]> 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 #10088 from keszybz/man-systemctl-return
[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);
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 /* inotify() does not receive IN_MODIFY events from file
1850 * accesses done via mmap(). After each access we hence
1851 * trigger IN_MODIFY by truncating the journal file to its
1852 * current size which triggers IN_MODIFY. */
1853
1854 __sync_synchronize();
1855
1856 if (ftruncate(f->fd, f->last_stat.st_size) < 0)
1857 log_debug_errno(errno, "Failed to truncate file to its own size: %m");
1858 }
1859
1860 static int post_change_thunk(sd_event_source *timer, uint64_t usec, void *userdata) {
1861 assert(userdata);
1862
1863 journal_file_post_change(userdata);
1864
1865 return 1;
1866 }
1867
1868 static void schedule_post_change(JournalFile *f) {
1869 sd_event_source *timer;
1870 int enabled, r;
1871 uint64_t now;
1872
1873 assert(f);
1874 assert(f->post_change_timer);
1875
1876 timer = f->post_change_timer;
1877
1878 r = sd_event_source_get_enabled(timer, &enabled);
1879 if (r < 0) {
1880 log_debug_errno(r, "Failed to get ftruncate timer state: %m");
1881 goto fail;
1882 }
1883
1884 if (enabled == SD_EVENT_ONESHOT)
1885 return;
1886
1887 r = sd_event_now(sd_event_source_get_event(timer), CLOCK_MONOTONIC, &now);
1888 if (r < 0) {
1889 log_debug_errno(r, "Failed to get clock's now for scheduling ftruncate: %m");
1890 goto fail;
1891 }
1892
1893 r = sd_event_source_set_time(timer, now+f->post_change_timer_period);
1894 if (r < 0) {
1895 log_debug_errno(r, "Failed to set time for scheduling ftruncate: %m");
1896 goto fail;
1897 }
1898
1899 r = sd_event_source_set_enabled(timer, SD_EVENT_ONESHOT);
1900 if (r < 0) {
1901 log_debug_errno(r, "Failed to enable scheduled ftruncate: %m");
1902 goto fail;
1903 }
1904
1905 return;
1906
1907 fail:
1908 /* On failure, let's simply post the change immediately. */
1909 journal_file_post_change(f);
1910 }
1911
1912 /* Enable coalesced change posting in a timer on the provided sd_event instance */
1913 int journal_file_enable_post_change_timer(JournalFile *f, sd_event *e, usec_t t) {
1914 _cleanup_(sd_event_source_unrefp) sd_event_source *timer = NULL;
1915 int r;
1916
1917 assert(f);
1918 assert_return(!f->post_change_timer, -EINVAL);
1919 assert(e);
1920 assert(t);
1921
1922 r = sd_event_add_time(e, &timer, CLOCK_MONOTONIC, 0, 0, post_change_thunk, f);
1923 if (r < 0)
1924 return r;
1925
1926 r = sd_event_source_set_enabled(timer, SD_EVENT_OFF);
1927 if (r < 0)
1928 return r;
1929
1930 f->post_change_timer = TAKE_PTR(timer);
1931 f->post_change_timer_period = t;
1932
1933 return r;
1934 }
1935
1936 static int entry_item_cmp(const EntryItem *a, const EntryItem *b) {
1937 return CMP(le64toh(a->object_offset), le64toh(b->object_offset));
1938 }
1939
1940 int journal_file_append_entry(
1941 JournalFile *f,
1942 const dual_timestamp *ts,
1943 const sd_id128_t *boot_id,
1944 const struct iovec iovec[], unsigned n_iovec,
1945 uint64_t *seqnum,
1946 Object **ret, uint64_t *offset) {
1947
1948 unsigned i;
1949 EntryItem *items;
1950 int r;
1951 uint64_t xor_hash = 0;
1952 struct dual_timestamp _ts;
1953
1954 assert(f);
1955 assert(f->header);
1956 assert(iovec || n_iovec == 0);
1957
1958 if (ts) {
1959 if (!VALID_REALTIME(ts->realtime)) {
1960 log_debug("Invalid realtime timestamp %"PRIu64", refusing entry.", ts->realtime);
1961 return -EBADMSG;
1962 }
1963 if (!VALID_MONOTONIC(ts->monotonic)) {
1964 log_debug("Invalid monotomic timestamp %"PRIu64", refusing entry.", ts->monotonic);
1965 return -EBADMSG;
1966 }
1967 } else {
1968 dual_timestamp_get(&_ts);
1969 ts = &_ts;
1970 }
1971
1972 #if HAVE_GCRYPT
1973 r = journal_file_maybe_append_tag(f, ts->realtime);
1974 if (r < 0)
1975 return r;
1976 #endif
1977
1978 /* alloca() can't take 0, hence let's allocate at least one */
1979 items = newa(EntryItem, MAX(1u, n_iovec));
1980
1981 for (i = 0; i < n_iovec; i++) {
1982 uint64_t p;
1983 Object *o;
1984
1985 r = journal_file_append_data(f, iovec[i].iov_base, iovec[i].iov_len, &o, &p);
1986 if (r < 0)
1987 return r;
1988
1989 xor_hash ^= le64toh(o->data.hash);
1990 items[i].object_offset = htole64(p);
1991 items[i].hash = o->data.hash;
1992 }
1993
1994 /* Order by the position on disk, in order to improve seek
1995 * times for rotating media. */
1996 typesafe_qsort(items, n_iovec, entry_item_cmp);
1997
1998 r = journal_file_append_entry_internal(f, ts, boot_id, xor_hash, items, n_iovec, seqnum, ret, offset);
1999
2000 /* If the memory mapping triggered a SIGBUS then we return an
2001 * IO error and ignore the error code passed down to us, since
2002 * it is very likely just an effect of a nullified replacement
2003 * mapping page */
2004
2005 if (mmap_cache_got_sigbus(f->mmap, f->cache_fd))
2006 r = -EIO;
2007
2008 if (f->post_change_timer)
2009 schedule_post_change(f);
2010 else
2011 journal_file_post_change(f);
2012
2013 return r;
2014 }
2015
2016 typedef struct ChainCacheItem {
2017 uint64_t first; /* the array at the beginning of the chain */
2018 uint64_t array; /* the cached array */
2019 uint64_t begin; /* the first item in the cached array */
2020 uint64_t total; /* the total number of items in all arrays before this one in the chain */
2021 uint64_t last_index; /* the last index we looked at, to optimize locality when bisecting */
2022 } ChainCacheItem;
2023
2024 static void chain_cache_put(
2025 OrderedHashmap *h,
2026 ChainCacheItem *ci,
2027 uint64_t first,
2028 uint64_t array,
2029 uint64_t begin,
2030 uint64_t total,
2031 uint64_t last_index) {
2032
2033 if (!ci) {
2034 /* If the chain item to cache for this chain is the
2035 * first one it's not worth caching anything */
2036 if (array == first)
2037 return;
2038
2039 if (ordered_hashmap_size(h) >= CHAIN_CACHE_MAX) {
2040 ci = ordered_hashmap_steal_first(h);
2041 assert(ci);
2042 } else {
2043 ci = new(ChainCacheItem, 1);
2044 if (!ci)
2045 return;
2046 }
2047
2048 ci->first = first;
2049
2050 if (ordered_hashmap_put(h, &ci->first, ci) < 0) {
2051 free(ci);
2052 return;
2053 }
2054 } else
2055 assert(ci->first == first);
2056
2057 ci->array = array;
2058 ci->begin = begin;
2059 ci->total = total;
2060 ci->last_index = last_index;
2061 }
2062
2063 static int generic_array_get(
2064 JournalFile *f,
2065 uint64_t first,
2066 uint64_t i,
2067 Object **ret, uint64_t *offset) {
2068
2069 Object *o;
2070 uint64_t p = 0, a, t = 0;
2071 int r;
2072 ChainCacheItem *ci;
2073
2074 assert(f);
2075
2076 a = first;
2077
2078 /* Try the chain cache first */
2079 ci = ordered_hashmap_get(f->chain_cache, &first);
2080 if (ci && i > ci->total) {
2081 a = ci->array;
2082 i -= ci->total;
2083 t = ci->total;
2084 }
2085
2086 while (a > 0) {
2087 uint64_t k;
2088
2089 r = journal_file_move_to_object(f, OBJECT_ENTRY_ARRAY, a, &o);
2090 if (r < 0)
2091 return r;
2092
2093 k = journal_file_entry_array_n_items(o);
2094 if (i < k) {
2095 p = le64toh(o->entry_array.items[i]);
2096 goto found;
2097 }
2098
2099 i -= k;
2100 t += k;
2101 a = le64toh(o->entry_array.next_entry_array_offset);
2102 }
2103
2104 return 0;
2105
2106 found:
2107 /* Let's cache this item for the next invocation */
2108 chain_cache_put(f->chain_cache, ci, first, a, le64toh(o->entry_array.items[0]), t, i);
2109
2110 r = journal_file_move_to_object(f, OBJECT_ENTRY, p, &o);
2111 if (r < 0)
2112 return r;
2113
2114 if (ret)
2115 *ret = o;
2116
2117 if (offset)
2118 *offset = p;
2119
2120 return 1;
2121 }
2122
2123 static int generic_array_get_plus_one(
2124 JournalFile *f,
2125 uint64_t extra,
2126 uint64_t first,
2127 uint64_t i,
2128 Object **ret, uint64_t *offset) {
2129
2130 Object *o;
2131
2132 assert(f);
2133
2134 if (i == 0) {
2135 int r;
2136
2137 r = journal_file_move_to_object(f, OBJECT_ENTRY, extra, &o);
2138 if (r < 0)
2139 return r;
2140
2141 if (ret)
2142 *ret = o;
2143
2144 if (offset)
2145 *offset = extra;
2146
2147 return 1;
2148 }
2149
2150 return generic_array_get(f, first, i-1, ret, offset);
2151 }
2152
2153 enum {
2154 TEST_FOUND,
2155 TEST_LEFT,
2156 TEST_RIGHT
2157 };
2158
2159 static int generic_array_bisect(
2160 JournalFile *f,
2161 uint64_t first,
2162 uint64_t n,
2163 uint64_t needle,
2164 int (*test_object)(JournalFile *f, uint64_t p, uint64_t needle),
2165 direction_t direction,
2166 Object **ret,
2167 uint64_t *offset,
2168 uint64_t *idx) {
2169
2170 uint64_t a, p, t = 0, i = 0, last_p = 0, last_index = (uint64_t) -1;
2171 bool subtract_one = false;
2172 Object *o, *array = NULL;
2173 int r;
2174 ChainCacheItem *ci;
2175
2176 assert(f);
2177 assert(test_object);
2178
2179 /* Start with the first array in the chain */
2180 a = first;
2181
2182 ci = ordered_hashmap_get(f->chain_cache, &first);
2183 if (ci && n > ci->total && ci->begin != 0) {
2184 /* Ah, we have iterated this bisection array chain
2185 * previously! Let's see if we can skip ahead in the
2186 * chain, as far as the last time. But we can't jump
2187 * backwards in the chain, so let's check that
2188 * first. */
2189
2190 r = test_object(f, ci->begin, needle);
2191 if (r < 0)
2192 return r;
2193
2194 if (r == TEST_LEFT) {
2195 /* OK, what we are looking for is right of the
2196 * begin of this EntryArray, so let's jump
2197 * straight to previously cached array in the
2198 * chain */
2199
2200 a = ci->array;
2201 n -= ci->total;
2202 t = ci->total;
2203 last_index = ci->last_index;
2204 }
2205 }
2206
2207 while (a > 0) {
2208 uint64_t left, right, k, lp;
2209
2210 r = journal_file_move_to_object(f, OBJECT_ENTRY_ARRAY, a, &array);
2211 if (r < 0)
2212 return r;
2213
2214 k = journal_file_entry_array_n_items(array);
2215 right = MIN(k, n);
2216 if (right <= 0)
2217 return 0;
2218
2219 i = right - 1;
2220 lp = p = le64toh(array->entry_array.items[i]);
2221 if (p <= 0)
2222 r = -EBADMSG;
2223 else
2224 r = test_object(f, p, needle);
2225 if (r == -EBADMSG) {
2226 log_debug_errno(r, "Encountered invalid entry while bisecting, cutting algorithm short. (1)");
2227 n = i;
2228 continue;
2229 }
2230 if (r < 0)
2231 return r;
2232
2233 if (r == TEST_FOUND)
2234 r = direction == DIRECTION_DOWN ? TEST_RIGHT : TEST_LEFT;
2235
2236 if (r == TEST_RIGHT) {
2237 left = 0;
2238 right -= 1;
2239
2240 if (last_index != (uint64_t) -1) {
2241 assert(last_index <= right);
2242
2243 /* If we cached the last index we
2244 * looked at, let's try to not to jump
2245 * too wildly around and see if we can
2246 * limit the range to look at early to
2247 * the immediate neighbors of the last
2248 * index we looked at. */
2249
2250 if (last_index > 0) {
2251 uint64_t x = last_index - 1;
2252
2253 p = le64toh(array->entry_array.items[x]);
2254 if (p <= 0)
2255 return -EBADMSG;
2256
2257 r = test_object(f, p, needle);
2258 if (r < 0)
2259 return r;
2260
2261 if (r == TEST_FOUND)
2262 r = direction == DIRECTION_DOWN ? TEST_RIGHT : TEST_LEFT;
2263
2264 if (r == TEST_RIGHT)
2265 right = x;
2266 else
2267 left = x + 1;
2268 }
2269
2270 if (last_index < right) {
2271 uint64_t y = last_index + 1;
2272
2273 p = le64toh(array->entry_array.items[y]);
2274 if (p <= 0)
2275 return -EBADMSG;
2276
2277 r = test_object(f, p, needle);
2278 if (r < 0)
2279 return r;
2280
2281 if (r == TEST_FOUND)
2282 r = direction == DIRECTION_DOWN ? TEST_RIGHT : TEST_LEFT;
2283
2284 if (r == TEST_RIGHT)
2285 right = y;
2286 else
2287 left = y + 1;
2288 }
2289 }
2290
2291 for (;;) {
2292 if (left == right) {
2293 if (direction == DIRECTION_UP)
2294 subtract_one = true;
2295
2296 i = left;
2297 goto found;
2298 }
2299
2300 assert(left < right);
2301 i = (left + right) / 2;
2302
2303 p = le64toh(array->entry_array.items[i]);
2304 if (p <= 0)
2305 r = -EBADMSG;
2306 else
2307 r = test_object(f, p, needle);
2308 if (r == -EBADMSG) {
2309 log_debug_errno(r, "Encountered invalid entry while bisecting, cutting algorithm short. (2)");
2310 right = n = i;
2311 continue;
2312 }
2313 if (r < 0)
2314 return r;
2315
2316 if (r == TEST_FOUND)
2317 r = direction == DIRECTION_DOWN ? TEST_RIGHT : TEST_LEFT;
2318
2319 if (r == TEST_RIGHT)
2320 right = i;
2321 else
2322 left = i + 1;
2323 }
2324 }
2325
2326 if (k >= n) {
2327 if (direction == DIRECTION_UP) {
2328 i = n;
2329 subtract_one = true;
2330 goto found;
2331 }
2332
2333 return 0;
2334 }
2335
2336 last_p = lp;
2337
2338 n -= k;
2339 t += k;
2340 last_index = (uint64_t) -1;
2341 a = le64toh(array->entry_array.next_entry_array_offset);
2342 }
2343
2344 return 0;
2345
2346 found:
2347 if (subtract_one && t == 0 && i == 0)
2348 return 0;
2349
2350 /* Let's cache this item for the next invocation */
2351 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);
2352
2353 if (subtract_one && i == 0)
2354 p = last_p;
2355 else if (subtract_one)
2356 p = le64toh(array->entry_array.items[i-1]);
2357 else
2358 p = le64toh(array->entry_array.items[i]);
2359
2360 r = journal_file_move_to_object(f, OBJECT_ENTRY, p, &o);
2361 if (r < 0)
2362 return r;
2363
2364 if (ret)
2365 *ret = o;
2366
2367 if (offset)
2368 *offset = p;
2369
2370 if (idx)
2371 *idx = t + i + (subtract_one ? -1 : 0);
2372
2373 return 1;
2374 }
2375
2376 static int generic_array_bisect_plus_one(
2377 JournalFile *f,
2378 uint64_t extra,
2379 uint64_t first,
2380 uint64_t n,
2381 uint64_t needle,
2382 int (*test_object)(JournalFile *f, uint64_t p, uint64_t needle),
2383 direction_t direction,
2384 Object **ret,
2385 uint64_t *offset,
2386 uint64_t *idx) {
2387
2388 int r;
2389 bool step_back = false;
2390 Object *o;
2391
2392 assert(f);
2393 assert(test_object);
2394
2395 if (n <= 0)
2396 return 0;
2397
2398 /* This bisects the array in object 'first', but first checks
2399 * an extra */
2400 r = test_object(f, extra, needle);
2401 if (r < 0)
2402 return r;
2403
2404 if (r == TEST_FOUND)
2405 r = direction == DIRECTION_DOWN ? TEST_RIGHT : TEST_LEFT;
2406
2407 /* if we are looking with DIRECTION_UP then we need to first
2408 see if in the actual array there is a matching entry, and
2409 return the last one of that. But if there isn't any we need
2410 to return this one. Hence remember this, and return it
2411 below. */
2412 if (r == TEST_LEFT)
2413 step_back = direction == DIRECTION_UP;
2414
2415 if (r == TEST_RIGHT) {
2416 if (direction == DIRECTION_DOWN)
2417 goto found;
2418 else
2419 return 0;
2420 }
2421
2422 r = generic_array_bisect(f, first, n-1, needle, test_object, direction, ret, offset, idx);
2423
2424 if (r == 0 && step_back)
2425 goto found;
2426
2427 if (r > 0 && idx)
2428 (*idx)++;
2429
2430 return r;
2431
2432 found:
2433 r = journal_file_move_to_object(f, OBJECT_ENTRY, extra, &o);
2434 if (r < 0)
2435 return r;
2436
2437 if (ret)
2438 *ret = o;
2439
2440 if (offset)
2441 *offset = extra;
2442
2443 if (idx)
2444 *idx = 0;
2445
2446 return 1;
2447 }
2448
2449 _pure_ static int test_object_offset(JournalFile *f, uint64_t p, uint64_t needle) {
2450 assert(f);
2451 assert(p > 0);
2452
2453 if (p == needle)
2454 return TEST_FOUND;
2455 else if (p < needle)
2456 return TEST_LEFT;
2457 else
2458 return TEST_RIGHT;
2459 }
2460
2461 static int test_object_seqnum(JournalFile *f, uint64_t p, uint64_t needle) {
2462 Object *o;
2463 int r;
2464
2465 assert(f);
2466 assert(p > 0);
2467
2468 r = journal_file_move_to_object(f, OBJECT_ENTRY, p, &o);
2469 if (r < 0)
2470 return r;
2471
2472 if (le64toh(o->entry.seqnum) == needle)
2473 return TEST_FOUND;
2474 else if (le64toh(o->entry.seqnum) < needle)
2475 return TEST_LEFT;
2476 else
2477 return TEST_RIGHT;
2478 }
2479
2480 int journal_file_move_to_entry_by_seqnum(
2481 JournalFile *f,
2482 uint64_t seqnum,
2483 direction_t direction,
2484 Object **ret,
2485 uint64_t *offset) {
2486 assert(f);
2487 assert(f->header);
2488
2489 return generic_array_bisect(f,
2490 le64toh(f->header->entry_array_offset),
2491 le64toh(f->header->n_entries),
2492 seqnum,
2493 test_object_seqnum,
2494 direction,
2495 ret, offset, NULL);
2496 }
2497
2498 static int test_object_realtime(JournalFile *f, uint64_t p, uint64_t needle) {
2499 Object *o;
2500 int r;
2501
2502 assert(f);
2503 assert(p > 0);
2504
2505 r = journal_file_move_to_object(f, OBJECT_ENTRY, p, &o);
2506 if (r < 0)
2507 return r;
2508
2509 if (le64toh(o->entry.realtime) == needle)
2510 return TEST_FOUND;
2511 else if (le64toh(o->entry.realtime) < needle)
2512 return TEST_LEFT;
2513 else
2514 return TEST_RIGHT;
2515 }
2516
2517 int journal_file_move_to_entry_by_realtime(
2518 JournalFile *f,
2519 uint64_t realtime,
2520 direction_t direction,
2521 Object **ret,
2522 uint64_t *offset) {
2523 assert(f);
2524 assert(f->header);
2525
2526 return generic_array_bisect(f,
2527 le64toh(f->header->entry_array_offset),
2528 le64toh(f->header->n_entries),
2529 realtime,
2530 test_object_realtime,
2531 direction,
2532 ret, offset, NULL);
2533 }
2534
2535 static int test_object_monotonic(JournalFile *f, uint64_t p, uint64_t needle) {
2536 Object *o;
2537 int r;
2538
2539 assert(f);
2540 assert(p > 0);
2541
2542 r = journal_file_move_to_object(f, OBJECT_ENTRY, p, &o);
2543 if (r < 0)
2544 return r;
2545
2546 if (le64toh(o->entry.monotonic) == needle)
2547 return TEST_FOUND;
2548 else if (le64toh(o->entry.monotonic) < needle)
2549 return TEST_LEFT;
2550 else
2551 return TEST_RIGHT;
2552 }
2553
2554 static int find_data_object_by_boot_id(
2555 JournalFile *f,
2556 sd_id128_t boot_id,
2557 Object **o,
2558 uint64_t *b) {
2559
2560 char t[STRLEN("_BOOT_ID=") + 32 + 1] = "_BOOT_ID=";
2561
2562 sd_id128_to_string(boot_id, t + 9);
2563 return journal_file_find_data_object(f, t, sizeof(t) - 1, o, b);
2564 }
2565
2566 int journal_file_move_to_entry_by_monotonic(
2567 JournalFile *f,
2568 sd_id128_t boot_id,
2569 uint64_t monotonic,
2570 direction_t direction,
2571 Object **ret,
2572 uint64_t *offset) {
2573
2574 Object *o;
2575 int r;
2576
2577 assert(f);
2578
2579 r = find_data_object_by_boot_id(f, boot_id, &o, NULL);
2580 if (r < 0)
2581 return r;
2582 if (r == 0)
2583 return -ENOENT;
2584
2585 return generic_array_bisect_plus_one(f,
2586 le64toh(o->data.entry_offset),
2587 le64toh(o->data.entry_array_offset),
2588 le64toh(o->data.n_entries),
2589 monotonic,
2590 test_object_monotonic,
2591 direction,
2592 ret, offset, NULL);
2593 }
2594
2595 void journal_file_reset_location(JournalFile *f) {
2596 f->location_type = LOCATION_HEAD;
2597 f->current_offset = 0;
2598 f->current_seqnum = 0;
2599 f->current_realtime = 0;
2600 f->current_monotonic = 0;
2601 zero(f->current_boot_id);
2602 f->current_xor_hash = 0;
2603 }
2604
2605 void journal_file_save_location(JournalFile *f, Object *o, uint64_t offset) {
2606 f->location_type = LOCATION_SEEK;
2607 f->current_offset = offset;
2608 f->current_seqnum = le64toh(o->entry.seqnum);
2609 f->current_realtime = le64toh(o->entry.realtime);
2610 f->current_monotonic = le64toh(o->entry.monotonic);
2611 f->current_boot_id = o->entry.boot_id;
2612 f->current_xor_hash = le64toh(o->entry.xor_hash);
2613 }
2614
2615 int journal_file_compare_locations(JournalFile *af, JournalFile *bf) {
2616 assert(af);
2617 assert(af->header);
2618 assert(bf);
2619 assert(bf->header);
2620 assert(af->location_type == LOCATION_SEEK);
2621 assert(bf->location_type == LOCATION_SEEK);
2622
2623 /* If contents and timestamps match, these entries are
2624 * identical, even if the seqnum does not match */
2625 if (sd_id128_equal(af->current_boot_id, bf->current_boot_id) &&
2626 af->current_monotonic == bf->current_monotonic &&
2627 af->current_realtime == bf->current_realtime &&
2628 af->current_xor_hash == bf->current_xor_hash)
2629 return 0;
2630
2631 if (sd_id128_equal(af->header->seqnum_id, bf->header->seqnum_id)) {
2632
2633 /* If this is from the same seqnum source, compare
2634 * seqnums */
2635 if (af->current_seqnum < bf->current_seqnum)
2636 return -1;
2637 if (af->current_seqnum > bf->current_seqnum)
2638 return 1;
2639
2640 /* Wow! This is weird, different data but the same
2641 * seqnums? Something is borked, but let's make the
2642 * best of it and compare by time. */
2643 }
2644
2645 if (sd_id128_equal(af->current_boot_id, bf->current_boot_id)) {
2646
2647 /* If the boot id matches, compare monotonic time */
2648 if (af->current_monotonic < bf->current_monotonic)
2649 return -1;
2650 if (af->current_monotonic > bf->current_monotonic)
2651 return 1;
2652 }
2653
2654 /* Otherwise, compare UTC time */
2655 if (af->current_realtime < bf->current_realtime)
2656 return -1;
2657 if (af->current_realtime > bf->current_realtime)
2658 return 1;
2659
2660 /* Finally, compare by contents */
2661 if (af->current_xor_hash < bf->current_xor_hash)
2662 return -1;
2663 if (af->current_xor_hash > bf->current_xor_hash)
2664 return 1;
2665
2666 return 0;
2667 }
2668
2669 static int bump_array_index(uint64_t *i, direction_t direction, uint64_t n) {
2670
2671 /* Increase or decrease the specified index, in the right direction. */
2672
2673 if (direction == DIRECTION_DOWN) {
2674 if (*i >= n - 1)
2675 return 0;
2676
2677 (*i) ++;
2678 } else {
2679 if (*i <= 0)
2680 return 0;
2681
2682 (*i) --;
2683 }
2684
2685 return 1;
2686 }
2687
2688 static bool check_properly_ordered(uint64_t new_offset, uint64_t old_offset, direction_t direction) {
2689
2690 /* Consider it an error if any of the two offsets is uninitialized */
2691 if (old_offset == 0 || new_offset == 0)
2692 return false;
2693
2694 /* If we go down, the new offset must be larger than the old one. */
2695 return direction == DIRECTION_DOWN ?
2696 new_offset > old_offset :
2697 new_offset < old_offset;
2698 }
2699
2700 int journal_file_next_entry(
2701 JournalFile *f,
2702 uint64_t p,
2703 direction_t direction,
2704 Object **ret, uint64_t *offset) {
2705
2706 uint64_t i, n, ofs;
2707 int r;
2708
2709 assert(f);
2710 assert(f->header);
2711
2712 n = le64toh(f->header->n_entries);
2713 if (n <= 0)
2714 return 0;
2715
2716 if (p == 0)
2717 i = direction == DIRECTION_DOWN ? 0 : n - 1;
2718 else {
2719 r = generic_array_bisect(f,
2720 le64toh(f->header->entry_array_offset),
2721 le64toh(f->header->n_entries),
2722 p,
2723 test_object_offset,
2724 DIRECTION_DOWN,
2725 NULL, NULL,
2726 &i);
2727 if (r <= 0)
2728 return r;
2729
2730 r = bump_array_index(&i, direction, n);
2731 if (r <= 0)
2732 return r;
2733 }
2734
2735 /* And jump to it */
2736 for (;;) {
2737 r = generic_array_get(f,
2738 le64toh(f->header->entry_array_offset),
2739 i,
2740 ret, &ofs);
2741 if (r > 0)
2742 break;
2743 if (r != -EBADMSG)
2744 return r;
2745
2746 /* OK, so this entry is borked. Most likely some entry didn't get synced to disk properly, let's see if
2747 * the next one might work for us instead. */
2748 log_debug_errno(r, "Entry item %" PRIu64 " is bad, skipping over it.", i);
2749
2750 r = bump_array_index(&i, direction, n);
2751 if (r <= 0)
2752 return r;
2753 }
2754
2755 /* Ensure our array is properly ordered. */
2756 if (p > 0 && !check_properly_ordered(ofs, p, direction)) {
2757 log_debug("%s: entry array not properly ordered at entry %" PRIu64, f->path, i);
2758 return -EBADMSG;
2759 }
2760
2761 if (offset)
2762 *offset = ofs;
2763
2764 return 1;
2765 }
2766
2767 int journal_file_next_entry_for_data(
2768 JournalFile *f,
2769 Object *o, uint64_t p,
2770 uint64_t data_offset,
2771 direction_t direction,
2772 Object **ret, uint64_t *offset) {
2773
2774 uint64_t i, n, ofs;
2775 Object *d;
2776 int r;
2777
2778 assert(f);
2779 assert(p > 0 || !o);
2780
2781 r = journal_file_move_to_object(f, OBJECT_DATA, data_offset, &d);
2782 if (r < 0)
2783 return r;
2784
2785 n = le64toh(d->data.n_entries);
2786 if (n <= 0)
2787 return n;
2788
2789 if (!o)
2790 i = direction == DIRECTION_DOWN ? 0 : n - 1;
2791 else {
2792 if (o->object.type != OBJECT_ENTRY)
2793 return -EINVAL;
2794
2795 r = generic_array_bisect_plus_one(f,
2796 le64toh(d->data.entry_offset),
2797 le64toh(d->data.entry_array_offset),
2798 le64toh(d->data.n_entries),
2799 p,
2800 test_object_offset,
2801 DIRECTION_DOWN,
2802 NULL, NULL,
2803 &i);
2804
2805 if (r <= 0)
2806 return r;
2807
2808 r = bump_array_index(&i, direction, n);
2809 if (r <= 0)
2810 return r;
2811 }
2812
2813 for (;;) {
2814 r = generic_array_get_plus_one(f,
2815 le64toh(d->data.entry_offset),
2816 le64toh(d->data.entry_array_offset),
2817 i,
2818 ret, &ofs);
2819 if (r > 0)
2820 break;
2821 if (r != -EBADMSG)
2822 return r;
2823
2824 log_debug_errno(r, "Data entry item %" PRIu64 " is bad, skipping over it.", i);
2825
2826 r = bump_array_index(&i, direction, n);
2827 if (r <= 0)
2828 return r;
2829 }
2830
2831 /* Ensure our array is properly ordered. */
2832 if (p > 0 && check_properly_ordered(ofs, p, direction)) {
2833 log_debug("%s data entry array not properly ordered at entry %" PRIu64, f->path, i);
2834 return -EBADMSG;
2835 }
2836
2837 if (offset)
2838 *offset = ofs;
2839
2840 return 1;
2841 }
2842
2843 int journal_file_move_to_entry_by_offset_for_data(
2844 JournalFile *f,
2845 uint64_t data_offset,
2846 uint64_t p,
2847 direction_t direction,
2848 Object **ret, uint64_t *offset) {
2849
2850 int r;
2851 Object *d;
2852
2853 assert(f);
2854
2855 r = journal_file_move_to_object(f, OBJECT_DATA, data_offset, &d);
2856 if (r < 0)
2857 return r;
2858
2859 return generic_array_bisect_plus_one(f,
2860 le64toh(d->data.entry_offset),
2861 le64toh(d->data.entry_array_offset),
2862 le64toh(d->data.n_entries),
2863 p,
2864 test_object_offset,
2865 direction,
2866 ret, offset, NULL);
2867 }
2868
2869 int journal_file_move_to_entry_by_monotonic_for_data(
2870 JournalFile *f,
2871 uint64_t data_offset,
2872 sd_id128_t boot_id,
2873 uint64_t monotonic,
2874 direction_t direction,
2875 Object **ret, uint64_t *offset) {
2876
2877 Object *o, *d;
2878 int r;
2879 uint64_t b, z;
2880
2881 assert(f);
2882
2883 /* First, seek by time */
2884 r = find_data_object_by_boot_id(f, boot_id, &o, &b);
2885 if (r < 0)
2886 return r;
2887 if (r == 0)
2888 return -ENOENT;
2889
2890 r = generic_array_bisect_plus_one(f,
2891 le64toh(o->data.entry_offset),
2892 le64toh(o->data.entry_array_offset),
2893 le64toh(o->data.n_entries),
2894 monotonic,
2895 test_object_monotonic,
2896 direction,
2897 NULL, &z, NULL);
2898 if (r <= 0)
2899 return r;
2900
2901 /* And now, continue seeking until we find an entry that
2902 * exists in both bisection arrays */
2903
2904 for (;;) {
2905 Object *qo;
2906 uint64_t p, q;
2907
2908 r = journal_file_move_to_object(f, OBJECT_DATA, data_offset, &d);
2909 if (r < 0)
2910 return r;
2911
2912 r = generic_array_bisect_plus_one(f,
2913 le64toh(d->data.entry_offset),
2914 le64toh(d->data.entry_array_offset),
2915 le64toh(d->data.n_entries),
2916 z,
2917 test_object_offset,
2918 direction,
2919 NULL, &p, NULL);
2920 if (r <= 0)
2921 return r;
2922
2923 r = journal_file_move_to_object(f, OBJECT_DATA, b, &o);
2924 if (r < 0)
2925 return r;
2926
2927 r = generic_array_bisect_plus_one(f,
2928 le64toh(o->data.entry_offset),
2929 le64toh(o->data.entry_array_offset),
2930 le64toh(o->data.n_entries),
2931 p,
2932 test_object_offset,
2933 direction,
2934 &qo, &q, NULL);
2935
2936 if (r <= 0)
2937 return r;
2938
2939 if (p == q) {
2940 if (ret)
2941 *ret = qo;
2942 if (offset)
2943 *offset = q;
2944
2945 return 1;
2946 }
2947
2948 z = q;
2949 }
2950 }
2951
2952 int journal_file_move_to_entry_by_seqnum_for_data(
2953 JournalFile *f,
2954 uint64_t data_offset,
2955 uint64_t seqnum,
2956 direction_t direction,
2957 Object **ret, uint64_t *offset) {
2958
2959 Object *d;
2960 int r;
2961
2962 assert(f);
2963
2964 r = journal_file_move_to_object(f, OBJECT_DATA, data_offset, &d);
2965 if (r < 0)
2966 return r;
2967
2968 return generic_array_bisect_plus_one(f,
2969 le64toh(d->data.entry_offset),
2970 le64toh(d->data.entry_array_offset),
2971 le64toh(d->data.n_entries),
2972 seqnum,
2973 test_object_seqnum,
2974 direction,
2975 ret, offset, NULL);
2976 }
2977
2978 int journal_file_move_to_entry_by_realtime_for_data(
2979 JournalFile *f,
2980 uint64_t data_offset,
2981 uint64_t realtime,
2982 direction_t direction,
2983 Object **ret, uint64_t *offset) {
2984
2985 Object *d;
2986 int r;
2987
2988 assert(f);
2989
2990 r = journal_file_move_to_object(f, OBJECT_DATA, data_offset, &d);
2991 if (r < 0)
2992 return r;
2993
2994 return generic_array_bisect_plus_one(f,
2995 le64toh(d->data.entry_offset),
2996 le64toh(d->data.entry_array_offset),
2997 le64toh(d->data.n_entries),
2998 realtime,
2999 test_object_realtime,
3000 direction,
3001 ret, offset, NULL);
3002 }
3003
3004 void journal_file_dump(JournalFile *f) {
3005 Object *o;
3006 int r;
3007 uint64_t p;
3008
3009 assert(f);
3010 assert(f->header);
3011
3012 journal_file_print_header(f);
3013
3014 p = le64toh(f->header->header_size);
3015 while (p != 0) {
3016 r = journal_file_move_to_object(f, OBJECT_UNUSED, p, &o);
3017 if (r < 0)
3018 goto fail;
3019
3020 switch (o->object.type) {
3021
3022 case OBJECT_UNUSED:
3023 printf("Type: OBJECT_UNUSED\n");
3024 break;
3025
3026 case OBJECT_DATA:
3027 printf("Type: OBJECT_DATA\n");
3028 break;
3029
3030 case OBJECT_FIELD:
3031 printf("Type: OBJECT_FIELD\n");
3032 break;
3033
3034 case OBJECT_ENTRY:
3035 printf("Type: OBJECT_ENTRY seqnum=%"PRIu64" monotonic=%"PRIu64" realtime=%"PRIu64"\n",
3036 le64toh(o->entry.seqnum),
3037 le64toh(o->entry.monotonic),
3038 le64toh(o->entry.realtime));
3039 break;
3040
3041 case OBJECT_FIELD_HASH_TABLE:
3042 printf("Type: OBJECT_FIELD_HASH_TABLE\n");
3043 break;
3044
3045 case OBJECT_DATA_HASH_TABLE:
3046 printf("Type: OBJECT_DATA_HASH_TABLE\n");
3047 break;
3048
3049 case OBJECT_ENTRY_ARRAY:
3050 printf("Type: OBJECT_ENTRY_ARRAY\n");
3051 break;
3052
3053 case OBJECT_TAG:
3054 printf("Type: OBJECT_TAG seqnum=%"PRIu64" epoch=%"PRIu64"\n",
3055 le64toh(o->tag.seqnum),
3056 le64toh(o->tag.epoch));
3057 break;
3058
3059 default:
3060 printf("Type: unknown (%i)\n", o->object.type);
3061 break;
3062 }
3063
3064 if (o->object.flags & OBJECT_COMPRESSION_MASK)
3065 printf("Flags: %s\n",
3066 object_compressed_to_string(o->object.flags & OBJECT_COMPRESSION_MASK));
3067
3068 if (p == le64toh(f->header->tail_object_offset))
3069 p = 0;
3070 else
3071 p = p + ALIGN64(le64toh(o->object.size));
3072 }
3073
3074 return;
3075 fail:
3076 log_error("File corrupt");
3077 }
3078
3079 static const char* format_timestamp_safe(char *buf, size_t l, usec_t t) {
3080 const char *x;
3081
3082 x = format_timestamp(buf, l, t);
3083 if (x)
3084 return x;
3085 return " --- ";
3086 }
3087
3088 void journal_file_print_header(JournalFile *f) {
3089 char a[33], b[33], c[33], d[33];
3090 char x[FORMAT_TIMESTAMP_MAX], y[FORMAT_TIMESTAMP_MAX], z[FORMAT_TIMESTAMP_MAX];
3091 struct stat st;
3092 char bytes[FORMAT_BYTES_MAX];
3093
3094 assert(f);
3095 assert(f->header);
3096
3097 printf("File Path: %s\n"
3098 "File ID: %s\n"
3099 "Machine ID: %s\n"
3100 "Boot ID: %s\n"
3101 "Sequential Number ID: %s\n"
3102 "State: %s\n"
3103 "Compatible Flags:%s%s\n"
3104 "Incompatible Flags:%s%s%s\n"
3105 "Header size: %"PRIu64"\n"
3106 "Arena size: %"PRIu64"\n"
3107 "Data Hash Table Size: %"PRIu64"\n"
3108 "Field Hash Table Size: %"PRIu64"\n"
3109 "Rotate Suggested: %s\n"
3110 "Head Sequential Number: %"PRIu64" (%"PRIx64")\n"
3111 "Tail Sequential Number: %"PRIu64" (%"PRIx64")\n"
3112 "Head Realtime Timestamp: %s (%"PRIx64")\n"
3113 "Tail Realtime Timestamp: %s (%"PRIx64")\n"
3114 "Tail Monotonic Timestamp: %s (%"PRIx64")\n"
3115 "Objects: %"PRIu64"\n"
3116 "Entry Objects: %"PRIu64"\n",
3117 f->path,
3118 sd_id128_to_string(f->header->file_id, a),
3119 sd_id128_to_string(f->header->machine_id, b),
3120 sd_id128_to_string(f->header->boot_id, c),
3121 sd_id128_to_string(f->header->seqnum_id, d),
3122 f->header->state == STATE_OFFLINE ? "OFFLINE" :
3123 f->header->state == STATE_ONLINE ? "ONLINE" :
3124 f->header->state == STATE_ARCHIVED ? "ARCHIVED" : "UNKNOWN",
3125 JOURNAL_HEADER_SEALED(f->header) ? " SEALED" : "",
3126 (le32toh(f->header->compatible_flags) & ~HEADER_COMPATIBLE_ANY) ? " ???" : "",
3127 JOURNAL_HEADER_COMPRESSED_XZ(f->header) ? " COMPRESSED-XZ" : "",
3128 JOURNAL_HEADER_COMPRESSED_LZ4(f->header) ? " COMPRESSED-LZ4" : "",
3129 (le32toh(f->header->incompatible_flags) & ~HEADER_INCOMPATIBLE_ANY) ? " ???" : "",
3130 le64toh(f->header->header_size),
3131 le64toh(f->header->arena_size),
3132 le64toh(f->header->data_hash_table_size) / sizeof(HashItem),
3133 le64toh(f->header->field_hash_table_size) / sizeof(HashItem),
3134 yes_no(journal_file_rotate_suggested(f, 0)),
3135 le64toh(f->header->head_entry_seqnum), le64toh(f->header->head_entry_seqnum),
3136 le64toh(f->header->tail_entry_seqnum), le64toh(f->header->tail_entry_seqnum),
3137 format_timestamp_safe(x, sizeof(x), le64toh(f->header->head_entry_realtime)), le64toh(f->header->head_entry_realtime),
3138 format_timestamp_safe(y, sizeof(y), le64toh(f->header->tail_entry_realtime)), le64toh(f->header->tail_entry_realtime),
3139 format_timespan(z, sizeof(z), le64toh(f->header->tail_entry_monotonic), USEC_PER_MSEC), le64toh(f->header->tail_entry_monotonic),
3140 le64toh(f->header->n_objects),
3141 le64toh(f->header->n_entries));
3142
3143 if (JOURNAL_HEADER_CONTAINS(f->header, n_data))
3144 printf("Data Objects: %"PRIu64"\n"
3145 "Data Hash Table Fill: %.1f%%\n",
3146 le64toh(f->header->n_data),
3147 100.0 * (double) le64toh(f->header->n_data) / ((double) (le64toh(f->header->data_hash_table_size) / sizeof(HashItem))));
3148
3149 if (JOURNAL_HEADER_CONTAINS(f->header, n_fields))
3150 printf("Field Objects: %"PRIu64"\n"
3151 "Field Hash Table Fill: %.1f%%\n",
3152 le64toh(f->header->n_fields),
3153 100.0 * (double) le64toh(f->header->n_fields) / ((double) (le64toh(f->header->field_hash_table_size) / sizeof(HashItem))));
3154
3155 if (JOURNAL_HEADER_CONTAINS(f->header, n_tags))
3156 printf("Tag Objects: %"PRIu64"\n",
3157 le64toh(f->header->n_tags));
3158 if (JOURNAL_HEADER_CONTAINS(f->header, n_entry_arrays))
3159 printf("Entry Array Objects: %"PRIu64"\n",
3160 le64toh(f->header->n_entry_arrays));
3161
3162 if (fstat(f->fd, &st) >= 0)
3163 printf("Disk usage: %s\n", format_bytes(bytes, sizeof(bytes), (uint64_t) st.st_blocks * 512ULL));
3164 }
3165
3166 static int journal_file_warn_btrfs(JournalFile *f) {
3167 unsigned attrs;
3168 int r;
3169
3170 assert(f);
3171
3172 /* Before we write anything, check if the COW logic is turned
3173 * off on btrfs. Given our write pattern that is quite
3174 * unfriendly to COW file systems this should greatly improve
3175 * performance on COW file systems, such as btrfs, at the
3176 * expense of data integrity features (which shouldn't be too
3177 * bad, given that we do our own checksumming). */
3178
3179 r = btrfs_is_filesystem(f->fd);
3180 if (r < 0)
3181 return log_warning_errno(r, "Failed to determine if journal is on btrfs: %m");
3182 if (!r)
3183 return 0;
3184
3185 r = read_attr_fd(f->fd, &attrs);
3186 if (r < 0)
3187 return log_warning_errno(r, "Failed to read file attributes: %m");
3188
3189 if (attrs & FS_NOCOW_FL) {
3190 log_debug("Detected btrfs file system with copy-on-write disabled, all is good.");
3191 return 0;
3192 }
3193
3194 log_notice("Creating journal file %s on a btrfs file system, and copy-on-write is enabled. "
3195 "This is likely to slow down journal access substantially, please consider turning "
3196 "off the copy-on-write file attribute on the journal directory, using chattr +C.", f->path);
3197
3198 return 1;
3199 }
3200
3201 int journal_file_open(
3202 int fd,
3203 const char *fname,
3204 int flags,
3205 mode_t mode,
3206 bool compress,
3207 uint64_t compress_threshold_bytes,
3208 bool seal,
3209 JournalMetrics *metrics,
3210 MMapCache *mmap_cache,
3211 Set *deferred_closes,
3212 JournalFile *template,
3213 JournalFile **ret) {
3214
3215 bool newly_created = false;
3216 JournalFile *f;
3217 void *h;
3218 int r;
3219 char bytes[FORMAT_BYTES_MAX];
3220
3221 assert(ret);
3222 assert(fd >= 0 || fname);
3223
3224 if (!IN_SET((flags & O_ACCMODE), O_RDONLY, O_RDWR))
3225 return -EINVAL;
3226
3227 if (fname && (flags & O_CREAT) && !endswith(fname, ".journal"))
3228 return -EINVAL;
3229
3230 f = new0(JournalFile, 1);
3231 if (!f)
3232 return -ENOMEM;
3233
3234 f->fd = fd;
3235 f->mode = mode;
3236
3237 f->flags = flags;
3238 f->prot = prot_from_flags(flags);
3239 f->writable = (flags & O_ACCMODE) != O_RDONLY;
3240 #if HAVE_LZ4
3241 f->compress_lz4 = compress;
3242 #elif HAVE_XZ
3243 f->compress_xz = compress;
3244 #endif
3245
3246 if (compress_threshold_bytes == (uint64_t) -1)
3247 f->compress_threshold_bytes = DEFAULT_COMPRESS_THRESHOLD;
3248 else
3249 f->compress_threshold_bytes = MAX(MIN_COMPRESS_THRESHOLD, compress_threshold_bytes);
3250
3251 #if HAVE_GCRYPT
3252 f->seal = seal;
3253 #endif
3254
3255 log_debug("Journal effective settings seal=%s compress=%s compress_threshold_bytes=%s",
3256 yes_no(f->seal), yes_no(JOURNAL_FILE_COMPRESS(f)),
3257 format_bytes(bytes, sizeof(bytes), f->compress_threshold_bytes));
3258
3259 if (mmap_cache)
3260 f->mmap = mmap_cache_ref(mmap_cache);
3261 else {
3262 f->mmap = mmap_cache_new();
3263 if (!f->mmap) {
3264 r = -ENOMEM;
3265 goto fail;
3266 }
3267 }
3268
3269 if (fname) {
3270 f->path = strdup(fname);
3271 if (!f->path) {
3272 r = -ENOMEM;
3273 goto fail;
3274 }
3275 } else {
3276 assert(fd >= 0);
3277
3278 /* If we don't know the path, fill in something explanatory and vaguely useful */
3279 if (asprintf(&f->path, "/proc/self/%i", fd) < 0) {
3280 r = -ENOMEM;
3281 goto fail;
3282 }
3283 }
3284
3285 f->chain_cache = ordered_hashmap_new(&uint64_hash_ops);
3286 if (!f->chain_cache) {
3287 r = -ENOMEM;
3288 goto fail;
3289 }
3290
3291 if (f->fd < 0) {
3292 /* We pass O_NONBLOCK here, so that in case somebody pointed us to some character device node or FIFO
3293 * or so, we likely fail quickly than block for long. For regular files O_NONBLOCK has no effect, hence
3294 * it doesn't hurt in that case. */
3295
3296 f->fd = open(f->path, f->flags|O_CLOEXEC|O_NONBLOCK, f->mode);
3297 if (f->fd < 0) {
3298 r = -errno;
3299 goto fail;
3300 }
3301
3302 /* fds we opened here by us should also be closed by us. */
3303 f->close_fd = true;
3304
3305 r = fd_nonblock(f->fd, false);
3306 if (r < 0)
3307 goto fail;
3308 }
3309
3310 f->cache_fd = mmap_cache_add_fd(f->mmap, f->fd);
3311 if (!f->cache_fd) {
3312 r = -ENOMEM;
3313 goto fail;
3314 }
3315
3316 r = journal_file_fstat(f);
3317 if (r < 0)
3318 goto fail;
3319
3320 if (f->last_stat.st_size == 0 && f->writable) {
3321
3322 (void) journal_file_warn_btrfs(f);
3323
3324 /* Let's attach the creation time to the journal file, so that the vacuuming code knows the age of this
3325 * file even if the file might end up corrupted one day... Ideally we'd just use the creation time many
3326 * file systems maintain for each file, but the API to query this is very new, hence let's emulate this
3327 * via extended attributes. If extended attributes are not supported we'll just skip this, and rely
3328 * solely on mtime/atime/ctime of the file. */
3329 (void) fd_setcrtime(f->fd, 0);
3330
3331 #if HAVE_GCRYPT
3332 /* Try to load the FSPRG state, and if we can't, then
3333 * just don't do sealing */
3334 if (f->seal) {
3335 r = journal_file_fss_load(f);
3336 if (r < 0)
3337 f->seal = false;
3338 }
3339 #endif
3340
3341 r = journal_file_init_header(f, template);
3342 if (r < 0)
3343 goto fail;
3344
3345 r = journal_file_fstat(f);
3346 if (r < 0)
3347 goto fail;
3348
3349 newly_created = true;
3350 }
3351
3352 if (f->last_stat.st_size < (off_t) HEADER_SIZE_MIN) {
3353 r = -ENODATA;
3354 goto fail;
3355 }
3356
3357 r = mmap_cache_get(f->mmap, f->cache_fd, f->prot, CONTEXT_HEADER, true, 0, PAGE_ALIGN(sizeof(Header)), &f->last_stat, &h, NULL);
3358 if (r < 0)
3359 goto fail;
3360
3361 f->header = h;
3362
3363 if (!newly_created) {
3364 set_clear_with_destructor(deferred_closes, journal_file_close);
3365
3366 r = journal_file_verify_header(f);
3367 if (r < 0)
3368 goto fail;
3369 }
3370
3371 #if HAVE_GCRYPT
3372 if (!newly_created && f->writable) {
3373 r = journal_file_fss_load(f);
3374 if (r < 0)
3375 goto fail;
3376 }
3377 #endif
3378
3379 if (f->writable) {
3380 if (metrics) {
3381 journal_default_metrics(metrics, f->fd);
3382 f->metrics = *metrics;
3383 } else if (template)
3384 f->metrics = template->metrics;
3385
3386 r = journal_file_refresh_header(f);
3387 if (r < 0)
3388 goto fail;
3389 }
3390
3391 #if HAVE_GCRYPT
3392 r = journal_file_hmac_setup(f);
3393 if (r < 0)
3394 goto fail;
3395 #endif
3396
3397 if (newly_created) {
3398 r = journal_file_setup_field_hash_table(f);
3399 if (r < 0)
3400 goto fail;
3401
3402 r = journal_file_setup_data_hash_table(f);
3403 if (r < 0)
3404 goto fail;
3405
3406 #if HAVE_GCRYPT
3407 r = journal_file_append_first_tag(f);
3408 if (r < 0)
3409 goto fail;
3410 #endif
3411 }
3412
3413 if (mmap_cache_got_sigbus(f->mmap, f->cache_fd)) {
3414 r = -EIO;
3415 goto fail;
3416 }
3417
3418 if (template && template->post_change_timer) {
3419 r = journal_file_enable_post_change_timer(
3420 f,
3421 sd_event_source_get_event(template->post_change_timer),
3422 template->post_change_timer_period);
3423
3424 if (r < 0)
3425 goto fail;
3426 }
3427
3428 /* The file is opened now successfully, thus we take possession of any passed in fd. */
3429 f->close_fd = true;
3430
3431 *ret = f;
3432 return 0;
3433
3434 fail:
3435 if (f->cache_fd && mmap_cache_got_sigbus(f->mmap, f->cache_fd))
3436 r = -EIO;
3437
3438 (void) journal_file_close(f);
3439
3440 return r;
3441 }
3442
3443 int journal_file_rotate(JournalFile **f, bool compress, uint64_t compress_threshold_bytes, bool seal, Set *deferred_closes) {
3444 _cleanup_free_ char *p = NULL;
3445 size_t l;
3446 JournalFile *old_file, *new_file = NULL;
3447 int r;
3448
3449 assert(f);
3450 assert(*f);
3451
3452 old_file = *f;
3453
3454 if (!old_file->writable)
3455 return -EINVAL;
3456
3457 /* Is this a journal file that was passed to us as fd? If so, we synthesized a path name for it, and we refuse
3458 * rotation, since we don't know the actual path, and couldn't rename the file hence. */
3459 if (path_startswith(old_file->path, "/proc/self/fd"))
3460 return -EINVAL;
3461
3462 if (!endswith(old_file->path, ".journal"))
3463 return -EINVAL;
3464
3465 l = strlen(old_file->path);
3466 r = asprintf(&p, "%.*s@" SD_ID128_FORMAT_STR "-%016"PRIx64"-%016"PRIx64".journal",
3467 (int) l - 8, old_file->path,
3468 SD_ID128_FORMAT_VAL(old_file->header->seqnum_id),
3469 le64toh((*f)->header->head_entry_seqnum),
3470 le64toh((*f)->header->head_entry_realtime));
3471 if (r < 0)
3472 return -ENOMEM;
3473
3474 /* Try to rename the file to the archived version. If the file
3475 * already was deleted, we'll get ENOENT, let's ignore that
3476 * case. */
3477 r = rename(old_file->path, p);
3478 if (r < 0 && errno != ENOENT)
3479 return -errno;
3480
3481 /* Sync the rename to disk */
3482 (void) fsync_directory_of_file(old_file->fd);
3483
3484 /* Set as archive so offlining commits w/state=STATE_ARCHIVED.
3485 * Previously we would set old_file->header->state to STATE_ARCHIVED directly here,
3486 * but journal_file_set_offline() short-circuits when state != STATE_ONLINE, which
3487 * would result in the rotated journal never getting fsync() called before closing.
3488 * Now we simply queue the archive state by setting an archive bit, leaving the state
3489 * as STATE_ONLINE so proper offlining occurs. */
3490 old_file->archive = true;
3491
3492 /* Currently, btrfs is not very good with out write patterns
3493 * and fragments heavily. Let's defrag our journal files when
3494 * we archive them */
3495 old_file->defrag_on_close = true;
3496
3497 r = journal_file_open(-1, old_file->path, old_file->flags, old_file->mode, compress,
3498 compress_threshold_bytes, seal, NULL, old_file->mmap, deferred_closes,
3499 old_file, &new_file);
3500
3501 if (deferred_closes &&
3502 set_put(deferred_closes, old_file) >= 0)
3503 (void) journal_file_set_offline(old_file, false);
3504 else
3505 (void) journal_file_close(old_file);
3506
3507 *f = new_file;
3508 return r;
3509 }
3510
3511 int journal_file_open_reliably(
3512 const char *fname,
3513 int flags,
3514 mode_t mode,
3515 bool compress,
3516 uint64_t compress_threshold_bytes,
3517 bool seal,
3518 JournalMetrics *metrics,
3519 MMapCache *mmap_cache,
3520 Set *deferred_closes,
3521 JournalFile *template,
3522 JournalFile **ret) {
3523
3524 int r;
3525 size_t l;
3526 _cleanup_free_ char *p = NULL;
3527
3528 r = journal_file_open(-1, fname, flags, mode, compress, compress_threshold_bytes, seal, metrics, mmap_cache,
3529 deferred_closes, template, ret);
3530 if (!IN_SET(r,
3531 -EBADMSG, /* Corrupted */
3532 -ENODATA, /* Truncated */
3533 -EHOSTDOWN, /* Other machine */
3534 -EPROTONOSUPPORT, /* Incompatible feature */
3535 -EBUSY, /* Unclean shutdown */
3536 -ESHUTDOWN, /* Already archived */
3537 -EIO, /* IO error, including SIGBUS on mmap */
3538 -EIDRM, /* File has been deleted */
3539 -ETXTBSY)) /* File is from the future */
3540 return r;
3541
3542 if ((flags & O_ACCMODE) == O_RDONLY)
3543 return r;
3544
3545 if (!(flags & O_CREAT))
3546 return r;
3547
3548 if (!endswith(fname, ".journal"))
3549 return r;
3550
3551 /* The file is corrupted. Rotate it away and try it again (but only once) */
3552
3553 l = strlen(fname);
3554 if (asprintf(&p, "%.*s@%016"PRIx64 "-%016"PRIx64 ".journal~",
3555 (int) l - 8, fname,
3556 now(CLOCK_REALTIME),
3557 random_u64()) < 0)
3558 return -ENOMEM;
3559
3560 if (rename(fname, p) < 0)
3561 return -errno;
3562
3563 /* btrfs doesn't cope well with our write pattern and
3564 * fragments heavily. Let's defrag all files we rotate */
3565
3566 (void) chattr_path(p, 0, FS_NOCOW_FL);
3567 (void) btrfs_defrag(p);
3568
3569 log_warning_errno(r, "File %s corrupted or uncleanly shut down, renaming and replacing.", fname);
3570
3571 return journal_file_open(-1, fname, flags, mode, compress, compress_threshold_bytes, seal, metrics, mmap_cache,
3572 deferred_closes, template, ret);
3573 }
3574
3575 int journal_file_copy_entry(JournalFile *from, JournalFile *to, Object *o, uint64_t p) {
3576 uint64_t i, n;
3577 uint64_t q, xor_hash = 0;
3578 int r;
3579 EntryItem *items;
3580 dual_timestamp ts;
3581 const sd_id128_t *boot_id;
3582
3583 assert(from);
3584 assert(to);
3585 assert(o);
3586 assert(p);
3587
3588 if (!to->writable)
3589 return -EPERM;
3590
3591 ts.monotonic = le64toh(o->entry.monotonic);
3592 ts.realtime = le64toh(o->entry.realtime);
3593 boot_id = &o->entry.boot_id;
3594
3595 n = journal_file_entry_n_items(o);
3596 /* alloca() can't take 0, hence let's allocate at least one */
3597 items = newa(EntryItem, MAX(1u, n));
3598
3599 for (i = 0; i < n; i++) {
3600 uint64_t l, h;
3601 le64_t le_hash;
3602 size_t t;
3603 void *data;
3604 Object *u;
3605
3606 q = le64toh(o->entry.items[i].object_offset);
3607 le_hash = o->entry.items[i].hash;
3608
3609 r = journal_file_move_to_object(from, OBJECT_DATA, q, &o);
3610 if (r < 0)
3611 return r;
3612
3613 if (le_hash != o->data.hash)
3614 return -EBADMSG;
3615
3616 l = le64toh(o->object.size) - offsetof(Object, data.payload);
3617 t = (size_t) l;
3618
3619 /* We hit the limit on 32bit machines */
3620 if ((uint64_t) t != l)
3621 return -E2BIG;
3622
3623 if (o->object.flags & OBJECT_COMPRESSION_MASK) {
3624 #if HAVE_XZ || HAVE_LZ4
3625 size_t rsize = 0;
3626
3627 r = decompress_blob(o->object.flags & OBJECT_COMPRESSION_MASK,
3628 o->data.payload, l, &from->compress_buffer, &from->compress_buffer_size, &rsize, 0);
3629 if (r < 0)
3630 return r;
3631
3632 data = from->compress_buffer;
3633 l = rsize;
3634 #else
3635 return -EPROTONOSUPPORT;
3636 #endif
3637 } else
3638 data = o->data.payload;
3639
3640 r = journal_file_append_data(to, data, l, &u, &h);
3641 if (r < 0)
3642 return r;
3643
3644 xor_hash ^= le64toh(u->data.hash);
3645 items[i].object_offset = htole64(h);
3646 items[i].hash = u->data.hash;
3647
3648 r = journal_file_move_to_object(from, OBJECT_ENTRY, p, &o);
3649 if (r < 0)
3650 return r;
3651 }
3652
3653 r = journal_file_append_entry_internal(to, &ts, boot_id, xor_hash, items, n,
3654 NULL, NULL, NULL);
3655
3656 if (mmap_cache_got_sigbus(to->mmap, to->cache_fd))
3657 return -EIO;
3658
3659 return r;
3660 }
3661
3662 void journal_reset_metrics(JournalMetrics *m) {
3663 assert(m);
3664
3665 /* Set everything to "pick automatic values". */
3666
3667 *m = (JournalMetrics) {
3668 .min_use = (uint64_t) -1,
3669 .max_use = (uint64_t) -1,
3670 .min_size = (uint64_t) -1,
3671 .max_size = (uint64_t) -1,
3672 .keep_free = (uint64_t) -1,
3673 .n_max_files = (uint64_t) -1,
3674 };
3675 }
3676
3677 void journal_default_metrics(JournalMetrics *m, int fd) {
3678 char a[FORMAT_BYTES_MAX], b[FORMAT_BYTES_MAX], c[FORMAT_BYTES_MAX], d[FORMAT_BYTES_MAX], e[FORMAT_BYTES_MAX];
3679 struct statvfs ss;
3680 uint64_t fs_size;
3681
3682 assert(m);
3683 assert(fd >= 0);
3684
3685 if (fstatvfs(fd, &ss) >= 0)
3686 fs_size = ss.f_frsize * ss.f_blocks;
3687 else {
3688 log_debug_errno(errno, "Failed to determine disk size: %m");
3689 fs_size = 0;
3690 }
3691
3692 if (m->max_use == (uint64_t) -1) {
3693
3694 if (fs_size > 0) {
3695 m->max_use = PAGE_ALIGN(fs_size / 10); /* 10% of file system size */
3696
3697 if (m->max_use > DEFAULT_MAX_USE_UPPER)
3698 m->max_use = DEFAULT_MAX_USE_UPPER;
3699
3700 if (m->max_use < DEFAULT_MAX_USE_LOWER)
3701 m->max_use = DEFAULT_MAX_USE_LOWER;
3702 } else
3703 m->max_use = DEFAULT_MAX_USE_LOWER;
3704 } else {
3705 m->max_use = PAGE_ALIGN(m->max_use);
3706
3707 if (m->max_use != 0 && m->max_use < JOURNAL_FILE_SIZE_MIN*2)
3708 m->max_use = JOURNAL_FILE_SIZE_MIN*2;
3709 }
3710
3711 if (m->min_use == (uint64_t) -1)
3712 m->min_use = DEFAULT_MIN_USE;
3713
3714 if (m->min_use > m->max_use)
3715 m->min_use = m->max_use;
3716
3717 if (m->max_size == (uint64_t) -1) {
3718 m->max_size = PAGE_ALIGN(m->max_use / 8); /* 8 chunks */
3719
3720 if (m->max_size > DEFAULT_MAX_SIZE_UPPER)
3721 m->max_size = DEFAULT_MAX_SIZE_UPPER;
3722 } else
3723 m->max_size = PAGE_ALIGN(m->max_size);
3724
3725 if (m->max_size != 0) {
3726 if (m->max_size < JOURNAL_FILE_SIZE_MIN)
3727 m->max_size = JOURNAL_FILE_SIZE_MIN;
3728
3729 if (m->max_use != 0 && m->max_size*2 > m->max_use)
3730 m->max_use = m->max_size*2;
3731 }
3732
3733 if (m->min_size == (uint64_t) -1)
3734 m->min_size = JOURNAL_FILE_SIZE_MIN;
3735 else {
3736 m->min_size = PAGE_ALIGN(m->min_size);
3737
3738 if (m->min_size < JOURNAL_FILE_SIZE_MIN)
3739 m->min_size = JOURNAL_FILE_SIZE_MIN;
3740
3741 if (m->max_size != 0 && m->min_size > m->max_size)
3742 m->max_size = m->min_size;
3743 }
3744
3745 if (m->keep_free == (uint64_t) -1) {
3746
3747 if (fs_size > 0) {
3748 m->keep_free = PAGE_ALIGN(fs_size * 3 / 20); /* 15% of file system size */
3749
3750 if (m->keep_free > DEFAULT_KEEP_FREE_UPPER)
3751 m->keep_free = DEFAULT_KEEP_FREE_UPPER;
3752
3753 } else
3754 m->keep_free = DEFAULT_KEEP_FREE;
3755 }
3756
3757 if (m->n_max_files == (uint64_t) -1)
3758 m->n_max_files = DEFAULT_N_MAX_FILES;
3759
3760 log_debug("Fixed min_use=%s max_use=%s max_size=%s min_size=%s keep_free=%s n_max_files=%" PRIu64,
3761 format_bytes(a, sizeof(a), m->min_use),
3762 format_bytes(b, sizeof(b), m->max_use),
3763 format_bytes(c, sizeof(c), m->max_size),
3764 format_bytes(d, sizeof(d), m->min_size),
3765 format_bytes(e, sizeof(e), m->keep_free),
3766 m->n_max_files);
3767 }
3768
3769 int journal_file_get_cutoff_realtime_usec(JournalFile *f, usec_t *from, usec_t *to) {
3770 assert(f);
3771 assert(f->header);
3772 assert(from || to);
3773
3774 if (from) {
3775 if (f->header->head_entry_realtime == 0)
3776 return -ENOENT;
3777
3778 *from = le64toh(f->header->head_entry_realtime);
3779 }
3780
3781 if (to) {
3782 if (f->header->tail_entry_realtime == 0)
3783 return -ENOENT;
3784
3785 *to = le64toh(f->header->tail_entry_realtime);
3786 }
3787
3788 return 1;
3789 }
3790
3791 int journal_file_get_cutoff_monotonic_usec(JournalFile *f, sd_id128_t boot_id, usec_t *from, usec_t *to) {
3792 Object *o;
3793 uint64_t p;
3794 int r;
3795
3796 assert(f);
3797 assert(from || to);
3798
3799 r = find_data_object_by_boot_id(f, boot_id, &o, &p);
3800 if (r <= 0)
3801 return r;
3802
3803 if (le64toh(o->data.n_entries) <= 0)
3804 return 0;
3805
3806 if (from) {
3807 r = journal_file_move_to_object(f, OBJECT_ENTRY, le64toh(o->data.entry_offset), &o);
3808 if (r < 0)
3809 return r;
3810
3811 *from = le64toh(o->entry.monotonic);
3812 }
3813
3814 if (to) {
3815 r = journal_file_move_to_object(f, OBJECT_DATA, p, &o);
3816 if (r < 0)
3817 return r;
3818
3819 r = generic_array_get_plus_one(f,
3820 le64toh(o->data.entry_offset),
3821 le64toh(o->data.entry_array_offset),
3822 le64toh(o->data.n_entries)-1,
3823 &o, NULL);
3824 if (r <= 0)
3825 return r;
3826
3827 *to = le64toh(o->entry.monotonic);
3828 }
3829
3830 return 1;
3831 }
3832
3833 bool journal_file_rotate_suggested(JournalFile *f, usec_t max_file_usec) {
3834 assert(f);
3835 assert(f->header);
3836
3837 /* If we gained new header fields we gained new features,
3838 * hence suggest a rotation */
3839 if (le64toh(f->header->header_size) < sizeof(Header)) {
3840 log_debug("%s uses an outdated header, suggesting rotation.", f->path);
3841 return true;
3842 }
3843
3844 /* Let's check if the hash tables grew over a certain fill
3845 * level (75%, borrowing this value from Java's hash table
3846 * implementation), and if so suggest a rotation. To calculate
3847 * the fill level we need the n_data field, which only exists
3848 * in newer versions. */
3849
3850 if (JOURNAL_HEADER_CONTAINS(f->header, n_data))
3851 if (le64toh(f->header->n_data) * 4ULL > (le64toh(f->header->data_hash_table_size) / sizeof(HashItem)) * 3ULL) {
3852 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.",
3853 f->path,
3854 100.0 * (double) le64toh(f->header->n_data) / ((double) (le64toh(f->header->data_hash_table_size) / sizeof(HashItem))),
3855 le64toh(f->header->n_data),
3856 le64toh(f->header->data_hash_table_size) / sizeof(HashItem),
3857 (unsigned long long) f->last_stat.st_size,
3858 f->last_stat.st_size / le64toh(f->header->n_data));
3859 return true;
3860 }
3861
3862 if (JOURNAL_HEADER_CONTAINS(f->header, n_fields))
3863 if (le64toh(f->header->n_fields) * 4ULL > (le64toh(f->header->field_hash_table_size) / sizeof(HashItem)) * 3ULL) {
3864 log_debug("Field hash table of %s has a fill level at %.1f (%"PRIu64" of %"PRIu64" items), suggesting rotation.",
3865 f->path,
3866 100.0 * (double) le64toh(f->header->n_fields) / ((double) (le64toh(f->header->field_hash_table_size) / sizeof(HashItem))),
3867 le64toh(f->header->n_fields),
3868 le64toh(f->header->field_hash_table_size) / sizeof(HashItem));
3869 return true;
3870 }
3871
3872 /* Are the data objects properly indexed by field objects? */
3873 if (JOURNAL_HEADER_CONTAINS(f->header, n_data) &&
3874 JOURNAL_HEADER_CONTAINS(f->header, n_fields) &&
3875 le64toh(f->header->n_data) > 0 &&
3876 le64toh(f->header->n_fields) == 0)
3877 return true;
3878
3879 if (max_file_usec > 0) {
3880 usec_t t, h;
3881
3882 h = le64toh(f->header->head_entry_realtime);
3883 t = now(CLOCK_REALTIME);
3884
3885 if (h > 0 && t > h + max_file_usec)
3886 return true;
3887 }
3888
3889 return false;
3890 }
Unnamed repository; edit this file 'description' to name the repository.