Print this page
3525 Persistent L2ARC
| 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 }
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
219 219 nv = fnvlist_alloc();
220 220
221 221 fnvlist_add_string(nv, ZPOOL_CONFIG_TYPE, vd->vdev_ops->vdev_op_type);
222 222 if (!(flags & (VDEV_CONFIG_SPARE | VDEV_CONFIG_L2CACHE)))
223 223 fnvlist_add_uint64(nv, ZPOOL_CONFIG_ID, vd->vdev_id);
224 224 fnvlist_add_uint64(nv, ZPOOL_CONFIG_GUID, vd->vdev_guid);
225 225
226 226 if (vd->vdev_path != NULL)
227 227 fnvlist_add_string(nv, ZPOOL_CONFIG_PATH, vd->vdev_path);
228 228
229 229 if (vd->vdev_devid != NULL)
230 230 fnvlist_add_string(nv, ZPOOL_CONFIG_DEVID, vd->vdev_devid);
231 231
232 232 if (vd->vdev_physpath != NULL)
233 233 fnvlist_add_string(nv, ZPOOL_CONFIG_PHYS_PATH,
234 234 vd->vdev_physpath);
235 235
236 236 if (vd->vdev_fru != NULL)
237 237 fnvlist_add_string(nv, ZPOOL_CONFIG_FRU, vd->vdev_fru);
238 238
239 239 if (vd->vdev_nparity != 0) {
240 240 ASSERT(strcmp(vd->vdev_ops->vdev_op_type,
241 241 VDEV_TYPE_RAIDZ) == 0);
242 242
243 243 /*
244 244 * Make sure someone hasn't managed to sneak a fancy new vdev
245 245 * into a crufty old storage pool.
246 246 */
247 247 ASSERT(vd->vdev_nparity == 1 ||
248 248 (vd->vdev_nparity <= 2 &&
249 249 spa_version(spa) >= SPA_VERSION_RAIDZ2) ||
250 250 (vd->vdev_nparity <= 3 &&
251 251 spa_version(spa) >= SPA_VERSION_RAIDZ3));
252 252
253 253 /*
254 254 * Note that we'll add the nparity tag even on storage pools
255 255 * that only support a single parity device -- older software
256 256 * will just ignore it.
257 257 */
258 258 fnvlist_add_uint64(nv, ZPOOL_CONFIG_NPARITY, vd->vdev_nparity);
259 259 }
260 260
261 261 if (vd->vdev_wholedisk != -1ULL)
262 262 fnvlist_add_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK,
263 263 vd->vdev_wholedisk);
264 264
265 265 if (vd->vdev_not_present)
266 266 fnvlist_add_uint64(nv, ZPOOL_CONFIG_NOT_PRESENT, 1);
267 267
268 268 if (vd->vdev_isspare)
269 269 fnvlist_add_uint64(nv, ZPOOL_CONFIG_IS_SPARE, 1);
270 270
271 271 if (!(flags & (VDEV_CONFIG_SPARE | VDEV_CONFIG_L2CACHE)) &&
272 272 vd == vd->vdev_top) {
273 273 fnvlist_add_uint64(nv, ZPOOL_CONFIG_METASLAB_ARRAY,
274 274 vd->vdev_ms_array);
275 275 fnvlist_add_uint64(nv, ZPOOL_CONFIG_METASLAB_SHIFT,
|
↓ open down ↓ |
275 lines elided |
↑ open up ↑ |
276 276 vd->vdev_ms_shift);
277 277 fnvlist_add_uint64(nv, ZPOOL_CONFIG_ASHIFT, vd->vdev_ashift);
278 278 fnvlist_add_uint64(nv, ZPOOL_CONFIG_ASIZE,
279 279 vd->vdev_asize);
280 280 fnvlist_add_uint64(nv, ZPOOL_CONFIG_IS_LOG, vd->vdev_islog);
281 281 if (vd->vdev_removing)
282 282 fnvlist_add_uint64(nv, ZPOOL_CONFIG_REMOVING,
283 283 vd->vdev_removing);
284 284 }
285 285
286 + if (flags & VDEV_CONFIG_L2CACHE)
287 + /* indicate that we support L2ARC persistency */
288 + VERIFY(nvlist_add_boolean_value(nv,
289 + ZPOOL_CONFIG_L2CACHE_PERSISTENT, B_TRUE) == 0);
290 +
286 291 if (vd->vdev_dtl_sm != NULL) {
287 292 fnvlist_add_uint64(nv, ZPOOL_CONFIG_DTL,
288 293 space_map_object(vd->vdev_dtl_sm));
289 294 }
290 295
291 296 if (vd->vdev_crtxg)
292 297 fnvlist_add_uint64(nv, ZPOOL_CONFIG_CREATE_TXG, vd->vdev_crtxg);
293 298
294 299 if (getstats) {
295 300 vdev_stat_t vs;
296 301 pool_scan_stat_t ps;
297 302
298 303 vdev_get_stats(vd, &vs);
299 304 fnvlist_add_uint64_array(nv, ZPOOL_CONFIG_VDEV_STATS,
300 305 (uint64_t *)&vs, sizeof (vs) / sizeof (uint64_t));
301 306
302 307 /* provide either current or previous scan information */
303 308 if (spa_scan_get_stats(spa, &ps) == 0) {
304 309 fnvlist_add_uint64_array(nv,
305 310 ZPOOL_CONFIG_SCAN_STATS, (uint64_t *)&ps,
306 311 sizeof (pool_scan_stat_t) / sizeof (uint64_t));
307 312 }
308 313 }
309 314
310 315 if (!vd->vdev_ops->vdev_op_leaf) {
311 316 nvlist_t **child;
312 317 int c, idx;
313 318
314 319 ASSERT(!vd->vdev_ishole);
315 320
316 321 child = kmem_alloc(vd->vdev_children * sizeof (nvlist_t *),
317 322 KM_SLEEP);
318 323
319 324 for (c = 0, idx = 0; c < vd->vdev_children; c++) {
320 325 vdev_t *cvd = vd->vdev_child[c];
321 326
322 327 /*
323 328 * If we're generating an nvlist of removing
324 329 * vdevs then skip over any device which is
325 330 * not being removed.
326 331 */
327 332 if ((flags & VDEV_CONFIG_REMOVING) &&
328 333 !cvd->vdev_removing)
329 334 continue;
330 335
331 336 child[idx++] = vdev_config_generate(spa, cvd,
332 337 getstats, flags);
333 338 }
334 339
335 340 if (idx) {
336 341 fnvlist_add_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
337 342 child, idx);
338 343 }
339 344
340 345 for (c = 0; c < idx; c++)
341 346 nvlist_free(child[c]);
342 347
343 348 kmem_free(child, vd->vdev_children * sizeof (nvlist_t *));
344 349
345 350 } else {
346 351 const char *aux = NULL;
347 352
348 353 if (vd->vdev_offline && !vd->vdev_tmpoffline)
349 354 fnvlist_add_uint64(nv, ZPOOL_CONFIG_OFFLINE, B_TRUE);
350 355 if (vd->vdev_resilver_txg != 0)
351 356 fnvlist_add_uint64(nv, ZPOOL_CONFIG_RESILVER_TXG,
352 357 vd->vdev_resilver_txg);
353 358 if (vd->vdev_faulted)
354 359 fnvlist_add_uint64(nv, ZPOOL_CONFIG_FAULTED, B_TRUE);
355 360 if (vd->vdev_degraded)
356 361 fnvlist_add_uint64(nv, ZPOOL_CONFIG_DEGRADED, B_TRUE);
357 362 if (vd->vdev_removed)
358 363 fnvlist_add_uint64(nv, ZPOOL_CONFIG_REMOVED, B_TRUE);
359 364 if (vd->vdev_unspare)
360 365 fnvlist_add_uint64(nv, ZPOOL_CONFIG_UNSPARE, B_TRUE);
361 366 if (vd->vdev_ishole)
362 367 fnvlist_add_uint64(nv, ZPOOL_CONFIG_IS_HOLE, B_TRUE);
363 368
364 369 switch (vd->vdev_stat.vs_aux) {
365 370 case VDEV_AUX_ERR_EXCEEDED:
366 371 aux = "err_exceeded";
367 372 break;
368 373
369 374 case VDEV_AUX_EXTERNAL:
370 375 aux = "external";
371 376 break;
372 377 }
373 378
374 379 if (aux != NULL)
375 380 fnvlist_add_string(nv, ZPOOL_CONFIG_AUX_STATE, aux);
376 381
377 382 if (vd->vdev_splitting && vd->vdev_orig_guid != 0LL) {
378 383 fnvlist_add_uint64(nv, ZPOOL_CONFIG_ORIG_GUID,
379 384 vd->vdev_orig_guid);
380 385 }
381 386 }
382 387
383 388 return (nv);
384 389 }
385 390
386 391 /*
387 392 * Generate a view of the top-level vdevs. If we currently have holes
388 393 * in the namespace, then generate an array which contains a list of holey
389 394 * vdevs. Additionally, add the number of top-level children that currently
390 395 * exist.
391 396 */
392 397 void
393 398 vdev_top_config_generate(spa_t *spa, nvlist_t *config)
394 399 {
395 400 vdev_t *rvd = spa->spa_root_vdev;
396 401 uint64_t *array;
397 402 uint_t c, idx;
398 403
399 404 array = kmem_alloc(rvd->vdev_children * sizeof (uint64_t), KM_SLEEP);
400 405
401 406 for (c = 0, idx = 0; c < rvd->vdev_children; c++) {
402 407 vdev_t *tvd = rvd->vdev_child[c];
403 408
404 409 if (tvd->vdev_ishole)
405 410 array[idx++] = c;
406 411 }
407 412
408 413 if (idx) {
409 414 VERIFY(nvlist_add_uint64_array(config, ZPOOL_CONFIG_HOLE_ARRAY,
410 415 array, idx) == 0);
411 416 }
412 417
413 418 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VDEV_CHILDREN,
414 419 rvd->vdev_children) == 0);
415 420
416 421 kmem_free(array, rvd->vdev_children * sizeof (uint64_t));
417 422 }
418 423
419 424 /*
420 425 * Returns the configuration from the label of the given vdev. For vdevs
421 426 * which don't have a txg value stored on their label (i.e. spares/cache)
422 427 * or have not been completely initialized (txg = 0) just return
423 428 * the configuration from the first valid label we find. Otherwise,
424 429 * find the most up-to-date label that does not exceed the specified
425 430 * 'txg' value.
426 431 */
427 432 nvlist_t *
428 433 vdev_label_read_config(vdev_t *vd, uint64_t txg)
429 434 {
430 435 spa_t *spa = vd->vdev_spa;
431 436 nvlist_t *config = NULL;
432 437 vdev_phys_t *vp;
433 438 zio_t *zio;
434 439 uint64_t best_txg = 0;
435 440 int error = 0;
436 441 int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL |
437 442 ZIO_FLAG_SPECULATIVE;
438 443
439 444 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
440 445
441 446 if (!vdev_readable(vd))
442 447 return (NULL);
443 448
444 449 vp = zio_buf_alloc(sizeof (vdev_phys_t));
445 450
446 451 retry:
447 452 for (int l = 0; l < VDEV_LABELS; l++) {
448 453 nvlist_t *label = NULL;
449 454
450 455 zio = zio_root(spa, NULL, NULL, flags);
451 456
452 457 vdev_label_read(zio, vd, l, vp,
453 458 offsetof(vdev_label_t, vl_vdev_phys),
454 459 sizeof (vdev_phys_t), NULL, NULL, flags);
455 460
456 461 if (zio_wait(zio) == 0 &&
457 462 nvlist_unpack(vp->vp_nvlist, sizeof (vp->vp_nvlist),
458 463 &label, 0) == 0) {
459 464 uint64_t label_txg = 0;
460 465
461 466 /*
462 467 * Auxiliary vdevs won't have txg values in their
463 468 * labels and newly added vdevs may not have been
464 469 * completely initialized so just return the
465 470 * configuration from the first valid label we
466 471 * encounter.
467 472 */
468 473 error = nvlist_lookup_uint64(label,
469 474 ZPOOL_CONFIG_POOL_TXG, &label_txg);
470 475 if ((error || label_txg == 0) && !config) {
471 476 config = label;
472 477 break;
473 478 } else if (label_txg <= txg && label_txg > best_txg) {
474 479 best_txg = label_txg;
475 480 nvlist_free(config);
476 481 config = fnvlist_dup(label);
477 482 }
478 483 }
479 484
480 485 if (label != NULL) {
481 486 nvlist_free(label);
482 487 label = NULL;
483 488 }
484 489 }
485 490
486 491 if (config == NULL && !(flags & ZIO_FLAG_TRYHARD)) {
487 492 flags |= ZIO_FLAG_TRYHARD;
488 493 goto retry;
489 494 }
490 495
491 496 zio_buf_free(vp, sizeof (vdev_phys_t));
492 497
493 498 return (config);
494 499 }
495 500
496 501 /*
497 502 * Determine if a device is in use. The 'spare_guid' parameter will be filled
498 503 * in with the device guid if this spare is active elsewhere on the system.
499 504 */
500 505 static boolean_t
501 506 vdev_inuse(vdev_t *vd, uint64_t crtxg, vdev_labeltype_t reason,
502 507 uint64_t *spare_guid, uint64_t *l2cache_guid)
503 508 {
504 509 spa_t *spa = vd->vdev_spa;
505 510 uint64_t state, pool_guid, device_guid, txg, spare_pool;
506 511 uint64_t vdtxg = 0;
507 512 nvlist_t *label;
508 513
509 514 if (spare_guid)
510 515 *spare_guid = 0ULL;
511 516 if (l2cache_guid)
512 517 *l2cache_guid = 0ULL;
513 518
514 519 /*
515 520 * Read the label, if any, and perform some basic sanity checks.
516 521 */
517 522 if ((label = vdev_label_read_config(vd, -1ULL)) == NULL)
518 523 return (B_FALSE);
519 524
520 525 (void) nvlist_lookup_uint64(label, ZPOOL_CONFIG_CREATE_TXG,
521 526 &vdtxg);
522 527
523 528 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE,
524 529 &state) != 0 ||
525 530 nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID,
526 531 &device_guid) != 0) {
527 532 nvlist_free(label);
528 533 return (B_FALSE);
529 534 }
530 535
531 536 if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
532 537 (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_GUID,
533 538 &pool_guid) != 0 ||
534 539 nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG,
535 540 &txg) != 0)) {
536 541 nvlist_free(label);
537 542 return (B_FALSE);
538 543 }
539 544
540 545 nvlist_free(label);
541 546
542 547 /*
543 548 * Check to see if this device indeed belongs to the pool it claims to
544 549 * be a part of. The only way this is allowed is if the device is a hot
545 550 * spare (which we check for later on).
546 551 */
547 552 if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
548 553 !spa_guid_exists(pool_guid, device_guid) &&
549 554 !spa_spare_exists(device_guid, NULL, NULL) &&
550 555 !spa_l2cache_exists(device_guid, NULL))
551 556 return (B_FALSE);
552 557
553 558 /*
554 559 * If the transaction group is zero, then this an initialized (but
555 560 * unused) label. This is only an error if the create transaction
556 561 * on-disk is the same as the one we're using now, in which case the
557 562 * user has attempted to add the same vdev multiple times in the same
558 563 * transaction.
559 564 */
560 565 if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
561 566 txg == 0 && vdtxg == crtxg)
562 567 return (B_TRUE);
563 568
564 569 /*
565 570 * Check to see if this is a spare device. We do an explicit check for
566 571 * spa_has_spare() here because it may be on our pending list of spares
567 572 * to add. We also check if it is an l2cache device.
568 573 */
569 574 if (spa_spare_exists(device_guid, &spare_pool, NULL) ||
570 575 spa_has_spare(spa, device_guid)) {
571 576 if (spare_guid)
572 577 *spare_guid = device_guid;
573 578
574 579 switch (reason) {
575 580 case VDEV_LABEL_CREATE:
576 581 case VDEV_LABEL_L2CACHE:
577 582 return (B_TRUE);
578 583
579 584 case VDEV_LABEL_REPLACE:
580 585 return (!spa_has_spare(spa, device_guid) ||
581 586 spare_pool != 0ULL);
582 587
583 588 case VDEV_LABEL_SPARE:
584 589 return (spa_has_spare(spa, device_guid));
585 590 }
586 591 }
587 592
588 593 /*
589 594 * Check to see if this is an l2cache device.
590 595 */
591 596 if (spa_l2cache_exists(device_guid, NULL))
592 597 return (B_TRUE);
593 598
594 599 /*
595 600 * We can't rely on a pool's state if it's been imported
596 601 * read-only. Instead we look to see if the pools is marked
597 602 * read-only in the namespace and set the state to active.
598 603 */
599 604 if ((spa = spa_by_guid(pool_guid, device_guid)) != NULL &&
600 605 spa_mode(spa) == FREAD)
601 606 state = POOL_STATE_ACTIVE;
602 607
603 608 /*
604 609 * If the device is marked ACTIVE, then this device is in use by another
605 610 * pool on the system.
606 611 */
607 612 return (state == POOL_STATE_ACTIVE);
608 613 }
609 614
610 615 /*
611 616 * Initialize a vdev label. We check to make sure each leaf device is not in
612 617 * use, and writable. We put down an initial label which we will later
613 618 * overwrite with a complete label. Note that it's important to do this
614 619 * sequentially, not in parallel, so that we catch cases of multiple use of the
615 620 * same leaf vdev in the vdev we're creating -- e.g. mirroring a disk with
616 621 * itself.
617 622 */
618 623 int
619 624 vdev_label_init(vdev_t *vd, uint64_t crtxg, vdev_labeltype_t reason)
620 625 {
621 626 spa_t *spa = vd->vdev_spa;
622 627 nvlist_t *label;
623 628 vdev_phys_t *vp;
624 629 char *pad2;
625 630 uberblock_t *ub;
626 631 zio_t *zio;
627 632 char *buf;
628 633 size_t buflen;
629 634 int error;
630 635 uint64_t spare_guid, l2cache_guid;
631 636 int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL;
632 637
633 638 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
634 639
635 640 for (int c = 0; c < vd->vdev_children; c++)
636 641 if ((error = vdev_label_init(vd->vdev_child[c],
637 642 crtxg, reason)) != 0)
638 643 return (error);
639 644
640 645 /* Track the creation time for this vdev */
641 646 vd->vdev_crtxg = crtxg;
642 647
643 648 if (!vd->vdev_ops->vdev_op_leaf)
644 649 return (0);
645 650
646 651 /*
647 652 * Dead vdevs cannot be initialized.
648 653 */
649 654 if (vdev_is_dead(vd))
650 655 return (SET_ERROR(EIO));
651 656
652 657 /*
653 658 * Determine if the vdev is in use.
654 659 */
655 660 if (reason != VDEV_LABEL_REMOVE && reason != VDEV_LABEL_SPLIT &&
656 661 vdev_inuse(vd, crtxg, reason, &spare_guid, &l2cache_guid))
657 662 return (SET_ERROR(EBUSY));
658 663
659 664 /*
660 665 * If this is a request to add or replace a spare or l2cache device
661 666 * that is in use elsewhere on the system, then we must update the
662 667 * guid (which was initialized to a random value) to reflect the
663 668 * actual GUID (which is shared between multiple pools).
664 669 */
665 670 if (reason != VDEV_LABEL_REMOVE && reason != VDEV_LABEL_L2CACHE &&
666 671 spare_guid != 0ULL) {
667 672 uint64_t guid_delta = spare_guid - vd->vdev_guid;
668 673
669 674 vd->vdev_guid += guid_delta;
670 675
671 676 for (vdev_t *pvd = vd; pvd != NULL; pvd = pvd->vdev_parent)
672 677 pvd->vdev_guid_sum += guid_delta;
673 678
674 679 /*
675 680 * If this is a replacement, then we want to fallthrough to the
676 681 * rest of the code. If we're adding a spare, then it's already
677 682 * labeled appropriately and we can just return.
678 683 */
679 684 if (reason == VDEV_LABEL_SPARE)
680 685 return (0);
681 686 ASSERT(reason == VDEV_LABEL_REPLACE ||
682 687 reason == VDEV_LABEL_SPLIT);
683 688 }
684 689
685 690 if (reason != VDEV_LABEL_REMOVE && reason != VDEV_LABEL_SPARE &&
686 691 l2cache_guid != 0ULL) {
687 692 uint64_t guid_delta = l2cache_guid - vd->vdev_guid;
688 693
689 694 vd->vdev_guid += guid_delta;
690 695
691 696 for (vdev_t *pvd = vd; pvd != NULL; pvd = pvd->vdev_parent)
692 697 pvd->vdev_guid_sum += guid_delta;
693 698
694 699 /*
695 700 * If this is a replacement, then we want to fallthrough to the
696 701 * rest of the code. If we're adding an l2cache, then it's
697 702 * already labeled appropriately and we can just return.
698 703 */
699 704 if (reason == VDEV_LABEL_L2CACHE)
700 705 return (0);
701 706 ASSERT(reason == VDEV_LABEL_REPLACE);
702 707 }
703 708
704 709 /*
705 710 * Initialize its label.
706 711 */
707 712 vp = zio_buf_alloc(sizeof (vdev_phys_t));
708 713 bzero(vp, sizeof (vdev_phys_t));
709 714
710 715 /*
711 716 * Generate a label describing the pool and our top-level vdev.
712 717 * We mark it as being from txg 0 to indicate that it's not
713 718 * really part of an active pool just yet. The labels will
714 719 * be written again with a meaningful txg by spa_sync().
715 720 */
716 721 if (reason == VDEV_LABEL_SPARE ||
717 722 (reason == VDEV_LABEL_REMOVE && vd->vdev_isspare)) {
718 723 /*
719 724 * For inactive hot spares, we generate a special label that
720 725 * identifies as a mutually shared hot spare. We write the
721 726 * label if we are adding a hot spare, or if we are removing an
722 727 * active hot spare (in which case we want to revert the
723 728 * labels).
724 729 */
725 730 VERIFY(nvlist_alloc(&label, NV_UNIQUE_NAME, KM_SLEEP) == 0);
726 731
727 732 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_VERSION,
728 733 spa_version(spa)) == 0);
729 734 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_POOL_STATE,
730 735 POOL_STATE_SPARE) == 0);
731 736 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_GUID,
732 737 vd->vdev_guid) == 0);
733 738 } else if (reason == VDEV_LABEL_L2CACHE ||
734 739 (reason == VDEV_LABEL_REMOVE && vd->vdev_isl2cache)) {
735 740 /*
736 741 * For level 2 ARC devices, add a special label.
737 742 */
738 743 VERIFY(nvlist_alloc(&label, NV_UNIQUE_NAME, KM_SLEEP) == 0);
739 744
740 745 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_VERSION,
741 746 spa_version(spa)) == 0);
742 747 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_POOL_STATE,
743 748 POOL_STATE_L2CACHE) == 0);
744 749 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_GUID,
745 750 vd->vdev_guid) == 0);
746 751 } else {
747 752 uint64_t txg = 0ULL;
748 753
749 754 if (reason == VDEV_LABEL_SPLIT)
750 755 txg = spa->spa_uberblock.ub_txg;
751 756 label = spa_config_generate(spa, vd, txg, B_FALSE);
752 757
753 758 /*
754 759 * Add our creation time. This allows us to detect multiple
755 760 * vdev uses as described above, and automatically expires if we
756 761 * fail.
757 762 */
758 763 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_CREATE_TXG,
759 764 crtxg) == 0);
760 765 }
761 766
762 767 buf = vp->vp_nvlist;
763 768 buflen = sizeof (vp->vp_nvlist);
764 769
765 770 error = nvlist_pack(label, &buf, &buflen, NV_ENCODE_XDR, KM_SLEEP);
766 771 if (error != 0) {
767 772 nvlist_free(label);
768 773 zio_buf_free(vp, sizeof (vdev_phys_t));
769 774 /* EFAULT means nvlist_pack ran out of room */
770 775 return (error == EFAULT ? ENAMETOOLONG : EINVAL);
771 776 }
772 777
773 778 /*
774 779 * Initialize uberblock template.
775 780 */
776 781 ub = zio_buf_alloc(VDEV_UBERBLOCK_RING);
777 782 bzero(ub, VDEV_UBERBLOCK_RING);
778 783 *ub = spa->spa_uberblock;
779 784 ub->ub_txg = 0;
780 785
781 786 /* Initialize the 2nd padding area. */
782 787 pad2 = zio_buf_alloc(VDEV_PAD_SIZE);
783 788 bzero(pad2, VDEV_PAD_SIZE);
784 789
785 790 /*
786 791 * Write everything in parallel.
787 792 */
788 793 retry:
789 794 zio = zio_root(spa, NULL, NULL, flags);
790 795
791 796 for (int l = 0; l < VDEV_LABELS; l++) {
792 797
793 798 vdev_label_write(zio, vd, l, vp,
794 799 offsetof(vdev_label_t, vl_vdev_phys),
795 800 sizeof (vdev_phys_t), NULL, NULL, flags);
796 801
797 802 /*
798 803 * Skip the 1st padding area.
799 804 * Zero out the 2nd padding area where it might have
800 805 * left over data from previous filesystem format.
801 806 */
802 807 vdev_label_write(zio, vd, l, pad2,
803 808 offsetof(vdev_label_t, vl_pad2),
804 809 VDEV_PAD_SIZE, NULL, NULL, flags);
805 810
806 811 vdev_label_write(zio, vd, l, ub,
807 812 offsetof(vdev_label_t, vl_uberblock),
808 813 VDEV_UBERBLOCK_RING, NULL, NULL, flags);
809 814 }
810 815
811 816 error = zio_wait(zio);
812 817
813 818 if (error != 0 && !(flags & ZIO_FLAG_TRYHARD)) {
814 819 flags |= ZIO_FLAG_TRYHARD;
815 820 goto retry;
816 821 }
817 822
818 823 nvlist_free(label);
819 824 zio_buf_free(pad2, VDEV_PAD_SIZE);
820 825 zio_buf_free(ub, VDEV_UBERBLOCK_RING);
821 826 zio_buf_free(vp, sizeof (vdev_phys_t));
822 827
823 828 /*
824 829 * If this vdev hasn't been previously identified as a spare, then we
825 830 * mark it as such only if a) we are labeling it as a spare, or b) it
826 831 * exists as a spare elsewhere in the system. Do the same for
827 832 * level 2 ARC devices.
828 833 */
829 834 if (error == 0 && !vd->vdev_isspare &&
830 835 (reason == VDEV_LABEL_SPARE ||
831 836 spa_spare_exists(vd->vdev_guid, NULL, NULL)))
832 837 spa_spare_add(vd);
833 838
834 839 if (error == 0 && !vd->vdev_isl2cache &&
835 840 (reason == VDEV_LABEL_L2CACHE ||
836 841 spa_l2cache_exists(vd->vdev_guid, NULL)))
837 842 spa_l2cache_add(vd);
838 843
839 844 return (error);
840 845 }
841 846
842 847 /*
843 848 * ==========================================================================
844 849 * uberblock load/sync
845 850 * ==========================================================================
846 851 */
847 852
848 853 /*
849 854 * Consider the following situation: txg is safely synced to disk. We've
850 855 * written the first uberblock for txg + 1, and then we lose power. When we
851 856 * come back up, we fail to see the uberblock for txg + 1 because, say,
852 857 * it was on a mirrored device and the replica to which we wrote txg + 1
853 858 * is now offline. If we then make some changes and sync txg + 1, and then
854 859 * the missing replica comes back, then for a few seconds we'll have two
855 860 * conflicting uberblocks on disk with the same txg. The solution is simple:
856 861 * among uberblocks with equal txg, choose the one with the latest timestamp.
857 862 */
858 863 static int
859 864 vdev_uberblock_compare(uberblock_t *ub1, uberblock_t *ub2)
860 865 {
861 866 if (ub1->ub_txg < ub2->ub_txg)
862 867 return (-1);
863 868 if (ub1->ub_txg > ub2->ub_txg)
864 869 return (1);
865 870
866 871 if (ub1->ub_timestamp < ub2->ub_timestamp)
867 872 return (-1);
868 873 if (ub1->ub_timestamp > ub2->ub_timestamp)
869 874 return (1);
870 875
871 876 return (0);
872 877 }
873 878
874 879 struct ubl_cbdata {
875 880 uberblock_t *ubl_ubbest; /* Best uberblock */
876 881 vdev_t *ubl_vd; /* vdev associated with the above */
877 882 };
878 883
879 884 static void
880 885 vdev_uberblock_load_done(zio_t *zio)
881 886 {
882 887 vdev_t *vd = zio->io_vd;
883 888 spa_t *spa = zio->io_spa;
884 889 zio_t *rio = zio->io_private;
885 890 uberblock_t *ub = zio->io_data;
886 891 struct ubl_cbdata *cbp = rio->io_private;
887 892
888 893 ASSERT3U(zio->io_size, ==, VDEV_UBERBLOCK_SIZE(vd));
889 894
890 895 if (zio->io_error == 0 && uberblock_verify(ub) == 0) {
891 896 mutex_enter(&rio->io_lock);
892 897 if (ub->ub_txg <= spa->spa_load_max_txg &&
893 898 vdev_uberblock_compare(ub, cbp->ubl_ubbest) > 0) {
894 899 /*
895 900 * Keep track of the vdev in which this uberblock
896 901 * was found. We will use this information later
897 902 * to obtain the config nvlist associated with
898 903 * this uberblock.
899 904 */
900 905 *cbp->ubl_ubbest = *ub;
901 906 cbp->ubl_vd = vd;
902 907 }
903 908 mutex_exit(&rio->io_lock);
904 909 }
905 910
906 911 zio_buf_free(zio->io_data, zio->io_size);
907 912 }
908 913
909 914 static void
910 915 vdev_uberblock_load_impl(zio_t *zio, vdev_t *vd, int flags,
911 916 struct ubl_cbdata *cbp)
912 917 {
913 918 for (int c = 0; c < vd->vdev_children; c++)
914 919 vdev_uberblock_load_impl(zio, vd->vdev_child[c], flags, cbp);
915 920
916 921 if (vd->vdev_ops->vdev_op_leaf && vdev_readable(vd)) {
917 922 for (int l = 0; l < VDEV_LABELS; l++) {
918 923 for (int n = 0; n < VDEV_UBERBLOCK_COUNT(vd); n++) {
919 924 vdev_label_read(zio, vd, l,
920 925 zio_buf_alloc(VDEV_UBERBLOCK_SIZE(vd)),
921 926 VDEV_UBERBLOCK_OFFSET(vd, n),
922 927 VDEV_UBERBLOCK_SIZE(vd),
923 928 vdev_uberblock_load_done, zio, flags);
924 929 }
925 930 }
926 931 }
927 932 }
928 933
929 934 /*
930 935 * Reads the 'best' uberblock from disk along with its associated
931 936 * configuration. First, we read the uberblock array of each label of each
932 937 * vdev, keeping track of the uberblock with the highest txg in each array.
933 938 * Then, we read the configuration from the same vdev as the best uberblock.
934 939 */
935 940 void
936 941 vdev_uberblock_load(vdev_t *rvd, uberblock_t *ub, nvlist_t **config)
937 942 {
938 943 zio_t *zio;
939 944 spa_t *spa = rvd->vdev_spa;
940 945 struct ubl_cbdata cb;
941 946 int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL |
942 947 ZIO_FLAG_SPECULATIVE | ZIO_FLAG_TRYHARD;
943 948
944 949 ASSERT(ub);
945 950 ASSERT(config);
946 951
947 952 bzero(ub, sizeof (uberblock_t));
948 953 *config = NULL;
949 954
950 955 cb.ubl_ubbest = ub;
951 956 cb.ubl_vd = NULL;
952 957
953 958 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
954 959 zio = zio_root(spa, NULL, &cb, flags);
955 960 vdev_uberblock_load_impl(zio, rvd, flags, &cb);
956 961 (void) zio_wait(zio);
957 962
958 963 /*
959 964 * It's possible that the best uberblock was discovered on a label
960 965 * that has a configuration which was written in a future txg.
961 966 * Search all labels on this vdev to find the configuration that
962 967 * matches the txg for our uberblock.
963 968 */
964 969 if (cb.ubl_vd != NULL)
965 970 *config = vdev_label_read_config(cb.ubl_vd, ub->ub_txg);
966 971 spa_config_exit(spa, SCL_ALL, FTAG);
967 972 }
968 973
969 974 /*
970 975 * On success, increment root zio's count of good writes.
971 976 * We only get credit for writes to known-visible vdevs; see spa_vdev_add().
972 977 */
973 978 static void
974 979 vdev_uberblock_sync_done(zio_t *zio)
975 980 {
976 981 uint64_t *good_writes = zio->io_private;
977 982
978 983 if (zio->io_error == 0 && zio->io_vd->vdev_top->vdev_ms_array != 0)
979 984 atomic_add_64(good_writes, 1);
980 985 }
981 986
982 987 /*
983 988 * Write the uberblock to all labels of all leaves of the specified vdev.
984 989 */
985 990 static void
986 991 vdev_uberblock_sync(zio_t *zio, uberblock_t *ub, vdev_t *vd, int flags)
987 992 {
988 993 uberblock_t *ubbuf;
989 994 int n;
990 995
991 996 for (int c = 0; c < vd->vdev_children; c++)
992 997 vdev_uberblock_sync(zio, ub, vd->vdev_child[c], flags);
993 998
994 999 if (!vd->vdev_ops->vdev_op_leaf)
995 1000 return;
996 1001
997 1002 if (!vdev_writeable(vd))
998 1003 return;
999 1004
1000 1005 n = ub->ub_txg & (VDEV_UBERBLOCK_COUNT(vd) - 1);
1001 1006
1002 1007 ubbuf = zio_buf_alloc(VDEV_UBERBLOCK_SIZE(vd));
1003 1008 bzero(ubbuf, VDEV_UBERBLOCK_SIZE(vd));
1004 1009 *ubbuf = *ub;
1005 1010
1006 1011 for (int l = 0; l < VDEV_LABELS; l++)
1007 1012 vdev_label_write(zio, vd, l, ubbuf,
1008 1013 VDEV_UBERBLOCK_OFFSET(vd, n), VDEV_UBERBLOCK_SIZE(vd),
1009 1014 vdev_uberblock_sync_done, zio->io_private,
1010 1015 flags | ZIO_FLAG_DONT_PROPAGATE);
1011 1016
1012 1017 zio_buf_free(ubbuf, VDEV_UBERBLOCK_SIZE(vd));
1013 1018 }
1014 1019
1015 1020 /* Sync the uberblocks to all vdevs in svd[] */
1016 1021 int
1017 1022 vdev_uberblock_sync_list(vdev_t **svd, int svdcount, uberblock_t *ub, int flags)
1018 1023 {
1019 1024 spa_t *spa = svd[0]->vdev_spa;
1020 1025 zio_t *zio;
1021 1026 uint64_t good_writes = 0;
1022 1027
1023 1028 zio = zio_root(spa, NULL, &good_writes, flags);
1024 1029
1025 1030 for (int v = 0; v < svdcount; v++)
1026 1031 vdev_uberblock_sync(zio, ub, svd[v], flags);
1027 1032
1028 1033 (void) zio_wait(zio);
1029 1034
1030 1035 /*
1031 1036 * Flush the uberblocks to disk. This ensures that the odd labels
1032 1037 * are no longer needed (because the new uberblocks and the even
1033 1038 * labels are safely on disk), so it is safe to overwrite them.
1034 1039 */
1035 1040 zio = zio_root(spa, NULL, NULL, flags);
1036 1041
1037 1042 for (int v = 0; v < svdcount; v++)
1038 1043 zio_flush(zio, svd[v]);
1039 1044
1040 1045 (void) zio_wait(zio);
1041 1046
1042 1047 return (good_writes >= 1 ? 0 : EIO);
1043 1048 }
1044 1049
1045 1050 /*
1046 1051 * On success, increment the count of good writes for our top-level vdev.
1047 1052 */
1048 1053 static void
1049 1054 vdev_label_sync_done(zio_t *zio)
1050 1055 {
1051 1056 uint64_t *good_writes = zio->io_private;
1052 1057
1053 1058 if (zio->io_error == 0)
1054 1059 atomic_add_64(good_writes, 1);
1055 1060 }
1056 1061
1057 1062 /*
1058 1063 * If there weren't enough good writes, indicate failure to the parent.
1059 1064 */
1060 1065 static void
1061 1066 vdev_label_sync_top_done(zio_t *zio)
1062 1067 {
1063 1068 uint64_t *good_writes = zio->io_private;
1064 1069
1065 1070 if (*good_writes == 0)
1066 1071 zio->io_error = SET_ERROR(EIO);
1067 1072
1068 1073 kmem_free(good_writes, sizeof (uint64_t));
1069 1074 }
1070 1075
1071 1076 /*
1072 1077 * We ignore errors for log and cache devices, simply free the private data.
1073 1078 */
1074 1079 static void
1075 1080 vdev_label_sync_ignore_done(zio_t *zio)
1076 1081 {
1077 1082 kmem_free(zio->io_private, sizeof (uint64_t));
1078 1083 }
1079 1084
1080 1085 /*
1081 1086 * Write all even or odd labels to all leaves of the specified vdev.
1082 1087 */
1083 1088 static void
1084 1089 vdev_label_sync(zio_t *zio, vdev_t *vd, int l, uint64_t txg, int flags)
1085 1090 {
1086 1091 nvlist_t *label;
1087 1092 vdev_phys_t *vp;
1088 1093 char *buf;
1089 1094 size_t buflen;
1090 1095
1091 1096 for (int c = 0; c < vd->vdev_children; c++)
1092 1097 vdev_label_sync(zio, vd->vdev_child[c], l, txg, flags);
1093 1098
1094 1099 if (!vd->vdev_ops->vdev_op_leaf)
1095 1100 return;
1096 1101
1097 1102 if (!vdev_writeable(vd))
1098 1103 return;
1099 1104
1100 1105 /*
1101 1106 * Generate a label describing the top-level config to which we belong.
1102 1107 */
1103 1108 label = spa_config_generate(vd->vdev_spa, vd, txg, B_FALSE);
1104 1109
1105 1110 vp = zio_buf_alloc(sizeof (vdev_phys_t));
1106 1111 bzero(vp, sizeof (vdev_phys_t));
1107 1112
1108 1113 buf = vp->vp_nvlist;
1109 1114 buflen = sizeof (vp->vp_nvlist);
1110 1115
1111 1116 if (nvlist_pack(label, &buf, &buflen, NV_ENCODE_XDR, KM_SLEEP) == 0) {
1112 1117 for (; l < VDEV_LABELS; l += 2) {
1113 1118 vdev_label_write(zio, vd, l, vp,
1114 1119 offsetof(vdev_label_t, vl_vdev_phys),
1115 1120 sizeof (vdev_phys_t),
1116 1121 vdev_label_sync_done, zio->io_private,
1117 1122 flags | ZIO_FLAG_DONT_PROPAGATE);
1118 1123 }
1119 1124 }
1120 1125
1121 1126 zio_buf_free(vp, sizeof (vdev_phys_t));
1122 1127 nvlist_free(label);
1123 1128 }
1124 1129
1125 1130 int
1126 1131 vdev_label_sync_list(spa_t *spa, int l, uint64_t txg, int flags)
1127 1132 {
1128 1133 list_t *dl = &spa->spa_config_dirty_list;
1129 1134 vdev_t *vd;
1130 1135 zio_t *zio;
1131 1136 int error;
1132 1137
1133 1138 /*
1134 1139 * Write the new labels to disk.
1135 1140 */
1136 1141 zio = zio_root(spa, NULL, NULL, flags);
1137 1142
1138 1143 for (vd = list_head(dl); vd != NULL; vd = list_next(dl, vd)) {
1139 1144 uint64_t *good_writes = kmem_zalloc(sizeof (uint64_t),
1140 1145 KM_SLEEP);
1141 1146
1142 1147 ASSERT(!vd->vdev_ishole);
1143 1148
1144 1149 zio_t *vio = zio_null(zio, spa, NULL,
1145 1150 (vd->vdev_islog || vd->vdev_aux != NULL) ?
1146 1151 vdev_label_sync_ignore_done : vdev_label_sync_top_done,
1147 1152 good_writes, flags);
1148 1153 vdev_label_sync(vio, vd, l, txg, flags);
1149 1154 zio_nowait(vio);
1150 1155 }
1151 1156
1152 1157 error = zio_wait(zio);
1153 1158
1154 1159 /*
1155 1160 * Flush the new labels to disk.
1156 1161 */
1157 1162 zio = zio_root(spa, NULL, NULL, flags);
1158 1163
1159 1164 for (vd = list_head(dl); vd != NULL; vd = list_next(dl, vd))
1160 1165 zio_flush(zio, vd);
1161 1166
1162 1167 (void) zio_wait(zio);
1163 1168
1164 1169 return (error);
1165 1170 }
1166 1171
1167 1172 /*
1168 1173 * Sync the uberblock and any changes to the vdev configuration.
1169 1174 *
1170 1175 * The order of operations is carefully crafted to ensure that
1171 1176 * if the system panics or loses power at any time, the state on disk
1172 1177 * is still transactionally consistent. The in-line comments below
1173 1178 * describe the failure semantics at each stage.
1174 1179 *
1175 1180 * Moreover, vdev_config_sync() is designed to be idempotent: if it fails
1176 1181 * at any time, you can just call it again, and it will resume its work.
1177 1182 */
1178 1183 int
1179 1184 vdev_config_sync(vdev_t **svd, int svdcount, uint64_t txg, boolean_t tryhard)
1180 1185 {
1181 1186 spa_t *spa = svd[0]->vdev_spa;
1182 1187 uberblock_t *ub = &spa->spa_uberblock;
1183 1188 vdev_t *vd;
1184 1189 zio_t *zio;
1185 1190 int error;
1186 1191 int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL;
1187 1192
1188 1193 /*
1189 1194 * Normally, we don't want to try too hard to write every label and
1190 1195 * uberblock. If there is a flaky disk, we don't want the rest of the
1191 1196 * sync process to block while we retry. But if we can't write a
1192 1197 * single label out, we should retry with ZIO_FLAG_TRYHARD before
1193 1198 * bailing out and declaring the pool faulted.
1194 1199 */
1195 1200 if (tryhard)
1196 1201 flags |= ZIO_FLAG_TRYHARD;
1197 1202
1198 1203 ASSERT(ub->ub_txg <= txg);
1199 1204
1200 1205 /*
1201 1206 * If this isn't a resync due to I/O errors,
1202 1207 * and nothing changed in this transaction group,
1203 1208 * and the vdev configuration hasn't changed,
1204 1209 * then there's nothing to do.
1205 1210 */
1206 1211 if (ub->ub_txg < txg &&
1207 1212 uberblock_update(ub, spa->spa_root_vdev, txg) == B_FALSE &&
1208 1213 list_is_empty(&spa->spa_config_dirty_list))
1209 1214 return (0);
1210 1215
1211 1216 if (txg > spa_freeze_txg(spa))
1212 1217 return (0);
1213 1218
1214 1219 ASSERT(txg <= spa->spa_final_txg);
1215 1220
1216 1221 /*
1217 1222 * Flush the write cache of every disk that's been written to
1218 1223 * in this transaction group. This ensures that all blocks
1219 1224 * written in this txg will be committed to stable storage
1220 1225 * before any uberblock that references them.
1221 1226 */
1222 1227 zio = zio_root(spa, NULL, NULL, flags);
1223 1228
1224 1229 for (vd = txg_list_head(&spa->spa_vdev_txg_list, TXG_CLEAN(txg)); vd;
1225 1230 vd = txg_list_next(&spa->spa_vdev_txg_list, vd, TXG_CLEAN(txg)))
1226 1231 zio_flush(zio, vd);
1227 1232
1228 1233 (void) zio_wait(zio);
1229 1234
1230 1235 /*
1231 1236 * Sync out the even labels (L0, L2) for every dirty vdev. If the
1232 1237 * system dies in the middle of this process, that's OK: all of the
1233 1238 * even labels that made it to disk will be newer than any uberblock,
1234 1239 * and will therefore be considered invalid. The odd labels (L1, L3),
1235 1240 * which have not yet been touched, will still be valid. We flush
1236 1241 * the new labels to disk to ensure that all even-label updates
1237 1242 * are committed to stable storage before the uberblock update.
1238 1243 */
1239 1244 if ((error = vdev_label_sync_list(spa, 0, txg, flags)) != 0)
1240 1245 return (error);
1241 1246
1242 1247 /*
1243 1248 * Sync the uberblocks to all vdevs in svd[].
1244 1249 * If the system dies in the middle of this step, there are two cases
1245 1250 * to consider, and the on-disk state is consistent either way:
1246 1251 *
1247 1252 * (1) If none of the new uberblocks made it to disk, then the
1248 1253 * previous uberblock will be the newest, and the odd labels
1249 1254 * (which had not yet been touched) will be valid with respect
1250 1255 * to that uberblock.
1251 1256 *
1252 1257 * (2) If one or more new uberblocks made it to disk, then they
1253 1258 * will be the newest, and the even labels (which had all
1254 1259 * been successfully committed) will be valid with respect
1255 1260 * to the new uberblocks.
1256 1261 */
1257 1262 if ((error = vdev_uberblock_sync_list(svd, svdcount, ub, flags)) != 0)
1258 1263 return (error);
1259 1264
1260 1265 /*
1261 1266 * Sync out odd labels for every dirty vdev. If the system dies
1262 1267 * in the middle of this process, the even labels and the new
1263 1268 * uberblocks will suffice to open the pool. The next time
1264 1269 * the pool is opened, the first thing we'll do -- before any
1265 1270 * user data is modified -- is mark every vdev dirty so that
1266 1271 * all labels will be brought up to date. We flush the new labels
1267 1272 * to disk to ensure that all odd-label updates are committed to
1268 1273 * stable storage before the next transaction group begins.
1269 1274 */
1270 1275 return (vdev_label_sync_list(spa, 1, txg, flags));
1271 1276 }
|
↓ open down ↓ |
976 lines elided |
↑ open up ↑ |
XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX