]>
Commit | Line | Data |
---|---|---|
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) 2013 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/zio.h> | |
147 | #include <sys/dsl_scan.h> | |
148 | #include <sys/fs/zfs.h> | |
149 | ||
150 | /* | |
151 | * Basic routines to read and write from a vdev label. | |
152 | * Used throughout the rest of this file. | |
153 | */ | |
154 | uint64_t | |
155 | vdev_label_offset(uint64_t psize, int l, uint64_t offset) | |
156 | { | |
157 | ASSERT(offset < sizeof (vdev_label_t)); | |
158 | ASSERT(P2PHASE_TYPED(psize, sizeof (vdev_label_t), uint64_t) == 0); | |
159 | ||
160 | return (offset + l * sizeof (vdev_label_t) + (l < VDEV_LABELS / 2 ? | |
161 | 0 : psize - VDEV_LABELS * sizeof (vdev_label_t))); | |
162 | } | |
163 | ||
164 | /* | |
165 | * Returns back the vdev label associated with the passed in offset. | |
166 | */ | |
167 | int | |
168 | vdev_label_number(uint64_t psize, uint64_t offset) | |
169 | { | |
170 | int l; | |
171 | ||
172 | if (offset >= psize - VDEV_LABEL_END_SIZE) { | |
173 | offset -= psize - VDEV_LABEL_END_SIZE; | |
174 | offset += (VDEV_LABELS / 2) * sizeof (vdev_label_t); | |
175 | } | |
176 | l = offset / sizeof (vdev_label_t); | |
177 | return (l < VDEV_LABELS ? l : -1); | |
178 | } | |
179 | ||
180 | static void | |
181 | vdev_label_read(zio_t *zio, vdev_t *vd, int l, void *buf, uint64_t offset, | |
182 | uint64_t size, zio_done_func_t *done, void *private, int flags) | |
183 | { | |
184 | ASSERT(spa_config_held(zio->io_spa, SCL_STATE_ALL, RW_WRITER) == | |
185 | SCL_STATE_ALL); | |
186 | ASSERT(flags & ZIO_FLAG_CONFIG_WRITER); | |
187 | ||
188 | zio_nowait(zio_read_phys(zio, vd, | |
189 | vdev_label_offset(vd->vdev_psize, l, offset), | |
190 | size, buf, ZIO_CHECKSUM_LABEL, done, private, | |
191 | ZIO_PRIORITY_SYNC_READ, flags, B_TRUE)); | |
192 | } | |
193 | ||
194 | static void | |
195 | vdev_label_write(zio_t *zio, vdev_t *vd, int l, void *buf, uint64_t offset, | |
196 | uint64_t size, zio_done_func_t *done, void *private, int flags) | |
197 | { | |
198 | ASSERT(spa_config_held(zio->io_spa, SCL_ALL, RW_WRITER) == SCL_ALL || | |
199 | (spa_config_held(zio->io_spa, SCL_CONFIG | SCL_STATE, RW_READER) == | |
200 | (SCL_CONFIG | SCL_STATE) && | |
201 | dsl_pool_sync_context(spa_get_dsl(zio->io_spa)))); | |
202 | ASSERT(flags & ZIO_FLAG_CONFIG_WRITER); | |
203 | ||
204 | zio_nowait(zio_write_phys(zio, vd, | |
205 | vdev_label_offset(vd->vdev_psize, l, offset), | |
206 | size, buf, ZIO_CHECKSUM_LABEL, done, private, | |
207 | ZIO_PRIORITY_SYNC_WRITE, flags, B_TRUE)); | |
208 | } | |
209 | ||
210 | /* | |
211 | * Generate the nvlist representing this vdev's config. | |
212 | */ | |
213 | nvlist_t * | |
214 | vdev_config_generate(spa_t *spa, vdev_t *vd, boolean_t getstats, | |
215 | vdev_config_flag_t flags) | |
216 | { | |
217 | nvlist_t *nv = NULL; | |
218 | ||
219 | nv = fnvlist_alloc(); | |
220 | ||
221 | fnvlist_add_string(nv, ZPOOL_CONFIG_TYPE, vd->vdev_ops->vdev_op_type); | |
222 | if (!(flags & (VDEV_CONFIG_SPARE | VDEV_CONFIG_L2CACHE))) | |
223 | fnvlist_add_uint64(nv, ZPOOL_CONFIG_ID, vd->vdev_id); | |
224 | fnvlist_add_uint64(nv, ZPOOL_CONFIG_GUID, vd->vdev_guid); | |
225 | ||
226 | if (vd->vdev_path != NULL) | |
227 | fnvlist_add_string(nv, ZPOOL_CONFIG_PATH, vd->vdev_path); | |
228 | ||
229 | if (vd->vdev_devid != NULL) | |
230 | fnvlist_add_string(nv, ZPOOL_CONFIG_DEVID, vd->vdev_devid); | |
231 | ||
232 | if (vd->vdev_physpath != NULL) | |
233 | fnvlist_add_string(nv, ZPOOL_CONFIG_PHYS_PATH, | |
234 | vd->vdev_physpath); | |
235 | ||
236 | if (vd->vdev_fru != NULL) | |
237 | fnvlist_add_string(nv, ZPOOL_CONFIG_FRU, vd->vdev_fru); | |
238 | ||
239 | if (vd->vdev_nparity != 0) { | |
240 | ASSERT(strcmp(vd->vdev_ops->vdev_op_type, | |
241 | VDEV_TYPE_RAIDZ) == 0); | |
242 | ||
243 | /* | |
244 | * Make sure someone hasn't managed to sneak a fancy new vdev | |
245 | * into a crufty old storage pool. | |
246 | */ | |
247 | ASSERT(vd->vdev_nparity == 1 || | |
248 | (vd->vdev_nparity <= 2 && | |
249 | spa_version(spa) >= SPA_VERSION_RAIDZ2) || | |
250 | (vd->vdev_nparity <= 3 && | |
251 | spa_version(spa) >= SPA_VERSION_RAIDZ3)); | |
252 | ||
253 | /* | |
254 | * Note that we'll add the nparity tag even on storage pools | |
255 | * that only support a single parity device -- older software | |
256 | * will just ignore it. | |
257 | */ | |
258 | fnvlist_add_uint64(nv, ZPOOL_CONFIG_NPARITY, vd->vdev_nparity); | |
259 | } | |
260 | ||
261 | if (vd->vdev_wholedisk != -1ULL) | |
262 | fnvlist_add_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK, | |
263 | vd->vdev_wholedisk); | |
264 | ||
265 | if (vd->vdev_not_present) | |
266 | fnvlist_add_uint64(nv, ZPOOL_CONFIG_NOT_PRESENT, 1); | |
267 | ||
268 | if (vd->vdev_isspare) | |
269 | fnvlist_add_uint64(nv, ZPOOL_CONFIG_IS_SPARE, 1); | |
270 | ||
271 | if (!(flags & (VDEV_CONFIG_SPARE | VDEV_CONFIG_L2CACHE)) && | |
272 | vd == vd->vdev_top) { | |
273 | fnvlist_add_uint64(nv, ZPOOL_CONFIG_METASLAB_ARRAY, | |
274 | vd->vdev_ms_array); | |
275 | fnvlist_add_uint64(nv, ZPOOL_CONFIG_METASLAB_SHIFT, | |
276 | vd->vdev_ms_shift); | |
277 | fnvlist_add_uint64(nv, ZPOOL_CONFIG_ASHIFT, vd->vdev_ashift); | |
278 | fnvlist_add_uint64(nv, ZPOOL_CONFIG_ASIZE, | |
279 | vd->vdev_asize); | |
280 | fnvlist_add_uint64(nv, ZPOOL_CONFIG_IS_LOG, vd->vdev_islog); | |
281 | if (vd->vdev_removing) | |
282 | fnvlist_add_uint64(nv, ZPOOL_CONFIG_REMOVING, | |
283 | vd->vdev_removing); | |
284 | } | |
285 | ||
286 | if (vd->vdev_dtl_sm != NULL) { | |
287 | fnvlist_add_uint64(nv, ZPOOL_CONFIG_DTL, | |
288 | space_map_object(vd->vdev_dtl_sm)); | |
289 | } | |
290 | ||
291 | if (vd->vdev_crtxg) | |
292 | fnvlist_add_uint64(nv, ZPOOL_CONFIG_CREATE_TXG, vd->vdev_crtxg); | |
293 | ||
294 | if (getstats) { | |
295 | vdev_stat_t vs; | |
296 | pool_scan_stat_t ps; | |
297 | ||
298 | vdev_get_stats(vd, &vs); | |
299 | fnvlist_add_uint64_array(nv, ZPOOL_CONFIG_VDEV_STATS, | |
300 | (uint64_t *)&vs, sizeof (vs) / sizeof (uint64_t)); | |
301 | ||
302 | /* provide either current or previous scan information */ | |
303 | if (spa_scan_get_stats(spa, &ps) == 0) { | |
304 | fnvlist_add_uint64_array(nv, | |
305 | ZPOOL_CONFIG_SCAN_STATS, (uint64_t *)&ps, | |
306 | sizeof (pool_scan_stat_t) / sizeof (uint64_t)); | |
307 | } | |
308 | } | |
309 | ||
310 | if (!vd->vdev_ops->vdev_op_leaf) { | |
311 | nvlist_t **child; | |
312 | int c, idx; | |
313 | ||
314 | ASSERT(!vd->vdev_ishole); | |
315 | ||
316 | child = kmem_alloc(vd->vdev_children * sizeof (nvlist_t *), | |
317 | KM_SLEEP); | |
318 | ||
319 | for (c = 0, idx = 0; c < vd->vdev_children; c++) { | |
320 | vdev_t *cvd = vd->vdev_child[c]; | |
321 | ||
322 | /* | |
323 | * If we're generating an nvlist of removing | |
324 | * vdevs then skip over any device which is | |
325 | * not being removed. | |
326 | */ | |
327 | if ((flags & VDEV_CONFIG_REMOVING) && | |
328 | !cvd->vdev_removing) | |
329 | continue; | |
330 | ||
331 | child[idx++] = vdev_config_generate(spa, cvd, | |
332 | getstats, flags); | |
333 | } | |
334 | ||
335 | if (idx) { | |
336 | fnvlist_add_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN, | |
337 | child, idx); | |
338 | } | |
339 | ||
340 | for (c = 0; c < idx; c++) | |
341 | nvlist_free(child[c]); | |
342 | ||
343 | kmem_free(child, vd->vdev_children * sizeof (nvlist_t *)); | |
344 | ||
345 | } else { | |
346 | const char *aux = NULL; | |
347 | ||
348 | if (vd->vdev_offline && !vd->vdev_tmpoffline) | |
349 | fnvlist_add_uint64(nv, ZPOOL_CONFIG_OFFLINE, B_TRUE); | |
350 | if (vd->vdev_resilver_txg != 0) | |
351 | fnvlist_add_uint64(nv, ZPOOL_CONFIG_RESILVER_TXG, | |
352 | vd->vdev_resilver_txg); | |
353 | if (vd->vdev_faulted) | |
354 | fnvlist_add_uint64(nv, ZPOOL_CONFIG_FAULTED, B_TRUE); | |
355 | if (vd->vdev_degraded) | |
356 | fnvlist_add_uint64(nv, ZPOOL_CONFIG_DEGRADED, B_TRUE); | |
357 | if (vd->vdev_removed) | |
358 | fnvlist_add_uint64(nv, ZPOOL_CONFIG_REMOVED, B_TRUE); | |
359 | if (vd->vdev_unspare) | |
360 | fnvlist_add_uint64(nv, ZPOOL_CONFIG_UNSPARE, B_TRUE); | |
361 | if (vd->vdev_ishole) | |
362 | fnvlist_add_uint64(nv, ZPOOL_CONFIG_IS_HOLE, B_TRUE); | |
363 | ||
364 | switch (vd->vdev_stat.vs_aux) { | |
365 | case VDEV_AUX_ERR_EXCEEDED: | |
366 | aux = "err_exceeded"; | |
367 | break; | |
368 | ||
369 | case VDEV_AUX_EXTERNAL: | |
370 | aux = "external"; | |
371 | break; | |
372 | } | |
373 | ||
374 | if (aux != NULL) | |
375 | fnvlist_add_string(nv, ZPOOL_CONFIG_AUX_STATE, aux); | |
376 | ||
377 | if (vd->vdev_splitting && vd->vdev_orig_guid != 0LL) { | |
378 | fnvlist_add_uint64(nv, ZPOOL_CONFIG_ORIG_GUID, | |
379 | vd->vdev_orig_guid); | |
380 | } | |
381 | } | |
382 | ||
383 | return (nv); | |
384 | } | |
385 | ||
386 | /* | |
387 | * Generate a view of the top-level vdevs. If we currently have holes | |
388 | * in the namespace, then generate an array which contains a list of holey | |
389 | * vdevs. Additionally, add the number of top-level children that currently | |
390 | * exist. | |
391 | */ | |
392 | void | |
393 | vdev_top_config_generate(spa_t *spa, nvlist_t *config) | |
394 | { | |
395 | vdev_t *rvd = spa->spa_root_vdev; | |
396 | uint64_t *array; | |
397 | uint_t c, idx; | |
398 | ||
399 | array = kmem_alloc(rvd->vdev_children * sizeof (uint64_t), KM_SLEEP); | |
400 | ||
401 | for (c = 0, idx = 0; c < rvd->vdev_children; c++) { | |
402 | vdev_t *tvd = rvd->vdev_child[c]; | |
403 | ||
404 | if (tvd->vdev_ishole) | |
405 | array[idx++] = c; | |
406 | } | |
407 | ||
408 | if (idx) { | |
409 | VERIFY(nvlist_add_uint64_array(config, ZPOOL_CONFIG_HOLE_ARRAY, | |
410 | array, idx) == 0); | |
411 | } | |
412 | ||
413 | VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VDEV_CHILDREN, | |
414 | rvd->vdev_children) == 0); | |
415 | ||
416 | kmem_free(array, rvd->vdev_children * sizeof (uint64_t)); | |
417 | } | |
418 | ||
419 | /* | |
420 | * Returns the configuration from the label of the given vdev. For vdevs | |
421 | * which don't have a txg value stored on their label (i.e. spares/cache) | |
422 | * or have not been completely initialized (txg = 0) just return | |
423 | * the configuration from the first valid label we find. Otherwise, | |
424 | * find the most up-to-date label that does not exceed the specified | |
425 | * 'txg' value. | |
426 | */ | |
427 | nvlist_t * | |
428 | vdev_label_read_config(vdev_t *vd, uint64_t txg) | |
429 | { | |
430 | spa_t *spa = vd->vdev_spa; | |
431 | nvlist_t *config = NULL; | |
432 | vdev_phys_t *vp; | |
433 | zio_t *zio; | |
434 | uint64_t best_txg = 0; | |
435 | int error = 0; | |
436 | int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL | | |
437 | ZIO_FLAG_SPECULATIVE; | |
438 | int l; | |
439 | ||
440 | ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); | |
441 | ||
442 | if (!vdev_readable(vd)) | |
443 | return (NULL); | |
444 | ||
445 | vp = zio_buf_alloc(sizeof (vdev_phys_t)); | |
446 | ||
447 | retry: | |
448 | for (l = 0; l < VDEV_LABELS; l++) { | |
449 | nvlist_t *label = NULL; | |
450 | ||
451 | zio = zio_root(spa, NULL, NULL, flags); | |
452 | ||
453 | vdev_label_read(zio, vd, l, vp, | |
454 | offsetof(vdev_label_t, vl_vdev_phys), | |
455 | sizeof (vdev_phys_t), NULL, NULL, flags); | |
456 | ||
457 | if (zio_wait(zio) == 0 && | |
458 | nvlist_unpack(vp->vp_nvlist, sizeof (vp->vp_nvlist), | |
459 | &label, 0) == 0) { | |
460 | uint64_t label_txg = 0; | |
461 | ||
462 | /* | |
463 | * Auxiliary vdevs won't have txg values in their | |
464 | * labels and newly added vdevs may not have been | |
465 | * completely initialized so just return the | |
466 | * configuration from the first valid label we | |
467 | * encounter. | |
468 | */ | |
469 | error = nvlist_lookup_uint64(label, | |
470 | ZPOOL_CONFIG_POOL_TXG, &label_txg); | |
471 | if ((error || label_txg == 0) && !config) { | |
472 | config = label; | |
473 | break; | |
474 | } else if (label_txg <= txg && label_txg > best_txg) { | |
475 | best_txg = label_txg; | |
476 | nvlist_free(config); | |
477 | config = fnvlist_dup(label); | |
478 | } | |
479 | } | |
480 | ||
481 | if (label != NULL) { | |
482 | nvlist_free(label); | |
483 | label = NULL; | |
484 | } | |
485 | } | |
486 | ||
487 | if (config == NULL && !(flags & ZIO_FLAG_TRYHARD)) { | |
488 | flags |= ZIO_FLAG_TRYHARD; | |
489 | goto retry; | |
490 | } | |
491 | ||
492 | zio_buf_free(vp, sizeof (vdev_phys_t)); | |
493 | ||
494 | return (config); | |
495 | } | |
496 | ||
497 | /* | |
498 | * Determine if a device is in use. The 'spare_guid' parameter will be filled | |
499 | * in with the device guid if this spare is active elsewhere on the system. | |
500 | */ | |
501 | static boolean_t | |
502 | vdev_inuse(vdev_t *vd, uint64_t crtxg, vdev_labeltype_t reason, | |
503 | uint64_t *spare_guid, uint64_t *l2cache_guid) | |
504 | { | |
505 | spa_t *spa = vd->vdev_spa; | |
506 | uint64_t state, pool_guid, device_guid, txg, spare_pool; | |
507 | uint64_t vdtxg = 0; | |
508 | nvlist_t *label; | |
509 | ||
510 | if (spare_guid) | |
511 | *spare_guid = 0ULL; | |
512 | if (l2cache_guid) | |
513 | *l2cache_guid = 0ULL; | |
514 | ||
515 | /* | |
516 | * Read the label, if any, and perform some basic sanity checks. | |
517 | */ | |
518 | if ((label = vdev_label_read_config(vd, -1ULL)) == NULL) | |
519 | return (B_FALSE); | |
520 | ||
521 | (void) nvlist_lookup_uint64(label, ZPOOL_CONFIG_CREATE_TXG, | |
522 | &vdtxg); | |
523 | ||
524 | if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE, | |
525 | &state) != 0 || | |
526 | nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, | |
527 | &device_guid) != 0) { | |
528 | nvlist_free(label); | |
529 | return (B_FALSE); | |
530 | } | |
531 | ||
532 | if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE && | |
533 | (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_GUID, | |
534 | &pool_guid) != 0 || | |
535 | nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG, | |
536 | &txg) != 0)) { | |
537 | nvlist_free(label); | |
538 | return (B_FALSE); | |
539 | } | |
540 | ||
541 | nvlist_free(label); | |
542 | ||
543 | /* | |
544 | * Check to see if this device indeed belongs to the pool it claims to | |
545 | * be a part of. The only way this is allowed is if the device is a hot | |
546 | * spare (which we check for later on). | |
547 | */ | |
548 | if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE && | |
549 | !spa_guid_exists(pool_guid, device_guid) && | |
550 | !spa_spare_exists(device_guid, NULL, NULL) && | |
551 | !spa_l2cache_exists(device_guid, NULL)) | |
552 | return (B_FALSE); | |
553 | ||
554 | /* | |
555 | * If the transaction group is zero, then this an initialized (but | |
556 | * unused) label. This is only an error if the create transaction | |
557 | * on-disk is the same as the one we're using now, in which case the | |
558 | * user has attempted to add the same vdev multiple times in the same | |
559 | * transaction. | |
560 | */ | |
561 | if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE && | |
562 | txg == 0 && vdtxg == crtxg) | |
563 | return (B_TRUE); | |
564 | ||
565 | /* | |
566 | * Check to see if this is a spare device. We do an explicit check for | |
567 | * spa_has_spare() here because it may be on our pending list of spares | |
568 | * to add. We also check if it is an l2cache device. | |
569 | */ | |
570 | if (spa_spare_exists(device_guid, &spare_pool, NULL) || | |
571 | spa_has_spare(spa, device_guid)) { | |
572 | if (spare_guid) | |
573 | *spare_guid = device_guid; | |
574 | ||
575 | switch (reason) { | |
576 | case VDEV_LABEL_CREATE: | |
577 | case VDEV_LABEL_L2CACHE: | |
578 | return (B_TRUE); | |
579 | ||
580 | case VDEV_LABEL_REPLACE: | |
581 | return (!spa_has_spare(spa, device_guid) || | |
582 | spare_pool != 0ULL); | |
583 | ||
584 | case VDEV_LABEL_SPARE: | |
585 | return (spa_has_spare(spa, device_guid)); | |
586 | default: | |
587 | break; | |
588 | } | |
589 | } | |
590 | ||
591 | /* | |
592 | * Check to see if this is an l2cache device. | |
593 | */ | |
594 | if (spa_l2cache_exists(device_guid, NULL)) | |
595 | return (B_TRUE); | |
596 | ||
597 | /* | |
598 | * We can't rely on a pool's state if it's been imported | |
599 | * read-only. Instead we look to see if the pools is marked | |
600 | * read-only in the namespace and set the state to active. | |
601 | */ | |
602 | if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE && | |
603 | (spa = spa_by_guid(pool_guid, device_guid)) != NULL && | |
604 | spa_mode(spa) == FREAD) | |
605 | state = POOL_STATE_ACTIVE; | |
606 | ||
607 | /* | |
608 | * If the device is marked ACTIVE, then this device is in use by another | |
609 | * pool on the system. | |
610 | */ | |
611 | return (state == POOL_STATE_ACTIVE); | |
612 | } | |
613 | ||
614 | /* | |
615 | * Initialize a vdev label. We check to make sure each leaf device is not in | |
616 | * use, and writable. We put down an initial label which we will later | |
617 | * overwrite with a complete label. Note that it's important to do this | |
618 | * sequentially, not in parallel, so that we catch cases of multiple use of the | |
619 | * same leaf vdev in the vdev we're creating -- e.g. mirroring a disk with | |
620 | * itself. | |
621 | */ | |
622 | int | |
623 | vdev_label_init(vdev_t *vd, uint64_t crtxg, vdev_labeltype_t reason) | |
624 | { | |
625 | spa_t *spa = vd->vdev_spa; | |
626 | nvlist_t *label; | |
627 | vdev_phys_t *vp; | |
628 | char *pad2; | |
629 | uberblock_t *ub; | |
630 | zio_t *zio; | |
631 | char *buf; | |
632 | size_t buflen; | |
633 | int error; | |
634 | uint64_t spare_guid = 0, l2cache_guid = 0; | |
635 | int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL; | |
636 | int c, l; | |
637 | vdev_t *pvd; | |
638 | ||
639 | ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); | |
640 | ||
641 | for (c = 0; c < vd->vdev_children; c++) | |
642 | if ((error = vdev_label_init(vd->vdev_child[c], | |
643 | crtxg, reason)) != 0) | |
644 | return (error); | |
645 | ||
646 | /* Track the creation time for this vdev */ | |
647 | vd->vdev_crtxg = crtxg; | |
648 | ||
649 | if (!vd->vdev_ops->vdev_op_leaf || !spa_writeable(spa)) | |
650 | return (0); | |
651 | ||
652 | /* | |
653 | * Dead vdevs cannot be initialized. | |
654 | */ | |
655 | if (vdev_is_dead(vd)) | |
656 | return (SET_ERROR(EIO)); | |
657 | ||
658 | /* | |
659 | * Determine if the vdev is in use. | |
660 | */ | |
661 | if (reason != VDEV_LABEL_REMOVE && reason != VDEV_LABEL_SPLIT && | |
662 | vdev_inuse(vd, crtxg, reason, &spare_guid, &l2cache_guid)) | |
663 | return (SET_ERROR(EBUSY)); | |
664 | ||
665 | /* | |
666 | * If this is a request to add or replace a spare or l2cache device | |
667 | * that is in use elsewhere on the system, then we must update the | |
668 | * guid (which was initialized to a random value) to reflect the | |
669 | * actual GUID (which is shared between multiple pools). | |
670 | */ | |
671 | if (reason != VDEV_LABEL_REMOVE && reason != VDEV_LABEL_L2CACHE && | |
672 | spare_guid != 0ULL) { | |
673 | uint64_t guid_delta = spare_guid - vd->vdev_guid; | |
674 | ||
675 | vd->vdev_guid += guid_delta; | |
676 | ||
677 | for (pvd = vd; pvd != NULL; pvd = pvd->vdev_parent) | |
678 | pvd->vdev_guid_sum += guid_delta; | |
679 | ||
680 | /* | |
681 | * If this is a replacement, then we want to fallthrough to the | |
682 | * rest of the code. If we're adding a spare, then it's already | |
683 | * labeled appropriately and we can just return. | |
684 | */ | |
685 | if (reason == VDEV_LABEL_SPARE) | |
686 | return (0); | |
687 | ASSERT(reason == VDEV_LABEL_REPLACE || | |
688 | reason == VDEV_LABEL_SPLIT); | |
689 | } | |
690 | ||
691 | if (reason != VDEV_LABEL_REMOVE && reason != VDEV_LABEL_SPARE && | |
692 | l2cache_guid != 0ULL) { | |
693 | uint64_t guid_delta = l2cache_guid - vd->vdev_guid; | |
694 | ||
695 | vd->vdev_guid += guid_delta; | |
696 | ||
697 | for (pvd = vd; pvd != NULL; pvd = pvd->vdev_parent) | |
698 | pvd->vdev_guid_sum += guid_delta; | |
699 | ||
700 | /* | |
701 | * If this is a replacement, then we want to fallthrough to the | |
702 | * rest of the code. If we're adding an l2cache, then it's | |
703 | * already labeled appropriately and we can just return. | |
704 | */ | |
705 | if (reason == VDEV_LABEL_L2CACHE) | |
706 | return (0); | |
707 | ASSERT(reason == VDEV_LABEL_REPLACE); | |
708 | } | |
709 | ||
710 | /* | |
711 | * Initialize its label. | |
712 | */ | |
713 | vp = zio_buf_alloc(sizeof (vdev_phys_t)); | |
714 | bzero(vp, sizeof (vdev_phys_t)); | |
715 | ||
716 | /* | |
717 | * Generate a label describing the pool and our top-level vdev. | |
718 | * We mark it as being from txg 0 to indicate that it's not | |
719 | * really part of an active pool just yet. The labels will | |
720 | * be written again with a meaningful txg by spa_sync(). | |
721 | */ | |
722 | if (reason == VDEV_LABEL_SPARE || | |
723 | (reason == VDEV_LABEL_REMOVE && vd->vdev_isspare)) { | |
724 | /* | |
725 | * For inactive hot spares, we generate a special label that | |
726 | * identifies as a mutually shared hot spare. We write the | |
727 | * label if we are adding a hot spare, or if we are removing an | |
728 | * active hot spare (in which case we want to revert the | |
729 | * labels). | |
730 | */ | |
731 | VERIFY(nvlist_alloc(&label, NV_UNIQUE_NAME, KM_SLEEP) == 0); | |
732 | ||
733 | VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_VERSION, | |
734 | spa_version(spa)) == 0); | |
735 | VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_POOL_STATE, | |
736 | POOL_STATE_SPARE) == 0); | |
737 | VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_GUID, | |
738 | vd->vdev_guid) == 0); | |
739 | } else if (reason == VDEV_LABEL_L2CACHE || | |
740 | (reason == VDEV_LABEL_REMOVE && vd->vdev_isl2cache)) { | |
741 | /* | |
742 | * For level 2 ARC devices, add a special label. | |
743 | */ | |
744 | VERIFY(nvlist_alloc(&label, NV_UNIQUE_NAME, KM_SLEEP) == 0); | |
745 | ||
746 | VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_VERSION, | |
747 | spa_version(spa)) == 0); | |
748 | VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_POOL_STATE, | |
749 | POOL_STATE_L2CACHE) == 0); | |
750 | VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_GUID, | |
751 | vd->vdev_guid) == 0); | |
752 | } else { | |
753 | uint64_t txg = 0ULL; | |
754 | ||
755 | if (reason == VDEV_LABEL_SPLIT) | |
756 | txg = spa->spa_uberblock.ub_txg; | |
757 | label = spa_config_generate(spa, vd, txg, B_FALSE); | |
758 | ||
759 | /* | |
760 | * Add our creation time. This allows us to detect multiple | |
761 | * vdev uses as described above, and automatically expires if we | |
762 | * fail. | |
763 | */ | |
764 | VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_CREATE_TXG, | |
765 | crtxg) == 0); | |
766 | } | |
767 | ||
768 | buf = vp->vp_nvlist; | |
769 | buflen = sizeof (vp->vp_nvlist); | |
770 | ||
771 | error = nvlist_pack(label, &buf, &buflen, NV_ENCODE_XDR, KM_SLEEP); | |
772 | if (error != 0) { | |
773 | nvlist_free(label); | |
774 | zio_buf_free(vp, sizeof (vdev_phys_t)); | |
775 | /* EFAULT means nvlist_pack ran out of room */ | |
776 | return (error == EFAULT ? ENAMETOOLONG : EINVAL); | |
777 | } | |
778 | ||
779 | /* | |
780 | * Initialize uberblock template. | |
781 | */ | |
782 | ub = zio_buf_alloc(VDEV_UBERBLOCK_RING); | |
783 | bzero(ub, VDEV_UBERBLOCK_RING); | |
784 | *ub = spa->spa_uberblock; | |
785 | ub->ub_txg = 0; | |
786 | ||
787 | /* Initialize the 2nd padding area. */ | |
788 | pad2 = zio_buf_alloc(VDEV_PAD_SIZE); | |
789 | bzero(pad2, VDEV_PAD_SIZE); | |
790 | ||
791 | /* | |
792 | * Write everything in parallel. | |
793 | */ | |
794 | retry: | |
795 | zio = zio_root(spa, NULL, NULL, flags); | |
796 | ||
797 | for (l = 0; l < VDEV_LABELS; l++) { | |
798 | ||
799 | vdev_label_write(zio, vd, l, vp, | |
800 | offsetof(vdev_label_t, vl_vdev_phys), | |
801 | sizeof (vdev_phys_t), NULL, NULL, flags); | |
802 | ||
803 | /* | |
804 | * Skip the 1st padding area. | |
805 | * Zero out the 2nd padding area where it might have | |
806 | * left over data from previous filesystem format. | |
807 | */ | |
808 | vdev_label_write(zio, vd, l, pad2, | |
809 | offsetof(vdev_label_t, vl_pad2), | |
810 | VDEV_PAD_SIZE, NULL, NULL, flags); | |
811 | ||
812 | vdev_label_write(zio, vd, l, ub, | |
813 | offsetof(vdev_label_t, vl_uberblock), | |
814 | VDEV_UBERBLOCK_RING, NULL, NULL, flags); | |
815 | } | |
816 | ||
817 | error = zio_wait(zio); | |
818 | ||
819 | if (error != 0 && !(flags & ZIO_FLAG_TRYHARD)) { | |
820 | flags |= ZIO_FLAG_TRYHARD; | |
821 | goto retry; | |
822 | } | |
823 | ||
824 | nvlist_free(label); | |
825 | zio_buf_free(pad2, VDEV_PAD_SIZE); | |
826 | zio_buf_free(ub, VDEV_UBERBLOCK_RING); | |
827 | zio_buf_free(vp, sizeof (vdev_phys_t)); | |
828 | ||
829 | /* | |
830 | * If this vdev hasn't been previously identified as a spare, then we | |
831 | * mark it as such only if a) we are labeling it as a spare, or b) it | |
832 | * exists as a spare elsewhere in the system. Do the same for | |
833 | * level 2 ARC devices. | |
834 | */ | |
835 | if (error == 0 && !vd->vdev_isspare && | |
836 | (reason == VDEV_LABEL_SPARE || | |
837 | spa_spare_exists(vd->vdev_guid, NULL, NULL))) | |
838 | spa_spare_add(vd); | |
839 | ||
840 | if (error == 0 && !vd->vdev_isl2cache && | |
841 | (reason == VDEV_LABEL_L2CACHE || | |
842 | spa_l2cache_exists(vd->vdev_guid, NULL))) | |
843 | spa_l2cache_add(vd); | |
844 | ||
845 | return (error); | |
846 | } | |
847 | ||
848 | /* | |
849 | * ========================================================================== | |
850 | * uberblock load/sync | |
851 | * ========================================================================== | |
852 | */ | |
853 | ||
854 | /* | |
855 | * Consider the following situation: txg is safely synced to disk. We've | |
856 | * written the first uberblock for txg + 1, and then we lose power. When we | |
857 | * come back up, we fail to see the uberblock for txg + 1 because, say, | |
858 | * it was on a mirrored device and the replica to which we wrote txg + 1 | |
859 | * is now offline. If we then make some changes and sync txg + 1, and then | |
860 | * the missing replica comes back, then for a few seconds we'll have two | |
861 | * conflicting uberblocks on disk with the same txg. The solution is simple: | |
862 | * among uberblocks with equal txg, choose the one with the latest timestamp. | |
863 | */ | |
864 | static int | |
865 | vdev_uberblock_compare(uberblock_t *ub1, uberblock_t *ub2) | |
866 | { | |
867 | if (ub1->ub_txg < ub2->ub_txg) | |
868 | return (-1); | |
869 | if (ub1->ub_txg > ub2->ub_txg) | |
870 | return (1); | |
871 | ||
872 | if (ub1->ub_timestamp < ub2->ub_timestamp) | |
873 | return (-1); | |
874 | if (ub1->ub_timestamp > ub2->ub_timestamp) | |
875 | return (1); | |
876 | ||
877 | return (0); | |
878 | } | |
879 | ||
880 | struct ubl_cbdata { | |
881 | uberblock_t *ubl_ubbest; /* Best uberblock */ | |
882 | vdev_t *ubl_vd; /* vdev associated with the above */ | |
883 | }; | |
884 | ||
885 | static void | |
886 | vdev_uberblock_load_done(zio_t *zio) | |
887 | { | |
888 | vdev_t *vd = zio->io_vd; | |
889 | spa_t *spa = zio->io_spa; | |
890 | zio_t *rio = zio->io_private; | |
891 | uberblock_t *ub = zio->io_data; | |
892 | struct ubl_cbdata *cbp = rio->io_private; | |
893 | ||
894 | ASSERT3U(zio->io_size, ==, VDEV_UBERBLOCK_SIZE(vd)); | |
895 | ||
896 | if (zio->io_error == 0 && uberblock_verify(ub) == 0) { | |
897 | mutex_enter(&rio->io_lock); | |
898 | if (ub->ub_txg <= spa->spa_load_max_txg && | |
899 | vdev_uberblock_compare(ub, cbp->ubl_ubbest) > 0) { | |
900 | /* | |
901 | * Keep track of the vdev in which this uberblock | |
902 | * was found. We will use this information later | |
903 | * to obtain the config nvlist associated with | |
904 | * this uberblock. | |
905 | */ | |
906 | *cbp->ubl_ubbest = *ub; | |
907 | cbp->ubl_vd = vd; | |
908 | } | |
909 | mutex_exit(&rio->io_lock); | |
910 | } | |
911 | ||
912 | zio_buf_free(zio->io_data, zio->io_size); | |
913 | } | |
914 | ||
915 | static void | |
916 | vdev_uberblock_load_impl(zio_t *zio, vdev_t *vd, int flags, | |
917 | struct ubl_cbdata *cbp) | |
918 | { | |
919 | int c, l, n; | |
920 | ||
921 | for (c = 0; c < vd->vdev_children; c++) | |
922 | vdev_uberblock_load_impl(zio, vd->vdev_child[c], flags, cbp); | |
923 | ||
924 | if (vd->vdev_ops->vdev_op_leaf && vdev_readable(vd)) { | |
925 | for (l = 0; l < VDEV_LABELS; l++) { | |
926 | for (n = 0; n < VDEV_UBERBLOCK_COUNT(vd); n++) { | |
927 | vdev_label_read(zio, vd, l, | |
928 | zio_buf_alloc(VDEV_UBERBLOCK_SIZE(vd)), | |
929 | VDEV_UBERBLOCK_OFFSET(vd, n), | |
930 | VDEV_UBERBLOCK_SIZE(vd), | |
931 | vdev_uberblock_load_done, zio, flags); | |
932 | } | |
933 | } | |
934 | } | |
935 | } | |
936 | ||
937 | /* | |
938 | * Reads the 'best' uberblock from disk along with its associated | |
939 | * configuration. First, we read the uberblock array of each label of each | |
940 | * vdev, keeping track of the uberblock with the highest txg in each array. | |
941 | * Then, we read the configuration from the same vdev as the best uberblock. | |
942 | */ | |
943 | void | |
944 | vdev_uberblock_load(vdev_t *rvd, uberblock_t *ub, nvlist_t **config) | |
945 | { | |
946 | zio_t *zio; | |
947 | spa_t *spa = rvd->vdev_spa; | |
948 | struct ubl_cbdata cb; | |
949 | int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL | | |
950 | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_TRYHARD; | |
951 | ||
952 | ASSERT(ub); | |
953 | ASSERT(config); | |
954 | ||
955 | bzero(ub, sizeof (uberblock_t)); | |
956 | *config = NULL; | |
957 | ||
958 | cb.ubl_ubbest = ub; | |
959 | cb.ubl_vd = NULL; | |
960 | ||
961 | spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); | |
962 | zio = zio_root(spa, NULL, &cb, flags); | |
963 | vdev_uberblock_load_impl(zio, rvd, flags, &cb); | |
964 | (void) zio_wait(zio); | |
965 | ||
966 | /* | |
967 | * It's possible that the best uberblock was discovered on a label | |
968 | * that has a configuration which was written in a future txg. | |
969 | * Search all labels on this vdev to find the configuration that | |
970 | * matches the txg for our uberblock. | |
971 | */ | |
972 | if (cb.ubl_vd != NULL) | |
973 | *config = vdev_label_read_config(cb.ubl_vd, ub->ub_txg); | |
974 | spa_config_exit(spa, SCL_ALL, FTAG); | |
975 | } | |
976 | ||
977 | /* | |
978 | * On success, increment root zio's count of good writes. | |
979 | * We only get credit for writes to known-visible vdevs; see spa_vdev_add(). | |
980 | */ | |
981 | static void | |
982 | vdev_uberblock_sync_done(zio_t *zio) | |
983 | { | |
984 | uint64_t *good_writes = zio->io_private; | |
985 | ||
986 | if (zio->io_error == 0 && zio->io_vd->vdev_top->vdev_ms_array != 0) | |
987 | atomic_add_64(good_writes, 1); | |
988 | } | |
989 | ||
990 | /* | |
991 | * Write the uberblock to all labels of all leaves of the specified vdev. | |
992 | */ | |
993 | static void | |
994 | vdev_uberblock_sync(zio_t *zio, uberblock_t *ub, vdev_t *vd, int flags) | |
995 | { | |
996 | uberblock_t *ubbuf; | |
997 | int c, l, n; | |
998 | ||
999 | for (c = 0; c < vd->vdev_children; c++) | |
1000 | vdev_uberblock_sync(zio, ub, vd->vdev_child[c], flags); | |
1001 | ||
1002 | if (!vd->vdev_ops->vdev_op_leaf) | |
1003 | return; | |
1004 | ||
1005 | if (!vdev_writeable(vd)) | |
1006 | return; | |
1007 | ||
1008 | n = ub->ub_txg & (VDEV_UBERBLOCK_COUNT(vd) - 1); | |
1009 | ||
1010 | ubbuf = zio_buf_alloc(VDEV_UBERBLOCK_SIZE(vd)); | |
1011 | bzero(ubbuf, VDEV_UBERBLOCK_SIZE(vd)); | |
1012 | *ubbuf = *ub; | |
1013 | ||
1014 | for (l = 0; l < VDEV_LABELS; l++) | |
1015 | vdev_label_write(zio, vd, l, ubbuf, | |
1016 | VDEV_UBERBLOCK_OFFSET(vd, n), VDEV_UBERBLOCK_SIZE(vd), | |
1017 | vdev_uberblock_sync_done, zio->io_private, | |
1018 | flags | ZIO_FLAG_DONT_PROPAGATE); | |
1019 | ||
1020 | zio_buf_free(ubbuf, VDEV_UBERBLOCK_SIZE(vd)); | |
1021 | } | |
1022 | ||
1023 | /* Sync the uberblocks to all vdevs in svd[] */ | |
1024 | int | |
1025 | vdev_uberblock_sync_list(vdev_t **svd, int svdcount, uberblock_t *ub, int flags) | |
1026 | { | |
1027 | spa_t *spa = svd[0]->vdev_spa; | |
1028 | zio_t *zio; | |
1029 | uint64_t good_writes = 0; | |
1030 | int v; | |
1031 | ||
1032 | zio = zio_root(spa, NULL, &good_writes, flags); | |
1033 | ||
1034 | for (v = 0; v < svdcount; v++) | |
1035 | vdev_uberblock_sync(zio, ub, svd[v], flags); | |
1036 | ||
1037 | (void) zio_wait(zio); | |
1038 | ||
1039 | /* | |
1040 | * Flush the uberblocks to disk. This ensures that the odd labels | |
1041 | * are no longer needed (because the new uberblocks and the even | |
1042 | * labels are safely on disk), so it is safe to overwrite them. | |
1043 | */ | |
1044 | zio = zio_root(spa, NULL, NULL, flags); | |
1045 | ||
1046 | for (v = 0; v < svdcount; v++) | |
1047 | zio_flush(zio, svd[v]); | |
1048 | ||
1049 | (void) zio_wait(zio); | |
1050 | ||
1051 | return (good_writes >= 1 ? 0 : EIO); | |
1052 | } | |
1053 | ||
1054 | /* | |
1055 | * On success, increment the count of good writes for our top-level vdev. | |
1056 | */ | |
1057 | static void | |
1058 | vdev_label_sync_done(zio_t *zio) | |
1059 | { | |
1060 | uint64_t *good_writes = zio->io_private; | |
1061 | ||
1062 | if (zio->io_error == 0) | |
1063 | atomic_add_64(good_writes, 1); | |
1064 | } | |
1065 | ||
1066 | /* | |
1067 | * If there weren't enough good writes, indicate failure to the parent. | |
1068 | */ | |
1069 | static void | |
1070 | vdev_label_sync_top_done(zio_t *zio) | |
1071 | { | |
1072 | uint64_t *good_writes = zio->io_private; | |
1073 | ||
1074 | if (*good_writes == 0) | |
1075 | zio->io_error = SET_ERROR(EIO); | |
1076 | ||
1077 | kmem_free(good_writes, sizeof (uint64_t)); | |
1078 | } | |
1079 | ||
1080 | /* | |
1081 | * We ignore errors for log and cache devices, simply free the private data. | |
1082 | */ | |
1083 | static void | |
1084 | vdev_label_sync_ignore_done(zio_t *zio) | |
1085 | { | |
1086 | kmem_free(zio->io_private, sizeof (uint64_t)); | |
1087 | } | |
1088 | ||
1089 | /* | |
1090 | * Write all even or odd labels to all leaves of the specified vdev. | |
1091 | */ | |
1092 | static void | |
1093 | vdev_label_sync(zio_t *zio, vdev_t *vd, int l, uint64_t txg, int flags) | |
1094 | { | |
1095 | nvlist_t *label; | |
1096 | vdev_phys_t *vp; | |
1097 | char *buf; | |
1098 | size_t buflen; | |
1099 | int c; | |
1100 | ||
1101 | for (c = 0; c < vd->vdev_children; c++) | |
1102 | vdev_label_sync(zio, vd->vdev_child[c], l, txg, flags); | |
1103 | ||
1104 | if (!vd->vdev_ops->vdev_op_leaf) | |
1105 | return; | |
1106 | ||
1107 | if (!vdev_writeable(vd)) | |
1108 | return; | |
1109 | ||
1110 | /* | |
1111 | * Generate a label describing the top-level config to which we belong. | |
1112 | */ | |
1113 | label = spa_config_generate(vd->vdev_spa, vd, txg, B_FALSE); | |
1114 | ||
1115 | vp = zio_buf_alloc(sizeof (vdev_phys_t)); | |
1116 | bzero(vp, sizeof (vdev_phys_t)); | |
1117 | ||
1118 | buf = vp->vp_nvlist; | |
1119 | buflen = sizeof (vp->vp_nvlist); | |
1120 | ||
1121 | if (!nvlist_pack(label, &buf, &buflen, NV_ENCODE_XDR, KM_SLEEP)) { | |
1122 | for (; l < VDEV_LABELS; l += 2) { | |
1123 | vdev_label_write(zio, vd, l, vp, | |
1124 | offsetof(vdev_label_t, vl_vdev_phys), | |
1125 | sizeof (vdev_phys_t), | |
1126 | vdev_label_sync_done, zio->io_private, | |
1127 | flags | ZIO_FLAG_DONT_PROPAGATE); | |
1128 | } | |
1129 | } | |
1130 | ||
1131 | zio_buf_free(vp, sizeof (vdev_phys_t)); | |
1132 | nvlist_free(label); | |
1133 | } | |
1134 | ||
1135 | int | |
1136 | vdev_label_sync_list(spa_t *spa, int l, uint64_t txg, int flags) | |
1137 | { | |
1138 | list_t *dl = &spa->spa_config_dirty_list; | |
1139 | vdev_t *vd; | |
1140 | zio_t *zio; | |
1141 | int error; | |
1142 | ||
1143 | /* | |
1144 | * Write the new labels to disk. | |
1145 | */ | |
1146 | zio = zio_root(spa, NULL, NULL, flags); | |
1147 | ||
1148 | for (vd = list_head(dl); vd != NULL; vd = list_next(dl, vd)) { | |
1149 | uint64_t *good_writes; | |
1150 | zio_t *vio; | |
1151 | ||
1152 | ASSERT(!vd->vdev_ishole); | |
1153 | ||
1154 | good_writes = kmem_zalloc(sizeof (uint64_t), KM_SLEEP); | |
1155 | vio = zio_null(zio, spa, NULL, | |
1156 | (vd->vdev_islog || vd->vdev_aux != NULL) ? | |
1157 | vdev_label_sync_ignore_done : vdev_label_sync_top_done, | |
1158 | good_writes, flags); | |
1159 | vdev_label_sync(vio, vd, l, txg, flags); | |
1160 | zio_nowait(vio); | |
1161 | } | |
1162 | ||
1163 | error = zio_wait(zio); | |
1164 | ||
1165 | /* | |
1166 | * Flush the new labels to disk. | |
1167 | */ | |
1168 | zio = zio_root(spa, NULL, NULL, flags); | |
1169 | ||
1170 | for (vd = list_head(dl); vd != NULL; vd = list_next(dl, vd)) | |
1171 | zio_flush(zio, vd); | |
1172 | ||
1173 | (void) zio_wait(zio); | |
1174 | ||
1175 | return (error); | |
1176 | } | |
1177 | ||
1178 | /* | |
1179 | * Sync the uberblock and any changes to the vdev configuration. | |
1180 | * | |
1181 | * The order of operations is carefully crafted to ensure that | |
1182 | * if the system panics or loses power at any time, the state on disk | |
1183 | * is still transactionally consistent. The in-line comments below | |
1184 | * describe the failure semantics at each stage. | |
1185 | * | |
1186 | * Moreover, vdev_config_sync() is designed to be idempotent: if it fails | |
1187 | * at any time, you can just call it again, and it will resume its work. | |
1188 | */ | |
1189 | int | |
1190 | vdev_config_sync(vdev_t **svd, int svdcount, uint64_t txg, boolean_t tryhard) | |
1191 | { | |
1192 | spa_t *spa = svd[0]->vdev_spa; | |
1193 | uberblock_t *ub = &spa->spa_uberblock; | |
1194 | vdev_t *vd; | |
1195 | zio_t *zio; | |
1196 | int error; | |
1197 | int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL; | |
1198 | ||
1199 | /* | |
1200 | * Normally, we don't want to try too hard to write every label and | |
1201 | * uberblock. If there is a flaky disk, we don't want the rest of the | |
1202 | * sync process to block while we retry. But if we can't write a | |
1203 | * single label out, we should retry with ZIO_FLAG_TRYHARD before | |
1204 | * bailing out and declaring the pool faulted. | |
1205 | */ | |
1206 | if (tryhard) | |
1207 | flags |= ZIO_FLAG_TRYHARD; | |
1208 | ||
1209 | ASSERT(ub->ub_txg <= txg); | |
1210 | ||
1211 | /* | |
1212 | * If this isn't a resync due to I/O errors, | |
1213 | * and nothing changed in this transaction group, | |
1214 | * and the vdev configuration hasn't changed, | |
1215 | * then there's nothing to do. | |
1216 | */ | |
1217 | if (ub->ub_txg < txg && | |
1218 | uberblock_update(ub, spa->spa_root_vdev, txg) == B_FALSE && | |
1219 | list_is_empty(&spa->spa_config_dirty_list)) | |
1220 | return (0); | |
1221 | ||
1222 | if (txg > spa_freeze_txg(spa)) | |
1223 | return (0); | |
1224 | ||
1225 | ASSERT(txg <= spa->spa_final_txg); | |
1226 | ||
1227 | /* | |
1228 | * Flush the write cache of every disk that's been written to | |
1229 | * in this transaction group. This ensures that all blocks | |
1230 | * written in this txg will be committed to stable storage | |
1231 | * before any uberblock that references them. | |
1232 | */ | |
1233 | zio = zio_root(spa, NULL, NULL, flags); | |
1234 | ||
1235 | for (vd = txg_list_head(&spa->spa_vdev_txg_list, TXG_CLEAN(txg)); vd; | |
1236 | vd = txg_list_next(&spa->spa_vdev_txg_list, vd, TXG_CLEAN(txg))) | |
1237 | zio_flush(zio, vd); | |
1238 | ||
1239 | (void) zio_wait(zio); | |
1240 | ||
1241 | /* | |
1242 | * Sync out the even labels (L0, L2) for every dirty vdev. If the | |
1243 | * system dies in the middle of this process, that's OK: all of the | |
1244 | * even labels that made it to disk will be newer than any uberblock, | |
1245 | * and will therefore be considered invalid. The odd labels (L1, L3), | |
1246 | * which have not yet been touched, will still be valid. We flush | |
1247 | * the new labels to disk to ensure that all even-label updates | |
1248 | * are committed to stable storage before the uberblock update. | |
1249 | */ | |
1250 | if ((error = vdev_label_sync_list(spa, 0, txg, flags)) != 0) | |
1251 | return (error); | |
1252 | ||
1253 | /* | |
1254 | * Sync the uberblocks to all vdevs in svd[]. | |
1255 | * If the system dies in the middle of this step, there are two cases | |
1256 | * to consider, and the on-disk state is consistent either way: | |
1257 | * | |
1258 | * (1) If none of the new uberblocks made it to disk, then the | |
1259 | * previous uberblock will be the newest, and the odd labels | |
1260 | * (which had not yet been touched) will be valid with respect | |
1261 | * to that uberblock. | |
1262 | * | |
1263 | * (2) If one or more new uberblocks made it to disk, then they | |
1264 | * will be the newest, and the even labels (which had all | |
1265 | * been successfully committed) will be valid with respect | |
1266 | * to the new uberblocks. | |
1267 | */ | |
1268 | if ((error = vdev_uberblock_sync_list(svd, svdcount, ub, flags)) != 0) | |
1269 | return (error); | |
1270 | ||
1271 | /* | |
1272 | * Sync out odd labels for every dirty vdev. If the system dies | |
1273 | * in the middle of this process, the even labels and the new | |
1274 | * uberblocks will suffice to open the pool. The next time | |
1275 | * the pool is opened, the first thing we'll do -- before any | |
1276 | * user data is modified -- is mark every vdev dirty so that | |
1277 | * all labels will be brought up to date. We flush the new labels | |
1278 | * to disk to ensure that all odd-label updates are committed to | |
1279 | * stable storage before the next transaction group begins. | |
1280 | */ | |
1281 | return (vdev_label_sync_list(spa, 1, txg, flags)); | |
1282 | } |