]> 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 #15840 from Werkov/mkosi-opensuse
[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 "sd-event.h"
14
15 #include "alloc-util.h"
16 #include "btrfs-util.h"
17 #include "chattr-util.h"
18 #include "compress.h"
19 #include "fd-util.h"
20 #include "format-util.h"
21 #include "fs-util.h"
22 #include "journal-authenticate.h"
23 #include "journal-def.h"
24 #include "journal-file.h"
25 #include "lookup3.h"
26 #include "memory-util.h"
27 #include "path-util.h"
28 #include "random-util.h"
29 #include "set.h"
30 #include "sort-util.h"
31 #include "stat-util.h"
32 #include "string-util.h"
33 #include "strv.h"
34 #include "xattr-util.h"
35
36 #define DEFAULT_DATA_HASH_TABLE_SIZE (2047ULL*sizeof(HashItem))
37 #define DEFAULT_FIELD_HASH_TABLE_SIZE (333ULL*sizeof(HashItem))
38
39 #define DEFAULT_COMPRESS_THRESHOLD (512ULL)
40 #define MIN_COMPRESS_THRESHOLD (8ULL)
41
42 /* This is the minimum journal file size */
43 #define JOURNAL_FILE_SIZE_MIN (512 * 1024ULL) /* 512 KiB */
44
45 /* These are the lower and upper bounds if we deduce the max_use value
46 * from the file system size */
47 #define MAX_USE_LOWER (1 * 1024 * 1024ULL) /* 1 MiB */
48 #define MAX_USE_UPPER (4 * 1024 * 1024 * 1024ULL) /* 4 GiB */
49
50 /* Those are the lower and upper bounds for the minimal use limit,
51 * i.e. how much we'll use even if keep_free suggests otherwise. */
52 #define MIN_USE_LOW (1 * 1024 * 1024ULL) /* 1 MiB */
53 #define MIN_USE_HIGH (16 * 1024 * 1024ULL) /* 16 MiB */
54
55 /* This is the upper bound if we deduce max_size from max_use */
56 #define MAX_SIZE_UPPER (128 * 1024 * 1024ULL) /* 128 MiB */
57
58 /* This is the upper bound if we deduce the keep_free value from the
59 * file system size */
60 #define KEEP_FREE_UPPER (4 * 1024 * 1024 * 1024ULL) /* 4 GiB */
61
62 /* This is the keep_free value when we can't determine the system
63 * size */
64 #define DEFAULT_KEEP_FREE (1024 * 1024ULL) /* 1 MB */
65
66 /* This is the default maximum number of journal files to keep around. */
67 #define DEFAULT_N_MAX_FILES 100
68
69 /* n_data was the first entry we added after the initial file format design */
70 #define HEADER_SIZE_MIN ALIGN64(offsetof(Header, n_data))
71
72 /* How many entries to keep in the entry array chain cache at max */
73 #define CHAIN_CACHE_MAX 20
74
75 /* How much to increase the journal file size at once each time we allocate something new. */
76 #define FILE_SIZE_INCREASE (8 * 1024 * 1024ULL) /* 8MB */
77
78 /* Reread fstat() of the file for detecting deletions at least this often */
79 #define LAST_STAT_REFRESH_USEC (5*USEC_PER_SEC)
80
81 /* The mmap context to use for the header we pick as one above the last defined typed */
82 #define CONTEXT_HEADER _OBJECT_TYPE_MAX
83
84 #ifdef __clang__
85 # pragma GCC diagnostic ignored "-Waddress-of-packed-member"
86 #endif
87
88 /* This may be called from a separate thread to prevent blocking the caller for the duration of fsync().
89 * As a result we use atomic operations on f->offline_state for inter-thread communications with
90 * journal_file_set_offline() and journal_file_set_online(). */
91 static void journal_file_set_offline_internal(JournalFile *f) {
92 assert(f);
93 assert(f->fd >= 0);
94 assert(f->header);
95
96 for (;;) {
97 switch (f->offline_state) {
98 case OFFLINE_CANCEL:
99 if (!__sync_bool_compare_and_swap(&f->offline_state, OFFLINE_CANCEL, OFFLINE_DONE))
100 continue;
101 return;
102
103 case OFFLINE_AGAIN_FROM_SYNCING:
104 if (!__sync_bool_compare_and_swap(&f->offline_state, OFFLINE_AGAIN_FROM_SYNCING, OFFLINE_SYNCING))
105 continue;
106 break;
107
108 case OFFLINE_AGAIN_FROM_OFFLINING:
109 if (!__sync_bool_compare_and_swap(&f->offline_state, OFFLINE_AGAIN_FROM_OFFLINING, OFFLINE_SYNCING))
110 continue;
111 break;
112
113 case OFFLINE_SYNCING:
114 (void) fsync(f->fd);
115
116 if (!__sync_bool_compare_and_swap(&f->offline_state, OFFLINE_SYNCING, OFFLINE_OFFLINING))
117 continue;
118
119 f->header->state = f->archive ? STATE_ARCHIVED : STATE_OFFLINE;
120 (void) fsync(f->fd);
121 break;
122
123 case OFFLINE_OFFLINING:
124 if (!__sync_bool_compare_and_swap(&f->offline_state, OFFLINE_OFFLINING, OFFLINE_DONE))
125 continue;
126 _fallthrough_;
127 case OFFLINE_DONE:
128 return;
129
130 case OFFLINE_JOINED:
131 log_debug("OFFLINE_JOINED unexpected offline state for journal_file_set_offline_internal()");
132 return;
133 }
134 }
135 }
136
137 static void * journal_file_set_offline_thread(void *arg) {
138 JournalFile *f = arg;
139
140 (void) pthread_setname_np(pthread_self(), "journal-offline");
141
142 journal_file_set_offline_internal(f);
143
144 return NULL;
145 }
146
147 static int journal_file_set_offline_thread_join(JournalFile *f) {
148 int r;
149
150 assert(f);
151
152 if (f->offline_state == OFFLINE_JOINED)
153 return 0;
154
155 r = pthread_join(f->offline_thread, NULL);
156 if (r)
157 return -r;
158
159 f->offline_state = OFFLINE_JOINED;
160
161 if (mmap_cache_got_sigbus(f->mmap, f->cache_fd))
162 return -EIO;
163
164 return 0;
165 }
166
167 /* Trigger a restart if the offline thread is mid-flight in a restartable state. */
168 static bool journal_file_set_offline_try_restart(JournalFile *f) {
169 for (;;) {
170 switch (f->offline_state) {
171 case OFFLINE_AGAIN_FROM_SYNCING:
172 case OFFLINE_AGAIN_FROM_OFFLINING:
173 return true;
174
175 case OFFLINE_CANCEL:
176 if (!__sync_bool_compare_and_swap(&f->offline_state, OFFLINE_CANCEL, OFFLINE_AGAIN_FROM_SYNCING))
177 continue;
178 return true;
179
180 case OFFLINE_SYNCING:
181 if (!__sync_bool_compare_and_swap(&f->offline_state, OFFLINE_SYNCING, OFFLINE_AGAIN_FROM_SYNCING))
182 continue;
183 return true;
184
185 case OFFLINE_OFFLINING:
186 if (!__sync_bool_compare_and_swap(&f->offline_state, OFFLINE_OFFLINING, OFFLINE_AGAIN_FROM_OFFLINING))
187 continue;
188 return true;
189
190 default:
191 return false;
192 }
193 }
194 }
195
196 /* Sets a journal offline.
197 *
198 * If wait is false then an offline is dispatched in a separate thread for a
199 * subsequent journal_file_set_offline() or journal_file_set_online() of the
200 * same journal to synchronize with.
201 *
202 * If wait is true, then either an existing offline thread will be restarted
203 * and joined, or if none exists the offline is simply performed in this
204 * context without involving another thread.
205 */
206 int journal_file_set_offline(JournalFile *f, bool wait) {
207 bool restarted;
208 int r;
209
210 assert(f);
211
212 if (!f->writable)
213 return -EPERM;
214
215 if (f->fd < 0 || !f->header)
216 return -EINVAL;
217
218 /* An offlining journal is implicitly online and may modify f->header->state,
219 * we must also join any potentially lingering offline thread when not online. */
220 if (!journal_file_is_offlining(f) && f->header->state != STATE_ONLINE)
221 return journal_file_set_offline_thread_join(f);
222
223 /* Restart an in-flight offline thread and wait if needed, or join a lingering done one. */
224 restarted = journal_file_set_offline_try_restart(f);
225 if ((restarted && wait) || !restarted) {
226 r = journal_file_set_offline_thread_join(f);
227 if (r < 0)
228 return r;
229 }
230
231 if (restarted)
232 return 0;
233
234 /* Initiate a new offline. */
235 f->offline_state = OFFLINE_SYNCING;
236
237 if (wait) /* Without using a thread if waiting. */
238 journal_file_set_offline_internal(f);
239 else {
240 sigset_t ss, saved_ss;
241 int k;
242
243 assert_se(sigfillset(&ss) >= 0);
244 /* Don't block SIGBUS since the offlining thread accesses a memory mapped file.
245 * Asynchronous SIGBUS signals can safely be handled by either thread. */
246 assert_se(sigdelset(&ss, SIGBUS) >= 0);
247
248 r = pthread_sigmask(SIG_BLOCK, &ss, &saved_ss);
249 if (r > 0)
250 return -r;
251
252 r = pthread_create(&f->offline_thread, NULL, journal_file_set_offline_thread, f);
253
254 k = pthread_sigmask(SIG_SETMASK, &saved_ss, NULL);
255 if (r > 0) {
256 f->offline_state = OFFLINE_JOINED;
257 return -r;
258 }
259 if (k > 0)
260 return -k;
261 }
262
263 return 0;
264 }
265
266 static int journal_file_set_online(JournalFile *f) {
267 bool wait = true;
268
269 assert(f);
270
271 if (!f->writable)
272 return -EPERM;
273
274 if (f->fd < 0 || !f->header)
275 return -EINVAL;
276
277 while (wait) {
278 switch (f->offline_state) {
279 case OFFLINE_JOINED:
280 /* No offline thread, no need to wait. */
281 wait = false;
282 break;
283
284 case OFFLINE_SYNCING:
285 if (!__sync_bool_compare_and_swap(&f->offline_state, OFFLINE_SYNCING, OFFLINE_CANCEL))
286 continue;
287 /* Canceled syncing prior to offlining, no need to wait. */
288 wait = false;
289 break;
290
291 case OFFLINE_AGAIN_FROM_SYNCING:
292 if (!__sync_bool_compare_and_swap(&f->offline_state, OFFLINE_AGAIN_FROM_SYNCING, OFFLINE_CANCEL))
293 continue;
294 /* Canceled restart from syncing, no need to wait. */
295 wait = false;
296 break;
297
298 case OFFLINE_AGAIN_FROM_OFFLINING:
299 if (!__sync_bool_compare_and_swap(&f->offline_state, OFFLINE_AGAIN_FROM_OFFLINING, OFFLINE_CANCEL))
300 continue;
301 /* Canceled restart from offlining, must wait for offlining to complete however. */
302 _fallthrough_;
303 default: {
304 int r;
305
306 r = journal_file_set_offline_thread_join(f);
307 if (r < 0)
308 return r;
309
310 wait = false;
311 break;
312 }
313 }
314 }
315
316 if (mmap_cache_got_sigbus(f->mmap, f->cache_fd))
317 return -EIO;
318
319 switch (f->header->state) {
320 case STATE_ONLINE:
321 return 0;
322
323 case STATE_OFFLINE:
324 f->header->state = STATE_ONLINE;
325 (void) fsync(f->fd);
326 return 0;
327
328 default:
329 return -EINVAL;
330 }
331 }
332
333 bool journal_file_is_offlining(JournalFile *f) {
334 assert(f);
335
336 __sync_synchronize();
337
338 if (IN_SET(f->offline_state, OFFLINE_DONE, OFFLINE_JOINED))
339 return false;
340
341 return true;
342 }
343
344 JournalFile* journal_file_close(JournalFile *f) {
345 if (!f)
346 return NULL;
347
348 #if HAVE_GCRYPT
349 /* Write the final tag */
350 if (f->seal && f->writable) {
351 int r;
352
353 r = journal_file_append_tag(f);
354 if (r < 0)
355 log_error_errno(r, "Failed to append tag when closing journal: %m");
356 }
357 #endif
358
359 if (f->post_change_timer) {
360 if (sd_event_source_get_enabled(f->post_change_timer, NULL) > 0)
361 journal_file_post_change(f);
362
363 sd_event_source_disable_unref(f->post_change_timer);
364 }
365
366 journal_file_set_offline(f, true);
367
368 if (f->mmap && f->cache_fd)
369 mmap_cache_free_fd(f->mmap, f->cache_fd);
370
371 if (f->fd >= 0 && f->defrag_on_close) {
372
373 /* Be friendly to btrfs: turn COW back on again now,
374 * and defragment the file. We won't write to the file
375 * ever again, hence remove all fragmentation, and
376 * reenable all the good bits COW usually provides
377 * (such as data checksumming). */
378
379 (void) chattr_fd(f->fd, 0, FS_NOCOW_FL, NULL);
380 (void) btrfs_defrag_fd(f->fd);
381 }
382
383 if (f->close_fd)
384 safe_close(f->fd);
385 free(f->path);
386
387 mmap_cache_unref(f->mmap);
388
389 ordered_hashmap_free_free(f->chain_cache);
390
391 #if HAVE_XZ || HAVE_LZ4
392 free(f->compress_buffer);
393 #endif
394
395 #if HAVE_GCRYPT
396 if (f->fss_file)
397 munmap(f->fss_file, PAGE_ALIGN(f->fss_file_size));
398 else
399 free(f->fsprg_state);
400
401 free(f->fsprg_seed);
402
403 if (f->hmac)
404 gcry_md_close(f->hmac);
405 #endif
406
407 return mfree(f);
408 }
409
410 static int journal_file_init_header(JournalFile *f, JournalFile *template) {
411 Header h = {};
412 ssize_t k;
413 int r;
414
415 assert(f);
416
417 memcpy(h.signature, HEADER_SIGNATURE, 8);
418 h.header_size = htole64(ALIGN64(sizeof(h)));
419
420 h.incompatible_flags |= htole32(
421 f->compress_xz * HEADER_INCOMPATIBLE_COMPRESSED_XZ |
422 f->compress_lz4 * HEADER_INCOMPATIBLE_COMPRESSED_LZ4);
423
424 h.compatible_flags = htole32(
425 f->seal * HEADER_COMPATIBLE_SEALED);
426
427 r = sd_id128_randomize(&h.file_id);
428 if (r < 0)
429 return r;
430
431 if (template) {
432 h.seqnum_id = template->header->seqnum_id;
433 h.tail_entry_seqnum = template->header->tail_entry_seqnum;
434 } else
435 h.seqnum_id = h.file_id;
436
437 k = pwrite(f->fd, &h, sizeof(h), 0);
438 if (k < 0)
439 return -errno;
440
441 if (k != sizeof(h))
442 return -EIO;
443
444 return 0;
445 }
446
447 static int journal_file_refresh_header(JournalFile *f) {
448 sd_id128_t boot_id;
449 int r;
450
451 assert(f);
452 assert(f->header);
453
454 r = sd_id128_get_machine(&f->header->machine_id);
455 if (IN_SET(r, -ENOENT, -ENOMEDIUM))
456 /* We don't have a machine-id, let's continue without */
457 zero(f->header->machine_id);
458 else if (r < 0)
459 return r;
460
461 r = sd_id128_get_boot(&boot_id);
462 if (r < 0)
463 return r;
464
465 f->header->boot_id = boot_id;
466
467 r = journal_file_set_online(f);
468
469 /* Sync the online state to disk */
470 (void) fsync(f->fd);
471
472 /* We likely just created a new file, also sync the directory this file is located in. */
473 (void) fsync_directory_of_file(f->fd);
474
475 return r;
476 }
477
478 static bool warn_wrong_flags(const JournalFile *f, bool compatible) {
479 const uint32_t any = compatible ? HEADER_COMPATIBLE_ANY : HEADER_INCOMPATIBLE_ANY,
480 supported = compatible ? HEADER_COMPATIBLE_SUPPORTED : HEADER_INCOMPATIBLE_SUPPORTED;
481 const char *type = compatible ? "compatible" : "incompatible";
482 uint32_t flags;
483
484 flags = le32toh(compatible ? f->header->compatible_flags : f->header->incompatible_flags);
485
486 if (flags & ~supported) {
487 if (flags & ~any)
488 log_debug("Journal file %s has unknown %s flags 0x%"PRIx32,
489 f->path, type, flags & ~any);
490 flags = (flags & any) & ~supported;
491 if (flags) {
492 const char* strv[3];
493 unsigned n = 0;
494 _cleanup_free_ char *t = NULL;
495
496 if (compatible && (flags & HEADER_COMPATIBLE_SEALED))
497 strv[n++] = "sealed";
498 if (!compatible && (flags & HEADER_INCOMPATIBLE_COMPRESSED_XZ))
499 strv[n++] = "xz-compressed";
500 if (!compatible && (flags & HEADER_INCOMPATIBLE_COMPRESSED_LZ4))
501 strv[n++] = "lz4-compressed";
502 strv[n] = NULL;
503 assert(n < ELEMENTSOF(strv));
504
505 t = strv_join((char**) strv, ", ");
506 log_debug("Journal file %s uses %s %s %s disabled at compilation time.",
507 f->path, type, n > 1 ? "flags" : "flag", strnull(t));
508 }
509 return true;
510 }
511
512 return false;
513 }
514
515 static int journal_file_verify_header(JournalFile *f) {
516 uint64_t arena_size, header_size;
517
518 assert(f);
519 assert(f->header);
520
521 if (memcmp(f->header->signature, HEADER_SIGNATURE, 8))
522 return -EBADMSG;
523
524 /* In both read and write mode we refuse to open files with incompatible
525 * flags we don't know. */
526 if (warn_wrong_flags(f, false))
527 return -EPROTONOSUPPORT;
528
529 /* When open for writing we refuse to open files with compatible flags, too. */
530 if (f->writable && warn_wrong_flags(f, true))
531 return -EPROTONOSUPPORT;
532
533 if (f->header->state >= _STATE_MAX)
534 return -EBADMSG;
535
536 header_size = le64toh(READ_NOW(f->header->header_size));
537
538 /* The first addition was n_data, so check that we are at least this large */
539 if (header_size < HEADER_SIZE_MIN)
540 return -EBADMSG;
541
542 if (JOURNAL_HEADER_SEALED(f->header) && !JOURNAL_HEADER_CONTAINS(f->header, n_entry_arrays))
543 return -EBADMSG;
544
545 arena_size = le64toh(READ_NOW(f->header->arena_size));
546
547 if (UINT64_MAX - header_size < arena_size || header_size + arena_size > (uint64_t) f->last_stat.st_size)
548 return -ENODATA;
549
550 if (le64toh(f->header->tail_object_offset) > header_size + arena_size)
551 return -ENODATA;
552
553 if (!VALID64(le64toh(f->header->data_hash_table_offset)) ||
554 !VALID64(le64toh(f->header->field_hash_table_offset)) ||
555 !VALID64(le64toh(f->header->tail_object_offset)) ||
556 !VALID64(le64toh(f->header->entry_array_offset)))
557 return -ENODATA;
558
559 if (f->writable) {
560 sd_id128_t machine_id;
561 uint8_t state;
562 int r;
563
564 r = sd_id128_get_machine(&machine_id);
565 if (r < 0)
566 return r;
567
568 if (!sd_id128_equal(machine_id, f->header->machine_id))
569 return -EHOSTDOWN;
570
571 state = f->header->state;
572
573 if (state == STATE_ARCHIVED)
574 return -ESHUTDOWN; /* Already archived */
575 else if (state == STATE_ONLINE)
576 return log_debug_errno(SYNTHETIC_ERRNO(EBUSY),
577 "Journal file %s is already online. Assuming unclean closing.",
578 f->path);
579 else if (state != STATE_OFFLINE)
580 return log_debug_errno(SYNTHETIC_ERRNO(EBUSY),
581 "Journal file %s has unknown state %i.",
582 f->path, state);
583
584 if (f->header->field_hash_table_size == 0 || f->header->data_hash_table_size == 0)
585 return -EBADMSG;
586
587 /* Don't permit appending to files from the future. Because otherwise the realtime timestamps wouldn't
588 * be strictly ordered in the entries in the file anymore, and we can't have that since it breaks
589 * bisection. */
590 if (le64toh(f->header->tail_entry_realtime) > now(CLOCK_REALTIME))
591 return log_debug_errno(SYNTHETIC_ERRNO(ETXTBSY),
592 "Journal file %s is from the future, refusing to append new data to it that'd be older.",
593 f->path);
594 }
595
596 f->compress_xz = JOURNAL_HEADER_COMPRESSED_XZ(f->header);
597 f->compress_lz4 = JOURNAL_HEADER_COMPRESSED_LZ4(f->header);
598
599 f->seal = JOURNAL_HEADER_SEALED(f->header);
600
601 return 0;
602 }
603
604 int journal_file_fstat(JournalFile *f) {
605 int r;
606
607 assert(f);
608 assert(f->fd >= 0);
609
610 if (fstat(f->fd, &f->last_stat) < 0)
611 return -errno;
612
613 f->last_stat_usec = now(CLOCK_MONOTONIC);
614
615 /* Refuse dealing with with files that aren't regular */
616 r = stat_verify_regular(&f->last_stat);
617 if (r < 0)
618 return r;
619
620 /* Refuse appending to files that are already deleted */
621 if (f->last_stat.st_nlink <= 0)
622 return -EIDRM;
623
624 return 0;
625 }
626
627 static int journal_file_allocate(JournalFile *f, uint64_t offset, uint64_t size) {
628 uint64_t old_size, new_size, old_header_size, old_arena_size;
629 int r;
630
631 assert(f);
632 assert(f->header);
633
634 /* We assume that this file is not sparse, and we know that for sure, since we always call
635 * posix_fallocate() ourselves */
636
637 if (size > PAGE_ALIGN_DOWN(UINT64_MAX) - offset)
638 return -EINVAL;
639
640 if (mmap_cache_got_sigbus(f->mmap, f->cache_fd))
641 return -EIO;
642
643 old_header_size = le64toh(READ_NOW(f->header->header_size));
644 old_arena_size = le64toh(READ_NOW(f->header->arena_size));
645 if (old_arena_size > PAGE_ALIGN_DOWN(UINT64_MAX) - old_header_size)
646 return -EBADMSG;
647
648 old_size = old_header_size + old_arena_size;
649
650 new_size = MAX(PAGE_ALIGN(offset + size), old_header_size);
651
652 if (new_size <= old_size) {
653
654 /* We already pre-allocated enough space, but before
655 * we write to it, let's check with fstat() if the
656 * file got deleted, in order make sure we don't throw
657 * away the data immediately. Don't check fstat() for
658 * all writes though, but only once ever 10s. */
659
660 if (f->last_stat_usec + LAST_STAT_REFRESH_USEC > now(CLOCK_MONOTONIC))
661 return 0;
662
663 return journal_file_fstat(f);
664 }
665
666 /* Allocate more space. */
667
668 if (f->metrics.max_size > 0 && new_size > f->metrics.max_size)
669 return -E2BIG;
670
671 if (new_size > f->metrics.min_size && f->metrics.keep_free > 0) {
672 struct statvfs svfs;
673
674 if (fstatvfs(f->fd, &svfs) >= 0) {
675 uint64_t available;
676
677 available = LESS_BY((uint64_t) svfs.f_bfree * (uint64_t) svfs.f_bsize, f->metrics.keep_free);
678
679 if (new_size - old_size > available)
680 return -E2BIG;
681 }
682 }
683
684 /* Increase by larger blocks at once */
685 new_size = DIV_ROUND_UP(new_size, FILE_SIZE_INCREASE) * FILE_SIZE_INCREASE;
686 if (f->metrics.max_size > 0 && new_size > f->metrics.max_size)
687 new_size = f->metrics.max_size;
688
689 /* Note that the glibc fallocate() fallback is very
690 inefficient, hence we try to minimize the allocation area
691 as we can. */
692 r = posix_fallocate(f->fd, old_size, new_size - old_size);
693 if (r != 0)
694 return -r;
695
696 f->header->arena_size = htole64(new_size - old_header_size);
697
698 return journal_file_fstat(f);
699 }
700
701 static unsigned type_to_context(ObjectType type) {
702 /* One context for each type, plus one catch-all for the rest */
703 assert_cc(_OBJECT_TYPE_MAX <= MMAP_CACHE_MAX_CONTEXTS);
704 assert_cc(CONTEXT_HEADER < MMAP_CACHE_MAX_CONTEXTS);
705 return type > OBJECT_UNUSED && type < _OBJECT_TYPE_MAX ? type : 0;
706 }
707
708 static int journal_file_move_to(
709 JournalFile *f,
710 ObjectType type,
711 bool keep_always,
712 uint64_t offset,
713 uint64_t size,
714 void **ret,
715 size_t *ret_size) {
716
717 int r;
718
719 assert(f);
720 assert(ret);
721
722 if (size <= 0)
723 return -EINVAL;
724
725 if (size > UINT64_MAX - offset)
726 return -EBADMSG;
727
728 /* Avoid SIGBUS on invalid accesses */
729 if (offset + size > (uint64_t) f->last_stat.st_size) {
730 /* Hmm, out of range? Let's refresh the fstat() data
731 * first, before we trust that check. */
732
733 r = journal_file_fstat(f);
734 if (r < 0)
735 return r;
736
737 if (offset + size > (uint64_t) f->last_stat.st_size)
738 return -EADDRNOTAVAIL;
739 }
740
741 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);
742 }
743
744 static uint64_t minimum_header_size(Object *o) {
745
746 static const uint64_t table[] = {
747 [OBJECT_DATA] = sizeof(DataObject),
748 [OBJECT_FIELD] = sizeof(FieldObject),
749 [OBJECT_ENTRY] = sizeof(EntryObject),
750 [OBJECT_DATA_HASH_TABLE] = sizeof(HashTableObject),
751 [OBJECT_FIELD_HASH_TABLE] = sizeof(HashTableObject),
752 [OBJECT_ENTRY_ARRAY] = sizeof(EntryArrayObject),
753 [OBJECT_TAG] = sizeof(TagObject),
754 };
755
756 if (o->object.type >= ELEMENTSOF(table) || table[o->object.type] <= 0)
757 return sizeof(ObjectHeader);
758
759 return table[o->object.type];
760 }
761
762 /* Lightweight object checks. We want this to be fast, so that we won't
763 * slowdown every journal_file_move_to_object() call too much. */
764 static int journal_file_check_object(JournalFile *f, uint64_t offset, Object *o) {
765 assert(f);
766 assert(o);
767
768 switch (o->object.type) {
769
770 case OBJECT_DATA:
771 if ((le64toh(o->data.entry_offset) == 0) ^ (le64toh(o->data.n_entries) == 0))
772 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG),
773 "Bad n_entries: %" PRIu64 ": %" PRIu64,
774 le64toh(o->data.n_entries),
775 offset);
776
777 if (le64toh(o->object.size) <= offsetof(DataObject, payload))
778 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG),
779 "Bad object size (<= %zu): %" PRIu64 ": %" PRIu64,
780 offsetof(DataObject, payload),
781 le64toh(o->object.size),
782 offset);
783
784 if (!VALID64(le64toh(o->data.next_hash_offset)) ||
785 !VALID64(le64toh(o->data.next_field_offset)) ||
786 !VALID64(le64toh(o->data.entry_offset)) ||
787 !VALID64(le64toh(o->data.entry_array_offset)))
788 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG),
789 "Invalid offset, next_hash_offset=" OFSfmt ", next_field_offset=" OFSfmt ", entry_offset=" OFSfmt ", entry_array_offset=" OFSfmt ": %" PRIu64,
790 le64toh(o->data.next_hash_offset),
791 le64toh(o->data.next_field_offset),
792 le64toh(o->data.entry_offset),
793 le64toh(o->data.entry_array_offset),
794 offset);
795
796 break;
797
798 case OBJECT_FIELD:
799 if (le64toh(o->object.size) <= offsetof(FieldObject, payload))
800 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG),
801 "Bad field size (<= %zu): %" PRIu64 ": %" PRIu64,
802 offsetof(FieldObject, payload),
803 le64toh(o->object.size),
804 offset);
805
806 if (!VALID64(le64toh(o->field.next_hash_offset)) ||
807 !VALID64(le64toh(o->field.head_data_offset)))
808 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG),
809 "Invalid offset, next_hash_offset=" OFSfmt ", head_data_offset=" OFSfmt ": %" PRIu64,
810 le64toh(o->field.next_hash_offset),
811 le64toh(o->field.head_data_offset),
812 offset);
813 break;
814
815 case OBJECT_ENTRY: {
816 uint64_t sz;
817
818 sz = le64toh(READ_NOW(o->object.size));
819 if (sz < offsetof(EntryObject, items) ||
820 (sz - offsetof(EntryObject, items)) % sizeof(EntryItem) != 0)
821 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG),
822 "Bad entry size (<= %zu): %" PRIu64 ": %" PRIu64,
823 offsetof(EntryObject, items),
824 sz,
825 offset);
826
827 if ((sz - offsetof(EntryObject, items)) / sizeof(EntryItem) <= 0)
828 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG),
829 "Invalid number items in entry: %" PRIu64 ": %" PRIu64,
830 (sz - offsetof(EntryObject, items)) / sizeof(EntryItem),
831 offset);
832
833 if (le64toh(o->entry.seqnum) <= 0)
834 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG),
835 "Invalid entry seqnum: %" PRIx64 ": %" PRIu64,
836 le64toh(o->entry.seqnum),
837 offset);
838
839 if (!VALID_REALTIME(le64toh(o->entry.realtime)))
840 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG),
841 "Invalid entry realtime timestamp: %" PRIu64 ": %" PRIu64,
842 le64toh(o->entry.realtime),
843 offset);
844
845 if (!VALID_MONOTONIC(le64toh(o->entry.monotonic)))
846 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG),
847 "Invalid entry monotonic timestamp: %" PRIu64 ": %" PRIu64,
848 le64toh(o->entry.monotonic),
849 offset);
850
851 break;
852 }
853
854 case OBJECT_DATA_HASH_TABLE:
855 case OBJECT_FIELD_HASH_TABLE: {
856 uint64_t sz;
857
858 sz = le64toh(READ_NOW(o->object.size));
859 if (sz < offsetof(HashTableObject, items) ||
860 (sz - offsetof(HashTableObject, items)) % sizeof(HashItem) != 0 ||
861 (sz - offsetof(HashTableObject, items)) / sizeof(HashItem) <= 0)
862 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG),
863 "Invalid %s hash table size: %" PRIu64 ": %" PRIu64,
864 o->object.type == OBJECT_DATA_HASH_TABLE ? "data" : "field",
865 sz,
866 offset);
867
868 break;
869 }
870
871 case OBJECT_ENTRY_ARRAY: {
872 uint64_t sz;
873
874 sz = le64toh(READ_NOW(o->object.size));
875 if (sz < offsetof(EntryArrayObject, items) ||
876 (sz - offsetof(EntryArrayObject, items)) % sizeof(le64_t) != 0 ||
877 (sz - offsetof(EntryArrayObject, items)) / sizeof(le64_t) <= 0)
878 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG),
879 "Invalid object entry array size: %" PRIu64 ": %" PRIu64,
880 sz,
881 offset);
882
883 if (!VALID64(le64toh(o->entry_array.next_entry_array_offset)))
884 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG),
885 "Invalid object entry array next_entry_array_offset: " OFSfmt ": %" PRIu64,
886 le64toh(o->entry_array.next_entry_array_offset),
887 offset);
888
889 break;
890 }
891
892 case OBJECT_TAG:
893 if (le64toh(o->object.size) != sizeof(TagObject))
894 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG),
895 "Invalid object tag size: %" PRIu64 ": %" PRIu64,
896 le64toh(o->object.size),
897 offset);
898
899 if (!VALID_EPOCH(le64toh(o->tag.epoch)))
900 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG),
901 "Invalid object tag epoch: %" PRIu64 ": %" PRIu64,
902 le64toh(o->tag.epoch), offset);
903
904 break;
905 }
906
907 return 0;
908 }
909
910 int journal_file_move_to_object(JournalFile *f, ObjectType type, uint64_t offset, Object **ret) {
911 int r;
912 void *t;
913 size_t tsize;
914 Object *o;
915 uint64_t s;
916
917 assert(f);
918 assert(ret);
919
920 /* Objects may only be located at multiple of 64 bit */
921 if (!VALID64(offset))
922 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG),
923 "Attempt to move to object at non-64bit boundary: %" PRIu64,
924 offset);
925
926 /* Object may not be located in the file header */
927 if (offset < le64toh(f->header->header_size))
928 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG),
929 "Attempt to move to object located in file header: %" PRIu64,
930 offset);
931
932 r = journal_file_move_to(f, type, false, offset, sizeof(ObjectHeader), &t, &tsize);
933 if (r < 0)
934 return r;
935
936 o = (Object*) t;
937 s = le64toh(READ_NOW(o->object.size));
938
939 if (s == 0)
940 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG),
941 "Attempt to move to uninitialized object: %" PRIu64,
942 offset);
943 if (s < sizeof(ObjectHeader))
944 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG),
945 "Attempt to move to overly short object: %" PRIu64,
946 offset);
947
948 if (o->object.type <= OBJECT_UNUSED)
949 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG),
950 "Attempt to move to object with invalid type: %" PRIu64,
951 offset);
952
953 if (s < minimum_header_size(o))
954 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG),
955 "Attempt to move to truncated object: %" PRIu64,
956 offset);
957
958 if (type > OBJECT_UNUSED && o->object.type != type)
959 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG),
960 "Attempt to move to object of unexpected type: %" PRIu64,
961 offset);
962
963 if (s > tsize) {
964 r = journal_file_move_to(f, type, false, offset, s, &t, NULL);
965 if (r < 0)
966 return r;
967
968 o = (Object*) t;
969 }
970
971 r = journal_file_check_object(f, offset, o);
972 if (r < 0)
973 return r;
974
975 *ret = o;
976 return 0;
977 }
978
979 static uint64_t journal_file_entry_seqnum(JournalFile *f, uint64_t *seqnum) {
980 uint64_t r;
981
982 assert(f);
983 assert(f->header);
984
985 r = le64toh(f->header->tail_entry_seqnum) + 1;
986
987 if (seqnum) {
988 /* If an external seqnum counter was passed, we update
989 * both the local and the external one, and set it to
990 * the maximum of both */
991
992 if (*seqnum + 1 > r)
993 r = *seqnum + 1;
994
995 *seqnum = r;
996 }
997
998 f->header->tail_entry_seqnum = htole64(r);
999
1000 if (f->header->head_entry_seqnum == 0)
1001 f->header->head_entry_seqnum = htole64(r);
1002
1003 return r;
1004 }
1005
1006 int journal_file_append_object(JournalFile *f, ObjectType type, uint64_t size, Object **ret, uint64_t *offset) {
1007 int r;
1008 uint64_t p;
1009 Object *tail, *o;
1010 void *t;
1011
1012 assert(f);
1013 assert(f->header);
1014 assert(type > OBJECT_UNUSED && type < _OBJECT_TYPE_MAX);
1015 assert(size >= sizeof(ObjectHeader));
1016 assert(offset);
1017 assert(ret);
1018
1019 r = journal_file_set_online(f);
1020 if (r < 0)
1021 return r;
1022
1023 p = le64toh(f->header->tail_object_offset);
1024 if (p == 0)
1025 p = le64toh(f->header->header_size);
1026 else {
1027 uint64_t sz;
1028
1029 r = journal_file_move_to_object(f, OBJECT_UNUSED, p, &tail);
1030 if (r < 0)
1031 return r;
1032
1033 sz = le64toh(READ_NOW(tail->object.size));
1034 if (sz > UINT64_MAX - sizeof(uint64_t) + 1)
1035 return -EBADMSG;
1036
1037 sz = ALIGN64(sz);
1038 if (p > UINT64_MAX - sz)
1039 return -EBADMSG;
1040
1041 p += sz;
1042 }
1043
1044 r = journal_file_allocate(f, p, size);
1045 if (r < 0)
1046 return r;
1047
1048 r = journal_file_move_to(f, type, false, p, size, &t, NULL);
1049 if (r < 0)
1050 return r;
1051
1052 o = (Object*) t;
1053 o->object = (ObjectHeader) {
1054 .type = type,
1055 .size = htole64(size),
1056 };
1057
1058 f->header->tail_object_offset = htole64(p);
1059 f->header->n_objects = htole64(le64toh(f->header->n_objects) + 1);
1060
1061 *ret = o;
1062 *offset = p;
1063
1064 return 0;
1065 }
1066
1067 static int journal_file_setup_data_hash_table(JournalFile *f) {
1068 uint64_t s, p;
1069 Object *o;
1070 int r;
1071
1072 assert(f);
1073 assert(f->header);
1074
1075 /* We estimate that we need 1 hash table entry per 768 bytes
1076 of journal file and we want to make sure we never get
1077 beyond 75% fill level. Calculate the hash table size for
1078 the maximum file size based on these metrics. */
1079
1080 s = (f->metrics.max_size * 4 / 768 / 3) * sizeof(HashItem);
1081 if (s < DEFAULT_DATA_HASH_TABLE_SIZE)
1082 s = DEFAULT_DATA_HASH_TABLE_SIZE;
1083
1084 log_debug("Reserving %"PRIu64" entries in hash table.", s / sizeof(HashItem));
1085
1086 r = journal_file_append_object(f,
1087 OBJECT_DATA_HASH_TABLE,
1088 offsetof(Object, hash_table.items) + s,
1089 &o, &p);
1090 if (r < 0)
1091 return r;
1092
1093 memzero(o->hash_table.items, s);
1094
1095 f->header->data_hash_table_offset = htole64(p + offsetof(Object, hash_table.items));
1096 f->header->data_hash_table_size = htole64(s);
1097
1098 return 0;
1099 }
1100
1101 static int journal_file_setup_field_hash_table(JournalFile *f) {
1102 uint64_t s, p;
1103 Object *o;
1104 int r;
1105
1106 assert(f);
1107 assert(f->header);
1108
1109 /* We use a fixed size hash table for the fields as this
1110 * number should grow very slowly only */
1111
1112 s = DEFAULT_FIELD_HASH_TABLE_SIZE;
1113 r = journal_file_append_object(f,
1114 OBJECT_FIELD_HASH_TABLE,
1115 offsetof(Object, hash_table.items) + s,
1116 &o, &p);
1117 if (r < 0)
1118 return r;
1119
1120 memzero(o->hash_table.items, s);
1121
1122 f->header->field_hash_table_offset = htole64(p + offsetof(Object, hash_table.items));
1123 f->header->field_hash_table_size = htole64(s);
1124
1125 return 0;
1126 }
1127
1128 int journal_file_map_data_hash_table(JournalFile *f) {
1129 uint64_t s, p;
1130 void *t;
1131 int r;
1132
1133 assert(f);
1134 assert(f->header);
1135
1136 if (f->data_hash_table)
1137 return 0;
1138
1139 p = le64toh(f->header->data_hash_table_offset);
1140 s = le64toh(f->header->data_hash_table_size);
1141
1142 r = journal_file_move_to(f,
1143 OBJECT_DATA_HASH_TABLE,
1144 true,
1145 p, s,
1146 &t, NULL);
1147 if (r < 0)
1148 return r;
1149
1150 f->data_hash_table = t;
1151 return 0;
1152 }
1153
1154 int journal_file_map_field_hash_table(JournalFile *f) {
1155 uint64_t s, p;
1156 void *t;
1157 int r;
1158
1159 assert(f);
1160 assert(f->header);
1161
1162 if (f->field_hash_table)
1163 return 0;
1164
1165 p = le64toh(f->header->field_hash_table_offset);
1166 s = le64toh(f->header->field_hash_table_size);
1167
1168 r = journal_file_move_to(f,
1169 OBJECT_FIELD_HASH_TABLE,
1170 true,
1171 p, s,
1172 &t, NULL);
1173 if (r < 0)
1174 return r;
1175
1176 f->field_hash_table = t;
1177 return 0;
1178 }
1179
1180 static int journal_file_link_field(
1181 JournalFile *f,
1182 Object *o,
1183 uint64_t offset,
1184 uint64_t hash) {
1185
1186 uint64_t p, h, m;
1187 int r;
1188
1189 assert(f);
1190 assert(f->header);
1191 assert(f->field_hash_table);
1192 assert(o);
1193 assert(offset > 0);
1194
1195 if (o->object.type != OBJECT_FIELD)
1196 return -EINVAL;
1197
1198 m = le64toh(READ_NOW(f->header->field_hash_table_size)) / sizeof(HashItem);
1199 if (m <= 0)
1200 return -EBADMSG;
1201
1202 /* This might alter the window we are looking at */
1203 o->field.next_hash_offset = o->field.head_data_offset = 0;
1204
1205 h = hash % m;
1206 p = le64toh(f->field_hash_table[h].tail_hash_offset);
1207 if (p == 0)
1208 f->field_hash_table[h].head_hash_offset = htole64(offset);
1209 else {
1210 r = journal_file_move_to_object(f, OBJECT_FIELD, p, &o);
1211 if (r < 0)
1212 return r;
1213
1214 o->field.next_hash_offset = htole64(offset);
1215 }
1216
1217 f->field_hash_table[h].tail_hash_offset = htole64(offset);
1218
1219 if (JOURNAL_HEADER_CONTAINS(f->header, n_fields))
1220 f->header->n_fields = htole64(le64toh(f->header->n_fields) + 1);
1221
1222 return 0;
1223 }
1224
1225 static int journal_file_link_data(
1226 JournalFile *f,
1227 Object *o,
1228 uint64_t offset,
1229 uint64_t hash) {
1230
1231 uint64_t p, h, m;
1232 int r;
1233
1234 assert(f);
1235 assert(f->header);
1236 assert(f->data_hash_table);
1237 assert(o);
1238 assert(offset > 0);
1239
1240 if (o->object.type != OBJECT_DATA)
1241 return -EINVAL;
1242
1243 m = le64toh(READ_NOW(f->header->data_hash_table_size)) / sizeof(HashItem);
1244 if (m <= 0)
1245 return -EBADMSG;
1246
1247 /* This might alter the window we are looking at */
1248 o->data.next_hash_offset = o->data.next_field_offset = 0;
1249 o->data.entry_offset = o->data.entry_array_offset = 0;
1250 o->data.n_entries = 0;
1251
1252 h = hash % m;
1253 p = le64toh(f->data_hash_table[h].tail_hash_offset);
1254 if (p == 0)
1255 /* Only entry in the hash table is easy */
1256 f->data_hash_table[h].head_hash_offset = htole64(offset);
1257 else {
1258 /* Move back to the previous data object, to patch in
1259 * pointer */
1260
1261 r = journal_file_move_to_object(f, OBJECT_DATA, p, &o);
1262 if (r < 0)
1263 return r;
1264
1265 o->data.next_hash_offset = htole64(offset);
1266 }
1267
1268 f->data_hash_table[h].tail_hash_offset = htole64(offset);
1269
1270 if (JOURNAL_HEADER_CONTAINS(f->header, n_data))
1271 f->header->n_data = htole64(le64toh(f->header->n_data) + 1);
1272
1273 return 0;
1274 }
1275
1276 int journal_file_find_field_object_with_hash(
1277 JournalFile *f,
1278 const void *field, uint64_t size, uint64_t hash,
1279 Object **ret, uint64_t *offset) {
1280
1281 uint64_t p, osize, h, m;
1282 int r;
1283
1284 assert(f);
1285 assert(f->header);
1286 assert(field && size > 0);
1287
1288 /* If the field hash table is empty, we can't find anything */
1289 if (le64toh(f->header->field_hash_table_size) <= 0)
1290 return 0;
1291
1292 /* Map the field hash table, if it isn't mapped yet. */
1293 r = journal_file_map_field_hash_table(f);
1294 if (r < 0)
1295 return r;
1296
1297 osize = offsetof(Object, field.payload) + size;
1298
1299 m = le64toh(READ_NOW(f->header->field_hash_table_size)) / sizeof(HashItem);
1300 if (m <= 0)
1301 return -EBADMSG;
1302
1303 h = hash % m;
1304 p = le64toh(f->field_hash_table[h].head_hash_offset);
1305
1306 while (p > 0) {
1307 Object *o;
1308
1309 r = journal_file_move_to_object(f, OBJECT_FIELD, p, &o);
1310 if (r < 0)
1311 return r;
1312
1313 if (le64toh(o->field.hash) == hash &&
1314 le64toh(o->object.size) == osize &&
1315 memcmp(o->field.payload, field, size) == 0) {
1316
1317 if (ret)
1318 *ret = o;
1319 if (offset)
1320 *offset = p;
1321
1322 return 1;
1323 }
1324
1325 p = le64toh(o->field.next_hash_offset);
1326 }
1327
1328 return 0;
1329 }
1330
1331 int journal_file_find_field_object(
1332 JournalFile *f,
1333 const void *field, uint64_t size,
1334 Object **ret, uint64_t *offset) {
1335
1336 uint64_t hash;
1337
1338 assert(f);
1339 assert(field && size > 0);
1340
1341 hash = hash64(field, size);
1342
1343 return journal_file_find_field_object_with_hash(f,
1344 field, size, hash,
1345 ret, offset);
1346 }
1347
1348 int journal_file_find_data_object_with_hash(
1349 JournalFile *f,
1350 const void *data, uint64_t size, uint64_t hash,
1351 Object **ret, uint64_t *offset) {
1352
1353 uint64_t p, osize, h, m;
1354 int r;
1355
1356 assert(f);
1357 assert(f->header);
1358 assert(data || size == 0);
1359
1360 /* If there's no data hash table, then there's no entry. */
1361 if (le64toh(f->header->data_hash_table_size) <= 0)
1362 return 0;
1363
1364 /* Map the data hash table, if it isn't mapped yet. */
1365 r = journal_file_map_data_hash_table(f);
1366 if (r < 0)
1367 return r;
1368
1369 osize = offsetof(Object, data.payload) + size;
1370
1371 m = le64toh(READ_NOW(f->header->data_hash_table_size)) / sizeof(HashItem);
1372 if (m <= 0)
1373 return -EBADMSG;
1374
1375 h = hash % m;
1376 p = le64toh(f->data_hash_table[h].head_hash_offset);
1377
1378 while (p > 0) {
1379 Object *o;
1380
1381 r = journal_file_move_to_object(f, OBJECT_DATA, p, &o);
1382 if (r < 0)
1383 return r;
1384
1385 if (le64toh(o->data.hash) != hash)
1386 goto next;
1387
1388 if (o->object.flags & OBJECT_COMPRESSION_MASK) {
1389 #if HAVE_XZ || HAVE_LZ4
1390 uint64_t l;
1391 size_t rsize = 0;
1392
1393 l = le64toh(READ_NOW(o->object.size));
1394 if (l <= offsetof(Object, data.payload))
1395 return -EBADMSG;
1396
1397 l -= offsetof(Object, data.payload);
1398
1399 r = decompress_blob(o->object.flags & OBJECT_COMPRESSION_MASK,
1400 o->data.payload, l, &f->compress_buffer, &f->compress_buffer_size, &rsize, 0);
1401 if (r < 0)
1402 return r;
1403
1404 if (rsize == size &&
1405 memcmp(f->compress_buffer, data, size) == 0) {
1406
1407 if (ret)
1408 *ret = o;
1409
1410 if (offset)
1411 *offset = p;
1412
1413 return 1;
1414 }
1415 #else
1416 return -EPROTONOSUPPORT;
1417 #endif
1418 } else if (le64toh(o->object.size) == osize &&
1419 memcmp(o->data.payload, data, size) == 0) {
1420
1421 if (ret)
1422 *ret = o;
1423
1424 if (offset)
1425 *offset = p;
1426
1427 return 1;
1428 }
1429
1430 next:
1431 p = le64toh(o->data.next_hash_offset);
1432 }
1433
1434 return 0;
1435 }
1436
1437 int journal_file_find_data_object(
1438 JournalFile *f,
1439 const void *data, uint64_t size,
1440 Object **ret, uint64_t *offset) {
1441
1442 uint64_t hash;
1443
1444 assert(f);
1445 assert(data || size == 0);
1446
1447 hash = hash64(data, size);
1448
1449 return journal_file_find_data_object_with_hash(f,
1450 data, size, hash,
1451 ret, offset);
1452 }
1453
1454 static int journal_file_append_field(
1455 JournalFile *f,
1456 const void *field, uint64_t size,
1457 Object **ret, uint64_t *offset) {
1458
1459 uint64_t hash, p;
1460 uint64_t osize;
1461 Object *o;
1462 int r;
1463
1464 assert(f);
1465 assert(field && size > 0);
1466
1467 hash = hash64(field, size);
1468
1469 r = journal_file_find_field_object_with_hash(f, field, size, hash, &o, &p);
1470 if (r < 0)
1471 return r;
1472 else if (r > 0) {
1473
1474 if (ret)
1475 *ret = o;
1476
1477 if (offset)
1478 *offset = p;
1479
1480 return 0;
1481 }
1482
1483 osize = offsetof(Object, field.payload) + size;
1484 r = journal_file_append_object(f, OBJECT_FIELD, osize, &o, &p);
1485 if (r < 0)
1486 return r;
1487
1488 o->field.hash = htole64(hash);
1489 memcpy(o->field.payload, field, size);
1490
1491 r = journal_file_link_field(f, o, p, hash);
1492 if (r < 0)
1493 return r;
1494
1495 /* The linking might have altered the window, so let's
1496 * refresh our pointer */
1497 r = journal_file_move_to_object(f, OBJECT_FIELD, p, &o);
1498 if (r < 0)
1499 return r;
1500
1501 #if HAVE_GCRYPT
1502 r = journal_file_hmac_put_object(f, OBJECT_FIELD, o, p);
1503 if (r < 0)
1504 return r;
1505 #endif
1506
1507 if (ret)
1508 *ret = o;
1509
1510 if (offset)
1511 *offset = p;
1512
1513 return 0;
1514 }
1515
1516 static int journal_file_append_data(
1517 JournalFile *f,
1518 const void *data, uint64_t size,
1519 Object **ret, uint64_t *offset) {
1520
1521 uint64_t hash, p;
1522 uint64_t osize;
1523 Object *o;
1524 int r, compression = 0;
1525 const void *eq;
1526
1527 assert(f);
1528 assert(data || size == 0);
1529
1530 hash = hash64(data, size);
1531
1532 r = journal_file_find_data_object_with_hash(f, data, size, hash, &o, &p);
1533 if (r < 0)
1534 return r;
1535 if (r > 0) {
1536
1537 if (ret)
1538 *ret = o;
1539
1540 if (offset)
1541 *offset = p;
1542
1543 return 0;
1544 }
1545
1546 osize = offsetof(Object, data.payload) + size;
1547 r = journal_file_append_object(f, OBJECT_DATA, osize, &o, &p);
1548 if (r < 0)
1549 return r;
1550
1551 o->data.hash = htole64(hash);
1552
1553 #if HAVE_XZ || HAVE_LZ4
1554 if (JOURNAL_FILE_COMPRESS(f) && size >= f->compress_threshold_bytes) {
1555 size_t rsize = 0;
1556
1557 compression = compress_blob(data, size, o->data.payload, size - 1, &rsize);
1558
1559 if (compression >= 0) {
1560 o->object.size = htole64(offsetof(Object, data.payload) + rsize);
1561 o->object.flags |= compression;
1562
1563 log_debug("Compressed data object %"PRIu64" -> %zu using %s",
1564 size, rsize, object_compressed_to_string(compression));
1565 } else
1566 /* Compression didn't work, we don't really care why, let's continue without compression */
1567 compression = 0;
1568 }
1569 #endif
1570
1571 if (compression == 0)
1572 memcpy_safe(o->data.payload, data, size);
1573
1574 r = journal_file_link_data(f, o, p, hash);
1575 if (r < 0)
1576 return r;
1577
1578 #if HAVE_GCRYPT
1579 r = journal_file_hmac_put_object(f, OBJECT_DATA, o, p);
1580 if (r < 0)
1581 return r;
1582 #endif
1583
1584 /* The linking might have altered the window, so let's
1585 * refresh our pointer */
1586 r = journal_file_move_to_object(f, OBJECT_DATA, p, &o);
1587 if (r < 0)
1588 return r;
1589
1590 if (!data)
1591 eq = NULL;
1592 else
1593 eq = memchr(data, '=', size);
1594 if (eq && eq > data) {
1595 Object *fo = NULL;
1596 uint64_t fp;
1597
1598 /* Create field object ... */
1599 r = journal_file_append_field(f, data, (uint8_t*) eq - (uint8_t*) data, &fo, &fp);
1600 if (r < 0)
1601 return r;
1602
1603 /* ... and link it in. */
1604 o->data.next_field_offset = fo->field.head_data_offset;
1605 fo->field.head_data_offset = le64toh(p);
1606 }
1607
1608 if (ret)
1609 *ret = o;
1610
1611 if (offset)
1612 *offset = p;
1613
1614 return 0;
1615 }
1616
1617 uint64_t journal_file_entry_n_items(Object *o) {
1618 uint64_t sz;
1619 assert(o);
1620
1621 if (o->object.type != OBJECT_ENTRY)
1622 return 0;
1623
1624 sz = le64toh(READ_NOW(o->object.size));
1625 if (sz < offsetof(Object, entry.items))
1626 return 0;
1627
1628 return (sz - offsetof(Object, entry.items)) / sizeof(EntryItem);
1629 }
1630
1631 uint64_t journal_file_entry_array_n_items(Object *o) {
1632 uint64_t sz;
1633
1634 assert(o);
1635
1636 if (o->object.type != OBJECT_ENTRY_ARRAY)
1637 return 0;
1638
1639 sz = le64toh(READ_NOW(o->object.size));
1640 if (sz < offsetof(Object, entry_array.items))
1641 return 0;
1642
1643 return (sz - offsetof(Object, entry_array.items)) / sizeof(uint64_t);
1644 }
1645
1646 uint64_t journal_file_hash_table_n_items(Object *o) {
1647 uint64_t sz;
1648
1649 assert(o);
1650
1651 if (!IN_SET(o->object.type, OBJECT_DATA_HASH_TABLE, OBJECT_FIELD_HASH_TABLE))
1652 return 0;
1653
1654 sz = le64toh(READ_NOW(o->object.size));
1655 if (sz < offsetof(Object, hash_table.items))
1656 return 0;
1657
1658 return (sz - offsetof(Object, hash_table.items)) / sizeof(HashItem);
1659 }
1660
1661 static int link_entry_into_array(JournalFile *f,
1662 le64_t *first,
1663 le64_t *idx,
1664 uint64_t p) {
1665 int r;
1666 uint64_t n = 0, ap = 0, q, i, a, hidx;
1667 Object *o;
1668
1669 assert(f);
1670 assert(f->header);
1671 assert(first);
1672 assert(idx);
1673 assert(p > 0);
1674
1675 a = le64toh(*first);
1676 i = hidx = le64toh(READ_NOW(*idx));
1677 while (a > 0) {
1678
1679 r = journal_file_move_to_object(f, OBJECT_ENTRY_ARRAY, a, &o);
1680 if (r < 0)
1681 return r;
1682
1683 n = journal_file_entry_array_n_items(o);
1684 if (i < n) {
1685 o->entry_array.items[i] = htole64(p);
1686 *idx = htole64(hidx + 1);
1687 return 0;
1688 }
1689
1690 i -= n;
1691 ap = a;
1692 a = le64toh(o->entry_array.next_entry_array_offset);
1693 }
1694
1695 if (hidx > n)
1696 n = (hidx+1) * 2;
1697 else
1698 n = n * 2;
1699
1700 if (n < 4)
1701 n = 4;
1702
1703 r = journal_file_append_object(f, OBJECT_ENTRY_ARRAY,
1704 offsetof(Object, entry_array.items) + n * sizeof(uint64_t),
1705 &o, &q);
1706 if (r < 0)
1707 return r;
1708
1709 #if HAVE_GCRYPT
1710 r = journal_file_hmac_put_object(f, OBJECT_ENTRY_ARRAY, o, q);
1711 if (r < 0)
1712 return r;
1713 #endif
1714
1715 o->entry_array.items[i] = htole64(p);
1716
1717 if (ap == 0)
1718 *first = htole64(q);
1719 else {
1720 r = journal_file_move_to_object(f, OBJECT_ENTRY_ARRAY, ap, &o);
1721 if (r < 0)
1722 return r;
1723
1724 o->entry_array.next_entry_array_offset = htole64(q);
1725 }
1726
1727 if (JOURNAL_HEADER_CONTAINS(f->header, n_entry_arrays))
1728 f->header->n_entry_arrays = htole64(le64toh(f->header->n_entry_arrays) + 1);
1729
1730 *idx = htole64(hidx + 1);
1731
1732 return 0;
1733 }
1734
1735 static int link_entry_into_array_plus_one(JournalFile *f,
1736 le64_t *extra,
1737 le64_t *first,
1738 le64_t *idx,
1739 uint64_t p) {
1740
1741 uint64_t hidx;
1742 int r;
1743
1744 assert(f);
1745 assert(extra);
1746 assert(first);
1747 assert(idx);
1748 assert(p > 0);
1749
1750 hidx = le64toh(READ_NOW(*idx));
1751 if (hidx == UINT64_MAX)
1752 return -EBADMSG;
1753 if (hidx == 0)
1754 *extra = htole64(p);
1755 else {
1756 le64_t i;
1757
1758 i = htole64(hidx - 1);
1759 r = link_entry_into_array(f, first, &i, p);
1760 if (r < 0)
1761 return r;
1762 }
1763
1764 *idx = htole64(hidx + 1);
1765 return 0;
1766 }
1767
1768 static int journal_file_link_entry_item(JournalFile *f, Object *o, uint64_t offset, uint64_t i) {
1769 uint64_t p;
1770 int r;
1771
1772 assert(f);
1773 assert(o);
1774 assert(offset > 0);
1775
1776 p = le64toh(o->entry.items[i].object_offset);
1777 r = journal_file_move_to_object(f, OBJECT_DATA, p, &o);
1778 if (r < 0)
1779 return r;
1780
1781 return link_entry_into_array_plus_one(f,
1782 &o->data.entry_offset,
1783 &o->data.entry_array_offset,
1784 &o->data.n_entries,
1785 offset);
1786 }
1787
1788 static int journal_file_link_entry(JournalFile *f, Object *o, uint64_t offset) {
1789 uint64_t n, i;
1790 int r;
1791
1792 assert(f);
1793 assert(f->header);
1794 assert(o);
1795 assert(offset > 0);
1796
1797 if (o->object.type != OBJECT_ENTRY)
1798 return -EINVAL;
1799
1800 __sync_synchronize();
1801
1802 /* Link up the entry itself */
1803 r = link_entry_into_array(f,
1804 &f->header->entry_array_offset,
1805 &f->header->n_entries,
1806 offset);
1807 if (r < 0)
1808 return r;
1809
1810 /* log_debug("=> %s seqnr=%"PRIu64" n_entries=%"PRIu64, f->path, o->entry.seqnum, f->header->n_entries); */
1811
1812 if (f->header->head_entry_realtime == 0)
1813 f->header->head_entry_realtime = o->entry.realtime;
1814
1815 f->header->tail_entry_realtime = o->entry.realtime;
1816 f->header->tail_entry_monotonic = o->entry.monotonic;
1817
1818 /* Link up the items */
1819 n = journal_file_entry_n_items(o);
1820 for (i = 0; i < n; i++) {
1821 r = journal_file_link_entry_item(f, o, offset, i);
1822 if (r < 0)
1823 return r;
1824 }
1825
1826 return 0;
1827 }
1828
1829 static int journal_file_append_entry_internal(
1830 JournalFile *f,
1831 const dual_timestamp *ts,
1832 const sd_id128_t *boot_id,
1833 uint64_t xor_hash,
1834 const EntryItem items[], unsigned n_items,
1835 uint64_t *seqnum,
1836 Object **ret, uint64_t *offset) {
1837 uint64_t np;
1838 uint64_t osize;
1839 Object *o;
1840 int r;
1841
1842 assert(f);
1843 assert(f->header);
1844 assert(items || n_items == 0);
1845 assert(ts);
1846
1847 osize = offsetof(Object, entry.items) + (n_items * sizeof(EntryItem));
1848
1849 r = journal_file_append_object(f, OBJECT_ENTRY, osize, &o, &np);
1850 if (r < 0)
1851 return r;
1852
1853 o->entry.seqnum = htole64(journal_file_entry_seqnum(f, seqnum));
1854 memcpy_safe(o->entry.items, items, n_items * sizeof(EntryItem));
1855 o->entry.realtime = htole64(ts->realtime);
1856 o->entry.monotonic = htole64(ts->monotonic);
1857 o->entry.xor_hash = htole64(xor_hash);
1858 if (boot_id)
1859 f->header->boot_id = *boot_id;
1860 o->entry.boot_id = f->header->boot_id;
1861
1862 #if HAVE_GCRYPT
1863 r = journal_file_hmac_put_object(f, OBJECT_ENTRY, o, np);
1864 if (r < 0)
1865 return r;
1866 #endif
1867
1868 r = journal_file_link_entry(f, o, np);
1869 if (r < 0)
1870 return r;
1871
1872 if (ret)
1873 *ret = o;
1874
1875 if (offset)
1876 *offset = np;
1877
1878 return 0;
1879 }
1880
1881 void journal_file_post_change(JournalFile *f) {
1882 assert(f);
1883
1884 if (f->fd < 0)
1885 return;
1886
1887 /* inotify() does not receive IN_MODIFY events from file
1888 * accesses done via mmap(). After each access we hence
1889 * trigger IN_MODIFY by truncating the journal file to its
1890 * current size which triggers IN_MODIFY. */
1891
1892 __sync_synchronize();
1893
1894 if (ftruncate(f->fd, f->last_stat.st_size) < 0)
1895 log_debug_errno(errno, "Failed to truncate file to its own size: %m");
1896 }
1897
1898 static int post_change_thunk(sd_event_source *timer, uint64_t usec, void *userdata) {
1899 assert(userdata);
1900
1901 journal_file_post_change(userdata);
1902
1903 return 1;
1904 }
1905
1906 static void schedule_post_change(JournalFile *f) {
1907 uint64_t now;
1908 int r;
1909
1910 assert(f);
1911 assert(f->post_change_timer);
1912
1913 r = sd_event_source_get_enabled(f->post_change_timer, NULL);
1914 if (r < 0) {
1915 log_debug_errno(r, "Failed to get ftruncate timer state: %m");
1916 goto fail;
1917 }
1918 if (r > 0)
1919 return;
1920
1921 r = sd_event_now(sd_event_source_get_event(f->post_change_timer), CLOCK_MONOTONIC, &now);
1922 if (r < 0) {
1923 log_debug_errno(r, "Failed to get clock's now for scheduling ftruncate: %m");
1924 goto fail;
1925 }
1926
1927 r = sd_event_source_set_time(f->post_change_timer, now + f->post_change_timer_period);
1928 if (r < 0) {
1929 log_debug_errno(r, "Failed to set time for scheduling ftruncate: %m");
1930 goto fail;
1931 }
1932
1933 r = sd_event_source_set_enabled(f->post_change_timer, SD_EVENT_ONESHOT);
1934 if (r < 0) {
1935 log_debug_errno(r, "Failed to enable scheduled ftruncate: %m");
1936 goto fail;
1937 }
1938
1939 return;
1940
1941 fail:
1942 /* On failure, let's simply post the change immediately. */
1943 journal_file_post_change(f);
1944 }
1945
1946 /* Enable coalesced change posting in a timer on the provided sd_event instance */
1947 int journal_file_enable_post_change_timer(JournalFile *f, sd_event *e, usec_t t) {
1948 _cleanup_(sd_event_source_unrefp) sd_event_source *timer = NULL;
1949 int r;
1950
1951 assert(f);
1952 assert_return(!f->post_change_timer, -EINVAL);
1953 assert(e);
1954 assert(t);
1955
1956 r = sd_event_add_time(e, &timer, CLOCK_MONOTONIC, 0, 0, post_change_thunk, f);
1957 if (r < 0)
1958 return r;
1959
1960 r = sd_event_source_set_enabled(timer, SD_EVENT_OFF);
1961 if (r < 0)
1962 return r;
1963
1964 f->post_change_timer = TAKE_PTR(timer);
1965 f->post_change_timer_period = t;
1966
1967 return r;
1968 }
1969
1970 static int entry_item_cmp(const EntryItem *a, const EntryItem *b) {
1971 return CMP(le64toh(a->object_offset), le64toh(b->object_offset));
1972 }
1973
1974 int journal_file_append_entry(
1975 JournalFile *f,
1976 const dual_timestamp *ts,
1977 const sd_id128_t *boot_id,
1978 const struct iovec iovec[], unsigned n_iovec,
1979 uint64_t *seqnum,
1980 Object **ret, uint64_t *offset) {
1981
1982 unsigned i;
1983 EntryItem *items;
1984 int r;
1985 uint64_t xor_hash = 0;
1986 struct dual_timestamp _ts;
1987
1988 assert(f);
1989 assert(f->header);
1990 assert(iovec || n_iovec == 0);
1991
1992 if (ts) {
1993 if (!VALID_REALTIME(ts->realtime))
1994 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG),
1995 "Invalid realtime timestamp %" PRIu64 ", refusing entry.",
1996 ts->realtime);
1997 if (!VALID_MONOTONIC(ts->monotonic))
1998 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG),
1999 "Invalid monotomic timestamp %" PRIu64 ", refusing entry.",
2000 ts->monotonic);
2001 } else {
2002 dual_timestamp_get(&_ts);
2003 ts = &_ts;
2004 }
2005
2006 #if HAVE_GCRYPT
2007 r = journal_file_maybe_append_tag(f, ts->realtime);
2008 if (r < 0)
2009 return r;
2010 #endif
2011
2012 /* alloca() can't take 0, hence let's allocate at least one */
2013 items = newa(EntryItem, MAX(1u, n_iovec));
2014
2015 for (i = 0; i < n_iovec; i++) {
2016 uint64_t p;
2017 Object *o;
2018
2019 r = journal_file_append_data(f, iovec[i].iov_base, iovec[i].iov_len, &o, &p);
2020 if (r < 0)
2021 return r;
2022
2023 xor_hash ^= le64toh(o->data.hash);
2024 items[i].object_offset = htole64(p);
2025 items[i].hash = o->data.hash;
2026 }
2027
2028 /* Order by the position on disk, in order to improve seek
2029 * times for rotating media. */
2030 typesafe_qsort(items, n_iovec, entry_item_cmp);
2031
2032 r = journal_file_append_entry_internal(f, ts, boot_id, xor_hash, items, n_iovec, seqnum, ret, offset);
2033
2034 /* If the memory mapping triggered a SIGBUS then we return an
2035 * IO error and ignore the error code passed down to us, since
2036 * it is very likely just an effect of a nullified replacement
2037 * mapping page */
2038
2039 if (mmap_cache_got_sigbus(f->mmap, f->cache_fd))
2040 r = -EIO;
2041
2042 if (f->post_change_timer)
2043 schedule_post_change(f);
2044 else
2045 journal_file_post_change(f);
2046
2047 return r;
2048 }
2049
2050 typedef struct ChainCacheItem {
2051 uint64_t first; /* the array at the beginning of the chain */
2052 uint64_t array; /* the cached array */
2053 uint64_t begin; /* the first item in the cached array */
2054 uint64_t total; /* the total number of items in all arrays before this one in the chain */
2055 uint64_t last_index; /* the last index we looked at, to optimize locality when bisecting */
2056 } ChainCacheItem;
2057
2058 static void chain_cache_put(
2059 OrderedHashmap *h,
2060 ChainCacheItem *ci,
2061 uint64_t first,
2062 uint64_t array,
2063 uint64_t begin,
2064 uint64_t total,
2065 uint64_t last_index) {
2066
2067 if (!ci) {
2068 /* If the chain item to cache for this chain is the
2069 * first one it's not worth caching anything */
2070 if (array == first)
2071 return;
2072
2073 if (ordered_hashmap_size(h) >= CHAIN_CACHE_MAX) {
2074 ci = ordered_hashmap_steal_first(h);
2075 assert(ci);
2076 } else {
2077 ci = new(ChainCacheItem, 1);
2078 if (!ci)
2079 return;
2080 }
2081
2082 ci->first = first;
2083
2084 if (ordered_hashmap_put(h, &ci->first, ci) < 0) {
2085 free(ci);
2086 return;
2087 }
2088 } else
2089 assert(ci->first == first);
2090
2091 ci->array = array;
2092 ci->begin = begin;
2093 ci->total = total;
2094 ci->last_index = last_index;
2095 }
2096
2097 static int generic_array_get(
2098 JournalFile *f,
2099 uint64_t first,
2100 uint64_t i,
2101 Object **ret, uint64_t *offset) {
2102
2103 Object *o;
2104 uint64_t p = 0, a, t = 0;
2105 int r;
2106 ChainCacheItem *ci;
2107
2108 assert(f);
2109
2110 a = first;
2111
2112 /* Try the chain cache first */
2113 ci = ordered_hashmap_get(f->chain_cache, &first);
2114 if (ci && i > ci->total) {
2115 a = ci->array;
2116 i -= ci->total;
2117 t = ci->total;
2118 }
2119
2120 while (a > 0) {
2121 uint64_t k;
2122
2123 r = journal_file_move_to_object(f, OBJECT_ENTRY_ARRAY, a, &o);
2124 if (r < 0)
2125 return r;
2126
2127 k = journal_file_entry_array_n_items(o);
2128 if (i < k) {
2129 p = le64toh(o->entry_array.items[i]);
2130 goto found;
2131 }
2132
2133 i -= k;
2134 t += k;
2135 a = le64toh(o->entry_array.next_entry_array_offset);
2136 }
2137
2138 return 0;
2139
2140 found:
2141 /* Let's cache this item for the next invocation */
2142 chain_cache_put(f->chain_cache, ci, first, a, le64toh(o->entry_array.items[0]), t, i);
2143
2144 r = journal_file_move_to_object(f, OBJECT_ENTRY, p, &o);
2145 if (r < 0)
2146 return r;
2147
2148 if (ret)
2149 *ret = o;
2150
2151 if (offset)
2152 *offset = p;
2153
2154 return 1;
2155 }
2156
2157 static int generic_array_get_plus_one(
2158 JournalFile *f,
2159 uint64_t extra,
2160 uint64_t first,
2161 uint64_t i,
2162 Object **ret, uint64_t *offset) {
2163
2164 Object *o;
2165
2166 assert(f);
2167
2168 if (i == 0) {
2169 int r;
2170
2171 r = journal_file_move_to_object(f, OBJECT_ENTRY, extra, &o);
2172 if (r < 0)
2173 return r;
2174
2175 if (ret)
2176 *ret = o;
2177
2178 if (offset)
2179 *offset = extra;
2180
2181 return 1;
2182 }
2183
2184 return generic_array_get(f, first, i-1, ret, offset);
2185 }
2186
2187 enum {
2188 TEST_FOUND,
2189 TEST_LEFT,
2190 TEST_RIGHT
2191 };
2192
2193 static int generic_array_bisect(
2194 JournalFile *f,
2195 uint64_t first,
2196 uint64_t n,
2197 uint64_t needle,
2198 int (*test_object)(JournalFile *f, uint64_t p, uint64_t needle),
2199 direction_t direction,
2200 Object **ret,
2201 uint64_t *offset,
2202 uint64_t *idx) {
2203
2204 uint64_t a, p, t = 0, i = 0, last_p = 0, last_index = (uint64_t) -1;
2205 bool subtract_one = false;
2206 Object *o, *array = NULL;
2207 int r;
2208 ChainCacheItem *ci;
2209
2210 assert(f);
2211 assert(test_object);
2212
2213 /* Start with the first array in the chain */
2214 a = first;
2215
2216 ci = ordered_hashmap_get(f->chain_cache, &first);
2217 if (ci && n > ci->total && ci->begin != 0) {
2218 /* Ah, we have iterated this bisection array chain
2219 * previously! Let's see if we can skip ahead in the
2220 * chain, as far as the last time. But we can't jump
2221 * backwards in the chain, so let's check that
2222 * first. */
2223
2224 r = test_object(f, ci->begin, needle);
2225 if (r < 0)
2226 return r;
2227
2228 if (r == TEST_LEFT) {
2229 /* OK, what we are looking for is right of the
2230 * begin of this EntryArray, so let's jump
2231 * straight to previously cached array in the
2232 * chain */
2233
2234 a = ci->array;
2235 n -= ci->total;
2236 t = ci->total;
2237 last_index = ci->last_index;
2238 }
2239 }
2240
2241 while (a > 0) {
2242 uint64_t left, right, k, lp;
2243
2244 r = journal_file_move_to_object(f, OBJECT_ENTRY_ARRAY, a, &array);
2245 if (r < 0)
2246 return r;
2247
2248 k = journal_file_entry_array_n_items(array);
2249 right = MIN(k, n);
2250 if (right <= 0)
2251 return 0;
2252
2253 i = right - 1;
2254 lp = p = le64toh(array->entry_array.items[i]);
2255 if (p <= 0)
2256 r = -EBADMSG;
2257 else
2258 r = test_object(f, p, needle);
2259 if (r == -EBADMSG) {
2260 log_debug_errno(r, "Encountered invalid entry while bisecting, cutting algorithm short. (1)");
2261 n = i;
2262 continue;
2263 }
2264 if (r < 0)
2265 return r;
2266
2267 if (r == TEST_FOUND)
2268 r = direction == DIRECTION_DOWN ? TEST_RIGHT : TEST_LEFT;
2269
2270 if (r == TEST_RIGHT) {
2271 left = 0;
2272 right -= 1;
2273
2274 if (last_index != (uint64_t) -1) {
2275 assert(last_index <= right);
2276
2277 /* If we cached the last index we
2278 * looked at, let's try to not to jump
2279 * too wildly around and see if we can
2280 * limit the range to look at early to
2281 * the immediate neighbors of the last
2282 * index we looked at. */
2283
2284 if (last_index > 0) {
2285 uint64_t x = last_index - 1;
2286
2287 p = le64toh(array->entry_array.items[x]);
2288 if (p <= 0)
2289 return -EBADMSG;
2290
2291 r = test_object(f, p, needle);
2292 if (r < 0)
2293 return r;
2294
2295 if (r == TEST_FOUND)
2296 r = direction == DIRECTION_DOWN ? TEST_RIGHT : TEST_LEFT;
2297
2298 if (r == TEST_RIGHT)
2299 right = x;
2300 else
2301 left = x + 1;
2302 }
2303
2304 if (last_index < right) {
2305 uint64_t y = last_index + 1;
2306
2307 p = le64toh(array->entry_array.items[y]);
2308 if (p <= 0)
2309 return -EBADMSG;
2310
2311 r = test_object(f, p, needle);
2312 if (r < 0)
2313 return r;
2314
2315 if (r == TEST_FOUND)
2316 r = direction == DIRECTION_DOWN ? TEST_RIGHT : TEST_LEFT;
2317
2318 if (r == TEST_RIGHT)
2319 right = y;
2320 else
2321 left = y + 1;
2322 }
2323 }
2324
2325 for (;;) {
2326 if (left == right) {
2327 if (direction == DIRECTION_UP)
2328 subtract_one = true;
2329
2330 i = left;
2331 goto found;
2332 }
2333
2334 assert(left < right);
2335 i = (left + right) / 2;
2336
2337 p = le64toh(array->entry_array.items[i]);
2338 if (p <= 0)
2339 r = -EBADMSG;
2340 else
2341 r = test_object(f, p, needle);
2342 if (r == -EBADMSG) {
2343 log_debug_errno(r, "Encountered invalid entry while bisecting, cutting algorithm short. (2)");
2344 right = n = i;
2345 continue;
2346 }
2347 if (r < 0)
2348 return r;
2349
2350 if (r == TEST_FOUND)
2351 r = direction == DIRECTION_DOWN ? TEST_RIGHT : TEST_LEFT;
2352
2353 if (r == TEST_RIGHT)
2354 right = i;
2355 else
2356 left = i + 1;
2357 }
2358 }
2359
2360 if (k >= n) {
2361 if (direction == DIRECTION_UP) {
2362 i = n;
2363 subtract_one = true;
2364 goto found;
2365 }
2366
2367 return 0;
2368 }
2369
2370 last_p = lp;
2371
2372 n -= k;
2373 t += k;
2374 last_index = (uint64_t) -1;
2375 a = le64toh(array->entry_array.next_entry_array_offset);
2376 }
2377
2378 return 0;
2379
2380 found:
2381 if (subtract_one && t == 0 && i == 0)
2382 return 0;
2383
2384 /* Let's cache this item for the next invocation */
2385 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);
2386
2387 if (subtract_one && i == 0)
2388 p = last_p;
2389 else if (subtract_one)
2390 p = le64toh(array->entry_array.items[i-1]);
2391 else
2392 p = le64toh(array->entry_array.items[i]);
2393
2394 r = journal_file_move_to_object(f, OBJECT_ENTRY, p, &o);
2395 if (r < 0)
2396 return r;
2397
2398 if (ret)
2399 *ret = o;
2400
2401 if (offset)
2402 *offset = p;
2403
2404 if (idx)
2405 *idx = t + i + (subtract_one ? -1 : 0);
2406
2407 return 1;
2408 }
2409
2410 static int generic_array_bisect_plus_one(
2411 JournalFile *f,
2412 uint64_t extra,
2413 uint64_t first,
2414 uint64_t n,
2415 uint64_t needle,
2416 int (*test_object)(JournalFile *f, uint64_t p, uint64_t needle),
2417 direction_t direction,
2418 Object **ret,
2419 uint64_t *offset,
2420 uint64_t *idx) {
2421
2422 int r;
2423 bool step_back = false;
2424 Object *o;
2425
2426 assert(f);
2427 assert(test_object);
2428
2429 if (n <= 0)
2430 return 0;
2431
2432 /* This bisects the array in object 'first', but first checks
2433 * an extra */
2434 r = test_object(f, extra, needle);
2435 if (r < 0)
2436 return r;
2437
2438 if (r == TEST_FOUND)
2439 r = direction == DIRECTION_DOWN ? TEST_RIGHT : TEST_LEFT;
2440
2441 /* if we are looking with DIRECTION_UP then we need to first
2442 see if in the actual array there is a matching entry, and
2443 return the last one of that. But if there isn't any we need
2444 to return this one. Hence remember this, and return it
2445 below. */
2446 if (r == TEST_LEFT)
2447 step_back = direction == DIRECTION_UP;
2448
2449 if (r == TEST_RIGHT) {
2450 if (direction == DIRECTION_DOWN)
2451 goto found;
2452 else
2453 return 0;
2454 }
2455
2456 r = generic_array_bisect(f, first, n-1, needle, test_object, direction, ret, offset, idx);
2457
2458 if (r == 0 && step_back)
2459 goto found;
2460
2461 if (r > 0 && idx)
2462 (*idx)++;
2463
2464 return r;
2465
2466 found:
2467 r = journal_file_move_to_object(f, OBJECT_ENTRY, extra, &o);
2468 if (r < 0)
2469 return r;
2470
2471 if (ret)
2472 *ret = o;
2473
2474 if (offset)
2475 *offset = extra;
2476
2477 if (idx)
2478 *idx = 0;
2479
2480 return 1;
2481 }
2482
2483 _pure_ static int test_object_offset(JournalFile *f, uint64_t p, uint64_t needle) {
2484 assert(f);
2485 assert(p > 0);
2486
2487 if (p == needle)
2488 return TEST_FOUND;
2489 else if (p < needle)
2490 return TEST_LEFT;
2491 else
2492 return TEST_RIGHT;
2493 }
2494
2495 static int test_object_seqnum(JournalFile *f, uint64_t p, uint64_t needle) {
2496 uint64_t sq;
2497 Object *o;
2498 int r;
2499
2500 assert(f);
2501 assert(p > 0);
2502
2503 r = journal_file_move_to_object(f, OBJECT_ENTRY, p, &o);
2504 if (r < 0)
2505 return r;
2506
2507 sq = le64toh(READ_NOW(o->entry.seqnum));
2508 if (sq == needle)
2509 return TEST_FOUND;
2510 else if (sq < needle)
2511 return TEST_LEFT;
2512 else
2513 return TEST_RIGHT;
2514 }
2515
2516 int journal_file_move_to_entry_by_seqnum(
2517 JournalFile *f,
2518 uint64_t seqnum,
2519 direction_t direction,
2520 Object **ret,
2521 uint64_t *offset) {
2522 assert(f);
2523 assert(f->header);
2524
2525 return generic_array_bisect(f,
2526 le64toh(f->header->entry_array_offset),
2527 le64toh(f->header->n_entries),
2528 seqnum,
2529 test_object_seqnum,
2530 direction,
2531 ret, offset, NULL);
2532 }
2533
2534 static int test_object_realtime(JournalFile *f, uint64_t p, uint64_t needle) {
2535 Object *o;
2536 uint64_t rt;
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 rt = le64toh(READ_NOW(o->entry.realtime));
2547 if (rt == needle)
2548 return TEST_FOUND;
2549 else if (rt < needle)
2550 return TEST_LEFT;
2551 else
2552 return TEST_RIGHT;
2553 }
2554
2555 int journal_file_move_to_entry_by_realtime(
2556 JournalFile *f,
2557 uint64_t realtime,
2558 direction_t direction,
2559 Object **ret,
2560 uint64_t *offset) {
2561 assert(f);
2562 assert(f->header);
2563
2564 return generic_array_bisect(f,
2565 le64toh(f->header->entry_array_offset),
2566 le64toh(f->header->n_entries),
2567 realtime,
2568 test_object_realtime,
2569 direction,
2570 ret, offset, NULL);
2571 }
2572
2573 static int test_object_monotonic(JournalFile *f, uint64_t p, uint64_t needle) {
2574 Object *o;
2575 uint64_t m;
2576 int r;
2577
2578 assert(f);
2579 assert(p > 0);
2580
2581 r = journal_file_move_to_object(f, OBJECT_ENTRY, p, &o);
2582 if (r < 0)
2583 return r;
2584
2585 m = le64toh(READ_NOW(o->entry.monotonic));
2586 if (m == needle)
2587 return TEST_FOUND;
2588 else if (m < needle)
2589 return TEST_LEFT;
2590 else
2591 return TEST_RIGHT;
2592 }
2593
2594 static int find_data_object_by_boot_id(
2595 JournalFile *f,
2596 sd_id128_t boot_id,
2597 Object **o,
2598 uint64_t *b) {
2599
2600 char t[STRLEN("_BOOT_ID=") + 32 + 1] = "_BOOT_ID=";
2601
2602 sd_id128_to_string(boot_id, t + 9);
2603 return journal_file_find_data_object(f, t, sizeof(t) - 1, o, b);
2604 }
2605
2606 int journal_file_move_to_entry_by_monotonic(
2607 JournalFile *f,
2608 sd_id128_t boot_id,
2609 uint64_t monotonic,
2610 direction_t direction,
2611 Object **ret,
2612 uint64_t *offset) {
2613
2614 Object *o;
2615 int r;
2616
2617 assert(f);
2618
2619 r = find_data_object_by_boot_id(f, boot_id, &o, NULL);
2620 if (r < 0)
2621 return r;
2622 if (r == 0)
2623 return -ENOENT;
2624
2625 return generic_array_bisect_plus_one(f,
2626 le64toh(o->data.entry_offset),
2627 le64toh(o->data.entry_array_offset),
2628 le64toh(o->data.n_entries),
2629 monotonic,
2630 test_object_monotonic,
2631 direction,
2632 ret, offset, NULL);
2633 }
2634
2635 void journal_file_reset_location(JournalFile *f) {
2636 f->location_type = LOCATION_HEAD;
2637 f->current_offset = 0;
2638 f->current_seqnum = 0;
2639 f->current_realtime = 0;
2640 f->current_monotonic = 0;
2641 zero(f->current_boot_id);
2642 f->current_xor_hash = 0;
2643 }
2644
2645 void journal_file_save_location(JournalFile *f, Object *o, uint64_t offset) {
2646 f->location_type = LOCATION_SEEK;
2647 f->current_offset = offset;
2648 f->current_seqnum = le64toh(o->entry.seqnum);
2649 f->current_realtime = le64toh(o->entry.realtime);
2650 f->current_monotonic = le64toh(o->entry.monotonic);
2651 f->current_boot_id = o->entry.boot_id;
2652 f->current_xor_hash = le64toh(o->entry.xor_hash);
2653 }
2654
2655 int journal_file_compare_locations(JournalFile *af, JournalFile *bf) {
2656 int r;
2657
2658 assert(af);
2659 assert(af->header);
2660 assert(bf);
2661 assert(bf->header);
2662 assert(af->location_type == LOCATION_SEEK);
2663 assert(bf->location_type == LOCATION_SEEK);
2664
2665 /* If contents and timestamps match, these entries are
2666 * identical, even if the seqnum does not match */
2667 if (sd_id128_equal(af->current_boot_id, bf->current_boot_id) &&
2668 af->current_monotonic == bf->current_monotonic &&
2669 af->current_realtime == bf->current_realtime &&
2670 af->current_xor_hash == bf->current_xor_hash)
2671 return 0;
2672
2673 if (sd_id128_equal(af->header->seqnum_id, bf->header->seqnum_id)) {
2674
2675 /* If this is from the same seqnum source, compare
2676 * seqnums */
2677 r = CMP(af->current_seqnum, bf->current_seqnum);
2678 if (r != 0)
2679 return r;
2680
2681 /* Wow! This is weird, different data but the same
2682 * seqnums? Something is borked, but let's make the
2683 * best of it and compare by time. */
2684 }
2685
2686 if (sd_id128_equal(af->current_boot_id, bf->current_boot_id)) {
2687
2688 /* If the boot id matches, compare monotonic time */
2689 r = CMP(af->current_monotonic, bf->current_monotonic);
2690 if (r != 0)
2691 return r;
2692 }
2693
2694 /* Otherwise, compare UTC time */
2695 r = CMP(af->current_realtime, bf->current_realtime);
2696 if (r != 0)
2697 return r;
2698
2699 /* Finally, compare by contents */
2700 return CMP(af->current_xor_hash, bf->current_xor_hash);
2701 }
2702
2703 static int bump_array_index(uint64_t *i, direction_t direction, uint64_t n) {
2704
2705 /* Increase or decrease the specified index, in the right direction. */
2706
2707 if (direction == DIRECTION_DOWN) {
2708 if (*i >= n - 1)
2709 return 0;
2710
2711 (*i) ++;
2712 } else {
2713 if (*i <= 0)
2714 return 0;
2715
2716 (*i) --;
2717 }
2718
2719 return 1;
2720 }
2721
2722 static bool check_properly_ordered(uint64_t new_offset, uint64_t old_offset, direction_t direction) {
2723
2724 /* Consider it an error if any of the two offsets is uninitialized */
2725 if (old_offset == 0 || new_offset == 0)
2726 return false;
2727
2728 /* If we go down, the new offset must be larger than the old one. */
2729 return direction == DIRECTION_DOWN ?
2730 new_offset > old_offset :
2731 new_offset < old_offset;
2732 }
2733
2734 int journal_file_next_entry(
2735 JournalFile *f,
2736 uint64_t p,
2737 direction_t direction,
2738 Object **ret, uint64_t *offset) {
2739
2740 uint64_t i, n, ofs;
2741 int r;
2742
2743 assert(f);
2744 assert(f->header);
2745
2746 n = le64toh(READ_NOW(f->header->n_entries));
2747 if (n <= 0)
2748 return 0;
2749
2750 if (p == 0)
2751 i = direction == DIRECTION_DOWN ? 0 : n - 1;
2752 else {
2753 r = generic_array_bisect(f,
2754 le64toh(f->header->entry_array_offset),
2755 le64toh(f->header->n_entries),
2756 p,
2757 test_object_offset,
2758 DIRECTION_DOWN,
2759 NULL, NULL,
2760 &i);
2761 if (r <= 0)
2762 return r;
2763
2764 r = bump_array_index(&i, direction, n);
2765 if (r <= 0)
2766 return r;
2767 }
2768
2769 /* And jump to it */
2770 for (;;) {
2771 r = generic_array_get(f,
2772 le64toh(f->header->entry_array_offset),
2773 i,
2774 ret, &ofs);
2775 if (r > 0)
2776 break;
2777 if (r != -EBADMSG)
2778 return r;
2779
2780 /* OK, so this entry is borked. Most likely some entry didn't get synced to disk properly, let's see if
2781 * the next one might work for us instead. */
2782 log_debug_errno(r, "Entry item %" PRIu64 " is bad, skipping over it.", i);
2783
2784 r = bump_array_index(&i, direction, n);
2785 if (r <= 0)
2786 return r;
2787 }
2788
2789 /* Ensure our array is properly ordered. */
2790 if (p > 0 && !check_properly_ordered(ofs, p, direction))
2791 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG),
2792 "%s: entry array not properly ordered at entry %" PRIu64,
2793 f->path, i);
2794
2795 if (offset)
2796 *offset = ofs;
2797
2798 return 1;
2799 }
2800
2801 int journal_file_next_entry_for_data(
2802 JournalFile *f,
2803 Object *o, uint64_t p,
2804 uint64_t data_offset,
2805 direction_t direction,
2806 Object **ret, uint64_t *offset) {
2807
2808 uint64_t i, n, ofs;
2809 Object *d;
2810 int r;
2811
2812 assert(f);
2813 assert(p > 0 || !o);
2814
2815 r = journal_file_move_to_object(f, OBJECT_DATA, data_offset, &d);
2816 if (r < 0)
2817 return r;
2818
2819 n = le64toh(READ_NOW(d->data.n_entries));
2820 if (n <= 0)
2821 return n;
2822
2823 if (!o)
2824 i = direction == DIRECTION_DOWN ? 0 : n - 1;
2825 else {
2826 if (o->object.type != OBJECT_ENTRY)
2827 return -EINVAL;
2828
2829 r = generic_array_bisect_plus_one(f,
2830 le64toh(d->data.entry_offset),
2831 le64toh(d->data.entry_array_offset),
2832 le64toh(d->data.n_entries),
2833 p,
2834 test_object_offset,
2835 DIRECTION_DOWN,
2836 NULL, NULL,
2837 &i);
2838
2839 if (r <= 0)
2840 return r;
2841
2842 r = bump_array_index(&i, direction, n);
2843 if (r <= 0)
2844 return r;
2845 }
2846
2847 for (;;) {
2848 r = generic_array_get_plus_one(f,
2849 le64toh(d->data.entry_offset),
2850 le64toh(d->data.entry_array_offset),
2851 i,
2852 ret, &ofs);
2853 if (r > 0)
2854 break;
2855 if (r != -EBADMSG)
2856 return r;
2857
2858 log_debug_errno(r, "Data entry item %" PRIu64 " is bad, skipping over it.", i);
2859
2860 r = bump_array_index(&i, direction, n);
2861 if (r <= 0)
2862 return r;
2863 }
2864
2865 /* Ensure our array is properly ordered. */
2866 if (p > 0 && check_properly_ordered(ofs, p, direction))
2867 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG),
2868 "%s data entry array not properly ordered at entry %" PRIu64,
2869 f->path, i);
2870
2871 if (offset)
2872 *offset = ofs;
2873
2874 return 1;
2875 }
2876
2877 int journal_file_move_to_entry_by_offset_for_data(
2878 JournalFile *f,
2879 uint64_t data_offset,
2880 uint64_t p,
2881 direction_t direction,
2882 Object **ret, uint64_t *offset) {
2883
2884 int r;
2885 Object *d;
2886
2887 assert(f);
2888
2889 r = journal_file_move_to_object(f, OBJECT_DATA, data_offset, &d);
2890 if (r < 0)
2891 return r;
2892
2893 return generic_array_bisect_plus_one(f,
2894 le64toh(d->data.entry_offset),
2895 le64toh(d->data.entry_array_offset),
2896 le64toh(d->data.n_entries),
2897 p,
2898 test_object_offset,
2899 direction,
2900 ret, offset, NULL);
2901 }
2902
2903 int journal_file_move_to_entry_by_monotonic_for_data(
2904 JournalFile *f,
2905 uint64_t data_offset,
2906 sd_id128_t boot_id,
2907 uint64_t monotonic,
2908 direction_t direction,
2909 Object **ret, uint64_t *offset) {
2910
2911 Object *o, *d;
2912 int r;
2913 uint64_t b, z;
2914
2915 assert(f);
2916
2917 /* First, seek by time */
2918 r = find_data_object_by_boot_id(f, boot_id, &o, &b);
2919 if (r < 0)
2920 return r;
2921 if (r == 0)
2922 return -ENOENT;
2923
2924 r = generic_array_bisect_plus_one(f,
2925 le64toh(o->data.entry_offset),
2926 le64toh(o->data.entry_array_offset),
2927 le64toh(o->data.n_entries),
2928 monotonic,
2929 test_object_monotonic,
2930 direction,
2931 NULL, &z, NULL);
2932 if (r <= 0)
2933 return r;
2934
2935 /* And now, continue seeking until we find an entry that
2936 * exists in both bisection arrays */
2937
2938 for (;;) {
2939 Object *qo;
2940 uint64_t p, q;
2941
2942 r = journal_file_move_to_object(f, OBJECT_DATA, data_offset, &d);
2943 if (r < 0)
2944 return r;
2945
2946 r = generic_array_bisect_plus_one(f,
2947 le64toh(d->data.entry_offset),
2948 le64toh(d->data.entry_array_offset),
2949 le64toh(d->data.n_entries),
2950 z,
2951 test_object_offset,
2952 direction,
2953 NULL, &p, NULL);
2954 if (r <= 0)
2955 return r;
2956
2957 r = journal_file_move_to_object(f, OBJECT_DATA, b, &o);
2958 if (r < 0)
2959 return r;
2960
2961 r = generic_array_bisect_plus_one(f,
2962 le64toh(o->data.entry_offset),
2963 le64toh(o->data.entry_array_offset),
2964 le64toh(o->data.n_entries),
2965 p,
2966 test_object_offset,
2967 direction,
2968 &qo, &q, NULL);
2969
2970 if (r <= 0)
2971 return r;
2972
2973 if (p == q) {
2974 if (ret)
2975 *ret = qo;
2976 if (offset)
2977 *offset = q;
2978
2979 return 1;
2980 }
2981
2982 z = q;
2983 }
2984 }
2985
2986 int journal_file_move_to_entry_by_seqnum_for_data(
2987 JournalFile *f,
2988 uint64_t data_offset,
2989 uint64_t seqnum,
2990 direction_t direction,
2991 Object **ret, uint64_t *offset) {
2992
2993 Object *d;
2994 int r;
2995
2996 assert(f);
2997
2998 r = journal_file_move_to_object(f, OBJECT_DATA, data_offset, &d);
2999 if (r < 0)
3000 return r;
3001
3002 return generic_array_bisect_plus_one(f,
3003 le64toh(d->data.entry_offset),
3004 le64toh(d->data.entry_array_offset),
3005 le64toh(d->data.n_entries),
3006 seqnum,
3007 test_object_seqnum,
3008 direction,
3009 ret, offset, NULL);
3010 }
3011
3012 int journal_file_move_to_entry_by_realtime_for_data(
3013 JournalFile *f,
3014 uint64_t data_offset,
3015 uint64_t realtime,
3016 direction_t direction,
3017 Object **ret, uint64_t *offset) {
3018
3019 Object *d;
3020 int r;
3021
3022 assert(f);
3023
3024 r = journal_file_move_to_object(f, OBJECT_DATA, data_offset, &d);
3025 if (r < 0)
3026 return r;
3027
3028 return generic_array_bisect_plus_one(f,
3029 le64toh(d->data.entry_offset),
3030 le64toh(d->data.entry_array_offset),
3031 le64toh(d->data.n_entries),
3032 realtime,
3033 test_object_realtime,
3034 direction,
3035 ret, offset, NULL);
3036 }
3037
3038 void journal_file_dump(JournalFile *f) {
3039 Object *o;
3040 int r;
3041 uint64_t p;
3042
3043 assert(f);
3044 assert(f->header);
3045
3046 journal_file_print_header(f);
3047
3048 p = le64toh(READ_NOW(f->header->header_size));
3049 while (p != 0) {
3050 r = journal_file_move_to_object(f, OBJECT_UNUSED, p, &o);
3051 if (r < 0)
3052 goto fail;
3053
3054 switch (o->object.type) {
3055
3056 case OBJECT_UNUSED:
3057 printf("Type: OBJECT_UNUSED\n");
3058 break;
3059
3060 case OBJECT_DATA:
3061 printf("Type: OBJECT_DATA\n");
3062 break;
3063
3064 case OBJECT_FIELD:
3065 printf("Type: OBJECT_FIELD\n");
3066 break;
3067
3068 case OBJECT_ENTRY:
3069 printf("Type: OBJECT_ENTRY seqnum=%"PRIu64" monotonic=%"PRIu64" realtime=%"PRIu64"\n",
3070 le64toh(o->entry.seqnum),
3071 le64toh(o->entry.monotonic),
3072 le64toh(o->entry.realtime));
3073 break;
3074
3075 case OBJECT_FIELD_HASH_TABLE:
3076 printf("Type: OBJECT_FIELD_HASH_TABLE\n");
3077 break;
3078
3079 case OBJECT_DATA_HASH_TABLE:
3080 printf("Type: OBJECT_DATA_HASH_TABLE\n");
3081 break;
3082
3083 case OBJECT_ENTRY_ARRAY:
3084 printf("Type: OBJECT_ENTRY_ARRAY\n");
3085 break;
3086
3087 case OBJECT_TAG:
3088 printf("Type: OBJECT_TAG seqnum=%"PRIu64" epoch=%"PRIu64"\n",
3089 le64toh(o->tag.seqnum),
3090 le64toh(o->tag.epoch));
3091 break;
3092
3093 default:
3094 printf("Type: unknown (%i)\n", o->object.type);
3095 break;
3096 }
3097
3098 if (o->object.flags & OBJECT_COMPRESSION_MASK)
3099 printf("Flags: %s\n",
3100 object_compressed_to_string(o->object.flags & OBJECT_COMPRESSION_MASK));
3101
3102 if (p == le64toh(f->header->tail_object_offset))
3103 p = 0;
3104 else
3105 p += ALIGN64(le64toh(o->object.size));
3106 }
3107
3108 return;
3109 fail:
3110 log_error("File corrupt");
3111 }
3112
3113 static const char* format_timestamp_safe(char *buf, size_t l, usec_t t) {
3114 const char *x;
3115
3116 x = format_timestamp(buf, l, t);
3117 if (x)
3118 return x;
3119 return " --- ";
3120 }
3121
3122 void journal_file_print_header(JournalFile *f) {
3123 char a[SD_ID128_STRING_MAX], b[SD_ID128_STRING_MAX], c[SD_ID128_STRING_MAX], d[SD_ID128_STRING_MAX];
3124 char x[FORMAT_TIMESTAMP_MAX], y[FORMAT_TIMESTAMP_MAX], z[FORMAT_TIMESTAMP_MAX];
3125 struct stat st;
3126 char bytes[FORMAT_BYTES_MAX];
3127
3128 assert(f);
3129 assert(f->header);
3130
3131 printf("File path: %s\n"
3132 "File ID: %s\n"
3133 "Machine ID: %s\n"
3134 "Boot ID: %s\n"
3135 "Sequential number ID: %s\n"
3136 "State: %s\n"
3137 "Compatible flags:%s%s\n"
3138 "Incompatible flags:%s%s%s\n"
3139 "Header size: %"PRIu64"\n"
3140 "Arena size: %"PRIu64"\n"
3141 "Data hash table size: %"PRIu64"\n"
3142 "Field hash table size: %"PRIu64"\n"
3143 "Rotate suggested: %s\n"
3144 "Head sequential number: %"PRIu64" (%"PRIx64")\n"
3145 "Tail sequential number: %"PRIu64" (%"PRIx64")\n"
3146 "Head realtime timestamp: %s (%"PRIx64")\n"
3147 "Tail realtime timestamp: %s (%"PRIx64")\n"
3148 "Tail monotonic timestamp: %s (%"PRIx64")\n"
3149 "Objects: %"PRIu64"\n"
3150 "Entry objects: %"PRIu64"\n",
3151 f->path,
3152 sd_id128_to_string(f->header->file_id, a),
3153 sd_id128_to_string(f->header->machine_id, b),
3154 sd_id128_to_string(f->header->boot_id, c),
3155 sd_id128_to_string(f->header->seqnum_id, d),
3156 f->header->state == STATE_OFFLINE ? "OFFLINE" :
3157 f->header->state == STATE_ONLINE ? "ONLINE" :
3158 f->header->state == STATE_ARCHIVED ? "ARCHIVED" : "UNKNOWN",
3159 JOURNAL_HEADER_SEALED(f->header) ? " SEALED" : "",
3160 (le32toh(f->header->compatible_flags) & ~HEADER_COMPATIBLE_ANY) ? " ???" : "",
3161 JOURNAL_HEADER_COMPRESSED_XZ(f->header) ? " COMPRESSED-XZ" : "",
3162 JOURNAL_HEADER_COMPRESSED_LZ4(f->header) ? " COMPRESSED-LZ4" : "",
3163 (le32toh(f->header->incompatible_flags) & ~HEADER_INCOMPATIBLE_ANY) ? " ???" : "",
3164 le64toh(f->header->header_size),
3165 le64toh(f->header->arena_size),
3166 le64toh(f->header->data_hash_table_size) / sizeof(HashItem),
3167 le64toh(f->header->field_hash_table_size) / sizeof(HashItem),
3168 yes_no(journal_file_rotate_suggested(f, 0)),
3169 le64toh(f->header->head_entry_seqnum), le64toh(f->header->head_entry_seqnum),
3170 le64toh(f->header->tail_entry_seqnum), le64toh(f->header->tail_entry_seqnum),
3171 format_timestamp_safe(x, sizeof(x), le64toh(f->header->head_entry_realtime)), le64toh(f->header->head_entry_realtime),
3172 format_timestamp_safe(y, sizeof(y), le64toh(f->header->tail_entry_realtime)), le64toh(f->header->tail_entry_realtime),
3173 format_timespan(z, sizeof(z), le64toh(f->header->tail_entry_monotonic), USEC_PER_MSEC), le64toh(f->header->tail_entry_monotonic),
3174 le64toh(f->header->n_objects),
3175 le64toh(f->header->n_entries));
3176
3177 if (JOURNAL_HEADER_CONTAINS(f->header, n_data))
3178 printf("Data objects: %"PRIu64"\n"
3179 "Data hash table fill: %.1f%%\n",
3180 le64toh(f->header->n_data),
3181 100.0 * (double) le64toh(f->header->n_data) / ((double) (le64toh(f->header->data_hash_table_size) / sizeof(HashItem))));
3182
3183 if (JOURNAL_HEADER_CONTAINS(f->header, n_fields))
3184 printf("Field objects: %"PRIu64"\n"
3185 "Field hash table fill: %.1f%%\n",
3186 le64toh(f->header->n_fields),
3187 100.0 * (double) le64toh(f->header->n_fields) / ((double) (le64toh(f->header->field_hash_table_size) / sizeof(HashItem))));
3188
3189 if (JOURNAL_HEADER_CONTAINS(f->header, n_tags))
3190 printf("Tag objects: %"PRIu64"\n",
3191 le64toh(f->header->n_tags));
3192 if (JOURNAL_HEADER_CONTAINS(f->header, n_entry_arrays))
3193 printf("Entry array objects: %"PRIu64"\n",
3194 le64toh(f->header->n_entry_arrays));
3195
3196 if (fstat(f->fd, &st) >= 0)
3197 printf("Disk usage: %s\n", format_bytes(bytes, sizeof(bytes), (uint64_t) st.st_blocks * 512ULL));
3198 }
3199
3200 static int journal_file_warn_btrfs(JournalFile *f) {
3201 unsigned attrs;
3202 int r;
3203
3204 assert(f);
3205
3206 /* Before we write anything, check if the COW logic is turned
3207 * off on btrfs. Given our write pattern that is quite
3208 * unfriendly to COW file systems this should greatly improve
3209 * performance on COW file systems, such as btrfs, at the
3210 * expense of data integrity features (which shouldn't be too
3211 * bad, given that we do our own checksumming). */
3212
3213 r = btrfs_is_filesystem(f->fd);
3214 if (r < 0)
3215 return log_warning_errno(r, "Failed to determine if journal is on btrfs: %m");
3216 if (!r)
3217 return 0;
3218
3219 r = read_attr_fd(f->fd, &attrs);
3220 if (r < 0)
3221 return log_warning_errno(r, "Failed to read file attributes: %m");
3222
3223 if (attrs & FS_NOCOW_FL) {
3224 log_debug("Detected btrfs file system with copy-on-write disabled, all is good.");
3225 return 0;
3226 }
3227
3228 log_notice("Creating journal file %s on a btrfs file system, and copy-on-write is enabled. "
3229 "This is likely to slow down journal access substantially, please consider turning "
3230 "off the copy-on-write file attribute on the journal directory, using chattr +C.", f->path);
3231
3232 return 1;
3233 }
3234
3235 int journal_file_open(
3236 int fd,
3237 const char *fname,
3238 int flags,
3239 mode_t mode,
3240 bool compress,
3241 uint64_t compress_threshold_bytes,
3242 bool seal,
3243 JournalMetrics *metrics,
3244 MMapCache *mmap_cache,
3245 Set *deferred_closes,
3246 JournalFile *template,
3247 JournalFile **ret) {
3248
3249 bool newly_created = false;
3250 JournalFile *f;
3251 void *h;
3252 int r;
3253
3254 assert(ret);
3255 assert(fd >= 0 || fname);
3256
3257 if (!IN_SET((flags & O_ACCMODE), O_RDONLY, O_RDWR))
3258 return -EINVAL;
3259
3260 if (fname && (flags & O_CREAT) && !endswith(fname, ".journal"))
3261 return -EINVAL;
3262
3263 f = new(JournalFile, 1);
3264 if (!f)
3265 return -ENOMEM;
3266
3267 *f = (JournalFile) {
3268 .fd = fd,
3269 .mode = mode,
3270
3271 .flags = flags,
3272 .prot = prot_from_flags(flags),
3273 .writable = (flags & O_ACCMODE) != O_RDONLY,
3274
3275 #if HAVE_LZ4
3276 .compress_lz4 = compress,
3277 #elif HAVE_XZ
3278 .compress_xz = compress,
3279 #endif
3280 .compress_threshold_bytes = compress_threshold_bytes == (uint64_t) -1 ?
3281 DEFAULT_COMPRESS_THRESHOLD :
3282 MAX(MIN_COMPRESS_THRESHOLD, compress_threshold_bytes),
3283 #if HAVE_GCRYPT
3284 .seal = seal,
3285 #endif
3286 };
3287
3288 if (DEBUG_LOGGING) {
3289 static int last_seal = -1, last_compress = -1;
3290 static uint64_t last_bytes = UINT64_MAX;
3291 char bytes[FORMAT_BYTES_MAX];
3292
3293 if (last_seal != f->seal ||
3294 last_compress != JOURNAL_FILE_COMPRESS(f) ||
3295 last_bytes != f->compress_threshold_bytes) {
3296
3297 log_debug("Journal effective settings seal=%s compress=%s compress_threshold_bytes=%s",
3298 yes_no(f->seal), yes_no(JOURNAL_FILE_COMPRESS(f)),
3299 format_bytes(bytes, sizeof bytes, f->compress_threshold_bytes));
3300 last_seal = f->seal;
3301 last_compress = JOURNAL_FILE_COMPRESS(f);
3302 last_bytes = f->compress_threshold_bytes;
3303 }
3304 }
3305
3306 if (mmap_cache)
3307 f->mmap = mmap_cache_ref(mmap_cache);
3308 else {
3309 f->mmap = mmap_cache_new();
3310 if (!f->mmap) {
3311 r = -ENOMEM;
3312 goto fail;
3313 }
3314 }
3315
3316 if (fname) {
3317 f->path = strdup(fname);
3318 if (!f->path) {
3319 r = -ENOMEM;
3320 goto fail;
3321 }
3322 } else {
3323 assert(fd >= 0);
3324
3325 /* If we don't know the path, fill in something explanatory and vaguely useful */
3326 if (asprintf(&f->path, "/proc/self/%i", fd) < 0) {
3327 r = -ENOMEM;
3328 goto fail;
3329 }
3330 }
3331
3332 f->chain_cache = ordered_hashmap_new(&uint64_hash_ops);
3333 if (!f->chain_cache) {
3334 r = -ENOMEM;
3335 goto fail;
3336 }
3337
3338 if (f->fd < 0) {
3339 /* We pass O_NONBLOCK here, so that in case somebody pointed us to some character device node or FIFO
3340 * or so, we likely fail quickly than block for long. For regular files O_NONBLOCK has no effect, hence
3341 * it doesn't hurt in that case. */
3342
3343 f->fd = open(f->path, f->flags|O_CLOEXEC|O_NONBLOCK, f->mode);
3344 if (f->fd < 0) {
3345 r = -errno;
3346 goto fail;
3347 }
3348
3349 /* fds we opened here by us should also be closed by us. */
3350 f->close_fd = true;
3351
3352 r = fd_nonblock(f->fd, false);
3353 if (r < 0)
3354 goto fail;
3355 }
3356
3357 f->cache_fd = mmap_cache_add_fd(f->mmap, f->fd);
3358 if (!f->cache_fd) {
3359 r = -ENOMEM;
3360 goto fail;
3361 }
3362
3363 r = journal_file_fstat(f);
3364 if (r < 0)
3365 goto fail;
3366
3367 if (f->last_stat.st_size == 0 && f->writable) {
3368
3369 (void) journal_file_warn_btrfs(f);
3370
3371 /* Let's attach the creation time to the journal file, so that the vacuuming code knows the age of this
3372 * file even if the file might end up corrupted one day... Ideally we'd just use the creation time many
3373 * file systems maintain for each file, but the API to query this is very new, hence let's emulate this
3374 * via extended attributes. If extended attributes are not supported we'll just skip this, and rely
3375 * solely on mtime/atime/ctime of the file. */
3376 (void) fd_setcrtime(f->fd, 0);
3377
3378 #if HAVE_GCRYPT
3379 /* Try to load the FSPRG state, and if we can't, then
3380 * just don't do sealing */
3381 if (f->seal) {
3382 r = journal_file_fss_load(f);
3383 if (r < 0)
3384 f->seal = false;
3385 }
3386 #endif
3387
3388 r = journal_file_init_header(f, template);
3389 if (r < 0)
3390 goto fail;
3391
3392 r = journal_file_fstat(f);
3393 if (r < 0)
3394 goto fail;
3395
3396 newly_created = true;
3397 }
3398
3399 if (f->last_stat.st_size < (off_t) HEADER_SIZE_MIN) {
3400 r = -ENODATA;
3401 goto fail;
3402 }
3403
3404 r = mmap_cache_get(f->mmap, f->cache_fd, f->prot, CONTEXT_HEADER, true, 0, PAGE_ALIGN(sizeof(Header)), &f->last_stat, &h, NULL);
3405 if (r == -EINVAL) {
3406 /* Some file systems (jffs2 or p9fs) don't support mmap() properly (or only read-only
3407 * mmap()), and return EINVAL in that case. Let's propagate that as a more recognizable error
3408 * code. */
3409 r = -EAFNOSUPPORT;
3410 goto fail;
3411 }
3412 if (r < 0)
3413 goto fail;
3414
3415 f->header = h;
3416
3417 if (!newly_created) {
3418 set_clear_with_destructor(deferred_closes, journal_file_close);
3419
3420 r = journal_file_verify_header(f);
3421 if (r < 0)
3422 goto fail;
3423 }
3424
3425 #if HAVE_GCRYPT
3426 if (!newly_created && f->writable) {
3427 r = journal_file_fss_load(f);
3428 if (r < 0)
3429 goto fail;
3430 }
3431 #endif
3432
3433 if (f->writable) {
3434 if (metrics) {
3435 journal_default_metrics(metrics, f->fd);
3436 f->metrics = *metrics;
3437 } else if (template)
3438 f->metrics = template->metrics;
3439
3440 r = journal_file_refresh_header(f);
3441 if (r < 0)
3442 goto fail;
3443 }
3444
3445 #if HAVE_GCRYPT
3446 r = journal_file_hmac_setup(f);
3447 if (r < 0)
3448 goto fail;
3449 #endif
3450
3451 if (newly_created) {
3452 r = journal_file_setup_field_hash_table(f);
3453 if (r < 0)
3454 goto fail;
3455
3456 r = journal_file_setup_data_hash_table(f);
3457 if (r < 0)
3458 goto fail;
3459
3460 #if HAVE_GCRYPT
3461 r = journal_file_append_first_tag(f);
3462 if (r < 0)
3463 goto fail;
3464 #endif
3465 }
3466
3467 if (mmap_cache_got_sigbus(f->mmap, f->cache_fd)) {
3468 r = -EIO;
3469 goto fail;
3470 }
3471
3472 if (template && template->post_change_timer) {
3473 r = journal_file_enable_post_change_timer(
3474 f,
3475 sd_event_source_get_event(template->post_change_timer),
3476 template->post_change_timer_period);
3477
3478 if (r < 0)
3479 goto fail;
3480 }
3481
3482 /* The file is opened now successfully, thus we take possession of any passed in fd. */
3483 f->close_fd = true;
3484
3485 *ret = f;
3486 return 0;
3487
3488 fail:
3489 if (f->cache_fd && mmap_cache_got_sigbus(f->mmap, f->cache_fd))
3490 r = -EIO;
3491
3492 (void) journal_file_close(f);
3493
3494 return r;
3495 }
3496
3497 int journal_file_archive(JournalFile *f) {
3498 _cleanup_free_ char *p = NULL;
3499
3500 assert(f);
3501
3502 if (!f->writable)
3503 return -EINVAL;
3504
3505 /* Is this a journal file that was passed to us as fd? If so, we synthesized a path name for it, and we refuse
3506 * rotation, since we don't know the actual path, and couldn't rename the file hence. */
3507 if (path_startswith(f->path, "/proc/self/fd"))
3508 return -EINVAL;
3509
3510 if (!endswith(f->path, ".journal"))
3511 return -EINVAL;
3512
3513 if (asprintf(&p, "%.*s@" SD_ID128_FORMAT_STR "-%016"PRIx64"-%016"PRIx64".journal",
3514 (int) strlen(f->path) - 8, f->path,
3515 SD_ID128_FORMAT_VAL(f->header->seqnum_id),
3516 le64toh(f->header->head_entry_seqnum),
3517 le64toh(f->header->head_entry_realtime)) < 0)
3518 return -ENOMEM;
3519
3520 /* Try to rename the file to the archived version. If the file already was deleted, we'll get ENOENT, let's
3521 * ignore that case. */
3522 if (rename(f->path, p) < 0 && errno != ENOENT)
3523 return -errno;
3524
3525 /* Sync the rename to disk */
3526 (void) fsync_directory_of_file(f->fd);
3527
3528 /* Set as archive so offlining commits w/state=STATE_ARCHIVED. Previously we would set old_file->header->state
3529 * to STATE_ARCHIVED directly here, but journal_file_set_offline() short-circuits when state != STATE_ONLINE,
3530 * which would result in the rotated journal never getting fsync() called before closing. Now we simply queue
3531 * the archive state by setting an archive bit, leaving the state as STATE_ONLINE so proper offlining
3532 * occurs. */
3533 f->archive = true;
3534
3535 /* Currently, btrfs is not very good with out write patterns and fragments heavily. Let's defrag our journal
3536 * files when we archive them */
3537 f->defrag_on_close = true;
3538
3539 return 0;
3540 }
3541
3542 JournalFile* journal_initiate_close(
3543 JournalFile *f,
3544 Set *deferred_closes) {
3545
3546 int r;
3547
3548 assert(f);
3549
3550 if (deferred_closes) {
3551
3552 r = set_put(deferred_closes, f);
3553 if (r < 0)
3554 log_debug_errno(r, "Failed to add file to deferred close set, closing immediately.");
3555 else {
3556 (void) journal_file_set_offline(f, false);
3557 return NULL;
3558 }
3559 }
3560
3561 return journal_file_close(f);
3562 }
3563
3564 int journal_file_rotate(
3565 JournalFile **f,
3566 bool compress,
3567 uint64_t compress_threshold_bytes,
3568 bool seal,
3569 Set *deferred_closes) {
3570
3571 JournalFile *new_file = NULL;
3572 int r;
3573
3574 assert(f);
3575 assert(*f);
3576
3577 r = journal_file_archive(*f);
3578 if (r < 0)
3579 return r;
3580
3581 r = journal_file_open(
3582 -1,
3583 (*f)->path,
3584 (*f)->flags,
3585 (*f)->mode,
3586 compress,
3587 compress_threshold_bytes,
3588 seal,
3589 NULL, /* metrics */
3590 (*f)->mmap,
3591 deferred_closes,
3592 *f, /* template */
3593 &new_file);
3594
3595 journal_initiate_close(*f, deferred_closes);
3596 *f = new_file;
3597
3598 return r;
3599 }
3600
3601 int journal_file_dispose(int dir_fd, const char *fname) {
3602 _cleanup_free_ char *p = NULL;
3603 _cleanup_close_ int fd = -1;
3604
3605 assert(fname);
3606
3607 /* Renames a journal file to *.journal~, i.e. to mark it as corruped or otherwise uncleanly shutdown. Note that
3608 * this is done without looking into the file or changing any of its contents. The idea is that this is called
3609 * whenever something is suspicious and we want to move the file away and make clear that it is not accessed
3610 * for writing anymore. */
3611
3612 if (!endswith(fname, ".journal"))
3613 return -EINVAL;
3614
3615 if (asprintf(&p, "%.*s@%016" PRIx64 "-%016" PRIx64 ".journal~",
3616 (int) strlen(fname) - 8, fname,
3617 now(CLOCK_REALTIME),
3618 random_u64()) < 0)
3619 return -ENOMEM;
3620
3621 if (renameat(dir_fd, fname, dir_fd, p) < 0)
3622 return -errno;
3623
3624 /* btrfs doesn't cope well with our write pattern and fragments heavily. Let's defrag all files we rotate */
3625 fd = openat(dir_fd, p, O_RDONLY|O_CLOEXEC|O_NOCTTY|O_NOFOLLOW);
3626 if (fd < 0)
3627 log_debug_errno(errno, "Failed to open file for defragmentation/FS_NOCOW_FL, ignoring: %m");
3628 else {
3629 (void) chattr_fd(fd, 0, FS_NOCOW_FL, NULL);
3630 (void) btrfs_defrag_fd(fd);
3631 }
3632
3633 return 0;
3634 }
3635
3636 int journal_file_open_reliably(
3637 const char *fname,
3638 int flags,
3639 mode_t mode,
3640 bool compress,
3641 uint64_t compress_threshold_bytes,
3642 bool seal,
3643 JournalMetrics *metrics,
3644 MMapCache *mmap_cache,
3645 Set *deferred_closes,
3646 JournalFile *template,
3647 JournalFile **ret) {
3648
3649 int r;
3650
3651 r = journal_file_open(-1, fname, flags, mode, compress, compress_threshold_bytes, seal, metrics, mmap_cache,
3652 deferred_closes, template, ret);
3653 if (!IN_SET(r,
3654 -EBADMSG, /* Corrupted */
3655 -ENODATA, /* Truncated */
3656 -EHOSTDOWN, /* Other machine */
3657 -EPROTONOSUPPORT, /* Incompatible feature */
3658 -EBUSY, /* Unclean shutdown */
3659 -ESHUTDOWN, /* Already archived */
3660 -EIO, /* IO error, including SIGBUS on mmap */
3661 -EIDRM, /* File has been deleted */
3662 -ETXTBSY)) /* File is from the future */
3663 return r;
3664
3665 if ((flags & O_ACCMODE) == O_RDONLY)
3666 return r;
3667
3668 if (!(flags & O_CREAT))
3669 return r;
3670
3671 if (!endswith(fname, ".journal"))
3672 return r;
3673
3674 /* The file is corrupted. Rotate it away and try it again (but only once) */
3675 log_warning_errno(r, "File %s corrupted or uncleanly shut down, renaming and replacing.", fname);
3676
3677 r = journal_file_dispose(AT_FDCWD, fname);
3678 if (r < 0)
3679 return r;
3680
3681 return journal_file_open(-1, fname, flags, mode, compress, compress_threshold_bytes, seal, metrics, mmap_cache,
3682 deferred_closes, template, ret);
3683 }
3684
3685 int journal_file_copy_entry(JournalFile *from, JournalFile *to, Object *o, uint64_t p) {
3686 uint64_t i, n;
3687 uint64_t q, xor_hash = 0;
3688 int r;
3689 EntryItem *items;
3690 dual_timestamp ts;
3691 const sd_id128_t *boot_id;
3692
3693 assert(from);
3694 assert(to);
3695 assert(o);
3696 assert(p);
3697
3698 if (!to->writable)
3699 return -EPERM;
3700
3701 ts.monotonic = le64toh(o->entry.monotonic);
3702 ts.realtime = le64toh(o->entry.realtime);
3703 boot_id = &o->entry.boot_id;
3704
3705 n = journal_file_entry_n_items(o);
3706 /* alloca() can't take 0, hence let's allocate at least one */
3707 items = newa(EntryItem, MAX(1u, n));
3708
3709 for (i = 0; i < n; i++) {
3710 uint64_t l, h;
3711 le64_t le_hash;
3712 size_t t;
3713 void *data;
3714 Object *u;
3715
3716 q = le64toh(o->entry.items[i].object_offset);
3717 le_hash = o->entry.items[i].hash;
3718
3719 r = journal_file_move_to_object(from, OBJECT_DATA, q, &o);
3720 if (r < 0)
3721 return r;
3722
3723 if (le_hash != o->data.hash)
3724 return -EBADMSG;
3725
3726 l = le64toh(READ_NOW(o->object.size));
3727 if (l < offsetof(Object, data.payload))
3728 return -EBADMSG;
3729
3730 l -= offsetof(Object, data.payload);
3731 t = (size_t) l;
3732
3733 /* We hit the limit on 32bit machines */
3734 if ((uint64_t) t != l)
3735 return -E2BIG;
3736
3737 if (o->object.flags & OBJECT_COMPRESSION_MASK) {
3738 #if HAVE_XZ || HAVE_LZ4
3739 size_t rsize = 0;
3740
3741 r = decompress_blob(o->object.flags & OBJECT_COMPRESSION_MASK,
3742 o->data.payload, l, &from->compress_buffer, &from->compress_buffer_size, &rsize, 0);
3743 if (r < 0)
3744 return r;
3745
3746 data = from->compress_buffer;
3747 l = rsize;
3748 #else
3749 return -EPROTONOSUPPORT;
3750 #endif
3751 } else
3752 data = o->data.payload;
3753
3754 r = journal_file_append_data(to, data, l, &u, &h);
3755 if (r < 0)
3756 return r;
3757
3758 xor_hash ^= le64toh(u->data.hash);
3759 items[i].object_offset = htole64(h);
3760 items[i].hash = u->data.hash;
3761
3762 r = journal_file_move_to_object(from, OBJECT_ENTRY, p, &o);
3763 if (r < 0)
3764 return r;
3765 }
3766
3767 r = journal_file_append_entry_internal(to, &ts, boot_id, xor_hash, items, n,
3768 NULL, NULL, NULL);
3769
3770 if (mmap_cache_got_sigbus(to->mmap, to->cache_fd))
3771 return -EIO;
3772
3773 return r;
3774 }
3775
3776 void journal_reset_metrics(JournalMetrics *m) {
3777 assert(m);
3778
3779 /* Set everything to "pick automatic values". */
3780
3781 *m = (JournalMetrics) {
3782 .min_use = (uint64_t) -1,
3783 .max_use = (uint64_t) -1,
3784 .min_size = (uint64_t) -1,
3785 .max_size = (uint64_t) -1,
3786 .keep_free = (uint64_t) -1,
3787 .n_max_files = (uint64_t) -1,
3788 };
3789 }
3790
3791 void journal_default_metrics(JournalMetrics *m, int fd) {
3792 char a[FORMAT_BYTES_MAX], b[FORMAT_BYTES_MAX], c[FORMAT_BYTES_MAX], d[FORMAT_BYTES_MAX], e[FORMAT_BYTES_MAX];
3793 struct statvfs ss;
3794 uint64_t fs_size = 0;
3795
3796 assert(m);
3797 assert(fd >= 0);
3798
3799 if (fstatvfs(fd, &ss) >= 0)
3800 fs_size = ss.f_frsize * ss.f_blocks;
3801 else
3802 log_debug_errno(errno, "Failed to determine disk size: %m");
3803
3804 if (m->max_use == (uint64_t) -1) {
3805
3806 if (fs_size > 0)
3807 m->max_use = CLAMP(PAGE_ALIGN(fs_size / 10), /* 10% of file system size */
3808 MAX_USE_LOWER, MAX_USE_UPPER);
3809 else
3810 m->max_use = MAX_USE_LOWER;
3811 } else {
3812 m->max_use = PAGE_ALIGN(m->max_use);
3813
3814 if (m->max_use != 0 && m->max_use < JOURNAL_FILE_SIZE_MIN*2)
3815 m->max_use = JOURNAL_FILE_SIZE_MIN*2;
3816 }
3817
3818 if (m->min_use == (uint64_t) -1) {
3819 if (fs_size > 0)
3820 m->min_use = CLAMP(PAGE_ALIGN(fs_size / 50), /* 2% of file system size */
3821 MIN_USE_LOW, MIN_USE_HIGH);
3822 else
3823 m->min_use = MIN_USE_LOW;
3824 }
3825
3826 if (m->min_use > m->max_use)
3827 m->min_use = m->max_use;
3828
3829 if (m->max_size == (uint64_t) -1)
3830 m->max_size = MIN(PAGE_ALIGN(m->max_use / 8), /* 8 chunks */
3831 MAX_SIZE_UPPER);
3832 else
3833 m->max_size = PAGE_ALIGN(m->max_size);
3834
3835 if (m->max_size != 0) {
3836 if (m->max_size < JOURNAL_FILE_SIZE_MIN)
3837 m->max_size = JOURNAL_FILE_SIZE_MIN;
3838
3839 if (m->max_use != 0 && m->max_size*2 > m->max_use)
3840 m->max_use = m->max_size*2;
3841 }
3842
3843 if (m->min_size == (uint64_t) -1)
3844 m->min_size = JOURNAL_FILE_SIZE_MIN;
3845 else
3846 m->min_size = CLAMP(PAGE_ALIGN(m->min_size),
3847 JOURNAL_FILE_SIZE_MIN,
3848 m->max_size ?: UINT64_MAX);
3849
3850 if (m->keep_free == (uint64_t) -1) {
3851 if (fs_size > 0)
3852 m->keep_free = MIN(PAGE_ALIGN(fs_size / 20), /* 5% of file system size */
3853 KEEP_FREE_UPPER);
3854 else
3855 m->keep_free = DEFAULT_KEEP_FREE;
3856 }
3857
3858 if (m->n_max_files == (uint64_t) -1)
3859 m->n_max_files = DEFAULT_N_MAX_FILES;
3860
3861 log_debug("Fixed min_use=%s max_use=%s max_size=%s min_size=%s keep_free=%s n_max_files=%" PRIu64,
3862 format_bytes(a, sizeof(a), m->min_use),
3863 format_bytes(b, sizeof(b), m->max_use),
3864 format_bytes(c, sizeof(c), m->max_size),
3865 format_bytes(d, sizeof(d), m->min_size),
3866 format_bytes(e, sizeof(e), m->keep_free),
3867 m->n_max_files);
3868 }
3869
3870 int journal_file_get_cutoff_realtime_usec(JournalFile *f, usec_t *from, usec_t *to) {
3871 assert(f);
3872 assert(f->header);
3873 assert(from || to);
3874
3875 if (from) {
3876 if (f->header->head_entry_realtime == 0)
3877 return -ENOENT;
3878
3879 *from = le64toh(f->header->head_entry_realtime);
3880 }
3881
3882 if (to) {
3883 if (f->header->tail_entry_realtime == 0)
3884 return -ENOENT;
3885
3886 *to = le64toh(f->header->tail_entry_realtime);
3887 }
3888
3889 return 1;
3890 }
3891
3892 int journal_file_get_cutoff_monotonic_usec(JournalFile *f, sd_id128_t boot_id, usec_t *from, usec_t *to) {
3893 Object *o;
3894 uint64_t p;
3895 int r;
3896
3897 assert(f);
3898 assert(from || to);
3899
3900 r = find_data_object_by_boot_id(f, boot_id, &o, &p);
3901 if (r <= 0)
3902 return r;
3903
3904 if (le64toh(o->data.n_entries) <= 0)
3905 return 0;
3906
3907 if (from) {
3908 r = journal_file_move_to_object(f, OBJECT_ENTRY, le64toh(o->data.entry_offset), &o);
3909 if (r < 0)
3910 return r;
3911
3912 *from = le64toh(o->entry.monotonic);
3913 }
3914
3915 if (to) {
3916 r = journal_file_move_to_object(f, OBJECT_DATA, p, &o);
3917 if (r < 0)
3918 return r;
3919
3920 r = generic_array_get_plus_one(f,
3921 le64toh(o->data.entry_offset),
3922 le64toh(o->data.entry_array_offset),
3923 le64toh(o->data.n_entries)-1,
3924 &o, NULL);
3925 if (r <= 0)
3926 return r;
3927
3928 *to = le64toh(o->entry.monotonic);
3929 }
3930
3931 return 1;
3932 }
3933
3934 bool journal_file_rotate_suggested(JournalFile *f, usec_t max_file_usec) {
3935 assert(f);
3936 assert(f->header);
3937
3938 /* If we gained new header fields we gained new features,
3939 * hence suggest a rotation */
3940 if (le64toh(f->header->header_size) < sizeof(Header)) {
3941 log_debug("%s uses an outdated header, suggesting rotation.", f->path);
3942 return true;
3943 }
3944
3945 /* Let's check if the hash tables grew over a certain fill
3946 * level (75%, borrowing this value from Java's hash table
3947 * implementation), and if so suggest a rotation. To calculate
3948 * the fill level we need the n_data field, which only exists
3949 * in newer versions. */
3950
3951 if (JOURNAL_HEADER_CONTAINS(f->header, n_data))
3952 if (le64toh(f->header->n_data) * 4ULL > (le64toh(f->header->data_hash_table_size) / sizeof(HashItem)) * 3ULL) {
3953 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.",
3954 f->path,
3955 100.0 * (double) le64toh(f->header->n_data) / ((double) (le64toh(f->header->data_hash_table_size) / sizeof(HashItem))),
3956 le64toh(f->header->n_data),
3957 le64toh(f->header->data_hash_table_size) / sizeof(HashItem),
3958 (unsigned long long) f->last_stat.st_size,
3959 f->last_stat.st_size / le64toh(f->header->n_data));
3960 return true;
3961 }
3962
3963 if (JOURNAL_HEADER_CONTAINS(f->header, n_fields))
3964 if (le64toh(f->header->n_fields) * 4ULL > (le64toh(f->header->field_hash_table_size) / sizeof(HashItem)) * 3ULL) {
3965 log_debug("Field hash table of %s has a fill level at %.1f (%"PRIu64" of %"PRIu64" items), suggesting rotation.",
3966 f->path,
3967 100.0 * (double) le64toh(f->header->n_fields) / ((double) (le64toh(f->header->field_hash_table_size) / sizeof(HashItem))),
3968 le64toh(f->header->n_fields),
3969 le64toh(f->header->field_hash_table_size) / sizeof(HashItem));
3970 return true;
3971 }
3972
3973 /* Are the data objects properly indexed by field objects? */
3974 if (JOURNAL_HEADER_CONTAINS(f->header, n_data) &&
3975 JOURNAL_HEADER_CONTAINS(f->header, n_fields) &&
3976 le64toh(f->header->n_data) > 0 &&
3977 le64toh(f->header->n_fields) == 0)
3978 return true;
3979
3980 if (max_file_usec > 0) {
3981 usec_t t, h;
3982
3983 h = le64toh(f->header->head_entry_realtime);
3984 t = now(CLOCK_REALTIME);
3985
3986 if (h > 0 && t > h + max_file_usec)
3987 return true;
3988 }
3989
3990 return false;
3991 }
Unnamed repository; edit this file 'description' to name the repository.