]> git.ipfire.org Git - thirdparty/systemd.git/blob - src/libsystemd/sd-journal/journal-file.c
projects / thirdparty / systemd.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Merge pull request #28728 from yuwata/fstab-no
[thirdparty/systemd.git] / src / libsystemd / sd-journal / journal-file.c
1 /* SPDX-License-Identifier: LGPL-2.1-or-later */
2
3 #include <errno.h>
4 #include <fcntl.h>
5 #include <linux/fs.h>
6 #include <linux/magic.h>
7 #include <pthread.h>
8 #include <stddef.h>
9 #include <sys/mman.h>
10 #include <sys/statvfs.h>
11 #include <sys/uio.h>
12 #include <unistd.h>
13
14 #include "sd-event.h"
15
16 #include "alloc-util.h"
17 #include "chattr-util.h"
18 #include "compress.h"
19 #include "env-util.h"
20 #include "fd-util.h"
21 #include "format-util.h"
22 #include "fs-util.h"
23 #include "id128-util.h"
24 #include "journal-authenticate.h"
25 #include "journal-def.h"
26 #include "journal-file.h"
27 #include "journal-internal.h"
28 #include "lookup3.h"
29 #include "memory-util.h"
30 #include "missing_threads.h"
31 #include "path-util.h"
32 #include "prioq.h"
33 #include "random-util.h"
34 #include "set.h"
35 #include "sort-util.h"
36 #include "stat-util.h"
37 #include "string-table.h"
38 #include "string-util.h"
39 #include "strv.h"
40 #include "sync-util.h"
41 #include "user-util.h"
42 #include "xattr-util.h"
43
44 #define DEFAULT_DATA_HASH_TABLE_SIZE (2047ULL*sizeof(HashItem))
45 #define DEFAULT_FIELD_HASH_TABLE_SIZE (333ULL*sizeof(HashItem))
46
47 #define DEFAULT_COMPRESS_THRESHOLD (512ULL)
48 #define MIN_COMPRESS_THRESHOLD (8ULL)
49
50 /* This is the minimum journal file size */
51 #define JOURNAL_FILE_SIZE_MIN (512 * 1024ULL) /* 512 KiB */
52 #define JOURNAL_COMPACT_SIZE_MAX UINT32_MAX /* 4 GiB */
53
54 /* These are the lower and upper bounds if we deduce the max_use value
55 * from the file system size */
56 #define MAX_USE_LOWER (1 * 1024 * 1024ULL) /* 1 MiB */
57 #define MAX_USE_UPPER (4 * 1024 * 1024 * 1024ULL) /* 4 GiB */
58
59 /* Those are the lower and upper bounds for the minimal use limit,
60 * i.e. how much we'll use even if keep_free suggests otherwise. */
61 #define MIN_USE_LOW (1 * 1024 * 1024ULL) /* 1 MiB */
62 #define MIN_USE_HIGH (16 * 1024 * 1024ULL) /* 16 MiB */
63
64 /* This is the upper bound if we deduce max_size from max_use */
65 #define MAX_SIZE_UPPER (128 * 1024 * 1024ULL) /* 128 MiB */
66
67 /* This is the upper bound if we deduce the keep_free value from the
68 * file system size */
69 #define KEEP_FREE_UPPER (4 * 1024 * 1024 * 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 (1024 * 1024ULL) /* 1 MB */
74
75 /* This is the default maximum number of journal files to keep around. */
76 #define DEFAULT_N_MAX_FILES 100
77
78 /* n_data was the first entry we added after the initial file format design */
79 #define HEADER_SIZE_MIN ALIGN64(offsetof(Header, n_data))
80
81 /* How many entries to keep in the entry array chain cache at max */
82 #define CHAIN_CACHE_MAX 20
83
84 /* How much to increase the journal file size at once each time we allocate something new. */
85 #define FILE_SIZE_INCREASE (8 * 1024 * 1024ULL) /* 8MB */
86
87 /* Reread fstat() of the file for detecting deletions at least this often */
88 #define LAST_STAT_REFRESH_USEC (5*USEC_PER_SEC)
89
90 /* The mmap context to use for the header we pick as one above the last defined typed */
91 #define CONTEXT_HEADER _OBJECT_TYPE_MAX
92
93 /* Longest hash chain to rotate after */
94 #define HASH_CHAIN_DEPTH_MAX 100
95
96 #ifdef __clang__
97 # pragma GCC diagnostic ignored "-Waddress-of-packed-member"
98 #endif
99
100 static int mmap_prot_from_open_flags(int flags) {
101 switch (flags & O_ACCMODE) {
102 case O_RDONLY:
103 return PROT_READ;
104 case O_WRONLY:
105 return PROT_WRITE;
106 case O_RDWR:
107 return PROT_READ|PROT_WRITE;
108 default:
109 assert_not_reached();
110 }
111 }
112
113 int journal_file_tail_end_by_pread(JournalFile *f, uint64_t *ret_offset) {
114 uint64_t p;
115 int r;
116
117 assert(f);
118 assert(f->header);
119 assert(ret_offset);
120
121 /* Same as journal_file_tail_end_by_mmap() below, but operates with pread() to avoid the mmap cache
122 * (and thus is thread safe) */
123
124 p = le64toh(f->header->tail_object_offset);
125 if (p == 0)
126 p = le64toh(f->header->header_size);
127 else {
128 Object tail;
129 uint64_t sz;
130
131 r = journal_file_read_object_header(f, OBJECT_UNUSED, p, &tail);
132 if (r < 0)
133 return r;
134
135 sz = le64toh(tail.object.size);
136 if (sz > UINT64_MAX - sizeof(uint64_t) + 1)
137 return -EBADMSG;
138
139 sz = ALIGN64(sz);
140 if (p > UINT64_MAX - sz)
141 return -EBADMSG;
142
143 p += sz;
144 }
145
146 *ret_offset = p;
147
148 return 0;
149 }
150
151 int journal_file_tail_end_by_mmap(JournalFile *f, uint64_t *ret_offset) {
152 uint64_t p;
153 int r;
154
155 assert(f);
156 assert(f->header);
157 assert(ret_offset);
158
159 /* Same as journal_file_tail_end_by_pread() above, but operates with the usual mmap logic */
160
161 p = le64toh(f->header->tail_object_offset);
162 if (p == 0)
163 p = le64toh(f->header->header_size);
164 else {
165 Object *tail;
166 uint64_t sz;
167
168 r = journal_file_move_to_object(f, OBJECT_UNUSED, p, &tail);
169 if (r < 0)
170 return r;
171
172 sz = le64toh(READ_NOW(tail->object.size));
173 if (sz > UINT64_MAX - sizeof(uint64_t) + 1)
174 return -EBADMSG;
175
176 sz = ALIGN64(sz);
177 if (p > UINT64_MAX - sz)
178 return -EBADMSG;
179
180 p += sz;
181 }
182
183 *ret_offset = p;
184
185 return 0;
186 }
187
188 int journal_file_set_offline_thread_join(JournalFile *f) {
189 int r;
190
191 assert(f);
192
193 if (f->offline_state == OFFLINE_JOINED)
194 return 0;
195
196 r = pthread_join(f->offline_thread, NULL);
197 if (r)
198 return -r;
199
200 f->offline_state = OFFLINE_JOINED;
201
202 if (mmap_cache_fd_got_sigbus(f->cache_fd))
203 return -EIO;
204
205 return 0;
206 }
207
208 static int journal_file_set_online(JournalFile *f) {
209 bool wait = true;
210
211 assert(f);
212
213 if (!journal_file_writable(f))
214 return -EPERM;
215
216 if (f->fd < 0 || !f->header)
217 return -EINVAL;
218
219 while (wait) {
220 switch (f->offline_state) {
221 case OFFLINE_JOINED:
222 /* No offline thread, no need to wait. */
223 wait = false;
224 break;
225
226 case OFFLINE_SYNCING: {
227 OfflineState tmp_state = OFFLINE_SYNCING;
228 if (!__atomic_compare_exchange_n(&f->offline_state, &tmp_state, OFFLINE_CANCEL,
229 false, __ATOMIC_SEQ_CST, __ATOMIC_SEQ_CST))
230 continue;
231 }
232 /* Canceled syncing prior to offlining, no need to wait. */
233 wait = false;
234 break;
235
236 case OFFLINE_AGAIN_FROM_SYNCING: {
237 OfflineState tmp_state = OFFLINE_AGAIN_FROM_SYNCING;
238 if (!__atomic_compare_exchange_n(&f->offline_state, &tmp_state, OFFLINE_CANCEL,
239 false, __ATOMIC_SEQ_CST, __ATOMIC_SEQ_CST))
240 continue;
241 }
242 /* Canceled restart from syncing, no need to wait. */
243 wait = false;
244 break;
245
246 case OFFLINE_AGAIN_FROM_OFFLINING: {
247 OfflineState tmp_state = OFFLINE_AGAIN_FROM_OFFLINING;
248 if (!__atomic_compare_exchange_n(&f->offline_state, &tmp_state, OFFLINE_CANCEL,
249 false, __ATOMIC_SEQ_CST, __ATOMIC_SEQ_CST))
250 continue;
251 }
252 /* Canceled restart from offlining, must wait for offlining to complete however. */
253 _fallthrough_;
254 default: {
255 int r;
256
257 r = journal_file_set_offline_thread_join(f);
258 if (r < 0)
259 return r;
260
261 wait = false;
262 break;
263 }
264 }
265 }
266
267 if (mmap_cache_fd_got_sigbus(f->cache_fd))
268 return -EIO;
269
270 switch (f->header->state) {
271 case STATE_ONLINE:
272 return 0;
273
274 case STATE_OFFLINE:
275 f->header->state = STATE_ONLINE;
276 (void) fsync(f->fd);
277 return 0;
278
279 default:
280 return -EINVAL;
281 }
282 }
283
284 JournalFile* journal_file_close(JournalFile *f) {
285 if (!f)
286 return NULL;
287
288 assert(f->newest_boot_id_prioq_idx == PRIOQ_IDX_NULL);
289
290 if (f->cache_fd)
291 mmap_cache_fd_free(f->cache_fd);
292
293 if (f->close_fd)
294 safe_close(f->fd);
295 free(f->path);
296
297 ordered_hashmap_free_free(f->chain_cache);
298
299 #if HAVE_COMPRESSION
300 free(f->compress_buffer);
301 #endif
302
303 #if HAVE_GCRYPT
304 if (f->fss_file)
305 munmap(f->fss_file, PAGE_ALIGN(f->fss_file_size));
306 else
307 free(f->fsprg_state);
308
309 free(f->fsprg_seed);
310
311 if (f->hmac)
312 gcry_md_close(f->hmac);
313 #endif
314
315 return mfree(f);
316 }
317
318 static bool keyed_hash_requested(void) {
319 static thread_local int cached = -1;
320 int r;
321
322 if (cached < 0) {
323 r = getenv_bool("SYSTEMD_JOURNAL_KEYED_HASH");
324 if (r < 0) {
325 if (r != -ENXIO)
326 log_debug_errno(r, "Failed to parse $SYSTEMD_JOURNAL_KEYED_HASH environment variable, ignoring: %m");
327 cached = true;
328 } else
329 cached = r;
330 }
331
332 return cached;
333 }
334
335 static bool compact_mode_requested(void) {
336 static thread_local int cached = -1;
337 int r;
338
339 if (cached < 0) {
340 r = getenv_bool("SYSTEMD_JOURNAL_COMPACT");
341 if (r < 0) {
342 if (r != -ENXIO)
343 log_debug_errno(r, "Failed to parse $SYSTEMD_JOURNAL_COMPACT environment variable, ignoring: %m");
344 cached = true;
345 } else
346 cached = r;
347 }
348
349 return cached;
350 }
351
352 #if HAVE_COMPRESSION
353 static Compression getenv_compression(void) {
354 Compression c;
355 const char *e;
356 int r;
357
358 e = getenv("SYSTEMD_JOURNAL_COMPRESS");
359 if (!e)
360 return DEFAULT_COMPRESSION;
361
362 r = parse_boolean(e);
363 if (r >= 0)
364 return r ? DEFAULT_COMPRESSION : COMPRESSION_NONE;
365
366 c = compression_from_string(e);
367 if (c < 0) {
368 log_debug_errno(c, "Failed to parse SYSTEMD_JOURNAL_COMPRESS value, ignoring: %s", e);
369 return DEFAULT_COMPRESSION;
370 }
371
372 if (!compression_supported(c)) {
373 log_debug("Unsupported compression algorithm specified, ignoring: %s", e);
374 return DEFAULT_COMPRESSION;
375 }
376
377 return c;
378 }
379 #endif
380
381 static Compression compression_requested(void) {
382 #if HAVE_COMPRESSION
383 static thread_local Compression cached = _COMPRESSION_INVALID;
384
385 if (cached < 0)
386 cached = getenv_compression();
387
388 return cached;
389 #else
390 return COMPRESSION_NONE;
391 #endif
392 }
393
394 static int journal_file_init_header(
395 JournalFile *f,
396 JournalFileFlags file_flags,
397 JournalFile *template) {
398
399 bool seal = false;
400 ssize_t k;
401 int r;
402
403 assert(f);
404
405 #if HAVE_GCRYPT
406 /* Try to load the FSPRG state, and if we can't, then just don't do sealing */
407 seal = FLAGS_SET(file_flags, JOURNAL_SEAL) && journal_file_fss_load(f) >= 0;
408 #endif
409
410 Header h = {
411 .header_size = htole64(ALIGN64(sizeof(h))),
412 .incompatible_flags = htole32(
413 FLAGS_SET(file_flags, JOURNAL_COMPRESS) * COMPRESSION_TO_HEADER_INCOMPATIBLE_FLAG(compression_requested()) |
414 keyed_hash_requested() * HEADER_INCOMPATIBLE_KEYED_HASH |
415 compact_mode_requested() * HEADER_INCOMPATIBLE_COMPACT),
416 .compatible_flags = htole32(
417 (seal * HEADER_COMPATIBLE_SEALED) |
418 HEADER_COMPATIBLE_TAIL_ENTRY_BOOT_ID),
419 };
420
421 assert_cc(sizeof(h.signature) == sizeof(HEADER_SIGNATURE));
422 memcpy(h.signature, HEADER_SIGNATURE, sizeof(HEADER_SIGNATURE));
423
424 r = sd_id128_randomize(&h.file_id);
425 if (r < 0)
426 return r;
427
428 r = sd_id128_get_machine(&h.machine_id);
429 if (r < 0 && !ERRNO_IS_MACHINE_ID_UNSET(r))
430 return r; /* If we have no valid machine ID (test environment?), let's simply leave the
431 * machine ID field all zeroes. */
432
433 if (template) {
434 h.seqnum_id = template->header->seqnum_id;
435 h.tail_entry_seqnum = template->header->tail_entry_seqnum;
436 } else
437 h.seqnum_id = h.file_id;
438
439 k = pwrite(f->fd, &h, sizeof(h), 0);
440 if (k < 0)
441 return -errno;
442 if (k != sizeof(h))
443 return -EIO;
444
445 return 0;
446 }
447
448 static int journal_file_refresh_header(JournalFile *f) {
449 int r;
450
451 assert(f);
452 assert(f->header);
453
454 /* We used to update the header's boot ID field here, but we don't do that anymore, as per
455 * HEADER_COMPATIBLE_TAIL_ENTRY_BOOT_ID */
456
457 r = journal_file_set_online(f);
458
459 /* Sync the online state to disk; likely just created a new file, also sync the directory this file
460 * is located in. */
461 (void) fsync_full(f->fd);
462
463 return r;
464 }
465
466 static bool warn_wrong_flags(const JournalFile *f, bool compatible) {
467 const uint32_t any = compatible ? HEADER_COMPATIBLE_ANY : HEADER_INCOMPATIBLE_ANY,
468 supported = compatible ? HEADER_COMPATIBLE_SUPPORTED : HEADER_INCOMPATIBLE_SUPPORTED;
469 const char *type = compatible ? "compatible" : "incompatible";
470 uint32_t flags;
471
472 assert(f);
473 assert(f->header);
474
475 flags = le32toh(compatible ? f->header->compatible_flags : f->header->incompatible_flags);
476
477 if (flags & ~supported) {
478 if (flags & ~any)
479 log_debug("Journal file %s has unknown %s flags 0x%"PRIx32,
480 f->path, type, flags & ~any);
481 flags = (flags & any) & ~supported;
482 if (flags) {
483 const char* strv[6];
484 size_t n = 0;
485 _cleanup_free_ char *t = NULL;
486
487 if (compatible) {
488 if (flags & HEADER_COMPATIBLE_SEALED)
489 strv[n++] = "sealed";
490 } else {
491 if (flags & HEADER_INCOMPATIBLE_COMPRESSED_XZ)
492 strv[n++] = "xz-compressed";
493 if (flags & HEADER_INCOMPATIBLE_COMPRESSED_LZ4)
494 strv[n++] = "lz4-compressed";
495 if (flags & HEADER_INCOMPATIBLE_COMPRESSED_ZSTD)
496 strv[n++] = "zstd-compressed";
497 if (flags & HEADER_INCOMPATIBLE_KEYED_HASH)
498 strv[n++] = "keyed-hash";
499 if (flags & HEADER_INCOMPATIBLE_COMPACT)
500 strv[n++] = "compact";
501 }
502 strv[n] = NULL;
503 assert(n < ELEMENTSOF(strv));
504
505 t = strv_join((char**) strv, ", ");
506 log_debug("Journal file %s uses %s %s %s disabled at compilation time.",
507 f->path, type, n > 1 ? "flags" : "flag", strnull(t));
508 }
509 return true;
510 }
511
512 return false;
513 }
514
515 static bool offset_is_valid(uint64_t offset, uint64_t header_size, uint64_t tail_object_offset) {
516 if (offset == 0)
517 return true;
518 if (!VALID64(offset))
519 return false;
520 if (offset < header_size)
521 return false;
522 if (offset > tail_object_offset)
523 return false;
524 return true;
525 }
526
527 static bool hash_table_is_valid(uint64_t offset, uint64_t size, uint64_t header_size, uint64_t arena_size, uint64_t tail_object_offset) {
528 if ((offset == 0) != (size == 0))
529 return false;
530 if (offset == 0)
531 return true;
532 if (offset <= offsetof(Object, hash_table.items))
533 return false;
534 offset -= offsetof(Object, hash_table.items);
535 if (!offset_is_valid(offset, header_size, tail_object_offset))
536 return false;
537 assert(offset <= header_size + arena_size);
538 if (size > header_size + arena_size - offset)
539 return false;
540 return true;
541 }
542
543 static int journal_file_verify_header(JournalFile *f) {
544 uint64_t arena_size, header_size;
545
546 assert(f);
547 assert(f->header);
548
549 if (memcmp(f->header->signature, HEADER_SIGNATURE, 8))
550 return -EBADMSG;
551
552 /* In both read and write mode we refuse to open files with incompatible
553 * flags we don't know. */
554 if (warn_wrong_flags(f, false))
555 return -EPROTONOSUPPORT;
556
557 /* When open for writing we refuse to open files with compatible flags, too. */
558 if (journal_file_writable(f) && warn_wrong_flags(f, true))
559 return -EPROTONOSUPPORT;
560
561 if (f->header->state >= _STATE_MAX)
562 return -EBADMSG;
563
564 header_size = le64toh(READ_NOW(f->header->header_size));
565
566 /* The first addition was n_data, so check that we are at least this large */
567 if (header_size < HEADER_SIZE_MIN)
568 return -EBADMSG;
569
570 /* When open for writing we refuse to open files with a mismatch of the header size, i.e. writing to
571 * files implementing older or new header structures. */
572 if (journal_file_writable(f) && header_size != sizeof(Header))
573 return -EPROTONOSUPPORT;
574
575 if (JOURNAL_HEADER_SEALED(f->header) && !JOURNAL_HEADER_CONTAINS(f->header, n_entry_arrays))
576 return -EBADMSG;
577
578 arena_size = le64toh(READ_NOW(f->header->arena_size));
579
580 if (UINT64_MAX - header_size < arena_size || header_size + arena_size > (uint64_t) f->last_stat.st_size)
581 return -ENODATA;
582
583 uint64_t tail_object_offset = le64toh(f->header->tail_object_offset);
584 if (!offset_is_valid(tail_object_offset, header_size, UINT64_MAX))
585 return -ENODATA;
586 if (header_size + arena_size < tail_object_offset)
587 return -ENODATA;
588 if (header_size + arena_size - tail_object_offset < sizeof(ObjectHeader))
589 return -ENODATA;
590
591 if (!hash_table_is_valid(le64toh(f->header->data_hash_table_offset),
592 le64toh(f->header->data_hash_table_size),
593 header_size, arena_size, tail_object_offset))
594 return -ENODATA;
595
596 if (!hash_table_is_valid(le64toh(f->header->field_hash_table_offset),
597 le64toh(f->header->field_hash_table_size),
598 header_size, arena_size, tail_object_offset))
599 return -ENODATA;
600
601 uint64_t entry_array_offset = le64toh(f->header->entry_array_offset);
602 if (!offset_is_valid(entry_array_offset, header_size, tail_object_offset))
603 return -ENODATA;
604
605 if (JOURNAL_HEADER_CONTAINS(f->header, tail_entry_array_offset)) {
606 uint32_t offset = le32toh(f->header->tail_entry_array_offset);
607 uint32_t n = le32toh(f->header->tail_entry_array_n_entries);
608
609 if (!offset_is_valid(offset, header_size, tail_object_offset))
610 return -ENODATA;
611 if (entry_array_offset > offset)
612 return -ENODATA;
613 if (entry_array_offset == 0 && offset != 0)
614 return -ENODATA;
615 if ((offset == 0) != (n == 0))
616 return -ENODATA;
617 assert(offset <= header_size + arena_size);
618 if ((uint64_t) n * journal_file_entry_array_item_size(f) > header_size + arena_size - offset)
619 return -ENODATA;
620 }
621
622 if (JOURNAL_HEADER_CONTAINS(f->header, tail_entry_offset))
623 if (!offset_is_valid(le64toh(f->header->tail_entry_offset), header_size, tail_object_offset))
624 return -ENODATA;
625
626 /* Verify number of objects */
627 uint64_t n_objects = le64toh(f->header->n_objects);
628 if (n_objects > arena_size / sizeof(ObjectHeader))
629 return -ENODATA;
630
631 uint64_t n_entries = le64toh(f->header->n_entries);
632 if (n_entries > n_objects)
633 return -ENODATA;
634
635 if (JOURNAL_HEADER_CONTAINS(f->header, n_data) &&
636 le64toh(f->header->n_data) > n_objects)
637 return -ENODATA;
638
639 if (JOURNAL_HEADER_CONTAINS(f->header, n_fields) &&
640 le64toh(f->header->n_fields) > n_objects)
641 return -ENODATA;
642
643 if (JOURNAL_HEADER_CONTAINS(f->header, n_tags) &&
644 le64toh(f->header->n_tags) > n_objects)
645 return -ENODATA;
646
647 if (JOURNAL_HEADER_CONTAINS(f->header, n_entry_arrays) &&
648 le64toh(f->header->n_entry_arrays) > n_objects)
649 return -ENODATA;
650
651 if (JOURNAL_HEADER_CONTAINS(f->header, tail_entry_array_n_entries) &&
652 le32toh(f->header->tail_entry_array_n_entries) > n_entries)
653 return -ENODATA;
654
655 if (journal_file_writable(f)) {
656 sd_id128_t machine_id;
657 uint8_t state;
658 int r;
659
660 r = sd_id128_get_machine(&machine_id);
661 if (r < 0) {
662 if (!ERRNO_IS_MACHINE_ID_UNSET(r)) /* handle graceful if machine ID is not initialized yet */
663 return r;
664
665 machine_id = SD_ID128_NULL;
666 }
667
668 if (!sd_id128_equal(machine_id, f->header->machine_id))
669 return log_debug_errno(SYNTHETIC_ERRNO(EHOSTDOWN),
670 "Trying to open journal file from different host for writing, refusing.");
671
672 state = f->header->state;
673
674 if (state == STATE_ARCHIVED)
675 return -ESHUTDOWN; /* Already archived */
676 if (state == STATE_ONLINE)
677 return log_debug_errno(SYNTHETIC_ERRNO(EBUSY),
678 "Journal file %s is already online. Assuming unclean closing.",
679 f->path);
680 if (state != STATE_OFFLINE)
681 return log_debug_errno(SYNTHETIC_ERRNO(EBUSY),
682 "Journal file %s has unknown state %i.",
683 f->path, state);
684
685 if (f->header->field_hash_table_size == 0 || f->header->data_hash_table_size == 0)
686 return -EBADMSG;
687 }
688
689 return 0;
690 }
691
692 int journal_file_fstat(JournalFile *f) {
693 int r;
694
695 assert(f);
696 assert(f->fd >= 0);
697
698 if (fstat(f->fd, &f->last_stat) < 0)
699 return -errno;
700
701 f->last_stat_usec = now(CLOCK_MONOTONIC);
702
703 /* Refuse dealing with files that aren't regular */
704 r = stat_verify_regular(&f->last_stat);
705 if (r < 0)
706 return r;
707
708 /* Refuse appending to files that are already deleted */
709 if (f->last_stat.st_nlink <= 0)
710 return -EIDRM;
711
712 return 0;
713 }
714
715 static int journal_file_allocate(JournalFile *f, uint64_t offset, uint64_t size) {
716 uint64_t old_size, new_size, old_header_size, old_arena_size;
717 int r;
718
719 assert(f);
720 assert(f->header);
721
722 /* We assume that this file is not sparse, and we know that for sure, since we always call
723 * posix_fallocate() ourselves */
724
725 if (size > PAGE_ALIGN_DOWN(UINT64_MAX) - offset)
726 return -EINVAL;
727
728 if (mmap_cache_fd_got_sigbus(f->cache_fd))
729 return -EIO;
730
731 old_header_size = le64toh(READ_NOW(f->header->header_size));
732 old_arena_size = le64toh(READ_NOW(f->header->arena_size));
733 if (old_arena_size > PAGE_ALIGN_DOWN(UINT64_MAX) - old_header_size)
734 return -EBADMSG;
735
736 old_size = old_header_size + old_arena_size;
737
738 new_size = MAX(PAGE_ALIGN(offset + size), old_header_size);
739
740 if (new_size <= old_size) {
741
742 /* We already pre-allocated enough space, but before
743 * we write to it, let's check with fstat() if the
744 * file got deleted, in order make sure we don't throw
745 * away the data immediately. Don't check fstat() for
746 * all writes though, but only once ever 10s. */
747
748 if (f->last_stat_usec + LAST_STAT_REFRESH_USEC > now(CLOCK_MONOTONIC))
749 return 0;
750
751 return journal_file_fstat(f);
752 }
753
754 /* Allocate more space. */
755
756 if (f->metrics.max_size > 0 && new_size > f->metrics.max_size)
757 return -E2BIG;
758
759 /* Refuse to go over 4G in compact mode so offsets can be stored in 32-bit. */
760 if (JOURNAL_HEADER_COMPACT(f->header) && new_size > UINT32_MAX)
761 return -E2BIG;
762
763 if (new_size > f->metrics.min_size && f->metrics.keep_free > 0) {
764 struct statvfs svfs;
765
766 if (fstatvfs(f->fd, &svfs) >= 0) {
767 uint64_t available;
768
769 available = LESS_BY((uint64_t) svfs.f_bfree * (uint64_t) svfs.f_bsize, f->metrics.keep_free);
770
771 if (new_size - old_size > available)
772 return -E2BIG;
773 }
774 }
775
776 /* Increase by larger blocks at once */
777 new_size = ROUND_UP(new_size, FILE_SIZE_INCREASE);
778 if (f->metrics.max_size > 0 && new_size > f->metrics.max_size)
779 new_size = f->metrics.max_size;
780
781 /* Note that the glibc fallocate() fallback is very
782 inefficient, hence we try to minimize the allocation area
783 as we can. */
784 r = posix_fallocate_loop(f->fd, old_size, new_size - old_size);
785 if (r < 0)
786 return r;
787
788 f->header->arena_size = htole64(new_size - old_header_size);
789
790 return journal_file_fstat(f);
791 }
792
793 static unsigned type_to_context(ObjectType type) {
794 /* One context for each type, plus one catch-all for the rest */
795 assert_cc(_OBJECT_TYPE_MAX <= MMAP_CACHE_MAX_CONTEXTS);
796 assert_cc(CONTEXT_HEADER < MMAP_CACHE_MAX_CONTEXTS);
797 return type > OBJECT_UNUSED && type < _OBJECT_TYPE_MAX ? type : 0;
798 }
799
800 static int journal_file_move_to(
801 JournalFile *f,
802 ObjectType type,
803 bool keep_always,
804 uint64_t offset,
805 uint64_t size,
806 void **ret) {
807
808 int r;
809
810 assert(f);
811 assert(ret);
812
813 /* This function may clear, overwrite, or alter previously cached entries. After this function has
814 * been called, all objects except for one obtained by this function are invalidated and must be
815 * re-read before use. */
816
817 if (size <= 0)
818 return -EINVAL;
819
820 if (size > UINT64_MAX - offset)
821 return -EBADMSG;
822
823 /* Avoid SIGBUS on invalid accesses */
824 if (offset + size > (uint64_t) f->last_stat.st_size) {
825 /* Hmm, out of range? Let's refresh the fstat() data
826 * first, before we trust that check. */
827
828 r = journal_file_fstat(f);
829 if (r < 0)
830 return r;
831
832 if (offset + size > (uint64_t) f->last_stat.st_size)
833 return -EADDRNOTAVAIL;
834 }
835
836 return mmap_cache_fd_get(f->cache_fd, type_to_context(type), keep_always, offset, size, &f->last_stat, ret);
837 }
838
839 static uint64_t minimum_header_size(JournalFile *f, Object *o) {
840
841 static const uint64_t table[] = {
842 [OBJECT_DATA] = sizeof(DataObject),
843 [OBJECT_FIELD] = sizeof(FieldObject),
844 [OBJECT_ENTRY] = sizeof(EntryObject),
845 [OBJECT_DATA_HASH_TABLE] = sizeof(HashTableObject),
846 [OBJECT_FIELD_HASH_TABLE] = sizeof(HashTableObject),
847 [OBJECT_ENTRY_ARRAY] = sizeof(EntryArrayObject),
848 [OBJECT_TAG] = sizeof(TagObject),
849 };
850
851 assert(f);
852 assert(o);
853
854 if (o->object.type == OBJECT_DATA)
855 return journal_file_data_payload_offset(f);
856
857 if (o->object.type >= ELEMENTSOF(table) || table[o->object.type] <= 0)
858 return sizeof(ObjectHeader);
859
860 return table[o->object.type];
861 }
862
863 static int check_object_header(JournalFile *f, Object *o, ObjectType type, uint64_t offset) {
864 uint64_t s;
865
866 assert(f);
867 assert(o);
868
869 s = le64toh(READ_NOW(o->object.size));
870 if (s == 0)
871 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG),
872 "Attempt to move to uninitialized object: %" PRIu64,
873 offset);
874
875 if (s < sizeof(ObjectHeader))
876 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG),
877 "Attempt to move to overly short object with size %"PRIu64": %" PRIu64,
878 s, offset);
879
880 if (o->object.type <= OBJECT_UNUSED || o->object.type >= _OBJECT_TYPE_MAX)
881 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG),
882 "Attempt to move to object with invalid type (%u): %" PRIu64,
883 o->object.type, offset);
884
885 if (type > OBJECT_UNUSED && o->object.type != type)
886 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG),
887 "Found %s object while expecting %s object: %" PRIu64,
888 journal_object_type_to_string(o->object.type),
889 journal_object_type_to_string(type),
890 offset);
891
892 if (s < minimum_header_size(f, o))
893 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG),
894 "Size of %s object (%"PRIu64") is smaller than the minimum object size (%"PRIu64"): %" PRIu64,
895 journal_object_type_to_string(o->object.type),
896 s,
897 minimum_header_size(f, o),
898 offset);
899
900 return 0;
901 }
902
903 /* Lightweight object checks. We want this to be fast, so that we won't
904 * slowdown every journal_file_move_to_object() call too much. */
905 static int check_object(JournalFile *f, Object *o, uint64_t offset) {
906 assert(f);
907 assert(o);
908
909 switch (o->object.type) {
910
911 case OBJECT_DATA:
912 if ((le64toh(o->data.entry_offset) == 0) ^ (le64toh(o->data.n_entries) == 0))
913 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG),
914 "Bad data n_entries: %" PRIu64 ": %" PRIu64,
915 le64toh(o->data.n_entries),
916 offset);
917
918 if (le64toh(o->object.size) <= journal_file_data_payload_offset(f))
919 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG),
920 "Bad data size (<= %zu): %" PRIu64 ": %" PRIu64,
921 journal_file_data_payload_offset(f),
922 le64toh(o->object.size),
923 offset);
924
925 if (!VALID64(le64toh(o->data.next_hash_offset)) ||
926 !VALID64(le64toh(o->data.next_field_offset)) ||
927 !VALID64(le64toh(o->data.entry_offset)) ||
928 !VALID64(le64toh(o->data.entry_array_offset)))
929 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG),
930 "Invalid offset, next_hash_offset=" OFSfmt ", next_field_offset=" OFSfmt ", entry_offset=" OFSfmt ", entry_array_offset=" OFSfmt ": %" PRIu64,
931 le64toh(o->data.next_hash_offset),
932 le64toh(o->data.next_field_offset),
933 le64toh(o->data.entry_offset),
934 le64toh(o->data.entry_array_offset),
935 offset);
936
937 break;
938
939 case OBJECT_FIELD:
940 if (le64toh(o->object.size) <= offsetof(Object, field.payload))
941 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG),
942 "Bad field size (<= %zu): %" PRIu64 ": %" PRIu64,
943 offsetof(Object, field.payload),
944 le64toh(o->object.size),
945 offset);
946
947 if (!VALID64(le64toh(o->field.next_hash_offset)) ||
948 !VALID64(le64toh(o->field.head_data_offset)))
949 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG),
950 "Invalid offset, next_hash_offset=" OFSfmt ", head_data_offset=" OFSfmt ": %" PRIu64,
951 le64toh(o->field.next_hash_offset),
952 le64toh(o->field.head_data_offset),
953 offset);
954 break;
955
956 case OBJECT_ENTRY: {
957 uint64_t sz;
958
959 sz = le64toh(READ_NOW(o->object.size));
960 if (sz < offsetof(Object, entry.items) ||
961 (sz - offsetof(Object, entry.items)) % journal_file_entry_item_size(f) != 0)
962 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG),
963 "Bad entry size (<= %zu): %" PRIu64 ": %" PRIu64,
964 offsetof(Object, entry.items),
965 sz,
966 offset);
967
968 if ((sz - offsetof(Object, entry.items)) / journal_file_entry_item_size(f) <= 0)
969 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG),
970 "Invalid number items in entry: %" PRIu64 ": %" PRIu64,
971 (sz - offsetof(Object, entry.items)) / journal_file_entry_item_size(f),
972 offset);
973
974 if (le64toh(o->entry.seqnum) <= 0)
975 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG),
976 "Invalid entry seqnum: %" PRIx64 ": %" PRIu64,
977 le64toh(o->entry.seqnum),
978 offset);
979
980 if (!VALID_REALTIME(le64toh(o->entry.realtime)))
981 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG),
982 "Invalid entry realtime timestamp: %" PRIu64 ": %" PRIu64,
983 le64toh(o->entry.realtime),
984 offset);
985
986 if (!VALID_MONOTONIC(le64toh(o->entry.monotonic)))
987 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG),
988 "Invalid entry monotonic timestamp: %" PRIu64 ": %" PRIu64,
989 le64toh(o->entry.monotonic),
990 offset);
991
992 break;
993 }
994
995 case OBJECT_DATA_HASH_TABLE:
996 case OBJECT_FIELD_HASH_TABLE: {
997 uint64_t sz;
998
999 sz = le64toh(READ_NOW(o->object.size));
1000 if (sz < offsetof(Object, hash_table.items) ||
1001 (sz - offsetof(Object, hash_table.items)) % sizeof(HashItem) != 0 ||
1002 (sz - offsetof(Object, hash_table.items)) / sizeof(HashItem) <= 0)
1003 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG),
1004 "Invalid %s hash table size: %" PRIu64 ": %" PRIu64,
1005 journal_object_type_to_string(o->object.type),
1006 sz,
1007 offset);
1008
1009 break;
1010 }
1011
1012 case OBJECT_ENTRY_ARRAY: {
1013 uint64_t sz, next;
1014
1015 sz = le64toh(READ_NOW(o->object.size));
1016 if (sz < offsetof(Object, entry_array.items) ||
1017 (sz - offsetof(Object, entry_array.items)) % journal_file_entry_array_item_size(f) != 0 ||
1018 (sz - offsetof(Object, entry_array.items)) / journal_file_entry_array_item_size(f) <= 0)
1019 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG),
1020 "Invalid object entry array size: %" PRIu64 ": %" PRIu64,
1021 sz,
1022 offset);
1023 /* Here, we request that the offset of each entry array object is in strictly increasing order. */
1024 next = le64toh(o->entry_array.next_entry_array_offset);
1025 if (!VALID64(next) || (next > 0 && next <= offset))
1026 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG),
1027 "Invalid object entry array next_entry_array_offset: %" PRIu64 ": %" PRIu64,
1028 next,
1029 offset);
1030
1031 break;
1032 }
1033
1034 case OBJECT_TAG:
1035 if (le64toh(o->object.size) != sizeof(TagObject))
1036 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG),
1037 "Invalid object tag size: %" PRIu64 ": %" PRIu64,
1038 le64toh(o->object.size),
1039 offset);
1040
1041 if (!VALID_EPOCH(le64toh(o->tag.epoch)))
1042 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG),
1043 "Invalid object tag epoch: %" PRIu64 ": %" PRIu64,
1044 le64toh(o->tag.epoch), offset);
1045
1046 break;
1047 }
1048
1049 return 0;
1050 }
1051
1052 int journal_file_move_to_object(JournalFile *f, ObjectType type, uint64_t offset, Object **ret) {
1053 int r;
1054 Object *o;
1055
1056 assert(f);
1057
1058 /* Even if this function fails, it may clear, overwrite, or alter previously cached entries. After
1059 * this function has been called, all objects except for one obtained by this function are
1060 * invalidated and must be re-read before use.. */
1061
1062 /* Objects may only be located at multiple of 64 bit */
1063 if (!VALID64(offset))
1064 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG),
1065 "Attempt to move to %s object at non-64-bit boundary: %" PRIu64,
1066 journal_object_type_to_string(type),
1067 offset);
1068
1069 /* Object may not be located in the file header */
1070 if (offset < le64toh(f->header->header_size))
1071 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG),
1072 "Attempt to move to %s object located in file header: %" PRIu64,
1073 journal_object_type_to_string(type),
1074 offset);
1075
1076 r = journal_file_move_to(f, type, false, offset, sizeof(ObjectHeader), (void**) &o);
1077 if (r < 0)
1078 return r;
1079
1080 r = check_object_header(f, o, type, offset);
1081 if (r < 0)
1082 return r;
1083
1084 r = journal_file_move_to(f, type, false, offset, le64toh(READ_NOW(o->object.size)), (void**) &o);
1085 if (r < 0)
1086 return r;
1087
1088 r = check_object_header(f, o, type, offset);
1089 if (r < 0)
1090 return r;
1091
1092 r = check_object(f, o, offset);
1093 if (r < 0)
1094 return r;
1095
1096 if (ret)
1097 *ret = o;
1098
1099 return 0;
1100 }
1101
1102 int journal_file_read_object_header(JournalFile *f, ObjectType type, uint64_t offset, Object *ret) {
1103 ssize_t n;
1104 Object o;
1105 int r;
1106
1107 assert(f);
1108
1109 /* Objects may only be located at multiple of 64 bit */
1110 if (!VALID64(offset))
1111 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG),
1112 "Attempt to read %s object at non-64-bit boundary: %" PRIu64,
1113 journal_object_type_to_string(type), offset);
1114
1115 /* Object may not be located in the file header */
1116 if (offset < le64toh(f->header->header_size))
1117 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG),
1118 "Attempt to read %s object located in file header: %" PRIu64,
1119 journal_object_type_to_string(type), offset);
1120
1121 /* This will likely read too much data but it avoids having to call pread() twice. */
1122 n = pread(f->fd, &o, sizeof(o), offset);
1123 if (n < 0)
1124 return log_debug_errno(errno, "Failed to read journal %s object at offset: %" PRIu64,
1125 journal_object_type_to_string(type), offset);
1126
1127 if ((size_t) n < sizeof(o.object))
1128 return log_debug_errno(SYNTHETIC_ERRNO(EIO),
1129 "Failed to read short %s object at offset: %" PRIu64,
1130 journal_object_type_to_string(type), offset);
1131
1132 r = check_object_header(f, &o, type, offset);
1133 if (r < 0)
1134 return r;
1135
1136 if ((size_t) n < minimum_header_size(f, &o))
1137 return log_debug_errno(SYNTHETIC_ERRNO(EIO),
1138 "Short read while reading %s object: %" PRIu64,
1139 journal_object_type_to_string(type), offset);
1140
1141 r = check_object(f, &o, offset);
1142 if (r < 0)
1143 return r;
1144
1145 if (ret)
1146 *ret = o;
1147
1148 return 0;
1149 }
1150
1151 static uint64_t inc_seqnum(uint64_t seqnum) {
1152 if (seqnum < UINT64_MAX-1)
1153 return seqnum + 1;
1154
1155 return 1; /* skip over UINT64_MAX and 0 when we run out of seqnums and start again */
1156 }
1157
1158 static uint64_t journal_file_entry_seqnum(
1159 JournalFile *f,
1160 uint64_t *seqnum) {
1161
1162 uint64_t next_seqnum;
1163
1164 assert(f);
1165 assert(f->header);
1166
1167 /* Picks a new sequence number for the entry we are about to add and returns it. */
1168
1169 next_seqnum = inc_seqnum(le64toh(f->header->tail_entry_seqnum));
1170
1171 /* If an external seqnum counter was passed, we update both the local and the external one, and set
1172 * it to the maximum of both */
1173 if (seqnum)
1174 *seqnum = next_seqnum = MAX(inc_seqnum(*seqnum), next_seqnum);
1175
1176 f->header->tail_entry_seqnum = htole64(next_seqnum);
1177
1178 if (f->header->head_entry_seqnum == 0)
1179 f->header->head_entry_seqnum = htole64(next_seqnum);
1180
1181 return next_seqnum;
1182 }
1183
1184 int journal_file_append_object(
1185 JournalFile *f,
1186 ObjectType type,
1187 uint64_t size,
1188 Object **ret_object,
1189 uint64_t *ret_offset) {
1190
1191 int r;
1192 uint64_t p;
1193 Object *o;
1194
1195 assert(f);
1196 assert(f->header);
1197 assert(type > OBJECT_UNUSED && type < _OBJECT_TYPE_MAX);
1198 assert(size >= sizeof(ObjectHeader));
1199
1200 r = journal_file_set_online(f);
1201 if (r < 0)
1202 return r;
1203
1204 r = journal_file_tail_end_by_mmap(f, &p);
1205 if (r < 0)
1206 return r;
1207
1208 r = journal_file_allocate(f, p, size);
1209 if (r < 0)
1210 return r;
1211
1212 r = journal_file_move_to(f, type, false, p, size, (void**) &o);
1213 if (r < 0)
1214 return r;
1215
1216 o->object = (ObjectHeader) {
1217 .type = type,
1218 .size = htole64(size),
1219 };
1220
1221 f->header->tail_object_offset = htole64(p);
1222 f->header->n_objects = htole64(le64toh(f->header->n_objects) + 1);
1223
1224 if (ret_object)
1225 *ret_object = o;
1226
1227 if (ret_offset)
1228 *ret_offset = p;
1229
1230 return 0;
1231 }
1232
1233 static int journal_file_setup_data_hash_table(JournalFile *f) {
1234 uint64_t s, p;
1235 Object *o;
1236 int r;
1237
1238 assert(f);
1239 assert(f->header);
1240
1241 /* We estimate that we need 1 hash table entry per 768 bytes
1242 of journal file and we want to make sure we never get
1243 beyond 75% fill level. Calculate the hash table size for
1244 the maximum file size based on these metrics. */
1245
1246 s = (f->metrics.max_size * 4 / 768 / 3) * sizeof(HashItem);
1247 if (s < DEFAULT_DATA_HASH_TABLE_SIZE)
1248 s = DEFAULT_DATA_HASH_TABLE_SIZE;
1249
1250 log_debug("Reserving %"PRIu64" entries in data hash table.", s / sizeof(HashItem));
1251
1252 r = journal_file_append_object(f,
1253 OBJECT_DATA_HASH_TABLE,
1254 offsetof(Object, hash_table.items) + s,
1255 &o, &p);
1256 if (r < 0)
1257 return r;
1258
1259 memzero(o->hash_table.items, s);
1260
1261 f->header->data_hash_table_offset = htole64(p + offsetof(Object, hash_table.items));
1262 f->header->data_hash_table_size = htole64(s);
1263
1264 return 0;
1265 }
1266
1267 static int journal_file_setup_field_hash_table(JournalFile *f) {
1268 uint64_t s, p;
1269 Object *o;
1270 int r;
1271
1272 assert(f);
1273 assert(f->header);
1274
1275 /* We use a fixed size hash table for the fields as this
1276 * number should grow very slowly only */
1277
1278 s = DEFAULT_FIELD_HASH_TABLE_SIZE;
1279 log_debug("Reserving %"PRIu64" entries in field hash table.", s / sizeof(HashItem));
1280
1281 r = journal_file_append_object(f,
1282 OBJECT_FIELD_HASH_TABLE,
1283 offsetof(Object, hash_table.items) + s,
1284 &o, &p);
1285 if (r < 0)
1286 return r;
1287
1288 memzero(o->hash_table.items, s);
1289
1290 f->header->field_hash_table_offset = htole64(p + offsetof(Object, hash_table.items));
1291 f->header->field_hash_table_size = htole64(s);
1292
1293 return 0;
1294 }
1295
1296 int journal_file_map_data_hash_table(JournalFile *f) {
1297 uint64_t s, p;
1298 void *t;
1299 int r;
1300
1301 assert(f);
1302 assert(f->header);
1303
1304 if (f->data_hash_table)
1305 return 0;
1306
1307 p = le64toh(f->header->data_hash_table_offset);
1308 s = le64toh(f->header->data_hash_table_size);
1309
1310 r = journal_file_move_to(f,
1311 OBJECT_DATA_HASH_TABLE,
1312 true,
1313 p, s,
1314 &t);
1315 if (r < 0)
1316 return r;
1317
1318 f->data_hash_table = t;
1319 return 0;
1320 }
1321
1322 int journal_file_map_field_hash_table(JournalFile *f) {
1323 uint64_t s, p;
1324 void *t;
1325 int r;
1326
1327 assert(f);
1328 assert(f->header);
1329
1330 if (f->field_hash_table)
1331 return 0;
1332
1333 p = le64toh(f->header->field_hash_table_offset);
1334 s = le64toh(f->header->field_hash_table_size);
1335
1336 r = journal_file_move_to(f,
1337 OBJECT_FIELD_HASH_TABLE,
1338 true,
1339 p, s,
1340 &t);
1341 if (r < 0)
1342 return r;
1343
1344 f->field_hash_table = t;
1345 return 0;
1346 }
1347
1348 static int journal_file_link_field(
1349 JournalFile *f,
1350 Object *o,
1351 uint64_t offset,
1352 uint64_t hash) {
1353
1354 uint64_t p, h, m;
1355 int r;
1356
1357 assert(f);
1358 assert(f->header);
1359 assert(f->field_hash_table);
1360 assert(o);
1361 assert(offset > 0);
1362
1363 if (o->object.type != OBJECT_FIELD)
1364 return -EINVAL;
1365
1366 m = le64toh(READ_NOW(f->header->field_hash_table_size)) / sizeof(HashItem);
1367 if (m <= 0)
1368 return -EBADMSG;
1369
1370 /* This might alter the window we are looking at */
1371 o->field.next_hash_offset = o->field.head_data_offset = 0;
1372
1373 h = hash % m;
1374 p = le64toh(f->field_hash_table[h].tail_hash_offset);
1375 if (p == 0)
1376 f->field_hash_table[h].head_hash_offset = htole64(offset);
1377 else {
1378 r = journal_file_move_to_object(f, OBJECT_FIELD, p, &o);
1379 if (r < 0)
1380 return r;
1381
1382 o->field.next_hash_offset = htole64(offset);
1383 }
1384
1385 f->field_hash_table[h].tail_hash_offset = htole64(offset);
1386
1387 if (JOURNAL_HEADER_CONTAINS(f->header, n_fields))
1388 f->header->n_fields = htole64(le64toh(f->header->n_fields) + 1);
1389
1390 return 0;
1391 }
1392
1393 static int journal_file_link_data(
1394 JournalFile *f,
1395 Object *o,
1396 uint64_t offset,
1397 uint64_t hash) {
1398
1399 uint64_t p, h, m;
1400 int r;
1401
1402 assert(f);
1403 assert(f->header);
1404 assert(f->data_hash_table);
1405 assert(o);
1406 assert(offset > 0);
1407
1408 if (o->object.type != OBJECT_DATA)
1409 return -EINVAL;
1410
1411 m = le64toh(READ_NOW(f->header->data_hash_table_size)) / sizeof(HashItem);
1412 if (m <= 0)
1413 return -EBADMSG;
1414
1415 /* This might alter the window we are looking at */
1416 o->data.next_hash_offset = o->data.next_field_offset = 0;
1417 o->data.entry_offset = o->data.entry_array_offset = 0;
1418 o->data.n_entries = 0;
1419
1420 h = hash % m;
1421 p = le64toh(f->data_hash_table[h].tail_hash_offset);
1422 if (p == 0)
1423 /* Only entry in the hash table is easy */
1424 f->data_hash_table[h].head_hash_offset = htole64(offset);
1425 else {
1426 /* Move back to the previous data object, to patch in
1427 * pointer */
1428
1429 r = journal_file_move_to_object(f, OBJECT_DATA, p, &o);
1430 if (r < 0)
1431 return r;
1432
1433 o->data.next_hash_offset = htole64(offset);
1434 }
1435
1436 f->data_hash_table[h].tail_hash_offset = htole64(offset);
1437
1438 if (JOURNAL_HEADER_CONTAINS(f->header, n_data))
1439 f->header->n_data = htole64(le64toh(f->header->n_data) + 1);
1440
1441 return 0;
1442 }
1443
1444 static int get_next_hash_offset(
1445 JournalFile *f,
1446 uint64_t *p,
1447 le64_t *next_hash_offset,
1448 uint64_t *depth,
1449 le64_t *header_max_depth) {
1450
1451 uint64_t nextp;
1452
1453 assert(f);
1454 assert(p);
1455 assert(next_hash_offset);
1456 assert(depth);
1457
1458 nextp = le64toh(READ_NOW(*next_hash_offset));
1459 if (nextp > 0) {
1460 if (nextp <= *p) /* Refuse going in loops */
1461 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG),
1462 "Detected hash item loop in %s, refusing.", f->path);
1463
1464 (*depth)++;
1465
1466 /* If the depth of this hash chain is larger than all others we have seen so far, record it */
1467 if (header_max_depth && journal_file_writable(f))
1468 *header_max_depth = htole64(MAX(*depth, le64toh(*header_max_depth)));
1469 }
1470
1471 *p = nextp;
1472 return 0;
1473 }
1474
1475 int journal_file_find_field_object_with_hash(
1476 JournalFile *f,
1477 const void *field,
1478 uint64_t size,
1479 uint64_t hash,
1480 Object **ret_object,
1481 uint64_t *ret_offset) {
1482
1483 uint64_t p, osize, h, m, depth = 0;
1484 int r;
1485
1486 assert(f);
1487 assert(f->header);
1488 assert(field);
1489 assert(size > 0);
1490
1491 /* If the field hash table is empty, we can't find anything */
1492 if (le64toh(f->header->field_hash_table_size) <= 0)
1493 return 0;
1494
1495 /* Map the field hash table, if it isn't mapped yet. */
1496 r = journal_file_map_field_hash_table(f);
1497 if (r < 0)
1498 return r;
1499
1500 osize = offsetof(Object, field.payload) + size;
1501
1502 m = le64toh(READ_NOW(f->header->field_hash_table_size)) / sizeof(HashItem);
1503 if (m <= 0)
1504 return -EBADMSG;
1505
1506 h = hash % m;
1507 p = le64toh(f->field_hash_table[h].head_hash_offset);
1508 while (p > 0) {
1509 Object *o;
1510
1511 r = journal_file_move_to_object(f, OBJECT_FIELD, p, &o);
1512 if (r < 0)
1513 return r;
1514
1515 if (le64toh(o->field.hash) == hash &&
1516 le64toh(o->object.size) == osize &&
1517 memcmp(o->field.payload, field, size) == 0) {
1518
1519 if (ret_object)
1520 *ret_object = o;
1521 if (ret_offset)
1522 *ret_offset = p;
1523
1524 return 1;
1525 }
1526
1527 r = get_next_hash_offset(
1528 f,
1529 &p,
1530 &o->field.next_hash_offset,
1531 &depth,
1532 JOURNAL_HEADER_CONTAINS(f->header, field_hash_chain_depth) ? &f->header->field_hash_chain_depth : NULL);
1533 if (r < 0)
1534 return r;
1535 }
1536
1537 return 0;
1538 }
1539
1540 uint64_t journal_file_hash_data(
1541 JournalFile *f,
1542 const void *data,
1543 size_t sz) {
1544
1545 assert(f);
1546 assert(f->header);
1547 assert(data || sz == 0);
1548
1549 /* We try to unify our codebase on siphash, hence new-styled journal files utilizing the keyed hash
1550 * function use siphash. Old journal files use the Jenkins hash. */
1551
1552 if (JOURNAL_HEADER_KEYED_HASH(f->header))
1553 return siphash24(data, sz, f->header->file_id.bytes);
1554
1555 return jenkins_hash64(data, sz);
1556 }
1557
1558 int journal_file_find_field_object(
1559 JournalFile *f,
1560 const void *field,
1561 uint64_t size,
1562 Object **ret_object,
1563 uint64_t *ret_offset) {
1564
1565 assert(f);
1566 assert(field);
1567 assert(size > 0);
1568
1569 return journal_file_find_field_object_with_hash(
1570 f,
1571 field, size,
1572 journal_file_hash_data(f, field, size),
1573 ret_object, ret_offset);
1574 }
1575
1576 int journal_file_find_data_object_with_hash(
1577 JournalFile *f,
1578 const void *data,
1579 uint64_t size,
1580 uint64_t hash,
1581 Object **ret_object,
1582 uint64_t *ret_offset) {
1583
1584 uint64_t p, h, m, depth = 0;
1585 int r;
1586
1587 assert(f);
1588 assert(f->header);
1589 assert(data || size == 0);
1590
1591 /* If there's no data hash table, then there's no entry. */
1592 if (le64toh(f->header->data_hash_table_size) <= 0)
1593 return 0;
1594
1595 /* Map the data hash table, if it isn't mapped yet. */
1596 r = journal_file_map_data_hash_table(f);
1597 if (r < 0)
1598 return r;
1599
1600 m = le64toh(READ_NOW(f->header->data_hash_table_size)) / sizeof(HashItem);
1601 if (m <= 0)
1602 return -EBADMSG;
1603
1604 h = hash % m;
1605 p = le64toh(f->data_hash_table[h].head_hash_offset);
1606
1607 while (p > 0) {
1608 Object *o;
1609 void *d;
1610 size_t rsize;
1611
1612 r = journal_file_move_to_object(f, OBJECT_DATA, p, &o);
1613 if (r < 0)
1614 return r;
1615
1616 if (le64toh(o->data.hash) != hash)
1617 goto next;
1618
1619 r = journal_file_data_payload(f, o, p, NULL, 0, 0, &d, &rsize);
1620 if (r < 0)
1621 return r;
1622 assert(r > 0); /* journal_file_data_payload() always returns > 0 if no field is provided. */
1623
1624 if (memcmp_nn(data, size, d, rsize) == 0) {
1625 if (ret_object)
1626 *ret_object = o;
1627
1628 if (ret_offset)
1629 *ret_offset = p;
1630
1631 return 1;
1632 }
1633
1634 next:
1635 r = get_next_hash_offset(
1636 f,
1637 &p,
1638 &o->data.next_hash_offset,
1639 &depth,
1640 JOURNAL_HEADER_CONTAINS(f->header, data_hash_chain_depth) ? &f->header->data_hash_chain_depth : NULL);
1641 if (r < 0)
1642 return r;
1643 }
1644
1645 return 0;
1646 }
1647
1648 int journal_file_find_data_object(
1649 JournalFile *f,
1650 const void *data,
1651 uint64_t size,
1652 Object **ret_object,
1653 uint64_t *ret_offset) {
1654
1655 assert(f);
1656 assert(data || size == 0);
1657
1658 return journal_file_find_data_object_with_hash(
1659 f,
1660 data, size,
1661 journal_file_hash_data(f, data, size),
1662 ret_object, ret_offset);
1663 }
1664
1665 bool journal_field_valid(const char *p, size_t l, bool allow_protected) {
1666 /* We kinda enforce POSIX syntax recommendations for
1667 environment variables here, but make a couple of additional
1668 requirements.
1669
1670 http://pubs.opengroup.org/onlinepubs/000095399/basedefs/xbd_chap08.html */
1671
1672 assert(p);
1673
1674 if (l == SIZE_MAX)
1675 l = strlen(p);
1676
1677 /* No empty field names */
1678 if (l <= 0)
1679 return false;
1680
1681 /* Don't allow names longer than 64 chars */
1682 if (l > 64)
1683 return false;
1684
1685 /* Variables starting with an underscore are protected */
1686 if (!allow_protected && p[0] == '_')
1687 return false;
1688
1689 /* Don't allow digits as first character */
1690 if (ascii_isdigit(p[0]))
1691 return false;
1692
1693 /* Only allow A-Z0-9 and '_' */
1694 for (const char *a = p; a < p + l; a++)
1695 if ((*a < 'A' || *a > 'Z') &&
1696 !ascii_isdigit(*a) &&
1697 *a != '_')
1698 return false;
1699
1700 return true;
1701 }
1702
1703 static int journal_file_append_field(
1704 JournalFile *f,
1705 const void *field,
1706 uint64_t size,
1707 Object **ret_object,
1708 uint64_t *ret_offset) {
1709
1710 uint64_t hash, p;
1711 uint64_t osize;
1712 Object *o;
1713 int r;
1714
1715 assert(f);
1716 assert(field);
1717 assert(size > 0);
1718
1719 if (!journal_field_valid(field, size, true))
1720 return -EBADMSG;
1721
1722 hash = journal_file_hash_data(f, field, size);
1723
1724 r = journal_file_find_field_object_with_hash(f, field, size, hash, ret_object, ret_offset);
1725 if (r < 0)
1726 return r;
1727 if (r > 0)
1728 return 0;
1729
1730 osize = offsetof(Object, field.payload) + size;
1731 r = journal_file_append_object(f, OBJECT_FIELD, osize, &o, &p);
1732 if (r < 0)
1733 return r;
1734
1735 o->field.hash = htole64(hash);
1736 memcpy(o->field.payload, field, size);
1737
1738 r = journal_file_link_field(f, o, p, hash);
1739 if (r < 0)
1740 return r;
1741
1742 /* The linking might have altered the window, so let's only pass the offset to hmac which will
1743 * move to the object again if needed. */
1744
1745 #if HAVE_GCRYPT
1746 r = journal_file_hmac_put_object(f, OBJECT_FIELD, NULL, p);
1747 if (r < 0)
1748 return r;
1749 #endif
1750
1751 if (ret_object) {
1752 r = journal_file_move_to_object(f, OBJECT_FIELD, p, ret_object);
1753 if (r < 0)
1754 return r;
1755 }
1756
1757 if (ret_offset)
1758 *ret_offset = p;
1759
1760 return 0;
1761 }
1762
1763 static int maybe_compress_payload(JournalFile *f, uint8_t *dst, const uint8_t *src, uint64_t size, size_t *rsize) {
1764 assert(f);
1765 assert(f->header);
1766
1767 #if HAVE_COMPRESSION
1768 Compression c;
1769 int r;
1770
1771 c = JOURNAL_FILE_COMPRESSION(f);
1772 if (c == COMPRESSION_NONE || size < f->compress_threshold_bytes)
1773 return 0;
1774
1775 r = compress_blob(c, src, size, dst, size - 1, rsize);
1776 if (r < 0)
1777 return log_debug_errno(r, "Failed to compress data object using %s, ignoring: %m", compression_to_string(c));
1778
1779 log_debug("Compressed data object %"PRIu64" -> %zu using %s", size, *rsize, compression_to_string(c));
1780
1781 return 1; /* compressed */
1782 #else
1783 return 0;
1784 #endif
1785 }
1786
1787 static int journal_file_append_data(
1788 JournalFile *f,
1789 const void *data,
1790 uint64_t size,
1791 Object **ret_object,
1792 uint64_t *ret_offset) {
1793
1794 uint64_t hash, p, osize;
1795 Object *o, *fo;
1796 size_t rsize = 0;
1797 const void *eq;
1798 int r;
1799
1800 assert(f);
1801
1802 if (!data || size == 0)
1803 return -EINVAL;
1804
1805 hash = journal_file_hash_data(f, data, size);
1806
1807 r = journal_file_find_data_object_with_hash(f, data, size, hash, ret_object, ret_offset);
1808 if (r < 0)
1809 return r;
1810 if (r > 0)
1811 return 0;
1812
1813 eq = memchr(data, '=', size);
1814 if (!eq)
1815 return -EINVAL;
1816
1817 osize = journal_file_data_payload_offset(f) + size;
1818 r = journal_file_append_object(f, OBJECT_DATA, osize, &o, &p);
1819 if (r < 0)
1820 return r;
1821
1822 o->data.hash = htole64(hash);
1823
1824 r = maybe_compress_payload(f, journal_file_data_payload_field(f, o), data, size, &rsize);
1825 if (r <= 0)
1826 /* We don't really care failures, let's continue without compression */
1827 memcpy_safe(journal_file_data_payload_field(f, o), data, size);
1828 else {
1829 Compression c = JOURNAL_FILE_COMPRESSION(f);
1830
1831 assert(c >= 0 && c < _COMPRESSION_MAX && c != COMPRESSION_NONE);
1832
1833 o->object.size = htole64(journal_file_data_payload_offset(f) + rsize);
1834 o->object.flags |= COMPRESSION_TO_OBJECT_FLAG(c);
1835 }
1836
1837 r = journal_file_link_data(f, o, p, hash);
1838 if (r < 0)
1839 return r;
1840
1841 /* The linking might have altered the window, so let's refresh our pointer. */
1842 r = journal_file_move_to_object(f, OBJECT_DATA, p, &o);
1843 if (r < 0)
1844 return r;
1845
1846 #if HAVE_GCRYPT
1847 r = journal_file_hmac_put_object(f, OBJECT_DATA, o, p);
1848 if (r < 0)
1849 return r;
1850 #endif
1851
1852 /* Create field object ... */
1853 r = journal_file_append_field(f, data, (uint8_t*) eq - (uint8_t*) data, &fo, NULL);
1854 if (r < 0)
1855 return r;
1856
1857 /* ... and link it in. */
1858 o->data.next_field_offset = fo->field.head_data_offset;
1859 fo->field.head_data_offset = le64toh(p);
1860
1861 if (ret_object)
1862 *ret_object = o;
1863
1864 if (ret_offset)
1865 *ret_offset = p;
1866
1867 return 0;
1868 }
1869
1870 static int maybe_decompress_payload(
1871 JournalFile *f,
1872 uint8_t *payload,
1873 uint64_t size,
1874 Compression compression,
1875 const char *field,
1876 size_t field_length,
1877 size_t data_threshold,
1878 void **ret_data,
1879 size_t *ret_size) {
1880
1881 assert(f);
1882
1883 /* We can't read objects larger than 4G on a 32-bit machine */
1884 if ((uint64_t) (size_t) size != size)
1885 return -E2BIG;
1886
1887 if (compression != COMPRESSION_NONE) {
1888 #if HAVE_COMPRESSION
1889 size_t rsize;
1890 int r;
1891
1892 if (field) {
1893 r = decompress_startswith(compression, payload, size, &f->compress_buffer, field,
1894 field_length, '=');
1895 if (r < 0)
1896 return log_debug_errno(r,
1897 "Cannot decompress %s object of length %" PRIu64 ": %m",
1898 compression_to_string(compression),
1899 size);
1900 if (r == 0) {
1901 if (ret_data)
1902 *ret_data = NULL;
1903 if (ret_size)
1904 *ret_size = 0;
1905 return 0;
1906 }
1907 }
1908
1909 r = decompress_blob(compression, payload, size, &f->compress_buffer, &rsize, 0);
1910 if (r < 0)
1911 return r;
1912
1913 if (ret_data)
1914 *ret_data = f->compress_buffer;
1915 if (ret_size)
1916 *ret_size = rsize;
1917 #else
1918 return -EPROTONOSUPPORT;
1919 #endif
1920 } else {
1921 if (field && (size < field_length + 1 || memcmp(payload, field, field_length) != 0 || payload[field_length] != '=')) {
1922 if (ret_data)
1923 *ret_data = NULL;
1924 if (ret_size)
1925 *ret_size = 0;
1926 return 0;
1927 }
1928
1929 if (ret_data)
1930 *ret_data = payload;
1931 if (ret_size)
1932 *ret_size = (size_t) size;
1933 }
1934
1935 return 1;
1936 }
1937
1938 int journal_file_data_payload(
1939 JournalFile *f,
1940 Object *o,
1941 uint64_t offset,
1942 const char *field,
1943 size_t field_length,
1944 size_t data_threshold,
1945 void **ret_data,
1946 size_t *ret_size) {
1947
1948 uint64_t size;
1949 Compression c;
1950 int r;
1951
1952 assert(f);
1953 assert(!field == (field_length == 0)); /* These must be specified together. */
1954
1955 if (!o) {
1956 r = journal_file_move_to_object(f, OBJECT_DATA, offset, &o);
1957 if (r < 0)
1958 return r;
1959 }
1960
1961 size = le64toh(READ_NOW(o->object.size));
1962 if (size < journal_file_data_payload_offset(f))
1963 return -EBADMSG;
1964
1965 size -= journal_file_data_payload_offset(f);
1966
1967 c = COMPRESSION_FROM_OBJECT(o);
1968 if (c < 0)
1969 return -EPROTONOSUPPORT;
1970
1971 return maybe_decompress_payload(f, journal_file_data_payload_field(f, o), size, c, field,
1972 field_length, data_threshold, ret_data, ret_size);
1973 }
1974
1975 uint64_t journal_file_entry_n_items(JournalFile *f, Object *o) {
1976 uint64_t sz;
1977
1978 assert(f);
1979 assert(o);
1980
1981 if (o->object.type != OBJECT_ENTRY)
1982 return 0;
1983
1984 sz = le64toh(READ_NOW(o->object.size));
1985 if (sz < offsetof(Object, entry.items))
1986 return 0;
1987
1988 return (sz - offsetof(Object, entry.items)) / journal_file_entry_item_size(f);
1989 }
1990
1991 uint64_t journal_file_entry_array_n_items(JournalFile *f, Object *o) {
1992 uint64_t sz;
1993
1994 assert(f);
1995 assert(o);
1996
1997 if (o->object.type != OBJECT_ENTRY_ARRAY)
1998 return 0;
1999
2000 sz = le64toh(READ_NOW(o->object.size));
2001 if (sz < offsetof(Object, entry_array.items))
2002 return 0;
2003
2004 return (sz - offsetof(Object, entry_array.items)) / journal_file_entry_array_item_size(f);
2005 }
2006
2007 uint64_t journal_file_hash_table_n_items(Object *o) {
2008 uint64_t sz;
2009
2010 assert(o);
2011
2012 if (!IN_SET(o->object.type, OBJECT_DATA_HASH_TABLE, OBJECT_FIELD_HASH_TABLE))
2013 return 0;
2014
2015 sz = le64toh(READ_NOW(o->object.size));
2016 if (sz < offsetof(Object, hash_table.items))
2017 return 0;
2018
2019 return (sz - offsetof(Object, hash_table.items)) / sizeof(HashItem);
2020 }
2021
2022 static void write_entry_array_item(JournalFile *f, Object *o, uint64_t i, uint64_t p) {
2023 assert(f);
2024 assert(o);
2025
2026 if (JOURNAL_HEADER_COMPACT(f->header)) {
2027 assert(p <= UINT32_MAX);
2028 o->entry_array.items.compact[i] = htole32(p);
2029 } else
2030 o->entry_array.items.regular[i] = htole64(p);
2031 }
2032
2033 static int link_entry_into_array(
2034 JournalFile *f,
2035 le64_t *first,
2036 le64_t *idx,
2037 le32_t *tail,
2038 le32_t *tidx,
2039 uint64_t p) {
2040
2041 uint64_t n = 0, ap = 0, q, i, a, hidx;
2042 Object *o;
2043 int r;
2044
2045 assert(f);
2046 assert(f->header);
2047 assert(first);
2048 assert(idx);
2049 assert(p > 0);
2050
2051 a = tail ? le32toh(*tail) : le64toh(*first);
2052 hidx = le64toh(READ_NOW(*idx));
2053 i = tidx ? le32toh(READ_NOW(*tidx)) : hidx;
2054
2055 while (a > 0) {
2056 r = journal_file_move_to_object(f, OBJECT_ENTRY_ARRAY, a, &o);
2057 if (r < 0)
2058 return r;
2059
2060 n = journal_file_entry_array_n_items(f, o);
2061 if (i < n) {
2062 write_entry_array_item(f, o, i, p);
2063 *idx = htole64(hidx + 1);
2064 if (tidx)
2065 *tidx = htole32(le32toh(*tidx) + 1);
2066 return 0;
2067 }
2068
2069 i -= n;
2070 ap = a;
2071 a = le64toh(o->entry_array.next_entry_array_offset);
2072 }
2073
2074 if (hidx > n)
2075 n = (hidx+1) * 2;
2076 else
2077 n = n * 2;
2078
2079 if (n < 4)
2080 n = 4;
2081
2082 r = journal_file_append_object(f, OBJECT_ENTRY_ARRAY,
2083 offsetof(Object, entry_array.items) + n * journal_file_entry_array_item_size(f),
2084 &o, &q);
2085 if (r < 0)
2086 return r;
2087
2088 #if HAVE_GCRYPT
2089 r = journal_file_hmac_put_object(f, OBJECT_ENTRY_ARRAY, o, q);
2090 if (r < 0)
2091 return r;
2092 #endif
2093
2094 write_entry_array_item(f, o, i, p);
2095
2096 if (ap == 0)
2097 *first = htole64(q);
2098 else {
2099 r = journal_file_move_to_object(f, OBJECT_ENTRY_ARRAY, ap, &o);
2100 if (r < 0)
2101 return r;
2102
2103 o->entry_array.next_entry_array_offset = htole64(q);
2104 }
2105
2106 if (tail)
2107 *tail = htole32(q);
2108
2109 if (JOURNAL_HEADER_CONTAINS(f->header, n_entry_arrays))
2110 f->header->n_entry_arrays = htole64(le64toh(f->header->n_entry_arrays) + 1);
2111
2112 *idx = htole64(hidx + 1);
2113 if (tidx)
2114 *tidx = htole32(1);
2115
2116 return 0;
2117 }
2118
2119 static int link_entry_into_array_plus_one(
2120 JournalFile *f,
2121 le64_t *extra,
2122 le64_t *first,
2123 le64_t *idx,
2124 le32_t *tail,
2125 le32_t *tidx,
2126 uint64_t p) {
2127
2128 uint64_t hidx;
2129 int r;
2130
2131 assert(f);
2132 assert(extra);
2133 assert(first);
2134 assert(idx);
2135 assert(p > 0);
2136
2137 hidx = le64toh(READ_NOW(*idx));
2138 if (hidx == UINT64_MAX)
2139 return -EBADMSG;
2140 if (hidx == 0)
2141 *extra = htole64(p);
2142 else {
2143 le64_t i;
2144
2145 i = htole64(hidx - 1);
2146 r = link_entry_into_array(f, first, &i, tail, tidx, p);
2147 if (r < 0)
2148 return r;
2149 }
2150
2151 *idx = htole64(hidx + 1);
2152 return 0;
2153 }
2154
2155 static int journal_file_link_entry_item(JournalFile *f, uint64_t offset, uint64_t p) {
2156 Object *o;
2157 int r;
2158
2159 assert(f);
2160 assert(offset > 0);
2161
2162 r = journal_file_move_to_object(f, OBJECT_DATA, p, &o);
2163 if (r < 0)
2164 return r;
2165
2166 return link_entry_into_array_plus_one(f,
2167 &o->data.entry_offset,
2168 &o->data.entry_array_offset,
2169 &o->data.n_entries,
2170 JOURNAL_HEADER_COMPACT(f->header) ? &o->data.compact.tail_entry_array_offset : NULL,
2171 JOURNAL_HEADER_COMPACT(f->header) ? &o->data.compact.tail_entry_array_n_entries : NULL,
2172 offset);
2173 }
2174
2175 static int journal_file_link_entry(
2176 JournalFile *f,
2177 Object *o,
2178 uint64_t offset,
2179 const EntryItem items[],
2180 size_t n_items) {
2181
2182 int r;
2183
2184 assert(f);
2185 assert(f->header);
2186 assert(o);
2187 assert(offset > 0);
2188
2189 if (o->object.type != OBJECT_ENTRY)
2190 return -EINVAL;
2191
2192 __atomic_thread_fence(__ATOMIC_SEQ_CST);
2193
2194 /* Link up the entry itself */
2195 r = link_entry_into_array(f,
2196 &f->header->entry_array_offset,
2197 &f->header->n_entries,
2198 JOURNAL_HEADER_CONTAINS(f->header, tail_entry_array_offset) ? &f->header->tail_entry_array_offset : NULL,
2199 JOURNAL_HEADER_CONTAINS(f->header, tail_entry_array_n_entries) ? &f->header->tail_entry_array_n_entries : NULL,
2200 offset);
2201 if (r < 0)
2202 return r;
2203
2204 /* log_debug("=> %s seqnr=%"PRIu64" n_entries=%"PRIu64, f->path, o->entry.seqnum, f->header->n_entries); */
2205
2206 if (f->header->head_entry_realtime == 0)
2207 f->header->head_entry_realtime = o->entry.realtime;
2208
2209 f->header->tail_entry_realtime = o->entry.realtime;
2210 f->header->tail_entry_monotonic = o->entry.monotonic;
2211 if (JOURNAL_HEADER_CONTAINS(f->header, tail_entry_offset))
2212 f->header->tail_entry_offset = htole64(offset);
2213 f->newest_mtime = 0; /* we have a new tail entry now, explicitly invalidate newest boot id/timestamp info */
2214
2215 /* Link up the items */
2216 for (uint64_t i = 0; i < n_items; i++) {
2217 int k;
2218
2219 /* If we fail to link an entry item because we can't allocate a new entry array, don't fail
2220 * immediately but try to link the other entry items since it might still be possible to link
2221 * those if they don't require a new entry array to be allocated. */
2222
2223 k = journal_file_link_entry_item(f, offset, items[i].object_offset);
2224 if (k == -E2BIG)
2225 r = k;
2226 else if (k < 0)
2227 return k;
2228 }
2229
2230 return r;
2231 }
2232
2233 static void write_entry_item(JournalFile *f, Object *o, uint64_t i, const EntryItem *item) {
2234 assert(f);
2235 assert(o);
2236 assert(item);
2237
2238 if (JOURNAL_HEADER_COMPACT(f->header)) {
2239 assert(item->object_offset <= UINT32_MAX);
2240 o->entry.items.compact[i].object_offset = htole32(item->object_offset);
2241 } else {
2242 o->entry.items.regular[i].object_offset = htole64(item->object_offset);
2243 o->entry.items.regular[i].hash = htole64(item->hash);
2244 }
2245 }
2246
2247 static int journal_file_append_entry_internal(
2248 JournalFile *f,
2249 const dual_timestamp *ts,
2250 const sd_id128_t *boot_id,
2251 const sd_id128_t *machine_id,
2252 uint64_t xor_hash,
2253 const EntryItem items[],
2254 size_t n_items,
2255 uint64_t *seqnum,
2256 sd_id128_t *seqnum_id,
2257 Object **ret_object,
2258 uint64_t *ret_offset) {
2259
2260 uint64_t np;
2261 uint64_t osize;
2262 Object *o;
2263 int r;
2264
2265 assert(f);
2266 assert(f->header);
2267 assert(ts);
2268 assert(items || n_items == 0);
2269
2270 if (f->strict_order) {
2271 /* If requested be stricter with ordering in this journal file, to make searching via
2272 * bisection fully deterministic. This is an optional feature, so that if desired journal
2273 * files can be written where the ordering is not strictly enforced (in which case bisection
2274 * will yield *a* result, but not the *only* result, when searching for points in
2275 * time). Strict ordering mode is enabled when journald originally writes the files, but
2276 * might not necessarily be if other tools (the remoting tools for example) write journal
2277 * files from combined sources.
2278 *
2279 * Typically, if any of the errors generated here are seen journald will just rotate the
2280 * journal files and start anew. */
2281
2282 if (ts->realtime < le64toh(f->header->tail_entry_realtime))
2283 return log_debug_errno(SYNTHETIC_ERRNO(EREMCHG),
2284 "Realtime timestamp %" PRIu64 " smaller than previous realtime "
2285 "timestamp %" PRIu64 ", refusing entry.",
2286 ts->realtime, le64toh(f->header->tail_entry_realtime));
2287
2288 if (!sd_id128_is_null(f->header->tail_entry_boot_id) && boot_id) {
2289
2290 if (!sd_id128_equal(f->header->tail_entry_boot_id, *boot_id))
2291 return log_debug_errno(SYNTHETIC_ERRNO(EREMOTE),
2292 "Boot ID to write is different from previous boot id, refusing entry.");
2293
2294 if (ts->monotonic < le64toh(f->header->tail_entry_monotonic))
2295 return log_debug_errno(SYNTHETIC_ERRNO(ENOTNAM),
2296 "Monotonic timestamp %" PRIu64 " smaller than previous monotonic "
2297 "timestamp %" PRIu64 ", refusing entry.",
2298 ts->monotonic, le64toh(f->header->tail_entry_monotonic));
2299 }
2300 }
2301
2302 if (seqnum_id) {
2303 /* Settle the passed in sequence number ID */
2304
2305 if (sd_id128_is_null(*seqnum_id))
2306 *seqnum_id = f->header->seqnum_id; /* Caller has none assigned, then copy the one from the file */
2307 else if (!sd_id128_equal(*seqnum_id, f->header->seqnum_id)) {
2308 /* Different seqnum IDs? We can't allow entries from multiple IDs end up in the same journal.*/
2309 if (le64toh(f->header->n_entries) == 0)
2310 f->header->seqnum_id = *seqnum_id; /* Caller has one, and file so far has no entries, then copy the one from the caller */
2311 else
2312 return log_debug_errno(SYNTHETIC_ERRNO(EILSEQ),
2313 "Sequence number IDs don't match, refusing entry.");
2314 }
2315 }
2316
2317 if (machine_id && sd_id128_is_null(f->header->machine_id))
2318 /* Initialize machine ID when not set yet */
2319 f->header->machine_id = *machine_id;
2320
2321 osize = offsetof(Object, entry.items) + (n_items * journal_file_entry_item_size(f));
2322
2323 r = journal_file_append_object(f, OBJECT_ENTRY, osize, &o, &np);
2324 if (r < 0)
2325 return r;
2326
2327 o->entry.seqnum = htole64(journal_file_entry_seqnum(f, seqnum));
2328 o->entry.realtime = htole64(ts->realtime);
2329 o->entry.monotonic = htole64(ts->monotonic);
2330 o->entry.xor_hash = htole64(xor_hash);
2331 if (boot_id)
2332 f->header->tail_entry_boot_id = *boot_id;
2333 o->entry.boot_id = f->header->tail_entry_boot_id;
2334
2335 for (size_t i = 0; i < n_items; i++)
2336 write_entry_item(f, o, i, &items[i]);
2337
2338 #if HAVE_GCRYPT
2339 r = journal_file_hmac_put_object(f, OBJECT_ENTRY, o, np);
2340 if (r < 0)
2341 return r;
2342 #endif
2343
2344 r = journal_file_link_entry(f, o, np, items, n_items);
2345 if (r < 0)
2346 return r;
2347
2348 if (ret_object)
2349 *ret_object = o;
2350
2351 if (ret_offset)
2352 *ret_offset = np;
2353
2354 return r;
2355 }
2356
2357 void journal_file_post_change(JournalFile *f) {
2358 assert(f);
2359
2360 if (f->fd < 0)
2361 return;
2362
2363 /* inotify() does not receive IN_MODIFY events from file
2364 * accesses done via mmap(). After each access we hence
2365 * trigger IN_MODIFY by truncating the journal file to its
2366 * current size which triggers IN_MODIFY. */
2367
2368 __atomic_thread_fence(__ATOMIC_SEQ_CST);
2369
2370 if (ftruncate(f->fd, f->last_stat.st_size) < 0)
2371 log_debug_errno(errno, "Failed to truncate file to its own size: %m");
2372 }
2373
2374 static int post_change_thunk(sd_event_source *timer, uint64_t usec, void *userdata) {
2375 assert(userdata);
2376
2377 journal_file_post_change(userdata);
2378
2379 return 1;
2380 }
2381
2382 static void schedule_post_change(JournalFile *f) {
2383 sd_event *e;
2384 int r;
2385
2386 assert(f);
2387 assert(f->post_change_timer);
2388
2389 assert_se(e = sd_event_source_get_event(f->post_change_timer));
2390
2391 /* If we are already going down, post the change immediately. */
2392 if (IN_SET(sd_event_get_state(e), SD_EVENT_EXITING, SD_EVENT_FINISHED))
2393 goto fail;
2394
2395 r = sd_event_source_get_enabled(f->post_change_timer, NULL);
2396 if (r < 0) {
2397 log_debug_errno(r, "Failed to get ftruncate timer state: %m");
2398 goto fail;
2399 }
2400 if (r > 0)
2401 return;
2402
2403 r = sd_event_source_set_time_relative(f->post_change_timer, f->post_change_timer_period);
2404 if (r < 0) {
2405 log_debug_errno(r, "Failed to set time for scheduling ftruncate: %m");
2406 goto fail;
2407 }
2408
2409 r = sd_event_source_set_enabled(f->post_change_timer, SD_EVENT_ONESHOT);
2410 if (r < 0) {
2411 log_debug_errno(r, "Failed to enable scheduled ftruncate: %m");
2412 goto fail;
2413 }
2414
2415 return;
2416
2417 fail:
2418 /* On failure, let's simply post the change immediately. */
2419 journal_file_post_change(f);
2420 }
2421
2422 /* Enable coalesced change posting in a timer on the provided sd_event instance */
2423 int journal_file_enable_post_change_timer(JournalFile *f, sd_event *e, usec_t t) {
2424 _cleanup_(sd_event_source_unrefp) sd_event_source *timer = NULL;
2425 int r;
2426
2427 assert(f);
2428 assert_return(!f->post_change_timer, -EINVAL);
2429 assert(e);
2430 assert(t);
2431
2432 r = sd_event_add_time(e, &timer, CLOCK_MONOTONIC, 0, 0, post_change_thunk, f);
2433 if (r < 0)
2434 return r;
2435
2436 r = sd_event_source_set_enabled(timer, SD_EVENT_OFF);
2437 if (r < 0)
2438 return r;
2439
2440 f->post_change_timer = TAKE_PTR(timer);
2441 f->post_change_timer_period = t;
2442
2443 return r;
2444 }
2445
2446 static int entry_item_cmp(const EntryItem *a, const EntryItem *b) {
2447 return CMP(ASSERT_PTR(a)->object_offset, ASSERT_PTR(b)->object_offset);
2448 }
2449
2450 static size_t remove_duplicate_entry_items(EntryItem items[], size_t n) {
2451 size_t j = 1;
2452
2453 assert(items || n == 0);
2454
2455 if (n <= 1)
2456 return n;
2457
2458 for (size_t i = 1; i < n; i++)
2459 if (items[i].object_offset != items[j - 1].object_offset)
2460 items[j++] = items[i];
2461
2462 return j;
2463 }
2464
2465 int journal_file_append_entry(
2466 JournalFile *f,
2467 const dual_timestamp *ts,
2468 const sd_id128_t *boot_id,
2469 const struct iovec iovec[],
2470 size_t n_iovec,
2471 uint64_t *seqnum,
2472 sd_id128_t *seqnum_id,
2473 Object **ret_object,
2474 uint64_t *ret_offset) {
2475
2476 _cleanup_free_ EntryItem *items_alloc = NULL;
2477 EntryItem *items;
2478 uint64_t xor_hash = 0;
2479 struct dual_timestamp _ts;
2480 sd_id128_t _boot_id, _machine_id, *machine_id;
2481 int r;
2482
2483 assert(f);
2484 assert(f->header);
2485 assert(iovec);
2486 assert(n_iovec > 0);
2487
2488 if (ts) {
2489 if (!VALID_REALTIME(ts->realtime))
2490 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG),
2491 "Invalid realtime timestamp %" PRIu64 ", refusing entry.",
2492 ts->realtime);
2493 if (!VALID_MONOTONIC(ts->monotonic))
2494 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG),
2495 "Invalid monotomic timestamp %" PRIu64 ", refusing entry.",
2496 ts->monotonic);
2497 } else {
2498 dual_timestamp_get(&_ts);
2499 ts = &_ts;
2500 }
2501
2502 if (!boot_id) {
2503 r = sd_id128_get_boot(&_boot_id);
2504 if (r < 0)
2505 return r;
2506
2507 boot_id = &_boot_id;
2508 }
2509
2510 r = sd_id128_get_machine(&_machine_id);
2511 if (r < 0) {
2512 if (!ERRNO_IS_MACHINE_ID_UNSET(r))
2513 return r;
2514
2515 /* If the machine ID is not initialized yet, handle gracefully */
2516 machine_id = NULL;
2517 } else
2518 machine_id = &_machine_id;
2519
2520 #if HAVE_GCRYPT
2521 r = journal_file_maybe_append_tag(f, ts->realtime);
2522 if (r < 0)
2523 return r;
2524 #endif
2525
2526 if (n_iovec < ALLOCA_MAX / sizeof(EntryItem) / 2)
2527 items = newa(EntryItem, n_iovec);
2528 else {
2529 items_alloc = new(EntryItem, n_iovec);
2530 if (!items_alloc)
2531 return -ENOMEM;
2532
2533 items = items_alloc;
2534 }
2535
2536 for (size_t i = 0; i < n_iovec; i++) {
2537 uint64_t p;
2538 Object *o;
2539
2540 r = journal_file_append_data(f, iovec[i].iov_base, iovec[i].iov_len, &o, &p);
2541 if (r < 0)
2542 return r;
2543
2544 /* When calculating the XOR hash field, we need to take special care if the "keyed-hash"
2545 * journal file flag is on. We use the XOR hash field to quickly determine the identity of a
2546 * specific record, and give records with otherwise identical position (i.e. match in seqno,
2547 * timestamp, …) a stable ordering. But for that we can't have it that the hash of the
2548 * objects in each file is different since they are keyed. Hence let's calculate the Jenkins
2549 * hash here for that. This also has the benefit that cursors for old and new journal files
2550 * are completely identical (they include the XOR hash after all). For classic Jenkins-hash
2551 * files things are easier, we can just take the value from the stored record directly. */
2552
2553 if (JOURNAL_HEADER_KEYED_HASH(f->header))
2554 xor_hash ^= jenkins_hash64(iovec[i].iov_base, iovec[i].iov_len);
2555 else
2556 xor_hash ^= le64toh(o->data.hash);
2557
2558 items[i] = (EntryItem) {
2559 .object_offset = p,
2560 .hash = le64toh(o->data.hash),
2561 };
2562 }
2563
2564 /* Order by the position on disk, in order to improve seek
2565 * times for rotating media. */
2566 typesafe_qsort(items, n_iovec, entry_item_cmp);
2567 n_iovec = remove_duplicate_entry_items(items, n_iovec);
2568
2569 r = journal_file_append_entry_internal(
2570 f,
2571 ts,
2572 boot_id,
2573 machine_id,
2574 xor_hash,
2575 items,
2576 n_iovec,
2577 seqnum,
2578 seqnum_id,
2579 ret_object,
2580 ret_offset);
2581
2582 /* If the memory mapping triggered a SIGBUS then we return an
2583 * IO error and ignore the error code passed down to us, since
2584 * it is very likely just an effect of a nullified replacement
2585 * mapping page */
2586
2587 if (mmap_cache_fd_got_sigbus(f->cache_fd))
2588 r = -EIO;
2589
2590 if (f->post_change_timer)
2591 schedule_post_change(f);
2592 else
2593 journal_file_post_change(f);
2594
2595 return r;
2596 }
2597
2598 typedef struct ChainCacheItem {
2599 uint64_t first; /* the array at the beginning of the chain */
2600 uint64_t array; /* the cached array */
2601 uint64_t begin; /* the first item in the cached array */
2602 uint64_t total; /* the total number of items in all arrays before this one in the chain */
2603 uint64_t last_index; /* the last index we looked at, to optimize locality when bisecting */
2604 } ChainCacheItem;
2605
2606 static void chain_cache_put(
2607 OrderedHashmap *h,
2608 ChainCacheItem *ci,
2609 uint64_t first,
2610 uint64_t array,
2611 uint64_t begin,
2612 uint64_t total,
2613 uint64_t last_index) {
2614
2615 assert(h);
2616
2617 if (!ci) {
2618 /* If the chain item to cache for this chain is the
2619 * first one it's not worth caching anything */
2620 if (array == first)
2621 return;
2622
2623 if (ordered_hashmap_size(h) >= CHAIN_CACHE_MAX) {
2624 ci = ordered_hashmap_steal_first(h);
2625 assert(ci);
2626 } else {
2627 ci = new(ChainCacheItem, 1);
2628 if (!ci)
2629 return;
2630 }
2631
2632 ci->first = first;
2633
2634 if (ordered_hashmap_put(h, &ci->first, ci) < 0) {
2635 free(ci);
2636 return;
2637 }
2638 } else
2639 assert(ci->first == first);
2640
2641 ci->array = array;
2642 ci->begin = begin;
2643 ci->total = total;
2644 ci->last_index = last_index;
2645 }
2646
2647 static int bump_array_index(uint64_t *i, direction_t direction, uint64_t n) {
2648 assert(i);
2649
2650 /* Increase or decrease the specified index, in the right direction. */
2651
2652 if (direction == DIRECTION_DOWN) {
2653 if (*i >= n - 1)
2654 return 0;
2655
2656 (*i)++;
2657 } else {
2658 if (*i <= 0)
2659 return 0;
2660
2661 (*i)--;
2662 }
2663
2664 return 1;
2665 }
2666
2667 static int bump_entry_array(
2668 JournalFile *f,
2669 Object *o,
2670 uint64_t offset,
2671 uint64_t first,
2672 direction_t direction,
2673 uint64_t *ret) {
2674
2675 uint64_t p, q = 0;
2676 int r;
2677
2678 assert(f);
2679 assert(offset);
2680 assert(ret);
2681
2682 if (direction == DIRECTION_DOWN) {
2683 assert(o);
2684 *ret = le64toh(o->entry_array.next_entry_array_offset);
2685 return 0;
2686 }
2687
2688 /* Entry array chains are a singly linked list, so to find the previous array in the chain, we have
2689 * to start iterating from the top. */
2690
2691 p = first;
2692
2693 while (p > 0 && p != offset) {
2694 r = journal_file_move_to_object(f, OBJECT_ENTRY_ARRAY, p, &o);
2695 if (r < 0)
2696 return r;
2697
2698 q = p;
2699 p = le64toh(o->entry_array.next_entry_array_offset);
2700 }
2701
2702 /* If we can't find the previous entry array in the entry array chain, we're likely dealing with a
2703 * corrupted journal file. */
2704 if (p == 0)
2705 return -EBADMSG;
2706
2707 *ret = q;
2708
2709 return 0;
2710 }
2711
2712 static int generic_array_get(
2713 JournalFile *f,
2714 uint64_t first,
2715 uint64_t i,
2716 direction_t direction,
2717 Object **ret_object,
2718 uint64_t *ret_offset) {
2719
2720 uint64_t a, t = 0, k;
2721 ChainCacheItem *ci;
2722 Object *o;
2723 int r;
2724
2725 assert(f);
2726
2727 /* FIXME: fix return value assignment on success. */
2728
2729 a = first;
2730
2731 /* Try the chain cache first */
2732 ci = ordered_hashmap_get(f->chain_cache, &first);
2733 if (ci && i > ci->total) {
2734 a = ci->array;
2735 i -= ci->total;
2736 t = ci->total;
2737 }
2738
2739 while (a > 0) {
2740 r = journal_file_move_to_object(f, OBJECT_ENTRY_ARRAY, a, &o);
2741 if (IN_SET(r, -EBADMSG, -EADDRNOTAVAIL)) {
2742 /* If there's corruption and we're going downwards, let's pretend we reached the
2743 * final entry in the entry array chain. */
2744
2745 if (direction == DIRECTION_DOWN)
2746 return 0;
2747
2748 /* If there's corruption and we're going upwards, move back to the previous entry
2749 * array and start iterating entries from there. */
2750
2751 r = bump_entry_array(f, NULL, a, first, DIRECTION_UP, &a);
2752 if (r < 0)
2753 return r;
2754
2755 i = UINT64_MAX;
2756
2757 break;
2758 }
2759 if (r < 0)
2760 return r;
2761
2762 k = journal_file_entry_array_n_items(f, o);
2763 if (i < k)
2764 break;
2765
2766 i -= k;
2767 t += k;
2768 a = le64toh(o->entry_array.next_entry_array_offset);
2769 }
2770
2771 /* If we've found the right location, now look for the first non-corrupt entry object (in the right
2772 * direction). */
2773
2774 while (a > 0) {
2775 /* In the first iteration of the while loop, we reuse i, k and o from the previous while
2776 * loop. */
2777 if (i == UINT64_MAX) {
2778 r = journal_file_move_to_object(f, OBJECT_ENTRY_ARRAY, a, &o);
2779 if (r < 0)
2780 return r;
2781
2782 k = journal_file_entry_array_n_items(f, o);
2783 if (k == 0)
2784 break;
2785
2786 i = direction == DIRECTION_DOWN ? 0 : k - 1;
2787 }
2788
2789 do {
2790 uint64_t p;
2791
2792 p = journal_file_entry_array_item(f, o, i);
2793
2794 r = journal_file_move_to_object(f, OBJECT_ENTRY, p, ret_object);
2795 if (r >= 0) {
2796 /* Let's cache this item for the next invocation */
2797 chain_cache_put(f->chain_cache, ci, first, a, journal_file_entry_array_item(f, o, 0), t, i);
2798
2799 if (ret_offset)
2800 *ret_offset = p;
2801
2802 return 1;
2803 }
2804 if (!IN_SET(r, -EADDRNOTAVAIL, -EBADMSG))
2805 return r;
2806
2807 /* OK, so this entry is borked. Most likely some entry didn't get synced to
2808 * disk properly, let's see if the next one might work for us instead. */
2809 log_debug_errno(r, "Entry item %" PRIu64 " is bad, skipping over it.", i);
2810
2811 r = journal_file_move_to_object(f, OBJECT_ENTRY_ARRAY, a, &o);
2812 if (r < 0)
2813 return r;
2814
2815 } while (bump_array_index(&i, direction, k) > 0);
2816
2817 r = bump_entry_array(f, o, a, first, direction, &a);
2818 if (r < 0)
2819 return r;
2820
2821 t += k;
2822 i = UINT64_MAX;
2823 }
2824
2825 return 0;
2826 }
2827
2828 static int generic_array_get_plus_one(
2829 JournalFile *f,
2830 uint64_t extra,
2831 uint64_t first,
2832 uint64_t i,
2833 direction_t direction,
2834 Object **ret_object,
2835 uint64_t *ret_offset) {
2836
2837 int r;
2838
2839 assert(f);
2840
2841 /* FIXME: fix return value assignment on success. */
2842
2843 if (i == 0) {
2844 r = journal_file_move_to_object(f, OBJECT_ENTRY, extra, ret_object);
2845 if (IN_SET(r, -EADDRNOTAVAIL, -EBADMSG))
2846 return generic_array_get(f, first, 0, direction, ret_object, ret_offset);
2847 if (r < 0)
2848 return r;
2849
2850 if (ret_offset)
2851 *ret_offset = extra;
2852
2853 return 1;
2854 }
2855
2856 return generic_array_get(f, first, i - 1, direction, ret_object, ret_offset);
2857 }
2858
2859 enum {
2860 TEST_FOUND,
2861 TEST_LEFT,
2862 TEST_RIGHT
2863 };
2864
2865 static int generic_array_bisect_one(
2866 JournalFile *f,
2867 uint64_t a, /* offset of entry array object. */
2868 uint64_t i, /* index of the entry item we will test. */
2869 uint64_t needle,
2870 int (*test_object)(JournalFile *f, uint64_t p, uint64_t needle),
2871 direction_t direction,
2872 uint64_t *left,
2873 uint64_t *right,
2874 uint64_t *ret_offset) {
2875
2876 Object *array;
2877 uint64_t p;
2878 int r;
2879
2880 assert(f);
2881 assert(test_object);
2882 assert(left);
2883 assert(right);
2884 assert(*left <= i);
2885 assert(i <= *right);
2886
2887 r = journal_file_move_to_object(f, OBJECT_ENTRY_ARRAY, a, &array);
2888 if (r < 0)
2889 return r;
2890
2891 p = journal_file_entry_array_item(f, array, i);
2892 if (p <= 0)
2893 r = -EBADMSG;
2894 else
2895 r = test_object(f, p, needle);
2896 if (IN_SET(r, -EBADMSG, -EADDRNOTAVAIL)) {
2897 log_debug_errno(r, "Encountered invalid entry while bisecting, cutting algorithm short.");
2898 *right = i;
2899 return -ENOANO; /* recognizable error */
2900 }
2901 if (r < 0)
2902 return r;
2903
2904 if (r == TEST_FOUND)
2905 r = direction == DIRECTION_DOWN ? TEST_RIGHT : TEST_LEFT;
2906
2907 if (r == TEST_RIGHT)
2908 *right = i;
2909 else
2910 *left = i + 1;
2911
2912 if (ret_offset)
2913 *ret_offset = p;
2914
2915 return r;
2916 }
2917
2918 static int generic_array_bisect(
2919 JournalFile *f,
2920 uint64_t first,
2921 uint64_t n,
2922 uint64_t needle,
2923 int (*test_object)(JournalFile *f, uint64_t p, uint64_t needle),
2924 direction_t direction,
2925 Object **ret_object,
2926 uint64_t *ret_offset,
2927 uint64_t *ret_idx) {
2928
2929 /* Given an entry array chain, this function finds the object "closest" to the given needle in the
2930 * chain, taking into account the provided direction. A function can be provided to determine how
2931 * an object is matched against the given needle.
2932 *
2933 * Given a journal file, the offset of an object and the needle, the test_object() function should
2934 * return TEST_LEFT if the needle is located earlier in the entry array chain, TEST_LEFT if the
2935 * needle is located later in the entry array chain and TEST_FOUND if the object matches the needle.
2936 * If test_object() returns TEST_FOUND for a specific object, that object's information will be used
2937 * to populate the return values of this function. If test_object() never returns TEST_FOUND, the
2938 * return values are populated with the details of one of the objects closest to the needle. If the
2939 * direction is DIRECTION_UP, the earlier object is used. Otherwise, the later object is used.
2940 */
2941
2942 uint64_t a, p, t = 0, i = 0, last_p = 0, last_index = UINT64_MAX;
2943 bool subtract_one = false;
2944 ChainCacheItem *ci;
2945 Object *array;
2946 int r;
2947
2948 assert(f);
2949 assert(test_object);
2950
2951 /* Start with the first array in the chain */
2952 a = first;
2953
2954 ci = ordered_hashmap_get(f->chain_cache, &first);
2955 if (ci && n > ci->total && ci->begin != 0) {
2956 /* Ah, we have iterated this bisection array chain previously! Let's see if we can skip ahead
2957 * in the chain, as far as the last time. But we can't jump backwards in the chain, so let's
2958 * check that first. */
2959
2960 r = test_object(f, ci->begin, needle);
2961 if (r < 0)
2962 return r;
2963
2964 if (r == TEST_LEFT) {
2965 /* OK, what we are looking for is right of the begin of this EntryArray, so let's
2966 * jump straight to previously cached array in the chain */
2967
2968 a = ci->array;
2969 n -= ci->total;
2970 t = ci->total;
2971 last_index = ci->last_index;
2972 }
2973 }
2974
2975 while (a > 0) {
2976 uint64_t left = 0, right, k, lp;
2977
2978 r = journal_file_move_to_object(f, OBJECT_ENTRY_ARRAY, a, &array);
2979 if (r < 0)
2980 return r;
2981
2982 k = journal_file_entry_array_n_items(f, array);
2983 right = MIN(k, n);
2984 if (right <= 0)
2985 return 0;
2986
2987 right--;
2988 r = generic_array_bisect_one(f, a, right, needle, test_object, direction, &left, &right, &lp);
2989 if (r == -ENOANO) {
2990 n = right;
2991 continue;
2992 }
2993 if (r < 0)
2994 return r;
2995
2996 if (r == TEST_RIGHT) {
2997 /* If we cached the last index we looked at, let's try to not to jump too wildly
2998 * around and see if we can limit the range to look at early to the immediate
2999 * neighbors of the last index we looked at. */
3000
3001 if (last_index > 0 && last_index - 1 < right) {
3002 r = generic_array_bisect_one(f, a, last_index - 1, needle, test_object, direction, &left, &right, NULL);
3003 if (r < 0 && r != -ENOANO)
3004 return r;
3005 }
3006
3007 if (last_index < right) {
3008 r = generic_array_bisect_one(f, a, last_index + 1, needle, test_object, direction, &left, &right, NULL);
3009 if (r < 0 && r != -ENOANO)
3010 return r;
3011 }
3012
3013 for (;;) {
3014 if (left == right) {
3015 if (direction == DIRECTION_UP)
3016 subtract_one = true;
3017
3018 i = left;
3019 goto found;
3020 }
3021
3022 assert(left < right);
3023 i = (left + right) / 2;
3024
3025 r = generic_array_bisect_one(f, a, i, needle, test_object, direction, &left, &right, NULL);
3026 if (r < 0 && r != -ENOANO)
3027 return r;
3028 }
3029 }
3030
3031 if (k >= n) {
3032 if (direction == DIRECTION_UP) {
3033 i = n;
3034 subtract_one = true;
3035 goto found;
3036 }
3037
3038 return 0;
3039 }
3040
3041 last_p = lp;
3042
3043 n -= k;
3044 t += k;
3045 last_index = UINT64_MAX;
3046 a = le64toh(array->entry_array.next_entry_array_offset);
3047 }
3048
3049 return 0;
3050
3051 found:
3052 if (subtract_one && t == 0 && i == 0)
3053 return 0;
3054
3055 r = journal_file_move_to_object(f, OBJECT_ENTRY_ARRAY, a, &array);
3056 if (r < 0)
3057 return r;
3058
3059 p = journal_file_entry_array_item(f, array, 0);
3060 if (p <= 0)
3061 return -EBADMSG;
3062
3063 /* Let's cache this item for the next invocation */
3064 chain_cache_put(f->chain_cache, ci, first, a, p, t, subtract_one ? (i > 0 ? i-1 : UINT64_MAX) : i);
3065
3066 if (subtract_one && i == 0)
3067 p = last_p;
3068 else if (subtract_one)
3069 p = journal_file_entry_array_item(f, array, i - 1);
3070 else
3071 p = journal_file_entry_array_item(f, array, i);
3072
3073 if (ret_object) {
3074 r = journal_file_move_to_object(f, OBJECT_ENTRY, p, ret_object);
3075 if (r < 0)
3076 return r;
3077 }
3078
3079 if (ret_offset)
3080 *ret_offset = p;
3081
3082 if (ret_idx)
3083 *ret_idx = t + i + (subtract_one ? -1 : 0);
3084
3085 return 1;
3086 }
3087
3088 static int generic_array_bisect_plus_one(
3089 JournalFile *f,
3090 uint64_t extra,
3091 uint64_t first,
3092 uint64_t n,
3093 uint64_t needle,
3094 int (*test_object)(JournalFile *f, uint64_t p, uint64_t needle),
3095 direction_t direction,
3096 Object **ret_object,
3097 uint64_t *ret_offset,
3098 uint64_t *ret_idx) {
3099
3100 int r;
3101 bool step_back = false;
3102
3103 assert(f);
3104 assert(test_object);
3105
3106 if (n <= 0)
3107 return 0;
3108
3109 /* This bisects the array in object 'first', but first checks
3110 * an extra */
3111 r = test_object(f, extra, needle);
3112 if (r < 0)
3113 return r;
3114
3115 if (r == TEST_FOUND)
3116 r = direction == DIRECTION_DOWN ? TEST_RIGHT : TEST_LEFT;
3117
3118 /* if we are looking with DIRECTION_UP then we need to first
3119 see if in the actual array there is a matching entry, and
3120 return the last one of that. But if there isn't any we need
3121 to return this one. Hence remember this, and return it
3122 below. */
3123 if (r == TEST_LEFT)
3124 step_back = direction == DIRECTION_UP;
3125
3126 if (r == TEST_RIGHT) {
3127 if (direction == DIRECTION_DOWN)
3128 goto found;
3129 else
3130 return 0;
3131 }
3132
3133 r = generic_array_bisect(f, first, n-1, needle, test_object, direction, ret_object, ret_offset, ret_idx);
3134
3135 if (r == 0 && step_back)
3136 goto found;
3137
3138 if (r > 0 && ret_idx)
3139 (*ret_idx)++;
3140
3141 return r;
3142
3143 found:
3144 if (ret_object) {
3145 r = journal_file_move_to_object(f, OBJECT_ENTRY, extra, ret_object);
3146 if (r < 0)
3147 return r;
3148 }
3149
3150 if (ret_offset)
3151 *ret_offset = extra;
3152
3153 if (ret_idx)
3154 *ret_idx = 0;
3155
3156 return 1;
3157 }
3158
3159 _pure_ static int test_object_offset(JournalFile *f, uint64_t p, uint64_t needle) {
3160 assert(f);
3161 assert(p > 0);
3162
3163 if (p == needle)
3164 return TEST_FOUND;
3165 else if (p < needle)
3166 return TEST_LEFT;
3167 else
3168 return TEST_RIGHT;
3169 }
3170
3171 int journal_file_move_to_entry_by_offset(
3172 JournalFile *f,
3173 uint64_t p,
3174 direction_t direction,
3175 Object **ret_object,
3176 uint64_t *ret_offset) {
3177
3178 assert(f);
3179 assert(f->header);
3180
3181 return generic_array_bisect(
3182 f,
3183 le64toh(f->header->entry_array_offset),
3184 le64toh(f->header->n_entries),
3185 p,
3186 test_object_offset,
3187 direction,
3188 ret_object, ret_offset, NULL);
3189 }
3190
3191 static int test_object_seqnum(JournalFile *f, uint64_t p, uint64_t needle) {
3192 uint64_t sq;
3193 Object *o;
3194 int r;
3195
3196 assert(f);
3197 assert(p > 0);
3198
3199 r = journal_file_move_to_object(f, OBJECT_ENTRY, p, &o);
3200 if (r < 0)
3201 return r;
3202
3203 sq = le64toh(READ_NOW(o->entry.seqnum));
3204 if (sq == needle)
3205 return TEST_FOUND;
3206 else if (sq < needle)
3207 return TEST_LEFT;
3208 else
3209 return TEST_RIGHT;
3210 }
3211
3212 int journal_file_move_to_entry_by_seqnum(
3213 JournalFile *f,
3214 uint64_t seqnum,
3215 direction_t direction,
3216 Object **ret_object,
3217 uint64_t *ret_offset) {
3218
3219 assert(f);
3220 assert(f->header);
3221
3222 return generic_array_bisect(
3223 f,
3224 le64toh(f->header->entry_array_offset),
3225 le64toh(f->header->n_entries),
3226 seqnum,
3227 test_object_seqnum,
3228 direction,
3229 ret_object, ret_offset, NULL);
3230 }
3231
3232 static int test_object_realtime(JournalFile *f, uint64_t p, uint64_t needle) {
3233 Object *o;
3234 uint64_t rt;
3235 int r;
3236
3237 assert(f);
3238 assert(p > 0);
3239
3240 r = journal_file_move_to_object(f, OBJECT_ENTRY, p, &o);
3241 if (r < 0)
3242 return r;
3243
3244 rt = le64toh(READ_NOW(o->entry.realtime));
3245 if (rt == needle)
3246 return TEST_FOUND;
3247 else if (rt < needle)
3248 return TEST_LEFT;
3249 else
3250 return TEST_RIGHT;
3251 }
3252
3253 int journal_file_move_to_entry_by_realtime(
3254 JournalFile *f,
3255 uint64_t realtime,
3256 direction_t direction,
3257 Object **ret_object,
3258 uint64_t *ret_offset) {
3259
3260 assert(f);
3261 assert(f->header);
3262
3263 return generic_array_bisect(
3264 f,
3265 le64toh(f->header->entry_array_offset),
3266 le64toh(f->header->n_entries),
3267 realtime,
3268 test_object_realtime,
3269 direction,
3270 ret_object, ret_offset, NULL);
3271 }
3272
3273 static int test_object_monotonic(JournalFile *f, uint64_t p, uint64_t needle) {
3274 Object *o;
3275 uint64_t m;
3276 int r;
3277
3278 assert(f);
3279 assert(p > 0);
3280
3281 r = journal_file_move_to_object(f, OBJECT_ENTRY, p, &o);
3282 if (r < 0)
3283 return r;
3284
3285 m = le64toh(READ_NOW(o->entry.monotonic));
3286 if (m == needle)
3287 return TEST_FOUND;
3288 else if (m < needle)
3289 return TEST_LEFT;
3290 else
3291 return TEST_RIGHT;
3292 }
3293
3294 static int find_data_object_by_boot_id(
3295 JournalFile *f,
3296 sd_id128_t boot_id,
3297 Object **ret_object,
3298 uint64_t *ret_offset) {
3299
3300 char t[STRLEN("_BOOT_ID=") + 32 + 1] = "_BOOT_ID=";
3301
3302 assert(f);
3303
3304 sd_id128_to_string(boot_id, t + 9);
3305 return journal_file_find_data_object(f, t, sizeof(t) - 1, ret_object, ret_offset);
3306 }
3307
3308 int journal_file_move_to_entry_by_monotonic(
3309 JournalFile *f,
3310 sd_id128_t boot_id,
3311 uint64_t monotonic,
3312 direction_t direction,
3313 Object **ret_object,
3314 uint64_t *ret_offset) {
3315
3316 Object *o;
3317 int r;
3318
3319 assert(f);
3320
3321 r = find_data_object_by_boot_id(f, boot_id, &o, NULL);
3322 if (r < 0)
3323 return r;
3324 if (r == 0)
3325 return -ENOENT;
3326
3327 return generic_array_bisect_plus_one(
3328 f,
3329 le64toh(o->data.entry_offset),
3330 le64toh(o->data.entry_array_offset),
3331 le64toh(o->data.n_entries),
3332 monotonic,
3333 test_object_monotonic,
3334 direction,
3335 ret_object, ret_offset, NULL);
3336 }
3337
3338 void journal_file_reset_location(JournalFile *f) {
3339 assert(f);
3340
3341 f->location_type = LOCATION_HEAD;
3342 f->current_offset = 0;
3343 f->current_seqnum = 0;
3344 f->current_realtime = 0;
3345 f->current_monotonic = 0;
3346 zero(f->current_boot_id);
3347 f->current_xor_hash = 0;
3348 }
3349
3350 void journal_file_save_location(JournalFile *f, Object *o, uint64_t offset) {
3351 assert(f);
3352 assert(o);
3353
3354 f->location_type = LOCATION_SEEK;
3355 f->current_offset = offset;
3356 f->current_seqnum = le64toh(o->entry.seqnum);
3357 f->current_realtime = le64toh(o->entry.realtime);
3358 f->current_monotonic = le64toh(o->entry.monotonic);
3359 f->current_boot_id = o->entry.boot_id;
3360 f->current_xor_hash = le64toh(o->entry.xor_hash);
3361 }
3362
3363 static bool check_properly_ordered(uint64_t new_offset, uint64_t old_offset, direction_t direction) {
3364
3365 /* Consider it an error if any of the two offsets is uninitialized */
3366 if (old_offset == 0 || new_offset == 0)
3367 return false;
3368
3369 /* If we go down, the new offset must be larger than the old one. */
3370 return direction == DIRECTION_DOWN ?
3371 new_offset > old_offset :
3372 new_offset < old_offset;
3373 }
3374
3375 int journal_file_next_entry(
3376 JournalFile *f,
3377 uint64_t p,
3378 direction_t direction,
3379 Object **ret_object,
3380 uint64_t *ret_offset) {
3381
3382 uint64_t i, n, ofs;
3383 int r;
3384
3385 assert(f);
3386 assert(f->header);
3387
3388 /* FIXME: fix return value assignment. */
3389
3390 n = le64toh(READ_NOW(f->header->n_entries));
3391 if (n <= 0)
3392 return 0;
3393
3394 if (p == 0)
3395 i = direction == DIRECTION_DOWN ? 0 : n - 1;
3396 else {
3397 r = generic_array_bisect(f,
3398 le64toh(f->header->entry_array_offset),
3399 le64toh(f->header->n_entries),
3400 p,
3401 test_object_offset,
3402 DIRECTION_DOWN,
3403 NULL, NULL,
3404 &i);
3405 if (r <= 0)
3406 return r;
3407
3408 r = bump_array_index(&i, direction, n);
3409 if (r <= 0)
3410 return r;
3411 }
3412
3413 /* And jump to it */
3414 r = generic_array_get(f, le64toh(f->header->entry_array_offset), i, direction, ret_object, &ofs);
3415 if (r <= 0)
3416 return r;
3417
3418 /* Ensure our array is properly ordered. */
3419 if (p > 0 && !check_properly_ordered(ofs, p, direction))
3420 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG),
3421 "%s: entry array not properly ordered at entry %" PRIu64,
3422 f->path, i);
3423
3424 if (ret_offset)
3425 *ret_offset = ofs;
3426
3427 return 1;
3428 }
3429
3430 int journal_file_next_entry_for_data(
3431 JournalFile *f,
3432 Object *d,
3433 direction_t direction,
3434 Object **ret_object,
3435 uint64_t *ret_offset) {
3436
3437 uint64_t i, n, ofs;
3438 int r;
3439
3440 assert(f);
3441 assert(d);
3442 assert(d->object.type == OBJECT_DATA);
3443
3444 /* FIXME: fix return value assignment. */
3445
3446 n = le64toh(READ_NOW(d->data.n_entries));
3447 if (n <= 0)
3448 return n;
3449
3450 i = direction == DIRECTION_DOWN ? 0 : n - 1;
3451
3452 r = generic_array_get_plus_one(f,
3453 le64toh(d->data.entry_offset),
3454 le64toh(d->data.entry_array_offset),
3455 i,
3456 direction,
3457 ret_object, &ofs);
3458 if (r <= 0)
3459 return r;
3460
3461 if (ret_offset)
3462 *ret_offset = ofs;
3463
3464 return 1;
3465 }
3466
3467 int journal_file_move_to_entry_by_offset_for_data(
3468 JournalFile *f,
3469 Object *d,
3470 uint64_t p,
3471 direction_t direction,
3472 Object **ret, uint64_t *ret_offset) {
3473
3474 assert(f);
3475 assert(d);
3476 assert(d->object.type == OBJECT_DATA);
3477
3478 return generic_array_bisect_plus_one(
3479 f,
3480 le64toh(d->data.entry_offset),
3481 le64toh(d->data.entry_array_offset),
3482 le64toh(d->data.n_entries),
3483 p,
3484 test_object_offset,
3485 direction,
3486 ret, ret_offset, NULL);
3487 }
3488
3489 int journal_file_move_to_entry_by_monotonic_for_data(
3490 JournalFile *f,
3491 Object *d,
3492 sd_id128_t boot_id,
3493 uint64_t monotonic,
3494 direction_t direction,
3495 Object **ret_object,
3496 uint64_t *ret_offset) {
3497
3498 uint64_t b, z, entry_offset, entry_array_offset, n_entries;
3499 Object *o;
3500 int r;
3501
3502 assert(f);
3503 assert(d);
3504 assert(d->object.type == OBJECT_DATA);
3505
3506 /* Save all the required data before the data object gets invalidated. */
3507 entry_offset = le64toh(READ_NOW(d->data.entry_offset));
3508 entry_array_offset = le64toh(READ_NOW(d->data.entry_array_offset));
3509 n_entries = le64toh(READ_NOW(d->data.n_entries));
3510
3511 /* First, seek by time */
3512 r = find_data_object_by_boot_id(f, boot_id, &o, &b);
3513 if (r < 0)
3514 return r;
3515 if (r == 0)
3516 return -ENOENT;
3517
3518 r = generic_array_bisect_plus_one(f,
3519 le64toh(o->data.entry_offset),
3520 le64toh(o->data.entry_array_offset),
3521 le64toh(o->data.n_entries),
3522 monotonic,
3523 test_object_monotonic,
3524 direction,
3525 NULL, &z, NULL);
3526 if (r <= 0)
3527 return r;
3528
3529 /* And now, continue seeking until we find an entry that
3530 * exists in both bisection arrays */
3531
3532 r = journal_file_move_to_object(f, OBJECT_DATA, b, &o);
3533 if (r < 0)
3534 return r;
3535
3536 for (;;) {
3537 uint64_t p, q;
3538
3539 r = generic_array_bisect_plus_one(f,
3540 entry_offset,
3541 entry_array_offset,
3542 n_entries,
3543 z,
3544 test_object_offset,
3545 direction,
3546 NULL, &p, NULL);
3547 if (r <= 0)
3548 return r;
3549
3550 r = generic_array_bisect_plus_one(f,
3551 le64toh(o->data.entry_offset),
3552 le64toh(o->data.entry_array_offset),
3553 le64toh(o->data.n_entries),
3554 p,
3555 test_object_offset,
3556 direction,
3557 NULL, &q, NULL);
3558
3559 if (r <= 0)
3560 return r;
3561
3562 if (p == q) {
3563 if (ret_object) {
3564 r = journal_file_move_to_object(f, OBJECT_ENTRY, q, ret_object);
3565 if (r < 0)
3566 return r;
3567 }
3568
3569 if (ret_offset)
3570 *ret_offset = q;
3571
3572 return 1;
3573 }
3574
3575 z = q;
3576 }
3577 }
3578
3579 int journal_file_move_to_entry_by_seqnum_for_data(
3580 JournalFile *f,
3581 Object *d,
3582 uint64_t seqnum,
3583 direction_t direction,
3584 Object **ret_object,
3585 uint64_t *ret_offset) {
3586
3587 assert(f);
3588 assert(d);
3589 assert(d->object.type == OBJECT_DATA);
3590
3591 return generic_array_bisect_plus_one(
3592 f,
3593 le64toh(d->data.entry_offset),
3594 le64toh(d->data.entry_array_offset),
3595 le64toh(d->data.n_entries),
3596 seqnum,
3597 test_object_seqnum,
3598 direction,
3599 ret_object, ret_offset, NULL);
3600 }
3601
3602 int journal_file_move_to_entry_by_realtime_for_data(
3603 JournalFile *f,
3604 Object *d,
3605 uint64_t realtime,
3606 direction_t direction,
3607 Object **ret, uint64_t *ret_offset) {
3608
3609 assert(f);
3610 assert(d);
3611 assert(d->object.type == OBJECT_DATA);
3612
3613 return generic_array_bisect_plus_one(
3614 f,
3615 le64toh(d->data.entry_offset),
3616 le64toh(d->data.entry_array_offset),
3617 le64toh(d->data.n_entries),
3618 realtime,
3619 test_object_realtime,
3620 direction,
3621 ret, ret_offset, NULL);
3622 }
3623
3624 void journal_file_dump(JournalFile *f) {
3625 Object *o;
3626 uint64_t p;
3627 int r;
3628
3629 assert(f);
3630 assert(f->header);
3631
3632 journal_file_print_header(f);
3633
3634 p = le64toh(READ_NOW(f->header->header_size));
3635 while (p != 0) {
3636 const char *s;
3637 Compression c;
3638
3639 r = journal_file_move_to_object(f, OBJECT_UNUSED, p, &o);
3640 if (r < 0)
3641 goto fail;
3642
3643 s = journal_object_type_to_string(o->object.type);
3644
3645 switch (o->object.type) {
3646
3647 case OBJECT_ENTRY:
3648 assert(s);
3649
3650 printf("Type: %s seqnum=%"PRIu64" monotonic=%"PRIu64" realtime=%"PRIu64"\n",
3651 s,
3652 le64toh(o->entry.seqnum),
3653 le64toh(o->entry.monotonic),
3654 le64toh(o->entry.realtime));
3655 break;
3656
3657 case OBJECT_TAG:
3658 assert(s);
3659
3660 printf("Type: %s seqnum=%"PRIu64" epoch=%"PRIu64"\n",
3661 s,
3662 le64toh(o->tag.seqnum),
3663 le64toh(o->tag.epoch));
3664 break;
3665
3666 default:
3667 if (s)
3668 printf("Type: %s \n", s);
3669 else
3670 printf("Type: unknown (%i)", o->object.type);
3671
3672 break;
3673 }
3674
3675 c = COMPRESSION_FROM_OBJECT(o);
3676 if (c > COMPRESSION_NONE)
3677 printf("Flags: %s\n",
3678 compression_to_string(c));
3679
3680 if (p == le64toh(f->header->tail_object_offset))
3681 p = 0;
3682 else
3683 p += ALIGN64(le64toh(o->object.size));
3684 }
3685
3686 return;
3687 fail:
3688 log_error("File corrupt");
3689 }
3690
3691 /* Note: the lifetime of the compound literal is the immediately surrounding block. */
3692 #define FORMAT_TIMESTAMP_SAFE(t) (FORMAT_TIMESTAMP(t) ?: " --- ")
3693
3694 void journal_file_print_header(JournalFile *f) {
3695 struct stat st;
3696
3697 assert(f);
3698 assert(f->header);
3699
3700 printf("File path: %s\n"
3701 "File ID: %s\n"
3702 "Machine ID: %s\n"
3703 "Boot ID: %s\n"
3704 "Sequential number ID: %s\n"
3705 "State: %s\n"
3706 "Compatible flags:%s%s%s\n"
3707 "Incompatible flags:%s%s%s%s%s%s\n"
3708 "Header size: %"PRIu64"\n"
3709 "Arena size: %"PRIu64"\n"
3710 "Data hash table size: %"PRIu64"\n"
3711 "Field hash table size: %"PRIu64"\n"
3712 "Rotate suggested: %s\n"
3713 "Head sequential number: %"PRIu64" (%"PRIx64")\n"
3714 "Tail sequential number: %"PRIu64" (%"PRIx64")\n"
3715 "Head realtime timestamp: %s (%"PRIx64")\n"
3716 "Tail realtime timestamp: %s (%"PRIx64")\n"
3717 "Tail monotonic timestamp: %s (%"PRIx64")\n"
3718 "Objects: %"PRIu64"\n"
3719 "Entry objects: %"PRIu64"\n",
3720 f->path,
3721 SD_ID128_TO_STRING(f->header->file_id),
3722 SD_ID128_TO_STRING(f->header->machine_id),
3723 SD_ID128_TO_STRING(f->header->tail_entry_boot_id),
3724 SD_ID128_TO_STRING(f->header->seqnum_id),
3725 f->header->state == STATE_OFFLINE ? "OFFLINE" :
3726 f->header->state == STATE_ONLINE ? "ONLINE" :
3727 f->header->state == STATE_ARCHIVED ? "ARCHIVED" : "UNKNOWN",
3728 JOURNAL_HEADER_SEALED(f->header) ? " SEALED" : "",
3729 JOURNAL_HEADER_TAIL_ENTRY_BOOT_ID(f->header) ? " TAIL_ENTRY_BOOT_ID" : "",
3730 (le32toh(f->header->compatible_flags) & ~HEADER_COMPATIBLE_ANY) ? " ???" : "",
3731 JOURNAL_HEADER_COMPRESSED_XZ(f->header) ? " COMPRESSED-XZ" : "",
3732 JOURNAL_HEADER_COMPRESSED_LZ4(f->header) ? " COMPRESSED-LZ4" : "",
3733 JOURNAL_HEADER_COMPRESSED_ZSTD(f->header) ? " COMPRESSED-ZSTD" : "",
3734 JOURNAL_HEADER_KEYED_HASH(f->header) ? " KEYED-HASH" : "",
3735 JOURNAL_HEADER_COMPACT(f->header) ? " COMPACT" : "",
3736 (le32toh(f->header->incompatible_flags) & ~HEADER_INCOMPATIBLE_ANY) ? " ???" : "",
3737 le64toh(f->header->header_size),
3738 le64toh(f->header->arena_size),
3739 le64toh(f->header->data_hash_table_size) / sizeof(HashItem),
3740 le64toh(f->header->field_hash_table_size) / sizeof(HashItem),
3741 yes_no(journal_file_rotate_suggested(f, 0, LOG_DEBUG)),
3742 le64toh(f->header->head_entry_seqnum), le64toh(f->header->head_entry_seqnum),
3743 le64toh(f->header->tail_entry_seqnum), le64toh(f->header->tail_entry_seqnum),
3744 FORMAT_TIMESTAMP_SAFE(le64toh(f->header->head_entry_realtime)), le64toh(f->header->head_entry_realtime),
3745 FORMAT_TIMESTAMP_SAFE(le64toh(f->header->tail_entry_realtime)), le64toh(f->header->tail_entry_realtime),
3746 FORMAT_TIMESPAN(le64toh(f->header->tail_entry_monotonic), USEC_PER_MSEC), le64toh(f->header->tail_entry_monotonic),
3747 le64toh(f->header->n_objects),
3748 le64toh(f->header->n_entries));
3749
3750 if (JOURNAL_HEADER_CONTAINS(f->header, n_data))
3751 printf("Data objects: %"PRIu64"\n"
3752 "Data hash table fill: %.1f%%\n",
3753 le64toh(f->header->n_data),
3754 100.0 * (double) le64toh(f->header->n_data) / ((double) (le64toh(f->header->data_hash_table_size) / sizeof(HashItem))));
3755
3756 if (JOURNAL_HEADER_CONTAINS(f->header, n_fields))
3757 printf("Field objects: %"PRIu64"\n"
3758 "Field hash table fill: %.1f%%\n",
3759 le64toh(f->header->n_fields),
3760 100.0 * (double) le64toh(f->header->n_fields) / ((double) (le64toh(f->header->field_hash_table_size) / sizeof(HashItem))));
3761
3762 if (JOURNAL_HEADER_CONTAINS(f->header, n_tags))
3763 printf("Tag objects: %"PRIu64"\n",
3764 le64toh(f->header->n_tags));
3765 if (JOURNAL_HEADER_CONTAINS(f->header, n_entry_arrays))
3766 printf("Entry array objects: %"PRIu64"\n",
3767 le64toh(f->header->n_entry_arrays));
3768
3769 if (JOURNAL_HEADER_CONTAINS(f->header, field_hash_chain_depth))
3770 printf("Deepest field hash chain: %" PRIu64"\n",
3771 f->header->field_hash_chain_depth);
3772
3773 if (JOURNAL_HEADER_CONTAINS(f->header, data_hash_chain_depth))
3774 printf("Deepest data hash chain: %" PRIu64"\n",
3775 f->header->data_hash_chain_depth);
3776
3777 if (fstat(f->fd, &st) >= 0)
3778 printf("Disk usage: %s\n", FORMAT_BYTES((uint64_t) st.st_blocks * 512ULL));
3779 }
3780
3781 static int journal_file_warn_btrfs(JournalFile *f) {
3782 unsigned attrs;
3783 int r;
3784
3785 assert(f);
3786
3787 /* Before we write anything, check if the COW logic is turned
3788 * off on btrfs. Given our write pattern that is quite
3789 * unfriendly to COW file systems this should greatly improve
3790 * performance on COW file systems, such as btrfs, at the
3791 * expense of data integrity features (which shouldn't be too
3792 * bad, given that we do our own checksumming). */
3793
3794 r = fd_is_fs_type(f->fd, BTRFS_SUPER_MAGIC);
3795 if (r < 0)
3796 return log_ratelimit_warning_errno(r, JOURNAL_LOG_RATELIMIT, "Failed to determine if journal is on btrfs: %m");
3797 if (!r)
3798 return 0;
3799
3800 r = read_attr_fd(f->fd, &attrs);
3801 if (r < 0)
3802 return log_ratelimit_warning_errno(r, JOURNAL_LOG_RATELIMIT, "Failed to read file attributes: %m");
3803
3804 if (attrs & FS_NOCOW_FL) {
3805 log_debug("Detected btrfs file system with copy-on-write disabled, all is good.");
3806 return 0;
3807 }
3808
3809 log_ratelimit_notice(JOURNAL_LOG_RATELIMIT,
3810 "Creating journal file %s on a btrfs file system, and copy-on-write is enabled. "
3811 "This is likely to slow down journal access substantially, please consider turning "
3812 "off the copy-on-write file attribute on the journal directory, using chattr +C.",
3813 f->path);
3814
3815 return 1;
3816 }
3817
3818 static void journal_default_metrics(JournalMetrics *m, int fd, bool compact) {
3819 struct statvfs ss;
3820 uint64_t fs_size = 0;
3821
3822 assert(m);
3823 assert(fd >= 0);
3824
3825 if (fstatvfs(fd, &ss) >= 0)
3826 fs_size = ss.f_frsize * ss.f_blocks;
3827 else
3828 log_debug_errno(errno, "Failed to determine disk size: %m");
3829
3830 if (m->max_use == UINT64_MAX) {
3831
3832 if (fs_size > 0)
3833 m->max_use = CLAMP(PAGE_ALIGN(fs_size / 10), /* 10% of file system size */
3834 MAX_USE_LOWER, MAX_USE_UPPER);
3835 else
3836 m->max_use = MAX_USE_LOWER;
3837 } else {
3838 m->max_use = PAGE_ALIGN(m->max_use);
3839
3840 if (m->max_use != 0 && m->max_use < JOURNAL_FILE_SIZE_MIN*2)
3841 m->max_use = JOURNAL_FILE_SIZE_MIN*2;
3842 }
3843
3844 if (m->min_use == UINT64_MAX) {
3845 if (fs_size > 0)
3846 m->min_use = CLAMP(PAGE_ALIGN(fs_size / 50), /* 2% of file system size */
3847 MIN_USE_LOW, MIN_USE_HIGH);
3848 else
3849 m->min_use = MIN_USE_LOW;
3850 }
3851
3852 if (m->min_use > m->max_use)
3853 m->min_use = m->max_use;
3854
3855 if (m->max_size == UINT64_MAX)
3856 m->max_size = MIN(PAGE_ALIGN(m->max_use / 8), /* 8 chunks */
3857 MAX_SIZE_UPPER);
3858 else
3859 m->max_size = PAGE_ALIGN(m->max_size);
3860
3861 if (compact && m->max_size > JOURNAL_COMPACT_SIZE_MAX)
3862 m->max_size = JOURNAL_COMPACT_SIZE_MAX;
3863
3864 if (m->max_size != 0) {
3865 if (m->max_size < JOURNAL_FILE_SIZE_MIN)
3866 m->max_size = JOURNAL_FILE_SIZE_MIN;
3867
3868 if (m->max_use != 0 && m->max_size*2 > m->max_use)
3869 m->max_use = m->max_size*2;
3870 }
3871
3872 if (m->min_size == UINT64_MAX)
3873 m->min_size = JOURNAL_FILE_SIZE_MIN;
3874 else
3875 m->min_size = CLAMP(PAGE_ALIGN(m->min_size),
3876 JOURNAL_FILE_SIZE_MIN,
3877 m->max_size ?: UINT64_MAX);
3878
3879 if (m->keep_free == UINT64_MAX) {
3880 if (fs_size > 0)
3881 m->keep_free = MIN(PAGE_ALIGN(fs_size / 20), /* 5% of file system size */
3882 KEEP_FREE_UPPER);
3883 else
3884 m->keep_free = DEFAULT_KEEP_FREE;
3885 }
3886
3887 if (m->n_max_files == UINT64_MAX)
3888 m->n_max_files = DEFAULT_N_MAX_FILES;
3889
3890 log_debug("Fixed min_use=%s max_use=%s max_size=%s min_size=%s keep_free=%s n_max_files=%" PRIu64,
3891 FORMAT_BYTES(m->min_use),
3892 FORMAT_BYTES(m->max_use),
3893 FORMAT_BYTES(m->max_size),
3894 FORMAT_BYTES(m->min_size),
3895 FORMAT_BYTES(m->keep_free),
3896 m->n_max_files);
3897 }
3898
3899 int journal_file_open(
3900 int fd,
3901 const char *fname,
3902 int open_flags,
3903 JournalFileFlags file_flags,
3904 mode_t mode,
3905 uint64_t compress_threshold_bytes,
3906 JournalMetrics *metrics,
3907 MMapCache *mmap_cache,
3908 JournalFile *template,
3909 JournalFile **ret) {
3910
3911 bool newly_created = false;
3912 JournalFile *f;
3913 void *h;
3914 int r;
3915
3916 assert(fd >= 0 || fname);
3917 assert(file_flags >= 0);
3918 assert(file_flags <= _JOURNAL_FILE_FLAGS_MAX);
3919 assert(mmap_cache);
3920 assert(ret);
3921
3922 if (!IN_SET((open_flags & O_ACCMODE), O_RDONLY, O_RDWR))
3923 return -EINVAL;
3924
3925 if ((open_flags & O_ACCMODE) == O_RDONLY && FLAGS_SET(open_flags, O_CREAT))
3926 return -EINVAL;
3927
3928 if (fname && (open_flags & O_CREAT) && !endswith(fname, ".journal"))
3929 return -EINVAL;
3930
3931 f = new(JournalFile, 1);
3932 if (!f)
3933 return -ENOMEM;
3934
3935 *f = (JournalFile) {
3936 .fd = fd,
3937 .mode = mode,
3938 .open_flags = open_flags,
3939 .compress_threshold_bytes = compress_threshold_bytes == UINT64_MAX ?
3940 DEFAULT_COMPRESS_THRESHOLD :
3941 MAX(MIN_COMPRESS_THRESHOLD, compress_threshold_bytes),
3942 .strict_order = FLAGS_SET(file_flags, JOURNAL_STRICT_ORDER),
3943 .newest_boot_id_prioq_idx = PRIOQ_IDX_NULL,
3944 };
3945
3946 if (fname) {
3947 f->path = strdup(fname);
3948 if (!f->path) {
3949 r = -ENOMEM;
3950 goto fail;
3951 }
3952 } else {
3953 assert(fd >= 0);
3954
3955 /* If we don't know the path, fill in something explanatory and vaguely useful */
3956 if (asprintf(&f->path, "/proc/self/%i", fd) < 0) {
3957 r = -ENOMEM;
3958 goto fail;
3959 }
3960 }
3961
3962 f->chain_cache = ordered_hashmap_new(&uint64_hash_ops);
3963 if (!f->chain_cache) {
3964 r = -ENOMEM;
3965 goto fail;
3966 }
3967
3968 if (f->fd < 0) {
3969 /* We pass O_NONBLOCK here, so that in case somebody pointed us to some character device node or FIFO
3970 * or so, we likely fail quickly than block for long. For regular files O_NONBLOCK has no effect, hence
3971 * it doesn't hurt in that case. */
3972
3973 f->fd = openat_report_new(AT_FDCWD, f->path, f->open_flags|O_CLOEXEC|O_NONBLOCK, f->mode, &newly_created);
3974 if (f->fd < 0) {
3975 r = f->fd;
3976 goto fail;
3977 }
3978
3979 /* fds we opened here by us should also be closed by us. */
3980 f->close_fd = true;
3981
3982 r = fd_nonblock(f->fd, false);
3983 if (r < 0)
3984 goto fail;
3985
3986 if (!newly_created) {
3987 r = journal_file_fstat(f);
3988 if (r < 0)
3989 goto fail;
3990 }
3991 } else {
3992 r = journal_file_fstat(f);
3993 if (r < 0)
3994 goto fail;
3995
3996 /* If we just got the fd passed in, we don't really know if we created the file anew */
3997 newly_created = f->last_stat.st_size == 0 && journal_file_writable(f);
3998 }
3999
4000 f->cache_fd = mmap_cache_add_fd(mmap_cache, f->fd, mmap_prot_from_open_flags(open_flags));
4001 if (!f->cache_fd) {
4002 r = -ENOMEM;
4003 goto fail;
4004 }
4005
4006 if (newly_created) {
4007 (void) journal_file_warn_btrfs(f);
4008
4009 /* Let's attach the creation time to the journal file, so that the vacuuming code knows the age of this
4010 * file even if the file might end up corrupted one day... Ideally we'd just use the creation time many
4011 * file systems maintain for each file, but the API to query this is very new, hence let's emulate this
4012 * via extended attributes. If extended attributes are not supported we'll just skip this, and rely
4013 * solely on mtime/atime/ctime of the file. */
4014 (void) fd_setcrtime(f->fd, 0);
4015
4016 r = journal_file_init_header(f, file_flags, template);
4017 if (r < 0)
4018 goto fail;
4019
4020 r = journal_file_fstat(f);
4021 if (r < 0)
4022 goto fail;
4023 }
4024
4025 if (f->last_stat.st_size < (off_t) HEADER_SIZE_MIN) {
4026 r = -ENODATA;
4027 goto fail;
4028 }
4029
4030 r = mmap_cache_fd_get(f->cache_fd, CONTEXT_HEADER, true, 0, PAGE_ALIGN(sizeof(Header)), &f->last_stat, &h);
4031 if (r == -EINVAL) {
4032 /* Some file systems (jffs2 or p9fs) don't support mmap() properly (or only read-only
4033 * mmap()), and return EINVAL in that case. Let's propagate that as a more recognizable error
4034 * code. */
4035 r = -EAFNOSUPPORT;
4036 goto fail;
4037 }
4038 if (r < 0)
4039 goto fail;
4040
4041 f->header = h;
4042
4043 if (!newly_created) {
4044 r = journal_file_verify_header(f);
4045 if (r < 0)
4046 goto fail;
4047 }
4048
4049 #if HAVE_GCRYPT
4050 if (!newly_created && journal_file_writable(f) && JOURNAL_HEADER_SEALED(f->header)) {
4051 r = journal_file_fss_load(f);
4052 if (r < 0)
4053 goto fail;
4054 }
4055 #endif
4056
4057 if (journal_file_writable(f)) {
4058 if (metrics) {
4059 journal_default_metrics(metrics, f->fd, JOURNAL_HEADER_COMPACT(f->header));
4060 f->metrics = *metrics;
4061 } else if (template)
4062 f->metrics = template->metrics;
4063
4064 r = journal_file_refresh_header(f);
4065 if (r < 0)
4066 goto fail;
4067 }
4068
4069 #if HAVE_GCRYPT
4070 r = journal_file_hmac_setup(f);
4071 if (r < 0)
4072 goto fail;
4073 #endif
4074
4075 if (newly_created) {
4076 r = journal_file_setup_field_hash_table(f);
4077 if (r < 0)
4078 goto fail;
4079
4080 r = journal_file_setup_data_hash_table(f);
4081 if (r < 0)
4082 goto fail;
4083
4084 #if HAVE_GCRYPT
4085 r = journal_file_append_first_tag(f);
4086 if (r < 0)
4087 goto fail;
4088 #endif
4089 }
4090
4091 if (mmap_cache_fd_got_sigbus(f->cache_fd)) {
4092 r = -EIO;
4093 goto fail;
4094 }
4095
4096 if (template && template->post_change_timer) {
4097 r = journal_file_enable_post_change_timer(
4098 f,
4099 sd_event_source_get_event(template->post_change_timer),
4100 template->post_change_timer_period);
4101
4102 if (r < 0)
4103 goto fail;
4104 }
4105
4106 /* The file is opened now successfully, thus we take possession of any passed in fd. */
4107 f->close_fd = true;
4108
4109 if (DEBUG_LOGGING) {
4110 static int last_seal = -1, last_keyed_hash = -1;
4111 static Compression last_compression = _COMPRESSION_INVALID;
4112 static uint64_t last_bytes = UINT64_MAX;
4113
4114 if (last_seal != JOURNAL_HEADER_SEALED(f->header) ||
4115 last_keyed_hash != JOURNAL_HEADER_KEYED_HASH(f->header) ||
4116 last_compression != JOURNAL_FILE_COMPRESSION(f) ||
4117 last_bytes != f->compress_threshold_bytes) {
4118
4119 log_debug("Journal effective settings seal=%s keyed_hash=%s compress=%s compress_threshold_bytes=%s",
4120 yes_no(JOURNAL_HEADER_SEALED(f->header)), yes_no(JOURNAL_HEADER_KEYED_HASH(f->header)),
4121 compression_to_string(JOURNAL_FILE_COMPRESSION(f)), FORMAT_BYTES(f->compress_threshold_bytes));
4122 last_seal = JOURNAL_HEADER_SEALED(f->header);
4123 last_keyed_hash = JOURNAL_HEADER_KEYED_HASH(f->header);
4124 last_compression = JOURNAL_FILE_COMPRESSION(f);
4125 last_bytes = f->compress_threshold_bytes;
4126 }
4127 }
4128
4129 *ret = f;
4130 return 0;
4131
4132 fail:
4133 if (f->cache_fd && mmap_cache_fd_got_sigbus(f->cache_fd))
4134 r = -EIO;
4135
4136 (void) journal_file_close(f);
4137
4138 if (newly_created && fd < 0)
4139 (void) unlink(fname);
4140
4141 return r;
4142 }
4143
4144 int journal_file_parse_uid_from_filename(const char *path, uid_t *ret_uid) {
4145 _cleanup_free_ char *buf = NULL, *p = NULL;
4146 const char *a, *b, *at;
4147 int r;
4148
4149 /* This helper returns -EREMOTE when the filename doesn't match user online/offline journal
4150 * pattern. Hence it currently doesn't parse archived or disposed user journals. */
4151
4152 assert(path);
4153 assert(ret_uid);
4154
4155 r = path_extract_filename(path, &p);
4156 if (r < 0)
4157 return r;
4158 if (r == O_DIRECTORY)
4159 return -EISDIR;
4160
4161 a = startswith(p, "user-");
4162 if (!a)
4163 return -EREMOTE;
4164 b = endswith(p, ".journal");
4165 if (!b)
4166 return -EREMOTE;
4167
4168 at = strchr(a, '@');
4169 if (at)
4170 return -EREMOTE;
4171
4172 buf = strndup(a, b-a);
4173 if (!buf)
4174 return -ENOMEM;
4175
4176 return parse_uid(buf, ret_uid);
4177 }
4178
4179 int journal_file_archive(JournalFile *f, char **ret_previous_path) {
4180 _cleanup_free_ char *p = NULL;
4181
4182 assert(f);
4183
4184 if (!journal_file_writable(f))
4185 return -EINVAL;
4186
4187 /* Is this a journal file that was passed to us as fd? If so, we synthesized a path name for it, and we refuse
4188 * rotation, since we don't know the actual path, and couldn't rename the file hence. */
4189 if (path_startswith(f->path, "/proc/self/fd"))
4190 return -EINVAL;
4191
4192 if (!endswith(f->path, ".journal"))
4193 return -EINVAL;
4194
4195 if (asprintf(&p, "%.*s@" SD_ID128_FORMAT_STR "-%016"PRIx64"-%016"PRIx64".journal",
4196 (int) strlen(f->path) - 8, f->path,
4197 SD_ID128_FORMAT_VAL(f->header->seqnum_id),
4198 le64toh(f->header->head_entry_seqnum),
4199 le64toh(f->header->head_entry_realtime)) < 0)
4200 return -ENOMEM;
4201
4202 /* Try to rename the file to the archived version. If the file already was deleted, we'll get ENOENT, let's
4203 * ignore that case. */
4204 if (rename(f->path, p) < 0 && errno != ENOENT)
4205 return -errno;
4206
4207 /* Sync the rename to disk */
4208 (void) fsync_directory_of_file(f->fd);
4209
4210 if (ret_previous_path)
4211 *ret_previous_path = f->path;
4212 else
4213 free(f->path);
4214
4215 f->path = TAKE_PTR(p);
4216
4217 /* Set as archive so offlining commits w/state=STATE_ARCHIVED. Previously we would set old_file->header->state
4218 * to STATE_ARCHIVED directly here, but journal_file_set_offline() short-circuits when state != STATE_ONLINE,
4219 * which would result in the rotated journal never getting fsync() called before closing. Now we simply queue
4220 * the archive state by setting an archive bit, leaving the state as STATE_ONLINE so proper offlining
4221 * occurs. */
4222 f->archive = true;
4223
4224 return 0;
4225 }
4226
4227 int journal_file_dispose(int dir_fd, const char *fname) {
4228 _cleanup_free_ char *p = NULL;
4229
4230 assert(fname);
4231
4232 /* Renames a journal file to *.journal~, i.e. to mark it as corrupted or otherwise uncleanly shutdown. Note that
4233 * this is done without looking into the file or changing any of its contents. The idea is that this is called
4234 * whenever something is suspicious and we want to move the file away and make clear that it is not accessed
4235 * for writing anymore. */
4236
4237 if (!endswith(fname, ".journal"))
4238 return -EINVAL;
4239
4240 if (asprintf(&p, "%.*s@%016" PRIx64 "-%016" PRIx64 ".journal~",
4241 (int) strlen(fname) - 8, fname,
4242 now(CLOCK_REALTIME),
4243 random_u64()) < 0)
4244 return -ENOMEM;
4245
4246 if (renameat(dir_fd, fname, dir_fd, p) < 0)
4247 return -errno;
4248
4249 return 0;
4250 }
4251
4252 int journal_file_copy_entry(
4253 JournalFile *from,
4254 JournalFile *to,
4255 Object *o,
4256 uint64_t p,
4257 uint64_t *seqnum,
4258 sd_id128_t *seqnum_id) {
4259
4260 _cleanup_free_ EntryItem *items_alloc = NULL;
4261 EntryItem *items;
4262 uint64_t n, m = 0, xor_hash = 0;
4263 sd_id128_t boot_id;
4264 dual_timestamp ts;
4265 int r;
4266
4267 assert(from);
4268 assert(to);
4269 assert(o);
4270 assert(p > 0);
4271
4272 if (!journal_file_writable(to))
4273 return -EPERM;
4274
4275 ts = (dual_timestamp) {
4276 .monotonic = le64toh(o->entry.monotonic),
4277 .realtime = le64toh(o->entry.realtime),
4278 };
4279 boot_id = o->entry.boot_id;
4280
4281 n = journal_file_entry_n_items(from, o);
4282 if (n == 0)
4283 return 0;
4284
4285 if (n < ALLOCA_MAX / sizeof(EntryItem) / 2)
4286 items = newa(EntryItem, n);
4287 else {
4288 items_alloc = new(EntryItem, n);
4289 if (!items_alloc)
4290 return -ENOMEM;
4291
4292 items = items_alloc;
4293 }
4294
4295 for (uint64_t i = 0; i < n; i++) {
4296 uint64_t h, q;
4297 void *data;
4298 size_t l;
4299 Object *u;
4300
4301 q = journal_file_entry_item_object_offset(from, o, i);
4302 r = journal_file_data_payload(from, NULL, q, NULL, 0, 0, &data, &l);
4303 if (IN_SET(r, -EADDRNOTAVAIL, -EBADMSG)) {
4304 log_debug_errno(r, "Entry item %"PRIu64" data object is bad, skipping over it: %m", i);
4305 goto next;
4306 }
4307 if (r < 0)
4308 return r;
4309 assert(r > 0);
4310
4311 if (l == 0)
4312 return -EBADMSG;
4313
4314 r = journal_file_append_data(to, data, l, &u, &h);
4315 if (r < 0)
4316 return r;
4317
4318 if (JOURNAL_HEADER_KEYED_HASH(to->header))
4319 xor_hash ^= jenkins_hash64(data, l);
4320 else
4321 xor_hash ^= le64toh(u->data.hash);
4322
4323 items[m++] = (EntryItem) {
4324 .object_offset = h,
4325 .hash = le64toh(u->data.hash),
4326 };
4327
4328 next:
4329 /* The above journal_file_data_payload() may clear or overwrite cached object. Hence, we need
4330 * to re-read the object from the cache. */
4331 r = journal_file_move_to_object(from, OBJECT_ENTRY, p, &o);
4332 if (r < 0)
4333 return r;
4334 }
4335
4336 if (m == 0)
4337 return 0;
4338
4339 r = journal_file_append_entry_internal(
4340 to,
4341 &ts,
4342 &boot_id,
4343 &from->header->machine_id,
4344 xor_hash,
4345 items,
4346 m,
4347 seqnum,
4348 seqnum_id,
4349 /* ret_object= */ NULL,
4350 /* ret_offset= */ NULL);
4351
4352 if (mmap_cache_fd_got_sigbus(to->cache_fd))
4353 return -EIO;
4354
4355 return r;
4356 }
4357
4358 void journal_reset_metrics(JournalMetrics *m) {
4359 assert(m);
4360
4361 /* Set everything to "pick automatic values". */
4362
4363 *m = (JournalMetrics) {
4364 .min_use = UINT64_MAX,
4365 .max_use = UINT64_MAX,
4366 .min_size = UINT64_MAX,
4367 .max_size = UINT64_MAX,
4368 .keep_free = UINT64_MAX,
4369 .n_max_files = UINT64_MAX,
4370 };
4371 }
4372
4373 int journal_file_get_cutoff_realtime_usec(JournalFile *f, usec_t *ret_from, usec_t *ret_to) {
4374 assert(f);
4375 assert(f->header);
4376 assert(ret_from || ret_to);
4377
4378 if (ret_from) {
4379 if (f->header->head_entry_realtime == 0)
4380 return -ENOENT;
4381
4382 *ret_from = le64toh(f->header->head_entry_realtime);
4383 }
4384
4385 if (ret_to) {
4386 if (f->header->tail_entry_realtime == 0)
4387 return -ENOENT;
4388
4389 *ret_to = le64toh(f->header->tail_entry_realtime);
4390 }
4391
4392 return 1;
4393 }
4394
4395 int journal_file_get_cutoff_monotonic_usec(JournalFile *f, sd_id128_t boot_id, usec_t *ret_from, usec_t *ret_to) {
4396 Object *o;
4397 uint64_t p;
4398 int r;
4399
4400 assert(f);
4401 assert(ret_from || ret_to);
4402
4403 /* FIXME: fix return value assignment on success with 0. */
4404
4405 r = find_data_object_by_boot_id(f, boot_id, &o, &p);
4406 if (r <= 0)
4407 return r;
4408
4409 if (le64toh(o->data.n_entries) <= 0)
4410 return 0;
4411
4412 if (ret_from) {
4413 r = journal_file_move_to_object(f, OBJECT_ENTRY, le64toh(o->data.entry_offset), &o);
4414 if (r < 0)
4415 return r;
4416
4417 *ret_from = le64toh(o->entry.monotonic);
4418 }
4419
4420 if (ret_to) {
4421 r = journal_file_move_to_object(f, OBJECT_DATA, p, &o);
4422 if (r < 0)
4423 return r;
4424
4425 r = generic_array_get_plus_one(f,
4426 le64toh(o->data.entry_offset),
4427 le64toh(o->data.entry_array_offset),
4428 le64toh(o->data.n_entries) - 1,
4429 DIRECTION_UP,
4430 &o, NULL);
4431 if (r <= 0)
4432 return r;
4433
4434 *ret_to = le64toh(o->entry.monotonic);
4435 }
4436
4437 return 1;
4438 }
4439
4440 bool journal_file_rotate_suggested(JournalFile *f, usec_t max_file_usec, int log_level) {
4441 assert(f);
4442 assert(f->header);
4443
4444 /* If we gained new header fields we gained new features,
4445 * hence suggest a rotation */
4446 if (le64toh(f->header->header_size) < sizeof(Header)) {
4447 log_ratelimit_full(log_level, JOURNAL_LOG_RATELIMIT,
4448 "%s uses an outdated header, suggesting rotation.", f->path);
4449 return true;
4450 }
4451
4452 /* Let's check if the hash tables grew over a certain fill level (75%, borrowing this value from
4453 * Java's hash table implementation), and if so suggest a rotation. To calculate the fill level we
4454 * need the n_data field, which only exists in newer versions. */
4455
4456 if (JOURNAL_HEADER_CONTAINS(f->header, n_data))
4457 if (le64toh(f->header->n_data) * 4ULL > (le64toh(f->header->data_hash_table_size) / sizeof(HashItem)) * 3ULL) {
4458 log_ratelimit_full(
4459 log_level, JOURNAL_LOG_RATELIMIT,
4460 "Data hash table of %s has a fill level at %.1f (%"PRIu64" of %"PRIu64" items, %"PRIu64" file size, %"PRIu64" bytes per hash table item), suggesting rotation.",
4461 f->path,
4462 100.0 * (double) le64toh(f->header->n_data) / ((double) (le64toh(f->header->data_hash_table_size) / sizeof(HashItem))),
4463 le64toh(f->header->n_data),
4464 le64toh(f->header->data_hash_table_size) / sizeof(HashItem),
4465 (uint64_t) f->last_stat.st_size,
4466 f->last_stat.st_size / le64toh(f->header->n_data));
4467 return true;
4468 }
4469
4470 if (JOURNAL_HEADER_CONTAINS(f->header, n_fields))
4471 if (le64toh(f->header->n_fields) * 4ULL > (le64toh(f->header->field_hash_table_size) / sizeof(HashItem)) * 3ULL) {
4472 log_ratelimit_full(
4473 log_level, JOURNAL_LOG_RATELIMIT,
4474 "Field hash table of %s has a fill level at %.1f (%"PRIu64" of %"PRIu64" items), suggesting rotation.",
4475 f->path,
4476 100.0 * (double) le64toh(f->header->n_fields) / ((double) (le64toh(f->header->field_hash_table_size) / sizeof(HashItem))),
4477 le64toh(f->header->n_fields),
4478 le64toh(f->header->field_hash_table_size) / sizeof(HashItem));
4479 return true;
4480 }
4481
4482 /* If there are too many hash collisions somebody is most likely playing games with us. Hence, if our
4483 * longest chain is longer than some threshold, let's suggest rotation. */
4484 if (JOURNAL_HEADER_CONTAINS(f->header, data_hash_chain_depth) &&
4485 le64toh(f->header->data_hash_chain_depth) > HASH_CHAIN_DEPTH_MAX) {
4486 log_ratelimit_full(
4487 log_level, JOURNAL_LOG_RATELIMIT,
4488 "Data hash table of %s has deepest hash chain of length %" PRIu64 ", suggesting rotation.",
4489 f->path, le64toh(f->header->data_hash_chain_depth));
4490 return true;
4491 }
4492
4493 if (JOURNAL_HEADER_CONTAINS(f->header, field_hash_chain_depth) &&
4494 le64toh(f->header->field_hash_chain_depth) > HASH_CHAIN_DEPTH_MAX) {
4495 log_ratelimit_full(
4496 log_level, JOURNAL_LOG_RATELIMIT,
4497 "Field hash table of %s has deepest hash chain of length at %" PRIu64 ", suggesting rotation.",
4498 f->path, le64toh(f->header->field_hash_chain_depth));
4499 return true;
4500 }
4501
4502 /* Are the data objects properly indexed by field objects? */
4503 if (JOURNAL_HEADER_CONTAINS(f->header, n_data) &&
4504 JOURNAL_HEADER_CONTAINS(f->header, n_fields) &&
4505 le64toh(f->header->n_data) > 0 &&
4506 le64toh(f->header->n_fields) == 0) {
4507 log_ratelimit_full(
4508 log_level, JOURNAL_LOG_RATELIMIT,
4509 "Data objects of %s are not indexed by field objects, suggesting rotation.",
4510 f->path);
4511 return true;
4512 }
4513
4514 if (max_file_usec > 0) {
4515 usec_t t, h;
4516
4517 h = le64toh(f->header->head_entry_realtime);
4518 t = now(CLOCK_REALTIME);
4519
4520 if (h > 0 && t > h + max_file_usec) {
4521 log_ratelimit_full(
4522 log_level, JOURNAL_LOG_RATELIMIT,
4523 "Oldest entry in %s is older than the configured file retention duration (%s), suggesting rotation.",
4524 f->path, FORMAT_TIMESPAN(max_file_usec, USEC_PER_SEC));
4525 return true;
4526 }
4527 }
4528
4529 return false;
4530 }
4531
4532 static const char * const journal_object_type_table[] = {
4533 [OBJECT_UNUSED] = "unused",
4534 [OBJECT_DATA] = "data",
4535 [OBJECT_FIELD] = "field",
4536 [OBJECT_ENTRY] = "entry",
4537 [OBJECT_DATA_HASH_TABLE] = "data hash table",
4538 [OBJECT_FIELD_HASH_TABLE] = "field hash table",
4539 [OBJECT_ENTRY_ARRAY] = "entry array",
4540 [OBJECT_TAG] = "tag",
4541 };
4542
4543 DEFINE_STRING_TABLE_LOOKUP_TO_STRING(journal_object_type, ObjectType);
Unnamed repository; edit this file 'description' to name the repository.