Print this page
4101 metaslab_debug should allow for fine-grained control
4102 space_maps should store more information about themselves
4103 space map object blocksize should be increased
4104 ::spa_space no longer works
4105 removing a mirrored log device results in a leaked object
4106 asynchronously load metaslab
Reviewed by: Matthew Ahrens <mahrens@delphix.com>
Reviewed by: Adam Leventhal <ahl@delphix.com>
Reviewed by: Sebastien Roy <seb@delphix.com>
| Split |
Close |
| Expand all |
| Collapse all |
--- old/usr/src/uts/common/fs/zfs/vdev_label.c
+++ new/usr/src/uts/common/fs/zfs/vdev_label.c
1 1 /*
2 2 * CDDL HEADER START
3 3 *
4 4 * The contents of this file are subject to the terms of the
5 5 * Common Development and Distribution License (the "License").
6 6 * You may not use this file except in compliance with the License.
7 7 *
8 8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 9 * or http://www.opensolaris.org/os/licensing.
10 10 * See the License for the specific language governing permissions
11 11 * and limitations under the License.
12 12 *
13 13 * When distributing Covered Code, include this CDDL HEADER in each
14 14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 15 * If applicable, add the following below this CDDL HEADER, with the
16 16 * fields enclosed by brackets "[]" replaced with your own identifying
17 17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 18 *
19 19 * CDDL HEADER END
20 20 */
21 21
22 22 /*
23 23 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
24 24 * Copyright (c) 2013 by Delphix. All rights reserved.
25 25 */
26 26
27 27 /*
28 28 * Virtual Device Labels
29 29 * ---------------------
30 30 *
31 31 * The vdev label serves several distinct purposes:
32 32 *
33 33 * 1. Uniquely identify this device as part of a ZFS pool and confirm its
34 34 * identity within the pool.
35 35 *
36 36 * 2. Verify that all the devices given in a configuration are present
37 37 * within the pool.
38 38 *
39 39 * 3. Determine the uberblock for the pool.
40 40 *
41 41 * 4. In case of an import operation, determine the configuration of the
42 42 * toplevel vdev of which it is a part.
43 43 *
44 44 * 5. If an import operation cannot find all the devices in the pool,
45 45 * provide enough information to the administrator to determine which
46 46 * devices are missing.
47 47 *
48 48 * It is important to note that while the kernel is responsible for writing the
49 49 * label, it only consumes the information in the first three cases. The
50 50 * latter information is only consumed in userland when determining the
51 51 * configuration to import a pool.
52 52 *
53 53 *
54 54 * Label Organization
55 55 * ------------------
56 56 *
57 57 * Before describing the contents of the label, it's important to understand how
58 58 * the labels are written and updated with respect to the uberblock.
59 59 *
60 60 * When the pool configuration is altered, either because it was newly created
61 61 * or a device was added, we want to update all the labels such that we can deal
62 62 * with fatal failure at any point. To this end, each disk has two labels which
63 63 * are updated before and after the uberblock is synced. Assuming we have
64 64 * labels and an uberblock with the following transaction groups:
65 65 *
66 66 * L1 UB L2
67 67 * +------+ +------+ +------+
68 68 * | | | | | |
69 69 * | t10 | | t10 | | t10 |
70 70 * | | | | | |
71 71 * +------+ +------+ +------+
72 72 *
73 73 * In this stable state, the labels and the uberblock were all updated within
74 74 * the same transaction group (10). Each label is mirrored and checksummed, so
75 75 * that we can detect when we fail partway through writing the label.
76 76 *
77 77 * In order to identify which labels are valid, the labels are written in the
78 78 * following manner:
79 79 *
80 80 * 1. For each vdev, update 'L1' to the new label
81 81 * 2. Update the uberblock
82 82 * 3. For each vdev, update 'L2' to the new label
83 83 *
84 84 * Given arbitrary failure, we can determine the correct label to use based on
85 85 * the transaction group. If we fail after updating L1 but before updating the
86 86 * UB, we will notice that L1's transaction group is greater than the uberblock,
87 87 * so L2 must be valid. If we fail after writing the uberblock but before
88 88 * writing L2, we will notice that L2's transaction group is less than L1, and
89 89 * therefore L1 is valid.
90 90 *
91 91 * Another added complexity is that not every label is updated when the config
92 92 * is synced. If we add a single device, we do not want to have to re-write
93 93 * every label for every device in the pool. This means that both L1 and L2 may
94 94 * be older than the pool uberblock, because the necessary information is stored
95 95 * on another vdev.
96 96 *
97 97 *
98 98 * On-disk Format
99 99 * --------------
100 100 *
101 101 * The vdev label consists of two distinct parts, and is wrapped within the
102 102 * vdev_label_t structure. The label includes 8k of padding to permit legacy
103 103 * VTOC disk labels, but is otherwise ignored.
104 104 *
105 105 * The first half of the label is a packed nvlist which contains pool wide
106 106 * properties, per-vdev properties, and configuration information. It is
107 107 * described in more detail below.
108 108 *
109 109 * The latter half of the label consists of a redundant array of uberblocks.
110 110 * These uberblocks are updated whenever a transaction group is committed,
111 111 * or when the configuration is updated. When a pool is loaded, we scan each
112 112 * vdev for the 'best' uberblock.
113 113 *
114 114 *
115 115 * Configuration Information
116 116 * -------------------------
117 117 *
118 118 * The nvlist describing the pool and vdev contains the following elements:
119 119 *
120 120 * version ZFS on-disk version
121 121 * name Pool name
122 122 * state Pool state
123 123 * txg Transaction group in which this label was written
124 124 * pool_guid Unique identifier for this pool
125 125 * vdev_tree An nvlist describing vdev tree.
126 126 * features_for_read
127 127 * An nvlist of the features necessary for reading the MOS.
128 128 *
129 129 * Each leaf device label also contains the following:
130 130 *
131 131 * top_guid Unique ID for top-level vdev in which this is contained
132 132 * guid Unique ID for the leaf vdev
133 133 *
134 134 * The 'vs' configuration follows the format described in 'spa_config.c'.
135 135 */
136 136
137 137 #include <sys/zfs_context.h>
138 138 #include <sys/spa.h>
139 139 #include <sys/spa_impl.h>
140 140 #include <sys/dmu.h>
141 141 #include <sys/zap.h>
142 142 #include <sys/vdev.h>
143 143 #include <sys/vdev_impl.h>
144 144 #include <sys/uberblock_impl.h>
145 145 #include <sys/metaslab.h>
146 146 #include <sys/zio.h>
147 147 #include <sys/dsl_scan.h>
148 148 #include <sys/fs/zfs.h>
149 149
150 150 /*
151 151 * Basic routines to read and write from a vdev label.
152 152 * Used throughout the rest of this file.
153 153 */
154 154 uint64_t
155 155 vdev_label_offset(uint64_t psize, int l, uint64_t offset)
156 156 {
157 157 ASSERT(offset < sizeof (vdev_label_t));
158 158 ASSERT(P2PHASE_TYPED(psize, sizeof (vdev_label_t), uint64_t) == 0);
159 159
160 160 return (offset + l * sizeof (vdev_label_t) + (l < VDEV_LABELS / 2 ?
161 161 0 : psize - VDEV_LABELS * sizeof (vdev_label_t)));
162 162 }
163 163
164 164 /*
165 165 * Returns back the vdev label associated with the passed in offset.
166 166 */
167 167 int
168 168 vdev_label_number(uint64_t psize, uint64_t offset)
169 169 {
170 170 int l;
171 171
172 172 if (offset >= psize - VDEV_LABEL_END_SIZE) {
173 173 offset -= psize - VDEV_LABEL_END_SIZE;
174 174 offset += (VDEV_LABELS / 2) * sizeof (vdev_label_t);
175 175 }
176 176 l = offset / sizeof (vdev_label_t);
177 177 return (l < VDEV_LABELS ? l : -1);
178 178 }
179 179
180 180 static void
181 181 vdev_label_read(zio_t *zio, vdev_t *vd, int l, void *buf, uint64_t offset,
182 182 uint64_t size, zio_done_func_t *done, void *private, int flags)
183 183 {
184 184 ASSERT(spa_config_held(zio->io_spa, SCL_STATE_ALL, RW_WRITER) ==
185 185 SCL_STATE_ALL);
186 186 ASSERT(flags & ZIO_FLAG_CONFIG_WRITER);
187 187
188 188 zio_nowait(zio_read_phys(zio, vd,
189 189 vdev_label_offset(vd->vdev_psize, l, offset),
190 190 size, buf, ZIO_CHECKSUM_LABEL, done, private,
191 191 ZIO_PRIORITY_SYNC_READ, flags, B_TRUE));
192 192 }
193 193
194 194 static void
195 195 vdev_label_write(zio_t *zio, vdev_t *vd, int l, void *buf, uint64_t offset,
196 196 uint64_t size, zio_done_func_t *done, void *private, int flags)
197 197 {
198 198 ASSERT(spa_config_held(zio->io_spa, SCL_ALL, RW_WRITER) == SCL_ALL ||
199 199 (spa_config_held(zio->io_spa, SCL_CONFIG | SCL_STATE, RW_READER) ==
200 200 (SCL_CONFIG | SCL_STATE) &&
201 201 dsl_pool_sync_context(spa_get_dsl(zio->io_spa))));
202 202 ASSERT(flags & ZIO_FLAG_CONFIG_WRITER);
203 203
204 204 zio_nowait(zio_write_phys(zio, vd,
205 205 vdev_label_offset(vd->vdev_psize, l, offset),
206 206 size, buf, ZIO_CHECKSUM_LABEL, done, private,
207 207 ZIO_PRIORITY_SYNC_WRITE, flags, B_TRUE));
208 208 }
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 (vd->vdev_dtl_smo.smo_object != 0)
286 + if (vd->vdev_dtl_sm != NULL) {
287 287 fnvlist_add_uint64(nv, ZPOOL_CONFIG_DTL,
288 - vd->vdev_dtl_smo.smo_object);
288 + space_map_object(vd->vdev_dtl_sm));
289 + }
289 290
290 291 if (vd->vdev_crtxg)
291 292 fnvlist_add_uint64(nv, ZPOOL_CONFIG_CREATE_TXG, vd->vdev_crtxg);
292 293
293 294 if (getstats) {
294 295 vdev_stat_t vs;
295 296 pool_scan_stat_t ps;
296 297
297 298 vdev_get_stats(vd, &vs);
298 299 fnvlist_add_uint64_array(nv, ZPOOL_CONFIG_VDEV_STATS,
299 300 (uint64_t *)&vs, sizeof (vs) / sizeof (uint64_t));
300 301
301 302 /* provide either current or previous scan information */
302 303 if (spa_scan_get_stats(spa, &ps) == 0) {
303 304 fnvlist_add_uint64_array(nv,
304 305 ZPOOL_CONFIG_SCAN_STATS, (uint64_t *)&ps,
305 306 sizeof (pool_scan_stat_t) / sizeof (uint64_t));
306 307 }
307 308 }
308 309
309 310 if (!vd->vdev_ops->vdev_op_leaf) {
310 311 nvlist_t **child;
311 312 int c, idx;
312 313
313 314 ASSERT(!vd->vdev_ishole);
314 315
315 316 child = kmem_alloc(vd->vdev_children * sizeof (nvlist_t *),
316 317 KM_SLEEP);
317 318
318 319 for (c = 0, idx = 0; c < vd->vdev_children; c++) {
319 320 vdev_t *cvd = vd->vdev_child[c];
320 321
321 322 /*
322 323 * If we're generating an nvlist of removing
323 324 * vdevs then skip over any device which is
324 325 * not being removed.
325 326 */
326 327 if ((flags & VDEV_CONFIG_REMOVING) &&
327 328 !cvd->vdev_removing)
328 329 continue;
329 330
330 331 child[idx++] = vdev_config_generate(spa, cvd,
331 332 getstats, flags);
332 333 }
333 334
334 335 if (idx) {
335 336 fnvlist_add_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
336 337 child, idx);
337 338 }
338 339
339 340 for (c = 0; c < idx; c++)
340 341 nvlist_free(child[c]);
341 342
342 343 kmem_free(child, vd->vdev_children * sizeof (nvlist_t *));
343 344
344 345 } else {
345 346 const char *aux = NULL;
346 347
347 348 if (vd->vdev_offline && !vd->vdev_tmpoffline)
348 349 fnvlist_add_uint64(nv, ZPOOL_CONFIG_OFFLINE, B_TRUE);
349 350 if (vd->vdev_resilver_txg != 0)
350 351 fnvlist_add_uint64(nv, ZPOOL_CONFIG_RESILVER_TXG,
351 352 vd->vdev_resilver_txg);
352 353 if (vd->vdev_faulted)
353 354 fnvlist_add_uint64(nv, ZPOOL_CONFIG_FAULTED, B_TRUE);
354 355 if (vd->vdev_degraded)
355 356 fnvlist_add_uint64(nv, ZPOOL_CONFIG_DEGRADED, B_TRUE);
356 357 if (vd->vdev_removed)
357 358 fnvlist_add_uint64(nv, ZPOOL_CONFIG_REMOVED, B_TRUE);
358 359 if (vd->vdev_unspare)
359 360 fnvlist_add_uint64(nv, ZPOOL_CONFIG_UNSPARE, B_TRUE);
360 361 if (vd->vdev_ishole)
361 362 fnvlist_add_uint64(nv, ZPOOL_CONFIG_IS_HOLE, B_TRUE);
362 363
363 364 switch (vd->vdev_stat.vs_aux) {
364 365 case VDEV_AUX_ERR_EXCEEDED:
365 366 aux = "err_exceeded";
366 367 break;
367 368
368 369 case VDEV_AUX_EXTERNAL:
369 370 aux = "external";
370 371 break;
371 372 }
372 373
373 374 if (aux != NULL)
374 375 fnvlist_add_string(nv, ZPOOL_CONFIG_AUX_STATE, aux);
375 376
376 377 if (vd->vdev_splitting && vd->vdev_orig_guid != 0LL) {
377 378 fnvlist_add_uint64(nv, ZPOOL_CONFIG_ORIG_GUID,
378 379 vd->vdev_orig_guid);
379 380 }
380 381 }
381 382
382 383 return (nv);
383 384 }
384 385
385 386 /*
386 387 * Generate a view of the top-level vdevs. If we currently have holes
387 388 * in the namespace, then generate an array which contains a list of holey
388 389 * vdevs. Additionally, add the number of top-level children that currently
389 390 * exist.
390 391 */
391 392 void
392 393 vdev_top_config_generate(spa_t *spa, nvlist_t *config)
393 394 {
394 395 vdev_t *rvd = spa->spa_root_vdev;
395 396 uint64_t *array;
396 397 uint_t c, idx;
397 398
398 399 array = kmem_alloc(rvd->vdev_children * sizeof (uint64_t), KM_SLEEP);
399 400
400 401 for (c = 0, idx = 0; c < rvd->vdev_children; c++) {
401 402 vdev_t *tvd = rvd->vdev_child[c];
402 403
403 404 if (tvd->vdev_ishole)
404 405 array[idx++] = c;
405 406 }
406 407
407 408 if (idx) {
408 409 VERIFY(nvlist_add_uint64_array(config, ZPOOL_CONFIG_HOLE_ARRAY,
409 410 array, idx) == 0);
410 411 }
411 412
412 413 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VDEV_CHILDREN,
413 414 rvd->vdev_children) == 0);
414 415
415 416 kmem_free(array, rvd->vdev_children * sizeof (uint64_t));
416 417 }
417 418
418 419 /*
419 420 * Returns the configuration from the label of the given vdev. For vdevs
420 421 * which don't have a txg value stored on their label (i.e. spares/cache)
421 422 * or have not been completely initialized (txg = 0) just return
422 423 * the configuration from the first valid label we find. Otherwise,
423 424 * find the most up-to-date label that does not exceed the specified
424 425 * 'txg' value.
425 426 */
426 427 nvlist_t *
427 428 vdev_label_read_config(vdev_t *vd, uint64_t txg)
428 429 {
429 430 spa_t *spa = vd->vdev_spa;
430 431 nvlist_t *config = NULL;
431 432 vdev_phys_t *vp;
432 433 zio_t *zio;
433 434 uint64_t best_txg = 0;
434 435 int error = 0;
435 436 int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL |
436 437 ZIO_FLAG_SPECULATIVE;
437 438
438 439 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
439 440
440 441 if (!vdev_readable(vd))
441 442 return (NULL);
442 443
443 444 vp = zio_buf_alloc(sizeof (vdev_phys_t));
444 445
445 446 retry:
446 447 for (int l = 0; l < VDEV_LABELS; l++) {
447 448 nvlist_t *label = NULL;
448 449
449 450 zio = zio_root(spa, NULL, NULL, flags);
450 451
451 452 vdev_label_read(zio, vd, l, vp,
452 453 offsetof(vdev_label_t, vl_vdev_phys),
453 454 sizeof (vdev_phys_t), NULL, NULL, flags);
454 455
455 456 if (zio_wait(zio) == 0 &&
456 457 nvlist_unpack(vp->vp_nvlist, sizeof (vp->vp_nvlist),
457 458 &label, 0) == 0) {
458 459 uint64_t label_txg = 0;
459 460
460 461 /*
461 462 * Auxiliary vdevs won't have txg values in their
462 463 * labels and newly added vdevs may not have been
463 464 * completely initialized so just return the
464 465 * configuration from the first valid label we
465 466 * encounter.
466 467 */
467 468 error = nvlist_lookup_uint64(label,
468 469 ZPOOL_CONFIG_POOL_TXG, &label_txg);
469 470 if ((error || label_txg == 0) && !config) {
470 471 config = label;
471 472 break;
472 473 } else if (label_txg <= txg && label_txg > best_txg) {
473 474 best_txg = label_txg;
474 475 nvlist_free(config);
475 476 config = fnvlist_dup(label);
476 477 }
477 478 }
478 479
479 480 if (label != NULL) {
480 481 nvlist_free(label);
481 482 label = NULL;
482 483 }
483 484 }
484 485
485 486 if (config == NULL && !(flags & ZIO_FLAG_TRYHARD)) {
486 487 flags |= ZIO_FLAG_TRYHARD;
487 488 goto retry;
488 489 }
489 490
490 491 zio_buf_free(vp, sizeof (vdev_phys_t));
491 492
492 493 return (config);
493 494 }
494 495
495 496 /*
496 497 * Determine if a device is in use. The 'spare_guid' parameter will be filled
497 498 * in with the device guid if this spare is active elsewhere on the system.
498 499 */
499 500 static boolean_t
500 501 vdev_inuse(vdev_t *vd, uint64_t crtxg, vdev_labeltype_t reason,
501 502 uint64_t *spare_guid, uint64_t *l2cache_guid)
502 503 {
503 504 spa_t *spa = vd->vdev_spa;
504 505 uint64_t state, pool_guid, device_guid, txg, spare_pool;
505 506 uint64_t vdtxg = 0;
506 507 nvlist_t *label;
507 508
508 509 if (spare_guid)
509 510 *spare_guid = 0ULL;
510 511 if (l2cache_guid)
511 512 *l2cache_guid = 0ULL;
512 513
513 514 /*
514 515 * Read the label, if any, and perform some basic sanity checks.
515 516 */
516 517 if ((label = vdev_label_read_config(vd, -1ULL)) == NULL)
517 518 return (B_FALSE);
518 519
519 520 (void) nvlist_lookup_uint64(label, ZPOOL_CONFIG_CREATE_TXG,
520 521 &vdtxg);
521 522
522 523 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE,
523 524 &state) != 0 ||
524 525 nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID,
525 526 &device_guid) != 0) {
526 527 nvlist_free(label);
527 528 return (B_FALSE);
528 529 }
529 530
530 531 if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
531 532 (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_GUID,
532 533 &pool_guid) != 0 ||
533 534 nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG,
534 535 &txg) != 0)) {
535 536 nvlist_free(label);
536 537 return (B_FALSE);
537 538 }
538 539
539 540 nvlist_free(label);
540 541
541 542 /*
542 543 * Check to see if this device indeed belongs to the pool it claims to
543 544 * be a part of. The only way this is allowed is if the device is a hot
544 545 * spare (which we check for later on).
545 546 */
546 547 if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
547 548 !spa_guid_exists(pool_guid, device_guid) &&
548 549 !spa_spare_exists(device_guid, NULL, NULL) &&
549 550 !spa_l2cache_exists(device_guid, NULL))
550 551 return (B_FALSE);
551 552
552 553 /*
553 554 * If the transaction group is zero, then this an initialized (but
554 555 * unused) label. This is only an error if the create transaction
555 556 * on-disk is the same as the one we're using now, in which case the
556 557 * user has attempted to add the same vdev multiple times in the same
557 558 * transaction.
558 559 */
559 560 if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
560 561 txg == 0 && vdtxg == crtxg)
561 562 return (B_TRUE);
562 563
563 564 /*
564 565 * Check to see if this is a spare device. We do an explicit check for
565 566 * spa_has_spare() here because it may be on our pending list of spares
566 567 * to add. We also check if it is an l2cache device.
567 568 */
568 569 if (spa_spare_exists(device_guid, &spare_pool, NULL) ||
569 570 spa_has_spare(spa, device_guid)) {
570 571 if (spare_guid)
571 572 *spare_guid = device_guid;
572 573
573 574 switch (reason) {
574 575 case VDEV_LABEL_CREATE:
575 576 case VDEV_LABEL_L2CACHE:
576 577 return (B_TRUE);
577 578
578 579 case VDEV_LABEL_REPLACE:
579 580 return (!spa_has_spare(spa, device_guid) ||
580 581 spare_pool != 0ULL);
581 582
582 583 case VDEV_LABEL_SPARE:
583 584 return (spa_has_spare(spa, device_guid));
584 585 }
585 586 }
586 587
587 588 /*
588 589 * Check to see if this is an l2cache device.
589 590 */
590 591 if (spa_l2cache_exists(device_guid, NULL))
591 592 return (B_TRUE);
592 593
593 594 /*
594 595 * We can't rely on a pool's state if it's been imported
595 596 * read-only. Instead we look to see if the pools is marked
596 597 * read-only in the namespace and set the state to active.
597 598 */
598 599 if ((spa = spa_by_guid(pool_guid, device_guid)) != NULL &&
599 600 spa_mode(spa) == FREAD)
600 601 state = POOL_STATE_ACTIVE;
601 602
602 603 /*
603 604 * If the device is marked ACTIVE, then this device is in use by another
604 605 * pool on the system.
605 606 */
606 607 return (state == POOL_STATE_ACTIVE);
607 608 }
608 609
609 610 /*
610 611 * Initialize a vdev label. We check to make sure each leaf device is not in
611 612 * use, and writable. We put down an initial label which we will later
612 613 * overwrite with a complete label. Note that it's important to do this
613 614 * sequentially, not in parallel, so that we catch cases of multiple use of the
614 615 * same leaf vdev in the vdev we're creating -- e.g. mirroring a disk with
615 616 * itself.
616 617 */
617 618 int
618 619 vdev_label_init(vdev_t *vd, uint64_t crtxg, vdev_labeltype_t reason)
619 620 {
620 621 spa_t *spa = vd->vdev_spa;
621 622 nvlist_t *label;
622 623 vdev_phys_t *vp;
623 624 char *pad2;
624 625 uberblock_t *ub;
625 626 zio_t *zio;
626 627 char *buf;
627 628 size_t buflen;
628 629 int error;
629 630 uint64_t spare_guid, l2cache_guid;
630 631 int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL;
631 632
632 633 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
633 634
634 635 for (int c = 0; c < vd->vdev_children; c++)
635 636 if ((error = vdev_label_init(vd->vdev_child[c],
636 637 crtxg, reason)) != 0)
637 638 return (error);
638 639
639 640 /* Track the creation time for this vdev */
640 641 vd->vdev_crtxg = crtxg;
641 642
642 643 if (!vd->vdev_ops->vdev_op_leaf)
643 644 return (0);
644 645
645 646 /*
646 647 * Dead vdevs cannot be initialized.
647 648 */
648 649 if (vdev_is_dead(vd))
649 650 return (SET_ERROR(EIO));
650 651
651 652 /*
652 653 * Determine if the vdev is in use.
653 654 */
654 655 if (reason != VDEV_LABEL_REMOVE && reason != VDEV_LABEL_SPLIT &&
655 656 vdev_inuse(vd, crtxg, reason, &spare_guid, &l2cache_guid))
656 657 return (SET_ERROR(EBUSY));
657 658
658 659 /*
659 660 * If this is a request to add or replace a spare or l2cache device
660 661 * that is in use elsewhere on the system, then we must update the
661 662 * guid (which was initialized to a random value) to reflect the
662 663 * actual GUID (which is shared between multiple pools).
663 664 */
664 665 if (reason != VDEV_LABEL_REMOVE && reason != VDEV_LABEL_L2CACHE &&
665 666 spare_guid != 0ULL) {
666 667 uint64_t guid_delta = spare_guid - vd->vdev_guid;
667 668
668 669 vd->vdev_guid += guid_delta;
669 670
670 671 for (vdev_t *pvd = vd; pvd != NULL; pvd = pvd->vdev_parent)
671 672 pvd->vdev_guid_sum += guid_delta;
672 673
673 674 /*
674 675 * If this is a replacement, then we want to fallthrough to the
675 676 * rest of the code. If we're adding a spare, then it's already
676 677 * labeled appropriately and we can just return.
677 678 */
678 679 if (reason == VDEV_LABEL_SPARE)
679 680 return (0);
680 681 ASSERT(reason == VDEV_LABEL_REPLACE ||
681 682 reason == VDEV_LABEL_SPLIT);
682 683 }
683 684
684 685 if (reason != VDEV_LABEL_REMOVE && reason != VDEV_LABEL_SPARE &&
685 686 l2cache_guid != 0ULL) {
686 687 uint64_t guid_delta = l2cache_guid - vd->vdev_guid;
687 688
688 689 vd->vdev_guid += guid_delta;
689 690
690 691 for (vdev_t *pvd = vd; pvd != NULL; pvd = pvd->vdev_parent)
691 692 pvd->vdev_guid_sum += guid_delta;
692 693
693 694 /*
694 695 * If this is a replacement, then we want to fallthrough to the
695 696 * rest of the code. If we're adding an l2cache, then it's
696 697 * already labeled appropriately and we can just return.
697 698 */
698 699 if (reason == VDEV_LABEL_L2CACHE)
699 700 return (0);
700 701 ASSERT(reason == VDEV_LABEL_REPLACE);
701 702 }
702 703
703 704 /*
704 705 * Initialize its label.
705 706 */
706 707 vp = zio_buf_alloc(sizeof (vdev_phys_t));
707 708 bzero(vp, sizeof (vdev_phys_t));
708 709
709 710 /*
710 711 * Generate a label describing the pool and our top-level vdev.
711 712 * We mark it as being from txg 0 to indicate that it's not
712 713 * really part of an active pool just yet. The labels will
713 714 * be written again with a meaningful txg by spa_sync().
714 715 */
715 716 if (reason == VDEV_LABEL_SPARE ||
716 717 (reason == VDEV_LABEL_REMOVE && vd->vdev_isspare)) {
717 718 /*
718 719 * For inactive hot spares, we generate a special label that
719 720 * identifies as a mutually shared hot spare. We write the
720 721 * label if we are adding a hot spare, or if we are removing an
721 722 * active hot spare (in which case we want to revert the
722 723 * labels).
723 724 */
724 725 VERIFY(nvlist_alloc(&label, NV_UNIQUE_NAME, KM_SLEEP) == 0);
725 726
726 727 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_VERSION,
727 728 spa_version(spa)) == 0);
728 729 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_POOL_STATE,
729 730 POOL_STATE_SPARE) == 0);
730 731 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_GUID,
731 732 vd->vdev_guid) == 0);
732 733 } else if (reason == VDEV_LABEL_L2CACHE ||
733 734 (reason == VDEV_LABEL_REMOVE && vd->vdev_isl2cache)) {
734 735 /*
735 736 * For level 2 ARC devices, add a special label.
736 737 */
737 738 VERIFY(nvlist_alloc(&label, NV_UNIQUE_NAME, KM_SLEEP) == 0);
738 739
739 740 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_VERSION,
740 741 spa_version(spa)) == 0);
741 742 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_POOL_STATE,
742 743 POOL_STATE_L2CACHE) == 0);
743 744 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_GUID,
744 745 vd->vdev_guid) == 0);
745 746 } else {
746 747 uint64_t txg = 0ULL;
747 748
748 749 if (reason == VDEV_LABEL_SPLIT)
749 750 txg = spa->spa_uberblock.ub_txg;
750 751 label = spa_config_generate(spa, vd, txg, B_FALSE);
751 752
752 753 /*
753 754 * Add our creation time. This allows us to detect multiple
754 755 * vdev uses as described above, and automatically expires if we
755 756 * fail.
756 757 */
757 758 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_CREATE_TXG,
758 759 crtxg) == 0);
759 760 }
760 761
761 762 buf = vp->vp_nvlist;
762 763 buflen = sizeof (vp->vp_nvlist);
763 764
764 765 error = nvlist_pack(label, &buf, &buflen, NV_ENCODE_XDR, KM_SLEEP);
765 766 if (error != 0) {
766 767 nvlist_free(label);
767 768 zio_buf_free(vp, sizeof (vdev_phys_t));
768 769 /* EFAULT means nvlist_pack ran out of room */
769 770 return (error == EFAULT ? ENAMETOOLONG : EINVAL);
770 771 }
771 772
772 773 /*
773 774 * Initialize uberblock template.
774 775 */
775 776 ub = zio_buf_alloc(VDEV_UBERBLOCK_RING);
776 777 bzero(ub, VDEV_UBERBLOCK_RING);
777 778 *ub = spa->spa_uberblock;
778 779 ub->ub_txg = 0;
779 780
780 781 /* Initialize the 2nd padding area. */
781 782 pad2 = zio_buf_alloc(VDEV_PAD_SIZE);
782 783 bzero(pad2, VDEV_PAD_SIZE);
783 784
784 785 /*
785 786 * Write everything in parallel.
786 787 */
787 788 retry:
788 789 zio = zio_root(spa, NULL, NULL, flags);
789 790
790 791 for (int l = 0; l < VDEV_LABELS; l++) {
791 792
792 793 vdev_label_write(zio, vd, l, vp,
793 794 offsetof(vdev_label_t, vl_vdev_phys),
794 795 sizeof (vdev_phys_t), NULL, NULL, flags);
795 796
796 797 /*
797 798 * Skip the 1st padding area.
798 799 * Zero out the 2nd padding area where it might have
799 800 * left over data from previous filesystem format.
800 801 */
801 802 vdev_label_write(zio, vd, l, pad2,
802 803 offsetof(vdev_label_t, vl_pad2),
803 804 VDEV_PAD_SIZE, NULL, NULL, flags);
804 805
805 806 vdev_label_write(zio, vd, l, ub,
806 807 offsetof(vdev_label_t, vl_uberblock),
807 808 VDEV_UBERBLOCK_RING, NULL, NULL, flags);
808 809 }
809 810
810 811 error = zio_wait(zio);
811 812
812 813 if (error != 0 && !(flags & ZIO_FLAG_TRYHARD)) {
813 814 flags |= ZIO_FLAG_TRYHARD;
814 815 goto retry;
815 816 }
816 817
817 818 nvlist_free(label);
818 819 zio_buf_free(pad2, VDEV_PAD_SIZE);
819 820 zio_buf_free(ub, VDEV_UBERBLOCK_RING);
820 821 zio_buf_free(vp, sizeof (vdev_phys_t));
821 822
822 823 /*
823 824 * If this vdev hasn't been previously identified as a spare, then we
824 825 * mark it as such only if a) we are labeling it as a spare, or b) it
825 826 * exists as a spare elsewhere in the system. Do the same for
826 827 * level 2 ARC devices.
827 828 */
828 829 if (error == 0 && !vd->vdev_isspare &&
829 830 (reason == VDEV_LABEL_SPARE ||
830 831 spa_spare_exists(vd->vdev_guid, NULL, NULL)))
831 832 spa_spare_add(vd);
832 833
833 834 if (error == 0 && !vd->vdev_isl2cache &&
834 835 (reason == VDEV_LABEL_L2CACHE ||
835 836 spa_l2cache_exists(vd->vdev_guid, NULL)))
836 837 spa_l2cache_add(vd);
837 838
838 839 return (error);
839 840 }
840 841
841 842 /*
842 843 * ==========================================================================
843 844 * uberblock load/sync
844 845 * ==========================================================================
845 846 */
846 847
847 848 /*
848 849 * Consider the following situation: txg is safely synced to disk. We've
849 850 * written the first uberblock for txg + 1, and then we lose power. When we
850 851 * come back up, we fail to see the uberblock for txg + 1 because, say,
851 852 * it was on a mirrored device and the replica to which we wrote txg + 1
852 853 * is now offline. If we then make some changes and sync txg + 1, and then
853 854 * the missing replica comes back, then for a few seconds we'll have two
854 855 * conflicting uberblocks on disk with the same txg. The solution is simple:
855 856 * among uberblocks with equal txg, choose the one with the latest timestamp.
856 857 */
857 858 static int
858 859 vdev_uberblock_compare(uberblock_t *ub1, uberblock_t *ub2)
859 860 {
860 861 if (ub1->ub_txg < ub2->ub_txg)
861 862 return (-1);
862 863 if (ub1->ub_txg > ub2->ub_txg)
863 864 return (1);
864 865
865 866 if (ub1->ub_timestamp < ub2->ub_timestamp)
866 867 return (-1);
867 868 if (ub1->ub_timestamp > ub2->ub_timestamp)
868 869 return (1);
869 870
870 871 return (0);
871 872 }
872 873
873 874 struct ubl_cbdata {
874 875 uberblock_t *ubl_ubbest; /* Best uberblock */
875 876 vdev_t *ubl_vd; /* vdev associated with the above */
876 877 };
877 878
878 879 static void
879 880 vdev_uberblock_load_done(zio_t *zio)
880 881 {
881 882 vdev_t *vd = zio->io_vd;
882 883 spa_t *spa = zio->io_spa;
883 884 zio_t *rio = zio->io_private;
884 885 uberblock_t *ub = zio->io_data;
885 886 struct ubl_cbdata *cbp = rio->io_private;
886 887
887 888 ASSERT3U(zio->io_size, ==, VDEV_UBERBLOCK_SIZE(vd));
888 889
889 890 if (zio->io_error == 0 && uberblock_verify(ub) == 0) {
890 891 mutex_enter(&rio->io_lock);
891 892 if (ub->ub_txg <= spa->spa_load_max_txg &&
892 893 vdev_uberblock_compare(ub, cbp->ubl_ubbest) > 0) {
893 894 /*
894 895 * Keep track of the vdev in which this uberblock
895 896 * was found. We will use this information later
896 897 * to obtain the config nvlist associated with
897 898 * this uberblock.
898 899 */
899 900 *cbp->ubl_ubbest = *ub;
900 901 cbp->ubl_vd = vd;
901 902 }
902 903 mutex_exit(&rio->io_lock);
903 904 }
904 905
905 906 zio_buf_free(zio->io_data, zio->io_size);
906 907 }
907 908
908 909 static void
909 910 vdev_uberblock_load_impl(zio_t *zio, vdev_t *vd, int flags,
910 911 struct ubl_cbdata *cbp)
911 912 {
912 913 for (int c = 0; c < vd->vdev_children; c++)
913 914 vdev_uberblock_load_impl(zio, vd->vdev_child[c], flags, cbp);
914 915
915 916 if (vd->vdev_ops->vdev_op_leaf && vdev_readable(vd)) {
916 917 for (int l = 0; l < VDEV_LABELS; l++) {
917 918 for (int n = 0; n < VDEV_UBERBLOCK_COUNT(vd); n++) {
918 919 vdev_label_read(zio, vd, l,
919 920 zio_buf_alloc(VDEV_UBERBLOCK_SIZE(vd)),
920 921 VDEV_UBERBLOCK_OFFSET(vd, n),
921 922 VDEV_UBERBLOCK_SIZE(vd),
922 923 vdev_uberblock_load_done, zio, flags);
923 924 }
924 925 }
925 926 }
926 927 }
927 928
928 929 /*
929 930 * Reads the 'best' uberblock from disk along with its associated
930 931 * configuration. First, we read the uberblock array of each label of each
931 932 * vdev, keeping track of the uberblock with the highest txg in each array.
932 933 * Then, we read the configuration from the same vdev as the best uberblock.
933 934 */
934 935 void
935 936 vdev_uberblock_load(vdev_t *rvd, uberblock_t *ub, nvlist_t **config)
936 937 {
937 938 zio_t *zio;
938 939 spa_t *spa = rvd->vdev_spa;
939 940 struct ubl_cbdata cb;
940 941 int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL |
941 942 ZIO_FLAG_SPECULATIVE | ZIO_FLAG_TRYHARD;
942 943
943 944 ASSERT(ub);
944 945 ASSERT(config);
945 946
946 947 bzero(ub, sizeof (uberblock_t));
947 948 *config = NULL;
948 949
949 950 cb.ubl_ubbest = ub;
950 951 cb.ubl_vd = NULL;
951 952
952 953 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
953 954 zio = zio_root(spa, NULL, &cb, flags);
954 955 vdev_uberblock_load_impl(zio, rvd, flags, &cb);
955 956 (void) zio_wait(zio);
956 957
957 958 /*
958 959 * It's possible that the best uberblock was discovered on a label
959 960 * that has a configuration which was written in a future txg.
960 961 * Search all labels on this vdev to find the configuration that
961 962 * matches the txg for our uberblock.
962 963 */
963 964 if (cb.ubl_vd != NULL)
964 965 *config = vdev_label_read_config(cb.ubl_vd, ub->ub_txg);
965 966 spa_config_exit(spa, SCL_ALL, FTAG);
966 967 }
967 968
968 969 /*
969 970 * On success, increment root zio's count of good writes.
970 971 * We only get credit for writes to known-visible vdevs; see spa_vdev_add().
971 972 */
972 973 static void
973 974 vdev_uberblock_sync_done(zio_t *zio)
974 975 {
975 976 uint64_t *good_writes = zio->io_private;
976 977
977 978 if (zio->io_error == 0 && zio->io_vd->vdev_top->vdev_ms_array != 0)
978 979 atomic_add_64(good_writes, 1);
979 980 }
980 981
981 982 /*
982 983 * Write the uberblock to all labels of all leaves of the specified vdev.
983 984 */
984 985 static void
985 986 vdev_uberblock_sync(zio_t *zio, uberblock_t *ub, vdev_t *vd, int flags)
986 987 {
987 988 uberblock_t *ubbuf;
988 989 int n;
989 990
990 991 for (int c = 0; c < vd->vdev_children; c++)
991 992 vdev_uberblock_sync(zio, ub, vd->vdev_child[c], flags);
992 993
993 994 if (!vd->vdev_ops->vdev_op_leaf)
994 995 return;
995 996
996 997 if (!vdev_writeable(vd))
997 998 return;
998 999
999 1000 n = ub->ub_txg & (VDEV_UBERBLOCK_COUNT(vd) - 1);
1000 1001
1001 1002 ubbuf = zio_buf_alloc(VDEV_UBERBLOCK_SIZE(vd));
1002 1003 bzero(ubbuf, VDEV_UBERBLOCK_SIZE(vd));
1003 1004 *ubbuf = *ub;
1004 1005
1005 1006 for (int l = 0; l < VDEV_LABELS; l++)
1006 1007 vdev_label_write(zio, vd, l, ubbuf,
1007 1008 VDEV_UBERBLOCK_OFFSET(vd, n), VDEV_UBERBLOCK_SIZE(vd),
1008 1009 vdev_uberblock_sync_done, zio->io_private,
1009 1010 flags | ZIO_FLAG_DONT_PROPAGATE);
1010 1011
1011 1012 zio_buf_free(ubbuf, VDEV_UBERBLOCK_SIZE(vd));
1012 1013 }
1013 1014
1014 1015 /* Sync the uberblocks to all vdevs in svd[] */
1015 1016 int
1016 1017 vdev_uberblock_sync_list(vdev_t **svd, int svdcount, uberblock_t *ub, int flags)
1017 1018 {
1018 1019 spa_t *spa = svd[0]->vdev_spa;
1019 1020 zio_t *zio;
1020 1021 uint64_t good_writes = 0;
1021 1022
1022 1023 zio = zio_root(spa, NULL, &good_writes, flags);
1023 1024
1024 1025 for (int v = 0; v < svdcount; v++)
1025 1026 vdev_uberblock_sync(zio, ub, svd[v], flags);
1026 1027
1027 1028 (void) zio_wait(zio);
1028 1029
1029 1030 /*
1030 1031 * Flush the uberblocks to disk. This ensures that the odd labels
1031 1032 * are no longer needed (because the new uberblocks and the even
1032 1033 * labels are safely on disk), so it is safe to overwrite them.
1033 1034 */
1034 1035 zio = zio_root(spa, NULL, NULL, flags);
1035 1036
1036 1037 for (int v = 0; v < svdcount; v++)
1037 1038 zio_flush(zio, svd[v]);
1038 1039
1039 1040 (void) zio_wait(zio);
1040 1041
1041 1042 return (good_writes >= 1 ? 0 : EIO);
1042 1043 }
1043 1044
1044 1045 /*
1045 1046 * On success, increment the count of good writes for our top-level vdev.
1046 1047 */
1047 1048 static void
1048 1049 vdev_label_sync_done(zio_t *zio)
1049 1050 {
1050 1051 uint64_t *good_writes = zio->io_private;
1051 1052
1052 1053 if (zio->io_error == 0)
1053 1054 atomic_add_64(good_writes, 1);
1054 1055 }
1055 1056
1056 1057 /*
1057 1058 * If there weren't enough good writes, indicate failure to the parent.
1058 1059 */
1059 1060 static void
1060 1061 vdev_label_sync_top_done(zio_t *zio)
1061 1062 {
1062 1063 uint64_t *good_writes = zio->io_private;
1063 1064
1064 1065 if (*good_writes == 0)
1065 1066 zio->io_error = SET_ERROR(EIO);
1066 1067
1067 1068 kmem_free(good_writes, sizeof (uint64_t));
1068 1069 }
1069 1070
1070 1071 /*
1071 1072 * We ignore errors for log and cache devices, simply free the private data.
1072 1073 */
1073 1074 static void
1074 1075 vdev_label_sync_ignore_done(zio_t *zio)
1075 1076 {
1076 1077 kmem_free(zio->io_private, sizeof (uint64_t));
1077 1078 }
1078 1079
1079 1080 /*
1080 1081 * Write all even or odd labels to all leaves of the specified vdev.
1081 1082 */
1082 1083 static void
1083 1084 vdev_label_sync(zio_t *zio, vdev_t *vd, int l, uint64_t txg, int flags)
1084 1085 {
1085 1086 nvlist_t *label;
1086 1087 vdev_phys_t *vp;
1087 1088 char *buf;
1088 1089 size_t buflen;
1089 1090
1090 1091 for (int c = 0; c < vd->vdev_children; c++)
1091 1092 vdev_label_sync(zio, vd->vdev_child[c], l, txg, flags);
1092 1093
1093 1094 if (!vd->vdev_ops->vdev_op_leaf)
1094 1095 return;
1095 1096
1096 1097 if (!vdev_writeable(vd))
1097 1098 return;
1098 1099
1099 1100 /*
1100 1101 * Generate a label describing the top-level config to which we belong.
1101 1102 */
1102 1103 label = spa_config_generate(vd->vdev_spa, vd, txg, B_FALSE);
1103 1104
1104 1105 vp = zio_buf_alloc(sizeof (vdev_phys_t));
1105 1106 bzero(vp, sizeof (vdev_phys_t));
1106 1107
1107 1108 buf = vp->vp_nvlist;
1108 1109 buflen = sizeof (vp->vp_nvlist);
1109 1110
1110 1111 if (nvlist_pack(label, &buf, &buflen, NV_ENCODE_XDR, KM_SLEEP) == 0) {
1111 1112 for (; l < VDEV_LABELS; l += 2) {
1112 1113 vdev_label_write(zio, vd, l, vp,
1113 1114 offsetof(vdev_label_t, vl_vdev_phys),
1114 1115 sizeof (vdev_phys_t),
1115 1116 vdev_label_sync_done, zio->io_private,
1116 1117 flags | ZIO_FLAG_DONT_PROPAGATE);
1117 1118 }
1118 1119 }
1119 1120
1120 1121 zio_buf_free(vp, sizeof (vdev_phys_t));
1121 1122 nvlist_free(label);
1122 1123 }
1123 1124
1124 1125 int
1125 1126 vdev_label_sync_list(spa_t *spa, int l, uint64_t txg, int flags)
1126 1127 {
1127 1128 list_t *dl = &spa->spa_config_dirty_list;
1128 1129 vdev_t *vd;
1129 1130 zio_t *zio;
1130 1131 int error;
1131 1132
1132 1133 /*
1133 1134 * Write the new labels to disk.
1134 1135 */
1135 1136 zio = zio_root(spa, NULL, NULL, flags);
1136 1137
1137 1138 for (vd = list_head(dl); vd != NULL; vd = list_next(dl, vd)) {
1138 1139 uint64_t *good_writes = kmem_zalloc(sizeof (uint64_t),
1139 1140 KM_SLEEP);
1140 1141
1141 1142 ASSERT(!vd->vdev_ishole);
1142 1143
1143 1144 zio_t *vio = zio_null(zio, spa, NULL,
1144 1145 (vd->vdev_islog || vd->vdev_aux != NULL) ?
1145 1146 vdev_label_sync_ignore_done : vdev_label_sync_top_done,
1146 1147 good_writes, flags);
1147 1148 vdev_label_sync(vio, vd, l, txg, flags);
1148 1149 zio_nowait(vio);
1149 1150 }
1150 1151
1151 1152 error = zio_wait(zio);
1152 1153
1153 1154 /*
1154 1155 * Flush the new labels to disk.
1155 1156 */
1156 1157 zio = zio_root(spa, NULL, NULL, flags);
1157 1158
1158 1159 for (vd = list_head(dl); vd != NULL; vd = list_next(dl, vd))
1159 1160 zio_flush(zio, vd);
1160 1161
1161 1162 (void) zio_wait(zio);
1162 1163
1163 1164 return (error);
1164 1165 }
1165 1166
1166 1167 /*
1167 1168 * Sync the uberblock and any changes to the vdev configuration.
1168 1169 *
1169 1170 * The order of operations is carefully crafted to ensure that
1170 1171 * if the system panics or loses power at any time, the state on disk
1171 1172 * is still transactionally consistent. The in-line comments below
1172 1173 * describe the failure semantics at each stage.
1173 1174 *
1174 1175 * Moreover, vdev_config_sync() is designed to be idempotent: if it fails
1175 1176 * at any time, you can just call it again, and it will resume its work.
1176 1177 */
1177 1178 int
1178 1179 vdev_config_sync(vdev_t **svd, int svdcount, uint64_t txg, boolean_t tryhard)
1179 1180 {
1180 1181 spa_t *spa = svd[0]->vdev_spa;
1181 1182 uberblock_t *ub = &spa->spa_uberblock;
1182 1183 vdev_t *vd;
1183 1184 zio_t *zio;
1184 1185 int error;
1185 1186 int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL;
1186 1187
1187 1188 /*
1188 1189 * Normally, we don't want to try too hard to write every label and
1189 1190 * uberblock. If there is a flaky disk, we don't want the rest of the
1190 1191 * sync process to block while we retry. But if we can't write a
1191 1192 * single label out, we should retry with ZIO_FLAG_TRYHARD before
1192 1193 * bailing out and declaring the pool faulted.
1193 1194 */
1194 1195 if (tryhard)
1195 1196 flags |= ZIO_FLAG_TRYHARD;
1196 1197
1197 1198 ASSERT(ub->ub_txg <= txg);
1198 1199
1199 1200 /*
1200 1201 * If this isn't a resync due to I/O errors,
1201 1202 * and nothing changed in this transaction group,
1202 1203 * and the vdev configuration hasn't changed,
1203 1204 * then there's nothing to do.
1204 1205 */
1205 1206 if (ub->ub_txg < txg &&
1206 1207 uberblock_update(ub, spa->spa_root_vdev, txg) == B_FALSE &&
1207 1208 list_is_empty(&spa->spa_config_dirty_list))
1208 1209 return (0);
1209 1210
1210 1211 if (txg > spa_freeze_txg(spa))
1211 1212 return (0);
1212 1213
1213 1214 ASSERT(txg <= spa->spa_final_txg);
1214 1215
1215 1216 /*
1216 1217 * Flush the write cache of every disk that's been written to
1217 1218 * in this transaction group. This ensures that all blocks
1218 1219 * written in this txg will be committed to stable storage
1219 1220 * before any uberblock that references them.
1220 1221 */
1221 1222 zio = zio_root(spa, NULL, NULL, flags);
1222 1223
1223 1224 for (vd = txg_list_head(&spa->spa_vdev_txg_list, TXG_CLEAN(txg)); vd;
1224 1225 vd = txg_list_next(&spa->spa_vdev_txg_list, vd, TXG_CLEAN(txg)))
1225 1226 zio_flush(zio, vd);
1226 1227
1227 1228 (void) zio_wait(zio);
1228 1229
1229 1230 /*
1230 1231 * Sync out the even labels (L0, L2) for every dirty vdev. If the
1231 1232 * system dies in the middle of this process, that's OK: all of the
1232 1233 * even labels that made it to disk will be newer than any uberblock,
1233 1234 * and will therefore be considered invalid. The odd labels (L1, L3),
1234 1235 * which have not yet been touched, will still be valid. We flush
1235 1236 * the new labels to disk to ensure that all even-label updates
1236 1237 * are committed to stable storage before the uberblock update.
1237 1238 */
1238 1239 if ((error = vdev_label_sync_list(spa, 0, txg, flags)) != 0)
1239 1240 return (error);
1240 1241
1241 1242 /*
1242 1243 * Sync the uberblocks to all vdevs in svd[].
1243 1244 * If the system dies in the middle of this step, there are two cases
1244 1245 * to consider, and the on-disk state is consistent either way:
1245 1246 *
1246 1247 * (1) If none of the new uberblocks made it to disk, then the
1247 1248 * previous uberblock will be the newest, and the odd labels
1248 1249 * (which had not yet been touched) will be valid with respect
1249 1250 * to that uberblock.
1250 1251 *
1251 1252 * (2) If one or more new uberblocks made it to disk, then they
1252 1253 * will be the newest, and the even labels (which had all
1253 1254 * been successfully committed) will be valid with respect
1254 1255 * to the new uberblocks.
1255 1256 */
1256 1257 if ((error = vdev_uberblock_sync_list(svd, svdcount, ub, flags)) != 0)
1257 1258 return (error);
1258 1259
1259 1260 /*
1260 1261 * Sync out odd labels for every dirty vdev. If the system dies
1261 1262 * in the middle of this process, the even labels and the new
1262 1263 * uberblocks will suffice to open the pool. The next time
1263 1264 * the pool is opened, the first thing we'll do -- before any
1264 1265 * user data is modified -- is mark every vdev dirty so that
1265 1266 * all labels will be brought up to date. We flush the new labels
1266 1267 * to disk to ensure that all odd-label updates are committed to
1267 1268 * stable storage before the next transaction group begins.
1268 1269 */
1269 1270 return (vdev_label_sync_list(spa, 1, txg, flags));
1270 1271 }
|
↓ open down ↓ |
972 lines elided |
↑ open up ↑ |
XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX