]>
Commit | Line | Data |
---|---|---|
1 | /* SPDX-License-Identifier: LGPL-2.1+ */ | |
2 | ||
3 | #include <errno.h> | |
4 | #include <fcntl.h> | |
5 | #include <linux/fs.h> | |
6 | #include <pthread.h> | |
7 | #include <stddef.h> | |
8 | #include <sys/mman.h> | |
9 | #include <sys/statvfs.h> | |
10 | #include <sys/uio.h> | |
11 | #include <unistd.h> | |
12 | ||
13 | #include "sd-event.h" | |
14 | ||
15 | #include "alloc-util.h" | |
16 | #include "btrfs-util.h" | |
17 | #include "chattr-util.h" | |
18 | #include "compress.h" | |
19 | #include "fd-util.h" | |
20 | #include "format-util.h" | |
21 | #include "fs-util.h" | |
22 | #include "journal-authenticate.h" | |
23 | #include "journal-def.h" | |
24 | #include "journal-file.h" | |
25 | #include "lookup3.h" | |
26 | #include "memory-util.h" | |
27 | #include "path-util.h" | |
28 | #include "random-util.h" | |
29 | #include "set.h" | |
30 | #include "sort-util.h" | |
31 | #include "stat-util.h" | |
32 | #include "string-util.h" | |
33 | #include "strv.h" | |
34 | #include "xattr-util.h" | |
35 | ||
36 | #define DEFAULT_DATA_HASH_TABLE_SIZE (2047ULL*sizeof(HashItem)) | |
37 | #define DEFAULT_FIELD_HASH_TABLE_SIZE (333ULL*sizeof(HashItem)) | |
38 | ||
39 | #define DEFAULT_COMPRESS_THRESHOLD (512ULL) | |
40 | #define MIN_COMPRESS_THRESHOLD (8ULL) | |
41 | ||
42 | /* This is the minimum journal file size */ | |
43 | #define JOURNAL_FILE_SIZE_MIN (512ULL*1024ULL) /* 512 KiB */ | |
44 | ||
45 | /* These are the lower and upper bounds if we deduce the max_use value | |
46 | * from the file system size */ | |
47 | #define DEFAULT_MAX_USE_LOWER (1ULL*1024ULL*1024ULL) /* 1 MiB */ | |
48 | #define DEFAULT_MAX_USE_UPPER (4ULL*1024ULL*1024ULL*1024ULL) /* 4 GiB */ | |
49 | ||
50 | /* This is the default minimal use limit, how much we'll use even if keep_free suggests otherwise. */ | |
51 | #define DEFAULT_MIN_USE (1ULL*1024ULL*1024ULL) /* 1 MiB */ | |
52 | ||
53 | /* This is the upper bound if we deduce max_size from max_use */ | |
54 | #define DEFAULT_MAX_SIZE_UPPER (128ULL*1024ULL*1024ULL) /* 128 MiB */ | |
55 | ||
56 | /* This is the upper bound if we deduce the keep_free value from the | |
57 | * file system size */ | |
58 | #define DEFAULT_KEEP_FREE_UPPER (4ULL*1024ULL*1024ULL*1024ULL) /* 4 GiB */ | |
59 | ||
60 | /* This is the keep_free value when we can't determine the system | |
61 | * size */ | |
62 | #define DEFAULT_KEEP_FREE (1024ULL*1024ULL) /* 1 MB */ | |
63 | ||
64 | /* This is the default maximum number of journal files to keep around. */ | |
65 | #define DEFAULT_N_MAX_FILES (100) | |
66 | ||
67 | /* n_data was the first entry we added after the initial file format design */ | |
68 | #define HEADER_SIZE_MIN ALIGN64(offsetof(Header, n_data)) | |
69 | ||
70 | /* How many entries to keep in the entry array chain cache at max */ | |
71 | #define CHAIN_CACHE_MAX 20 | |
72 | ||
73 | /* How much to increase the journal file size at once each time we allocate something new. */ | |
74 | #define FILE_SIZE_INCREASE (8ULL*1024ULL*1024ULL) /* 8MB */ | |
75 | ||
76 | /* Reread fstat() of the file for detecting deletions at least this often */ | |
77 | #define LAST_STAT_REFRESH_USEC (5*USEC_PER_SEC) | |
78 | ||
79 | /* The mmap context to use for the header we pick as one above the last defined typed */ | |
80 | #define CONTEXT_HEADER _OBJECT_TYPE_MAX | |
81 | ||
82 | #ifdef __clang__ | |
83 | # pragma GCC diagnostic ignored "-Waddress-of-packed-member" | |
84 | #endif | |
85 | ||
86 | /* This may be called from a separate thread to prevent blocking the caller for the duration of fsync(). | |
87 | * As a result we use atomic operations on f->offline_state for inter-thread communications with | |
88 | * journal_file_set_offline() and journal_file_set_online(). */ | |
89 | static void journal_file_set_offline_internal(JournalFile *f) { | |
90 | assert(f); | |
91 | assert(f->fd >= 0); | |
92 | assert(f->header); | |
93 | ||
94 | for (;;) { | |
95 | switch (f->offline_state) { | |
96 | case OFFLINE_CANCEL: | |
97 | if (!__sync_bool_compare_and_swap(&f->offline_state, OFFLINE_CANCEL, OFFLINE_DONE)) | |
98 | continue; | |
99 | return; | |
100 | ||
101 | case OFFLINE_AGAIN_FROM_SYNCING: | |
102 | if (!__sync_bool_compare_and_swap(&f->offline_state, OFFLINE_AGAIN_FROM_SYNCING, OFFLINE_SYNCING)) | |
103 | continue; | |
104 | break; | |
105 | ||
106 | case OFFLINE_AGAIN_FROM_OFFLINING: | |
107 | if (!__sync_bool_compare_and_swap(&f->offline_state, OFFLINE_AGAIN_FROM_OFFLINING, OFFLINE_SYNCING)) | |
108 | continue; | |
109 | break; | |
110 | ||
111 | case OFFLINE_SYNCING: | |
112 | (void) fsync(f->fd); | |
113 | ||
114 | if (!__sync_bool_compare_and_swap(&f->offline_state, OFFLINE_SYNCING, OFFLINE_OFFLINING)) | |
115 | continue; | |
116 | ||
117 | f->header->state = f->archive ? STATE_ARCHIVED : STATE_OFFLINE; | |
118 | (void) fsync(f->fd); | |
119 | break; | |
120 | ||
121 | case OFFLINE_OFFLINING: | |
122 | if (!__sync_bool_compare_and_swap(&f->offline_state, OFFLINE_OFFLINING, OFFLINE_DONE)) | |
123 | continue; | |
124 | _fallthrough_; | |
125 | case OFFLINE_DONE: | |
126 | return; | |
127 | ||
128 | case OFFLINE_JOINED: | |
129 | log_debug("OFFLINE_JOINED unexpected offline state for journal_file_set_offline_internal()"); | |
130 | return; | |
131 | } | |
132 | } | |
133 | } | |
134 | ||
135 | static void * journal_file_set_offline_thread(void *arg) { | |
136 | JournalFile *f = arg; | |
137 | ||
138 | (void) pthread_setname_np(pthread_self(), "journal-offline"); | |
139 | ||
140 | journal_file_set_offline_internal(f); | |
141 | ||
142 | return NULL; | |
143 | } | |
144 | ||
145 | static int journal_file_set_offline_thread_join(JournalFile *f) { | |
146 | int r; | |
147 | ||
148 | assert(f); | |
149 | ||
150 | if (f->offline_state == OFFLINE_JOINED) | |
151 | return 0; | |
152 | ||
153 | r = pthread_join(f->offline_thread, NULL); | |
154 | if (r) | |
155 | return -r; | |
156 | ||
157 | f->offline_state = OFFLINE_JOINED; | |
158 | ||
159 | if (mmap_cache_got_sigbus(f->mmap, f->cache_fd)) | |
160 | return -EIO; | |
161 | ||
162 | return 0; | |
163 | } | |
164 | ||
165 | /* Trigger a restart if the offline thread is mid-flight in a restartable state. */ | |
166 | static bool journal_file_set_offline_try_restart(JournalFile *f) { | |
167 | for (;;) { | |
168 | switch (f->offline_state) { | |
169 | case OFFLINE_AGAIN_FROM_SYNCING: | |
170 | case OFFLINE_AGAIN_FROM_OFFLINING: | |
171 | return true; | |
172 | ||
173 | case OFFLINE_CANCEL: | |
174 | if (!__sync_bool_compare_and_swap(&f->offline_state, OFFLINE_CANCEL, OFFLINE_AGAIN_FROM_SYNCING)) | |
175 | continue; | |
176 | return true; | |
177 | ||
178 | case OFFLINE_SYNCING: | |
179 | if (!__sync_bool_compare_and_swap(&f->offline_state, OFFLINE_SYNCING, OFFLINE_AGAIN_FROM_SYNCING)) | |
180 | continue; | |
181 | return true; | |
182 | ||
183 | case OFFLINE_OFFLINING: | |
184 | if (!__sync_bool_compare_and_swap(&f->offline_state, OFFLINE_OFFLINING, OFFLINE_AGAIN_FROM_OFFLINING)) | |
185 | continue; | |
186 | return true; | |
187 | ||
188 | default: | |
189 | return false; | |
190 | } | |
191 | } | |
192 | } | |
193 | ||
194 | /* Sets a journal offline. | |
195 | * | |
196 | * If wait is false then an offline is dispatched in a separate thread for a | |
197 | * subsequent journal_file_set_offline() or journal_file_set_online() of the | |
198 | * same journal to synchronize with. | |
199 | * | |
200 | * If wait is true, then either an existing offline thread will be restarted | |
201 | * and joined, or if none exists the offline is simply performed in this | |
202 | * context without involving another thread. | |
203 | */ | |
204 | int journal_file_set_offline(JournalFile *f, bool wait) { | |
205 | bool restarted; | |
206 | int r; | |
207 | ||
208 | assert(f); | |
209 | ||
210 | if (!f->writable) | |
211 | return -EPERM; | |
212 | ||
213 | if (f->fd < 0 || !f->header) | |
214 | return -EINVAL; | |
215 | ||
216 | /* An offlining journal is implicitly online and may modify f->header->state, | |
217 | * we must also join any potentially lingering offline thread when not online. */ | |
218 | if (!journal_file_is_offlining(f) && f->header->state != STATE_ONLINE) | |
219 | return journal_file_set_offline_thread_join(f); | |
220 | ||
221 | /* Restart an in-flight offline thread and wait if needed, or join a lingering done one. */ | |
222 | restarted = journal_file_set_offline_try_restart(f); | |
223 | if ((restarted && wait) || !restarted) { | |
224 | r = journal_file_set_offline_thread_join(f); | |
225 | if (r < 0) | |
226 | return r; | |
227 | } | |
228 | ||
229 | if (restarted) | |
230 | return 0; | |
231 | ||
232 | /* Initiate a new offline. */ | |
233 | f->offline_state = OFFLINE_SYNCING; | |
234 | ||
235 | if (wait) /* Without using a thread if waiting. */ | |
236 | journal_file_set_offline_internal(f); | |
237 | else { | |
238 | sigset_t ss, saved_ss; | |
239 | int k; | |
240 | ||
241 | assert_se(sigfillset(&ss) >= 0); | |
242 | /* Don't block SIGBUS since the offlining thread accesses a memory mapped file. | |
243 | * Asynchronous SIGBUS signals can safely be handled by either thread. */ | |
244 | assert_se(sigdelset(&ss, SIGBUS) >= 0); | |
245 | ||
246 | r = pthread_sigmask(SIG_BLOCK, &ss, &saved_ss); | |
247 | if (r > 0) | |
248 | return -r; | |
249 | ||
250 | r = pthread_create(&f->offline_thread, NULL, journal_file_set_offline_thread, f); | |
251 | ||
252 | k = pthread_sigmask(SIG_SETMASK, &saved_ss, NULL); | |
253 | if (r > 0) { | |
254 | f->offline_state = OFFLINE_JOINED; | |
255 | return -r; | |
256 | } | |
257 | if (k > 0) | |
258 | return -k; | |
259 | } | |
260 | ||
261 | return 0; | |
262 | } | |
263 | ||
264 | static int journal_file_set_online(JournalFile *f) { | |
265 | bool wait = true; | |
266 | ||
267 | assert(f); | |
268 | ||
269 | if (!f->writable) | |
270 | return -EPERM; | |
271 | ||
272 | if (f->fd < 0 || !f->header) | |
273 | return -EINVAL; | |
274 | ||
275 | while (wait) { | |
276 | switch (f->offline_state) { | |
277 | case OFFLINE_JOINED: | |
278 | /* No offline thread, no need to wait. */ | |
279 | wait = false; | |
280 | break; | |
281 | ||
282 | case OFFLINE_SYNCING: | |
283 | if (!__sync_bool_compare_and_swap(&f->offline_state, OFFLINE_SYNCING, OFFLINE_CANCEL)) | |
284 | continue; | |
285 | /* Canceled syncing prior to offlining, no need to wait. */ | |
286 | wait = false; | |
287 | break; | |
288 | ||
289 | case OFFLINE_AGAIN_FROM_SYNCING: | |
290 | if (!__sync_bool_compare_and_swap(&f->offline_state, OFFLINE_AGAIN_FROM_SYNCING, OFFLINE_CANCEL)) | |
291 | continue; | |
292 | /* Canceled restart from syncing, no need to wait. */ | |
293 | wait = false; | |
294 | break; | |
295 | ||
296 | case OFFLINE_AGAIN_FROM_OFFLINING: | |
297 | if (!__sync_bool_compare_and_swap(&f->offline_state, OFFLINE_AGAIN_FROM_OFFLINING, OFFLINE_CANCEL)) | |
298 | continue; | |
299 | /* Canceled restart from offlining, must wait for offlining to complete however. */ | |
300 | _fallthrough_; | |
301 | default: { | |
302 | int r; | |
303 | ||
304 | r = journal_file_set_offline_thread_join(f); | |
305 | if (r < 0) | |
306 | return r; | |
307 | ||
308 | wait = false; | |
309 | break; | |
310 | } | |
311 | } | |
312 | } | |
313 | ||
314 | if (mmap_cache_got_sigbus(f->mmap, f->cache_fd)) | |
315 | return -EIO; | |
316 | ||
317 | switch (f->header->state) { | |
318 | case STATE_ONLINE: | |
319 | return 0; | |
320 | ||
321 | case STATE_OFFLINE: | |
322 | f->header->state = STATE_ONLINE; | |
323 | (void) fsync(f->fd); | |
324 | return 0; | |
325 | ||
326 | default: | |
327 | return -EINVAL; | |
328 | } | |
329 | } | |
330 | ||
331 | bool journal_file_is_offlining(JournalFile *f) { | |
332 | assert(f); | |
333 | ||
334 | __sync_synchronize(); | |
335 | ||
336 | if (IN_SET(f->offline_state, OFFLINE_DONE, OFFLINE_JOINED)) | |
337 | return false; | |
338 | ||
339 | return true; | |
340 | } | |
341 | ||
342 | JournalFile* journal_file_close(JournalFile *f) { | |
343 | if (!f) | |
344 | return NULL; | |
345 | ||
346 | #if HAVE_GCRYPT | |
347 | /* Write the final tag */ | |
348 | if (f->seal && f->writable) { | |
349 | int r; | |
350 | ||
351 | r = journal_file_append_tag(f); | |
352 | if (r < 0) | |
353 | log_error_errno(r, "Failed to append tag when closing journal: %m"); | |
354 | } | |
355 | #endif | |
356 | ||
357 | if (f->post_change_timer) { | |
358 | if (sd_event_source_get_enabled(f->post_change_timer, NULL) > 0) | |
359 | journal_file_post_change(f); | |
360 | ||
361 | sd_event_source_disable_unref(f->post_change_timer); | |
362 | } | |
363 | ||
364 | journal_file_set_offline(f, true); | |
365 | ||
366 | if (f->mmap && f->cache_fd) | |
367 | mmap_cache_free_fd(f->mmap, f->cache_fd); | |
368 | ||
369 | if (f->fd >= 0 && f->defrag_on_close) { | |
370 | ||
371 | /* Be friendly to btrfs: turn COW back on again now, | |
372 | * and defragment the file. We won't write to the file | |
373 | * ever again, hence remove all fragmentation, and | |
374 | * reenable all the good bits COW usually provides | |
375 | * (such as data checksumming). */ | |
376 | ||
377 | (void) chattr_fd(f->fd, 0, FS_NOCOW_FL, NULL); | |
378 | (void) btrfs_defrag_fd(f->fd); | |
379 | } | |
380 | ||
381 | if (f->close_fd) | |
382 | safe_close(f->fd); | |
383 | free(f->path); | |
384 | ||
385 | mmap_cache_unref(f->mmap); | |
386 | ||
387 | ordered_hashmap_free_free(f->chain_cache); | |
388 | ||
389 | #if HAVE_XZ || HAVE_LZ4 | |
390 | free(f->compress_buffer); | |
391 | #endif | |
392 | ||
393 | #if HAVE_GCRYPT | |
394 | if (f->fss_file) | |
395 | munmap(f->fss_file, PAGE_ALIGN(f->fss_file_size)); | |
396 | else | |
397 | free(f->fsprg_state); | |
398 | ||
399 | free(f->fsprg_seed); | |
400 | ||
401 | if (f->hmac) | |
402 | gcry_md_close(f->hmac); | |
403 | #endif | |
404 | ||
405 | return mfree(f); | |
406 | } | |
407 | ||
408 | static int journal_file_init_header(JournalFile *f, JournalFile *template) { | |
409 | Header h = {}; | |
410 | ssize_t k; | |
411 | int r; | |
412 | ||
413 | assert(f); | |
414 | ||
415 | memcpy(h.signature, HEADER_SIGNATURE, 8); | |
416 | h.header_size = htole64(ALIGN64(sizeof(h))); | |
417 | ||
418 | h.incompatible_flags |= htole32( | |
419 | f->compress_xz * HEADER_INCOMPATIBLE_COMPRESSED_XZ | | |
420 | f->compress_lz4 * HEADER_INCOMPATIBLE_COMPRESSED_LZ4); | |
421 | ||
422 | h.compatible_flags = htole32( | |
423 | f->seal * HEADER_COMPATIBLE_SEALED); | |
424 | ||
425 | r = sd_id128_randomize(&h.file_id); | |
426 | if (r < 0) | |
427 | return r; | |
428 | ||
429 | if (template) { | |
430 | h.seqnum_id = template->header->seqnum_id; | |
431 | h.tail_entry_seqnum = template->header->tail_entry_seqnum; | |
432 | } else | |
433 | h.seqnum_id = h.file_id; | |
434 | ||
435 | k = pwrite(f->fd, &h, sizeof(h), 0); | |
436 | if (k < 0) | |
437 | return -errno; | |
438 | ||
439 | if (k != sizeof(h)) | |
440 | return -EIO; | |
441 | ||
442 | return 0; | |
443 | } | |
444 | ||
445 | static int journal_file_refresh_header(JournalFile *f) { | |
446 | sd_id128_t boot_id; | |
447 | int r; | |
448 | ||
449 | assert(f); | |
450 | assert(f->header); | |
451 | ||
452 | r = sd_id128_get_machine(&f->header->machine_id); | |
453 | if (IN_SET(r, -ENOENT, -ENOMEDIUM)) | |
454 | /* We don't have a machine-id, let's continue without */ | |
455 | zero(f->header->machine_id); | |
456 | else if (r < 0) | |
457 | return r; | |
458 | ||
459 | r = sd_id128_get_boot(&boot_id); | |
460 | if (r < 0) | |
461 | return r; | |
462 | ||
463 | f->header->boot_id = boot_id; | |
464 | ||
465 | r = journal_file_set_online(f); | |
466 | ||
467 | /* Sync the online state to disk */ | |
468 | (void) fsync(f->fd); | |
469 | ||
470 | /* We likely just created a new file, also sync the directory this file is located in. */ | |
471 | (void) fsync_directory_of_file(f->fd); | |
472 | ||
473 | return r; | |
474 | } | |
475 | ||
476 | static bool warn_wrong_flags(const JournalFile *f, bool compatible) { | |
477 | const uint32_t any = compatible ? HEADER_COMPATIBLE_ANY : HEADER_INCOMPATIBLE_ANY, | |
478 | supported = compatible ? HEADER_COMPATIBLE_SUPPORTED : HEADER_INCOMPATIBLE_SUPPORTED; | |
479 | const char *type = compatible ? "compatible" : "incompatible"; | |
480 | uint32_t flags; | |
481 | ||
482 | flags = le32toh(compatible ? f->header->compatible_flags : f->header->incompatible_flags); | |
483 | ||
484 | if (flags & ~supported) { | |
485 | if (flags & ~any) | |
486 | log_debug("Journal file %s has unknown %s flags 0x%"PRIx32, | |
487 | f->path, type, flags & ~any); | |
488 | flags = (flags & any) & ~supported; | |
489 | if (flags) { | |
490 | const char* strv[3]; | |
491 | unsigned n = 0; | |
492 | _cleanup_free_ char *t = NULL; | |
493 | ||
494 | if (compatible && (flags & HEADER_COMPATIBLE_SEALED)) | |
495 | strv[n++] = "sealed"; | |
496 | if (!compatible && (flags & HEADER_INCOMPATIBLE_COMPRESSED_XZ)) | |
497 | strv[n++] = "xz-compressed"; | |
498 | if (!compatible && (flags & HEADER_INCOMPATIBLE_COMPRESSED_LZ4)) | |
499 | strv[n++] = "lz4-compressed"; | |
500 | strv[n] = NULL; | |
501 | assert(n < ELEMENTSOF(strv)); | |
502 | ||
503 | t = strv_join((char**) strv, ", "); | |
504 | log_debug("Journal file %s uses %s %s %s disabled at compilation time.", | |
505 | f->path, type, n > 1 ? "flags" : "flag", strnull(t)); | |
506 | } | |
507 | return true; | |
508 | } | |
509 | ||
510 | return false; | |
511 | } | |
512 | ||
513 | static int journal_file_verify_header(JournalFile *f) { | |
514 | uint64_t arena_size, header_size; | |
515 | ||
516 | assert(f); | |
517 | assert(f->header); | |
518 | ||
519 | if (memcmp(f->header->signature, HEADER_SIGNATURE, 8)) | |
520 | return -EBADMSG; | |
521 | ||
522 | /* In both read and write mode we refuse to open files with incompatible | |
523 | * flags we don't know. */ | |
524 | if (warn_wrong_flags(f, false)) | |
525 | return -EPROTONOSUPPORT; | |
526 | ||
527 | /* When open for writing we refuse to open files with compatible flags, too. */ | |
528 | if (f->writable && warn_wrong_flags(f, true)) | |
529 | return -EPROTONOSUPPORT; | |
530 | ||
531 | if (f->header->state >= _STATE_MAX) | |
532 | return -EBADMSG; | |
533 | ||
534 | header_size = le64toh(f->header->header_size); | |
535 | ||
536 | /* The first addition was n_data, so check that we are at least this large */ | |
537 | if (header_size < HEADER_SIZE_MIN) | |
538 | return -EBADMSG; | |
539 | ||
540 | if (JOURNAL_HEADER_SEALED(f->header) && !JOURNAL_HEADER_CONTAINS(f->header, n_entry_arrays)) | |
541 | return -EBADMSG; | |
542 | ||
543 | arena_size = le64toh(f->header->arena_size); | |
544 | ||
545 | if (UINT64_MAX - header_size < arena_size || header_size + arena_size > (uint64_t) f->last_stat.st_size) | |
546 | return -ENODATA; | |
547 | ||
548 | if (le64toh(f->header->tail_object_offset) > header_size + arena_size) | |
549 | return -ENODATA; | |
550 | ||
551 | if (!VALID64(le64toh(f->header->data_hash_table_offset)) || | |
552 | !VALID64(le64toh(f->header->field_hash_table_offset)) || | |
553 | !VALID64(le64toh(f->header->tail_object_offset)) || | |
554 | !VALID64(le64toh(f->header->entry_array_offset))) | |
555 | return -ENODATA; | |
556 | ||
557 | if (f->writable) { | |
558 | sd_id128_t machine_id; | |
559 | uint8_t state; | |
560 | int r; | |
561 | ||
562 | r = sd_id128_get_machine(&machine_id); | |
563 | if (r < 0) | |
564 | return r; | |
565 | ||
566 | if (!sd_id128_equal(machine_id, f->header->machine_id)) | |
567 | return -EHOSTDOWN; | |
568 | ||
569 | state = f->header->state; | |
570 | ||
571 | if (state == STATE_ARCHIVED) | |
572 | return -ESHUTDOWN; /* Already archived */ | |
573 | else if (state == STATE_ONLINE) | |
574 | return log_debug_errno(SYNTHETIC_ERRNO(EBUSY), | |
575 | "Journal file %s is already online. Assuming unclean closing.", | |
576 | f->path); | |
577 | else if (state != STATE_OFFLINE) | |
578 | return log_debug_errno(SYNTHETIC_ERRNO(EBUSY), | |
579 | "Journal file %s has unknown state %i.", | |
580 | f->path, state); | |
581 | ||
582 | if (f->header->field_hash_table_size == 0 || f->header->data_hash_table_size == 0) | |
583 | return -EBADMSG; | |
584 | ||
585 | /* Don't permit appending to files from the future. Because otherwise the realtime timestamps wouldn't | |
586 | * be strictly ordered in the entries in the file anymore, and we can't have that since it breaks | |
587 | * bisection. */ | |
588 | if (le64toh(f->header->tail_entry_realtime) > now(CLOCK_REALTIME)) | |
589 | return log_debug_errno(SYNTHETIC_ERRNO(ETXTBSY), | |
590 | "Journal file %s is from the future, refusing to append new data to it that'd be older.", | |
591 | f->path); | |
592 | } | |
593 | ||
594 | f->compress_xz = JOURNAL_HEADER_COMPRESSED_XZ(f->header); | |
595 | f->compress_lz4 = JOURNAL_HEADER_COMPRESSED_LZ4(f->header); | |
596 | ||
597 | f->seal = JOURNAL_HEADER_SEALED(f->header); | |
598 | ||
599 | return 0; | |
600 | } | |
601 | ||
602 | static int journal_file_fstat(JournalFile *f) { | |
603 | int r; | |
604 | ||
605 | assert(f); | |
606 | assert(f->fd >= 0); | |
607 | ||
608 | if (fstat(f->fd, &f->last_stat) < 0) | |
609 | return -errno; | |
610 | ||
611 | f->last_stat_usec = now(CLOCK_MONOTONIC); | |
612 | ||
613 | /* Refuse dealing with with files that aren't regular */ | |
614 | r = stat_verify_regular(&f->last_stat); | |
615 | if (r < 0) | |
616 | return r; | |
617 | ||
618 | /* Refuse appending to files that are already deleted */ | |
619 | if (f->last_stat.st_nlink <= 0) | |
620 | return -EIDRM; | |
621 | ||
622 | return 0; | |
623 | } | |
624 | ||
625 | static int journal_file_allocate(JournalFile *f, uint64_t offset, uint64_t size) { | |
626 | uint64_t old_size, new_size; | |
627 | int r; | |
628 | ||
629 | assert(f); | |
630 | assert(f->header); | |
631 | ||
632 | /* We assume that this file is not sparse, and we know that | |
633 | * for sure, since we always call posix_fallocate() | |
634 | * ourselves */ | |
635 | ||
636 | if (mmap_cache_got_sigbus(f->mmap, f->cache_fd)) | |
637 | return -EIO; | |
638 | ||
639 | old_size = | |
640 | le64toh(f->header->header_size) + | |
641 | le64toh(f->header->arena_size); | |
642 | ||
643 | new_size = PAGE_ALIGN(offset + size); | |
644 | if (new_size < le64toh(f->header->header_size)) | |
645 | new_size = le64toh(f->header->header_size); | |
646 | ||
647 | if (new_size <= old_size) { | |
648 | ||
649 | /* We already pre-allocated enough space, but before | |
650 | * we write to it, let's check with fstat() if the | |
651 | * file got deleted, in order make sure we don't throw | |
652 | * away the data immediately. Don't check fstat() for | |
653 | * all writes though, but only once ever 10s. */ | |
654 | ||
655 | if (f->last_stat_usec + LAST_STAT_REFRESH_USEC > now(CLOCK_MONOTONIC)) | |
656 | return 0; | |
657 | ||
658 | return journal_file_fstat(f); | |
659 | } | |
660 | ||
661 | /* Allocate more space. */ | |
662 | ||
663 | if (f->metrics.max_size > 0 && new_size > f->metrics.max_size) | |
664 | return -E2BIG; | |
665 | ||
666 | if (new_size > f->metrics.min_size && f->metrics.keep_free > 0) { | |
667 | struct statvfs svfs; | |
668 | ||
669 | if (fstatvfs(f->fd, &svfs) >= 0) { | |
670 | uint64_t available; | |
671 | ||
672 | available = LESS_BY((uint64_t) svfs.f_bfree * (uint64_t) svfs.f_bsize, f->metrics.keep_free); | |
673 | ||
674 | if (new_size - old_size > available) | |
675 | return -E2BIG; | |
676 | } | |
677 | } | |
678 | ||
679 | /* Increase by larger blocks at once */ | |
680 | new_size = DIV_ROUND_UP(new_size, FILE_SIZE_INCREASE) * FILE_SIZE_INCREASE; | |
681 | if (f->metrics.max_size > 0 && new_size > f->metrics.max_size) | |
682 | new_size = f->metrics.max_size; | |
683 | ||
684 | /* Note that the glibc fallocate() fallback is very | |
685 | inefficient, hence we try to minimize the allocation area | |
686 | as we can. */ | |
687 | r = posix_fallocate(f->fd, old_size, new_size - old_size); | |
688 | if (r != 0) | |
689 | return -r; | |
690 | ||
691 | f->header->arena_size = htole64(new_size - le64toh(f->header->header_size)); | |
692 | ||
693 | return journal_file_fstat(f); | |
694 | } | |
695 | ||
696 | static unsigned type_to_context(ObjectType type) { | |
697 | /* One context for each type, plus one catch-all for the rest */ | |
698 | assert_cc(_OBJECT_TYPE_MAX <= MMAP_CACHE_MAX_CONTEXTS); | |
699 | assert_cc(CONTEXT_HEADER < MMAP_CACHE_MAX_CONTEXTS); | |
700 | return type > OBJECT_UNUSED && type < _OBJECT_TYPE_MAX ? type : 0; | |
701 | } | |
702 | ||
703 | static int journal_file_move_to(JournalFile *f, ObjectType type, bool keep_always, uint64_t offset, uint64_t size, void **ret, size_t *ret_size) { | |
704 | int r; | |
705 | ||
706 | assert(f); | |
707 | assert(ret); | |
708 | ||
709 | if (size <= 0) | |
710 | return -EINVAL; | |
711 | ||
712 | /* Avoid SIGBUS on invalid accesses */ | |
713 | if (offset + size > (uint64_t) f->last_stat.st_size) { | |
714 | /* Hmm, out of range? Let's refresh the fstat() data | |
715 | * first, before we trust that check. */ | |
716 | ||
717 | r = journal_file_fstat(f); | |
718 | if (r < 0) | |
719 | return r; | |
720 | ||
721 | if (offset + size > (uint64_t) f->last_stat.st_size) | |
722 | return -EADDRNOTAVAIL; | |
723 | } | |
724 | ||
725 | return mmap_cache_get(f->mmap, f->cache_fd, f->prot, type_to_context(type), keep_always, offset, size, &f->last_stat, ret, ret_size); | |
726 | } | |
727 | ||
728 | static uint64_t minimum_header_size(Object *o) { | |
729 | ||
730 | static const uint64_t table[] = { | |
731 | [OBJECT_DATA] = sizeof(DataObject), | |
732 | [OBJECT_FIELD] = sizeof(FieldObject), | |
733 | [OBJECT_ENTRY] = sizeof(EntryObject), | |
734 | [OBJECT_DATA_HASH_TABLE] = sizeof(HashTableObject), | |
735 | [OBJECT_FIELD_HASH_TABLE] = sizeof(HashTableObject), | |
736 | [OBJECT_ENTRY_ARRAY] = sizeof(EntryArrayObject), | |
737 | [OBJECT_TAG] = sizeof(TagObject), | |
738 | }; | |
739 | ||
740 | if (o->object.type >= ELEMENTSOF(table) || table[o->object.type] <= 0) | |
741 | return sizeof(ObjectHeader); | |
742 | ||
743 | return table[o->object.type]; | |
744 | } | |
745 | ||
746 | /* Lightweight object checks. We want this to be fast, so that we won't | |
747 | * slowdown every journal_file_move_to_object() call too much. */ | |
748 | static int journal_file_check_object(JournalFile *f, uint64_t offset, Object *o) { | |
749 | assert(f); | |
750 | assert(o); | |
751 | ||
752 | switch (o->object.type) { | |
753 | ||
754 | case OBJECT_DATA: { | |
755 | if ((le64toh(o->data.entry_offset) == 0) ^ (le64toh(o->data.n_entries) == 0)) | |
756 | return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG), | |
757 | "Bad n_entries: %" PRIu64 ": %" PRIu64, | |
758 | le64toh(o->data.n_entries), | |
759 | offset); | |
760 | ||
761 | if (le64toh(o->object.size) - offsetof(DataObject, payload) <= 0) | |
762 | return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG), | |
763 | "Bad object size (<= %zu): %" PRIu64 ": %" PRIu64, | |
764 | offsetof(DataObject, payload), | |
765 | le64toh(o->object.size), | |
766 | offset); | |
767 | ||
768 | if (!VALID64(le64toh(o->data.next_hash_offset)) || | |
769 | !VALID64(le64toh(o->data.next_field_offset)) || | |
770 | !VALID64(le64toh(o->data.entry_offset)) || | |
771 | !VALID64(le64toh(o->data.entry_array_offset))) | |
772 | return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG), | |
773 | "Invalid offset, next_hash_offset=" OFSfmt ", next_field_offset=" OFSfmt ", entry_offset=" OFSfmt ", entry_array_offset=" OFSfmt ": %" PRIu64, | |
774 | le64toh(o->data.next_hash_offset), | |
775 | le64toh(o->data.next_field_offset), | |
776 | le64toh(o->data.entry_offset), | |
777 | le64toh(o->data.entry_array_offset), | |
778 | offset); | |
779 | ||
780 | break; | |
781 | } | |
782 | ||
783 | case OBJECT_FIELD: | |
784 | if (le64toh(o->object.size) - offsetof(FieldObject, payload) <= 0) | |
785 | return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG), | |
786 | "Bad field size (<= %zu): %" PRIu64 ": %" PRIu64, | |
787 | offsetof(FieldObject, payload), | |
788 | le64toh(o->object.size), | |
789 | offset); | |
790 | ||
791 | if (!VALID64(le64toh(o->field.next_hash_offset)) || | |
792 | !VALID64(le64toh(o->field.head_data_offset))) | |
793 | return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG), | |
794 | "Invalid offset, next_hash_offset=" OFSfmt ", head_data_offset=" OFSfmt ": %" PRIu64, | |
795 | le64toh(o->field.next_hash_offset), | |
796 | le64toh(o->field.head_data_offset), | |
797 | offset); | |
798 | break; | |
799 | ||
800 | case OBJECT_ENTRY: | |
801 | if ((le64toh(o->object.size) - offsetof(EntryObject, items)) % sizeof(EntryItem) != 0) | |
802 | return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG), | |
803 | "Bad entry size (<= %zu): %" PRIu64 ": %" PRIu64, | |
804 | offsetof(EntryObject, items), | |
805 | le64toh(o->object.size), | |
806 | offset); | |
807 | ||
808 | if ((le64toh(o->object.size) - offsetof(EntryObject, items)) / sizeof(EntryItem) <= 0) | |
809 | return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG), | |
810 | "Invalid number items in entry: %" PRIu64 ": %" PRIu64, | |
811 | (le64toh(o->object.size) - offsetof(EntryObject, items)) / sizeof(EntryItem), | |
812 | offset); | |
813 | ||
814 | if (le64toh(o->entry.seqnum) <= 0) | |
815 | return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG), | |
816 | "Invalid entry seqnum: %" PRIx64 ": %" PRIu64, | |
817 | le64toh(o->entry.seqnum), | |
818 | offset); | |
819 | ||
820 | if (!VALID_REALTIME(le64toh(o->entry.realtime))) | |
821 | return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG), | |
822 | "Invalid entry realtime timestamp: %" PRIu64 ": %" PRIu64, | |
823 | le64toh(o->entry.realtime), | |
824 | offset); | |
825 | ||
826 | if (!VALID_MONOTONIC(le64toh(o->entry.monotonic))) | |
827 | return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG), | |
828 | "Invalid entry monotonic timestamp: %" PRIu64 ": %" PRIu64, | |
829 | le64toh(o->entry.monotonic), | |
830 | offset); | |
831 | ||
832 | break; | |
833 | ||
834 | case OBJECT_DATA_HASH_TABLE: | |
835 | case OBJECT_FIELD_HASH_TABLE: | |
836 | if ((le64toh(o->object.size) - offsetof(HashTableObject, items)) % sizeof(HashItem) != 0 || | |
837 | (le64toh(o->object.size) - offsetof(HashTableObject, items)) / sizeof(HashItem) <= 0) | |
838 | return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG), | |
839 | "Invalid %s hash table size: %" PRIu64 ": %" PRIu64, | |
840 | o->object.type == OBJECT_DATA_HASH_TABLE ? "data" : "field", | |
841 | le64toh(o->object.size), | |
842 | offset); | |
843 | ||
844 | break; | |
845 | ||
846 | case OBJECT_ENTRY_ARRAY: | |
847 | if ((le64toh(o->object.size) - offsetof(EntryArrayObject, items)) % sizeof(le64_t) != 0 || | |
848 | (le64toh(o->object.size) - offsetof(EntryArrayObject, items)) / sizeof(le64_t) <= 0) | |
849 | return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG), | |
850 | "Invalid object entry array size: %" PRIu64 ": %" PRIu64, | |
851 | le64toh(o->object.size), | |
852 | offset); | |
853 | ||
854 | if (!VALID64(le64toh(o->entry_array.next_entry_array_offset))) | |
855 | return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG), | |
856 | "Invalid object entry array next_entry_array_offset: " OFSfmt ": %" PRIu64, | |
857 | le64toh(o->entry_array.next_entry_array_offset), | |
858 | offset); | |
859 | ||
860 | break; | |
861 | ||
862 | case OBJECT_TAG: | |
863 | if (le64toh(o->object.size) != sizeof(TagObject)) | |
864 | return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG), | |
865 | "Invalid object tag size: %" PRIu64 ": %" PRIu64, | |
866 | le64toh(o->object.size), | |
867 | offset); | |
868 | ||
869 | if (!VALID_EPOCH(le64toh(o->tag.epoch))) | |
870 | return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG), | |
871 | "Invalid object tag epoch: %" PRIu64 ": %" PRIu64, | |
872 | le64toh(o->tag.epoch), offset); | |
873 | ||
874 | break; | |
875 | } | |
876 | ||
877 | return 0; | |
878 | } | |
879 | ||
880 | int journal_file_move_to_object(JournalFile *f, ObjectType type, uint64_t offset, Object **ret) { | |
881 | int r; | |
882 | void *t; | |
883 | size_t tsize; | |
884 | Object *o; | |
885 | uint64_t s; | |
886 | ||
887 | assert(f); | |
888 | assert(ret); | |
889 | ||
890 | /* Objects may only be located at multiple of 64 bit */ | |
891 | if (!VALID64(offset)) | |
892 | return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG), | |
893 | "Attempt to move to object at non-64bit boundary: %" PRIu64, | |
894 | offset); | |
895 | ||
896 | /* Object may not be located in the file header */ | |
897 | if (offset < le64toh(f->header->header_size)) | |
898 | return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG), | |
899 | "Attempt to move to object located in file header: %" PRIu64, | |
900 | offset); | |
901 | ||
902 | r = journal_file_move_to(f, type, false, offset, sizeof(ObjectHeader), &t, &tsize); | |
903 | if (r < 0) | |
904 | return r; | |
905 | ||
906 | o = (Object*) t; | |
907 | s = le64toh(o->object.size); | |
908 | ||
909 | if (s == 0) | |
910 | return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG), | |
911 | "Attempt to move to uninitialized object: %" PRIu64, | |
912 | offset); | |
913 | if (s < sizeof(ObjectHeader)) | |
914 | return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG), | |
915 | "Attempt to move to overly short object: %" PRIu64, | |
916 | offset); | |
917 | ||
918 | if (o->object.type <= OBJECT_UNUSED) | |
919 | return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG), | |
920 | "Attempt to move to object with invalid type: %" PRIu64, | |
921 | offset); | |
922 | ||
923 | if (s < minimum_header_size(o)) | |
924 | return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG), | |
925 | "Attempt to move to truncated object: %" PRIu64, | |
926 | offset); | |
927 | ||
928 | if (type > OBJECT_UNUSED && o->object.type != type) | |
929 | return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG), | |
930 | "Attempt to move to object of unexpected type: %" PRIu64, | |
931 | offset); | |
932 | ||
933 | if (s > tsize) { | |
934 | r = journal_file_move_to(f, type, false, offset, s, &t, NULL); | |
935 | if (r < 0) | |
936 | return r; | |
937 | ||
938 | o = (Object*) t; | |
939 | } | |
940 | ||
941 | r = journal_file_check_object(f, offset, o); | |
942 | if (r < 0) | |
943 | return r; | |
944 | ||
945 | *ret = o; | |
946 | return 0; | |
947 | } | |
948 | ||
949 | static uint64_t journal_file_entry_seqnum(JournalFile *f, uint64_t *seqnum) { | |
950 | uint64_t r; | |
951 | ||
952 | assert(f); | |
953 | assert(f->header); | |
954 | ||
955 | r = le64toh(f->header->tail_entry_seqnum) + 1; | |
956 | ||
957 | if (seqnum) { | |
958 | /* If an external seqnum counter was passed, we update | |
959 | * both the local and the external one, and set it to | |
960 | * the maximum of both */ | |
961 | ||
962 | if (*seqnum + 1 > r) | |
963 | r = *seqnum + 1; | |
964 | ||
965 | *seqnum = r; | |
966 | } | |
967 | ||
968 | f->header->tail_entry_seqnum = htole64(r); | |
969 | ||
970 | if (f->header->head_entry_seqnum == 0) | |
971 | f->header->head_entry_seqnum = htole64(r); | |
972 | ||
973 | return r; | |
974 | } | |
975 | ||
976 | int journal_file_append_object(JournalFile *f, ObjectType type, uint64_t size, Object **ret, uint64_t *offset) { | |
977 | int r; | |
978 | uint64_t p; | |
979 | Object *tail, *o; | |
980 | void *t; | |
981 | ||
982 | assert(f); | |
983 | assert(f->header); | |
984 | assert(type > OBJECT_UNUSED && type < _OBJECT_TYPE_MAX); | |
985 | assert(size >= sizeof(ObjectHeader)); | |
986 | assert(offset); | |
987 | assert(ret); | |
988 | ||
989 | r = journal_file_set_online(f); | |
990 | if (r < 0) | |
991 | return r; | |
992 | ||
993 | p = le64toh(f->header->tail_object_offset); | |
994 | if (p == 0) | |
995 | p = le64toh(f->header->header_size); | |
996 | else { | |
997 | r = journal_file_move_to_object(f, OBJECT_UNUSED, p, &tail); | |
998 | if (r < 0) | |
999 | return r; | |
1000 | ||
1001 | p += ALIGN64(le64toh(tail->object.size)); | |
1002 | } | |
1003 | ||
1004 | r = journal_file_allocate(f, p, size); | |
1005 | if (r < 0) | |
1006 | return r; | |
1007 | ||
1008 | r = journal_file_move_to(f, type, false, p, size, &t, NULL); | |
1009 | if (r < 0) | |
1010 | return r; | |
1011 | ||
1012 | o = (Object*) t; | |
1013 | ||
1014 | zero(o->object); | |
1015 | o->object.type = type; | |
1016 | o->object.size = htole64(size); | |
1017 | ||
1018 | f->header->tail_object_offset = htole64(p); | |
1019 | f->header->n_objects = htole64(le64toh(f->header->n_objects) + 1); | |
1020 | ||
1021 | *ret = o; | |
1022 | *offset = p; | |
1023 | ||
1024 | return 0; | |
1025 | } | |
1026 | ||
1027 | static int journal_file_setup_data_hash_table(JournalFile *f) { | |
1028 | uint64_t s, p; | |
1029 | Object *o; | |
1030 | int r; | |
1031 | ||
1032 | assert(f); | |
1033 | assert(f->header); | |
1034 | ||
1035 | /* We estimate that we need 1 hash table entry per 768 bytes | |
1036 | of journal file and we want to make sure we never get | |
1037 | beyond 75% fill level. Calculate the hash table size for | |
1038 | the maximum file size based on these metrics. */ | |
1039 | ||
1040 | s = (f->metrics.max_size * 4 / 768 / 3) * sizeof(HashItem); | |
1041 | if (s < DEFAULT_DATA_HASH_TABLE_SIZE) | |
1042 | s = DEFAULT_DATA_HASH_TABLE_SIZE; | |
1043 | ||
1044 | log_debug("Reserving %"PRIu64" entries in hash table.", s / sizeof(HashItem)); | |
1045 | ||
1046 | r = journal_file_append_object(f, | |
1047 | OBJECT_DATA_HASH_TABLE, | |
1048 | offsetof(Object, hash_table.items) + s, | |
1049 | &o, &p); | |
1050 | if (r < 0) | |
1051 | return r; | |
1052 | ||
1053 | memzero(o->hash_table.items, s); | |
1054 | ||
1055 | f->header->data_hash_table_offset = htole64(p + offsetof(Object, hash_table.items)); | |
1056 | f->header->data_hash_table_size = htole64(s); | |
1057 | ||
1058 | return 0; | |
1059 | } | |
1060 | ||
1061 | static int journal_file_setup_field_hash_table(JournalFile *f) { | |
1062 | uint64_t s, p; | |
1063 | Object *o; | |
1064 | int r; | |
1065 | ||
1066 | assert(f); | |
1067 | assert(f->header); | |
1068 | ||
1069 | /* We use a fixed size hash table for the fields as this | |
1070 | * number should grow very slowly only */ | |
1071 | ||
1072 | s = DEFAULT_FIELD_HASH_TABLE_SIZE; | |
1073 | r = journal_file_append_object(f, | |
1074 | OBJECT_FIELD_HASH_TABLE, | |
1075 | offsetof(Object, hash_table.items) + s, | |
1076 | &o, &p); | |
1077 | if (r < 0) | |
1078 | return r; | |
1079 | ||
1080 | memzero(o->hash_table.items, s); | |
1081 | ||
1082 | f->header->field_hash_table_offset = htole64(p + offsetof(Object, hash_table.items)); | |
1083 | f->header->field_hash_table_size = htole64(s); | |
1084 | ||
1085 | return 0; | |
1086 | } | |
1087 | ||
1088 | int journal_file_map_data_hash_table(JournalFile *f) { | |
1089 | uint64_t s, p; | |
1090 | void *t; | |
1091 | int r; | |
1092 | ||
1093 | assert(f); | |
1094 | assert(f->header); | |
1095 | ||
1096 | if (f->data_hash_table) | |
1097 | return 0; | |
1098 | ||
1099 | p = le64toh(f->header->data_hash_table_offset); | |
1100 | s = le64toh(f->header->data_hash_table_size); | |
1101 | ||
1102 | r = journal_file_move_to(f, | |
1103 | OBJECT_DATA_HASH_TABLE, | |
1104 | true, | |
1105 | p, s, | |
1106 | &t, NULL); | |
1107 | if (r < 0) | |
1108 | return r; | |
1109 | ||
1110 | f->data_hash_table = t; | |
1111 | return 0; | |
1112 | } | |
1113 | ||
1114 | int journal_file_map_field_hash_table(JournalFile *f) { | |
1115 | uint64_t s, p; | |
1116 | void *t; | |
1117 | int r; | |
1118 | ||
1119 | assert(f); | |
1120 | assert(f->header); | |
1121 | ||
1122 | if (f->field_hash_table) | |
1123 | return 0; | |
1124 | ||
1125 | p = le64toh(f->header->field_hash_table_offset); | |
1126 | s = le64toh(f->header->field_hash_table_size); | |
1127 | ||
1128 | r = journal_file_move_to(f, | |
1129 | OBJECT_FIELD_HASH_TABLE, | |
1130 | true, | |
1131 | p, s, | |
1132 | &t, NULL); | |
1133 | if (r < 0) | |
1134 | return r; | |
1135 | ||
1136 | f->field_hash_table = t; | |
1137 | return 0; | |
1138 | } | |
1139 | ||
1140 | static int journal_file_link_field( | |
1141 | JournalFile *f, | |
1142 | Object *o, | |
1143 | uint64_t offset, | |
1144 | uint64_t hash) { | |
1145 | ||
1146 | uint64_t p, h, m; | |
1147 | int r; | |
1148 | ||
1149 | assert(f); | |
1150 | assert(f->header); | |
1151 | assert(f->field_hash_table); | |
1152 | assert(o); | |
1153 | assert(offset > 0); | |
1154 | ||
1155 | if (o->object.type != OBJECT_FIELD) | |
1156 | return -EINVAL; | |
1157 | ||
1158 | m = le64toh(f->header->field_hash_table_size) / sizeof(HashItem); | |
1159 | if (m <= 0) | |
1160 | return -EBADMSG; | |
1161 | ||
1162 | /* This might alter the window we are looking at */ | |
1163 | o->field.next_hash_offset = o->field.head_data_offset = 0; | |
1164 | ||
1165 | h = hash % m; | |
1166 | p = le64toh(f->field_hash_table[h].tail_hash_offset); | |
1167 | if (p == 0) | |
1168 | f->field_hash_table[h].head_hash_offset = htole64(offset); | |
1169 | else { | |
1170 | r = journal_file_move_to_object(f, OBJECT_FIELD, p, &o); | |
1171 | if (r < 0) | |
1172 | return r; | |
1173 | ||
1174 | o->field.next_hash_offset = htole64(offset); | |
1175 | } | |
1176 | ||
1177 | f->field_hash_table[h].tail_hash_offset = htole64(offset); | |
1178 | ||
1179 | if (JOURNAL_HEADER_CONTAINS(f->header, n_fields)) | |
1180 | f->header->n_fields = htole64(le64toh(f->header->n_fields) + 1); | |
1181 | ||
1182 | return 0; | |
1183 | } | |
1184 | ||
1185 | static int journal_file_link_data( | |
1186 | JournalFile *f, | |
1187 | Object *o, | |
1188 | uint64_t offset, | |
1189 | uint64_t hash) { | |
1190 | ||
1191 | uint64_t p, h, m; | |
1192 | int r; | |
1193 | ||
1194 | assert(f); | |
1195 | assert(f->header); | |
1196 | assert(f->data_hash_table); | |
1197 | assert(o); | |
1198 | assert(offset > 0); | |
1199 | ||
1200 | if (o->object.type != OBJECT_DATA) | |
1201 | return -EINVAL; | |
1202 | ||
1203 | m = le64toh(f->header->data_hash_table_size) / sizeof(HashItem); | |
1204 | if (m <= 0) | |
1205 | return -EBADMSG; | |
1206 | ||
1207 | /* This might alter the window we are looking at */ | |
1208 | o->data.next_hash_offset = o->data.next_field_offset = 0; | |
1209 | o->data.entry_offset = o->data.entry_array_offset = 0; | |
1210 | o->data.n_entries = 0; | |
1211 | ||
1212 | h = hash % m; | |
1213 | p = le64toh(f->data_hash_table[h].tail_hash_offset); | |
1214 | if (p == 0) | |
1215 | /* Only entry in the hash table is easy */ | |
1216 | f->data_hash_table[h].head_hash_offset = htole64(offset); | |
1217 | else { | |
1218 | /* Move back to the previous data object, to patch in | |
1219 | * pointer */ | |
1220 | ||
1221 | r = journal_file_move_to_object(f, OBJECT_DATA, p, &o); | |
1222 | if (r < 0) | |
1223 | return r; | |
1224 | ||
1225 | o->data.next_hash_offset = htole64(offset); | |
1226 | } | |
1227 | ||
1228 | f->data_hash_table[h].tail_hash_offset = htole64(offset); | |
1229 | ||
1230 | if (JOURNAL_HEADER_CONTAINS(f->header, n_data)) | |
1231 | f->header->n_data = htole64(le64toh(f->header->n_data) + 1); | |
1232 | ||
1233 | return 0; | |
1234 | } | |
1235 | ||
1236 | int journal_file_find_field_object_with_hash( | |
1237 | JournalFile *f, | |
1238 | const void *field, uint64_t size, uint64_t hash, | |
1239 | Object **ret, uint64_t *offset) { | |
1240 | ||
1241 | uint64_t p, osize, h, m; | |
1242 | int r; | |
1243 | ||
1244 | assert(f); | |
1245 | assert(f->header); | |
1246 | assert(field && size > 0); | |
1247 | ||
1248 | /* If the field hash table is empty, we can't find anything */ | |
1249 | if (le64toh(f->header->field_hash_table_size) <= 0) | |
1250 | return 0; | |
1251 | ||
1252 | /* Map the field hash table, if it isn't mapped yet. */ | |
1253 | r = journal_file_map_field_hash_table(f); | |
1254 | if (r < 0) | |
1255 | return r; | |
1256 | ||
1257 | osize = offsetof(Object, field.payload) + size; | |
1258 | ||
1259 | m = le64toh(f->header->field_hash_table_size) / sizeof(HashItem); | |
1260 | if (m <= 0) | |
1261 | return -EBADMSG; | |
1262 | ||
1263 | h = hash % m; | |
1264 | p = le64toh(f->field_hash_table[h].head_hash_offset); | |
1265 | ||
1266 | while (p > 0) { | |
1267 | Object *o; | |
1268 | ||
1269 | r = journal_file_move_to_object(f, OBJECT_FIELD, p, &o); | |
1270 | if (r < 0) | |
1271 | return r; | |
1272 | ||
1273 | if (le64toh(o->field.hash) == hash && | |
1274 | le64toh(o->object.size) == osize && | |
1275 | memcmp(o->field.payload, field, size) == 0) { | |
1276 | ||
1277 | if (ret) | |
1278 | *ret = o; | |
1279 | if (offset) | |
1280 | *offset = p; | |
1281 | ||
1282 | return 1; | |
1283 | } | |
1284 | ||
1285 | p = le64toh(o->field.next_hash_offset); | |
1286 | } | |
1287 | ||
1288 | return 0; | |
1289 | } | |
1290 | ||
1291 | int journal_file_find_field_object( | |
1292 | JournalFile *f, | |
1293 | const void *field, uint64_t size, | |
1294 | Object **ret, uint64_t *offset) { | |
1295 | ||
1296 | uint64_t hash; | |
1297 | ||
1298 | assert(f); | |
1299 | assert(field && size > 0); | |
1300 | ||
1301 | hash = hash64(field, size); | |
1302 | ||
1303 | return journal_file_find_field_object_with_hash(f, | |
1304 | field, size, hash, | |
1305 | ret, offset); | |
1306 | } | |
1307 | ||
1308 | int journal_file_find_data_object_with_hash( | |
1309 | JournalFile *f, | |
1310 | const void *data, uint64_t size, uint64_t hash, | |
1311 | Object **ret, uint64_t *offset) { | |
1312 | ||
1313 | uint64_t p, osize, h, m; | |
1314 | int r; | |
1315 | ||
1316 | assert(f); | |
1317 | assert(f->header); | |
1318 | assert(data || size == 0); | |
1319 | ||
1320 | /* If there's no data hash table, then there's no entry. */ | |
1321 | if (le64toh(f->header->data_hash_table_size) <= 0) | |
1322 | return 0; | |
1323 | ||
1324 | /* Map the data hash table, if it isn't mapped yet. */ | |
1325 | r = journal_file_map_data_hash_table(f); | |
1326 | if (r < 0) | |
1327 | return r; | |
1328 | ||
1329 | osize = offsetof(Object, data.payload) + size; | |
1330 | ||
1331 | m = le64toh(f->header->data_hash_table_size) / sizeof(HashItem); | |
1332 | if (m <= 0) | |
1333 | return -EBADMSG; | |
1334 | ||
1335 | h = hash % m; | |
1336 | p = le64toh(f->data_hash_table[h].head_hash_offset); | |
1337 | ||
1338 | while (p > 0) { | |
1339 | Object *o; | |
1340 | ||
1341 | r = journal_file_move_to_object(f, OBJECT_DATA, p, &o); | |
1342 | if (r < 0) | |
1343 | return r; | |
1344 | ||
1345 | if (le64toh(o->data.hash) != hash) | |
1346 | goto next; | |
1347 | ||
1348 | if (o->object.flags & OBJECT_COMPRESSION_MASK) { | |
1349 | #if HAVE_XZ || HAVE_LZ4 | |
1350 | uint64_t l; | |
1351 | size_t rsize = 0; | |
1352 | ||
1353 | l = le64toh(o->object.size); | |
1354 | if (l <= offsetof(Object, data.payload)) | |
1355 | return -EBADMSG; | |
1356 | ||
1357 | l -= offsetof(Object, data.payload); | |
1358 | ||
1359 | r = decompress_blob(o->object.flags & OBJECT_COMPRESSION_MASK, | |
1360 | o->data.payload, l, &f->compress_buffer, &f->compress_buffer_size, &rsize, 0); | |
1361 | if (r < 0) | |
1362 | return r; | |
1363 | ||
1364 | if (rsize == size && | |
1365 | memcmp(f->compress_buffer, data, size) == 0) { | |
1366 | ||
1367 | if (ret) | |
1368 | *ret = o; | |
1369 | ||
1370 | if (offset) | |
1371 | *offset = p; | |
1372 | ||
1373 | return 1; | |
1374 | } | |
1375 | #else | |
1376 | return -EPROTONOSUPPORT; | |
1377 | #endif | |
1378 | } else if (le64toh(o->object.size) == osize && | |
1379 | memcmp(o->data.payload, data, size) == 0) { | |
1380 | ||
1381 | if (ret) | |
1382 | *ret = o; | |
1383 | ||
1384 | if (offset) | |
1385 | *offset = p; | |
1386 | ||
1387 | return 1; | |
1388 | } | |
1389 | ||
1390 | next: | |
1391 | p = le64toh(o->data.next_hash_offset); | |
1392 | } | |
1393 | ||
1394 | return 0; | |
1395 | } | |
1396 | ||
1397 | int journal_file_find_data_object( | |
1398 | JournalFile *f, | |
1399 | const void *data, uint64_t size, | |
1400 | Object **ret, uint64_t *offset) { | |
1401 | ||
1402 | uint64_t hash; | |
1403 | ||
1404 | assert(f); | |
1405 | assert(data || size == 0); | |
1406 | ||
1407 | hash = hash64(data, size); | |
1408 | ||
1409 | return journal_file_find_data_object_with_hash(f, | |
1410 | data, size, hash, | |
1411 | ret, offset); | |
1412 | } | |
1413 | ||
1414 | static int journal_file_append_field( | |
1415 | JournalFile *f, | |
1416 | const void *field, uint64_t size, | |
1417 | Object **ret, uint64_t *offset) { | |
1418 | ||
1419 | uint64_t hash, p; | |
1420 | uint64_t osize; | |
1421 | Object *o; | |
1422 | int r; | |
1423 | ||
1424 | assert(f); | |
1425 | assert(field && size > 0); | |
1426 | ||
1427 | hash = hash64(field, size); | |
1428 | ||
1429 | r = journal_file_find_field_object_with_hash(f, field, size, hash, &o, &p); | |
1430 | if (r < 0) | |
1431 | return r; | |
1432 | else if (r > 0) { | |
1433 | ||
1434 | if (ret) | |
1435 | *ret = o; | |
1436 | ||
1437 | if (offset) | |
1438 | *offset = p; | |
1439 | ||
1440 | return 0; | |
1441 | } | |
1442 | ||
1443 | osize = offsetof(Object, field.payload) + size; | |
1444 | r = journal_file_append_object(f, OBJECT_FIELD, osize, &o, &p); | |
1445 | if (r < 0) | |
1446 | return r; | |
1447 | ||
1448 | o->field.hash = htole64(hash); | |
1449 | memcpy(o->field.payload, field, size); | |
1450 | ||
1451 | r = journal_file_link_field(f, o, p, hash); | |
1452 | if (r < 0) | |
1453 | return r; | |
1454 | ||
1455 | /* The linking might have altered the window, so let's | |
1456 | * refresh our pointer */ | |
1457 | r = journal_file_move_to_object(f, OBJECT_FIELD, p, &o); | |
1458 | if (r < 0) | |
1459 | return r; | |
1460 | ||
1461 | #if HAVE_GCRYPT | |
1462 | r = journal_file_hmac_put_object(f, OBJECT_FIELD, o, p); | |
1463 | if (r < 0) | |
1464 | return r; | |
1465 | #endif | |
1466 | ||
1467 | if (ret) | |
1468 | *ret = o; | |
1469 | ||
1470 | if (offset) | |
1471 | *offset = p; | |
1472 | ||
1473 | return 0; | |
1474 | } | |
1475 | ||
1476 | static int journal_file_append_data( | |
1477 | JournalFile *f, | |
1478 | const void *data, uint64_t size, | |
1479 | Object **ret, uint64_t *offset) { | |
1480 | ||
1481 | uint64_t hash, p; | |
1482 | uint64_t osize; | |
1483 | Object *o; | |
1484 | int r, compression = 0; | |
1485 | const void *eq; | |
1486 | ||
1487 | assert(f); | |
1488 | assert(data || size == 0); | |
1489 | ||
1490 | hash = hash64(data, size); | |
1491 | ||
1492 | r = journal_file_find_data_object_with_hash(f, data, size, hash, &o, &p); | |
1493 | if (r < 0) | |
1494 | return r; | |
1495 | if (r > 0) { | |
1496 | ||
1497 | if (ret) | |
1498 | *ret = o; | |
1499 | ||
1500 | if (offset) | |
1501 | *offset = p; | |
1502 | ||
1503 | return 0; | |
1504 | } | |
1505 | ||
1506 | osize = offsetof(Object, data.payload) + size; | |
1507 | r = journal_file_append_object(f, OBJECT_DATA, osize, &o, &p); | |
1508 | if (r < 0) | |
1509 | return r; | |
1510 | ||
1511 | o->data.hash = htole64(hash); | |
1512 | ||
1513 | #if HAVE_XZ || HAVE_LZ4 | |
1514 | if (JOURNAL_FILE_COMPRESS(f) && size >= f->compress_threshold_bytes) { | |
1515 | size_t rsize = 0; | |
1516 | ||
1517 | compression = compress_blob(data, size, o->data.payload, size - 1, &rsize); | |
1518 | ||
1519 | if (compression >= 0) { | |
1520 | o->object.size = htole64(offsetof(Object, data.payload) + rsize); | |
1521 | o->object.flags |= compression; | |
1522 | ||
1523 | log_debug("Compressed data object %"PRIu64" -> %zu using %s", | |
1524 | size, rsize, object_compressed_to_string(compression)); | |
1525 | } else | |
1526 | /* Compression didn't work, we don't really care why, let's continue without compression */ | |
1527 | compression = 0; | |
1528 | } | |
1529 | #endif | |
1530 | ||
1531 | if (compression == 0) | |
1532 | memcpy_safe(o->data.payload, data, size); | |
1533 | ||
1534 | r = journal_file_link_data(f, o, p, hash); | |
1535 | if (r < 0) | |
1536 | return r; | |
1537 | ||
1538 | #if HAVE_GCRYPT | |
1539 | r = journal_file_hmac_put_object(f, OBJECT_DATA, o, p); | |
1540 | if (r < 0) | |
1541 | return r; | |
1542 | #endif | |
1543 | ||
1544 | /* The linking might have altered the window, so let's | |
1545 | * refresh our pointer */ | |
1546 | r = journal_file_move_to_object(f, OBJECT_DATA, p, &o); | |
1547 | if (r < 0) | |
1548 | return r; | |
1549 | ||
1550 | if (!data) | |
1551 | eq = NULL; | |
1552 | else | |
1553 | eq = memchr(data, '=', size); | |
1554 | if (eq && eq > data) { | |
1555 | Object *fo = NULL; | |
1556 | uint64_t fp; | |
1557 | ||
1558 | /* Create field object ... */ | |
1559 | r = journal_file_append_field(f, data, (uint8_t*) eq - (uint8_t*) data, &fo, &fp); | |
1560 | if (r < 0) | |
1561 | return r; | |
1562 | ||
1563 | /* ... and link it in. */ | |
1564 | o->data.next_field_offset = fo->field.head_data_offset; | |
1565 | fo->field.head_data_offset = le64toh(p); | |
1566 | } | |
1567 | ||
1568 | if (ret) | |
1569 | *ret = o; | |
1570 | ||
1571 | if (offset) | |
1572 | *offset = p; | |
1573 | ||
1574 | return 0; | |
1575 | } | |
1576 | ||
1577 | uint64_t journal_file_entry_n_items(Object *o) { | |
1578 | assert(o); | |
1579 | ||
1580 | if (o->object.type != OBJECT_ENTRY) | |
1581 | return 0; | |
1582 | ||
1583 | return (le64toh(o->object.size) - offsetof(Object, entry.items)) / sizeof(EntryItem); | |
1584 | } | |
1585 | ||
1586 | uint64_t journal_file_entry_array_n_items(Object *o) { | |
1587 | assert(o); | |
1588 | ||
1589 | if (o->object.type != OBJECT_ENTRY_ARRAY) | |
1590 | return 0; | |
1591 | ||
1592 | return (le64toh(o->object.size) - offsetof(Object, entry_array.items)) / sizeof(uint64_t); | |
1593 | } | |
1594 | ||
1595 | uint64_t journal_file_hash_table_n_items(Object *o) { | |
1596 | assert(o); | |
1597 | ||
1598 | if (!IN_SET(o->object.type, OBJECT_DATA_HASH_TABLE, OBJECT_FIELD_HASH_TABLE)) | |
1599 | return 0; | |
1600 | ||
1601 | return (le64toh(o->object.size) - offsetof(Object, hash_table.items)) / sizeof(HashItem); | |
1602 | } | |
1603 | ||
1604 | static int link_entry_into_array(JournalFile *f, | |
1605 | le64_t *first, | |
1606 | le64_t *idx, | |
1607 | uint64_t p) { | |
1608 | int r; | |
1609 | uint64_t n = 0, ap = 0, q, i, a, hidx; | |
1610 | Object *o; | |
1611 | ||
1612 | assert(f); | |
1613 | assert(f->header); | |
1614 | assert(first); | |
1615 | assert(idx); | |
1616 | assert(p > 0); | |
1617 | ||
1618 | a = le64toh(*first); | |
1619 | i = hidx = le64toh(*idx); | |
1620 | while (a > 0) { | |
1621 | ||
1622 | r = journal_file_move_to_object(f, OBJECT_ENTRY_ARRAY, a, &o); | |
1623 | if (r < 0) | |
1624 | return r; | |
1625 | ||
1626 | n = journal_file_entry_array_n_items(o); | |
1627 | if (i < n) { | |
1628 | o->entry_array.items[i] = htole64(p); | |
1629 | *idx = htole64(hidx + 1); | |
1630 | return 0; | |
1631 | } | |
1632 | ||
1633 | i -= n; | |
1634 | ap = a; | |
1635 | a = le64toh(o->entry_array.next_entry_array_offset); | |
1636 | } | |
1637 | ||
1638 | if (hidx > n) | |
1639 | n = (hidx+1) * 2; | |
1640 | else | |
1641 | n = n * 2; | |
1642 | ||
1643 | if (n < 4) | |
1644 | n = 4; | |
1645 | ||
1646 | r = journal_file_append_object(f, OBJECT_ENTRY_ARRAY, | |
1647 | offsetof(Object, entry_array.items) + n * sizeof(uint64_t), | |
1648 | &o, &q); | |
1649 | if (r < 0) | |
1650 | return r; | |
1651 | ||
1652 | #if HAVE_GCRYPT | |
1653 | r = journal_file_hmac_put_object(f, OBJECT_ENTRY_ARRAY, o, q); | |
1654 | if (r < 0) | |
1655 | return r; | |
1656 | #endif | |
1657 | ||
1658 | o->entry_array.items[i] = htole64(p); | |
1659 | ||
1660 | if (ap == 0) | |
1661 | *first = htole64(q); | |
1662 | else { | |
1663 | r = journal_file_move_to_object(f, OBJECT_ENTRY_ARRAY, ap, &o); | |
1664 | if (r < 0) | |
1665 | return r; | |
1666 | ||
1667 | o->entry_array.next_entry_array_offset = htole64(q); | |
1668 | } | |
1669 | ||
1670 | if (JOURNAL_HEADER_CONTAINS(f->header, n_entry_arrays)) | |
1671 | f->header->n_entry_arrays = htole64(le64toh(f->header->n_entry_arrays) + 1); | |
1672 | ||
1673 | *idx = htole64(hidx + 1); | |
1674 | ||
1675 | return 0; | |
1676 | } | |
1677 | ||
1678 | static int link_entry_into_array_plus_one(JournalFile *f, | |
1679 | le64_t *extra, | |
1680 | le64_t *first, | |
1681 | le64_t *idx, | |
1682 | uint64_t p) { | |
1683 | ||
1684 | int r; | |
1685 | ||
1686 | assert(f); | |
1687 | assert(extra); | |
1688 | assert(first); | |
1689 | assert(idx); | |
1690 | assert(p > 0); | |
1691 | ||
1692 | if (*idx == 0) | |
1693 | *extra = htole64(p); | |
1694 | else { | |
1695 | le64_t i; | |
1696 | ||
1697 | i = htole64(le64toh(*idx) - 1); | |
1698 | r = link_entry_into_array(f, first, &i, p); | |
1699 | if (r < 0) | |
1700 | return r; | |
1701 | } | |
1702 | ||
1703 | *idx = htole64(le64toh(*idx) + 1); | |
1704 | return 0; | |
1705 | } | |
1706 | ||
1707 | static int journal_file_link_entry_item(JournalFile *f, Object *o, uint64_t offset, uint64_t i) { | |
1708 | uint64_t p; | |
1709 | int r; | |
1710 | assert(f); | |
1711 | assert(o); | |
1712 | assert(offset > 0); | |
1713 | ||
1714 | p = le64toh(o->entry.items[i].object_offset); | |
1715 | if (p == 0) | |
1716 | return -EINVAL; | |
1717 | ||
1718 | r = journal_file_move_to_object(f, OBJECT_DATA, p, &o); | |
1719 | if (r < 0) | |
1720 | return r; | |
1721 | ||
1722 | return link_entry_into_array_plus_one(f, | |
1723 | &o->data.entry_offset, | |
1724 | &o->data.entry_array_offset, | |
1725 | &o->data.n_entries, | |
1726 | offset); | |
1727 | } | |
1728 | ||
1729 | static int journal_file_link_entry(JournalFile *f, Object *o, uint64_t offset) { | |
1730 | uint64_t n, i; | |
1731 | int r; | |
1732 | ||
1733 | assert(f); | |
1734 | assert(f->header); | |
1735 | assert(o); | |
1736 | assert(offset > 0); | |
1737 | ||
1738 | if (o->object.type != OBJECT_ENTRY) | |
1739 | return -EINVAL; | |
1740 | ||
1741 | __sync_synchronize(); | |
1742 | ||
1743 | /* Link up the entry itself */ | |
1744 | r = link_entry_into_array(f, | |
1745 | &f->header->entry_array_offset, | |
1746 | &f->header->n_entries, | |
1747 | offset); | |
1748 | if (r < 0) | |
1749 | return r; | |
1750 | ||
1751 | /* log_debug("=> %s seqnr=%"PRIu64" n_entries=%"PRIu64, f->path, o->entry.seqnum, f->header->n_entries); */ | |
1752 | ||
1753 | if (f->header->head_entry_realtime == 0) | |
1754 | f->header->head_entry_realtime = o->entry.realtime; | |
1755 | ||
1756 | f->header->tail_entry_realtime = o->entry.realtime; | |
1757 | f->header->tail_entry_monotonic = o->entry.monotonic; | |
1758 | ||
1759 | /* Link up the items */ | |
1760 | n = journal_file_entry_n_items(o); | |
1761 | for (i = 0; i < n; i++) { | |
1762 | r = journal_file_link_entry_item(f, o, offset, i); | |
1763 | if (r < 0) | |
1764 | return r; | |
1765 | } | |
1766 | ||
1767 | return 0; | |
1768 | } | |
1769 | ||
1770 | static int journal_file_append_entry_internal( | |
1771 | JournalFile *f, | |
1772 | const dual_timestamp *ts, | |
1773 | const sd_id128_t *boot_id, | |
1774 | uint64_t xor_hash, | |
1775 | const EntryItem items[], unsigned n_items, | |
1776 | uint64_t *seqnum, | |
1777 | Object **ret, uint64_t *offset) { | |
1778 | uint64_t np; | |
1779 | uint64_t osize; | |
1780 | Object *o; | |
1781 | int r; | |
1782 | ||
1783 | assert(f); | |
1784 | assert(f->header); | |
1785 | assert(items || n_items == 0); | |
1786 | assert(ts); | |
1787 | ||
1788 | osize = offsetof(Object, entry.items) + (n_items * sizeof(EntryItem)); | |
1789 | ||
1790 | r = journal_file_append_object(f, OBJECT_ENTRY, osize, &o, &np); | |
1791 | if (r < 0) | |
1792 | return r; | |
1793 | ||
1794 | o->entry.seqnum = htole64(journal_file_entry_seqnum(f, seqnum)); | |
1795 | memcpy_safe(o->entry.items, items, n_items * sizeof(EntryItem)); | |
1796 | o->entry.realtime = htole64(ts->realtime); | |
1797 | o->entry.monotonic = htole64(ts->monotonic); | |
1798 | o->entry.xor_hash = htole64(xor_hash); | |
1799 | if (boot_id) | |
1800 | f->header->boot_id = *boot_id; | |
1801 | o->entry.boot_id = f->header->boot_id; | |
1802 | ||
1803 | #if HAVE_GCRYPT | |
1804 | r = journal_file_hmac_put_object(f, OBJECT_ENTRY, o, np); | |
1805 | if (r < 0) | |
1806 | return r; | |
1807 | #endif | |
1808 | ||
1809 | r = journal_file_link_entry(f, o, np); | |
1810 | if (r < 0) | |
1811 | return r; | |
1812 | ||
1813 | if (ret) | |
1814 | *ret = o; | |
1815 | ||
1816 | if (offset) | |
1817 | *offset = np; | |
1818 | ||
1819 | return 0; | |
1820 | } | |
1821 | ||
1822 | void journal_file_post_change(JournalFile *f) { | |
1823 | assert(f); | |
1824 | ||
1825 | if (f->fd < 0) | |
1826 | return; | |
1827 | ||
1828 | /* inotify() does not receive IN_MODIFY events from file | |
1829 | * accesses done via mmap(). After each access we hence | |
1830 | * trigger IN_MODIFY by truncating the journal file to its | |
1831 | * current size which triggers IN_MODIFY. */ | |
1832 | ||
1833 | __sync_synchronize(); | |
1834 | ||
1835 | if (ftruncate(f->fd, f->last_stat.st_size) < 0) | |
1836 | log_debug_errno(errno, "Failed to truncate file to its own size: %m"); | |
1837 | } | |
1838 | ||
1839 | static int post_change_thunk(sd_event_source *timer, uint64_t usec, void *userdata) { | |
1840 | assert(userdata); | |
1841 | ||
1842 | journal_file_post_change(userdata); | |
1843 | ||
1844 | return 1; | |
1845 | } | |
1846 | ||
1847 | static void schedule_post_change(JournalFile *f) { | |
1848 | uint64_t now; | |
1849 | int r; | |
1850 | ||
1851 | assert(f); | |
1852 | assert(f->post_change_timer); | |
1853 | ||
1854 | r = sd_event_source_get_enabled(f->post_change_timer, NULL); | |
1855 | if (r < 0) { | |
1856 | log_debug_errno(r, "Failed to get ftruncate timer state: %m"); | |
1857 | goto fail; | |
1858 | } | |
1859 | if (r > 0) | |
1860 | return; | |
1861 | ||
1862 | r = sd_event_now(sd_event_source_get_event(f->post_change_timer), CLOCK_MONOTONIC, &now); | |
1863 | if (r < 0) { | |
1864 | log_debug_errno(r, "Failed to get clock's now for scheduling ftruncate: %m"); | |
1865 | goto fail; | |
1866 | } | |
1867 | ||
1868 | r = sd_event_source_set_time(f->post_change_timer, now + f->post_change_timer_period); | |
1869 | if (r < 0) { | |
1870 | log_debug_errno(r, "Failed to set time for scheduling ftruncate: %m"); | |
1871 | goto fail; | |
1872 | } | |
1873 | ||
1874 | r = sd_event_source_set_enabled(f->post_change_timer, SD_EVENT_ONESHOT); | |
1875 | if (r < 0) { | |
1876 | log_debug_errno(r, "Failed to enable scheduled ftruncate: %m"); | |
1877 | goto fail; | |
1878 | } | |
1879 | ||
1880 | return; | |
1881 | ||
1882 | fail: | |
1883 | /* On failure, let's simply post the change immediately. */ | |
1884 | journal_file_post_change(f); | |
1885 | } | |
1886 | ||
1887 | /* Enable coalesced change posting in a timer on the provided sd_event instance */ | |
1888 | int journal_file_enable_post_change_timer(JournalFile *f, sd_event *e, usec_t t) { | |
1889 | _cleanup_(sd_event_source_unrefp) sd_event_source *timer = NULL; | |
1890 | int r; | |
1891 | ||
1892 | assert(f); | |
1893 | assert_return(!f->post_change_timer, -EINVAL); | |
1894 | assert(e); | |
1895 | assert(t); | |
1896 | ||
1897 | r = sd_event_add_time(e, &timer, CLOCK_MONOTONIC, 0, 0, post_change_thunk, f); | |
1898 | if (r < 0) | |
1899 | return r; | |
1900 | ||
1901 | r = sd_event_source_set_enabled(timer, SD_EVENT_OFF); | |
1902 | if (r < 0) | |
1903 | return r; | |
1904 | ||
1905 | f->post_change_timer = TAKE_PTR(timer); | |
1906 | f->post_change_timer_period = t; | |
1907 | ||
1908 | return r; | |
1909 | } | |
1910 | ||
1911 | static int entry_item_cmp(const EntryItem *a, const EntryItem *b) { | |
1912 | return CMP(le64toh(a->object_offset), le64toh(b->object_offset)); | |
1913 | } | |
1914 | ||
1915 | int journal_file_append_entry( | |
1916 | JournalFile *f, | |
1917 | const dual_timestamp *ts, | |
1918 | const sd_id128_t *boot_id, | |
1919 | const struct iovec iovec[], unsigned n_iovec, | |
1920 | uint64_t *seqnum, | |
1921 | Object **ret, uint64_t *offset) { | |
1922 | ||
1923 | unsigned i; | |
1924 | EntryItem *items; | |
1925 | int r; | |
1926 | uint64_t xor_hash = 0; | |
1927 | struct dual_timestamp _ts; | |
1928 | ||
1929 | assert(f); | |
1930 | assert(f->header); | |
1931 | assert(iovec || n_iovec == 0); | |
1932 | ||
1933 | if (ts) { | |
1934 | if (!VALID_REALTIME(ts->realtime)) | |
1935 | return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG), | |
1936 | "Invalid realtime timestamp %" PRIu64 ", refusing entry.", | |
1937 | ts->realtime); | |
1938 | if (!VALID_MONOTONIC(ts->monotonic)) | |
1939 | return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG), | |
1940 | "Invalid monotomic timestamp %" PRIu64 ", refusing entry.", | |
1941 | ts->monotonic); | |
1942 | } else { | |
1943 | dual_timestamp_get(&_ts); | |
1944 | ts = &_ts; | |
1945 | } | |
1946 | ||
1947 | #if HAVE_GCRYPT | |
1948 | r = journal_file_maybe_append_tag(f, ts->realtime); | |
1949 | if (r < 0) | |
1950 | return r; | |
1951 | #endif | |
1952 | ||
1953 | /* alloca() can't take 0, hence let's allocate at least one */ | |
1954 | items = newa(EntryItem, MAX(1u, n_iovec)); | |
1955 | ||
1956 | for (i = 0; i < n_iovec; i++) { | |
1957 | uint64_t p; | |
1958 | Object *o; | |
1959 | ||
1960 | r = journal_file_append_data(f, iovec[i].iov_base, iovec[i].iov_len, &o, &p); | |
1961 | if (r < 0) | |
1962 | return r; | |
1963 | ||
1964 | xor_hash ^= le64toh(o->data.hash); | |
1965 | items[i].object_offset = htole64(p); | |
1966 | items[i].hash = o->data.hash; | |
1967 | } | |
1968 | ||
1969 | /* Order by the position on disk, in order to improve seek | |
1970 | * times for rotating media. */ | |
1971 | typesafe_qsort(items, n_iovec, entry_item_cmp); | |
1972 | ||
1973 | r = journal_file_append_entry_internal(f, ts, boot_id, xor_hash, items, n_iovec, seqnum, ret, offset); | |
1974 | ||
1975 | /* If the memory mapping triggered a SIGBUS then we return an | |
1976 | * IO error and ignore the error code passed down to us, since | |
1977 | * it is very likely just an effect of a nullified replacement | |
1978 | * mapping page */ | |
1979 | ||
1980 | if (mmap_cache_got_sigbus(f->mmap, f->cache_fd)) | |
1981 | r = -EIO; | |
1982 | ||
1983 | if (f->post_change_timer) | |
1984 | schedule_post_change(f); | |
1985 | else | |
1986 | journal_file_post_change(f); | |
1987 | ||
1988 | return r; | |
1989 | } | |
1990 | ||
1991 | typedef struct ChainCacheItem { | |
1992 | uint64_t first; /* the array at the beginning of the chain */ | |
1993 | uint64_t array; /* the cached array */ | |
1994 | uint64_t begin; /* the first item in the cached array */ | |
1995 | uint64_t total; /* the total number of items in all arrays before this one in the chain */ | |
1996 | uint64_t last_index; /* the last index we looked at, to optimize locality when bisecting */ | |
1997 | } ChainCacheItem; | |
1998 | ||
1999 | static void chain_cache_put( | |
2000 | OrderedHashmap *h, | |
2001 | ChainCacheItem *ci, | |
2002 | uint64_t first, | |
2003 | uint64_t array, | |
2004 | uint64_t begin, | |
2005 | uint64_t total, | |
2006 | uint64_t last_index) { | |
2007 | ||
2008 | if (!ci) { | |
2009 | /* If the chain item to cache for this chain is the | |
2010 | * first one it's not worth caching anything */ | |
2011 | if (array == first) | |
2012 | return; | |
2013 | ||
2014 | if (ordered_hashmap_size(h) >= CHAIN_CACHE_MAX) { | |
2015 | ci = ordered_hashmap_steal_first(h); | |
2016 | assert(ci); | |
2017 | } else { | |
2018 | ci = new(ChainCacheItem, 1); | |
2019 | if (!ci) | |
2020 | return; | |
2021 | } | |
2022 | ||
2023 | ci->first = first; | |
2024 | ||
2025 | if (ordered_hashmap_put(h, &ci->first, ci) < 0) { | |
2026 | free(ci); | |
2027 | return; | |
2028 | } | |
2029 | } else | |
2030 | assert(ci->first == first); | |
2031 | ||
2032 | ci->array = array; | |
2033 | ci->begin = begin; | |
2034 | ci->total = total; | |
2035 | ci->last_index = last_index; | |
2036 | } | |
2037 | ||
2038 | static int generic_array_get( | |
2039 | JournalFile *f, | |
2040 | uint64_t first, | |
2041 | uint64_t i, | |
2042 | Object **ret, uint64_t *offset) { | |
2043 | ||
2044 | Object *o; | |
2045 | uint64_t p = 0, a, t = 0; | |
2046 | int r; | |
2047 | ChainCacheItem *ci; | |
2048 | ||
2049 | assert(f); | |
2050 | ||
2051 | a = first; | |
2052 | ||
2053 | /* Try the chain cache first */ | |
2054 | ci = ordered_hashmap_get(f->chain_cache, &first); | |
2055 | if (ci && i > ci->total) { | |
2056 | a = ci->array; | |
2057 | i -= ci->total; | |
2058 | t = ci->total; | |
2059 | } | |
2060 | ||
2061 | while (a > 0) { | |
2062 | uint64_t k; | |
2063 | ||
2064 | r = journal_file_move_to_object(f, OBJECT_ENTRY_ARRAY, a, &o); | |
2065 | if (r < 0) | |
2066 | return r; | |
2067 | ||
2068 | k = journal_file_entry_array_n_items(o); | |
2069 | if (i < k) { | |
2070 | p = le64toh(o->entry_array.items[i]); | |
2071 | goto found; | |
2072 | } | |
2073 | ||
2074 | i -= k; | |
2075 | t += k; | |
2076 | a = le64toh(o->entry_array.next_entry_array_offset); | |
2077 | } | |
2078 | ||
2079 | return 0; | |
2080 | ||
2081 | found: | |
2082 | /* Let's cache this item for the next invocation */ | |
2083 | chain_cache_put(f->chain_cache, ci, first, a, le64toh(o->entry_array.items[0]), t, i); | |
2084 | ||
2085 | r = journal_file_move_to_object(f, OBJECT_ENTRY, p, &o); | |
2086 | if (r < 0) | |
2087 | return r; | |
2088 | ||
2089 | if (ret) | |
2090 | *ret = o; | |
2091 | ||
2092 | if (offset) | |
2093 | *offset = p; | |
2094 | ||
2095 | return 1; | |
2096 | } | |
2097 | ||
2098 | static int generic_array_get_plus_one( | |
2099 | JournalFile *f, | |
2100 | uint64_t extra, | |
2101 | uint64_t first, | |
2102 | uint64_t i, | |
2103 | Object **ret, uint64_t *offset) { | |
2104 | ||
2105 | Object *o; | |
2106 | ||
2107 | assert(f); | |
2108 | ||
2109 | if (i == 0) { | |
2110 | int r; | |
2111 | ||
2112 | r = journal_file_move_to_object(f, OBJECT_ENTRY, extra, &o); | |
2113 | if (r < 0) | |
2114 | return r; | |
2115 | ||
2116 | if (ret) | |
2117 | *ret = o; | |
2118 | ||
2119 | if (offset) | |
2120 | *offset = extra; | |
2121 | ||
2122 | return 1; | |
2123 | } | |
2124 | ||
2125 | return generic_array_get(f, first, i-1, ret, offset); | |
2126 | } | |
2127 | ||
2128 | enum { | |
2129 | TEST_FOUND, | |
2130 | TEST_LEFT, | |
2131 | TEST_RIGHT | |
2132 | }; | |
2133 | ||
2134 | static int generic_array_bisect( | |
2135 | JournalFile *f, | |
2136 | uint64_t first, | |
2137 | uint64_t n, | |
2138 | uint64_t needle, | |
2139 | int (*test_object)(JournalFile *f, uint64_t p, uint64_t needle), | |
2140 | direction_t direction, | |
2141 | Object **ret, | |
2142 | uint64_t *offset, | |
2143 | uint64_t *idx) { | |
2144 | ||
2145 | uint64_t a, p, t = 0, i = 0, last_p = 0, last_index = (uint64_t) -1; | |
2146 | bool subtract_one = false; | |
2147 | Object *o, *array = NULL; | |
2148 | int r; | |
2149 | ChainCacheItem *ci; | |
2150 | ||
2151 | assert(f); | |
2152 | assert(test_object); | |
2153 | ||
2154 | /* Start with the first array in the chain */ | |
2155 | a = first; | |
2156 | ||
2157 | ci = ordered_hashmap_get(f->chain_cache, &first); | |
2158 | if (ci && n > ci->total && ci->begin != 0) { | |
2159 | /* Ah, we have iterated this bisection array chain | |
2160 | * previously! Let's see if we can skip ahead in the | |
2161 | * chain, as far as the last time. But we can't jump | |
2162 | * backwards in the chain, so let's check that | |
2163 | * first. */ | |
2164 | ||
2165 | r = test_object(f, ci->begin, needle); | |
2166 | if (r < 0) | |
2167 | return r; | |
2168 | ||
2169 | if (r == TEST_LEFT) { | |
2170 | /* OK, what we are looking for is right of the | |
2171 | * begin of this EntryArray, so let's jump | |
2172 | * straight to previously cached array in the | |
2173 | * chain */ | |
2174 | ||
2175 | a = ci->array; | |
2176 | n -= ci->total; | |
2177 | t = ci->total; | |
2178 | last_index = ci->last_index; | |
2179 | } | |
2180 | } | |
2181 | ||
2182 | while (a > 0) { | |
2183 | uint64_t left, right, k, lp; | |
2184 | ||
2185 | r = journal_file_move_to_object(f, OBJECT_ENTRY_ARRAY, a, &array); | |
2186 | if (r < 0) | |
2187 | return r; | |
2188 | ||
2189 | k = journal_file_entry_array_n_items(array); | |
2190 | right = MIN(k, n); | |
2191 | if (right <= 0) | |
2192 | return 0; | |
2193 | ||
2194 | i = right - 1; | |
2195 | lp = p = le64toh(array->entry_array.items[i]); | |
2196 | if (p <= 0) | |
2197 | r = -EBADMSG; | |
2198 | else | |
2199 | r = test_object(f, p, needle); | |
2200 | if (r == -EBADMSG) { | |
2201 | log_debug_errno(r, "Encountered invalid entry while bisecting, cutting algorithm short. (1)"); | |
2202 | n = i; | |
2203 | continue; | |
2204 | } | |
2205 | if (r < 0) | |
2206 | return r; | |
2207 | ||
2208 | if (r == TEST_FOUND) | |
2209 | r = direction == DIRECTION_DOWN ? TEST_RIGHT : TEST_LEFT; | |
2210 | ||
2211 | if (r == TEST_RIGHT) { | |
2212 | left = 0; | |
2213 | right -= 1; | |
2214 | ||
2215 | if (last_index != (uint64_t) -1) { | |
2216 | assert(last_index <= right); | |
2217 | ||
2218 | /* If we cached the last index we | |
2219 | * looked at, let's try to not to jump | |
2220 | * too wildly around and see if we can | |
2221 | * limit the range to look at early to | |
2222 | * the immediate neighbors of the last | |
2223 | * index we looked at. */ | |
2224 | ||
2225 | if (last_index > 0) { | |
2226 | uint64_t x = last_index - 1; | |
2227 | ||
2228 | p = le64toh(array->entry_array.items[x]); | |
2229 | if (p <= 0) | |
2230 | return -EBADMSG; | |
2231 | ||
2232 | r = test_object(f, p, needle); | |
2233 | if (r < 0) | |
2234 | return r; | |
2235 | ||
2236 | if (r == TEST_FOUND) | |
2237 | r = direction == DIRECTION_DOWN ? TEST_RIGHT : TEST_LEFT; | |
2238 | ||
2239 | if (r == TEST_RIGHT) | |
2240 | right = x; | |
2241 | else | |
2242 | left = x + 1; | |
2243 | } | |
2244 | ||
2245 | if (last_index < right) { | |
2246 | uint64_t y = last_index + 1; | |
2247 | ||
2248 | p = le64toh(array->entry_array.items[y]); | |
2249 | if (p <= 0) | |
2250 | return -EBADMSG; | |
2251 | ||
2252 | r = test_object(f, p, needle); | |
2253 | if (r < 0) | |
2254 | return r; | |
2255 | ||
2256 | if (r == TEST_FOUND) | |
2257 | r = direction == DIRECTION_DOWN ? TEST_RIGHT : TEST_LEFT; | |
2258 | ||
2259 | if (r == TEST_RIGHT) | |
2260 | right = y; | |
2261 | else | |
2262 | left = y + 1; | |
2263 | } | |
2264 | } | |
2265 | ||
2266 | for (;;) { | |
2267 | if (left == right) { | |
2268 | if (direction == DIRECTION_UP) | |
2269 | subtract_one = true; | |
2270 | ||
2271 | i = left; | |
2272 | goto found; | |
2273 | } | |
2274 | ||
2275 | assert(left < right); | |
2276 | i = (left + right) / 2; | |
2277 | ||
2278 | p = le64toh(array->entry_array.items[i]); | |
2279 | if (p <= 0) | |
2280 | r = -EBADMSG; | |
2281 | else | |
2282 | r = test_object(f, p, needle); | |
2283 | if (r == -EBADMSG) { | |
2284 | log_debug_errno(r, "Encountered invalid entry while bisecting, cutting algorithm short. (2)"); | |
2285 | right = n = i; | |
2286 | continue; | |
2287 | } | |
2288 | if (r < 0) | |
2289 | return r; | |
2290 | ||
2291 | if (r == TEST_FOUND) | |
2292 | r = direction == DIRECTION_DOWN ? TEST_RIGHT : TEST_LEFT; | |
2293 | ||
2294 | if (r == TEST_RIGHT) | |
2295 | right = i; | |
2296 | else | |
2297 | left = i + 1; | |
2298 | } | |
2299 | } | |
2300 | ||
2301 | if (k >= n) { | |
2302 | if (direction == DIRECTION_UP) { | |
2303 | i = n; | |
2304 | subtract_one = true; | |
2305 | goto found; | |
2306 | } | |
2307 | ||
2308 | return 0; | |
2309 | } | |
2310 | ||
2311 | last_p = lp; | |
2312 | ||
2313 | n -= k; | |
2314 | t += k; | |
2315 | last_index = (uint64_t) -1; | |
2316 | a = le64toh(array->entry_array.next_entry_array_offset); | |
2317 | } | |
2318 | ||
2319 | return 0; | |
2320 | ||
2321 | found: | |
2322 | if (subtract_one && t == 0 && i == 0) | |
2323 | return 0; | |
2324 | ||
2325 | /* Let's cache this item for the next invocation */ | |
2326 | 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); | |
2327 | ||
2328 | if (subtract_one && i == 0) | |
2329 | p = last_p; | |
2330 | else if (subtract_one) | |
2331 | p = le64toh(array->entry_array.items[i-1]); | |
2332 | else | |
2333 | p = le64toh(array->entry_array.items[i]); | |
2334 | ||
2335 | r = journal_file_move_to_object(f, OBJECT_ENTRY, p, &o); | |
2336 | if (r < 0) | |
2337 | return r; | |
2338 | ||
2339 | if (ret) | |
2340 | *ret = o; | |
2341 | ||
2342 | if (offset) | |
2343 | *offset = p; | |
2344 | ||
2345 | if (idx) | |
2346 | *idx = t + i + (subtract_one ? -1 : 0); | |
2347 | ||
2348 | return 1; | |
2349 | } | |
2350 | ||
2351 | static int generic_array_bisect_plus_one( | |
2352 | JournalFile *f, | |
2353 | uint64_t extra, | |
2354 | uint64_t first, | |
2355 | uint64_t n, | |
2356 | uint64_t needle, | |
2357 | int (*test_object)(JournalFile *f, uint64_t p, uint64_t needle), | |
2358 | direction_t direction, | |
2359 | Object **ret, | |
2360 | uint64_t *offset, | |
2361 | uint64_t *idx) { | |
2362 | ||
2363 | int r; | |
2364 | bool step_back = false; | |
2365 | Object *o; | |
2366 | ||
2367 | assert(f); | |
2368 | assert(test_object); | |
2369 | ||
2370 | if (n <= 0) | |
2371 | return 0; | |
2372 | ||
2373 | /* This bisects the array in object 'first', but first checks | |
2374 | * an extra */ | |
2375 | r = test_object(f, extra, needle); | |
2376 | if (r < 0) | |
2377 | return r; | |
2378 | ||
2379 | if (r == TEST_FOUND) | |
2380 | r = direction == DIRECTION_DOWN ? TEST_RIGHT : TEST_LEFT; | |
2381 | ||
2382 | /* if we are looking with DIRECTION_UP then we need to first | |
2383 | see if in the actual array there is a matching entry, and | |
2384 | return the last one of that. But if there isn't any we need | |
2385 | to return this one. Hence remember this, and return it | |
2386 | below. */ | |
2387 | if (r == TEST_LEFT) | |
2388 | step_back = direction == DIRECTION_UP; | |
2389 | ||
2390 | if (r == TEST_RIGHT) { | |
2391 | if (direction == DIRECTION_DOWN) | |
2392 | goto found; | |
2393 | else | |
2394 | return 0; | |
2395 | } | |
2396 | ||
2397 | r = generic_array_bisect(f, first, n-1, needle, test_object, direction, ret, offset, idx); | |
2398 | ||
2399 | if (r == 0 && step_back) | |
2400 | goto found; | |
2401 | ||
2402 | if (r > 0 && idx) | |
2403 | (*idx)++; | |
2404 | ||
2405 | return r; | |
2406 | ||
2407 | found: | |
2408 | r = journal_file_move_to_object(f, OBJECT_ENTRY, extra, &o); | |
2409 | if (r < 0) | |
2410 | return r; | |
2411 | ||
2412 | if (ret) | |
2413 | *ret = o; | |
2414 | ||
2415 | if (offset) | |
2416 | *offset = extra; | |
2417 | ||
2418 | if (idx) | |
2419 | *idx = 0; | |
2420 | ||
2421 | return 1; | |
2422 | } | |
2423 | ||
2424 | _pure_ static int test_object_offset(JournalFile *f, uint64_t p, uint64_t needle) { | |
2425 | assert(f); | |
2426 | assert(p > 0); | |
2427 | ||
2428 | if (p == needle) | |
2429 | return TEST_FOUND; | |
2430 | else if (p < needle) | |
2431 | return TEST_LEFT; | |
2432 | else | |
2433 | return TEST_RIGHT; | |
2434 | } | |
2435 | ||
2436 | static int test_object_seqnum(JournalFile *f, uint64_t p, uint64_t needle) { | |
2437 | Object *o; | |
2438 | int r; | |
2439 | ||
2440 | assert(f); | |
2441 | assert(p > 0); | |
2442 | ||
2443 | r = journal_file_move_to_object(f, OBJECT_ENTRY, p, &o); | |
2444 | if (r < 0) | |
2445 | return r; | |
2446 | ||
2447 | if (le64toh(o->entry.seqnum) == needle) | |
2448 | return TEST_FOUND; | |
2449 | else if (le64toh(o->entry.seqnum) < needle) | |
2450 | return TEST_LEFT; | |
2451 | else | |
2452 | return TEST_RIGHT; | |
2453 | } | |
2454 | ||
2455 | int journal_file_move_to_entry_by_seqnum( | |
2456 | JournalFile *f, | |
2457 | uint64_t seqnum, | |
2458 | direction_t direction, | |
2459 | Object **ret, | |
2460 | uint64_t *offset) { | |
2461 | assert(f); | |
2462 | assert(f->header); | |
2463 | ||
2464 | return generic_array_bisect(f, | |
2465 | le64toh(f->header->entry_array_offset), | |
2466 | le64toh(f->header->n_entries), | |
2467 | seqnum, | |
2468 | test_object_seqnum, | |
2469 | direction, | |
2470 | ret, offset, NULL); | |
2471 | } | |
2472 | ||
2473 | static int test_object_realtime(JournalFile *f, uint64_t p, uint64_t needle) { | |
2474 | Object *o; | |
2475 | int r; | |
2476 | ||
2477 | assert(f); | |
2478 | assert(p > 0); | |
2479 | ||
2480 | r = journal_file_move_to_object(f, OBJECT_ENTRY, p, &o); | |
2481 | if (r < 0) | |
2482 | return r; | |
2483 | ||
2484 | if (le64toh(o->entry.realtime) == needle) | |
2485 | return TEST_FOUND; | |
2486 | else if (le64toh(o->entry.realtime) < needle) | |
2487 | return TEST_LEFT; | |
2488 | else | |
2489 | return TEST_RIGHT; | |
2490 | } | |
2491 | ||
2492 | int journal_file_move_to_entry_by_realtime( | |
2493 | JournalFile *f, | |
2494 | uint64_t realtime, | |
2495 | direction_t direction, | |
2496 | Object **ret, | |
2497 | uint64_t *offset) { | |
2498 | assert(f); | |
2499 | assert(f->header); | |
2500 | ||
2501 | return generic_array_bisect(f, | |
2502 | le64toh(f->header->entry_array_offset), | |
2503 | le64toh(f->header->n_entries), | |
2504 | realtime, | |
2505 | test_object_realtime, | |
2506 | direction, | |
2507 | ret, offset, NULL); | |
2508 | } | |
2509 | ||
2510 | static int test_object_monotonic(JournalFile *f, uint64_t p, uint64_t needle) { | |
2511 | Object *o; | |
2512 | int r; | |
2513 | ||
2514 | assert(f); | |
2515 | assert(p > 0); | |
2516 | ||
2517 | r = journal_file_move_to_object(f, OBJECT_ENTRY, p, &o); | |
2518 | if (r < 0) | |
2519 | return r; | |
2520 | ||
2521 | if (le64toh(o->entry.monotonic) == needle) | |
2522 | return TEST_FOUND; | |
2523 | else if (le64toh(o->entry.monotonic) < needle) | |
2524 | return TEST_LEFT; | |
2525 | else | |
2526 | return TEST_RIGHT; | |
2527 | } | |
2528 | ||
2529 | static int find_data_object_by_boot_id( | |
2530 | JournalFile *f, | |
2531 | sd_id128_t boot_id, | |
2532 | Object **o, | |
2533 | uint64_t *b) { | |
2534 | ||
2535 | char t[STRLEN("_BOOT_ID=") + 32 + 1] = "_BOOT_ID="; | |
2536 | ||
2537 | sd_id128_to_string(boot_id, t + 9); | |
2538 | return journal_file_find_data_object(f, t, sizeof(t) - 1, o, b); | |
2539 | } | |
2540 | ||
2541 | int journal_file_move_to_entry_by_monotonic( | |
2542 | JournalFile *f, | |
2543 | sd_id128_t boot_id, | |
2544 | uint64_t monotonic, | |
2545 | direction_t direction, | |
2546 | Object **ret, | |
2547 | uint64_t *offset) { | |
2548 | ||
2549 | Object *o; | |
2550 | int r; | |
2551 | ||
2552 | assert(f); | |
2553 | ||
2554 | r = find_data_object_by_boot_id(f, boot_id, &o, NULL); | |
2555 | if (r < 0) | |
2556 | return r; | |
2557 | if (r == 0) | |
2558 | return -ENOENT; | |
2559 | ||
2560 | return generic_array_bisect_plus_one(f, | |
2561 | le64toh(o->data.entry_offset), | |
2562 | le64toh(o->data.entry_array_offset), | |
2563 | le64toh(o->data.n_entries), | |
2564 | monotonic, | |
2565 | test_object_monotonic, | |
2566 | direction, | |
2567 | ret, offset, NULL); | |
2568 | } | |
2569 | ||
2570 | void journal_file_reset_location(JournalFile *f) { | |
2571 | f->location_type = LOCATION_HEAD; | |
2572 | f->current_offset = 0; | |
2573 | f->current_seqnum = 0; | |
2574 | f->current_realtime = 0; | |
2575 | f->current_monotonic = 0; | |
2576 | zero(f->current_boot_id); | |
2577 | f->current_xor_hash = 0; | |
2578 | } | |
2579 | ||
2580 | void journal_file_save_location(JournalFile *f, Object *o, uint64_t offset) { | |
2581 | f->location_type = LOCATION_SEEK; | |
2582 | f->current_offset = offset; | |
2583 | f->current_seqnum = le64toh(o->entry.seqnum); | |
2584 | f->current_realtime = le64toh(o->entry.realtime); | |
2585 | f->current_monotonic = le64toh(o->entry.monotonic); | |
2586 | f->current_boot_id = o->entry.boot_id; | |
2587 | f->current_xor_hash = le64toh(o->entry.xor_hash); | |
2588 | } | |
2589 | ||
2590 | int journal_file_compare_locations(JournalFile *af, JournalFile *bf) { | |
2591 | int r; | |
2592 | ||
2593 | assert(af); | |
2594 | assert(af->header); | |
2595 | assert(bf); | |
2596 | assert(bf->header); | |
2597 | assert(af->location_type == LOCATION_SEEK); | |
2598 | assert(bf->location_type == LOCATION_SEEK); | |
2599 | ||
2600 | /* If contents and timestamps match, these entries are | |
2601 | * identical, even if the seqnum does not match */ | |
2602 | if (sd_id128_equal(af->current_boot_id, bf->current_boot_id) && | |
2603 | af->current_monotonic == bf->current_monotonic && | |
2604 | af->current_realtime == bf->current_realtime && | |
2605 | af->current_xor_hash == bf->current_xor_hash) | |
2606 | return 0; | |
2607 | ||
2608 | if (sd_id128_equal(af->header->seqnum_id, bf->header->seqnum_id)) { | |
2609 | ||
2610 | /* If this is from the same seqnum source, compare | |
2611 | * seqnums */ | |
2612 | r = CMP(af->current_seqnum, bf->current_seqnum); | |
2613 | if (r != 0) | |
2614 | return r; | |
2615 | ||
2616 | /* Wow! This is weird, different data but the same | |
2617 | * seqnums? Something is borked, but let's make the | |
2618 | * best of it and compare by time. */ | |
2619 | } | |
2620 | ||
2621 | if (sd_id128_equal(af->current_boot_id, bf->current_boot_id)) { | |
2622 | ||
2623 | /* If the boot id matches, compare monotonic time */ | |
2624 | r = CMP(af->current_monotonic, bf->current_monotonic); | |
2625 | if (r != 0) | |
2626 | return r; | |
2627 | } | |
2628 | ||
2629 | /* Otherwise, compare UTC time */ | |
2630 | r = CMP(af->current_realtime, bf->current_realtime); | |
2631 | if (r != 0) | |
2632 | return r; | |
2633 | ||
2634 | /* Finally, compare by contents */ | |
2635 | return CMP(af->current_xor_hash, bf->current_xor_hash); | |
2636 | } | |
2637 | ||
2638 | static int bump_array_index(uint64_t *i, direction_t direction, uint64_t n) { | |
2639 | ||
2640 | /* Increase or decrease the specified index, in the right direction. */ | |
2641 | ||
2642 | if (direction == DIRECTION_DOWN) { | |
2643 | if (*i >= n - 1) | |
2644 | return 0; | |
2645 | ||
2646 | (*i) ++; | |
2647 | } else { | |
2648 | if (*i <= 0) | |
2649 | return 0; | |
2650 | ||
2651 | (*i) --; | |
2652 | } | |
2653 | ||
2654 | return 1; | |
2655 | } | |
2656 | ||
2657 | static bool check_properly_ordered(uint64_t new_offset, uint64_t old_offset, direction_t direction) { | |
2658 | ||
2659 | /* Consider it an error if any of the two offsets is uninitialized */ | |
2660 | if (old_offset == 0 || new_offset == 0) | |
2661 | return false; | |
2662 | ||
2663 | /* If we go down, the new offset must be larger than the old one. */ | |
2664 | return direction == DIRECTION_DOWN ? | |
2665 | new_offset > old_offset : | |
2666 | new_offset < old_offset; | |
2667 | } | |
2668 | ||
2669 | int journal_file_next_entry( | |
2670 | JournalFile *f, | |
2671 | uint64_t p, | |
2672 | direction_t direction, | |
2673 | Object **ret, uint64_t *offset) { | |
2674 | ||
2675 | uint64_t i, n, ofs; | |
2676 | int r; | |
2677 | ||
2678 | assert(f); | |
2679 | assert(f->header); | |
2680 | ||
2681 | n = le64toh(f->header->n_entries); | |
2682 | if (n <= 0) | |
2683 | return 0; | |
2684 | ||
2685 | if (p == 0) | |
2686 | i = direction == DIRECTION_DOWN ? 0 : n - 1; | |
2687 | else { | |
2688 | r = generic_array_bisect(f, | |
2689 | le64toh(f->header->entry_array_offset), | |
2690 | le64toh(f->header->n_entries), | |
2691 | p, | |
2692 | test_object_offset, | |
2693 | DIRECTION_DOWN, | |
2694 | NULL, NULL, | |
2695 | &i); | |
2696 | if (r <= 0) | |
2697 | return r; | |
2698 | ||
2699 | r = bump_array_index(&i, direction, n); | |
2700 | if (r <= 0) | |
2701 | return r; | |
2702 | } | |
2703 | ||
2704 | /* And jump to it */ | |
2705 | for (;;) { | |
2706 | r = generic_array_get(f, | |
2707 | le64toh(f->header->entry_array_offset), | |
2708 | i, | |
2709 | ret, &ofs); | |
2710 | if (r > 0) | |
2711 | break; | |
2712 | if (r != -EBADMSG) | |
2713 | return r; | |
2714 | ||
2715 | /* OK, so this entry is borked. Most likely some entry didn't get synced to disk properly, let's see if | |
2716 | * the next one might work for us instead. */ | |
2717 | log_debug_errno(r, "Entry item %" PRIu64 " is bad, skipping over it.", i); | |
2718 | ||
2719 | r = bump_array_index(&i, direction, n); | |
2720 | if (r <= 0) | |
2721 | return r; | |
2722 | } | |
2723 | ||
2724 | /* Ensure our array is properly ordered. */ | |
2725 | if (p > 0 && !check_properly_ordered(ofs, p, direction)) | |
2726 | return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG), | |
2727 | "%s: entry array not properly ordered at entry %" PRIu64, | |
2728 | f->path, i); | |
2729 | ||
2730 | if (offset) | |
2731 | *offset = ofs; | |
2732 | ||
2733 | return 1; | |
2734 | } | |
2735 | ||
2736 | int journal_file_next_entry_for_data( | |
2737 | JournalFile *f, | |
2738 | Object *o, uint64_t p, | |
2739 | uint64_t data_offset, | |
2740 | direction_t direction, | |
2741 | Object **ret, uint64_t *offset) { | |
2742 | ||
2743 | uint64_t i, n, ofs; | |
2744 | Object *d; | |
2745 | int r; | |
2746 | ||
2747 | assert(f); | |
2748 | assert(p > 0 || !o); | |
2749 | ||
2750 | r = journal_file_move_to_object(f, OBJECT_DATA, data_offset, &d); | |
2751 | if (r < 0) | |
2752 | return r; | |
2753 | ||
2754 | n = le64toh(d->data.n_entries); | |
2755 | if (n <= 0) | |
2756 | return n; | |
2757 | ||
2758 | if (!o) | |
2759 | i = direction == DIRECTION_DOWN ? 0 : n - 1; | |
2760 | else { | |
2761 | if (o->object.type != OBJECT_ENTRY) | |
2762 | return -EINVAL; | |
2763 | ||
2764 | r = generic_array_bisect_plus_one(f, | |
2765 | le64toh(d->data.entry_offset), | |
2766 | le64toh(d->data.entry_array_offset), | |
2767 | le64toh(d->data.n_entries), | |
2768 | p, | |
2769 | test_object_offset, | |
2770 | DIRECTION_DOWN, | |
2771 | NULL, NULL, | |
2772 | &i); | |
2773 | ||
2774 | if (r <= 0) | |
2775 | return r; | |
2776 | ||
2777 | r = bump_array_index(&i, direction, n); | |
2778 | if (r <= 0) | |
2779 | return r; | |
2780 | } | |
2781 | ||
2782 | for (;;) { | |
2783 | r = generic_array_get_plus_one(f, | |
2784 | le64toh(d->data.entry_offset), | |
2785 | le64toh(d->data.entry_array_offset), | |
2786 | i, | |
2787 | ret, &ofs); | |
2788 | if (r > 0) | |
2789 | break; | |
2790 | if (r != -EBADMSG) | |
2791 | return r; | |
2792 | ||
2793 | log_debug_errno(r, "Data entry item %" PRIu64 " is bad, skipping over it.", i); | |
2794 | ||
2795 | r = bump_array_index(&i, direction, n); | |
2796 | if (r <= 0) | |
2797 | return r; | |
2798 | } | |
2799 | ||
2800 | /* Ensure our array is properly ordered. */ | |
2801 | if (p > 0 && check_properly_ordered(ofs, p, direction)) | |
2802 | return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG), | |
2803 | "%s data entry array not properly ordered at entry %" PRIu64, | |
2804 | f->path, i); | |
2805 | ||
2806 | if (offset) | |
2807 | *offset = ofs; | |
2808 | ||
2809 | return 1; | |
2810 | } | |
2811 | ||
2812 | int journal_file_move_to_entry_by_offset_for_data( | |
2813 | JournalFile *f, | |
2814 | uint64_t data_offset, | |
2815 | uint64_t p, | |
2816 | direction_t direction, | |
2817 | Object **ret, uint64_t *offset) { | |
2818 | ||
2819 | int r; | |
2820 | Object *d; | |
2821 | ||
2822 | assert(f); | |
2823 | ||
2824 | r = journal_file_move_to_object(f, OBJECT_DATA, data_offset, &d); | |
2825 | if (r < 0) | |
2826 | return r; | |
2827 | ||
2828 | return generic_array_bisect_plus_one(f, | |
2829 | le64toh(d->data.entry_offset), | |
2830 | le64toh(d->data.entry_array_offset), | |
2831 | le64toh(d->data.n_entries), | |
2832 | p, | |
2833 | test_object_offset, | |
2834 | direction, | |
2835 | ret, offset, NULL); | |
2836 | } | |
2837 | ||
2838 | int journal_file_move_to_entry_by_monotonic_for_data( | |
2839 | JournalFile *f, | |
2840 | uint64_t data_offset, | |
2841 | sd_id128_t boot_id, | |
2842 | uint64_t monotonic, | |
2843 | direction_t direction, | |
2844 | Object **ret, uint64_t *offset) { | |
2845 | ||
2846 | Object *o, *d; | |
2847 | int r; | |
2848 | uint64_t b, z; | |
2849 | ||
2850 | assert(f); | |
2851 | ||
2852 | /* First, seek by time */ | |
2853 | r = find_data_object_by_boot_id(f, boot_id, &o, &b); | |
2854 | if (r < 0) | |
2855 | return r; | |
2856 | if (r == 0) | |
2857 | return -ENOENT; | |
2858 | ||
2859 | r = generic_array_bisect_plus_one(f, | |
2860 | le64toh(o->data.entry_offset), | |
2861 | le64toh(o->data.entry_array_offset), | |
2862 | le64toh(o->data.n_entries), | |
2863 | monotonic, | |
2864 | test_object_monotonic, | |
2865 | direction, | |
2866 | NULL, &z, NULL); | |
2867 | if (r <= 0) | |
2868 | return r; | |
2869 | ||
2870 | /* And now, continue seeking until we find an entry that | |
2871 | * exists in both bisection arrays */ | |
2872 | ||
2873 | for (;;) { | |
2874 | Object *qo; | |
2875 | uint64_t p, q; | |
2876 | ||
2877 | r = journal_file_move_to_object(f, OBJECT_DATA, data_offset, &d); | |
2878 | if (r < 0) | |
2879 | return r; | |
2880 | ||
2881 | r = generic_array_bisect_plus_one(f, | |
2882 | le64toh(d->data.entry_offset), | |
2883 | le64toh(d->data.entry_array_offset), | |
2884 | le64toh(d->data.n_entries), | |
2885 | z, | |
2886 | test_object_offset, | |
2887 | direction, | |
2888 | NULL, &p, NULL); | |
2889 | if (r <= 0) | |
2890 | return r; | |
2891 | ||
2892 | r = journal_file_move_to_object(f, OBJECT_DATA, b, &o); | |
2893 | if (r < 0) | |
2894 | return r; | |
2895 | ||
2896 | r = generic_array_bisect_plus_one(f, | |
2897 | le64toh(o->data.entry_offset), | |
2898 | le64toh(o->data.entry_array_offset), | |
2899 | le64toh(o->data.n_entries), | |
2900 | p, | |
2901 | test_object_offset, | |
2902 | direction, | |
2903 | &qo, &q, NULL); | |
2904 | ||
2905 | if (r <= 0) | |
2906 | return r; | |
2907 | ||
2908 | if (p == q) { | |
2909 | if (ret) | |
2910 | *ret = qo; | |
2911 | if (offset) | |
2912 | *offset = q; | |
2913 | ||
2914 | return 1; | |
2915 | } | |
2916 | ||
2917 | z = q; | |
2918 | } | |
2919 | } | |
2920 | ||
2921 | int journal_file_move_to_entry_by_seqnum_for_data( | |
2922 | JournalFile *f, | |
2923 | uint64_t data_offset, | |
2924 | uint64_t seqnum, | |
2925 | direction_t direction, | |
2926 | Object **ret, uint64_t *offset) { | |
2927 | ||
2928 | Object *d; | |
2929 | int r; | |
2930 | ||
2931 | assert(f); | |
2932 | ||
2933 | r = journal_file_move_to_object(f, OBJECT_DATA, data_offset, &d); | |
2934 | if (r < 0) | |
2935 | return r; | |
2936 | ||
2937 | return generic_array_bisect_plus_one(f, | |
2938 | le64toh(d->data.entry_offset), | |
2939 | le64toh(d->data.entry_array_offset), | |
2940 | le64toh(d->data.n_entries), | |
2941 | seqnum, | |
2942 | test_object_seqnum, | |
2943 | direction, | |
2944 | ret, offset, NULL); | |
2945 | } | |
2946 | ||
2947 | int journal_file_move_to_entry_by_realtime_for_data( | |
2948 | JournalFile *f, | |
2949 | uint64_t data_offset, | |
2950 | uint64_t realtime, | |
2951 | direction_t direction, | |
2952 | Object **ret, uint64_t *offset) { | |
2953 | ||
2954 | Object *d; | |
2955 | int r; | |
2956 | ||
2957 | assert(f); | |
2958 | ||
2959 | r = journal_file_move_to_object(f, OBJECT_DATA, data_offset, &d); | |
2960 | if (r < 0) | |
2961 | return r; | |
2962 | ||
2963 | return generic_array_bisect_plus_one(f, | |
2964 | le64toh(d->data.entry_offset), | |
2965 | le64toh(d->data.entry_array_offset), | |
2966 | le64toh(d->data.n_entries), | |
2967 | realtime, | |
2968 | test_object_realtime, | |
2969 | direction, | |
2970 | ret, offset, NULL); | |
2971 | } | |
2972 | ||
2973 | void journal_file_dump(JournalFile *f) { | |
2974 | Object *o; | |
2975 | int r; | |
2976 | uint64_t p; | |
2977 | ||
2978 | assert(f); | |
2979 | assert(f->header); | |
2980 | ||
2981 | journal_file_print_header(f); | |
2982 | ||
2983 | p = le64toh(f->header->header_size); | |
2984 | while (p != 0) { | |
2985 | r = journal_file_move_to_object(f, OBJECT_UNUSED, p, &o); | |
2986 | if (r < 0) | |
2987 | goto fail; | |
2988 | ||
2989 | switch (o->object.type) { | |
2990 | ||
2991 | case OBJECT_UNUSED: | |
2992 | printf("Type: OBJECT_UNUSED\n"); | |
2993 | break; | |
2994 | ||
2995 | case OBJECT_DATA: | |
2996 | printf("Type: OBJECT_DATA\n"); | |
2997 | break; | |
2998 | ||
2999 | case OBJECT_FIELD: | |
3000 | printf("Type: OBJECT_FIELD\n"); | |
3001 | break; | |
3002 | ||
3003 | case OBJECT_ENTRY: | |
3004 | printf("Type: OBJECT_ENTRY seqnum=%"PRIu64" monotonic=%"PRIu64" realtime=%"PRIu64"\n", | |
3005 | le64toh(o->entry.seqnum), | |
3006 | le64toh(o->entry.monotonic), | |
3007 | le64toh(o->entry.realtime)); | |
3008 | break; | |
3009 | ||
3010 | case OBJECT_FIELD_HASH_TABLE: | |
3011 | printf("Type: OBJECT_FIELD_HASH_TABLE\n"); | |
3012 | break; | |
3013 | ||
3014 | case OBJECT_DATA_HASH_TABLE: | |
3015 | printf("Type: OBJECT_DATA_HASH_TABLE\n"); | |
3016 | break; | |
3017 | ||
3018 | case OBJECT_ENTRY_ARRAY: | |
3019 | printf("Type: OBJECT_ENTRY_ARRAY\n"); | |
3020 | break; | |
3021 | ||
3022 | case OBJECT_TAG: | |
3023 | printf("Type: OBJECT_TAG seqnum=%"PRIu64" epoch=%"PRIu64"\n", | |
3024 | le64toh(o->tag.seqnum), | |
3025 | le64toh(o->tag.epoch)); | |
3026 | break; | |
3027 | ||
3028 | default: | |
3029 | printf("Type: unknown (%i)\n", o->object.type); | |
3030 | break; | |
3031 | } | |
3032 | ||
3033 | if (o->object.flags & OBJECT_COMPRESSION_MASK) | |
3034 | printf("Flags: %s\n", | |
3035 | object_compressed_to_string(o->object.flags & OBJECT_COMPRESSION_MASK)); | |
3036 | ||
3037 | if (p == le64toh(f->header->tail_object_offset)) | |
3038 | p = 0; | |
3039 | else | |
3040 | p = p + ALIGN64(le64toh(o->object.size)); | |
3041 | } | |
3042 | ||
3043 | return; | |
3044 | fail: | |
3045 | log_error("File corrupt"); | |
3046 | } | |
3047 | ||
3048 | static const char* format_timestamp_safe(char *buf, size_t l, usec_t t) { | |
3049 | const char *x; | |
3050 | ||
3051 | x = format_timestamp(buf, l, t); | |
3052 | if (x) | |
3053 | return x; | |
3054 | return " --- "; | |
3055 | } | |
3056 | ||
3057 | void journal_file_print_header(JournalFile *f) { | |
3058 | char a[33], b[33], c[33], d[33]; | |
3059 | char x[FORMAT_TIMESTAMP_MAX], y[FORMAT_TIMESTAMP_MAX], z[FORMAT_TIMESTAMP_MAX]; | |
3060 | struct stat st; | |
3061 | char bytes[FORMAT_BYTES_MAX]; | |
3062 | ||
3063 | assert(f); | |
3064 | assert(f->header); | |
3065 | ||
3066 | printf("File Path: %s\n" | |
3067 | "File ID: %s\n" | |
3068 | "Machine ID: %s\n" | |
3069 | "Boot ID: %s\n" | |
3070 | "Sequential Number ID: %s\n" | |
3071 | "State: %s\n" | |
3072 | "Compatible Flags:%s%s\n" | |
3073 | "Incompatible Flags:%s%s%s\n" | |
3074 | "Header size: %"PRIu64"\n" | |
3075 | "Arena size: %"PRIu64"\n" | |
3076 | "Data Hash Table Size: %"PRIu64"\n" | |
3077 | "Field Hash Table Size: %"PRIu64"\n" | |
3078 | "Rotate Suggested: %s\n" | |
3079 | "Head Sequential Number: %"PRIu64" (%"PRIx64")\n" | |
3080 | "Tail Sequential Number: %"PRIu64" (%"PRIx64")\n" | |
3081 | "Head Realtime Timestamp: %s (%"PRIx64")\n" | |
3082 | "Tail Realtime Timestamp: %s (%"PRIx64")\n" | |
3083 | "Tail Monotonic Timestamp: %s (%"PRIx64")\n" | |
3084 | "Objects: %"PRIu64"\n" | |
3085 | "Entry Objects: %"PRIu64"\n", | |
3086 | f->path, | |
3087 | sd_id128_to_string(f->header->file_id, a), | |
3088 | sd_id128_to_string(f->header->machine_id, b), | |
3089 | sd_id128_to_string(f->header->boot_id, c), | |
3090 | sd_id128_to_string(f->header->seqnum_id, d), | |
3091 | f->header->state == STATE_OFFLINE ? "OFFLINE" : | |
3092 | f->header->state == STATE_ONLINE ? "ONLINE" : | |
3093 | f->header->state == STATE_ARCHIVED ? "ARCHIVED" : "UNKNOWN", | |
3094 | JOURNAL_HEADER_SEALED(f->header) ? " SEALED" : "", | |
3095 | (le32toh(f->header->compatible_flags) & ~HEADER_COMPATIBLE_ANY) ? " ???" : "", | |
3096 | JOURNAL_HEADER_COMPRESSED_XZ(f->header) ? " COMPRESSED-XZ" : "", | |
3097 | JOURNAL_HEADER_COMPRESSED_LZ4(f->header) ? " COMPRESSED-LZ4" : "", | |
3098 | (le32toh(f->header->incompatible_flags) & ~HEADER_INCOMPATIBLE_ANY) ? " ???" : "", | |
3099 | le64toh(f->header->header_size), | |
3100 | le64toh(f->header->arena_size), | |
3101 | le64toh(f->header->data_hash_table_size) / sizeof(HashItem), | |
3102 | le64toh(f->header->field_hash_table_size) / sizeof(HashItem), | |
3103 | yes_no(journal_file_rotate_suggested(f, 0)), | |
3104 | le64toh(f->header->head_entry_seqnum), le64toh(f->header->head_entry_seqnum), | |
3105 | le64toh(f->header->tail_entry_seqnum), le64toh(f->header->tail_entry_seqnum), | |
3106 | format_timestamp_safe(x, sizeof(x), le64toh(f->header->head_entry_realtime)), le64toh(f->header->head_entry_realtime), | |
3107 | format_timestamp_safe(y, sizeof(y), le64toh(f->header->tail_entry_realtime)), le64toh(f->header->tail_entry_realtime), | |
3108 | format_timespan(z, sizeof(z), le64toh(f->header->tail_entry_monotonic), USEC_PER_MSEC), le64toh(f->header->tail_entry_monotonic), | |
3109 | le64toh(f->header->n_objects), | |
3110 | le64toh(f->header->n_entries)); | |
3111 | ||
3112 | if (JOURNAL_HEADER_CONTAINS(f->header, n_data)) | |
3113 | printf("Data Objects: %"PRIu64"\n" | |
3114 | "Data Hash Table Fill: %.1f%%\n", | |
3115 | le64toh(f->header->n_data), | |
3116 | 100.0 * (double) le64toh(f->header->n_data) / ((double) (le64toh(f->header->data_hash_table_size) / sizeof(HashItem)))); | |
3117 | ||
3118 | if (JOURNAL_HEADER_CONTAINS(f->header, n_fields)) | |
3119 | printf("Field Objects: %"PRIu64"\n" | |
3120 | "Field Hash Table Fill: %.1f%%\n", | |
3121 | le64toh(f->header->n_fields), | |
3122 | 100.0 * (double) le64toh(f->header->n_fields) / ((double) (le64toh(f->header->field_hash_table_size) / sizeof(HashItem)))); | |
3123 | ||
3124 | if (JOURNAL_HEADER_CONTAINS(f->header, n_tags)) | |
3125 | printf("Tag Objects: %"PRIu64"\n", | |
3126 | le64toh(f->header->n_tags)); | |
3127 | if (JOURNAL_HEADER_CONTAINS(f->header, n_entry_arrays)) | |
3128 | printf("Entry Array Objects: %"PRIu64"\n", | |
3129 | le64toh(f->header->n_entry_arrays)); | |
3130 | ||
3131 | if (fstat(f->fd, &st) >= 0) | |
3132 | printf("Disk usage: %s\n", format_bytes(bytes, sizeof(bytes), (uint64_t) st.st_blocks * 512ULL)); | |
3133 | } | |
3134 | ||
3135 | static int journal_file_warn_btrfs(JournalFile *f) { | |
3136 | unsigned attrs; | |
3137 | int r; | |
3138 | ||
3139 | assert(f); | |
3140 | ||
3141 | /* Before we write anything, check if the COW logic is turned | |
3142 | * off on btrfs. Given our write pattern that is quite | |
3143 | * unfriendly to COW file systems this should greatly improve | |
3144 | * performance on COW file systems, such as btrfs, at the | |
3145 | * expense of data integrity features (which shouldn't be too | |
3146 | * bad, given that we do our own checksumming). */ | |
3147 | ||
3148 | r = btrfs_is_filesystem(f->fd); | |
3149 | if (r < 0) | |
3150 | return log_warning_errno(r, "Failed to determine if journal is on btrfs: %m"); | |
3151 | if (!r) | |
3152 | return 0; | |
3153 | ||
3154 | r = read_attr_fd(f->fd, &attrs); | |
3155 | if (r < 0) | |
3156 | return log_warning_errno(r, "Failed to read file attributes: %m"); | |
3157 | ||
3158 | if (attrs & FS_NOCOW_FL) { | |
3159 | log_debug("Detected btrfs file system with copy-on-write disabled, all is good."); | |
3160 | return 0; | |
3161 | } | |
3162 | ||
3163 | log_notice("Creating journal file %s on a btrfs file system, and copy-on-write is enabled. " | |
3164 | "This is likely to slow down journal access substantially, please consider turning " | |
3165 | "off the copy-on-write file attribute on the journal directory, using chattr +C.", f->path); | |
3166 | ||
3167 | return 1; | |
3168 | } | |
3169 | ||
3170 | int journal_file_open( | |
3171 | int fd, | |
3172 | const char *fname, | |
3173 | int flags, | |
3174 | mode_t mode, | |
3175 | bool compress, | |
3176 | uint64_t compress_threshold_bytes, | |
3177 | bool seal, | |
3178 | JournalMetrics *metrics, | |
3179 | MMapCache *mmap_cache, | |
3180 | Set *deferred_closes, | |
3181 | JournalFile *template, | |
3182 | JournalFile **ret) { | |
3183 | ||
3184 | bool newly_created = false; | |
3185 | JournalFile *f; | |
3186 | void *h; | |
3187 | int r; | |
3188 | ||
3189 | assert(ret); | |
3190 | assert(fd >= 0 || fname); | |
3191 | ||
3192 | if (!IN_SET((flags & O_ACCMODE), O_RDONLY, O_RDWR)) | |
3193 | return -EINVAL; | |
3194 | ||
3195 | if (fname && (flags & O_CREAT) && !endswith(fname, ".journal")) | |
3196 | return -EINVAL; | |
3197 | ||
3198 | f = new(JournalFile, 1); | |
3199 | if (!f) | |
3200 | return -ENOMEM; | |
3201 | ||
3202 | *f = (JournalFile) { | |
3203 | .fd = fd, | |
3204 | .mode = mode, | |
3205 | ||
3206 | .flags = flags, | |
3207 | .prot = prot_from_flags(flags), | |
3208 | .writable = (flags & O_ACCMODE) != O_RDONLY, | |
3209 | ||
3210 | #if HAVE_LZ4 | |
3211 | .compress_lz4 = compress, | |
3212 | #elif HAVE_XZ | |
3213 | .compress_xz = compress, | |
3214 | #endif | |
3215 | .compress_threshold_bytes = compress_threshold_bytes == (uint64_t) -1 ? | |
3216 | DEFAULT_COMPRESS_THRESHOLD : | |
3217 | MAX(MIN_COMPRESS_THRESHOLD, compress_threshold_bytes), | |
3218 | #if HAVE_GCRYPT | |
3219 | .seal = seal, | |
3220 | #endif | |
3221 | }; | |
3222 | ||
3223 | if (DEBUG_LOGGING) { | |
3224 | static int last_seal = -1, last_compress = -1; | |
3225 | static uint64_t last_bytes = UINT64_MAX; | |
3226 | char bytes[FORMAT_BYTES_MAX]; | |
3227 | ||
3228 | if (last_seal != f->seal || | |
3229 | last_compress != JOURNAL_FILE_COMPRESS(f) || | |
3230 | last_bytes != f->compress_threshold_bytes) { | |
3231 | ||
3232 | log_debug("Journal effective settings seal=%s compress=%s compress_threshold_bytes=%s", | |
3233 | yes_no(f->seal), yes_no(JOURNAL_FILE_COMPRESS(f)), | |
3234 | format_bytes(bytes, sizeof bytes, f->compress_threshold_bytes)); | |
3235 | last_seal = f->seal; | |
3236 | last_compress = JOURNAL_FILE_COMPRESS(f); | |
3237 | last_bytes = f->compress_threshold_bytes; | |
3238 | } | |
3239 | } | |
3240 | ||
3241 | if (mmap_cache) | |
3242 | f->mmap = mmap_cache_ref(mmap_cache); | |
3243 | else { | |
3244 | f->mmap = mmap_cache_new(); | |
3245 | if (!f->mmap) { | |
3246 | r = -ENOMEM; | |
3247 | goto fail; | |
3248 | } | |
3249 | } | |
3250 | ||
3251 | if (fname) { | |
3252 | f->path = strdup(fname); | |
3253 | if (!f->path) { | |
3254 | r = -ENOMEM; | |
3255 | goto fail; | |
3256 | } | |
3257 | } else { | |
3258 | assert(fd >= 0); | |
3259 | ||
3260 | /* If we don't know the path, fill in something explanatory and vaguely useful */ | |
3261 | if (asprintf(&f->path, "/proc/self/%i", fd) < 0) { | |
3262 | r = -ENOMEM; | |
3263 | goto fail; | |
3264 | } | |
3265 | } | |
3266 | ||
3267 | f->chain_cache = ordered_hashmap_new(&uint64_hash_ops); | |
3268 | if (!f->chain_cache) { | |
3269 | r = -ENOMEM; | |
3270 | goto fail; | |
3271 | } | |
3272 | ||
3273 | if (f->fd < 0) { | |
3274 | /* We pass O_NONBLOCK here, so that in case somebody pointed us to some character device node or FIFO | |
3275 | * or so, we likely fail quickly than block for long. For regular files O_NONBLOCK has no effect, hence | |
3276 | * it doesn't hurt in that case. */ | |
3277 | ||
3278 | f->fd = open(f->path, f->flags|O_CLOEXEC|O_NONBLOCK, f->mode); | |
3279 | if (f->fd < 0) { | |
3280 | r = -errno; | |
3281 | goto fail; | |
3282 | } | |
3283 | ||
3284 | /* fds we opened here by us should also be closed by us. */ | |
3285 | f->close_fd = true; | |
3286 | ||
3287 | r = fd_nonblock(f->fd, false); | |
3288 | if (r < 0) | |
3289 | goto fail; | |
3290 | } | |
3291 | ||
3292 | f->cache_fd = mmap_cache_add_fd(f->mmap, f->fd); | |
3293 | if (!f->cache_fd) { | |
3294 | r = -ENOMEM; | |
3295 | goto fail; | |
3296 | } | |
3297 | ||
3298 | r = journal_file_fstat(f); | |
3299 | if (r < 0) | |
3300 | goto fail; | |
3301 | ||
3302 | if (f->last_stat.st_size == 0 && f->writable) { | |
3303 | ||
3304 | (void) journal_file_warn_btrfs(f); | |
3305 | ||
3306 | /* Let's attach the creation time to the journal file, so that the vacuuming code knows the age of this | |
3307 | * file even if the file might end up corrupted one day... Ideally we'd just use the creation time many | |
3308 | * file systems maintain for each file, but the API to query this is very new, hence let's emulate this | |
3309 | * via extended attributes. If extended attributes are not supported we'll just skip this, and rely | |
3310 | * solely on mtime/atime/ctime of the file. */ | |
3311 | (void) fd_setcrtime(f->fd, 0); | |
3312 | ||
3313 | #if HAVE_GCRYPT | |
3314 | /* Try to load the FSPRG state, and if we can't, then | |
3315 | * just don't do sealing */ | |
3316 | if (f->seal) { | |
3317 | r = journal_file_fss_load(f); | |
3318 | if (r < 0) | |
3319 | f->seal = false; | |
3320 | } | |
3321 | #endif | |
3322 | ||
3323 | r = journal_file_init_header(f, template); | |
3324 | if (r < 0) | |
3325 | goto fail; | |
3326 | ||
3327 | r = journal_file_fstat(f); | |
3328 | if (r < 0) | |
3329 | goto fail; | |
3330 | ||
3331 | newly_created = true; | |
3332 | } | |
3333 | ||
3334 | if (f->last_stat.st_size < (off_t) HEADER_SIZE_MIN) { | |
3335 | r = -ENODATA; | |
3336 | goto fail; | |
3337 | } | |
3338 | ||
3339 | r = mmap_cache_get(f->mmap, f->cache_fd, f->prot, CONTEXT_HEADER, true, 0, PAGE_ALIGN(sizeof(Header)), &f->last_stat, &h, NULL); | |
3340 | if (r == -EINVAL) { | |
3341 | /* Some file systems (jffs2 or p9fs) don't support mmap() properly (or only read-only | |
3342 | * mmap()), and return EINVAL in that case. Let's propagate that as a more recognizable error | |
3343 | * code. */ | |
3344 | r = -EAFNOSUPPORT; | |
3345 | goto fail; | |
3346 | } | |
3347 | if (r < 0) | |
3348 | goto fail; | |
3349 | ||
3350 | f->header = h; | |
3351 | ||
3352 | if (!newly_created) { | |
3353 | set_clear_with_destructor(deferred_closes, journal_file_close); | |
3354 | ||
3355 | r = journal_file_verify_header(f); | |
3356 | if (r < 0) | |
3357 | goto fail; | |
3358 | } | |
3359 | ||
3360 | #if HAVE_GCRYPT | |
3361 | if (!newly_created && f->writable) { | |
3362 | r = journal_file_fss_load(f); | |
3363 | if (r < 0) | |
3364 | goto fail; | |
3365 | } | |
3366 | #endif | |
3367 | ||
3368 | if (f->writable) { | |
3369 | if (metrics) { | |
3370 | journal_default_metrics(metrics, f->fd); | |
3371 | f->metrics = *metrics; | |
3372 | } else if (template) | |
3373 | f->metrics = template->metrics; | |
3374 | ||
3375 | r = journal_file_refresh_header(f); | |
3376 | if (r < 0) | |
3377 | goto fail; | |
3378 | } | |
3379 | ||
3380 | #if HAVE_GCRYPT | |
3381 | r = journal_file_hmac_setup(f); | |
3382 | if (r < 0) | |
3383 | goto fail; | |
3384 | #endif | |
3385 | ||
3386 | if (newly_created) { | |
3387 | r = journal_file_setup_field_hash_table(f); | |
3388 | if (r < 0) | |
3389 | goto fail; | |
3390 | ||
3391 | r = journal_file_setup_data_hash_table(f); | |
3392 | if (r < 0) | |
3393 | goto fail; | |
3394 | ||
3395 | #if HAVE_GCRYPT | |
3396 | r = journal_file_append_first_tag(f); | |
3397 | if (r < 0) | |
3398 | goto fail; | |
3399 | #endif | |
3400 | } | |
3401 | ||
3402 | if (mmap_cache_got_sigbus(f->mmap, f->cache_fd)) { | |
3403 | r = -EIO; | |
3404 | goto fail; | |
3405 | } | |
3406 | ||
3407 | if (template && template->post_change_timer) { | |
3408 | r = journal_file_enable_post_change_timer( | |
3409 | f, | |
3410 | sd_event_source_get_event(template->post_change_timer), | |
3411 | template->post_change_timer_period); | |
3412 | ||
3413 | if (r < 0) | |
3414 | goto fail; | |
3415 | } | |
3416 | ||
3417 | /* The file is opened now successfully, thus we take possession of any passed in fd. */ | |
3418 | f->close_fd = true; | |
3419 | ||
3420 | *ret = f; | |
3421 | return 0; | |
3422 | ||
3423 | fail: | |
3424 | if (f->cache_fd && mmap_cache_got_sigbus(f->mmap, f->cache_fd)) | |
3425 | r = -EIO; | |
3426 | ||
3427 | (void) journal_file_close(f); | |
3428 | ||
3429 | return r; | |
3430 | } | |
3431 | ||
3432 | int journal_file_archive(JournalFile *f) { | |
3433 | _cleanup_free_ char *p = NULL; | |
3434 | ||
3435 | assert(f); | |
3436 | ||
3437 | if (!f->writable) | |
3438 | return -EINVAL; | |
3439 | ||
3440 | /* Is this a journal file that was passed to us as fd? If so, we synthesized a path name for it, and we refuse | |
3441 | * rotation, since we don't know the actual path, and couldn't rename the file hence. */ | |
3442 | if (path_startswith(f->path, "/proc/self/fd")) | |
3443 | return -EINVAL; | |
3444 | ||
3445 | if (!endswith(f->path, ".journal")) | |
3446 | return -EINVAL; | |
3447 | ||
3448 | if (asprintf(&p, "%.*s@" SD_ID128_FORMAT_STR "-%016"PRIx64"-%016"PRIx64".journal", | |
3449 | (int) strlen(f->path) - 8, f->path, | |
3450 | SD_ID128_FORMAT_VAL(f->header->seqnum_id), | |
3451 | le64toh(f->header->head_entry_seqnum), | |
3452 | le64toh(f->header->head_entry_realtime)) < 0) | |
3453 | return -ENOMEM; | |
3454 | ||
3455 | /* Try to rename the file to the archived version. If the file already was deleted, we'll get ENOENT, let's | |
3456 | * ignore that case. */ | |
3457 | if (rename(f->path, p) < 0 && errno != ENOENT) | |
3458 | return -errno; | |
3459 | ||
3460 | /* Sync the rename to disk */ | |
3461 | (void) fsync_directory_of_file(f->fd); | |
3462 | ||
3463 | /* Set as archive so offlining commits w/state=STATE_ARCHIVED. Previously we would set old_file->header->state | |
3464 | * to STATE_ARCHIVED directly here, but journal_file_set_offline() short-circuits when state != STATE_ONLINE, | |
3465 | * which would result in the rotated journal never getting fsync() called before closing. Now we simply queue | |
3466 | * the archive state by setting an archive bit, leaving the state as STATE_ONLINE so proper offlining | |
3467 | * occurs. */ | |
3468 | f->archive = true; | |
3469 | ||
3470 | /* Currently, btrfs is not very good with out write patterns and fragments heavily. Let's defrag our journal | |
3471 | * files when we archive them */ | |
3472 | f->defrag_on_close = true; | |
3473 | ||
3474 | return 0; | |
3475 | } | |
3476 | ||
3477 | JournalFile* journal_initiate_close( | |
3478 | JournalFile *f, | |
3479 | Set *deferred_closes) { | |
3480 | ||
3481 | int r; | |
3482 | ||
3483 | assert(f); | |
3484 | ||
3485 | if (deferred_closes) { | |
3486 | ||
3487 | r = set_put(deferred_closes, f); | |
3488 | if (r < 0) | |
3489 | log_debug_errno(r, "Failed to add file to deferred close set, closing immediately."); | |
3490 | else { | |
3491 | (void) journal_file_set_offline(f, false); | |
3492 | return NULL; | |
3493 | } | |
3494 | } | |
3495 | ||
3496 | return journal_file_close(f); | |
3497 | } | |
3498 | ||
3499 | int journal_file_rotate( | |
3500 | JournalFile **f, | |
3501 | bool compress, | |
3502 | uint64_t compress_threshold_bytes, | |
3503 | bool seal, | |
3504 | Set *deferred_closes) { | |
3505 | ||
3506 | JournalFile *new_file = NULL; | |
3507 | int r; | |
3508 | ||
3509 | assert(f); | |
3510 | assert(*f); | |
3511 | ||
3512 | r = journal_file_archive(*f); | |
3513 | if (r < 0) | |
3514 | return r; | |
3515 | ||
3516 | r = journal_file_open( | |
3517 | -1, | |
3518 | (*f)->path, | |
3519 | (*f)->flags, | |
3520 | (*f)->mode, | |
3521 | compress, | |
3522 | compress_threshold_bytes, | |
3523 | seal, | |
3524 | NULL, /* metrics */ | |
3525 | (*f)->mmap, | |
3526 | deferred_closes, | |
3527 | *f, /* template */ | |
3528 | &new_file); | |
3529 | ||
3530 | journal_initiate_close(*f, deferred_closes); | |
3531 | *f = new_file; | |
3532 | ||
3533 | return r; | |
3534 | } | |
3535 | ||
3536 | int journal_file_dispose(int dir_fd, const char *fname) { | |
3537 | _cleanup_free_ char *p = NULL; | |
3538 | _cleanup_close_ int fd = -1; | |
3539 | ||
3540 | assert(fname); | |
3541 | ||
3542 | /* Renames a journal file to *.journal~, i.e. to mark it as corruped or otherwise uncleanly shutdown. Note that | |
3543 | * this is done without looking into the file or changing any of its contents. The idea is that this is called | |
3544 | * whenever something is suspicious and we want to move the file away and make clear that it is not accessed | |
3545 | * for writing anymore. */ | |
3546 | ||
3547 | if (!endswith(fname, ".journal")) | |
3548 | return -EINVAL; | |
3549 | ||
3550 | if (asprintf(&p, "%.*s@%016" PRIx64 "-%016" PRIx64 ".journal~", | |
3551 | (int) strlen(fname) - 8, fname, | |
3552 | now(CLOCK_REALTIME), | |
3553 | random_u64()) < 0) | |
3554 | return -ENOMEM; | |
3555 | ||
3556 | if (renameat(dir_fd, fname, dir_fd, p) < 0) | |
3557 | return -errno; | |
3558 | ||
3559 | /* btrfs doesn't cope well with our write pattern and fragments heavily. Let's defrag all files we rotate */ | |
3560 | fd = openat(dir_fd, p, O_RDONLY|O_CLOEXEC|O_NOCTTY|O_NOFOLLOW); | |
3561 | if (fd < 0) | |
3562 | log_debug_errno(errno, "Failed to open file for defragmentation/FS_NOCOW_FL, ignoring: %m"); | |
3563 | else { | |
3564 | (void) chattr_fd(fd, 0, FS_NOCOW_FL, NULL); | |
3565 | (void) btrfs_defrag_fd(fd); | |
3566 | } | |
3567 | ||
3568 | return 0; | |
3569 | } | |
3570 | ||
3571 | int journal_file_open_reliably( | |
3572 | const char *fname, | |
3573 | int flags, | |
3574 | mode_t mode, | |
3575 | bool compress, | |
3576 | uint64_t compress_threshold_bytes, | |
3577 | bool seal, | |
3578 | JournalMetrics *metrics, | |
3579 | MMapCache *mmap_cache, | |
3580 | Set *deferred_closes, | |
3581 | JournalFile *template, | |
3582 | JournalFile **ret) { | |
3583 | ||
3584 | int r; | |
3585 | ||
3586 | r = journal_file_open(-1, fname, flags, mode, compress, compress_threshold_bytes, seal, metrics, mmap_cache, | |
3587 | deferred_closes, template, ret); | |
3588 | if (!IN_SET(r, | |
3589 | -EBADMSG, /* Corrupted */ | |
3590 | -ENODATA, /* Truncated */ | |
3591 | -EHOSTDOWN, /* Other machine */ | |
3592 | -EPROTONOSUPPORT, /* Incompatible feature */ | |
3593 | -EBUSY, /* Unclean shutdown */ | |
3594 | -ESHUTDOWN, /* Already archived */ | |
3595 | -EIO, /* IO error, including SIGBUS on mmap */ | |
3596 | -EIDRM, /* File has been deleted */ | |
3597 | -ETXTBSY)) /* File is from the future */ | |
3598 | return r; | |
3599 | ||
3600 | if ((flags & O_ACCMODE) == O_RDONLY) | |
3601 | return r; | |
3602 | ||
3603 | if (!(flags & O_CREAT)) | |
3604 | return r; | |
3605 | ||
3606 | if (!endswith(fname, ".journal")) | |
3607 | return r; | |
3608 | ||
3609 | /* The file is corrupted. Rotate it away and try it again (but only once) */ | |
3610 | log_warning_errno(r, "File %s corrupted or uncleanly shut down, renaming and replacing.", fname); | |
3611 | ||
3612 | r = journal_file_dispose(AT_FDCWD, fname); | |
3613 | if (r < 0) | |
3614 | return r; | |
3615 | ||
3616 | return journal_file_open(-1, fname, flags, mode, compress, compress_threshold_bytes, seal, metrics, mmap_cache, | |
3617 | deferred_closes, template, ret); | |
3618 | } | |
3619 | ||
3620 | int journal_file_copy_entry(JournalFile *from, JournalFile *to, Object *o, uint64_t p) { | |
3621 | uint64_t i, n; | |
3622 | uint64_t q, xor_hash = 0; | |
3623 | int r; | |
3624 | EntryItem *items; | |
3625 | dual_timestamp ts; | |
3626 | const sd_id128_t *boot_id; | |
3627 | ||
3628 | assert(from); | |
3629 | assert(to); | |
3630 | assert(o); | |
3631 | assert(p); | |
3632 | ||
3633 | if (!to->writable) | |
3634 | return -EPERM; | |
3635 | ||
3636 | ts.monotonic = le64toh(o->entry.monotonic); | |
3637 | ts.realtime = le64toh(o->entry.realtime); | |
3638 | boot_id = &o->entry.boot_id; | |
3639 | ||
3640 | n = journal_file_entry_n_items(o); | |
3641 | /* alloca() can't take 0, hence let's allocate at least one */ | |
3642 | items = newa(EntryItem, MAX(1u, n)); | |
3643 | ||
3644 | for (i = 0; i < n; i++) { | |
3645 | uint64_t l, h; | |
3646 | le64_t le_hash; | |
3647 | size_t t; | |
3648 | void *data; | |
3649 | Object *u; | |
3650 | ||
3651 | q = le64toh(o->entry.items[i].object_offset); | |
3652 | le_hash = o->entry.items[i].hash; | |
3653 | ||
3654 | r = journal_file_move_to_object(from, OBJECT_DATA, q, &o); | |
3655 | if (r < 0) | |
3656 | return r; | |
3657 | ||
3658 | if (le_hash != o->data.hash) | |
3659 | return -EBADMSG; | |
3660 | ||
3661 | l = le64toh(o->object.size) - offsetof(Object, data.payload); | |
3662 | t = (size_t) l; | |
3663 | ||
3664 | /* We hit the limit on 32bit machines */ | |
3665 | if ((uint64_t) t != l) | |
3666 | return -E2BIG; | |
3667 | ||
3668 | if (o->object.flags & OBJECT_COMPRESSION_MASK) { | |
3669 | #if HAVE_XZ || HAVE_LZ4 | |
3670 | size_t rsize = 0; | |
3671 | ||
3672 | r = decompress_blob(o->object.flags & OBJECT_COMPRESSION_MASK, | |
3673 | o->data.payload, l, &from->compress_buffer, &from->compress_buffer_size, &rsize, 0); | |
3674 | if (r < 0) | |
3675 | return r; | |
3676 | ||
3677 | data = from->compress_buffer; | |
3678 | l = rsize; | |
3679 | #else | |
3680 | return -EPROTONOSUPPORT; | |
3681 | #endif | |
3682 | } else | |
3683 | data = o->data.payload; | |
3684 | ||
3685 | r = journal_file_append_data(to, data, l, &u, &h); | |
3686 | if (r < 0) | |
3687 | return r; | |
3688 | ||
3689 | xor_hash ^= le64toh(u->data.hash); | |
3690 | items[i].object_offset = htole64(h); | |
3691 | items[i].hash = u->data.hash; | |
3692 | ||
3693 | r = journal_file_move_to_object(from, OBJECT_ENTRY, p, &o); | |
3694 | if (r < 0) | |
3695 | return r; | |
3696 | } | |
3697 | ||
3698 | r = journal_file_append_entry_internal(to, &ts, boot_id, xor_hash, items, n, | |
3699 | NULL, NULL, NULL); | |
3700 | ||
3701 | if (mmap_cache_got_sigbus(to->mmap, to->cache_fd)) | |
3702 | return -EIO; | |
3703 | ||
3704 | return r; | |
3705 | } | |
3706 | ||
3707 | void journal_reset_metrics(JournalMetrics *m) { | |
3708 | assert(m); | |
3709 | ||
3710 | /* Set everything to "pick automatic values". */ | |
3711 | ||
3712 | *m = (JournalMetrics) { | |
3713 | .min_use = (uint64_t) -1, | |
3714 | .max_use = (uint64_t) -1, | |
3715 | .min_size = (uint64_t) -1, | |
3716 | .max_size = (uint64_t) -1, | |
3717 | .keep_free = (uint64_t) -1, | |
3718 | .n_max_files = (uint64_t) -1, | |
3719 | }; | |
3720 | } | |
3721 | ||
3722 | void journal_default_metrics(JournalMetrics *m, int fd) { | |
3723 | char a[FORMAT_BYTES_MAX], b[FORMAT_BYTES_MAX], c[FORMAT_BYTES_MAX], d[FORMAT_BYTES_MAX], e[FORMAT_BYTES_MAX]; | |
3724 | struct statvfs ss; | |
3725 | uint64_t fs_size; | |
3726 | ||
3727 | assert(m); | |
3728 | assert(fd >= 0); | |
3729 | ||
3730 | if (fstatvfs(fd, &ss) >= 0) | |
3731 | fs_size = ss.f_frsize * ss.f_blocks; | |
3732 | else { | |
3733 | log_debug_errno(errno, "Failed to determine disk size: %m"); | |
3734 | fs_size = 0; | |
3735 | } | |
3736 | ||
3737 | if (m->max_use == (uint64_t) -1) { | |
3738 | ||
3739 | if (fs_size > 0) { | |
3740 | m->max_use = PAGE_ALIGN(fs_size / 10); /* 10% of file system size */ | |
3741 | ||
3742 | if (m->max_use > DEFAULT_MAX_USE_UPPER) | |
3743 | m->max_use = DEFAULT_MAX_USE_UPPER; | |
3744 | ||
3745 | if (m->max_use < DEFAULT_MAX_USE_LOWER) | |
3746 | m->max_use = DEFAULT_MAX_USE_LOWER; | |
3747 | } else | |
3748 | m->max_use = DEFAULT_MAX_USE_LOWER; | |
3749 | } else { | |
3750 | m->max_use = PAGE_ALIGN(m->max_use); | |
3751 | ||
3752 | if (m->max_use != 0 && m->max_use < JOURNAL_FILE_SIZE_MIN*2) | |
3753 | m->max_use = JOURNAL_FILE_SIZE_MIN*2; | |
3754 | } | |
3755 | ||
3756 | if (m->min_use == (uint64_t) -1) | |
3757 | m->min_use = DEFAULT_MIN_USE; | |
3758 | ||
3759 | if (m->min_use > m->max_use) | |
3760 | m->min_use = m->max_use; | |
3761 | ||
3762 | if (m->max_size == (uint64_t) -1) { | |
3763 | m->max_size = PAGE_ALIGN(m->max_use / 8); /* 8 chunks */ | |
3764 | ||
3765 | if (m->max_size > DEFAULT_MAX_SIZE_UPPER) | |
3766 | m->max_size = DEFAULT_MAX_SIZE_UPPER; | |
3767 | } else | |
3768 | m->max_size = PAGE_ALIGN(m->max_size); | |
3769 | ||
3770 | if (m->max_size != 0) { | |
3771 | if (m->max_size < JOURNAL_FILE_SIZE_MIN) | |
3772 | m->max_size = JOURNAL_FILE_SIZE_MIN; | |
3773 | ||
3774 | if (m->max_use != 0 && m->max_size*2 > m->max_use) | |
3775 | m->max_use = m->max_size*2; | |
3776 | } | |
3777 | ||
3778 | if (m->min_size == (uint64_t) -1) | |
3779 | m->min_size = JOURNAL_FILE_SIZE_MIN; | |
3780 | else { | |
3781 | m->min_size = PAGE_ALIGN(m->min_size); | |
3782 | ||
3783 | if (m->min_size < JOURNAL_FILE_SIZE_MIN) | |
3784 | m->min_size = JOURNAL_FILE_SIZE_MIN; | |
3785 | ||
3786 | if (m->max_size != 0 && m->min_size > m->max_size) | |
3787 | m->max_size = m->min_size; | |
3788 | } | |
3789 | ||
3790 | if (m->keep_free == (uint64_t) -1) { | |
3791 | ||
3792 | if (fs_size > 0) { | |
3793 | m->keep_free = PAGE_ALIGN(fs_size * 3 / 20); /* 15% of file system size */ | |
3794 | ||
3795 | if (m->keep_free > DEFAULT_KEEP_FREE_UPPER) | |
3796 | m->keep_free = DEFAULT_KEEP_FREE_UPPER; | |
3797 | ||
3798 | } else | |
3799 | m->keep_free = DEFAULT_KEEP_FREE; | |
3800 | } | |
3801 | ||
3802 | if (m->n_max_files == (uint64_t) -1) | |
3803 | m->n_max_files = DEFAULT_N_MAX_FILES; | |
3804 | ||
3805 | log_debug("Fixed min_use=%s max_use=%s max_size=%s min_size=%s keep_free=%s n_max_files=%" PRIu64, | |
3806 | format_bytes(a, sizeof(a), m->min_use), | |
3807 | format_bytes(b, sizeof(b), m->max_use), | |
3808 | format_bytes(c, sizeof(c), m->max_size), | |
3809 | format_bytes(d, sizeof(d), m->min_size), | |
3810 | format_bytes(e, sizeof(e), m->keep_free), | |
3811 | m->n_max_files); | |
3812 | } | |
3813 | ||
3814 | int journal_file_get_cutoff_realtime_usec(JournalFile *f, usec_t *from, usec_t *to) { | |
3815 | assert(f); | |
3816 | assert(f->header); | |
3817 | assert(from || to); | |
3818 | ||
3819 | if (from) { | |
3820 | if (f->header->head_entry_realtime == 0) | |
3821 | return -ENOENT; | |
3822 | ||
3823 | *from = le64toh(f->header->head_entry_realtime); | |
3824 | } | |
3825 | ||
3826 | if (to) { | |
3827 | if (f->header->tail_entry_realtime == 0) | |
3828 | return -ENOENT; | |
3829 | ||
3830 | *to = le64toh(f->header->tail_entry_realtime); | |
3831 | } | |
3832 | ||
3833 | return 1; | |
3834 | } | |
3835 | ||
3836 | int journal_file_get_cutoff_monotonic_usec(JournalFile *f, sd_id128_t boot_id, usec_t *from, usec_t *to) { | |
3837 | Object *o; | |
3838 | uint64_t p; | |
3839 | int r; | |
3840 | ||
3841 | assert(f); | |
3842 | assert(from || to); | |
3843 | ||
3844 | r = find_data_object_by_boot_id(f, boot_id, &o, &p); | |
3845 | if (r <= 0) | |
3846 | return r; | |
3847 | ||
3848 | if (le64toh(o->data.n_entries) <= 0) | |
3849 | return 0; | |
3850 | ||
3851 | if (from) { | |
3852 | r = journal_file_move_to_object(f, OBJECT_ENTRY, le64toh(o->data.entry_offset), &o); | |
3853 | if (r < 0) | |
3854 | return r; | |
3855 | ||
3856 | *from = le64toh(o->entry.monotonic); | |
3857 | } | |
3858 | ||
3859 | if (to) { | |
3860 | r = journal_file_move_to_object(f, OBJECT_DATA, p, &o); | |
3861 | if (r < 0) | |
3862 | return r; | |
3863 | ||
3864 | r = generic_array_get_plus_one(f, | |
3865 | le64toh(o->data.entry_offset), | |
3866 | le64toh(o->data.entry_array_offset), | |
3867 | le64toh(o->data.n_entries)-1, | |
3868 | &o, NULL); | |
3869 | if (r <= 0) | |
3870 | return r; | |
3871 | ||
3872 | *to = le64toh(o->entry.monotonic); | |
3873 | } | |
3874 | ||
3875 | return 1; | |
3876 | } | |
3877 | ||
3878 | bool journal_file_rotate_suggested(JournalFile *f, usec_t max_file_usec) { | |
3879 | assert(f); | |
3880 | assert(f->header); | |
3881 | ||
3882 | /* If we gained new header fields we gained new features, | |
3883 | * hence suggest a rotation */ | |
3884 | if (le64toh(f->header->header_size) < sizeof(Header)) { | |
3885 | log_debug("%s uses an outdated header, suggesting rotation.", f->path); | |
3886 | return true; | |
3887 | } | |
3888 | ||
3889 | /* Let's check if the hash tables grew over a certain fill | |
3890 | * level (75%, borrowing this value from Java's hash table | |
3891 | * implementation), and if so suggest a rotation. To calculate | |
3892 | * the fill level we need the n_data field, which only exists | |
3893 | * in newer versions. */ | |
3894 | ||
3895 | if (JOURNAL_HEADER_CONTAINS(f->header, n_data)) | |
3896 | if (le64toh(f->header->n_data) * 4ULL > (le64toh(f->header->data_hash_table_size) / sizeof(HashItem)) * 3ULL) { | |
3897 | 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.", | |
3898 | f->path, | |
3899 | 100.0 * (double) le64toh(f->header->n_data) / ((double) (le64toh(f->header->data_hash_table_size) / sizeof(HashItem))), | |
3900 | le64toh(f->header->n_data), | |
3901 | le64toh(f->header->data_hash_table_size) / sizeof(HashItem), | |
3902 | (unsigned long long) f->last_stat.st_size, | |
3903 | f->last_stat.st_size / le64toh(f->header->n_data)); | |
3904 | return true; | |
3905 | } | |
3906 | ||
3907 | if (JOURNAL_HEADER_CONTAINS(f->header, n_fields)) | |
3908 | if (le64toh(f->header->n_fields) * 4ULL > (le64toh(f->header->field_hash_table_size) / sizeof(HashItem)) * 3ULL) { | |
3909 | log_debug("Field hash table of %s has a fill level at %.1f (%"PRIu64" of %"PRIu64" items), suggesting rotation.", | |
3910 | f->path, | |
3911 | 100.0 * (double) le64toh(f->header->n_fields) / ((double) (le64toh(f->header->field_hash_table_size) / sizeof(HashItem))), | |
3912 | le64toh(f->header->n_fields), | |
3913 | le64toh(f->header->field_hash_table_size) / sizeof(HashItem)); | |
3914 | return true; | |
3915 | } | |
3916 | ||
3917 | /* Are the data objects properly indexed by field objects? */ | |
3918 | if (JOURNAL_HEADER_CONTAINS(f->header, n_data) && | |
3919 | JOURNAL_HEADER_CONTAINS(f->header, n_fields) && | |
3920 | le64toh(f->header->n_data) > 0 && | |
3921 | le64toh(f->header->n_fields) == 0) | |
3922 | return true; | |
3923 | ||
3924 | if (max_file_usec > 0) { | |
3925 | usec_t t, h; | |
3926 | ||
3927 | h = le64toh(f->header->head_entry_realtime); | |
3928 | t = now(CLOCK_REALTIME); | |
3929 | ||
3930 | if (h > 0 && t > h + max_file_usec) | |
3931 | return true; | |
3932 | } | |
3933 | ||
3934 | return false; | |
3935 | } |