]>
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 | /* | |
428870ff | 22 | * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. |
34dc7c2f BB |
23 | */ |
24 | ||
34dc7c2f BB |
25 | /* |
26 | * Virtual Device Labels | |
27 | * --------------------- | |
28 | * | |
29 | * The vdev label serves several distinct purposes: | |
30 | * | |
31 | * 1. Uniquely identify this device as part of a ZFS pool and confirm its | |
32 | * identity within the pool. | |
33 | * | |
34 | * 2. Verify that all the devices given in a configuration are present | |
35 | * within the pool. | |
36 | * | |
37 | * 3. Determine the uberblock for the pool. | |
38 | * | |
39 | * 4. In case of an import operation, determine the configuration of the | |
40 | * toplevel vdev of which it is a part. | |
41 | * | |
42 | * 5. If an import operation cannot find all the devices in the pool, | |
43 | * provide enough information to the administrator to determine which | |
44 | * devices are missing. | |
45 | * | |
46 | * It is important to note that while the kernel is responsible for writing the | |
47 | * label, it only consumes the information in the first three cases. The | |
48 | * latter information is only consumed in userland when determining the | |
49 | * configuration to import a pool. | |
50 | * | |
51 | * | |
52 | * Label Organization | |
53 | * ------------------ | |
54 | * | |
55 | * Before describing the contents of the label, it's important to understand how | |
56 | * the labels are written and updated with respect to the uberblock. | |
57 | * | |
58 | * When the pool configuration is altered, either because it was newly created | |
59 | * or a device was added, we want to update all the labels such that we can deal | |
60 | * with fatal failure at any point. To this end, each disk has two labels which | |
61 | * are updated before and after the uberblock is synced. Assuming we have | |
62 | * labels and an uberblock with the following transaction groups: | |
63 | * | |
64 | * L1 UB L2 | |
65 | * +------+ +------+ +------+ | |
66 | * | | | | | | | |
67 | * | t10 | | t10 | | t10 | | |
68 | * | | | | | | | |
69 | * +------+ +------+ +------+ | |
70 | * | |
71 | * In this stable state, the labels and the uberblock were all updated within | |
72 | * the same transaction group (10). Each label is mirrored and checksummed, so | |
73 | * that we can detect when we fail partway through writing the label. | |
74 | * | |
75 | * In order to identify which labels are valid, the labels are written in the | |
76 | * following manner: | |
77 | * | |
78 | * 1. For each vdev, update 'L1' to the new label | |
79 | * 2. Update the uberblock | |
80 | * 3. For each vdev, update 'L2' to the new label | |
81 | * | |
82 | * Given arbitrary failure, we can determine the correct label to use based on | |
83 | * the transaction group. If we fail after updating L1 but before updating the | |
84 | * UB, we will notice that L1's transaction group is greater than the uberblock, | |
85 | * so L2 must be valid. If we fail after writing the uberblock but before | |
86 | * writing L2, we will notice that L2's transaction group is less than L1, and | |
87 | * therefore L1 is valid. | |
88 | * | |
89 | * Another added complexity is that not every label is updated when the config | |
90 | * is synced. If we add a single device, we do not want to have to re-write | |
91 | * every label for every device in the pool. This means that both L1 and L2 may | |
92 | * be older than the pool uberblock, because the necessary information is stored | |
93 | * on another vdev. | |
94 | * | |
95 | * | |
96 | * On-disk Format | |
97 | * -------------- | |
98 | * | |
99 | * The vdev label consists of two distinct parts, and is wrapped within the | |
100 | * vdev_label_t structure. The label includes 8k of padding to permit legacy | |
101 | * VTOC disk labels, but is otherwise ignored. | |
102 | * | |
103 | * The first half of the label is a packed nvlist which contains pool wide | |
104 | * properties, per-vdev properties, and configuration information. It is | |
105 | * described in more detail below. | |
106 | * | |
107 | * The latter half of the label consists of a redundant array of uberblocks. | |
108 | * These uberblocks are updated whenever a transaction group is committed, | |
109 | * or when the configuration is updated. When a pool is loaded, we scan each | |
110 | * vdev for the 'best' uberblock. | |
111 | * | |
112 | * | |
113 | * Configuration Information | |
114 | * ------------------------- | |
115 | * | |
116 | * The nvlist describing the pool and vdev contains the following elements: | |
117 | * | |
118 | * version ZFS on-disk version | |
119 | * name Pool name | |
120 | * state Pool state | |
121 | * txg Transaction group in which this label was written | |
122 | * pool_guid Unique identifier for this pool | |
123 | * vdev_tree An nvlist describing vdev tree. | |
124 | * | |
125 | * Each leaf device label also contains the following: | |
126 | * | |
127 | * top_guid Unique ID for top-level vdev in which this is contained | |
128 | * guid Unique ID for the leaf vdev | |
129 | * | |
130 | * The 'vs' configuration follows the format described in 'spa_config.c'. | |
131 | */ | |
132 | ||
133 | #include <sys/zfs_context.h> | |
134 | #include <sys/spa.h> | |
135 | #include <sys/spa_impl.h> | |
136 | #include <sys/dmu.h> | |
137 | #include <sys/zap.h> | |
138 | #include <sys/vdev.h> | |
139 | #include <sys/vdev_impl.h> | |
140 | #include <sys/uberblock_impl.h> | |
141 | #include <sys/metaslab.h> | |
142 | #include <sys/zio.h> | |
428870ff | 143 | #include <sys/dsl_scan.h> |
34dc7c2f BB |
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, | |
428870ff | 211 | vdev_config_flag_t flags) |
34dc7c2f BB |
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); | |
428870ff | 219 | if (!(flags & (VDEV_CONFIG_SPARE | VDEV_CONFIG_L2CACHE))) |
34dc7c2f BB |
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 | ||
428870ff BB |
273 | if (!(flags & (VDEV_CONFIG_SPARE | VDEV_CONFIG_L2CACHE)) && |
274 | vd == vd->vdev_top) { | |
34dc7c2f BB |
275 | VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_METASLAB_ARRAY, |
276 | vd->vdev_ms_array) == 0); | |
277 | VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_METASLAB_SHIFT, | |
278 | vd->vdev_ms_shift) == 0); | |
279 | VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_ASHIFT, | |
280 | vd->vdev_ashift) == 0); | |
281 | VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_ASIZE, | |
282 | vd->vdev_asize) == 0); | |
283 | VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_IS_LOG, | |
284 | vd->vdev_islog) == 0); | |
428870ff BB |
285 | if (vd->vdev_removing) |
286 | VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_REMOVING, | |
287 | vd->vdev_removing) == 0); | |
34dc7c2f BB |
288 | } |
289 | ||
fb5f0bc8 | 290 | if (vd->vdev_dtl_smo.smo_object != 0) |
34dc7c2f | 291 | VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_DTL, |
fb5f0bc8 | 292 | vd->vdev_dtl_smo.smo_object) == 0); |
34dc7c2f | 293 | |
428870ff BB |
294 | if (vd->vdev_crtxg) |
295 | VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_CREATE_TXG, | |
296 | vd->vdev_crtxg) == 0); | |
297 | ||
34dc7c2f BB |
298 | if (getstats) { |
299 | vdev_stat_t vs; | |
428870ff BB |
300 | pool_scan_stat_t ps; |
301 | ||
34dc7c2f | 302 | vdev_get_stats(vd, &vs); |
428870ff | 303 | VERIFY(nvlist_add_uint64_array(nv, ZPOOL_CONFIG_VDEV_STATS, |
34dc7c2f | 304 | (uint64_t *)&vs, sizeof (vs) / sizeof (uint64_t)) == 0); |
428870ff BB |
305 | |
306 | /* provide either current or previous scan information */ | |
307 | if (spa_scan_get_stats(spa, &ps) == 0) { | |
308 | VERIFY(nvlist_add_uint64_array(nv, | |
309 | ZPOOL_CONFIG_SCAN_STATS, (uint64_t *)&ps, | |
310 | sizeof (pool_scan_stat_t) / sizeof (uint64_t)) | |
311 | == 0); | |
312 | } | |
34dc7c2f BB |
313 | } |
314 | ||
315 | if (!vd->vdev_ops->vdev_op_leaf) { | |
316 | nvlist_t **child; | |
428870ff BB |
317 | int c, idx; |
318 | ||
319 | ASSERT(!vd->vdev_ishole); | |
34dc7c2f BB |
320 | |
321 | child = kmem_alloc(vd->vdev_children * sizeof (nvlist_t *), | |
322 | KM_SLEEP); | |
323 | ||
428870ff BB |
324 | for (c = 0, idx = 0; c < vd->vdev_children; c++) { |
325 | vdev_t *cvd = vd->vdev_child[c]; | |
34dc7c2f | 326 | |
428870ff BB |
327 | /* |
328 | * If we're generating an nvlist of removing | |
329 | * vdevs then skip over any device which is | |
330 | * not being removed. | |
331 | */ | |
332 | if ((flags & VDEV_CONFIG_REMOVING) && | |
333 | !cvd->vdev_removing) | |
334 | continue; | |
34dc7c2f | 335 | |
428870ff BB |
336 | child[idx++] = vdev_config_generate(spa, cvd, |
337 | getstats, flags); | |
338 | } | |
339 | ||
340 | if (idx) { | |
341 | VERIFY(nvlist_add_nvlist_array(nv, | |
342 | ZPOOL_CONFIG_CHILDREN, child, idx) == 0); | |
343 | } | |
344 | ||
345 | for (c = 0; c < idx; c++) | |
34dc7c2f BB |
346 | nvlist_free(child[c]); |
347 | ||
348 | kmem_free(child, vd->vdev_children * sizeof (nvlist_t *)); | |
349 | ||
350 | } else { | |
428870ff BB |
351 | const char *aux = NULL; |
352 | ||
34dc7c2f BB |
353 | if (vd->vdev_offline && !vd->vdev_tmpoffline) |
354 | VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_OFFLINE, | |
355 | B_TRUE) == 0); | |
572e2857 BB |
356 | if (vd->vdev_resilvering) |
357 | VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_RESILVERING, | |
358 | B_TRUE) == 0); | |
34dc7c2f BB |
359 | if (vd->vdev_faulted) |
360 | VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_FAULTED, | |
361 | B_TRUE) == 0); | |
362 | if (vd->vdev_degraded) | |
363 | VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_DEGRADED, | |
364 | B_TRUE) == 0); | |
365 | if (vd->vdev_removed) | |
366 | VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_REMOVED, | |
367 | B_TRUE) == 0); | |
368 | if (vd->vdev_unspare) | |
369 | VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_UNSPARE, | |
370 | B_TRUE) == 0); | |
428870ff BB |
371 | if (vd->vdev_ishole) |
372 | VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_IS_HOLE, | |
373 | B_TRUE) == 0); | |
374 | ||
375 | switch (vd->vdev_stat.vs_aux) { | |
376 | case VDEV_AUX_ERR_EXCEEDED: | |
377 | aux = "err_exceeded"; | |
378 | break; | |
379 | ||
380 | case VDEV_AUX_EXTERNAL: | |
381 | aux = "external"; | |
382 | break; | |
383 | } | |
384 | ||
385 | if (aux != NULL) | |
386 | VERIFY(nvlist_add_string(nv, ZPOOL_CONFIG_AUX_STATE, | |
387 | aux) == 0); | |
388 | ||
389 | if (vd->vdev_splitting && vd->vdev_orig_guid != 0LL) { | |
390 | VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_ORIG_GUID, | |
391 | vd->vdev_orig_guid) == 0); | |
392 | } | |
34dc7c2f BB |
393 | } |
394 | ||
395 | return (nv); | |
396 | } | |
397 | ||
428870ff BB |
398 | /* |
399 | * Generate a view of the top-level vdevs. If we currently have holes | |
400 | * in the namespace, then generate an array which contains a list of holey | |
401 | * vdevs. Additionally, add the number of top-level children that currently | |
402 | * exist. | |
403 | */ | |
404 | void | |
405 | vdev_top_config_generate(spa_t *spa, nvlist_t *config) | |
406 | { | |
407 | vdev_t *rvd = spa->spa_root_vdev; | |
408 | uint64_t *array; | |
409 | uint_t c, idx; | |
410 | ||
411 | array = kmem_alloc(rvd->vdev_children * sizeof (uint64_t), KM_SLEEP); | |
412 | ||
413 | for (c = 0, idx = 0; c < rvd->vdev_children; c++) { | |
414 | vdev_t *tvd = rvd->vdev_child[c]; | |
415 | ||
416 | if (tvd->vdev_ishole) | |
417 | array[idx++] = c; | |
418 | } | |
419 | ||
420 | if (idx) { | |
421 | VERIFY(nvlist_add_uint64_array(config, ZPOOL_CONFIG_HOLE_ARRAY, | |
422 | array, idx) == 0); | |
423 | } | |
424 | ||
425 | VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VDEV_CHILDREN, | |
426 | rvd->vdev_children) == 0); | |
427 | ||
428 | kmem_free(array, rvd->vdev_children * sizeof (uint64_t)); | |
429 | } | |
430 | ||
34dc7c2f BB |
431 | nvlist_t * |
432 | vdev_label_read_config(vdev_t *vd) | |
433 | { | |
434 | spa_t *spa = vd->vdev_spa; | |
435 | nvlist_t *config = NULL; | |
436 | vdev_phys_t *vp; | |
437 | zio_t *zio; | |
9babb374 BB |
438 | int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL | |
439 | ZIO_FLAG_SPECULATIVE; | |
d6320ddb | 440 | int l; |
34dc7c2f | 441 | |
b128c09f | 442 | ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); |
34dc7c2f BB |
443 | |
444 | if (!vdev_readable(vd)) | |
445 | return (NULL); | |
446 | ||
447 | vp = zio_buf_alloc(sizeof (vdev_phys_t)); | |
448 | ||
9babb374 | 449 | retry: |
d6320ddb | 450 | for (l = 0; l < VDEV_LABELS; l++) { |
34dc7c2f | 451 | |
b128c09f | 452 | zio = zio_root(spa, NULL, NULL, flags); |
34dc7c2f BB |
453 | |
454 | vdev_label_read(zio, vd, l, vp, | |
455 | offsetof(vdev_label_t, vl_vdev_phys), | |
b128c09f | 456 | sizeof (vdev_phys_t), NULL, NULL, flags); |
34dc7c2f BB |
457 | |
458 | if (zio_wait(zio) == 0 && | |
459 | nvlist_unpack(vp->vp_nvlist, sizeof (vp->vp_nvlist), | |
460 | &config, 0) == 0) | |
461 | break; | |
462 | ||
463 | if (config != NULL) { | |
464 | nvlist_free(config); | |
465 | config = NULL; | |
466 | } | |
467 | } | |
468 | ||
9babb374 BB |
469 | if (config == NULL && !(flags & ZIO_FLAG_TRYHARD)) { |
470 | flags |= ZIO_FLAG_TRYHARD; | |
471 | goto retry; | |
472 | } | |
473 | ||
34dc7c2f BB |
474 | zio_buf_free(vp, sizeof (vdev_phys_t)); |
475 | ||
476 | return (config); | |
477 | } | |
478 | ||
479 | /* | |
480 | * Determine if a device is in use. The 'spare_guid' parameter will be filled | |
481 | * in with the device guid if this spare is active elsewhere on the system. | |
482 | */ | |
483 | static boolean_t | |
484 | vdev_inuse(vdev_t *vd, uint64_t crtxg, vdev_labeltype_t reason, | |
485 | uint64_t *spare_guid, uint64_t *l2cache_guid) | |
486 | { | |
487 | spa_t *spa = vd->vdev_spa; | |
488 | uint64_t state, pool_guid, device_guid, txg, spare_pool; | |
489 | uint64_t vdtxg = 0; | |
490 | nvlist_t *label; | |
491 | ||
492 | if (spare_guid) | |
493 | *spare_guid = 0ULL; | |
494 | if (l2cache_guid) | |
495 | *l2cache_guid = 0ULL; | |
496 | ||
497 | /* | |
498 | * Read the label, if any, and perform some basic sanity checks. | |
499 | */ | |
500 | if ((label = vdev_label_read_config(vd)) == NULL) | |
501 | return (B_FALSE); | |
502 | ||
503 | (void) nvlist_lookup_uint64(label, ZPOOL_CONFIG_CREATE_TXG, | |
504 | &vdtxg); | |
505 | ||
506 | if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE, | |
507 | &state) != 0 || | |
508 | nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, | |
509 | &device_guid) != 0) { | |
510 | nvlist_free(label); | |
511 | return (B_FALSE); | |
512 | } | |
513 | ||
514 | if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE && | |
515 | (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_GUID, | |
516 | &pool_guid) != 0 || | |
517 | nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG, | |
518 | &txg) != 0)) { | |
519 | nvlist_free(label); | |
520 | return (B_FALSE); | |
521 | } | |
522 | ||
523 | nvlist_free(label); | |
524 | ||
525 | /* | |
526 | * Check to see if this device indeed belongs to the pool it claims to | |
527 | * be a part of. The only way this is allowed is if the device is a hot | |
528 | * spare (which we check for later on). | |
529 | */ | |
530 | if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE && | |
531 | !spa_guid_exists(pool_guid, device_guid) && | |
b128c09f | 532 | !spa_spare_exists(device_guid, NULL, NULL) && |
34dc7c2f BB |
533 | !spa_l2cache_exists(device_guid, NULL)) |
534 | return (B_FALSE); | |
535 | ||
536 | /* | |
537 | * If the transaction group is zero, then this an initialized (but | |
538 | * unused) label. This is only an error if the create transaction | |
539 | * on-disk is the same as the one we're using now, in which case the | |
540 | * user has attempted to add the same vdev multiple times in the same | |
541 | * transaction. | |
542 | */ | |
543 | if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE && | |
544 | txg == 0 && vdtxg == crtxg) | |
545 | return (B_TRUE); | |
546 | ||
547 | /* | |
548 | * Check to see if this is a spare device. We do an explicit check for | |
549 | * spa_has_spare() here because it may be on our pending list of spares | |
550 | * to add. We also check if it is an l2cache device. | |
551 | */ | |
b128c09f | 552 | if (spa_spare_exists(device_guid, &spare_pool, NULL) || |
34dc7c2f BB |
553 | spa_has_spare(spa, device_guid)) { |
554 | if (spare_guid) | |
555 | *spare_guid = device_guid; | |
556 | ||
557 | switch (reason) { | |
558 | case VDEV_LABEL_CREATE: | |
559 | case VDEV_LABEL_L2CACHE: | |
560 | return (B_TRUE); | |
561 | ||
562 | case VDEV_LABEL_REPLACE: | |
563 | return (!spa_has_spare(spa, device_guid) || | |
564 | spare_pool != 0ULL); | |
565 | ||
566 | case VDEV_LABEL_SPARE: | |
567 | return (spa_has_spare(spa, device_guid)); | |
e75c13c3 BB |
568 | default: |
569 | break; | |
34dc7c2f BB |
570 | } |
571 | } | |
572 | ||
573 | /* | |
574 | * Check to see if this is an l2cache device. | |
575 | */ | |
576 | if (spa_l2cache_exists(device_guid, NULL)) | |
577 | return (B_TRUE); | |
578 | ||
572e2857 BB |
579 | /* |
580 | * We can't rely on a pool's state if it's been imported | |
581 | * read-only. Instead we look to see if the pools is marked | |
582 | * read-only in the namespace and set the state to active. | |
583 | */ | |
584 | if ((spa = spa_by_guid(pool_guid, device_guid)) != NULL && | |
585 | spa_mode(spa) == FREAD) | |
586 | state = POOL_STATE_ACTIVE; | |
587 | ||
34dc7c2f BB |
588 | /* |
589 | * If the device is marked ACTIVE, then this device is in use by another | |
590 | * pool on the system. | |
591 | */ | |
592 | return (state == POOL_STATE_ACTIVE); | |
593 | } | |
594 | ||
595 | /* | |
596 | * Initialize a vdev label. We check to make sure each leaf device is not in | |
597 | * use, and writable. We put down an initial label which we will later | |
598 | * overwrite with a complete label. Note that it's important to do this | |
599 | * sequentially, not in parallel, so that we catch cases of multiple use of the | |
600 | * same leaf vdev in the vdev we're creating -- e.g. mirroring a disk with | |
601 | * itself. | |
602 | */ | |
603 | int | |
604 | vdev_label_init(vdev_t *vd, uint64_t crtxg, vdev_labeltype_t reason) | |
605 | { | |
606 | spa_t *spa = vd->vdev_spa; | |
607 | nvlist_t *label; | |
608 | vdev_phys_t *vp; | |
9babb374 | 609 | char *pad2; |
34dc7c2f BB |
610 | uberblock_t *ub; |
611 | zio_t *zio; | |
34dc7c2f BB |
612 | char *buf; |
613 | size_t buflen; | |
614 | int error; | |
615 | uint64_t spare_guid, l2cache_guid; | |
b128c09f | 616 | int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL; |
d6320ddb BB |
617 | int c, l; |
618 | vdev_t *pvd; | |
34dc7c2f | 619 | |
b128c09f | 620 | ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); |
34dc7c2f | 621 | |
d6320ddb | 622 | for (c = 0; c < vd->vdev_children; c++) |
34dc7c2f BB |
623 | if ((error = vdev_label_init(vd->vdev_child[c], |
624 | crtxg, reason)) != 0) | |
625 | return (error); | |
626 | ||
428870ff BB |
627 | /* Track the creation time for this vdev */ |
628 | vd->vdev_crtxg = crtxg; | |
629 | ||
34dc7c2f BB |
630 | if (!vd->vdev_ops->vdev_op_leaf) |
631 | return (0); | |
632 | ||
633 | /* | |
634 | * Dead vdevs cannot be initialized. | |
635 | */ | |
636 | if (vdev_is_dead(vd)) | |
637 | return (EIO); | |
638 | ||
639 | /* | |
640 | * Determine if the vdev is in use. | |
641 | */ | |
428870ff | 642 | if (reason != VDEV_LABEL_REMOVE && reason != VDEV_LABEL_SPLIT && |
34dc7c2f BB |
643 | vdev_inuse(vd, crtxg, reason, &spare_guid, &l2cache_guid)) |
644 | return (EBUSY); | |
645 | ||
34dc7c2f BB |
646 | /* |
647 | * If this is a request to add or replace a spare or l2cache device | |
648 | * that is in use elsewhere on the system, then we must update the | |
649 | * guid (which was initialized to a random value) to reflect the | |
650 | * actual GUID (which is shared between multiple pools). | |
651 | */ | |
652 | if (reason != VDEV_LABEL_REMOVE && reason != VDEV_LABEL_L2CACHE && | |
653 | spare_guid != 0ULL) { | |
b128c09f | 654 | uint64_t guid_delta = spare_guid - vd->vdev_guid; |
34dc7c2f | 655 | |
b128c09f | 656 | vd->vdev_guid += guid_delta; |
34dc7c2f | 657 | |
d6320ddb | 658 | for (pvd = vd; pvd != NULL; pvd = pvd->vdev_parent) |
b128c09f | 659 | pvd->vdev_guid_sum += guid_delta; |
34dc7c2f BB |
660 | |
661 | /* | |
662 | * If this is a replacement, then we want to fallthrough to the | |
663 | * rest of the code. If we're adding a spare, then it's already | |
664 | * labeled appropriately and we can just return. | |
665 | */ | |
666 | if (reason == VDEV_LABEL_SPARE) | |
667 | return (0); | |
428870ff BB |
668 | ASSERT(reason == VDEV_LABEL_REPLACE || |
669 | reason == VDEV_LABEL_SPLIT); | |
34dc7c2f BB |
670 | } |
671 | ||
672 | if (reason != VDEV_LABEL_REMOVE && reason != VDEV_LABEL_SPARE && | |
673 | l2cache_guid != 0ULL) { | |
b128c09f | 674 | uint64_t guid_delta = l2cache_guid - vd->vdev_guid; |
34dc7c2f | 675 | |
b128c09f | 676 | vd->vdev_guid += guid_delta; |
34dc7c2f | 677 | |
d6320ddb | 678 | for (pvd = vd; pvd != NULL; pvd = pvd->vdev_parent) |
b128c09f | 679 | pvd->vdev_guid_sum += guid_delta; |
34dc7c2f BB |
680 | |
681 | /* | |
682 | * If this is a replacement, then we want to fallthrough to the | |
683 | * rest of the code. If we're adding an l2cache, then it's | |
684 | * already labeled appropriately and we can just return. | |
685 | */ | |
686 | if (reason == VDEV_LABEL_L2CACHE) | |
687 | return (0); | |
688 | ASSERT(reason == VDEV_LABEL_REPLACE); | |
689 | } | |
690 | ||
691 | /* | |
692 | * Initialize its label. | |
693 | */ | |
694 | vp = zio_buf_alloc(sizeof (vdev_phys_t)); | |
695 | bzero(vp, sizeof (vdev_phys_t)); | |
696 | ||
697 | /* | |
698 | * Generate a label describing the pool and our top-level vdev. | |
699 | * We mark it as being from txg 0 to indicate that it's not | |
700 | * really part of an active pool just yet. The labels will | |
701 | * be written again with a meaningful txg by spa_sync(). | |
702 | */ | |
703 | if (reason == VDEV_LABEL_SPARE || | |
704 | (reason == VDEV_LABEL_REMOVE && vd->vdev_isspare)) { | |
705 | /* | |
706 | * For inactive hot spares, we generate a special label that | |
707 | * identifies as a mutually shared hot spare. We write the | |
708 | * label if we are adding a hot spare, or if we are removing an | |
709 | * active hot spare (in which case we want to revert the | |
710 | * labels). | |
711 | */ | |
712 | VERIFY(nvlist_alloc(&label, NV_UNIQUE_NAME, KM_SLEEP) == 0); | |
713 | ||
714 | VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_VERSION, | |
715 | spa_version(spa)) == 0); | |
716 | VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_POOL_STATE, | |
717 | POOL_STATE_SPARE) == 0); | |
718 | VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_GUID, | |
719 | vd->vdev_guid) == 0); | |
720 | } else if (reason == VDEV_LABEL_L2CACHE || | |
721 | (reason == VDEV_LABEL_REMOVE && vd->vdev_isl2cache)) { | |
722 | /* | |
723 | * For level 2 ARC devices, add a special label. | |
724 | */ | |
725 | VERIFY(nvlist_alloc(&label, NV_UNIQUE_NAME, KM_SLEEP) == 0); | |
726 | ||
727 | VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_VERSION, | |
728 | spa_version(spa)) == 0); | |
729 | VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_POOL_STATE, | |
730 | POOL_STATE_L2CACHE) == 0); | |
731 | VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_GUID, | |
732 | vd->vdev_guid) == 0); | |
733 | } else { | |
428870ff BB |
734 | uint64_t txg = 0ULL; |
735 | ||
736 | if (reason == VDEV_LABEL_SPLIT) | |
737 | txg = spa->spa_uberblock.ub_txg; | |
738 | label = spa_config_generate(spa, vd, txg, B_FALSE); | |
34dc7c2f BB |
739 | |
740 | /* | |
741 | * Add our creation time. This allows us to detect multiple | |
742 | * vdev uses as described above, and automatically expires if we | |
743 | * fail. | |
744 | */ | |
745 | VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_CREATE_TXG, | |
746 | crtxg) == 0); | |
747 | } | |
748 | ||
749 | buf = vp->vp_nvlist; | |
750 | buflen = sizeof (vp->vp_nvlist); | |
751 | ||
752 | error = nvlist_pack(label, &buf, &buflen, NV_ENCODE_XDR, KM_SLEEP); | |
753 | if (error != 0) { | |
754 | nvlist_free(label); | |
755 | zio_buf_free(vp, sizeof (vdev_phys_t)); | |
756 | /* EFAULT means nvlist_pack ran out of room */ | |
757 | return (error == EFAULT ? ENAMETOOLONG : EINVAL); | |
758 | } | |
759 | ||
34dc7c2f BB |
760 | /* |
761 | * Initialize uberblock template. | |
762 | */ | |
45d1cae3 BB |
763 | ub = zio_buf_alloc(VDEV_UBERBLOCK_RING); |
764 | bzero(ub, VDEV_UBERBLOCK_RING); | |
34dc7c2f BB |
765 | *ub = spa->spa_uberblock; |
766 | ub->ub_txg = 0; | |
767 | ||
9babb374 BB |
768 | /* Initialize the 2nd padding area. */ |
769 | pad2 = zio_buf_alloc(VDEV_PAD_SIZE); | |
770 | bzero(pad2, VDEV_PAD_SIZE); | |
771 | ||
34dc7c2f BB |
772 | /* |
773 | * Write everything in parallel. | |
774 | */ | |
9babb374 | 775 | retry: |
34dc7c2f BB |
776 | zio = zio_root(spa, NULL, NULL, flags); |
777 | ||
d6320ddb | 778 | for (l = 0; l < VDEV_LABELS; l++) { |
34dc7c2f BB |
779 | |
780 | vdev_label_write(zio, vd, l, vp, | |
781 | offsetof(vdev_label_t, vl_vdev_phys), | |
782 | sizeof (vdev_phys_t), NULL, NULL, flags); | |
783 | ||
9babb374 BB |
784 | /* |
785 | * Skip the 1st padding area. | |
786 | * Zero out the 2nd padding area where it might have | |
787 | * left over data from previous filesystem format. | |
788 | */ | |
789 | vdev_label_write(zio, vd, l, pad2, | |
790 | offsetof(vdev_label_t, vl_pad2), | |
791 | VDEV_PAD_SIZE, NULL, NULL, flags); | |
34dc7c2f | 792 | |
45d1cae3 BB |
793 | vdev_label_write(zio, vd, l, ub, |
794 | offsetof(vdev_label_t, vl_uberblock), | |
795 | VDEV_UBERBLOCK_RING, NULL, NULL, flags); | |
34dc7c2f BB |
796 | } |
797 | ||
798 | error = zio_wait(zio); | |
799 | ||
9babb374 BB |
800 | if (error != 0 && !(flags & ZIO_FLAG_TRYHARD)) { |
801 | flags |= ZIO_FLAG_TRYHARD; | |
802 | goto retry; | |
803 | } | |
804 | ||
34dc7c2f | 805 | nvlist_free(label); |
9babb374 | 806 | zio_buf_free(pad2, VDEV_PAD_SIZE); |
45d1cae3 | 807 | zio_buf_free(ub, VDEV_UBERBLOCK_RING); |
34dc7c2f BB |
808 | zio_buf_free(vp, sizeof (vdev_phys_t)); |
809 | ||
810 | /* | |
811 | * If this vdev hasn't been previously identified as a spare, then we | |
812 | * mark it as such only if a) we are labeling it as a spare, or b) it | |
813 | * exists as a spare elsewhere in the system. Do the same for | |
814 | * level 2 ARC devices. | |
815 | */ | |
816 | if (error == 0 && !vd->vdev_isspare && | |
817 | (reason == VDEV_LABEL_SPARE || | |
b128c09f | 818 | spa_spare_exists(vd->vdev_guid, NULL, NULL))) |
34dc7c2f BB |
819 | spa_spare_add(vd); |
820 | ||
821 | if (error == 0 && !vd->vdev_isl2cache && | |
822 | (reason == VDEV_LABEL_L2CACHE || | |
823 | spa_l2cache_exists(vd->vdev_guid, NULL))) | |
824 | spa_l2cache_add(vd); | |
825 | ||
826 | return (error); | |
827 | } | |
828 | ||
829 | /* | |
830 | * ========================================================================== | |
831 | * uberblock load/sync | |
832 | * ========================================================================== | |
833 | */ | |
834 | ||
835 | /* | |
836 | * Consider the following situation: txg is safely synced to disk. We've | |
837 | * written the first uberblock for txg + 1, and then we lose power. When we | |
838 | * come back up, we fail to see the uberblock for txg + 1 because, say, | |
839 | * it was on a mirrored device and the replica to which we wrote txg + 1 | |
840 | * is now offline. If we then make some changes and sync txg + 1, and then | |
841 | * the missing replica comes back, then for a new seconds we'll have two | |
842 | * conflicting uberblocks on disk with the same txg. The solution is simple: | |
843 | * among uberblocks with equal txg, choose the one with the latest timestamp. | |
844 | */ | |
845 | static int | |
846 | vdev_uberblock_compare(uberblock_t *ub1, uberblock_t *ub2) | |
847 | { | |
848 | if (ub1->ub_txg < ub2->ub_txg) | |
849 | return (-1); | |
850 | if (ub1->ub_txg > ub2->ub_txg) | |
851 | return (1); | |
852 | ||
853 | if (ub1->ub_timestamp < ub2->ub_timestamp) | |
854 | return (-1); | |
855 | if (ub1->ub_timestamp > ub2->ub_timestamp) | |
856 | return (1); | |
857 | ||
858 | return (0); | |
859 | } | |
860 | ||
861 | static void | |
862 | vdev_uberblock_load_done(zio_t *zio) | |
863 | { | |
428870ff | 864 | spa_t *spa = zio->io_spa; |
b128c09f | 865 | zio_t *rio = zio->io_private; |
34dc7c2f | 866 | uberblock_t *ub = zio->io_data; |
b128c09f | 867 | uberblock_t *ubbest = rio->io_private; |
34dc7c2f BB |
868 | |
869 | ASSERT3U(zio->io_size, ==, VDEV_UBERBLOCK_SIZE(zio->io_vd)); | |
870 | ||
871 | if (zio->io_error == 0 && uberblock_verify(ub) == 0) { | |
b128c09f | 872 | mutex_enter(&rio->io_lock); |
428870ff | 873 | if (ub->ub_txg <= spa->spa_load_max_txg && |
fb5f0bc8 | 874 | vdev_uberblock_compare(ub, ubbest) > 0) |
34dc7c2f | 875 | *ubbest = *ub; |
b128c09f | 876 | mutex_exit(&rio->io_lock); |
34dc7c2f BB |
877 | } |
878 | ||
879 | zio_buf_free(zio->io_data, zio->io_size); | |
880 | } | |
881 | ||
882 | void | |
883 | vdev_uberblock_load(zio_t *zio, vdev_t *vd, uberblock_t *ubbest) | |
884 | { | |
b128c09f BB |
885 | spa_t *spa = vd->vdev_spa; |
886 | vdev_t *rvd = spa->spa_root_vdev; | |
9babb374 BB |
887 | int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL | |
888 | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_TRYHARD; | |
d6320ddb | 889 | int c, l, n; |
b128c09f BB |
890 | |
891 | if (vd == rvd) { | |
892 | ASSERT(zio == NULL); | |
893 | spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); | |
894 | zio = zio_root(spa, NULL, ubbest, flags); | |
895 | bzero(ubbest, sizeof (uberblock_t)); | |
896 | } | |
34dc7c2f | 897 | |
b128c09f | 898 | ASSERT(zio != NULL); |
34dc7c2f | 899 | |
d6320ddb | 900 | for (c = 0; c < vd->vdev_children; c++) |
b128c09f | 901 | vdev_uberblock_load(zio, vd->vdev_child[c], ubbest); |
34dc7c2f | 902 | |
b128c09f | 903 | if (vd->vdev_ops->vdev_op_leaf && vdev_readable(vd)) { |
d6320ddb BB |
904 | for (l = 0; l < VDEV_LABELS; l++) { |
905 | for (n = 0; n < VDEV_UBERBLOCK_COUNT(vd); n++) { | |
b128c09f BB |
906 | vdev_label_read(zio, vd, l, |
907 | zio_buf_alloc(VDEV_UBERBLOCK_SIZE(vd)), | |
908 | VDEV_UBERBLOCK_OFFSET(vd, n), | |
909 | VDEV_UBERBLOCK_SIZE(vd), | |
910 | vdev_uberblock_load_done, zio, flags); | |
911 | } | |
34dc7c2f BB |
912 | } |
913 | } | |
b128c09f BB |
914 | |
915 | if (vd == rvd) { | |
916 | (void) zio_wait(zio); | |
917 | spa_config_exit(spa, SCL_ALL, FTAG); | |
918 | } | |
34dc7c2f BB |
919 | } |
920 | ||
921 | /* | |
922 | * On success, increment root zio's count of good writes. | |
923 | * We only get credit for writes to known-visible vdevs; see spa_vdev_add(). | |
924 | */ | |
925 | static void | |
926 | vdev_uberblock_sync_done(zio_t *zio) | |
927 | { | |
928 | uint64_t *good_writes = zio->io_private; | |
929 | ||
930 | if (zio->io_error == 0 && zio->io_vd->vdev_top->vdev_ms_array != 0) | |
931 | atomic_add_64(good_writes, 1); | |
932 | } | |
933 | ||
934 | /* | |
935 | * Write the uberblock to all labels of all leaves of the specified vdev. | |
936 | */ | |
937 | static void | |
b128c09f | 938 | vdev_uberblock_sync(zio_t *zio, uberblock_t *ub, vdev_t *vd, int flags) |
34dc7c2f | 939 | { |
34dc7c2f | 940 | uberblock_t *ubbuf; |
d6320ddb | 941 | int c, l, n; |
34dc7c2f | 942 | |
d6320ddb | 943 | for (c = 0; c < vd->vdev_children; c++) |
b128c09f | 944 | vdev_uberblock_sync(zio, ub, vd->vdev_child[c], flags); |
34dc7c2f BB |
945 | |
946 | if (!vd->vdev_ops->vdev_op_leaf) | |
947 | return; | |
948 | ||
b128c09f | 949 | if (!vdev_writeable(vd)) |
34dc7c2f BB |
950 | return; |
951 | ||
952 | n = ub->ub_txg & (VDEV_UBERBLOCK_COUNT(vd) - 1); | |
953 | ||
954 | ubbuf = zio_buf_alloc(VDEV_UBERBLOCK_SIZE(vd)); | |
955 | bzero(ubbuf, VDEV_UBERBLOCK_SIZE(vd)); | |
956 | *ubbuf = *ub; | |
957 | ||
d6320ddb | 958 | for (l = 0; l < VDEV_LABELS; l++) |
34dc7c2f | 959 | vdev_label_write(zio, vd, l, ubbuf, |
b128c09f | 960 | VDEV_UBERBLOCK_OFFSET(vd, n), VDEV_UBERBLOCK_SIZE(vd), |
34dc7c2f | 961 | vdev_uberblock_sync_done, zio->io_private, |
b128c09f | 962 | flags | ZIO_FLAG_DONT_PROPAGATE); |
34dc7c2f BB |
963 | |
964 | zio_buf_free(ubbuf, VDEV_UBERBLOCK_SIZE(vd)); | |
965 | } | |
966 | ||
967 | int | |
968 | vdev_uberblock_sync_list(vdev_t **svd, int svdcount, uberblock_t *ub, int flags) | |
969 | { | |
970 | spa_t *spa = svd[0]->vdev_spa; | |
34dc7c2f BB |
971 | zio_t *zio; |
972 | uint64_t good_writes = 0; | |
d6320ddb | 973 | int v; |
34dc7c2f BB |
974 | |
975 | zio = zio_root(spa, NULL, &good_writes, flags); | |
976 | ||
d6320ddb | 977 | for (v = 0; v < svdcount; v++) |
b128c09f | 978 | vdev_uberblock_sync(zio, ub, svd[v], flags); |
34dc7c2f BB |
979 | |
980 | (void) zio_wait(zio); | |
981 | ||
982 | /* | |
983 | * Flush the uberblocks to disk. This ensures that the odd labels | |
984 | * are no longer needed (because the new uberblocks and the even | |
985 | * labels are safely on disk), so it is safe to overwrite them. | |
986 | */ | |
987 | zio = zio_root(spa, NULL, NULL, flags); | |
988 | ||
d6320ddb | 989 | for (v = 0; v < svdcount; v++) |
34dc7c2f BB |
990 | zio_flush(zio, svd[v]); |
991 | ||
992 | (void) zio_wait(zio); | |
993 | ||
994 | return (good_writes >= 1 ? 0 : EIO); | |
995 | } | |
996 | ||
997 | /* | |
998 | * On success, increment the count of good writes for our top-level vdev. | |
999 | */ | |
1000 | static void | |
1001 | vdev_label_sync_done(zio_t *zio) | |
1002 | { | |
1003 | uint64_t *good_writes = zio->io_private; | |
1004 | ||
1005 | if (zio->io_error == 0) | |
1006 | atomic_add_64(good_writes, 1); | |
1007 | } | |
1008 | ||
1009 | /* | |
1010 | * If there weren't enough good writes, indicate failure to the parent. | |
1011 | */ | |
1012 | static void | |
1013 | vdev_label_sync_top_done(zio_t *zio) | |
1014 | { | |
1015 | uint64_t *good_writes = zio->io_private; | |
1016 | ||
1017 | if (*good_writes == 0) | |
1018 | zio->io_error = EIO; | |
1019 | ||
1020 | kmem_free(good_writes, sizeof (uint64_t)); | |
1021 | } | |
1022 | ||
b128c09f BB |
1023 | /* |
1024 | * We ignore errors for log and cache devices, simply free the private data. | |
1025 | */ | |
1026 | static void | |
1027 | vdev_label_sync_ignore_done(zio_t *zio) | |
1028 | { | |
1029 | kmem_free(zio->io_private, sizeof (uint64_t)); | |
1030 | } | |
1031 | ||
34dc7c2f BB |
1032 | /* |
1033 | * Write all even or odd labels to all leaves of the specified vdev. | |
1034 | */ | |
1035 | static void | |
b128c09f | 1036 | vdev_label_sync(zio_t *zio, vdev_t *vd, int l, uint64_t txg, int flags) |
34dc7c2f BB |
1037 | { |
1038 | nvlist_t *label; | |
1039 | vdev_phys_t *vp; | |
1040 | char *buf; | |
1041 | size_t buflen; | |
d6320ddb | 1042 | int c; |
34dc7c2f | 1043 | |
d6320ddb | 1044 | for (c = 0; c < vd->vdev_children; c++) |
b128c09f | 1045 | vdev_label_sync(zio, vd->vdev_child[c], l, txg, flags); |
34dc7c2f BB |
1046 | |
1047 | if (!vd->vdev_ops->vdev_op_leaf) | |
1048 | return; | |
1049 | ||
b128c09f | 1050 | if (!vdev_writeable(vd)) |
34dc7c2f BB |
1051 | return; |
1052 | ||
1053 | /* | |
1054 | * Generate a label describing the top-level config to which we belong. | |
1055 | */ | |
1056 | label = spa_config_generate(vd->vdev_spa, vd, txg, B_FALSE); | |
1057 | ||
1058 | vp = zio_buf_alloc(sizeof (vdev_phys_t)); | |
1059 | bzero(vp, sizeof (vdev_phys_t)); | |
1060 | ||
1061 | buf = vp->vp_nvlist; | |
1062 | buflen = sizeof (vp->vp_nvlist); | |
1063 | ||
1064 | if (nvlist_pack(label, &buf, &buflen, NV_ENCODE_XDR, KM_SLEEP) == 0) { | |
1065 | for (; l < VDEV_LABELS; l += 2) { | |
1066 | vdev_label_write(zio, vd, l, vp, | |
1067 | offsetof(vdev_label_t, vl_vdev_phys), | |
1068 | sizeof (vdev_phys_t), | |
1069 | vdev_label_sync_done, zio->io_private, | |
b128c09f | 1070 | flags | ZIO_FLAG_DONT_PROPAGATE); |
34dc7c2f BB |
1071 | } |
1072 | } | |
1073 | ||
1074 | zio_buf_free(vp, sizeof (vdev_phys_t)); | |
1075 | nvlist_free(label); | |
1076 | } | |
1077 | ||
1078 | int | |
b128c09f | 1079 | vdev_label_sync_list(spa_t *spa, int l, uint64_t txg, int flags) |
34dc7c2f | 1080 | { |
b128c09f | 1081 | list_t *dl = &spa->spa_config_dirty_list; |
34dc7c2f BB |
1082 | vdev_t *vd; |
1083 | zio_t *zio; | |
1084 | int error; | |
1085 | ||
1086 | /* | |
1087 | * Write the new labels to disk. | |
1088 | */ | |
1089 | zio = zio_root(spa, NULL, NULL, flags); | |
1090 | ||
1091 | for (vd = list_head(dl); vd != NULL; vd = list_next(dl, vd)) { | |
d6320ddb BB |
1092 | uint64_t *good_writes; |
1093 | zio_t *vio; | |
428870ff BB |
1094 | |
1095 | ASSERT(!vd->vdev_ishole); | |
1096 | ||
d6320ddb BB |
1097 | good_writes = kmem_zalloc(sizeof (uint64_t), KM_SLEEP); |
1098 | vio = zio_null(zio, spa, NULL, | |
b128c09f BB |
1099 | (vd->vdev_islog || vd->vdev_aux != NULL) ? |
1100 | vdev_label_sync_ignore_done : vdev_label_sync_top_done, | |
34dc7c2f | 1101 | good_writes, flags); |
b128c09f | 1102 | vdev_label_sync(vio, vd, l, txg, flags); |
34dc7c2f BB |
1103 | zio_nowait(vio); |
1104 | } | |
1105 | ||
1106 | error = zio_wait(zio); | |
1107 | ||
1108 | /* | |
1109 | * Flush the new labels to disk. | |
1110 | */ | |
1111 | zio = zio_root(spa, NULL, NULL, flags); | |
1112 | ||
1113 | for (vd = list_head(dl); vd != NULL; vd = list_next(dl, vd)) | |
1114 | zio_flush(zio, vd); | |
1115 | ||
1116 | (void) zio_wait(zio); | |
1117 | ||
1118 | return (error); | |
1119 | } | |
1120 | ||
1121 | /* | |
1122 | * Sync the uberblock and any changes to the vdev configuration. | |
1123 | * | |
1124 | * The order of operations is carefully crafted to ensure that | |
1125 | * if the system panics or loses power at any time, the state on disk | |
1126 | * is still transactionally consistent. The in-line comments below | |
1127 | * describe the failure semantics at each stage. | |
1128 | * | |
1129 | * Moreover, vdev_config_sync() is designed to be idempotent: if it fails | |
1130 | * at any time, you can just call it again, and it will resume its work. | |
1131 | */ | |
1132 | int | |
9babb374 | 1133 | vdev_config_sync(vdev_t **svd, int svdcount, uint64_t txg, boolean_t tryhard) |
34dc7c2f BB |
1134 | { |
1135 | spa_t *spa = svd[0]->vdev_spa; | |
1136 | uberblock_t *ub = &spa->spa_uberblock; | |
1137 | vdev_t *vd; | |
1138 | zio_t *zio; | |
1139 | int error; | |
b128c09f | 1140 | int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL; |
34dc7c2f | 1141 | |
9babb374 BB |
1142 | /* |
1143 | * Normally, we don't want to try too hard to write every label and | |
1144 | * uberblock. If there is a flaky disk, we don't want the rest of the | |
1145 | * sync process to block while we retry. But if we can't write a | |
1146 | * single label out, we should retry with ZIO_FLAG_TRYHARD before | |
1147 | * bailing out and declaring the pool faulted. | |
1148 | */ | |
1149 | if (tryhard) | |
1150 | flags |= ZIO_FLAG_TRYHARD; | |
1151 | ||
34dc7c2f BB |
1152 | ASSERT(ub->ub_txg <= txg); |
1153 | ||
1154 | /* | |
1155 | * If this isn't a resync due to I/O errors, | |
1156 | * and nothing changed in this transaction group, | |
1157 | * and the vdev configuration hasn't changed, | |
1158 | * then there's nothing to do. | |
1159 | */ | |
1160 | if (ub->ub_txg < txg && | |
1161 | uberblock_update(ub, spa->spa_root_vdev, txg) == B_FALSE && | |
b128c09f | 1162 | list_is_empty(&spa->spa_config_dirty_list)) |
34dc7c2f BB |
1163 | return (0); |
1164 | ||
1165 | if (txg > spa_freeze_txg(spa)) | |
1166 | return (0); | |
1167 | ||
1168 | ASSERT(txg <= spa->spa_final_txg); | |
1169 | ||
1170 | /* | |
1171 | * Flush the write cache of every disk that's been written to | |
1172 | * in this transaction group. This ensures that all blocks | |
1173 | * written in this txg will be committed to stable storage | |
1174 | * before any uberblock that references them. | |
1175 | */ | |
1176 | zio = zio_root(spa, NULL, NULL, flags); | |
1177 | ||
1178 | for (vd = txg_list_head(&spa->spa_vdev_txg_list, TXG_CLEAN(txg)); vd; | |
1179 | vd = txg_list_next(&spa->spa_vdev_txg_list, vd, TXG_CLEAN(txg))) | |
1180 | zio_flush(zio, vd); | |
1181 | ||
1182 | (void) zio_wait(zio); | |
1183 | ||
1184 | /* | |
1185 | * Sync out the even labels (L0, L2) for every dirty vdev. If the | |
1186 | * system dies in the middle of this process, that's OK: all of the | |
1187 | * even labels that made it to disk will be newer than any uberblock, | |
1188 | * and will therefore be considered invalid. The odd labels (L1, L3), | |
1189 | * which have not yet been touched, will still be valid. We flush | |
1190 | * the new labels to disk to ensure that all even-label updates | |
1191 | * are committed to stable storage before the uberblock update. | |
1192 | */ | |
b128c09f | 1193 | if ((error = vdev_label_sync_list(spa, 0, txg, flags)) != 0) |
34dc7c2f BB |
1194 | return (error); |
1195 | ||
1196 | /* | |
1197 | * Sync the uberblocks to all vdevs in svd[]. | |
1198 | * If the system dies in the middle of this step, there are two cases | |
1199 | * to consider, and the on-disk state is consistent either way: | |
1200 | * | |
1201 | * (1) If none of the new uberblocks made it to disk, then the | |
1202 | * previous uberblock will be the newest, and the odd labels | |
1203 | * (which had not yet been touched) will be valid with respect | |
1204 | * to that uberblock. | |
1205 | * | |
1206 | * (2) If one or more new uberblocks made it to disk, then they | |
1207 | * will be the newest, and the even labels (which had all | |
1208 | * been successfully committed) will be valid with respect | |
1209 | * to the new uberblocks. | |
1210 | */ | |
1211 | if ((error = vdev_uberblock_sync_list(svd, svdcount, ub, flags)) != 0) | |
1212 | return (error); | |
1213 | ||
1214 | /* | |
1215 | * Sync out odd labels for every dirty vdev. If the system dies | |
1216 | * in the middle of this process, the even labels and the new | |
1217 | * uberblocks will suffice to open the pool. The next time | |
1218 | * the pool is opened, the first thing we'll do -- before any | |
1219 | * user data is modified -- is mark every vdev dirty so that | |
1220 | * all labels will be brought up to date. We flush the new labels | |
1221 | * to disk to ensure that all odd-label updates are committed to | |
1222 | * stable storage before the next transaction group begins. | |
1223 | */ | |
b128c09f | 1224 | return (vdev_label_sync_list(spa, 1, txg, flags)); |
34dc7c2f | 1225 | } |