]> git.proxmox.com Git - mirror_zfs.git/blob - lib/libzutil/zutil_import.c
projects / mirror_zfs.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Rescan enclosure sysfs path on import
[mirror_zfs.git] / lib / libzutil / zutil_import.c
1 /*
2 * CDDL HEADER START
3 *
4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
7 *
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
12 *
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 *
19 * CDDL HEADER END
20 */
21 /*
22 * Copyright 2015 Nexenta Systems, Inc. All rights reserved.
23 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Copyright (c) 2012, 2018 by Delphix. All rights reserved.
25 * Copyright 2015 RackTop Systems.
26 * Copyright (c) 2016, Intel Corporation.
27 * Copyright (c) 2021, Colm Buckley <colm@tuatha.org>
28 */
29
30 /*
31 * Pool import support functions.
32 *
33 * Used by zpool, ztest, zdb, and zhack to locate importable configs. Since
34 * these commands are expected to run in the global zone, we can assume
35 * that the devices are all readable when called.
36 *
37 * To import a pool, we rely on reading the configuration information from the
38 * ZFS label of each device. If we successfully read the label, then we
39 * organize the configuration information in the following hierarchy:
40 *
41 * pool guid -> toplevel vdev guid -> label txg
42 *
43 * Duplicate entries matching this same tuple will be discarded. Once we have
44 * examined every device, we pick the best label txg config for each toplevel
45 * vdev. We then arrange these toplevel vdevs into a complete pool config, and
46 * update any paths that have changed. Finally, we attempt to import the pool
47 * using our derived config, and record the results.
48 */
49
50 #include <aio.h>
51 #include <ctype.h>
52 #include <dirent.h>
53 #include <errno.h>
54 #include <libintl.h>
55 #include <libgen.h>
56 #include <stddef.h>
57 #include <stdlib.h>
58 #include <string.h>
59 #include <sys/stat.h>
60 #include <unistd.h>
61 #include <fcntl.h>
62 #include <sys/dktp/fdisk.h>
63 #include <sys/vdev_impl.h>
64 #include <sys/fs/zfs.h>
65
66 #include <thread_pool.h>
67 #include <libzutil.h>
68 #include <libnvpair.h>
69
70 #include "zutil_import.h"
71
72 static __attribute__((format(printf, 2, 3))) void
73 zutil_error_aux(libpc_handle_t *hdl, const char *fmt, ...)
74 {
75 va_list ap;
76
77 va_start(ap, fmt);
78
79 (void) vsnprintf(hdl->lpc_desc, sizeof (hdl->lpc_desc), fmt, ap);
80 hdl->lpc_desc_active = B_TRUE;
81
82 va_end(ap);
83 }
84
85 static void
86 zutil_verror(libpc_handle_t *hdl, const char *error, const char *fmt,
87 va_list ap)
88 {
89 char action[1024];
90
91 (void) vsnprintf(action, sizeof (action), fmt, ap);
92
93 if (hdl->lpc_desc_active)
94 hdl->lpc_desc_active = B_FALSE;
95 else
96 hdl->lpc_desc[0] = '\0';
97
98 if (hdl->lpc_printerr) {
99 if (hdl->lpc_desc[0] != '\0')
100 error = hdl->lpc_desc;
101
102 (void) fprintf(stderr, "%s: %s\n", action, error);
103 }
104 }
105
106 static __attribute__((format(printf, 3, 4))) int
107 zutil_error_fmt(libpc_handle_t *hdl, const char *error, const char *fmt, ...)
108 {
109 va_list ap;
110
111 va_start(ap, fmt);
112
113 zutil_verror(hdl, error, fmt, ap);
114
115 va_end(ap);
116
117 return (-1);
118 }
119
120 static int
121 zutil_error(libpc_handle_t *hdl, const char *error, const char *msg)
122 {
123 return (zutil_error_fmt(hdl, error, "%s", msg));
124 }
125
126 static int
127 zutil_no_memory(libpc_handle_t *hdl)
128 {
129 zutil_error(hdl, EZFS_NOMEM, "internal error");
130 exit(1);
131 }
132
133 void *
134 zutil_alloc(libpc_handle_t *hdl, size_t size)
135 {
136 void *data;
137
138 if ((data = calloc(1, size)) == NULL)
139 (void) zutil_no_memory(hdl);
140
141 return (data);
142 }
143
144 char *
145 zutil_strdup(libpc_handle_t *hdl, const char *str)
146 {
147 char *ret;
148
149 if ((ret = strdup(str)) == NULL)
150 (void) zutil_no_memory(hdl);
151
152 return (ret);
153 }
154
155 static char *
156 zutil_strndup(libpc_handle_t *hdl, const char *str, size_t n)
157 {
158 char *ret;
159
160 if ((ret = strndup(str, n)) == NULL)
161 (void) zutil_no_memory(hdl);
162
163 return (ret);
164 }
165
166 /*
167 * Intermediate structures used to gather configuration information.
168 */
169 typedef struct config_entry {
170 uint64_t ce_txg;
171 nvlist_t *ce_config;
172 struct config_entry *ce_next;
173 } config_entry_t;
174
175 typedef struct vdev_entry {
176 uint64_t ve_guid;
177 config_entry_t *ve_configs;
178 struct vdev_entry *ve_next;
179 } vdev_entry_t;
180
181 typedef struct pool_entry {
182 uint64_t pe_guid;
183 vdev_entry_t *pe_vdevs;
184 struct pool_entry *pe_next;
185 } pool_entry_t;
186
187 typedef struct name_entry {
188 char *ne_name;
189 uint64_t ne_guid;
190 uint64_t ne_order;
191 uint64_t ne_num_labels;
192 struct name_entry *ne_next;
193 } name_entry_t;
194
195 typedef struct pool_list {
196 pool_entry_t *pools;
197 name_entry_t *names;
198 } pool_list_t;
199
200 /*
201 * Go through and fix up any path and/or devid information for the given vdev
202 * configuration.
203 */
204 static int
205 fix_paths(libpc_handle_t *hdl, nvlist_t *nv, name_entry_t *names)
206 {
207 nvlist_t **child;
208 uint_t c, children;
209 uint64_t guid;
210 name_entry_t *ne, *best;
211 char *path;
212
213 if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
214 &child, &children) == 0) {
215 for (c = 0; c < children; c++)
216 if (fix_paths(hdl, child[c], names) != 0)
217 return (-1);
218 return (0);
219 }
220
221 /*
222 * This is a leaf (file or disk) vdev. In either case, go through
223 * the name list and see if we find a matching guid. If so, replace
224 * the path and see if we can calculate a new devid.
225 *
226 * There may be multiple names associated with a particular guid, in
227 * which case we have overlapping partitions or multiple paths to the
228 * same disk. In this case we prefer to use the path name which
229 * matches the ZPOOL_CONFIG_PATH. If no matching entry is found we
230 * use the lowest order device which corresponds to the first match
231 * while traversing the ZPOOL_IMPORT_PATH search path.
232 */
233 verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) == 0);
234 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &path) != 0)
235 path = NULL;
236
237 best = NULL;
238 for (ne = names; ne != NULL; ne = ne->ne_next) {
239 if (ne->ne_guid == guid) {
240 if (path == NULL) {
241 best = ne;
242 break;
243 }
244
245 if ((strlen(path) == strlen(ne->ne_name)) &&
246 strncmp(path, ne->ne_name, strlen(path)) == 0) {
247 best = ne;
248 break;
249 }
250
251 if (best == NULL) {
252 best = ne;
253 continue;
254 }
255
256 /* Prefer paths with move vdev labels. */
257 if (ne->ne_num_labels > best->ne_num_labels) {
258 best = ne;
259 continue;
260 }
261
262 /* Prefer paths earlier in the search order. */
263 if (ne->ne_num_labels == best->ne_num_labels &&
264 ne->ne_order < best->ne_order) {
265 best = ne;
266 continue;
267 }
268 }
269 }
270
271 if (best == NULL)
272 return (0);
273
274 if (nvlist_add_string(nv, ZPOOL_CONFIG_PATH, best->ne_name) != 0)
275 return (-1);
276
277 update_vdev_config_dev_strs(nv);
278
279 return (0);
280 }
281
282 /*
283 * Add the given configuration to the list of known devices.
284 */
285 static int
286 add_config(libpc_handle_t *hdl, pool_list_t *pl, const char *path,
287 int order, int num_labels, nvlist_t *config)
288 {
289 uint64_t pool_guid, vdev_guid, top_guid, txg, state;
290 pool_entry_t *pe;
291 vdev_entry_t *ve;
292 config_entry_t *ce;
293 name_entry_t *ne;
294
295 /*
296 * If this is a hot spare not currently in use or level 2 cache
297 * device, add it to the list of names to translate, but don't do
298 * anything else.
299 */
300 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_STATE,
301 &state) == 0 &&
302 (state == POOL_STATE_SPARE || state == POOL_STATE_L2CACHE) &&
303 nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID, &vdev_guid) == 0) {
304 if ((ne = zutil_alloc(hdl, sizeof (name_entry_t))) == NULL)
305 return (-1);
306
307 if ((ne->ne_name = zutil_strdup(hdl, path)) == NULL) {
308 free(ne);
309 return (-1);
310 }
311 ne->ne_guid = vdev_guid;
312 ne->ne_order = order;
313 ne->ne_num_labels = num_labels;
314 ne->ne_next = pl->names;
315 pl->names = ne;
316
317 return (0);
318 }
319
320 /*
321 * If we have a valid config but cannot read any of these fields, then
322 * it means we have a half-initialized label. In vdev_label_init()
323 * we write a label with txg == 0 so that we can identify the device
324 * in case the user refers to the same disk later on. If we fail to
325 * create the pool, we'll be left with a label in this state
326 * which should not be considered part of a valid pool.
327 */
328 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
329 &pool_guid) != 0 ||
330 nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID,
331 &vdev_guid) != 0 ||
332 nvlist_lookup_uint64(config, ZPOOL_CONFIG_TOP_GUID,
333 &top_guid) != 0 ||
334 nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG,
335 &txg) != 0 || txg == 0) {
336 return (0);
337 }
338
339 /*
340 * First, see if we know about this pool. If not, then add it to the
341 * list of known pools.
342 */
343 for (pe = pl->pools; pe != NULL; pe = pe->pe_next) {
344 if (pe->pe_guid == pool_guid)
345 break;
346 }
347
348 if (pe == NULL) {
349 if ((pe = zutil_alloc(hdl, sizeof (pool_entry_t))) == NULL) {
350 return (-1);
351 }
352 pe->pe_guid = pool_guid;
353 pe->pe_next = pl->pools;
354 pl->pools = pe;
355 }
356
357 /*
358 * Second, see if we know about this toplevel vdev. Add it if its
359 * missing.
360 */
361 for (ve = pe->pe_vdevs; ve != NULL; ve = ve->ve_next) {
362 if (ve->ve_guid == top_guid)
363 break;
364 }
365
366 if (ve == NULL) {
367 if ((ve = zutil_alloc(hdl, sizeof (vdev_entry_t))) == NULL) {
368 return (-1);
369 }
370 ve->ve_guid = top_guid;
371 ve->ve_next = pe->pe_vdevs;
372 pe->pe_vdevs = ve;
373 }
374
375 /*
376 * Third, see if we have a config with a matching transaction group. If
377 * so, then we do nothing. Otherwise, add it to the list of known
378 * configs.
379 */
380 for (ce = ve->ve_configs; ce != NULL; ce = ce->ce_next) {
381 if (ce->ce_txg == txg)
382 break;
383 }
384
385 if (ce == NULL) {
386 if ((ce = zutil_alloc(hdl, sizeof (config_entry_t))) == NULL) {
387 return (-1);
388 }
389 ce->ce_txg = txg;
390 ce->ce_config = fnvlist_dup(config);
391 ce->ce_next = ve->ve_configs;
392 ve->ve_configs = ce;
393 }
394
395 /*
396 * At this point we've successfully added our config to the list of
397 * known configs. The last thing to do is add the vdev guid -> path
398 * mappings so that we can fix up the configuration as necessary before
399 * doing the import.
400 */
401 if ((ne = zutil_alloc(hdl, sizeof (name_entry_t))) == NULL)
402 return (-1);
403
404 if ((ne->ne_name = zutil_strdup(hdl, path)) == NULL) {
405 free(ne);
406 return (-1);
407 }
408
409 ne->ne_guid = vdev_guid;
410 ne->ne_order = order;
411 ne->ne_num_labels = num_labels;
412 ne->ne_next = pl->names;
413 pl->names = ne;
414
415 return (0);
416 }
417
418 static int
419 zutil_pool_active(libpc_handle_t *hdl, const char *name, uint64_t guid,
420 boolean_t *isactive)
421 {
422 ASSERT(hdl->lpc_ops->pco_pool_active != NULL);
423
424 int error = hdl->lpc_ops->pco_pool_active(hdl->lpc_lib_handle, name,
425 guid, isactive);
426
427 return (error);
428 }
429
430 static nvlist_t *
431 zutil_refresh_config(libpc_handle_t *hdl, nvlist_t *tryconfig)
432 {
433 ASSERT(hdl->lpc_ops->pco_refresh_config != NULL);
434
435 return (hdl->lpc_ops->pco_refresh_config(hdl->lpc_lib_handle,
436 tryconfig));
437 }
438
439 /*
440 * Determine if the vdev id is a hole in the namespace.
441 */
442 static boolean_t
443 vdev_is_hole(uint64_t *hole_array, uint_t holes, uint_t id)
444 {
445 int c;
446
447 for (c = 0; c < holes; c++) {
448
449 /* Top-level is a hole */
450 if (hole_array[c] == id)
451 return (B_TRUE);
452 }
453 return (B_FALSE);
454 }
455
456 /*
457 * Convert our list of pools into the definitive set of configurations. We
458 * start by picking the best config for each toplevel vdev. Once that's done,
459 * we assemble the toplevel vdevs into a full config for the pool. We make a
460 * pass to fix up any incorrect paths, and then add it to the main list to
461 * return to the user.
462 */
463 static nvlist_t *
464 get_configs(libpc_handle_t *hdl, pool_list_t *pl, boolean_t active_ok,
465 nvlist_t *policy)
466 {
467 pool_entry_t *pe;
468 vdev_entry_t *ve;
469 config_entry_t *ce;
470 nvlist_t *ret = NULL, *config = NULL, *tmp = NULL, *nvtop, *nvroot;
471 nvlist_t **spares, **l2cache;
472 uint_t i, nspares, nl2cache;
473 boolean_t config_seen;
474 uint64_t best_txg;
475 char *name, *hostname = NULL;
476 uint64_t guid;
477 uint_t children = 0;
478 nvlist_t **child = NULL;
479 uint_t holes;
480 uint64_t *hole_array, max_id;
481 uint_t c;
482 boolean_t isactive;
483 uint64_t hostid;
484 nvlist_t *nvl;
485 boolean_t valid_top_config = B_FALSE;
486
487 if (nvlist_alloc(&ret, 0, 0) != 0)
488 goto nomem;
489
490 for (pe = pl->pools; pe != NULL; pe = pe->pe_next) {
491 uint64_t id, max_txg = 0;
492
493 if (nvlist_alloc(&config, NV_UNIQUE_NAME, 0) != 0)
494 goto nomem;
495 config_seen = B_FALSE;
496
497 /*
498 * Iterate over all toplevel vdevs. Grab the pool configuration
499 * from the first one we find, and then go through the rest and
500 * add them as necessary to the 'vdevs' member of the config.
501 */
502 for (ve = pe->pe_vdevs; ve != NULL; ve = ve->ve_next) {
503
504 /*
505 * Determine the best configuration for this vdev by
506 * selecting the config with the latest transaction
507 * group.
508 */
509 best_txg = 0;
510 for (ce = ve->ve_configs; ce != NULL;
511 ce = ce->ce_next) {
512
513 if (ce->ce_txg > best_txg) {
514 tmp = ce->ce_config;
515 best_txg = ce->ce_txg;
516 }
517 }
518
519 /*
520 * We rely on the fact that the max txg for the
521 * pool will contain the most up-to-date information
522 * about the valid top-levels in the vdev namespace.
523 */
524 if (best_txg > max_txg) {
525 (void) nvlist_remove(config,
526 ZPOOL_CONFIG_VDEV_CHILDREN,
527 DATA_TYPE_UINT64);
528 (void) nvlist_remove(config,
529 ZPOOL_CONFIG_HOLE_ARRAY,
530 DATA_TYPE_UINT64_ARRAY);
531
532 max_txg = best_txg;
533 hole_array = NULL;
534 holes = 0;
535 max_id = 0;
536 valid_top_config = B_FALSE;
537
538 if (nvlist_lookup_uint64(tmp,
539 ZPOOL_CONFIG_VDEV_CHILDREN, &max_id) == 0) {
540 verify(nvlist_add_uint64(config,
541 ZPOOL_CONFIG_VDEV_CHILDREN,
542 max_id) == 0);
543 valid_top_config = B_TRUE;
544 }
545
546 if (nvlist_lookup_uint64_array(tmp,
547 ZPOOL_CONFIG_HOLE_ARRAY, &hole_array,
548 &holes) == 0) {
549 verify(nvlist_add_uint64_array(config,
550 ZPOOL_CONFIG_HOLE_ARRAY,
551 hole_array, holes) == 0);
552 }
553 }
554
555 if (!config_seen) {
556 /*
557 * Copy the relevant pieces of data to the pool
558 * configuration:
559 *
560 * version
561 * pool guid
562 * name
563 * comment (if available)
564 * compatibility features (if available)
565 * pool state
566 * hostid (if available)
567 * hostname (if available)
568 */
569 uint64_t state, version;
570 char *comment = NULL;
571 char *compatibility = NULL;
572
573 version = fnvlist_lookup_uint64(tmp,
574 ZPOOL_CONFIG_VERSION);
575 fnvlist_add_uint64(config,
576 ZPOOL_CONFIG_VERSION, version);
577 guid = fnvlist_lookup_uint64(tmp,
578 ZPOOL_CONFIG_POOL_GUID);
579 fnvlist_add_uint64(config,
580 ZPOOL_CONFIG_POOL_GUID, guid);
581 name = fnvlist_lookup_string(tmp,
582 ZPOOL_CONFIG_POOL_NAME);
583 fnvlist_add_string(config,
584 ZPOOL_CONFIG_POOL_NAME, name);
585
586 if (nvlist_lookup_string(tmp,
587 ZPOOL_CONFIG_COMMENT, &comment) == 0)
588 fnvlist_add_string(config,
589 ZPOOL_CONFIG_COMMENT, comment);
590
591 if (nvlist_lookup_string(tmp,
592 ZPOOL_CONFIG_COMPATIBILITY,
593 &compatibility) == 0)
594 fnvlist_add_string(config,
595 ZPOOL_CONFIG_COMPATIBILITY,
596 compatibility);
597
598 state = fnvlist_lookup_uint64(tmp,
599 ZPOOL_CONFIG_POOL_STATE);
600 fnvlist_add_uint64(config,
601 ZPOOL_CONFIG_POOL_STATE, state);
602
603 hostid = 0;
604 if (nvlist_lookup_uint64(tmp,
605 ZPOOL_CONFIG_HOSTID, &hostid) == 0) {
606 fnvlist_add_uint64(config,
607 ZPOOL_CONFIG_HOSTID, hostid);
608 hostname = fnvlist_lookup_string(tmp,
609 ZPOOL_CONFIG_HOSTNAME);
610 fnvlist_add_string(config,
611 ZPOOL_CONFIG_HOSTNAME, hostname);
612 }
613
614 config_seen = B_TRUE;
615 }
616
617 /*
618 * Add this top-level vdev to the child array.
619 */
620 verify(nvlist_lookup_nvlist(tmp,
621 ZPOOL_CONFIG_VDEV_TREE, &nvtop) == 0);
622 verify(nvlist_lookup_uint64(nvtop, ZPOOL_CONFIG_ID,
623 &id) == 0);
624
625 if (id >= children) {
626 nvlist_t **newchild;
627
628 newchild = zutil_alloc(hdl, (id + 1) *
629 sizeof (nvlist_t *));
630 if (newchild == NULL)
631 goto nomem;
632
633 for (c = 0; c < children; c++)
634 newchild[c] = child[c];
635
636 free(child);
637 child = newchild;
638 children = id + 1;
639 }
640 if (nvlist_dup(nvtop, &child[id], 0) != 0)
641 goto nomem;
642
643 }
644
645 /*
646 * If we have information about all the top-levels then
647 * clean up the nvlist which we've constructed. This
648 * means removing any extraneous devices that are
649 * beyond the valid range or adding devices to the end
650 * of our array which appear to be missing.
651 */
652 if (valid_top_config) {
653 if (max_id < children) {
654 for (c = max_id; c < children; c++)
655 nvlist_free(child[c]);
656 children = max_id;
657 } else if (max_id > children) {
658 nvlist_t **newchild;
659
660 newchild = zutil_alloc(hdl, (max_id) *
661 sizeof (nvlist_t *));
662 if (newchild == NULL)
663 goto nomem;
664
665 for (c = 0; c < children; c++)
666 newchild[c] = child[c];
667
668 free(child);
669 child = newchild;
670 children = max_id;
671 }
672 }
673
674 verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
675 &guid) == 0);
676
677 /*
678 * The vdev namespace may contain holes as a result of
679 * device removal. We must add them back into the vdev
680 * tree before we process any missing devices.
681 */
682 if (holes > 0) {
683 ASSERT(valid_top_config);
684
685 for (c = 0; c < children; c++) {
686 nvlist_t *holey;
687
688 if (child[c] != NULL ||
689 !vdev_is_hole(hole_array, holes, c))
690 continue;
691
692 if (nvlist_alloc(&holey, NV_UNIQUE_NAME,
693 0) != 0)
694 goto nomem;
695
696 /*
697 * Holes in the namespace are treated as
698 * "hole" top-level vdevs and have a
699 * special flag set on them.
700 */
701 if (nvlist_add_string(holey,
702 ZPOOL_CONFIG_TYPE,
703 VDEV_TYPE_HOLE) != 0 ||
704 nvlist_add_uint64(holey,
705 ZPOOL_CONFIG_ID, c) != 0 ||
706 nvlist_add_uint64(holey,
707 ZPOOL_CONFIG_GUID, 0ULL) != 0) {
708 nvlist_free(holey);
709 goto nomem;
710 }
711 child[c] = holey;
712 }
713 }
714
715 /*
716 * Look for any missing top-level vdevs. If this is the case,
717 * create a faked up 'missing' vdev as a placeholder. We cannot
718 * simply compress the child array, because the kernel performs
719 * certain checks to make sure the vdev IDs match their location
720 * in the configuration.
721 */
722 for (c = 0; c < children; c++) {
723 if (child[c] == NULL) {
724 nvlist_t *missing;
725 if (nvlist_alloc(&missing, NV_UNIQUE_NAME,
726 0) != 0)
727 goto nomem;
728 if (nvlist_add_string(missing,
729 ZPOOL_CONFIG_TYPE,
730 VDEV_TYPE_MISSING) != 0 ||
731 nvlist_add_uint64(missing,
732 ZPOOL_CONFIG_ID, c) != 0 ||
733 nvlist_add_uint64(missing,
734 ZPOOL_CONFIG_GUID, 0ULL) != 0) {
735 nvlist_free(missing);
736 goto nomem;
737 }
738 child[c] = missing;
739 }
740 }
741
742 /*
743 * Put all of this pool's top-level vdevs into a root vdev.
744 */
745 if (nvlist_alloc(&nvroot, NV_UNIQUE_NAME, 0) != 0)
746 goto nomem;
747 if (nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
748 VDEV_TYPE_ROOT) != 0 ||
749 nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) != 0 ||
750 nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, guid) != 0 ||
751 nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
752 child, children) != 0) {
753 nvlist_free(nvroot);
754 goto nomem;
755 }
756
757 for (c = 0; c < children; c++)
758 nvlist_free(child[c]);
759 free(child);
760 children = 0;
761 child = NULL;
762
763 /*
764 * Go through and fix up any paths and/or devids based on our
765 * known list of vdev GUID -> path mappings.
766 */
767 if (fix_paths(hdl, nvroot, pl->names) != 0) {
768 nvlist_free(nvroot);
769 goto nomem;
770 }
771
772 /*
773 * Add the root vdev to this pool's configuration.
774 */
775 if (nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
776 nvroot) != 0) {
777 nvlist_free(nvroot);
778 goto nomem;
779 }
780 nvlist_free(nvroot);
781
782 /*
783 * zdb uses this path to report on active pools that were
784 * imported or created using -R.
785 */
786 if (active_ok)
787 goto add_pool;
788
789 /*
790 * Determine if this pool is currently active, in which case we
791 * can't actually import it.
792 */
793 verify(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
794 &name) == 0);
795 verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
796 &guid) == 0);
797
798 if (zutil_pool_active(hdl, name, guid, &isactive) != 0)
799 goto error;
800
801 if (isactive) {
802 nvlist_free(config);
803 config = NULL;
804 continue;
805 }
806
807 if (policy != NULL) {
808 if (nvlist_add_nvlist(config, ZPOOL_LOAD_POLICY,
809 policy) != 0)
810 goto nomem;
811 }
812
813 if ((nvl = zutil_refresh_config(hdl, config)) == NULL) {
814 nvlist_free(config);
815 config = NULL;
816 continue;
817 }
818
819 nvlist_free(config);
820 config = nvl;
821
822 /*
823 * Go through and update the paths for spares, now that we have
824 * them.
825 */
826 verify(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
827 &nvroot) == 0);
828 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
829 &spares, &nspares) == 0) {
830 for (i = 0; i < nspares; i++) {
831 if (fix_paths(hdl, spares[i], pl->names) != 0)
832 goto nomem;
833 }
834 }
835
836 /*
837 * Update the paths for l2cache devices.
838 */
839 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
840 &l2cache, &nl2cache) == 0) {
841 for (i = 0; i < nl2cache; i++) {
842 if (fix_paths(hdl, l2cache[i], pl->names) != 0)
843 goto nomem;
844 }
845 }
846
847 /*
848 * Restore the original information read from the actual label.
849 */
850 (void) nvlist_remove(config, ZPOOL_CONFIG_HOSTID,
851 DATA_TYPE_UINT64);
852 (void) nvlist_remove(config, ZPOOL_CONFIG_HOSTNAME,
853 DATA_TYPE_STRING);
854 if (hostid != 0) {
855 verify(nvlist_add_uint64(config, ZPOOL_CONFIG_HOSTID,
856 hostid) == 0);
857 verify(nvlist_add_string(config, ZPOOL_CONFIG_HOSTNAME,
858 hostname) == 0);
859 }
860
861 add_pool:
862 /*
863 * Add this pool to the list of configs.
864 */
865 verify(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
866 &name) == 0);
867
868 if (nvlist_add_nvlist(ret, name, config) != 0)
869 goto nomem;
870
871 nvlist_free(config);
872 config = NULL;
873 }
874
875 return (ret);
876
877 nomem:
878 (void) zutil_no_memory(hdl);
879 error:
880 nvlist_free(config);
881 nvlist_free(ret);
882 for (c = 0; c < children; c++)
883 nvlist_free(child[c]);
884 free(child);
885
886 return (NULL);
887 }
888
889 /*
890 * Return the offset of the given label.
891 */
892 static uint64_t
893 label_offset(uint64_t size, int l)
894 {
895 ASSERT(P2PHASE_TYPED(size, sizeof (vdev_label_t), uint64_t) == 0);
896 return (l * sizeof (vdev_label_t) + (l < VDEV_LABELS / 2 ?
897 0 : size - VDEV_LABELS * sizeof (vdev_label_t)));
898 }
899
900 /*
901 * The same description applies as to zpool_read_label below,
902 * except here we do it without aio, presumably because an aio call
903 * errored out in a way we think not using it could circumvent.
904 */
905 static int
906 zpool_read_label_slow(int fd, nvlist_t **config, int *num_labels)
907 {
908 struct stat64 statbuf;
909 int l, count = 0;
910 vdev_phys_t *label;
911 nvlist_t *expected_config = NULL;
912 uint64_t expected_guid = 0, size;
913 int error;
914
915 *config = NULL;
916
917 if (fstat64_blk(fd, &statbuf) == -1)
918 return (0);
919 size = P2ALIGN_TYPED(statbuf.st_size, sizeof (vdev_label_t), uint64_t);
920
921 error = posix_memalign((void **)&label, PAGESIZE, sizeof (*label));
922 if (error)
923 return (-1);
924
925 for (l = 0; l < VDEV_LABELS; l++) {
926 uint64_t state, guid, txg;
927 off_t offset = label_offset(size, l) + VDEV_SKIP_SIZE;
928
929 if (pread64(fd, label, sizeof (vdev_phys_t),
930 offset) != sizeof (vdev_phys_t))
931 continue;
932
933 if (nvlist_unpack(label->vp_nvlist,
934 sizeof (label->vp_nvlist), config, 0) != 0)
935 continue;
936
937 if (nvlist_lookup_uint64(*config, ZPOOL_CONFIG_GUID,
938 &guid) != 0 || guid == 0) {
939 nvlist_free(*config);
940 continue;
941 }
942
943 if (nvlist_lookup_uint64(*config, ZPOOL_CONFIG_POOL_STATE,
944 &state) != 0 || state > POOL_STATE_L2CACHE) {
945 nvlist_free(*config);
946 continue;
947 }
948
949 if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
950 (nvlist_lookup_uint64(*config, ZPOOL_CONFIG_POOL_TXG,
951 &txg) != 0 || txg == 0)) {
952 nvlist_free(*config);
953 continue;
954 }
955
956 if (expected_guid) {
957 if (expected_guid == guid)
958 count++;
959
960 nvlist_free(*config);
961 } else {
962 expected_config = *config;
963 expected_guid = guid;
964 count++;
965 }
966 }
967
968 if (num_labels != NULL)
969 *num_labels = count;
970
971 free(label);
972 *config = expected_config;
973
974 return (0);
975 }
976
977 /*
978 * Given a file descriptor, read the label information and return an nvlist
979 * describing the configuration, if there is one. The number of valid
980 * labels found will be returned in num_labels when non-NULL.
981 */
982 int
983 zpool_read_label(int fd, nvlist_t **config, int *num_labels)
984 {
985 struct stat64 statbuf;
986 struct aiocb aiocbs[VDEV_LABELS];
987 struct aiocb *aiocbps[VDEV_LABELS];
988 vdev_phys_t *labels;
989 nvlist_t *expected_config = NULL;
990 uint64_t expected_guid = 0, size;
991 int error, l, count = 0;
992
993 *config = NULL;
994
995 if (fstat64_blk(fd, &statbuf) == -1)
996 return (0);
997 size = P2ALIGN_TYPED(statbuf.st_size, sizeof (vdev_label_t), uint64_t);
998
999 error = posix_memalign((void **)&labels, PAGESIZE,
1000 VDEV_LABELS * sizeof (*labels));
1001 if (error)
1002 return (-1);
1003
1004 memset(aiocbs, 0, sizeof (aiocbs));
1005 for (l = 0; l < VDEV_LABELS; l++) {
1006 off_t offset = label_offset(size, l) + VDEV_SKIP_SIZE;
1007
1008 aiocbs[l].aio_fildes = fd;
1009 aiocbs[l].aio_offset = offset;
1010 aiocbs[l].aio_buf = &labels[l];
1011 aiocbs[l].aio_nbytes = sizeof (vdev_phys_t);
1012 aiocbs[l].aio_lio_opcode = LIO_READ;
1013 aiocbps[l] = &aiocbs[l];
1014 }
1015
1016 if (lio_listio(LIO_WAIT, aiocbps, VDEV_LABELS, NULL) != 0) {
1017 int saved_errno = errno;
1018 boolean_t do_slow = B_FALSE;
1019 error = -1;
1020
1021 if (errno == EAGAIN || errno == EINTR || errno == EIO) {
1022 /*
1023 * A portion of the requests may have been submitted.
1024 * Clean them up.
1025 */
1026 for (l = 0; l < VDEV_LABELS; l++) {
1027 errno = 0;
1028 switch (aio_error(&aiocbs[l])) {
1029 case EINVAL:
1030 break;
1031 case EINPROGRESS:
1032 // This shouldn't be possible to
1033 // encounter, die if we do.
1034 ASSERT(B_FALSE);
1035 fallthrough;
1036 case EOPNOTSUPP:
1037 case ENOSYS:
1038 do_slow = B_TRUE;
1039 fallthrough;
1040 case 0:
1041 default:
1042 (void) aio_return(&aiocbs[l]);
1043 }
1044 }
1045 }
1046 if (do_slow) {
1047 /*
1048 * At least some IO involved access unsafe-for-AIO
1049 * files. Let's try again, without AIO this time.
1050 */
1051 error = zpool_read_label_slow(fd, config, num_labels);
1052 saved_errno = errno;
1053 }
1054 free(labels);
1055 errno = saved_errno;
1056 return (error);
1057 }
1058
1059 for (l = 0; l < VDEV_LABELS; l++) {
1060 uint64_t state, guid, txg;
1061
1062 if (aio_return(&aiocbs[l]) != sizeof (vdev_phys_t))
1063 continue;
1064
1065 if (nvlist_unpack(labels[l].vp_nvlist,
1066 sizeof (labels[l].vp_nvlist), config, 0) != 0)
1067 continue;
1068
1069 if (nvlist_lookup_uint64(*config, ZPOOL_CONFIG_GUID,
1070 &guid) != 0 || guid == 0) {
1071 nvlist_free(*config);
1072 continue;
1073 }
1074
1075 if (nvlist_lookup_uint64(*config, ZPOOL_CONFIG_POOL_STATE,
1076 &state) != 0 || state > POOL_STATE_L2CACHE) {
1077 nvlist_free(*config);
1078 continue;
1079 }
1080
1081 if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
1082 (nvlist_lookup_uint64(*config, ZPOOL_CONFIG_POOL_TXG,
1083 &txg) != 0 || txg == 0)) {
1084 nvlist_free(*config);
1085 continue;
1086 }
1087
1088 if (expected_guid) {
1089 if (expected_guid == guid)
1090 count++;
1091
1092 nvlist_free(*config);
1093 } else {
1094 expected_config = *config;
1095 expected_guid = guid;
1096 count++;
1097 }
1098 }
1099
1100 if (num_labels != NULL)
1101 *num_labels = count;
1102
1103 free(labels);
1104 *config = expected_config;
1105
1106 return (0);
1107 }
1108
1109 /*
1110 * Sorted by full path and then vdev guid to allow for multiple entries with
1111 * the same full path name. This is required because it's possible to
1112 * have multiple block devices with labels that refer to the same
1113 * ZPOOL_CONFIG_PATH yet have different vdev guids. In this case both
1114 * entries need to be added to the cache. Scenarios where this can occur
1115 * include overwritten pool labels, devices which are visible from multiple
1116 * hosts and multipath devices.
1117 */
1118 int
1119 slice_cache_compare(const void *arg1, const void *arg2)
1120 {
1121 const char *nm1 = ((rdsk_node_t *)arg1)->rn_name;
1122 const char *nm2 = ((rdsk_node_t *)arg2)->rn_name;
1123 uint64_t guid1 = ((rdsk_node_t *)arg1)->rn_vdev_guid;
1124 uint64_t guid2 = ((rdsk_node_t *)arg2)->rn_vdev_guid;
1125 int rv;
1126
1127 rv = TREE_ISIGN(strcmp(nm1, nm2));
1128 if (rv)
1129 return (rv);
1130
1131 return (TREE_CMP(guid1, guid2));
1132 }
1133
1134 static int
1135 label_paths_impl(libpc_handle_t *hdl, nvlist_t *nvroot, uint64_t pool_guid,
1136 uint64_t vdev_guid, char **path, char **devid)
1137 {
1138 nvlist_t **child;
1139 uint_t c, children;
1140 uint64_t guid;
1141 char *val;
1142 int error;
1143
1144 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
1145 &child, &children) == 0) {
1146 for (c = 0; c < children; c++) {
1147 error = label_paths_impl(hdl, child[c],
1148 pool_guid, vdev_guid, path, devid);
1149 if (error)
1150 return (error);
1151 }
1152 return (0);
1153 }
1154
1155 if (nvroot == NULL)
1156 return (0);
1157
1158 error = nvlist_lookup_uint64(nvroot, ZPOOL_CONFIG_GUID, &guid);
1159 if ((error != 0) || (guid != vdev_guid))
1160 return (0);
1161
1162 error = nvlist_lookup_string(nvroot, ZPOOL_CONFIG_PATH, &val);
1163 if (error == 0)
1164 *path = val;
1165
1166 error = nvlist_lookup_string(nvroot, ZPOOL_CONFIG_DEVID, &val);
1167 if (error == 0)
1168 *devid = val;
1169
1170 return (0);
1171 }
1172
1173 /*
1174 * Given a disk label fetch the ZPOOL_CONFIG_PATH and ZPOOL_CONFIG_DEVID
1175 * and store these strings as config_path and devid_path respectively.
1176 * The returned pointers are only valid as long as label remains valid.
1177 */
1178 int
1179 label_paths(libpc_handle_t *hdl, nvlist_t *label, char **path, char **devid)
1180 {
1181 nvlist_t *nvroot;
1182 uint64_t pool_guid;
1183 uint64_t vdev_guid;
1184
1185 *path = NULL;
1186 *devid = NULL;
1187
1188 if (nvlist_lookup_nvlist(label, ZPOOL_CONFIG_VDEV_TREE, &nvroot) ||
1189 nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_GUID, &pool_guid) ||
1190 nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, &vdev_guid))
1191 return (ENOENT);
1192
1193 return (label_paths_impl(hdl, nvroot, pool_guid, vdev_guid, path,
1194 devid));
1195 }
1196
1197 static void
1198 zpool_find_import_scan_add_slice(libpc_handle_t *hdl, pthread_mutex_t *lock,
1199 avl_tree_t *cache, const char *path, const char *name, int order)
1200 {
1201 avl_index_t where;
1202 rdsk_node_t *slice;
1203
1204 slice = zutil_alloc(hdl, sizeof (rdsk_node_t));
1205 if (asprintf(&slice->rn_name, "%s/%s", path, name) == -1) {
1206 free(slice);
1207 return;
1208 }
1209 slice->rn_vdev_guid = 0;
1210 slice->rn_lock = lock;
1211 slice->rn_avl = cache;
1212 slice->rn_hdl = hdl;
1213 slice->rn_order = order + IMPORT_ORDER_SCAN_OFFSET;
1214 slice->rn_labelpaths = B_FALSE;
1215
1216 pthread_mutex_lock(lock);
1217 if (avl_find(cache, slice, &where)) {
1218 free(slice->rn_name);
1219 free(slice);
1220 } else {
1221 avl_insert(cache, slice, where);
1222 }
1223 pthread_mutex_unlock(lock);
1224 }
1225
1226 static int
1227 zpool_find_import_scan_dir(libpc_handle_t *hdl, pthread_mutex_t *lock,
1228 avl_tree_t *cache, const char *dir, int order)
1229 {
1230 int error;
1231 char path[MAXPATHLEN];
1232 struct dirent64 *dp;
1233 DIR *dirp;
1234
1235 if (realpath(dir, path) == NULL) {
1236 error = errno;
1237 if (error == ENOENT)
1238 return (0);
1239
1240 zutil_error_aux(hdl, "%s", strerror(error));
1241 (void) zutil_error_fmt(hdl, EZFS_BADPATH, dgettext(
1242 TEXT_DOMAIN, "cannot resolve path '%s'"), dir);
1243 return (error);
1244 }
1245
1246 dirp = opendir(path);
1247 if (dirp == NULL) {
1248 error = errno;
1249 zutil_error_aux(hdl, "%s", strerror(error));
1250 (void) zutil_error_fmt(hdl, EZFS_BADPATH,
1251 dgettext(TEXT_DOMAIN, "cannot open '%s'"), path);
1252 return (error);
1253 }
1254
1255 while ((dp = readdir64(dirp)) != NULL) {
1256 const char *name = dp->d_name;
1257 if (strcmp(name, ".") == 0 || strcmp(name, "..") == 0)
1258 continue;
1259
1260 switch (dp->d_type) {
1261 case DT_UNKNOWN:
1262 case DT_BLK:
1263 case DT_LNK:
1264 #ifdef __FreeBSD__
1265 case DT_CHR:
1266 #endif
1267 case DT_REG:
1268 break;
1269 default:
1270 continue;
1271 }
1272
1273 zpool_find_import_scan_add_slice(hdl, lock, cache, path, name,
1274 order);
1275 }
1276
1277 (void) closedir(dirp);
1278 return (0);
1279 }
1280
1281 static int
1282 zpool_find_import_scan_path(libpc_handle_t *hdl, pthread_mutex_t *lock,
1283 avl_tree_t *cache, const char *dir, int order)
1284 {
1285 int error = 0;
1286 char path[MAXPATHLEN];
1287 char *d = NULL;
1288 ssize_t dl;
1289 const char *dpath, *name;
1290
1291 /*
1292 * Separate the directory and the basename.
1293 * We do this so that we can get the realpath of
1294 * the directory. We don't get the realpath on the
1295 * whole path because if it's a symlink, we want the
1296 * path of the symlink not where it points to.
1297 */
1298 name = zfs_basename(dir);
1299 if ((dl = zfs_dirnamelen(dir)) == -1)
1300 dpath = ".";
1301 else
1302 dpath = d = zutil_strndup(hdl, dir, dl);
1303
1304 if (realpath(dpath, path) == NULL) {
1305 error = errno;
1306 if (error == ENOENT) {
1307 error = 0;
1308 goto out;
1309 }
1310
1311 zutil_error_aux(hdl, "%s", strerror(error));
1312 (void) zutil_error_fmt(hdl, EZFS_BADPATH, dgettext(
1313 TEXT_DOMAIN, "cannot resolve path '%s'"), dir);
1314 goto out;
1315 }
1316
1317 zpool_find_import_scan_add_slice(hdl, lock, cache, path, name, order);
1318
1319 out:
1320 free(d);
1321 return (error);
1322 }
1323
1324 /*
1325 * Scan a list of directories for zfs devices.
1326 */
1327 static int
1328 zpool_find_import_scan(libpc_handle_t *hdl, pthread_mutex_t *lock,
1329 avl_tree_t **slice_cache, const char * const *dir, size_t dirs)
1330 {
1331 avl_tree_t *cache;
1332 rdsk_node_t *slice;
1333 void *cookie;
1334 int i, error;
1335
1336 *slice_cache = NULL;
1337 cache = zutil_alloc(hdl, sizeof (avl_tree_t));
1338 avl_create(cache, slice_cache_compare, sizeof (rdsk_node_t),
1339 offsetof(rdsk_node_t, rn_node));
1340
1341 for (i = 0; i < dirs; i++) {
1342 struct stat sbuf;
1343
1344 if (stat(dir[i], &sbuf) != 0) {
1345 error = errno;
1346 if (error == ENOENT)
1347 continue;
1348
1349 zutil_error_aux(hdl, "%s", strerror(error));
1350 (void) zutil_error_fmt(hdl, EZFS_BADPATH, dgettext(
1351 TEXT_DOMAIN, "cannot resolve path '%s'"), dir[i]);
1352 goto error;
1353 }
1354
1355 /*
1356 * If dir[i] is a directory, we walk through it and add all
1357 * the entries to the cache. If it's not a directory, we just
1358 * add it to the cache.
1359 */
1360 if (S_ISDIR(sbuf.st_mode)) {
1361 if ((error = zpool_find_import_scan_dir(hdl, lock,
1362 cache, dir[i], i)) != 0)
1363 goto error;
1364 } else {
1365 if ((error = zpool_find_import_scan_path(hdl, lock,
1366 cache, dir[i], i)) != 0)
1367 goto error;
1368 }
1369 }
1370
1371 *slice_cache = cache;
1372 return (0);
1373
1374 error:
1375 cookie = NULL;
1376 while ((slice = avl_destroy_nodes(cache, &cookie)) != NULL) {
1377 free(slice->rn_name);
1378 free(slice);
1379 }
1380 free(cache);
1381
1382 return (error);
1383 }
1384
1385 /*
1386 * Given a list of directories to search, find all pools stored on disk. This
1387 * includes partial pools which are not available to import. If no args are
1388 * given (argc is 0), then the default directory (/dev/dsk) is searched.
1389 * poolname or guid (but not both) are provided by the caller when trying
1390 * to import a specific pool.
1391 */
1392 static nvlist_t *
1393 zpool_find_import_impl(libpc_handle_t *hdl, importargs_t *iarg,
1394 pthread_mutex_t *lock, avl_tree_t *cache)
1395 {
1396 nvlist_t *ret = NULL;
1397 pool_list_t pools = { 0 };
1398 pool_entry_t *pe, *penext;
1399 vdev_entry_t *ve, *venext;
1400 config_entry_t *ce, *cenext;
1401 name_entry_t *ne, *nenext;
1402 rdsk_node_t *slice;
1403 void *cookie;
1404 tpool_t *t;
1405
1406 verify(iarg->poolname == NULL || iarg->guid == 0);
1407
1408 /*
1409 * Create a thread pool to parallelize the process of reading and
1410 * validating labels, a large number of threads can be used due to
1411 * minimal contention.
1412 */
1413 t = tpool_create(1, 2 * sysconf(_SC_NPROCESSORS_ONLN), 0, NULL);
1414 for (slice = avl_first(cache); slice;
1415 (slice = avl_walk(cache, slice, AVL_AFTER)))
1416 (void) tpool_dispatch(t, zpool_open_func, slice);
1417
1418 tpool_wait(t);
1419 tpool_destroy(t);
1420
1421 /*
1422 * Process the cache, filtering out any entries which are not
1423 * for the specified pool then adding matching label configs.
1424 */
1425 cookie = NULL;
1426 while ((slice = avl_destroy_nodes(cache, &cookie)) != NULL) {
1427 if (slice->rn_config != NULL) {
1428 nvlist_t *config = slice->rn_config;
1429 boolean_t matched = B_TRUE;
1430 boolean_t aux = B_FALSE;
1431 int fd;
1432
1433 /*
1434 * Check if it's a spare or l2cache device. If it is,
1435 * we need to skip the name and guid check since they
1436 * don't exist on aux device label.
1437 */
1438 if (iarg->poolname != NULL || iarg->guid != 0) {
1439 uint64_t state;
1440 aux = nvlist_lookup_uint64(config,
1441 ZPOOL_CONFIG_POOL_STATE, &state) == 0 &&
1442 (state == POOL_STATE_SPARE ||
1443 state == POOL_STATE_L2CACHE);
1444 }
1445
1446 if (iarg->poolname != NULL && !aux) {
1447 char *pname;
1448
1449 matched = nvlist_lookup_string(config,
1450 ZPOOL_CONFIG_POOL_NAME, &pname) == 0 &&
1451 strcmp(iarg->poolname, pname) == 0;
1452 } else if (iarg->guid != 0 && !aux) {
1453 uint64_t this_guid;
1454
1455 matched = nvlist_lookup_uint64(config,
1456 ZPOOL_CONFIG_POOL_GUID, &this_guid) == 0 &&
1457 iarg->guid == this_guid;
1458 }
1459 if (matched) {
1460 /*
1461 * Verify all remaining entries can be opened
1462 * exclusively. This will prune all underlying
1463 * multipath devices which otherwise could
1464 * result in the vdev appearing as UNAVAIL.
1465 *
1466 * Under zdb, this step isn't required and
1467 * would prevent a zdb -e of active pools with
1468 * no cachefile.
1469 */
1470 fd = open(slice->rn_name,
1471 O_RDONLY | O_EXCL | O_CLOEXEC);
1472 if (fd >= 0 || iarg->can_be_active) {
1473 if (fd >= 0)
1474 close(fd);
1475 add_config(hdl, &pools,
1476 slice->rn_name, slice->rn_order,
1477 slice->rn_num_labels, config);
1478 }
1479 }
1480 nvlist_free(config);
1481 }
1482 free(slice->rn_name);
1483 free(slice);
1484 }
1485 avl_destroy(cache);
1486 free(cache);
1487
1488 ret = get_configs(hdl, &pools, iarg->can_be_active, iarg->policy);
1489
1490 for (pe = pools.pools; pe != NULL; pe = penext) {
1491 penext = pe->pe_next;
1492 for (ve = pe->pe_vdevs; ve != NULL; ve = venext) {
1493 venext = ve->ve_next;
1494 for (ce = ve->ve_configs; ce != NULL; ce = cenext) {
1495 cenext = ce->ce_next;
1496 nvlist_free(ce->ce_config);
1497 free(ce);
1498 }
1499 free(ve);
1500 }
1501 free(pe);
1502 }
1503
1504 for (ne = pools.names; ne != NULL; ne = nenext) {
1505 nenext = ne->ne_next;
1506 free(ne->ne_name);
1507 free(ne);
1508 }
1509
1510 return (ret);
1511 }
1512
1513 /*
1514 * Given a config, discover the paths for the devices which
1515 * exist in the config.
1516 */
1517 static int
1518 discover_cached_paths(libpc_handle_t *hdl, nvlist_t *nv,
1519 avl_tree_t *cache, pthread_mutex_t *lock)
1520 {
1521 char *path = NULL;
1522 ssize_t dl;
1523 uint_t children;
1524 nvlist_t **child;
1525
1526 if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
1527 &child, &children) == 0) {
1528 for (int c = 0; c < children; c++) {
1529 discover_cached_paths(hdl, child[c], cache, lock);
1530 }
1531 }
1532
1533 /*
1534 * Once we have the path, we need to add the directory to
1535 * our directory cache.
1536 */
1537 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &path) == 0) {
1538 if ((dl = zfs_dirnamelen(path)) == -1)
1539 path = ".";
1540 else
1541 path[dl] = '\0';
1542 return (zpool_find_import_scan_dir(hdl, lock, cache,
1543 path, 0));
1544 }
1545 return (0);
1546 }
1547
1548 /*
1549 * Given a cache file, return the contents as a list of importable pools.
1550 * poolname or guid (but not both) are provided by the caller when trying
1551 * to import a specific pool.
1552 */
1553 static nvlist_t *
1554 zpool_find_import_cached(libpc_handle_t *hdl, importargs_t *iarg)
1555 {
1556 char *buf;
1557 int fd;
1558 struct stat64 statbuf;
1559 nvlist_t *raw, *src, *dst;
1560 nvlist_t *pools;
1561 nvpair_t *elem;
1562 char *name;
1563 uint64_t this_guid;
1564 boolean_t active;
1565
1566 verify(iarg->poolname == NULL || iarg->guid == 0);
1567
1568 if ((fd = open(iarg->cachefile, O_RDONLY | O_CLOEXEC)) < 0) {
1569 zutil_error_aux(hdl, "%s", strerror(errno));
1570 (void) zutil_error(hdl, EZFS_BADCACHE,
1571 dgettext(TEXT_DOMAIN, "failed to open cache file"));
1572 return (NULL);
1573 }
1574
1575 if (fstat64(fd, &statbuf) != 0) {
1576 zutil_error_aux(hdl, "%s", strerror(errno));
1577 (void) close(fd);
1578 (void) zutil_error(hdl, EZFS_BADCACHE,
1579 dgettext(TEXT_DOMAIN, "failed to get size of cache file"));
1580 return (NULL);
1581 }
1582
1583 if ((buf = zutil_alloc(hdl, statbuf.st_size)) == NULL) {
1584 (void) close(fd);
1585 return (NULL);
1586 }
1587
1588 if (read(fd, buf, statbuf.st_size) != statbuf.st_size) {
1589 (void) close(fd);
1590 free(buf);
1591 (void) zutil_error(hdl, EZFS_BADCACHE,
1592 dgettext(TEXT_DOMAIN,
1593 "failed to read cache file contents"));
1594 return (NULL);
1595 }
1596
1597 (void) close(fd);
1598
1599 if (nvlist_unpack(buf, statbuf.st_size, &raw, 0) != 0) {
1600 free(buf);
1601 (void) zutil_error(hdl, EZFS_BADCACHE,
1602 dgettext(TEXT_DOMAIN,
1603 "invalid or corrupt cache file contents"));
1604 return (NULL);
1605 }
1606
1607 free(buf);
1608
1609 /*
1610 * Go through and get the current state of the pools and refresh their
1611 * state.
1612 */
1613 if (nvlist_alloc(&pools, 0, 0) != 0) {
1614 (void) zutil_no_memory(hdl);
1615 nvlist_free(raw);
1616 return (NULL);
1617 }
1618
1619 elem = NULL;
1620 while ((elem = nvlist_next_nvpair(raw, elem)) != NULL) {
1621 src = fnvpair_value_nvlist(elem);
1622
1623 name = fnvlist_lookup_string(src, ZPOOL_CONFIG_POOL_NAME);
1624 if (iarg->poolname != NULL && strcmp(iarg->poolname, name) != 0)
1625 continue;
1626
1627 this_guid = fnvlist_lookup_uint64(src, ZPOOL_CONFIG_POOL_GUID);
1628 if (iarg->guid != 0 && iarg->guid != this_guid)
1629 continue;
1630
1631 if (zutil_pool_active(hdl, name, this_guid, &active) != 0) {
1632 nvlist_free(raw);
1633 nvlist_free(pools);
1634 return (NULL);
1635 }
1636
1637 if (active)
1638 continue;
1639
1640 if (iarg->scan) {
1641 uint64_t saved_guid = iarg->guid;
1642 const char *saved_poolname = iarg->poolname;
1643 pthread_mutex_t lock;
1644
1645 /*
1646 * Create the device cache that will hold the
1647 * devices we will scan based on the cachefile.
1648 * This will get destroyed and freed by
1649 * zpool_find_import_impl.
1650 */
1651 avl_tree_t *cache = zutil_alloc(hdl,
1652 sizeof (avl_tree_t));
1653 avl_create(cache, slice_cache_compare,
1654 sizeof (rdsk_node_t),
1655 offsetof(rdsk_node_t, rn_node));
1656 nvlist_t *nvroot = fnvlist_lookup_nvlist(src,
1657 ZPOOL_CONFIG_VDEV_TREE);
1658
1659 /*
1660 * We only want to find the pool with this_guid.
1661 * We will reset these values back later.
1662 */
1663 iarg->guid = this_guid;
1664 iarg->poolname = NULL;
1665
1666 /*
1667 * We need to build up a cache of devices that exists
1668 * in the paths pointed to by the cachefile. This allows
1669 * us to preserve the device namespace that was
1670 * originally specified by the user but also lets us
1671 * scan devices in those directories in case they had
1672 * been renamed.
1673 */
1674 pthread_mutex_init(&lock, NULL);
1675 discover_cached_paths(hdl, nvroot, cache, &lock);
1676 nvlist_t *nv = zpool_find_import_impl(hdl, iarg,
1677 &lock, cache);
1678 pthread_mutex_destroy(&lock);
1679
1680 /*
1681 * zpool_find_import_impl will return back
1682 * a list of pools that it found based on the
1683 * device cache. There should only be one pool
1684 * since we're looking for a specific guid.
1685 * We will use that pool to build up the final
1686 * pool nvlist which is returned back to the
1687 * caller.
1688 */
1689 nvpair_t *pair = nvlist_next_nvpair(nv, NULL);
1690 fnvlist_add_nvlist(pools, nvpair_name(pair),
1691 fnvpair_value_nvlist(pair));
1692
1693 VERIFY3P(nvlist_next_nvpair(nv, pair), ==, NULL);
1694
1695 iarg->guid = saved_guid;
1696 iarg->poolname = saved_poolname;
1697 continue;
1698 }
1699
1700 if (nvlist_add_string(src, ZPOOL_CONFIG_CACHEFILE,
1701 iarg->cachefile) != 0) {
1702 (void) zutil_no_memory(hdl);
1703 nvlist_free(raw);
1704 nvlist_free(pools);
1705 return (NULL);
1706 }
1707
1708 update_vdevs_config_dev_sysfs_path(src);
1709
1710 if ((dst = zutil_refresh_config(hdl, src)) == NULL) {
1711 nvlist_free(raw);
1712 nvlist_free(pools);
1713 return (NULL);
1714 }
1715
1716 if (nvlist_add_nvlist(pools, nvpair_name(elem), dst) != 0) {
1717 (void) zutil_no_memory(hdl);
1718 nvlist_free(dst);
1719 nvlist_free(raw);
1720 nvlist_free(pools);
1721 return (NULL);
1722 }
1723 nvlist_free(dst);
1724 }
1725 nvlist_free(raw);
1726 return (pools);
1727 }
1728
1729 static nvlist_t *
1730 zpool_find_import(libpc_handle_t *hdl, importargs_t *iarg)
1731 {
1732 pthread_mutex_t lock;
1733 avl_tree_t *cache;
1734 nvlist_t *pools = NULL;
1735
1736 verify(iarg->poolname == NULL || iarg->guid == 0);
1737 pthread_mutex_init(&lock, NULL);
1738
1739 /*
1740 * Locate pool member vdevs by blkid or by directory scanning.
1741 * On success a newly allocated AVL tree which is populated with an
1742 * entry for each discovered vdev will be returned in the cache.
1743 * It's the caller's responsibility to consume and destroy this tree.
1744 */
1745 if (iarg->scan || iarg->paths != 0) {
1746 size_t dirs = iarg->paths;
1747 const char * const *dir = (const char * const *)iarg->path;
1748
1749 if (dirs == 0)
1750 dir = zpool_default_search_paths(&dirs);
1751
1752 if (zpool_find_import_scan(hdl, &lock, &cache,
1753 dir, dirs) != 0) {
1754 pthread_mutex_destroy(&lock);
1755 return (NULL);
1756 }
1757 } else {
1758 if (zpool_find_import_blkid(hdl, &lock, &cache) != 0) {
1759 pthread_mutex_destroy(&lock);
1760 return (NULL);
1761 }
1762 }
1763
1764 pools = zpool_find_import_impl(hdl, iarg, &lock, cache);
1765 pthread_mutex_destroy(&lock);
1766 return (pools);
1767 }
1768
1769
1770 nvlist_t *
1771 zpool_search_import(void *hdl, importargs_t *import,
1772 const pool_config_ops_t *pco)
1773 {
1774 libpc_handle_t handle = { 0 };
1775 nvlist_t *pools = NULL;
1776
1777 handle.lpc_lib_handle = hdl;
1778 handle.lpc_ops = pco;
1779 handle.lpc_printerr = B_TRUE;
1780
1781 verify(import->poolname == NULL || import->guid == 0);
1782
1783 if (import->cachefile != NULL)
1784 pools = zpool_find_import_cached(&handle, import);
1785 else
1786 pools = zpool_find_import(&handle, import);
1787
1788 if ((pools == NULL || nvlist_empty(pools)) &&
1789 handle.lpc_open_access_error && geteuid() != 0) {
1790 (void) zutil_error(&handle, EZFS_EACESS, dgettext(TEXT_DOMAIN,
1791 "no pools found"));
1792 }
1793
1794 return (pools);
1795 }
1796
1797 static boolean_t
1798 pool_match(nvlist_t *cfg, char *tgt)
1799 {
1800 uint64_t v, guid = strtoull(tgt, NULL, 0);
1801 char *s;
1802
1803 if (guid != 0) {
1804 if (nvlist_lookup_uint64(cfg, ZPOOL_CONFIG_POOL_GUID, &v) == 0)
1805 return (v == guid);
1806 } else {
1807 if (nvlist_lookup_string(cfg, ZPOOL_CONFIG_POOL_NAME, &s) == 0)
1808 return (strcmp(s, tgt) == 0);
1809 }
1810 return (B_FALSE);
1811 }
1812
1813 int
1814 zpool_find_config(void *hdl, const char *target, nvlist_t **configp,
1815 importargs_t *args, const pool_config_ops_t *pco)
1816 {
1817 nvlist_t *pools;
1818 nvlist_t *match = NULL;
1819 nvlist_t *config = NULL;
1820 char *sepp = NULL;
1821 int count = 0;
1822 char *targetdup = strdup(target);
1823
1824 *configp = NULL;
1825
1826 if ((sepp = strpbrk(targetdup, "/@")) != NULL)
1827 *sepp = '\0';
1828
1829 pools = zpool_search_import(hdl, args, pco);
1830
1831 if (pools != NULL) {
1832 nvpair_t *elem = NULL;
1833 while ((elem = nvlist_next_nvpair(pools, elem)) != NULL) {
1834 VERIFY0(nvpair_value_nvlist(elem, &config));
1835 if (pool_match(config, targetdup)) {
1836 count++;
1837 if (match != NULL) {
1838 /* multiple matches found */
1839 continue;
1840 } else {
1841 match = fnvlist_dup(config);
1842 }
1843 }
1844 }
1845 fnvlist_free(pools);
1846 }
1847
1848 if (count == 0) {
1849 free(targetdup);
1850 return (ENOENT);
1851 }
1852
1853 if (count > 1) {
1854 free(targetdup);
1855 fnvlist_free(match);
1856 return (EINVAL);
1857 }
1858
1859 *configp = match;
1860 free(targetdup);
1861
1862 return (0);
1863 }
1864
1865 /*
1866 * Internal function for iterating over the vdevs.
1867 *
1868 * For each vdev, func() will be called and will be passed 'zhp' (which is
1869 * typically the zpool_handle_t cast as a void pointer), the vdev's nvlist, and
1870 * a user-defined data pointer).
1871 *
1872 * The return values from all the func() calls will be OR'd together and
1873 * returned.
1874 */
1875 int
1876 for_each_vdev_cb(void *zhp, nvlist_t *nv, pool_vdev_iter_f func,
1877 void *data)
1878 {
1879 nvlist_t **child;
1880 uint_t c, children;
1881 int ret = 0;
1882 int i;
1883 char *type;
1884
1885 const char *list[] = {
1886 ZPOOL_CONFIG_SPARES,
1887 ZPOOL_CONFIG_L2CACHE,
1888 ZPOOL_CONFIG_CHILDREN
1889 };
1890
1891 for (i = 0; i < ARRAY_SIZE(list); i++) {
1892 if (nvlist_lookup_nvlist_array(nv, list[i], &child,
1893 &children) == 0) {
1894 for (c = 0; c < children; c++) {
1895 uint64_t ishole = 0;
1896
1897 (void) nvlist_lookup_uint64(child[c],
1898 ZPOOL_CONFIG_IS_HOLE, &ishole);
1899
1900 if (ishole)
1901 continue;
1902
1903 ret |= for_each_vdev_cb(zhp, child[c],
1904 func, data);
1905 }
1906 }
1907 }
1908
1909 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type) != 0)
1910 return (ret);
1911
1912 /* Don't run our function on root vdevs */
1913 if (strcmp(type, VDEV_TYPE_ROOT) != 0) {
1914 ret |= func(zhp, nv, data);
1915 }
1916
1917 return (ret);
1918 }
1919
1920 /*
1921 * Given an ZPOOL_CONFIG_VDEV_TREE nvpair, iterate over all the vdevs, calling
1922 * func() for each one. func() is passed the vdev's nvlist and an optional
1923 * user-defined 'data' pointer.
1924 */
1925 int
1926 for_each_vdev_in_nvlist(nvlist_t *nvroot, pool_vdev_iter_f func, void *data)
1927 {
1928 return (for_each_vdev_cb(NULL, nvroot, func, data));
1929 }
ZFS On Linux mirror for PVE