Print this page
2619 asynchronous destruction of ZFS file systems
2747 SPA versioning with zfs feature flags
Reviewed by: Matt Ahrens <mahrens@delphix.com>
Reviewed by: George Wilson <gwilson@delphix.com>
Reviewed by: Richard Lowe <richlowe@richlowe.net>
Reviewed by: Dan Kruchinin <dan.kruchinin@gmail.com>
Approved by: Dan McDonald <danmcd@nexenta.com>
| Split |
Close |
| Expand all |
| Collapse all |
--- old/usr/src/uts/common/fs/zfs/spa.c
+++ new/usr/src/uts/common/fs/zfs/spa.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 2011 Nexenta Systems, Inc. All rights reserved.
25 25 * Copyright (c) 2012 by Delphix. All rights reserved.
26 26 */
27 27
28 28 /*
29 29 * This file contains all the routines used when modifying on-disk SPA state.
30 30 * This includes opening, importing, destroying, exporting a pool, and syncing a
31 31 * pool.
32 32 */
33 33
34 34 #include <sys/zfs_context.h>
35 35 #include <sys/fm/fs/zfs.h>
36 36 #include <sys/spa_impl.h>
37 37 #include <sys/zio.h>
38 38 #include <sys/zio_checksum.h>
39 39 #include <sys/dmu.h>
40 40 #include <sys/dmu_tx.h>
41 41 #include <sys/zap.h>
42 42 #include <sys/zil.h>
43 43 #include <sys/ddt.h>
44 44 #include <sys/vdev_impl.h>
45 45 #include <sys/metaslab.h>
46 46 #include <sys/metaslab_impl.h>
47 47 #include <sys/uberblock_impl.h>
48 48 #include <sys/txg.h>
49 49 #include <sys/avl.h>
50 50 #include <sys/dmu_traverse.h>
51 51 #include <sys/dmu_objset.h>
52 52 #include <sys/unique.h>
53 53 #include <sys/dsl_pool.h>
54 54 #include <sys/dsl_dataset.h>
|
↓ open down ↓ |
54 lines elided |
↑ open up ↑ |
55 55 #include <sys/dsl_dir.h>
56 56 #include <sys/dsl_prop.h>
57 57 #include <sys/dsl_synctask.h>
58 58 #include <sys/fs/zfs.h>
59 59 #include <sys/arc.h>
60 60 #include <sys/callb.h>
61 61 #include <sys/systeminfo.h>
62 62 #include <sys/spa_boot.h>
63 63 #include <sys/zfs_ioctl.h>
64 64 #include <sys/dsl_scan.h>
65 +#include <sys/zfeature.h>
65 66
66 67 #ifdef _KERNEL
67 68 #include <sys/bootprops.h>
68 69 #include <sys/callb.h>
69 70 #include <sys/cpupart.h>
70 71 #include <sys/pool.h>
71 72 #include <sys/sysdc.h>
72 73 #include <sys/zone.h>
73 74 #endif /* _KERNEL */
74 75
75 76 #include "zfs_prop.h"
76 77 #include "zfs_comutil.h"
77 78
78 79 typedef enum zti_modes {
79 80 zti_mode_fixed, /* value is # of threads (min 1) */
80 81 zti_mode_online_percent, /* value is % of online CPUs */
81 82 zti_mode_batch, /* cpu-intensive; value is ignored */
82 83 zti_mode_null, /* don't create a taskq */
83 84 zti_nmodes
84 85 } zti_modes_t;
85 86
86 87 #define ZTI_FIX(n) { zti_mode_fixed, (n) }
87 88 #define ZTI_PCT(n) { zti_mode_online_percent, (n) }
88 89 #define ZTI_BATCH { zti_mode_batch, 0 }
89 90 #define ZTI_NULL { zti_mode_null, 0 }
90 91
91 92 #define ZTI_ONE ZTI_FIX(1)
92 93
93 94 typedef struct zio_taskq_info {
94 95 enum zti_modes zti_mode;
95 96 uint_t zti_value;
96 97 } zio_taskq_info_t;
97 98
98 99 static const char *const zio_taskq_types[ZIO_TASKQ_TYPES] = {
99 100 "issue", "issue_high", "intr", "intr_high"
100 101 };
101 102
102 103 /*
103 104 * Define the taskq threads for the following I/O types:
104 105 * NULL, READ, WRITE, FREE, CLAIM, and IOCTL
105 106 */
|
↓ open down ↓ |
31 lines elided |
↑ open up ↑ |
106 107 const zio_taskq_info_t zio_taskqs[ZIO_TYPES][ZIO_TASKQ_TYPES] = {
107 108 /* ISSUE ISSUE_HIGH INTR INTR_HIGH */
108 109 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL },
109 110 { ZTI_FIX(8), ZTI_NULL, ZTI_BATCH, ZTI_NULL },
110 111 { ZTI_BATCH, ZTI_FIX(5), ZTI_FIX(8), ZTI_FIX(5) },
111 112 { ZTI_FIX(100), ZTI_NULL, ZTI_ONE, ZTI_NULL },
112 113 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL },
113 114 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL },
114 115 };
115 116
117 +static dsl_syncfunc_t spa_sync_version;
116 118 static dsl_syncfunc_t spa_sync_props;
117 119 static boolean_t spa_has_active_shared_spare(spa_t *spa);
118 120 static int spa_load_impl(spa_t *spa, uint64_t, nvlist_t *config,
119 121 spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig,
120 122 char **ereport);
121 123 static void spa_vdev_resilver_done(spa_t *spa);
122 124
123 125 uint_t zio_taskq_batch_pct = 100; /* 1 thread per cpu in pset */
124 126 id_t zio_taskq_psrset_bind = PS_NONE;
125 127 boolean_t zio_taskq_sysdc = B_TRUE; /* use SDC scheduling class */
126 128 uint_t zio_taskq_basedc = 80; /* base duty cycle */
127 129
128 130 boolean_t spa_create_process = B_TRUE; /* no process ==> no sysdc */
129 131
130 132 /*
131 133 * This (illegal) pool name is used when temporarily importing a spa_t in order
132 134 * to get the vdev stats associated with the imported devices.
133 135 */
134 136 #define TRYIMPORT_NAME "$import"
135 137
136 138 /*
137 139 * ==========================================================================
138 140 * SPA properties routines
139 141 * ==========================================================================
140 142 */
141 143
142 144 /*
143 145 * Add a (source=src, propname=propval) list to an nvlist.
144 146 */
145 147 static void
146 148 spa_prop_add_list(nvlist_t *nvl, zpool_prop_t prop, char *strval,
147 149 uint64_t intval, zprop_source_t src)
148 150 {
149 151 const char *propname = zpool_prop_to_name(prop);
150 152 nvlist_t *propval;
151 153
152 154 VERIFY(nvlist_alloc(&propval, NV_UNIQUE_NAME, KM_SLEEP) == 0);
153 155 VERIFY(nvlist_add_uint64(propval, ZPROP_SOURCE, src) == 0);
154 156
155 157 if (strval != NULL)
156 158 VERIFY(nvlist_add_string(propval, ZPROP_VALUE, strval) == 0);
157 159 else
158 160 VERIFY(nvlist_add_uint64(propval, ZPROP_VALUE, intval) == 0);
159 161
160 162 VERIFY(nvlist_add_nvlist(nvl, propname, propval) == 0);
|
↓ open down ↓ |
35 lines elided |
↑ open up ↑ |
161 163 nvlist_free(propval);
162 164 }
163 165
164 166 /*
165 167 * Get property values from the spa configuration.
166 168 */
167 169 static void
168 170 spa_prop_get_config(spa_t *spa, nvlist_t **nvp)
169 171 {
170 172 vdev_t *rvd = spa->spa_root_vdev;
173 + dsl_pool_t *pool = spa->spa_dsl_pool;
171 174 uint64_t size;
172 175 uint64_t alloc;
173 176 uint64_t space;
174 177 uint64_t cap, version;
175 178 zprop_source_t src = ZPROP_SRC_NONE;
176 179 spa_config_dirent_t *dp;
177 180
178 181 ASSERT(MUTEX_HELD(&spa->spa_props_lock));
179 182
180 183 if (rvd != NULL) {
181 184 alloc = metaslab_class_get_alloc(spa_normal_class(spa));
182 185 size = metaslab_class_get_space(spa_normal_class(spa));
183 186 spa_prop_add_list(*nvp, ZPOOL_PROP_NAME, spa_name(spa), 0, src);
184 187 spa_prop_add_list(*nvp, ZPOOL_PROP_SIZE, NULL, size, src);
185 188 spa_prop_add_list(*nvp, ZPOOL_PROP_ALLOCATED, NULL, alloc, src);
186 189 spa_prop_add_list(*nvp, ZPOOL_PROP_FREE, NULL,
187 190 size - alloc, src);
188 191
189 192 space = 0;
190 193 for (int c = 0; c < rvd->vdev_children; c++) {
191 194 vdev_t *tvd = rvd->vdev_child[c];
192 195 space += tvd->vdev_max_asize - tvd->vdev_asize;
193 196 }
194 197 spa_prop_add_list(*nvp, ZPOOL_PROP_EXPANDSZ, NULL, space,
195 198 src);
196 199
197 200 spa_prop_add_list(*nvp, ZPOOL_PROP_READONLY, NULL,
198 201 (spa_mode(spa) == FREAD), src);
199 202
200 203 cap = (size == 0) ? 0 : (alloc * 100 / size);
201 204 spa_prop_add_list(*nvp, ZPOOL_PROP_CAPACITY, NULL, cap, src);
202 205
203 206 spa_prop_add_list(*nvp, ZPOOL_PROP_DEDUPRATIO, NULL,
204 207 ddt_get_pool_dedup_ratio(spa), src);
205 208
206 209 spa_prop_add_list(*nvp, ZPOOL_PROP_HEALTH, NULL,
|
↓ open down ↓ |
26 lines elided |
↑ open up ↑ |
207 210 rvd->vdev_state, src);
208 211
209 212 version = spa_version(spa);
210 213 if (version == zpool_prop_default_numeric(ZPOOL_PROP_VERSION))
211 214 src = ZPROP_SRC_DEFAULT;
212 215 else
213 216 src = ZPROP_SRC_LOCAL;
214 217 spa_prop_add_list(*nvp, ZPOOL_PROP_VERSION, NULL, version, src);
215 218 }
216 219
220 + if (pool != NULL) {
221 + dsl_dir_t *freedir = pool->dp_free_dir;
222 +
223 + /*
224 + * The $FREE directory was introduced in SPA_VERSION_DEADLISTS,
225 + * when opening pools before this version freedir will be NULL.
226 + */
227 + if (freedir != NULL) {
228 + spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING, NULL,
229 + freedir->dd_phys->dd_used_bytes, src);
230 + } else {
231 + spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING,
232 + NULL, 0, src);
233 + }
234 + }
235 +
217 236 spa_prop_add_list(*nvp, ZPOOL_PROP_GUID, NULL, spa_guid(spa), src);
218 237
219 238 if (spa->spa_comment != NULL) {
220 239 spa_prop_add_list(*nvp, ZPOOL_PROP_COMMENT, spa->spa_comment,
221 240 0, ZPROP_SRC_LOCAL);
222 241 }
223 242
224 243 if (spa->spa_root != NULL)
225 244 spa_prop_add_list(*nvp, ZPOOL_PROP_ALTROOT, spa->spa_root,
226 245 0, ZPROP_SRC_LOCAL);
227 246
228 247 if ((dp = list_head(&spa->spa_config_list)) != NULL) {
229 248 if (dp->scd_path == NULL) {
230 249 spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
231 250 "none", 0, ZPROP_SRC_LOCAL);
232 251 } else if (strcmp(dp->scd_path, spa_config_path) != 0) {
233 252 spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
234 253 dp->scd_path, 0, ZPROP_SRC_LOCAL);
235 254 }
236 255 }
237 256 }
238 257
239 258 /*
240 259 * Get zpool property values.
241 260 */
242 261 int
243 262 spa_prop_get(spa_t *spa, nvlist_t **nvp)
244 263 {
245 264 objset_t *mos = spa->spa_meta_objset;
246 265 zap_cursor_t zc;
247 266 zap_attribute_t za;
248 267 int err;
249 268
250 269 VERIFY(nvlist_alloc(nvp, NV_UNIQUE_NAME, KM_SLEEP) == 0);
251 270
252 271 mutex_enter(&spa->spa_props_lock);
253 272
254 273 /*
255 274 * Get properties from the spa config.
256 275 */
257 276 spa_prop_get_config(spa, nvp);
258 277
259 278 /* If no pool property object, no more prop to get. */
260 279 if (mos == NULL || spa->spa_pool_props_object == 0) {
261 280 mutex_exit(&spa->spa_props_lock);
262 281 return (0);
263 282 }
264 283
265 284 /*
266 285 * Get properties from the MOS pool property object.
267 286 */
268 287 for (zap_cursor_init(&zc, mos, spa->spa_pool_props_object);
269 288 (err = zap_cursor_retrieve(&zc, &za)) == 0;
270 289 zap_cursor_advance(&zc)) {
271 290 uint64_t intval = 0;
272 291 char *strval = NULL;
273 292 zprop_source_t src = ZPROP_SRC_DEFAULT;
274 293 zpool_prop_t prop;
275 294
276 295 if ((prop = zpool_name_to_prop(za.za_name)) == ZPROP_INVAL)
277 296 continue;
278 297
279 298 switch (za.za_integer_length) {
280 299 case 8:
281 300 /* integer property */
282 301 if (za.za_first_integer !=
283 302 zpool_prop_default_numeric(prop))
284 303 src = ZPROP_SRC_LOCAL;
285 304
286 305 if (prop == ZPOOL_PROP_BOOTFS) {
287 306 dsl_pool_t *dp;
288 307 dsl_dataset_t *ds = NULL;
289 308
290 309 dp = spa_get_dsl(spa);
291 310 rw_enter(&dp->dp_config_rwlock, RW_READER);
292 311 if (err = dsl_dataset_hold_obj(dp,
293 312 za.za_first_integer, FTAG, &ds)) {
294 313 rw_exit(&dp->dp_config_rwlock);
295 314 break;
296 315 }
297 316
298 317 strval = kmem_alloc(
299 318 MAXNAMELEN + strlen(MOS_DIR_NAME) + 1,
300 319 KM_SLEEP);
301 320 dsl_dataset_name(ds, strval);
302 321 dsl_dataset_rele(ds, FTAG);
303 322 rw_exit(&dp->dp_config_rwlock);
304 323 } else {
305 324 strval = NULL;
306 325 intval = za.za_first_integer;
307 326 }
308 327
309 328 spa_prop_add_list(*nvp, prop, strval, intval, src);
310 329
311 330 if (strval != NULL)
312 331 kmem_free(strval,
313 332 MAXNAMELEN + strlen(MOS_DIR_NAME) + 1);
314 333
315 334 break;
316 335
317 336 case 1:
318 337 /* string property */
319 338 strval = kmem_alloc(za.za_num_integers, KM_SLEEP);
320 339 err = zap_lookup(mos, spa->spa_pool_props_object,
321 340 za.za_name, 1, za.za_num_integers, strval);
322 341 if (err) {
323 342 kmem_free(strval, za.za_num_integers);
324 343 break;
325 344 }
326 345 spa_prop_add_list(*nvp, prop, strval, 0, src);
327 346 kmem_free(strval, za.za_num_integers);
328 347 break;
329 348
330 349 default:
331 350 break;
332 351 }
333 352 }
334 353 zap_cursor_fini(&zc);
335 354 mutex_exit(&spa->spa_props_lock);
336 355 out:
337 356 if (err && err != ENOENT) {
338 357 nvlist_free(*nvp);
339 358 *nvp = NULL;
340 359 return (err);
341 360 }
342 361
343 362 return (0);
344 363 }
345 364
|
↓ open down ↓ |
119 lines elided |
↑ open up ↑ |
346 365 /*
347 366 * Validate the given pool properties nvlist and modify the list
348 367 * for the property values to be set.
349 368 */
350 369 static int
351 370 spa_prop_validate(spa_t *spa, nvlist_t *props)
352 371 {
353 372 nvpair_t *elem;
354 373 int error = 0, reset_bootfs = 0;
355 374 uint64_t objnum;
375 + boolean_t has_feature = B_FALSE;
356 376
357 377 elem = NULL;
358 378 while ((elem = nvlist_next_nvpair(props, elem)) != NULL) {
359 - zpool_prop_t prop;
360 - char *propname, *strval;
361 379 uint64_t intval;
362 - objset_t *os;
363 - char *slash, *check;
380 + char *strval, *slash, *check, *fname;
381 + const char *propname = nvpair_name(elem);
382 + zpool_prop_t prop = zpool_name_to_prop(propname);
364 383
365 - propname = nvpair_name(elem);
384 + switch (prop) {
385 + case ZPROP_INVAL:
386 + if (!zpool_prop_feature(propname)) {
387 + error = EINVAL;
388 + break;
389 + }
366 390
367 - if ((prop = zpool_name_to_prop(propname)) == ZPROP_INVAL)
368 - return (EINVAL);
391 + /*
392 + * Sanitize the input.
393 + */
394 + if (nvpair_type(elem) != DATA_TYPE_UINT64) {
395 + error = EINVAL;
396 + break;
397 + }
369 398
370 - switch (prop) {
399 + if (nvpair_value_uint64(elem, &intval) != 0) {
400 + error = EINVAL;
401 + break;
402 + }
403 +
404 + if (intval != 0) {
405 + error = EINVAL;
406 + break;
407 + }
408 +
409 + fname = strchr(propname, '@') + 1;
410 + if (zfeature_lookup_name(fname, NULL) != 0) {
411 + error = EINVAL;
412 + break;
413 + }
414 +
415 + has_feature = B_TRUE;
416 + break;
417 +
371 418 case ZPOOL_PROP_VERSION:
372 419 error = nvpair_value_uint64(elem, &intval);
373 420 if (!error &&
374 - (intval < spa_version(spa) || intval > SPA_VERSION))
421 + (intval < spa_version(spa) ||
422 + intval > SPA_VERSION_BEFORE_FEATURES ||
423 + has_feature))
375 424 error = EINVAL;
376 425 break;
377 426
378 427 case ZPOOL_PROP_DELEGATION:
379 428 case ZPOOL_PROP_AUTOREPLACE:
380 429 case ZPOOL_PROP_LISTSNAPS:
381 430 case ZPOOL_PROP_AUTOEXPAND:
382 431 error = nvpair_value_uint64(elem, &intval);
383 432 if (!error && intval > 1)
384 433 error = EINVAL;
385 434 break;
386 435
387 436 case ZPOOL_PROP_BOOTFS:
388 437 /*
389 438 * If the pool version is less than SPA_VERSION_BOOTFS,
390 439 * or the pool is still being created (version == 0),
391 440 * the bootfs property cannot be set.
392 441 */
393 442 if (spa_version(spa) < SPA_VERSION_BOOTFS) {
394 443 error = ENOTSUP;
395 444 break;
396 445 }
397 446
398 447 /*
399 448 * Make sure the vdev config is bootable
400 449 */
|
↓ open down ↓ |
16 lines elided |
↑ open up ↑ |
401 450 if (!vdev_is_bootable(spa->spa_root_vdev)) {
402 451 error = ENOTSUP;
403 452 break;
404 453 }
405 454
406 455 reset_bootfs = 1;
407 456
408 457 error = nvpair_value_string(elem, &strval);
409 458
410 459 if (!error) {
460 + objset_t *os;
411 461 uint64_t compress;
412 462
413 463 if (strval == NULL || strval[0] == '\0') {
414 464 objnum = zpool_prop_default_numeric(
415 465 ZPOOL_PROP_BOOTFS);
416 466 break;
417 467 }
418 468
419 469 if (error = dmu_objset_hold(strval, FTAG, &os))
420 470 break;
421 471
422 472 /* Must be ZPL and not gzip compressed. */
423 473
424 474 if (dmu_objset_type(os) != DMU_OST_ZFS) {
425 475 error = ENOTSUP;
426 476 } else if ((error = dsl_prop_get_integer(strval,
427 477 zfs_prop_to_name(ZFS_PROP_COMPRESSION),
428 478 &compress, NULL)) == 0 &&
429 479 !BOOTFS_COMPRESS_VALID(compress)) {
430 480 error = ENOTSUP;
431 481 } else {
432 482 objnum = dmu_objset_id(os);
433 483 }
434 484 dmu_objset_rele(os, FTAG);
435 485 }
436 486 break;
437 487
438 488 case ZPOOL_PROP_FAILUREMODE:
439 489 error = nvpair_value_uint64(elem, &intval);
440 490 if (!error && (intval < ZIO_FAILURE_MODE_WAIT ||
441 491 intval > ZIO_FAILURE_MODE_PANIC))
442 492 error = EINVAL;
443 493
444 494 /*
445 495 * This is a special case which only occurs when
446 496 * the pool has completely failed. This allows
447 497 * the user to change the in-core failmode property
448 498 * without syncing it out to disk (I/Os might
449 499 * currently be blocked). We do this by returning
450 500 * EIO to the caller (spa_prop_set) to trick it
451 501 * into thinking we encountered a property validation
452 502 * error.
453 503 */
454 504 if (!error && spa_suspended(spa)) {
455 505 spa->spa_failmode = intval;
456 506 error = EIO;
457 507 }
458 508 break;
459 509
460 510 case ZPOOL_PROP_CACHEFILE:
461 511 if ((error = nvpair_value_string(elem, &strval)) != 0)
462 512 break;
463 513
464 514 if (strval[0] == '\0')
465 515 break;
466 516
467 517 if (strcmp(strval, "none") == 0)
468 518 break;
469 519
470 520 if (strval[0] != '/') {
471 521 error = EINVAL;
472 522 break;
473 523 }
474 524
475 525 slash = strrchr(strval, '/');
476 526 ASSERT(slash != NULL);
477 527
478 528 if (slash[1] == '\0' || strcmp(slash, "/.") == 0 ||
479 529 strcmp(slash, "/..") == 0)
480 530 error = EINVAL;
481 531 break;
482 532
483 533 case ZPOOL_PROP_COMMENT:
484 534 if ((error = nvpair_value_string(elem, &strval)) != 0)
485 535 break;
486 536 for (check = strval; *check != '\0'; check++) {
487 537 /*
488 538 * The kernel doesn't have an easy isprint()
489 539 * check. For this kernel check, we merely
490 540 * check ASCII apart from DEL. Fix this if
491 541 * there is an easy-to-use kernel isprint().
492 542 */
493 543 if (*check >= 0x7f) {
494 544 error = EINVAL;
495 545 break;
496 546 }
497 547 check++;
498 548 }
499 549 if (strlen(strval) > ZPROP_MAX_COMMENT)
500 550 error = E2BIG;
501 551 break;
502 552
503 553 case ZPOOL_PROP_DEDUPDITTO:
504 554 if (spa_version(spa) < SPA_VERSION_DEDUP)
505 555 error = ENOTSUP;
506 556 else
507 557 error = nvpair_value_uint64(elem, &intval);
508 558 if (error == 0 &&
509 559 intval != 0 && intval < ZIO_DEDUPDITTO_MIN)
510 560 error = EINVAL;
511 561 break;
512 562 }
513 563
514 564 if (error)
515 565 break;
516 566 }
517 567
518 568 if (!error && reset_bootfs) {
519 569 error = nvlist_remove(props,
520 570 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), DATA_TYPE_STRING);
521 571
522 572 if (!error) {
523 573 error = nvlist_add_uint64(props,
524 574 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), objnum);
525 575 }
526 576 }
527 577
528 578 return (error);
529 579 }
530 580
531 581 void
532 582 spa_configfile_set(spa_t *spa, nvlist_t *nvp, boolean_t need_sync)
533 583 {
534 584 char *cachefile;
535 585 spa_config_dirent_t *dp;
536 586
537 587 if (nvlist_lookup_string(nvp, zpool_prop_to_name(ZPOOL_PROP_CACHEFILE),
538 588 &cachefile) != 0)
539 589 return;
540 590
541 591 dp = kmem_alloc(sizeof (spa_config_dirent_t),
542 592 KM_SLEEP);
543 593
544 594 if (cachefile[0] == '\0')
545 595 dp->scd_path = spa_strdup(spa_config_path);
546 596 else if (strcmp(cachefile, "none") == 0)
547 597 dp->scd_path = NULL;
548 598 else
549 599 dp->scd_path = spa_strdup(cachefile);
|
↓ open down ↓ |
129 lines elided |
↑ open up ↑ |
550 600
551 601 list_insert_head(&spa->spa_config_list, dp);
552 602 if (need_sync)
553 603 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
554 604 }
555 605
556 606 int
557 607 spa_prop_set(spa_t *spa, nvlist_t *nvp)
558 608 {
559 609 int error;
560 - nvpair_t *elem;
610 + nvpair_t *elem = NULL;
561 611 boolean_t need_sync = B_FALSE;
562 - zpool_prop_t prop;
563 612
564 613 if ((error = spa_prop_validate(spa, nvp)) != 0)
565 614 return (error);
566 615
567 - elem = NULL;
568 616 while ((elem = nvlist_next_nvpair(nvp, elem)) != NULL) {
569 - if ((prop = zpool_name_to_prop(
570 - nvpair_name(elem))) == ZPROP_INVAL)
571 - return (EINVAL);
617 + zpool_prop_t prop = zpool_name_to_prop(nvpair_name(elem));
572 618
573 619 if (prop == ZPOOL_PROP_CACHEFILE ||
574 620 prop == ZPOOL_PROP_ALTROOT ||
575 621 prop == ZPOOL_PROP_READONLY)
576 622 continue;
577 623
624 + if (prop == ZPOOL_PROP_VERSION || prop == ZPROP_INVAL) {
625 + uint64_t ver;
626 +
627 + if (prop == ZPOOL_PROP_VERSION) {
628 + VERIFY(nvpair_value_uint64(elem, &ver) == 0);
629 + } else {
630 + ASSERT(zpool_prop_feature(nvpair_name(elem)));
631 + ver = SPA_VERSION_FEATURES;
632 + need_sync = B_TRUE;
633 + }
634 +
635 + /* Save time if the version is already set. */
636 + if (ver == spa_version(spa))
637 + continue;
638 +
639 + /*
640 + * In addition to the pool directory object, we might
641 + * create the pool properties object, the features for
642 + * read object, the features for write object, or the
643 + * feature descriptions object.
644 + */
645 + error = dsl_sync_task_do(spa_get_dsl(spa), NULL,
646 + spa_sync_version, spa, &ver, 6);
647 + if (error)
648 + return (error);
649 + continue;
650 + }
651 +
578 652 need_sync = B_TRUE;
579 653 break;
580 654 }
581 655
582 - if (need_sync)
656 + if (need_sync) {
583 657 return (dsl_sync_task_do(spa_get_dsl(spa), NULL, spa_sync_props,
584 - spa, nvp, 3));
585 - else
586 - return (0);
658 + spa, nvp, 6));
659 + }
660 +
661 + return (0);
587 662 }
588 663
589 664 /*
590 665 * If the bootfs property value is dsobj, clear it.
591 666 */
592 667 void
593 668 spa_prop_clear_bootfs(spa_t *spa, uint64_t dsobj, dmu_tx_t *tx)
594 669 {
595 670 if (spa->spa_bootfs == dsobj && spa->spa_pool_props_object != 0) {
596 671 VERIFY(zap_remove(spa->spa_meta_objset,
597 672 spa->spa_pool_props_object,
598 673 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), tx) == 0);
599 674 spa->spa_bootfs = 0;
600 675 }
601 676 }
602 677
603 678 /*
604 679 * Change the GUID for the pool. This is done so that we can later
605 680 * re-import a pool built from a clone of our own vdevs. We will modify
606 681 * the root vdev's guid, our own pool guid, and then mark all of our
607 682 * vdevs dirty. Note that we must make sure that all our vdevs are
608 683 * online when we do this, or else any vdevs that weren't present
609 684 * would be orphaned from our pool. We are also going to issue a
610 685 * sysevent to update any watchers.
611 686 */
612 687 int
613 688 spa_change_guid(spa_t *spa)
614 689 {
615 690 uint64_t oldguid, newguid;
616 691 uint64_t txg;
617 692
618 693 if (!(spa_mode_global & FWRITE))
619 694 return (EROFS);
620 695
621 696 txg = spa_vdev_enter(spa);
622 697
623 698 if (spa->spa_root_vdev->vdev_state != VDEV_STATE_HEALTHY)
624 699 return (spa_vdev_exit(spa, NULL, txg, ENXIO));
625 700
626 701 oldguid = spa_guid(spa);
627 702 newguid = spa_generate_guid(NULL);
628 703 ASSERT3U(oldguid, !=, newguid);
629 704
630 705 spa->spa_root_vdev->vdev_guid = newguid;
631 706 spa->spa_root_vdev->vdev_guid_sum += (newguid - oldguid);
632 707
633 708 vdev_config_dirty(spa->spa_root_vdev);
634 709
635 710 spa_event_notify(spa, NULL, ESC_ZFS_POOL_REGUID);
636 711
637 712 return (spa_vdev_exit(spa, NULL, txg, 0));
638 713 }
639 714
640 715 /*
641 716 * ==========================================================================
642 717 * SPA state manipulation (open/create/destroy/import/export)
643 718 * ==========================================================================
644 719 */
645 720
646 721 static int
647 722 spa_error_entry_compare(const void *a, const void *b)
648 723 {
649 724 spa_error_entry_t *sa = (spa_error_entry_t *)a;
650 725 spa_error_entry_t *sb = (spa_error_entry_t *)b;
651 726 int ret;
652 727
653 728 ret = bcmp(&sa->se_bookmark, &sb->se_bookmark,
654 729 sizeof (zbookmark_t));
655 730
656 731 if (ret < 0)
657 732 return (-1);
658 733 else if (ret > 0)
659 734 return (1);
660 735 else
661 736 return (0);
662 737 }
663 738
664 739 /*
665 740 * Utility function which retrieves copies of the current logs and
666 741 * re-initializes them in the process.
667 742 */
668 743 void
669 744 spa_get_errlists(spa_t *spa, avl_tree_t *last, avl_tree_t *scrub)
670 745 {
671 746 ASSERT(MUTEX_HELD(&spa->spa_errlist_lock));
672 747
673 748 bcopy(&spa->spa_errlist_last, last, sizeof (avl_tree_t));
674 749 bcopy(&spa->spa_errlist_scrub, scrub, sizeof (avl_tree_t));
675 750
676 751 avl_create(&spa->spa_errlist_scrub,
677 752 spa_error_entry_compare, sizeof (spa_error_entry_t),
678 753 offsetof(spa_error_entry_t, se_avl));
679 754 avl_create(&spa->spa_errlist_last,
680 755 spa_error_entry_compare, sizeof (spa_error_entry_t),
681 756 offsetof(spa_error_entry_t, se_avl));
682 757 }
683 758
684 759 static taskq_t *
685 760 spa_taskq_create(spa_t *spa, const char *name, enum zti_modes mode,
686 761 uint_t value)
687 762 {
688 763 uint_t flags = 0;
689 764 boolean_t batch = B_FALSE;
690 765
691 766 switch (mode) {
692 767 case zti_mode_null:
693 768 return (NULL); /* no taskq needed */
694 769
695 770 case zti_mode_fixed:
696 771 ASSERT3U(value, >=, 1);
697 772 value = MAX(value, 1);
698 773 break;
699 774
700 775 case zti_mode_batch:
701 776 batch = B_TRUE;
702 777 flags |= TASKQ_THREADS_CPU_PCT;
703 778 value = zio_taskq_batch_pct;
704 779 break;
705 780
706 781 case zti_mode_online_percent:
707 782 flags |= TASKQ_THREADS_CPU_PCT;
708 783 break;
709 784
710 785 default:
711 786 panic("unrecognized mode for %s taskq (%u:%u) in "
712 787 "spa_activate()",
713 788 name, mode, value);
714 789 break;
715 790 }
716 791
717 792 if (zio_taskq_sysdc && spa->spa_proc != &p0) {
718 793 if (batch)
719 794 flags |= TASKQ_DC_BATCH;
720 795
721 796 return (taskq_create_sysdc(name, value, 50, INT_MAX,
722 797 spa->spa_proc, zio_taskq_basedc, flags));
723 798 }
724 799 return (taskq_create_proc(name, value, maxclsyspri, 50, INT_MAX,
725 800 spa->spa_proc, flags));
726 801 }
727 802
728 803 static void
729 804 spa_create_zio_taskqs(spa_t *spa)
730 805 {
731 806 for (int t = 0; t < ZIO_TYPES; t++) {
732 807 for (int q = 0; q < ZIO_TASKQ_TYPES; q++) {
733 808 const zio_taskq_info_t *ztip = &zio_taskqs[t][q];
734 809 enum zti_modes mode = ztip->zti_mode;
735 810 uint_t value = ztip->zti_value;
736 811 char name[32];
737 812
738 813 (void) snprintf(name, sizeof (name),
739 814 "%s_%s", zio_type_name[t], zio_taskq_types[q]);
740 815
741 816 spa->spa_zio_taskq[t][q] =
742 817 spa_taskq_create(spa, name, mode, value);
743 818 }
744 819 }
745 820 }
746 821
747 822 #ifdef _KERNEL
748 823 static void
749 824 spa_thread(void *arg)
750 825 {
751 826 callb_cpr_t cprinfo;
752 827
753 828 spa_t *spa = arg;
754 829 user_t *pu = PTOU(curproc);
755 830
756 831 CALLB_CPR_INIT(&cprinfo, &spa->spa_proc_lock, callb_generic_cpr,
757 832 spa->spa_name);
758 833
759 834 ASSERT(curproc != &p0);
760 835 (void) snprintf(pu->u_psargs, sizeof (pu->u_psargs),
761 836 "zpool-%s", spa->spa_name);
762 837 (void) strlcpy(pu->u_comm, pu->u_psargs, sizeof (pu->u_comm));
763 838
764 839 /* bind this thread to the requested psrset */
765 840 if (zio_taskq_psrset_bind != PS_NONE) {
766 841 pool_lock();
767 842 mutex_enter(&cpu_lock);
768 843 mutex_enter(&pidlock);
769 844 mutex_enter(&curproc->p_lock);
770 845
771 846 if (cpupart_bind_thread(curthread, zio_taskq_psrset_bind,
772 847 0, NULL, NULL) == 0) {
773 848 curthread->t_bind_pset = zio_taskq_psrset_bind;
774 849 } else {
775 850 cmn_err(CE_WARN,
776 851 "Couldn't bind process for zfs pool \"%s\" to "
777 852 "pset %d\n", spa->spa_name, zio_taskq_psrset_bind);
778 853 }
779 854
780 855 mutex_exit(&curproc->p_lock);
781 856 mutex_exit(&pidlock);
782 857 mutex_exit(&cpu_lock);
783 858 pool_unlock();
784 859 }
785 860
786 861 if (zio_taskq_sysdc) {
787 862 sysdc_thread_enter(curthread, 100, 0);
788 863 }
789 864
790 865 spa->spa_proc = curproc;
791 866 spa->spa_did = curthread->t_did;
792 867
793 868 spa_create_zio_taskqs(spa);
794 869
795 870 mutex_enter(&spa->spa_proc_lock);
796 871 ASSERT(spa->spa_proc_state == SPA_PROC_CREATED);
797 872
798 873 spa->spa_proc_state = SPA_PROC_ACTIVE;
799 874 cv_broadcast(&spa->spa_proc_cv);
800 875
801 876 CALLB_CPR_SAFE_BEGIN(&cprinfo);
802 877 while (spa->spa_proc_state == SPA_PROC_ACTIVE)
803 878 cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock);
804 879 CALLB_CPR_SAFE_END(&cprinfo, &spa->spa_proc_lock);
805 880
806 881 ASSERT(spa->spa_proc_state == SPA_PROC_DEACTIVATE);
807 882 spa->spa_proc_state = SPA_PROC_GONE;
808 883 spa->spa_proc = &p0;
809 884 cv_broadcast(&spa->spa_proc_cv);
810 885 CALLB_CPR_EXIT(&cprinfo); /* drops spa_proc_lock */
811 886
812 887 mutex_enter(&curproc->p_lock);
813 888 lwp_exit();
814 889 }
815 890 #endif
816 891
817 892 /*
818 893 * Activate an uninitialized pool.
819 894 */
820 895 static void
821 896 spa_activate(spa_t *spa, int mode)
822 897 {
823 898 ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED);
824 899
825 900 spa->spa_state = POOL_STATE_ACTIVE;
826 901 spa->spa_mode = mode;
827 902
828 903 spa->spa_normal_class = metaslab_class_create(spa, zfs_metaslab_ops);
829 904 spa->spa_log_class = metaslab_class_create(spa, zfs_metaslab_ops);
830 905
831 906 /* Try to create a covering process */
832 907 mutex_enter(&spa->spa_proc_lock);
833 908 ASSERT(spa->spa_proc_state == SPA_PROC_NONE);
834 909 ASSERT(spa->spa_proc == &p0);
835 910 spa->spa_did = 0;
836 911
837 912 /* Only create a process if we're going to be around a while. */
838 913 if (spa_create_process && strcmp(spa->spa_name, TRYIMPORT_NAME) != 0) {
839 914 if (newproc(spa_thread, (caddr_t)spa, syscid, maxclsyspri,
840 915 NULL, 0) == 0) {
841 916 spa->spa_proc_state = SPA_PROC_CREATED;
842 917 while (spa->spa_proc_state == SPA_PROC_CREATED) {
843 918 cv_wait(&spa->spa_proc_cv,
844 919 &spa->spa_proc_lock);
845 920 }
846 921 ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE);
847 922 ASSERT(spa->spa_proc != &p0);
848 923 ASSERT(spa->spa_did != 0);
849 924 } else {
850 925 #ifdef _KERNEL
851 926 cmn_err(CE_WARN,
852 927 "Couldn't create process for zfs pool \"%s\"\n",
853 928 spa->spa_name);
854 929 #endif
855 930 }
856 931 }
857 932 mutex_exit(&spa->spa_proc_lock);
858 933
859 934 /* If we didn't create a process, we need to create our taskqs. */
860 935 if (spa->spa_proc == &p0) {
861 936 spa_create_zio_taskqs(spa);
862 937 }
863 938
864 939 list_create(&spa->spa_config_dirty_list, sizeof (vdev_t),
865 940 offsetof(vdev_t, vdev_config_dirty_node));
866 941 list_create(&spa->spa_state_dirty_list, sizeof (vdev_t),
867 942 offsetof(vdev_t, vdev_state_dirty_node));
868 943
869 944 txg_list_create(&spa->spa_vdev_txg_list,
870 945 offsetof(struct vdev, vdev_txg_node));
871 946
872 947 avl_create(&spa->spa_errlist_scrub,
873 948 spa_error_entry_compare, sizeof (spa_error_entry_t),
874 949 offsetof(spa_error_entry_t, se_avl));
875 950 avl_create(&spa->spa_errlist_last,
876 951 spa_error_entry_compare, sizeof (spa_error_entry_t),
877 952 offsetof(spa_error_entry_t, se_avl));
878 953 }
879 954
880 955 /*
881 956 * Opposite of spa_activate().
882 957 */
883 958 static void
884 959 spa_deactivate(spa_t *spa)
885 960 {
886 961 ASSERT(spa->spa_sync_on == B_FALSE);
887 962 ASSERT(spa->spa_dsl_pool == NULL);
888 963 ASSERT(spa->spa_root_vdev == NULL);
889 964 ASSERT(spa->spa_async_zio_root == NULL);
890 965 ASSERT(spa->spa_state != POOL_STATE_UNINITIALIZED);
891 966
892 967 txg_list_destroy(&spa->spa_vdev_txg_list);
893 968
894 969 list_destroy(&spa->spa_config_dirty_list);
895 970 list_destroy(&spa->spa_state_dirty_list);
896 971
897 972 for (int t = 0; t < ZIO_TYPES; t++) {
898 973 for (int q = 0; q < ZIO_TASKQ_TYPES; q++) {
899 974 if (spa->spa_zio_taskq[t][q] != NULL)
900 975 taskq_destroy(spa->spa_zio_taskq[t][q]);
901 976 spa->spa_zio_taskq[t][q] = NULL;
902 977 }
903 978 }
904 979
905 980 metaslab_class_destroy(spa->spa_normal_class);
906 981 spa->spa_normal_class = NULL;
907 982
908 983 metaslab_class_destroy(spa->spa_log_class);
909 984 spa->spa_log_class = NULL;
910 985
911 986 /*
912 987 * If this was part of an import or the open otherwise failed, we may
913 988 * still have errors left in the queues. Empty them just in case.
914 989 */
915 990 spa_errlog_drain(spa);
916 991
917 992 avl_destroy(&spa->spa_errlist_scrub);
918 993 avl_destroy(&spa->spa_errlist_last);
919 994
920 995 spa->spa_state = POOL_STATE_UNINITIALIZED;
921 996
922 997 mutex_enter(&spa->spa_proc_lock);
923 998 if (spa->spa_proc_state != SPA_PROC_NONE) {
924 999 ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE);
925 1000 spa->spa_proc_state = SPA_PROC_DEACTIVATE;
926 1001 cv_broadcast(&spa->spa_proc_cv);
927 1002 while (spa->spa_proc_state == SPA_PROC_DEACTIVATE) {
928 1003 ASSERT(spa->spa_proc != &p0);
929 1004 cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock);
930 1005 }
931 1006 ASSERT(spa->spa_proc_state == SPA_PROC_GONE);
932 1007 spa->spa_proc_state = SPA_PROC_NONE;
933 1008 }
934 1009 ASSERT(spa->spa_proc == &p0);
935 1010 mutex_exit(&spa->spa_proc_lock);
936 1011
937 1012 /*
938 1013 * We want to make sure spa_thread() has actually exited the ZFS
939 1014 * module, so that the module can't be unloaded out from underneath
940 1015 * it.
941 1016 */
942 1017 if (spa->spa_did != 0) {
943 1018 thread_join(spa->spa_did);
944 1019 spa->spa_did = 0;
945 1020 }
946 1021 }
947 1022
948 1023 /*
949 1024 * Verify a pool configuration, and construct the vdev tree appropriately. This
950 1025 * will create all the necessary vdevs in the appropriate layout, with each vdev
951 1026 * in the CLOSED state. This will prep the pool before open/creation/import.
952 1027 * All vdev validation is done by the vdev_alloc() routine.
953 1028 */
954 1029 static int
955 1030 spa_config_parse(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent,
956 1031 uint_t id, int atype)
957 1032 {
958 1033 nvlist_t **child;
959 1034 uint_t children;
960 1035 int error;
961 1036
962 1037 if ((error = vdev_alloc(spa, vdp, nv, parent, id, atype)) != 0)
963 1038 return (error);
964 1039
965 1040 if ((*vdp)->vdev_ops->vdev_op_leaf)
966 1041 return (0);
967 1042
968 1043 error = nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
969 1044 &child, &children);
970 1045
971 1046 if (error == ENOENT)
972 1047 return (0);
973 1048
974 1049 if (error) {
975 1050 vdev_free(*vdp);
976 1051 *vdp = NULL;
977 1052 return (EINVAL);
978 1053 }
979 1054
980 1055 for (int c = 0; c < children; c++) {
981 1056 vdev_t *vd;
982 1057 if ((error = spa_config_parse(spa, &vd, child[c], *vdp, c,
983 1058 atype)) != 0) {
984 1059 vdev_free(*vdp);
985 1060 *vdp = NULL;
986 1061 return (error);
987 1062 }
988 1063 }
989 1064
990 1065 ASSERT(*vdp != NULL);
991 1066
992 1067 return (0);
993 1068 }
994 1069
995 1070 /*
996 1071 * Opposite of spa_load().
997 1072 */
998 1073 static void
999 1074 spa_unload(spa_t *spa)
1000 1075 {
1001 1076 int i;
1002 1077
1003 1078 ASSERT(MUTEX_HELD(&spa_namespace_lock));
1004 1079
1005 1080 /*
1006 1081 * Stop async tasks.
1007 1082 */
1008 1083 spa_async_suspend(spa);
1009 1084
1010 1085 /*
1011 1086 * Stop syncing.
1012 1087 */
1013 1088 if (spa->spa_sync_on) {
1014 1089 txg_sync_stop(spa->spa_dsl_pool);
1015 1090 spa->spa_sync_on = B_FALSE;
1016 1091 }
1017 1092
1018 1093 /*
1019 1094 * Wait for any outstanding async I/O to complete.
1020 1095 */
1021 1096 if (spa->spa_async_zio_root != NULL) {
1022 1097 (void) zio_wait(spa->spa_async_zio_root);
1023 1098 spa->spa_async_zio_root = NULL;
1024 1099 }
1025 1100
1026 1101 bpobj_close(&spa->spa_deferred_bpobj);
1027 1102
1028 1103 /*
1029 1104 * Close the dsl pool.
1030 1105 */
1031 1106 if (spa->spa_dsl_pool) {
1032 1107 dsl_pool_close(spa->spa_dsl_pool);
1033 1108 spa->spa_dsl_pool = NULL;
1034 1109 spa->spa_meta_objset = NULL;
1035 1110 }
1036 1111
1037 1112 ddt_unload(spa);
1038 1113
1039 1114 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1040 1115
1041 1116 /*
1042 1117 * Drop and purge level 2 cache
1043 1118 */
1044 1119 spa_l2cache_drop(spa);
1045 1120
1046 1121 /*
1047 1122 * Close all vdevs.
1048 1123 */
1049 1124 if (spa->spa_root_vdev)
1050 1125 vdev_free(spa->spa_root_vdev);
1051 1126 ASSERT(spa->spa_root_vdev == NULL);
1052 1127
1053 1128 for (i = 0; i < spa->spa_spares.sav_count; i++)
1054 1129 vdev_free(spa->spa_spares.sav_vdevs[i]);
1055 1130 if (spa->spa_spares.sav_vdevs) {
1056 1131 kmem_free(spa->spa_spares.sav_vdevs,
1057 1132 spa->spa_spares.sav_count * sizeof (void *));
1058 1133 spa->spa_spares.sav_vdevs = NULL;
1059 1134 }
1060 1135 if (spa->spa_spares.sav_config) {
1061 1136 nvlist_free(spa->spa_spares.sav_config);
1062 1137 spa->spa_spares.sav_config = NULL;
1063 1138 }
1064 1139 spa->spa_spares.sav_count = 0;
1065 1140
1066 1141 for (i = 0; i < spa->spa_l2cache.sav_count; i++) {
1067 1142 vdev_clear_stats(spa->spa_l2cache.sav_vdevs[i]);
1068 1143 vdev_free(spa->spa_l2cache.sav_vdevs[i]);
1069 1144 }
1070 1145 if (spa->spa_l2cache.sav_vdevs) {
1071 1146 kmem_free(spa->spa_l2cache.sav_vdevs,
1072 1147 spa->spa_l2cache.sav_count * sizeof (void *));
1073 1148 spa->spa_l2cache.sav_vdevs = NULL;
1074 1149 }
1075 1150 if (spa->spa_l2cache.sav_config) {
1076 1151 nvlist_free(spa->spa_l2cache.sav_config);
1077 1152 spa->spa_l2cache.sav_config = NULL;
1078 1153 }
1079 1154 spa->spa_l2cache.sav_count = 0;
1080 1155
1081 1156 spa->spa_async_suspended = 0;
1082 1157
1083 1158 if (spa->spa_comment != NULL) {
1084 1159 spa_strfree(spa->spa_comment);
1085 1160 spa->spa_comment = NULL;
1086 1161 }
1087 1162
1088 1163 spa_config_exit(spa, SCL_ALL, FTAG);
1089 1164 }
1090 1165
1091 1166 /*
1092 1167 * Load (or re-load) the current list of vdevs describing the active spares for
1093 1168 * this pool. When this is called, we have some form of basic information in
1094 1169 * 'spa_spares.sav_config'. We parse this into vdevs, try to open them, and
1095 1170 * then re-generate a more complete list including status information.
1096 1171 */
1097 1172 static void
1098 1173 spa_load_spares(spa_t *spa)
1099 1174 {
1100 1175 nvlist_t **spares;
1101 1176 uint_t nspares;
1102 1177 int i;
1103 1178 vdev_t *vd, *tvd;
1104 1179
1105 1180 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1106 1181
1107 1182 /*
1108 1183 * First, close and free any existing spare vdevs.
1109 1184 */
1110 1185 for (i = 0; i < spa->spa_spares.sav_count; i++) {
1111 1186 vd = spa->spa_spares.sav_vdevs[i];
1112 1187
1113 1188 /* Undo the call to spa_activate() below */
1114 1189 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1115 1190 B_FALSE)) != NULL && tvd->vdev_isspare)
1116 1191 spa_spare_remove(tvd);
1117 1192 vdev_close(vd);
1118 1193 vdev_free(vd);
1119 1194 }
1120 1195
1121 1196 if (spa->spa_spares.sav_vdevs)
1122 1197 kmem_free(spa->spa_spares.sav_vdevs,
1123 1198 spa->spa_spares.sav_count * sizeof (void *));
1124 1199
1125 1200 if (spa->spa_spares.sav_config == NULL)
1126 1201 nspares = 0;
1127 1202 else
1128 1203 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
1129 1204 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
1130 1205
1131 1206 spa->spa_spares.sav_count = (int)nspares;
1132 1207 spa->spa_spares.sav_vdevs = NULL;
1133 1208
1134 1209 if (nspares == 0)
1135 1210 return;
1136 1211
1137 1212 /*
1138 1213 * Construct the array of vdevs, opening them to get status in the
1139 1214 * process. For each spare, there is potentially two different vdev_t
1140 1215 * structures associated with it: one in the list of spares (used only
1141 1216 * for basic validation purposes) and one in the active vdev
1142 1217 * configuration (if it's spared in). During this phase we open and
1143 1218 * validate each vdev on the spare list. If the vdev also exists in the
1144 1219 * active configuration, then we also mark this vdev as an active spare.
1145 1220 */
1146 1221 spa->spa_spares.sav_vdevs = kmem_alloc(nspares * sizeof (void *),
1147 1222 KM_SLEEP);
1148 1223 for (i = 0; i < spa->spa_spares.sav_count; i++) {
1149 1224 VERIFY(spa_config_parse(spa, &vd, spares[i], NULL, 0,
1150 1225 VDEV_ALLOC_SPARE) == 0);
1151 1226 ASSERT(vd != NULL);
1152 1227
1153 1228 spa->spa_spares.sav_vdevs[i] = vd;
1154 1229
1155 1230 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1156 1231 B_FALSE)) != NULL) {
1157 1232 if (!tvd->vdev_isspare)
1158 1233 spa_spare_add(tvd);
1159 1234
1160 1235 /*
1161 1236 * We only mark the spare active if we were successfully
1162 1237 * able to load the vdev. Otherwise, importing a pool
1163 1238 * with a bad active spare would result in strange
1164 1239 * behavior, because multiple pool would think the spare
1165 1240 * is actively in use.
1166 1241 *
1167 1242 * There is a vulnerability here to an equally bizarre
1168 1243 * circumstance, where a dead active spare is later
1169 1244 * brought back to life (onlined or otherwise). Given
1170 1245 * the rarity of this scenario, and the extra complexity
1171 1246 * it adds, we ignore the possibility.
1172 1247 */
1173 1248 if (!vdev_is_dead(tvd))
1174 1249 spa_spare_activate(tvd);
1175 1250 }
1176 1251
1177 1252 vd->vdev_top = vd;
1178 1253 vd->vdev_aux = &spa->spa_spares;
1179 1254
1180 1255 if (vdev_open(vd) != 0)
1181 1256 continue;
1182 1257
1183 1258 if (vdev_validate_aux(vd) == 0)
1184 1259 spa_spare_add(vd);
1185 1260 }
1186 1261
1187 1262 /*
1188 1263 * Recompute the stashed list of spares, with status information
1189 1264 * this time.
1190 1265 */
1191 1266 VERIFY(nvlist_remove(spa->spa_spares.sav_config, ZPOOL_CONFIG_SPARES,
1192 1267 DATA_TYPE_NVLIST_ARRAY) == 0);
1193 1268
1194 1269 spares = kmem_alloc(spa->spa_spares.sav_count * sizeof (void *),
1195 1270 KM_SLEEP);
1196 1271 for (i = 0; i < spa->spa_spares.sav_count; i++)
1197 1272 spares[i] = vdev_config_generate(spa,
1198 1273 spa->spa_spares.sav_vdevs[i], B_TRUE, VDEV_CONFIG_SPARE);
1199 1274 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
1200 1275 ZPOOL_CONFIG_SPARES, spares, spa->spa_spares.sav_count) == 0);
1201 1276 for (i = 0; i < spa->spa_spares.sav_count; i++)
1202 1277 nvlist_free(spares[i]);
1203 1278 kmem_free(spares, spa->spa_spares.sav_count * sizeof (void *));
1204 1279 }
1205 1280
1206 1281 /*
1207 1282 * Load (or re-load) the current list of vdevs describing the active l2cache for
1208 1283 * this pool. When this is called, we have some form of basic information in
1209 1284 * 'spa_l2cache.sav_config'. We parse this into vdevs, try to open them, and
1210 1285 * then re-generate a more complete list including status information.
1211 1286 * Devices which are already active have their details maintained, and are
1212 1287 * not re-opened.
1213 1288 */
1214 1289 static void
1215 1290 spa_load_l2cache(spa_t *spa)
1216 1291 {
1217 1292 nvlist_t **l2cache;
1218 1293 uint_t nl2cache;
1219 1294 int i, j, oldnvdevs;
1220 1295 uint64_t guid;
1221 1296 vdev_t *vd, **oldvdevs, **newvdevs;
1222 1297 spa_aux_vdev_t *sav = &spa->spa_l2cache;
1223 1298
1224 1299 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1225 1300
1226 1301 if (sav->sav_config != NULL) {
1227 1302 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config,
1228 1303 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
1229 1304 newvdevs = kmem_alloc(nl2cache * sizeof (void *), KM_SLEEP);
1230 1305 } else {
1231 1306 nl2cache = 0;
1232 1307 }
1233 1308
1234 1309 oldvdevs = sav->sav_vdevs;
1235 1310 oldnvdevs = sav->sav_count;
1236 1311 sav->sav_vdevs = NULL;
1237 1312 sav->sav_count = 0;
1238 1313
1239 1314 /*
1240 1315 * Process new nvlist of vdevs.
1241 1316 */
1242 1317 for (i = 0; i < nl2cache; i++) {
1243 1318 VERIFY(nvlist_lookup_uint64(l2cache[i], ZPOOL_CONFIG_GUID,
1244 1319 &guid) == 0);
1245 1320
1246 1321 newvdevs[i] = NULL;
1247 1322 for (j = 0; j < oldnvdevs; j++) {
1248 1323 vd = oldvdevs[j];
1249 1324 if (vd != NULL && guid == vd->vdev_guid) {
1250 1325 /*
1251 1326 * Retain previous vdev for add/remove ops.
1252 1327 */
1253 1328 newvdevs[i] = vd;
1254 1329 oldvdevs[j] = NULL;
1255 1330 break;
1256 1331 }
1257 1332 }
1258 1333
1259 1334 if (newvdevs[i] == NULL) {
1260 1335 /*
1261 1336 * Create new vdev
1262 1337 */
1263 1338 VERIFY(spa_config_parse(spa, &vd, l2cache[i], NULL, 0,
1264 1339 VDEV_ALLOC_L2CACHE) == 0);
1265 1340 ASSERT(vd != NULL);
1266 1341 newvdevs[i] = vd;
1267 1342
1268 1343 /*
1269 1344 * Commit this vdev as an l2cache device,
1270 1345 * even if it fails to open.
1271 1346 */
1272 1347 spa_l2cache_add(vd);
1273 1348
1274 1349 vd->vdev_top = vd;
1275 1350 vd->vdev_aux = sav;
1276 1351
1277 1352 spa_l2cache_activate(vd);
1278 1353
1279 1354 if (vdev_open(vd) != 0)
1280 1355 continue;
1281 1356
1282 1357 (void) vdev_validate_aux(vd);
1283 1358
1284 1359 if (!vdev_is_dead(vd))
1285 1360 l2arc_add_vdev(spa, vd);
1286 1361 }
1287 1362 }
1288 1363
1289 1364 /*
1290 1365 * Purge vdevs that were dropped
1291 1366 */
1292 1367 for (i = 0; i < oldnvdevs; i++) {
1293 1368 uint64_t pool;
1294 1369
1295 1370 vd = oldvdevs[i];
1296 1371 if (vd != NULL) {
1297 1372 ASSERT(vd->vdev_isl2cache);
1298 1373
1299 1374 if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
1300 1375 pool != 0ULL && l2arc_vdev_present(vd))
1301 1376 l2arc_remove_vdev(vd);
1302 1377 vdev_clear_stats(vd);
1303 1378 vdev_free(vd);
1304 1379 }
1305 1380 }
1306 1381
1307 1382 if (oldvdevs)
1308 1383 kmem_free(oldvdevs, oldnvdevs * sizeof (void *));
1309 1384
1310 1385 if (sav->sav_config == NULL)
1311 1386 goto out;
1312 1387
1313 1388 sav->sav_vdevs = newvdevs;
1314 1389 sav->sav_count = (int)nl2cache;
1315 1390
1316 1391 /*
1317 1392 * Recompute the stashed list of l2cache devices, with status
1318 1393 * information this time.
1319 1394 */
1320 1395 VERIFY(nvlist_remove(sav->sav_config, ZPOOL_CONFIG_L2CACHE,
1321 1396 DATA_TYPE_NVLIST_ARRAY) == 0);
1322 1397
1323 1398 l2cache = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
1324 1399 for (i = 0; i < sav->sav_count; i++)
1325 1400 l2cache[i] = vdev_config_generate(spa,
1326 1401 sav->sav_vdevs[i], B_TRUE, VDEV_CONFIG_L2CACHE);
1327 1402 VERIFY(nvlist_add_nvlist_array(sav->sav_config,
1328 1403 ZPOOL_CONFIG_L2CACHE, l2cache, sav->sav_count) == 0);
1329 1404 out:
1330 1405 for (i = 0; i < sav->sav_count; i++)
1331 1406 nvlist_free(l2cache[i]);
1332 1407 if (sav->sav_count)
1333 1408 kmem_free(l2cache, sav->sav_count * sizeof (void *));
1334 1409 }
1335 1410
1336 1411 static int
1337 1412 load_nvlist(spa_t *spa, uint64_t obj, nvlist_t **value)
1338 1413 {
1339 1414 dmu_buf_t *db;
1340 1415 char *packed = NULL;
1341 1416 size_t nvsize = 0;
1342 1417 int error;
1343 1418 *value = NULL;
1344 1419
1345 1420 VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
1346 1421 nvsize = *(uint64_t *)db->db_data;
1347 1422 dmu_buf_rele(db, FTAG);
1348 1423
1349 1424 packed = kmem_alloc(nvsize, KM_SLEEP);
1350 1425 error = dmu_read(spa->spa_meta_objset, obj, 0, nvsize, packed,
1351 1426 DMU_READ_PREFETCH);
1352 1427 if (error == 0)
1353 1428 error = nvlist_unpack(packed, nvsize, value, 0);
1354 1429 kmem_free(packed, nvsize);
1355 1430
1356 1431 return (error);
1357 1432 }
1358 1433
1359 1434 /*
1360 1435 * Checks to see if the given vdev could not be opened, in which case we post a
1361 1436 * sysevent to notify the autoreplace code that the device has been removed.
1362 1437 */
1363 1438 static void
1364 1439 spa_check_removed(vdev_t *vd)
1365 1440 {
1366 1441 for (int c = 0; c < vd->vdev_children; c++)
1367 1442 spa_check_removed(vd->vdev_child[c]);
1368 1443
1369 1444 if (vd->vdev_ops->vdev_op_leaf && vdev_is_dead(vd)) {
1370 1445 zfs_post_autoreplace(vd->vdev_spa, vd);
1371 1446 spa_event_notify(vd->vdev_spa, vd, ESC_ZFS_VDEV_CHECK);
1372 1447 }
1373 1448 }
1374 1449
1375 1450 /*
1376 1451 * Validate the current config against the MOS config
1377 1452 */
1378 1453 static boolean_t
1379 1454 spa_config_valid(spa_t *spa, nvlist_t *config)
1380 1455 {
1381 1456 vdev_t *mrvd, *rvd = spa->spa_root_vdev;
1382 1457 nvlist_t *nv;
1383 1458
1384 1459 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nv) == 0);
1385 1460
1386 1461 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1387 1462 VERIFY(spa_config_parse(spa, &mrvd, nv, NULL, 0, VDEV_ALLOC_LOAD) == 0);
1388 1463
1389 1464 ASSERT3U(rvd->vdev_children, ==, mrvd->vdev_children);
1390 1465
1391 1466 /*
1392 1467 * If we're doing a normal import, then build up any additional
1393 1468 * diagnostic information about missing devices in this config.
1394 1469 * We'll pass this up to the user for further processing.
1395 1470 */
1396 1471 if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG)) {
1397 1472 nvlist_t **child, *nv;
1398 1473 uint64_t idx = 0;
1399 1474
1400 1475 child = kmem_alloc(rvd->vdev_children * sizeof (nvlist_t **),
1401 1476 KM_SLEEP);
1402 1477 VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0);
1403 1478
1404 1479 for (int c = 0; c < rvd->vdev_children; c++) {
1405 1480 vdev_t *tvd = rvd->vdev_child[c];
1406 1481 vdev_t *mtvd = mrvd->vdev_child[c];
1407 1482
1408 1483 if (tvd->vdev_ops == &vdev_missing_ops &&
1409 1484 mtvd->vdev_ops != &vdev_missing_ops &&
1410 1485 mtvd->vdev_islog)
1411 1486 child[idx++] = vdev_config_generate(spa, mtvd,
1412 1487 B_FALSE, 0);
1413 1488 }
1414 1489
1415 1490 if (idx) {
1416 1491 VERIFY(nvlist_add_nvlist_array(nv,
1417 1492 ZPOOL_CONFIG_CHILDREN, child, idx) == 0);
1418 1493 VERIFY(nvlist_add_nvlist(spa->spa_load_info,
1419 1494 ZPOOL_CONFIG_MISSING_DEVICES, nv) == 0);
1420 1495
1421 1496 for (int i = 0; i < idx; i++)
1422 1497 nvlist_free(child[i]);
1423 1498 }
1424 1499 nvlist_free(nv);
1425 1500 kmem_free(child, rvd->vdev_children * sizeof (char **));
1426 1501 }
1427 1502
1428 1503 /*
1429 1504 * Compare the root vdev tree with the information we have
1430 1505 * from the MOS config (mrvd). Check each top-level vdev
1431 1506 * with the corresponding MOS config top-level (mtvd).
1432 1507 */
1433 1508 for (int c = 0; c < rvd->vdev_children; c++) {
1434 1509 vdev_t *tvd = rvd->vdev_child[c];
1435 1510 vdev_t *mtvd = mrvd->vdev_child[c];
1436 1511
1437 1512 /*
1438 1513 * Resolve any "missing" vdevs in the current configuration.
1439 1514 * If we find that the MOS config has more accurate information
1440 1515 * about the top-level vdev then use that vdev instead.
1441 1516 */
1442 1517 if (tvd->vdev_ops == &vdev_missing_ops &&
1443 1518 mtvd->vdev_ops != &vdev_missing_ops) {
1444 1519
1445 1520 if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG))
1446 1521 continue;
1447 1522
1448 1523 /*
1449 1524 * Device specific actions.
1450 1525 */
1451 1526 if (mtvd->vdev_islog) {
1452 1527 spa_set_log_state(spa, SPA_LOG_CLEAR);
1453 1528 } else {
1454 1529 /*
1455 1530 * XXX - once we have 'readonly' pool
1456 1531 * support we should be able to handle
1457 1532 * missing data devices by transitioning
1458 1533 * the pool to readonly.
1459 1534 */
1460 1535 continue;
1461 1536 }
1462 1537
1463 1538 /*
1464 1539 * Swap the missing vdev with the data we were
1465 1540 * able to obtain from the MOS config.
1466 1541 */
1467 1542 vdev_remove_child(rvd, tvd);
1468 1543 vdev_remove_child(mrvd, mtvd);
1469 1544
1470 1545 vdev_add_child(rvd, mtvd);
1471 1546 vdev_add_child(mrvd, tvd);
1472 1547
1473 1548 spa_config_exit(spa, SCL_ALL, FTAG);
1474 1549 vdev_load(mtvd);
1475 1550 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1476 1551
1477 1552 vdev_reopen(rvd);
1478 1553 } else if (mtvd->vdev_islog) {
1479 1554 /*
1480 1555 * Load the slog device's state from the MOS config
1481 1556 * since it's possible that the label does not
1482 1557 * contain the most up-to-date information.
1483 1558 */
1484 1559 vdev_load_log_state(tvd, mtvd);
1485 1560 vdev_reopen(tvd);
1486 1561 }
1487 1562 }
1488 1563 vdev_free(mrvd);
1489 1564 spa_config_exit(spa, SCL_ALL, FTAG);
1490 1565
1491 1566 /*
1492 1567 * Ensure we were able to validate the config.
1493 1568 */
1494 1569 return (rvd->vdev_guid_sum == spa->spa_uberblock.ub_guid_sum);
1495 1570 }
1496 1571
1497 1572 /*
1498 1573 * Check for missing log devices
1499 1574 */
1500 1575 static int
1501 1576 spa_check_logs(spa_t *spa)
1502 1577 {
1503 1578 switch (spa->spa_log_state) {
1504 1579 case SPA_LOG_MISSING:
1505 1580 /* need to recheck in case slog has been restored */
1506 1581 case SPA_LOG_UNKNOWN:
1507 1582 if (dmu_objset_find(spa->spa_name, zil_check_log_chain, NULL,
1508 1583 DS_FIND_CHILDREN)) {
1509 1584 spa_set_log_state(spa, SPA_LOG_MISSING);
1510 1585 return (1);
1511 1586 }
1512 1587 break;
1513 1588 }
1514 1589 return (0);
1515 1590 }
1516 1591
1517 1592 static boolean_t
1518 1593 spa_passivate_log(spa_t *spa)
1519 1594 {
1520 1595 vdev_t *rvd = spa->spa_root_vdev;
1521 1596 boolean_t slog_found = B_FALSE;
1522 1597
1523 1598 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1524 1599
1525 1600 if (!spa_has_slogs(spa))
1526 1601 return (B_FALSE);
1527 1602
1528 1603 for (int c = 0; c < rvd->vdev_children; c++) {
1529 1604 vdev_t *tvd = rvd->vdev_child[c];
1530 1605 metaslab_group_t *mg = tvd->vdev_mg;
1531 1606
1532 1607 if (tvd->vdev_islog) {
1533 1608 metaslab_group_passivate(mg);
1534 1609 slog_found = B_TRUE;
1535 1610 }
1536 1611 }
1537 1612
1538 1613 return (slog_found);
1539 1614 }
1540 1615
1541 1616 static void
1542 1617 spa_activate_log(spa_t *spa)
1543 1618 {
1544 1619 vdev_t *rvd = spa->spa_root_vdev;
1545 1620
1546 1621 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1547 1622
1548 1623 for (int c = 0; c < rvd->vdev_children; c++) {
1549 1624 vdev_t *tvd = rvd->vdev_child[c];
1550 1625 metaslab_group_t *mg = tvd->vdev_mg;
1551 1626
1552 1627 if (tvd->vdev_islog)
1553 1628 metaslab_group_activate(mg);
1554 1629 }
1555 1630 }
1556 1631
1557 1632 int
1558 1633 spa_offline_log(spa_t *spa)
1559 1634 {
1560 1635 int error = 0;
1561 1636
1562 1637 if ((error = dmu_objset_find(spa_name(spa), zil_vdev_offline,
1563 1638 NULL, DS_FIND_CHILDREN)) == 0) {
1564 1639
1565 1640 /*
1566 1641 * We successfully offlined the log device, sync out the
1567 1642 * current txg so that the "stubby" block can be removed
1568 1643 * by zil_sync().
1569 1644 */
1570 1645 txg_wait_synced(spa->spa_dsl_pool, 0);
1571 1646 }
1572 1647 return (error);
1573 1648 }
1574 1649
1575 1650 static void
1576 1651 spa_aux_check_removed(spa_aux_vdev_t *sav)
1577 1652 {
1578 1653 for (int i = 0; i < sav->sav_count; i++)
1579 1654 spa_check_removed(sav->sav_vdevs[i]);
1580 1655 }
1581 1656
1582 1657 void
1583 1658 spa_claim_notify(zio_t *zio)
1584 1659 {
1585 1660 spa_t *spa = zio->io_spa;
1586 1661
1587 1662 if (zio->io_error)
1588 1663 return;
1589 1664
1590 1665 mutex_enter(&spa->spa_props_lock); /* any mutex will do */
1591 1666 if (spa->spa_claim_max_txg < zio->io_bp->blk_birth)
1592 1667 spa->spa_claim_max_txg = zio->io_bp->blk_birth;
1593 1668 mutex_exit(&spa->spa_props_lock);
1594 1669 }
1595 1670
1596 1671 typedef struct spa_load_error {
1597 1672 uint64_t sle_meta_count;
1598 1673 uint64_t sle_data_count;
1599 1674 } spa_load_error_t;
|
↓ open down ↓ |
1003 lines elided |
↑ open up ↑ |
1600 1675
1601 1676 static void
1602 1677 spa_load_verify_done(zio_t *zio)
1603 1678 {
1604 1679 blkptr_t *bp = zio->io_bp;
1605 1680 spa_load_error_t *sle = zio->io_private;
1606 1681 dmu_object_type_t type = BP_GET_TYPE(bp);
1607 1682 int error = zio->io_error;
1608 1683
1609 1684 if (error) {
1610 - if ((BP_GET_LEVEL(bp) != 0 || dmu_ot[type].ot_metadata) &&
1685 + if ((BP_GET_LEVEL(bp) != 0 || DMU_OT_IS_METADATA(type)) &&
1611 1686 type != DMU_OT_INTENT_LOG)
1612 1687 atomic_add_64(&sle->sle_meta_count, 1);
1613 1688 else
1614 1689 atomic_add_64(&sle->sle_data_count, 1);
1615 1690 }
1616 1691 zio_data_buf_free(zio->io_data, zio->io_size);
1617 1692 }
1618 1693
1619 1694 /*ARGSUSED*/
1620 1695 static int
1621 1696 spa_load_verify_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
1622 1697 arc_buf_t *pbuf, const zbookmark_t *zb, const dnode_phys_t *dnp, void *arg)
1623 1698 {
1624 1699 if (bp != NULL) {
1625 1700 zio_t *rio = arg;
1626 1701 size_t size = BP_GET_PSIZE(bp);
1627 1702 void *data = zio_data_buf_alloc(size);
1628 1703
1629 1704 zio_nowait(zio_read(rio, spa, bp, data, size,
1630 1705 spa_load_verify_done, rio->io_private, ZIO_PRIORITY_SCRUB,
1631 1706 ZIO_FLAG_SPECULATIVE | ZIO_FLAG_CANFAIL |
1632 1707 ZIO_FLAG_SCRUB | ZIO_FLAG_RAW, zb));
1633 1708 }
1634 1709 return (0);
1635 1710 }
1636 1711
1637 1712 static int
1638 1713 spa_load_verify(spa_t *spa)
1639 1714 {
1640 1715 zio_t *rio;
1641 1716 spa_load_error_t sle = { 0 };
1642 1717 zpool_rewind_policy_t policy;
1643 1718 boolean_t verify_ok = B_FALSE;
1644 1719 int error;
1645 1720
1646 1721 zpool_get_rewind_policy(spa->spa_config, &policy);
1647 1722
1648 1723 if (policy.zrp_request & ZPOOL_NEVER_REWIND)
1649 1724 return (0);
1650 1725
1651 1726 rio = zio_root(spa, NULL, &sle,
1652 1727 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE);
1653 1728
1654 1729 error = traverse_pool(spa, spa->spa_verify_min_txg,
1655 1730 TRAVERSE_PRE | TRAVERSE_PREFETCH, spa_load_verify_cb, rio);
1656 1731
1657 1732 (void) zio_wait(rio);
1658 1733
1659 1734 spa->spa_load_meta_errors = sle.sle_meta_count;
1660 1735 spa->spa_load_data_errors = sle.sle_data_count;
1661 1736
1662 1737 if (!error && sle.sle_meta_count <= policy.zrp_maxmeta &&
1663 1738 sle.sle_data_count <= policy.zrp_maxdata) {
1664 1739 int64_t loss = 0;
1665 1740
1666 1741 verify_ok = B_TRUE;
1667 1742 spa->spa_load_txg = spa->spa_uberblock.ub_txg;
1668 1743 spa->spa_load_txg_ts = spa->spa_uberblock.ub_timestamp;
1669 1744
1670 1745 loss = spa->spa_last_ubsync_txg_ts - spa->spa_load_txg_ts;
1671 1746 VERIFY(nvlist_add_uint64(spa->spa_load_info,
1672 1747 ZPOOL_CONFIG_LOAD_TIME, spa->spa_load_txg_ts) == 0);
1673 1748 VERIFY(nvlist_add_int64(spa->spa_load_info,
1674 1749 ZPOOL_CONFIG_REWIND_TIME, loss) == 0);
1675 1750 VERIFY(nvlist_add_uint64(spa->spa_load_info,
1676 1751 ZPOOL_CONFIG_LOAD_DATA_ERRORS, sle.sle_data_count) == 0);
1677 1752 } else {
1678 1753 spa->spa_load_max_txg = spa->spa_uberblock.ub_txg;
1679 1754 }
1680 1755
1681 1756 if (error) {
1682 1757 if (error != ENXIO && error != EIO)
1683 1758 error = EIO;
1684 1759 return (error);
1685 1760 }
1686 1761
1687 1762 return (verify_ok ? 0 : EIO);
1688 1763 }
1689 1764
1690 1765 /*
1691 1766 * Find a value in the pool props object.
1692 1767 */
1693 1768 static void
1694 1769 spa_prop_find(spa_t *spa, zpool_prop_t prop, uint64_t *val)
1695 1770 {
1696 1771 (void) zap_lookup(spa->spa_meta_objset, spa->spa_pool_props_object,
1697 1772 zpool_prop_to_name(prop), sizeof (uint64_t), 1, val);
1698 1773 }
1699 1774
1700 1775 /*
1701 1776 * Find a value in the pool directory object.
1702 1777 */
1703 1778 static int
1704 1779 spa_dir_prop(spa_t *spa, const char *name, uint64_t *val)
1705 1780 {
1706 1781 return (zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
1707 1782 name, sizeof (uint64_t), 1, val));
1708 1783 }
1709 1784
1710 1785 static int
1711 1786 spa_vdev_err(vdev_t *vdev, vdev_aux_t aux, int err)
1712 1787 {
1713 1788 vdev_set_state(vdev, B_TRUE, VDEV_STATE_CANT_OPEN, aux);
1714 1789 return (err);
1715 1790 }
1716 1791
1717 1792 /*
1718 1793 * Fix up config after a partly-completed split. This is done with the
1719 1794 * ZPOOL_CONFIG_SPLIT nvlist. Both the splitting pool and the split-off
1720 1795 * pool have that entry in their config, but only the splitting one contains
1721 1796 * a list of all the guids of the vdevs that are being split off.
1722 1797 *
1723 1798 * This function determines what to do with that list: either rejoin
1724 1799 * all the disks to the pool, or complete the splitting process. To attempt
1725 1800 * the rejoin, each disk that is offlined is marked online again, and
1726 1801 * we do a reopen() call. If the vdev label for every disk that was
1727 1802 * marked online indicates it was successfully split off (VDEV_AUX_SPLIT_POOL)
1728 1803 * then we call vdev_split() on each disk, and complete the split.
1729 1804 *
1730 1805 * Otherwise we leave the config alone, with all the vdevs in place in
1731 1806 * the original pool.
1732 1807 */
1733 1808 static void
1734 1809 spa_try_repair(spa_t *spa, nvlist_t *config)
1735 1810 {
1736 1811 uint_t extracted;
1737 1812 uint64_t *glist;
1738 1813 uint_t i, gcount;
1739 1814 nvlist_t *nvl;
1740 1815 vdev_t **vd;
1741 1816 boolean_t attempt_reopen;
1742 1817
1743 1818 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, &nvl) != 0)
1744 1819 return;
1745 1820
1746 1821 /* check that the config is complete */
1747 1822 if (nvlist_lookup_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
1748 1823 &glist, &gcount) != 0)
1749 1824 return;
1750 1825
1751 1826 vd = kmem_zalloc(gcount * sizeof (vdev_t *), KM_SLEEP);
1752 1827
1753 1828 /* attempt to online all the vdevs & validate */
1754 1829 attempt_reopen = B_TRUE;
1755 1830 for (i = 0; i < gcount; i++) {
1756 1831 if (glist[i] == 0) /* vdev is hole */
1757 1832 continue;
1758 1833
1759 1834 vd[i] = spa_lookup_by_guid(spa, glist[i], B_FALSE);
1760 1835 if (vd[i] == NULL) {
1761 1836 /*
1762 1837 * Don't bother attempting to reopen the disks;
1763 1838 * just do the split.
1764 1839 */
1765 1840 attempt_reopen = B_FALSE;
1766 1841 } else {
1767 1842 /* attempt to re-online it */
1768 1843 vd[i]->vdev_offline = B_FALSE;
1769 1844 }
1770 1845 }
1771 1846
1772 1847 if (attempt_reopen) {
1773 1848 vdev_reopen(spa->spa_root_vdev);
1774 1849
1775 1850 /* check each device to see what state it's in */
1776 1851 for (extracted = 0, i = 0; i < gcount; i++) {
1777 1852 if (vd[i] != NULL &&
1778 1853 vd[i]->vdev_stat.vs_aux != VDEV_AUX_SPLIT_POOL)
1779 1854 break;
1780 1855 ++extracted;
1781 1856 }
1782 1857 }
1783 1858
1784 1859 /*
1785 1860 * If every disk has been moved to the new pool, or if we never
1786 1861 * even attempted to look at them, then we split them off for
1787 1862 * good.
1788 1863 */
1789 1864 if (!attempt_reopen || gcount == extracted) {
1790 1865 for (i = 0; i < gcount; i++)
1791 1866 if (vd[i] != NULL)
1792 1867 vdev_split(vd[i]);
1793 1868 vdev_reopen(spa->spa_root_vdev);
1794 1869 }
1795 1870
1796 1871 kmem_free(vd, gcount * sizeof (vdev_t *));
1797 1872 }
1798 1873
1799 1874 static int
1800 1875 spa_load(spa_t *spa, spa_load_state_t state, spa_import_type_t type,
1801 1876 boolean_t mosconfig)
1802 1877 {
1803 1878 nvlist_t *config = spa->spa_config;
1804 1879 char *ereport = FM_EREPORT_ZFS_POOL;
1805 1880 char *comment;
1806 1881 int error;
1807 1882 uint64_t pool_guid;
1808 1883 nvlist_t *nvl;
1809 1884
1810 1885 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, &pool_guid))
1811 1886 return (EINVAL);
1812 1887
1813 1888 ASSERT(spa->spa_comment == NULL);
1814 1889 if (nvlist_lookup_string(config, ZPOOL_CONFIG_COMMENT, &comment) == 0)
1815 1890 spa->spa_comment = spa_strdup(comment);
1816 1891
1817 1892 /*
1818 1893 * Versioning wasn't explicitly added to the label until later, so if
1819 1894 * it's not present treat it as the initial version.
1820 1895 */
1821 1896 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION,
1822 1897 &spa->spa_ubsync.ub_version) != 0)
1823 1898 spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL;
1824 1899
1825 1900 (void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG,
1826 1901 &spa->spa_config_txg);
1827 1902
1828 1903 if ((state == SPA_LOAD_IMPORT || state == SPA_LOAD_TRYIMPORT) &&
1829 1904 spa_guid_exists(pool_guid, 0)) {
|
↓ open down ↓ |
209 lines elided |
↑ open up ↑ |
1830 1905 error = EEXIST;
1831 1906 } else {
1832 1907 spa->spa_config_guid = pool_guid;
1833 1908
1834 1909 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT,
1835 1910 &nvl) == 0) {
1836 1911 VERIFY(nvlist_dup(nvl, &spa->spa_config_splitting,
1837 1912 KM_SLEEP) == 0);
1838 1913 }
1839 1914
1915 + nvlist_free(spa->spa_load_info);
1916 + spa->spa_load_info = fnvlist_alloc();
1917 +
1840 1918 gethrestime(&spa->spa_loaded_ts);
1841 1919 error = spa_load_impl(spa, pool_guid, config, state, type,
1842 1920 mosconfig, &ereport);
1843 1921 }
1844 1922
1845 1923 spa->spa_minref = refcount_count(&spa->spa_refcount);
1846 1924 if (error) {
1847 1925 if (error != EEXIST) {
1848 1926 spa->spa_loaded_ts.tv_sec = 0;
1849 1927 spa->spa_loaded_ts.tv_nsec = 0;
1850 1928 }
1851 1929 if (error != EBADF) {
1852 1930 zfs_ereport_post(ereport, spa, NULL, NULL, 0, 0);
1853 1931 }
1854 1932 }
1855 1933 spa->spa_load_state = error ? SPA_LOAD_ERROR : SPA_LOAD_NONE;
1856 1934 spa->spa_ena = 0;
1857 1935
1858 1936 return (error);
1859 1937 }
1860 1938
1861 1939 /*
|
↓ open down ↓ |
12 lines elided |
↑ open up ↑ |
1862 1940 * Load an existing storage pool, using the pool's builtin spa_config as a
1863 1941 * source of configuration information.
1864 1942 */
1865 1943 static int
1866 1944 spa_load_impl(spa_t *spa, uint64_t pool_guid, nvlist_t *config,
1867 1945 spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig,
1868 1946 char **ereport)
1869 1947 {
1870 1948 int error = 0;
1871 1949 nvlist_t *nvroot = NULL;
1950 + nvlist_t *label;
1872 1951 vdev_t *rvd;
1873 1952 uberblock_t *ub = &spa->spa_uberblock;
1874 1953 uint64_t children, config_cache_txg = spa->spa_config_txg;
1875 1954 int orig_mode = spa->spa_mode;
1876 1955 int parse;
1877 1956 uint64_t obj;
1957 + boolean_t missing_feat_write = B_FALSE;
1878 1958
1879 1959 /*
1880 1960 * If this is an untrusted config, access the pool in read-only mode.
1881 1961 * This prevents things like resilvering recently removed devices.
1882 1962 */
1883 1963 if (!mosconfig)
1884 1964 spa->spa_mode = FREAD;
1885 1965
1886 1966 ASSERT(MUTEX_HELD(&spa_namespace_lock));
1887 1967
1888 1968 spa->spa_load_state = state;
1889 1969
1890 1970 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvroot))
1891 1971 return (EINVAL);
1892 1972
1893 1973 parse = (type == SPA_IMPORT_EXISTING ?
1894 1974 VDEV_ALLOC_LOAD : VDEV_ALLOC_SPLIT);
1895 1975
1896 1976 /*
1897 1977 * Create "The Godfather" zio to hold all async IOs
1898 1978 */
1899 1979 spa->spa_async_zio_root = zio_root(spa, NULL, NULL,
1900 1980 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_GODFATHER);
1901 1981
1902 1982 /*
1903 1983 * Parse the configuration into a vdev tree. We explicitly set the
1904 1984 * value that will be returned by spa_version() since parsing the
1905 1985 * configuration requires knowing the version number.
1906 1986 */
1907 1987 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1908 1988 error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, parse);
1909 1989 spa_config_exit(spa, SCL_ALL, FTAG);
1910 1990
1911 1991 if (error != 0)
1912 1992 return (error);
1913 1993
1914 1994 ASSERT(spa->spa_root_vdev == rvd);
1915 1995
1916 1996 if (type != SPA_IMPORT_ASSEMBLE) {
1917 1997 ASSERT(spa_guid(spa) == pool_guid);
1918 1998 }
1919 1999
1920 2000 /*
1921 2001 * Try to open all vdevs, loading each label in the process.
1922 2002 */
1923 2003 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1924 2004 error = vdev_open(rvd);
1925 2005 spa_config_exit(spa, SCL_ALL, FTAG);
1926 2006 if (error != 0)
1927 2007 return (error);
1928 2008
1929 2009 /*
1930 2010 * We need to validate the vdev labels against the configuration that
1931 2011 * we have in hand, which is dependent on the setting of mosconfig. If
1932 2012 * mosconfig is true then we're validating the vdev labels based on
1933 2013 * that config. Otherwise, we're validating against the cached config
1934 2014 * (zpool.cache) that was read when we loaded the zfs module, and then
1935 2015 * later we will recursively call spa_load() and validate against
1936 2016 * the vdev config.
1937 2017 *
1938 2018 * If we're assembling a new pool that's been split off from an
1939 2019 * existing pool, the labels haven't yet been updated so we skip
1940 2020 * validation for now.
1941 2021 */
1942 2022 if (type != SPA_IMPORT_ASSEMBLE) {
1943 2023 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1944 2024 error = vdev_validate(rvd, mosconfig);
1945 2025 spa_config_exit(spa, SCL_ALL, FTAG);
1946 2026
|
↓ open down ↓ |
59 lines elided |
↑ open up ↑ |
1947 2027 if (error != 0)
1948 2028 return (error);
1949 2029
1950 2030 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN)
1951 2031 return (ENXIO);
1952 2032 }
1953 2033
1954 2034 /*
1955 2035 * Find the best uberblock.
1956 2036 */
1957 - vdev_uberblock_load(NULL, rvd, ub);
2037 + vdev_uberblock_load(rvd, ub, &label);
1958 2038
1959 2039 /*
1960 2040 * If we weren't able to find a single valid uberblock, return failure.
1961 2041 */
1962 - if (ub->ub_txg == 0)
2042 + if (ub->ub_txg == 0) {
2043 + nvlist_free(label);
1963 2044 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, ENXIO));
2045 + }
1964 2046
1965 2047 /*
1966 - * If the pool is newer than the code, we can't open it.
2048 + * If the pool has an unsupported version we can't open it.
1967 2049 */
1968 - if (ub->ub_version > SPA_VERSION)
2050 + if (!SPA_VERSION_IS_SUPPORTED(ub->ub_version)) {
2051 + nvlist_free(label);
1969 2052 return (spa_vdev_err(rvd, VDEV_AUX_VERSION_NEWER, ENOTSUP));
2053 + }
1970 2054
2055 + if (ub->ub_version >= SPA_VERSION_FEATURES) {
2056 + nvlist_t *features;
2057 +
2058 + /*
2059 + * If we weren't able to find what's necessary for reading the
2060 + * MOS in the label, return failure.
2061 + */
2062 + if (label == NULL || nvlist_lookup_nvlist(label,
2063 + ZPOOL_CONFIG_FEATURES_FOR_READ, &features) != 0) {
2064 + nvlist_free(label);
2065 + return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
2066 + ENXIO));
2067 + }
2068 +
2069 + /*
2070 + * Update our in-core representation with the definitive values
2071 + * from the label.
2072 + */
2073 + nvlist_free(spa->spa_label_features);
2074 + VERIFY(nvlist_dup(features, &spa->spa_label_features, 0) == 0);
2075 + }
2076 +
2077 + nvlist_free(label);
2078 +
1971 2079 /*
2080 + * Look through entries in the label nvlist's features_for_read. If
2081 + * there is a feature listed there which we don't understand then we
2082 + * cannot open a pool.
2083 + */
2084 + if (ub->ub_version >= SPA_VERSION_FEATURES) {
2085 + nvlist_t *unsup_feat;
2086 +
2087 + VERIFY(nvlist_alloc(&unsup_feat, NV_UNIQUE_NAME, KM_SLEEP) ==
2088 + 0);
2089 +
2090 + for (nvpair_t *nvp = nvlist_next_nvpair(spa->spa_label_features,
2091 + NULL); nvp != NULL;
2092 + nvp = nvlist_next_nvpair(spa->spa_label_features, nvp)) {
2093 + if (!zfeature_is_supported(nvpair_name(nvp))) {
2094 + VERIFY(nvlist_add_string(unsup_feat,
2095 + nvpair_name(nvp), "") == 0);
2096 + }
2097 + }
2098 +
2099 + if (!nvlist_empty(unsup_feat)) {
2100 + VERIFY(nvlist_add_nvlist(spa->spa_load_info,
2101 + ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat) == 0);
2102 + nvlist_free(unsup_feat);
2103 + return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT,
2104 + ENOTSUP));
2105 + }
2106 +
2107 + nvlist_free(unsup_feat);
2108 + }
2109 +
2110 + /*
1972 2111 * If the vdev guid sum doesn't match the uberblock, we have an
1973 2112 * incomplete configuration. We first check to see if the pool
1974 2113 * is aware of the complete config (i.e ZPOOL_CONFIG_VDEV_CHILDREN).
1975 2114 * If it is, defer the vdev_guid_sum check till later so we
1976 2115 * can handle missing vdevs.
1977 2116 */
1978 2117 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VDEV_CHILDREN,
1979 2118 &children) != 0 && mosconfig && type != SPA_IMPORT_ASSEMBLE &&
1980 2119 rvd->vdev_guid_sum != ub->ub_guid_sum)
1981 2120 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM, ENXIO));
1982 2121
1983 2122 if (type != SPA_IMPORT_ASSEMBLE && spa->spa_config_splitting) {
1984 2123 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1985 2124 spa_try_repair(spa, config);
1986 2125 spa_config_exit(spa, SCL_ALL, FTAG);
1987 2126 nvlist_free(spa->spa_config_splitting);
1988 2127 spa->spa_config_splitting = NULL;
1989 2128 }
1990 2129
1991 2130 /*
1992 2131 * Initialize internal SPA structures.
|
↓ open down ↓ |
11 lines elided |
↑ open up ↑ |
1993 2132 */
1994 2133 spa->spa_state = POOL_STATE_ACTIVE;
1995 2134 spa->spa_ubsync = spa->spa_uberblock;
1996 2135 spa->spa_verify_min_txg = spa->spa_extreme_rewind ?
1997 2136 TXG_INITIAL - 1 : spa_last_synced_txg(spa) - TXG_DEFER_SIZE - 1;
1998 2137 spa->spa_first_txg = spa->spa_last_ubsync_txg ?
1999 2138 spa->spa_last_ubsync_txg : spa_last_synced_txg(spa) + 1;
2000 2139 spa->spa_claim_max_txg = spa->spa_first_txg;
2001 2140 spa->spa_prev_software_version = ub->ub_software_version;
2002 2141
2003 - error = dsl_pool_open(spa, spa->spa_first_txg, &spa->spa_dsl_pool);
2142 + error = dsl_pool_init(spa, spa->spa_first_txg, &spa->spa_dsl_pool);
2004 2143 if (error)
2005 2144 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2006 2145 spa->spa_meta_objset = spa->spa_dsl_pool->dp_meta_objset;
2007 2146
2008 2147 if (spa_dir_prop(spa, DMU_POOL_CONFIG, &spa->spa_config_object) != 0)
2009 2148 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2010 2149
2150 + if (spa_version(spa) >= SPA_VERSION_FEATURES) {
2151 + boolean_t missing_feat_read = B_FALSE;
2152 + nvlist_t *unsup_feat;
2153 +
2154 + if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_READ,
2155 + &spa->spa_feat_for_read_obj) != 0) {
2156 + return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2157 + }
2158 +
2159 + if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_WRITE,
2160 + &spa->spa_feat_for_write_obj) != 0) {
2161 + return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2162 + }
2163 +
2164 + if (spa_dir_prop(spa, DMU_POOL_FEATURE_DESCRIPTIONS,
2165 + &spa->spa_feat_desc_obj) != 0) {
2166 + return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2167 + }
2168 +
2169 + VERIFY(nvlist_alloc(&unsup_feat, NV_UNIQUE_NAME, KM_SLEEP) ==
2170 + 0);
2171 +
2172 + if (!feature_is_supported(spa->spa_meta_objset,
2173 + spa->spa_feat_for_read_obj, spa->spa_feat_desc_obj,
2174 + unsup_feat))
2175 + missing_feat_read = B_TRUE;
2176 +
2177 + if (spa_writeable(spa) || state == SPA_LOAD_TRYIMPORT) {
2178 + if (!feature_is_supported(spa->spa_meta_objset,
2179 + spa->spa_feat_for_write_obj, spa->spa_feat_desc_obj,
2180 + unsup_feat))
2181 + missing_feat_write = B_TRUE;
2182 + }
2183 +
2184 + if (!nvlist_empty(unsup_feat)) {
2185 + VERIFY(nvlist_add_nvlist(spa->spa_load_info,
2186 + ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat) == 0);
2187 + }
2188 +
2189 + nvlist_free(unsup_feat);
2190 +
2191 + if (!missing_feat_read) {
2192 + fnvlist_add_boolean(spa->spa_load_info,
2193 + ZPOOL_CONFIG_CAN_RDONLY);
2194 + }
2195 +
2196 + /*
2197 + * If the state is SPA_LOAD_TRYIMPORT, our objective is
2198 + * twofold: to determine whether the pool is available for
2199 + * import in read-write mode and (if it is not) whether the
2200 + * pool is available for import in read-only mode. If the pool
2201 + * is available for import in read-write mode, it is displayed
2202 + * as available in userland; if it is not available for import
2203 + * in read-only mode, it is displayed as unavailable in
2204 + * userland. If the pool is available for import in read-only
2205 + * mode but not read-write mode, it is displayed as unavailable
2206 + * in userland with a special note that the pool is actually
2207 + * available for open in read-only mode.
2208 + *
2209 + * As a result, if the state is SPA_LOAD_TRYIMPORT and we are
2210 + * missing a feature for write, we must first determine whether
2211 + * the pool can be opened read-only before returning to
2212 + * userland in order to know whether to display the
2213 + * abovementioned note.
2214 + */
2215 + if (missing_feat_read || (missing_feat_write &&
2216 + spa_writeable(spa))) {
2217 + return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT,
2218 + ENOTSUP));
2219 + }
2220 + }
2221 +
2222 + spa->spa_is_initializing = B_TRUE;
2223 + error = dsl_pool_open(spa->spa_dsl_pool);
2224 + spa->spa_is_initializing = B_FALSE;
2225 + if (error != 0)
2226 + return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2227 +
2011 2228 if (!mosconfig) {
2012 2229 uint64_t hostid;
2013 2230 nvlist_t *policy = NULL, *nvconfig;
2014 2231
2015 2232 if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0)
2016 2233 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2017 2234
2018 2235 if (!spa_is_root(spa) && nvlist_lookup_uint64(nvconfig,
2019 2236 ZPOOL_CONFIG_HOSTID, &hostid) == 0) {
2020 2237 char *hostname;
2021 2238 unsigned long myhostid = 0;
2022 2239
2023 2240 VERIFY(nvlist_lookup_string(nvconfig,
2024 2241 ZPOOL_CONFIG_HOSTNAME, &hostname) == 0);
2025 2242
2026 2243 #ifdef _KERNEL
2027 2244 myhostid = zone_get_hostid(NULL);
2028 2245 #else /* _KERNEL */
2029 2246 /*
2030 2247 * We're emulating the system's hostid in userland, so
2031 2248 * we can't use zone_get_hostid().
2032 2249 */
2033 2250 (void) ddi_strtoul(hw_serial, NULL, 10, &myhostid);
2034 2251 #endif /* _KERNEL */
2035 2252 if (hostid != 0 && myhostid != 0 &&
2036 2253 hostid != myhostid) {
2037 2254 nvlist_free(nvconfig);
2038 2255 cmn_err(CE_WARN, "pool '%s' could not be "
2039 2256 "loaded as it was last accessed by "
2040 2257 "another system (host: %s hostid: 0x%lx). "
2041 2258 "See: http://illumos.org/msg/ZFS-8000-EY",
2042 2259 spa_name(spa), hostname,
2043 2260 (unsigned long)hostid);
2044 2261 return (EBADF);
2045 2262 }
2046 2263 }
2047 2264 if (nvlist_lookup_nvlist(spa->spa_config,
2048 2265 ZPOOL_REWIND_POLICY, &policy) == 0)
2049 2266 VERIFY(nvlist_add_nvlist(nvconfig,
2050 2267 ZPOOL_REWIND_POLICY, policy) == 0);
2051 2268
2052 2269 spa_config_set(spa, nvconfig);
2053 2270 spa_unload(spa);
2054 2271 spa_deactivate(spa);
2055 2272 spa_activate(spa, orig_mode);
2056 2273
2057 2274 return (spa_load(spa, state, SPA_IMPORT_EXISTING, B_TRUE));
2058 2275 }
2059 2276
2060 2277 if (spa_dir_prop(spa, DMU_POOL_SYNC_BPOBJ, &obj) != 0)
2061 2278 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2062 2279 error = bpobj_open(&spa->spa_deferred_bpobj, spa->spa_meta_objset, obj);
2063 2280 if (error != 0)
2064 2281 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2065 2282
2066 2283 /*
2067 2284 * Load the bit that tells us to use the new accounting function
2068 2285 * (raid-z deflation). If we have an older pool, this will not
2069 2286 * be present.
2070 2287 */
2071 2288 error = spa_dir_prop(spa, DMU_POOL_DEFLATE, &spa->spa_deflate);
2072 2289 if (error != 0 && error != ENOENT)
2073 2290 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2074 2291
2075 2292 error = spa_dir_prop(spa, DMU_POOL_CREATION_VERSION,
2076 2293 &spa->spa_creation_version);
2077 2294 if (error != 0 && error != ENOENT)
2078 2295 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2079 2296
2080 2297 /*
2081 2298 * Load the persistent error log. If we have an older pool, this will
2082 2299 * not be present.
2083 2300 */
2084 2301 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_LAST, &spa->spa_errlog_last);
2085 2302 if (error != 0 && error != ENOENT)
2086 2303 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2087 2304
2088 2305 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_SCRUB,
2089 2306 &spa->spa_errlog_scrub);
2090 2307 if (error != 0 && error != ENOENT)
2091 2308 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2092 2309
2093 2310 /*
2094 2311 * Load the history object. If we have an older pool, this
2095 2312 * will not be present.
2096 2313 */
2097 2314 error = spa_dir_prop(spa, DMU_POOL_HISTORY, &spa->spa_history);
2098 2315 if (error != 0 && error != ENOENT)
2099 2316 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2100 2317
2101 2318 /*
2102 2319 * If we're assembling the pool from the split-off vdevs of
2103 2320 * an existing pool, we don't want to attach the spares & cache
2104 2321 * devices.
2105 2322 */
2106 2323
2107 2324 /*
2108 2325 * Load any hot spares for this pool.
2109 2326 */
2110 2327 error = spa_dir_prop(spa, DMU_POOL_SPARES, &spa->spa_spares.sav_object);
2111 2328 if (error != 0 && error != ENOENT)
2112 2329 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2113 2330 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
2114 2331 ASSERT(spa_version(spa) >= SPA_VERSION_SPARES);
2115 2332 if (load_nvlist(spa, spa->spa_spares.sav_object,
2116 2333 &spa->spa_spares.sav_config) != 0)
2117 2334 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2118 2335
2119 2336 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2120 2337 spa_load_spares(spa);
2121 2338 spa_config_exit(spa, SCL_ALL, FTAG);
2122 2339 } else if (error == 0) {
2123 2340 spa->spa_spares.sav_sync = B_TRUE;
2124 2341 }
2125 2342
2126 2343 /*
2127 2344 * Load any level 2 ARC devices for this pool.
2128 2345 */
2129 2346 error = spa_dir_prop(spa, DMU_POOL_L2CACHE,
2130 2347 &spa->spa_l2cache.sav_object);
2131 2348 if (error != 0 && error != ENOENT)
2132 2349 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2133 2350 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
2134 2351 ASSERT(spa_version(spa) >= SPA_VERSION_L2CACHE);
2135 2352 if (load_nvlist(spa, spa->spa_l2cache.sav_object,
2136 2353 &spa->spa_l2cache.sav_config) != 0)
2137 2354 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2138 2355
2139 2356 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2140 2357 spa_load_l2cache(spa);
2141 2358 spa_config_exit(spa, SCL_ALL, FTAG);
2142 2359 } else if (error == 0) {
2143 2360 spa->spa_l2cache.sav_sync = B_TRUE;
2144 2361 }
2145 2362
2146 2363 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
2147 2364
2148 2365 error = spa_dir_prop(spa, DMU_POOL_PROPS, &spa->spa_pool_props_object);
2149 2366 if (error && error != ENOENT)
2150 2367 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2151 2368
2152 2369 if (error == 0) {
2153 2370 uint64_t autoreplace;
2154 2371
2155 2372 spa_prop_find(spa, ZPOOL_PROP_BOOTFS, &spa->spa_bootfs);
2156 2373 spa_prop_find(spa, ZPOOL_PROP_AUTOREPLACE, &autoreplace);
2157 2374 spa_prop_find(spa, ZPOOL_PROP_DELEGATION, &spa->spa_delegation);
2158 2375 spa_prop_find(spa, ZPOOL_PROP_FAILUREMODE, &spa->spa_failmode);
2159 2376 spa_prop_find(spa, ZPOOL_PROP_AUTOEXPAND, &spa->spa_autoexpand);
2160 2377 spa_prop_find(spa, ZPOOL_PROP_DEDUPDITTO,
2161 2378 &spa->spa_dedup_ditto);
2162 2379
2163 2380 spa->spa_autoreplace = (autoreplace != 0);
2164 2381 }
2165 2382
2166 2383 /*
2167 2384 * If the 'autoreplace' property is set, then post a resource notifying
2168 2385 * the ZFS DE that it should not issue any faults for unopenable
2169 2386 * devices. We also iterate over the vdevs, and post a sysevent for any
2170 2387 * unopenable vdevs so that the normal autoreplace handler can take
2171 2388 * over.
2172 2389 */
2173 2390 if (spa->spa_autoreplace && state != SPA_LOAD_TRYIMPORT) {
2174 2391 spa_check_removed(spa->spa_root_vdev);
2175 2392 /*
2176 2393 * For the import case, this is done in spa_import(), because
2177 2394 * at this point we're using the spare definitions from
2178 2395 * the MOS config, not necessarily from the userland config.
2179 2396 */
2180 2397 if (state != SPA_LOAD_IMPORT) {
2181 2398 spa_aux_check_removed(&spa->spa_spares);
2182 2399 spa_aux_check_removed(&spa->spa_l2cache);
2183 2400 }
2184 2401 }
2185 2402
2186 2403 /*
2187 2404 * Load the vdev state for all toplevel vdevs.
2188 2405 */
2189 2406 vdev_load(rvd);
2190 2407
2191 2408 /*
2192 2409 * Propagate the leaf DTLs we just loaded all the way up the tree.
2193 2410 */
2194 2411 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2195 2412 vdev_dtl_reassess(rvd, 0, 0, B_FALSE);
2196 2413 spa_config_exit(spa, SCL_ALL, FTAG);
2197 2414
2198 2415 /*
2199 2416 * Load the DDTs (dedup tables).
2200 2417 */
2201 2418 error = ddt_load(spa);
2202 2419 if (error != 0)
2203 2420 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2204 2421
2205 2422 spa_update_dspace(spa);
2206 2423
2207 2424 /*
2208 2425 * Validate the config, using the MOS config to fill in any
2209 2426 * information which might be missing. If we fail to validate
2210 2427 * the config then declare the pool unfit for use. If we're
2211 2428 * assembling a pool from a split, the log is not transferred
2212 2429 * over.
2213 2430 */
2214 2431 if (type != SPA_IMPORT_ASSEMBLE) {
2215 2432 nvlist_t *nvconfig;
2216 2433
2217 2434 if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0)
|
↓ open down ↓ |
197 lines elided |
↑ open up ↑ |
2218 2435 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2219 2436
2220 2437 if (!spa_config_valid(spa, nvconfig)) {
2221 2438 nvlist_free(nvconfig);
2222 2439 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM,
2223 2440 ENXIO));
2224 2441 }
2225 2442 nvlist_free(nvconfig);
2226 2443
2227 2444 /*
2228 - * Now that we've validate the config, check the state of the
2445 + * Now that we've validated the config, check the state of the
2229 2446 * root vdev. If it can't be opened, it indicates one or
2230 2447 * more toplevel vdevs are faulted.
2231 2448 */
2232 2449 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN)
2233 2450 return (ENXIO);
2234 2451
2235 2452 if (spa_check_logs(spa)) {
2236 2453 *ereport = FM_EREPORT_ZFS_LOG_REPLAY;
2237 2454 return (spa_vdev_err(rvd, VDEV_AUX_BAD_LOG, ENXIO));
2238 2455 }
2239 2456 }
2240 2457
2458 + if (missing_feat_write) {
2459 + ASSERT(state == SPA_LOAD_TRYIMPORT);
2460 +
2461 + /*
2462 + * At this point, we know that we can open the pool in
2463 + * read-only mode but not read-write mode. We now have enough
2464 + * information and can return to userland.
2465 + */
2466 + return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT, ENOTSUP));
2467 + }
2468 +
2241 2469 /*
2242 2470 * We've successfully opened the pool, verify that we're ready
2243 2471 * to start pushing transactions.
2244 2472 */
2245 2473 if (state != SPA_LOAD_TRYIMPORT) {
2246 2474 if (error = spa_load_verify(spa))
2247 2475 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
2248 2476 error));
2249 2477 }
2250 2478
2251 2479 if (spa_writeable(spa) && (state == SPA_LOAD_RECOVER ||
2252 2480 spa->spa_load_max_txg == UINT64_MAX)) {
2253 2481 dmu_tx_t *tx;
2254 2482 int need_update = B_FALSE;
2255 2483
2256 2484 ASSERT(state != SPA_LOAD_TRYIMPORT);
2257 2485
2258 2486 /*
2259 2487 * Claim log blocks that haven't been committed yet.
2260 2488 * This must all happen in a single txg.
2261 2489 * Note: spa_claim_max_txg is updated by spa_claim_notify(),
2262 2490 * invoked from zil_claim_log_block()'s i/o done callback.
2263 2491 * Price of rollback is that we abandon the log.
2264 2492 */
2265 2493 spa->spa_claiming = B_TRUE;
2266 2494
2267 2495 tx = dmu_tx_create_assigned(spa_get_dsl(spa),
2268 2496 spa_first_txg(spa));
2269 2497 (void) dmu_objset_find(spa_name(spa),
2270 2498 zil_claim, tx, DS_FIND_CHILDREN);
2271 2499 dmu_tx_commit(tx);
2272 2500
2273 2501 spa->spa_claiming = B_FALSE;
2274 2502
2275 2503 spa_set_log_state(spa, SPA_LOG_GOOD);
2276 2504 spa->spa_sync_on = B_TRUE;
2277 2505 txg_sync_start(spa->spa_dsl_pool);
2278 2506
2279 2507 /*
2280 2508 * Wait for all claims to sync. We sync up to the highest
2281 2509 * claimed log block birth time so that claimed log blocks
2282 2510 * don't appear to be from the future. spa_claim_max_txg
2283 2511 * will have been set for us by either zil_check_log_chain()
2284 2512 * (invoked from spa_check_logs()) or zil_claim() above.
2285 2513 */
2286 2514 txg_wait_synced(spa->spa_dsl_pool, spa->spa_claim_max_txg);
2287 2515
2288 2516 /*
2289 2517 * If the config cache is stale, or we have uninitialized
2290 2518 * metaslabs (see spa_vdev_add()), then update the config.
2291 2519 *
2292 2520 * If this is a verbatim import, trust the current
2293 2521 * in-core spa_config and update the disk labels.
2294 2522 */
2295 2523 if (config_cache_txg != spa->spa_config_txg ||
2296 2524 state == SPA_LOAD_IMPORT ||
2297 2525 state == SPA_LOAD_RECOVER ||
2298 2526 (spa->spa_import_flags & ZFS_IMPORT_VERBATIM))
2299 2527 need_update = B_TRUE;
2300 2528
2301 2529 for (int c = 0; c < rvd->vdev_children; c++)
2302 2530 if (rvd->vdev_child[c]->vdev_ms_array == 0)
2303 2531 need_update = B_TRUE;
2304 2532
2305 2533 /*
2306 2534 * Update the config cache asychronously in case we're the
2307 2535 * root pool, in which case the config cache isn't writable yet.
2308 2536 */
2309 2537 if (need_update)
2310 2538 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
2311 2539
2312 2540 /*
2313 2541 * Check all DTLs to see if anything needs resilvering.
2314 2542 */
2315 2543 if (!dsl_scan_resilvering(spa->spa_dsl_pool) &&
2316 2544 vdev_resilver_needed(rvd, NULL, NULL))
2317 2545 spa_async_request(spa, SPA_ASYNC_RESILVER);
2318 2546
2319 2547 /*
2320 2548 * Delete any inconsistent datasets.
2321 2549 */
2322 2550 (void) dmu_objset_find(spa_name(spa),
2323 2551 dsl_destroy_inconsistent, NULL, DS_FIND_CHILDREN);
2324 2552
2325 2553 /*
2326 2554 * Clean up any stale temporary dataset userrefs.
2327 2555 */
2328 2556 dsl_pool_clean_tmp_userrefs(spa->spa_dsl_pool);
2329 2557 }
2330 2558
2331 2559 return (0);
2332 2560 }
2333 2561
2334 2562 static int
2335 2563 spa_load_retry(spa_t *spa, spa_load_state_t state, int mosconfig)
2336 2564 {
2337 2565 int mode = spa->spa_mode;
2338 2566
2339 2567 spa_unload(spa);
|
↓ open down ↓ |
89 lines elided |
↑ open up ↑ |
2340 2568 spa_deactivate(spa);
2341 2569
2342 2570 spa->spa_load_max_txg--;
2343 2571
2344 2572 spa_activate(spa, mode);
2345 2573 spa_async_suspend(spa);
2346 2574
2347 2575 return (spa_load(spa, state, SPA_IMPORT_EXISTING, mosconfig));
2348 2576 }
2349 2577
2578 +/*
2579 + * If spa_load() fails this function will try loading prior txg's. If
2580 + * 'state' is SPA_LOAD_RECOVER and one of these loads succeeds the pool
2581 + * will be rewound to that txg. If 'state' is not SPA_LOAD_RECOVER this
2582 + * function will not rewind the pool and will return the same error as
2583 + * spa_load().
2584 + */
2350 2585 static int
2351 2586 spa_load_best(spa_t *spa, spa_load_state_t state, int mosconfig,
2352 2587 uint64_t max_request, int rewind_flags)
2353 2588 {
2589 + nvlist_t *loadinfo = NULL;
2354 2590 nvlist_t *config = NULL;
2355 2591 int load_error, rewind_error;
2356 2592 uint64_t safe_rewind_txg;
2357 2593 uint64_t min_txg;
2358 2594
2359 2595 if (spa->spa_load_txg && state == SPA_LOAD_RECOVER) {
2360 2596 spa->spa_load_max_txg = spa->spa_load_txg;
2361 2597 spa_set_log_state(spa, SPA_LOG_CLEAR);
2362 2598 } else {
2363 2599 spa->spa_load_max_txg = max_request;
2364 2600 }
2365 2601
2366 2602 load_error = rewind_error = spa_load(spa, state, SPA_IMPORT_EXISTING,
2367 2603 mosconfig);
2368 2604 if (load_error == 0)
2369 2605 return (0);
2370 2606
2371 2607 if (spa->spa_root_vdev != NULL)
|
↓ open down ↓ |
8 lines elided |
↑ open up ↑ |
2372 2608 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
2373 2609
2374 2610 spa->spa_last_ubsync_txg = spa->spa_uberblock.ub_txg;
2375 2611 spa->spa_last_ubsync_txg_ts = spa->spa_uberblock.ub_timestamp;
2376 2612
2377 2613 if (rewind_flags & ZPOOL_NEVER_REWIND) {
2378 2614 nvlist_free(config);
2379 2615 return (load_error);
2380 2616 }
2381 2617
2382 - /* Price of rolling back is discarding txgs, including log */
2383 - if (state == SPA_LOAD_RECOVER)
2618 + if (state == SPA_LOAD_RECOVER) {
2619 + /* Price of rolling back is discarding txgs, including log */
2384 2620 spa_set_log_state(spa, SPA_LOG_CLEAR);
2621 + } else {
2622 + /*
2623 + * If we aren't rolling back save the load info from our first
2624 + * import attempt so that we can restore it after attempting
2625 + * to rewind.
2626 + */
2627 + loadinfo = spa->spa_load_info;
2628 + spa->spa_load_info = fnvlist_alloc();
2629 + }
2385 2630
2386 2631 spa->spa_load_max_txg = spa->spa_last_ubsync_txg;
2387 2632 safe_rewind_txg = spa->spa_last_ubsync_txg - TXG_DEFER_SIZE;
2388 2633 min_txg = (rewind_flags & ZPOOL_EXTREME_REWIND) ?
2389 2634 TXG_INITIAL : safe_rewind_txg;
2390 2635
2391 2636 /*
2392 2637 * Continue as long as we're finding errors, we're still within
2393 2638 * the acceptable rewind range, and we're still finding uberblocks
2394 2639 */
2395 2640 while (rewind_error && spa->spa_uberblock.ub_txg >= min_txg &&
2396 2641 spa->spa_uberblock.ub_txg <= spa->spa_load_max_txg) {
2397 2642 if (spa->spa_load_max_txg < safe_rewind_txg)
|
↓ open down ↓ |
3 lines elided |
↑ open up ↑ |
2398 2643 spa->spa_extreme_rewind = B_TRUE;
2399 2644 rewind_error = spa_load_retry(spa, state, mosconfig);
2400 2645 }
2401 2646
2402 2647 spa->spa_extreme_rewind = B_FALSE;
2403 2648 spa->spa_load_max_txg = UINT64_MAX;
2404 2649
2405 2650 if (config && (rewind_error || state != SPA_LOAD_RECOVER))
2406 2651 spa_config_set(spa, config);
2407 2652
2408 - return (state == SPA_LOAD_RECOVER ? rewind_error : load_error);
2653 + if (state == SPA_LOAD_RECOVER) {
2654 + ASSERT3P(loadinfo, ==, NULL);
2655 + return (rewind_error);
2656 + } else {
2657 + /* Store the rewind info as part of the initial load info */
2658 + fnvlist_add_nvlist(loadinfo, ZPOOL_CONFIG_REWIND_INFO,
2659 + spa->spa_load_info);
2660 +
2661 + /* Restore the initial load info */
2662 + fnvlist_free(spa->spa_load_info);
2663 + spa->spa_load_info = loadinfo;
2664 +
2665 + return (load_error);
2666 + }
2409 2667 }
2410 2668
2411 2669 /*
2412 2670 * Pool Open/Import
2413 2671 *
2414 2672 * The import case is identical to an open except that the configuration is sent
2415 2673 * down from userland, instead of grabbed from the configuration cache. For the
2416 2674 * case of an open, the pool configuration will exist in the
2417 2675 * POOL_STATE_UNINITIALIZED state.
2418 2676 *
2419 2677 * The stats information (gen/count/ustats) is used to gather vdev statistics at
2420 2678 * the same time open the pool, without having to keep around the spa_t in some
2421 2679 * ambiguous state.
2422 2680 */
2423 2681 static int
2424 2682 spa_open_common(const char *pool, spa_t **spapp, void *tag, nvlist_t *nvpolicy,
2425 2683 nvlist_t **config)
2426 2684 {
2427 2685 spa_t *spa;
2428 2686 spa_load_state_t state = SPA_LOAD_OPEN;
2429 2687 int error;
2430 2688 int locked = B_FALSE;
2431 2689
2432 2690 *spapp = NULL;
2433 2691
2434 2692 /*
2435 2693 * As disgusting as this is, we need to support recursive calls to this
2436 2694 * function because dsl_dir_open() is called during spa_load(), and ends
2437 2695 * up calling spa_open() again. The real fix is to figure out how to
2438 2696 * avoid dsl_dir_open() calling this in the first place.
2439 2697 */
2440 2698 if (mutex_owner(&spa_namespace_lock) != curthread) {
2441 2699 mutex_enter(&spa_namespace_lock);
2442 2700 locked = B_TRUE;
2443 2701 }
2444 2702
2445 2703 if ((spa = spa_lookup(pool)) == NULL) {
2446 2704 if (locked)
2447 2705 mutex_exit(&spa_namespace_lock);
2448 2706 return (ENOENT);
2449 2707 }
2450 2708
2451 2709 if (spa->spa_state == POOL_STATE_UNINITIALIZED) {
2452 2710 zpool_rewind_policy_t policy;
2453 2711
2454 2712 zpool_get_rewind_policy(nvpolicy ? nvpolicy : spa->spa_config,
2455 2713 &policy);
2456 2714 if (policy.zrp_request & ZPOOL_DO_REWIND)
2457 2715 state = SPA_LOAD_RECOVER;
2458 2716
2459 2717 spa_activate(spa, spa_mode_global);
2460 2718
2461 2719 if (state != SPA_LOAD_RECOVER)
2462 2720 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
2463 2721
2464 2722 error = spa_load_best(spa, state, B_FALSE, policy.zrp_txg,
2465 2723 policy.zrp_request);
2466 2724
2467 2725 if (error == EBADF) {
2468 2726 /*
2469 2727 * If vdev_validate() returns failure (indicated by
2470 2728 * EBADF), it indicates that one of the vdevs indicates
2471 2729 * that the pool has been exported or destroyed. If
2472 2730 * this is the case, the config cache is out of sync and
2473 2731 * we should remove the pool from the namespace.
2474 2732 */
2475 2733 spa_unload(spa);
2476 2734 spa_deactivate(spa);
2477 2735 spa_config_sync(spa, B_TRUE, B_TRUE);
2478 2736 spa_remove(spa);
2479 2737 if (locked)
2480 2738 mutex_exit(&spa_namespace_lock);
2481 2739 return (ENOENT);
2482 2740 }
2483 2741
2484 2742 if (error) {
2485 2743 /*
2486 2744 * We can't open the pool, but we still have useful
2487 2745 * information: the state of each vdev after the
2488 2746 * attempted vdev_open(). Return this to the user.
2489 2747 */
2490 2748 if (config != NULL && spa->spa_config) {
2491 2749 VERIFY(nvlist_dup(spa->spa_config, config,
2492 2750 KM_SLEEP) == 0);
2493 2751 VERIFY(nvlist_add_nvlist(*config,
2494 2752 ZPOOL_CONFIG_LOAD_INFO,
2495 2753 spa->spa_load_info) == 0);
2496 2754 }
2497 2755 spa_unload(spa);
2498 2756 spa_deactivate(spa);
2499 2757 spa->spa_last_open_failed = error;
2500 2758 if (locked)
2501 2759 mutex_exit(&spa_namespace_lock);
2502 2760 *spapp = NULL;
2503 2761 return (error);
2504 2762 }
2505 2763 }
2506 2764
2507 2765 spa_open_ref(spa, tag);
2508 2766
2509 2767 if (config != NULL)
2510 2768 *config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
2511 2769
2512 2770 /*
2513 2771 * If we've recovered the pool, pass back any information we
2514 2772 * gathered while doing the load.
2515 2773 */
2516 2774 if (state == SPA_LOAD_RECOVER) {
2517 2775 VERIFY(nvlist_add_nvlist(*config, ZPOOL_CONFIG_LOAD_INFO,
2518 2776 spa->spa_load_info) == 0);
2519 2777 }
2520 2778
2521 2779 if (locked) {
2522 2780 spa->spa_last_open_failed = 0;
2523 2781 spa->spa_last_ubsync_txg = 0;
2524 2782 spa->spa_load_txg = 0;
2525 2783 mutex_exit(&spa_namespace_lock);
2526 2784 }
2527 2785
2528 2786 *spapp = spa;
2529 2787
2530 2788 return (0);
2531 2789 }
2532 2790
2533 2791 int
2534 2792 spa_open_rewind(const char *name, spa_t **spapp, void *tag, nvlist_t *policy,
2535 2793 nvlist_t **config)
2536 2794 {
2537 2795 return (spa_open_common(name, spapp, tag, policy, config));
2538 2796 }
2539 2797
2540 2798 int
2541 2799 spa_open(const char *name, spa_t **spapp, void *tag)
2542 2800 {
2543 2801 return (spa_open_common(name, spapp, tag, NULL, NULL));
2544 2802 }
2545 2803
2546 2804 /*
2547 2805 * Lookup the given spa_t, incrementing the inject count in the process,
2548 2806 * preventing it from being exported or destroyed.
2549 2807 */
2550 2808 spa_t *
2551 2809 spa_inject_addref(char *name)
2552 2810 {
2553 2811 spa_t *spa;
2554 2812
2555 2813 mutex_enter(&spa_namespace_lock);
2556 2814 if ((spa = spa_lookup(name)) == NULL) {
2557 2815 mutex_exit(&spa_namespace_lock);
2558 2816 return (NULL);
2559 2817 }
2560 2818 spa->spa_inject_ref++;
2561 2819 mutex_exit(&spa_namespace_lock);
2562 2820
2563 2821 return (spa);
2564 2822 }
2565 2823
2566 2824 void
2567 2825 spa_inject_delref(spa_t *spa)
2568 2826 {
2569 2827 mutex_enter(&spa_namespace_lock);
2570 2828 spa->spa_inject_ref--;
2571 2829 mutex_exit(&spa_namespace_lock);
2572 2830 }
2573 2831
2574 2832 /*
2575 2833 * Add spares device information to the nvlist.
2576 2834 */
2577 2835 static void
2578 2836 spa_add_spares(spa_t *spa, nvlist_t *config)
2579 2837 {
2580 2838 nvlist_t **spares;
2581 2839 uint_t i, nspares;
2582 2840 nvlist_t *nvroot;
2583 2841 uint64_t guid;
2584 2842 vdev_stat_t *vs;
2585 2843 uint_t vsc;
2586 2844 uint64_t pool;
2587 2845
2588 2846 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
2589 2847
2590 2848 if (spa->spa_spares.sav_count == 0)
2591 2849 return;
2592 2850
2593 2851 VERIFY(nvlist_lookup_nvlist(config,
2594 2852 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
2595 2853 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
2596 2854 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
2597 2855 if (nspares != 0) {
2598 2856 VERIFY(nvlist_add_nvlist_array(nvroot,
2599 2857 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
2600 2858 VERIFY(nvlist_lookup_nvlist_array(nvroot,
2601 2859 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
2602 2860
2603 2861 /*
2604 2862 * Go through and find any spares which have since been
2605 2863 * repurposed as an active spare. If this is the case, update
2606 2864 * their status appropriately.
2607 2865 */
2608 2866 for (i = 0; i < nspares; i++) {
2609 2867 VERIFY(nvlist_lookup_uint64(spares[i],
2610 2868 ZPOOL_CONFIG_GUID, &guid) == 0);
2611 2869 if (spa_spare_exists(guid, &pool, NULL) &&
2612 2870 pool != 0ULL) {
2613 2871 VERIFY(nvlist_lookup_uint64_array(
2614 2872 spares[i], ZPOOL_CONFIG_VDEV_STATS,
2615 2873 (uint64_t **)&vs, &vsc) == 0);
2616 2874 vs->vs_state = VDEV_STATE_CANT_OPEN;
2617 2875 vs->vs_aux = VDEV_AUX_SPARED;
2618 2876 }
2619 2877 }
2620 2878 }
2621 2879 }
2622 2880
2623 2881 /*
2624 2882 * Add l2cache device information to the nvlist, including vdev stats.
2625 2883 */
2626 2884 static void
2627 2885 spa_add_l2cache(spa_t *spa, nvlist_t *config)
2628 2886 {
2629 2887 nvlist_t **l2cache;
2630 2888 uint_t i, j, nl2cache;
2631 2889 nvlist_t *nvroot;
2632 2890 uint64_t guid;
2633 2891 vdev_t *vd;
2634 2892 vdev_stat_t *vs;
2635 2893 uint_t vsc;
2636 2894
2637 2895 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
2638 2896
2639 2897 if (spa->spa_l2cache.sav_count == 0)
2640 2898 return;
2641 2899
2642 2900 VERIFY(nvlist_lookup_nvlist(config,
2643 2901 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
2644 2902 VERIFY(nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
2645 2903 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
2646 2904 if (nl2cache != 0) {
2647 2905 VERIFY(nvlist_add_nvlist_array(nvroot,
2648 2906 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
2649 2907 VERIFY(nvlist_lookup_nvlist_array(nvroot,
2650 2908 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
2651 2909
2652 2910 /*
2653 2911 * Update level 2 cache device stats.
2654 2912 */
2655 2913
2656 2914 for (i = 0; i < nl2cache; i++) {
2657 2915 VERIFY(nvlist_lookup_uint64(l2cache[i],
2658 2916 ZPOOL_CONFIG_GUID, &guid) == 0);
2659 2917
2660 2918 vd = NULL;
2661 2919 for (j = 0; j < spa->spa_l2cache.sav_count; j++) {
2662 2920 if (guid ==
2663 2921 spa->spa_l2cache.sav_vdevs[j]->vdev_guid) {
2664 2922 vd = spa->spa_l2cache.sav_vdevs[j];
2665 2923 break;
2666 2924 }
2667 2925 }
|
↓ open down ↓ |
249 lines elided |
↑ open up ↑ |
2668 2926 ASSERT(vd != NULL);
2669 2927
2670 2928 VERIFY(nvlist_lookup_uint64_array(l2cache[i],
2671 2929 ZPOOL_CONFIG_VDEV_STATS, (uint64_t **)&vs, &vsc)
2672 2930 == 0);
2673 2931 vdev_get_stats(vd, vs);
2674 2932 }
2675 2933 }
2676 2934 }
2677 2935
2936 +static void
2937 +spa_add_feature_stats(spa_t *spa, nvlist_t *config)
2938 +{
2939 + nvlist_t *features;
2940 + zap_cursor_t zc;
2941 + zap_attribute_t za;
2942 +
2943 + ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
2944 + VERIFY(nvlist_alloc(&features, NV_UNIQUE_NAME, KM_SLEEP) == 0);
2945 +
2946 + if (spa->spa_feat_for_read_obj != 0) {
2947 + for (zap_cursor_init(&zc, spa->spa_meta_objset,
2948 + spa->spa_feat_for_read_obj);
2949 + zap_cursor_retrieve(&zc, &za) == 0;
2950 + zap_cursor_advance(&zc)) {
2951 + ASSERT(za.za_integer_length == sizeof (uint64_t) &&
2952 + za.za_num_integers == 1);
2953 + VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name,
2954 + za.za_first_integer));
2955 + }
2956 + zap_cursor_fini(&zc);
2957 + }
2958 +
2959 + if (spa->spa_feat_for_write_obj != 0) {
2960 + for (zap_cursor_init(&zc, spa->spa_meta_objset,
2961 + spa->spa_feat_for_write_obj);
2962 + zap_cursor_retrieve(&zc, &za) == 0;
2963 + zap_cursor_advance(&zc)) {
2964 + ASSERT(za.za_integer_length == sizeof (uint64_t) &&
2965 + za.za_num_integers == 1);
2966 + VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name,
2967 + za.za_first_integer));
2968 + }
2969 + zap_cursor_fini(&zc);
2970 + }
2971 +
2972 + VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_FEATURE_STATS,
2973 + features) == 0);
2974 + nvlist_free(features);
2975 +}
2976 +
2678 2977 int
2679 -spa_get_stats(const char *name, nvlist_t **config, char *altroot, size_t buflen)
2978 +spa_get_stats(const char *name, nvlist_t **config,
2979 + char *altroot, size_t buflen)
2680 2980 {
2681 2981 int error;
2682 2982 spa_t *spa;
2683 2983
2684 2984 *config = NULL;
2685 2985 error = spa_open_common(name, &spa, FTAG, NULL, config);
2686 2986
2687 2987 if (spa != NULL) {
2688 2988 /*
2689 2989 * This still leaves a window of inconsistency where the spares
2690 2990 * or l2cache devices could change and the config would be
2691 2991 * self-inconsistent.
2692 2992 */
2693 2993 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
2694 2994
2695 2995 if (*config != NULL) {
2696 2996 uint64_t loadtimes[2];
2697 2997
2698 2998 loadtimes[0] = spa->spa_loaded_ts.tv_sec;
2699 2999 loadtimes[1] = spa->spa_loaded_ts.tv_nsec;
2700 3000 VERIFY(nvlist_add_uint64_array(*config,
2701 3001 ZPOOL_CONFIG_LOADED_TIME, loadtimes, 2) == 0);
2702 3002
2703 3003 VERIFY(nvlist_add_uint64(*config,
|
↓ open down ↓ |
14 lines elided |
↑ open up ↑ |
2704 3004 ZPOOL_CONFIG_ERRCOUNT,
2705 3005 spa_get_errlog_size(spa)) == 0);
2706 3006
2707 3007 if (spa_suspended(spa))
2708 3008 VERIFY(nvlist_add_uint64(*config,
2709 3009 ZPOOL_CONFIG_SUSPENDED,
2710 3010 spa->spa_failmode) == 0);
2711 3011
2712 3012 spa_add_spares(spa, *config);
2713 3013 spa_add_l2cache(spa, *config);
3014 + spa_add_feature_stats(spa, *config);
2714 3015 }
2715 3016 }
2716 3017
2717 3018 /*
2718 3019 * We want to get the alternate root even for faulted pools, so we cheat
2719 3020 * and call spa_lookup() directly.
2720 3021 */
2721 3022 if (altroot) {
2722 3023 if (spa == NULL) {
2723 3024 mutex_enter(&spa_namespace_lock);
2724 3025 spa = spa_lookup(name);
2725 3026 if (spa)
2726 3027 spa_altroot(spa, altroot, buflen);
2727 3028 else
2728 3029 altroot[0] = '\0';
2729 3030 spa = NULL;
2730 3031 mutex_exit(&spa_namespace_lock);
2731 3032 } else {
2732 3033 spa_altroot(spa, altroot, buflen);
2733 3034 }
2734 3035 }
2735 3036
2736 3037 if (spa != NULL) {
2737 3038 spa_config_exit(spa, SCL_CONFIG, FTAG);
2738 3039 spa_close(spa, FTAG);
2739 3040 }
2740 3041
2741 3042 return (error);
2742 3043 }
2743 3044
2744 3045 /*
2745 3046 * Validate that the auxiliary device array is well formed. We must have an
2746 3047 * array of nvlists, each which describes a valid leaf vdev. If this is an
2747 3048 * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be
2748 3049 * specified, as long as they are well-formed.
2749 3050 */
2750 3051 static int
2751 3052 spa_validate_aux_devs(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode,
2752 3053 spa_aux_vdev_t *sav, const char *config, uint64_t version,
2753 3054 vdev_labeltype_t label)
2754 3055 {
2755 3056 nvlist_t **dev;
2756 3057 uint_t i, ndev;
2757 3058 vdev_t *vd;
2758 3059 int error;
2759 3060
2760 3061 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
2761 3062
2762 3063 /*
2763 3064 * It's acceptable to have no devs specified.
2764 3065 */
2765 3066 if (nvlist_lookup_nvlist_array(nvroot, config, &dev, &ndev) != 0)
2766 3067 return (0);
2767 3068
2768 3069 if (ndev == 0)
2769 3070 return (EINVAL);
2770 3071
2771 3072 /*
2772 3073 * Make sure the pool is formatted with a version that supports this
2773 3074 * device type.
2774 3075 */
2775 3076 if (spa_version(spa) < version)
2776 3077 return (ENOTSUP);
2777 3078
2778 3079 /*
2779 3080 * Set the pending device list so we correctly handle device in-use
2780 3081 * checking.
2781 3082 */
2782 3083 sav->sav_pending = dev;
2783 3084 sav->sav_npending = ndev;
2784 3085
2785 3086 for (i = 0; i < ndev; i++) {
2786 3087 if ((error = spa_config_parse(spa, &vd, dev[i], NULL, 0,
2787 3088 mode)) != 0)
2788 3089 goto out;
2789 3090
2790 3091 if (!vd->vdev_ops->vdev_op_leaf) {
2791 3092 vdev_free(vd);
2792 3093 error = EINVAL;
2793 3094 goto out;
2794 3095 }
2795 3096
2796 3097 /*
2797 3098 * The L2ARC currently only supports disk devices in
2798 3099 * kernel context. For user-level testing, we allow it.
2799 3100 */
2800 3101 #ifdef _KERNEL
2801 3102 if ((strcmp(config, ZPOOL_CONFIG_L2CACHE) == 0) &&
2802 3103 strcmp(vd->vdev_ops->vdev_op_type, VDEV_TYPE_DISK) != 0) {
2803 3104 error = ENOTBLK;
2804 3105 vdev_free(vd);
2805 3106 goto out;
2806 3107 }
2807 3108 #endif
2808 3109 vd->vdev_top = vd;
2809 3110
2810 3111 if ((error = vdev_open(vd)) == 0 &&
2811 3112 (error = vdev_label_init(vd, crtxg, label)) == 0) {
2812 3113 VERIFY(nvlist_add_uint64(dev[i], ZPOOL_CONFIG_GUID,
2813 3114 vd->vdev_guid) == 0);
2814 3115 }
2815 3116
2816 3117 vdev_free(vd);
2817 3118
2818 3119 if (error &&
2819 3120 (mode != VDEV_ALLOC_SPARE && mode != VDEV_ALLOC_L2CACHE))
2820 3121 goto out;
2821 3122 else
2822 3123 error = 0;
2823 3124 }
2824 3125
2825 3126 out:
2826 3127 sav->sav_pending = NULL;
2827 3128 sav->sav_npending = 0;
2828 3129 return (error);
2829 3130 }
2830 3131
2831 3132 static int
2832 3133 spa_validate_aux(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode)
2833 3134 {
2834 3135 int error;
2835 3136
2836 3137 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
2837 3138
2838 3139 if ((error = spa_validate_aux_devs(spa, nvroot, crtxg, mode,
2839 3140 &spa->spa_spares, ZPOOL_CONFIG_SPARES, SPA_VERSION_SPARES,
2840 3141 VDEV_LABEL_SPARE)) != 0) {
2841 3142 return (error);
2842 3143 }
2843 3144
2844 3145 return (spa_validate_aux_devs(spa, nvroot, crtxg, mode,
2845 3146 &spa->spa_l2cache, ZPOOL_CONFIG_L2CACHE, SPA_VERSION_L2CACHE,
2846 3147 VDEV_LABEL_L2CACHE));
2847 3148 }
2848 3149
2849 3150 static void
2850 3151 spa_set_aux_vdevs(spa_aux_vdev_t *sav, nvlist_t **devs, int ndevs,
2851 3152 const char *config)
2852 3153 {
2853 3154 int i;
2854 3155
2855 3156 if (sav->sav_config != NULL) {
2856 3157 nvlist_t **olddevs;
2857 3158 uint_t oldndevs;
2858 3159 nvlist_t **newdevs;
2859 3160
2860 3161 /*
2861 3162 * Generate new dev list by concatentating with the
2862 3163 * current dev list.
2863 3164 */
2864 3165 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, config,
2865 3166 &olddevs, &oldndevs) == 0);
2866 3167
2867 3168 newdevs = kmem_alloc(sizeof (void *) *
2868 3169 (ndevs + oldndevs), KM_SLEEP);
2869 3170 for (i = 0; i < oldndevs; i++)
2870 3171 VERIFY(nvlist_dup(olddevs[i], &newdevs[i],
2871 3172 KM_SLEEP) == 0);
2872 3173 for (i = 0; i < ndevs; i++)
2873 3174 VERIFY(nvlist_dup(devs[i], &newdevs[i + oldndevs],
2874 3175 KM_SLEEP) == 0);
2875 3176
2876 3177 VERIFY(nvlist_remove(sav->sav_config, config,
2877 3178 DATA_TYPE_NVLIST_ARRAY) == 0);
2878 3179
2879 3180 VERIFY(nvlist_add_nvlist_array(sav->sav_config,
2880 3181 config, newdevs, ndevs + oldndevs) == 0);
2881 3182 for (i = 0; i < oldndevs + ndevs; i++)
2882 3183 nvlist_free(newdevs[i]);
2883 3184 kmem_free(newdevs, (oldndevs + ndevs) * sizeof (void *));
2884 3185 } else {
2885 3186 /*
2886 3187 * Generate a new dev list.
2887 3188 */
2888 3189 VERIFY(nvlist_alloc(&sav->sav_config, NV_UNIQUE_NAME,
2889 3190 KM_SLEEP) == 0);
2890 3191 VERIFY(nvlist_add_nvlist_array(sav->sav_config, config,
2891 3192 devs, ndevs) == 0);
2892 3193 }
2893 3194 }
2894 3195
2895 3196 /*
2896 3197 * Stop and drop level 2 ARC devices
2897 3198 */
2898 3199 void
2899 3200 spa_l2cache_drop(spa_t *spa)
2900 3201 {
2901 3202 vdev_t *vd;
2902 3203 int i;
2903 3204 spa_aux_vdev_t *sav = &spa->spa_l2cache;
2904 3205
2905 3206 for (i = 0; i < sav->sav_count; i++) {
2906 3207 uint64_t pool;
2907 3208
2908 3209 vd = sav->sav_vdevs[i];
2909 3210 ASSERT(vd != NULL);
2910 3211
2911 3212 if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
2912 3213 pool != 0ULL && l2arc_vdev_present(vd))
2913 3214 l2arc_remove_vdev(vd);
2914 3215 }
2915 3216 }
2916 3217
2917 3218 /*
2918 3219 * Pool Creation
2919 3220 */
2920 3221 int
2921 3222 spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props,
2922 3223 const char *history_str, nvlist_t *zplprops)
2923 3224 {
|
↓ open down ↓ |
200 lines elided |
↑ open up ↑ |
2924 3225 spa_t *spa;
2925 3226 char *altroot = NULL;
2926 3227 vdev_t *rvd;
2927 3228 dsl_pool_t *dp;
2928 3229 dmu_tx_t *tx;
2929 3230 int error = 0;
2930 3231 uint64_t txg = TXG_INITIAL;
2931 3232 nvlist_t **spares, **l2cache;
2932 3233 uint_t nspares, nl2cache;
2933 3234 uint64_t version, obj;
3235 + boolean_t has_features;
2934 3236
2935 3237 /*
2936 3238 * If this pool already exists, return failure.
2937 3239 */
2938 3240 mutex_enter(&spa_namespace_lock);
2939 3241 if (spa_lookup(pool) != NULL) {
2940 3242 mutex_exit(&spa_namespace_lock);
2941 3243 return (EEXIST);
2942 3244 }
2943 3245
2944 3246 /*
2945 3247 * Allocate a new spa_t structure.
2946 3248 */
2947 3249 (void) nvlist_lookup_string(props,
2948 3250 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
|
↓ open down ↓ |
5 lines elided |
↑ open up ↑ |
2949 3251 spa = spa_add(pool, NULL, altroot);
2950 3252 spa_activate(spa, spa_mode_global);
2951 3253
2952 3254 if (props && (error = spa_prop_validate(spa, props))) {
2953 3255 spa_deactivate(spa);
2954 3256 spa_remove(spa);
2955 3257 mutex_exit(&spa_namespace_lock);
2956 3258 return (error);
2957 3259 }
2958 3260
2959 - if (nvlist_lookup_uint64(props, zpool_prop_to_name(ZPOOL_PROP_VERSION),
2960 - &version) != 0)
3261 + has_features = B_FALSE;
3262 + for (nvpair_t *elem = nvlist_next_nvpair(props, NULL);
3263 + elem != NULL; elem = nvlist_next_nvpair(props, elem)) {
3264 + if (zpool_prop_feature(nvpair_name(elem)))
3265 + has_features = B_TRUE;
3266 + }
3267 +
3268 + if (has_features || nvlist_lookup_uint64(props,
3269 + zpool_prop_to_name(ZPOOL_PROP_VERSION), &version) != 0) {
2961 3270 version = SPA_VERSION;
2962 - ASSERT(version <= SPA_VERSION);
3271 + }
3272 + ASSERT(SPA_VERSION_IS_SUPPORTED(version));
2963 3273
2964 3274 spa->spa_first_txg = txg;
2965 3275 spa->spa_uberblock.ub_txg = txg - 1;
2966 3276 spa->spa_uberblock.ub_version = version;
2967 3277 spa->spa_ubsync = spa->spa_uberblock;
2968 3278
2969 3279 /*
2970 3280 * Create "The Godfather" zio to hold all async IOs
2971 3281 */
2972 3282 spa->spa_async_zio_root = zio_root(spa, NULL, NULL,
2973 3283 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_GODFATHER);
2974 3284
2975 3285 /*
2976 3286 * Create the root vdev.
2977 3287 */
2978 3288 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2979 3289
2980 3290 error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_ADD);
2981 3291
2982 3292 ASSERT(error != 0 || rvd != NULL);
2983 3293 ASSERT(error != 0 || spa->spa_root_vdev == rvd);
2984 3294
2985 3295 if (error == 0 && !zfs_allocatable_devs(nvroot))
2986 3296 error = EINVAL;
2987 3297
2988 3298 if (error == 0 &&
2989 3299 (error = vdev_create(rvd, txg, B_FALSE)) == 0 &&
2990 3300 (error = spa_validate_aux(spa, nvroot, txg,
2991 3301 VDEV_ALLOC_ADD)) == 0) {
2992 3302 for (int c = 0; c < rvd->vdev_children; c++) {
2993 3303 vdev_metaslab_set_size(rvd->vdev_child[c]);
2994 3304 vdev_expand(rvd->vdev_child[c], txg);
2995 3305 }
2996 3306 }
2997 3307
2998 3308 spa_config_exit(spa, SCL_ALL, FTAG);
2999 3309
3000 3310 if (error != 0) {
3001 3311 spa_unload(spa);
3002 3312 spa_deactivate(spa);
3003 3313 spa_remove(spa);
3004 3314 mutex_exit(&spa_namespace_lock);
3005 3315 return (error);
3006 3316 }
3007 3317
3008 3318 /*
3009 3319 * Get the list of spares, if specified.
3010 3320 */
3011 3321 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
3012 3322 &spares, &nspares) == 0) {
3013 3323 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, NV_UNIQUE_NAME,
3014 3324 KM_SLEEP) == 0);
3015 3325 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
3016 3326 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
3017 3327 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3018 3328 spa_load_spares(spa);
3019 3329 spa_config_exit(spa, SCL_ALL, FTAG);
3020 3330 spa->spa_spares.sav_sync = B_TRUE;
3021 3331 }
3022 3332
3023 3333 /*
3024 3334 * Get the list of level 2 cache devices, if specified.
3025 3335 */
3026 3336 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
3027 3337 &l2cache, &nl2cache) == 0) {
|
↓ open down ↓ |
55 lines elided |
↑ open up ↑ |
3028 3338 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
3029 3339 NV_UNIQUE_NAME, KM_SLEEP) == 0);
3030 3340 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
3031 3341 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3032 3342 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3033 3343 spa_load_l2cache(spa);
3034 3344 spa_config_exit(spa, SCL_ALL, FTAG);
3035 3345 spa->spa_l2cache.sav_sync = B_TRUE;
3036 3346 }
3037 3347
3348 + spa->spa_is_initializing = B_TRUE;
3038 3349 spa->spa_dsl_pool = dp = dsl_pool_create(spa, zplprops, txg);
3039 3350 spa->spa_meta_objset = dp->dp_meta_objset;
3351 + spa->spa_is_initializing = B_FALSE;
3040 3352
3041 3353 /*
3042 3354 * Create DDTs (dedup tables).
3043 3355 */
3044 3356 ddt_create(spa);
3045 3357
3046 3358 spa_update_dspace(spa);
3047 3359
3048 3360 tx = dmu_tx_create_assigned(dp, txg);
3049 3361
3050 3362 /*
3051 3363 * Create the pool config object.
3052 3364 */
|
↓ open down ↓ |
3 lines elided |
↑ open up ↑ |
3053 3365 spa->spa_config_object = dmu_object_alloc(spa->spa_meta_objset,
3054 3366 DMU_OT_PACKED_NVLIST, SPA_CONFIG_BLOCKSIZE,
3055 3367 DMU_OT_PACKED_NVLIST_SIZE, sizeof (uint64_t), tx);
3056 3368
3057 3369 if (zap_add(spa->spa_meta_objset,
3058 3370 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG,
3059 3371 sizeof (uint64_t), 1, &spa->spa_config_object, tx) != 0) {
3060 3372 cmn_err(CE_PANIC, "failed to add pool config");
3061 3373 }
3062 3374
3375 + if (spa_version(spa) >= SPA_VERSION_FEATURES)
3376 + spa_feature_create_zap_objects(spa, tx);
3377 +
3063 3378 if (zap_add(spa->spa_meta_objset,
3064 3379 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CREATION_VERSION,
3065 3380 sizeof (uint64_t), 1, &version, tx) != 0) {
3066 3381 cmn_err(CE_PANIC, "failed to add pool version");
3067 3382 }
3068 3383
3069 3384 /* Newly created pools with the right version are always deflated. */
3070 3385 if (version >= SPA_VERSION_RAIDZ_DEFLATE) {
3071 3386 spa->spa_deflate = TRUE;
3072 3387 if (zap_add(spa->spa_meta_objset,
3073 3388 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
3074 3389 sizeof (uint64_t), 1, &spa->spa_deflate, tx) != 0) {
3075 3390 cmn_err(CE_PANIC, "failed to add deflate");
3076 3391 }
3077 3392 }
3078 3393
3079 3394 /*
3080 3395 * Create the deferred-free bpobj. Turn off compression
3081 3396 * because sync-to-convergence takes longer if the blocksize
3082 3397 * keeps changing.
3083 3398 */
3084 3399 obj = bpobj_alloc(spa->spa_meta_objset, 1 << 14, tx);
3085 3400 dmu_object_set_compress(spa->spa_meta_objset, obj,
3086 3401 ZIO_COMPRESS_OFF, tx);
3087 3402 if (zap_add(spa->spa_meta_objset,
3088 3403 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPOBJ,
3089 3404 sizeof (uint64_t), 1, &obj, tx) != 0) {
3090 3405 cmn_err(CE_PANIC, "failed to add bpobj");
3091 3406 }
3092 3407 VERIFY3U(0, ==, bpobj_open(&spa->spa_deferred_bpobj,
3093 3408 spa->spa_meta_objset, obj));
3094 3409
3095 3410 /*
3096 3411 * Create the pool's history object.
3097 3412 */
3098 3413 if (version >= SPA_VERSION_ZPOOL_HISTORY)
3099 3414 spa_history_create_obj(spa, tx);
3100 3415
3101 3416 /*
3102 3417 * Set pool properties.
3103 3418 */
3104 3419 spa->spa_bootfs = zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS);
3105 3420 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
3106 3421 spa->spa_failmode = zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE);
3107 3422 spa->spa_autoexpand = zpool_prop_default_numeric(ZPOOL_PROP_AUTOEXPAND);
3108 3423
3109 3424 if (props != NULL) {
3110 3425 spa_configfile_set(spa, props, B_FALSE);
3111 3426 spa_sync_props(spa, props, tx);
3112 3427 }
3113 3428
3114 3429 dmu_tx_commit(tx);
3115 3430
3116 3431 spa->spa_sync_on = B_TRUE;
3117 3432 txg_sync_start(spa->spa_dsl_pool);
3118 3433
3119 3434 /*
3120 3435 * We explicitly wait for the first transaction to complete so that our
3121 3436 * bean counters are appropriately updated.
3122 3437 */
3123 3438 txg_wait_synced(spa->spa_dsl_pool, txg);
3124 3439
3125 3440 spa_config_sync(spa, B_FALSE, B_TRUE);
3126 3441
3127 3442 if (version >= SPA_VERSION_ZPOOL_HISTORY && history_str != NULL)
3128 3443 (void) spa_history_log(spa, history_str, LOG_CMD_POOL_CREATE);
3129 3444 spa_history_log_version(spa, LOG_POOL_CREATE);
3130 3445
3131 3446 spa->spa_minref = refcount_count(&spa->spa_refcount);
3132 3447
3133 3448 mutex_exit(&spa_namespace_lock);
3134 3449
3135 3450 return (0);
3136 3451 }
3137 3452
3138 3453 #ifdef _KERNEL
3139 3454 /*
3140 3455 * Get the root pool information from the root disk, then import the root pool
3141 3456 * during the system boot up time.
3142 3457 */
3143 3458 extern int vdev_disk_read_rootlabel(char *, char *, nvlist_t **);
3144 3459
3145 3460 static nvlist_t *
3146 3461 spa_generate_rootconf(char *devpath, char *devid, uint64_t *guid)
3147 3462 {
3148 3463 nvlist_t *config;
3149 3464 nvlist_t *nvtop, *nvroot;
3150 3465 uint64_t pgid;
3151 3466
3152 3467 if (vdev_disk_read_rootlabel(devpath, devid, &config) != 0)
3153 3468 return (NULL);
3154 3469
3155 3470 /*
3156 3471 * Add this top-level vdev to the child array.
3157 3472 */
3158 3473 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3159 3474 &nvtop) == 0);
3160 3475 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
3161 3476 &pgid) == 0);
3162 3477 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID, guid) == 0);
3163 3478
3164 3479 /*
3165 3480 * Put this pool's top-level vdevs into a root vdev.
3166 3481 */
3167 3482 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
3168 3483 VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
3169 3484 VDEV_TYPE_ROOT) == 0);
3170 3485 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0);
3171 3486 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0);
3172 3487 VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
3173 3488 &nvtop, 1) == 0);
3174 3489
3175 3490 /*
3176 3491 * Replace the existing vdev_tree with the new root vdev in
3177 3492 * this pool's configuration (remove the old, add the new).
3178 3493 */
3179 3494 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0);
3180 3495 nvlist_free(nvroot);
3181 3496 return (config);
3182 3497 }
3183 3498
3184 3499 /*
3185 3500 * Walk the vdev tree and see if we can find a device with "better"
3186 3501 * configuration. A configuration is "better" if the label on that
3187 3502 * device has a more recent txg.
3188 3503 */
3189 3504 static void
3190 3505 spa_alt_rootvdev(vdev_t *vd, vdev_t **avd, uint64_t *txg)
3191 3506 {
3192 3507 for (int c = 0; c < vd->vdev_children; c++)
3193 3508 spa_alt_rootvdev(vd->vdev_child[c], avd, txg);
3194 3509
3195 3510 if (vd->vdev_ops->vdev_op_leaf) {
3196 3511 nvlist_t *label;
3197 3512 uint64_t label_txg;
3198 3513
3199 3514 if (vdev_disk_read_rootlabel(vd->vdev_physpath, vd->vdev_devid,
3200 3515 &label) != 0)
3201 3516 return;
3202 3517
3203 3518 VERIFY(nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG,
3204 3519 &label_txg) == 0);
3205 3520
3206 3521 /*
3207 3522 * Do we have a better boot device?
3208 3523 */
3209 3524 if (label_txg > *txg) {
3210 3525 *txg = label_txg;
3211 3526 *avd = vd;
3212 3527 }
3213 3528 nvlist_free(label);
3214 3529 }
3215 3530 }
3216 3531
3217 3532 /*
3218 3533 * Import a root pool.
3219 3534 *
3220 3535 * For x86. devpath_list will consist of devid and/or physpath name of
3221 3536 * the vdev (e.g. "id1,sd@SSEAGATE..." or "/pci@1f,0/ide@d/disk@0,0:a").
3222 3537 * The GRUB "findroot" command will return the vdev we should boot.
3223 3538 *
3224 3539 * For Sparc, devpath_list consists the physpath name of the booting device
3225 3540 * no matter the rootpool is a single device pool or a mirrored pool.
3226 3541 * e.g.
3227 3542 * "/pci@1f,0/ide@d/disk@0,0:a"
3228 3543 */
3229 3544 int
3230 3545 spa_import_rootpool(char *devpath, char *devid)
3231 3546 {
3232 3547 spa_t *spa;
3233 3548 vdev_t *rvd, *bvd, *avd = NULL;
3234 3549 nvlist_t *config, *nvtop;
3235 3550 uint64_t guid, txg;
3236 3551 char *pname;
3237 3552 int error;
3238 3553
3239 3554 /*
3240 3555 * Read the label from the boot device and generate a configuration.
3241 3556 */
3242 3557 config = spa_generate_rootconf(devpath, devid, &guid);
|
↓ open down ↓ |
170 lines elided |
↑ open up ↑ |
3243 3558 #if defined(_OBP) && defined(_KERNEL)
3244 3559 if (config == NULL) {
3245 3560 if (strstr(devpath, "/iscsi/ssd") != NULL) {
3246 3561 /* iscsi boot */
3247 3562 get_iscsi_bootpath_phy(devpath);
3248 3563 config = spa_generate_rootconf(devpath, devid, &guid);
3249 3564 }
3250 3565 }
3251 3566 #endif
3252 3567 if (config == NULL) {
3253 - cmn_err(CE_NOTE, "Can not read the pool label from '%s'",
3568 + cmn_err(CE_NOTE, "Cannot read the pool label from '%s'",
3254 3569 devpath);
3255 3570 return (EIO);
3256 3571 }
3257 3572
3258 3573 VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
3259 3574 &pname) == 0);
3260 3575 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg) == 0);
3261 3576
3262 3577 mutex_enter(&spa_namespace_lock);
3263 3578 if ((spa = spa_lookup(pname)) != NULL) {
3264 3579 /*
3265 3580 * Remove the existing root pool from the namespace so that we
3266 3581 * can replace it with the correct config we just read in.
3267 3582 */
3268 3583 spa_remove(spa);
3269 3584 }
3270 3585
3271 3586 spa = spa_add(pname, config, NULL);
3272 3587 spa->spa_is_root = B_TRUE;
3273 3588 spa->spa_import_flags = ZFS_IMPORT_VERBATIM;
3274 3589
3275 3590 /*
3276 3591 * Build up a vdev tree based on the boot device's label config.
3277 3592 */
3278 3593 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3279 3594 &nvtop) == 0);
3280 3595 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3281 3596 error = spa_config_parse(spa, &rvd, nvtop, NULL, 0,
3282 3597 VDEV_ALLOC_ROOTPOOL);
3283 3598 spa_config_exit(spa, SCL_ALL, FTAG);
3284 3599 if (error) {
3285 3600 mutex_exit(&spa_namespace_lock);
3286 3601 nvlist_free(config);
3287 3602 cmn_err(CE_NOTE, "Can not parse the config for pool '%s'",
3288 3603 pname);
3289 3604 return (error);
3290 3605 }
3291 3606
3292 3607 /*
3293 3608 * Get the boot vdev.
3294 3609 */
3295 3610 if ((bvd = vdev_lookup_by_guid(rvd, guid)) == NULL) {
3296 3611 cmn_err(CE_NOTE, "Can not find the boot vdev for guid %llu",
3297 3612 (u_longlong_t)guid);
3298 3613 error = ENOENT;
3299 3614 goto out;
3300 3615 }
3301 3616
3302 3617 /*
3303 3618 * Determine if there is a better boot device.
3304 3619 */
3305 3620 avd = bvd;
3306 3621 spa_alt_rootvdev(rvd, &avd, &txg);
3307 3622 if (avd != bvd) {
3308 3623 cmn_err(CE_NOTE, "The boot device is 'degraded'. Please "
3309 3624 "try booting from '%s'", avd->vdev_path);
3310 3625 error = EINVAL;
3311 3626 goto out;
3312 3627 }
3313 3628
3314 3629 /*
3315 3630 * If the boot device is part of a spare vdev then ensure that
3316 3631 * we're booting off the active spare.
3317 3632 */
3318 3633 if (bvd->vdev_parent->vdev_ops == &vdev_spare_ops &&
3319 3634 !bvd->vdev_isspare) {
3320 3635 cmn_err(CE_NOTE, "The boot device is currently spared. Please "
3321 3636 "try booting from '%s'",
3322 3637 bvd->vdev_parent->
3323 3638 vdev_child[bvd->vdev_parent->vdev_children - 1]->vdev_path);
3324 3639 error = EINVAL;
3325 3640 goto out;
3326 3641 }
3327 3642
3328 3643 error = 0;
3329 3644 spa_history_log_version(spa, LOG_POOL_IMPORT);
3330 3645 out:
3331 3646 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3332 3647 vdev_free(rvd);
3333 3648 spa_config_exit(spa, SCL_ALL, FTAG);
3334 3649 mutex_exit(&spa_namespace_lock);
3335 3650
3336 3651 nvlist_free(config);
3337 3652 return (error);
3338 3653 }
3339 3654
3340 3655 #endif
3341 3656
3342 3657 /*
3343 3658 * Import a non-root pool into the system.
3344 3659 */
3345 3660 int
3346 3661 spa_import(const char *pool, nvlist_t *config, nvlist_t *props, uint64_t flags)
3347 3662 {
3348 3663 spa_t *spa;
3349 3664 char *altroot = NULL;
3350 3665 spa_load_state_t state = SPA_LOAD_IMPORT;
3351 3666 zpool_rewind_policy_t policy;
3352 3667 uint64_t mode = spa_mode_global;
3353 3668 uint64_t readonly = B_FALSE;
3354 3669 int error;
3355 3670 nvlist_t *nvroot;
3356 3671 nvlist_t **spares, **l2cache;
3357 3672 uint_t nspares, nl2cache;
3358 3673
3359 3674 /*
3360 3675 * If a pool with this name exists, return failure.
3361 3676 */
3362 3677 mutex_enter(&spa_namespace_lock);
3363 3678 if (spa_lookup(pool) != NULL) {
3364 3679 mutex_exit(&spa_namespace_lock);
3365 3680 return (EEXIST);
3366 3681 }
3367 3682
3368 3683 /*
3369 3684 * Create and initialize the spa structure.
3370 3685 */
3371 3686 (void) nvlist_lookup_string(props,
3372 3687 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
3373 3688 (void) nvlist_lookup_uint64(props,
3374 3689 zpool_prop_to_name(ZPOOL_PROP_READONLY), &readonly);
3375 3690 if (readonly)
3376 3691 mode = FREAD;
3377 3692 spa = spa_add(pool, config, altroot);
3378 3693 spa->spa_import_flags = flags;
3379 3694
3380 3695 /*
3381 3696 * Verbatim import - Take a pool and insert it into the namespace
3382 3697 * as if it had been loaded at boot.
3383 3698 */
3384 3699 if (spa->spa_import_flags & ZFS_IMPORT_VERBATIM) {
3385 3700 if (props != NULL)
3386 3701 spa_configfile_set(spa, props, B_FALSE);
3387 3702
3388 3703 spa_config_sync(spa, B_FALSE, B_TRUE);
3389 3704
3390 3705 mutex_exit(&spa_namespace_lock);
3391 3706 spa_history_log_version(spa, LOG_POOL_IMPORT);
3392 3707
3393 3708 return (0);
3394 3709 }
3395 3710
3396 3711 spa_activate(spa, mode);
3397 3712
3398 3713 /*
3399 3714 * Don't start async tasks until we know everything is healthy.
3400 3715 */
3401 3716 spa_async_suspend(spa);
3402 3717
3403 3718 zpool_get_rewind_policy(config, &policy);
3404 3719 if (policy.zrp_request & ZPOOL_DO_REWIND)
3405 3720 state = SPA_LOAD_RECOVER;
3406 3721
3407 3722 /*
3408 3723 * Pass off the heavy lifting to spa_load(). Pass TRUE for mosconfig
3409 3724 * because the user-supplied config is actually the one to trust when
3410 3725 * doing an import.
3411 3726 */
3412 3727 if (state != SPA_LOAD_RECOVER)
3413 3728 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
3414 3729
3415 3730 error = spa_load_best(spa, state, B_TRUE, policy.zrp_txg,
3416 3731 policy.zrp_request);
3417 3732
3418 3733 /*
3419 3734 * Propagate anything learned while loading the pool and pass it
3420 3735 * back to caller (i.e. rewind info, missing devices, etc).
3421 3736 */
3422 3737 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO,
3423 3738 spa->spa_load_info) == 0);
3424 3739
3425 3740 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3426 3741 /*
3427 3742 * Toss any existing sparelist, as it doesn't have any validity
3428 3743 * anymore, and conflicts with spa_has_spare().
3429 3744 */
3430 3745 if (spa->spa_spares.sav_config) {
3431 3746 nvlist_free(spa->spa_spares.sav_config);
3432 3747 spa->spa_spares.sav_config = NULL;
3433 3748 spa_load_spares(spa);
3434 3749 }
3435 3750 if (spa->spa_l2cache.sav_config) {
3436 3751 nvlist_free(spa->spa_l2cache.sav_config);
3437 3752 spa->spa_l2cache.sav_config = NULL;
3438 3753 spa_load_l2cache(spa);
3439 3754 }
3440 3755
3441 3756 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3442 3757 &nvroot) == 0);
3443 3758 if (error == 0)
3444 3759 error = spa_validate_aux(spa, nvroot, -1ULL,
3445 3760 VDEV_ALLOC_SPARE);
3446 3761 if (error == 0)
3447 3762 error = spa_validate_aux(spa, nvroot, -1ULL,
3448 3763 VDEV_ALLOC_L2CACHE);
3449 3764 spa_config_exit(spa, SCL_ALL, FTAG);
3450 3765
3451 3766 if (props != NULL)
3452 3767 spa_configfile_set(spa, props, B_FALSE);
3453 3768
3454 3769 if (error != 0 || (props && spa_writeable(spa) &&
3455 3770 (error = spa_prop_set(spa, props)))) {
3456 3771 spa_unload(spa);
3457 3772 spa_deactivate(spa);
3458 3773 spa_remove(spa);
3459 3774 mutex_exit(&spa_namespace_lock);
3460 3775 return (error);
3461 3776 }
3462 3777
3463 3778 spa_async_resume(spa);
3464 3779
3465 3780 /*
3466 3781 * Override any spares and level 2 cache devices as specified by
3467 3782 * the user, as these may have correct device names/devids, etc.
3468 3783 */
3469 3784 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
3470 3785 &spares, &nspares) == 0) {
3471 3786 if (spa->spa_spares.sav_config)
3472 3787 VERIFY(nvlist_remove(spa->spa_spares.sav_config,
3473 3788 ZPOOL_CONFIG_SPARES, DATA_TYPE_NVLIST_ARRAY) == 0);
3474 3789 else
3475 3790 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config,
3476 3791 NV_UNIQUE_NAME, KM_SLEEP) == 0);
3477 3792 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
3478 3793 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
3479 3794 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3480 3795 spa_load_spares(spa);
3481 3796 spa_config_exit(spa, SCL_ALL, FTAG);
3482 3797 spa->spa_spares.sav_sync = B_TRUE;
3483 3798 }
3484 3799 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
3485 3800 &l2cache, &nl2cache) == 0) {
3486 3801 if (spa->spa_l2cache.sav_config)
3487 3802 VERIFY(nvlist_remove(spa->spa_l2cache.sav_config,
3488 3803 ZPOOL_CONFIG_L2CACHE, DATA_TYPE_NVLIST_ARRAY) == 0);
3489 3804 else
3490 3805 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
3491 3806 NV_UNIQUE_NAME, KM_SLEEP) == 0);
3492 3807 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
3493 3808 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3494 3809 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3495 3810 spa_load_l2cache(spa);
3496 3811 spa_config_exit(spa, SCL_ALL, FTAG);
3497 3812 spa->spa_l2cache.sav_sync = B_TRUE;
3498 3813 }
3499 3814
3500 3815 /*
3501 3816 * Check for any removed devices.
3502 3817 */
3503 3818 if (spa->spa_autoreplace) {
3504 3819 spa_aux_check_removed(&spa->spa_spares);
3505 3820 spa_aux_check_removed(&spa->spa_l2cache);
3506 3821 }
3507 3822
3508 3823 if (spa_writeable(spa)) {
3509 3824 /*
3510 3825 * Update the config cache to include the newly-imported pool.
3511 3826 */
3512 3827 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
3513 3828 }
3514 3829
3515 3830 /*
3516 3831 * It's possible that the pool was expanded while it was exported.
3517 3832 * We kick off an async task to handle this for us.
3518 3833 */
3519 3834 spa_async_request(spa, SPA_ASYNC_AUTOEXPAND);
3520 3835
3521 3836 mutex_exit(&spa_namespace_lock);
3522 3837 spa_history_log_version(spa, LOG_POOL_IMPORT);
3523 3838
3524 3839 return (0);
3525 3840 }
3526 3841
3527 3842 nvlist_t *
3528 3843 spa_tryimport(nvlist_t *tryconfig)
3529 3844 {
3530 3845 nvlist_t *config = NULL;
3531 3846 char *poolname;
3532 3847 spa_t *spa;
3533 3848 uint64_t state;
3534 3849 int error;
3535 3850
3536 3851 if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_POOL_NAME, &poolname))
3537 3852 return (NULL);
3538 3853
3539 3854 if (nvlist_lookup_uint64(tryconfig, ZPOOL_CONFIG_POOL_STATE, &state))
3540 3855 return (NULL);
3541 3856
3542 3857 /*
3543 3858 * Create and initialize the spa structure.
3544 3859 */
3545 3860 mutex_enter(&spa_namespace_lock);
3546 3861 spa = spa_add(TRYIMPORT_NAME, tryconfig, NULL);
3547 3862 spa_activate(spa, FREAD);
3548 3863
3549 3864 /*
3550 3865 * Pass off the heavy lifting to spa_load().
3551 3866 * Pass TRUE for mosconfig because the user-supplied config
3552 3867 * is actually the one to trust when doing an import.
3553 3868 */
3554 3869 error = spa_load(spa, SPA_LOAD_TRYIMPORT, SPA_IMPORT_EXISTING, B_TRUE);
3555 3870
3556 3871 /*
|
↓ open down ↓ |
293 lines elided |
↑ open up ↑ |
3557 3872 * If 'tryconfig' was at least parsable, return the current config.
3558 3873 */
3559 3874 if (spa->spa_root_vdev != NULL) {
3560 3875 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
3561 3876 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME,
3562 3877 poolname) == 0);
3563 3878 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
3564 3879 state) == 0);
3565 3880 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TIMESTAMP,
3566 3881 spa->spa_uberblock.ub_timestamp) == 0);
3882 + VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO,
3883 + spa->spa_load_info) == 0);
3567 3884
3568 3885 /*
3569 3886 * If the bootfs property exists on this pool then we
3570 3887 * copy it out so that external consumers can tell which
3571 3888 * pools are bootable.
3572 3889 */
3573 3890 if ((!error || error == EEXIST) && spa->spa_bootfs) {
3574 3891 char *tmpname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
3575 3892
3576 3893 /*
3577 3894 * We have to play games with the name since the
3578 3895 * pool was opened as TRYIMPORT_NAME.
3579 3896 */
3580 3897 if (dsl_dsobj_to_dsname(spa_name(spa),
3581 3898 spa->spa_bootfs, tmpname) == 0) {
3582 3899 char *cp;
3583 3900 char *dsname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
3584 3901
3585 3902 cp = strchr(tmpname, '/');
3586 3903 if (cp == NULL) {
3587 3904 (void) strlcpy(dsname, tmpname,
3588 3905 MAXPATHLEN);
3589 3906 } else {
3590 3907 (void) snprintf(dsname, MAXPATHLEN,
3591 3908 "%s/%s", poolname, ++cp);
3592 3909 }
3593 3910 VERIFY(nvlist_add_string(config,
3594 3911 ZPOOL_CONFIG_BOOTFS, dsname) == 0);
3595 3912 kmem_free(dsname, MAXPATHLEN);
3596 3913 }
3597 3914 kmem_free(tmpname, MAXPATHLEN);
3598 3915 }
3599 3916
3600 3917 /*
3601 3918 * Add the list of hot spares and level 2 cache devices.
3602 3919 */
3603 3920 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
3604 3921 spa_add_spares(spa, config);
3605 3922 spa_add_l2cache(spa, config);
3606 3923 spa_config_exit(spa, SCL_CONFIG, FTAG);
3607 3924 }
3608 3925
3609 3926 spa_unload(spa);
3610 3927 spa_deactivate(spa);
3611 3928 spa_remove(spa);
3612 3929 mutex_exit(&spa_namespace_lock);
3613 3930
3614 3931 return (config);
3615 3932 }
3616 3933
3617 3934 /*
3618 3935 * Pool export/destroy
3619 3936 *
3620 3937 * The act of destroying or exporting a pool is very simple. We make sure there
3621 3938 * is no more pending I/O and any references to the pool are gone. Then, we
3622 3939 * update the pool state and sync all the labels to disk, removing the
3623 3940 * configuration from the cache afterwards. If the 'hardforce' flag is set, then
3624 3941 * we don't sync the labels or remove the configuration cache.
3625 3942 */
3626 3943 static int
3627 3944 spa_export_common(char *pool, int new_state, nvlist_t **oldconfig,
3628 3945 boolean_t force, boolean_t hardforce)
3629 3946 {
3630 3947 spa_t *spa;
3631 3948
3632 3949 if (oldconfig)
3633 3950 *oldconfig = NULL;
3634 3951
3635 3952 if (!(spa_mode_global & FWRITE))
3636 3953 return (EROFS);
3637 3954
3638 3955 mutex_enter(&spa_namespace_lock);
3639 3956 if ((spa = spa_lookup(pool)) == NULL) {
3640 3957 mutex_exit(&spa_namespace_lock);
3641 3958 return (ENOENT);
3642 3959 }
3643 3960
3644 3961 /*
3645 3962 * Put a hold on the pool, drop the namespace lock, stop async tasks,
3646 3963 * reacquire the namespace lock, and see if we can export.
3647 3964 */
3648 3965 spa_open_ref(spa, FTAG);
3649 3966 mutex_exit(&spa_namespace_lock);
3650 3967 spa_async_suspend(spa);
3651 3968 mutex_enter(&spa_namespace_lock);
3652 3969 spa_close(spa, FTAG);
3653 3970
3654 3971 /*
3655 3972 * The pool will be in core if it's openable,
3656 3973 * in which case we can modify its state.
3657 3974 */
3658 3975 if (spa->spa_state != POOL_STATE_UNINITIALIZED && spa->spa_sync_on) {
3659 3976 /*
3660 3977 * Objsets may be open only because they're dirty, so we
3661 3978 * have to force it to sync before checking spa_refcnt.
3662 3979 */
3663 3980 txg_wait_synced(spa->spa_dsl_pool, 0);
3664 3981
3665 3982 /*
3666 3983 * A pool cannot be exported or destroyed if there are active
3667 3984 * references. If we are resetting a pool, allow references by
3668 3985 * fault injection handlers.
3669 3986 */
3670 3987 if (!spa_refcount_zero(spa) ||
3671 3988 (spa->spa_inject_ref != 0 &&
3672 3989 new_state != POOL_STATE_UNINITIALIZED)) {
3673 3990 spa_async_resume(spa);
3674 3991 mutex_exit(&spa_namespace_lock);
3675 3992 return (EBUSY);
3676 3993 }
3677 3994
3678 3995 /*
3679 3996 * A pool cannot be exported if it has an active shared spare.
3680 3997 * This is to prevent other pools stealing the active spare
3681 3998 * from an exported pool. At user's own will, such pool can
3682 3999 * be forcedly exported.
3683 4000 */
3684 4001 if (!force && new_state == POOL_STATE_EXPORTED &&
3685 4002 spa_has_active_shared_spare(spa)) {
3686 4003 spa_async_resume(spa);
3687 4004 mutex_exit(&spa_namespace_lock);
3688 4005 return (EXDEV);
3689 4006 }
3690 4007
3691 4008 /*
3692 4009 * We want this to be reflected on every label,
3693 4010 * so mark them all dirty. spa_unload() will do the
3694 4011 * final sync that pushes these changes out.
3695 4012 */
3696 4013 if (new_state != POOL_STATE_UNINITIALIZED && !hardforce) {
3697 4014 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3698 4015 spa->spa_state = new_state;
3699 4016 spa->spa_final_txg = spa_last_synced_txg(spa) +
3700 4017 TXG_DEFER_SIZE + 1;
3701 4018 vdev_config_dirty(spa->spa_root_vdev);
3702 4019 spa_config_exit(spa, SCL_ALL, FTAG);
3703 4020 }
3704 4021 }
3705 4022
3706 4023 spa_event_notify(spa, NULL, ESC_ZFS_POOL_DESTROY);
3707 4024
3708 4025 if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
3709 4026 spa_unload(spa);
3710 4027 spa_deactivate(spa);
3711 4028 }
3712 4029
3713 4030 if (oldconfig && spa->spa_config)
3714 4031 VERIFY(nvlist_dup(spa->spa_config, oldconfig, 0) == 0);
3715 4032
3716 4033 if (new_state != POOL_STATE_UNINITIALIZED) {
3717 4034 if (!hardforce)
3718 4035 spa_config_sync(spa, B_TRUE, B_TRUE);
3719 4036 spa_remove(spa);
3720 4037 }
3721 4038 mutex_exit(&spa_namespace_lock);
3722 4039
3723 4040 return (0);
3724 4041 }
3725 4042
3726 4043 /*
3727 4044 * Destroy a storage pool.
3728 4045 */
3729 4046 int
3730 4047 spa_destroy(char *pool)
3731 4048 {
3732 4049 return (spa_export_common(pool, POOL_STATE_DESTROYED, NULL,
3733 4050 B_FALSE, B_FALSE));
3734 4051 }
3735 4052
3736 4053 /*
3737 4054 * Export a storage pool.
3738 4055 */
3739 4056 int
3740 4057 spa_export(char *pool, nvlist_t **oldconfig, boolean_t force,
3741 4058 boolean_t hardforce)
3742 4059 {
3743 4060 return (spa_export_common(pool, POOL_STATE_EXPORTED, oldconfig,
3744 4061 force, hardforce));
3745 4062 }
3746 4063
3747 4064 /*
3748 4065 * Similar to spa_export(), this unloads the spa_t without actually removing it
3749 4066 * from the namespace in any way.
3750 4067 */
3751 4068 int
3752 4069 spa_reset(char *pool)
3753 4070 {
3754 4071 return (spa_export_common(pool, POOL_STATE_UNINITIALIZED, NULL,
3755 4072 B_FALSE, B_FALSE));
3756 4073 }
3757 4074
3758 4075 /*
3759 4076 * ==========================================================================
3760 4077 * Device manipulation
3761 4078 * ==========================================================================
3762 4079 */
3763 4080
3764 4081 /*
3765 4082 * Add a device to a storage pool.
3766 4083 */
3767 4084 int
3768 4085 spa_vdev_add(spa_t *spa, nvlist_t *nvroot)
3769 4086 {
3770 4087 uint64_t txg, id;
3771 4088 int error;
3772 4089 vdev_t *rvd = spa->spa_root_vdev;
3773 4090 vdev_t *vd, *tvd;
3774 4091 nvlist_t **spares, **l2cache;
3775 4092 uint_t nspares, nl2cache;
3776 4093
3777 4094 ASSERT(spa_writeable(spa));
3778 4095
3779 4096 txg = spa_vdev_enter(spa);
3780 4097
3781 4098 if ((error = spa_config_parse(spa, &vd, nvroot, NULL, 0,
3782 4099 VDEV_ALLOC_ADD)) != 0)
3783 4100 return (spa_vdev_exit(spa, NULL, txg, error));
3784 4101
3785 4102 spa->spa_pending_vdev = vd; /* spa_vdev_exit() will clear this */
3786 4103
3787 4104 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, &spares,
3788 4105 &nspares) != 0)
3789 4106 nspares = 0;
3790 4107
3791 4108 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, &l2cache,
3792 4109 &nl2cache) != 0)
3793 4110 nl2cache = 0;
3794 4111
3795 4112 if (vd->vdev_children == 0 && nspares == 0 && nl2cache == 0)
3796 4113 return (spa_vdev_exit(spa, vd, txg, EINVAL));
3797 4114
3798 4115 if (vd->vdev_children != 0 &&
3799 4116 (error = vdev_create(vd, txg, B_FALSE)) != 0)
3800 4117 return (spa_vdev_exit(spa, vd, txg, error));
3801 4118
3802 4119 /*
3803 4120 * We must validate the spares and l2cache devices after checking the
3804 4121 * children. Otherwise, vdev_inuse() will blindly overwrite the spare.
3805 4122 */
3806 4123 if ((error = spa_validate_aux(spa, nvroot, txg, VDEV_ALLOC_ADD)) != 0)
3807 4124 return (spa_vdev_exit(spa, vd, txg, error));
3808 4125
3809 4126 /*
3810 4127 * Transfer each new top-level vdev from vd to rvd.
3811 4128 */
3812 4129 for (int c = 0; c < vd->vdev_children; c++) {
3813 4130
3814 4131 /*
3815 4132 * Set the vdev id to the first hole, if one exists.
3816 4133 */
3817 4134 for (id = 0; id < rvd->vdev_children; id++) {
3818 4135 if (rvd->vdev_child[id]->vdev_ishole) {
3819 4136 vdev_free(rvd->vdev_child[id]);
3820 4137 break;
3821 4138 }
3822 4139 }
3823 4140 tvd = vd->vdev_child[c];
3824 4141 vdev_remove_child(vd, tvd);
3825 4142 tvd->vdev_id = id;
3826 4143 vdev_add_child(rvd, tvd);
3827 4144 vdev_config_dirty(tvd);
3828 4145 }
3829 4146
3830 4147 if (nspares != 0) {
3831 4148 spa_set_aux_vdevs(&spa->spa_spares, spares, nspares,
3832 4149 ZPOOL_CONFIG_SPARES);
3833 4150 spa_load_spares(spa);
3834 4151 spa->spa_spares.sav_sync = B_TRUE;
3835 4152 }
3836 4153
3837 4154 if (nl2cache != 0) {
3838 4155 spa_set_aux_vdevs(&spa->spa_l2cache, l2cache, nl2cache,
3839 4156 ZPOOL_CONFIG_L2CACHE);
3840 4157 spa_load_l2cache(spa);
3841 4158 spa->spa_l2cache.sav_sync = B_TRUE;
3842 4159 }
3843 4160
3844 4161 /*
3845 4162 * We have to be careful when adding new vdevs to an existing pool.
3846 4163 * If other threads start allocating from these vdevs before we
3847 4164 * sync the config cache, and we lose power, then upon reboot we may
3848 4165 * fail to open the pool because there are DVAs that the config cache
3849 4166 * can't translate. Therefore, we first add the vdevs without
3850 4167 * initializing metaslabs; sync the config cache (via spa_vdev_exit());
3851 4168 * and then let spa_config_update() initialize the new metaslabs.
3852 4169 *
3853 4170 * spa_load() checks for added-but-not-initialized vdevs, so that
3854 4171 * if we lose power at any point in this sequence, the remaining
3855 4172 * steps will be completed the next time we load the pool.
3856 4173 */
3857 4174 (void) spa_vdev_exit(spa, vd, txg, 0);
3858 4175
3859 4176 mutex_enter(&spa_namespace_lock);
3860 4177 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
3861 4178 mutex_exit(&spa_namespace_lock);
3862 4179
3863 4180 return (0);
3864 4181 }
3865 4182
3866 4183 /*
3867 4184 * Attach a device to a mirror. The arguments are the path to any device
3868 4185 * in the mirror, and the nvroot for the new device. If the path specifies
3869 4186 * a device that is not mirrored, we automatically insert the mirror vdev.
3870 4187 *
3871 4188 * If 'replacing' is specified, the new device is intended to replace the
3872 4189 * existing device; in this case the two devices are made into their own
3873 4190 * mirror using the 'replacing' vdev, which is functionally identical to
3874 4191 * the mirror vdev (it actually reuses all the same ops) but has a few
3875 4192 * extra rules: you can't attach to it after it's been created, and upon
3876 4193 * completion of resilvering, the first disk (the one being replaced)
3877 4194 * is automatically detached.
3878 4195 */
3879 4196 int
3880 4197 spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot, int replacing)
3881 4198 {
3882 4199 uint64_t txg, dtl_max_txg;
3883 4200 vdev_t *rvd = spa->spa_root_vdev;
3884 4201 vdev_t *oldvd, *newvd, *newrootvd, *pvd, *tvd;
3885 4202 vdev_ops_t *pvops;
3886 4203 char *oldvdpath, *newvdpath;
3887 4204 int newvd_isspare;
3888 4205 int error;
3889 4206
3890 4207 ASSERT(spa_writeable(spa));
3891 4208
3892 4209 txg = spa_vdev_enter(spa);
3893 4210
3894 4211 oldvd = spa_lookup_by_guid(spa, guid, B_FALSE);
3895 4212
3896 4213 if (oldvd == NULL)
3897 4214 return (spa_vdev_exit(spa, NULL, txg, ENODEV));
3898 4215
3899 4216 if (!oldvd->vdev_ops->vdev_op_leaf)
3900 4217 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
3901 4218
3902 4219 pvd = oldvd->vdev_parent;
3903 4220
3904 4221 if ((error = spa_config_parse(spa, &newrootvd, nvroot, NULL, 0,
3905 4222 VDEV_ALLOC_ATTACH)) != 0)
3906 4223 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
3907 4224
3908 4225 if (newrootvd->vdev_children != 1)
3909 4226 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
3910 4227
3911 4228 newvd = newrootvd->vdev_child[0];
3912 4229
3913 4230 if (!newvd->vdev_ops->vdev_op_leaf)
3914 4231 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
3915 4232
3916 4233 if ((error = vdev_create(newrootvd, txg, replacing)) != 0)
3917 4234 return (spa_vdev_exit(spa, newrootvd, txg, error));
3918 4235
3919 4236 /*
3920 4237 * Spares can't replace logs
3921 4238 */
3922 4239 if (oldvd->vdev_top->vdev_islog && newvd->vdev_isspare)
3923 4240 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
3924 4241
3925 4242 if (!replacing) {
3926 4243 /*
3927 4244 * For attach, the only allowable parent is a mirror or the root
3928 4245 * vdev.
3929 4246 */
3930 4247 if (pvd->vdev_ops != &vdev_mirror_ops &&
3931 4248 pvd->vdev_ops != &vdev_root_ops)
3932 4249 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
3933 4250
3934 4251 pvops = &vdev_mirror_ops;
3935 4252 } else {
3936 4253 /*
3937 4254 * Active hot spares can only be replaced by inactive hot
3938 4255 * spares.
3939 4256 */
3940 4257 if (pvd->vdev_ops == &vdev_spare_ops &&
3941 4258 oldvd->vdev_isspare &&
3942 4259 !spa_has_spare(spa, newvd->vdev_guid))
3943 4260 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
3944 4261
3945 4262 /*
3946 4263 * If the source is a hot spare, and the parent isn't already a
3947 4264 * spare, then we want to create a new hot spare. Otherwise, we
3948 4265 * want to create a replacing vdev. The user is not allowed to
3949 4266 * attach to a spared vdev child unless the 'isspare' state is
3950 4267 * the same (spare replaces spare, non-spare replaces
3951 4268 * non-spare).
3952 4269 */
3953 4270 if (pvd->vdev_ops == &vdev_replacing_ops &&
3954 4271 spa_version(spa) < SPA_VERSION_MULTI_REPLACE) {
3955 4272 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
3956 4273 } else if (pvd->vdev_ops == &vdev_spare_ops &&
3957 4274 newvd->vdev_isspare != oldvd->vdev_isspare) {
3958 4275 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
3959 4276 }
3960 4277
3961 4278 if (newvd->vdev_isspare)
3962 4279 pvops = &vdev_spare_ops;
3963 4280 else
3964 4281 pvops = &vdev_replacing_ops;
3965 4282 }
3966 4283
3967 4284 /*
3968 4285 * Make sure the new device is big enough.
3969 4286 */
3970 4287 if (newvd->vdev_asize < vdev_get_min_asize(oldvd))
3971 4288 return (spa_vdev_exit(spa, newrootvd, txg, EOVERFLOW));
3972 4289
3973 4290 /*
3974 4291 * The new device cannot have a higher alignment requirement
3975 4292 * than the top-level vdev.
3976 4293 */
3977 4294 if (newvd->vdev_ashift > oldvd->vdev_top->vdev_ashift)
3978 4295 return (spa_vdev_exit(spa, newrootvd, txg, EDOM));
3979 4296
3980 4297 /*
3981 4298 * If this is an in-place replacement, update oldvd's path and devid
3982 4299 * to make it distinguishable from newvd, and unopenable from now on.
3983 4300 */
3984 4301 if (strcmp(oldvd->vdev_path, newvd->vdev_path) == 0) {
3985 4302 spa_strfree(oldvd->vdev_path);
3986 4303 oldvd->vdev_path = kmem_alloc(strlen(newvd->vdev_path) + 5,
3987 4304 KM_SLEEP);
3988 4305 (void) sprintf(oldvd->vdev_path, "%s/%s",
3989 4306 newvd->vdev_path, "old");
3990 4307 if (oldvd->vdev_devid != NULL) {
3991 4308 spa_strfree(oldvd->vdev_devid);
3992 4309 oldvd->vdev_devid = NULL;
3993 4310 }
3994 4311 }
3995 4312
3996 4313 /* mark the device being resilvered */
3997 4314 newvd->vdev_resilvering = B_TRUE;
3998 4315
3999 4316 /*
4000 4317 * If the parent is not a mirror, or if we're replacing, insert the new
4001 4318 * mirror/replacing/spare vdev above oldvd.
4002 4319 */
4003 4320 if (pvd->vdev_ops != pvops)
4004 4321 pvd = vdev_add_parent(oldvd, pvops);
4005 4322
4006 4323 ASSERT(pvd->vdev_top->vdev_parent == rvd);
4007 4324 ASSERT(pvd->vdev_ops == pvops);
4008 4325 ASSERT(oldvd->vdev_parent == pvd);
4009 4326
4010 4327 /*
4011 4328 * Extract the new device from its root and add it to pvd.
4012 4329 */
4013 4330 vdev_remove_child(newrootvd, newvd);
4014 4331 newvd->vdev_id = pvd->vdev_children;
4015 4332 newvd->vdev_crtxg = oldvd->vdev_crtxg;
4016 4333 vdev_add_child(pvd, newvd);
4017 4334
4018 4335 tvd = newvd->vdev_top;
4019 4336 ASSERT(pvd->vdev_top == tvd);
4020 4337 ASSERT(tvd->vdev_parent == rvd);
4021 4338
4022 4339 vdev_config_dirty(tvd);
4023 4340
4024 4341 /*
4025 4342 * Set newvd's DTL to [TXG_INITIAL, dtl_max_txg) so that we account
4026 4343 * for any dmu_sync-ed blocks. It will propagate upward when
4027 4344 * spa_vdev_exit() calls vdev_dtl_reassess().
4028 4345 */
4029 4346 dtl_max_txg = txg + TXG_CONCURRENT_STATES;
4030 4347
4031 4348 vdev_dtl_dirty(newvd, DTL_MISSING, TXG_INITIAL,
4032 4349 dtl_max_txg - TXG_INITIAL);
4033 4350
4034 4351 if (newvd->vdev_isspare) {
4035 4352 spa_spare_activate(newvd);
4036 4353 spa_event_notify(spa, newvd, ESC_ZFS_VDEV_SPARE);
4037 4354 }
4038 4355
4039 4356 oldvdpath = spa_strdup(oldvd->vdev_path);
4040 4357 newvdpath = spa_strdup(newvd->vdev_path);
4041 4358 newvd_isspare = newvd->vdev_isspare;
4042 4359
4043 4360 /*
4044 4361 * Mark newvd's DTL dirty in this txg.
4045 4362 */
4046 4363 vdev_dirty(tvd, VDD_DTL, newvd, txg);
4047 4364
4048 4365 /*
4049 4366 * Restart the resilver
4050 4367 */
4051 4368 dsl_resilver_restart(spa->spa_dsl_pool, dtl_max_txg);
4052 4369
4053 4370 /*
4054 4371 * Commit the config
4055 4372 */
4056 4373 (void) spa_vdev_exit(spa, newrootvd, dtl_max_txg, 0);
4057 4374
4058 4375 spa_history_log_internal(LOG_POOL_VDEV_ATTACH, spa, NULL,
4059 4376 "%s vdev=%s %s vdev=%s",
4060 4377 replacing && newvd_isspare ? "spare in" :
4061 4378 replacing ? "replace" : "attach", newvdpath,
4062 4379 replacing ? "for" : "to", oldvdpath);
4063 4380
4064 4381 spa_strfree(oldvdpath);
4065 4382 spa_strfree(newvdpath);
4066 4383
4067 4384 if (spa->spa_bootfs)
4068 4385 spa_event_notify(spa, newvd, ESC_ZFS_BOOTFS_VDEV_ATTACH);
4069 4386
4070 4387 return (0);
4071 4388 }
4072 4389
4073 4390 /*
4074 4391 * Detach a device from a mirror or replacing vdev.
4075 4392 * If 'replace_done' is specified, only detach if the parent
4076 4393 * is a replacing vdev.
4077 4394 */
4078 4395 int
4079 4396 spa_vdev_detach(spa_t *spa, uint64_t guid, uint64_t pguid, int replace_done)
4080 4397 {
4081 4398 uint64_t txg;
4082 4399 int error;
4083 4400 vdev_t *rvd = spa->spa_root_vdev;
4084 4401 vdev_t *vd, *pvd, *cvd, *tvd;
4085 4402 boolean_t unspare = B_FALSE;
4086 4403 uint64_t unspare_guid;
4087 4404 char *vdpath;
4088 4405
4089 4406 ASSERT(spa_writeable(spa));
4090 4407
4091 4408 txg = spa_vdev_enter(spa);
4092 4409
4093 4410 vd = spa_lookup_by_guid(spa, guid, B_FALSE);
4094 4411
4095 4412 if (vd == NULL)
4096 4413 return (spa_vdev_exit(spa, NULL, txg, ENODEV));
4097 4414
4098 4415 if (!vd->vdev_ops->vdev_op_leaf)
4099 4416 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4100 4417
4101 4418 pvd = vd->vdev_parent;
4102 4419
4103 4420 /*
4104 4421 * If the parent/child relationship is not as expected, don't do it.
4105 4422 * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing
4106 4423 * vdev that's replacing B with C. The user's intent in replacing
4107 4424 * is to go from M(A,B) to M(A,C). If the user decides to cancel
4108 4425 * the replace by detaching C, the expected behavior is to end up
4109 4426 * M(A,B). But suppose that right after deciding to detach C,
4110 4427 * the replacement of B completes. We would have M(A,C), and then
4111 4428 * ask to detach C, which would leave us with just A -- not what
4112 4429 * the user wanted. To prevent this, we make sure that the
4113 4430 * parent/child relationship hasn't changed -- in this example,
4114 4431 * that C's parent is still the replacing vdev R.
4115 4432 */
4116 4433 if (pvd->vdev_guid != pguid && pguid != 0)
4117 4434 return (spa_vdev_exit(spa, NULL, txg, EBUSY));
4118 4435
4119 4436 /*
4120 4437 * Only 'replacing' or 'spare' vdevs can be replaced.
4121 4438 */
4122 4439 if (replace_done && pvd->vdev_ops != &vdev_replacing_ops &&
4123 4440 pvd->vdev_ops != &vdev_spare_ops)
4124 4441 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4125 4442
4126 4443 ASSERT(pvd->vdev_ops != &vdev_spare_ops ||
4127 4444 spa_version(spa) >= SPA_VERSION_SPARES);
4128 4445
4129 4446 /*
4130 4447 * Only mirror, replacing, and spare vdevs support detach.
4131 4448 */
4132 4449 if (pvd->vdev_ops != &vdev_replacing_ops &&
4133 4450 pvd->vdev_ops != &vdev_mirror_ops &&
4134 4451 pvd->vdev_ops != &vdev_spare_ops)
4135 4452 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4136 4453
4137 4454 /*
4138 4455 * If this device has the only valid copy of some data,
4139 4456 * we cannot safely detach it.
4140 4457 */
4141 4458 if (vdev_dtl_required(vd))
4142 4459 return (spa_vdev_exit(spa, NULL, txg, EBUSY));
4143 4460
4144 4461 ASSERT(pvd->vdev_children >= 2);
4145 4462
4146 4463 /*
4147 4464 * If we are detaching the second disk from a replacing vdev, then
4148 4465 * check to see if we changed the original vdev's path to have "/old"
4149 4466 * at the end in spa_vdev_attach(). If so, undo that change now.
4150 4467 */
4151 4468 if (pvd->vdev_ops == &vdev_replacing_ops && vd->vdev_id > 0 &&
4152 4469 vd->vdev_path != NULL) {
4153 4470 size_t len = strlen(vd->vdev_path);
4154 4471
4155 4472 for (int c = 0; c < pvd->vdev_children; c++) {
4156 4473 cvd = pvd->vdev_child[c];
4157 4474
4158 4475 if (cvd == vd || cvd->vdev_path == NULL)
4159 4476 continue;
4160 4477
4161 4478 if (strncmp(cvd->vdev_path, vd->vdev_path, len) == 0 &&
4162 4479 strcmp(cvd->vdev_path + len, "/old") == 0) {
4163 4480 spa_strfree(cvd->vdev_path);
4164 4481 cvd->vdev_path = spa_strdup(vd->vdev_path);
4165 4482 break;
4166 4483 }
4167 4484 }
4168 4485 }
4169 4486
4170 4487 /*
4171 4488 * If we are detaching the original disk from a spare, then it implies
4172 4489 * that the spare should become a real disk, and be removed from the
4173 4490 * active spare list for the pool.
4174 4491 */
4175 4492 if (pvd->vdev_ops == &vdev_spare_ops &&
4176 4493 vd->vdev_id == 0 &&
4177 4494 pvd->vdev_child[pvd->vdev_children - 1]->vdev_isspare)
4178 4495 unspare = B_TRUE;
4179 4496
4180 4497 /*
4181 4498 * Erase the disk labels so the disk can be used for other things.
4182 4499 * This must be done after all other error cases are handled,
4183 4500 * but before we disembowel vd (so we can still do I/O to it).
4184 4501 * But if we can't do it, don't treat the error as fatal --
4185 4502 * it may be that the unwritability of the disk is the reason
4186 4503 * it's being detached!
4187 4504 */
4188 4505 error = vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
4189 4506
4190 4507 /*
4191 4508 * Remove vd from its parent and compact the parent's children.
4192 4509 */
4193 4510 vdev_remove_child(pvd, vd);
4194 4511 vdev_compact_children(pvd);
4195 4512
4196 4513 /*
4197 4514 * Remember one of the remaining children so we can get tvd below.
4198 4515 */
4199 4516 cvd = pvd->vdev_child[pvd->vdev_children - 1];
4200 4517
4201 4518 /*
4202 4519 * If we need to remove the remaining child from the list of hot spares,
4203 4520 * do it now, marking the vdev as no longer a spare in the process.
4204 4521 * We must do this before vdev_remove_parent(), because that can
4205 4522 * change the GUID if it creates a new toplevel GUID. For a similar
4206 4523 * reason, we must remove the spare now, in the same txg as the detach;
4207 4524 * otherwise someone could attach a new sibling, change the GUID, and
4208 4525 * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail.
4209 4526 */
4210 4527 if (unspare) {
4211 4528 ASSERT(cvd->vdev_isspare);
4212 4529 spa_spare_remove(cvd);
4213 4530 unspare_guid = cvd->vdev_guid;
4214 4531 (void) spa_vdev_remove(spa, unspare_guid, B_TRUE);
4215 4532 cvd->vdev_unspare = B_TRUE;
4216 4533 }
4217 4534
4218 4535 /*
4219 4536 * If the parent mirror/replacing vdev only has one child,
4220 4537 * the parent is no longer needed. Remove it from the tree.
4221 4538 */
4222 4539 if (pvd->vdev_children == 1) {
4223 4540 if (pvd->vdev_ops == &vdev_spare_ops)
4224 4541 cvd->vdev_unspare = B_FALSE;
4225 4542 vdev_remove_parent(cvd);
4226 4543 cvd->vdev_resilvering = B_FALSE;
4227 4544 }
4228 4545
4229 4546
4230 4547 /*
4231 4548 * We don't set tvd until now because the parent we just removed
4232 4549 * may have been the previous top-level vdev.
4233 4550 */
4234 4551 tvd = cvd->vdev_top;
4235 4552 ASSERT(tvd->vdev_parent == rvd);
4236 4553
4237 4554 /*
4238 4555 * Reevaluate the parent vdev state.
4239 4556 */
4240 4557 vdev_propagate_state(cvd);
4241 4558
4242 4559 /*
4243 4560 * If the 'autoexpand' property is set on the pool then automatically
4244 4561 * try to expand the size of the pool. For example if the device we
4245 4562 * just detached was smaller than the others, it may be possible to
4246 4563 * add metaslabs (i.e. grow the pool). We need to reopen the vdev
4247 4564 * first so that we can obtain the updated sizes of the leaf vdevs.
4248 4565 */
4249 4566 if (spa->spa_autoexpand) {
4250 4567 vdev_reopen(tvd);
4251 4568 vdev_expand(tvd, txg);
4252 4569 }
4253 4570
4254 4571 vdev_config_dirty(tvd);
4255 4572
4256 4573 /*
4257 4574 * Mark vd's DTL as dirty in this txg. vdev_dtl_sync() will see that
4258 4575 * vd->vdev_detached is set and free vd's DTL object in syncing context.
4259 4576 * But first make sure we're not on any *other* txg's DTL list, to
4260 4577 * prevent vd from being accessed after it's freed.
4261 4578 */
4262 4579 vdpath = spa_strdup(vd->vdev_path);
4263 4580 for (int t = 0; t < TXG_SIZE; t++)
4264 4581 (void) txg_list_remove_this(&tvd->vdev_dtl_list, vd, t);
4265 4582 vd->vdev_detached = B_TRUE;
4266 4583 vdev_dirty(tvd, VDD_DTL, vd, txg);
4267 4584
4268 4585 spa_event_notify(spa, vd, ESC_ZFS_VDEV_REMOVE);
4269 4586
4270 4587 /* hang on to the spa before we release the lock */
4271 4588 spa_open_ref(spa, FTAG);
4272 4589
4273 4590 error = spa_vdev_exit(spa, vd, txg, 0);
4274 4591
4275 4592 spa_history_log_internal(LOG_POOL_VDEV_DETACH, spa, NULL,
4276 4593 "vdev=%s", vdpath);
4277 4594 spa_strfree(vdpath);
4278 4595
4279 4596 /*
4280 4597 * If this was the removal of the original device in a hot spare vdev,
4281 4598 * then we want to go through and remove the device from the hot spare
4282 4599 * list of every other pool.
4283 4600 */
4284 4601 if (unspare) {
4285 4602 spa_t *altspa = NULL;
4286 4603
4287 4604 mutex_enter(&spa_namespace_lock);
4288 4605 while ((altspa = spa_next(altspa)) != NULL) {
4289 4606 if (altspa->spa_state != POOL_STATE_ACTIVE ||
4290 4607 altspa == spa)
4291 4608 continue;
4292 4609
4293 4610 spa_open_ref(altspa, FTAG);
4294 4611 mutex_exit(&spa_namespace_lock);
4295 4612 (void) spa_vdev_remove(altspa, unspare_guid, B_TRUE);
4296 4613 mutex_enter(&spa_namespace_lock);
4297 4614 spa_close(altspa, FTAG);
4298 4615 }
4299 4616 mutex_exit(&spa_namespace_lock);
4300 4617
4301 4618 /* search the rest of the vdevs for spares to remove */
4302 4619 spa_vdev_resilver_done(spa);
4303 4620 }
4304 4621
4305 4622 /* all done with the spa; OK to release */
4306 4623 mutex_enter(&spa_namespace_lock);
4307 4624 spa_close(spa, FTAG);
4308 4625 mutex_exit(&spa_namespace_lock);
4309 4626
4310 4627 return (error);
4311 4628 }
4312 4629
4313 4630 /*
4314 4631 * Split a set of devices from their mirrors, and create a new pool from them.
4315 4632 */
4316 4633 int
4317 4634 spa_vdev_split_mirror(spa_t *spa, char *newname, nvlist_t *config,
4318 4635 nvlist_t *props, boolean_t exp)
4319 4636 {
4320 4637 int error = 0;
4321 4638 uint64_t txg, *glist;
4322 4639 spa_t *newspa;
4323 4640 uint_t c, children, lastlog;
4324 4641 nvlist_t **child, *nvl, *tmp;
4325 4642 dmu_tx_t *tx;
4326 4643 char *altroot = NULL;
4327 4644 vdev_t *rvd, **vml = NULL; /* vdev modify list */
4328 4645 boolean_t activate_slog;
4329 4646
4330 4647 ASSERT(spa_writeable(spa));
4331 4648
4332 4649 txg = spa_vdev_enter(spa);
4333 4650
4334 4651 /* clear the log and flush everything up to now */
4335 4652 activate_slog = spa_passivate_log(spa);
4336 4653 (void) spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
4337 4654 error = spa_offline_log(spa);
4338 4655 txg = spa_vdev_config_enter(spa);
4339 4656
4340 4657 if (activate_slog)
4341 4658 spa_activate_log(spa);
4342 4659
4343 4660 if (error != 0)
4344 4661 return (spa_vdev_exit(spa, NULL, txg, error));
4345 4662
4346 4663 /* check new spa name before going any further */
4347 4664 if (spa_lookup(newname) != NULL)
4348 4665 return (spa_vdev_exit(spa, NULL, txg, EEXIST));
4349 4666
4350 4667 /*
4351 4668 * scan through all the children to ensure they're all mirrors
4352 4669 */
4353 4670 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvl) != 0 ||
4354 4671 nvlist_lookup_nvlist_array(nvl, ZPOOL_CONFIG_CHILDREN, &child,
4355 4672 &children) != 0)
4356 4673 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4357 4674
4358 4675 /* first, check to ensure we've got the right child count */
4359 4676 rvd = spa->spa_root_vdev;
4360 4677 lastlog = 0;
4361 4678 for (c = 0; c < rvd->vdev_children; c++) {
4362 4679 vdev_t *vd = rvd->vdev_child[c];
4363 4680
4364 4681 /* don't count the holes & logs as children */
4365 4682 if (vd->vdev_islog || vd->vdev_ishole) {
4366 4683 if (lastlog == 0)
4367 4684 lastlog = c;
4368 4685 continue;
4369 4686 }
4370 4687
4371 4688 lastlog = 0;
4372 4689 }
4373 4690 if (children != (lastlog != 0 ? lastlog : rvd->vdev_children))
4374 4691 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4375 4692
4376 4693 /* next, ensure no spare or cache devices are part of the split */
4377 4694 if (nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_SPARES, &tmp) == 0 ||
4378 4695 nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_L2CACHE, &tmp) == 0)
4379 4696 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4380 4697
4381 4698 vml = kmem_zalloc(children * sizeof (vdev_t *), KM_SLEEP);
4382 4699 glist = kmem_zalloc(children * sizeof (uint64_t), KM_SLEEP);
4383 4700
4384 4701 /* then, loop over each vdev and validate it */
4385 4702 for (c = 0; c < children; c++) {
4386 4703 uint64_t is_hole = 0;
4387 4704
4388 4705 (void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_HOLE,
4389 4706 &is_hole);
4390 4707
4391 4708 if (is_hole != 0) {
4392 4709 if (spa->spa_root_vdev->vdev_child[c]->vdev_ishole ||
4393 4710 spa->spa_root_vdev->vdev_child[c]->vdev_islog) {
4394 4711 continue;
4395 4712 } else {
4396 4713 error = EINVAL;
4397 4714 break;
4398 4715 }
4399 4716 }
4400 4717
4401 4718 /* which disk is going to be split? */
4402 4719 if (nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_GUID,
4403 4720 &glist[c]) != 0) {
4404 4721 error = EINVAL;
4405 4722 break;
4406 4723 }
4407 4724
4408 4725 /* look it up in the spa */
4409 4726 vml[c] = spa_lookup_by_guid(spa, glist[c], B_FALSE);
4410 4727 if (vml[c] == NULL) {
4411 4728 error = ENODEV;
4412 4729 break;
4413 4730 }
4414 4731
4415 4732 /* make sure there's nothing stopping the split */
4416 4733 if (vml[c]->vdev_parent->vdev_ops != &vdev_mirror_ops ||
4417 4734 vml[c]->vdev_islog ||
4418 4735 vml[c]->vdev_ishole ||
4419 4736 vml[c]->vdev_isspare ||
4420 4737 vml[c]->vdev_isl2cache ||
4421 4738 !vdev_writeable(vml[c]) ||
4422 4739 vml[c]->vdev_children != 0 ||
4423 4740 vml[c]->vdev_state != VDEV_STATE_HEALTHY ||
4424 4741 c != spa->spa_root_vdev->vdev_child[c]->vdev_id) {
4425 4742 error = EINVAL;
4426 4743 break;
4427 4744 }
4428 4745
4429 4746 if (vdev_dtl_required(vml[c])) {
4430 4747 error = EBUSY;
4431 4748 break;
4432 4749 }
4433 4750
4434 4751 /* we need certain info from the top level */
4435 4752 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_ARRAY,
4436 4753 vml[c]->vdev_top->vdev_ms_array) == 0);
4437 4754 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_SHIFT,
4438 4755 vml[c]->vdev_top->vdev_ms_shift) == 0);
4439 4756 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASIZE,
4440 4757 vml[c]->vdev_top->vdev_asize) == 0);
4441 4758 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASHIFT,
4442 4759 vml[c]->vdev_top->vdev_ashift) == 0);
4443 4760 }
4444 4761
4445 4762 if (error != 0) {
4446 4763 kmem_free(vml, children * sizeof (vdev_t *));
4447 4764 kmem_free(glist, children * sizeof (uint64_t));
4448 4765 return (spa_vdev_exit(spa, NULL, txg, error));
4449 4766 }
4450 4767
4451 4768 /* stop writers from using the disks */
4452 4769 for (c = 0; c < children; c++) {
4453 4770 if (vml[c] != NULL)
4454 4771 vml[c]->vdev_offline = B_TRUE;
4455 4772 }
4456 4773 vdev_reopen(spa->spa_root_vdev);
4457 4774
4458 4775 /*
4459 4776 * Temporarily record the splitting vdevs in the spa config. This
4460 4777 * will disappear once the config is regenerated.
4461 4778 */
4462 4779 VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_SLEEP) == 0);
4463 4780 VERIFY(nvlist_add_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
4464 4781 glist, children) == 0);
4465 4782 kmem_free(glist, children * sizeof (uint64_t));
4466 4783
4467 4784 mutex_enter(&spa->spa_props_lock);
4468 4785 VERIFY(nvlist_add_nvlist(spa->spa_config, ZPOOL_CONFIG_SPLIT,
4469 4786 nvl) == 0);
4470 4787 mutex_exit(&spa->spa_props_lock);
4471 4788 spa->spa_config_splitting = nvl;
4472 4789 vdev_config_dirty(spa->spa_root_vdev);
4473 4790
4474 4791 /* configure and create the new pool */
4475 4792 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, newname) == 0);
4476 4793 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
4477 4794 exp ? POOL_STATE_EXPORTED : POOL_STATE_ACTIVE) == 0);
4478 4795 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
4479 4796 spa_version(spa)) == 0);
4480 4797 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG,
4481 4798 spa->spa_config_txg) == 0);
4482 4799 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID,
4483 4800 spa_generate_guid(NULL)) == 0);
4484 4801 (void) nvlist_lookup_string(props,
4485 4802 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
4486 4803
4487 4804 /* add the new pool to the namespace */
4488 4805 newspa = spa_add(newname, config, altroot);
4489 4806 newspa->spa_config_txg = spa->spa_config_txg;
4490 4807 spa_set_log_state(newspa, SPA_LOG_CLEAR);
4491 4808
4492 4809 /* release the spa config lock, retaining the namespace lock */
4493 4810 spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
4494 4811
4495 4812 if (zio_injection_enabled)
4496 4813 zio_handle_panic_injection(spa, FTAG, 1);
4497 4814
4498 4815 spa_activate(newspa, spa_mode_global);
4499 4816 spa_async_suspend(newspa);
4500 4817
4501 4818 /* create the new pool from the disks of the original pool */
4502 4819 error = spa_load(newspa, SPA_LOAD_IMPORT, SPA_IMPORT_ASSEMBLE, B_TRUE);
4503 4820 if (error)
4504 4821 goto out;
4505 4822
4506 4823 /* if that worked, generate a real config for the new pool */
4507 4824 if (newspa->spa_root_vdev != NULL) {
4508 4825 VERIFY(nvlist_alloc(&newspa->spa_config_splitting,
4509 4826 NV_UNIQUE_NAME, KM_SLEEP) == 0);
4510 4827 VERIFY(nvlist_add_uint64(newspa->spa_config_splitting,
4511 4828 ZPOOL_CONFIG_SPLIT_GUID, spa_guid(spa)) == 0);
4512 4829 spa_config_set(newspa, spa_config_generate(newspa, NULL, -1ULL,
4513 4830 B_TRUE));
4514 4831 }
4515 4832
4516 4833 /* set the props */
4517 4834 if (props != NULL) {
4518 4835 spa_configfile_set(newspa, props, B_FALSE);
4519 4836 error = spa_prop_set(newspa, props);
4520 4837 if (error)
4521 4838 goto out;
4522 4839 }
4523 4840
4524 4841 /* flush everything */
4525 4842 txg = spa_vdev_config_enter(newspa);
4526 4843 vdev_config_dirty(newspa->spa_root_vdev);
4527 4844 (void) spa_vdev_config_exit(newspa, NULL, txg, 0, FTAG);
4528 4845
4529 4846 if (zio_injection_enabled)
4530 4847 zio_handle_panic_injection(spa, FTAG, 2);
4531 4848
4532 4849 spa_async_resume(newspa);
4533 4850
4534 4851 /* finally, update the original pool's config */
4535 4852 txg = spa_vdev_config_enter(spa);
4536 4853 tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir);
4537 4854 error = dmu_tx_assign(tx, TXG_WAIT);
4538 4855 if (error != 0)
4539 4856 dmu_tx_abort(tx);
4540 4857 for (c = 0; c < children; c++) {
4541 4858 if (vml[c] != NULL) {
4542 4859 vdev_split(vml[c]);
4543 4860 if (error == 0)
4544 4861 spa_history_log_internal(LOG_POOL_VDEV_DETACH,
4545 4862 spa, tx, "vdev=%s",
4546 4863 vml[c]->vdev_path);
4547 4864 vdev_free(vml[c]);
4548 4865 }
4549 4866 }
4550 4867 vdev_config_dirty(spa->spa_root_vdev);
4551 4868 spa->spa_config_splitting = NULL;
4552 4869 nvlist_free(nvl);
4553 4870 if (error == 0)
4554 4871 dmu_tx_commit(tx);
4555 4872 (void) spa_vdev_exit(spa, NULL, txg, 0);
4556 4873
4557 4874 if (zio_injection_enabled)
4558 4875 zio_handle_panic_injection(spa, FTAG, 3);
4559 4876
4560 4877 /* split is complete; log a history record */
4561 4878 spa_history_log_internal(LOG_POOL_SPLIT, newspa, NULL,
4562 4879 "split new pool %s from pool %s", newname, spa_name(spa));
4563 4880
4564 4881 kmem_free(vml, children * sizeof (vdev_t *));
4565 4882
4566 4883 /* if we're not going to mount the filesystems in userland, export */
4567 4884 if (exp)
4568 4885 error = spa_export_common(newname, POOL_STATE_EXPORTED, NULL,
4569 4886 B_FALSE, B_FALSE);
4570 4887
4571 4888 return (error);
4572 4889
4573 4890 out:
4574 4891 spa_unload(newspa);
4575 4892 spa_deactivate(newspa);
4576 4893 spa_remove(newspa);
4577 4894
4578 4895 txg = spa_vdev_config_enter(spa);
4579 4896
4580 4897 /* re-online all offlined disks */
4581 4898 for (c = 0; c < children; c++) {
4582 4899 if (vml[c] != NULL)
4583 4900 vml[c]->vdev_offline = B_FALSE;
4584 4901 }
4585 4902 vdev_reopen(spa->spa_root_vdev);
4586 4903
4587 4904 nvlist_free(spa->spa_config_splitting);
4588 4905 spa->spa_config_splitting = NULL;
4589 4906 (void) spa_vdev_exit(spa, NULL, txg, error);
4590 4907
4591 4908 kmem_free(vml, children * sizeof (vdev_t *));
4592 4909 return (error);
4593 4910 }
4594 4911
4595 4912 static nvlist_t *
4596 4913 spa_nvlist_lookup_by_guid(nvlist_t **nvpp, int count, uint64_t target_guid)
4597 4914 {
4598 4915 for (int i = 0; i < count; i++) {
4599 4916 uint64_t guid;
4600 4917
4601 4918 VERIFY(nvlist_lookup_uint64(nvpp[i], ZPOOL_CONFIG_GUID,
4602 4919 &guid) == 0);
4603 4920
4604 4921 if (guid == target_guid)
4605 4922 return (nvpp[i]);
4606 4923 }
4607 4924
4608 4925 return (NULL);
4609 4926 }
4610 4927
4611 4928 static void
4612 4929 spa_vdev_remove_aux(nvlist_t *config, char *name, nvlist_t **dev, int count,
4613 4930 nvlist_t *dev_to_remove)
4614 4931 {
4615 4932 nvlist_t **newdev = NULL;
4616 4933
4617 4934 if (count > 1)
4618 4935 newdev = kmem_alloc((count - 1) * sizeof (void *), KM_SLEEP);
4619 4936
4620 4937 for (int i = 0, j = 0; i < count; i++) {
4621 4938 if (dev[i] == dev_to_remove)
4622 4939 continue;
4623 4940 VERIFY(nvlist_dup(dev[i], &newdev[j++], KM_SLEEP) == 0);
4624 4941 }
4625 4942
4626 4943 VERIFY(nvlist_remove(config, name, DATA_TYPE_NVLIST_ARRAY) == 0);
4627 4944 VERIFY(nvlist_add_nvlist_array(config, name, newdev, count - 1) == 0);
4628 4945
4629 4946 for (int i = 0; i < count - 1; i++)
4630 4947 nvlist_free(newdev[i]);
4631 4948
4632 4949 if (count > 1)
4633 4950 kmem_free(newdev, (count - 1) * sizeof (void *));
4634 4951 }
4635 4952
4636 4953 /*
4637 4954 * Evacuate the device.
4638 4955 */
4639 4956 static int
4640 4957 spa_vdev_remove_evacuate(spa_t *spa, vdev_t *vd)
4641 4958 {
4642 4959 uint64_t txg;
4643 4960 int error = 0;
4644 4961
4645 4962 ASSERT(MUTEX_HELD(&spa_namespace_lock));
4646 4963 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
4647 4964 ASSERT(vd == vd->vdev_top);
4648 4965
4649 4966 /*
4650 4967 * Evacuate the device. We don't hold the config lock as writer
4651 4968 * since we need to do I/O but we do keep the
4652 4969 * spa_namespace_lock held. Once this completes the device
4653 4970 * should no longer have any blocks allocated on it.
4654 4971 */
4655 4972 if (vd->vdev_islog) {
4656 4973 if (vd->vdev_stat.vs_alloc != 0)
4657 4974 error = spa_offline_log(spa);
4658 4975 } else {
4659 4976 error = ENOTSUP;
4660 4977 }
4661 4978
4662 4979 if (error)
4663 4980 return (error);
4664 4981
4665 4982 /*
4666 4983 * The evacuation succeeded. Remove any remaining MOS metadata
4667 4984 * associated with this vdev, and wait for these changes to sync.
4668 4985 */
4669 4986 ASSERT3U(vd->vdev_stat.vs_alloc, ==, 0);
4670 4987 txg = spa_vdev_config_enter(spa);
4671 4988 vd->vdev_removing = B_TRUE;
4672 4989 vdev_dirty(vd, 0, NULL, txg);
4673 4990 vdev_config_dirty(vd);
4674 4991 spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
4675 4992
4676 4993 return (0);
4677 4994 }
4678 4995
4679 4996 /*
4680 4997 * Complete the removal by cleaning up the namespace.
4681 4998 */
4682 4999 static void
4683 5000 spa_vdev_remove_from_namespace(spa_t *spa, vdev_t *vd)
4684 5001 {
4685 5002 vdev_t *rvd = spa->spa_root_vdev;
4686 5003 uint64_t id = vd->vdev_id;
4687 5004 boolean_t last_vdev = (id == (rvd->vdev_children - 1));
4688 5005
4689 5006 ASSERT(MUTEX_HELD(&spa_namespace_lock));
4690 5007 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
4691 5008 ASSERT(vd == vd->vdev_top);
4692 5009
4693 5010 /*
4694 5011 * Only remove any devices which are empty.
4695 5012 */
4696 5013 if (vd->vdev_stat.vs_alloc != 0)
4697 5014 return;
4698 5015
4699 5016 (void) vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
4700 5017
4701 5018 if (list_link_active(&vd->vdev_state_dirty_node))
4702 5019 vdev_state_clean(vd);
4703 5020 if (list_link_active(&vd->vdev_config_dirty_node))
4704 5021 vdev_config_clean(vd);
4705 5022
4706 5023 vdev_free(vd);
4707 5024
4708 5025 if (last_vdev) {
4709 5026 vdev_compact_children(rvd);
4710 5027 } else {
4711 5028 vd = vdev_alloc_common(spa, id, 0, &vdev_hole_ops);
4712 5029 vdev_add_child(rvd, vd);
4713 5030 }
4714 5031 vdev_config_dirty(rvd);
4715 5032
4716 5033 /*
4717 5034 * Reassess the health of our root vdev.
4718 5035 */
4719 5036 vdev_reopen(rvd);
4720 5037 }
4721 5038
4722 5039 /*
4723 5040 * Remove a device from the pool -
4724 5041 *
4725 5042 * Removing a device from the vdev namespace requires several steps
4726 5043 * and can take a significant amount of time. As a result we use
4727 5044 * the spa_vdev_config_[enter/exit] functions which allow us to
4728 5045 * grab and release the spa_config_lock while still holding the namespace
4729 5046 * lock. During each step the configuration is synced out.
4730 5047 */
4731 5048
4732 5049 /*
4733 5050 * Remove a device from the pool. Currently, this supports removing only hot
4734 5051 * spares, slogs, and level 2 ARC devices.
4735 5052 */
4736 5053 int
4737 5054 spa_vdev_remove(spa_t *spa, uint64_t guid, boolean_t unspare)
4738 5055 {
4739 5056 vdev_t *vd;
4740 5057 metaslab_group_t *mg;
4741 5058 nvlist_t **spares, **l2cache, *nv;
4742 5059 uint64_t txg = 0;
4743 5060 uint_t nspares, nl2cache;
4744 5061 int error = 0;
4745 5062 boolean_t locked = MUTEX_HELD(&spa_namespace_lock);
4746 5063
4747 5064 ASSERT(spa_writeable(spa));
4748 5065
4749 5066 if (!locked)
4750 5067 txg = spa_vdev_enter(spa);
4751 5068
4752 5069 vd = spa_lookup_by_guid(spa, guid, B_FALSE);
4753 5070
4754 5071 if (spa->spa_spares.sav_vdevs != NULL &&
4755 5072 nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
4756 5073 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0 &&
4757 5074 (nv = spa_nvlist_lookup_by_guid(spares, nspares, guid)) != NULL) {
4758 5075 /*
4759 5076 * Only remove the hot spare if it's not currently in use
4760 5077 * in this pool.
4761 5078 */
4762 5079 if (vd == NULL || unspare) {
4763 5080 spa_vdev_remove_aux(spa->spa_spares.sav_config,
4764 5081 ZPOOL_CONFIG_SPARES, spares, nspares, nv);
4765 5082 spa_load_spares(spa);
4766 5083 spa->spa_spares.sav_sync = B_TRUE;
4767 5084 } else {
4768 5085 error = EBUSY;
4769 5086 }
4770 5087 } else if (spa->spa_l2cache.sav_vdevs != NULL &&
4771 5088 nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
4772 5089 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0 &&
4773 5090 (nv = spa_nvlist_lookup_by_guid(l2cache, nl2cache, guid)) != NULL) {
4774 5091 /*
4775 5092 * Cache devices can always be removed.
4776 5093 */
4777 5094 spa_vdev_remove_aux(spa->spa_l2cache.sav_config,
4778 5095 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache, nv);
4779 5096 spa_load_l2cache(spa);
4780 5097 spa->spa_l2cache.sav_sync = B_TRUE;
4781 5098 } else if (vd != NULL && vd->vdev_islog) {
4782 5099 ASSERT(!locked);
4783 5100 ASSERT(vd == vd->vdev_top);
4784 5101
4785 5102 /*
4786 5103 * XXX - Once we have bp-rewrite this should
4787 5104 * become the common case.
4788 5105 */
4789 5106
4790 5107 mg = vd->vdev_mg;
4791 5108
4792 5109 /*
4793 5110 * Stop allocating from this vdev.
4794 5111 */
4795 5112 metaslab_group_passivate(mg);
4796 5113
4797 5114 /*
4798 5115 * Wait for the youngest allocations and frees to sync,
4799 5116 * and then wait for the deferral of those frees to finish.
4800 5117 */
4801 5118 spa_vdev_config_exit(spa, NULL,
4802 5119 txg + TXG_CONCURRENT_STATES + TXG_DEFER_SIZE, 0, FTAG);
4803 5120
4804 5121 /*
4805 5122 * Attempt to evacuate the vdev.
4806 5123 */
4807 5124 error = spa_vdev_remove_evacuate(spa, vd);
4808 5125
4809 5126 txg = spa_vdev_config_enter(spa);
4810 5127
4811 5128 /*
4812 5129 * If we couldn't evacuate the vdev, unwind.
4813 5130 */
4814 5131 if (error) {
4815 5132 metaslab_group_activate(mg);
4816 5133 return (spa_vdev_exit(spa, NULL, txg, error));
4817 5134 }
4818 5135
4819 5136 /*
4820 5137 * Clean up the vdev namespace.
4821 5138 */
4822 5139 spa_vdev_remove_from_namespace(spa, vd);
4823 5140
4824 5141 } else if (vd != NULL) {
4825 5142 /*
4826 5143 * Normal vdevs cannot be removed (yet).
4827 5144 */
4828 5145 error = ENOTSUP;
4829 5146 } else {
4830 5147 /*
4831 5148 * There is no vdev of any kind with the specified guid.
4832 5149 */
4833 5150 error = ENOENT;
4834 5151 }
4835 5152
4836 5153 if (!locked)
4837 5154 return (spa_vdev_exit(spa, NULL, txg, error));
4838 5155
4839 5156 return (error);
4840 5157 }
4841 5158
4842 5159 /*
4843 5160 * Find any device that's done replacing, or a vdev marked 'unspare' that's
4844 5161 * current spared, so we can detach it.
4845 5162 */
4846 5163 static vdev_t *
4847 5164 spa_vdev_resilver_done_hunt(vdev_t *vd)
4848 5165 {
4849 5166 vdev_t *newvd, *oldvd;
4850 5167
4851 5168 for (int c = 0; c < vd->vdev_children; c++) {
4852 5169 oldvd = spa_vdev_resilver_done_hunt(vd->vdev_child[c]);
4853 5170 if (oldvd != NULL)
4854 5171 return (oldvd);
4855 5172 }
4856 5173
4857 5174 /*
4858 5175 * Check for a completed replacement. We always consider the first
4859 5176 * vdev in the list to be the oldest vdev, and the last one to be
4860 5177 * the newest (see spa_vdev_attach() for how that works). In
4861 5178 * the case where the newest vdev is faulted, we will not automatically
4862 5179 * remove it after a resilver completes. This is OK as it will require
4863 5180 * user intervention to determine which disk the admin wishes to keep.
4864 5181 */
4865 5182 if (vd->vdev_ops == &vdev_replacing_ops) {
4866 5183 ASSERT(vd->vdev_children > 1);
4867 5184
4868 5185 newvd = vd->vdev_child[vd->vdev_children - 1];
4869 5186 oldvd = vd->vdev_child[0];
4870 5187
4871 5188 if (vdev_dtl_empty(newvd, DTL_MISSING) &&
4872 5189 vdev_dtl_empty(newvd, DTL_OUTAGE) &&
4873 5190 !vdev_dtl_required(oldvd))
4874 5191 return (oldvd);
4875 5192 }
4876 5193
4877 5194 /*
4878 5195 * Check for a completed resilver with the 'unspare' flag set.
4879 5196 */
4880 5197 if (vd->vdev_ops == &vdev_spare_ops) {
4881 5198 vdev_t *first = vd->vdev_child[0];
4882 5199 vdev_t *last = vd->vdev_child[vd->vdev_children - 1];
4883 5200
4884 5201 if (last->vdev_unspare) {
4885 5202 oldvd = first;
4886 5203 newvd = last;
4887 5204 } else if (first->vdev_unspare) {
4888 5205 oldvd = last;
4889 5206 newvd = first;
4890 5207 } else {
4891 5208 oldvd = NULL;
4892 5209 }
4893 5210
4894 5211 if (oldvd != NULL &&
4895 5212 vdev_dtl_empty(newvd, DTL_MISSING) &&
4896 5213 vdev_dtl_empty(newvd, DTL_OUTAGE) &&
4897 5214 !vdev_dtl_required(oldvd))
4898 5215 return (oldvd);
4899 5216
4900 5217 /*
4901 5218 * If there are more than two spares attached to a disk,
4902 5219 * and those spares are not required, then we want to
4903 5220 * attempt to free them up now so that they can be used
4904 5221 * by other pools. Once we're back down to a single
4905 5222 * disk+spare, we stop removing them.
4906 5223 */
4907 5224 if (vd->vdev_children > 2) {
4908 5225 newvd = vd->vdev_child[1];
4909 5226
4910 5227 if (newvd->vdev_isspare && last->vdev_isspare &&
4911 5228 vdev_dtl_empty(last, DTL_MISSING) &&
4912 5229 vdev_dtl_empty(last, DTL_OUTAGE) &&
4913 5230 !vdev_dtl_required(newvd))
4914 5231 return (newvd);
4915 5232 }
4916 5233 }
4917 5234
4918 5235 return (NULL);
4919 5236 }
4920 5237
4921 5238 static void
4922 5239 spa_vdev_resilver_done(spa_t *spa)
4923 5240 {
4924 5241 vdev_t *vd, *pvd, *ppvd;
4925 5242 uint64_t guid, sguid, pguid, ppguid;
4926 5243
4927 5244 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4928 5245
4929 5246 while ((vd = spa_vdev_resilver_done_hunt(spa->spa_root_vdev)) != NULL) {
4930 5247 pvd = vd->vdev_parent;
4931 5248 ppvd = pvd->vdev_parent;
4932 5249 guid = vd->vdev_guid;
4933 5250 pguid = pvd->vdev_guid;
4934 5251 ppguid = ppvd->vdev_guid;
4935 5252 sguid = 0;
4936 5253 /*
4937 5254 * If we have just finished replacing a hot spared device, then
4938 5255 * we need to detach the parent's first child (the original hot
4939 5256 * spare) as well.
4940 5257 */
4941 5258 if (ppvd->vdev_ops == &vdev_spare_ops && pvd->vdev_id == 0 &&
4942 5259 ppvd->vdev_children == 2) {
4943 5260 ASSERT(pvd->vdev_ops == &vdev_replacing_ops);
4944 5261 sguid = ppvd->vdev_child[1]->vdev_guid;
4945 5262 }
4946 5263 spa_config_exit(spa, SCL_ALL, FTAG);
4947 5264 if (spa_vdev_detach(spa, guid, pguid, B_TRUE) != 0)
4948 5265 return;
4949 5266 if (sguid && spa_vdev_detach(spa, sguid, ppguid, B_TRUE) != 0)
4950 5267 return;
4951 5268 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4952 5269 }
4953 5270
4954 5271 spa_config_exit(spa, SCL_ALL, FTAG);
4955 5272 }
4956 5273
4957 5274 /*
4958 5275 * Update the stored path or FRU for this vdev.
4959 5276 */
4960 5277 int
4961 5278 spa_vdev_set_common(spa_t *spa, uint64_t guid, const char *value,
4962 5279 boolean_t ispath)
4963 5280 {
4964 5281 vdev_t *vd;
4965 5282 boolean_t sync = B_FALSE;
4966 5283
4967 5284 ASSERT(spa_writeable(spa));
4968 5285
4969 5286 spa_vdev_state_enter(spa, SCL_ALL);
4970 5287
4971 5288 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
4972 5289 return (spa_vdev_state_exit(spa, NULL, ENOENT));
4973 5290
4974 5291 if (!vd->vdev_ops->vdev_op_leaf)
4975 5292 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
4976 5293
4977 5294 if (ispath) {
4978 5295 if (strcmp(value, vd->vdev_path) != 0) {
4979 5296 spa_strfree(vd->vdev_path);
4980 5297 vd->vdev_path = spa_strdup(value);
4981 5298 sync = B_TRUE;
4982 5299 }
4983 5300 } else {
4984 5301 if (vd->vdev_fru == NULL) {
4985 5302 vd->vdev_fru = spa_strdup(value);
4986 5303 sync = B_TRUE;
4987 5304 } else if (strcmp(value, vd->vdev_fru) != 0) {
4988 5305 spa_strfree(vd->vdev_fru);
4989 5306 vd->vdev_fru = spa_strdup(value);
4990 5307 sync = B_TRUE;
4991 5308 }
4992 5309 }
4993 5310
4994 5311 return (spa_vdev_state_exit(spa, sync ? vd : NULL, 0));
4995 5312 }
4996 5313
4997 5314 int
4998 5315 spa_vdev_setpath(spa_t *spa, uint64_t guid, const char *newpath)
4999 5316 {
5000 5317 return (spa_vdev_set_common(spa, guid, newpath, B_TRUE));
5001 5318 }
5002 5319
5003 5320 int
5004 5321 spa_vdev_setfru(spa_t *spa, uint64_t guid, const char *newfru)
5005 5322 {
5006 5323 return (spa_vdev_set_common(spa, guid, newfru, B_FALSE));
5007 5324 }
5008 5325
5009 5326 /*
5010 5327 * ==========================================================================
5011 5328 * SPA Scanning
5012 5329 * ==========================================================================
5013 5330 */
5014 5331
5015 5332 int
5016 5333 spa_scan_stop(spa_t *spa)
5017 5334 {
5018 5335 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5019 5336 if (dsl_scan_resilvering(spa->spa_dsl_pool))
5020 5337 return (EBUSY);
5021 5338 return (dsl_scan_cancel(spa->spa_dsl_pool));
5022 5339 }
5023 5340
5024 5341 int
5025 5342 spa_scan(spa_t *spa, pool_scan_func_t func)
5026 5343 {
5027 5344 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5028 5345
5029 5346 if (func >= POOL_SCAN_FUNCS || func == POOL_SCAN_NONE)
5030 5347 return (ENOTSUP);
5031 5348
5032 5349 /*
5033 5350 * If a resilver was requested, but there is no DTL on a
5034 5351 * writeable leaf device, we have nothing to do.
5035 5352 */
5036 5353 if (func == POOL_SCAN_RESILVER &&
5037 5354 !vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL)) {
5038 5355 spa_async_request(spa, SPA_ASYNC_RESILVER_DONE);
5039 5356 return (0);
5040 5357 }
5041 5358
5042 5359 return (dsl_scan(spa->spa_dsl_pool, func));
5043 5360 }
5044 5361
5045 5362 /*
5046 5363 * ==========================================================================
5047 5364 * SPA async task processing
5048 5365 * ==========================================================================
5049 5366 */
5050 5367
5051 5368 static void
5052 5369 spa_async_remove(spa_t *spa, vdev_t *vd)
5053 5370 {
5054 5371 if (vd->vdev_remove_wanted) {
5055 5372 vd->vdev_remove_wanted = B_FALSE;
5056 5373 vd->vdev_delayed_close = B_FALSE;
5057 5374 vdev_set_state(vd, B_FALSE, VDEV_STATE_REMOVED, VDEV_AUX_NONE);
5058 5375
5059 5376 /*
5060 5377 * We want to clear the stats, but we don't want to do a full
5061 5378 * vdev_clear() as that will cause us to throw away
5062 5379 * degraded/faulted state as well as attempt to reopen the
5063 5380 * device, all of which is a waste.
5064 5381 */
5065 5382 vd->vdev_stat.vs_read_errors = 0;
5066 5383 vd->vdev_stat.vs_write_errors = 0;
5067 5384 vd->vdev_stat.vs_checksum_errors = 0;
5068 5385
5069 5386 vdev_state_dirty(vd->vdev_top);
5070 5387 }
5071 5388
5072 5389 for (int c = 0; c < vd->vdev_children; c++)
5073 5390 spa_async_remove(spa, vd->vdev_child[c]);
5074 5391 }
5075 5392
5076 5393 static void
5077 5394 spa_async_probe(spa_t *spa, vdev_t *vd)
5078 5395 {
5079 5396 if (vd->vdev_probe_wanted) {
5080 5397 vd->vdev_probe_wanted = B_FALSE;
5081 5398 vdev_reopen(vd); /* vdev_open() does the actual probe */
5082 5399 }
5083 5400
5084 5401 for (int c = 0; c < vd->vdev_children; c++)
5085 5402 spa_async_probe(spa, vd->vdev_child[c]);
5086 5403 }
5087 5404
5088 5405 static void
5089 5406 spa_async_autoexpand(spa_t *spa, vdev_t *vd)
5090 5407 {
5091 5408 sysevent_id_t eid;
5092 5409 nvlist_t *attr;
5093 5410 char *physpath;
5094 5411
5095 5412 if (!spa->spa_autoexpand)
5096 5413 return;
5097 5414
5098 5415 for (int c = 0; c < vd->vdev_children; c++) {
5099 5416 vdev_t *cvd = vd->vdev_child[c];
5100 5417 spa_async_autoexpand(spa, cvd);
5101 5418 }
5102 5419
5103 5420 if (!vd->vdev_ops->vdev_op_leaf || vd->vdev_physpath == NULL)
5104 5421 return;
5105 5422
5106 5423 physpath = kmem_zalloc(MAXPATHLEN, KM_SLEEP);
5107 5424 (void) snprintf(physpath, MAXPATHLEN, "/devices%s", vd->vdev_physpath);
5108 5425
5109 5426 VERIFY(nvlist_alloc(&attr, NV_UNIQUE_NAME, KM_SLEEP) == 0);
5110 5427 VERIFY(nvlist_add_string(attr, DEV_PHYS_PATH, physpath) == 0);
5111 5428
5112 5429 (void) ddi_log_sysevent(zfs_dip, SUNW_VENDOR, EC_DEV_STATUS,
5113 5430 ESC_DEV_DLE, attr, &eid, DDI_SLEEP);
5114 5431
5115 5432 nvlist_free(attr);
5116 5433 kmem_free(physpath, MAXPATHLEN);
5117 5434 }
5118 5435
5119 5436 static void
5120 5437 spa_async_thread(spa_t *spa)
5121 5438 {
5122 5439 int tasks;
5123 5440
5124 5441 ASSERT(spa->spa_sync_on);
5125 5442
5126 5443 mutex_enter(&spa->spa_async_lock);
5127 5444 tasks = spa->spa_async_tasks;
5128 5445 spa->spa_async_tasks = 0;
5129 5446 mutex_exit(&spa->spa_async_lock);
5130 5447
5131 5448 /*
5132 5449 * See if the config needs to be updated.
5133 5450 */
5134 5451 if (tasks & SPA_ASYNC_CONFIG_UPDATE) {
5135 5452 uint64_t old_space, new_space;
5136 5453
5137 5454 mutex_enter(&spa_namespace_lock);
5138 5455 old_space = metaslab_class_get_space(spa_normal_class(spa));
5139 5456 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
5140 5457 new_space = metaslab_class_get_space(spa_normal_class(spa));
5141 5458 mutex_exit(&spa_namespace_lock);
5142 5459
5143 5460 /*
5144 5461 * If the pool grew as a result of the config update,
5145 5462 * then log an internal history event.
5146 5463 */
5147 5464 if (new_space != old_space) {
5148 5465 spa_history_log_internal(LOG_POOL_VDEV_ONLINE,
5149 5466 spa, NULL,
5150 5467 "pool '%s' size: %llu(+%llu)",
5151 5468 spa_name(spa), new_space, new_space - old_space);
5152 5469 }
5153 5470 }
5154 5471
5155 5472 /*
5156 5473 * See if any devices need to be marked REMOVED.
5157 5474 */
5158 5475 if (tasks & SPA_ASYNC_REMOVE) {
5159 5476 spa_vdev_state_enter(spa, SCL_NONE);
5160 5477 spa_async_remove(spa, spa->spa_root_vdev);
5161 5478 for (int i = 0; i < spa->spa_l2cache.sav_count; i++)
5162 5479 spa_async_remove(spa, spa->spa_l2cache.sav_vdevs[i]);
5163 5480 for (int i = 0; i < spa->spa_spares.sav_count; i++)
5164 5481 spa_async_remove(spa, spa->spa_spares.sav_vdevs[i]);
5165 5482 (void) spa_vdev_state_exit(spa, NULL, 0);
5166 5483 }
5167 5484
5168 5485 if ((tasks & SPA_ASYNC_AUTOEXPAND) && !spa_suspended(spa)) {
5169 5486 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
5170 5487 spa_async_autoexpand(spa, spa->spa_root_vdev);
5171 5488 spa_config_exit(spa, SCL_CONFIG, FTAG);
5172 5489 }
5173 5490
5174 5491 /*
5175 5492 * See if any devices need to be probed.
5176 5493 */
5177 5494 if (tasks & SPA_ASYNC_PROBE) {
5178 5495 spa_vdev_state_enter(spa, SCL_NONE);
5179 5496 spa_async_probe(spa, spa->spa_root_vdev);
5180 5497 (void) spa_vdev_state_exit(spa, NULL, 0);
5181 5498 }
5182 5499
5183 5500 /*
5184 5501 * If any devices are done replacing, detach them.
5185 5502 */
5186 5503 if (tasks & SPA_ASYNC_RESILVER_DONE)
5187 5504 spa_vdev_resilver_done(spa);
5188 5505
5189 5506 /*
5190 5507 * Kick off a resilver.
5191 5508 */
5192 5509 if (tasks & SPA_ASYNC_RESILVER)
5193 5510 dsl_resilver_restart(spa->spa_dsl_pool, 0);
5194 5511
5195 5512 /*
5196 5513 * Let the world know that we're done.
5197 5514 */
5198 5515 mutex_enter(&spa->spa_async_lock);
5199 5516 spa->spa_async_thread = NULL;
5200 5517 cv_broadcast(&spa->spa_async_cv);
5201 5518 mutex_exit(&spa->spa_async_lock);
5202 5519 thread_exit();
5203 5520 }
5204 5521
5205 5522 void
5206 5523 spa_async_suspend(spa_t *spa)
5207 5524 {
5208 5525 mutex_enter(&spa->spa_async_lock);
5209 5526 spa->spa_async_suspended++;
5210 5527 while (spa->spa_async_thread != NULL)
5211 5528 cv_wait(&spa->spa_async_cv, &spa->spa_async_lock);
5212 5529 mutex_exit(&spa->spa_async_lock);
5213 5530 }
5214 5531
5215 5532 void
5216 5533 spa_async_resume(spa_t *spa)
5217 5534 {
5218 5535 mutex_enter(&spa->spa_async_lock);
5219 5536 ASSERT(spa->spa_async_suspended != 0);
5220 5537 spa->spa_async_suspended--;
5221 5538 mutex_exit(&spa->spa_async_lock);
5222 5539 }
5223 5540
5224 5541 static void
5225 5542 spa_async_dispatch(spa_t *spa)
5226 5543 {
5227 5544 mutex_enter(&spa->spa_async_lock);
5228 5545 if (spa->spa_async_tasks && !spa->spa_async_suspended &&
5229 5546 spa->spa_async_thread == NULL &&
5230 5547 rootdir != NULL && !vn_is_readonly(rootdir))
5231 5548 spa->spa_async_thread = thread_create(NULL, 0,
5232 5549 spa_async_thread, spa, 0, &p0, TS_RUN, maxclsyspri);
5233 5550 mutex_exit(&spa->spa_async_lock);
5234 5551 }
5235 5552
5236 5553 void
5237 5554 spa_async_request(spa_t *spa, int task)
5238 5555 {
5239 5556 zfs_dbgmsg("spa=%s async request task=%u", spa->spa_name, task);
5240 5557 mutex_enter(&spa->spa_async_lock);
5241 5558 spa->spa_async_tasks |= task;
5242 5559 mutex_exit(&spa->spa_async_lock);
5243 5560 }
5244 5561
5245 5562 /*
5246 5563 * ==========================================================================
5247 5564 * SPA syncing routines
5248 5565 * ==========================================================================
5249 5566 */
5250 5567
5251 5568 static int
5252 5569 bpobj_enqueue_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
5253 5570 {
5254 5571 bpobj_t *bpo = arg;
5255 5572 bpobj_enqueue(bpo, bp, tx);
5256 5573 return (0);
5257 5574 }
5258 5575
5259 5576 static int
5260 5577 spa_free_sync_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
5261 5578 {
5262 5579 zio_t *zio = arg;
5263 5580
5264 5581 zio_nowait(zio_free_sync(zio, zio->io_spa, dmu_tx_get_txg(tx), bp,
5265 5582 zio->io_flags));
5266 5583 return (0);
5267 5584 }
5268 5585
5269 5586 static void
5270 5587 spa_sync_nvlist(spa_t *spa, uint64_t obj, nvlist_t *nv, dmu_tx_t *tx)
5271 5588 {
5272 5589 char *packed = NULL;
5273 5590 size_t bufsize;
|
↓ open down ↓ |
1697 lines elided |
↑ open up ↑ |
5274 5591 size_t nvsize = 0;
5275 5592 dmu_buf_t *db;
5276 5593
5277 5594 VERIFY(nvlist_size(nv, &nvsize, NV_ENCODE_XDR) == 0);
5278 5595
5279 5596 /*
5280 5597 * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration
5281 5598 * information. This avoids the dbuf_will_dirty() path and
5282 5599 * saves us a pre-read to get data we don't actually care about.
5283 5600 */
5284 - bufsize = P2ROUNDUP(nvsize, SPA_CONFIG_BLOCKSIZE);
5601 + bufsize = P2ROUNDUP((uint64_t)nvsize, SPA_CONFIG_BLOCKSIZE);
5285 5602 packed = kmem_alloc(bufsize, KM_SLEEP);
5286 5603
5287 5604 VERIFY(nvlist_pack(nv, &packed, &nvsize, NV_ENCODE_XDR,
5288 5605 KM_SLEEP) == 0);
5289 5606 bzero(packed + nvsize, bufsize - nvsize);
5290 5607
5291 5608 dmu_write(spa->spa_meta_objset, obj, 0, bufsize, packed, tx);
5292 5609
5293 5610 kmem_free(packed, bufsize);
5294 5611
5295 5612 VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
5296 5613 dmu_buf_will_dirty(db, tx);
5297 5614 *(uint64_t *)db->db_data = nvsize;
5298 5615 dmu_buf_rele(db, FTAG);
5299 5616 }
5300 5617
5301 5618 static void
5302 5619 spa_sync_aux_dev(spa_t *spa, spa_aux_vdev_t *sav, dmu_tx_t *tx,
5303 5620 const char *config, const char *entry)
5304 5621 {
5305 5622 nvlist_t *nvroot;
5306 5623 nvlist_t **list;
5307 5624 int i;
5308 5625
5309 5626 if (!sav->sav_sync)
5310 5627 return;
5311 5628
5312 5629 /*
5313 5630 * Update the MOS nvlist describing the list of available devices.
5314 5631 * spa_validate_aux() will have already made sure this nvlist is
5315 5632 * valid and the vdevs are labeled appropriately.
5316 5633 */
5317 5634 if (sav->sav_object == 0) {
5318 5635 sav->sav_object = dmu_object_alloc(spa->spa_meta_objset,
5319 5636 DMU_OT_PACKED_NVLIST, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE,
5320 5637 sizeof (uint64_t), tx);
5321 5638 VERIFY(zap_update(spa->spa_meta_objset,
5322 5639 DMU_POOL_DIRECTORY_OBJECT, entry, sizeof (uint64_t), 1,
5323 5640 &sav->sav_object, tx) == 0);
5324 5641 }
5325 5642
5326 5643 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
5327 5644 if (sav->sav_count == 0) {
5328 5645 VERIFY(nvlist_add_nvlist_array(nvroot, config, NULL, 0) == 0);
5329 5646 } else {
5330 5647 list = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
5331 5648 for (i = 0; i < sav->sav_count; i++)
5332 5649 list[i] = vdev_config_generate(spa, sav->sav_vdevs[i],
5333 5650 B_FALSE, VDEV_CONFIG_L2CACHE);
5334 5651 VERIFY(nvlist_add_nvlist_array(nvroot, config, list,
5335 5652 sav->sav_count) == 0);
5336 5653 for (i = 0; i < sav->sav_count; i++)
5337 5654 nvlist_free(list[i]);
5338 5655 kmem_free(list, sav->sav_count * sizeof (void *));
5339 5656 }
5340 5657
5341 5658 spa_sync_nvlist(spa, sav->sav_object, nvroot, tx);
5342 5659 nvlist_free(nvroot);
5343 5660
5344 5661 sav->sav_sync = B_FALSE;
5345 5662 }
5346 5663
5347 5664 static void
5348 5665 spa_sync_config_object(spa_t *spa, dmu_tx_t *tx)
5349 5666 {
5350 5667 nvlist_t *config;
5351 5668
5352 5669 if (list_is_empty(&spa->spa_config_dirty_list))
5353 5670 return;
5354 5671
5355 5672 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
5356 5673
5357 5674 config = spa_config_generate(spa, spa->spa_root_vdev,
5358 5675 dmu_tx_get_txg(tx), B_FALSE);
|
↓ open down ↓ |
64 lines elided |
↑ open up ↑ |
5359 5676
5360 5677 spa_config_exit(spa, SCL_STATE, FTAG);
5361 5678
5362 5679 if (spa->spa_config_syncing)
5363 5680 nvlist_free(spa->spa_config_syncing);
5364 5681 spa->spa_config_syncing = config;
5365 5682
5366 5683 spa_sync_nvlist(spa, spa->spa_config_object, config, tx);
5367 5684 }
5368 5685
5686 +static void
5687 +spa_sync_version(void *arg1, void *arg2, dmu_tx_t *tx)
5688 +{
5689 + spa_t *spa = arg1;
5690 + uint64_t version = *(uint64_t *)arg2;
5691 +
5692 + /*
5693 + * Setting the version is special cased when first creating the pool.
5694 + */
5695 + ASSERT(tx->tx_txg != TXG_INITIAL);
5696 +
5697 + ASSERT(version <= SPA_VERSION);
5698 + ASSERT(version >= spa_version(spa));
5699 +
5700 + spa->spa_uberblock.ub_version = version;
5701 + vdev_config_dirty(spa->spa_root_vdev);
5702 +}
5703 +
5369 5704 /*
5370 5705 * Set zpool properties.
5371 5706 */
5372 5707 static void
5373 5708 spa_sync_props(void *arg1, void *arg2, dmu_tx_t *tx)
5374 5709 {
5375 5710 spa_t *spa = arg1;
5376 5711 objset_t *mos = spa->spa_meta_objset;
5377 5712 nvlist_t *nvp = arg2;
5378 - nvpair_t *elem;
5379 - uint64_t intval;
5380 - char *strval;
5381 - zpool_prop_t prop;
5382 - const char *propname;
5383 - zprop_type_t proptype;
5713 + nvpair_t *elem = NULL;
5384 5714
5385 5715 mutex_enter(&spa->spa_props_lock);
5386 5716
5387 - elem = NULL;
5388 5717 while ((elem = nvlist_next_nvpair(nvp, elem))) {
5718 + uint64_t intval;
5719 + char *strval, *fname;
5720 + zpool_prop_t prop;
5721 + const char *propname;
5722 + zprop_type_t proptype;
5723 + zfeature_info_t *feature;
5724 +
5389 5725 switch (prop = zpool_name_to_prop(nvpair_name(elem))) {
5726 + case ZPROP_INVAL:
5727 + /*
5728 + * We checked this earlier in spa_prop_validate().
5729 + */
5730 + ASSERT(zpool_prop_feature(nvpair_name(elem)));
5731 +
5732 + fname = strchr(nvpair_name(elem), '@') + 1;
5733 + VERIFY3U(0, ==, zfeature_lookup_name(fname, &feature));
5734 +
5735 + spa_feature_enable(spa, feature, tx);
5736 + break;
5737 +
5390 5738 case ZPOOL_PROP_VERSION:
5739 + VERIFY(nvpair_value_uint64(elem, &intval) == 0);
5391 5740 /*
5392 - * Only set version for non-zpool-creation cases
5393 - * (set/import). spa_create() needs special care
5394 - * for version setting.
5741 + * The version is synced seperatly before other
5742 + * properties and should be correct by now.
5395 5743 */
5396 - if (tx->tx_txg != TXG_INITIAL) {
5397 - VERIFY(nvpair_value_uint64(elem,
5398 - &intval) == 0);
5399 - ASSERT(intval <= SPA_VERSION);
5400 - ASSERT(intval >= spa_version(spa));
5401 - spa->spa_uberblock.ub_version = intval;
5402 - vdev_config_dirty(spa->spa_root_vdev);
5403 - }
5744 + ASSERT3U(spa_version(spa), >=, intval);
5404 5745 break;
5405 5746
5406 5747 case ZPOOL_PROP_ALTROOT:
5407 5748 /*
5408 5749 * 'altroot' is a non-persistent property. It should
5409 5750 * have been set temporarily at creation or import time.
5410 5751 */
5411 5752 ASSERT(spa->spa_root != NULL);
5412 5753 break;
5413 5754
5414 5755 case ZPOOL_PROP_READONLY:
5415 5756 case ZPOOL_PROP_CACHEFILE:
5416 5757 /*
5417 5758 * 'readonly' and 'cachefile' are also non-persisitent
5418 5759 * properties.
5419 5760 */
5420 5761 break;
5421 5762 case ZPOOL_PROP_COMMENT:
5422 5763 VERIFY(nvpair_value_string(elem, &strval) == 0);
5423 5764 if (spa->spa_comment != NULL)
5424 5765 spa_strfree(spa->spa_comment);
5425 5766 spa->spa_comment = spa_strdup(strval);
5426 5767 /*
5427 5768 * We need to dirty the configuration on all the vdevs
5428 5769 * so that their labels get updated. It's unnecessary
5429 5770 * to do this for pool creation since the vdev's
|
↓ open down ↓ |
16 lines elided |
↑ open up ↑ |
5430 5771 * configuratoin has already been dirtied.
5431 5772 */
5432 5773 if (tx->tx_txg != TXG_INITIAL)
5433 5774 vdev_config_dirty(spa->spa_root_vdev);
5434 5775 break;
5435 5776 default:
5436 5777 /*
5437 5778 * Set pool property values in the poolprops mos object.
5438 5779 */
5439 5780 if (spa->spa_pool_props_object == 0) {
5440 - VERIFY((spa->spa_pool_props_object =
5441 - zap_create(mos, DMU_OT_POOL_PROPS,
5442 - DMU_OT_NONE, 0, tx)) > 0);
5443 -
5444 - VERIFY(zap_update(mos,
5781 + spa->spa_pool_props_object =
5782 + zap_create_link(mos, DMU_OT_POOL_PROPS,
5445 5783 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_PROPS,
5446 - 8, 1, &spa->spa_pool_props_object, tx)
5447 - == 0);
5784 + tx);
5448 5785 }
5449 5786
5450 5787 /* normalize the property name */
5451 5788 propname = zpool_prop_to_name(prop);
5452 5789 proptype = zpool_prop_get_type(prop);
5453 5790
5454 5791 if (nvpair_type(elem) == DATA_TYPE_STRING) {
5455 5792 ASSERT(proptype == PROP_TYPE_STRING);
5456 5793 VERIFY(nvpair_value_string(elem, &strval) == 0);
5457 5794 VERIFY(zap_update(mos,
5458 5795 spa->spa_pool_props_object, propname,
5459 5796 1, strlen(strval) + 1, strval, tx) == 0);
5460 5797
5461 5798 } else if (nvpair_type(elem) == DATA_TYPE_UINT64) {
5462 5799 VERIFY(nvpair_value_uint64(elem, &intval) == 0);
5463 5800
5464 5801 if (proptype == PROP_TYPE_INDEX) {
5465 5802 const char *unused;
5466 5803 VERIFY(zpool_prop_index_to_string(
5467 5804 prop, intval, &unused) == 0);
5468 5805 }
5469 5806 VERIFY(zap_update(mos,
5470 5807 spa->spa_pool_props_object, propname,
5471 5808 8, 1, &intval, tx) == 0);
5472 5809 } else {
5473 5810 ASSERT(0); /* not allowed */
5474 5811 }
5475 5812
5476 5813 switch (prop) {
5477 5814 case ZPOOL_PROP_DELEGATION:
5478 5815 spa->spa_delegation = intval;
5479 5816 break;
5480 5817 case ZPOOL_PROP_BOOTFS:
5481 5818 spa->spa_bootfs = intval;
5482 5819 break;
5483 5820 case ZPOOL_PROP_FAILUREMODE:
5484 5821 spa->spa_failmode = intval;
5485 5822 break;
5486 5823 case ZPOOL_PROP_AUTOEXPAND:
5487 5824 spa->spa_autoexpand = intval;
5488 5825 if (tx->tx_txg != TXG_INITIAL)
5489 5826 spa_async_request(spa,
5490 5827 SPA_ASYNC_AUTOEXPAND);
5491 5828 break;
5492 5829 case ZPOOL_PROP_DEDUPDITTO:
5493 5830 spa->spa_dedup_ditto = intval;
5494 5831 break;
5495 5832 default:
5496 5833 break;
5497 5834 }
5498 5835 }
5499 5836
5500 5837 /* log internal history if this is not a zpool create */
5501 5838 if (spa_version(spa) >= SPA_VERSION_ZPOOL_HISTORY &&
5502 5839 tx->tx_txg != TXG_INITIAL) {
5503 5840 spa_history_log_internal(LOG_POOL_PROPSET,
5504 5841 spa, tx, "%s %lld %s",
5505 5842 nvpair_name(elem), intval, spa_name(spa));
5506 5843 }
5507 5844 }
5508 5845
5509 5846 mutex_exit(&spa->spa_props_lock);
5510 5847 }
5511 5848
5512 5849 /*
5513 5850 * Perform one-time upgrade on-disk changes. spa_version() does not
5514 5851 * reflect the new version this txg, so there must be no changes this
5515 5852 * txg to anything that the upgrade code depends on after it executes.
5516 5853 * Therefore this must be called after dsl_pool_sync() does the sync
5517 5854 * tasks.
5518 5855 */
5519 5856 static void
5520 5857 spa_sync_upgrades(spa_t *spa, dmu_tx_t *tx)
5521 5858 {
5522 5859 dsl_pool_t *dp = spa->spa_dsl_pool;
5523 5860
5524 5861 ASSERT(spa->spa_sync_pass == 1);
5525 5862
5526 5863 if (spa->spa_ubsync.ub_version < SPA_VERSION_ORIGIN &&
5527 5864 spa->spa_uberblock.ub_version >= SPA_VERSION_ORIGIN) {
5528 5865 dsl_pool_create_origin(dp, tx);
5529 5866
5530 5867 /* Keeping the origin open increases spa_minref */
5531 5868 spa->spa_minref += 3;
5532 5869 }
5533 5870
5534 5871 if (spa->spa_ubsync.ub_version < SPA_VERSION_NEXT_CLONES &&
5535 5872 spa->spa_uberblock.ub_version >= SPA_VERSION_NEXT_CLONES) {
|
↓ open down ↓ |
78 lines elided |
↑ open up ↑ |
5536 5873 dsl_pool_upgrade_clones(dp, tx);
5537 5874 }
5538 5875
5539 5876 if (spa->spa_ubsync.ub_version < SPA_VERSION_DIR_CLONES &&
5540 5877 spa->spa_uberblock.ub_version >= SPA_VERSION_DIR_CLONES) {
5541 5878 dsl_pool_upgrade_dir_clones(dp, tx);
5542 5879
5543 5880 /* Keeping the freedir open increases spa_minref */
5544 5881 spa->spa_minref += 3;
5545 5882 }
5883 +
5884 + if (spa->spa_ubsync.ub_version < SPA_VERSION_FEATURES &&
5885 + spa->spa_uberblock.ub_version >= SPA_VERSION_FEATURES) {
5886 + spa_feature_create_zap_objects(spa, tx);
5887 + }
5546 5888 }
5547 5889
5548 5890 /*
5549 5891 * Sync the specified transaction group. New blocks may be dirtied as
5550 5892 * part of the process, so we iterate until it converges.
5551 5893 */
5552 5894 void
5553 5895 spa_sync(spa_t *spa, uint64_t txg)
5554 5896 {
5555 5897 dsl_pool_t *dp = spa->spa_dsl_pool;
5556 5898 objset_t *mos = spa->spa_meta_objset;
5557 5899 bpobj_t *defer_bpo = &spa->spa_deferred_bpobj;
5558 5900 bplist_t *free_bpl = &spa->spa_free_bplist[txg & TXG_MASK];
5559 5901 vdev_t *rvd = spa->spa_root_vdev;
5560 5902 vdev_t *vd;
5561 5903 dmu_tx_t *tx;
5562 5904 int error;
5563 5905
5564 5906 VERIFY(spa_writeable(spa));
5565 5907
5566 5908 /*
5567 5909 * Lock out configuration changes.
5568 5910 */
5569 5911 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
5570 5912
5571 5913 spa->spa_syncing_txg = txg;
5572 5914 spa->spa_sync_pass = 0;
5573 5915
5574 5916 /*
5575 5917 * If there are any pending vdev state changes, convert them
5576 5918 * into config changes that go out with this transaction group.
5577 5919 */
5578 5920 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
5579 5921 while (list_head(&spa->spa_state_dirty_list) != NULL) {
5580 5922 /*
5581 5923 * We need the write lock here because, for aux vdevs,
5582 5924 * calling vdev_config_dirty() modifies sav_config.
5583 5925 * This is ugly and will become unnecessary when we
5584 5926 * eliminate the aux vdev wart by integrating all vdevs
5585 5927 * into the root vdev tree.
5586 5928 */
5587 5929 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
5588 5930 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_WRITER);
5589 5931 while ((vd = list_head(&spa->spa_state_dirty_list)) != NULL) {
5590 5932 vdev_state_clean(vd);
5591 5933 vdev_config_dirty(vd);
5592 5934 }
5593 5935 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
5594 5936 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER);
5595 5937 }
5596 5938 spa_config_exit(spa, SCL_STATE, FTAG);
5597 5939
5598 5940 tx = dmu_tx_create_assigned(dp, txg);
5599 5941
5600 5942 /*
5601 5943 * If we are upgrading to SPA_VERSION_RAIDZ_DEFLATE this txg,
5602 5944 * set spa_deflate if we have no raid-z vdevs.
5603 5945 */
5604 5946 if (spa->spa_ubsync.ub_version < SPA_VERSION_RAIDZ_DEFLATE &&
5605 5947 spa->spa_uberblock.ub_version >= SPA_VERSION_RAIDZ_DEFLATE) {
5606 5948 int i;
5607 5949
5608 5950 for (i = 0; i < rvd->vdev_children; i++) {
5609 5951 vd = rvd->vdev_child[i];
5610 5952 if (vd->vdev_deflate_ratio != SPA_MINBLOCKSIZE)
5611 5953 break;
5612 5954 }
5613 5955 if (i == rvd->vdev_children) {
5614 5956 spa->spa_deflate = TRUE;
5615 5957 VERIFY(0 == zap_add(spa->spa_meta_objset,
5616 5958 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
5617 5959 sizeof (uint64_t), 1, &spa->spa_deflate, tx));
5618 5960 }
5619 5961 }
5620 5962
5621 5963 /*
5622 5964 * If anything has changed in this txg, or if someone is waiting
5623 5965 * for this txg to sync (eg, spa_vdev_remove()), push the
5624 5966 * deferred frees from the previous txg. If not, leave them
5625 5967 * alone so that we don't generate work on an otherwise idle
5626 5968 * system.
5627 5969 */
5628 5970 if (!txg_list_empty(&dp->dp_dirty_datasets, txg) ||
5629 5971 !txg_list_empty(&dp->dp_dirty_dirs, txg) ||
5630 5972 !txg_list_empty(&dp->dp_sync_tasks, txg) ||
5631 5973 ((dsl_scan_active(dp->dp_scan) ||
5632 5974 txg_sync_waiting(dp)) && !spa_shutting_down(spa))) {
5633 5975 zio_t *zio = zio_root(spa, NULL, NULL, 0);
5634 5976 VERIFY3U(bpobj_iterate(defer_bpo,
5635 5977 spa_free_sync_cb, zio, tx), ==, 0);
5636 5978 VERIFY3U(zio_wait(zio), ==, 0);
5637 5979 }
5638 5980
5639 5981 /*
5640 5982 * Iterate to convergence.
5641 5983 */
5642 5984 do {
5643 5985 int pass = ++spa->spa_sync_pass;
5644 5986
5645 5987 spa_sync_config_object(spa, tx);
5646 5988 spa_sync_aux_dev(spa, &spa->spa_spares, tx,
5647 5989 ZPOOL_CONFIG_SPARES, DMU_POOL_SPARES);
5648 5990 spa_sync_aux_dev(spa, &spa->spa_l2cache, tx,
5649 5991 ZPOOL_CONFIG_L2CACHE, DMU_POOL_L2CACHE);
5650 5992 spa_errlog_sync(spa, txg);
5651 5993 dsl_pool_sync(dp, txg);
5652 5994
5653 5995 if (pass <= SYNC_PASS_DEFERRED_FREE) {
5654 5996 zio_t *zio = zio_root(spa, NULL, NULL, 0);
5655 5997 bplist_iterate(free_bpl, spa_free_sync_cb,
5656 5998 zio, tx);
5657 5999 VERIFY(zio_wait(zio) == 0);
5658 6000 } else {
5659 6001 bplist_iterate(free_bpl, bpobj_enqueue_cb,
5660 6002 defer_bpo, tx);
5661 6003 }
5662 6004
5663 6005 ddt_sync(spa, txg);
5664 6006 dsl_scan_sync(dp, tx);
5665 6007
5666 6008 while (vd = txg_list_remove(&spa->spa_vdev_txg_list, txg))
5667 6009 vdev_sync(vd, txg);
5668 6010
5669 6011 if (pass == 1)
5670 6012 spa_sync_upgrades(spa, tx);
5671 6013
5672 6014 } while (dmu_objset_is_dirty(mos, txg));
5673 6015
5674 6016 /*
5675 6017 * Rewrite the vdev configuration (which includes the uberblock)
5676 6018 * to commit the transaction group.
5677 6019 *
5678 6020 * If there are no dirty vdevs, we sync the uberblock to a few
5679 6021 * random top-level vdevs that are known to be visible in the
5680 6022 * config cache (see spa_vdev_add() for a complete description).
5681 6023 * If there *are* dirty vdevs, sync the uberblock to all vdevs.
5682 6024 */
5683 6025 for (;;) {
5684 6026 /*
5685 6027 * We hold SCL_STATE to prevent vdev open/close/etc.
5686 6028 * while we're attempting to write the vdev labels.
5687 6029 */
5688 6030 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
5689 6031
5690 6032 if (list_is_empty(&spa->spa_config_dirty_list)) {
5691 6033 vdev_t *svd[SPA_DVAS_PER_BP];
5692 6034 int svdcount = 0;
5693 6035 int children = rvd->vdev_children;
5694 6036 int c0 = spa_get_random(children);
5695 6037
5696 6038 for (int c = 0; c < children; c++) {
5697 6039 vd = rvd->vdev_child[(c0 + c) % children];
5698 6040 if (vd->vdev_ms_array == 0 || vd->vdev_islog)
5699 6041 continue;
5700 6042 svd[svdcount++] = vd;
5701 6043 if (svdcount == SPA_DVAS_PER_BP)
5702 6044 break;
5703 6045 }
5704 6046 error = vdev_config_sync(svd, svdcount, txg, B_FALSE);
5705 6047 if (error != 0)
5706 6048 error = vdev_config_sync(svd, svdcount, txg,
5707 6049 B_TRUE);
5708 6050 } else {
5709 6051 error = vdev_config_sync(rvd->vdev_child,
5710 6052 rvd->vdev_children, txg, B_FALSE);
5711 6053 if (error != 0)
5712 6054 error = vdev_config_sync(rvd->vdev_child,
5713 6055 rvd->vdev_children, txg, B_TRUE);
5714 6056 }
5715 6057
5716 6058 spa_config_exit(spa, SCL_STATE, FTAG);
5717 6059
5718 6060 if (error == 0)
5719 6061 break;
5720 6062 zio_suspend(spa, NULL);
5721 6063 zio_resume_wait(spa);
5722 6064 }
5723 6065 dmu_tx_commit(tx);
5724 6066
5725 6067 /*
5726 6068 * Clear the dirty config list.
5727 6069 */
5728 6070 while ((vd = list_head(&spa->spa_config_dirty_list)) != NULL)
5729 6071 vdev_config_clean(vd);
5730 6072
5731 6073 /*
5732 6074 * Now that the new config has synced transactionally,
5733 6075 * let it become visible to the config cache.
5734 6076 */
5735 6077 if (spa->spa_config_syncing != NULL) {
5736 6078 spa_config_set(spa, spa->spa_config_syncing);
5737 6079 spa->spa_config_txg = txg;
5738 6080 spa->spa_config_syncing = NULL;
5739 6081 }
5740 6082
5741 6083 spa->spa_ubsync = spa->spa_uberblock;
5742 6084
5743 6085 dsl_pool_sync_done(dp, txg);
5744 6086
5745 6087 /*
5746 6088 * Update usable space statistics.
5747 6089 */
5748 6090 while (vd = txg_list_remove(&spa->spa_vdev_txg_list, TXG_CLEAN(txg)))
5749 6091 vdev_sync_done(vd, txg);
5750 6092
5751 6093 spa_update_dspace(spa);
5752 6094
5753 6095 /*
5754 6096 * It had better be the case that we didn't dirty anything
5755 6097 * since vdev_config_sync().
5756 6098 */
5757 6099 ASSERT(txg_list_empty(&dp->dp_dirty_datasets, txg));
5758 6100 ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg));
5759 6101 ASSERT(txg_list_empty(&spa->spa_vdev_txg_list, txg));
5760 6102
5761 6103 spa->spa_sync_pass = 0;
5762 6104
5763 6105 spa_config_exit(spa, SCL_CONFIG, FTAG);
5764 6106
5765 6107 spa_handle_ignored_writes(spa);
5766 6108
5767 6109 /*
5768 6110 * If any async tasks have been requested, kick them off.
5769 6111 */
5770 6112 spa_async_dispatch(spa);
5771 6113 }
5772 6114
5773 6115 /*
5774 6116 * Sync all pools. We don't want to hold the namespace lock across these
5775 6117 * operations, so we take a reference on the spa_t and drop the lock during the
5776 6118 * sync.
5777 6119 */
5778 6120 void
5779 6121 spa_sync_allpools(void)
5780 6122 {
5781 6123 spa_t *spa = NULL;
5782 6124 mutex_enter(&spa_namespace_lock);
5783 6125 while ((spa = spa_next(spa)) != NULL) {
5784 6126 if (spa_state(spa) != POOL_STATE_ACTIVE ||
5785 6127 !spa_writeable(spa) || spa_suspended(spa))
5786 6128 continue;
5787 6129 spa_open_ref(spa, FTAG);
5788 6130 mutex_exit(&spa_namespace_lock);
5789 6131 txg_wait_synced(spa_get_dsl(spa), 0);
5790 6132 mutex_enter(&spa_namespace_lock);
5791 6133 spa_close(spa, FTAG);
5792 6134 }
5793 6135 mutex_exit(&spa_namespace_lock);
5794 6136 }
5795 6137
5796 6138 /*
5797 6139 * ==========================================================================
5798 6140 * Miscellaneous routines
5799 6141 * ==========================================================================
5800 6142 */
5801 6143
5802 6144 /*
5803 6145 * Remove all pools in the system.
5804 6146 */
5805 6147 void
5806 6148 spa_evict_all(void)
5807 6149 {
5808 6150 spa_t *spa;
5809 6151
5810 6152 /*
5811 6153 * Remove all cached state. All pools should be closed now,
5812 6154 * so every spa in the AVL tree should be unreferenced.
5813 6155 */
5814 6156 mutex_enter(&spa_namespace_lock);
5815 6157 while ((spa = spa_next(NULL)) != NULL) {
5816 6158 /*
5817 6159 * Stop async tasks. The async thread may need to detach
5818 6160 * a device that's been replaced, which requires grabbing
5819 6161 * spa_namespace_lock, so we must drop it here.
5820 6162 */
5821 6163 spa_open_ref(spa, FTAG);
5822 6164 mutex_exit(&spa_namespace_lock);
5823 6165 spa_async_suspend(spa);
5824 6166 mutex_enter(&spa_namespace_lock);
5825 6167 spa_close(spa, FTAG);
5826 6168
5827 6169 if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
5828 6170 spa_unload(spa);
5829 6171 spa_deactivate(spa);
5830 6172 }
5831 6173 spa_remove(spa);
5832 6174 }
5833 6175 mutex_exit(&spa_namespace_lock);
5834 6176 }
5835 6177
5836 6178 vdev_t *
5837 6179 spa_lookup_by_guid(spa_t *spa, uint64_t guid, boolean_t aux)
5838 6180 {
5839 6181 vdev_t *vd;
5840 6182 int i;
5841 6183
5842 6184 if ((vd = vdev_lookup_by_guid(spa->spa_root_vdev, guid)) != NULL)
5843 6185 return (vd);
5844 6186
5845 6187 if (aux) {
5846 6188 for (i = 0; i < spa->spa_l2cache.sav_count; i++) {
5847 6189 vd = spa->spa_l2cache.sav_vdevs[i];
5848 6190 if (vd->vdev_guid == guid)
5849 6191 return (vd);
5850 6192 }
5851 6193
5852 6194 for (i = 0; i < spa->spa_spares.sav_count; i++) {
5853 6195 vd = spa->spa_spares.sav_vdevs[i];
5854 6196 if (vd->vdev_guid == guid)
5855 6197 return (vd);
5856 6198 }
5857 6199 }
5858 6200
5859 6201 return (NULL);
5860 6202 }
5861 6203
5862 6204 void
5863 6205 spa_upgrade(spa_t *spa, uint64_t version)
5864 6206 {
5865 6207 ASSERT(spa_writeable(spa));
5866 6208
5867 6209 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5868 6210
5869 6211 /*
5870 6212 * This should only be called for a non-faulted pool, and since a
5871 6213 * future version would result in an unopenable pool, this shouldn't be
5872 6214 * possible.
5873 6215 */
5874 6216 ASSERT(spa->spa_uberblock.ub_version <= SPA_VERSION);
5875 6217 ASSERT(version >= spa->spa_uberblock.ub_version);
5876 6218
5877 6219 spa->spa_uberblock.ub_version = version;
5878 6220 vdev_config_dirty(spa->spa_root_vdev);
5879 6221
5880 6222 spa_config_exit(spa, SCL_ALL, FTAG);
5881 6223
5882 6224 txg_wait_synced(spa_get_dsl(spa), 0);
5883 6225 }
5884 6226
5885 6227 boolean_t
5886 6228 spa_has_spare(spa_t *spa, uint64_t guid)
5887 6229 {
5888 6230 int i;
5889 6231 uint64_t spareguid;
5890 6232 spa_aux_vdev_t *sav = &spa->spa_spares;
5891 6233
5892 6234 for (i = 0; i < sav->sav_count; i++)
5893 6235 if (sav->sav_vdevs[i]->vdev_guid == guid)
5894 6236 return (B_TRUE);
5895 6237
5896 6238 for (i = 0; i < sav->sav_npending; i++) {
5897 6239 if (nvlist_lookup_uint64(sav->sav_pending[i], ZPOOL_CONFIG_GUID,
5898 6240 &spareguid) == 0 && spareguid == guid)
5899 6241 return (B_TRUE);
5900 6242 }
5901 6243
5902 6244 return (B_FALSE);
5903 6245 }
5904 6246
5905 6247 /*
5906 6248 * Check if a pool has an active shared spare device.
5907 6249 * Note: reference count of an active spare is 2, as a spare and as a replace
5908 6250 */
5909 6251 static boolean_t
5910 6252 spa_has_active_shared_spare(spa_t *spa)
5911 6253 {
5912 6254 int i, refcnt;
5913 6255 uint64_t pool;
5914 6256 spa_aux_vdev_t *sav = &spa->spa_spares;
5915 6257
5916 6258 for (i = 0; i < sav->sav_count; i++) {
5917 6259 if (spa_spare_exists(sav->sav_vdevs[i]->vdev_guid, &pool,
5918 6260 &refcnt) && pool != 0ULL && pool == spa_guid(spa) &&
5919 6261 refcnt > 2)
5920 6262 return (B_TRUE);
5921 6263 }
5922 6264
5923 6265 return (B_FALSE);
5924 6266 }
5925 6267
5926 6268 /*
5927 6269 * Post a sysevent corresponding to the given event. The 'name' must be one of
5928 6270 * the event definitions in sys/sysevent/eventdefs.h. The payload will be
5929 6271 * filled in from the spa and (optionally) the vdev. This doesn't do anything
5930 6272 * in the userland libzpool, as we don't want consumers to misinterpret ztest
5931 6273 * or zdb as real changes.
5932 6274 */
5933 6275 void
5934 6276 spa_event_notify(spa_t *spa, vdev_t *vd, const char *name)
5935 6277 {
5936 6278 #ifdef _KERNEL
5937 6279 sysevent_t *ev;
5938 6280 sysevent_attr_list_t *attr = NULL;
5939 6281 sysevent_value_t value;
5940 6282 sysevent_id_t eid;
5941 6283
5942 6284 ev = sysevent_alloc(EC_ZFS, (char *)name, SUNW_KERN_PUB "zfs",
5943 6285 SE_SLEEP);
5944 6286
5945 6287 value.value_type = SE_DATA_TYPE_STRING;
5946 6288 value.value.sv_string = spa_name(spa);
5947 6289 if (sysevent_add_attr(&attr, ZFS_EV_POOL_NAME, &value, SE_SLEEP) != 0)
5948 6290 goto done;
5949 6291
5950 6292 value.value_type = SE_DATA_TYPE_UINT64;
5951 6293 value.value.sv_uint64 = spa_guid(spa);
5952 6294 if (sysevent_add_attr(&attr, ZFS_EV_POOL_GUID, &value, SE_SLEEP) != 0)
5953 6295 goto done;
5954 6296
5955 6297 if (vd) {
5956 6298 value.value_type = SE_DATA_TYPE_UINT64;
5957 6299 value.value.sv_uint64 = vd->vdev_guid;
5958 6300 if (sysevent_add_attr(&attr, ZFS_EV_VDEV_GUID, &value,
5959 6301 SE_SLEEP) != 0)
5960 6302 goto done;
5961 6303
5962 6304 if (vd->vdev_path) {
5963 6305 value.value_type = SE_DATA_TYPE_STRING;
5964 6306 value.value.sv_string = vd->vdev_path;
5965 6307 if (sysevent_add_attr(&attr, ZFS_EV_VDEV_PATH,
5966 6308 &value, SE_SLEEP) != 0)
5967 6309 goto done;
5968 6310 }
5969 6311 }
5970 6312
5971 6313 if (sysevent_attach_attributes(ev, attr) != 0)
5972 6314 goto done;
5973 6315 attr = NULL;
5974 6316
5975 6317 (void) log_sysevent(ev, SE_SLEEP, &eid);
5976 6318
5977 6319 done:
5978 6320 if (attr)
5979 6321 sysevent_free_attr(attr);
5980 6322 sysevent_free(ev);
5981 6323 #endif
5982 6324 }
|
↓ open down ↓ |
427 lines elided |
↑ open up ↑ |
XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX