]> git.ipfire.org Git - thirdparty/systemd.git/blob - src/journal/journal-file.c
journal: increase compression threshold for objects from 64 to 512
[thirdparty/systemd.git] / src / journal / journal-file.c
1 /*-*- Mode: C; c-basic-offset: 8; indent-tabs-mode: nil -*-*/
2
3 /***
4 This file is part of systemd.
5
6 Copyright 2011 Lennart Poettering
7
8 systemd is free software; you can redistribute it and/or modify it
9 under the terms of the GNU General Public License as published by
10 the Free Software Foundation; either version 2 of the License, or
11 (at your option) any later version.
12
13 systemd is distributed in the hope that it will be useful, but
14 WITHOUT ANY WARRANTY; without even the implied warranty of
15 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
16 General Public License for more details.
17
18 You should have received a copy of the GNU General Public License
19 along with systemd; If not, see <http://www.gnu.org/licenses/>.
20 ***/
21
22 #include <sys/mman.h>
23 #include <errno.h>
24 #include <sys/uio.h>
25 #include <unistd.h>
26 #include <sys/statvfs.h>
27 #include <fcntl.h>
28 #include <stddef.h>
29
30 #include "journal-def.h"
31 #include "journal-file.h"
32 #include "lookup3.h"
33 #include "compress.h"
34
35 #define DEFAULT_DATA_HASH_TABLE_SIZE (2047ULL*16ULL)
36 #define DEFAULT_FIELD_HASH_TABLE_SIZE (2047ULL*16ULL)
37
38 #define DEFAULT_WINDOW_SIZE (128ULL*1024ULL*1024ULL)
39
40 #define COMPRESSION_SIZE_THRESHOLD (512ULL)
41
42 /* This is the minimum journal file size */
43 #define JOURNAL_FILE_SIZE_MIN (64ULL*1024ULL)
44
45 /* These are the lower and upper bounds if we deduce the max_use value
46 * from the file system size */
47 #define DEFAULT_MAX_USE_LOWER (1ULL*1024ULL*1024ULL) /* 1 MiB */
48 #define DEFAULT_MAX_USE_UPPER (4ULL*1024ULL*1024ULL*1024ULL) /* 4 GiB */
49
50 /* This is the upper bound if we deduce max_size from max_use */
51 #define DEFAULT_MAX_SIZE_UPPER (16ULL*1024ULL*1024ULL) /* 16 MiB */
52
53 /* This is the upper bound if we deduce the keep_free value from the
54 * file system size */
55 #define DEFAULT_KEEP_FREE_UPPER (4ULL*1024ULL*1024ULL*1024ULL) /* 4 GiB */
56
57 /* This is the keep_free value when we can't determine the system
58 * size */
59 #define DEFAULT_KEEP_FREE (1024ULL*1024ULL) /* 1 MB */
60
61 static const char signature[] = { 'L', 'P', 'K', 'S', 'H', 'H', 'R', 'H' };
62
63 #define ALIGN64(x) (((x) + 7ULL) & ~7ULL)
64
65 void journal_file_close(JournalFile *f) {
66 int t;
67
68 assert(f);
69
70 if (f->header && f->writable)
71 f->header->state = STATE_OFFLINE;
72
73
74 for (t = 0; t < _WINDOW_MAX; t++)
75 if (f->windows[t].ptr)
76 munmap(f->windows[t].ptr, f->windows[t].size);
77
78 if (f->fd >= 0)
79 close_nointr_nofail(f->fd);
80
81 free(f->path);
82
83 #ifdef HAVE_XZ
84 free(f->compress_buffer);
85 #endif
86
87 free(f);
88 }
89
90 static int journal_file_init_header(JournalFile *f, JournalFile *template) {
91 Header h;
92 ssize_t k;
93 int r;
94
95 assert(f);
96
97 zero(h);
98 memcpy(h.signature, signature, 8);
99 h.arena_offset = htole64(ALIGN64(sizeof(h)));
100
101 r = sd_id128_randomize(&h.file_id);
102 if (r < 0)
103 return r;
104
105 if (template) {
106 h.seqnum_id = template->header->seqnum_id;
107 h.seqnum = template->header->seqnum;
108 } else
109 h.seqnum_id = h.file_id;
110
111 k = pwrite(f->fd, &h, sizeof(h), 0);
112 if (k < 0)
113 return -errno;
114
115 if (k != sizeof(h))
116 return -EIO;
117
118 return 0;
119 }
120
121 static int journal_file_refresh_header(JournalFile *f) {
122 int r;
123 sd_id128_t boot_id;
124
125 assert(f);
126
127 r = sd_id128_get_machine(&f->header->machine_id);
128 if (r < 0)
129 return r;
130
131 r = sd_id128_get_boot(&boot_id);
132 if (r < 0)
133 return r;
134
135 if (sd_id128_equal(boot_id, f->header->boot_id))
136 f->tail_entry_monotonic_valid = true;
137
138 f->header->boot_id = boot_id;
139
140 f->header->state = STATE_ONLINE;
141
142 __sync_synchronize();
143
144 return 0;
145 }
146
147 static int journal_file_verify_header(JournalFile *f) {
148 assert(f);
149
150 if (memcmp(f->header, signature, 8))
151 return -EBADMSG;
152
153 #ifdef HAVE_XZ
154 if ((le64toh(f->header->incompatible_flags) & ~HEADER_INCOMPATIBLE_COMPRESSED) != 0)
155 return -EPROTONOSUPPORT;
156 #else
157 if (f->header->incompatible_flags != 0)
158 return -EPROTONOSUPPORT;
159 #endif
160
161 if ((uint64_t) f->last_stat.st_size < (le64toh(f->header->arena_offset) + le64toh(f->header->arena_size)))
162 return -ENODATA;
163
164 if (f->writable) {
165 uint32_t state;
166 sd_id128_t machine_id;
167 int r;
168
169 r = sd_id128_get_machine(&machine_id);
170 if (r < 0)
171 return r;
172
173 if (!sd_id128_equal(machine_id, f->header->machine_id))
174 return -EHOSTDOWN;
175
176 state = f->header->state;
177
178 if (state == STATE_ONLINE)
179 log_debug("Journal file %s is already online. Assuming unclean closing. Ignoring.", f->path);
180 else if (state == STATE_ARCHIVED)
181 return -ESHUTDOWN;
182 else if (state != STATE_OFFLINE)
183 log_debug("Journal file %s has unknown state %u. Ignoring.", f->path, state);
184 }
185
186 return 0;
187 }
188
189 static int journal_file_allocate(JournalFile *f, uint64_t offset, uint64_t size) {
190 uint64_t old_size, new_size;
191
192 assert(f);
193
194 /* We assume that this file is not sparse, and we know that
195 * for sure, since we always call posix_fallocate()
196 * ourselves */
197
198 old_size =
199 le64toh(f->header->arena_offset) +
200 le64toh(f->header->arena_size);
201
202 new_size = PAGE_ALIGN(offset + size);
203 if (new_size < le64toh(f->header->arena_offset))
204 new_size = le64toh(f->header->arena_offset);
205
206 if (new_size <= old_size)
207 return 0;
208
209 if (f->metrics.max_size > 0 &&
210 new_size > f->metrics.max_size)
211 return -E2BIG;
212
213 if (new_size > f->metrics.min_size &&
214 f->metrics.keep_free > 0) {
215 struct statvfs svfs;
216
217 if (fstatvfs(f->fd, &svfs) >= 0) {
218 uint64_t available;
219
220 available = svfs.f_bfree * svfs.f_bsize;
221
222 if (available >= f->metrics.keep_free)
223 available -= f->metrics.keep_free;
224 else
225 available = 0;
226
227 if (new_size - old_size > available)
228 return -E2BIG;
229 }
230 }
231
232 /* Note that the glibc fallocate() fallback is very
233 inefficient, hence we try to minimize the allocation area
234 as we can. */
235 if (posix_fallocate(f->fd, old_size, new_size - old_size) < 0)
236 return -errno;
237
238 if (fstat(f->fd, &f->last_stat) < 0)
239 return -errno;
240
241 f->header->arena_size = new_size - htole64(f->header->arena_offset);
242
243 return 0;
244 }
245
246 static int journal_file_map(
247 JournalFile *f,
248 uint64_t offset,
249 uint64_t size,
250 void **_window,
251 uint64_t *_woffset,
252 uint64_t *_wsize,
253 void **ret) {
254
255 uint64_t woffset, wsize;
256 void *window;
257
258 assert(f);
259 assert(size > 0);
260 assert(ret);
261
262 woffset = offset & ~((uint64_t) page_size() - 1ULL);
263 wsize = size + (offset - woffset);
264 wsize = PAGE_ALIGN(wsize);
265
266 /* Avoid SIGBUS on invalid accesses */
267 if (woffset + wsize > (uint64_t) PAGE_ALIGN(f->last_stat.st_size))
268 return -EADDRNOTAVAIL;
269
270 window = mmap(NULL, wsize, f->prot, MAP_SHARED, f->fd, woffset);
271 if (window == MAP_FAILED)
272 return -errno;
273
274 if (_window)
275 *_window = window;
276
277 if (_woffset)
278 *_woffset = woffset;
279
280 if (_wsize)
281 *_wsize = wsize;
282
283 *ret = (uint8_t*) window + (offset - woffset);
284
285 return 0;
286 }
287
288 static int journal_file_move_to(JournalFile *f, int wt, uint64_t offset, uint64_t size, void **ret) {
289 void *p = NULL;
290 uint64_t delta;
291 int r;
292 Window *w;
293
294 assert(f);
295 assert(ret);
296 assert(wt >= 0);
297 assert(wt < _WINDOW_MAX);
298
299 if (offset + size > (uint64_t) f->last_stat.st_size) {
300 /* Hmm, out of range? Let's refresh the fstat() data
301 * first, before we trust that check. */
302
303 if (fstat(f->fd, &f->last_stat) < 0 ||
304 offset + size > (uint64_t) f->last_stat.st_size)
305 return -EADDRNOTAVAIL;
306 }
307
308 w = f->windows + wt;
309
310 if (_likely_(w->ptr &&
311 w->offset <= offset &&
312 w->offset + w->size >= offset + size)) {
313
314 *ret = (uint8_t*) w->ptr + (offset - w->offset);
315 return 0;
316 }
317
318 if (w->ptr) {
319 if (munmap(w->ptr, w->size) < 0)
320 return -errno;
321
322 w->ptr = NULL;
323 w->size = w->offset = 0;
324 }
325
326 if (size < DEFAULT_WINDOW_SIZE) {
327 /* If the default window size is larger then what was
328 * asked for extend the mapping a bit in the hope to
329 * minimize needed remappings later on. We add half
330 * the window space before and half behind the
331 * requested mapping */
332
333 delta = (DEFAULT_WINDOW_SIZE - size) / 2;
334
335 if (delta > offset)
336 delta = offset;
337
338 offset -= delta;
339 size = DEFAULT_WINDOW_SIZE;
340 } else
341 delta = 0;
342
343 if (offset + size > (uint64_t) f->last_stat.st_size)
344 size = (uint64_t) f->last_stat.st_size - offset;
345
346 if (size <= 0)
347 return -EADDRNOTAVAIL;
348
349 r = journal_file_map(f,
350 offset, size,
351 &w->ptr, &w->offset, &w->size,
352 &p);
353
354 if (r < 0)
355 return r;
356
357 *ret = (uint8_t*) p + delta;
358 return 0;
359 }
360
361 static bool verify_hash(Object *o) {
362 uint64_t h1, h2;
363
364 assert(o);
365
366 if (o->object.type == OBJECT_DATA && !(o->object.flags & OBJECT_COMPRESSED)) {
367 h1 = le64toh(o->data.hash);
368 h2 = hash64(o->data.payload, le64toh(o->object.size) - offsetof(Object, data.payload));
369 } else if (o->object.type == OBJECT_FIELD) {
370 h1 = le64toh(o->field.hash);
371 h2 = hash64(o->field.payload, le64toh(o->object.size) - offsetof(Object, field.payload));
372 } else
373 return true;
374
375 return h1 == h2;
376 }
377
378 int journal_file_move_to_object(JournalFile *f, int type, uint64_t offset, Object **ret) {
379 int r;
380 void *t;
381 Object *o;
382 uint64_t s;
383
384 assert(f);
385 assert(ret);
386 assert(type < _OBJECT_TYPE_MAX);
387
388 r = journal_file_move_to(f, type >= 0 ? type : WINDOW_UNKNOWN, offset, sizeof(ObjectHeader), &t);
389 if (r < 0)
390 return r;
391
392 o = (Object*) t;
393 s = le64toh(o->object.size);
394
395 if (s < sizeof(ObjectHeader))
396 return -EBADMSG;
397
398 if (type >= 0 && o->object.type != type)
399 return -EBADMSG;
400
401 if (s > sizeof(ObjectHeader)) {
402 r = journal_file_move_to(f, o->object.type, offset, s, &t);
403 if (r < 0)
404 return r;
405
406 o = (Object*) t;
407 }
408
409 if (!verify_hash(o))
410 return -EBADMSG;
411
412 *ret = o;
413 return 0;
414 }
415
416 static uint64_t journal_file_seqnum(JournalFile *f, uint64_t *seqnum) {
417 uint64_t r;
418
419 assert(f);
420
421 r = le64toh(f->header->seqnum) + 1;
422
423 if (seqnum) {
424 /* If an external seqnum counter was passed, we update
425 * both the local and the external one, and set it to
426 * the maximum of both */
427
428 if (*seqnum + 1 > r)
429 r = *seqnum + 1;
430
431 *seqnum = r;
432 }
433
434 f->header->seqnum = htole64(r);
435
436 if (f->header->first_seqnum == 0)
437 f->header->first_seqnum = htole64(r);
438
439 return r;
440 }
441
442 static int journal_file_append_object(JournalFile *f, int type, uint64_t size, Object **ret, uint64_t *offset) {
443 int r;
444 uint64_t p;
445 Object *tail, *o;
446 void *t;
447
448 assert(f);
449 assert(size >= sizeof(ObjectHeader));
450 assert(offset);
451 assert(ret);
452
453 p = le64toh(f->header->tail_object_offset);
454 if (p == 0)
455 p = le64toh(f->header->arena_offset);
456 else {
457 r = journal_file_move_to_object(f, -1, p, &tail);
458 if (r < 0)
459 return r;
460
461 p += ALIGN64(le64toh(tail->object.size));
462 }
463
464 r = journal_file_allocate(f, p, size);
465 if (r < 0)
466 return r;
467
468 r = journal_file_move_to(f, type, p, size, &t);
469 if (r < 0)
470 return r;
471
472 o = (Object*) t;
473
474 zero(o->object);
475 o->object.type = type;
476 o->object.size = htole64(size);
477
478 f->header->tail_object_offset = htole64(p);
479 f->header->n_objects = htole64(le64toh(f->header->n_objects) + 1);
480
481 *ret = o;
482 *offset = p;
483
484 return 0;
485 }
486
487 static int journal_file_setup_data_hash_table(JournalFile *f) {
488 uint64_t s, p;
489 Object *o;
490 int r;
491
492 assert(f);
493
494 s = DEFAULT_DATA_HASH_TABLE_SIZE;
495 r = journal_file_append_object(f,
496 OBJECT_DATA_HASH_TABLE,
497 offsetof(Object, hash_table.items) + s,
498 &o, &p);
499 if (r < 0)
500 return r;
501
502 memset(o->hash_table.items, 0, s);
503
504 f->header->data_hash_table_offset = htole64(p + offsetof(Object, hash_table.items));
505 f->header->data_hash_table_size = htole64(s);
506
507 return 0;
508 }
509
510 static int journal_file_setup_field_hash_table(JournalFile *f) {
511 uint64_t s, p;
512 Object *o;
513 int r;
514
515 assert(f);
516
517 s = DEFAULT_FIELD_HASH_TABLE_SIZE;
518 r = journal_file_append_object(f,
519 OBJECT_FIELD_HASH_TABLE,
520 offsetof(Object, hash_table.items) + s,
521 &o, &p);
522 if (r < 0)
523 return r;
524
525 memset(o->hash_table.items, 0, s);
526
527 f->header->field_hash_table_offset = htole64(p + offsetof(Object, hash_table.items));
528 f->header->field_hash_table_size = htole64(s);
529
530 return 0;
531 }
532
533 static int journal_file_map_data_hash_table(JournalFile *f) {
534 uint64_t s, p;
535 void *t;
536 int r;
537
538 assert(f);
539
540 p = le64toh(f->header->data_hash_table_offset);
541 s = le64toh(f->header->data_hash_table_size);
542
543 r = journal_file_move_to(f,
544 WINDOW_DATA_HASH_TABLE,
545 p, s,
546 &t);
547 if (r < 0)
548 return r;
549
550 f->data_hash_table = t;
551 return 0;
552 }
553
554 static int journal_file_map_field_hash_table(JournalFile *f) {
555 uint64_t s, p;
556 void *t;
557 int r;
558
559 assert(f);
560
561 p = le64toh(f->header->field_hash_table_offset);
562 s = le64toh(f->header->field_hash_table_size);
563
564 r = journal_file_move_to(f,
565 WINDOW_FIELD_HASH_TABLE,
566 p, s,
567 &t);
568 if (r < 0)
569 return r;
570
571 f->field_hash_table = t;
572 return 0;
573 }
574
575 static int journal_file_link_data(JournalFile *f, Object *o, uint64_t offset, uint64_t hash) {
576 uint64_t p, h;
577 int r;
578
579 assert(f);
580 assert(o);
581 assert(offset > 0);
582 assert(o->object.type == OBJECT_DATA);
583
584 o->data.next_hash_offset = o->data.next_field_offset = 0;
585 o->data.entry_offset = o->data.entry_array_offset = 0;
586 o->data.n_entries = 0;
587
588 h = hash % (le64toh(f->header->data_hash_table_size) / sizeof(HashItem));
589 p = le64toh(f->data_hash_table[h].head_hash_offset);
590 if (p == 0) {
591 /* Only entry in the hash table is easy */
592 f->data_hash_table[h].head_hash_offset = htole64(offset);
593 } else {
594 /* Temporarily move back to the previous data object,
595 * to patch in pointer */
596
597 r = journal_file_move_to_object(f, OBJECT_DATA, p, &o);
598 if (r < 0)
599 return r;
600
601 o->data.next_hash_offset = htole64(offset);
602
603 r = journal_file_move_to_object(f, OBJECT_DATA, offset, &o);
604 if (r < 0)
605 return r;
606 }
607
608 f->data_hash_table[h].tail_hash_offset = htole64(offset);
609
610 return 0;
611 }
612
613 int journal_file_find_data_object_with_hash(
614 JournalFile *f,
615 const void *data, uint64_t size, uint64_t hash,
616 Object **ret, uint64_t *offset) {
617 uint64_t p, osize, h;
618 int r;
619
620 assert(f);
621 assert(data || size == 0);
622
623 osize = offsetof(Object, data.payload) + size;
624
625 if (f->header->data_hash_table_size == 0)
626 return -EBADMSG;
627
628 h = hash % (le64toh(f->header->data_hash_table_size) / sizeof(HashItem));
629 p = le64toh(f->data_hash_table[h].head_hash_offset);
630
631 while (p > 0) {
632 Object *o;
633
634 r = journal_file_move_to_object(f, OBJECT_DATA, p, &o);
635 if (r < 0)
636 return r;
637
638 if (le64toh(o->data.hash) != hash)
639 goto next;
640
641 if (o->object.flags & OBJECT_COMPRESSED) {
642 #ifdef HAVE_XZ
643 uint64_t l, rsize;
644
645 l = le64toh(o->object.size);
646 if (l <= offsetof(Object, data.payload))
647 return -EBADMSG;
648
649 l -= offsetof(Object, data.payload);
650
651 if (!uncompress_blob(o->data.payload, l, &f->compress_buffer, &f->compress_buffer_size, &rsize))
652 return -EBADMSG;
653
654 if (rsize == size &&
655 memcmp(f->compress_buffer, data, size) == 0) {
656
657 if (ret)
658 *ret = o;
659
660 if (offset)
661 *offset = p;
662
663 return 1;
664 }
665 #else
666 return -EPROTONOSUPPORT;
667 #endif
668
669 } else if (le64toh(o->object.size) == osize &&
670 memcmp(o->data.payload, data, size) == 0) {
671
672 if (ret)
673 *ret = o;
674
675 if (offset)
676 *offset = p;
677
678 return 1;
679 }
680
681 next:
682 p = le64toh(o->data.next_hash_offset);
683 }
684
685 return 0;
686 }
687
688 int journal_file_find_data_object(
689 JournalFile *f,
690 const void *data, uint64_t size,
691 Object **ret, uint64_t *offset) {
692
693 uint64_t hash;
694
695 assert(f);
696 assert(data || size == 0);
697
698 hash = hash64(data, size);
699
700 return journal_file_find_data_object_with_hash(f,
701 data, size, hash,
702 ret, offset);
703 }
704
705 static int journal_file_append_data(JournalFile *f, const void *data, uint64_t size, Object **ret, uint64_t *offset) {
706 uint64_t hash, p;
707 uint64_t osize;
708 Object *o;
709 int r;
710 bool compressed = false;
711
712 assert(f);
713 assert(data || size == 0);
714
715 hash = hash64(data, size);
716
717 r = journal_file_find_data_object_with_hash(f, data, size, hash, &o, &p);
718 if (r < 0)
719 return r;
720 else if (r > 0) {
721
722 if (ret)
723 *ret = o;
724
725 if (offset)
726 *offset = p;
727
728 return 0;
729 }
730
731 osize = offsetof(Object, data.payload) + size;
732 r = journal_file_append_object(f, OBJECT_DATA, osize, &o, &p);
733 if (r < 0)
734 return r;
735
736 o->data.hash = htole64(hash);
737
738 #ifdef HAVE_XZ
739 if (f->compress &&
740 size >= COMPRESSION_SIZE_THRESHOLD) {
741 uint64_t rsize;
742
743 compressed = compress_blob(data, size, o->data.payload, &rsize);
744
745 if (compressed) {
746 o->object.size = htole64(offsetof(Object, data.payload) + rsize);
747 o->object.flags |= OBJECT_COMPRESSED;
748
749 f->header->incompatible_flags = htole32(le32toh(f->header->incompatible_flags) | HEADER_INCOMPATIBLE_COMPRESSED);
750
751 log_debug("Compressed data object %lu -> %lu", (unsigned long) size, (unsigned long) rsize);
752 }
753 }
754 #endif
755
756 if (!compressed)
757 memcpy(o->data.payload, data, size);
758
759 r = journal_file_link_data(f, o, p, hash);
760 if (r < 0)
761 return r;
762
763 if (ret)
764 *ret = o;
765
766 if (offset)
767 *offset = p;
768
769 return 0;
770 }
771
772 uint64_t journal_file_entry_n_items(Object *o) {
773 assert(o);
774 assert(o->object.type == htole64(OBJECT_ENTRY));
775
776 return (le64toh(o->object.size) - offsetof(Object, entry.items)) / sizeof(EntryItem);
777 }
778
779 static uint64_t journal_file_entry_array_n_items(Object *o) {
780 assert(o);
781 assert(o->object.type == htole64(OBJECT_ENTRY_ARRAY));
782
783 return (le64toh(o->object.size) - offsetof(Object, entry_array.items)) / sizeof(uint64_t);
784 }
785
786 static int link_entry_into_array(JournalFile *f,
787 uint64_t *first,
788 uint64_t *idx,
789 uint64_t p) {
790 int r;
791 uint64_t n = 0, ap = 0, q, i, a, hidx;
792 Object *o;
793
794 assert(f);
795 assert(first);
796 assert(idx);
797 assert(p > 0);
798
799 a = le64toh(*first);
800 i = hidx = le64toh(*idx);
801 while (a > 0) {
802
803 r = journal_file_move_to_object(f, OBJECT_ENTRY_ARRAY, a, &o);
804 if (r < 0)
805 return r;
806
807 n = journal_file_entry_array_n_items(o);
808 if (i < n) {
809 o->entry_array.items[i] = htole64(p);
810 *idx = htole64(hidx + 1);
811 return 0;
812 }
813
814 i -= n;
815 ap = a;
816 a = le64toh(o->entry_array.next_entry_array_offset);
817 }
818
819 if (hidx > n)
820 n = (hidx+1) * 2;
821 else
822 n = n * 2;
823
824 if (n < 4)
825 n = 4;
826
827 r = journal_file_append_object(f, OBJECT_ENTRY_ARRAY,
828 offsetof(Object, entry_array.items) + n * sizeof(uint64_t),
829 &o, &q);
830 if (r < 0)
831 return r;
832
833 o->entry_array.items[i] = htole64(p);
834
835 if (ap == 0)
836 *first = q;
837 else {
838 r = journal_file_move_to_object(f, OBJECT_ENTRY_ARRAY, ap, &o);
839 if (r < 0)
840 return r;
841
842 o->entry_array.next_entry_array_offset = htole64(q);
843 }
844
845 *idx = htole64(hidx + 1);
846
847 return 0;
848 }
849
850 static int link_entry_into_array_plus_one(JournalFile *f,
851 uint64_t *extra,
852 uint64_t *first,
853 uint64_t *idx,
854 uint64_t p) {
855
856 int r;
857
858 assert(f);
859 assert(extra);
860 assert(first);
861 assert(idx);
862 assert(p > 0);
863
864 if (*idx == 0)
865 *extra = htole64(p);
866 else {
867 uint64_t i;
868
869 i = le64toh(*idx) - 1;
870 r = link_entry_into_array(f, first, &i, p);
871 if (r < 0)
872 return r;
873 }
874
875 *idx = htole64(le64toh(*idx) + 1);
876 return 0;
877 }
878
879 static int journal_file_link_entry_item(JournalFile *f, Object *o, uint64_t offset, uint64_t i) {
880 uint64_t p;
881 int r;
882 assert(f);
883 assert(o);
884 assert(offset > 0);
885
886 p = le64toh(o->entry.items[i].object_offset);
887 if (p == 0)
888 return -EINVAL;
889
890 r = journal_file_move_to_object(f, OBJECT_DATA, p, &o);
891 if (r < 0)
892 return r;
893
894 return link_entry_into_array_plus_one(f,
895 &o->data.entry_offset,
896 &o->data.entry_array_offset,
897 &o->data.n_entries,
898 offset);
899 }
900
901 static int journal_file_link_entry(JournalFile *f, Object *o, uint64_t offset) {
902 uint64_t n, i;
903 int r;
904
905 assert(f);
906 assert(o);
907 assert(offset > 0);
908 assert(o->object.type == OBJECT_ENTRY);
909
910 __sync_synchronize();
911
912 /* Link up the entry itself */
913 r = link_entry_into_array(f,
914 &f->header->entry_array_offset,
915 &f->header->n_entries,
916 offset);
917 if (r < 0)
918 return r;
919
920 /* log_debug("=> %s seqnr=%lu n_entries=%lu", f->path, (unsigned long) o->entry.seqnum, (unsigned long) f->header->n_entries); */
921
922 if (f->header->head_entry_realtime == 0)
923 f->header->head_entry_realtime = o->entry.realtime;
924
925 f->header->tail_entry_realtime = o->entry.realtime;
926 f->header->tail_entry_monotonic = o->entry.monotonic;
927
928 f->tail_entry_monotonic_valid = true;
929
930 /* Link up the items */
931 n = journal_file_entry_n_items(o);
932 for (i = 0; i < n; i++) {
933 r = journal_file_link_entry_item(f, o, offset, i);
934 if (r < 0)
935 return r;
936 }
937
938 return 0;
939 }
940
941 static int journal_file_append_entry_internal(
942 JournalFile *f,
943 const dual_timestamp *ts,
944 uint64_t xor_hash,
945 const EntryItem items[], unsigned n_items,
946 uint64_t *seqnum,
947 Object **ret, uint64_t *offset) {
948 uint64_t np;
949 uint64_t osize;
950 Object *o;
951 int r;
952
953 assert(f);
954 assert(items || n_items == 0);
955 assert(ts);
956
957 osize = offsetof(Object, entry.items) + (n_items * sizeof(EntryItem));
958
959 r = journal_file_append_object(f, OBJECT_ENTRY, osize, &o, &np);
960 if (r < 0)
961 return r;
962
963 o->entry.seqnum = htole64(journal_file_seqnum(f, seqnum));
964 memcpy(o->entry.items, items, n_items * sizeof(EntryItem));
965 o->entry.realtime = htole64(ts->realtime);
966 o->entry.monotonic = htole64(ts->monotonic);
967 o->entry.xor_hash = htole64(xor_hash);
968 o->entry.boot_id = f->header->boot_id;
969
970 r = journal_file_link_entry(f, o, np);
971 if (r < 0)
972 return r;
973
974 if (ret)
975 *ret = o;
976
977 if (offset)
978 *offset = np;
979
980 return 0;
981 }
982
983 void journal_file_post_change(JournalFile *f) {
984 assert(f);
985
986 /* inotify() does not receive IN_MODIFY events from file
987 * accesses done via mmap(). After each access we hence
988 * trigger IN_MODIFY by truncating the journal file to its
989 * current size which triggers IN_MODIFY. */
990
991 __sync_synchronize();
992
993 if (ftruncate(f->fd, f->last_stat.st_size) < 0)
994 log_error("Failed to to truncate file to its own size: %m");
995 }
996
997 int journal_file_append_entry(JournalFile *f, const dual_timestamp *ts, const struct iovec iovec[], unsigned n_iovec, uint64_t *seqnum, Object **ret, uint64_t *offset) {
998 unsigned i;
999 EntryItem *items;
1000 int r;
1001 uint64_t xor_hash = 0;
1002 struct dual_timestamp _ts;
1003
1004 assert(f);
1005 assert(iovec || n_iovec == 0);
1006
1007 if (!f->writable)
1008 return -EPERM;
1009
1010 if (!ts) {
1011 dual_timestamp_get(&_ts);
1012 ts = &_ts;
1013 }
1014
1015 if (f->tail_entry_monotonic_valid &&
1016 ts->monotonic < le64toh(f->header->tail_entry_monotonic))
1017 return -EINVAL;
1018
1019 items = alloca(sizeof(EntryItem) * n_iovec);
1020
1021 for (i = 0; i < n_iovec; i++) {
1022 uint64_t p;
1023 Object *o;
1024
1025 r = journal_file_append_data(f, iovec[i].iov_base, iovec[i].iov_len, &o, &p);
1026 if (r < 0)
1027 return r;
1028
1029 xor_hash ^= le64toh(o->data.hash);
1030 items[i].object_offset = htole64(p);
1031 items[i].hash = o->data.hash;
1032 }
1033
1034 r = journal_file_append_entry_internal(f, ts, xor_hash, items, n_iovec, seqnum, ret, offset);
1035
1036 journal_file_post_change(f);
1037
1038 return r;
1039 }
1040
1041 static int generic_array_get(JournalFile *f,
1042 uint64_t first,
1043 uint64_t i,
1044 Object **ret, uint64_t *offset) {
1045
1046 Object *o;
1047 uint64_t p = 0, a;
1048 int r;
1049
1050 assert(f);
1051
1052 a = first;
1053 while (a > 0) {
1054 uint64_t n;
1055
1056 r = journal_file_move_to_object(f, OBJECT_ENTRY_ARRAY, a, &o);
1057 if (r < 0)
1058 return r;
1059
1060 n = journal_file_entry_array_n_items(o);
1061 if (i < n) {
1062 p = le64toh(o->entry_array.items[i]);
1063 break;
1064 }
1065
1066 i -= n;
1067 a = le64toh(o->entry_array.next_entry_array_offset);
1068 }
1069
1070 if (a <= 0 || p <= 0)
1071 return 0;
1072
1073 r = journal_file_move_to_object(f, OBJECT_ENTRY, p, &o);
1074 if (r < 0)
1075 return r;
1076
1077 if (ret)
1078 *ret = o;
1079
1080 if (offset)
1081 *offset = p;
1082
1083 return 1;
1084 }
1085
1086 static int generic_array_get_plus_one(JournalFile *f,
1087 uint64_t extra,
1088 uint64_t first,
1089 uint64_t i,
1090 Object **ret, uint64_t *offset) {
1091
1092 Object *o;
1093
1094 assert(f);
1095
1096 if (i == 0) {
1097 int r;
1098
1099 r = journal_file_move_to_object(f, OBJECT_ENTRY, extra, &o);
1100 if (r < 0)
1101 return r;
1102
1103 if (ret)
1104 *ret = o;
1105
1106 if (offset)
1107 *offset = extra;
1108
1109 return 1;
1110 }
1111
1112 return generic_array_get(f, first, i-1, ret, offset);
1113 }
1114
1115 enum {
1116 TEST_FOUND,
1117 TEST_LEFT,
1118 TEST_RIGHT
1119 };
1120
1121 static int generic_array_bisect(JournalFile *f,
1122 uint64_t first,
1123 uint64_t n,
1124 uint64_t needle,
1125 int (*test_object)(JournalFile *f, uint64_t p, uint64_t needle),
1126 direction_t direction,
1127 Object **ret,
1128 uint64_t *offset,
1129 uint64_t *idx) {
1130
1131 uint64_t a, p, t = 0, i = 0, last_p = 0;
1132 bool subtract_one = false;
1133 Object *o, *array = NULL;
1134 int r;
1135
1136 assert(f);
1137 assert(test_object);
1138
1139 a = first;
1140 while (a > 0) {
1141 uint64_t left, right, k, lp;
1142
1143 r = journal_file_move_to_object(f, OBJECT_ENTRY_ARRAY, a, &array);
1144 if (r < 0)
1145 return r;
1146
1147 k = journal_file_entry_array_n_items(array);
1148 right = MIN(k, n);
1149 if (right <= 0)
1150 return 0;
1151
1152 i = right - 1;
1153 lp = p = le64toh(array->entry_array.items[i]);
1154 if (p <= 0)
1155 return -EBADMSG;
1156
1157 r = test_object(f, p, needle);
1158 if (r < 0)
1159 return r;
1160
1161 if (r == TEST_FOUND)
1162 r = direction == DIRECTION_DOWN ? TEST_RIGHT : TEST_LEFT;
1163
1164 if (r == TEST_RIGHT) {
1165 left = 0;
1166 right -= 1;
1167 for (;;) {
1168 if (left == right) {
1169 if (direction == DIRECTION_UP)
1170 subtract_one = true;
1171
1172 i = left;
1173 goto found;
1174 }
1175
1176 assert(left < right);
1177
1178 i = (left + right) / 2;
1179 p = le64toh(array->entry_array.items[i]);
1180 if (p <= 0)
1181 return -EBADMSG;
1182
1183 r = test_object(f, p, needle);
1184 if (r < 0)
1185 return r;
1186
1187 if (r == TEST_FOUND)
1188 r = direction == DIRECTION_DOWN ? TEST_RIGHT : TEST_LEFT;
1189
1190 if (r == TEST_RIGHT)
1191 right = i;
1192 else
1193 left = i + 1;
1194 }
1195 }
1196
1197 if (k > n)
1198 return 0;
1199
1200 last_p = lp;
1201
1202 n -= k;
1203 t += k;
1204 a = le64toh(array->entry_array.next_entry_array_offset);
1205 }
1206
1207 return 0;
1208
1209 found:
1210 if (subtract_one && t == 0 && i == 0)
1211 return 0;
1212
1213 if (subtract_one && i == 0)
1214 p = last_p;
1215 else if (subtract_one)
1216 p = le64toh(array->entry_array.items[i-1]);
1217 else
1218 p = le64toh(array->entry_array.items[i]);
1219
1220 r = journal_file_move_to_object(f, OBJECT_ENTRY, p, &o);
1221 if (r < 0)
1222 return r;
1223
1224 if (ret)
1225 *ret = o;
1226
1227 if (offset)
1228 *offset = p;
1229
1230 if (idx)
1231 *idx = t + i - (subtract_one ? 1 : 0);
1232
1233 return 1;
1234 }
1235
1236 static int generic_array_bisect_plus_one(JournalFile *f,
1237 uint64_t extra,
1238 uint64_t first,
1239 uint64_t n,
1240 uint64_t needle,
1241 int (*test_object)(JournalFile *f, uint64_t p, uint64_t needle),
1242 direction_t direction,
1243 Object **ret,
1244 uint64_t *offset,
1245 uint64_t *idx) {
1246
1247 int r;
1248
1249 assert(f);
1250 assert(test_object);
1251
1252 if (n <= 0)
1253 return 0;
1254
1255 /* This bisects the array in object 'first', but first checks
1256 * an extra */
1257 r = test_object(f, extra, needle);
1258 if (r < 0)
1259 return r;
1260 else if (r == TEST_FOUND) {
1261 Object *o;
1262
1263 r = journal_file_move_to_object(f, OBJECT_ENTRY, extra, &o);
1264 if (r < 0)
1265 return r;
1266
1267 if (ret)
1268 *ret = o;
1269
1270 if (offset)
1271 *offset = extra;
1272
1273 if (idx)
1274 *idx = 0;
1275
1276 return 1;
1277 } else if (r == TEST_RIGHT)
1278 return 0;
1279
1280 r = generic_array_bisect(f, first, n-1, needle, test_object, direction, ret, offset, idx);
1281
1282 if (r > 0)
1283 (*idx) ++;
1284
1285 return r;
1286 }
1287
1288 static int test_object_seqnum(JournalFile *f, uint64_t p, uint64_t needle) {
1289 Object *o;
1290 int r;
1291
1292 assert(f);
1293 assert(p > 0);
1294
1295 r = journal_file_move_to_object(f, OBJECT_ENTRY, p, &o);
1296 if (r < 0)
1297 return r;
1298
1299 if (le64toh(o->entry.seqnum) == needle)
1300 return TEST_FOUND;
1301 else if (le64toh(o->entry.seqnum) < needle)
1302 return TEST_LEFT;
1303 else
1304 return TEST_RIGHT;
1305 }
1306
1307 int journal_file_move_to_entry_by_seqnum(
1308 JournalFile *f,
1309 uint64_t seqnum,
1310 direction_t direction,
1311 Object **ret,
1312 uint64_t *offset) {
1313
1314 return generic_array_bisect(f,
1315 le64toh(f->header->entry_array_offset),
1316 le64toh(f->header->n_entries),
1317 seqnum,
1318 test_object_seqnum,
1319 direction,
1320 ret, offset, NULL);
1321 }
1322
1323 static int test_object_realtime(JournalFile *f, uint64_t p, uint64_t needle) {
1324 Object *o;
1325 int r;
1326
1327 assert(f);
1328 assert(p > 0);
1329
1330 r = journal_file_move_to_object(f, OBJECT_ENTRY, p, &o);
1331 if (r < 0)
1332 return r;
1333
1334 if (le64toh(o->entry.realtime) == needle)
1335 return TEST_FOUND;
1336 else if (le64toh(o->entry.realtime) < needle)
1337 return TEST_LEFT;
1338 else
1339 return TEST_RIGHT;
1340 }
1341
1342 int journal_file_move_to_entry_by_realtime(
1343 JournalFile *f,
1344 uint64_t realtime,
1345 direction_t direction,
1346 Object **ret,
1347 uint64_t *offset) {
1348
1349 return generic_array_bisect(f,
1350 le64toh(f->header->entry_array_offset),
1351 le64toh(f->header->n_entries),
1352 realtime,
1353 test_object_realtime,
1354 direction,
1355 ret, offset, NULL);
1356 }
1357
1358 static int test_object_monotonic(JournalFile *f, uint64_t p, uint64_t needle) {
1359 Object *o;
1360 int r;
1361
1362 assert(f);
1363 assert(p > 0);
1364
1365 r = journal_file_move_to_object(f, OBJECT_ENTRY, p, &o);
1366 if (r < 0)
1367 return r;
1368
1369 if (le64toh(o->entry.monotonic) == needle)
1370 return TEST_FOUND;
1371 else if (le64toh(o->entry.monotonic) < needle)
1372 return TEST_LEFT;
1373 else
1374 return TEST_RIGHT;
1375 }
1376
1377 int journal_file_move_to_entry_by_monotonic(
1378 JournalFile *f,
1379 sd_id128_t boot_id,
1380 uint64_t monotonic,
1381 direction_t direction,
1382 Object **ret,
1383 uint64_t *offset) {
1384
1385 char t[8+32+1] = "_BOOT_ID=";
1386 Object *o;
1387 int r;
1388
1389 sd_id128_to_string(boot_id, t + 8);
1390
1391 r = journal_file_find_data_object(f, t, strlen(t), &o, NULL);
1392 if (r < 0)
1393 return r;
1394 else if (r == 0)
1395 return -ENOENT;
1396
1397 return generic_array_bisect_plus_one(f,
1398 le64toh(o->data.entry_offset),
1399 le64toh(o->data.entry_array_offset),
1400 le64toh(o->data.n_entries),
1401 monotonic,
1402 test_object_monotonic,
1403 direction,
1404 ret, offset, NULL);
1405 }
1406
1407 static int test_object_offset(JournalFile *f, uint64_t p, uint64_t needle) {
1408 assert(f);
1409 assert(p > 0);
1410
1411 if (p == needle)
1412 return TEST_FOUND;
1413 else if (p < needle)
1414 return TEST_LEFT;
1415 else
1416 return TEST_RIGHT;
1417 }
1418
1419 int journal_file_next_entry(
1420 JournalFile *f,
1421 Object *o, uint64_t p,
1422 direction_t direction,
1423 Object **ret, uint64_t *offset) {
1424
1425 uint64_t i, n;
1426 int r;
1427
1428 assert(f);
1429 assert(p > 0 || !o);
1430
1431 n = le64toh(f->header->n_entries);
1432 if (n <= 0)
1433 return 0;
1434
1435 if (!o)
1436 i = direction == DIRECTION_DOWN ? 0 : n - 1;
1437 else {
1438 if (o->object.type != OBJECT_ENTRY)
1439 return -EINVAL;
1440
1441 r = generic_array_bisect(f,
1442 le64toh(f->header->entry_array_offset),
1443 le64toh(f->header->n_entries),
1444 p,
1445 test_object_offset,
1446 DIRECTION_DOWN,
1447 NULL, NULL,
1448 &i);
1449 if (r <= 0)
1450 return r;
1451
1452 if (direction == DIRECTION_DOWN) {
1453 if (i >= n - 1)
1454 return 0;
1455
1456 i++;
1457 } else {
1458 if (i <= 0)
1459 return 0;
1460
1461 i--;
1462 }
1463 }
1464
1465 /* And jump to it */
1466 return generic_array_get(f,
1467 le64toh(f->header->entry_array_offset),
1468 i,
1469 ret, offset);
1470 }
1471
1472 int journal_file_skip_entry(
1473 JournalFile *f,
1474 Object *o, uint64_t p,
1475 int64_t skip,
1476 Object **ret, uint64_t *offset) {
1477
1478 uint64_t i, n;
1479 int r;
1480
1481 assert(f);
1482 assert(o);
1483 assert(p > 0);
1484
1485 if (o->object.type != OBJECT_ENTRY)
1486 return -EINVAL;
1487
1488 r = generic_array_bisect(f,
1489 le64toh(f->header->entry_array_offset),
1490 le64toh(f->header->n_entries),
1491 p,
1492 test_object_offset,
1493 DIRECTION_DOWN,
1494 NULL, NULL,
1495 &i);
1496 if (r <= 0)
1497 return r;
1498
1499 /* Calculate new index */
1500 if (skip < 0) {
1501 if ((uint64_t) -skip >= i)
1502 i = 0;
1503 else
1504 i = i - (uint64_t) -skip;
1505 } else
1506 i += (uint64_t) skip;
1507
1508 n = le64toh(f->header->n_entries);
1509 if (n <= 0)
1510 return -EBADMSG;
1511
1512 if (i >= n)
1513 i = n-1;
1514
1515 return generic_array_get(f,
1516 le64toh(f->header->entry_array_offset),
1517 i,
1518 ret, offset);
1519 }
1520
1521 int journal_file_next_entry_for_data(
1522 JournalFile *f,
1523 Object *o, uint64_t p,
1524 uint64_t data_offset,
1525 direction_t direction,
1526 Object **ret, uint64_t *offset) {
1527
1528 uint64_t n, i;
1529 int r;
1530 Object *d;
1531
1532 assert(f);
1533 assert(p > 0 || !o);
1534
1535 r = journal_file_move_to_object(f, OBJECT_DATA, data_offset, &d);
1536 if (r < 0)
1537 return r;
1538
1539 n = le64toh(d->data.n_entries);
1540 if (n <= 0)
1541 return n;
1542
1543 if (!o)
1544 i = direction == DIRECTION_DOWN ? 0 : n - 1;
1545 else {
1546 if (o->object.type != OBJECT_ENTRY)
1547 return -EINVAL;
1548
1549 r = generic_array_bisect_plus_one(f,
1550 le64toh(d->data.entry_offset),
1551 le64toh(d->data.entry_array_offset),
1552 le64toh(d->data.n_entries),
1553 p,
1554 test_object_offset,
1555 DIRECTION_DOWN,
1556 NULL, NULL,
1557 &i);
1558
1559 if (r <= 0)
1560 return r;
1561
1562 if (direction == DIRECTION_DOWN) {
1563 if (i >= n - 1)
1564 return 0;
1565
1566 i++;
1567 } else {
1568 if (i <= 0)
1569 return 0;
1570
1571 i--;
1572 }
1573
1574 }
1575
1576 return generic_array_get_plus_one(f,
1577 le64toh(d->data.entry_offset),
1578 le64toh(d->data.entry_array_offset),
1579 i,
1580 ret, offset);
1581 }
1582
1583 int journal_file_move_to_entry_by_seqnum_for_data(
1584 JournalFile *f,
1585 uint64_t data_offset,
1586 uint64_t seqnum,
1587 direction_t direction,
1588 Object **ret, uint64_t *offset) {
1589
1590 Object *d;
1591 int r;
1592
1593 r = journal_file_move_to_object(f, OBJECT_DATA, data_offset, &d);
1594 if (r <= 0)
1595 return r;
1596
1597 return generic_array_bisect_plus_one(f,
1598 le64toh(d->data.entry_offset),
1599 le64toh(d->data.entry_array_offset),
1600 le64toh(d->data.n_entries),
1601 seqnum,
1602 test_object_seqnum,
1603 direction,
1604 ret, offset, NULL);
1605 }
1606
1607 int journal_file_move_to_entry_by_realtime_for_data(
1608 JournalFile *f,
1609 uint64_t data_offset,
1610 uint64_t realtime,
1611 direction_t direction,
1612 Object **ret, uint64_t *offset) {
1613
1614 Object *d;
1615 int r;
1616
1617 r = journal_file_move_to_object(f, OBJECT_DATA, data_offset, &d);
1618 if (r <= 0)
1619 return r;
1620
1621 return generic_array_bisect_plus_one(f,
1622 le64toh(d->data.entry_offset),
1623 le64toh(d->data.entry_array_offset),
1624 le64toh(d->data.n_entries),
1625 realtime,
1626 test_object_realtime,
1627 direction,
1628 ret, offset, NULL);
1629 }
1630
1631 void journal_file_dump(JournalFile *f) {
1632 char a[33], b[33], c[33];
1633 Object *o;
1634 int r;
1635 uint64_t p;
1636
1637 assert(f);
1638
1639 printf("File Path: %s\n"
1640 "File ID: %s\n"
1641 "Machine ID: %s\n"
1642 "Boot ID: %s\n"
1643 "Arena size: %llu\n"
1644 "Objects: %lu\n"
1645 "Entries: %lu\n",
1646 f->path,
1647 sd_id128_to_string(f->header->file_id, a),
1648 sd_id128_to_string(f->header->machine_id, b),
1649 sd_id128_to_string(f->header->boot_id, c),
1650 (unsigned long long) le64toh(f->header->arena_size),
1651 (unsigned long) le64toh(f->header->n_objects),
1652 (unsigned long) le64toh(f->header->n_entries));
1653
1654 p = le64toh(f->header->arena_offset);
1655 while (p != 0) {
1656 r = journal_file_move_to_object(f, -1, p, &o);
1657 if (r < 0)
1658 goto fail;
1659
1660 switch (o->object.type) {
1661
1662 case OBJECT_UNUSED:
1663 printf("Type: OBJECT_UNUSED\n");
1664 break;
1665
1666 case OBJECT_DATA:
1667 printf("Type: OBJECT_DATA\n");
1668 break;
1669
1670 case OBJECT_ENTRY:
1671 printf("Type: OBJECT_ENTRY %llu %llu %llu\n",
1672 (unsigned long long) le64toh(o->entry.seqnum),
1673 (unsigned long long) le64toh(o->entry.monotonic),
1674 (unsigned long long) le64toh(o->entry.realtime));
1675 break;
1676
1677 case OBJECT_FIELD_HASH_TABLE:
1678 printf("Type: OBJECT_FIELD_HASH_TABLE\n");
1679 break;
1680
1681 case OBJECT_DATA_HASH_TABLE:
1682 printf("Type: OBJECT_DATA_HASH_TABLE\n");
1683 break;
1684
1685 case OBJECT_ENTRY_ARRAY:
1686 printf("Type: OBJECT_ENTRY_ARRAY\n");
1687 break;
1688 }
1689
1690 if (o->object.flags & OBJECT_COMPRESSED)
1691 printf("Flags: COMPRESSED\n");
1692
1693 if (p == le64toh(f->header->tail_object_offset))
1694 p = 0;
1695 else
1696 p = p + ALIGN64(le64toh(o->object.size));
1697 }
1698
1699 return;
1700 fail:
1701 log_error("File corrupt");
1702 }
1703
1704 int journal_file_open(
1705 const char *fname,
1706 int flags,
1707 mode_t mode,
1708 JournalFile *template,
1709 JournalFile **ret) {
1710
1711 JournalFile *f;
1712 int r;
1713 bool newly_created = false;
1714
1715 assert(fname);
1716
1717 if ((flags & O_ACCMODE) != O_RDONLY &&
1718 (flags & O_ACCMODE) != O_RDWR)
1719 return -EINVAL;
1720
1721 f = new0(JournalFile, 1);
1722 if (!f)
1723 return -ENOMEM;
1724
1725 f->fd = -1;
1726 f->flags = flags;
1727 f->mode = mode;
1728 f->writable = (flags & O_ACCMODE) != O_RDONLY;
1729 f->prot = prot_from_flags(flags);
1730
1731 f->path = strdup(fname);
1732 if (!f->path) {
1733 r = -ENOMEM;
1734 goto fail;
1735 }
1736
1737 f->fd = open(f->path, f->flags|O_CLOEXEC, f->mode);
1738 if (f->fd < 0) {
1739 r = -errno;
1740 goto fail;
1741 }
1742
1743 if (fstat(f->fd, &f->last_stat) < 0) {
1744 r = -errno;
1745 goto fail;
1746 }
1747
1748 if (f->last_stat.st_size == 0 && f->writable) {
1749 newly_created = true;
1750
1751 r = journal_file_init_header(f, template);
1752 if (r < 0)
1753 goto fail;
1754
1755 if (fstat(f->fd, &f->last_stat) < 0) {
1756 r = -errno;
1757 goto fail;
1758 }
1759 }
1760
1761 if (f->last_stat.st_size < (off_t) sizeof(Header)) {
1762 r = -EIO;
1763 goto fail;
1764 }
1765
1766 f->header = mmap(NULL, PAGE_ALIGN(sizeof(Header)), prot_from_flags(flags), MAP_SHARED, f->fd, 0);
1767 if (f->header == MAP_FAILED) {
1768 f->header = NULL;
1769 r = -errno;
1770 goto fail;
1771 }
1772
1773 if (!newly_created) {
1774 r = journal_file_verify_header(f);
1775 if (r < 0)
1776 goto fail;
1777 }
1778
1779 if (f->writable) {
1780 r = journal_file_refresh_header(f);
1781 if (r < 0)
1782 goto fail;
1783 }
1784
1785 if (newly_created) {
1786
1787 r = journal_file_setup_field_hash_table(f);
1788 if (r < 0)
1789 goto fail;
1790
1791 r = journal_file_setup_data_hash_table(f);
1792 if (r < 0)
1793 goto fail;
1794 }
1795
1796 r = journal_file_map_field_hash_table(f);
1797 if (r < 0)
1798 goto fail;
1799
1800 r = journal_file_map_data_hash_table(f);
1801 if (r < 0)
1802 goto fail;
1803
1804 if (ret)
1805 *ret = f;
1806
1807 return 0;
1808
1809 fail:
1810 journal_file_close(f);
1811
1812 return r;
1813 }
1814
1815 int journal_file_rotate(JournalFile **f) {
1816 char *p;
1817 size_t l;
1818 JournalFile *old_file, *new_file = NULL;
1819 int r;
1820
1821 assert(f);
1822 assert(*f);
1823
1824 old_file = *f;
1825
1826 if (!old_file->writable)
1827 return -EINVAL;
1828
1829 if (!endswith(old_file->path, ".journal"))
1830 return -EINVAL;
1831
1832 l = strlen(old_file->path);
1833
1834 p = new(char, l + 1 + 16 + 1 + 32 + 1 + 16 + 1);
1835 if (!p)
1836 return -ENOMEM;
1837
1838 memcpy(p, old_file->path, l - 8);
1839 p[l-8] = '@';
1840 sd_id128_to_string(old_file->header->seqnum_id, p + l - 8 + 1);
1841 snprintf(p + l - 8 + 1 + 32, 1 + 16 + 1 + 16 + 8 + 1,
1842 "-%016llx-%016llx.journal",
1843 (unsigned long long) le64toh((*f)->header->seqnum),
1844 (unsigned long long) le64toh((*f)->header->tail_entry_realtime));
1845
1846 r = rename(old_file->path, p);
1847 free(p);
1848
1849 if (r < 0)
1850 return -errno;
1851
1852 old_file->header->state = le32toh(STATE_ARCHIVED);
1853
1854 r = journal_file_open(old_file->path, old_file->flags, old_file->mode, old_file, &new_file);
1855 journal_file_close(old_file);
1856
1857 *f = new_file;
1858 return r;
1859 }
1860
1861 struct vacuum_info {
1862 off_t usage;
1863 char *filename;
1864
1865 uint64_t realtime;
1866 sd_id128_t seqnum_id;
1867 uint64_t seqnum;
1868 };
1869
1870 static int vacuum_compare(const void *_a, const void *_b) {
1871 const struct vacuum_info *a, *b;
1872
1873 a = _a;
1874 b = _b;
1875
1876 if (sd_id128_equal(a->seqnum_id, b->seqnum_id)) {
1877 if (a->seqnum < b->seqnum)
1878 return -1;
1879 else if (a->seqnum > b->seqnum)
1880 return 1;
1881 else
1882 return 0;
1883 }
1884
1885 if (a->realtime < b->realtime)
1886 return -1;
1887 else if (a->realtime > b->realtime)
1888 return 1;
1889 else
1890 return memcmp(&a->seqnum_id, &b->seqnum_id, 16);
1891 }
1892
1893 int journal_directory_vacuum(const char *directory, uint64_t max_use, uint64_t min_free) {
1894 DIR *d;
1895 int r = 0;
1896 struct vacuum_info *list = NULL;
1897 unsigned n_list = 0, n_allocated = 0, i;
1898 uint64_t sum = 0;
1899
1900 assert(directory);
1901
1902 if (max_use <= 0)
1903 return 0;
1904
1905 d = opendir(directory);
1906 if (!d)
1907 return -errno;
1908
1909 for (;;) {
1910 int k;
1911 struct dirent buf, *de;
1912 size_t q;
1913 struct stat st;
1914 char *p;
1915 unsigned long long seqnum, realtime;
1916 sd_id128_t seqnum_id;
1917
1918 k = readdir_r(d, &buf, &de);
1919 if (k != 0) {
1920 r = -k;
1921 goto finish;
1922 }
1923
1924 if (!de)
1925 break;
1926
1927 if (!dirent_is_file_with_suffix(de, ".journal"))
1928 continue;
1929
1930 q = strlen(de->d_name);
1931
1932 if (q < 1 + 32 + 1 + 16 + 1 + 16 + 8)
1933 continue;
1934
1935 if (de->d_name[q-8-16-1] != '-' ||
1936 de->d_name[q-8-16-1-16-1] != '-' ||
1937 de->d_name[q-8-16-1-16-1-32-1] != '@')
1938 continue;
1939
1940 if (fstatat(dirfd(d), de->d_name, &st, AT_SYMLINK_NOFOLLOW) < 0)
1941 continue;
1942
1943 if (!S_ISREG(st.st_mode))
1944 continue;
1945
1946 p = strdup(de->d_name);
1947 if (!p) {
1948 r = -ENOMEM;
1949 goto finish;
1950 }
1951
1952 de->d_name[q-8-16-1-16-1] = 0;
1953 if (sd_id128_from_string(de->d_name + q-8-16-1-16-1-32, &seqnum_id) < 0) {
1954 free(p);
1955 continue;
1956 }
1957
1958 if (sscanf(de->d_name + q-8-16-1-16, "%16llx-%16llx.journal", &seqnum, &realtime) != 2) {
1959 free(p);
1960 continue;
1961 }
1962
1963 if (n_list >= n_allocated) {
1964 struct vacuum_info *j;
1965
1966 n_allocated = MAX(n_allocated * 2U, 8U);
1967 j = realloc(list, n_allocated * sizeof(struct vacuum_info));
1968 if (!j) {
1969 free(p);
1970 r = -ENOMEM;
1971 goto finish;
1972 }
1973
1974 list = j;
1975 }
1976
1977 list[n_list].filename = p;
1978 list[n_list].usage = (uint64_t) st.st_blksize * (uint64_t) st.st_blocks;
1979 list[n_list].seqnum = seqnum;
1980 list[n_list].realtime = realtime;
1981 list[n_list].seqnum_id = seqnum_id;
1982
1983 sum += list[n_list].usage;
1984
1985 n_list ++;
1986 }
1987
1988 qsort(list, n_list, sizeof(struct vacuum_info), vacuum_compare);
1989
1990 for(i = 0; i < n_list; i++) {
1991 struct statvfs ss;
1992
1993 if (fstatvfs(dirfd(d), &ss) < 0) {
1994 r = -errno;
1995 goto finish;
1996 }
1997
1998 if (sum <= max_use &&
1999 (uint64_t) ss.f_bavail * (uint64_t) ss.f_bsize >= min_free)
2000 break;
2001
2002 if (unlinkat(dirfd(d), list[i].filename, 0) >= 0) {
2003 log_debug("Deleted archived journal %s/%s.", directory, list[i].filename);
2004 sum -= list[i].usage;
2005 } else if (errno != ENOENT)
2006 log_warning("Failed to delete %s/%s: %m", directory, list[i].filename);
2007 }
2008
2009 finish:
2010 for (i = 0; i < n_list; i++)
2011 free(list[i].filename);
2012
2013 free(list);
2014
2015 if (d)
2016 closedir(d);
2017
2018 return r;
2019 }
2020
2021 int journal_file_copy_entry(JournalFile *from, JournalFile *to, Object *o, uint64_t p, uint64_t *seqnum, Object **ret, uint64_t *offset) {
2022 uint64_t i, n;
2023 uint64_t q, xor_hash = 0;
2024 int r;
2025 EntryItem *items;
2026 dual_timestamp ts;
2027
2028 assert(from);
2029 assert(to);
2030 assert(o);
2031 assert(p);
2032
2033 if (!to->writable)
2034 return -EPERM;
2035
2036 ts.monotonic = le64toh(o->entry.monotonic);
2037 ts.realtime = le64toh(o->entry.realtime);
2038
2039 if (to->tail_entry_monotonic_valid &&
2040 ts.monotonic < le64toh(to->header->tail_entry_monotonic))
2041 return -EINVAL;
2042
2043 if (ts.realtime < le64toh(to->header->tail_entry_realtime))
2044 return -EINVAL;
2045
2046 n = journal_file_entry_n_items(o);
2047 items = alloca(sizeof(EntryItem) * n);
2048
2049 for (i = 0; i < n; i++) {
2050 uint64_t le_hash, l, h;
2051 size_t t;
2052 void *data;
2053 Object *u;
2054
2055 q = le64toh(o->entry.items[i].object_offset);
2056 le_hash = o->entry.items[i].hash;
2057
2058 r = journal_file_move_to_object(from, OBJECT_DATA, q, &o);
2059 if (r < 0)
2060 return r;
2061
2062 if (le_hash != o->data.hash)
2063 return -EBADMSG;
2064
2065 l = le64toh(o->object.size) - offsetof(Object, data.payload);
2066 t = (size_t) l;
2067
2068 /* We hit the limit on 32bit machines */
2069 if ((uint64_t) t != l)
2070 return -E2BIG;
2071
2072 if (o->object.flags & OBJECT_COMPRESSED) {
2073 #ifdef HAVE_XZ
2074 uint64_t rsize;
2075
2076 if (!uncompress_blob(o->data.payload, l, &from->compress_buffer, &from->compress_buffer_size, &rsize))
2077 return -EBADMSG;
2078
2079 data = from->compress_buffer;
2080 l = rsize;
2081 #else
2082 return -EPROTONOSUPPORT;
2083 #endif
2084 } else
2085 data = o->data.payload;
2086
2087 r = journal_file_append_data(to, data, l, &u, &h);
2088 if (r < 0)
2089 return r;
2090
2091 xor_hash ^= le64toh(u->data.hash);
2092 items[i].object_offset = htole64(h);
2093 items[i].hash = u->data.hash;
2094
2095 r = journal_file_move_to_object(from, OBJECT_ENTRY, p, &o);
2096 if (r < 0)
2097 return r;
2098 }
2099
2100 return journal_file_append_entry_internal(to, &ts, xor_hash, items, n, seqnum, ret, offset);
2101 }
2102
2103 void journal_default_metrics(JournalMetrics *m, int fd) {
2104 uint64_t fs_size = 0;
2105 struct statvfs ss;
2106 char a[FORMAT_BYTES_MAX], b[FORMAT_BYTES_MAX], c[FORMAT_BYTES_MAX], d[FORMAT_BYTES_MAX];
2107
2108 assert(m);
2109 assert(fd >= 0);
2110
2111 if (fstatvfs(fd, &ss) >= 0)
2112 fs_size = ss.f_frsize * ss.f_blocks;
2113
2114 if (m->max_use == (uint64_t) -1) {
2115
2116 if (fs_size > 0) {
2117 m->max_use = PAGE_ALIGN(fs_size / 10); /* 10% of file system size */
2118
2119 if (m->max_use > DEFAULT_MAX_USE_UPPER)
2120 m->max_use = DEFAULT_MAX_USE_UPPER;
2121
2122 if (m->max_use < DEFAULT_MAX_USE_LOWER)
2123 m->max_use = DEFAULT_MAX_USE_LOWER;
2124 } else
2125 m->max_use = DEFAULT_MAX_USE_LOWER;
2126 } else {
2127 m->max_use = PAGE_ALIGN(m->max_use);
2128
2129 if (m->max_use < JOURNAL_FILE_SIZE_MIN*2)
2130 m->max_use = JOURNAL_FILE_SIZE_MIN*2;
2131 }
2132
2133 if (m->max_size == (uint64_t) -1) {
2134 m->max_size = PAGE_ALIGN(m->max_use / 8); /* 8 chunks */
2135
2136 if (m->max_size > DEFAULT_MAX_SIZE_UPPER)
2137 m->max_size = DEFAULT_MAX_SIZE_UPPER;
2138 } else
2139 m->max_size = PAGE_ALIGN(m->max_size);
2140
2141 if (m->max_size < JOURNAL_FILE_SIZE_MIN)
2142 m->max_size = JOURNAL_FILE_SIZE_MIN;
2143
2144 if (m->max_size*2 > m->max_use)
2145 m->max_use = m->max_size*2;
2146
2147 if (m->min_size == (uint64_t) -1)
2148 m->min_size = JOURNAL_FILE_SIZE_MIN;
2149 else {
2150 m->min_size = PAGE_ALIGN(m->min_size);
2151
2152 if (m->min_size < JOURNAL_FILE_SIZE_MIN)
2153 m->min_size = JOURNAL_FILE_SIZE_MIN;
2154
2155 if (m->min_size > m->max_size)
2156 m->max_size = m->min_size;
2157 }
2158
2159 if (m->keep_free == (uint64_t) -1) {
2160
2161 if (fs_size > 0) {
2162 m->keep_free = PAGE_ALIGN(fs_size / 20); /* 5% of file system size */
2163
2164 if (m->keep_free > DEFAULT_KEEP_FREE_UPPER)
2165 m->keep_free = DEFAULT_KEEP_FREE_UPPER;
2166
2167 } else
2168 m->keep_free = DEFAULT_KEEP_FREE;
2169 }
2170
2171 log_debug("Fixed max_use=%s max_size=%s min_size=%s keep_free=%s",
2172 format_bytes(a, sizeof(a), m->max_use),
2173 format_bytes(b, sizeof(b), m->max_size),
2174 format_bytes(c, sizeof(c), m->min_size),
2175 format_bytes(d, sizeof(d), m->keep_free));
2176 }