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 /*
23 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Copyright (c) 2012, 2018 by Delphix. All rights reserved.
25 */
26
27 /*
28 * Virtual Device Labels
29 * ---------------------
30 *
31 * The vdev label serves several distinct purposes:
32 *
33 * 1. Uniquely identify this device as part of a ZFS pool and confirm its
34 * identity within the pool.
35 *
36 * 2. Verify that all the devices given in a configuration are present
37 * within the pool.
38 *
39 * 3. Determine the uberblock for the pool.
40 *
41 * 4. In case of an import operation, determine the configuration of the
42 * toplevel vdev of which it is a part.
43 *
44 * 5. If an import operation cannot find all the devices in the pool,
45 * provide enough information to the administrator to determine which
46 * devices are missing.
47 *
48 * It is important to note that while the kernel is responsible for writing the
49 * label, it only consumes the information in the first three cases. The
50 * latter information is only consumed in userland when determining the
51 * configuration to import a pool.
52 *
53 *
54 * Label Organization
55 * ------------------
56 *
57 * Before describing the contents of the label, it's important to understand how
58 * the labels are written and updated with respect to the uberblock.
59 *
60 * When the pool configuration is altered, either because it was newly created
61 * or a device was added, we want to update all the labels such that we can deal
62 * with fatal failure at any point. To this end, each disk has two labels which
63 * are updated before and after the uberblock is synced. Assuming we have
64 * labels and an uberblock with the following transaction groups:
65 *
66 * L1 UB L2
67 * +------+ +------+ +------+
68 * | | | | | |
69 * | t10 | | t10 | | t10 |
70 * | | | | | |
71 * +------+ +------+ +------+
72 *
73 * In this stable state, the labels and the uberblock were all updated within
74 * the same transaction group (10). Each label is mirrored and checksummed, so
75 * that we can detect when we fail partway through writing the label.
76 *
77 * In order to identify which labels are valid, the labels are written in the
78 * following manner:
79 *
80 * 1. For each vdev, update 'L1' to the new label
81 * 2. Update the uberblock
82 * 3. For each vdev, update 'L2' to the new label
83 *
84 * Given arbitrary failure, we can determine the correct label to use based on
85 * the transaction group. If we fail after updating L1 but before updating the
86 * UB, we will notice that L1's transaction group is greater than the uberblock,
87 * so L2 must be valid. If we fail after writing the uberblock but before
88 * writing L2, we will notice that L2's transaction group is less than L1, and
89 * therefore L1 is valid.
90 *
91 * Another added complexity is that not every label is updated when the config
92 * is synced. If we add a single device, we do not want to have to re-write
93 * every label for every device in the pool. This means that both L1 and L2 may
94 * be older than the pool uberblock, because the necessary information is stored
95 * on another vdev.
96 *
97 *
98 * On-disk Format
99 * --------------
100 *
101 * The vdev label consists of two distinct parts, and is wrapped within the
102 * vdev_label_t structure. The label includes 8k of padding to permit legacy
103 * VTOC disk labels, but is otherwise ignored.
104 *
105 * The first half of the label is a packed nvlist which contains pool wide
106 * properties, per-vdev properties, and configuration information. It is
107 * described in more detail below.
108 *
109 * The latter half of the label consists of a redundant array of uberblocks.
110 * These uberblocks are updated whenever a transaction group is committed,
111 * or when the configuration is updated. When a pool is loaded, we scan each
112 * vdev for the 'best' uberblock.
113 *
114 *
115 * Configuration Information
116 * -------------------------
117 *
118 * The nvlist describing the pool and vdev contains the following elements:
119 *
120 * version ZFS on-disk version
121 * name Pool name
122 * state Pool state
123 * txg Transaction group in which this label was written
124 * pool_guid Unique identifier for this pool
125 * vdev_tree An nvlist describing vdev tree.
126 * features_for_read
127 * An nvlist of the features necessary for reading the MOS.
128 *
129 * Each leaf device label also contains the following:
130 *
131 * top_guid Unique ID for top-level vdev in which this is contained
132 * guid Unique ID for the leaf vdev
133 *
134 * The 'vs' configuration follows the format described in 'spa_config.c'.
135 */
136
137 #include <sys/zfs_context.h>
138 #include <sys/spa.h>
139 #include <sys/spa_impl.h>
140 #include <sys/dmu.h>
141 #include <sys/zap.h>
142 #include <sys/vdev.h>
143 #include <sys/vdev_impl.h>
144 #include <sys/uberblock_impl.h>
145 #include <sys/metaslab.h>
146 #include <sys/metaslab_impl.h>
147 #include <sys/zio.h>
148 #include <sys/dsl_scan.h>
149 #include <sys/abd.h>
150 #include <sys/fs/zfs.h>
151 #include <sys/trim_map.h>
152
153 static boolean_t vdev_trim_on_init = B_TRUE;
154 SYSCTL_DECL(_vfs_zfs_vdev);
155 SYSCTL_INT(_vfs_zfs_vdev, OID_AUTO, trim_on_init, CTLFLAG_RWTUN,
156 &vdev_trim_on_init, 0, "Enable/disable full vdev trim on initialisation");
157
158 /*
159 * Basic routines to read and write from a vdev label.
160 * Used throughout the rest of this file.
161 */
162 uint64_t
vdev_label_offset(uint64_t psize,int l,uint64_t offset)163 vdev_label_offset(uint64_t psize, int l, uint64_t offset)
164 {
165 ASSERT(offset < sizeof (vdev_label_t));
166 ASSERT(P2PHASE_TYPED(psize, sizeof (vdev_label_t), uint64_t) == 0);
167
168 return (offset + l * sizeof (vdev_label_t) + (l < VDEV_LABELS / 2 ?
169 0 : psize - VDEV_LABELS * sizeof (vdev_label_t)));
170 }
171
172 /*
173 * Returns back the vdev label associated with the passed in offset.
174 */
175 int
vdev_label_number(uint64_t psize,uint64_t offset)176 vdev_label_number(uint64_t psize, uint64_t offset)
177 {
178 int l;
179
180 if (offset >= psize - VDEV_LABEL_END_SIZE) {
181 offset -= psize - VDEV_LABEL_END_SIZE;
182 offset += (VDEV_LABELS / 2) * sizeof (vdev_label_t);
183 }
184 l = offset / sizeof (vdev_label_t);
185 return (l < VDEV_LABELS ? l : -1);
186 }
187
188 static void
vdev_label_read(zio_t * zio,vdev_t * vd,int l,abd_t * buf,uint64_t offset,uint64_t size,zio_done_func_t * done,void * private,int flags)189 vdev_label_read(zio_t *zio, vdev_t *vd, int l, abd_t *buf, uint64_t offset,
190 uint64_t size, zio_done_func_t *done, void *private, int flags)
191 {
192 ASSERT(spa_config_held(zio->io_spa, SCL_STATE_ALL, RW_WRITER) ==
193 SCL_STATE_ALL);
194 ASSERT(flags & ZIO_FLAG_CONFIG_WRITER);
195
196 zio_nowait(zio_read_phys(zio, vd,
197 vdev_label_offset(vd->vdev_psize, l, offset),
198 size, buf, ZIO_CHECKSUM_LABEL, done, private,
199 ZIO_PRIORITY_SYNC_READ, flags, B_TRUE));
200 }
201
202 static void
vdev_label_write(zio_t * zio,vdev_t * vd,int l,abd_t * buf,uint64_t offset,uint64_t size,zio_done_func_t * done,void * private,int flags)203 vdev_label_write(zio_t *zio, vdev_t *vd, int l, abd_t *buf, uint64_t offset,
204 uint64_t size, zio_done_func_t *done, void *private, int flags)
205 {
206 ASSERT(spa_config_held(zio->io_spa, SCL_ALL, RW_WRITER) == SCL_ALL ||
207 (spa_config_held(zio->io_spa, SCL_CONFIG | SCL_STATE, RW_READER) ==
208 (SCL_CONFIG | SCL_STATE) &&
209 dsl_pool_sync_context(spa_get_dsl(zio->io_spa))));
210 ASSERT(flags & ZIO_FLAG_CONFIG_WRITER);
211
212 zio_nowait(zio_write_phys(zio, vd,
213 vdev_label_offset(vd->vdev_psize, l, offset),
214 size, buf, ZIO_CHECKSUM_LABEL, done, private,
215 ZIO_PRIORITY_SYNC_WRITE, flags, B_TRUE));
216 }
217
218 static void
root_vdev_actions_getprogress(vdev_t * vd,nvlist_t * nvl)219 root_vdev_actions_getprogress(vdev_t *vd, nvlist_t *nvl)
220 {
221 spa_t *spa = vd->vdev_spa;
222
223 if (vd != spa->spa_root_vdev)
224 return;
225
226 /* provide either current or previous scan information */
227 pool_scan_stat_t ps;
228 if (spa_scan_get_stats(spa, &ps) == 0) {
229 fnvlist_add_uint64_array(nvl,
230 ZPOOL_CONFIG_SCAN_STATS, (uint64_t *)&ps,
231 sizeof (pool_scan_stat_t) / sizeof (uint64_t));
232 }
233
234 pool_removal_stat_t prs;
235 if (spa_removal_get_stats(spa, &prs) == 0) {
236 fnvlist_add_uint64_array(nvl,
237 ZPOOL_CONFIG_REMOVAL_STATS, (uint64_t *)&prs,
238 sizeof (prs) / sizeof (uint64_t));
239 }
240
241 pool_checkpoint_stat_t pcs;
242 if (spa_checkpoint_get_stats(spa, &pcs) == 0) {
243 fnvlist_add_uint64_array(nvl,
244 ZPOOL_CONFIG_CHECKPOINT_STATS, (uint64_t *)&pcs,
245 sizeof (pcs) / sizeof (uint64_t));
246 }
247 }
248
249 /*
250 * Generate the nvlist representing this vdev's config.
251 */
252 nvlist_t *
vdev_config_generate(spa_t * spa,vdev_t * vd,boolean_t getstats,vdev_config_flag_t flags)253 vdev_config_generate(spa_t *spa, vdev_t *vd, boolean_t getstats,
254 vdev_config_flag_t flags)
255 {
256 nvlist_t *nv = NULL;
257 vdev_indirect_config_t *vic = &vd->vdev_indirect_config;
258
259 nv = fnvlist_alloc();
260
261 fnvlist_add_string(nv, ZPOOL_CONFIG_TYPE, vd->vdev_ops->vdev_op_type);
262 if (!(flags & (VDEV_CONFIG_SPARE | VDEV_CONFIG_L2CACHE)))
263 fnvlist_add_uint64(nv, ZPOOL_CONFIG_ID, vd->vdev_id);
264 fnvlist_add_uint64(nv, ZPOOL_CONFIG_GUID, vd->vdev_guid);
265
266 if (vd->vdev_path != NULL)
267 fnvlist_add_string(nv, ZPOOL_CONFIG_PATH, vd->vdev_path);
268
269 if (vd->vdev_devid != NULL)
270 fnvlist_add_string(nv, ZPOOL_CONFIG_DEVID, vd->vdev_devid);
271
272 if (vd->vdev_physpath != NULL)
273 fnvlist_add_string(nv, ZPOOL_CONFIG_PHYS_PATH,
274 vd->vdev_physpath);
275
276 if (vd->vdev_fru != NULL)
277 fnvlist_add_string(nv, ZPOOL_CONFIG_FRU, vd->vdev_fru);
278
279 if (vd->vdev_nparity != 0) {
280 ASSERT(strcmp(vd->vdev_ops->vdev_op_type,
281 VDEV_TYPE_RAIDZ) == 0);
282
283 /*
284 * Make sure someone hasn't managed to sneak a fancy new vdev
285 * into a crufty old storage pool.
286 */
287 ASSERT(vd->vdev_nparity == 1 ||
288 (vd->vdev_nparity <= 2 &&
289 spa_version(spa) >= SPA_VERSION_RAIDZ2) ||
290 (vd->vdev_nparity <= 3 &&
291 spa_version(spa) >= SPA_VERSION_RAIDZ3));
292
293 /*
294 * Note that we'll add the nparity tag even on storage pools
295 * that only support a single parity device -- older software
296 * will just ignore it.
297 */
298 fnvlist_add_uint64(nv, ZPOOL_CONFIG_NPARITY, vd->vdev_nparity);
299 }
300
301 if (vd->vdev_wholedisk != -1ULL)
302 fnvlist_add_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK,
303 vd->vdev_wholedisk);
304
305 if (vd->vdev_not_present && !(flags & VDEV_CONFIG_MISSING))
306 fnvlist_add_uint64(nv, ZPOOL_CONFIG_NOT_PRESENT, 1);
307
308 if (vd->vdev_isspare)
309 fnvlist_add_uint64(nv, ZPOOL_CONFIG_IS_SPARE, 1);
310
311 if (!(flags & (VDEV_CONFIG_SPARE | VDEV_CONFIG_L2CACHE)) &&
312 vd == vd->vdev_top) {
313 fnvlist_add_uint64(nv, ZPOOL_CONFIG_METASLAB_ARRAY,
314 vd->vdev_ms_array);
315 fnvlist_add_uint64(nv, ZPOOL_CONFIG_METASLAB_SHIFT,
316 vd->vdev_ms_shift);
317 fnvlist_add_uint64(nv, ZPOOL_CONFIG_ASHIFT, vd->vdev_ashift);
318 fnvlist_add_uint64(nv, ZPOOL_CONFIG_ASIZE,
319 vd->vdev_asize);
320 fnvlist_add_uint64(nv, ZPOOL_CONFIG_IS_LOG, vd->vdev_islog);
321 if (vd->vdev_removing) {
322 fnvlist_add_uint64(nv, ZPOOL_CONFIG_REMOVING,
323 vd->vdev_removing);
324 }
325 }
326
327 if (vd->vdev_dtl_sm != NULL) {
328 fnvlist_add_uint64(nv, ZPOOL_CONFIG_DTL,
329 space_map_object(vd->vdev_dtl_sm));
330 }
331
332 if (vic->vic_mapping_object != 0) {
333 fnvlist_add_uint64(nv, ZPOOL_CONFIG_INDIRECT_OBJECT,
334 vic->vic_mapping_object);
335 }
336
337 if (vic->vic_births_object != 0) {
338 fnvlist_add_uint64(nv, ZPOOL_CONFIG_INDIRECT_BIRTHS,
339 vic->vic_births_object);
340 }
341
342 if (vic->vic_prev_indirect_vdev != UINT64_MAX) {
343 fnvlist_add_uint64(nv, ZPOOL_CONFIG_PREV_INDIRECT_VDEV,
344 vic->vic_prev_indirect_vdev);
345 }
346
347 if (vd->vdev_crtxg)
348 fnvlist_add_uint64(nv, ZPOOL_CONFIG_CREATE_TXG, vd->vdev_crtxg);
349
350 if (flags & VDEV_CONFIG_MOS) {
351 if (vd->vdev_leaf_zap != 0) {
352 ASSERT(vd->vdev_ops->vdev_op_leaf);
353 fnvlist_add_uint64(nv, ZPOOL_CONFIG_VDEV_LEAF_ZAP,
354 vd->vdev_leaf_zap);
355 }
356
357 if (vd->vdev_top_zap != 0) {
358 ASSERT(vd == vd->vdev_top);
359 fnvlist_add_uint64(nv, ZPOOL_CONFIG_VDEV_TOP_ZAP,
360 vd->vdev_top_zap);
361 }
362 }
363
364 if (getstats) {
365 vdev_stat_t vs;
366
367 vdev_get_stats(vd, &vs);
368 fnvlist_add_uint64_array(nv, ZPOOL_CONFIG_VDEV_STATS,
369 (uint64_t *)&vs, sizeof (vs) / sizeof (uint64_t));
370
371 root_vdev_actions_getprogress(vd, nv);
372
373 /*
374 * Note: this can be called from open context
375 * (spa_get_stats()), so we need the rwlock to prevent
376 * the mapping from being changed by condensing.
377 */
378 rw_enter(&vd->vdev_indirect_rwlock, RW_READER);
379 if (vd->vdev_indirect_mapping != NULL) {
380 ASSERT(vd->vdev_indirect_births != NULL);
381 vdev_indirect_mapping_t *vim =
382 vd->vdev_indirect_mapping;
383 fnvlist_add_uint64(nv, ZPOOL_CONFIG_INDIRECT_SIZE,
384 vdev_indirect_mapping_size(vim));
385 }
386 rw_exit(&vd->vdev_indirect_rwlock);
387 if (vd->vdev_mg != NULL &&
388 vd->vdev_mg->mg_fragmentation != ZFS_FRAG_INVALID) {
389 /*
390 * Compute approximately how much memory would be used
391 * for the indirect mapping if this device were to
392 * be removed.
393 *
394 * Note: If the frag metric is invalid, then not
395 * enough metaslabs have been converted to have
396 * histograms.
397 */
398 uint64_t seg_count = 0;
399 uint64_t to_alloc = vd->vdev_stat.vs_alloc;
400
401 /*
402 * There are the same number of allocated segments
403 * as free segments, so we will have at least one
404 * entry per free segment. However, small free
405 * segments (smaller than vdev_removal_max_span)
406 * will be combined with adjacent allocated segments
407 * as a single mapping.
408 */
409 for (int i = 0; i < RANGE_TREE_HISTOGRAM_SIZE; i++) {
410 if (1ULL << (i + 1) < vdev_removal_max_span) {
411 to_alloc +=
412 vd->vdev_mg->mg_histogram[i] <<
413 i + 1;
414 } else {
415 seg_count +=
416 vd->vdev_mg->mg_histogram[i];
417 }
418 }
419
420 /*
421 * The maximum length of a mapping is
422 * zfs_remove_max_segment, so we need at least one entry
423 * per zfs_remove_max_segment of allocated data.
424 */
425 seg_count += to_alloc / zfs_remove_max_segment;
426
427 fnvlist_add_uint64(nv, ZPOOL_CONFIG_INDIRECT_SIZE,
428 seg_count *
429 sizeof (vdev_indirect_mapping_entry_phys_t));
430 }
431 }
432
433 if (!vd->vdev_ops->vdev_op_leaf) {
434 nvlist_t **child;
435 int c, idx;
436
437 ASSERT(!vd->vdev_ishole);
438
439 child = kmem_alloc(vd->vdev_children * sizeof (nvlist_t *),
440 KM_SLEEP);
441
442 for (c = 0, idx = 0; c < vd->vdev_children; c++) {
443 vdev_t *cvd = vd->vdev_child[c];
444
445 /*
446 * If we're generating an nvlist of removing
447 * vdevs then skip over any device which is
448 * not being removed.
449 */
450 if ((flags & VDEV_CONFIG_REMOVING) &&
451 !cvd->vdev_removing)
452 continue;
453
454 child[idx++] = vdev_config_generate(spa, cvd,
455 getstats, flags);
456 }
457
458 if (idx) {
459 fnvlist_add_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
460 child, idx);
461 }
462
463 for (c = 0; c < idx; c++)
464 nvlist_free(child[c]);
465
466 kmem_free(child, vd->vdev_children * sizeof (nvlist_t *));
467
468 } else {
469 const char *aux = NULL;
470
471 if (vd->vdev_offline && !vd->vdev_tmpoffline)
472 fnvlist_add_uint64(nv, ZPOOL_CONFIG_OFFLINE, B_TRUE);
473 if (vd->vdev_resilver_txg != 0)
474 fnvlist_add_uint64(nv, ZPOOL_CONFIG_RESILVER_TXG,
475 vd->vdev_resilver_txg);
476 if (vd->vdev_faulted)
477 fnvlist_add_uint64(nv, ZPOOL_CONFIG_FAULTED, B_TRUE);
478 if (vd->vdev_degraded)
479 fnvlist_add_uint64(nv, ZPOOL_CONFIG_DEGRADED, B_TRUE);
480 if (vd->vdev_removed)
481 fnvlist_add_uint64(nv, ZPOOL_CONFIG_REMOVED, B_TRUE);
482 if (vd->vdev_unspare)
483 fnvlist_add_uint64(nv, ZPOOL_CONFIG_UNSPARE, B_TRUE);
484 if (vd->vdev_ishole)
485 fnvlist_add_uint64(nv, ZPOOL_CONFIG_IS_HOLE, B_TRUE);
486
487 switch (vd->vdev_stat.vs_aux) {
488 case VDEV_AUX_ERR_EXCEEDED:
489 aux = "err_exceeded";
490 break;
491
492 case VDEV_AUX_EXTERNAL:
493 aux = "external";
494 break;
495 }
496
497 if (aux != NULL)
498 fnvlist_add_string(nv, ZPOOL_CONFIG_AUX_STATE, aux);
499
500 if (vd->vdev_splitting && vd->vdev_orig_guid != 0LL) {
501 fnvlist_add_uint64(nv, ZPOOL_CONFIG_ORIG_GUID,
502 vd->vdev_orig_guid);
503 }
504 }
505
506 return (nv);
507 }
508
509 /*
510 * Generate a view of the top-level vdevs. If we currently have holes
511 * in the namespace, then generate an array which contains a list of holey
512 * vdevs. Additionally, add the number of top-level children that currently
513 * exist.
514 */
515 void
vdev_top_config_generate(spa_t * spa,nvlist_t * config)516 vdev_top_config_generate(spa_t *spa, nvlist_t *config)
517 {
518 vdev_t *rvd = spa->spa_root_vdev;
519 uint64_t *array;
520 uint_t c, idx;
521
522 array = kmem_alloc(rvd->vdev_children * sizeof (uint64_t), KM_SLEEP);
523
524 for (c = 0, idx = 0; c < rvd->vdev_children; c++) {
525 vdev_t *tvd = rvd->vdev_child[c];
526
527 if (tvd->vdev_ishole) {
528 array[idx++] = c;
529 }
530 }
531
532 if (idx) {
533 VERIFY(nvlist_add_uint64_array(config, ZPOOL_CONFIG_HOLE_ARRAY,
534 array, idx) == 0);
535 }
536
537 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VDEV_CHILDREN,
538 rvd->vdev_children) == 0);
539
540 kmem_free(array, rvd->vdev_children * sizeof (uint64_t));
541 }
542
543 /*
544 * Returns the configuration from the label of the given vdev. For vdevs
545 * which don't have a txg value stored on their label (i.e. spares/cache)
546 * or have not been completely initialized (txg = 0) just return
547 * the configuration from the first valid label we find. Otherwise,
548 * find the most up-to-date label that does not exceed the specified
549 * 'txg' value.
550 */
551 nvlist_t *
vdev_label_read_config(vdev_t * vd,uint64_t txg)552 vdev_label_read_config(vdev_t *vd, uint64_t txg)
553 {
554 spa_t *spa = vd->vdev_spa;
555 nvlist_t *config = NULL;
556 vdev_phys_t *vp;
557 abd_t *vp_abd;
558 zio_t *zio;
559 uint64_t best_txg = 0;
560 uint64_t label_txg = 0;
561 int error = 0;
562 int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL |
563 ZIO_FLAG_SPECULATIVE;
564
565 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
566
567 if (!vdev_readable(vd))
568 return (NULL);
569
570 vp_abd = abd_alloc_linear(sizeof (vdev_phys_t), B_TRUE);
571 vp = abd_to_buf(vp_abd);
572
573 retry:
574 for (int l = 0; l < VDEV_LABELS; l++) {
575 nvlist_t *label = NULL;
576
577 zio = zio_root(spa, NULL, NULL, flags);
578
579 vdev_label_read(zio, vd, l, vp_abd,
580 offsetof(vdev_label_t, vl_vdev_phys),
581 sizeof (vdev_phys_t), NULL, NULL, flags);
582
583 if (zio_wait(zio) == 0 &&
584 nvlist_unpack(vp->vp_nvlist, sizeof (vp->vp_nvlist),
585 &label, 0) == 0) {
586 /*
587 * Auxiliary vdevs won't have txg values in their
588 * labels and newly added vdevs may not have been
589 * completely initialized so just return the
590 * configuration from the first valid label we
591 * encounter.
592 */
593 error = nvlist_lookup_uint64(label,
594 ZPOOL_CONFIG_POOL_TXG, &label_txg);
595 if ((error || label_txg == 0) && !config) {
596 config = label;
597 break;
598 } else if (label_txg <= txg && label_txg > best_txg) {
599 best_txg = label_txg;
600 nvlist_free(config);
601 config = fnvlist_dup(label);
602 }
603 }
604
605 if (label != NULL) {
606 nvlist_free(label);
607 label = NULL;
608 }
609 }
610
611 if (config == NULL && !(flags & ZIO_FLAG_TRYHARD)) {
612 flags |= ZIO_FLAG_TRYHARD;
613 goto retry;
614 }
615
616 /*
617 * We found a valid label but it didn't pass txg restrictions.
618 */
619 if (config == NULL && label_txg != 0) {
620 vdev_dbgmsg(vd, "label discarded as txg is too large "
621 "(%llu > %llu)", (u_longlong_t)label_txg,
622 (u_longlong_t)txg);
623 }
624
625 abd_free(vp_abd);
626
627 return (config);
628 }
629
630 /*
631 * Determine if a device is in use. The 'spare_guid' parameter will be filled
632 * in with the device guid if this spare is active elsewhere on the system.
633 */
634 static boolean_t
vdev_inuse(vdev_t * vd,uint64_t crtxg,vdev_labeltype_t reason,uint64_t * spare_guid,uint64_t * l2cache_guid)635 vdev_inuse(vdev_t *vd, uint64_t crtxg, vdev_labeltype_t reason,
636 uint64_t *spare_guid, uint64_t *l2cache_guid)
637 {
638 spa_t *spa = vd->vdev_spa;
639 uint64_t state, pool_guid, device_guid, txg, spare_pool;
640 uint64_t vdtxg = 0;
641 nvlist_t *label;
642
643 if (spare_guid)
644 *spare_guid = 0ULL;
645 if (l2cache_guid)
646 *l2cache_guid = 0ULL;
647
648 /*
649 * Read the label, if any, and perform some basic sanity checks.
650 */
651 if ((label = vdev_label_read_config(vd, -1ULL)) == NULL)
652 return (B_FALSE);
653
654 (void) nvlist_lookup_uint64(label, ZPOOL_CONFIG_CREATE_TXG,
655 &vdtxg);
656
657 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE,
658 &state) != 0 ||
659 nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID,
660 &device_guid) != 0) {
661 nvlist_free(label);
662 return (B_FALSE);
663 }
664
665 if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
666 (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_GUID,
667 &pool_guid) != 0 ||
668 nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG,
669 &txg) != 0)) {
670 nvlist_free(label);
671 return (B_FALSE);
672 }
673
674 nvlist_free(label);
675
676 /*
677 * Check to see if this device indeed belongs to the pool it claims to
678 * be a part of. The only way this is allowed is if the device is a hot
679 * spare (which we check for later on).
680 */
681 if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
682 !spa_guid_exists(pool_guid, device_guid) &&
683 !spa_spare_exists(device_guid, NULL, NULL) &&
684 !spa_l2cache_exists(device_guid, NULL))
685 return (B_FALSE);
686
687 /*
688 * If the transaction group is zero, then this an initialized (but
689 * unused) label. This is only an error if the create transaction
690 * on-disk is the same as the one we're using now, in which case the
691 * user has attempted to add the same vdev multiple times in the same
692 * transaction.
693 */
694 if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
695 txg == 0 && vdtxg == crtxg)
696 return (B_TRUE);
697
698 /*
699 * Check to see if this is a spare device. We do an explicit check for
700 * spa_has_spare() here because it may be on our pending list of spares
701 * to add. We also check if it is an l2cache device.
702 */
703 if (spa_spare_exists(device_guid, &spare_pool, NULL) ||
704 spa_has_spare(spa, device_guid)) {
705 if (spare_guid)
706 *spare_guid = device_guid;
707
708 switch (reason) {
709 case VDEV_LABEL_CREATE:
710 case VDEV_LABEL_L2CACHE:
711 return (B_TRUE);
712
713 case VDEV_LABEL_REPLACE:
714 return (!spa_has_spare(spa, device_guid) ||
715 spare_pool != 0ULL);
716
717 case VDEV_LABEL_SPARE:
718 return (spa_has_spare(spa, device_guid));
719 }
720 }
721
722 /*
723 * Check to see if this is an l2cache device.
724 */
725 if (spa_l2cache_exists(device_guid, NULL))
726 return (B_TRUE);
727
728 /*
729 * We can't rely on a pool's state if it's been imported
730 * read-only. Instead we look to see if the pools is marked
731 * read-only in the namespace and set the state to active.
732 */
733 if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
734 (spa = spa_by_guid(pool_guid, device_guid)) != NULL &&
735 spa_mode(spa) == FREAD)
736 state = POOL_STATE_ACTIVE;
737
738 /*
739 * If the device is marked ACTIVE, then this device is in use by another
740 * pool on the system.
741 */
742 return (state == POOL_STATE_ACTIVE);
743 }
744
745 /*
746 * Initialize a vdev label. We check to make sure each leaf device is not in
747 * use, and writable. We put down an initial label which we will later
748 * overwrite with a complete label. Note that it's important to do this
749 * sequentially, not in parallel, so that we catch cases of multiple use of the
750 * same leaf vdev in the vdev we're creating -- e.g. mirroring a disk with
751 * itself.
752 */
753 int
vdev_label_init(vdev_t * vd,uint64_t crtxg,vdev_labeltype_t reason)754 vdev_label_init(vdev_t *vd, uint64_t crtxg, vdev_labeltype_t reason)
755 {
756 spa_t *spa = vd->vdev_spa;
757 nvlist_t *label;
758 vdev_phys_t *vp;
759 abd_t *vp_abd;
760 abd_t *pad2;
761 uberblock_t *ub;
762 abd_t *ub_abd;
763 zio_t *zio;
764 char *buf;
765 size_t buflen;
766 int error;
767 uint64_t spare_guid, l2cache_guid;
768 int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL;
769
770 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
771
772 for (int c = 0; c < vd->vdev_children; c++)
773 if ((error = vdev_label_init(vd->vdev_child[c],
774 crtxg, reason)) != 0)
775 return (error);
776
777 /* Track the creation time for this vdev */
778 vd->vdev_crtxg = crtxg;
779
780 if (!vd->vdev_ops->vdev_op_leaf || !spa_writeable(spa))
781 return (0);
782
783 /*
784 * Dead vdevs cannot be initialized.
785 */
786 if (vdev_is_dead(vd))
787 return (SET_ERROR(EIO));
788
789 /*
790 * Determine if the vdev is in use.
791 */
792 if (reason != VDEV_LABEL_REMOVE && reason != VDEV_LABEL_SPLIT &&
793 vdev_inuse(vd, crtxg, reason, &spare_guid, &l2cache_guid))
794 return (SET_ERROR(EBUSY));
795
796 /*
797 * If this is a request to add or replace a spare or l2cache device
798 * that is in use elsewhere on the system, then we must update the
799 * guid (which was initialized to a random value) to reflect the
800 * actual GUID (which is shared between multiple pools).
801 */
802 if (reason != VDEV_LABEL_REMOVE && reason != VDEV_LABEL_L2CACHE &&
803 spare_guid != 0ULL) {
804 uint64_t guid_delta = spare_guid - vd->vdev_guid;
805
806 vd->vdev_guid += guid_delta;
807
808 for (vdev_t *pvd = vd; pvd != NULL; pvd = pvd->vdev_parent)
809 pvd->vdev_guid_sum += guid_delta;
810
811 /*
812 * If this is a replacement, then we want to fallthrough to the
813 * rest of the code. If we're adding a spare, then it's already
814 * labeled appropriately and we can just return.
815 */
816 if (reason == VDEV_LABEL_SPARE)
817 return (0);
818 ASSERT(reason == VDEV_LABEL_REPLACE ||
819 reason == VDEV_LABEL_SPLIT);
820 }
821
822 if (reason != VDEV_LABEL_REMOVE && reason != VDEV_LABEL_SPARE &&
823 l2cache_guid != 0ULL) {
824 uint64_t guid_delta = l2cache_guid - vd->vdev_guid;
825
826 vd->vdev_guid += guid_delta;
827
828 for (vdev_t *pvd = vd; pvd != NULL; pvd = pvd->vdev_parent)
829 pvd->vdev_guid_sum += guid_delta;
830
831 /*
832 * If this is a replacement, then we want to fallthrough to the
833 * rest of the code. If we're adding an l2cache, then it's
834 * already labeled appropriately and we can just return.
835 */
836 if (reason == VDEV_LABEL_L2CACHE)
837 return (0);
838 ASSERT(reason == VDEV_LABEL_REPLACE);
839 }
840
841 /*
842 * TRIM the whole thing, excluding the blank space and boot header
843 * as specified by ZFS On-Disk Specification (section 1.3), so that
844 * we start with a clean slate.
845 * It's just an optimization, so we don't care if it fails.
846 * Don't TRIM if removing so that we don't interfere with zpool
847 * disaster recovery.
848 */
849 if (zfs_trim_enabled && vdev_trim_on_init && !vd->vdev_notrim &&
850 (reason == VDEV_LABEL_CREATE || reason == VDEV_LABEL_SPARE ||
851 reason == VDEV_LABEL_L2CACHE))
852 zio_wait(zio_trim(NULL, spa, vd, VDEV_SKIP_SIZE,
853 vd->vdev_psize - VDEV_SKIP_SIZE));
854
855 /*
856 * Initialize its label.
857 */
858 vp_abd = abd_alloc_linear(sizeof (vdev_phys_t), B_TRUE);
859 abd_zero(vp_abd, sizeof (vdev_phys_t));
860 vp = abd_to_buf(vp_abd);
861
862 /*
863 * Generate a label describing the pool and our top-level vdev.
864 * We mark it as being from txg 0 to indicate that it's not
865 * really part of an active pool just yet. The labels will
866 * be written again with a meaningful txg by spa_sync().
867 */
868 if (reason == VDEV_LABEL_SPARE ||
869 (reason == VDEV_LABEL_REMOVE && vd->vdev_isspare)) {
870 /*
871 * For inactive hot spares, we generate a special label that
872 * identifies as a mutually shared hot spare. We write the
873 * label if we are adding a hot spare, or if we are removing an
874 * active hot spare (in which case we want to revert the
875 * labels).
876 */
877 VERIFY(nvlist_alloc(&label, NV_UNIQUE_NAME, KM_SLEEP) == 0);
878
879 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_VERSION,
880 spa_version(spa)) == 0);
881 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_POOL_STATE,
882 POOL_STATE_SPARE) == 0);
883 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_GUID,
884 vd->vdev_guid) == 0);
885 } else if (reason == VDEV_LABEL_L2CACHE ||
886 (reason == VDEV_LABEL_REMOVE && vd->vdev_isl2cache)) {
887 /*
888 * For level 2 ARC devices, add a special label.
889 */
890 VERIFY(nvlist_alloc(&label, NV_UNIQUE_NAME, KM_SLEEP) == 0);
891
892 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_VERSION,
893 spa_version(spa)) == 0);
894 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_POOL_STATE,
895 POOL_STATE_L2CACHE) == 0);
896 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_GUID,
897 vd->vdev_guid) == 0);
898 } else {
899 uint64_t txg = 0ULL;
900
901 if (reason == VDEV_LABEL_SPLIT)
902 txg = spa->spa_uberblock.ub_txg;
903 label = spa_config_generate(spa, vd, txg, B_FALSE);
904
905 /*
906 * Add our creation time. This allows us to detect multiple
907 * vdev uses as described above, and automatically expires if we
908 * fail.
909 */
910 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_CREATE_TXG,
911 crtxg) == 0);
912 }
913
914 buf = vp->vp_nvlist;
915 buflen = sizeof (vp->vp_nvlist);
916
917 error = nvlist_pack(label, &buf, &buflen, NV_ENCODE_XDR, KM_SLEEP);
918 if (error != 0) {
919 nvlist_free(label);
920 abd_free(vp_abd);
921 /* EFAULT means nvlist_pack ran out of room */
922 return (error == EFAULT ? ENAMETOOLONG : EINVAL);
923 }
924
925 /*
926 * Initialize uberblock template.
927 */
928 ub_abd = abd_alloc_linear(VDEV_UBERBLOCK_RING, B_TRUE);
929 abd_zero(ub_abd, VDEV_UBERBLOCK_RING);
930 abd_copy_from_buf(ub_abd, &spa->spa_uberblock, sizeof (uberblock_t));
931 ub = abd_to_buf(ub_abd);
932 ub->ub_txg = 0;
933
934 /* Initialize the 2nd padding area. */
935 pad2 = abd_alloc_for_io(VDEV_PAD_SIZE, B_TRUE);
936 abd_zero(pad2, VDEV_PAD_SIZE);
937
938 /*
939 * Write everything in parallel.
940 */
941 retry:
942 zio = zio_root(spa, NULL, NULL, flags);
943
944 for (int l = 0; l < VDEV_LABELS; l++) {
945
946 vdev_label_write(zio, vd, l, vp_abd,
947 offsetof(vdev_label_t, vl_vdev_phys),
948 sizeof (vdev_phys_t), NULL, NULL, flags);
949
950 /*
951 * Skip the 1st padding area.
952 * Zero out the 2nd padding area where it might have
953 * left over data from previous filesystem format.
954 */
955 vdev_label_write(zio, vd, l, pad2,
956 offsetof(vdev_label_t, vl_pad2),
957 VDEV_PAD_SIZE, NULL, NULL, flags);
958
959 vdev_label_write(zio, vd, l, ub_abd,
960 offsetof(vdev_label_t, vl_uberblock),
961 VDEV_UBERBLOCK_RING, NULL, NULL, flags);
962 }
963
964 error = zio_wait(zio);
965
966 if (error != 0 && !(flags & ZIO_FLAG_TRYHARD)) {
967 flags |= ZIO_FLAG_TRYHARD;
968 goto retry;
969 }
970
971 nvlist_free(label);
972 abd_free(pad2);
973 abd_free(ub_abd);
974 abd_free(vp_abd);
975
976 /*
977 * If this vdev hasn't been previously identified as a spare, then we
978 * mark it as such only if a) we are labeling it as a spare, or b) it
979 * exists as a spare elsewhere in the system. Do the same for
980 * level 2 ARC devices.
981 */
982 if (error == 0 && !vd->vdev_isspare &&
983 (reason == VDEV_LABEL_SPARE ||
984 spa_spare_exists(vd->vdev_guid, NULL, NULL)))
985 spa_spare_add(vd);
986
987 if (error == 0 && !vd->vdev_isl2cache &&
988 (reason == VDEV_LABEL_L2CACHE ||
989 spa_l2cache_exists(vd->vdev_guid, NULL)))
990 spa_l2cache_add(vd);
991
992 return (error);
993 }
994
995 int
vdev_label_write_pad2(vdev_t * vd,const char * buf,size_t size)996 vdev_label_write_pad2(vdev_t *vd, const char *buf, size_t size)
997 {
998 spa_t *spa = vd->vdev_spa;
999 zio_t *zio;
1000 abd_t *pad2;
1001 int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL;
1002 int error;
1003
1004 if (size > VDEV_PAD_SIZE)
1005 return (EINVAL);
1006
1007 if (!vd->vdev_ops->vdev_op_leaf)
1008 return (ENODEV);
1009 if (vdev_is_dead(vd))
1010 return (ENXIO);
1011
1012 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1013
1014 pad2 = abd_alloc_for_io(VDEV_PAD_SIZE, B_TRUE);
1015 abd_zero(pad2, VDEV_PAD_SIZE);
1016 abd_copy_from_buf(pad2, buf, size);
1017
1018 retry:
1019 zio = zio_root(spa, NULL, NULL, flags);
1020 vdev_label_write(zio, vd, 0, pad2,
1021 offsetof(vdev_label_t, vl_pad2),
1022 VDEV_PAD_SIZE, NULL, NULL, flags);
1023 error = zio_wait(zio);
1024 if (error != 0 && !(flags & ZIO_FLAG_TRYHARD)) {
1025 flags |= ZIO_FLAG_TRYHARD;
1026 goto retry;
1027 }
1028
1029 abd_free(pad2);
1030 return (error);
1031 }
1032
1033 /*
1034 * ==========================================================================
1035 * uberblock load/sync
1036 * ==========================================================================
1037 */
1038
1039 /*
1040 * Consider the following situation: txg is safely synced to disk. We've
1041 * written the first uberblock for txg + 1, and then we lose power. When we
1042 * come back up, we fail to see the uberblock for txg + 1 because, say,
1043 * it was on a mirrored device and the replica to which we wrote txg + 1
1044 * is now offline. If we then make some changes and sync txg + 1, and then
1045 * the missing replica comes back, then for a few seconds we'll have two
1046 * conflicting uberblocks on disk with the same txg. The solution is simple:
1047 * among uberblocks with equal txg, choose the one with the latest timestamp.
1048 */
1049 static int
vdev_uberblock_compare(const uberblock_t * ub1,const uberblock_t * ub2)1050 vdev_uberblock_compare(const uberblock_t *ub1, const uberblock_t *ub2)
1051 {
1052 int cmp = AVL_CMP(ub1->ub_txg, ub2->ub_txg);
1053 if (likely(cmp))
1054 return (cmp);
1055
1056 return (AVL_CMP(ub1->ub_timestamp, ub2->ub_timestamp));
1057 }
1058
1059 struct ubl_cbdata {
1060 uberblock_t *ubl_ubbest; /* Best uberblock */
1061 vdev_t *ubl_vd; /* vdev associated with the above */
1062 };
1063
1064 static void
vdev_uberblock_load_done(zio_t * zio)1065 vdev_uberblock_load_done(zio_t *zio)
1066 {
1067 vdev_t *vd = zio->io_vd;
1068 spa_t *spa = zio->io_spa;
1069 zio_t *rio = zio->io_private;
1070 uberblock_t *ub = abd_to_buf(zio->io_abd);
1071 struct ubl_cbdata *cbp = rio->io_private;
1072
1073 ASSERT3U(zio->io_size, ==, VDEV_UBERBLOCK_SIZE(vd));
1074
1075 if (zio->io_error == 0 && uberblock_verify(ub) == 0) {
1076 mutex_enter(&rio->io_lock);
1077 if (ub->ub_txg <= spa->spa_load_max_txg &&
1078 vdev_uberblock_compare(ub, cbp->ubl_ubbest) > 0) {
1079 /*
1080 * Keep track of the vdev in which this uberblock
1081 * was found. We will use this information later
1082 * to obtain the config nvlist associated with
1083 * this uberblock.
1084 */
1085 *cbp->ubl_ubbest = *ub;
1086 cbp->ubl_vd = vd;
1087 }
1088 mutex_exit(&rio->io_lock);
1089 }
1090
1091 abd_free(zio->io_abd);
1092 }
1093
1094 static void
vdev_uberblock_load_impl(zio_t * zio,vdev_t * vd,int flags,struct ubl_cbdata * cbp)1095 vdev_uberblock_load_impl(zio_t *zio, vdev_t *vd, int flags,
1096 struct ubl_cbdata *cbp)
1097 {
1098 for (int c = 0; c < vd->vdev_children; c++)
1099 vdev_uberblock_load_impl(zio, vd->vdev_child[c], flags, cbp);
1100
1101 if (vd->vdev_ops->vdev_op_leaf && vdev_readable(vd)) {
1102 for (int l = 0; l < VDEV_LABELS; l++) {
1103 for (int n = 0; n < VDEV_UBERBLOCK_COUNT(vd); n++) {
1104 vdev_label_read(zio, vd, l,
1105 abd_alloc_linear(VDEV_UBERBLOCK_SIZE(vd),
1106 B_TRUE), VDEV_UBERBLOCK_OFFSET(vd, n),
1107 VDEV_UBERBLOCK_SIZE(vd),
1108 vdev_uberblock_load_done, zio, flags);
1109 }
1110 }
1111 }
1112 }
1113
1114 /*
1115 * Reads the 'best' uberblock from disk along with its associated
1116 * configuration. First, we read the uberblock array of each label of each
1117 * vdev, keeping track of the uberblock with the highest txg in each array.
1118 * Then, we read the configuration from the same vdev as the best uberblock.
1119 */
1120 void
vdev_uberblock_load(vdev_t * rvd,uberblock_t * ub,nvlist_t ** config)1121 vdev_uberblock_load(vdev_t *rvd, uberblock_t *ub, nvlist_t **config)
1122 {
1123 zio_t *zio;
1124 spa_t *spa = rvd->vdev_spa;
1125 struct ubl_cbdata cb;
1126 int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL |
1127 ZIO_FLAG_SPECULATIVE | ZIO_FLAG_TRYHARD;
1128
1129 ASSERT(ub);
1130 ASSERT(config);
1131
1132 bzero(ub, sizeof (uberblock_t));
1133 *config = NULL;
1134
1135 cb.ubl_ubbest = ub;
1136 cb.ubl_vd = NULL;
1137
1138 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1139 zio = zio_root(spa, NULL, &cb, flags);
1140 vdev_uberblock_load_impl(zio, rvd, flags, &cb);
1141 (void) zio_wait(zio);
1142
1143 /*
1144 * It's possible that the best uberblock was discovered on a label
1145 * that has a configuration which was written in a future txg.
1146 * Search all labels on this vdev to find the configuration that
1147 * matches the txg for our uberblock.
1148 */
1149 if (cb.ubl_vd != NULL) {
1150 vdev_dbgmsg(cb.ubl_vd, "best uberblock found for spa %s. "
1151 "txg %llu", spa->spa_name, (u_longlong_t)ub->ub_txg);
1152
1153 *config = vdev_label_read_config(cb.ubl_vd, ub->ub_txg);
1154 if (*config == NULL && spa->spa_extreme_rewind) {
1155 vdev_dbgmsg(cb.ubl_vd, "failed to read label config. "
1156 "Trying again without txg restrictions.");
1157 *config = vdev_label_read_config(cb.ubl_vd, UINT64_MAX);
1158 }
1159 if (*config == NULL) {
1160 vdev_dbgmsg(cb.ubl_vd, "failed to read label config");
1161 }
1162 }
1163 spa_config_exit(spa, SCL_ALL, FTAG);
1164 }
1165
1166 /*
1167 * On success, increment root zio's count of good writes.
1168 * We only get credit for writes to known-visible vdevs; see spa_vdev_add().
1169 */
1170 static void
vdev_uberblock_sync_done(zio_t * zio)1171 vdev_uberblock_sync_done(zio_t *zio)
1172 {
1173 uint64_t *good_writes = zio->io_private;
1174
1175 if (zio->io_error == 0 && zio->io_vd->vdev_top->vdev_ms_array != 0)
1176 atomic_inc_64(good_writes);
1177 }
1178
1179 /*
1180 * Write the uberblock to all labels of all leaves of the specified vdev.
1181 */
1182 static void
vdev_uberblock_sync(zio_t * zio,uint64_t * good_writes,uberblock_t * ub,vdev_t * vd,int flags)1183 vdev_uberblock_sync(zio_t *zio, uint64_t *good_writes,
1184 uberblock_t *ub, vdev_t *vd, int flags)
1185 {
1186 for (uint64_t c = 0; c < vd->vdev_children; c++) {
1187 vdev_uberblock_sync(zio, good_writes,
1188 ub, vd->vdev_child[c], flags);
1189 }
1190
1191 if (!vd->vdev_ops->vdev_op_leaf)
1192 return;
1193
1194 if (!vdev_writeable(vd))
1195 return;
1196
1197 int n = ub->ub_txg & (VDEV_UBERBLOCK_COUNT(vd) - 1);
1198
1199 /* Copy the uberblock_t into the ABD */
1200 abd_t *ub_abd = abd_alloc_for_io(VDEV_UBERBLOCK_SIZE(vd), B_TRUE);
1201 abd_zero(ub_abd, VDEV_UBERBLOCK_SIZE(vd));
1202 abd_copy_from_buf(ub_abd, ub, sizeof (uberblock_t));
1203
1204 for (int l = 0; l < VDEV_LABELS; l++)
1205 vdev_label_write(zio, vd, l, ub_abd,
1206 VDEV_UBERBLOCK_OFFSET(vd, n), VDEV_UBERBLOCK_SIZE(vd),
1207 vdev_uberblock_sync_done, good_writes,
1208 flags | ZIO_FLAG_DONT_PROPAGATE);
1209
1210 abd_free(ub_abd);
1211 }
1212
1213 /* Sync the uberblocks to all vdevs in svd[] */
1214 int
vdev_uberblock_sync_list(vdev_t ** svd,int svdcount,uberblock_t * ub,int flags)1215 vdev_uberblock_sync_list(vdev_t **svd, int svdcount, uberblock_t *ub, int flags)
1216 {
1217 spa_t *spa = svd[0]->vdev_spa;
1218 zio_t *zio;
1219 uint64_t good_writes = 0;
1220
1221 zio = zio_root(spa, NULL, NULL, flags);
1222
1223 for (int v = 0; v < svdcount; v++)
1224 vdev_uberblock_sync(zio, &good_writes, ub, svd[v], flags);
1225
1226 (void) zio_wait(zio);
1227
1228 /*
1229 * Flush the uberblocks to disk. This ensures that the odd labels
1230 * are no longer needed (because the new uberblocks and the even
1231 * labels are safely on disk), so it is safe to overwrite them.
1232 */
1233 zio = zio_root(spa, NULL, NULL, flags);
1234
1235 for (int v = 0; v < svdcount; v++) {
1236 if (vdev_writeable(svd[v])) {
1237 zio_flush(zio, svd[v]);
1238 }
1239 }
1240
1241 (void) zio_wait(zio);
1242
1243 return (good_writes >= 1 ? 0 : EIO);
1244 }
1245
1246 /*
1247 * On success, increment the count of good writes for our top-level vdev.
1248 */
1249 static void
vdev_label_sync_done(zio_t * zio)1250 vdev_label_sync_done(zio_t *zio)
1251 {
1252 uint64_t *good_writes = zio->io_private;
1253
1254 if (zio->io_error == 0)
1255 atomic_inc_64(good_writes);
1256 }
1257
1258 /*
1259 * If there weren't enough good writes, indicate failure to the parent.
1260 */
1261 static void
vdev_label_sync_top_done(zio_t * zio)1262 vdev_label_sync_top_done(zio_t *zio)
1263 {
1264 uint64_t *good_writes = zio->io_private;
1265
1266 if (*good_writes == 0)
1267 zio->io_error = SET_ERROR(EIO);
1268
1269 kmem_free(good_writes, sizeof (uint64_t));
1270 }
1271
1272 /*
1273 * We ignore errors for log and cache devices, simply free the private data.
1274 */
1275 static void
vdev_label_sync_ignore_done(zio_t * zio)1276 vdev_label_sync_ignore_done(zio_t *zio)
1277 {
1278 kmem_free(zio->io_private, sizeof (uint64_t));
1279 }
1280
1281 /*
1282 * Write all even or odd labels to all leaves of the specified vdev.
1283 */
1284 static void
vdev_label_sync(zio_t * zio,uint64_t * good_writes,vdev_t * vd,int l,uint64_t txg,int flags)1285 vdev_label_sync(zio_t *zio, uint64_t *good_writes,
1286 vdev_t *vd, int l, uint64_t txg, int flags)
1287 {
1288 nvlist_t *label;
1289 vdev_phys_t *vp;
1290 abd_t *vp_abd;
1291 char *buf;
1292 size_t buflen;
1293
1294 for (int c = 0; c < vd->vdev_children; c++) {
1295 vdev_label_sync(zio, good_writes,
1296 vd->vdev_child[c], l, txg, flags);
1297 }
1298
1299 if (!vd->vdev_ops->vdev_op_leaf)
1300 return;
1301
1302 if (!vdev_writeable(vd))
1303 return;
1304
1305 /*
1306 * Generate a label describing the top-level config to which we belong.
1307 */
1308 label = spa_config_generate(vd->vdev_spa, vd, txg, B_FALSE);
1309
1310 vp_abd = abd_alloc_linear(sizeof (vdev_phys_t), B_TRUE);
1311 abd_zero(vp_abd, sizeof (vdev_phys_t));
1312 vp = abd_to_buf(vp_abd);
1313
1314 buf = vp->vp_nvlist;
1315 buflen = sizeof (vp->vp_nvlist);
1316
1317 if (nvlist_pack(label, &buf, &buflen, NV_ENCODE_XDR, KM_SLEEP) == 0) {
1318 for (; l < VDEV_LABELS; l += 2) {
1319 vdev_label_write(zio, vd, l, vp_abd,
1320 offsetof(vdev_label_t, vl_vdev_phys),
1321 sizeof (vdev_phys_t),
1322 vdev_label_sync_done, good_writes,
1323 flags | ZIO_FLAG_DONT_PROPAGATE);
1324 }
1325 }
1326
1327 abd_free(vp_abd);
1328 nvlist_free(label);
1329 }
1330
1331 int
vdev_label_sync_list(spa_t * spa,int l,uint64_t txg,int flags)1332 vdev_label_sync_list(spa_t *spa, int l, uint64_t txg, int flags)
1333 {
1334 list_t *dl = &spa->spa_config_dirty_list;
1335 vdev_t *vd;
1336 zio_t *zio;
1337 int error;
1338
1339 /*
1340 * Write the new labels to disk.
1341 */
1342 zio = zio_root(spa, NULL, NULL, flags);
1343
1344 for (vd = list_head(dl); vd != NULL; vd = list_next(dl, vd)) {
1345 uint64_t *good_writes = kmem_zalloc(sizeof (uint64_t),
1346 KM_SLEEP);
1347
1348 ASSERT(!vd->vdev_ishole);
1349
1350 zio_t *vio = zio_null(zio, spa, NULL,
1351 (vd->vdev_islog || vd->vdev_aux != NULL) ?
1352 vdev_label_sync_ignore_done : vdev_label_sync_top_done,
1353 good_writes, flags);
1354 vdev_label_sync(vio, good_writes, vd, l, txg, flags);
1355 zio_nowait(vio);
1356 }
1357
1358 error = zio_wait(zio);
1359
1360 /*
1361 * Flush the new labels to disk.
1362 */
1363 zio = zio_root(spa, NULL, NULL, flags);
1364
1365 for (vd = list_head(dl); vd != NULL; vd = list_next(dl, vd))
1366 zio_flush(zio, vd);
1367
1368 (void) zio_wait(zio);
1369
1370 return (error);
1371 }
1372
1373 /*
1374 * Sync the uberblock and any changes to the vdev configuration.
1375 *
1376 * The order of operations is carefully crafted to ensure that
1377 * if the system panics or loses power at any time, the state on disk
1378 * is still transactionally consistent. The in-line comments below
1379 * describe the failure semantics at each stage.
1380 *
1381 * Moreover, vdev_config_sync() is designed to be idempotent: if it fails
1382 * at any time, you can just call it again, and it will resume its work.
1383 */
1384 int
vdev_config_sync(vdev_t ** svd,int svdcount,uint64_t txg)1385 vdev_config_sync(vdev_t **svd, int svdcount, uint64_t txg)
1386 {
1387 spa_t *spa = svd[0]->vdev_spa;
1388 uberblock_t *ub = &spa->spa_uberblock;
1389 int error = 0;
1390 int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL;
1391
1392 ASSERT(svdcount != 0);
1393 retry:
1394 /*
1395 * Normally, we don't want to try too hard to write every label and
1396 * uberblock. If there is a flaky disk, we don't want the rest of the
1397 * sync process to block while we retry. But if we can't write a
1398 * single label out, we should retry with ZIO_FLAG_TRYHARD before
1399 * bailing out and declaring the pool faulted.
1400 */
1401 if (error != 0) {
1402 if ((flags & ZIO_FLAG_TRYHARD) != 0)
1403 return (error);
1404 flags |= ZIO_FLAG_TRYHARD;
1405 }
1406
1407 ASSERT(ub->ub_txg <= txg);
1408
1409 /*
1410 * If this isn't a resync due to I/O errors,
1411 * and nothing changed in this transaction group,
1412 * and the vdev configuration hasn't changed,
1413 * then there's nothing to do.
1414 */
1415 if (ub->ub_txg < txg &&
1416 uberblock_update(ub, spa->spa_root_vdev, txg) == B_FALSE &&
1417 list_is_empty(&spa->spa_config_dirty_list))
1418 return (0);
1419
1420 if (txg > spa_freeze_txg(spa))
1421 return (0);
1422
1423 ASSERT(txg <= spa->spa_final_txg);
1424
1425 /*
1426 * Flush the write cache of every disk that's been written to
1427 * in this transaction group. This ensures that all blocks
1428 * written in this txg will be committed to stable storage
1429 * before any uberblock that references them.
1430 */
1431 zio_t *zio = zio_root(spa, NULL, NULL, flags);
1432
1433 for (vdev_t *vd =
1434 txg_list_head(&spa->spa_vdev_txg_list, TXG_CLEAN(txg)); vd != NULL;
1435 vd = txg_list_next(&spa->spa_vdev_txg_list, vd, TXG_CLEAN(txg)))
1436 zio_flush(zio, vd);
1437
1438 (void) zio_wait(zio);
1439
1440 /*
1441 * Sync out the even labels (L0, L2) for every dirty vdev. If the
1442 * system dies in the middle of this process, that's OK: all of the
1443 * even labels that made it to disk will be newer than any uberblock,
1444 * and will therefore be considered invalid. The odd labels (L1, L3),
1445 * which have not yet been touched, will still be valid. We flush
1446 * the new labels to disk to ensure that all even-label updates
1447 * are committed to stable storage before the uberblock update.
1448 */
1449 if ((error = vdev_label_sync_list(spa, 0, txg, flags)) != 0) {
1450 if ((flags & ZIO_FLAG_TRYHARD) != 0) {
1451 zfs_dbgmsg("vdev_label_sync_list() returned error %d "
1452 "for pool '%s' when syncing out the even labels "
1453 "of dirty vdevs", error, spa_name(spa));
1454 }
1455 goto retry;
1456 }
1457
1458 /*
1459 * Sync the uberblocks to all vdevs in svd[].
1460 * If the system dies in the middle of this step, there are two cases
1461 * to consider, and the on-disk state is consistent either way:
1462 *
1463 * (1) If none of the new uberblocks made it to disk, then the
1464 * previous uberblock will be the newest, and the odd labels
1465 * (which had not yet been touched) will be valid with respect
1466 * to that uberblock.
1467 *
1468 * (2) If one or more new uberblocks made it to disk, then they
1469 * will be the newest, and the even labels (which had all
1470 * been successfully committed) will be valid with respect
1471 * to the new uberblocks.
1472 */
1473 if ((error = vdev_uberblock_sync_list(svd, svdcount, ub, flags)) != 0) {
1474 if ((flags & ZIO_FLAG_TRYHARD) != 0) {
1475 zfs_dbgmsg("vdev_uberblock_sync_list() returned error "
1476 "%d for pool '%s'", error, spa_name(spa));
1477 }
1478 goto retry;
1479 }
1480
1481 /*
1482 * Sync out odd labels for every dirty vdev. If the system dies
1483 * in the middle of this process, the even labels and the new
1484 * uberblocks will suffice to open the pool. The next time
1485 * the pool is opened, the first thing we'll do -- before any
1486 * user data is modified -- is mark every vdev dirty so that
1487 * all labels will be brought up to date. We flush the new labels
1488 * to disk to ensure that all odd-label updates are committed to
1489 * stable storage before the next transaction group begins.
1490 */
1491 if ((error = vdev_label_sync_list(spa, 1, txg, flags)) != 0) {
1492 if ((flags & ZIO_FLAG_TRYHARD) != 0) {
1493 zfs_dbgmsg("vdev_label_sync_list() returned error %d "
1494 "for pool '%s' when syncing out the odd labels of "
1495 "dirty vdevs", error, spa_name(spa));
1496 }
1497 goto retry;;
1498 }
1499
1500 trim_thread_wakeup(spa);
1501
1502 return (0);
1503 }
1504