Print this page
4045 zfs write throttle & i/o scheduler performance work
Reviewed by: George Wilson <george.wilson@delphix.com>
Reviewed by: Adam Leventhal <ahl@delphix.com>
Reviewed by: Christopher Siden <christopher.siden@delphix.com>
| 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.
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 25 */
26 26
27 27 #include <sys/zfs_context.h>
28 28 #include <sys/spa_impl.h>
29 29 #include <sys/spa_boot.h>
30 30 #include <sys/zio.h>
31 31 #include <sys/zio_checksum.h>
32 32 #include <sys/zio_compress.h>
33 33 #include <sys/dmu.h>
34 34 #include <sys/dmu_tx.h>
35 35 #include <sys/zap.h>
36 36 #include <sys/zil.h>
37 37 #include <sys/vdev_impl.h>
38 38 #include <sys/metaslab.h>
39 39 #include <sys/uberblock_impl.h>
40 40 #include <sys/txg.h>
41 41 #include <sys/avl.h>
42 42 #include <sys/unique.h>
43 43 #include <sys/dsl_pool.h>
44 44 #include <sys/dsl_dir.h>
45 45 #include <sys/dsl_prop.h>
46 46 #include <sys/dsl_scan.h>
47 47 #include <sys/fs/zfs.h>
48 48 #include <sys/metaslab_impl.h>
49 49 #include <sys/arc.h>
50 50 #include <sys/ddt.h>
51 51 #include "zfs_prop.h"
52 52 #include "zfeature_common.h"
53 53
54 54 /*
55 55 * SPA locking
56 56 *
57 57 * There are four basic locks for managing spa_t structures:
58 58 *
59 59 * spa_namespace_lock (global mutex)
60 60 *
61 61 * This lock must be acquired to do any of the following:
62 62 *
63 63 * - Lookup a spa_t by name
64 64 * - Add or remove a spa_t from the namespace
65 65 * - Increase spa_refcount from non-zero
66 66 * - Check if spa_refcount is zero
67 67 * - Rename a spa_t
68 68 * - add/remove/attach/detach devices
69 69 * - Held for the duration of create/destroy/import/export
70 70 *
71 71 * It does not need to handle recursion. A create or destroy may
72 72 * reference objects (files or zvols) in other pools, but by
73 73 * definition they must have an existing reference, and will never need
74 74 * to lookup a spa_t by name.
75 75 *
76 76 * spa_refcount (per-spa refcount_t protected by mutex)
77 77 *
78 78 * This reference count keep track of any active users of the spa_t. The
79 79 * spa_t cannot be destroyed or freed while this is non-zero. Internally,
80 80 * the refcount is never really 'zero' - opening a pool implicitly keeps
81 81 * some references in the DMU. Internally we check against spa_minref, but
82 82 * present the image of a zero/non-zero value to consumers.
83 83 *
84 84 * spa_config_lock[] (per-spa array of rwlocks)
85 85 *
86 86 * This protects the spa_t from config changes, and must be held in
87 87 * the following circumstances:
88 88 *
89 89 * - RW_READER to perform I/O to the spa
90 90 * - RW_WRITER to change the vdev config
91 91 *
92 92 * The locking order is fairly straightforward:
93 93 *
94 94 * spa_namespace_lock -> spa_refcount
95 95 *
96 96 * The namespace lock must be acquired to increase the refcount from 0
97 97 * or to check if it is zero.
98 98 *
99 99 * spa_refcount -> spa_config_lock[]
100 100 *
101 101 * There must be at least one valid reference on the spa_t to acquire
102 102 * the config lock.
103 103 *
104 104 * spa_namespace_lock -> spa_config_lock[]
105 105 *
106 106 * The namespace lock must always be taken before the config lock.
107 107 *
108 108 *
109 109 * The spa_namespace_lock can be acquired directly and is globally visible.
110 110 *
111 111 * The namespace is manipulated using the following functions, all of which
112 112 * require the spa_namespace_lock to be held.
113 113 *
114 114 * spa_lookup() Lookup a spa_t by name.
115 115 *
116 116 * spa_add() Create a new spa_t in the namespace.
117 117 *
118 118 * spa_remove() Remove a spa_t from the namespace. This also
119 119 * frees up any memory associated with the spa_t.
120 120 *
121 121 * spa_next() Returns the next spa_t in the system, or the
122 122 * first if NULL is passed.
123 123 *
124 124 * spa_evict_all() Shutdown and remove all spa_t structures in
125 125 * the system.
126 126 *
127 127 * spa_guid_exists() Determine whether a pool/device guid exists.
128 128 *
129 129 * The spa_refcount is manipulated using the following functions:
130 130 *
131 131 * spa_open_ref() Adds a reference to the given spa_t. Must be
132 132 * called with spa_namespace_lock held if the
133 133 * refcount is currently zero.
134 134 *
135 135 * spa_close() Remove a reference from the spa_t. This will
136 136 * not free the spa_t or remove it from the
137 137 * namespace. No locking is required.
138 138 *
139 139 * spa_refcount_zero() Returns true if the refcount is currently
140 140 * zero. Must be called with spa_namespace_lock
141 141 * held.
142 142 *
143 143 * The spa_config_lock[] is an array of rwlocks, ordered as follows:
144 144 * SCL_CONFIG > SCL_STATE > SCL_ALLOC > SCL_ZIO > SCL_FREE > SCL_VDEV.
145 145 * spa_config_lock[] is manipulated with spa_config_{enter,exit,held}().
146 146 *
147 147 * To read the configuration, it suffices to hold one of these locks as reader.
148 148 * To modify the configuration, you must hold all locks as writer. To modify
149 149 * vdev state without altering the vdev tree's topology (e.g. online/offline),
150 150 * you must hold SCL_STATE and SCL_ZIO as writer.
151 151 *
152 152 * We use these distinct config locks to avoid recursive lock entry.
153 153 * For example, spa_sync() (which holds SCL_CONFIG as reader) induces
154 154 * block allocations (SCL_ALLOC), which may require reading space maps
155 155 * from disk (dmu_read() -> zio_read() -> SCL_ZIO).
156 156 *
157 157 * The spa config locks cannot be normal rwlocks because we need the
158 158 * ability to hand off ownership. For example, SCL_ZIO is acquired
159 159 * by the issuing thread and later released by an interrupt thread.
160 160 * They do, however, obey the usual write-wanted semantics to prevent
161 161 * writer (i.e. system administrator) starvation.
162 162 *
163 163 * The lock acquisition rules are as follows:
164 164 *
165 165 * SCL_CONFIG
166 166 * Protects changes to the vdev tree topology, such as vdev
167 167 * add/remove/attach/detach. Protects the dirty config list
168 168 * (spa_config_dirty_list) and the set of spares and l2arc devices.
169 169 *
170 170 * SCL_STATE
171 171 * Protects changes to pool state and vdev state, such as vdev
172 172 * online/offline/fault/degrade/clear. Protects the dirty state list
173 173 * (spa_state_dirty_list) and global pool state (spa_state).
174 174 *
175 175 * SCL_ALLOC
176 176 * Protects changes to metaslab groups and classes.
177 177 * Held as reader by metaslab_alloc() and metaslab_claim().
178 178 *
179 179 * SCL_ZIO
180 180 * Held by bp-level zios (those which have no io_vd upon entry)
181 181 * to prevent changes to the vdev tree. The bp-level zio implicitly
182 182 * protects all of its vdev child zios, which do not hold SCL_ZIO.
183 183 *
184 184 * SCL_FREE
185 185 * Protects changes to metaslab groups and classes.
186 186 * Held as reader by metaslab_free(). SCL_FREE is distinct from
187 187 * SCL_ALLOC, and lower than SCL_ZIO, so that we can safely free
188 188 * blocks in zio_done() while another i/o that holds either
189 189 * SCL_ALLOC or SCL_ZIO is waiting for this i/o to complete.
190 190 *
191 191 * SCL_VDEV
192 192 * Held as reader to prevent changes to the vdev tree during trivial
193 193 * inquiries such as bp_get_dsize(). SCL_VDEV is distinct from the
194 194 * other locks, and lower than all of them, to ensure that it's safe
195 195 * to acquire regardless of caller context.
196 196 *
197 197 * In addition, the following rules apply:
198 198 *
199 199 * (a) spa_props_lock protects pool properties, spa_config and spa_config_list.
200 200 * The lock ordering is SCL_CONFIG > spa_props_lock.
201 201 *
202 202 * (b) I/O operations on leaf vdevs. For any zio operation that takes
203 203 * an explicit vdev_t argument -- such as zio_ioctl(), zio_read_phys(),
204 204 * or zio_write_phys() -- the caller must ensure that the config cannot
205 205 * cannot change in the interim, and that the vdev cannot be reopened.
206 206 * SCL_STATE as reader suffices for both.
207 207 *
208 208 * The vdev configuration is protected by spa_vdev_enter() / spa_vdev_exit().
209 209 *
210 210 * spa_vdev_enter() Acquire the namespace lock and the config lock
211 211 * for writing.
212 212 *
213 213 * spa_vdev_exit() Release the config lock, wait for all I/O
214 214 * to complete, sync the updated configs to the
215 215 * cache, and release the namespace lock.
216 216 *
217 217 * vdev state is protected by spa_vdev_state_enter() / spa_vdev_state_exit().
218 218 * Like spa_vdev_enter/exit, these are convenience wrappers -- the actual
219 219 * locking is, always, based on spa_namespace_lock and spa_config_lock[].
220 220 *
221 221 * spa_rename() is also implemented within this file since it requires
222 222 * manipulation of the namespace.
223 223 */
224 224
225 225 static avl_tree_t spa_namespace_avl;
226 226 kmutex_t spa_namespace_lock;
227 227 static kcondvar_t spa_namespace_cv;
228 228 static int spa_active_count;
229 229 int spa_max_replication_override = SPA_DVAS_PER_BP;
230 230
231 231 static kmutex_t spa_spare_lock;
232 232 static avl_tree_t spa_spare_avl;
233 233 static kmutex_t spa_l2cache_lock;
234 234 static avl_tree_t spa_l2cache_avl;
235 235
236 236 kmem_cache_t *spa_buffer_pool;
237 237 int spa_mode_global;
238 238
239 239 #ifdef ZFS_DEBUG
240 240 /* Everything except dprintf and spa is on by default in debug builds */
241 241 int zfs_flags = ~(ZFS_DEBUG_DPRINTF | ZFS_DEBUG_SPA);
242 242 #else
|
↓ open down ↓ |
242 lines elided |
↑ open up ↑ |
243 243 int zfs_flags = 0;
244 244 #endif
245 245
246 246 /*
247 247 * zfs_recover can be set to nonzero to attempt to recover from
248 248 * otherwise-fatal errors, typically caused by on-disk corruption. When
249 249 * set, calls to zfs_panic_recover() will turn into warning messages.
250 250 */
251 251 int zfs_recover = 0;
252 252
253 -extern int zfs_txg_synctime_ms;
253 +/*
254 + * Expiration time in milliseconds. This value has two meanings. First it is
255 + * used to determine when the spa_deadman() logic should fire. By default the
256 + * spa_deadman() will fire if spa_sync() has not completed in 1000 seconds.
257 + * Secondly, the value determines if an I/O is considered "hung". Any I/O that
258 + * has not completed in zfs_deadman_synctime_ms is considered "hung" resulting
259 + * in a system panic.
260 + */
261 +uint64_t zfs_deadman_synctime_ms = 1000000ULL;
254 262
255 263 /*
256 - * Expiration time in units of zfs_txg_synctime_ms. This value has two
257 - * meanings. First it is used to determine when the spa_deadman logic
258 - * should fire. By default the spa_deadman will fire if spa_sync has
259 - * not completed in 1000 * zfs_txg_synctime_ms (i.e. 1000 seconds).
260 - * Secondly, the value determines if an I/O is considered "hung".
261 - * Any I/O that has not completed in zfs_deadman_synctime is considered
262 - * "hung" resulting in a system panic.
264 + * Check time in milliseconds. This defines the frequency at which we check
265 + * for hung I/O.
263 266 */
264 -uint64_t zfs_deadman_synctime = 1000ULL;
267 +uint64_t zfs_deadman_checktime_ms = 5000ULL;
265 268
266 269 /*
267 270 * Override the zfs deadman behavior via /etc/system. By default the
268 271 * deadman is enabled except on VMware and sparc deployments.
269 272 */
270 273 int zfs_deadman_enabled = -1;
271 274
275 +/*
276 + * The worst case is single-sector max-parity RAID-Z blocks, in which
277 + * case the space requirement is exactly (VDEV_RAIDZ_MAXPARITY + 1)
278 + * times the size; so just assume that. Add to this the fact that
279 + * we can have up to 3 DVAs per bp, and one more factor of 2 because
280 + * the block may be dittoed with up to 3 DVAs by ddt_sync(). All together,
281 + * the worst case is:
282 + * (VDEV_RAIDZ_MAXPARITY + 1) * SPA_DVAS_PER_BP * 2 == 24
283 + */
284 +int spa_asize_inflation = 24;
272 285
273 286 /*
274 287 * ==========================================================================
275 288 * SPA config locking
276 289 * ==========================================================================
277 290 */
278 291 static void
279 292 spa_config_lock_init(spa_t *spa)
280 293 {
281 294 for (int i = 0; i < SCL_LOCKS; i++) {
282 295 spa_config_lock_t *scl = &spa->spa_config_lock[i];
283 296 mutex_init(&scl->scl_lock, NULL, MUTEX_DEFAULT, NULL);
284 297 cv_init(&scl->scl_cv, NULL, CV_DEFAULT, NULL);
285 298 refcount_create_untracked(&scl->scl_count);
286 299 scl->scl_writer = NULL;
287 300 scl->scl_write_wanted = 0;
288 301 }
289 302 }
290 303
291 304 static void
292 305 spa_config_lock_destroy(spa_t *spa)
293 306 {
294 307 for (int i = 0; i < SCL_LOCKS; i++) {
295 308 spa_config_lock_t *scl = &spa->spa_config_lock[i];
296 309 mutex_destroy(&scl->scl_lock);
297 310 cv_destroy(&scl->scl_cv);
298 311 refcount_destroy(&scl->scl_count);
299 312 ASSERT(scl->scl_writer == NULL);
300 313 ASSERT(scl->scl_write_wanted == 0);
301 314 }
302 315 }
303 316
304 317 int
305 318 spa_config_tryenter(spa_t *spa, int locks, void *tag, krw_t rw)
306 319 {
307 320 for (int i = 0; i < SCL_LOCKS; i++) {
308 321 spa_config_lock_t *scl = &spa->spa_config_lock[i];
309 322 if (!(locks & (1 << i)))
310 323 continue;
311 324 mutex_enter(&scl->scl_lock);
312 325 if (rw == RW_READER) {
313 326 if (scl->scl_writer || scl->scl_write_wanted) {
314 327 mutex_exit(&scl->scl_lock);
315 328 spa_config_exit(spa, locks ^ (1 << i), tag);
316 329 return (0);
317 330 }
318 331 } else {
319 332 ASSERT(scl->scl_writer != curthread);
320 333 if (!refcount_is_zero(&scl->scl_count)) {
321 334 mutex_exit(&scl->scl_lock);
322 335 spa_config_exit(spa, locks ^ (1 << i), tag);
323 336 return (0);
324 337 }
325 338 scl->scl_writer = curthread;
326 339 }
327 340 (void) refcount_add(&scl->scl_count, tag);
328 341 mutex_exit(&scl->scl_lock);
329 342 }
330 343 return (1);
331 344 }
332 345
333 346 void
334 347 spa_config_enter(spa_t *spa, int locks, void *tag, krw_t rw)
335 348 {
336 349 int wlocks_held = 0;
337 350
338 351 ASSERT3U(SCL_LOCKS, <, sizeof (wlocks_held) * NBBY);
339 352
340 353 for (int i = 0; i < SCL_LOCKS; i++) {
341 354 spa_config_lock_t *scl = &spa->spa_config_lock[i];
342 355 if (scl->scl_writer == curthread)
343 356 wlocks_held |= (1 << i);
344 357 if (!(locks & (1 << i)))
345 358 continue;
346 359 mutex_enter(&scl->scl_lock);
347 360 if (rw == RW_READER) {
348 361 while (scl->scl_writer || scl->scl_write_wanted) {
349 362 cv_wait(&scl->scl_cv, &scl->scl_lock);
350 363 }
351 364 } else {
352 365 ASSERT(scl->scl_writer != curthread);
353 366 while (!refcount_is_zero(&scl->scl_count)) {
354 367 scl->scl_write_wanted++;
355 368 cv_wait(&scl->scl_cv, &scl->scl_lock);
356 369 scl->scl_write_wanted--;
357 370 }
358 371 scl->scl_writer = curthread;
359 372 }
360 373 (void) refcount_add(&scl->scl_count, tag);
361 374 mutex_exit(&scl->scl_lock);
362 375 }
363 376 ASSERT(wlocks_held <= locks);
364 377 }
365 378
366 379 void
367 380 spa_config_exit(spa_t *spa, int locks, void *tag)
368 381 {
369 382 for (int i = SCL_LOCKS - 1; i >= 0; i--) {
370 383 spa_config_lock_t *scl = &spa->spa_config_lock[i];
371 384 if (!(locks & (1 << i)))
372 385 continue;
373 386 mutex_enter(&scl->scl_lock);
374 387 ASSERT(!refcount_is_zero(&scl->scl_count));
375 388 if (refcount_remove(&scl->scl_count, tag) == 0) {
376 389 ASSERT(scl->scl_writer == NULL ||
377 390 scl->scl_writer == curthread);
378 391 scl->scl_writer = NULL; /* OK in either case */
379 392 cv_broadcast(&scl->scl_cv);
380 393 }
381 394 mutex_exit(&scl->scl_lock);
382 395 }
383 396 }
384 397
385 398 int
386 399 spa_config_held(spa_t *spa, int locks, krw_t rw)
387 400 {
388 401 int locks_held = 0;
389 402
390 403 for (int i = 0; i < SCL_LOCKS; i++) {
391 404 spa_config_lock_t *scl = &spa->spa_config_lock[i];
392 405 if (!(locks & (1 << i)))
393 406 continue;
394 407 if ((rw == RW_READER && !refcount_is_zero(&scl->scl_count)) ||
395 408 (rw == RW_WRITER && scl->scl_writer == curthread))
396 409 locks_held |= 1 << i;
397 410 }
398 411
399 412 return (locks_held);
400 413 }
401 414
402 415 /*
403 416 * ==========================================================================
404 417 * SPA namespace functions
405 418 * ==========================================================================
406 419 */
407 420
408 421 /*
409 422 * Lookup the named spa_t in the AVL tree. The spa_namespace_lock must be held.
410 423 * Returns NULL if no matching spa_t is found.
411 424 */
412 425 spa_t *
413 426 spa_lookup(const char *name)
414 427 {
415 428 static spa_t search; /* spa_t is large; don't allocate on stack */
416 429 spa_t *spa;
417 430 avl_index_t where;
418 431 char *cp;
419 432
420 433 ASSERT(MUTEX_HELD(&spa_namespace_lock));
421 434
422 435 (void) strlcpy(search.spa_name, name, sizeof (search.spa_name));
423 436
424 437 /*
425 438 * If it's a full dataset name, figure out the pool name and
426 439 * just use that.
427 440 */
428 441 cp = strpbrk(search.spa_name, "/@");
429 442 if (cp != NULL)
430 443 *cp = '\0';
431 444
432 445 spa = avl_find(&spa_namespace_avl, &search, &where);
433 446
434 447 return (spa);
435 448 }
436 449
437 450 /*
438 451 * Fires when spa_sync has not completed within zfs_deadman_synctime_ms.
439 452 * If the zfs_deadman_enabled flag is set then it inspects all vdev queues
440 453 * looking for potentially hung I/Os.
441 454 */
442 455 void
443 456 spa_deadman(void *arg)
444 457 {
445 458 spa_t *spa = arg;
446 459
447 460 zfs_dbgmsg("slow spa_sync: started %llu seconds ago, calls %llu",
448 461 (gethrtime() - spa->spa_sync_starttime) / NANOSEC,
449 462 ++spa->spa_deadman_calls);
450 463 if (zfs_deadman_enabled)
451 464 vdev_deadman(spa->spa_root_vdev);
452 465 }
453 466
454 467 /*
455 468 * Create an uninitialized spa_t with the given name. Requires
456 469 * spa_namespace_lock. The caller must ensure that the spa_t doesn't already
457 470 * exist by calling spa_lookup() first.
458 471 */
459 472 spa_t *
460 473 spa_add(const char *name, nvlist_t *config, const char *altroot)
461 474 {
462 475 spa_t *spa;
463 476 spa_config_dirent_t *dp;
464 477 cyc_handler_t hdlr;
465 478 cyc_time_t when;
466 479
467 480 ASSERT(MUTEX_HELD(&spa_namespace_lock));
468 481
469 482 spa = kmem_zalloc(sizeof (spa_t), KM_SLEEP);
470 483
471 484 mutex_init(&spa->spa_async_lock, NULL, MUTEX_DEFAULT, NULL);
472 485 mutex_init(&spa->spa_errlist_lock, NULL, MUTEX_DEFAULT, NULL);
473 486 mutex_init(&spa->spa_errlog_lock, NULL, MUTEX_DEFAULT, NULL);
474 487 mutex_init(&spa->spa_history_lock, NULL, MUTEX_DEFAULT, NULL);
475 488 mutex_init(&spa->spa_proc_lock, NULL, MUTEX_DEFAULT, NULL);
476 489 mutex_init(&spa->spa_props_lock, NULL, MUTEX_DEFAULT, NULL);
477 490 mutex_init(&spa->spa_scrub_lock, NULL, MUTEX_DEFAULT, NULL);
478 491 mutex_init(&spa->spa_suspend_lock, NULL, MUTEX_DEFAULT, NULL);
479 492 mutex_init(&spa->spa_vdev_top_lock, NULL, MUTEX_DEFAULT, NULL);
480 493 mutex_init(&spa->spa_iokstat_lock, NULL, MUTEX_DEFAULT, NULL);
481 494
482 495 cv_init(&spa->spa_async_cv, NULL, CV_DEFAULT, NULL);
483 496 cv_init(&spa->spa_proc_cv, NULL, CV_DEFAULT, NULL);
484 497 cv_init(&spa->spa_scrub_io_cv, NULL, CV_DEFAULT, NULL);
485 498 cv_init(&spa->spa_suspend_cv, NULL, CV_DEFAULT, NULL);
486 499
487 500 for (int t = 0; t < TXG_SIZE; t++)
488 501 bplist_create(&spa->spa_free_bplist[t]);
489 502
490 503 (void) strlcpy(spa->spa_name, name, sizeof (spa->spa_name));
491 504 spa->spa_state = POOL_STATE_UNINITIALIZED;
|
↓ open down ↓ |
210 lines elided |
↑ open up ↑ |
492 505 spa->spa_freeze_txg = UINT64_MAX;
493 506 spa->spa_final_txg = UINT64_MAX;
494 507 spa->spa_load_max_txg = UINT64_MAX;
495 508 spa->spa_proc = &p0;
496 509 spa->spa_proc_state = SPA_PROC_NONE;
497 510
498 511 hdlr.cyh_func = spa_deadman;
499 512 hdlr.cyh_arg = spa;
500 513 hdlr.cyh_level = CY_LOW_LEVEL;
501 514
502 - spa->spa_deadman_synctime = MSEC2NSEC(zfs_deadman_synctime *
503 - zfs_txg_synctime_ms);
515 + spa->spa_deadman_synctime = MSEC2NSEC(zfs_deadman_synctime_ms);
504 516
505 517 /*
506 518 * This determines how often we need to check for hung I/Os after
507 519 * the cyclic has already fired. Since checking for hung I/Os is
508 520 * an expensive operation we don't want to check too frequently.
509 - * Instead wait for 5 synctimes before checking again.
521 + * Instead wait for 5 seconds before checking again.
510 522 */
511 - when.cyt_interval = MSEC2NSEC(5 * zfs_txg_synctime_ms);
523 + when.cyt_interval = MSEC2NSEC(zfs_deadman_checktime_ms);
512 524 when.cyt_when = CY_INFINITY;
513 525 mutex_enter(&cpu_lock);
514 526 spa->spa_deadman_cycid = cyclic_add(&hdlr, &when);
515 527 mutex_exit(&cpu_lock);
516 528
517 529 refcount_create(&spa->spa_refcount);
518 530 spa_config_lock_init(spa);
519 531
520 532 avl_add(&spa_namespace_avl, spa);
521 533
522 534 /*
523 535 * Set the alternate root, if there is one.
524 536 */
525 537 if (altroot) {
526 538 spa->spa_root = spa_strdup(altroot);
527 539 spa_active_count++;
528 540 }
529 541
530 542 /*
531 543 * Every pool starts with the default cachefile
532 544 */
533 545 list_create(&spa->spa_config_list, sizeof (spa_config_dirent_t),
534 546 offsetof(spa_config_dirent_t, scd_link));
535 547
536 548 dp = kmem_zalloc(sizeof (spa_config_dirent_t), KM_SLEEP);
537 549 dp->scd_path = altroot ? NULL : spa_strdup(spa_config_path);
538 550 list_insert_head(&spa->spa_config_list, dp);
539 551
540 552 VERIFY(nvlist_alloc(&spa->spa_load_info, NV_UNIQUE_NAME,
541 553 KM_SLEEP) == 0);
542 554
543 555 if (config != NULL) {
544 556 nvlist_t *features;
545 557
546 558 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_FEATURES_FOR_READ,
547 559 &features) == 0) {
548 560 VERIFY(nvlist_dup(features, &spa->spa_label_features,
549 561 0) == 0);
550 562 }
551 563
552 564 VERIFY(nvlist_dup(config, &spa->spa_config, 0) == 0);
553 565 }
554 566
555 567 if (spa->spa_label_features == NULL) {
556 568 VERIFY(nvlist_alloc(&spa->spa_label_features, NV_UNIQUE_NAME,
557 569 KM_SLEEP) == 0);
558 570 }
559 571
560 572 spa->spa_iokstat = kstat_create("zfs", 0, name,
561 573 "disk", KSTAT_TYPE_IO, 1, 0);
562 574 if (spa->spa_iokstat) {
563 575 spa->spa_iokstat->ks_lock = &spa->spa_iokstat_lock;
564 576 kstat_install(spa->spa_iokstat);
565 577 }
566 578
567 579 spa->spa_debug = ((zfs_flags & ZFS_DEBUG_SPA) != 0);
568 580
569 581 return (spa);
570 582 }
571 583
572 584 /*
573 585 * Removes a spa_t from the namespace, freeing up any memory used. Requires
574 586 * spa_namespace_lock. This is called only after the spa_t has been closed and
575 587 * deactivated.
576 588 */
577 589 void
578 590 spa_remove(spa_t *spa)
579 591 {
580 592 spa_config_dirent_t *dp;
581 593
582 594 ASSERT(MUTEX_HELD(&spa_namespace_lock));
583 595 ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED);
584 596
585 597 nvlist_free(spa->spa_config_splitting);
586 598
587 599 avl_remove(&spa_namespace_avl, spa);
588 600 cv_broadcast(&spa_namespace_cv);
589 601
590 602 if (spa->spa_root) {
591 603 spa_strfree(spa->spa_root);
592 604 spa_active_count--;
593 605 }
594 606
595 607 while ((dp = list_head(&spa->spa_config_list)) != NULL) {
596 608 list_remove(&spa->spa_config_list, dp);
597 609 if (dp->scd_path != NULL)
598 610 spa_strfree(dp->scd_path);
599 611 kmem_free(dp, sizeof (spa_config_dirent_t));
600 612 }
601 613
602 614 list_destroy(&spa->spa_config_list);
603 615
604 616 nvlist_free(spa->spa_label_features);
605 617 nvlist_free(spa->spa_load_info);
606 618 spa_config_set(spa, NULL);
607 619
608 620 mutex_enter(&cpu_lock);
609 621 if (spa->spa_deadman_cycid != CYCLIC_NONE)
610 622 cyclic_remove(spa->spa_deadman_cycid);
611 623 mutex_exit(&cpu_lock);
612 624 spa->spa_deadman_cycid = CYCLIC_NONE;
613 625
614 626 refcount_destroy(&spa->spa_refcount);
615 627
616 628 spa_config_lock_destroy(spa);
617 629
618 630 kstat_delete(spa->spa_iokstat);
619 631 spa->spa_iokstat = NULL;
620 632
621 633 for (int t = 0; t < TXG_SIZE; t++)
622 634 bplist_destroy(&spa->spa_free_bplist[t]);
623 635
624 636 cv_destroy(&spa->spa_async_cv);
625 637 cv_destroy(&spa->spa_proc_cv);
626 638 cv_destroy(&spa->spa_scrub_io_cv);
627 639 cv_destroy(&spa->spa_suspend_cv);
628 640
629 641 mutex_destroy(&spa->spa_async_lock);
630 642 mutex_destroy(&spa->spa_errlist_lock);
631 643 mutex_destroy(&spa->spa_errlog_lock);
632 644 mutex_destroy(&spa->spa_history_lock);
633 645 mutex_destroy(&spa->spa_proc_lock);
634 646 mutex_destroy(&spa->spa_props_lock);
635 647 mutex_destroy(&spa->spa_scrub_lock);
636 648 mutex_destroy(&spa->spa_suspend_lock);
637 649 mutex_destroy(&spa->spa_vdev_top_lock);
638 650 mutex_destroy(&spa->spa_iokstat_lock);
639 651
640 652 kmem_free(spa, sizeof (spa_t));
641 653 }
642 654
643 655 /*
644 656 * Given a pool, return the next pool in the namespace, or NULL if there is
645 657 * none. If 'prev' is NULL, return the first pool.
646 658 */
647 659 spa_t *
648 660 spa_next(spa_t *prev)
649 661 {
650 662 ASSERT(MUTEX_HELD(&spa_namespace_lock));
651 663
652 664 if (prev)
653 665 return (AVL_NEXT(&spa_namespace_avl, prev));
654 666 else
655 667 return (avl_first(&spa_namespace_avl));
656 668 }
657 669
658 670 /*
659 671 * ==========================================================================
660 672 * SPA refcount functions
661 673 * ==========================================================================
662 674 */
663 675
664 676 /*
665 677 * Add a reference to the given spa_t. Must have at least one reference, or
666 678 * have the namespace lock held.
667 679 */
668 680 void
669 681 spa_open_ref(spa_t *spa, void *tag)
670 682 {
671 683 ASSERT(refcount_count(&spa->spa_refcount) >= spa->spa_minref ||
672 684 MUTEX_HELD(&spa_namespace_lock));
673 685 (void) refcount_add(&spa->spa_refcount, tag);
674 686 }
675 687
676 688 /*
677 689 * Remove a reference to the given spa_t. Must have at least one reference, or
678 690 * have the namespace lock held.
679 691 */
680 692 void
681 693 spa_close(spa_t *spa, void *tag)
682 694 {
683 695 ASSERT(refcount_count(&spa->spa_refcount) > spa->spa_minref ||
684 696 MUTEX_HELD(&spa_namespace_lock));
685 697 (void) refcount_remove(&spa->spa_refcount, tag);
686 698 }
687 699
688 700 /*
689 701 * Check to see if the spa refcount is zero. Must be called with
690 702 * spa_namespace_lock held. We really compare against spa_minref, which is the
691 703 * number of references acquired when opening a pool
692 704 */
693 705 boolean_t
694 706 spa_refcount_zero(spa_t *spa)
695 707 {
696 708 ASSERT(MUTEX_HELD(&spa_namespace_lock));
697 709
698 710 return (refcount_count(&spa->spa_refcount) == spa->spa_minref);
699 711 }
700 712
701 713 /*
702 714 * ==========================================================================
703 715 * SPA spare and l2cache tracking
704 716 * ==========================================================================
705 717 */
706 718
707 719 /*
708 720 * Hot spares and cache devices are tracked using the same code below,
709 721 * for 'auxiliary' devices.
710 722 */
711 723
712 724 typedef struct spa_aux {
713 725 uint64_t aux_guid;
714 726 uint64_t aux_pool;
715 727 avl_node_t aux_avl;
716 728 int aux_count;
717 729 } spa_aux_t;
718 730
719 731 static int
720 732 spa_aux_compare(const void *a, const void *b)
721 733 {
722 734 const spa_aux_t *sa = a;
723 735 const spa_aux_t *sb = b;
724 736
725 737 if (sa->aux_guid < sb->aux_guid)
726 738 return (-1);
727 739 else if (sa->aux_guid > sb->aux_guid)
728 740 return (1);
729 741 else
730 742 return (0);
731 743 }
732 744
733 745 void
734 746 spa_aux_add(vdev_t *vd, avl_tree_t *avl)
735 747 {
736 748 avl_index_t where;
737 749 spa_aux_t search;
738 750 spa_aux_t *aux;
739 751
740 752 search.aux_guid = vd->vdev_guid;
741 753 if ((aux = avl_find(avl, &search, &where)) != NULL) {
742 754 aux->aux_count++;
743 755 } else {
744 756 aux = kmem_zalloc(sizeof (spa_aux_t), KM_SLEEP);
745 757 aux->aux_guid = vd->vdev_guid;
746 758 aux->aux_count = 1;
747 759 avl_insert(avl, aux, where);
748 760 }
749 761 }
750 762
751 763 void
752 764 spa_aux_remove(vdev_t *vd, avl_tree_t *avl)
753 765 {
754 766 spa_aux_t search;
755 767 spa_aux_t *aux;
756 768 avl_index_t where;
757 769
758 770 search.aux_guid = vd->vdev_guid;
759 771 aux = avl_find(avl, &search, &where);
760 772
761 773 ASSERT(aux != NULL);
762 774
763 775 if (--aux->aux_count == 0) {
764 776 avl_remove(avl, aux);
765 777 kmem_free(aux, sizeof (spa_aux_t));
766 778 } else if (aux->aux_pool == spa_guid(vd->vdev_spa)) {
767 779 aux->aux_pool = 0ULL;
768 780 }
769 781 }
770 782
771 783 boolean_t
772 784 spa_aux_exists(uint64_t guid, uint64_t *pool, int *refcnt, avl_tree_t *avl)
773 785 {
774 786 spa_aux_t search, *found;
775 787
776 788 search.aux_guid = guid;
777 789 found = avl_find(avl, &search, NULL);
778 790
779 791 if (pool) {
780 792 if (found)
781 793 *pool = found->aux_pool;
782 794 else
783 795 *pool = 0ULL;
784 796 }
785 797
786 798 if (refcnt) {
787 799 if (found)
788 800 *refcnt = found->aux_count;
789 801 else
790 802 *refcnt = 0;
791 803 }
792 804
793 805 return (found != NULL);
794 806 }
795 807
796 808 void
797 809 spa_aux_activate(vdev_t *vd, avl_tree_t *avl)
798 810 {
799 811 spa_aux_t search, *found;
800 812 avl_index_t where;
801 813
802 814 search.aux_guid = vd->vdev_guid;
803 815 found = avl_find(avl, &search, &where);
804 816 ASSERT(found != NULL);
805 817 ASSERT(found->aux_pool == 0ULL);
806 818
807 819 found->aux_pool = spa_guid(vd->vdev_spa);
808 820 }
809 821
810 822 /*
811 823 * Spares are tracked globally due to the following constraints:
812 824 *
813 825 * - A spare may be part of multiple pools.
814 826 * - A spare may be added to a pool even if it's actively in use within
815 827 * another pool.
816 828 * - A spare in use in any pool can only be the source of a replacement if
817 829 * the target is a spare in the same pool.
818 830 *
819 831 * We keep track of all spares on the system through the use of a reference
820 832 * counted AVL tree. When a vdev is added as a spare, or used as a replacement
821 833 * spare, then we bump the reference count in the AVL tree. In addition, we set
822 834 * the 'vdev_isspare' member to indicate that the device is a spare (active or
823 835 * inactive). When a spare is made active (used to replace a device in the
824 836 * pool), we also keep track of which pool its been made a part of.
825 837 *
826 838 * The 'spa_spare_lock' protects the AVL tree. These functions are normally
827 839 * called under the spa_namespace lock as part of vdev reconfiguration. The
828 840 * separate spare lock exists for the status query path, which does not need to
829 841 * be completely consistent with respect to other vdev configuration changes.
830 842 */
831 843
832 844 static int
833 845 spa_spare_compare(const void *a, const void *b)
834 846 {
835 847 return (spa_aux_compare(a, b));
836 848 }
837 849
838 850 void
839 851 spa_spare_add(vdev_t *vd)
840 852 {
841 853 mutex_enter(&spa_spare_lock);
842 854 ASSERT(!vd->vdev_isspare);
843 855 spa_aux_add(vd, &spa_spare_avl);
844 856 vd->vdev_isspare = B_TRUE;
845 857 mutex_exit(&spa_spare_lock);
846 858 }
847 859
848 860 void
849 861 spa_spare_remove(vdev_t *vd)
850 862 {
851 863 mutex_enter(&spa_spare_lock);
852 864 ASSERT(vd->vdev_isspare);
853 865 spa_aux_remove(vd, &spa_spare_avl);
854 866 vd->vdev_isspare = B_FALSE;
855 867 mutex_exit(&spa_spare_lock);
856 868 }
857 869
858 870 boolean_t
859 871 spa_spare_exists(uint64_t guid, uint64_t *pool, int *refcnt)
860 872 {
861 873 boolean_t found;
862 874
863 875 mutex_enter(&spa_spare_lock);
864 876 found = spa_aux_exists(guid, pool, refcnt, &spa_spare_avl);
865 877 mutex_exit(&spa_spare_lock);
866 878
867 879 return (found);
868 880 }
869 881
870 882 void
871 883 spa_spare_activate(vdev_t *vd)
872 884 {
873 885 mutex_enter(&spa_spare_lock);
874 886 ASSERT(vd->vdev_isspare);
875 887 spa_aux_activate(vd, &spa_spare_avl);
876 888 mutex_exit(&spa_spare_lock);
877 889 }
878 890
879 891 /*
880 892 * Level 2 ARC devices are tracked globally for the same reasons as spares.
881 893 * Cache devices currently only support one pool per cache device, and so
882 894 * for these devices the aux reference count is currently unused beyond 1.
883 895 */
884 896
885 897 static int
886 898 spa_l2cache_compare(const void *a, const void *b)
887 899 {
888 900 return (spa_aux_compare(a, b));
889 901 }
890 902
891 903 void
892 904 spa_l2cache_add(vdev_t *vd)
893 905 {
894 906 mutex_enter(&spa_l2cache_lock);
895 907 ASSERT(!vd->vdev_isl2cache);
896 908 spa_aux_add(vd, &spa_l2cache_avl);
897 909 vd->vdev_isl2cache = B_TRUE;
898 910 mutex_exit(&spa_l2cache_lock);
899 911 }
900 912
901 913 void
902 914 spa_l2cache_remove(vdev_t *vd)
903 915 {
904 916 mutex_enter(&spa_l2cache_lock);
905 917 ASSERT(vd->vdev_isl2cache);
906 918 spa_aux_remove(vd, &spa_l2cache_avl);
907 919 vd->vdev_isl2cache = B_FALSE;
908 920 mutex_exit(&spa_l2cache_lock);
909 921 }
910 922
911 923 boolean_t
912 924 spa_l2cache_exists(uint64_t guid, uint64_t *pool)
913 925 {
914 926 boolean_t found;
915 927
916 928 mutex_enter(&spa_l2cache_lock);
917 929 found = spa_aux_exists(guid, pool, NULL, &spa_l2cache_avl);
918 930 mutex_exit(&spa_l2cache_lock);
919 931
920 932 return (found);
921 933 }
922 934
923 935 void
924 936 spa_l2cache_activate(vdev_t *vd)
925 937 {
926 938 mutex_enter(&spa_l2cache_lock);
927 939 ASSERT(vd->vdev_isl2cache);
928 940 spa_aux_activate(vd, &spa_l2cache_avl);
929 941 mutex_exit(&spa_l2cache_lock);
930 942 }
931 943
932 944 /*
933 945 * ==========================================================================
934 946 * SPA vdev locking
935 947 * ==========================================================================
936 948 */
937 949
938 950 /*
939 951 * Lock the given spa_t for the purpose of adding or removing a vdev.
940 952 * Grabs the global spa_namespace_lock plus the spa config lock for writing.
941 953 * It returns the next transaction group for the spa_t.
942 954 */
943 955 uint64_t
944 956 spa_vdev_enter(spa_t *spa)
945 957 {
946 958 mutex_enter(&spa->spa_vdev_top_lock);
947 959 mutex_enter(&spa_namespace_lock);
948 960 return (spa_vdev_config_enter(spa));
949 961 }
950 962
951 963 /*
952 964 * Internal implementation for spa_vdev_enter(). Used when a vdev
953 965 * operation requires multiple syncs (i.e. removing a device) while
954 966 * keeping the spa_namespace_lock held.
955 967 */
956 968 uint64_t
957 969 spa_vdev_config_enter(spa_t *spa)
958 970 {
959 971 ASSERT(MUTEX_HELD(&spa_namespace_lock));
960 972
961 973 spa_config_enter(spa, SCL_ALL, spa, RW_WRITER);
962 974
963 975 return (spa_last_synced_txg(spa) + 1);
964 976 }
965 977
966 978 /*
967 979 * Used in combination with spa_vdev_config_enter() to allow the syncing
968 980 * of multiple transactions without releasing the spa_namespace_lock.
969 981 */
970 982 void
971 983 spa_vdev_config_exit(spa_t *spa, vdev_t *vd, uint64_t txg, int error, char *tag)
972 984 {
973 985 ASSERT(MUTEX_HELD(&spa_namespace_lock));
974 986
975 987 int config_changed = B_FALSE;
976 988
977 989 ASSERT(txg > spa_last_synced_txg(spa));
978 990
979 991 spa->spa_pending_vdev = NULL;
980 992
981 993 /*
982 994 * Reassess the DTLs.
983 995 */
984 996 vdev_dtl_reassess(spa->spa_root_vdev, 0, 0, B_FALSE);
985 997
986 998 if (error == 0 && !list_is_empty(&spa->spa_config_dirty_list)) {
987 999 config_changed = B_TRUE;
988 1000 spa->spa_config_generation++;
989 1001 }
990 1002
991 1003 /*
992 1004 * Verify the metaslab classes.
993 1005 */
994 1006 ASSERT(metaslab_class_validate(spa_normal_class(spa)) == 0);
995 1007 ASSERT(metaslab_class_validate(spa_log_class(spa)) == 0);
996 1008
997 1009 spa_config_exit(spa, SCL_ALL, spa);
998 1010
999 1011 /*
1000 1012 * Panic the system if the specified tag requires it. This
1001 1013 * is useful for ensuring that configurations are updated
1002 1014 * transactionally.
1003 1015 */
1004 1016 if (zio_injection_enabled)
1005 1017 zio_handle_panic_injection(spa, tag, 0);
1006 1018
1007 1019 /*
1008 1020 * Note: this txg_wait_synced() is important because it ensures
1009 1021 * that there won't be more than one config change per txg.
1010 1022 * This allows us to use the txg as the generation number.
1011 1023 */
1012 1024 if (error == 0)
1013 1025 txg_wait_synced(spa->spa_dsl_pool, txg);
1014 1026
1015 1027 if (vd != NULL) {
1016 1028 ASSERT(!vd->vdev_detached || vd->vdev_dtl_smo.smo_object == 0);
1017 1029 spa_config_enter(spa, SCL_ALL, spa, RW_WRITER);
1018 1030 vdev_free(vd);
1019 1031 spa_config_exit(spa, SCL_ALL, spa);
1020 1032 }
1021 1033
1022 1034 /*
1023 1035 * If the config changed, update the config cache.
1024 1036 */
1025 1037 if (config_changed)
1026 1038 spa_config_sync(spa, B_FALSE, B_TRUE);
1027 1039 }
1028 1040
1029 1041 /*
1030 1042 * Unlock the spa_t after adding or removing a vdev. Besides undoing the
1031 1043 * locking of spa_vdev_enter(), we also want make sure the transactions have
1032 1044 * synced to disk, and then update the global configuration cache with the new
1033 1045 * information.
1034 1046 */
1035 1047 int
1036 1048 spa_vdev_exit(spa_t *spa, vdev_t *vd, uint64_t txg, int error)
1037 1049 {
1038 1050 spa_vdev_config_exit(spa, vd, txg, error, FTAG);
1039 1051 mutex_exit(&spa_namespace_lock);
1040 1052 mutex_exit(&spa->spa_vdev_top_lock);
1041 1053
1042 1054 return (error);
1043 1055 }
1044 1056
1045 1057 /*
1046 1058 * Lock the given spa_t for the purpose of changing vdev state.
1047 1059 */
1048 1060 void
1049 1061 spa_vdev_state_enter(spa_t *spa, int oplocks)
1050 1062 {
1051 1063 int locks = SCL_STATE_ALL | oplocks;
1052 1064
1053 1065 /*
1054 1066 * Root pools may need to read of the underlying devfs filesystem
1055 1067 * when opening up a vdev. Unfortunately if we're holding the
1056 1068 * SCL_ZIO lock it will result in a deadlock when we try to issue
1057 1069 * the read from the root filesystem. Instead we "prefetch"
1058 1070 * the associated vnodes that we need prior to opening the
1059 1071 * underlying devices and cache them so that we can prevent
1060 1072 * any I/O when we are doing the actual open.
1061 1073 */
1062 1074 if (spa_is_root(spa)) {
1063 1075 int low = locks & ~(SCL_ZIO - 1);
1064 1076 int high = locks & ~low;
1065 1077
1066 1078 spa_config_enter(spa, high, spa, RW_WRITER);
1067 1079 vdev_hold(spa->spa_root_vdev);
1068 1080 spa_config_enter(spa, low, spa, RW_WRITER);
1069 1081 } else {
1070 1082 spa_config_enter(spa, locks, spa, RW_WRITER);
1071 1083 }
1072 1084 spa->spa_vdev_locks = locks;
1073 1085 }
1074 1086
1075 1087 int
1076 1088 spa_vdev_state_exit(spa_t *spa, vdev_t *vd, int error)
1077 1089 {
1078 1090 boolean_t config_changed = B_FALSE;
1079 1091
1080 1092 if (vd != NULL || error == 0)
1081 1093 vdev_dtl_reassess(vd ? vd->vdev_top : spa->spa_root_vdev,
1082 1094 0, 0, B_FALSE);
1083 1095
1084 1096 if (vd != NULL) {
1085 1097 vdev_state_dirty(vd->vdev_top);
1086 1098 config_changed = B_TRUE;
1087 1099 spa->spa_config_generation++;
1088 1100 }
1089 1101
1090 1102 if (spa_is_root(spa))
1091 1103 vdev_rele(spa->spa_root_vdev);
1092 1104
1093 1105 ASSERT3U(spa->spa_vdev_locks, >=, SCL_STATE_ALL);
1094 1106 spa_config_exit(spa, spa->spa_vdev_locks, spa);
1095 1107
1096 1108 /*
1097 1109 * If anything changed, wait for it to sync. This ensures that,
1098 1110 * from the system administrator's perspective, zpool(1M) commands
1099 1111 * are synchronous. This is important for things like zpool offline:
1100 1112 * when the command completes, you expect no further I/O from ZFS.
1101 1113 */
1102 1114 if (vd != NULL)
1103 1115 txg_wait_synced(spa->spa_dsl_pool, 0);
1104 1116
1105 1117 /*
1106 1118 * If the config changed, update the config cache.
1107 1119 */
1108 1120 if (config_changed) {
1109 1121 mutex_enter(&spa_namespace_lock);
1110 1122 spa_config_sync(spa, B_FALSE, B_TRUE);
1111 1123 mutex_exit(&spa_namespace_lock);
1112 1124 }
1113 1125
1114 1126 return (error);
1115 1127 }
1116 1128
1117 1129 /*
1118 1130 * ==========================================================================
1119 1131 * Miscellaneous functions
1120 1132 * ==========================================================================
1121 1133 */
1122 1134
1123 1135 void
1124 1136 spa_activate_mos_feature(spa_t *spa, const char *feature)
1125 1137 {
1126 1138 (void) nvlist_add_boolean(spa->spa_label_features, feature);
1127 1139 vdev_config_dirty(spa->spa_root_vdev);
1128 1140 }
1129 1141
1130 1142 void
1131 1143 spa_deactivate_mos_feature(spa_t *spa, const char *feature)
1132 1144 {
1133 1145 (void) nvlist_remove_all(spa->spa_label_features, feature);
1134 1146 vdev_config_dirty(spa->spa_root_vdev);
1135 1147 }
1136 1148
1137 1149 /*
1138 1150 * Rename a spa_t.
1139 1151 */
1140 1152 int
1141 1153 spa_rename(const char *name, const char *newname)
1142 1154 {
1143 1155 spa_t *spa;
1144 1156 int err;
1145 1157
1146 1158 /*
1147 1159 * Lookup the spa_t and grab the config lock for writing. We need to
1148 1160 * actually open the pool so that we can sync out the necessary labels.
1149 1161 * It's OK to call spa_open() with the namespace lock held because we
1150 1162 * allow recursive calls for other reasons.
1151 1163 */
1152 1164 mutex_enter(&spa_namespace_lock);
1153 1165 if ((err = spa_open(name, &spa, FTAG)) != 0) {
1154 1166 mutex_exit(&spa_namespace_lock);
1155 1167 return (err);
1156 1168 }
1157 1169
1158 1170 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1159 1171
1160 1172 avl_remove(&spa_namespace_avl, spa);
1161 1173 (void) strlcpy(spa->spa_name, newname, sizeof (spa->spa_name));
1162 1174 avl_add(&spa_namespace_avl, spa);
1163 1175
1164 1176 /*
1165 1177 * Sync all labels to disk with the new names by marking the root vdev
1166 1178 * dirty and waiting for it to sync. It will pick up the new pool name
1167 1179 * during the sync.
1168 1180 */
1169 1181 vdev_config_dirty(spa->spa_root_vdev);
1170 1182
1171 1183 spa_config_exit(spa, SCL_ALL, FTAG);
1172 1184
1173 1185 txg_wait_synced(spa->spa_dsl_pool, 0);
1174 1186
1175 1187 /*
1176 1188 * Sync the updated config cache.
1177 1189 */
1178 1190 spa_config_sync(spa, B_FALSE, B_TRUE);
1179 1191
1180 1192 spa_close(spa, FTAG);
1181 1193
1182 1194 mutex_exit(&spa_namespace_lock);
1183 1195
1184 1196 return (0);
1185 1197 }
1186 1198
1187 1199 /*
1188 1200 * Return the spa_t associated with given pool_guid, if it exists. If
1189 1201 * device_guid is non-zero, determine whether the pool exists *and* contains
1190 1202 * a device with the specified device_guid.
1191 1203 */
1192 1204 spa_t *
1193 1205 spa_by_guid(uint64_t pool_guid, uint64_t device_guid)
1194 1206 {
1195 1207 spa_t *spa;
1196 1208 avl_tree_t *t = &spa_namespace_avl;
1197 1209
1198 1210 ASSERT(MUTEX_HELD(&spa_namespace_lock));
1199 1211
1200 1212 for (spa = avl_first(t); spa != NULL; spa = AVL_NEXT(t, spa)) {
1201 1213 if (spa->spa_state == POOL_STATE_UNINITIALIZED)
1202 1214 continue;
1203 1215 if (spa->spa_root_vdev == NULL)
1204 1216 continue;
1205 1217 if (spa_guid(spa) == pool_guid) {
1206 1218 if (device_guid == 0)
1207 1219 break;
1208 1220
1209 1221 if (vdev_lookup_by_guid(spa->spa_root_vdev,
1210 1222 device_guid) != NULL)
1211 1223 break;
1212 1224
1213 1225 /*
1214 1226 * Check any devices we may be in the process of adding.
1215 1227 */
1216 1228 if (spa->spa_pending_vdev) {
1217 1229 if (vdev_lookup_by_guid(spa->spa_pending_vdev,
1218 1230 device_guid) != NULL)
1219 1231 break;
1220 1232 }
1221 1233 }
1222 1234 }
1223 1235
1224 1236 return (spa);
1225 1237 }
1226 1238
1227 1239 /*
1228 1240 * Determine whether a pool with the given pool_guid exists.
1229 1241 */
1230 1242 boolean_t
1231 1243 spa_guid_exists(uint64_t pool_guid, uint64_t device_guid)
1232 1244 {
1233 1245 return (spa_by_guid(pool_guid, device_guid) != NULL);
1234 1246 }
1235 1247
1236 1248 char *
1237 1249 spa_strdup(const char *s)
1238 1250 {
1239 1251 size_t len;
1240 1252 char *new;
1241 1253
1242 1254 len = strlen(s);
1243 1255 new = kmem_alloc(len + 1, KM_SLEEP);
1244 1256 bcopy(s, new, len);
1245 1257 new[len] = '\0';
1246 1258
1247 1259 return (new);
1248 1260 }
1249 1261
1250 1262 void
1251 1263 spa_strfree(char *s)
1252 1264 {
1253 1265 kmem_free(s, strlen(s) + 1);
1254 1266 }
1255 1267
1256 1268 uint64_t
1257 1269 spa_get_random(uint64_t range)
1258 1270 {
1259 1271 uint64_t r;
1260 1272
1261 1273 ASSERT(range != 0);
1262 1274
1263 1275 (void) random_get_pseudo_bytes((void *)&r, sizeof (uint64_t));
1264 1276
1265 1277 return (r % range);
1266 1278 }
1267 1279
1268 1280 uint64_t
1269 1281 spa_generate_guid(spa_t *spa)
1270 1282 {
1271 1283 uint64_t guid = spa_get_random(-1ULL);
1272 1284
1273 1285 if (spa != NULL) {
1274 1286 while (guid == 0 || spa_guid_exists(spa_guid(spa), guid))
1275 1287 guid = spa_get_random(-1ULL);
1276 1288 } else {
1277 1289 while (guid == 0 || spa_guid_exists(guid, 0))
1278 1290 guid = spa_get_random(-1ULL);
1279 1291 }
1280 1292
1281 1293 return (guid);
1282 1294 }
1283 1295
1284 1296 void
1285 1297 sprintf_blkptr(char *buf, const blkptr_t *bp)
1286 1298 {
1287 1299 char type[256];
1288 1300 char *checksum = NULL;
1289 1301 char *compress = NULL;
1290 1302
1291 1303 if (bp != NULL) {
1292 1304 if (BP_GET_TYPE(bp) & DMU_OT_NEWTYPE) {
1293 1305 dmu_object_byteswap_t bswap =
1294 1306 DMU_OT_BYTESWAP(BP_GET_TYPE(bp));
1295 1307 (void) snprintf(type, sizeof (type), "bswap %s %s",
1296 1308 DMU_OT_IS_METADATA(BP_GET_TYPE(bp)) ?
1297 1309 "metadata" : "data",
1298 1310 dmu_ot_byteswap[bswap].ob_name);
1299 1311 } else {
1300 1312 (void) strlcpy(type, dmu_ot[BP_GET_TYPE(bp)].ot_name,
1301 1313 sizeof (type));
1302 1314 }
1303 1315 checksum = zio_checksum_table[BP_GET_CHECKSUM(bp)].ci_name;
1304 1316 compress = zio_compress_table[BP_GET_COMPRESS(bp)].ci_name;
1305 1317 }
1306 1318
1307 1319 SPRINTF_BLKPTR(snprintf, ' ', buf, bp, type, checksum, compress);
1308 1320 }
1309 1321
1310 1322 void
1311 1323 spa_freeze(spa_t *spa)
1312 1324 {
1313 1325 uint64_t freeze_txg = 0;
1314 1326
1315 1327 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1316 1328 if (spa->spa_freeze_txg == UINT64_MAX) {
1317 1329 freeze_txg = spa_last_synced_txg(spa) + TXG_SIZE;
1318 1330 spa->spa_freeze_txg = freeze_txg;
1319 1331 }
1320 1332 spa_config_exit(spa, SCL_ALL, FTAG);
1321 1333 if (freeze_txg != 0)
1322 1334 txg_wait_synced(spa_get_dsl(spa), freeze_txg);
1323 1335 }
1324 1336
1325 1337 void
1326 1338 zfs_panic_recover(const char *fmt, ...)
1327 1339 {
1328 1340 va_list adx;
1329 1341
1330 1342 va_start(adx, fmt);
1331 1343 vcmn_err(zfs_recover ? CE_WARN : CE_PANIC, fmt, adx);
1332 1344 va_end(adx);
1333 1345 }
1334 1346
1335 1347 /*
1336 1348 * This is a stripped-down version of strtoull, suitable only for converting
1337 1349 * lowercase hexadecimal numbers that don't overflow.
1338 1350 */
1339 1351 uint64_t
1340 1352 strtonum(const char *str, char **nptr)
1341 1353 {
1342 1354 uint64_t val = 0;
1343 1355 char c;
1344 1356 int digit;
1345 1357
1346 1358 while ((c = *str) != '\0') {
1347 1359 if (c >= '0' && c <= '9')
1348 1360 digit = c - '0';
1349 1361 else if (c >= 'a' && c <= 'f')
1350 1362 digit = 10 + c - 'a';
1351 1363 else
1352 1364 break;
1353 1365
1354 1366 val *= 16;
1355 1367 val += digit;
1356 1368
1357 1369 str++;
1358 1370 }
1359 1371
1360 1372 if (nptr)
1361 1373 *nptr = (char *)str;
1362 1374
1363 1375 return (val);
1364 1376 }
1365 1377
1366 1378 /*
1367 1379 * ==========================================================================
1368 1380 * Accessor functions
1369 1381 * ==========================================================================
1370 1382 */
1371 1383
1372 1384 boolean_t
1373 1385 spa_shutting_down(spa_t *spa)
1374 1386 {
1375 1387 return (spa->spa_async_suspended);
1376 1388 }
1377 1389
1378 1390 dsl_pool_t *
1379 1391 spa_get_dsl(spa_t *spa)
1380 1392 {
1381 1393 return (spa->spa_dsl_pool);
1382 1394 }
1383 1395
1384 1396 boolean_t
1385 1397 spa_is_initializing(spa_t *spa)
1386 1398 {
1387 1399 return (spa->spa_is_initializing);
1388 1400 }
1389 1401
1390 1402 blkptr_t *
1391 1403 spa_get_rootblkptr(spa_t *spa)
1392 1404 {
1393 1405 return (&spa->spa_ubsync.ub_rootbp);
1394 1406 }
1395 1407
1396 1408 void
1397 1409 spa_set_rootblkptr(spa_t *spa, const blkptr_t *bp)
1398 1410 {
1399 1411 spa->spa_uberblock.ub_rootbp = *bp;
1400 1412 }
1401 1413
1402 1414 void
1403 1415 spa_altroot(spa_t *spa, char *buf, size_t buflen)
1404 1416 {
1405 1417 if (spa->spa_root == NULL)
1406 1418 buf[0] = '\0';
1407 1419 else
1408 1420 (void) strncpy(buf, spa->spa_root, buflen);
1409 1421 }
1410 1422
1411 1423 int
1412 1424 spa_sync_pass(spa_t *spa)
1413 1425 {
1414 1426 return (spa->spa_sync_pass);
1415 1427 }
1416 1428
1417 1429 char *
1418 1430 spa_name(spa_t *spa)
1419 1431 {
1420 1432 return (spa->spa_name);
1421 1433 }
1422 1434
1423 1435 uint64_t
1424 1436 spa_guid(spa_t *spa)
1425 1437 {
1426 1438 dsl_pool_t *dp = spa_get_dsl(spa);
1427 1439 uint64_t guid;
1428 1440
1429 1441 /*
1430 1442 * If we fail to parse the config during spa_load(), we can go through
1431 1443 * the error path (which posts an ereport) and end up here with no root
1432 1444 * vdev. We stash the original pool guid in 'spa_config_guid' to handle
1433 1445 * this case.
1434 1446 */
1435 1447 if (spa->spa_root_vdev == NULL)
1436 1448 return (spa->spa_config_guid);
1437 1449
1438 1450 guid = spa->spa_last_synced_guid != 0 ?
1439 1451 spa->spa_last_synced_guid : spa->spa_root_vdev->vdev_guid;
1440 1452
1441 1453 /*
1442 1454 * Return the most recently synced out guid unless we're
1443 1455 * in syncing context.
1444 1456 */
1445 1457 if (dp && dsl_pool_sync_context(dp))
1446 1458 return (spa->spa_root_vdev->vdev_guid);
1447 1459 else
1448 1460 return (guid);
1449 1461 }
1450 1462
1451 1463 uint64_t
1452 1464 spa_load_guid(spa_t *spa)
1453 1465 {
1454 1466 /*
1455 1467 * This is a GUID that exists solely as a reference for the
1456 1468 * purposes of the arc. It is generated at load time, and
1457 1469 * is never written to persistent storage.
1458 1470 */
1459 1471 return (spa->spa_load_guid);
1460 1472 }
1461 1473
1462 1474 uint64_t
1463 1475 spa_last_synced_txg(spa_t *spa)
1464 1476 {
1465 1477 return (spa->spa_ubsync.ub_txg);
1466 1478 }
1467 1479
1468 1480 uint64_t
1469 1481 spa_first_txg(spa_t *spa)
1470 1482 {
1471 1483 return (spa->spa_first_txg);
1472 1484 }
1473 1485
1474 1486 uint64_t
1475 1487 spa_syncing_txg(spa_t *spa)
1476 1488 {
1477 1489 return (spa->spa_syncing_txg);
1478 1490 }
1479 1491
1480 1492 pool_state_t
1481 1493 spa_state(spa_t *spa)
1482 1494 {
1483 1495 return (spa->spa_state);
1484 1496 }
1485 1497
1486 1498 spa_load_state_t
1487 1499 spa_load_state(spa_t *spa)
1488 1500 {
1489 1501 return (spa->spa_load_state);
1490 1502 }
1491 1503
|
↓ open down ↓ |
970 lines elided |
↑ open up ↑ |
1492 1504 uint64_t
1493 1505 spa_freeze_txg(spa_t *spa)
1494 1506 {
1495 1507 return (spa->spa_freeze_txg);
1496 1508 }
1497 1509
1498 1510 /* ARGSUSED */
1499 1511 uint64_t
1500 1512 spa_get_asize(spa_t *spa, uint64_t lsize)
1501 1513 {
1502 - /*
1503 - * The worst case is single-sector max-parity RAID-Z blocks, in which
1504 - * case the space requirement is exactly (VDEV_RAIDZ_MAXPARITY + 1)
1505 - * times the size; so just assume that. Add to this the fact that
1506 - * we can have up to 3 DVAs per bp, and one more factor of 2 because
1507 - * the block may be dittoed with up to 3 DVAs by ddt_sync().
1508 - */
1509 - return (lsize * (VDEV_RAIDZ_MAXPARITY + 1) * SPA_DVAS_PER_BP * 2);
1514 + return (lsize * spa_asize_inflation);
1510 1515 }
1511 1516
1512 1517 uint64_t
1513 1518 spa_get_dspace(spa_t *spa)
1514 1519 {
1515 1520 return (spa->spa_dspace);
1516 1521 }
1517 1522
1518 1523 void
1519 1524 spa_update_dspace(spa_t *spa)
1520 1525 {
1521 1526 spa->spa_dspace = metaslab_class_get_dspace(spa_normal_class(spa)) +
1522 1527 ddt_get_dedup_dspace(spa);
1523 1528 }
1524 1529
1525 1530 /*
1526 1531 * Return the failure mode that has been set to this pool. The default
1527 1532 * behavior will be to block all I/Os when a complete failure occurs.
1528 1533 */
1529 1534 uint8_t
1530 1535 spa_get_failmode(spa_t *spa)
1531 1536 {
1532 1537 return (spa->spa_failmode);
1533 1538 }
1534 1539
1535 1540 boolean_t
1536 1541 spa_suspended(spa_t *spa)
1537 1542 {
1538 1543 return (spa->spa_suspended);
1539 1544 }
1540 1545
1541 1546 uint64_t
1542 1547 spa_version(spa_t *spa)
1543 1548 {
1544 1549 return (spa->spa_ubsync.ub_version);
1545 1550 }
1546 1551
1547 1552 boolean_t
1548 1553 spa_deflate(spa_t *spa)
1549 1554 {
1550 1555 return (spa->spa_deflate);
1551 1556 }
1552 1557
1553 1558 metaslab_class_t *
1554 1559 spa_normal_class(spa_t *spa)
1555 1560 {
1556 1561 return (spa->spa_normal_class);
1557 1562 }
1558 1563
1559 1564 metaslab_class_t *
1560 1565 spa_log_class(spa_t *spa)
1561 1566 {
1562 1567 return (spa->spa_log_class);
1563 1568 }
1564 1569
1565 1570 int
1566 1571 spa_max_replication(spa_t *spa)
1567 1572 {
1568 1573 /*
1569 1574 * As of SPA_VERSION == SPA_VERSION_DITTO_BLOCKS, we are able to
1570 1575 * handle BPs with more than one DVA allocated. Set our max
1571 1576 * replication level accordingly.
1572 1577 */
1573 1578 if (spa_version(spa) < SPA_VERSION_DITTO_BLOCKS)
1574 1579 return (1);
1575 1580 return (MIN(SPA_DVAS_PER_BP, spa_max_replication_override));
1576 1581 }
1577 1582
1578 1583 int
1579 1584 spa_prev_software_version(spa_t *spa)
1580 1585 {
1581 1586 return (spa->spa_prev_software_version);
1582 1587 }
1583 1588
1584 1589 uint64_t
1585 1590 spa_deadman_synctime(spa_t *spa)
1586 1591 {
1587 1592 return (spa->spa_deadman_synctime);
1588 1593 }
1589 1594
1590 1595 uint64_t
1591 1596 dva_get_dsize_sync(spa_t *spa, const dva_t *dva)
1592 1597 {
1593 1598 uint64_t asize = DVA_GET_ASIZE(dva);
1594 1599 uint64_t dsize = asize;
1595 1600
1596 1601 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
1597 1602
1598 1603 if (asize != 0 && spa->spa_deflate) {
1599 1604 vdev_t *vd = vdev_lookup_top(spa, DVA_GET_VDEV(dva));
1600 1605 dsize = (asize >> SPA_MINBLOCKSHIFT) * vd->vdev_deflate_ratio;
1601 1606 }
1602 1607
1603 1608 return (dsize);
1604 1609 }
1605 1610
1606 1611 uint64_t
1607 1612 bp_get_dsize_sync(spa_t *spa, const blkptr_t *bp)
1608 1613 {
1609 1614 uint64_t dsize = 0;
1610 1615
1611 1616 for (int d = 0; d < SPA_DVAS_PER_BP; d++)
1612 1617 dsize += dva_get_dsize_sync(spa, &bp->blk_dva[d]);
1613 1618
1614 1619 return (dsize);
1615 1620 }
1616 1621
1617 1622 uint64_t
1618 1623 bp_get_dsize(spa_t *spa, const blkptr_t *bp)
1619 1624 {
1620 1625 uint64_t dsize = 0;
1621 1626
1622 1627 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
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 spa_config_exit(spa, SCL_VDEV, FTAG);
1628 1633
1629 1634 return (dsize);
1630 1635 }
1631 1636
1632 1637 /*
1633 1638 * ==========================================================================
1634 1639 * Initialization and Termination
1635 1640 * ==========================================================================
1636 1641 */
1637 1642
1638 1643 static int
1639 1644 spa_name_compare(const void *a1, const void *a2)
1640 1645 {
1641 1646 const spa_t *s1 = a1;
1642 1647 const spa_t *s2 = a2;
1643 1648 int s;
1644 1649
1645 1650 s = strcmp(s1->spa_name, s2->spa_name);
1646 1651 if (s > 0)
1647 1652 return (1);
1648 1653 if (s < 0)
1649 1654 return (-1);
1650 1655 return (0);
1651 1656 }
1652 1657
1653 1658 int
1654 1659 spa_busy(void)
1655 1660 {
1656 1661 return (spa_active_count);
1657 1662 }
1658 1663
1659 1664 void
1660 1665 spa_boot_init()
1661 1666 {
1662 1667 spa_config_load();
1663 1668 }
1664 1669
1665 1670 void
1666 1671 spa_init(int mode)
1667 1672 {
1668 1673 mutex_init(&spa_namespace_lock, NULL, MUTEX_DEFAULT, NULL);
1669 1674 mutex_init(&spa_spare_lock, NULL, MUTEX_DEFAULT, NULL);
1670 1675 mutex_init(&spa_l2cache_lock, NULL, MUTEX_DEFAULT, NULL);
1671 1676 cv_init(&spa_namespace_cv, NULL, CV_DEFAULT, NULL);
1672 1677
1673 1678 avl_create(&spa_namespace_avl, spa_name_compare, sizeof (spa_t),
1674 1679 offsetof(spa_t, spa_avl));
1675 1680
1676 1681 avl_create(&spa_spare_avl, spa_spare_compare, sizeof (spa_aux_t),
1677 1682 offsetof(spa_aux_t, aux_avl));
1678 1683
1679 1684 avl_create(&spa_l2cache_avl, spa_l2cache_compare, sizeof (spa_aux_t),
1680 1685 offsetof(spa_aux_t, aux_avl));
1681 1686
1682 1687 spa_mode_global = mode;
1683 1688
1684 1689 #ifdef _KERNEL
1685 1690 spa_arch_init();
1686 1691 #else
1687 1692 if (spa_mode_global != FREAD && dprintf_find_string("watch")) {
1688 1693 arc_procfd = open("/proc/self/ctl", O_WRONLY);
1689 1694 if (arc_procfd == -1) {
1690 1695 perror("could not enable watchpoints: "
1691 1696 "opening /proc/self/ctl failed: ");
1692 1697 } else {
1693 1698 arc_watch = B_TRUE;
1694 1699 }
1695 1700 }
1696 1701 #endif
1697 1702
1698 1703 refcount_init();
1699 1704 unique_init();
1700 1705 space_map_init();
1701 1706 zio_init();
1702 1707 dmu_init();
1703 1708 zil_init();
1704 1709 vdev_cache_stat_init();
1705 1710 zfs_prop_init();
1706 1711 zpool_prop_init();
1707 1712 zpool_feature_init();
1708 1713 spa_config_load();
1709 1714 l2arc_start();
1710 1715 }
1711 1716
1712 1717 void
1713 1718 spa_fini(void)
1714 1719 {
1715 1720 l2arc_stop();
1716 1721
1717 1722 spa_evict_all();
1718 1723
1719 1724 vdev_cache_stat_fini();
1720 1725 zil_fini();
1721 1726 dmu_fini();
1722 1727 zio_fini();
1723 1728 space_map_fini();
1724 1729 unique_fini();
1725 1730 refcount_fini();
1726 1731
1727 1732 avl_destroy(&spa_namespace_avl);
1728 1733 avl_destroy(&spa_spare_avl);
1729 1734 avl_destroy(&spa_l2cache_avl);
1730 1735
1731 1736 cv_destroy(&spa_namespace_cv);
1732 1737 mutex_destroy(&spa_namespace_lock);
1733 1738 mutex_destroy(&spa_spare_lock);
1734 1739 mutex_destroy(&spa_l2cache_lock);
1735 1740 }
1736 1741
1737 1742 /*
1738 1743 * Return whether this pool has slogs. No locking needed.
1739 1744 * It's not a problem if the wrong answer is returned as it's only for
1740 1745 * performance and not correctness
1741 1746 */
1742 1747 boolean_t
1743 1748 spa_has_slogs(spa_t *spa)
1744 1749 {
1745 1750 return (spa->spa_log_class->mc_rotor != NULL);
1746 1751 }
1747 1752
1748 1753 spa_log_state_t
1749 1754 spa_get_log_state(spa_t *spa)
1750 1755 {
1751 1756 return (spa->spa_log_state);
1752 1757 }
1753 1758
1754 1759 void
1755 1760 spa_set_log_state(spa_t *spa, spa_log_state_t state)
1756 1761 {
1757 1762 spa->spa_log_state = state;
1758 1763 }
1759 1764
1760 1765 boolean_t
1761 1766 spa_is_root(spa_t *spa)
1762 1767 {
1763 1768 return (spa->spa_is_root);
1764 1769 }
1765 1770
1766 1771 boolean_t
1767 1772 spa_writeable(spa_t *spa)
1768 1773 {
1769 1774 return (!!(spa->spa_mode & FWRITE));
1770 1775 }
1771 1776
1772 1777 int
1773 1778 spa_mode(spa_t *spa)
1774 1779 {
1775 1780 return (spa->spa_mode);
1776 1781 }
1777 1782
1778 1783 uint64_t
1779 1784 spa_bootfs(spa_t *spa)
1780 1785 {
1781 1786 return (spa->spa_bootfs);
1782 1787 }
1783 1788
1784 1789 uint64_t
1785 1790 spa_delegation(spa_t *spa)
1786 1791 {
1787 1792 return (spa->spa_delegation);
1788 1793 }
1789 1794
1790 1795 objset_t *
1791 1796 spa_meta_objset(spa_t *spa)
1792 1797 {
1793 1798 return (spa->spa_meta_objset);
1794 1799 }
1795 1800
1796 1801 enum zio_checksum
1797 1802 spa_dedup_checksum(spa_t *spa)
1798 1803 {
1799 1804 return (spa->spa_dedup_checksum);
1800 1805 }
1801 1806
1802 1807 /*
1803 1808 * Reset pool scan stat per scan pass (or reboot).
1804 1809 */
1805 1810 void
1806 1811 spa_scan_stat_init(spa_t *spa)
1807 1812 {
1808 1813 /* data not stored on disk */
1809 1814 spa->spa_scan_pass_start = gethrestime_sec();
1810 1815 spa->spa_scan_pass_exam = 0;
1811 1816 vdev_scan_stat_init(spa->spa_root_vdev);
1812 1817 }
1813 1818
1814 1819 /*
1815 1820 * Get scan stats for zpool status reports
1816 1821 */
1817 1822 int
1818 1823 spa_scan_get_stats(spa_t *spa, pool_scan_stat_t *ps)
1819 1824 {
1820 1825 dsl_scan_t *scn = spa->spa_dsl_pool ? spa->spa_dsl_pool->dp_scan : NULL;
1821 1826
1822 1827 if (scn == NULL || scn->scn_phys.scn_func == POOL_SCAN_NONE)
1823 1828 return (SET_ERROR(ENOENT));
1824 1829 bzero(ps, sizeof (pool_scan_stat_t));
1825 1830
1826 1831 /* data stored on disk */
1827 1832 ps->pss_func = scn->scn_phys.scn_func;
1828 1833 ps->pss_start_time = scn->scn_phys.scn_start_time;
1829 1834 ps->pss_end_time = scn->scn_phys.scn_end_time;
1830 1835 ps->pss_to_examine = scn->scn_phys.scn_to_examine;
1831 1836 ps->pss_examined = scn->scn_phys.scn_examined;
1832 1837 ps->pss_to_process = scn->scn_phys.scn_to_process;
1833 1838 ps->pss_processed = scn->scn_phys.scn_processed;
1834 1839 ps->pss_errors = scn->scn_phys.scn_errors;
1835 1840 ps->pss_state = scn->scn_phys.scn_state;
1836 1841
1837 1842 /* data not stored on disk */
1838 1843 ps->pss_pass_start = spa->spa_scan_pass_start;
1839 1844 ps->pss_pass_exam = spa->spa_scan_pass_exam;
1840 1845
1841 1846 return (0);
1842 1847 }
1843 1848
1844 1849 boolean_t
1845 1850 spa_debug_enabled(spa_t *spa)
1846 1851 {
1847 1852 return (spa->spa_debug);
1848 1853 }
|
↓ open down ↓ |
329 lines elided |
↑ open up ↑ |
XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX