]> git.ipfire.org Git - thirdparty/systemd.git/blame_incremental - src/journal/journal-file.c
projects / thirdparty / systemd.git / blame_incremental
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (non-incremental) | history | HEAD
Merge pull request #5164 from Werkov/ordering-for-_netdev-devices
[thirdparty/systemd.git] / src / journal / journal-file.c
... / ...
CommitLineData
1/***
2 This file is part of systemd.
3
4 Copyright 2011 Lennart Poettering
5
6 systemd is free software; you can redistribute it and/or modify it
7 under the terms of the GNU Lesser General Public License as published by
8 the Free Software Foundation; either version 2.1 of the License, or
9 (at your option) any later version.
10
11 systemd is distributed in the hope that it will be useful, but
12 WITHOUT ANY WARRANTY; without even the implied warranty of
13 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14 Lesser General Public License for more details.
15
16 You should have received a copy of the GNU Lesser General Public License
17 along with systemd; If not, see <http://www.gnu.org/licenses/>.
18***/
19
20#include <errno.h>
21#include <fcntl.h>
22#include <linux/fs.h>
23#include <pthread.h>
24#include <stddef.h>
25#include <sys/mman.h>
26#include <sys/statvfs.h>
27#include <sys/uio.h>
28#include <unistd.h>
29
30#include "alloc-util.h"
31#include "btrfs-util.h"
32#include "chattr-util.h"
33#include "compress.h"
34#include "fd-util.h"
35#include "journal-authenticate.h"
36#include "journal-def.h"
37#include "journal-file.h"
38#include "lookup3.h"
39#include "parse-util.h"
40#include "path-util.h"
41#include "random-util.h"
42#include "sd-event.h"
43#include "set.h"
44#include "string-util.h"
45#include "strv.h"
46#include "xattr-util.h"
47
48#define DEFAULT_DATA_HASH_TABLE_SIZE (2047ULL*sizeof(HashItem))
49#define DEFAULT_FIELD_HASH_TABLE_SIZE (333ULL*sizeof(HashItem))
50
51#define COMPRESSION_SIZE_THRESHOLD (512ULL)
52
53/* This is the minimum journal file size */
54#define JOURNAL_FILE_SIZE_MIN (512ULL*1024ULL) /* 512 KiB */
55
56/* These are the lower and upper bounds if we deduce the max_use value
57 * from the file system size */
58#define DEFAULT_MAX_USE_LOWER (1ULL*1024ULL*1024ULL) /* 1 MiB */
59#define DEFAULT_MAX_USE_UPPER (4ULL*1024ULL*1024ULL*1024ULL) /* 4 GiB */
60
61/* This is the default minimal use limit, how much we'll use even if keep_free suggests otherwise. */
62#define DEFAULT_MIN_USE (1ULL*1024ULL*1024ULL) /* 1 MiB */
63
64/* This is the upper bound if we deduce max_size from max_use */
65#define DEFAULT_MAX_SIZE_UPPER (128ULL*1024ULL*1024ULL) /* 128 MiB */
66
67/* This is the upper bound if we deduce the keep_free value from the
68 * file system size */
69#define DEFAULT_KEEP_FREE_UPPER (4ULL*1024ULL*1024ULL*1024ULL) /* 4 GiB */
70
71/* This is the keep_free value when we can't determine the system
72 * size */
73#define DEFAULT_KEEP_FREE (1024ULL*1024ULL) /* 1 MB */
74
75/* This is the default maximum number of journal files to keep around. */
76#define DEFAULT_N_MAX_FILES (100)
77
78/* n_data was the first entry we added after the initial file format design */
79#define HEADER_SIZE_MIN ALIGN64(offsetof(Header, n_data))
80
81/* How many entries to keep in the entry array chain cache at max */
82#define CHAIN_CACHE_MAX 20
83
84/* How much to increase the journal file size at once each time we allocate something new. */
85#define FILE_SIZE_INCREASE (8ULL*1024ULL*1024ULL) /* 8MB */
86
87/* Reread fstat() of the file for detecting deletions at least this often */
88#define LAST_STAT_REFRESH_USEC (5*USEC_PER_SEC)
89
90/* The mmap context to use for the header we pick as one above the last defined typed */
91#define CONTEXT_HEADER _OBJECT_TYPE_MAX
92
93/* This may be called from a separate thread to prevent blocking the caller for the duration of fsync().
94 * As a result we use atomic operations on f->offline_state for inter-thread communications with
95 * journal_file_set_offline() and journal_file_set_online(). */
96static void journal_file_set_offline_internal(JournalFile *f) {
97 assert(f);
98 assert(f->fd >= 0);
99 assert(f->header);
100
101 for (;;) {
102 switch (f->offline_state) {
103 case OFFLINE_CANCEL:
104 if (!__sync_bool_compare_and_swap(&f->offline_state, OFFLINE_CANCEL, OFFLINE_DONE))
105 continue;
106 return;
107
108 case OFFLINE_AGAIN_FROM_SYNCING:
109 if (!__sync_bool_compare_and_swap(&f->offline_state, OFFLINE_AGAIN_FROM_SYNCING, OFFLINE_SYNCING))
110 continue;
111 break;
112
113 case OFFLINE_AGAIN_FROM_OFFLINING:
114 if (!__sync_bool_compare_and_swap(&f->offline_state, OFFLINE_AGAIN_FROM_OFFLINING, OFFLINE_SYNCING))
115 continue;
116 break;
117
118 case OFFLINE_SYNCING:
119 (void) fsync(f->fd);
120
121 if (!__sync_bool_compare_and_swap(&f->offline_state, OFFLINE_SYNCING, OFFLINE_OFFLINING))
122 continue;
123
124 f->header->state = f->archive ? STATE_ARCHIVED : STATE_OFFLINE;
125 (void) fsync(f->fd);
126 break;
127
128 case OFFLINE_OFFLINING:
129 if (!__sync_bool_compare_and_swap(&f->offline_state, OFFLINE_OFFLINING, OFFLINE_DONE))
130 continue;
131 /* fall through */
132
133 case OFFLINE_DONE:
134 return;
135
136 case OFFLINE_JOINED:
137 log_debug("OFFLINE_JOINED unexpected offline state for journal_file_set_offline_internal()");
138 return;
139 }
140 }
141}
142
143static void * journal_file_set_offline_thread(void *arg) {
144 JournalFile *f = arg;
145
146 journal_file_set_offline_internal(f);
147
148 return NULL;
149}
150
151static int journal_file_set_offline_thread_join(JournalFile *f) {
152 int r;
153
154 assert(f);
155
156 if (f->offline_state == OFFLINE_JOINED)
157 return 0;
158
159 r = pthread_join(f->offline_thread, NULL);
160 if (r)
161 return -r;
162
163 f->offline_state = OFFLINE_JOINED;
164
165 if (mmap_cache_got_sigbus(f->mmap, f->fd))
166 return -EIO;
167
168 return 0;
169}
170
171/* Trigger a restart if the offline thread is mid-flight in a restartable state. */
172static bool journal_file_set_offline_try_restart(JournalFile *f) {
173 for (;;) {
174 switch (f->offline_state) {
175 case OFFLINE_AGAIN_FROM_SYNCING:
176 case OFFLINE_AGAIN_FROM_OFFLINING:
177 return true;
178
179 case OFFLINE_CANCEL:
180 if (!__sync_bool_compare_and_swap(&f->offline_state, OFFLINE_CANCEL, OFFLINE_AGAIN_FROM_SYNCING))
181 continue;
182 return true;
183
184 case OFFLINE_SYNCING:
185 if (!__sync_bool_compare_and_swap(&f->offline_state, OFFLINE_SYNCING, OFFLINE_AGAIN_FROM_SYNCING))
186 continue;
187 return true;
188
189 case OFFLINE_OFFLINING:
190 if (!__sync_bool_compare_and_swap(&f->offline_state, OFFLINE_OFFLINING, OFFLINE_AGAIN_FROM_OFFLINING))
191 continue;
192 return true;
193
194 default:
195 return false;
196 }
197 }
198}
199
200/* Sets a journal offline.
201 *
202 * If wait is false then an offline is dispatched in a separate thread for a
203 * subsequent journal_file_set_offline() or journal_file_set_online() of the
204 * same journal to synchronize with.
205 *
206 * If wait is true, then either an existing offline thread will be restarted
207 * and joined, or if none exists the offline is simply performed in this
208 * context without involving another thread.
209 */
210int journal_file_set_offline(JournalFile *f, bool wait) {
211 bool restarted;
212 int r;
213
214 assert(f);
215
216 if (!f->writable)
217 return -EPERM;
218
219 if (!(f->fd >= 0 && f->header))
220 return -EINVAL;
221
222 /* An offlining journal is implicitly online and may modify f->header->state,
223 * we must also join any potentially lingering offline thread when not online. */
224 if (!journal_file_is_offlining(f) && f->header->state != STATE_ONLINE)
225 return journal_file_set_offline_thread_join(f);
226
227 /* Restart an in-flight offline thread and wait if needed, or join a lingering done one. */
228 restarted = journal_file_set_offline_try_restart(f);
229 if ((restarted && wait) || !restarted) {
230 r = journal_file_set_offline_thread_join(f);
231 if (r < 0)
232 return r;
233 }
234
235 if (restarted)
236 return 0;
237
238 /* Initiate a new offline. */
239 f->offline_state = OFFLINE_SYNCING;
240
241 if (wait) /* Without using a thread if waiting. */
242 journal_file_set_offline_internal(f);
243 else {
244 r = pthread_create(&f->offline_thread, NULL, journal_file_set_offline_thread, f);
245 if (r > 0) {
246 f->offline_state = OFFLINE_JOINED;
247 return -r;
248 }
249 }
250
251 return 0;
252}
253
254static int journal_file_set_online(JournalFile *f) {
255 bool joined = false;
256
257 assert(f);
258
259 if (!f->writable)
260 return -EPERM;
261
262 if (!(f->fd >= 0 && f->header))
263 return -EINVAL;
264
265 while (!joined) {
266 switch (f->offline_state) {
267 case OFFLINE_JOINED:
268 /* No offline thread, no need to wait. */
269 joined = true;
270 break;
271
272 case OFFLINE_SYNCING:
273 if (!__sync_bool_compare_and_swap(&f->offline_state, OFFLINE_SYNCING, OFFLINE_CANCEL))
274 continue;
275 /* Canceled syncing prior to offlining, no need to wait. */
276 break;
277
278 case OFFLINE_AGAIN_FROM_SYNCING:
279 if (!__sync_bool_compare_and_swap(&f->offline_state, OFFLINE_AGAIN_FROM_SYNCING, OFFLINE_CANCEL))
280 continue;
281 /* Canceled restart from syncing, no need to wait. */
282 break;
283
284 case OFFLINE_AGAIN_FROM_OFFLINING:
285 if (!__sync_bool_compare_and_swap(&f->offline_state, OFFLINE_AGAIN_FROM_OFFLINING, OFFLINE_CANCEL))
286 continue;
287 /* Canceled restart from offlining, must wait for offlining to complete however. */
288
289 /* fall through */
290 default: {
291 int r;
292
293 r = journal_file_set_offline_thread_join(f);
294 if (r < 0)
295 return r;
296
297 joined = true;
298 break;
299 }
300 }
301 }
302
303 if (mmap_cache_got_sigbus(f->mmap, f->fd))
304 return -EIO;
305
306 switch (f->header->state) {
307 case STATE_ONLINE:
308 return 0;
309
310 case STATE_OFFLINE:
311 f->header->state = STATE_ONLINE;
312 (void) fsync(f->fd);
313 return 0;
314
315 default:
316 return -EINVAL;
317 }
318}
319
320bool journal_file_is_offlining(JournalFile *f) {
321 assert(f);
322
323 __sync_synchronize();
324
325 if (f->offline_state == OFFLINE_DONE ||
326 f->offline_state == OFFLINE_JOINED)
327 return false;
328
329 return true;
330}
331
332JournalFile* journal_file_close(JournalFile *f) {
333 assert(f);
334
335#ifdef HAVE_GCRYPT
336 /* Write the final tag */
337 if (f->seal && f->writable) {
338 int r;
339
340 r = journal_file_append_tag(f);
341 if (r < 0)
342 log_error_errno(r, "Failed to append tag when closing journal: %m");
343 }
344#endif
345
346 if (f->post_change_timer) {
347 int enabled;
348
349 if (sd_event_source_get_enabled(f->post_change_timer, &enabled) >= 0)
350 if (enabled == SD_EVENT_ONESHOT)
351 journal_file_post_change(f);
352
353 (void) sd_event_source_set_enabled(f->post_change_timer, SD_EVENT_OFF);
354 sd_event_source_unref(f->post_change_timer);
355 }
356
357 journal_file_set_offline(f, true);
358
359 if (f->mmap && f->fd >= 0)
360 mmap_cache_close_fd(f->mmap, f->fd);
361
362 if (f->fd >= 0 && f->defrag_on_close) {
363
364 /* Be friendly to btrfs: turn COW back on again now,
365 * and defragment the file. We won't write to the file
366 * ever again, hence remove all fragmentation, and
367 * reenable all the good bits COW usually provides
368 * (such as data checksumming). */
369
370 (void) chattr_fd(f->fd, 0, FS_NOCOW_FL);
371 (void) btrfs_defrag_fd(f->fd);
372 }
373
374 if (f->close_fd)
375 safe_close(f->fd);
376 free(f->path);
377
378 mmap_cache_unref(f->mmap);
379
380 ordered_hashmap_free_free(f->chain_cache);
381
382#if defined(HAVE_XZ) || defined(HAVE_LZ4)
383 free(f->compress_buffer);
384#endif
385
386#ifdef HAVE_GCRYPT
387 if (f->fss_file)
388 munmap(f->fss_file, PAGE_ALIGN(f->fss_file_size));
389 else
390 free(f->fsprg_state);
391
392 free(f->fsprg_seed);
393
394 if (f->hmac)
395 gcry_md_close(f->hmac);
396#endif
397
398 return mfree(f);
399}
400
401void journal_file_close_set(Set *s) {
402 JournalFile *f;
403
404 assert(s);
405
406 while ((f = set_steal_first(s)))
407 (void) journal_file_close(f);
408}
409
410static 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
447static int fsync_directory_of_file(int fd) {
448 _cleanup_free_ char *path = NULL, *dn = NULL;
449 _cleanup_close_ int dfd = -1;
450 struct stat st;
451 int r;
452
453 if (fstat(fd, &st) < 0)
454 return -errno;
455
456 if (!S_ISREG(st.st_mode))
457 return -EBADFD;
458
459 r = fd_get_path(fd, &path);
460 if (r < 0)
461 return r;
462
463 if (!path_is_absolute(path))
464 return -EINVAL;
465
466 dn = dirname_malloc(path);
467 if (!dn)
468 return -ENOMEM;
469
470 dfd = open(dn, O_RDONLY|O_CLOEXEC|O_DIRECTORY);
471 if (dfd < 0)
472 return -errno;
473
474 if (fsync(dfd) < 0)
475 return -errno;
476
477 return 0;
478}
479
480static int journal_file_refresh_header(JournalFile *f) {
481 sd_id128_t boot_id;
482 int r;
483
484 assert(f);
485 assert(f->header);
486
487 r = sd_id128_get_machine(&f->header->machine_id);
488 if (r < 0)
489 return r;
490
491 r = sd_id128_get_boot(&boot_id);
492 if (r < 0)
493 return r;
494
495 if (sd_id128_equal(boot_id, f->header->boot_id))
496 f->tail_entry_monotonic_valid = true;
497
498 f->header->boot_id = boot_id;
499
500 r = journal_file_set_online(f);
501
502 /* Sync the online state to disk */
503 (void) fsync(f->fd);
504
505 /* We likely just created a new file, also sync the directory this file is located in. */
506 (void) fsync_directory_of_file(f->fd);
507
508 return r;
509}
510
511static bool warn_wrong_flags(const JournalFile *f, bool compatible) {
512 const uint32_t any = compatible ? HEADER_COMPATIBLE_ANY : HEADER_INCOMPATIBLE_ANY,
513 supported = compatible ? HEADER_COMPATIBLE_SUPPORTED : HEADER_INCOMPATIBLE_SUPPORTED;
514 const char *type = compatible ? "compatible" : "incompatible";
515 uint32_t flags;
516
517 flags = le32toh(compatible ? f->header->compatible_flags : f->header->incompatible_flags);
518
519 if (flags & ~supported) {
520 if (flags & ~any)
521 log_debug("Journal file %s has unknown %s flags 0x%"PRIx32,
522 f->path, type, flags & ~any);
523 flags = (flags & any) & ~supported;
524 if (flags) {
525 const char* strv[3];
526 unsigned n = 0;
527 _cleanup_free_ char *t = NULL;
528
529 if (compatible && (flags & HEADER_COMPATIBLE_SEALED))
530 strv[n++] = "sealed";
531 if (!compatible && (flags & HEADER_INCOMPATIBLE_COMPRESSED_XZ))
532 strv[n++] = "xz-compressed";
533 if (!compatible && (flags & HEADER_INCOMPATIBLE_COMPRESSED_LZ4))
534 strv[n++] = "lz4-compressed";
535 strv[n] = NULL;
536 assert(n < ELEMENTSOF(strv));
537
538 t = strv_join((char**) strv, ", ");
539 log_debug("Journal file %s uses %s %s %s disabled at compilation time.",
540 f->path, type, n > 1 ? "flags" : "flag", strnull(t));
541 }
542 return true;
543 }
544
545 return false;
546}
547
548static int journal_file_verify_header(JournalFile *f) {
549 uint64_t arena_size, header_size;
550
551 assert(f);
552 assert(f->header);
553
554 if (memcmp(f->header->signature, HEADER_SIGNATURE, 8))
555 return -EBADMSG;
556
557 /* In both read and write mode we refuse to open files with incompatible
558 * flags we don't know. */
559 if (warn_wrong_flags(f, false))
560 return -EPROTONOSUPPORT;
561
562 /* When open for writing we refuse to open files with compatible flags, too. */
563 if (f->writable && warn_wrong_flags(f, true))
564 return -EPROTONOSUPPORT;
565
566 if (f->header->state >= _STATE_MAX)
567 return -EBADMSG;
568
569 header_size = le64toh(f->header->header_size);
570
571 /* The first addition was n_data, so check that we are at least this large */
572 if (header_size < HEADER_SIZE_MIN)
573 return -EBADMSG;
574
575 if (JOURNAL_HEADER_SEALED(f->header) && !JOURNAL_HEADER_CONTAINS(f->header, n_entry_arrays))
576 return -EBADMSG;
577
578 arena_size = le64toh(f->header->arena_size);
579
580 if (UINT64_MAX - header_size < arena_size || header_size + arena_size > (uint64_t) f->last_stat.st_size)
581 return -ENODATA;
582
583 if (le64toh(f->header->tail_object_offset) > header_size + arena_size)
584 return -ENODATA;
585
586 if (!VALID64(le64toh(f->header->data_hash_table_offset)) ||
587 !VALID64(le64toh(f->header->field_hash_table_offset)) ||
588 !VALID64(le64toh(f->header->tail_object_offset)) ||
589 !VALID64(le64toh(f->header->entry_array_offset)))
590 return -ENODATA;
591
592 if (f->writable) {
593 sd_id128_t machine_id;
594 uint8_t state;
595 int r;
596
597 r = sd_id128_get_machine(&machine_id);
598 if (r < 0)
599 return r;
600
601 if (!sd_id128_equal(machine_id, f->header->machine_id))
602 return -EHOSTDOWN;
603
604 state = f->header->state;
605
606 if (state == STATE_ARCHIVED)
607 return -ESHUTDOWN; /* Already archived */
608 else if (state == STATE_ONLINE) {
609 log_debug("Journal file %s is already online. Assuming unclean closing.", f->path);
610 return -EBUSY;
611 } else if (state != STATE_OFFLINE) {
612 log_debug("Journal file %s has unknown state %i.", f->path, state);
613 return -EBUSY;
614 }
615
616 /* Don't permit appending to files from the future. Because otherwise the realtime timestamps wouldn't
617 * be strictly ordered in the entries in the file anymore, and we can't have that since it breaks
618 * bisection. */
619 if (le64toh(f->header->tail_entry_realtime) > now(CLOCK_REALTIME)) {
620 log_debug("Journal file %s is from the future, refusing to append new data to it that'd be older.", f->path);
621 return -ETXTBSY;
622 }
623 }
624
625 f->compress_xz = JOURNAL_HEADER_COMPRESSED_XZ(f->header);
626 f->compress_lz4 = JOURNAL_HEADER_COMPRESSED_LZ4(f->header);
627
628 f->seal = JOURNAL_HEADER_SEALED(f->header);
629
630 return 0;
631}
632
633static int journal_file_fstat(JournalFile *f) {
634 assert(f);
635 assert(f->fd >= 0);
636
637 if (fstat(f->fd, &f->last_stat) < 0)
638 return -errno;
639
640 f->last_stat_usec = now(CLOCK_MONOTONIC);
641
642 /* Refuse appending to files that are already deleted */
643 if (f->last_stat.st_nlink <= 0)
644 return -EIDRM;
645
646 return 0;
647}
648
649static int journal_file_allocate(JournalFile *f, uint64_t offset, uint64_t size) {
650 uint64_t old_size, new_size;
651 int r;
652
653 assert(f);
654 assert(f->header);
655
656 /* We assume that this file is not sparse, and we know that
657 * for sure, since we always call posix_fallocate()
658 * ourselves */
659
660 if (mmap_cache_got_sigbus(f->mmap, f->fd))
661 return -EIO;
662
663 old_size =
664 le64toh(f->header->header_size) +
665 le64toh(f->header->arena_size);
666
667 new_size = PAGE_ALIGN(offset + size);
668 if (new_size < le64toh(f->header->header_size))
669 new_size = le64toh(f->header->header_size);
670
671 if (new_size <= old_size) {
672
673 /* We already pre-allocated enough space, but before
674 * we write to it, let's check with fstat() if the
675 * file got deleted, in order make sure we don't throw
676 * away the data immediately. Don't check fstat() for
677 * all writes though, but only once ever 10s. */
678
679 if (f->last_stat_usec + LAST_STAT_REFRESH_USEC > now(CLOCK_MONOTONIC))
680 return 0;
681
682 return journal_file_fstat(f);
683 }
684
685 /* Allocate more space. */
686
687 if (f->metrics.max_size > 0 && new_size > f->metrics.max_size)
688 return -E2BIG;
689
690 if (new_size > f->metrics.min_size && f->metrics.keep_free > 0) {
691 struct statvfs svfs;
692
693 if (fstatvfs(f->fd, &svfs) >= 0) {
694 uint64_t available;
695
696 available = LESS_BY((uint64_t) svfs.f_bfree * (uint64_t) svfs.f_bsize, f->metrics.keep_free);
697
698 if (new_size - old_size > available)
699 return -E2BIG;
700 }
701 }
702
703 /* Increase by larger blocks at once */
704 new_size = ((new_size+FILE_SIZE_INCREASE-1) / FILE_SIZE_INCREASE) * FILE_SIZE_INCREASE;
705 if (f->metrics.max_size > 0 && new_size > f->metrics.max_size)
706 new_size = f->metrics.max_size;
707
708 /* Note that the glibc fallocate() fallback is very
709 inefficient, hence we try to minimize the allocation area
710 as we can. */
711 r = posix_fallocate(f->fd, old_size, new_size - old_size);
712 if (r != 0)
713 return -r;
714
715 f->header->arena_size = htole64(new_size - le64toh(f->header->header_size));
716
717 return journal_file_fstat(f);
718}
719
720static unsigned type_to_context(ObjectType type) {
721 /* One context for each type, plus one catch-all for the rest */
722 assert_cc(_OBJECT_TYPE_MAX <= MMAP_CACHE_MAX_CONTEXTS);
723 assert_cc(CONTEXT_HEADER < MMAP_CACHE_MAX_CONTEXTS);
724 return type > OBJECT_UNUSED && type < _OBJECT_TYPE_MAX ? type : 0;
725}
726
727static int journal_file_move_to(JournalFile *f, ObjectType type, bool keep_always, uint64_t offset, uint64_t size, void **ret) {
728 int r;
729
730 assert(f);
731 assert(ret);
732
733 if (size <= 0)
734 return -EINVAL;
735
736 /* Avoid SIGBUS on invalid accesses */
737 if (offset + size > (uint64_t) f->last_stat.st_size) {
738 /* Hmm, out of range? Let's refresh the fstat() data
739 * first, before we trust that check. */
740
741 r = journal_file_fstat(f);
742 if (r < 0)
743 return r;
744
745 if (offset + size > (uint64_t) f->last_stat.st_size)
746 return -EADDRNOTAVAIL;
747 }
748
749 return mmap_cache_get(f->mmap, f->fd, f->prot, type_to_context(type), keep_always, offset, size, &f->last_stat, ret);
750}
751
752static uint64_t minimum_header_size(Object *o) {
753
754 static const uint64_t table[] = {
755 [OBJECT_DATA] = sizeof(DataObject),
756 [OBJECT_FIELD] = sizeof(FieldObject),
757 [OBJECT_ENTRY] = sizeof(EntryObject),
758 [OBJECT_DATA_HASH_TABLE] = sizeof(HashTableObject),
759 [OBJECT_FIELD_HASH_TABLE] = sizeof(HashTableObject),
760 [OBJECT_ENTRY_ARRAY] = sizeof(EntryArrayObject),
761 [OBJECT_TAG] = sizeof(TagObject),
762 };
763
764 if (o->object.type >= ELEMENTSOF(table) || table[o->object.type] <= 0)
765 return sizeof(ObjectHeader);
766
767 return table[o->object.type];
768}
769
770int journal_file_move_to_object(JournalFile *f, ObjectType type, uint64_t offset, Object **ret) {
771 int r;
772 void *t;
773 Object *o;
774 uint64_t s;
775
776 assert(f);
777 assert(ret);
778
779 /* Objects may only be located at multiple of 64 bit */
780 if (!VALID64(offset)) {
781 log_debug("Attempt to move to object at non-64bit boundary: %" PRIu64, offset);
782 return -EBADMSG;
783 }
784
785 /* Object may not be located in the file header */
786 if (offset < le64toh(f->header->header_size)) {
787 log_debug("Attempt to move to object located in file header: %" PRIu64, offset);
788 return -EBADMSG;
789 }
790
791 r = journal_file_move_to(f, type, false, offset, sizeof(ObjectHeader), &t);
792 if (r < 0)
793 return r;
794
795 o = (Object*) t;
796 s = le64toh(o->object.size);
797
798 if (s == 0) {
799 log_debug("Attempt to move to uninitialized object: %" PRIu64, offset);
800 return -EBADMSG;
801 }
802 if (s < sizeof(ObjectHeader)) {
803 log_debug("Attempt to move to overly short object: %" PRIu64, offset);
804 return -EBADMSG;
805 }
806
807 if (o->object.type <= OBJECT_UNUSED) {
808 log_debug("Attempt to move to object with invalid type: %" PRIu64, offset);
809 return -EBADMSG;
810 }
811
812 if (s < minimum_header_size(o)) {
813 log_debug("Attempt to move to truncated object: %" PRIu64, offset);
814 return -EBADMSG;
815 }
816
817 if (type > OBJECT_UNUSED && o->object.type != type) {
818 log_debug("Attempt to move to object of unexpected type: %" PRIu64, offset);
819 return -EBADMSG;
820 }
821
822 if (s > sizeof(ObjectHeader)) {
823 r = journal_file_move_to(f, type, false, offset, s, &t);
824 if (r < 0)
825 return r;
826
827 o = (Object*) t;
828 }
829
830 *ret = o;
831 return 0;
832}
833
834static uint64_t journal_file_entry_seqnum(JournalFile *f, uint64_t *seqnum) {
835 uint64_t r;
836
837 assert(f);
838 assert(f->header);
839
840 r = le64toh(f->header->tail_entry_seqnum) + 1;
841
842 if (seqnum) {
843 /* If an external seqnum counter was passed, we update
844 * both the local and the external one, and set it to
845 * the maximum of both */
846
847 if (*seqnum + 1 > r)
848 r = *seqnum + 1;
849
850 *seqnum = r;
851 }
852
853 f->header->tail_entry_seqnum = htole64(r);
854
855 if (f->header->head_entry_seqnum == 0)
856 f->header->head_entry_seqnum = htole64(r);
857
858 return r;
859}
860
861int journal_file_append_object(JournalFile *f, ObjectType type, uint64_t size, Object **ret, uint64_t *offset) {
862 int r;
863 uint64_t p;
864 Object *tail, *o;
865 void *t;
866
867 assert(f);
868 assert(f->header);
869 assert(type > OBJECT_UNUSED && type < _OBJECT_TYPE_MAX);
870 assert(size >= sizeof(ObjectHeader));
871 assert(offset);
872 assert(ret);
873
874 r = journal_file_set_online(f);
875 if (r < 0)
876 return r;
877
878 p = le64toh(f->header->tail_object_offset);
879 if (p == 0)
880 p = le64toh(f->header->header_size);
881 else {
882 r = journal_file_move_to_object(f, OBJECT_UNUSED, p, &tail);
883 if (r < 0)
884 return r;
885
886 p += ALIGN64(le64toh(tail->object.size));
887 }
888
889 r = journal_file_allocate(f, p, size);
890 if (r < 0)
891 return r;
892
893 r = journal_file_move_to(f, type, false, p, size, &t);
894 if (r < 0)
895 return r;
896
897 o = (Object*) t;
898
899 zero(o->object);
900 o->object.type = type;
901 o->object.size = htole64(size);
902
903 f->header->tail_object_offset = htole64(p);
904 f->header->n_objects = htole64(le64toh(f->header->n_objects) + 1);
905
906 *ret = o;
907 *offset = p;
908
909 return 0;
910}
911
912static int journal_file_setup_data_hash_table(JournalFile *f) {
913 uint64_t s, p;
914 Object *o;
915 int r;
916
917 assert(f);
918 assert(f->header);
919
920 /* We estimate that we need 1 hash table entry per 768 bytes
921 of journal file and we want to make sure we never get
922 beyond 75% fill level. Calculate the hash table size for
923 the maximum file size based on these metrics. */
924
925 s = (f->metrics.max_size * 4 / 768 / 3) * sizeof(HashItem);
926 if (s < DEFAULT_DATA_HASH_TABLE_SIZE)
927 s = DEFAULT_DATA_HASH_TABLE_SIZE;
928
929 log_debug("Reserving %"PRIu64" entries in hash table.", s / sizeof(HashItem));
930
931 r = journal_file_append_object(f,
932 OBJECT_DATA_HASH_TABLE,
933 offsetof(Object, hash_table.items) + s,
934 &o, &p);
935 if (r < 0)
936 return r;
937
938 memzero(o->hash_table.items, s);
939
940 f->header->data_hash_table_offset = htole64(p + offsetof(Object, hash_table.items));
941 f->header->data_hash_table_size = htole64(s);
942
943 return 0;
944}
945
946static int journal_file_setup_field_hash_table(JournalFile *f) {
947 uint64_t s, p;
948 Object *o;
949 int r;
950
951 assert(f);
952 assert(f->header);
953
954 /* We use a fixed size hash table for the fields as this
955 * number should grow very slowly only */
956
957 s = DEFAULT_FIELD_HASH_TABLE_SIZE;
958 r = journal_file_append_object(f,
959 OBJECT_FIELD_HASH_TABLE,
960 offsetof(Object, hash_table.items) + s,
961 &o, &p);
962 if (r < 0)
963 return r;
964
965 memzero(o->hash_table.items, s);
966
967 f->header->field_hash_table_offset = htole64(p + offsetof(Object, hash_table.items));
968 f->header->field_hash_table_size = htole64(s);
969
970 return 0;
971}
972
973int journal_file_map_data_hash_table(JournalFile *f) {
974 uint64_t s, p;
975 void *t;
976 int r;
977
978 assert(f);
979 assert(f->header);
980
981 if (f->data_hash_table)
982 return 0;
983
984 p = le64toh(f->header->data_hash_table_offset);
985 s = le64toh(f->header->data_hash_table_size);
986
987 r = journal_file_move_to(f,
988 OBJECT_DATA_HASH_TABLE,
989 true,
990 p, s,
991 &t);
992 if (r < 0)
993 return r;
994
995 f->data_hash_table = t;
996 return 0;
997}
998
999int journal_file_map_field_hash_table(JournalFile *f) {
1000 uint64_t s, p;
1001 void *t;
1002 int r;
1003
1004 assert(f);
1005 assert(f->header);
1006
1007 if (f->field_hash_table)
1008 return 0;
1009
1010 p = le64toh(f->header->field_hash_table_offset);
1011 s = le64toh(f->header->field_hash_table_size);
1012
1013 r = journal_file_move_to(f,
1014 OBJECT_FIELD_HASH_TABLE,
1015 true,
1016 p, s,
1017 &t);
1018 if (r < 0)
1019 return r;
1020
1021 f->field_hash_table = t;
1022 return 0;
1023}
1024
1025static int journal_file_link_field(
1026 JournalFile *f,
1027 Object *o,
1028 uint64_t offset,
1029 uint64_t hash) {
1030
1031 uint64_t p, h, m;
1032 int r;
1033
1034 assert(f);
1035 assert(f->header);
1036 assert(f->field_hash_table);
1037 assert(o);
1038 assert(offset > 0);
1039
1040 if (o->object.type != OBJECT_FIELD)
1041 return -EINVAL;
1042
1043 m = le64toh(f->header->field_hash_table_size) / sizeof(HashItem);
1044 if (m <= 0)
1045 return -EBADMSG;
1046
1047 /* This might alter the window we are looking at */
1048 o->field.next_hash_offset = o->field.head_data_offset = 0;
1049
1050 h = hash % m;
1051 p = le64toh(f->field_hash_table[h].tail_hash_offset);
1052 if (p == 0)
1053 f->field_hash_table[h].head_hash_offset = htole64(offset);
1054 else {
1055 r = journal_file_move_to_object(f, OBJECT_FIELD, p, &o);
1056 if (r < 0)
1057 return r;
1058
1059 o->field.next_hash_offset = htole64(offset);
1060 }
1061
1062 f->field_hash_table[h].tail_hash_offset = htole64(offset);
1063
1064 if (JOURNAL_HEADER_CONTAINS(f->header, n_fields))
1065 f->header->n_fields = htole64(le64toh(f->header->n_fields) + 1);
1066
1067 return 0;
1068}
1069
1070static int journal_file_link_data(
1071 JournalFile *f,
1072 Object *o,
1073 uint64_t offset,
1074 uint64_t hash) {
1075
1076 uint64_t p, h, m;
1077 int r;
1078
1079 assert(f);
1080 assert(f->header);
1081 assert(f->data_hash_table);
1082 assert(o);
1083 assert(offset > 0);
1084
1085 if (o->object.type != OBJECT_DATA)
1086 return -EINVAL;
1087
1088 m = le64toh(f->header->data_hash_table_size) / sizeof(HashItem);
1089 if (m <= 0)
1090 return -EBADMSG;
1091
1092 /* This might alter the window we are looking at */
1093 o->data.next_hash_offset = o->data.next_field_offset = 0;
1094 o->data.entry_offset = o->data.entry_array_offset = 0;
1095 o->data.n_entries = 0;
1096
1097 h = hash % m;
1098 p = le64toh(f->data_hash_table[h].tail_hash_offset);
1099 if (p == 0)
1100 /* Only entry in the hash table is easy */
1101 f->data_hash_table[h].head_hash_offset = htole64(offset);
1102 else {
1103 /* Move back to the previous data object, to patch in
1104 * pointer */
1105
1106 r = journal_file_move_to_object(f, OBJECT_DATA, p, &o);
1107 if (r < 0)
1108 return r;
1109
1110 o->data.next_hash_offset = htole64(offset);
1111 }
1112
1113 f->data_hash_table[h].tail_hash_offset = htole64(offset);
1114
1115 if (JOURNAL_HEADER_CONTAINS(f->header, n_data))
1116 f->header->n_data = htole64(le64toh(f->header->n_data) + 1);
1117
1118 return 0;
1119}
1120
1121int journal_file_find_field_object_with_hash(
1122 JournalFile *f,
1123 const void *field, uint64_t size, uint64_t hash,
1124 Object **ret, uint64_t *offset) {
1125
1126 uint64_t p, osize, h, m;
1127 int r;
1128
1129 assert(f);
1130 assert(f->header);
1131 assert(field && size > 0);
1132
1133 /* If the field hash table is empty, we can't find anything */
1134 if (le64toh(f->header->field_hash_table_size) <= 0)
1135 return 0;
1136
1137 /* Map the field hash table, if it isn't mapped yet. */
1138 r = journal_file_map_field_hash_table(f);
1139 if (r < 0)
1140 return r;
1141
1142 osize = offsetof(Object, field.payload) + size;
1143
1144 m = le64toh(f->header->field_hash_table_size) / sizeof(HashItem);
1145 if (m <= 0)
1146 return -EBADMSG;
1147
1148 h = hash % m;
1149 p = le64toh(f->field_hash_table[h].head_hash_offset);
1150
1151 while (p > 0) {
1152 Object *o;
1153
1154 r = journal_file_move_to_object(f, OBJECT_FIELD, p, &o);
1155 if (r < 0)
1156 return r;
1157
1158 if (le64toh(o->field.hash) == hash &&
1159 le64toh(o->object.size) == osize &&
1160 memcmp(o->field.payload, field, size) == 0) {
1161
1162 if (ret)
1163 *ret = o;
1164 if (offset)
1165 *offset = p;
1166
1167 return 1;
1168 }
1169
1170 p = le64toh(o->field.next_hash_offset);
1171 }
1172
1173 return 0;
1174}
1175
1176int journal_file_find_field_object(
1177 JournalFile *f,
1178 const void *field, uint64_t size,
1179 Object **ret, uint64_t *offset) {
1180
1181 uint64_t hash;
1182
1183 assert(f);
1184 assert(field && size > 0);
1185
1186 hash = hash64(field, size);
1187
1188 return journal_file_find_field_object_with_hash(f,
1189 field, size, hash,
1190 ret, offset);
1191}
1192
1193int journal_file_find_data_object_with_hash(
1194 JournalFile *f,
1195 const void *data, uint64_t size, uint64_t hash,
1196 Object **ret, uint64_t *offset) {
1197
1198 uint64_t p, osize, h, m;
1199 int r;
1200
1201 assert(f);
1202 assert(f->header);
1203 assert(data || size == 0);
1204
1205 /* If there's no data hash table, then there's no entry. */
1206 if (le64toh(f->header->data_hash_table_size) <= 0)
1207 return 0;
1208
1209 /* Map the data hash table, if it isn't mapped yet. */
1210 r = journal_file_map_data_hash_table(f);
1211 if (r < 0)
1212 return r;
1213
1214 osize = offsetof(Object, data.payload) + size;
1215
1216 m = le64toh(f->header->data_hash_table_size) / sizeof(HashItem);
1217 if (m <= 0)
1218 return -EBADMSG;
1219
1220 h = hash % m;
1221 p = le64toh(f->data_hash_table[h].head_hash_offset);
1222
1223 while (p > 0) {
1224 Object *o;
1225
1226 r = journal_file_move_to_object(f, OBJECT_DATA, p, &o);
1227 if (r < 0)
1228 return r;
1229
1230 if (le64toh(o->data.hash) != hash)
1231 goto next;
1232
1233 if (o->object.flags & OBJECT_COMPRESSION_MASK) {
1234#if defined(HAVE_XZ) || defined(HAVE_LZ4)
1235 uint64_t l;
1236 size_t rsize = 0;
1237
1238 l = le64toh(o->object.size);
1239 if (l <= offsetof(Object, data.payload))
1240 return -EBADMSG;
1241
1242 l -= offsetof(Object, data.payload);
1243
1244 r = decompress_blob(o->object.flags & OBJECT_COMPRESSION_MASK,
1245 o->data.payload, l, &f->compress_buffer, &f->compress_buffer_size, &rsize, 0);
1246 if (r < 0)
1247 return r;
1248
1249 if (rsize == size &&
1250 memcmp(f->compress_buffer, data, size) == 0) {
1251
1252 if (ret)
1253 *ret = o;
1254
1255 if (offset)
1256 *offset = p;
1257
1258 return 1;
1259 }
1260#else
1261 return -EPROTONOSUPPORT;
1262#endif
1263 } else if (le64toh(o->object.size) == osize &&
1264 memcmp(o->data.payload, data, size) == 0) {
1265
1266 if (ret)
1267 *ret = o;
1268
1269 if (offset)
1270 *offset = p;
1271
1272 return 1;
1273 }
1274
1275 next:
1276 p = le64toh(o->data.next_hash_offset);
1277 }
1278
1279 return 0;
1280}
1281
1282int journal_file_find_data_object(
1283 JournalFile *f,
1284 const void *data, uint64_t size,
1285 Object **ret, uint64_t *offset) {
1286
1287 uint64_t hash;
1288
1289 assert(f);
1290 assert(data || size == 0);
1291
1292 hash = hash64(data, size);
1293
1294 return journal_file_find_data_object_with_hash(f,
1295 data, size, hash,
1296 ret, offset);
1297}
1298
1299static int journal_file_append_field(
1300 JournalFile *f,
1301 const void *field, uint64_t size,
1302 Object **ret, uint64_t *offset) {
1303
1304 uint64_t hash, p;
1305 uint64_t osize;
1306 Object *o;
1307 int r;
1308
1309 assert(f);
1310 assert(field && size > 0);
1311
1312 hash = hash64(field, size);
1313
1314 r = journal_file_find_field_object_with_hash(f, field, size, hash, &o, &p);
1315 if (r < 0)
1316 return r;
1317 else if (r > 0) {
1318
1319 if (ret)
1320 *ret = o;
1321
1322 if (offset)
1323 *offset = p;
1324
1325 return 0;
1326 }
1327
1328 osize = offsetof(Object, field.payload) + size;
1329 r = journal_file_append_object(f, OBJECT_FIELD, osize, &o, &p);
1330 if (r < 0)
1331 return r;
1332
1333 o->field.hash = htole64(hash);
1334 memcpy(o->field.payload, field, size);
1335
1336 r = journal_file_link_field(f, o, p, hash);
1337 if (r < 0)
1338 return r;
1339
1340 /* The linking might have altered the window, so let's
1341 * refresh our pointer */
1342 r = journal_file_move_to_object(f, OBJECT_FIELD, p, &o);
1343 if (r < 0)
1344 return r;
1345
1346#ifdef HAVE_GCRYPT
1347 r = journal_file_hmac_put_object(f, OBJECT_FIELD, o, p);
1348 if (r < 0)
1349 return r;
1350#endif
1351
1352 if (ret)
1353 *ret = o;
1354
1355 if (offset)
1356 *offset = p;
1357
1358 return 0;
1359}
1360
1361static int journal_file_append_data(
1362 JournalFile *f,
1363 const void *data, uint64_t size,
1364 Object **ret, uint64_t *offset) {
1365
1366 uint64_t hash, p;
1367 uint64_t osize;
1368 Object *o;
1369 int r, compression = 0;
1370 const void *eq;
1371
1372 assert(f);
1373 assert(data || size == 0);
1374
1375 hash = hash64(data, size);
1376
1377 r = journal_file_find_data_object_with_hash(f, data, size, hash, &o, &p);
1378 if (r < 0)
1379 return r;
1380 if (r > 0) {
1381
1382 if (ret)
1383 *ret = o;
1384
1385 if (offset)
1386 *offset = p;
1387
1388 return 0;
1389 }
1390
1391 osize = offsetof(Object, data.payload) + size;
1392 r = journal_file_append_object(f, OBJECT_DATA, osize, &o, &p);
1393 if (r < 0)
1394 return r;
1395
1396 o->data.hash = htole64(hash);
1397
1398#if defined(HAVE_XZ) || defined(HAVE_LZ4)
1399 if (JOURNAL_FILE_COMPRESS(f) && size >= COMPRESSION_SIZE_THRESHOLD) {
1400 size_t rsize = 0;
1401
1402 compression = compress_blob(data, size, o->data.payload, size - 1, &rsize);
1403
1404 if (compression >= 0) {
1405 o->object.size = htole64(offsetof(Object, data.payload) + rsize);
1406 o->object.flags |= compression;
1407
1408 log_debug("Compressed data object %"PRIu64" -> %zu using %s",
1409 size, rsize, object_compressed_to_string(compression));
1410 } else
1411 /* Compression didn't work, we don't really care why, let's continue without compression */
1412 compression = 0;
1413 }
1414#endif
1415
1416 if (compression == 0)
1417 memcpy_safe(o->data.payload, data, size);
1418
1419 r = journal_file_link_data(f, o, p, hash);
1420 if (r < 0)
1421 return r;
1422
1423#ifdef HAVE_GCRYPT
1424 r = journal_file_hmac_put_object(f, OBJECT_DATA, o, p);
1425 if (r < 0)
1426 return r;
1427#endif
1428
1429 /* The linking might have altered the window, so let's
1430 * refresh our pointer */
1431 r = journal_file_move_to_object(f, OBJECT_DATA, p, &o);
1432 if (r < 0)
1433 return r;
1434
1435 if (!data)
1436 eq = NULL;
1437 else
1438 eq = memchr(data, '=', size);
1439 if (eq && eq > data) {
1440 Object *fo = NULL;
1441 uint64_t fp;
1442
1443 /* Create field object ... */
1444 r = journal_file_append_field(f, data, (uint8_t*) eq - (uint8_t*) data, &fo, &fp);
1445 if (r < 0)
1446 return r;
1447
1448 /* ... and link it in. */
1449 o->data.next_field_offset = fo->field.head_data_offset;
1450 fo->field.head_data_offset = le64toh(p);
1451 }
1452
1453 if (ret)
1454 *ret = o;
1455
1456 if (offset)
1457 *offset = p;
1458
1459 return 0;
1460}
1461
1462uint64_t journal_file_entry_n_items(Object *o) {
1463 assert(o);
1464
1465 if (o->object.type != OBJECT_ENTRY)
1466 return 0;
1467
1468 return (le64toh(o->object.size) - offsetof(Object, entry.items)) / sizeof(EntryItem);
1469}
1470
1471uint64_t journal_file_entry_array_n_items(Object *o) {
1472 assert(o);
1473
1474 if (o->object.type != OBJECT_ENTRY_ARRAY)
1475 return 0;
1476
1477 return (le64toh(o->object.size) - offsetof(Object, entry_array.items)) / sizeof(uint64_t);
1478}
1479
1480uint64_t journal_file_hash_table_n_items(Object *o) {
1481 assert(o);
1482
1483 if (o->object.type != OBJECT_DATA_HASH_TABLE &&
1484 o->object.type != OBJECT_FIELD_HASH_TABLE)
1485 return 0;
1486
1487 return (le64toh(o->object.size) - offsetof(Object, hash_table.items)) / sizeof(HashItem);
1488}
1489
1490static int link_entry_into_array(JournalFile *f,
1491 le64_t *first,
1492 le64_t *idx,
1493 uint64_t p) {
1494 int r;
1495 uint64_t n = 0, ap = 0, q, i, a, hidx;
1496 Object *o;
1497
1498 assert(f);
1499 assert(f->header);
1500 assert(first);
1501 assert(idx);
1502 assert(p > 0);
1503
1504 a = le64toh(*first);
1505 i = hidx = le64toh(*idx);
1506 while (a > 0) {
1507
1508 r = journal_file_move_to_object(f, OBJECT_ENTRY_ARRAY, a, &o);
1509 if (r < 0)
1510 return r;
1511
1512 n = journal_file_entry_array_n_items(o);
1513 if (i < n) {
1514 o->entry_array.items[i] = htole64(p);
1515 *idx = htole64(hidx + 1);
1516 return 0;
1517 }
1518
1519 i -= n;
1520 ap = a;
1521 a = le64toh(o->entry_array.next_entry_array_offset);
1522 }
1523
1524 if (hidx > n)
1525 n = (hidx+1) * 2;
1526 else
1527 n = n * 2;
1528
1529 if (n < 4)
1530 n = 4;
1531
1532 r = journal_file_append_object(f, OBJECT_ENTRY_ARRAY,
1533 offsetof(Object, entry_array.items) + n * sizeof(uint64_t),
1534 &o, &q);
1535 if (r < 0)
1536 return r;
1537
1538#ifdef HAVE_GCRYPT
1539 r = journal_file_hmac_put_object(f, OBJECT_ENTRY_ARRAY, o, q);
1540 if (r < 0)
1541 return r;
1542#endif
1543
1544 o->entry_array.items[i] = htole64(p);
1545
1546 if (ap == 0)
1547 *first = htole64(q);
1548 else {
1549 r = journal_file_move_to_object(f, OBJECT_ENTRY_ARRAY, ap, &o);
1550 if (r < 0)
1551 return r;
1552
1553 o->entry_array.next_entry_array_offset = htole64(q);
1554 }
1555
1556 if (JOURNAL_HEADER_CONTAINS(f->header, n_entry_arrays))
1557 f->header->n_entry_arrays = htole64(le64toh(f->header->n_entry_arrays) + 1);
1558
1559 *idx = htole64(hidx + 1);
1560
1561 return 0;
1562}
1563
1564static int link_entry_into_array_plus_one(JournalFile *f,
1565 le64_t *extra,
1566 le64_t *first,
1567 le64_t *idx,
1568 uint64_t p) {
1569
1570 int r;
1571
1572 assert(f);
1573 assert(extra);
1574 assert(first);
1575 assert(idx);
1576 assert(p > 0);
1577
1578 if (*idx == 0)
1579 *extra = htole64(p);
1580 else {
1581 le64_t i;
1582
1583 i = htole64(le64toh(*idx) - 1);
1584 r = link_entry_into_array(f, first, &i, p);
1585 if (r < 0)
1586 return r;
1587 }
1588
1589 *idx = htole64(le64toh(*idx) + 1);
1590 return 0;
1591}
1592
1593static int journal_file_link_entry_item(JournalFile *f, Object *o, uint64_t offset, uint64_t i) {
1594 uint64_t p;
1595 int r;
1596 assert(f);
1597 assert(o);
1598 assert(offset > 0);
1599
1600 p = le64toh(o->entry.items[i].object_offset);
1601 if (p == 0)
1602 return -EINVAL;
1603
1604 r = journal_file_move_to_object(f, OBJECT_DATA, p, &o);
1605 if (r < 0)
1606 return r;
1607
1608 return link_entry_into_array_plus_one(f,
1609 &o->data.entry_offset,
1610 &o->data.entry_array_offset,
1611 &o->data.n_entries,
1612 offset);
1613}
1614
1615static int journal_file_link_entry(JournalFile *f, Object *o, uint64_t offset) {
1616 uint64_t n, i;
1617 int r;
1618
1619 assert(f);
1620 assert(f->header);
1621 assert(o);
1622 assert(offset > 0);
1623
1624 if (o->object.type != OBJECT_ENTRY)
1625 return -EINVAL;
1626
1627 __sync_synchronize();
1628
1629 /* Link up the entry itself */
1630 r = link_entry_into_array(f,
1631 &f->header->entry_array_offset,
1632 &f->header->n_entries,
1633 offset);
1634 if (r < 0)
1635 return r;
1636
1637 /* log_debug("=> %s seqnr=%"PRIu64" n_entries=%"PRIu64, f->path, o->entry.seqnum, f->header->n_entries); */
1638
1639 if (f->header->head_entry_realtime == 0)
1640 f->header->head_entry_realtime = o->entry.realtime;
1641
1642 f->header->tail_entry_realtime = o->entry.realtime;
1643 f->header->tail_entry_monotonic = o->entry.monotonic;
1644
1645 f->tail_entry_monotonic_valid = true;
1646
1647 /* Link up the items */
1648 n = journal_file_entry_n_items(o);
1649 for (i = 0; i < n; i++) {
1650 r = journal_file_link_entry_item(f, o, offset, i);
1651 if (r < 0)
1652 return r;
1653 }
1654
1655 return 0;
1656}
1657
1658static int journal_file_append_entry_internal(
1659 JournalFile *f,
1660 const dual_timestamp *ts,
1661 uint64_t xor_hash,
1662 const EntryItem items[], unsigned n_items,
1663 uint64_t *seqnum,
1664 Object **ret, uint64_t *offset) {
1665 uint64_t np;
1666 uint64_t osize;
1667 Object *o;
1668 int r;
1669
1670 assert(f);
1671 assert(f->header);
1672 assert(items || n_items == 0);
1673 assert(ts);
1674
1675 osize = offsetof(Object, entry.items) + (n_items * sizeof(EntryItem));
1676
1677 r = journal_file_append_object(f, OBJECT_ENTRY, osize, &o, &np);
1678 if (r < 0)
1679 return r;
1680
1681 o->entry.seqnum = htole64(journal_file_entry_seqnum(f, seqnum));
1682 memcpy_safe(o->entry.items, items, n_items * sizeof(EntryItem));
1683 o->entry.realtime = htole64(ts->realtime);
1684 o->entry.monotonic = htole64(ts->monotonic);
1685 o->entry.xor_hash = htole64(xor_hash);
1686 o->entry.boot_id = f->header->boot_id;
1687
1688#ifdef HAVE_GCRYPT
1689 r = journal_file_hmac_put_object(f, OBJECT_ENTRY, o, np);
1690 if (r < 0)
1691 return r;
1692#endif
1693
1694 r = journal_file_link_entry(f, o, np);
1695 if (r < 0)
1696 return r;
1697
1698 if (ret)
1699 *ret = o;
1700
1701 if (offset)
1702 *offset = np;
1703
1704 return 0;
1705}
1706
1707void journal_file_post_change(JournalFile *f) {
1708 assert(f);
1709
1710 /* inotify() does not receive IN_MODIFY events from file
1711 * accesses done via mmap(). After each access we hence
1712 * trigger IN_MODIFY by truncating the journal file to its
1713 * current size which triggers IN_MODIFY. */
1714
1715 __sync_synchronize();
1716
1717 if (ftruncate(f->fd, f->last_stat.st_size) < 0)
1718 log_debug_errno(errno, "Failed to truncate file to its own size: %m");
1719}
1720
1721static int post_change_thunk(sd_event_source *timer, uint64_t usec, void *userdata) {
1722 assert(userdata);
1723
1724 journal_file_post_change(userdata);
1725
1726 return 1;
1727}
1728
1729static void schedule_post_change(JournalFile *f) {
1730 sd_event_source *timer;
1731 int enabled, r;
1732 uint64_t now;
1733
1734 assert(f);
1735 assert(f->post_change_timer);
1736
1737 timer = f->post_change_timer;
1738
1739 r = sd_event_source_get_enabled(timer, &enabled);
1740 if (r < 0) {
1741 log_debug_errno(r, "Failed to get ftruncate timer state: %m");
1742 goto fail;
1743 }
1744
1745 if (enabled == SD_EVENT_ONESHOT)
1746 return;
1747
1748 r = sd_event_now(sd_event_source_get_event(timer), CLOCK_MONOTONIC, &now);
1749 if (r < 0) {
1750 log_debug_errno(r, "Failed to get clock's now for scheduling ftruncate: %m");
1751 goto fail;
1752 }
1753
1754 r = sd_event_source_set_time(timer, now+f->post_change_timer_period);
1755 if (r < 0) {
1756 log_debug_errno(r, "Failed to set time for scheduling ftruncate: %m");
1757 goto fail;
1758 }
1759
1760 r = sd_event_source_set_enabled(timer, SD_EVENT_ONESHOT);
1761 if (r < 0) {
1762 log_debug_errno(r, "Failed to enable scheduled ftruncate: %m");
1763 goto fail;
1764 }
1765
1766 return;
1767
1768fail:
1769 /* On failure, let's simply post the change immediately. */
1770 journal_file_post_change(f);
1771}
1772
1773/* Enable coalesced change posting in a timer on the provided sd_event instance */
1774int journal_file_enable_post_change_timer(JournalFile *f, sd_event *e, usec_t t) {
1775 _cleanup_(sd_event_source_unrefp) sd_event_source *timer = NULL;
1776 int r;
1777
1778 assert(f);
1779 assert_return(!f->post_change_timer, -EINVAL);
1780 assert(e);
1781 assert(t);
1782
1783 r = sd_event_add_time(e, &timer, CLOCK_MONOTONIC, 0, 0, post_change_thunk, f);
1784 if (r < 0)
1785 return r;
1786
1787 r = sd_event_source_set_enabled(timer, SD_EVENT_OFF);
1788 if (r < 0)
1789 return r;
1790
1791 f->post_change_timer = timer;
1792 timer = NULL;
1793 f->post_change_timer_period = t;
1794
1795 return r;
1796}
1797
1798static int entry_item_cmp(const void *_a, const void *_b) {
1799 const EntryItem *a = _a, *b = _b;
1800
1801 if (le64toh(a->object_offset) < le64toh(b->object_offset))
1802 return -1;
1803 if (le64toh(a->object_offset) > le64toh(b->object_offset))
1804 return 1;
1805 return 0;
1806}
1807
1808int journal_file_append_entry(JournalFile *f, const dual_timestamp *ts, const struct iovec iovec[], unsigned n_iovec, uint64_t *seqnum, Object **ret, uint64_t *offset) {
1809 unsigned i;
1810 EntryItem *items;
1811 int r;
1812 uint64_t xor_hash = 0;
1813 struct dual_timestamp _ts;
1814
1815 assert(f);
1816 assert(f->header);
1817 assert(iovec || n_iovec == 0);
1818
1819 if (!ts) {
1820 dual_timestamp_get(&_ts);
1821 ts = &_ts;
1822 }
1823
1824#ifdef HAVE_GCRYPT
1825 r = journal_file_maybe_append_tag(f, ts->realtime);
1826 if (r < 0)
1827 return r;
1828#endif
1829
1830 /* alloca() can't take 0, hence let's allocate at least one */
1831 items = alloca(sizeof(EntryItem) * MAX(1u, n_iovec));
1832
1833 for (i = 0; i < n_iovec; i++) {
1834 uint64_t p;
1835 Object *o;
1836
1837 r = journal_file_append_data(f, iovec[i].iov_base, iovec[i].iov_len, &o, &p);
1838 if (r < 0)
1839 return r;
1840
1841 xor_hash ^= le64toh(o->data.hash);
1842 items[i].object_offset = htole64(p);
1843 items[i].hash = o->data.hash;
1844 }
1845
1846 /* Order by the position on disk, in order to improve seek
1847 * times for rotating media. */
1848 qsort_safe(items, n_iovec, sizeof(EntryItem), entry_item_cmp);
1849
1850 r = journal_file_append_entry_internal(f, ts, xor_hash, items, n_iovec, seqnum, ret, offset);
1851
1852 /* If the memory mapping triggered a SIGBUS then we return an
1853 * IO error and ignore the error code passed down to us, since
1854 * it is very likely just an effect of a nullified replacement
1855 * mapping page */
1856
1857 if (mmap_cache_got_sigbus(f->mmap, f->fd))
1858 r = -EIO;
1859
1860 if (f->post_change_timer)
1861 schedule_post_change(f);
1862 else
1863 journal_file_post_change(f);
1864
1865 return r;
1866}
1867
1868typedef struct ChainCacheItem {
1869 uint64_t first; /* the array at the beginning of the chain */
1870 uint64_t array; /* the cached array */
1871 uint64_t begin; /* the first item in the cached array */
1872 uint64_t total; /* the total number of items in all arrays before this one in the chain */
1873 uint64_t last_index; /* the last index we looked at, to optimize locality when bisecting */
1874} ChainCacheItem;
1875
1876static void chain_cache_put(
1877 OrderedHashmap *h,
1878 ChainCacheItem *ci,
1879 uint64_t first,
1880 uint64_t array,
1881 uint64_t begin,
1882 uint64_t total,
1883 uint64_t last_index) {
1884
1885 if (!ci) {
1886 /* If the chain item to cache for this chain is the
1887 * first one it's not worth caching anything */
1888 if (array == first)
1889 return;
1890
1891 if (ordered_hashmap_size(h) >= CHAIN_CACHE_MAX) {
1892 ci = ordered_hashmap_steal_first(h);
1893 assert(ci);
1894 } else {
1895 ci = new(ChainCacheItem, 1);
1896 if (!ci)
1897 return;
1898 }
1899
1900 ci->first = first;
1901
1902 if (ordered_hashmap_put(h, &ci->first, ci) < 0) {
1903 free(ci);
1904 return;
1905 }
1906 } else
1907 assert(ci->first == first);
1908
1909 ci->array = array;
1910 ci->begin = begin;
1911 ci->total = total;
1912 ci->last_index = last_index;
1913}
1914
1915static int generic_array_get(
1916 JournalFile *f,
1917 uint64_t first,
1918 uint64_t i,
1919 Object **ret, uint64_t *offset) {
1920
1921 Object *o;
1922 uint64_t p = 0, a, t = 0;
1923 int r;
1924 ChainCacheItem *ci;
1925
1926 assert(f);
1927
1928 a = first;
1929
1930 /* Try the chain cache first */
1931 ci = ordered_hashmap_get(f->chain_cache, &first);
1932 if (ci && i > ci->total) {
1933 a = ci->array;
1934 i -= ci->total;
1935 t = ci->total;
1936 }
1937
1938 while (a > 0) {
1939 uint64_t k;
1940
1941 r = journal_file_move_to_object(f, OBJECT_ENTRY_ARRAY, a, &o);
1942 if (r < 0)
1943 return r;
1944
1945 k = journal_file_entry_array_n_items(o);
1946 if (i < k) {
1947 p = le64toh(o->entry_array.items[i]);
1948 goto found;
1949 }
1950
1951 i -= k;
1952 t += k;
1953 a = le64toh(o->entry_array.next_entry_array_offset);
1954 }
1955
1956 return 0;
1957
1958found:
1959 /* Let's cache this item for the next invocation */
1960 chain_cache_put(f->chain_cache, ci, first, a, le64toh(o->entry_array.items[0]), t, i);
1961
1962 r = journal_file_move_to_object(f, OBJECT_ENTRY, p, &o);
1963 if (r < 0)
1964 return r;
1965
1966 if (ret)
1967 *ret = o;
1968
1969 if (offset)
1970 *offset = p;
1971
1972 return 1;
1973}
1974
1975static int generic_array_get_plus_one(
1976 JournalFile *f,
1977 uint64_t extra,
1978 uint64_t first,
1979 uint64_t i,
1980 Object **ret, uint64_t *offset) {
1981
1982 Object *o;
1983
1984 assert(f);
1985
1986 if (i == 0) {
1987 int r;
1988
1989 r = journal_file_move_to_object(f, OBJECT_ENTRY, extra, &o);
1990 if (r < 0)
1991 return r;
1992
1993 if (ret)
1994 *ret = o;
1995
1996 if (offset)
1997 *offset = extra;
1998
1999 return 1;
2000 }
2001
2002 return generic_array_get(f, first, i-1, ret, offset);
2003}
2004
2005enum {
2006 TEST_FOUND,
2007 TEST_LEFT,
2008 TEST_RIGHT
2009};
2010
2011static int generic_array_bisect(
2012 JournalFile *f,
2013 uint64_t first,
2014 uint64_t n,
2015 uint64_t needle,
2016 int (*test_object)(JournalFile *f, uint64_t p, uint64_t needle),
2017 direction_t direction,
2018 Object **ret,
2019 uint64_t *offset,
2020 uint64_t *idx) {
2021
2022 uint64_t a, p, t = 0, i = 0, last_p = 0, last_index = (uint64_t) -1;
2023 bool subtract_one = false;
2024 Object *o, *array = NULL;
2025 int r;
2026 ChainCacheItem *ci;
2027
2028 assert(f);
2029 assert(test_object);
2030
2031 /* Start with the first array in the chain */
2032 a = first;
2033
2034 ci = ordered_hashmap_get(f->chain_cache, &first);
2035 if (ci && n > ci->total) {
2036 /* Ah, we have iterated this bisection array chain
2037 * previously! Let's see if we can skip ahead in the
2038 * chain, as far as the last time. But we can't jump
2039 * backwards in the chain, so let's check that
2040 * first. */
2041
2042 r = test_object(f, ci->begin, needle);
2043 if (r < 0)
2044 return r;
2045
2046 if (r == TEST_LEFT) {
2047 /* OK, what we are looking for is right of the
2048 * begin of this EntryArray, so let's jump
2049 * straight to previously cached array in the
2050 * chain */
2051
2052 a = ci->array;
2053 n -= ci->total;
2054 t = ci->total;
2055 last_index = ci->last_index;
2056 }
2057 }
2058
2059 while (a > 0) {
2060 uint64_t left, right, k, lp;
2061
2062 r = journal_file_move_to_object(f, OBJECT_ENTRY_ARRAY, a, &array);
2063 if (r < 0)
2064 return r;
2065
2066 k = journal_file_entry_array_n_items(array);
2067 right = MIN(k, n);
2068 if (right <= 0)
2069 return 0;
2070
2071 i = right - 1;
2072 lp = p = le64toh(array->entry_array.items[i]);
2073 if (p <= 0)
2074 r = -EBADMSG;
2075 else
2076 r = test_object(f, p, needle);
2077 if (r == -EBADMSG) {
2078 log_debug_errno(r, "Encountered invalid entry while bisecting, cutting algorithm short. (1)");
2079 n = i;
2080 continue;
2081 }
2082 if (r < 0)
2083 return r;
2084
2085 if (r == TEST_FOUND)
2086 r = direction == DIRECTION_DOWN ? TEST_RIGHT : TEST_LEFT;
2087
2088 if (r == TEST_RIGHT) {
2089 left = 0;
2090 right -= 1;
2091
2092 if (last_index != (uint64_t) -1) {
2093 assert(last_index <= right);
2094
2095 /* If we cached the last index we
2096 * looked at, let's try to not to jump
2097 * too wildly around and see if we can
2098 * limit the range to look at early to
2099 * the immediate neighbors of the last
2100 * index we looked at. */
2101
2102 if (last_index > 0) {
2103 uint64_t x = last_index - 1;
2104
2105 p = le64toh(array->entry_array.items[x]);
2106 if (p <= 0)
2107 return -EBADMSG;
2108
2109 r = test_object(f, p, needle);
2110 if (r < 0)
2111 return r;
2112
2113 if (r == TEST_FOUND)
2114 r = direction == DIRECTION_DOWN ? TEST_RIGHT : TEST_LEFT;
2115
2116 if (r == TEST_RIGHT)
2117 right = x;
2118 else
2119 left = x + 1;
2120 }
2121
2122 if (last_index < right) {
2123 uint64_t y = last_index + 1;
2124
2125 p = le64toh(array->entry_array.items[y]);
2126 if (p <= 0)
2127 return -EBADMSG;
2128
2129 r = test_object(f, p, needle);
2130 if (r < 0)
2131 return r;
2132
2133 if (r == TEST_FOUND)
2134 r = direction == DIRECTION_DOWN ? TEST_RIGHT : TEST_LEFT;
2135
2136 if (r == TEST_RIGHT)
2137 right = y;
2138 else
2139 left = y + 1;
2140 }
2141 }
2142
2143 for (;;) {
2144 if (left == right) {
2145 if (direction == DIRECTION_UP)
2146 subtract_one = true;
2147
2148 i = left;
2149 goto found;
2150 }
2151
2152 assert(left < right);
2153 i = (left + right) / 2;
2154
2155 p = le64toh(array->entry_array.items[i]);
2156 if (p <= 0)
2157 r = -EBADMSG;
2158 else
2159 r = test_object(f, p, needle);
2160 if (r == -EBADMSG) {
2161 log_debug_errno(r, "Encountered invalid entry while bisecting, cutting algorithm short. (2)");
2162 right = n = i;
2163 continue;
2164 }
2165 if (r < 0)
2166 return r;
2167
2168 if (r == TEST_FOUND)
2169 r = direction == DIRECTION_DOWN ? TEST_RIGHT : TEST_LEFT;
2170
2171 if (r == TEST_RIGHT)
2172 right = i;
2173 else
2174 left = i + 1;
2175 }
2176 }
2177
2178 if (k >= n) {
2179 if (direction == DIRECTION_UP) {
2180 i = n;
2181 subtract_one = true;
2182 goto found;
2183 }
2184
2185 return 0;
2186 }
2187
2188 last_p = lp;
2189
2190 n -= k;
2191 t += k;
2192 last_index = (uint64_t) -1;
2193 a = le64toh(array->entry_array.next_entry_array_offset);
2194 }
2195
2196 return 0;
2197
2198found:
2199 if (subtract_one && t == 0 && i == 0)
2200 return 0;
2201
2202 /* Let's cache this item for the next invocation */
2203 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);
2204
2205 if (subtract_one && i == 0)
2206 p = last_p;
2207 else if (subtract_one)
2208 p = le64toh(array->entry_array.items[i-1]);
2209 else
2210 p = le64toh(array->entry_array.items[i]);
2211
2212 r = journal_file_move_to_object(f, OBJECT_ENTRY, p, &o);
2213 if (r < 0)
2214 return r;
2215
2216 if (ret)
2217 *ret = o;
2218
2219 if (offset)
2220 *offset = p;
2221
2222 if (idx)
2223 *idx = t + i + (subtract_one ? -1 : 0);
2224
2225 return 1;
2226}
2227
2228static int generic_array_bisect_plus_one(
2229 JournalFile *f,
2230 uint64_t extra,
2231 uint64_t first,
2232 uint64_t n,
2233 uint64_t needle,
2234 int (*test_object)(JournalFile *f, uint64_t p, uint64_t needle),
2235 direction_t direction,
2236 Object **ret,
2237 uint64_t *offset,
2238 uint64_t *idx) {
2239
2240 int r;
2241 bool step_back = false;
2242 Object *o;
2243
2244 assert(f);
2245 assert(test_object);
2246
2247 if (n <= 0)
2248 return 0;
2249
2250 /* This bisects the array in object 'first', but first checks
2251 * an extra */
2252 r = test_object(f, extra, needle);
2253 if (r < 0)
2254 return r;
2255
2256 if (r == TEST_FOUND)
2257 r = direction == DIRECTION_DOWN ? TEST_RIGHT : TEST_LEFT;
2258
2259 /* if we are looking with DIRECTION_UP then we need to first
2260 see if in the actual array there is a matching entry, and
2261 return the last one of that. But if there isn't any we need
2262 to return this one. Hence remember this, and return it
2263 below. */
2264 if (r == TEST_LEFT)
2265 step_back = direction == DIRECTION_UP;
2266
2267 if (r == TEST_RIGHT) {
2268 if (direction == DIRECTION_DOWN)
2269 goto found;
2270 else
2271 return 0;
2272 }
2273
2274 r = generic_array_bisect(f, first, n-1, needle, test_object, direction, ret, offset, idx);
2275
2276 if (r == 0 && step_back)
2277 goto found;
2278
2279 if (r > 0 && idx)
2280 (*idx)++;
2281
2282 return r;
2283
2284found:
2285 r = journal_file_move_to_object(f, OBJECT_ENTRY, extra, &o);
2286 if (r < 0)
2287 return r;
2288
2289 if (ret)
2290 *ret = o;
2291
2292 if (offset)
2293 *offset = extra;
2294
2295 if (idx)
2296 *idx = 0;
2297
2298 return 1;
2299}
2300
2301_pure_ static int test_object_offset(JournalFile *f, uint64_t p, uint64_t needle) {
2302 assert(f);
2303 assert(p > 0);
2304
2305 if (p == needle)
2306 return TEST_FOUND;
2307 else if (p < needle)
2308 return TEST_LEFT;
2309 else
2310 return TEST_RIGHT;
2311}
2312
2313static int test_object_seqnum(JournalFile *f, uint64_t p, uint64_t needle) {
2314 Object *o;
2315 int r;
2316
2317 assert(f);
2318 assert(p > 0);
2319
2320 r = journal_file_move_to_object(f, OBJECT_ENTRY, p, &o);
2321 if (r < 0)
2322 return r;
2323
2324 if (le64toh(o->entry.seqnum) == needle)
2325 return TEST_FOUND;
2326 else if (le64toh(o->entry.seqnum) < needle)
2327 return TEST_LEFT;
2328 else
2329 return TEST_RIGHT;
2330}
2331
2332int journal_file_move_to_entry_by_seqnum(
2333 JournalFile *f,
2334 uint64_t seqnum,
2335 direction_t direction,
2336 Object **ret,
2337 uint64_t *offset) {
2338 assert(f);
2339 assert(f->header);
2340
2341 return generic_array_bisect(f,
2342 le64toh(f->header->entry_array_offset),
2343 le64toh(f->header->n_entries),
2344 seqnum,
2345 test_object_seqnum,
2346 direction,
2347 ret, offset, NULL);
2348}
2349
2350static int test_object_realtime(JournalFile *f, uint64_t p, uint64_t needle) {
2351 Object *o;
2352 int r;
2353
2354 assert(f);
2355 assert(p > 0);
2356
2357 r = journal_file_move_to_object(f, OBJECT_ENTRY, p, &o);
2358 if (r < 0)
2359 return r;
2360
2361 if (le64toh(o->entry.realtime) == needle)
2362 return TEST_FOUND;
2363 else if (le64toh(o->entry.realtime) < needle)
2364 return TEST_LEFT;
2365 else
2366 return TEST_RIGHT;
2367}
2368
2369int journal_file_move_to_entry_by_realtime(
2370 JournalFile *f,
2371 uint64_t realtime,
2372 direction_t direction,
2373 Object **ret,
2374 uint64_t *offset) {
2375 assert(f);
2376 assert(f->header);
2377
2378 return generic_array_bisect(f,
2379 le64toh(f->header->entry_array_offset),
2380 le64toh(f->header->n_entries),
2381 realtime,
2382 test_object_realtime,
2383 direction,
2384 ret, offset, NULL);
2385}
2386
2387static int test_object_monotonic(JournalFile *f, uint64_t p, uint64_t needle) {
2388 Object *o;
2389 int r;
2390
2391 assert(f);
2392 assert(p > 0);
2393
2394 r = journal_file_move_to_object(f, OBJECT_ENTRY, p, &o);
2395 if (r < 0)
2396 return r;
2397
2398 if (le64toh(o->entry.monotonic) == needle)
2399 return TEST_FOUND;
2400 else if (le64toh(o->entry.monotonic) < needle)
2401 return TEST_LEFT;
2402 else
2403 return TEST_RIGHT;
2404}
2405
2406static int find_data_object_by_boot_id(
2407 JournalFile *f,
2408 sd_id128_t boot_id,
2409 Object **o,
2410 uint64_t *b) {
2411
2412 char t[sizeof("_BOOT_ID=")-1 + 32 + 1] = "_BOOT_ID=";
2413
2414 sd_id128_to_string(boot_id, t + 9);
2415 return journal_file_find_data_object(f, t, sizeof(t) - 1, o, b);
2416}
2417
2418int journal_file_move_to_entry_by_monotonic(
2419 JournalFile *f,
2420 sd_id128_t boot_id,
2421 uint64_t monotonic,
2422 direction_t direction,
2423 Object **ret,
2424 uint64_t *offset) {
2425
2426 Object *o;
2427 int r;
2428
2429 assert(f);
2430
2431 r = find_data_object_by_boot_id(f, boot_id, &o, NULL);
2432 if (r < 0)
2433 return r;
2434 if (r == 0)
2435 return -ENOENT;
2436
2437 return generic_array_bisect_plus_one(f,
2438 le64toh(o->data.entry_offset),
2439 le64toh(o->data.entry_array_offset),
2440 le64toh(o->data.n_entries),
2441 monotonic,
2442 test_object_monotonic,
2443 direction,
2444 ret, offset, NULL);
2445}
2446
2447void journal_file_reset_location(JournalFile *f) {
2448 f->location_type = LOCATION_HEAD;
2449 f->current_offset = 0;
2450 f->current_seqnum = 0;
2451 f->current_realtime = 0;
2452 f->current_monotonic = 0;
2453 zero(f->current_boot_id);
2454 f->current_xor_hash = 0;
2455}
2456
2457void journal_file_save_location(JournalFile *f, Object *o, uint64_t offset) {
2458 f->location_type = LOCATION_SEEK;
2459 f->current_offset = offset;
2460 f->current_seqnum = le64toh(o->entry.seqnum);
2461 f->current_realtime = le64toh(o->entry.realtime);
2462 f->current_monotonic = le64toh(o->entry.monotonic);
2463 f->current_boot_id = o->entry.boot_id;
2464 f->current_xor_hash = le64toh(o->entry.xor_hash);
2465}
2466
2467int journal_file_compare_locations(JournalFile *af, JournalFile *bf) {
2468 assert(af);
2469 assert(af->header);
2470 assert(bf);
2471 assert(bf->header);
2472 assert(af->location_type == LOCATION_SEEK);
2473 assert(bf->location_type == LOCATION_SEEK);
2474
2475 /* If contents and timestamps match, these entries are
2476 * identical, even if the seqnum does not match */
2477 if (sd_id128_equal(af->current_boot_id, bf->current_boot_id) &&
2478 af->current_monotonic == bf->current_monotonic &&
2479 af->current_realtime == bf->current_realtime &&
2480 af->current_xor_hash == bf->current_xor_hash)
2481 return 0;
2482
2483 if (sd_id128_equal(af->header->seqnum_id, bf->header->seqnum_id)) {
2484
2485 /* If this is from the same seqnum source, compare
2486 * seqnums */
2487 if (af->current_seqnum < bf->current_seqnum)
2488 return -1;
2489 if (af->current_seqnum > bf->current_seqnum)
2490 return 1;
2491
2492 /* Wow! This is weird, different data but the same
2493 * seqnums? Something is borked, but let's make the
2494 * best of it and compare by time. */
2495 }
2496
2497 if (sd_id128_equal(af->current_boot_id, bf->current_boot_id)) {
2498
2499 /* If the boot id matches, compare monotonic time */
2500 if (af->current_monotonic < bf->current_monotonic)
2501 return -1;
2502 if (af->current_monotonic > bf->current_monotonic)
2503 return 1;
2504 }
2505
2506 /* Otherwise, compare UTC time */
2507 if (af->current_realtime < bf->current_realtime)
2508 return -1;
2509 if (af->current_realtime > bf->current_realtime)
2510 return 1;
2511
2512 /* Finally, compare by contents */
2513 if (af->current_xor_hash < bf->current_xor_hash)
2514 return -1;
2515 if (af->current_xor_hash > bf->current_xor_hash)
2516 return 1;
2517
2518 return 0;
2519}
2520
2521static int bump_array_index(uint64_t *i, direction_t direction, uint64_t n) {
2522
2523 /* Increase or decrease the specified index, in the right direction. */
2524
2525 if (direction == DIRECTION_DOWN) {
2526 if (*i >= n - 1)
2527 return 0;
2528
2529 (*i) ++;
2530 } else {
2531 if (*i <= 0)
2532 return 0;
2533
2534 (*i) --;
2535 }
2536
2537 return 1;
2538}
2539
2540static bool check_properly_ordered(uint64_t new_offset, uint64_t old_offset, direction_t direction) {
2541
2542 /* Consider it an error if any of the two offsets is uninitialized */
2543 if (old_offset == 0 || new_offset == 0)
2544 return false;
2545
2546 /* If we go down, the new offset must be larger than the old one. */
2547 return direction == DIRECTION_DOWN ?
2548 new_offset > old_offset :
2549 new_offset < old_offset;
2550}
2551
2552int journal_file_next_entry(
2553 JournalFile *f,
2554 uint64_t p,
2555 direction_t direction,
2556 Object **ret, uint64_t *offset) {
2557
2558 uint64_t i, n, ofs;
2559 int r;
2560
2561 assert(f);
2562 assert(f->header);
2563
2564 n = le64toh(f->header->n_entries);
2565 if (n <= 0)
2566 return 0;
2567
2568 if (p == 0)
2569 i = direction == DIRECTION_DOWN ? 0 : n - 1;
2570 else {
2571 r = generic_array_bisect(f,
2572 le64toh(f->header->entry_array_offset),
2573 le64toh(f->header->n_entries),
2574 p,
2575 test_object_offset,
2576 DIRECTION_DOWN,
2577 NULL, NULL,
2578 &i);
2579 if (r <= 0)
2580 return r;
2581
2582 r = bump_array_index(&i, direction, n);
2583 if (r <= 0)
2584 return r;
2585 }
2586
2587 /* And jump to it */
2588 for (;;) {
2589 r = generic_array_get(f,
2590 le64toh(f->header->entry_array_offset),
2591 i,
2592 ret, &ofs);
2593 if (r > 0)
2594 break;
2595 if (r != -EBADMSG)
2596 return r;
2597
2598 /* OK, so this entry is borked. Most likely some entry didn't get synced to disk properly, let's see if
2599 * the next one might work for us instead. */
2600 log_debug_errno(r, "Entry item %" PRIu64 " is bad, skipping over it.", i);
2601
2602 r = bump_array_index(&i, direction, n);
2603 if (r <= 0)
2604 return r;
2605 }
2606
2607 /* Ensure our array is properly ordered. */
2608 if (p > 0 && !check_properly_ordered(ofs, p, direction)) {
2609 log_debug("%s: entry array not properly ordered at entry %" PRIu64, f->path, i);
2610 return -EBADMSG;
2611 }
2612
2613 if (offset)
2614 *offset = ofs;
2615
2616 return 1;
2617}
2618
2619int journal_file_next_entry_for_data(
2620 JournalFile *f,
2621 Object *o, uint64_t p,
2622 uint64_t data_offset,
2623 direction_t direction,
2624 Object **ret, uint64_t *offset) {
2625
2626 uint64_t i, n, ofs;
2627 Object *d;
2628 int r;
2629
2630 assert(f);
2631 assert(p > 0 || !o);
2632
2633 r = journal_file_move_to_object(f, OBJECT_DATA, data_offset, &d);
2634 if (r < 0)
2635 return r;
2636
2637 n = le64toh(d->data.n_entries);
2638 if (n <= 0)
2639 return n;
2640
2641 if (!o)
2642 i = direction == DIRECTION_DOWN ? 0 : n - 1;
2643 else {
2644 if (o->object.type != OBJECT_ENTRY)
2645 return -EINVAL;
2646
2647 r = generic_array_bisect_plus_one(f,
2648 le64toh(d->data.entry_offset),
2649 le64toh(d->data.entry_array_offset),
2650 le64toh(d->data.n_entries),
2651 p,
2652 test_object_offset,
2653 DIRECTION_DOWN,
2654 NULL, NULL,
2655 &i);
2656
2657 if (r <= 0)
2658 return r;
2659
2660 r = bump_array_index(&i, direction, n);
2661 if (r <= 0)
2662 return r;
2663 }
2664
2665 for (;;) {
2666 r = generic_array_get_plus_one(f,
2667 le64toh(d->data.entry_offset),
2668 le64toh(d->data.entry_array_offset),
2669 i,
2670 ret, &ofs);
2671 if (r > 0)
2672 break;
2673 if (r != -EBADMSG)
2674 return r;
2675
2676 log_debug_errno(r, "Data entry item %" PRIu64 " is bad, skipping over it.", i);
2677
2678 r = bump_array_index(&i, direction, n);
2679 if (r <= 0)
2680 return r;
2681 }
2682
2683 /* Ensure our array is properly ordered. */
2684 if (p > 0 && check_properly_ordered(ofs, p, direction)) {
2685 log_debug("%s data entry array not properly ordered at entry %" PRIu64, f->path, i);
2686 return -EBADMSG;
2687 }
2688
2689 if (offset)
2690 *offset = ofs;
2691
2692 return 1;
2693}
2694
2695int journal_file_move_to_entry_by_offset_for_data(
2696 JournalFile *f,
2697 uint64_t data_offset,
2698 uint64_t p,
2699 direction_t direction,
2700 Object **ret, uint64_t *offset) {
2701
2702 int r;
2703 Object *d;
2704
2705 assert(f);
2706
2707 r = journal_file_move_to_object(f, OBJECT_DATA, data_offset, &d);
2708 if (r < 0)
2709 return r;
2710
2711 return generic_array_bisect_plus_one(f,
2712 le64toh(d->data.entry_offset),
2713 le64toh(d->data.entry_array_offset),
2714 le64toh(d->data.n_entries),
2715 p,
2716 test_object_offset,
2717 direction,
2718 ret, offset, NULL);
2719}
2720
2721int journal_file_move_to_entry_by_monotonic_for_data(
2722 JournalFile *f,
2723 uint64_t data_offset,
2724 sd_id128_t boot_id,
2725 uint64_t monotonic,
2726 direction_t direction,
2727 Object **ret, uint64_t *offset) {
2728
2729 Object *o, *d;
2730 int r;
2731 uint64_t b, z;
2732
2733 assert(f);
2734
2735 /* First, seek by time */
2736 r = find_data_object_by_boot_id(f, boot_id, &o, &b);
2737 if (r < 0)
2738 return r;
2739 if (r == 0)
2740 return -ENOENT;
2741
2742 r = generic_array_bisect_plus_one(f,
2743 le64toh(o->data.entry_offset),
2744 le64toh(o->data.entry_array_offset),
2745 le64toh(o->data.n_entries),
2746 monotonic,
2747 test_object_monotonic,
2748 direction,
2749 NULL, &z, NULL);
2750 if (r <= 0)
2751 return r;
2752
2753 /* And now, continue seeking until we find an entry that
2754 * exists in both bisection arrays */
2755
2756 for (;;) {
2757 Object *qo;
2758 uint64_t p, q;
2759
2760 r = journal_file_move_to_object(f, OBJECT_DATA, data_offset, &d);
2761 if (r < 0)
2762 return r;
2763
2764 r = generic_array_bisect_plus_one(f,
2765 le64toh(d->data.entry_offset),
2766 le64toh(d->data.entry_array_offset),
2767 le64toh(d->data.n_entries),
2768 z,
2769 test_object_offset,
2770 direction,
2771 NULL, &p, NULL);
2772 if (r <= 0)
2773 return r;
2774
2775 r = journal_file_move_to_object(f, OBJECT_DATA, b, &o);
2776 if (r < 0)
2777 return r;
2778
2779 r = generic_array_bisect_plus_one(f,
2780 le64toh(o->data.entry_offset),
2781 le64toh(o->data.entry_array_offset),
2782 le64toh(o->data.n_entries),
2783 p,
2784 test_object_offset,
2785 direction,
2786 &qo, &q, NULL);
2787
2788 if (r <= 0)
2789 return r;
2790
2791 if (p == q) {
2792 if (ret)
2793 *ret = qo;
2794 if (offset)
2795 *offset = q;
2796
2797 return 1;
2798 }
2799
2800 z = q;
2801 }
2802}
2803
2804int journal_file_move_to_entry_by_seqnum_for_data(
2805 JournalFile *f,
2806 uint64_t data_offset,
2807 uint64_t seqnum,
2808 direction_t direction,
2809 Object **ret, uint64_t *offset) {
2810
2811 Object *d;
2812 int r;
2813
2814 assert(f);
2815
2816 r = journal_file_move_to_object(f, OBJECT_DATA, data_offset, &d);
2817 if (r < 0)
2818 return r;
2819
2820 return generic_array_bisect_plus_one(f,
2821 le64toh(d->data.entry_offset),
2822 le64toh(d->data.entry_array_offset),
2823 le64toh(d->data.n_entries),
2824 seqnum,
2825 test_object_seqnum,
2826 direction,
2827 ret, offset, NULL);
2828}
2829
2830int journal_file_move_to_entry_by_realtime_for_data(
2831 JournalFile *f,
2832 uint64_t data_offset,
2833 uint64_t realtime,
2834 direction_t direction,
2835 Object **ret, uint64_t *offset) {
2836
2837 Object *d;
2838 int r;
2839
2840 assert(f);
2841
2842 r = journal_file_move_to_object(f, OBJECT_DATA, data_offset, &d);
2843 if (r < 0)
2844 return r;
2845
2846 return generic_array_bisect_plus_one(f,
2847 le64toh(d->data.entry_offset),
2848 le64toh(d->data.entry_array_offset),
2849 le64toh(d->data.n_entries),
2850 realtime,
2851 test_object_realtime,
2852 direction,
2853 ret, offset, NULL);
2854}
2855
2856void journal_file_dump(JournalFile *f) {
2857 Object *o;
2858 int r;
2859 uint64_t p;
2860
2861 assert(f);
2862 assert(f->header);
2863
2864 journal_file_print_header(f);
2865
2866 p = le64toh(f->header->header_size);
2867 while (p != 0) {
2868 r = journal_file_move_to_object(f, OBJECT_UNUSED, p, &o);
2869 if (r < 0)
2870 goto fail;
2871
2872 switch (o->object.type) {
2873
2874 case OBJECT_UNUSED:
2875 printf("Type: OBJECT_UNUSED\n");
2876 break;
2877
2878 case OBJECT_DATA:
2879 printf("Type: OBJECT_DATA\n");
2880 break;
2881
2882 case OBJECT_FIELD:
2883 printf("Type: OBJECT_FIELD\n");
2884 break;
2885
2886 case OBJECT_ENTRY:
2887 printf("Type: OBJECT_ENTRY seqnum=%"PRIu64" monotonic=%"PRIu64" realtime=%"PRIu64"\n",
2888 le64toh(o->entry.seqnum),
2889 le64toh(o->entry.monotonic),
2890 le64toh(o->entry.realtime));
2891 break;
2892
2893 case OBJECT_FIELD_HASH_TABLE:
2894 printf("Type: OBJECT_FIELD_HASH_TABLE\n");
2895 break;
2896
2897 case OBJECT_DATA_HASH_TABLE:
2898 printf("Type: OBJECT_DATA_HASH_TABLE\n");
2899 break;
2900
2901 case OBJECT_ENTRY_ARRAY:
2902 printf("Type: OBJECT_ENTRY_ARRAY\n");
2903 break;
2904
2905 case OBJECT_TAG:
2906 printf("Type: OBJECT_TAG seqnum=%"PRIu64" epoch=%"PRIu64"\n",
2907 le64toh(o->tag.seqnum),
2908 le64toh(o->tag.epoch));
2909 break;
2910
2911 default:
2912 printf("Type: unknown (%i)\n", o->object.type);
2913 break;
2914 }
2915
2916 if (o->object.flags & OBJECT_COMPRESSION_MASK)
2917 printf("Flags: %s\n",
2918 object_compressed_to_string(o->object.flags & OBJECT_COMPRESSION_MASK));
2919
2920 if (p == le64toh(f->header->tail_object_offset))
2921 p = 0;
2922 else
2923 p = p + ALIGN64(le64toh(o->object.size));
2924 }
2925
2926 return;
2927fail:
2928 log_error("File corrupt");
2929}
2930
2931static const char* format_timestamp_safe(char *buf, size_t l, usec_t t) {
2932 const char *x;
2933
2934 x = format_timestamp(buf, l, t);
2935 if (x)
2936 return x;
2937 return " --- ";
2938}
2939
2940void journal_file_print_header(JournalFile *f) {
2941 char a[33], b[33], c[33], d[33];
2942 char x[FORMAT_TIMESTAMP_MAX], y[FORMAT_TIMESTAMP_MAX], z[FORMAT_TIMESTAMP_MAX];
2943 struct stat st;
2944 char bytes[FORMAT_BYTES_MAX];
2945
2946 assert(f);
2947 assert(f->header);
2948
2949 printf("File Path: %s\n"
2950 "File ID: %s\n"
2951 "Machine ID: %s\n"
2952 "Boot ID: %s\n"
2953 "Sequential Number ID: %s\n"
2954 "State: %s\n"
2955 "Compatible Flags:%s%s\n"
2956 "Incompatible Flags:%s%s%s\n"
2957 "Header size: %"PRIu64"\n"
2958 "Arena size: %"PRIu64"\n"
2959 "Data Hash Table Size: %"PRIu64"\n"
2960 "Field Hash Table Size: %"PRIu64"\n"
2961 "Rotate Suggested: %s\n"
2962 "Head Sequential Number: %"PRIu64" (%"PRIx64")\n"
2963 "Tail Sequential Number: %"PRIu64" (%"PRIx64")\n"
2964 "Head Realtime Timestamp: %s (%"PRIx64")\n"
2965 "Tail Realtime Timestamp: %s (%"PRIx64")\n"
2966 "Tail Monotonic Timestamp: %s (%"PRIx64")\n"
2967 "Objects: %"PRIu64"\n"
2968 "Entry Objects: %"PRIu64"\n",
2969 f->path,
2970 sd_id128_to_string(f->header->file_id, a),
2971 sd_id128_to_string(f->header->machine_id, b),
2972 sd_id128_to_string(f->header->boot_id, c),
2973 sd_id128_to_string(f->header->seqnum_id, d),
2974 f->header->state == STATE_OFFLINE ? "OFFLINE" :
2975 f->header->state == STATE_ONLINE ? "ONLINE" :
2976 f->header->state == STATE_ARCHIVED ? "ARCHIVED" : "UNKNOWN",
2977 JOURNAL_HEADER_SEALED(f->header) ? " SEALED" : "",
2978 (le32toh(f->header->compatible_flags) & ~HEADER_COMPATIBLE_ANY) ? " ???" : "",
2979 JOURNAL_HEADER_COMPRESSED_XZ(f->header) ? " COMPRESSED-XZ" : "",
2980 JOURNAL_HEADER_COMPRESSED_LZ4(f->header) ? " COMPRESSED-LZ4" : "",
2981 (le32toh(f->header->incompatible_flags) & ~HEADER_INCOMPATIBLE_ANY) ? " ???" : "",
2982 le64toh(f->header->header_size),
2983 le64toh(f->header->arena_size),
2984 le64toh(f->header->data_hash_table_size) / sizeof(HashItem),
2985 le64toh(f->header->field_hash_table_size) / sizeof(HashItem),
2986 yes_no(journal_file_rotate_suggested(f, 0)),
2987 le64toh(f->header->head_entry_seqnum), le64toh(f->header->head_entry_seqnum),
2988 le64toh(f->header->tail_entry_seqnum), le64toh(f->header->tail_entry_seqnum),
2989 format_timestamp_safe(x, sizeof(x), le64toh(f->header->head_entry_realtime)), le64toh(f->header->head_entry_realtime),
2990 format_timestamp_safe(y, sizeof(y), le64toh(f->header->tail_entry_realtime)), le64toh(f->header->tail_entry_realtime),
2991 format_timespan(z, sizeof(z), le64toh(f->header->tail_entry_monotonic), USEC_PER_MSEC), le64toh(f->header->tail_entry_monotonic),
2992 le64toh(f->header->n_objects),
2993 le64toh(f->header->n_entries));
2994
2995 if (JOURNAL_HEADER_CONTAINS(f->header, n_data))
2996 printf("Data Objects: %"PRIu64"\n"
2997 "Data Hash Table Fill: %.1f%%\n",
2998 le64toh(f->header->n_data),
2999 100.0 * (double) le64toh(f->header->n_data) / ((double) (le64toh(f->header->data_hash_table_size) / sizeof(HashItem))));
3000
3001 if (JOURNAL_HEADER_CONTAINS(f->header, n_fields))
3002 printf("Field Objects: %"PRIu64"\n"
3003 "Field Hash Table Fill: %.1f%%\n",
3004 le64toh(f->header->n_fields),
3005 100.0 * (double) le64toh(f->header->n_fields) / ((double) (le64toh(f->header->field_hash_table_size) / sizeof(HashItem))));
3006
3007 if (JOURNAL_HEADER_CONTAINS(f->header, n_tags))
3008 printf("Tag Objects: %"PRIu64"\n",
3009 le64toh(f->header->n_tags));
3010 if (JOURNAL_HEADER_CONTAINS(f->header, n_entry_arrays))
3011 printf("Entry Array Objects: %"PRIu64"\n",
3012 le64toh(f->header->n_entry_arrays));
3013
3014 if (fstat(f->fd, &st) >= 0)
3015 printf("Disk usage: %s\n", format_bytes(bytes, sizeof(bytes), (uint64_t) st.st_blocks * 512ULL));
3016}
3017
3018static int journal_file_warn_btrfs(JournalFile *f) {
3019 unsigned attrs;
3020 int r;
3021
3022 assert(f);
3023
3024 /* Before we write anything, check if the COW logic is turned
3025 * off on btrfs. Given our write pattern that is quite
3026 * unfriendly to COW file systems this should greatly improve
3027 * performance on COW file systems, such as btrfs, at the
3028 * expense of data integrity features (which shouldn't be too
3029 * bad, given that we do our own checksumming). */
3030
3031 r = btrfs_is_filesystem(f->fd);
3032 if (r < 0)
3033 return log_warning_errno(r, "Failed to determine if journal is on btrfs: %m");
3034 if (!r)
3035 return 0;
3036
3037 r = read_attr_fd(f->fd, &attrs);
3038 if (r < 0)
3039 return log_warning_errno(r, "Failed to read file attributes: %m");
3040
3041 if (attrs & FS_NOCOW_FL) {
3042 log_debug("Detected btrfs file system with copy-on-write disabled, all is good.");
3043 return 0;
3044 }
3045
3046 log_notice("Creating journal file %s on a btrfs file system, and copy-on-write is enabled. "
3047 "This is likely to slow down journal access substantially, please consider turning "
3048 "off the copy-on-write file attribute on the journal directory, using chattr +C.", f->path);
3049
3050 return 1;
3051}
3052
3053int journal_file_open(
3054 int fd,
3055 const char *fname,
3056 int flags,
3057 mode_t mode,
3058 bool compress,
3059 bool seal,
3060 JournalMetrics *metrics,
3061 MMapCache *mmap_cache,
3062 Set *deferred_closes,
3063 JournalFile *template,
3064 JournalFile **ret) {
3065
3066 bool newly_created = false;
3067 JournalFile *f;
3068 void *h;
3069 int r;
3070
3071 assert(ret);
3072 assert(fd >= 0 || fname);
3073
3074 if ((flags & O_ACCMODE) != O_RDONLY &&
3075 (flags & O_ACCMODE) != O_RDWR)
3076 return -EINVAL;
3077
3078 if (fname) {
3079 if (!endswith(fname, ".journal") &&
3080 !endswith(fname, ".journal~"))
3081 return -EINVAL;
3082 }
3083
3084 f = new0(JournalFile, 1);
3085 if (!f)
3086 return -ENOMEM;
3087
3088 f->fd = fd;
3089 f->mode = mode;
3090
3091 f->flags = flags;
3092 f->prot = prot_from_flags(flags);
3093 f->writable = (flags & O_ACCMODE) != O_RDONLY;
3094#if defined(HAVE_LZ4)
3095 f->compress_lz4 = compress;
3096#elif defined(HAVE_XZ)
3097 f->compress_xz = compress;
3098#endif
3099#ifdef HAVE_GCRYPT
3100 f->seal = seal;
3101#endif
3102
3103 if (mmap_cache)
3104 f->mmap = mmap_cache_ref(mmap_cache);
3105 else {
3106 f->mmap = mmap_cache_new();
3107 if (!f->mmap) {
3108 r = -ENOMEM;
3109 goto fail;
3110 }
3111 }
3112
3113 if (fname) {
3114 f->path = strdup(fname);
3115 if (!f->path) {
3116 r = -ENOMEM;
3117 goto fail;
3118 }
3119 } else {
3120 /* If we don't know the path, fill in something explanatory and vaguely useful */
3121 if (asprintf(&f->path, "/proc/self/%i", fd) < 0) {
3122 r = -ENOMEM;
3123 goto fail;
3124 }
3125 }
3126
3127 f->chain_cache = ordered_hashmap_new(&uint64_hash_ops);
3128 if (!f->chain_cache) {
3129 r = -ENOMEM;
3130 goto fail;
3131 }
3132
3133 if (f->fd < 0) {
3134 f->fd = open(f->path, f->flags|O_CLOEXEC, f->mode);
3135 if (f->fd < 0) {
3136 r = -errno;
3137 goto fail;
3138 }
3139
3140 /* fds we opened here by us should also be closed by us. */
3141 f->close_fd = true;
3142 }
3143
3144 r = journal_file_fstat(f);
3145 if (r < 0)
3146 goto fail;
3147
3148 if (f->last_stat.st_size == 0 && f->writable) {
3149
3150 (void) journal_file_warn_btrfs(f);
3151
3152 /* Let's attach the creation time to the journal file,
3153 * so that the vacuuming code knows the age of this
3154 * file even if the file might end up corrupted one
3155 * day... Ideally we'd just use the creation time many
3156 * file systems maintain for each file, but there is
3157 * currently no usable API to query this, hence let's
3158 * emulate this via extended attributes. If extended
3159 * attributes are not supported we'll just skip this,
3160 * and rely solely on mtime/atime/ctime of the file. */
3161
3162 fd_setcrtime(f->fd, 0);
3163
3164#ifdef HAVE_GCRYPT
3165 /* Try to load the FSPRG state, and if we can't, then
3166 * just don't do sealing */
3167 if (f->seal) {
3168 r = journal_file_fss_load(f);
3169 if (r < 0)
3170 f->seal = false;
3171 }
3172#endif
3173
3174 r = journal_file_init_header(f, template);
3175 if (r < 0)
3176 goto fail;
3177
3178 r = journal_file_fstat(f);
3179 if (r < 0)
3180 goto fail;
3181
3182 newly_created = true;
3183 }
3184
3185 if (f->last_stat.st_size < (off_t) HEADER_SIZE_MIN) {
3186 r = -ENODATA;
3187 goto fail;
3188 }
3189
3190 r = mmap_cache_get(f->mmap, f->fd, f->prot, CONTEXT_HEADER, true, 0, PAGE_ALIGN(sizeof(Header)), &f->last_stat, &h);
3191 if (r < 0)
3192 goto fail;
3193
3194 f->header = h;
3195
3196 if (!newly_created) {
3197 if (deferred_closes)
3198 journal_file_close_set(deferred_closes);
3199
3200 r = journal_file_verify_header(f);
3201 if (r < 0)
3202 goto fail;
3203 }
3204
3205#ifdef HAVE_GCRYPT
3206 if (!newly_created && f->writable) {
3207 r = journal_file_fss_load(f);
3208 if (r < 0)
3209 goto fail;
3210 }
3211#endif
3212
3213 if (f->writable) {
3214 if (metrics) {
3215 journal_default_metrics(metrics, f->fd);
3216 f->metrics = *metrics;
3217 } else if (template)
3218 f->metrics = template->metrics;
3219
3220 r = journal_file_refresh_header(f);
3221 if (r < 0)
3222 goto fail;
3223 }
3224
3225#ifdef HAVE_GCRYPT
3226 r = journal_file_hmac_setup(f);
3227 if (r < 0)
3228 goto fail;
3229#endif
3230
3231 if (newly_created) {
3232 r = journal_file_setup_field_hash_table(f);
3233 if (r < 0)
3234 goto fail;
3235
3236 r = journal_file_setup_data_hash_table(f);
3237 if (r < 0)
3238 goto fail;
3239
3240#ifdef HAVE_GCRYPT
3241 r = journal_file_append_first_tag(f);
3242 if (r < 0)
3243 goto fail;
3244#endif
3245 }
3246
3247 if (mmap_cache_got_sigbus(f->mmap, f->fd)) {
3248 r = -EIO;
3249 goto fail;
3250 }
3251
3252 if (template && template->post_change_timer) {
3253 r = journal_file_enable_post_change_timer(
3254 f,
3255 sd_event_source_get_event(template->post_change_timer),
3256 template->post_change_timer_period);
3257
3258 if (r < 0)
3259 goto fail;
3260 }
3261
3262 /* The file is opened now successfully, thus we take possession of any passed in fd. */
3263 f->close_fd = true;
3264
3265 *ret = f;
3266 return 0;
3267
3268fail:
3269 if (f->fd >= 0 && mmap_cache_got_sigbus(f->mmap, f->fd))
3270 r = -EIO;
3271
3272 (void) journal_file_close(f);
3273
3274 return r;
3275}
3276
3277int journal_file_rotate(JournalFile **f, bool compress, bool seal, Set *deferred_closes) {
3278 _cleanup_free_ char *p = NULL;
3279 size_t l;
3280 JournalFile *old_file, *new_file = NULL;
3281 int r;
3282
3283 assert(f);
3284 assert(*f);
3285
3286 old_file = *f;
3287
3288 if (!old_file->writable)
3289 return -EINVAL;
3290
3291 /* Is this a journal file that was passed to us as fd? If so, we synthesized a path name for it, and we refuse
3292 * rotation, since we don't know the actual path, and couldn't rename the file hence. */
3293 if (path_startswith(old_file->path, "/proc/self/fd"))
3294 return -EINVAL;
3295
3296 if (!endswith(old_file->path, ".journal"))
3297 return -EINVAL;
3298
3299 l = strlen(old_file->path);
3300 r = asprintf(&p, "%.*s@" SD_ID128_FORMAT_STR "-%016"PRIx64"-%016"PRIx64".journal",
3301 (int) l - 8, old_file->path,
3302 SD_ID128_FORMAT_VAL(old_file->header->seqnum_id),
3303 le64toh((*f)->header->head_entry_seqnum),
3304 le64toh((*f)->header->head_entry_realtime));
3305 if (r < 0)
3306 return -ENOMEM;
3307
3308 /* Try to rename the file to the archived version. If the file
3309 * already was deleted, we'll get ENOENT, let's ignore that
3310 * case. */
3311 r = rename(old_file->path, p);
3312 if (r < 0 && errno != ENOENT)
3313 return -errno;
3314
3315 /* Sync the rename to disk */
3316 (void) fsync_directory_of_file(old_file->fd);
3317
3318 /* Set as archive so offlining commits w/state=STATE_ARCHIVED.
3319 * Previously we would set old_file->header->state to STATE_ARCHIVED directly here,
3320 * but journal_file_set_offline() short-circuits when state != STATE_ONLINE, which
3321 * would result in the rotated journal never getting fsync() called before closing.
3322 * Now we simply queue the archive state by setting an archive bit, leaving the state
3323 * as STATE_ONLINE so proper offlining occurs. */
3324 old_file->archive = true;
3325
3326 /* Currently, btrfs is not very good with out write patterns
3327 * and fragments heavily. Let's defrag our journal files when
3328 * we archive them */
3329 old_file->defrag_on_close = true;
3330
3331 r = journal_file_open(-1, old_file->path, old_file->flags, old_file->mode, compress, seal, NULL, old_file->mmap, deferred_closes, old_file, &new_file);
3332
3333 if (deferred_closes &&
3334 set_put(deferred_closes, old_file) >= 0)
3335 (void) journal_file_set_offline(old_file, false);
3336 else
3337 (void) journal_file_close(old_file);
3338
3339 *f = new_file;
3340 return r;
3341}
3342
3343int journal_file_open_reliably(
3344 const char *fname,
3345 int flags,
3346 mode_t mode,
3347 bool compress,
3348 bool seal,
3349 JournalMetrics *metrics,
3350 MMapCache *mmap_cache,
3351 Set *deferred_closes,
3352 JournalFile *template,
3353 JournalFile **ret) {
3354
3355 int r;
3356 size_t l;
3357 _cleanup_free_ char *p = NULL;
3358
3359 r = journal_file_open(-1, fname, flags, mode, compress, seal, metrics, mmap_cache, deferred_closes, template, ret);
3360 if (!IN_SET(r,
3361 -EBADMSG, /* Corrupted */
3362 -ENODATA, /* Truncated */
3363 -EHOSTDOWN, /* Other machine */
3364 -EPROTONOSUPPORT, /* Incompatible feature */
3365 -EBUSY, /* Unclean shutdown */
3366 -ESHUTDOWN, /* Already archived */
3367 -EIO, /* IO error, including SIGBUS on mmap */
3368 -EIDRM, /* File has been deleted */
3369 -ETXTBSY)) /* File is from the future */
3370 return r;
3371
3372 if ((flags & O_ACCMODE) == O_RDONLY)
3373 return r;
3374
3375 if (!(flags & O_CREAT))
3376 return r;
3377
3378 if (!endswith(fname, ".journal"))
3379 return r;
3380
3381 /* The file is corrupted. Rotate it away and try it again (but only once) */
3382
3383 l = strlen(fname);
3384 if (asprintf(&p, "%.*s@%016"PRIx64 "-%016"PRIx64 ".journal~",
3385 (int) l - 8, fname,
3386 now(CLOCK_REALTIME),
3387 random_u64()) < 0)
3388 return -ENOMEM;
3389
3390 if (rename(fname, p) < 0)
3391 return -errno;
3392
3393 /* btrfs doesn't cope well with our write pattern and
3394 * fragments heavily. Let's defrag all files we rotate */
3395
3396 (void) chattr_path(p, 0, FS_NOCOW_FL);
3397 (void) btrfs_defrag(p);
3398
3399 log_warning_errno(r, "File %s corrupted or uncleanly shut down, renaming and replacing.", fname);
3400
3401 return journal_file_open(-1, fname, flags, mode, compress, seal, metrics, mmap_cache, deferred_closes, template, ret);
3402}
3403
3404int journal_file_copy_entry(JournalFile *from, JournalFile *to, Object *o, uint64_t p, uint64_t *seqnum, Object **ret, uint64_t *offset) {
3405 uint64_t i, n;
3406 uint64_t q, xor_hash = 0;
3407 int r;
3408 EntryItem *items;
3409 dual_timestamp ts;
3410
3411 assert(from);
3412 assert(to);
3413 assert(o);
3414 assert(p);
3415
3416 if (!to->writable)
3417 return -EPERM;
3418
3419 ts.monotonic = le64toh(o->entry.monotonic);
3420 ts.realtime = le64toh(o->entry.realtime);
3421
3422 n = journal_file_entry_n_items(o);
3423 /* alloca() can't take 0, hence let's allocate at least one */
3424 items = alloca(sizeof(EntryItem) * MAX(1u, n));
3425
3426 for (i = 0; i < n; i++) {
3427 uint64_t l, h;
3428 le64_t le_hash;
3429 size_t t;
3430 void *data;
3431 Object *u;
3432
3433 q = le64toh(o->entry.items[i].object_offset);
3434 le_hash = o->entry.items[i].hash;
3435
3436 r = journal_file_move_to_object(from, OBJECT_DATA, q, &o);
3437 if (r < 0)
3438 return r;
3439
3440 if (le_hash != o->data.hash)
3441 return -EBADMSG;
3442
3443 l = le64toh(o->object.size) - offsetof(Object, data.payload);
3444 t = (size_t) l;
3445
3446 /* We hit the limit on 32bit machines */
3447 if ((uint64_t) t != l)
3448 return -E2BIG;
3449
3450 if (o->object.flags & OBJECT_COMPRESSION_MASK) {
3451#if defined(HAVE_XZ) || defined(HAVE_LZ4)
3452 size_t rsize = 0;
3453
3454 r = decompress_blob(o->object.flags & OBJECT_COMPRESSION_MASK,
3455 o->data.payload, l, &from->compress_buffer, &from->compress_buffer_size, &rsize, 0);
3456 if (r < 0)
3457 return r;
3458
3459 data = from->compress_buffer;
3460 l = rsize;
3461#else
3462 return -EPROTONOSUPPORT;
3463#endif
3464 } else
3465 data = o->data.payload;
3466
3467 r = journal_file_append_data(to, data, l, &u, &h);
3468 if (r < 0)
3469 return r;
3470
3471 xor_hash ^= le64toh(u->data.hash);
3472 items[i].object_offset = htole64(h);
3473 items[i].hash = u->data.hash;
3474
3475 r = journal_file_move_to_object(from, OBJECT_ENTRY, p, &o);
3476 if (r < 0)
3477 return r;
3478 }
3479
3480 r = journal_file_append_entry_internal(to, &ts, xor_hash, items, n, seqnum, ret, offset);
3481
3482 if (mmap_cache_got_sigbus(to->mmap, to->fd))
3483 return -EIO;
3484
3485 return r;
3486}
3487
3488void journal_reset_metrics(JournalMetrics *m) {
3489 assert(m);
3490
3491 /* Set everything to "pick automatic values". */
3492
3493 *m = (JournalMetrics) {
3494 .min_use = (uint64_t) -1,
3495 .max_use = (uint64_t) -1,
3496 .min_size = (uint64_t) -1,
3497 .max_size = (uint64_t) -1,
3498 .keep_free = (uint64_t) -1,
3499 .n_max_files = (uint64_t) -1,
3500 };
3501}
3502
3503void journal_default_metrics(JournalMetrics *m, int fd) {
3504 char a[FORMAT_BYTES_MAX], b[FORMAT_BYTES_MAX], c[FORMAT_BYTES_MAX], d[FORMAT_BYTES_MAX], e[FORMAT_BYTES_MAX];
3505 struct statvfs ss;
3506 uint64_t fs_size;
3507
3508 assert(m);
3509 assert(fd >= 0);
3510
3511 if (fstatvfs(fd, &ss) >= 0)
3512 fs_size = ss.f_frsize * ss.f_blocks;
3513 else {
3514 log_debug_errno(errno, "Failed to detremine disk size: %m");
3515 fs_size = 0;
3516 }
3517
3518 if (m->max_use == (uint64_t) -1) {
3519
3520 if (fs_size > 0) {
3521 m->max_use = PAGE_ALIGN(fs_size / 10); /* 10% of file system size */
3522
3523 if (m->max_use > DEFAULT_MAX_USE_UPPER)
3524 m->max_use = DEFAULT_MAX_USE_UPPER;
3525
3526 if (m->max_use < DEFAULT_MAX_USE_LOWER)
3527 m->max_use = DEFAULT_MAX_USE_LOWER;
3528 } else
3529 m->max_use = DEFAULT_MAX_USE_LOWER;
3530 } else {
3531 m->max_use = PAGE_ALIGN(m->max_use);
3532
3533 if (m->max_use != 0 && m->max_use < JOURNAL_FILE_SIZE_MIN*2)
3534 m->max_use = JOURNAL_FILE_SIZE_MIN*2;
3535 }
3536
3537 if (m->min_use == (uint64_t) -1)
3538 m->min_use = DEFAULT_MIN_USE;
3539
3540 if (m->min_use > m->max_use)
3541 m->min_use = m->max_use;
3542
3543 if (m->max_size == (uint64_t) -1) {
3544 m->max_size = PAGE_ALIGN(m->max_use / 8); /* 8 chunks */
3545
3546 if (m->max_size > DEFAULT_MAX_SIZE_UPPER)
3547 m->max_size = DEFAULT_MAX_SIZE_UPPER;
3548 } else
3549 m->max_size = PAGE_ALIGN(m->max_size);
3550
3551 if (m->max_size != 0) {
3552 if (m->max_size < JOURNAL_FILE_SIZE_MIN)
3553 m->max_size = JOURNAL_FILE_SIZE_MIN;
3554
3555 if (m->max_use != 0 && m->max_size*2 > m->max_use)
3556 m->max_use = m->max_size*2;
3557 }
3558
3559 if (m->min_size == (uint64_t) -1)
3560 m->min_size = JOURNAL_FILE_SIZE_MIN;
3561 else {
3562 m->min_size = PAGE_ALIGN(m->min_size);
3563
3564 if (m->min_size < JOURNAL_FILE_SIZE_MIN)
3565 m->min_size = JOURNAL_FILE_SIZE_MIN;
3566
3567 if (m->max_size != 0 && m->min_size > m->max_size)
3568 m->max_size = m->min_size;
3569 }
3570
3571 if (m->keep_free == (uint64_t) -1) {
3572
3573 if (fs_size > 0) {
3574 m->keep_free = PAGE_ALIGN(fs_size * 3 / 20); /* 15% of file system size */
3575
3576 if (m->keep_free > DEFAULT_KEEP_FREE_UPPER)
3577 m->keep_free = DEFAULT_KEEP_FREE_UPPER;
3578
3579 } else
3580 m->keep_free = DEFAULT_KEEP_FREE;
3581 }
3582
3583 if (m->n_max_files == (uint64_t) -1)
3584 m->n_max_files = DEFAULT_N_MAX_FILES;
3585
3586 log_debug("Fixed min_use=%s max_use=%s max_size=%s min_size=%s keep_free=%s n_max_files=%" PRIu64,
3587 format_bytes(a, sizeof(a), m->min_use),
3588 format_bytes(b, sizeof(b), m->max_use),
3589 format_bytes(c, sizeof(c), m->max_size),
3590 format_bytes(d, sizeof(d), m->min_size),
3591 format_bytes(e, sizeof(e), m->keep_free),
3592 m->n_max_files);
3593}
3594
3595int journal_file_get_cutoff_realtime_usec(JournalFile *f, usec_t *from, usec_t *to) {
3596 assert(f);
3597 assert(f->header);
3598 assert(from || to);
3599
3600 if (from) {
3601 if (f->header->head_entry_realtime == 0)
3602 return -ENOENT;
3603
3604 *from = le64toh(f->header->head_entry_realtime);
3605 }
3606
3607 if (to) {
3608 if (f->header->tail_entry_realtime == 0)
3609 return -ENOENT;
3610
3611 *to = le64toh(f->header->tail_entry_realtime);
3612 }
3613
3614 return 1;
3615}
3616
3617int journal_file_get_cutoff_monotonic_usec(JournalFile *f, sd_id128_t boot_id, usec_t *from, usec_t *to) {
3618 Object *o;
3619 uint64_t p;
3620 int r;
3621
3622 assert(f);
3623 assert(from || to);
3624
3625 r = find_data_object_by_boot_id(f, boot_id, &o, &p);
3626 if (r <= 0)
3627 return r;
3628
3629 if (le64toh(o->data.n_entries) <= 0)
3630 return 0;
3631
3632 if (from) {
3633 r = journal_file_move_to_object(f, OBJECT_ENTRY, le64toh(o->data.entry_offset), &o);
3634 if (r < 0)
3635 return r;
3636
3637 *from = le64toh(o->entry.monotonic);
3638 }
3639
3640 if (to) {
3641 r = journal_file_move_to_object(f, OBJECT_DATA, p, &o);
3642 if (r < 0)
3643 return r;
3644
3645 r = generic_array_get_plus_one(f,
3646 le64toh(o->data.entry_offset),
3647 le64toh(o->data.entry_array_offset),
3648 le64toh(o->data.n_entries)-1,
3649 &o, NULL);
3650 if (r <= 0)
3651 return r;
3652
3653 *to = le64toh(o->entry.monotonic);
3654 }
3655
3656 return 1;
3657}
3658
3659bool journal_file_rotate_suggested(JournalFile *f, usec_t max_file_usec) {
3660 assert(f);
3661 assert(f->header);
3662
3663 /* If we gained new header fields we gained new features,
3664 * hence suggest a rotation */
3665 if (le64toh(f->header->header_size) < sizeof(Header)) {
3666 log_debug("%s uses an outdated header, suggesting rotation.", f->path);
3667 return true;
3668 }
3669
3670 /* Let's check if the hash tables grew over a certain fill
3671 * level (75%, borrowing this value from Java's hash table
3672 * implementation), and if so suggest a rotation. To calculate
3673 * the fill level we need the n_data field, which only exists
3674 * in newer versions. */
3675
3676 if (JOURNAL_HEADER_CONTAINS(f->header, n_data))
3677 if (le64toh(f->header->n_data) * 4ULL > (le64toh(f->header->data_hash_table_size) / sizeof(HashItem)) * 3ULL) {
3678 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.",
3679 f->path,
3680 100.0 * (double) le64toh(f->header->n_data) / ((double) (le64toh(f->header->data_hash_table_size) / sizeof(HashItem))),
3681 le64toh(f->header->n_data),
3682 le64toh(f->header->data_hash_table_size) / sizeof(HashItem),
3683 (unsigned long long) f->last_stat.st_size,
3684 f->last_stat.st_size / le64toh(f->header->n_data));
3685 return true;
3686 }
3687
3688 if (JOURNAL_HEADER_CONTAINS(f->header, n_fields))
3689 if (le64toh(f->header->n_fields) * 4ULL > (le64toh(f->header->field_hash_table_size) / sizeof(HashItem)) * 3ULL) {
3690 log_debug("Field hash table of %s has a fill level at %.1f (%"PRIu64" of %"PRIu64" items), suggesting rotation.",
3691 f->path,
3692 100.0 * (double) le64toh(f->header->n_fields) / ((double) (le64toh(f->header->field_hash_table_size) / sizeof(HashItem))),
3693 le64toh(f->header->n_fields),
3694 le64toh(f->header->field_hash_table_size) / sizeof(HashItem));
3695 return true;
3696 }
3697
3698 /* Are the data objects properly indexed by field objects? */
3699 if (JOURNAL_HEADER_CONTAINS(f->header, n_data) &&
3700 JOURNAL_HEADER_CONTAINS(f->header, n_fields) &&
3701 le64toh(f->header->n_data) > 0 &&
3702 le64toh(f->header->n_fields) == 0)
3703 return true;
3704
3705 if (max_file_usec > 0) {
3706 usec_t t, h;
3707
3708 h = le64toh(f->header->head_entry_realtime);
3709 t = now(CLOCK_REALTIME);
3710
3711 if (h > 0 && t > h + max_file_usec)
3712 return true;
3713 }
3714
3715 return false;
3716}
Unnamed repository; edit this file 'description' to name the repository.