]>
Commit | Line | Data |
---|---|---|
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 Lesser General Public License as published by | |
10 | the Free Software Foundation; either version 2.1 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 | Lesser General Public License for more details. | |
17 | ||
18 | You should have received a copy of the GNU Lesser 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 | #include <linux/fs.h> | |
30 | ||
31 | #include "btrfs-util.h" | |
32 | #include "journal-def.h" | |
33 | #include "journal-file.h" | |
34 | #include "journal-authenticate.h" | |
35 | #include "lookup3.h" | |
36 | #include "compress.h" | |
37 | #include "random-util.h" | |
38 | ||
39 | #define DEFAULT_DATA_HASH_TABLE_SIZE (2047ULL*sizeof(HashItem)) | |
40 | #define DEFAULT_FIELD_HASH_TABLE_SIZE (333ULL*sizeof(HashItem)) | |
41 | ||
42 | #define COMPRESSION_SIZE_THRESHOLD (512ULL) | |
43 | ||
44 | /* This is the minimum journal file size */ | |
45 | #define JOURNAL_FILE_SIZE_MIN (4ULL*1024ULL*1024ULL) /* 4 MiB */ | |
46 | ||
47 | /* These are the lower and upper bounds if we deduce the max_use value | |
48 | * from the file system size */ | |
49 | #define DEFAULT_MAX_USE_LOWER (1ULL*1024ULL*1024ULL) /* 1 MiB */ | |
50 | #define DEFAULT_MAX_USE_UPPER (4ULL*1024ULL*1024ULL*1024ULL) /* 4 GiB */ | |
51 | ||
52 | /* This is the default minimal use limit, how much we'll use even if keep_free suggests otherwise. */ | |
53 | #define DEFAULT_MIN_USE (1ULL*1024ULL*1024ULL) /* 1 MiB */ | |
54 | ||
55 | /* This is the upper bound if we deduce max_size from max_use */ | |
56 | #define DEFAULT_MAX_SIZE_UPPER (128ULL*1024ULL*1024ULL) /* 128 MiB */ | |
57 | ||
58 | /* This is the upper bound if we deduce the keep_free value from the | |
59 | * file system size */ | |
60 | #define DEFAULT_KEEP_FREE_UPPER (4ULL*1024ULL*1024ULL*1024ULL) /* 4 GiB */ | |
61 | ||
62 | /* This is the keep_free value when we can't determine the system | |
63 | * size */ | |
64 | #define DEFAULT_KEEP_FREE (1024ULL*1024ULL) /* 1 MB */ | |
65 | ||
66 | /* This is the default maximum number of journal files to keep around. */ | |
67 | #define DEFAULT_N_MAX_FILES (100) | |
68 | ||
69 | /* n_data was the first entry we added after the initial file format design */ | |
70 | #define HEADER_SIZE_MIN ALIGN64(offsetof(Header, n_data)) | |
71 | ||
72 | /* How many entries to keep in the entry array chain cache at max */ | |
73 | #define CHAIN_CACHE_MAX 20 | |
74 | ||
75 | /* How much to increase the journal file size at once each time we allocate something new. */ | |
76 | #define FILE_SIZE_INCREASE (8ULL*1024ULL*1024ULL) /* 8MB */ | |
77 | ||
78 | /* Reread fstat() of the file for detecting deletions at least this often */ | |
79 | #define LAST_STAT_REFRESH_USEC (5*USEC_PER_SEC) | |
80 | ||
81 | /* The mmap context to use for the header we pick as one above the last defined typed */ | |
82 | #define CONTEXT_HEADER _OBJECT_TYPE_MAX | |
83 | ||
84 | static int journal_file_set_online(JournalFile *f) { | |
85 | assert(f); | |
86 | ||
87 | if (!f->writable) | |
88 | return -EPERM; | |
89 | ||
90 | if (!(f->fd >= 0 && f->header)) | |
91 | return -EINVAL; | |
92 | ||
93 | if (mmap_cache_got_sigbus(f->mmap, f->fd)) | |
94 | return -EIO; | |
95 | ||
96 | switch(f->header->state) { | |
97 | case STATE_ONLINE: | |
98 | return 0; | |
99 | ||
100 | case STATE_OFFLINE: | |
101 | f->header->state = STATE_ONLINE; | |
102 | fsync(f->fd); | |
103 | return 0; | |
104 | ||
105 | default: | |
106 | return -EINVAL; | |
107 | } | |
108 | } | |
109 | ||
110 | int journal_file_set_offline(JournalFile *f) { | |
111 | assert(f); | |
112 | ||
113 | if (!f->writable) | |
114 | return -EPERM; | |
115 | ||
116 | if (!(f->fd >= 0 && f->header)) | |
117 | return -EINVAL; | |
118 | ||
119 | if (f->header->state != STATE_ONLINE) | |
120 | return 0; | |
121 | ||
122 | fsync(f->fd); | |
123 | ||
124 | if (mmap_cache_got_sigbus(f->mmap, f->fd)) | |
125 | return -EIO; | |
126 | ||
127 | f->header->state = STATE_OFFLINE; | |
128 | ||
129 | if (mmap_cache_got_sigbus(f->mmap, f->fd)) | |
130 | return -EIO; | |
131 | ||
132 | fsync(f->fd); | |
133 | ||
134 | return 0; | |
135 | } | |
136 | ||
137 | JournalFile* journal_file_close(JournalFile *f) { | |
138 | assert(f); | |
139 | ||
140 | #ifdef HAVE_GCRYPT | |
141 | /* Write the final tag */ | |
142 | if (f->seal && f->writable) | |
143 | journal_file_append_tag(f); | |
144 | #endif | |
145 | ||
146 | journal_file_set_offline(f); | |
147 | ||
148 | if (f->mmap && f->fd >= 0) | |
149 | mmap_cache_close_fd(f->mmap, f->fd); | |
150 | ||
151 | if (f->fd >= 0 && f->defrag_on_close) { | |
152 | ||
153 | /* Be friendly to btrfs: turn COW back on again now, | |
154 | * and defragment the file. We won't write to the file | |
155 | * ever again, hence remove all fragmentation, and | |
156 | * reenable all the good bits COW usually provides | |
157 | * (such as data checksumming). */ | |
158 | ||
159 | (void) chattr_fd(f->fd, 0, FS_NOCOW_FL); | |
160 | (void) btrfs_defrag_fd(f->fd); | |
161 | } | |
162 | ||
163 | safe_close(f->fd); | |
164 | free(f->path); | |
165 | ||
166 | if (f->mmap) | |
167 | mmap_cache_unref(f->mmap); | |
168 | ||
169 | ordered_hashmap_free_free(f->chain_cache); | |
170 | ||
171 | #if defined(HAVE_XZ) || defined(HAVE_LZ4) | |
172 | free(f->compress_buffer); | |
173 | #endif | |
174 | ||
175 | #ifdef HAVE_GCRYPT | |
176 | if (f->fss_file) | |
177 | munmap(f->fss_file, PAGE_ALIGN(f->fss_file_size)); | |
178 | else | |
179 | free(f->fsprg_state); | |
180 | ||
181 | free(f->fsprg_seed); | |
182 | ||
183 | if (f->hmac) | |
184 | gcry_md_close(f->hmac); | |
185 | #endif | |
186 | ||
187 | free(f); | |
188 | return NULL; | |
189 | } | |
190 | ||
191 | static int journal_file_init_header(JournalFile *f, JournalFile *template) { | |
192 | Header h = {}; | |
193 | ssize_t k; | |
194 | int r; | |
195 | ||
196 | assert(f); | |
197 | ||
198 | memcpy(h.signature, HEADER_SIGNATURE, 8); | |
199 | h.header_size = htole64(ALIGN64(sizeof(h))); | |
200 | ||
201 | h.incompatible_flags |= htole32( | |
202 | f->compress_xz * HEADER_INCOMPATIBLE_COMPRESSED_XZ | | |
203 | f->compress_lz4 * HEADER_INCOMPATIBLE_COMPRESSED_LZ4); | |
204 | ||
205 | h.compatible_flags = htole32( | |
206 | f->seal * HEADER_COMPATIBLE_SEALED); | |
207 | ||
208 | r = sd_id128_randomize(&h.file_id); | |
209 | if (r < 0) | |
210 | return r; | |
211 | ||
212 | if (template) { | |
213 | h.seqnum_id = template->header->seqnum_id; | |
214 | h.tail_entry_seqnum = template->header->tail_entry_seqnum; | |
215 | } else | |
216 | h.seqnum_id = h.file_id; | |
217 | ||
218 | k = pwrite(f->fd, &h, sizeof(h), 0); | |
219 | if (k < 0) | |
220 | return -errno; | |
221 | ||
222 | if (k != sizeof(h)) | |
223 | return -EIO; | |
224 | ||
225 | return 0; | |
226 | } | |
227 | ||
228 | static int journal_file_refresh_header(JournalFile *f) { | |
229 | sd_id128_t boot_id; | |
230 | int r; | |
231 | ||
232 | assert(f); | |
233 | ||
234 | r = sd_id128_get_machine(&f->header->machine_id); | |
235 | if (r < 0) | |
236 | return r; | |
237 | ||
238 | r = sd_id128_get_boot(&boot_id); | |
239 | if (r < 0) | |
240 | return r; | |
241 | ||
242 | if (sd_id128_equal(boot_id, f->header->boot_id)) | |
243 | f->tail_entry_monotonic_valid = true; | |
244 | ||
245 | f->header->boot_id = boot_id; | |
246 | ||
247 | r = journal_file_set_online(f); | |
248 | ||
249 | /* Sync the online state to disk */ | |
250 | fsync(f->fd); | |
251 | ||
252 | return r; | |
253 | } | |
254 | ||
255 | static int journal_file_verify_header(JournalFile *f) { | |
256 | uint32_t flags; | |
257 | ||
258 | assert(f); | |
259 | ||
260 | if (memcmp(f->header->signature, HEADER_SIGNATURE, 8)) | |
261 | return -EBADMSG; | |
262 | ||
263 | /* In both read and write mode we refuse to open files with | |
264 | * incompatible flags we don't know */ | |
265 | flags = le32toh(f->header->incompatible_flags); | |
266 | if (flags & ~HEADER_INCOMPATIBLE_SUPPORTED) { | |
267 | if (flags & ~HEADER_INCOMPATIBLE_ANY) | |
268 | log_debug("Journal file %s has unknown incompatible flags %"PRIx32, | |
269 | f->path, flags & ~HEADER_INCOMPATIBLE_ANY); | |
270 | flags = (flags & HEADER_INCOMPATIBLE_ANY) & ~HEADER_INCOMPATIBLE_SUPPORTED; | |
271 | if (flags) | |
272 | log_debug("Journal file %s uses incompatible flags %"PRIx32 | |
273 | " disabled at compilation time.", f->path, flags); | |
274 | return -EPROTONOSUPPORT; | |
275 | } | |
276 | ||
277 | /* When open for writing we refuse to open files with | |
278 | * compatible flags, too */ | |
279 | flags = le32toh(f->header->compatible_flags); | |
280 | if (f->writable && (flags & ~HEADER_COMPATIBLE_SUPPORTED)) { | |
281 | if (flags & ~HEADER_COMPATIBLE_ANY) | |
282 | log_debug("Journal file %s has unknown compatible flags %"PRIx32, | |
283 | f->path, flags & ~HEADER_COMPATIBLE_ANY); | |
284 | flags = (flags & HEADER_COMPATIBLE_ANY) & ~HEADER_COMPATIBLE_SUPPORTED; | |
285 | if (flags) | |
286 | log_debug("Journal file %s uses compatible flags %"PRIx32 | |
287 | " disabled at compilation time.", f->path, flags); | |
288 | return -EPROTONOSUPPORT; | |
289 | } | |
290 | ||
291 | if (f->header->state >= _STATE_MAX) | |
292 | return -EBADMSG; | |
293 | ||
294 | /* The first addition was n_data, so check that we are at least this large */ | |
295 | if (le64toh(f->header->header_size) < HEADER_SIZE_MIN) | |
296 | return -EBADMSG; | |
297 | ||
298 | if (JOURNAL_HEADER_SEALED(f->header) && !JOURNAL_HEADER_CONTAINS(f->header, n_entry_arrays)) | |
299 | return -EBADMSG; | |
300 | ||
301 | if ((le64toh(f->header->header_size) + le64toh(f->header->arena_size)) > (uint64_t) f->last_stat.st_size) | |
302 | return -ENODATA; | |
303 | ||
304 | if (le64toh(f->header->tail_object_offset) > (le64toh(f->header->header_size) + le64toh(f->header->arena_size))) | |
305 | return -ENODATA; | |
306 | ||
307 | if (!VALID64(le64toh(f->header->data_hash_table_offset)) || | |
308 | !VALID64(le64toh(f->header->field_hash_table_offset)) || | |
309 | !VALID64(le64toh(f->header->tail_object_offset)) || | |
310 | !VALID64(le64toh(f->header->entry_array_offset))) | |
311 | return -ENODATA; | |
312 | ||
313 | if (f->writable) { | |
314 | uint8_t state; | |
315 | sd_id128_t machine_id; | |
316 | int r; | |
317 | ||
318 | r = sd_id128_get_machine(&machine_id); | |
319 | if (r < 0) | |
320 | return r; | |
321 | ||
322 | if (!sd_id128_equal(machine_id, f->header->machine_id)) | |
323 | return -EHOSTDOWN; | |
324 | ||
325 | state = f->header->state; | |
326 | ||
327 | if (state == STATE_ONLINE) { | |
328 | log_debug("Journal file %s is already online. Assuming unclean closing.", f->path); | |
329 | return -EBUSY; | |
330 | } else if (state == STATE_ARCHIVED) | |
331 | return -ESHUTDOWN; | |
332 | else if (state != STATE_OFFLINE) { | |
333 | log_debug("Journal file %s has unknown state %i.", f->path, state); | |
334 | return -EBUSY; | |
335 | } | |
336 | } | |
337 | ||
338 | f->compress_xz = JOURNAL_HEADER_COMPRESSED_XZ(f->header); | |
339 | f->compress_lz4 = JOURNAL_HEADER_COMPRESSED_LZ4(f->header); | |
340 | ||
341 | f->seal = JOURNAL_HEADER_SEALED(f->header); | |
342 | ||
343 | return 0; | |
344 | } | |
345 | ||
346 | static int journal_file_fstat(JournalFile *f) { | |
347 | assert(f); | |
348 | assert(f->fd >= 0); | |
349 | ||
350 | if (fstat(f->fd, &f->last_stat) < 0) | |
351 | return -errno; | |
352 | ||
353 | f->last_stat_usec = now(CLOCK_MONOTONIC); | |
354 | ||
355 | /* Refuse appending to files that are already deleted */ | |
356 | if (f->last_stat.st_nlink <= 0) | |
357 | return -EIDRM; | |
358 | ||
359 | return 0; | |
360 | } | |
361 | ||
362 | static int journal_file_allocate(JournalFile *f, uint64_t offset, uint64_t size) { | |
363 | uint64_t old_size, new_size; | |
364 | int r; | |
365 | ||
366 | assert(f); | |
367 | ||
368 | /* We assume that this file is not sparse, and we know that | |
369 | * for sure, since we always call posix_fallocate() | |
370 | * ourselves */ | |
371 | ||
372 | if (mmap_cache_got_sigbus(f->mmap, f->fd)) | |
373 | return -EIO; | |
374 | ||
375 | old_size = | |
376 | le64toh(f->header->header_size) + | |
377 | le64toh(f->header->arena_size); | |
378 | ||
379 | new_size = PAGE_ALIGN(offset + size); | |
380 | if (new_size < le64toh(f->header->header_size)) | |
381 | new_size = le64toh(f->header->header_size); | |
382 | ||
383 | if (new_size <= old_size) { | |
384 | ||
385 | /* We already pre-allocated enough space, but before | |
386 | * we write to it, let's check with fstat() if the | |
387 | * file got deleted, in order make sure we don't throw | |
388 | * away the data immediately. Don't check fstat() for | |
389 | * all writes though, but only once ever 10s. */ | |
390 | ||
391 | if (f->last_stat_usec + LAST_STAT_REFRESH_USEC > now(CLOCK_MONOTONIC)) | |
392 | return 0; | |
393 | ||
394 | return journal_file_fstat(f); | |
395 | } | |
396 | ||
397 | /* Allocate more space. */ | |
398 | ||
399 | if (f->metrics.max_size > 0 && new_size > f->metrics.max_size) | |
400 | return -E2BIG; | |
401 | ||
402 | if (new_size > f->metrics.min_size && f->metrics.keep_free > 0) { | |
403 | struct statvfs svfs; | |
404 | ||
405 | if (fstatvfs(f->fd, &svfs) >= 0) { | |
406 | uint64_t available; | |
407 | ||
408 | available = LESS_BY((uint64_t) svfs.f_bfree * (uint64_t) svfs.f_bsize, f->metrics.keep_free); | |
409 | ||
410 | if (new_size - old_size > available) | |
411 | return -E2BIG; | |
412 | } | |
413 | } | |
414 | ||
415 | /* Increase by larger blocks at once */ | |
416 | new_size = ((new_size+FILE_SIZE_INCREASE-1) / FILE_SIZE_INCREASE) * FILE_SIZE_INCREASE; | |
417 | if (f->metrics.max_size > 0 && new_size > f->metrics.max_size) | |
418 | new_size = f->metrics.max_size; | |
419 | ||
420 | /* Note that the glibc fallocate() fallback is very | |
421 | inefficient, hence we try to minimize the allocation area | |
422 | as we can. */ | |
423 | r = posix_fallocate(f->fd, old_size, new_size - old_size); | |
424 | if (r != 0) | |
425 | return -r; | |
426 | ||
427 | f->header->arena_size = htole64(new_size - le64toh(f->header->header_size)); | |
428 | ||
429 | return journal_file_fstat(f); | |
430 | } | |
431 | ||
432 | static unsigned type_to_context(ObjectType type) { | |
433 | /* One context for each type, plus one catch-all for the rest */ | |
434 | assert_cc(_OBJECT_TYPE_MAX <= MMAP_CACHE_MAX_CONTEXTS); | |
435 | assert_cc(CONTEXT_HEADER < MMAP_CACHE_MAX_CONTEXTS); | |
436 | return type > OBJECT_UNUSED && type < _OBJECT_TYPE_MAX ? type : 0; | |
437 | } | |
438 | ||
439 | static int journal_file_move_to(JournalFile *f, ObjectType type, bool keep_always, uint64_t offset, uint64_t size, void **ret) { | |
440 | int r; | |
441 | ||
442 | assert(f); | |
443 | assert(ret); | |
444 | ||
445 | if (size <= 0) | |
446 | return -EINVAL; | |
447 | ||
448 | /* Avoid SIGBUS on invalid accesses */ | |
449 | if (offset + size > (uint64_t) f->last_stat.st_size) { | |
450 | /* Hmm, out of range? Let's refresh the fstat() data | |
451 | * first, before we trust that check. */ | |
452 | ||
453 | r = journal_file_fstat(f); | |
454 | if (r < 0) | |
455 | return r; | |
456 | ||
457 | if (offset + size > (uint64_t) f->last_stat.st_size) | |
458 | return -EADDRNOTAVAIL; | |
459 | } | |
460 | ||
461 | return mmap_cache_get(f->mmap, f->fd, f->prot, type_to_context(type), keep_always, offset, size, &f->last_stat, ret); | |
462 | } | |
463 | ||
464 | static uint64_t minimum_header_size(Object *o) { | |
465 | ||
466 | static const uint64_t table[] = { | |
467 | [OBJECT_DATA] = sizeof(DataObject), | |
468 | [OBJECT_FIELD] = sizeof(FieldObject), | |
469 | [OBJECT_ENTRY] = sizeof(EntryObject), | |
470 | [OBJECT_DATA_HASH_TABLE] = sizeof(HashTableObject), | |
471 | [OBJECT_FIELD_HASH_TABLE] = sizeof(HashTableObject), | |
472 | [OBJECT_ENTRY_ARRAY] = sizeof(EntryArrayObject), | |
473 | [OBJECT_TAG] = sizeof(TagObject), | |
474 | }; | |
475 | ||
476 | if (o->object.type >= ELEMENTSOF(table) || table[o->object.type] <= 0) | |
477 | return sizeof(ObjectHeader); | |
478 | ||
479 | return table[o->object.type]; | |
480 | } | |
481 | ||
482 | int journal_file_move_to_object(JournalFile *f, ObjectType type, uint64_t offset, Object **ret) { | |
483 | int r; | |
484 | void *t; | |
485 | Object *o; | |
486 | uint64_t s; | |
487 | ||
488 | assert(f); | |
489 | assert(ret); | |
490 | ||
491 | /* Objects may only be located at multiple of 64 bit */ | |
492 | if (!VALID64(offset)) | |
493 | return -EFAULT; | |
494 | ||
495 | r = journal_file_move_to(f, type, false, offset, sizeof(ObjectHeader), &t); | |
496 | if (r < 0) | |
497 | return r; | |
498 | ||
499 | o = (Object*) t; | |
500 | s = le64toh(o->object.size); | |
501 | ||
502 | if (s < sizeof(ObjectHeader)) | |
503 | return -EBADMSG; | |
504 | ||
505 | if (o->object.type <= OBJECT_UNUSED) | |
506 | return -EBADMSG; | |
507 | ||
508 | if (s < minimum_header_size(o)) | |
509 | return -EBADMSG; | |
510 | ||
511 | if (type > OBJECT_UNUSED && o->object.type != type) | |
512 | return -EBADMSG; | |
513 | ||
514 | if (s > sizeof(ObjectHeader)) { | |
515 | r = journal_file_move_to(f, type, false, offset, s, &t); | |
516 | if (r < 0) | |
517 | return r; | |
518 | ||
519 | o = (Object*) t; | |
520 | } | |
521 | ||
522 | *ret = o; | |
523 | return 0; | |
524 | } | |
525 | ||
526 | static uint64_t journal_file_entry_seqnum(JournalFile *f, uint64_t *seqnum) { | |
527 | uint64_t r; | |
528 | ||
529 | assert(f); | |
530 | ||
531 | r = le64toh(f->header->tail_entry_seqnum) + 1; | |
532 | ||
533 | if (seqnum) { | |
534 | /* If an external seqnum counter was passed, we update | |
535 | * both the local and the external one, and set it to | |
536 | * the maximum of both */ | |
537 | ||
538 | if (*seqnum + 1 > r) | |
539 | r = *seqnum + 1; | |
540 | ||
541 | *seqnum = r; | |
542 | } | |
543 | ||
544 | f->header->tail_entry_seqnum = htole64(r); | |
545 | ||
546 | if (f->header->head_entry_seqnum == 0) | |
547 | f->header->head_entry_seqnum = htole64(r); | |
548 | ||
549 | return r; | |
550 | } | |
551 | ||
552 | int journal_file_append_object(JournalFile *f, ObjectType type, uint64_t size, Object **ret, uint64_t *offset) { | |
553 | int r; | |
554 | uint64_t p; | |
555 | Object *tail, *o; | |
556 | void *t; | |
557 | ||
558 | assert(f); | |
559 | assert(type > OBJECT_UNUSED && type < _OBJECT_TYPE_MAX); | |
560 | assert(size >= sizeof(ObjectHeader)); | |
561 | assert(offset); | |
562 | assert(ret); | |
563 | ||
564 | r = journal_file_set_online(f); | |
565 | if (r < 0) | |
566 | return r; | |
567 | ||
568 | p = le64toh(f->header->tail_object_offset); | |
569 | if (p == 0) | |
570 | p = le64toh(f->header->header_size); | |
571 | else { | |
572 | r = journal_file_move_to_object(f, OBJECT_UNUSED, p, &tail); | |
573 | if (r < 0) | |
574 | return r; | |
575 | ||
576 | p += ALIGN64(le64toh(tail->object.size)); | |
577 | } | |
578 | ||
579 | r = journal_file_allocate(f, p, size); | |
580 | if (r < 0) | |
581 | return r; | |
582 | ||
583 | r = journal_file_move_to(f, type, false, p, size, &t); | |
584 | if (r < 0) | |
585 | return r; | |
586 | ||
587 | o = (Object*) t; | |
588 | ||
589 | zero(o->object); | |
590 | o->object.type = type; | |
591 | o->object.size = htole64(size); | |
592 | ||
593 | f->header->tail_object_offset = htole64(p); | |
594 | f->header->n_objects = htole64(le64toh(f->header->n_objects) + 1); | |
595 | ||
596 | *ret = o; | |
597 | *offset = p; | |
598 | ||
599 | return 0; | |
600 | } | |
601 | ||
602 | static int journal_file_setup_data_hash_table(JournalFile *f) { | |
603 | uint64_t s, p; | |
604 | Object *o; | |
605 | int r; | |
606 | ||
607 | assert(f); | |
608 | ||
609 | /* We estimate that we need 1 hash table entry per 768 bytes | |
610 | of journal file and we want to make sure we never get | |
611 | beyond 75% fill level. Calculate the hash table size for | |
612 | the maximum file size based on these metrics. */ | |
613 | ||
614 | s = (f->metrics.max_size * 4 / 768 / 3) * sizeof(HashItem); | |
615 | if (s < DEFAULT_DATA_HASH_TABLE_SIZE) | |
616 | s = DEFAULT_DATA_HASH_TABLE_SIZE; | |
617 | ||
618 | log_debug("Reserving %"PRIu64" entries in hash table.", s / sizeof(HashItem)); | |
619 | ||
620 | r = journal_file_append_object(f, | |
621 | OBJECT_DATA_HASH_TABLE, | |
622 | offsetof(Object, hash_table.items) + s, | |
623 | &o, &p); | |
624 | if (r < 0) | |
625 | return r; | |
626 | ||
627 | memzero(o->hash_table.items, s); | |
628 | ||
629 | f->header->data_hash_table_offset = htole64(p + offsetof(Object, hash_table.items)); | |
630 | f->header->data_hash_table_size = htole64(s); | |
631 | ||
632 | return 0; | |
633 | } | |
634 | ||
635 | static int journal_file_setup_field_hash_table(JournalFile *f) { | |
636 | uint64_t s, p; | |
637 | Object *o; | |
638 | int r; | |
639 | ||
640 | assert(f); | |
641 | ||
642 | /* We use a fixed size hash table for the fields as this | |
643 | * number should grow very slowly only */ | |
644 | ||
645 | s = DEFAULT_FIELD_HASH_TABLE_SIZE; | |
646 | r = journal_file_append_object(f, | |
647 | OBJECT_FIELD_HASH_TABLE, | |
648 | offsetof(Object, hash_table.items) + s, | |
649 | &o, &p); | |
650 | if (r < 0) | |
651 | return r; | |
652 | ||
653 | memzero(o->hash_table.items, s); | |
654 | ||
655 | f->header->field_hash_table_offset = htole64(p + offsetof(Object, hash_table.items)); | |
656 | f->header->field_hash_table_size = htole64(s); | |
657 | ||
658 | return 0; | |
659 | } | |
660 | ||
661 | int journal_file_map_data_hash_table(JournalFile *f) { | |
662 | uint64_t s, p; | |
663 | void *t; | |
664 | int r; | |
665 | ||
666 | assert(f); | |
667 | ||
668 | if (f->data_hash_table) | |
669 | return 0; | |
670 | ||
671 | p = le64toh(f->header->data_hash_table_offset); | |
672 | s = le64toh(f->header->data_hash_table_size); | |
673 | ||
674 | r = journal_file_move_to(f, | |
675 | OBJECT_DATA_HASH_TABLE, | |
676 | true, | |
677 | p, s, | |
678 | &t); | |
679 | if (r < 0) | |
680 | return r; | |
681 | ||
682 | f->data_hash_table = t; | |
683 | return 0; | |
684 | } | |
685 | ||
686 | int journal_file_map_field_hash_table(JournalFile *f) { | |
687 | uint64_t s, p; | |
688 | void *t; | |
689 | int r; | |
690 | ||
691 | assert(f); | |
692 | ||
693 | if (f->field_hash_table) | |
694 | return 0; | |
695 | ||
696 | p = le64toh(f->header->field_hash_table_offset); | |
697 | s = le64toh(f->header->field_hash_table_size); | |
698 | ||
699 | r = journal_file_move_to(f, | |
700 | OBJECT_FIELD_HASH_TABLE, | |
701 | true, | |
702 | p, s, | |
703 | &t); | |
704 | if (r < 0) | |
705 | return r; | |
706 | ||
707 | f->field_hash_table = t; | |
708 | return 0; | |
709 | } | |
710 | ||
711 | static int journal_file_link_field( | |
712 | JournalFile *f, | |
713 | Object *o, | |
714 | uint64_t offset, | |
715 | uint64_t hash) { | |
716 | ||
717 | uint64_t p, h, m; | |
718 | int r; | |
719 | ||
720 | assert(f); | |
721 | assert(o); | |
722 | assert(offset > 0); | |
723 | ||
724 | if (o->object.type != OBJECT_FIELD) | |
725 | return -EINVAL; | |
726 | ||
727 | m = le64toh(f->header->field_hash_table_size) / sizeof(HashItem); | |
728 | if (m <= 0) | |
729 | return -EBADMSG; | |
730 | ||
731 | /* This might alter the window we are looking at */ | |
732 | o->field.next_hash_offset = o->field.head_data_offset = 0; | |
733 | ||
734 | h = hash % m; | |
735 | p = le64toh(f->field_hash_table[h].tail_hash_offset); | |
736 | if (p == 0) | |
737 | f->field_hash_table[h].head_hash_offset = htole64(offset); | |
738 | else { | |
739 | r = journal_file_move_to_object(f, OBJECT_FIELD, p, &o); | |
740 | if (r < 0) | |
741 | return r; | |
742 | ||
743 | o->field.next_hash_offset = htole64(offset); | |
744 | } | |
745 | ||
746 | f->field_hash_table[h].tail_hash_offset = htole64(offset); | |
747 | ||
748 | if (JOURNAL_HEADER_CONTAINS(f->header, n_fields)) | |
749 | f->header->n_fields = htole64(le64toh(f->header->n_fields) + 1); | |
750 | ||
751 | return 0; | |
752 | } | |
753 | ||
754 | static int journal_file_link_data( | |
755 | JournalFile *f, | |
756 | Object *o, | |
757 | uint64_t offset, | |
758 | uint64_t hash) { | |
759 | ||
760 | uint64_t p, h, m; | |
761 | int r; | |
762 | ||
763 | assert(f); | |
764 | assert(o); | |
765 | assert(offset > 0); | |
766 | ||
767 | if (o->object.type != OBJECT_DATA) | |
768 | return -EINVAL; | |
769 | ||
770 | m = le64toh(f->header->data_hash_table_size) / sizeof(HashItem); | |
771 | if (m <= 0) | |
772 | return -EBADMSG; | |
773 | ||
774 | /* This might alter the window we are looking at */ | |
775 | o->data.next_hash_offset = o->data.next_field_offset = 0; | |
776 | o->data.entry_offset = o->data.entry_array_offset = 0; | |
777 | o->data.n_entries = 0; | |
778 | ||
779 | h = hash % m; | |
780 | p = le64toh(f->data_hash_table[h].tail_hash_offset); | |
781 | if (p == 0) | |
782 | /* Only entry in the hash table is easy */ | |
783 | f->data_hash_table[h].head_hash_offset = htole64(offset); | |
784 | else { | |
785 | /* Move back to the previous data object, to patch in | |
786 | * pointer */ | |
787 | ||
788 | r = journal_file_move_to_object(f, OBJECT_DATA, p, &o); | |
789 | if (r < 0) | |
790 | return r; | |
791 | ||
792 | o->data.next_hash_offset = htole64(offset); | |
793 | } | |
794 | ||
795 | f->data_hash_table[h].tail_hash_offset = htole64(offset); | |
796 | ||
797 | if (JOURNAL_HEADER_CONTAINS(f->header, n_data)) | |
798 | f->header->n_data = htole64(le64toh(f->header->n_data) + 1); | |
799 | ||
800 | return 0; | |
801 | } | |
802 | ||
803 | int journal_file_find_field_object_with_hash( | |
804 | JournalFile *f, | |
805 | const void *field, uint64_t size, uint64_t hash, | |
806 | Object **ret, uint64_t *offset) { | |
807 | ||
808 | uint64_t p, osize, h, m; | |
809 | int r; | |
810 | ||
811 | assert(f); | |
812 | assert(field && size > 0); | |
813 | ||
814 | /* If the field hash table is empty, we can't find anything */ | |
815 | if (le64toh(f->header->field_hash_table_size) <= 0) | |
816 | return 0; | |
817 | ||
818 | /* Map the field hash table, if it isn't mapped yet. */ | |
819 | r = journal_file_map_field_hash_table(f); | |
820 | if (r < 0) | |
821 | return r; | |
822 | ||
823 | osize = offsetof(Object, field.payload) + size; | |
824 | ||
825 | m = le64toh(f->header->field_hash_table_size) / sizeof(HashItem); | |
826 | if (m <= 0) | |
827 | return -EBADMSG; | |
828 | ||
829 | h = hash % m; | |
830 | p = le64toh(f->field_hash_table[h].head_hash_offset); | |
831 | ||
832 | while (p > 0) { | |
833 | Object *o; | |
834 | ||
835 | r = journal_file_move_to_object(f, OBJECT_FIELD, p, &o); | |
836 | if (r < 0) | |
837 | return r; | |
838 | ||
839 | if (le64toh(o->field.hash) == hash && | |
840 | le64toh(o->object.size) == osize && | |
841 | memcmp(o->field.payload, field, size) == 0) { | |
842 | ||
843 | if (ret) | |
844 | *ret = o; | |
845 | if (offset) | |
846 | *offset = p; | |
847 | ||
848 | return 1; | |
849 | } | |
850 | ||
851 | p = le64toh(o->field.next_hash_offset); | |
852 | } | |
853 | ||
854 | return 0; | |
855 | } | |
856 | ||
857 | int journal_file_find_field_object( | |
858 | JournalFile *f, | |
859 | const void *field, uint64_t size, | |
860 | Object **ret, uint64_t *offset) { | |
861 | ||
862 | uint64_t hash; | |
863 | ||
864 | assert(f); | |
865 | assert(field && size > 0); | |
866 | ||
867 | hash = hash64(field, size); | |
868 | ||
869 | return journal_file_find_field_object_with_hash(f, | |
870 | field, size, hash, | |
871 | ret, offset); | |
872 | } | |
873 | ||
874 | int journal_file_find_data_object_with_hash( | |
875 | JournalFile *f, | |
876 | const void *data, uint64_t size, uint64_t hash, | |
877 | Object **ret, uint64_t *offset) { | |
878 | ||
879 | uint64_t p, osize, h, m; | |
880 | int r; | |
881 | ||
882 | assert(f); | |
883 | assert(data || size == 0); | |
884 | ||
885 | /* If there's no data hash table, then there's no entry. */ | |
886 | if (le64toh(f->header->data_hash_table_size) <= 0) | |
887 | return 0; | |
888 | ||
889 | /* Map the data hash table, if it isn't mapped yet. */ | |
890 | r = journal_file_map_data_hash_table(f); | |
891 | if (r < 0) | |
892 | return r; | |
893 | ||
894 | osize = offsetof(Object, data.payload) + size; | |
895 | ||
896 | m = le64toh(f->header->data_hash_table_size) / sizeof(HashItem); | |
897 | if (m <= 0) | |
898 | return -EBADMSG; | |
899 | ||
900 | h = hash % m; | |
901 | p = le64toh(f->data_hash_table[h].head_hash_offset); | |
902 | ||
903 | while (p > 0) { | |
904 | Object *o; | |
905 | ||
906 | r = journal_file_move_to_object(f, OBJECT_DATA, p, &o); | |
907 | if (r < 0) | |
908 | return r; | |
909 | ||
910 | if (le64toh(o->data.hash) != hash) | |
911 | goto next; | |
912 | ||
913 | if (o->object.flags & OBJECT_COMPRESSION_MASK) { | |
914 | #if defined(HAVE_XZ) || defined(HAVE_LZ4) | |
915 | uint64_t l; | |
916 | size_t rsize = 0; | |
917 | ||
918 | l = le64toh(o->object.size); | |
919 | if (l <= offsetof(Object, data.payload)) | |
920 | return -EBADMSG; | |
921 | ||
922 | l -= offsetof(Object, data.payload); | |
923 | ||
924 | r = decompress_blob(o->object.flags & OBJECT_COMPRESSION_MASK, | |
925 | o->data.payload, l, &f->compress_buffer, &f->compress_buffer_size, &rsize, 0); | |
926 | if (r < 0) | |
927 | return r; | |
928 | ||
929 | if (rsize == size && | |
930 | memcmp(f->compress_buffer, data, size) == 0) { | |
931 | ||
932 | if (ret) | |
933 | *ret = o; | |
934 | ||
935 | if (offset) | |
936 | *offset = p; | |
937 | ||
938 | return 1; | |
939 | } | |
940 | #else | |
941 | return -EPROTONOSUPPORT; | |
942 | #endif | |
943 | } else if (le64toh(o->object.size) == osize && | |
944 | memcmp(o->data.payload, data, size) == 0) { | |
945 | ||
946 | if (ret) | |
947 | *ret = o; | |
948 | ||
949 | if (offset) | |
950 | *offset = p; | |
951 | ||
952 | return 1; | |
953 | } | |
954 | ||
955 | next: | |
956 | p = le64toh(o->data.next_hash_offset); | |
957 | } | |
958 | ||
959 | return 0; | |
960 | } | |
961 | ||
962 | int journal_file_find_data_object( | |
963 | JournalFile *f, | |
964 | const void *data, uint64_t size, | |
965 | Object **ret, uint64_t *offset) { | |
966 | ||
967 | uint64_t hash; | |
968 | ||
969 | assert(f); | |
970 | assert(data || size == 0); | |
971 | ||
972 | hash = hash64(data, size); | |
973 | ||
974 | return journal_file_find_data_object_with_hash(f, | |
975 | data, size, hash, | |
976 | ret, offset); | |
977 | } | |
978 | ||
979 | static int journal_file_append_field( | |
980 | JournalFile *f, | |
981 | const void *field, uint64_t size, | |
982 | Object **ret, uint64_t *offset) { | |
983 | ||
984 | uint64_t hash, p; | |
985 | uint64_t osize; | |
986 | Object *o; | |
987 | int r; | |
988 | ||
989 | assert(f); | |
990 | assert(field && size > 0); | |
991 | ||
992 | hash = hash64(field, size); | |
993 | ||
994 | r = journal_file_find_field_object_with_hash(f, field, size, hash, &o, &p); | |
995 | if (r < 0) | |
996 | return r; | |
997 | else if (r > 0) { | |
998 | ||
999 | if (ret) | |
1000 | *ret = o; | |
1001 | ||
1002 | if (offset) | |
1003 | *offset = p; | |
1004 | ||
1005 | return 0; | |
1006 | } | |
1007 | ||
1008 | osize = offsetof(Object, field.payload) + size; | |
1009 | r = journal_file_append_object(f, OBJECT_FIELD, osize, &o, &p); | |
1010 | if (r < 0) | |
1011 | return r; | |
1012 | ||
1013 | o->field.hash = htole64(hash); | |
1014 | memcpy(o->field.payload, field, size); | |
1015 | ||
1016 | r = journal_file_link_field(f, o, p, hash); | |
1017 | if (r < 0) | |
1018 | return r; | |
1019 | ||
1020 | /* The linking might have altered the window, so let's | |
1021 | * refresh our pointer */ | |
1022 | r = journal_file_move_to_object(f, OBJECT_FIELD, p, &o); | |
1023 | if (r < 0) | |
1024 | return r; | |
1025 | ||
1026 | #ifdef HAVE_GCRYPT | |
1027 | r = journal_file_hmac_put_object(f, OBJECT_FIELD, o, p); | |
1028 | if (r < 0) | |
1029 | return r; | |
1030 | #endif | |
1031 | ||
1032 | if (ret) | |
1033 | *ret = o; | |
1034 | ||
1035 | if (offset) | |
1036 | *offset = p; | |
1037 | ||
1038 | return 0; | |
1039 | } | |
1040 | ||
1041 | static int journal_file_append_data( | |
1042 | JournalFile *f, | |
1043 | const void *data, uint64_t size, | |
1044 | Object **ret, uint64_t *offset) { | |
1045 | ||
1046 | uint64_t hash, p; | |
1047 | uint64_t osize; | |
1048 | Object *o; | |
1049 | int r, compression = 0; | |
1050 | const void *eq; | |
1051 | ||
1052 | assert(f); | |
1053 | assert(data || size == 0); | |
1054 | ||
1055 | hash = hash64(data, size); | |
1056 | ||
1057 | r = journal_file_find_data_object_with_hash(f, data, size, hash, &o, &p); | |
1058 | if (r < 0) | |
1059 | return r; | |
1060 | else if (r > 0) { | |
1061 | ||
1062 | if (ret) | |
1063 | *ret = o; | |
1064 | ||
1065 | if (offset) | |
1066 | *offset = p; | |
1067 | ||
1068 | return 0; | |
1069 | } | |
1070 | ||
1071 | osize = offsetof(Object, data.payload) + size; | |
1072 | r = journal_file_append_object(f, OBJECT_DATA, osize, &o, &p); | |
1073 | if (r < 0) | |
1074 | return r; | |
1075 | ||
1076 | o->data.hash = htole64(hash); | |
1077 | ||
1078 | #if defined(HAVE_XZ) || defined(HAVE_LZ4) | |
1079 | if (JOURNAL_FILE_COMPRESS(f) && size >= COMPRESSION_SIZE_THRESHOLD) { | |
1080 | size_t rsize = 0; | |
1081 | ||
1082 | compression = compress_blob(data, size, o->data.payload, &rsize); | |
1083 | ||
1084 | if (compression >= 0) { | |
1085 | o->object.size = htole64(offsetof(Object, data.payload) + rsize); | |
1086 | o->object.flags |= compression; | |
1087 | ||
1088 | log_debug("Compressed data object %"PRIu64" -> %zu using %s", | |
1089 | size, rsize, object_compressed_to_string(compression)); | |
1090 | } else | |
1091 | /* Compression didn't work, we don't really care why, let's continue without compression */ | |
1092 | compression = 0; | |
1093 | } | |
1094 | #endif | |
1095 | ||
1096 | if (compression == 0 && size > 0) | |
1097 | memcpy(o->data.payload, data, size); | |
1098 | ||
1099 | r = journal_file_link_data(f, o, p, hash); | |
1100 | if (r < 0) | |
1101 | return r; | |
1102 | ||
1103 | /* The linking might have altered the window, so let's | |
1104 | * refresh our pointer */ | |
1105 | r = journal_file_move_to_object(f, OBJECT_DATA, p, &o); | |
1106 | if (r < 0) | |
1107 | return r; | |
1108 | ||
1109 | if (!data) | |
1110 | eq = NULL; | |
1111 | else | |
1112 | eq = memchr(data, '=', size); | |
1113 | if (eq && eq > data) { | |
1114 | Object *fo = NULL; | |
1115 | uint64_t fp; | |
1116 | ||
1117 | /* Create field object ... */ | |
1118 | r = journal_file_append_field(f, data, (uint8_t*) eq - (uint8_t*) data, &fo, &fp); | |
1119 | if (r < 0) | |
1120 | return r; | |
1121 | ||
1122 | /* ... and link it in. */ | |
1123 | o->data.next_field_offset = fo->field.head_data_offset; | |
1124 | fo->field.head_data_offset = le64toh(p); | |
1125 | } | |
1126 | ||
1127 | #ifdef HAVE_GCRYPT | |
1128 | r = journal_file_hmac_put_object(f, OBJECT_DATA, o, p); | |
1129 | if (r < 0) | |
1130 | return r; | |
1131 | #endif | |
1132 | ||
1133 | if (ret) | |
1134 | *ret = o; | |
1135 | ||
1136 | if (offset) | |
1137 | *offset = p; | |
1138 | ||
1139 | return 0; | |
1140 | } | |
1141 | ||
1142 | uint64_t journal_file_entry_n_items(Object *o) { | |
1143 | assert(o); | |
1144 | ||
1145 | if (o->object.type != OBJECT_ENTRY) | |
1146 | return 0; | |
1147 | ||
1148 | return (le64toh(o->object.size) - offsetof(Object, entry.items)) / sizeof(EntryItem); | |
1149 | } | |
1150 | ||
1151 | uint64_t journal_file_entry_array_n_items(Object *o) { | |
1152 | assert(o); | |
1153 | ||
1154 | if (o->object.type != OBJECT_ENTRY_ARRAY) | |
1155 | return 0; | |
1156 | ||
1157 | return (le64toh(o->object.size) - offsetof(Object, entry_array.items)) / sizeof(uint64_t); | |
1158 | } | |
1159 | ||
1160 | uint64_t journal_file_hash_table_n_items(Object *o) { | |
1161 | assert(o); | |
1162 | ||
1163 | if (o->object.type != OBJECT_DATA_HASH_TABLE && | |
1164 | o->object.type != OBJECT_FIELD_HASH_TABLE) | |
1165 | return 0; | |
1166 | ||
1167 | return (le64toh(o->object.size) - offsetof(Object, hash_table.items)) / sizeof(HashItem); | |
1168 | } | |
1169 | ||
1170 | static int link_entry_into_array(JournalFile *f, | |
1171 | le64_t *first, | |
1172 | le64_t *idx, | |
1173 | uint64_t p) { | |
1174 | int r; | |
1175 | uint64_t n = 0, ap = 0, q, i, a, hidx; | |
1176 | Object *o; | |
1177 | ||
1178 | assert(f); | |
1179 | assert(first); | |
1180 | assert(idx); | |
1181 | assert(p > 0); | |
1182 | ||
1183 | a = le64toh(*first); | |
1184 | i = hidx = le64toh(*idx); | |
1185 | while (a > 0) { | |
1186 | ||
1187 | r = journal_file_move_to_object(f, OBJECT_ENTRY_ARRAY, a, &o); | |
1188 | if (r < 0) | |
1189 | return r; | |
1190 | ||
1191 | n = journal_file_entry_array_n_items(o); | |
1192 | if (i < n) { | |
1193 | o->entry_array.items[i] = htole64(p); | |
1194 | *idx = htole64(hidx + 1); | |
1195 | return 0; | |
1196 | } | |
1197 | ||
1198 | i -= n; | |
1199 | ap = a; | |
1200 | a = le64toh(o->entry_array.next_entry_array_offset); | |
1201 | } | |
1202 | ||
1203 | if (hidx > n) | |
1204 | n = (hidx+1) * 2; | |
1205 | else | |
1206 | n = n * 2; | |
1207 | ||
1208 | if (n < 4) | |
1209 | n = 4; | |
1210 | ||
1211 | r = journal_file_append_object(f, OBJECT_ENTRY_ARRAY, | |
1212 | offsetof(Object, entry_array.items) + n * sizeof(uint64_t), | |
1213 | &o, &q); | |
1214 | if (r < 0) | |
1215 | return r; | |
1216 | ||
1217 | #ifdef HAVE_GCRYPT | |
1218 | r = journal_file_hmac_put_object(f, OBJECT_ENTRY_ARRAY, o, q); | |
1219 | if (r < 0) | |
1220 | return r; | |
1221 | #endif | |
1222 | ||
1223 | o->entry_array.items[i] = htole64(p); | |
1224 | ||
1225 | if (ap == 0) | |
1226 | *first = htole64(q); | |
1227 | else { | |
1228 | r = journal_file_move_to_object(f, OBJECT_ENTRY_ARRAY, ap, &o); | |
1229 | if (r < 0) | |
1230 | return r; | |
1231 | ||
1232 | o->entry_array.next_entry_array_offset = htole64(q); | |
1233 | } | |
1234 | ||
1235 | if (JOURNAL_HEADER_CONTAINS(f->header, n_entry_arrays)) | |
1236 | f->header->n_entry_arrays = htole64(le64toh(f->header->n_entry_arrays) + 1); | |
1237 | ||
1238 | *idx = htole64(hidx + 1); | |
1239 | ||
1240 | return 0; | |
1241 | } | |
1242 | ||
1243 | static int link_entry_into_array_plus_one(JournalFile *f, | |
1244 | le64_t *extra, | |
1245 | le64_t *first, | |
1246 | le64_t *idx, | |
1247 | uint64_t p) { | |
1248 | ||
1249 | int r; | |
1250 | ||
1251 | assert(f); | |
1252 | assert(extra); | |
1253 | assert(first); | |
1254 | assert(idx); | |
1255 | assert(p > 0); | |
1256 | ||
1257 | if (*idx == 0) | |
1258 | *extra = htole64(p); | |
1259 | else { | |
1260 | le64_t i; | |
1261 | ||
1262 | i = htole64(le64toh(*idx) - 1); | |
1263 | r = link_entry_into_array(f, first, &i, p); | |
1264 | if (r < 0) | |
1265 | return r; | |
1266 | } | |
1267 | ||
1268 | *idx = htole64(le64toh(*idx) + 1); | |
1269 | return 0; | |
1270 | } | |
1271 | ||
1272 | static int journal_file_link_entry_item(JournalFile *f, Object *o, uint64_t offset, uint64_t i) { | |
1273 | uint64_t p; | |
1274 | int r; | |
1275 | assert(f); | |
1276 | assert(o); | |
1277 | assert(offset > 0); | |
1278 | ||
1279 | p = le64toh(o->entry.items[i].object_offset); | |
1280 | if (p == 0) | |
1281 | return -EINVAL; | |
1282 | ||
1283 | r = journal_file_move_to_object(f, OBJECT_DATA, p, &o); | |
1284 | if (r < 0) | |
1285 | return r; | |
1286 | ||
1287 | return link_entry_into_array_plus_one(f, | |
1288 | &o->data.entry_offset, | |
1289 | &o->data.entry_array_offset, | |
1290 | &o->data.n_entries, | |
1291 | offset); | |
1292 | } | |
1293 | ||
1294 | static int journal_file_link_entry(JournalFile *f, Object *o, uint64_t offset) { | |
1295 | uint64_t n, i; | |
1296 | int r; | |
1297 | ||
1298 | assert(f); | |
1299 | assert(o); | |
1300 | assert(offset > 0); | |
1301 | ||
1302 | if (o->object.type != OBJECT_ENTRY) | |
1303 | return -EINVAL; | |
1304 | ||
1305 | __sync_synchronize(); | |
1306 | ||
1307 | /* Link up the entry itself */ | |
1308 | r = link_entry_into_array(f, | |
1309 | &f->header->entry_array_offset, | |
1310 | &f->header->n_entries, | |
1311 | offset); | |
1312 | if (r < 0) | |
1313 | return r; | |
1314 | ||
1315 | /* log_debug("=> %s seqnr=%"PRIu64" n_entries=%"PRIu64, f->path, o->entry.seqnum, f->header->n_entries); */ | |
1316 | ||
1317 | if (f->header->head_entry_realtime == 0) | |
1318 | f->header->head_entry_realtime = o->entry.realtime; | |
1319 | ||
1320 | f->header->tail_entry_realtime = o->entry.realtime; | |
1321 | f->header->tail_entry_monotonic = o->entry.monotonic; | |
1322 | ||
1323 | f->tail_entry_monotonic_valid = true; | |
1324 | ||
1325 | /* Link up the items */ | |
1326 | n = journal_file_entry_n_items(o); | |
1327 | for (i = 0; i < n; i++) { | |
1328 | r = journal_file_link_entry_item(f, o, offset, i); | |
1329 | if (r < 0) | |
1330 | return r; | |
1331 | } | |
1332 | ||
1333 | return 0; | |
1334 | } | |
1335 | ||
1336 | static int journal_file_append_entry_internal( | |
1337 | JournalFile *f, | |
1338 | const dual_timestamp *ts, | |
1339 | uint64_t xor_hash, | |
1340 | const EntryItem items[], unsigned n_items, | |
1341 | uint64_t *seqnum, | |
1342 | Object **ret, uint64_t *offset) { | |
1343 | uint64_t np; | |
1344 | uint64_t osize; | |
1345 | Object *o; | |
1346 | int r; | |
1347 | ||
1348 | assert(f); | |
1349 | assert(items || n_items == 0); | |
1350 | assert(ts); | |
1351 | ||
1352 | osize = offsetof(Object, entry.items) + (n_items * sizeof(EntryItem)); | |
1353 | ||
1354 | r = journal_file_append_object(f, OBJECT_ENTRY, osize, &o, &np); | |
1355 | if (r < 0) | |
1356 | return r; | |
1357 | ||
1358 | o->entry.seqnum = htole64(journal_file_entry_seqnum(f, seqnum)); | |
1359 | memcpy(o->entry.items, items, n_items * sizeof(EntryItem)); | |
1360 | o->entry.realtime = htole64(ts->realtime); | |
1361 | o->entry.monotonic = htole64(ts->monotonic); | |
1362 | o->entry.xor_hash = htole64(xor_hash); | |
1363 | o->entry.boot_id = f->header->boot_id; | |
1364 | ||
1365 | #ifdef HAVE_GCRYPT | |
1366 | r = journal_file_hmac_put_object(f, OBJECT_ENTRY, o, np); | |
1367 | if (r < 0) | |
1368 | return r; | |
1369 | #endif | |
1370 | ||
1371 | r = journal_file_link_entry(f, o, np); | |
1372 | if (r < 0) | |
1373 | return r; | |
1374 | ||
1375 | if (ret) | |
1376 | *ret = o; | |
1377 | ||
1378 | if (offset) | |
1379 | *offset = np; | |
1380 | ||
1381 | return 0; | |
1382 | } | |
1383 | ||
1384 | void journal_file_post_change(JournalFile *f) { | |
1385 | assert(f); | |
1386 | ||
1387 | /* inotify() does not receive IN_MODIFY events from file | |
1388 | * accesses done via mmap(). After each access we hence | |
1389 | * trigger IN_MODIFY by truncating the journal file to its | |
1390 | * current size which triggers IN_MODIFY. */ | |
1391 | ||
1392 | __sync_synchronize(); | |
1393 | ||
1394 | if (ftruncate(f->fd, f->last_stat.st_size) < 0) | |
1395 | log_error_errno(errno, "Failed to truncate file to its own size: %m"); | |
1396 | } | |
1397 | ||
1398 | static int entry_item_cmp(const void *_a, const void *_b) { | |
1399 | const EntryItem *a = _a, *b = _b; | |
1400 | ||
1401 | if (le64toh(a->object_offset) < le64toh(b->object_offset)) | |
1402 | return -1; | |
1403 | if (le64toh(a->object_offset) > le64toh(b->object_offset)) | |
1404 | return 1; | |
1405 | return 0; | |
1406 | } | |
1407 | ||
1408 | 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) { | |
1409 | unsigned i; | |
1410 | EntryItem *items; | |
1411 | int r; | |
1412 | uint64_t xor_hash = 0; | |
1413 | struct dual_timestamp _ts; | |
1414 | ||
1415 | assert(f); | |
1416 | assert(iovec || n_iovec == 0); | |
1417 | ||
1418 | if (!ts) { | |
1419 | dual_timestamp_get(&_ts); | |
1420 | ts = &_ts; | |
1421 | } | |
1422 | ||
1423 | if (f->tail_entry_monotonic_valid && | |
1424 | ts->monotonic < le64toh(f->header->tail_entry_monotonic)) | |
1425 | return -EINVAL; | |
1426 | ||
1427 | #ifdef HAVE_GCRYPT | |
1428 | r = journal_file_maybe_append_tag(f, ts->realtime); | |
1429 | if (r < 0) | |
1430 | return r; | |
1431 | #endif | |
1432 | ||
1433 | /* alloca() can't take 0, hence let's allocate at least one */ | |
1434 | items = alloca(sizeof(EntryItem) * MAX(1u, n_iovec)); | |
1435 | ||
1436 | for (i = 0; i < n_iovec; i++) { | |
1437 | uint64_t p; | |
1438 | Object *o; | |
1439 | ||
1440 | r = journal_file_append_data(f, iovec[i].iov_base, iovec[i].iov_len, &o, &p); | |
1441 | if (r < 0) | |
1442 | return r; | |
1443 | ||
1444 | xor_hash ^= le64toh(o->data.hash); | |
1445 | items[i].object_offset = htole64(p); | |
1446 | items[i].hash = o->data.hash; | |
1447 | } | |
1448 | ||
1449 | /* Order by the position on disk, in order to improve seek | |
1450 | * times for rotating media. */ | |
1451 | qsort_safe(items, n_iovec, sizeof(EntryItem), entry_item_cmp); | |
1452 | ||
1453 | r = journal_file_append_entry_internal(f, ts, xor_hash, items, n_iovec, seqnum, ret, offset); | |
1454 | ||
1455 | /* If the memory mapping triggered a SIGBUS then we return an | |
1456 | * IO error and ignore the error code passed down to us, since | |
1457 | * it is very likely just an effect of a nullified replacement | |
1458 | * mapping page */ | |
1459 | ||
1460 | if (mmap_cache_got_sigbus(f->mmap, f->fd)) | |
1461 | r = -EIO; | |
1462 | ||
1463 | journal_file_post_change(f); | |
1464 | ||
1465 | return r; | |
1466 | } | |
1467 | ||
1468 | typedef struct ChainCacheItem { | |
1469 | uint64_t first; /* the array at the beginning of the chain */ | |
1470 | uint64_t array; /* the cached array */ | |
1471 | uint64_t begin; /* the first item in the cached array */ | |
1472 | uint64_t total; /* the total number of items in all arrays before this one in the chain */ | |
1473 | uint64_t last_index; /* the last index we looked at, to optimize locality when bisecting */ | |
1474 | } ChainCacheItem; | |
1475 | ||
1476 | static void chain_cache_put( | |
1477 | OrderedHashmap *h, | |
1478 | ChainCacheItem *ci, | |
1479 | uint64_t first, | |
1480 | uint64_t array, | |
1481 | uint64_t begin, | |
1482 | uint64_t total, | |
1483 | uint64_t last_index) { | |
1484 | ||
1485 | if (!ci) { | |
1486 | /* If the chain item to cache for this chain is the | |
1487 | * first one it's not worth caching anything */ | |
1488 | if (array == first) | |
1489 | return; | |
1490 | ||
1491 | if (ordered_hashmap_size(h) >= CHAIN_CACHE_MAX) { | |
1492 | ci = ordered_hashmap_steal_first(h); | |
1493 | assert(ci); | |
1494 | } else { | |
1495 | ci = new(ChainCacheItem, 1); | |
1496 | if (!ci) | |
1497 | return; | |
1498 | } | |
1499 | ||
1500 | ci->first = first; | |
1501 | ||
1502 | if (ordered_hashmap_put(h, &ci->first, ci) < 0) { | |
1503 | free(ci); | |
1504 | return; | |
1505 | } | |
1506 | } else | |
1507 | assert(ci->first == first); | |
1508 | ||
1509 | ci->array = array; | |
1510 | ci->begin = begin; | |
1511 | ci->total = total; | |
1512 | ci->last_index = last_index; | |
1513 | } | |
1514 | ||
1515 | static int generic_array_get( | |
1516 | JournalFile *f, | |
1517 | uint64_t first, | |
1518 | uint64_t i, | |
1519 | Object **ret, uint64_t *offset) { | |
1520 | ||
1521 | Object *o; | |
1522 | uint64_t p = 0, a, t = 0; | |
1523 | int r; | |
1524 | ChainCacheItem *ci; | |
1525 | ||
1526 | assert(f); | |
1527 | ||
1528 | a = first; | |
1529 | ||
1530 | /* Try the chain cache first */ | |
1531 | ci = ordered_hashmap_get(f->chain_cache, &first); | |
1532 | if (ci && i > ci->total) { | |
1533 | a = ci->array; | |
1534 | i -= ci->total; | |
1535 | t = ci->total; | |
1536 | } | |
1537 | ||
1538 | while (a > 0) { | |
1539 | uint64_t k; | |
1540 | ||
1541 | r = journal_file_move_to_object(f, OBJECT_ENTRY_ARRAY, a, &o); | |
1542 | if (r < 0) | |
1543 | return r; | |
1544 | ||
1545 | k = journal_file_entry_array_n_items(o); | |
1546 | if (i < k) { | |
1547 | p = le64toh(o->entry_array.items[i]); | |
1548 | goto found; | |
1549 | } | |
1550 | ||
1551 | i -= k; | |
1552 | t += k; | |
1553 | a = le64toh(o->entry_array.next_entry_array_offset); | |
1554 | } | |
1555 | ||
1556 | return 0; | |
1557 | ||
1558 | found: | |
1559 | /* Let's cache this item for the next invocation */ | |
1560 | chain_cache_put(f->chain_cache, ci, first, a, le64toh(o->entry_array.items[0]), t, i); | |
1561 | ||
1562 | r = journal_file_move_to_object(f, OBJECT_ENTRY, p, &o); | |
1563 | if (r < 0) | |
1564 | return r; | |
1565 | ||
1566 | if (ret) | |
1567 | *ret = o; | |
1568 | ||
1569 | if (offset) | |
1570 | *offset = p; | |
1571 | ||
1572 | return 1; | |
1573 | } | |
1574 | ||
1575 | static int generic_array_get_plus_one( | |
1576 | JournalFile *f, | |
1577 | uint64_t extra, | |
1578 | uint64_t first, | |
1579 | uint64_t i, | |
1580 | Object **ret, uint64_t *offset) { | |
1581 | ||
1582 | Object *o; | |
1583 | ||
1584 | assert(f); | |
1585 | ||
1586 | if (i == 0) { | |
1587 | int r; | |
1588 | ||
1589 | r = journal_file_move_to_object(f, OBJECT_ENTRY, extra, &o); | |
1590 | if (r < 0) | |
1591 | return r; | |
1592 | ||
1593 | if (ret) | |
1594 | *ret = o; | |
1595 | ||
1596 | if (offset) | |
1597 | *offset = extra; | |
1598 | ||
1599 | return 1; | |
1600 | } | |
1601 | ||
1602 | return generic_array_get(f, first, i-1, ret, offset); | |
1603 | } | |
1604 | ||
1605 | enum { | |
1606 | TEST_FOUND, | |
1607 | TEST_LEFT, | |
1608 | TEST_RIGHT | |
1609 | }; | |
1610 | ||
1611 | static int generic_array_bisect( | |
1612 | JournalFile *f, | |
1613 | uint64_t first, | |
1614 | uint64_t n, | |
1615 | uint64_t needle, | |
1616 | int (*test_object)(JournalFile *f, uint64_t p, uint64_t needle), | |
1617 | direction_t direction, | |
1618 | Object **ret, | |
1619 | uint64_t *offset, | |
1620 | uint64_t *idx) { | |
1621 | ||
1622 | uint64_t a, p, t = 0, i = 0, last_p = 0, last_index = (uint64_t) -1; | |
1623 | bool subtract_one = false; | |
1624 | Object *o, *array = NULL; | |
1625 | int r; | |
1626 | ChainCacheItem *ci; | |
1627 | ||
1628 | assert(f); | |
1629 | assert(test_object); | |
1630 | ||
1631 | /* Start with the first array in the chain */ | |
1632 | a = first; | |
1633 | ||
1634 | ci = ordered_hashmap_get(f->chain_cache, &first); | |
1635 | if (ci && n > ci->total) { | |
1636 | /* Ah, we have iterated this bisection array chain | |
1637 | * previously! Let's see if we can skip ahead in the | |
1638 | * chain, as far as the last time. But we can't jump | |
1639 | * backwards in the chain, so let's check that | |
1640 | * first. */ | |
1641 | ||
1642 | r = test_object(f, ci->begin, needle); | |
1643 | if (r < 0) | |
1644 | return r; | |
1645 | ||
1646 | if (r == TEST_LEFT) { | |
1647 | /* OK, what we are looking for is right of the | |
1648 | * begin of this EntryArray, so let's jump | |
1649 | * straight to previously cached array in the | |
1650 | * chain */ | |
1651 | ||
1652 | a = ci->array; | |
1653 | n -= ci->total; | |
1654 | t = ci->total; | |
1655 | last_index = ci->last_index; | |
1656 | } | |
1657 | } | |
1658 | ||
1659 | while (a > 0) { | |
1660 | uint64_t left, right, k, lp; | |
1661 | ||
1662 | r = journal_file_move_to_object(f, OBJECT_ENTRY_ARRAY, a, &array); | |
1663 | if (r < 0) | |
1664 | return r; | |
1665 | ||
1666 | k = journal_file_entry_array_n_items(array); | |
1667 | right = MIN(k, n); | |
1668 | if (right <= 0) | |
1669 | return 0; | |
1670 | ||
1671 | i = right - 1; | |
1672 | lp = p = le64toh(array->entry_array.items[i]); | |
1673 | if (p <= 0) | |
1674 | return -EBADMSG; | |
1675 | ||
1676 | r = test_object(f, p, needle); | |
1677 | if (r < 0) | |
1678 | return r; | |
1679 | ||
1680 | if (r == TEST_FOUND) | |
1681 | r = direction == DIRECTION_DOWN ? TEST_RIGHT : TEST_LEFT; | |
1682 | ||
1683 | if (r == TEST_RIGHT) { | |
1684 | left = 0; | |
1685 | right -= 1; | |
1686 | ||
1687 | if (last_index != (uint64_t) -1) { | |
1688 | assert(last_index <= right); | |
1689 | ||
1690 | /* If we cached the last index we | |
1691 | * looked at, let's try to not to jump | |
1692 | * too wildly around and see if we can | |
1693 | * limit the range to look at early to | |
1694 | * the immediate neighbors of the last | |
1695 | * index we looked at. */ | |
1696 | ||
1697 | if (last_index > 0) { | |
1698 | uint64_t x = last_index - 1; | |
1699 | ||
1700 | p = le64toh(array->entry_array.items[x]); | |
1701 | if (p <= 0) | |
1702 | return -EBADMSG; | |
1703 | ||
1704 | r = test_object(f, p, needle); | |
1705 | if (r < 0) | |
1706 | return r; | |
1707 | ||
1708 | if (r == TEST_FOUND) | |
1709 | r = direction == DIRECTION_DOWN ? TEST_RIGHT : TEST_LEFT; | |
1710 | ||
1711 | if (r == TEST_RIGHT) | |
1712 | right = x; | |
1713 | else | |
1714 | left = x + 1; | |
1715 | } | |
1716 | ||
1717 | if (last_index < right) { | |
1718 | uint64_t y = last_index + 1; | |
1719 | ||
1720 | p = le64toh(array->entry_array.items[y]); | |
1721 | if (p <= 0) | |
1722 | return -EBADMSG; | |
1723 | ||
1724 | r = test_object(f, p, needle); | |
1725 | if (r < 0) | |
1726 | return r; | |
1727 | ||
1728 | if (r == TEST_FOUND) | |
1729 | r = direction == DIRECTION_DOWN ? TEST_RIGHT : TEST_LEFT; | |
1730 | ||
1731 | if (r == TEST_RIGHT) | |
1732 | right = y; | |
1733 | else | |
1734 | left = y + 1; | |
1735 | } | |
1736 | } | |
1737 | ||
1738 | for (;;) { | |
1739 | if (left == right) { | |
1740 | if (direction == DIRECTION_UP) | |
1741 | subtract_one = true; | |
1742 | ||
1743 | i = left; | |
1744 | goto found; | |
1745 | } | |
1746 | ||
1747 | assert(left < right); | |
1748 | i = (left + right) / 2; | |
1749 | ||
1750 | p = le64toh(array->entry_array.items[i]); | |
1751 | if (p <= 0) | |
1752 | return -EBADMSG; | |
1753 | ||
1754 | r = test_object(f, p, needle); | |
1755 | if (r < 0) | |
1756 | return r; | |
1757 | ||
1758 | if (r == TEST_FOUND) | |
1759 | r = direction == DIRECTION_DOWN ? TEST_RIGHT : TEST_LEFT; | |
1760 | ||
1761 | if (r == TEST_RIGHT) | |
1762 | right = i; | |
1763 | else | |
1764 | left = i + 1; | |
1765 | } | |
1766 | } | |
1767 | ||
1768 | if (k >= n) { | |
1769 | if (direction == DIRECTION_UP) { | |
1770 | i = n; | |
1771 | subtract_one = true; | |
1772 | goto found; | |
1773 | } | |
1774 | ||
1775 | return 0; | |
1776 | } | |
1777 | ||
1778 | last_p = lp; | |
1779 | ||
1780 | n -= k; | |
1781 | t += k; | |
1782 | last_index = (uint64_t) -1; | |
1783 | a = le64toh(array->entry_array.next_entry_array_offset); | |
1784 | } | |
1785 | ||
1786 | return 0; | |
1787 | ||
1788 | found: | |
1789 | if (subtract_one && t == 0 && i == 0) | |
1790 | return 0; | |
1791 | ||
1792 | /* Let's cache this item for the next invocation */ | |
1793 | chain_cache_put(f->chain_cache, ci, first, a, le64toh(array->entry_array.items[0]), t, subtract_one ? (i > 0 ? i-1 : (uint64_t) -1) : i); | |
1794 | ||
1795 | if (subtract_one && i == 0) | |
1796 | p = last_p; | |
1797 | else if (subtract_one) | |
1798 | p = le64toh(array->entry_array.items[i-1]); | |
1799 | else | |
1800 | p = le64toh(array->entry_array.items[i]); | |
1801 | ||
1802 | r = journal_file_move_to_object(f, OBJECT_ENTRY, p, &o); | |
1803 | if (r < 0) | |
1804 | return r; | |
1805 | ||
1806 | if (ret) | |
1807 | *ret = o; | |
1808 | ||
1809 | if (offset) | |
1810 | *offset = p; | |
1811 | ||
1812 | if (idx) | |
1813 | *idx = t + i + (subtract_one ? -1 : 0); | |
1814 | ||
1815 | return 1; | |
1816 | } | |
1817 | ||
1818 | static int generic_array_bisect_plus_one( | |
1819 | JournalFile *f, | |
1820 | uint64_t extra, | |
1821 | uint64_t first, | |
1822 | uint64_t n, | |
1823 | uint64_t needle, | |
1824 | int (*test_object)(JournalFile *f, uint64_t p, uint64_t needle), | |
1825 | direction_t direction, | |
1826 | Object **ret, | |
1827 | uint64_t *offset, | |
1828 | uint64_t *idx) { | |
1829 | ||
1830 | int r; | |
1831 | bool step_back = false; | |
1832 | Object *o; | |
1833 | ||
1834 | assert(f); | |
1835 | assert(test_object); | |
1836 | ||
1837 | if (n <= 0) | |
1838 | return 0; | |
1839 | ||
1840 | /* This bisects the array in object 'first', but first checks | |
1841 | * an extra */ | |
1842 | r = test_object(f, extra, needle); | |
1843 | if (r < 0) | |
1844 | return r; | |
1845 | ||
1846 | if (r == TEST_FOUND) | |
1847 | r = direction == DIRECTION_DOWN ? TEST_RIGHT : TEST_LEFT; | |
1848 | ||
1849 | /* if we are looking with DIRECTION_UP then we need to first | |
1850 | see if in the actual array there is a matching entry, and | |
1851 | return the last one of that. But if there isn't any we need | |
1852 | to return this one. Hence remember this, and return it | |
1853 | below. */ | |
1854 | if (r == TEST_LEFT) | |
1855 | step_back = direction == DIRECTION_UP; | |
1856 | ||
1857 | if (r == TEST_RIGHT) { | |
1858 | if (direction == DIRECTION_DOWN) | |
1859 | goto found; | |
1860 | else | |
1861 | return 0; | |
1862 | } | |
1863 | ||
1864 | r = generic_array_bisect(f, first, n-1, needle, test_object, direction, ret, offset, idx); | |
1865 | ||
1866 | if (r == 0 && step_back) | |
1867 | goto found; | |
1868 | ||
1869 | if (r > 0 && idx) | |
1870 | (*idx) ++; | |
1871 | ||
1872 | return r; | |
1873 | ||
1874 | found: | |
1875 | r = journal_file_move_to_object(f, OBJECT_ENTRY, extra, &o); | |
1876 | if (r < 0) | |
1877 | return r; | |
1878 | ||
1879 | if (ret) | |
1880 | *ret = o; | |
1881 | ||
1882 | if (offset) | |
1883 | *offset = extra; | |
1884 | ||
1885 | if (idx) | |
1886 | *idx = 0; | |
1887 | ||
1888 | return 1; | |
1889 | } | |
1890 | ||
1891 | _pure_ static int test_object_offset(JournalFile *f, uint64_t p, uint64_t needle) { | |
1892 | assert(f); | |
1893 | assert(p > 0); | |
1894 | ||
1895 | if (p == needle) | |
1896 | return TEST_FOUND; | |
1897 | else if (p < needle) | |
1898 | return TEST_LEFT; | |
1899 | else | |
1900 | return TEST_RIGHT; | |
1901 | } | |
1902 | ||
1903 | static int test_object_seqnum(JournalFile *f, uint64_t p, uint64_t needle) { | |
1904 | Object *o; | |
1905 | int r; | |
1906 | ||
1907 | assert(f); | |
1908 | assert(p > 0); | |
1909 | ||
1910 | r = journal_file_move_to_object(f, OBJECT_ENTRY, p, &o); | |
1911 | if (r < 0) | |
1912 | return r; | |
1913 | ||
1914 | if (le64toh(o->entry.seqnum) == needle) | |
1915 | return TEST_FOUND; | |
1916 | else if (le64toh(o->entry.seqnum) < needle) | |
1917 | return TEST_LEFT; | |
1918 | else | |
1919 | return TEST_RIGHT; | |
1920 | } | |
1921 | ||
1922 | int journal_file_move_to_entry_by_seqnum( | |
1923 | JournalFile *f, | |
1924 | uint64_t seqnum, | |
1925 | direction_t direction, | |
1926 | Object **ret, | |
1927 | uint64_t *offset) { | |
1928 | ||
1929 | return generic_array_bisect(f, | |
1930 | le64toh(f->header->entry_array_offset), | |
1931 | le64toh(f->header->n_entries), | |
1932 | seqnum, | |
1933 | test_object_seqnum, | |
1934 | direction, | |
1935 | ret, offset, NULL); | |
1936 | } | |
1937 | ||
1938 | static int test_object_realtime(JournalFile *f, uint64_t p, uint64_t needle) { | |
1939 | Object *o; | |
1940 | int r; | |
1941 | ||
1942 | assert(f); | |
1943 | assert(p > 0); | |
1944 | ||
1945 | r = journal_file_move_to_object(f, OBJECT_ENTRY, p, &o); | |
1946 | if (r < 0) | |
1947 | return r; | |
1948 | ||
1949 | if (le64toh(o->entry.realtime) == needle) | |
1950 | return TEST_FOUND; | |
1951 | else if (le64toh(o->entry.realtime) < needle) | |
1952 | return TEST_LEFT; | |
1953 | else | |
1954 | return TEST_RIGHT; | |
1955 | } | |
1956 | ||
1957 | int journal_file_move_to_entry_by_realtime( | |
1958 | JournalFile *f, | |
1959 | uint64_t realtime, | |
1960 | direction_t direction, | |
1961 | Object **ret, | |
1962 | uint64_t *offset) { | |
1963 | ||
1964 | return generic_array_bisect(f, | |
1965 | le64toh(f->header->entry_array_offset), | |
1966 | le64toh(f->header->n_entries), | |
1967 | realtime, | |
1968 | test_object_realtime, | |
1969 | direction, | |
1970 | ret, offset, NULL); | |
1971 | } | |
1972 | ||
1973 | static int test_object_monotonic(JournalFile *f, uint64_t p, uint64_t needle) { | |
1974 | Object *o; | |
1975 | int r; | |
1976 | ||
1977 | assert(f); | |
1978 | assert(p > 0); | |
1979 | ||
1980 | r = journal_file_move_to_object(f, OBJECT_ENTRY, p, &o); | |
1981 | if (r < 0) | |
1982 | return r; | |
1983 | ||
1984 | if (le64toh(o->entry.monotonic) == needle) | |
1985 | return TEST_FOUND; | |
1986 | else if (le64toh(o->entry.monotonic) < needle) | |
1987 | return TEST_LEFT; | |
1988 | else | |
1989 | return TEST_RIGHT; | |
1990 | } | |
1991 | ||
1992 | static int find_data_object_by_boot_id( | |
1993 | JournalFile *f, | |
1994 | sd_id128_t boot_id, | |
1995 | Object **o, | |
1996 | uint64_t *b) { | |
1997 | ||
1998 | char t[sizeof("_BOOT_ID=")-1 + 32 + 1] = "_BOOT_ID="; | |
1999 | ||
2000 | sd_id128_to_string(boot_id, t + 9); | |
2001 | return journal_file_find_data_object(f, t, sizeof(t) - 1, o, b); | |
2002 | } | |
2003 | ||
2004 | int journal_file_move_to_entry_by_monotonic( | |
2005 | JournalFile *f, | |
2006 | sd_id128_t boot_id, | |
2007 | uint64_t monotonic, | |
2008 | direction_t direction, | |
2009 | Object **ret, | |
2010 | uint64_t *offset) { | |
2011 | ||
2012 | Object *o; | |
2013 | int r; | |
2014 | ||
2015 | assert(f); | |
2016 | ||
2017 | r = find_data_object_by_boot_id(f, boot_id, &o, NULL); | |
2018 | if (r < 0) | |
2019 | return r; | |
2020 | if (r == 0) | |
2021 | return -ENOENT; | |
2022 | ||
2023 | return generic_array_bisect_plus_one(f, | |
2024 | le64toh(o->data.entry_offset), | |
2025 | le64toh(o->data.entry_array_offset), | |
2026 | le64toh(o->data.n_entries), | |
2027 | monotonic, | |
2028 | test_object_monotonic, | |
2029 | direction, | |
2030 | ret, offset, NULL); | |
2031 | } | |
2032 | ||
2033 | void journal_file_reset_location(JournalFile *f) { | |
2034 | f->location_type = LOCATION_HEAD; | |
2035 | f->current_offset = 0; | |
2036 | f->current_seqnum = 0; | |
2037 | f->current_realtime = 0; | |
2038 | f->current_monotonic = 0; | |
2039 | zero(f->current_boot_id); | |
2040 | f->current_xor_hash = 0; | |
2041 | } | |
2042 | ||
2043 | void journal_file_save_location(JournalFile *f, Object *o, uint64_t offset) { | |
2044 | f->location_type = LOCATION_SEEK; | |
2045 | f->current_offset = offset; | |
2046 | f->current_seqnum = le64toh(o->entry.seqnum); | |
2047 | f->current_realtime = le64toh(o->entry.realtime); | |
2048 | f->current_monotonic = le64toh(o->entry.monotonic); | |
2049 | f->current_boot_id = o->entry.boot_id; | |
2050 | f->current_xor_hash = le64toh(o->entry.xor_hash); | |
2051 | } | |
2052 | ||
2053 | int journal_file_compare_locations(JournalFile *af, JournalFile *bf) { | |
2054 | assert(af); | |
2055 | assert(bf); | |
2056 | assert(af->location_type == LOCATION_SEEK); | |
2057 | assert(bf->location_type == LOCATION_SEEK); | |
2058 | ||
2059 | /* If contents and timestamps match, these entries are | |
2060 | * identical, even if the seqnum does not match */ | |
2061 | if (sd_id128_equal(af->current_boot_id, bf->current_boot_id) && | |
2062 | af->current_monotonic == bf->current_monotonic && | |
2063 | af->current_realtime == bf->current_realtime && | |
2064 | af->current_xor_hash == bf->current_xor_hash) | |
2065 | return 0; | |
2066 | ||
2067 | if (sd_id128_equal(af->header->seqnum_id, bf->header->seqnum_id)) { | |
2068 | ||
2069 | /* If this is from the same seqnum source, compare | |
2070 | * seqnums */ | |
2071 | if (af->current_seqnum < bf->current_seqnum) | |
2072 | return -1; | |
2073 | if (af->current_seqnum > bf->current_seqnum) | |
2074 | return 1; | |
2075 | ||
2076 | /* Wow! This is weird, different data but the same | |
2077 | * seqnums? Something is borked, but let's make the | |
2078 | * best of it and compare by time. */ | |
2079 | } | |
2080 | ||
2081 | if (sd_id128_equal(af->current_boot_id, bf->current_boot_id)) { | |
2082 | ||
2083 | /* If the boot id matches, compare monotonic time */ | |
2084 | if (af->current_monotonic < bf->current_monotonic) | |
2085 | return -1; | |
2086 | if (af->current_monotonic > bf->current_monotonic) | |
2087 | return 1; | |
2088 | } | |
2089 | ||
2090 | /* Otherwise, compare UTC time */ | |
2091 | if (af->current_realtime < bf->current_realtime) | |
2092 | return -1; | |
2093 | if (af->current_realtime > bf->current_realtime) | |
2094 | return 1; | |
2095 | ||
2096 | /* Finally, compare by contents */ | |
2097 | if (af->current_xor_hash < bf->current_xor_hash) | |
2098 | return -1; | |
2099 | if (af->current_xor_hash > bf->current_xor_hash) | |
2100 | return 1; | |
2101 | ||
2102 | return 0; | |
2103 | } | |
2104 | ||
2105 | int journal_file_next_entry( | |
2106 | JournalFile *f, | |
2107 | uint64_t p, | |
2108 | direction_t direction, | |
2109 | Object **ret, uint64_t *offset) { | |
2110 | ||
2111 | uint64_t i, n, ofs; | |
2112 | int r; | |
2113 | ||
2114 | assert(f); | |
2115 | ||
2116 | n = le64toh(f->header->n_entries); | |
2117 | if (n <= 0) | |
2118 | return 0; | |
2119 | ||
2120 | if (p == 0) | |
2121 | i = direction == DIRECTION_DOWN ? 0 : n - 1; | |
2122 | else { | |
2123 | r = generic_array_bisect(f, | |
2124 | le64toh(f->header->entry_array_offset), | |
2125 | le64toh(f->header->n_entries), | |
2126 | p, | |
2127 | test_object_offset, | |
2128 | DIRECTION_DOWN, | |
2129 | NULL, NULL, | |
2130 | &i); | |
2131 | if (r <= 0) | |
2132 | return r; | |
2133 | ||
2134 | if (direction == DIRECTION_DOWN) { | |
2135 | if (i >= n - 1) | |
2136 | return 0; | |
2137 | ||
2138 | i++; | |
2139 | } else { | |
2140 | if (i <= 0) | |
2141 | return 0; | |
2142 | ||
2143 | i--; | |
2144 | } | |
2145 | } | |
2146 | ||
2147 | /* And jump to it */ | |
2148 | r = generic_array_get(f, | |
2149 | le64toh(f->header->entry_array_offset), | |
2150 | i, | |
2151 | ret, &ofs); | |
2152 | if (r <= 0) | |
2153 | return r; | |
2154 | ||
2155 | if (p > 0 && | |
2156 | (direction == DIRECTION_DOWN ? ofs <= p : ofs >= p)) { | |
2157 | log_debug("%s: entry array corrupted at entry %"PRIu64, | |
2158 | f->path, i); | |
2159 | return -EBADMSG; | |
2160 | } | |
2161 | ||
2162 | if (offset) | |
2163 | *offset = ofs; | |
2164 | ||
2165 | return 1; | |
2166 | } | |
2167 | ||
2168 | int journal_file_next_entry_for_data( | |
2169 | JournalFile *f, | |
2170 | Object *o, uint64_t p, | |
2171 | uint64_t data_offset, | |
2172 | direction_t direction, | |
2173 | Object **ret, uint64_t *offset) { | |
2174 | ||
2175 | uint64_t n, i; | |
2176 | int r; | |
2177 | Object *d; | |
2178 | ||
2179 | assert(f); | |
2180 | assert(p > 0 || !o); | |
2181 | ||
2182 | r = journal_file_move_to_object(f, OBJECT_DATA, data_offset, &d); | |
2183 | if (r < 0) | |
2184 | return r; | |
2185 | ||
2186 | n = le64toh(d->data.n_entries); | |
2187 | if (n <= 0) | |
2188 | return n; | |
2189 | ||
2190 | if (!o) | |
2191 | i = direction == DIRECTION_DOWN ? 0 : n - 1; | |
2192 | else { | |
2193 | if (o->object.type != OBJECT_ENTRY) | |
2194 | return -EINVAL; | |
2195 | ||
2196 | r = generic_array_bisect_plus_one(f, | |
2197 | le64toh(d->data.entry_offset), | |
2198 | le64toh(d->data.entry_array_offset), | |
2199 | le64toh(d->data.n_entries), | |
2200 | p, | |
2201 | test_object_offset, | |
2202 | DIRECTION_DOWN, | |
2203 | NULL, NULL, | |
2204 | &i); | |
2205 | ||
2206 | if (r <= 0) | |
2207 | return r; | |
2208 | ||
2209 | if (direction == DIRECTION_DOWN) { | |
2210 | if (i >= n - 1) | |
2211 | return 0; | |
2212 | ||
2213 | i++; | |
2214 | } else { | |
2215 | if (i <= 0) | |
2216 | return 0; | |
2217 | ||
2218 | i--; | |
2219 | } | |
2220 | ||
2221 | } | |
2222 | ||
2223 | return generic_array_get_plus_one(f, | |
2224 | le64toh(d->data.entry_offset), | |
2225 | le64toh(d->data.entry_array_offset), | |
2226 | i, | |
2227 | ret, offset); | |
2228 | } | |
2229 | ||
2230 | int journal_file_move_to_entry_by_offset_for_data( | |
2231 | JournalFile *f, | |
2232 | uint64_t data_offset, | |
2233 | uint64_t p, | |
2234 | direction_t direction, | |
2235 | Object **ret, uint64_t *offset) { | |
2236 | ||
2237 | int r; | |
2238 | Object *d; | |
2239 | ||
2240 | assert(f); | |
2241 | ||
2242 | r = journal_file_move_to_object(f, OBJECT_DATA, data_offset, &d); | |
2243 | if (r < 0) | |
2244 | return r; | |
2245 | ||
2246 | return generic_array_bisect_plus_one(f, | |
2247 | le64toh(d->data.entry_offset), | |
2248 | le64toh(d->data.entry_array_offset), | |
2249 | le64toh(d->data.n_entries), | |
2250 | p, | |
2251 | test_object_offset, | |
2252 | direction, | |
2253 | ret, offset, NULL); | |
2254 | } | |
2255 | ||
2256 | int journal_file_move_to_entry_by_monotonic_for_data( | |
2257 | JournalFile *f, | |
2258 | uint64_t data_offset, | |
2259 | sd_id128_t boot_id, | |
2260 | uint64_t monotonic, | |
2261 | direction_t direction, | |
2262 | Object **ret, uint64_t *offset) { | |
2263 | ||
2264 | Object *o, *d; | |
2265 | int r; | |
2266 | uint64_t b, z; | |
2267 | ||
2268 | assert(f); | |
2269 | ||
2270 | /* First, seek by time */ | |
2271 | r = find_data_object_by_boot_id(f, boot_id, &o, &b); | |
2272 | if (r < 0) | |
2273 | return r; | |
2274 | if (r == 0) | |
2275 | return -ENOENT; | |
2276 | ||
2277 | r = generic_array_bisect_plus_one(f, | |
2278 | le64toh(o->data.entry_offset), | |
2279 | le64toh(o->data.entry_array_offset), | |
2280 | le64toh(o->data.n_entries), | |
2281 | monotonic, | |
2282 | test_object_monotonic, | |
2283 | direction, | |
2284 | NULL, &z, NULL); | |
2285 | if (r <= 0) | |
2286 | return r; | |
2287 | ||
2288 | /* And now, continue seeking until we find an entry that | |
2289 | * exists in both bisection arrays */ | |
2290 | ||
2291 | for (;;) { | |
2292 | Object *qo; | |
2293 | uint64_t p, q; | |
2294 | ||
2295 | r = journal_file_move_to_object(f, OBJECT_DATA, data_offset, &d); | |
2296 | if (r < 0) | |
2297 | return r; | |
2298 | ||
2299 | r = generic_array_bisect_plus_one(f, | |
2300 | le64toh(d->data.entry_offset), | |
2301 | le64toh(d->data.entry_array_offset), | |
2302 | le64toh(d->data.n_entries), | |
2303 | z, | |
2304 | test_object_offset, | |
2305 | direction, | |
2306 | NULL, &p, NULL); | |
2307 | if (r <= 0) | |
2308 | return r; | |
2309 | ||
2310 | r = journal_file_move_to_object(f, OBJECT_DATA, b, &o); | |
2311 | if (r < 0) | |
2312 | return r; | |
2313 | ||
2314 | r = generic_array_bisect_plus_one(f, | |
2315 | le64toh(o->data.entry_offset), | |
2316 | le64toh(o->data.entry_array_offset), | |
2317 | le64toh(o->data.n_entries), | |
2318 | p, | |
2319 | test_object_offset, | |
2320 | direction, | |
2321 | &qo, &q, NULL); | |
2322 | ||
2323 | if (r <= 0) | |
2324 | return r; | |
2325 | ||
2326 | if (p == q) { | |
2327 | if (ret) | |
2328 | *ret = qo; | |
2329 | if (offset) | |
2330 | *offset = q; | |
2331 | ||
2332 | return 1; | |
2333 | } | |
2334 | ||
2335 | z = q; | |
2336 | } | |
2337 | } | |
2338 | ||
2339 | int journal_file_move_to_entry_by_seqnum_for_data( | |
2340 | JournalFile *f, | |
2341 | uint64_t data_offset, | |
2342 | uint64_t seqnum, | |
2343 | direction_t direction, | |
2344 | Object **ret, uint64_t *offset) { | |
2345 | ||
2346 | Object *d; | |
2347 | int r; | |
2348 | ||
2349 | assert(f); | |
2350 | ||
2351 | r = journal_file_move_to_object(f, OBJECT_DATA, data_offset, &d); | |
2352 | if (r < 0) | |
2353 | return r; | |
2354 | ||
2355 | return generic_array_bisect_plus_one(f, | |
2356 | le64toh(d->data.entry_offset), | |
2357 | le64toh(d->data.entry_array_offset), | |
2358 | le64toh(d->data.n_entries), | |
2359 | seqnum, | |
2360 | test_object_seqnum, | |
2361 | direction, | |
2362 | ret, offset, NULL); | |
2363 | } | |
2364 | ||
2365 | int journal_file_move_to_entry_by_realtime_for_data( | |
2366 | JournalFile *f, | |
2367 | uint64_t data_offset, | |
2368 | uint64_t realtime, | |
2369 | direction_t direction, | |
2370 | Object **ret, uint64_t *offset) { | |
2371 | ||
2372 | Object *d; | |
2373 | int r; | |
2374 | ||
2375 | assert(f); | |
2376 | ||
2377 | r = journal_file_move_to_object(f, OBJECT_DATA, data_offset, &d); | |
2378 | if (r < 0) | |
2379 | return r; | |
2380 | ||
2381 | return generic_array_bisect_plus_one(f, | |
2382 | le64toh(d->data.entry_offset), | |
2383 | le64toh(d->data.entry_array_offset), | |
2384 | le64toh(d->data.n_entries), | |
2385 | realtime, | |
2386 | test_object_realtime, | |
2387 | direction, | |
2388 | ret, offset, NULL); | |
2389 | } | |
2390 | ||
2391 | void journal_file_dump(JournalFile *f) { | |
2392 | Object *o; | |
2393 | int r; | |
2394 | uint64_t p; | |
2395 | ||
2396 | assert(f); | |
2397 | ||
2398 | journal_file_print_header(f); | |
2399 | ||
2400 | p = le64toh(f->header->header_size); | |
2401 | while (p != 0) { | |
2402 | r = journal_file_move_to_object(f, OBJECT_UNUSED, p, &o); | |
2403 | if (r < 0) | |
2404 | goto fail; | |
2405 | ||
2406 | switch (o->object.type) { | |
2407 | ||
2408 | case OBJECT_UNUSED: | |
2409 | printf("Type: OBJECT_UNUSED\n"); | |
2410 | break; | |
2411 | ||
2412 | case OBJECT_DATA: | |
2413 | printf("Type: OBJECT_DATA\n"); | |
2414 | break; | |
2415 | ||
2416 | case OBJECT_FIELD: | |
2417 | printf("Type: OBJECT_FIELD\n"); | |
2418 | break; | |
2419 | ||
2420 | case OBJECT_ENTRY: | |
2421 | printf("Type: OBJECT_ENTRY seqnum=%"PRIu64" monotonic=%"PRIu64" realtime=%"PRIu64"\n", | |
2422 | le64toh(o->entry.seqnum), | |
2423 | le64toh(o->entry.monotonic), | |
2424 | le64toh(o->entry.realtime)); | |
2425 | break; | |
2426 | ||
2427 | case OBJECT_FIELD_HASH_TABLE: | |
2428 | printf("Type: OBJECT_FIELD_HASH_TABLE\n"); | |
2429 | break; | |
2430 | ||
2431 | case OBJECT_DATA_HASH_TABLE: | |
2432 | printf("Type: OBJECT_DATA_HASH_TABLE\n"); | |
2433 | break; | |
2434 | ||
2435 | case OBJECT_ENTRY_ARRAY: | |
2436 | printf("Type: OBJECT_ENTRY_ARRAY\n"); | |
2437 | break; | |
2438 | ||
2439 | case OBJECT_TAG: | |
2440 | printf("Type: OBJECT_TAG seqnum=%"PRIu64" epoch=%"PRIu64"\n", | |
2441 | le64toh(o->tag.seqnum), | |
2442 | le64toh(o->tag.epoch)); | |
2443 | break; | |
2444 | ||
2445 | default: | |
2446 | printf("Type: unknown (%i)\n", o->object.type); | |
2447 | break; | |
2448 | } | |
2449 | ||
2450 | if (o->object.flags & OBJECT_COMPRESSION_MASK) | |
2451 | printf("Flags: %s\n", | |
2452 | object_compressed_to_string(o->object.flags & OBJECT_COMPRESSION_MASK)); | |
2453 | ||
2454 | if (p == le64toh(f->header->tail_object_offset)) | |
2455 | p = 0; | |
2456 | else | |
2457 | p = p + ALIGN64(le64toh(o->object.size)); | |
2458 | } | |
2459 | ||
2460 | return; | |
2461 | fail: | |
2462 | log_error("File corrupt"); | |
2463 | } | |
2464 | ||
2465 | static const char* format_timestamp_safe(char *buf, size_t l, usec_t t) { | |
2466 | const char *x; | |
2467 | ||
2468 | x = format_timestamp(buf, l, t); | |
2469 | if (x) | |
2470 | return x; | |
2471 | return " --- "; | |
2472 | } | |
2473 | ||
2474 | void journal_file_print_header(JournalFile *f) { | |
2475 | char a[33], b[33], c[33], d[33]; | |
2476 | char x[FORMAT_TIMESTAMP_MAX], y[FORMAT_TIMESTAMP_MAX], z[FORMAT_TIMESTAMP_MAX]; | |
2477 | struct stat st; | |
2478 | char bytes[FORMAT_BYTES_MAX]; | |
2479 | ||
2480 | assert(f); | |
2481 | ||
2482 | printf("File Path: %s\n" | |
2483 | "File ID: %s\n" | |
2484 | "Machine ID: %s\n" | |
2485 | "Boot ID: %s\n" | |
2486 | "Sequential Number ID: %s\n" | |
2487 | "State: %s\n" | |
2488 | "Compatible Flags:%s%s\n" | |
2489 | "Incompatible Flags:%s%s%s\n" | |
2490 | "Header size: %"PRIu64"\n" | |
2491 | "Arena size: %"PRIu64"\n" | |
2492 | "Data Hash Table Size: %"PRIu64"\n" | |
2493 | "Field Hash Table Size: %"PRIu64"\n" | |
2494 | "Rotate Suggested: %s\n" | |
2495 | "Head Sequential Number: %"PRIu64"\n" | |
2496 | "Tail Sequential Number: %"PRIu64"\n" | |
2497 | "Head Realtime Timestamp: %s\n" | |
2498 | "Tail Realtime Timestamp: %s\n" | |
2499 | "Tail Monotonic Timestamp: %s\n" | |
2500 | "Objects: %"PRIu64"\n" | |
2501 | "Entry Objects: %"PRIu64"\n", | |
2502 | f->path, | |
2503 | sd_id128_to_string(f->header->file_id, a), | |
2504 | sd_id128_to_string(f->header->machine_id, b), | |
2505 | sd_id128_to_string(f->header->boot_id, c), | |
2506 | sd_id128_to_string(f->header->seqnum_id, d), | |
2507 | f->header->state == STATE_OFFLINE ? "OFFLINE" : | |
2508 | f->header->state == STATE_ONLINE ? "ONLINE" : | |
2509 | f->header->state == STATE_ARCHIVED ? "ARCHIVED" : "UNKNOWN", | |
2510 | JOURNAL_HEADER_SEALED(f->header) ? " SEALED" : "", | |
2511 | (le32toh(f->header->compatible_flags) & ~HEADER_COMPATIBLE_ANY) ? " ???" : "", | |
2512 | JOURNAL_HEADER_COMPRESSED_XZ(f->header) ? " COMPRESSED-XZ" : "", | |
2513 | JOURNAL_HEADER_COMPRESSED_LZ4(f->header) ? " COMPRESSED-LZ4" : "", | |
2514 | (le32toh(f->header->incompatible_flags) & ~HEADER_INCOMPATIBLE_ANY) ? " ???" : "", | |
2515 | le64toh(f->header->header_size), | |
2516 | le64toh(f->header->arena_size), | |
2517 | le64toh(f->header->data_hash_table_size) / sizeof(HashItem), | |
2518 | le64toh(f->header->field_hash_table_size) / sizeof(HashItem), | |
2519 | yes_no(journal_file_rotate_suggested(f, 0)), | |
2520 | le64toh(f->header->head_entry_seqnum), | |
2521 | le64toh(f->header->tail_entry_seqnum), | |
2522 | format_timestamp_safe(x, sizeof(x), le64toh(f->header->head_entry_realtime)), | |
2523 | format_timestamp_safe(y, sizeof(y), le64toh(f->header->tail_entry_realtime)), | |
2524 | format_timespan(z, sizeof(z), le64toh(f->header->tail_entry_monotonic), USEC_PER_MSEC), | |
2525 | le64toh(f->header->n_objects), | |
2526 | le64toh(f->header->n_entries)); | |
2527 | ||
2528 | if (JOURNAL_HEADER_CONTAINS(f->header, n_data)) | |
2529 | printf("Data Objects: %"PRIu64"\n" | |
2530 | "Data Hash Table Fill: %.1f%%\n", | |
2531 | le64toh(f->header->n_data), | |
2532 | 100.0 * (double) le64toh(f->header->n_data) / ((double) (le64toh(f->header->data_hash_table_size) / sizeof(HashItem)))); | |
2533 | ||
2534 | if (JOURNAL_HEADER_CONTAINS(f->header, n_fields)) | |
2535 | printf("Field Objects: %"PRIu64"\n" | |
2536 | "Field Hash Table Fill: %.1f%%\n", | |
2537 | le64toh(f->header->n_fields), | |
2538 | 100.0 * (double) le64toh(f->header->n_fields) / ((double) (le64toh(f->header->field_hash_table_size) / sizeof(HashItem)))); | |
2539 | ||
2540 | if (JOURNAL_HEADER_CONTAINS(f->header, n_tags)) | |
2541 | printf("Tag Objects: %"PRIu64"\n", | |
2542 | le64toh(f->header->n_tags)); | |
2543 | if (JOURNAL_HEADER_CONTAINS(f->header, n_entry_arrays)) | |
2544 | printf("Entry Array Objects: %"PRIu64"\n", | |
2545 | le64toh(f->header->n_entry_arrays)); | |
2546 | ||
2547 | if (fstat(f->fd, &st) >= 0) | |
2548 | printf("Disk usage: %s\n", format_bytes(bytes, sizeof(bytes), (uint64_t) st.st_blocks * 512ULL)); | |
2549 | } | |
2550 | ||
2551 | static int journal_file_warn_btrfs(JournalFile *f) { | |
2552 | unsigned attrs; | |
2553 | int r; | |
2554 | ||
2555 | assert(f); | |
2556 | ||
2557 | /* Before we write anything, check if the COW logic is turned | |
2558 | * off on btrfs. Given our write pattern that is quite | |
2559 | * unfriendly to COW file systems this should greatly improve | |
2560 | * performance on COW file systems, such as btrfs, at the | |
2561 | * expense of data integrity features (which shouldn't be too | |
2562 | * bad, given that we do our own checksumming). */ | |
2563 | ||
2564 | r = btrfs_is_filesystem(f->fd); | |
2565 | if (r < 0) | |
2566 | return log_warning_errno(r, "Failed to determine if journal is on btrfs: %m"); | |
2567 | if (!r) | |
2568 | return 0; | |
2569 | ||
2570 | r = read_attr_fd(f->fd, &attrs); | |
2571 | if (r < 0) | |
2572 | return log_warning_errno(r, "Failed to read file attributes: %m"); | |
2573 | ||
2574 | if (attrs & FS_NOCOW_FL) { | |
2575 | log_debug("Detected btrfs file system with copy-on-write disabled, all is good."); | |
2576 | return 0; | |
2577 | } | |
2578 | ||
2579 | log_notice("Creating journal file %s on a btrfs file system, and copy-on-write is enabled. " | |
2580 | "This is likely to slow down journal access substantially, please consider turning " | |
2581 | "off the copy-on-write file attribute on the journal directory, using chattr +C.", f->path); | |
2582 | ||
2583 | return 1; | |
2584 | } | |
2585 | ||
2586 | int journal_file_open( | |
2587 | const char *fname, | |
2588 | int flags, | |
2589 | mode_t mode, | |
2590 | bool compress, | |
2591 | bool seal, | |
2592 | JournalMetrics *metrics, | |
2593 | MMapCache *mmap_cache, | |
2594 | JournalFile *template, | |
2595 | JournalFile **ret) { | |
2596 | ||
2597 | bool newly_created = false; | |
2598 | JournalFile *f; | |
2599 | void *h; | |
2600 | int r; | |
2601 | ||
2602 | assert(fname); | |
2603 | assert(ret); | |
2604 | ||
2605 | if ((flags & O_ACCMODE) != O_RDONLY && | |
2606 | (flags & O_ACCMODE) != O_RDWR) | |
2607 | return -EINVAL; | |
2608 | ||
2609 | if (!endswith(fname, ".journal") && | |
2610 | !endswith(fname, ".journal~")) | |
2611 | return -EINVAL; | |
2612 | ||
2613 | f = new0(JournalFile, 1); | |
2614 | if (!f) | |
2615 | return -ENOMEM; | |
2616 | ||
2617 | f->fd = -1; | |
2618 | f->mode = mode; | |
2619 | ||
2620 | f->flags = flags; | |
2621 | f->prot = prot_from_flags(flags); | |
2622 | f->writable = (flags & O_ACCMODE) != O_RDONLY; | |
2623 | #if defined(HAVE_LZ4) | |
2624 | f->compress_lz4 = compress; | |
2625 | #elif defined(HAVE_XZ) | |
2626 | f->compress_xz = compress; | |
2627 | #endif | |
2628 | #ifdef HAVE_GCRYPT | |
2629 | f->seal = seal; | |
2630 | #endif | |
2631 | ||
2632 | if (mmap_cache) | |
2633 | f->mmap = mmap_cache_ref(mmap_cache); | |
2634 | else { | |
2635 | f->mmap = mmap_cache_new(); | |
2636 | if (!f->mmap) { | |
2637 | r = -ENOMEM; | |
2638 | goto fail; | |
2639 | } | |
2640 | } | |
2641 | ||
2642 | f->path = strdup(fname); | |
2643 | if (!f->path) { | |
2644 | r = -ENOMEM; | |
2645 | goto fail; | |
2646 | } | |
2647 | ||
2648 | f->chain_cache = ordered_hashmap_new(&uint64_hash_ops); | |
2649 | if (!f->chain_cache) { | |
2650 | r = -ENOMEM; | |
2651 | goto fail; | |
2652 | } | |
2653 | ||
2654 | f->fd = open(f->path, f->flags|O_CLOEXEC, f->mode); | |
2655 | if (f->fd < 0) { | |
2656 | r = -errno; | |
2657 | goto fail; | |
2658 | } | |
2659 | ||
2660 | r = journal_file_fstat(f); | |
2661 | if (r < 0) | |
2662 | goto fail; | |
2663 | ||
2664 | if (f->last_stat.st_size == 0 && f->writable) { | |
2665 | ||
2666 | (void) journal_file_warn_btrfs(f); | |
2667 | ||
2668 | /* Let's attach the creation time to the journal file, | |
2669 | * so that the vacuuming code knows the age of this | |
2670 | * file even if the file might end up corrupted one | |
2671 | * day... Ideally we'd just use the creation time many | |
2672 | * file systems maintain for each file, but there is | |
2673 | * currently no usable API to query this, hence let's | |
2674 | * emulate this via extended attributes. If extended | |
2675 | * attributes are not supported we'll just skip this, | |
2676 | * and rely solely on mtime/atime/ctime of the file. */ | |
2677 | ||
2678 | fd_setcrtime(f->fd, 0); | |
2679 | ||
2680 | #ifdef HAVE_GCRYPT | |
2681 | /* Try to load the FSPRG state, and if we can't, then | |
2682 | * just don't do sealing */ | |
2683 | if (f->seal) { | |
2684 | r = journal_file_fss_load(f); | |
2685 | if (r < 0) | |
2686 | f->seal = false; | |
2687 | } | |
2688 | #endif | |
2689 | ||
2690 | r = journal_file_init_header(f, template); | |
2691 | if (r < 0) | |
2692 | goto fail; | |
2693 | ||
2694 | r = journal_file_fstat(f); | |
2695 | if (r < 0) | |
2696 | goto fail; | |
2697 | ||
2698 | newly_created = true; | |
2699 | } | |
2700 | ||
2701 | if (f->last_stat.st_size < (off_t) HEADER_SIZE_MIN) { | |
2702 | r = -EIO; | |
2703 | goto fail; | |
2704 | } | |
2705 | ||
2706 | r = mmap_cache_get(f->mmap, f->fd, f->prot, CONTEXT_HEADER, true, 0, PAGE_ALIGN(sizeof(Header)), &f->last_stat, &h); | |
2707 | if (r < 0) | |
2708 | goto fail; | |
2709 | ||
2710 | f->header = h; | |
2711 | ||
2712 | if (!newly_created) { | |
2713 | r = journal_file_verify_header(f); | |
2714 | if (r < 0) | |
2715 | goto fail; | |
2716 | } | |
2717 | ||
2718 | #ifdef HAVE_GCRYPT | |
2719 | if (!newly_created && f->writable) { | |
2720 | r = journal_file_fss_load(f); | |
2721 | if (r < 0) | |
2722 | goto fail; | |
2723 | } | |
2724 | #endif | |
2725 | ||
2726 | if (f->writable) { | |
2727 | if (metrics) { | |
2728 | journal_default_metrics(metrics, f->fd); | |
2729 | f->metrics = *metrics; | |
2730 | } else if (template) | |
2731 | f->metrics = template->metrics; | |
2732 | ||
2733 | r = journal_file_refresh_header(f); | |
2734 | if (r < 0) | |
2735 | goto fail; | |
2736 | } | |
2737 | ||
2738 | #ifdef HAVE_GCRYPT | |
2739 | r = journal_file_hmac_setup(f); | |
2740 | if (r < 0) | |
2741 | goto fail; | |
2742 | #endif | |
2743 | ||
2744 | if (newly_created) { | |
2745 | r = journal_file_setup_field_hash_table(f); | |
2746 | if (r < 0) | |
2747 | goto fail; | |
2748 | ||
2749 | r = journal_file_setup_data_hash_table(f); | |
2750 | if (r < 0) | |
2751 | goto fail; | |
2752 | ||
2753 | #ifdef HAVE_GCRYPT | |
2754 | r = journal_file_append_first_tag(f); | |
2755 | if (r < 0) | |
2756 | goto fail; | |
2757 | #endif | |
2758 | } | |
2759 | ||
2760 | if (mmap_cache_got_sigbus(f->mmap, f->fd)) { | |
2761 | r = -EIO; | |
2762 | goto fail; | |
2763 | } | |
2764 | ||
2765 | *ret = f; | |
2766 | return 0; | |
2767 | ||
2768 | fail: | |
2769 | if (f->fd >= 0 && mmap_cache_got_sigbus(f->mmap, f->fd)) | |
2770 | r = -EIO; | |
2771 | ||
2772 | journal_file_close(f); | |
2773 | ||
2774 | return r; | |
2775 | } | |
2776 | ||
2777 | int journal_file_rotate(JournalFile **f, bool compress, bool seal) { | |
2778 | _cleanup_free_ char *p = NULL; | |
2779 | size_t l; | |
2780 | JournalFile *old_file, *new_file = NULL; | |
2781 | int r; | |
2782 | ||
2783 | assert(f); | |
2784 | assert(*f); | |
2785 | ||
2786 | old_file = *f; | |
2787 | ||
2788 | if (!old_file->writable) | |
2789 | return -EINVAL; | |
2790 | ||
2791 | if (!endswith(old_file->path, ".journal")) | |
2792 | return -EINVAL; | |
2793 | ||
2794 | l = strlen(old_file->path); | |
2795 | r = asprintf(&p, "%.*s@" SD_ID128_FORMAT_STR "-%016"PRIx64"-%016"PRIx64".journal", | |
2796 | (int) l - 8, old_file->path, | |
2797 | SD_ID128_FORMAT_VAL(old_file->header->seqnum_id), | |
2798 | le64toh((*f)->header->head_entry_seqnum), | |
2799 | le64toh((*f)->header->head_entry_realtime)); | |
2800 | if (r < 0) | |
2801 | return -ENOMEM; | |
2802 | ||
2803 | /* Try to rename the file to the archived version. If the file | |
2804 | * already was deleted, we'll get ENOENT, let's ignore that | |
2805 | * case. */ | |
2806 | r = rename(old_file->path, p); | |
2807 | if (r < 0 && errno != ENOENT) | |
2808 | return -errno; | |
2809 | ||
2810 | old_file->header->state = STATE_ARCHIVED; | |
2811 | ||
2812 | /* Currently, btrfs is not very good with out write patterns | |
2813 | * and fragments heavily. Let's defrag our journal files when | |
2814 | * we archive them */ | |
2815 | old_file->defrag_on_close = true; | |
2816 | ||
2817 | r = journal_file_open(old_file->path, old_file->flags, old_file->mode, compress, seal, NULL, old_file->mmap, old_file, &new_file); | |
2818 | journal_file_close(old_file); | |
2819 | ||
2820 | *f = new_file; | |
2821 | return r; | |
2822 | } | |
2823 | ||
2824 | int journal_file_open_reliably( | |
2825 | const char *fname, | |
2826 | int flags, | |
2827 | mode_t mode, | |
2828 | bool compress, | |
2829 | bool seal, | |
2830 | JournalMetrics *metrics, | |
2831 | MMapCache *mmap_cache, | |
2832 | JournalFile *template, | |
2833 | JournalFile **ret) { | |
2834 | ||
2835 | int r; | |
2836 | size_t l; | |
2837 | _cleanup_free_ char *p = NULL; | |
2838 | ||
2839 | r = journal_file_open(fname, flags, mode, compress, seal, metrics, mmap_cache, template, ret); | |
2840 | if (!IN_SET(r, | |
2841 | -EBADMSG, /* corrupted */ | |
2842 | -ENODATA, /* truncated */ | |
2843 | -EHOSTDOWN, /* other machine */ | |
2844 | -EPROTONOSUPPORT, /* incompatible feature */ | |
2845 | -EBUSY, /* unclean shutdown */ | |
2846 | -ESHUTDOWN, /* already archived */ | |
2847 | -EIO, /* IO error, including SIGBUS on mmap */ | |
2848 | -EIDRM /* File has been deleted */)) | |
2849 | return r; | |
2850 | ||
2851 | if ((flags & O_ACCMODE) == O_RDONLY) | |
2852 | return r; | |
2853 | ||
2854 | if (!(flags & O_CREAT)) | |
2855 | return r; | |
2856 | ||
2857 | if (!endswith(fname, ".journal")) | |
2858 | return r; | |
2859 | ||
2860 | /* The file is corrupted. Rotate it away and try it again (but only once) */ | |
2861 | ||
2862 | l = strlen(fname); | |
2863 | if (asprintf(&p, "%.*s@%016"PRIx64 "-%016"PRIx64 ".journal~", | |
2864 | (int) l - 8, fname, | |
2865 | now(CLOCK_REALTIME), | |
2866 | random_u64()) < 0) | |
2867 | return -ENOMEM; | |
2868 | ||
2869 | if (rename(fname, p) < 0) | |
2870 | return -errno; | |
2871 | ||
2872 | /* btrfs doesn't cope well with our write pattern and | |
2873 | * fragments heavily. Let's defrag all files we rotate */ | |
2874 | ||
2875 | (void) chattr_path(p, false, FS_NOCOW_FL); | |
2876 | (void) btrfs_defrag(p); | |
2877 | ||
2878 | log_warning_errno(r, "File %s corrupted or uncleanly shut down, renaming and replacing.", fname); | |
2879 | ||
2880 | return journal_file_open(fname, flags, mode, compress, seal, metrics, mmap_cache, template, ret); | |
2881 | } | |
2882 | ||
2883 | int journal_file_copy_entry(JournalFile *from, JournalFile *to, Object *o, uint64_t p, uint64_t *seqnum, Object **ret, uint64_t *offset) { | |
2884 | uint64_t i, n; | |
2885 | uint64_t q, xor_hash = 0; | |
2886 | int r; | |
2887 | EntryItem *items; | |
2888 | dual_timestamp ts; | |
2889 | ||
2890 | assert(from); | |
2891 | assert(to); | |
2892 | assert(o); | |
2893 | assert(p); | |
2894 | ||
2895 | if (!to->writable) | |
2896 | return -EPERM; | |
2897 | ||
2898 | ts.monotonic = le64toh(o->entry.monotonic); | |
2899 | ts.realtime = le64toh(o->entry.realtime); | |
2900 | ||
2901 | n = journal_file_entry_n_items(o); | |
2902 | /* alloca() can't take 0, hence let's allocate at least one */ | |
2903 | items = alloca(sizeof(EntryItem) * MAX(1u, n)); | |
2904 | ||
2905 | for (i = 0; i < n; i++) { | |
2906 | uint64_t l, h; | |
2907 | le64_t le_hash; | |
2908 | size_t t; | |
2909 | void *data; | |
2910 | Object *u; | |
2911 | ||
2912 | q = le64toh(o->entry.items[i].object_offset); | |
2913 | le_hash = o->entry.items[i].hash; | |
2914 | ||
2915 | r = journal_file_move_to_object(from, OBJECT_DATA, q, &o); | |
2916 | if (r < 0) | |
2917 | return r; | |
2918 | ||
2919 | if (le_hash != o->data.hash) | |
2920 | return -EBADMSG; | |
2921 | ||
2922 | l = le64toh(o->object.size) - offsetof(Object, data.payload); | |
2923 | t = (size_t) l; | |
2924 | ||
2925 | /* We hit the limit on 32bit machines */ | |
2926 | if ((uint64_t) t != l) | |
2927 | return -E2BIG; | |
2928 | ||
2929 | if (o->object.flags & OBJECT_COMPRESSION_MASK) { | |
2930 | #if defined(HAVE_XZ) || defined(HAVE_LZ4) | |
2931 | size_t rsize = 0; | |
2932 | ||
2933 | r = decompress_blob(o->object.flags & OBJECT_COMPRESSION_MASK, | |
2934 | o->data.payload, l, &from->compress_buffer, &from->compress_buffer_size, &rsize, 0); | |
2935 | if (r < 0) | |
2936 | return r; | |
2937 | ||
2938 | data = from->compress_buffer; | |
2939 | l = rsize; | |
2940 | #else | |
2941 | return -EPROTONOSUPPORT; | |
2942 | #endif | |
2943 | } else | |
2944 | data = o->data.payload; | |
2945 | ||
2946 | r = journal_file_append_data(to, data, l, &u, &h); | |
2947 | if (r < 0) | |
2948 | return r; | |
2949 | ||
2950 | xor_hash ^= le64toh(u->data.hash); | |
2951 | items[i].object_offset = htole64(h); | |
2952 | items[i].hash = u->data.hash; | |
2953 | ||
2954 | r = journal_file_move_to_object(from, OBJECT_ENTRY, p, &o); | |
2955 | if (r < 0) | |
2956 | return r; | |
2957 | } | |
2958 | ||
2959 | r = journal_file_append_entry_internal(to, &ts, xor_hash, items, n, seqnum, ret, offset); | |
2960 | ||
2961 | if (mmap_cache_got_sigbus(to->mmap, to->fd)) | |
2962 | return -EIO; | |
2963 | ||
2964 | return r; | |
2965 | } | |
2966 | ||
2967 | void journal_reset_metrics(JournalMetrics *m) { | |
2968 | assert(m); | |
2969 | ||
2970 | /* Set everything to "pick automatic values". */ | |
2971 | ||
2972 | *m = (JournalMetrics) { | |
2973 | .min_use = (uint64_t) -1, | |
2974 | .max_use = (uint64_t) -1, | |
2975 | .min_size = (uint64_t) -1, | |
2976 | .max_size = (uint64_t) -1, | |
2977 | .keep_free = (uint64_t) -1, | |
2978 | .n_max_files = (uint64_t) -1, | |
2979 | }; | |
2980 | } | |
2981 | ||
2982 | void journal_default_metrics(JournalMetrics *m, int fd) { | |
2983 | char a[FORMAT_BYTES_MAX], b[FORMAT_BYTES_MAX], c[FORMAT_BYTES_MAX], d[FORMAT_BYTES_MAX], e[FORMAT_BYTES_MAX]; | |
2984 | struct statvfs ss; | |
2985 | uint64_t fs_size; | |
2986 | ||
2987 | assert(m); | |
2988 | assert(fd >= 0); | |
2989 | ||
2990 | if (fstatvfs(fd, &ss) >= 0) | |
2991 | fs_size = ss.f_frsize * ss.f_blocks; | |
2992 | else { | |
2993 | log_debug_errno(errno, "Failed to detremine disk size: %m"); | |
2994 | fs_size = 0; | |
2995 | } | |
2996 | ||
2997 | if (m->max_use == (uint64_t) -1) { | |
2998 | ||
2999 | if (fs_size > 0) { | |
3000 | m->max_use = PAGE_ALIGN(fs_size / 10); /* 10% of file system size */ | |
3001 | ||
3002 | if (m->max_use > DEFAULT_MAX_USE_UPPER) | |
3003 | m->max_use = DEFAULT_MAX_USE_UPPER; | |
3004 | ||
3005 | if (m->max_use < DEFAULT_MAX_USE_LOWER) | |
3006 | m->max_use = DEFAULT_MAX_USE_LOWER; | |
3007 | } else | |
3008 | m->max_use = DEFAULT_MAX_USE_LOWER; | |
3009 | } else { | |
3010 | m->max_use = PAGE_ALIGN(m->max_use); | |
3011 | ||
3012 | if (m->max_use != 0 && m->max_use < JOURNAL_FILE_SIZE_MIN*2) | |
3013 | m->max_use = JOURNAL_FILE_SIZE_MIN*2; | |
3014 | } | |
3015 | ||
3016 | if (m->min_use == (uint64_t) -1) | |
3017 | m->min_use = DEFAULT_MIN_USE; | |
3018 | ||
3019 | if (m->min_use > m->max_use) | |
3020 | m->min_use = m->max_use; | |
3021 | ||
3022 | if (m->max_size == (uint64_t) -1) { | |
3023 | m->max_size = PAGE_ALIGN(m->max_use / 8); /* 8 chunks */ | |
3024 | ||
3025 | if (m->max_size > DEFAULT_MAX_SIZE_UPPER) | |
3026 | m->max_size = DEFAULT_MAX_SIZE_UPPER; | |
3027 | } else | |
3028 | m->max_size = PAGE_ALIGN(m->max_size); | |
3029 | ||
3030 | if (m->max_size != 0) { | |
3031 | if (m->max_size < JOURNAL_FILE_SIZE_MIN) | |
3032 | m->max_size = JOURNAL_FILE_SIZE_MIN; | |
3033 | ||
3034 | if (m->max_use != 0 && m->max_size*2 > m->max_use) | |
3035 | m->max_use = m->max_size*2; | |
3036 | } | |
3037 | ||
3038 | if (m->min_size == (uint64_t) -1) | |
3039 | m->min_size = JOURNAL_FILE_SIZE_MIN; | |
3040 | else { | |
3041 | m->min_size = PAGE_ALIGN(m->min_size); | |
3042 | ||
3043 | if (m->min_size < JOURNAL_FILE_SIZE_MIN) | |
3044 | m->min_size = JOURNAL_FILE_SIZE_MIN; | |
3045 | ||
3046 | if (m->max_size != 0 && m->min_size > m->max_size) | |
3047 | m->max_size = m->min_size; | |
3048 | } | |
3049 | ||
3050 | if (m->keep_free == (uint64_t) -1) { | |
3051 | ||
3052 | if (fs_size > 0) { | |
3053 | m->keep_free = PAGE_ALIGN(fs_size * 3 / 20); /* 15% of file system size */ | |
3054 | ||
3055 | if (m->keep_free > DEFAULT_KEEP_FREE_UPPER) | |
3056 | m->keep_free = DEFAULT_KEEP_FREE_UPPER; | |
3057 | ||
3058 | } else | |
3059 | m->keep_free = DEFAULT_KEEP_FREE; | |
3060 | } | |
3061 | ||
3062 | if (m->n_max_files == (uint64_t) -1) | |
3063 | m->n_max_files = DEFAULT_N_MAX_FILES; | |
3064 | ||
3065 | log_debug("Fixed min_use=%s max_use=%s max_size=%s min_size=%s keep_free=%s n_max_files=%" PRIu64, | |
3066 | format_bytes(a, sizeof(a), m->min_use), | |
3067 | format_bytes(b, sizeof(b), m->max_use), | |
3068 | format_bytes(c, sizeof(c), m->max_size), | |
3069 | format_bytes(d, sizeof(d), m->min_size), | |
3070 | format_bytes(e, sizeof(e), m->keep_free), | |
3071 | m->n_max_files); | |
3072 | } | |
3073 | ||
3074 | int journal_file_get_cutoff_realtime_usec(JournalFile *f, usec_t *from, usec_t *to) { | |
3075 | assert(f); | |
3076 | assert(from || to); | |
3077 | ||
3078 | if (from) { | |
3079 | if (f->header->head_entry_realtime == 0) | |
3080 | return -ENOENT; | |
3081 | ||
3082 | *from = le64toh(f->header->head_entry_realtime); | |
3083 | } | |
3084 | ||
3085 | if (to) { | |
3086 | if (f->header->tail_entry_realtime == 0) | |
3087 | return -ENOENT; | |
3088 | ||
3089 | *to = le64toh(f->header->tail_entry_realtime); | |
3090 | } | |
3091 | ||
3092 | return 1; | |
3093 | } | |
3094 | ||
3095 | int journal_file_get_cutoff_monotonic_usec(JournalFile *f, sd_id128_t boot_id, usec_t *from, usec_t *to) { | |
3096 | Object *o; | |
3097 | uint64_t p; | |
3098 | int r; | |
3099 | ||
3100 | assert(f); | |
3101 | assert(from || to); | |
3102 | ||
3103 | r = find_data_object_by_boot_id(f, boot_id, &o, &p); | |
3104 | if (r <= 0) | |
3105 | return r; | |
3106 | ||
3107 | if (le64toh(o->data.n_entries) <= 0) | |
3108 | return 0; | |
3109 | ||
3110 | if (from) { | |
3111 | r = journal_file_move_to_object(f, OBJECT_ENTRY, le64toh(o->data.entry_offset), &o); | |
3112 | if (r < 0) | |
3113 | return r; | |
3114 | ||
3115 | *from = le64toh(o->entry.monotonic); | |
3116 | } | |
3117 | ||
3118 | if (to) { | |
3119 | r = journal_file_move_to_object(f, OBJECT_DATA, p, &o); | |
3120 | if (r < 0) | |
3121 | return r; | |
3122 | ||
3123 | r = generic_array_get_plus_one(f, | |
3124 | le64toh(o->data.entry_offset), | |
3125 | le64toh(o->data.entry_array_offset), | |
3126 | le64toh(o->data.n_entries)-1, | |
3127 | &o, NULL); | |
3128 | if (r <= 0) | |
3129 | return r; | |
3130 | ||
3131 | *to = le64toh(o->entry.monotonic); | |
3132 | } | |
3133 | ||
3134 | return 1; | |
3135 | } | |
3136 | ||
3137 | bool journal_file_rotate_suggested(JournalFile *f, usec_t max_file_usec) { | |
3138 | assert(f); | |
3139 | ||
3140 | /* If we gained new header fields we gained new features, | |
3141 | * hence suggest a rotation */ | |
3142 | if (le64toh(f->header->header_size) < sizeof(Header)) { | |
3143 | log_debug("%s uses an outdated header, suggesting rotation.", f->path); | |
3144 | return true; | |
3145 | } | |
3146 | ||
3147 | /* Let's check if the hash tables grew over a certain fill | |
3148 | * level (75%, borrowing this value from Java's hash table | |
3149 | * implementation), and if so suggest a rotation. To calculate | |
3150 | * the fill level we need the n_data field, which only exists | |
3151 | * in newer versions. */ | |
3152 | ||
3153 | if (JOURNAL_HEADER_CONTAINS(f->header, n_data)) | |
3154 | if (le64toh(f->header->n_data) * 4ULL > (le64toh(f->header->data_hash_table_size) / sizeof(HashItem)) * 3ULL) { | |
3155 | log_debug("Data hash table of %s has a fill level at %.1f (%"PRIu64" of %"PRIu64" items, %llu file size, %"PRIu64" bytes per hash table item), suggesting rotation.", | |
3156 | f->path, | |
3157 | 100.0 * (double) le64toh(f->header->n_data) / ((double) (le64toh(f->header->data_hash_table_size) / sizeof(HashItem))), | |
3158 | le64toh(f->header->n_data), | |
3159 | le64toh(f->header->data_hash_table_size) / sizeof(HashItem), | |
3160 | (unsigned long long) f->last_stat.st_size, | |
3161 | f->last_stat.st_size / le64toh(f->header->n_data)); | |
3162 | return true; | |
3163 | } | |
3164 | ||
3165 | if (JOURNAL_HEADER_CONTAINS(f->header, n_fields)) | |
3166 | if (le64toh(f->header->n_fields) * 4ULL > (le64toh(f->header->field_hash_table_size) / sizeof(HashItem)) * 3ULL) { | |
3167 | log_debug("Field hash table of %s has a fill level at %.1f (%"PRIu64" of %"PRIu64" items), suggesting rotation.", | |
3168 | f->path, | |
3169 | 100.0 * (double) le64toh(f->header->n_fields) / ((double) (le64toh(f->header->field_hash_table_size) / sizeof(HashItem))), | |
3170 | le64toh(f->header->n_fields), | |
3171 | le64toh(f->header->field_hash_table_size) / sizeof(HashItem)); | |
3172 | return true; | |
3173 | } | |
3174 | ||
3175 | /* Are the data objects properly indexed by field objects? */ | |
3176 | if (JOURNAL_HEADER_CONTAINS(f->header, n_data) && | |
3177 | JOURNAL_HEADER_CONTAINS(f->header, n_fields) && | |
3178 | le64toh(f->header->n_data) > 0 && | |
3179 | le64toh(f->header->n_fields) == 0) | |
3180 | return true; | |
3181 | ||
3182 | if (max_file_usec > 0) { | |
3183 | usec_t t, h; | |
3184 | ||
3185 | h = le64toh(f->header->head_entry_realtime); | |
3186 | t = now(CLOCK_REALTIME); | |
3187 | ||
3188 | if (h > 0 && t > h + max_file_usec) | |
3189 | return true; | |
3190 | } | |
3191 | ||
3192 | return false; | |
3193 | } |