1 /*
2 * CDDL HEADER START
3 *
4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
7 *
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
12 *
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 *
19 * CDDL HEADER END
20 */
21 /*
22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23 */
24
25 /*
26 * Virtual Device Labels
27 * ---------------------
28 *
29 * The vdev label serves several distinct purposes:
30 *
31 * 1. Uniquely identify this device as part of a ZFS pool and confirm its
32 * identity within the pool.
33 *
34 * 2. Verify that all the devices given in a configuration are present
35 * within the pool.
36 *
37 * 3. Determine the uberblock for the pool.
38 *
39 * 4. In case of an import operation, determine the configuration of the
40 * toplevel vdev of which it is a part.
41 *
42 * 5. If an import operation cannot find all the devices in the pool,
43 * provide enough information to the administrator to determine which
44 * devices are missing.
45 *
46 * It is important to note that while the kernel is responsible for writing the
47 * label, it only consumes the information in the first three cases. The
48 * latter information is only consumed in userland when determining the
49 * configuration to import a pool.
50 *
51 *
52 * Label Organization
53 * ------------------
54 *
55 * Before describing the contents of the label, it's important to understand how
56 * the labels are written and updated with respect to the uberblock.
57 *
58 * When the pool configuration is altered, either because it was newly created
59 * or a device was added, we want to update all the labels such that we can deal
60 * with fatal failure at any point. To this end, each disk has two labels which
61 * are updated before and after the uberblock is synced. Assuming we have
62 * labels and an uberblock with the following transaction groups:
63 *
64 * L1 UB L2
65 * +------+ +------+ +------+
66 * | | | | | |
67 * | t10 | | t10 | | t10 |
68 * | | | | | |
69 * +------+ +------+ +------+
70 *
71 * In this stable state, the labels and the uberblock were all updated within
72 * the same transaction group (10). Each label is mirrored and checksummed, so
73 * that we can detect when we fail partway through writing the label.
74 *
75 * In order to identify which labels are valid, the labels are written in the
76 * following manner:
77 *
78 * 1. For each vdev, update 'L1' to the new label
79 * 2. Update the uberblock
80 * 3. For each vdev, update 'L2' to the new label
81 *
82 * Given arbitrary failure, we can determine the correct label to use based on
83 * the transaction group. If we fail after updating L1 but before updating the
84 * UB, we will notice that L1's transaction group is greater than the uberblock,
85 * so L2 must be valid. If we fail after writing the uberblock but before
86 * writing L2, we will notice that L2's transaction group is less than L1, and
87 * therefore L1 is valid.
88 *
89 * Another added complexity is that not every label is updated when the config
90 * is synced. If we add a single device, we do not want to have to re-write
91 * every label for every device in the pool. This means that both L1 and L2 may
92 * be older than the pool uberblock, because the necessary information is stored
93 * on another vdev.
94 *
95 *
96 * On-disk Format
97 * --------------
98 *
99 * The vdev label consists of two distinct parts, and is wrapped within the
100 * vdev_label_t structure. The label includes 8k of padding to permit legacy
101 * VTOC disk labels, but is otherwise ignored.
102 *
103 * The first half of the label is a packed nvlist which contains pool wide
104 * properties, per-vdev properties, and configuration information. It is
105 * described in more detail below.
106 *
107 * The latter half of the label consists of a redundant array of uberblocks.
108 * These uberblocks are updated whenever a transaction group is committed,
109 * or when the configuration is updated. When a pool is loaded, we scan each
110 * vdev for the 'best' uberblock.
111 *
112 *
113 * Configuration Information
114 * -------------------------
115 *
116 * The nvlist describing the pool and vdev contains the following elements:
117 *
118 * version ZFS on-disk version
119 * name Pool name
120 * state Pool state
121 * txg Transaction group in which this label was written
122 * pool_guid Unique identifier for this pool
123 * vdev_tree An nvlist describing vdev tree.
124 *
125 * Each leaf device label also contains the following:
126 *
127 * top_guid Unique ID for top-level vdev in which this is contained
128 * guid Unique ID for the leaf vdev
129 *
130 * The 'vs' configuration follows the format described in 'spa_config.c'.
131 */
132
133 #include <sys/zfs_context.h>
134 #include <sys/spa.h>
135 #include <sys/spa_impl.h>
136 #include <sys/dmu.h>
137 #include <sys/zap.h>
138 #include <sys/vdev.h>
139 #include <sys/vdev_impl.h>
140 #include <sys/uberblock_impl.h>
141 #include <sys/metaslab.h>
142 #include <sys/zio.h>
143 #include <sys/dsl_scan.h>
144 #include <sys/fs/zfs.h>
145
146 /*
147 * Basic routines to read and write from a vdev label.
148 * Used throughout the rest of this file.
149 */
150 uint64_t
151 vdev_label_offset(uint64_t psize, int l, uint64_t offset)
152 {
153 ASSERT(offset < sizeof (vdev_label_t));
154 ASSERT(P2PHASE_TYPED(psize, sizeof (vdev_label_t), uint64_t) == 0);
155
156 return (offset + l * sizeof (vdev_label_t) + (l < VDEV_LABELS / 2 ?
157 0 : psize - VDEV_LABELS * sizeof (vdev_label_t)));
158 }
159
160 /*
161 * Returns back the vdev label associated with the passed in offset.
162 */
163 int
164 vdev_label_number(uint64_t psize, uint64_t offset)
165 {
166 int l;
167
168 if (offset >= psize - VDEV_LABEL_END_SIZE) {
169 offset -= psize - VDEV_LABEL_END_SIZE;
170 offset += (VDEV_LABELS / 2) * sizeof (vdev_label_t);
171 }
172 l = offset / sizeof (vdev_label_t);
173 return (l < VDEV_LABELS ? l : -1);
174 }
175
176 static void
177 vdev_label_read(zio_t *zio, vdev_t *vd, int l, void *buf, uint64_t offset,
178 uint64_t size, zio_done_func_t *done, void *private, int flags)
179 {
180 ASSERT(spa_config_held(zio->io_spa, SCL_STATE_ALL, RW_WRITER) ==
181 SCL_STATE_ALL);
182 ASSERT(flags & ZIO_FLAG_CONFIG_WRITER);
183
184 zio_nowait(zio_read_phys(zio, vd,
185 vdev_label_offset(vd->vdev_psize, l, offset),
186 size, buf, ZIO_CHECKSUM_LABEL, done, private,
187 ZIO_PRIORITY_SYNC_READ, flags, B_TRUE));
188 }
189
190 static void
191 vdev_label_write(zio_t *zio, vdev_t *vd, int l, void *buf, uint64_t offset,
192 uint64_t size, zio_done_func_t *done, void *private, int flags)
193 {
194 ASSERT(spa_config_held(zio->io_spa, SCL_ALL, RW_WRITER) == SCL_ALL ||
195 (spa_config_held(zio->io_spa, SCL_CONFIG | SCL_STATE, RW_READER) ==
196 (SCL_CONFIG | SCL_STATE) &&
197 dsl_pool_sync_context(spa_get_dsl(zio->io_spa))));
198 ASSERT(flags & ZIO_FLAG_CONFIG_WRITER);
199
200 zio_nowait(zio_write_phys(zio, vd,
201 vdev_label_offset(vd->vdev_psize, l, offset),
202 size, buf, ZIO_CHECKSUM_LABEL, done, private,
203 ZIO_PRIORITY_SYNC_WRITE, flags, B_TRUE));
204 }
205
206 /*
207 * Generate the nvlist representing this vdev's config.
208 */
209 nvlist_t *
210 vdev_config_generate(spa_t *spa, vdev_t *vd, boolean_t getstats,
211 vdev_config_flag_t flags)
212 {
213 nvlist_t *nv = NULL;
214
215 VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0);
216
217 VERIFY(nvlist_add_string(nv, ZPOOL_CONFIG_TYPE,
218 vd->vdev_ops->vdev_op_type) == 0);
219 if (!(flags & (VDEV_CONFIG_SPARE | VDEV_CONFIG_L2CACHE)))
220 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_ID, vd->vdev_id)
221 == 0);
222 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_GUID, vd->vdev_guid) == 0);
223
224 if (vd->vdev_path != NULL)
225 VERIFY(nvlist_add_string(nv, ZPOOL_CONFIG_PATH,
226 vd->vdev_path) == 0);
227
228 if (vd->vdev_devid != NULL)
229 VERIFY(nvlist_add_string(nv, ZPOOL_CONFIG_DEVID,
230 vd->vdev_devid) == 0);
231
232 if (vd->vdev_physpath != NULL)
233 VERIFY(nvlist_add_string(nv, ZPOOL_CONFIG_PHYS_PATH,
234 vd->vdev_physpath) == 0);
235
236 if (vd->vdev_fru != NULL)
237 VERIFY(nvlist_add_string(nv, ZPOOL_CONFIG_FRU,
238 vd->vdev_fru) == 0);
239
240 if (vd->vdev_nparity != 0) {
241 ASSERT(strcmp(vd->vdev_ops->vdev_op_type,
242 VDEV_TYPE_RAIDZ) == 0);
243
244 /*
245 * Make sure someone hasn't managed to sneak a fancy new vdev
246 * into a crufty old storage pool.
247 */
248 ASSERT(vd->vdev_nparity == 1 ||
249 (vd->vdev_nparity <= 2 &&
250 spa_version(spa) >= SPA_VERSION_RAIDZ2) ||
251 (vd->vdev_nparity <= 3 &&
252 spa_version(spa) >= SPA_VERSION_RAIDZ3));
253
254 /*
255 * Note that we'll add the nparity tag even on storage pools
256 * that only support a single parity device -- older software
257 * will just ignore it.
258 */
259 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_NPARITY,
260 vd->vdev_nparity) == 0);
261 }
262
263 if (vd->vdev_wholedisk != -1ULL)
264 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK,
265 vd->vdev_wholedisk) == 0);
266
267 if (vd->vdev_not_present)
268 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_NOT_PRESENT, 1) == 0);
269
270 if (vd->vdev_isspare)
271 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_IS_SPARE, 1) == 0);
272
273 if (!(flags & (VDEV_CONFIG_SPARE | VDEV_CONFIG_L2CACHE)) &&
274 vd == vd->vdev_top) {
275 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_METASLAB_ARRAY,
276 vd->vdev_ms_array) == 0);
277 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_METASLAB_SHIFT,
278 vd->vdev_ms_shift) == 0);
279 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_ASHIFT,
280 vd->vdev_ashift) == 0);
281 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_ASIZE,
282 vd->vdev_asize) == 0);
283 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_IS_LOG,
284 vd->vdev_islog) == 0);
285 if (vd->vdev_removing)
286 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_REMOVING,
287 vd->vdev_removing) == 0);
288 }
289
290 if (vd->vdev_dtl_smo.smo_object != 0)
291 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_DTL,
292 vd->vdev_dtl_smo.smo_object) == 0);
293
294 if (vd->vdev_crtxg)
295 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_CREATE_TXG,
296 vd->vdev_crtxg) == 0);
297
298 if (getstats) {
299 vdev_stat_t vs;
300 pool_scan_stat_t ps;
301
302 vdev_get_stats(vd, &vs);
303 VERIFY(nvlist_add_uint64_array(nv, ZPOOL_CONFIG_VDEV_STATS,
304 (uint64_t *)&vs, sizeof (vs) / sizeof (uint64_t)) == 0);
305
306 /* provide either current or previous scan information */
307 if (spa_scan_get_stats(spa, &ps) == 0) {
308 VERIFY(nvlist_add_uint64_array(nv,
309 ZPOOL_CONFIG_SCAN_STATS, (uint64_t *)&ps,
310 sizeof (pool_scan_stat_t) / sizeof (uint64_t))
311 == 0);
312 }
313 }
314
315 if (!vd->vdev_ops->vdev_op_leaf) {
316 nvlist_t **child;
317 int c, idx;
318
319 ASSERT(!vd->vdev_ishole);
320
321 child = kmem_alloc(vd->vdev_children * sizeof (nvlist_t *),
322 KM_SLEEP);
323
324 for (c = 0, idx = 0; c < vd->vdev_children; c++) {
325 vdev_t *cvd = vd->vdev_child[c];
326
327 /*
328 * If we're generating an nvlist of removing
329 * vdevs then skip over any device which is
330 * not being removed.
331 */
332 if ((flags & VDEV_CONFIG_REMOVING) &&
333 !cvd->vdev_removing)
334 continue;
335
336 child[idx++] = vdev_config_generate(spa, cvd,
337 getstats, flags);
338 }
339
340 if (idx) {
341 VERIFY(nvlist_add_nvlist_array(nv,
342 ZPOOL_CONFIG_CHILDREN, child, idx) == 0);
343 }
344
345 for (c = 0; c < idx; c++)
346 nvlist_free(child[c]);
347
348 kmem_free(child, vd->vdev_children * sizeof (nvlist_t *));
349
350 } else {
351 const char *aux = NULL;
352
353 if (vd->vdev_offline && !vd->vdev_tmpoffline)
354 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_OFFLINE,
355 B_TRUE) == 0);
356 if (vd->vdev_resilvering)
357 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_RESILVERING,
358 B_TRUE) == 0);
359 if (vd->vdev_faulted)
360 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_FAULTED,
361 B_TRUE) == 0);
362 if (vd->vdev_degraded)
363 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_DEGRADED,
364 B_TRUE) == 0);
365 if (vd->vdev_removed)
366 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_REMOVED,
367 B_TRUE) == 0);
368 if (vd->vdev_unspare)
369 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_UNSPARE,
370 B_TRUE) == 0);
371 if (vd->vdev_ishole)
372 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_IS_HOLE,
373 B_TRUE) == 0);
374
375 switch (vd->vdev_stat.vs_aux) {
376 case VDEV_AUX_ERR_EXCEEDED:
377 aux = "err_exceeded";
378 break;
379
380 case VDEV_AUX_EXTERNAL:
381 aux = "external";
382 break;
383 }
384
385 if (aux != NULL)
386 VERIFY(nvlist_add_string(nv, ZPOOL_CONFIG_AUX_STATE,
387 aux) == 0);
388
389 if (vd->vdev_splitting && vd->vdev_orig_guid != 0LL) {
390 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_ORIG_GUID,
391 vd->vdev_orig_guid) == 0);
392 }
393 }
394
395 return (nv);
396 }
397
398 /*
399 * Generate a view of the top-level vdevs. If we currently have holes
400 * in the namespace, then generate an array which contains a list of holey
401 * vdevs. Additionally, add the number of top-level children that currently
402 * exist.
403 */
404 void
405 vdev_top_config_generate(spa_t *spa, nvlist_t *config)
406 {
407 vdev_t *rvd = spa->spa_root_vdev;
408 uint64_t *array;
409 uint_t c, idx;
410
411 array = kmem_alloc(rvd->vdev_children * sizeof (uint64_t), KM_SLEEP);
412
413 for (c = 0, idx = 0; c < rvd->vdev_children; c++) {
414 vdev_t *tvd = rvd->vdev_child[c];
415
416 if (tvd->vdev_ishole)
417 array[idx++] = c;
418 }
419
420 if (idx) {
421 VERIFY(nvlist_add_uint64_array(config, ZPOOL_CONFIG_HOLE_ARRAY,
422 array, idx) == 0);
423 }
424
425 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VDEV_CHILDREN,
426 rvd->vdev_children) == 0);
427
428 kmem_free(array, rvd->vdev_children * sizeof (uint64_t));
429 }
430
431 nvlist_t *
432 vdev_label_read_config(vdev_t *vd)
433 {
434 spa_t *spa = vd->vdev_spa;
435 nvlist_t *config = NULL;
436 vdev_phys_t *vp;
437 zio_t *zio;
438 int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL |
439 ZIO_FLAG_SPECULATIVE;
440
441 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
442
443 if (!vdev_readable(vd))
444 return (NULL);
445
446 vp = zio_buf_alloc(sizeof (vdev_phys_t));
447
448 retry:
449 for (int l = 0; l < VDEV_LABELS; l++) {
450
451 zio = zio_root(spa, NULL, NULL, flags);
452
453 vdev_label_read(zio, vd, l, vp,
454 offsetof(vdev_label_t, vl_vdev_phys),
455 sizeof (vdev_phys_t), NULL, NULL, flags);
456
457 if (zio_wait(zio) == 0 &&
458 nvlist_unpack(vp->vp_nvlist, sizeof (vp->vp_nvlist),
459 &config, 0) == 0)
460 break;
461
462 if (config != NULL) {
463 nvlist_free(config);
464 config = NULL;
465 }
466 }
467
468 if (config == NULL && !(flags & ZIO_FLAG_TRYHARD)) {
469 flags |= ZIO_FLAG_TRYHARD;
470 goto retry;
471 }
472
473 zio_buf_free(vp, sizeof (vdev_phys_t));
474
475 return (config);
476 }
477
478 /*
479 * Determine if a device is in use. The 'spare_guid' parameter will be filled
480 * in with the device guid if this spare is active elsewhere on the system.
481 */
482 static boolean_t
483 vdev_inuse(vdev_t *vd, uint64_t crtxg, vdev_labeltype_t reason,
484 uint64_t *spare_guid, uint64_t *l2cache_guid)
485 {
486 spa_t *spa = vd->vdev_spa;
487 uint64_t state, pool_guid, device_guid, txg, spare_pool;
488 uint64_t vdtxg = 0;
489 nvlist_t *label;
490
491 if (spare_guid)
492 *spare_guid = 0ULL;
493 if (l2cache_guid)
494 *l2cache_guid = 0ULL;
495
496 /*
497 * Read the label, if any, and perform some basic sanity checks.
498 */
499 if ((label = vdev_label_read_config(vd)) == NULL)
500 return (B_FALSE);
501
502 (void) nvlist_lookup_uint64(label, ZPOOL_CONFIG_CREATE_TXG,
503 &vdtxg);
504
505 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE,
506 &state) != 0 ||
507 nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID,
508 &device_guid) != 0) {
509 nvlist_free(label);
510 return (B_FALSE);
511 }
512
513 if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
514 (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_GUID,
515 &pool_guid) != 0 ||
516 nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG,
517 &txg) != 0)) {
518 nvlist_free(label);
519 return (B_FALSE);
520 }
521
522 nvlist_free(label);
523
524 /*
525 * Check to see if this device indeed belongs to the pool it claims to
526 * be a part of. The only way this is allowed is if the device is a hot
527 * spare (which we check for later on).
528 */
529 if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
530 !spa_guid_exists(pool_guid, device_guid) &&
531 !spa_spare_exists(device_guid, NULL, NULL) &&
532 !spa_l2cache_exists(device_guid, NULL))
533 return (B_FALSE);
534
535 /*
536 * If the transaction group is zero, then this an initialized (but
537 * unused) label. This is only an error if the create transaction
538 * on-disk is the same as the one we're using now, in which case the
539 * user has attempted to add the same vdev multiple times in the same
540 * transaction.
541 */
542 if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
543 txg == 0 && vdtxg == crtxg)
544 return (B_TRUE);
545
546 /*
547 * Check to see if this is a spare device. We do an explicit check for
548 * spa_has_spare() here because it may be on our pending list of spares
549 * to add. We also check if it is an l2cache device.
550 */
551 if (spa_spare_exists(device_guid, &spare_pool, NULL) ||
552 spa_has_spare(spa, device_guid)) {
553 if (spare_guid)
554 *spare_guid = device_guid;
555
556 switch (reason) {
557 case VDEV_LABEL_CREATE:
558 case VDEV_LABEL_L2CACHE:
559 return (B_TRUE);
560
561 case VDEV_LABEL_REPLACE:
562 return (!spa_has_spare(spa, device_guid) ||
563 spare_pool != 0ULL);
564
565 case VDEV_LABEL_SPARE:
566 return (spa_has_spare(spa, device_guid));
567 }
568 }
569
570 /*
571 * Check to see if this is an l2cache device.
572 */
573 if (spa_l2cache_exists(device_guid, NULL))
574 return (B_TRUE);
575
576 /*
577 * We can't rely on a pool's state if it's been imported
578 * read-only. Instead we look to see if the pools is marked
579 * read-only in the namespace and set the state to active.
580 */
581 if ((spa = spa_by_guid(pool_guid, device_guid)) != NULL &&
582 spa_mode(spa) == FREAD)
583 state = POOL_STATE_ACTIVE;
584
585 /*
586 * If the device is marked ACTIVE, then this device is in use by another
587 * pool on the system.
588 */
589 return (state == POOL_STATE_ACTIVE);
590 }
591
592 /*
593 * Initialize a vdev label. We check to make sure each leaf device is not in
594 * use, and writable. We put down an initial label which we will later
595 * overwrite with a complete label. Note that it's important to do this
596 * sequentially, not in parallel, so that we catch cases of multiple use of the
597 * same leaf vdev in the vdev we're creating -- e.g. mirroring a disk with
598 * itself.
599 */
600 int
601 vdev_label_init(vdev_t *vd, uint64_t crtxg, vdev_labeltype_t reason)
602 {
603 spa_t *spa = vd->vdev_spa;
604 nvlist_t *label;
605 vdev_phys_t *vp;
606 char *pad2;
607 uberblock_t *ub;
608 zio_t *zio;
609 char *buf;
610 size_t buflen;
611 int error;
612 uint64_t spare_guid, l2cache_guid;
613 int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL;
614
615 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
616
617 for (int c = 0; c < vd->vdev_children; c++)
618 if ((error = vdev_label_init(vd->vdev_child[c],
619 crtxg, reason)) != 0)
620 return (error);
621
622 /* Track the creation time for this vdev */
623 vd->vdev_crtxg = crtxg;
624
625 if (!vd->vdev_ops->vdev_op_leaf)
626 return (0);
627
628 /*
629 * Dead vdevs cannot be initialized.
630 */
631 if (vdev_is_dead(vd))
632 return (EIO);
633
634 /*
635 * Determine if the vdev is in use.
636 */
637 if (reason != VDEV_LABEL_REMOVE && reason != VDEV_LABEL_SPLIT &&
638 vdev_inuse(vd, crtxg, reason, &spare_guid, &l2cache_guid))
639 return (EBUSY);
640
641 /*
642 * If this is a request to add or replace a spare or l2cache device
643 * that is in use elsewhere on the system, then we must update the
644 * guid (which was initialized to a random value) to reflect the
645 * actual GUID (which is shared between multiple pools).
646 */
647 if (reason != VDEV_LABEL_REMOVE && reason != VDEV_LABEL_L2CACHE &&
648 spare_guid != 0ULL) {
649 uint64_t guid_delta = spare_guid - vd->vdev_guid;
650
651 vd->vdev_guid += guid_delta;
652
653 for (vdev_t *pvd = vd; pvd != NULL; pvd = pvd->vdev_parent)
654 pvd->vdev_guid_sum += guid_delta;
655
656 /*
657 * If this is a replacement, then we want to fallthrough to the
658 * rest of the code. If we're adding a spare, then it's already
659 * labeled appropriately and we can just return.
660 */
661 if (reason == VDEV_LABEL_SPARE)
662 return (0);
663 ASSERT(reason == VDEV_LABEL_REPLACE ||
664 reason == VDEV_LABEL_SPLIT);
665 }
666
667 if (reason != VDEV_LABEL_REMOVE && reason != VDEV_LABEL_SPARE &&
668 l2cache_guid != 0ULL) {
669 uint64_t guid_delta = l2cache_guid - vd->vdev_guid;
670
671 vd->vdev_guid += guid_delta;
672
673 for (vdev_t *pvd = vd; pvd != NULL; pvd = pvd->vdev_parent)
674 pvd->vdev_guid_sum += guid_delta;
675
676 /*
677 * If this is a replacement, then we want to fallthrough to the
678 * rest of the code. If we're adding an l2cache, then it's
679 * already labeled appropriately and we can just return.
680 */
681 if (reason == VDEV_LABEL_L2CACHE)
682 return (0);
683 ASSERT(reason == VDEV_LABEL_REPLACE);
684 }
685
686 /*
687 * Initialize its label.
688 */
689 vp = zio_buf_alloc(sizeof (vdev_phys_t));
690 bzero(vp, sizeof (vdev_phys_t));
691
692 /*
693 * Generate a label describing the pool and our top-level vdev.
694 * We mark it as being from txg 0 to indicate that it's not
695 * really part of an active pool just yet. The labels will
696 * be written again with a meaningful txg by spa_sync().
697 */
698 if (reason == VDEV_LABEL_SPARE ||
699 (reason == VDEV_LABEL_REMOVE && vd->vdev_isspare)) {
700 /*
701 * For inactive hot spares, we generate a special label that
702 * identifies as a mutually shared hot spare. We write the
703 * label if we are adding a hot spare, or if we are removing an
704 * active hot spare (in which case we want to revert the
705 * labels).
706 */
707 VERIFY(nvlist_alloc(&label, NV_UNIQUE_NAME, KM_SLEEP) == 0);
708
709 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_VERSION,
710 spa_version(spa)) == 0);
711 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_POOL_STATE,
712 POOL_STATE_SPARE) == 0);
713 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_GUID,
714 vd->vdev_guid) == 0);
715 } else if (reason == VDEV_LABEL_L2CACHE ||
716 (reason == VDEV_LABEL_REMOVE && vd->vdev_isl2cache)) {
717 /*
718 * For level 2 ARC devices, add a special label.
719 */
720 VERIFY(nvlist_alloc(&label, NV_UNIQUE_NAME, KM_SLEEP) == 0);
721
722 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_VERSION,
723 spa_version(spa)) == 0);
724 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_POOL_STATE,
725 POOL_STATE_L2CACHE) == 0);
726 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_GUID,
727 vd->vdev_guid) == 0);
728 } else {
729 uint64_t txg = 0ULL;
730
731 if (reason == VDEV_LABEL_SPLIT)
732 txg = spa->spa_uberblock.ub_txg;
733 label = spa_config_generate(spa, vd, txg, B_FALSE);
734
735 /*
736 * Add our creation time. This allows us to detect multiple
737 * vdev uses as described above, and automatically expires if we
738 * fail.
739 */
740 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_CREATE_TXG,
741 crtxg) == 0);
742 }
743
744 buf = vp->vp_nvlist;
745 buflen = sizeof (vp->vp_nvlist);
746
747 error = nvlist_pack(label, &buf, &buflen, NV_ENCODE_XDR, KM_SLEEP);
748 if (error != 0) {
749 nvlist_free(label);
750 zio_buf_free(vp, sizeof (vdev_phys_t));
751 /* EFAULT means nvlist_pack ran out of room */
752 return (error == EFAULT ? ENAMETOOLONG : EINVAL);
753 }
754
755 /*
756 * Initialize uberblock template.
757 */
758 ub = zio_buf_alloc(VDEV_UBERBLOCK_RING);
759 bzero(ub, VDEV_UBERBLOCK_RING);
760 *ub = spa->spa_uberblock;
761 ub->ub_txg = 0;
762
763 /* Initialize the 2nd padding area. */
764 pad2 = zio_buf_alloc(VDEV_PAD_SIZE);
765 bzero(pad2, VDEV_PAD_SIZE);
766
767 /*
768 * Write everything in parallel.
769 */
770 retry:
771 zio = zio_root(spa, NULL, NULL, flags);
772
773 for (int l = 0; l < VDEV_LABELS; l++) {
774
775 vdev_label_write(zio, vd, l, vp,
776 offsetof(vdev_label_t, vl_vdev_phys),
777 sizeof (vdev_phys_t), NULL, NULL, flags);
778
779 /*
780 * Skip the 1st padding area.
781 * Zero out the 2nd padding area where it might have
782 * left over data from previous filesystem format.
783 */
784 vdev_label_write(zio, vd, l, pad2,
785 offsetof(vdev_label_t, vl_pad2),
786 VDEV_PAD_SIZE, NULL, NULL, flags);
787
788 vdev_label_write(zio, vd, l, ub,
789 offsetof(vdev_label_t, vl_uberblock),
790 VDEV_UBERBLOCK_RING, NULL, NULL, flags);
791 }
792
793 error = zio_wait(zio);
794
795 if (error != 0 && !(flags & ZIO_FLAG_TRYHARD)) {
796 flags |= ZIO_FLAG_TRYHARD;
797 goto retry;
798 }
799
800 nvlist_free(label);
801 zio_buf_free(pad2, VDEV_PAD_SIZE);
802 zio_buf_free(ub, VDEV_UBERBLOCK_RING);
803 zio_buf_free(vp, sizeof (vdev_phys_t));
804
805 /*
806 * If this vdev hasn't been previously identified as a spare, then we
807 * mark it as such only if a) we are labeling it as a spare, or b) it
808 * exists as a spare elsewhere in the system. Do the same for
809 * level 2 ARC devices.
810 */
811 if (error == 0 && !vd->vdev_isspare &&
812 (reason == VDEV_LABEL_SPARE ||
813 spa_spare_exists(vd->vdev_guid, NULL, NULL)))
814 spa_spare_add(vd);
815
816 if (error == 0 && !vd->vdev_isl2cache &&
817 (reason == VDEV_LABEL_L2CACHE ||
818 spa_l2cache_exists(vd->vdev_guid, NULL)))
819 spa_l2cache_add(vd);
820
821 return (error);
822 }
823
824 /*
825 * ==========================================================================
826 * uberblock load/sync
827 * ==========================================================================
828 */
829
830 /*
831 * Consider the following situation: txg is safely synced to disk. We've
832 * written the first uberblock for txg + 1, and then we lose power. When we
833 * come back up, we fail to see the uberblock for txg + 1 because, say,
834 * it was on a mirrored device and the replica to which we wrote txg + 1
835 * is now offline. If we then make some changes and sync txg + 1, and then
836 * the missing replica comes back, then for a new seconds we'll have two
837 * conflicting uberblocks on disk with the same txg. The solution is simple:
838 * among uberblocks with equal txg, choose the one with the latest timestamp.
839 */
840 static int
841 vdev_uberblock_compare(uberblock_t *ub1, uberblock_t *ub2)
842 {
843 if (ub1->ub_txg < ub2->ub_txg)
844 return (-1);
845 if (ub1->ub_txg > ub2->ub_txg)
846 return (1);
847
848 if (ub1->ub_timestamp < ub2->ub_timestamp)
849 return (-1);
850 if (ub1->ub_timestamp > ub2->ub_timestamp)
851 return (1);
852
853 return (0);
854 }
855
856 static void
857 vdev_uberblock_load_done(zio_t *zio)
858 {
859 spa_t *spa = zio->io_spa;
860 zio_t *rio = zio->io_private;
861 uberblock_t *ub = zio->io_data;
862 uberblock_t *ubbest = rio->io_private;
863
864 ASSERT3U(zio->io_size, ==, VDEV_UBERBLOCK_SIZE(zio->io_vd));
865
866 if (zio->io_error == 0 && uberblock_verify(ub) == 0) {
867 mutex_enter(&rio->io_lock);
868 if (ub->ub_txg <= spa->spa_load_max_txg &&
869 vdev_uberblock_compare(ub, ubbest) > 0)
870 *ubbest = *ub;
871 mutex_exit(&rio->io_lock);
872 }
873
874 zio_buf_free(zio->io_data, zio->io_size);
875 }
876
877 void
878 vdev_uberblock_load(zio_t *zio, vdev_t *vd, uberblock_t *ubbest)
879 {
880 spa_t *spa = vd->vdev_spa;
881 vdev_t *rvd = spa->spa_root_vdev;
882 int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL |
883 ZIO_FLAG_SPECULATIVE | ZIO_FLAG_TRYHARD;
884
885 if (vd == rvd) {
886 ASSERT(zio == NULL);
887 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
888 zio = zio_root(spa, NULL, ubbest, flags);
889 bzero(ubbest, sizeof (uberblock_t));
890 }
891
892 ASSERT(zio != NULL);
893
894 for (int c = 0; c < vd->vdev_children; c++)
895 vdev_uberblock_load(zio, vd->vdev_child[c], ubbest);
896
897 if (vd->vdev_ops->vdev_op_leaf && vdev_readable(vd)) {
898 for (int l = 0; l < VDEV_LABELS; l++) {
899 for (int n = 0; n < VDEV_UBERBLOCK_COUNT(vd); n++) {
900 vdev_label_read(zio, vd, l,
901 zio_buf_alloc(VDEV_UBERBLOCK_SIZE(vd)),
902 VDEV_UBERBLOCK_OFFSET(vd, n),
903 VDEV_UBERBLOCK_SIZE(vd),
904 vdev_uberblock_load_done, zio, flags);
905 }
906 }
907 }
908
909 if (vd == rvd) {
910 (void) zio_wait(zio);
911 spa_config_exit(spa, SCL_ALL, FTAG);
912 }
913 }
914
915 /*
916 * On success, increment root zio's count of good writes.
917 * We only get credit for writes to known-visible vdevs; see spa_vdev_add().
918 */
919 static void
920 vdev_uberblock_sync_done(zio_t *zio)
921 {
922 uint64_t *good_writes = zio->io_private;
923
924 if (zio->io_error == 0 && zio->io_vd->vdev_top->vdev_ms_array != 0)
925 atomic_add_64(good_writes, 1);
926 }
927
928 /*
929 * Write the uberblock to all labels of all leaves of the specified vdev.
930 */
931 static void
932 vdev_uberblock_sync(zio_t *zio, uberblock_t *ub, vdev_t *vd, int flags)
933 {
934 uberblock_t *ubbuf;
935 int n;
936
937 for (int c = 0; c < vd->vdev_children; c++)
938 vdev_uberblock_sync(zio, ub, vd->vdev_child[c], flags);
939
940 if (!vd->vdev_ops->vdev_op_leaf)
941 return;
942
943 if (!vdev_writeable(vd))
944 return;
945
946 n = ub->ub_txg & (VDEV_UBERBLOCK_COUNT(vd) - 1);
947
948 ubbuf = zio_buf_alloc(VDEV_UBERBLOCK_SIZE(vd));
949 bzero(ubbuf, VDEV_UBERBLOCK_SIZE(vd));
950 *ubbuf = *ub;
951
952 for (int l = 0; l < VDEV_LABELS; l++)
953 vdev_label_write(zio, vd, l, ubbuf,
954 VDEV_UBERBLOCK_OFFSET(vd, n), VDEV_UBERBLOCK_SIZE(vd),
955 vdev_uberblock_sync_done, zio->io_private,
956 flags | ZIO_FLAG_DONT_PROPAGATE);
957
958 zio_buf_free(ubbuf, VDEV_UBERBLOCK_SIZE(vd));
959 }
960
961 int
962 vdev_uberblock_sync_list(vdev_t **svd, int svdcount, uberblock_t *ub, int flags)
963 {
964 spa_t *spa = svd[0]->vdev_spa;
965 zio_t *zio;
966 uint64_t good_writes = 0;
967
968 zio = zio_root(spa, NULL, &good_writes, flags);
969
970 for (int v = 0; v < svdcount; v++)
971 vdev_uberblock_sync(zio, ub, svd[v], flags);
972
973 (void) zio_wait(zio);
974
975 /*
976 * Flush the uberblocks to disk. This ensures that the odd labels
977 * are no longer needed (because the new uberblocks and the even
978 * labels are safely on disk), so it is safe to overwrite them.
979 */
980 zio = zio_root(spa, NULL, NULL, flags);
981
982 for (int v = 0; v < svdcount; v++)
983 zio_flush(zio, svd[v]);
984
985 (void) zio_wait(zio);
986
987 return (good_writes >= 1 ? 0 : EIO);
988 }
989
990 /*
991 * On success, increment the count of good writes for our top-level vdev.
992 */
993 static void
994 vdev_label_sync_done(zio_t *zio)
995 {
996 uint64_t *good_writes = zio->io_private;
997
998 if (zio->io_error == 0)
999 atomic_add_64(good_writes, 1);
1000 }
1001
1002 /*
1003 * If there weren't enough good writes, indicate failure to the parent.
1004 */
1005 static void
1006 vdev_label_sync_top_done(zio_t *zio)
1007 {
1008 uint64_t *good_writes = zio->io_private;
1009
1010 if (*good_writes == 0)
1011 zio->io_error = EIO;
1012
1013 kmem_free(good_writes, sizeof (uint64_t));
1014 }
1015
1016 /*
1017 * We ignore errors for log and cache devices, simply free the private data.
1018 */
1019 static void
1020 vdev_label_sync_ignore_done(zio_t *zio)
1021 {
1022 kmem_free(zio->io_private, sizeof (uint64_t));
1023 }
1024
1025 /*
1026 * Write all even or odd labels to all leaves of the specified vdev.
1027 */
1028 static void
1029 vdev_label_sync(zio_t *zio, vdev_t *vd, int l, uint64_t txg, int flags)
1030 {
1031 nvlist_t *label;
1032 vdev_phys_t *vp;
1033 char *buf;
1034 size_t buflen;
1035
1036 for (int c = 0; c < vd->vdev_children; c++)
1037 vdev_label_sync(zio, vd->vdev_child[c], l, txg, flags);
1038
1039 if (!vd->vdev_ops->vdev_op_leaf)
1040 return;
1041
1042 if (!vdev_writeable(vd))
1043 return;
1044
1045 /*
1046 * Generate a label describing the top-level config to which we belong.
1047 */
1048 label = spa_config_generate(vd->vdev_spa, vd, txg, B_FALSE);
1049
1050 vp = zio_buf_alloc(sizeof (vdev_phys_t));
1051 bzero(vp, sizeof (vdev_phys_t));
1052
1053 buf = vp->vp_nvlist;
1054 buflen = sizeof (vp->vp_nvlist);
1055
1056 if (nvlist_pack(label, &buf, &buflen, NV_ENCODE_XDR, KM_SLEEP) == 0) {
1057 for (; l < VDEV_LABELS; l += 2) {
1058 vdev_label_write(zio, vd, l, vp,
1059 offsetof(vdev_label_t, vl_vdev_phys),
1060 sizeof (vdev_phys_t),
1061 vdev_label_sync_done, zio->io_private,
1062 flags | ZIO_FLAG_DONT_PROPAGATE);
1063 }
1064 }
1065
1066 zio_buf_free(vp, sizeof (vdev_phys_t));
1067 nvlist_free(label);
1068 }
1069
1070 int
1071 vdev_label_sync_list(spa_t *spa, int l, uint64_t txg, int flags)
1072 {
1073 list_t *dl = &spa->spa_config_dirty_list;
1074 vdev_t *vd;
1075 zio_t *zio;
1076 int error;
1077
1078 /*
1079 * Write the new labels to disk.
1080 */
1081 zio = zio_root(spa, NULL, NULL, flags);
1082
1083 for (vd = list_head(dl); vd != NULL; vd = list_next(dl, vd)) {
1084 uint64_t *good_writes = kmem_zalloc(sizeof (uint64_t),
1085 KM_SLEEP);
1086
1087 ASSERT(!vd->vdev_ishole);
1088
1089 zio_t *vio = zio_null(zio, spa, NULL,
1090 (vd->vdev_islog || vd->vdev_aux != NULL) ?
1091 vdev_label_sync_ignore_done : vdev_label_sync_top_done,
1092 good_writes, flags);
1093 vdev_label_sync(vio, vd, l, txg, flags);
1094 zio_nowait(vio);
1095 }
1096
1097 error = zio_wait(zio);
1098
1099 /*
1100 * Flush the new labels to disk.
1101 */
1102 zio = zio_root(spa, NULL, NULL, flags);
1103
1104 for (vd = list_head(dl); vd != NULL; vd = list_next(dl, vd))
1105 zio_flush(zio, vd);
1106
1107 (void) zio_wait(zio);
1108
1109 return (error);
1110 }
1111
1112 /*
1113 * Sync the uberblock and any changes to the vdev configuration.
1114 *
1115 * The order of operations is carefully crafted to ensure that
1116 * if the system panics or loses power at any time, the state on disk
1117 * is still transactionally consistent. The in-line comments below
1118 * describe the failure semantics at each stage.
1119 *
1120 * Moreover, vdev_config_sync() is designed to be idempotent: if it fails
1121 * at any time, you can just call it again, and it will resume its work.
1122 */
1123 int
1124 vdev_config_sync(vdev_t **svd, int svdcount, uint64_t txg, boolean_t tryhard)
1125 {
1126 spa_t *spa = svd[0]->vdev_spa;
1127 uberblock_t *ub = &spa->spa_uberblock;
1128 vdev_t *vd;
1129 zio_t *zio;
1130 int error;
1131 int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL;
1132
1133 /*
1134 * Normally, we don't want to try too hard to write every label and
1135 * uberblock. If there is a flaky disk, we don't want the rest of the
1136 * sync process to block while we retry. But if we can't write a
1137 * single label out, we should retry with ZIO_FLAG_TRYHARD before
1138 * bailing out and declaring the pool faulted.
1139 */
1140 if (tryhard)
1141 flags |= ZIO_FLAG_TRYHARD;
1142
1143 ASSERT(ub->ub_txg <= txg);
1144
1145 /*
1146 * If this isn't a resync due to I/O errors,
1147 * and nothing changed in this transaction group,
1148 * and the vdev configuration hasn't changed,
1149 * then there's nothing to do.
1150 */
1151 if (ub->ub_txg < txg &&
1152 uberblock_update(ub, spa->spa_root_vdev, txg) == B_FALSE &&
1153 list_is_empty(&spa->spa_config_dirty_list))
1154 return (0);
1155
1156 if (txg > spa_freeze_txg(spa))
1157 return (0);
1158
1159 ASSERT(txg <= spa->spa_final_txg);
1160
1161 /*
1162 * Flush the write cache of every disk that's been written to
1163 * in this transaction group. This ensures that all blocks
1164 * written in this txg will be committed to stable storage
1165 * before any uberblock that references them.
1166 */
1167 zio = zio_root(spa, NULL, NULL, flags);
1168
1169 for (vd = txg_list_head(&spa->spa_vdev_txg_list, TXG_CLEAN(txg)); vd;
1170 vd = txg_list_next(&spa->spa_vdev_txg_list, vd, TXG_CLEAN(txg)))
1171 zio_flush(zio, vd);
1172
1173 (void) zio_wait(zio);
1174
1175 /*
1176 * Sync out the even labels (L0, L2) for every dirty vdev. If the
1177 * system dies in the middle of this process, that's OK: all of the
1178 * even labels that made it to disk will be newer than any uberblock,
1179 * and will therefore be considered invalid. The odd labels (L1, L3),
1180 * which have not yet been touched, will still be valid. We flush
1181 * the new labels to disk to ensure that all even-label updates
1182 * are committed to stable storage before the uberblock update.
1183 */
1184 if ((error = vdev_label_sync_list(spa, 0, txg, flags)) != 0)
1185 return (error);
1186
1187 /*
1188 * Sync the uberblocks to all vdevs in svd[].
1189 * If the system dies in the middle of this step, there are two cases
1190 * to consider, and the on-disk state is consistent either way:
1191 *
1192 * (1) If none of the new uberblocks made it to disk, then the
1193 * previous uberblock will be the newest, and the odd labels
1194 * (which had not yet been touched) will be valid with respect
1195 * to that uberblock.
1196 *
1197 * (2) If one or more new uberblocks made it to disk, then they
1198 * will be the newest, and the even labels (which had all
1199 * been successfully committed) will be valid with respect
1200 * to the new uberblocks.
1201 */
1202 if ((error = vdev_uberblock_sync_list(svd, svdcount, ub, flags)) != 0)
1203 return (error);
1204
1205 /*
1206 * Sync out odd labels for every dirty vdev. If the system dies
1207 * in the middle of this process, the even labels and the new
1208 * uberblocks will suffice to open the pool. The next time
1209 * the pool is opened, the first thing we'll do -- before any
1210 * user data is modified -- is mark every vdev dirty so that
1211 * all labels will be brought up to date. We flush the new labels
1212 * to disk to ensure that all odd-label updates are committed to
1213 * stable storage before the next transaction group begins.
1214 */
1215 return (vdev_label_sync_list(spa, 1, txg, flags));
1216 }