]> 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
Merge pull request #5571 from keszybz/tarball-generation-tweaks
[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"
4761fd0f 45#include "strv.h"
89a5a90c 46#include "xattr-util.h"
cec736d2 47
4a92baf3
LP		 48#define DEFAULT_DATA_HASH_TABLE_SIZE (2047ULL*sizeof(HashItem))
49#define DEFAULT_FIELD_HASH_TABLE_SIZE (333ULL*sizeof(HashItem))
cec736d2 50
be19b7df 51#define COMPRESSION_SIZE_THRESHOLD (512ULL)
807e17f0 52
babfc091 53/* This is the minimum journal file size */
16098e93 54#define JOURNAL_FILE_SIZE_MIN (512ULL*1024ULL) /* 512 KiB */
babfc091
LP		 55
56/* These are the lower and upper bounds if we deduce the max_use value
57 * from the file system size */
58#define DEFAULT_MAX_USE_LOWER (1ULL*1024ULL*1024ULL) /* 1 MiB */
59#define DEFAULT_MAX_USE_UPPER (4ULL*1024ULL*1024ULL*1024ULL) /* 4 GiB */
60
8580d1f7
LP		 61/* This is the default minimal use limit, how much we'll use even if keep_free suggests otherwise. */
62#define DEFAULT_MIN_USE (1ULL*1024ULL*1024ULL) /* 1 MiB */
63
babfc091 64/* This is the upper bound if we deduce max_size from max_use */
71100051 65#define DEFAULT_MAX_SIZE_UPPER (128ULL*1024ULL*1024ULL) /* 128 MiB */
babfc091
LP		 66
67/* This is the upper bound if we deduce the keep_free value from the
68 * file system size */
69#define DEFAULT_KEEP_FREE_UPPER (4ULL*1024ULL*1024ULL*1024ULL) /* 4 GiB */
70
71/* This is the keep_free value when we can't determine the system
72 * size */
73#define DEFAULT_KEEP_FREE (1024ULL*1024ULL) /* 1 MB */
74
8580d1f7
LP		 75/* This is the default maximum number of journal files to keep around. */
76#define DEFAULT_N_MAX_FILES (100)
77
dca6219e
LP		 78/* n_data was the first entry we added after the initial file format design */
79#define HEADER_SIZE_MIN ALIGN64(offsetof(Header, n_data))
cec736d2 80
a4bcff5b
LP		 81/* How many entries to keep in the entry array chain cache at max */
82#define CHAIN_CACHE_MAX 20
83
a676e665
LP		 84/* How much to increase the journal file size at once each time we allocate something new. */
85#define FILE_SIZE_INCREASE (8ULL*1024ULL*1024ULL) /* 8MB */
86
2678031a
LP		 87/* Reread fstat() of the file for detecting deletions at least this often */
88#define LAST_STAT_REFRESH_USEC (5*USEC_PER_SEC)
89
fa6ac760
LP		 90/* The mmap context to use for the header we pick as one above the last defined typed */
91#define CONTEXT_HEADER _OBJECT_TYPE_MAX
92
ac2e41f5
VC		 93/* This may be called from a separate thread to prevent blocking the caller for the duration of fsync().
94 * As a result we use atomic operations on f->offline_state for inter-thread communications with
95 * journal_file_set_offline() and journal_file_set_online(). */
96static void journal_file_set_offline_internal(JournalFile *f) {
26687bf8 97 assert(f);
ac2e41f5
VC		 98 assert(f->fd >= 0);
99 assert(f->header);
100
101 for (;;) {
102 switch (f->offline_state) {
103 case OFFLINE_CANCEL:
104 if (!__sync_bool_compare_and_swap(&f->offline_state, OFFLINE_CANCEL, OFFLINE_DONE))
105 continue;
106 return;
107
108 case OFFLINE_AGAIN_FROM_SYNCING:
109 if (!__sync_bool_compare_and_swap(&f->offline_state, OFFLINE_AGAIN_FROM_SYNCING, OFFLINE_SYNCING))
110 continue;
111 break;
112
113 case OFFLINE_AGAIN_FROM_OFFLINING:
114 if (!__sync_bool_compare_and_swap(&f->offline_state, OFFLINE_AGAIN_FROM_OFFLINING, OFFLINE_SYNCING))
115 continue;
116 break;
117
118 case OFFLINE_SYNCING:
119 (void) fsync(f->fd);
26687bf8 120
ac2e41f5
VC		 121 if (!__sync_bool_compare_and_swap(&f->offline_state, OFFLINE_SYNCING, OFFLINE_OFFLINING))
122 continue;
26687bf8 123
8eb85171 124 f->header->state = f->archive ? STATE_ARCHIVED : STATE_OFFLINE;
ac2e41f5
VC		 125 (void) fsync(f->fd);
126 break;
127
128 case OFFLINE_OFFLINING:
129 if (!__sync_bool_compare_and_swap(&f->offline_state, OFFLINE_OFFLINING, OFFLINE_DONE))
130 continue;
131 /* fall through */
132
133 case OFFLINE_DONE:
134 return;
135
136 case OFFLINE_JOINED:
137 log_debug("OFFLINE_JOINED unexpected offline state for journal_file_set_offline_internal()");
138 return;
139 }
140 }
141}
142
143static void * journal_file_set_offline_thread(void *arg) {
144 JournalFile *f = arg;
145
146 journal_file_set_offline_internal(f);
147
148 return NULL;
149}
150
151static int journal_file_set_offline_thread_join(JournalFile *f) {
152 int r;
153
154 assert(f);
155
156 if (f->offline_state == OFFLINE_JOINED)
157 return 0;
158
159 r = pthread_join(f->offline_thread, NULL);
160 if (r)
161 return -r;
162
163 f->offline_state = OFFLINE_JOINED;
26687bf8 164
fa6ac760
LP		 165 if (mmap_cache_got_sigbus(f->mmap, f->fd))
166 return -EIO;
167
ac2e41f5
VC		 168 return 0;
169}
26687bf8 170
ac2e41f5
VC		 171/* Trigger a restart if the offline thread is mid-flight in a restartable state. */
172static bool journal_file_set_offline_try_restart(JournalFile *f) {
173 for (;;) {
174 switch (f->offline_state) {
175 case OFFLINE_AGAIN_FROM_SYNCING:
176 case OFFLINE_AGAIN_FROM_OFFLINING:
177 return true;
178
179 case OFFLINE_CANCEL:
180 if (!__sync_bool_compare_and_swap(&f->offline_state, OFFLINE_CANCEL, OFFLINE_AGAIN_FROM_SYNCING))
181 continue;
182 return true;
183
184 case OFFLINE_SYNCING:
185 if (!__sync_bool_compare_and_swap(&f->offline_state, OFFLINE_SYNCING, OFFLINE_AGAIN_FROM_SYNCING))
186 continue;
187 return true;
188
189 case OFFLINE_OFFLINING:
190 if (!__sync_bool_compare_and_swap(&f->offline_state, OFFLINE_OFFLINING, OFFLINE_AGAIN_FROM_OFFLINING))
191 continue;
192 return true;
26687bf8
OS		 193
194 default:
ac2e41f5
VC		 195 return false;
196 }
26687bf8
OS		 197 }
198}
199
ac2e41f5
VC		 200/* Sets a journal offline.
201 *
202 * If wait is false then an offline is dispatched in a separate thread for a
203 * subsequent journal_file_set_offline() or journal_file_set_online() of the
204 * same journal to synchronize with.
205 *
206 * If wait is true, then either an existing offline thread will be restarted
207 * and joined, or if none exists the offline is simply performed in this
208 * context without involving another thread.
209 */
210int journal_file_set_offline(JournalFile *f, bool wait) {
211 bool restarted;
212 int r;
213
26687bf8
OS		 214 assert(f);
215
216 if (!f->writable)
217 return -EPERM;
218
219 if (!(f->fd >= 0 && f->header))
220 return -EINVAL;
221
b8f99e27
VC		 222 /* An offlining journal is implicitly online and may modify f->header->state,
223 * we must also join any potentially lingering offline thread when not online. */
224 if (!journal_file_is_offlining(f) && f->header->state != STATE_ONLINE)
225 return journal_file_set_offline_thread_join(f);
26687bf8 226
ac2e41f5
VC		 227 /* Restart an in-flight offline thread and wait if needed, or join a lingering done one. */
228 restarted = journal_file_set_offline_try_restart(f);
229 if ((restarted && wait) || !restarted) {
230 r = journal_file_set_offline_thread_join(f);
231 if (r < 0)
232 return r;
233 }
26687bf8 234
ac2e41f5
VC		 235 if (restarted)
236 return 0;
237
238 /* Initiate a new offline. */
239 f->offline_state = OFFLINE_SYNCING;
fa6ac760 240
ac2e41f5
VC		 241 if (wait) /* Without using a thread if waiting. */
242 journal_file_set_offline_internal(f);
243 else {
244 r = pthread_create(&f->offline_thread, NULL, journal_file_set_offline_thread, f);
ec9ffa2c
VC		 245 if (r > 0) {
246 f->offline_state = OFFLINE_JOINED;
ac2e41f5 247 return -r;
ec9ffa2c 248 }
ac2e41f5
VC		 249 }
250
251 return 0;
252}
253
254static int journal_file_set_online(JournalFile *f) {
255 bool joined = false;
256
257 assert(f);
258
259 if (!f->writable)
260 return -EPERM;
261
262 if (!(f->fd >= 0 && f->header))
263 return -EINVAL;
264
265 while (!joined) {
266 switch (f->offline_state) {
267 case OFFLINE_JOINED:
268 /* No offline thread, no need to wait. */
269 joined = true;
270 break;
271
272 case OFFLINE_SYNCING:
273 if (!__sync_bool_compare_and_swap(&f->offline_state, OFFLINE_SYNCING, OFFLINE_CANCEL))
274 continue;
275 /* Canceled syncing prior to offlining, no need to wait. */
276 break;
277
278 case OFFLINE_AGAIN_FROM_SYNCING:
279 if (!__sync_bool_compare_and_swap(&f->offline_state, OFFLINE_AGAIN_FROM_SYNCING, OFFLINE_CANCEL))
280 continue;
281 /* Canceled restart from syncing, no need to wait. */
282 break;
283
284 case OFFLINE_AGAIN_FROM_OFFLINING:
285 if (!__sync_bool_compare_and_swap(&f->offline_state, OFFLINE_AGAIN_FROM_OFFLINING, OFFLINE_CANCEL))
286 continue;
287 /* Canceled restart from offlining, must wait for offlining to complete however. */
288
ec251fe7 289 /* fall through */
ac2e41f5
VC		 290 default: {
291 int r;
292
293 r = journal_file_set_offline_thread_join(f);
294 if (r < 0)
295 return r;
296
297 joined = true;
298 break;
299 }
300 }
301 }
26687bf8 302
fa6ac760
LP		 303 if (mmap_cache_got_sigbus(f->mmap, f->fd))
304 return -EIO;
305
ac2e41f5
VC		 306 switch (f->header->state) {
307 case STATE_ONLINE:
308 return 0;
26687bf8 309
ac2e41f5
VC		 310 case STATE_OFFLINE:
311 f->header->state = STATE_ONLINE;
312 (void) fsync(f->fd);
313 return 0;
314
315 default:
316 return -EINVAL;
317 }
26687bf8
OS		 318}
319
b58c888f
VC		 320bool journal_file_is_offlining(JournalFile *f) {
321 assert(f);
322
323 __sync_synchronize();
324
325 if (f->offline_state == OFFLINE_DONE ||
326 f->offline_state == OFFLINE_JOINED)
327 return false;
328
329 return true;
330}
331
804ae586 332JournalFile* journal_file_close(JournalFile *f) {
de190aef 333 assert(f);
cec736d2 334
feb12d3e 335#ifdef HAVE_GCRYPT
b0af6f41 336 /* Write the final tag */
43cd8794
FB		 337 if (f->seal && f->writable) {
338 int r;
339
340 r = journal_file_append_tag(f);
341 if (r < 0)
342 log_error_errno(r, "Failed to append tag when closing journal: %m");
343 }
feb12d3e 344#endif
b0af6f41 345
7a24f3bf
VC		 346 if (f->post_change_timer) {
347 int enabled;
348
349 if (sd_event_source_get_enabled(f->post_change_timer, &enabled) >= 0)
350 if (enabled == SD_EVENT_ONESHOT)
351 journal_file_post_change(f);
352
e167d7fd 353 (void) sd_event_source_set_enabled(f->post_change_timer, SD_EVENT_OFF);
7a24f3bf
VC		 354 sd_event_source_unref(f->post_change_timer);
355 }
356
ac2e41f5 357 journal_file_set_offline(f, true);
cec736d2 358
fa6ac760
LP		 359 if (f->mmap && f->fd >= 0)
360 mmap_cache_close_fd(f->mmap, f->fd);
cec736d2 361
11689d2a
LP		 362 if (f->fd >= 0 && f->defrag_on_close) {
363
364 /* Be friendly to btrfs: turn COW back on again now,
365 * and defragment the file. We won't write to the file
366 * ever again, hence remove all fragmentation, and
367 * reenable all the good bits COW usually provides
368 * (such as data checksumming). */
369
1ed8f8c1 370 (void) chattr_fd(f->fd, 0, FS_NOCOW_FL);
11689d2a
LP		 371 (void) btrfs_defrag_fd(f->fd);
372 }
f27a3864 373
5d1ce257
LP		 374 if (f->close_fd)
375 safe_close(f->fd);
cec736d2 376 free(f->path);
807e17f0 377
f649045c 378 mmap_cache_unref(f->mmap);
16e9f408 379
4743015d 380 ordered_hashmap_free_free(f->chain_cache);
a4bcff5b 381
d89c8fdf 382#if defined(HAVE_XZ) || defined(HAVE_LZ4)
807e17f0
LP		 383 free(f->compress_buffer);
384#endif
385
7560fffc 386#ifdef HAVE_GCRYPT
baed47c3
LP		 387 if (f->fss_file)
388 munmap(f->fss_file, PAGE_ALIGN(f->fss_file_size));
dc4ebc07 389 else
b7c9ae91
LP		 390 free(f->fsprg_state);
391
392 free(f->fsprg_seed);
7560fffc
LP		 393
394 if (f->hmac)
395 gcry_md_close(f->hmac);
396#endif
397
6b430fdb 398 return mfree(f);
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
4214009f
ZJS		 511static bool warn_wrong_flags(const JournalFile *f, bool compatible) {
512 const uint32_t any = compatible ? HEADER_COMPATIBLE_ANY : HEADER_INCOMPATIBLE_ANY,
513 supported = compatible ? HEADER_COMPATIBLE_SUPPORTED : HEADER_INCOMPATIBLE_SUPPORTED;
514 const char *type = compatible ? "compatible" : "incompatible";
d89c8fdf
ZJS		 515 uint32_t flags;
516
4214009f
ZJS		 517 flags = le32toh(compatible ? f->header->compatible_flags : f->header->incompatible_flags);
518
519 if (flags & ~supported) {
520 if (flags & ~any)
4761fd0f 521 log_debug("Journal file %s has unknown %s flags 0x%"PRIx32,
4214009f
ZJS		 522 f->path, type, flags & ~any);
523 flags = (flags & any) & ~supported;
4761fd0f
ZJS		 524 if (flags) {
525 const char* strv[3];
526 unsigned n = 0;
527 _cleanup_free_ char *t = NULL;
528
529 if (compatible && (flags & HEADER_COMPATIBLE_SEALED))
530 strv[n++] = "sealed";
531 if (!compatible && (flags & HEADER_INCOMPATIBLE_COMPRESSED_XZ))
532 strv[n++] = "xz-compressed";
533 if (!compatible && (flags & HEADER_INCOMPATIBLE_COMPRESSED_LZ4))
534 strv[n++] = "lz4-compressed";
535 strv[n] = NULL;
536 assert(n < ELEMENTSOF(strv));
537
538 t = strv_join((char**) strv, ", ");
539 log_debug("Journal file %s uses %s %s %s disabled at compilation time.",
540 f->path, type, n > 1 ? "flags" : "flag", strnull(t));
541 }
4214009f
ZJS		 542 return true;
543 }
544
545 return false;
546}
547
548static int journal_file_verify_header(JournalFile *f) {
cec736d2 549 assert(f);
c88cc6af 550 assert(f->header);
cec736d2 551
7560fffc 552 if (memcmp(f->header->signature, HEADER_SIGNATURE, 8))
cec736d2
LP		 553 return -EBADMSG;
554
4214009f
ZJS		 555 /* In both read and write mode we refuse to open files with incompatible
556 * flags we don't know. */
557 if (warn_wrong_flags(f, false))
cec736d2
LP		 558 return -EPROTONOSUPPORT;
559
4214009f
ZJS		 560 /* When open for writing we refuse to open files with compatible flags, too. */
561 if (f->writable && warn_wrong_flags(f, true))
d89c8fdf 562 return -EPROTONOSUPPORT;
7560fffc 563
db11ac1a
LP		 564 if (f->header->state >= _STATE_MAX)
565 return -EBADMSG;
566
dca6219e
LP		 567 /* The first addition was n_data, so check that we are at least this large */
568 if (le64toh(f->header->header_size) < HEADER_SIZE_MIN)
23b0b2b2
LP		 569 return -EBADMSG;
570
8088cbd3 571 if (JOURNAL_HEADER_SEALED(f->header) && !JOURNAL_HEADER_CONTAINS(f->header, n_entry_arrays))
beec0085
LP		 572 return -EBADMSG;
573
db11ac1a
LP		 574 if ((le64toh(f->header->header_size) + le64toh(f->header->arena_size)) > (uint64_t) f->last_stat.st_size)
575 return -ENODATA;
576
577 if (le64toh(f->header->tail_object_offset) > (le64toh(f->header->header_size) + le64toh(f->header->arena_size)))
578 return -ENODATA;
579
7762e02b
LP		 580 if (!VALID64(le64toh(f->header->data_hash_table_offset)) ||
581 !VALID64(le64toh(f->header->field_hash_table_offset)) ||
582 !VALID64(le64toh(f->header->tail_object_offset)) ||
583 !VALID64(le64toh(f->header->entry_array_offset)))
584 return -ENODATA;
585
cec736d2 586 if (f->writable) {
cec736d2 587 sd_id128_t machine_id;
ae739cc1 588 uint8_t state;
cec736d2
LP		 589 int r;
590
591 r = sd_id128_get_machine(&machine_id);
592 if (r < 0)
593 return r;
594
595 if (!sd_id128_equal(machine_id, f->header->machine_id))
596 return -EHOSTDOWN;
597
de190aef 598 state = f->header->state;
cec736d2 599
b288cdeb
ZJS		 600 if (state == STATE_ARCHIVED)
601 return -ESHUTDOWN; /* Already archived */
602 else if (state == STATE_ONLINE) {
71fa6f00
LP		 603 log_debug("Journal file %s is already online. Assuming unclean closing.", f->path);
604 return -EBUSY;
b288cdeb 605 } else if (state != STATE_OFFLINE) {
8facc349 606 log_debug("Journal file %s has unknown state %i.", f->path, state);
71fa6f00
LP		 607 return -EBUSY;
608 }
ae739cc1
LP		 609
610 /* Don't permit appending to files from the future. Because otherwise the realtime timestamps wouldn't
611 * be strictly ordered in the entries in the file anymore, and we can't have that since it breaks
612 * bisection. */
613 if (le64toh(f->header->tail_entry_realtime) > now(CLOCK_REALTIME)) {
614 log_debug("Journal file %s is from the future, refusing to append new data to it that'd be older.", f->path);
615 return -ETXTBSY;
616 }
cec736d2
LP		 617 }
618
d89c8fdf
ZJS		 619 f->compress_xz = JOURNAL_HEADER_COMPRESSED_XZ(f->header);
620 f->compress_lz4 = JOURNAL_HEADER_COMPRESSED_LZ4(f->header);
c586dbf1 621
f1889c91 622 f->seal = JOURNAL_HEADER_SEALED(f->header);
7560fffc 623
cec736d2
LP		 624 return 0;
625}
626
2678031a
LP		 627static int journal_file_fstat(JournalFile *f) {
628 assert(f);
629 assert(f->fd >= 0);
630
631 if (fstat(f->fd, &f->last_stat) < 0)
632 return -errno;
633
634 f->last_stat_usec = now(CLOCK_MONOTONIC);
635
636 /* Refuse appending to files that are already deleted */
637 if (f->last_stat.st_nlink <= 0)
638 return -EIDRM;
639
640 return 0;
641}
642
cec736d2 643static int journal_file_allocate(JournalFile *f, uint64_t offset, uint64_t size) {
eda4b58b 644 uint64_t old_size, new_size;
fec2aa2f 645 int r;
cec736d2
LP		 646
647 assert(f);
c88cc6af 648 assert(f->header);
cec736d2 649
cec736d2 650 /* We assume that this file is not sparse, and we know that
38ac38b2 651 * for sure, since we always call posix_fallocate()
cec736d2
LP		 652 * ourselves */
653
fa6ac760
LP		 654 if (mmap_cache_got_sigbus(f->mmap, f->fd))
655 return -EIO;
656
cec736d2 657 old_size =
23b0b2b2 658 le64toh(f->header->header_size) +
cec736d2
LP		 659 le64toh(f->header->arena_size);
660
bc85bfee 661 new_size = PAGE_ALIGN(offset + size);
23b0b2b2
LP		 662 if (new_size < le64toh(f->header->header_size))
663 new_size = le64toh(f->header->header_size);
bc85bfee 664
2678031a
LP		 665 if (new_size <= old_size) {
666
667 /* We already pre-allocated enough space, but before
668 * we write to it, let's check with fstat() if the
669 * file got deleted, in order make sure we don't throw
670 * away the data immediately. Don't check fstat() for
671 * all writes though, but only once ever 10s. */
672
673 if (f->last_stat_usec + LAST_STAT_REFRESH_USEC > now(CLOCK_MONOTONIC))
674 return 0;
675
676 return journal_file_fstat(f);
677 }
678
679 /* Allocate more space. */
cec736d2 680
a676e665 681 if (f->metrics.max_size > 0 && new_size > f->metrics.max_size)
bc85bfee 682 return -E2BIG;
cec736d2 683
a676e665 684 if (new_size > f->metrics.min_size && f->metrics.keep_free > 0) {
cec736d2
LP		 685 struct statvfs svfs;
686
687 if (fstatvfs(f->fd, &svfs) >= 0) {
688 uint64_t available;
689
070052ab 690 available = LESS_BY((uint64_t) svfs.f_bfree * (uint64_t) svfs.f_bsize, f->metrics.keep_free);
cec736d2
LP		 691
692 if (new_size - old_size > available)
693 return -E2BIG;
694 }
695 }
696
eda4b58b
LP		 697 /* Increase by larger blocks at once */
698 new_size = ((new_size+FILE_SIZE_INCREASE-1) / FILE_SIZE_INCREASE) * FILE_SIZE_INCREASE;
699 if (f->metrics.max_size > 0 && new_size > f->metrics.max_size)
700 new_size = f->metrics.max_size;
701
bc85bfee
LP		 702 /* Note that the glibc fallocate() fallback is very
703 inefficient, hence we try to minimize the allocation area
704 as we can. */
fec2aa2f
GV		 705 r = posix_fallocate(f->fd, old_size, new_size - old_size);
706 if (r != 0)
707 return -r;
cec736d2 708
23b0b2b2 709 f->header->arena_size = htole64(new_size - le64toh(f->header->header_size));
cec736d2 710
2678031a 711 return journal_file_fstat(f);
cec736d2
LP		 712}
713
78519831 714static unsigned type_to_context(ObjectType type) {
d3d3208f 715 /* One context for each type, plus one catch-all for the rest */
69adae51 716 assert_cc(_OBJECT_TYPE_MAX <= MMAP_CACHE_MAX_CONTEXTS);
fa6ac760 717 assert_cc(CONTEXT_HEADER < MMAP_CACHE_MAX_CONTEXTS);
d05089d8 718 return type > OBJECT_UNUSED && type < _OBJECT_TYPE_MAX ? type : 0;
d3d3208f
MS		 719}
720
7a9dabea 721static int journal_file_move_to(JournalFile *f, ObjectType type, bool keep_always, uint64_t offset, uint64_t size, void **ret) {
2678031a
LP		 722 int r;
723
cec736d2 724 assert(f);
cec736d2
LP		 725 assert(ret);
726
7762e02b
LP		 727 if (size <= 0)
728 return -EINVAL;
729
2a59ea54 730 /* Avoid SIGBUS on invalid accesses */
4bbdcdb3
LP		 731 if (offset + size > (uint64_t) f->last_stat.st_size) {
732 /* Hmm, out of range? Let's refresh the fstat() data
733 * first, before we trust that check. */
734
2678031a
LP		 735 r = journal_file_fstat(f);
736 if (r < 0)
737 return r;
738
739 if (offset + size > (uint64_t) f->last_stat.st_size)
4bbdcdb3
LP		 740 return -EADDRNOTAVAIL;
741 }
742
7a9dabea 743 return mmap_cache_get(f->mmap, f->fd, f->prot, type_to_context(type), keep_always, offset, size, &f->last_stat, ret);
cec736d2
LP		 744}
745
16e9f408
LP		 746static uint64_t minimum_header_size(Object *o) {
747
b8e891e6 748 static const uint64_t table[] = {
16e9f408
LP		 749 [OBJECT_DATA] = sizeof(DataObject),
750 [OBJECT_FIELD] = sizeof(FieldObject),
751 [OBJECT_ENTRY] = sizeof(EntryObject),
752 [OBJECT_DATA_HASH_TABLE] = sizeof(HashTableObject),
753 [OBJECT_FIELD_HASH_TABLE] = sizeof(HashTableObject),
754 [OBJECT_ENTRY_ARRAY] = sizeof(EntryArrayObject),
755 [OBJECT_TAG] = sizeof(TagObject),
756 };
757
758 if (o->object.type >= ELEMENTSOF(table) || table[o->object.type] <= 0)
759 return sizeof(ObjectHeader);
760
761 return table[o->object.type];
762}
763
78519831 764int journal_file_move_to_object(JournalFile *f, ObjectType type, uint64_t offset, Object **ret) {
cec736d2
LP		 765 int r;
766 void *t;
767 Object *o;
768 uint64_t s;
769
770 assert(f);
771 assert(ret);
772
db11ac1a 773 /* Objects may only be located at multiple of 64 bit */
202fd896
LP		 774 if (!VALID64(offset)) {
775 log_debug("Attempt to move to object at non-64bit boundary: %" PRIu64, offset);
bd30fdf2 776 return -EBADMSG;
202fd896 777 }
db11ac1a 778
50809d7a 779 /* Object may not be located in the file header */
202fd896
LP		 780 if (offset < le64toh(f->header->header_size)) {
781 log_debug("Attempt to move to object located in file header: %" PRIu64, offset);
50809d7a 782 return -EBADMSG;
202fd896 783 }
50809d7a 784
7a9dabea 785 r = journal_file_move_to(f, type, false, offset, sizeof(ObjectHeader), &t);
cec736d2
LP		 786 if (r < 0)
787 return r;
788
789 o = (Object*) t;
790 s = le64toh(o->object.size);
791
1c69f096
LP		 792 if (s == 0) {
793 log_debug("Attempt to move to uninitialized object: %" PRIu64, offset);
794 return -EBADMSG;
795 }
202fd896
LP		 796 if (s < sizeof(ObjectHeader)) {
797 log_debug("Attempt to move to overly short object: %" PRIu64, offset);
cec736d2 798 return -EBADMSG;
202fd896 799 }
cec736d2 800
202fd896
LP		 801 if (o->object.type <= OBJECT_UNUSED) {
802 log_debug("Attempt to move to object with invalid type: %" PRIu64, offset);
16e9f408 803 return -EBADMSG;
202fd896 804 }
16e9f408 805
202fd896
LP		 806 if (s < minimum_header_size(o)) {
807 log_debug("Attempt to move to truncated object: %" PRIu64, offset);
16e9f408 808 return -EBADMSG;
202fd896 809 }
16e9f408 810
202fd896
LP		 811 if (type > OBJECT_UNUSED && o->object.type != type) {
812 log_debug("Attempt to move to object of unexpected type: %" PRIu64, offset);
cec736d2 813 return -EBADMSG;
202fd896 814 }
cec736d2
LP		 815
816 if (s > sizeof(ObjectHeader)) {
7a9dabea 817 r = journal_file_move_to(f, type, false, offset, s, &t);
cec736d2
LP		 818 if (r < 0)
819 return r;
820
821 o = (Object*) t;
822 }
823
cec736d2
LP		 824 *ret = o;
825 return 0;
826}
827
d98cc1f2 828static uint64_t journal_file_entry_seqnum(JournalFile *f, uint64_t *seqnum) {
cec736d2
LP		 829 uint64_t r;
830
831 assert(f);
c88cc6af 832 assert(f->header);
cec736d2 833
beec0085 834 r = le64toh(f->header->tail_entry_seqnum) + 1;
c2373f84
LP		 835
836 if (seqnum) {
de190aef 837 /* If an external seqnum counter was passed, we update
c2373f84
LP		 838 * both the local and the external one, and set it to
839 * the maximum of both */
840
841 if (*seqnum + 1 > r)
842 r = *seqnum + 1;
843
844 *seqnum = r;
845 }
846
beec0085 847 f->header->tail_entry_seqnum = htole64(r);
cec736d2 848
beec0085
LP		 849 if (f->header->head_entry_seqnum == 0)
850 f->header->head_entry_seqnum = htole64(r);
de190aef 851
cec736d2
LP		 852 return r;
853}
854
78519831 855int journal_file_append_object(JournalFile *f, ObjectType type, uint64_t size, Object **ret, uint64_t *offset) {
cec736d2
LP		 856 int r;
857 uint64_t p;
858 Object *tail, *o;
859 void *t;
860
861 assert(f);
c88cc6af 862 assert(f->header);
d05089d8 863 assert(type > OBJECT_UNUSED && type < _OBJECT_TYPE_MAX);
cec736d2
LP		 864 assert(size >= sizeof(ObjectHeader));
865 assert(offset);
866 assert(ret);
867
26687bf8
OS		 868 r = journal_file_set_online(f);
869 if (r < 0)
870 return r;
871
cec736d2 872 p = le64toh(f->header->tail_object_offset);
cec736d2 873 if (p == 0)
23b0b2b2 874 p = le64toh(f->header->header_size);
cec736d2 875 else {
d05089d8 876 r = journal_file_move_to_object(f, OBJECT_UNUSED, p, &tail);
cec736d2
LP		 877 if (r < 0)
878 return r;
879
880 p += ALIGN64(le64toh(tail->object.size));
881 }
882
883 r = journal_file_allocate(f, p, size);
884 if (r < 0)
885 return r;
886
fcde2389 887 r = journal_file_move_to(f, type, false, p, size, &t);
cec736d2
LP		 888 if (r < 0)
889 return r;
890
891 o = (Object*) t;
892
893 zero(o->object);
de190aef 894 o->object.type = type;
cec736d2
LP		 895 o->object.size = htole64(size);
896
897 f->header->tail_object_offset = htole64(p);
cec736d2
LP		 898 f->header->n_objects = htole64(le64toh(f->header->n_objects) + 1);
899
900 *ret = o;
901 *offset = p;
902
903 return 0;
904}
905
de190aef 906static int journal_file_setup_data_hash_table(JournalFile *f) {
cec736d2
LP		 907 uint64_t s, p;
908 Object *o;
909 int r;
910
911 assert(f);
c88cc6af 912 assert(f->header);
cec736d2 913
070052ab
LP		 914 /* We estimate that we need 1 hash table entry per 768 bytes
915 of journal file and we want to make sure we never get
916 beyond 75% fill level. Calculate the hash table size for
917 the maximum file size based on these metrics. */
4a92baf3 918
dfabe643 919 s = (f->metrics.max_size * 4 / 768 / 3) * sizeof(HashItem);
4a92baf3
LP		 920 if (s < DEFAULT_DATA_HASH_TABLE_SIZE)
921 s = DEFAULT_DATA_HASH_TABLE_SIZE;
922
507f22bd 923 log_debug("Reserving %"PRIu64" entries in hash table.", s / sizeof(HashItem));
4a92baf3 924
de190aef
LP		 925 r = journal_file_append_object(f,
926 OBJECT_DATA_HASH_TABLE,
927 offsetof(Object, hash_table.items) + s,
928 &o, &p);
cec736d2
LP		 929 if (r < 0)
930 return r;
931
29804cc1 932 memzero(o->hash_table.items, s);
cec736d2 933
de190aef
LP		 934 f->header->data_hash_table_offset = htole64(p + offsetof(Object, hash_table.items));
935 f->header->data_hash_table_size = htole64(s);
cec736d2
LP		 936
937 return 0;
938}
939
de190aef 940static int journal_file_setup_field_hash_table(JournalFile *f) {
cec736d2
LP		 941 uint64_t s, p;
942 Object *o;
943 int r;
944
945 assert(f);
c88cc6af 946 assert(f->header);
cec736d2 947
3c1668da
LP		 948 /* We use a fixed size hash table for the fields as this
949 * number should grow very slowly only */
950
de190aef
LP		 951 s = DEFAULT_FIELD_HASH_TABLE_SIZE;
952 r = journal_file_append_object(f,
953 OBJECT_FIELD_HASH_TABLE,
954 offsetof(Object, hash_table.items) + s,
955 &o, &p);
cec736d2
LP		 956 if (r < 0)
957 return r;
958
29804cc1 959 memzero(o->hash_table.items, s);
cec736d2 960
de190aef
LP		 961 f->header->field_hash_table_offset = htole64(p + offsetof(Object, hash_table.items));
962 f->header->field_hash_table_size = htole64(s);
cec736d2
LP		 963
964 return 0;
965}
966
dade37d4 967int journal_file_map_data_hash_table(JournalFile *f) {
cec736d2
LP		 968 uint64_t s, p;
969 void *t;
970 int r;
971
972 assert(f);
c88cc6af 973 assert(f->header);
cec736d2 974
dade37d4
LP		 975 if (f->data_hash_table)
976 return 0;
977
de190aef
LP		 978 p = le64toh(f->header->data_hash_table_offset);
979 s = le64toh(f->header->data_hash_table_size);
cec736d2 980
de190aef 981 r = journal_file_move_to(f,
16e9f408 982 OBJECT_DATA_HASH_TABLE,
fcde2389 983 true,
de190aef
LP		 984 p, s,
985 &t);
cec736d2
LP		 986 if (r < 0)
987 return r;
988
de190aef 989 f->data_hash_table = t;
cec736d2
LP		 990 return 0;
991}
992
dade37d4 993int journal_file_map_field_hash_table(JournalFile *f) {
cec736d2
LP		 994 uint64_t s, p;
995 void *t;
996 int r;
997
998 assert(f);
c88cc6af 999 assert(f->header);
cec736d2 1000
dade37d4
LP		 1001 if (f->field_hash_table)
1002 return 0;
1003
de190aef
LP		 1004 p = le64toh(f->header->field_hash_table_offset);
1005 s = le64toh(f->header->field_hash_table_size);
cec736d2 1006
de190aef 1007 r = journal_file_move_to(f,
16e9f408 1008 OBJECT_FIELD_HASH_TABLE,
fcde2389 1009 true,
de190aef
LP		 1010 p, s,
1011 &t);
cec736d2
LP		 1012 if (r < 0)
1013 return r;
1014
de190aef 1015 f->field_hash_table = t;
cec736d2
LP		 1016 return 0;
1017}
1018
3c1668da
LP		 1019static int journal_file_link_field(
1020 JournalFile *f,
1021 Object *o,
1022 uint64_t offset,
1023 uint64_t hash) {
1024
805d1486 1025 uint64_t p, h, m;
3c1668da
LP		 1026 int r;
1027
1028 assert(f);
c88cc6af 1029 assert(f->header);
90d222c1 1030 assert(f->field_hash_table);
3c1668da
LP		 1031 assert(o);
1032 assert(offset > 0);
1033
1034 if (o->object.type != OBJECT_FIELD)
1035 return -EINVAL;
1036
805d1486
LP		 1037 m = le64toh(f->header->field_hash_table_size) / sizeof(HashItem);
1038 if (m <= 0)
1039 return -EBADMSG;
3c1668da 1040
805d1486 1041 /* This might alter the window we are looking at */
3c1668da
LP		 1042 o->field.next_hash_offset = o->field.head_data_offset = 0;
1043
805d1486 1044 h = hash % m;
3c1668da
LP		 1045 p = le64toh(f->field_hash_table[h].tail_hash_offset);
1046 if (p == 0)
1047 f->field_hash_table[h].head_hash_offset = htole64(offset);
1048 else {
1049 r = journal_file_move_to_object(f, OBJECT_FIELD, p, &o);
1050 if (r < 0)
1051 return r;
1052
1053 o->field.next_hash_offset = htole64(offset);
1054 }
1055
1056 f->field_hash_table[h].tail_hash_offset = htole64(offset);
1057
1058 if (JOURNAL_HEADER_CONTAINS(f->header, n_fields))
1059 f->header->n_fields = htole64(le64toh(f->header->n_fields) + 1);
1060
1061 return 0;
1062}
1063
1064static int journal_file_link_data(
1065 JournalFile *f,
1066 Object *o,
1067 uint64_t offset,
1068 uint64_t hash) {
1069
805d1486 1070 uint64_t p, h, m;
cec736d2
LP		 1071 int r;
1072
1073 assert(f);
c88cc6af 1074 assert(f->header);
90d222c1 1075 assert(f->data_hash_table);
cec736d2
LP		 1076 assert(o);
1077 assert(offset > 0);
b588975f
LP		 1078
1079 if (o->object.type != OBJECT_DATA)
1080 return -EINVAL;
cec736d2 1081
805d1486
LP		 1082 m = le64toh(f->header->data_hash_table_size) / sizeof(HashItem);
1083 if (m <= 0)
1084 return -EBADMSG;
48496df6 1085
805d1486 1086 /* This might alter the window we are looking at */
de190aef
LP		 1087 o->data.next_hash_offset = o->data.next_field_offset = 0;
1088 o->data.entry_offset = o->data.entry_array_offset = 0;
1089 o->data.n_entries = 0;
cec736d2 1090
805d1486 1091 h = hash % m;
8db4213e 1092 p = le64toh(f->data_hash_table[h].tail_hash_offset);
3c1668da 1093 if (p == 0)
cec736d2 1094 /* Only entry in the hash table is easy */
de190aef 1095 f->data_hash_table[h].head_hash_offset = htole64(offset);
3c1668da 1096 else {
48496df6
LP		 1097 /* Move back to the previous data object, to patch in
1098 * pointer */
cec736d2 1099
de190aef 1100 r = journal_file_move_to_object(f, OBJECT_DATA, p, &o);
cec736d2
LP		 1101 if (r < 0)
1102 return r;
1103
de190aef 1104 o->data.next_hash_offset = htole64(offset);
cec736d2
LP		 1105 }
1106
de190aef 1107 f->data_hash_table[h].tail_hash_offset = htole64(offset);
cec736d2 1108
dca6219e
LP		 1109 if (JOURNAL_HEADER_CONTAINS(f->header, n_data))
1110 f->header->n_data = htole64(le64toh(f->header->n_data) + 1);
1111
cec736d2
LP		 1112 return 0;
1113}
1114
3c1668da
LP		 1115int journal_file_find_field_object_with_hash(
1116 JournalFile *f,
1117 const void *field, uint64_t size, uint64_t hash,
1118 Object **ret, uint64_t *offset) {
1119
805d1486 1120 uint64_t p, osize, h, m;
3c1668da
LP		 1121 int r;
1122
1123 assert(f);
c88cc6af 1124 assert(f->header);
3c1668da
LP		 1125 assert(field && size > 0);
1126
dade37d4
LP		 1127 /* If the field hash table is empty, we can't find anything */
1128 if (le64toh(f->header->field_hash_table_size) <= 0)
1129 return 0;
1130
1131 /* Map the field hash table, if it isn't mapped yet. */
1132 r = journal_file_map_field_hash_table(f);
1133 if (r < 0)
1134 return r;
1135
3c1668da
LP		 1136 osize = offsetof(Object, field.payload) + size;
1137
805d1486 1138 m = le64toh(f->header->field_hash_table_size) / sizeof(HashItem);
805d1486 1139 if (m <= 0)
3c1668da
LP		 1140 return -EBADMSG;
1141
805d1486 1142 h = hash % m;
3c1668da
LP		 1143 p = le64toh(f->field_hash_table[h].head_hash_offset);
1144
1145 while (p > 0) {
1146 Object *o;
1147
1148 r = journal_file_move_to_object(f, OBJECT_FIELD, p, &o);
1149 if (r < 0)
1150 return r;
1151
1152 if (le64toh(o->field.hash) == hash &&
1153 le64toh(o->object.size) == osize &&
1154 memcmp(o->field.payload, field, size) == 0) {
1155
1156 if (ret)
1157 *ret = o;
1158 if (offset)
1159 *offset = p;
1160
1161 return 1;
1162 }
1163
1164 p = le64toh(o->field.next_hash_offset);
1165 }
1166
1167 return 0;
1168}
1169
1170int journal_file_find_field_object(
1171 JournalFile *f,
1172 const void *field, uint64_t size,
1173 Object **ret, uint64_t *offset) {
1174
1175 uint64_t hash;
1176
1177 assert(f);
1178 assert(field && size > 0);
1179
1180 hash = hash64(field, size);
1181
1182 return journal_file_find_field_object_with_hash(f,
1183 field, size, hash,
1184 ret, offset);
1185}
1186
de190aef
LP		 1187int journal_file_find_data_object_with_hash(
1188 JournalFile *f,
1189 const void *data, uint64_t size, uint64_t hash,
1190 Object **ret, uint64_t *offset) {
48496df6 1191
805d1486 1192 uint64_t p, osize, h, m;
cec736d2
LP		 1193 int r;
1194
1195 assert(f);
c88cc6af 1196 assert(f->header);
cec736d2
LP		 1197 assert(data || size == 0);
1198
dade37d4
LP		 1199 /* If there's no data hash table, then there's no entry. */
1200 if (le64toh(f->header->data_hash_table_size) <= 0)
1201 return 0;
1202
1203 /* Map the data hash table, if it isn't mapped yet. */
1204 r = journal_file_map_data_hash_table(f);
1205 if (r < 0)
1206 return r;
1207
cec736d2
LP		 1208 osize = offsetof(Object, data.payload) + size;
1209
805d1486
LP		 1210 m = le64toh(f->header->data_hash_table_size) / sizeof(HashItem);
1211 if (m <= 0)
bc85bfee
LP		 1212 return -EBADMSG;
1213
805d1486 1214 h = hash % m;
de190aef 1215 p = le64toh(f->data_hash_table[h].head_hash_offset);
cec736d2 1216
de190aef
LP		 1217 while (p > 0) {
1218 Object *o;
cec736d2 1219
de190aef 1220 r = journal_file_move_to_object(f, OBJECT_DATA, p, &o);
cec736d2
LP		 1221 if (r < 0)
1222 return r;
1223
807e17f0 1224 if (le64toh(o->data.hash) != hash)
85a131e8 1225 goto next;
807e17f0 1226
d89c8fdf 1227 if (o->object.flags & OBJECT_COMPRESSION_MASK) {
3b1a55e1 1228#if defined(HAVE_XZ) || defined(HAVE_LZ4)
fa1c4b51 1229 uint64_t l;
a7f7d1bd 1230 size_t rsize = 0;
cec736d2 1231
807e17f0
LP		 1232 l = le64toh(o->object.size);
1233 if (l <= offsetof(Object, data.payload))
cec736d2
LP		 1234 return -EBADMSG;
1235
807e17f0
LP		 1236 l -= offsetof(Object, data.payload);
1237
d89c8fdf
ZJS		 1238 r = decompress_blob(o->object.flags & OBJECT_COMPRESSION_MASK,
1239 o->data.payload, l, &f->compress_buffer, &f->compress_buffer_size, &rsize, 0);
1240 if (r < 0)
1241 return r;
807e17f0 1242
b785c858 1243 if (rsize == size &&
807e17f0
LP		 1244 memcmp(f->compress_buffer, data, size) == 0) {
1245
1246 if (ret)
1247 *ret = o;
1248
1249 if (offset)
1250 *offset = p;
1251
1252 return 1;
1253 }
3b1a55e1
ZJS		 1254#else
1255 return -EPROTONOSUPPORT;
1256#endif
807e17f0
LP		 1257 } else if (le64toh(o->object.size) == osize &&
1258 memcmp(o->data.payload, data, size) == 0) {
1259
cec736d2
LP		 1260 if (ret)
1261 *ret = o;
1262
1263 if (offset)
1264 *offset = p;
1265
de190aef 1266 return 1;
cec736d2
LP		 1267 }
1268
85a131e8 1269 next:
cec736d2
LP		 1270 p = le64toh(o->data.next_hash_offset);
1271 }
1272
de190aef
LP		 1273 return 0;
1274}
1275
1276int journal_file_find_data_object(
1277 JournalFile *f,
1278 const void *data, uint64_t size,
1279 Object **ret, uint64_t *offset) {
1280
1281 uint64_t hash;
1282
1283 assert(f);
1284 assert(data || size == 0);
1285
1286 hash = hash64(data, size);
1287
1288 return journal_file_find_data_object_with_hash(f,
1289 data, size, hash,
1290 ret, offset);
1291}
1292
3c1668da
LP		 1293static int journal_file_append_field(
1294 JournalFile *f,
1295 const void *field, uint64_t size,
1296 Object **ret, uint64_t *offset) {
1297
1298 uint64_t hash, p;
1299 uint64_t osize;
1300 Object *o;
1301 int r;
1302
1303 assert(f);
1304 assert(field && size > 0);
1305
1306 hash = hash64(field, size);
1307
1308 r = journal_file_find_field_object_with_hash(f, field, size, hash, &o, &p);
1309 if (r < 0)
1310 return r;
1311 else if (r > 0) {
1312
1313 if (ret)
1314 *ret = o;
1315
1316 if (offset)
1317 *offset = p;
1318
1319 return 0;
1320 }
1321
1322 osize = offsetof(Object, field.payload) + size;
1323 r = journal_file_append_object(f, OBJECT_FIELD, osize, &o, &p);
8c92d4bb
LP		 1324 if (r < 0)
1325 return r;
3c1668da
LP		 1326
1327 o->field.hash = htole64(hash);
1328 memcpy(o->field.payload, field, size);
1329
1330 r = journal_file_link_field(f, o, p, hash);
1331 if (r < 0)
1332 return r;
1333
1334 /* The linking might have altered the window, so let's
1335 * refresh our pointer */
1336 r = journal_file_move_to_object(f, OBJECT_FIELD, p, &o);
1337 if (r < 0)
1338 return r;
1339
1340#ifdef HAVE_GCRYPT
1341 r = journal_file_hmac_put_object(f, OBJECT_FIELD, o, p);
1342 if (r < 0)
1343 return r;
1344#endif
1345
1346 if (ret)
1347 *ret = o;
1348
1349 if (offset)
1350 *offset = p;
1351
1352 return 0;
1353}
1354
48496df6
LP		 1355static int journal_file_append_data(
1356 JournalFile *f,
1357 const void *data, uint64_t size,
1358 Object **ret, uint64_t *offset) {
1359
de190aef
LP		 1360 uint64_t hash, p;
1361 uint64_t osize;
1362 Object *o;
d89c8fdf 1363 int r, compression = 0;
3c1668da 1364 const void *eq;
de190aef
LP		 1365
1366 assert(f);
1367 assert(data || size == 0);
1368
1369 hash = hash64(data, size);
1370
1371 r = journal_file_find_data_object_with_hash(f, data, size, hash, &o, &p);
1372 if (r < 0)
1373 return r;
0240c603 1374 if (r > 0) {
de190aef
LP		 1375
1376 if (ret)
1377 *ret = o;
1378
1379 if (offset)
1380 *offset = p;
1381
1382 return 0;
1383 }
1384
1385 osize = offsetof(Object, data.payload) + size;
1386 r = journal_file_append_object(f, OBJECT_DATA, osize, &o, &p);
cec736d2
LP		 1387 if (r < 0)
1388 return r;
1389
cec736d2 1390 o->data.hash = htole64(hash);
807e17f0 1391
d89c8fdf 1392#if defined(HAVE_XZ) || defined(HAVE_LZ4)
d1afbcd2 1393 if (JOURNAL_FILE_COMPRESS(f) && size >= COMPRESSION_SIZE_THRESHOLD) {
a7f7d1bd 1394 size_t rsize = 0;
807e17f0 1395
5d6f46b6 1396 compression = compress_blob(data, size, o->data.payload, size - 1, &rsize);
807e17f0 1397
d1afbcd2 1398 if (compression >= 0) {
807e17f0 1399 o->object.size = htole64(offsetof(Object, data.payload) + rsize);
d89c8fdf 1400 o->object.flags |= compression;
807e17f0 1401
fa1c4b51 1402 log_debug("Compressed data object %"PRIu64" -> %zu using %s",
d89c8fdf 1403 size, rsize, object_compressed_to_string(compression));
d1afbcd2
LP		 1404 } else
1405 /* Compression didn't work, we don't really care why, let's continue without compression */
1406 compression = 0;
807e17f0
LP		 1407 }
1408#endif
1409
75f32f04
ZJS		 1410 if (compression == 0)
1411 memcpy_safe(o->data.payload, data, size);
cec736d2 1412
de190aef 1413 r = journal_file_link_data(f, o, p, hash);
cec736d2
LP		 1414 if (r < 0)
1415 return r;
1416
33685a5a
FB		 1417#ifdef HAVE_GCRYPT
1418 r = journal_file_hmac_put_object(f, OBJECT_DATA, o, p);
1419 if (r < 0)
1420 return r;
1421#endif
1422
48496df6
LP		 1423 /* The linking might have altered the window, so let's
1424 * refresh our pointer */
1425 r = journal_file_move_to_object(f, OBJECT_DATA, p, &o);
1426 if (r < 0)
1427 return r;
1428
08c6f819
SL		 1429 if (!data)
1430 eq = NULL;
1431 else
1432 eq = memchr(data, '=', size);
3c1668da 1433 if (eq && eq > data) {
748db592 1434 Object *fo = NULL;
3c1668da 1435 uint64_t fp;
3c1668da
LP		 1436
1437 /* Create field object ... */
1438 r = journal_file_append_field(f, data, (uint8_t*) eq - (uint8_t*) data, &fo, &fp);
1439 if (r < 0)
1440 return r;
1441
1442 /* ... and link it in. */
1443 o->data.next_field_offset = fo->field.head_data_offset;
1444 fo->field.head_data_offset = le64toh(p);
1445 }
1446
cec736d2
LP		 1447 if (ret)
1448 *ret = o;
1449
1450 if (offset)
de190aef 1451 *offset = p;
cec736d2
LP		 1452
1453 return 0;
1454}
1455
1456uint64_t journal_file_entry_n_items(Object *o) {
1457 assert(o);
b588975f
LP		 1458
1459 if (o->object.type != OBJECT_ENTRY)
1460 return 0;
cec736d2
LP		 1461
1462 return (le64toh(o->object.size) - offsetof(Object, entry.items)) / sizeof(EntryItem);
1463}
1464
0284adc6 1465uint64_t journal_file_entry_array_n_items(Object *o) {
de190aef 1466 assert(o);
b588975f
LP		 1467
1468 if (o->object.type != OBJECT_ENTRY_ARRAY)
1469 return 0;
de190aef
LP		 1470
1471 return (le64toh(o->object.size) - offsetof(Object, entry_array.items)) / sizeof(uint64_t);
1472}
1473
fb9a24b6
LP		 1474uint64_t journal_file_hash_table_n_items(Object *o) {
1475 assert(o);
b588975f
LP		 1476
1477 if (o->object.type != OBJECT_DATA_HASH_TABLE &&
1478 o->object.type != OBJECT_FIELD_HASH_TABLE)
1479 return 0;
fb9a24b6
LP		 1480
1481 return (le64toh(o->object.size) - offsetof(Object, hash_table.items)) / sizeof(HashItem);
1482}
1483
de190aef 1484static int link_entry_into_array(JournalFile *f,
4fd052ae
FC		 1485 le64_t *first,
1486 le64_t *idx,
de190aef 1487 uint64_t p) {
cec736d2 1488 int r;
de190aef
LP		 1489 uint64_t n = 0, ap = 0, q, i, a, hidx;
1490 Object *o;
1491
cec736d2 1492 assert(f);
c88cc6af 1493 assert(f->header);
de190aef
LP		 1494 assert(first);
1495 assert(idx);
1496 assert(p > 0);
cec736d2 1497
de190aef
LP		 1498 a = le64toh(*first);
1499 i = hidx = le64toh(*idx);
1500 while (a > 0) {
1501
1502 r = journal_file_move_to_object(f, OBJECT_ENTRY_ARRAY, a, &o);
1503 if (r < 0)
1504 return r;
cec736d2 1505
de190aef
LP		 1506 n = journal_file_entry_array_n_items(o);
1507 if (i < n) {
1508 o->entry_array.items[i] = htole64(p);
1509 *idx = htole64(hidx + 1);
1510 return 0;
1511 }
cec736d2 1512
de190aef
LP		 1513 i -= n;
1514 ap = a;
1515 a = le64toh(o->entry_array.next_entry_array_offset);
1516 }
1517
1518 if (hidx > n)
1519 n = (hidx+1) * 2;
1520 else
1521 n = n * 2;
1522
1523 if (n < 4)
1524 n = 4;
1525
1526 r = journal_file_append_object(f, OBJECT_ENTRY_ARRAY,
1527 offsetof(Object, entry_array.items) + n * sizeof(uint64_t),
1528 &o, &q);
cec736d2
LP		 1529 if (r < 0)
1530 return r;
1531
feb12d3e 1532#ifdef HAVE_GCRYPT
5996c7c2 1533 r = journal_file_hmac_put_object(f, OBJECT_ENTRY_ARRAY, o, q);
b0af6f41
LP		 1534 if (r < 0)
1535 return r;
feb12d3e 1536#endif
b0af6f41 1537
de190aef 1538 o->entry_array.items[i] = htole64(p);
cec736d2 1539
de190aef 1540 if (ap == 0)
7be3aa17 1541 *first = htole64(q);
cec736d2 1542 else {
de190aef 1543 r = journal_file_move_to_object(f, OBJECT_ENTRY_ARRAY, ap, &o);
cec736d2
LP		 1544 if (r < 0)
1545 return r;
1546
de190aef
LP		 1547 o->entry_array.next_entry_array_offset = htole64(q);
1548 }
cec736d2 1549
2dee23eb
LP		 1550 if (JOURNAL_HEADER_CONTAINS(f->header, n_entry_arrays))
1551 f->header->n_entry_arrays = htole64(le64toh(f->header->n_entry_arrays) + 1);
1552
de190aef
LP		 1553 *idx = htole64(hidx + 1);
1554
1555 return 0;
1556}
cec736d2 1557
de190aef 1558static int link_entry_into_array_plus_one(JournalFile *f,
4fd052ae
FC		 1559 le64_t *extra,
1560 le64_t *first,
1561 le64_t *idx,
de190aef
LP		 1562 uint64_t p) {
1563
1564 int r;
1565
1566 assert(f);
1567 assert(extra);
1568 assert(first);
1569 assert(idx);
1570 assert(p > 0);
1571
1572 if (*idx == 0)
1573 *extra = htole64(p);
1574 else {
4fd052ae 1575 le64_t i;
de190aef 1576
7be3aa17 1577 i = htole64(le64toh(*idx) - 1);
de190aef
LP		 1578 r = link_entry_into_array(f, first, &i, p);
1579 if (r < 0)
1580 return r;
cec736d2
LP		 1581 }
1582
de190aef
LP		 1583 *idx = htole64(le64toh(*idx) + 1);
1584 return 0;
1585}
1586
1587static int journal_file_link_entry_item(JournalFile *f, Object *o, uint64_t offset, uint64_t i) {
1588 uint64_t p;
1589 int r;
1590 assert(f);
1591 assert(o);
1592 assert(offset > 0);
1593
1594 p = le64toh(o->entry.items[i].object_offset);
1595 if (p == 0)
1596 return -EINVAL;
1597
1598 r = journal_file_move_to_object(f, OBJECT_DATA, p, &o);
cec736d2
LP		 1599 if (r < 0)
1600 return r;
1601
de190aef
LP		 1602 return link_entry_into_array_plus_one(f,
1603 &o->data.entry_offset,
1604 &o->data.entry_array_offset,
1605 &o->data.n_entries,
1606 offset);
cec736d2
LP		 1607}
1608
1609static int journal_file_link_entry(JournalFile *f, Object *o, uint64_t offset) {
de190aef 1610 uint64_t n, i;
cec736d2
LP		 1611 int r;
1612
1613 assert(f);
c88cc6af 1614 assert(f->header);
cec736d2
LP		 1615 assert(o);
1616 assert(offset > 0);
b588975f
LP		 1617
1618 if (o->object.type != OBJECT_ENTRY)
1619 return -EINVAL;
cec736d2 1620
b788cc23
LP		 1621 __sync_synchronize();
1622
cec736d2 1623 /* Link up the entry itself */
de190aef
LP		 1624 r = link_entry_into_array(f,
1625 &f->header->entry_array_offset,
1626 &f->header->n_entries,
1627 offset);
1628 if (r < 0)
1629 return r;
cec736d2 1630
507f22bd 1631 /* log_debug("=> %s seqnr=%"PRIu64" n_entries=%"PRIu64, f->path, o->entry.seqnum, f->header->n_entries); */
cec736d2 1632
de190aef 1633 if (f->header->head_entry_realtime == 0)
0ac38b70 1634 f->header->head_entry_realtime = o->entry.realtime;
cec736d2 1635
0ac38b70 1636 f->header->tail_entry_realtime = o->entry.realtime;
de190aef
LP		 1637 f->header->tail_entry_monotonic = o->entry.monotonic;
1638
1639 f->tail_entry_monotonic_valid = true;
cec736d2
LP		 1640
1641 /* Link up the items */
1642 n = journal_file_entry_n_items(o);
1643 for (i = 0; i < n; i++) {
1644 r = journal_file_link_entry_item(f, o, offset, i);
1645 if (r < 0)
1646 return r;
1647 }
1648
cec736d2
LP		 1649 return 0;
1650}
1651
1652static int journal_file_append_entry_internal(
1653 JournalFile *f,
1654 const dual_timestamp *ts,
1655 uint64_t xor_hash,
1656 const EntryItem items[], unsigned n_items,
de190aef 1657 uint64_t *seqnum,
cec736d2
LP		 1658 Object **ret, uint64_t *offset) {
1659 uint64_t np;
1660 uint64_t osize;
1661 Object *o;
1662 int r;
1663
1664 assert(f);
c88cc6af 1665 assert(f->header);
cec736d2 1666 assert(items || n_items == 0);
de190aef 1667 assert(ts);
cec736d2
LP		 1668
1669 osize = offsetof(Object, entry.items) + (n_items * sizeof(EntryItem));
1670
de190aef 1671 r = journal_file_append_object(f, OBJECT_ENTRY, osize, &o, &np);
cec736d2
LP		 1672 if (r < 0)
1673 return r;
1674
d98cc1f2 1675 o->entry.seqnum = htole64(journal_file_entry_seqnum(f, seqnum));
75f32f04 1676 memcpy_safe(o->entry.items, items, n_items * sizeof(EntryItem));
de190aef
LP		 1677 o->entry.realtime = htole64(ts->realtime);
1678 o->entry.monotonic = htole64(ts->monotonic);
cec736d2
LP		 1679 o->entry.xor_hash = htole64(xor_hash);
1680 o->entry.boot_id = f->header->boot_id;
1681
feb12d3e 1682#ifdef HAVE_GCRYPT
5996c7c2 1683 r = journal_file_hmac_put_object(f, OBJECT_ENTRY, o, np);
b0af6f41
LP		 1684 if (r < 0)
1685 return r;
feb12d3e 1686#endif
b0af6f41 1687
cec736d2
LP		 1688 r = journal_file_link_entry(f, o, np);
1689 if (r < 0)
1690 return r;
1691
1692 if (ret)
1693 *ret = o;
1694
1695 if (offset)
1696 *offset = np;
1697
1698 return 0;
1699}
1700
cf244689 1701void journal_file_post_change(JournalFile *f) {
50f20cfd
LP		 1702 assert(f);
1703
1704 /* inotify() does not receive IN_MODIFY events from file
1705 * accesses done via mmap(). After each access we hence
1706 * trigger IN_MODIFY by truncating the journal file to its
1707 * current size which triggers IN_MODIFY. */
1708
bc85bfee
LP		 1709 __sync_synchronize();
1710
50f20cfd 1711 if (ftruncate(f->fd, f->last_stat.st_size) < 0)
e167d7fd 1712 log_debug_errno(errno, "Failed to truncate file to its own size: %m");
50f20cfd
LP		 1713}
1714
7a24f3bf
VC		 1715static int post_change_thunk(sd_event_source *timer, uint64_t usec, void *userdata) {
1716 assert(userdata);
1717
1718 journal_file_post_change(userdata);
1719
1720 return 1;
1721}
1722
1723static void schedule_post_change(JournalFile *f) {
1724 sd_event_source *timer;
1725 int enabled, r;
1726 uint64_t now;
1727
1728 assert(f);
1729 assert(f->post_change_timer);
1730
1731 timer = f->post_change_timer;
1732
1733 r = sd_event_source_get_enabled(timer, &enabled);
1734 if (r < 0) {
e167d7fd
LP		 1735 log_debug_errno(r, "Failed to get ftruncate timer state: %m");
1736 goto fail;
7a24f3bf
VC		 1737 }
1738
1739 if (enabled == SD_EVENT_ONESHOT)
1740 return;
1741
1742 r = sd_event_now(sd_event_source_get_event(timer), CLOCK_MONOTONIC, &now);
1743 if (r < 0) {
e167d7fd
LP		 1744 log_debug_errno(r, "Failed to get clock's now for scheduling ftruncate: %m");
1745 goto fail;
7a24f3bf
VC		 1746 }
1747
1748 r = sd_event_source_set_time(timer, now+f->post_change_timer_period);
1749 if (r < 0) {
e167d7fd
LP		 1750 log_debug_errno(r, "Failed to set time for scheduling ftruncate: %m");
1751 goto fail;
7a24f3bf
VC		 1752 }
1753
1754 r = sd_event_source_set_enabled(timer, SD_EVENT_ONESHOT);
1755 if (r < 0) {
e167d7fd
LP		 1756 log_debug_errno(r, "Failed to enable scheduled ftruncate: %m");
1757 goto fail;
7a24f3bf 1758 }
e167d7fd
LP		 1759
1760 return;
1761
1762fail:
1763 /* On failure, let's simply post the change immediately. */
1764 journal_file_post_change(f);
7a24f3bf
VC		 1765}
1766
1767/* Enable coalesced change posting in a timer on the provided sd_event instance */
1768int journal_file_enable_post_change_timer(JournalFile *f, sd_event *e, usec_t t) {
1769 _cleanup_(sd_event_source_unrefp) sd_event_source *timer = NULL;
1770 int r;
1771
1772 assert(f);
1773 assert_return(!f->post_change_timer, -EINVAL);
1774 assert(e);
1775 assert(t);
1776
1777 r = sd_event_add_time(e, &timer, CLOCK_MONOTONIC, 0, 0, post_change_thunk, f);
1778 if (r < 0)
1779 return r;
1780
1781 r = sd_event_source_set_enabled(timer, SD_EVENT_OFF);
1782 if (r < 0)
1783 return r;
1784
1785 f->post_change_timer = timer;
1786 timer = NULL;
1787 f->post_change_timer_period = t;
1788
1789 return r;
1790}
1791
1f2da9ec
LP		 1792static int entry_item_cmp(const void *_a, const void *_b) {
1793 const EntryItem *a = _a, *b = _b;
1794
1795 if (le64toh(a->object_offset) < le64toh(b->object_offset))
1796 return -1;
1797 if (le64toh(a->object_offset) > le64toh(b->object_offset))
1798 return 1;
1799 return 0;
1800}
1801
de190aef 1802int 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		 1803 unsigned i;
1804 EntryItem *items;
1805 int r;
1806 uint64_t xor_hash = 0;
de190aef 1807 struct dual_timestamp _ts;
cec736d2
LP		 1808
1809 assert(f);
c88cc6af 1810 assert(f->header);
cec736d2
LP		 1811 assert(iovec || n_iovec == 0);
1812
de190aef
LP		 1813 if (!ts) {
1814 dual_timestamp_get(&_ts);
1815 ts = &_ts;
1816 }
1817
feb12d3e 1818#ifdef HAVE_GCRYPT
7560fffc
LP		 1819 r = journal_file_maybe_append_tag(f, ts->realtime);
1820 if (r < 0)
1821 return r;
feb12d3e 1822#endif
7560fffc 1823
64825d3c 1824 /* alloca() can't take 0, hence let's allocate at least one */
9607d947 1825 items = alloca(sizeof(EntryItem) * MAX(1u, n_iovec));
cec736d2
LP		 1826
1827 for (i = 0; i < n_iovec; i++) {
1828 uint64_t p;
1829 Object *o;
1830
1831 r = journal_file_append_data(f, iovec[i].iov_base, iovec[i].iov_len, &o, &p);
1832 if (r < 0)
cf244689 1833 return r;
cec736d2
LP		 1834
1835 xor_hash ^= le64toh(o->data.hash);
1836 items[i].object_offset = htole64(p);
de7b95cd 1837 items[i].hash = o->data.hash;
cec736d2
LP		 1838 }
1839
1f2da9ec
LP		 1840 /* Order by the position on disk, in order to improve seek
1841 * times for rotating media. */
7ff7394d 1842 qsort_safe(items, n_iovec, sizeof(EntryItem), entry_item_cmp);
1f2da9ec 1843
de190aef 1844 r = journal_file_append_entry_internal(f, ts, xor_hash, items, n_iovec, seqnum, ret, offset);
cec736d2 1845
fa6ac760
LP		 1846 /* If the memory mapping triggered a SIGBUS then we return an
1847 * IO error and ignore the error code passed down to us, since
1848 * it is very likely just an effect of a nullified replacement
1849 * mapping page */
1850
1851 if (mmap_cache_got_sigbus(f->mmap, f->fd))
1852 r = -EIO;
1853
7a24f3bf
VC		 1854 if (f->post_change_timer)
1855 schedule_post_change(f);
1856 else
1857 journal_file_post_change(f);
50f20cfd 1858
cec736d2
LP		 1859 return r;
1860}
1861
a4bcff5b 1862typedef struct ChainCacheItem {
fb099c8d 1863 uint64_t first; /* the array at the beginning of the chain */
a4bcff5b
LP		 1864 uint64_t array; /* the cached array */
1865 uint64_t begin; /* the first item in the cached array */
1866 uint64_t total; /* the total number of items in all arrays before this one in the chain */
f268980d 1867 uint64_t last_index; /* the last index we looked at, to optimize locality when bisecting */
a4bcff5b
LP		 1868} ChainCacheItem;
1869
1870static void chain_cache_put(
4743015d 1871 OrderedHashmap *h,
a4bcff5b
LP		 1872 ChainCacheItem *ci,
1873 uint64_t first,
1874 uint64_t array,
1875 uint64_t begin,
f268980d
LP		 1876 uint64_t total,
1877 uint64_t last_index) {
a4bcff5b
LP		 1878
1879 if (!ci) {
34741aa3
LP		 1880 /* If the chain item to cache for this chain is the
1881 * first one it's not worth caching anything */
1882 if (array == first)
1883 return;
1884
29433089 1885 if (ordered_hashmap_size(h) >= CHAIN_CACHE_MAX) {
4743015d 1886 ci = ordered_hashmap_steal_first(h);
29433089
LP		 1887 assert(ci);
1888 } else {
a4bcff5b
LP		 1889 ci = new(ChainCacheItem, 1);
1890 if (!ci)
1891 return;
1892 }
1893
1894 ci->first = first;
1895
4743015d 1896 if (ordered_hashmap_put(h, &ci->first, ci) < 0) {
a4bcff5b
LP		 1897 free(ci);
1898 return;
1899 }
1900 } else
1901 assert(ci->first == first);
1902
1903 ci->array = array;
1904 ci->begin = begin;
1905 ci->total = total;
f268980d 1906 ci->last_index = last_index;
a4bcff5b
LP		 1907}
1908
f268980d
LP		 1909static int generic_array_get(
1910 JournalFile *f,
1911 uint64_t first,
1912 uint64_t i,
1913 Object **ret, uint64_t *offset) {
de190aef 1914
cec736d2 1915 Object *o;
a4bcff5b 1916 uint64_t p = 0, a, t = 0;
cec736d2 1917 int r;
a4bcff5b 1918 ChainCacheItem *ci;
cec736d2
LP		 1919
1920 assert(f);
1921
de190aef 1922 a = first;
a4bcff5b
LP		 1923
1924 /* Try the chain cache first */
4743015d 1925 ci = ordered_hashmap_get(f->chain_cache, &first);
a4bcff5b
LP		 1926 if (ci && i > ci->total) {
1927 a = ci->array;
1928 i -= ci->total;
1929 t = ci->total;
1930 }
1931
de190aef 1932 while (a > 0) {
a4bcff5b 1933 uint64_t k;
cec736d2 1934
de190aef
LP		 1935 r = journal_file_move_to_object(f, OBJECT_ENTRY_ARRAY, a, &o);
1936 if (r < 0)
1937 return r;
cec736d2 1938
a4bcff5b
LP		 1939 k = journal_file_entry_array_n_items(o);
1940 if (i < k) {
de190aef 1941 p = le64toh(o->entry_array.items[i]);
a4bcff5b 1942 goto found;
cec736d2
LP		 1943 }
1944
a4bcff5b
LP		 1945 i -= k;
1946 t += k;
de190aef
LP		 1947 a = le64toh(o->entry_array.next_entry_array_offset);
1948 }
1949
a4bcff5b
LP		 1950 return 0;
1951
1952found:
1953 /* Let's cache this item for the next invocation */
af13a6b0 1954 chain_cache_put(f->chain_cache, ci, first, a, le64toh(o->entry_array.items[0]), t, i);
de190aef
LP		 1955
1956 r = journal_file_move_to_object(f, OBJECT_ENTRY, p, &o);
1957 if (r < 0)
1958 return r;
1959
1960 if (ret)
1961 *ret = o;
1962
1963 if (offset)
1964 *offset = p;
1965
1966 return 1;
1967}
1968
f268980d
LP		 1969static int generic_array_get_plus_one(
1970 JournalFile *f,
1971 uint64_t extra,
1972 uint64_t first,
1973 uint64_t i,
1974 Object **ret, uint64_t *offset) {
de190aef
LP		 1975
1976 Object *o;
1977
1978 assert(f);
1979
1980 if (i == 0) {
1981 int r;
1982
1983 r = journal_file_move_to_object(f, OBJECT_ENTRY, extra, &o);
cec736d2
LP		 1984 if (r < 0)
1985 return r;
1986
de190aef
LP		 1987 if (ret)
1988 *ret = o;
cec736d2 1989
de190aef
LP		 1990 if (offset)
1991 *offset = extra;
cec736d2 1992
de190aef 1993 return 1;
cec736d2
LP		 1994 }
1995
de190aef
LP		 1996 return generic_array_get(f, first, i-1, ret, offset);
1997}
cec736d2 1998
de190aef
LP		 1999enum {
2000 TEST_FOUND,
2001 TEST_LEFT,
2002 TEST_RIGHT
2003};
cec736d2 2004
f268980d
LP		 2005static int generic_array_bisect(
2006 JournalFile *f,
2007 uint64_t first,
2008 uint64_t n,
2009 uint64_t needle,
2010 int (*test_object)(JournalFile *f, uint64_t p, uint64_t needle),
2011 direction_t direction,
2012 Object **ret,
2013 uint64_t *offset,
2014 uint64_t *idx) {
2015
2016 uint64_t a, p, t = 0, i = 0, last_p = 0, last_index = (uint64_t) -1;
de190aef
LP		 2017 bool subtract_one = false;
2018 Object *o, *array = NULL;
2019 int r;
a4bcff5b 2020 ChainCacheItem *ci;
cec736d2 2021
de190aef
LP		 2022 assert(f);
2023 assert(test_object);
cec736d2 2024
a4bcff5b 2025 /* Start with the first array in the chain */
de190aef 2026 a = first;
a4bcff5b 2027
4743015d 2028 ci = ordered_hashmap_get(f->chain_cache, &first);
a4bcff5b
LP		 2029 if (ci && n > ci->total) {
2030 /* Ah, we have iterated this bisection array chain
2031 * previously! Let's see if we can skip ahead in the
2032 * chain, as far as the last time. But we can't jump
2033 * backwards in the chain, so let's check that
2034 * first. */
2035
2036 r = test_object(f, ci->begin, needle);
2037 if (r < 0)
2038 return r;
2039
2040 if (r == TEST_LEFT) {
f268980d 2041 /* OK, what we are looking for is right of the
a4bcff5b
LP		 2042 * begin of this EntryArray, so let's jump
2043 * straight to previously cached array in the
2044 * chain */
2045
2046 a = ci->array;
2047 n -= ci->total;
2048 t = ci->total;
f268980d 2049 last_index = ci->last_index;
a4bcff5b
LP		 2050 }
2051 }
2052
de190aef
LP		 2053 while (a > 0) {
2054 uint64_t left, right, k, lp;
2055
2056 r = journal_file_move_to_object(f, OBJECT_ENTRY_ARRAY, a, &array);
cec736d2
LP		 2057 if (r < 0)
2058 return r;
2059
de190aef
LP		 2060 k = journal_file_entry_array_n_items(array);
2061 right = MIN(k, n);
2062 if (right <= 0)
2063 return 0;
cec736d2 2064
de190aef
LP		 2065 i = right - 1;
2066 lp = p = le64toh(array->entry_array.items[i]);
2067 if (p <= 0)
bee6a291
LP		 2068 r = -EBADMSG;
2069 else
2070 r = test_object(f, p, needle);
2071 if (r == -EBADMSG) {
2072 log_debug_errno(r, "Encountered invalid entry while bisecting, cutting algorithm short. (1)");
2073 n = i;
2074 continue;
2075 }
de190aef
LP		 2076 if (r < 0)
2077 return r;
cec736d2 2078
de190aef
LP		 2079 if (r == TEST_FOUND)
2080 r = direction == DIRECTION_DOWN ? TEST_RIGHT : TEST_LEFT;
2081
2082 if (r == TEST_RIGHT) {
2083 left = 0;
2084 right -= 1;
f268980d
LP		 2085
2086 if (last_index != (uint64_t) -1) {
2087 assert(last_index <= right);
2088
2089 /* If we cached the last index we
2090 * looked at, let's try to not to jump
2091 * too wildly around and see if we can
2092 * limit the range to look at early to
2093 * the immediate neighbors of the last
2094 * index we looked at. */
2095
2096 if (last_index > 0) {
2097 uint64_t x = last_index - 1;
2098
2099 p = le64toh(array->entry_array.items[x]);
2100 if (p <= 0)
2101 return -EBADMSG;
2102
2103 r = test_object(f, p, needle);
2104 if (r < 0)
2105 return r;
2106
2107 if (r == TEST_FOUND)
2108 r = direction == DIRECTION_DOWN ? TEST_RIGHT : TEST_LEFT;
2109
2110 if (r == TEST_RIGHT)
2111 right = x;
2112 else
2113 left = x + 1;
2114 }
2115
2116 if (last_index < right) {
2117 uint64_t y = last_index + 1;
2118
2119 p = le64toh(array->entry_array.items[y]);
2120 if (p <= 0)
2121 return -EBADMSG;
2122
2123 r = test_object(f, p, needle);
2124 if (r < 0)
2125 return r;
2126
2127 if (r == TEST_FOUND)
2128 r = direction == DIRECTION_DOWN ? TEST_RIGHT : TEST_LEFT;
2129
2130 if (r == TEST_RIGHT)
2131 right = y;
2132 else
2133 left = y + 1;
2134 }
f268980d
LP		 2135 }
2136
de190aef
LP		 2137 for (;;) {
2138 if (left == right) {
2139 if (direction == DIRECTION_UP)
2140 subtract_one = true;
2141
2142 i = left;
2143 goto found;
2144 }
2145
2146 assert(left < right);
de190aef 2147 i = (left + right) / 2;
f268980d 2148
de190aef
LP		 2149 p = le64toh(array->entry_array.items[i]);
2150 if (p <= 0)
bee6a291
LP		 2151 r = -EBADMSG;
2152 else
2153 r = test_object(f, p, needle);
2154 if (r == -EBADMSG) {
2155 log_debug_errno(r, "Encountered invalid entry while bisecting, cutting algorithm short. (2)");
2156 right = n = i;
2157 continue;
2158 }
de190aef
LP		 2159 if (r < 0)
2160 return r;
cec736d2 2161
de190aef
LP		 2162 if (r == TEST_FOUND)
2163 r = direction == DIRECTION_DOWN ? TEST_RIGHT : TEST_LEFT;
2164
2165 if (r == TEST_RIGHT)
2166 right = i;
2167 else
2168 left = i + 1;
2169 }
2170 }
2171
2173cbf8 2172 if (k >= n) {
cbdca852
LP		 2173 if (direction == DIRECTION_UP) {
2174 i = n;
2175 subtract_one = true;
2176 goto found;
2177 }
2178
cec736d2 2179 return 0;
cbdca852 2180 }
cec736d2 2181
de190aef
LP		 2182 last_p = lp;
2183
2184 n -= k;
2185 t += k;
f268980d 2186 last_index = (uint64_t) -1;
de190aef 2187 a = le64toh(array->entry_array.next_entry_array_offset);
cec736d2
LP		 2188 }
2189
2190 return 0;
de190aef
LP		 2191
2192found:
2193 if (subtract_one && t == 0 && i == 0)
2194 return 0;
2195
a4bcff5b 2196 /* Let's cache this item for the next invocation */
af13a6b0 2197 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 2198
de190aef
LP		 2199 if (subtract_one && i == 0)
2200 p = last_p;
2201 else if (subtract_one)
2202 p = le64toh(array->entry_array.items[i-1]);
2203 else
2204 p = le64toh(array->entry_array.items[i]);
2205
2206 r = journal_file_move_to_object(f, OBJECT_ENTRY, p, &o);
2207 if (r < 0)
2208 return r;
2209
2210 if (ret)
2211 *ret = o;
2212
2213 if (offset)
2214 *offset = p;
2215
2216 if (idx)
cbdca852 2217 *idx = t + i + (subtract_one ? -1 : 0);
de190aef
LP		 2218
2219 return 1;
cec736d2
LP		 2220}
2221
f268980d
LP		 2222static int generic_array_bisect_plus_one(
2223 JournalFile *f,
2224 uint64_t extra,
2225 uint64_t first,
2226 uint64_t n,
2227 uint64_t needle,
2228 int (*test_object)(JournalFile *f, uint64_t p, uint64_t needle),
2229 direction_t direction,
2230 Object **ret,
2231 uint64_t *offset,
2232 uint64_t *idx) {
de190aef 2233
cec736d2 2234 int r;
cbdca852
LP		 2235 bool step_back = false;
2236 Object *o;
cec736d2
LP		 2237
2238 assert(f);
de190aef 2239 assert(test_object);
cec736d2 2240
de190aef
LP		 2241 if (n <= 0)
2242 return 0;
cec736d2 2243
de190aef
LP		 2244 /* This bisects the array in object 'first', but first checks
2245 * an extra */
de190aef
LP		 2246 r = test_object(f, extra, needle);
2247 if (r < 0)
2248 return r;
a536e261
LP		 2249
2250 if (r == TEST_FOUND)
2251 r = direction == DIRECTION_DOWN ? TEST_RIGHT : TEST_LEFT;
2252
cbdca852
LP		 2253 /* if we are looking with DIRECTION_UP then we need to first
2254 see if in the actual array there is a matching entry, and
2255 return the last one of that. But if there isn't any we need
2256 to return this one. Hence remember this, and return it
2257 below. */
2258 if (r == TEST_LEFT)
2259 step_back = direction == DIRECTION_UP;
de190aef 2260
cbdca852
LP		 2261 if (r == TEST_RIGHT) {
2262 if (direction == DIRECTION_DOWN)
2263 goto found;
2264 else
2265 return 0;
a536e261 2266 }
cec736d2 2267
de190aef
LP		 2268 r = generic_array_bisect(f, first, n-1, needle, test_object, direction, ret, offset, idx);
2269
cbdca852
LP		 2270 if (r == 0 && step_back)
2271 goto found;
2272
ecf68b1d 2273 if (r > 0 && idx)
313cefa1 2274 (*idx)++;
de190aef
LP		 2275
2276 return r;
cbdca852
LP		 2277
2278found:
2279 r = journal_file_move_to_object(f, OBJECT_ENTRY, extra, &o);
2280 if (r < 0)
2281 return r;
2282
2283 if (ret)
2284 *ret = o;
2285
2286 if (offset)
2287 *offset = extra;
2288
2289 if (idx)
2290 *idx = 0;
2291
2292 return 1;
2293}
2294
44a6b1b6 2295_pure_ static int test_object_offset(JournalFile *f, uint64_t p, uint64_t needle) {
cbdca852
LP		 2296 assert(f);
2297 assert(p > 0);
2298
2299 if (p == needle)
2300 return TEST_FOUND;
2301 else if (p < needle)
2302 return TEST_LEFT;
2303 else
2304 return TEST_RIGHT;
2305}
2306
de190aef
LP		 2307static int test_object_seqnum(JournalFile *f, uint64_t p, uint64_t needle) {
2308 Object *o;
2309 int r;
2310
2311 assert(f);
2312 assert(p > 0);
2313
2314 r = journal_file_move_to_object(f, OBJECT_ENTRY, p, &o);
cec736d2
LP		 2315 if (r < 0)
2316 return r;
2317
de190aef
LP		 2318 if (le64toh(o->entry.seqnum) == needle)
2319 return TEST_FOUND;
2320 else if (le64toh(o->entry.seqnum) < needle)
2321 return TEST_LEFT;
2322 else
2323 return TEST_RIGHT;
2324}
cec736d2 2325
de190aef
LP		 2326int journal_file_move_to_entry_by_seqnum(
2327 JournalFile *f,
2328 uint64_t seqnum,
2329 direction_t direction,
2330 Object **ret,
2331 uint64_t *offset) {
c88cc6af
VC		 2332 assert(f);
2333 assert(f->header);
de190aef
LP		 2334
2335 return generic_array_bisect(f,
2336 le64toh(f->header->entry_array_offset),
2337 le64toh(f->header->n_entries),
2338 seqnum,
2339 test_object_seqnum,
2340 direction,
2341 ret, offset, NULL);
2342}
cec736d2 2343
de190aef
LP		 2344static int test_object_realtime(JournalFile *f, uint64_t p, uint64_t needle) {
2345 Object *o;
2346 int r;
2347
2348 assert(f);
2349 assert(p > 0);
2350
2351 r = journal_file_move_to_object(f, OBJECT_ENTRY, p, &o);
2352 if (r < 0)
2353 return r;
2354
2355 if (le64toh(o->entry.realtime) == needle)
2356 return TEST_FOUND;
2357 else if (le64toh(o->entry.realtime) < needle)
2358 return TEST_LEFT;
2359 else
2360 return TEST_RIGHT;
cec736d2
LP		 2361}
2362
de190aef
LP		 2363int journal_file_move_to_entry_by_realtime(
2364 JournalFile *f,
2365 uint64_t realtime,
2366 direction_t direction,
2367 Object **ret,
2368 uint64_t *offset) {
c88cc6af
VC		 2369 assert(f);
2370 assert(f->header);
de190aef
LP		 2371
2372 return generic_array_bisect(f,
2373 le64toh(f->header->entry_array_offset),
2374 le64toh(f->header->n_entries),
2375 realtime,
2376 test_object_realtime,
2377 direction,
2378 ret, offset, NULL);
2379}
2380
2381static int test_object_monotonic(JournalFile *f, uint64_t p, uint64_t needle) {
2382 Object *o;
2383 int r;
2384
2385 assert(f);
2386 assert(p > 0);
2387
2388 r = journal_file_move_to_object(f, OBJECT_ENTRY, p, &o);
2389 if (r < 0)
2390 return r;
2391
2392 if (le64toh(o->entry.monotonic) == needle)
2393 return TEST_FOUND;
2394 else if (le64toh(o->entry.monotonic) < needle)
2395 return TEST_LEFT;
2396 else
2397 return TEST_RIGHT;
2398}
2399
2a560338 2400static int find_data_object_by_boot_id(
47838ab3
ZJS		 2401 JournalFile *f,
2402 sd_id128_t boot_id,
2403 Object **o,
2404 uint64_t *b) {
2a560338 2405
47838ab3
ZJS		 2406 char t[sizeof("_BOOT_ID=")-1 + 32 + 1] = "_BOOT_ID=";
2407
2408 sd_id128_to_string(boot_id, t + 9);
2409 return journal_file_find_data_object(f, t, sizeof(t) - 1, o, b);
2410}
2411
de190aef
LP		 2412int journal_file_move_to_entry_by_monotonic(
2413 JournalFile *f,
2414 sd_id128_t boot_id,
2415 uint64_t monotonic,
2416 direction_t direction,
2417 Object **ret,
2418 uint64_t *offset) {
2419
de190aef
LP		 2420 Object *o;
2421 int r;
2422
cbdca852 2423 assert(f);
de190aef 2424
47838ab3 2425 r = find_data_object_by_boot_id(f, boot_id, &o, NULL);
de190aef
LP		 2426 if (r < 0)
2427 return r;
cbdca852 2428 if (r == 0)
de190aef
LP		 2429 return -ENOENT;
2430
2431 return generic_array_bisect_plus_one(f,
2432 le64toh(o->data.entry_offset),
2433 le64toh(o->data.entry_array_offset),
2434 le64toh(o->data.n_entries),
2435 monotonic,
2436 test_object_monotonic,
2437 direction,
2438 ret, offset, NULL);
2439}
2440
1fc605b0 2441void journal_file_reset_location(JournalFile *f) {
6573ef05 2442 f->location_type = LOCATION_HEAD;
1fc605b0 2443 f->current_offset = 0;
6573ef05
MS		 2444 f->current_seqnum = 0;
2445 f->current_realtime = 0;
2446 f->current_monotonic = 0;
2447 zero(f->current_boot_id);
2448 f->current_xor_hash = 0;
2449}
2450
950c07d4 2451void journal_file_save_location(JournalFile *f, Object *o, uint64_t offset) {
6573ef05
MS		 2452 f->location_type = LOCATION_SEEK;
2453 f->current_offset = offset;
2454 f->current_seqnum = le64toh(o->entry.seqnum);
2455 f->current_realtime = le64toh(o->entry.realtime);
2456 f->current_monotonic = le64toh(o->entry.monotonic);
2457 f->current_boot_id = o->entry.boot_id;
2458 f->current_xor_hash = le64toh(o->entry.xor_hash);
1fc605b0
MS		 2459}
2460
d8ae66d7
MS		 2461int journal_file_compare_locations(JournalFile *af, JournalFile *bf) {
2462 assert(af);
c88cc6af 2463 assert(af->header);
d8ae66d7 2464 assert(bf);
c88cc6af 2465 assert(bf->header);
d8ae66d7
MS		 2466 assert(af->location_type == LOCATION_SEEK);
2467 assert(bf->location_type == LOCATION_SEEK);
2468
2469 /* If contents and timestamps match, these entries are
2470 * identical, even if the seqnum does not match */
2471 if (sd_id128_equal(af->current_boot_id, bf->current_boot_id) &&
2472 af->current_monotonic == bf->current_monotonic &&
2473 af->current_realtime == bf->current_realtime &&
2474 af->current_xor_hash == bf->current_xor_hash)
2475 return 0;
2476
2477 if (sd_id128_equal(af->header->seqnum_id, bf->header->seqnum_id)) {
2478
2479 /* If this is from the same seqnum source, compare
2480 * seqnums */
2481 if (af->current_seqnum < bf->current_seqnum)
2482 return -1;
2483 if (af->current_seqnum > bf->current_seqnum)
2484 return 1;
2485
2486 /* Wow! This is weird, different data but the same
2487 * seqnums? Something is borked, but let's make the
2488 * best of it and compare by time. */
2489 }
2490
2491 if (sd_id128_equal(af->current_boot_id, bf->current_boot_id)) {
2492
2493 /* If the boot id matches, compare monotonic time */
2494 if (af->current_monotonic < bf->current_monotonic)
2495 return -1;
2496 if (af->current_monotonic > bf->current_monotonic)
2497 return 1;
2498 }
2499
2500 /* Otherwise, compare UTC time */
2501 if (af->current_realtime < bf->current_realtime)
2502 return -1;
2503 if (af->current_realtime > bf->current_realtime)
2504 return 1;
2505
2506 /* Finally, compare by contents */
2507 if (af->current_xor_hash < bf->current_xor_hash)
2508 return -1;
2509 if (af->current_xor_hash > bf->current_xor_hash)
2510 return 1;
2511
2512 return 0;
2513}
2514
aa598ba5
LP		 2515static int bump_array_index(uint64_t *i, direction_t direction, uint64_t n) {
2516
2517 /* Increase or decrease the specified index, in the right direction. */
2518
2519 if (direction == DIRECTION_DOWN) {
2520 if (*i >= n - 1)
2521 return 0;
2522
2523 (*i) ++;
2524 } else {
2525 if (*i <= 0)
2526 return 0;
2527
2528 (*i) --;
2529 }
2530
2531 return 1;
2532}
2533
b6da4ed0
LP		 2534static bool check_properly_ordered(uint64_t new_offset, uint64_t old_offset, direction_t direction) {
2535
2536 /* Consider it an error if any of the two offsets is uninitialized */
2537 if (old_offset == 0 || new_offset == 0)
2538 return false;
2539
2540 /* If we go down, the new offset must be larger than the old one. */
2541 return direction == DIRECTION_DOWN ?
2542 new_offset > old_offset :
2543 new_offset < old_offset;
2544}
2545
de190aef
LP		 2546int journal_file_next_entry(
2547 JournalFile *f,
f534928a 2548 uint64_t p,
de190aef
LP		 2549 direction_t direction,
2550 Object **ret, uint64_t *offset) {
2551
fb099c8d 2552 uint64_t i, n, ofs;
cec736d2
LP		 2553 int r;
2554
2555 assert(f);
c88cc6af 2556 assert(f->header);
de190aef
LP		 2557
2558 n = le64toh(f->header->n_entries);
2559 if (n <= 0)
2560 return 0;
cec736d2 2561
f534928a 2562 if (p == 0)
de190aef 2563 i = direction == DIRECTION_DOWN ? 0 : n - 1;
cec736d2 2564 else {
de190aef
LP		 2565 r = generic_array_bisect(f,
2566 le64toh(f->header->entry_array_offset),
2567 le64toh(f->header->n_entries),
2568 p,
2569 test_object_offset,
2570 DIRECTION_DOWN,
2571 NULL, NULL,
2572 &i);
2573 if (r <= 0)
2574 return r;
2575
aa598ba5
LP		 2576 r = bump_array_index(&i, direction, n);
2577 if (r <= 0)
2578 return r;
cec736d2
LP		 2579 }
2580
de190aef 2581 /* And jump to it */
989793d3
LP		 2582 for (;;) {
2583 r = generic_array_get(f,
2584 le64toh(f->header->entry_array_offset),
2585 i,
2586 ret, &ofs);
2587 if (r > 0)
2588 break;
2589 if (r != -EBADMSG)
2590 return r;
2591
2592 /* OK, so this entry is borked. Most likely some entry didn't get synced to disk properly, let's see if
2593 * the next one might work for us instead. */
2594 log_debug_errno(r, "Entry item %" PRIu64 " is bad, skipping over it.", i);
2595
2596 r = bump_array_index(&i, direction, n);
2597 if (r <= 0)
2598 return r;
caeab8f6 2599 }
fb099c8d 2600
b6da4ed0
LP		 2601 /* Ensure our array is properly ordered. */
2602 if (p > 0 && !check_properly_ordered(ofs, p, direction)) {
2603 log_debug("%s: entry array not properly ordered at entry %" PRIu64, f->path, i);
fb099c8d
ZJS		 2604 return -EBADMSG;
2605 }
2606
2607 if (offset)
2608 *offset = ofs;
2609
2610 return 1;
de190aef 2611}
cec736d2 2612
de190aef
LP		 2613int journal_file_next_entry_for_data(
2614 JournalFile *f,
2615 Object *o, uint64_t p,
2616 uint64_t data_offset,
2617 direction_t direction,
2618 Object **ret, uint64_t *offset) {
2619
ded5034e 2620 uint64_t i, n, ofs;
de190aef 2621 Object *d;
989793d3 2622 int r;
cec736d2
LP		 2623
2624 assert(f);
de190aef 2625 assert(p > 0 || !o);
cec736d2 2626
de190aef 2627 r = journal_file_move_to_object(f, OBJECT_DATA, data_offset, &d);
466ccd92 2628 if (r < 0)
de190aef 2629 return r;
cec736d2 2630
de190aef
LP		 2631 n = le64toh(d->data.n_entries);
2632 if (n <= 0)
2633 return n;
cec736d2 2634
de190aef
LP		 2635 if (!o)
2636 i = direction == DIRECTION_DOWN ? 0 : n - 1;
2637 else {
2638 if (o->object.type != OBJECT_ENTRY)
2639 return -EINVAL;
cec736d2 2640
de190aef
LP		 2641 r = generic_array_bisect_plus_one(f,
2642 le64toh(d->data.entry_offset),
2643 le64toh(d->data.entry_array_offset),
2644 le64toh(d->data.n_entries),
2645 p,
2646 test_object_offset,
2647 DIRECTION_DOWN,
2648 NULL, NULL,
2649 &i);
2650
2651 if (r <= 0)
cec736d2
LP		 2652 return r;
2653
aa598ba5
LP		 2654 r = bump_array_index(&i, direction, n);
2655 if (r <= 0)
2656 return r;
de190aef 2657 }
cec736d2 2658
989793d3
LP		 2659 for (;;) {
2660 r = generic_array_get_plus_one(f,
2661 le64toh(d->data.entry_offset),
2662 le64toh(d->data.entry_array_offset),
2663 i,
2664 ret, &ofs);
2665 if (r > 0)
2666 break;
2667 if (r != -EBADMSG)
2668 return r;
2669
2670 log_debug_errno(r, "Data entry item %" PRIu64 " is bad, skipping over it.", i);
2671
2672 r = bump_array_index(&i, direction, n);
2673 if (r <= 0)
2674 return r;
2675 }
ded5034e
LP		 2676
2677 /* Ensure our array is properly ordered. */
2678 if (p > 0 && check_properly_ordered(ofs, p, direction)) {
2679 log_debug("%s data entry array not properly ordered at entry %" PRIu64, f->path, i);
2680 return -EBADMSG;
2681 }
2682
2683 if (offset)
2684 *offset = ofs;
2685
2686 return 1;
de190aef 2687}
cec736d2 2688
cbdca852
LP		 2689int journal_file_move_to_entry_by_offset_for_data(
2690 JournalFile *f,
2691 uint64_t data_offset,
2692 uint64_t p,
2693 direction_t direction,
2694 Object **ret, uint64_t *offset) {
2695
2696 int r;
2697 Object *d;
2698
2699 assert(f);
2700
2701 r = journal_file_move_to_object(f, OBJECT_DATA, data_offset, &d);
2702 if (r < 0)
2703 return r;
2704
2705 return generic_array_bisect_plus_one(f,
2706 le64toh(d->data.entry_offset),
2707 le64toh(d->data.entry_array_offset),
2708 le64toh(d->data.n_entries),
2709 p,
2710 test_object_offset,
2711 direction,
2712 ret, offset, NULL);
2713}
2714
2715int journal_file_move_to_entry_by_monotonic_for_data(
2716 JournalFile *f,
2717 uint64_t data_offset,
2718 sd_id128_t boot_id,
2719 uint64_t monotonic,
2720 direction_t direction,
2721 Object **ret, uint64_t *offset) {
2722
cbdca852
LP		 2723 Object *o, *d;
2724 int r;
2725 uint64_t b, z;
2726
2727 assert(f);
2728
2729 /* First, seek by time */
47838ab3 2730 r = find_data_object_by_boot_id(f, boot_id, &o, &b);
cbdca852
LP		 2731 if (r < 0)
2732 return r;
2733 if (r == 0)
2734 return -ENOENT;
2735
2736 r = generic_array_bisect_plus_one(f,
2737 le64toh(o->data.entry_offset),
2738 le64toh(o->data.entry_array_offset),
2739 le64toh(o->data.n_entries),
2740 monotonic,
2741 test_object_monotonic,
2742 direction,
2743 NULL, &z, NULL);
2744 if (r <= 0)
2745 return r;
2746
2747 /* And now, continue seeking until we find an entry that
2748 * exists in both bisection arrays */
2749
2750 for (;;) {
2751 Object *qo;
2752 uint64_t p, q;
2753
2754 r = journal_file_move_to_object(f, OBJECT_DATA, data_offset, &d);
2755 if (r < 0)
2756 return r;
2757
2758 r = generic_array_bisect_plus_one(f,
2759 le64toh(d->data.entry_offset),
2760 le64toh(d->data.entry_array_offset),
2761 le64toh(d->data.n_entries),
2762 z,
2763 test_object_offset,
2764 direction,
2765 NULL, &p, NULL);
2766 if (r <= 0)
2767 return r;
2768
2769 r = journal_file_move_to_object(f, OBJECT_DATA, b, &o);
2770 if (r < 0)
2771 return r;
2772
2773 r = generic_array_bisect_plus_one(f,
2774 le64toh(o->data.entry_offset),
2775 le64toh(o->data.entry_array_offset),
2776 le64toh(o->data.n_entries),
2777 p,
2778 test_object_offset,
2779 direction,
2780 &qo, &q, NULL);
2781
2782 if (r <= 0)
2783 return r;
2784
2785 if (p == q) {
2786 if (ret)
2787 *ret = qo;
2788 if (offset)
2789 *offset = q;
2790
2791 return 1;
2792 }
2793
2794 z = q;
2795 }
cbdca852
LP		 2796}
2797
de190aef
LP		 2798int journal_file_move_to_entry_by_seqnum_for_data(
2799 JournalFile *f,
2800 uint64_t data_offset,
2801 uint64_t seqnum,
2802 direction_t direction,
2803 Object **ret, uint64_t *offset) {
cec736d2 2804
de190aef
LP		 2805 Object *d;
2806 int r;
cec736d2 2807
91a31dde
LP		 2808 assert(f);
2809
de190aef 2810 r = journal_file_move_to_object(f, OBJECT_DATA, data_offset, &d);
91a31dde 2811 if (r < 0)
de190aef 2812 return r;
cec736d2 2813
de190aef
LP		 2814 return generic_array_bisect_plus_one(f,
2815 le64toh(d->data.entry_offset),
2816 le64toh(d->data.entry_array_offset),
2817 le64toh(d->data.n_entries),
2818 seqnum,
2819 test_object_seqnum,
2820 direction,
2821 ret, offset, NULL);
2822}
cec736d2 2823
de190aef
LP		 2824int journal_file_move_to_entry_by_realtime_for_data(
2825 JournalFile *f,
2826 uint64_t data_offset,
2827 uint64_t realtime,
2828 direction_t direction,
2829 Object **ret, uint64_t *offset) {
2830
2831 Object *d;
2832 int r;
2833
91a31dde
LP		 2834 assert(f);
2835
de190aef 2836 r = journal_file_move_to_object(f, OBJECT_DATA, data_offset, &d);
91a31dde 2837 if (r < 0)
de190aef
LP		 2838 return r;
2839
2840 return generic_array_bisect_plus_one(f,
2841 le64toh(d->data.entry_offset),
2842 le64toh(d->data.entry_array_offset),
2843 le64toh(d->data.n_entries),
2844 realtime,
2845 test_object_realtime,
2846 direction,
2847 ret, offset, NULL);
cec736d2
LP		 2848}
2849
0284adc6 2850void journal_file_dump(JournalFile *f) {
7560fffc 2851 Object *o;
7560fffc 2852 int r;
0284adc6 2853 uint64_t p;
7560fffc
LP		 2854
2855 assert(f);
c88cc6af 2856 assert(f->header);
7560fffc 2857
0284adc6 2858 journal_file_print_header(f);
7560fffc 2859
0284adc6
LP		 2860 p = le64toh(f->header->header_size);
2861 while (p != 0) {
d05089d8 2862 r = journal_file_move_to_object(f, OBJECT_UNUSED, p, &o);
0284adc6
LP		 2863 if (r < 0)
2864 goto fail;
7560fffc 2865
0284adc6 2866 switch (o->object.type) {
d98cc1f2 2867
0284adc6
LP		 2868 case OBJECT_UNUSED:
2869 printf("Type: OBJECT_UNUSED\n");
2870 break;
d98cc1f2 2871
0284adc6
LP		 2872 case OBJECT_DATA:
2873 printf("Type: OBJECT_DATA\n");
2874 break;
7560fffc 2875
3c1668da
LP		 2876 case OBJECT_FIELD:
2877 printf("Type: OBJECT_FIELD\n");
2878 break;
2879
0284adc6 2880 case OBJECT_ENTRY:
507f22bd
ZJS		 2881 printf("Type: OBJECT_ENTRY seqnum=%"PRIu64" monotonic=%"PRIu64" realtime=%"PRIu64"\n",
2882 le64toh(o->entry.seqnum),
2883 le64toh(o->entry.monotonic),
2884 le64toh(o->entry.realtime));
0284adc6 2885 break;
7560fffc 2886
0284adc6
LP		 2887 case OBJECT_FIELD_HASH_TABLE:
2888 printf("Type: OBJECT_FIELD_HASH_TABLE\n");
2889 break;
7560fffc 2890
0284adc6
LP		 2891 case OBJECT_DATA_HASH_TABLE:
2892 printf("Type: OBJECT_DATA_HASH_TABLE\n");
2893 break;
7560fffc 2894
0284adc6
LP		 2895 case OBJECT_ENTRY_ARRAY:
2896 printf("Type: OBJECT_ENTRY_ARRAY\n");
2897 break;
7560fffc 2898
0284adc6 2899 case OBJECT_TAG:
507f22bd
ZJS		 2900 printf("Type: OBJECT_TAG seqnum=%"PRIu64" epoch=%"PRIu64"\n",
2901 le64toh(o->tag.seqnum),
2902 le64toh(o->tag.epoch));
0284adc6 2903 break;
3c1668da
LP		 2904
2905 default:
8facc349 2906 printf("Type: unknown (%i)\n", o->object.type);
3c1668da 2907 break;
0284adc6 2908 }
7560fffc 2909
d89c8fdf
ZJS		 2910 if (o->object.flags & OBJECT_COMPRESSION_MASK)
2911 printf("Flags: %s\n",
2912 object_compressed_to_string(o->object.flags & OBJECT_COMPRESSION_MASK));
7560fffc 2913
0284adc6
LP		 2914 if (p == le64toh(f->header->tail_object_offset))
2915 p = 0;
2916 else
2917 p = p + ALIGN64(le64toh(o->object.size));
2918 }
7560fffc 2919
0284adc6
LP		 2920 return;
2921fail:
2922 log_error("File corrupt");
7560fffc
LP		 2923}
2924
718fe4b1
ZJS		 2925static const char* format_timestamp_safe(char *buf, size_t l, usec_t t) {
2926 const char *x;
2927
2928 x = format_timestamp(buf, l, t);
2929 if (x)
2930 return x;
2931 return " --- ";
2932}
2933
0284adc6 2934void journal_file_print_header(JournalFile *f) {
2765b7bb 2935 char a[33], b[33], c[33], d[33];
ed375beb 2936 char x[FORMAT_TIMESTAMP_MAX], y[FORMAT_TIMESTAMP_MAX], z[FORMAT_TIMESTAMP_MAX];
a1a03e30
LP		 2937 struct stat st;
2938 char bytes[FORMAT_BYTES_MAX];
7560fffc
LP		 2939
2940 assert(f);
c88cc6af 2941 assert(f->header);
7560fffc 2942
0284adc6
LP		 2943 printf("File Path: %s\n"
2944 "File ID: %s\n"
2945 "Machine ID: %s\n"
2946 "Boot ID: %s\n"
2947 "Sequential Number ID: %s\n"
2948 "State: %s\n"
2949 "Compatible Flags:%s%s\n"
d89c8fdf 2950 "Incompatible Flags:%s%s%s\n"
507f22bd
ZJS		 2951 "Header size: %"PRIu64"\n"
2952 "Arena size: %"PRIu64"\n"
2953 "Data Hash Table Size: %"PRIu64"\n"
2954 "Field Hash Table Size: %"PRIu64"\n"
0284adc6 2955 "Rotate Suggested: %s\n"
0808b92f
LP		 2956 "Head Sequential Number: %"PRIu64" (%"PRIx64")\n"
2957 "Tail Sequential Number: %"PRIu64" (%"PRIx64")\n"
2958 "Head Realtime Timestamp: %s (%"PRIx64")\n"
2959 "Tail Realtime Timestamp: %s (%"PRIx64")\n"
2960 "Tail Monotonic Timestamp: %s (%"PRIx64")\n"
507f22bd
ZJS		 2961 "Objects: %"PRIu64"\n"
2962 "Entry Objects: %"PRIu64"\n",
0284adc6
LP		 2963 f->path,
2964 sd_id128_to_string(f->header->file_id, a),
2965 sd_id128_to_string(f->header->machine_id, b),
2966 sd_id128_to_string(f->header->boot_id, c),
2765b7bb 2967 sd_id128_to_string(f->header->seqnum_id, d),
3223f44f
LP		 2968 f->header->state == STATE_OFFLINE ? "OFFLINE" :
2969 f->header->state == STATE_ONLINE ? "ONLINE" :
2970 f->header->state == STATE_ARCHIVED ? "ARCHIVED" : "UNKNOWN",
8088cbd3 2971 JOURNAL_HEADER_SEALED(f->header) ? " SEALED" : "",
d89c8fdf
ZJS		 2972 (le32toh(f->header->compatible_flags) & ~HEADER_COMPATIBLE_ANY) ? " ???" : "",
2973 JOURNAL_HEADER_COMPRESSED_XZ(f->header) ? " COMPRESSED-XZ" : "",
2974 JOURNAL_HEADER_COMPRESSED_LZ4(f->header) ? " COMPRESSED-LZ4" : "",
2975 (le32toh(f->header->incompatible_flags) & ~HEADER_INCOMPATIBLE_ANY) ? " ???" : "",
507f22bd
ZJS		 2976 le64toh(f->header->header_size),
2977 le64toh(f->header->arena_size),
2978 le64toh(f->header->data_hash_table_size) / sizeof(HashItem),
2979 le64toh(f->header->field_hash_table_size) / sizeof(HashItem),
fb0951b0 2980 yes_no(journal_file_rotate_suggested(f, 0)),
0808b92f
LP		 2981 le64toh(f->header->head_entry_seqnum), le64toh(f->header->head_entry_seqnum),
2982 le64toh(f->header->tail_entry_seqnum), le64toh(f->header->tail_entry_seqnum),
2983 format_timestamp_safe(x, sizeof(x), le64toh(f->header->head_entry_realtime)), le64toh(f->header->head_entry_realtime),
2984 format_timestamp_safe(y, sizeof(y), le64toh(f->header->tail_entry_realtime)), le64toh(f->header->tail_entry_realtime),
2985 format_timespan(z, sizeof(z), le64toh(f->header->tail_entry_monotonic), USEC_PER_MSEC), le64toh(f->header->tail_entry_monotonic),
507f22bd
ZJS		 2986 le64toh(f->header->n_objects),
2987 le64toh(f->header->n_entries));
7560fffc 2988
0284adc6 2989 if (JOURNAL_HEADER_CONTAINS(f->header, n_data))
507f22bd 2990 printf("Data Objects: %"PRIu64"\n"
0284adc6 2991 "Data Hash Table Fill: %.1f%%\n",
507f22bd 2992 le64toh(f->header->n_data),
0284adc6 2993 100.0 * (double) le64toh(f->header->n_data) / ((double) (le64toh(f->header->data_hash_table_size) / sizeof(HashItem))));
7560fffc 2994
0284adc6 2995 if (JOURNAL_HEADER_CONTAINS(f->header, n_fields))
507f22bd 2996 printf("Field Objects: %"PRIu64"\n"
0284adc6 2997 "Field Hash Table Fill: %.1f%%\n",
507f22bd 2998 le64toh(f->header->n_fields),
0284adc6 2999 100.0 * (double) le64toh(f->header->n_fields) / ((double) (le64toh(f->header->field_hash_table_size) / sizeof(HashItem))));
3223f44f
LP		 3000
3001 if (JOURNAL_HEADER_CONTAINS(f->header, n_tags))
507f22bd
ZJS		 3002 printf("Tag Objects: %"PRIu64"\n",
3003 le64toh(f->header->n_tags));
3223f44f 3004 if (JOURNAL_HEADER_CONTAINS(f->header, n_entry_arrays))
507f22bd
ZJS		 3005 printf("Entry Array Objects: %"PRIu64"\n",
3006 le64toh(f->header->n_entry_arrays));
a1a03e30
LP		 3007
3008 if (fstat(f->fd, &st) >= 0)
59f448cf 3009 printf("Disk usage: %s\n", format_bytes(bytes, sizeof(bytes), (uint64_t) st.st_blocks * 512ULL));
7560fffc
LP		 3010}
3011
fc68c929
LP		 3012static int journal_file_warn_btrfs(JournalFile *f) {
3013 unsigned attrs;
3014 int r;
3015
3016 assert(f);
3017
3018 /* Before we write anything, check if the COW logic is turned
3019 * off on btrfs. Given our write pattern that is quite
3020 * unfriendly to COW file systems this should greatly improve
3021 * performance on COW file systems, such as btrfs, at the
3022 * expense of data integrity features (which shouldn't be too
3023 * bad, given that we do our own checksumming). */
3024
3025 r = btrfs_is_filesystem(f->fd);
3026 if (r < 0)
3027 return log_warning_errno(r, "Failed to determine if journal is on btrfs: %m");
3028 if (!r)
3029 return 0;
3030
3031 r = read_attr_fd(f->fd, &attrs);
3032 if (r < 0)
3033 return log_warning_errno(r, "Failed to read file attributes: %m");
3034
3035 if (attrs & FS_NOCOW_FL) {
3036 log_debug("Detected btrfs file system with copy-on-write disabled, all is good.");
3037 return 0;
3038 }
3039
3040 log_notice("Creating journal file %s on a btrfs file system, and copy-on-write is enabled. "
3041 "This is likely to slow down journal access substantially, please consider turning "
3042 "off the copy-on-write file attribute on the journal directory, using chattr +C.", f->path);
3043
3044 return 1;
3045}
3046
0284adc6 3047int journal_file_open(
5d1ce257 3048 int fd,
0284adc6
LP		 3049 const char *fname,
3050 int flags,
3051 mode_t mode,
3052 bool compress,
baed47c3 3053 bool seal,
0284adc6
LP		 3054 JournalMetrics *metrics,
3055 MMapCache *mmap_cache,
b58c888f 3056 Set *deferred_closes,
0284adc6
LP		 3057 JournalFile *template,
3058 JournalFile **ret) {
7560fffc 3059
fa6ac760 3060 bool newly_created = false;
0284adc6 3061 JournalFile *f;
fa6ac760 3062 void *h;
0284adc6 3063 int r;
7560fffc 3064
0559d3a5 3065 assert(ret);
5d1ce257 3066 assert(fd >= 0 || fname);
7560fffc 3067
0284adc6
LP		 3068 if ((flags & O_ACCMODE) != O_RDONLY &&
3069 (flags & O_ACCMODE) != O_RDWR)
3070 return -EINVAL;
7560fffc 3071
5d1ce257
LP		 3072 if (fname) {
3073 if (!endswith(fname, ".journal") &&
3074 !endswith(fname, ".journal~"))
3075 return -EINVAL;
3076 }
7560fffc 3077
0284adc6
LP		 3078 f = new0(JournalFile, 1);
3079 if (!f)
3080 return -ENOMEM;
7560fffc 3081
5d1ce257 3082 f->fd = fd;
0284adc6 3083 f->mode = mode;
7560fffc 3084
0284adc6
LP		 3085 f->flags = flags;
3086 f->prot = prot_from_flags(flags);
3087 f->writable = (flags & O_ACCMODE) != O_RDONLY;
92261977 3088#if defined(HAVE_LZ4)
d89c8fdf
ZJS		 3089 f->compress_lz4 = compress;
3090#elif defined(HAVE_XZ)
3091 f->compress_xz = compress;
48b61739 3092#endif
49a32d43 3093#ifdef HAVE_GCRYPT
baed47c3 3094 f->seal = seal;
49a32d43 3095#endif
7560fffc 3096
0284adc6
LP		 3097 if (mmap_cache)
3098 f->mmap = mmap_cache_ref(mmap_cache);
3099 else {
84168d80 3100 f->mmap = mmap_cache_new();
0284adc6
LP		 3101 if (!f->mmap) {
3102 r = -ENOMEM;
3103 goto fail;
3104 }
3105 }
7560fffc 3106
7645c77b 3107 if (fname) {
5d1ce257 3108 f->path = strdup(fname);
7645c77b
ZJS		 3109 if (!f->path) {
3110 r = -ENOMEM;
3111 goto fail;
3112 }
3113 } else {
3114 /* If we don't know the path, fill in something explanatory and vaguely useful */
3115 if (asprintf(&f->path, "/proc/self/%i", fd) < 0) {
3116 r = -ENOMEM;
3117 goto fail;
3118 }
0284adc6 3119 }
7560fffc 3120
4743015d 3121 f->chain_cache = ordered_hashmap_new(&uint64_hash_ops);
a4bcff5b
LP		 3122 if (!f->chain_cache) {
3123 r = -ENOMEM;
3124 goto fail;
3125 }
3126
0284adc6 3127 if (f->fd < 0) {
5d1ce257
LP		 3128 f->fd = open(f->path, f->flags|O_CLOEXEC, f->mode);
3129 if (f->fd < 0) {
3130 r = -errno;
3131 goto fail;
3132 }
3133
3134 /* fds we opened here by us should also be closed by us. */
3135 f->close_fd = true;
7560fffc 3136 }
7560fffc 3137
2678031a
LP		 3138 r = journal_file_fstat(f);
3139 if (r < 0)
0284adc6 3140 goto fail;
7560fffc 3141
0284adc6 3142 if (f->last_stat.st_size == 0 && f->writable) {
11689d2a 3143
fc68c929 3144 (void) journal_file_warn_btrfs(f);
11689d2a 3145
fb0951b0
LP		 3146 /* Let's attach the creation time to the journal file,
3147 * so that the vacuuming code knows the age of this
3148 * file even if the file might end up corrupted one
3149 * day... Ideally we'd just use the creation time many
3150 * file systems maintain for each file, but there is
3151 * currently no usable API to query this, hence let's
3152 * emulate this via extended attributes. If extended
3153 * attributes are not supported we'll just skip this,
7517e174 3154 * and rely solely on mtime/atime/ctime of the file. */
fb0951b0 3155
d61b600d 3156 fd_setcrtime(f->fd, 0);
7560fffc 3157
feb12d3e 3158#ifdef HAVE_GCRYPT
0284adc6 3159 /* Try to load the FSPRG state, and if we can't, then
baed47c3 3160 * just don't do sealing */
49a32d43
LP		 3161 if (f->seal) {
3162 r = journal_file_fss_load(f);
3163 if (r < 0)
3164 f->seal = false;
3165 }
feb12d3e 3166#endif
7560fffc 3167
0284adc6
LP		 3168 r = journal_file_init_header(f, template);
3169 if (r < 0)
3170 goto fail;
7560fffc 3171
2678031a
LP		 3172 r = journal_file_fstat(f);
3173 if (r < 0)
0284adc6 3174 goto fail;
fb0951b0
LP		 3175
3176 newly_created = true;
0284adc6 3177 }
7560fffc 3178
0284adc6 3179 if (f->last_stat.st_size < (off_t) HEADER_SIZE_MIN) {
cfb571f3 3180 r = -ENODATA;
0284adc6
LP		 3181 goto fail;
3182 }
7560fffc 3183
fa6ac760 3184 r = mmap_cache_get(f->mmap, f->fd, f->prot, CONTEXT_HEADER, true, 0, PAGE_ALIGN(sizeof(Header)), &f->last_stat, &h);
977eaa1e 3185 if (r < 0)
0284adc6 3186 goto fail;
7560fffc 3187
fa6ac760
LP		 3188 f->header = h;
3189
0284adc6 3190 if (!newly_created) {
b58c888f
VC		 3191 if (deferred_closes)
3192 journal_file_close_set(deferred_closes);
3193
0284adc6
LP		 3194 r = journal_file_verify_header(f);
3195 if (r < 0)
3196 goto fail;
3197 }
7560fffc 3198
feb12d3e 3199#ifdef HAVE_GCRYPT
0284adc6 3200 if (!newly_created && f->writable) {
baed47c3 3201 r = journal_file_fss_load(f);
0284adc6
LP		 3202 if (r < 0)
3203 goto fail;
3204 }
feb12d3e 3205#endif
cec736d2
LP		 3206
3207 if (f->writable) {
4a92baf3
LP		 3208 if (metrics) {
3209 journal_default_metrics(metrics, f->fd);
3210 f->metrics = *metrics;
3211 } else if (template)
3212 f->metrics = template->metrics;
3213
cec736d2
LP		 3214 r = journal_file_refresh_header(f);
3215 if (r < 0)
3216 goto fail;
3217 }
3218
feb12d3e 3219#ifdef HAVE_GCRYPT
baed47c3 3220 r = journal_file_hmac_setup(f);
14d10188
LP		 3221 if (r < 0)
3222 goto fail;
feb12d3e 3223#endif
14d10188 3224
cec736d2 3225 if (newly_created) {
de190aef 3226 r = journal_file_setup_field_hash_table(f);
cec736d2
LP		 3227 if (r < 0)
3228 goto fail;
3229
de190aef 3230 r = journal_file_setup_data_hash_table(f);
cec736d2
LP		 3231 if (r < 0)
3232 goto fail;
7560fffc 3233
feb12d3e 3234#ifdef HAVE_GCRYPT
7560fffc
LP		 3235 r = journal_file_append_first_tag(f);
3236 if (r < 0)
3237 goto fail;
feb12d3e 3238#endif
cec736d2
LP		 3239 }
3240
fa6ac760
LP		 3241 if (mmap_cache_got_sigbus(f->mmap, f->fd)) {
3242 r = -EIO;
3243 goto fail;
3244 }
3245
7a24f3bf 3246 if (template && template->post_change_timer) {
e167d7fd
LP		 3247 r = journal_file_enable_post_change_timer(
3248 f,
3249 sd_event_source_get_event(template->post_change_timer),
3250 template->post_change_timer_period);
7a24f3bf 3251
7a24f3bf
VC		 3252 if (r < 0)
3253 goto fail;
3254 }
3255
f8e2f4d6 3256 /* The file is opened now successfully, thus we take possession of any passed in fd. */
5d1ce257
LP		 3257 f->close_fd = true;
3258
0559d3a5 3259 *ret = f;
cec736d2
LP		 3260 return 0;
3261
3262fail:
fa6ac760
LP		 3263 if (f->fd >= 0 && mmap_cache_got_sigbus(f->mmap, f->fd))
3264 r = -EIO;
3265
69a3a6fd 3266 (void) journal_file_close(f);
cec736d2
LP		 3267
3268 return r;
3269}
0ac38b70 3270
b58c888f 3271int journal_file_rotate(JournalFile **f, bool compress, bool seal, Set *deferred_closes) {
57535f47 3272 _cleanup_free_ char *p = NULL;
0ac38b70
LP		 3273 size_t l;
3274 JournalFile *old_file, *new_file = NULL;
3275 int r;
3276
3277 assert(f);
3278 assert(*f);
3279
3280 old_file = *f;
3281
3282 if (!old_file->writable)
3283 return -EINVAL;
3284
5d1ce257 3285 /* Is this a journal file that was passed to us as fd? If so, we synthesized a path name for it, and we refuse
13e785f7 3286 * rotation, since we don't know the actual path, and couldn't rename the file hence. */
5d1ce257
LP		 3287 if (path_startswith(old_file->path, "/proc/self/fd"))
3288 return -EINVAL;
3289
0ac38b70
LP		 3290 if (!endswith(old_file->path, ".journal"))
3291 return -EINVAL;
3292
3293 l = strlen(old_file->path);
57535f47
ZJS		 3294 r = asprintf(&p, "%.*s@" SD_ID128_FORMAT_STR "-%016"PRIx64"-%016"PRIx64".journal",
3295 (int) l - 8, old_file->path,
3296 SD_ID128_FORMAT_VAL(old_file->header->seqnum_id),
3297 le64toh((*f)->header->head_entry_seqnum),
3298 le64toh((*f)->header->head_entry_realtime));
3299 if (r < 0)
0ac38b70
LP		 3300 return -ENOMEM;
3301
2678031a
LP		 3302 /* Try to rename the file to the archived version. If the file
3303 * already was deleted, we'll get ENOENT, let's ignore that
3304 * case. */
0ac38b70 3305 r = rename(old_file->path, p);
2678031a 3306 if (r < 0 && errno != ENOENT)
0ac38b70
LP		 3307 return -errno;
3308
1fcefd88
LP		 3309 /* Sync the rename to disk */
3310 (void) fsync_directory_of_file(old_file->fd);
3311
8eb85171
VC		 3312 /* Set as archive so offlining commits w/state=STATE_ARCHIVED.
3313 * Previously we would set old_file->header->state to STATE_ARCHIVED directly here,
3314 * but journal_file_set_offline() short-circuits when state != STATE_ONLINE, which
3315 * would result in the rotated journal never getting fsync() called before closing.
3316 * Now we simply queue the archive state by setting an archive bit, leaving the state
3317 * as STATE_ONLINE so proper offlining occurs. */
3318 old_file->archive = true;
0ac38b70 3319
f27a3864
LP		 3320 /* Currently, btrfs is not very good with out write patterns
3321 * and fragments heavily. Let's defrag our journal files when
3322 * we archive them */
3323 old_file->defrag_on_close = true;
3324
5d1ce257 3325 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		 3326
3327 if (deferred_closes &&
3328 set_put(deferred_closes, old_file) >= 0)
3329 (void) journal_file_set_offline(old_file, false);
3330 else
3331 (void) journal_file_close(old_file);
0ac38b70
LP		 3332
3333 *f = new_file;
3334 return r;
3335}
3336
9447a7f1
LP		 3337int journal_file_open_reliably(
3338 const char *fname,
3339 int flags,
3340 mode_t mode,
7560fffc 3341 bool compress,
baed47c3 3342 bool seal,
4a92baf3 3343 JournalMetrics *metrics,
27370278 3344 MMapCache *mmap_cache,
b58c888f 3345 Set *deferred_closes,
9447a7f1
LP		 3346 JournalFile *template,
3347 JournalFile **ret) {
3348
3349 int r;
3350 size_t l;
ed375beb 3351 _cleanup_free_ char *p = NULL;
9447a7f1 3352
5d1ce257 3353 r = journal_file_open(-1, fname, flags, mode, compress, seal, metrics, mmap_cache, deferred_closes, template, ret);
288359db 3354 if (!IN_SET(r,
b288cdeb
ZJS		 3355 -EBADMSG, /* Corrupted */
3356 -ENODATA, /* Truncated */
3357 -EHOSTDOWN, /* Other machine */
3358 -EPROTONOSUPPORT, /* Incompatible feature */
3359 -EBUSY, /* Unclean shutdown */
3360 -ESHUTDOWN, /* Already archived */
288359db 3361 -EIO, /* IO error, including SIGBUS on mmap */
ae739cc1
LP		 3362 -EIDRM, /* File has been deleted */
3363 -ETXTBSY)) /* File is from the future */
9447a7f1
LP		 3364 return r;
3365
3366 if ((flags & O_ACCMODE) == O_RDONLY)
3367 return r;
3368
3369 if (!(flags & O_CREAT))
3370 return r;
3371
7560fffc
LP		 3372 if (!endswith(fname, ".journal"))
3373 return r;
3374
5c70eab4
LP		 3375 /* The file is corrupted. Rotate it away and try it again (but only once) */
3376
9447a7f1 3377 l = strlen(fname);
d587eca5 3378 if (asprintf(&p, "%.*s@%016"PRIx64 "-%016"PRIx64 ".journal~",
57535f47 3379 (int) l - 8, fname,
d587eca5 3380 now(CLOCK_REALTIME),
9bf3b535 3381 random_u64()) < 0)
9447a7f1
LP		 3382 return -ENOMEM;
3383
65089b82 3384 if (rename(fname, p) < 0)
9447a7f1
LP		 3385 return -errno;
3386
f27a3864
LP		 3387 /* btrfs doesn't cope well with our write pattern and
3388 * fragments heavily. Let's defrag all files we rotate */
11689d2a 3389
a67d68b8 3390 (void) chattr_path(p, 0, FS_NOCOW_FL);
f27a3864
LP		 3391 (void) btrfs_defrag(p);
3392
65089b82 3393 log_warning_errno(r, "File %s corrupted or uncleanly shut down, renaming and replacing.", fname);
9447a7f1 3394
5d1ce257 3395 return journal_file_open(-1, fname, flags, mode, compress, seal, metrics, mmap_cache, deferred_closes, template, ret);
9447a7f1
LP		 3396}
3397
cf244689
LP		 3398int journal_file_copy_entry(JournalFile *from, JournalFile *to, Object *o, uint64_t p, uint64_t *seqnum, Object **ret, uint64_t *offset) {
3399 uint64_t i, n;
3400 uint64_t q, xor_hash = 0;
3401 int r;
3402 EntryItem *items;
3403 dual_timestamp ts;
3404
3405 assert(from);
3406 assert(to);
3407 assert(o);
3408 assert(p);
3409
3410 if (!to->writable)
3411 return -EPERM;
3412
3413 ts.monotonic = le64toh(o->entry.monotonic);
3414 ts.realtime = le64toh(o->entry.realtime);
3415
cf244689 3416 n = journal_file_entry_n_items(o);
4faa7004
TA		 3417 /* alloca() can't take 0, hence let's allocate at least one */
3418 items = alloca(sizeof(EntryItem) * MAX(1u, n));
cf244689
LP		 3419
3420 for (i = 0; i < n; i++) {
4fd052ae
FC		 3421 uint64_t l, h;
3422 le64_t le_hash;
cf244689
LP		 3423 size_t t;
3424 void *data;
3425 Object *u;
3426
3427 q = le64toh(o->entry.items[i].object_offset);
3428 le_hash = o->entry.items[i].hash;
3429
3430 r = journal_file_move_to_object(from, OBJECT_DATA, q, &o);
3431 if (r < 0)
3432 return r;
3433
3434 if (le_hash != o->data.hash)
3435 return -EBADMSG;
3436
3437 l = le64toh(o->object.size) - offsetof(Object, data.payload);
3438 t = (size_t) l;
3439
3440 /* We hit the limit on 32bit machines */
3441 if ((uint64_t) t != l)
3442 return -E2BIG;
3443
d89c8fdf 3444 if (o->object.flags & OBJECT_COMPRESSION_MASK) {
3b1a55e1 3445#if defined(HAVE_XZ) || defined(HAVE_LZ4)
a7f7d1bd 3446 size_t rsize = 0;
cf244689 3447
d89c8fdf
ZJS		 3448 r = decompress_blob(o->object.flags & OBJECT_COMPRESSION_MASK,
3449 o->data.payload, l, &from->compress_buffer, &from->compress_buffer_size, &rsize, 0);
3450 if (r < 0)
3451 return r;
cf244689
LP		 3452
3453 data = from->compress_buffer;
3454 l = rsize;
3b1a55e1
ZJS		 3455#else
3456 return -EPROTONOSUPPORT;
3457#endif
cf244689
LP		 3458 } else
3459 data = o->data.payload;
3460
3461 r = journal_file_append_data(to, data, l, &u, &h);
3462 if (r < 0)
3463 return r;
3464
3465 xor_hash ^= le64toh(u->data.hash);
3466 items[i].object_offset = htole64(h);
3467 items[i].hash = u->data.hash;
3468
3469 r = journal_file_move_to_object(from, OBJECT_ENTRY, p, &o);
3470 if (r < 0)
3471 return r;
3472 }
3473
fa6ac760
LP		 3474 r = journal_file_append_entry_internal(to, &ts, xor_hash, items, n, seqnum, ret, offset);
3475
3476 if (mmap_cache_got_sigbus(to->mmap, to->fd))
3477 return -EIO;
3478
3479 return r;
cf244689 3480}
babfc091 3481
8580d1f7
LP		 3482void journal_reset_metrics(JournalMetrics *m) {
3483 assert(m);
3484
3485 /* Set everything to "pick automatic values". */
3486
3487 *m = (JournalMetrics) {
3488 .min_use = (uint64_t) -1,
3489 .max_use = (uint64_t) -1,
3490 .min_size = (uint64_t) -1,
3491 .max_size = (uint64_t) -1,
3492 .keep_free = (uint64_t) -1,
3493 .n_max_files = (uint64_t) -1,
3494 };
3495}
3496
babfc091 3497void journal_default_metrics(JournalMetrics *m, int fd) {
8580d1f7 3498 char a[FORMAT_BYTES_MAX], b[FORMAT_BYTES_MAX], c[FORMAT_BYTES_MAX], d[FORMAT_BYTES_MAX], e[FORMAT_BYTES_MAX];
babfc091 3499 struct statvfs ss;
8580d1f7 3500 uint64_t fs_size;
babfc091
LP		 3501
3502 assert(m);
3503 assert(fd >= 0);
3504
3505 if (fstatvfs(fd, &ss) >= 0)
3506 fs_size = ss.f_frsize * ss.f_blocks;
8580d1f7
LP		 3507 else {
3508 log_debug_errno(errno, "Failed to detremine disk size: %m");
3509 fs_size = 0;
3510 }
babfc091
LP		 3511
3512 if (m->max_use == (uint64_t) -1) {
3513
3514 if (fs_size > 0) {
3515 m->max_use = PAGE_ALIGN(fs_size / 10); /* 10% of file system size */
3516
3517 if (m->max_use > DEFAULT_MAX_USE_UPPER)
3518 m->max_use = DEFAULT_MAX_USE_UPPER;
3519
3520 if (m->max_use < DEFAULT_MAX_USE_LOWER)
3521 m->max_use = DEFAULT_MAX_USE_LOWER;
3522 } else
3523 m->max_use = DEFAULT_MAX_USE_LOWER;
3524 } else {
3525 m->max_use = PAGE_ALIGN(m->max_use);
3526
8580d1f7 3527 if (m->max_use != 0 && m->max_use < JOURNAL_FILE_SIZE_MIN*2)
babfc091
LP		 3528 m->max_use = JOURNAL_FILE_SIZE_MIN*2;
3529 }
3530
8580d1f7
LP		 3531 if (m->min_use == (uint64_t) -1)
3532 m->min_use = DEFAULT_MIN_USE;
3533
3534 if (m->min_use > m->max_use)
3535 m->min_use = m->max_use;
3536
babfc091
LP		 3537 if (m->max_size == (uint64_t) -1) {
3538 m->max_size = PAGE_ALIGN(m->max_use / 8); /* 8 chunks */
3539
3540 if (m->max_size > DEFAULT_MAX_SIZE_UPPER)
3541 m->max_size = DEFAULT_MAX_SIZE_UPPER;
3542 } else
3543 m->max_size = PAGE_ALIGN(m->max_size);
3544
8580d1f7
LP		 3545 if (m->max_size != 0) {
3546 if (m->max_size < JOURNAL_FILE_SIZE_MIN)
3547 m->max_size = JOURNAL_FILE_SIZE_MIN;
babfc091 3548
8580d1f7
LP		 3549 if (m->max_use != 0 && m->max_size*2 > m->max_use)
3550 m->max_use = m->max_size*2;
3551 }
babfc091
LP		 3552
3553 if (m->min_size == (uint64_t) -1)
3554 m->min_size = JOURNAL_FILE_SIZE_MIN;
3555 else {
3556 m->min_size = PAGE_ALIGN(m->min_size);
3557
3558 if (m->min_size < JOURNAL_FILE_SIZE_MIN)
3559 m->min_size = JOURNAL_FILE_SIZE_MIN;
3560
8580d1f7 3561 if (m->max_size != 0 && m->min_size > m->max_size)
babfc091
LP		 3562 m->max_size = m->min_size;
3563 }
3564
3565 if (m->keep_free == (uint64_t) -1) {
3566
3567 if (fs_size > 0) {
8621b110 3568 m->keep_free = PAGE_ALIGN(fs_size * 3 / 20); /* 15% of file system size */
babfc091
LP		 3569
3570 if (m->keep_free > DEFAULT_KEEP_FREE_UPPER)
3571 m->keep_free = DEFAULT_KEEP_FREE_UPPER;
3572
3573 } else
3574 m->keep_free = DEFAULT_KEEP_FREE;
3575 }
3576
8580d1f7
LP		 3577 if (m->n_max_files == (uint64_t) -1)
3578 m->n_max_files = DEFAULT_N_MAX_FILES;
3579
3580 log_debug("Fixed min_use=%s max_use=%s max_size=%s min_size=%s keep_free=%s n_max_files=%" PRIu64,
3581 format_bytes(a, sizeof(a), m->min_use),
3582 format_bytes(b, sizeof(b), m->max_use),
3583 format_bytes(c, sizeof(c), m->max_size),
3584 format_bytes(d, sizeof(d), m->min_size),
3585 format_bytes(e, sizeof(e), m->keep_free),
3586 m->n_max_files);
babfc091 3587}
08984293
LP		 3588
3589int journal_file_get_cutoff_realtime_usec(JournalFile *f, usec_t *from, usec_t *to) {
08984293 3590 assert(f);
c88cc6af 3591 assert(f->header);
08984293
LP		 3592 assert(from || to);
3593
3594 if (from) {
162566a4
LP		 3595 if (f->header->head_entry_realtime == 0)
3596 return -ENOENT;
08984293 3597
162566a4 3598 *from = le64toh(f->header->head_entry_realtime);
08984293
LP		 3599 }
3600
3601 if (to) {
162566a4
LP		 3602 if (f->header->tail_entry_realtime == 0)
3603 return -ENOENT;
08984293 3604
162566a4 3605 *to = le64toh(f->header->tail_entry_realtime);
08984293
LP		 3606 }
3607
3608 return 1;
3609}
3610
3611int journal_file_get_cutoff_monotonic_usec(JournalFile *f, sd_id128_t boot_id, usec_t *from, usec_t *to) {
08984293
LP		 3612 Object *o;
3613 uint64_t p;
3614 int r;
3615
3616 assert(f);
3617 assert(from || to);
3618
47838ab3 3619 r = find_data_object_by_boot_id(f, boot_id, &o, &p);
08984293
LP		 3620 if (r <= 0)
3621 return r;
3622
3623 if (le64toh(o->data.n_entries) <= 0)
3624 return 0;
3625
3626 if (from) {
3627 r = journal_file_move_to_object(f, OBJECT_ENTRY, le64toh(o->data.entry_offset), &o);
3628 if (r < 0)
3629 return r;
3630
3631 *from = le64toh(o->entry.monotonic);
3632 }
3633
3634 if (to) {
3635 r = journal_file_move_to_object(f, OBJECT_DATA, p, &o);
3636 if (r < 0)
3637 return r;
3638
3639 r = generic_array_get_plus_one(f,
3640 le64toh(o->data.entry_offset),
3641 le64toh(o->data.entry_array_offset),
3642 le64toh(o->data.n_entries)-1,
3643 &o, NULL);
3644 if (r <= 0)
3645 return r;
3646
3647 *to = le64toh(o->entry.monotonic);
3648 }
3649
3650 return 1;
3651}
dca6219e 3652
fb0951b0 3653bool journal_file_rotate_suggested(JournalFile *f, usec_t max_file_usec) {
dca6219e 3654 assert(f);
c88cc6af 3655 assert(f->header);
dca6219e
LP		 3656
3657 /* If we gained new header fields we gained new features,
3658 * hence suggest a rotation */
361f9cbc
LP		 3659 if (le64toh(f->header->header_size) < sizeof(Header)) {
3660 log_debug("%s uses an outdated header, suggesting rotation.", f->path);
dca6219e 3661 return true;
361f9cbc 3662 }
dca6219e
LP		 3663
3664 /* Let's check if the hash tables grew over a certain fill
3665 * level (75%, borrowing this value from Java's hash table
3666 * implementation), and if so suggest a rotation. To calculate
3667 * the fill level we need the n_data field, which only exists
3668 * in newer versions. */
3669
3670 if (JOURNAL_HEADER_CONTAINS(f->header, n_data))
361f9cbc 3671 if (le64toh(f->header->n_data) * 4ULL > (le64toh(f->header->data_hash_table_size) / sizeof(HashItem)) * 3ULL) {
507f22bd 3672 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		 3673 f->path,
3674 100.0 * (double) le64toh(f->header->n_data) / ((double) (le64toh(f->header->data_hash_table_size) / sizeof(HashItem))),
507f22bd
ZJS		 3675 le64toh(f->header->n_data),
3676 le64toh(f->header->data_hash_table_size) / sizeof(HashItem),
3677 (unsigned long long) f->last_stat.st_size,
3678 f->last_stat.st_size / le64toh(f->header->n_data));
dca6219e 3679 return true;
361f9cbc 3680 }
dca6219e
LP		 3681
3682 if (JOURNAL_HEADER_CONTAINS(f->header, n_fields))
361f9cbc 3683 if (le64toh(f->header->n_fields) * 4ULL > (le64toh(f->header->field_hash_table_size) / sizeof(HashItem)) * 3ULL) {
507f22bd 3684 log_debug("Field hash table of %s has a fill level at %.1f (%"PRIu64" of %"PRIu64" items), suggesting rotation.",
361f9cbc
LP		 3685 f->path,
3686 100.0 * (double) le64toh(f->header->n_fields) / ((double) (le64toh(f->header->field_hash_table_size) / sizeof(HashItem))),
507f22bd
ZJS		 3687 le64toh(f->header->n_fields),
3688 le64toh(f->header->field_hash_table_size) / sizeof(HashItem));
dca6219e 3689 return true;
361f9cbc 3690 }
dca6219e 3691
0598fd4a
LP		 3692 /* Are the data objects properly indexed by field objects? */
3693 if (JOURNAL_HEADER_CONTAINS(f->header, n_data) &&
3694 JOURNAL_HEADER_CONTAINS(f->header, n_fields) &&
3695 le64toh(f->header->n_data) > 0 &&
3696 le64toh(f->header->n_fields) == 0)
3697 return true;
3698
fb0951b0
LP		 3699 if (max_file_usec > 0) {
3700 usec_t t, h;
3701
3702 h = le64toh(f->header->head_entry_realtime);
3703 t = now(CLOCK_REALTIME);
3704
3705 if (h > 0 && t > h + max_file_usec)
3706 return true;
3707 }
3708
dca6219e
LP		 3709 return false;
3710}
Unnamed repository; edit this file 'description' to name the repository.