Print this page
4185 New hash algorithm support
| Split |
Close |
| Expand all |
| Collapse all |
--- old/usr/src/uts/common/fs/zfs/spa_misc.c
+++ new/usr/src/uts/common/fs/zfs/spa_misc.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.
|
↓ open down ↓ |
14 lines elided |
↑ open up ↑ |
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 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23 23 * Copyright (c) 2013 by Delphix. All rights reserved.
24 24 * Copyright 2011 Nexenta Systems, Inc. All rights reserved.
25 + * Copyright 2013 Saso Kiselkov. All rights reserved.
25 26 */
26 27
27 28 #include <sys/zfs_context.h>
28 29 #include <sys/spa_impl.h>
29 30 #include <sys/spa_boot.h>
30 31 #include <sys/zio.h>
31 32 #include <sys/zio_checksum.h>
32 33 #include <sys/zio_compress.h>
33 34 #include <sys/dmu.h>
34 35 #include <sys/dmu_tx.h>
35 36 #include <sys/zap.h>
36 37 #include <sys/zil.h>
37 38 #include <sys/vdev_impl.h>
38 39 #include <sys/metaslab.h>
39 40 #include <sys/uberblock_impl.h>
40 41 #include <sys/txg.h>
41 42 #include <sys/avl.h>
|
↓ open down ↓ |
7 lines elided |
↑ open up ↑ |
42 43 #include <sys/unique.h>
43 44 #include <sys/dsl_pool.h>
44 45 #include <sys/dsl_dir.h>
45 46 #include <sys/dsl_prop.h>
46 47 #include <sys/dsl_scan.h>
47 48 #include <sys/fs/zfs.h>
48 49 #include <sys/metaslab_impl.h>
49 50 #include <sys/arc.h>
50 51 #include <sys/ddt.h>
51 52 #include "zfs_prop.h"
52 -#include "zfeature_common.h"
53 +#include <sys/zfeature.h>
53 54
54 55 /*
55 56 * SPA locking
56 57 *
57 58 * There are four basic locks for managing spa_t structures:
58 59 *
59 60 * spa_namespace_lock (global mutex)
60 61 *
61 62 * This lock must be acquired to do any of the following:
62 63 *
63 64 * - Lookup a spa_t by name
64 65 * - Add or remove a spa_t from the namespace
65 66 * - Increase spa_refcount from non-zero
66 67 * - Check if spa_refcount is zero
67 68 * - Rename a spa_t
68 69 * - add/remove/attach/detach devices
69 70 * - Held for the duration of create/destroy/import/export
70 71 *
71 72 * It does not need to handle recursion. A create or destroy may
72 73 * reference objects (files or zvols) in other pools, but by
73 74 * definition they must have an existing reference, and will never need
74 75 * to lookup a spa_t by name.
75 76 *
76 77 * spa_refcount (per-spa refcount_t protected by mutex)
77 78 *
78 79 * This reference count keep track of any active users of the spa_t. The
79 80 * spa_t cannot be destroyed or freed while this is non-zero. Internally,
80 81 * the refcount is never really 'zero' - opening a pool implicitly keeps
81 82 * some references in the DMU. Internally we check against spa_minref, but
82 83 * present the image of a zero/non-zero value to consumers.
83 84 *
84 85 * spa_config_lock[] (per-spa array of rwlocks)
85 86 *
86 87 * This protects the spa_t from config changes, and must be held in
87 88 * the following circumstances:
88 89 *
89 90 * - RW_READER to perform I/O to the spa
90 91 * - RW_WRITER to change the vdev config
91 92 *
92 93 * The locking order is fairly straightforward:
93 94 *
94 95 * spa_namespace_lock -> spa_refcount
95 96 *
96 97 * The namespace lock must be acquired to increase the refcount from 0
97 98 * or to check if it is zero.
98 99 *
99 100 * spa_refcount -> spa_config_lock[]
100 101 *
101 102 * There must be at least one valid reference on the spa_t to acquire
102 103 * the config lock.
103 104 *
104 105 * spa_namespace_lock -> spa_config_lock[]
105 106 *
106 107 * The namespace lock must always be taken before the config lock.
107 108 *
108 109 *
109 110 * The spa_namespace_lock can be acquired directly and is globally visible.
110 111 *
111 112 * The namespace is manipulated using the following functions, all of which
112 113 * require the spa_namespace_lock to be held.
113 114 *
114 115 * spa_lookup() Lookup a spa_t by name.
115 116 *
116 117 * spa_add() Create a new spa_t in the namespace.
117 118 *
118 119 * spa_remove() Remove a spa_t from the namespace. This also
119 120 * frees up any memory associated with the spa_t.
120 121 *
121 122 * spa_next() Returns the next spa_t in the system, or the
122 123 * first if NULL is passed.
123 124 *
124 125 * spa_evict_all() Shutdown and remove all spa_t structures in
125 126 * the system.
126 127 *
127 128 * spa_guid_exists() Determine whether a pool/device guid exists.
128 129 *
129 130 * The spa_refcount is manipulated using the following functions:
130 131 *
131 132 * spa_open_ref() Adds a reference to the given spa_t. Must be
132 133 * called with spa_namespace_lock held if the
133 134 * refcount is currently zero.
134 135 *
135 136 * spa_close() Remove a reference from the spa_t. This will
136 137 * not free the spa_t or remove it from the
137 138 * namespace. No locking is required.
138 139 *
139 140 * spa_refcount_zero() Returns true if the refcount is currently
140 141 * zero. Must be called with spa_namespace_lock
141 142 * held.
142 143 *
143 144 * The spa_config_lock[] is an array of rwlocks, ordered as follows:
144 145 * SCL_CONFIG > SCL_STATE > SCL_ALLOC > SCL_ZIO > SCL_FREE > SCL_VDEV.
145 146 * spa_config_lock[] is manipulated with spa_config_{enter,exit,held}().
146 147 *
147 148 * To read the configuration, it suffices to hold one of these locks as reader.
148 149 * To modify the configuration, you must hold all locks as writer. To modify
149 150 * vdev state without altering the vdev tree's topology (e.g. online/offline),
150 151 * you must hold SCL_STATE and SCL_ZIO as writer.
151 152 *
152 153 * We use these distinct config locks to avoid recursive lock entry.
153 154 * For example, spa_sync() (which holds SCL_CONFIG as reader) induces
154 155 * block allocations (SCL_ALLOC), which may require reading space maps
155 156 * from disk (dmu_read() -> zio_read() -> SCL_ZIO).
156 157 *
157 158 * The spa config locks cannot be normal rwlocks because we need the
158 159 * ability to hand off ownership. For example, SCL_ZIO is acquired
159 160 * by the issuing thread and later released by an interrupt thread.
160 161 * They do, however, obey the usual write-wanted semantics to prevent
161 162 * writer (i.e. system administrator) starvation.
162 163 *
163 164 * The lock acquisition rules are as follows:
164 165 *
165 166 * SCL_CONFIG
166 167 * Protects changes to the vdev tree topology, such as vdev
167 168 * add/remove/attach/detach. Protects the dirty config list
168 169 * (spa_config_dirty_list) and the set of spares and l2arc devices.
169 170 *
170 171 * SCL_STATE
171 172 * Protects changes to pool state and vdev state, such as vdev
172 173 * online/offline/fault/degrade/clear. Protects the dirty state list
173 174 * (spa_state_dirty_list) and global pool state (spa_state).
174 175 *
175 176 * SCL_ALLOC
176 177 * Protects changes to metaslab groups and classes.
177 178 * Held as reader by metaslab_alloc() and metaslab_claim().
178 179 *
179 180 * SCL_ZIO
180 181 * Held by bp-level zios (those which have no io_vd upon entry)
181 182 * to prevent changes to the vdev tree. The bp-level zio implicitly
182 183 * protects all of its vdev child zios, which do not hold SCL_ZIO.
183 184 *
184 185 * SCL_FREE
185 186 * Protects changes to metaslab groups and classes.
186 187 * Held as reader by metaslab_free(). SCL_FREE is distinct from
187 188 * SCL_ALLOC, and lower than SCL_ZIO, so that we can safely free
188 189 * blocks in zio_done() while another i/o that holds either
189 190 * SCL_ALLOC or SCL_ZIO is waiting for this i/o to complete.
190 191 *
191 192 * SCL_VDEV
192 193 * Held as reader to prevent changes to the vdev tree during trivial
193 194 * inquiries such as bp_get_dsize(). SCL_VDEV is distinct from the
194 195 * other locks, and lower than all of them, to ensure that it's safe
195 196 * to acquire regardless of caller context.
196 197 *
197 198 * In addition, the following rules apply:
198 199 *
199 200 * (a) spa_props_lock protects pool properties, spa_config and spa_config_list.
200 201 * The lock ordering is SCL_CONFIG > spa_props_lock.
201 202 *
202 203 * (b) I/O operations on leaf vdevs. For any zio operation that takes
203 204 * an explicit vdev_t argument -- such as zio_ioctl(), zio_read_phys(),
204 205 * or zio_write_phys() -- the caller must ensure that the config cannot
205 206 * cannot change in the interim, and that the vdev cannot be reopened.
206 207 * SCL_STATE as reader suffices for both.
207 208 *
208 209 * The vdev configuration is protected by spa_vdev_enter() / spa_vdev_exit().
209 210 *
210 211 * spa_vdev_enter() Acquire the namespace lock and the config lock
211 212 * for writing.
212 213 *
213 214 * spa_vdev_exit() Release the config lock, wait for all I/O
214 215 * to complete, sync the updated configs to the
215 216 * cache, and release the namespace lock.
216 217 *
217 218 * vdev state is protected by spa_vdev_state_enter() / spa_vdev_state_exit().
218 219 * Like spa_vdev_enter/exit, these are convenience wrappers -- the actual
219 220 * locking is, always, based on spa_namespace_lock and spa_config_lock[].
220 221 *
221 222 * spa_rename() is also implemented within this file since it requires
222 223 * manipulation of the namespace.
223 224 */
224 225
225 226 static avl_tree_t spa_namespace_avl;
226 227 kmutex_t spa_namespace_lock;
227 228 static kcondvar_t spa_namespace_cv;
228 229 static int spa_active_count;
229 230 int spa_max_replication_override = SPA_DVAS_PER_BP;
230 231
231 232 static kmutex_t spa_spare_lock;
232 233 static avl_tree_t spa_spare_avl;
233 234 static kmutex_t spa_l2cache_lock;
234 235 static avl_tree_t spa_l2cache_avl;
235 236
236 237 kmem_cache_t *spa_buffer_pool;
237 238 int spa_mode_global;
238 239
239 240 #ifdef ZFS_DEBUG
240 241 /* Everything except dprintf and spa is on by default in debug builds */
241 242 int zfs_flags = ~(ZFS_DEBUG_DPRINTF | ZFS_DEBUG_SPA);
242 243 #else
243 244 int zfs_flags = 0;
244 245 #endif
245 246
246 247 /*
247 248 * zfs_recover can be set to nonzero to attempt to recover from
248 249 * otherwise-fatal errors, typically caused by on-disk corruption. When
249 250 * set, calls to zfs_panic_recover() will turn into warning messages.
250 251 */
251 252 int zfs_recover = 0;
252 253
253 254 /*
254 255 * Expiration time in milliseconds. This value has two meanings. First it is
255 256 * used to determine when the spa_deadman() logic should fire. By default the
256 257 * spa_deadman() will fire if spa_sync() has not completed in 1000 seconds.
257 258 * Secondly, the value determines if an I/O is considered "hung". Any I/O that
258 259 * has not completed in zfs_deadman_synctime_ms is considered "hung" resulting
259 260 * in a system panic.
260 261 */
261 262 uint64_t zfs_deadman_synctime_ms = 1000000ULL;
262 263
263 264 /*
264 265 * Check time in milliseconds. This defines the frequency at which we check
265 266 * for hung I/O.
266 267 */
267 268 uint64_t zfs_deadman_checktime_ms = 5000ULL;
268 269
269 270 /*
270 271 * Override the zfs deadman behavior via /etc/system. By default the
271 272 * deadman is enabled except on VMware and sparc deployments.
272 273 */
273 274 int zfs_deadman_enabled = -1;
274 275
275 276 /*
276 277 * The worst case is single-sector max-parity RAID-Z blocks, in which
277 278 * case the space requirement is exactly (VDEV_RAIDZ_MAXPARITY + 1)
278 279 * times the size; so just assume that. Add to this the fact that
279 280 * we can have up to 3 DVAs per bp, and one more factor of 2 because
280 281 * the block may be dittoed with up to 3 DVAs by ddt_sync(). All together,
281 282 * the worst case is:
282 283 * (VDEV_RAIDZ_MAXPARITY + 1) * SPA_DVAS_PER_BP * 2 == 24
283 284 */
284 285 int spa_asize_inflation = 24;
285 286
286 287 /*
287 288 * ==========================================================================
288 289 * SPA config locking
289 290 * ==========================================================================
290 291 */
291 292 static void
292 293 spa_config_lock_init(spa_t *spa)
293 294 {
294 295 for (int i = 0; i < SCL_LOCKS; i++) {
295 296 spa_config_lock_t *scl = &spa->spa_config_lock[i];
296 297 mutex_init(&scl->scl_lock, NULL, MUTEX_DEFAULT, NULL);
297 298 cv_init(&scl->scl_cv, NULL, CV_DEFAULT, NULL);
298 299 refcount_create_untracked(&scl->scl_count);
299 300 scl->scl_writer = NULL;
300 301 scl->scl_write_wanted = 0;
301 302 }
302 303 }
303 304
304 305 static void
305 306 spa_config_lock_destroy(spa_t *spa)
306 307 {
307 308 for (int i = 0; i < SCL_LOCKS; i++) {
308 309 spa_config_lock_t *scl = &spa->spa_config_lock[i];
309 310 mutex_destroy(&scl->scl_lock);
310 311 cv_destroy(&scl->scl_cv);
311 312 refcount_destroy(&scl->scl_count);
312 313 ASSERT(scl->scl_writer == NULL);
313 314 ASSERT(scl->scl_write_wanted == 0);
314 315 }
315 316 }
316 317
317 318 int
318 319 spa_config_tryenter(spa_t *spa, int locks, void *tag, krw_t rw)
319 320 {
320 321 for (int i = 0; i < SCL_LOCKS; i++) {
321 322 spa_config_lock_t *scl = &spa->spa_config_lock[i];
322 323 if (!(locks & (1 << i)))
323 324 continue;
324 325 mutex_enter(&scl->scl_lock);
325 326 if (rw == RW_READER) {
326 327 if (scl->scl_writer || scl->scl_write_wanted) {
327 328 mutex_exit(&scl->scl_lock);
328 329 spa_config_exit(spa, locks ^ (1 << i), tag);
329 330 return (0);
330 331 }
331 332 } else {
332 333 ASSERT(scl->scl_writer != curthread);
333 334 if (!refcount_is_zero(&scl->scl_count)) {
334 335 mutex_exit(&scl->scl_lock);
335 336 spa_config_exit(spa, locks ^ (1 << i), tag);
336 337 return (0);
337 338 }
338 339 scl->scl_writer = curthread;
339 340 }
340 341 (void) refcount_add(&scl->scl_count, tag);
341 342 mutex_exit(&scl->scl_lock);
342 343 }
343 344 return (1);
344 345 }
345 346
346 347 void
347 348 spa_config_enter(spa_t *spa, int locks, void *tag, krw_t rw)
348 349 {
349 350 int wlocks_held = 0;
350 351
351 352 ASSERT3U(SCL_LOCKS, <, sizeof (wlocks_held) * NBBY);
352 353
353 354 for (int i = 0; i < SCL_LOCKS; i++) {
354 355 spa_config_lock_t *scl = &spa->spa_config_lock[i];
355 356 if (scl->scl_writer == curthread)
356 357 wlocks_held |= (1 << i);
357 358 if (!(locks & (1 << i)))
358 359 continue;
359 360 mutex_enter(&scl->scl_lock);
360 361 if (rw == RW_READER) {
361 362 while (scl->scl_writer || scl->scl_write_wanted) {
362 363 cv_wait(&scl->scl_cv, &scl->scl_lock);
363 364 }
364 365 } else {
365 366 ASSERT(scl->scl_writer != curthread);
366 367 while (!refcount_is_zero(&scl->scl_count)) {
367 368 scl->scl_write_wanted++;
368 369 cv_wait(&scl->scl_cv, &scl->scl_lock);
369 370 scl->scl_write_wanted--;
370 371 }
371 372 scl->scl_writer = curthread;
372 373 }
373 374 (void) refcount_add(&scl->scl_count, tag);
374 375 mutex_exit(&scl->scl_lock);
375 376 }
376 377 ASSERT(wlocks_held <= locks);
377 378 }
378 379
379 380 void
380 381 spa_config_exit(spa_t *spa, int locks, void *tag)
381 382 {
382 383 for (int i = SCL_LOCKS - 1; i >= 0; i--) {
383 384 spa_config_lock_t *scl = &spa->spa_config_lock[i];
384 385 if (!(locks & (1 << i)))
385 386 continue;
386 387 mutex_enter(&scl->scl_lock);
387 388 ASSERT(!refcount_is_zero(&scl->scl_count));
388 389 if (refcount_remove(&scl->scl_count, tag) == 0) {
389 390 ASSERT(scl->scl_writer == NULL ||
390 391 scl->scl_writer == curthread);
391 392 scl->scl_writer = NULL; /* OK in either case */
392 393 cv_broadcast(&scl->scl_cv);
393 394 }
394 395 mutex_exit(&scl->scl_lock);
395 396 }
396 397 }
397 398
398 399 int
399 400 spa_config_held(spa_t *spa, int locks, krw_t rw)
400 401 {
401 402 int locks_held = 0;
402 403
403 404 for (int i = 0; i < SCL_LOCKS; i++) {
404 405 spa_config_lock_t *scl = &spa->spa_config_lock[i];
405 406 if (!(locks & (1 << i)))
406 407 continue;
407 408 if ((rw == RW_READER && !refcount_is_zero(&scl->scl_count)) ||
408 409 (rw == RW_WRITER && scl->scl_writer == curthread))
409 410 locks_held |= 1 << i;
410 411 }
411 412
412 413 return (locks_held);
413 414 }
414 415
415 416 /*
416 417 * ==========================================================================
417 418 * SPA namespace functions
418 419 * ==========================================================================
419 420 */
420 421
421 422 /*
422 423 * Lookup the named spa_t in the AVL tree. The spa_namespace_lock must be held.
423 424 * Returns NULL if no matching spa_t is found.
424 425 */
425 426 spa_t *
426 427 spa_lookup(const char *name)
427 428 {
428 429 static spa_t search; /* spa_t is large; don't allocate on stack */
429 430 spa_t *spa;
430 431 avl_index_t where;
431 432 char *cp;
432 433
433 434 ASSERT(MUTEX_HELD(&spa_namespace_lock));
434 435
435 436 (void) strlcpy(search.spa_name, name, sizeof (search.spa_name));
436 437
437 438 /*
438 439 * If it's a full dataset name, figure out the pool name and
439 440 * just use that.
440 441 */
441 442 cp = strpbrk(search.spa_name, "/@");
442 443 if (cp != NULL)
443 444 *cp = '\0';
444 445
445 446 spa = avl_find(&spa_namespace_avl, &search, &where);
446 447
447 448 return (spa);
448 449 }
449 450
450 451 /*
451 452 * Fires when spa_sync has not completed within zfs_deadman_synctime_ms.
452 453 * If the zfs_deadman_enabled flag is set then it inspects all vdev queues
453 454 * looking for potentially hung I/Os.
454 455 */
455 456 void
456 457 spa_deadman(void *arg)
457 458 {
458 459 spa_t *spa = arg;
459 460
460 461 /*
461 462 * Disable the deadman timer if the pool is suspended.
462 463 */
463 464 if (spa_suspended(spa)) {
464 465 VERIFY(cyclic_reprogram(spa->spa_deadman_cycid, CY_INFINITY));
465 466 return;
466 467 }
467 468
468 469 zfs_dbgmsg("slow spa_sync: started %llu seconds ago, calls %llu",
469 470 (gethrtime() - spa->spa_sync_starttime) / NANOSEC,
470 471 ++spa->spa_deadman_calls);
471 472 if (zfs_deadman_enabled)
472 473 vdev_deadman(spa->spa_root_vdev);
473 474 }
474 475
475 476 /*
476 477 * Create an uninitialized spa_t with the given name. Requires
477 478 * spa_namespace_lock. The caller must ensure that the spa_t doesn't already
478 479 * exist by calling spa_lookup() first.
479 480 */
480 481 spa_t *
481 482 spa_add(const char *name, nvlist_t *config, const char *altroot)
482 483 {
483 484 spa_t *spa;
484 485 spa_config_dirent_t *dp;
485 486 cyc_handler_t hdlr;
486 487 cyc_time_t when;
487 488
|
↓ open down ↓ |
425 lines elided |
↑ open up ↑ |
488 489 ASSERT(MUTEX_HELD(&spa_namespace_lock));
489 490
490 491 spa = kmem_zalloc(sizeof (spa_t), KM_SLEEP);
491 492
492 493 mutex_init(&spa->spa_async_lock, NULL, MUTEX_DEFAULT, NULL);
493 494 mutex_init(&spa->spa_errlist_lock, NULL, MUTEX_DEFAULT, NULL);
494 495 mutex_init(&spa->spa_errlog_lock, NULL, MUTEX_DEFAULT, NULL);
495 496 mutex_init(&spa->spa_history_lock, NULL, MUTEX_DEFAULT, NULL);
496 497 mutex_init(&spa->spa_proc_lock, NULL, MUTEX_DEFAULT, NULL);
497 498 mutex_init(&spa->spa_props_lock, NULL, MUTEX_DEFAULT, NULL);
499 + mutex_init(&spa->spa_cksum_tmpls_lock, NULL, MUTEX_DEFAULT, NULL);
498 500 mutex_init(&spa->spa_scrub_lock, NULL, MUTEX_DEFAULT, NULL);
499 501 mutex_init(&spa->spa_suspend_lock, NULL, MUTEX_DEFAULT, NULL);
500 502 mutex_init(&spa->spa_vdev_top_lock, NULL, MUTEX_DEFAULT, NULL);
501 503 mutex_init(&spa->spa_iokstat_lock, NULL, MUTEX_DEFAULT, NULL);
502 504
503 505 cv_init(&spa->spa_async_cv, NULL, CV_DEFAULT, NULL);
504 506 cv_init(&spa->spa_proc_cv, NULL, CV_DEFAULT, NULL);
505 507 cv_init(&spa->spa_scrub_io_cv, NULL, CV_DEFAULT, NULL);
506 508 cv_init(&spa->spa_suspend_cv, NULL, CV_DEFAULT, NULL);
507 509
508 510 for (int t = 0; t < TXG_SIZE; t++)
509 511 bplist_create(&spa->spa_free_bplist[t]);
510 512
511 513 (void) strlcpy(spa->spa_name, name, sizeof (spa->spa_name));
512 514 spa->spa_state = POOL_STATE_UNINITIALIZED;
513 515 spa->spa_freeze_txg = UINT64_MAX;
514 516 spa->spa_final_txg = UINT64_MAX;
515 517 spa->spa_load_max_txg = UINT64_MAX;
516 518 spa->spa_proc = &p0;
517 519 spa->spa_proc_state = SPA_PROC_NONE;
518 520
519 521 hdlr.cyh_func = spa_deadman;
520 522 hdlr.cyh_arg = spa;
521 523 hdlr.cyh_level = CY_LOW_LEVEL;
522 524
523 525 spa->spa_deadman_synctime = MSEC2NSEC(zfs_deadman_synctime_ms);
524 526
525 527 /*
526 528 * This determines how often we need to check for hung I/Os after
527 529 * the cyclic has already fired. Since checking for hung I/Os is
528 530 * an expensive operation we don't want to check too frequently.
529 531 * Instead wait for 5 seconds before checking again.
530 532 */
531 533 when.cyt_interval = MSEC2NSEC(zfs_deadman_checktime_ms);
532 534 when.cyt_when = CY_INFINITY;
533 535 mutex_enter(&cpu_lock);
534 536 spa->spa_deadman_cycid = cyclic_add(&hdlr, &when);
535 537 mutex_exit(&cpu_lock);
536 538
537 539 refcount_create(&spa->spa_refcount);
538 540 spa_config_lock_init(spa);
539 541
540 542 avl_add(&spa_namespace_avl, spa);
541 543
542 544 /*
543 545 * Set the alternate root, if there is one.
544 546 */
545 547 if (altroot) {
546 548 spa->spa_root = spa_strdup(altroot);
547 549 spa_active_count++;
548 550 }
549 551
550 552 /*
551 553 * Every pool starts with the default cachefile
552 554 */
553 555 list_create(&spa->spa_config_list, sizeof (spa_config_dirent_t),
554 556 offsetof(spa_config_dirent_t, scd_link));
555 557
556 558 dp = kmem_zalloc(sizeof (spa_config_dirent_t), KM_SLEEP);
557 559 dp->scd_path = altroot ? NULL : spa_strdup(spa_config_path);
558 560 list_insert_head(&spa->spa_config_list, dp);
559 561
560 562 VERIFY(nvlist_alloc(&spa->spa_load_info, NV_UNIQUE_NAME,
561 563 KM_SLEEP) == 0);
562 564
563 565 if (config != NULL) {
564 566 nvlist_t *features;
565 567
566 568 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_FEATURES_FOR_READ,
567 569 &features) == 0) {
568 570 VERIFY(nvlist_dup(features, &spa->spa_label_features,
569 571 0) == 0);
570 572 }
571 573
572 574 VERIFY(nvlist_dup(config, &spa->spa_config, 0) == 0);
573 575 }
574 576
575 577 if (spa->spa_label_features == NULL) {
576 578 VERIFY(nvlist_alloc(&spa->spa_label_features, NV_UNIQUE_NAME,
577 579 KM_SLEEP) == 0);
578 580 }
579 581
580 582 spa->spa_iokstat = kstat_create("zfs", 0, name,
581 583 "disk", KSTAT_TYPE_IO, 1, 0);
582 584 if (spa->spa_iokstat) {
583 585 spa->spa_iokstat->ks_lock = &spa->spa_iokstat_lock;
584 586 kstat_install(spa->spa_iokstat);
585 587 }
586 588
587 589 spa->spa_debug = ((zfs_flags & ZFS_DEBUG_SPA) != 0);
588 590
589 591 return (spa);
590 592 }
591 593
592 594 /*
593 595 * Removes a spa_t from the namespace, freeing up any memory used. Requires
594 596 * spa_namespace_lock. This is called only after the spa_t has been closed and
595 597 * deactivated.
596 598 */
597 599 void
598 600 spa_remove(spa_t *spa)
599 601 {
600 602 spa_config_dirent_t *dp;
601 603
602 604 ASSERT(MUTEX_HELD(&spa_namespace_lock));
603 605 ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED);
604 606
605 607 nvlist_free(spa->spa_config_splitting);
606 608
607 609 avl_remove(&spa_namespace_avl, spa);
608 610 cv_broadcast(&spa_namespace_cv);
609 611
610 612 if (spa->spa_root) {
611 613 spa_strfree(spa->spa_root);
612 614 spa_active_count--;
613 615 }
614 616
615 617 while ((dp = list_head(&spa->spa_config_list)) != NULL) {
616 618 list_remove(&spa->spa_config_list, dp);
617 619 if (dp->scd_path != NULL)
618 620 spa_strfree(dp->scd_path);
619 621 kmem_free(dp, sizeof (spa_config_dirent_t));
620 622 }
621 623
622 624 list_destroy(&spa->spa_config_list);
623 625
624 626 nvlist_free(spa->spa_label_features);
625 627 nvlist_free(spa->spa_load_info);
626 628 spa_config_set(spa, NULL);
627 629
628 630 mutex_enter(&cpu_lock);
629 631 if (spa->spa_deadman_cycid != CYCLIC_NONE)
630 632 cyclic_remove(spa->spa_deadman_cycid);
631 633 mutex_exit(&cpu_lock);
632 634 spa->spa_deadman_cycid = CYCLIC_NONE;
633 635
|
↓ open down ↓ |
126 lines elided |
↑ open up ↑ |
634 636 refcount_destroy(&spa->spa_refcount);
635 637
636 638 spa_config_lock_destroy(spa);
637 639
638 640 kstat_delete(spa->spa_iokstat);
639 641 spa->spa_iokstat = NULL;
640 642
641 643 for (int t = 0; t < TXG_SIZE; t++)
642 644 bplist_destroy(&spa->spa_free_bplist[t]);
643 645
646 + zio_checksum_templates_free(spa);
647 +
644 648 cv_destroy(&spa->spa_async_cv);
645 649 cv_destroy(&spa->spa_proc_cv);
646 650 cv_destroy(&spa->spa_scrub_io_cv);
647 651 cv_destroy(&spa->spa_suspend_cv);
648 652
649 653 mutex_destroy(&spa->spa_async_lock);
650 654 mutex_destroy(&spa->spa_errlist_lock);
651 655 mutex_destroy(&spa->spa_errlog_lock);
652 656 mutex_destroy(&spa->spa_history_lock);
653 657 mutex_destroy(&spa->spa_proc_lock);
654 658 mutex_destroy(&spa->spa_props_lock);
659 + mutex_destroy(&spa->spa_cksum_tmpls_lock);
655 660 mutex_destroy(&spa->spa_scrub_lock);
656 661 mutex_destroy(&spa->spa_suspend_lock);
657 662 mutex_destroy(&spa->spa_vdev_top_lock);
658 663 mutex_destroy(&spa->spa_iokstat_lock);
659 664
660 665 kmem_free(spa, sizeof (spa_t));
661 666 }
662 667
663 668 /*
664 669 * Given a pool, return the next pool in the namespace, or NULL if there is
665 670 * none. If 'prev' is NULL, return the first pool.
666 671 */
667 672 spa_t *
668 673 spa_next(spa_t *prev)
669 674 {
670 675 ASSERT(MUTEX_HELD(&spa_namespace_lock));
671 676
672 677 if (prev)
673 678 return (AVL_NEXT(&spa_namespace_avl, prev));
674 679 else
675 680 return (avl_first(&spa_namespace_avl));
676 681 }
677 682
678 683 /*
679 684 * ==========================================================================
680 685 * SPA refcount functions
681 686 * ==========================================================================
682 687 */
683 688
684 689 /*
685 690 * Add a reference to the given spa_t. Must have at least one reference, or
686 691 * have the namespace lock held.
687 692 */
688 693 void
689 694 spa_open_ref(spa_t *spa, void *tag)
690 695 {
691 696 ASSERT(refcount_count(&spa->spa_refcount) >= spa->spa_minref ||
692 697 MUTEX_HELD(&spa_namespace_lock));
693 698 (void) refcount_add(&spa->spa_refcount, tag);
694 699 }
695 700
696 701 /*
697 702 * Remove a reference to the given spa_t. Must have at least one reference, or
698 703 * have the namespace lock held.
699 704 */
700 705 void
701 706 spa_close(spa_t *spa, void *tag)
702 707 {
703 708 ASSERT(refcount_count(&spa->spa_refcount) > spa->spa_minref ||
704 709 MUTEX_HELD(&spa_namespace_lock));
705 710 (void) refcount_remove(&spa->spa_refcount, tag);
706 711 }
707 712
708 713 /*
709 714 * Check to see if the spa refcount is zero. Must be called with
710 715 * spa_namespace_lock held. We really compare against spa_minref, which is the
711 716 * number of references acquired when opening a pool
712 717 */
713 718 boolean_t
714 719 spa_refcount_zero(spa_t *spa)
715 720 {
716 721 ASSERT(MUTEX_HELD(&spa_namespace_lock));
717 722
718 723 return (refcount_count(&spa->spa_refcount) == spa->spa_minref);
719 724 }
720 725
721 726 /*
722 727 * ==========================================================================
723 728 * SPA spare and l2cache tracking
724 729 * ==========================================================================
725 730 */
726 731
727 732 /*
728 733 * Hot spares and cache devices are tracked using the same code below,
729 734 * for 'auxiliary' devices.
730 735 */
731 736
732 737 typedef struct spa_aux {
733 738 uint64_t aux_guid;
734 739 uint64_t aux_pool;
735 740 avl_node_t aux_avl;
736 741 int aux_count;
737 742 } spa_aux_t;
738 743
739 744 static int
740 745 spa_aux_compare(const void *a, const void *b)
741 746 {
742 747 const spa_aux_t *sa = a;
743 748 const spa_aux_t *sb = b;
744 749
745 750 if (sa->aux_guid < sb->aux_guid)
746 751 return (-1);
747 752 else if (sa->aux_guid > sb->aux_guid)
748 753 return (1);
749 754 else
750 755 return (0);
751 756 }
752 757
753 758 void
754 759 spa_aux_add(vdev_t *vd, avl_tree_t *avl)
755 760 {
756 761 avl_index_t where;
757 762 spa_aux_t search;
758 763 spa_aux_t *aux;
759 764
760 765 search.aux_guid = vd->vdev_guid;
761 766 if ((aux = avl_find(avl, &search, &where)) != NULL) {
762 767 aux->aux_count++;
763 768 } else {
764 769 aux = kmem_zalloc(sizeof (spa_aux_t), KM_SLEEP);
765 770 aux->aux_guid = vd->vdev_guid;
766 771 aux->aux_count = 1;
767 772 avl_insert(avl, aux, where);
768 773 }
769 774 }
770 775
771 776 void
772 777 spa_aux_remove(vdev_t *vd, avl_tree_t *avl)
773 778 {
774 779 spa_aux_t search;
775 780 spa_aux_t *aux;
776 781 avl_index_t where;
777 782
778 783 search.aux_guid = vd->vdev_guid;
779 784 aux = avl_find(avl, &search, &where);
780 785
781 786 ASSERT(aux != NULL);
782 787
783 788 if (--aux->aux_count == 0) {
784 789 avl_remove(avl, aux);
785 790 kmem_free(aux, sizeof (spa_aux_t));
786 791 } else if (aux->aux_pool == spa_guid(vd->vdev_spa)) {
787 792 aux->aux_pool = 0ULL;
788 793 }
789 794 }
790 795
791 796 boolean_t
792 797 spa_aux_exists(uint64_t guid, uint64_t *pool, int *refcnt, avl_tree_t *avl)
793 798 {
794 799 spa_aux_t search, *found;
795 800
796 801 search.aux_guid = guid;
797 802 found = avl_find(avl, &search, NULL);
798 803
799 804 if (pool) {
800 805 if (found)
801 806 *pool = found->aux_pool;
802 807 else
803 808 *pool = 0ULL;
804 809 }
805 810
806 811 if (refcnt) {
807 812 if (found)
808 813 *refcnt = found->aux_count;
809 814 else
810 815 *refcnt = 0;
811 816 }
812 817
813 818 return (found != NULL);
814 819 }
815 820
816 821 void
817 822 spa_aux_activate(vdev_t *vd, avl_tree_t *avl)
818 823 {
819 824 spa_aux_t search, *found;
820 825 avl_index_t where;
821 826
822 827 search.aux_guid = vd->vdev_guid;
823 828 found = avl_find(avl, &search, &where);
824 829 ASSERT(found != NULL);
825 830 ASSERT(found->aux_pool == 0ULL);
826 831
827 832 found->aux_pool = spa_guid(vd->vdev_spa);
828 833 }
829 834
830 835 /*
831 836 * Spares are tracked globally due to the following constraints:
832 837 *
833 838 * - A spare may be part of multiple pools.
834 839 * - A spare may be added to a pool even if it's actively in use within
835 840 * another pool.
836 841 * - A spare in use in any pool can only be the source of a replacement if
837 842 * the target is a spare in the same pool.
838 843 *
839 844 * We keep track of all spares on the system through the use of a reference
840 845 * counted AVL tree. When a vdev is added as a spare, or used as a replacement
841 846 * spare, then we bump the reference count in the AVL tree. In addition, we set
842 847 * the 'vdev_isspare' member to indicate that the device is a spare (active or
843 848 * inactive). When a spare is made active (used to replace a device in the
844 849 * pool), we also keep track of which pool its been made a part of.
845 850 *
846 851 * The 'spa_spare_lock' protects the AVL tree. These functions are normally
847 852 * called under the spa_namespace lock as part of vdev reconfiguration. The
848 853 * separate spare lock exists for the status query path, which does not need to
849 854 * be completely consistent with respect to other vdev configuration changes.
850 855 */
851 856
852 857 static int
853 858 spa_spare_compare(const void *a, const void *b)
854 859 {
855 860 return (spa_aux_compare(a, b));
856 861 }
857 862
858 863 void
859 864 spa_spare_add(vdev_t *vd)
860 865 {
861 866 mutex_enter(&spa_spare_lock);
862 867 ASSERT(!vd->vdev_isspare);
863 868 spa_aux_add(vd, &spa_spare_avl);
864 869 vd->vdev_isspare = B_TRUE;
865 870 mutex_exit(&spa_spare_lock);
866 871 }
867 872
868 873 void
869 874 spa_spare_remove(vdev_t *vd)
870 875 {
871 876 mutex_enter(&spa_spare_lock);
872 877 ASSERT(vd->vdev_isspare);
873 878 spa_aux_remove(vd, &spa_spare_avl);
874 879 vd->vdev_isspare = B_FALSE;
875 880 mutex_exit(&spa_spare_lock);
876 881 }
877 882
878 883 boolean_t
879 884 spa_spare_exists(uint64_t guid, uint64_t *pool, int *refcnt)
880 885 {
881 886 boolean_t found;
882 887
883 888 mutex_enter(&spa_spare_lock);
884 889 found = spa_aux_exists(guid, pool, refcnt, &spa_spare_avl);
885 890 mutex_exit(&spa_spare_lock);
886 891
887 892 return (found);
888 893 }
889 894
890 895 void
891 896 spa_spare_activate(vdev_t *vd)
892 897 {
893 898 mutex_enter(&spa_spare_lock);
894 899 ASSERT(vd->vdev_isspare);
895 900 spa_aux_activate(vd, &spa_spare_avl);
896 901 mutex_exit(&spa_spare_lock);
897 902 }
898 903
899 904 /*
900 905 * Level 2 ARC devices are tracked globally for the same reasons as spares.
901 906 * Cache devices currently only support one pool per cache device, and so
902 907 * for these devices the aux reference count is currently unused beyond 1.
903 908 */
904 909
905 910 static int
906 911 spa_l2cache_compare(const void *a, const void *b)
907 912 {
908 913 return (spa_aux_compare(a, b));
909 914 }
910 915
911 916 void
912 917 spa_l2cache_add(vdev_t *vd)
913 918 {
914 919 mutex_enter(&spa_l2cache_lock);
915 920 ASSERT(!vd->vdev_isl2cache);
916 921 spa_aux_add(vd, &spa_l2cache_avl);
917 922 vd->vdev_isl2cache = B_TRUE;
918 923 mutex_exit(&spa_l2cache_lock);
919 924 }
920 925
921 926 void
922 927 spa_l2cache_remove(vdev_t *vd)
923 928 {
924 929 mutex_enter(&spa_l2cache_lock);
925 930 ASSERT(vd->vdev_isl2cache);
926 931 spa_aux_remove(vd, &spa_l2cache_avl);
927 932 vd->vdev_isl2cache = B_FALSE;
928 933 mutex_exit(&spa_l2cache_lock);
929 934 }
930 935
931 936 boolean_t
932 937 spa_l2cache_exists(uint64_t guid, uint64_t *pool)
933 938 {
934 939 boolean_t found;
935 940
936 941 mutex_enter(&spa_l2cache_lock);
937 942 found = spa_aux_exists(guid, pool, NULL, &spa_l2cache_avl);
938 943 mutex_exit(&spa_l2cache_lock);
939 944
940 945 return (found);
941 946 }
942 947
943 948 void
944 949 spa_l2cache_activate(vdev_t *vd)
945 950 {
946 951 mutex_enter(&spa_l2cache_lock);
947 952 ASSERT(vd->vdev_isl2cache);
948 953 spa_aux_activate(vd, &spa_l2cache_avl);
949 954 mutex_exit(&spa_l2cache_lock);
950 955 }
951 956
952 957 /*
953 958 * ==========================================================================
954 959 * SPA vdev locking
955 960 * ==========================================================================
956 961 */
957 962
958 963 /*
959 964 * Lock the given spa_t for the purpose of adding or removing a vdev.
960 965 * Grabs the global spa_namespace_lock plus the spa config lock for writing.
961 966 * It returns the next transaction group for the spa_t.
962 967 */
963 968 uint64_t
964 969 spa_vdev_enter(spa_t *spa)
965 970 {
966 971 mutex_enter(&spa->spa_vdev_top_lock);
967 972 mutex_enter(&spa_namespace_lock);
968 973 return (spa_vdev_config_enter(spa));
969 974 }
970 975
971 976 /*
972 977 * Internal implementation for spa_vdev_enter(). Used when a vdev
973 978 * operation requires multiple syncs (i.e. removing a device) while
974 979 * keeping the spa_namespace_lock held.
975 980 */
976 981 uint64_t
977 982 spa_vdev_config_enter(spa_t *spa)
978 983 {
979 984 ASSERT(MUTEX_HELD(&spa_namespace_lock));
980 985
981 986 spa_config_enter(spa, SCL_ALL, spa, RW_WRITER);
982 987
983 988 return (spa_last_synced_txg(spa) + 1);
984 989 }
985 990
986 991 /*
987 992 * Used in combination with spa_vdev_config_enter() to allow the syncing
988 993 * of multiple transactions without releasing the spa_namespace_lock.
989 994 */
990 995 void
991 996 spa_vdev_config_exit(spa_t *spa, vdev_t *vd, uint64_t txg, int error, char *tag)
992 997 {
993 998 ASSERT(MUTEX_HELD(&spa_namespace_lock));
994 999
995 1000 int config_changed = B_FALSE;
996 1001
997 1002 ASSERT(txg > spa_last_synced_txg(spa));
998 1003
999 1004 spa->spa_pending_vdev = NULL;
1000 1005
1001 1006 /*
1002 1007 * Reassess the DTLs.
1003 1008 */
1004 1009 vdev_dtl_reassess(spa->spa_root_vdev, 0, 0, B_FALSE);
1005 1010
1006 1011 if (error == 0 && !list_is_empty(&spa->spa_config_dirty_list)) {
1007 1012 config_changed = B_TRUE;
1008 1013 spa->spa_config_generation++;
1009 1014 }
1010 1015
1011 1016 /*
1012 1017 * Verify the metaslab classes.
1013 1018 */
1014 1019 ASSERT(metaslab_class_validate(spa_normal_class(spa)) == 0);
1015 1020 ASSERT(metaslab_class_validate(spa_log_class(spa)) == 0);
1016 1021
1017 1022 spa_config_exit(spa, SCL_ALL, spa);
1018 1023
1019 1024 /*
1020 1025 * Panic the system if the specified tag requires it. This
1021 1026 * is useful for ensuring that configurations are updated
1022 1027 * transactionally.
1023 1028 */
1024 1029 if (zio_injection_enabled)
1025 1030 zio_handle_panic_injection(spa, tag, 0);
1026 1031
1027 1032 /*
1028 1033 * Note: this txg_wait_synced() is important because it ensures
1029 1034 * that there won't be more than one config change per txg.
1030 1035 * This allows us to use the txg as the generation number.
1031 1036 */
1032 1037 if (error == 0)
1033 1038 txg_wait_synced(spa->spa_dsl_pool, txg);
1034 1039
1035 1040 if (vd != NULL) {
1036 1041 ASSERT(!vd->vdev_detached || vd->vdev_dtl_sm == NULL);
1037 1042 spa_config_enter(spa, SCL_ALL, spa, RW_WRITER);
1038 1043 vdev_free(vd);
1039 1044 spa_config_exit(spa, SCL_ALL, spa);
1040 1045 }
1041 1046
1042 1047 /*
1043 1048 * If the config changed, update the config cache.
1044 1049 */
1045 1050 if (config_changed)
1046 1051 spa_config_sync(spa, B_FALSE, B_TRUE);
1047 1052 }
1048 1053
1049 1054 /*
1050 1055 * Unlock the spa_t after adding or removing a vdev. Besides undoing the
1051 1056 * locking of spa_vdev_enter(), we also want make sure the transactions have
1052 1057 * synced to disk, and then update the global configuration cache with the new
1053 1058 * information.
1054 1059 */
1055 1060 int
1056 1061 spa_vdev_exit(spa_t *spa, vdev_t *vd, uint64_t txg, int error)
1057 1062 {
1058 1063 spa_vdev_config_exit(spa, vd, txg, error, FTAG);
1059 1064 mutex_exit(&spa_namespace_lock);
1060 1065 mutex_exit(&spa->spa_vdev_top_lock);
1061 1066
1062 1067 return (error);
1063 1068 }
1064 1069
1065 1070 /*
1066 1071 * Lock the given spa_t for the purpose of changing vdev state.
1067 1072 */
1068 1073 void
1069 1074 spa_vdev_state_enter(spa_t *spa, int oplocks)
1070 1075 {
1071 1076 int locks = SCL_STATE_ALL | oplocks;
1072 1077
1073 1078 /*
1074 1079 * Root pools may need to read of the underlying devfs filesystem
1075 1080 * when opening up a vdev. Unfortunately if we're holding the
1076 1081 * SCL_ZIO lock it will result in a deadlock when we try to issue
1077 1082 * the read from the root filesystem. Instead we "prefetch"
1078 1083 * the associated vnodes that we need prior to opening the
1079 1084 * underlying devices and cache them so that we can prevent
1080 1085 * any I/O when we are doing the actual open.
1081 1086 */
1082 1087 if (spa_is_root(spa)) {
1083 1088 int low = locks & ~(SCL_ZIO - 1);
1084 1089 int high = locks & ~low;
1085 1090
1086 1091 spa_config_enter(spa, high, spa, RW_WRITER);
1087 1092 vdev_hold(spa->spa_root_vdev);
1088 1093 spa_config_enter(spa, low, spa, RW_WRITER);
1089 1094 } else {
1090 1095 spa_config_enter(spa, locks, spa, RW_WRITER);
1091 1096 }
1092 1097 spa->spa_vdev_locks = locks;
1093 1098 }
1094 1099
1095 1100 int
1096 1101 spa_vdev_state_exit(spa_t *spa, vdev_t *vd, int error)
1097 1102 {
1098 1103 boolean_t config_changed = B_FALSE;
1099 1104
1100 1105 if (vd != NULL || error == 0)
1101 1106 vdev_dtl_reassess(vd ? vd->vdev_top : spa->spa_root_vdev,
1102 1107 0, 0, B_FALSE);
1103 1108
1104 1109 if (vd != NULL) {
1105 1110 vdev_state_dirty(vd->vdev_top);
1106 1111 config_changed = B_TRUE;
1107 1112 spa->spa_config_generation++;
1108 1113 }
1109 1114
1110 1115 if (spa_is_root(spa))
1111 1116 vdev_rele(spa->spa_root_vdev);
1112 1117
1113 1118 ASSERT3U(spa->spa_vdev_locks, >=, SCL_STATE_ALL);
1114 1119 spa_config_exit(spa, spa->spa_vdev_locks, spa);
1115 1120
1116 1121 /*
1117 1122 * If anything changed, wait for it to sync. This ensures that,
1118 1123 * from the system administrator's perspective, zpool(1M) commands
1119 1124 * are synchronous. This is important for things like zpool offline:
1120 1125 * when the command completes, you expect no further I/O from ZFS.
1121 1126 */
1122 1127 if (vd != NULL)
1123 1128 txg_wait_synced(spa->spa_dsl_pool, 0);
1124 1129
1125 1130 /*
1126 1131 * If the config changed, update the config cache.
1127 1132 */
1128 1133 if (config_changed) {
1129 1134 mutex_enter(&spa_namespace_lock);
1130 1135 spa_config_sync(spa, B_FALSE, B_TRUE);
1131 1136 mutex_exit(&spa_namespace_lock);
1132 1137 }
1133 1138
1134 1139 return (error);
1135 1140 }
1136 1141
1137 1142 /*
1138 1143 * ==========================================================================
1139 1144 * Miscellaneous functions
1140 1145 * ==========================================================================
1141 1146 */
1142 1147
1143 1148 void
1144 1149 spa_activate_mos_feature(spa_t *spa, const char *feature)
1145 1150 {
1146 1151 (void) nvlist_add_boolean(spa->spa_label_features, feature);
1147 1152 vdev_config_dirty(spa->spa_root_vdev);
1148 1153 }
1149 1154
1150 1155 void
1151 1156 spa_deactivate_mos_feature(spa_t *spa, const char *feature)
1152 1157 {
1153 1158 (void) nvlist_remove_all(spa->spa_label_features, feature);
1154 1159 vdev_config_dirty(spa->spa_root_vdev);
1155 1160 }
1156 1161
1157 1162 /*
1158 1163 * Rename a spa_t.
1159 1164 */
1160 1165 int
1161 1166 spa_rename(const char *name, const char *newname)
1162 1167 {
1163 1168 spa_t *spa;
1164 1169 int err;
1165 1170
1166 1171 /*
1167 1172 * Lookup the spa_t and grab the config lock for writing. We need to
1168 1173 * actually open the pool so that we can sync out the necessary labels.
1169 1174 * It's OK to call spa_open() with the namespace lock held because we
1170 1175 * allow recursive calls for other reasons.
1171 1176 */
1172 1177 mutex_enter(&spa_namespace_lock);
1173 1178 if ((err = spa_open(name, &spa, FTAG)) != 0) {
1174 1179 mutex_exit(&spa_namespace_lock);
1175 1180 return (err);
1176 1181 }
1177 1182
1178 1183 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1179 1184
1180 1185 avl_remove(&spa_namespace_avl, spa);
1181 1186 (void) strlcpy(spa->spa_name, newname, sizeof (spa->spa_name));
1182 1187 avl_add(&spa_namespace_avl, spa);
1183 1188
1184 1189 /*
1185 1190 * Sync all labels to disk with the new names by marking the root vdev
1186 1191 * dirty and waiting for it to sync. It will pick up the new pool name
1187 1192 * during the sync.
1188 1193 */
1189 1194 vdev_config_dirty(spa->spa_root_vdev);
1190 1195
1191 1196 spa_config_exit(spa, SCL_ALL, FTAG);
1192 1197
1193 1198 txg_wait_synced(spa->spa_dsl_pool, 0);
1194 1199
1195 1200 /*
1196 1201 * Sync the updated config cache.
1197 1202 */
1198 1203 spa_config_sync(spa, B_FALSE, B_TRUE);
1199 1204
1200 1205 spa_close(spa, FTAG);
1201 1206
1202 1207 mutex_exit(&spa_namespace_lock);
1203 1208
1204 1209 return (0);
1205 1210 }
1206 1211
1207 1212 /*
1208 1213 * Return the spa_t associated with given pool_guid, if it exists. If
1209 1214 * device_guid is non-zero, determine whether the pool exists *and* contains
1210 1215 * a device with the specified device_guid.
1211 1216 */
1212 1217 spa_t *
1213 1218 spa_by_guid(uint64_t pool_guid, uint64_t device_guid)
1214 1219 {
1215 1220 spa_t *spa;
1216 1221 avl_tree_t *t = &spa_namespace_avl;
1217 1222
1218 1223 ASSERT(MUTEX_HELD(&spa_namespace_lock));
1219 1224
1220 1225 for (spa = avl_first(t); spa != NULL; spa = AVL_NEXT(t, spa)) {
1221 1226 if (spa->spa_state == POOL_STATE_UNINITIALIZED)
1222 1227 continue;
1223 1228 if (spa->spa_root_vdev == NULL)
1224 1229 continue;
1225 1230 if (spa_guid(spa) == pool_guid) {
1226 1231 if (device_guid == 0)
1227 1232 break;
1228 1233
1229 1234 if (vdev_lookup_by_guid(spa->spa_root_vdev,
1230 1235 device_guid) != NULL)
1231 1236 break;
1232 1237
1233 1238 /*
1234 1239 * Check any devices we may be in the process of adding.
1235 1240 */
1236 1241 if (spa->spa_pending_vdev) {
1237 1242 if (vdev_lookup_by_guid(spa->spa_pending_vdev,
1238 1243 device_guid) != NULL)
1239 1244 break;
1240 1245 }
1241 1246 }
1242 1247 }
1243 1248
1244 1249 return (spa);
1245 1250 }
1246 1251
1247 1252 /*
1248 1253 * Determine whether a pool with the given pool_guid exists.
1249 1254 */
1250 1255 boolean_t
1251 1256 spa_guid_exists(uint64_t pool_guid, uint64_t device_guid)
1252 1257 {
1253 1258 return (spa_by_guid(pool_guid, device_guid) != NULL);
1254 1259 }
1255 1260
1256 1261 char *
1257 1262 spa_strdup(const char *s)
1258 1263 {
1259 1264 size_t len;
1260 1265 char *new;
1261 1266
1262 1267 len = strlen(s);
1263 1268 new = kmem_alloc(len + 1, KM_SLEEP);
1264 1269 bcopy(s, new, len);
1265 1270 new[len] = '\0';
1266 1271
1267 1272 return (new);
1268 1273 }
1269 1274
1270 1275 void
1271 1276 spa_strfree(char *s)
1272 1277 {
1273 1278 kmem_free(s, strlen(s) + 1);
1274 1279 }
1275 1280
1276 1281 uint64_t
1277 1282 spa_get_random(uint64_t range)
1278 1283 {
1279 1284 uint64_t r;
1280 1285
1281 1286 ASSERT(range != 0);
1282 1287
1283 1288 (void) random_get_pseudo_bytes((void *)&r, sizeof (uint64_t));
1284 1289
1285 1290 return (r % range);
1286 1291 }
1287 1292
1288 1293 uint64_t
1289 1294 spa_generate_guid(spa_t *spa)
1290 1295 {
1291 1296 uint64_t guid = spa_get_random(-1ULL);
1292 1297
1293 1298 if (spa != NULL) {
1294 1299 while (guid == 0 || spa_guid_exists(spa_guid(spa), guid))
1295 1300 guid = spa_get_random(-1ULL);
1296 1301 } else {
1297 1302 while (guid == 0 || spa_guid_exists(guid, 0))
1298 1303 guid = spa_get_random(-1ULL);
1299 1304 }
1300 1305
1301 1306 return (guid);
1302 1307 }
1303 1308
1304 1309 void
1305 1310 sprintf_blkptr(char *buf, const blkptr_t *bp)
1306 1311 {
1307 1312 char type[256];
1308 1313 char *checksum = NULL;
1309 1314 char *compress = NULL;
1310 1315
1311 1316 if (bp != NULL) {
1312 1317 if (BP_GET_TYPE(bp) & DMU_OT_NEWTYPE) {
1313 1318 dmu_object_byteswap_t bswap =
1314 1319 DMU_OT_BYTESWAP(BP_GET_TYPE(bp));
1315 1320 (void) snprintf(type, sizeof (type), "bswap %s %s",
1316 1321 DMU_OT_IS_METADATA(BP_GET_TYPE(bp)) ?
1317 1322 "metadata" : "data",
1318 1323 dmu_ot_byteswap[bswap].ob_name);
1319 1324 } else {
1320 1325 (void) strlcpy(type, dmu_ot[BP_GET_TYPE(bp)].ot_name,
1321 1326 sizeof (type));
1322 1327 }
1323 1328 checksum = zio_checksum_table[BP_GET_CHECKSUM(bp)].ci_name;
1324 1329 compress = zio_compress_table[BP_GET_COMPRESS(bp)].ci_name;
1325 1330 }
1326 1331
1327 1332 SPRINTF_BLKPTR(snprintf, ' ', buf, bp, type, checksum, compress);
1328 1333 }
1329 1334
1330 1335 void
1331 1336 spa_freeze(spa_t *spa)
1332 1337 {
1333 1338 uint64_t freeze_txg = 0;
1334 1339
1335 1340 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1336 1341 if (spa->spa_freeze_txg == UINT64_MAX) {
1337 1342 freeze_txg = spa_last_synced_txg(spa) + TXG_SIZE;
1338 1343 spa->spa_freeze_txg = freeze_txg;
1339 1344 }
1340 1345 spa_config_exit(spa, SCL_ALL, FTAG);
1341 1346 if (freeze_txg != 0)
1342 1347 txg_wait_synced(spa_get_dsl(spa), freeze_txg);
1343 1348 }
1344 1349
1345 1350 void
1346 1351 zfs_panic_recover(const char *fmt, ...)
1347 1352 {
1348 1353 va_list adx;
1349 1354
1350 1355 va_start(adx, fmt);
1351 1356 vcmn_err(zfs_recover ? CE_WARN : CE_PANIC, fmt, adx);
1352 1357 va_end(adx);
1353 1358 }
1354 1359
1355 1360 /*
1356 1361 * This is a stripped-down version of strtoull, suitable only for converting
1357 1362 * lowercase hexadecimal numbers that don't overflow.
1358 1363 */
1359 1364 uint64_t
1360 1365 strtonum(const char *str, char **nptr)
1361 1366 {
1362 1367 uint64_t val = 0;
1363 1368 char c;
1364 1369 int digit;
1365 1370
1366 1371 while ((c = *str) != '\0') {
1367 1372 if (c >= '0' && c <= '9')
1368 1373 digit = c - '0';
1369 1374 else if (c >= 'a' && c <= 'f')
1370 1375 digit = 10 + c - 'a';
1371 1376 else
1372 1377 break;
1373 1378
1374 1379 val *= 16;
1375 1380 val += digit;
1376 1381
1377 1382 str++;
1378 1383 }
1379 1384
1380 1385 if (nptr)
1381 1386 *nptr = (char *)str;
1382 1387
1383 1388 return (val);
1384 1389 }
1385 1390
1386 1391 /*
1387 1392 * ==========================================================================
1388 1393 * Accessor functions
1389 1394 * ==========================================================================
1390 1395 */
1391 1396
1392 1397 boolean_t
1393 1398 spa_shutting_down(spa_t *spa)
1394 1399 {
1395 1400 return (spa->spa_async_suspended);
1396 1401 }
1397 1402
1398 1403 dsl_pool_t *
1399 1404 spa_get_dsl(spa_t *spa)
1400 1405 {
1401 1406 return (spa->spa_dsl_pool);
1402 1407 }
1403 1408
1404 1409 boolean_t
1405 1410 spa_is_initializing(spa_t *spa)
1406 1411 {
1407 1412 return (spa->spa_is_initializing);
1408 1413 }
1409 1414
1410 1415 blkptr_t *
1411 1416 spa_get_rootblkptr(spa_t *spa)
1412 1417 {
1413 1418 return (&spa->spa_ubsync.ub_rootbp);
1414 1419 }
1415 1420
1416 1421 void
1417 1422 spa_set_rootblkptr(spa_t *spa, const blkptr_t *bp)
1418 1423 {
1419 1424 spa->spa_uberblock.ub_rootbp = *bp;
1420 1425 }
1421 1426
1422 1427 void
1423 1428 spa_altroot(spa_t *spa, char *buf, size_t buflen)
1424 1429 {
1425 1430 if (spa->spa_root == NULL)
1426 1431 buf[0] = '\0';
1427 1432 else
1428 1433 (void) strncpy(buf, spa->spa_root, buflen);
1429 1434 }
1430 1435
1431 1436 int
1432 1437 spa_sync_pass(spa_t *spa)
1433 1438 {
1434 1439 return (spa->spa_sync_pass);
1435 1440 }
1436 1441
1437 1442 char *
1438 1443 spa_name(spa_t *spa)
1439 1444 {
1440 1445 return (spa->spa_name);
1441 1446 }
1442 1447
1443 1448 uint64_t
1444 1449 spa_guid(spa_t *spa)
1445 1450 {
1446 1451 dsl_pool_t *dp = spa_get_dsl(spa);
1447 1452 uint64_t guid;
1448 1453
1449 1454 /*
1450 1455 * If we fail to parse the config during spa_load(), we can go through
1451 1456 * the error path (which posts an ereport) and end up here with no root
1452 1457 * vdev. We stash the original pool guid in 'spa_config_guid' to handle
1453 1458 * this case.
1454 1459 */
1455 1460 if (spa->spa_root_vdev == NULL)
1456 1461 return (spa->spa_config_guid);
1457 1462
1458 1463 guid = spa->spa_last_synced_guid != 0 ?
1459 1464 spa->spa_last_synced_guid : spa->spa_root_vdev->vdev_guid;
1460 1465
1461 1466 /*
1462 1467 * Return the most recently synced out guid unless we're
1463 1468 * in syncing context.
1464 1469 */
1465 1470 if (dp && dsl_pool_sync_context(dp))
1466 1471 return (spa->spa_root_vdev->vdev_guid);
1467 1472 else
1468 1473 return (guid);
1469 1474 }
1470 1475
1471 1476 uint64_t
1472 1477 spa_load_guid(spa_t *spa)
1473 1478 {
1474 1479 /*
1475 1480 * This is a GUID that exists solely as a reference for the
1476 1481 * purposes of the arc. It is generated at load time, and
1477 1482 * is never written to persistent storage.
1478 1483 */
1479 1484 return (spa->spa_load_guid);
1480 1485 }
1481 1486
1482 1487 uint64_t
1483 1488 spa_last_synced_txg(spa_t *spa)
1484 1489 {
1485 1490 return (spa->spa_ubsync.ub_txg);
1486 1491 }
1487 1492
1488 1493 uint64_t
1489 1494 spa_first_txg(spa_t *spa)
1490 1495 {
1491 1496 return (spa->spa_first_txg);
1492 1497 }
1493 1498
1494 1499 uint64_t
1495 1500 spa_syncing_txg(spa_t *spa)
1496 1501 {
1497 1502 return (spa->spa_syncing_txg);
1498 1503 }
1499 1504
1500 1505 pool_state_t
1501 1506 spa_state(spa_t *spa)
1502 1507 {
1503 1508 return (spa->spa_state);
1504 1509 }
1505 1510
1506 1511 spa_load_state_t
1507 1512 spa_load_state(spa_t *spa)
1508 1513 {
1509 1514 return (spa->spa_load_state);
1510 1515 }
1511 1516
1512 1517 uint64_t
1513 1518 spa_freeze_txg(spa_t *spa)
1514 1519 {
1515 1520 return (spa->spa_freeze_txg);
1516 1521 }
1517 1522
1518 1523 /* ARGSUSED */
1519 1524 uint64_t
1520 1525 spa_get_asize(spa_t *spa, uint64_t lsize)
1521 1526 {
1522 1527 return (lsize * spa_asize_inflation);
1523 1528 }
1524 1529
1525 1530 uint64_t
1526 1531 spa_get_dspace(spa_t *spa)
1527 1532 {
1528 1533 return (spa->spa_dspace);
1529 1534 }
1530 1535
1531 1536 void
1532 1537 spa_update_dspace(spa_t *spa)
1533 1538 {
1534 1539 spa->spa_dspace = metaslab_class_get_dspace(spa_normal_class(spa)) +
1535 1540 ddt_get_dedup_dspace(spa);
1536 1541 }
1537 1542
1538 1543 /*
1539 1544 * Return the failure mode that has been set to this pool. The default
1540 1545 * behavior will be to block all I/Os when a complete failure occurs.
1541 1546 */
1542 1547 uint8_t
1543 1548 spa_get_failmode(spa_t *spa)
1544 1549 {
1545 1550 return (spa->spa_failmode);
1546 1551 }
1547 1552
1548 1553 boolean_t
1549 1554 spa_suspended(spa_t *spa)
1550 1555 {
1551 1556 return (spa->spa_suspended);
1552 1557 }
1553 1558
1554 1559 uint64_t
1555 1560 spa_version(spa_t *spa)
1556 1561 {
1557 1562 return (spa->spa_ubsync.ub_version);
1558 1563 }
1559 1564
1560 1565 boolean_t
1561 1566 spa_deflate(spa_t *spa)
1562 1567 {
1563 1568 return (spa->spa_deflate);
1564 1569 }
1565 1570
1566 1571 metaslab_class_t *
1567 1572 spa_normal_class(spa_t *spa)
1568 1573 {
1569 1574 return (spa->spa_normal_class);
1570 1575 }
1571 1576
1572 1577 metaslab_class_t *
1573 1578 spa_log_class(spa_t *spa)
1574 1579 {
1575 1580 return (spa->spa_log_class);
1576 1581 }
1577 1582
1578 1583 int
1579 1584 spa_max_replication(spa_t *spa)
1580 1585 {
1581 1586 /*
1582 1587 * As of SPA_VERSION == SPA_VERSION_DITTO_BLOCKS, we are able to
1583 1588 * handle BPs with more than one DVA allocated. Set our max
1584 1589 * replication level accordingly.
1585 1590 */
1586 1591 if (spa_version(spa) < SPA_VERSION_DITTO_BLOCKS)
1587 1592 return (1);
1588 1593 return (MIN(SPA_DVAS_PER_BP, spa_max_replication_override));
1589 1594 }
1590 1595
1591 1596 int
1592 1597 spa_prev_software_version(spa_t *spa)
1593 1598 {
1594 1599 return (spa->spa_prev_software_version);
1595 1600 }
1596 1601
1597 1602 uint64_t
1598 1603 spa_deadman_synctime(spa_t *spa)
1599 1604 {
1600 1605 return (spa->spa_deadman_synctime);
1601 1606 }
1602 1607
1603 1608 uint64_t
1604 1609 dva_get_dsize_sync(spa_t *spa, const dva_t *dva)
1605 1610 {
1606 1611 uint64_t asize = DVA_GET_ASIZE(dva);
1607 1612 uint64_t dsize = asize;
1608 1613
1609 1614 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
1610 1615
1611 1616 if (asize != 0 && spa->spa_deflate) {
1612 1617 vdev_t *vd = vdev_lookup_top(spa, DVA_GET_VDEV(dva));
1613 1618 dsize = (asize >> SPA_MINBLOCKSHIFT) * vd->vdev_deflate_ratio;
1614 1619 }
1615 1620
1616 1621 return (dsize);
1617 1622 }
1618 1623
1619 1624 uint64_t
1620 1625 bp_get_dsize_sync(spa_t *spa, const blkptr_t *bp)
1621 1626 {
1622 1627 uint64_t dsize = 0;
1623 1628
1624 1629 for (int d = 0; d < SPA_DVAS_PER_BP; d++)
1625 1630 dsize += dva_get_dsize_sync(spa, &bp->blk_dva[d]);
1626 1631
1627 1632 return (dsize);
1628 1633 }
1629 1634
1630 1635 uint64_t
1631 1636 bp_get_dsize(spa_t *spa, const blkptr_t *bp)
1632 1637 {
1633 1638 uint64_t dsize = 0;
1634 1639
1635 1640 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
1636 1641
1637 1642 for (int d = 0; d < SPA_DVAS_PER_BP; d++)
1638 1643 dsize += dva_get_dsize_sync(spa, &bp->blk_dva[d]);
1639 1644
1640 1645 spa_config_exit(spa, SCL_VDEV, FTAG);
1641 1646
1642 1647 return (dsize);
1643 1648 }
1644 1649
1645 1650 /*
1646 1651 * ==========================================================================
1647 1652 * Initialization and Termination
1648 1653 * ==========================================================================
1649 1654 */
1650 1655
1651 1656 static int
1652 1657 spa_name_compare(const void *a1, const void *a2)
1653 1658 {
1654 1659 const spa_t *s1 = a1;
1655 1660 const spa_t *s2 = a2;
1656 1661 int s;
1657 1662
1658 1663 s = strcmp(s1->spa_name, s2->spa_name);
1659 1664 if (s > 0)
1660 1665 return (1);
1661 1666 if (s < 0)
1662 1667 return (-1);
1663 1668 return (0);
1664 1669 }
1665 1670
1666 1671 int
1667 1672 spa_busy(void)
1668 1673 {
1669 1674 return (spa_active_count);
1670 1675 }
1671 1676
1672 1677 void
1673 1678 spa_boot_init()
1674 1679 {
1675 1680 spa_config_load();
1676 1681 }
1677 1682
1678 1683 void
1679 1684 spa_init(int mode)
1680 1685 {
1681 1686 mutex_init(&spa_namespace_lock, NULL, MUTEX_DEFAULT, NULL);
1682 1687 mutex_init(&spa_spare_lock, NULL, MUTEX_DEFAULT, NULL);
1683 1688 mutex_init(&spa_l2cache_lock, NULL, MUTEX_DEFAULT, NULL);
1684 1689 cv_init(&spa_namespace_cv, NULL, CV_DEFAULT, NULL);
1685 1690
1686 1691 avl_create(&spa_namespace_avl, spa_name_compare, sizeof (spa_t),
1687 1692 offsetof(spa_t, spa_avl));
1688 1693
1689 1694 avl_create(&spa_spare_avl, spa_spare_compare, sizeof (spa_aux_t),
1690 1695 offsetof(spa_aux_t, aux_avl));
1691 1696
1692 1697 avl_create(&spa_l2cache_avl, spa_l2cache_compare, sizeof (spa_aux_t),
1693 1698 offsetof(spa_aux_t, aux_avl));
1694 1699
1695 1700 spa_mode_global = mode;
1696 1701
1697 1702 #ifdef _KERNEL
1698 1703 spa_arch_init();
1699 1704 #else
1700 1705 if (spa_mode_global != FREAD && dprintf_find_string("watch")) {
1701 1706 arc_procfd = open("/proc/self/ctl", O_WRONLY);
1702 1707 if (arc_procfd == -1) {
1703 1708 perror("could not enable watchpoints: "
1704 1709 "opening /proc/self/ctl failed: ");
1705 1710 } else {
1706 1711 arc_watch = B_TRUE;
1707 1712 }
1708 1713 }
1709 1714 #endif
1710 1715
1711 1716 refcount_init();
1712 1717 unique_init();
1713 1718 range_tree_init();
1714 1719 zio_init();
1715 1720 dmu_init();
1716 1721 zil_init();
1717 1722 vdev_cache_stat_init();
1718 1723 zfs_prop_init();
1719 1724 zpool_prop_init();
1720 1725 zpool_feature_init();
1721 1726 spa_config_load();
1722 1727 l2arc_start();
1723 1728 }
1724 1729
1725 1730 void
1726 1731 spa_fini(void)
1727 1732 {
1728 1733 l2arc_stop();
1729 1734
1730 1735 spa_evict_all();
1731 1736
1732 1737 vdev_cache_stat_fini();
1733 1738 zil_fini();
1734 1739 dmu_fini();
1735 1740 zio_fini();
1736 1741 range_tree_fini();
1737 1742 unique_fini();
1738 1743 refcount_fini();
1739 1744
1740 1745 avl_destroy(&spa_namespace_avl);
1741 1746 avl_destroy(&spa_spare_avl);
1742 1747 avl_destroy(&spa_l2cache_avl);
1743 1748
1744 1749 cv_destroy(&spa_namespace_cv);
1745 1750 mutex_destroy(&spa_namespace_lock);
1746 1751 mutex_destroy(&spa_spare_lock);
1747 1752 mutex_destroy(&spa_l2cache_lock);
1748 1753 }
1749 1754
1750 1755 /*
1751 1756 * Return whether this pool has slogs. No locking needed.
1752 1757 * It's not a problem if the wrong answer is returned as it's only for
1753 1758 * performance and not correctness
1754 1759 */
1755 1760 boolean_t
1756 1761 spa_has_slogs(spa_t *spa)
1757 1762 {
1758 1763 return (spa->spa_log_class->mc_rotor != NULL);
1759 1764 }
1760 1765
1761 1766 spa_log_state_t
1762 1767 spa_get_log_state(spa_t *spa)
1763 1768 {
1764 1769 return (spa->spa_log_state);
1765 1770 }
1766 1771
1767 1772 void
1768 1773 spa_set_log_state(spa_t *spa, spa_log_state_t state)
1769 1774 {
1770 1775 spa->spa_log_state = state;
1771 1776 }
1772 1777
1773 1778 boolean_t
1774 1779 spa_is_root(spa_t *spa)
|
↓ open down ↓ |
1110 lines elided |
↑ open up ↑ |
1775 1780 {
1776 1781 return (spa->spa_is_root);
1777 1782 }
1778 1783
1779 1784 boolean_t
1780 1785 spa_writeable(spa_t *spa)
1781 1786 {
1782 1787 return (!!(spa->spa_mode & FWRITE));
1783 1788 }
1784 1789
1790 +static int
1791 +activate_salted_cksum_check(zfeature_info_t *feature, dmu_tx_t *tx)
1792 +{
1793 + spa_t *spa = dmu_tx_pool(tx)->dp_spa;
1794 +
1795 + if (!spa_feature_is_active(spa, feature))
1796 + return (0);
1797 + else
1798 + return (SET_ERROR(EBUSY));
1799 +}
1800 +
1801 +static void
1802 +activate_salted_cksum_sync(zfeature_info_t *feature, dmu_tx_t *tx)
1803 +{
1804 + spa_t *spa = dmu_tx_pool(tx)->dp_spa;
1805 +
1806 + spa_feature_incr(spa, feature, tx);
1807 + /*
1808 + * This is the first salted checksum that's been activated, so
1809 + * create the persistent checksum salt object now.
1810 + */
1811 + if (spa->spa_cksum_salt_obj == 0) {
1812 + spa->spa_cksum_salt_obj = zap_create_link(spa->spa_meta_objset,
1813 + DMU_OTN_ZAP_METADATA, DMU_POOL_DIRECTORY_OBJECT,
1814 + DMU_POOL_CHECKSUM_SALT, tx);
1815 + VERIFY3U(zap_add(spa->spa_meta_objset,
1816 + spa->spa_cksum_salt_obj, DMU_POOL_CHECKSUM_SALT, 1,
1817 + sizeof (spa->spa_cksum_salt.zcs_bytes),
1818 + spa->spa_cksum_salt.zcs_bytes, tx), ==, 0);
1819 + }
1820 +}
1821 +
1822 +/*
1823 + * Activates a feature associated with a salted checksum. You must call this
1824 + * function instead of calling spa_feature_incr() directly, because we may
1825 + * also need to sync the MOS object holding the checksum salt.
1826 + * Arguments:
1827 + * spa Pool on which to activate the salted checksum feature.
1828 + * feature Salted checksum algorithm feature to activate (see
1829 + * spa_feature_table).
1830 + */
1831 +int
1832 +spa_activate_salted_cksum(spa_t *spa, struct zfeature_info *feature)
1833 +{
1834 + int err;
1835 +
1836 + /* EBUSY here indicates that the feature is already active */
1837 + err = dsl_sync_task(spa_name(spa),
1838 + (dsl_checkfunc_t *)activate_salted_cksum_check,
1839 + (dsl_syncfunc_t *)activate_salted_cksum_sync, feature, 2);
1840 +
1841 + if (err != 0 && err != EBUSY)
1842 + return (err);
1843 + else
1844 + return (0);
1845 +}
1846 +
1785 1847 int
1786 1848 spa_mode(spa_t *spa)
1787 1849 {
1788 1850 return (spa->spa_mode);
1789 1851 }
1790 1852
1791 1853 uint64_t
1792 1854 spa_bootfs(spa_t *spa)
1793 1855 {
1794 1856 return (spa->spa_bootfs);
1795 1857 }
1796 1858
1797 1859 uint64_t
1798 1860 spa_delegation(spa_t *spa)
1799 1861 {
1800 1862 return (spa->spa_delegation);
1801 1863 }
1802 1864
1803 1865 objset_t *
1804 1866 spa_meta_objset(spa_t *spa)
1805 1867 {
1806 1868 return (spa->spa_meta_objset);
1807 1869 }
1808 1870
1809 1871 enum zio_checksum
1810 1872 spa_dedup_checksum(spa_t *spa)
1811 1873 {
1812 1874 return (spa->spa_dedup_checksum);
1813 1875 }
1814 1876
1815 1877 /*
1816 1878 * Reset pool scan stat per scan pass (or reboot).
1817 1879 */
1818 1880 void
1819 1881 spa_scan_stat_init(spa_t *spa)
1820 1882 {
1821 1883 /* data not stored on disk */
1822 1884 spa->spa_scan_pass_start = gethrestime_sec();
1823 1885 spa->spa_scan_pass_exam = 0;
1824 1886 vdev_scan_stat_init(spa->spa_root_vdev);
1825 1887 }
1826 1888
1827 1889 /*
1828 1890 * Get scan stats for zpool status reports
1829 1891 */
1830 1892 int
1831 1893 spa_scan_get_stats(spa_t *spa, pool_scan_stat_t *ps)
1832 1894 {
1833 1895 dsl_scan_t *scn = spa->spa_dsl_pool ? spa->spa_dsl_pool->dp_scan : NULL;
1834 1896
1835 1897 if (scn == NULL || scn->scn_phys.scn_func == POOL_SCAN_NONE)
1836 1898 return (SET_ERROR(ENOENT));
1837 1899 bzero(ps, sizeof (pool_scan_stat_t));
1838 1900
1839 1901 /* data stored on disk */
1840 1902 ps->pss_func = scn->scn_phys.scn_func;
1841 1903 ps->pss_start_time = scn->scn_phys.scn_start_time;
1842 1904 ps->pss_end_time = scn->scn_phys.scn_end_time;
1843 1905 ps->pss_to_examine = scn->scn_phys.scn_to_examine;
1844 1906 ps->pss_examined = scn->scn_phys.scn_examined;
1845 1907 ps->pss_to_process = scn->scn_phys.scn_to_process;
1846 1908 ps->pss_processed = scn->scn_phys.scn_processed;
1847 1909 ps->pss_errors = scn->scn_phys.scn_errors;
1848 1910 ps->pss_state = scn->scn_phys.scn_state;
1849 1911
1850 1912 /* data not stored on disk */
1851 1913 ps->pss_pass_start = spa->spa_scan_pass_start;
1852 1914 ps->pss_pass_exam = spa->spa_scan_pass_exam;
1853 1915
1854 1916 return (0);
1855 1917 }
1856 1918
1857 1919 boolean_t
1858 1920 spa_debug_enabled(spa_t *spa)
1859 1921 {
1860 1922 return (spa->spa_debug);
1861 1923 }
|
↓ open down ↓ |
67 lines elided |
↑ open up ↑ |
XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX