Print this page
2619 asynchronous destruction of ZFS file systems
2747 SPA versioning with zfs feature flags
Reviewed by: Matt Ahrens <mahrens@delphix.com>
Reviewed by: George Wilson <gwilson@delphix.com>
Reviewed by: Richard Lowe <richlowe@richlowe.net>
Reviewed by: Dan Kruchinin <dan.kruchinin@gmail.com>
Approved by: Dan McDonald <danmcd@nexenta.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
|
↓ open down ↓ |
10 lines elided |
↑ open up ↑ |
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 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
24 + * Copyright (c) 2012 by Delphix. All rights reserved.
23 25 */
24 26
25 27 /*
26 28 * Virtual Device Labels
27 29 * ---------------------
28 30 *
29 31 * The vdev label serves several distinct purposes:
30 32 *
31 33 * 1. Uniquely identify this device as part of a ZFS pool and confirm its
32 34 * identity within the pool.
33 35 *
34 36 * 2. Verify that all the devices given in a configuration are present
35 37 * within the pool.
36 38 *
37 39 * 3. Determine the uberblock for the pool.
38 40 *
39 41 * 4. In case of an import operation, determine the configuration of the
40 42 * toplevel vdev of which it is a part.
41 43 *
42 44 * 5. If an import operation cannot find all the devices in the pool,
43 45 * provide enough information to the administrator to determine which
44 46 * devices are missing.
45 47 *
46 48 * It is important to note that while the kernel is responsible for writing the
47 49 * label, it only consumes the information in the first three cases. The
48 50 * latter information is only consumed in userland when determining the
49 51 * configuration to import a pool.
50 52 *
51 53 *
52 54 * Label Organization
53 55 * ------------------
54 56 *
55 57 * Before describing the contents of the label, it's important to understand how
56 58 * the labels are written and updated with respect to the uberblock.
57 59 *
58 60 * When the pool configuration is altered, either because it was newly created
59 61 * or a device was added, we want to update all the labels such that we can deal
60 62 * with fatal failure at any point. To this end, each disk has two labels which
61 63 * are updated before and after the uberblock is synced. Assuming we have
62 64 * labels and an uberblock with the following transaction groups:
63 65 *
64 66 * L1 UB L2
65 67 * +------+ +------+ +------+
66 68 * | | | | | |
67 69 * | t10 | | t10 | | t10 |
68 70 * | | | | | |
69 71 * +------+ +------+ +------+
70 72 *
71 73 * In this stable state, the labels and the uberblock were all updated within
72 74 * the same transaction group (10). Each label is mirrored and checksummed, so
73 75 * that we can detect when we fail partway through writing the label.
74 76 *
75 77 * In order to identify which labels are valid, the labels are written in the
76 78 * following manner:
77 79 *
78 80 * 1. For each vdev, update 'L1' to the new label
79 81 * 2. Update the uberblock
80 82 * 3. For each vdev, update 'L2' to the new label
81 83 *
82 84 * Given arbitrary failure, we can determine the correct label to use based on
83 85 * the transaction group. If we fail after updating L1 but before updating the
84 86 * UB, we will notice that L1's transaction group is greater than the uberblock,
85 87 * so L2 must be valid. If we fail after writing the uberblock but before
86 88 * writing L2, we will notice that L2's transaction group is less than L1, and
87 89 * therefore L1 is valid.
88 90 *
89 91 * Another added complexity is that not every label is updated when the config
90 92 * is synced. If we add a single device, we do not want to have to re-write
91 93 * every label for every device in the pool. This means that both L1 and L2 may
92 94 * be older than the pool uberblock, because the necessary information is stored
93 95 * on another vdev.
94 96 *
95 97 *
96 98 * On-disk Format
97 99 * --------------
98 100 *
99 101 * The vdev label consists of two distinct parts, and is wrapped within the
100 102 * vdev_label_t structure. The label includes 8k of padding to permit legacy
101 103 * VTOC disk labels, but is otherwise ignored.
102 104 *
103 105 * The first half of the label is a packed nvlist which contains pool wide
104 106 * properties, per-vdev properties, and configuration information. It is
105 107 * described in more detail below.
106 108 *
107 109 * The latter half of the label consists of a redundant array of uberblocks.
108 110 * These uberblocks are updated whenever a transaction group is committed,
109 111 * or when the configuration is updated. When a pool is loaded, we scan each
110 112 * vdev for the 'best' uberblock.
111 113 *
112 114 *
113 115 * Configuration Information
|
↓ open down ↓ |
81 lines elided |
↑ open up ↑ |
114 116 * -------------------------
115 117 *
116 118 * The nvlist describing the pool and vdev contains the following elements:
117 119 *
118 120 * version ZFS on-disk version
119 121 * name Pool name
120 122 * state Pool state
121 123 * txg Transaction group in which this label was written
122 124 * pool_guid Unique identifier for this pool
123 125 * vdev_tree An nvlist describing vdev tree.
126 + * features_for_read
127 + * An nvlist of the features necessary for reading the MOS.
124 128 *
125 129 * Each leaf device label also contains the following:
126 130 *
127 131 * top_guid Unique ID for top-level vdev in which this is contained
128 132 * guid Unique ID for the leaf vdev
129 133 *
130 134 * The 'vs' configuration follows the format described in 'spa_config.c'.
131 135 */
132 136
133 137 #include <sys/zfs_context.h>
134 138 #include <sys/spa.h>
135 139 #include <sys/spa_impl.h>
136 140 #include <sys/dmu.h>
137 141 #include <sys/zap.h>
138 142 #include <sys/vdev.h>
139 143 #include <sys/vdev_impl.h>
140 144 #include <sys/uberblock_impl.h>
141 145 #include <sys/metaslab.h>
142 146 #include <sys/zio.h>
143 147 #include <sys/dsl_scan.h>
144 148 #include <sys/fs/zfs.h>
145 149
146 150 /*
147 151 * Basic routines to read and write from a vdev label.
148 152 * Used throughout the rest of this file.
149 153 */
150 154 uint64_t
151 155 vdev_label_offset(uint64_t psize, int l, uint64_t offset)
152 156 {
153 157 ASSERT(offset < sizeof (vdev_label_t));
154 158 ASSERT(P2PHASE_TYPED(psize, sizeof (vdev_label_t), uint64_t) == 0);
155 159
156 160 return (offset + l * sizeof (vdev_label_t) + (l < VDEV_LABELS / 2 ?
157 161 0 : psize - VDEV_LABELS * sizeof (vdev_label_t)));
158 162 }
159 163
160 164 /*
161 165 * Returns back the vdev label associated with the passed in offset.
162 166 */
163 167 int
164 168 vdev_label_number(uint64_t psize, uint64_t offset)
165 169 {
166 170 int l;
167 171
168 172 if (offset >= psize - VDEV_LABEL_END_SIZE) {
169 173 offset -= psize - VDEV_LABEL_END_SIZE;
170 174 offset += (VDEV_LABELS / 2) * sizeof (vdev_label_t);
171 175 }
172 176 l = offset / sizeof (vdev_label_t);
173 177 return (l < VDEV_LABELS ? l : -1);
174 178 }
175 179
176 180 static void
177 181 vdev_label_read(zio_t *zio, vdev_t *vd, int l, void *buf, uint64_t offset,
178 182 uint64_t size, zio_done_func_t *done, void *private, int flags)
179 183 {
180 184 ASSERT(spa_config_held(zio->io_spa, SCL_STATE_ALL, RW_WRITER) ==
181 185 SCL_STATE_ALL);
182 186 ASSERT(flags & ZIO_FLAG_CONFIG_WRITER);
183 187
184 188 zio_nowait(zio_read_phys(zio, vd,
185 189 vdev_label_offset(vd->vdev_psize, l, offset),
186 190 size, buf, ZIO_CHECKSUM_LABEL, done, private,
187 191 ZIO_PRIORITY_SYNC_READ, flags, B_TRUE));
188 192 }
189 193
190 194 static void
191 195 vdev_label_write(zio_t *zio, vdev_t *vd, int l, void *buf, uint64_t offset,
192 196 uint64_t size, zio_done_func_t *done, void *private, int flags)
193 197 {
194 198 ASSERT(spa_config_held(zio->io_spa, SCL_ALL, RW_WRITER) == SCL_ALL ||
195 199 (spa_config_held(zio->io_spa, SCL_CONFIG | SCL_STATE, RW_READER) ==
196 200 (SCL_CONFIG | SCL_STATE) &&
197 201 dsl_pool_sync_context(spa_get_dsl(zio->io_spa))));
198 202 ASSERT(flags & ZIO_FLAG_CONFIG_WRITER);
199 203
200 204 zio_nowait(zio_write_phys(zio, vd,
201 205 vdev_label_offset(vd->vdev_psize, l, offset),
202 206 size, buf, ZIO_CHECKSUM_LABEL, done, private,
203 207 ZIO_PRIORITY_SYNC_WRITE, flags, B_TRUE));
204 208 }
205 209
206 210 /*
207 211 * Generate the nvlist representing this vdev's config.
208 212 */
209 213 nvlist_t *
210 214 vdev_config_generate(spa_t *spa, vdev_t *vd, boolean_t getstats,
211 215 vdev_config_flag_t flags)
212 216 {
213 217 nvlist_t *nv = NULL;
214 218
215 219 VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0);
216 220
217 221 VERIFY(nvlist_add_string(nv, ZPOOL_CONFIG_TYPE,
218 222 vd->vdev_ops->vdev_op_type) == 0);
219 223 if (!(flags & (VDEV_CONFIG_SPARE | VDEV_CONFIG_L2CACHE)))
220 224 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_ID, vd->vdev_id)
221 225 == 0);
222 226 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_GUID, vd->vdev_guid) == 0);
223 227
224 228 if (vd->vdev_path != NULL)
225 229 VERIFY(nvlist_add_string(nv, ZPOOL_CONFIG_PATH,
226 230 vd->vdev_path) == 0);
227 231
228 232 if (vd->vdev_devid != NULL)
229 233 VERIFY(nvlist_add_string(nv, ZPOOL_CONFIG_DEVID,
230 234 vd->vdev_devid) == 0);
231 235
232 236 if (vd->vdev_physpath != NULL)
233 237 VERIFY(nvlist_add_string(nv, ZPOOL_CONFIG_PHYS_PATH,
234 238 vd->vdev_physpath) == 0);
235 239
236 240 if (vd->vdev_fru != NULL)
237 241 VERIFY(nvlist_add_string(nv, ZPOOL_CONFIG_FRU,
238 242 vd->vdev_fru) == 0);
239 243
240 244 if (vd->vdev_nparity != 0) {
241 245 ASSERT(strcmp(vd->vdev_ops->vdev_op_type,
242 246 VDEV_TYPE_RAIDZ) == 0);
243 247
244 248 /*
245 249 * Make sure someone hasn't managed to sneak a fancy new vdev
246 250 * into a crufty old storage pool.
247 251 */
248 252 ASSERT(vd->vdev_nparity == 1 ||
249 253 (vd->vdev_nparity <= 2 &&
250 254 spa_version(spa) >= SPA_VERSION_RAIDZ2) ||
251 255 (vd->vdev_nparity <= 3 &&
252 256 spa_version(spa) >= SPA_VERSION_RAIDZ3));
253 257
254 258 /*
255 259 * Note that we'll add the nparity tag even on storage pools
256 260 * that only support a single parity device -- older software
257 261 * will just ignore it.
258 262 */
259 263 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_NPARITY,
260 264 vd->vdev_nparity) == 0);
261 265 }
262 266
263 267 if (vd->vdev_wholedisk != -1ULL)
264 268 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK,
265 269 vd->vdev_wholedisk) == 0);
266 270
267 271 if (vd->vdev_not_present)
268 272 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_NOT_PRESENT, 1) == 0);
269 273
270 274 if (vd->vdev_isspare)
271 275 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_IS_SPARE, 1) == 0);
272 276
273 277 if (!(flags & (VDEV_CONFIG_SPARE | VDEV_CONFIG_L2CACHE)) &&
274 278 vd == vd->vdev_top) {
275 279 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_METASLAB_ARRAY,
276 280 vd->vdev_ms_array) == 0);
277 281 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_METASLAB_SHIFT,
278 282 vd->vdev_ms_shift) == 0);
279 283 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_ASHIFT,
280 284 vd->vdev_ashift) == 0);
281 285 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_ASIZE,
282 286 vd->vdev_asize) == 0);
283 287 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_IS_LOG,
284 288 vd->vdev_islog) == 0);
285 289 if (vd->vdev_removing)
286 290 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_REMOVING,
287 291 vd->vdev_removing) == 0);
288 292 }
289 293
290 294 if (vd->vdev_dtl_smo.smo_object != 0)
291 295 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_DTL,
292 296 vd->vdev_dtl_smo.smo_object) == 0);
293 297
294 298 if (vd->vdev_crtxg)
295 299 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_CREATE_TXG,
296 300 vd->vdev_crtxg) == 0);
297 301
298 302 if (getstats) {
299 303 vdev_stat_t vs;
300 304 pool_scan_stat_t ps;
301 305
302 306 vdev_get_stats(vd, &vs);
303 307 VERIFY(nvlist_add_uint64_array(nv, ZPOOL_CONFIG_VDEV_STATS,
304 308 (uint64_t *)&vs, sizeof (vs) / sizeof (uint64_t)) == 0);
305 309
306 310 /* provide either current or previous scan information */
307 311 if (spa_scan_get_stats(spa, &ps) == 0) {
308 312 VERIFY(nvlist_add_uint64_array(nv,
309 313 ZPOOL_CONFIG_SCAN_STATS, (uint64_t *)&ps,
310 314 sizeof (pool_scan_stat_t) / sizeof (uint64_t))
311 315 == 0);
312 316 }
313 317 }
314 318
315 319 if (!vd->vdev_ops->vdev_op_leaf) {
316 320 nvlist_t **child;
317 321 int c, idx;
318 322
319 323 ASSERT(!vd->vdev_ishole);
320 324
321 325 child = kmem_alloc(vd->vdev_children * sizeof (nvlist_t *),
322 326 KM_SLEEP);
323 327
324 328 for (c = 0, idx = 0; c < vd->vdev_children; c++) {
325 329 vdev_t *cvd = vd->vdev_child[c];
326 330
327 331 /*
328 332 * If we're generating an nvlist of removing
329 333 * vdevs then skip over any device which is
330 334 * not being removed.
331 335 */
332 336 if ((flags & VDEV_CONFIG_REMOVING) &&
333 337 !cvd->vdev_removing)
334 338 continue;
335 339
336 340 child[idx++] = vdev_config_generate(spa, cvd,
337 341 getstats, flags);
338 342 }
339 343
340 344 if (idx) {
341 345 VERIFY(nvlist_add_nvlist_array(nv,
342 346 ZPOOL_CONFIG_CHILDREN, child, idx) == 0);
343 347 }
344 348
345 349 for (c = 0; c < idx; c++)
346 350 nvlist_free(child[c]);
347 351
348 352 kmem_free(child, vd->vdev_children * sizeof (nvlist_t *));
349 353
350 354 } else {
351 355 const char *aux = NULL;
352 356
353 357 if (vd->vdev_offline && !vd->vdev_tmpoffline)
354 358 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_OFFLINE,
355 359 B_TRUE) == 0);
356 360 if (vd->vdev_resilvering)
357 361 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_RESILVERING,
358 362 B_TRUE) == 0);
359 363 if (vd->vdev_faulted)
360 364 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_FAULTED,
361 365 B_TRUE) == 0);
362 366 if (vd->vdev_degraded)
363 367 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_DEGRADED,
364 368 B_TRUE) == 0);
365 369 if (vd->vdev_removed)
366 370 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_REMOVED,
367 371 B_TRUE) == 0);
368 372 if (vd->vdev_unspare)
369 373 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_UNSPARE,
370 374 B_TRUE) == 0);
371 375 if (vd->vdev_ishole)
372 376 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_IS_HOLE,
373 377 B_TRUE) == 0);
374 378
375 379 switch (vd->vdev_stat.vs_aux) {
376 380 case VDEV_AUX_ERR_EXCEEDED:
377 381 aux = "err_exceeded";
378 382 break;
379 383
380 384 case VDEV_AUX_EXTERNAL:
381 385 aux = "external";
382 386 break;
383 387 }
384 388
385 389 if (aux != NULL)
386 390 VERIFY(nvlist_add_string(nv, ZPOOL_CONFIG_AUX_STATE,
387 391 aux) == 0);
388 392
389 393 if (vd->vdev_splitting && vd->vdev_orig_guid != 0LL) {
390 394 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_ORIG_GUID,
391 395 vd->vdev_orig_guid) == 0);
392 396 }
393 397 }
394 398
395 399 return (nv);
396 400 }
397 401
398 402 /*
399 403 * Generate a view of the top-level vdevs. If we currently have holes
400 404 * in the namespace, then generate an array which contains a list of holey
401 405 * vdevs. Additionally, add the number of top-level children that currently
402 406 * exist.
403 407 */
404 408 void
405 409 vdev_top_config_generate(spa_t *spa, nvlist_t *config)
406 410 {
407 411 vdev_t *rvd = spa->spa_root_vdev;
408 412 uint64_t *array;
409 413 uint_t c, idx;
410 414
411 415 array = kmem_alloc(rvd->vdev_children * sizeof (uint64_t), KM_SLEEP);
412 416
413 417 for (c = 0, idx = 0; c < rvd->vdev_children; c++) {
414 418 vdev_t *tvd = rvd->vdev_child[c];
415 419
416 420 if (tvd->vdev_ishole)
417 421 array[idx++] = c;
418 422 }
419 423
420 424 if (idx) {
|
↓ open down ↓ |
287 lines elided |
↑ open up ↑ |
421 425 VERIFY(nvlist_add_uint64_array(config, ZPOOL_CONFIG_HOLE_ARRAY,
422 426 array, idx) == 0);
423 427 }
424 428
425 429 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VDEV_CHILDREN,
426 430 rvd->vdev_children) == 0);
427 431
428 432 kmem_free(array, rvd->vdev_children * sizeof (uint64_t));
429 433 }
430 434
435 +/*
436 + * Returns the configuration from the label of the given vdev. If 'label' is
437 + * VDEV_BEST_LABEL, each label of the vdev will be read until a valid
438 + * configuration is found; otherwise, only the specified label will be read.
439 + */
431 440 nvlist_t *
432 -vdev_label_read_config(vdev_t *vd)
441 +vdev_label_read_config(vdev_t *vd, int label)
433 442 {
434 443 spa_t *spa = vd->vdev_spa;
435 444 nvlist_t *config = NULL;
436 445 vdev_phys_t *vp;
437 446 zio_t *zio;
438 447 int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL |
439 448 ZIO_FLAG_SPECULATIVE;
440 449
441 450 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
442 451
443 452 if (!vdev_readable(vd))
444 453 return (NULL);
445 454
446 455 vp = zio_buf_alloc(sizeof (vdev_phys_t));
447 456
448 457 retry:
449 458 for (int l = 0; l < VDEV_LABELS; l++) {
459 + if (label >= 0 && label < VDEV_LABELS && label != l)
460 + continue;
450 461
451 462 zio = zio_root(spa, NULL, NULL, flags);
452 463
453 464 vdev_label_read(zio, vd, l, vp,
454 465 offsetof(vdev_label_t, vl_vdev_phys),
455 466 sizeof (vdev_phys_t), NULL, NULL, flags);
456 467
457 468 if (zio_wait(zio) == 0 &&
458 469 nvlist_unpack(vp->vp_nvlist, sizeof (vp->vp_nvlist),
459 470 &config, 0) == 0)
460 471 break;
461 472
462 473 if (config != NULL) {
463 474 nvlist_free(config);
464 475 config = NULL;
465 476 }
466 477 }
467 478
468 479 if (config == NULL && !(flags & ZIO_FLAG_TRYHARD)) {
469 480 flags |= ZIO_FLAG_TRYHARD;
470 481 goto retry;
471 482 }
472 483
473 484 zio_buf_free(vp, sizeof (vdev_phys_t));
474 485
475 486 return (config);
476 487 }
477 488
478 489 /*
479 490 * Determine if a device is in use. The 'spare_guid' parameter will be filled
480 491 * in with the device guid if this spare is active elsewhere on the system.
481 492 */
482 493 static boolean_t
483 494 vdev_inuse(vdev_t *vd, uint64_t crtxg, vdev_labeltype_t reason,
484 495 uint64_t *spare_guid, uint64_t *l2cache_guid)
485 496 {
486 497 spa_t *spa = vd->vdev_spa;
487 498 uint64_t state, pool_guid, device_guid, txg, spare_pool;
488 499 uint64_t vdtxg = 0;
|
↓ open down ↓ |
29 lines elided |
↑ open up ↑ |
489 500 nvlist_t *label;
490 501
491 502 if (spare_guid)
492 503 *spare_guid = 0ULL;
493 504 if (l2cache_guid)
494 505 *l2cache_guid = 0ULL;
495 506
496 507 /*
497 508 * Read the label, if any, and perform some basic sanity checks.
498 509 */
499 - if ((label = vdev_label_read_config(vd)) == NULL)
510 + if ((label = vdev_label_read_config(vd, VDEV_BEST_LABEL)) == NULL)
500 511 return (B_FALSE);
501 512
502 513 (void) nvlist_lookup_uint64(label, ZPOOL_CONFIG_CREATE_TXG,
503 514 &vdtxg);
504 515
505 516 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE,
506 517 &state) != 0 ||
507 518 nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID,
508 519 &device_guid) != 0) {
509 520 nvlist_free(label);
510 521 return (B_FALSE);
511 522 }
512 523
513 524 if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
514 525 (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_GUID,
515 526 &pool_guid) != 0 ||
516 527 nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG,
517 528 &txg) != 0)) {
518 529 nvlist_free(label);
519 530 return (B_FALSE);
520 531 }
521 532
522 533 nvlist_free(label);
523 534
524 535 /*
525 536 * Check to see if this device indeed belongs to the pool it claims to
526 537 * be a part of. The only way this is allowed is if the device is a hot
527 538 * spare (which we check for later on).
528 539 */
529 540 if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
530 541 !spa_guid_exists(pool_guid, device_guid) &&
531 542 !spa_spare_exists(device_guid, NULL, NULL) &&
532 543 !spa_l2cache_exists(device_guid, NULL))
533 544 return (B_FALSE);
534 545
535 546 /*
536 547 * If the transaction group is zero, then this an initialized (but
537 548 * unused) label. This is only an error if the create transaction
538 549 * on-disk is the same as the one we're using now, in which case the
539 550 * user has attempted to add the same vdev multiple times in the same
540 551 * transaction.
541 552 */
542 553 if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
543 554 txg == 0 && vdtxg == crtxg)
544 555 return (B_TRUE);
545 556
546 557 /*
547 558 * Check to see if this is a spare device. We do an explicit check for
548 559 * spa_has_spare() here because it may be on our pending list of spares
549 560 * to add. We also check if it is an l2cache device.
550 561 */
551 562 if (spa_spare_exists(device_guid, &spare_pool, NULL) ||
552 563 spa_has_spare(spa, device_guid)) {
553 564 if (spare_guid)
554 565 *spare_guid = device_guid;
555 566
556 567 switch (reason) {
557 568 case VDEV_LABEL_CREATE:
558 569 case VDEV_LABEL_L2CACHE:
559 570 return (B_TRUE);
560 571
561 572 case VDEV_LABEL_REPLACE:
562 573 return (!spa_has_spare(spa, device_guid) ||
563 574 spare_pool != 0ULL);
564 575
565 576 case VDEV_LABEL_SPARE:
566 577 return (spa_has_spare(spa, device_guid));
567 578 }
568 579 }
569 580
570 581 /*
571 582 * Check to see if this is an l2cache device.
572 583 */
573 584 if (spa_l2cache_exists(device_guid, NULL))
574 585 return (B_TRUE);
575 586
576 587 /*
577 588 * We can't rely on a pool's state if it's been imported
578 589 * read-only. Instead we look to see if the pools is marked
579 590 * read-only in the namespace and set the state to active.
580 591 */
581 592 if ((spa = spa_by_guid(pool_guid, device_guid)) != NULL &&
582 593 spa_mode(spa) == FREAD)
583 594 state = POOL_STATE_ACTIVE;
584 595
585 596 /*
586 597 * If the device is marked ACTIVE, then this device is in use by another
587 598 * pool on the system.
588 599 */
589 600 return (state == POOL_STATE_ACTIVE);
590 601 }
591 602
592 603 /*
593 604 * Initialize a vdev label. We check to make sure each leaf device is not in
594 605 * use, and writable. We put down an initial label which we will later
595 606 * overwrite with a complete label. Note that it's important to do this
596 607 * sequentially, not in parallel, so that we catch cases of multiple use of the
597 608 * same leaf vdev in the vdev we're creating -- e.g. mirroring a disk with
598 609 * itself.
599 610 */
600 611 int
601 612 vdev_label_init(vdev_t *vd, uint64_t crtxg, vdev_labeltype_t reason)
602 613 {
603 614 spa_t *spa = vd->vdev_spa;
604 615 nvlist_t *label;
605 616 vdev_phys_t *vp;
606 617 char *pad2;
607 618 uberblock_t *ub;
608 619 zio_t *zio;
609 620 char *buf;
610 621 size_t buflen;
611 622 int error;
612 623 uint64_t spare_guid, l2cache_guid;
613 624 int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL;
614 625
615 626 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
616 627
617 628 for (int c = 0; c < vd->vdev_children; c++)
618 629 if ((error = vdev_label_init(vd->vdev_child[c],
619 630 crtxg, reason)) != 0)
620 631 return (error);
621 632
622 633 /* Track the creation time for this vdev */
623 634 vd->vdev_crtxg = crtxg;
624 635
625 636 if (!vd->vdev_ops->vdev_op_leaf)
626 637 return (0);
627 638
628 639 /*
629 640 * Dead vdevs cannot be initialized.
630 641 */
631 642 if (vdev_is_dead(vd))
632 643 return (EIO);
633 644
634 645 /*
635 646 * Determine if the vdev is in use.
636 647 */
637 648 if (reason != VDEV_LABEL_REMOVE && reason != VDEV_LABEL_SPLIT &&
638 649 vdev_inuse(vd, crtxg, reason, &spare_guid, &l2cache_guid))
639 650 return (EBUSY);
640 651
641 652 /*
642 653 * If this is a request to add or replace a spare or l2cache device
643 654 * that is in use elsewhere on the system, then we must update the
644 655 * guid (which was initialized to a random value) to reflect the
645 656 * actual GUID (which is shared between multiple pools).
646 657 */
647 658 if (reason != VDEV_LABEL_REMOVE && reason != VDEV_LABEL_L2CACHE &&
648 659 spare_guid != 0ULL) {
649 660 uint64_t guid_delta = spare_guid - vd->vdev_guid;
650 661
651 662 vd->vdev_guid += guid_delta;
652 663
653 664 for (vdev_t *pvd = vd; pvd != NULL; pvd = pvd->vdev_parent)
654 665 pvd->vdev_guid_sum += guid_delta;
655 666
656 667 /*
657 668 * If this is a replacement, then we want to fallthrough to the
658 669 * rest of the code. If we're adding a spare, then it's already
659 670 * labeled appropriately and we can just return.
660 671 */
661 672 if (reason == VDEV_LABEL_SPARE)
662 673 return (0);
663 674 ASSERT(reason == VDEV_LABEL_REPLACE ||
664 675 reason == VDEV_LABEL_SPLIT);
665 676 }
666 677
667 678 if (reason != VDEV_LABEL_REMOVE && reason != VDEV_LABEL_SPARE &&
668 679 l2cache_guid != 0ULL) {
669 680 uint64_t guid_delta = l2cache_guid - vd->vdev_guid;
670 681
671 682 vd->vdev_guid += guid_delta;
672 683
673 684 for (vdev_t *pvd = vd; pvd != NULL; pvd = pvd->vdev_parent)
674 685 pvd->vdev_guid_sum += guid_delta;
675 686
676 687 /*
677 688 * If this is a replacement, then we want to fallthrough to the
678 689 * rest of the code. If we're adding an l2cache, then it's
679 690 * already labeled appropriately and we can just return.
680 691 */
681 692 if (reason == VDEV_LABEL_L2CACHE)
682 693 return (0);
683 694 ASSERT(reason == VDEV_LABEL_REPLACE);
684 695 }
685 696
686 697 /*
687 698 * Initialize its label.
688 699 */
689 700 vp = zio_buf_alloc(sizeof (vdev_phys_t));
690 701 bzero(vp, sizeof (vdev_phys_t));
691 702
692 703 /*
693 704 * Generate a label describing the pool and our top-level vdev.
694 705 * We mark it as being from txg 0 to indicate that it's not
695 706 * really part of an active pool just yet. The labels will
696 707 * be written again with a meaningful txg by spa_sync().
697 708 */
698 709 if (reason == VDEV_LABEL_SPARE ||
699 710 (reason == VDEV_LABEL_REMOVE && vd->vdev_isspare)) {
700 711 /*
701 712 * For inactive hot spares, we generate a special label that
702 713 * identifies as a mutually shared hot spare. We write the
703 714 * label if we are adding a hot spare, or if we are removing an
704 715 * active hot spare (in which case we want to revert the
705 716 * labels).
706 717 */
707 718 VERIFY(nvlist_alloc(&label, NV_UNIQUE_NAME, KM_SLEEP) == 0);
708 719
709 720 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_VERSION,
710 721 spa_version(spa)) == 0);
711 722 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_POOL_STATE,
712 723 POOL_STATE_SPARE) == 0);
713 724 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_GUID,
714 725 vd->vdev_guid) == 0);
715 726 } else if (reason == VDEV_LABEL_L2CACHE ||
716 727 (reason == VDEV_LABEL_REMOVE && vd->vdev_isl2cache)) {
717 728 /*
718 729 * For level 2 ARC devices, add a special label.
719 730 */
720 731 VERIFY(nvlist_alloc(&label, NV_UNIQUE_NAME, KM_SLEEP) == 0);
721 732
722 733 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_VERSION,
723 734 spa_version(spa)) == 0);
724 735 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_POOL_STATE,
725 736 POOL_STATE_L2CACHE) == 0);
726 737 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_GUID,
727 738 vd->vdev_guid) == 0);
728 739 } else {
729 740 uint64_t txg = 0ULL;
730 741
731 742 if (reason == VDEV_LABEL_SPLIT)
732 743 txg = spa->spa_uberblock.ub_txg;
733 744 label = spa_config_generate(spa, vd, txg, B_FALSE);
734 745
735 746 /*
736 747 * Add our creation time. This allows us to detect multiple
737 748 * vdev uses as described above, and automatically expires if we
738 749 * fail.
739 750 */
740 751 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_CREATE_TXG,
741 752 crtxg) == 0);
742 753 }
743 754
744 755 buf = vp->vp_nvlist;
745 756 buflen = sizeof (vp->vp_nvlist);
746 757
747 758 error = nvlist_pack(label, &buf, &buflen, NV_ENCODE_XDR, KM_SLEEP);
748 759 if (error != 0) {
749 760 nvlist_free(label);
750 761 zio_buf_free(vp, sizeof (vdev_phys_t));
751 762 /* EFAULT means nvlist_pack ran out of room */
752 763 return (error == EFAULT ? ENAMETOOLONG : EINVAL);
753 764 }
754 765
755 766 /*
756 767 * Initialize uberblock template.
757 768 */
758 769 ub = zio_buf_alloc(VDEV_UBERBLOCK_RING);
759 770 bzero(ub, VDEV_UBERBLOCK_RING);
760 771 *ub = spa->spa_uberblock;
761 772 ub->ub_txg = 0;
762 773
763 774 /* Initialize the 2nd padding area. */
764 775 pad2 = zio_buf_alloc(VDEV_PAD_SIZE);
765 776 bzero(pad2, VDEV_PAD_SIZE);
766 777
767 778 /*
768 779 * Write everything in parallel.
769 780 */
770 781 retry:
771 782 zio = zio_root(spa, NULL, NULL, flags);
772 783
773 784 for (int l = 0; l < VDEV_LABELS; l++) {
774 785
775 786 vdev_label_write(zio, vd, l, vp,
776 787 offsetof(vdev_label_t, vl_vdev_phys),
777 788 sizeof (vdev_phys_t), NULL, NULL, flags);
778 789
779 790 /*
780 791 * Skip the 1st padding area.
781 792 * Zero out the 2nd padding area where it might have
782 793 * left over data from previous filesystem format.
783 794 */
784 795 vdev_label_write(zio, vd, l, pad2,
785 796 offsetof(vdev_label_t, vl_pad2),
786 797 VDEV_PAD_SIZE, NULL, NULL, flags);
787 798
788 799 vdev_label_write(zio, vd, l, ub,
789 800 offsetof(vdev_label_t, vl_uberblock),
790 801 VDEV_UBERBLOCK_RING, NULL, NULL, flags);
791 802 }
792 803
793 804 error = zio_wait(zio);
794 805
795 806 if (error != 0 && !(flags & ZIO_FLAG_TRYHARD)) {
796 807 flags |= ZIO_FLAG_TRYHARD;
797 808 goto retry;
798 809 }
799 810
800 811 nvlist_free(label);
801 812 zio_buf_free(pad2, VDEV_PAD_SIZE);
802 813 zio_buf_free(ub, VDEV_UBERBLOCK_RING);
803 814 zio_buf_free(vp, sizeof (vdev_phys_t));
804 815
805 816 /*
806 817 * If this vdev hasn't been previously identified as a spare, then we
807 818 * mark it as such only if a) we are labeling it as a spare, or b) it
808 819 * exists as a spare elsewhere in the system. Do the same for
809 820 * level 2 ARC devices.
810 821 */
811 822 if (error == 0 && !vd->vdev_isspare &&
812 823 (reason == VDEV_LABEL_SPARE ||
813 824 spa_spare_exists(vd->vdev_guid, NULL, NULL)))
814 825 spa_spare_add(vd);
815 826
816 827 if (error == 0 && !vd->vdev_isl2cache &&
817 828 (reason == VDEV_LABEL_L2CACHE ||
818 829 spa_l2cache_exists(vd->vdev_guid, NULL)))
819 830 spa_l2cache_add(vd);
820 831
821 832 return (error);
822 833 }
823 834
824 835 /*
825 836 * ==========================================================================
|
↓ open down ↓ |
316 lines elided |
↑ open up ↑ |
826 837 * uberblock load/sync
827 838 * ==========================================================================
828 839 */
829 840
830 841 /*
831 842 * Consider the following situation: txg is safely synced to disk. We've
832 843 * written the first uberblock for txg + 1, and then we lose power. When we
833 844 * come back up, we fail to see the uberblock for txg + 1 because, say,
834 845 * it was on a mirrored device and the replica to which we wrote txg + 1
835 846 * is now offline. If we then make some changes and sync txg + 1, and then
836 - * the missing replica comes back, then for a new seconds we'll have two
847 + * the missing replica comes back, then for a few seconds we'll have two
837 848 * conflicting uberblocks on disk with the same txg. The solution is simple:
838 849 * among uberblocks with equal txg, choose the one with the latest timestamp.
839 850 */
840 851 static int
841 852 vdev_uberblock_compare(uberblock_t *ub1, uberblock_t *ub2)
842 853 {
843 854 if (ub1->ub_txg < ub2->ub_txg)
844 855 return (-1);
845 856 if (ub1->ub_txg > ub2->ub_txg)
846 857 return (1);
847 858
848 859 if (ub1->ub_timestamp < ub2->ub_timestamp)
849 860 return (-1);
850 861 if (ub1->ub_timestamp > ub2->ub_timestamp)
851 862 return (1);
852 863
853 864 return (0);
854 865 }
855 866
867 +struct ubl_cbdata {
868 + uberblock_t *ubl_ubbest; /* Best uberblock */
869 + vdev_t *ubl_vd; /* vdev associated with the above */
870 + int ubl_label; /* Label associated with the above */
871 +};
872 +
856 873 static void
857 874 vdev_uberblock_load_done(zio_t *zio)
858 875 {
876 + vdev_t *vd = zio->io_vd;
859 877 spa_t *spa = zio->io_spa;
860 878 zio_t *rio = zio->io_private;
861 879 uberblock_t *ub = zio->io_data;
862 - uberblock_t *ubbest = rio->io_private;
880 + struct ubl_cbdata *cbp = rio->io_private;
863 881
864 - ASSERT3U(zio->io_size, ==, VDEV_UBERBLOCK_SIZE(zio->io_vd));
882 + ASSERT3U(zio->io_size, ==, VDEV_UBERBLOCK_SIZE(vd));
865 883
866 884 if (zio->io_error == 0 && uberblock_verify(ub) == 0) {
867 885 mutex_enter(&rio->io_lock);
868 886 if (ub->ub_txg <= spa->spa_load_max_txg &&
869 - vdev_uberblock_compare(ub, ubbest) > 0)
870 - *ubbest = *ub;
887 + vdev_uberblock_compare(ub, cbp->ubl_ubbest) > 0) {
888 + /*
889 + * Keep track of the vdev and label in which this
890 + * uberblock was found. We will use this information
891 + * later to obtain the config nvlist associated with
892 + * this uberblock.
893 + */
894 + *cbp->ubl_ubbest = *ub;
895 + cbp->ubl_vd = vd;
896 + cbp->ubl_label = vdev_label_number(vd->vdev_psize,
897 + zio->io_offset);
898 + }
871 899 mutex_exit(&rio->io_lock);
872 900 }
873 901
874 902 zio_buf_free(zio->io_data, zio->io_size);
875 903 }
876 904
877 -void
878 -vdev_uberblock_load(zio_t *zio, vdev_t *vd, uberblock_t *ubbest)
905 +static void
906 +vdev_uberblock_load_impl(zio_t *zio, vdev_t *vd, int flags,
907 + struct ubl_cbdata *cbp)
879 908 {
880 - spa_t *spa = vd->vdev_spa;
881 - vdev_t *rvd = spa->spa_root_vdev;
882 - int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL |
883 - ZIO_FLAG_SPECULATIVE | ZIO_FLAG_TRYHARD;
884 -
885 - if (vd == rvd) {
886 - ASSERT(zio == NULL);
887 - spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
888 - zio = zio_root(spa, NULL, ubbest, flags);
889 - bzero(ubbest, sizeof (uberblock_t));
890 - }
891 -
892 - ASSERT(zio != NULL);
893 -
894 909 for (int c = 0; c < vd->vdev_children; c++)
895 - vdev_uberblock_load(zio, vd->vdev_child[c], ubbest);
910 + vdev_uberblock_load_impl(zio, vd->vdev_child[c], flags, cbp);
896 911
897 912 if (vd->vdev_ops->vdev_op_leaf && vdev_readable(vd)) {
898 913 for (int l = 0; l < VDEV_LABELS; l++) {
899 914 for (int n = 0; n < VDEV_UBERBLOCK_COUNT(vd); n++) {
900 915 vdev_label_read(zio, vd, l,
901 916 zio_buf_alloc(VDEV_UBERBLOCK_SIZE(vd)),
902 917 VDEV_UBERBLOCK_OFFSET(vd, n),
903 918 VDEV_UBERBLOCK_SIZE(vd),
904 919 vdev_uberblock_load_done, zio, flags);
905 920 }
906 921 }
907 922 }
923 +}
908 924
909 - if (vd == rvd) {
910 - (void) zio_wait(zio);
911 - spa_config_exit(spa, SCL_ALL, FTAG);
925 +/*
926 + * Reads the 'best' uberblock from disk along with its associated
927 + * configuration. First, we read the uberblock array of each label of each
928 + * vdev, keeping track of the uberblock with the highest txg in each array.
929 + * Then, we read the configuration from the same label as the best uberblock.
930 + */
931 +void
932 +vdev_uberblock_load(vdev_t *rvd, uberblock_t *ub, nvlist_t **config)
933 +{
934 + int i;
935 + zio_t *zio;
936 + spa_t *spa = rvd->vdev_spa;
937 + struct ubl_cbdata cb;
938 + int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL |
939 + ZIO_FLAG_SPECULATIVE | ZIO_FLAG_TRYHARD;
940 +
941 + ASSERT(ub);
942 + ASSERT(config);
943 +
944 + bzero(ub, sizeof (uberblock_t));
945 + *config = NULL;
946 +
947 + cb.ubl_ubbest = ub;
948 + cb.ubl_vd = NULL;
949 +
950 + spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
951 + zio = zio_root(spa, NULL, &cb, flags);
952 + vdev_uberblock_load_impl(zio, rvd, flags, &cb);
953 + (void) zio_wait(zio);
954 + if (cb.ubl_vd != NULL) {
955 + for (i = cb.ubl_label % 2; i < VDEV_LABELS; i += 2) {
956 + *config = vdev_label_read_config(cb.ubl_vd, i);
957 + if (*config != NULL)
958 + break;
959 + }
912 960 }
961 + spa_config_exit(spa, SCL_ALL, FTAG);
913 962 }
914 963
915 964 /*
916 965 * On success, increment root zio's count of good writes.
917 966 * We only get credit for writes to known-visible vdevs; see spa_vdev_add().
918 967 */
919 968 static void
920 969 vdev_uberblock_sync_done(zio_t *zio)
921 970 {
922 971 uint64_t *good_writes = zio->io_private;
923 972
924 973 if (zio->io_error == 0 && zio->io_vd->vdev_top->vdev_ms_array != 0)
925 974 atomic_add_64(good_writes, 1);
926 975 }
927 976
928 977 /*
929 978 * Write the uberblock to all labels of all leaves of the specified vdev.
930 979 */
931 980 static void
932 981 vdev_uberblock_sync(zio_t *zio, uberblock_t *ub, vdev_t *vd, int flags)
933 982 {
934 983 uberblock_t *ubbuf;
935 984 int n;
936 985
937 986 for (int c = 0; c < vd->vdev_children; c++)
938 987 vdev_uberblock_sync(zio, ub, vd->vdev_child[c], flags);
939 988
940 989 if (!vd->vdev_ops->vdev_op_leaf)
941 990 return;
942 991
943 992 if (!vdev_writeable(vd))
944 993 return;
945 994
946 995 n = ub->ub_txg & (VDEV_UBERBLOCK_COUNT(vd) - 1);
947 996
948 997 ubbuf = zio_buf_alloc(VDEV_UBERBLOCK_SIZE(vd));
949 998 bzero(ubbuf, VDEV_UBERBLOCK_SIZE(vd));
950 999 *ubbuf = *ub;
951 1000
952 1001 for (int l = 0; l < VDEV_LABELS; l++)
953 1002 vdev_label_write(zio, vd, l, ubbuf,
954 1003 VDEV_UBERBLOCK_OFFSET(vd, n), VDEV_UBERBLOCK_SIZE(vd),
955 1004 vdev_uberblock_sync_done, zio->io_private,
956 1005 flags | ZIO_FLAG_DONT_PROPAGATE);
957 1006
958 1007 zio_buf_free(ubbuf, VDEV_UBERBLOCK_SIZE(vd));
959 1008 }
960 1009
961 1010 int
962 1011 vdev_uberblock_sync_list(vdev_t **svd, int svdcount, uberblock_t *ub, int flags)
963 1012 {
964 1013 spa_t *spa = svd[0]->vdev_spa;
965 1014 zio_t *zio;
966 1015 uint64_t good_writes = 0;
967 1016
968 1017 zio = zio_root(spa, NULL, &good_writes, flags);
969 1018
970 1019 for (int v = 0; v < svdcount; v++)
971 1020 vdev_uberblock_sync(zio, ub, svd[v], flags);
972 1021
973 1022 (void) zio_wait(zio);
974 1023
975 1024 /*
976 1025 * Flush the uberblocks to disk. This ensures that the odd labels
977 1026 * are no longer needed (because the new uberblocks and the even
978 1027 * labels are safely on disk), so it is safe to overwrite them.
979 1028 */
980 1029 zio = zio_root(spa, NULL, NULL, flags);
981 1030
982 1031 for (int v = 0; v < svdcount; v++)
983 1032 zio_flush(zio, svd[v]);
984 1033
985 1034 (void) zio_wait(zio);
986 1035
987 1036 return (good_writes >= 1 ? 0 : EIO);
988 1037 }
989 1038
990 1039 /*
991 1040 * On success, increment the count of good writes for our top-level vdev.
992 1041 */
993 1042 static void
994 1043 vdev_label_sync_done(zio_t *zio)
995 1044 {
996 1045 uint64_t *good_writes = zio->io_private;
997 1046
998 1047 if (zio->io_error == 0)
999 1048 atomic_add_64(good_writes, 1);
1000 1049 }
1001 1050
1002 1051 /*
1003 1052 * If there weren't enough good writes, indicate failure to the parent.
1004 1053 */
1005 1054 static void
1006 1055 vdev_label_sync_top_done(zio_t *zio)
1007 1056 {
1008 1057 uint64_t *good_writes = zio->io_private;
1009 1058
1010 1059 if (*good_writes == 0)
1011 1060 zio->io_error = EIO;
1012 1061
1013 1062 kmem_free(good_writes, sizeof (uint64_t));
1014 1063 }
1015 1064
1016 1065 /*
1017 1066 * We ignore errors for log and cache devices, simply free the private data.
1018 1067 */
1019 1068 static void
1020 1069 vdev_label_sync_ignore_done(zio_t *zio)
1021 1070 {
1022 1071 kmem_free(zio->io_private, sizeof (uint64_t));
1023 1072 }
1024 1073
1025 1074 /*
1026 1075 * Write all even or odd labels to all leaves of the specified vdev.
1027 1076 */
1028 1077 static void
1029 1078 vdev_label_sync(zio_t *zio, vdev_t *vd, int l, uint64_t txg, int flags)
1030 1079 {
1031 1080 nvlist_t *label;
1032 1081 vdev_phys_t *vp;
1033 1082 char *buf;
1034 1083 size_t buflen;
1035 1084
1036 1085 for (int c = 0; c < vd->vdev_children; c++)
1037 1086 vdev_label_sync(zio, vd->vdev_child[c], l, txg, flags);
1038 1087
1039 1088 if (!vd->vdev_ops->vdev_op_leaf)
1040 1089 return;
1041 1090
1042 1091 if (!vdev_writeable(vd))
1043 1092 return;
1044 1093
1045 1094 /*
1046 1095 * Generate a label describing the top-level config to which we belong.
1047 1096 */
1048 1097 label = spa_config_generate(vd->vdev_spa, vd, txg, B_FALSE);
1049 1098
1050 1099 vp = zio_buf_alloc(sizeof (vdev_phys_t));
1051 1100 bzero(vp, sizeof (vdev_phys_t));
1052 1101
1053 1102 buf = vp->vp_nvlist;
1054 1103 buflen = sizeof (vp->vp_nvlist);
1055 1104
1056 1105 if (nvlist_pack(label, &buf, &buflen, NV_ENCODE_XDR, KM_SLEEP) == 0) {
1057 1106 for (; l < VDEV_LABELS; l += 2) {
1058 1107 vdev_label_write(zio, vd, l, vp,
1059 1108 offsetof(vdev_label_t, vl_vdev_phys),
1060 1109 sizeof (vdev_phys_t),
1061 1110 vdev_label_sync_done, zio->io_private,
1062 1111 flags | ZIO_FLAG_DONT_PROPAGATE);
1063 1112 }
1064 1113 }
1065 1114
1066 1115 zio_buf_free(vp, sizeof (vdev_phys_t));
1067 1116 nvlist_free(label);
1068 1117 }
1069 1118
1070 1119 int
1071 1120 vdev_label_sync_list(spa_t *spa, int l, uint64_t txg, int flags)
1072 1121 {
1073 1122 list_t *dl = &spa->spa_config_dirty_list;
1074 1123 vdev_t *vd;
1075 1124 zio_t *zio;
1076 1125 int error;
1077 1126
1078 1127 /*
1079 1128 * Write the new labels to disk.
1080 1129 */
1081 1130 zio = zio_root(spa, NULL, NULL, flags);
1082 1131
1083 1132 for (vd = list_head(dl); vd != NULL; vd = list_next(dl, vd)) {
1084 1133 uint64_t *good_writes = kmem_zalloc(sizeof (uint64_t),
1085 1134 KM_SLEEP);
1086 1135
1087 1136 ASSERT(!vd->vdev_ishole);
1088 1137
1089 1138 zio_t *vio = zio_null(zio, spa, NULL,
1090 1139 (vd->vdev_islog || vd->vdev_aux != NULL) ?
1091 1140 vdev_label_sync_ignore_done : vdev_label_sync_top_done,
1092 1141 good_writes, flags);
1093 1142 vdev_label_sync(vio, vd, l, txg, flags);
1094 1143 zio_nowait(vio);
1095 1144 }
1096 1145
1097 1146 error = zio_wait(zio);
1098 1147
1099 1148 /*
1100 1149 * Flush the new labels to disk.
1101 1150 */
1102 1151 zio = zio_root(spa, NULL, NULL, flags);
1103 1152
1104 1153 for (vd = list_head(dl); vd != NULL; vd = list_next(dl, vd))
1105 1154 zio_flush(zio, vd);
1106 1155
1107 1156 (void) zio_wait(zio);
1108 1157
1109 1158 return (error);
1110 1159 }
1111 1160
1112 1161 /*
1113 1162 * Sync the uberblock and any changes to the vdev configuration.
1114 1163 *
1115 1164 * The order of operations is carefully crafted to ensure that
1116 1165 * if the system panics or loses power at any time, the state on disk
1117 1166 * is still transactionally consistent. The in-line comments below
1118 1167 * describe the failure semantics at each stage.
1119 1168 *
1120 1169 * Moreover, vdev_config_sync() is designed to be idempotent: if it fails
1121 1170 * at any time, you can just call it again, and it will resume its work.
1122 1171 */
1123 1172 int
1124 1173 vdev_config_sync(vdev_t **svd, int svdcount, uint64_t txg, boolean_t tryhard)
1125 1174 {
1126 1175 spa_t *spa = svd[0]->vdev_spa;
1127 1176 uberblock_t *ub = &spa->spa_uberblock;
1128 1177 vdev_t *vd;
1129 1178 zio_t *zio;
1130 1179 int error;
1131 1180 int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL;
1132 1181
1133 1182 /*
1134 1183 * Normally, we don't want to try too hard to write every label and
1135 1184 * uberblock. If there is a flaky disk, we don't want the rest of the
1136 1185 * sync process to block while we retry. But if we can't write a
1137 1186 * single label out, we should retry with ZIO_FLAG_TRYHARD before
1138 1187 * bailing out and declaring the pool faulted.
1139 1188 */
1140 1189 if (tryhard)
1141 1190 flags |= ZIO_FLAG_TRYHARD;
1142 1191
1143 1192 ASSERT(ub->ub_txg <= txg);
1144 1193
1145 1194 /*
1146 1195 * If this isn't a resync due to I/O errors,
1147 1196 * and nothing changed in this transaction group,
1148 1197 * and the vdev configuration hasn't changed,
1149 1198 * then there's nothing to do.
1150 1199 */
1151 1200 if (ub->ub_txg < txg &&
1152 1201 uberblock_update(ub, spa->spa_root_vdev, txg) == B_FALSE &&
1153 1202 list_is_empty(&spa->spa_config_dirty_list))
1154 1203 return (0);
1155 1204
1156 1205 if (txg > spa_freeze_txg(spa))
1157 1206 return (0);
1158 1207
1159 1208 ASSERT(txg <= spa->spa_final_txg);
1160 1209
1161 1210 /*
1162 1211 * Flush the write cache of every disk that's been written to
1163 1212 * in this transaction group. This ensures that all blocks
1164 1213 * written in this txg will be committed to stable storage
1165 1214 * before any uberblock that references them.
1166 1215 */
1167 1216 zio = zio_root(spa, NULL, NULL, flags);
1168 1217
1169 1218 for (vd = txg_list_head(&spa->spa_vdev_txg_list, TXG_CLEAN(txg)); vd;
1170 1219 vd = txg_list_next(&spa->spa_vdev_txg_list, vd, TXG_CLEAN(txg)))
1171 1220 zio_flush(zio, vd);
1172 1221
1173 1222 (void) zio_wait(zio);
1174 1223
1175 1224 /*
1176 1225 * Sync out the even labels (L0, L2) for every dirty vdev. If the
1177 1226 * system dies in the middle of this process, that's OK: all of the
1178 1227 * even labels that made it to disk will be newer than any uberblock,
1179 1228 * and will therefore be considered invalid. The odd labels (L1, L3),
1180 1229 * which have not yet been touched, will still be valid. We flush
1181 1230 * the new labels to disk to ensure that all even-label updates
1182 1231 * are committed to stable storage before the uberblock update.
1183 1232 */
1184 1233 if ((error = vdev_label_sync_list(spa, 0, txg, flags)) != 0)
1185 1234 return (error);
1186 1235
1187 1236 /*
1188 1237 * Sync the uberblocks to all vdevs in svd[].
1189 1238 * If the system dies in the middle of this step, there are two cases
1190 1239 * to consider, and the on-disk state is consistent either way:
1191 1240 *
1192 1241 * (1) If none of the new uberblocks made it to disk, then the
1193 1242 * previous uberblock will be the newest, and the odd labels
1194 1243 * (which had not yet been touched) will be valid with respect
1195 1244 * to that uberblock.
1196 1245 *
1197 1246 * (2) If one or more new uberblocks made it to disk, then they
1198 1247 * will be the newest, and the even labels (which had all
1199 1248 * been successfully committed) will be valid with respect
1200 1249 * to the new uberblocks.
1201 1250 */
1202 1251 if ((error = vdev_uberblock_sync_list(svd, svdcount, ub, flags)) != 0)
1203 1252 return (error);
1204 1253
1205 1254 /*
1206 1255 * Sync out odd labels for every dirty vdev. If the system dies
1207 1256 * in the middle of this process, the even labels and the new
1208 1257 * uberblocks will suffice to open the pool. The next time
1209 1258 * the pool is opened, the first thing we'll do -- before any
1210 1259 * user data is modified -- is mark every vdev dirty so that
1211 1260 * all labels will be brought up to date. We flush the new labels
1212 1261 * to disk to ensure that all odd-label updates are committed to
1213 1262 * stable storage before the next transaction group begins.
1214 1263 */
1215 1264 return (vdev_label_sync_list(spa, 1, txg, flags));
1216 1265 }
|
↓ open down ↓ |
294 lines elided |
↑ open up ↑ |
XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX