]> git.ipfire.org Git - thirdparty/systemd.git/blame - src/journal/journal-file.c
projects / thirdparty / systemd.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
journal: when we encounter a broken journal file, add some debug logging
[thirdparty/systemd.git] / src / journal / journal-file.c
CommitLineData
cec736d2
LP		 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
5430f7f2
LP		 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
cec736d2
LP		 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
5430f7f2 14 Lesser General Public License for more details.
cec736d2 15
5430f7f2 16 You should have received a copy of the GNU Lesser General Public License
cec736d2
LP		 17 along with systemd; If not, see <http://www.gnu.org/licenses/>.
18***/
19
cec736d2 20#include <errno.h>
cec736d2 21#include <fcntl.h>
11689d2a 22#include <linux/fs.h>
ac2e41f5 23#include <pthread.h>
07630cea
LP		 24#include <stddef.h>
25#include <sys/mman.h>
26#include <sys/statvfs.h>
27#include <sys/uio.h>
28#include <unistd.h>
fb0951b0 29
b5efdb8a 30#include "alloc-util.h"
f27a3864 31#include "btrfs-util.h"
c8b3094d 32#include "chattr-util.h"
07630cea 33#include "compress.h"
3ffd4af2 34#include "fd-util.h"
0284adc6 35#include "journal-authenticate.h"
cec736d2
LP		 36#include "journal-def.h"
37#include "journal-file.h"
38#include "lookup3.h"
6bedfcbb 39#include "parse-util.h"
5d1ce257 40#include "path-util.h"
3df3e884 41#include "random-util.h"
7a24f3bf 42#include "sd-event.h"
b58c888f 43#include "set.h"
07630cea 44#include "string-util.h"
89a5a90c 45#include "xattr-util.h"
cec736d2 46
4a92baf3
LP		 47#define DEFAULT_DATA_HASH_TABLE_SIZE (2047ULL*sizeof(HashItem))
48#define DEFAULT_FIELD_HASH_TABLE_SIZE (333ULL*sizeof(HashItem))
cec736d2 49
be19b7df 50#define COMPRESSION_SIZE_THRESHOLD (512ULL)
807e17f0 51
babfc091 52/* This is the minimum journal file size */
16098e93 53#define JOURNAL_FILE_SIZE_MIN (512ULL*1024ULL) /* 512 KiB */
babfc091
LP		 54
55/* These are the lower and upper bounds if we deduce the max_use value
56 * from the file system size */
57#define DEFAULT_MAX_USE_LOWER (1ULL*1024ULL*1024ULL) /* 1 MiB */
58#define DEFAULT_MAX_USE_UPPER (4ULL*1024ULL*1024ULL*1024ULL) /* 4 GiB */
59
8580d1f7
LP		 60/* This is the default minimal use limit, how much we'll use even if keep_free suggests otherwise. */
61#define DEFAULT_MIN_USE (1ULL*1024ULL*1024ULL) /* 1 MiB */
62
babfc091 63/* This is the upper bound if we deduce max_size from max_use */
71100051 64#define DEFAULT_MAX_SIZE_UPPER (128ULL*1024ULL*1024ULL) /* 128 MiB */
babfc091
LP		 65
66/* This is the upper bound if we deduce the keep_free value from the
67 * file system size */
68#define DEFAULT_KEEP_FREE_UPPER (4ULL*1024ULL*1024ULL*1024ULL) /* 4 GiB */
69
70/* This is the keep_free value when we can't determine the system
71 * size */
72#define DEFAULT_KEEP_FREE (1024ULL*1024ULL) /* 1 MB */
73
8580d1f7
LP		 74/* This is the default maximum number of journal files to keep around. */
75#define DEFAULT_N_MAX_FILES (100)
76
dca6219e
LP		 77/* n_data was the first entry we added after the initial file format design */
78#define HEADER_SIZE_MIN ALIGN64(offsetof(Header, n_data))
cec736d2 79
a4bcff5b
LP		 80/* How many entries to keep in the entry array chain cache at max */
81#define CHAIN_CACHE_MAX 20
82
a676e665
LP		 83/* How much to increase the journal file size at once each time we allocate something new. */
84#define FILE_SIZE_INCREASE (8ULL*1024ULL*1024ULL) /* 8MB */
85
2678031a
LP		 86/* Reread fstat() of the file for detecting deletions at least this often */
87#define LAST_STAT_REFRESH_USEC (5*USEC_PER_SEC)
88
fa6ac760
LP		 89/* The mmap context to use for the header we pick as one above the last defined typed */
90#define CONTEXT_HEADER _OBJECT_TYPE_MAX
91
ac2e41f5
VC		 92/* This may be called from a separate thread to prevent blocking the caller for the duration of fsync().
93 * As a result we use atomic operations on f->offline_state for inter-thread communications with
94 * journal_file_set_offline() and journal_file_set_online(). */
95static void journal_file_set_offline_internal(JournalFile *f) {
26687bf8 96 assert(f);
ac2e41f5
VC		 97 assert(f->fd >= 0);
98 assert(f->header);
99
100 for (;;) {
101 switch (f->offline_state) {
102 case OFFLINE_CANCEL:
103 if (!__sync_bool_compare_and_swap(&f->offline_state, OFFLINE_CANCEL, OFFLINE_DONE))
104 continue;
105 return;
106
107 case OFFLINE_AGAIN_FROM_SYNCING:
108 if (!__sync_bool_compare_and_swap(&f->offline_state, OFFLINE_AGAIN_FROM_SYNCING, OFFLINE_SYNCING))
109 continue;
110 break;
111
112 case OFFLINE_AGAIN_FROM_OFFLINING:
113 if (!__sync_bool_compare_and_swap(&f->offline_state, OFFLINE_AGAIN_FROM_OFFLINING, OFFLINE_SYNCING))
114 continue;
115 break;
116
117 case OFFLINE_SYNCING:
118 (void) fsync(f->fd);
26687bf8 119
ac2e41f5
VC		 120 if (!__sync_bool_compare_and_swap(&f->offline_state, OFFLINE_SYNCING, OFFLINE_OFFLINING))
121 continue;
26687bf8 122
8eb85171 123 f->header->state = f->archive ? STATE_ARCHIVED : STATE_OFFLINE;
ac2e41f5
VC		 124 (void) fsync(f->fd);
125 break;
126
127 case OFFLINE_OFFLINING:
128 if (!__sync_bool_compare_and_swap(&f->offline_state, OFFLINE_OFFLINING, OFFLINE_DONE))
129 continue;
130 /* fall through */
131
132 case OFFLINE_DONE:
133 return;
134
135 case OFFLINE_JOINED:
136 log_debug("OFFLINE_JOINED unexpected offline state for journal_file_set_offline_internal()");
137 return;
138 }
139 }
140}
141
142static void * journal_file_set_offline_thread(void *arg) {
143 JournalFile *f = arg;
144
145 journal_file_set_offline_internal(f);
146
147 return NULL;
148}
149
150static int journal_file_set_offline_thread_join(JournalFile *f) {
151 int r;
152
153 assert(f);
154
155 if (f->offline_state == OFFLINE_JOINED)
156 return 0;
157
158 r = pthread_join(f->offline_thread, NULL);
159 if (r)
160 return -r;
161
162 f->offline_state = OFFLINE_JOINED;
26687bf8 163
fa6ac760
LP		 164 if (mmap_cache_got_sigbus(f->mmap, f->fd))
165 return -EIO;
166
ac2e41f5
VC		 167 return 0;
168}
26687bf8 169
ac2e41f5
VC		 170/* Trigger a restart if the offline thread is mid-flight in a restartable state. */
171static bool journal_file_set_offline_try_restart(JournalFile *f) {
172 for (;;) {
173 switch (f->offline_state) {
174 case OFFLINE_AGAIN_FROM_SYNCING:
175 case OFFLINE_AGAIN_FROM_OFFLINING:
176 return true;
177
178 case OFFLINE_CANCEL:
179 if (!__sync_bool_compare_and_swap(&f->offline_state, OFFLINE_CANCEL, OFFLINE_AGAIN_FROM_SYNCING))
180 continue;
181 return true;
182
183 case OFFLINE_SYNCING:
184 if (!__sync_bool_compare_and_swap(&f->offline_state, OFFLINE_SYNCING, OFFLINE_AGAIN_FROM_SYNCING))
185 continue;
186 return true;
187
188 case OFFLINE_OFFLINING:
189 if (!__sync_bool_compare_and_swap(&f->offline_state, OFFLINE_OFFLINING, OFFLINE_AGAIN_FROM_OFFLINING))
190 continue;
191 return true;
26687bf8
OS		 192
193 default:
ac2e41f5
VC		 194 return false;
195 }
26687bf8
OS		 196 }
197}
198
ac2e41f5
VC		 199/* Sets a journal offline.
200 *
201 * If wait is false then an offline is dispatched in a separate thread for a
202 * subsequent journal_file_set_offline() or journal_file_set_online() of the
203 * same journal to synchronize with.
204 *
205 * If wait is true, then either an existing offline thread will be restarted
206 * and joined, or if none exists the offline is simply performed in this
207 * context without involving another thread.
208 */
209int journal_file_set_offline(JournalFile *f, bool wait) {
210 bool restarted;
211 int r;
212
26687bf8
OS		 213 assert(f);
214
215 if (!f->writable)
216 return -EPERM;
217
218 if (!(f->fd >= 0 && f->header))
219 return -EINVAL;
220
b8f99e27
VC		 221 /* An offlining journal is implicitly online and may modify f->header->state,
222 * we must also join any potentially lingering offline thread when not online. */
223 if (!journal_file_is_offlining(f) && f->header->state != STATE_ONLINE)
224 return journal_file_set_offline_thread_join(f);
26687bf8 225
ac2e41f5
VC		 226 /* Restart an in-flight offline thread and wait if needed, or join a lingering done one. */
227 restarted = journal_file_set_offline_try_restart(f);
228 if ((restarted && wait) || !restarted) {
229 r = journal_file_set_offline_thread_join(f);
230 if (r < 0)
231 return r;
232 }
26687bf8 233
ac2e41f5
VC		 234 if (restarted)
235 return 0;
236
237 /* Initiate a new offline. */
238 f->offline_state = OFFLINE_SYNCING;
fa6ac760 239
ac2e41f5
VC		 240 if (wait) /* Without using a thread if waiting. */
241 journal_file_set_offline_internal(f);
242 else {
243 r = pthread_create(&f->offline_thread, NULL, journal_file_set_offline_thread, f);
ec9ffa2c
VC		 244 if (r > 0) {
245 f->offline_state = OFFLINE_JOINED;
ac2e41f5 246 return -r;
ec9ffa2c 247 }
ac2e41f5
VC		 248 }
249
250 return 0;
251}
252
253static int journal_file_set_online(JournalFile *f) {
254 bool joined = false;
255
256 assert(f);
257
258 if (!f->writable)
259 return -EPERM;
260
261 if (!(f->fd >= 0 && f->header))
262 return -EINVAL;
263
264 while (!joined) {
265 switch (f->offline_state) {
266 case OFFLINE_JOINED:
267 /* No offline thread, no need to wait. */
268 joined = true;
269 break;
270
271 case OFFLINE_SYNCING:
272 if (!__sync_bool_compare_and_swap(&f->offline_state, OFFLINE_SYNCING, OFFLINE_CANCEL))
273 continue;
274 /* Canceled syncing prior to offlining, no need to wait. */
275 break;
276
277 case OFFLINE_AGAIN_FROM_SYNCING:
278 if (!__sync_bool_compare_and_swap(&f->offline_state, OFFLINE_AGAIN_FROM_SYNCING, OFFLINE_CANCEL))
279 continue;
280 /* Canceled restart from syncing, no need to wait. */
281 break;
282
283 case OFFLINE_AGAIN_FROM_OFFLINING:
284 if (!__sync_bool_compare_and_swap(&f->offline_state, OFFLINE_AGAIN_FROM_OFFLINING, OFFLINE_CANCEL))
285 continue;
286 /* Canceled restart from offlining, must wait for offlining to complete however. */
287
288 /* fall through to wait */
289 default: {
290 int r;
291
292 r = journal_file_set_offline_thread_join(f);
293 if (r < 0)
294 return r;
295
296 joined = true;
297 break;
298 }
299 }
300 }
26687bf8 301
fa6ac760
LP		 302 if (mmap_cache_got_sigbus(f->mmap, f->fd))
303 return -EIO;
304
ac2e41f5
VC		 305 switch (f->header->state) {
306 case STATE_ONLINE:
307 return 0;
26687bf8 308
ac2e41f5
VC		 309 case STATE_OFFLINE:
310 f->header->state = STATE_ONLINE;
311 (void) fsync(f->fd);
312 return 0;
313
314 default:
315 return -EINVAL;
316 }
26687bf8
OS		 317}
318
b58c888f
VC		 319bool journal_file_is_offlining(JournalFile *f) {
320 assert(f);
321
322 __sync_synchronize();
323
324 if (f->offline_state == OFFLINE_DONE ||
325 f->offline_state == OFFLINE_JOINED)
326 return false;
327
328 return true;
329}
330
804ae586 331JournalFile* journal_file_close(JournalFile *f) {
de190aef 332 assert(f);
cec736d2 333
feb12d3e 334#ifdef HAVE_GCRYPT
b0af6f41 335 /* Write the final tag */
43cd8794
FB		 336 if (f->seal && f->writable) {
337 int r;
338
339 r = journal_file_append_tag(f);
340 if (r < 0)
341 log_error_errno(r, "Failed to append tag when closing journal: %m");
342 }
feb12d3e 343#endif
b0af6f41 344
7a24f3bf
VC		 345 if (f->post_change_timer) {
346 int enabled;
347
348 if (sd_event_source_get_enabled(f->post_change_timer, &enabled) >= 0)
349 if (enabled == SD_EVENT_ONESHOT)
350 journal_file_post_change(f);
351
e167d7fd 352 (void) sd_event_source_set_enabled(f->post_change_timer, SD_EVENT_OFF);
7a24f3bf
VC		 353 sd_event_source_unref(f->post_change_timer);
354 }
355
ac2e41f5 356 journal_file_set_offline(f, true);
cec736d2 357
fa6ac760
LP		 358 if (f->mmap && f->fd >= 0)
359 mmap_cache_close_fd(f->mmap, f->fd);
cec736d2 360
11689d2a
LP		 361 if (f->fd >= 0 && f->defrag_on_close) {
362
363 /* Be friendly to btrfs: turn COW back on again now,
364 * and defragment the file. We won't write to the file
365 * ever again, hence remove all fragmentation, and
366 * reenable all the good bits COW usually provides
367 * (such as data checksumming). */
368
1ed8f8c1 369 (void) chattr_fd(f->fd, 0, FS_NOCOW_FL);
11689d2a
LP		 370 (void) btrfs_defrag_fd(f->fd);
371 }
f27a3864 372
5d1ce257
LP		 373 if (f->close_fd)
374 safe_close(f->fd);
cec736d2 375 free(f->path);
807e17f0 376
f649045c 377 mmap_cache_unref(f->mmap);
16e9f408 378
4743015d 379 ordered_hashmap_free_free(f->chain_cache);
a4bcff5b 380
d89c8fdf 381#if defined(HAVE_XZ) || defined(HAVE_LZ4)
807e17f0
LP		 382 free(f->compress_buffer);
383#endif
384
7560fffc 385#ifdef HAVE_GCRYPT
baed47c3
LP		 386 if (f->fss_file)
387 munmap(f->fss_file, PAGE_ALIGN(f->fss_file_size));
dc4ebc07 388 else
b7c9ae91
LP		 389 free(f->fsprg_state);
390
391 free(f->fsprg_seed);
7560fffc
LP		 392
393 if (f->hmac)
394 gcry_md_close(f->hmac);
395#endif
396
cec736d2 397 free(f);
804ae586 398 return NULL;
cec736d2
LP		 399}
400
b58c888f
VC		 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
0ac38b70 410static int journal_file_init_header(JournalFile *f, JournalFile *template) {
d89c8fdf 411 Header h = {};
cec736d2
LP		 412 ssize_t k;
413 int r;
414
415 assert(f);
416
7560fffc 417 memcpy(h.signature, HEADER_SIGNATURE, 8);
23b0b2b2 418 h.header_size = htole64(ALIGN64(sizeof(h)));
cec736d2 419
d89c8fdf
ZJS		 420 h.incompatible_flags |= htole32(
421 f->compress_xz * HEADER_INCOMPATIBLE_COMPRESSED_XZ |
422 f->compress_lz4 * HEADER_INCOMPATIBLE_COMPRESSED_LZ4);
7560fffc 423
d89c8fdf
ZJS		 424 h.compatible_flags = htole32(
425 f->seal * HEADER_COMPATIBLE_SEALED);
7560fffc 426
cec736d2
LP		 427 r = sd_id128_randomize(&h.file_id);
428 if (r < 0)
429 return r;
430
0ac38b70
LP		 431 if (template) {
432 h.seqnum_id = template->header->seqnum_id;
beec0085 433 h.tail_entry_seqnum = template->header->tail_entry_seqnum;
0ac38b70
LP		 434 } else
435 h.seqnum_id = h.file_id;
cec736d2
LP		 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
a0fe2a2d
LP		 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
cec736d2 480static int journal_file_refresh_header(JournalFile *f) {
de190aef 481 sd_id128_t boot_id;
fa6ac760 482 int r;
cec736d2
LP		 483
484 assert(f);
c88cc6af 485 assert(f->header);
cec736d2
LP		 486
487 r = sd_id128_get_machine(&f->header->machine_id);
488 if (r < 0)
489 return r;
490
de190aef 491 r = sd_id128_get_boot(&boot_id);
cec736d2
LP		 492 if (r < 0)
493 return r;
494
de190aef
LP		 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
fa6ac760 500 r = journal_file_set_online(f);
b788cc23 501
7560fffc 502 /* Sync the online state to disk */
fb426037 503 (void) fsync(f->fd);
b788cc23 504
a0fe2a2d
LP		 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
fa6ac760 508 return r;
cec736d2
LP		 509}
510
511static int journal_file_verify_header(JournalFile *f) {
d89c8fdf
ZJS		 512 uint32_t flags;
513
cec736d2 514 assert(f);
c88cc6af 515 assert(f->header);
cec736d2 516
7560fffc 517 if (memcmp(f->header->signature, HEADER_SIGNATURE, 8))
cec736d2
LP		 518 return -EBADMSG;
519
7560fffc
LP		 520 /* In both read and write mode we refuse to open files with
521 * incompatible flags we don't know */
d89c8fdf
ZJS		 522 flags = le32toh(f->header->incompatible_flags);
523 if (flags & ~HEADER_INCOMPATIBLE_SUPPORTED) {
524 if (flags & ~HEADER_INCOMPATIBLE_ANY)
525 log_debug("Journal file %s has unknown incompatible flags %"PRIx32,
526 f->path, flags & ~HEADER_INCOMPATIBLE_ANY);
527 flags = (flags & HEADER_INCOMPATIBLE_ANY) & ~HEADER_INCOMPATIBLE_SUPPORTED;
528 if (flags)
529 log_debug("Journal file %s uses incompatible flags %"PRIx32
530 " disabled at compilation time.", f->path, flags);
cec736d2 531 return -EPROTONOSUPPORT;
d89c8fdf 532 }
cec736d2 533
7560fffc
LP		 534 /* When open for writing we refuse to open files with
535 * compatible flags, too */
d89c8fdf
ZJS		 536 flags = le32toh(f->header->compatible_flags);
537 if (f->writable && (flags & ~HEADER_COMPATIBLE_SUPPORTED)) {
538 if (flags & ~HEADER_COMPATIBLE_ANY)
539 log_debug("Journal file %s has unknown compatible flags %"PRIx32,
540 f->path, flags & ~HEADER_COMPATIBLE_ANY);
541 flags = (flags & HEADER_COMPATIBLE_ANY) & ~HEADER_COMPATIBLE_SUPPORTED;
542 if (flags)
543 log_debug("Journal file %s uses compatible flags %"PRIx32
544 " disabled at compilation time.", f->path, flags);
545 return -EPROTONOSUPPORT;
7560fffc
LP		 546 }
547
db11ac1a
LP		 548 if (f->header->state >= _STATE_MAX)
549 return -EBADMSG;
550
dca6219e
LP		 551 /* The first addition was n_data, so check that we are at least this large */
552 if (le64toh(f->header->header_size) < HEADER_SIZE_MIN)
23b0b2b2
LP		 553 return -EBADMSG;
554
8088cbd3 555 if (JOURNAL_HEADER_SEALED(f->header) && !JOURNAL_HEADER_CONTAINS(f->header, n_entry_arrays))
beec0085
LP		 556 return -EBADMSG;
557
db11ac1a
LP		 558 if ((le64toh(f->header->header_size) + le64toh(f->header->arena_size)) > (uint64_t) f->last_stat.st_size)
559 return -ENODATA;
560
561 if (le64toh(f->header->tail_object_offset) > (le64toh(f->header->header_size) + le64toh(f->header->arena_size)))
562 return -ENODATA;
563
7762e02b
LP		 564 if (!VALID64(le64toh(f->header->data_hash_table_offset)) ||
565 !VALID64(le64toh(f->header->field_hash_table_offset)) ||
566 !VALID64(le64toh(f->header->tail_object_offset)) ||
567 !VALID64(le64toh(f->header->entry_array_offset)))
568 return -ENODATA;
569
cec736d2 570 if (f->writable) {
ccdbaf91 571 uint8_t state;
cec736d2
LP		 572 sd_id128_t machine_id;
573 int r;
574
575 r = sd_id128_get_machine(&machine_id);
576 if (r < 0)
577 return r;
578
579 if (!sd_id128_equal(machine_id, f->header->machine_id))
580 return -EHOSTDOWN;
581
de190aef 582 state = f->header->state;
cec736d2 583
71fa6f00
LP		 584 if (state == STATE_ONLINE) {
585 log_debug("Journal file %s is already online. Assuming unclean closing.", f->path);
586 return -EBUSY;
587 } else if (state == STATE_ARCHIVED)
cec736d2 588 return -ESHUTDOWN;
71fa6f00 589 else if (state != STATE_OFFLINE) {
8facc349 590 log_debug("Journal file %s has unknown state %i.", f->path, state);
71fa6f00
LP		 591 return -EBUSY;
592 }
cec736d2
LP		 593 }
594
d89c8fdf
ZJS		 595 f->compress_xz = JOURNAL_HEADER_COMPRESSED_XZ(f->header);
596 f->compress_lz4 = JOURNAL_HEADER_COMPRESSED_LZ4(f->header);
c586dbf1 597
f1889c91 598 f->seal = JOURNAL_HEADER_SEALED(f->header);
7560fffc 599
cec736d2
LP		 600 return 0;
601}
602
2678031a
LP		 603static int journal_file_fstat(JournalFile *f) {
604 assert(f);
605 assert(f->fd >= 0);
606
607 if (fstat(f->fd, &f->last_stat) < 0)
608 return -errno;
609
610 f->last_stat_usec = now(CLOCK_MONOTONIC);
611
612 /* Refuse appending to files that are already deleted */
613 if (f->last_stat.st_nlink <= 0)
614 return -EIDRM;
615
616 return 0;
617}
618
cec736d2 619static int journal_file_allocate(JournalFile *f, uint64_t offset, uint64_t size) {
eda4b58b 620 uint64_t old_size, new_size;
fec2aa2f 621 int r;
cec736d2
LP		 622
623 assert(f);
c88cc6af 624 assert(f->header);
cec736d2 625
cec736d2 626 /* We assume that this file is not sparse, and we know that
38ac38b2 627 * for sure, since we always call posix_fallocate()
cec736d2
LP		 628 * ourselves */
629
fa6ac760
LP		 630 if (mmap_cache_got_sigbus(f->mmap, f->fd))
631 return -EIO;
632
cec736d2 633 old_size =
23b0b2b2 634 le64toh(f->header->header_size) +
cec736d2
LP		 635 le64toh(f->header->arena_size);
636
bc85bfee 637 new_size = PAGE_ALIGN(offset + size);
23b0b2b2
LP		 638 if (new_size < le64toh(f->header->header_size))
639 new_size = le64toh(f->header->header_size);
bc85bfee 640
2678031a
LP		 641 if (new_size <= old_size) {
642
643 /* We already pre-allocated enough space, but before
644 * we write to it, let's check with fstat() if the
645 * file got deleted, in order make sure we don't throw
646 * away the data immediately. Don't check fstat() for
647 * all writes though, but only once ever 10s. */
648
649 if (f->last_stat_usec + LAST_STAT_REFRESH_USEC > now(CLOCK_MONOTONIC))
650 return 0;
651
652 return journal_file_fstat(f);
653 }
654
655 /* Allocate more space. */
cec736d2 656
a676e665 657 if (f->metrics.max_size > 0 && new_size > f->metrics.max_size)
bc85bfee 658 return -E2BIG;
cec736d2 659
a676e665 660 if (new_size > f->metrics.min_size && f->metrics.keep_free > 0) {
cec736d2
LP		 661 struct statvfs svfs;
662
663 if (fstatvfs(f->fd, &svfs) >= 0) {
664 uint64_t available;
665
070052ab 666 available = LESS_BY((uint64_t) svfs.f_bfree * (uint64_t) svfs.f_bsize, f->metrics.keep_free);
cec736d2
LP		 667
668 if (new_size - old_size > available)
669 return -E2BIG;
670 }
671 }
672
eda4b58b
LP		 673 /* Increase by larger blocks at once */
674 new_size = ((new_size+FILE_SIZE_INCREASE-1) / FILE_SIZE_INCREASE) * FILE_SIZE_INCREASE;
675 if (f->metrics.max_size > 0 && new_size > f->metrics.max_size)
676 new_size = f->metrics.max_size;
677
bc85bfee
LP		 678 /* Note that the glibc fallocate() fallback is very
679 inefficient, hence we try to minimize the allocation area
680 as we can. */
fec2aa2f
GV		 681 r = posix_fallocate(f->fd, old_size, new_size - old_size);
682 if (r != 0)
683 return -r;
cec736d2 684
23b0b2b2 685 f->header->arena_size = htole64(new_size - le64toh(f->header->header_size));
cec736d2 686
2678031a 687 return journal_file_fstat(f);
cec736d2
LP		 688}
689
78519831 690static unsigned type_to_context(ObjectType type) {
d3d3208f 691 /* One context for each type, plus one catch-all for the rest */
69adae51 692 assert_cc(_OBJECT_TYPE_MAX <= MMAP_CACHE_MAX_CONTEXTS);
fa6ac760 693 assert_cc(CONTEXT_HEADER < MMAP_CACHE_MAX_CONTEXTS);
d05089d8 694 return type > OBJECT_UNUSED && type < _OBJECT_TYPE_MAX ? type : 0;
d3d3208f
MS		 695}
696
7a9dabea 697static int journal_file_move_to(JournalFile *f, ObjectType type, bool keep_always, uint64_t offset, uint64_t size, void **ret) {
2678031a
LP		 698 int r;
699
cec736d2 700 assert(f);
cec736d2
LP		 701 assert(ret);
702
7762e02b
LP		 703 if (size <= 0)
704 return -EINVAL;
705
2a59ea54 706 /* Avoid SIGBUS on invalid accesses */
4bbdcdb3
LP		 707 if (offset + size > (uint64_t) f->last_stat.st_size) {
708 /* Hmm, out of range? Let's refresh the fstat() data
709 * first, before we trust that check. */
710
2678031a
LP		 711 r = journal_file_fstat(f);
712 if (r < 0)
713 return r;
714
715 if (offset + size > (uint64_t) f->last_stat.st_size)
4bbdcdb3
LP		 716 return -EADDRNOTAVAIL;
717 }
718
7a9dabea 719 return mmap_cache_get(f->mmap, f->fd, f->prot, type_to_context(type), keep_always, offset, size, &f->last_stat, ret);
cec736d2
LP		 720}
721
16e9f408
LP		 722static uint64_t minimum_header_size(Object *o) {
723
b8e891e6 724 static const uint64_t table[] = {
16e9f408
LP		 725 [OBJECT_DATA] = sizeof(DataObject),
726 [OBJECT_FIELD] = sizeof(FieldObject),
727 [OBJECT_ENTRY] = sizeof(EntryObject),
728 [OBJECT_DATA_HASH_TABLE] = sizeof(HashTableObject),
729 [OBJECT_FIELD_HASH_TABLE] = sizeof(HashTableObject),
730 [OBJECT_ENTRY_ARRAY] = sizeof(EntryArrayObject),
731 [OBJECT_TAG] = sizeof(TagObject),
732 };
733
734 if (o->object.type >= ELEMENTSOF(table) || table[o->object.type] <= 0)
735 return sizeof(ObjectHeader);
736
737 return table[o->object.type];
738}
739
78519831 740int journal_file_move_to_object(JournalFile *f, ObjectType type, uint64_t offset, Object **ret) {
cec736d2
LP		 741 int r;
742 void *t;
743 Object *o;
744 uint64_t s;
745
746 assert(f);
747 assert(ret);
748
db11ac1a 749 /* Objects may only be located at multiple of 64 bit */
202fd896
LP		 750 if (!VALID64(offset)) {
751 log_debug("Attempt to move to object at non-64bit boundary: %" PRIu64, offset);
bd30fdf2 752 return -EBADMSG;
202fd896 753 }
db11ac1a 754
50809d7a 755 /* Object may not be located in the file header */
202fd896
LP		 756 if (offset < le64toh(f->header->header_size)) {
757 log_debug("Attempt to move to object located in file header: %" PRIu64, offset);
50809d7a 758 return -EBADMSG;
202fd896 759 }
50809d7a 760
7a9dabea 761 r = journal_file_move_to(f, type, false, offset, sizeof(ObjectHeader), &t);
cec736d2
LP		 762 if (r < 0)
763 return r;
764
765 o = (Object*) t;
766 s = le64toh(o->object.size);
767
202fd896
LP		 768 if (s < sizeof(ObjectHeader)) {
769 log_debug("Attempt to move to overly short object: %" PRIu64, offset);
cec736d2 770 return -EBADMSG;
202fd896 771 }
cec736d2 772
202fd896
LP		 773 if (o->object.type <= OBJECT_UNUSED) {
774 log_debug("Attempt to move to object with invalid type: %" PRIu64, offset);
16e9f408 775 return -EBADMSG;
202fd896 776 }
16e9f408 777
202fd896
LP		 778 if (s < minimum_header_size(o)) {
779 log_debug("Attempt to move to truncated object: %" PRIu64, offset);
16e9f408 780 return -EBADMSG;
202fd896 781 }
16e9f408 782
202fd896
LP		 783 if (type > OBJECT_UNUSED && o->object.type != type) {
784 log_debug("Attempt to move to object of unexpected type: %" PRIu64, offset);
cec736d2 785 return -EBADMSG;
202fd896 786 }
cec736d2
LP		 787
788 if (s > sizeof(ObjectHeader)) {
7a9dabea 789 r = journal_file_move_to(f, type, false, offset, s, &t);
cec736d2
LP		 790 if (r < 0)
791 return r;
792
793 o = (Object*) t;
794 }
795
cec736d2
LP		 796 *ret = o;
797 return 0;
798}
799
d98cc1f2 800static uint64_t journal_file_entry_seqnum(JournalFile *f, uint64_t *seqnum) {
cec736d2
LP		 801 uint64_t r;
802
803 assert(f);
c88cc6af 804 assert(f->header);
cec736d2 805
beec0085 806 r = le64toh(f->header->tail_entry_seqnum) + 1;
c2373f84
LP		 807
808 if (seqnum) {
de190aef 809 /* If an external seqnum counter was passed, we update
c2373f84
LP		 810 * both the local and the external one, and set it to
811 * the maximum of both */
812
813 if (*seqnum + 1 > r)
814 r = *seqnum + 1;
815
816 *seqnum = r;
817 }
818
beec0085 819 f->header->tail_entry_seqnum = htole64(r);
cec736d2 820
beec0085
LP		 821 if (f->header->head_entry_seqnum == 0)
822 f->header->head_entry_seqnum = htole64(r);
de190aef 823
cec736d2
LP		 824 return r;
825}
826
78519831 827int journal_file_append_object(JournalFile *f, ObjectType type, uint64_t size, Object **ret, uint64_t *offset) {
cec736d2
LP		 828 int r;
829 uint64_t p;
830 Object *tail, *o;
831 void *t;
832
833 assert(f);
c88cc6af 834 assert(f->header);
d05089d8 835 assert(type > OBJECT_UNUSED && type < _OBJECT_TYPE_MAX);
cec736d2
LP		 836 assert(size >= sizeof(ObjectHeader));
837 assert(offset);
838 assert(ret);
839
26687bf8
OS		 840 r = journal_file_set_online(f);
841 if (r < 0)
842 return r;
843
cec736d2 844 p = le64toh(f->header->tail_object_offset);
cec736d2 845 if (p == 0)
23b0b2b2 846 p = le64toh(f->header->header_size);
cec736d2 847 else {
d05089d8 848 r = journal_file_move_to_object(f, OBJECT_UNUSED, p, &tail);
cec736d2
LP		 849 if (r < 0)
850 return r;
851
852 p += ALIGN64(le64toh(tail->object.size));
853 }
854
855 r = journal_file_allocate(f, p, size);
856 if (r < 0)
857 return r;
858
fcde2389 859 r = journal_file_move_to(f, type, false, p, size, &t);
cec736d2
LP		 860 if (r < 0)
861 return r;
862
863 o = (Object*) t;
864
865 zero(o->object);
de190aef 866 o->object.type = type;
cec736d2
LP		 867 o->object.size = htole64(size);
868
869 f->header->tail_object_offset = htole64(p);
cec736d2
LP		 870 f->header->n_objects = htole64(le64toh(f->header->n_objects) + 1);
871
872 *ret = o;
873 *offset = p;
874
875 return 0;
876}
877
de190aef 878static int journal_file_setup_data_hash_table(JournalFile *f) {
cec736d2
LP		 879 uint64_t s, p;
880 Object *o;
881 int r;
882
883 assert(f);
c88cc6af 884 assert(f->header);
cec736d2 885
070052ab
LP		 886 /* We estimate that we need 1 hash table entry per 768 bytes
887 of journal file and we want to make sure we never get
888 beyond 75% fill level. Calculate the hash table size for
889 the maximum file size based on these metrics. */
4a92baf3 890
dfabe643 891 s = (f->metrics.max_size * 4 / 768 / 3) * sizeof(HashItem);
4a92baf3
LP		 892 if (s < DEFAULT_DATA_HASH_TABLE_SIZE)
893 s = DEFAULT_DATA_HASH_TABLE_SIZE;
894
507f22bd 895 log_debug("Reserving %"PRIu64" entries in hash table.", s / sizeof(HashItem));
4a92baf3 896
de190aef
LP		 897 r = journal_file_append_object(f,
898 OBJECT_DATA_HASH_TABLE,
899 offsetof(Object, hash_table.items) + s,
900 &o, &p);
cec736d2
LP		 901 if (r < 0)
902 return r;
903
29804cc1 904 memzero(o->hash_table.items, s);
cec736d2 905
de190aef
LP		 906 f->header->data_hash_table_offset = htole64(p + offsetof(Object, hash_table.items));
907 f->header->data_hash_table_size = htole64(s);
cec736d2
LP		 908
909 return 0;
910}
911
de190aef 912static int journal_file_setup_field_hash_table(JournalFile *f) {
cec736d2
LP		 913 uint64_t s, p;
914 Object *o;
915 int r;
916
917 assert(f);
c88cc6af 918 assert(f->header);
cec736d2 919
3c1668da
LP		 920 /* We use a fixed size hash table for the fields as this
921 * number should grow very slowly only */
922
de190aef
LP		 923 s = DEFAULT_FIELD_HASH_TABLE_SIZE;
924 r = journal_file_append_object(f,
925 OBJECT_FIELD_HASH_TABLE,
926 offsetof(Object, hash_table.items) + s,
927 &o, &p);
cec736d2
LP		 928 if (r < 0)
929 return r;
930
29804cc1 931 memzero(o->hash_table.items, s);
cec736d2 932
de190aef
LP		 933 f->header->field_hash_table_offset = htole64(p + offsetof(Object, hash_table.items));
934 f->header->field_hash_table_size = htole64(s);
cec736d2
LP		 935
936 return 0;
937}
938
dade37d4 939int journal_file_map_data_hash_table(JournalFile *f) {
cec736d2
LP		 940 uint64_t s, p;
941 void *t;
942 int r;
943
944 assert(f);
c88cc6af 945 assert(f->header);
cec736d2 946
dade37d4
LP		 947 if (f->data_hash_table)
948 return 0;
949
de190aef
LP		 950 p = le64toh(f->header->data_hash_table_offset);
951 s = le64toh(f->header->data_hash_table_size);
cec736d2 952
de190aef 953 r = journal_file_move_to(f,
16e9f408 954 OBJECT_DATA_HASH_TABLE,
fcde2389 955 true,
de190aef
LP		 956 p, s,
957 &t);
cec736d2
LP		 958 if (r < 0)
959 return r;
960
de190aef 961 f->data_hash_table = t;
cec736d2
LP		 962 return 0;
963}
964
dade37d4 965int journal_file_map_field_hash_table(JournalFile *f) {
cec736d2
LP		 966 uint64_t s, p;
967 void *t;
968 int r;
969
970 assert(f);
c88cc6af 971 assert(f->header);
cec736d2 972
dade37d4
LP		 973 if (f->field_hash_table)
974 return 0;
975
de190aef
LP		 976 p = le64toh(f->header->field_hash_table_offset);
977 s = le64toh(f->header->field_hash_table_size);
cec736d2 978
de190aef 979 r = journal_file_move_to(f,
16e9f408 980 OBJECT_FIELD_HASH_TABLE,
fcde2389 981 true,
de190aef
LP		 982 p, s,
983 &t);
cec736d2
LP		 984 if (r < 0)
985 return r;
986
de190aef 987 f->field_hash_table = t;
cec736d2
LP		 988 return 0;
989}
990
3c1668da
LP		 991static int journal_file_link_field(
992 JournalFile *f,
993 Object *o,
994 uint64_t offset,
995 uint64_t hash) {
996
805d1486 997 uint64_t p, h, m;
3c1668da
LP		 998 int r;
999
1000 assert(f);
c88cc6af 1001 assert(f->header);
90d222c1 1002 assert(f->field_hash_table);
3c1668da
LP		 1003 assert(o);
1004 assert(offset > 0);
1005
1006 if (o->object.type != OBJECT_FIELD)
1007 return -EINVAL;
1008
805d1486
LP		 1009 m = le64toh(f->header->field_hash_table_size) / sizeof(HashItem);
1010 if (m <= 0)
1011 return -EBADMSG;
3c1668da 1012
805d1486 1013 /* This might alter the window we are looking at */
3c1668da
LP		 1014 o->field.next_hash_offset = o->field.head_data_offset = 0;
1015
805d1486 1016 h = hash % m;
3c1668da
LP		 1017 p = le64toh(f->field_hash_table[h].tail_hash_offset);
1018 if (p == 0)
1019 f->field_hash_table[h].head_hash_offset = htole64(offset);
1020 else {
1021 r = journal_file_move_to_object(f, OBJECT_FIELD, p, &o);
1022 if (r < 0)
1023 return r;
1024
1025 o->field.next_hash_offset = htole64(offset);
1026 }
1027
1028 f->field_hash_table[h].tail_hash_offset = htole64(offset);
1029
1030 if (JOURNAL_HEADER_CONTAINS(f->header, n_fields))
1031 f->header->n_fields = htole64(le64toh(f->header->n_fields) + 1);
1032
1033 return 0;
1034}
1035
1036static int journal_file_link_data(
1037 JournalFile *f,
1038 Object *o,
1039 uint64_t offset,
1040 uint64_t hash) {
1041
805d1486 1042 uint64_t p, h, m;
cec736d2
LP		 1043 int r;
1044
1045 assert(f);
c88cc6af 1046 assert(f->header);
90d222c1 1047 assert(f->data_hash_table);
cec736d2
LP		 1048 assert(o);
1049 assert(offset > 0);
b588975f
LP		 1050
1051 if (o->object.type != OBJECT_DATA)
1052 return -EINVAL;
cec736d2 1053
805d1486
LP		 1054 m = le64toh(f->header->data_hash_table_size) / sizeof(HashItem);
1055 if (m <= 0)
1056 return -EBADMSG;
48496df6 1057
805d1486 1058 /* This might alter the window we are looking at */
de190aef
LP		 1059 o->data.next_hash_offset = o->data.next_field_offset = 0;
1060 o->data.entry_offset = o->data.entry_array_offset = 0;
1061 o->data.n_entries = 0;
cec736d2 1062
805d1486 1063 h = hash % m;
8db4213e 1064 p = le64toh(f->data_hash_table[h].tail_hash_offset);
3c1668da 1065 if (p == 0)
cec736d2 1066 /* Only entry in the hash table is easy */
de190aef 1067 f->data_hash_table[h].head_hash_offset = htole64(offset);
3c1668da 1068 else {
48496df6
LP		 1069 /* Move back to the previous data object, to patch in
1070 * pointer */
cec736d2 1071
de190aef 1072 r = journal_file_move_to_object(f, OBJECT_DATA, p, &o);
cec736d2
LP		 1073 if (r < 0)
1074 return r;
1075
de190aef 1076 o->data.next_hash_offset = htole64(offset);
cec736d2
LP		 1077 }
1078
de190aef 1079 f->data_hash_table[h].tail_hash_offset = htole64(offset);
cec736d2 1080
dca6219e
LP		 1081 if (JOURNAL_HEADER_CONTAINS(f->header, n_data))
1082 f->header->n_data = htole64(le64toh(f->header->n_data) + 1);
1083
cec736d2
LP		 1084 return 0;
1085}
1086
3c1668da
LP		 1087int journal_file_find_field_object_with_hash(
1088 JournalFile *f,
1089 const void *field, uint64_t size, uint64_t hash,
1090 Object **ret, uint64_t *offset) {
1091
805d1486 1092 uint64_t p, osize, h, m;
3c1668da
LP		 1093 int r;
1094
1095 assert(f);
c88cc6af 1096 assert(f->header);
3c1668da
LP		 1097 assert(field && size > 0);
1098
dade37d4
LP		 1099 /* If the field hash table is empty, we can't find anything */
1100 if (le64toh(f->header->field_hash_table_size) <= 0)
1101 return 0;
1102
1103 /* Map the field hash table, if it isn't mapped yet. */
1104 r = journal_file_map_field_hash_table(f);
1105 if (r < 0)
1106 return r;
1107
3c1668da
LP		 1108 osize = offsetof(Object, field.payload) + size;
1109
805d1486 1110 m = le64toh(f->header->field_hash_table_size) / sizeof(HashItem);
805d1486 1111 if (m <= 0)
3c1668da
LP		 1112 return -EBADMSG;
1113
805d1486 1114 h = hash % m;
3c1668da
LP		 1115 p = le64toh(f->field_hash_table[h].head_hash_offset);
1116
1117 while (p > 0) {
1118 Object *o;
1119
1120 r = journal_file_move_to_object(f, OBJECT_FIELD, p, &o);
1121 if (r < 0)
1122 return r;
1123
1124 if (le64toh(o->field.hash) == hash &&
1125 le64toh(o->object.size) == osize &&
1126 memcmp(o->field.payload, field, size) == 0) {
1127
1128 if (ret)
1129 *ret = o;
1130 if (offset)
1131 *offset = p;
1132
1133 return 1;
1134 }
1135
1136 p = le64toh(o->field.next_hash_offset);
1137 }
1138
1139 return 0;
1140}
1141
1142int journal_file_find_field_object(
1143 JournalFile *f,
1144 const void *field, uint64_t size,
1145 Object **ret, uint64_t *offset) {
1146
1147 uint64_t hash;
1148
1149 assert(f);
1150 assert(field && size > 0);
1151
1152 hash = hash64(field, size);
1153
1154 return journal_file_find_field_object_with_hash(f,
1155 field, size, hash,
1156 ret, offset);
1157}
1158
de190aef
LP		 1159int journal_file_find_data_object_with_hash(
1160 JournalFile *f,
1161 const void *data, uint64_t size, uint64_t hash,
1162 Object **ret, uint64_t *offset) {
48496df6 1163
805d1486 1164 uint64_t p, osize, h, m;
cec736d2
LP		 1165 int r;
1166
1167 assert(f);
c88cc6af 1168 assert(f->header);
cec736d2
LP		 1169 assert(data || size == 0);
1170
dade37d4
LP		 1171 /* If there's no data hash table, then there's no entry. */
1172 if (le64toh(f->header->data_hash_table_size) <= 0)
1173 return 0;
1174
1175 /* Map the data hash table, if it isn't mapped yet. */
1176 r = journal_file_map_data_hash_table(f);
1177 if (r < 0)
1178 return r;
1179
cec736d2
LP		 1180 osize = offsetof(Object, data.payload) + size;
1181
805d1486
LP		 1182 m = le64toh(f->header->data_hash_table_size) / sizeof(HashItem);
1183 if (m <= 0)
bc85bfee
LP		 1184 return -EBADMSG;
1185
805d1486 1186 h = hash % m;
de190aef 1187 p = le64toh(f->data_hash_table[h].head_hash_offset);
cec736d2 1188
de190aef
LP		 1189 while (p > 0) {
1190 Object *o;
cec736d2 1191
de190aef 1192 r = journal_file_move_to_object(f, OBJECT_DATA, p, &o);
cec736d2
LP		 1193 if (r < 0)
1194 return r;
1195
807e17f0 1196 if (le64toh(o->data.hash) != hash)
85a131e8 1197 goto next;
807e17f0 1198
d89c8fdf 1199 if (o->object.flags & OBJECT_COMPRESSION_MASK) {
3b1a55e1 1200#if defined(HAVE_XZ) || defined(HAVE_LZ4)
fa1c4b51 1201 uint64_t l;
a7f7d1bd 1202 size_t rsize = 0;
cec736d2 1203
807e17f0
LP		 1204 l = le64toh(o->object.size);
1205 if (l <= offsetof(Object, data.payload))
cec736d2
LP		 1206 return -EBADMSG;
1207
807e17f0
LP		 1208 l -= offsetof(Object, data.payload);
1209
d89c8fdf
ZJS		 1210 r = decompress_blob(o->object.flags & OBJECT_COMPRESSION_MASK,
1211 o->data.payload, l, &f->compress_buffer, &f->compress_buffer_size, &rsize, 0);
1212 if (r < 0)
1213 return r;
807e17f0 1214
b785c858 1215 if (rsize == size &&
807e17f0
LP		 1216 memcmp(f->compress_buffer, data, size) == 0) {
1217
1218 if (ret)
1219 *ret = o;
1220
1221 if (offset)
1222 *offset = p;
1223
1224 return 1;
1225 }
3b1a55e1
ZJS		 1226#else
1227 return -EPROTONOSUPPORT;
1228#endif
807e17f0
LP		 1229 } else if (le64toh(o->object.size) == osize &&
1230 memcmp(o->data.payload, data, size) == 0) {
1231
cec736d2
LP		 1232 if (ret)
1233 *ret = o;
1234
1235 if (offset)
1236 *offset = p;
1237
de190aef 1238 return 1;
cec736d2
LP		 1239 }
1240
85a131e8 1241 next:
cec736d2
LP		 1242 p = le64toh(o->data.next_hash_offset);
1243 }
1244
de190aef
LP		 1245 return 0;
1246}
1247
1248int journal_file_find_data_object(
1249 JournalFile *f,
1250 const void *data, uint64_t size,
1251 Object **ret, uint64_t *offset) {
1252
1253 uint64_t hash;
1254
1255 assert(f);
1256 assert(data || size == 0);
1257
1258 hash = hash64(data, size);
1259
1260 return journal_file_find_data_object_with_hash(f,
1261 data, size, hash,
1262 ret, offset);
1263}
1264
3c1668da
LP		 1265static int journal_file_append_field(
1266 JournalFile *f,
1267 const void *field, uint64_t size,
1268 Object **ret, uint64_t *offset) {
1269
1270 uint64_t hash, p;
1271 uint64_t osize;
1272 Object *o;
1273 int r;
1274
1275 assert(f);
1276 assert(field && size > 0);
1277
1278 hash = hash64(field, size);
1279
1280 r = journal_file_find_field_object_with_hash(f, field, size, hash, &o, &p);
1281 if (r < 0)
1282 return r;
1283 else if (r > 0) {
1284
1285 if (ret)
1286 *ret = o;
1287
1288 if (offset)
1289 *offset = p;
1290
1291 return 0;
1292 }
1293
1294 osize = offsetof(Object, field.payload) + size;
1295 r = journal_file_append_object(f, OBJECT_FIELD, osize, &o, &p);
8c92d4bb
LP		 1296 if (r < 0)
1297 return r;
3c1668da
LP		 1298
1299 o->field.hash = htole64(hash);
1300 memcpy(o->field.payload, field, size);
1301
1302 r = journal_file_link_field(f, o, p, hash);
1303 if (r < 0)
1304 return r;
1305
1306 /* The linking might have altered the window, so let's
1307 * refresh our pointer */
1308 r = journal_file_move_to_object(f, OBJECT_FIELD, p, &o);
1309 if (r < 0)
1310 return r;
1311
1312#ifdef HAVE_GCRYPT
1313 r = journal_file_hmac_put_object(f, OBJECT_FIELD, o, p);
1314 if (r < 0)
1315 return r;
1316#endif
1317
1318 if (ret)
1319 *ret = o;
1320
1321 if (offset)
1322 *offset = p;
1323
1324 return 0;
1325}
1326
48496df6
LP		 1327static int journal_file_append_data(
1328 JournalFile *f,
1329 const void *data, uint64_t size,
1330 Object **ret, uint64_t *offset) {
1331
de190aef
LP		 1332 uint64_t hash, p;
1333 uint64_t osize;
1334 Object *o;
d89c8fdf 1335 int r, compression = 0;
3c1668da 1336 const void *eq;
de190aef
LP		 1337
1338 assert(f);
1339 assert(data || size == 0);
1340
1341 hash = hash64(data, size);
1342
1343 r = journal_file_find_data_object_with_hash(f, data, size, hash, &o, &p);
1344 if (r < 0)
1345 return r;
0240c603 1346 if (r > 0) {
de190aef
LP		 1347
1348 if (ret)
1349 *ret = o;
1350
1351 if (offset)
1352 *offset = p;
1353
1354 return 0;
1355 }
1356
1357 osize = offsetof(Object, data.payload) + size;
1358 r = journal_file_append_object(f, OBJECT_DATA, osize, &o, &p);
cec736d2
LP		 1359 if (r < 0)
1360 return r;
1361
cec736d2 1362 o->data.hash = htole64(hash);
807e17f0 1363
d89c8fdf 1364#if defined(HAVE_XZ) || defined(HAVE_LZ4)
d1afbcd2 1365 if (JOURNAL_FILE_COMPRESS(f) && size >= COMPRESSION_SIZE_THRESHOLD) {
a7f7d1bd 1366 size_t rsize = 0;
807e17f0 1367
5d6f46b6 1368 compression = compress_blob(data, size, o->data.payload, size - 1, &rsize);
807e17f0 1369
d1afbcd2 1370 if (compression >= 0) {
807e17f0 1371 o->object.size = htole64(offsetof(Object, data.payload) + rsize);
d89c8fdf 1372 o->object.flags |= compression;
807e17f0 1373
fa1c4b51 1374 log_debug("Compressed data object %"PRIu64" -> %zu using %s",
d89c8fdf 1375 size, rsize, object_compressed_to_string(compression));
d1afbcd2
LP		 1376 } else
1377 /* Compression didn't work, we don't really care why, let's continue without compression */
1378 compression = 0;
807e17f0
LP		 1379 }
1380#endif
1381
75f32f04
ZJS		 1382 if (compression == 0)
1383 memcpy_safe(o->data.payload, data, size);
cec736d2 1384
de190aef 1385 r = journal_file_link_data(f, o, p, hash);
cec736d2
LP		 1386 if (r < 0)
1387 return r;
1388
33685a5a
FB		 1389#ifdef HAVE_GCRYPT
1390 r = journal_file_hmac_put_object(f, OBJECT_DATA, o, p);
1391 if (r < 0)
1392 return r;
1393#endif
1394
48496df6
LP		 1395 /* The linking might have altered the window, so let's
1396 * refresh our pointer */
1397 r = journal_file_move_to_object(f, OBJECT_DATA, p, &o);
1398 if (r < 0)
1399 return r;
1400
08c6f819
SL		 1401 if (!data)
1402 eq = NULL;
1403 else
1404 eq = memchr(data, '=', size);
3c1668da 1405 if (eq && eq > data) {
748db592 1406 Object *fo = NULL;
3c1668da 1407 uint64_t fp;
3c1668da
LP		 1408
1409 /* Create field object ... */
1410 r = journal_file_append_field(f, data, (uint8_t*) eq - (uint8_t*) data, &fo, &fp);
1411 if (r < 0)
1412 return r;
1413
1414 /* ... and link it in. */
1415 o->data.next_field_offset = fo->field.head_data_offset;
1416 fo->field.head_data_offset = le64toh(p);
1417 }
1418
cec736d2
LP		 1419 if (ret)
1420 *ret = o;
1421
1422 if (offset)
de190aef 1423 *offset = p;
cec736d2
LP		 1424
1425 return 0;
1426}
1427
1428uint64_t journal_file_entry_n_items(Object *o) {
1429 assert(o);
b588975f
LP		 1430
1431 if (o->object.type != OBJECT_ENTRY)
1432 return 0;
cec736d2
LP		 1433
1434 return (le64toh(o->object.size) - offsetof(Object, entry.items)) / sizeof(EntryItem);
1435}
1436
0284adc6 1437uint64_t journal_file_entry_array_n_items(Object *o) {
de190aef 1438 assert(o);
b588975f
LP		 1439
1440 if (o->object.type != OBJECT_ENTRY_ARRAY)
1441 return 0;
de190aef
LP		 1442
1443 return (le64toh(o->object.size) - offsetof(Object, entry_array.items)) / sizeof(uint64_t);
1444}
1445
fb9a24b6
LP		 1446uint64_t journal_file_hash_table_n_items(Object *o) {
1447 assert(o);
b588975f
LP		 1448
1449 if (o->object.type != OBJECT_DATA_HASH_TABLE &&
1450 o->object.type != OBJECT_FIELD_HASH_TABLE)
1451 return 0;
fb9a24b6
LP		 1452
1453 return (le64toh(o->object.size) - offsetof(Object, hash_table.items)) / sizeof(HashItem);
1454}
1455
de190aef 1456static int link_entry_into_array(JournalFile *f,
4fd052ae
FC		 1457 le64_t *first,
1458 le64_t *idx,
de190aef 1459 uint64_t p) {
cec736d2 1460 int r;
de190aef
LP		 1461 uint64_t n = 0, ap = 0, q, i, a, hidx;
1462 Object *o;
1463
cec736d2 1464 assert(f);
c88cc6af 1465 assert(f->header);
de190aef
LP		 1466 assert(first);
1467 assert(idx);
1468 assert(p > 0);
cec736d2 1469
de190aef
LP		 1470 a = le64toh(*first);
1471 i = hidx = le64toh(*idx);
1472 while (a > 0) {
1473
1474 r = journal_file_move_to_object(f, OBJECT_ENTRY_ARRAY, a, &o);
1475 if (r < 0)
1476 return r;
cec736d2 1477
de190aef
LP		 1478 n = journal_file_entry_array_n_items(o);
1479 if (i < n) {
1480 o->entry_array.items[i] = htole64(p);
1481 *idx = htole64(hidx + 1);
1482 return 0;
1483 }
cec736d2 1484
de190aef
LP		 1485 i -= n;
1486 ap = a;
1487 a = le64toh(o->entry_array.next_entry_array_offset);
1488 }
1489
1490 if (hidx > n)
1491 n = (hidx+1) * 2;
1492 else
1493 n = n * 2;
1494
1495 if (n < 4)
1496 n = 4;
1497
1498 r = journal_file_append_object(f, OBJECT_ENTRY_ARRAY,
1499 offsetof(Object, entry_array.items) + n * sizeof(uint64_t),
1500 &o, &q);
cec736d2
LP		 1501 if (r < 0)
1502 return r;
1503
feb12d3e 1504#ifdef HAVE_GCRYPT
5996c7c2 1505 r = journal_file_hmac_put_object(f, OBJECT_ENTRY_ARRAY, o, q);
b0af6f41
LP		 1506 if (r < 0)
1507 return r;
feb12d3e 1508#endif
b0af6f41 1509
de190aef 1510 o->entry_array.items[i] = htole64(p);
cec736d2 1511
de190aef 1512 if (ap == 0)
7be3aa17 1513 *first = htole64(q);
cec736d2 1514 else {
de190aef 1515 r = journal_file_move_to_object(f, OBJECT_ENTRY_ARRAY, ap, &o);
cec736d2
LP		 1516 if (r < 0)
1517 return r;
1518
de190aef
LP		 1519 o->entry_array.next_entry_array_offset = htole64(q);
1520 }
cec736d2 1521
2dee23eb
LP		 1522 if (JOURNAL_HEADER_CONTAINS(f->header, n_entry_arrays))
1523 f->header->n_entry_arrays = htole64(le64toh(f->header->n_entry_arrays) + 1);
1524
de190aef
LP		 1525 *idx = htole64(hidx + 1);
1526
1527 return 0;
1528}
cec736d2 1529
de190aef 1530static int link_entry_into_array_plus_one(JournalFile *f,
4fd052ae
FC		 1531 le64_t *extra,
1532 le64_t *first,
1533 le64_t *idx,
de190aef
LP		 1534 uint64_t p) {
1535
1536 int r;
1537
1538 assert(f);
1539 assert(extra);
1540 assert(first);
1541 assert(idx);
1542 assert(p > 0);
1543
1544 if (*idx == 0)
1545 *extra = htole64(p);
1546 else {
4fd052ae 1547 le64_t i;
de190aef 1548
7be3aa17 1549 i = htole64(le64toh(*idx) - 1);
de190aef
LP		 1550 r = link_entry_into_array(f, first, &i, p);
1551 if (r < 0)
1552 return r;
cec736d2
LP		 1553 }
1554
de190aef
LP		 1555 *idx = htole64(le64toh(*idx) + 1);
1556 return 0;
1557}
1558
1559static int journal_file_link_entry_item(JournalFile *f, Object *o, uint64_t offset, uint64_t i) {
1560 uint64_t p;
1561 int r;
1562 assert(f);
1563 assert(o);
1564 assert(offset > 0);
1565
1566 p = le64toh(o->entry.items[i].object_offset);
1567 if (p == 0)
1568 return -EINVAL;
1569
1570 r = journal_file_move_to_object(f, OBJECT_DATA, p, &o);
cec736d2
LP		 1571 if (r < 0)
1572 return r;
1573
de190aef
LP		 1574 return link_entry_into_array_plus_one(f,
1575 &o->data.entry_offset,
1576 &o->data.entry_array_offset,
1577 &o->data.n_entries,
1578 offset);
cec736d2
LP		 1579}
1580
1581static int journal_file_link_entry(JournalFile *f, Object *o, uint64_t offset) {
de190aef 1582 uint64_t n, i;
cec736d2
LP		 1583 int r;
1584
1585 assert(f);
c88cc6af 1586 assert(f->header);
cec736d2
LP		 1587 assert(o);
1588 assert(offset > 0);
b588975f
LP		 1589
1590 if (o->object.type != OBJECT_ENTRY)
1591 return -EINVAL;
cec736d2 1592
b788cc23
LP		 1593 __sync_synchronize();
1594
cec736d2 1595 /* Link up the entry itself */
de190aef
LP		 1596 r = link_entry_into_array(f,
1597 &f->header->entry_array_offset,
1598 &f->header->n_entries,
1599 offset);
1600 if (r < 0)
1601 return r;
cec736d2 1602
507f22bd 1603 /* log_debug("=> %s seqnr=%"PRIu64" n_entries=%"PRIu64, f->path, o->entry.seqnum, f->header->n_entries); */
cec736d2 1604
de190aef 1605 if (f->header->head_entry_realtime == 0)
0ac38b70 1606 f->header->head_entry_realtime = o->entry.realtime;
cec736d2 1607
0ac38b70 1608 f->header->tail_entry_realtime = o->entry.realtime;
de190aef
LP		 1609 f->header->tail_entry_monotonic = o->entry.monotonic;
1610
1611 f->tail_entry_monotonic_valid = true;
cec736d2
LP		 1612
1613 /* Link up the items */
1614 n = journal_file_entry_n_items(o);
1615 for (i = 0; i < n; i++) {
1616 r = journal_file_link_entry_item(f, o, offset, i);
1617 if (r < 0)
1618 return r;
1619 }
1620
cec736d2
LP		 1621 return 0;
1622}
1623
1624static int journal_file_append_entry_internal(
1625 JournalFile *f,
1626 const dual_timestamp *ts,
1627 uint64_t xor_hash,
1628 const EntryItem items[], unsigned n_items,
de190aef 1629 uint64_t *seqnum,
cec736d2
LP		 1630 Object **ret, uint64_t *offset) {
1631 uint64_t np;
1632 uint64_t osize;
1633 Object *o;
1634 int r;
1635
1636 assert(f);
c88cc6af 1637 assert(f->header);
cec736d2 1638 assert(items || n_items == 0);
de190aef 1639 assert(ts);
cec736d2
LP		 1640
1641 osize = offsetof(Object, entry.items) + (n_items * sizeof(EntryItem));
1642
de190aef 1643 r = journal_file_append_object(f, OBJECT_ENTRY, osize, &o, &np);
cec736d2
LP		 1644 if (r < 0)
1645 return r;
1646
d98cc1f2 1647 o->entry.seqnum = htole64(journal_file_entry_seqnum(f, seqnum));
75f32f04 1648 memcpy_safe(o->entry.items, items, n_items * sizeof(EntryItem));
de190aef
LP		 1649 o->entry.realtime = htole64(ts->realtime);
1650 o->entry.monotonic = htole64(ts->monotonic);
cec736d2
LP		 1651 o->entry.xor_hash = htole64(xor_hash);
1652 o->entry.boot_id = f->header->boot_id;
1653
feb12d3e 1654#ifdef HAVE_GCRYPT
5996c7c2 1655 r = journal_file_hmac_put_object(f, OBJECT_ENTRY, o, np);
b0af6f41
LP		 1656 if (r < 0)
1657 return r;
feb12d3e 1658#endif
b0af6f41 1659
cec736d2
LP		 1660 r = journal_file_link_entry(f, o, np);
1661 if (r < 0)
1662 return r;
1663
1664 if (ret)
1665 *ret = o;
1666
1667 if (offset)
1668 *offset = np;
1669
1670 return 0;
1671}
1672
cf244689 1673void journal_file_post_change(JournalFile *f) {
50f20cfd
LP		 1674 assert(f);
1675
1676 /* inotify() does not receive IN_MODIFY events from file
1677 * accesses done via mmap(). After each access we hence
1678 * trigger IN_MODIFY by truncating the journal file to its
1679 * current size which triggers IN_MODIFY. */
1680
bc85bfee
LP		 1681 __sync_synchronize();
1682
50f20cfd 1683 if (ftruncate(f->fd, f->last_stat.st_size) < 0)
e167d7fd 1684 log_debug_errno(errno, "Failed to truncate file to its own size: %m");
50f20cfd
LP		 1685}
1686
7a24f3bf
VC		 1687static int post_change_thunk(sd_event_source *timer, uint64_t usec, void *userdata) {
1688 assert(userdata);
1689
1690 journal_file_post_change(userdata);
1691
1692 return 1;
1693}
1694
1695static void schedule_post_change(JournalFile *f) {
1696 sd_event_source *timer;
1697 int enabled, r;
1698 uint64_t now;
1699
1700 assert(f);
1701 assert(f->post_change_timer);
1702
1703 timer = f->post_change_timer;
1704
1705 r = sd_event_source_get_enabled(timer, &enabled);
1706 if (r < 0) {
e167d7fd
LP		 1707 log_debug_errno(r, "Failed to get ftruncate timer state: %m");
1708 goto fail;
7a24f3bf
VC		 1709 }
1710
1711 if (enabled == SD_EVENT_ONESHOT)
1712 return;
1713
1714 r = sd_event_now(sd_event_source_get_event(timer), CLOCK_MONOTONIC, &now);
1715 if (r < 0) {
e167d7fd
LP		 1716 log_debug_errno(r, "Failed to get clock's now for scheduling ftruncate: %m");
1717 goto fail;
7a24f3bf
VC		 1718 }
1719
1720 r = sd_event_source_set_time(timer, now+f->post_change_timer_period);
1721 if (r < 0) {
e167d7fd
LP		 1722 log_debug_errno(r, "Failed to set time for scheduling ftruncate: %m");
1723 goto fail;
7a24f3bf
VC		 1724 }
1725
1726 r = sd_event_source_set_enabled(timer, SD_EVENT_ONESHOT);
1727 if (r < 0) {
e167d7fd
LP		 1728 log_debug_errno(r, "Failed to enable scheduled ftruncate: %m");
1729 goto fail;
7a24f3bf 1730 }
e167d7fd
LP		 1731
1732 return;
1733
1734fail:
1735 /* On failure, let's simply post the change immediately. */
1736 journal_file_post_change(f);
7a24f3bf
VC		 1737}
1738
1739/* Enable coalesced change posting in a timer on the provided sd_event instance */
1740int journal_file_enable_post_change_timer(JournalFile *f, sd_event *e, usec_t t) {
1741 _cleanup_(sd_event_source_unrefp) sd_event_source *timer = NULL;
1742 int r;
1743
1744 assert(f);
1745 assert_return(!f->post_change_timer, -EINVAL);
1746 assert(e);
1747 assert(t);
1748
1749 r = sd_event_add_time(e, &timer, CLOCK_MONOTONIC, 0, 0, post_change_thunk, f);
1750 if (r < 0)
1751 return r;
1752
1753 r = sd_event_source_set_enabled(timer, SD_EVENT_OFF);
1754 if (r < 0)
1755 return r;
1756
1757 f->post_change_timer = timer;
1758 timer = NULL;
1759 f->post_change_timer_period = t;
1760
1761 return r;
1762}
1763
1f2da9ec
LP		 1764static int entry_item_cmp(const void *_a, const void *_b) {
1765 const EntryItem *a = _a, *b = _b;
1766
1767 if (le64toh(a->object_offset) < le64toh(b->object_offset))
1768 return -1;
1769 if (le64toh(a->object_offset) > le64toh(b->object_offset))
1770 return 1;
1771 return 0;
1772}
1773
de190aef 1774int 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) {
cec736d2
LP		 1775 unsigned i;
1776 EntryItem *items;
1777 int r;
1778 uint64_t xor_hash = 0;
de190aef 1779 struct dual_timestamp _ts;
cec736d2
LP		 1780
1781 assert(f);
c88cc6af 1782 assert(f->header);
cec736d2
LP		 1783 assert(iovec || n_iovec == 0);
1784
de190aef
LP		 1785 if (!ts) {
1786 dual_timestamp_get(&_ts);
1787 ts = &_ts;
1788 }
1789
feb12d3e 1790#ifdef HAVE_GCRYPT
7560fffc
LP		 1791 r = journal_file_maybe_append_tag(f, ts->realtime);
1792 if (r < 0)
1793 return r;
feb12d3e 1794#endif
7560fffc 1795
64825d3c 1796 /* alloca() can't take 0, hence let's allocate at least one */
9607d947 1797 items = alloca(sizeof(EntryItem) * MAX(1u, n_iovec));
cec736d2
LP		 1798
1799 for (i = 0; i < n_iovec; i++) {
1800 uint64_t p;
1801 Object *o;
1802
1803 r = journal_file_append_data(f, iovec[i].iov_base, iovec[i].iov_len, &o, &p);
1804 if (r < 0)
cf244689 1805 return r;
cec736d2
LP		 1806
1807 xor_hash ^= le64toh(o->data.hash);
1808 items[i].object_offset = htole64(p);
de7b95cd 1809 items[i].hash = o->data.hash;
cec736d2
LP		 1810 }
1811
1f2da9ec
LP		 1812 /* Order by the position on disk, in order to improve seek
1813 * times for rotating media. */
7ff7394d 1814 qsort_safe(items, n_iovec, sizeof(EntryItem), entry_item_cmp);
1f2da9ec 1815
de190aef 1816 r = journal_file_append_entry_internal(f, ts, xor_hash, items, n_iovec, seqnum, ret, offset);
cec736d2 1817
fa6ac760
LP		 1818 /* If the memory mapping triggered a SIGBUS then we return an
1819 * IO error and ignore the error code passed down to us, since
1820 * it is very likely just an effect of a nullified replacement
1821 * mapping page */
1822
1823 if (mmap_cache_got_sigbus(f->mmap, f->fd))
1824 r = -EIO;
1825
7a24f3bf
VC		 1826 if (f->post_change_timer)
1827 schedule_post_change(f);
1828 else
1829 journal_file_post_change(f);
50f20cfd 1830
cec736d2
LP		 1831 return r;
1832}
1833
a4bcff5b 1834typedef struct ChainCacheItem {
fb099c8d 1835 uint64_t first; /* the array at the beginning of the chain */
a4bcff5b
LP		 1836 uint64_t array; /* the cached array */
1837 uint64_t begin; /* the first item in the cached array */
1838 uint64_t total; /* the total number of items in all arrays before this one in the chain */
f268980d 1839 uint64_t last_index; /* the last index we looked at, to optimize locality when bisecting */
a4bcff5b
LP		 1840} ChainCacheItem;
1841
1842static void chain_cache_put(
4743015d 1843 OrderedHashmap *h,
a4bcff5b
LP		 1844 ChainCacheItem *ci,
1845 uint64_t first,
1846 uint64_t array,
1847 uint64_t begin,
f268980d
LP		 1848 uint64_t total,
1849 uint64_t last_index) {
a4bcff5b
LP		 1850
1851 if (!ci) {
34741aa3
LP		 1852 /* If the chain item to cache for this chain is the
1853 * first one it's not worth caching anything */
1854 if (array == first)
1855 return;
1856
29433089 1857 if (ordered_hashmap_size(h) >= CHAIN_CACHE_MAX) {
4743015d 1858 ci = ordered_hashmap_steal_first(h);
29433089
LP		 1859 assert(ci);
1860 } else {
a4bcff5b
LP		 1861 ci = new(ChainCacheItem, 1);
1862 if (!ci)
1863 return;
1864 }
1865
1866 ci->first = first;
1867
4743015d 1868 if (ordered_hashmap_put(h, &ci->first, ci) < 0) {
a4bcff5b
LP		 1869 free(ci);
1870 return;
1871 }
1872 } else
1873 assert(ci->first == first);
1874
1875 ci->array = array;
1876 ci->begin = begin;
1877 ci->total = total;
f268980d 1878 ci->last_index = last_index;
a4bcff5b
LP		 1879}
1880
f268980d
LP		 1881static int generic_array_get(
1882 JournalFile *f,
1883 uint64_t first,
1884 uint64_t i,
1885 Object **ret, uint64_t *offset) {
de190aef 1886
cec736d2 1887 Object *o;
a4bcff5b 1888 uint64_t p = 0, a, t = 0;
cec736d2 1889 int r;
a4bcff5b 1890 ChainCacheItem *ci;
cec736d2
LP		 1891
1892 assert(f);
1893
de190aef 1894 a = first;
a4bcff5b
LP		 1895
1896 /* Try the chain cache first */
4743015d 1897 ci = ordered_hashmap_get(f->chain_cache, &first);
a4bcff5b
LP		 1898 if (ci && i > ci->total) {
1899 a = ci->array;
1900 i -= ci->total;
1901 t = ci->total;
1902 }
1903
de190aef 1904 while (a > 0) {
a4bcff5b 1905 uint64_t k;
cec736d2 1906
de190aef
LP		 1907 r = journal_file_move_to_object(f, OBJECT_ENTRY_ARRAY, a, &o);
1908 if (r < 0)
1909 return r;
cec736d2 1910
a4bcff5b
LP		 1911 k = journal_file_entry_array_n_items(o);
1912 if (i < k) {
de190aef 1913 p = le64toh(o->entry_array.items[i]);
a4bcff5b 1914 goto found;
cec736d2
LP		 1915 }
1916
a4bcff5b
LP		 1917 i -= k;
1918 t += k;
de190aef
LP		 1919 a = le64toh(o->entry_array.next_entry_array_offset);
1920 }
1921
a4bcff5b
LP		 1922 return 0;
1923
1924found:
1925 /* Let's cache this item for the next invocation */
af13a6b0 1926 chain_cache_put(f->chain_cache, ci, first, a, le64toh(o->entry_array.items[0]), t, i);
de190aef
LP		 1927
1928 r = journal_file_move_to_object(f, OBJECT_ENTRY, p, &o);
1929 if (r < 0)
1930 return r;
1931
1932 if (ret)
1933 *ret = o;
1934
1935 if (offset)
1936 *offset = p;
1937
1938 return 1;
1939}
1940
f268980d
LP		 1941static int generic_array_get_plus_one(
1942 JournalFile *f,
1943 uint64_t extra,
1944 uint64_t first,
1945 uint64_t i,
1946 Object **ret, uint64_t *offset) {
de190aef
LP		 1947
1948 Object *o;
1949
1950 assert(f);
1951
1952 if (i == 0) {
1953 int r;
1954
1955 r = journal_file_move_to_object(f, OBJECT_ENTRY, extra, &o);
cec736d2
LP		 1956 if (r < 0)
1957 return r;
1958
de190aef
LP		 1959 if (ret)
1960 *ret = o;
cec736d2 1961
de190aef
LP		 1962 if (offset)
1963 *offset = extra;
cec736d2 1964
de190aef 1965 return 1;
cec736d2
LP		 1966 }
1967
de190aef
LP		 1968 return generic_array_get(f, first, i-1, ret, offset);
1969}
cec736d2 1970
de190aef
LP		 1971enum {
1972 TEST_FOUND,
1973 TEST_LEFT,
1974 TEST_RIGHT
1975};
cec736d2 1976
f268980d
LP		 1977static int generic_array_bisect(
1978 JournalFile *f,
1979 uint64_t first,
1980 uint64_t n,
1981 uint64_t needle,
1982 int (*test_object)(JournalFile *f, uint64_t p, uint64_t needle),
1983 direction_t direction,
1984 Object **ret,
1985 uint64_t *offset,
1986 uint64_t *idx) {
1987
1988 uint64_t a, p, t = 0, i = 0, last_p = 0, last_index = (uint64_t) -1;
de190aef
LP		 1989 bool subtract_one = false;
1990 Object *o, *array = NULL;
1991 int r;
a4bcff5b 1992 ChainCacheItem *ci;
cec736d2 1993
de190aef
LP		 1994 assert(f);
1995 assert(test_object);
cec736d2 1996
a4bcff5b 1997 /* Start with the first array in the chain */
de190aef 1998 a = first;
a4bcff5b 1999
4743015d 2000 ci = ordered_hashmap_get(f->chain_cache, &first);
a4bcff5b
LP		 2001 if (ci && n > ci->total) {
2002 /* Ah, we have iterated this bisection array chain
2003 * previously! Let's see if we can skip ahead in the
2004 * chain, as far as the last time. But we can't jump
2005 * backwards in the chain, so let's check that
2006 * first. */
2007
2008 r = test_object(f, ci->begin, needle);
2009 if (r < 0)
2010 return r;
2011
2012 if (r == TEST_LEFT) {
f268980d 2013 /* OK, what we are looking for is right of the
a4bcff5b
LP		 2014 * begin of this EntryArray, so let's jump
2015 * straight to previously cached array in the
2016 * chain */
2017
2018 a = ci->array;
2019 n -= ci->total;
2020 t = ci->total;
f268980d 2021 last_index = ci->last_index;
a4bcff5b
LP		 2022 }
2023 }
2024
de190aef
LP		 2025 while (a > 0) {
2026 uint64_t left, right, k, lp;
2027
2028 r = journal_file_move_to_object(f, OBJECT_ENTRY_ARRAY, a, &array);
cec736d2
LP		 2029 if (r < 0)
2030 return r;
2031
de190aef
LP		 2032 k = journal_file_entry_array_n_items(array);
2033 right = MIN(k, n);
2034 if (right <= 0)
2035 return 0;
cec736d2 2036
de190aef
LP		 2037 i = right - 1;
2038 lp = p = le64toh(array->entry_array.items[i]);
2039 if (p <= 0)
bee6a291
LP		 2040 r = -EBADMSG;
2041 else
2042 r = test_object(f, p, needle);
2043 if (r == -EBADMSG) {
2044 log_debug_errno(r, "Encountered invalid entry while bisecting, cutting algorithm short. (1)");
2045 n = i;
2046 continue;
2047 }
de190aef
LP		 2048 if (r < 0)
2049 return r;
cec736d2 2050
de190aef
LP		 2051 if (r == TEST_FOUND)
2052 r = direction == DIRECTION_DOWN ? TEST_RIGHT : TEST_LEFT;
2053
2054 if (r == TEST_RIGHT) {
2055 left = 0;
2056 right -= 1;
f268980d
LP		 2057
2058 if (last_index != (uint64_t) -1) {
2059 assert(last_index <= right);
2060
2061 /* If we cached the last index we
2062 * looked at, let's try to not to jump
2063 * too wildly around and see if we can
2064 * limit the range to look at early to
2065 * the immediate neighbors of the last
2066 * index we looked at. */
2067
2068 if (last_index > 0) {
2069 uint64_t x = last_index - 1;
2070
2071 p = le64toh(array->entry_array.items[x]);
2072 if (p <= 0)
2073 return -EBADMSG;
2074
2075 r = test_object(f, p, needle);
2076 if (r < 0)
2077 return r;
2078
2079 if (r == TEST_FOUND)
2080 r = direction == DIRECTION_DOWN ? TEST_RIGHT : TEST_LEFT;
2081
2082 if (r == TEST_RIGHT)
2083 right = x;
2084 else
2085 left = x + 1;
2086 }
2087
2088 if (last_index < right) {
2089 uint64_t y = last_index + 1;
2090
2091 p = le64toh(array->entry_array.items[y]);
2092 if (p <= 0)
2093 return -EBADMSG;
2094
2095 r = test_object(f, p, needle);
2096 if (r < 0)
2097 return r;
2098
2099 if (r == TEST_FOUND)
2100 r = direction == DIRECTION_DOWN ? TEST_RIGHT : TEST_LEFT;
2101
2102 if (r == TEST_RIGHT)
2103 right = y;
2104 else
2105 left = y + 1;
2106 }
f268980d
LP		 2107 }
2108
de190aef
LP		 2109 for (;;) {
2110 if (left == right) {
2111 if (direction == DIRECTION_UP)
2112 subtract_one = true;
2113
2114 i = left;
2115 goto found;
2116 }
2117
2118 assert(left < right);
de190aef 2119 i = (left + right) / 2;
f268980d 2120
de190aef
LP		 2121 p = le64toh(array->entry_array.items[i]);
2122 if (p <= 0)
bee6a291
LP		 2123 r = -EBADMSG;
2124 else
2125 r = test_object(f, p, needle);
2126 if (r == -EBADMSG) {
2127 log_debug_errno(r, "Encountered invalid entry while bisecting, cutting algorithm short. (2)");
2128 right = n = i;
2129 continue;
2130 }
de190aef
LP		 2131 if (r < 0)
2132 return r;
cec736d2 2133
de190aef
LP		 2134 if (r == TEST_FOUND)
2135 r = direction == DIRECTION_DOWN ? TEST_RIGHT : TEST_LEFT;
2136
2137 if (r == TEST_RIGHT)
2138 right = i;
2139 else
2140 left = i + 1;
2141 }
2142 }
2143
2173cbf8 2144 if (k >= n) {
cbdca852
LP		 2145 if (direction == DIRECTION_UP) {
2146 i = n;
2147 subtract_one = true;
2148 goto found;
2149 }
2150
cec736d2 2151 return 0;
cbdca852 2152 }
cec736d2 2153
de190aef
LP		 2154 last_p = lp;
2155
2156 n -= k;
2157 t += k;
f268980d 2158 last_index = (uint64_t) -1;
de190aef 2159 a = le64toh(array->entry_array.next_entry_array_offset);
cec736d2
LP		 2160 }
2161
2162 return 0;
de190aef
LP		 2163
2164found:
2165 if (subtract_one && t == 0 && i == 0)
2166 return 0;
2167
a4bcff5b 2168 /* Let's cache this item for the next invocation */
af13a6b0 2169 chain_cache_put(f->chain_cache, ci, first, a, le64toh(array->entry_array.items[0]), t, subtract_one ? (i > 0 ? i-1 : (uint64_t) -1) : i);
a4bcff5b 2170
de190aef
LP		 2171 if (subtract_one && i == 0)
2172 p = last_p;
2173 else if (subtract_one)
2174 p = le64toh(array->entry_array.items[i-1]);
2175 else
2176 p = le64toh(array->entry_array.items[i]);
2177
2178 r = journal_file_move_to_object(f, OBJECT_ENTRY, p, &o);
2179 if (r < 0)
2180 return r;
2181
2182 if (ret)
2183 *ret = o;
2184
2185 if (offset)
2186 *offset = p;
2187
2188 if (idx)
cbdca852 2189 *idx = t + i + (subtract_one ? -1 : 0);
de190aef
LP		 2190
2191 return 1;
cec736d2
LP		 2192}
2193
f268980d
LP		 2194static int generic_array_bisect_plus_one(
2195 JournalFile *f,
2196 uint64_t extra,
2197 uint64_t first,
2198 uint64_t n,
2199 uint64_t needle,
2200 int (*test_object)(JournalFile *f, uint64_t p, uint64_t needle),
2201 direction_t direction,
2202 Object **ret,
2203 uint64_t *offset,
2204 uint64_t *idx) {
de190aef 2205
cec736d2 2206 int r;
cbdca852
LP		 2207 bool step_back = false;
2208 Object *o;
cec736d2
LP		 2209
2210 assert(f);
de190aef 2211 assert(test_object);
cec736d2 2212
de190aef
LP		 2213 if (n <= 0)
2214 return 0;
cec736d2 2215
de190aef
LP		 2216 /* This bisects the array in object 'first', but first checks
2217 * an extra */
de190aef
LP		 2218 r = test_object(f, extra, needle);
2219 if (r < 0)
2220 return r;
a536e261
LP		 2221
2222 if (r == TEST_FOUND)
2223 r = direction == DIRECTION_DOWN ? TEST_RIGHT : TEST_LEFT;
2224
cbdca852
LP		 2225 /* if we are looking with DIRECTION_UP then we need to first
2226 see if in the actual array there is a matching entry, and
2227 return the last one of that. But if there isn't any we need
2228 to return this one. Hence remember this, and return it
2229 below. */
2230 if (r == TEST_LEFT)
2231 step_back = direction == DIRECTION_UP;
de190aef 2232
cbdca852
LP		 2233 if (r == TEST_RIGHT) {
2234 if (direction == DIRECTION_DOWN)
2235 goto found;
2236 else
2237 return 0;
a536e261 2238 }
cec736d2 2239
de190aef
LP		 2240 r = generic_array_bisect(f, first, n-1, needle, test_object, direction, ret, offset, idx);
2241
cbdca852
LP		 2242 if (r == 0 && step_back)
2243 goto found;
2244
ecf68b1d 2245 if (r > 0 && idx)
313cefa1 2246 (*idx)++;
de190aef
LP		 2247
2248 return r;
cbdca852
LP		 2249
2250found:
2251 r = journal_file_move_to_object(f, OBJECT_ENTRY, extra, &o);
2252 if (r < 0)
2253 return r;
2254
2255 if (ret)
2256 *ret = o;
2257
2258 if (offset)
2259 *offset = extra;
2260
2261 if (idx)
2262 *idx = 0;
2263
2264 return 1;
2265}
2266
44a6b1b6 2267_pure_ static int test_object_offset(JournalFile *f, uint64_t p, uint64_t needle) {
cbdca852
LP		 2268 assert(f);
2269 assert(p > 0);
2270
2271 if (p == needle)
2272 return TEST_FOUND;
2273 else if (p < needle)
2274 return TEST_LEFT;
2275 else
2276 return TEST_RIGHT;
2277}
2278
de190aef
LP		 2279static int test_object_seqnum(JournalFile *f, uint64_t p, uint64_t needle) {
2280 Object *o;
2281 int r;
2282
2283 assert(f);
2284 assert(p > 0);
2285
2286 r = journal_file_move_to_object(f, OBJECT_ENTRY, p, &o);
cec736d2
LP		 2287 if (r < 0)
2288 return r;
2289
de190aef
LP		 2290 if (le64toh(o->entry.seqnum) == needle)
2291 return TEST_FOUND;
2292 else if (le64toh(o->entry.seqnum) < needle)
2293 return TEST_LEFT;
2294 else
2295 return TEST_RIGHT;
2296}
cec736d2 2297
de190aef
LP		 2298int journal_file_move_to_entry_by_seqnum(
2299 JournalFile *f,
2300 uint64_t seqnum,
2301 direction_t direction,
2302 Object **ret,
2303 uint64_t *offset) {
c88cc6af
VC		 2304 assert(f);
2305 assert(f->header);
de190aef
LP		 2306
2307 return generic_array_bisect(f,
2308 le64toh(f->header->entry_array_offset),
2309 le64toh(f->header->n_entries),
2310 seqnum,
2311 test_object_seqnum,
2312 direction,
2313 ret, offset, NULL);
2314}
cec736d2 2315
de190aef
LP		 2316static int test_object_realtime(JournalFile *f, uint64_t p, uint64_t needle) {
2317 Object *o;
2318 int r;
2319
2320 assert(f);
2321 assert(p > 0);
2322
2323 r = journal_file_move_to_object(f, OBJECT_ENTRY, p, &o);
2324 if (r < 0)
2325 return r;
2326
2327 if (le64toh(o->entry.realtime) == needle)
2328 return TEST_FOUND;
2329 else if (le64toh(o->entry.realtime) < needle)
2330 return TEST_LEFT;
2331 else
2332 return TEST_RIGHT;
cec736d2
LP		 2333}
2334
de190aef
LP		 2335int journal_file_move_to_entry_by_realtime(
2336 JournalFile *f,
2337 uint64_t realtime,
2338 direction_t direction,
2339 Object **ret,
2340 uint64_t *offset) {
c88cc6af
VC		 2341 assert(f);
2342 assert(f->header);
de190aef
LP		 2343
2344 return generic_array_bisect(f,
2345 le64toh(f->header->entry_array_offset),
2346 le64toh(f->header->n_entries),
2347 realtime,
2348 test_object_realtime,
2349 direction,
2350 ret, offset, NULL);
2351}
2352
2353static int test_object_monotonic(JournalFile *f, uint64_t p, uint64_t needle) {
2354 Object *o;
2355 int r;
2356
2357 assert(f);
2358 assert(p > 0);
2359
2360 r = journal_file_move_to_object(f, OBJECT_ENTRY, p, &o);
2361 if (r < 0)
2362 return r;
2363
2364 if (le64toh(o->entry.monotonic) == needle)
2365 return TEST_FOUND;
2366 else if (le64toh(o->entry.monotonic) < needle)
2367 return TEST_LEFT;
2368 else
2369 return TEST_RIGHT;
2370}
2371
2a560338 2372static int find_data_object_by_boot_id(
47838ab3
ZJS		 2373 JournalFile *f,
2374 sd_id128_t boot_id,
2375 Object **o,
2376 uint64_t *b) {
2a560338 2377
47838ab3
ZJS		 2378 char t[sizeof("_BOOT_ID=")-1 + 32 + 1] = "_BOOT_ID=";
2379
2380 sd_id128_to_string(boot_id, t + 9);
2381 return journal_file_find_data_object(f, t, sizeof(t) - 1, o, b);
2382}
2383
de190aef
LP		 2384int journal_file_move_to_entry_by_monotonic(
2385 JournalFile *f,
2386 sd_id128_t boot_id,
2387 uint64_t monotonic,
2388 direction_t direction,
2389 Object **ret,
2390 uint64_t *offset) {
2391
de190aef
LP		 2392 Object *o;
2393 int r;
2394
cbdca852 2395 assert(f);
de190aef 2396
47838ab3 2397 r = find_data_object_by_boot_id(f, boot_id, &o, NULL);
de190aef
LP		 2398 if (r < 0)
2399 return r;
cbdca852 2400 if (r == 0)
de190aef
LP		 2401 return -ENOENT;
2402
2403 return generic_array_bisect_plus_one(f,
2404 le64toh(o->data.entry_offset),
2405 le64toh(o->data.entry_array_offset),
2406 le64toh(o->data.n_entries),
2407 monotonic,
2408 test_object_monotonic,
2409 direction,
2410 ret, offset, NULL);
2411}
2412
1fc605b0 2413void journal_file_reset_location(JournalFile *f) {
6573ef05 2414 f->location_type = LOCATION_HEAD;
1fc605b0 2415 f->current_offset = 0;
6573ef05
MS		 2416 f->current_seqnum = 0;
2417 f->current_realtime = 0;
2418 f->current_monotonic = 0;
2419 zero(f->current_boot_id);
2420 f->current_xor_hash = 0;
2421}
2422
950c07d4 2423void journal_file_save_location(JournalFile *f, Object *o, uint64_t offset) {
6573ef05
MS		 2424 f->location_type = LOCATION_SEEK;
2425 f->current_offset = offset;
2426 f->current_seqnum = le64toh(o->entry.seqnum);
2427 f->current_realtime = le64toh(o->entry.realtime);
2428 f->current_monotonic = le64toh(o->entry.monotonic);
2429 f->current_boot_id = o->entry.boot_id;
2430 f->current_xor_hash = le64toh(o->entry.xor_hash);
1fc605b0
MS		 2431}
2432
d8ae66d7
MS		 2433int journal_file_compare_locations(JournalFile *af, JournalFile *bf) {
2434 assert(af);
c88cc6af 2435 assert(af->header);
d8ae66d7 2436 assert(bf);
c88cc6af 2437 assert(bf->header);
d8ae66d7
MS		 2438 assert(af->location_type == LOCATION_SEEK);
2439 assert(bf->location_type == LOCATION_SEEK);
2440
2441 /* If contents and timestamps match, these entries are
2442 * identical, even if the seqnum does not match */
2443 if (sd_id128_equal(af->current_boot_id, bf->current_boot_id) &&
2444 af->current_monotonic == bf->current_monotonic &&
2445 af->current_realtime == bf->current_realtime &&
2446 af->current_xor_hash == bf->current_xor_hash)
2447 return 0;
2448
2449 if (sd_id128_equal(af->header->seqnum_id, bf->header->seqnum_id)) {
2450
2451 /* If this is from the same seqnum source, compare
2452 * seqnums */
2453 if (af->current_seqnum < bf->current_seqnum)
2454 return -1;
2455 if (af->current_seqnum > bf->current_seqnum)
2456 return 1;
2457
2458 /* Wow! This is weird, different data but the same
2459 * seqnums? Something is borked, but let's make the
2460 * best of it and compare by time. */
2461 }
2462
2463 if (sd_id128_equal(af->current_boot_id, bf->current_boot_id)) {
2464
2465 /* If the boot id matches, compare monotonic time */
2466 if (af->current_monotonic < bf->current_monotonic)
2467 return -1;
2468 if (af->current_monotonic > bf->current_monotonic)
2469 return 1;
2470 }
2471
2472 /* Otherwise, compare UTC time */
2473 if (af->current_realtime < bf->current_realtime)
2474 return -1;
2475 if (af->current_realtime > bf->current_realtime)
2476 return 1;
2477
2478 /* Finally, compare by contents */
2479 if (af->current_xor_hash < bf->current_xor_hash)
2480 return -1;
2481 if (af->current_xor_hash > bf->current_xor_hash)
2482 return 1;
2483
2484 return 0;
2485}
2486
de190aef
LP		 2487int journal_file_next_entry(
2488 JournalFile *f,
f534928a 2489 uint64_t p,
de190aef
LP		 2490 direction_t direction,
2491 Object **ret, uint64_t *offset) {
2492
fb099c8d 2493 uint64_t i, n, ofs;
cec736d2
LP		 2494 int r;
2495
2496 assert(f);
c88cc6af 2497 assert(f->header);
de190aef
LP		 2498
2499 n = le64toh(f->header->n_entries);
2500 if (n <= 0)
2501 return 0;
cec736d2 2502
f534928a 2503 if (p == 0)
de190aef 2504 i = direction == DIRECTION_DOWN ? 0 : n - 1;
cec736d2 2505 else {
de190aef
LP		 2506 r = generic_array_bisect(f,
2507 le64toh(f->header->entry_array_offset),
2508 le64toh(f->header->n_entries),
2509 p,
2510 test_object_offset,
2511 DIRECTION_DOWN,
2512 NULL, NULL,
2513 &i);
2514 if (r <= 0)
2515 return r;
2516
2517 if (direction == DIRECTION_DOWN) {
2518 if (i >= n - 1)
2519 return 0;
2520
2521 i++;
2522 } else {
2523 if (i <= 0)
2524 return 0;
2525
2526 i--;
2527 }
cec736d2
LP		 2528 }
2529
de190aef 2530 /* And jump to it */
fb099c8d
ZJS		 2531 r = generic_array_get(f,
2532 le64toh(f->header->entry_array_offset),
2533 i,
2534 ret, &ofs);
caeab8f6
LP		 2535 if (r == -EBADMSG && direction == DIRECTION_DOWN) {
2536 /* Special case: when we iterate throught the journal file linearly, and hit an entry we can't read,
2537 * consider this the end of the journal file. */
2538 log_debug_errno(r, "Encountered entry we can't read while iterating through journal file. Considering this the end of the file.");
2539 return 0;
2540 }
fb099c8d
ZJS		 2541 if (r <= 0)
2542 return r;
2543
2544 if (p > 0 &&
2545 (direction == DIRECTION_DOWN ? ofs <= p : ofs >= p)) {
caeab8f6 2546 log_debug("%s: entry array corrupted at entry %" PRIu64, f->path, i);
fb099c8d
ZJS		 2547 return -EBADMSG;
2548 }
2549
2550 if (offset)
2551 *offset = ofs;
2552
2553 return 1;
de190aef 2554}
cec736d2 2555
de190aef
LP		 2556int journal_file_next_entry_for_data(
2557 JournalFile *f,
2558 Object *o, uint64_t p,
2559 uint64_t data_offset,
2560 direction_t direction,
2561 Object **ret, uint64_t *offset) {
2562
2563 uint64_t n, i;
cec736d2 2564 int r;
de190aef 2565 Object *d;
cec736d2
LP		 2566
2567 assert(f);
de190aef 2568 assert(p > 0 || !o);
cec736d2 2569
de190aef 2570 r = journal_file_move_to_object(f, OBJECT_DATA, data_offset, &d);
466ccd92 2571 if (r < 0)
de190aef 2572 return r;
cec736d2 2573
de190aef
LP		 2574 n = le64toh(d->data.n_entries);
2575 if (n <= 0)
2576 return n;
cec736d2 2577
de190aef
LP		 2578 if (!o)
2579 i = direction == DIRECTION_DOWN ? 0 : n - 1;
2580 else {
2581 if (o->object.type != OBJECT_ENTRY)
2582 return -EINVAL;
cec736d2 2583
de190aef
LP		 2584 r = generic_array_bisect_plus_one(f,
2585 le64toh(d->data.entry_offset),
2586 le64toh(d->data.entry_array_offset),
2587 le64toh(d->data.n_entries),
2588 p,
2589 test_object_offset,
2590 DIRECTION_DOWN,
2591 NULL, NULL,
2592 &i);
2593
2594 if (r <= 0)
cec736d2
LP		 2595 return r;
2596
de190aef
LP		 2597 if (direction == DIRECTION_DOWN) {
2598 if (i >= n - 1)
2599 return 0;
cec736d2 2600
de190aef
LP		 2601 i++;
2602 } else {
2603 if (i <= 0)
2604 return 0;
cec736d2 2605
de190aef
LP		 2606 i--;
2607 }
cec736d2 2608
de190aef 2609 }
cec736d2 2610
de190aef
LP		 2611 return generic_array_get_plus_one(f,
2612 le64toh(d->data.entry_offset),
2613 le64toh(d->data.entry_array_offset),
2614 i,
2615 ret, offset);
2616}
cec736d2 2617
cbdca852
LP		 2618int journal_file_move_to_entry_by_offset_for_data(
2619 JournalFile *f,
2620 uint64_t data_offset,
2621 uint64_t p,
2622 direction_t direction,
2623 Object **ret, uint64_t *offset) {
2624
2625 int r;
2626 Object *d;
2627
2628 assert(f);
2629
2630 r = journal_file_move_to_object(f, OBJECT_DATA, data_offset, &d);
2631 if (r < 0)
2632 return r;
2633
2634 return generic_array_bisect_plus_one(f,
2635 le64toh(d->data.entry_offset),
2636 le64toh(d->data.entry_array_offset),
2637 le64toh(d->data.n_entries),
2638 p,
2639 test_object_offset,
2640 direction,
2641 ret, offset, NULL);
2642}
2643
2644int journal_file_move_to_entry_by_monotonic_for_data(
2645 JournalFile *f,
2646 uint64_t data_offset,
2647 sd_id128_t boot_id,
2648 uint64_t monotonic,
2649 direction_t direction,
2650 Object **ret, uint64_t *offset) {
2651
cbdca852
LP		 2652 Object *o, *d;
2653 int r;
2654 uint64_t b, z;
2655
2656 assert(f);
2657
2658 /* First, seek by time */
47838ab3 2659 r = find_data_object_by_boot_id(f, boot_id, &o, &b);
cbdca852
LP		 2660 if (r < 0)
2661 return r;
2662 if (r == 0)
2663 return -ENOENT;
2664
2665 r = generic_array_bisect_plus_one(f,
2666 le64toh(o->data.entry_offset),
2667 le64toh(o->data.entry_array_offset),
2668 le64toh(o->data.n_entries),
2669 monotonic,
2670 test_object_monotonic,
2671 direction,
2672 NULL, &z, NULL);
2673 if (r <= 0)
2674 return r;
2675
2676 /* And now, continue seeking until we find an entry that
2677 * exists in both bisection arrays */
2678
2679 for (;;) {
2680 Object *qo;
2681 uint64_t p, q;
2682
2683 r = journal_file_move_to_object(f, OBJECT_DATA, data_offset, &d);
2684 if (r < 0)
2685 return r;
2686
2687 r = generic_array_bisect_plus_one(f,
2688 le64toh(d->data.entry_offset),
2689 le64toh(d->data.entry_array_offset),
2690 le64toh(d->data.n_entries),
2691 z,
2692 test_object_offset,
2693 direction,
2694 NULL, &p, NULL);
2695 if (r <= 0)
2696 return r;
2697
2698 r = journal_file_move_to_object(f, OBJECT_DATA, b, &o);
2699 if (r < 0)
2700 return r;
2701
2702 r = generic_array_bisect_plus_one(f,
2703 le64toh(o->data.entry_offset),
2704 le64toh(o->data.entry_array_offset),
2705 le64toh(o->data.n_entries),
2706 p,
2707 test_object_offset,
2708 direction,
2709 &qo, &q, NULL);
2710
2711 if (r <= 0)
2712 return r;
2713
2714 if (p == q) {
2715 if (ret)
2716 *ret = qo;
2717 if (offset)
2718 *offset = q;
2719
2720 return 1;
2721 }
2722
2723 z = q;
2724 }
cbdca852
LP		 2725}
2726
de190aef
LP		 2727int journal_file_move_to_entry_by_seqnum_for_data(
2728 JournalFile *f,
2729 uint64_t data_offset,
2730 uint64_t seqnum,
2731 direction_t direction,
2732 Object **ret, uint64_t *offset) {
cec736d2 2733
de190aef
LP		 2734 Object *d;
2735 int r;
cec736d2 2736
91a31dde
LP		 2737 assert(f);
2738
de190aef 2739 r = journal_file_move_to_object(f, OBJECT_DATA, data_offset, &d);
91a31dde 2740 if (r < 0)
de190aef 2741 return r;
cec736d2 2742
de190aef
LP		 2743 return generic_array_bisect_plus_one(f,
2744 le64toh(d->data.entry_offset),
2745 le64toh(d->data.entry_array_offset),
2746 le64toh(d->data.n_entries),
2747 seqnum,
2748 test_object_seqnum,
2749 direction,
2750 ret, offset, NULL);
2751}
cec736d2 2752
de190aef
LP		 2753int journal_file_move_to_entry_by_realtime_for_data(
2754 JournalFile *f,
2755 uint64_t data_offset,
2756 uint64_t realtime,
2757 direction_t direction,
2758 Object **ret, uint64_t *offset) {
2759
2760 Object *d;
2761 int r;
2762
91a31dde
LP		 2763 assert(f);
2764
de190aef 2765 r = journal_file_move_to_object(f, OBJECT_DATA, data_offset, &d);
91a31dde 2766 if (r < 0)
de190aef
LP		 2767 return r;
2768
2769 return generic_array_bisect_plus_one(f,
2770 le64toh(d->data.entry_offset),
2771 le64toh(d->data.entry_array_offset),
2772 le64toh(d->data.n_entries),
2773 realtime,
2774 test_object_realtime,
2775 direction,
2776 ret, offset, NULL);
cec736d2
LP		 2777}
2778
0284adc6 2779void journal_file_dump(JournalFile *f) {
7560fffc 2780 Object *o;
7560fffc 2781 int r;
0284adc6 2782 uint64_t p;
7560fffc
LP		 2783
2784 assert(f);
c88cc6af 2785 assert(f->header);
7560fffc 2786
0284adc6 2787 journal_file_print_header(f);
7560fffc 2788
0284adc6
LP		 2789 p = le64toh(f->header->header_size);
2790 while (p != 0) {
d05089d8 2791 r = journal_file_move_to_object(f, OBJECT_UNUSED, p, &o);
0284adc6
LP		 2792 if (r < 0)
2793 goto fail;
7560fffc 2794
0284adc6 2795 switch (o->object.type) {
d98cc1f2 2796
0284adc6
LP		 2797 case OBJECT_UNUSED:
2798 printf("Type: OBJECT_UNUSED\n");
2799 break;
d98cc1f2 2800
0284adc6
LP		 2801 case OBJECT_DATA:
2802 printf("Type: OBJECT_DATA\n");
2803 break;
7560fffc 2804
3c1668da
LP		 2805 case OBJECT_FIELD:
2806 printf("Type: OBJECT_FIELD\n");
2807 break;
2808
0284adc6 2809 case OBJECT_ENTRY:
507f22bd
ZJS		 2810 printf("Type: OBJECT_ENTRY seqnum=%"PRIu64" monotonic=%"PRIu64" realtime=%"PRIu64"\n",
2811 le64toh(o->entry.seqnum),
2812 le64toh(o->entry.monotonic),
2813 le64toh(o->entry.realtime));
0284adc6 2814 break;
7560fffc 2815
0284adc6
LP		 2816 case OBJECT_FIELD_HASH_TABLE:
2817 printf("Type: OBJECT_FIELD_HASH_TABLE\n");
2818 break;
7560fffc 2819
0284adc6
LP		 2820 case OBJECT_DATA_HASH_TABLE:
2821 printf("Type: OBJECT_DATA_HASH_TABLE\n");
2822 break;
7560fffc 2823
0284adc6
LP		 2824 case OBJECT_ENTRY_ARRAY:
2825 printf("Type: OBJECT_ENTRY_ARRAY\n");
2826 break;
7560fffc 2827
0284adc6 2828 case OBJECT_TAG:
507f22bd
ZJS		 2829 printf("Type: OBJECT_TAG seqnum=%"PRIu64" epoch=%"PRIu64"\n",
2830 le64toh(o->tag.seqnum),
2831 le64toh(o->tag.epoch));
0284adc6 2832 break;
3c1668da
LP		 2833
2834 default:
8facc349 2835 printf("Type: unknown (%i)\n", o->object.type);
3c1668da 2836 break;
0284adc6 2837 }
7560fffc 2838
d89c8fdf
ZJS		 2839 if (o->object.flags & OBJECT_COMPRESSION_MASK)
2840 printf("Flags: %s\n",
2841 object_compressed_to_string(o->object.flags & OBJECT_COMPRESSION_MASK));
7560fffc 2842
0284adc6
LP		 2843 if (p == le64toh(f->header->tail_object_offset))
2844 p = 0;
2845 else
2846 p = p + ALIGN64(le64toh(o->object.size));
2847 }
7560fffc 2848
0284adc6
LP		 2849 return;
2850fail:
2851 log_error("File corrupt");
7560fffc
LP		 2852}
2853
718fe4b1
ZJS		 2854static const char* format_timestamp_safe(char *buf, size_t l, usec_t t) {
2855 const char *x;
2856
2857 x = format_timestamp(buf, l, t);
2858 if (x)
2859 return x;
2860 return " --- ";
2861}
2862
0284adc6 2863void journal_file_print_header(JournalFile *f) {
2765b7bb 2864 char a[33], b[33], c[33], d[33];
ed375beb 2865 char x[FORMAT_TIMESTAMP_MAX], y[FORMAT_TIMESTAMP_MAX], z[FORMAT_TIMESTAMP_MAX];
a1a03e30
LP		 2866 struct stat st;
2867 char bytes[FORMAT_BYTES_MAX];
7560fffc
LP		 2868
2869 assert(f);
c88cc6af 2870 assert(f->header);
7560fffc 2871
0284adc6
LP		 2872 printf("File Path: %s\n"
2873 "File ID: %s\n"
2874 "Machine ID: %s\n"
2875 "Boot ID: %s\n"
2876 "Sequential Number ID: %s\n"
2877 "State: %s\n"
2878 "Compatible Flags:%s%s\n"
d89c8fdf 2879 "Incompatible Flags:%s%s%s\n"
507f22bd
ZJS		 2880 "Header size: %"PRIu64"\n"
2881 "Arena size: %"PRIu64"\n"
2882 "Data Hash Table Size: %"PRIu64"\n"
2883 "Field Hash Table Size: %"PRIu64"\n"
0284adc6 2884 "Rotate Suggested: %s\n"
0808b92f
LP		 2885 "Head Sequential Number: %"PRIu64" (%"PRIx64")\n"
2886 "Tail Sequential Number: %"PRIu64" (%"PRIx64")\n"
2887 "Head Realtime Timestamp: %s (%"PRIx64")\n"
2888 "Tail Realtime Timestamp: %s (%"PRIx64")\n"
2889 "Tail Monotonic Timestamp: %s (%"PRIx64")\n"
507f22bd
ZJS		 2890 "Objects: %"PRIu64"\n"
2891 "Entry Objects: %"PRIu64"\n",
0284adc6
LP		 2892 f->path,
2893 sd_id128_to_string(f->header->file_id, a),
2894 sd_id128_to_string(f->header->machine_id, b),
2895 sd_id128_to_string(f->header->boot_id, c),
2765b7bb 2896 sd_id128_to_string(f->header->seqnum_id, d),
3223f44f
LP		 2897 f->header->state == STATE_OFFLINE ? "OFFLINE" :
2898 f->header->state == STATE_ONLINE ? "ONLINE" :
2899 f->header->state == STATE_ARCHIVED ? "ARCHIVED" : "UNKNOWN",
8088cbd3 2900 JOURNAL_HEADER_SEALED(f->header) ? " SEALED" : "",
d89c8fdf
ZJS		 2901 (le32toh(f->header->compatible_flags) & ~HEADER_COMPATIBLE_ANY) ? " ???" : "",
2902 JOURNAL_HEADER_COMPRESSED_XZ(f->header) ? " COMPRESSED-XZ" : "",
2903 JOURNAL_HEADER_COMPRESSED_LZ4(f->header) ? " COMPRESSED-LZ4" : "",
2904 (le32toh(f->header->incompatible_flags) & ~HEADER_INCOMPATIBLE_ANY) ? " ???" : "",
507f22bd
ZJS		 2905 le64toh(f->header->header_size),
2906 le64toh(f->header->arena_size),
2907 le64toh(f->header->data_hash_table_size) / sizeof(HashItem),
2908 le64toh(f->header->field_hash_table_size) / sizeof(HashItem),
fb0951b0 2909 yes_no(journal_file_rotate_suggested(f, 0)),
0808b92f
LP		 2910 le64toh(f->header->head_entry_seqnum), le64toh(f->header->head_entry_seqnum),
2911 le64toh(f->header->tail_entry_seqnum), le64toh(f->header->tail_entry_seqnum),
2912 format_timestamp_safe(x, sizeof(x), le64toh(f->header->head_entry_realtime)), le64toh(f->header->head_entry_realtime),
2913 format_timestamp_safe(y, sizeof(y), le64toh(f->header->tail_entry_realtime)), le64toh(f->header->tail_entry_realtime),
2914 format_timespan(z, sizeof(z), le64toh(f->header->tail_entry_monotonic), USEC_PER_MSEC), le64toh(f->header->tail_entry_monotonic),
507f22bd
ZJS		 2915 le64toh(f->header->n_objects),
2916 le64toh(f->header->n_entries));
7560fffc 2917
0284adc6 2918 if (JOURNAL_HEADER_CONTAINS(f->header, n_data))
507f22bd 2919 printf("Data Objects: %"PRIu64"\n"
0284adc6 2920 "Data Hash Table Fill: %.1f%%\n",
507f22bd 2921 le64toh(f->header->n_data),
0284adc6 2922 100.0 * (double) le64toh(f->header->n_data) / ((double) (le64toh(f->header->data_hash_table_size) / sizeof(HashItem))));
7560fffc 2923
0284adc6 2924 if (JOURNAL_HEADER_CONTAINS(f->header, n_fields))
507f22bd 2925 printf("Field Objects: %"PRIu64"\n"
0284adc6 2926 "Field Hash Table Fill: %.1f%%\n",
507f22bd 2927 le64toh(f->header->n_fields),
0284adc6 2928 100.0 * (double) le64toh(f->header->n_fields) / ((double) (le64toh(f->header->field_hash_table_size) / sizeof(HashItem))));
3223f44f
LP		 2929
2930 if (JOURNAL_HEADER_CONTAINS(f->header, n_tags))
507f22bd
ZJS		 2931 printf("Tag Objects: %"PRIu64"\n",
2932 le64toh(f->header->n_tags));
3223f44f 2933 if (JOURNAL_HEADER_CONTAINS(f->header, n_entry_arrays))
507f22bd
ZJS		 2934 printf("Entry Array Objects: %"PRIu64"\n",
2935 le64toh(f->header->n_entry_arrays));
a1a03e30
LP		 2936
2937 if (fstat(f->fd, &st) >= 0)
59f448cf 2938 printf("Disk usage: %s\n", format_bytes(bytes, sizeof(bytes), (uint64_t) st.st_blocks * 512ULL));
7560fffc
LP		 2939}
2940
fc68c929
LP		 2941static int journal_file_warn_btrfs(JournalFile *f) {
2942 unsigned attrs;
2943 int r;
2944
2945 assert(f);
2946
2947 /* Before we write anything, check if the COW logic is turned
2948 * off on btrfs. Given our write pattern that is quite
2949 * unfriendly to COW file systems this should greatly improve
2950 * performance on COW file systems, such as btrfs, at the
2951 * expense of data integrity features (which shouldn't be too
2952 * bad, given that we do our own checksumming). */
2953
2954 r = btrfs_is_filesystem(f->fd);
2955 if (r < 0)
2956 return log_warning_errno(r, "Failed to determine if journal is on btrfs: %m");
2957 if (!r)
2958 return 0;
2959
2960 r = read_attr_fd(f->fd, &attrs);
2961 if (r < 0)
2962 return log_warning_errno(r, "Failed to read file attributes: %m");
2963
2964 if (attrs & FS_NOCOW_FL) {
2965 log_debug("Detected btrfs file system with copy-on-write disabled, all is good.");
2966 return 0;
2967 }
2968
2969 log_notice("Creating journal file %s on a btrfs file system, and copy-on-write is enabled. "
2970 "This is likely to slow down journal access substantially, please consider turning "
2971 "off the copy-on-write file attribute on the journal directory, using chattr +C.", f->path);
2972
2973 return 1;
2974}
2975
0284adc6 2976int journal_file_open(
5d1ce257 2977 int fd,
0284adc6
LP		 2978 const char *fname,
2979 int flags,
2980 mode_t mode,
2981 bool compress,
baed47c3 2982 bool seal,
0284adc6
LP		 2983 JournalMetrics *metrics,
2984 MMapCache *mmap_cache,
b58c888f 2985 Set *deferred_closes,
0284adc6
LP		 2986 JournalFile *template,
2987 JournalFile **ret) {
7560fffc 2988
fa6ac760 2989 bool newly_created = false;
0284adc6 2990 JournalFile *f;
fa6ac760 2991 void *h;
0284adc6 2992 int r;
7560fffc 2993
0559d3a5 2994 assert(ret);
5d1ce257 2995 assert(fd >= 0 || fname);
7560fffc 2996
0284adc6
LP		 2997 if ((flags & O_ACCMODE) != O_RDONLY &&
2998 (flags & O_ACCMODE) != O_RDWR)
2999 return -EINVAL;
7560fffc 3000
5d1ce257
LP		 3001 if (fname) {
3002 if (!endswith(fname, ".journal") &&
3003 !endswith(fname, ".journal~"))
3004 return -EINVAL;
3005 }
7560fffc 3006
0284adc6
LP		 3007 f = new0(JournalFile, 1);
3008 if (!f)
3009 return -ENOMEM;
7560fffc 3010
5d1ce257 3011 f->fd = fd;
0284adc6 3012 f->mode = mode;
7560fffc 3013
0284adc6
LP		 3014 f->flags = flags;
3015 f->prot = prot_from_flags(flags);
3016 f->writable = (flags & O_ACCMODE) != O_RDONLY;
92261977 3017#if defined(HAVE_LZ4)
d89c8fdf
ZJS		 3018 f->compress_lz4 = compress;
3019#elif defined(HAVE_XZ)
3020 f->compress_xz = compress;
48b61739 3021#endif
49a32d43 3022#ifdef HAVE_GCRYPT
baed47c3 3023 f->seal = seal;
49a32d43 3024#endif
7560fffc 3025
0284adc6
LP		 3026 if (mmap_cache)
3027 f->mmap = mmap_cache_ref(mmap_cache);
3028 else {
84168d80 3029 f->mmap = mmap_cache_new();
0284adc6
LP		 3030 if (!f->mmap) {
3031 r = -ENOMEM;
3032 goto fail;
3033 }
3034 }
7560fffc 3035
5d1ce257
LP		 3036 if (fname)
3037 f->path = strdup(fname);
3038 else /* If we don't know the path, fill in something explanatory and vaguely useful */
3039 asprintf(&f->path, "/proc/self/%i", fd);
0284adc6
LP		 3040 if (!f->path) {
3041 r = -ENOMEM;
3042 goto fail;
3043 }
7560fffc 3044
4743015d 3045 f->chain_cache = ordered_hashmap_new(&uint64_hash_ops);
a4bcff5b
LP		 3046 if (!f->chain_cache) {
3047 r = -ENOMEM;
3048 goto fail;
3049 }
3050
0284adc6 3051 if (f->fd < 0) {
5d1ce257
LP		 3052 f->fd = open(f->path, f->flags|O_CLOEXEC, f->mode);
3053 if (f->fd < 0) {
3054 r = -errno;
3055 goto fail;
3056 }
3057
3058 /* fds we opened here by us should also be closed by us. */
3059 f->close_fd = true;
7560fffc 3060 }
7560fffc 3061
2678031a
LP		 3062 r = journal_file_fstat(f);
3063 if (r < 0)
0284adc6 3064 goto fail;
7560fffc 3065
0284adc6 3066 if (f->last_stat.st_size == 0 && f->writable) {
11689d2a 3067
fc68c929 3068 (void) journal_file_warn_btrfs(f);
11689d2a 3069
fb0951b0
LP		 3070 /* Let's attach the creation time to the journal file,
3071 * so that the vacuuming code knows the age of this
3072 * file even if the file might end up corrupted one
3073 * day... Ideally we'd just use the creation time many
3074 * file systems maintain for each file, but there is
3075 * currently no usable API to query this, hence let's
3076 * emulate this via extended attributes. If extended
3077 * attributes are not supported we'll just skip this,
7517e174 3078 * and rely solely on mtime/atime/ctime of the file. */
fb0951b0 3079
d61b600d 3080 fd_setcrtime(f->fd, 0);
7560fffc 3081
feb12d3e 3082#ifdef HAVE_GCRYPT
0284adc6 3083 /* Try to load the FSPRG state, and if we can't, then
baed47c3 3084 * just don't do sealing */
49a32d43
LP		 3085 if (f->seal) {
3086 r = journal_file_fss_load(f);
3087 if (r < 0)
3088 f->seal = false;
3089 }
feb12d3e 3090#endif
7560fffc 3091
0284adc6
LP		 3092 r = journal_file_init_header(f, template);
3093 if (r < 0)
3094 goto fail;
7560fffc 3095
2678031a
LP		 3096 r = journal_file_fstat(f);
3097 if (r < 0)
0284adc6 3098 goto fail;
fb0951b0
LP		 3099
3100 newly_created = true;
0284adc6 3101 }
7560fffc 3102
0284adc6 3103 if (f->last_stat.st_size < (off_t) HEADER_SIZE_MIN) {
cfb571f3 3104 r = -ENODATA;
0284adc6
LP		 3105 goto fail;
3106 }
7560fffc 3107
fa6ac760 3108 r = mmap_cache_get(f->mmap, f->fd, f->prot, CONTEXT_HEADER, true, 0, PAGE_ALIGN(sizeof(Header)), &f->last_stat, &h);
977eaa1e 3109 if (r < 0)
0284adc6 3110 goto fail;
7560fffc 3111
fa6ac760
LP		 3112 f->header = h;
3113
0284adc6 3114 if (!newly_created) {
b58c888f
VC		 3115 if (deferred_closes)
3116 journal_file_close_set(deferred_closes);
3117
0284adc6
LP		 3118 r = journal_file_verify_header(f);
3119 if (r < 0)
3120 goto fail;
3121 }
7560fffc 3122
feb12d3e 3123#ifdef HAVE_GCRYPT
0284adc6 3124 if (!newly_created && f->writable) {
baed47c3 3125 r = journal_file_fss_load(f);
0284adc6
LP		 3126 if (r < 0)
3127 goto fail;
3128 }
feb12d3e 3129#endif
cec736d2
LP		 3130
3131 if (f->writable) {
4a92baf3
LP		 3132 if (metrics) {
3133 journal_default_metrics(metrics, f->fd);
3134 f->metrics = *metrics;
3135 } else if (template)
3136 f->metrics = template->metrics;
3137
cec736d2
LP		 3138 r = journal_file_refresh_header(f);
3139 if (r < 0)
3140 goto fail;
3141 }
3142
feb12d3e 3143#ifdef HAVE_GCRYPT
baed47c3 3144 r = journal_file_hmac_setup(f);
14d10188
LP		 3145 if (r < 0)
3146 goto fail;
feb12d3e 3147#endif
14d10188 3148
cec736d2 3149 if (newly_created) {
de190aef 3150 r = journal_file_setup_field_hash_table(f);
cec736d2
LP		 3151 if (r < 0)
3152 goto fail;
3153
de190aef 3154 r = journal_file_setup_data_hash_table(f);
cec736d2
LP		 3155 if (r < 0)
3156 goto fail;
7560fffc 3157
feb12d3e 3158#ifdef HAVE_GCRYPT
7560fffc
LP		 3159 r = journal_file_append_first_tag(f);
3160 if (r < 0)
3161 goto fail;
feb12d3e 3162#endif
cec736d2
LP		 3163 }
3164
fa6ac760
LP		 3165 if (mmap_cache_got_sigbus(f->mmap, f->fd)) {
3166 r = -EIO;
3167 goto fail;
3168 }
3169
7a24f3bf 3170 if (template && template->post_change_timer) {
e167d7fd
LP		 3171 r = journal_file_enable_post_change_timer(
3172 f,
3173 sd_event_source_get_event(template->post_change_timer),
3174 template->post_change_timer_period);
7a24f3bf 3175
7a24f3bf
VC		 3176 if (r < 0)
3177 goto fail;
3178 }
3179
f8e2f4d6 3180 /* The file is opened now successfully, thus we take possession of any passed in fd. */
5d1ce257
LP		 3181 f->close_fd = true;
3182
0559d3a5 3183 *ret = f;
cec736d2
LP		 3184 return 0;
3185
3186fail:
fa6ac760
LP		 3187 if (f->fd >= 0 && mmap_cache_got_sigbus(f->mmap, f->fd))
3188 r = -EIO;
3189
69a3a6fd 3190 (void) journal_file_close(f);
cec736d2
LP		 3191
3192 return r;
3193}
0ac38b70 3194
b58c888f 3195int journal_file_rotate(JournalFile **f, bool compress, bool seal, Set *deferred_closes) {
57535f47 3196 _cleanup_free_ char *p = NULL;
0ac38b70
LP		 3197 size_t l;
3198 JournalFile *old_file, *new_file = NULL;
3199 int r;
3200
3201 assert(f);
3202 assert(*f);
3203
3204 old_file = *f;
3205
3206 if (!old_file->writable)
3207 return -EINVAL;
3208
5d1ce257
LP		 3209 /* Is this a journal file that was passed to us as fd? If so, we synthesized a path name for it, and we refuse
3210 * rotation, since we don't know the actual path, and couldn't rename the file hence.*/
3211 if (path_startswith(old_file->path, "/proc/self/fd"))
3212 return -EINVAL;
3213
0ac38b70
LP		 3214 if (!endswith(old_file->path, ".journal"))
3215 return -EINVAL;
3216
3217 l = strlen(old_file->path);
57535f47
ZJS		 3218 r = asprintf(&p, "%.*s@" SD_ID128_FORMAT_STR "-%016"PRIx64"-%016"PRIx64".journal",
3219 (int) l - 8, old_file->path,
3220 SD_ID128_FORMAT_VAL(old_file->header->seqnum_id),
3221 le64toh((*f)->header->head_entry_seqnum),
3222 le64toh((*f)->header->head_entry_realtime));
3223 if (r < 0)
0ac38b70
LP		 3224 return -ENOMEM;
3225
2678031a
LP		 3226 /* Try to rename the file to the archived version. If the file
3227 * already was deleted, we'll get ENOENT, let's ignore that
3228 * case. */
0ac38b70 3229 r = rename(old_file->path, p);
2678031a 3230 if (r < 0 && errno != ENOENT)
0ac38b70
LP		 3231 return -errno;
3232
1fcefd88
LP		 3233 /* Sync the rename to disk */
3234 (void) fsync_directory_of_file(old_file->fd);
3235
8eb85171
VC		 3236 /* Set as archive so offlining commits w/state=STATE_ARCHIVED.
3237 * Previously we would set old_file->header->state to STATE_ARCHIVED directly here,
3238 * but journal_file_set_offline() short-circuits when state != STATE_ONLINE, which
3239 * would result in the rotated journal never getting fsync() called before closing.
3240 * Now we simply queue the archive state by setting an archive bit, leaving the state
3241 * as STATE_ONLINE so proper offlining occurs. */
3242 old_file->archive = true;
0ac38b70 3243
f27a3864
LP		 3244 /* Currently, btrfs is not very good with out write patterns
3245 * and fragments heavily. Let's defrag our journal files when
3246 * we archive them */
3247 old_file->defrag_on_close = true;
3248
5d1ce257 3249 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);
b58c888f
VC		 3250
3251 if (deferred_closes &&
3252 set_put(deferred_closes, old_file) >= 0)
3253 (void) journal_file_set_offline(old_file, false);
3254 else
3255 (void) journal_file_close(old_file);
0ac38b70
LP		 3256
3257 *f = new_file;
3258 return r;
3259}
3260
9447a7f1
LP		 3261int journal_file_open_reliably(
3262 const char *fname,
3263 int flags,
3264 mode_t mode,
7560fffc 3265 bool compress,
baed47c3 3266 bool seal,
4a92baf3 3267 JournalMetrics *metrics,
27370278 3268 MMapCache *mmap_cache,
b58c888f 3269 Set *deferred_closes,
9447a7f1
LP		 3270 JournalFile *template,
3271 JournalFile **ret) {
3272
3273 int r;
3274 size_t l;
ed375beb 3275 _cleanup_free_ char *p = NULL;
9447a7f1 3276
5d1ce257 3277 r = journal_file_open(-1, fname, flags, mode, compress, seal, metrics, mmap_cache, deferred_closes, template, ret);
288359db
ZJS		 3278 if (!IN_SET(r,
3279 -EBADMSG, /* corrupted */
3280 -ENODATA, /* truncated */
3281 -EHOSTDOWN, /* other machine */
3282 -EPROTONOSUPPORT, /* incompatible feature */
3283 -EBUSY, /* unclean shutdown */
3284 -ESHUTDOWN, /* already archived */
3285 -EIO, /* IO error, including SIGBUS on mmap */
3286 -EIDRM /* File has been deleted */))
9447a7f1
LP		 3287 return r;
3288
3289 if ((flags & O_ACCMODE) == O_RDONLY)
3290 return r;
3291
3292 if (!(flags & O_CREAT))
3293 return r;
3294
7560fffc
LP		 3295 if (!endswith(fname, ".journal"))
3296 return r;
3297
5c70eab4
LP		 3298 /* The file is corrupted. Rotate it away and try it again (but only once) */
3299
9447a7f1 3300 l = strlen(fname);
d587eca5 3301 if (asprintf(&p, "%.*s@%016"PRIx64 "-%016"PRIx64 ".journal~",
57535f47 3302 (int) l - 8, fname,
d587eca5 3303 now(CLOCK_REALTIME),
9bf3b535 3304 random_u64()) < 0)
9447a7f1
LP		 3305 return -ENOMEM;
3306
65089b82 3307 if (rename(fname, p) < 0)
9447a7f1
LP		 3308 return -errno;
3309
f27a3864
LP		 3310 /* btrfs doesn't cope well with our write pattern and
3311 * fragments heavily. Let's defrag all files we rotate */
11689d2a 3312
a67d68b8 3313 (void) chattr_path(p, 0, FS_NOCOW_FL);
f27a3864
LP		 3314 (void) btrfs_defrag(p);
3315
65089b82 3316 log_warning_errno(r, "File %s corrupted or uncleanly shut down, renaming and replacing.", fname);
9447a7f1 3317
5d1ce257 3318 return journal_file_open(-1, fname, flags, mode, compress, seal, metrics, mmap_cache, deferred_closes, template, ret);
9447a7f1
LP		 3319}
3320
cf244689
LP		 3321int journal_file_copy_entry(JournalFile *from, JournalFile *to, Object *o, uint64_t p, uint64_t *seqnum, Object **ret, uint64_t *offset) {
3322 uint64_t i, n;
3323 uint64_t q, xor_hash = 0;
3324 int r;
3325 EntryItem *items;
3326 dual_timestamp ts;
3327
3328 assert(from);
3329 assert(to);
3330 assert(o);
3331 assert(p);
3332
3333 if (!to->writable)
3334 return -EPERM;
3335
3336 ts.monotonic = le64toh(o->entry.monotonic);
3337 ts.realtime = le64toh(o->entry.realtime);
3338
cf244689 3339 n = journal_file_entry_n_items(o);
4faa7004
TA		 3340 /* alloca() can't take 0, hence let's allocate at least one */
3341 items = alloca(sizeof(EntryItem) * MAX(1u, n));
cf244689
LP		 3342
3343 for (i = 0; i < n; i++) {
4fd052ae
FC		 3344 uint64_t l, h;
3345 le64_t le_hash;
cf244689
LP		 3346 size_t t;
3347 void *data;
3348 Object *u;
3349
3350 q = le64toh(o->entry.items[i].object_offset);
3351 le_hash = o->entry.items[i].hash;
3352
3353 r = journal_file_move_to_object(from, OBJECT_DATA, q, &o);
3354 if (r < 0)
3355 return r;
3356
3357 if (le_hash != o->data.hash)
3358 return -EBADMSG;
3359
3360 l = le64toh(o->object.size) - offsetof(Object, data.payload);
3361 t = (size_t) l;
3362
3363 /* We hit the limit on 32bit machines */
3364 if ((uint64_t) t != l)
3365 return -E2BIG;
3366
d89c8fdf 3367 if (o->object.flags & OBJECT_COMPRESSION_MASK) {
3b1a55e1 3368#if defined(HAVE_XZ) || defined(HAVE_LZ4)
a7f7d1bd 3369 size_t rsize = 0;
cf244689 3370
d89c8fdf
ZJS		 3371 r = decompress_blob(o->object.flags & OBJECT_COMPRESSION_MASK,
3372 o->data.payload, l, &from->compress_buffer, &from->compress_buffer_size, &rsize, 0);
3373 if (r < 0)
3374 return r;
cf244689
LP		 3375
3376 data = from->compress_buffer;
3377 l = rsize;
3b1a55e1
ZJS		 3378#else
3379 return -EPROTONOSUPPORT;
3380#endif
cf244689
LP		 3381 } else
3382 data = o->data.payload;
3383
3384 r = journal_file_append_data(to, data, l, &u, &h);
3385 if (r < 0)
3386 return r;
3387
3388 xor_hash ^= le64toh(u->data.hash);
3389 items[i].object_offset = htole64(h);
3390 items[i].hash = u->data.hash;
3391
3392 r = journal_file_move_to_object(from, OBJECT_ENTRY, p, &o);
3393 if (r < 0)
3394 return r;
3395 }
3396
fa6ac760
LP		 3397 r = journal_file_append_entry_internal(to, &ts, xor_hash, items, n, seqnum, ret, offset);
3398
3399 if (mmap_cache_got_sigbus(to->mmap, to->fd))
3400 return -EIO;
3401
3402 return r;
cf244689 3403}
babfc091 3404
8580d1f7
LP		 3405void journal_reset_metrics(JournalMetrics *m) {
3406 assert(m);
3407
3408 /* Set everything to "pick automatic values". */
3409
3410 *m = (JournalMetrics) {
3411 .min_use = (uint64_t) -1,
3412 .max_use = (uint64_t) -1,
3413 .min_size = (uint64_t) -1,
3414 .max_size = (uint64_t) -1,
3415 .keep_free = (uint64_t) -1,
3416 .n_max_files = (uint64_t) -1,
3417 };
3418}
3419
babfc091 3420void journal_default_metrics(JournalMetrics *m, int fd) {
8580d1f7 3421 char a[FORMAT_BYTES_MAX], b[FORMAT_BYTES_MAX], c[FORMAT_BYTES_MAX], d[FORMAT_BYTES_MAX], e[FORMAT_BYTES_MAX];
babfc091 3422 struct statvfs ss;
8580d1f7 3423 uint64_t fs_size;
babfc091
LP		 3424
3425 assert(m);
3426 assert(fd >= 0);
3427
3428 if (fstatvfs(fd, &ss) >= 0)
3429 fs_size = ss.f_frsize * ss.f_blocks;
8580d1f7
LP		 3430 else {
3431 log_debug_errno(errno, "Failed to detremine disk size: %m");
3432 fs_size = 0;
3433 }
babfc091
LP		 3434
3435 if (m->max_use == (uint64_t) -1) {
3436
3437 if (fs_size > 0) {
3438 m->max_use = PAGE_ALIGN(fs_size / 10); /* 10% of file system size */
3439
3440 if (m->max_use > DEFAULT_MAX_USE_UPPER)
3441 m->max_use = DEFAULT_MAX_USE_UPPER;
3442
3443 if (m->max_use < DEFAULT_MAX_USE_LOWER)
3444 m->max_use = DEFAULT_MAX_USE_LOWER;
3445 } else
3446 m->max_use = DEFAULT_MAX_USE_LOWER;
3447 } else {
3448 m->max_use = PAGE_ALIGN(m->max_use);
3449
8580d1f7 3450 if (m->max_use != 0 && m->max_use < JOURNAL_FILE_SIZE_MIN*2)
babfc091
LP		 3451 m->max_use = JOURNAL_FILE_SIZE_MIN*2;
3452 }
3453
8580d1f7
LP		 3454 if (m->min_use == (uint64_t) -1)
3455 m->min_use = DEFAULT_MIN_USE;
3456
3457 if (m->min_use > m->max_use)
3458 m->min_use = m->max_use;
3459
babfc091
LP		 3460 if (m->max_size == (uint64_t) -1) {
3461 m->max_size = PAGE_ALIGN(m->max_use / 8); /* 8 chunks */
3462
3463 if (m->max_size > DEFAULT_MAX_SIZE_UPPER)
3464 m->max_size = DEFAULT_MAX_SIZE_UPPER;
3465 } else
3466 m->max_size = PAGE_ALIGN(m->max_size);
3467
8580d1f7
LP		 3468 if (m->max_size != 0) {
3469 if (m->max_size < JOURNAL_FILE_SIZE_MIN)
3470 m->max_size = JOURNAL_FILE_SIZE_MIN;
babfc091 3471
8580d1f7
LP		 3472 if (m->max_use != 0 && m->max_size*2 > m->max_use)
3473 m->max_use = m->max_size*2;
3474 }
babfc091
LP		 3475
3476 if (m->min_size == (uint64_t) -1)
3477 m->min_size = JOURNAL_FILE_SIZE_MIN;
3478 else {
3479 m->min_size = PAGE_ALIGN(m->min_size);
3480
3481 if (m->min_size < JOURNAL_FILE_SIZE_MIN)
3482 m->min_size = JOURNAL_FILE_SIZE_MIN;
3483
8580d1f7 3484 if (m->max_size != 0 && m->min_size > m->max_size)
babfc091
LP		 3485 m->max_size = m->min_size;
3486 }
3487
3488 if (m->keep_free == (uint64_t) -1) {
3489
3490 if (fs_size > 0) {
8621b110 3491 m->keep_free = PAGE_ALIGN(fs_size * 3 / 20); /* 15% of file system size */
babfc091
LP		 3492
3493 if (m->keep_free > DEFAULT_KEEP_FREE_UPPER)
3494 m->keep_free = DEFAULT_KEEP_FREE_UPPER;
3495
3496 } else
3497 m->keep_free = DEFAULT_KEEP_FREE;
3498 }
3499
8580d1f7
LP		 3500 if (m->n_max_files == (uint64_t) -1)
3501 m->n_max_files = DEFAULT_N_MAX_FILES;
3502
3503 log_debug("Fixed min_use=%s max_use=%s max_size=%s min_size=%s keep_free=%s n_max_files=%" PRIu64,
3504 format_bytes(a, sizeof(a), m->min_use),
3505 format_bytes(b, sizeof(b), m->max_use),
3506 format_bytes(c, sizeof(c), m->max_size),
3507 format_bytes(d, sizeof(d), m->min_size),
3508 format_bytes(e, sizeof(e), m->keep_free),
3509 m->n_max_files);
babfc091 3510}
08984293
LP		 3511
3512int journal_file_get_cutoff_realtime_usec(JournalFile *f, usec_t *from, usec_t *to) {
08984293 3513 assert(f);
c88cc6af 3514 assert(f->header);
08984293
LP		 3515 assert(from || to);
3516
3517 if (from) {
162566a4
LP		 3518 if (f->header->head_entry_realtime == 0)
3519 return -ENOENT;
08984293 3520
162566a4 3521 *from = le64toh(f->header->head_entry_realtime);
08984293
LP		 3522 }
3523
3524 if (to) {
162566a4
LP		 3525 if (f->header->tail_entry_realtime == 0)
3526 return -ENOENT;
08984293 3527
162566a4 3528 *to = le64toh(f->header->tail_entry_realtime);
08984293
LP		 3529 }
3530
3531 return 1;
3532}
3533
3534int journal_file_get_cutoff_monotonic_usec(JournalFile *f, sd_id128_t boot_id, usec_t *from, usec_t *to) {
08984293
LP		 3535 Object *o;
3536 uint64_t p;
3537 int r;
3538
3539 assert(f);
3540 assert(from || to);
3541
47838ab3 3542 r = find_data_object_by_boot_id(f, boot_id, &o, &p);
08984293
LP		 3543 if (r <= 0)
3544 return r;
3545
3546 if (le64toh(o->data.n_entries) <= 0)
3547 return 0;
3548
3549 if (from) {
3550 r = journal_file_move_to_object(f, OBJECT_ENTRY, le64toh(o->data.entry_offset), &o);
3551 if (r < 0)
3552 return r;
3553
3554 *from = le64toh(o->entry.monotonic);
3555 }
3556
3557 if (to) {
3558 r = journal_file_move_to_object(f, OBJECT_DATA, p, &o);
3559 if (r < 0)
3560 return r;
3561
3562 r = generic_array_get_plus_one(f,
3563 le64toh(o->data.entry_offset),
3564 le64toh(o->data.entry_array_offset),
3565 le64toh(o->data.n_entries)-1,
3566 &o, NULL);
3567 if (r <= 0)
3568 return r;
3569
3570 *to = le64toh(o->entry.monotonic);
3571 }
3572
3573 return 1;
3574}
dca6219e 3575
fb0951b0 3576bool journal_file_rotate_suggested(JournalFile *f, usec_t max_file_usec) {
dca6219e 3577 assert(f);
c88cc6af 3578 assert(f->header);
dca6219e
LP		 3579
3580 /* If we gained new header fields we gained new features,
3581 * hence suggest a rotation */
361f9cbc
LP		 3582 if (le64toh(f->header->header_size) < sizeof(Header)) {
3583 log_debug("%s uses an outdated header, suggesting rotation.", f->path);
dca6219e 3584 return true;
361f9cbc 3585 }
dca6219e
LP		 3586
3587 /* Let's check if the hash tables grew over a certain fill
3588 * level (75%, borrowing this value from Java's hash table
3589 * implementation), and if so suggest a rotation. To calculate
3590 * the fill level we need the n_data field, which only exists
3591 * in newer versions. */
3592
3593 if (JOURNAL_HEADER_CONTAINS(f->header, n_data))
361f9cbc 3594 if (le64toh(f->header->n_data) * 4ULL > (le64toh(f->header->data_hash_table_size) / sizeof(HashItem)) * 3ULL) {
507f22bd 3595 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.",
361f9cbc
LP		 3596 f->path,
3597 100.0 * (double) le64toh(f->header->n_data) / ((double) (le64toh(f->header->data_hash_table_size) / sizeof(HashItem))),
507f22bd
ZJS		 3598 le64toh(f->header->n_data),
3599 le64toh(f->header->data_hash_table_size) / sizeof(HashItem),
3600 (unsigned long long) f->last_stat.st_size,
3601 f->last_stat.st_size / le64toh(f->header->n_data));
dca6219e 3602 return true;
361f9cbc 3603 }
dca6219e
LP		 3604
3605 if (JOURNAL_HEADER_CONTAINS(f->header, n_fields))
361f9cbc 3606 if (le64toh(f->header->n_fields) * 4ULL > (le64toh(f->header->field_hash_table_size) / sizeof(HashItem)) * 3ULL) {
507f22bd 3607 log_debug("Field hash table of %s has a fill level at %.1f (%"PRIu64" of %"PRIu64" items), suggesting rotation.",
361f9cbc
LP		 3608 f->path,
3609 100.0 * (double) le64toh(f->header->n_fields) / ((double) (le64toh(f->header->field_hash_table_size) / sizeof(HashItem))),
507f22bd
ZJS		 3610 le64toh(f->header->n_fields),
3611 le64toh(f->header->field_hash_table_size) / sizeof(HashItem));
dca6219e 3612 return true;
361f9cbc 3613 }
dca6219e 3614
0598fd4a
LP		 3615 /* Are the data objects properly indexed by field objects? */
3616 if (JOURNAL_HEADER_CONTAINS(f->header, n_data) &&
3617 JOURNAL_HEADER_CONTAINS(f->header, n_fields) &&
3618 le64toh(f->header->n_data) > 0 &&
3619 le64toh(f->header->n_fields) == 0)
3620 return true;
3621
fb0951b0
LP		 3622 if (max_file_usec > 0) {
3623 usec_t t, h;
3624
3625 h = le64toh(f->header->head_entry_realtime);
3626 t = now(CLOCK_REALTIME);
3627
3628 if (h > 0 && t > h + max_file_usec)
3629 return true;
3630 }
3631
dca6219e
LP		 3632 return false;
3633}
Unnamed repository; edit this file 'description' to name the repository.