Print this page
3956 ::vdev -r should work with pipelines
3957 ztest should update the cachefile before killing itself
3958 multiple scans can lead to partial resilvering
3959 ddt entries are not always resilvered
3960 dsl_scan can skip over dedup-ed blocks if physical birth != logical birth
3961 freed gang blocks are not resilvered and can cause pool to suspend
3962 ztest should print out zfs debug buffer before exiting
Reviewed by: Matthew Ahrens <mahrens@delphix.com>
Reviewed by: Adam Leventhal <ahl@delphix.com>
| Split |
Close |
| Expand all |
| Collapse all |
--- old/usr/src/uts/common/fs/zfs/vdev_label.c
+++ new/usr/src/uts/common/fs/zfs/vdev_label.c
1 1 /*
2 2 * CDDL HEADER START
3 3 *
4 4 * The contents of this file are subject to the terms of the
5 5 * Common Development and Distribution License (the "License").
6 6 * You may not use this file except in compliance with the License.
7 7 *
8 8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 9 * or http://www.opensolaris.org/os/licensing.
10 10 * See the License for the specific language governing permissions
11 11 * and limitations under the License.
12 12 *
13 13 * When distributing Covered Code, include this CDDL HEADER in each
14 14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 15 * If applicable, add the following below this CDDL HEADER, with the
16 16 * fields enclosed by brackets "[]" replaced with your own identifying
17 17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 18 *
19 19 * CDDL HEADER END
20 20 */
21 21
22 22 /*
23 23 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
24 24 * Copyright (c) 2013 by Delphix. All rights reserved.
25 25 */
26 26
27 27 /*
28 28 * Virtual Device Labels
29 29 * ---------------------
30 30 *
31 31 * The vdev label serves several distinct purposes:
32 32 *
33 33 * 1. Uniquely identify this device as part of a ZFS pool and confirm its
34 34 * identity within the pool.
35 35 *
36 36 * 2. Verify that all the devices given in a configuration are present
37 37 * within the pool.
38 38 *
39 39 * 3. Determine the uberblock for the pool.
40 40 *
41 41 * 4. In case of an import operation, determine the configuration of the
42 42 * toplevel vdev of which it is a part.
43 43 *
44 44 * 5. If an import operation cannot find all the devices in the pool,
45 45 * provide enough information to the administrator to determine which
46 46 * devices are missing.
47 47 *
48 48 * It is important to note that while the kernel is responsible for writing the
49 49 * label, it only consumes the information in the first three cases. The
50 50 * latter information is only consumed in userland when determining the
51 51 * configuration to import a pool.
52 52 *
53 53 *
54 54 * Label Organization
55 55 * ------------------
56 56 *
57 57 * Before describing the contents of the label, it's important to understand how
58 58 * the labels are written and updated with respect to the uberblock.
59 59 *
60 60 * When the pool configuration is altered, either because it was newly created
61 61 * or a device was added, we want to update all the labels such that we can deal
62 62 * with fatal failure at any point. To this end, each disk has two labels which
63 63 * are updated before and after the uberblock is synced. Assuming we have
64 64 * labels and an uberblock with the following transaction groups:
65 65 *
66 66 * L1 UB L2
67 67 * +------+ +------+ +------+
68 68 * | | | | | |
69 69 * | t10 | | t10 | | t10 |
70 70 * | | | | | |
71 71 * +------+ +------+ +------+
72 72 *
73 73 * In this stable state, the labels and the uberblock were all updated within
74 74 * the same transaction group (10). Each label is mirrored and checksummed, so
75 75 * that we can detect when we fail partway through writing the label.
76 76 *
77 77 * In order to identify which labels are valid, the labels are written in the
78 78 * following manner:
79 79 *
80 80 * 1. For each vdev, update 'L1' to the new label
81 81 * 2. Update the uberblock
82 82 * 3. For each vdev, update 'L2' to the new label
83 83 *
84 84 * Given arbitrary failure, we can determine the correct label to use based on
85 85 * the transaction group. If we fail after updating L1 but before updating the
86 86 * UB, we will notice that L1's transaction group is greater than the uberblock,
87 87 * so L2 must be valid. If we fail after writing the uberblock but before
88 88 * writing L2, we will notice that L2's transaction group is less than L1, and
89 89 * therefore L1 is valid.
90 90 *
91 91 * Another added complexity is that not every label is updated when the config
92 92 * is synced. If we add a single device, we do not want to have to re-write
93 93 * every label for every device in the pool. This means that both L1 and L2 may
94 94 * be older than the pool uberblock, because the necessary information is stored
95 95 * on another vdev.
96 96 *
97 97 *
98 98 * On-disk Format
99 99 * --------------
100 100 *
101 101 * The vdev label consists of two distinct parts, and is wrapped within the
102 102 * vdev_label_t structure. The label includes 8k of padding to permit legacy
103 103 * VTOC disk labels, but is otherwise ignored.
104 104 *
105 105 * The first half of the label is a packed nvlist which contains pool wide
106 106 * properties, per-vdev properties, and configuration information. It is
107 107 * described in more detail below.
108 108 *
109 109 * The latter half of the label consists of a redundant array of uberblocks.
110 110 * These uberblocks are updated whenever a transaction group is committed,
111 111 * or when the configuration is updated. When a pool is loaded, we scan each
112 112 * vdev for the 'best' uberblock.
113 113 *
114 114 *
115 115 * Configuration Information
116 116 * -------------------------
117 117 *
118 118 * The nvlist describing the pool and vdev contains the following elements:
119 119 *
120 120 * version ZFS on-disk version
121 121 * name Pool name
122 122 * state Pool state
123 123 * txg Transaction group in which this label was written
124 124 * pool_guid Unique identifier for this pool
125 125 * vdev_tree An nvlist describing vdev tree.
126 126 * features_for_read
127 127 * An nvlist of the features necessary for reading the MOS.
128 128 *
129 129 * Each leaf device label also contains the following:
130 130 *
131 131 * top_guid Unique ID for top-level vdev in which this is contained
132 132 * guid Unique ID for the leaf vdev
133 133 *
134 134 * The 'vs' configuration follows the format described in 'spa_config.c'.
135 135 */
136 136
137 137 #include <sys/zfs_context.h>
138 138 #include <sys/spa.h>
139 139 #include <sys/spa_impl.h>
140 140 #include <sys/dmu.h>
141 141 #include <sys/zap.h>
142 142 #include <sys/vdev.h>
143 143 #include <sys/vdev_impl.h>
144 144 #include <sys/uberblock_impl.h>
145 145 #include <sys/metaslab.h>
146 146 #include <sys/zio.h>
147 147 #include <sys/dsl_scan.h>
148 148 #include <sys/fs/zfs.h>
149 149
150 150 /*
151 151 * Basic routines to read and write from a vdev label.
152 152 * Used throughout the rest of this file.
153 153 */
154 154 uint64_t
155 155 vdev_label_offset(uint64_t psize, int l, uint64_t offset)
156 156 {
157 157 ASSERT(offset < sizeof (vdev_label_t));
158 158 ASSERT(P2PHASE_TYPED(psize, sizeof (vdev_label_t), uint64_t) == 0);
159 159
160 160 return (offset + l * sizeof (vdev_label_t) + (l < VDEV_LABELS / 2 ?
161 161 0 : psize - VDEV_LABELS * sizeof (vdev_label_t)));
162 162 }
163 163
164 164 /*
165 165 * Returns back the vdev label associated with the passed in offset.
166 166 */
167 167 int
168 168 vdev_label_number(uint64_t psize, uint64_t offset)
169 169 {
170 170 int l;
171 171
172 172 if (offset >= psize - VDEV_LABEL_END_SIZE) {
173 173 offset -= psize - VDEV_LABEL_END_SIZE;
174 174 offset += (VDEV_LABELS / 2) * sizeof (vdev_label_t);
175 175 }
176 176 l = offset / sizeof (vdev_label_t);
177 177 return (l < VDEV_LABELS ? l : -1);
178 178 }
179 179
180 180 static void
181 181 vdev_label_read(zio_t *zio, vdev_t *vd, int l, void *buf, uint64_t offset,
182 182 uint64_t size, zio_done_func_t *done, void *private, int flags)
183 183 {
184 184 ASSERT(spa_config_held(zio->io_spa, SCL_STATE_ALL, RW_WRITER) ==
185 185 SCL_STATE_ALL);
186 186 ASSERT(flags & ZIO_FLAG_CONFIG_WRITER);
187 187
188 188 zio_nowait(zio_read_phys(zio, vd,
189 189 vdev_label_offset(vd->vdev_psize, l, offset),
190 190 size, buf, ZIO_CHECKSUM_LABEL, done, private,
191 191 ZIO_PRIORITY_SYNC_READ, flags, B_TRUE));
192 192 }
193 193
194 194 static void
195 195 vdev_label_write(zio_t *zio, vdev_t *vd, int l, void *buf, uint64_t offset,
196 196 uint64_t size, zio_done_func_t *done, void *private, int flags)
197 197 {
198 198 ASSERT(spa_config_held(zio->io_spa, SCL_ALL, RW_WRITER) == SCL_ALL ||
199 199 (spa_config_held(zio->io_spa, SCL_CONFIG | SCL_STATE, RW_READER) ==
200 200 (SCL_CONFIG | SCL_STATE) &&
201 201 dsl_pool_sync_context(spa_get_dsl(zio->io_spa))));
202 202 ASSERT(flags & ZIO_FLAG_CONFIG_WRITER);
203 203
204 204 zio_nowait(zio_write_phys(zio, vd,
205 205 vdev_label_offset(vd->vdev_psize, l, offset),
206 206 size, buf, ZIO_CHECKSUM_LABEL, done, private,
207 207 ZIO_PRIORITY_SYNC_WRITE, flags, B_TRUE));
208 208 }
|
↓ open down ↓ |
208 lines elided |
↑ open up ↑ |
209 209
210 210 /*
211 211 * Generate the nvlist representing this vdev's config.
212 212 */
213 213 nvlist_t *
214 214 vdev_config_generate(spa_t *spa, vdev_t *vd, boolean_t getstats,
215 215 vdev_config_flag_t flags)
216 216 {
217 217 nvlist_t *nv = NULL;
218 218
219 - VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0);
219 + nv = fnvlist_alloc();
220 220
221 - VERIFY(nvlist_add_string(nv, ZPOOL_CONFIG_TYPE,
222 - vd->vdev_ops->vdev_op_type) == 0);
221 + fnvlist_add_string(nv, ZPOOL_CONFIG_TYPE, vd->vdev_ops->vdev_op_type);
223 222 if (!(flags & (VDEV_CONFIG_SPARE | VDEV_CONFIG_L2CACHE)))
224 - VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_ID, vd->vdev_id)
225 - == 0);
226 - VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_GUID, vd->vdev_guid) == 0);
223 + fnvlist_add_uint64(nv, ZPOOL_CONFIG_ID, vd->vdev_id);
224 + fnvlist_add_uint64(nv, ZPOOL_CONFIG_GUID, vd->vdev_guid);
227 225
228 226 if (vd->vdev_path != NULL)
229 - VERIFY(nvlist_add_string(nv, ZPOOL_CONFIG_PATH,
230 - vd->vdev_path) == 0);
227 + fnvlist_add_string(nv, ZPOOL_CONFIG_PATH, vd->vdev_path);
231 228
232 229 if (vd->vdev_devid != NULL)
233 - VERIFY(nvlist_add_string(nv, ZPOOL_CONFIG_DEVID,
234 - vd->vdev_devid) == 0);
230 + fnvlist_add_string(nv, ZPOOL_CONFIG_DEVID, vd->vdev_devid);
235 231
236 232 if (vd->vdev_physpath != NULL)
237 - VERIFY(nvlist_add_string(nv, ZPOOL_CONFIG_PHYS_PATH,
238 - vd->vdev_physpath) == 0);
233 + fnvlist_add_string(nv, ZPOOL_CONFIG_PHYS_PATH,
234 + vd->vdev_physpath);
239 235
240 236 if (vd->vdev_fru != NULL)
241 - VERIFY(nvlist_add_string(nv, ZPOOL_CONFIG_FRU,
242 - vd->vdev_fru) == 0);
237 + fnvlist_add_string(nv, ZPOOL_CONFIG_FRU, vd->vdev_fru);
243 238
244 239 if (vd->vdev_nparity != 0) {
245 240 ASSERT(strcmp(vd->vdev_ops->vdev_op_type,
246 241 VDEV_TYPE_RAIDZ) == 0);
247 242
248 243 /*
249 244 * Make sure someone hasn't managed to sneak a fancy new vdev
250 245 * into a crufty old storage pool.
251 246 */
252 247 ASSERT(vd->vdev_nparity == 1 ||
253 248 (vd->vdev_nparity <= 2 &&
254 249 spa_version(spa) >= SPA_VERSION_RAIDZ2) ||
255 250 (vd->vdev_nparity <= 3 &&
256 251 spa_version(spa) >= SPA_VERSION_RAIDZ3));
257 252
258 253 /*
259 254 * Note that we'll add the nparity tag even on storage pools
260 255 * that only support a single parity device -- older software
261 256 * will just ignore it.
262 257 */
263 - VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_NPARITY,
264 - vd->vdev_nparity) == 0);
258 + fnvlist_add_uint64(nv, ZPOOL_CONFIG_NPARITY, vd->vdev_nparity);
265 259 }
266 260
267 261 if (vd->vdev_wholedisk != -1ULL)
268 - VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK,
269 - vd->vdev_wholedisk) == 0);
262 + fnvlist_add_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK,
263 + vd->vdev_wholedisk);
270 264
271 265 if (vd->vdev_not_present)
272 - VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_NOT_PRESENT, 1) == 0);
266 + fnvlist_add_uint64(nv, ZPOOL_CONFIG_NOT_PRESENT, 1);
273 267
274 268 if (vd->vdev_isspare)
275 - VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_IS_SPARE, 1) == 0);
269 + fnvlist_add_uint64(nv, ZPOOL_CONFIG_IS_SPARE, 1);
276 270
277 271 if (!(flags & (VDEV_CONFIG_SPARE | VDEV_CONFIG_L2CACHE)) &&
278 272 vd == vd->vdev_top) {
279 - VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_METASLAB_ARRAY,
280 - vd->vdev_ms_array) == 0);
281 - VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_METASLAB_SHIFT,
282 - vd->vdev_ms_shift) == 0);
283 - VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_ASHIFT,
284 - vd->vdev_ashift) == 0);
285 - VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_ASIZE,
286 - vd->vdev_asize) == 0);
287 - VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_IS_LOG,
288 - vd->vdev_islog) == 0);
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);
289 281 if (vd->vdev_removing)
290 - VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_REMOVING,
291 - vd->vdev_removing) == 0);
282 + fnvlist_add_uint64(nv, ZPOOL_CONFIG_REMOVING,
283 + vd->vdev_removing);
292 284 }
293 285
294 286 if (vd->vdev_dtl_smo.smo_object != 0)
295 - VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_DTL,
296 - vd->vdev_dtl_smo.smo_object) == 0);
287 + fnvlist_add_uint64(nv, ZPOOL_CONFIG_DTL,
288 + vd->vdev_dtl_smo.smo_object);
297 289
298 290 if (vd->vdev_crtxg)
299 - VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_CREATE_TXG,
300 - vd->vdev_crtxg) == 0);
291 + fnvlist_add_uint64(nv, ZPOOL_CONFIG_CREATE_TXG, vd->vdev_crtxg);
301 292
302 293 if (getstats) {
303 294 vdev_stat_t vs;
304 295 pool_scan_stat_t ps;
305 296
306 297 vdev_get_stats(vd, &vs);
307 - VERIFY(nvlist_add_uint64_array(nv, ZPOOL_CONFIG_VDEV_STATS,
308 - (uint64_t *)&vs, sizeof (vs) / sizeof (uint64_t)) == 0);
298 + fnvlist_add_uint64_array(nv, ZPOOL_CONFIG_VDEV_STATS,
299 + (uint64_t *)&vs, sizeof (vs) / sizeof (uint64_t));
309 300
310 301 /* provide either current or previous scan information */
311 302 if (spa_scan_get_stats(spa, &ps) == 0) {
312 - VERIFY(nvlist_add_uint64_array(nv,
303 + fnvlist_add_uint64_array(nv,
313 304 ZPOOL_CONFIG_SCAN_STATS, (uint64_t *)&ps,
314 - sizeof (pool_scan_stat_t) / sizeof (uint64_t))
315 - == 0);
305 + sizeof (pool_scan_stat_t) / sizeof (uint64_t));
316 306 }
317 307 }
318 308
319 309 if (!vd->vdev_ops->vdev_op_leaf) {
320 310 nvlist_t **child;
321 311 int c, idx;
322 312
323 313 ASSERT(!vd->vdev_ishole);
324 314
325 315 child = kmem_alloc(vd->vdev_children * sizeof (nvlist_t *),
326 316 KM_SLEEP);
327 317
328 318 for (c = 0, idx = 0; c < vd->vdev_children; c++) {
329 319 vdev_t *cvd = vd->vdev_child[c];
330 320
331 321 /*
332 322 * If we're generating an nvlist of removing
333 323 * vdevs then skip over any device which is
334 324 * not being removed.
|
↓ open down ↓ |
9 lines elided |
↑ open up ↑ |
335 325 */
336 326 if ((flags & VDEV_CONFIG_REMOVING) &&
337 327 !cvd->vdev_removing)
338 328 continue;
339 329
340 330 child[idx++] = vdev_config_generate(spa, cvd,
341 331 getstats, flags);
342 332 }
343 333
344 334 if (idx) {
345 - VERIFY(nvlist_add_nvlist_array(nv,
346 - ZPOOL_CONFIG_CHILDREN, child, idx) == 0);
335 + fnvlist_add_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
336 + child, idx);
347 337 }
348 338
349 339 for (c = 0; c < idx; c++)
350 340 nvlist_free(child[c]);
351 341
352 342 kmem_free(child, vd->vdev_children * sizeof (nvlist_t *));
353 343
354 344 } else {
355 345 const char *aux = NULL;
356 346
357 347 if (vd->vdev_offline && !vd->vdev_tmpoffline)
358 - VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_OFFLINE,
359 - B_TRUE) == 0);
360 - if (vd->vdev_resilvering)
361 - VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_RESILVERING,
362 - B_TRUE) == 0);
348 + fnvlist_add_uint64(nv, ZPOOL_CONFIG_OFFLINE, B_TRUE);
349 + if (vd->vdev_resilver_txg != 0)
350 + fnvlist_add_uint64(nv, ZPOOL_CONFIG_RESILVER_TXG,
351 + vd->vdev_resilver_txg);
363 352 if (vd->vdev_faulted)
364 - VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_FAULTED,
365 - B_TRUE) == 0);
353 + fnvlist_add_uint64(nv, ZPOOL_CONFIG_FAULTED, B_TRUE);
366 354 if (vd->vdev_degraded)
367 - VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_DEGRADED,
368 - B_TRUE) == 0);
355 + fnvlist_add_uint64(nv, ZPOOL_CONFIG_DEGRADED, B_TRUE);
369 356 if (vd->vdev_removed)
370 - VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_REMOVED,
371 - B_TRUE) == 0);
357 + fnvlist_add_uint64(nv, ZPOOL_CONFIG_REMOVED, B_TRUE);
372 358 if (vd->vdev_unspare)
373 - VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_UNSPARE,
374 - B_TRUE) == 0);
359 + fnvlist_add_uint64(nv, ZPOOL_CONFIG_UNSPARE, B_TRUE);
375 360 if (vd->vdev_ishole)
376 - VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_IS_HOLE,
377 - B_TRUE) == 0);
361 + fnvlist_add_uint64(nv, ZPOOL_CONFIG_IS_HOLE, B_TRUE);
378 362
379 363 switch (vd->vdev_stat.vs_aux) {
380 364 case VDEV_AUX_ERR_EXCEEDED:
381 365 aux = "err_exceeded";
382 366 break;
383 367
384 368 case VDEV_AUX_EXTERNAL:
385 369 aux = "external";
386 370 break;
387 371 }
388 372
389 373 if (aux != NULL)
390 - VERIFY(nvlist_add_string(nv, ZPOOL_CONFIG_AUX_STATE,
391 - aux) == 0);
374 + fnvlist_add_string(nv, ZPOOL_CONFIG_AUX_STATE, aux);
392 375
393 376 if (vd->vdev_splitting && vd->vdev_orig_guid != 0LL) {
394 - VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_ORIG_GUID,
395 - vd->vdev_orig_guid) == 0);
377 + fnvlist_add_uint64(nv, ZPOOL_CONFIG_ORIG_GUID,
378 + vd->vdev_orig_guid);
396 379 }
397 380 }
398 381
399 382 return (nv);
400 383 }
401 384
402 385 /*
403 386 * Generate a view of the top-level vdevs. If we currently have holes
404 387 * in the namespace, then generate an array which contains a list of holey
405 388 * vdevs. Additionally, add the number of top-level children that currently
406 389 * exist.
407 390 */
408 391 void
409 392 vdev_top_config_generate(spa_t *spa, nvlist_t *config)
410 393 {
411 394 vdev_t *rvd = spa->spa_root_vdev;
412 395 uint64_t *array;
413 396 uint_t c, idx;
414 397
415 398 array = kmem_alloc(rvd->vdev_children * sizeof (uint64_t), KM_SLEEP);
416 399
417 400 for (c = 0, idx = 0; c < rvd->vdev_children; c++) {
418 401 vdev_t *tvd = rvd->vdev_child[c];
419 402
420 403 if (tvd->vdev_ishole)
421 404 array[idx++] = c;
422 405 }
423 406
424 407 if (idx) {
425 408 VERIFY(nvlist_add_uint64_array(config, ZPOOL_CONFIG_HOLE_ARRAY,
426 409 array, idx) == 0);
427 410 }
428 411
429 412 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VDEV_CHILDREN,
430 413 rvd->vdev_children) == 0);
431 414
432 415 kmem_free(array, rvd->vdev_children * sizeof (uint64_t));
433 416 }
434 417
435 418 /*
436 419 * Returns the configuration from the label of the given vdev. For vdevs
437 420 * which don't have a txg value stored on their label (i.e. spares/cache)
438 421 * or have not been completely initialized (txg = 0) just return
439 422 * the configuration from the first valid label we find. Otherwise,
440 423 * find the most up-to-date label that does not exceed the specified
441 424 * 'txg' value.
442 425 */
443 426 nvlist_t *
444 427 vdev_label_read_config(vdev_t *vd, uint64_t txg)
445 428 {
446 429 spa_t *spa = vd->vdev_spa;
447 430 nvlist_t *config = NULL;
448 431 vdev_phys_t *vp;
449 432 zio_t *zio;
450 433 uint64_t best_txg = 0;
451 434 int error = 0;
452 435 int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL |
453 436 ZIO_FLAG_SPECULATIVE;
454 437
455 438 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
456 439
457 440 if (!vdev_readable(vd))
458 441 return (NULL);
459 442
460 443 vp = zio_buf_alloc(sizeof (vdev_phys_t));
461 444
462 445 retry:
463 446 for (int l = 0; l < VDEV_LABELS; l++) {
464 447 nvlist_t *label = NULL;
465 448
466 449 zio = zio_root(spa, NULL, NULL, flags);
467 450
468 451 vdev_label_read(zio, vd, l, vp,
469 452 offsetof(vdev_label_t, vl_vdev_phys),
470 453 sizeof (vdev_phys_t), NULL, NULL, flags);
471 454
472 455 if (zio_wait(zio) == 0 &&
473 456 nvlist_unpack(vp->vp_nvlist, sizeof (vp->vp_nvlist),
474 457 &label, 0) == 0) {
475 458 uint64_t label_txg = 0;
476 459
477 460 /*
478 461 * Auxiliary vdevs won't have txg values in their
479 462 * labels and newly added vdevs may not have been
480 463 * completely initialized so just return the
481 464 * configuration from the first valid label we
482 465 * encounter.
483 466 */
484 467 error = nvlist_lookup_uint64(label,
485 468 ZPOOL_CONFIG_POOL_TXG, &label_txg);
486 469 if ((error || label_txg == 0) && !config) {
487 470 config = label;
488 471 break;
489 472 } else if (label_txg <= txg && label_txg > best_txg) {
490 473 best_txg = label_txg;
491 474 nvlist_free(config);
492 475 config = fnvlist_dup(label);
493 476 }
494 477 }
495 478
496 479 if (label != NULL) {
497 480 nvlist_free(label);
498 481 label = NULL;
499 482 }
500 483 }
501 484
502 485 if (config == NULL && !(flags & ZIO_FLAG_TRYHARD)) {
503 486 flags |= ZIO_FLAG_TRYHARD;
504 487 goto retry;
505 488 }
506 489
507 490 zio_buf_free(vp, sizeof (vdev_phys_t));
508 491
509 492 return (config);
510 493 }
511 494
512 495 /*
513 496 * Determine if a device is in use. The 'spare_guid' parameter will be filled
514 497 * in with the device guid if this spare is active elsewhere on the system.
515 498 */
516 499 static boolean_t
517 500 vdev_inuse(vdev_t *vd, uint64_t crtxg, vdev_labeltype_t reason,
518 501 uint64_t *spare_guid, uint64_t *l2cache_guid)
519 502 {
520 503 spa_t *spa = vd->vdev_spa;
521 504 uint64_t state, pool_guid, device_guid, txg, spare_pool;
522 505 uint64_t vdtxg = 0;
523 506 nvlist_t *label;
524 507
525 508 if (spare_guid)
526 509 *spare_guid = 0ULL;
527 510 if (l2cache_guid)
528 511 *l2cache_guid = 0ULL;
529 512
530 513 /*
531 514 * Read the label, if any, and perform some basic sanity checks.
532 515 */
533 516 if ((label = vdev_label_read_config(vd, -1ULL)) == NULL)
534 517 return (B_FALSE);
535 518
536 519 (void) nvlist_lookup_uint64(label, ZPOOL_CONFIG_CREATE_TXG,
537 520 &vdtxg);
538 521
539 522 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE,
540 523 &state) != 0 ||
541 524 nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID,
542 525 &device_guid) != 0) {
543 526 nvlist_free(label);
544 527 return (B_FALSE);
545 528 }
546 529
547 530 if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
548 531 (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_GUID,
549 532 &pool_guid) != 0 ||
550 533 nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG,
551 534 &txg) != 0)) {
552 535 nvlist_free(label);
553 536 return (B_FALSE);
554 537 }
555 538
556 539 nvlist_free(label);
557 540
558 541 /*
559 542 * Check to see if this device indeed belongs to the pool it claims to
560 543 * be a part of. The only way this is allowed is if the device is a hot
561 544 * spare (which we check for later on).
562 545 */
563 546 if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
564 547 !spa_guid_exists(pool_guid, device_guid) &&
565 548 !spa_spare_exists(device_guid, NULL, NULL) &&
566 549 !spa_l2cache_exists(device_guid, NULL))
567 550 return (B_FALSE);
568 551
569 552 /*
570 553 * If the transaction group is zero, then this an initialized (but
571 554 * unused) label. This is only an error if the create transaction
572 555 * on-disk is the same as the one we're using now, in which case the
573 556 * user has attempted to add the same vdev multiple times in the same
574 557 * transaction.
575 558 */
576 559 if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
577 560 txg == 0 && vdtxg == crtxg)
578 561 return (B_TRUE);
579 562
580 563 /*
581 564 * Check to see if this is a spare device. We do an explicit check for
582 565 * spa_has_spare() here because it may be on our pending list of spares
583 566 * to add. We also check if it is an l2cache device.
584 567 */
585 568 if (spa_spare_exists(device_guid, &spare_pool, NULL) ||
586 569 spa_has_spare(spa, device_guid)) {
587 570 if (spare_guid)
588 571 *spare_guid = device_guid;
589 572
590 573 switch (reason) {
591 574 case VDEV_LABEL_CREATE:
592 575 case VDEV_LABEL_L2CACHE:
593 576 return (B_TRUE);
594 577
595 578 case VDEV_LABEL_REPLACE:
596 579 return (!spa_has_spare(spa, device_guid) ||
597 580 spare_pool != 0ULL);
598 581
599 582 case VDEV_LABEL_SPARE:
600 583 return (spa_has_spare(spa, device_guid));
601 584 }
602 585 }
603 586
604 587 /*
605 588 * Check to see if this is an l2cache device.
606 589 */
607 590 if (spa_l2cache_exists(device_guid, NULL))
608 591 return (B_TRUE);
609 592
610 593 /*
611 594 * We can't rely on a pool's state if it's been imported
612 595 * read-only. Instead we look to see if the pools is marked
613 596 * read-only in the namespace and set the state to active.
614 597 */
615 598 if ((spa = spa_by_guid(pool_guid, device_guid)) != NULL &&
616 599 spa_mode(spa) == FREAD)
617 600 state = POOL_STATE_ACTIVE;
618 601
619 602 /*
620 603 * If the device is marked ACTIVE, then this device is in use by another
621 604 * pool on the system.
622 605 */
623 606 return (state == POOL_STATE_ACTIVE);
624 607 }
625 608
626 609 /*
627 610 * Initialize a vdev label. We check to make sure each leaf device is not in
628 611 * use, and writable. We put down an initial label which we will later
629 612 * overwrite with a complete label. Note that it's important to do this
630 613 * sequentially, not in parallel, so that we catch cases of multiple use of the
631 614 * same leaf vdev in the vdev we're creating -- e.g. mirroring a disk with
632 615 * itself.
633 616 */
634 617 int
635 618 vdev_label_init(vdev_t *vd, uint64_t crtxg, vdev_labeltype_t reason)
636 619 {
637 620 spa_t *spa = vd->vdev_spa;
638 621 nvlist_t *label;
639 622 vdev_phys_t *vp;
640 623 char *pad2;
641 624 uberblock_t *ub;
642 625 zio_t *zio;
643 626 char *buf;
644 627 size_t buflen;
645 628 int error;
646 629 uint64_t spare_guid, l2cache_guid;
647 630 int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL;
648 631
649 632 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
650 633
651 634 for (int c = 0; c < vd->vdev_children; c++)
652 635 if ((error = vdev_label_init(vd->vdev_child[c],
653 636 crtxg, reason)) != 0)
654 637 return (error);
655 638
656 639 /* Track the creation time for this vdev */
657 640 vd->vdev_crtxg = crtxg;
658 641
659 642 if (!vd->vdev_ops->vdev_op_leaf)
660 643 return (0);
661 644
662 645 /*
663 646 * Dead vdevs cannot be initialized.
664 647 */
665 648 if (vdev_is_dead(vd))
666 649 return (SET_ERROR(EIO));
667 650
668 651 /*
669 652 * Determine if the vdev is in use.
670 653 */
671 654 if (reason != VDEV_LABEL_REMOVE && reason != VDEV_LABEL_SPLIT &&
672 655 vdev_inuse(vd, crtxg, reason, &spare_guid, &l2cache_guid))
673 656 return (SET_ERROR(EBUSY));
674 657
675 658 /*
676 659 * If this is a request to add or replace a spare or l2cache device
677 660 * that is in use elsewhere on the system, then we must update the
678 661 * guid (which was initialized to a random value) to reflect the
679 662 * actual GUID (which is shared between multiple pools).
680 663 */
681 664 if (reason != VDEV_LABEL_REMOVE && reason != VDEV_LABEL_L2CACHE &&
682 665 spare_guid != 0ULL) {
683 666 uint64_t guid_delta = spare_guid - vd->vdev_guid;
684 667
685 668 vd->vdev_guid += guid_delta;
686 669
687 670 for (vdev_t *pvd = vd; pvd != NULL; pvd = pvd->vdev_parent)
688 671 pvd->vdev_guid_sum += guid_delta;
689 672
690 673 /*
691 674 * If this is a replacement, then we want to fallthrough to the
692 675 * rest of the code. If we're adding a spare, then it's already
693 676 * labeled appropriately and we can just return.
694 677 */
695 678 if (reason == VDEV_LABEL_SPARE)
696 679 return (0);
697 680 ASSERT(reason == VDEV_LABEL_REPLACE ||
698 681 reason == VDEV_LABEL_SPLIT);
699 682 }
700 683
701 684 if (reason != VDEV_LABEL_REMOVE && reason != VDEV_LABEL_SPARE &&
702 685 l2cache_guid != 0ULL) {
703 686 uint64_t guid_delta = l2cache_guid - vd->vdev_guid;
704 687
705 688 vd->vdev_guid += guid_delta;
706 689
707 690 for (vdev_t *pvd = vd; pvd != NULL; pvd = pvd->vdev_parent)
708 691 pvd->vdev_guid_sum += guid_delta;
709 692
710 693 /*
711 694 * If this is a replacement, then we want to fallthrough to the
712 695 * rest of the code. If we're adding an l2cache, then it's
713 696 * already labeled appropriately and we can just return.
714 697 */
715 698 if (reason == VDEV_LABEL_L2CACHE)
716 699 return (0);
717 700 ASSERT(reason == VDEV_LABEL_REPLACE);
718 701 }
719 702
720 703 /*
721 704 * Initialize its label.
722 705 */
723 706 vp = zio_buf_alloc(sizeof (vdev_phys_t));
724 707 bzero(vp, sizeof (vdev_phys_t));
725 708
726 709 /*
727 710 * Generate a label describing the pool and our top-level vdev.
728 711 * We mark it as being from txg 0 to indicate that it's not
729 712 * really part of an active pool just yet. The labels will
730 713 * be written again with a meaningful txg by spa_sync().
731 714 */
732 715 if (reason == VDEV_LABEL_SPARE ||
733 716 (reason == VDEV_LABEL_REMOVE && vd->vdev_isspare)) {
734 717 /*
735 718 * For inactive hot spares, we generate a special label that
736 719 * identifies as a mutually shared hot spare. We write the
737 720 * label if we are adding a hot spare, or if we are removing an
738 721 * active hot spare (in which case we want to revert the
739 722 * labels).
740 723 */
741 724 VERIFY(nvlist_alloc(&label, NV_UNIQUE_NAME, KM_SLEEP) == 0);
742 725
743 726 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_VERSION,
744 727 spa_version(spa)) == 0);
745 728 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_POOL_STATE,
746 729 POOL_STATE_SPARE) == 0);
747 730 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_GUID,
748 731 vd->vdev_guid) == 0);
749 732 } else if (reason == VDEV_LABEL_L2CACHE ||
750 733 (reason == VDEV_LABEL_REMOVE && vd->vdev_isl2cache)) {
751 734 /*
752 735 * For level 2 ARC devices, add a special label.
753 736 */
754 737 VERIFY(nvlist_alloc(&label, NV_UNIQUE_NAME, KM_SLEEP) == 0);
755 738
756 739 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_VERSION,
757 740 spa_version(spa)) == 0);
758 741 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_POOL_STATE,
759 742 POOL_STATE_L2CACHE) == 0);
760 743 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_GUID,
761 744 vd->vdev_guid) == 0);
762 745 } else {
763 746 uint64_t txg = 0ULL;
764 747
765 748 if (reason == VDEV_LABEL_SPLIT)
766 749 txg = spa->spa_uberblock.ub_txg;
767 750 label = spa_config_generate(spa, vd, txg, B_FALSE);
768 751
769 752 /*
770 753 * Add our creation time. This allows us to detect multiple
771 754 * vdev uses as described above, and automatically expires if we
772 755 * fail.
773 756 */
774 757 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_CREATE_TXG,
775 758 crtxg) == 0);
776 759 }
777 760
778 761 buf = vp->vp_nvlist;
779 762 buflen = sizeof (vp->vp_nvlist);
780 763
781 764 error = nvlist_pack(label, &buf, &buflen, NV_ENCODE_XDR, KM_SLEEP);
782 765 if (error != 0) {
783 766 nvlist_free(label);
784 767 zio_buf_free(vp, sizeof (vdev_phys_t));
785 768 /* EFAULT means nvlist_pack ran out of room */
786 769 return (error == EFAULT ? ENAMETOOLONG : EINVAL);
787 770 }
788 771
789 772 /*
790 773 * Initialize uberblock template.
791 774 */
792 775 ub = zio_buf_alloc(VDEV_UBERBLOCK_RING);
793 776 bzero(ub, VDEV_UBERBLOCK_RING);
794 777 *ub = spa->spa_uberblock;
795 778 ub->ub_txg = 0;
796 779
797 780 /* Initialize the 2nd padding area. */
798 781 pad2 = zio_buf_alloc(VDEV_PAD_SIZE);
799 782 bzero(pad2, VDEV_PAD_SIZE);
800 783
801 784 /*
802 785 * Write everything in parallel.
803 786 */
804 787 retry:
805 788 zio = zio_root(spa, NULL, NULL, flags);
806 789
807 790 for (int l = 0; l < VDEV_LABELS; l++) {
808 791
809 792 vdev_label_write(zio, vd, l, vp,
810 793 offsetof(vdev_label_t, vl_vdev_phys),
811 794 sizeof (vdev_phys_t), NULL, NULL, flags);
812 795
813 796 /*
814 797 * Skip the 1st padding area.
815 798 * Zero out the 2nd padding area where it might have
816 799 * left over data from previous filesystem format.
817 800 */
818 801 vdev_label_write(zio, vd, l, pad2,
819 802 offsetof(vdev_label_t, vl_pad2),
820 803 VDEV_PAD_SIZE, NULL, NULL, flags);
821 804
822 805 vdev_label_write(zio, vd, l, ub,
823 806 offsetof(vdev_label_t, vl_uberblock),
824 807 VDEV_UBERBLOCK_RING, NULL, NULL, flags);
825 808 }
826 809
827 810 error = zio_wait(zio);
828 811
829 812 if (error != 0 && !(flags & ZIO_FLAG_TRYHARD)) {
830 813 flags |= ZIO_FLAG_TRYHARD;
831 814 goto retry;
832 815 }
833 816
834 817 nvlist_free(label);
835 818 zio_buf_free(pad2, VDEV_PAD_SIZE);
836 819 zio_buf_free(ub, VDEV_UBERBLOCK_RING);
837 820 zio_buf_free(vp, sizeof (vdev_phys_t));
838 821
839 822 /*
840 823 * If this vdev hasn't been previously identified as a spare, then we
841 824 * mark it as such only if a) we are labeling it as a spare, or b) it
842 825 * exists as a spare elsewhere in the system. Do the same for
843 826 * level 2 ARC devices.
844 827 */
845 828 if (error == 0 && !vd->vdev_isspare &&
846 829 (reason == VDEV_LABEL_SPARE ||
847 830 spa_spare_exists(vd->vdev_guid, NULL, NULL)))
848 831 spa_spare_add(vd);
849 832
850 833 if (error == 0 && !vd->vdev_isl2cache &&
851 834 (reason == VDEV_LABEL_L2CACHE ||
852 835 spa_l2cache_exists(vd->vdev_guid, NULL)))
853 836 spa_l2cache_add(vd);
854 837
855 838 return (error);
856 839 }
857 840
858 841 /*
859 842 * ==========================================================================
860 843 * uberblock load/sync
861 844 * ==========================================================================
862 845 */
863 846
864 847 /*
865 848 * Consider the following situation: txg is safely synced to disk. We've
866 849 * written the first uberblock for txg + 1, and then we lose power. When we
867 850 * come back up, we fail to see the uberblock for txg + 1 because, say,
868 851 * it was on a mirrored device and the replica to which we wrote txg + 1
869 852 * is now offline. If we then make some changes and sync txg + 1, and then
870 853 * the missing replica comes back, then for a few seconds we'll have two
871 854 * conflicting uberblocks on disk with the same txg. The solution is simple:
872 855 * among uberblocks with equal txg, choose the one with the latest timestamp.
873 856 */
874 857 static int
875 858 vdev_uberblock_compare(uberblock_t *ub1, uberblock_t *ub2)
876 859 {
877 860 if (ub1->ub_txg < ub2->ub_txg)
878 861 return (-1);
879 862 if (ub1->ub_txg > ub2->ub_txg)
880 863 return (1);
881 864
882 865 if (ub1->ub_timestamp < ub2->ub_timestamp)
883 866 return (-1);
884 867 if (ub1->ub_timestamp > ub2->ub_timestamp)
885 868 return (1);
886 869
887 870 return (0);
888 871 }
889 872
890 873 struct ubl_cbdata {
891 874 uberblock_t *ubl_ubbest; /* Best uberblock */
892 875 vdev_t *ubl_vd; /* vdev associated with the above */
893 876 };
894 877
895 878 static void
896 879 vdev_uberblock_load_done(zio_t *zio)
897 880 {
898 881 vdev_t *vd = zio->io_vd;
899 882 spa_t *spa = zio->io_spa;
900 883 zio_t *rio = zio->io_private;
901 884 uberblock_t *ub = zio->io_data;
902 885 struct ubl_cbdata *cbp = rio->io_private;
903 886
904 887 ASSERT3U(zio->io_size, ==, VDEV_UBERBLOCK_SIZE(vd));
905 888
906 889 if (zio->io_error == 0 && uberblock_verify(ub) == 0) {
907 890 mutex_enter(&rio->io_lock);
908 891 if (ub->ub_txg <= spa->spa_load_max_txg &&
909 892 vdev_uberblock_compare(ub, cbp->ubl_ubbest) > 0) {
910 893 /*
911 894 * Keep track of the vdev in which this uberblock
912 895 * was found. We will use this information later
913 896 * to obtain the config nvlist associated with
914 897 * this uberblock.
915 898 */
916 899 *cbp->ubl_ubbest = *ub;
917 900 cbp->ubl_vd = vd;
918 901 }
919 902 mutex_exit(&rio->io_lock);
920 903 }
921 904
922 905 zio_buf_free(zio->io_data, zio->io_size);
923 906 }
924 907
925 908 static void
926 909 vdev_uberblock_load_impl(zio_t *zio, vdev_t *vd, int flags,
927 910 struct ubl_cbdata *cbp)
928 911 {
929 912 for (int c = 0; c < vd->vdev_children; c++)
930 913 vdev_uberblock_load_impl(zio, vd->vdev_child[c], flags, cbp);
931 914
932 915 if (vd->vdev_ops->vdev_op_leaf && vdev_readable(vd)) {
933 916 for (int l = 0; l < VDEV_LABELS; l++) {
934 917 for (int n = 0; n < VDEV_UBERBLOCK_COUNT(vd); n++) {
935 918 vdev_label_read(zio, vd, l,
936 919 zio_buf_alloc(VDEV_UBERBLOCK_SIZE(vd)),
937 920 VDEV_UBERBLOCK_OFFSET(vd, n),
938 921 VDEV_UBERBLOCK_SIZE(vd),
939 922 vdev_uberblock_load_done, zio, flags);
940 923 }
941 924 }
942 925 }
943 926 }
944 927
945 928 /*
946 929 * Reads the 'best' uberblock from disk along with its associated
947 930 * configuration. First, we read the uberblock array of each label of each
948 931 * vdev, keeping track of the uberblock with the highest txg in each array.
949 932 * Then, we read the configuration from the same vdev as the best uberblock.
950 933 */
951 934 void
952 935 vdev_uberblock_load(vdev_t *rvd, uberblock_t *ub, nvlist_t **config)
953 936 {
954 937 zio_t *zio;
955 938 spa_t *spa = rvd->vdev_spa;
956 939 struct ubl_cbdata cb;
957 940 int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL |
958 941 ZIO_FLAG_SPECULATIVE | ZIO_FLAG_TRYHARD;
959 942
960 943 ASSERT(ub);
961 944 ASSERT(config);
962 945
963 946 bzero(ub, sizeof (uberblock_t));
964 947 *config = NULL;
965 948
966 949 cb.ubl_ubbest = ub;
967 950 cb.ubl_vd = NULL;
968 951
969 952 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
970 953 zio = zio_root(spa, NULL, &cb, flags);
971 954 vdev_uberblock_load_impl(zio, rvd, flags, &cb);
972 955 (void) zio_wait(zio);
973 956
974 957 /*
975 958 * It's possible that the best uberblock was discovered on a label
976 959 * that has a configuration which was written in a future txg.
977 960 * Search all labels on this vdev to find the configuration that
978 961 * matches the txg for our uberblock.
979 962 */
980 963 if (cb.ubl_vd != NULL)
981 964 *config = vdev_label_read_config(cb.ubl_vd, ub->ub_txg);
982 965 spa_config_exit(spa, SCL_ALL, FTAG);
983 966 }
984 967
985 968 /*
986 969 * On success, increment root zio's count of good writes.
987 970 * We only get credit for writes to known-visible vdevs; see spa_vdev_add().
988 971 */
989 972 static void
990 973 vdev_uberblock_sync_done(zio_t *zio)
991 974 {
992 975 uint64_t *good_writes = zio->io_private;
993 976
994 977 if (zio->io_error == 0 && zio->io_vd->vdev_top->vdev_ms_array != 0)
995 978 atomic_add_64(good_writes, 1);
996 979 }
997 980
998 981 /*
999 982 * Write the uberblock to all labels of all leaves of the specified vdev.
1000 983 */
1001 984 static void
1002 985 vdev_uberblock_sync(zio_t *zio, uberblock_t *ub, vdev_t *vd, int flags)
1003 986 {
1004 987 uberblock_t *ubbuf;
1005 988 int n;
1006 989
1007 990 for (int c = 0; c < vd->vdev_children; c++)
1008 991 vdev_uberblock_sync(zio, ub, vd->vdev_child[c], flags);
1009 992
1010 993 if (!vd->vdev_ops->vdev_op_leaf)
1011 994 return;
1012 995
1013 996 if (!vdev_writeable(vd))
1014 997 return;
1015 998
1016 999 n = ub->ub_txg & (VDEV_UBERBLOCK_COUNT(vd) - 1);
1017 1000
1018 1001 ubbuf = zio_buf_alloc(VDEV_UBERBLOCK_SIZE(vd));
1019 1002 bzero(ubbuf, VDEV_UBERBLOCK_SIZE(vd));
1020 1003 *ubbuf = *ub;
1021 1004
1022 1005 for (int l = 0; l < VDEV_LABELS; l++)
1023 1006 vdev_label_write(zio, vd, l, ubbuf,
1024 1007 VDEV_UBERBLOCK_OFFSET(vd, n), VDEV_UBERBLOCK_SIZE(vd),
1025 1008 vdev_uberblock_sync_done, zio->io_private,
1026 1009 flags | ZIO_FLAG_DONT_PROPAGATE);
1027 1010
1028 1011 zio_buf_free(ubbuf, VDEV_UBERBLOCK_SIZE(vd));
1029 1012 }
1030 1013
1031 1014 /* Sync the uberblocks to all vdevs in svd[] */
1032 1015 int
1033 1016 vdev_uberblock_sync_list(vdev_t **svd, int svdcount, uberblock_t *ub, int flags)
1034 1017 {
1035 1018 spa_t *spa = svd[0]->vdev_spa;
1036 1019 zio_t *zio;
1037 1020 uint64_t good_writes = 0;
1038 1021
1039 1022 zio = zio_root(spa, NULL, &good_writes, flags);
1040 1023
1041 1024 for (int v = 0; v < svdcount; v++)
1042 1025 vdev_uberblock_sync(zio, ub, svd[v], flags);
1043 1026
1044 1027 (void) zio_wait(zio);
1045 1028
1046 1029 /*
1047 1030 * Flush the uberblocks to disk. This ensures that the odd labels
1048 1031 * are no longer needed (because the new uberblocks and the even
1049 1032 * labels are safely on disk), so it is safe to overwrite them.
1050 1033 */
1051 1034 zio = zio_root(spa, NULL, NULL, flags);
1052 1035
1053 1036 for (int v = 0; v < svdcount; v++)
1054 1037 zio_flush(zio, svd[v]);
1055 1038
1056 1039 (void) zio_wait(zio);
1057 1040
1058 1041 return (good_writes >= 1 ? 0 : EIO);
1059 1042 }
1060 1043
1061 1044 /*
1062 1045 * On success, increment the count of good writes for our top-level vdev.
1063 1046 */
1064 1047 static void
1065 1048 vdev_label_sync_done(zio_t *zio)
1066 1049 {
1067 1050 uint64_t *good_writes = zio->io_private;
1068 1051
1069 1052 if (zio->io_error == 0)
1070 1053 atomic_add_64(good_writes, 1);
1071 1054 }
1072 1055
1073 1056 /*
1074 1057 * If there weren't enough good writes, indicate failure to the parent.
1075 1058 */
1076 1059 static void
1077 1060 vdev_label_sync_top_done(zio_t *zio)
1078 1061 {
1079 1062 uint64_t *good_writes = zio->io_private;
1080 1063
1081 1064 if (*good_writes == 0)
1082 1065 zio->io_error = SET_ERROR(EIO);
1083 1066
1084 1067 kmem_free(good_writes, sizeof (uint64_t));
1085 1068 }
1086 1069
1087 1070 /*
1088 1071 * We ignore errors for log and cache devices, simply free the private data.
1089 1072 */
1090 1073 static void
1091 1074 vdev_label_sync_ignore_done(zio_t *zio)
1092 1075 {
1093 1076 kmem_free(zio->io_private, sizeof (uint64_t));
1094 1077 }
1095 1078
1096 1079 /*
1097 1080 * Write all even or odd labels to all leaves of the specified vdev.
1098 1081 */
1099 1082 static void
1100 1083 vdev_label_sync(zio_t *zio, vdev_t *vd, int l, uint64_t txg, int flags)
1101 1084 {
1102 1085 nvlist_t *label;
1103 1086 vdev_phys_t *vp;
1104 1087 char *buf;
1105 1088 size_t buflen;
1106 1089
1107 1090 for (int c = 0; c < vd->vdev_children; c++)
1108 1091 vdev_label_sync(zio, vd->vdev_child[c], l, txg, flags);
1109 1092
1110 1093 if (!vd->vdev_ops->vdev_op_leaf)
1111 1094 return;
1112 1095
1113 1096 if (!vdev_writeable(vd))
1114 1097 return;
1115 1098
1116 1099 /*
1117 1100 * Generate a label describing the top-level config to which we belong.
1118 1101 */
1119 1102 label = spa_config_generate(vd->vdev_spa, vd, txg, B_FALSE);
1120 1103
1121 1104 vp = zio_buf_alloc(sizeof (vdev_phys_t));
1122 1105 bzero(vp, sizeof (vdev_phys_t));
1123 1106
1124 1107 buf = vp->vp_nvlist;
1125 1108 buflen = sizeof (vp->vp_nvlist);
1126 1109
1127 1110 if (nvlist_pack(label, &buf, &buflen, NV_ENCODE_XDR, KM_SLEEP) == 0) {
1128 1111 for (; l < VDEV_LABELS; l += 2) {
1129 1112 vdev_label_write(zio, vd, l, vp,
1130 1113 offsetof(vdev_label_t, vl_vdev_phys),
1131 1114 sizeof (vdev_phys_t),
1132 1115 vdev_label_sync_done, zio->io_private,
1133 1116 flags | ZIO_FLAG_DONT_PROPAGATE);
1134 1117 }
1135 1118 }
1136 1119
1137 1120 zio_buf_free(vp, sizeof (vdev_phys_t));
1138 1121 nvlist_free(label);
1139 1122 }
1140 1123
1141 1124 int
1142 1125 vdev_label_sync_list(spa_t *spa, int l, uint64_t txg, int flags)
1143 1126 {
1144 1127 list_t *dl = &spa->spa_config_dirty_list;
1145 1128 vdev_t *vd;
1146 1129 zio_t *zio;
1147 1130 int error;
1148 1131
1149 1132 /*
1150 1133 * Write the new labels to disk.
1151 1134 */
1152 1135 zio = zio_root(spa, NULL, NULL, flags);
1153 1136
1154 1137 for (vd = list_head(dl); vd != NULL; vd = list_next(dl, vd)) {
1155 1138 uint64_t *good_writes = kmem_zalloc(sizeof (uint64_t),
1156 1139 KM_SLEEP);
1157 1140
1158 1141 ASSERT(!vd->vdev_ishole);
1159 1142
1160 1143 zio_t *vio = zio_null(zio, spa, NULL,
1161 1144 (vd->vdev_islog || vd->vdev_aux != NULL) ?
1162 1145 vdev_label_sync_ignore_done : vdev_label_sync_top_done,
1163 1146 good_writes, flags);
1164 1147 vdev_label_sync(vio, vd, l, txg, flags);
1165 1148 zio_nowait(vio);
1166 1149 }
1167 1150
1168 1151 error = zio_wait(zio);
1169 1152
1170 1153 /*
1171 1154 * Flush the new labels to disk.
1172 1155 */
1173 1156 zio = zio_root(spa, NULL, NULL, flags);
1174 1157
1175 1158 for (vd = list_head(dl); vd != NULL; vd = list_next(dl, vd))
1176 1159 zio_flush(zio, vd);
1177 1160
1178 1161 (void) zio_wait(zio);
1179 1162
1180 1163 return (error);
1181 1164 }
1182 1165
1183 1166 /*
1184 1167 * Sync the uberblock and any changes to the vdev configuration.
1185 1168 *
1186 1169 * The order of operations is carefully crafted to ensure that
1187 1170 * if the system panics or loses power at any time, the state on disk
1188 1171 * is still transactionally consistent. The in-line comments below
1189 1172 * describe the failure semantics at each stage.
1190 1173 *
1191 1174 * Moreover, vdev_config_sync() is designed to be idempotent: if it fails
1192 1175 * at any time, you can just call it again, and it will resume its work.
1193 1176 */
1194 1177 int
1195 1178 vdev_config_sync(vdev_t **svd, int svdcount, uint64_t txg, boolean_t tryhard)
1196 1179 {
1197 1180 spa_t *spa = svd[0]->vdev_spa;
1198 1181 uberblock_t *ub = &spa->spa_uberblock;
1199 1182 vdev_t *vd;
1200 1183 zio_t *zio;
1201 1184 int error;
1202 1185 int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL;
1203 1186
1204 1187 /*
1205 1188 * Normally, we don't want to try too hard to write every label and
1206 1189 * uberblock. If there is a flaky disk, we don't want the rest of the
1207 1190 * sync process to block while we retry. But if we can't write a
1208 1191 * single label out, we should retry with ZIO_FLAG_TRYHARD before
1209 1192 * bailing out and declaring the pool faulted.
1210 1193 */
1211 1194 if (tryhard)
1212 1195 flags |= ZIO_FLAG_TRYHARD;
1213 1196
1214 1197 ASSERT(ub->ub_txg <= txg);
1215 1198
1216 1199 /*
1217 1200 * If this isn't a resync due to I/O errors,
1218 1201 * and nothing changed in this transaction group,
1219 1202 * and the vdev configuration hasn't changed,
1220 1203 * then there's nothing to do.
1221 1204 */
1222 1205 if (ub->ub_txg < txg &&
1223 1206 uberblock_update(ub, spa->spa_root_vdev, txg) == B_FALSE &&
1224 1207 list_is_empty(&spa->spa_config_dirty_list))
1225 1208 return (0);
1226 1209
1227 1210 if (txg > spa_freeze_txg(spa))
1228 1211 return (0);
1229 1212
1230 1213 ASSERT(txg <= spa->spa_final_txg);
1231 1214
1232 1215 /*
1233 1216 * Flush the write cache of every disk that's been written to
1234 1217 * in this transaction group. This ensures that all blocks
1235 1218 * written in this txg will be committed to stable storage
1236 1219 * before any uberblock that references them.
1237 1220 */
1238 1221 zio = zio_root(spa, NULL, NULL, flags);
1239 1222
1240 1223 for (vd = txg_list_head(&spa->spa_vdev_txg_list, TXG_CLEAN(txg)); vd;
1241 1224 vd = txg_list_next(&spa->spa_vdev_txg_list, vd, TXG_CLEAN(txg)))
1242 1225 zio_flush(zio, vd);
1243 1226
1244 1227 (void) zio_wait(zio);
1245 1228
1246 1229 /*
1247 1230 * Sync out the even labels (L0, L2) for every dirty vdev. If the
1248 1231 * system dies in the middle of this process, that's OK: all of the
1249 1232 * even labels that made it to disk will be newer than any uberblock,
1250 1233 * and will therefore be considered invalid. The odd labels (L1, L3),
1251 1234 * which have not yet been touched, will still be valid. We flush
1252 1235 * the new labels to disk to ensure that all even-label updates
1253 1236 * are committed to stable storage before the uberblock update.
1254 1237 */
1255 1238 if ((error = vdev_label_sync_list(spa, 0, txg, flags)) != 0)
1256 1239 return (error);
1257 1240
1258 1241 /*
1259 1242 * Sync the uberblocks to all vdevs in svd[].
1260 1243 * If the system dies in the middle of this step, there are two cases
1261 1244 * to consider, and the on-disk state is consistent either way:
1262 1245 *
1263 1246 * (1) If none of the new uberblocks made it to disk, then the
1264 1247 * previous uberblock will be the newest, and the odd labels
1265 1248 * (which had not yet been touched) will be valid with respect
1266 1249 * to that uberblock.
1267 1250 *
1268 1251 * (2) If one or more new uberblocks made it to disk, then they
1269 1252 * will be the newest, and the even labels (which had all
1270 1253 * been successfully committed) will be valid with respect
1271 1254 * to the new uberblocks.
1272 1255 */
1273 1256 if ((error = vdev_uberblock_sync_list(svd, svdcount, ub, flags)) != 0)
1274 1257 return (error);
1275 1258
1276 1259 /*
1277 1260 * Sync out odd labels for every dirty vdev. If the system dies
1278 1261 * in the middle of this process, the even labels and the new
1279 1262 * uberblocks will suffice to open the pool. The next time
1280 1263 * the pool is opened, the first thing we'll do -- before any
1281 1264 * user data is modified -- is mark every vdev dirty so that
1282 1265 * all labels will be brought up to date. We flush the new labels
1283 1266 * to disk to ensure that all odd-label updates are committed to
1284 1267 * stable storage before the next transaction group begins.
1285 1268 */
1286 1269 return (vdev_label_sync_list(spa, 1, txg, flags));
1287 1270 }
|
↓ open down ↓ |
882 lines elided |
↑ open up ↑ |
XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX