]>
Commit | Line | Data |
---|---|---|
a4b75251 TL |
1 | // SPDX-License-Identifier: BSD-3-Clause |
2 | /* Copyright 2016-2020, Intel Corporation */ | |
3 | ||
4 | /* | |
5 | * replica.c -- groups all commands for replica manipulation | |
6 | */ | |
7 | ||
8 | #include "replica.h" | |
9 | ||
10 | #include <errno.h> | |
11 | #include <sys/mman.h> | |
12 | #include <errno.h> | |
13 | #include <fcntl.h> | |
14 | #include <stddef.h> | |
15 | #include <stdint.h> | |
16 | #include <string.h> | |
17 | #include <unistd.h> | |
18 | #include <libgen.h> | |
19 | ||
20 | #include "obj.h" | |
21 | #include "palloc.h" | |
22 | #include "file.h" | |
23 | #include "os.h" | |
24 | #include "out.h" | |
25 | #include "pool_hdr.h" | |
26 | #include "set.h" | |
27 | #include "util.h" | |
28 | #include "uuid.h" | |
29 | #include "shutdown_state.h" | |
30 | #include "badblocks.h" | |
31 | #include "set_badblocks.h" | |
32 | ||
33 | /* | |
34 | * check_flags_sync -- (internal) check if flags are supported for sync | |
35 | */ | |
36 | static int | |
37 | check_flags_sync(unsigned flags) | |
38 | { | |
39 | flags &= ~(PMEMPOOL_SYNC_DRY_RUN | PMEMPOOL_SYNC_FIX_BAD_BLOCKS); | |
40 | return flags > 0; | |
41 | } | |
42 | ||
43 | /* | |
44 | * check_flags_transform -- (internal) check if flags are supported for | |
45 | * transform | |
46 | */ | |
47 | static int | |
48 | check_flags_transform(unsigned flags) | |
49 | { | |
50 | flags &= ~PMEMPOOL_TRANSFORM_DRY_RUN; | |
51 | return flags > 0; | |
52 | } | |
53 | ||
54 | /* | |
55 | * replica_align_badblock_offset_length -- align offset and length | |
56 | * of the bad block for the given part | |
57 | */ | |
58 | void | |
59 | replica_align_badblock_offset_length(size_t *offset, size_t *length, | |
60 | struct pool_set *set_in, unsigned repn, unsigned partn) | |
61 | { | |
62 | LOG(3, "offset %zu, length %zu, pool_set %p, replica %u, part %u", | |
63 | *offset, *length, set_in, repn, partn); | |
64 | ||
65 | size_t alignment = set_in->replica[repn]->part[partn].alignment; | |
66 | ||
67 | size_t off = ALIGN_DOWN(*offset, alignment); | |
68 | size_t len = ALIGN_UP(*length + (*offset - off), alignment); | |
69 | ||
70 | *offset = off; | |
71 | *length = len; | |
72 | } | |
73 | ||
74 | /* | |
75 | * replica_get_part_data_len -- get data length for given part | |
76 | */ | |
77 | size_t | |
78 | replica_get_part_data_len(struct pool_set *set_in, unsigned repn, | |
79 | unsigned partn) | |
80 | { | |
81 | size_t alignment = set_in->replica[repn]->part[partn].alignment; | |
82 | size_t hdrsize = (set_in->options & OPTION_SINGLEHDR) ? 0 : alignment; | |
83 | return ALIGN_DOWN(set_in->replica[repn]->part[partn].filesize, | |
84 | alignment) - ((partn == 0) ? POOL_HDR_SIZE : hdrsize); | |
85 | } | |
86 | ||
87 | /* | |
88 | * replica_get_part_offset -- get part's offset from the beginning of replica | |
89 | */ | |
90 | uint64_t | |
91 | replica_get_part_offset(struct pool_set *set, unsigned repn, unsigned partn) | |
92 | { | |
93 | return (uint64_t)set->replica[repn]->part[partn].addr - | |
94 | (uint64_t)set->replica[repn]->part[0].addr; | |
95 | } | |
96 | ||
97 | /* | |
98 | * replica_get_part_data_offset -- get data length before given part | |
99 | */ | |
100 | uint64_t | |
101 | replica_get_part_data_offset(struct pool_set *set, unsigned repn, | |
102 | unsigned partn) | |
103 | { | |
104 | if (partn == 0) | |
105 | return POOL_HDR_SIZE; | |
106 | ||
107 | return (uint64_t)set->replica[repn]->part[partn].addr - | |
108 | (uint64_t)set->replica[repn]->part[0].addr; | |
109 | } | |
110 | ||
111 | /* | |
112 | * replica_remove_part -- unlink part from replica | |
113 | */ | |
114 | int | |
115 | replica_remove_part(struct pool_set *set, unsigned repn, unsigned partn, | |
116 | int fix_bad_blocks) | |
117 | { | |
118 | LOG(3, "set %p repn %u partn %u fix_bad_blocks %i", | |
119 | set, repn, partn, fix_bad_blocks); | |
120 | ||
121 | struct pool_set_part *part = PART(REP(set, repn), partn); | |
122 | if (part->fd != -1) { | |
123 | os_close(part->fd); | |
124 | part->fd = -1; | |
125 | } | |
126 | ||
127 | int olderrno = errno; | |
128 | enum file_type type = util_file_get_type(part->path); | |
129 | if (type == OTHER_ERROR) | |
130 | return -1; | |
131 | ||
132 | /* if the part is a device dax, clear its bad blocks */ | |
133 | if (type == TYPE_DEVDAX && fix_bad_blocks && | |
134 | badblocks_clear_all(part->path)) { | |
135 | ERR("clearing bad blocks in device dax failed -- '%s'", | |
136 | part->path); | |
137 | errno = EIO; | |
138 | return -1; | |
139 | } | |
140 | ||
141 | if (type == TYPE_NORMAL && util_unlink(part->path)) { | |
142 | ERR("!removing part %u from replica %u failed", | |
143 | partn, repn); | |
144 | return -1; | |
145 | } | |
146 | ||
147 | errno = olderrno; | |
148 | LOG(4, "Removed part %s number %u from replica %u", part->path, partn, | |
149 | repn); | |
150 | return 0; | |
151 | } | |
152 | ||
153 | /* | |
154 | * create_replica_health_status -- (internal) create helping structure for | |
155 | * storing replica's health status | |
156 | */ | |
157 | static struct replica_health_status * | |
158 | create_replica_health_status(struct pool_set *set, unsigned repn) | |
159 | { | |
160 | LOG(3, "set %p, repn %u", set, repn); | |
161 | ||
162 | unsigned nparts = set->replica[repn]->nparts; | |
163 | struct replica_health_status *replica_hs; | |
164 | ||
165 | replica_hs = Zalloc(sizeof(struct replica_health_status) | |
166 | + nparts * sizeof(struct part_health_status)); | |
167 | if (replica_hs == NULL) { | |
168 | ERR("!Zalloc for replica health status"); | |
169 | return NULL; | |
170 | } | |
171 | ||
172 | replica_hs->nparts = nparts; | |
173 | replica_hs->nhdrs = set->replica[repn]->nhdrs; | |
174 | ||
175 | return replica_hs; | |
176 | } | |
177 | ||
178 | /* | |
179 | * replica_part_remove_recovery_file -- remove bad blocks' recovery file | |
180 | */ | |
181 | static int | |
182 | replica_part_remove_recovery_file(struct part_health_status *phs) | |
183 | { | |
184 | LOG(3, "phs %p", phs); | |
185 | ||
186 | if (phs->recovery_file_name == NULL || phs->recovery_file_exists == 0) | |
187 | return 0; | |
188 | ||
189 | if (os_unlink(phs->recovery_file_name) < 0) { | |
190 | ERR("!removing the bad block recovery file failed -- '%s'", | |
191 | phs->recovery_file_name); | |
192 | return -1; | |
193 | } | |
194 | ||
195 | LOG(3, "bad block recovery file removed -- '%s'", | |
196 | phs->recovery_file_name); | |
197 | ||
198 | phs->recovery_file_exists = 0; | |
199 | ||
200 | return 0; | |
201 | } | |
202 | ||
203 | /* | |
204 | * replica_remove_all_recovery_files -- remove all recovery files | |
205 | */ | |
206 | int | |
207 | replica_remove_all_recovery_files(struct poolset_health_status *set_hs) | |
208 | { | |
209 | LOG(3, "set_hs %p", set_hs); | |
210 | ||
211 | int ret = 0; | |
212 | ||
213 | for (unsigned r = 0; r < set_hs->nreplicas; ++r) { | |
214 | struct replica_health_status *rhs = set_hs->replica[r]; | |
215 | for (unsigned p = 0; p < rhs->nparts; ++p) | |
216 | ret |= replica_part_remove_recovery_file(&rhs->part[p]); | |
217 | } | |
218 | ||
219 | return ret; | |
220 | } | |
221 | ||
222 | /* | |
223 | * replica_free_poolset_health_status -- free memory allocated for helping | |
224 | * structure | |
225 | */ | |
226 | void | |
227 | replica_free_poolset_health_status(struct poolset_health_status *set_hs) | |
228 | { | |
229 | LOG(3, "set_hs %p", set_hs); | |
230 | ||
231 | for (unsigned r = 0; r < set_hs->nreplicas; ++r) { | |
232 | struct replica_health_status *rep_hs = set_hs->replica[r]; | |
233 | ||
234 | for (unsigned p = 0; p < rep_hs->nparts; ++p) { | |
235 | Free(rep_hs->part[p].recovery_file_name); | |
236 | Free(rep_hs->part[p].bbs.bbv); | |
237 | } | |
238 | ||
239 | Free(set_hs->replica[r]); | |
240 | } | |
241 | ||
242 | Free(set_hs); | |
243 | } | |
244 | ||
245 | /* | |
246 | * replica_create_poolset_health_status -- create helping structure for storing | |
247 | * poolset's health status | |
248 | */ | |
249 | int | |
250 | replica_create_poolset_health_status(struct pool_set *set, | |
251 | struct poolset_health_status **set_hsp) | |
252 | { | |
253 | LOG(3, "set %p, set_hsp %p", set, set_hsp); | |
254 | unsigned nreplicas = set->nreplicas; | |
255 | struct poolset_health_status *set_hs; | |
256 | set_hs = Zalloc(sizeof(struct poolset_health_status) + | |
257 | nreplicas * sizeof(struct replica_health_status *)); | |
258 | if (set_hs == NULL) { | |
259 | ERR("!Zalloc for poolset health state"); | |
260 | return -1; | |
261 | } | |
262 | set_hs->nreplicas = nreplicas; | |
263 | for (unsigned i = 0; i < nreplicas; ++i) { | |
264 | struct replica_health_status *replica_hs = | |
265 | create_replica_health_status(set, i); | |
266 | if (replica_hs == NULL) { | |
267 | replica_free_poolset_health_status(set_hs); | |
268 | return -1; | |
269 | } | |
270 | set_hs->replica[i] = replica_hs; | |
271 | } | |
272 | *set_hsp = set_hs; | |
273 | return 0; | |
274 | } | |
275 | ||
276 | /* | |
277 | * replica_is_part_broken -- check if part is marked as broken in the helping | |
278 | * structure | |
279 | */ | |
280 | int | |
281 | replica_is_part_broken(unsigned repn, unsigned partn, | |
282 | struct poolset_health_status *set_hs) | |
283 | { | |
284 | struct replica_health_status *rhs = REP_HEALTH(set_hs, repn); | |
285 | return (rhs->flags & IS_BROKEN) || | |
286 | (PART_HEALTH(rhs, partn) & IS_BROKEN); | |
287 | } | |
288 | ||
289 | /* | |
290 | * is_replica_broken -- check if any part in the replica is marked as broken | |
291 | */ | |
292 | int | |
293 | replica_is_replica_broken(unsigned repn, struct poolset_health_status *set_hs) | |
294 | { | |
295 | LOG(3, "repn %u, set_hs %p", repn, set_hs); | |
296 | struct replica_health_status *r_hs = REP_HEALTH(set_hs, repn); | |
297 | if (r_hs->flags & IS_BROKEN) | |
298 | return 1; | |
299 | ||
300 | for (unsigned p = 0; p < r_hs->nparts; ++p) { | |
301 | if (replica_is_part_broken(repn, p, set_hs)) | |
302 | return 1; | |
303 | } | |
304 | return 0; | |
305 | } | |
306 | ||
307 | /* | |
308 | * replica_is_replica_consistent -- check if replica is not marked as | |
309 | * inconsistent | |
310 | */ | |
311 | int | |
312 | replica_is_replica_consistent(unsigned repn, | |
313 | struct poolset_health_status *set_hs) | |
314 | { | |
315 | return !(REP_HEALTH(set_hs, repn)->flags & IS_INCONSISTENT); | |
316 | } | |
317 | ||
318 | /* | |
319 | * replica_has_bad_blocks -- check if replica has bad blocks | |
320 | */ | |
321 | int | |
322 | replica_has_bad_blocks(unsigned repn, struct poolset_health_status *set_hs) | |
323 | { | |
324 | return REP_HEALTH(set_hs, repn)->flags & HAS_BAD_BLOCKS; | |
325 | } | |
326 | ||
327 | /* | |
328 | * replica_part_has_bad_blocks -- check if replica's part has bad blocks | |
329 | */ | |
330 | int | |
331 | replica_part_has_bad_blocks(struct part_health_status *phs) | |
332 | { | |
333 | return phs->flags & HAS_BAD_BLOCKS; | |
334 | } | |
335 | ||
336 | /* | |
337 | * replica_part_has_corrupted_header -- (internal) check if replica's part | |
338 | * has bad blocks in the header (corrupted header) | |
339 | */ | |
340 | int | |
341 | replica_part_has_corrupted_header(unsigned repn, unsigned partn, | |
342 | struct poolset_health_status *set_hs) | |
343 | { | |
344 | struct replica_health_status *rhs = REP_HEALTH(set_hs, repn); | |
345 | return PART_HEALTH(rhs, partn) & HAS_CORRUPTED_HEADER; | |
346 | } | |
347 | ||
348 | /* | |
349 | * replica_has_corrupted_header -- (internal) check if replica has bad blocks | |
350 | * in the header (corrupted header) | |
351 | */ | |
352 | static int | |
353 | replica_has_corrupted_header(unsigned repn, | |
354 | struct poolset_health_status *set_hs) | |
355 | { | |
356 | return REP_HEALTH(set_hs, repn)->flags & HAS_CORRUPTED_HEADER; | |
357 | } | |
358 | ||
359 | /* | |
360 | * replica_is_replica_healthy -- check if replica is unbroken and consistent | |
361 | */ | |
362 | int | |
363 | replica_is_replica_healthy(unsigned repn, struct poolset_health_status *set_hs) | |
364 | { | |
365 | LOG(3, "repn %u, set_hs %p", repn, set_hs); | |
366 | ||
367 | int ret = !replica_is_replica_broken(repn, set_hs) && | |
368 | replica_is_replica_consistent(repn, set_hs) && | |
369 | !replica_has_bad_blocks(repn, set_hs); | |
370 | ||
371 | LOG(4, "return %i", ret); | |
372 | ||
373 | return ret; | |
374 | } | |
375 | ||
376 | /* | |
377 | * replica_has_healthy_header -- (internal) check if replica has healthy headers | |
378 | */ | |
379 | static int | |
380 | replica_has_healthy_header(unsigned repn, struct poolset_health_status *set_hs) | |
381 | { | |
382 | LOG(3, "repn %u, set_hs %p", repn, set_hs); | |
383 | ||
384 | int ret = !replica_is_replica_broken(repn, set_hs) && | |
385 | replica_is_replica_consistent(repn, set_hs) && | |
386 | !replica_has_corrupted_header(repn, set_hs); | |
387 | ||
388 | LOG(4, "return %i", ret); | |
389 | ||
390 | return ret; | |
391 | } | |
392 | ||
393 | /* | |
394 | * replica_is_poolset_healthy -- check if all replicas in a poolset are not | |
395 | * marked as broken nor inconsistent in the | |
396 | * helping structure | |
397 | */ | |
398 | int | |
399 | replica_is_poolset_healthy(struct poolset_health_status *set_hs) | |
400 | { | |
401 | LOG(3, "set_hs %p", set_hs); | |
402 | for (unsigned r = 0; r < set_hs->nreplicas; ++r) { | |
403 | if (!replica_is_replica_healthy(r, set_hs)) | |
404 | return 0; | |
405 | } | |
406 | return 1; | |
407 | } | |
408 | ||
409 | /* | |
410 | * replica_is_poolset_transformed -- check if the flag indicating a call from | |
411 | * pmempool_transform is on | |
412 | */ | |
413 | int | |
414 | replica_is_poolset_transformed(unsigned flags) | |
415 | { | |
416 | return flags & IS_TRANSFORMED; | |
417 | } | |
418 | ||
419 | /* | |
420 | * replica_find_unbroken_part_with_header -- find a part number in a given | |
421 | * replica, which is not marked as broken in the helping structure and contains | |
422 | * a pool header | |
423 | */ | |
424 | unsigned | |
425 | replica_find_unbroken_part(unsigned repn, struct poolset_health_status *set_hs) | |
426 | { | |
427 | LOG(3, "repn %u, set_hs %p", repn, set_hs); | |
428 | for (unsigned p = 0; p < REP_HEALTH(set_hs, repn)->nhdrs; ++p) { | |
429 | if (!replica_is_part_broken(repn, p, set_hs)) | |
430 | return p; | |
431 | } | |
432 | return UNDEF_PART; | |
433 | } | |
434 | ||
435 | /* | |
436 | * replica_find_healthy_replica -- find a replica which is a good source of data | |
437 | */ | |
438 | unsigned | |
439 | replica_find_healthy_replica(struct poolset_health_status *set_hs) | |
440 | { | |
441 | LOG(3, "set_hs %p", set_hs); | |
442 | ||
443 | for (unsigned r = 0; r < set_hs->nreplicas; ++r) { | |
444 | if (replica_is_replica_healthy(r, set_hs)) { | |
445 | LOG(4, "return %i", r); | |
446 | return r; | |
447 | } | |
448 | } | |
449 | ||
450 | LOG(4, "return %i", UNDEF_REPLICA); | |
451 | return UNDEF_REPLICA; | |
452 | } | |
453 | ||
454 | /* | |
455 | * replica_find_replica_healthy_header -- find a replica with a healthy header | |
456 | */ | |
457 | unsigned | |
458 | replica_find_replica_healthy_header(struct poolset_health_status *set_hs) | |
459 | { | |
460 | LOG(3, "set_hs %p", set_hs); | |
461 | ||
462 | for (unsigned r = 0; r < set_hs->nreplicas; ++r) { | |
463 | if (replica_has_healthy_header(r, set_hs)) { | |
464 | LOG(4, "return %i", r); | |
465 | return r; | |
466 | } | |
467 | } | |
468 | ||
469 | LOG(4, "return %i", UNDEF_REPLICA); | |
470 | return UNDEF_REPLICA; | |
471 | } | |
472 | ||
473 | /* | |
474 | * replica_check_store_size -- (internal) store size from pool descriptor for | |
475 | * replica | |
476 | */ | |
477 | static int | |
478 | replica_check_store_size(struct pool_set *set, | |
479 | struct poolset_health_status *set_hs, unsigned repn) | |
480 | { | |
481 | LOG(3, "set %p, set_hs %p, repn %u", set, set_hs, repn); | |
482 | struct pool_replica *rep = set->replica[repn]; | |
483 | struct pmemobjpool pop; | |
484 | ||
485 | if (rep->remote) { | |
486 | memcpy(&pop.hdr, rep->part[0].hdr, sizeof(pop.hdr)); | |
487 | void *descr = (void *)((uintptr_t)&pop + POOL_HDR_SIZE); | |
488 | if (Rpmem_read(rep->remote->rpp, descr, POOL_HDR_SIZE, | |
489 | sizeof(pop) - POOL_HDR_SIZE, 0)) { | |
490 | return -1; | |
491 | } | |
492 | } else { | |
493 | /* round up map size to Mmap align size */ | |
494 | if (util_map_part(&rep->part[0], NULL, | |
495 | ALIGN_UP(sizeof(pop), rep->part[0].alignment), | |
496 | 0, MAP_SHARED, 1)) { | |
497 | return -1; | |
498 | } | |
499 | ||
500 | memcpy(&pop, rep->part[0].addr, sizeof(pop)); | |
501 | ||
502 | util_unmap_part(&rep->part[0]); | |
503 | } | |
504 | ||
505 | void *dscp = (void *)((uintptr_t)&pop + sizeof(pop.hdr)); | |
506 | ||
507 | if (!util_checksum(dscp, OBJ_DSC_P_SIZE, &pop.checksum, 0, | |
508 | 0)) { | |
509 | set_hs->replica[repn]->flags |= IS_BROKEN; | |
510 | return 0; | |
511 | } | |
512 | ||
513 | set_hs->replica[repn]->pool_size = pop.heap_offset + pop.heap_size; | |
514 | ||
515 | return 0; | |
516 | } | |
517 | ||
518 | /* | |
519 | * check_store_all_sizes -- (internal) store sizes from pool descriptor for all | |
520 | * healthy replicas | |
521 | */ | |
522 | static int | |
523 | check_store_all_sizes(struct pool_set *set, | |
524 | struct poolset_health_status *set_hs) | |
525 | { | |
526 | LOG(3, "set %p, set_hs %p", set, set_hs); | |
527 | for (unsigned r = 0; r < set->nreplicas; ++r) { | |
528 | if (!replica_has_healthy_header(r, set_hs)) | |
529 | continue; | |
530 | ||
531 | if (replica_check_store_size(set, set_hs, r)) | |
532 | return -1; | |
533 | } | |
534 | ||
535 | return 0; | |
536 | } | |
537 | ||
538 | /* | |
539 | * check_and_open_poolset_part_files -- (internal) for each part in a poolset | |
540 | * check if the part files are accessible, and if not, mark it as broken | |
541 | * in a helping structure; then open the part file | |
542 | */ | |
543 | static int | |
544 | check_and_open_poolset_part_files(struct pool_set *set, | |
545 | struct poolset_health_status *set_hs, unsigned flags) | |
546 | { | |
547 | LOG(3, "set %p, set_hs %p, flags %u", set, set_hs, flags); | |
548 | for (unsigned r = 0; r < set->nreplicas; ++r) { | |
549 | struct pool_replica *rep = set->replica[r]; | |
550 | struct replica_health_status *rep_hs = set_hs->replica[r]; | |
551 | if (rep->remote) { | |
552 | if (util_replica_open_remote(set, r, 0)) { | |
553 | LOG(1, "cannot open remote replica no %u", r); | |
554 | return -1; | |
555 | } | |
556 | ||
557 | unsigned nlanes = REMOTE_NLANES; | |
558 | int ret = util_poolset_remote_open(rep, r, | |
559 | rep->repsize, 0, | |
560 | rep->part[0].addr, | |
561 | rep->resvsize, &nlanes); | |
562 | if (ret) { | |
563 | rep_hs->flags |= IS_BROKEN; | |
564 | LOG(1, "remote replica #%u marked as BROKEN", | |
565 | r); | |
566 | } | |
567 | ||
568 | continue; | |
569 | } | |
570 | ||
571 | for (unsigned p = 0; p < rep->nparts; ++p) { | |
572 | const char *path = rep->part[p].path; | |
573 | enum file_type type = util_file_get_type(path); | |
574 | ||
575 | if (type < 0 || os_access(path, R_OK|W_OK) != 0) { | |
576 | LOG(1, "part file %s is not accessible", path); | |
577 | errno = 0; | |
578 | rep_hs->part[p].flags |= IS_BROKEN; | |
579 | if (is_dry_run(flags)) | |
580 | continue; | |
581 | } | |
582 | ||
583 | if (util_part_open(&rep->part[p], 0, 0)) { | |
584 | if (type == TYPE_DEVDAX) { | |
585 | LOG(1, | |
586 | "opening part on Device DAX %s failed", | |
587 | path); | |
588 | return -1; | |
589 | } | |
590 | LOG(1, "opening part %s failed", path); | |
591 | errno = 0; | |
592 | rep_hs->part[p].flags |= IS_BROKEN; | |
593 | } | |
594 | } | |
595 | } | |
596 | return 0; | |
597 | } | |
598 | ||
599 | /* | |
600 | * map_all_unbroken_headers -- (internal) map all headers in a poolset, | |
601 | * skipping those marked as broken in a helping | |
602 | * structure | |
603 | */ | |
604 | static int | |
605 | map_all_unbroken_headers(struct pool_set *set, | |
606 | struct poolset_health_status *set_hs) | |
607 | { | |
608 | LOG(3, "set %p, set_hs %p", set, set_hs); | |
609 | for (unsigned r = 0; r < set->nreplicas; ++r) { | |
610 | struct pool_replica *rep = set->replica[r]; | |
611 | struct replica_health_status *rep_hs = set_hs->replica[r]; | |
612 | if (rep->remote) | |
613 | continue; | |
614 | ||
615 | for (unsigned p = 0; p < rep->nhdrs; ++p) { | |
616 | /* skip broken parts */ | |
617 | if (replica_is_part_broken(r, p, set_hs)) | |
618 | continue; | |
619 | ||
620 | LOG(4, "mapping header for part %u, replica %u", p, r); | |
621 | if (util_map_hdr(&rep->part[p], MAP_SHARED, 0) != 0) { | |
622 | LOG(1, "header mapping failed - part #%d", p); | |
623 | rep_hs->part[p].flags |= IS_BROKEN; | |
624 | } | |
625 | } | |
626 | } | |
627 | return 0; | |
628 | } | |
629 | ||
630 | /* | |
631 | * unmap_all_headers -- (internal) unmap all headers in a poolset | |
632 | */ | |
633 | static int | |
634 | unmap_all_headers(struct pool_set *set) | |
635 | { | |
636 | LOG(3, "set %p", set); | |
637 | for (unsigned r = 0; r < set->nreplicas; ++r) { | |
638 | struct pool_replica *rep = set->replica[r]; | |
639 | util_replica_close(set, r); | |
640 | ||
641 | if (rep->remote && rep->remote->rpp) { | |
642 | Rpmem_close(rep->remote->rpp); | |
643 | rep->remote->rpp = NULL; | |
644 | } | |
645 | } | |
646 | ||
647 | return 0; | |
648 | } | |
649 | ||
650 | /* | |
651 | * check_checksums_and_signatures -- (internal) check if checksums | |
652 | * and signatures are correct for parts | |
653 | * in a given replica | |
654 | */ | |
655 | static int | |
656 | check_checksums_and_signatures(struct pool_set *set, | |
657 | struct poolset_health_status *set_hs) | |
658 | { | |
659 | LOG(3, "set %p, set_hs %p", set, set_hs); | |
660 | for (unsigned r = 0; r < set->nreplicas; ++r) { | |
661 | struct pool_replica *rep = REP(set, r); | |
662 | struct replica_health_status *rep_hs = REP_HEALTH(set_hs, r); | |
663 | ||
664 | /* | |
665 | * Checksums and signatures of remote replicas are checked | |
666 | * during opening them on the remote side by the rpmem daemon. | |
667 | * The local version of remote headers does not contain | |
668 | * such data. | |
669 | */ | |
670 | if (rep->remote) | |
671 | continue; | |
672 | ||
673 | for (unsigned p = 0; p < rep->nhdrs; ++p) { | |
674 | ||
675 | /* skip broken parts */ | |
676 | if (replica_is_part_broken(r, p, set_hs)) | |
677 | continue; | |
678 | ||
679 | /* check part's checksum */ | |
680 | LOG(4, "checking checksum for part %u, replica %u", | |
681 | p, r); | |
682 | ||
683 | struct pool_hdr *hdr = HDR(rep, p); | |
684 | ||
685 | if (!util_checksum(hdr, sizeof(*hdr), &hdr->checksum, 0, | |
686 | POOL_HDR_CSUM_END_OFF(hdr))) { | |
687 | ERR("invalid checksum of pool header"); | |
688 | rep_hs->part[p].flags |= IS_BROKEN; | |
689 | } else if (util_is_zeroed(hdr, sizeof(*hdr))) { | |
690 | rep_hs->part[p].flags |= IS_BROKEN; | |
691 | } | |
692 | ||
693 | enum pool_type type = pool_hdr_get_type(hdr); | |
694 | if (type == POOL_TYPE_UNKNOWN) { | |
695 | ERR("invalid signature"); | |
696 | rep_hs->part[p].flags |= IS_BROKEN; | |
697 | } | |
698 | } | |
699 | } | |
700 | return 0; | |
701 | } | |
702 | ||
703 | /* | |
704 | * replica_badblocks_recovery_file_save -- save bad blocks in the bad blocks | |
705 | * recovery file before clearing them | |
706 | */ | |
707 | static int | |
708 | replica_badblocks_recovery_file_save(struct part_health_status *part_hs) | |
709 | { | |
710 | LOG(3, "part_health_status %p", part_hs); | |
711 | ||
712 | ASSERTeq(part_hs->recovery_file_exists, 1); | |
713 | ASSERTne(part_hs->recovery_file_name, NULL); | |
714 | ||
715 | struct badblocks *bbs = &part_hs->bbs; | |
716 | char *path = part_hs->recovery_file_name; | |
717 | int ret = -1; | |
718 | ||
719 | int fd = os_open(path, O_WRONLY | O_TRUNC); | |
720 | if (fd < 0) { | |
721 | ERR("!opening bad block recovery file failed -- '%s'", path); | |
722 | return -1; | |
723 | } | |
724 | ||
725 | FILE *recovery_file_name = os_fdopen(fd, "w"); | |
726 | if (recovery_file_name == NULL) { | |
727 | ERR( | |
728 | "!opening a file stream for bad block recovery file failed -- '%s'", | |
729 | path); | |
730 | os_close(fd); | |
731 | return -1; | |
732 | } | |
733 | ||
734 | /* save bad blocks */ | |
735 | for (unsigned i = 0; i < bbs->bb_cnt; i++) { | |
736 | ASSERT(bbs->bbv[i].length != 0); | |
737 | fprintf(recovery_file_name, "%zu %zu\n", | |
738 | bbs->bbv[i].offset, bbs->bbv[i].length); | |
739 | } | |
740 | ||
741 | if (fflush(recovery_file_name) == EOF) { | |
742 | ERR("!flushing bad block recovery file failed -- '%s'", path); | |
743 | goto exit_error; | |
744 | } | |
745 | ||
746 | if (os_fsync(fd) < 0) { | |
747 | ERR("!syncing bad block recovery file failed -- '%s'", path); | |
748 | goto exit_error; | |
749 | } | |
750 | ||
751 | /* save the finish flag */ | |
752 | fprintf(recovery_file_name, "0 0\n"); | |
753 | ||
754 | if (fflush(recovery_file_name) == EOF) { | |
755 | ERR("!flushing bad block recovery file failed -- '%s'", path); | |
756 | goto exit_error; | |
757 | } | |
758 | ||
759 | if (os_fsync(fd) < 0) { | |
760 | ERR("!syncing bad block recovery file failed -- '%s'", path); | |
761 | goto exit_error; | |
762 | } | |
763 | ||
764 | LOG(3, "bad blocks saved in the recovery file -- '%s'", path); | |
765 | ret = 0; | |
766 | ||
767 | exit_error: | |
768 | os_fclose(recovery_file_name); | |
769 | ||
770 | return ret; | |
771 | } | |
772 | ||
773 | /* | |
774 | * replica_part_badblocks_recovery_file_read -- read bad blocks | |
775 | * from the bad block recovery file | |
776 | * for the current part | |
777 | */ | |
778 | static int | |
779 | replica_part_badblocks_recovery_file_read(struct part_health_status *part_hs) | |
780 | { | |
781 | LOG(3, "part_health_status %p", part_hs); | |
782 | ||
783 | ASSERT(part_hs->recovery_file_exists); | |
784 | ASSERTne(part_hs->recovery_file_name, NULL); | |
785 | ||
786 | VEC(bbsvec, struct bad_block) bbv = VEC_INITIALIZER; | |
787 | char *path = part_hs->recovery_file_name; | |
788 | struct bad_block bb; | |
789 | int ret = -1; | |
790 | ||
791 | FILE *recovery_file = os_fopen(path, "r"); | |
792 | if (!recovery_file) { | |
793 | ERR("!opening the recovery file for reading failed -- '%s'", | |
794 | path); | |
795 | return -1; | |
796 | } | |
797 | ||
798 | unsigned long long min_offset = 0; /* minimum possible offset */ | |
799 | ||
800 | do { | |
801 | if (fscanf(recovery_file, "%zu %zu\n", | |
802 | &bb.offset, &bb.length) < 2) { | |
803 | LOG(1, "incomplete bad block recovery file -- '%s'", | |
804 | path); | |
805 | ret = 1; | |
806 | goto error_exit; | |
807 | } | |
808 | ||
809 | if (bb.offset == 0 && bb.length == 0) { | |
810 | /* finish_flag */ | |
811 | break; | |
812 | } | |
813 | ||
814 | /* check if bad blocks build an increasing sequence */ | |
815 | if (bb.offset < min_offset) { | |
816 | ERR( | |
817 | "wrong format of bad block recovery file (bad blocks are not sorted by the offset in ascending order) -- '%s'", | |
818 | path); | |
819 | errno = EINVAL; | |
820 | ret = -1; | |
821 | goto error_exit; | |
822 | } | |
823 | ||
824 | /* update the minimum possible offset */ | |
825 | min_offset = bb.offset + bb.length; | |
826 | ||
827 | bb.nhealthy = NO_HEALTHY_REPLICA; /* unknown healthy replica */ | |
828 | ||
829 | /* add the new bad block to the vector */ | |
830 | if (VEC_PUSH_BACK(&bbv, bb)) | |
831 | goto error_exit; | |
832 | } while (1); | |
833 | ||
834 | part_hs->bbs.bbv = VEC_ARR(&bbv); | |
835 | part_hs->bbs.bb_cnt = (unsigned)VEC_SIZE(&bbv); | |
836 | ||
837 | os_fclose(recovery_file); | |
838 | ||
839 | LOG(1, "bad blocks read from the recovery file -- '%s'", path); | |
840 | ||
841 | return 0; | |
842 | ||
843 | error_exit: | |
844 | VEC_DELETE(&bbv); | |
845 | os_fclose(recovery_file); | |
846 | return ret; | |
847 | } | |
848 | ||
849 | /* status returned by the replica_badblocks_recovery_files_check() function */ | |
850 | enum badblocks_recovery_files_status { | |
851 | RECOVERY_FILES_ERROR = -1, | |
852 | RECOVERY_FILES_DO_NOT_EXIST = 0, | |
853 | RECOVERY_FILES_EXIST_ALL = 1, | |
854 | RECOVERY_FILES_NOT_ALL_EXIST = 2 | |
855 | }; | |
856 | ||
857 | /* | |
858 | * replica_badblocks_recovery_files_check -- (internal) check if bad blocks | |
859 | * recovery files exist | |
860 | */ | |
861 | static enum badblocks_recovery_files_status | |
862 | replica_badblocks_recovery_files_check(struct pool_set *set, | |
863 | struct poolset_health_status *set_hs) | |
864 | { | |
865 | LOG(3, "set %p, set_hs %p", set, set_hs); | |
866 | ||
867 | int recovery_file_exists = 0; | |
868 | int recovery_file_does_not_exist = 0; | |
869 | ||
870 | for (unsigned r = 0; r < set->nreplicas; ++r) { | |
871 | struct pool_replica *rep = set->replica[r]; | |
872 | struct replica_health_status *rep_hs = set_hs->replica[r]; | |
873 | ||
874 | if (rep->remote) { | |
875 | /* | |
876 | * Bad blocks in remote replicas currently are fixed | |
877 | * during opening by removing and recreating | |
878 | * the whole remote replica. | |
879 | */ | |
880 | continue; | |
881 | } | |
882 | ||
883 | for (unsigned p = 0; p < rep->nparts; ++p) { | |
884 | const char *path = PART(rep, p)->path; | |
885 | struct part_health_status *part_hs = &rep_hs->part[p]; | |
886 | ||
887 | int exists = util_file_exists(path); | |
888 | if (exists < 0) | |
889 | return -1; | |
890 | ||
891 | if (!exists) { | |
892 | /* part file does not exist - skip it */ | |
893 | continue; | |
894 | } | |
895 | ||
896 | part_hs->recovery_file_name = | |
897 | badblocks_recovery_file_alloc(set->path, | |
898 | r, p); | |
899 | if (part_hs->recovery_file_name == NULL) { | |
900 | LOG(1, | |
901 | "allocating name of bad block recovery file failed"); | |
902 | return RECOVERY_FILES_ERROR; | |
903 | } | |
904 | ||
905 | exists = util_file_exists(part_hs->recovery_file_name); | |
906 | if (exists < 0) | |
907 | return -1; | |
908 | ||
909 | part_hs->recovery_file_exists = exists; | |
910 | ||
911 | if (part_hs->recovery_file_exists) { | |
912 | LOG(3, "bad block recovery file exists: %s", | |
913 | part_hs->recovery_file_name); | |
914 | ||
915 | recovery_file_exists = 1; | |
916 | ||
917 | } else { | |
918 | LOG(3, | |
919 | "bad block recovery file does not exist: %s", | |
920 | part_hs->recovery_file_name); | |
921 | ||
922 | recovery_file_does_not_exist = 1; | |
923 | } | |
924 | } | |
925 | } | |
926 | ||
927 | if (recovery_file_exists) { | |
928 | if (recovery_file_does_not_exist) { | |
929 | LOG(4, "return RECOVERY_FILES_NOT_ALL_EXIST"); | |
930 | return RECOVERY_FILES_NOT_ALL_EXIST; | |
931 | } else { | |
932 | LOG(4, "return RECOVERY_FILES_EXIST_ALL"); | |
933 | return RECOVERY_FILES_EXIST_ALL; | |
934 | } | |
935 | } | |
936 | ||
937 | LOG(4, "return RECOVERY_FILES_DO_NOT_EXIST"); | |
938 | return RECOVERY_FILES_DO_NOT_EXIST; | |
939 | } | |
940 | ||
941 | /* | |
942 | * replica_badblocks_recovery_files_read -- (internal) read bad blocks from all | |
943 | * bad block recovery files for all parts | |
944 | */ | |
945 | static int | |
946 | replica_badblocks_recovery_files_read(struct pool_set *set, | |
947 | struct poolset_health_status *set_hs) | |
948 | { | |
949 | LOG(3, "set %p, set_hs %p", set, set_hs); | |
950 | ||
951 | int ret; | |
952 | ||
953 | for (unsigned r = 0; r < set->nreplicas; ++r) { | |
954 | struct pool_replica *rep = set->replica[r]; | |
955 | struct replica_health_status *rep_hs = set_hs->replica[r]; | |
956 | ||
957 | /* XXX: not supported yet */ | |
958 | if (rep->remote) | |
959 | continue; | |
960 | ||
961 | for (unsigned p = 0; p < rep->nparts; ++p) { | |
962 | const char *path = PART(rep, p)->path; | |
963 | struct part_health_status *part_hs = &rep_hs->part[p]; | |
964 | ||
965 | int exists = util_file_exists(path); | |
966 | if (exists < 0) | |
967 | return -1; | |
968 | ||
969 | if (!exists) { | |
970 | /* the part does not exist */ | |
971 | continue; | |
972 | } | |
973 | ||
974 | LOG(1, | |
975 | "reading bad blocks from the recovery file -- '%s'", | |
976 | part_hs->recovery_file_name); | |
977 | ||
978 | ret = replica_part_badblocks_recovery_file_read( | |
979 | part_hs); | |
980 | if (ret < 0) { | |
981 | LOG(1, | |
982 | "reading bad blocks from the recovery file failed -- '%s'", | |
983 | part_hs->recovery_file_name); | |
984 | return -1; | |
985 | } | |
986 | ||
987 | if (ret > 0) { | |
988 | LOG(1, | |
989 | "incomplete bad block recovery file detected -- '%s'", | |
990 | part_hs->recovery_file_name); | |
991 | return 1; | |
992 | } | |
993 | ||
994 | if (part_hs->bbs.bb_cnt) { | |
995 | LOG(3, "part %u contains %u bad blocks -- '%s'", | |
996 | p, part_hs->bbs.bb_cnt, path); | |
997 | } | |
998 | } | |
999 | } | |
1000 | ||
1001 | return 0; | |
1002 | } | |
1003 | ||
1004 | /* | |
1005 | * replica_badblocks_recovery_files_create_empty -- (internal) create one empty | |
1006 | * bad block recovery file | |
1007 | * for each part file | |
1008 | */ | |
1009 | static int | |
1010 | replica_badblocks_recovery_files_create_empty(struct pool_set *set, | |
1011 | struct poolset_health_status *set_hs) | |
1012 | { | |
1013 | LOG(3, "set %p, set_hs %p", set, set_hs); | |
1014 | ||
1015 | struct part_health_status *part_hs; | |
1016 | const char *path; | |
1017 | int fd; | |
1018 | ||
1019 | for (unsigned r = 0; r < set->nreplicas; ++r) { | |
1020 | struct pool_replica *rep = set->replica[r]; | |
1021 | struct replica_health_status *rep_hs = set_hs->replica[r]; | |
1022 | ||
1023 | /* XXX: not supported yet */ | |
1024 | if (rep->remote) | |
1025 | continue; | |
1026 | ||
1027 | for (unsigned p = 0; p < rep->nparts; ++p) { | |
1028 | part_hs = &rep_hs->part[p]; | |
1029 | path = PART(rep, p)->path; | |
1030 | ||
1031 | if (!part_hs->recovery_file_name) | |
1032 | continue; | |
1033 | ||
1034 | fd = os_open(part_hs->recovery_file_name, | |
1035 | O_RDWR | O_CREAT | O_EXCL, | |
1036 | 0600); | |
1037 | if (fd < 0) { | |
1038 | ERR( | |
1039 | "!creating an empty bad block recovery file failed -- '%s' (part file '%s')", | |
1040 | part_hs->recovery_file_name, path); | |
1041 | return -1; | |
1042 | } | |
1043 | ||
1044 | os_close(fd); | |
1045 | ||
1046 | char *file_name = Strdup(part_hs->recovery_file_name); | |
1047 | if (file_name == NULL) { | |
1048 | ERR("!Strdup"); | |
1049 | return -1; | |
1050 | } | |
1051 | ||
1052 | char *dir_name = dirname(file_name); | |
1053 | ||
1054 | /* fsync the file's directory */ | |
1055 | if (os_fsync_dir(dir_name) < 0) { | |
1056 | ERR( | |
1057 | "!syncing the directory of the bad block recovery file failed -- '%s' (part file '%s')", | |
1058 | dir_name, path); | |
1059 | Free(file_name); | |
1060 | return -1; | |
1061 | } | |
1062 | ||
1063 | Free(file_name); | |
1064 | ||
1065 | part_hs->recovery_file_exists = 1; | |
1066 | } | |
1067 | } | |
1068 | ||
1069 | return 0; | |
1070 | } | |
1071 | ||
1072 | /* | |
1073 | * replica_badblocks_recovery_files_save -- (internal) save bad blocks | |
1074 | * in the bad block recovery files | |
1075 | */ | |
1076 | static int | |
1077 | replica_badblocks_recovery_files_save(struct pool_set *set, | |
1078 | struct poolset_health_status *set_hs) | |
1079 | { | |
1080 | LOG(3, "set %p, set_hs %p", set, set_hs); | |
1081 | ||
1082 | for (unsigned r = 0; r < set->nreplicas; ++r) { | |
1083 | struct pool_replica *rep = set->replica[r]; | |
1084 | struct replica_health_status *rep_hs = set_hs->replica[r]; | |
1085 | ||
1086 | /* XXX: not supported yet */ | |
1087 | if (rep->remote) | |
1088 | continue; | |
1089 | ||
1090 | for (unsigned p = 0; p < rep->nparts; ++p) { | |
1091 | struct part_health_status *part_hs = &rep_hs->part[p]; | |
1092 | ||
1093 | if (!part_hs->recovery_file_name) | |
1094 | continue; | |
1095 | ||
1096 | int ret = replica_badblocks_recovery_file_save(part_hs); | |
1097 | if (ret < 0) { | |
1098 | LOG(1, | |
1099 | "opening bad block recovery file failed -- '%s'", | |
1100 | part_hs->recovery_file_name); | |
1101 | return -1; | |
1102 | } | |
1103 | } | |
1104 | } | |
1105 | ||
1106 | return 0; | |
1107 | } | |
1108 | ||
1109 | /* | |
1110 | * replica_badblocks_get -- (internal) get all bad blocks and save them | |
1111 | * in part_hs->bbs structures. | |
1112 | * Returns 1 if any bad block was found, 0 otherwise. | |
1113 | */ | |
1114 | static int | |
1115 | replica_badblocks_get(struct pool_set *set, | |
1116 | struct poolset_health_status *set_hs) | |
1117 | { | |
1118 | LOG(3, "set %p, set_hs %p", set, set_hs); | |
1119 | ||
1120 | int bad_blocks_found = 0; | |
1121 | ||
1122 | for (unsigned r = 0; r < set->nreplicas; ++r) { | |
1123 | struct pool_replica *rep = set->replica[r]; | |
1124 | struct replica_health_status *rep_hs = set_hs->replica[r]; | |
1125 | ||
1126 | /* XXX: not supported yet */ | |
1127 | if (rep->remote) | |
1128 | continue; | |
1129 | ||
1130 | for (unsigned p = 0; p < rep->nparts; ++p) { | |
1131 | const char *path = PART(rep, p)->path; | |
1132 | struct part_health_status *part_hs = &rep_hs->part[p]; | |
1133 | ||
1134 | int exists = util_file_exists(path); | |
1135 | if (exists < 0) | |
1136 | return -1; | |
1137 | ||
1138 | if (!exists) | |
1139 | continue; | |
1140 | ||
1141 | int ret = badblocks_get(path, &part_hs->bbs); | |
1142 | if (ret < 0) { | |
1143 | ERR( | |
1144 | "!checking the pool part for bad blocks failed -- '%s'", | |
1145 | path); | |
1146 | return -1; | |
1147 | } | |
1148 | ||
1149 | if (part_hs->bbs.bb_cnt) { | |
1150 | LOG(3, "part %u contains %u bad blocks -- '%s'", | |
1151 | p, part_hs->bbs.bb_cnt, path); | |
1152 | ||
1153 | bad_blocks_found = 1; | |
1154 | } | |
1155 | } | |
1156 | } | |
1157 | ||
1158 | return bad_blocks_found; | |
1159 | } | |
1160 | ||
1161 | /* | |
1162 | * check_badblocks_in_header -- (internal) check if bad blocks corrupted | |
1163 | * the header | |
1164 | */ | |
1165 | static int | |
1166 | check_badblocks_in_header(struct badblocks *bbs) | |
1167 | { | |
1168 | for (unsigned b = 0; b < bbs->bb_cnt; b++) | |
1169 | if (bbs->bbv[b].offset < POOL_HDR_SIZE) | |
1170 | return 1; | |
1171 | ||
1172 | return 0; | |
1173 | } | |
1174 | ||
1175 | /* | |
1176 | * replica_badblocks_clear -- (internal) clear all bad blocks | |
1177 | */ | |
1178 | static int | |
1179 | replica_badblocks_clear(struct pool_set *set, | |
1180 | struct poolset_health_status *set_hs) | |
1181 | { | |
1182 | LOG(3, "set %p, set_hs %p", set, set_hs); | |
1183 | ||
1184 | int ret; | |
1185 | ||
1186 | for (unsigned r = 0; r < set->nreplicas; ++r) { | |
1187 | struct pool_replica *rep = set->replica[r]; | |
1188 | struct replica_health_status *rep_hs = set_hs->replica[r]; | |
1189 | ||
1190 | /* XXX: not supported yet */ | |
1191 | if (rep->remote) | |
1192 | continue; | |
1193 | ||
1194 | for (unsigned p = 0; p < rep->nparts; ++p) { | |
1195 | const char *path = PART(rep, p)->path; | |
1196 | struct part_health_status *part_hs = &rep_hs->part[p]; | |
1197 | ||
1198 | int exists = util_file_exists(path); | |
1199 | if (exists < 0) | |
1200 | return -1; | |
1201 | ||
1202 | if (!exists) { | |
1203 | /* the part does not exist */ | |
1204 | continue; | |
1205 | } | |
1206 | ||
1207 | if (part_hs->bbs.bb_cnt == 0) { | |
1208 | /* no bad blocks found */ | |
1209 | continue; | |
1210 | } | |
1211 | ||
1212 | /* bad blocks were found */ | |
1213 | part_hs->flags |= HAS_BAD_BLOCKS; | |
1214 | rep_hs->flags |= HAS_BAD_BLOCKS; | |
1215 | ||
1216 | if (check_badblocks_in_header(&part_hs->bbs)) { | |
1217 | part_hs->flags |= HAS_CORRUPTED_HEADER; | |
1218 | if (p == 0) | |
1219 | rep_hs->flags |= HAS_CORRUPTED_HEADER; | |
1220 | } | |
1221 | ||
1222 | ret = badblocks_clear(path, &part_hs->bbs); | |
1223 | if (ret < 0) { | |
1224 | LOG(1, | |
1225 | "clearing bad blocks in replica failed -- '%s'", | |
1226 | path); | |
1227 | return -1; | |
1228 | } | |
1229 | } | |
1230 | } | |
1231 | ||
1232 | return 0; | |
1233 | } | |
1234 | ||
1235 | /* | |
1236 | * replica_badblocks_check_or_clear -- (internal) check if replica contains | |
1237 | * bad blocks when in dry run | |
1238 | * or clear them otherwise | |
1239 | */ | |
1240 | static int | |
1241 | replica_badblocks_check_or_clear(struct pool_set *set, | |
1242 | struct poolset_health_status *set_hs, | |
1243 | int dry_run, int called_from_sync, | |
1244 | int check_bad_blocks, int fix_bad_blocks) | |
1245 | { | |
1246 | LOG(3, | |
1247 | "set %p, set_hs %p, dry_run %i, called_from_sync %i, " | |
1248 | "check_bad_blocks %i, fix_bad_blocks %i", | |
1249 | set, set_hs, dry_run, called_from_sync, | |
1250 | check_bad_blocks, fix_bad_blocks); | |
1251 | ||
1252 | #define ERR_MSG_BB \ | |
1253 | " please read the manual first and use this option\n"\ | |
1254 | " ONLY IF you are sure that you know what you are doing" | |
1255 | ||
1256 | enum badblocks_recovery_files_status status; | |
1257 | int ret; | |
1258 | ||
1259 | /* check all bad block recovery files */ | |
1260 | status = replica_badblocks_recovery_files_check(set, set_hs); | |
1261 | ||
1262 | /* phase #1 - error handling */ | |
1263 | switch (status) { | |
1264 | case RECOVERY_FILES_ERROR: | |
1265 | LOG(1, "checking bad block recovery files failed"); | |
1266 | return -1; | |
1267 | ||
1268 | case RECOVERY_FILES_EXIST_ALL: | |
1269 | case RECOVERY_FILES_NOT_ALL_EXIST: | |
1270 | if (!called_from_sync) { | |
1271 | ERR( | |
1272 | "error: a bad block recovery file exists, run 'pmempool sync --bad-blocks' to fix bad blocks first"); | |
1273 | return -1; | |
1274 | } | |
1275 | ||
1276 | if (!fix_bad_blocks) { | |
1277 | ERR( | |
1278 | "error: a bad block recovery file exists, but the '--bad-blocks' option is not set\n" | |
1279 | ERR_MSG_BB); | |
1280 | return -1; | |
1281 | } | |
1282 | break; | |
1283 | ||
1284 | default: | |
1285 | break; | |
1286 | }; | |
1287 | ||
1288 | /* | |
1289 | * The pool is checked for bad blocks only if: | |
1290 | * 1) compat feature POOL_FEAT_CHECK_BAD_BLOCKS is set | |
1291 | * OR: | |
1292 | * 2) the '--bad-blocks' option is set | |
1293 | * | |
1294 | * Bad blocks are cleared and fixed only if: | |
1295 | * - the '--bad-blocks' option is set | |
1296 | */ | |
1297 | if (!fix_bad_blocks && !check_bad_blocks) { | |
1298 | LOG(3, "skipping bad blocks checking"); | |
1299 | return 0; | |
1300 | } | |
1301 | ||
1302 | /* phase #2 - reading recovery files */ | |
1303 | switch (status) { | |
1304 | case RECOVERY_FILES_EXIST_ALL: | |
1305 | /* read all bad block recovery files */ | |
1306 | ret = replica_badblocks_recovery_files_read(set, set_hs); | |
1307 | if (ret < 0) { | |
1308 | LOG(1, "checking bad block recovery files failed"); | |
1309 | return -1; | |
1310 | } | |
1311 | ||
1312 | if (ret > 0) { | |
1313 | /* incomplete bad block recovery file was detected */ | |
1314 | ||
1315 | LOG(1, | |
1316 | "warning: incomplete bad block recovery file detected\n" | |
1317 | " - all recovery files will be removed"); | |
1318 | ||
1319 | /* changing status to RECOVERY_FILES_NOT_ALL_EXIST */ | |
1320 | status = RECOVERY_FILES_NOT_ALL_EXIST; | |
1321 | } | |
1322 | break; | |
1323 | ||
1324 | case RECOVERY_FILES_NOT_ALL_EXIST: | |
1325 | LOG(1, | |
1326 | "warning: one of bad block recovery files does not exist\n" | |
1327 | " - all recovery files will be removed"); | |
1328 | break; | |
1329 | ||
1330 | default: | |
1331 | break; | |
1332 | }; | |
1333 | ||
1334 | if (status == RECOVERY_FILES_NOT_ALL_EXIST) { | |
1335 | /* | |
1336 | * At least one of bad block recovery files does not exist, | |
1337 | * or an incomplete bad block recovery file was detected, | |
1338 | * so all recovery files have to be removed. | |
1339 | */ | |
1340 | ||
1341 | if (!dry_run) { | |
1342 | LOG(1, "removing all bad block recovery files..."); | |
1343 | ret = replica_remove_all_recovery_files(set_hs); | |
1344 | if (ret < 0) { | |
1345 | LOG(1, | |
1346 | "removing bad block recovery files failed"); | |
1347 | return -1; | |
1348 | } | |
1349 | } else { | |
1350 | LOG(1, "all bad block recovery files would be removed"); | |
1351 | } | |
1352 | ||
1353 | /* changing status to RECOVERY_FILES_DO_NOT_EXIST */ | |
1354 | status = RECOVERY_FILES_DO_NOT_EXIST; | |
1355 | } | |
1356 | ||
1357 | if (status == RECOVERY_FILES_DO_NOT_EXIST) { | |
1358 | /* | |
1359 | * There are no bad block recovery files, | |
1360 | * so let's check bad blocks. | |
1361 | */ | |
1362 | ||
1363 | int bad_blocks_found = replica_badblocks_get(set, set_hs); | |
1364 | if (bad_blocks_found < 0) { | |
1365 | if (errno == ENOTSUP) { | |
1366 | LOG(1, BB_NOT_SUPP); | |
1367 | return -1; | |
1368 | } | |
1369 | ||
1370 | LOG(1, "checking bad blocks failed"); | |
1371 | return -1; | |
1372 | } | |
1373 | ||
1374 | if (!bad_blocks_found) { | |
1375 | LOG(4, "no bad blocks found"); | |
1376 | return 0; | |
1377 | } | |
1378 | ||
1379 | /* bad blocks were found */ | |
1380 | ||
1381 | if (!called_from_sync) { | |
1382 | ERR( | |
1383 | "error: bad blocks found, run 'pmempool sync --bad-blocks' to fix bad blocks first"); | |
1384 | return -1; | |
1385 | } | |
1386 | ||
1387 | if (!fix_bad_blocks) { | |
1388 | ERR( | |
1389 | "error: bad blocks found, but the '--bad-blocks' option is not set\n" | |
1390 | ERR_MSG_BB); | |
1391 | return -1; | |
1392 | } | |
1393 | ||
1394 | if (dry_run) { | |
1395 | /* dry-run - do nothing */ | |
1396 | LOG(1, "warning: bad blocks were found"); | |
1397 | return 0; | |
1398 | } | |
1399 | ||
1400 | /* create one empty recovery file for each part file */ | |
1401 | ret = replica_badblocks_recovery_files_create_empty(set, | |
1402 | set_hs); | |
1403 | if (ret < 0) { | |
1404 | LOG(1, | |
1405 | "creating empty bad block recovery files failed"); | |
1406 | return -1; | |
1407 | } | |
1408 | ||
1409 | /* save bad blocks in recovery files */ | |
1410 | ret = replica_badblocks_recovery_files_save(set, set_hs); | |
1411 | if (ret < 0) { | |
1412 | LOG(1, "saving bad block recovery files failed"); | |
1413 | return -1; | |
1414 | } | |
1415 | } | |
1416 | ||
1417 | if (dry_run) { | |
1418 | /* dry-run - do nothing */ | |
1419 | LOG(1, "bad blocks would be cleared"); | |
1420 | return 0; | |
1421 | } | |
1422 | ||
1423 | ret = replica_badblocks_clear(set, set_hs); | |
1424 | if (ret < 0) { | |
1425 | ERR("clearing bad blocks failed"); | |
1426 | return -1; | |
1427 | } | |
1428 | ||
1429 | return 0; | |
1430 | } | |
1431 | ||
1432 | /* | |
1433 | * check_shutdown_state -- (internal) check if replica has | |
1434 | * healthy shutdown_state | |
1435 | */ | |
1436 | static int | |
1437 | check_shutdown_state(struct pool_set *set, | |
1438 | struct poolset_health_status *set_hs) | |
1439 | { | |
1440 | LOG(3, "set %p, set_hs %p", set, set_hs); | |
1441 | for (unsigned r = 0; r < set->nreplicas; ++r) {\ | |
1442 | struct pool_replica *rep = set->replica[r]; | |
1443 | struct replica_health_status *rep_hs = set_hs->replica[r]; | |
1444 | struct pool_hdr *hdrp = HDR(rep, 0); | |
1445 | ||
1446 | if (rep->remote) | |
1447 | continue; | |
1448 | ||
1449 | if (hdrp == NULL) { | |
1450 | /* cannot verify shutdown state */ | |
1451 | rep_hs->flags |= IS_BROKEN; | |
1452 | continue; | |
1453 | } | |
1454 | ||
1455 | struct shutdown_state curr_sds; | |
1456 | shutdown_state_init(&curr_sds, NULL); | |
1457 | for (unsigned p = 0; p < rep->nparts; ++p) { | |
1458 | if (PART(rep, p)->fd < 0) | |
1459 | continue; | |
1460 | ||
1461 | if (shutdown_state_add_part(&curr_sds, | |
1462 | PART(rep, p)->fd, NULL)) { | |
1463 | rep_hs->flags |= IS_BROKEN; | |
1464 | break; | |
1465 | } | |
1466 | } | |
1467 | ||
1468 | if (rep_hs->flags & IS_BROKEN) | |
1469 | continue; | |
1470 | ||
1471 | /* make a copy of sds as we shouldn't modify a pool */ | |
1472 | struct shutdown_state pool_sds = hdrp->sds; | |
1473 | ||
1474 | if (shutdown_state_check(&curr_sds, &pool_sds, NULL)) | |
1475 | rep_hs->flags |= IS_BROKEN; | |
1476 | ||
1477 | } | |
1478 | return 0; | |
1479 | } | |
1480 | ||
1481 | /* | |
1482 | * check_uuids_between_parts -- (internal) check if uuids between adjacent | |
1483 | * parts are consistent for a given replica | |
1484 | */ | |
1485 | static int | |
1486 | check_uuids_between_parts(struct pool_set *set, unsigned repn, | |
1487 | struct poolset_health_status *set_hs) | |
1488 | { | |
1489 | LOG(3, "set %p, repn %u, set_hs %p", set, repn, set_hs); | |
1490 | struct pool_replica *rep = REP(set, repn); | |
1491 | ||
1492 | /* check poolset_uuid consistency between replica's parts */ | |
1493 | LOG(4, "checking consistency of poolset uuid in replica %u", repn); | |
1494 | uuid_t poolset_uuid; | |
1495 | int uuid_stored = 0; | |
1496 | unsigned part_stored = UNDEF_PART; | |
1497 | for (unsigned p = 0; p < rep->nhdrs; ++p) { | |
1498 | /* skip broken parts */ | |
1499 | if (replica_is_part_broken(repn, p, set_hs)) | |
1500 | continue; | |
1501 | ||
1502 | if (!uuid_stored) { | |
1503 | memcpy(poolset_uuid, HDR(rep, p)->poolset_uuid, | |
1504 | POOL_HDR_UUID_LEN); | |
1505 | uuid_stored = 1; | |
1506 | part_stored = p; | |
1507 | continue; | |
1508 | } | |
1509 | ||
1510 | if (uuidcmp(HDR(rep, p)->poolset_uuid, poolset_uuid)) { | |
1511 | ERR( | |
1512 | "different poolset uuids in parts from the same replica (repn %u, parts %u and %u) - cannot synchronize", | |
1513 | repn, part_stored, p); | |
1514 | errno = EINVAL; | |
1515 | return -1; | |
1516 | } | |
1517 | } | |
1518 | ||
1519 | /* check if all uuids for adjacent replicas are the same across parts */ | |
1520 | LOG(4, "checking consistency of adjacent replicas' uuids in replica %u", | |
1521 | repn); | |
1522 | unsigned unbroken_p = UNDEF_PART; | |
1523 | for (unsigned p = 0; p < rep->nhdrs; ++p) { | |
1524 | /* skip broken parts */ | |
1525 | if (replica_is_part_broken(repn, p, set_hs)) | |
1526 | continue; | |
1527 | ||
1528 | if (unbroken_p == UNDEF_PART) { | |
1529 | unbroken_p = p; | |
1530 | continue; | |
1531 | } | |
1532 | ||
1533 | struct pool_hdr *hdrp = HDR(rep, p); | |
1534 | int prev_differ = uuidcmp(HDR(rep, unbroken_p)->prev_repl_uuid, | |
1535 | hdrp->prev_repl_uuid); | |
1536 | int next_differ = uuidcmp(HDR(rep, unbroken_p)->next_repl_uuid, | |
1537 | hdrp->next_repl_uuid); | |
1538 | ||
1539 | if (prev_differ || next_differ) { | |
1540 | ERR( | |
1541 | "different adjacent replica UUID between parts (repn %u, parts %u and %u) - cannot synchronize", | |
1542 | repn, unbroken_p, p); | |
1543 | errno = EINVAL; | |
1544 | return -1; | |
1545 | } | |
1546 | } | |
1547 | ||
1548 | /* check parts linkage */ | |
1549 | LOG(4, "checking parts linkage in replica %u", repn); | |
1550 | for (unsigned p = 0; p < rep->nhdrs; ++p) { | |
1551 | /* skip broken parts */ | |
1552 | if (replica_is_part_broken(repn, p, set_hs)) | |
1553 | continue; | |
1554 | ||
1555 | struct pool_hdr *hdrp = HDR(rep, p); | |
1556 | struct pool_hdr *next_hdrp = HDRN(rep, p); | |
1557 | int next_is_broken = replica_is_part_broken(repn, p + 1, | |
1558 | set_hs); | |
1559 | ||
1560 | if (!next_is_broken) { | |
1561 | int next_decoupled = | |
1562 | uuidcmp(next_hdrp->prev_part_uuid, | |
1563 | hdrp->uuid) || | |
1564 | uuidcmp(hdrp->next_part_uuid, next_hdrp->uuid); | |
1565 | if (next_decoupled) { | |
1566 | ERR( | |
1567 | "two consecutive unbroken parts are not linked to each other (repn %u, parts %u and %u) - cannot synchronize", | |
1568 | repn, p, p + 1); | |
1569 | errno = EINVAL; | |
1570 | return -1; | |
1571 | } | |
1572 | } | |
1573 | } | |
1574 | return 0; | |
1575 | } | |
1576 | ||
1577 | /* | |
1578 | * check_replicas_consistency -- (internal) check if all uuids within each | |
1579 | * replica are consistent | |
1580 | */ | |
1581 | static int | |
1582 | check_replicas_consistency(struct pool_set *set, | |
1583 | struct poolset_health_status *set_hs) | |
1584 | { | |
1585 | LOG(3, "set %p, set_hs %p", set, set_hs); | |
1586 | for (unsigned r = 0; r < set->nreplicas; ++r) { | |
1587 | if (check_uuids_between_parts(set, r, set_hs)) | |
1588 | return -1; | |
1589 | } | |
1590 | return 0; | |
1591 | } | |
1592 | ||
1593 | /* | |
1594 | * check_replica_options -- (internal) check if options are consistent in the | |
1595 | * replica | |
1596 | */ | |
1597 | static int | |
1598 | check_replica_options(struct pool_set *set, unsigned repn, | |
1599 | struct poolset_health_status *set_hs) | |
1600 | { | |
1601 | LOG(3, "set %p, repn %u, set_hs %p", set, repn, set_hs); | |
1602 | struct pool_replica *rep = REP(set, repn); | |
1603 | struct replica_health_status *rep_hs = REP_HEALTH(set_hs, repn); | |
1604 | for (unsigned p = 0; p < rep->nhdrs; ++p) { | |
1605 | /* skip broken parts */ | |
1606 | if (replica_is_part_broken(repn, p, set_hs)) | |
1607 | continue; | |
1608 | ||
1609 | struct pool_hdr *hdr = HDR(rep, p); | |
1610 | if (((hdr->features.incompat & POOL_FEAT_SINGLEHDR) == 0) != | |
1611 | ((set->options & OPTION_SINGLEHDR) == 0)) { | |
1612 | LOG(1, | |
1613 | "improper options are set in part %u's header in replica %u", | |
1614 | p, repn); | |
1615 | rep_hs->part[p].flags |= IS_BROKEN; | |
1616 | } | |
1617 | } | |
1618 | return 0; | |
1619 | } | |
1620 | ||
1621 | /* | |
1622 | * check_options -- (internal) check if options are consistent in all replicas | |
1623 | */ | |
1624 | static int | |
1625 | check_options(struct pool_set *set, struct poolset_health_status *set_hs) | |
1626 | { | |
1627 | LOG(3, "set %p, set_hs %p", set, set_hs); | |
1628 | for (unsigned r = 0; r < set->nreplicas; ++r) { | |
1629 | if (check_replica_options(set, r, set_hs)) | |
1630 | return -1; | |
1631 | } | |
1632 | return 0; | |
1633 | } | |
1634 | ||
1635 | /* | |
1636 | * check_replica_poolset_uuids - (internal) check if poolset_uuid fields are | |
1637 | * consistent among all parts of a replica; | |
1638 | * the replica is initially considered as | |
1639 | * consistent | |
1640 | */ | |
1641 | static int | |
1642 | check_replica_poolset_uuids(struct pool_set *set, unsigned repn, | |
1643 | uuid_t poolset_uuid, struct poolset_health_status *set_hs) | |
1644 | { | |
1645 | LOG(3, "set %p, repn %u, poolset_uuid %p, set_hs %p", set, repn, | |
1646 | poolset_uuid, set_hs); | |
1647 | struct pool_replica *rep = REP(set, repn); | |
1648 | for (unsigned p = 0; p < rep->nhdrs; ++p) { | |
1649 | /* skip broken parts */ | |
1650 | if (replica_is_part_broken(repn, p, set_hs)) | |
1651 | continue; | |
1652 | ||
1653 | if (uuidcmp(HDR(rep, p)->poolset_uuid, poolset_uuid)) { | |
1654 | /* | |
1655 | * two internally consistent replicas have | |
1656 | * different poolset_uuid | |
1657 | */ | |
1658 | return -1; | |
1659 | } else { | |
1660 | /* | |
1661 | * it is sufficient to check only one part | |
1662 | * from internally consistent replica | |
1663 | */ | |
1664 | break; | |
1665 | } | |
1666 | } | |
1667 | return 0; | |
1668 | } | |
1669 | ||
1670 | /* | |
1671 | * check_poolset_uuids -- (internal) check if poolset_uuid fields are consistent | |
1672 | * among all internally consistent replicas | |
1673 | */ | |
1674 | static int | |
1675 | check_poolset_uuids(struct pool_set *set, | |
1676 | struct poolset_health_status *set_hs) | |
1677 | { | |
1678 | LOG(3, "set %p, set_hs %p", set, set_hs); | |
1679 | ||
1680 | /* find a replica with healthy header */ | |
1681 | unsigned r_h = replica_find_replica_healthy_header(set_hs); | |
1682 | if (r_h == UNDEF_REPLICA) { | |
1683 | ERR("no healthy replica found"); | |
1684 | return -1; | |
1685 | } | |
1686 | ||
1687 | uuid_t poolset_uuid; | |
1688 | memcpy(poolset_uuid, HDR(REP(set, r_h), 0)->poolset_uuid, | |
1689 | POOL_HDR_UUID_LEN); | |
1690 | ||
1691 | for (unsigned r = 0; r < set->nreplicas; ++r) { | |
1692 | /* skip inconsistent replicas */ | |
1693 | if (!replica_is_replica_consistent(r, set_hs) || r == r_h) | |
1694 | continue; | |
1695 | ||
1696 | if (check_replica_poolset_uuids(set, r, poolset_uuid, set_hs)) { | |
1697 | ERR( | |
1698 | "inconsistent poolset uuids between replicas %u and %u - cannot synchronize", | |
1699 | r_h, r); | |
1700 | return -1; | |
1701 | } | |
1702 | } | |
1703 | return 0; | |
1704 | } | |
1705 | ||
1706 | /* | |
1707 | * get_replica_uuid -- (internal) get replica uuid | |
1708 | */ | |
1709 | static int | |
1710 | get_replica_uuid(struct pool_replica *rep, unsigned repn, | |
1711 | struct poolset_health_status *set_hs, uuid_t **uuidpp) | |
1712 | { | |
1713 | unsigned nhdrs = rep->nhdrs; | |
1714 | if (!replica_is_part_broken(repn, 0, set_hs)) { | |
1715 | /* the first part is not broken */ | |
1716 | *uuidpp = &HDR(rep, 0)->uuid; | |
1717 | return 0; | |
1718 | } else if (nhdrs > 1 && !replica_is_part_broken(repn, 1, set_hs)) { | |
1719 | /* the second part is not broken */ | |
1720 | *uuidpp = &HDR(rep, 1)->prev_part_uuid; | |
1721 | return 0; | |
1722 | } else if (nhdrs > 1 && | |
1723 | !replica_is_part_broken(repn, nhdrs - 1, set_hs)) { | |
1724 | /* the last part is not broken */ | |
1725 | *uuidpp = &HDR(rep, nhdrs - 1)->next_part_uuid; | |
1726 | return 0; | |
1727 | } else { | |
1728 | /* cannot get replica uuid */ | |
1729 | return -1; | |
1730 | } | |
1731 | } | |
1732 | ||
1733 | /* | |
1734 | * check_uuids_between_replicas -- (internal) check if uuids between internally | |
1735 | * consistent adjacent replicas are consistent | |
1736 | */ | |
1737 | static int | |
1738 | check_uuids_between_replicas(struct pool_set *set, | |
1739 | struct poolset_health_status *set_hs) | |
1740 | { | |
1741 | LOG(3, "set %p, set_hs %p", set, set_hs); | |
1742 | for (unsigned r = 0; r < set->nreplicas; ++r) { | |
1743 | /* skip comparing inconsistent pairs of replicas */ | |
1744 | if (!replica_is_replica_consistent(r, set_hs) || | |
1745 | !replica_is_replica_consistent(r + 1, set_hs)) | |
1746 | continue; | |
1747 | ||
1748 | struct pool_replica *rep = REP(set, r); | |
1749 | struct pool_replica *rep_n = REPN(set, r); | |
1750 | ||
1751 | /* get uuids of the two adjacent replicas */ | |
1752 | uuid_t *rep_uuidp = NULL; | |
1753 | uuid_t *rep_n_uuidp = NULL; | |
1754 | unsigned r_n = REPN_HEALTHidx(set_hs, r); | |
1755 | if (get_replica_uuid(rep, r, set_hs, &rep_uuidp)) | |
1756 | LOG(2, "cannot get replica uuid, replica %u", r); | |
1757 | if (get_replica_uuid(rep_n, r_n, set_hs, &rep_n_uuidp)) | |
1758 | LOG(2, "cannot get replica uuid, replica %u", r_n); | |
1759 | ||
1760 | /* | |
1761 | * check if replica uuids are consistent between two adjacent | |
1762 | * replicas | |
1763 | */ | |
1764 | unsigned p = replica_find_unbroken_part(r, set_hs); | |
1765 | unsigned p_n = replica_find_unbroken_part(r_n, set_hs); | |
1766 | if (p_n != UNDEF_PART && rep_uuidp != NULL && | |
1767 | uuidcmp(*rep_uuidp, | |
1768 | HDR(rep_n, p_n)->prev_repl_uuid)) { | |
1769 | ERR( | |
1770 | "inconsistent replica uuids between replicas %u and %u", | |
1771 | r, r_n); | |
1772 | return -1; | |
1773 | } | |
1774 | if (p != UNDEF_PART && rep_n_uuidp != NULL && | |
1775 | uuidcmp(*rep_n_uuidp, | |
1776 | HDR(rep, p)->next_repl_uuid)) { | |
1777 | ERR( | |
1778 | "inconsistent replica uuids between replicas %u and %u", | |
1779 | r, r_n); | |
1780 | return -1; | |
1781 | } | |
1782 | ||
1783 | /* | |
1784 | * check if replica uuids on borders of a broken replica are | |
1785 | * consistent | |
1786 | */ | |
1787 | unsigned r_nn = REPN_HEALTHidx(set_hs, r_n); | |
1788 | if (set->nreplicas > 1 && p != UNDEF_PART && | |
1789 | replica_is_replica_broken(r_n, set_hs) && | |
1790 | replica_is_replica_consistent(r_nn, set_hs)) { | |
1791 | unsigned p_nn = | |
1792 | replica_find_unbroken_part(r_nn, set_hs); | |
1793 | if (p_nn == UNDEF_PART) { | |
1794 | LOG(2, | |
1795 | "cannot compare uuids on borders of replica %u", | |
1796 | r); | |
1797 | continue; | |
1798 | } | |
1799 | struct pool_replica *rep_nn = REP(set, r_nn); | |
1800 | if (uuidcmp(HDR(rep, p)->next_repl_uuid, | |
1801 | HDR(rep_nn, p_nn)->prev_repl_uuid)) { | |
1802 | ERR( | |
1803 | "inconsistent replica uuids on borders of replica %u", | |
1804 | r); | |
1805 | return -1; | |
1806 | } | |
1807 | } | |
1808 | } | |
1809 | return 0; | |
1810 | } | |
1811 | ||
1812 | /* | |
1813 | * check_replica_cycles -- (internal) check if healthy replicas form cycles | |
1814 | * shorter than the number of all replicas | |
1815 | */ | |
1816 | static int | |
1817 | check_replica_cycles(struct pool_set *set, | |
1818 | struct poolset_health_status *set_hs) | |
1819 | { | |
1820 | LOG(3, "set %p, set_hs %p", set, set_hs); | |
1821 | unsigned first_healthy; | |
1822 | unsigned count_healthy = 0; | |
1823 | for (unsigned r = 0; r < set->nreplicas; ++r) { | |
1824 | if (!replica_is_replica_healthy(r, set_hs)) { | |
1825 | count_healthy = 0; | |
1826 | continue; | |
1827 | } | |
1828 | ||
1829 | if (count_healthy == 0) | |
1830 | first_healthy = r; | |
1831 | ||
1832 | ++count_healthy; | |
1833 | struct pool_hdr *hdrh = | |
1834 | PART(REP(set, first_healthy), 0)->hdr; | |
1835 | struct pool_hdr *hdr = PART(REP(set, r), 0)->hdr; | |
1836 | if (uuidcmp(hdrh->uuid, hdr->next_repl_uuid) == 0 && | |
1837 | count_healthy < set->nreplicas) { | |
1838 | /* | |
1839 | * Healthy replicas form a cycle shorter than | |
1840 | * the number of all replicas; for the user it | |
1841 | * means that: | |
1842 | */ | |
1843 | ERR( | |
1844 | "alien replica found (probably coming from a different poolset)"); | |
1845 | return -1; | |
1846 | } | |
1847 | } | |
1848 | return 0; | |
1849 | } | |
1850 | ||
1851 | /* | |
1852 | * check_replica_sizes -- (internal) check if all replicas are large | |
1853 | * enough to hold data from a healthy replica | |
1854 | */ | |
1855 | static int | |
1856 | check_replica_sizes(struct pool_set *set, struct poolset_health_status *set_hs) | |
1857 | { | |
1858 | LOG(3, "set %p, set_hs %p", set, set_hs); | |
1859 | ssize_t pool_size = -1; | |
1860 | for (unsigned r = 0; r < set->nreplicas; ++r) { | |
1861 | /* skip broken replicas */ | |
1862 | if (!replica_is_replica_healthy(r, set_hs)) | |
1863 | continue; | |
1864 | ||
1865 | /* get the size of a pool in the replica */ | |
1866 | ssize_t replica_pool_size; | |
1867 | if (REP(set, r)->remote) | |
1868 | /* XXX: no way to get the size of a remote pool yet */ | |
1869 | replica_pool_size = (ssize_t)set->poolsize; | |
1870 | else | |
1871 | replica_pool_size = replica_get_pool_size(set, r); | |
1872 | ||
1873 | if (replica_pool_size < 0) { | |
1874 | LOG(1, "getting pool size from replica %u failed", r); | |
1875 | set_hs->replica[r]->flags |= IS_BROKEN; | |
1876 | continue; | |
1877 | } | |
1878 | ||
1879 | /* check if the pool is bigger than minimum size */ | |
1880 | enum pool_type type = pool_hdr_get_type(HDR(REP(set, r), 0)); | |
1881 | if ((size_t)replica_pool_size < pool_get_min_size(type)) { | |
1882 | LOG(1, | |
1883 | "pool size from replica %u is smaller than the minimum size allowed for the pool", | |
1884 | r); | |
1885 | set_hs->replica[r]->flags |= IS_BROKEN; | |
1886 | continue; | |
1887 | } | |
1888 | ||
1889 | /* check if each replica is big enough to hold the pool data */ | |
1890 | if (set->poolsize < (size_t)replica_pool_size) { | |
1891 | ERR( | |
1892 | "some replicas are too small to hold synchronized data"); | |
1893 | return -1; | |
1894 | } | |
1895 | ||
1896 | if (pool_size < 0) { | |
1897 | pool_size = replica_pool_size; | |
1898 | continue; | |
1899 | } | |
1900 | ||
1901 | /* check if pools in all healthy replicas are of equal size */ | |
1902 | if (pool_size != replica_pool_size) { | |
1903 | ERR("pool sizes from different replicas differ"); | |
1904 | return -1; | |
1905 | } | |
1906 | } | |
1907 | return 0; | |
1908 | } | |
1909 | ||
1910 | /* | |
1911 | * replica_read_features -- (internal) read features from the header | |
1912 | */ | |
1913 | static int | |
1914 | replica_read_features(struct pool_set *set, | |
1915 | struct poolset_health_status *set_hs, | |
1916 | features_t *features) | |
1917 | { | |
1918 | LOG(3, "set %p set_hs %p features %p", set, set_hs, features); | |
1919 | ||
1920 | ASSERTne(features, NULL); | |
1921 | ||
1922 | for (unsigned r = 0; r < set->nreplicas; r++) { | |
1923 | struct pool_replica *rep = set->replica[r]; | |
1924 | struct replica_health_status *rep_hs = set_hs->replica[r]; | |
1925 | ||
1926 | if (rep->remote) { | |
1927 | if (rep_hs->flags & IS_BROKEN) | |
1928 | continue; | |
1929 | ||
1930 | struct pool_hdr *hdrp = rep->part[0].hdr; | |
1931 | memcpy(features, &hdrp->features, sizeof(*features)); | |
1932 | ||
1933 | return 0; | |
1934 | } | |
1935 | ||
1936 | for (unsigned p = 0; p < rep->nparts; p++) { | |
1937 | struct pool_set_part *part = &rep->part[p]; | |
1938 | ||
1939 | if (part->fd == -1) | |
1940 | continue; | |
1941 | ||
1942 | if (util_map_hdr(part, MAP_SHARED, 0) != 0) { | |
1943 | LOG(1, "header mapping failed"); | |
1944 | return -1; | |
1945 | } | |
1946 | ||
1947 | struct pool_hdr *hdrp = part->hdr; | |
1948 | memcpy(features, &hdrp->features, sizeof(*features)); | |
1949 | ||
1950 | util_unmap_hdr(part); | |
1951 | ||
1952 | return 0; | |
1953 | } | |
1954 | } | |
1955 | ||
1956 | /* no healthy replica/part found */ | |
1957 | return -1; | |
1958 | } | |
1959 | ||
1960 | /* | |
1961 | * replica_check_poolset_health -- check if a given poolset can be considered as | |
1962 | * healthy, and store the status in a helping structure | |
1963 | */ | |
1964 | int | |
1965 | replica_check_poolset_health(struct pool_set *set, | |
1966 | struct poolset_health_status **set_hsp, | |
1967 | int called_from_sync, unsigned flags) | |
1968 | { | |
1969 | LOG(3, "set %p, set_hsp %p, called_from_sync %i, flags %u", | |
1970 | set, set_hsp, called_from_sync, flags); | |
1971 | ||
1972 | if (replica_create_poolset_health_status(set, set_hsp)) { | |
1973 | LOG(1, "creating poolset health status failed"); | |
1974 | return -1; | |
1975 | } | |
1976 | ||
1977 | struct poolset_health_status *set_hs = *set_hsp; | |
1978 | ||
1979 | /* check if part files exist and are accessible */ | |
1980 | if (check_and_open_poolset_part_files(set, set_hs, flags)) { | |
1981 | LOG(1, "poolset part files check failed"); | |
1982 | goto err; | |
1983 | } | |
1984 | ||
1985 | features_t features; | |
1986 | int check_bad_blks; | |
1987 | int fix_bad_blks = called_from_sync && fix_bad_blocks(flags); | |
1988 | ||
1989 | if (fix_bad_blks) { | |
1990 | /* | |
1991 | * We will fix bad blocks, so we cannot read features here, | |
1992 | * because reading could fail, because of bad blocks. | |
1993 | * We will read features after having bad blocks fixed. | |
1994 | * | |
1995 | * Fixing bad blocks implies checking bad blocks. | |
1996 | */ | |
1997 | check_bad_blks = 1; | |
1998 | } else { | |
1999 | /* | |
2000 | * We will not fix bad blocks, so we have to read features here. | |
2001 | */ | |
2002 | if (replica_read_features(set, set_hs, &features)) { | |
2003 | LOG(1, "reading features failed"); | |
2004 | goto err; | |
2005 | } | |
2006 | check_bad_blks = features.compat & POOL_FEAT_CHECK_BAD_BLOCKS; | |
2007 | } | |
2008 | ||
2009 | /* check for bad blocks when in dry run or clear them otherwise */ | |
2010 | if (replica_badblocks_check_or_clear(set, set_hs, is_dry_run(flags), | |
2011 | called_from_sync, check_bad_blks, fix_bad_blks)) { | |
2012 | LOG(1, "replica bad_blocks check failed"); | |
2013 | goto err; | |
2014 | } | |
2015 | ||
2016 | /* read features after fixing bad blocks */ | |
2017 | if (fix_bad_blks && replica_read_features(set, set_hs, &features)) { | |
2018 | LOG(1, "reading features failed"); | |
2019 | goto err; | |
2020 | } | |
2021 | ||
2022 | /* set ignore_sds flag basing on features read from the header */ | |
2023 | set->ignore_sds = !(features.incompat & POOL_FEAT_SDS); | |
2024 | ||
2025 | /* map all headers */ | |
2026 | map_all_unbroken_headers(set, set_hs); | |
2027 | ||
2028 | /* | |
2029 | * Check if checksums and signatures are correct for all parts | |
2030 | * in all replicas. | |
2031 | */ | |
2032 | check_checksums_and_signatures(set, set_hs); | |
2033 | ||
2034 | /* check if option flags are consistent */ | |
2035 | if (check_options(set, set_hs)) { | |
2036 | LOG(1, "flags check failed"); | |
2037 | goto err; | |
2038 | } | |
2039 | ||
2040 | if (!set->ignore_sds && check_shutdown_state(set, set_hs)) { | |
2041 | LOG(1, "replica shutdown_state check failed"); | |
2042 | goto err; | |
2043 | } | |
2044 | ||
2045 | /* check if uuids in parts across each replica are consistent */ | |
2046 | if (check_replicas_consistency(set, set_hs)) { | |
2047 | LOG(1, "replica consistency check failed"); | |
2048 | goto err; | |
2049 | } | |
2050 | ||
2051 | /* check poolset_uuid values between replicas */ | |
2052 | if (check_poolset_uuids(set, set_hs)) { | |
2053 | LOG(1, "poolset uuids check failed"); | |
2054 | goto err; | |
2055 | } | |
2056 | ||
2057 | /* check if uuids for adjacent replicas are consistent */ | |
2058 | if (check_uuids_between_replicas(set, set_hs)) { | |
2059 | LOG(1, "replica uuids check failed"); | |
2060 | goto err; | |
2061 | } | |
2062 | ||
2063 | /* check if healthy replicas make up another poolset */ | |
2064 | if (check_replica_cycles(set, set_hs)) { | |
2065 | LOG(1, "replica cycles check failed"); | |
2066 | goto err; | |
2067 | } | |
2068 | ||
2069 | /* check if replicas are large enough */ | |
2070 | if (check_replica_sizes(set, set_hs)) { | |
2071 | LOG(1, "replica sizes check failed"); | |
2072 | goto err; | |
2073 | } | |
2074 | ||
2075 | if (check_store_all_sizes(set, set_hs)) { | |
2076 | LOG(1, "reading pool sizes failed"); | |
2077 | goto err; | |
2078 | } | |
2079 | ||
2080 | unmap_all_headers(set); | |
2081 | util_poolset_fdclose_always(set); | |
2082 | return 0; | |
2083 | ||
2084 | err: | |
2085 | errno = EINVAL; | |
2086 | unmap_all_headers(set); | |
2087 | util_poolset_fdclose_always(set); | |
2088 | replica_free_poolset_health_status(set_hs); | |
2089 | return -1; | |
2090 | } | |
2091 | ||
2092 | /* | |
2093 | * replica_get_pool_size -- find the effective size (mapped) of a pool based | |
2094 | * on metadata from given replica | |
2095 | */ | |
2096 | ssize_t | |
2097 | replica_get_pool_size(struct pool_set *set, unsigned repn) | |
2098 | { | |
2099 | LOG(3, "set %p, repn %u", set, repn); | |
2100 | struct pool_set_part *part = PART(REP(set, repn), 0); | |
2101 | int should_close_part = 0; | |
2102 | int should_unmap_part = 0; | |
2103 | if (part->fd == -1) { | |
2104 | if (util_part_open(part, 0, 0)) | |
2105 | return -1; | |
2106 | ||
2107 | should_close_part = 1; | |
2108 | } | |
2109 | ||
2110 | if (part->addr == NULL) { | |
2111 | if (util_map_part(part, NULL, | |
2112 | ALIGN_UP(sizeof(PMEMobjpool), part->alignment), 0, | |
2113 | MAP_SHARED, 1)) { | |
2114 | util_part_fdclose(part); | |
2115 | return -1; | |
2116 | } | |
2117 | should_unmap_part = 1; | |
2118 | } | |
2119 | ||
2120 | PMEMobjpool *pop = (PMEMobjpool *)part->addr; | |
2121 | ssize_t ret = (ssize_t)(pop->heap_offset + pop->heap_size); | |
2122 | ||
2123 | if (should_unmap_part) | |
2124 | util_unmap_part(part); | |
2125 | if (should_close_part) | |
2126 | util_part_fdclose(part); | |
2127 | ||
2128 | return ret; | |
2129 | } | |
2130 | ||
2131 | /* | |
2132 | * replica_check_part_sizes -- check if all parts are large enough | |
2133 | */ | |
2134 | int | |
2135 | replica_check_part_sizes(struct pool_set *set, size_t min_size) | |
2136 | { | |
2137 | LOG(3, "set %p, min_size %zu", set, min_size); | |
2138 | for (unsigned r = 0; r < set->nreplicas; ++r) { | |
2139 | struct pool_replica *rep = set->replica[r]; | |
2140 | if (rep->remote != NULL) | |
2141 | /* skip remote replicas */ | |
2142 | continue; | |
2143 | ||
2144 | for (unsigned p = 0; p < rep->nparts; ++p) { | |
2145 | if (PART(rep, p)->filesize < min_size) { | |
2146 | ERR("replica %u, part %u: file is too small", | |
2147 | r, p); | |
2148 | errno = EINVAL; | |
2149 | return -1; | |
2150 | } | |
2151 | } | |
2152 | } | |
2153 | return 0; | |
2154 | } | |
2155 | ||
2156 | /* | |
2157 | * replica_check_local_part_dir -- check if directory for the part file | |
2158 | * exists | |
2159 | */ | |
2160 | int | |
2161 | replica_check_local_part_dir(struct pool_set *set, unsigned repn, | |
2162 | unsigned partn) | |
2163 | { | |
2164 | LOG(3, "set %p, repn %u, partn %u", set, repn, partn); | |
2165 | char *path = Strdup(PART(REP(set, repn), partn)->path); | |
2166 | const char *dir = dirname(path); | |
2167 | os_stat_t sb; | |
2168 | if (os_stat(dir, &sb) != 0 || !(sb.st_mode & S_IFDIR)) { | |
2169 | ERR( | |
2170 | "directory %s for part %u in replica %u does not exist or is not accessible", | |
2171 | path, partn, repn); | |
2172 | Free(path); | |
2173 | return -1; | |
2174 | } | |
2175 | Free(path); | |
2176 | return 0; | |
2177 | } | |
2178 | ||
2179 | /* | |
2180 | * replica_check_part_dirs -- (internal) check if directories for part files | |
2181 | * exist | |
2182 | */ | |
2183 | int | |
2184 | replica_check_part_dirs(struct pool_set *set) | |
2185 | { | |
2186 | LOG(3, "set %p", set); | |
2187 | for (unsigned r = 0; r < set->nreplicas; ++r) { | |
2188 | struct pool_replica *rep = set->replica[r]; | |
2189 | if (rep->remote != NULL) | |
2190 | /* skip remote replicas */ | |
2191 | continue; | |
2192 | ||
2193 | for (unsigned p = 0; p < rep->nparts; ++p) { | |
2194 | if (replica_check_local_part_dir(set, r, p)) | |
2195 | return -1; | |
2196 | } | |
2197 | } | |
2198 | return 0; | |
2199 | } | |
2200 | ||
2201 | /* | |
2202 | * replica_open_replica_part_files -- open all part files for a replica | |
2203 | */ | |
2204 | int | |
2205 | replica_open_replica_part_files(struct pool_set *set, unsigned repn) | |
2206 | { | |
2207 | LOG(3, "set %p, repn %u", set, repn); | |
2208 | struct pool_replica *rep = set->replica[repn]; | |
2209 | for (unsigned p = 0; p < rep->nparts; ++p) { | |
2210 | /* skip already opened files */ | |
2211 | if (rep->part[p].fd != -1) | |
2212 | continue; | |
2213 | ||
2214 | if (util_part_open(&rep->part[p], 0, 0)) { | |
2215 | LOG(1, "part files open failed for replica %u, part %u", | |
2216 | repn, p); | |
2217 | errno = EINVAL; | |
2218 | goto err; | |
2219 | } | |
2220 | } | |
2221 | return 0; | |
2222 | ||
2223 | err: | |
2224 | util_replica_fdclose(set->replica[repn]); | |
2225 | return -1; | |
2226 | } | |
2227 | ||
2228 | /* | |
2229 | * replica_open_poolset_part_files -- open all part files for a poolset | |
2230 | */ | |
2231 | int | |
2232 | replica_open_poolset_part_files(struct pool_set *set) | |
2233 | { | |
2234 | LOG(3, "set %p", set); | |
2235 | for (unsigned r = 0; r < set->nreplicas; ++r) { | |
2236 | if (set->replica[r]->remote) | |
2237 | continue; | |
2238 | if (replica_open_replica_part_files(set, r)) { | |
2239 | LOG(1, "opening replica %u, part files failed", r); | |
2240 | goto err; | |
2241 | } | |
2242 | } | |
2243 | ||
2244 | return 0; | |
2245 | ||
2246 | err: | |
2247 | util_poolset_fdclose_always(set); | |
2248 | return -1; | |
2249 | } | |
2250 | ||
2251 | /* | |
2252 | * pmempool_syncU -- synchronize replicas within a poolset | |
2253 | */ | |
2254 | #ifndef _WIN32 | |
2255 | static inline | |
2256 | #endif | |
2257 | int | |
2258 | pmempool_syncU(const char *poolset, unsigned flags) | |
2259 | { | |
2260 | LOG(3, "poolset %s, flags %u", poolset, flags); | |
2261 | ASSERTne(poolset, NULL); | |
2262 | ||
2263 | /* check if poolset has correct signature */ | |
2264 | if (util_is_poolset_file(poolset) != 1) { | |
2265 | ERR("file is not a poolset file"); | |
2266 | goto err; | |
2267 | } | |
2268 | ||
2269 | /* check if flags are supported */ | |
2270 | if (check_flags_sync(flags)) { | |
2271 | ERR("unsupported flags"); | |
2272 | errno = EINVAL; | |
2273 | goto err; | |
2274 | } | |
2275 | ||
2276 | /* open poolset file */ | |
2277 | int fd = util_file_open(poolset, NULL, 0, O_RDONLY); | |
2278 | if (fd < 0) { | |
2279 | ERR("cannot open a poolset file"); | |
2280 | goto err; | |
2281 | } | |
2282 | ||
2283 | /* fill up pool_set structure */ | |
2284 | struct pool_set *set = NULL; | |
2285 | if (util_poolset_parse(&set, poolset, fd)) { | |
2286 | ERR("parsing input poolset failed"); | |
2287 | goto err_close_file; | |
2288 | } | |
2289 | ||
2290 | if (set->nreplicas == 1) { | |
2291 | ERR("no replica(s) found in the pool set"); | |
2292 | errno = EINVAL; | |
2293 | goto err_close_file; | |
2294 | } | |
2295 | ||
2296 | if (set->remote && util_remote_load()) { | |
2297 | ERR("remote replication not available"); | |
2298 | errno = ENOTSUP; | |
2299 | goto err_close_file; | |
2300 | } | |
2301 | ||
2302 | /* sync all replicas */ | |
2303 | if (replica_sync(set, NULL, flags)) { | |
2304 | LOG(1, "synchronization failed"); | |
2305 | goto err_close_all; | |
2306 | } | |
2307 | ||
2308 | util_poolset_close(set, DO_NOT_DELETE_PARTS); | |
2309 | os_close(fd); | |
2310 | return 0; | |
2311 | ||
2312 | err_close_all: | |
2313 | util_poolset_close(set, DO_NOT_DELETE_PARTS); | |
2314 | ||
2315 | err_close_file: | |
2316 | os_close(fd); | |
2317 | ||
2318 | err: | |
2319 | if (errno == 0) | |
2320 | errno = EINVAL; | |
2321 | ||
2322 | return -1; | |
2323 | } | |
2324 | ||
2325 | #ifndef _WIN32 | |
2326 | /* | |
2327 | * pmempool_sync -- synchronize replicas within a poolset | |
2328 | */ | |
2329 | int | |
2330 | pmempool_sync(const char *poolset, unsigned flags) | |
2331 | { | |
2332 | return pmempool_syncU(poolset, flags); | |
2333 | } | |
2334 | #else | |
2335 | /* | |
2336 | * pmempool_syncW -- synchronize replicas within a poolset in widechar | |
2337 | */ | |
2338 | int | |
2339 | pmempool_syncW(const wchar_t *poolset, unsigned flags) | |
2340 | { | |
2341 | char *path = util_toUTF8(poolset); | |
2342 | if (path == NULL) { | |
2343 | ERR("Invalid poolest file path."); | |
2344 | return -1; | |
2345 | } | |
2346 | ||
2347 | int ret = pmempool_syncU(path, flags); | |
2348 | ||
2349 | util_free_UTF8(path); | |
2350 | return ret; | |
2351 | } | |
2352 | #endif | |
2353 | ||
2354 | /* | |
2355 | * pmempool_transformU -- alter poolset structure | |
2356 | */ | |
2357 | #ifndef _WIN32 | |
2358 | static inline | |
2359 | #endif | |
2360 | int | |
2361 | pmempool_transformU(const char *poolset_src, | |
2362 | const char *poolset_dst, unsigned flags) | |
2363 | { | |
2364 | LOG(3, "poolset_src %s, poolset_dst %s, flags %u", poolset_src, | |
2365 | poolset_dst, flags); | |
2366 | ASSERTne(poolset_src, NULL); | |
2367 | ASSERTne(poolset_dst, NULL); | |
2368 | ||
2369 | /* check if the source poolset has correct signature */ | |
2370 | if (util_is_poolset_file(poolset_src) != 1) { | |
2371 | ERR("source file is not a poolset file"); | |
2372 | goto err; | |
2373 | } | |
2374 | ||
2375 | /* check if the destination poolset has correct signature */ | |
2376 | if (util_is_poolset_file(poolset_dst) != 1) { | |
2377 | ERR("destination file is not a poolset file"); | |
2378 | goto err; | |
2379 | } | |
2380 | ||
2381 | /* check if flags are supported */ | |
2382 | if (check_flags_transform(flags)) { | |
2383 | ERR("unsupported flags"); | |
2384 | errno = EINVAL; | |
2385 | goto err; | |
2386 | } | |
2387 | ||
2388 | /* open the source poolset file */ | |
2389 | int fd_in = util_file_open(poolset_src, NULL, 0, O_RDONLY); | |
2390 | if (fd_in < 0) { | |
2391 | ERR("cannot open source poolset file"); | |
2392 | goto err; | |
2393 | } | |
2394 | ||
2395 | /* parse the source poolset file */ | |
2396 | struct pool_set *set_in = NULL; | |
2397 | if (util_poolset_parse(&set_in, poolset_src, fd_in)) { | |
2398 | ERR("parsing source poolset failed"); | |
2399 | os_close(fd_in); | |
2400 | goto err; | |
2401 | } | |
2402 | os_close(fd_in); | |
2403 | ||
2404 | /* open the destination poolset file */ | |
2405 | int fd_out = util_file_open(poolset_dst, NULL, 0, O_RDONLY); | |
2406 | if (fd_out < 0) { | |
2407 | ERR("cannot open destination poolset file"); | |
2408 | goto err; | |
2409 | } | |
2410 | ||
2411 | enum del_parts_mode del = DO_NOT_DELETE_PARTS; | |
2412 | ||
2413 | /* parse the destination poolset file */ | |
2414 | struct pool_set *set_out = NULL; | |
2415 | if (util_poolset_parse(&set_out, poolset_dst, fd_out)) { | |
2416 | ERR("parsing destination poolset failed"); | |
2417 | os_close(fd_out); | |
2418 | goto err_free_poolin; | |
2419 | } | |
2420 | os_close(fd_out); | |
2421 | ||
2422 | /* check if the source poolset is of a correct type */ | |
2423 | enum pool_type ptype = pool_set_type(set_in); | |
2424 | if (ptype != POOL_TYPE_OBJ) { | |
2425 | errno = EINVAL; | |
2426 | ERR("transform is not supported for given pool type: %s", | |
2427 | pool_get_pool_type_str(ptype)); | |
2428 | goto err_free_poolout; | |
2429 | } | |
2430 | ||
2431 | /* load remote library if needed */ | |
2432 | if (set_in->remote && util_remote_load()) { | |
2433 | ERR("remote replication not available"); | |
2434 | goto err_free_poolout; | |
2435 | } | |
2436 | if (set_out->remote && util_remote_load()) { | |
2437 | ERR("remote replication not available"); | |
2438 | goto err_free_poolout; | |
2439 | } | |
2440 | ||
2441 | del = is_dry_run(flags) ? DO_NOT_DELETE_PARTS : DELETE_CREATED_PARTS; | |
2442 | ||
2443 | /* transform poolset */ | |
2444 | if (replica_transform(set_in, set_out, flags)) { | |
2445 | LOG(1, "transformation failed"); | |
2446 | goto err_free_poolout; | |
2447 | } | |
2448 | ||
2449 | util_poolset_close(set_in, DO_NOT_DELETE_PARTS); | |
2450 | util_poolset_close(set_out, DO_NOT_DELETE_PARTS); | |
2451 | return 0; | |
2452 | ||
2453 | err_free_poolout: | |
2454 | util_poolset_close(set_out, del); | |
2455 | ||
2456 | err_free_poolin: | |
2457 | util_poolset_close(set_in, DO_NOT_DELETE_PARTS); | |
2458 | ||
2459 | err: | |
2460 | if (errno == 0) | |
2461 | errno = EINVAL; | |
2462 | ||
2463 | return -1; | |
2464 | } | |
2465 | ||
2466 | #ifndef _WIN32 | |
2467 | /* | |
2468 | * pmempool_transform -- alter poolset structure | |
2469 | */ | |
2470 | int | |
2471 | pmempool_transform(const char *poolset_src, | |
2472 | const char *poolset_dst, unsigned flags) | |
2473 | { | |
2474 | return pmempool_transformU(poolset_src, poolset_dst, flags); | |
2475 | } | |
2476 | #else | |
2477 | /* | |
2478 | * pmempool_transformW -- alter poolset structure in widechar | |
2479 | */ | |
2480 | int | |
2481 | pmempool_transformW(const wchar_t *poolset_src, | |
2482 | const wchar_t *poolset_dst, unsigned flags) | |
2483 | { | |
2484 | char *path_src = util_toUTF8(poolset_src); | |
2485 | if (path_src == NULL) { | |
2486 | ERR("Invalid source poolest file path."); | |
2487 | return -1; | |
2488 | } | |
2489 | ||
2490 | char *path_dst = util_toUTF8(poolset_dst); | |
2491 | if (path_dst == NULL) { | |
2492 | ERR("Invalid destination poolest file path."); | |
2493 | Free(path_src); | |
2494 | return -1; | |
2495 | } | |
2496 | ||
2497 | int ret = pmempool_transformU(path_src, path_dst, flags); | |
2498 | ||
2499 | util_free_UTF8(path_src); | |
2500 | util_free_UTF8(path_dst); | |
2501 | return ret; | |
2502 | } | |
2503 | #endif |