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