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 2010 Sun Microsystems, Inc. All rights reserved.
23 * Use is subject to license terms.
24 */
25
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 */
150 uint64_t
vdev_label_offset(uint64_t psize,int l,uint64_t offset)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
vdev_label_number(uint64_t psize,uint64_t offset)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
vdev_label_read(zio_t * zio,vdev_t * vd,int l,void * buf,uint64_t offset,uint64_t size,zio_done_func_t * done,void * private,int flags)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
vdev_label_write(zio_t * zio,vdev_t * vd,int l,void * buf,uint64_t offset,uint64_t size,zio_done_func_t * done,void * private,int flags)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 *
vdev_config_generate(spa_t * spa,vdev_t * vd,boolean_t getstats,boolean_t isspare,boolean_t isl2cache)210 vdev_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_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 (!isspare && !isl2cache && vd == vd->vdev_top) {
274 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_METASLAB_ARRAY,
275 vd->vdev_ms_array) == 0);
276 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_METASLAB_SHIFT,
277 vd->vdev_ms_shift) == 0);
278 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_ASHIFT,
279 vd->vdev_ashift) == 0);
280 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_ASIZE,
281 vd->vdev_asize) == 0);
282 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_IS_LOG,
283 vd->vdev_islog) == 0);
284 }
285
286 if (vd->vdev_dtl_smo.smo_object != 0)
287 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_DTL,
288 vd->vdev_dtl_smo.smo_object) == 0);
289
290 if (vd->vdev_crtxg)
291 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_CREATE_TXG,
292 vd->vdev_crtxg) == 0);
293
294 if (getstats) {
295 vdev_stat_t vs;
296 vdev_get_stats(vd, &vs);
297 VERIFY(nvlist_add_uint64_array(nv, ZPOOL_CONFIG_STATS,
298 (uint64_t *)&vs, sizeof (vs) / sizeof (uint64_t)) == 0);
299 }
300
301 if (!vd->vdev_ops->vdev_op_leaf) {
302 nvlist_t **child;
303 int c;
304
305 ASSERT(!vd->vdev_ishole);
306
307 child = kmem_alloc(vd->vdev_children * sizeof (nvlist_t *),
308 KM_SLEEP);
309
310 for (c = 0; c < vd->vdev_children; c++)
311 child[c] = vdev_config_generate(spa, vd->vdev_child[c],
312 getstats, isspare, isl2cache);
313
314 VERIFY(nvlist_add_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
315 child, vd->vdev_children) == 0);
316
317 for (c = 0; c < vd->vdev_children; c++)
318 nvlist_free(child[c]);
319
320 kmem_free(child, vd->vdev_children * sizeof (nvlist_t *));
321
322 } else {
323 const char *aux = NULL;
324
325 if (vd->vdev_offline && !vd->vdev_tmpoffline)
326 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_OFFLINE,
327 B_TRUE) == 0);
328 if (vd->vdev_faulted)
329 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_FAULTED,
330 B_TRUE) == 0);
331 if (vd->vdev_degraded)
332 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_DEGRADED,
333 B_TRUE) == 0);
334 if (vd->vdev_removed)
335 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_REMOVED,
336 B_TRUE) == 0);
337 if (vd->vdev_unspare)
338 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_UNSPARE,
339 B_TRUE) == 0);
340 if (vd->vdev_ishole)
341 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_IS_HOLE,
342 B_TRUE) == 0);
343
344 switch (vd->vdev_stat.vs_aux) {
345 case VDEV_AUX_ERR_EXCEEDED:
346 aux = "err_exceeded";
347 break;
348
349 case VDEV_AUX_EXTERNAL:
350 aux = "external";
351 break;
352 }
353
354 if (aux != NULL)
355 VERIFY(nvlist_add_string(nv, ZPOOL_CONFIG_AUX_STATE,
356 aux) == 0);
357
358 if (vd->vdev_splitting && vd->vdev_orig_guid != 0LL) {
359 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_ORIG_GUID,
360 vd->vdev_orig_guid) == 0);
361 }
362 }
363
364 return (nv);
365 }
366
367 /*
368 * Generate a view of the top-level vdevs. If we currently have holes
369 * in the namespace, then generate an array which contains a list of holey
370 * vdevs. Additionally, add the number of top-level children that currently
371 * exist.
372 */
373 void
vdev_top_config_generate(spa_t * spa,nvlist_t * config)374 vdev_top_config_generate(spa_t *spa, nvlist_t *config)
375 {
376 vdev_t *rvd = spa->spa_root_vdev;
377 uint64_t *array;
378 uint_t idx;
379
380 array = kmem_alloc(rvd->vdev_children * sizeof (uint64_t), KM_SLEEP);
381
382 idx = 0;
383 for (int c = 0; c < rvd->vdev_children; c++) {
384 vdev_t *tvd = rvd->vdev_child[c];
385
386 if (tvd->vdev_ishole)
387 array[idx++] = c;
388 }
389
390 if (idx) {
391 VERIFY(nvlist_add_uint64_array(config, ZPOOL_CONFIG_HOLE_ARRAY,
392 array, idx) == 0);
393 }
394
395 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VDEV_CHILDREN,
396 rvd->vdev_children) == 0);
397
398 kmem_free(array, rvd->vdev_children * sizeof (uint64_t));
399 }
400
401 nvlist_t *
vdev_label_read_config(vdev_t * vd)402 vdev_label_read_config(vdev_t *vd)
403 {
404 spa_t *spa = vd->vdev_spa;
405 nvlist_t *config = NULL;
406 vdev_phys_t *vp;
407 zio_t *zio;
408 int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL |
409 ZIO_FLAG_SPECULATIVE;
410
411 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
412
413 if (!vdev_readable(vd))
414 return (NULL);
415
416 vp = zio_buf_alloc(sizeof (vdev_phys_t));
417
418 retry:
419 for (int l = 0; l < VDEV_LABELS; l++) {
420
421 zio = zio_root(spa, NULL, NULL, flags);
422
423 vdev_label_read(zio, vd, l, vp,
424 offsetof(vdev_label_t, vl_vdev_phys),
425 sizeof (vdev_phys_t), NULL, NULL, flags);
426
427 if (zio_wait(zio) == 0 &&
428 nvlist_unpack(vp->vp_nvlist, sizeof (vp->vp_nvlist),
429 &config, 0) == 0)
430 break;
431
432 if (config != NULL) {
433 nvlist_free(config);
434 config = NULL;
435 }
436 }
437
438 if (config == NULL && !(flags & ZIO_FLAG_TRYHARD)) {
439 flags |= ZIO_FLAG_TRYHARD;
440 goto retry;
441 }
442
443 zio_buf_free(vp, sizeof (vdev_phys_t));
444
445 return (config);
446 }
447
448 /*
449 * Determine if a device is in use. The 'spare_guid' parameter will be filled
450 * in with the device guid if this spare is active elsewhere on the system.
451 */
452 static boolean_t
vdev_inuse(vdev_t * vd,uint64_t crtxg,vdev_labeltype_t reason,uint64_t * spare_guid,uint64_t * l2cache_guid)453 vdev_inuse(vdev_t *vd, uint64_t crtxg, vdev_labeltype_t reason,
454 uint64_t *spare_guid, uint64_t *l2cache_guid)
455 {
456 spa_t *spa = vd->vdev_spa;
457 uint64_t state, pool_guid, device_guid, txg, spare_pool;
458 uint64_t vdtxg = 0;
459 nvlist_t *label;
460
461 if (spare_guid)
462 *spare_guid = 0ULL;
463 if (l2cache_guid)
464 *l2cache_guid = 0ULL;
465
466 /*
467 * Read the label, if any, and perform some basic sanity checks.
468 */
469 if ((label = vdev_label_read_config(vd)) == NULL)
470 return (B_FALSE);
471
472 (void) nvlist_lookup_uint64(label, ZPOOL_CONFIG_CREATE_TXG,
473 &vdtxg);
474
475 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE,
476 &state) != 0 ||
477 nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID,
478 &device_guid) != 0) {
479 nvlist_free(label);
480 return (B_FALSE);
481 }
482
483 if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
484 (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_GUID,
485 &pool_guid) != 0 ||
486 nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG,
487 &txg) != 0)) {
488 nvlist_free(label);
489 return (B_FALSE);
490 }
491
492 nvlist_free(label);
493
494 /*
495 * Check to see if this device indeed belongs to the pool it claims to
496 * be a part of. The only way this is allowed is if the device is a hot
497 * spare (which we check for later on).
498 */
499 if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
500 !spa_guid_exists(pool_guid, device_guid) &&
501 !spa_spare_exists(device_guid, NULL, NULL) &&
502 !spa_l2cache_exists(device_guid, NULL))
503 return (B_FALSE);
504
505 /*
506 * If the transaction group is zero, then this an initialized (but
507 * unused) label. This is only an error if the create transaction
508 * on-disk is the same as the one we're using now, in which case the
509 * user has attempted to add the same vdev multiple times in the same
510 * transaction.
511 */
512 if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
513 txg == 0 && vdtxg == crtxg)
514 return (B_TRUE);
515
516 /*
517 * Check to see if this is a spare device. We do an explicit check for
518 * spa_has_spare() here because it may be on our pending list of spares
519 * to add. We also check if it is an l2cache device.
520 */
521 if (spa_spare_exists(device_guid, &spare_pool, NULL) ||
522 spa_has_spare(spa, device_guid)) {
523 if (spare_guid)
524 *spare_guid = device_guid;
525
526 switch (reason) {
527 case VDEV_LABEL_CREATE:
528 case VDEV_LABEL_L2CACHE:
529 return (B_TRUE);
530
531 case VDEV_LABEL_REPLACE:
532 return (!spa_has_spare(spa, device_guid) ||
533 spare_pool != 0ULL);
534
535 case VDEV_LABEL_SPARE:
536 return (spa_has_spare(spa, device_guid));
537 }
538 }
539
540 /*
541 * Check to see if this is an l2cache device.
542 */
543 if (spa_l2cache_exists(device_guid, NULL))
544 return (B_TRUE);
545
546 /*
547 * If the device is marked ACTIVE, then this device is in use by another
548 * pool on the system.
549 */
550 return (state == POOL_STATE_ACTIVE);
551 }
552
553 /*
554 * Initialize a vdev label. We check to make sure each leaf device is not in
555 * use, and writable. We put down an initial label which we will later
556 * overwrite with a complete label. Note that it's important to do this
557 * sequentially, not in parallel, so that we catch cases of multiple use of the
558 * same leaf vdev in the vdev we're creating -- e.g. mirroring a disk with
559 * itself.
560 */
561 int
vdev_label_init(vdev_t * vd,uint64_t crtxg,vdev_labeltype_t reason)562 vdev_label_init(vdev_t *vd, uint64_t crtxg, vdev_labeltype_t reason)
563 {
564 spa_t *spa = vd->vdev_spa;
565 nvlist_t *label;
566 vdev_phys_t *vp;
567 char *pad2;
568 uberblock_t *ub;
569 zio_t *zio;
570 char *buf;
571 size_t buflen;
572 int error;
573 uint64_t spare_guid, l2cache_guid;
574 int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL;
575
576 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
577
578 for (int c = 0; c < vd->vdev_children; c++)
579 if ((error = vdev_label_init(vd->vdev_child[c],
580 crtxg, reason)) != 0)
581 return (error);
582
583 /* Track the creation time for this vdev */
584 vd->vdev_crtxg = crtxg;
585
586 if (!vd->vdev_ops->vdev_op_leaf)
587 return (0);
588
589 /*
590 * Dead vdevs cannot be initialized.
591 */
592 if (vdev_is_dead(vd))
593 return (EIO);
594
595 /*
596 * Determine if the vdev is in use.
597 */
598 if (reason != VDEV_LABEL_REMOVE && reason != VDEV_LABEL_SPLIT &&
599 vdev_inuse(vd, crtxg, reason, &spare_guid, &l2cache_guid))
600 return (EBUSY);
601
602 /*
603 * If this is a request to add or replace a spare or l2cache device
604 * that is in use elsewhere on the system, then we must update the
605 * guid (which was initialized to a random value) to reflect the
606 * actual GUID (which is shared between multiple pools).
607 */
608 if (reason != VDEV_LABEL_REMOVE && reason != VDEV_LABEL_L2CACHE &&
609 spare_guid != 0ULL) {
610 uint64_t guid_delta = spare_guid - vd->vdev_guid;
611
612 vd->vdev_guid += guid_delta;
613
614 for (vdev_t *pvd = vd; pvd != NULL; pvd = pvd->vdev_parent)
615 pvd->vdev_guid_sum += guid_delta;
616
617 /*
618 * If this is a replacement, then we want to fallthrough to the
619 * rest of the code. If we're adding a spare, then it's already
620 * labeled appropriately and we can just return.
621 */
622 if (reason == VDEV_LABEL_SPARE)
623 return (0);
624 ASSERT(reason == VDEV_LABEL_REPLACE ||
625 reason == VDEV_LABEL_SPLIT);
626 }
627
628 if (reason != VDEV_LABEL_REMOVE && reason != VDEV_LABEL_SPARE &&
629 l2cache_guid != 0ULL) {
630 uint64_t guid_delta = l2cache_guid - vd->vdev_guid;
631
632 vd->vdev_guid += guid_delta;
633
634 for (vdev_t *pvd = vd; pvd != NULL; pvd = pvd->vdev_parent)
635 pvd->vdev_guid_sum += guid_delta;
636
637 /*
638 * If this is a replacement, then we want to fallthrough to the
639 * rest of the code. If we're adding an l2cache, then it's
640 * already labeled appropriately and we can just return.
641 */
642 if (reason == VDEV_LABEL_L2CACHE)
643 return (0);
644 ASSERT(reason == VDEV_LABEL_REPLACE);
645 }
646
647 /*
648 * Initialize its label.
649 */
650 vp = zio_buf_alloc(sizeof (vdev_phys_t));
651 bzero(vp, sizeof (vdev_phys_t));
652
653 /*
654 * Generate a label describing the pool and our top-level vdev.
655 * We mark it as being from txg 0 to indicate that it's not
656 * really part of an active pool just yet. The labels will
657 * be written again with a meaningful txg by spa_sync().
658 */
659 if (reason == VDEV_LABEL_SPARE ||
660 (reason == VDEV_LABEL_REMOVE && vd->vdev_isspare)) {
661 /*
662 * For inactive hot spares, we generate a special label that
663 * identifies as a mutually shared hot spare. We write the
664 * label if we are adding a hot spare, or if we are removing an
665 * active hot spare (in which case we want to revert the
666 * labels).
667 */
668 VERIFY(nvlist_alloc(&label, NV_UNIQUE_NAME, KM_SLEEP) == 0);
669
670 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_VERSION,
671 spa_version(spa)) == 0);
672 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_POOL_STATE,
673 POOL_STATE_SPARE) == 0);
674 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_GUID,
675 vd->vdev_guid) == 0);
676 } else if (reason == VDEV_LABEL_L2CACHE ||
677 (reason == VDEV_LABEL_REMOVE && vd->vdev_isl2cache)) {
678 /*
679 * For level 2 ARC devices, add a special label.
680 */
681 VERIFY(nvlist_alloc(&label, NV_UNIQUE_NAME, KM_SLEEP) == 0);
682
683 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_VERSION,
684 spa_version(spa)) == 0);
685 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_POOL_STATE,
686 POOL_STATE_L2CACHE) == 0);
687 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_GUID,
688 vd->vdev_guid) == 0);
689 } else {
690 uint64_t txg = 0ULL;
691
692 if (reason == VDEV_LABEL_SPLIT)
693 txg = spa->spa_uberblock.ub_txg;
694 label = spa_config_generate(spa, vd, txg, B_FALSE);
695
696 /*
697 * Add our creation time. This allows us to detect multiple
698 * vdev uses as described above, and automatically expires if we
699 * fail.
700 */
701 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_CREATE_TXG,
702 crtxg) == 0);
703 }
704
705 buf = vp->vp_nvlist;
706 buflen = sizeof (vp->vp_nvlist);
707
708 error = nvlist_pack(label, &buf, &buflen, NV_ENCODE_XDR, KM_SLEEP);
709 if (error != 0) {
710 nvlist_free(label);
711 zio_buf_free(vp, sizeof (vdev_phys_t));
712 /* EFAULT means nvlist_pack ran out of room */
713 return (error == EFAULT ? ENAMETOOLONG : EINVAL);
714 }
715
716 /*
717 * Initialize uberblock template.
718 */
719 ub = zio_buf_alloc(VDEV_UBERBLOCK_RING);
720 bzero(ub, VDEV_UBERBLOCK_RING);
721 *ub = spa->spa_uberblock;
722 ub->ub_txg = 0;
723
724 /* Initialize the 2nd padding area. */
725 pad2 = zio_buf_alloc(VDEV_PAD_SIZE);
726 bzero(pad2, VDEV_PAD_SIZE);
727
728 /*
729 * Write everything in parallel.
730 */
731 retry:
732 zio = zio_root(spa, NULL, NULL, flags);
733
734 for (int l = 0; l < VDEV_LABELS; l++) {
735
736 vdev_label_write(zio, vd, l, vp,
737 offsetof(vdev_label_t, vl_vdev_phys),
738 sizeof (vdev_phys_t), NULL, NULL, flags);
739
740 /*
741 * Skip the 1st padding area.
742 * Zero out the 2nd padding area where it might have
743 * left over data from previous filesystem format.
744 */
745 vdev_label_write(zio, vd, l, pad2,
746 offsetof(vdev_label_t, vl_pad2),
747 VDEV_PAD_SIZE, NULL, NULL, flags);
748
749 vdev_label_write(zio, vd, l, ub,
750 offsetof(vdev_label_t, vl_uberblock),
751 VDEV_UBERBLOCK_RING, NULL, NULL, flags);
752 }
753
754 error = zio_wait(zio);
755
756 if (error != 0 && !(flags & ZIO_FLAG_TRYHARD)) {
757 flags |= ZIO_FLAG_TRYHARD;
758 goto retry;
759 }
760
761 nvlist_free(label);
762 zio_buf_free(pad2, VDEV_PAD_SIZE);
763 zio_buf_free(ub, VDEV_UBERBLOCK_RING);
764 zio_buf_free(vp, sizeof (vdev_phys_t));
765
766 /*
767 * If this vdev hasn't been previously identified as a spare, then we
768 * mark it as such only if a) we are labeling it as a spare, or b) it
769 * exists as a spare elsewhere in the system. Do the same for
770 * level 2 ARC devices.
771 */
772 if (error == 0 && !vd->vdev_isspare &&
773 (reason == VDEV_LABEL_SPARE ||
774 spa_spare_exists(vd->vdev_guid, NULL, NULL)))
775 spa_spare_add(vd);
776
777 if (error == 0 && !vd->vdev_isl2cache &&
778 (reason == VDEV_LABEL_L2CACHE ||
779 spa_l2cache_exists(vd->vdev_guid, NULL)))
780 spa_l2cache_add(vd);
781
782 return (error);
783 }
784
785 /*
786 * ==========================================================================
787 * uberblock load/sync
788 * ==========================================================================
789 */
790
791 /*
792 * Consider the following situation: txg is safely synced to disk. We've
793 * written the first uberblock for txg + 1, and then we lose power. When we
794 * come back up, we fail to see the uberblock for txg + 1 because, say,
795 * it was on a mirrored device and the replica to which we wrote txg + 1
796 * is now offline. If we then make some changes and sync txg + 1, and then
797 * the missing replica comes back, then for a new seconds we'll have two
798 * conflicting uberblocks on disk with the same txg. The solution is simple:
799 * among uberblocks with equal txg, choose the one with the latest timestamp.
800 */
801 static int
vdev_uberblock_compare(uberblock_t * ub1,uberblock_t * ub2)802 vdev_uberblock_compare(uberblock_t *ub1, uberblock_t *ub2)
803 {
804 if (ub1->ub_txg < ub2->ub_txg)
805 return (-1);
806 if (ub1->ub_txg > ub2->ub_txg)
807 return (1);
808
809 if (ub1->ub_timestamp < ub2->ub_timestamp)
810 return (-1);
811 if (ub1->ub_timestamp > ub2->ub_timestamp)
812 return (1);
813
814 return (0);
815 }
816
817 static void
vdev_uberblock_load_done(zio_t * zio)818 vdev_uberblock_load_done(zio_t *zio)
819 {
820 spa_t *spa = zio->io_spa;
821 zio_t *rio = zio->io_private;
822 uberblock_t *ub = zio->io_data;
823 uberblock_t *ubbest = rio->io_private;
824
825 ASSERT3U(zio->io_size, ==, VDEV_UBERBLOCK_SIZE(zio->io_vd));
826
827 if (zio->io_error == 0 && uberblock_verify(ub) == 0) {
828 mutex_enter(&rio->io_lock);
829 if (ub->ub_txg <= spa->spa_load_max_txg &&
830 vdev_uberblock_compare(ub, ubbest) > 0)
831 *ubbest = *ub;
832 mutex_exit(&rio->io_lock);
833 }
834
835 zio_buf_free(zio->io_data, zio->io_size);
836 }
837
838 void
vdev_uberblock_load(zio_t * zio,vdev_t * vd,uberblock_t * ubbest)839 vdev_uberblock_load(zio_t *zio, vdev_t *vd, uberblock_t *ubbest)
840 {
841 spa_t *spa = vd->vdev_spa;
842 vdev_t *rvd = spa->spa_root_vdev;
843 int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL |
844 ZIO_FLAG_SPECULATIVE | ZIO_FLAG_TRYHARD;
845
846 if (vd == rvd) {
847 ASSERT(zio == NULL);
848 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
849 zio = zio_root(spa, NULL, ubbest, flags);
850 bzero(ubbest, sizeof (uberblock_t));
851 }
852
853 ASSERT(zio != NULL);
854
855 for (int c = 0; c < vd->vdev_children; c++)
856 vdev_uberblock_load(zio, vd->vdev_child[c], ubbest);
857
858 if (vd->vdev_ops->vdev_op_leaf && vdev_readable(vd)) {
859 for (int l = 0; l < VDEV_LABELS; l++) {
860 for (int n = 0; n < VDEV_UBERBLOCK_COUNT(vd); n++) {
861 vdev_label_read(zio, vd, l,
862 zio_buf_alloc(VDEV_UBERBLOCK_SIZE(vd)),
863 VDEV_UBERBLOCK_OFFSET(vd, n),
864 VDEV_UBERBLOCK_SIZE(vd),
865 vdev_uberblock_load_done, zio, flags);
866 }
867 }
868 }
869
870 if (vd == rvd) {
871 (void) zio_wait(zio);
872 spa_config_exit(spa, SCL_ALL, FTAG);
873 }
874 }
875
876 /*
877 * On success, increment root zio's count of good writes.
878 * We only get credit for writes to known-visible vdevs; see spa_vdev_add().
879 */
880 static void
vdev_uberblock_sync_done(zio_t * zio)881 vdev_uberblock_sync_done(zio_t *zio)
882 {
883 uint64_t *good_writes = zio->io_private;
884
885 if (zio->io_error == 0 && zio->io_vd->vdev_top->vdev_ms_array != 0)
886 atomic_add_64(good_writes, 1);
887 }
888
889 /*
890 * Write the uberblock to all labels of all leaves of the specified vdev.
891 */
892 static void
vdev_uberblock_sync(zio_t * zio,uberblock_t * ub,vdev_t * vd,int flags)893 vdev_uberblock_sync(zio_t *zio, uberblock_t *ub, vdev_t *vd, int flags)
894 {
895 uberblock_t *ubbuf;
896 int n;
897
898 for (int c = 0; c < vd->vdev_children; c++)
899 vdev_uberblock_sync(zio, ub, vd->vdev_child[c], flags);
900
901 if (!vd->vdev_ops->vdev_op_leaf)
902 return;
903
904 if (!vdev_writeable(vd))
905 return;
906
907 n = ub->ub_txg & (VDEV_UBERBLOCK_COUNT(vd) - 1);
908
909 ubbuf = zio_buf_alloc(VDEV_UBERBLOCK_SIZE(vd));
910 bzero(ubbuf, VDEV_UBERBLOCK_SIZE(vd));
911 *ubbuf = *ub;
912
913 for (int l = 0; l < VDEV_LABELS; l++)
914 vdev_label_write(zio, vd, l, ubbuf,
915 VDEV_UBERBLOCK_OFFSET(vd, n), VDEV_UBERBLOCK_SIZE(vd),
916 vdev_uberblock_sync_done, zio->io_private,
917 flags | ZIO_FLAG_DONT_PROPAGATE);
918
919 zio_buf_free(ubbuf, VDEV_UBERBLOCK_SIZE(vd));
920 }
921
922 int
vdev_uberblock_sync_list(vdev_t ** svd,int svdcount,uberblock_t * ub,int flags)923 vdev_uberblock_sync_list(vdev_t **svd, int svdcount, uberblock_t *ub, int flags)
924 {
925 spa_t *spa = svd[0]->vdev_spa;
926 zio_t *zio;
927 uint64_t good_writes = 0;
928
929 zio = zio_root(spa, NULL, &good_writes, flags);
930
931 for (int v = 0; v < svdcount; v++)
932 vdev_uberblock_sync(zio, ub, svd[v], flags);
933
934 (void) zio_wait(zio);
935
936 /*
937 * Flush the uberblocks to disk. This ensures that the odd labels
938 * are no longer needed (because the new uberblocks and the even
939 * labels are safely on disk), so it is safe to overwrite them.
940 */
941 zio = zio_root(spa, NULL, NULL, flags);
942
943 for (int v = 0; v < svdcount; v++)
944 zio_flush(zio, svd[v]);
945
946 (void) zio_wait(zio);
947
948 return (good_writes >= 1 ? 0 : EIO);
949 }
950
951 /*
952 * On success, increment the count of good writes for our top-level vdev.
953 */
954 static void
vdev_label_sync_done(zio_t * zio)955 vdev_label_sync_done(zio_t *zio)
956 {
957 uint64_t *good_writes = zio->io_private;
958
959 if (zio->io_error == 0)
960 atomic_add_64(good_writes, 1);
961 }
962
963 /*
964 * If there weren't enough good writes, indicate failure to the parent.
965 */
966 static void
vdev_label_sync_top_done(zio_t * zio)967 vdev_label_sync_top_done(zio_t *zio)
968 {
969 uint64_t *good_writes = zio->io_private;
970
971 if (*good_writes == 0)
972 zio->io_error = EIO;
973
974 kmem_free(good_writes, sizeof (uint64_t));
975 }
976
977 /*
978 * We ignore errors for log and cache devices, simply free the private data.
979 */
980 static void
vdev_label_sync_ignore_done(zio_t * zio)981 vdev_label_sync_ignore_done(zio_t *zio)
982 {
983 kmem_free(zio->io_private, sizeof (uint64_t));
984 }
985
986 /*
987 * Write all even or odd labels to all leaves of the specified vdev.
988 */
989 static void
vdev_label_sync(zio_t * zio,vdev_t * vd,int l,uint64_t txg,int flags)990 vdev_label_sync(zio_t *zio, vdev_t *vd, int l, uint64_t txg, int flags)
991 {
992 nvlist_t *label;
993 vdev_phys_t *vp;
994 char *buf;
995 size_t buflen;
996
997 for (int c = 0; c < vd->vdev_children; c++)
998 vdev_label_sync(zio, vd->vdev_child[c], l, txg, flags);
999
1000 if (!vd->vdev_ops->vdev_op_leaf)
1001 return;
1002
1003 if (!vdev_writeable(vd))
1004 return;
1005
1006 /*
1007 * Generate a label describing the top-level config to which we belong.
1008 */
1009 label = spa_config_generate(vd->vdev_spa, vd, txg, B_FALSE);
1010
1011 vp = zio_buf_alloc(sizeof (vdev_phys_t));
1012 bzero(vp, sizeof (vdev_phys_t));
1013
1014 buf = vp->vp_nvlist;
1015 buflen = sizeof (vp->vp_nvlist);
1016
1017 if (nvlist_pack(label, &buf, &buflen, NV_ENCODE_XDR, KM_SLEEP) == 0) {
1018 for (; l < VDEV_LABELS; l += 2) {
1019 vdev_label_write(zio, vd, l, vp,
1020 offsetof(vdev_label_t, vl_vdev_phys),
1021 sizeof (vdev_phys_t),
1022 vdev_label_sync_done, zio->io_private,
1023 flags | ZIO_FLAG_DONT_PROPAGATE);
1024 }
1025 }
1026
1027 zio_buf_free(vp, sizeof (vdev_phys_t));
1028 nvlist_free(label);
1029 }
1030
1031 int
vdev_label_sync_list(spa_t * spa,int l,uint64_t txg,int flags)1032 vdev_label_sync_list(spa_t *spa, int l, uint64_t txg, int flags)
1033 {
1034 list_t *dl = &spa->spa_config_dirty_list;
1035 vdev_t *vd;
1036 zio_t *zio;
1037 int error;
1038
1039 /*
1040 * Write the new labels to disk.
1041 */
1042 zio = zio_root(spa, NULL, NULL, flags);
1043
1044 for (vd = list_head(dl); vd != NULL; vd = list_next(dl, vd)) {
1045 uint64_t *good_writes = kmem_zalloc(sizeof (uint64_t),
1046 KM_SLEEP);
1047
1048 ASSERT(!vd->vdev_ishole);
1049
1050 zio_t *vio = zio_null(zio, spa, NULL,
1051 (vd->vdev_islog || vd->vdev_aux != NULL) ?
1052 vdev_label_sync_ignore_done : vdev_label_sync_top_done,
1053 good_writes, flags);
1054 vdev_label_sync(vio, vd, l, txg, flags);
1055 zio_nowait(vio);
1056 }
1057
1058 error = zio_wait(zio);
1059
1060 /*
1061 * Flush the new labels to disk.
1062 */
1063 zio = zio_root(spa, NULL, NULL, flags);
1064
1065 for (vd = list_head(dl); vd != NULL; vd = list_next(dl, vd))
1066 zio_flush(zio, vd);
1067
1068 (void) zio_wait(zio);
1069
1070 return (error);
1071 }
1072
1073 /*
1074 * Sync the uberblock and any changes to the vdev configuration.
1075 *
1076 * The order of operations is carefully crafted to ensure that
1077 * if the system panics or loses power at any time, the state on disk
1078 * is still transactionally consistent. The in-line comments below
1079 * describe the failure semantics at each stage.
1080 *
1081 * Moreover, vdev_config_sync() is designed to be idempotent: if it fails
1082 * at any time, you can just call it again, and it will resume its work.
1083 */
1084 int
vdev_config_sync(vdev_t ** svd,int svdcount,uint64_t txg,boolean_t tryhard)1085 vdev_config_sync(vdev_t **svd, int svdcount, uint64_t txg, boolean_t tryhard)
1086 {
1087 spa_t *spa = svd[0]->vdev_spa;
1088 uberblock_t *ub = &spa->spa_uberblock;
1089 vdev_t *vd;
1090 zio_t *zio;
1091 int error;
1092 int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL;
1093
1094 /*
1095 * Normally, we don't want to try too hard to write every label and
1096 * uberblock. If there is a flaky disk, we don't want the rest of the
1097 * sync process to block while we retry. But if we can't write a
1098 * single label out, we should retry with ZIO_FLAG_TRYHARD before
1099 * bailing out and declaring the pool faulted.
1100 */
1101 if (tryhard)
1102 flags |= ZIO_FLAG_TRYHARD;
1103
1104 ASSERT(ub->ub_txg <= txg);
1105
1106 /*
1107 * If this isn't a resync due to I/O errors,
1108 * and nothing changed in this transaction group,
1109 * and the vdev configuration hasn't changed,
1110 * then there's nothing to do.
1111 */
1112 if (ub->ub_txg < txg &&
1113 uberblock_update(ub, spa->spa_root_vdev, txg) == B_FALSE &&
1114 list_is_empty(&spa->spa_config_dirty_list))
1115 return (0);
1116
1117 if (txg > spa_freeze_txg(spa))
1118 return (0);
1119
1120 ASSERT(txg <= spa->spa_final_txg);
1121
1122 /*
1123 * Flush the write cache of every disk that's been written to
1124 * in this transaction group. This ensures that all blocks
1125 * written in this txg will be committed to stable storage
1126 * before any uberblock that references them.
1127 */
1128 zio = zio_root(spa, NULL, NULL, flags);
1129
1130 for (vd = txg_list_head(&spa->spa_vdev_txg_list, TXG_CLEAN(txg)); vd;
1131 vd = txg_list_next(&spa->spa_vdev_txg_list, vd, TXG_CLEAN(txg)))
1132 zio_flush(zio, vd);
1133
1134 (void) zio_wait(zio);
1135
1136 /*
1137 * Sync out the even labels (L0, L2) for every dirty vdev. If the
1138 * system dies in the middle of this process, that's OK: all of the
1139 * even labels that made it to disk will be newer than any uberblock,
1140 * and will therefore be considered invalid. The odd labels (L1, L3),
1141 * which have not yet been touched, will still be valid. We flush
1142 * the new labels to disk to ensure that all even-label updates
1143 * are committed to stable storage before the uberblock update.
1144 */
1145 if ((error = vdev_label_sync_list(spa, 0, txg, flags)) != 0)
1146 return (error);
1147
1148 /*
1149 * Sync the uberblocks to all vdevs in svd[].
1150 * If the system dies in the middle of this step, there are two cases
1151 * to consider, and the on-disk state is consistent either way:
1152 *
1153 * (1) If none of the new uberblocks made it to disk, then the
1154 * previous uberblock will be the newest, and the odd labels
1155 * (which had not yet been touched) will be valid with respect
1156 * to that uberblock.
1157 *
1158 * (2) If one or more new uberblocks made it to disk, then they
1159 * will be the newest, and the even labels (which had all
1160 * been successfully committed) will be valid with respect
1161 * to the new uberblocks.
1162 */
1163 if ((error = vdev_uberblock_sync_list(svd, svdcount, ub, flags)) != 0)
1164 return (error);
1165
1166 /*
1167 * Sync out odd labels for every dirty vdev. If the system dies
1168 * in the middle of this process, the even labels and the new
1169 * uberblocks will suffice to open the pool. The next time
1170 * the pool is opened, the first thing we'll do -- before any
1171 * user data is modified -- is mark every vdev dirty so that
1172 * all labels will be brought up to date. We flush the new labels
1173 * to disk to ensure that all odd-label updates are committed to
1174 * stable storage before the next transaction group begins.
1175 */
1176 return (vdev_label_sync_list(spa, 1, txg, flags));
1177 }
1178