]>
Commit | Line | Data |
---|---|---|
34dc7c2f BB |
1 | /* |
2 | * CDDL HEADER START | |
3 | * | |
4 | * The contents of this file are subject to the terms of the | |
5 | * Common Development and Distribution License (the "License"). | |
6 | * You may not use this file except in compliance with the License. | |
7 | * | |
8 | * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE | |
9 | * or http://www.opensolaris.org/os/licensing. | |
10 | * See the License for the specific language governing permissions | |
11 | * and limitations under the License. | |
12 | * | |
13 | * When distributing Covered Code, include this CDDL HEADER in each | |
14 | * file and include the License file at usr/src/OPENSOLARIS.LICENSE. | |
15 | * If applicable, add the following below this CDDL HEADER, with the | |
16 | * fields enclosed by brackets "[]" replaced with your own identifying | |
17 | * information: Portions Copyright [yyyy] [name of copyright owner] | |
18 | * | |
19 | * CDDL HEADER END | |
20 | */ | |
21 | /* | |
d164b209 | 22 | * Copyright 2009 Sun Microsystems, Inc. All rights reserved. |
34dc7c2f BB |
23 | * Use is subject to license terms. |
24 | */ | |
25 | ||
34dc7c2f BB |
26 | /* |
27 | * Virtual Device Labels | |
28 | * --------------------- | |
29 | * | |
30 | * The vdev label serves several distinct purposes: | |
31 | * | |
32 | * 1. Uniquely identify this device as part of a ZFS pool and confirm its | |
33 | * identity within the pool. | |
34 | * | |
35 | * 2. Verify that all the devices given in a configuration are present | |
36 | * within the pool. | |
37 | * | |
38 | * 3. Determine the uberblock for the pool. | |
39 | * | |
40 | * 4. In case of an import operation, determine the configuration of the | |
41 | * toplevel vdev of which it is a part. | |
42 | * | |
43 | * 5. If an import operation cannot find all the devices in the pool, | |
44 | * provide enough information to the administrator to determine which | |
45 | * devices are missing. | |
46 | * | |
47 | * It is important to note that while the kernel is responsible for writing the | |
48 | * label, it only consumes the information in the first three cases. The | |
49 | * latter information is only consumed in userland when determining the | |
50 | * configuration to import a pool. | |
51 | * | |
52 | * | |
53 | * Label Organization | |
54 | * ------------------ | |
55 | * | |
56 | * Before describing the contents of the label, it's important to understand how | |
57 | * the labels are written and updated with respect to the uberblock. | |
58 | * | |
59 | * When the pool configuration is altered, either because it was newly created | |
60 | * or a device was added, we want to update all the labels such that we can deal | |
61 | * with fatal failure at any point. To this end, each disk has two labels which | |
62 | * are updated before and after the uberblock is synced. Assuming we have | |
63 | * labels and an uberblock with the following transaction groups: | |
64 | * | |
65 | * L1 UB L2 | |
66 | * +------+ +------+ +------+ | |
67 | * | | | | | | | |
68 | * | t10 | | t10 | | t10 | | |
69 | * | | | | | | | |
70 | * +------+ +------+ +------+ | |
71 | * | |
72 | * In this stable state, the labels and the uberblock were all updated within | |
73 | * the same transaction group (10). Each label is mirrored and checksummed, so | |
74 | * that we can detect when we fail partway through writing the label. | |
75 | * | |
76 | * In order to identify which labels are valid, the labels are written in the | |
77 | * following manner: | |
78 | * | |
79 | * 1. For each vdev, update 'L1' to the new label | |
80 | * 2. Update the uberblock | |
81 | * 3. For each vdev, update 'L2' to the new label | |
82 | * | |
83 | * Given arbitrary failure, we can determine the correct label to use based on | |
84 | * the transaction group. If we fail after updating L1 but before updating the | |
85 | * UB, we will notice that L1's transaction group is greater than the uberblock, | |
86 | * so L2 must be valid. If we fail after writing the uberblock but before | |
87 | * writing L2, we will notice that L2's transaction group is less than L1, and | |
88 | * therefore L1 is valid. | |
89 | * | |
90 | * Another added complexity is that not every label is updated when the config | |
91 | * is synced. If we add a single device, we do not want to have to re-write | |
92 | * every label for every device in the pool. This means that both L1 and L2 may | |
93 | * be older than the pool uberblock, because the necessary information is stored | |
94 | * on another vdev. | |
95 | * | |
96 | * | |
97 | * On-disk Format | |
98 | * -------------- | |
99 | * | |
100 | * The vdev label consists of two distinct parts, and is wrapped within the | |
101 | * vdev_label_t structure. The label includes 8k of padding to permit legacy | |
102 | * VTOC disk labels, but is otherwise ignored. | |
103 | * | |
104 | * The first half of the label is a packed nvlist which contains pool wide | |
105 | * properties, per-vdev properties, and configuration information. It is | |
106 | * described in more detail below. | |
107 | * | |
108 | * The latter half of the label consists of a redundant array of uberblocks. | |
109 | * These uberblocks are updated whenever a transaction group is committed, | |
110 | * or when the configuration is updated. When a pool is loaded, we scan each | |
111 | * vdev for the 'best' uberblock. | |
112 | * | |
113 | * | |
114 | * Configuration Information | |
115 | * ------------------------- | |
116 | * | |
117 | * The nvlist describing the pool and vdev contains the following elements: | |
118 | * | |
119 | * version ZFS on-disk version | |
120 | * name Pool name | |
121 | * state Pool state | |
122 | * txg Transaction group in which this label was written | |
123 | * pool_guid Unique identifier for this pool | |
124 | * vdev_tree An nvlist describing vdev tree. | |
125 | * | |
126 | * Each leaf device label also contains the following: | |
127 | * | |
128 | * top_guid Unique ID for top-level vdev in which this is contained | |
129 | * guid Unique ID for the leaf vdev | |
130 | * | |
131 | * The 'vs' configuration follows the format described in 'spa_config.c'. | |
132 | */ | |
133 | ||
134 | #include <sys/zfs_context.h> | |
135 | #include <sys/spa.h> | |
136 | #include <sys/spa_impl.h> | |
137 | #include <sys/dmu.h> | |
138 | #include <sys/zap.h> | |
139 | #include <sys/vdev.h> | |
140 | #include <sys/vdev_impl.h> | |
141 | #include <sys/uberblock_impl.h> | |
142 | #include <sys/metaslab.h> | |
143 | #include <sys/zio.h> | |
144 | #include <sys/fs/zfs.h> | |
145 | ||
146 | /* | |
147 | * Basic routines to read and write from a vdev label. | |
148 | * Used throughout the rest of this file. | |
149 | */ | |
150 | uint64_t | |
151 | vdev_label_offset(uint64_t psize, int l, uint64_t offset) | |
152 | { | |
153 | ASSERT(offset < sizeof (vdev_label_t)); | |
154 | ASSERT(P2PHASE_TYPED(psize, sizeof (vdev_label_t), uint64_t) == 0); | |
155 | ||
156 | return (offset + l * sizeof (vdev_label_t) + (l < VDEV_LABELS / 2 ? | |
157 | 0 : psize - VDEV_LABELS * sizeof (vdev_label_t))); | |
158 | } | |
159 | ||
b128c09f BB |
160 | /* |
161 | * Returns back the vdev label associated with the passed in offset. | |
162 | */ | |
163 | int | |
164 | vdev_label_number(uint64_t psize, uint64_t offset) | |
165 | { | |
166 | int l; | |
167 | ||
168 | if (offset >= psize - VDEV_LABEL_END_SIZE) { | |
169 | offset -= psize - VDEV_LABEL_END_SIZE; | |
170 | offset += (VDEV_LABELS / 2) * sizeof (vdev_label_t); | |
171 | } | |
172 | l = offset / sizeof (vdev_label_t); | |
173 | return (l < VDEV_LABELS ? l : -1); | |
174 | } | |
175 | ||
34dc7c2f BB |
176 | static void |
177 | vdev_label_read(zio_t *zio, vdev_t *vd, int l, void *buf, uint64_t offset, | |
b128c09f | 178 | uint64_t size, zio_done_func_t *done, void *private, int flags) |
34dc7c2f | 179 | { |
b128c09f BB |
180 | ASSERT(spa_config_held(zio->io_spa, SCL_STATE_ALL, RW_WRITER) == |
181 | SCL_STATE_ALL); | |
182 | ASSERT(flags & ZIO_FLAG_CONFIG_WRITER); | |
34dc7c2f BB |
183 | |
184 | zio_nowait(zio_read_phys(zio, vd, | |
185 | vdev_label_offset(vd->vdev_psize, l, offset), | |
186 | size, buf, ZIO_CHECKSUM_LABEL, done, private, | |
b128c09f | 187 | ZIO_PRIORITY_SYNC_READ, flags, B_TRUE)); |
34dc7c2f BB |
188 | } |
189 | ||
190 | static void | |
191 | vdev_label_write(zio_t *zio, vdev_t *vd, int l, void *buf, uint64_t offset, | |
192 | uint64_t size, zio_done_func_t *done, void *private, int flags) | |
193 | { | |
b128c09f BB |
194 | ASSERT(spa_config_held(zio->io_spa, SCL_ALL, RW_WRITER) == SCL_ALL || |
195 | (spa_config_held(zio->io_spa, SCL_CONFIG | SCL_STATE, RW_READER) == | |
196 | (SCL_CONFIG | SCL_STATE) && | |
197 | dsl_pool_sync_context(spa_get_dsl(zio->io_spa)))); | |
198 | ASSERT(flags & ZIO_FLAG_CONFIG_WRITER); | |
34dc7c2f BB |
199 | |
200 | zio_nowait(zio_write_phys(zio, vd, | |
201 | vdev_label_offset(vd->vdev_psize, l, offset), | |
202 | size, buf, ZIO_CHECKSUM_LABEL, done, private, | |
203 | ZIO_PRIORITY_SYNC_WRITE, flags, B_TRUE)); | |
204 | } | |
205 | ||
206 | /* | |
207 | * Generate the nvlist representing this vdev's config. | |
208 | */ | |
209 | nvlist_t * | |
210 | vdev_config_generate(spa_t *spa, vdev_t *vd, boolean_t getstats, | |
211 | boolean_t isspare, boolean_t isl2cache) | |
212 | { | |
213 | nvlist_t *nv = NULL; | |
214 | ||
215 | VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0); | |
216 | ||
217 | VERIFY(nvlist_add_string(nv, ZPOOL_CONFIG_TYPE, | |
218 | vd->vdev_ops->vdev_op_type) == 0); | |
219 | if (!isspare && !isl2cache) | |
220 | VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_ID, vd->vdev_id) | |
221 | == 0); | |
222 | VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_GUID, vd->vdev_guid) == 0); | |
223 | ||
224 | if (vd->vdev_path != NULL) | |
225 | VERIFY(nvlist_add_string(nv, ZPOOL_CONFIG_PATH, | |
226 | vd->vdev_path) == 0); | |
227 | ||
228 | if (vd->vdev_devid != NULL) | |
229 | VERIFY(nvlist_add_string(nv, ZPOOL_CONFIG_DEVID, | |
230 | vd->vdev_devid) == 0); | |
231 | ||
232 | if (vd->vdev_physpath != NULL) | |
233 | VERIFY(nvlist_add_string(nv, ZPOOL_CONFIG_PHYS_PATH, | |
234 | vd->vdev_physpath) == 0); | |
235 | ||
9babb374 BB |
236 | if (vd->vdev_fru != NULL) |
237 | VERIFY(nvlist_add_string(nv, ZPOOL_CONFIG_FRU, | |
238 | vd->vdev_fru) == 0); | |
239 | ||
34dc7c2f BB |
240 | if (vd->vdev_nparity != 0) { |
241 | ASSERT(strcmp(vd->vdev_ops->vdev_op_type, | |
242 | VDEV_TYPE_RAIDZ) == 0); | |
243 | ||
244 | /* | |
245 | * Make sure someone hasn't managed to sneak a fancy new vdev | |
246 | * into a crufty old storage pool. | |
247 | */ | |
248 | ASSERT(vd->vdev_nparity == 1 || | |
45d1cae3 BB |
249 | (vd->vdev_nparity <= 2 && |
250 | spa_version(spa) >= SPA_VERSION_RAIDZ2) || | |
251 | (vd->vdev_nparity <= 3 && | |
252 | spa_version(spa) >= SPA_VERSION_RAIDZ3)); | |
34dc7c2f BB |
253 | |
254 | /* | |
255 | * Note that we'll add the nparity tag even on storage pools | |
256 | * that only support a single parity device -- older software | |
257 | * will just ignore it. | |
258 | */ | |
259 | VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_NPARITY, | |
260 | vd->vdev_nparity) == 0); | |
261 | } | |
262 | ||
263 | if (vd->vdev_wholedisk != -1ULL) | |
264 | VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK, | |
265 | vd->vdev_wholedisk) == 0); | |
266 | ||
267 | if (vd->vdev_not_present) | |
268 | VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_NOT_PRESENT, 1) == 0); | |
269 | ||
270 | if (vd->vdev_isspare) | |
271 | VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_IS_SPARE, 1) == 0); | |
272 | ||
273 | if (!isspare && !isl2cache && vd == vd->vdev_top) { | |
274 | VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_METASLAB_ARRAY, | |
275 | vd->vdev_ms_array) == 0); | |
276 | VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_METASLAB_SHIFT, | |
277 | vd->vdev_ms_shift) == 0); | |
278 | VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_ASHIFT, | |
279 | vd->vdev_ashift) == 0); | |
280 | VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_ASIZE, | |
281 | vd->vdev_asize) == 0); | |
282 | VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_IS_LOG, | |
283 | vd->vdev_islog) == 0); | |
284 | } | |
285 | ||
fb5f0bc8 | 286 | if (vd->vdev_dtl_smo.smo_object != 0) |
34dc7c2f | 287 | VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_DTL, |
fb5f0bc8 | 288 | vd->vdev_dtl_smo.smo_object) == 0); |
34dc7c2f BB |
289 | |
290 | if (getstats) { | |
291 | vdev_stat_t vs; | |
292 | vdev_get_stats(vd, &vs); | |
293 | VERIFY(nvlist_add_uint64_array(nv, ZPOOL_CONFIG_STATS, | |
294 | (uint64_t *)&vs, sizeof (vs) / sizeof (uint64_t)) == 0); | |
295 | } | |
296 | ||
297 | if (!vd->vdev_ops->vdev_op_leaf) { | |
298 | nvlist_t **child; | |
299 | int c; | |
300 | ||
301 | child = kmem_alloc(vd->vdev_children * sizeof (nvlist_t *), | |
302 | KM_SLEEP); | |
303 | ||
304 | for (c = 0; c < vd->vdev_children; c++) | |
305 | child[c] = vdev_config_generate(spa, vd->vdev_child[c], | |
306 | getstats, isspare, isl2cache); | |
307 | ||
308 | VERIFY(nvlist_add_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN, | |
309 | child, vd->vdev_children) == 0); | |
310 | ||
311 | for (c = 0; c < vd->vdev_children; c++) | |
312 | nvlist_free(child[c]); | |
313 | ||
314 | kmem_free(child, vd->vdev_children * sizeof (nvlist_t *)); | |
315 | ||
316 | } else { | |
317 | if (vd->vdev_offline && !vd->vdev_tmpoffline) | |
318 | VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_OFFLINE, | |
319 | B_TRUE) == 0); | |
320 | if (vd->vdev_faulted) | |
321 | VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_FAULTED, | |
322 | B_TRUE) == 0); | |
323 | if (vd->vdev_degraded) | |
324 | VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_DEGRADED, | |
325 | B_TRUE) == 0); | |
326 | if (vd->vdev_removed) | |
327 | VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_REMOVED, | |
328 | B_TRUE) == 0); | |
329 | if (vd->vdev_unspare) | |
330 | VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_UNSPARE, | |
331 | B_TRUE) == 0); | |
332 | } | |
333 | ||
334 | return (nv); | |
335 | } | |
336 | ||
337 | nvlist_t * | |
338 | vdev_label_read_config(vdev_t *vd) | |
339 | { | |
340 | spa_t *spa = vd->vdev_spa; | |
341 | nvlist_t *config = NULL; | |
342 | vdev_phys_t *vp; | |
343 | zio_t *zio; | |
9babb374 BB |
344 | int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL | |
345 | ZIO_FLAG_SPECULATIVE; | |
34dc7c2f | 346 | |
b128c09f | 347 | ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); |
34dc7c2f BB |
348 | |
349 | if (!vdev_readable(vd)) | |
350 | return (NULL); | |
351 | ||
352 | vp = zio_buf_alloc(sizeof (vdev_phys_t)); | |
353 | ||
9babb374 | 354 | retry: |
b128c09f | 355 | for (int l = 0; l < VDEV_LABELS; l++) { |
34dc7c2f | 356 | |
b128c09f | 357 | zio = zio_root(spa, NULL, NULL, flags); |
34dc7c2f BB |
358 | |
359 | vdev_label_read(zio, vd, l, vp, | |
360 | offsetof(vdev_label_t, vl_vdev_phys), | |
b128c09f | 361 | sizeof (vdev_phys_t), NULL, NULL, flags); |
34dc7c2f BB |
362 | |
363 | if (zio_wait(zio) == 0 && | |
364 | nvlist_unpack(vp->vp_nvlist, sizeof (vp->vp_nvlist), | |
365 | &config, 0) == 0) | |
366 | break; | |
367 | ||
368 | if (config != NULL) { | |
369 | nvlist_free(config); | |
370 | config = NULL; | |
371 | } | |
372 | } | |
373 | ||
9babb374 BB |
374 | if (config == NULL && !(flags & ZIO_FLAG_TRYHARD)) { |
375 | flags |= ZIO_FLAG_TRYHARD; | |
376 | goto retry; | |
377 | } | |
378 | ||
34dc7c2f BB |
379 | zio_buf_free(vp, sizeof (vdev_phys_t)); |
380 | ||
381 | return (config); | |
382 | } | |
383 | ||
384 | /* | |
385 | * Determine if a device is in use. The 'spare_guid' parameter will be filled | |
386 | * in with the device guid if this spare is active elsewhere on the system. | |
387 | */ | |
388 | static boolean_t | |
389 | vdev_inuse(vdev_t *vd, uint64_t crtxg, vdev_labeltype_t reason, | |
390 | uint64_t *spare_guid, uint64_t *l2cache_guid) | |
391 | { | |
392 | spa_t *spa = vd->vdev_spa; | |
393 | uint64_t state, pool_guid, device_guid, txg, spare_pool; | |
394 | uint64_t vdtxg = 0; | |
395 | nvlist_t *label; | |
396 | ||
397 | if (spare_guid) | |
398 | *spare_guid = 0ULL; | |
399 | if (l2cache_guid) | |
400 | *l2cache_guid = 0ULL; | |
401 | ||
402 | /* | |
403 | * Read the label, if any, and perform some basic sanity checks. | |
404 | */ | |
405 | if ((label = vdev_label_read_config(vd)) == NULL) | |
406 | return (B_FALSE); | |
407 | ||
408 | (void) nvlist_lookup_uint64(label, ZPOOL_CONFIG_CREATE_TXG, | |
409 | &vdtxg); | |
410 | ||
411 | if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE, | |
412 | &state) != 0 || | |
413 | nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, | |
414 | &device_guid) != 0) { | |
415 | nvlist_free(label); | |
416 | return (B_FALSE); | |
417 | } | |
418 | ||
419 | if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE && | |
420 | (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_GUID, | |
421 | &pool_guid) != 0 || | |
422 | nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG, | |
423 | &txg) != 0)) { | |
424 | nvlist_free(label); | |
425 | return (B_FALSE); | |
426 | } | |
427 | ||
428 | nvlist_free(label); | |
429 | ||
430 | /* | |
431 | * Check to see if this device indeed belongs to the pool it claims to | |
432 | * be a part of. The only way this is allowed is if the device is a hot | |
433 | * spare (which we check for later on). | |
434 | */ | |
435 | if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE && | |
436 | !spa_guid_exists(pool_guid, device_guid) && | |
b128c09f | 437 | !spa_spare_exists(device_guid, NULL, NULL) && |
34dc7c2f BB |
438 | !spa_l2cache_exists(device_guid, NULL)) |
439 | return (B_FALSE); | |
440 | ||
441 | /* | |
442 | * If the transaction group is zero, then this an initialized (but | |
443 | * unused) label. This is only an error if the create transaction | |
444 | * on-disk is the same as the one we're using now, in which case the | |
445 | * user has attempted to add the same vdev multiple times in the same | |
446 | * transaction. | |
447 | */ | |
448 | if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE && | |
449 | txg == 0 && vdtxg == crtxg) | |
450 | return (B_TRUE); | |
451 | ||
452 | /* | |
453 | * Check to see if this is a spare device. We do an explicit check for | |
454 | * spa_has_spare() here because it may be on our pending list of spares | |
455 | * to add. We also check if it is an l2cache device. | |
456 | */ | |
b128c09f | 457 | if (spa_spare_exists(device_guid, &spare_pool, NULL) || |
34dc7c2f BB |
458 | spa_has_spare(spa, device_guid)) { |
459 | if (spare_guid) | |
460 | *spare_guid = device_guid; | |
461 | ||
462 | switch (reason) { | |
463 | case VDEV_LABEL_CREATE: | |
464 | case VDEV_LABEL_L2CACHE: | |
465 | return (B_TRUE); | |
466 | ||
467 | case VDEV_LABEL_REPLACE: | |
468 | return (!spa_has_spare(spa, device_guid) || | |
469 | spare_pool != 0ULL); | |
470 | ||
471 | case VDEV_LABEL_SPARE: | |
472 | return (spa_has_spare(spa, device_guid)); | |
473 | } | |
474 | } | |
475 | ||
476 | /* | |
477 | * Check to see if this is an l2cache device. | |
478 | */ | |
479 | if (spa_l2cache_exists(device_guid, NULL)) | |
480 | return (B_TRUE); | |
481 | ||
482 | /* | |
483 | * If the device is marked ACTIVE, then this device is in use by another | |
484 | * pool on the system. | |
485 | */ | |
486 | return (state == POOL_STATE_ACTIVE); | |
487 | } | |
488 | ||
489 | /* | |
490 | * Initialize a vdev label. We check to make sure each leaf device is not in | |
491 | * use, and writable. We put down an initial label which we will later | |
492 | * overwrite with a complete label. Note that it's important to do this | |
493 | * sequentially, not in parallel, so that we catch cases of multiple use of the | |
494 | * same leaf vdev in the vdev we're creating -- e.g. mirroring a disk with | |
495 | * itself. | |
496 | */ | |
497 | int | |
498 | vdev_label_init(vdev_t *vd, uint64_t crtxg, vdev_labeltype_t reason) | |
499 | { | |
500 | spa_t *spa = vd->vdev_spa; | |
501 | nvlist_t *label; | |
502 | vdev_phys_t *vp; | |
9babb374 | 503 | char *pad2; |
34dc7c2f BB |
504 | uberblock_t *ub; |
505 | zio_t *zio; | |
34dc7c2f BB |
506 | char *buf; |
507 | size_t buflen; | |
508 | int error; | |
509 | uint64_t spare_guid, l2cache_guid; | |
b128c09f | 510 | int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL; |
34dc7c2f | 511 | |
b128c09f | 512 | ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); |
34dc7c2f | 513 | |
b128c09f | 514 | for (int c = 0; c < vd->vdev_children; c++) |
34dc7c2f BB |
515 | if ((error = vdev_label_init(vd->vdev_child[c], |
516 | crtxg, reason)) != 0) | |
517 | return (error); | |
518 | ||
519 | if (!vd->vdev_ops->vdev_op_leaf) | |
520 | return (0); | |
521 | ||
522 | /* | |
523 | * Dead vdevs cannot be initialized. | |
524 | */ | |
525 | if (vdev_is_dead(vd)) | |
526 | return (EIO); | |
527 | ||
528 | /* | |
529 | * Determine if the vdev is in use. | |
530 | */ | |
531 | if (reason != VDEV_LABEL_REMOVE && | |
532 | vdev_inuse(vd, crtxg, reason, &spare_guid, &l2cache_guid)) | |
533 | return (EBUSY); | |
534 | ||
34dc7c2f BB |
535 | /* |
536 | * If this is a request to add or replace a spare or l2cache device | |
537 | * that is in use elsewhere on the system, then we must update the | |
538 | * guid (which was initialized to a random value) to reflect the | |
539 | * actual GUID (which is shared between multiple pools). | |
540 | */ | |
541 | if (reason != VDEV_LABEL_REMOVE && reason != VDEV_LABEL_L2CACHE && | |
542 | spare_guid != 0ULL) { | |
b128c09f | 543 | uint64_t guid_delta = spare_guid - vd->vdev_guid; |
34dc7c2f | 544 | |
b128c09f | 545 | vd->vdev_guid += guid_delta; |
34dc7c2f | 546 | |
b128c09f BB |
547 | for (vdev_t *pvd = vd; pvd != NULL; pvd = pvd->vdev_parent) |
548 | pvd->vdev_guid_sum += guid_delta; | |
34dc7c2f BB |
549 | |
550 | /* | |
551 | * If this is a replacement, then we want to fallthrough to the | |
552 | * rest of the code. If we're adding a spare, then it's already | |
553 | * labeled appropriately and we can just return. | |
554 | */ | |
555 | if (reason == VDEV_LABEL_SPARE) | |
556 | return (0); | |
557 | ASSERT(reason == VDEV_LABEL_REPLACE); | |
558 | } | |
559 | ||
560 | if (reason != VDEV_LABEL_REMOVE && reason != VDEV_LABEL_SPARE && | |
561 | l2cache_guid != 0ULL) { | |
b128c09f | 562 | uint64_t guid_delta = l2cache_guid - vd->vdev_guid; |
34dc7c2f | 563 | |
b128c09f | 564 | vd->vdev_guid += guid_delta; |
34dc7c2f | 565 | |
b128c09f BB |
566 | for (vdev_t *pvd = vd; pvd != NULL; pvd = pvd->vdev_parent) |
567 | pvd->vdev_guid_sum += guid_delta; | |
34dc7c2f BB |
568 | |
569 | /* | |
570 | * If this is a replacement, then we want to fallthrough to the | |
571 | * rest of the code. If we're adding an l2cache, then it's | |
572 | * already labeled appropriately and we can just return. | |
573 | */ | |
574 | if (reason == VDEV_LABEL_L2CACHE) | |
575 | return (0); | |
576 | ASSERT(reason == VDEV_LABEL_REPLACE); | |
577 | } | |
578 | ||
579 | /* | |
580 | * Initialize its label. | |
581 | */ | |
582 | vp = zio_buf_alloc(sizeof (vdev_phys_t)); | |
583 | bzero(vp, sizeof (vdev_phys_t)); | |
584 | ||
585 | /* | |
586 | * Generate a label describing the pool and our top-level vdev. | |
587 | * We mark it as being from txg 0 to indicate that it's not | |
588 | * really part of an active pool just yet. The labels will | |
589 | * be written again with a meaningful txg by spa_sync(). | |
590 | */ | |
591 | if (reason == VDEV_LABEL_SPARE || | |
592 | (reason == VDEV_LABEL_REMOVE && vd->vdev_isspare)) { | |
593 | /* | |
594 | * For inactive hot spares, we generate a special label that | |
595 | * identifies as a mutually shared hot spare. We write the | |
596 | * label if we are adding a hot spare, or if we are removing an | |
597 | * active hot spare (in which case we want to revert the | |
598 | * labels). | |
599 | */ | |
600 | VERIFY(nvlist_alloc(&label, NV_UNIQUE_NAME, KM_SLEEP) == 0); | |
601 | ||
602 | VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_VERSION, | |
603 | spa_version(spa)) == 0); | |
604 | VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_POOL_STATE, | |
605 | POOL_STATE_SPARE) == 0); | |
606 | VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_GUID, | |
607 | vd->vdev_guid) == 0); | |
608 | } else if (reason == VDEV_LABEL_L2CACHE || | |
609 | (reason == VDEV_LABEL_REMOVE && vd->vdev_isl2cache)) { | |
610 | /* | |
611 | * For level 2 ARC devices, add a special label. | |
612 | */ | |
613 | VERIFY(nvlist_alloc(&label, NV_UNIQUE_NAME, KM_SLEEP) == 0); | |
614 | ||
615 | VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_VERSION, | |
616 | spa_version(spa)) == 0); | |
617 | VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_POOL_STATE, | |
618 | POOL_STATE_L2CACHE) == 0); | |
619 | VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_GUID, | |
620 | vd->vdev_guid) == 0); | |
621 | } else { | |
622 | label = spa_config_generate(spa, vd, 0ULL, B_FALSE); | |
623 | ||
624 | /* | |
625 | * Add our creation time. This allows us to detect multiple | |
626 | * vdev uses as described above, and automatically expires if we | |
627 | * fail. | |
628 | */ | |
629 | VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_CREATE_TXG, | |
630 | crtxg) == 0); | |
631 | } | |
632 | ||
633 | buf = vp->vp_nvlist; | |
634 | buflen = sizeof (vp->vp_nvlist); | |
635 | ||
636 | error = nvlist_pack(label, &buf, &buflen, NV_ENCODE_XDR, KM_SLEEP); | |
637 | if (error != 0) { | |
638 | nvlist_free(label); | |
639 | zio_buf_free(vp, sizeof (vdev_phys_t)); | |
640 | /* EFAULT means nvlist_pack ran out of room */ | |
641 | return (error == EFAULT ? ENAMETOOLONG : EINVAL); | |
642 | } | |
643 | ||
34dc7c2f BB |
644 | /* |
645 | * Initialize uberblock template. | |
646 | */ | |
45d1cae3 BB |
647 | ub = zio_buf_alloc(VDEV_UBERBLOCK_RING); |
648 | bzero(ub, VDEV_UBERBLOCK_RING); | |
34dc7c2f BB |
649 | *ub = spa->spa_uberblock; |
650 | ub->ub_txg = 0; | |
651 | ||
9babb374 BB |
652 | /* Initialize the 2nd padding area. */ |
653 | pad2 = zio_buf_alloc(VDEV_PAD_SIZE); | |
654 | bzero(pad2, VDEV_PAD_SIZE); | |
655 | ||
34dc7c2f BB |
656 | /* |
657 | * Write everything in parallel. | |
658 | */ | |
9babb374 | 659 | retry: |
34dc7c2f BB |
660 | zio = zio_root(spa, NULL, NULL, flags); |
661 | ||
b128c09f | 662 | for (int l = 0; l < VDEV_LABELS; l++) { |
34dc7c2f BB |
663 | |
664 | vdev_label_write(zio, vd, l, vp, | |
665 | offsetof(vdev_label_t, vl_vdev_phys), | |
666 | sizeof (vdev_phys_t), NULL, NULL, flags); | |
667 | ||
9babb374 BB |
668 | /* |
669 | * Skip the 1st padding area. | |
670 | * Zero out the 2nd padding area where it might have | |
671 | * left over data from previous filesystem format. | |
672 | */ | |
673 | vdev_label_write(zio, vd, l, pad2, | |
674 | offsetof(vdev_label_t, vl_pad2), | |
675 | VDEV_PAD_SIZE, NULL, NULL, flags); | |
34dc7c2f | 676 | |
45d1cae3 BB |
677 | vdev_label_write(zio, vd, l, ub, |
678 | offsetof(vdev_label_t, vl_uberblock), | |
679 | VDEV_UBERBLOCK_RING, NULL, NULL, flags); | |
34dc7c2f BB |
680 | } |
681 | ||
682 | error = zio_wait(zio); | |
683 | ||
9babb374 BB |
684 | if (error != 0 && !(flags & ZIO_FLAG_TRYHARD)) { |
685 | flags |= ZIO_FLAG_TRYHARD; | |
686 | goto retry; | |
687 | } | |
688 | ||
34dc7c2f | 689 | nvlist_free(label); |
9babb374 | 690 | zio_buf_free(pad2, VDEV_PAD_SIZE); |
45d1cae3 | 691 | zio_buf_free(ub, VDEV_UBERBLOCK_RING); |
34dc7c2f BB |
692 | zio_buf_free(vp, sizeof (vdev_phys_t)); |
693 | ||
694 | /* | |
695 | * If this vdev hasn't been previously identified as a spare, then we | |
696 | * mark it as such only if a) we are labeling it as a spare, or b) it | |
697 | * exists as a spare elsewhere in the system. Do the same for | |
698 | * level 2 ARC devices. | |
699 | */ | |
700 | if (error == 0 && !vd->vdev_isspare && | |
701 | (reason == VDEV_LABEL_SPARE || | |
b128c09f | 702 | spa_spare_exists(vd->vdev_guid, NULL, NULL))) |
34dc7c2f BB |
703 | spa_spare_add(vd); |
704 | ||
705 | if (error == 0 && !vd->vdev_isl2cache && | |
706 | (reason == VDEV_LABEL_L2CACHE || | |
707 | spa_l2cache_exists(vd->vdev_guid, NULL))) | |
708 | spa_l2cache_add(vd); | |
709 | ||
710 | return (error); | |
711 | } | |
712 | ||
713 | /* | |
714 | * ========================================================================== | |
715 | * uberblock load/sync | |
716 | * ========================================================================== | |
717 | */ | |
718 | ||
fb5f0bc8 BB |
719 | /* |
720 | * For use by zdb and debugging purposes only | |
721 | */ | |
722 | uint64_t ub_max_txg = UINT64_MAX; | |
723 | ||
34dc7c2f BB |
724 | /* |
725 | * Consider the following situation: txg is safely synced to disk. We've | |
726 | * written the first uberblock for txg + 1, and then we lose power. When we | |
727 | * come back up, we fail to see the uberblock for txg + 1 because, say, | |
728 | * it was on a mirrored device and the replica to which we wrote txg + 1 | |
729 | * is now offline. If we then make some changes and sync txg + 1, and then | |
730 | * the missing replica comes back, then for a new seconds we'll have two | |
731 | * conflicting uberblocks on disk with the same txg. The solution is simple: | |
732 | * among uberblocks with equal txg, choose the one with the latest timestamp. | |
733 | */ | |
734 | static int | |
735 | vdev_uberblock_compare(uberblock_t *ub1, uberblock_t *ub2) | |
736 | { | |
737 | if (ub1->ub_txg < ub2->ub_txg) | |
738 | return (-1); | |
739 | if (ub1->ub_txg > ub2->ub_txg) | |
740 | return (1); | |
741 | ||
742 | if (ub1->ub_timestamp < ub2->ub_timestamp) | |
743 | return (-1); | |
744 | if (ub1->ub_timestamp > ub2->ub_timestamp) | |
745 | return (1); | |
746 | ||
747 | return (0); | |
748 | } | |
749 | ||
750 | static void | |
751 | vdev_uberblock_load_done(zio_t *zio) | |
752 | { | |
b128c09f | 753 | zio_t *rio = zio->io_private; |
34dc7c2f | 754 | uberblock_t *ub = zio->io_data; |
b128c09f | 755 | uberblock_t *ubbest = rio->io_private; |
34dc7c2f BB |
756 | |
757 | ASSERT3U(zio->io_size, ==, VDEV_UBERBLOCK_SIZE(zio->io_vd)); | |
758 | ||
759 | if (zio->io_error == 0 && uberblock_verify(ub) == 0) { | |
b128c09f | 760 | mutex_enter(&rio->io_lock); |
fb5f0bc8 BB |
761 | if (ub->ub_txg <= ub_max_txg && |
762 | vdev_uberblock_compare(ub, ubbest) > 0) | |
34dc7c2f | 763 | *ubbest = *ub; |
b128c09f | 764 | mutex_exit(&rio->io_lock); |
34dc7c2f BB |
765 | } |
766 | ||
767 | zio_buf_free(zio->io_data, zio->io_size); | |
768 | } | |
769 | ||
770 | void | |
771 | vdev_uberblock_load(zio_t *zio, vdev_t *vd, uberblock_t *ubbest) | |
772 | { | |
b128c09f BB |
773 | spa_t *spa = vd->vdev_spa; |
774 | vdev_t *rvd = spa->spa_root_vdev; | |
9babb374 BB |
775 | int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL | |
776 | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_TRYHARD; | |
b128c09f BB |
777 | |
778 | if (vd == rvd) { | |
779 | ASSERT(zio == NULL); | |
780 | spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); | |
781 | zio = zio_root(spa, NULL, ubbest, flags); | |
782 | bzero(ubbest, sizeof (uberblock_t)); | |
783 | } | |
34dc7c2f | 784 | |
b128c09f | 785 | ASSERT(zio != NULL); |
34dc7c2f | 786 | |
b128c09f BB |
787 | for (int c = 0; c < vd->vdev_children; c++) |
788 | vdev_uberblock_load(zio, vd->vdev_child[c], ubbest); | |
34dc7c2f | 789 | |
b128c09f BB |
790 | if (vd->vdev_ops->vdev_op_leaf && vdev_readable(vd)) { |
791 | for (int l = 0; l < VDEV_LABELS; l++) { | |
792 | for (int n = 0; n < VDEV_UBERBLOCK_COUNT(vd); n++) { | |
793 | vdev_label_read(zio, vd, l, | |
794 | zio_buf_alloc(VDEV_UBERBLOCK_SIZE(vd)), | |
795 | VDEV_UBERBLOCK_OFFSET(vd, n), | |
796 | VDEV_UBERBLOCK_SIZE(vd), | |
797 | vdev_uberblock_load_done, zio, flags); | |
798 | } | |
34dc7c2f BB |
799 | } |
800 | } | |
b128c09f BB |
801 | |
802 | if (vd == rvd) { | |
803 | (void) zio_wait(zio); | |
804 | spa_config_exit(spa, SCL_ALL, FTAG); | |
805 | } | |
34dc7c2f BB |
806 | } |
807 | ||
808 | /* | |
809 | * On success, increment root zio's count of good writes. | |
810 | * We only get credit for writes to known-visible vdevs; see spa_vdev_add(). | |
811 | */ | |
812 | static void | |
813 | vdev_uberblock_sync_done(zio_t *zio) | |
814 | { | |
815 | uint64_t *good_writes = zio->io_private; | |
816 | ||
817 | if (zio->io_error == 0 && zio->io_vd->vdev_top->vdev_ms_array != 0) | |
818 | atomic_add_64(good_writes, 1); | |
819 | } | |
820 | ||
821 | /* | |
822 | * Write the uberblock to all labels of all leaves of the specified vdev. | |
823 | */ | |
824 | static void | |
b128c09f | 825 | vdev_uberblock_sync(zio_t *zio, uberblock_t *ub, vdev_t *vd, int flags) |
34dc7c2f | 826 | { |
34dc7c2f | 827 | uberblock_t *ubbuf; |
b128c09f | 828 | int n; |
34dc7c2f | 829 | |
b128c09f BB |
830 | for (int c = 0; c < vd->vdev_children; c++) |
831 | vdev_uberblock_sync(zio, ub, vd->vdev_child[c], flags); | |
34dc7c2f BB |
832 | |
833 | if (!vd->vdev_ops->vdev_op_leaf) | |
834 | return; | |
835 | ||
b128c09f | 836 | if (!vdev_writeable(vd)) |
34dc7c2f BB |
837 | return; |
838 | ||
839 | n = ub->ub_txg & (VDEV_UBERBLOCK_COUNT(vd) - 1); | |
840 | ||
841 | ubbuf = zio_buf_alloc(VDEV_UBERBLOCK_SIZE(vd)); | |
842 | bzero(ubbuf, VDEV_UBERBLOCK_SIZE(vd)); | |
843 | *ubbuf = *ub; | |
844 | ||
b128c09f | 845 | for (int l = 0; l < VDEV_LABELS; l++) |
34dc7c2f | 846 | vdev_label_write(zio, vd, l, ubbuf, |
b128c09f | 847 | VDEV_UBERBLOCK_OFFSET(vd, n), VDEV_UBERBLOCK_SIZE(vd), |
34dc7c2f | 848 | vdev_uberblock_sync_done, zio->io_private, |
b128c09f | 849 | flags | ZIO_FLAG_DONT_PROPAGATE); |
34dc7c2f BB |
850 | |
851 | zio_buf_free(ubbuf, VDEV_UBERBLOCK_SIZE(vd)); | |
852 | } | |
853 | ||
854 | int | |
855 | vdev_uberblock_sync_list(vdev_t **svd, int svdcount, uberblock_t *ub, int flags) | |
856 | { | |
857 | spa_t *spa = svd[0]->vdev_spa; | |
34dc7c2f BB |
858 | zio_t *zio; |
859 | uint64_t good_writes = 0; | |
860 | ||
861 | zio = zio_root(spa, NULL, &good_writes, flags); | |
862 | ||
b128c09f BB |
863 | for (int v = 0; v < svdcount; v++) |
864 | vdev_uberblock_sync(zio, ub, svd[v], flags); | |
34dc7c2f BB |
865 | |
866 | (void) zio_wait(zio); | |
867 | ||
868 | /* | |
869 | * Flush the uberblocks to disk. This ensures that the odd labels | |
870 | * are no longer needed (because the new uberblocks and the even | |
871 | * labels are safely on disk), so it is safe to overwrite them. | |
872 | */ | |
873 | zio = zio_root(spa, NULL, NULL, flags); | |
874 | ||
b128c09f | 875 | for (int v = 0; v < svdcount; v++) |
34dc7c2f BB |
876 | zio_flush(zio, svd[v]); |
877 | ||
878 | (void) zio_wait(zio); | |
879 | ||
880 | return (good_writes >= 1 ? 0 : EIO); | |
881 | } | |
882 | ||
883 | /* | |
884 | * On success, increment the count of good writes for our top-level vdev. | |
885 | */ | |
886 | static void | |
887 | vdev_label_sync_done(zio_t *zio) | |
888 | { | |
889 | uint64_t *good_writes = zio->io_private; | |
890 | ||
891 | if (zio->io_error == 0) | |
892 | atomic_add_64(good_writes, 1); | |
893 | } | |
894 | ||
895 | /* | |
896 | * If there weren't enough good writes, indicate failure to the parent. | |
897 | */ | |
898 | static void | |
899 | vdev_label_sync_top_done(zio_t *zio) | |
900 | { | |
901 | uint64_t *good_writes = zio->io_private; | |
902 | ||
903 | if (*good_writes == 0) | |
904 | zio->io_error = EIO; | |
905 | ||
906 | kmem_free(good_writes, sizeof (uint64_t)); | |
907 | } | |
908 | ||
b128c09f BB |
909 | /* |
910 | * We ignore errors for log and cache devices, simply free the private data. | |
911 | */ | |
912 | static void | |
913 | vdev_label_sync_ignore_done(zio_t *zio) | |
914 | { | |
915 | kmem_free(zio->io_private, sizeof (uint64_t)); | |
916 | } | |
917 | ||
34dc7c2f BB |
918 | /* |
919 | * Write all even or odd labels to all leaves of the specified vdev. | |
920 | */ | |
921 | static void | |
b128c09f | 922 | vdev_label_sync(zio_t *zio, vdev_t *vd, int l, uint64_t txg, int flags) |
34dc7c2f BB |
923 | { |
924 | nvlist_t *label; | |
925 | vdev_phys_t *vp; | |
926 | char *buf; | |
927 | size_t buflen; | |
34dc7c2f | 928 | |
b128c09f BB |
929 | for (int c = 0; c < vd->vdev_children; c++) |
930 | vdev_label_sync(zio, vd->vdev_child[c], l, txg, flags); | |
34dc7c2f BB |
931 | |
932 | if (!vd->vdev_ops->vdev_op_leaf) | |
933 | return; | |
934 | ||
b128c09f | 935 | if (!vdev_writeable(vd)) |
34dc7c2f BB |
936 | return; |
937 | ||
938 | /* | |
939 | * Generate a label describing the top-level config to which we belong. | |
940 | */ | |
941 | label = spa_config_generate(vd->vdev_spa, vd, txg, B_FALSE); | |
942 | ||
943 | vp = zio_buf_alloc(sizeof (vdev_phys_t)); | |
944 | bzero(vp, sizeof (vdev_phys_t)); | |
945 | ||
946 | buf = vp->vp_nvlist; | |
947 | buflen = sizeof (vp->vp_nvlist); | |
948 | ||
949 | if (nvlist_pack(label, &buf, &buflen, NV_ENCODE_XDR, KM_SLEEP) == 0) { | |
950 | for (; l < VDEV_LABELS; l += 2) { | |
951 | vdev_label_write(zio, vd, l, vp, | |
952 | offsetof(vdev_label_t, vl_vdev_phys), | |
953 | sizeof (vdev_phys_t), | |
954 | vdev_label_sync_done, zio->io_private, | |
b128c09f | 955 | flags | ZIO_FLAG_DONT_PROPAGATE); |
34dc7c2f BB |
956 | } |
957 | } | |
958 | ||
959 | zio_buf_free(vp, sizeof (vdev_phys_t)); | |
960 | nvlist_free(label); | |
961 | } | |
962 | ||
963 | int | |
b128c09f | 964 | vdev_label_sync_list(spa_t *spa, int l, uint64_t txg, int flags) |
34dc7c2f | 965 | { |
b128c09f | 966 | list_t *dl = &spa->spa_config_dirty_list; |
34dc7c2f BB |
967 | vdev_t *vd; |
968 | zio_t *zio; | |
969 | int error; | |
970 | ||
971 | /* | |
972 | * Write the new labels to disk. | |
973 | */ | |
974 | zio = zio_root(spa, NULL, NULL, flags); | |
975 | ||
976 | for (vd = list_head(dl); vd != NULL; vd = list_next(dl, vd)) { | |
977 | uint64_t *good_writes = kmem_zalloc(sizeof (uint64_t), | |
978 | KM_SLEEP); | |
d164b209 | 979 | zio_t *vio = zio_null(zio, spa, NULL, |
b128c09f BB |
980 | (vd->vdev_islog || vd->vdev_aux != NULL) ? |
981 | vdev_label_sync_ignore_done : vdev_label_sync_top_done, | |
34dc7c2f | 982 | good_writes, flags); |
b128c09f | 983 | vdev_label_sync(vio, vd, l, txg, flags); |
34dc7c2f BB |
984 | zio_nowait(vio); |
985 | } | |
986 | ||
987 | error = zio_wait(zio); | |
988 | ||
989 | /* | |
990 | * Flush the new labels to disk. | |
991 | */ | |
992 | zio = zio_root(spa, NULL, NULL, flags); | |
993 | ||
994 | for (vd = list_head(dl); vd != NULL; vd = list_next(dl, vd)) | |
995 | zio_flush(zio, vd); | |
996 | ||
997 | (void) zio_wait(zio); | |
998 | ||
999 | return (error); | |
1000 | } | |
1001 | ||
1002 | /* | |
1003 | * Sync the uberblock and any changes to the vdev configuration. | |
1004 | * | |
1005 | * The order of operations is carefully crafted to ensure that | |
1006 | * if the system panics or loses power at any time, the state on disk | |
1007 | * is still transactionally consistent. The in-line comments below | |
1008 | * describe the failure semantics at each stage. | |
1009 | * | |
1010 | * Moreover, vdev_config_sync() is designed to be idempotent: if it fails | |
1011 | * at any time, you can just call it again, and it will resume its work. | |
1012 | */ | |
1013 | int | |
9babb374 | 1014 | vdev_config_sync(vdev_t **svd, int svdcount, uint64_t txg, boolean_t tryhard) |
34dc7c2f BB |
1015 | { |
1016 | spa_t *spa = svd[0]->vdev_spa; | |
1017 | uberblock_t *ub = &spa->spa_uberblock; | |
1018 | vdev_t *vd; | |
1019 | zio_t *zio; | |
1020 | int error; | |
b128c09f | 1021 | int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL; |
34dc7c2f | 1022 | |
9babb374 BB |
1023 | /* |
1024 | * Normally, we don't want to try too hard to write every label and | |
1025 | * uberblock. If there is a flaky disk, we don't want the rest of the | |
1026 | * sync process to block while we retry. But if we can't write a | |
1027 | * single label out, we should retry with ZIO_FLAG_TRYHARD before | |
1028 | * bailing out and declaring the pool faulted. | |
1029 | */ | |
1030 | if (tryhard) | |
1031 | flags |= ZIO_FLAG_TRYHARD; | |
1032 | ||
34dc7c2f BB |
1033 | ASSERT(ub->ub_txg <= txg); |
1034 | ||
1035 | /* | |
1036 | * If this isn't a resync due to I/O errors, | |
1037 | * and nothing changed in this transaction group, | |
1038 | * and the vdev configuration hasn't changed, | |
1039 | * then there's nothing to do. | |
1040 | */ | |
1041 | if (ub->ub_txg < txg && | |
1042 | uberblock_update(ub, spa->spa_root_vdev, txg) == B_FALSE && | |
b128c09f | 1043 | list_is_empty(&spa->spa_config_dirty_list)) |
34dc7c2f BB |
1044 | return (0); |
1045 | ||
1046 | if (txg > spa_freeze_txg(spa)) | |
1047 | return (0); | |
1048 | ||
1049 | ASSERT(txg <= spa->spa_final_txg); | |
1050 | ||
1051 | /* | |
1052 | * Flush the write cache of every disk that's been written to | |
1053 | * in this transaction group. This ensures that all blocks | |
1054 | * written in this txg will be committed to stable storage | |
1055 | * before any uberblock that references them. | |
1056 | */ | |
1057 | zio = zio_root(spa, NULL, NULL, flags); | |
1058 | ||
1059 | for (vd = txg_list_head(&spa->spa_vdev_txg_list, TXG_CLEAN(txg)); vd; | |
1060 | vd = txg_list_next(&spa->spa_vdev_txg_list, vd, TXG_CLEAN(txg))) | |
1061 | zio_flush(zio, vd); | |
1062 | ||
1063 | (void) zio_wait(zio); | |
1064 | ||
1065 | /* | |
1066 | * Sync out the even labels (L0, L2) for every dirty vdev. If the | |
1067 | * system dies in the middle of this process, that's OK: all of the | |
1068 | * even labels that made it to disk will be newer than any uberblock, | |
1069 | * and will therefore be considered invalid. The odd labels (L1, L3), | |
1070 | * which have not yet been touched, will still be valid. We flush | |
1071 | * the new labels to disk to ensure that all even-label updates | |
1072 | * are committed to stable storage before the uberblock update. | |
1073 | */ | |
b128c09f | 1074 | if ((error = vdev_label_sync_list(spa, 0, txg, flags)) != 0) |
34dc7c2f BB |
1075 | return (error); |
1076 | ||
1077 | /* | |
1078 | * Sync the uberblocks to all vdevs in svd[]. | |
1079 | * If the system dies in the middle of this step, there are two cases | |
1080 | * to consider, and the on-disk state is consistent either way: | |
1081 | * | |
1082 | * (1) If none of the new uberblocks made it to disk, then the | |
1083 | * previous uberblock will be the newest, and the odd labels | |
1084 | * (which had not yet been touched) will be valid with respect | |
1085 | * to that uberblock. | |
1086 | * | |
1087 | * (2) If one or more new uberblocks made it to disk, then they | |
1088 | * will be the newest, and the even labels (which had all | |
1089 | * been successfully committed) will be valid with respect | |
1090 | * to the new uberblocks. | |
1091 | */ | |
1092 | if ((error = vdev_uberblock_sync_list(svd, svdcount, ub, flags)) != 0) | |
1093 | return (error); | |
1094 | ||
1095 | /* | |
1096 | * Sync out odd labels for every dirty vdev. If the system dies | |
1097 | * in the middle of this process, the even labels and the new | |
1098 | * uberblocks will suffice to open the pool. The next time | |
1099 | * the pool is opened, the first thing we'll do -- before any | |
1100 | * user data is modified -- is mark every vdev dirty so that | |
1101 | * all labels will be brought up to date. We flush the new labels | |
1102 | * to disk to ensure that all odd-label updates are committed to | |
1103 | * stable storage before the next transaction group begins. | |
1104 | */ | |
b128c09f | 1105 | return (vdev_label_sync_list(spa, 1, txg, flags)); |
34dc7c2f | 1106 | } |