]> git.proxmox.com Git - mirror_zfs-debian.git/blame - module/zfs/vdev_label.c
Rebase master to b108
[mirror_zfs-debian.git] / module / zfs / 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/*
d164b209 22 * Copyright 2009 Sun Microsystems, Inc. All rights reserved.
34dc7c2f
BB
23 * Use is subject to license terms.
24 */
25
34dc7c2f
BB
26/*
27 * Virtual Device Labels
28 * ---------------------
29 *
30 * The vdev label serves several distinct purposes:
31 *
32 * 1. Uniquely identify this device as part of a ZFS pool and confirm its
33 * identity within the pool.
34 *
35 * 2. Verify that all the devices given in a configuration are present
36 * within the pool.
37 *
38 * 3. Determine the uberblock for the pool.
39 *
40 * 4. In case of an import operation, determine the configuration of the
41 * toplevel vdev of which it is a part.
42 *
43 * 5. If an import operation cannot find all the devices in the pool,
44 * provide enough information to the administrator to determine which
45 * devices are missing.
46 *
47 * It is important to note that while the kernel is responsible for writing the
48 * label, it only consumes the information in the first three cases. The
49 * latter information is only consumed in userland when determining the
50 * configuration to import a pool.
51 *
52 *
53 * Label Organization
54 * ------------------
55 *
56 * Before describing the contents of the label, it's important to understand how
57 * the labels are written and updated with respect to the uberblock.
58 *
59 * When the pool configuration is altered, either because it was newly created
60 * or a device was added, we want to update all the labels such that we can deal
61 * with fatal failure at any point. To this end, each disk has two labels which
62 * are updated before and after the uberblock is synced. Assuming we have
63 * labels and an uberblock with the following transaction groups:
64 *
65 * L1 UB L2
66 * +------+ +------+ +------+
67 * | | | | | |
68 * | t10 | | t10 | | t10 |
69 * | | | | | |
70 * +------+ +------+ +------+
71 *
72 * In this stable state, the labels and the uberblock were all updated within
73 * the same transaction group (10). Each label is mirrored and checksummed, so
74 * that we can detect when we fail partway through writing the label.
75 *
76 * In order to identify which labels are valid, the labels are written in the
77 * following manner:
78 *
79 * 1. For each vdev, update 'L1' to the new label
80 * 2. Update the uberblock
81 * 3. For each vdev, update 'L2' to the new label
82 *
83 * Given arbitrary failure, we can determine the correct label to use based on
84 * the transaction group. If we fail after updating L1 but before updating the
85 * UB, we will notice that L1's transaction group is greater than the uberblock,
86 * so L2 must be valid. If we fail after writing the uberblock but before
87 * writing L2, we will notice that L2's transaction group is less than L1, and
88 * therefore L1 is valid.
89 *
90 * Another added complexity is that not every label is updated when the config
91 * is synced. If we add a single device, we do not want to have to re-write
92 * every label for every device in the pool. This means that both L1 and L2 may
93 * be older than the pool uberblock, because the necessary information is stored
94 * on another vdev.
95 *
96 *
97 * On-disk Format
98 * --------------
99 *
100 * The vdev label consists of two distinct parts, and is wrapped within the
101 * vdev_label_t structure. The label includes 8k of padding to permit legacy
102 * VTOC disk labels, but is otherwise ignored.
103 *
104 * The first half of the label is a packed nvlist which contains pool wide
105 * properties, per-vdev properties, and configuration information. It is
106 * described in more detail below.
107 *
108 * The latter half of the label consists of a redundant array of uberblocks.
109 * These uberblocks are updated whenever a transaction group is committed,
110 * or when the configuration is updated. When a pool is loaded, we scan each
111 * vdev for the 'best' uberblock.
112 *
113 *
114 * Configuration Information
115 * -------------------------
116 *
117 * The nvlist describing the pool and vdev contains the following elements:
118 *
119 * version ZFS on-disk version
120 * name Pool name
121 * state Pool state
122 * txg Transaction group in which this label was written
123 * pool_guid Unique identifier for this pool
124 * vdev_tree An nvlist describing vdev tree.
125 *
126 * Each leaf device label also contains the following:
127 *
128 * top_guid Unique ID for top-level vdev in which this is contained
129 * guid Unique ID for the leaf vdev
130 *
131 * The 'vs' configuration follows the format described in 'spa_config.c'.
132 */
133
134#include <sys/zfs_context.h>
135#include <sys/spa.h>
136#include <sys/spa_impl.h>
137#include <sys/dmu.h>
138#include <sys/zap.h>
139#include <sys/vdev.h>
140#include <sys/vdev_impl.h>
141#include <sys/uberblock_impl.h>
142#include <sys/metaslab.h>
143#include <sys/zio.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 */
150uint64_t
151vdev_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
b128c09f
BB
160/*
161 * Returns back the vdev label associated with the passed in offset.
162 */
163int
164vdev_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
34dc7c2f
BB
176static void
177vdev_label_read(zio_t *zio, vdev_t *vd, int l, void *buf, uint64_t offset,
b128c09f 178 uint64_t size, zio_done_func_t *done, void *private, int flags)
34dc7c2f 179{
b128c09f
BB
180 ASSERT(spa_config_held(zio->io_spa, SCL_STATE_ALL, RW_WRITER) ==
181 SCL_STATE_ALL);
182 ASSERT(flags & ZIO_FLAG_CONFIG_WRITER);
34dc7c2f
BB
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,
b128c09f 187 ZIO_PRIORITY_SYNC_READ, flags, B_TRUE));
34dc7c2f
BB
188}
189
190static void
191vdev_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{
b128c09f
BB
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);
34dc7c2f
BB
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 */
209nvlist_t *
210vdev_config_generate(spa_t *spa, vdev_t *vd, boolean_t getstats,
211 boolean_t isspare, boolean_t isl2cache)
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 (!isspare && !isl2cache)
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_nparity != 0) {
237 ASSERT(strcmp(vd->vdev_ops->vdev_op_type,
238 VDEV_TYPE_RAIDZ) == 0);
239
240 /*
241 * Make sure someone hasn't managed to sneak a fancy new vdev
242 * into a crufty old storage pool.
243 */
244 ASSERT(vd->vdev_nparity == 1 ||
245 (vd->vdev_nparity == 2 &&
246 spa_version(spa) >= SPA_VERSION_RAID6));
247
248 /*
249 * Note that we'll add the nparity tag even on storage pools
250 * that only support a single parity device -- older software
251 * will just ignore it.
252 */
253 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_NPARITY,
254 vd->vdev_nparity) == 0);
255 }
256
257 if (vd->vdev_wholedisk != -1ULL)
258 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK,
259 vd->vdev_wholedisk) == 0);
260
261 if (vd->vdev_not_present)
262 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_NOT_PRESENT, 1) == 0);
263
264 if (vd->vdev_isspare)
265 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_IS_SPARE, 1) == 0);
266
267 if (!isspare && !isl2cache && vd == vd->vdev_top) {
268 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_METASLAB_ARRAY,
269 vd->vdev_ms_array) == 0);
270 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_METASLAB_SHIFT,
271 vd->vdev_ms_shift) == 0);
272 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_ASHIFT,
273 vd->vdev_ashift) == 0);
274 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_ASIZE,
275 vd->vdev_asize) == 0);
276 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_IS_LOG,
277 vd->vdev_islog) == 0);
278 }
279
fb5f0bc8 280 if (vd->vdev_dtl_smo.smo_object != 0)
34dc7c2f 281 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_DTL,
fb5f0bc8 282 vd->vdev_dtl_smo.smo_object) == 0);
34dc7c2f
BB
283
284 if (getstats) {
285 vdev_stat_t vs;
286 vdev_get_stats(vd, &vs);
287 VERIFY(nvlist_add_uint64_array(nv, ZPOOL_CONFIG_STATS,
288 (uint64_t *)&vs, sizeof (vs) / sizeof (uint64_t)) == 0);
289 }
290
291 if (!vd->vdev_ops->vdev_op_leaf) {
292 nvlist_t **child;
293 int c;
294
295 child = kmem_alloc(vd->vdev_children * sizeof (nvlist_t *),
296 KM_SLEEP);
297
298 for (c = 0; c < vd->vdev_children; c++)
299 child[c] = vdev_config_generate(spa, vd->vdev_child[c],
300 getstats, isspare, isl2cache);
301
302 VERIFY(nvlist_add_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
303 child, vd->vdev_children) == 0);
304
305 for (c = 0; c < vd->vdev_children; c++)
306 nvlist_free(child[c]);
307
308 kmem_free(child, vd->vdev_children * sizeof (nvlist_t *));
309
310 } else {
311 if (vd->vdev_offline && !vd->vdev_tmpoffline)
312 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_OFFLINE,
313 B_TRUE) == 0);
314 if (vd->vdev_faulted)
315 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_FAULTED,
316 B_TRUE) == 0);
317 if (vd->vdev_degraded)
318 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_DEGRADED,
319 B_TRUE) == 0);
320 if (vd->vdev_removed)
321 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_REMOVED,
322 B_TRUE) == 0);
323 if (vd->vdev_unspare)
324 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_UNSPARE,
325 B_TRUE) == 0);
326 }
327
328 return (nv);
329}
330
331nvlist_t *
332vdev_label_read_config(vdev_t *vd)
333{
334 spa_t *spa = vd->vdev_spa;
335 nvlist_t *config = NULL;
336 vdev_phys_t *vp;
337 zio_t *zio;
b128c09f
BB
338 int flags =
339 ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE;
34dc7c2f 340
b128c09f 341 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
34dc7c2f
BB
342
343 if (!vdev_readable(vd))
344 return (NULL);
345
346 vp = zio_buf_alloc(sizeof (vdev_phys_t));
347
b128c09f 348 for (int l = 0; l < VDEV_LABELS; l++) {
34dc7c2f 349
b128c09f 350 zio = zio_root(spa, NULL, NULL, flags);
34dc7c2f
BB
351
352 vdev_label_read(zio, vd, l, vp,
353 offsetof(vdev_label_t, vl_vdev_phys),
b128c09f 354 sizeof (vdev_phys_t), NULL, NULL, flags);
34dc7c2f
BB
355
356 if (zio_wait(zio) == 0 &&
357 nvlist_unpack(vp->vp_nvlist, sizeof (vp->vp_nvlist),
358 &config, 0) == 0)
359 break;
360
361 if (config != NULL) {
362 nvlist_free(config);
363 config = NULL;
364 }
365 }
366
367 zio_buf_free(vp, sizeof (vdev_phys_t));
368
369 return (config);
370}
371
372/*
373 * Determine if a device is in use. The 'spare_guid' parameter will be filled
374 * in with the device guid if this spare is active elsewhere on the system.
375 */
376static boolean_t
377vdev_inuse(vdev_t *vd, uint64_t crtxg, vdev_labeltype_t reason,
378 uint64_t *spare_guid, uint64_t *l2cache_guid)
379{
380 spa_t *spa = vd->vdev_spa;
381 uint64_t state, pool_guid, device_guid, txg, spare_pool;
382 uint64_t vdtxg = 0;
383 nvlist_t *label;
384
385 if (spare_guid)
386 *spare_guid = 0ULL;
387 if (l2cache_guid)
388 *l2cache_guid = 0ULL;
389
390 /*
391 * Read the label, if any, and perform some basic sanity checks.
392 */
393 if ((label = vdev_label_read_config(vd)) == NULL)
394 return (B_FALSE);
395
396 (void) nvlist_lookup_uint64(label, ZPOOL_CONFIG_CREATE_TXG,
397 &vdtxg);
398
399 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE,
400 &state) != 0 ||
401 nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID,
402 &device_guid) != 0) {
403 nvlist_free(label);
404 return (B_FALSE);
405 }
406
407 if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
408 (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_GUID,
409 &pool_guid) != 0 ||
410 nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG,
411 &txg) != 0)) {
412 nvlist_free(label);
413 return (B_FALSE);
414 }
415
416 nvlist_free(label);
417
418 /*
419 * Check to see if this device indeed belongs to the pool it claims to
420 * be a part of. The only way this is allowed is if the device is a hot
421 * spare (which we check for later on).
422 */
423 if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
424 !spa_guid_exists(pool_guid, device_guid) &&
b128c09f 425 !spa_spare_exists(device_guid, NULL, NULL) &&
34dc7c2f
BB
426 !spa_l2cache_exists(device_guid, NULL))
427 return (B_FALSE);
428
429 /*
430 * If the transaction group is zero, then this an initialized (but
431 * unused) label. This is only an error if the create transaction
432 * on-disk is the same as the one we're using now, in which case the
433 * user has attempted to add the same vdev multiple times in the same
434 * transaction.
435 */
436 if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
437 txg == 0 && vdtxg == crtxg)
438 return (B_TRUE);
439
440 /*
441 * Check to see if this is a spare device. We do an explicit check for
442 * spa_has_spare() here because it may be on our pending list of spares
443 * to add. We also check if it is an l2cache device.
444 */
b128c09f 445 if (spa_spare_exists(device_guid, &spare_pool, NULL) ||
34dc7c2f
BB
446 spa_has_spare(spa, device_guid)) {
447 if (spare_guid)
448 *spare_guid = device_guid;
449
450 switch (reason) {
451 case VDEV_LABEL_CREATE:
452 case VDEV_LABEL_L2CACHE:
453 return (B_TRUE);
454
455 case VDEV_LABEL_REPLACE:
456 return (!spa_has_spare(spa, device_guid) ||
457 spare_pool != 0ULL);
458
459 case VDEV_LABEL_SPARE:
460 return (spa_has_spare(spa, device_guid));
461 }
462 }
463
464 /*
465 * Check to see if this is an l2cache device.
466 */
467 if (spa_l2cache_exists(device_guid, NULL))
468 return (B_TRUE);
469
470 /*
471 * If the device is marked ACTIVE, then this device is in use by another
472 * pool on the system.
473 */
474 return (state == POOL_STATE_ACTIVE);
475}
476
477/*
478 * Initialize a vdev label. We check to make sure each leaf device is not in
479 * use, and writable. We put down an initial label which we will later
480 * overwrite with a complete label. Note that it's important to do this
481 * sequentially, not in parallel, so that we catch cases of multiple use of the
482 * same leaf vdev in the vdev we're creating -- e.g. mirroring a disk with
483 * itself.
484 */
485int
486vdev_label_init(vdev_t *vd, uint64_t crtxg, vdev_labeltype_t reason)
487{
488 spa_t *spa = vd->vdev_spa;
489 nvlist_t *label;
490 vdev_phys_t *vp;
491 vdev_boot_header_t *vb;
492 uberblock_t *ub;
493 zio_t *zio;
34dc7c2f
BB
494 char *buf;
495 size_t buflen;
496 int error;
497 uint64_t spare_guid, l2cache_guid;
b128c09f 498 int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL;
34dc7c2f 499
b128c09f 500 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
34dc7c2f 501
b128c09f 502 for (int c = 0; c < vd->vdev_children; c++)
34dc7c2f
BB
503 if ((error = vdev_label_init(vd->vdev_child[c],
504 crtxg, reason)) != 0)
505 return (error);
506
507 if (!vd->vdev_ops->vdev_op_leaf)
508 return (0);
509
510 /*
511 * Dead vdevs cannot be initialized.
512 */
513 if (vdev_is_dead(vd))
514 return (EIO);
515
516 /*
517 * Determine if the vdev is in use.
518 */
519 if (reason != VDEV_LABEL_REMOVE &&
520 vdev_inuse(vd, crtxg, reason, &spare_guid, &l2cache_guid))
521 return (EBUSY);
522
34dc7c2f
BB
523 /*
524 * If this is a request to add or replace a spare or l2cache device
525 * that is in use elsewhere on the system, then we must update the
526 * guid (which was initialized to a random value) to reflect the
527 * actual GUID (which is shared between multiple pools).
528 */
529 if (reason != VDEV_LABEL_REMOVE && reason != VDEV_LABEL_L2CACHE &&
530 spare_guid != 0ULL) {
b128c09f 531 uint64_t guid_delta = spare_guid - vd->vdev_guid;
34dc7c2f 532
b128c09f 533 vd->vdev_guid += guid_delta;
34dc7c2f 534
b128c09f
BB
535 for (vdev_t *pvd = vd; pvd != NULL; pvd = pvd->vdev_parent)
536 pvd->vdev_guid_sum += guid_delta;
34dc7c2f
BB
537
538 /*
539 * If this is a replacement, then we want to fallthrough to the
540 * rest of the code. If we're adding a spare, then it's already
541 * labeled appropriately and we can just return.
542 */
543 if (reason == VDEV_LABEL_SPARE)
544 return (0);
545 ASSERT(reason == VDEV_LABEL_REPLACE);
546 }
547
548 if (reason != VDEV_LABEL_REMOVE && reason != VDEV_LABEL_SPARE &&
549 l2cache_guid != 0ULL) {
b128c09f 550 uint64_t guid_delta = l2cache_guid - vd->vdev_guid;
34dc7c2f 551
b128c09f 552 vd->vdev_guid += guid_delta;
34dc7c2f 553
b128c09f
BB
554 for (vdev_t *pvd = vd; pvd != NULL; pvd = pvd->vdev_parent)
555 pvd->vdev_guid_sum += guid_delta;
34dc7c2f
BB
556
557 /*
558 * If this is a replacement, then we want to fallthrough to the
559 * rest of the code. If we're adding an l2cache, then it's
560 * already labeled appropriately and we can just return.
561 */
562 if (reason == VDEV_LABEL_L2CACHE)
563 return (0);
564 ASSERT(reason == VDEV_LABEL_REPLACE);
565 }
566
567 /*
568 * Initialize its label.
569 */
570 vp = zio_buf_alloc(sizeof (vdev_phys_t));
571 bzero(vp, sizeof (vdev_phys_t));
572
573 /*
574 * Generate a label describing the pool and our top-level vdev.
575 * We mark it as being from txg 0 to indicate that it's not
576 * really part of an active pool just yet. The labels will
577 * be written again with a meaningful txg by spa_sync().
578 */
579 if (reason == VDEV_LABEL_SPARE ||
580 (reason == VDEV_LABEL_REMOVE && vd->vdev_isspare)) {
581 /*
582 * For inactive hot spares, we generate a special label that
583 * identifies as a mutually shared hot spare. We write the
584 * label if we are adding a hot spare, or if we are removing an
585 * active hot spare (in which case we want to revert the
586 * labels).
587 */
588 VERIFY(nvlist_alloc(&label, NV_UNIQUE_NAME, KM_SLEEP) == 0);
589
590 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_VERSION,
591 spa_version(spa)) == 0);
592 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_POOL_STATE,
593 POOL_STATE_SPARE) == 0);
594 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_GUID,
595 vd->vdev_guid) == 0);
596 } else if (reason == VDEV_LABEL_L2CACHE ||
597 (reason == VDEV_LABEL_REMOVE && vd->vdev_isl2cache)) {
598 /*
599 * For level 2 ARC devices, add a special label.
600 */
601 VERIFY(nvlist_alloc(&label, NV_UNIQUE_NAME, KM_SLEEP) == 0);
602
603 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_VERSION,
604 spa_version(spa)) == 0);
605 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_POOL_STATE,
606 POOL_STATE_L2CACHE) == 0);
607 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_GUID,
608 vd->vdev_guid) == 0);
609 } else {
610 label = spa_config_generate(spa, vd, 0ULL, B_FALSE);
611
612 /*
613 * Add our creation time. This allows us to detect multiple
614 * vdev uses as described above, and automatically expires if we
615 * fail.
616 */
617 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_CREATE_TXG,
618 crtxg) == 0);
619 }
620
621 buf = vp->vp_nvlist;
622 buflen = sizeof (vp->vp_nvlist);
623
624 error = nvlist_pack(label, &buf, &buflen, NV_ENCODE_XDR, KM_SLEEP);
625 if (error != 0) {
626 nvlist_free(label);
627 zio_buf_free(vp, sizeof (vdev_phys_t));
628 /* EFAULT means nvlist_pack ran out of room */
629 return (error == EFAULT ? ENAMETOOLONG : EINVAL);
630 }
631
632 /*
633 * Initialize boot block header.
634 */
635 vb = zio_buf_alloc(sizeof (vdev_boot_header_t));
636 bzero(vb, sizeof (vdev_boot_header_t));
637 vb->vb_magic = VDEV_BOOT_MAGIC;
638 vb->vb_version = VDEV_BOOT_VERSION;
639 vb->vb_offset = VDEV_BOOT_OFFSET;
640 vb->vb_size = VDEV_BOOT_SIZE;
641
642 /*
643 * Initialize uberblock template.
644 */
645 ub = zio_buf_alloc(VDEV_UBERBLOCK_SIZE(vd));
646 bzero(ub, VDEV_UBERBLOCK_SIZE(vd));
647 *ub = spa->spa_uberblock;
648 ub->ub_txg = 0;
649
650 /*
651 * Write everything in parallel.
652 */
653 zio = zio_root(spa, NULL, NULL, flags);
654
b128c09f 655 for (int l = 0; l < VDEV_LABELS; l++) {
34dc7c2f
BB
656
657 vdev_label_write(zio, vd, l, vp,
658 offsetof(vdev_label_t, vl_vdev_phys),
659 sizeof (vdev_phys_t), NULL, NULL, flags);
660
661 vdev_label_write(zio, vd, l, vb,
662 offsetof(vdev_label_t, vl_boot_header),
663 sizeof (vdev_boot_header_t), NULL, NULL, flags);
664
b128c09f 665 for (int n = 0; n < VDEV_UBERBLOCK_COUNT(vd); n++) {
34dc7c2f
BB
666 vdev_label_write(zio, vd, l, ub,
667 VDEV_UBERBLOCK_OFFSET(vd, n),
668 VDEV_UBERBLOCK_SIZE(vd), NULL, NULL, flags);
669 }
670 }
671
672 error = zio_wait(zio);
673
674 nvlist_free(label);
675 zio_buf_free(ub, VDEV_UBERBLOCK_SIZE(vd));
676 zio_buf_free(vb, sizeof (vdev_boot_header_t));
677 zio_buf_free(vp, sizeof (vdev_phys_t));
678
679 /*
680 * If this vdev hasn't been previously identified as a spare, then we
681 * mark it as such only if a) we are labeling it as a spare, or b) it
682 * exists as a spare elsewhere in the system. Do the same for
683 * level 2 ARC devices.
684 */
685 if (error == 0 && !vd->vdev_isspare &&
686 (reason == VDEV_LABEL_SPARE ||
b128c09f 687 spa_spare_exists(vd->vdev_guid, NULL, NULL)))
34dc7c2f
BB
688 spa_spare_add(vd);
689
690 if (error == 0 && !vd->vdev_isl2cache &&
691 (reason == VDEV_LABEL_L2CACHE ||
692 spa_l2cache_exists(vd->vdev_guid, NULL)))
693 spa_l2cache_add(vd);
694
695 return (error);
696}
697
698/*
699 * ==========================================================================
700 * uberblock load/sync
701 * ==========================================================================
702 */
703
fb5f0bc8
BB
704/*
705 * For use by zdb and debugging purposes only
706 */
707uint64_t ub_max_txg = UINT64_MAX;
708
34dc7c2f
BB
709/*
710 * Consider the following situation: txg is safely synced to disk. We've
711 * written the first uberblock for txg + 1, and then we lose power. When we
712 * come back up, we fail to see the uberblock for txg + 1 because, say,
713 * it was on a mirrored device and the replica to which we wrote txg + 1
714 * is now offline. If we then make some changes and sync txg + 1, and then
715 * the missing replica comes back, then for a new seconds we'll have two
716 * conflicting uberblocks on disk with the same txg. The solution is simple:
717 * among uberblocks with equal txg, choose the one with the latest timestamp.
718 */
719static int
720vdev_uberblock_compare(uberblock_t *ub1, uberblock_t *ub2)
721{
722 if (ub1->ub_txg < ub2->ub_txg)
723 return (-1);
724 if (ub1->ub_txg > ub2->ub_txg)
725 return (1);
726
727 if (ub1->ub_timestamp < ub2->ub_timestamp)
728 return (-1);
729 if (ub1->ub_timestamp > ub2->ub_timestamp)
730 return (1);
731
732 return (0);
733}
734
735static void
736vdev_uberblock_load_done(zio_t *zio)
737{
b128c09f 738 zio_t *rio = zio->io_private;
34dc7c2f 739 uberblock_t *ub = zio->io_data;
b128c09f 740 uberblock_t *ubbest = rio->io_private;
34dc7c2f
BB
741
742 ASSERT3U(zio->io_size, ==, VDEV_UBERBLOCK_SIZE(zio->io_vd));
743
744 if (zio->io_error == 0 && uberblock_verify(ub) == 0) {
b128c09f 745 mutex_enter(&rio->io_lock);
fb5f0bc8
BB
746 if (ub->ub_txg <= ub_max_txg &&
747 vdev_uberblock_compare(ub, ubbest) > 0)
34dc7c2f 748 *ubbest = *ub;
b128c09f 749 mutex_exit(&rio->io_lock);
34dc7c2f
BB
750 }
751
752 zio_buf_free(zio->io_data, zio->io_size);
753}
754
755void
756vdev_uberblock_load(zio_t *zio, vdev_t *vd, uberblock_t *ubbest)
757{
b128c09f
BB
758 spa_t *spa = vd->vdev_spa;
759 vdev_t *rvd = spa->spa_root_vdev;
760 int flags =
761 ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE;
762
763 if (vd == rvd) {
764 ASSERT(zio == NULL);
765 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
766 zio = zio_root(spa, NULL, ubbest, flags);
767 bzero(ubbest, sizeof (uberblock_t));
768 }
34dc7c2f 769
b128c09f 770 ASSERT(zio != NULL);
34dc7c2f 771
b128c09f
BB
772 for (int c = 0; c < vd->vdev_children; c++)
773 vdev_uberblock_load(zio, vd->vdev_child[c], ubbest);
34dc7c2f 774
b128c09f
BB
775 if (vd->vdev_ops->vdev_op_leaf && vdev_readable(vd)) {
776 for (int l = 0; l < VDEV_LABELS; l++) {
777 for (int n = 0; n < VDEV_UBERBLOCK_COUNT(vd); n++) {
778 vdev_label_read(zio, vd, l,
779 zio_buf_alloc(VDEV_UBERBLOCK_SIZE(vd)),
780 VDEV_UBERBLOCK_OFFSET(vd, n),
781 VDEV_UBERBLOCK_SIZE(vd),
782 vdev_uberblock_load_done, zio, flags);
783 }
34dc7c2f
BB
784 }
785 }
b128c09f
BB
786
787 if (vd == rvd) {
788 (void) zio_wait(zio);
789 spa_config_exit(spa, SCL_ALL, FTAG);
790 }
34dc7c2f
BB
791}
792
793/*
794 * On success, increment root zio's count of good writes.
795 * We only get credit for writes to known-visible vdevs; see spa_vdev_add().
796 */
797static void
798vdev_uberblock_sync_done(zio_t *zio)
799{
800 uint64_t *good_writes = zio->io_private;
801
802 if (zio->io_error == 0 && zio->io_vd->vdev_top->vdev_ms_array != 0)
803 atomic_add_64(good_writes, 1);
804}
805
806/*
807 * Write the uberblock to all labels of all leaves of the specified vdev.
808 */
809static void
b128c09f 810vdev_uberblock_sync(zio_t *zio, uberblock_t *ub, vdev_t *vd, int flags)
34dc7c2f 811{
34dc7c2f 812 uberblock_t *ubbuf;
b128c09f 813 int n;
34dc7c2f 814
b128c09f
BB
815 for (int c = 0; c < vd->vdev_children; c++)
816 vdev_uberblock_sync(zio, ub, vd->vdev_child[c], flags);
34dc7c2f
BB
817
818 if (!vd->vdev_ops->vdev_op_leaf)
819 return;
820
b128c09f 821 if (!vdev_writeable(vd))
34dc7c2f
BB
822 return;
823
824 n = ub->ub_txg & (VDEV_UBERBLOCK_COUNT(vd) - 1);
825
826 ubbuf = zio_buf_alloc(VDEV_UBERBLOCK_SIZE(vd));
827 bzero(ubbuf, VDEV_UBERBLOCK_SIZE(vd));
828 *ubbuf = *ub;
829
b128c09f 830 for (int l = 0; l < VDEV_LABELS; l++)
34dc7c2f 831 vdev_label_write(zio, vd, l, ubbuf,
b128c09f 832 VDEV_UBERBLOCK_OFFSET(vd, n), VDEV_UBERBLOCK_SIZE(vd),
34dc7c2f 833 vdev_uberblock_sync_done, zio->io_private,
b128c09f 834 flags | ZIO_FLAG_DONT_PROPAGATE);
34dc7c2f
BB
835
836 zio_buf_free(ubbuf, VDEV_UBERBLOCK_SIZE(vd));
837}
838
839int
840vdev_uberblock_sync_list(vdev_t **svd, int svdcount, uberblock_t *ub, int flags)
841{
842 spa_t *spa = svd[0]->vdev_spa;
34dc7c2f
BB
843 zio_t *zio;
844 uint64_t good_writes = 0;
845
846 zio = zio_root(spa, NULL, &good_writes, flags);
847
b128c09f
BB
848 for (int v = 0; v < svdcount; v++)
849 vdev_uberblock_sync(zio, ub, svd[v], flags);
34dc7c2f
BB
850
851 (void) zio_wait(zio);
852
853 /*
854 * Flush the uberblocks to disk. This ensures that the odd labels
855 * are no longer needed (because the new uberblocks and the even
856 * labels are safely on disk), so it is safe to overwrite them.
857 */
858 zio = zio_root(spa, NULL, NULL, flags);
859
b128c09f 860 for (int v = 0; v < svdcount; v++)
34dc7c2f
BB
861 zio_flush(zio, svd[v]);
862
863 (void) zio_wait(zio);
864
865 return (good_writes >= 1 ? 0 : EIO);
866}
867
868/*
869 * On success, increment the count of good writes for our top-level vdev.
870 */
871static void
872vdev_label_sync_done(zio_t *zio)
873{
874 uint64_t *good_writes = zio->io_private;
875
876 if (zio->io_error == 0)
877 atomic_add_64(good_writes, 1);
878}
879
880/*
881 * If there weren't enough good writes, indicate failure to the parent.
882 */
883static void
884vdev_label_sync_top_done(zio_t *zio)
885{
886 uint64_t *good_writes = zio->io_private;
887
888 if (*good_writes == 0)
889 zio->io_error = EIO;
890
891 kmem_free(good_writes, sizeof (uint64_t));
892}
893
b128c09f
BB
894/*
895 * We ignore errors for log and cache devices, simply free the private data.
896 */
897static void
898vdev_label_sync_ignore_done(zio_t *zio)
899{
900 kmem_free(zio->io_private, sizeof (uint64_t));
901}
902
34dc7c2f
BB
903/*
904 * Write all even or odd labels to all leaves of the specified vdev.
905 */
906static void
b128c09f 907vdev_label_sync(zio_t *zio, vdev_t *vd, int l, uint64_t txg, int flags)
34dc7c2f
BB
908{
909 nvlist_t *label;
910 vdev_phys_t *vp;
911 char *buf;
912 size_t buflen;
34dc7c2f 913
b128c09f
BB
914 for (int c = 0; c < vd->vdev_children; c++)
915 vdev_label_sync(zio, vd->vdev_child[c], l, txg, flags);
34dc7c2f
BB
916
917 if (!vd->vdev_ops->vdev_op_leaf)
918 return;
919
b128c09f 920 if (!vdev_writeable(vd))
34dc7c2f
BB
921 return;
922
923 /*
924 * Generate a label describing the top-level config to which we belong.
925 */
926 label = spa_config_generate(vd->vdev_spa, vd, txg, B_FALSE);
927
928 vp = zio_buf_alloc(sizeof (vdev_phys_t));
929 bzero(vp, sizeof (vdev_phys_t));
930
931 buf = vp->vp_nvlist;
932 buflen = sizeof (vp->vp_nvlist);
933
934 if (nvlist_pack(label, &buf, &buflen, NV_ENCODE_XDR, KM_SLEEP) == 0) {
935 for (; l < VDEV_LABELS; l += 2) {
936 vdev_label_write(zio, vd, l, vp,
937 offsetof(vdev_label_t, vl_vdev_phys),
938 sizeof (vdev_phys_t),
939 vdev_label_sync_done, zio->io_private,
b128c09f 940 flags | ZIO_FLAG_DONT_PROPAGATE);
34dc7c2f
BB
941 }
942 }
943
944 zio_buf_free(vp, sizeof (vdev_phys_t));
945 nvlist_free(label);
946}
947
948int
b128c09f 949vdev_label_sync_list(spa_t *spa, int l, uint64_t txg, int flags)
34dc7c2f 950{
b128c09f 951 list_t *dl = &spa->spa_config_dirty_list;
34dc7c2f
BB
952 vdev_t *vd;
953 zio_t *zio;
954 int error;
955
956 /*
957 * Write the new labels to disk.
958 */
959 zio = zio_root(spa, NULL, NULL, flags);
960
961 for (vd = list_head(dl); vd != NULL; vd = list_next(dl, vd)) {
962 uint64_t *good_writes = kmem_zalloc(sizeof (uint64_t),
963 KM_SLEEP);
d164b209 964 zio_t *vio = zio_null(zio, spa, NULL,
b128c09f
BB
965 (vd->vdev_islog || vd->vdev_aux != NULL) ?
966 vdev_label_sync_ignore_done : vdev_label_sync_top_done,
34dc7c2f 967 good_writes, flags);
b128c09f 968 vdev_label_sync(vio, vd, l, txg, flags);
34dc7c2f
BB
969 zio_nowait(vio);
970 }
971
972 error = zio_wait(zio);
973
974 /*
975 * Flush the new labels to disk.
976 */
977 zio = zio_root(spa, NULL, NULL, flags);
978
979 for (vd = list_head(dl); vd != NULL; vd = list_next(dl, vd))
980 zio_flush(zio, vd);
981
982 (void) zio_wait(zio);
983
984 return (error);
985}
986
987/*
988 * Sync the uberblock and any changes to the vdev configuration.
989 *
990 * The order of operations is carefully crafted to ensure that
991 * if the system panics or loses power at any time, the state on disk
992 * is still transactionally consistent. The in-line comments below
993 * describe the failure semantics at each stage.
994 *
995 * Moreover, vdev_config_sync() is designed to be idempotent: if it fails
996 * at any time, you can just call it again, and it will resume its work.
997 */
998int
999vdev_config_sync(vdev_t **svd, int svdcount, uint64_t txg)
1000{
1001 spa_t *spa = svd[0]->vdev_spa;
1002 uberblock_t *ub = &spa->spa_uberblock;
1003 vdev_t *vd;
1004 zio_t *zio;
1005 int error;
b128c09f 1006 int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL;
34dc7c2f
BB
1007
1008 ASSERT(ub->ub_txg <= txg);
1009
1010 /*
1011 * If this isn't a resync due to I/O errors,
1012 * and nothing changed in this transaction group,
1013 * and the vdev configuration hasn't changed,
1014 * then there's nothing to do.
1015 */
1016 if (ub->ub_txg < txg &&
1017 uberblock_update(ub, spa->spa_root_vdev, txg) == B_FALSE &&
b128c09f 1018 list_is_empty(&spa->spa_config_dirty_list))
34dc7c2f
BB
1019 return (0);
1020
1021 if (txg > spa_freeze_txg(spa))
1022 return (0);
1023
1024 ASSERT(txg <= spa->spa_final_txg);
1025
1026 /*
1027 * Flush the write cache of every disk that's been written to
1028 * in this transaction group. This ensures that all blocks
1029 * written in this txg will be committed to stable storage
1030 * before any uberblock that references them.
1031 */
1032 zio = zio_root(spa, NULL, NULL, flags);
1033
1034 for (vd = txg_list_head(&spa->spa_vdev_txg_list, TXG_CLEAN(txg)); vd;
1035 vd = txg_list_next(&spa->spa_vdev_txg_list, vd, TXG_CLEAN(txg)))
1036 zio_flush(zio, vd);
1037
1038 (void) zio_wait(zio);
1039
1040 /*
1041 * Sync out the even labels (L0, L2) for every dirty vdev. If the
1042 * system dies in the middle of this process, that's OK: all of the
1043 * even labels that made it to disk will be newer than any uberblock,
1044 * and will therefore be considered invalid. The odd labels (L1, L3),
1045 * which have not yet been touched, will still be valid. We flush
1046 * the new labels to disk to ensure that all even-label updates
1047 * are committed to stable storage before the uberblock update.
1048 */
b128c09f 1049 if ((error = vdev_label_sync_list(spa, 0, txg, flags)) != 0)
34dc7c2f
BB
1050 return (error);
1051
1052 /*
1053 * Sync the uberblocks to all vdevs in svd[].
1054 * If the system dies in the middle of this step, there are two cases
1055 * to consider, and the on-disk state is consistent either way:
1056 *
1057 * (1) If none of the new uberblocks made it to disk, then the
1058 * previous uberblock will be the newest, and the odd labels
1059 * (which had not yet been touched) will be valid with respect
1060 * to that uberblock.
1061 *
1062 * (2) If one or more new uberblocks made it to disk, then they
1063 * will be the newest, and the even labels (which had all
1064 * been successfully committed) will be valid with respect
1065 * to the new uberblocks.
1066 */
1067 if ((error = vdev_uberblock_sync_list(svd, svdcount, ub, flags)) != 0)
1068 return (error);
1069
1070 /*
1071 * Sync out odd labels for every dirty vdev. If the system dies
1072 * in the middle of this process, the even labels and the new
1073 * uberblocks will suffice to open the pool. The next time
1074 * the pool is opened, the first thing we'll do -- before any
1075 * user data is modified -- is mark every vdev dirty so that
1076 * all labels will be brought up to date. We flush the new labels
1077 * to disk to ensure that all odd-label updates are committed to
1078 * stable storage before the next transaction group begins.
1079 */
b128c09f 1080 return (vdev_label_sync_list(spa, 1, txg, flags));
34dc7c2f 1081}