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 VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0);
220 220
221 221 VERIFY(nvlist_add_string(nv, ZPOOL_CONFIG_TYPE,
222 222 vd->vdev_ops->vdev_op_type) == 0);
223 223 if (!(flags & (VDEV_CONFIG_SPARE | VDEV_CONFIG_L2CACHE)))
224 224 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_ID, vd->vdev_id)
225 225 == 0);
226 226 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_GUID, vd->vdev_guid) == 0);
227 227
228 228 if (vd->vdev_path != NULL)
229 229 VERIFY(nvlist_add_string(nv, ZPOOL_CONFIG_PATH,
230 230 vd->vdev_path) == 0);
231 231
232 232 if (vd->vdev_devid != NULL)
233 233 VERIFY(nvlist_add_string(nv, ZPOOL_CONFIG_DEVID,
234 234 vd->vdev_devid) == 0);
235 235
236 236 if (vd->vdev_physpath != NULL)
237 237 VERIFY(nvlist_add_string(nv, ZPOOL_CONFIG_PHYS_PATH,
238 238 vd->vdev_physpath) == 0);
239 239
240 240 if (vd->vdev_fru != NULL)
241 241 VERIFY(nvlist_add_string(nv, ZPOOL_CONFIG_FRU,
242 242 vd->vdev_fru) == 0);
243 243
244 244 if (vd->vdev_nparity != 0) {
245 245 ASSERT(strcmp(vd->vdev_ops->vdev_op_type,
246 246 VDEV_TYPE_RAIDZ) == 0);
247 247
248 248 /*
249 249 * Make sure someone hasn't managed to sneak a fancy new vdev
250 250 * into a crufty old storage pool.
251 251 */
252 252 ASSERT(vd->vdev_nparity == 1 ||
253 253 (vd->vdev_nparity <= 2 &&
254 254 spa_version(spa) >= SPA_VERSION_RAIDZ2) ||
255 255 (vd->vdev_nparity <= 3 &&
256 256 spa_version(spa) >= SPA_VERSION_RAIDZ3));
257 257
258 258 /*
259 259 * Note that we'll add the nparity tag even on storage pools
260 260 * that only support a single parity device -- older software
261 261 * will just ignore it.
262 262 */
263 263 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_NPARITY,
264 264 vd->vdev_nparity) == 0);
265 265 }
266 266
267 267 if (vd->vdev_wholedisk != -1ULL)
268 268 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK,
269 269 vd->vdev_wholedisk) == 0);
270 270
271 271 if (vd->vdev_not_present)
272 272 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_NOT_PRESENT, 1) == 0);
273 273
274 274 if (vd->vdev_isspare)
275 275 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_IS_SPARE, 1) == 0);
276 276
277 277 if (!(flags & (VDEV_CONFIG_SPARE | VDEV_CONFIG_L2CACHE)) &&
278 278 vd == vd->vdev_top) {
279 279 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_METASLAB_ARRAY,
280 280 vd->vdev_ms_array) == 0);
281 281 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_METASLAB_SHIFT,
282 282 vd->vdev_ms_shift) == 0);
283 283 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_ASHIFT,
|
↓ open down ↓ |
283 lines elided |
↑ open up ↑ |
284 284 vd->vdev_ashift) == 0);
285 285 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_ASIZE,
286 286 vd->vdev_asize) == 0);
287 287 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_IS_LOG,
288 288 vd->vdev_islog) == 0);
289 289 if (vd->vdev_removing)
290 290 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_REMOVING,
291 291 vd->vdev_removing) == 0);
292 292 }
293 293
294 + if (flags & VDEV_CONFIG_L2CACHE)
295 + /* indicate that we support L2ARC persistency */
296 + VERIFY(nvlist_add_boolean_value(nv,
297 + ZPOOL_CONFIG_L2CACHE_PERSISTENT, B_TRUE) == 0);
298 +
294 299 if (vd->vdev_dtl_smo.smo_object != 0)
295 300 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_DTL,
296 301 vd->vdev_dtl_smo.smo_object) == 0);
297 302
298 303 if (vd->vdev_crtxg)
299 304 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_CREATE_TXG,
300 305 vd->vdev_crtxg) == 0);
301 306
302 307 if (getstats) {
303 308 vdev_stat_t vs;
304 309 pool_scan_stat_t ps;
305 310
306 311 vdev_get_stats(vd, &vs);
307 312 VERIFY(nvlist_add_uint64_array(nv, ZPOOL_CONFIG_VDEV_STATS,
308 313 (uint64_t *)&vs, sizeof (vs) / sizeof (uint64_t)) == 0);
309 314
310 315 /* provide either current or previous scan information */
311 316 if (spa_scan_get_stats(spa, &ps) == 0) {
312 317 VERIFY(nvlist_add_uint64_array(nv,
313 318 ZPOOL_CONFIG_SCAN_STATS, (uint64_t *)&ps,
314 319 sizeof (pool_scan_stat_t) / sizeof (uint64_t))
315 320 == 0);
316 321 }
317 322 }
318 323
319 324 if (!vd->vdev_ops->vdev_op_leaf) {
320 325 nvlist_t **child;
321 326 int c, idx;
322 327
323 328 ASSERT(!vd->vdev_ishole);
324 329
325 330 child = kmem_alloc(vd->vdev_children * sizeof (nvlist_t *),
326 331 KM_SLEEP);
327 332
328 333 for (c = 0, idx = 0; c < vd->vdev_children; c++) {
329 334 vdev_t *cvd = vd->vdev_child[c];
330 335
331 336 /*
332 337 * If we're generating an nvlist of removing
333 338 * vdevs then skip over any device which is
334 339 * not being removed.
335 340 */
336 341 if ((flags & VDEV_CONFIG_REMOVING) &&
337 342 !cvd->vdev_removing)
338 343 continue;
339 344
340 345 child[idx++] = vdev_config_generate(spa, cvd,
341 346 getstats, flags);
342 347 }
343 348
344 349 if (idx) {
345 350 VERIFY(nvlist_add_nvlist_array(nv,
346 351 ZPOOL_CONFIG_CHILDREN, child, idx) == 0);
347 352 }
348 353
349 354 for (c = 0; c < idx; c++)
350 355 nvlist_free(child[c]);
351 356
352 357 kmem_free(child, vd->vdev_children * sizeof (nvlist_t *));
353 358
354 359 } else {
355 360 const char *aux = NULL;
356 361
357 362 if (vd->vdev_offline && !vd->vdev_tmpoffline)
358 363 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_OFFLINE,
359 364 B_TRUE) == 0);
360 365 if (vd->vdev_resilvering)
361 366 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_RESILVERING,
362 367 B_TRUE) == 0);
363 368 if (vd->vdev_faulted)
364 369 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_FAULTED,
365 370 B_TRUE) == 0);
366 371 if (vd->vdev_degraded)
367 372 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_DEGRADED,
368 373 B_TRUE) == 0);
369 374 if (vd->vdev_removed)
370 375 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_REMOVED,
371 376 B_TRUE) == 0);
372 377 if (vd->vdev_unspare)
373 378 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_UNSPARE,
374 379 B_TRUE) == 0);
375 380 if (vd->vdev_ishole)
376 381 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_IS_HOLE,
377 382 B_TRUE) == 0);
378 383
379 384 switch (vd->vdev_stat.vs_aux) {
380 385 case VDEV_AUX_ERR_EXCEEDED:
381 386 aux = "err_exceeded";
382 387 break;
383 388
384 389 case VDEV_AUX_EXTERNAL:
385 390 aux = "external";
386 391 break;
387 392 }
388 393
389 394 if (aux != NULL)
390 395 VERIFY(nvlist_add_string(nv, ZPOOL_CONFIG_AUX_STATE,
391 396 aux) == 0);
392 397
393 398 if (vd->vdev_splitting && vd->vdev_orig_guid != 0LL) {
394 399 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_ORIG_GUID,
395 400 vd->vdev_orig_guid) == 0);
396 401 }
397 402 }
398 403
399 404 return (nv);
400 405 }
401 406
402 407 /*
403 408 * Generate a view of the top-level vdevs. If we currently have holes
404 409 * in the namespace, then generate an array which contains a list of holey
405 410 * vdevs. Additionally, add the number of top-level children that currently
406 411 * exist.
407 412 */
408 413 void
409 414 vdev_top_config_generate(spa_t *spa, nvlist_t *config)
410 415 {
411 416 vdev_t *rvd = spa->spa_root_vdev;
412 417 uint64_t *array;
413 418 uint_t c, idx;
414 419
415 420 array = kmem_alloc(rvd->vdev_children * sizeof (uint64_t), KM_SLEEP);
416 421
417 422 for (c = 0, idx = 0; c < rvd->vdev_children; c++) {
418 423 vdev_t *tvd = rvd->vdev_child[c];
419 424
420 425 if (tvd->vdev_ishole)
421 426 array[idx++] = c;
422 427 }
423 428
424 429 if (idx) {
425 430 VERIFY(nvlist_add_uint64_array(config, ZPOOL_CONFIG_HOLE_ARRAY,
426 431 array, idx) == 0);
427 432 }
428 433
429 434 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VDEV_CHILDREN,
430 435 rvd->vdev_children) == 0);
431 436
432 437 kmem_free(array, rvd->vdev_children * sizeof (uint64_t));
433 438 }
434 439
435 440 /*
436 441 * Returns the configuration from the label of the given vdev. For vdevs
437 442 * which don't have a txg value stored on their label (i.e. spares/cache)
438 443 * or have not been completely initialized (txg = 0) just return
439 444 * the configuration from the first valid label we find. Otherwise,
440 445 * find the most up-to-date label that does not exceed the specified
441 446 * 'txg' value.
442 447 */
443 448 nvlist_t *
444 449 vdev_label_read_config(vdev_t *vd, uint64_t txg)
445 450 {
446 451 spa_t *spa = vd->vdev_spa;
447 452 nvlist_t *config = NULL;
448 453 vdev_phys_t *vp;
449 454 zio_t *zio;
450 455 uint64_t best_txg = 0;
451 456 int error = 0;
452 457 int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL |
453 458 ZIO_FLAG_SPECULATIVE;
454 459
455 460 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
456 461
457 462 if (!vdev_readable(vd))
458 463 return (NULL);
459 464
460 465 vp = zio_buf_alloc(sizeof (vdev_phys_t));
461 466
462 467 retry:
463 468 for (int l = 0; l < VDEV_LABELS; l++) {
464 469 nvlist_t *label = NULL;
465 470
466 471 zio = zio_root(spa, NULL, NULL, flags);
467 472
468 473 vdev_label_read(zio, vd, l, vp,
469 474 offsetof(vdev_label_t, vl_vdev_phys),
470 475 sizeof (vdev_phys_t), NULL, NULL, flags);
471 476
472 477 if (zio_wait(zio) == 0 &&
473 478 nvlist_unpack(vp->vp_nvlist, sizeof (vp->vp_nvlist),
474 479 &label, 0) == 0) {
475 480 uint64_t label_txg = 0;
476 481
477 482 /*
478 483 * Auxiliary vdevs won't have txg values in their
479 484 * labels and newly added vdevs may not have been
480 485 * completely initialized so just return the
481 486 * configuration from the first valid label we
482 487 * encounter.
483 488 */
484 489 error = nvlist_lookup_uint64(label,
485 490 ZPOOL_CONFIG_POOL_TXG, &label_txg);
486 491 if ((error || label_txg == 0) && !config) {
487 492 config = label;
488 493 break;
489 494 } else if (label_txg <= txg && label_txg > best_txg) {
490 495 best_txg = label_txg;
491 496 nvlist_free(config);
492 497 config = fnvlist_dup(label);
493 498 }
494 499 }
495 500
496 501 if (label != NULL) {
497 502 nvlist_free(label);
498 503 label = NULL;
499 504 }
500 505 }
501 506
502 507 if (config == NULL && !(flags & ZIO_FLAG_TRYHARD)) {
503 508 flags |= ZIO_FLAG_TRYHARD;
504 509 goto retry;
505 510 }
506 511
507 512 zio_buf_free(vp, sizeof (vdev_phys_t));
508 513
509 514 return (config);
510 515 }
511 516
512 517 /*
513 518 * Determine if a device is in use. The 'spare_guid' parameter will be filled
514 519 * in with the device guid if this spare is active elsewhere on the system.
515 520 */
516 521 static boolean_t
517 522 vdev_inuse(vdev_t *vd, uint64_t crtxg, vdev_labeltype_t reason,
518 523 uint64_t *spare_guid, uint64_t *l2cache_guid)
519 524 {
520 525 spa_t *spa = vd->vdev_spa;
521 526 uint64_t state, pool_guid, device_guid, txg, spare_pool;
522 527 uint64_t vdtxg = 0;
523 528 nvlist_t *label;
524 529
525 530 if (spare_guid)
526 531 *spare_guid = 0ULL;
527 532 if (l2cache_guid)
528 533 *l2cache_guid = 0ULL;
529 534
530 535 /*
531 536 * Read the label, if any, and perform some basic sanity checks.
532 537 */
533 538 if ((label = vdev_label_read_config(vd, -1ULL)) == NULL)
534 539 return (B_FALSE);
535 540
536 541 (void) nvlist_lookup_uint64(label, ZPOOL_CONFIG_CREATE_TXG,
537 542 &vdtxg);
538 543
539 544 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE,
540 545 &state) != 0 ||
541 546 nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID,
542 547 &device_guid) != 0) {
543 548 nvlist_free(label);
544 549 return (B_FALSE);
545 550 }
546 551
547 552 if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
548 553 (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_GUID,
549 554 &pool_guid) != 0 ||
550 555 nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG,
551 556 &txg) != 0)) {
552 557 nvlist_free(label);
553 558 return (B_FALSE);
554 559 }
555 560
556 561 nvlist_free(label);
557 562
558 563 /*
559 564 * Check to see if this device indeed belongs to the pool it claims to
560 565 * be a part of. The only way this is allowed is if the device is a hot
561 566 * spare (which we check for later on).
562 567 */
563 568 if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
564 569 !spa_guid_exists(pool_guid, device_guid) &&
565 570 !spa_spare_exists(device_guid, NULL, NULL) &&
566 571 !spa_l2cache_exists(device_guid, NULL))
567 572 return (B_FALSE);
568 573
569 574 /*
570 575 * If the transaction group is zero, then this an initialized (but
571 576 * unused) label. This is only an error if the create transaction
572 577 * on-disk is the same as the one we're using now, in which case the
573 578 * user has attempted to add the same vdev multiple times in the same
574 579 * transaction.
575 580 */
576 581 if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
577 582 txg == 0 && vdtxg == crtxg)
578 583 return (B_TRUE);
579 584
580 585 /*
581 586 * Check to see if this is a spare device. We do an explicit check for
582 587 * spa_has_spare() here because it may be on our pending list of spares
583 588 * to add. We also check if it is an l2cache device.
584 589 */
585 590 if (spa_spare_exists(device_guid, &spare_pool, NULL) ||
586 591 spa_has_spare(spa, device_guid)) {
587 592 if (spare_guid)
588 593 *spare_guid = device_guid;
589 594
590 595 switch (reason) {
591 596 case VDEV_LABEL_CREATE:
592 597 case VDEV_LABEL_L2CACHE:
593 598 return (B_TRUE);
594 599
595 600 case VDEV_LABEL_REPLACE:
596 601 return (!spa_has_spare(spa, device_guid) ||
597 602 spare_pool != 0ULL);
598 603
599 604 case VDEV_LABEL_SPARE:
600 605 return (spa_has_spare(spa, device_guid));
601 606 }
602 607 }
603 608
604 609 /*
605 610 * Check to see if this is an l2cache device.
606 611 */
607 612 if (spa_l2cache_exists(device_guid, NULL))
608 613 return (B_TRUE);
609 614
610 615 /*
611 616 * We can't rely on a pool's state if it's been imported
612 617 * read-only. Instead we look to see if the pools is marked
613 618 * read-only in the namespace and set the state to active.
614 619 */
615 620 if ((spa = spa_by_guid(pool_guid, device_guid)) != NULL &&
616 621 spa_mode(spa) == FREAD)
617 622 state = POOL_STATE_ACTIVE;
618 623
619 624 /*
620 625 * If the device is marked ACTIVE, then this device is in use by another
621 626 * pool on the system.
622 627 */
623 628 return (state == POOL_STATE_ACTIVE);
624 629 }
625 630
626 631 /*
627 632 * Initialize a vdev label. We check to make sure each leaf device is not in
628 633 * use, and writable. We put down an initial label which we will later
629 634 * overwrite with a complete label. Note that it's important to do this
630 635 * sequentially, not in parallel, so that we catch cases of multiple use of the
631 636 * same leaf vdev in the vdev we're creating -- e.g. mirroring a disk with
632 637 * itself.
633 638 */
634 639 int
635 640 vdev_label_init(vdev_t *vd, uint64_t crtxg, vdev_labeltype_t reason)
636 641 {
637 642 spa_t *spa = vd->vdev_spa;
638 643 nvlist_t *label;
639 644 vdev_phys_t *vp;
640 645 char *pad2;
641 646 uberblock_t *ub;
642 647 zio_t *zio;
643 648 char *buf;
644 649 size_t buflen;
645 650 int error;
646 651 uint64_t spare_guid, l2cache_guid;
647 652 int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL;
648 653
649 654 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
650 655
651 656 for (int c = 0; c < vd->vdev_children; c++)
652 657 if ((error = vdev_label_init(vd->vdev_child[c],
653 658 crtxg, reason)) != 0)
654 659 return (error);
655 660
656 661 /* Track the creation time for this vdev */
657 662 vd->vdev_crtxg = crtxg;
658 663
659 664 if (!vd->vdev_ops->vdev_op_leaf)
660 665 return (0);
661 666
662 667 /*
663 668 * Dead vdevs cannot be initialized.
664 669 */
665 670 if (vdev_is_dead(vd))
666 671 return (SET_ERROR(EIO));
667 672
668 673 /*
669 674 * Determine if the vdev is in use.
670 675 */
671 676 if (reason != VDEV_LABEL_REMOVE && reason != VDEV_LABEL_SPLIT &&
672 677 vdev_inuse(vd, crtxg, reason, &spare_guid, &l2cache_guid))
673 678 return (SET_ERROR(EBUSY));
674 679
675 680 /*
676 681 * If this is a request to add or replace a spare or l2cache device
677 682 * that is in use elsewhere on the system, then we must update the
678 683 * guid (which was initialized to a random value) to reflect the
679 684 * actual GUID (which is shared between multiple pools).
680 685 */
681 686 if (reason != VDEV_LABEL_REMOVE && reason != VDEV_LABEL_L2CACHE &&
682 687 spare_guid != 0ULL) {
683 688 uint64_t guid_delta = spare_guid - vd->vdev_guid;
684 689
685 690 vd->vdev_guid += guid_delta;
686 691
687 692 for (vdev_t *pvd = vd; pvd != NULL; pvd = pvd->vdev_parent)
688 693 pvd->vdev_guid_sum += guid_delta;
689 694
690 695 /*
691 696 * If this is a replacement, then we want to fallthrough to the
692 697 * rest of the code. If we're adding a spare, then it's already
693 698 * labeled appropriately and we can just return.
694 699 */
695 700 if (reason == VDEV_LABEL_SPARE)
696 701 return (0);
697 702 ASSERT(reason == VDEV_LABEL_REPLACE ||
698 703 reason == VDEV_LABEL_SPLIT);
699 704 }
700 705
701 706 if (reason != VDEV_LABEL_REMOVE && reason != VDEV_LABEL_SPARE &&
702 707 l2cache_guid != 0ULL) {
703 708 uint64_t guid_delta = l2cache_guid - vd->vdev_guid;
704 709
705 710 vd->vdev_guid += guid_delta;
706 711
707 712 for (vdev_t *pvd = vd; pvd != NULL; pvd = pvd->vdev_parent)
708 713 pvd->vdev_guid_sum += guid_delta;
709 714
710 715 /*
711 716 * If this is a replacement, then we want to fallthrough to the
712 717 * rest of the code. If we're adding an l2cache, then it's
713 718 * already labeled appropriately and we can just return.
714 719 */
715 720 if (reason == VDEV_LABEL_L2CACHE)
716 721 return (0);
717 722 ASSERT(reason == VDEV_LABEL_REPLACE);
718 723 }
719 724
720 725 /*
721 726 * Initialize its label.
722 727 */
723 728 vp = zio_buf_alloc(sizeof (vdev_phys_t));
724 729 bzero(vp, sizeof (vdev_phys_t));
725 730
726 731 /*
727 732 * Generate a label describing the pool and our top-level vdev.
728 733 * We mark it as being from txg 0 to indicate that it's not
729 734 * really part of an active pool just yet. The labels will
730 735 * be written again with a meaningful txg by spa_sync().
731 736 */
732 737 if (reason == VDEV_LABEL_SPARE ||
733 738 (reason == VDEV_LABEL_REMOVE && vd->vdev_isspare)) {
734 739 /*
735 740 * For inactive hot spares, we generate a special label that
736 741 * identifies as a mutually shared hot spare. We write the
737 742 * label if we are adding a hot spare, or if we are removing an
738 743 * active hot spare (in which case we want to revert the
739 744 * labels).
740 745 */
741 746 VERIFY(nvlist_alloc(&label, NV_UNIQUE_NAME, KM_SLEEP) == 0);
742 747
743 748 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_VERSION,
744 749 spa_version(spa)) == 0);
745 750 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_POOL_STATE,
746 751 POOL_STATE_SPARE) == 0);
747 752 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_GUID,
748 753 vd->vdev_guid) == 0);
749 754 } else if (reason == VDEV_LABEL_L2CACHE ||
750 755 (reason == VDEV_LABEL_REMOVE && vd->vdev_isl2cache)) {
751 756 /*
752 757 * For level 2 ARC devices, add a special label.
753 758 */
754 759 VERIFY(nvlist_alloc(&label, NV_UNIQUE_NAME, KM_SLEEP) == 0);
755 760
756 761 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_VERSION,
757 762 spa_version(spa)) == 0);
758 763 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_POOL_STATE,
759 764 POOL_STATE_L2CACHE) == 0);
760 765 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_GUID,
761 766 vd->vdev_guid) == 0);
762 767 } else {
763 768 uint64_t txg = 0ULL;
764 769
765 770 if (reason == VDEV_LABEL_SPLIT)
766 771 txg = spa->spa_uberblock.ub_txg;
767 772 label = spa_config_generate(spa, vd, txg, B_FALSE);
768 773
769 774 /*
770 775 * Add our creation time. This allows us to detect multiple
771 776 * vdev uses as described above, and automatically expires if we
772 777 * fail.
773 778 */
774 779 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_CREATE_TXG,
775 780 crtxg) == 0);
776 781 }
777 782
778 783 buf = vp->vp_nvlist;
779 784 buflen = sizeof (vp->vp_nvlist);
780 785
781 786 error = nvlist_pack(label, &buf, &buflen, NV_ENCODE_XDR, KM_SLEEP);
782 787 if (error != 0) {
783 788 nvlist_free(label);
784 789 zio_buf_free(vp, sizeof (vdev_phys_t));
785 790 /* EFAULT means nvlist_pack ran out of room */
786 791 return (error == EFAULT ? ENAMETOOLONG : EINVAL);
787 792 }
788 793
789 794 /*
790 795 * Initialize uberblock template.
791 796 */
792 797 ub = zio_buf_alloc(VDEV_UBERBLOCK_RING);
793 798 bzero(ub, VDEV_UBERBLOCK_RING);
794 799 *ub = spa->spa_uberblock;
795 800 ub->ub_txg = 0;
796 801
797 802 /* Initialize the 2nd padding area. */
798 803 pad2 = zio_buf_alloc(VDEV_PAD_SIZE);
799 804 bzero(pad2, VDEV_PAD_SIZE);
800 805
801 806 /*
802 807 * Write everything in parallel.
803 808 */
804 809 retry:
805 810 zio = zio_root(spa, NULL, NULL, flags);
806 811
807 812 for (int l = 0; l < VDEV_LABELS; l++) {
808 813
809 814 vdev_label_write(zio, vd, l, vp,
810 815 offsetof(vdev_label_t, vl_vdev_phys),
811 816 sizeof (vdev_phys_t), NULL, NULL, flags);
812 817
813 818 /*
814 819 * Skip the 1st padding area.
815 820 * Zero out the 2nd padding area where it might have
816 821 * left over data from previous filesystem format.
817 822 */
818 823 vdev_label_write(zio, vd, l, pad2,
819 824 offsetof(vdev_label_t, vl_pad2),
820 825 VDEV_PAD_SIZE, NULL, NULL, flags);
821 826
822 827 vdev_label_write(zio, vd, l, ub,
823 828 offsetof(vdev_label_t, vl_uberblock),
824 829 VDEV_UBERBLOCK_RING, NULL, NULL, flags);
825 830 }
826 831
827 832 error = zio_wait(zio);
828 833
829 834 if (error != 0 && !(flags & ZIO_FLAG_TRYHARD)) {
830 835 flags |= ZIO_FLAG_TRYHARD;
831 836 goto retry;
832 837 }
833 838
834 839 nvlist_free(label);
835 840 zio_buf_free(pad2, VDEV_PAD_SIZE);
836 841 zio_buf_free(ub, VDEV_UBERBLOCK_RING);
837 842 zio_buf_free(vp, sizeof (vdev_phys_t));
838 843
839 844 /*
840 845 * If this vdev hasn't been previously identified as a spare, then we
841 846 * mark it as such only if a) we are labeling it as a spare, or b) it
842 847 * exists as a spare elsewhere in the system. Do the same for
843 848 * level 2 ARC devices.
844 849 */
845 850 if (error == 0 && !vd->vdev_isspare &&
846 851 (reason == VDEV_LABEL_SPARE ||
847 852 spa_spare_exists(vd->vdev_guid, NULL, NULL)))
848 853 spa_spare_add(vd);
849 854
850 855 if (error == 0 && !vd->vdev_isl2cache &&
851 856 (reason == VDEV_LABEL_L2CACHE ||
852 857 spa_l2cache_exists(vd->vdev_guid, NULL)))
853 858 spa_l2cache_add(vd);
854 859
855 860 return (error);
856 861 }
857 862
858 863 /*
859 864 * ==========================================================================
860 865 * uberblock load/sync
861 866 * ==========================================================================
862 867 */
863 868
864 869 /*
865 870 * Consider the following situation: txg is safely synced to disk. We've
866 871 * written the first uberblock for txg + 1, and then we lose power. When we
867 872 * come back up, we fail to see the uberblock for txg + 1 because, say,
868 873 * it was on a mirrored device and the replica to which we wrote txg + 1
869 874 * is now offline. If we then make some changes and sync txg + 1, and then
870 875 * the missing replica comes back, then for a few seconds we'll have two
871 876 * conflicting uberblocks on disk with the same txg. The solution is simple:
872 877 * among uberblocks with equal txg, choose the one with the latest timestamp.
873 878 */
874 879 static int
875 880 vdev_uberblock_compare(uberblock_t *ub1, uberblock_t *ub2)
876 881 {
877 882 if (ub1->ub_txg < ub2->ub_txg)
878 883 return (-1);
879 884 if (ub1->ub_txg > ub2->ub_txg)
880 885 return (1);
881 886
882 887 if (ub1->ub_timestamp < ub2->ub_timestamp)
883 888 return (-1);
884 889 if (ub1->ub_timestamp > ub2->ub_timestamp)
885 890 return (1);
886 891
887 892 return (0);
888 893 }
889 894
890 895 struct ubl_cbdata {
891 896 uberblock_t *ubl_ubbest; /* Best uberblock */
892 897 vdev_t *ubl_vd; /* vdev associated with the above */
893 898 };
894 899
895 900 static void
896 901 vdev_uberblock_load_done(zio_t *zio)
897 902 {
898 903 vdev_t *vd = zio->io_vd;
899 904 spa_t *spa = zio->io_spa;
900 905 zio_t *rio = zio->io_private;
901 906 uberblock_t *ub = zio->io_data;
902 907 struct ubl_cbdata *cbp = rio->io_private;
903 908
904 909 ASSERT3U(zio->io_size, ==, VDEV_UBERBLOCK_SIZE(vd));
905 910
906 911 if (zio->io_error == 0 && uberblock_verify(ub) == 0) {
907 912 mutex_enter(&rio->io_lock);
908 913 if (ub->ub_txg <= spa->spa_load_max_txg &&
909 914 vdev_uberblock_compare(ub, cbp->ubl_ubbest) > 0) {
910 915 /*
911 916 * Keep track of the vdev in which this uberblock
912 917 * was found. We will use this information later
913 918 * to obtain the config nvlist associated with
914 919 * this uberblock.
915 920 */
916 921 *cbp->ubl_ubbest = *ub;
917 922 cbp->ubl_vd = vd;
918 923 }
919 924 mutex_exit(&rio->io_lock);
920 925 }
921 926
922 927 zio_buf_free(zio->io_data, zio->io_size);
923 928 }
924 929
925 930 static void
926 931 vdev_uberblock_load_impl(zio_t *zio, vdev_t *vd, int flags,
927 932 struct ubl_cbdata *cbp)
928 933 {
929 934 for (int c = 0; c < vd->vdev_children; c++)
930 935 vdev_uberblock_load_impl(zio, vd->vdev_child[c], flags, cbp);
931 936
932 937 if (vd->vdev_ops->vdev_op_leaf && vdev_readable(vd)) {
933 938 for (int l = 0; l < VDEV_LABELS; l++) {
934 939 for (int n = 0; n < VDEV_UBERBLOCK_COUNT(vd); n++) {
935 940 vdev_label_read(zio, vd, l,
936 941 zio_buf_alloc(VDEV_UBERBLOCK_SIZE(vd)),
937 942 VDEV_UBERBLOCK_OFFSET(vd, n),
938 943 VDEV_UBERBLOCK_SIZE(vd),
939 944 vdev_uberblock_load_done, zio, flags);
940 945 }
941 946 }
942 947 }
943 948 }
944 949
945 950 /*
946 951 * Reads the 'best' uberblock from disk along with its associated
947 952 * configuration. First, we read the uberblock array of each label of each
948 953 * vdev, keeping track of the uberblock with the highest txg in each array.
949 954 * Then, we read the configuration from the same vdev as the best uberblock.
950 955 */
951 956 void
952 957 vdev_uberblock_load(vdev_t *rvd, uberblock_t *ub, nvlist_t **config)
953 958 {
954 959 zio_t *zio;
955 960 spa_t *spa = rvd->vdev_spa;
956 961 struct ubl_cbdata cb;
957 962 int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL |
958 963 ZIO_FLAG_SPECULATIVE | ZIO_FLAG_TRYHARD;
959 964
960 965 ASSERT(ub);
961 966 ASSERT(config);
962 967
963 968 bzero(ub, sizeof (uberblock_t));
964 969 *config = NULL;
965 970
966 971 cb.ubl_ubbest = ub;
967 972 cb.ubl_vd = NULL;
968 973
969 974 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
970 975 zio = zio_root(spa, NULL, &cb, flags);
971 976 vdev_uberblock_load_impl(zio, rvd, flags, &cb);
972 977 (void) zio_wait(zio);
973 978
974 979 /*
975 980 * It's possible that the best uberblock was discovered on a label
976 981 * that has a configuration which was written in a future txg.
977 982 * Search all labels on this vdev to find the configuration that
978 983 * matches the txg for our uberblock.
979 984 */
980 985 if (cb.ubl_vd != NULL)
981 986 *config = vdev_label_read_config(cb.ubl_vd, ub->ub_txg);
982 987 spa_config_exit(spa, SCL_ALL, FTAG);
983 988 }
984 989
985 990 /*
986 991 * On success, increment root zio's count of good writes.
987 992 * We only get credit for writes to known-visible vdevs; see spa_vdev_add().
988 993 */
989 994 static void
990 995 vdev_uberblock_sync_done(zio_t *zio)
991 996 {
992 997 uint64_t *good_writes = zio->io_private;
993 998
994 999 if (zio->io_error == 0 && zio->io_vd->vdev_top->vdev_ms_array != 0)
995 1000 atomic_add_64(good_writes, 1);
996 1001 }
997 1002
998 1003 /*
999 1004 * Write the uberblock to all labels of all leaves of the specified vdev.
1000 1005 */
1001 1006 static void
1002 1007 vdev_uberblock_sync(zio_t *zio, uberblock_t *ub, vdev_t *vd, int flags)
1003 1008 {
1004 1009 uberblock_t *ubbuf;
1005 1010 int n;
1006 1011
1007 1012 for (int c = 0; c < vd->vdev_children; c++)
1008 1013 vdev_uberblock_sync(zio, ub, vd->vdev_child[c], flags);
1009 1014
1010 1015 if (!vd->vdev_ops->vdev_op_leaf)
1011 1016 return;
1012 1017
1013 1018 if (!vdev_writeable(vd))
1014 1019 return;
1015 1020
1016 1021 n = ub->ub_txg & (VDEV_UBERBLOCK_COUNT(vd) - 1);
1017 1022
1018 1023 ubbuf = zio_buf_alloc(VDEV_UBERBLOCK_SIZE(vd));
1019 1024 bzero(ubbuf, VDEV_UBERBLOCK_SIZE(vd));
1020 1025 *ubbuf = *ub;
1021 1026
1022 1027 for (int l = 0; l < VDEV_LABELS; l++)
1023 1028 vdev_label_write(zio, vd, l, ubbuf,
1024 1029 VDEV_UBERBLOCK_OFFSET(vd, n), VDEV_UBERBLOCK_SIZE(vd),
1025 1030 vdev_uberblock_sync_done, zio->io_private,
1026 1031 flags | ZIO_FLAG_DONT_PROPAGATE);
1027 1032
1028 1033 zio_buf_free(ubbuf, VDEV_UBERBLOCK_SIZE(vd));
1029 1034 }
1030 1035
1031 1036 /* Sync the uberblocks to all vdevs in svd[] */
1032 1037 int
1033 1038 vdev_uberblock_sync_list(vdev_t **svd, int svdcount, uberblock_t *ub, int flags)
1034 1039 {
1035 1040 spa_t *spa = svd[0]->vdev_spa;
1036 1041 zio_t *zio;
1037 1042 uint64_t good_writes = 0;
1038 1043
1039 1044 zio = zio_root(spa, NULL, &good_writes, flags);
1040 1045
1041 1046 for (int v = 0; v < svdcount; v++)
1042 1047 vdev_uberblock_sync(zio, ub, svd[v], flags);
1043 1048
1044 1049 (void) zio_wait(zio);
1045 1050
1046 1051 /*
1047 1052 * Flush the uberblocks to disk. This ensures that the odd labels
1048 1053 * are no longer needed (because the new uberblocks and the even
1049 1054 * labels are safely on disk), so it is safe to overwrite them.
1050 1055 */
1051 1056 zio = zio_root(spa, NULL, NULL, flags);
1052 1057
1053 1058 for (int v = 0; v < svdcount; v++)
1054 1059 zio_flush(zio, svd[v]);
1055 1060
1056 1061 (void) zio_wait(zio);
1057 1062
1058 1063 return (good_writes >= 1 ? 0 : EIO);
1059 1064 }
1060 1065
1061 1066 /*
1062 1067 * On success, increment the count of good writes for our top-level vdev.
1063 1068 */
1064 1069 static void
1065 1070 vdev_label_sync_done(zio_t *zio)
1066 1071 {
1067 1072 uint64_t *good_writes = zio->io_private;
1068 1073
1069 1074 if (zio->io_error == 0)
1070 1075 atomic_add_64(good_writes, 1);
1071 1076 }
1072 1077
1073 1078 /*
1074 1079 * If there weren't enough good writes, indicate failure to the parent.
1075 1080 */
1076 1081 static void
1077 1082 vdev_label_sync_top_done(zio_t *zio)
1078 1083 {
1079 1084 uint64_t *good_writes = zio->io_private;
1080 1085
1081 1086 if (*good_writes == 0)
1082 1087 zio->io_error = SET_ERROR(EIO);
1083 1088
1084 1089 kmem_free(good_writes, sizeof (uint64_t));
1085 1090 }
1086 1091
1087 1092 /*
1088 1093 * We ignore errors for log and cache devices, simply free the private data.
1089 1094 */
1090 1095 static void
1091 1096 vdev_label_sync_ignore_done(zio_t *zio)
1092 1097 {
1093 1098 kmem_free(zio->io_private, sizeof (uint64_t));
1094 1099 }
1095 1100
1096 1101 /*
1097 1102 * Write all even or odd labels to all leaves of the specified vdev.
1098 1103 */
1099 1104 static void
1100 1105 vdev_label_sync(zio_t *zio, vdev_t *vd, int l, uint64_t txg, int flags)
1101 1106 {
1102 1107 nvlist_t *label;
1103 1108 vdev_phys_t *vp;
1104 1109 char *buf;
1105 1110 size_t buflen;
1106 1111
1107 1112 for (int c = 0; c < vd->vdev_children; c++)
1108 1113 vdev_label_sync(zio, vd->vdev_child[c], l, txg, flags);
1109 1114
1110 1115 if (!vd->vdev_ops->vdev_op_leaf)
1111 1116 return;
1112 1117
1113 1118 if (!vdev_writeable(vd))
1114 1119 return;
1115 1120
1116 1121 /*
1117 1122 * Generate a label describing the top-level config to which we belong.
1118 1123 */
1119 1124 label = spa_config_generate(vd->vdev_spa, vd, txg, B_FALSE);
1120 1125
1121 1126 vp = zio_buf_alloc(sizeof (vdev_phys_t));
1122 1127 bzero(vp, sizeof (vdev_phys_t));
1123 1128
1124 1129 buf = vp->vp_nvlist;
1125 1130 buflen = sizeof (vp->vp_nvlist);
1126 1131
1127 1132 if (nvlist_pack(label, &buf, &buflen, NV_ENCODE_XDR, KM_SLEEP) == 0) {
1128 1133 for (; l < VDEV_LABELS; l += 2) {
1129 1134 vdev_label_write(zio, vd, l, vp,
1130 1135 offsetof(vdev_label_t, vl_vdev_phys),
1131 1136 sizeof (vdev_phys_t),
1132 1137 vdev_label_sync_done, zio->io_private,
1133 1138 flags | ZIO_FLAG_DONT_PROPAGATE);
1134 1139 }
1135 1140 }
1136 1141
1137 1142 zio_buf_free(vp, sizeof (vdev_phys_t));
1138 1143 nvlist_free(label);
1139 1144 }
1140 1145
1141 1146 int
1142 1147 vdev_label_sync_list(spa_t *spa, int l, uint64_t txg, int flags)
1143 1148 {
1144 1149 list_t *dl = &spa->spa_config_dirty_list;
1145 1150 vdev_t *vd;
1146 1151 zio_t *zio;
1147 1152 int error;
1148 1153
1149 1154 /*
1150 1155 * Write the new labels to disk.
1151 1156 */
1152 1157 zio = zio_root(spa, NULL, NULL, flags);
1153 1158
1154 1159 for (vd = list_head(dl); vd != NULL; vd = list_next(dl, vd)) {
1155 1160 uint64_t *good_writes = kmem_zalloc(sizeof (uint64_t),
1156 1161 KM_SLEEP);
1157 1162
1158 1163 ASSERT(!vd->vdev_ishole);
1159 1164
1160 1165 zio_t *vio = zio_null(zio, spa, NULL,
1161 1166 (vd->vdev_islog || vd->vdev_aux != NULL) ?
1162 1167 vdev_label_sync_ignore_done : vdev_label_sync_top_done,
1163 1168 good_writes, flags);
1164 1169 vdev_label_sync(vio, vd, l, txg, flags);
1165 1170 zio_nowait(vio);
1166 1171 }
1167 1172
1168 1173 error = zio_wait(zio);
1169 1174
1170 1175 /*
1171 1176 * Flush the new labels to disk.
1172 1177 */
1173 1178 zio = zio_root(spa, NULL, NULL, flags);
1174 1179
1175 1180 for (vd = list_head(dl); vd != NULL; vd = list_next(dl, vd))
1176 1181 zio_flush(zio, vd);
1177 1182
1178 1183 (void) zio_wait(zio);
1179 1184
1180 1185 return (error);
1181 1186 }
1182 1187
1183 1188 /*
1184 1189 * Sync the uberblock and any changes to the vdev configuration.
1185 1190 *
1186 1191 * The order of operations is carefully crafted to ensure that
1187 1192 * if the system panics or loses power at any time, the state on disk
1188 1193 * is still transactionally consistent. The in-line comments below
1189 1194 * describe the failure semantics at each stage.
1190 1195 *
1191 1196 * Moreover, vdev_config_sync() is designed to be idempotent: if it fails
1192 1197 * at any time, you can just call it again, and it will resume its work.
1193 1198 */
1194 1199 int
1195 1200 vdev_config_sync(vdev_t **svd, int svdcount, uint64_t txg, boolean_t tryhard)
1196 1201 {
1197 1202 spa_t *spa = svd[0]->vdev_spa;
1198 1203 uberblock_t *ub = &spa->spa_uberblock;
1199 1204 vdev_t *vd;
1200 1205 zio_t *zio;
1201 1206 int error;
1202 1207 int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL;
1203 1208
1204 1209 /*
1205 1210 * Normally, we don't want to try too hard to write every label and
1206 1211 * uberblock. If there is a flaky disk, we don't want the rest of the
1207 1212 * sync process to block while we retry. But if we can't write a
1208 1213 * single label out, we should retry with ZIO_FLAG_TRYHARD before
1209 1214 * bailing out and declaring the pool faulted.
1210 1215 */
1211 1216 if (tryhard)
1212 1217 flags |= ZIO_FLAG_TRYHARD;
1213 1218
1214 1219 ASSERT(ub->ub_txg <= txg);
1215 1220
1216 1221 /*
1217 1222 * If this isn't a resync due to I/O errors,
1218 1223 * and nothing changed in this transaction group,
1219 1224 * and the vdev configuration hasn't changed,
1220 1225 * then there's nothing to do.
1221 1226 */
1222 1227 if (ub->ub_txg < txg &&
1223 1228 uberblock_update(ub, spa->spa_root_vdev, txg) == B_FALSE &&
1224 1229 list_is_empty(&spa->spa_config_dirty_list))
1225 1230 return (0);
1226 1231
1227 1232 if (txg > spa_freeze_txg(spa))
1228 1233 return (0);
1229 1234
1230 1235 ASSERT(txg <= spa->spa_final_txg);
1231 1236
1232 1237 /*
1233 1238 * Flush the write cache of every disk that's been written to
1234 1239 * in this transaction group. This ensures that all blocks
1235 1240 * written in this txg will be committed to stable storage
1236 1241 * before any uberblock that references them.
1237 1242 */
1238 1243 zio = zio_root(spa, NULL, NULL, flags);
1239 1244
1240 1245 for (vd = txg_list_head(&spa->spa_vdev_txg_list, TXG_CLEAN(txg)); vd;
1241 1246 vd = txg_list_next(&spa->spa_vdev_txg_list, vd, TXG_CLEAN(txg)))
1242 1247 zio_flush(zio, vd);
1243 1248
1244 1249 (void) zio_wait(zio);
1245 1250
1246 1251 /*
1247 1252 * Sync out the even labels (L0, L2) for every dirty vdev. If the
1248 1253 * system dies in the middle of this process, that's OK: all of the
1249 1254 * even labels that made it to disk will be newer than any uberblock,
1250 1255 * and will therefore be considered invalid. The odd labels (L1, L3),
1251 1256 * which have not yet been touched, will still be valid. We flush
1252 1257 * the new labels to disk to ensure that all even-label updates
1253 1258 * are committed to stable storage before the uberblock update.
1254 1259 */
1255 1260 if ((error = vdev_label_sync_list(spa, 0, txg, flags)) != 0)
1256 1261 return (error);
1257 1262
1258 1263 /*
1259 1264 * Sync the uberblocks to all vdevs in svd[].
1260 1265 * If the system dies in the middle of this step, there are two cases
1261 1266 * to consider, and the on-disk state is consistent either way:
1262 1267 *
1263 1268 * (1) If none of the new uberblocks made it to disk, then the
1264 1269 * previous uberblock will be the newest, and the odd labels
1265 1270 * (which had not yet been touched) will be valid with respect
1266 1271 * to that uberblock.
1267 1272 *
1268 1273 * (2) If one or more new uberblocks made it to disk, then they
1269 1274 * will be the newest, and the even labels (which had all
1270 1275 * been successfully committed) will be valid with respect
1271 1276 * to the new uberblocks.
1272 1277 */
1273 1278 if ((error = vdev_uberblock_sync_list(svd, svdcount, ub, flags)) != 0)
1274 1279 return (error);
1275 1280
1276 1281 /*
1277 1282 * Sync out odd labels for every dirty vdev. If the system dies
1278 1283 * in the middle of this process, the even labels and the new
1279 1284 * uberblocks will suffice to open the pool. The next time
1280 1285 * the pool is opened, the first thing we'll do -- before any
1281 1286 * user data is modified -- is mark every vdev dirty so that
1282 1287 * all labels will be brought up to date. We flush the new labels
1283 1288 * to disk to ensure that all odd-label updates are committed to
1284 1289 * stable storage before the next transaction group begins.
1285 1290 */
1286 1291 return (vdev_label_sync_list(spa, 1, txg, flags));
1287 1292 }
|
↓ open down ↓ |
984 lines elided |
↑ open up ↑ |
XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX