]> git.proxmox.com Git - mirror_zfs-debian.git/blame - zfs/lib/libzpool/vdev_label.c
Initial Linux ZFS GIT Repo
[mirror_zfs-debian.git] / zfs / lib / libzpool / vdev_label.c
CommitLineData
34dc7c2f
BB
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 2007 Sun Microsystems, Inc. All rights reserved.
23 * Use is subject to license terms.
24 */
25
26#pragma ident "@(#)vdev_label.c 1.18 07/12/12 SMI"
27
28/*
29 * Virtual Device Labels
30 * ---------------------
31 *
32 * The vdev label serves several distinct purposes:
33 *
34 * 1. Uniquely identify this device as part of a ZFS pool and confirm its
35 * identity within the pool.
36 *
37 * 2. Verify that all the devices given in a configuration are present
38 * within the pool.
39 *
40 * 3. Determine the uberblock for the pool.
41 *
42 * 4. In case of an import operation, determine the configuration of the
43 * toplevel vdev of which it is a part.
44 *
45 * 5. If an import operation cannot find all the devices in the pool,
46 * provide enough information to the administrator to determine which
47 * devices are missing.
48 *
49 * It is important to note that while the kernel is responsible for writing the
50 * label, it only consumes the information in the first three cases. The
51 * latter information is only consumed in userland when determining the
52 * configuration to import a pool.
53 *
54 *
55 * Label Organization
56 * ------------------
57 *
58 * Before describing the contents of the label, it's important to understand how
59 * the labels are written and updated with respect to the uberblock.
60 *
61 * When the pool configuration is altered, either because it was newly created
62 * or a device was added, we want to update all the labels such that we can deal
63 * with fatal failure at any point. To this end, each disk has two labels which
64 * are updated before and after the uberblock is synced. Assuming we have
65 * labels and an uberblock with the following transaction groups:
66 *
67 * L1 UB L2
68 * +------+ +------+ +------+
69 * | | | | | |
70 * | t10 | | t10 | | t10 |
71 * | | | | | |
72 * +------+ +------+ +------+
73 *
74 * In this stable state, the labels and the uberblock were all updated within
75 * the same transaction group (10). Each label is mirrored and checksummed, so
76 * that we can detect when we fail partway through writing the label.
77 *
78 * In order to identify which labels are valid, the labels are written in the
79 * following manner:
80 *
81 * 1. For each vdev, update 'L1' to the new label
82 * 2. Update the uberblock
83 * 3. For each vdev, update 'L2' to the new label
84 *
85 * Given arbitrary failure, we can determine the correct label to use based on
86 * the transaction group. If we fail after updating L1 but before updating the
87 * UB, we will notice that L1's transaction group is greater than the uberblock,
88 * so L2 must be valid. If we fail after writing the uberblock but before
89 * writing L2, we will notice that L2's transaction group is less than L1, and
90 * therefore L1 is valid.
91 *
92 * Another added complexity is that not every label is updated when the config
93 * is synced. If we add a single device, we do not want to have to re-write
94 * every label for every device in the pool. This means that both L1 and L2 may
95 * be older than the pool uberblock, because the necessary information is stored
96 * on another vdev.
97 *
98 *
99 * On-disk Format
100 * --------------
101 *
102 * The vdev label consists of two distinct parts, and is wrapped within the
103 * vdev_label_t structure. The label includes 8k of padding to permit legacy
104 * VTOC disk labels, but is otherwise ignored.
105 *
106 * The first half of the label is a packed nvlist which contains pool wide
107 * properties, per-vdev properties, and configuration information. It is
108 * described in more detail below.
109 *
110 * The latter half of the label consists of a redundant array of uberblocks.
111 * These uberblocks are updated whenever a transaction group is committed,
112 * or when the configuration is updated. When a pool is loaded, we scan each
113 * vdev for the 'best' uberblock.
114 *
115 *
116 * Configuration Information
117 * -------------------------
118 *
119 * The nvlist describing the pool and vdev contains the following elements:
120 *
121 * version ZFS on-disk version
122 * name Pool name
123 * state Pool state
124 * txg Transaction group in which this label was written
125 * pool_guid Unique identifier for this pool
126 * vdev_tree An nvlist describing vdev tree.
127 *
128 * Each leaf device label also contains the following:
129 *
130 * top_guid Unique ID for top-level vdev in which this is contained
131 * guid Unique ID for the leaf vdev
132 *
133 * The 'vs' configuration follows the format described in 'spa_config.c'.
134 */
135
136#include <sys/zfs_context.h>
137#include <sys/spa.h>
138#include <sys/spa_impl.h>
139#include <sys/dmu.h>
140#include <sys/zap.h>
141#include <sys/vdev.h>
142#include <sys/vdev_impl.h>
143#include <sys/uberblock_impl.h>
144#include <sys/metaslab.h>
145#include <sys/zio.h>
146#include <sys/fs/zfs.h>
147
148/*
149 * Basic routines to read and write from a vdev label.
150 * Used throughout the rest of this file.
151 */
152uint64_t
153vdev_label_offset(uint64_t psize, int l, uint64_t offset)
154{
155 ASSERT(offset < sizeof (vdev_label_t));
156 ASSERT(P2PHASE_TYPED(psize, sizeof (vdev_label_t), uint64_t) == 0);
157
158 return (offset + l * sizeof (vdev_label_t) + (l < VDEV_LABELS / 2 ?
159 0 : psize - VDEV_LABELS * sizeof (vdev_label_t)));
160}
161
162static void
163vdev_label_read(zio_t *zio, vdev_t *vd, int l, void *buf, uint64_t offset,
164 uint64_t size, zio_done_func_t *done, void *private)
165{
166 ASSERT(vd->vdev_children == 0);
167
168 zio_nowait(zio_read_phys(zio, vd,
169 vdev_label_offset(vd->vdev_psize, l, offset),
170 size, buf, ZIO_CHECKSUM_LABEL, done, private,
171 ZIO_PRIORITY_SYNC_READ,
172 ZIO_FLAG_CONFIG_HELD | ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE,
173 B_TRUE));
174}
175
176static void
177vdev_label_write(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(vd->vdev_children == 0);
181
182 zio_nowait(zio_write_phys(zio, vd,
183 vdev_label_offset(vd->vdev_psize, l, offset),
184 size, buf, ZIO_CHECKSUM_LABEL, done, private,
185 ZIO_PRIORITY_SYNC_WRITE, flags, B_TRUE));
186}
187
188/*
189 * Generate the nvlist representing this vdev's config.
190 */
191nvlist_t *
192vdev_config_generate(spa_t *spa, vdev_t *vd, boolean_t getstats,
193 boolean_t isspare, boolean_t isl2cache)
194{
195 nvlist_t *nv = NULL;
196
197 VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0);
198
199 VERIFY(nvlist_add_string(nv, ZPOOL_CONFIG_TYPE,
200 vd->vdev_ops->vdev_op_type) == 0);
201 if (!isspare && !isl2cache)
202 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_ID, vd->vdev_id)
203 == 0);
204 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_GUID, vd->vdev_guid) == 0);
205
206 if (vd->vdev_path != NULL)
207 VERIFY(nvlist_add_string(nv, ZPOOL_CONFIG_PATH,
208 vd->vdev_path) == 0);
209
210 if (vd->vdev_devid != NULL)
211 VERIFY(nvlist_add_string(nv, ZPOOL_CONFIG_DEVID,
212 vd->vdev_devid) == 0);
213
214 if (vd->vdev_physpath != NULL)
215 VERIFY(nvlist_add_string(nv, ZPOOL_CONFIG_PHYS_PATH,
216 vd->vdev_physpath) == 0);
217
218 if (vd->vdev_nparity != 0) {
219 ASSERT(strcmp(vd->vdev_ops->vdev_op_type,
220 VDEV_TYPE_RAIDZ) == 0);
221
222 /*
223 * Make sure someone hasn't managed to sneak a fancy new vdev
224 * into a crufty old storage pool.
225 */
226 ASSERT(vd->vdev_nparity == 1 ||
227 (vd->vdev_nparity == 2 &&
228 spa_version(spa) >= SPA_VERSION_RAID6));
229
230 /*
231 * Note that we'll add the nparity tag even on storage pools
232 * that only support a single parity device -- older software
233 * will just ignore it.
234 */
235 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_NPARITY,
236 vd->vdev_nparity) == 0);
237 }
238
239 if (vd->vdev_wholedisk != -1ULL)
240 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK,
241 vd->vdev_wholedisk) == 0);
242
243 if (vd->vdev_not_present)
244 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_NOT_PRESENT, 1) == 0);
245
246 if (vd->vdev_isspare)
247 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_IS_SPARE, 1) == 0);
248
249 if (!isspare && !isl2cache && vd == vd->vdev_top) {
250 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_METASLAB_ARRAY,
251 vd->vdev_ms_array) == 0);
252 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_METASLAB_SHIFT,
253 vd->vdev_ms_shift) == 0);
254 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_ASHIFT,
255 vd->vdev_ashift) == 0);
256 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_ASIZE,
257 vd->vdev_asize) == 0);
258 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_IS_LOG,
259 vd->vdev_islog) == 0);
260 }
261
262 if (vd->vdev_dtl.smo_object != 0)
263 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_DTL,
264 vd->vdev_dtl.smo_object) == 0);
265
266 if (getstats) {
267 vdev_stat_t vs;
268 vdev_get_stats(vd, &vs);
269 VERIFY(nvlist_add_uint64_array(nv, ZPOOL_CONFIG_STATS,
270 (uint64_t *)&vs, sizeof (vs) / sizeof (uint64_t)) == 0);
271 }
272
273 if (!vd->vdev_ops->vdev_op_leaf) {
274 nvlist_t **child;
275 int c;
276
277 child = kmem_alloc(vd->vdev_children * sizeof (nvlist_t *),
278 KM_SLEEP);
279
280 for (c = 0; c < vd->vdev_children; c++)
281 child[c] = vdev_config_generate(spa, vd->vdev_child[c],
282 getstats, isspare, isl2cache);
283
284 VERIFY(nvlist_add_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
285 child, vd->vdev_children) == 0);
286
287 for (c = 0; c < vd->vdev_children; c++)
288 nvlist_free(child[c]);
289
290 kmem_free(child, vd->vdev_children * sizeof (nvlist_t *));
291
292 } else {
293 if (vd->vdev_offline && !vd->vdev_tmpoffline)
294 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_OFFLINE,
295 B_TRUE) == 0);
296 if (vd->vdev_faulted)
297 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_FAULTED,
298 B_TRUE) == 0);
299 if (vd->vdev_degraded)
300 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_DEGRADED,
301 B_TRUE) == 0);
302 if (vd->vdev_removed)
303 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_REMOVED,
304 B_TRUE) == 0);
305 if (vd->vdev_unspare)
306 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_UNSPARE,
307 B_TRUE) == 0);
308 }
309
310 return (nv);
311}
312
313nvlist_t *
314vdev_label_read_config(vdev_t *vd)
315{
316 spa_t *spa = vd->vdev_spa;
317 nvlist_t *config = NULL;
318 vdev_phys_t *vp;
319 zio_t *zio;
320 int l;
321
322 ASSERT(spa_config_held(spa, RW_READER) ||
323 spa_config_held(spa, RW_WRITER));
324
325 if (!vdev_readable(vd))
326 return (NULL);
327
328 vp = zio_buf_alloc(sizeof (vdev_phys_t));
329
330 for (l = 0; l < VDEV_LABELS; l++) {
331
332 zio = zio_root(spa, NULL, NULL, ZIO_FLAG_CANFAIL |
333 ZIO_FLAG_SPECULATIVE | ZIO_FLAG_CONFIG_HELD);
334
335 vdev_label_read(zio, vd, l, vp,
336 offsetof(vdev_label_t, vl_vdev_phys),
337 sizeof (vdev_phys_t), NULL, NULL);
338
339 if (zio_wait(zio) == 0 &&
340 nvlist_unpack(vp->vp_nvlist, sizeof (vp->vp_nvlist),
341 &config, 0) == 0)
342 break;
343
344 if (config != NULL) {
345 nvlist_free(config);
346 config = NULL;
347 }
348 }
349
350 zio_buf_free(vp, sizeof (vdev_phys_t));
351
352 return (config);
353}
354
355/*
356 * Determine if a device is in use. The 'spare_guid' parameter will be filled
357 * in with the device guid if this spare is active elsewhere on the system.
358 */
359static boolean_t
360vdev_inuse(vdev_t *vd, uint64_t crtxg, vdev_labeltype_t reason,
361 uint64_t *spare_guid, uint64_t *l2cache_guid)
362{
363 spa_t *spa = vd->vdev_spa;
364 uint64_t state, pool_guid, device_guid, txg, spare_pool;
365 uint64_t vdtxg = 0;
366 nvlist_t *label;
367
368 if (spare_guid)
369 *spare_guid = 0ULL;
370 if (l2cache_guid)
371 *l2cache_guid = 0ULL;
372
373 /*
374 * Read the label, if any, and perform some basic sanity checks.
375 */
376 if ((label = vdev_label_read_config(vd)) == NULL)
377 return (B_FALSE);
378
379 (void) nvlist_lookup_uint64(label, ZPOOL_CONFIG_CREATE_TXG,
380 &vdtxg);
381
382 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE,
383 &state) != 0 ||
384 nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID,
385 &device_guid) != 0) {
386 nvlist_free(label);
387 return (B_FALSE);
388 }
389
390 if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
391 (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_GUID,
392 &pool_guid) != 0 ||
393 nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG,
394 &txg) != 0)) {
395 nvlist_free(label);
396 return (B_FALSE);
397 }
398
399 nvlist_free(label);
400
401 /*
402 * Check to see if this device indeed belongs to the pool it claims to
403 * be a part of. The only way this is allowed is if the device is a hot
404 * spare (which we check for later on).
405 */
406 if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
407 !spa_guid_exists(pool_guid, device_guid) &&
408 !spa_spare_exists(device_guid, NULL) &&
409 !spa_l2cache_exists(device_guid, NULL))
410 return (B_FALSE);
411
412 /*
413 * If the transaction group is zero, then this an initialized (but
414 * unused) label. This is only an error if the create transaction
415 * on-disk is the same as the one we're using now, in which case the
416 * user has attempted to add the same vdev multiple times in the same
417 * transaction.
418 */
419 if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
420 txg == 0 && vdtxg == crtxg)
421 return (B_TRUE);
422
423 /*
424 * Check to see if this is a spare device. We do an explicit check for
425 * spa_has_spare() here because it may be on our pending list of spares
426 * to add. We also check if it is an l2cache device.
427 */
428 if (spa_spare_exists(device_guid, &spare_pool) ||
429 spa_has_spare(spa, device_guid)) {
430 if (spare_guid)
431 *spare_guid = device_guid;
432
433 switch (reason) {
434 case VDEV_LABEL_CREATE:
435 case VDEV_LABEL_L2CACHE:
436 return (B_TRUE);
437
438 case VDEV_LABEL_REPLACE:
439 return (!spa_has_spare(spa, device_guid) ||
440 spare_pool != 0ULL);
441
442 case VDEV_LABEL_SPARE:
443 return (spa_has_spare(spa, device_guid));
444 }
445 }
446
447 /*
448 * Check to see if this is an l2cache device.
449 */
450 if (spa_l2cache_exists(device_guid, NULL))
451 return (B_TRUE);
452
453 /*
454 * If the device is marked ACTIVE, then this device is in use by another
455 * pool on the system.
456 */
457 return (state == POOL_STATE_ACTIVE);
458}
459
460/*
461 * Initialize a vdev label. We check to make sure each leaf device is not in
462 * use, and writable. We put down an initial label which we will later
463 * overwrite with a complete label. Note that it's important to do this
464 * sequentially, not in parallel, so that we catch cases of multiple use of the
465 * same leaf vdev in the vdev we're creating -- e.g. mirroring a disk with
466 * itself.
467 */
468int
469vdev_label_init(vdev_t *vd, uint64_t crtxg, vdev_labeltype_t reason)
470{
471 spa_t *spa = vd->vdev_spa;
472 nvlist_t *label;
473 vdev_phys_t *vp;
474 vdev_boot_header_t *vb;
475 uberblock_t *ub;
476 zio_t *zio;
477 int l, c, n;
478 char *buf;
479 size_t buflen;
480 int error;
481 uint64_t spare_guid, l2cache_guid;
482 int flags = ZIO_FLAG_CONFIG_HELD | ZIO_FLAG_CANFAIL;
483
484 ASSERT(spa_config_held(spa, RW_WRITER));
485
486 for (c = 0; c < vd->vdev_children; c++)
487 if ((error = vdev_label_init(vd->vdev_child[c],
488 crtxg, reason)) != 0)
489 return (error);
490
491 if (!vd->vdev_ops->vdev_op_leaf)
492 return (0);
493
494 /*
495 * Dead vdevs cannot be initialized.
496 */
497 if (vdev_is_dead(vd))
498 return (EIO);
499
500 /*
501 * Determine if the vdev is in use.
502 */
503 if (reason != VDEV_LABEL_REMOVE &&
504 vdev_inuse(vd, crtxg, reason, &spare_guid, &l2cache_guid))
505 return (EBUSY);
506
507 ASSERT(reason != VDEV_LABEL_REMOVE ||
508 vdev_inuse(vd, crtxg, reason, NULL, NULL));
509
510 /*
511 * If this is a request to add or replace a spare or l2cache device
512 * that is in use elsewhere on the system, then we must update the
513 * guid (which was initialized to a random value) to reflect the
514 * actual GUID (which is shared between multiple pools).
515 */
516 if (reason != VDEV_LABEL_REMOVE && reason != VDEV_LABEL_L2CACHE &&
517 spare_guid != 0ULL) {
518 vdev_t *pvd = vd->vdev_parent;
519
520 for (; pvd != NULL; pvd = pvd->vdev_parent) {
521 pvd->vdev_guid_sum -= vd->vdev_guid;
522 pvd->vdev_guid_sum += spare_guid;
523 }
524
525 vd->vdev_guid = vd->vdev_guid_sum = spare_guid;
526
527 /*
528 * If this is a replacement, then we want to fallthrough to the
529 * rest of the code. If we're adding a spare, then it's already
530 * labeled appropriately and we can just return.
531 */
532 if (reason == VDEV_LABEL_SPARE)
533 return (0);
534 ASSERT(reason == VDEV_LABEL_REPLACE);
535 }
536
537 if (reason != VDEV_LABEL_REMOVE && reason != VDEV_LABEL_SPARE &&
538 l2cache_guid != 0ULL) {
539 vdev_t *pvd = vd->vdev_parent;
540
541 for (; pvd != NULL; pvd = pvd->vdev_parent) {
542 pvd->vdev_guid_sum -= vd->vdev_guid;
543 pvd->vdev_guid_sum += l2cache_guid;
544 }
545
546 vd->vdev_guid = vd->vdev_guid_sum = l2cache_guid;
547
548 /*
549 * If this is a replacement, then we want to fallthrough to the
550 * rest of the code. If we're adding an l2cache, then it's
551 * already labeled appropriately and we can just return.
552 */
553 if (reason == VDEV_LABEL_L2CACHE)
554 return (0);
555 ASSERT(reason == VDEV_LABEL_REPLACE);
556 }
557
558 /*
559 * Initialize its label.
560 */
561 vp = zio_buf_alloc(sizeof (vdev_phys_t));
562 bzero(vp, sizeof (vdev_phys_t));
563
564 /*
565 * Generate a label describing the pool and our top-level vdev.
566 * We mark it as being from txg 0 to indicate that it's not
567 * really part of an active pool just yet. The labels will
568 * be written again with a meaningful txg by spa_sync().
569 */
570 if (reason == VDEV_LABEL_SPARE ||
571 (reason == VDEV_LABEL_REMOVE && vd->vdev_isspare)) {
572 /*
573 * For inactive hot spares, we generate a special label that
574 * identifies as a mutually shared hot spare. We write the
575 * label if we are adding a hot spare, or if we are removing an
576 * active hot spare (in which case we want to revert the
577 * labels).
578 */
579 VERIFY(nvlist_alloc(&label, NV_UNIQUE_NAME, KM_SLEEP) == 0);
580
581 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_VERSION,
582 spa_version(spa)) == 0);
583 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_POOL_STATE,
584 POOL_STATE_SPARE) == 0);
585 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_GUID,
586 vd->vdev_guid) == 0);
587 } else if (reason == VDEV_LABEL_L2CACHE ||
588 (reason == VDEV_LABEL_REMOVE && vd->vdev_isl2cache)) {
589 /*
590 * For level 2 ARC devices, add a special label.
591 */
592 VERIFY(nvlist_alloc(&label, NV_UNIQUE_NAME, KM_SLEEP) == 0);
593
594 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_VERSION,
595 spa_version(spa)) == 0);
596 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_POOL_STATE,
597 POOL_STATE_L2CACHE) == 0);
598 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_GUID,
599 vd->vdev_guid) == 0);
600 } else {
601 label = spa_config_generate(spa, vd, 0ULL, B_FALSE);
602
603 /*
604 * Add our creation time. This allows us to detect multiple
605 * vdev uses as described above, and automatically expires if we
606 * fail.
607 */
608 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_CREATE_TXG,
609 crtxg) == 0);
610 }
611
612 buf = vp->vp_nvlist;
613 buflen = sizeof (vp->vp_nvlist);
614
615 error = nvlist_pack(label, &buf, &buflen, NV_ENCODE_XDR, KM_SLEEP);
616 if (error != 0) {
617 nvlist_free(label);
618 zio_buf_free(vp, sizeof (vdev_phys_t));
619 /* EFAULT means nvlist_pack ran out of room */
620 return (error == EFAULT ? ENAMETOOLONG : EINVAL);
621 }
622
623 /*
624 * Initialize boot block header.
625 */
626 vb = zio_buf_alloc(sizeof (vdev_boot_header_t));
627 bzero(vb, sizeof (vdev_boot_header_t));
628 vb->vb_magic = VDEV_BOOT_MAGIC;
629 vb->vb_version = VDEV_BOOT_VERSION;
630 vb->vb_offset = VDEV_BOOT_OFFSET;
631 vb->vb_size = VDEV_BOOT_SIZE;
632
633 /*
634 * Initialize uberblock template.
635 */
636 ub = zio_buf_alloc(VDEV_UBERBLOCK_SIZE(vd));
637 bzero(ub, VDEV_UBERBLOCK_SIZE(vd));
638 *ub = spa->spa_uberblock;
639 ub->ub_txg = 0;
640
641 /*
642 * Write everything in parallel.
643 */
644 zio = zio_root(spa, NULL, NULL, flags);
645
646 for (l = 0; l < VDEV_LABELS; l++) {
647
648 vdev_label_write(zio, vd, l, vp,
649 offsetof(vdev_label_t, vl_vdev_phys),
650 sizeof (vdev_phys_t), NULL, NULL, flags);
651
652 vdev_label_write(zio, vd, l, vb,
653 offsetof(vdev_label_t, vl_boot_header),
654 sizeof (vdev_boot_header_t), NULL, NULL, flags);
655
656 for (n = 0; n < VDEV_UBERBLOCK_COUNT(vd); n++) {
657 vdev_label_write(zio, vd, l, ub,
658 VDEV_UBERBLOCK_OFFSET(vd, n),
659 VDEV_UBERBLOCK_SIZE(vd), NULL, NULL, flags);
660 }
661 }
662
663 error = zio_wait(zio);
664
665 nvlist_free(label);
666 zio_buf_free(ub, VDEV_UBERBLOCK_SIZE(vd));
667 zio_buf_free(vb, sizeof (vdev_boot_header_t));
668 zio_buf_free(vp, sizeof (vdev_phys_t));
669
670 /*
671 * If this vdev hasn't been previously identified as a spare, then we
672 * mark it as such only if a) we are labeling it as a spare, or b) it
673 * exists as a spare elsewhere in the system. Do the same for
674 * level 2 ARC devices.
675 */
676 if (error == 0 && !vd->vdev_isspare &&
677 (reason == VDEV_LABEL_SPARE ||
678 spa_spare_exists(vd->vdev_guid, NULL)))
679 spa_spare_add(vd);
680
681 if (error == 0 && !vd->vdev_isl2cache &&
682 (reason == VDEV_LABEL_L2CACHE ||
683 spa_l2cache_exists(vd->vdev_guid, NULL)))
684 spa_l2cache_add(vd);
685
686 return (error);
687}
688
689/*
690 * ==========================================================================
691 * uberblock load/sync
692 * ==========================================================================
693 */
694
695/*
696 * Consider the following situation: txg is safely synced to disk. We've
697 * written the first uberblock for txg + 1, and then we lose power. When we
698 * come back up, we fail to see the uberblock for txg + 1 because, say,
699 * it was on a mirrored device and the replica to which we wrote txg + 1
700 * is now offline. If we then make some changes and sync txg + 1, and then
701 * the missing replica comes back, then for a new seconds we'll have two
702 * conflicting uberblocks on disk with the same txg. The solution is simple:
703 * among uberblocks with equal txg, choose the one with the latest timestamp.
704 */
705static int
706vdev_uberblock_compare(uberblock_t *ub1, uberblock_t *ub2)
707{
708 if (ub1->ub_txg < ub2->ub_txg)
709 return (-1);
710 if (ub1->ub_txg > ub2->ub_txg)
711 return (1);
712
713 if (ub1->ub_timestamp < ub2->ub_timestamp)
714 return (-1);
715 if (ub1->ub_timestamp > ub2->ub_timestamp)
716 return (1);
717
718 return (0);
719}
720
721static void
722vdev_uberblock_load_done(zio_t *zio)
723{
724 uberblock_t *ub = zio->io_data;
725 uberblock_t *ubbest = zio->io_private;
726 spa_t *spa = zio->io_spa;
727
728 ASSERT3U(zio->io_size, ==, VDEV_UBERBLOCK_SIZE(zio->io_vd));
729
730 if (zio->io_error == 0 && uberblock_verify(ub) == 0) {
731 mutex_enter(&spa->spa_uberblock_lock);
732 if (vdev_uberblock_compare(ub, ubbest) > 0)
733 *ubbest = *ub;
734 mutex_exit(&spa->spa_uberblock_lock);
735 }
736
737 zio_buf_free(zio->io_data, zio->io_size);
738}
739
740void
741vdev_uberblock_load(zio_t *zio, vdev_t *vd, uberblock_t *ubbest)
742{
743 int l, c, n;
744
745 for (c = 0; c < vd->vdev_children; c++)
746 vdev_uberblock_load(zio, vd->vdev_child[c], ubbest);
747
748 if (!vd->vdev_ops->vdev_op_leaf)
749 return;
750
751 if (vdev_is_dead(vd))
752 return;
753
754 for (l = 0; l < VDEV_LABELS; l++) {
755 for (n = 0; n < VDEV_UBERBLOCK_COUNT(vd); n++) {
756 vdev_label_read(zio, vd, l,
757 zio_buf_alloc(VDEV_UBERBLOCK_SIZE(vd)),
758 VDEV_UBERBLOCK_OFFSET(vd, n),
759 VDEV_UBERBLOCK_SIZE(vd),
760 vdev_uberblock_load_done, ubbest);
761 }
762 }
763}
764
765/*
766 * On success, increment root zio's count of good writes.
767 * We only get credit for writes to known-visible vdevs; see spa_vdev_add().
768 */
769static void
770vdev_uberblock_sync_done(zio_t *zio)
771{
772 uint64_t *good_writes = zio->io_private;
773
774 if (zio->io_error == 0 && zio->io_vd->vdev_top->vdev_ms_array != 0)
775 atomic_add_64(good_writes, 1);
776}
777
778/*
779 * Write the uberblock to all labels of all leaves of the specified vdev.
780 */
781static void
782vdev_uberblock_sync(zio_t *zio, uberblock_t *ub, vdev_t *vd)
783{
784 int l, c, n;
785 uberblock_t *ubbuf;
786
787 for (c = 0; c < vd->vdev_children; c++)
788 vdev_uberblock_sync(zio, ub, vd->vdev_child[c]);
789
790 if (!vd->vdev_ops->vdev_op_leaf)
791 return;
792
793 if (vdev_is_dead(vd))
794 return;
795
796 n = ub->ub_txg & (VDEV_UBERBLOCK_COUNT(vd) - 1);
797
798 ubbuf = zio_buf_alloc(VDEV_UBERBLOCK_SIZE(vd));
799 bzero(ubbuf, VDEV_UBERBLOCK_SIZE(vd));
800 *ubbuf = *ub;
801
802 for (l = 0; l < VDEV_LABELS; l++)
803 vdev_label_write(zio, vd, l, ubbuf,
804 VDEV_UBERBLOCK_OFFSET(vd, n),
805 VDEV_UBERBLOCK_SIZE(vd),
806 vdev_uberblock_sync_done, zio->io_private,
807 ZIO_FLAG_CANFAIL | ZIO_FLAG_DONT_PROPAGATE);
808
809 zio_buf_free(ubbuf, VDEV_UBERBLOCK_SIZE(vd));
810}
811
812int
813vdev_uberblock_sync_list(vdev_t **svd, int svdcount, uberblock_t *ub, int flags)
814{
815 spa_t *spa = svd[0]->vdev_spa;
816 int v;
817 zio_t *zio;
818 uint64_t good_writes = 0;
819
820 zio = zio_root(spa, NULL, &good_writes, flags);
821
822 for (v = 0; v < svdcount; v++)
823 vdev_uberblock_sync(zio, ub, svd[v]);
824
825 (void) zio_wait(zio);
826
827 /*
828 * Flush the uberblocks to disk. This ensures that the odd labels
829 * are no longer needed (because the new uberblocks and the even
830 * labels are safely on disk), so it is safe to overwrite them.
831 */
832 zio = zio_root(spa, NULL, NULL, flags);
833
834 for (v = 0; v < svdcount; v++)
835 zio_flush(zio, svd[v]);
836
837 (void) zio_wait(zio);
838
839 return (good_writes >= 1 ? 0 : EIO);
840}
841
842/*
843 * On success, increment the count of good writes for our top-level vdev.
844 */
845static void
846vdev_label_sync_done(zio_t *zio)
847{
848 uint64_t *good_writes = zio->io_private;
849
850 if (zio->io_error == 0)
851 atomic_add_64(good_writes, 1);
852}
853
854/*
855 * If there weren't enough good writes, indicate failure to the parent.
856 */
857static void
858vdev_label_sync_top_done(zio_t *zio)
859{
860 uint64_t *good_writes = zio->io_private;
861
862 if (*good_writes == 0)
863 zio->io_error = EIO;
864
865 kmem_free(good_writes, sizeof (uint64_t));
866}
867
868/*
869 * Write all even or odd labels to all leaves of the specified vdev.
870 */
871static void
872vdev_label_sync(zio_t *zio, vdev_t *vd, int l, uint64_t txg)
873{
874 nvlist_t *label;
875 vdev_phys_t *vp;
876 char *buf;
877 size_t buflen;
878 int c;
879
880 for (c = 0; c < vd->vdev_children; c++)
881 vdev_label_sync(zio, vd->vdev_child[c], l, txg);
882
883 if (!vd->vdev_ops->vdev_op_leaf)
884 return;
885
886 if (vdev_is_dead(vd))
887 return;
888
889 /*
890 * Generate a label describing the top-level config to which we belong.
891 */
892 label = spa_config_generate(vd->vdev_spa, vd, txg, B_FALSE);
893
894 vp = zio_buf_alloc(sizeof (vdev_phys_t));
895 bzero(vp, sizeof (vdev_phys_t));
896
897 buf = vp->vp_nvlist;
898 buflen = sizeof (vp->vp_nvlist);
899
900 if (nvlist_pack(label, &buf, &buflen, NV_ENCODE_XDR, KM_SLEEP) == 0) {
901 for (; l < VDEV_LABELS; l += 2) {
902 vdev_label_write(zio, vd, l, vp,
903 offsetof(vdev_label_t, vl_vdev_phys),
904 sizeof (vdev_phys_t),
905 vdev_label_sync_done, zio->io_private,
906 ZIO_FLAG_CANFAIL | ZIO_FLAG_DONT_PROPAGATE);
907 }
908 }
909
910 zio_buf_free(vp, sizeof (vdev_phys_t));
911 nvlist_free(label);
912}
913
914int
915vdev_label_sync_list(spa_t *spa, int l, int flags, uint64_t txg)
916{
917 list_t *dl = &spa->spa_dirty_list;
918 vdev_t *vd;
919 zio_t *zio;
920 int error;
921
922 /*
923 * Write the new labels to disk.
924 */
925 zio = zio_root(spa, NULL, NULL, flags);
926
927 for (vd = list_head(dl); vd != NULL; vd = list_next(dl, vd)) {
928 uint64_t *good_writes = kmem_zalloc(sizeof (uint64_t),
929 KM_SLEEP);
930 zio_t *vio = zio_null(zio, spa, vdev_label_sync_top_done,
931 good_writes, flags);
932 vdev_label_sync(vio, vd, l, txg);
933 zio_nowait(vio);
934 }
935
936 error = zio_wait(zio);
937
938 /*
939 * Flush the new labels to disk.
940 */
941 zio = zio_root(spa, NULL, NULL, flags);
942
943 for (vd = list_head(dl); vd != NULL; vd = list_next(dl, vd))
944 zio_flush(zio, vd);
945
946 (void) zio_wait(zio);
947
948 return (error);
949}
950
951/*
952 * Sync the uberblock and any changes to the vdev configuration.
953 *
954 * The order of operations is carefully crafted to ensure that
955 * if the system panics or loses power at any time, the state on disk
956 * is still transactionally consistent. The in-line comments below
957 * describe the failure semantics at each stage.
958 *
959 * Moreover, vdev_config_sync() is designed to be idempotent: if it fails
960 * at any time, you can just call it again, and it will resume its work.
961 */
962int
963vdev_config_sync(vdev_t **svd, int svdcount, uint64_t txg)
964{
965 spa_t *spa = svd[0]->vdev_spa;
966 uberblock_t *ub = &spa->spa_uberblock;
967 vdev_t *vd;
968 zio_t *zio;
969 int error;
970 int flags = ZIO_FLAG_CONFIG_HELD | ZIO_FLAG_CANFAIL;
971
972 ASSERT(ub->ub_txg <= txg);
973
974 /*
975 * If this isn't a resync due to I/O errors,
976 * and nothing changed in this transaction group,
977 * and the vdev configuration hasn't changed,
978 * then there's nothing to do.
979 */
980 if (ub->ub_txg < txg &&
981 uberblock_update(ub, spa->spa_root_vdev, txg) == B_FALSE &&
982 list_is_empty(&spa->spa_dirty_list))
983 return (0);
984
985 if (txg > spa_freeze_txg(spa))
986 return (0);
987
988 ASSERT(txg <= spa->spa_final_txg);
989
990 /*
991 * Flush the write cache of every disk that's been written to
992 * in this transaction group. This ensures that all blocks
993 * written in this txg will be committed to stable storage
994 * before any uberblock that references them.
995 */
996 zio = zio_root(spa, NULL, NULL, flags);
997
998 for (vd = txg_list_head(&spa->spa_vdev_txg_list, TXG_CLEAN(txg)); vd;
999 vd = txg_list_next(&spa->spa_vdev_txg_list, vd, TXG_CLEAN(txg)))
1000 zio_flush(zio, vd);
1001
1002 (void) zio_wait(zio);
1003
1004 /*
1005 * Sync out the even labels (L0, L2) for every dirty vdev. If the
1006 * system dies in the middle of this process, that's OK: all of the
1007 * even labels that made it to disk will be newer than any uberblock,
1008 * and will therefore be considered invalid. The odd labels (L1, L3),
1009 * which have not yet been touched, will still be valid. We flush
1010 * the new labels to disk to ensure that all even-label updates
1011 * are committed to stable storage before the uberblock update.
1012 */
1013 if ((error = vdev_label_sync_list(spa, 0, flags, txg)) != 0)
1014 return (error);
1015
1016 /*
1017 * Sync the uberblocks to all vdevs in svd[].
1018 * If the system dies in the middle of this step, there are two cases
1019 * to consider, and the on-disk state is consistent either way:
1020 *
1021 * (1) If none of the new uberblocks made it to disk, then the
1022 * previous uberblock will be the newest, and the odd labels
1023 * (which had not yet been touched) will be valid with respect
1024 * to that uberblock.
1025 *
1026 * (2) If one or more new uberblocks made it to disk, then they
1027 * will be the newest, and the even labels (which had all
1028 * been successfully committed) will be valid with respect
1029 * to the new uberblocks.
1030 */
1031 if ((error = vdev_uberblock_sync_list(svd, svdcount, ub, flags)) != 0)
1032 return (error);
1033
1034 /*
1035 * Sync out odd labels for every dirty vdev. If the system dies
1036 * in the middle of this process, the even labels and the new
1037 * uberblocks will suffice to open the pool. The next time
1038 * the pool is opened, the first thing we'll do -- before any
1039 * user data is modified -- is mark every vdev dirty so that
1040 * all labels will be brought up to date. We flush the new labels
1041 * to disk to ensure that all odd-label updates are committed to
1042 * stable storage before the next transaction group begins.
1043 */
1044 return (vdev_label_sync_list(spa, 1, flags, txg));
1045}