]> 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 #30464 from CodethinkLabs/misc-integration-test-fixes
[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 * U64_KB) /* 512 KiB */
52 #define JOURNAL_COMPACT_SIZE_MAX ((uint64_t) UINT32_MAX) /* 4 GiB */
53
54 /* These are the lower and upper bounds if we deduce the max_use value from the file system size */
55 #define MAX_USE_LOWER (1 * U64_MB) /* 1 MiB */
56 #define MAX_USE_UPPER (4 * U64_GB) /* 4 GiB */
57
58 /* Those are the lower and upper bounds for the minimal use limit,
59 * i.e. how much we'll use even if keep_free suggests otherwise. */
60 #define MIN_USE_LOW (1 * U64_MB) /* 1 MiB */
61 #define MIN_USE_HIGH (16 * U64_MB) /* 16 MiB */
62
63 /* This is the upper bound if we deduce max_size from max_use */
64 #define MAX_SIZE_UPPER (128 * U64_MB) /* 128 MiB */
65
66 /* This is the upper bound if we deduce the keep_free value from the file system size */
67 #define KEEP_FREE_UPPER (4 * U64_GB) /* 4 GiB */
68
69 /* This is the keep_free value when we can't determine the system size */
70 #define DEFAULT_KEEP_FREE (1 * U64_MB) /* 1 MB */
71
72 /* This is the default maximum number of journal files to keep around. */
73 #define DEFAULT_N_MAX_FILES 100
74
75 /* n_data was the first entry we added after the initial file format design */
76 #define HEADER_SIZE_MIN ALIGN64(offsetof(Header, n_data))
77
78 /* How many entries to keep in the entry array chain cache at max */
79 #define CHAIN_CACHE_MAX 20
80
81 /* How much to increase the journal file size at once each time we allocate something new. */
82 #define FILE_SIZE_INCREASE (8 * U64_MB) /* 8MB */
83
84 /* Reread fstat() of the file for detecting deletions at least this often */
85 #define LAST_STAT_REFRESH_USEC (5*USEC_PER_SEC)
86
87 /* Longest hash chain to rotate after */
88 #define HASH_CHAIN_DEPTH_MAX 100
89
90 #ifdef __clang__
91 # pragma GCC diagnostic ignored "-Waddress-of-packed-member"
92 #endif
93
94 static int mmap_prot_from_open_flags(int flags) {
95 switch (flags & O_ACCMODE) {
96 case O_RDONLY:
97 return PROT_READ;
98 case O_WRONLY:
99 return PROT_WRITE;
100 case O_RDWR:
101 return PROT_READ|PROT_WRITE;
102 default:
103 assert_not_reached();
104 }
105 }
106
107 int journal_file_tail_end_by_pread(JournalFile *f, uint64_t *ret_offset) {
108 uint64_t p;
109 int r;
110
111 assert(f);
112 assert(f->header);
113 assert(ret_offset);
114
115 /* Same as journal_file_tail_end_by_mmap() below, but operates with pread() to avoid the mmap cache
116 * (and thus is thread safe) */
117
118 p = le64toh(f->header->tail_object_offset);
119 if (p == 0)
120 p = le64toh(f->header->header_size);
121 else {
122 Object tail;
123 uint64_t sz;
124
125 r = journal_file_read_object_header(f, OBJECT_UNUSED, p, &tail);
126 if (r < 0)
127 return r;
128
129 sz = le64toh(tail.object.size);
130 if (sz > UINT64_MAX - sizeof(uint64_t) + 1)
131 return -EBADMSG;
132
133 sz = ALIGN64(sz);
134 if (p > UINT64_MAX - sz)
135 return -EBADMSG;
136
137 p += sz;
138 }
139
140 *ret_offset = p;
141
142 return 0;
143 }
144
145 int journal_file_tail_end_by_mmap(JournalFile *f, uint64_t *ret_offset) {
146 uint64_t p;
147 int r;
148
149 assert(f);
150 assert(f->header);
151 assert(ret_offset);
152
153 /* Same as journal_file_tail_end_by_pread() above, but operates with the usual mmap logic */
154
155 p = le64toh(f->header->tail_object_offset);
156 if (p == 0)
157 p = le64toh(f->header->header_size);
158 else {
159 Object *tail;
160 uint64_t sz;
161
162 r = journal_file_move_to_object(f, OBJECT_UNUSED, p, &tail);
163 if (r < 0)
164 return r;
165
166 sz = le64toh(READ_NOW(tail->object.size));
167 if (sz > UINT64_MAX - sizeof(uint64_t) + 1)
168 return -EBADMSG;
169
170 sz = ALIGN64(sz);
171 if (p > UINT64_MAX - sz)
172 return -EBADMSG;
173
174 p += sz;
175 }
176
177 *ret_offset = p;
178
179 return 0;
180 }
181
182 int journal_file_set_offline_thread_join(JournalFile *f) {
183 int r;
184
185 assert(f);
186
187 if (f->offline_state == OFFLINE_JOINED)
188 return 0;
189
190 r = pthread_join(f->offline_thread, NULL);
191 if (r)
192 return -r;
193
194 f->offline_state = OFFLINE_JOINED;
195
196 if (mmap_cache_fd_got_sigbus(f->cache_fd))
197 return -EIO;
198
199 return 0;
200 }
201
202 static int journal_file_set_online(JournalFile *f) {
203 bool wait = true;
204
205 assert(f);
206
207 if (!journal_file_writable(f))
208 return -EPERM;
209
210 if (f->fd < 0 || !f->header)
211 return -EINVAL;
212
213 while (wait) {
214 switch (f->offline_state) {
215 case OFFLINE_JOINED:
216 /* No offline thread, no need to wait. */
217 wait = false;
218 break;
219
220 case OFFLINE_SYNCING: {
221 OfflineState tmp_state = OFFLINE_SYNCING;
222 if (!__atomic_compare_exchange_n(&f->offline_state, &tmp_state, OFFLINE_CANCEL,
223 false, __ATOMIC_SEQ_CST, __ATOMIC_SEQ_CST))
224 continue;
225 }
226 /* Canceled syncing prior to offlining, no need to wait. */
227 wait = false;
228 break;
229
230 case OFFLINE_AGAIN_FROM_SYNCING: {
231 OfflineState tmp_state = OFFLINE_AGAIN_FROM_SYNCING;
232 if (!__atomic_compare_exchange_n(&f->offline_state, &tmp_state, OFFLINE_CANCEL,
233 false, __ATOMIC_SEQ_CST, __ATOMIC_SEQ_CST))
234 continue;
235 }
236 /* Canceled restart from syncing, no need to wait. */
237 wait = false;
238 break;
239
240 case OFFLINE_AGAIN_FROM_OFFLINING: {
241 OfflineState tmp_state = OFFLINE_AGAIN_FROM_OFFLINING;
242 if (!__atomic_compare_exchange_n(&f->offline_state, &tmp_state, OFFLINE_CANCEL,
243 false, __ATOMIC_SEQ_CST, __ATOMIC_SEQ_CST))
244 continue;
245 }
246 /* Canceled restart from offlining, must wait for offlining to complete however. */
247 _fallthrough_;
248 default: {
249 int r;
250
251 r = journal_file_set_offline_thread_join(f);
252 if (r < 0)
253 return r;
254
255 wait = false;
256 break;
257 }
258 }
259 }
260
261 if (mmap_cache_fd_got_sigbus(f->cache_fd))
262 return -EIO;
263
264 switch (f->header->state) {
265 case STATE_ONLINE:
266 return 0;
267
268 case STATE_OFFLINE:
269 f->header->state = STATE_ONLINE;
270 (void) fsync(f->fd);
271 return 0;
272
273 default:
274 return -EINVAL;
275 }
276 }
277
278 JournalFile* journal_file_close(JournalFile *f) {
279 if (!f)
280 return NULL;
281
282 assert(f->newest_boot_id_prioq_idx == PRIOQ_IDX_NULL);
283
284 if (f->cache_fd)
285 mmap_cache_fd_free(f->cache_fd);
286
287 if (f->close_fd)
288 safe_close(f->fd);
289 free(f->path);
290
291 ordered_hashmap_free_free(f->chain_cache);
292
293 #if HAVE_COMPRESSION
294 free(f->compress_buffer);
295 #endif
296
297 #if HAVE_GCRYPT
298 if (f->fss_file) {
299 size_t sz = PAGE_ALIGN(f->fss_file_size);
300 assert(sz < SIZE_MAX);
301 munmap(f->fss_file, sz);
302 } else
303 free(f->fsprg_state);
304
305 free(f->fsprg_seed);
306
307 if (f->hmac)
308 gcry_md_close(f->hmac);
309 #endif
310
311 return mfree(f);
312 }
313
314 static bool keyed_hash_requested(void) {
315 static thread_local int cached = -1;
316 int r;
317
318 if (cached < 0) {
319 r = getenv_bool("SYSTEMD_JOURNAL_KEYED_HASH");
320 if (r < 0) {
321 if (r != -ENXIO)
322 log_debug_errno(r, "Failed to parse $SYSTEMD_JOURNAL_KEYED_HASH environment variable, ignoring: %m");
323 cached = true;
324 } else
325 cached = r;
326 }
327
328 return cached;
329 }
330
331 static bool compact_mode_requested(void) {
332 static thread_local int cached = -1;
333 int r;
334
335 if (cached < 0) {
336 r = getenv_bool("SYSTEMD_JOURNAL_COMPACT");
337 if (r < 0) {
338 if (r != -ENXIO)
339 log_debug_errno(r, "Failed to parse $SYSTEMD_JOURNAL_COMPACT environment variable, ignoring: %m");
340 cached = true;
341 } else
342 cached = r;
343 }
344
345 return cached;
346 }
347
348 #if HAVE_COMPRESSION
349 static Compression getenv_compression(void) {
350 Compression c;
351 const char *e;
352 int r;
353
354 e = getenv("SYSTEMD_JOURNAL_COMPRESS");
355 if (!e)
356 return DEFAULT_COMPRESSION;
357
358 r = parse_boolean(e);
359 if (r >= 0)
360 return r ? DEFAULT_COMPRESSION : COMPRESSION_NONE;
361
362 c = compression_from_string(e);
363 if (c < 0) {
364 log_debug_errno(c, "Failed to parse SYSTEMD_JOURNAL_COMPRESS value, ignoring: %s", e);
365 return DEFAULT_COMPRESSION;
366 }
367
368 if (!compression_supported(c)) {
369 log_debug("Unsupported compression algorithm specified, ignoring: %s", e);
370 return DEFAULT_COMPRESSION;
371 }
372
373 return c;
374 }
375 #endif
376
377 static Compression compression_requested(void) {
378 #if HAVE_COMPRESSION
379 static thread_local Compression cached = _COMPRESSION_INVALID;
380
381 if (cached < 0)
382 cached = getenv_compression();
383
384 return cached;
385 #else
386 return COMPRESSION_NONE;
387 #endif
388 }
389
390 static int journal_file_init_header(
391 JournalFile *f,
392 JournalFileFlags file_flags,
393 JournalFile *template) {
394
395 bool seal = false;
396 ssize_t k;
397 int r;
398
399 assert(f);
400
401 #if HAVE_GCRYPT
402 /* Try to load the FSPRG state, and if we can't, then just don't do sealing */
403 seal = FLAGS_SET(file_flags, JOURNAL_SEAL) && journal_file_fss_load(f) >= 0;
404 #endif
405
406 Header h = {
407 .header_size = htole64(ALIGN64(sizeof(h))),
408 .incompatible_flags = htole32(
409 FLAGS_SET(file_flags, JOURNAL_COMPRESS) * COMPRESSION_TO_HEADER_INCOMPATIBLE_FLAG(compression_requested()) |
410 keyed_hash_requested() * HEADER_INCOMPATIBLE_KEYED_HASH |
411 compact_mode_requested() * HEADER_INCOMPATIBLE_COMPACT),
412 .compatible_flags = htole32(
413 (seal * (HEADER_COMPATIBLE_SEALED | HEADER_COMPATIBLE_SEALED_CONTINUOUS) ) |
414 HEADER_COMPATIBLE_TAIL_ENTRY_BOOT_ID),
415 };
416
417 assert_cc(sizeof(h.signature) == sizeof(HEADER_SIGNATURE));
418 memcpy(h.signature, HEADER_SIGNATURE, sizeof(HEADER_SIGNATURE));
419
420 r = sd_id128_randomize(&h.file_id);
421 if (r < 0)
422 return r;
423
424 r = sd_id128_get_machine(&h.machine_id);
425 if (r < 0 && !ERRNO_IS_MACHINE_ID_UNSET(r))
426 return r; /* If we have no valid machine ID (test environment?), let's simply leave the
427 * machine ID field all zeroes. */
428
429 if (template) {
430 h.seqnum_id = template->header->seqnum_id;
431 h.tail_entry_seqnum = template->header->tail_entry_seqnum;
432 } else
433 h.seqnum_id = h.file_id;
434
435 k = pwrite(f->fd, &h, sizeof(h), 0);
436 if (k < 0)
437 return -errno;
438 if (k != sizeof(h))
439 return -EIO;
440
441 return 0;
442 }
443
444 static int journal_file_refresh_header(JournalFile *f) {
445 int r;
446
447 assert(f);
448 assert(f->header);
449
450 /* We used to update the header's boot ID field here, but we don't do that anymore, as per
451 * HEADER_COMPATIBLE_TAIL_ENTRY_BOOT_ID */
452
453 r = journal_file_set_online(f);
454
455 /* Sync the online state to disk; likely just created a new file, also sync the directory this file
456 * is located in. */
457 (void) fsync_full(f->fd);
458
459 return r;
460 }
461
462 static bool warn_wrong_flags(const JournalFile *f, bool compatible) {
463 const uint32_t any = compatible ? HEADER_COMPATIBLE_ANY : HEADER_INCOMPATIBLE_ANY,
464 supported = compatible ? HEADER_COMPATIBLE_SUPPORTED : HEADER_INCOMPATIBLE_SUPPORTED;
465 const char *type = compatible ? "compatible" : "incompatible";
466 uint32_t flags;
467
468 assert(f);
469 assert(f->header);
470
471 flags = le32toh(compatible ? f->header->compatible_flags : f->header->incompatible_flags);
472
473 if (flags & ~supported) {
474 if (flags & ~any)
475 log_debug("Journal file %s has unknown %s flags 0x%"PRIx32,
476 f->path, type, flags & ~any);
477 flags = (flags & any) & ~supported;
478 if (flags) {
479 const char* strv[6];
480 size_t n = 0;
481 _cleanup_free_ char *t = NULL;
482
483 if (compatible) {
484 if (flags & HEADER_COMPATIBLE_SEALED)
485 strv[n++] = "sealed";
486 if (flags & HEADER_COMPATIBLE_SEALED_CONTINUOUS)
487 strv[n++] = "sealed-continuous";
488 } else {
489 if (flags & HEADER_INCOMPATIBLE_COMPRESSED_XZ)
490 strv[n++] = "xz-compressed";
491 if (flags & HEADER_INCOMPATIBLE_COMPRESSED_LZ4)
492 strv[n++] = "lz4-compressed";
493 if (flags & HEADER_INCOMPATIBLE_COMPRESSED_ZSTD)
494 strv[n++] = "zstd-compressed";
495 if (flags & HEADER_INCOMPATIBLE_KEYED_HASH)
496 strv[n++] = "keyed-hash";
497 if (flags & HEADER_INCOMPATIBLE_COMPACT)
498 strv[n++] = "compact";
499 }
500 strv[n] = NULL;
501 assert(n < ELEMENTSOF(strv));
502
503 t = strv_join((char**) strv, ", ");
504 log_debug("Journal file %s uses %s %s %s disabled at compilation time.",
505 f->path, type, n > 1 ? "flags" : "flag", strnull(t));
506 }
507 return true;
508 }
509
510 return false;
511 }
512
513 static bool offset_is_valid(uint64_t offset, uint64_t header_size, uint64_t tail_object_offset) {
514 if (offset == 0)
515 return true;
516 if (!VALID64(offset))
517 return false;
518 if (offset < header_size)
519 return false;
520 if (offset > tail_object_offset)
521 return false;
522 return true;
523 }
524
525 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) {
526 if ((offset == 0) != (size == 0))
527 return false;
528 if (offset == 0)
529 return true;
530 if (offset <= offsetof(Object, hash_table.items))
531 return false;
532 offset -= offsetof(Object, hash_table.items);
533 if (!offset_is_valid(offset, header_size, tail_object_offset))
534 return false;
535 assert(offset <= header_size + arena_size);
536 if (size > header_size + arena_size - offset)
537 return false;
538 return true;
539 }
540
541 static int journal_file_verify_header(JournalFile *f) {
542 uint64_t arena_size, header_size;
543
544 assert(f);
545 assert(f->header);
546
547 if (memcmp(f->header->signature, HEADER_SIGNATURE, 8))
548 return -EBADMSG;
549
550 /* In both read and write mode we refuse to open files with incompatible
551 * flags we don't know. */
552 if (warn_wrong_flags(f, false))
553 return -EPROTONOSUPPORT;
554
555 /* When open for writing we refuse to open files with compatible flags, too. */
556 if (journal_file_writable(f) && warn_wrong_flags(f, true))
557 return -EPROTONOSUPPORT;
558
559 if (f->header->state >= _STATE_MAX)
560 return -EBADMSG;
561
562 header_size = le64toh(READ_NOW(f->header->header_size));
563
564 /* The first addition was n_data, so check that we are at least this large */
565 if (header_size < HEADER_SIZE_MIN)
566 return -EBADMSG;
567
568 /* When open for writing we refuse to open files with a mismatch of the header size, i.e. writing to
569 * files implementing older or new header structures. */
570 if (journal_file_writable(f) && header_size != sizeof(Header))
571 return -EPROTONOSUPPORT;
572
573 /* Don't write to journal files without the new boot ID update behavior guarantee. */
574 if (journal_file_writable(f) && !JOURNAL_HEADER_TAIL_ENTRY_BOOT_ID(f->header))
575 return -EPROTONOSUPPORT;
576
577 if (JOURNAL_HEADER_SEALED(f->header) && !JOURNAL_HEADER_CONTAINS(f->header, n_entry_arrays))
578 return -EBADMSG;
579
580 arena_size = le64toh(READ_NOW(f->header->arena_size));
581
582 if (UINT64_MAX - header_size < arena_size || header_size + arena_size > (uint64_t) f->last_stat.st_size)
583 return -ENODATA;
584
585 uint64_t tail_object_offset = le64toh(f->header->tail_object_offset);
586 if (!offset_is_valid(tail_object_offset, header_size, UINT64_MAX))
587 return -ENODATA;
588 if (header_size + arena_size < tail_object_offset)
589 return -ENODATA;
590 if (header_size + arena_size - tail_object_offset < sizeof(ObjectHeader))
591 return -ENODATA;
592
593 if (!hash_table_is_valid(le64toh(f->header->data_hash_table_offset),
594 le64toh(f->header->data_hash_table_size),
595 header_size, arena_size, tail_object_offset))
596 return -ENODATA;
597
598 if (!hash_table_is_valid(le64toh(f->header->field_hash_table_offset),
599 le64toh(f->header->field_hash_table_size),
600 header_size, arena_size, tail_object_offset))
601 return -ENODATA;
602
603 uint64_t entry_array_offset = le64toh(f->header->entry_array_offset);
604 if (!offset_is_valid(entry_array_offset, header_size, tail_object_offset))
605 return -ENODATA;
606
607 if (JOURNAL_HEADER_CONTAINS(f->header, tail_entry_array_offset)) {
608 uint32_t offset = le32toh(f->header->tail_entry_array_offset);
609 uint32_t n = le32toh(f->header->tail_entry_array_n_entries);
610
611 if (!offset_is_valid(offset, header_size, tail_object_offset))
612 return -ENODATA;
613 if (entry_array_offset > offset)
614 return -ENODATA;
615 if (entry_array_offset == 0 && offset != 0)
616 return -ENODATA;
617 if ((offset == 0) != (n == 0))
618 return -ENODATA;
619 assert(offset <= header_size + arena_size);
620 if ((uint64_t) n * journal_file_entry_array_item_size(f) > header_size + arena_size - offset)
621 return -ENODATA;
622 }
623
624 if (JOURNAL_HEADER_CONTAINS(f->header, tail_entry_offset)) {
625 uint64_t offset = le64toh(f->header->tail_entry_offset);
626
627 if (!offset_is_valid(offset, header_size, tail_object_offset))
628 return -ENODATA;
629
630 if (offset > 0) {
631 /* When there is an entry object, then these fields must be filled. */
632 if (sd_id128_is_null(f->header->tail_entry_boot_id))
633 return -ENODATA;
634 if (!VALID_REALTIME(le64toh(f->header->head_entry_realtime)))
635 return -ENODATA;
636 if (!VALID_REALTIME(le64toh(f->header->tail_entry_realtime)))
637 return -ENODATA;
638 if (!VALID_MONOTONIC(le64toh(f->header->tail_entry_realtime)))
639 return -ENODATA;
640 } else {
641 /* Otherwise, the fields must be zero. */
642 if (JOURNAL_HEADER_TAIL_ENTRY_BOOT_ID(f->header) &&
643 !sd_id128_is_null(f->header->tail_entry_boot_id))
644 return -ENODATA;
645 if (f->header->head_entry_realtime != 0)
646 return -ENODATA;
647 if (f->header->tail_entry_realtime != 0)
648 return -ENODATA;
649 if (f->header->tail_entry_realtime != 0)
650 return -ENODATA;
651 }
652 }
653
654 /* Verify number of objects */
655 uint64_t n_objects = le64toh(f->header->n_objects);
656 if (n_objects > arena_size / sizeof(ObjectHeader))
657 return -ENODATA;
658
659 uint64_t n_entries = le64toh(f->header->n_entries);
660 if (n_entries > n_objects)
661 return -ENODATA;
662
663 if (JOURNAL_HEADER_CONTAINS(f->header, n_data) &&
664 le64toh(f->header->n_data) > n_objects)
665 return -ENODATA;
666
667 if (JOURNAL_HEADER_CONTAINS(f->header, n_fields) &&
668 le64toh(f->header->n_fields) > n_objects)
669 return -ENODATA;
670
671 if (JOURNAL_HEADER_CONTAINS(f->header, n_tags) &&
672 le64toh(f->header->n_tags) > n_objects)
673 return -ENODATA;
674
675 if (JOURNAL_HEADER_CONTAINS(f->header, n_entry_arrays) &&
676 le64toh(f->header->n_entry_arrays) > n_objects)
677 return -ENODATA;
678
679 if (JOURNAL_HEADER_CONTAINS(f->header, tail_entry_array_n_entries) &&
680 le32toh(f->header->tail_entry_array_n_entries) > n_entries)
681 return -ENODATA;
682
683 if (journal_file_writable(f)) {
684 sd_id128_t machine_id;
685 uint8_t state;
686 int r;
687
688 r = sd_id128_get_machine(&machine_id);
689 if (ERRNO_IS_NEG_MACHINE_ID_UNSET(r)) /* Gracefully handle the machine ID not being initialized yet */
690 machine_id = SD_ID128_NULL;
691 else if (r < 0)
692 return r;
693
694 if (!sd_id128_equal(machine_id, f->header->machine_id))
695 return log_debug_errno(SYNTHETIC_ERRNO(EHOSTDOWN),
696 "Trying to open journal file from different host for writing, refusing.");
697
698 state = f->header->state;
699
700 if (state == STATE_ARCHIVED)
701 return -ESHUTDOWN; /* Already archived */
702 if (state == STATE_ONLINE)
703 return log_debug_errno(SYNTHETIC_ERRNO(EBUSY),
704 "Journal file %s is already online. Assuming unclean closing.",
705 f->path);
706 if (state != STATE_OFFLINE)
707 return log_debug_errno(SYNTHETIC_ERRNO(EBUSY),
708 "Journal file %s has unknown state %i.",
709 f->path, state);
710
711 if (f->header->field_hash_table_size == 0 || f->header->data_hash_table_size == 0)
712 return -EBADMSG;
713 }
714
715 return 0;
716 }
717
718 int journal_file_fstat(JournalFile *f) {
719 int r;
720
721 assert(f);
722 assert(f->fd >= 0);
723
724 if (fstat(f->fd, &f->last_stat) < 0)
725 return -errno;
726
727 f->last_stat_usec = now(CLOCK_MONOTONIC);
728
729 /* Refuse dealing with files that aren't regular */
730 r = stat_verify_regular(&f->last_stat);
731 if (r < 0)
732 return r;
733
734 /* Refuse appending to files that are already deleted */
735 if (f->last_stat.st_nlink <= 0)
736 return -EIDRM;
737
738 return 0;
739 }
740
741 static int journal_file_allocate(JournalFile *f, uint64_t offset, uint64_t size) {
742 uint64_t old_size, new_size, old_header_size, old_arena_size;
743 int r;
744
745 assert(f);
746 assert(f->header);
747
748 /* We assume that this file is not sparse, and we know that for sure, since we always call
749 * posix_fallocate() ourselves */
750
751 if (size > PAGE_ALIGN_DOWN_U64(UINT64_MAX) - offset)
752 return -EINVAL;
753
754 if (mmap_cache_fd_got_sigbus(f->cache_fd))
755 return -EIO;
756
757 old_header_size = le64toh(READ_NOW(f->header->header_size));
758 old_arena_size = le64toh(READ_NOW(f->header->arena_size));
759 if (old_arena_size > PAGE_ALIGN_DOWN_U64(UINT64_MAX) - old_header_size)
760 return -EBADMSG;
761
762 old_size = old_header_size + old_arena_size;
763
764 new_size = MAX(PAGE_ALIGN_U64(offset + size), old_header_size);
765
766 if (new_size <= old_size) {
767
768 /* We already pre-allocated enough space, but before
769 * we write to it, let's check with fstat() if the
770 * file got deleted, in order make sure we don't throw
771 * away the data immediately. Don't check fstat() for
772 * all writes though, but only once ever 10s. */
773
774 if (f->last_stat_usec + LAST_STAT_REFRESH_USEC > now(CLOCK_MONOTONIC))
775 return 0;
776
777 return journal_file_fstat(f);
778 }
779
780 /* Allocate more space. */
781
782 if (f->metrics.max_size > 0 && new_size > f->metrics.max_size)
783 return -E2BIG;
784
785 /* Refuse to go over 4G in compact mode so offsets can be stored in 32-bit. */
786 if (JOURNAL_HEADER_COMPACT(f->header) && new_size > UINT32_MAX)
787 return -E2BIG;
788
789 if (new_size > f->metrics.min_size && f->metrics.keep_free > 0) {
790 struct statvfs svfs;
791
792 if (fstatvfs(f->fd, &svfs) >= 0) {
793 uint64_t available;
794
795 available = LESS_BY(u64_multiply_safe(svfs.f_bfree, svfs.f_bsize), f->metrics.keep_free);
796
797 if (new_size - old_size > available)
798 return -E2BIG;
799 }
800 }
801
802 /* Increase by larger blocks at once */
803 new_size = ROUND_UP(new_size, FILE_SIZE_INCREASE);
804 if (f->metrics.max_size > 0 && new_size > f->metrics.max_size)
805 new_size = f->metrics.max_size;
806
807 /* Note that the glibc fallocate() fallback is very
808 inefficient, hence we try to minimize the allocation area
809 as we can. */
810 r = posix_fallocate_loop(f->fd, old_size, new_size - old_size);
811 if (r < 0)
812 return r;
813
814 f->header->arena_size = htole64(new_size - old_header_size);
815
816 return journal_file_fstat(f);
817 }
818
819 static int journal_file_move_to(
820 JournalFile *f,
821 ObjectType type,
822 bool keep_always,
823 uint64_t offset,
824 uint64_t size,
825 void **ret) {
826
827 int r;
828
829 assert(f);
830 assert(ret);
831
832 /* This function may clear, overwrite, or alter previously cached entries with the same type. After
833 * this function has been called, all previously read objects with the same type may be invalidated,
834 * hence must be re-read before use. */
835
836 if (size <= 0)
837 return -EINVAL;
838
839 if (size > UINT64_MAX - offset)
840 return -EBADMSG;
841
842 /* Avoid SIGBUS on invalid accesses */
843 if (offset + size > (uint64_t) f->last_stat.st_size) {
844 /* Hmm, out of range? Let's refresh the fstat() data
845 * first, before we trust that check. */
846
847 r = journal_file_fstat(f);
848 if (r < 0)
849 return r;
850
851 if (offset + size > (uint64_t) f->last_stat.st_size)
852 return -EADDRNOTAVAIL;
853 }
854
855 return mmap_cache_fd_get(f->cache_fd, type_to_category(type), keep_always, offset, size, &f->last_stat, ret);
856 }
857
858 static uint64_t minimum_header_size(JournalFile *f, Object *o) {
859
860 static const uint64_t table[] = {
861 [OBJECT_DATA] = sizeof(DataObject),
862 [OBJECT_FIELD] = sizeof(FieldObject),
863 [OBJECT_ENTRY] = sizeof(EntryObject),
864 [OBJECT_DATA_HASH_TABLE] = sizeof(HashTableObject),
865 [OBJECT_FIELD_HASH_TABLE] = sizeof(HashTableObject),
866 [OBJECT_ENTRY_ARRAY] = sizeof(EntryArrayObject),
867 [OBJECT_TAG] = sizeof(TagObject),
868 };
869
870 assert(f);
871 assert(o);
872
873 if (o->object.type == OBJECT_DATA)
874 return journal_file_data_payload_offset(f);
875
876 if (o->object.type >= ELEMENTSOF(table) || table[o->object.type] <= 0)
877 return sizeof(ObjectHeader);
878
879 return table[o->object.type];
880 }
881
882 static int check_object_header(JournalFile *f, Object *o, ObjectType type, uint64_t offset) {
883 uint64_t s;
884
885 assert(f);
886 assert(o);
887
888 s = le64toh(READ_NOW(o->object.size));
889 if (s == 0)
890 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG),
891 "Attempt to move to uninitialized object: %" PRIu64,
892 offset);
893
894 if (s < sizeof(ObjectHeader))
895 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG),
896 "Attempt to move to overly short object with size %"PRIu64": %" PRIu64,
897 s, offset);
898
899 if (o->object.type <= OBJECT_UNUSED || o->object.type >= _OBJECT_TYPE_MAX)
900 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG),
901 "Attempt to move to object with invalid type (%u): %" PRIu64,
902 o->object.type, offset);
903
904 if (type > OBJECT_UNUSED && o->object.type != type)
905 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG),
906 "Found %s object while expecting %s object: %" PRIu64,
907 journal_object_type_to_string(o->object.type),
908 journal_object_type_to_string(type),
909 offset);
910
911 if (s < minimum_header_size(f, o))
912 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG),
913 "Size of %s object (%"PRIu64") is smaller than the minimum object size (%"PRIu64"): %" PRIu64,
914 journal_object_type_to_string(o->object.type),
915 s,
916 minimum_header_size(f, o),
917 offset);
918
919 return 0;
920 }
921
922 /* Lightweight object checks. We want this to be fast, so that we won't
923 * slowdown every journal_file_move_to_object() call too much. */
924 static int check_object(JournalFile *f, Object *o, uint64_t offset) {
925 assert(f);
926 assert(o);
927
928 switch (o->object.type) {
929
930 case OBJECT_DATA:
931 if ((le64toh(o->data.entry_offset) == 0) ^ (le64toh(o->data.n_entries) == 0))
932 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG),
933 "Bad data n_entries: %" PRIu64 ": %" PRIu64,
934 le64toh(o->data.n_entries),
935 offset);
936
937 if (le64toh(o->object.size) <= journal_file_data_payload_offset(f))
938 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG),
939 "Bad data size (<= %zu): %" PRIu64 ": %" PRIu64,
940 journal_file_data_payload_offset(f),
941 le64toh(o->object.size),
942 offset);
943
944 if (!VALID64(le64toh(o->data.next_hash_offset)) ||
945 !VALID64(le64toh(o->data.next_field_offset)) ||
946 !VALID64(le64toh(o->data.entry_offset)) ||
947 !VALID64(le64toh(o->data.entry_array_offset)))
948 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG),
949 "Invalid offset, next_hash_offset=" OFSfmt ", next_field_offset=" OFSfmt ", entry_offset=" OFSfmt ", entry_array_offset=" OFSfmt ": %" PRIu64,
950 le64toh(o->data.next_hash_offset),
951 le64toh(o->data.next_field_offset),
952 le64toh(o->data.entry_offset),
953 le64toh(o->data.entry_array_offset),
954 offset);
955
956 break;
957
958 case OBJECT_FIELD:
959 if (le64toh(o->object.size) <= offsetof(Object, field.payload))
960 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG),
961 "Bad field size (<= %zu): %" PRIu64 ": %" PRIu64,
962 offsetof(Object, field.payload),
963 le64toh(o->object.size),
964 offset);
965
966 if (!VALID64(le64toh(o->field.next_hash_offset)) ||
967 !VALID64(le64toh(o->field.head_data_offset)))
968 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG),
969 "Invalid offset, next_hash_offset=" OFSfmt ", head_data_offset=" OFSfmt ": %" PRIu64,
970 le64toh(o->field.next_hash_offset),
971 le64toh(o->field.head_data_offset),
972 offset);
973 break;
974
975 case OBJECT_ENTRY: {
976 uint64_t sz;
977
978 sz = le64toh(READ_NOW(o->object.size));
979 if (sz < offsetof(Object, entry.items) ||
980 (sz - offsetof(Object, entry.items)) % journal_file_entry_item_size(f) != 0)
981 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG),
982 "Bad entry size (<= %zu): %" PRIu64 ": %" PRIu64,
983 offsetof(Object, entry.items),
984 sz,
985 offset);
986
987 if ((sz - offsetof(Object, entry.items)) / journal_file_entry_item_size(f) <= 0)
988 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG),
989 "Invalid number items in entry: %" PRIu64 ": %" PRIu64,
990 (sz - offsetof(Object, entry.items)) / journal_file_entry_item_size(f),
991 offset);
992
993 if (le64toh(o->entry.seqnum) <= 0)
994 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG),
995 "Invalid entry seqnum: %" PRIx64 ": %" PRIu64,
996 le64toh(o->entry.seqnum),
997 offset);
998
999 if (!VALID_REALTIME(le64toh(o->entry.realtime)))
1000 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG),
1001 "Invalid entry realtime timestamp: %" PRIu64 ": %" PRIu64,
1002 le64toh(o->entry.realtime),
1003 offset);
1004
1005 if (!VALID_MONOTONIC(le64toh(o->entry.monotonic)))
1006 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG),
1007 "Invalid entry monotonic timestamp: %" PRIu64 ": %" PRIu64,
1008 le64toh(o->entry.monotonic),
1009 offset);
1010
1011 if (sd_id128_is_null(o->entry.boot_id))
1012 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG),
1013 "Invalid object entry with an empty boot ID: %" PRIu64,
1014 offset);
1015
1016 break;
1017 }
1018
1019 case OBJECT_DATA_HASH_TABLE:
1020 case OBJECT_FIELD_HASH_TABLE: {
1021 uint64_t sz;
1022
1023 sz = le64toh(READ_NOW(o->object.size));
1024 if (sz < offsetof(Object, hash_table.items) ||
1025 (sz - offsetof(Object, hash_table.items)) % sizeof(HashItem) != 0 ||
1026 (sz - offsetof(Object, hash_table.items)) / sizeof(HashItem) <= 0)
1027 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG),
1028 "Invalid %s hash table size: %" PRIu64 ": %" PRIu64,
1029 journal_object_type_to_string(o->object.type),
1030 sz,
1031 offset);
1032
1033 break;
1034 }
1035
1036 case OBJECT_ENTRY_ARRAY: {
1037 uint64_t sz, next;
1038
1039 sz = le64toh(READ_NOW(o->object.size));
1040 if (sz < offsetof(Object, entry_array.items) ||
1041 (sz - offsetof(Object, entry_array.items)) % journal_file_entry_array_item_size(f) != 0 ||
1042 (sz - offsetof(Object, entry_array.items)) / journal_file_entry_array_item_size(f) <= 0)
1043 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG),
1044 "Invalid object entry array size: %" PRIu64 ": %" PRIu64,
1045 sz,
1046 offset);
1047 /* Here, we request that the offset of each entry array object is in strictly increasing order. */
1048 next = le64toh(o->entry_array.next_entry_array_offset);
1049 if (!VALID64(next) || (next > 0 && next <= offset))
1050 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG),
1051 "Invalid object entry array next_entry_array_offset: %" PRIu64 ": %" PRIu64,
1052 next,
1053 offset);
1054
1055 break;
1056 }
1057
1058 case OBJECT_TAG:
1059 if (le64toh(o->object.size) != sizeof(TagObject))
1060 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG),
1061 "Invalid object tag size: %" PRIu64 ": %" PRIu64,
1062 le64toh(o->object.size),
1063 offset);
1064
1065 if (!VALID_EPOCH(le64toh(o->tag.epoch)))
1066 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG),
1067 "Invalid object tag epoch: %" PRIu64 ": %" PRIu64,
1068 le64toh(o->tag.epoch), offset);
1069
1070 break;
1071 }
1072
1073 return 0;
1074 }
1075
1076 int journal_file_move_to_object(JournalFile *f, ObjectType type, uint64_t offset, Object **ret) {
1077 int r;
1078 Object *o;
1079
1080 assert(f);
1081
1082 /* Even if this function fails, it may clear, overwrite, or alter previously cached entries with the
1083 * same type. After this function has been called, all previously read objects with the same type may
1084 * be invalidated, hence must be re-read before use. */
1085
1086 /* Objects may only be located at multiple of 64 bit */
1087 if (!VALID64(offset))
1088 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG),
1089 "Attempt to move to %s object at non-64-bit boundary: %" PRIu64,
1090 journal_object_type_to_string(type),
1091 offset);
1092
1093 /* Object may not be located in the file header */
1094 if (offset < le64toh(f->header->header_size))
1095 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG),
1096 "Attempt to move to %s object located in file header: %" PRIu64,
1097 journal_object_type_to_string(type),
1098 offset);
1099
1100 r = journal_file_move_to(f, type, false, offset, sizeof(ObjectHeader), (void**) &o);
1101 if (r < 0)
1102 return r;
1103
1104 r = check_object_header(f, o, type, offset);
1105 if (r < 0)
1106 return r;
1107
1108 r = journal_file_move_to(f, type, false, offset, le64toh(READ_NOW(o->object.size)), (void**) &o);
1109 if (r < 0)
1110 return r;
1111
1112 r = check_object_header(f, o, type, offset);
1113 if (r < 0)
1114 return r;
1115
1116 r = check_object(f, o, offset);
1117 if (r < 0)
1118 return r;
1119
1120 if (ret)
1121 *ret = o;
1122
1123 return 0;
1124 }
1125
1126 int journal_file_pin_object(JournalFile *f, Object *o) {
1127 assert(f);
1128 assert(o);
1129
1130 /* This attaches the mmap window that provides the object to the 'pinning' category. So, reading
1131 * another object with the same type will not invalidate the object, until this function is called
1132 * for another object. */
1133 return mmap_cache_fd_pin(f->cache_fd, type_to_category(o->object.type), o, le64toh(o->object.size));
1134 }
1135
1136 int journal_file_read_object_header(JournalFile *f, ObjectType type, uint64_t offset, Object *ret) {
1137 ssize_t n;
1138 Object o;
1139 int r;
1140
1141 assert(f);
1142
1143 /* Objects may only be located at multiple of 64 bit */
1144 if (!VALID64(offset))
1145 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG),
1146 "Attempt to read %s object at non-64-bit boundary: %" PRIu64,
1147 journal_object_type_to_string(type), offset);
1148
1149 /* Object may not be located in the file header */
1150 if (offset < le64toh(f->header->header_size))
1151 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG),
1152 "Attempt to read %s object located in file header: %" PRIu64,
1153 journal_object_type_to_string(type), offset);
1154
1155 /* This will likely read too much data but it avoids having to call pread() twice. */
1156 n = pread(f->fd, &o, sizeof(o), offset);
1157 if (n < 0)
1158 return log_debug_errno(errno, "Failed to read journal %s object at offset: %" PRIu64,
1159 journal_object_type_to_string(type), offset);
1160
1161 if ((size_t) n < sizeof(o.object))
1162 return log_debug_errno(SYNTHETIC_ERRNO(EIO),
1163 "Failed to read short %s object at offset: %" PRIu64,
1164 journal_object_type_to_string(type), offset);
1165
1166 r = check_object_header(f, &o, type, offset);
1167 if (r < 0)
1168 return r;
1169
1170 if ((size_t) n < minimum_header_size(f, &o))
1171 return log_debug_errno(SYNTHETIC_ERRNO(EIO),
1172 "Short read while reading %s object: %" PRIu64,
1173 journal_object_type_to_string(type), offset);
1174
1175 r = check_object(f, &o, offset);
1176 if (r < 0)
1177 return r;
1178
1179 if (ret)
1180 *ret = o;
1181
1182 return 0;
1183 }
1184
1185 static uint64_t inc_seqnum(uint64_t seqnum) {
1186 if (seqnum < UINT64_MAX-1)
1187 return seqnum + 1;
1188
1189 return 1; /* skip over UINT64_MAX and 0 when we run out of seqnums and start again */
1190 }
1191
1192 static uint64_t journal_file_entry_seqnum(
1193 JournalFile *f,
1194 uint64_t *seqnum) {
1195
1196 uint64_t next_seqnum;
1197
1198 assert(f);
1199 assert(f->header);
1200
1201 /* Picks a new sequence number for the entry we are about to add and returns it. */
1202
1203 next_seqnum = inc_seqnum(le64toh(f->header->tail_entry_seqnum));
1204
1205 /* If an external seqnum counter was passed, we update both the local and the external one, and set
1206 * it to the maximum of both */
1207 if (seqnum)
1208 *seqnum = next_seqnum = MAX(inc_seqnum(*seqnum), next_seqnum);
1209
1210 f->header->tail_entry_seqnum = htole64(next_seqnum);
1211
1212 if (f->header->head_entry_seqnum == 0)
1213 f->header->head_entry_seqnum = htole64(next_seqnum);
1214
1215 return next_seqnum;
1216 }
1217
1218 int journal_file_append_object(
1219 JournalFile *f,
1220 ObjectType type,
1221 uint64_t size,
1222 Object **ret_object,
1223 uint64_t *ret_offset) {
1224
1225 int r;
1226 uint64_t p;
1227 Object *o;
1228
1229 assert(f);
1230 assert(f->header);
1231 assert(type > OBJECT_UNUSED && type < _OBJECT_TYPE_MAX);
1232 assert(size >= sizeof(ObjectHeader));
1233
1234 r = journal_file_set_online(f);
1235 if (r < 0)
1236 return r;
1237
1238 r = journal_file_tail_end_by_mmap(f, &p);
1239 if (r < 0)
1240 return r;
1241
1242 r = journal_file_allocate(f, p, size);
1243 if (r < 0)
1244 return r;
1245
1246 r = journal_file_move_to(f, type, false, p, size, (void**) &o);
1247 if (r < 0)
1248 return r;
1249
1250 o->object = (ObjectHeader) {
1251 .type = type,
1252 .size = htole64(size),
1253 };
1254
1255 f->header->tail_object_offset = htole64(p);
1256 f->header->n_objects = htole64(le64toh(f->header->n_objects) + 1);
1257
1258 if (ret_object)
1259 *ret_object = o;
1260
1261 if (ret_offset)
1262 *ret_offset = p;
1263
1264 return 0;
1265 }
1266
1267 static int journal_file_setup_data_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 estimate that we need 1 hash table entry per 768 bytes
1276 of journal file and we want to make sure we never get
1277 beyond 75% fill level. Calculate the hash table size for
1278 the maximum file size based on these metrics. */
1279
1280 s = (f->metrics.max_size * 4 / 768 / 3) * sizeof(HashItem);
1281 if (s < DEFAULT_DATA_HASH_TABLE_SIZE)
1282 s = DEFAULT_DATA_HASH_TABLE_SIZE;
1283
1284 log_debug("Reserving %"PRIu64" entries in data hash table.", s / sizeof(HashItem));
1285
1286 r = journal_file_append_object(f,
1287 OBJECT_DATA_HASH_TABLE,
1288 offsetof(Object, hash_table.items) + s,
1289 &o, &p);
1290 if (r < 0)
1291 return r;
1292
1293 memzero(o->hash_table.items, s);
1294
1295 f->header->data_hash_table_offset = htole64(p + offsetof(Object, hash_table.items));
1296 f->header->data_hash_table_size = htole64(s);
1297
1298 return 0;
1299 }
1300
1301 static int journal_file_setup_field_hash_table(JournalFile *f) {
1302 uint64_t s, p;
1303 Object *o;
1304 int r;
1305
1306 assert(f);
1307 assert(f->header);
1308
1309 /* We use a fixed size hash table for the fields as this
1310 * number should grow very slowly only */
1311
1312 s = DEFAULT_FIELD_HASH_TABLE_SIZE;
1313 log_debug("Reserving %"PRIu64" entries in field hash table.", s / sizeof(HashItem));
1314
1315 r = journal_file_append_object(f,
1316 OBJECT_FIELD_HASH_TABLE,
1317 offsetof(Object, hash_table.items) + s,
1318 &o, &p);
1319 if (r < 0)
1320 return r;
1321
1322 memzero(o->hash_table.items, s);
1323
1324 f->header->field_hash_table_offset = htole64(p + offsetof(Object, hash_table.items));
1325 f->header->field_hash_table_size = htole64(s);
1326
1327 return 0;
1328 }
1329
1330 int journal_file_map_data_hash_table(JournalFile *f) {
1331 uint64_t s, p;
1332 void *t;
1333 int r;
1334
1335 assert(f);
1336 assert(f->header);
1337
1338 if (f->data_hash_table)
1339 return 0;
1340
1341 p = le64toh(f->header->data_hash_table_offset);
1342 s = le64toh(f->header->data_hash_table_size);
1343
1344 r = journal_file_move_to(f,
1345 OBJECT_DATA_HASH_TABLE,
1346 true,
1347 p, s,
1348 &t);
1349 if (r < 0)
1350 return r;
1351
1352 f->data_hash_table = t;
1353 return 0;
1354 }
1355
1356 int journal_file_map_field_hash_table(JournalFile *f) {
1357 uint64_t s, p;
1358 void *t;
1359 int r;
1360
1361 assert(f);
1362 assert(f->header);
1363
1364 if (f->field_hash_table)
1365 return 0;
1366
1367 p = le64toh(f->header->field_hash_table_offset);
1368 s = le64toh(f->header->field_hash_table_size);
1369
1370 r = journal_file_move_to(f,
1371 OBJECT_FIELD_HASH_TABLE,
1372 true,
1373 p, s,
1374 &t);
1375 if (r < 0)
1376 return r;
1377
1378 f->field_hash_table = t;
1379 return 0;
1380 }
1381
1382 static int journal_file_link_field(
1383 JournalFile *f,
1384 Object *o,
1385 uint64_t offset,
1386 uint64_t hash) {
1387
1388 uint64_t p, h, m;
1389 int r;
1390
1391 assert(f);
1392 assert(f->header);
1393 assert(f->field_hash_table);
1394 assert(o);
1395 assert(offset > 0);
1396
1397 if (o->object.type != OBJECT_FIELD)
1398 return -EINVAL;
1399
1400 m = le64toh(READ_NOW(f->header->field_hash_table_size)) / sizeof(HashItem);
1401 if (m <= 0)
1402 return -EBADMSG;
1403
1404 /* This might alter the window we are looking at */
1405 o->field.next_hash_offset = o->field.head_data_offset = 0;
1406
1407 h = hash % m;
1408 p = le64toh(f->field_hash_table[h].tail_hash_offset);
1409 if (p == 0)
1410 f->field_hash_table[h].head_hash_offset = htole64(offset);
1411 else {
1412 r = journal_file_move_to_object(f, OBJECT_FIELD, p, &o);
1413 if (r < 0)
1414 return r;
1415
1416 o->field.next_hash_offset = htole64(offset);
1417 }
1418
1419 f->field_hash_table[h].tail_hash_offset = htole64(offset);
1420
1421 if (JOURNAL_HEADER_CONTAINS(f->header, n_fields))
1422 f->header->n_fields = htole64(le64toh(f->header->n_fields) + 1);
1423
1424 return 0;
1425 }
1426
1427 static int journal_file_link_data(
1428 JournalFile *f,
1429 Object *o,
1430 uint64_t offset,
1431 uint64_t hash) {
1432
1433 uint64_t p, h, m;
1434 int r;
1435
1436 assert(f);
1437 assert(f->header);
1438 assert(f->data_hash_table);
1439 assert(o);
1440 assert(offset > 0);
1441
1442 if (o->object.type != OBJECT_DATA)
1443 return -EINVAL;
1444
1445 m = le64toh(READ_NOW(f->header->data_hash_table_size)) / sizeof(HashItem);
1446 if (m <= 0)
1447 return -EBADMSG;
1448
1449 /* This might alter the window we are looking at */
1450 o->data.next_hash_offset = o->data.next_field_offset = 0;
1451 o->data.entry_offset = o->data.entry_array_offset = 0;
1452 o->data.n_entries = 0;
1453
1454 h = hash % m;
1455 p = le64toh(f->data_hash_table[h].tail_hash_offset);
1456 if (p == 0)
1457 /* Only entry in the hash table is easy */
1458 f->data_hash_table[h].head_hash_offset = htole64(offset);
1459 else {
1460 /* Move back to the previous data object, to patch in
1461 * pointer */
1462
1463 r = journal_file_move_to_object(f, OBJECT_DATA, p, &o);
1464 if (r < 0)
1465 return r;
1466
1467 o->data.next_hash_offset = htole64(offset);
1468 }
1469
1470 f->data_hash_table[h].tail_hash_offset = htole64(offset);
1471
1472 if (JOURNAL_HEADER_CONTAINS(f->header, n_data))
1473 f->header->n_data = htole64(le64toh(f->header->n_data) + 1);
1474
1475 return 0;
1476 }
1477
1478 static int get_next_hash_offset(
1479 JournalFile *f,
1480 uint64_t *p,
1481 le64_t *next_hash_offset,
1482 uint64_t *depth,
1483 le64_t *header_max_depth) {
1484
1485 uint64_t nextp;
1486
1487 assert(f);
1488 assert(p);
1489 assert(next_hash_offset);
1490 assert(depth);
1491
1492 nextp = le64toh(READ_NOW(*next_hash_offset));
1493 if (nextp > 0) {
1494 if (nextp <= *p) /* Refuse going in loops */
1495 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG),
1496 "Detected hash item loop in %s, refusing.", f->path);
1497
1498 (*depth)++;
1499
1500 /* If the depth of this hash chain is larger than all others we have seen so far, record it */
1501 if (header_max_depth && journal_file_writable(f))
1502 *header_max_depth = htole64(MAX(*depth, le64toh(*header_max_depth)));
1503 }
1504
1505 *p = nextp;
1506 return 0;
1507 }
1508
1509 int journal_file_find_field_object_with_hash(
1510 JournalFile *f,
1511 const void *field,
1512 uint64_t size,
1513 uint64_t hash,
1514 Object **ret_object,
1515 uint64_t *ret_offset) {
1516
1517 uint64_t p, osize, h, m, depth = 0;
1518 int r;
1519
1520 assert(f);
1521 assert(f->header);
1522 assert(field);
1523 assert(size > 0);
1524
1525 /* If the field hash table is empty, we can't find anything */
1526 if (le64toh(f->header->field_hash_table_size) <= 0)
1527 return 0;
1528
1529 /* Map the field hash table, if it isn't mapped yet. */
1530 r = journal_file_map_field_hash_table(f);
1531 if (r < 0)
1532 return r;
1533
1534 osize = offsetof(Object, field.payload) + size;
1535
1536 m = le64toh(READ_NOW(f->header->field_hash_table_size)) / sizeof(HashItem);
1537 if (m <= 0)
1538 return -EBADMSG;
1539
1540 h = hash % m;
1541 p = le64toh(f->field_hash_table[h].head_hash_offset);
1542 while (p > 0) {
1543 Object *o;
1544
1545 r = journal_file_move_to_object(f, OBJECT_FIELD, p, &o);
1546 if (r < 0)
1547 return r;
1548
1549 if (le64toh(o->field.hash) == hash &&
1550 le64toh(o->object.size) == osize &&
1551 memcmp(o->field.payload, field, size) == 0) {
1552
1553 if (ret_object)
1554 *ret_object = o;
1555 if (ret_offset)
1556 *ret_offset = p;
1557
1558 return 1;
1559 }
1560
1561 r = get_next_hash_offset(
1562 f,
1563 &p,
1564 &o->field.next_hash_offset,
1565 &depth,
1566 JOURNAL_HEADER_CONTAINS(f->header, field_hash_chain_depth) ? &f->header->field_hash_chain_depth : NULL);
1567 if (r < 0)
1568 return r;
1569 }
1570
1571 return 0;
1572 }
1573
1574 uint64_t journal_file_hash_data(
1575 JournalFile *f,
1576 const void *data,
1577 size_t sz) {
1578
1579 assert(f);
1580 assert(f->header);
1581 assert(data || sz == 0);
1582
1583 /* We try to unify our codebase on siphash, hence new-styled journal files utilizing the keyed hash
1584 * function use siphash. Old journal files use the Jenkins hash. */
1585
1586 if (JOURNAL_HEADER_KEYED_HASH(f->header))
1587 return siphash24(data, sz, f->header->file_id.bytes);
1588
1589 return jenkins_hash64(data, sz);
1590 }
1591
1592 int journal_file_find_field_object(
1593 JournalFile *f,
1594 const void *field,
1595 uint64_t size,
1596 Object **ret_object,
1597 uint64_t *ret_offset) {
1598
1599 assert(f);
1600 assert(field);
1601 assert(size > 0);
1602
1603 return journal_file_find_field_object_with_hash(
1604 f,
1605 field, size,
1606 journal_file_hash_data(f, field, size),
1607 ret_object, ret_offset);
1608 }
1609
1610 int journal_file_find_data_object_with_hash(
1611 JournalFile *f,
1612 const void *data,
1613 uint64_t size,
1614 uint64_t hash,
1615 Object **ret_object,
1616 uint64_t *ret_offset) {
1617
1618 uint64_t p, h, m, depth = 0;
1619 int r;
1620
1621 assert(f);
1622 assert(f->header);
1623 assert(data || size == 0);
1624
1625 /* If there's no data hash table, then there's no entry. */
1626 if (le64toh(f->header->data_hash_table_size) <= 0)
1627 return 0;
1628
1629 /* Map the data hash table, if it isn't mapped yet. */
1630 r = journal_file_map_data_hash_table(f);
1631 if (r < 0)
1632 return r;
1633
1634 m = le64toh(READ_NOW(f->header->data_hash_table_size)) / sizeof(HashItem);
1635 if (m <= 0)
1636 return -EBADMSG;
1637
1638 h = hash % m;
1639 p = le64toh(f->data_hash_table[h].head_hash_offset);
1640
1641 while (p > 0) {
1642 Object *o;
1643 void *d;
1644 size_t rsize;
1645
1646 r = journal_file_move_to_object(f, OBJECT_DATA, p, &o);
1647 if (r < 0)
1648 return r;
1649
1650 if (le64toh(o->data.hash) != hash)
1651 goto next;
1652
1653 r = journal_file_data_payload(f, o, p, NULL, 0, 0, &d, &rsize);
1654 if (r < 0)
1655 return r;
1656 assert(r > 0); /* journal_file_data_payload() always returns > 0 if no field is provided. */
1657
1658 if (memcmp_nn(data, size, d, rsize) == 0) {
1659 if (ret_object)
1660 *ret_object = o;
1661
1662 if (ret_offset)
1663 *ret_offset = p;
1664
1665 return 1;
1666 }
1667
1668 next:
1669 r = get_next_hash_offset(
1670 f,
1671 &p,
1672 &o->data.next_hash_offset,
1673 &depth,
1674 JOURNAL_HEADER_CONTAINS(f->header, data_hash_chain_depth) ? &f->header->data_hash_chain_depth : NULL);
1675 if (r < 0)
1676 return r;
1677 }
1678
1679 return 0;
1680 }
1681
1682 int journal_file_find_data_object(
1683 JournalFile *f,
1684 const void *data,
1685 uint64_t size,
1686 Object **ret_object,
1687 uint64_t *ret_offset) {
1688
1689 assert(f);
1690 assert(data || size == 0);
1691
1692 return journal_file_find_data_object_with_hash(
1693 f,
1694 data, size,
1695 journal_file_hash_data(f, data, size),
1696 ret_object, ret_offset);
1697 }
1698
1699 bool journal_field_valid(const char *p, size_t l, bool allow_protected) {
1700 /* We kinda enforce POSIX syntax recommendations for
1701 environment variables here, but make a couple of additional
1702 requirements.
1703
1704 http://pubs.opengroup.org/onlinepubs/000095399/basedefs/xbd_chap08.html */
1705
1706 assert(p);
1707
1708 if (l == SIZE_MAX)
1709 l = strlen(p);
1710
1711 /* No empty field names */
1712 if (l <= 0)
1713 return false;
1714
1715 /* Don't allow names longer than 64 chars */
1716 if (l > 64)
1717 return false;
1718
1719 /* Variables starting with an underscore are protected */
1720 if (!allow_protected && p[0] == '_')
1721 return false;
1722
1723 /* Don't allow digits as first character */
1724 if (ascii_isdigit(p[0]))
1725 return false;
1726
1727 /* Only allow A-Z0-9 and '_' */
1728 for (const char *a = p; a < p + l; a++)
1729 if ((*a < 'A' || *a > 'Z') &&
1730 !ascii_isdigit(*a) &&
1731 *a != '_')
1732 return false;
1733
1734 return true;
1735 }
1736
1737 static int journal_file_append_field(
1738 JournalFile *f,
1739 const void *field,
1740 uint64_t size,
1741 Object **ret_object,
1742 uint64_t *ret_offset) {
1743
1744 uint64_t hash, p;
1745 uint64_t osize;
1746 Object *o;
1747 int r;
1748
1749 assert(f);
1750 assert(field);
1751 assert(size > 0);
1752
1753 if (!journal_field_valid(field, size, true))
1754 return -EBADMSG;
1755
1756 hash = journal_file_hash_data(f, field, size);
1757
1758 r = journal_file_find_field_object_with_hash(f, field, size, hash, ret_object, ret_offset);
1759 if (r < 0)
1760 return r;
1761 if (r > 0)
1762 return 0;
1763
1764 osize = offsetof(Object, field.payload) + size;
1765 r = journal_file_append_object(f, OBJECT_FIELD, osize, &o, &p);
1766 if (r < 0)
1767 return r;
1768
1769 o->field.hash = htole64(hash);
1770 memcpy(o->field.payload, field, size);
1771
1772 r = journal_file_link_field(f, o, p, hash);
1773 if (r < 0)
1774 return r;
1775
1776 /* The linking might have altered the window, so let's only pass the offset to hmac which will
1777 * move to the object again if needed. */
1778
1779 #if HAVE_GCRYPT
1780 r = journal_file_hmac_put_object(f, OBJECT_FIELD, NULL, p);
1781 if (r < 0)
1782 return r;
1783 #endif
1784
1785 if (ret_object) {
1786 r = journal_file_move_to_object(f, OBJECT_FIELD, p, ret_object);
1787 if (r < 0)
1788 return r;
1789 }
1790
1791 if (ret_offset)
1792 *ret_offset = p;
1793
1794 return 0;
1795 }
1796
1797 static int maybe_compress_payload(JournalFile *f, uint8_t *dst, const uint8_t *src, uint64_t size, size_t *rsize) {
1798 assert(f);
1799 assert(f->header);
1800
1801 #if HAVE_COMPRESSION
1802 Compression c;
1803 int r;
1804
1805 c = JOURNAL_FILE_COMPRESSION(f);
1806 if (c == COMPRESSION_NONE || size < f->compress_threshold_bytes)
1807 return 0;
1808
1809 r = compress_blob(c, src, size, dst, size - 1, rsize);
1810 if (r < 0)
1811 return log_debug_errno(r, "Failed to compress data object using %s, ignoring: %m", compression_to_string(c));
1812
1813 log_debug("Compressed data object %"PRIu64" -> %zu using %s", size, *rsize, compression_to_string(c));
1814
1815 return 1; /* compressed */
1816 #else
1817 return 0;
1818 #endif
1819 }
1820
1821 static int journal_file_append_data(
1822 JournalFile *f,
1823 const void *data,
1824 uint64_t size,
1825 Object **ret_object,
1826 uint64_t *ret_offset) {
1827
1828 uint64_t hash, p, osize;
1829 Object *o, *fo;
1830 size_t rsize = 0;
1831 const void *eq;
1832 int r;
1833
1834 assert(f);
1835
1836 if (!data || size == 0)
1837 return -EINVAL;
1838
1839 hash = journal_file_hash_data(f, data, size);
1840
1841 r = journal_file_find_data_object_with_hash(f, data, size, hash, ret_object, ret_offset);
1842 if (r < 0)
1843 return r;
1844 if (r > 0)
1845 return 0;
1846
1847 eq = memchr(data, '=', size);
1848 if (!eq)
1849 return -EINVAL;
1850
1851 osize = journal_file_data_payload_offset(f) + size;
1852 r = journal_file_append_object(f, OBJECT_DATA, osize, &o, &p);
1853 if (r < 0)
1854 return r;
1855
1856 o->data.hash = htole64(hash);
1857
1858 r = maybe_compress_payload(f, journal_file_data_payload_field(f, o), data, size, &rsize);
1859 if (r <= 0)
1860 /* We don't really care failures, let's continue without compression */
1861 memcpy_safe(journal_file_data_payload_field(f, o), data, size);
1862 else {
1863 Compression c = JOURNAL_FILE_COMPRESSION(f);
1864
1865 assert(c >= 0 && c < _COMPRESSION_MAX && c != COMPRESSION_NONE);
1866
1867 o->object.size = htole64(journal_file_data_payload_offset(f) + rsize);
1868 o->object.flags |= COMPRESSION_TO_OBJECT_FLAG(c);
1869 }
1870
1871 r = journal_file_link_data(f, o, p, hash);
1872 if (r < 0)
1873 return r;
1874
1875 /* The linking might have altered the window, so let's refresh our pointer. */
1876 r = journal_file_move_to_object(f, OBJECT_DATA, p, &o);
1877 if (r < 0)
1878 return r;
1879
1880 #if HAVE_GCRYPT
1881 r = journal_file_hmac_put_object(f, OBJECT_DATA, o, p);
1882 if (r < 0)
1883 return r;
1884 #endif
1885
1886 /* Create field object ... */
1887 r = journal_file_append_field(f, data, (uint8_t*) eq - (uint8_t*) data, &fo, NULL);
1888 if (r < 0)
1889 return r;
1890
1891 /* ... and link it in. */
1892 o->data.next_field_offset = fo->field.head_data_offset;
1893 fo->field.head_data_offset = le64toh(p);
1894
1895 if (ret_object)
1896 *ret_object = o;
1897
1898 if (ret_offset)
1899 *ret_offset = p;
1900
1901 return 0;
1902 }
1903
1904 static int maybe_decompress_payload(
1905 JournalFile *f,
1906 uint8_t *payload,
1907 uint64_t size,
1908 Compression compression,
1909 const char *field,
1910 size_t field_length,
1911 size_t data_threshold,
1912 void **ret_data,
1913 size_t *ret_size) {
1914
1915 assert(f);
1916
1917 /* We can't read objects larger than 4G on a 32-bit machine */
1918 if ((uint64_t) (size_t) size != size)
1919 return -E2BIG;
1920
1921 if (compression != COMPRESSION_NONE) {
1922 #if HAVE_COMPRESSION
1923 size_t rsize;
1924 int r;
1925
1926 if (field) {
1927 r = decompress_startswith(compression, payload, size, &f->compress_buffer, field,
1928 field_length, '=');
1929 if (r < 0)
1930 return log_debug_errno(r,
1931 "Cannot decompress %s object of length %" PRIu64 ": %m",
1932 compression_to_string(compression),
1933 size);
1934 if (r == 0) {
1935 if (ret_data)
1936 *ret_data = NULL;
1937 if (ret_size)
1938 *ret_size = 0;
1939 return 0;
1940 }
1941 }
1942
1943 r = decompress_blob(compression, payload, size, &f->compress_buffer, &rsize, 0);
1944 if (r < 0)
1945 return r;
1946
1947 if (ret_data)
1948 *ret_data = f->compress_buffer;
1949 if (ret_size)
1950 *ret_size = rsize;
1951 #else
1952 return -EPROTONOSUPPORT;
1953 #endif
1954 } else {
1955 if (field && (size < field_length + 1 || memcmp(payload, field, field_length) != 0 || payload[field_length] != '=')) {
1956 if (ret_data)
1957 *ret_data = NULL;
1958 if (ret_size)
1959 *ret_size = 0;
1960 return 0;
1961 }
1962
1963 if (ret_data)
1964 *ret_data = payload;
1965 if (ret_size)
1966 *ret_size = (size_t) size;
1967 }
1968
1969 return 1;
1970 }
1971
1972 int journal_file_data_payload(
1973 JournalFile *f,
1974 Object *o,
1975 uint64_t offset,
1976 const char *field,
1977 size_t field_length,
1978 size_t data_threshold,
1979 void **ret_data,
1980 size_t *ret_size) {
1981
1982 uint64_t size;
1983 Compression c;
1984 int r;
1985
1986 assert(f);
1987 assert(!field == (field_length == 0)); /* These must be specified together. */
1988
1989 if (!o) {
1990 r = journal_file_move_to_object(f, OBJECT_DATA, offset, &o);
1991 if (r < 0)
1992 return r;
1993 }
1994
1995 size = le64toh(READ_NOW(o->object.size));
1996 if (size < journal_file_data_payload_offset(f))
1997 return -EBADMSG;
1998
1999 size -= journal_file_data_payload_offset(f);
2000
2001 c = COMPRESSION_FROM_OBJECT(o);
2002 if (c < 0)
2003 return -EPROTONOSUPPORT;
2004
2005 return maybe_decompress_payload(f, journal_file_data_payload_field(f, o), size, c, field,
2006 field_length, data_threshold, ret_data, ret_size);
2007 }
2008
2009 uint64_t journal_file_entry_n_items(JournalFile *f, Object *o) {
2010 uint64_t sz;
2011
2012 assert(f);
2013 assert(o);
2014
2015 if (o->object.type != OBJECT_ENTRY)
2016 return 0;
2017
2018 sz = le64toh(READ_NOW(o->object.size));
2019 if (sz < offsetof(Object, entry.items))
2020 return 0;
2021
2022 return (sz - offsetof(Object, entry.items)) / journal_file_entry_item_size(f);
2023 }
2024
2025 uint64_t journal_file_entry_array_n_items(JournalFile *f, Object *o) {
2026 uint64_t sz;
2027
2028 assert(f);
2029 assert(o);
2030
2031 if (o->object.type != OBJECT_ENTRY_ARRAY)
2032 return 0;
2033
2034 sz = le64toh(READ_NOW(o->object.size));
2035 if (sz < offsetof(Object, entry_array.items))
2036 return 0;
2037
2038 return (sz - offsetof(Object, entry_array.items)) / journal_file_entry_array_item_size(f);
2039 }
2040
2041 uint64_t journal_file_hash_table_n_items(Object *o) {
2042 uint64_t sz;
2043
2044 assert(o);
2045
2046 if (!IN_SET(o->object.type, OBJECT_DATA_HASH_TABLE, OBJECT_FIELD_HASH_TABLE))
2047 return 0;
2048
2049 sz = le64toh(READ_NOW(o->object.size));
2050 if (sz < offsetof(Object, hash_table.items))
2051 return 0;
2052
2053 return (sz - offsetof(Object, hash_table.items)) / sizeof(HashItem);
2054 }
2055
2056 static void write_entry_array_item(JournalFile *f, Object *o, uint64_t i, uint64_t p) {
2057 assert(f);
2058 assert(o);
2059
2060 if (JOURNAL_HEADER_COMPACT(f->header)) {
2061 assert(p <= UINT32_MAX);
2062 o->entry_array.items.compact[i] = htole32(p);
2063 } else
2064 o->entry_array.items.regular[i] = htole64(p);
2065 }
2066
2067 static int link_entry_into_array(
2068 JournalFile *f,
2069 le64_t *first,
2070 le64_t *idx,
2071 le32_t *tail,
2072 le32_t *tidx,
2073 uint64_t p) {
2074
2075 uint64_t n = 0, ap = 0, q, i, a, hidx;
2076 Object *o;
2077 int r;
2078
2079 assert(f);
2080 assert(f->header);
2081 assert(first);
2082 assert(idx);
2083 assert(p > 0);
2084
2085 a = tail ? le32toh(*tail) : le64toh(*first);
2086 hidx = le64toh(READ_NOW(*idx));
2087 i = tidx ? le32toh(READ_NOW(*tidx)) : hidx;
2088
2089 while (a > 0) {
2090 r = journal_file_move_to_object(f, OBJECT_ENTRY_ARRAY, a, &o);
2091 if (r < 0)
2092 return r;
2093
2094 n = journal_file_entry_array_n_items(f, o);
2095 if (i < n) {
2096 write_entry_array_item(f, o, i, p);
2097 *idx = htole64(hidx + 1);
2098 if (tidx)
2099 *tidx = htole32(le32toh(*tidx) + 1);
2100 return 0;
2101 }
2102
2103 i -= n;
2104 ap = a;
2105 a = le64toh(o->entry_array.next_entry_array_offset);
2106 }
2107
2108 if (hidx > n)
2109 n = (hidx+1) * 2;
2110 else
2111 n = n * 2;
2112
2113 if (n < 4)
2114 n = 4;
2115
2116 r = journal_file_append_object(f, OBJECT_ENTRY_ARRAY,
2117 offsetof(Object, entry_array.items) + n * journal_file_entry_array_item_size(f),
2118 &o, &q);
2119 if (r < 0)
2120 return r;
2121
2122 #if HAVE_GCRYPT
2123 r = journal_file_hmac_put_object(f, OBJECT_ENTRY_ARRAY, o, q);
2124 if (r < 0)
2125 return r;
2126 #endif
2127
2128 write_entry_array_item(f, o, i, p);
2129
2130 if (ap == 0)
2131 *first = htole64(q);
2132 else {
2133 r = journal_file_move_to_object(f, OBJECT_ENTRY_ARRAY, ap, &o);
2134 if (r < 0)
2135 return r;
2136
2137 o->entry_array.next_entry_array_offset = htole64(q);
2138 }
2139
2140 if (tail)
2141 *tail = htole32(q);
2142
2143 if (JOURNAL_HEADER_CONTAINS(f->header, n_entry_arrays))
2144 f->header->n_entry_arrays = htole64(le64toh(f->header->n_entry_arrays) + 1);
2145
2146 *idx = htole64(hidx + 1);
2147 if (tidx)
2148 *tidx = htole32(1);
2149
2150 return 0;
2151 }
2152
2153 static int link_entry_into_array_plus_one(
2154 JournalFile *f,
2155 le64_t *extra,
2156 le64_t *first,
2157 le64_t *idx,
2158 le32_t *tail,
2159 le32_t *tidx,
2160 uint64_t p) {
2161
2162 uint64_t hidx;
2163 int r;
2164
2165 assert(f);
2166 assert(extra);
2167 assert(first);
2168 assert(idx);
2169 assert(p > 0);
2170
2171 hidx = le64toh(READ_NOW(*idx));
2172 if (hidx == UINT64_MAX)
2173 return -EBADMSG;
2174 if (hidx == 0)
2175 *extra = htole64(p);
2176 else {
2177 le64_t i;
2178
2179 i = htole64(hidx - 1);
2180 r = link_entry_into_array(f, first, &i, tail, tidx, p);
2181 if (r < 0)
2182 return r;
2183 }
2184
2185 *idx = htole64(hidx + 1);
2186 return 0;
2187 }
2188
2189 static int journal_file_link_entry_item(JournalFile *f, uint64_t offset, uint64_t p) {
2190 Object *o;
2191 int r;
2192
2193 assert(f);
2194 assert(offset > 0);
2195
2196 r = journal_file_move_to_object(f, OBJECT_DATA, p, &o);
2197 if (r < 0)
2198 return r;
2199
2200 return link_entry_into_array_plus_one(f,
2201 &o->data.entry_offset,
2202 &o->data.entry_array_offset,
2203 &o->data.n_entries,
2204 JOURNAL_HEADER_COMPACT(f->header) ? &o->data.compact.tail_entry_array_offset : NULL,
2205 JOURNAL_HEADER_COMPACT(f->header) ? &o->data.compact.tail_entry_array_n_entries : NULL,
2206 offset);
2207 }
2208
2209 static int journal_file_link_entry(
2210 JournalFile *f,
2211 Object *o,
2212 uint64_t offset,
2213 const EntryItem items[],
2214 size_t n_items) {
2215
2216 int r;
2217
2218 assert(f);
2219 assert(f->header);
2220 assert(o);
2221 assert(offset > 0);
2222
2223 if (o->object.type != OBJECT_ENTRY)
2224 return -EINVAL;
2225
2226 __atomic_thread_fence(__ATOMIC_SEQ_CST);
2227
2228 /* Link up the entry itself */
2229 r = link_entry_into_array(f,
2230 &f->header->entry_array_offset,
2231 &f->header->n_entries,
2232 JOURNAL_HEADER_CONTAINS(f->header, tail_entry_array_offset) ? &f->header->tail_entry_array_offset : NULL,
2233 JOURNAL_HEADER_CONTAINS(f->header, tail_entry_array_n_entries) ? &f->header->tail_entry_array_n_entries : NULL,
2234 offset);
2235 if (r < 0)
2236 return r;
2237
2238 /* log_debug("=> %s seqnr=%"PRIu64" n_entries=%"PRIu64, f->path, o->entry.seqnum, f->header->n_entries); */
2239
2240 if (f->header->head_entry_realtime == 0)
2241 f->header->head_entry_realtime = o->entry.realtime;
2242
2243 f->header->tail_entry_realtime = o->entry.realtime;
2244 f->header->tail_entry_monotonic = o->entry.monotonic;
2245 if (JOURNAL_HEADER_CONTAINS(f->header, tail_entry_offset))
2246 f->header->tail_entry_offset = htole64(offset);
2247 f->newest_mtime = 0; /* we have a new tail entry now, explicitly invalidate newest boot id/timestamp info */
2248
2249 /* Link up the items */
2250 for (uint64_t i = 0; i < n_items; i++) {
2251 int k;
2252
2253 /* If we fail to link an entry item because we can't allocate a new entry array, don't fail
2254 * immediately but try to link the other entry items since it might still be possible to link
2255 * those if they don't require a new entry array to be allocated. */
2256
2257 k = journal_file_link_entry_item(f, offset, items[i].object_offset);
2258 if (k == -E2BIG)
2259 r = k;
2260 else if (k < 0)
2261 return k;
2262 }
2263
2264 return r;
2265 }
2266
2267 static void write_entry_item(JournalFile *f, Object *o, uint64_t i, const EntryItem *item) {
2268 assert(f);
2269 assert(o);
2270 assert(item);
2271
2272 if (JOURNAL_HEADER_COMPACT(f->header)) {
2273 assert(item->object_offset <= UINT32_MAX);
2274 o->entry.items.compact[i].object_offset = htole32(item->object_offset);
2275 } else {
2276 o->entry.items.regular[i].object_offset = htole64(item->object_offset);
2277 o->entry.items.regular[i].hash = htole64(item->hash);
2278 }
2279 }
2280
2281 static int journal_file_append_entry_internal(
2282 JournalFile *f,
2283 const dual_timestamp *ts,
2284 const sd_id128_t *boot_id,
2285 const sd_id128_t *machine_id,
2286 uint64_t xor_hash,
2287 const EntryItem items[],
2288 size_t n_items,
2289 uint64_t *seqnum,
2290 sd_id128_t *seqnum_id,
2291 Object **ret_object,
2292 uint64_t *ret_offset) {
2293
2294 uint64_t np;
2295 uint64_t osize;
2296 Object *o;
2297 int r;
2298
2299 assert(f);
2300 assert(f->header);
2301 assert(ts);
2302 assert(boot_id);
2303 assert(!sd_id128_is_null(*boot_id));
2304 assert(items || n_items == 0);
2305
2306 if (f->strict_order) {
2307 /* If requested be stricter with ordering in this journal file, to make searching via
2308 * bisection fully deterministic. This is an optional feature, so that if desired journal
2309 * files can be written where the ordering is not strictly enforced (in which case bisection
2310 * will yield *a* result, but not the *only* result, when searching for points in
2311 * time). Strict ordering mode is enabled when journald originally writes the files, but
2312 * might not necessarily be if other tools (the remoting tools for example) write journal
2313 * files from combined sources.
2314 *
2315 * Typically, if any of the errors generated here are seen journald will just rotate the
2316 * journal files and start anew. */
2317
2318 if (ts->realtime < le64toh(f->header->tail_entry_realtime))
2319 return log_debug_errno(SYNTHETIC_ERRNO(EREMCHG),
2320 "Realtime timestamp %" PRIu64 " smaller than previous realtime "
2321 "timestamp %" PRIu64 ", refusing entry.",
2322 ts->realtime, le64toh(f->header->tail_entry_realtime));
2323
2324 if (sd_id128_equal(*boot_id, f->header->tail_entry_boot_id) &&
2325 ts->monotonic < le64toh(f->header->tail_entry_monotonic))
2326 return log_debug_errno(
2327 SYNTHETIC_ERRNO(ENOTNAM),
2328 "Monotonic timestamp %" PRIu64
2329 " smaller than previous monotonic timestamp %" PRIu64
2330 " while having the same boot ID, refusing entry.",
2331 ts->monotonic,
2332 le64toh(f->header->tail_entry_monotonic));
2333 }
2334
2335 if (seqnum_id) {
2336 /* Settle the passed in sequence number ID */
2337
2338 if (sd_id128_is_null(*seqnum_id))
2339 *seqnum_id = f->header->seqnum_id; /* Caller has none assigned, then copy the one from the file */
2340 else if (!sd_id128_equal(*seqnum_id, f->header->seqnum_id)) {
2341 /* Different seqnum IDs? We can't allow entries from multiple IDs end up in the same journal.*/
2342 if (le64toh(f->header->n_entries) == 0)
2343 f->header->seqnum_id = *seqnum_id; /* Caller has one, and file so far has no entries, then copy the one from the caller */
2344 else
2345 return log_debug_errno(SYNTHETIC_ERRNO(EILSEQ),
2346 "Sequence number IDs don't match, refusing entry.");
2347 }
2348 }
2349
2350 if (machine_id && sd_id128_is_null(f->header->machine_id))
2351 /* Initialize machine ID when not set yet */
2352 f->header->machine_id = *machine_id;
2353
2354 osize = offsetof(Object, entry.items) + (n_items * journal_file_entry_item_size(f));
2355
2356 r = journal_file_append_object(f, OBJECT_ENTRY, osize, &o, &np);
2357 if (r < 0)
2358 return r;
2359
2360 o->entry.seqnum = htole64(journal_file_entry_seqnum(f, seqnum));
2361 o->entry.realtime = htole64(ts->realtime);
2362 o->entry.monotonic = htole64(ts->monotonic);
2363 o->entry.xor_hash = htole64(xor_hash);
2364 o->entry.boot_id = f->header->tail_entry_boot_id = *boot_id;
2365
2366 for (size_t i = 0; i < n_items; i++)
2367 write_entry_item(f, o, i, &items[i]);
2368
2369 #if HAVE_GCRYPT
2370 r = journal_file_hmac_put_object(f, OBJECT_ENTRY, o, np);
2371 if (r < 0)
2372 return r;
2373 #endif
2374
2375 r = journal_file_link_entry(f, o, np, items, n_items);
2376 if (r < 0)
2377 return r;
2378
2379 if (ret_object)
2380 *ret_object = o;
2381
2382 if (ret_offset)
2383 *ret_offset = np;
2384
2385 return r;
2386 }
2387
2388 void journal_file_post_change(JournalFile *f) {
2389 assert(f);
2390
2391 if (f->fd < 0)
2392 return;
2393
2394 /* inotify() does not receive IN_MODIFY events from file
2395 * accesses done via mmap(). After each access we hence
2396 * trigger IN_MODIFY by truncating the journal file to its
2397 * current size which triggers IN_MODIFY. */
2398
2399 __atomic_thread_fence(__ATOMIC_SEQ_CST);
2400
2401 if (ftruncate(f->fd, f->last_stat.st_size) < 0)
2402 log_debug_errno(errno, "Failed to truncate file to its own size: %m");
2403 }
2404
2405 static int post_change_thunk(sd_event_source *timer, uint64_t usec, void *userdata) {
2406 assert(userdata);
2407
2408 journal_file_post_change(userdata);
2409
2410 return 1;
2411 }
2412
2413 static void schedule_post_change(JournalFile *f) {
2414 sd_event *e;
2415 int r;
2416
2417 assert(f);
2418 assert(f->post_change_timer);
2419
2420 assert_se(e = sd_event_source_get_event(f->post_change_timer));
2421
2422 /* If we are already going down, post the change immediately. */
2423 if (IN_SET(sd_event_get_state(e), SD_EVENT_EXITING, SD_EVENT_FINISHED))
2424 goto fail;
2425
2426 r = sd_event_source_get_enabled(f->post_change_timer, NULL);
2427 if (r < 0) {
2428 log_debug_errno(r, "Failed to get ftruncate timer state: %m");
2429 goto fail;
2430 }
2431 if (r > 0)
2432 return;
2433
2434 r = sd_event_source_set_time_relative(f->post_change_timer, f->post_change_timer_period);
2435 if (r < 0) {
2436 log_debug_errno(r, "Failed to set time for scheduling ftruncate: %m");
2437 goto fail;
2438 }
2439
2440 r = sd_event_source_set_enabled(f->post_change_timer, SD_EVENT_ONESHOT);
2441 if (r < 0) {
2442 log_debug_errno(r, "Failed to enable scheduled ftruncate: %m");
2443 goto fail;
2444 }
2445
2446 return;
2447
2448 fail:
2449 /* On failure, let's simply post the change immediately. */
2450 journal_file_post_change(f);
2451 }
2452
2453 /* Enable coalesced change posting in a timer on the provided sd_event instance */
2454 int journal_file_enable_post_change_timer(JournalFile *f, sd_event *e, usec_t t) {
2455 _cleanup_(sd_event_source_unrefp) sd_event_source *timer = NULL;
2456 int r;
2457
2458 assert(f);
2459 assert_return(!f->post_change_timer, -EINVAL);
2460 assert(e);
2461 assert(t);
2462
2463 r = sd_event_add_time(e, &timer, CLOCK_MONOTONIC, 0, 0, post_change_thunk, f);
2464 if (r < 0)
2465 return r;
2466
2467 r = sd_event_source_set_enabled(timer, SD_EVENT_OFF);
2468 if (r < 0)
2469 return r;
2470
2471 f->post_change_timer = TAKE_PTR(timer);
2472 f->post_change_timer_period = t;
2473
2474 return r;
2475 }
2476
2477 static int entry_item_cmp(const EntryItem *a, const EntryItem *b) {
2478 return CMP(ASSERT_PTR(a)->object_offset, ASSERT_PTR(b)->object_offset);
2479 }
2480
2481 static size_t remove_duplicate_entry_items(EntryItem items[], size_t n) {
2482 size_t j = 1;
2483
2484 assert(items || n == 0);
2485
2486 if (n <= 1)
2487 return n;
2488
2489 for (size_t i = 1; i < n; i++)
2490 if (items[i].object_offset != items[j - 1].object_offset)
2491 items[j++] = items[i];
2492
2493 return j;
2494 }
2495
2496 int journal_file_append_entry(
2497 JournalFile *f,
2498 const dual_timestamp *ts,
2499 const sd_id128_t *boot_id,
2500 const struct iovec iovec[],
2501 size_t n_iovec,
2502 uint64_t *seqnum,
2503 sd_id128_t *seqnum_id,
2504 Object **ret_object,
2505 uint64_t *ret_offset) {
2506
2507 _cleanup_free_ EntryItem *items_alloc = NULL;
2508 EntryItem *items;
2509 uint64_t xor_hash = 0;
2510 struct dual_timestamp _ts;
2511 sd_id128_t _boot_id, _machine_id, *machine_id;
2512 int r;
2513
2514 assert(f);
2515 assert(f->header);
2516 assert(iovec);
2517 assert(n_iovec > 0);
2518
2519 if (ts) {
2520 if (!VALID_REALTIME(ts->realtime))
2521 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG),
2522 "Invalid realtime timestamp %" PRIu64 ", refusing entry.",
2523 ts->realtime);
2524 if (!VALID_MONOTONIC(ts->monotonic))
2525 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG),
2526 "Invalid monotomic timestamp %" PRIu64 ", refusing entry.",
2527 ts->monotonic);
2528 } else {
2529 dual_timestamp_now(&_ts);
2530 ts = &_ts;
2531 }
2532
2533 if (boot_id) {
2534 if (sd_id128_is_null(*boot_id))
2535 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG), "Empty boot ID, refusing entry.");
2536 } else {
2537 r = sd_id128_get_boot(&_boot_id);
2538 if (r < 0)
2539 return r;
2540
2541 boot_id = &_boot_id;
2542 }
2543
2544 r = sd_id128_get_machine(&_machine_id);
2545 if (ERRNO_IS_NEG_MACHINE_ID_UNSET(r))
2546 /* Gracefully handle the machine ID not being initialized yet */
2547 machine_id = NULL;
2548 else if (r < 0)
2549 return r;
2550 else
2551 machine_id = &_machine_id;
2552
2553 #if HAVE_GCRYPT
2554 r = journal_file_maybe_append_tag(f, ts->realtime);
2555 if (r < 0)
2556 return r;
2557 #endif
2558
2559 if (n_iovec < ALLOCA_MAX / sizeof(EntryItem) / 2)
2560 items = newa(EntryItem, n_iovec);
2561 else {
2562 items_alloc = new(EntryItem, n_iovec);
2563 if (!items_alloc)
2564 return -ENOMEM;
2565
2566 items = items_alloc;
2567 }
2568
2569 for (size_t i = 0; i < n_iovec; i++) {
2570 uint64_t p;
2571 Object *o;
2572
2573 r = journal_file_append_data(f, iovec[i].iov_base, iovec[i].iov_len, &o, &p);
2574 if (r < 0)
2575 return r;
2576
2577 /* When calculating the XOR hash field, we need to take special care if the "keyed-hash"
2578 * journal file flag is on. We use the XOR hash field to quickly determine the identity of a
2579 * specific record, and give records with otherwise identical position (i.e. match in seqno,
2580 * timestamp, …) a stable ordering. But for that we can't have it that the hash of the
2581 * objects in each file is different since they are keyed. Hence let's calculate the Jenkins
2582 * hash here for that. This also has the benefit that cursors for old and new journal files
2583 * are completely identical (they include the XOR hash after all). For classic Jenkins-hash
2584 * files things are easier, we can just take the value from the stored record directly. */
2585
2586 if (JOURNAL_HEADER_KEYED_HASH(f->header))
2587 xor_hash ^= jenkins_hash64(iovec[i].iov_base, iovec[i].iov_len);
2588 else
2589 xor_hash ^= le64toh(o->data.hash);
2590
2591 items[i] = (EntryItem) {
2592 .object_offset = p,
2593 .hash = le64toh(o->data.hash),
2594 };
2595 }
2596
2597 /* Order by the position on disk, in order to improve seek
2598 * times for rotating media. */
2599 typesafe_qsort(items, n_iovec, entry_item_cmp);
2600 n_iovec = remove_duplicate_entry_items(items, n_iovec);
2601
2602 r = journal_file_append_entry_internal(
2603 f,
2604 ts,
2605 boot_id,
2606 machine_id,
2607 xor_hash,
2608 items,
2609 n_iovec,
2610 seqnum,
2611 seqnum_id,
2612 ret_object,
2613 ret_offset);
2614
2615 /* If the memory mapping triggered a SIGBUS then we return an
2616 * IO error and ignore the error code passed down to us, since
2617 * it is very likely just an effect of a nullified replacement
2618 * mapping page */
2619
2620 if (mmap_cache_fd_got_sigbus(f->cache_fd))
2621 r = -EIO;
2622
2623 if (f->post_change_timer)
2624 schedule_post_change(f);
2625 else
2626 journal_file_post_change(f);
2627
2628 return r;
2629 }
2630
2631 typedef struct ChainCacheItem {
2632 uint64_t first; /* The offset of the entry array object at the beginning of the chain,
2633 * i.e., le64toh(f->header->entry_array_offset), or le64toh(o->data.entry_offset). */
2634 uint64_t array; /* The offset of the cached entry array object. */
2635 uint64_t begin; /* The offset of the first item in the cached array. */
2636 uint64_t total; /* The total number of items in all arrays before the cached one in the chain. */
2637 uint64_t last_index; /* The last index we looked at in the cached array, to optimize locality when bisecting. */
2638 } ChainCacheItem;
2639
2640 static void chain_cache_put(
2641 OrderedHashmap *h,
2642 ChainCacheItem *ci,
2643 uint64_t first,
2644 uint64_t array,
2645 uint64_t begin,
2646 uint64_t total,
2647 uint64_t last_index) {
2648
2649 assert(h);
2650
2651 if (!ci) {
2652 /* If the chain item to cache for this chain is the
2653 * first one it's not worth caching anything */
2654 if (array == first)
2655 return;
2656
2657 if (ordered_hashmap_size(h) >= CHAIN_CACHE_MAX) {
2658 ci = ordered_hashmap_steal_first(h);
2659 assert(ci);
2660 } else {
2661 ci = new(ChainCacheItem, 1);
2662 if (!ci)
2663 return;
2664 }
2665
2666 ci->first = first;
2667
2668 if (ordered_hashmap_put(h, &ci->first, ci) < 0) {
2669 free(ci);
2670 return;
2671 }
2672 } else
2673 assert(ci->first == first);
2674
2675 ci->array = array;
2676 ci->begin = begin;
2677 ci->total = total;
2678 ci->last_index = last_index;
2679 }
2680
2681 static int bump_array_index(uint64_t *i, direction_t direction, uint64_t n) {
2682 assert(i);
2683
2684 /* Increase or decrease the specified index, in the right direction. */
2685
2686 if (direction == DIRECTION_DOWN) {
2687 if (*i >= n - 1)
2688 return 0;
2689
2690 (*i)++;
2691 } else {
2692 if (*i <= 0)
2693 return 0;
2694
2695 (*i)--;
2696 }
2697
2698 return 1;
2699 }
2700
2701 static int bump_entry_array(
2702 JournalFile *f,
2703 Object *o, /* the current entry array object. */
2704 uint64_t offset, /* the offset of the entry array object. */
2705 uint64_t first, /* The offset of the first entry array object in the chain. */
2706 direction_t direction,
2707 uint64_t *ret) {
2708
2709 int r;
2710
2711 assert(f);
2712 assert(ret);
2713
2714 if (direction == DIRECTION_DOWN) {
2715 assert(o);
2716 assert(o->object.type == OBJECT_ENTRY_ARRAY);
2717
2718 *ret = le64toh(o->entry_array.next_entry_array_offset);
2719 } else {
2720
2721 /* Entry array chains are a singly linked list, so to find the previous array in the chain, we have
2722 * to start iterating from the top. */
2723
2724 assert(offset > 0);
2725
2726 uint64_t p = first, q = 0;
2727 while (p > 0 && p != offset) {
2728 r = journal_file_move_to_object(f, OBJECT_ENTRY_ARRAY, p, &o);
2729 if (r < 0)
2730 return r;
2731
2732 q = p;
2733 p = le64toh(o->entry_array.next_entry_array_offset);
2734 }
2735
2736 /* If we can't find the previous entry array in the entry array chain, we're likely dealing with a
2737 * corrupted journal file. */
2738 if (p == 0)
2739 return -EBADMSG;
2740
2741 *ret = q;
2742 }
2743
2744 return *ret > 0;
2745 }
2746
2747 static int generic_array_get(
2748 JournalFile *f,
2749 uint64_t first, /* The offset of the first entry array object in the chain. */
2750 uint64_t i, /* The index of the target object counted from the beginning of the entry array chain. */
2751 direction_t direction,
2752 Object **ret_object, /* The found object. */
2753 uint64_t *ret_offset) { /* The offset of the found object. */
2754
2755 uint64_t a, t = 0, k;
2756 ChainCacheItem *ci;
2757 Object *o = NULL;
2758 int r;
2759
2760 assert(f);
2761
2762 /* FIXME: fix return value assignment on success. */
2763
2764 a = first;
2765
2766 /* Try the chain cache first */
2767 ci = ordered_hashmap_get(f->chain_cache, &first);
2768 if (ci && i > ci->total) {
2769 a = ci->array;
2770 i -= ci->total;
2771 t = ci->total;
2772 }
2773
2774 while (a > 0) {
2775 r = journal_file_move_to_object(f, OBJECT_ENTRY_ARRAY, a, &o);
2776 if (IN_SET(r, -EBADMSG, -EADDRNOTAVAIL)) {
2777 /* If there's corruption and we're going downwards, let's pretend we reached the
2778 * final entry in the entry array chain. */
2779
2780 if (direction == DIRECTION_DOWN)
2781 return 0;
2782
2783 /* If there's corruption and we're going upwards, move back to the previous entry
2784 * array and start iterating entries from there. */
2785
2786 i = UINT64_MAX;
2787 break;
2788 }
2789 if (r < 0)
2790 return r;
2791
2792 k = journal_file_entry_array_n_items(f, o);
2793 if (k == 0)
2794 return 0;
2795
2796 if (i < k)
2797 break;
2798
2799 /* The index is larger than the number of elements in the array. Let's move to the next array. */
2800 i -= k;
2801 t += k;
2802 a = le64toh(o->entry_array.next_entry_array_offset);
2803 }
2804
2805 /* If we've found the right location, now look for the first non-corrupt entry object (in the right
2806 * direction). */
2807
2808 while (a > 0) {
2809 if (i == UINT64_MAX) {
2810 r = bump_entry_array(f, o, a, first, direction, &a);
2811 if (r <= 0)
2812 return r;
2813
2814 r = journal_file_move_to_object(f, OBJECT_ENTRY_ARRAY, a, &o);
2815 if (r < 0)
2816 return r;
2817
2818 k = journal_file_entry_array_n_items(f, o);
2819 if (k == 0)
2820 break;
2821
2822 if (direction == DIRECTION_DOWN)
2823 i = 0;
2824 else {
2825 /* We moved to the previous array. The total must be decreased. */
2826 if (t < k)
2827 return -EBADMSG; /* chain cache is broken ? */
2828
2829 i = k - 1;
2830 t -= k;
2831 }
2832 }
2833
2834 do {
2835 uint64_t p;
2836
2837 p = journal_file_entry_array_item(f, o, i);
2838
2839 r = journal_file_move_to_object(f, OBJECT_ENTRY, p, ret_object);
2840 if (r >= 0) {
2841 /* Let's cache this item for the next invocation */
2842 chain_cache_put(f->chain_cache, ci, first, a, journal_file_entry_array_item(f, o, 0), t, i);
2843
2844 if (ret_offset)
2845 *ret_offset = p;
2846
2847 return 1;
2848 }
2849 if (!IN_SET(r, -EADDRNOTAVAIL, -EBADMSG))
2850 return r;
2851
2852 /* OK, so this entry is borked. Most likely some entry didn't get synced to
2853 * disk properly, let's see if the next one might work for us instead. */
2854 log_debug_errno(r, "Entry item %" PRIu64 " is bad, skipping over it.", i);
2855
2856 } while (bump_array_index(&i, direction, k) > 0);
2857
2858 /* All entries tried in the above do-while loop are broken. Let's move to the next (or previous) array. */
2859
2860 if (direction == DIRECTION_DOWN)
2861 /* We are going to the next array, the total must be incremented. */
2862 t += k;
2863
2864 i = UINT64_MAX;
2865 }
2866
2867 return 0;
2868 }
2869
2870 enum {
2871 TEST_FOUND, /* The current object passes the test. */
2872 TEST_LEFT, /* The current object is in an earlier position, and the object we are looking
2873 * for should exist in a later position. */
2874 TEST_RIGHT, /* The current object is in a later position, and the object we are looking for
2875 * should exist in an earlier position. */
2876 TEST_GOTO_NEXT, /* No matching object exists in this array and earlier arrays, go to the next array. */
2877 TEST_GOTO_PREVIOUS, /* No matching object exists in this array and later arrays, go to the previous array. */
2878 };
2879
2880 static int generic_array_bisect_step(
2881 JournalFile *f,
2882 Object *array, /* entry array object */
2883 uint64_t i, /* index of the entry item in the array we will test. */
2884 uint64_t needle,
2885 int (*test_object)(JournalFile *f, uint64_t p, uint64_t needle),
2886 direction_t direction,
2887 uint64_t *m, /* The maximum number of the entries we will check in the array. */
2888 uint64_t *left, /* The index of the left boundary in the array. */
2889 uint64_t *right) { /* The index of the right boundary in the array. */
2890
2891 uint64_t p;
2892 int r;
2893
2894 assert(f);
2895 assert(array);
2896 assert(test_object);
2897 assert(m);
2898 assert(left);
2899 assert(right);
2900 assert(*left <= i);
2901 assert(i <= *right);
2902 assert(*right < *m);
2903
2904 p = journal_file_entry_array_item(f, array, i);
2905 if (p <= 0)
2906 r = -EBADMSG;
2907 else
2908 r = test_object(f, p, needle);
2909 if (IN_SET(r, -EBADMSG, -EADDRNOTAVAIL)) {
2910 log_debug_errno(r, "Encountered invalid entry while bisecting, cutting algorithm short.");
2911
2912 if (i == *left) {
2913 /* This happens on two situations:
2914 *
2915 * a) i == 0 (hence, *left == 0):
2916 * The first entry in the array is corrupted, let's go back to the previous array.
2917 *
2918 * b) *right == *left or *left + 1, and we are going to downwards:
2919 * In that case, the (i-1)-th object has been already tested in the previous call,
2920 * which returned TEST_LEFT. See below. So, there is no matching entry in this
2921 * array nor in the whole entry array chain. */
2922 assert(i == 0 || (*right - *left <= 1 && direction == DIRECTION_DOWN));
2923 return TEST_GOTO_PREVIOUS;
2924 }
2925
2926 /* Otherwise, cutting the array short. So, here we limit the number of elements we will see
2927 * in this array, and set the right boundary to the last possibly non-corrupted object. */
2928 *m = i;
2929 *right = i - 1;
2930 return TEST_RIGHT;
2931 }
2932 if (r < 0)
2933 return r;
2934
2935 if (r == TEST_FOUND)
2936 /* There may be multiple entries that match with the needle. When the direction is down, we
2937 * need to find the first matching entry, hence the right boundary can be moved, but the left
2938 * one cannot. Similarly, when the direction is up, we need to find the last matching entry,
2939 * hence the left boundary can be moved, but the right one cannot. */
2940 r = direction == DIRECTION_DOWN ? TEST_RIGHT : TEST_LEFT;
2941
2942 if (r == TEST_RIGHT) {
2943 /* Currently, left --- needle --- i --- right, hence we can move the right boundary to i. */
2944 if (direction == DIRECTION_DOWN)
2945 *right = i;
2946 else {
2947 if (i == 0)
2948 return TEST_GOTO_PREVIOUS;
2949 *right = i - 1;
2950 }
2951 } else {
2952 /* Currently, left --- i --- needle --- right, hence we can move the left boundary to i. */
2953 if (direction == DIRECTION_DOWN) {
2954 /* Note, here *m is always positive, as by the assertions at the beginning, we have
2955 * 0 <= *left <= i <= *right < m */
2956 if (i == *m - 1)
2957 return TEST_GOTO_NEXT;
2958
2959 *left = i + 1;
2960 } else
2961 *left = i;
2962 }
2963
2964 return r;
2965 }
2966
2967 static int generic_array_bisect(
2968 JournalFile *f,
2969 uint64_t first, /* The offset of the first entry array object in the chain. */
2970 uint64_t n, /* The total number of elements in the chain of the entry array. */
2971 uint64_t needle, /* The target value (e.g. seqnum, monotonic, realtime, ...). */
2972 int (*test_object)(JournalFile *f,
2973 uint64_t p, /* the offset of the (data or entry) object that will be tested. */
2974 uint64_t needle),
2975 direction_t direction,
2976 Object **ret_object, /* The found object. */
2977 uint64_t *ret_offset, /* The offset of the found object. */
2978 uint64_t *ret_idx) { /* The index of the found object counted from the beginning of the entry array chain. */
2979
2980 /* Given an entry array chain, this function finds the object "closest" to the given needle in the
2981 * chain, taking into account the provided direction. A function can be provided to determine how
2982 * an object is matched against the given needle.
2983 *
2984 * Given a journal file, the offset of an object and the needle, the test_object() function should
2985 * return TEST_RIGHT if the needle is located earlier in the entry array chain, TEST_LEFT if the
2986 * needle is located later in the entry array chain, and TEST_FOUND if the object matches the needle.
2987 * If test_object() returns TEST_FOUND for a specific object, that object's information will be used
2988 * to populate the return values of this function. If test_object() never returns TEST_FOUND, the
2989 * return values are populated with the details of one of the objects closest to the needle. If the
2990 * direction is DIRECTION_UP, the earlier object is used. Otherwise, the later object is used.
2991 * If there are multiple objects that test_object() return TEST_FOUND for, then the first matching
2992 * object returned when direction is DIRECTION_DOWN. Otherwise the last object is returned. */
2993
2994 uint64_t a, p, t = 0, i, last_index = UINT64_MAX;
2995 ChainCacheItem *ci;
2996 Object *array;
2997 int r;
2998
2999 assert(f);
3000 assert(test_object);
3001
3002 if (n <= 0)
3003 return 0;
3004
3005 /* Start with the first array in the chain */
3006 a = first;
3007
3008 ci = ordered_hashmap_get(f->chain_cache, &first);
3009 if (ci && n > ci->total && ci->begin != 0) {
3010 /* Ah, we have iterated this bisection array chain previously! Let's see if we can skip ahead
3011 * in the chain, as far as the last time. But we can't jump backwards in the chain, so let's
3012 * check that first. */
3013
3014 r = test_object(f, ci->begin, needle);
3015 if (IN_SET(r, -EBADMSG, -EADDRNOTAVAIL))
3016 log_debug_errno(r, "Cached entry is corrupted, ignoring: %m");
3017 else if (r < 0)
3018 return r;
3019 else if (r == TEST_LEFT) {
3020 /* OK, what we are looking for is right of the begin of this EntryArray, so let's
3021 * jump straight to previously cached array in the chain */
3022
3023 a = ci->array;
3024 n -= ci->total;
3025 t = ci->total;
3026 last_index = ci->last_index;
3027 }
3028 }
3029
3030 while (a > 0) {
3031 uint64_t left, right, k, m, m_original;
3032
3033 r = journal_file_move_to_object(f, OBJECT_ENTRY_ARRAY, a, &array);
3034 if (r < 0)
3035 return r;
3036
3037 k = journal_file_entry_array_n_items(f, array);
3038 m = m_original = MIN(k, n);
3039 if (m <= 0)
3040 return 0;
3041
3042 left = 0;
3043 right = m - 1;
3044
3045 if (direction == DIRECTION_UP) {
3046 /* If we're going upwards, the last entry of the previous array may pass the test,
3047 * and the first entry of the current array may not pass. In that case, the last
3048 * entry of the previous array must be returned. Hence, we need to test the first
3049 * entry of the current array. */
3050 r = generic_array_bisect_step(f, array, 0, needle, test_object, direction, &m, &left, &right);
3051 if (r < 0)
3052 return r;
3053 if (r == TEST_GOTO_PREVIOUS)
3054 goto previous;
3055 }
3056
3057 /* Test the last entry of this array, to determine if we should go to the next array. */
3058 r = generic_array_bisect_step(f, array, right, needle, test_object, direction, &m, &left, &right);
3059 if (r < 0)
3060 return r;
3061 if (r == TEST_GOTO_PREVIOUS)
3062 goto previous;
3063
3064 /* The expected entry should be in this array, (or the last entry of the previous array). */
3065 if (r == TEST_RIGHT) {
3066
3067 /* If we cached the last index we looked at, let's try to not to jump too wildly
3068 * around and see if we can limit the range to look at early to the immediate
3069 * neighbors of the last index we looked at. */
3070
3071 if (last_index > 0 && left < last_index - 1 && last_index - 1 < right) {
3072 r = generic_array_bisect_step(f, array, last_index - 1, needle, test_object, direction, &m, &left, &right);
3073 if (r < 0)
3074 return r;
3075 if (r == TEST_GOTO_PREVIOUS)
3076 goto previous;
3077 }
3078
3079 if (last_index < UINT64_MAX && left < last_index + 1 && last_index + 1 < right) {
3080 r = generic_array_bisect_step(f, array, last_index + 1, needle, test_object, direction, &m, &left, &right);
3081 if (r < 0)
3082 return r;
3083 if (r == TEST_GOTO_PREVIOUS)
3084 goto previous;
3085 }
3086
3087 for (;;) {
3088 if (left == right) {
3089 /* We found one or more corrupted entries in generic_array_bisect_step().
3090 * In that case, the entry pointed by 'right' may not be tested.
3091 *
3092 * When we are going to downwards, the entry object pointed by 'left'
3093 * has not been tested yet, Hence, even if left == right, we still
3094 * have to check the final entry to see if it actually matches.
3095 *
3096 * On the other hand, when we are going to upwards, the entry pointed
3097 * by 'left' is always tested, So, it is not necessary to test the
3098 * final entry again. */
3099 if (m != m_original && direction == DIRECTION_DOWN) {
3100 r = generic_array_bisect_step(f, array, left, needle, test_object, direction, &m, &left, &right);
3101 if (r < 0)
3102 return r;
3103 if (IN_SET(r, TEST_GOTO_PREVIOUS, TEST_GOTO_NEXT))
3104 return 0; /* The entry does not pass the test, or is corrupted */
3105
3106 assert(TEST_RIGHT);
3107 assert(left == right);
3108 }
3109
3110 i = left;
3111 goto found;
3112 }
3113
3114 assert(left < right);
3115 i = (left + right + (direction == DIRECTION_UP)) / 2;
3116
3117 r = generic_array_bisect_step(f, array, i, needle, test_object, direction, &m, &left, &right);
3118 if (r < 0)
3119 return r;
3120 if (r == TEST_GOTO_PREVIOUS)
3121 goto previous;
3122 if (r == TEST_GOTO_NEXT)
3123 return 0; /* Found a corrupt entry, and the array was cut short. */
3124 }
3125 }
3126
3127 /* Not found in this array (or the last entry of this array should be returned), go to the next array. */
3128 assert(r == (direction == DIRECTION_DOWN ? TEST_GOTO_NEXT : TEST_LEFT));
3129
3130 if (k >= n) {
3131 if (direction == DIRECTION_UP) {
3132 assert(n > 0);
3133 i = n - 1;
3134 goto found;
3135 }
3136
3137 return 0;
3138 }
3139
3140 n -= k;
3141 t += k;
3142 last_index = UINT64_MAX;
3143 a = le64toh(array->entry_array.next_entry_array_offset);
3144 }
3145
3146 return 0;
3147
3148 previous:
3149 /* Not found in the current array, return the last entry of the previous array. */
3150 assert(r == TEST_GOTO_PREVIOUS);
3151
3152 /* The current array is the first in the chain. no previous array. */
3153 if (t == 0)
3154 return 0;
3155
3156 /* When we are going downwards, there is no matching entries in the previous array. */
3157 if (direction == DIRECTION_DOWN)
3158 return 0;
3159
3160 /* Indicate to go to the previous array later. Note, do not move to the previous array here,
3161 * as that may invalidate the current array object in the mmap cache and
3162 * journal_file_entry_array_item() below may read invalid address. */
3163 i = UINT64_MAX;
3164
3165 found:
3166 p = journal_file_entry_array_item(f, array, 0);
3167 if (p <= 0)
3168 return -EBADMSG;
3169
3170 /* Let's cache this item for the next invocation */
3171 chain_cache_put(f->chain_cache, ci, first, a, p, t, i);
3172
3173 if (i == UINT64_MAX) {
3174 uint64_t m;
3175
3176 /* Get the last entry of the previous array. */
3177
3178 r = bump_entry_array(f, NULL, a, first, DIRECTION_UP, &a);
3179 if (r <= 0)
3180 return r;
3181
3182 r = journal_file_move_to_object(f, OBJECT_ENTRY_ARRAY, a, &array);
3183 if (r < 0)
3184 return r;
3185
3186 m = journal_file_entry_array_n_items(f, array);
3187 if (m == 0 || t < m)
3188 return -EBADMSG;
3189
3190 t -= m;
3191 i = m - 1;
3192 }
3193
3194 p = journal_file_entry_array_item(f, array, i);
3195 if (p == 0)
3196 return -EBADMSG;
3197
3198 if (ret_object) {
3199 r = journal_file_move_to_object(f, OBJECT_ENTRY, p, ret_object);
3200 if (r < 0)
3201 return r;
3202 }
3203
3204 if (ret_offset)
3205 *ret_offset = p;
3206
3207 if (ret_idx)
3208 *ret_idx = t + i;
3209
3210 return 1;
3211 }
3212
3213 static int generic_array_bisect_for_data(
3214 JournalFile *f,
3215 Object *d,
3216 uint64_t needle,
3217 int (*test_object)(JournalFile *f, uint64_t p, uint64_t needle),
3218 direction_t direction,
3219 Object **ret_object,
3220 uint64_t *ret_offset) {
3221
3222 uint64_t extra, first, n;
3223 int r;
3224
3225 assert(f);
3226 assert(d);
3227 assert(d->object.type == OBJECT_DATA);
3228 assert(test_object);
3229
3230 n = le64toh(d->data.n_entries);
3231 if (n <= 0)
3232 return 0;
3233 n--; /* n_entries is the number of entries linked to the data object, including the 'extra' entry. */
3234
3235 extra = le64toh(d->data.entry_offset);
3236 first = le64toh(d->data.entry_array_offset);
3237
3238 /* This bisects the array in object 'first', but first checks an extra. */
3239 r = test_object(f, extra, needle);
3240 if (r < 0)
3241 return r;
3242
3243 if (direction == DIRECTION_DOWN) {
3244 /* If we are going downwards, then we need to return the first object that passes the test.
3245 * When there is no object that passes the test, we need to return the first object that
3246 * test_object() returns TEST_RIGHT for. */
3247 if (IN_SET(r,
3248 TEST_FOUND, /* The 'extra' object passes the test. Hence, this is the first
3249 * object that passes the test. */
3250 TEST_RIGHT)) /* The 'extra' object is the first object that test_object() returns
3251 * TEST_RIGHT for, and no object exists even in the chained arrays
3252 * that passes the test. */
3253 goto use_extra; /* The 'extra' object is exactly the one we are looking for. It is
3254 * not necessary to bisect the chained arrays. */
3255
3256 /* Otherwise, the 'extra' object is not the one we are looking for. Search in the arrays. */
3257
3258 } else {
3259 /* If we are going upwards, then we need to return the last object that passes the test.
3260 * When there is no object that passes the test, we need to return the the last object that
3261 * test_object() returns TEST_LEFT for. */
3262 if (r == TEST_RIGHT)
3263 return 0; /* Not only the 'extra' object, but also all objects in the chained arrays
3264 * will never get TEST_FOUND or TEST_LEFT. The object we are looking for
3265 * does not exist. */
3266
3267 /* Even if the 'extra' object passes the test, there may be multiple objects in the arrays
3268 * that also pass the test. Hence, we need to bisect the arrays for finding the last matching
3269 * object. */
3270 }
3271
3272 r = generic_array_bisect(f, first, n, needle, test_object, direction, ret_object, ret_offset, NULL);
3273 if (r != 0)
3274 return r; /* When > 0, the found object is the first (or last, when DIRECTION_UP) object.
3275 * Hence, return the found object now. */
3276
3277 /* No matching object found in the chained arrays.
3278 * DIRECTION_DOWN : the 'extra' object neither matches the condition. There is no matching object.
3279 * DIRECTION_UP : the 'extra' object matches the condition. So, return it. */
3280 if (direction == DIRECTION_DOWN)
3281 return 0;
3282
3283 use_extra:
3284 if (ret_object) {
3285 r = journal_file_move_to_object(f, OBJECT_ENTRY, extra, ret_object);
3286 if (r < 0)
3287 return r;
3288 }
3289
3290 if (ret_offset)
3291 *ret_offset = extra;
3292
3293 return 1;
3294 }
3295
3296 static int test_object_offset(JournalFile *f, uint64_t p, uint64_t needle) {
3297 assert(f);
3298 assert(p > 0);
3299
3300 if (p == needle)
3301 return TEST_FOUND;
3302 else if (p < needle)
3303 return TEST_LEFT;
3304 else
3305 return TEST_RIGHT;
3306 }
3307
3308 int journal_file_move_to_entry_by_offset(
3309 JournalFile *f,
3310 uint64_t p,
3311 direction_t direction,
3312 Object **ret_object,
3313 uint64_t *ret_offset) {
3314
3315 assert(f);
3316 assert(f->header);
3317
3318 return generic_array_bisect(
3319 f,
3320 le64toh(f->header->entry_array_offset),
3321 le64toh(f->header->n_entries),
3322 p,
3323 test_object_offset,
3324 direction,
3325 ret_object, ret_offset, NULL);
3326 }
3327
3328 static int test_object_seqnum(JournalFile *f, uint64_t p, uint64_t needle) {
3329 uint64_t sq;
3330 Object *o;
3331 int r;
3332
3333 assert(f);
3334 assert(p > 0);
3335
3336 r = journal_file_move_to_object(f, OBJECT_ENTRY, p, &o);
3337 if (r < 0)
3338 return r;
3339
3340 sq = le64toh(READ_NOW(o->entry.seqnum));
3341 if (sq == needle)
3342 return TEST_FOUND;
3343 else if (sq < needle)
3344 return TEST_LEFT;
3345 else
3346 return TEST_RIGHT;
3347 }
3348
3349 int journal_file_move_to_entry_by_seqnum(
3350 JournalFile *f,
3351 uint64_t seqnum,
3352 direction_t direction,
3353 Object **ret_object,
3354 uint64_t *ret_offset) {
3355
3356 assert(f);
3357 assert(f->header);
3358
3359 return generic_array_bisect(
3360 f,
3361 le64toh(f->header->entry_array_offset),
3362 le64toh(f->header->n_entries),
3363 seqnum,
3364 test_object_seqnum,
3365 direction,
3366 ret_object, ret_offset, NULL);
3367 }
3368
3369 static int test_object_realtime(JournalFile *f, uint64_t p, uint64_t needle) {
3370 Object *o;
3371 uint64_t rt;
3372 int r;
3373
3374 assert(f);
3375 assert(p > 0);
3376
3377 r = journal_file_move_to_object(f, OBJECT_ENTRY, p, &o);
3378 if (r < 0)
3379 return r;
3380
3381 rt = le64toh(READ_NOW(o->entry.realtime));
3382 if (rt == needle)
3383 return TEST_FOUND;
3384 else if (rt < needle)
3385 return TEST_LEFT;
3386 else
3387 return TEST_RIGHT;
3388 }
3389
3390 int journal_file_move_to_entry_by_realtime(
3391 JournalFile *f,
3392 uint64_t realtime,
3393 direction_t direction,
3394 Object **ret_object,
3395 uint64_t *ret_offset) {
3396
3397 assert(f);
3398 assert(f->header);
3399
3400 return generic_array_bisect(
3401 f,
3402 le64toh(f->header->entry_array_offset),
3403 le64toh(f->header->n_entries),
3404 realtime,
3405 test_object_realtime,
3406 direction,
3407 ret_object, ret_offset, NULL);
3408 }
3409
3410 static int test_object_monotonic(JournalFile *f, uint64_t p, uint64_t needle) {
3411 Object *o;
3412 uint64_t m;
3413 int r;
3414
3415 assert(f);
3416 assert(p > 0);
3417
3418 r = journal_file_move_to_object(f, OBJECT_ENTRY, p, &o);
3419 if (r < 0)
3420 return r;
3421
3422 m = le64toh(READ_NOW(o->entry.monotonic));
3423 if (m == needle)
3424 return TEST_FOUND;
3425 else if (m < needle)
3426 return TEST_LEFT;
3427 else
3428 return TEST_RIGHT;
3429 }
3430
3431 static int find_data_object_by_boot_id(
3432 JournalFile *f,
3433 sd_id128_t boot_id,
3434 Object **ret_object,
3435 uint64_t *ret_offset) {
3436
3437 char t[STRLEN("_BOOT_ID=") + 32 + 1] = "_BOOT_ID=";
3438
3439 assert(f);
3440
3441 sd_id128_to_string(boot_id, t + 9);
3442 return journal_file_find_data_object(f, t, sizeof(t) - 1, ret_object, ret_offset);
3443 }
3444
3445 int journal_file_move_to_entry_by_monotonic(
3446 JournalFile *f,
3447 sd_id128_t boot_id,
3448 uint64_t monotonic,
3449 direction_t direction,
3450 Object **ret_object,
3451 uint64_t *ret_offset) {
3452
3453 Object *o;
3454 int r;
3455
3456 assert(f);
3457
3458 r = find_data_object_by_boot_id(f, boot_id, &o, NULL);
3459 if (r <= 0)
3460 return r;
3461
3462 return generic_array_bisect_for_data(
3463 f,
3464 o,
3465 monotonic,
3466 test_object_monotonic,
3467 direction,
3468 ret_object, ret_offset);
3469 }
3470
3471 void journal_file_reset_location(JournalFile *f) {
3472 assert(f);
3473
3474 f->location_type = LOCATION_HEAD;
3475 f->current_offset = 0;
3476 f->current_seqnum = 0;
3477 f->current_realtime = 0;
3478 f->current_monotonic = 0;
3479 zero(f->current_boot_id);
3480 f->current_xor_hash = 0;
3481
3482 /* Also reset the previous reading direction. Otherwise, next_beyond_location() may wrongly handle we
3483 * already hit EOF. See issue #29216. */
3484 f->last_direction = _DIRECTION_INVALID;
3485 }
3486
3487 void journal_file_save_location(JournalFile *f, Object *o, uint64_t offset) {
3488 assert(f);
3489 assert(o);
3490
3491 f->location_type = LOCATION_SEEK;
3492 f->current_offset = offset;
3493 f->current_seqnum = le64toh(o->entry.seqnum);
3494 f->current_realtime = le64toh(o->entry.realtime);
3495 f->current_monotonic = le64toh(o->entry.monotonic);
3496 f->current_boot_id = o->entry.boot_id;
3497 f->current_xor_hash = le64toh(o->entry.xor_hash);
3498 }
3499
3500 static bool check_properly_ordered(uint64_t new_offset, uint64_t old_offset, direction_t direction) {
3501
3502 /* Consider it an error if any of the two offsets is uninitialized */
3503 if (old_offset == 0 || new_offset == 0)
3504 return false;
3505
3506 /* If we go down, the new offset must be larger than the old one. */
3507 return direction == DIRECTION_DOWN ?
3508 new_offset > old_offset :
3509 new_offset < old_offset;
3510 }
3511
3512 int journal_file_next_entry(
3513 JournalFile *f,
3514 uint64_t p,
3515 direction_t direction,
3516 Object **ret_object,
3517 uint64_t *ret_offset) {
3518
3519 uint64_t i, n, q;
3520 Object *o;
3521 int r;
3522
3523 assert(f);
3524 assert(f->header);
3525
3526 /* FIXME: fix return value assignment. */
3527
3528 n = le64toh(READ_NOW(f->header->n_entries));
3529 if (n <= 0)
3530 return 0;
3531
3532 /* When the input offset 'p' is zero, return the first (or last on DIRECTION_UP) entry. */
3533 if (p == 0)
3534 return generic_array_get(f,
3535 le64toh(f->header->entry_array_offset),
3536 direction == DIRECTION_DOWN ? 0 : n - 1,
3537 direction,
3538 ret_object, ret_offset);
3539
3540 /* Otherwise, first find the nearest entry object. */
3541 r = generic_array_bisect(f,
3542 le64toh(f->header->entry_array_offset),
3543 le64toh(f->header->n_entries),
3544 p,
3545 test_object_offset,
3546 direction,
3547 ret_object ? &o : NULL, &q, &i);
3548 if (r <= 0)
3549 return r;
3550
3551 assert(direction == DIRECTION_DOWN ? p <= q : q <= p);
3552
3553 /* If the input offset 'p' points to an entry object, generic_array_bisect() should provides
3554 * the same offset, and the index needs to be shifted. Otherwise, use the found object as is,
3555 * as it is the nearest entry object from the input offset 'p'. */
3556
3557 if (p != q)
3558 goto found;
3559
3560 r = bump_array_index(&i, direction, n);
3561 if (r <= 0)
3562 return r;
3563
3564 /* And jump to it */
3565 r = generic_array_get(f, le64toh(f->header->entry_array_offset), i, direction, ret_object ? &o : NULL, &q);
3566 if (r <= 0)
3567 return r;
3568
3569 /* Ensure our array is properly ordered. */
3570 if (!check_properly_ordered(q, p, direction))
3571 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG),
3572 "%s: entry array not properly ordered at entry index %" PRIu64,
3573 f->path, i);
3574 found:
3575 if (ret_object)
3576 *ret_object = o;
3577 if (ret_offset)
3578 *ret_offset = q;
3579
3580 return 1;
3581 }
3582
3583 int journal_file_move_to_entry_for_data(
3584 JournalFile *f,
3585 Object *d,
3586 direction_t direction,
3587 Object **ret_object,
3588 uint64_t *ret_offset) {
3589
3590 uint64_t extra, first, n;
3591 int r = 0;
3592
3593 assert(f);
3594 assert(d);
3595 assert(d->object.type == OBJECT_DATA);
3596 assert(IN_SET(direction, DIRECTION_DOWN, DIRECTION_UP));
3597
3598 /* FIXME: fix return value assignment. */
3599
3600 /* This returns the first (when the direction is down, otherwise the last) entry linked to the
3601 * specified data object. */
3602
3603 n = le64toh(d->data.n_entries);
3604 if (n <= 0)
3605 return 0;
3606 n--; /* n_entries is the number of entries linked to the data object, including the 'extra' entry. */
3607
3608 extra = le64toh(d->data.entry_offset);
3609 first = le64toh(d->data.entry_array_offset);
3610
3611 if (direction == DIRECTION_DOWN && extra > 0) {
3612 /* When we are going downwards, first try to read the extra entry. */
3613 r = journal_file_move_to_object(f, OBJECT_ENTRY, extra, ret_object);
3614 if (r >= 0)
3615 goto use_extra;
3616 if (!IN_SET(r, -EADDRNOTAVAIL, -EBADMSG))
3617 return r;
3618 }
3619
3620 if (n > 0) {
3621 /* DIRECTION_DOWN : The extra entry is broken, falling back to the entries in the array.
3622 * DIRECTION_UP : Try to find a valid entry in the array from the tail. */
3623 r = generic_array_get(f,
3624 first,
3625 direction == DIRECTION_DOWN ? 0 : n - 1,
3626 direction,
3627 ret_object, ret_offset);
3628 if (!IN_SET(r, 0, -EADDRNOTAVAIL, -EBADMSG))
3629 return r; /* found or critical error. */
3630 }
3631
3632 if (direction == DIRECTION_UP && extra > 0) {
3633 /* No valid entry exists in the chained array, falling back to the extra entry. */
3634 r = journal_file_move_to_object(f, OBJECT_ENTRY, extra, ret_object);
3635 if (r >= 0)
3636 goto use_extra;
3637 }
3638
3639 return r;
3640
3641 use_extra:
3642 if (ret_offset)
3643 *ret_offset = extra;
3644
3645 return 1;
3646 }
3647
3648 int journal_file_move_to_entry_by_offset_for_data(
3649 JournalFile *f,
3650 Object *d,
3651 uint64_t p,
3652 direction_t direction,
3653 Object **ret, uint64_t *ret_offset) {
3654
3655 assert(f);
3656 assert(d);
3657 assert(d->object.type == OBJECT_DATA);
3658
3659 return generic_array_bisect_for_data(
3660 f,
3661 d,
3662 p,
3663 test_object_offset,
3664 direction,
3665 ret, ret_offset);
3666 }
3667
3668 int journal_file_move_to_entry_by_monotonic_for_data(
3669 JournalFile *f,
3670 Object *d,
3671 sd_id128_t boot_id,
3672 uint64_t monotonic,
3673 direction_t direction,
3674 Object **ret_object,
3675 uint64_t *ret_offset) {
3676
3677 Object *o, *entry;
3678 uint64_t z;
3679 int r;
3680
3681 assert(f);
3682 assert(d);
3683 assert(d->object.type == OBJECT_DATA);
3684
3685 /* First, pin the given data object, before reading the _BOOT_ID= data object below. */
3686 r = journal_file_pin_object(f, d);
3687 if (r < 0)
3688 return r;
3689
3690 /* Then, read a data object for _BOOT_ID= and seek by time. */
3691 r = find_data_object_by_boot_id(f, boot_id, &o, NULL);
3692 if (r <= 0)
3693 return r;
3694
3695 r = generic_array_bisect_for_data(f,
3696 o,
3697 monotonic,
3698 test_object_monotonic,
3699 direction,
3700 NULL, &z);
3701 if (r <= 0)
3702 return r;
3703
3704 /* And now, continue seeking until we find an entry that exists in both bisection arrays. */
3705 for (;;) {
3706 uint64_t p;
3707
3708 /* The journal entry found by the above bisect_plus_one() may not have the specified data,
3709 * that is, it may not be linked in the data object. So, we need to check that. */
3710
3711 r = journal_file_move_to_entry_by_offset_for_data(
3712 f, d, z, direction, ret_object ? &entry : NULL, &p);
3713 if (r <= 0)
3714 return r;
3715 if (p == z)
3716 break; /* The journal entry has the specified data. Yay! */
3717
3718 /* If the entry does not have the data, then move to the next (or previous, depends on the
3719 * 'direction') entry linked to the data object. But, the next entry may be in another boot.
3720 * So, we need to check that the entry has the matching boot ID. */
3721
3722 r = journal_file_move_to_entry_by_offset_for_data(
3723 f, o, p, direction, ret_object ? &entry : NULL, &z);
3724 if (r <= 0)
3725 return r;
3726 if (p == z)
3727 break; /* The journal entry has the specified boot ID. Yay! */
3728
3729 /* If not, let's try to the next entry... */
3730 }
3731
3732 if (ret_object)
3733 *ret_object = entry;
3734 if (ret_offset)
3735 *ret_offset = z;
3736 return 1;
3737 }
3738
3739 int journal_file_move_to_entry_by_seqnum_for_data(
3740 JournalFile *f,
3741 Object *d,
3742 uint64_t seqnum,
3743 direction_t direction,
3744 Object **ret_object,
3745 uint64_t *ret_offset) {
3746
3747 assert(f);
3748 assert(d);
3749 assert(d->object.type == OBJECT_DATA);
3750
3751 return generic_array_bisect_for_data(
3752 f,
3753 d,
3754 seqnum,
3755 test_object_seqnum,
3756 direction,
3757 ret_object, ret_offset);
3758 }
3759
3760 int journal_file_move_to_entry_by_realtime_for_data(
3761 JournalFile *f,
3762 Object *d,
3763 uint64_t realtime,
3764 direction_t direction,
3765 Object **ret, uint64_t *ret_offset) {
3766
3767 assert(f);
3768 assert(d);
3769 assert(d->object.type == OBJECT_DATA);
3770
3771 return generic_array_bisect_for_data(
3772 f,
3773 d,
3774 realtime,
3775 test_object_realtime,
3776 direction,
3777 ret, ret_offset);
3778 }
3779
3780 void journal_file_dump(JournalFile *f) {
3781 Object *o;
3782 uint64_t p;
3783 int r;
3784
3785 assert(f);
3786 assert(f->header);
3787
3788 journal_file_print_header(f);
3789
3790 p = le64toh(READ_NOW(f->header->header_size));
3791 while (p != 0) {
3792 const char *s;
3793 Compression c;
3794
3795 r = journal_file_move_to_object(f, OBJECT_UNUSED, p, &o);
3796 if (r < 0)
3797 goto fail;
3798
3799 s = journal_object_type_to_string(o->object.type);
3800
3801 switch (o->object.type) {
3802
3803 case OBJECT_ENTRY:
3804 assert(s);
3805
3806 printf("Type: %s seqnum=%"PRIu64" monotonic=%"PRIu64" realtime=%"PRIu64"\n",
3807 s,
3808 le64toh(o->entry.seqnum),
3809 le64toh(o->entry.monotonic),
3810 le64toh(o->entry.realtime));
3811 break;
3812
3813 case OBJECT_TAG:
3814 assert(s);
3815
3816 printf("Type: %s seqnum=%"PRIu64" epoch=%"PRIu64"\n",
3817 s,
3818 le64toh(o->tag.seqnum),
3819 le64toh(o->tag.epoch));
3820 break;
3821
3822 default:
3823 if (s)
3824 printf("Type: %s \n", s);
3825 else
3826 printf("Type: unknown (%i)", o->object.type);
3827
3828 break;
3829 }
3830
3831 c = COMPRESSION_FROM_OBJECT(o);
3832 if (c > COMPRESSION_NONE)
3833 printf("Flags: %s\n",
3834 compression_to_string(c));
3835
3836 if (p == le64toh(f->header->tail_object_offset))
3837 p = 0;
3838 else
3839 p += ALIGN64(le64toh(o->object.size));
3840 }
3841
3842 return;
3843 fail:
3844 log_error("File corrupt");
3845 }
3846
3847 /* Note: the lifetime of the compound literal is the immediately surrounding block. */
3848 #define FORMAT_TIMESTAMP_SAFE(t) (FORMAT_TIMESTAMP(t) ?: " --- ")
3849
3850 void journal_file_print_header(JournalFile *f) {
3851 struct stat st;
3852
3853 assert(f);
3854 assert(f->header);
3855
3856 printf("File path: %s\n"
3857 "File ID: %s\n"
3858 "Machine ID: %s\n"
3859 "Boot ID: %s\n"
3860 "Sequential number ID: %s\n"
3861 "State: %s\n"
3862 "Compatible flags:%s%s%s%s\n"
3863 "Incompatible flags:%s%s%s%s%s%s\n"
3864 "Header size: %"PRIu64"\n"
3865 "Arena size: %"PRIu64"\n"
3866 "Data hash table size: %"PRIu64"\n"
3867 "Field hash table size: %"PRIu64"\n"
3868 "Rotate suggested: %s\n"
3869 "Head sequential number: %"PRIu64" (%"PRIx64")\n"
3870 "Tail sequential number: %"PRIu64" (%"PRIx64")\n"
3871 "Head realtime timestamp: %s (%"PRIx64")\n"
3872 "Tail realtime timestamp: %s (%"PRIx64")\n"
3873 "Tail monotonic timestamp: %s (%"PRIx64")\n"
3874 "Objects: %"PRIu64"\n"
3875 "Entry objects: %"PRIu64"\n",
3876 f->path,
3877 SD_ID128_TO_STRING(f->header->file_id),
3878 SD_ID128_TO_STRING(f->header->machine_id),
3879 SD_ID128_TO_STRING(f->header->tail_entry_boot_id),
3880 SD_ID128_TO_STRING(f->header->seqnum_id),
3881 f->header->state == STATE_OFFLINE ? "OFFLINE" :
3882 f->header->state == STATE_ONLINE ? "ONLINE" :
3883 f->header->state == STATE_ARCHIVED ? "ARCHIVED" : "UNKNOWN",
3884 JOURNAL_HEADER_SEALED(f->header) ? " SEALED" : "",
3885 JOURNAL_HEADER_SEALED_CONTINUOUS(f->header) ? " SEALED_CONTINUOUS" : "",
3886 JOURNAL_HEADER_TAIL_ENTRY_BOOT_ID(f->header) ? " TAIL_ENTRY_BOOT_ID" : "",
3887 (le32toh(f->header->compatible_flags) & ~HEADER_COMPATIBLE_ANY) ? " ???" : "",
3888 JOURNAL_HEADER_COMPRESSED_XZ(f->header) ? " COMPRESSED-XZ" : "",
3889 JOURNAL_HEADER_COMPRESSED_LZ4(f->header) ? " COMPRESSED-LZ4" : "",
3890 JOURNAL_HEADER_COMPRESSED_ZSTD(f->header) ? " COMPRESSED-ZSTD" : "",
3891 JOURNAL_HEADER_KEYED_HASH(f->header) ? " KEYED-HASH" : "",
3892 JOURNAL_HEADER_COMPACT(f->header) ? " COMPACT" : "",
3893 (le32toh(f->header->incompatible_flags) & ~HEADER_INCOMPATIBLE_ANY) ? " ???" : "",
3894 le64toh(f->header->header_size),
3895 le64toh(f->header->arena_size),
3896 le64toh(f->header->data_hash_table_size) / sizeof(HashItem),
3897 le64toh(f->header->field_hash_table_size) / sizeof(HashItem),
3898 yes_no(journal_file_rotate_suggested(f, 0, LOG_DEBUG)),
3899 le64toh(f->header->head_entry_seqnum), le64toh(f->header->head_entry_seqnum),
3900 le64toh(f->header->tail_entry_seqnum), le64toh(f->header->tail_entry_seqnum),
3901 FORMAT_TIMESTAMP_SAFE(le64toh(f->header->head_entry_realtime)), le64toh(f->header->head_entry_realtime),
3902 FORMAT_TIMESTAMP_SAFE(le64toh(f->header->tail_entry_realtime)), le64toh(f->header->tail_entry_realtime),
3903 FORMAT_TIMESPAN(le64toh(f->header->tail_entry_monotonic), USEC_PER_MSEC), le64toh(f->header->tail_entry_monotonic),
3904 le64toh(f->header->n_objects),
3905 le64toh(f->header->n_entries));
3906
3907 if (JOURNAL_HEADER_CONTAINS(f->header, n_data))
3908 printf("Data objects: %"PRIu64"\n"
3909 "Data hash table fill: %.1f%%\n",
3910 le64toh(f->header->n_data),
3911 100.0 * (double) le64toh(f->header->n_data) / ((double) (le64toh(f->header->data_hash_table_size) / sizeof(HashItem))));
3912
3913 if (JOURNAL_HEADER_CONTAINS(f->header, n_fields))
3914 printf("Field objects: %"PRIu64"\n"
3915 "Field hash table fill: %.1f%%\n",
3916 le64toh(f->header->n_fields),
3917 100.0 * (double) le64toh(f->header->n_fields) / ((double) (le64toh(f->header->field_hash_table_size) / sizeof(HashItem))));
3918
3919 if (JOURNAL_HEADER_CONTAINS(f->header, n_tags))
3920 printf("Tag objects: %"PRIu64"\n",
3921 le64toh(f->header->n_tags));
3922 if (JOURNAL_HEADER_CONTAINS(f->header, n_entry_arrays))
3923 printf("Entry array objects: %"PRIu64"\n",
3924 le64toh(f->header->n_entry_arrays));
3925
3926 if (JOURNAL_HEADER_CONTAINS(f->header, field_hash_chain_depth))
3927 printf("Deepest field hash chain: %" PRIu64"\n",
3928 f->header->field_hash_chain_depth);
3929
3930 if (JOURNAL_HEADER_CONTAINS(f->header, data_hash_chain_depth))
3931 printf("Deepest data hash chain: %" PRIu64"\n",
3932 f->header->data_hash_chain_depth);
3933
3934 if (fstat(f->fd, &st) >= 0)
3935 printf("Disk usage: %s\n", FORMAT_BYTES((uint64_t) st.st_blocks * 512ULL));
3936 }
3937
3938 static int journal_file_warn_btrfs(JournalFile *f) {
3939 unsigned attrs;
3940 int r;
3941
3942 assert(f);
3943
3944 /* Before we write anything, check if the COW logic is turned
3945 * off on btrfs. Given our write pattern that is quite
3946 * unfriendly to COW file systems this should greatly improve
3947 * performance on COW file systems, such as btrfs, at the
3948 * expense of data integrity features (which shouldn't be too
3949 * bad, given that we do our own checksumming). */
3950
3951 r = fd_is_fs_type(f->fd, BTRFS_SUPER_MAGIC);
3952 if (r < 0)
3953 return log_ratelimit_warning_errno(r, JOURNAL_LOG_RATELIMIT, "Failed to determine if journal is on btrfs: %m");
3954 if (r == 0)
3955 return 0;
3956
3957 r = read_attr_fd(f->fd, &attrs);
3958 if (r < 0)
3959 return log_ratelimit_warning_errno(r, JOURNAL_LOG_RATELIMIT, "Failed to read file attributes: %m");
3960
3961 if (attrs & FS_NOCOW_FL) {
3962 log_debug("Detected btrfs file system with copy-on-write disabled, all is good.");
3963 return 0;
3964 }
3965
3966 log_ratelimit_notice(JOURNAL_LOG_RATELIMIT,
3967 "Creating journal file %s on a btrfs file system, and copy-on-write is enabled. "
3968 "This is likely to slow down journal access substantially, please consider turning "
3969 "off the copy-on-write file attribute on the journal directory, using chattr +C.",
3970 f->path);
3971
3972 return 1;
3973 }
3974
3975 static void journal_default_metrics(JournalMetrics *m, int fd, bool compact) {
3976 struct statvfs ss;
3977 uint64_t fs_size = 0;
3978
3979 assert(m);
3980 assert(fd >= 0);
3981
3982 if (fstatvfs(fd, &ss) >= 0)
3983 fs_size = u64_multiply_safe(ss.f_frsize, ss.f_blocks);
3984 else
3985 log_debug_errno(errno, "Failed to determine disk size: %m");
3986
3987 if (m->max_use == UINT64_MAX) {
3988
3989 if (fs_size > 0)
3990 m->max_use = CLAMP(PAGE_ALIGN_U64(fs_size / 10), /* 10% of file system size */
3991 MAX_USE_LOWER, MAX_USE_UPPER);
3992 else
3993 m->max_use = MAX_USE_LOWER;
3994 } else {
3995 m->max_use = PAGE_ALIGN_U64(m->max_use);
3996
3997 if (m->max_use != 0 && m->max_use < JOURNAL_FILE_SIZE_MIN*2)
3998 m->max_use = JOURNAL_FILE_SIZE_MIN*2;
3999 }
4000
4001 if (m->min_use == UINT64_MAX) {
4002 if (fs_size > 0)
4003 m->min_use = CLAMP(PAGE_ALIGN_U64(fs_size / 50), /* 2% of file system size */
4004 MIN_USE_LOW, MIN_USE_HIGH);
4005 else
4006 m->min_use = MIN_USE_LOW;
4007 }
4008
4009 if (m->min_use > m->max_use)
4010 m->min_use = m->max_use;
4011
4012 if (m->max_size == UINT64_MAX)
4013 m->max_size = MIN(PAGE_ALIGN_U64(m->max_use / 8), /* 8 chunks */
4014 MAX_SIZE_UPPER);
4015 else
4016 m->max_size = PAGE_ALIGN_U64(m->max_size);
4017
4018 if (compact && m->max_size > JOURNAL_COMPACT_SIZE_MAX)
4019 m->max_size = JOURNAL_COMPACT_SIZE_MAX;
4020
4021 if (m->max_size != 0) {
4022 if (m->max_size < JOURNAL_FILE_SIZE_MIN)
4023 m->max_size = JOURNAL_FILE_SIZE_MIN;
4024
4025 if (m->max_use != 0 && m->max_size*2 > m->max_use)
4026 m->max_use = m->max_size*2;
4027 }
4028
4029 if (m->min_size == UINT64_MAX)
4030 m->min_size = JOURNAL_FILE_SIZE_MIN;
4031 else
4032 m->min_size = CLAMP(PAGE_ALIGN_U64(m->min_size),
4033 JOURNAL_FILE_SIZE_MIN,
4034 m->max_size ?: UINT64_MAX);
4035
4036 if (m->keep_free == UINT64_MAX) {
4037 if (fs_size > 0)
4038 m->keep_free = MIN(PAGE_ALIGN_U64(fs_size / 20), /* 5% of file system size */
4039 KEEP_FREE_UPPER);
4040 else
4041 m->keep_free = DEFAULT_KEEP_FREE;
4042 }
4043
4044 if (m->n_max_files == UINT64_MAX)
4045 m->n_max_files = DEFAULT_N_MAX_FILES;
4046
4047 log_debug("Fixed min_use=%s max_use=%s max_size=%s min_size=%s keep_free=%s n_max_files=%" PRIu64,
4048 FORMAT_BYTES(m->min_use),
4049 FORMAT_BYTES(m->max_use),
4050 FORMAT_BYTES(m->max_size),
4051 FORMAT_BYTES(m->min_size),
4052 FORMAT_BYTES(m->keep_free),
4053 m->n_max_files);
4054 }
4055
4056 int journal_file_open(
4057 int fd,
4058 const char *fname,
4059 int open_flags,
4060 JournalFileFlags file_flags,
4061 mode_t mode,
4062 uint64_t compress_threshold_bytes,
4063 JournalMetrics *metrics,
4064 MMapCache *mmap_cache,
4065 JournalFile *template,
4066 JournalFile **ret) {
4067
4068 bool newly_created = false;
4069 JournalFile *f;
4070 void *h;
4071 int r;
4072
4073 assert(fd >= 0 || fname);
4074 assert(file_flags >= 0);
4075 assert(file_flags <= _JOURNAL_FILE_FLAGS_MAX);
4076 assert(mmap_cache);
4077 assert(ret);
4078
4079 if (!IN_SET((open_flags & O_ACCMODE), O_RDONLY, O_RDWR))
4080 return -EINVAL;
4081
4082 if ((open_flags & O_ACCMODE) == O_RDONLY && FLAGS_SET(open_flags, O_CREAT))
4083 return -EINVAL;
4084
4085 if (fname && (open_flags & O_CREAT) && !endswith(fname, ".journal"))
4086 return -EINVAL;
4087
4088 f = new(JournalFile, 1);
4089 if (!f)
4090 return -ENOMEM;
4091
4092 *f = (JournalFile) {
4093 .fd = fd,
4094 .mode = mode,
4095 .open_flags = open_flags,
4096 .compress_threshold_bytes = compress_threshold_bytes == UINT64_MAX ?
4097 DEFAULT_COMPRESS_THRESHOLD :
4098 MAX(MIN_COMPRESS_THRESHOLD, compress_threshold_bytes),
4099 .strict_order = FLAGS_SET(file_flags, JOURNAL_STRICT_ORDER),
4100 .newest_boot_id_prioq_idx = PRIOQ_IDX_NULL,
4101 .last_direction = _DIRECTION_INVALID,
4102 };
4103
4104 if (fname) {
4105 f->path = strdup(fname);
4106 if (!f->path) {
4107 r = -ENOMEM;
4108 goto fail;
4109 }
4110 } else {
4111 assert(fd >= 0);
4112
4113 /* If we don't know the path, fill in something explanatory and vaguely useful */
4114 if (asprintf(&f->path, "/proc/self/%i", fd) < 0) {
4115 r = -ENOMEM;
4116 goto fail;
4117 }
4118 }
4119
4120 f->chain_cache = ordered_hashmap_new(&uint64_hash_ops);
4121 if (!f->chain_cache) {
4122 r = -ENOMEM;
4123 goto fail;
4124 }
4125
4126 if (f->fd < 0) {
4127 /* We pass O_NONBLOCK here, so that in case somebody pointed us to some character device node or FIFO
4128 * or so, we likely fail quickly than block for long. For regular files O_NONBLOCK has no effect, hence
4129 * it doesn't hurt in that case. */
4130
4131 f->fd = openat_report_new(AT_FDCWD, f->path, f->open_flags|O_CLOEXEC|O_NONBLOCK, f->mode, &newly_created);
4132 if (f->fd < 0) {
4133 r = f->fd;
4134 goto fail;
4135 }
4136
4137 /* fds we opened here by us should also be closed by us. */
4138 f->close_fd = true;
4139
4140 r = fd_nonblock(f->fd, false);
4141 if (r < 0)
4142 goto fail;
4143
4144 if (!newly_created) {
4145 r = journal_file_fstat(f);
4146 if (r < 0)
4147 goto fail;
4148 }
4149 } else {
4150 r = journal_file_fstat(f);
4151 if (r < 0)
4152 goto fail;
4153
4154 /* If we just got the fd passed in, we don't really know if we created the file anew */
4155 newly_created = f->last_stat.st_size == 0 && journal_file_writable(f);
4156 }
4157
4158 r = mmap_cache_add_fd(mmap_cache, f->fd, mmap_prot_from_open_flags(open_flags), &f->cache_fd);
4159 if (r < 0)
4160 goto fail;
4161
4162 if (newly_created) {
4163 (void) journal_file_warn_btrfs(f);
4164
4165 /* Let's attach the creation time to the journal file, so that the vacuuming code knows the age of this
4166 * file even if the file might end up corrupted one day... Ideally we'd just use the creation time many
4167 * file systems maintain for each file, but the API to query this is very new, hence let's emulate this
4168 * via extended attributes. If extended attributes are not supported we'll just skip this, and rely
4169 * solely on mtime/atime/ctime of the file. */
4170 (void) fd_setcrtime(f->fd, 0);
4171
4172 r = journal_file_init_header(f, file_flags, template);
4173 if (r < 0)
4174 goto fail;
4175
4176 r = journal_file_fstat(f);
4177 if (r < 0)
4178 goto fail;
4179 }
4180
4181 if (f->last_stat.st_size < (off_t) HEADER_SIZE_MIN) {
4182 r = -ENODATA;
4183 goto fail;
4184 }
4185
4186 r = mmap_cache_fd_get(f->cache_fd, MMAP_CACHE_CATEGORY_HEADER, true, 0, PAGE_ALIGN(sizeof(Header)), &f->last_stat, &h);
4187 if (r == -EINVAL) {
4188 /* Some file systems (jffs2 or p9fs) don't support mmap() properly (or only read-only
4189 * mmap()), and return EINVAL in that case. Let's propagate that as a more recognizable error
4190 * code. */
4191 r = -EAFNOSUPPORT;
4192 goto fail;
4193 }
4194 if (r < 0)
4195 goto fail;
4196
4197 f->header = h;
4198
4199 if (!newly_created) {
4200 r = journal_file_verify_header(f);
4201 if (r < 0)
4202 goto fail;
4203 }
4204
4205 #if HAVE_GCRYPT
4206 if (!newly_created && journal_file_writable(f) && JOURNAL_HEADER_SEALED(f->header)) {
4207 r = journal_file_fss_load(f);
4208 if (r < 0)
4209 goto fail;
4210 }
4211 #endif
4212
4213 if (journal_file_writable(f)) {
4214 if (metrics) {
4215 journal_default_metrics(metrics, f->fd, JOURNAL_HEADER_COMPACT(f->header));
4216 f->metrics = *metrics;
4217 } else if (template)
4218 f->metrics = template->metrics;
4219
4220 r = journal_file_refresh_header(f);
4221 if (r < 0)
4222 goto fail;
4223 }
4224
4225 #if HAVE_GCRYPT
4226 r = journal_file_hmac_setup(f);
4227 if (r < 0)
4228 goto fail;
4229 #endif
4230
4231 if (newly_created) {
4232 r = journal_file_setup_field_hash_table(f);
4233 if (r < 0)
4234 goto fail;
4235
4236 r = journal_file_setup_data_hash_table(f);
4237 if (r < 0)
4238 goto fail;
4239
4240 #if HAVE_GCRYPT
4241 r = journal_file_append_first_tag(f);
4242 if (r < 0)
4243 goto fail;
4244 #endif
4245 }
4246
4247 if (mmap_cache_fd_got_sigbus(f->cache_fd)) {
4248 r = -EIO;
4249 goto fail;
4250 }
4251
4252 if (template && template->post_change_timer) {
4253 r = journal_file_enable_post_change_timer(
4254 f,
4255 sd_event_source_get_event(template->post_change_timer),
4256 template->post_change_timer_period);
4257
4258 if (r < 0)
4259 goto fail;
4260 }
4261
4262 /* The file is opened now successfully, thus we take possession of any passed in fd. */
4263 f->close_fd = true;
4264
4265 if (DEBUG_LOGGING) {
4266 static int last_seal = -1, last_keyed_hash = -1;
4267 static Compression last_compression = _COMPRESSION_INVALID;
4268 static uint64_t last_bytes = UINT64_MAX;
4269
4270 if (last_seal != JOURNAL_HEADER_SEALED(f->header) ||
4271 last_keyed_hash != JOURNAL_HEADER_KEYED_HASH(f->header) ||
4272 last_compression != JOURNAL_FILE_COMPRESSION(f) ||
4273 last_bytes != f->compress_threshold_bytes) {
4274
4275 log_debug("Journal effective settings seal=%s keyed_hash=%s compress=%s compress_threshold_bytes=%s",
4276 yes_no(JOURNAL_HEADER_SEALED(f->header)), yes_no(JOURNAL_HEADER_KEYED_HASH(f->header)),
4277 compression_to_string(JOURNAL_FILE_COMPRESSION(f)), FORMAT_BYTES(f->compress_threshold_bytes));
4278 last_seal = JOURNAL_HEADER_SEALED(f->header);
4279 last_keyed_hash = JOURNAL_HEADER_KEYED_HASH(f->header);
4280 last_compression = JOURNAL_FILE_COMPRESSION(f);
4281 last_bytes = f->compress_threshold_bytes;
4282 }
4283 }
4284
4285 *ret = f;
4286 return 0;
4287
4288 fail:
4289 if (f->cache_fd && mmap_cache_fd_got_sigbus(f->cache_fd))
4290 r = -EIO;
4291
4292 (void) journal_file_close(f);
4293
4294 if (newly_created && fd < 0)
4295 (void) unlink(fname);
4296
4297 return r;
4298 }
4299
4300 int journal_file_parse_uid_from_filename(const char *path, uid_t *ret_uid) {
4301 _cleanup_free_ char *buf = NULL, *p = NULL;
4302 const char *a, *b, *at;
4303 int r;
4304
4305 /* This helper returns -EREMOTE when the filename doesn't match user online/offline journal
4306 * pattern. Hence it currently doesn't parse archived or disposed user journals. */
4307
4308 assert(path);
4309 assert(ret_uid);
4310
4311 r = path_extract_filename(path, &p);
4312 if (r < 0)
4313 return r;
4314 if (r == O_DIRECTORY)
4315 return -EISDIR;
4316
4317 a = startswith(p, "user-");
4318 if (!a)
4319 return -EREMOTE;
4320 b = endswith(p, ".journal");
4321 if (!b)
4322 return -EREMOTE;
4323
4324 at = strchr(a, '@');
4325 if (at)
4326 return -EREMOTE;
4327
4328 buf = strndup(a, b-a);
4329 if (!buf)
4330 return -ENOMEM;
4331
4332 return parse_uid(buf, ret_uid);
4333 }
4334
4335 int journal_file_archive(JournalFile *f, char **ret_previous_path) {
4336 _cleanup_free_ char *p = NULL;
4337
4338 assert(f);
4339
4340 if (!journal_file_writable(f))
4341 return -EINVAL;
4342
4343 /* Is this a journal file that was passed to us as fd? If so, we synthesized a path name for it, and we refuse
4344 * rotation, since we don't know the actual path, and couldn't rename the file hence. */
4345 if (path_startswith(f->path, "/proc/self/fd"))
4346 return -EINVAL;
4347
4348 if (!endswith(f->path, ".journal"))
4349 return -EINVAL;
4350
4351 if (asprintf(&p, "%.*s@" SD_ID128_FORMAT_STR "-%016"PRIx64"-%016"PRIx64".journal",
4352 (int) strlen(f->path) - 8, f->path,
4353 SD_ID128_FORMAT_VAL(f->header->seqnum_id),
4354 le64toh(f->header->head_entry_seqnum),
4355 le64toh(f->header->head_entry_realtime)) < 0)
4356 return -ENOMEM;
4357
4358 /* Try to rename the file to the archived version. If the file already was deleted, we'll get ENOENT, let's
4359 * ignore that case. */
4360 if (rename(f->path, p) < 0 && errno != ENOENT)
4361 return -errno;
4362
4363 /* Sync the rename to disk */
4364 (void) fsync_directory_of_file(f->fd);
4365
4366 if (ret_previous_path)
4367 *ret_previous_path = f->path;
4368 else
4369 free(f->path);
4370
4371 f->path = TAKE_PTR(p);
4372
4373 /* Set as archive so offlining commits w/state=STATE_ARCHIVED. Previously we would set old_file->header->state
4374 * to STATE_ARCHIVED directly here, but journal_file_set_offline() short-circuits when state != STATE_ONLINE,
4375 * which would result in the rotated journal never getting fsync() called before closing. Now we simply queue
4376 * the archive state by setting an archive bit, leaving the state as STATE_ONLINE so proper offlining
4377 * occurs. */
4378 f->archive = true;
4379
4380 return 0;
4381 }
4382
4383 int journal_file_dispose(int dir_fd, const char *fname) {
4384 _cleanup_free_ char *p = NULL;
4385
4386 assert(fname);
4387
4388 /* Renames a journal file to *.journal~, i.e. to mark it as corrupted or otherwise uncleanly shutdown. Note that
4389 * this is done without looking into the file or changing any of its contents. The idea is that this is called
4390 * whenever something is suspicious and we want to move the file away and make clear that it is not accessed
4391 * for writing anymore. */
4392
4393 if (!endswith(fname, ".journal"))
4394 return -EINVAL;
4395
4396 if (asprintf(&p, "%.*s@%016" PRIx64 "-%016" PRIx64 ".journal~",
4397 (int) strlen(fname) - 8, fname,
4398 now(CLOCK_REALTIME),
4399 random_u64()) < 0)
4400 return -ENOMEM;
4401
4402 if (renameat(dir_fd, fname, dir_fd, p) < 0)
4403 return -errno;
4404
4405 return 0;
4406 }
4407
4408 int journal_file_copy_entry(
4409 JournalFile *from,
4410 JournalFile *to,
4411 Object *o,
4412 uint64_t p,
4413 uint64_t *seqnum,
4414 sd_id128_t *seqnum_id) {
4415
4416 _cleanup_free_ EntryItem *items_alloc = NULL;
4417 EntryItem *items;
4418 uint64_t n, m = 0, xor_hash = 0;
4419 sd_id128_t boot_id;
4420 dual_timestamp ts;
4421 int r;
4422
4423 assert(from);
4424 assert(to);
4425 assert(o);
4426 assert(p > 0);
4427
4428 if (!journal_file_writable(to))
4429 return -EPERM;
4430
4431 ts = (dual_timestamp) {
4432 .monotonic = le64toh(o->entry.monotonic),
4433 .realtime = le64toh(o->entry.realtime),
4434 };
4435 boot_id = o->entry.boot_id;
4436
4437 n = journal_file_entry_n_items(from, o);
4438 if (n == 0)
4439 return 0;
4440
4441 if (n < ALLOCA_MAX / sizeof(EntryItem) / 2)
4442 items = newa(EntryItem, n);
4443 else {
4444 items_alloc = new(EntryItem, n);
4445 if (!items_alloc)
4446 return -ENOMEM;
4447
4448 items = items_alloc;
4449 }
4450
4451 for (uint64_t i = 0; i < n; i++) {
4452 uint64_t h, q;
4453 void *data;
4454 size_t l;
4455 Object *u;
4456
4457 q = journal_file_entry_item_object_offset(from, o, i);
4458 r = journal_file_data_payload(from, NULL, q, NULL, 0, 0, &data, &l);
4459 if (IN_SET(r, -EADDRNOTAVAIL, -EBADMSG)) {
4460 log_debug_errno(r, "Entry item %"PRIu64" data object is bad, skipping over it: %m", i);
4461 continue;
4462 }
4463 if (r < 0)
4464 return r;
4465 assert(r > 0);
4466
4467 if (l == 0)
4468 return -EBADMSG;
4469
4470 r = journal_file_append_data(to, data, l, &u, &h);
4471 if (r < 0)
4472 return r;
4473
4474 if (JOURNAL_HEADER_KEYED_HASH(to->header))
4475 xor_hash ^= jenkins_hash64(data, l);
4476 else
4477 xor_hash ^= le64toh(u->data.hash);
4478
4479 items[m++] = (EntryItem) {
4480 .object_offset = h,
4481 .hash = le64toh(u->data.hash),
4482 };
4483 }
4484
4485 if (m == 0)
4486 return 0;
4487
4488 r = journal_file_append_entry_internal(
4489 to,
4490 &ts,
4491 &boot_id,
4492 &from->header->machine_id,
4493 xor_hash,
4494 items,
4495 m,
4496 seqnum,
4497 seqnum_id,
4498 /* ret_object= */ NULL,
4499 /* ret_offset= */ NULL);
4500
4501 if (mmap_cache_fd_got_sigbus(to->cache_fd))
4502 return -EIO;
4503
4504 return r;
4505 }
4506
4507 void journal_reset_metrics(JournalMetrics *m) {
4508 assert(m);
4509
4510 /* Set everything to "pick automatic values". */
4511
4512 *m = (JournalMetrics) {
4513 .min_use = UINT64_MAX,
4514 .max_use = UINT64_MAX,
4515 .min_size = UINT64_MAX,
4516 .max_size = UINT64_MAX,
4517 .keep_free = UINT64_MAX,
4518 .n_max_files = UINT64_MAX,
4519 };
4520 }
4521
4522 int journal_file_get_cutoff_realtime_usec(JournalFile *f, usec_t *ret_from, usec_t *ret_to) {
4523 assert(f);
4524 assert(f->header);
4525 assert(ret_from || ret_to);
4526
4527 if (ret_from) {
4528 if (f->header->head_entry_realtime == 0)
4529 return -ENOENT;
4530
4531 *ret_from = le64toh(f->header->head_entry_realtime);
4532 }
4533
4534 if (ret_to) {
4535 if (f->header->tail_entry_realtime == 0)
4536 return -ENOENT;
4537
4538 *ret_to = le64toh(f->header->tail_entry_realtime);
4539 }
4540
4541 return 1;
4542 }
4543
4544 int journal_file_get_cutoff_monotonic_usec(JournalFile *f, sd_id128_t boot_id, usec_t *ret_from, usec_t *ret_to) {
4545 Object *o;
4546 uint64_t p;
4547 int r;
4548
4549 assert(f);
4550 assert(ret_from || ret_to);
4551
4552 /* FIXME: fix return value assignment on success with 0. */
4553
4554 r = find_data_object_by_boot_id(f, boot_id, &o, &p);
4555 if (r <= 0)
4556 return r;
4557
4558 if (le64toh(o->data.n_entries) <= 0)
4559 return 0;
4560
4561 if (ret_from) {
4562 r = journal_file_move_to_object(f, OBJECT_ENTRY, le64toh(o->data.entry_offset), &o);
4563 if (r < 0)
4564 return r;
4565
4566 *ret_from = le64toh(o->entry.monotonic);
4567 }
4568
4569 if (ret_to) {
4570 r = journal_file_move_to_object(f, OBJECT_DATA, p, &o);
4571 if (r < 0)
4572 return r;
4573
4574 r = journal_file_move_to_entry_for_data(f, o, DIRECTION_UP, &o, NULL);
4575 if (r <= 0)
4576 return r;
4577
4578 *ret_to = le64toh(o->entry.monotonic);
4579 }
4580
4581 return 1;
4582 }
4583
4584 bool journal_file_rotate_suggested(JournalFile *f, usec_t max_file_usec, int log_level) {
4585 assert(f);
4586 assert(f->header);
4587
4588 /* If we gained new header fields we gained new features,
4589 * hence suggest a rotation */
4590 if (le64toh(f->header->header_size) < sizeof(Header)) {
4591 log_ratelimit_full(log_level, JOURNAL_LOG_RATELIMIT,
4592 "%s uses an outdated header, suggesting rotation.", f->path);
4593 return true;
4594 }
4595
4596 /* Let's check if the hash tables grew over a certain fill level (75%, borrowing this value from
4597 * Java's hash table implementation), and if so suggest a rotation. To calculate the fill level we
4598 * need the n_data field, which only exists in newer versions. */
4599
4600 if (JOURNAL_HEADER_CONTAINS(f->header, n_data))
4601 if (le64toh(f->header->n_data) * 4ULL > (le64toh(f->header->data_hash_table_size) / sizeof(HashItem)) * 3ULL) {
4602 log_ratelimit_full(
4603 log_level, JOURNAL_LOG_RATELIMIT,
4604 "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.",
4605 f->path,
4606 100.0 * (double) le64toh(f->header->n_data) / ((double) (le64toh(f->header->data_hash_table_size) / sizeof(HashItem))),
4607 le64toh(f->header->n_data),
4608 le64toh(f->header->data_hash_table_size) / sizeof(HashItem),
4609 (uint64_t) f->last_stat.st_size,
4610 f->last_stat.st_size / le64toh(f->header->n_data));
4611 return true;
4612 }
4613
4614 if (JOURNAL_HEADER_CONTAINS(f->header, n_fields))
4615 if (le64toh(f->header->n_fields) * 4ULL > (le64toh(f->header->field_hash_table_size) / sizeof(HashItem)) * 3ULL) {
4616 log_ratelimit_full(
4617 log_level, JOURNAL_LOG_RATELIMIT,
4618 "Field hash table of %s has a fill level at %.1f (%"PRIu64" of %"PRIu64" items), suggesting rotation.",
4619 f->path,
4620 100.0 * (double) le64toh(f->header->n_fields) / ((double) (le64toh(f->header->field_hash_table_size) / sizeof(HashItem))),
4621 le64toh(f->header->n_fields),
4622 le64toh(f->header->field_hash_table_size) / sizeof(HashItem));
4623 return true;
4624 }
4625
4626 /* If there are too many hash collisions somebody is most likely playing games with us. Hence, if our
4627 * longest chain is longer than some threshold, let's suggest rotation. */
4628 if (JOURNAL_HEADER_CONTAINS(f->header, data_hash_chain_depth) &&
4629 le64toh(f->header->data_hash_chain_depth) > HASH_CHAIN_DEPTH_MAX) {
4630 log_ratelimit_full(
4631 log_level, JOURNAL_LOG_RATELIMIT,
4632 "Data hash table of %s has deepest hash chain of length %" PRIu64 ", suggesting rotation.",
4633 f->path, le64toh(f->header->data_hash_chain_depth));
4634 return true;
4635 }
4636
4637 if (JOURNAL_HEADER_CONTAINS(f->header, field_hash_chain_depth) &&
4638 le64toh(f->header->field_hash_chain_depth) > HASH_CHAIN_DEPTH_MAX) {
4639 log_ratelimit_full(
4640 log_level, JOURNAL_LOG_RATELIMIT,
4641 "Field hash table of %s has deepest hash chain of length at %" PRIu64 ", suggesting rotation.",
4642 f->path, le64toh(f->header->field_hash_chain_depth));
4643 return true;
4644 }
4645
4646 /* Are the data objects properly indexed by field objects? */
4647 if (JOURNAL_HEADER_CONTAINS(f->header, n_data) &&
4648 JOURNAL_HEADER_CONTAINS(f->header, n_fields) &&
4649 le64toh(f->header->n_data) > 0 &&
4650 le64toh(f->header->n_fields) == 0) {
4651 log_ratelimit_full(
4652 log_level, JOURNAL_LOG_RATELIMIT,
4653 "Data objects of %s are not indexed by field objects, suggesting rotation.",
4654 f->path);
4655 return true;
4656 }
4657
4658 if (max_file_usec > 0) {
4659 usec_t t, h;
4660
4661 h = le64toh(f->header->head_entry_realtime);
4662 t = now(CLOCK_REALTIME);
4663
4664 if (h > 0 && t > h + max_file_usec) {
4665 log_ratelimit_full(
4666 log_level, JOURNAL_LOG_RATELIMIT,
4667 "Oldest entry in %s is older than the configured file retention duration (%s), suggesting rotation.",
4668 f->path, FORMAT_TIMESPAN(max_file_usec, USEC_PER_SEC));
4669 return true;
4670 }
4671 }
4672
4673 return false;
4674 }
4675
4676 static const char * const journal_object_type_table[] = {
4677 [OBJECT_UNUSED] = "unused",
4678 [OBJECT_DATA] = "data",
4679 [OBJECT_FIELD] = "field",
4680 [OBJECT_ENTRY] = "entry",
4681 [OBJECT_DATA_HASH_TABLE] = "data hash table",
4682 [OBJECT_FIELD_HASH_TABLE] = "field hash table",
4683 [OBJECT_ENTRY_ARRAY] = "entry array",
4684 [OBJECT_TAG] = "tag",
4685 };
4686
4687 DEFINE_STRING_TABLE_LOOKUP_TO_STRING(journal_object_type, ObjectType);
Unnamed repository; edit this file 'description' to name the repository.