANCEL))
286 continue;
287 /* Canceled restart from syncing, no need to wait. */
288 wait = false;
289 break;
290
291 case OFFLINE_AGAIN_FROM_OFFLINING:
292 if (!__sync_bool_compare_and_swap(&f->offline_state, OFFLINE_AGAIN_FROM_OFFLINING, OFFLINE_CANCEL))
293 continue;
294 /* Canceled restart from offlining, must wait for offlining to complete however. */
295 _fallthrough_;
296 default: {
297 int r;
298
299 r = journal_file_set_offline_thread_join(f);
300 if (r < 0)
301 return r;
302
303 wait = false;
304 break;
305 }
306 }
307 }
308
309 if (mmap_cache_got_sigbus(f->mmap, f->cache_fd))
310 return -EIO;
311
312 switch (f->header->state) {
313 case STATE_ONLINE:
314 return 0;
315
316 case STATE_OFFLINE:
317 f->header->state = STATE_ONLINE;
318 (void) fsync(f->fd);
319 return 0;
320
321 default:
322 return -EINVAL;
323 }
324 }
325
326 bool journal_file_is_offlining(JournalFile *f) {
327 assert(f);
328
329 __sync_synchronize();
330
331 if (IN_SET(f->offline_state, OFFLINE_DONE, OFFLINE_JOINED))
332 return false;
333
334 return true;
335 }
336
337 JournalFile* journal_file_close(JournalFile *f) {
338 assert(f);
339
340 #if HAVE_GCRYPT
341 /* Write the final tag */
342 if (f->seal && f->writable) {
343 int r;
344
345 r = journal_file_append_tag(f);
346 if (r < 0)
347 log_error_errno(r, "Failed to append tag when closing journal: %m");
348 }
349 #endif
350
351 if (f->post_change_timer) {
352 int enabled;
353
354 if (sd_event_source_get_enabled(f->post_change_timer, &enabled) >= 0)
355 if (enabled == SD_EVENT_ONESHOT)
356 journal_file_post_change(f);
357
358 (void) sd_event_source_set_enabled(f->post_change_timer, SD_EVENT_OFF);
359 sd_event_source_unref(f->post_change_timer);
360 }
361
362 journal_file_set_offline(f, true);
363
364 if (f->mmap && f->cache_fd)
365 mmap_cache_free_fd(f->mmap, f->cache_fd);
366
367 if (f->fd >= 0 && f->defrag_on_close) {
368
369 /* Be friendly to btrfs: turn COW back on again now,
370 * and defragment the file. We won't write to the file
371 * ever again, hence remove all fragmentation, and
372 * reenable all the good bits COW usually provides
373 * (such as data checksumming). */
374
375 (void) chattr_fd(f->fd, 0, FS_NOCOW_FL, NULL);
376 (void) btrfs_defrag_fd(f->fd);
377 }
378
379 if (f->close_fd)
380 safe_close(f->fd);
381 free(f->path);
382
383 mmap_cache_unref(f->mmap);
384
385 ordered_hashmap_free_free(f->chain_cache);
386
387 #if HAVE_XZ || HAVE_LZ4
388 free(f->compress_buffer);
389 #endif
390
391 #if HAVE_GCRYPT
392 if (f->fss_file)
393 munmap(f->fss_file, PAGE_ALIGN(f->fss_file_size));
394 else
395 free(f->fsprg_state);
396
397 free(f->fsprg_seed);
398
399 if (f->hmac)
400 gcry_md_close(f->hmac);
401 #endif
402
403 return mfree(f);
404 }
405
406 static int journal_file_init_header(JournalFile *f, JournalFile *template) {
407 Header h = {};
408 ssize_t k;
409 int r;
410
411 assert(f);
412
413 memcpy(h.signature, HEADER_SIGNATURE, 8);
414 h.header_size = htole64(ALIGN64(sizeof(h)));
415
416 h.incompatible_flags |= htole32(
417 f->compress_xz * HEADER_INCOMPATIBLE_COMPRESSED_XZ |
418 f->compress_lz4 * HEADER_INCOMPATIBLE_COMPRESSED_LZ4);
419
420 h.compatible_flags = htole32(
421 f->seal * HEADER_COMPATIBLE_SEALED);
422
423 r = sd_id128_randomize(&h.file_id);
424 if (r < 0)
425 return r;
426
427 if (template) {
428 h.seqnum_id = template->header->seqnum_id;
429 h.tail_entry_seqnum = template->header->tail_entry_seqnum;
430 } else
431 h.seqnum_id = h.file_id;
432
433 k = pwrite(f->fd, &h, sizeof(h), 0);
434 if (k < 0)
435 return -errno;
436
437 if (k != sizeof(h))
438 return -EIO;
439
440 return 0;
441 }
442
443 static int journal_file_refresh_header(JournalFile *f) {
444 sd_id128_t boot_id;
445 int r;
446
447 assert(f);
448 assert(f->header);
449
450 r = sd_id128_get_machine(&f->header->machine_id);
451 if (IN_SET(r, -ENOENT, -ENOMEDIUM))
452 /* We don't have a machine-id, let's continue without */
453 zero(f->header->machine_id);
454 else if (r < 0)
455 return r;
456
457 r = sd_id128_get_boot(&boot_id);
458 if (r < 0)
459 return r;
460
461 f->header->boot_id = boot_id;
462
463 r = journal_file_set_online(f);
464
465 /* Sync the online state to disk */
466 (void) fsync(f->fd);
467
468 /* We likely just created a new file, also sync the directory this file is located in. */
469 (void) fsync_directory_of_file(f->fd);
470
471 return r;
472 }
473
474 static bool warn_wrong_flags(const JournalFile *f, bool compatible) {
475 const uint32_t any = compatible ? HEADER_COMPATIBLE_ANY : HEADER_INCOMPATIBLE_ANY,
476 supported = compatible ? HEADER_COMPATIBLE_SUPPORTED : HEADER_INCOMPATIBLE_SUPPORTED;
477 const char *type = compatible ? "compatible" : "incompatible";
478 uint32_t flags;
479
480 flags = le32toh(compatible ? f->header->compatible_flags : f->header->incompatible_flags);
481
482 if (flags & ~supported) {
483 if (flags & ~any)
484 log_debug("Journal file %s has unknown %s flags 0x%"PRIx32,
485 f->path, type, flags & ~any);
486 flags = (flags & any) & ~supported;
487 if (flags) {
488 const char* strv[3];
489 unsigned n = 0;
490 _cleanup_free_ char *t = NULL;
491
492 if (compatible && (flags & HEADER_COMPATIBLE_SEALED))
493 strv[n++] = "sealed";
494 if (!compatible && (flags & HEADER_INCOMPATIBLE_COMPRESSED_XZ))
495 strv[n++] = "xz-compressed";
496 if (!compatible && (flags & HEADER_INCOMPATIBLE_COMPRESSED_LZ4))
497 strv[n++] = "lz4-compressed";
498 strv[n] = NULL;
499 assert(n < ELEMENTSOF(strv));
500
501 t = strv_join((char**) strv, ", ");
502 log_debug("Journal file %s uses %s %s %s disabled at compilation time.",
503 f->path, type, n > 1 ? "flags" : "flag", strnull(t));
504 }
505 return true;
506 }
507
508 return false;
509 }
510
511 static int journal_file_verify_header(JournalFile *f) {
512 uint64_t arena_size, header_size;
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 incompatible
521 * flags we don't know. */
522 if (warn_wrong_flags(f, false))
523 return -EPROTONOSUPPORT;
524
525 /* When open for writing we refuse to open files with compatible flags, too. */
526 if (f->writable && warn_wrong_flags(f, true))
527 return -EPROTONOSUPPORT;
528
529 if (f->header->state >= _STATE_MAX)
530 return -EBADMSG;
531
532 header_size = le64toh(f->header->header_size);
533
534 /* The first addition was n_data, so check that we are at least this large */
535 if (header_size < HEADER_SIZE_MIN)
536 return -EBADMSG;
537
538 if (JOURNAL_HEADER_SEALED(f->header) && !JOURNAL_HEADER_CONTAINS(f->header, n_entry_arrays))
539 return -EBADMSG;
540
541 arena_size = le64toh(f->header->arena_size);
542
543 if (UINT64_MAX - header_size < arena_size || header_size + arena_size > (uint64_t) f->last_stat.st_size)
544 return -ENODATA;
545
546 if (le64toh(f->header->tail_object_offset) > header_size + arena_size)
547 return -ENODATA;
548
549 if (!VALID64(le64toh(f->header->data_hash_table_offset)) ||
550 !VALID64(le64toh(f->header->field_hash_table_offset)) ||
551 !VALID64(le64toh(f->header->tail_object_offset)) ||
552 !VALID64(le64toh(f->header->entry_array_offset)))
553 return -ENODATA;
554
555 if (f->writable) {
556 sd_id128_t machine_id;
557 uint8_t state;
558 int r;
559
560 r = sd_id128_get_machine(&machine_id);
561 if (r < 0)
562 return r;
563
564 if (!sd_id128_equal(machine_id, f->header->machine_id))
565 return -EHOSTDOWN;
566
567 state = f->header->state;
568
569 if (state == STATE_ARCHIVED)
570 return -ESHUTDOWN; /* Already archived */
571 else if (state == STATE_ONLINE) {
572 log_debug("Journal file %s is already online. Assuming unclean closing.", f->path);
573 return -EBUSY;
574 } else if (state != STATE_OFFLINE) {
575 log_debug("Journal file %s has unknown state %i.", f->path, state);
576 return -EBUSY;
577 }
578
579 if (f->header->field_hash_table_size == 0 || f->header->data_hash_table_size == 0)
580 return -EBADMSG;
581
582 /* Don't permit appending to files from the future. Because otherwise the realtime timestamps wouldn't
583 * be strictly ordered in the entries in the file anymore, and we can't have that since it breaks
584 * bisection. */
585 if (le64toh(f->header->tail_entry_realtime) > now(CLOCK_REALTIME)) {
586 log_debug("Journal file %s is from the future, refusing to append new data to it that'd be older.", f->path);
587 return -ETXTBSY;
588 }
589 }
590
591 f->compress_xz = JOURNAL_HEADER_COMPRESSED_XZ(f->header);
592 f->compress_lz4 = JOURNAL_HEADER_COMPRESSED_LZ4(f->header);
593
594 f->seal = JOURNAL_HEADER_SEALED(f->header);
595
596 return 0;
597 }
598
599 static int journal_file_fstat(JournalFile *f) {
600 int r;
601
602 assert(f);
603 assert(f->fd >= 0);
604
605 if (fstat(f->fd, &f->last_stat) < 0)
606 return -errno;
607
608 f->last_stat_usec = now(CLOCK_MONOTONIC);
609
610 /* Refuse dealing with with files that aren't regular */
611 r = stat_verify_regular(&f->last_stat);
612 if (r < 0)
613 return r;
614
615 /* Refuse appending to files that are already deleted */
616 if (f->last_stat.st_nlink <= 0)
617 return -EIDRM;
618
619 return 0;
620 }
621
622 static int journal_file_allocate(JournalFile *f, uint64_t offset, uint64_t size) {
623 uint64_t old_size, new_size;
624 int r;
625
626 assert(f);
627 assert(f->header);
628
629 /* We assume that this file is not sparse, and we know that
630 * for sure, since we always call posix_fallocate()
631 * ourselves */
632
633 if (mmap_cache_got_sigbus(f->mmap, f->cache_fd))
634 return -EIO;
635
636 old_size =
637 le64toh(f->header->header_size) +
638 le64toh(f->header->arena_size);
639
640 new_size = PAGE_ALIGN(offset + size);
641 if (new_size < le64toh(f->header->header_size))
642 new_size = le64toh(f->header->header_size);
643
644 if (new_size <= old_size) {
645
646 /* We already pre-allocated enough space, but before
647 * we write to it, let's check with fstat() if the
648 * file got deleted, in order make sure we don't throw
649 * away the data immediately. Don't check fstat() for
650 * all writes though, but only once ever 10s. */
651
652 if (f->last_stat_usec + LAST_STAT_REFRESH_USEC > now(CLOCK_MONOTONIC))
653 return 0;
654
655 return journal_file_fstat(f);
656 }
657
658 /* Allocate more space. */
659
660 if (f->metrics.max_size > 0 && new_size > f->metrics.max_size)
661 return -E2BIG;
662
663 if (new_size > f->metrics.min_size && f->metrics.keep_free > 0) {
664 struct statvfs svfs;
665
666 if (fstatvfs(f->fd, &svfs) >= 0) {
667 uint64_t available;
668
669 available = LESS_BY((uint64_t) svfs.f_bfree * (uint64_t) svfs.f_bsize, f->metrics.keep_free);
670
671 if (new_size - old_size > available)
672 return -E2BIG;
673 }
674 }
675
676 /* Increase by larger blocks at once */
677 new_size = DIV_ROUND_UP(new_size, FILE_SIZE_INCREASE) * FILE_SIZE_INCREASE;
678 if (f->metrics.max_size > 0 && new_size > f->metrics.max_size)
679 new_size = f->metrics.max_size;
680
681 /* Note that the glibc fallocate() fallback is very
682 inefficient, hence we try to minimize the allocation area
683 as we can. */
684 r = posix_fallocate(f->fd, old_size, new_size - old_size);
685 if (r != 0)
686 return -r;
687
688 f->header->arena_size = htole64(new_size - le64toh(f->header->header_size));
689
690 return journal_file_fstat(f);
691 }
692
693 static unsigned type_to_context(ObjectType type) {
694 /* One context for each type, plus one catch-all for the rest */
695 assert_cc(_OBJECT_TYPE_MAX <= MMAP_CACHE_MAX_CONTEXTS);
696 assert_cc(CONTEXT_HEADER < MMAP_CACHE_MAX_CONTEXTS);
697 return type > OBJECT_UNUSED && type < _OBJECT_TYPE_MAX ? type : 0;
698 }
699
700 static int journal_file_move_to(JournalFile *f, ObjectType type, bool keep_always, uint64_t offset, uint64_t size, void **ret, size_t *ret_size) {
701 int r;
702
703 assert(f);
704 assert(ret);
705
706 if (size <= 0)
707 return -EINVAL;
708
709 /* Avoid SIGBUS on invalid accesses */
710 if (offset + size > (uint64_t) f->last_stat.st_size) {
711 /* Hmm, out of range? Let's refresh the fstat() data
712 * first, before we trust that check. */
713
714 r = journal_file_fstat(f);
715 if (r < 0)
716 return r;
717
718 if (offset + size > (uint64_t) f->last_stat.st_size)
719 return -EADDRNOTAVAIL;
720 }
721
722 return mmap_cache_get(f->mmap, f->cache_fd, f->prot, type_to_context(type), keep_always, offset, size, &f->last_stat, ret, ret_size);
723 }
724
725 static uint64_t minimum_header_size(Object *o) {
726
727 static const uint64_t table[] = {
728 [OBJECT_DATA] = sizeof(DataObject),
729 [OBJECT_FIELD] = sizeof(FieldObject),
730 [OBJECT_ENTRY] = sizeof(EntryObject),
731 [OBJECT_DATA_HASH_TABLE] = sizeof(HashTableObject),
732 [OBJECT_FIELD_HASH_TABLE] = sizeof(HashTableObject),
733 [OBJECT_ENTRY_ARRAY] = sizeof(EntryArrayObject),
734 [OBJECT_TAG] = sizeof(TagObject),
735 };
736
737 if (o->object.type >= ELEMENTSOF(table) || table[o->object.type] <= 0)
738 return sizeof(ObjectHeader);
739
740 return table[o->object.type];
741 }
742
743 /* Lightweight object checks. We want this to be fast, so that we won't
744 * slowdown every journal_file_move_to_object() call too much. */
745 static int journal_file_check_object(JournalFile *f, uint64_t offset, Object *o) {
746 assert(f);
747 assert(o);
748
749 switch (o->object.type) {
750
751 case OBJECT_DATA: {
752 if ((le64toh(o->data.entry_offset) == 0) ^ (le64toh(o->data.n_entries) == 0)) {
753 log_debug("Bad n_entries: %"PRIu64": %"PRIu64,
754 le64toh(o->data.n_entries), offset);
755 return -EBADMSG;
756 }
757
758 if (le64toh(o->object.size) - offsetof(DataObject, payload) <= 0) {
759 log_debug("Bad object size (<= %zu): %"PRIu64": %"PRIu64,
760 offsetof(DataObject, payload),
761 le64toh(o->object.size),
762 offset);
763 return -EBADMSG;
764 }
765
766 if (!VALID64(le64toh(o->data.next_hash_offset)) ||
767 !VALID64(le64toh(o->data.next_field_offset)) ||
768 !VALID64(le64toh(o->data.entry_offset)) ||
769 !VALID64(le64toh(o->data.entry_array_offset))) {
770 log_debug("Invalid offset, next_hash_offset="OFSfmt", next_field_offset="OFSfmt
771 ", entry_offset="OFSfmt", entry_array_offset="OFSfmt": %"PRIu64,
772 le64toh(o->data.next_hash_offset),
773 le64toh(o->data.next_field_offset),
774 le64toh(o->data.entry_offset),
775 le64toh(o->data.entry_array_offset),
776 offset);
777 return -EBADMSG;
778 }
779
780 break;
781 }
782
783 case OBJECT_FIELD:
784 if (le64toh(o->object.size) - offsetof(FieldObject, payload) <= 0) {
785 log_debug(
786 "Bad field size (<= %zu): %"PRIu64": %"PRIu64,
787 offsetof(FieldObject, payload),
788 le64toh(o->object.size),
789 offset);
790 return -EBADMSG;
791 }
792
793 if (!VALID64(le64toh(o->field.next_hash_offset)) ||
794 !VALID64(le64toh(o->field.head_data_offset))) {
795 log_debug(
796 "Invalid offset, next_hash_offset="OFSfmt
797 ", head_data_offset="OFSfmt": %"PRIu64,
798 le64toh(o->field.next_hash_offset),
799 le64toh(o->field.head_data_offset),
800 offset);
801 return -EBADMSG;
802 }
803 break;
804
805 case OBJECT_ENTRY:
806 if ((le64toh(o->object.size) - offsetof(EntryObject, items)) % sizeof(EntryItem) != 0) {
807 log_debug(
808 "Bad entry size (<= %zu): %"PRIu64": %"PRIu64,
809 offsetof(EntryObject, items),
810 le64toh(o->object.size),
811 offset);
812 return -EBADMSG;
813 }
814
815 if ((le64toh(o->object.size) - offsetof(EntryObject, items)) / sizeof(EntryItem) <= 0) {
816 log_debug(
817 "Invalid number items in entry: %"PRIu64": %"PRIu64,
818 (le64toh(o->object.size) - offsetof(EntryObject, items)) / sizeof(EntryItem),
819 offset);
820 return -EBADMSG;
821 }
822
823 if (le64toh(o->entry.seqnum) <= 0) {
824 log_debug(
825 "Invalid entry seqnum: %"PRIx64": %"PRIu64,
826 le64toh(o->entry.seqnum),
827 offset);
828 return -EBADMSG;
829 }
830
831 if (!VALID_REALTIME(le64toh(o->entry.realtime))) {
832 log_debug(
833 "Invalid entry realtime timestamp: %"PRIu64": %"PRIu64,
834 le64toh(o->entry.realtime),
835 offset);
836 return -EBADMSG;
837 }
838
839 if (!VALID_MONOTONIC(le64toh(o->entry.monotonic))) {
840 log_debug(
841 "Invalid entry monotonic timestamp: %"PRIu64": %"PRIu64,
842 le64toh(o->entry.monotonic),
843 offset);
844 return -EBADMSG;
845 }
846
847 break;
848
849 case OBJECT_DATA_HASH_TABLE:
850 case OBJECT_FIELD_HASH_TABLE:
851 if ((le64toh(o->object.size) - offsetof(HashTableObject, items)) % sizeof(HashItem) != 0 ||
852 (le64toh(o->object.size) - offsetof(HashTableObject, items)) / sizeof(HashItem) <= 0) {
853 log_debug(
854 "Invalid %s hash table size: %"PRIu64": %"PRIu64,
855 o->object.type == OBJECT_DATA_HASH_TABLE ? "data" : "field",
856 le64toh(o->object.size),
857 offset);
858 return -EBADMSG;
859 }
860
861 break;
862
863 case OBJECT_ENTRY_ARRAY:
864 if ((le64toh(o->object.size) - offsetof(EntryArrayObject, items)) % sizeof(le64_t) != 0 ||
865 (le64toh(o->object.size) - offsetof(EntryArrayObject, items)) / sizeof(le64_t) <= 0) {
866 log_debug(
867 "Invalid object entry array size: %"PRIu64": %"PRIu64,
868 le64toh(o->object.size),
869 offset);
870 return -EBADMSG;
871 }
872
873 if (!VALID64(le64toh(o->entry_array.next_entry_array_offset))) {
874 log_debug(
875 "Invalid object entry array next_entry_array_offset: "OFSfmt": %"PRIu64,
876 le64toh(o->entry_array.next_entry_array_offset),
877 offset);
878 return -EBADMSG;
879 }
880
881 break;
882
883 case OBJECT_TAG:
884 if (le64toh(o->object.size) != sizeof(TagObject)) {
885 log_debug(
886 "Invalid object tag size: %"PRIu64": %"PRIu64,
887 le64toh(o->object.size),
888 offset);
889 return -EBADMSG;
890 }
891
892 if (!VALID_EPOCH(le64toh(o->tag.epoch))) {
893 log_debug(
894 "Invalid object tag epoch: %"PRIu64": %"PRIu64,
895 le64toh(o->tag.epoch),
896 offset);
897 return -EBADMSG;
898 }
899
900 break;
901 }
902
903 return 0;
904 }
905
906 int journal_file_move_to_object(JournalFile *f, ObjectType type, uint64_t offset, Object **ret) {
907 int r;
908 void *t;
909 size_t tsize;
910 Object *o;
911 uint64_t s;
912
913 assert(f);
914 assert(ret);
915
916 /* Objects may only be located at multiple of 64 bit */
917 if (!VALID64(offset)) {
918 log_debug("Attempt to move to object at non-64bit boundary: %" PRIu64, offset);
919 return -EBADMSG;
920 }
921
922 /* Object may not be located in the file header */
923 if (offset < le64toh(f->header->header_size)) {
924 log_debug("Attempt to move to object located in file header: %" PRIu64, offset);
925 return -EBADMSG;
926 }
927
928 r = journal_file_move_to(f, type, false, offset, sizeof(ObjectHeader), &t, &tsize);
929 if (r < 0)
930 return r;
931
932 o = (Object*) t;
933 s = le64toh(o->object.size);
934
935 if (s == 0) {
936 log_debug("Attempt to move to uninitialized object: %" PRIu64, offset);
937 return -EBADMSG;
938 }
939 if (s < sizeof(ObjectHeader)) {
940 log_debug("Attempt to move to overly short object: %" PRIu64, offset);
941 return -EBADMSG;
942 }
943
944 if (o->object.type <= OBJECT_UNUSED) {
945 log_debug("Attempt to move to object with invalid type: %" PRIu64, offset);
946 return -EBADMSG;
947 }
948
949 if (s < minimum_header_size(o)) {
950 log_debug("Attempt to move to truncated object: %" PRIu64, offset);
951 return -EBADMSG;
952 }
953
954 if (type > OBJECT_UNUSED && o->object.type != type) {
955 log_debug("Attempt to move to object of unexpected type: %" PRIu64, offset);
956 return -EBADMSG;
957 }
958
959 if (s > tsize) {
960 r = journal_file_move_to(f, type, false, offset, s, &t, NULL);
961 if (r < 0)
962 return r;
963
964 o = (Object*) t;
965 }
966
967 r = journal_file_check_object(f, offset, o);
968 if (r < 0)
969 return r;
970
971 *ret = o;
972 return 0;
973 }
974
975 static uint64_t journal_file_entry_seqnum(JournalFile *f, uint64_t *seqnum) {
976 uint64_t r;
977
978 assert(f);
979 assert(f->header);
980
981 r = le64toh(f->header->tail_entry_seqnum) + 1;
982
983 if (seqnum) {
984 /* If an external seqnum counter was passed, we update
985 * both the local and the external one, and set it to
986 * the maximum of both */
987
988 if (*seqnum + 1 > r)
989 r = *seqnum + 1;
990
991 *seqnum = r;
992 }
993
994 f->header->tail_entry_seqnum = htole64(r);
995
996 if (f->header->head_entry_seqnum == 0)
997 f->header->head_entry_seqnum = htole64(r);
998
999 return r;
1000 }
1001
1002 int journal_file_append_object(JournalFile *f, ObjectType type, uint64_t size, Object **ret, uint64_t *offset) {
1003 int r;
1004 uint64_t p;
1005 Object *tail, *o;
1006 void *t;
1007
1008 assert(f);
1009 assert(f->header);
1010 assert(type > OBJECT_UNUSED && type < _OBJECT_TYPE_MAX);
1011 assert(size >= sizeof(ObjectHeader));
1012 assert(offset);
1013 assert(ret);
1014
1015 r = journal_file_set_online(f);
1016 if (r < 0)
1017 return r;
1018
1019 p = le64toh(f->header->tail_object_offset);
1020 if (p == 0)
1021 p = le64toh(f->header->header_size);
1022 else {
1023 r = journal_file_move_to_object(f, OBJECT_UNUSED, p, &tail);
1024 if (r < 0)
1025 return r;
1026
1027 p += ALIGN64(le64toh(tail->object.size));
1028 }
1029
1030 r = journal_file_allocate(f, p, size);
1031 if (r < 0)
1032 return r;
1033
1034 r = journal_file_move_to(f, type, false, p, size, &t, NULL);
1035 if (r < 0)
1036 return r;
1037
1038 o = (Object*) t;
1039
1040 zero(o->object);
1041 o->object.type = type;
1042 o->object.size = htole64(size);
1043
1044 f->header->tail_object_offset = htole64(p);
1045 f->header->n_objects = htole64(le64toh(f->header->n_objects) + 1);
1046
1047 *ret = o;
1048 *offset = p;
1049
1050 return 0;
1051 }
1052
1053 static int journal_file_setup_data_hash_table(JournalFile *f) {
1054 uint64_t s, p;
1055 Object *o;
1056 int r;
1057
1058 assert(f);
1059 assert(f->header);
1060
1061 /* We estimate that we need 1 hash table entry per 768 bytes
1062 of journal file and we want to make sure we never get
1063 beyond 75% fill level. Calculate the hash table size for
1064 the maximum file size based on these metrics. */
1065
1066 s = (f->metrics.max_size * 4 / 768 / 3) * sizeof(HashItem);
1067 if (s < DEFAULT_DATA_HASH_TABLE_SIZE)
1068 s = DEFAULT_DATA_HASH_TABLE_SIZE;
1069
1070 log_debug("Reserving %"PRIu64" entries in hash table.", s / sizeof(HashItem));
1071
1072 r = journal_file_append_object(f,
1073 OBJECT_DATA_HASH_TABLE,
1074 offsetof(Object, hash_table.items) + s,
1075 &o, &p);
1076 if (r < 0)
1077 return r;
1078
1079 memzero(o->hash_table.items, s);
1080
1081 f->header->data_hash_table_offset = htole64(p + offsetof(Object, hash_table.items));
1082 f->header->data_hash_table_size = htole64(s);
1083
1084 return 0;
1085 }
1086
1087 static int journal_file_setup_field_hash_table(JournalFile *f) {
1088 uint64_t s, p;
1089 Object *o;
1090 int r;
1091
1092 assert(f);
1093 assert(f->header);
1094
1095 /* We use a fixed size hash table for the fields as this
1096 * number should grow very slowly only */
1097
1098 s = DEFAULT_FIELD_HASH_TABLE_SIZE;
1099 r = journal_file_append_object(f,
1100 OBJECT_FIELD_HASH_TABLE,
1101 offsetof(Object, hash_table.items) + s,
1102 &o, &p);
1103 if (r < 0)
1104 return r;
1105
1106 memzero(o->hash_table.items, s);
1107
1108 f->header->field_hash_table_offset = htole64(p + offsetof(Object, hash_table.items));
1109 f->header->field_hash_table_size = htole64(s);
1110
1111 return 0;
1112 }
1113
1114 int journal_file_map_data_hash_table(JournalFile *f) {
1115 uint64_t s, p;
1116 void *t;
1117 int r;
1118
1119 assert(f);
1120 assert(f->header);
1121
1122 if (f->data_hash_table)
1123 return 0;
1124
1125 p = le64toh(f->header->data_hash_table_offset);
1126 s = le64toh(f->header->data_hash_table_size);
1127
1128 r = journal_file_move_to(f,
1129 OBJECT_DATA_HASH_TABLE,
1130 true,
1131 p, s,
1132 &t, NULL);
1133 if (r < 0)
1134 return r;
1135
1136 f->data_hash_table = t;
1137 return 0;
1138 }
1139
1140 int journal_file_map_field_hash_table(JournalFile *f) {
1141 uint64_t s, p;
1142 void *t;
1143 int r;
1144
1145 assert(f);
1146 assert(f->header);
1147
1148 if (f->field_hash_table)
1149 return 0;
1150
1151 p = le64toh(f->header->field_hash_table_offset);
1152 s = le64toh(f->header->field_hash_table_size);
1153
1154 r = journal_file_move_to(f,
1155 OBJECT_FIELD_HASH_TABLE,
1156 true,
1157 p, s,
1158 &t, NULL);
1159 if (r < 0)
1160 return r;
1161
1162 f->field_hash_table = t;
1163 return 0;
1164 }
1165
1166 static int journal_file_link_field(
1167 JournalFile *f,
1168 Object *o,
1169 uint64_t offset,
1170 uint64_t hash) {
1171
1172 uint64_t p, h, m;
1173 int r;
1174
1175 assert(f);
1176 assert(f->header);
1177 assert(f->field_hash_table);
1178 assert(o);
1179 assert(offset > 0);
1180
1181 if (o->object.type != OBJECT_FIELD)
1182 return -EINVAL;
1183
1184 m = le64toh(f->header->field_hash_table_size) / sizeof(HashItem);
1185 if (m <= 0)
1186 return -EBADMSG;
1187
1188 /* This might alter the window we are looking at */
1189 o->field.next_hash_offset = o->field.head_data_offset = 0;
1190
1191 h = hash % m;
1192 p = le64toh(f->field_hash_table[h].tail_hash_offset);
1193 if (p == 0)
1194 f->field_hash_table[h].head_hash_offset = htole64(offset);
1195 else {
1196 r = journal_file_move_to_object(f, OBJECT_FIELD, p, &o);
1197 if (r < 0)
1198 return r;
1199
1200 o->field.next_hash_offset = htole64(offset);
1201 }
1202
1203 f->field_hash_table[h].tail_hash_offset = htole64(offset);
1204
1205 if (JOURNAL_HEADER_CONTAINS(f->header, n_fields))
1206 f->header->n_fields = htole64(le64toh(f->header->n_fields) + 1);
1207
1208 return 0;
1209 }
1210
1211 static int journal_file_link_data(
1212 JournalFile *f,
1213 Object *o,
1214 uint64_t offset,
1215 uint64_t hash) {
1216
1217 uint64_t p, h, m;
1218 int r;
1219
1220 assert(f);
1221 assert(f->header);
1222 assert(f->data_hash_table);
1223 assert(o);
1224 assert(offset > 0);
1225
1226 if (o->object.type != OBJECT_DATA)
1227 return -EINVAL;
1228
1229 m = le64toh(f->header->data_hash_table_size) / sizeof(HashItem);
1230 if (m <= 0)
1231 return -EBADMSG;
1232
1233 /* This might alter the window we are looking at */
1234 o->data.next_hash_offset = o->data.next_field_offset = 0;
1235 o->data.entry_offset = o->data.entry_array_offset = 0;
1236 o->data.n_entries = 0;
1237
1238 h = hash % m;
1239 p = le64toh(f->data_hash_table[h].tail_hash_offset);
1240 if (p == 0)
1241 /* Only entry in the hash table is easy */
1242 f->data_hash_table[h].head_hash_offset = htole64(offset);
1243 else {
1244 /* Move back to the previous data object, to patch in
1245 * pointer */
1246
1247 r = journal_file_move_to_object(f, OBJECT_DATA, p, &o);
1248 if (r < 0)
1249 return r;
1250
1251 o->data.next_hash_offset = htole64(offset);
1252 }
1253
1254 f->data_hash_table[h].tail_hash_offset = htole64(offset);
1255
1256 if (JOURNAL_HEADER_CONTAINS(f->header, n_data))
1257 f->header->n_data = htole64(le64toh(f->header->n_data) + 1);
1258
1259 return 0;
1260 }
1261
1262 int journal_file_find_field_object_with_hash(
1263 JournalFile *f,
1264 const void *field, uint64_t size, uint64_t hash,
1265 Object **ret, uint64_t *offset) {
1266
1267 uint64_t p, osize, h, m;
1268 int r;
1269
1270 assert(f);
1271 assert(f->header);
1272 assert(field && size > 0);
1273
1274 /* If the field hash table is empty, we can't find anything */
1275 if (le64toh(f->header->field_hash_table_size) <= 0)
1276 return 0;
1277
1278 /* Map the field hash table, if it isn't mapped yet. */
1279 r = journal_file_map_field_hash_table(f);
1280 if (r < 0)
1281 return r;
1282
1283 osize = offsetof(Object, field.payload) + size;
1284
1285 m = le64toh(f->header->field_hash_table_size) / sizeof(HashItem);
1286 if (m <= 0)
1287 return -EBADMSG;
1288
1289 h = hash % m;
1290 p = le64toh(f->field_hash_table[h].head_hash_offset);
1291
1292 while (p > 0) {
1293 Object *o;
1294
1295 r = journal_file_move_to_object(f, OBJECT_FIELD, p, &o);
1296 if (r < 0)
1297 return r;
1298
1299 if (le64toh(o->field.hash) == hash &&
1300 le64toh(o->object.size) == osize &&
1301 memcmp(o->field.payload, field, size) == 0) {
1302
1303 if (ret)
1304 *ret = o;
1305 if (offset)
1306 *offset = p;
1307
1308 return 1;
1309 }
1310
1311 p = le64toh(o->field.next_hash_offset);
1312 }
1313
1314 return 0;
1315 }
1316
1317 int journal_file_find_field_object(
1318 JournalFile *f,
1319 const void *field, uint64_t size,
1320 Object **ret, uint64_t *offset) {
1321
1322 uint64_t hash;
1323
1324 assert(f);
1325 assert(field && size > 0);
1326
1327 hash = hash64(field, size);
1328
1329 return journal_file_find_field_object_with_hash(f,
1330 field, size, hash,
1331 ret, offset);
1332 }
1333
1334 int journal_file_find_data_object_with_hash(
1335 JournalFile *f,
1336 const void *data, uint64_t size, uint64_t hash,
1337 Object **ret, uint64_t *offset) {
1338
1339 uint64_t p, osize, h, m;
1340 int r;
1341
1342 assert(f);
1343 assert(f->header);
1344 assert(data || size == 0);
1345
1346 /* If there's no data hash table, then there's no entry. */
1347 if (le64toh(f->header->data_hash_table_size) <= 0)
1348 return 0;
1349
1350 /* Map the data hash table, if it isn't mapped yet. */
1351 r = journal_file_map_data_hash_table(f);
1352 if (r < 0)
1353 return r;
1354
1355 osize = offsetof(Object, data.payload) + size;
1356
1357 m = le64toh(f->header->data_hash_table_size) / sizeof(HashItem);
1358 if (m <= 0)
1359 return -EBADMSG;
1360
1361 h = hash % m;
1362 p = le64toh(f->data_hash_table[h].head_hash_offset);
1363
1364 while (p > 0) {
1365 Object *o;
1366
1367 r = journal_file_move_to_object(f, OBJECT_DATA, p, &o);
1368 if (r < 0)
1369 return r;
1370
1371 if (le64toh(o->data.hash) != hash)
1372 goto next;
1373
1374 if (o->object.flags & OBJECT_COMPRESSION_MASK) {
1375 #if HAVE_XZ || HAVE_LZ4
1376 uint64_t l;
1377 size_t rsize = 0;
1378
1379 l = le64toh(o->object.size);
1380 if (l <= offsetof(Object, data.payload))
1381 return -EBADMSG;
1382
1383 l -= offsetof(Object, data.payload);
1384
1385 r = decompress_blob(o->object.flags & OBJECT_COMPRESSION_MASK,
1386 o->data.payload, l, &f->compress_buffer, &f->compress_buffer_size, &rsize, 0);
1387 if (r < 0)
1388 return r;
1389
1390 if (rsize == size &&
1391 memcmp(f->compress_buffer, data, size) == 0) {
1392
1393 if (ret)
1394 *ret = o;
1395
1396 if (offset)
1397 *offset = p;
1398
1399 return 1;
1400 }
1401 #else
1402 return -EPROTONOSUPPORT;
1403 #endif
1404 } else if (le64toh(o->object.size) == osize &&
1405 memcmp(o->data.payload, data, size) == 0) {
1406
1407 if (ret)
1408 *ret = o;
1409
1410 if (offset)
1411 *offset = p;
1412
1413 return 1;
1414 }
1415
1416 next:
1417 p = le64toh(o->data.next_hash_offset);
1418 }
1419
1420 return 0;
1421 }
1422
1423 int journal_file_find_data_object(
1424 JournalFile *f,
1425 const void *data, uint64_t size,
1426 Object **ret, uint64_t *offset) {
1427
1428 uint64_t hash;
1429
1430 assert(f);
1431 assert(data || size == 0);
1432
1433 hash = hash64(data, size);
1434
1435 return journal_file_find_data_object_with_hash(f,
1436 data, size, hash,
1437 ret, offset);
1438 }
1439
1440 static int journal_file_append_field(
1441 JournalFile *f,
1442 const void *field, uint64_t size,
1443 Object **ret, uint64_t *offset) {
1444
1445 uint64_t hash, p;
1446 uint64_t osize;
1447 Object *o;
1448 int r;
1449
1450 assert(f);
1451 assert(field && size > 0);
1452
1453 hash = hash64(field, size);
1454
1455 r = journal_file_find_field_object_with_hash(f, field, size, hash, &o, &p);
1456 if (r < 0)
1457 return r;
1458 else if (r > 0) {
1459
1460 if (ret)
1461 *ret = o;
1462
1463 if (offset)
1464 *offset = p;
1465
1466 return 0;
1467 }
1468
1469 osize = offsetof(Object, field.payload) + size;
1470 r = journal_file_append_object(f, OBJECT_FIELD, osize, &o, &p);
1471 if (r < 0)
1472 return r;
1473
1474 o->field.hash = htole64(hash);
1475 memcpy(o->field.payload, field, size);
1476
1477 r = journal_file_link_field(f, o, p, hash);
1478 if (r < 0)
1479 return r;
1480
1481 /* The linking might have altered the window, so let's
1482 * refresh our pointer */
1483 r = journal_file_move_to_object(f, OBJECT_FIELD, p, &o);
1484 if (r < 0)
1485 return r;
1486
1487 #if HAVE_GCRYPT
1488 r = journal_file_hmac_put_object(f, OBJECT_FIELD, o, p);
1489 if (r < 0)
1490 return r;
1491 #endif
1492
1493 if (ret)
1494 *ret = o;
1495
1496 if (offset)
1497 *offset = p;
1498
1499 return 0;
1500 }
1501
1502 static int journal_file_append_data(
1503 JournalFile *f,
1504 const void *data, uint64_t size,
1505 Object **ret, uint64_t *offset) {
1506
1507 uint64_t hash, p;
1508 uint64_t osize;
1509 Object *o;
1510 int r, compression = 0;
1511 const void *eq;
1512
1513 assert(f);
1514 assert(data || size == 0);
1515
1516 hash = hash64(data, size);
1517
1518 r = journal_file_find_data_object_with_hash(f, data, size, hash, &o, &p);
1519 if (r < 0)
1520 return r;
1521 if (r > 0) {
1522
1523 if (ret)
1524 *ret = o;
1525
1526 if (offset)
1527 *offset = p;
1528
1529 return 0;
1530 }
1531
1532 osize = offsetof(Object, data.payload) + size;
1533 r = journal_file_append_object(f, OBJECT_DATA, osize, &o, &p);
1534 if (r < 0)
1535 return r;
1536
1537 o->data.hash = htole64(hash);
1538
1539 #if HAVE_XZ || HAVE_LZ4
1540 if (JOURNAL_FILE_COMPRESS(f) && size >= f->compress_threshold_bytes) {
1541 size_t rsize = 0;
1542
1543 compression = compress_blob(data, size, o->data.payload, size - 1, &rsize);
1544
1545 if (compression >= 0) {
1546 o->object.size = htole64(offsetof(Object, data.payload) + rsize);
1547 o->object.flags |= compression;
1548
1549 log_debug("Compressed data object %"PRIu64" -> %zu using %s",
1550 size, rsize, object_compressed_to_string(compression));
1551 } else
1552 /* Compression didn't work, we don't really care why, let's continue without compression */
1553 compression = 0;
1554 }
1555 #endif
1556
1557 if (compression == 0)
1558 memcpy_safe(o->data.payload, data, size);
1559
1560 r = journal_file_link_data(f, o, p, hash);
1561 if (r < 0)
1562 return r;
1563
1564 #if HAVE_GCRYPT
1565 r = journal_file_hmac_put_object(f, OBJECT_DATA, o, p);
1566 if (r < 0)
1567 return r;
1568 #endif
1569
1570 /* The linking might have altered the window, so let's
1571 * refresh our pointer */
1572 r = journal_file_move_to_object(f, OBJECT_DATA, p, &o);
1573 if (r < 0)
1574 return r;
1575
1576 if (!data)
1577 eq = NULL;
1578 else
1579 eq = memchr(data, '=', size);
1580 if (eq && eq > data) {
1581 Object *fo = NULL;
1582 uint64_t fp;
1583
1584 /* Create field object ... */
1585 r = journal_file_append_field(f, data, (uint8_t*) eq - (uint8_t*) data, &fo, &fp);
1586 if (r < 0)
1587 return r;
1588
1589 /* ... and link it in. */
1590 o->data.next_field_offset = fo->field.head_data_offset;
1591 fo->field.head_data_offset = le64toh(p);
1592 }
1593
1594 if (ret)
1595 *ret = o;
1596
1597 if (offset)
1598 *offset = p;
1599
1600 return 0;
1601 }
1602
1603 uint64_t journal_file_entry_n_items(Object *o) {
1604 assert(o);
1605
1606 if (o->object.type != OBJECT_ENTRY)
1607 return 0;
1608
1609 return (le64toh(o->object.size) - offsetof(Object, entry.items)) / sizeof(EntryItem);
1610 }
1611
1612 uint64_t journal_file_entry_array_n_items(Object *o) {
1613 assert(o);
1614
1615 if (o->object.type != OBJECT_ENTRY_ARRAY)
1616 return 0;
1617
1618 return (le64toh(o->object.size) - offsetof(Object, entry_array.items)) / sizeof(uint64_t);
1619 }
1620
1621 uint64_t journal_file_hash_table_n_items(Object *o) {
1622 assert(o);
1623
1624 if (!IN_SET(o->object.type, OBJECT_DATA_HASH_TABLE, OBJECT_FIELD_HASH_TABLE))
1625 return 0;
1626
1627 return (le64toh(o->object.size) - offsetof(Object, hash_table.items)) / sizeof(HashItem);
1628 }
1629
1630 static int link_entry_into_array(JournalFile *f,
1631 le64_t *first,
1632 le64_t *idx,
1633 uint64_t p) {
1634 int r;
1635 uint64_t n = 0, ap = 0, q, i, a, hidx;
1636 Object *o;
1637
1638 assert(f);
1639 assert(f->header);
1640 assert(first);
1641 assert(idx);
1642 assert(p > 0);
1643
1644 a = le64toh(*first);
1645 i = hidx = le64toh(*idx);
1646 while (a > 0) {
1647
1648 r = journal_file_move_to_object(f, OBJECT_ENTRY_ARRAY, a, &o);
1649 if (r < 0)
1650 return r;
1651
1652 n = journal_file_entry_array_n_items(o);
1653 if (i < n) {
1654 o->entry_array.items[i] = htole64(p);
1655 *idx = htole64(hidx + 1);
1656 return 0;
1657 }
1658
1659 i -= n;
1660 ap = a;
1661 a = le64toh(o->entry_array.next_entry_array_offset);
1662 }
1663
1664 if (hidx > n)
1665 n = (hidx+1) * 2;
1666 else
1667 n = n * 2;
1668
1669 if (n < 4)
1670 n = 4;
1671
1672 r = journal_file_append_object(f, OBJECT_ENTRY_ARRAY,
1673 offsetof(Object, entry_array.items) + n * sizeof(uint64_t),
1674 &o, &q);
1675 if (r < 0)
1676 return r;
1677
1678 #if HAVE_GCRYPT
1679 r = journal_file_hmac_put_object(f, OBJECT_ENTRY_ARRAY, o, q);
1680 if (r < 0)
1681 return r;
1682 #endif
1683
1684 o->entry_array.items[i] = htole64(p);
1685
1686 if (ap == 0)
1687 *first = htole64(q);
1688 else {
1689 r = journal_file_move_to_object(f, OBJECT_ENTRY_ARRAY, ap, &o);
1690 if (r < 0)
1691 return r;
1692
1693 o->entry_array.next_entry_array_offset = htole64(q);
1694 }
1695
1696 if (JOURNAL_HEADER_CONTAINS(f->header, n_entry_arrays))
1697 f->header->n_entry_arrays = htole64(le64toh(f->header->n_entry_arrays) + 1);
1698
1699 *idx = htole64(hidx + 1);
1700
1701 return 0;
1702 }
1703
1704 static int link_entry_into_array_plus_one(JournalFile *f,
1705 le64_t *extra,
1706 le64_t *first,
1707 le64_t *idx,
1708 uint64_t p) {
1709
1710 int r;
1711
1712 assert(f);
1713 assert(extra);
1714 assert(first);
1715 assert(idx);
1716 assert(p > 0);
1717
1718 if (*idx == 0)
1719 *extra = htole64(p);
1720 else {
1721 le64_t i;
1722
1723 i = htole64(le64toh(*idx) - 1);
1724 r = link_entry_into_array(f, first, &i, p);
1725 if (r < 0)
1726 return r;
1727 }
1728
1729 *idx = htole64(le64toh(*idx) + 1);
1730 return 0;
1731 }
1732
1733 static int journal_file_link_entry_item(JournalFile *f, Object *o, uint64_t offset, uint64_t i) {
1734 uint64_t p;
1735 int r;
1736 assert(f);
1737 assert(o);
1738 assert(offset > 0);
1739
1740 p = le64toh(o->entry.items[i].object_offset);
1741 if (p == 0)
1742 return -EINVAL;
1743
1744 r = journal_file_move_to_object(f, OBJECT_DATA, p, &o);
1745 if (r < 0)
1746 return r;
1747
1748 return link_entry_into_array_plus_one(f,
1749 &o->data.entry_offset,
1750 &o->data.entry_array_offset,
1751 &o->data.n_entries,
1752 offset);
1753 }
1754
1755 static int journal_file_link_entry(JournalFile *f, Object *o, uint64_t offset) {
1756 uint64_t n, i;
1757 int r;
1758
1759 assert(f);
1760 assert(f->header);
1761 assert(o);
1762 assert(offset > 0);
1763
1764 if (o->object.type != OBJECT_ENTRY)
1765 return -EINVAL;
1766
1767 __sync_synchronize();
1768
1769 /* Link up the entry itself */
1770 r = link_entry_into_array(f,
1771 &f->header->entry_array_offset,
1772 &f->header->n_entries,
1773 offset);
1774 if (r < 0)
1775 return r;
1776
1777 /* log_debug("=> %s seqnr=%"PRIu64" n_entries=%"PRIu64, f->path, o->entry.seqnum, f->header->n_entries); */
1778
1779 if (f->header->head_entry_realtime == 0)
1780 f->header->head_entry_realtime = o->entry.realtime;
1781
1782 f->header->tail_entry_realtime = o->entry.realtime;
1783 f->header->tail_entry_monotonic = o->entry.monotonic;
1784
1785 /* Link up the items */
1786 n = journal_file_entry_n_items(o);
1787 for (i = 0; i < n; i++) {
1788 r = journal_file_link_entry_item(f, o, offset, i);
1789 if (r < 0)
1790 return r;
1791 }
1792
1793 return 0;
1794 }
1795
1796 static int journal_file_append_entry_internal(
1797 JournalFile *f,
1798 const dual_timestamp *ts,
1799 const sd_id128_t *boot_id,
1800 uint64_t xor_hash,
1801 const EntryItem items[], unsigned n_items,
1802 uint64_t *seqnum,
1803 Object **ret, uint64_t *offset) {
1804 uint64_t np;
1805 uint64_t osize;
1806 Object *o;
1807 int r;
1808
1809 assert(f);
1810 assert(f->header);
1811 assert(items || n_items == 0);
1812 assert(ts);
1813
1814 osize = offsetof(Object, entry.items) + (n_items * sizeof(EntryItem));
1815
1816 r = journal_file_append_object(f, OBJECT_ENTRY, osize, &o, &np);
1817 if (r < 0)
1818 return r;
1819
1820 o->entry.seqnum = htole64(journal_file_entry_seqnum(f, seqnum));
1821 memcpy_safe(o->entry.items, items, n_items * sizeof(EntryItem));
1822 o->entry.realtime = htole64(ts->realtime);
1823 o->entry.monotonic = htole64(ts->monotonic);
1824 o->entry.xor_hash = htole64(xor_hash);
1825 o->entry.boot_id = boot_id ? *boot_id : f->header->boot_id;
1826
1827 #if HAVE_GCRYPT
1828 r = journal_file_hmac_put_object(f, OBJECT_ENTRY, o, np);
1829 if (r < 0)
1830 return r;
1831 #endif
1832
1833 r = journal_file_link_entry(f, o, np);
1834 if (r < 0)
1835 return r;
1836
1837 if (ret)
1838 *ret = o;
1839
1840 if (offset)
1841 *offset = np;
1842
1843 return 0;
1844 }
1845
1846 void journal_file_post_change(JournalFile *f) {
1847 assert(f);
1848
1849 if (f->fd < 0)
1850 return;
1851
1852 /* inotify() does not receive IN_MODIFY events from file
1853 * accesses done via mmap(). After each access we hence
1854 * trigger IN_MODIFY by truncating the journal file to its
1855 * current size which triggers IN_MODIFY. */
1856
1857 __sync_synchronize();
1858
1859 if (ftruncate(f->fd, f->last_stat.st_size) < 0)
1860 log_debug_errno(errno, "Failed to truncate file to its own size: %m");
1861 }
1862
1863 static int post_change_thunk(sd_event_source *timer, uint64_t usec, void *userdata) {
1864 assert(userdata);
1865
1866 journal_file_post_change(userdata);
1867
1868 return 1;
1869 }
1870
1871 static void schedule_post_change(JournalFile *f) {
1872 sd_event_source *timer;
1873 int enabled, r;
1874 uint64_t now;
1875
1876 assert(f);
1877 assert(f->post_change_timer);
1878
1879 timer = f->post_change_timer;
1880
1881 r = sd_event_source_get_enabled(timer, &enabled);
1882 if (r < 0) {
1883 log_debug_errno(r, "Failed to get ftruncate timer state: %m");
1884 goto fail;
1885 }
1886
1887 if (enabled == SD_EVENT_ONESHOT)
1888 return;
1889
1890 r = sd_event_now(sd_event_source_get_event(timer), CLOCK_MONOTONIC, &now);
1891 if (r < 0) {
1892 log_debug_errno(r, "Failed to get clock's now for scheduling ftruncate: %m");
1893 goto fail;
1894 }
1895
1896 r = sd_event_source_set_time(timer, now+f->post_change_timer_period);
1897 if (r < 0) {
1898 log_debug_errno(r, "Failed to set time for scheduling ftruncate: %m");
1899 goto fail;
1900 }
1901
1902 r = sd_event_source_set_enabled(timer, SD_EVENT_ONESHOT);
1903 if (r < 0) {
1904 log_debug_errno(r, "Failed to enable scheduled ftruncate: %m");
1905 goto fail;
1906 }
1907
1908 return;
1909
1910 fail:
1911 /* On failure, let's simply post the change immediately. */
1912 journal_file_post_change(f);
1913 }
1914
1915 /* Enable coalesced change posting in a timer on the provided sd_event instance */
1916 int journal_file_enable_post_change_timer(JournalFile *f, sd_event *e, usec_t t) {
1917 _cleanup_(sd_event_source_unrefp) sd_event_source *timer = NULL;
1918 int r;
1919
1920 assert(f);
1921 assert_return(!f->post_change_timer, -EINVAL);
1922 assert(e);
1923 assert(t);
1924
1925 r = sd_event_add_time(e, &timer, CLOCK_MONOTONIC, 0, 0, post_change_thunk, f);
1926 if (r < 0)
1927 return r;
1928
1929 r = sd_event_source_set_enabled(timer, SD_EVENT_OFF);
1930 if (r < 0)
1931 return r;
1932
1933 f->post_change_timer = TAKE_PTR(timer);
1934 f->post_change_timer_period = t;
1935
1936 return r;
1937 }
1938
1939 static int entry_item_cmp(const EntryItem *a, const EntryItem *b) {
1940 return CMP(le64toh(a->object_offset), le64toh(b->object_offset));
1941 }
1942
1943 int journal_file_append_entry(
1944 JournalFile *f,
1945 const dual_timestamp *ts,
1946 const sd_id128_t *boot_id,
1947 const struct iovec iovec[], unsigned n_iovec,
1948 uint64_t *seqnum,
1949 Object **ret, uint64_t *offset) {
1950
1951 unsigned i;
1952 EntryItem *items;
1953 int r;
1954 uint64_t xor_hash = 0;
1955 struct dual_timestamp _ts;
1956
1957 assert(f);
1958 assert(f->header);
1959 assert(iovec || n_iovec == 0);
1960
1961 if (ts) {
1962 if (!VALID_REALTIME(ts->realtime)) {
1963 log_debug("Invalid realtime timestamp %"PRIu64", refusing entry.", ts->realtime);
1964 return -EBADMSG;
1965 }
1966 if (!VALID_MONOTONIC(ts->monotonic)) {
1967 log_debug("Invalid monotomic timestamp %"PRIu64", refusing entry.", ts->monotonic);
1968 return -EBADMSG;
1969 }
1970 } else {
1971 dual_timestamp_get(&_ts);
1972 ts = &_ts;
1973 }
1974
1975 #if HAVE_GCRYPT
1976 r = journal_file_maybe_append_tag(f, ts->realtime);
1977 if (r < 0)
1978 return r;
1979 #endif
1980
1981 /* alloca() can't take 0, hence let's allocate at least one */
1982 items = newa(EntryItem, MAX(1u, n_iovec));
1983
1984 for (i = 0; i < n_iovec; i++) {
1985 uint64_t p;
1986 Object *o;
1987
1988 r = journal_file_append_data(f, iovec[i].iov_base, iovec[i].iov_len, &o, &p);
1989 if (r < 0)
1990 return r;
1991
1992 xor_hash ^= le64toh(o->data.hash);
1993 items[i].object_offset = htole64(p);
1994 items[i].hash = o->data.hash;
1995 }
1996
1997 /* Order by the position on disk, in order to improve seek
1998 * times for rotating media. */
1999 typesafe_qsort(items, n_iovec, entry_item_cmp);
2000
2001 r = journal_file_append_entry_internal(f, ts, boot_id, xor_hash, items, n_iovec, seqnum, ret, offset);
2002
2003 /* If the memory mapping triggered a SIGBUS then we return an
2004 * IO error and ignore the error code passed down to us, since
2005 * it is very likely just an effect of a nullified replacement
2006 * mapping page */
2007
2008 if (mmap_cache_got_sigbus(f->mmap, f->cache_fd))
2009 r = -EIO;
2010
2011 if (f->post_change_timer)
2012 schedule_post_change(f);
2013 else
2014 journal_file_post_change(f);
2015
2016 return r;
2017 }
2018
2019 typedef struct ChainCacheItem {
2020 uint64_t first; /* the array at the beginning of the chain */
2021 uint64_t array; /* the cached array */
2022 uint64_t begin; /* the first item in the cached array */
2023 uint64_t total; /* the total number of items in all arrays before this one in the chain */
2024 uint64_t last_index; /* the last index we looked at, to optimize locality when bisecting */
2025 } ChainCacheItem;
2026
2027 static void chain_cache_put(
2028 OrderedHashmap *h,
2029 ChainCacheItem *ci,
2030 uint64_t first,
2031 uint64_t array,
2032 uint64_t begin,
2033 uint64_t total,
2034 uint64_t last_index) {
2035
2036 if (!ci) {
2037 /* If the chain item to cache for this chain is the
2038 * first one it's not worth caching anything */
2039 if (array == first)
2040 return;
2041
2042 if (ordered_hashmap_size(h) >= CHAIN_CACHE_MAX) {
2043 ci = ordered_hashmap_steal_first(h);
2044 assert(ci);
2045 } else {
2046 ci = new(ChainCacheItem, 1);
2047 if (!ci)
2048 return;
2049 }
2050
2051 ci->first = first;
2052
2053 if (ordered_hashmap_put(h, &ci->first, ci) < 0) {
2054 free(ci);
2055 return;
2056 }
2057 } else
2058 assert(ci->first == first);
2059
2060 ci->array = array;
2061 ci->begin = begin;
2062 ci->total = total;
2063 ci->last_index = last_index;
2064 }
2065
2066 static int generic_array_get(
2067 JournalFile *f,
2068 uint64_t first,
2069 uint64_t i,
2070 Object **ret, uint64_t *offset) {
2071
2072 Object *o;
2073 uint64_t p = 0, a, t = 0;
2074 int r;
2075 ChainCacheItem *ci;
2076
2077 assert(f);
2078
2079 a = first;
2080
2081 /* Try the chain cache first */
2082 ci = ordered_hashmap_get(f->chain_cache, &first);
2083 if (ci && i > ci->total) {
2084 a = ci->array;
2085 i -= ci->total;
2086 t = ci->total;
2087 }
2088
2089 while (a > 0) {
2090 uint64_t k;
2091
2092 r = journal_file_move_to_object(f, OBJECT_ENTRY_ARRAY, a, &o);
2093 if (r < 0)
2094 return r;
2095
2096 k = journal_file_entry_array_n_items(o);
2097 if (i < k) {
2098 p = le64toh(o->entry_array.items[i]);
2099 goto found;
2100 }
2101
2102 i -= k;
2103 t += k;
2104 a = le64toh(o->entry_array.next_entry_array_offset);
2105 }
2106
2107 return 0;
2108
2109 found:
2110 /* Let's cache this item for the next invocation */
2111 chain_cache_put(f->chain_cache, ci, first, a, le64toh(o->entry_array.items[0]), t, i);
2112
2113 r = journal_file_move_to_object(f, OBJECT_ENTRY, p, &o);
2114 if (r < 0)
2115 return r;
2116
2117 if (ret)
2118 *ret = o;
2119
2120 if (offset)
2121 *offset = p;
2122
2123 return 1;
2124 }
2125
2126 static int generic_array_get_plus_one(
2127 JournalFile *f,
2128 uint64_t extra,
2129 uint64_t first,
2130 uint64_t i,
2131 Object **ret, uint64_t *offset) {
2132
2133 Object *o;
2134
2135 assert(f);
2136
2137 if (i == 0) {
2138 int r;
2139
2140 r = journal_file_move_to_object(f, OBJECT_ENTRY, extra, &o);
2141 if (r < 0)
2142 return r;
2143
2144 if (ret)
2145 *ret = o;
2146
2147 if (offset)
2148 *offset = extra;
2149
2150 return 1;
2151 }
2152
2153 return generic_array_get(f, first, i-1, ret, offset);
2154 }
2155
2156 enum {
2157 TEST_FOUND,
2158 TEST_LEFT,
2159 TEST_RIGHT
2160 };
2161
2162 static int generic_array_bisect(
2163 JournalFile *f,
2164 uint64_t first,
2165 uint64_t n,
2166 uint64_t needle,
2167 int (*test_object)(JournalFile *f, uint64_t p, uint64_t needle),
2168 direction_t direction,
2169 Object **ret,
2170 uint64_t *offset,
2171 uint64_t *idx) {
2172
2173 uint64_t a, p, t = 0, i = 0, last_p = 0, last_index = (uint64_t) -1;
2174 bool subtract_one = false;
2175 Object *o, *array = NULL;
2176 int r;
2177 ChainCacheItem *ci;
2178
2179 assert(f);
2180 assert(test_object);
2181
2182 /* Start with the first array in the chain */
2183 a = first;
2184
2185 ci = ordered_hashmap_get(f->chain_cache, &first);
2186 if (ci && n > ci->total && ci->begin != 0) {
2187 /* Ah, we have iterated this bisection array chain
2188 * previously! Let's see if we can skip ahead in the
2189 * chain, as far as the last time. But we can't jump
2190 * backwards in the chain, so let's check that
2191 * first. */
2192
2193 r = test_object(f, ci->begin, needle);
2194 if (r < 0)
2195 return r;
2196
2197 if (r == TEST_LEFT) {
2198 /* OK, what we are looking for is right of the
2199 * begin of this EntryArray, so let's jump
2200 * straight to previously cached array in the
2201 * chain */
2202
2203 a = ci->array;
2204 n -= ci->total;
2205 t = ci->total;
2206 last_index = ci->last_index;
2207 }
2208 }
2209
2210 while (a > 0) {
2211 uint64_t left, right, k, lp;
2212
2213 r = journal_file_move_to_object(f, OBJECT_ENTRY_ARRAY, a, &array);
2214 if (r < 0)
2215 return r;
2216
2217 k = journal_file_entry_array_n_items(array);
2218 right = MIN(k, n);
2219 if (right <= 0)
2220 return 0;
2221
2222 i = right - 1;
2223 lp = p = le64toh(array->entry_array.items[i]);
2224 if (p <= 0)
2225 r = -EBADMSG;
2226 else
2227 r = test_object(f, p, needle);
2228 if (r == -EBADMSG) {
2229 log_debug_errno(r, "Encountered invalid entry while bisecting, cutting algorithm short. (1)");
2230 n = i;
2231 continue;
2232 }
2233 if (r < 0)
2234 return r;
2235
2236 if (r == TEST_FOUND)
2237 r = direction == DIRECTION_DOWN ? TEST_RIGHT : TEST_LEFT;
2238
2239 if (r == TEST_RIGHT) {
2240 left = 0;
2241 right -= 1;
2242
2243 if (last_index != (uint64_t) -1) {
2244 assert(last_index <= right);
2245
2246 /* If we cached the last index we
2247 * looked at, let's try to not to jump
2248 * too wildly around and see if we can
2249 * limit the range to look at early to
2250 * the immediate neighbors of the last
2251 * index we looked at. */
2252
2253 if (last_index > 0) {
2254 uint64_t x = last_index - 1;
2255
2256 p = le64toh(array->entry_array.items[x]);
2257 if (p <= 0)
2258 return -EBADMSG;
2259
2260 r = test_object(f, p, needle);
2261 if (r < 0)
2262 return r;
2263
2264 if (r == TEST_FOUND)
2265 r = direction == DIRECTION_DOWN ? TEST_RIGHT : TEST_LEFT;
2266
2267 if (r == TEST_RIGHT)
2268 right = x;
2269 else
2270 left = x + 1;
2271 }
2272
2273 if (last_index < right) {
2274 uint64_t y = last_index + 1;
2275
2276 p = le64toh(array->entry_array.items[y]);
2277 if (p <= 0)
2278 return -EBADMSG;
2279
2280 r = test_object(f, p, needle);
2281 if (r < 0)
2282 return r;
2283
2284 if (r == TEST_FOUND)
2285 r = direction == DIRECTION_DOWN ? TEST_RIGHT : TEST_LEFT;
2286
2287 if (r == TEST_RIGHT)
2288 right = y;
2289 else
2290 left = y + 1;
2291 }
2292 }
2293
2294 for (;;) {
2295 if (left == right) {
2296 if (direction == DIRECTION_UP)
2297 subtract_one = true;
2298
2299 i = left;
2300 goto found;
2301 }
2302
2303 assert(left < right);
2304 i = (left + right) / 2;
2305
2306 p = le64toh(array->entry_array.items[i]);
2307 if (p <= 0)
2308 r = -EBADMSG;
2309 else
2310 r = test_object(f, p, needle);
2311 if (r == -EBADMSG) {
2312 log_debug_errno(r, "Encountered invalid entry while bisecting, cutting algorithm short. (2)");
2313 right = n = i;
2314 continue;
2315 }
2316 if (r < 0)
2317 return r;
2318
2319 if (r == TEST_FOUND)
2320 r = direction == DIRECTION_DOWN ? TEST_RIGHT : TEST_LEFT;
2321
2322 if (r == TEST_RIGHT)
2323 right = i;
2324 else
2325 left = i + 1;
2326 }
2327 }
2328
2329 if (k >= n) {
2330 if (direction == DIRECTION_UP) {
2331 i = n;
2332 subtract_one = true;
2333 goto found;
2334 }
2335
2336 return 0;
2337 }
2338
2339 last_p = lp;
2340
2341 n -= k;
2342 t += k;
2343 last_index = (uint64_t) -1;
2344 a = le64toh(array->entry_array.next_entry_array_offset);
2345 }
2346
2347 return 0;
2348
2349 found:
2350 if (subtract_one && t == 0 && i == 0)
2351 return 0;
2352
2353 /* Let's cache this item for the next invocation */
2354 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);
2355
2356 if (subtract_one && i == 0)
2357 p = last_p;
2358 else if (subtract_one)
2359 p = le64toh(array->entry_array.items[i-1]);
2360 else
2361 p = le64toh(array->entry_array.items[i]);
2362
2363 r = journal_file_move_to_object(f, OBJECT_ENTRY, p, &o);
2364 if (r < 0)
2365 return r;
2366
2367 if (ret)
2368 *ret = o;
2369
2370 if (offset)
2371 *offset = p;
2372
2373 if (idx)
2374 *idx = t + i + (subtract_one ? -1 : 0);
2375
2376 return 1;
2377 }
2378
2379 static int generic_array_bisect_plus_one(
2380 JournalFile *f,
2381 uint64_t extra,
2382 uint64_t first,
2383 uint64_t n,
2384 uint64_t needle,
2385 int (*test_object)(JournalFile *f, uint64_t p, uint64_t needle),
2386 direction_t direction,
2387 Object **ret,
2388 uint64_t *offset,
2389 uint64_t *idx) {
2390
2391 int r;
2392 bool step_back = false;
2393 Object *o;
2394
2395 assert(f);
2396 assert(test_object);
2397
2398 if (n <= 0)
2399 return 0;
2400
2401 /* This bisects the array in object 'first', but first checks
2402 * an extra */
2403 r = test_object(f, extra, needle);
2404 if (r < 0)
2405 return r;
2406
2407 if (r == TEST_FOUND)
2408 r = direction == DIRECTION_DOWN ? TEST_RIGHT : TEST_LEFT;
2409
2410 /* if we are looking with DIRECTION_UP then we need to first
2411 see if in the actual array there is a matching entry, and
2412 return the last one of that. But if there isn't any we need
2413 to return this one. Hence remember this, and return it
2414 below. */
2415 if (r == TEST_LEFT)
2416 step_back = direction == DIRECTION_UP;
2417
2418 if (r == TEST_RIGHT) {
2419 if (direction == DIRECTION_DOWN)
2420 goto found;
2421 else
2422 return 0;
2423 }
2424
2425 r = generic_array_bisect(f, first, n-1, needle, test_object, direction, ret, offset, idx);
2426
2427 if (r == 0 && step_back)
2428 goto found;
2429
2430 if (r > 0 && idx)
2431 (*idx)++;
2432
2433 return r;
2434
2435 found:
2436 r = journal_file_move_to_object(f, OBJECT_ENTRY, extra, &o);
2437 if (r < 0)
2438 return r;
2439
2440 if (ret)
2441 *ret = o;
2442
2443 if (offset)
2444 *offset = extra;
2445
2446 if (idx)
2447 *idx = 0;
2448
2449 return 1;
2450 }
2451
2452 _pure_ static int test_object_offset(JournalFile *f, uint64_t p, uint64_t needle) {
2453 assert(f);
2454 assert(p > 0);
2455
2456 if (p == needle)
2457 return TEST_FOUND;
2458 else if (p < needle)
2459 return TEST_LEFT;
2460 else
2461 return TEST_RIGHT;
2462 }
2463
2464 static int test_object_seqnum(JournalFile *f, uint64_t p, uint64_t needle) {
2465 Object *o;
2466 int r;
2467
2468 assert(f);
2469 assert(p > 0);
2470
2471 r = journal_file_move_to_object(f, OBJECT_ENTRY, p, &o);
2472 if (r < 0)
2473 return r;
2474
2475 if (le64toh(o->entry.seqnum) == needle)
2476 return TEST_FOUND;
2477 else if (le64toh(o->entry.seqnum) < needle)
2478 return TEST_LEFT;
2479 else
2480 return TEST_RIGHT;
2481 }
2482
2483 int journal_file_move_to_entry_by_seqnum(
2484 JournalFile *f,
2485 uint64_t seqnum,
2486 direction_t direction,
2487 Object **ret,
2488 uint64_t *offset) {
2489 assert(f);
2490 assert(f->header);
2491
2492 return generic_array_bisect(f,
2493 le64toh(f->header->entry_array_offset),
2494 le64toh(f->header->n_entries),
2495 seqnum,
2496 test_object_seqnum,
2497 direction,
2498 ret, offset, NULL);
2499 }
2500
2501 static int test_object_realtime(JournalFile *f, uint64_t p, uint64_t needle) {
2502 Object *o;
2503 int r;
2504
2505 assert(f);
2506 assert(p > 0);
2507
2508 r = journal_file_move_to_object(f, OBJECT_ENTRY, p, &o);
2509 if (r < 0)
2510 return r;
2511
2512 if (le64toh(o->entry.realtime) == needle)
2513 return TEST_FOUND;
2514 else if (le64toh(o->entry.realtime) < needle)
2515 return TEST_LEFT;
2516 else
2517 return TEST_RIGHT;
2518 }
2519
2520 int journal_file_move_to_entry_by_realtime(
2521 JournalFile *f,
2522 uint64_t realtime,
2523 direction_t direction,
2524 Object **ret,
2525 uint64_t *offset) {
2526 assert(f);
2527 assert(f->header);
2528
2529 return generic_array_bisect(f,
2530 le64toh(f->header->entry_array_offset),
2531 le64toh(f->header->n_entries),
2532 realtime,
2533 test_object_realtime,
2534 direction,
2535 ret, offset, NULL);
2536 }
2537
2538 static int test_object_monotonic(JournalFile *f, uint64_t p, uint64_t needle) {
2539 Object *o;
2540 int r;
2541
2542 assert(f);
2543 assert(p > 0);
2544
2545 r = journal_file_move_to_object(f, OBJECT_ENTRY, p, &o);
2546 if (r < 0)
2547 return r;
2548
2549 if (le64toh(o->entry.monotonic) == needle)
2550 return TEST_FOUND;
2551 else if (le64toh(o->entry.monotonic) < needle)
2552 return TEST_LEFT;
2553 else
2554 return TEST_RIGHT;
2555 }
2556
2557 static int find_data_object_by_boot_id(
2558 JournalFile *f,
2559 sd_id128_t boot_id,
2560 Object **o,
2561 uint64_t *b) {
2562
2563 char t[STRLEN("_BOOT_ID=") + 32 + 1] = "_BOOT_ID=";
2564
2565 sd_id128_to_string(boot_id, t + 9);
2566 return journal_file_find_data_object(f, t, sizeof(t) - 1, o, b);
2567 }
2568
2569 int journal_file_move_to_entry_by_monotonic(
2570 JournalFile *f,
2571 sd_id128_t boot_id,
2572 uint64_t monotonic,
2573 direction_t direction,
2574 Object **ret,
2575 uint64_t *offset) {
2576
2577 Object *o;
2578 int r;
2579
2580 assert(f);
2581
2582 r = find_data_object_by_boot_id(f, boot_id, &o, NULL);
2583 if (r < 0)
2584 return r;
2585 if (r == 0)
2586 return -ENOENT;
2587
2588 return generic_array_bisect_plus_one(f,
2589 le64toh(o->data.entry_offset),
2590 le64toh(o->data.entry_array_offset),
2591 le64toh(o->data.n_entries),
2592 monotonic,
2593 test_object_monotonic,
2594 direction,
2595 ret, offset, NULL);
2596 }
2597
2598 void journal_file_reset_location(JournalFile *f) {
2599 f->location_type = LOCATION_HEAD;
2600 f->current_offset = 0;
2601 f->current_seqnum = 0;
2602 f->current_realtime = 0;
2603 f->current_monotonic = 0;
2604 zero(f->current_boot_id);
2605 f->current_xor_hash = 0;
2606 }
2607
2608 void journal_file_save_location(JournalFile *f, Object *o, uint64_t offset) {
2609 f->location_type = LOCATION_SEEK;
2610 f->current_offset = offset;
2611 f->current_seqnum = le64toh(o->entry.seqnum);
2612 f->current_realtime = le64toh(o->entry.realtime);
2613 f->current_monotonic = le64toh(o->entry.monotonic);
2614 f->current_boot_id = o->entry.boot_id;
2615 f->current_xor_hash = le64toh(o->entry.xor_hash);
2616 }
2617
2618 int journal_file_compare_locations(JournalFile *af, JournalFile *bf) {
2619 int r;
2620
2621 assert(af);
2622 assert(af->header);
2623 assert(bf);
2624 assert(bf->header);
2625 assert(af->location_type == LOCATION_SEEK);
2626 assert(bf->location_type == LOCATION_SEEK);
2627
2628 /* If contents and timestamps match, these entries are
2629 * identical, even if the seqnum does not match */
2630 if (sd_id128_equal(af->current_boot_id, bf->current_boot_id) &&
2631 af->current_monotonic == bf->current_monotonic &&
2632 af->current_realtime == bf->current_realtime &&
2633 af->current_xor_hash == bf->current_xor_hash)
2634 return 0;
2635
2636 if (sd_id128_equal(af->header->seqnum_id, bf->header->seqnum_id)) {
2637
2638 /* If this is from the same seqnum source, compare
2639 * seqnums */
2640 r = CMP(af->current_seqnum, bf->current_seqnum);
2641 if (r != 0)
2642 return r;
2643
2644 /* Wow! This is weird, different data but the same
2645 * seqnums? Something is borked, but let's make the
2646 * best of it and compare by time. */
2647 }
2648
2649 if (sd_id128_equal(af->current_boot_id, bf->current_boot_id)) {
2650
2651 /* If the boot id matches, compare monotonic time */
2652 r = CMP(af->current_monotonic, bf->current_monotonic);
2653 if (r != 0)
2654 return r;
2655 }
2656
2657 /* Otherwise, compare UTC time */
2658 r = CMP(af->current_realtime, bf->current_realtime);
2659 if (r != 0)
2660 return r;
2661
2662 /* Finally, compare by contents */
2663 return CMP(af->current_xor_hash, bf->current_xor_hash);
2664 }
2665
2666 static int bump_array_index(uint64_t *i, direction_t direction, uint64_t n) {
2667
2668 /* Increase or decrease the specified index, in the right direction. */
2669
2670 if (direction == DIRECTION_DOWN) {
2671 if (*i >= n - 1)
2672 return 0;
2673
2674 (*i) ++;
2675 } else {
2676 if (*i <= 0)
2677 return 0;
2678
2679 (*i) --;
2680 }
2681
2682 return 1;
2683 }
2684
2685 static bool check_properly_ordered(uint64_t new_offset, uint64_t old_offset, direction_t direction) {
2686
2687 /* Consider it an error if any of the two offsets is uninitialized */
2688 if (old_offset == 0 || new_offset == 0)
2689 return false;
2690
2691 /* If we go down, the new offset must be larger than the old one. */
2692 return direction == DIRECTION_DOWN ?
2693 new_offset > old_offset :
2694 new_offset < old_offset;
2695 }
2696
2697 int journal_file_next_entry(
2698 JournalFile *f,
2699 uint64_t p,
2700 direction_t direction,
2701 Object **ret, uint64_t *offset) {
2702
2703 uint64_t i, n, ofs;
2704 int r;
2705
2706 assert(f);
2707 assert(f->header);
2708
2709 n = le64toh(f->header->n_entries);
2710 if (n <= 0)
2711 return 0;
2712
2713 if (p == 0)
2714 i = direction == DIRECTION_DOWN ? 0 : n - 1;
2715 else {
2716 r = generic_array_bisect(f,
2717 le64toh(f->header->entry_array_offset),
2718 le64toh(f->header->n_entries),
2719 p,
2720 test_object_offset,
2721 DIRECTION_DOWN,
2722 NULL, NULL,
2723 &i);
2724 if (r <= 0)
2725 return r;
2726
2727 r = bump_array_index(&i, direction, n);
2728 if (r <= 0)
2729 return r;
2730 }
2731
2732 /* And jump to it */
2733 for (;;) {
2734 r = generic_array_get(f,
2735 le64toh(f->header->entry_array_offset),
2736 i,
2737 ret, &ofs);
2738 if (r > 0)
2739 break;
2740 if (r != -EBADMSG)
2741 return r;
2742
2743 /* OK, so this entry is borked. Most likely some entry didn't get synced to disk properly, let's see if
2744 * the next one might work for us instead. */
2745 log_debug_errno(r, "Entry item %" PRIu64 " is bad, skipping over it.", i);
2746
2747 r = bump_array_index(&i, direction, n);
2748 if (r <= 0)
2749 return r;
2750 }
2751
2752 /* Ensure our array is properly ordered. */
2753 if (p > 0 && !check_properly_ordered(ofs, p, direction)) {
2754 log_debug("%s: entry array not properly ordered at entry %" PRIu64, f->path, i);
2755 return -EBADMSG;
2756 }
2757
2758 if (offset)
2759 *offset = ofs;
2760
2761 return 1;
2762 }
2763
2764 int journal_file_next_entry_for_data(
2765 JournalFile *f,
2766 Object *o, uint64_t p,
2767 uint64_t data_offset,
2768 direction_t direction,
2769 Object **ret, uint64_t *offset) {
2770
2771 uint64_t i, n, ofs;
2772 Object *d;
2773 int r;
2774
2775 assert(f);
2776 assert(p > 0 || !o);
2777
2778 r = journal_file_move_to_object(f, OBJECT_DATA, data_offset, &d);
2779 if (r < 0)
2780 return r;
2781
2782 n = le64toh(d->data.n_entries);
2783 if (n <= 0)
2784 return n;
2785
2786 if (!o)
2787 i = direction == DIRECTION_DOWN ? 0 : n - 1;
2788 else {
2789 if (o->object.type != OBJECT_ENTRY)
2790 return -EINVAL;
2791
2792 r = generic_array_bisect_plus_one(f,
2793 le64toh(d->data.entry_offset),
2794 le64toh(d->data.entry_array_offset),
2795 le64toh(d->data.n_entries),
2796 p,
2797 test_object_offset,
2798 DIRECTION_DOWN,
2799 NULL, NULL,
2800 &i);
2801
2802 if (r <= 0)
2803 return r;
2804
2805 r = bump_array_index(&i, direction, n);
2806 if (r <= 0)
2807 return r;
2808 }
2809
2810 for (;;) {
2811 r = generic_array_get_plus_one(f,
2812 le64toh(d->data.entry_offset),
2813 le64toh(d->data.entry_array_offset),
2814 i,
2815 ret, &ofs);
2816 if (r > 0)
2817 break;
2818 if (r != -EBADMSG)
2819 return r;
2820
2821 log_debug_errno(r, "Data entry item %" PRIu64 " is bad, skipping over it.", i);
2822
2823 r = bump_array_index(&i, direction, n);
2824 if (r <= 0)
2825 return r;
2826 }
2827
2828 /* Ensure our array is properly ordered. */
2829 if (p > 0 && check_properly_ordered(ofs, p, direction)) {
2830 log_debug("%s data entry array not properly ordered at entry %" PRIu64, f->path, i);
2831 return -EBADMSG;
2832 }
2833
2834 if (offset)
2835 *offset = ofs;
2836
2837 return 1;
2838 }
2839
2840 int journal_file_move_to_entry_by_offset_for_data(
2841 JournalFile *f,
2842 uint64_t data_offset,
2843 uint64_t p,
2844 direction_t direction,
2845 Object **ret, uint64_t *offset) {
2846
2847 int r;
2848 Object *d;
2849
2850 assert(f);
2851
2852 r = journal_file_move_to_object(f, OBJECT_DATA, data_offset, &d);
2853 if (r < 0)
2854 return r;
2855
2856 return generic_array_bisect_plus_one(f,
2857 le64toh(d->data.entry_offset),
2858 le64toh(d->data.entry_array_offset),
2859 le64toh(d->data.n_entries),
2860 p,
2861 test_object_offset,
2862 direction,
2863 ret, offset, NULL);
2864 }
2865
2866 int journal_file_move_to_entry_by_monotonic_for_data(
2867 JournalFile *f,
2868 uint64_t data_offset,
2869 sd_id128_t boot_id,
2870 uint64_t monotonic,
2871 direction_t direction,
2872 Object **ret, uint64_t *offset) {
2873
2874 Object *o, *d;
2875 int r;
2876 uint64_t b, z;
2877
2878 assert(f);
2879
2880 /* First, seek by time */
2881 r = find_data_object_by_boot_id(f, boot_id, &o, &b);
2882 if (r < 0)
2883 return r;
2884 if (r == 0)
2885 return -ENOENT;
2886
2887 r = generic_array_bisect_plus_one(f,
2888 le64toh(o->data.entry_offset),
2889 le64toh(o->data.entry_array_offset),
2890 le64toh(o->data.n_entries),
2891 monotonic,
2892 test_object_monotonic,
2893 direction,
2894 NULL, &z, NULL);
2895 if (r <= 0)
2896 return r;
2897
2898 /* And now, continue seeking until we find an entry that
2899 * exists in both bisection arrays */
2900
2901 for (;;) {
2902 Object *qo;
2903 uint64_t p, q;
2904
2905 r = journal_file_move_to_object(f, OBJECT_DATA, data_offset, &d);
2906 if (r < 0)
2907 return r;
2908
2909 r = generic_array_bisect_plus_one(f,
2910 le64toh(d->data.entry_offset),
2911 le64toh(d->data.entry_array_offset),
2912 le64toh(d->data.n_entries),
2913 z,
2914 test_object_offset,
2915 direction,
2916 NULL, &p, NULL);
2917 if (r <= 0)
2918 return r;
2919
2920 r = journal_file_move_to_object(f, OBJECT_DATA, b, &o);
2921 if (r < 0)
2922 return r;
2923
2924 r = generic_array_bisect_plus_one(f,
2925 le64toh(o->data.entry_offset),
2926 le64toh(o->data.entry_array_offset),
2927 le64toh(o->data.n_entries),
2928 p,
2929 test_object_offset,
2930 direction,
2931 &qo, &q, NULL);
2932
2933 if (r <= 0)
2934 return r;
2935
2936 if (p == q) {
2937 if (ret)
2938 *ret = qo;
2939 if (offset)
2940 *offset = q;
2941
2942 return 1;
2943 }
2944
2945 z = q;
2946 }
2947 }
2948
2949 int journal_file_move_to_entry_by_seqnum_for_data(
2950 JournalFile *f,
2951 uint64_t data_offset,
2952 uint64_t seqnum,
2953 direction_t direction,
2954 Object **ret, uint64_t *offset) {
2955
2956 Object *d;
2957 int r;
2958
2959 assert(f);
2960
2961 r = journal_file_move_to_object(f, OBJECT_DATA, data_offset, &d);
2962 if (r < 0)
2963 return r;
2964
2965 return generic_array_bisect_plus_one(f,
2966 le64toh(d->data.entry_offset),
2967 le64toh(d->data.entry_array_offset),
2968 le64toh(d->data.n_entries),
2969 seqnum,
2970 test_object_seqnum,
2971 direction,
2972 ret, offset, NULL);
2973 }
2974
2975 int journal_file_move_to_entry_by_realtime_for_data(
2976 JournalFile *f,
2977 uint64_t data_offset,
2978 uint64_t realtime,
2979 direction_t direction,
2980 Object **ret, uint64_t *offset) {
2981
2982 Object *d;
2983 int r;
2984
2985 assert(f);
2986
2987 r = journal_file_move_to_object(f, OBJECT_DATA, data_offset, &d);
2988 if (r < 0)
2989 return r;
2990
2991 return generic_array_bisect_plus_one(f,
2992 le64toh(d->data.entry_offset),
2993 le64toh(d->data.entry_array_offset),
2994 le64toh(d->data.n_entries),
2995 realtime,
2996 test_object_realtime,
2997 direction,
2998 ret, offset, NULL);
2999 }
3000
3001 void journal_file_dump(JournalFile *f) {
3002 Object *o;
3003 int r;
3004 uint64_t p;
3005
3006 assert(f);
3007 assert(f->header);
3008
3009 journal_file_print_header(f);
3010
3011 p = le64toh(f->header->header_size);
3012 while (p != 0) {
3013 r = journal_file_move_to_object(f, OBJECT_UNUSED, p, &o);
3014 if (r < 0)
3015 goto fail;
3016
3017 switch (o->object.type) {
3018
3019 case OBJECT_UNUSED:
3020 printf("Type: OBJECT_UNUSED\n");
3021 break;
3022
3023 case OBJECT_DATA:
3024 printf("Type: OBJECT_DATA\n");
3025 break;
3026
3027 case OBJECT_FIELD:
3028 printf("Type: OBJECT_FIELD\n");
3029 break;
3030
3031 case OBJECT_ENTRY:
3032 printf("Type: OBJECT_ENTRY seqnum=%"PRIu64" monotonic=%"PRIu64" realtime=%"PRIu64"\n",
3033 le64toh(o->entry.seqnum),
3034 le64toh(o->entry.monotonic),
3035 le64toh(o->entry.realtime));
3036 break;
3037
3038 case OBJECT_FIELD_HASH_TABLE:
3039 printf("Type: OBJECT_FIELD_HASH_TABLE\n");
3040 break;
3041
3042 case OBJECT_DATA_HASH_TABLE:
3043 printf("Type: OBJECT_DATA_HASH_TABLE\n");
3044 break;
3045
3046 case OBJECT_ENTRY_ARRAY:
3047 printf("Type: OBJECT_ENTRY_ARRAY\n");
3048 break;
3049
3050 case OBJECT_TAG:
3051 printf("Type: OBJECT_TAG seqnum=%"PRIu64" epoch=%"PRIu64"\n",
3052 le64toh(o->tag.seqnum),
3053 le64toh(o->tag.epoch));
3054 break;
3055
3056 default:
3057 printf("Type: unknown (%i)\n", o->object.type);
3058 break;
3059 }
3060
3061 if (o->object.flags & OBJECT_COMPRESSION_MASK)
3062 printf("Flags: %s\n",
3063 object_compressed_to_string(o->object.flags & OBJECT_COMPRESSION_MASK));
3064
3065 if (p == le64toh(f->header->tail_object_offset))
3066 p = 0;
3067 else
3068 p = p + ALIGN64(le64toh(o->object.size));
3069 }
3070
3071 return;
3072 fail:
3073 log_error("File corrupt");
3074 }
3075
3076 static const char* format_timestamp_safe(char *buf, size_t l, usec_t t) {
3077 const char *x;
3078
3079 x = format_timestamp(buf, l, t);
3080 if (x)
3081 return x;
3082 return " --- ";
3083 }
3084
3085 void journal_file_print_header(JournalFile *f) {
3086 char a[33], b[33], c[33], d[33];
3087 char x[FORMAT_TIMESTAMP_MAX], y[FORMAT_TIMESTAMP_MAX], z[FORMAT_TIMESTAMP_MAX];
3088 struct stat st;
3089 char bytes[FORMAT_BYTES_MAX];
3090
3091 assert(f);
3092 assert(f->header);
3093
3094 printf("File Path: %s\n"
3095 "File ID: %s\n"
3096 "Machine ID: %s\n"
3097 "Boot ID: %s\n"
3098 "Sequential Number ID: %s\n"
3099 "State: %s\n"
3100 "Compatible Flags:%s%s\n"
3101 "Incompatible Flags:%s%s%s\n"
3102 "Header size: %"PRIu64"\n"
3103 "Arena size: %"PRIu64"\n"
3104 "Data Hash Table Size: %"PRIu64"\n"
3105 "Field Hash Table Size: %"PRIu64"\n"
3106 "Rotate Suggested: %s\n"
3107 "Head Sequential Number: %"PRIu64" (%"PRIx64")\n"
3108 "Tail Sequential Number: %"PRIu64" (%"PRIx64")\n"
3109 "Head Realtime Timestamp: %s (%"PRIx64")\n"
3110 "Tail Realtime Timestamp: %s (%"PRIx64")\n"
3111 "Tail Monotonic Timestamp: %s (%"PRIx64")\n"
3112 "Objects: %"PRIu64"\n"
3113 "Entry Objects: %"PRIu64"\n",
3114 f->path,
3115 sd_id128_to_string(f->header->file_id, a),
3116 sd_id128_to_string(f->header->machine_id, b),
3117 sd_id128_to_string(f->header->boot_id, c),
3118 sd_id128_to_string(f->header->seqnum_id, d),
3119 f->header->state == STATE_OFFLINE ? "OFFLINE" :
3120 f->header->state == STATE_ONLINE ? "ONLINE" :
3121 f->header->state == STATE_ARCHIVED ? "ARCHIVED" : "UNKNOWN",
3122 JOURNAL_HEADER_SEALED(f->header) ? " SEALED" : "",
3123 (le32toh(f->header->compatible_flags) & ~HEADER_COMPATIBLE_ANY) ? " ???" : "",
3124 JOURNAL_HEADER_COMPRESSED_XZ(f->header) ? " COMPRESSED-XZ" : "",
3125 JOURNAL_HEADER_COMPRESSED_LZ4(f->header) ? " COMPRESSED-LZ4" : "",
3126 (le32toh(f->header->incompatible_flags) & ~HEADER_INCOMPATIBLE_ANY) ? " ???" : "",
3127 le64toh(f->header->header_size),
3128 le64toh(f->header->arena_size),
3129 le64toh(f->header->data_hash_table_size) / sizeof(HashItem),
3130 le64toh(f->header->field_hash_table_size) / sizeof(HashItem),
3131 yes_no(journal_file_rotate_suggested(f, 0)),
3132 le64toh(f->header->head_entry_seqnum), le64toh(f->header->head_entry_seqnum),
3133 le64toh(f->header->tail_entry_seqnum), le64toh(f->header->tail_entry_seqnum),
3134 format_timestamp_safe(x, sizeof(x), le64toh(f->header->head_entry_realtime)), le64toh(f->header->head_entry_realtime),
3135 format_timestamp_safe(y, sizeof(y), le64toh(f->header->tail_entry_realtime)), le64toh(f->header->tail_entry_realtime),
3136 format_timespan(z, sizeof(z), le64toh(f->header->tail_entry_monotonic), USEC_PER_MSEC), le64toh(f->header->tail_entry_monotonic),
3137 le64toh(f->header->n_objects),
3138 le64toh(f->header->n_entries));
3139
3140 if (JOURNAL_HEADER_CONTAINS(f->header, n_data))
3141 printf("Data Objects: %"PRIu64"\n"
3142 "Data Hash Table Fill: %.1f%%\n",
3143 le64toh(f->header->n_data),
3144 100.0 * (double) le64toh(f->header->n_data) / ((double) (le64toh(f->header->data_hash_table_size) / sizeof(HashItem))));
3145
3146 if (JOURNAL_HEADER_CONTAINS(f->header, n_fields))
3147 printf("Field Objects: %"PRIu64"\n"
3148 "Field Hash Table Fill: %.1f%%\n",
3149 le64toh(f->header->n_fields),
3150 100.0 * (double) le64toh(f->header->n_fields) / ((double) (le64toh(f->header->field_hash_table_size) / sizeof(HashItem))));
3151
3152 if (JOURNAL_HEADER_CONTAINS(f->header, n_tags))
3153 printf("Tag Objects: %"PRIu64"\n",
3154 le64toh(f->header->n_tags));
3155 if (JOURNAL_HEADER_CONTAINS(f->header, n_entry_arrays))
3156 printf("Entry Array Objects: %"PRIu64"\n",
3157 le64toh(f->header->n_entry_arrays));
3158
3159 if (fstat(f->fd, &st) >= 0)
3160 printf("Disk usage: %s\n", format_bytes(bytes, sizeof(bytes), (uint64_t) st.st_blocks * 512ULL));
3161 }
3162
3163 static int journal_file_warn_btrfs(JournalFile *f) {
3164 unsigned attrs;
3165 int r;
3166
3167 assert(f);
3168
3169 /* Before we write anything, check if the COW logic is turned
3170 * off on btrfs. Given our write pattern that is quite
3171 * unfriendly to COW file systems this should greatly improve
3172 * performance on COW file systems, such as btrfs, at the
3173 * expense of data integrity features (which shouldn't be too
3174 * bad, given that we do our own checksumming). */
3175
3176 r = btrfs_is_filesystem(f->fd);
3177 if (r < 0)
3178 return log_warning_errno(r, "Failed to determine if journal is on btrfs: %m");
3179 if (!r)
3180 return 0;
3181
3182 r = read_attr_fd(f->fd, &attrs);
3183 if (r < 0)
3184 return log_warning_errno(r, "Failed to read file attributes: %m");
3185
3186 if (attrs & FS_NOCOW_FL) {
3187 log_debug("Detected btrfs file system with copy-on-write disabled, all is good.");
3188 return 0;
3189 }
3190
3191 log_notice("Creating journal file %s on a btrfs file system, and copy-on-write is enabled. "
3192 "This is likely to slow down journal access substantially, please consider turning "
3193 "off the copy-on-write file attribute on the journal directory, using chattr +C.", f->path);
3194
3195 return 1;
3196 }
3197
3198 int journal_file_open(
3199 int fd,
3200 const char *fname,
3201 int flags,
3202 mode_t mode,
3203 bool compress,
3204 uint64_t compress_threshold_bytes,
3205 bool seal,
3206 JournalMetrics *metrics,
3207 MMapCache *mmap_cache,
3208 Set *deferred_closes,
3209 JournalFile *template,
3210 JournalFile **ret) {
3211
3212 bool newly_created = false;
3213 JournalFile *f;
3214 void *h;
3215 int r;
3216 char bytes[FORMAT_BYTES_MAX];
3217
3218 assert(ret);
3219 assert(fd >= 0 || fname);
3220
3221 if (!IN_SET((flags & O_ACCMODE), O_RDONLY, O_RDWR))
3222 return -EINVAL;
3223
3224 if (fname && (flags & O_CREAT) && !endswith(fname, ".journal"))
3225 return -EINVAL;
3226
3227 f = new(JournalFile, 1);
3228 if (!f)
3229 return -ENOMEM;
3230
3231 *f = (JournalFile) {
3232 .fd = fd,
3233 .mode = mode,
3234
3235 .flags = flags,
3236 .prot = prot_from_flags(flags),
3237 .writable = (flags & O_ACCMODE) != O_RDONLY,
3238
3239 #if HAVE_LZ4
3240 .compress_lz4 = compress,
3241 #elif HAVE_XZ
3242 .compress_xz = compress,
3243 #endif
3244 .compress_threshold_bytes = compress_threshold_bytes == (uint64_t) -1 ?
3245 DEFAULT_COMPRESS_THRESHOLD :
3246 MAX(MIN_COMPRESS_THRESHOLD, compress_threshold_bytes),
3247 #if HAVE_GCRYPT
3248 .seal = seal,
3249 #endif
3250 };
3251
3252 log_debug("Journal effective settings seal=%s compress=%s compress_threshold_bytes=%s",
3253 yes_no(f->seal), yes_no(JOURNAL_FILE_COMPRESS(f)),
3254 format_bytes(bytes, sizeof(bytes), f->compress_threshold_bytes));
3255
3256 if (mmap_cache)
3257 f->mmap = mmap_cache_ref(mmap_cache);
3258 else {
3259 f->mmap = mmap_cache_new();
3260 if (!f->mmap) {
3261 r = -ENOMEM;
3262 goto fail;
3263 }
3264 }
3265
3266 if (fname) {
3267 f->path = strdup(fname);
3268 if (!f->path) {
3269 r = -ENOMEM;
3270 goto fail;
3271 }
3272 } else {
3273 assert(fd >= 0);
3274
3275 /* If we don't know the path, fill in something explanatory and vaguely useful */
3276 if (asprintf(&f->path, "/proc/self/%i", fd) < 0) {
3277 r = -ENOMEM;
3278 goto fail;
3279 }
3280 }
3281
3282 f->chain_cache = ordered_hashmap_new(&uint64_hash_ops);
3283 if (!f->chain_cache) {
3284 r = -ENOMEM;
3285 goto fail;
3286 }
3287
3288 if (f->fd < 0) {
3289 /* We pass O_NONBLOCK here, so that in case somebody pointed us to some character device node or FIFO
3290 * or so, we likely fail quickly than block for long. For regular files O_NONBLOCK has no effect, hence
3291 * it doesn't hurt in that case. */
3292
3293 f->fd = open(f->path, f->flags|O_CLOEXEC|O_NONBLOCK, f->mode);
3294 if (f->fd < 0) {
3295 r = -errno;
3296 goto fail;
3297 }
3298
3299 /* fds we opened here by us should also be closed by us. */
3300 f->close_fd = true;
3301
3302 r = fd_nonblock(f->fd, false);
3303 if (r < 0)
3304 goto fail;
3305 }
3306
3307 f->cache_fd = mmap_cache_add_fd(f->mmap, f->fd);
3308 if (!f->cache_fd) {
3309 r = -ENOMEM;
3310 goto fail;
3311 }
3312
3313 r = journal_file_fstat(f);
3314 if (r < 0)
3315 goto fail;
3316
3317 if (f->last_stat.st_size == 0 && f->writable) {
3318
3319 (void) journal_file_warn_btrfs(f);
3320
3321 /* Let's attach the creation time to the journal file, so that the vacuuming code knows the age of this
3322 * file even if the file might end up corrupted one day... Ideally we'd just use the creation time many
3323 * file systems maintain for each file, but the API to query this is very new, hence let's emulate this
3324 * via extended attributes. If extended attributes are not supported we'll just skip this, and rely
3325 * solely on mtime/atime/ctime of the file. */
3326 (void) fd_setcrtime(f->fd, 0);
3327
3328 #if HAVE_GCRYPT
3329 /* Try to load the FSPRG state, and if we can't, then
3330 * just don't do sealing */
3331 if (f->seal) {
3332 r = journal_file_fss_load(f);
3333 if (r < 0)
3334 f->seal = false;
3335 }
3336 #endif
3337
3338 r = journal_file_init_header(f, template);
3339 if (r < 0)
3340 goto fail;
3341
3342 r = journal_file_fstat(f);
3343 if (r < 0)
3344 goto fail;
3345
3346 newly_created = true;
3347 }
3348
3349 if (f->last_stat.st_size < (off_t) HEADER_SIZE_MIN) {
3350 r = -ENODATA;
3351 goto fail;
3352 }
3353
3354 r = mmap_cache_get(f->mmap, f->cache_fd, f->prot, CONTEXT_HEADER, true, 0, PAGE_ALIGN(sizeof(Header)), &f->last_stat, &h, NULL);
3355 if (r < 0)
3356 goto fail;
3357
3358 f->header = h;
3359
3360 if (!newly_created) {
3361 set_clear_with_destructor(deferred_closes, journal_file_close);
3362
3363 r = journal_file_verify_header(f);
3364 if (r < 0)
3365 goto fail;
3366 }
3367
3368 #if HAVE_GCRYPT
3369 if (!newly_created && f->writable) {
3370 r = journal_file_fss_load(f);
3371 if (r < 0)
3372 goto fail;
3373 }
3374 #endif
3375
3376 if (f->writable) {
3377 if (metrics) {
3378 journal_default_metrics(metrics, f->fd);
3379 f->metrics = *metrics;
3380 } else if (template)
3381 f->metrics = template->metrics;
3382
3383 r = journal_file_refresh_header(f);
3384 if (r < 0)
3385 goto fail;
3386 }
3387
3388 #if HAVE_GCRYPT
3389 r = journal_file_hmac_setup(f);
3390 if (r < 0)
3391 goto fail;
3392 #endif
3393
3394 if (newly_created) {
3395 r = journal_file_setup_field_hash_table(f);
3396 if (r < 0)
3397 goto fail;
3398
3399 r = journal_file_setup_data_hash_table(f);
3400 if (r < 0)
3401 goto fail;
3402
3403 #if HAVE_GCRYPT
3404 r = journal_file_append_first_tag(f);
3405 if (r < 0)
3406 goto fail;
3407 #endif
3408 }
3409
3410 if (mmap_cache_got_sigbus(f->mmap, f->cache_fd)) {
3411 r = -EIO;
3412 goto fail;
3413 }
3414
3415 if (template && template->post_change_timer) {
3416 r = journal_file_enable_post_change_timer(
3417 f,
3418 sd_event_source_get_event(template->post_change_timer),
3419 template->post_change_timer_period);
3420
3421 if (r < 0)
3422 goto fail;
3423 }
3424
3425 /* The file is opened now successfully, thus we take possession of any passed in fd. */
3426 f->close_fd = true;
3427
3428 *ret = f;
3429 return 0;
3430
3431 fail:
3432 if (f->cache_fd && mmap_cache_got_sigbus(f->mmap, f->cache_fd))
3433 r = -EIO;
3434
3435 (void) journal_file_close(f);
3436
3437 return r;
3438 }
3439
3440 int journal_file_archive(JournalFile *f) {
3441 _cleanup_free_ char *p = NULL;
3442
3443 assert(f);
3444
3445 if (!f->writable)
3446 return -EINVAL;
3447
3448 /* Is this a journal file that was passed to us as fd? If so, we synthesized a path name for it, and we refuse
3449 * rotation, since we don't know the actual path, and couldn't rename the file hence. */
3450 if (path_startswith(f->path, "/proc/self/fd"))
3451 return -EINVAL;
3452
3453 if (!endswith(f->path, ".journal"))
3454 return -EINVAL;
3455
3456 if (asprintf(&p, "%.*s@" SD_ID128_FORMAT_STR "-%016"PRIx64"-%016"PRIx64".journal",
3457 (int) strlen(f->path) - 8, f->path,
3458 SD_ID128_FORMAT_VAL(f->header->seqnum_id),
3459 le64toh(f->header->head_entry_seqnum),
3460 le64toh(f->header->head_entry_realtime)) < 0)
3461 return -ENOMEM;
3462
3463 /* Try to rename the file to the archived version. If the file already was deleted, we'll get ENOENT, let's
3464 * ignore that case. */
3465 if (rename(f->path, p) < 0 && errno != ENOENT)
3466 return -errno;
3467
3468 /* Sync the rename to disk */
3469 (void) fsync_directory_of_file(f->fd);
3470
3471 /* Set as archive so offlining commits w/state=STATE_ARCHIVED. Previously we would set old_file->header->state
3472 * to STATE_ARCHIVED directly here, but journal_file_set_offline() short-circuits when state != STATE_ONLINE,
3473 * which would result in the rotated journal never getting fsync() called before closing. Now we simply queue
3474 * the archive state by setting an archive bit, leaving the state as STATE_ONLINE so proper offlining
3475 * occurs. */
3476 f->archive = true;
3477
3478 /* Currently, btrfs is not very good with out write patterns and fragments heavily. Let's defrag our journal
3479 * files when we archive them */
3480 f->defrag_on_close = true;
3481
3482 return 0;
3483 }
3484
3485 JournalFile* journal_initiate_close(
3486 JournalFile *f,
3487 Set *deferred_closes) {
3488
3489 int r;
3490
3491 assert(f);
3492
3493 if (deferred_closes) {
3494
3495 r = set_put(deferred_closes, f);
3496 if (r < 0)
3497 log_debug_errno(r, "Failed to add file to deferred close set, closing immediately.");
3498 else {
3499 (void) journal_file_set_offline(f, false);
3500 return NULL;
3501 }
3502 }
3503
3504 return journal_file_close(f);
3505 }
3506
3507 int journal_file_rotate(
3508 JournalFile **f,
3509 bool compress,
3510 uint64_t compress_threshold_bytes,
3511 bool seal,
3512 Set *deferred_closes) {
3513
3514 JournalFile *new_file = NULL;
3515 int r;
3516
3517 assert(f);
3518 assert(*f);
3519
3520 r = journal_file_archive(*f);
3521 if (r < 0)
3522 return r;
3523
3524 r = journal_file_open(
3525 -1,
3526 (*f)->path,
3527 (*f)->flags,
3528 (*f)->mode,
3529 compress,
3530 compress_threshold_bytes,
3531 seal,
3532 NULL, /* metrics */
3533 (*f)->mmap,
3534 deferred_closes,
3535 *f, /* template */
3536 &new_file);
3537
3538 journal_initiate_close(*f, deferred_closes);
3539 *f = new_file;
3540
3541 return r;
3542 }
3543
3544 int journal_file_dispose(int dir_fd, const char *fname) {
3545 _cleanup_free_ char *p = NULL;
3546 _cleanup_close_ int fd = -1;
3547
3548 assert(fname);
3549
3550 /* Renames a journal file to *.journal~, i.e. to mark it as corruped or otherwise uncleanly shutdown. Note that
3551 * this is done without looking into the file or changing any of its contents. The idea is that this is called
3552 * whenever something is suspicious and we want to move the file away and make clear that it is not accessed
3553 * for writing anymore. */
3554
3555 if (!endswith(fname, ".journal"))
3556 return -EINVAL;
3557
3558 if (asprintf(&p, "%.*s@%016" PRIx64 "-%016" PRIx64 ".journal~",
3559 (int) strlen(fname) - 8, fname,
3560 now(CLOCK_REALTIME),
3561 random_u64()) < 0)
3562 return -ENOMEM;
3563
3564 if (renameat(dir_fd, fname, dir_fd, p) < 0)
3565 return -errno;
3566
3567 /* btrfs doesn't cope well with our write pattern and fragments heavily. Let's defrag all files we rotate */
3568 fd = openat(dir_fd, p, O_RDONLY|O_CLOEXEC|O_NOCTTY|O_NOFOLLOW);
3569 if (fd < 0)
3570 log_debug_errno(errno, "Failed to open file for defragmentation/FS_NOCOW_FL, ignoring: %m");
3571 else {
3572 (void) chattr_fd(fd, 0, FS_NOCOW_FL, NULL);
3573 (void) btrfs_defrag_fd(fd);
3574 }
3575
3576 return 0;
3577 }
3578
3579 int journal_file_open_reliably(
3580 const char *fname,
3581 int flags,
3582 mode_t mode,
3583 bool compress,
3584 uint64_t compress_threshold_bytes,
3585 bool seal,
3586 JournalMetrics *metrics,
3587 MMapCache *mmap_cache,
3588 Set *deferred_closes,
3589 JournalFile *template,
3590 JournalFile **ret) {
3591
3592 _cleanup_free_ char *p = NULL;
3593 int r;
3594
3595 r = journal_file_open(-1, fname, flags, mode, compress, compress_threshold_bytes, seal, metrics, mmap_cache,
3596 deferred_closes, template, ret);
3597 if (!IN_SET(r,
3598 -EBADMSG, /* Corrupted */
3599 -ENODATA, /* Truncated */
3600 -EHOSTDOWN, /* Other machine */
3601 -EPROTONOSUPPORT, /* Incompatible feature */
3602 -EBUSY, /* Unclean shutdown */
3603 -ESHUTDOWN, /* Already archived */
3604 -EIO, /* IO error, including SIGBUS on mmap */
3605 -EIDRM, /* File has been deleted */
3606 -ETXTBSY)) /* File is from the future */
3607 return r;
3608
3609 if ((flags & O_ACCMODE) == O_RDONLY)
3610 return r;
3611
3612 if (!(flags & O_CREAT))
3613 return r;
3614
3615 if (!endswith(fname, ".journal"))
3616 return r;
3617
3618 /* The file is corrupted. Rotate it away and try it again (but only once) */
3619 log_warning_errno(r, "File %s corrupted or uncleanly shut down, renaming and replacing.", fname);
3620
3621 r = journal_file_dispose(AT_FDCWD, fname);
3622 if (r < 0)
3623 return r;
3624
3625 return journal_file_open(-1, fname, flags, mode, compress, compress_threshold_bytes, seal, metrics, mmap_cache,
3626 deferred_closes, template, ret);
3627 }
3628
3629 int journal_file_copy_entry(JournalFile *from, JournalFile *to, Object *o, uint64_t p) {
3630 uint64_t i, n;
3631 uint64_t q, xor_hash = 0;
3632 int r;
3633 EntryItem *items;
3634 dual_timestamp ts;
3635 const sd_id128_t *boot_id;
3636
3637 assert(from);
3638 assert(to);
3639 assert(o);
3640 assert(p);
3641
3642 if (!to->writable)
3643 return -EPERM;
3644
3645 ts.monotonic = le64toh(o->entry.monotonic);
3646 ts.realtime = le64toh(o->entry.realtime);
3647 boot_id = &o->entry.boot_id;
3648
3649 n = journal_file_entry_n_items(o);
3650 /* alloca() can't take 0, hence let's allocate at least one */
3651 items = newa(EntryItem, MAX(1u, n));
3652
3653 for (i = 0; i < n; i++) {
3654 uint64_t l, h;
3655 le64_t le_hash;
3656 size_t t;
3657 void *data;
3658 Object *u;
3659
3660 q = le64toh(o->entry.items[i].object_offset);
3661 le_hash = o->entry.items[i].hash;
3662
3663 r = journal_file_move_to_object(from, OBJECT_DATA, q, &o);
3664 if (r < 0)
3665 return r;
3666
3667 if (le_hash != o->data.hash)
3668 return -EBADMSG;
3669
3670 l = le64toh(o->object.size) - offsetof(Object, data.payload);
3671 t = (size_t) l;
3672
3673 /* We hit the limit on 32bit machines */
3674 if ((uint64_t) t != l)
3675 return -E2BIG;
3676
3677 if (o->object.flags & OBJECT_COMPRESSION_MASK) {
3678 #if HAVE_XZ || HAVE_LZ4
3679 size_t rsize = 0;
3680
3681 r = decompress_blob(o->object.flags & OBJECT_COMPRESSION_MASK,
3682 o->data.payload, l, &from->compress_buffer, &from->compress_buffer_size, &rsize, 0);
3683 if (r < 0)
3684 return r;
3685
3686 data = from->compress_buffer;
3687 l = rsize;
3688 #else
3689 return -EPROTONOSUPPORT;
3690 #endif
3691 } else
3692 data = o->data.payload;
3693
3694 r = journal_file_append_data(to, data, l, &u, &h);
3695 if (r < 0)
3696 return r;
3697
3698 xor_hash ^= le64toh(u->data.hash);
3699 items[i].object_offset = htole64(h);
3700 items[i].hash = u->data.hash;
3701
3702 r = journal_file_move_to_object(from, OBJECT_ENTRY, p, &o);
3703 if (r < 0)
3704 return r;
3705 }
3706
3707 r = journal_file_append_entry_internal(to, &ts, boot_id, xor_hash, items, n,
3708 NULL, NULL, NULL);
3709
3710 if (mmap_cache_got_sigbus(to->mmap, to->cache_fd))
3711 return -EIO;
3712
3713 return r;
3714 }
3715
3716 void journal_reset_metrics(JournalMetrics *m) {
3717 assert(m);
3718
3719 /* Set everything to "pick automatic values". */
3720
3721 *m = (JournalMetrics) {
3722 .min_use = (uint64_t) -1,
3723 .max_use = (uint64_t) -1,
3724 .min_size = (uint64_t) -1,
3725 .max_size = (uint64_t) -1,
3726 .keep_free = (uint64_t) -1,
3727 .n_max_files = (uint64_t) -1,
3728 };
3729 }
3730
3731 void journal_default_metrics(JournalMetrics *m, int fd) {
3732 char a[FORMAT_BYTES_MAX], b[FORMAT_BYTES_MAX], c[FORMAT_BYTES_MAX], d[FORMAT_BYTES_MAX], e[FORMAT_BYTES_MAX];
3733 struct statvfs ss;
3734 uint64_t fs_size;
3735
3736 assert(m);
3737 assert(fd >= 0);
3738
3739 if (fstatvfs(fd, &ss) >= 0)
3740 fs_size = ss.f_frsize * ss.f_blocks;
3741 else {
3742 log_debug_errno(errno, "Failed to determine disk size: %m");
3743 fs_size = 0;
3744 }
3745
3746 if (m->max_use == (uint64_t) -1) {
3747
3748 if (fs_size > 0) {
3749 m->max_use = PAGE_ALIGN(fs_size / 10); /* 10% of file system size */
3750
3751 if (m->max_use > DEFAULT_MAX_USE_UPPER)
3752 m->max_use = DEFAULT_MAX_USE_UPPER;
3753
3754 if (m->max_use < DEFAULT_MAX_USE_LOWER)
3755 m->max_use = DEFAULT_MAX_USE_LOWER;
3756 } else
3757 m->max_use = DEFAULT_MAX_USE_LOWER;
3758 } else {
3759 m->max_use = PAGE_ALIGN(m->max_use);
3760
3761 if (m->max_use != 0 && m->max_use < JOURNAL_FILE_SIZE_MIN*2)
3762 m->max_use = JOURNAL_FILE_SIZE_MIN*2;
3763 }
3764
3765 if (m->min_use == (uint64_t) -1)
3766 m->min_use = DEFAULT_MIN_USE;
3767
3768 if (m->min_use > m->max_use)
3769 m->min_use = m->max_use;
3770
3771 if (m->max_size == (uint64_t) -1) {
3772 m->max_size = PAGE_ALIGN(m->max_use / 8); /* 8 chunks */
3773
3774 if (m->max_size > DEFAULT_MAX_SIZE_UPPER)
3775 m->max_size = DEFAULT_MAX_SIZE_UPPER;
3776 } else
3777 m->max_size = PAGE_ALIGN(m->max_size);
3778
3779 if (m->max_size != 0) {
3780 if (m->max_size < JOURNAL_FILE_SIZE_MIN)
3781 m->max_size = JOURNAL_FILE_SIZE_MIN;
3782
3783 if (m->max_use != 0 && m->max_size*2 > m->max_use)
3784 m->max_use = m->max_size*2;
3785 }
3786
3787 if (m->min_size == (uint64_t) -1)
3788 m->min_size = JOURNAL_FILE_SIZE_MIN;
3789 else {
3790 m->min_size = PAGE_ALIGN(m->min_size);
3791
3792 if (m->min_size < JOURNAL_FILE_SIZE_MIN)
3793 m->min_size = JOURNAL_FILE_SIZE_MIN;
3794
3795 if (m->max_size != 0 && m->min_size > m->max_size)
3796 m->max_size = m->min_size;
3797 }
3798
3799 if (m->keep_free == (uint64_t) -1) {
3800
3801 if (fs_size > 0) {
3802 m->keep_free = PAGE_ALIGN(fs_size * 3 / 20); /* 15% of file system size */
3803
3804 if (m->keep_free > DEFAULT_KEEP_FREE_UPPER)
3805 m->keep_free = DEFAULT_KEEP_FREE_UPPER;
3806
3807 } else
3808 m->keep_free = DEFAULT_KEEP_FREE;
3809 }
3810
3811 if (m->n_max_files == (uint64_t) -1)
3812 m->n_max_files = DEFAULT_N_MAX_FILES;
3813
3814 log_debug("Fixed min_use=%s max_use=%s max_size=%s min_size=%s keep_free=%s n_max_files=%" PRIu64,
3815 format_bytes(a, sizeof(a), m->min_use),
3816 format_bytes(b, sizeof(b), m->max_use),
3817 format_bytes(c, sizeof(c), m->max_size),
3818 format_bytes(d, sizeof(d), m->min_size),
3819 format_bytes(e, sizeof(e), m->keep_free),
3820 m->n_max_files);
3821 }
3822
3823 int journal_file_get_cutoff_realtime_usec(JournalFile *f, usec_t *from, usec_t *to) {
3824 assert(f);
3825 assert(f->header);
3826 assert(from || to);
3827
3828 if (from) {
3829 if (f->header->head_entry_realtime == 0)
3830 return -ENOENT;
3831
3832 *from = le64toh(f->header->head_entry_realtime);
3833 }
3834
3835 if (to) {
3836 if (f->header->tail_entry_realtime == 0)
3837 return -ENOENT;
3838
3839 *to = le64toh(f->header->tail_entry_realtime);
3840 }
3841
3842 return 1;
3843 }
3844
3845 int journal_file_get_cutoff_monotonic_usec(JournalFile *f, sd_id128_t boot_id, usec_t *from, usec_t *to) {
3846 Object *o;
3847 uint64_t p;
3848 int r;
3849
3850 assert(f);
3851 assert(from || to);
3852
3853 r = find_data_object_by_boot_id(f, boot_id, &o, &p);
3854 if (r <= 0)
3855 return r;
3856
3857 if (le64toh(o->data.n_entries) <= 0)
3858 return 0;
3859
3860 if (from) {
3861 r = journal_file_move_to_object(f, OBJECT_ENTRY, le64toh(o->data.entry_offset), &o);
3862 if (r < 0)
3863 return r;
3864
3865 *from = le64toh(o->entry.monotonic);
3866 }
3867
3868 if (to) {
3869 r = journal_file_move_to_object(f, OBJECT_DATA, p, &o);
3870 if (r < 0)
3871 return r;
3872
3873 r = generic_array_get_plus_one(f,
3874 le64toh(o->data.entry_offset),
3875 le64toh(o->data.entry_array_offset),
3876 le64toh(o->data.n_entries)-1,
3877 &o, NULL);
3878 if (r <= 0)
3879 return r;
3880
3881 *to = le64toh(o->entry.monotonic);
3882 }
3883
3884 return 1;
3885 }
3886
3887 bool journal_file_rotate_suggested(JournalFile *f, usec_t max_file_usec) {
3888 assert(f);
3889 assert(f->header);
3890
3891 /* If we gained new header fields we gained new features,
3892 * hence suggest a rotation */
3893 if (le64toh(f->header->header_size) < sizeof(Header)) {
3894 log_debug("%s uses an outdated header, suggesting rotation.", f->path);
3895 return true;
3896 }
3897
3898 /* Let's check if the hash tables grew over a certain fill
3899 * level (75%, borrowing this value from Java's hash table
3900 * implementation), and if so suggest a rotation. To calculate
3901 * the fill level we need the n_data field, which only exists
3902 * in newer versions. */
3903
3904 if (JOURNAL_HEADER_CONTAINS(f->header, n_data))
3905 if (le64toh(f->header->n_data) * 4ULL > (le64toh(f->header->data_hash_table_size) / sizeof(HashItem)) * 3ULL) {
3906 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.",
3907 f->path,
3908 100.0 * (double) le64toh(f->header->n_data) / ((double) (le64toh(f->header->data_hash_table_size) / sizeof(HashItem))),
3909 le64toh(f->header->n_data),
3910 le64toh(f->header->data_hash_table_size) / sizeof(HashItem),
3911 (unsigned long long) f->last_stat.st_size,
3912 f->last_stat.st_size / le64toh(f->header->n_data));
3913 return true;
3914 }
3915
3916 if (JOURNAL_HEADER_CONTAINS(f->header, n_fields))
3917 if (le64toh(f->header->n_fields) * 4ULL > (le64toh(f->header->field_hash_table_size) / sizeof(HashItem)) * 3ULL) {
3918 log_debug("Field hash table of %s has a fill level at %.1f (%"PRIu64" of %"PRIu64" items), suggesting rotation.",
3919 f->path,
3920 100.0 * (double) le64toh(f->header->n_fields) / ((double) (le64toh(f->header->field_hash_table_size) / sizeof(HashItem))),
3921 le64toh(f->header->n_fields),
3922 le64toh(f->header->field_hash_table_size) / sizeof(HashItem));
3923 return true;
3924 }
3925
3926 /* Are the data objects properly indexed by field objects? */
3927 if (JOURNAL_HEADER_CONTAINS(f->header, n_data) &&
3928 JOURNAL_HEADER_CONTAINS(f->header, n_fields) &&
3929 le64toh(f->header->n_data) > 0 &&
3930 le64toh(f->header->n_fields) == 0)
3931 return true;
3932
3933 if (max_file_usec > 0) {
3934 usec_t t, h;
3935
3936 h = le64toh(f->header->head_entry_realtime);
3937 t = now(CLOCK_REALTIME);
3938
3939 if (h > 0 && t > h + max_file_usec)
3940 return true;
3941 }
3942
3943 return false;
3944 }
Unnamed repository; edit this file 'description' to name the repository.