?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
zinject(8): rename "ioctl" to "flush"
[zfs.git] / lib / libzutil / zutil_import.c
blob77fa0ce38b2a11af65926338a4d6d209714b7ff9
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 https://opensource.org/licenses/CDDL-1.0.
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 #ifdef HAVE_AIO_H
51 #include <aio.h>
52 #endif
53 #include <ctype.h>
54 #include <dirent.h>
55 #include <errno.h>
56 #include <libintl.h>
57 #include <libgen.h>
58 #include <stddef.h>
59 #include <stdlib.h>
60 #include <string.h>
61 #include <sys/stat.h>
62 #include <unistd.h>
63 #include <fcntl.h>
64 #include <sys/dktp/fdisk.h>
65 #include <sys/vdev_impl.h>
66 #include <sys/fs/zfs.h>
67
68 #include <thread_pool.h>
69 #include <libzutil.h>
70 #include <libnvpair.h>
71
72 #include "zutil_import.h"
73
74 const char *
75 libpc_error_description(libpc_handle_t *hdl)
76 {
77 if (hdl->lpc_desc[0] != '\0')
78 return (hdl->lpc_desc);
79
80 switch (hdl->lpc_error) {
81 case LPC_BADCACHE:
82 return (dgettext(TEXT_DOMAIN, "invalid or missing cache file"));
83 case LPC_BADPATH:
84 return (dgettext(TEXT_DOMAIN, "must be an absolute path"));
85 case LPC_NOMEM:
86 return (dgettext(TEXT_DOMAIN, "out of memory"));
87 case LPC_EACCESS:
88 return (dgettext(TEXT_DOMAIN, "some devices require root "
89 "privileges"));
90 case LPC_UNKNOWN:
91 return (dgettext(TEXT_DOMAIN, "unknown error"));
92 default:
93 assert(hdl->lpc_error == 0);
94 return (dgettext(TEXT_DOMAIN, "no error"));
95 }
96 }
97
98 static __attribute__((format(printf, 2, 3))) void
99 zutil_error_aux(libpc_handle_t *hdl, const char *fmt, ...)
100 {
101 va_list ap;
102
103 va_start(ap, fmt);
104
105 (void) vsnprintf(hdl->lpc_desc, sizeof (hdl->lpc_desc), fmt, ap);
106 hdl->lpc_desc_active = B_TRUE;
107
108 va_end(ap);
109 }
110
111 static void
112 zutil_verror(libpc_handle_t *hdl, lpc_error_t error, const char *fmt,
113 va_list ap)
114 {
115 char action[1024];
116
117 (void) vsnprintf(action, sizeof (action), fmt, ap);
118 hdl->lpc_error = error;
119
120 if (hdl->lpc_desc_active)
121 hdl->lpc_desc_active = B_FALSE;
122 else
123 hdl->lpc_desc[0] = '\0';
124
125 if (hdl->lpc_printerr)
126 (void) fprintf(stderr, "%s: %s\n", action,
127 libpc_error_description(hdl));
128 }
129
130 static __attribute__((format(printf, 3, 4))) int
131 zutil_error_fmt(libpc_handle_t *hdl, lpc_error_t error,
132 const char *fmt, ...)
133 {
134 va_list ap;
135
136 va_start(ap, fmt);
137
138 zutil_verror(hdl, error, fmt, ap);
139
140 va_end(ap);
141
142 return (-1);
143 }
144
145 static int
146 zutil_error(libpc_handle_t *hdl, lpc_error_t error, const char *msg)
147 {
148 return (zutil_error_fmt(hdl, error, "%s", msg));
149 }
150
151 static int
152 zutil_no_memory(libpc_handle_t *hdl)
153 {
154 zutil_error(hdl, LPC_NOMEM, "internal error");
155 exit(1);
156 }
157
158 void *
159 zutil_alloc(libpc_handle_t *hdl, size_t size)
160 {
161 void *data;
162
163 if ((data = calloc(1, size)) == NULL)
164 (void) zutil_no_memory(hdl);
165
166 return (data);
167 }
168
169 char *
170 zutil_strdup(libpc_handle_t *hdl, const char *str)
171 {
172 char *ret;
173
174 if ((ret = strdup(str)) == NULL)
175 (void) zutil_no_memory(hdl);
176
177 return (ret);
178 }
179
180 static char *
181 zutil_strndup(libpc_handle_t *hdl, const char *str, size_t n)
182 {
183 char *ret;
184
185 if ((ret = strndup(str, n)) == NULL)
186 (void) zutil_no_memory(hdl);
187
188 return (ret);
189 }
190
191 /*
192 * Intermediate structures used to gather configuration information.
193 */
194 typedef struct config_entry {
195 uint64_t ce_txg;
196 nvlist_t *ce_config;
197 struct config_entry *ce_next;
198 } config_entry_t;
199
200 typedef struct vdev_entry {
201 uint64_t ve_guid;
202 config_entry_t *ve_configs;
203 struct vdev_entry *ve_next;
204 } vdev_entry_t;
205
206 typedef struct pool_entry {
207 uint64_t pe_guid;
208 vdev_entry_t *pe_vdevs;
209 struct pool_entry *pe_next;
210 } pool_entry_t;
211
212 typedef struct name_entry {
213 char *ne_name;
214 uint64_t ne_guid;
215 uint64_t ne_order;
216 uint64_t ne_num_labels;
217 struct name_entry *ne_next;
218 } name_entry_t;
219
220 typedef struct pool_list {
221 pool_entry_t *pools;
222 name_entry_t *names;
223 } pool_list_t;
224
225 /*
226 * Go through and fix up any path and/or devid information for the given vdev
227 * configuration.
228 */
229 static int
230 fix_paths(libpc_handle_t *hdl, nvlist_t *nv, name_entry_t *names)
231 {
232 nvlist_t **child;
233 uint_t c, children;
234 uint64_t guid;
235 name_entry_t *ne, *best;
236 const char *path;
237
238 if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
239 &child, &children) == 0) {
240 for (c = 0; c < children; c++)
241 if (fix_paths(hdl, child[c], names) != 0)
242 return (-1);
243 return (0);
244 }
245
246 /*
247 * This is a leaf (file or disk) vdev. In either case, go through
248 * the name list and see if we find a matching guid. If so, replace
249 * the path and see if we can calculate a new devid.
250 *
251 * There may be multiple names associated with a particular guid, in
252 * which case we have overlapping partitions or multiple paths to the
253 * same disk. In this case we prefer to use the path name which
254 * matches the ZPOOL_CONFIG_PATH. If no matching entry is found we
255 * use the lowest order device which corresponds to the first match
256 * while traversing the ZPOOL_IMPORT_PATH search path.
257 */
258 verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) == 0);
259 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &path) != 0)
260 path = NULL;
261
262 best = NULL;
263 for (ne = names; ne != NULL; ne = ne->ne_next) {
264 if (ne->ne_guid == guid) {
265 if (path == NULL) {
266 best = ne;
267 break;
268 }
269
270 if ((strlen(path) == strlen(ne->ne_name)) &&
271 strncmp(path, ne->ne_name, strlen(path)) == 0) {
272 best = ne;
273 break;
274 }
275
276 if (best == NULL) {
277 best = ne;
278 continue;
279 }
280
281 /* Prefer paths with move vdev labels. */
282 if (ne->ne_num_labels > best->ne_num_labels) {
283 best = ne;
284 continue;
285 }
286
287 /* Prefer paths earlier in the search order. */
288 if (ne->ne_num_labels == best->ne_num_labels &&
289 ne->ne_order < best->ne_order) {
290 best = ne;
291 continue;
292 }
293 }
294 }
295
296 if (best == NULL)
297 return (0);
298
299 if (nvlist_add_string(nv, ZPOOL_CONFIG_PATH, best->ne_name) != 0)
300 return (-1);
301
302 update_vdev_config_dev_strs(nv);
303
304 return (0);
305 }
306
307 /*
308 * Add the given configuration to the list of known devices.
309 */
310 static int
311 add_config(libpc_handle_t *hdl, pool_list_t *pl, const char *path,
312 int order, int num_labels, nvlist_t *config)
313 {
314 uint64_t pool_guid, vdev_guid, top_guid, txg, state;
315 pool_entry_t *pe;
316 vdev_entry_t *ve;
317 config_entry_t *ce;
318 name_entry_t *ne;
319
320 /*
321 * If this is a hot spare not currently in use or level 2 cache
322 * device, add it to the list of names to translate, but don't do
323 * anything else.
324 */
325 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_STATE,
326 &state) == 0 &&
327 (state == POOL_STATE_SPARE || state == POOL_STATE_L2CACHE) &&
328 nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID, &vdev_guid) == 0) {
329 if ((ne = zutil_alloc(hdl, sizeof (name_entry_t))) == NULL)
330 return (-1);
331
332 if ((ne->ne_name = zutil_strdup(hdl, path)) == NULL) {
333 free(ne);
334 return (-1);
335 }
336 ne->ne_guid = vdev_guid;
337 ne->ne_order = order;
338 ne->ne_num_labels = num_labels;
339 ne->ne_next = pl->names;
340 pl->names = ne;
341
342 return (0);
343 }
344
345 /*
346 * If we have a valid config but cannot read any of these fields, then
347 * it means we have a half-initialized label. In vdev_label_init()
348 * we write a label with txg == 0 so that we can identify the device
349 * in case the user refers to the same disk later on. If we fail to
350 * create the pool, we'll be left with a label in this state
351 * which should not be considered part of a valid pool.
352 */
353 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
354 &pool_guid) != 0 ||
355 nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID,
356 &vdev_guid) != 0 ||
357 nvlist_lookup_uint64(config, ZPOOL_CONFIG_TOP_GUID,
358 &top_guid) != 0 ||
359 nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG,
360 &txg) != 0 || txg == 0) {
361 return (0);
362 }
363
364 /*
365 * First, see if we know about this pool. If not, then add it to the
366 * list of known pools.
367 */
368 for (pe = pl->pools; pe != NULL; pe = pe->pe_next) {
369 if (pe->pe_guid == pool_guid)
370 break;
371 }
372
373 if (pe == NULL) {
374 if ((pe = zutil_alloc(hdl, sizeof (pool_entry_t))) == NULL) {
375 return (-1);
376 }
377 pe->pe_guid = pool_guid;
378 pe->pe_next = pl->pools;
379 pl->pools = pe;
380 }
381
382 /*
383 * Second, see if we know about this toplevel vdev. Add it if its
384 * missing.
385 */
386 for (ve = pe->pe_vdevs; ve != NULL; ve = ve->ve_next) {
387 if (ve->ve_guid == top_guid)
388 break;
389 }
390
391 if (ve == NULL) {
392 if ((ve = zutil_alloc(hdl, sizeof (vdev_entry_t))) == NULL) {
393 return (-1);
394 }
395 ve->ve_guid = top_guid;
396 ve->ve_next = pe->pe_vdevs;
397 pe->pe_vdevs = ve;
398 }
399
400 /*
401 * Third, see if we have a config with a matching transaction group. If
402 * so, then we do nothing. Otherwise, add it to the list of known
403 * configs.
404 */
405 for (ce = ve->ve_configs; ce != NULL; ce = ce->ce_next) {
406 if (ce->ce_txg == txg)
407 break;
408 }
409
410 if (ce == NULL) {
411 if ((ce = zutil_alloc(hdl, sizeof (config_entry_t))) == NULL) {
412 return (-1);
413 }
414 ce->ce_txg = txg;
415 ce->ce_config = fnvlist_dup(config);
416 ce->ce_next = ve->ve_configs;
417 ve->ve_configs = ce;
418 }
419
420 /*
421 * At this point we've successfully added our config to the list of
422 * known configs. The last thing to do is add the vdev guid -> path
423 * mappings so that we can fix up the configuration as necessary before
424 * doing the import.
425 */
426 if ((ne = zutil_alloc(hdl, sizeof (name_entry_t))) == NULL)
427 return (-1);
428
429 if ((ne->ne_name = zutil_strdup(hdl, path)) == NULL) {
430 free(ne);
431 return (-1);
432 }
433
434 ne->ne_guid = vdev_guid;
435 ne->ne_order = order;
436 ne->ne_num_labels = num_labels;
437 ne->ne_next = pl->names;
438 pl->names = ne;
439
440 return (0);
441 }
442
443 static int
444 zutil_pool_active(libpc_handle_t *hdl, const char *name, uint64_t guid,
445 boolean_t *isactive)
446 {
447 ASSERT(hdl->lpc_ops->pco_pool_active != NULL);
448
449 int error = hdl->lpc_ops->pco_pool_active(hdl->lpc_lib_handle, name,
450 guid, isactive);
451
452 return (error);
453 }
454
455 static nvlist_t *
456 zutil_refresh_config(libpc_handle_t *hdl, nvlist_t *tryconfig)
457 {
458 ASSERT(hdl->lpc_ops->pco_refresh_config != NULL);
459
460 return (hdl->lpc_ops->pco_refresh_config(hdl->lpc_lib_handle,
461 tryconfig));
462 }
463
464 /*
465 * Determine if the vdev id is a hole in the namespace.
466 */
467 static boolean_t
468 vdev_is_hole(uint64_t *hole_array, uint_t holes, uint_t id)
469 {
470 int c;
471
472 for (c = 0; c < holes; c++) {
473
474 /* Top-level is a hole */
475 if (hole_array[c] == id)
476 return (B_TRUE);
477 }
478 return (B_FALSE);
479 }
480
481 /*
482 * Convert our list of pools into the definitive set of configurations. We
483 * start by picking the best config for each toplevel vdev. Once that's done,
484 * we assemble the toplevel vdevs into a full config for the pool. We make a
485 * pass to fix up any incorrect paths, and then add it to the main list to
486 * return to the user.
487 */
488 static nvlist_t *
489 get_configs(libpc_handle_t *hdl, pool_list_t *pl, boolean_t active_ok,
490 nvlist_t *policy)
491 {
492 pool_entry_t *pe;
493 vdev_entry_t *ve;
494 config_entry_t *ce;
495 nvlist_t *ret = NULL, *config = NULL, *tmp = NULL, *nvtop, *nvroot;
496 nvlist_t **spares, **l2cache;
497 uint_t i, nspares, nl2cache;
498 boolean_t config_seen;
499 uint64_t best_txg;
500 const char *name, *hostname = NULL;
501 uint64_t guid;
502 uint_t children = 0;
503 nvlist_t **child = NULL;
504 uint64_t *hole_array, max_id;
505 uint_t c;
506 boolean_t isactive;
507 nvlist_t *nvl;
508 boolean_t valid_top_config = B_FALSE;
509
510 if (nvlist_alloc(&ret, 0, 0) != 0)
511 goto nomem;
512
513 for (pe = pl->pools; pe != NULL; pe = pe->pe_next) {
514 uint64_t id, max_txg = 0, hostid = 0;
515 uint_t holes = 0;
516
517 if (nvlist_alloc(&config, NV_UNIQUE_NAME, 0) != 0)
518 goto nomem;
519 config_seen = B_FALSE;
520
521 /*
522 * Iterate over all toplevel vdevs. Grab the pool configuration
523 * from the first one we find, and then go through the rest and
524 * add them as necessary to the 'vdevs' member of the config.
525 */
526 for (ve = pe->pe_vdevs; ve != NULL; ve = ve->ve_next) {
527
528 /*
529 * Determine the best configuration for this vdev by
530 * selecting the config with the latest transaction
531 * group.
532 */
533 best_txg = 0;
534 for (ce = ve->ve_configs; ce != NULL;
535 ce = ce->ce_next) {
536
537 if (ce->ce_txg > best_txg) {
538 tmp = ce->ce_config;
539 best_txg = ce->ce_txg;
540 }
541 }
542
543 /*
544 * We rely on the fact that the max txg for the
545 * pool will contain the most up-to-date information
546 * about the valid top-levels in the vdev namespace.
547 */
548 if (best_txg > max_txg) {
549 (void) nvlist_remove(config,
550 ZPOOL_CONFIG_VDEV_CHILDREN,
551 DATA_TYPE_UINT64);
552 (void) nvlist_remove(config,
553 ZPOOL_CONFIG_HOLE_ARRAY,
554 DATA_TYPE_UINT64_ARRAY);
555
556 max_txg = best_txg;
557 hole_array = NULL;
558 holes = 0;
559 max_id = 0;
560 valid_top_config = B_FALSE;
561
562 if (nvlist_lookup_uint64(tmp,
563 ZPOOL_CONFIG_VDEV_CHILDREN, &max_id) == 0) {
564 verify(nvlist_add_uint64(config,
565 ZPOOL_CONFIG_VDEV_CHILDREN,
566 max_id) == 0);
567 valid_top_config = B_TRUE;
568 }
569
570 if (nvlist_lookup_uint64_array(tmp,
571 ZPOOL_CONFIG_HOLE_ARRAY, &hole_array,
572 &holes) == 0) {
573 verify(nvlist_add_uint64_array(config,
574 ZPOOL_CONFIG_HOLE_ARRAY,
575 hole_array, holes) == 0);
576 }
577 }
578
579 if (!config_seen) {
580 /*
581 * Copy the relevant pieces of data to the pool
582 * configuration:
583 *
584 * version
585 * pool guid
586 * name
587 * comment (if available)
588 * compatibility features (if available)
589 * pool state
590 * hostid (if available)
591 * hostname (if available)
592 */
593 uint64_t state, version;
594 const char *comment = NULL;
595 const char *compatibility = NULL;
596
597 version = fnvlist_lookup_uint64(tmp,
598 ZPOOL_CONFIG_VERSION);
599 fnvlist_add_uint64(config,
600 ZPOOL_CONFIG_VERSION, version);
601 guid = fnvlist_lookup_uint64(tmp,
602 ZPOOL_CONFIG_POOL_GUID);
603 fnvlist_add_uint64(config,
604 ZPOOL_CONFIG_POOL_GUID, guid);
605 name = fnvlist_lookup_string(tmp,
606 ZPOOL_CONFIG_POOL_NAME);
607 fnvlist_add_string(config,
608 ZPOOL_CONFIG_POOL_NAME, name);
609
610 if (nvlist_lookup_string(tmp,
611 ZPOOL_CONFIG_COMMENT, &comment) == 0)
612 fnvlist_add_string(config,
613 ZPOOL_CONFIG_COMMENT, comment);
614
615 if (nvlist_lookup_string(tmp,
616 ZPOOL_CONFIG_COMPATIBILITY,
617 &compatibility) == 0)
618 fnvlist_add_string(config,
619 ZPOOL_CONFIG_COMPATIBILITY,
620 compatibility);
621
622 state = fnvlist_lookup_uint64(tmp,
623 ZPOOL_CONFIG_POOL_STATE);
624 fnvlist_add_uint64(config,
625 ZPOOL_CONFIG_POOL_STATE, state);
626
627 hostid = 0;
628 if (nvlist_lookup_uint64(tmp,
629 ZPOOL_CONFIG_HOSTID, &hostid) == 0) {
630 fnvlist_add_uint64(config,
631 ZPOOL_CONFIG_HOSTID, hostid);
632 hostname = fnvlist_lookup_string(tmp,
633 ZPOOL_CONFIG_HOSTNAME);
634 fnvlist_add_string(config,
635 ZPOOL_CONFIG_HOSTNAME, hostname);
636 }
637
638 config_seen = B_TRUE;
639 }
640
641 /*
642 * Add this top-level vdev to the child array.
643 */
644 verify(nvlist_lookup_nvlist(tmp,
645 ZPOOL_CONFIG_VDEV_TREE, &nvtop) == 0);
646 verify(nvlist_lookup_uint64(nvtop, ZPOOL_CONFIG_ID,
647 &id) == 0);
648
649 if (id >= children) {
650 nvlist_t **newchild;
651
652 newchild = zutil_alloc(hdl, (id + 1) *
653 sizeof (nvlist_t *));
654 if (newchild == NULL)
655 goto nomem;
656
657 for (c = 0; c < children; c++)
658 newchild[c] = child[c];
659
660 free(child);
661 child = newchild;
662 children = id + 1;
663 }
664 if (nvlist_dup(nvtop, &child[id], 0) != 0)
665 goto nomem;
666
667 }
668
669 /*
670 * If we have information about all the top-levels then
671 * clean up the nvlist which we've constructed. This
672 * means removing any extraneous devices that are
673 * beyond the valid range or adding devices to the end
674 * of our array which appear to be missing.
675 */
676 if (valid_top_config) {
677 if (max_id < children) {
678 for (c = max_id; c < children; c++)
679 nvlist_free(child[c]);
680 children = max_id;
681 } else if (max_id > children) {
682 nvlist_t **newchild;
683
684 newchild = zutil_alloc(hdl, (max_id) *
685 sizeof (nvlist_t *));
686 if (newchild == NULL)
687 goto nomem;
688
689 for (c = 0; c < children; c++)
690 newchild[c] = child[c];
691
692 free(child);
693 child = newchild;
694 children = max_id;
695 }
696 }
697
698 verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
699 &guid) == 0);
700
701 /*
702 * The vdev namespace may contain holes as a result of
703 * device removal. We must add them back into the vdev
704 * tree before we process any missing devices.
705 */
706 if (holes > 0) {
707 ASSERT(valid_top_config);
708
709 for (c = 0; c < children; c++) {
710 nvlist_t *holey;
711
712 if (child[c] != NULL ||
713 !vdev_is_hole(hole_array, holes, c))
714 continue;
715
716 if (nvlist_alloc(&holey, NV_UNIQUE_NAME,
717 0) != 0)
718 goto nomem;
719
720 /*
721 * Holes in the namespace are treated as
722 * "hole" top-level vdevs and have a
723 * special flag set on them.
724 */
725 if (nvlist_add_string(holey,
726 ZPOOL_CONFIG_TYPE,
727 VDEV_TYPE_HOLE) != 0 ||
728 nvlist_add_uint64(holey,
729 ZPOOL_CONFIG_ID, c) != 0 ||
730 nvlist_add_uint64(holey,
731 ZPOOL_CONFIG_GUID, 0ULL) != 0) {
732 nvlist_free(holey);
733 goto nomem;
734 }
735 child[c] = holey;
736 }
737 }
738
739 /*
740 * Look for any missing top-level vdevs. If this is the case,
741 * create a faked up 'missing' vdev as a placeholder. We cannot
742 * simply compress the child array, because the kernel performs
743 * certain checks to make sure the vdev IDs match their location
744 * in the configuration.
745 */
746 for (c = 0; c < children; c++) {
747 if (child[c] == NULL) {
748 nvlist_t *missing;
749 if (nvlist_alloc(&missing, NV_UNIQUE_NAME,
750 0) != 0)
751 goto nomem;
752 if (nvlist_add_string(missing,
753 ZPOOL_CONFIG_TYPE,
754 VDEV_TYPE_MISSING) != 0 ||
755 nvlist_add_uint64(missing,
756 ZPOOL_CONFIG_ID, c) != 0 ||
757 nvlist_add_uint64(missing,
758 ZPOOL_CONFIG_GUID, 0ULL) != 0) {
759 nvlist_free(missing);
760 goto nomem;
761 }
762 child[c] = missing;
763 }
764 }
765
766 /*
767 * Put all of this pool's top-level vdevs into a root vdev.
768 */
769 if (nvlist_alloc(&nvroot, NV_UNIQUE_NAME, 0) != 0)
770 goto nomem;
771 if (nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
772 VDEV_TYPE_ROOT) != 0 ||
773 nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) != 0 ||
774 nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, guid) != 0 ||
775 nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
776 (const nvlist_t **)child, children) != 0) {
777 nvlist_free(nvroot);
778 goto nomem;
779 }
780
781 for (c = 0; c < children; c++)
782 nvlist_free(child[c]);
783 free(child);
784 children = 0;
785 child = NULL;
786
787 /*
788 * Go through and fix up any paths and/or devids based on our
789 * known list of vdev GUID -> path mappings.
790 */
791 if (fix_paths(hdl, nvroot, pl->names) != 0) {
792 nvlist_free(nvroot);
793 goto nomem;
794 }
795
796 /*
797 * Add the root vdev to this pool's configuration.
798 */
799 if (nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
800 nvroot) != 0) {
801 nvlist_free(nvroot);
802 goto nomem;
803 }
804 nvlist_free(nvroot);
805
806 /*
807 * zdb uses this path to report on active pools that were
808 * imported or created using -R.
809 */
810 if (active_ok)
811 goto add_pool;
812
813 /*
814 * Determine if this pool is currently active, in which case we
815 * can't actually import it.
816 */
817 verify(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
818 &name) == 0);
819 verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
820 &guid) == 0);
821
822 if (zutil_pool_active(hdl, name, guid, &isactive) != 0)
823 goto error;
824
825 if (isactive) {
826 nvlist_free(config);
827 config = NULL;
828 continue;
829 }
830
831 if (policy != NULL) {
832 if (nvlist_add_nvlist(config, ZPOOL_LOAD_POLICY,
833 policy) != 0)
834 goto nomem;
835 }
836
837 if ((nvl = zutil_refresh_config(hdl, config)) == NULL) {
838 nvlist_free(config);
839 config = NULL;
840 continue;
841 }
842
843 nvlist_free(config);
844 config = nvl;
845
846 /*
847 * Go through and update the paths for spares, now that we have
848 * them.
849 */
850 verify(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
851 &nvroot) == 0);
852 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
853 &spares, &nspares) == 0) {
854 for (i = 0; i < nspares; i++) {
855 if (fix_paths(hdl, spares[i], pl->names) != 0)
856 goto nomem;
857 }
858 }
859
860 /*
861 * Update the paths for l2cache devices.
862 */
863 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
864 &l2cache, &nl2cache) == 0) {
865 for (i = 0; i < nl2cache; i++) {
866 if (fix_paths(hdl, l2cache[i], pl->names) != 0)
867 goto nomem;
868 }
869 }
870
871 /*
872 * Restore the original information read from the actual label.
873 */
874 (void) nvlist_remove(config, ZPOOL_CONFIG_HOSTID,
875 DATA_TYPE_UINT64);
876 (void) nvlist_remove(config, ZPOOL_CONFIG_HOSTNAME,
877 DATA_TYPE_STRING);
878 if (hostid != 0) {
879 verify(nvlist_add_uint64(config, ZPOOL_CONFIG_HOSTID,
880 hostid) == 0);
881 verify(nvlist_add_string(config, ZPOOL_CONFIG_HOSTNAME,
882 hostname) == 0);
883 }
884
885 add_pool:
886 /*
887 * Add this pool to the list of configs.
888 */
889 verify(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
890 &name) == 0);
891
892 if (nvlist_add_nvlist(ret, name, config) != 0)
893 goto nomem;
894
895 nvlist_free(config);
896 config = NULL;
897 }
898
899 return (ret);
900
901 nomem:
902 (void) zutil_no_memory(hdl);
903 error:
904 nvlist_free(config);
905 nvlist_free(ret);
906 for (c = 0; c < children; c++)
907 nvlist_free(child[c]);
908 free(child);
909
910 return (NULL);
911 }
912
913 /*
914 * Return the offset of the given label.
915 */
916 static uint64_t
917 label_offset(uint64_t size, int l)
918 {
919 ASSERT(P2PHASE_TYPED(size, sizeof (vdev_label_t), uint64_t) == 0);
920 return (l * sizeof (vdev_label_t) + (l < VDEV_LABELS / 2 ?
921 0 : size - VDEV_LABELS * sizeof (vdev_label_t)));
922 }
923
924 /*
925 * The same description applies as to zpool_read_label below,
926 * except here we do it without aio, presumably because an aio call
927 * errored out in a way we think not using it could circumvent.
928 */
929 static int
930 zpool_read_label_slow(int fd, nvlist_t **config, int *num_labels)
931 {
932 struct stat64 statbuf;
933 int l, count = 0;
934 vdev_phys_t *label;
935 nvlist_t *expected_config = NULL;
936 uint64_t expected_guid = 0, size;
937
938 *config = NULL;
939
940 if (fstat64_blk(fd, &statbuf) == -1)
941 return (0);
942 size = P2ALIGN_TYPED(statbuf.st_size, sizeof (vdev_label_t), uint64_t);
943
944 label = (vdev_phys_t *)umem_alloc_aligned(sizeof (*label), PAGESIZE,
945 UMEM_DEFAULT);
946 if (label == NULL)
947 return (-1);
948
949 for (l = 0; l < VDEV_LABELS; l++) {
950 uint64_t state, guid, txg;
951 off_t offset = label_offset(size, l) + VDEV_SKIP_SIZE;
952
953 if (pread64(fd, label, sizeof (vdev_phys_t),
954 offset) != sizeof (vdev_phys_t))
955 continue;
956
957 if (nvlist_unpack(label->vp_nvlist,
958 sizeof (label->vp_nvlist), config, 0) != 0)
959 continue;
960
961 if (nvlist_lookup_uint64(*config, ZPOOL_CONFIG_GUID,
962 &guid) != 0 || guid == 0) {
963 nvlist_free(*config);
964 continue;
965 }
966
967 if (nvlist_lookup_uint64(*config, ZPOOL_CONFIG_POOL_STATE,
968 &state) != 0 || state > POOL_STATE_L2CACHE) {
969 nvlist_free(*config);
970 continue;
971 }
972
973 if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
974 (nvlist_lookup_uint64(*config, ZPOOL_CONFIG_POOL_TXG,
975 &txg) != 0 || txg == 0)) {
976 nvlist_free(*config);
977 continue;
978 }
979
980 if (expected_guid) {
981 if (expected_guid == guid)
982 count++;
983
984 nvlist_free(*config);
985 } else {
986 expected_config = *config;
987 expected_guid = guid;
988 count++;
989 }
990 }
991
992 if (num_labels != NULL)
993 *num_labels = count;
994
995 umem_free_aligned(label, sizeof (*label));
996 *config = expected_config;
997
998 return (0);
999 }
1000
1001 /*
1002 * Given a file descriptor, read the label information and return an nvlist
1003 * describing the configuration, if there is one. The number of valid
1004 * labels found will be returned in num_labels when non-NULL.
1005 */
1006 int
1007 zpool_read_label(int fd, nvlist_t **config, int *num_labels)
1008 {
1009 #ifndef HAVE_AIO_H
1010 return (zpool_read_label_slow(fd, config, num_labels));
1011 #else
1012 struct stat64 statbuf;
1013 struct aiocb aiocbs[VDEV_LABELS];
1014 struct aiocb *aiocbps[VDEV_LABELS];
1015 vdev_phys_t *labels;
1016 nvlist_t *expected_config = NULL;
1017 uint64_t expected_guid = 0, size;
1018 int error, l, count = 0;
1019
1020 *config = NULL;
1021
1022 if (fstat64_blk(fd, &statbuf) == -1)
1023 return (0);
1024 size = P2ALIGN_TYPED(statbuf.st_size, sizeof (vdev_label_t), uint64_t);
1025
1026 labels = (vdev_phys_t *)umem_alloc_aligned(
1027 VDEV_LABELS * sizeof (*labels), PAGESIZE, UMEM_DEFAULT);
1028 if (labels == NULL)
1029 return (-1);
1030
1031 memset(aiocbs, 0, sizeof (aiocbs));
1032 for (l = 0; l < VDEV_LABELS; l++) {
1033 off_t offset = label_offset(size, l) + VDEV_SKIP_SIZE;
1034
1035 aiocbs[l].aio_fildes = fd;
1036 aiocbs[l].aio_offset = offset;
1037 aiocbs[l].aio_buf = &labels[l];
1038 aiocbs[l].aio_nbytes = sizeof (vdev_phys_t);
1039 aiocbs[l].aio_lio_opcode = LIO_READ;
1040 aiocbps[l] = &aiocbs[l];
1041 }
1042
1043 if (lio_listio(LIO_WAIT, aiocbps, VDEV_LABELS, NULL) != 0) {
1044 int saved_errno = errno;
1045 boolean_t do_slow = B_FALSE;
1046 error = -1;
1047
1048 if (errno == EAGAIN || errno == EINTR || errno == EIO) {
1049 /*
1050 * A portion of the requests may have been submitted.
1051 * Clean them up.
1052 */
1053 for (l = 0; l < VDEV_LABELS; l++) {
1054 errno = 0;
1055 switch (aio_error(&aiocbs[l])) {
1056 case EINVAL:
1057 break;
1058 case EINPROGRESS:
1059 /*
1060 * This shouldn't be possible to
1061 * encounter, die if we do.
1062 */
1063 ASSERT(B_FALSE);
1064 zfs_fallthrough;
1065 case EREMOTEIO:
1066 /*
1067 * May be returned by an NVMe device
1068 * which is visible in /dev/ but due
1069 * to a low-level format change, or
1070 * other error, needs to be rescanned.
1071 * Try the slow method.
1072 */
1073 zfs_fallthrough;
1074 case EAGAIN:
1075 case EOPNOTSUPP:
1076 case ENOSYS:
1077 do_slow = B_TRUE;
1078 zfs_fallthrough;
1079 case 0:
1080 default:
1081 (void) aio_return(&aiocbs[l]);
1082 }
1083 }
1084 }
1085 if (do_slow) {
1086 /*
1087 * At least some IO involved access unsafe-for-AIO
1088 * files. Let's try again, without AIO this time.
1089 */
1090 error = zpool_read_label_slow(fd, config, num_labels);
1091 saved_errno = errno;
1092 }
1093 umem_free_aligned(labels, VDEV_LABELS * sizeof (*labels));
1094 errno = saved_errno;
1095 return (error);
1096 }
1097
1098 for (l = 0; l < VDEV_LABELS; l++) {
1099 uint64_t state, guid, txg;
1100
1101 if (aio_return(&aiocbs[l]) != sizeof (vdev_phys_t))
1102 continue;
1103
1104 if (nvlist_unpack(labels[l].vp_nvlist,
1105 sizeof (labels[l].vp_nvlist), config, 0) != 0)
1106 continue;
1107
1108 if (nvlist_lookup_uint64(*config, ZPOOL_CONFIG_GUID,
1109 &guid) != 0 || guid == 0) {
1110 nvlist_free(*config);
1111 continue;
1112 }
1113
1114 if (nvlist_lookup_uint64(*config, ZPOOL_CONFIG_POOL_STATE,
1115 &state) != 0 || state > POOL_STATE_L2CACHE) {
1116 nvlist_free(*config);
1117 continue;
1118 }
1119
1120 if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
1121 (nvlist_lookup_uint64(*config, ZPOOL_CONFIG_POOL_TXG,
1122 &txg) != 0 || txg == 0)) {
1123 nvlist_free(*config);
1124 continue;
1125 }
1126
1127 if (expected_guid) {
1128 if (expected_guid == guid)
1129 count++;
1130
1131 nvlist_free(*config);
1132 } else {
1133 expected_config = *config;
1134 expected_guid = guid;
1135 count++;
1136 }
1137 }
1138
1139 if (num_labels != NULL)
1140 *num_labels = count;
1141
1142 umem_free_aligned(labels, VDEV_LABELS * sizeof (*labels));
1143 *config = expected_config;
1144
1145 return (0);
1146 #endif
1147 }
1148
1149 /*
1150 * Sorted by full path and then vdev guid to allow for multiple entries with
1151 * the same full path name. This is required because it's possible to
1152 * have multiple block devices with labels that refer to the same
1153 * ZPOOL_CONFIG_PATH yet have different vdev guids. In this case both
1154 * entries need to be added to the cache. Scenarios where this can occur
1155 * include overwritten pool labels, devices which are visible from multiple
1156 * hosts and multipath devices.
1157 */
1158 int
1159 slice_cache_compare(const void *arg1, const void *arg2)
1160 {
1161 const char *nm1 = ((rdsk_node_t *)arg1)->rn_name;
1162 const char *nm2 = ((rdsk_node_t *)arg2)->rn_name;
1163 uint64_t guid1 = ((rdsk_node_t *)arg1)->rn_vdev_guid;
1164 uint64_t guid2 = ((rdsk_node_t *)arg2)->rn_vdev_guid;
1165 int rv;
1166
1167 rv = TREE_ISIGN(strcmp(nm1, nm2));
1168 if (rv)
1169 return (rv);
1170
1171 return (TREE_CMP(guid1, guid2));
1172 }
1173
1174 static int
1175 label_paths_impl(libpc_handle_t *hdl, nvlist_t *nvroot, uint64_t pool_guid,
1176 uint64_t vdev_guid, const char **path, const char **devid)
1177 {
1178 nvlist_t **child;
1179 uint_t c, children;
1180 uint64_t guid;
1181 const char *val;
1182 int error;
1183
1184 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
1185 &child, &children) == 0) {
1186 for (c = 0; c < children; c++) {
1187 error = label_paths_impl(hdl, child[c],
1188 pool_guid, vdev_guid, path, devid);
1189 if (error)
1190 return (error);
1191 }
1192 return (0);
1193 }
1194
1195 if (nvroot == NULL)
1196 return (0);
1197
1198 error = nvlist_lookup_uint64(nvroot, ZPOOL_CONFIG_GUID, &guid);
1199 if ((error != 0) || (guid != vdev_guid))
1200 return (0);
1201
1202 error = nvlist_lookup_string(nvroot, ZPOOL_CONFIG_PATH, &val);
1203 if (error == 0)
1204 *path = val;
1205
1206 error = nvlist_lookup_string(nvroot, ZPOOL_CONFIG_DEVID, &val);
1207 if (error == 0)
1208 *devid = val;
1209
1210 return (0);
1211 }
1212
1213 /*
1214 * Given a disk label fetch the ZPOOL_CONFIG_PATH and ZPOOL_CONFIG_DEVID
1215 * and store these strings as config_path and devid_path respectively.
1216 * The returned pointers are only valid as long as label remains valid.
1217 */
1218 int
1219 label_paths(libpc_handle_t *hdl, nvlist_t *label, const char **path,
1220 const char **devid)
1221 {
1222 nvlist_t *nvroot;
1223 uint64_t pool_guid;
1224 uint64_t vdev_guid;
1225 uint64_t state;
1226
1227 *path = NULL;
1228 *devid = NULL;
1229 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, &vdev_guid) != 0)
1230 return (ENOENT);
1231
1232 /*
1233 * In case of spare or l2cache, we directly return path/devid from the
1234 * label.
1235 */
1236 if (!(nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE, &state)) &&
1237 (state == POOL_STATE_SPARE || state == POOL_STATE_L2CACHE)) {
1238 (void) nvlist_lookup_string(label, ZPOOL_CONFIG_PATH, path);
1239 (void) nvlist_lookup_string(label, ZPOOL_CONFIG_DEVID, devid);
1240 return (0);
1241 }
1242
1243 if (nvlist_lookup_nvlist(label, ZPOOL_CONFIG_VDEV_TREE, &nvroot) ||
1244 nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_GUID, &pool_guid))
1245 return (ENOENT);
1246
1247 return (label_paths_impl(hdl, nvroot, pool_guid, vdev_guid, path,
1248 devid));
1249 }
1250
1251 static void
1252 zpool_find_import_scan_add_slice(libpc_handle_t *hdl, pthread_mutex_t *lock,
1253 avl_tree_t *cache, const char *path, const char *name, int order)
1254 {
1255 avl_index_t where;
1256 rdsk_node_t *slice;
1257
1258 slice = zutil_alloc(hdl, sizeof (rdsk_node_t));
1259 if (asprintf(&slice->rn_name, "%s/%s", path, name) == -1) {
1260 free(slice);
1261 return;
1262 }
1263 slice->rn_vdev_guid = 0;
1264 slice->rn_lock = lock;
1265 slice->rn_avl = cache;
1266 slice->rn_hdl = hdl;
1267 slice->rn_order = order + IMPORT_ORDER_SCAN_OFFSET;
1268 slice->rn_labelpaths = B_FALSE;
1269
1270 pthread_mutex_lock(lock);
1271 if (avl_find(cache, slice, &where)) {
1272 free(slice->rn_name);
1273 free(slice);
1274 } else {
1275 avl_insert(cache, slice, where);
1276 }
1277 pthread_mutex_unlock(lock);
1278 }
1279
1280 static int
1281 zpool_find_import_scan_dir(libpc_handle_t *hdl, pthread_mutex_t *lock,
1282 avl_tree_t *cache, const char *dir, int order)
1283 {
1284 int error;
1285 char path[MAXPATHLEN];
1286 struct dirent64 *dp;
1287 DIR *dirp;
1288
1289 if (realpath(dir, path) == NULL) {
1290 error = errno;
1291 if (error == ENOENT)
1292 return (0);
1293
1294 zutil_error_aux(hdl, "%s", zfs_strerror(error));
1295 (void) zutil_error_fmt(hdl, LPC_BADPATH, dgettext(TEXT_DOMAIN,
1296 "cannot resolve path '%s'"), dir);
1297 return (error);
1298 }
1299
1300 dirp = opendir(path);
1301 if (dirp == NULL) {
1302 error = errno;
1303 zutil_error_aux(hdl, "%s", zfs_strerror(error));
1304 (void) zutil_error_fmt(hdl, LPC_BADPATH, dgettext(TEXT_DOMAIN,
1305 "cannot open '%s'"), path);
1306 return (error);
1307 }
1308
1309 while ((dp = readdir64(dirp)) != NULL) {
1310 const char *name = dp->d_name;
1311 if (strcmp(name, ".") == 0 || strcmp(name, "..") == 0)
1312 continue;
1313
1314 switch (dp->d_type) {
1315 case DT_UNKNOWN:
1316 case DT_BLK:
1317 case DT_LNK:
1318 #ifdef __FreeBSD__
1319 case DT_CHR:
1320 #endif
1321 case DT_REG:
1322 break;
1323 default:
1324 continue;
1325 }
1326
1327 zpool_find_import_scan_add_slice(hdl, lock, cache, path, name,
1328 order);
1329 }
1330
1331 (void) closedir(dirp);
1332 return (0);
1333 }
1334
1335 static int
1336 zpool_find_import_scan_path(libpc_handle_t *hdl, pthread_mutex_t *lock,
1337 avl_tree_t *cache, const char *dir, int order)
1338 {
1339 int error = 0;
1340 char path[MAXPATHLEN];
1341 char *d = NULL;
1342 ssize_t dl;
1343 const char *dpath, *name;
1344
1345 /*
1346 * Separate the directory and the basename.
1347 * We do this so that we can get the realpath of
1348 * the directory. We don't get the realpath on the
1349 * whole path because if it's a symlink, we want the
1350 * path of the symlink not where it points to.
1351 */
1352 name = zfs_basename(dir);
1353 if ((dl = zfs_dirnamelen(dir)) == -1)
1354 dpath = ".";
1355 else
1356 dpath = d = zutil_strndup(hdl, dir, dl);
1357
1358 if (realpath(dpath, path) == NULL) {
1359 error = errno;
1360 if (error == ENOENT) {
1361 error = 0;
1362 goto out;
1363 }
1364
1365 zutil_error_aux(hdl, "%s", zfs_strerror(error));
1366 (void) zutil_error_fmt(hdl, LPC_BADPATH, dgettext(TEXT_DOMAIN,
1367 "cannot resolve path '%s'"), dir);
1368 goto out;
1369 }
1370
1371 zpool_find_import_scan_add_slice(hdl, lock, cache, path, name, order);
1372
1373 out:
1374 free(d);
1375 return (error);
1376 }
1377
1378 /*
1379 * Scan a list of directories for zfs devices.
1380 */
1381 static int
1382 zpool_find_import_scan(libpc_handle_t *hdl, pthread_mutex_t *lock,
1383 avl_tree_t **slice_cache, const char * const *dir, size_t dirs)
1384 {
1385 avl_tree_t *cache;
1386 rdsk_node_t *slice;
1387 void *cookie;
1388 int i, error;
1389
1390 *slice_cache = NULL;
1391 cache = zutil_alloc(hdl, sizeof (avl_tree_t));
1392 avl_create(cache, slice_cache_compare, sizeof (rdsk_node_t),
1393 offsetof(rdsk_node_t, rn_node));
1394
1395 for (i = 0; i < dirs; i++) {
1396 struct stat sbuf;
1397
1398 if (stat(dir[i], &sbuf) != 0) {
1399 error = errno;
1400 if (error == ENOENT)
1401 continue;
1402
1403 zutil_error_aux(hdl, "%s", zfs_strerror(error));
1404 (void) zutil_error_fmt(hdl, LPC_BADPATH, dgettext(
1405 TEXT_DOMAIN, "cannot resolve path '%s'"), dir[i]);
1406 goto error;
1407 }
1408
1409 /*
1410 * If dir[i] is a directory, we walk through it and add all
1411 * the entries to the cache. If it's not a directory, we just
1412 * add it to the cache.
1413 */
1414 if (S_ISDIR(sbuf.st_mode)) {
1415 if ((error = zpool_find_import_scan_dir(hdl, lock,
1416 cache, dir[i], i)) != 0)
1417 goto error;
1418 } else {
1419 if ((error = zpool_find_import_scan_path(hdl, lock,
1420 cache, dir[i], i)) != 0)
1421 goto error;
1422 }
1423 }
1424
1425 *slice_cache = cache;
1426 return (0);
1427
1428 error:
1429 cookie = NULL;
1430 while ((slice = avl_destroy_nodes(cache, &cookie)) != NULL) {
1431 free(slice->rn_name);
1432 free(slice);
1433 }
1434 free(cache);
1435
1436 return (error);
1437 }
1438
1439 /*
1440 * Given a list of directories to search, find all pools stored on disk. This
1441 * includes partial pools which are not available to import. If no args are
1442 * given (argc is 0), then the default directory (/dev/dsk) is searched.
1443 * poolname or guid (but not both) are provided by the caller when trying
1444 * to import a specific pool.
1445 */
1446 static nvlist_t *
1447 zpool_find_import_impl(libpc_handle_t *hdl, importargs_t *iarg,
1448 pthread_mutex_t *lock, avl_tree_t *cache)
1449 {
1450 (void) lock;
1451 nvlist_t *ret = NULL;
1452 pool_list_t pools = { 0 };
1453 pool_entry_t *pe, *penext;
1454 vdev_entry_t *ve, *venext;
1455 config_entry_t *ce, *cenext;
1456 name_entry_t *ne, *nenext;
1457 rdsk_node_t *slice;
1458 void *cookie;
1459 tpool_t *t;
1460
1461 verify(iarg->poolname == NULL || iarg->guid == 0);
1462
1463 /*
1464 * Create a thread pool to parallelize the process of reading and
1465 * validating labels, a large number of threads can be used due to
1466 * minimal contention.
1467 */
1468 long threads = 2 * sysconf(_SC_NPROCESSORS_ONLN);
1469 #ifdef HAVE_AIO_H
1470 long am;
1471 #ifdef _SC_AIO_LISTIO_MAX
1472 am = sysconf(_SC_AIO_LISTIO_MAX);
1473 if (am >= VDEV_LABELS)
1474 threads = MIN(threads, am / VDEV_LABELS);
1475 #endif
1476 #ifdef _SC_AIO_MAX
1477 am = sysconf(_SC_AIO_MAX);
1478 if (am >= VDEV_LABELS)
1479 threads = MIN(threads, am / VDEV_LABELS);
1480 #endif
1481 #endif
1482 t = tpool_create(1, threads, 0, NULL);
1483 for (slice = avl_first(cache); slice;
1484 (slice = avl_walk(cache, slice, AVL_AFTER)))
1485 (void) tpool_dispatch(t, zpool_open_func, slice);
1486
1487 tpool_wait(t);
1488 tpool_destroy(t);
1489
1490 /*
1491 * Process the cache, filtering out any entries which are not
1492 * for the specified pool then adding matching label configs.
1493 */
1494 cookie = NULL;
1495 while ((slice = avl_destroy_nodes(cache, &cookie)) != NULL) {
1496 if (slice->rn_config != NULL) {
1497 nvlist_t *config = slice->rn_config;
1498 boolean_t matched = B_TRUE;
1499 boolean_t aux = B_FALSE;
1500 int fd;
1501
1502 /*
1503 * Check if it's a spare or l2cache device. If it is,
1504 * we need to skip the name and guid check since they
1505 * don't exist on aux device label.
1506 */
1507 if (iarg->poolname != NULL || iarg->guid != 0) {
1508 uint64_t state;
1509 aux = nvlist_lookup_uint64(config,
1510 ZPOOL_CONFIG_POOL_STATE, &state) == 0 &&
1511 (state == POOL_STATE_SPARE ||
1512 state == POOL_STATE_L2CACHE);
1513 }
1514
1515 if (iarg->poolname != NULL && !aux) {
1516 const char *pname;
1517
1518 matched = nvlist_lookup_string(config,
1519 ZPOOL_CONFIG_POOL_NAME, &pname) == 0 &&
1520 strcmp(iarg->poolname, pname) == 0;
1521 } else if (iarg->guid != 0 && !aux) {
1522 uint64_t this_guid;
1523
1524 matched = nvlist_lookup_uint64(config,
1525 ZPOOL_CONFIG_POOL_GUID, &this_guid) == 0 &&
1526 iarg->guid == this_guid;
1527 }
1528 if (matched) {
1529 /*
1530 * Verify all remaining entries can be opened
1531 * exclusively. This will prune all underlying
1532 * multipath devices which otherwise could
1533 * result in the vdev appearing as UNAVAIL.
1534 *
1535 * Under zdb, this step isn't required and
1536 * would prevent a zdb -e of active pools with
1537 * no cachefile.
1538 */
1539 fd = open(slice->rn_name,
1540 O_RDONLY | O_EXCL | O_CLOEXEC);
1541 if (fd >= 0 || iarg->can_be_active) {
1542 if (fd >= 0)
1543 close(fd);
1544 add_config(hdl, &pools,
1545 slice->rn_name, slice->rn_order,
1546 slice->rn_num_labels, config);
1547 }
1548 }
1549 nvlist_free(config);
1550 }
1551 free(slice->rn_name);
1552 free(slice);
1553 }
1554 avl_destroy(cache);
1555 free(cache);
1556
1557 ret = get_configs(hdl, &pools, iarg->can_be_active, iarg->policy);
1558
1559 for (pe = pools.pools; pe != NULL; pe = penext) {
1560 penext = pe->pe_next;
1561 for (ve = pe->pe_vdevs; ve != NULL; ve = venext) {
1562 venext = ve->ve_next;
1563 for (ce = ve->ve_configs; ce != NULL; ce = cenext) {
1564 cenext = ce->ce_next;
1565 nvlist_free(ce->ce_config);
1566 free(ce);
1567 }
1568 free(ve);
1569 }
1570 free(pe);
1571 }
1572
1573 for (ne = pools.names; ne != NULL; ne = nenext) {
1574 nenext = ne->ne_next;
1575 free(ne->ne_name);
1576 free(ne);
1577 }
1578
1579 return (ret);
1580 }
1581
1582 /*
1583 * Given a config, discover the paths for the devices which
1584 * exist in the config.
1585 */
1586 static int
1587 discover_cached_paths(libpc_handle_t *hdl, nvlist_t *nv,
1588 avl_tree_t *cache, pthread_mutex_t *lock)
1589 {
1590 const char *path = NULL;
1591 ssize_t dl;
1592 uint_t children;
1593 nvlist_t **child;
1594
1595 if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
1596 &child, &children) == 0) {
1597 for (int c = 0; c < children; c++) {
1598 discover_cached_paths(hdl, child[c], cache, lock);
1599 }
1600 }
1601
1602 /*
1603 * Once we have the path, we need to add the directory to
1604 * our directory cache.
1605 */
1606 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &path) == 0) {
1607 int ret;
1608 char c = '\0';
1609 if ((dl = zfs_dirnamelen(path)) == -1) {
1610 path = ".";
1611 } else {
1612 c = path[dl];
1613 ((char *)path)[dl] = '\0';
1614
1615 }
1616 ret = zpool_find_import_scan_dir(hdl, lock, cache,
1617 path, 0);
1618 if (c != '\0')
1619 ((char *)path)[dl] = c;
1620
1621 return (ret);
1622 }
1623 return (0);
1624 }
1625
1626 /*
1627 * Given a cache file, return the contents as a list of importable pools.
1628 * poolname or guid (but not both) are provided by the caller when trying
1629 * to import a specific pool.
1630 */
1631 static nvlist_t *
1632 zpool_find_import_cached(libpc_handle_t *hdl, importargs_t *iarg)
1633 {
1634 char *buf;
1635 int fd;
1636 struct stat64 statbuf;
1637 nvlist_t *raw, *src, *dst;
1638 nvlist_t *pools;
1639 nvpair_t *elem;
1640 const char *name;
1641 uint64_t this_guid;
1642 boolean_t active;
1643
1644 verify(iarg->poolname == NULL || iarg->guid == 0);
1645
1646 if ((fd = open(iarg->cachefile, O_RDONLY | O_CLOEXEC)) < 0) {
1647 zutil_error_aux(hdl, "%s", zfs_strerror(errno));
1648 (void) zutil_error(hdl, LPC_BADCACHE, dgettext(TEXT_DOMAIN,
1649 "failed to open cache file"));
1650 return (NULL);
1651 }
1652
1653 if (fstat64(fd, &statbuf) != 0) {
1654 zutil_error_aux(hdl, "%s", zfs_strerror(errno));
1655 (void) close(fd);
1656 (void) zutil_error(hdl, LPC_BADCACHE, dgettext(TEXT_DOMAIN,
1657 "failed to get size of cache file"));
1658 return (NULL);
1659 }
1660
1661 if ((buf = zutil_alloc(hdl, statbuf.st_size)) == NULL) {
1662 (void) close(fd);
1663 return (NULL);
1664 }
1665
1666 if (read(fd, buf, statbuf.st_size) != statbuf.st_size) {
1667 (void) close(fd);
1668 free(buf);
1669 (void) zutil_error(hdl, LPC_BADCACHE, dgettext(TEXT_DOMAIN,
1670 "failed to read cache file contents"));
1671 return (NULL);
1672 }
1673
1674 (void) close(fd);
1675
1676 if (nvlist_unpack(buf, statbuf.st_size, &raw, 0) != 0) {
1677 free(buf);
1678 (void) zutil_error(hdl, LPC_BADCACHE, dgettext(TEXT_DOMAIN,
1679 "invalid or corrupt cache file contents"));
1680 return (NULL);
1681 }
1682
1683 free(buf);
1684
1685 /*
1686 * Go through and get the current state of the pools and refresh their
1687 * state.
1688 */
1689 if (nvlist_alloc(&pools, 0, 0) != 0) {
1690 (void) zutil_no_memory(hdl);
1691 nvlist_free(raw);
1692 return (NULL);
1693 }
1694
1695 elem = NULL;
1696 while ((elem = nvlist_next_nvpair(raw, elem)) != NULL) {
1697 src = fnvpair_value_nvlist(elem);
1698
1699 name = fnvlist_lookup_string(src, ZPOOL_CONFIG_POOL_NAME);
1700 if (iarg->poolname != NULL && strcmp(iarg->poolname, name) != 0)
1701 continue;
1702
1703 this_guid = fnvlist_lookup_uint64(src, ZPOOL_CONFIG_POOL_GUID);
1704 if (iarg->guid != 0 && iarg->guid != this_guid)
1705 continue;
1706
1707 if (zutil_pool_active(hdl, name, this_guid, &active) != 0) {
1708 nvlist_free(raw);
1709 nvlist_free(pools);
1710 return (NULL);
1711 }
1712
1713 if (active)
1714 continue;
1715
1716 if (iarg->scan) {
1717 uint64_t saved_guid = iarg->guid;
1718 const char *saved_poolname = iarg->poolname;
1719 pthread_mutex_t lock;
1720
1721 /*
1722 * Create the device cache that will hold the
1723 * devices we will scan based on the cachefile.
1724 * This will get destroyed and freed by
1725 * zpool_find_import_impl.
1726 */
1727 avl_tree_t *cache = zutil_alloc(hdl,
1728 sizeof (avl_tree_t));
1729 avl_create(cache, slice_cache_compare,
1730 sizeof (rdsk_node_t),
1731 offsetof(rdsk_node_t, rn_node));
1732 nvlist_t *nvroot = fnvlist_lookup_nvlist(src,
1733 ZPOOL_CONFIG_VDEV_TREE);
1734
1735 /*
1736 * We only want to find the pool with this_guid.
1737 * We will reset these values back later.
1738 */
1739 iarg->guid = this_guid;
1740 iarg->poolname = NULL;
1741
1742 /*
1743 * We need to build up a cache of devices that exists
1744 * in the paths pointed to by the cachefile. This allows
1745 * us to preserve the device namespace that was
1746 * originally specified by the user but also lets us
1747 * scan devices in those directories in case they had
1748 * been renamed.
1749 */
1750 pthread_mutex_init(&lock, NULL);
1751 discover_cached_paths(hdl, nvroot, cache, &lock);
1752 nvlist_t *nv = zpool_find_import_impl(hdl, iarg,
1753 &lock, cache);
1754 pthread_mutex_destroy(&lock);
1755
1756 /*
1757 * zpool_find_import_impl will return back
1758 * a list of pools that it found based on the
1759 * device cache. There should only be one pool
1760 * since we're looking for a specific guid.
1761 * We will use that pool to build up the final
1762 * pool nvlist which is returned back to the
1763 * caller.
1764 */
1765 nvpair_t *pair = nvlist_next_nvpair(nv, NULL);
1766 if (pair == NULL)
1767 continue;
1768 fnvlist_add_nvlist(pools, nvpair_name(pair),
1769 fnvpair_value_nvlist(pair));
1770
1771 VERIFY3P(nvlist_next_nvpair(nv, pair), ==, NULL);
1772
1773 iarg->guid = saved_guid;
1774 iarg->poolname = saved_poolname;
1775 continue;
1776 }
1777
1778 if (nvlist_add_string(src, ZPOOL_CONFIG_CACHEFILE,
1779 iarg->cachefile) != 0) {
1780 (void) zutil_no_memory(hdl);
1781 nvlist_free(raw);
1782 nvlist_free(pools);
1783 return (NULL);
1784 }
1785
1786 update_vdevs_config_dev_sysfs_path(src);
1787
1788 if ((dst = zutil_refresh_config(hdl, src)) == NULL) {
1789 nvlist_free(raw);
1790 nvlist_free(pools);
1791 return (NULL);
1792 }
1793
1794 if (nvlist_add_nvlist(pools, nvpair_name(elem), dst) != 0) {
1795 (void) zutil_no_memory(hdl);
1796 nvlist_free(dst);
1797 nvlist_free(raw);
1798 nvlist_free(pools);
1799 return (NULL);
1800 }
1801 nvlist_free(dst);
1802 }
1803 nvlist_free(raw);
1804 return (pools);
1805 }
1806
1807 static nvlist_t *
1808 zpool_find_import(libpc_handle_t *hdl, importargs_t *iarg)
1809 {
1810 pthread_mutex_t lock;
1811 avl_tree_t *cache;
1812 nvlist_t *pools = NULL;
1813
1814 verify(iarg->poolname == NULL || iarg->guid == 0);
1815 pthread_mutex_init(&lock, NULL);
1816
1817 /*
1818 * Locate pool member vdevs by blkid or by directory scanning.
1819 * On success a newly allocated AVL tree which is populated with an
1820 * entry for each discovered vdev will be returned in the cache.
1821 * It's the caller's responsibility to consume and destroy this tree.
1822 */
1823 if (iarg->scan || iarg->paths != 0) {
1824 size_t dirs = iarg->paths;
1825 const char * const *dir = (const char * const *)iarg->path;
1826
1827 if (dirs == 0)
1828 dir = zpool_default_search_paths(&dirs);
1829
1830 if (zpool_find_import_scan(hdl, &lock, &cache,
1831 dir, dirs) != 0) {
1832 pthread_mutex_destroy(&lock);
1833 return (NULL);
1834 }
1835 } else {
1836 if (zpool_find_import_blkid(hdl, &lock, &cache) != 0) {
1837 pthread_mutex_destroy(&lock);
1838 return (NULL);
1839 }
1840 }
1841
1842 pools = zpool_find_import_impl(hdl, iarg, &lock, cache);
1843 pthread_mutex_destroy(&lock);
1844 return (pools);
1845 }
1846
1847
1848 nvlist_t *
1849 zpool_search_import(libpc_handle_t *hdl, importargs_t *import)
1850 {
1851 nvlist_t *pools = NULL;
1852
1853 verify(import->poolname == NULL || import->guid == 0);
1854
1855 if (import->cachefile != NULL)
1856 pools = zpool_find_import_cached(hdl, import);
1857 else
1858 pools = zpool_find_import(hdl, import);
1859
1860 if ((pools == NULL || nvlist_empty(pools)) &&
1861 hdl->lpc_open_access_error && geteuid() != 0) {
1862 (void) zutil_error(hdl, LPC_EACCESS, dgettext(TEXT_DOMAIN,
1863 "no pools found"));
1864 }
1865
1866 return (pools);
1867 }
1868
1869 static boolean_t
1870 pool_match(nvlist_t *cfg, const char *tgt)
1871 {
1872 uint64_t v, guid = strtoull(tgt, NULL, 0);
1873 const char *s;
1874
1875 if (guid != 0) {
1876 if (nvlist_lookup_uint64(cfg, ZPOOL_CONFIG_POOL_GUID, &v) == 0)
1877 return (v == guid);
1878 } else {
1879 if (nvlist_lookup_string(cfg, ZPOOL_CONFIG_POOL_NAME, &s) == 0)
1880 return (strcmp(s, tgt) == 0);
1881 }
1882 return (B_FALSE);
1883 }
1884
1885 int
1886 zpool_find_config(libpc_handle_t *hdl, const char *target, nvlist_t **configp,
1887 importargs_t *args)
1888 {
1889 nvlist_t *pools;
1890 nvlist_t *match = NULL;
1891 nvlist_t *config = NULL;
1892 char *sepp = NULL;
1893 int count = 0;
1894 char *targetdup = strdup(target);
1895
1896 if (targetdup == NULL)
1897 return (ENOMEM);
1898
1899 *configp = NULL;
1900
1901 if ((sepp = strpbrk(targetdup, "/@")) != NULL)
1902 *sepp = '\0';
1903
1904 pools = zpool_search_import(hdl, args);
1905
1906 if (pools != NULL) {
1907 nvpair_t *elem = NULL;
1908 while ((elem = nvlist_next_nvpair(pools, elem)) != NULL) {
1909 VERIFY0(nvpair_value_nvlist(elem, &config));
1910 if (pool_match(config, targetdup)) {
1911 count++;
1912 if (match != NULL) {
1913 /* multiple matches found */
1914 continue;
1915 } else {
1916 match = fnvlist_dup(config);
1917 }
1918 }
1919 }
1920 fnvlist_free(pools);
1921 }
1922
1923 if (count == 0) {
1924 free(targetdup);
1925 return (ENOENT);
1926 }
1927
1928 if (count > 1) {
1929 free(targetdup);
1930 fnvlist_free(match);
1931 return (EINVAL);
1932 }
1933
1934 *configp = match;
1935 free(targetdup);
1936
1937 return (0);
1938 }
1939
1940 /* Return if a vdev is a leaf vdev. Note: draid spares are leaf vdevs. */
1941 static boolean_t
1942 vdev_is_leaf(nvlist_t *nv)
1943 {
1944 uint_t children = 0;
1945 nvlist_t **child;
1946
1947 (void) nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
1948 &child, &children);
1949
1950 return (children == 0);
1951 }
1952
1953 /* Return if a vdev is a leaf vdev and a real device (disk or file) */
1954 static boolean_t
1955 vdev_is_real_leaf(nvlist_t *nv)
1956 {
1957 const char *type = NULL;
1958 if (!vdev_is_leaf(nv))
1959 return (B_FALSE);
1960
1961 (void) nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type);
1962 if ((strcmp(type, VDEV_TYPE_DISK) == 0) ||
1963 (strcmp(type, VDEV_TYPE_FILE) == 0)) {
1964 return (B_TRUE);
1965 }
1966
1967 return (B_FALSE);
1968 }
1969
1970 /*
1971 * This function is called by our FOR_EACH_VDEV() macros.
1972 *
1973 * state: State machine status (stored inside of a (nvlist_t *))
1974 * nv: The current vdev nvlist_t we are iterating over.
1975 * last_nv: The previous vdev nvlist_t we returned to the user in
1976 * the last iteration of FOR_EACH_VDEV(). We use it
1977 * to find the next vdev nvlist_t we should return.
1978 * real_leaves_only: Only return leaf vdevs.
1979 *
1980 * Returns 1 if we found the next vdev nvlist_t for this iteration. 0 if
1981 * we're still searching for it.
1982 */
1983 static int
1984 __for_each_vdev_macro_helper_func(void *state, nvlist_t *nv, void *last_nv,
1985 boolean_t real_leaves_only)
1986 {
1987 enum {FIRST_NV = 0, NEXT_IS_MATCH = 1, STOP_LOOKING = 2};
1988
1989 /* The very first entry in the NV list is a special case */
1990 if (*((nvlist_t **)state) == (nvlist_t *)FIRST_NV) {
1991 if (real_leaves_only && !vdev_is_real_leaf(nv))
1992 return (0);
1993
1994 *((nvlist_t **)last_nv) = nv;
1995 *((nvlist_t **)state) = (nvlist_t *)STOP_LOOKING;
1996 return (1);
1997 }
1998
1999 /*
2000 * We came across our last_nv, meaning the next one is the one we
2001 * want
2002 */
2003 if (nv == *((nvlist_t **)last_nv)) {
2004 /* Next iteration of this function will return the nvlist_t */
2005 *((nvlist_t **)state) = (nvlist_t *)NEXT_IS_MATCH;
2006 return (0);
2007 }
2008
2009 /*
2010 * We marked NEXT_IS_MATCH on the previous iteration, so this is the one
2011 * we want.
2012 */
2013 if (*(nvlist_t **)state == (nvlist_t *)NEXT_IS_MATCH) {
2014 if (real_leaves_only && !vdev_is_real_leaf(nv))
2015 return (0);
2016
2017 *((nvlist_t **)last_nv) = nv;
2018 *((nvlist_t **)state) = (nvlist_t *)STOP_LOOKING;
2019 return (1);
2020 }
2021
2022 return (0);
2023 }
2024
2025 int
2026 for_each_vdev_macro_helper_func(void *state, nvlist_t *nv, void *last_nv)
2027 {
2028 return (__for_each_vdev_macro_helper_func(state, nv, last_nv, B_FALSE));
2029 }
2030
2031 int
2032 for_each_real_leaf_vdev_macro_helper_func(void *state, nvlist_t *nv,
2033 void *last_nv)
2034 {
2035 return (__for_each_vdev_macro_helper_func(state, nv, last_nv, B_TRUE));
2036 }
2037
2038 /*
2039 * Internal function for iterating over the vdevs.
2040 *
2041 * For each vdev, func() will be called and will be passed 'zhp' (which is
2042 * typically the zpool_handle_t cast as a void pointer), the vdev's nvlist, and
2043 * a user-defined data pointer).
2044 *
2045 * The return values from all the func() calls will be OR'd together and
2046 * returned.
2047 */
2048 int
2049 for_each_vdev_cb(void *zhp, nvlist_t *nv, pool_vdev_iter_f func,
2050 void *data)
2051 {
2052 nvlist_t **child;
2053 uint_t c, children;
2054 int ret = 0;
2055 int i;
2056 const char *type;
2057
2058 const char *list[] = {
2059 ZPOOL_CONFIG_SPARES,
2060 ZPOOL_CONFIG_L2CACHE,
2061 ZPOOL_CONFIG_CHILDREN
2062 };
2063
2064 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type) != 0)
2065 return (ret);
2066
2067 /* Don't run our function on indirect vdevs */
2068 if (strcmp(type, VDEV_TYPE_INDIRECT) != 0) {
2069 ret |= func(zhp, nv, data);
2070 }
2071
2072 for (i = 0; i < ARRAY_SIZE(list); i++) {
2073 if (nvlist_lookup_nvlist_array(nv, list[i], &child,
2074 &children) == 0) {
2075 for (c = 0; c < children; c++) {
2076 uint64_t ishole = 0;
2077
2078 (void) nvlist_lookup_uint64(child[c],
2079 ZPOOL_CONFIG_IS_HOLE, &ishole);
2080
2081 if (ishole)
2082 continue;
2083
2084 ret |= for_each_vdev_cb(zhp, child[c],
2085 func, data);
2086 }
2087 }
2088 }
2089
2090 return (ret);
2091 }
2092
2093 /*
2094 * Given an ZPOOL_CONFIG_VDEV_TREE nvpair, iterate over all the vdevs, calling
2095 * func() for each one. func() is passed the vdev's nvlist and an optional
2096 * user-defined 'data' pointer.
2097 */
2098 int
2099 for_each_vdev_in_nvlist(nvlist_t *nvroot, pool_vdev_iter_f func, void *data)
2100 {
2101 return (for_each_vdev_cb(NULL, nvroot, func, data));
2102 }
OpenZFS on Linux and FreeBSD