Print this page
6659 nvlist_free(NULL) is a no-op
Split |
Close |
Expand all |
Collapse all |
--- old/usr/src/lib/libzfs/common/libzfs_import.c
+++ new/usr/src/lib/libzfs/common/libzfs_import.c
1 1 /*
2 2 * CDDL HEADER START
3 3 *
4 4 * The contents of this file are subject to the terms of the
5 5 * Common Development and Distribution License (the "License").
6 6 * You may not use this file except in compliance with the License.
7 7 *
8 8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 9 * or http://www.opensolaris.org/os/licensing.
10 10 * See the License for the specific language governing permissions
11 11 * and limitations under the License.
12 12 *
13 13 * When distributing Covered Code, include this CDDL HEADER in each
14 14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 15 * If applicable, add the following below this CDDL HEADER, with the
16 16 * fields enclosed by brackets "[]" replaced with your own identifying
17 17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 18 *
19 19 * CDDL HEADER END
20 20 */
21 21
22 22 /*
23 23 * Copyright 2015 Nexenta Systems, Inc. All rights reserved.
24 24 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
25 25 * Copyright (c) 2013 by Delphix. All rights reserved.
26 26 * Copyright 2015 RackTop Systems.
27 27 */
28 28
29 29 /*
30 30 * Pool import support functions.
31 31 *
32 32 * To import a pool, we rely on reading the configuration information from the
33 33 * ZFS label of each device. If we successfully read the label, then we
34 34 * organize the configuration information in the following hierarchy:
35 35 *
36 36 * pool guid -> toplevel vdev guid -> label txg
37 37 *
38 38 * Duplicate entries matching this same tuple will be discarded. Once we have
39 39 * examined every device, we pick the best label txg config for each toplevel
40 40 * vdev. We then arrange these toplevel vdevs into a complete pool config, and
41 41 * update any paths that have changed. Finally, we attempt to import the pool
42 42 * using our derived config, and record the results.
43 43 */
44 44
45 45 #include <ctype.h>
46 46 #include <devid.h>
47 47 #include <dirent.h>
48 48 #include <errno.h>
49 49 #include <libintl.h>
50 50 #include <stddef.h>
51 51 #include <stdlib.h>
52 52 #include <string.h>
53 53 #include <sys/stat.h>
54 54 #include <unistd.h>
55 55 #include <fcntl.h>
56 56 #include <sys/vtoc.h>
57 57 #include <sys/dktp/fdisk.h>
58 58 #include <sys/efi_partition.h>
59 59 #include <thread_pool.h>
60 60
61 61 #include <sys/vdev_impl.h>
62 62
63 63 #include "libzfs.h"
64 64 #include "libzfs_impl.h"
65 65
66 66 /*
67 67 * Intermediate structures used to gather configuration information.
68 68 */
69 69 typedef struct config_entry {
70 70 uint64_t ce_txg;
71 71 nvlist_t *ce_config;
72 72 struct config_entry *ce_next;
73 73 } config_entry_t;
74 74
75 75 typedef struct vdev_entry {
76 76 uint64_t ve_guid;
77 77 config_entry_t *ve_configs;
78 78 struct vdev_entry *ve_next;
79 79 } vdev_entry_t;
80 80
81 81 typedef struct pool_entry {
82 82 uint64_t pe_guid;
83 83 vdev_entry_t *pe_vdevs;
84 84 struct pool_entry *pe_next;
85 85 } pool_entry_t;
86 86
87 87 typedef struct name_entry {
88 88 char *ne_name;
89 89 uint64_t ne_guid;
90 90 struct name_entry *ne_next;
91 91 } name_entry_t;
92 92
93 93 typedef struct pool_list {
94 94 pool_entry_t *pools;
95 95 name_entry_t *names;
96 96 } pool_list_t;
97 97
98 98 static char *
99 99 get_devid(const char *path)
100 100 {
101 101 int fd;
102 102 ddi_devid_t devid;
103 103 char *minor, *ret;
104 104
105 105 if ((fd = open(path, O_RDONLY)) < 0)
106 106 return (NULL);
107 107
108 108 minor = NULL;
109 109 ret = NULL;
110 110 if (devid_get(fd, &devid) == 0) {
111 111 if (devid_get_minor_name(fd, &minor) == 0)
112 112 ret = devid_str_encode(devid, minor);
113 113 if (minor != NULL)
114 114 devid_str_free(minor);
115 115 devid_free(devid);
116 116 }
117 117 (void) close(fd);
118 118
119 119 return (ret);
120 120 }
121 121
122 122
123 123 /*
124 124 * Go through and fix up any path and/or devid information for the given vdev
125 125 * configuration.
126 126 */
127 127 static int
128 128 fix_paths(nvlist_t *nv, name_entry_t *names)
129 129 {
130 130 nvlist_t **child;
131 131 uint_t c, children;
132 132 uint64_t guid;
133 133 name_entry_t *ne, *best;
134 134 char *path, *devid;
135 135 int matched;
136 136
137 137 if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
138 138 &child, &children) == 0) {
139 139 for (c = 0; c < children; c++)
140 140 if (fix_paths(child[c], names) != 0)
141 141 return (-1);
142 142 return (0);
143 143 }
144 144
145 145 /*
146 146 * This is a leaf (file or disk) vdev. In either case, go through
147 147 * the name list and see if we find a matching guid. If so, replace
148 148 * the path and see if we can calculate a new devid.
149 149 *
150 150 * There may be multiple names associated with a particular guid, in
151 151 * which case we have overlapping slices or multiple paths to the same
152 152 * disk. If this is the case, then we want to pick the path that is
153 153 * the most similar to the original, where "most similar" is the number
154 154 * of matching characters starting from the end of the path. This will
155 155 * preserve slice numbers even if the disks have been reorganized, and
156 156 * will also catch preferred disk names if multiple paths exist.
157 157 */
158 158 verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) == 0);
159 159 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &path) != 0)
160 160 path = NULL;
161 161
162 162 matched = 0;
163 163 best = NULL;
164 164 for (ne = names; ne != NULL; ne = ne->ne_next) {
165 165 if (ne->ne_guid == guid) {
166 166 const char *src, *dst;
167 167 int count;
168 168
169 169 if (path == NULL) {
170 170 best = ne;
171 171 break;
172 172 }
173 173
174 174 src = ne->ne_name + strlen(ne->ne_name) - 1;
175 175 dst = path + strlen(path) - 1;
176 176 for (count = 0; src >= ne->ne_name && dst >= path;
177 177 src--, dst--, count++)
178 178 if (*src != *dst)
179 179 break;
180 180
181 181 /*
182 182 * At this point, 'count' is the number of characters
183 183 * matched from the end.
184 184 */
185 185 if (count > matched || best == NULL) {
186 186 best = ne;
187 187 matched = count;
188 188 }
189 189 }
190 190 }
191 191
192 192 if (best == NULL)
193 193 return (0);
194 194
195 195 if (nvlist_add_string(nv, ZPOOL_CONFIG_PATH, best->ne_name) != 0)
196 196 return (-1);
197 197
198 198 if ((devid = get_devid(best->ne_name)) == NULL) {
199 199 (void) nvlist_remove_all(nv, ZPOOL_CONFIG_DEVID);
200 200 } else {
201 201 if (nvlist_add_string(nv, ZPOOL_CONFIG_DEVID, devid) != 0) {
202 202 devid_str_free(devid);
203 203 return (-1);
204 204 }
205 205 devid_str_free(devid);
206 206 }
207 207
208 208 return (0);
209 209 }
210 210
211 211 /*
212 212 * Add the given configuration to the list of known devices.
213 213 */
214 214 static int
215 215 add_config(libzfs_handle_t *hdl, pool_list_t *pl, const char *path,
216 216 nvlist_t *config)
217 217 {
218 218 uint64_t pool_guid, vdev_guid, top_guid, txg, state;
219 219 pool_entry_t *pe;
220 220 vdev_entry_t *ve;
221 221 config_entry_t *ce;
222 222 name_entry_t *ne;
223 223
224 224 /*
225 225 * If this is a hot spare not currently in use or level 2 cache
226 226 * device, add it to the list of names to translate, but don't do
227 227 * anything else.
228 228 */
229 229 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_STATE,
230 230 &state) == 0 &&
231 231 (state == POOL_STATE_SPARE || state == POOL_STATE_L2CACHE) &&
232 232 nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID, &vdev_guid) == 0) {
233 233 if ((ne = zfs_alloc(hdl, sizeof (name_entry_t))) == NULL)
234 234 return (-1);
235 235
236 236 if ((ne->ne_name = zfs_strdup(hdl, path)) == NULL) {
237 237 free(ne);
238 238 return (-1);
239 239 }
240 240 ne->ne_guid = vdev_guid;
241 241 ne->ne_next = pl->names;
242 242 pl->names = ne;
243 243 return (0);
244 244 }
245 245
246 246 /*
247 247 * If we have a valid config but cannot read any of these fields, then
248 248 * it means we have a half-initialized label. In vdev_label_init()
249 249 * we write a label with txg == 0 so that we can identify the device
250 250 * in case the user refers to the same disk later on. If we fail to
251 251 * create the pool, we'll be left with a label in this state
252 252 * which should not be considered part of a valid pool.
253 253 */
254 254 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
255 255 &pool_guid) != 0 ||
256 256 nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID,
257 257 &vdev_guid) != 0 ||
258 258 nvlist_lookup_uint64(config, ZPOOL_CONFIG_TOP_GUID,
259 259 &top_guid) != 0 ||
260 260 nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG,
261 261 &txg) != 0 || txg == 0) {
262 262 nvlist_free(config);
263 263 return (0);
264 264 }
265 265
266 266 /*
267 267 * First, see if we know about this pool. If not, then add it to the
268 268 * list of known pools.
269 269 */
270 270 for (pe = pl->pools; pe != NULL; pe = pe->pe_next) {
271 271 if (pe->pe_guid == pool_guid)
272 272 break;
273 273 }
274 274
275 275 if (pe == NULL) {
276 276 if ((pe = zfs_alloc(hdl, sizeof (pool_entry_t))) == NULL) {
277 277 nvlist_free(config);
278 278 return (-1);
279 279 }
280 280 pe->pe_guid = pool_guid;
281 281 pe->pe_next = pl->pools;
282 282 pl->pools = pe;
283 283 }
284 284
285 285 /*
286 286 * Second, see if we know about this toplevel vdev. Add it if its
287 287 * missing.
288 288 */
289 289 for (ve = pe->pe_vdevs; ve != NULL; ve = ve->ve_next) {
290 290 if (ve->ve_guid == top_guid)
291 291 break;
292 292 }
293 293
294 294 if (ve == NULL) {
295 295 if ((ve = zfs_alloc(hdl, sizeof (vdev_entry_t))) == NULL) {
296 296 nvlist_free(config);
297 297 return (-1);
298 298 }
299 299 ve->ve_guid = top_guid;
300 300 ve->ve_next = pe->pe_vdevs;
301 301 pe->pe_vdevs = ve;
302 302 }
303 303
304 304 /*
305 305 * Third, see if we have a config with a matching transaction group. If
306 306 * so, then we do nothing. Otherwise, add it to the list of known
307 307 * configs.
308 308 */
309 309 for (ce = ve->ve_configs; ce != NULL; ce = ce->ce_next) {
310 310 if (ce->ce_txg == txg)
311 311 break;
312 312 }
313 313
314 314 if (ce == NULL) {
315 315 if ((ce = zfs_alloc(hdl, sizeof (config_entry_t))) == NULL) {
316 316 nvlist_free(config);
317 317 return (-1);
318 318 }
319 319 ce->ce_txg = txg;
320 320 ce->ce_config = config;
321 321 ce->ce_next = ve->ve_configs;
322 322 ve->ve_configs = ce;
323 323 } else {
324 324 nvlist_free(config);
325 325 }
326 326
327 327 /*
328 328 * At this point we've successfully added our config to the list of
329 329 * known configs. The last thing to do is add the vdev guid -> path
330 330 * mappings so that we can fix up the configuration as necessary before
331 331 * doing the import.
332 332 */
333 333 if ((ne = zfs_alloc(hdl, sizeof (name_entry_t))) == NULL)
334 334 return (-1);
335 335
336 336 if ((ne->ne_name = zfs_strdup(hdl, path)) == NULL) {
337 337 free(ne);
338 338 return (-1);
339 339 }
340 340
341 341 ne->ne_guid = vdev_guid;
342 342 ne->ne_next = pl->names;
343 343 pl->names = ne;
344 344
345 345 return (0);
346 346 }
347 347
348 348 /*
349 349 * Returns true if the named pool matches the given GUID.
350 350 */
351 351 static int
352 352 pool_active(libzfs_handle_t *hdl, const char *name, uint64_t guid,
353 353 boolean_t *isactive)
354 354 {
355 355 zpool_handle_t *zhp;
356 356 uint64_t theguid;
357 357
358 358 if (zpool_open_silent(hdl, name, &zhp) != 0)
359 359 return (-1);
360 360
361 361 if (zhp == NULL) {
362 362 *isactive = B_FALSE;
363 363 return (0);
364 364 }
365 365
366 366 verify(nvlist_lookup_uint64(zhp->zpool_config, ZPOOL_CONFIG_POOL_GUID,
367 367 &theguid) == 0);
368 368
369 369 zpool_close(zhp);
370 370
371 371 *isactive = (theguid == guid);
372 372 return (0);
373 373 }
374 374
375 375 static nvlist_t *
376 376 refresh_config(libzfs_handle_t *hdl, nvlist_t *config)
377 377 {
378 378 nvlist_t *nvl;
379 379 zfs_cmd_t zc = { 0 };
380 380 int err;
381 381
382 382 if (zcmd_write_conf_nvlist(hdl, &zc, config) != 0)
383 383 return (NULL);
384 384
385 385 if (zcmd_alloc_dst_nvlist(hdl, &zc,
386 386 zc.zc_nvlist_conf_size * 2) != 0) {
387 387 zcmd_free_nvlists(&zc);
388 388 return (NULL);
389 389 }
390 390
391 391 while ((err = ioctl(hdl->libzfs_fd, ZFS_IOC_POOL_TRYIMPORT,
392 392 &zc)) != 0 && errno == ENOMEM) {
393 393 if (zcmd_expand_dst_nvlist(hdl, &zc) != 0) {
394 394 zcmd_free_nvlists(&zc);
395 395 return (NULL);
396 396 }
397 397 }
398 398
399 399 if (err) {
400 400 zcmd_free_nvlists(&zc);
401 401 return (NULL);
402 402 }
403 403
404 404 if (zcmd_read_dst_nvlist(hdl, &zc, &nvl) != 0) {
405 405 zcmd_free_nvlists(&zc);
406 406 return (NULL);
407 407 }
408 408
409 409 zcmd_free_nvlists(&zc);
410 410 return (nvl);
411 411 }
412 412
413 413 /*
414 414 * Determine if the vdev id is a hole in the namespace.
415 415 */
416 416 boolean_t
417 417 vdev_is_hole(uint64_t *hole_array, uint_t holes, uint_t id)
418 418 {
419 419 for (int c = 0; c < holes; c++) {
420 420
421 421 /* Top-level is a hole */
422 422 if (hole_array[c] == id)
423 423 return (B_TRUE);
424 424 }
425 425 return (B_FALSE);
426 426 }
427 427
428 428 /*
429 429 * Convert our list of pools into the definitive set of configurations. We
430 430 * start by picking the best config for each toplevel vdev. Once that's done,
431 431 * we assemble the toplevel vdevs into a full config for the pool. We make a
432 432 * pass to fix up any incorrect paths, and then add it to the main list to
433 433 * return to the user.
434 434 */
435 435 static nvlist_t *
436 436 get_configs(libzfs_handle_t *hdl, pool_list_t *pl, boolean_t active_ok)
437 437 {
438 438 pool_entry_t *pe;
439 439 vdev_entry_t *ve;
440 440 config_entry_t *ce;
441 441 nvlist_t *ret = NULL, *config = NULL, *tmp, *nvtop, *nvroot;
442 442 nvlist_t **spares, **l2cache;
443 443 uint_t i, nspares, nl2cache;
444 444 boolean_t config_seen;
445 445 uint64_t best_txg;
446 446 char *name, *hostname;
447 447 uint64_t guid;
448 448 uint_t children = 0;
449 449 nvlist_t **child = NULL;
450 450 uint_t holes;
451 451 uint64_t *hole_array, max_id;
452 452 uint_t c;
453 453 boolean_t isactive;
454 454 uint64_t hostid;
455 455 nvlist_t *nvl;
456 456 boolean_t found_one = B_FALSE;
457 457 boolean_t valid_top_config = B_FALSE;
458 458
459 459 if (nvlist_alloc(&ret, 0, 0) != 0)
460 460 goto nomem;
461 461
462 462 for (pe = pl->pools; pe != NULL; pe = pe->pe_next) {
463 463 uint64_t id, max_txg = 0;
464 464
465 465 if (nvlist_alloc(&config, NV_UNIQUE_NAME, 0) != 0)
466 466 goto nomem;
467 467 config_seen = B_FALSE;
468 468
469 469 /*
470 470 * Iterate over all toplevel vdevs. Grab the pool configuration
471 471 * from the first one we find, and then go through the rest and
472 472 * add them as necessary to the 'vdevs' member of the config.
473 473 */
474 474 for (ve = pe->pe_vdevs; ve != NULL; ve = ve->ve_next) {
475 475
476 476 /*
477 477 * Determine the best configuration for this vdev by
478 478 * selecting the config with the latest transaction
479 479 * group.
480 480 */
481 481 best_txg = 0;
482 482 for (ce = ve->ve_configs; ce != NULL;
483 483 ce = ce->ce_next) {
484 484
485 485 if (ce->ce_txg > best_txg) {
486 486 tmp = ce->ce_config;
487 487 best_txg = ce->ce_txg;
488 488 }
489 489 }
490 490
491 491 /*
492 492 * We rely on the fact that the max txg for the
493 493 * pool will contain the most up-to-date information
494 494 * about the valid top-levels in the vdev namespace.
495 495 */
496 496 if (best_txg > max_txg) {
497 497 (void) nvlist_remove(config,
498 498 ZPOOL_CONFIG_VDEV_CHILDREN,
499 499 DATA_TYPE_UINT64);
500 500 (void) nvlist_remove(config,
501 501 ZPOOL_CONFIG_HOLE_ARRAY,
502 502 DATA_TYPE_UINT64_ARRAY);
503 503
504 504 max_txg = best_txg;
505 505 hole_array = NULL;
506 506 holes = 0;
507 507 max_id = 0;
508 508 valid_top_config = B_FALSE;
509 509
510 510 if (nvlist_lookup_uint64(tmp,
511 511 ZPOOL_CONFIG_VDEV_CHILDREN, &max_id) == 0) {
512 512 verify(nvlist_add_uint64(config,
513 513 ZPOOL_CONFIG_VDEV_CHILDREN,
514 514 max_id) == 0);
515 515 valid_top_config = B_TRUE;
516 516 }
517 517
518 518 if (nvlist_lookup_uint64_array(tmp,
519 519 ZPOOL_CONFIG_HOLE_ARRAY, &hole_array,
520 520 &holes) == 0) {
521 521 verify(nvlist_add_uint64_array(config,
522 522 ZPOOL_CONFIG_HOLE_ARRAY,
523 523 hole_array, holes) == 0);
524 524 }
525 525 }
526 526
527 527 if (!config_seen) {
528 528 /*
529 529 * Copy the relevant pieces of data to the pool
530 530 * configuration:
531 531 *
532 532 * version
533 533 * pool guid
534 534 * name
535 535 * comment (if available)
536 536 * pool state
537 537 * hostid (if available)
538 538 * hostname (if available)
539 539 */
540 540 uint64_t state, version;
541 541 char *comment = NULL;
542 542
543 543 version = fnvlist_lookup_uint64(tmp,
544 544 ZPOOL_CONFIG_VERSION);
545 545 fnvlist_add_uint64(config,
546 546 ZPOOL_CONFIG_VERSION, version);
547 547 guid = fnvlist_lookup_uint64(tmp,
548 548 ZPOOL_CONFIG_POOL_GUID);
549 549 fnvlist_add_uint64(config,
550 550 ZPOOL_CONFIG_POOL_GUID, guid);
551 551 name = fnvlist_lookup_string(tmp,
552 552 ZPOOL_CONFIG_POOL_NAME);
553 553 fnvlist_add_string(config,
554 554 ZPOOL_CONFIG_POOL_NAME, name);
555 555
556 556 if (nvlist_lookup_string(tmp,
557 557 ZPOOL_CONFIG_COMMENT, &comment) == 0)
558 558 fnvlist_add_string(config,
559 559 ZPOOL_CONFIG_COMMENT, comment);
560 560
561 561 state = fnvlist_lookup_uint64(tmp,
562 562 ZPOOL_CONFIG_POOL_STATE);
563 563 fnvlist_add_uint64(config,
564 564 ZPOOL_CONFIG_POOL_STATE, state);
565 565
566 566 hostid = 0;
567 567 if (nvlist_lookup_uint64(tmp,
568 568 ZPOOL_CONFIG_HOSTID, &hostid) == 0) {
569 569 fnvlist_add_uint64(config,
570 570 ZPOOL_CONFIG_HOSTID, hostid);
571 571 hostname = fnvlist_lookup_string(tmp,
572 572 ZPOOL_CONFIG_HOSTNAME);
573 573 fnvlist_add_string(config,
574 574 ZPOOL_CONFIG_HOSTNAME, hostname);
575 575 }
576 576
577 577 config_seen = B_TRUE;
578 578 }
579 579
580 580 /*
581 581 * Add this top-level vdev to the child array.
582 582 */
583 583 verify(nvlist_lookup_nvlist(tmp,
584 584 ZPOOL_CONFIG_VDEV_TREE, &nvtop) == 0);
585 585 verify(nvlist_lookup_uint64(nvtop, ZPOOL_CONFIG_ID,
586 586 &id) == 0);
587 587
588 588 if (id >= children) {
589 589 nvlist_t **newchild;
590 590
591 591 newchild = zfs_alloc(hdl, (id + 1) *
592 592 sizeof (nvlist_t *));
593 593 if (newchild == NULL)
594 594 goto nomem;
595 595
596 596 for (c = 0; c < children; c++)
597 597 newchild[c] = child[c];
598 598
599 599 free(child);
600 600 child = newchild;
601 601 children = id + 1;
602 602 }
603 603 if (nvlist_dup(nvtop, &child[id], 0) != 0)
604 604 goto nomem;
605 605
606 606 }
607 607
608 608 /*
609 609 * If we have information about all the top-levels then
610 610 * clean up the nvlist which we've constructed. This
611 611 * means removing any extraneous devices that are
612 612 * beyond the valid range or adding devices to the end
613 613 * of our array which appear to be missing.
614 614 */
615 615 if (valid_top_config) {
616 616 if (max_id < children) {
617 617 for (c = max_id; c < children; c++)
618 618 nvlist_free(child[c]);
619 619 children = max_id;
620 620 } else if (max_id > children) {
621 621 nvlist_t **newchild;
622 622
623 623 newchild = zfs_alloc(hdl, (max_id) *
624 624 sizeof (nvlist_t *));
625 625 if (newchild == NULL)
626 626 goto nomem;
627 627
628 628 for (c = 0; c < children; c++)
629 629 newchild[c] = child[c];
630 630
631 631 free(child);
632 632 child = newchild;
633 633 children = max_id;
634 634 }
635 635 }
636 636
637 637 verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
638 638 &guid) == 0);
639 639
640 640 /*
641 641 * The vdev namespace may contain holes as a result of
642 642 * device removal. We must add them back into the vdev
643 643 * tree before we process any missing devices.
644 644 */
645 645 if (holes > 0) {
646 646 ASSERT(valid_top_config);
647 647
648 648 for (c = 0; c < children; c++) {
649 649 nvlist_t *holey;
650 650
651 651 if (child[c] != NULL ||
652 652 !vdev_is_hole(hole_array, holes, c))
653 653 continue;
654 654
655 655 if (nvlist_alloc(&holey, NV_UNIQUE_NAME,
656 656 0) != 0)
657 657 goto nomem;
658 658
659 659 /*
660 660 * Holes in the namespace are treated as
661 661 * "hole" top-level vdevs and have a
662 662 * special flag set on them.
663 663 */
664 664 if (nvlist_add_string(holey,
665 665 ZPOOL_CONFIG_TYPE,
666 666 VDEV_TYPE_HOLE) != 0 ||
667 667 nvlist_add_uint64(holey,
668 668 ZPOOL_CONFIG_ID, c) != 0 ||
669 669 nvlist_add_uint64(holey,
670 670 ZPOOL_CONFIG_GUID, 0ULL) != 0) {
671 671 nvlist_free(holey);
672 672 goto nomem;
673 673 }
674 674 child[c] = holey;
675 675 }
676 676 }
677 677
678 678 /*
679 679 * Look for any missing top-level vdevs. If this is the case,
680 680 * create a faked up 'missing' vdev as a placeholder. We cannot
681 681 * simply compress the child array, because the kernel performs
682 682 * certain checks to make sure the vdev IDs match their location
683 683 * in the configuration.
684 684 */
685 685 for (c = 0; c < children; c++) {
686 686 if (child[c] == NULL) {
687 687 nvlist_t *missing;
688 688 if (nvlist_alloc(&missing, NV_UNIQUE_NAME,
689 689 0) != 0)
690 690 goto nomem;
691 691 if (nvlist_add_string(missing,
692 692 ZPOOL_CONFIG_TYPE,
693 693 VDEV_TYPE_MISSING) != 0 ||
694 694 nvlist_add_uint64(missing,
695 695 ZPOOL_CONFIG_ID, c) != 0 ||
696 696 nvlist_add_uint64(missing,
697 697 ZPOOL_CONFIG_GUID, 0ULL) != 0) {
698 698 nvlist_free(missing);
699 699 goto nomem;
700 700 }
701 701 child[c] = missing;
702 702 }
703 703 }
704 704
705 705 /*
706 706 * Put all of this pool's top-level vdevs into a root vdev.
707 707 */
708 708 if (nvlist_alloc(&nvroot, NV_UNIQUE_NAME, 0) != 0)
709 709 goto nomem;
710 710 if (nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
711 711 VDEV_TYPE_ROOT) != 0 ||
712 712 nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) != 0 ||
713 713 nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, guid) != 0 ||
714 714 nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
715 715 child, children) != 0) {
716 716 nvlist_free(nvroot);
717 717 goto nomem;
718 718 }
719 719
720 720 for (c = 0; c < children; c++)
721 721 nvlist_free(child[c]);
722 722 free(child);
723 723 children = 0;
724 724 child = NULL;
725 725
726 726 /*
727 727 * Go through and fix up any paths and/or devids based on our
728 728 * known list of vdev GUID -> path mappings.
729 729 */
730 730 if (fix_paths(nvroot, pl->names) != 0) {
731 731 nvlist_free(nvroot);
732 732 goto nomem;
733 733 }
734 734
735 735 /*
736 736 * Add the root vdev to this pool's configuration.
737 737 */
738 738 if (nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
739 739 nvroot) != 0) {
740 740 nvlist_free(nvroot);
741 741 goto nomem;
742 742 }
743 743 nvlist_free(nvroot);
744 744
745 745 /*
746 746 * zdb uses this path to report on active pools that were
747 747 * imported or created using -R.
748 748 */
749 749 if (active_ok)
750 750 goto add_pool;
751 751
752 752 /*
753 753 * Determine if this pool is currently active, in which case we
754 754 * can't actually import it.
755 755 */
756 756 verify(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
757 757 &name) == 0);
758 758 verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
759 759 &guid) == 0);
760 760
761 761 if (pool_active(hdl, name, guid, &isactive) != 0)
762 762 goto error;
763 763
764 764 if (isactive) {
765 765 nvlist_free(config);
766 766 config = NULL;
767 767 continue;
768 768 }
769 769
770 770 if ((nvl = refresh_config(hdl, config)) == NULL) {
771 771 nvlist_free(config);
772 772 config = NULL;
773 773 continue;
774 774 }
775 775
776 776 nvlist_free(config);
777 777 config = nvl;
778 778
779 779 /*
780 780 * Go through and update the paths for spares, now that we have
781 781 * them.
782 782 */
783 783 verify(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
784 784 &nvroot) == 0);
785 785 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
786 786 &spares, &nspares) == 0) {
787 787 for (i = 0; i < nspares; i++) {
788 788 if (fix_paths(spares[i], pl->names) != 0)
789 789 goto nomem;
790 790 }
791 791 }
792 792
793 793 /*
794 794 * Update the paths for l2cache devices.
795 795 */
796 796 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
797 797 &l2cache, &nl2cache) == 0) {
798 798 for (i = 0; i < nl2cache; i++) {
799 799 if (fix_paths(l2cache[i], pl->names) != 0)
800 800 goto nomem;
801 801 }
802 802 }
803 803
804 804 /*
805 805 * Restore the original information read from the actual label.
806 806 */
807 807 (void) nvlist_remove(config, ZPOOL_CONFIG_HOSTID,
808 808 DATA_TYPE_UINT64);
809 809 (void) nvlist_remove(config, ZPOOL_CONFIG_HOSTNAME,
810 810 DATA_TYPE_STRING);
811 811 if (hostid != 0) {
812 812 verify(nvlist_add_uint64(config, ZPOOL_CONFIG_HOSTID,
813 813 hostid) == 0);
814 814 verify(nvlist_add_string(config, ZPOOL_CONFIG_HOSTNAME,
815 815 hostname) == 0);
816 816 }
817 817
818 818 add_pool:
819 819 /*
820 820 * Add this pool to the list of configs.
821 821 */
822 822 verify(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
823 823 &name) == 0);
824 824 if (nvlist_add_nvlist(ret, name, config) != 0)
825 825 goto nomem;
826 826
827 827 found_one = B_TRUE;
828 828 nvlist_free(config);
829 829 config = NULL;
830 830 }
831 831
832 832 if (!found_one) {
833 833 nvlist_free(ret);
834 834 ret = NULL;
835 835 }
836 836
837 837 return (ret);
838 838
839 839 nomem:
840 840 (void) no_memory(hdl);
841 841 error:
842 842 nvlist_free(config);
843 843 nvlist_free(ret);
844 844 for (c = 0; c < children; c++)
845 845 nvlist_free(child[c]);
846 846 free(child);
847 847
848 848 return (NULL);
849 849 }
850 850
851 851 /*
852 852 * Return the offset of the given label.
853 853 */
854 854 static uint64_t
855 855 label_offset(uint64_t size, int l)
856 856 {
857 857 ASSERT(P2PHASE_TYPED(size, sizeof (vdev_label_t), uint64_t) == 0);
858 858 return (l * sizeof (vdev_label_t) + (l < VDEV_LABELS / 2 ?
859 859 0 : size - VDEV_LABELS * sizeof (vdev_label_t)));
860 860 }
861 861
862 862 /*
863 863 * Given a file descriptor, read the label information and return an nvlist
864 864 * describing the configuration, if there is one.
865 865 */
866 866 int
867 867 zpool_read_label(int fd, nvlist_t **config)
868 868 {
869 869 struct stat64 statbuf;
870 870 int l;
871 871 vdev_label_t *label;
872 872 uint64_t state, txg, size;
873 873
874 874 *config = NULL;
875 875
876 876 if (fstat64(fd, &statbuf) == -1)
877 877 return (0);
878 878 size = P2ALIGN_TYPED(statbuf.st_size, sizeof (vdev_label_t), uint64_t);
879 879
880 880 if ((label = malloc(sizeof (vdev_label_t))) == NULL)
881 881 return (-1);
882 882
883 883 for (l = 0; l < VDEV_LABELS; l++) {
884 884 if (pread64(fd, label, sizeof (vdev_label_t),
885 885 label_offset(size, l)) != sizeof (vdev_label_t))
886 886 continue;
887 887
888 888 if (nvlist_unpack(label->vl_vdev_phys.vp_nvlist,
889 889 sizeof (label->vl_vdev_phys.vp_nvlist), config, 0) != 0)
890 890 continue;
891 891
892 892 if (nvlist_lookup_uint64(*config, ZPOOL_CONFIG_POOL_STATE,
893 893 &state) != 0 || state > POOL_STATE_L2CACHE) {
894 894 nvlist_free(*config);
895 895 continue;
896 896 }
897 897
898 898 if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
899 899 (nvlist_lookup_uint64(*config, ZPOOL_CONFIG_POOL_TXG,
900 900 &txg) != 0 || txg == 0)) {
901 901 nvlist_free(*config);
902 902 continue;
903 903 }
904 904
905 905 free(label);
906 906 return (0);
907 907 }
908 908
909 909 free(label);
910 910 *config = NULL;
911 911 return (0);
912 912 }
913 913
914 914 typedef struct rdsk_node {
915 915 char *rn_name;
916 916 int rn_dfd;
917 917 libzfs_handle_t *rn_hdl;
918 918 nvlist_t *rn_config;
919 919 avl_tree_t *rn_avl;
920 920 avl_node_t rn_node;
921 921 boolean_t rn_nozpool;
922 922 } rdsk_node_t;
923 923
924 924 static int
925 925 slice_cache_compare(const void *arg1, const void *arg2)
926 926 {
927 927 const char *nm1 = ((rdsk_node_t *)arg1)->rn_name;
928 928 const char *nm2 = ((rdsk_node_t *)arg2)->rn_name;
929 929 char *nm1slice, *nm2slice;
930 930 int rv;
931 931
932 932 /*
933 933 * slices zero and two are the most likely to provide results,
934 934 * so put those first
935 935 */
936 936 nm1slice = strstr(nm1, "s0");
937 937 nm2slice = strstr(nm2, "s0");
938 938 if (nm1slice && !nm2slice) {
939 939 return (-1);
940 940 }
941 941 if (!nm1slice && nm2slice) {
942 942 return (1);
943 943 }
944 944 nm1slice = strstr(nm1, "s2");
945 945 nm2slice = strstr(nm2, "s2");
946 946 if (nm1slice && !nm2slice) {
947 947 return (-1);
948 948 }
949 949 if (!nm1slice && nm2slice) {
950 950 return (1);
951 951 }
952 952
953 953 rv = strcmp(nm1, nm2);
954 954 if (rv == 0)
955 955 return (0);
956 956 return (rv > 0 ? 1 : -1);
957 957 }
958 958
959 959 static void
960 960 check_one_slice(avl_tree_t *r, char *diskname, uint_t partno,
961 961 diskaddr_t size, uint_t blksz)
962 962 {
963 963 rdsk_node_t tmpnode;
964 964 rdsk_node_t *node;
965 965 char sname[MAXNAMELEN];
966 966
967 967 tmpnode.rn_name = &sname[0];
968 968 (void) snprintf(tmpnode.rn_name, MAXNAMELEN, "%s%u",
969 969 diskname, partno);
970 970 /*
971 971 * protect against division by zero for disk labels that
972 972 * contain a bogus sector size
973 973 */
974 974 if (blksz == 0)
975 975 blksz = DEV_BSIZE;
976 976 /* too small to contain a zpool? */
977 977 if ((size < (SPA_MINDEVSIZE / blksz)) &&
978 978 (node = avl_find(r, &tmpnode, NULL)))
979 979 node->rn_nozpool = B_TRUE;
980 980 }
981 981
982 982 static void
983 983 nozpool_all_slices(avl_tree_t *r, const char *sname)
984 984 {
985 985 char diskname[MAXNAMELEN];
986 986 char *ptr;
987 987 int i;
988 988
989 989 (void) strncpy(diskname, sname, MAXNAMELEN);
990 990 if (((ptr = strrchr(diskname, 's')) == NULL) &&
991 991 ((ptr = strrchr(diskname, 'p')) == NULL))
992 992 return;
993 993 ptr[0] = 's';
994 994 ptr[1] = '\0';
995 995 for (i = 0; i < NDKMAP; i++)
996 996 check_one_slice(r, diskname, i, 0, 1);
997 997 ptr[0] = 'p';
998 998 for (i = 0; i <= FD_NUMPART; i++)
999 999 check_one_slice(r, diskname, i, 0, 1);
1000 1000 }
1001 1001
1002 1002 static void
1003 1003 check_slices(avl_tree_t *r, int fd, const char *sname)
1004 1004 {
1005 1005 struct extvtoc vtoc;
1006 1006 struct dk_gpt *gpt;
1007 1007 char diskname[MAXNAMELEN];
1008 1008 char *ptr;
1009 1009 int i;
1010 1010
1011 1011 (void) strncpy(diskname, sname, MAXNAMELEN);
1012 1012 if ((ptr = strrchr(diskname, 's')) == NULL || !isdigit(ptr[1]))
1013 1013 return;
1014 1014 ptr[1] = '\0';
1015 1015
1016 1016 if (read_extvtoc(fd, &vtoc) >= 0) {
1017 1017 for (i = 0; i < NDKMAP; i++)
1018 1018 check_one_slice(r, diskname, i,
1019 1019 vtoc.v_part[i].p_size, vtoc.v_sectorsz);
1020 1020 } else if (efi_alloc_and_read(fd, &gpt) >= 0) {
1021 1021 /*
1022 1022 * on x86 we'll still have leftover links that point
1023 1023 * to slices s[9-15], so use NDKMAP instead
1024 1024 */
1025 1025 for (i = 0; i < NDKMAP; i++)
1026 1026 check_one_slice(r, diskname, i,
1027 1027 gpt->efi_parts[i].p_size, gpt->efi_lbasize);
1028 1028 /* nodes p[1-4] are never used with EFI labels */
1029 1029 ptr[0] = 'p';
1030 1030 for (i = 1; i <= FD_NUMPART; i++)
1031 1031 check_one_slice(r, diskname, i, 0, 1);
1032 1032 efi_free(gpt);
1033 1033 }
1034 1034 }
1035 1035
1036 1036 static void
1037 1037 zpool_open_func(void *arg)
1038 1038 {
1039 1039 rdsk_node_t *rn = arg;
1040 1040 struct stat64 statbuf;
1041 1041 nvlist_t *config;
1042 1042 int fd;
1043 1043
1044 1044 if (rn->rn_nozpool)
1045 1045 return;
1046 1046 if ((fd = openat64(rn->rn_dfd, rn->rn_name, O_RDONLY)) < 0) {
1047 1047 /* symlink to a device that's no longer there */
1048 1048 if (errno == ENOENT)
1049 1049 nozpool_all_slices(rn->rn_avl, rn->rn_name);
1050 1050 return;
1051 1051 }
1052 1052 /*
1053 1053 * Ignore failed stats. We only want regular
1054 1054 * files, character devs and block devs.
1055 1055 */
1056 1056 if (fstat64(fd, &statbuf) != 0 ||
1057 1057 (!S_ISREG(statbuf.st_mode) &&
1058 1058 !S_ISCHR(statbuf.st_mode) &&
1059 1059 !S_ISBLK(statbuf.st_mode))) {
1060 1060 (void) close(fd);
1061 1061 return;
1062 1062 }
1063 1063 /* this file is too small to hold a zpool */
1064 1064 if (S_ISREG(statbuf.st_mode) &&
1065 1065 statbuf.st_size < SPA_MINDEVSIZE) {
1066 1066 (void) close(fd);
1067 1067 return;
1068 1068 } else if (!S_ISREG(statbuf.st_mode)) {
1069 1069 /*
1070 1070 * Try to read the disk label first so we don't have to
1071 1071 * open a bunch of minor nodes that can't have a zpool.
1072 1072 */
1073 1073 check_slices(rn->rn_avl, fd, rn->rn_name);
1074 1074 }
1075 1075
1076 1076 if ((zpool_read_label(fd, &config)) != 0) {
1077 1077 (void) close(fd);
1078 1078 (void) no_memory(rn->rn_hdl);
1079 1079 return;
1080 1080 }
1081 1081 (void) close(fd);
1082 1082
1083 1083 rn->rn_config = config;
1084 1084 }
1085 1085
1086 1086 /*
1087 1087 * Given a file descriptor, clear (zero) the label information. This function
1088 1088 * is currently only used in the appliance stack as part of the ZFS sysevent
1089 1089 * module.
1090 1090 */
1091 1091 int
1092 1092 zpool_clear_label(int fd)
1093 1093 {
1094 1094 struct stat64 statbuf;
1095 1095 int l;
1096 1096 vdev_label_t *label;
1097 1097 uint64_t size;
1098 1098
1099 1099 if (fstat64(fd, &statbuf) == -1)
1100 1100 return (0);
1101 1101 size = P2ALIGN_TYPED(statbuf.st_size, sizeof (vdev_label_t), uint64_t);
1102 1102
1103 1103 if ((label = calloc(sizeof (vdev_label_t), 1)) == NULL)
1104 1104 return (-1);
1105 1105
1106 1106 for (l = 0; l < VDEV_LABELS; l++) {
1107 1107 if (pwrite64(fd, label, sizeof (vdev_label_t),
1108 1108 label_offset(size, l)) != sizeof (vdev_label_t)) {
1109 1109 free(label);
1110 1110 return (-1);
1111 1111 }
1112 1112 }
1113 1113
1114 1114 free(label);
1115 1115 return (0);
1116 1116 }
1117 1117
1118 1118 /*
1119 1119 * Given a list of directories to search, find all pools stored on disk. This
1120 1120 * includes partial pools which are not available to import. If no args are
1121 1121 * given (argc is 0), then the default directory (/dev/dsk) is searched.
1122 1122 * poolname or guid (but not both) are provided by the caller when trying
1123 1123 * to import a specific pool.
1124 1124 */
1125 1125 static nvlist_t *
1126 1126 zpool_find_import_impl(libzfs_handle_t *hdl, importargs_t *iarg)
1127 1127 {
1128 1128 int i, dirs = iarg->paths;
1129 1129 struct dirent64 *dp;
1130 1130 char path[MAXPATHLEN];
1131 1131 char *end, **dir = iarg->path;
1132 1132 size_t pathleft;
1133 1133 nvlist_t *ret = NULL;
1134 1134 static char *default_dir = "/dev/dsk";
1135 1135 pool_list_t pools = { 0 };
1136 1136 pool_entry_t *pe, *penext;
1137 1137 vdev_entry_t *ve, *venext;
1138 1138 config_entry_t *ce, *cenext;
1139 1139 name_entry_t *ne, *nenext;
1140 1140 avl_tree_t slice_cache;
1141 1141 rdsk_node_t *slice;
1142 1142 void *cookie;
1143 1143
1144 1144 if (dirs == 0) {
1145 1145 dirs = 1;
1146 1146 dir = &default_dir;
1147 1147 }
1148 1148
1149 1149 /*
1150 1150 * Go through and read the label configuration information from every
1151 1151 * possible device, organizing the information according to pool GUID
1152 1152 * and toplevel GUID.
1153 1153 */
1154 1154 for (i = 0; i < dirs; i++) {
1155 1155 tpool_t *t;
1156 1156 char *rdsk;
1157 1157 int dfd;
1158 1158 boolean_t config_failed = B_FALSE;
1159 1159 DIR *dirp;
1160 1160
1161 1161 /* use realpath to normalize the path */
1162 1162 if (realpath(dir[i], path) == 0) {
1163 1163 (void) zfs_error_fmt(hdl, EZFS_BADPATH,
1164 1164 dgettext(TEXT_DOMAIN, "cannot open '%s'"), dir[i]);
1165 1165 goto error;
1166 1166 }
1167 1167 end = &path[strlen(path)];
1168 1168 *end++ = '/';
1169 1169 *end = 0;
1170 1170 pathleft = &path[sizeof (path)] - end;
1171 1171
1172 1172 /*
1173 1173 * Using raw devices instead of block devices when we're
1174 1174 * reading the labels skips a bunch of slow operations during
1175 1175 * close(2) processing, so we replace /dev/dsk with /dev/rdsk.
1176 1176 */
1177 1177 if (strcmp(path, "/dev/dsk/") == 0)
1178 1178 rdsk = "/dev/rdsk/";
1179 1179 else
1180 1180 rdsk = path;
1181 1181
1182 1182 if ((dfd = open64(rdsk, O_RDONLY)) < 0 ||
1183 1183 (dirp = fdopendir(dfd)) == NULL) {
1184 1184 if (dfd >= 0)
1185 1185 (void) close(dfd);
1186 1186 zfs_error_aux(hdl, strerror(errno));
1187 1187 (void) zfs_error_fmt(hdl, EZFS_BADPATH,
1188 1188 dgettext(TEXT_DOMAIN, "cannot open '%s'"),
1189 1189 rdsk);
1190 1190 goto error;
1191 1191 }
1192 1192
1193 1193 avl_create(&slice_cache, slice_cache_compare,
1194 1194 sizeof (rdsk_node_t), offsetof(rdsk_node_t, rn_node));
1195 1195 /*
1196 1196 * This is not MT-safe, but we have no MT consumers of libzfs
1197 1197 */
1198 1198 while ((dp = readdir64(dirp)) != NULL) {
1199 1199 const char *name = dp->d_name;
1200 1200 if (name[0] == '.' &&
1201 1201 (name[1] == 0 || (name[1] == '.' && name[2] == 0)))
1202 1202 continue;
1203 1203
1204 1204 slice = zfs_alloc(hdl, sizeof (rdsk_node_t));
1205 1205 slice->rn_name = zfs_strdup(hdl, name);
1206 1206 slice->rn_avl = &slice_cache;
1207 1207 slice->rn_dfd = dfd;
1208 1208 slice->rn_hdl = hdl;
1209 1209 slice->rn_nozpool = B_FALSE;
1210 1210 avl_add(&slice_cache, slice);
1211 1211 }
1212 1212 /*
1213 1213 * create a thread pool to do all of this in parallel;
1214 1214 * rn_nozpool is not protected, so this is racy in that
1215 1215 * multiple tasks could decide that the same slice can
1216 1216 * not hold a zpool, which is benign. Also choose
1217 1217 * double the number of processors; we hold a lot of
1218 1218 * locks in the kernel, so going beyond this doesn't
1219 1219 * buy us much.
1220 1220 */
1221 1221 t = tpool_create(1, 2 * sysconf(_SC_NPROCESSORS_ONLN),
1222 1222 0, NULL);
1223 1223 for (slice = avl_first(&slice_cache); slice;
1224 1224 (slice = avl_walk(&slice_cache, slice,
1225 1225 AVL_AFTER)))
1226 1226 (void) tpool_dispatch(t, zpool_open_func, slice);
1227 1227 tpool_wait(t);
1228 1228 tpool_destroy(t);
1229 1229
1230 1230 cookie = NULL;
1231 1231 while ((slice = avl_destroy_nodes(&slice_cache,
1232 1232 &cookie)) != NULL) {
1233 1233 if (slice->rn_config != NULL && !config_failed) {
1234 1234 nvlist_t *config = slice->rn_config;
1235 1235 boolean_t matched = B_TRUE;
1236 1236
1237 1237 if (iarg->poolname != NULL) {
1238 1238 char *pname;
1239 1239
1240 1240 matched = nvlist_lookup_string(config,
1241 1241 ZPOOL_CONFIG_POOL_NAME,
1242 1242 &pname) == 0 &&
1243 1243 strcmp(iarg->poolname, pname) == 0;
1244 1244 } else if (iarg->guid != 0) {
1245 1245 uint64_t this_guid;
1246 1246
1247 1247 matched = nvlist_lookup_uint64(config,
1248 1248 ZPOOL_CONFIG_POOL_GUID,
1249 1249 &this_guid) == 0 &&
1250 1250 iarg->guid == this_guid;
1251 1251 }
1252 1252 if (!matched) {
1253 1253 nvlist_free(config);
1254 1254 } else {
1255 1255 /*
1256 1256 * use the non-raw path for the config
1257 1257 */
1258 1258 (void) strlcpy(end, slice->rn_name,
1259 1259 pathleft);
1260 1260 if (add_config(hdl, &pools, path,
1261 1261 config) != 0)
1262 1262 config_failed = B_TRUE;
1263 1263 }
1264 1264 }
1265 1265 free(slice->rn_name);
1266 1266 free(slice);
1267 1267 }
1268 1268 avl_destroy(&slice_cache);
1269 1269
1270 1270 (void) closedir(dirp);
1271 1271
1272 1272 if (config_failed)
1273 1273 goto error;
1274 1274 }
↓ open down ↓ |
1274 lines elided |
↑ open up ↑ |
1275 1275
1276 1276 ret = get_configs(hdl, &pools, iarg->can_be_active);
1277 1277
1278 1278 error:
1279 1279 for (pe = pools.pools; pe != NULL; pe = penext) {
1280 1280 penext = pe->pe_next;
1281 1281 for (ve = pe->pe_vdevs; ve != NULL; ve = venext) {
1282 1282 venext = ve->ve_next;
1283 1283 for (ce = ve->ve_configs; ce != NULL; ce = cenext) {
1284 1284 cenext = ce->ce_next;
1285 - if (ce->ce_config)
1286 - nvlist_free(ce->ce_config);
1285 + nvlist_free(ce->ce_config);
1287 1286 free(ce);
1288 1287 }
1289 1288 free(ve);
1290 1289 }
1291 1290 free(pe);
1292 1291 }
1293 1292
1294 1293 for (ne = pools.names; ne != NULL; ne = nenext) {
1295 1294 nenext = ne->ne_next;
1296 1295 free(ne->ne_name);
1297 1296 free(ne);
1298 1297 }
1299 1298
1300 1299 return (ret);
1301 1300 }
1302 1301
1303 1302 nvlist_t *
1304 1303 zpool_find_import(libzfs_handle_t *hdl, int argc, char **argv)
1305 1304 {
1306 1305 importargs_t iarg = { 0 };
1307 1306
1308 1307 iarg.paths = argc;
1309 1308 iarg.path = argv;
1310 1309
1311 1310 return (zpool_find_import_impl(hdl, &iarg));
1312 1311 }
1313 1312
1314 1313 /*
1315 1314 * Given a cache file, return the contents as a list of importable pools.
1316 1315 * poolname or guid (but not both) are provided by the caller when trying
1317 1316 * to import a specific pool.
1318 1317 */
1319 1318 nvlist_t *
1320 1319 zpool_find_import_cached(libzfs_handle_t *hdl, const char *cachefile,
1321 1320 char *poolname, uint64_t guid)
1322 1321 {
1323 1322 char *buf;
1324 1323 int fd;
1325 1324 struct stat64 statbuf;
1326 1325 nvlist_t *raw, *src, *dst;
1327 1326 nvlist_t *pools;
1328 1327 nvpair_t *elem;
1329 1328 char *name;
1330 1329 uint64_t this_guid;
1331 1330 boolean_t active;
1332 1331
1333 1332 verify(poolname == NULL || guid == 0);
1334 1333
1335 1334 if ((fd = open(cachefile, O_RDONLY)) < 0) {
1336 1335 zfs_error_aux(hdl, "%s", strerror(errno));
1337 1336 (void) zfs_error(hdl, EZFS_BADCACHE,
1338 1337 dgettext(TEXT_DOMAIN, "failed to open cache file"));
1339 1338 return (NULL);
1340 1339 }
1341 1340
1342 1341 if (fstat64(fd, &statbuf) != 0) {
1343 1342 zfs_error_aux(hdl, "%s", strerror(errno));
1344 1343 (void) close(fd);
1345 1344 (void) zfs_error(hdl, EZFS_BADCACHE,
1346 1345 dgettext(TEXT_DOMAIN, "failed to get size of cache file"));
1347 1346 return (NULL);
1348 1347 }
1349 1348
1350 1349 if ((buf = zfs_alloc(hdl, statbuf.st_size)) == NULL) {
1351 1350 (void) close(fd);
1352 1351 return (NULL);
1353 1352 }
1354 1353
1355 1354 if (read(fd, buf, statbuf.st_size) != statbuf.st_size) {
1356 1355 (void) close(fd);
1357 1356 free(buf);
1358 1357 (void) zfs_error(hdl, EZFS_BADCACHE,
1359 1358 dgettext(TEXT_DOMAIN,
1360 1359 "failed to read cache file contents"));
1361 1360 return (NULL);
1362 1361 }
1363 1362
1364 1363 (void) close(fd);
1365 1364
1366 1365 if (nvlist_unpack(buf, statbuf.st_size, &raw, 0) != 0) {
1367 1366 free(buf);
1368 1367 (void) zfs_error(hdl, EZFS_BADCACHE,
1369 1368 dgettext(TEXT_DOMAIN,
1370 1369 "invalid or corrupt cache file contents"));
1371 1370 return (NULL);
1372 1371 }
1373 1372
1374 1373 free(buf);
1375 1374
1376 1375 /*
1377 1376 * Go through and get the current state of the pools and refresh their
1378 1377 * state.
1379 1378 */
1380 1379 if (nvlist_alloc(&pools, 0, 0) != 0) {
1381 1380 (void) no_memory(hdl);
1382 1381 nvlist_free(raw);
1383 1382 return (NULL);
1384 1383 }
1385 1384
1386 1385 elem = NULL;
1387 1386 while ((elem = nvlist_next_nvpair(raw, elem)) != NULL) {
1388 1387 src = fnvpair_value_nvlist(elem);
1389 1388
1390 1389 name = fnvlist_lookup_string(src, ZPOOL_CONFIG_POOL_NAME);
1391 1390 if (poolname != NULL && strcmp(poolname, name) != 0)
1392 1391 continue;
1393 1392
1394 1393 this_guid = fnvlist_lookup_uint64(src, ZPOOL_CONFIG_POOL_GUID);
1395 1394 if (guid != 0 && guid != this_guid)
1396 1395 continue;
1397 1396
1398 1397 if (pool_active(hdl, name, this_guid, &active) != 0) {
1399 1398 nvlist_free(raw);
1400 1399 nvlist_free(pools);
1401 1400 return (NULL);
1402 1401 }
1403 1402
1404 1403 if (active)
1405 1404 continue;
1406 1405
1407 1406 if ((dst = refresh_config(hdl, src)) == NULL) {
1408 1407 nvlist_free(raw);
1409 1408 nvlist_free(pools);
1410 1409 return (NULL);
1411 1410 }
1412 1411
1413 1412 if (nvlist_add_nvlist(pools, nvpair_name(elem), dst) != 0) {
1414 1413 (void) no_memory(hdl);
1415 1414 nvlist_free(dst);
1416 1415 nvlist_free(raw);
1417 1416 nvlist_free(pools);
1418 1417 return (NULL);
1419 1418 }
1420 1419 nvlist_free(dst);
1421 1420 }
1422 1421
1423 1422 nvlist_free(raw);
1424 1423 return (pools);
1425 1424 }
1426 1425
1427 1426 static int
1428 1427 name_or_guid_exists(zpool_handle_t *zhp, void *data)
1429 1428 {
1430 1429 importargs_t *import = data;
1431 1430 int found = 0;
1432 1431
1433 1432 if (import->poolname != NULL) {
1434 1433 char *pool_name;
1435 1434
1436 1435 verify(nvlist_lookup_string(zhp->zpool_config,
1437 1436 ZPOOL_CONFIG_POOL_NAME, &pool_name) == 0);
1438 1437 if (strcmp(pool_name, import->poolname) == 0)
1439 1438 found = 1;
1440 1439 } else {
1441 1440 uint64_t pool_guid;
1442 1441
1443 1442 verify(nvlist_lookup_uint64(zhp->zpool_config,
1444 1443 ZPOOL_CONFIG_POOL_GUID, &pool_guid) == 0);
1445 1444 if (pool_guid == import->guid)
1446 1445 found = 1;
1447 1446 }
1448 1447
1449 1448 zpool_close(zhp);
1450 1449 return (found);
1451 1450 }
1452 1451
1453 1452 nvlist_t *
1454 1453 zpool_search_import(libzfs_handle_t *hdl, importargs_t *import)
1455 1454 {
1456 1455 verify(import->poolname == NULL || import->guid == 0);
1457 1456
1458 1457 if (import->unique)
1459 1458 import->exists = zpool_iter(hdl, name_or_guid_exists, import);
1460 1459
1461 1460 if (import->cachefile != NULL)
1462 1461 return (zpool_find_import_cached(hdl, import->cachefile,
1463 1462 import->poolname, import->guid));
1464 1463
1465 1464 return (zpool_find_import_impl(hdl, import));
1466 1465 }
1467 1466
1468 1467 boolean_t
1469 1468 find_guid(nvlist_t *nv, uint64_t guid)
1470 1469 {
1471 1470 uint64_t tmp;
1472 1471 nvlist_t **child;
1473 1472 uint_t c, children;
1474 1473
1475 1474 verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &tmp) == 0);
1476 1475 if (tmp == guid)
1477 1476 return (B_TRUE);
1478 1477
1479 1478 if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
1480 1479 &child, &children) == 0) {
1481 1480 for (c = 0; c < children; c++)
1482 1481 if (find_guid(child[c], guid))
1483 1482 return (B_TRUE);
1484 1483 }
1485 1484
1486 1485 return (B_FALSE);
1487 1486 }
1488 1487
1489 1488 typedef struct aux_cbdata {
1490 1489 const char *cb_type;
1491 1490 uint64_t cb_guid;
1492 1491 zpool_handle_t *cb_zhp;
1493 1492 } aux_cbdata_t;
1494 1493
1495 1494 static int
1496 1495 find_aux(zpool_handle_t *zhp, void *data)
1497 1496 {
1498 1497 aux_cbdata_t *cbp = data;
1499 1498 nvlist_t **list;
1500 1499 uint_t i, count;
1501 1500 uint64_t guid;
1502 1501 nvlist_t *nvroot;
1503 1502
1504 1503 verify(nvlist_lookup_nvlist(zhp->zpool_config, ZPOOL_CONFIG_VDEV_TREE,
1505 1504 &nvroot) == 0);
1506 1505
1507 1506 if (nvlist_lookup_nvlist_array(nvroot, cbp->cb_type,
1508 1507 &list, &count) == 0) {
1509 1508 for (i = 0; i < count; i++) {
1510 1509 verify(nvlist_lookup_uint64(list[i],
1511 1510 ZPOOL_CONFIG_GUID, &guid) == 0);
1512 1511 if (guid == cbp->cb_guid) {
1513 1512 cbp->cb_zhp = zhp;
1514 1513 return (1);
1515 1514 }
1516 1515 }
1517 1516 }
1518 1517
1519 1518 zpool_close(zhp);
1520 1519 return (0);
1521 1520 }
1522 1521
1523 1522 /*
1524 1523 * Determines if the pool is in use. If so, it returns true and the state of
1525 1524 * the pool as well as the name of the pool. Both strings are allocated and
1526 1525 * must be freed by the caller.
1527 1526 */
1528 1527 int
1529 1528 zpool_in_use(libzfs_handle_t *hdl, int fd, pool_state_t *state, char **namestr,
1530 1529 boolean_t *inuse)
1531 1530 {
1532 1531 nvlist_t *config;
1533 1532 char *name;
1534 1533 boolean_t ret;
1535 1534 uint64_t guid, vdev_guid;
1536 1535 zpool_handle_t *zhp;
1537 1536 nvlist_t *pool_config;
1538 1537 uint64_t stateval, isspare;
1539 1538 aux_cbdata_t cb = { 0 };
1540 1539 boolean_t isactive;
1541 1540
1542 1541 *inuse = B_FALSE;
1543 1542
1544 1543 if (zpool_read_label(fd, &config) != 0) {
1545 1544 (void) no_memory(hdl);
1546 1545 return (-1);
1547 1546 }
1548 1547
1549 1548 if (config == NULL)
1550 1549 return (0);
1551 1550
1552 1551 verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_STATE,
1553 1552 &stateval) == 0);
1554 1553 verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID,
1555 1554 &vdev_guid) == 0);
1556 1555
1557 1556 if (stateval != POOL_STATE_SPARE && stateval != POOL_STATE_L2CACHE) {
1558 1557 verify(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
1559 1558 &name) == 0);
1560 1559 verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
1561 1560 &guid) == 0);
1562 1561 }
1563 1562
1564 1563 switch (stateval) {
1565 1564 case POOL_STATE_EXPORTED:
1566 1565 /*
1567 1566 * A pool with an exported state may in fact be imported
1568 1567 * read-only, so check the in-core state to see if it's
1569 1568 * active and imported read-only. If it is, set
1570 1569 * its state to active.
1571 1570 */
1572 1571 if (pool_active(hdl, name, guid, &isactive) == 0 && isactive &&
1573 1572 (zhp = zpool_open_canfail(hdl, name)) != NULL) {
1574 1573 if (zpool_get_prop_int(zhp, ZPOOL_PROP_READONLY, NULL))
1575 1574 stateval = POOL_STATE_ACTIVE;
1576 1575
1577 1576 /*
1578 1577 * All we needed the zpool handle for is the
1579 1578 * readonly prop check.
1580 1579 */
1581 1580 zpool_close(zhp);
1582 1581 }
1583 1582
1584 1583 ret = B_TRUE;
1585 1584 break;
1586 1585
1587 1586 case POOL_STATE_ACTIVE:
1588 1587 /*
1589 1588 * For an active pool, we have to determine if it's really part
1590 1589 * of a currently active pool (in which case the pool will exist
1591 1590 * and the guid will be the same), or whether it's part of an
1592 1591 * active pool that was disconnected without being explicitly
1593 1592 * exported.
1594 1593 */
1595 1594 if (pool_active(hdl, name, guid, &isactive) != 0) {
1596 1595 nvlist_free(config);
1597 1596 return (-1);
1598 1597 }
1599 1598
1600 1599 if (isactive) {
1601 1600 /*
1602 1601 * Because the device may have been removed while
1603 1602 * offlined, we only report it as active if the vdev is
1604 1603 * still present in the config. Otherwise, pretend like
1605 1604 * it's not in use.
1606 1605 */
1607 1606 if ((zhp = zpool_open_canfail(hdl, name)) != NULL &&
1608 1607 (pool_config = zpool_get_config(zhp, NULL))
1609 1608 != NULL) {
1610 1609 nvlist_t *nvroot;
1611 1610
1612 1611 verify(nvlist_lookup_nvlist(pool_config,
1613 1612 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
1614 1613 ret = find_guid(nvroot, vdev_guid);
1615 1614 } else {
1616 1615 ret = B_FALSE;
1617 1616 }
1618 1617
1619 1618 /*
1620 1619 * If this is an active spare within another pool, we
1621 1620 * treat it like an unused hot spare. This allows the
1622 1621 * user to create a pool with a hot spare that currently
1623 1622 * in use within another pool. Since we return B_TRUE,
1624 1623 * libdiskmgt will continue to prevent generic consumers
1625 1624 * from using the device.
1626 1625 */
1627 1626 if (ret && nvlist_lookup_uint64(config,
1628 1627 ZPOOL_CONFIG_IS_SPARE, &isspare) == 0 && isspare)
1629 1628 stateval = POOL_STATE_SPARE;
1630 1629
1631 1630 if (zhp != NULL)
1632 1631 zpool_close(zhp);
1633 1632 } else {
1634 1633 stateval = POOL_STATE_POTENTIALLY_ACTIVE;
1635 1634 ret = B_TRUE;
1636 1635 }
1637 1636 break;
1638 1637
1639 1638 case POOL_STATE_SPARE:
1640 1639 /*
1641 1640 * For a hot spare, it can be either definitively in use, or
1642 1641 * potentially active. To determine if it's in use, we iterate
1643 1642 * over all pools in the system and search for one with a spare
1644 1643 * with a matching guid.
1645 1644 *
1646 1645 * Due to the shared nature of spares, we don't actually report
1647 1646 * the potentially active case as in use. This means the user
1648 1647 * can freely create pools on the hot spares of exported pools,
1649 1648 * but to do otherwise makes the resulting code complicated, and
1650 1649 * we end up having to deal with this case anyway.
1651 1650 */
1652 1651 cb.cb_zhp = NULL;
1653 1652 cb.cb_guid = vdev_guid;
1654 1653 cb.cb_type = ZPOOL_CONFIG_SPARES;
1655 1654 if (zpool_iter(hdl, find_aux, &cb) == 1) {
1656 1655 name = (char *)zpool_get_name(cb.cb_zhp);
1657 1656 ret = B_TRUE;
1658 1657 } else {
1659 1658 ret = B_FALSE;
1660 1659 }
1661 1660 break;
1662 1661
1663 1662 case POOL_STATE_L2CACHE:
1664 1663
1665 1664 /*
1666 1665 * Check if any pool is currently using this l2cache device.
1667 1666 */
1668 1667 cb.cb_zhp = NULL;
1669 1668 cb.cb_guid = vdev_guid;
1670 1669 cb.cb_type = ZPOOL_CONFIG_L2CACHE;
1671 1670 if (zpool_iter(hdl, find_aux, &cb) == 1) {
1672 1671 name = (char *)zpool_get_name(cb.cb_zhp);
1673 1672 ret = B_TRUE;
1674 1673 } else {
1675 1674 ret = B_FALSE;
1676 1675 }
1677 1676 break;
1678 1677
1679 1678 default:
1680 1679 ret = B_FALSE;
1681 1680 }
1682 1681
1683 1682
1684 1683 if (ret) {
1685 1684 if ((*namestr = zfs_strdup(hdl, name)) == NULL) {
1686 1685 if (cb.cb_zhp)
1687 1686 zpool_close(cb.cb_zhp);
1688 1687 nvlist_free(config);
1689 1688 return (-1);
1690 1689 }
1691 1690 *state = (pool_state_t)stateval;
1692 1691 }
1693 1692
1694 1693 if (cb.cb_zhp)
1695 1694 zpool_close(cb.cb_zhp);
1696 1695
1697 1696 nvlist_free(config);
1698 1697 *inuse = ret;
1699 1698 return (0);
1700 1699 }
↓ open down ↓ |
404 lines elided |
↑ open up ↑ |
XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX