1 /*
   2  * CDDL HEADER START
   3  *
   4  * The contents of this file are subject to the terms of the
   5  * Common Development and Distribution License (the "License").
   6  * You may not use this file except in compliance with the License.
   7  *
   8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
   9  * or http://www.opensolaris.org/os/licensing.
  10  * See the License for the specific language governing permissions
  11  * and limitations under the License.
  12  *
  13  * When distributing Covered Code, include this CDDL HEADER in each
  14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
  15  * If applicable, add the following below this CDDL HEADER, with the
  16  * fields enclosed by brackets "[]" replaced with your own identifying
  17  * information: Portions Copyright [yyyy] [name of copyright owner]
  18  *
  19  * CDDL HEADER END
  20  */
  21 /*
  22  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
  23  * Copyright (c) 2013 by Delphix. All rights reserved.
  24  * Copyright 2011 Nexenta Systems, Inc.  All rights reserved.
  25  */
  26 
  27 #include <sys/zfs_context.h>
  28 #include <sys/spa_impl.h>
  29 #include <sys/spa_boot.h>
  30 #include <sys/zio.h>
  31 #include <sys/zio_checksum.h>
  32 #include <sys/zio_compress.h>
  33 #include <sys/dmu.h>
  34 #include <sys/dmu_tx.h>
  35 #include <sys/zap.h>
  36 #include <sys/zil.h>
  37 #include <sys/vdev_impl.h>
  38 #include <sys/metaslab.h>
  39 #include <sys/uberblock_impl.h>
  40 #include <sys/txg.h>
  41 #include <sys/avl.h>
  42 #include <sys/unique.h>
  43 #include <sys/dsl_pool.h>
  44 #include <sys/dsl_dir.h>
  45 #include <sys/dsl_prop.h>
  46 #include <sys/dsl_scan.h>
  47 #include <sys/fs/zfs.h>
  48 #include <sys/metaslab_impl.h>
  49 #include <sys/arc.h>
  50 #include <sys/ddt.h>
  51 #include "zfs_prop.h"
  52 #include "zfeature_common.h"
  53 
  54 /*
  55  * SPA locking
  56  *
  57  * There are four basic locks for managing spa_t structures:
  58  *
  59  * spa_namespace_lock (global mutex)
  60  *
  61  *      This lock must be acquired to do any of the following:
  62  *
  63  *              - Lookup a spa_t by name
  64  *              - Add or remove a spa_t from the namespace
  65  *              - Increase spa_refcount from non-zero
  66  *              - Check if spa_refcount is zero
  67  *              - Rename a spa_t
  68  *              - add/remove/attach/detach devices
  69  *              - Held for the duration of create/destroy/import/export
  70  *
  71  *      It does not need to handle recursion.  A create or destroy may
  72  *      reference objects (files or zvols) in other pools, but by
  73  *      definition they must have an existing reference, and will never need
  74  *      to lookup a spa_t by name.
  75  *
  76  * spa_refcount (per-spa refcount_t protected by mutex)
  77  *
  78  *      This reference count keep track of any active users of the spa_t.  The
  79  *      spa_t cannot be destroyed or freed while this is non-zero.  Internally,
  80  *      the refcount is never really 'zero' - opening a pool implicitly keeps
  81  *      some references in the DMU.  Internally we check against spa_minref, but
  82  *      present the image of a zero/non-zero value to consumers.
  83  *
  84  * spa_config_lock[] (per-spa array of rwlocks)
  85  *
  86  *      This protects the spa_t from config changes, and must be held in
  87  *      the following circumstances:
  88  *
  89  *              - RW_READER to perform I/O to the spa
  90  *              - RW_WRITER to change the vdev config
  91  *
  92  * The locking order is fairly straightforward:
  93  *
  94  *              spa_namespace_lock      ->   spa_refcount
  95  *
  96  *      The namespace lock must be acquired to increase the refcount from 0
  97  *      or to check if it is zero.
  98  *
  99  *              spa_refcount            ->   spa_config_lock[]
 100  *
 101  *      There must be at least one valid reference on the spa_t to acquire
 102  *      the config lock.
 103  *
 104  *              spa_namespace_lock      ->   spa_config_lock[]
 105  *
 106  *      The namespace lock must always be taken before the config lock.
 107  *
 108  *
 109  * The spa_namespace_lock can be acquired directly and is globally visible.
 110  *
 111  * The namespace is manipulated using the following functions, all of which
 112  * require the spa_namespace_lock to be held.
 113  *
 114  *      spa_lookup()            Lookup a spa_t by name.
 115  *
 116  *      spa_add()               Create a new spa_t in the namespace.
 117  *
 118  *      spa_remove()            Remove a spa_t from the namespace.  This also
 119  *                              frees up any memory associated with the spa_t.
 120  *
 121  *      spa_next()              Returns the next spa_t in the system, or the
 122  *                              first if NULL is passed.
 123  *
 124  *      spa_evict_all()         Shutdown and remove all spa_t structures in
 125  *                              the system.
 126  *
 127  *      spa_guid_exists()       Determine whether a pool/device guid exists.
 128  *
 129  * The spa_refcount is manipulated using the following functions:
 130  *
 131  *      spa_open_ref()          Adds a reference to the given spa_t.  Must be
 132  *                              called with spa_namespace_lock held if the
 133  *                              refcount is currently zero.
 134  *
 135  *      spa_close()             Remove a reference from the spa_t.  This will
 136  *                              not free the spa_t or remove it from the
 137  *                              namespace.  No locking is required.
 138  *
 139  *      spa_refcount_zero()     Returns true if the refcount is currently
 140  *                              zero.  Must be called with spa_namespace_lock
 141  *                              held.
 142  *
 143  * The spa_config_lock[] is an array of rwlocks, ordered as follows:
 144  * SCL_CONFIG > SCL_STATE > SCL_ALLOC > SCL_ZIO > SCL_FREE > SCL_VDEV.
 145  * spa_config_lock[] is manipulated with spa_config_{enter,exit,held}().
 146  *
 147  * To read the configuration, it suffices to hold one of these locks as reader.
 148  * To modify the configuration, you must hold all locks as writer.  To modify
 149  * vdev state without altering the vdev tree's topology (e.g. online/offline),
 150  * you must hold SCL_STATE and SCL_ZIO as writer.
 151  *
 152  * We use these distinct config locks to avoid recursive lock entry.
 153  * For example, spa_sync() (which holds SCL_CONFIG as reader) induces
 154  * block allocations (SCL_ALLOC), which may require reading space maps
 155  * from disk (dmu_read() -> zio_read() -> SCL_ZIO).
 156  *
 157  * The spa config locks cannot be normal rwlocks because we need the
 158  * ability to hand off ownership.  For example, SCL_ZIO is acquired
 159  * by the issuing thread and later released by an interrupt thread.
 160  * They do, however, obey the usual write-wanted semantics to prevent
 161  * writer (i.e. system administrator) starvation.
 162  *
 163  * The lock acquisition rules are as follows:
 164  *
 165  * SCL_CONFIG
 166  *      Protects changes to the vdev tree topology, such as vdev
 167  *      add/remove/attach/detach.  Protects the dirty config list
 168  *      (spa_config_dirty_list) and the set of spares and l2arc devices.
 169  *
 170  * SCL_STATE
 171  *      Protects changes to pool state and vdev state, such as vdev
 172  *      online/offline/fault/degrade/clear.  Protects the dirty state list
 173  *      (spa_state_dirty_list) and global pool state (spa_state).
 174  *
 175  * SCL_ALLOC
 176  *      Protects changes to metaslab groups and classes.
 177  *      Held as reader by metaslab_alloc() and metaslab_claim().
 178  *
 179  * SCL_ZIO
 180  *      Held by bp-level zios (those which have no io_vd upon entry)
 181  *      to prevent changes to the vdev tree.  The bp-level zio implicitly
 182  *      protects all of its vdev child zios, which do not hold SCL_ZIO.
 183  *
 184  * SCL_FREE
 185  *      Protects changes to metaslab groups and classes.
 186  *      Held as reader by metaslab_free().  SCL_FREE is distinct from
 187  *      SCL_ALLOC, and lower than SCL_ZIO, so that we can safely free
 188  *      blocks in zio_done() while another i/o that holds either
 189  *      SCL_ALLOC or SCL_ZIO is waiting for this i/o to complete.
 190  *
 191  * SCL_VDEV
 192  *      Held as reader to prevent changes to the vdev tree during trivial
 193  *      inquiries such as bp_get_dsize().  SCL_VDEV is distinct from the
 194  *      other locks, and lower than all of them, to ensure that it's safe
 195  *      to acquire regardless of caller context.
 196  *
 197  * In addition, the following rules apply:
 198  *
 199  * (a)  spa_props_lock protects pool properties, spa_config and spa_config_list.
 200  *      The lock ordering is SCL_CONFIG > spa_props_lock.
 201  *
 202  * (b)  I/O operations on leaf vdevs.  For any zio operation that takes
 203  *      an explicit vdev_t argument -- such as zio_ioctl(), zio_read_phys(),
 204  *      or zio_write_phys() -- the caller must ensure that the config cannot
 205  *      cannot change in the interim, and that the vdev cannot be reopened.
 206  *      SCL_STATE as reader suffices for both.
 207  *
 208  * The vdev configuration is protected by spa_vdev_enter() / spa_vdev_exit().
 209  *
 210  *      spa_vdev_enter()        Acquire the namespace lock and the config lock
 211  *                              for writing.
 212  *
 213  *      spa_vdev_exit()         Release the config lock, wait for all I/O
 214  *                              to complete, sync the updated configs to the
 215  *                              cache, and release the namespace lock.
 216  *
 217  * vdev state is protected by spa_vdev_state_enter() / spa_vdev_state_exit().
 218  * Like spa_vdev_enter/exit, these are convenience wrappers -- the actual
 219  * locking is, always, based on spa_namespace_lock and spa_config_lock[].
 220  *
 221  * spa_rename() is also implemented within this file since it requires
 222  * manipulation of the namespace.
 223  */
 224 
 225 static avl_tree_t spa_namespace_avl;
 226 kmutex_t spa_namespace_lock;
 227 static kcondvar_t spa_namespace_cv;
 228 static int spa_active_count;
 229 int spa_max_replication_override = SPA_DVAS_PER_BP;
 230 
 231 static kmutex_t spa_spare_lock;
 232 static avl_tree_t spa_spare_avl;
 233 static kmutex_t spa_l2cache_lock;
 234 static avl_tree_t spa_l2cache_avl;
 235 
 236 kmem_cache_t *spa_buffer_pool;
 237 int spa_mode_global;
 238 
 239 #ifdef ZFS_DEBUG
 240 /* Everything except dprintf and spa is on by default in debug builds */
 241 int zfs_flags = ~(ZFS_DEBUG_DPRINTF | ZFS_DEBUG_SPA);
 242 #else
 243 int zfs_flags = 0;
 244 #endif
 245 
 246 /*
 247  * zfs_recover can be set to nonzero to attempt to recover from
 248  * otherwise-fatal errors, typically caused by on-disk corruption.  When
 249  * set, calls to zfs_panic_recover() will turn into warning messages.
 250  */
 251 int zfs_recover = 0;
 252 
 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;
 262 
 263 /*
 264  * Check time in milliseconds. This defines the frequency at which we check
 265  * for hung I/O.
 266  */
 267 uint64_t zfs_deadman_checktime_ms = 5000ULL;
 268 
 269 /*
 270  * Override the zfs deadman behavior via /etc/system. By default the
 271  * deadman is enabled except on VMware and sparc deployments.
 272  */
 273 int zfs_deadman_enabled = -1;
 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;
 285 
 286 /*
 287  * ==========================================================================
 288  * SPA config locking
 289  * ==========================================================================
 290  */
 291 static void
 292 spa_config_lock_init(spa_t *spa)
 293 {
 294         for (int i = 0; i < SCL_LOCKS; i++) {
 295                 spa_config_lock_t *scl = &spa->spa_config_lock[i];
 296                 mutex_init(&scl->scl_lock, NULL, MUTEX_DEFAULT, NULL);
 297                 cv_init(&scl->scl_cv, NULL, CV_DEFAULT, NULL);
 298                 refcount_create_untracked(&scl->scl_count);
 299                 scl->scl_writer = NULL;
 300                 scl->scl_write_wanted = 0;
 301         }
 302 }
 303 
 304 static void
 305 spa_config_lock_destroy(spa_t *spa)
 306 {
 307         for (int i = 0; i < SCL_LOCKS; i++) {
 308                 spa_config_lock_t *scl = &spa->spa_config_lock[i];
 309                 mutex_destroy(&scl->scl_lock);
 310                 cv_destroy(&scl->scl_cv);
 311                 refcount_destroy(&scl->scl_count);
 312                 ASSERT(scl->scl_writer == NULL);
 313                 ASSERT(scl->scl_write_wanted == 0);
 314         }
 315 }
 316 
 317 int
 318 spa_config_tryenter(spa_t *spa, int locks, void *tag, krw_t rw)
 319 {
 320         for (int i = 0; i < SCL_LOCKS; i++) {
 321                 spa_config_lock_t *scl = &spa->spa_config_lock[i];
 322                 if (!(locks & (1 << i)))
 323                         continue;
 324                 mutex_enter(&scl->scl_lock);
 325                 if (rw == RW_READER) {
 326                         if (scl->scl_writer || scl->scl_write_wanted) {
 327                                 mutex_exit(&scl->scl_lock);
 328                                 spa_config_exit(spa, locks ^ (1 << i), tag);
 329                                 return (0);
 330                         }
 331                 } else {
 332                         ASSERT(scl->scl_writer != curthread);
 333                         if (!refcount_is_zero(&scl->scl_count)) {
 334                                 mutex_exit(&scl->scl_lock);
 335                                 spa_config_exit(spa, locks ^ (1 << i), tag);
 336                                 return (0);
 337                         }
 338                         scl->scl_writer = curthread;
 339                 }
 340                 (void) refcount_add(&scl->scl_count, tag);
 341                 mutex_exit(&scl->scl_lock);
 342         }
 343         return (1);
 344 }
 345 
 346 void
 347 spa_config_enter(spa_t *spa, int locks, void *tag, krw_t rw)
 348 {
 349         int wlocks_held = 0;
 350 
 351         ASSERT3U(SCL_LOCKS, <, sizeof (wlocks_held) * NBBY);
 352 
 353         for (int i = 0; i < SCL_LOCKS; i++) {
 354                 spa_config_lock_t *scl = &spa->spa_config_lock[i];
 355                 if (scl->scl_writer == curthread)
 356                         wlocks_held |= (1 << i);
 357                 if (!(locks & (1 << i)))
 358                         continue;
 359                 mutex_enter(&scl->scl_lock);
 360                 if (rw == RW_READER) {
 361                         while (scl->scl_writer || scl->scl_write_wanted) {
 362                                 cv_wait(&scl->scl_cv, &scl->scl_lock);
 363                         }
 364                 } else {
 365                         ASSERT(scl->scl_writer != curthread);
 366                         while (!refcount_is_zero(&scl->scl_count)) {
 367                                 scl->scl_write_wanted++;
 368                                 cv_wait(&scl->scl_cv, &scl->scl_lock);
 369                                 scl->scl_write_wanted--;
 370                         }
 371                         scl->scl_writer = curthread;
 372                 }
 373                 (void) refcount_add(&scl->scl_count, tag);
 374                 mutex_exit(&scl->scl_lock);
 375         }
 376         ASSERT(wlocks_held <= locks);
 377 }
 378 
 379 void
 380 spa_config_exit(spa_t *spa, int locks, void *tag)
 381 {
 382         for (int i = SCL_LOCKS - 1; i >= 0; i--) {
 383                 spa_config_lock_t *scl = &spa->spa_config_lock[i];
 384                 if (!(locks & (1 << i)))
 385                         continue;
 386                 mutex_enter(&scl->scl_lock);
 387                 ASSERT(!refcount_is_zero(&scl->scl_count));
 388                 if (refcount_remove(&scl->scl_count, tag) == 0) {
 389                         ASSERT(scl->scl_writer == NULL ||
 390                             scl->scl_writer == curthread);
 391                         scl->scl_writer = NULL;      /* OK in either case */
 392                         cv_broadcast(&scl->scl_cv);
 393                 }
 394                 mutex_exit(&scl->scl_lock);
 395         }
 396 }
 397 
 398 int
 399 spa_config_held(spa_t *spa, int locks, krw_t rw)
 400 {
 401         int locks_held = 0;
 402 
 403         for (int i = 0; i < SCL_LOCKS; i++) {
 404                 spa_config_lock_t *scl = &spa->spa_config_lock[i];
 405                 if (!(locks & (1 << i)))
 406                         continue;
 407                 if ((rw == RW_READER && !refcount_is_zero(&scl->scl_count)) ||
 408                     (rw == RW_WRITER && scl->scl_writer == curthread))
 409                         locks_held |= 1 << i;
 410         }
 411 
 412         return (locks_held);
 413 }
 414 
 415 /*
 416  * ==========================================================================
 417  * SPA namespace functions
 418  * ==========================================================================
 419  */
 420 
 421 /*
 422  * Lookup the named spa_t in the AVL tree.  The spa_namespace_lock must be held.
 423  * Returns NULL if no matching spa_t is found.
 424  */
 425 spa_t *
 426 spa_lookup(const char *name)
 427 {
 428         static spa_t search;    /* spa_t is large; don't allocate on stack */
 429         spa_t *spa;
 430         avl_index_t where;
 431         char *cp;
 432 
 433         ASSERT(MUTEX_HELD(&spa_namespace_lock));
 434 
 435         (void) strlcpy(search.spa_name, name, sizeof (search.spa_name));
 436 
 437         /*
 438          * If it's a full dataset name, figure out the pool name and
 439          * just use that.
 440          */
 441         cp = strpbrk(search.spa_name, "/@");
 442         if (cp != NULL)
 443                 *cp = '\0';
 444 
 445         spa = avl_find(&spa_namespace_avl, &search, &where);
 446 
 447         return (spa);
 448 }
 449 
 450 /*
 451  * Fires when spa_sync has not completed within zfs_deadman_synctime_ms.
 452  * If the zfs_deadman_enabled flag is set then it inspects all vdev queues
 453  * looking for potentially hung I/Os.
 454  */
 455 void
 456 spa_deadman(void *arg)
 457 {
 458         spa_t *spa = arg;
 459 
 460         /*
 461          * Disable the deadman timer if the pool is suspended.
 462          */
 463         if (spa_suspended(spa)) {
 464                 VERIFY(cyclic_reprogram(spa->spa_deadman_cycid, CY_INFINITY));
 465                 return;
 466         }
 467 
 468         zfs_dbgmsg("slow spa_sync: started %llu seconds ago, calls %llu",
 469             (gethrtime() - spa->spa_sync_starttime) / NANOSEC,
 470             ++spa->spa_deadman_calls);
 471         if (zfs_deadman_enabled)
 472                 vdev_deadman(spa->spa_root_vdev);
 473 }
 474 
 475 /*
 476  * Create an uninitialized spa_t with the given name.  Requires
 477  * spa_namespace_lock.  The caller must ensure that the spa_t doesn't already
 478  * exist by calling spa_lookup() first.
 479  */
 480 spa_t *
 481 spa_add(const char *name, nvlist_t *config, const char *altroot)
 482 {
 483         spa_t *spa;
 484         spa_config_dirent_t *dp;
 485         cyc_handler_t hdlr;
 486         cyc_time_t when;
 487 
 488         ASSERT(MUTEX_HELD(&spa_namespace_lock));
 489 
 490         spa = kmem_zalloc(sizeof (spa_t), KM_SLEEP);
 491 
 492         mutex_init(&spa->spa_async_lock, NULL, MUTEX_DEFAULT, NULL);
 493         mutex_init(&spa->spa_errlist_lock, NULL, MUTEX_DEFAULT, NULL);
 494         mutex_init(&spa->spa_errlog_lock, NULL, MUTEX_DEFAULT, NULL);
 495         mutex_init(&spa->spa_history_lock, NULL, MUTEX_DEFAULT, NULL);
 496         mutex_init(&spa->spa_proc_lock, NULL, MUTEX_DEFAULT, NULL);
 497         mutex_init(&spa->spa_props_lock, NULL, MUTEX_DEFAULT, NULL);
 498         mutex_init(&spa->spa_scrub_lock, NULL, MUTEX_DEFAULT, NULL);
 499         mutex_init(&spa->spa_suspend_lock, NULL, MUTEX_DEFAULT, NULL);
 500         mutex_init(&spa->spa_vdev_top_lock, NULL, MUTEX_DEFAULT, NULL);
 501         mutex_init(&spa->spa_iokstat_lock, NULL, MUTEX_DEFAULT, NULL);
 502 
 503         cv_init(&spa->spa_async_cv, NULL, CV_DEFAULT, NULL);
 504         cv_init(&spa->spa_proc_cv, NULL, CV_DEFAULT, NULL);
 505         cv_init(&spa->spa_scrub_io_cv, NULL, CV_DEFAULT, NULL);
 506         cv_init(&spa->spa_suspend_cv, NULL, CV_DEFAULT, NULL);
 507 
 508         for (int t = 0; t < TXG_SIZE; t++)
 509                 bplist_create(&spa->spa_free_bplist[t]);
 510 
 511         (void) strlcpy(spa->spa_name, name, sizeof (spa->spa_name));
 512         spa->spa_state = POOL_STATE_UNINITIALIZED;
 513         spa->spa_freeze_txg = UINT64_MAX;
 514         spa->spa_final_txg = UINT64_MAX;
 515         spa->spa_load_max_txg = UINT64_MAX;
 516         spa->spa_proc = &p0;
 517         spa->spa_proc_state = SPA_PROC_NONE;
 518 
 519         hdlr.cyh_func = spa_deadman;
 520         hdlr.cyh_arg = spa;
 521         hdlr.cyh_level = CY_LOW_LEVEL;
 522 
 523         spa->spa_deadman_synctime = MSEC2NSEC(zfs_deadman_synctime_ms);
 524 
 525         /*
 526          * This determines how often we need to check for hung I/Os after
 527          * the cyclic has already fired. Since checking for hung I/Os is
 528          * an expensive operation we don't want to check too frequently.
 529          * Instead wait for 5 seconds before checking again.
 530          */
 531         when.cyt_interval = MSEC2NSEC(zfs_deadman_checktime_ms);
 532         when.cyt_when = CY_INFINITY;
 533         mutex_enter(&cpu_lock);
 534         spa->spa_deadman_cycid = cyclic_add(&hdlr, &when);
 535         mutex_exit(&cpu_lock);
 536 
 537         refcount_create(&spa->spa_refcount);
 538         spa_config_lock_init(spa);
 539 
 540         avl_add(&spa_namespace_avl, spa);
 541 
 542         /*
 543          * Set the alternate root, if there is one.
 544          */
 545         if (altroot) {
 546                 spa->spa_root = spa_strdup(altroot);
 547                 spa_active_count++;
 548         }
 549 
 550         /*
 551          * Every pool starts with the default cachefile
 552          */
 553         list_create(&spa->spa_config_list, sizeof (spa_config_dirent_t),
 554             offsetof(spa_config_dirent_t, scd_link));
 555 
 556         dp = kmem_zalloc(sizeof (spa_config_dirent_t), KM_SLEEP);
 557         dp->scd_path = altroot ? NULL : spa_strdup(spa_config_path);
 558         list_insert_head(&spa->spa_config_list, dp);
 559 
 560         VERIFY(nvlist_alloc(&spa->spa_load_info, NV_UNIQUE_NAME,
 561             KM_SLEEP) == 0);
 562 
 563         if (config != NULL) {
 564                 nvlist_t *features;
 565 
 566                 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_FEATURES_FOR_READ,
 567                     &features) == 0) {
 568                         VERIFY(nvlist_dup(features, &spa->spa_label_features,
 569                             0) == 0);
 570                 }
 571 
 572                 VERIFY(nvlist_dup(config, &spa->spa_config, 0) == 0);
 573         }
 574 
 575         if (spa->spa_label_features == NULL) {
 576                 VERIFY(nvlist_alloc(&spa->spa_label_features, NV_UNIQUE_NAME,
 577                     KM_SLEEP) == 0);
 578         }
 579 
 580         spa->spa_iokstat = kstat_create("zfs", 0, name,
 581             "disk", KSTAT_TYPE_IO, 1, 0);
 582         if (spa->spa_iokstat) {
 583                 spa->spa_iokstat->ks_lock = &spa->spa_iokstat_lock;
 584                 kstat_install(spa->spa_iokstat);
 585         }
 586 
 587         spa->spa_debug = ((zfs_flags & ZFS_DEBUG_SPA) != 0);
 588 
 589         return (spa);
 590 }
 591 
 592 /*
 593  * Removes a spa_t from the namespace, freeing up any memory used.  Requires
 594  * spa_namespace_lock.  This is called only after the spa_t has been closed and
 595  * deactivated.
 596  */
 597 void
 598 spa_remove(spa_t *spa)
 599 {
 600         spa_config_dirent_t *dp;
 601 
 602         ASSERT(MUTEX_HELD(&spa_namespace_lock));
 603         ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED);
 604 
 605         nvlist_free(spa->spa_config_splitting);
 606 
 607         avl_remove(&spa_namespace_avl, spa);
 608         cv_broadcast(&spa_namespace_cv);
 609 
 610         if (spa->spa_root) {
 611                 spa_strfree(spa->spa_root);
 612                 spa_active_count--;
 613         }
 614 
 615         while ((dp = list_head(&spa->spa_config_list)) != NULL) {
 616                 list_remove(&spa->spa_config_list, dp);
 617                 if (dp->scd_path != NULL)
 618                         spa_strfree(dp->scd_path);
 619                 kmem_free(dp, sizeof (spa_config_dirent_t));
 620         }
 621 
 622         list_destroy(&spa->spa_config_list);
 623 
 624         nvlist_free(spa->spa_label_features);
 625         nvlist_free(spa->spa_load_info);
 626         spa_config_set(spa, NULL);
 627 
 628         mutex_enter(&cpu_lock);
 629         if (spa->spa_deadman_cycid != CYCLIC_NONE)
 630                 cyclic_remove(spa->spa_deadman_cycid);
 631         mutex_exit(&cpu_lock);
 632         spa->spa_deadman_cycid = CYCLIC_NONE;
 633 
 634         refcount_destroy(&spa->spa_refcount);
 635 
 636         spa_config_lock_destroy(spa);
 637 
 638         kstat_delete(spa->spa_iokstat);
 639         spa->spa_iokstat = NULL;
 640 
 641         for (int t = 0; t < TXG_SIZE; t++)
 642                 bplist_destroy(&spa->spa_free_bplist[t]);
 643 
 644         cv_destroy(&spa->spa_async_cv);
 645         cv_destroy(&spa->spa_proc_cv);
 646         cv_destroy(&spa->spa_scrub_io_cv);
 647         cv_destroy(&spa->spa_suspend_cv);
 648 
 649         mutex_destroy(&spa->spa_async_lock);
 650         mutex_destroy(&spa->spa_errlist_lock);
 651         mutex_destroy(&spa->spa_errlog_lock);
 652         mutex_destroy(&spa->spa_history_lock);
 653         mutex_destroy(&spa->spa_proc_lock);
 654         mutex_destroy(&spa->spa_props_lock);
 655         mutex_destroy(&spa->spa_scrub_lock);
 656         mutex_destroy(&spa->spa_suspend_lock);
 657         mutex_destroy(&spa->spa_vdev_top_lock);
 658         mutex_destroy(&spa->spa_iokstat_lock);
 659 
 660         kmem_free(spa, sizeof (spa_t));
 661 }
 662 
 663 /*
 664  * Given a pool, return the next pool in the namespace, or NULL if there is
 665  * none.  If 'prev' is NULL, return the first pool.
 666  */
 667 spa_t *
 668 spa_next(spa_t *prev)
 669 {
 670         ASSERT(MUTEX_HELD(&spa_namespace_lock));
 671 
 672         if (prev)
 673                 return (AVL_NEXT(&spa_namespace_avl, prev));
 674         else
 675                 return (avl_first(&spa_namespace_avl));
 676 }
 677 
 678 /*
 679  * ==========================================================================
 680  * SPA refcount functions
 681  * ==========================================================================
 682  */
 683 
 684 /*
 685  * Add a reference to the given spa_t.  Must have at least one reference, or
 686  * have the namespace lock held.
 687  */
 688 void
 689 spa_open_ref(spa_t *spa, void *tag)
 690 {
 691         ASSERT(refcount_count(&spa->spa_refcount) >= spa->spa_minref ||
 692             MUTEX_HELD(&spa_namespace_lock));
 693         (void) refcount_add(&spa->spa_refcount, tag);
 694 }
 695 
 696 /*
 697  * Remove a reference to the given spa_t.  Must have at least one reference, or
 698  * have the namespace lock held.
 699  */
 700 void
 701 spa_close(spa_t *spa, void *tag)
 702 {
 703         ASSERT(refcount_count(&spa->spa_refcount) > spa->spa_minref ||
 704             MUTEX_HELD(&spa_namespace_lock));
 705         (void) refcount_remove(&spa->spa_refcount, tag);
 706 }
 707 
 708 /*
 709  * Check to see if the spa refcount is zero.  Must be called with
 710  * spa_namespace_lock held.  We really compare against spa_minref, which is the
 711  * number of references acquired when opening a pool
 712  */
 713 boolean_t
 714 spa_refcount_zero(spa_t *spa)
 715 {
 716         ASSERT(MUTEX_HELD(&spa_namespace_lock));
 717 
 718         return (refcount_count(&spa->spa_refcount) == spa->spa_minref);
 719 }
 720 
 721 /*
 722  * ==========================================================================
 723  * SPA spare and l2cache tracking
 724  * ==========================================================================
 725  */
 726 
 727 /*
 728  * Hot spares and cache devices are tracked using the same code below,
 729  * for 'auxiliary' devices.
 730  */
 731 
 732 typedef struct spa_aux {
 733         uint64_t        aux_guid;
 734         uint64_t        aux_pool;
 735         avl_node_t      aux_avl;
 736         int             aux_count;
 737 } spa_aux_t;
 738 
 739 static int
 740 spa_aux_compare(const void *a, const void *b)
 741 {
 742         const spa_aux_t *sa = a;
 743         const spa_aux_t *sb = b;
 744 
 745         if (sa->aux_guid < sb->aux_guid)
 746                 return (-1);
 747         else if (sa->aux_guid > sb->aux_guid)
 748                 return (1);
 749         else
 750                 return (0);
 751 }
 752 
 753 void
 754 spa_aux_add(vdev_t *vd, avl_tree_t *avl)
 755 {
 756         avl_index_t where;
 757         spa_aux_t search;
 758         spa_aux_t *aux;
 759 
 760         search.aux_guid = vd->vdev_guid;
 761         if ((aux = avl_find(avl, &search, &where)) != NULL) {
 762                 aux->aux_count++;
 763         } else {
 764                 aux = kmem_zalloc(sizeof (spa_aux_t), KM_SLEEP);
 765                 aux->aux_guid = vd->vdev_guid;
 766                 aux->aux_count = 1;
 767                 avl_insert(avl, aux, where);
 768         }
 769 }
 770 
 771 void
 772 spa_aux_remove(vdev_t *vd, avl_tree_t *avl)
 773 {
 774         spa_aux_t search;
 775         spa_aux_t *aux;
 776         avl_index_t where;
 777 
 778         search.aux_guid = vd->vdev_guid;
 779         aux = avl_find(avl, &search, &where);
 780 
 781         ASSERT(aux != NULL);
 782 
 783         if (--aux->aux_count == 0) {
 784                 avl_remove(avl, aux);
 785                 kmem_free(aux, sizeof (spa_aux_t));
 786         } else if (aux->aux_pool == spa_guid(vd->vdev_spa)) {
 787                 aux->aux_pool = 0ULL;
 788         }
 789 }
 790 
 791 boolean_t
 792 spa_aux_exists(uint64_t guid, uint64_t *pool, int *refcnt, avl_tree_t *avl)
 793 {
 794         spa_aux_t search, *found;
 795 
 796         search.aux_guid = guid;
 797         found = avl_find(avl, &search, NULL);
 798 
 799         if (pool) {
 800                 if (found)
 801                         *pool = found->aux_pool;
 802                 else
 803                         *pool = 0ULL;
 804         }
 805 
 806         if (refcnt) {
 807                 if (found)
 808                         *refcnt = found->aux_count;
 809                 else
 810                         *refcnt = 0;
 811         }
 812 
 813         return (found != NULL);
 814 }
 815 
 816 void
 817 spa_aux_activate(vdev_t *vd, avl_tree_t *avl)
 818 {
 819         spa_aux_t search, *found;
 820         avl_index_t where;
 821 
 822         search.aux_guid = vd->vdev_guid;
 823         found = avl_find(avl, &search, &where);
 824         ASSERT(found != NULL);
 825         ASSERT(found->aux_pool == 0ULL);
 826 
 827         found->aux_pool = spa_guid(vd->vdev_spa);
 828 }
 829 
 830 /*
 831  * Spares are tracked globally due to the following constraints:
 832  *
 833  *      - A spare may be part of multiple pools.
 834  *      - A spare may be added to a pool even if it's actively in use within
 835  *        another pool.
 836  *      - A spare in use in any pool can only be the source of a replacement if
 837  *        the target is a spare in the same pool.
 838  *
 839  * We keep track of all spares on the system through the use of a reference
 840  * counted AVL tree.  When a vdev is added as a spare, or used as a replacement
 841  * spare, then we bump the reference count in the AVL tree.  In addition, we set
 842  * the 'vdev_isspare' member to indicate that the device is a spare (active or
 843  * inactive).  When a spare is made active (used to replace a device in the
 844  * pool), we also keep track of which pool its been made a part of.
 845  *
 846  * The 'spa_spare_lock' protects the AVL tree.  These functions are normally
 847  * called under the spa_namespace lock as part of vdev reconfiguration.  The
 848  * separate spare lock exists for the status query path, which does not need to
 849  * be completely consistent with respect to other vdev configuration changes.
 850  */
 851 
 852 static int
 853 spa_spare_compare(const void *a, const void *b)
 854 {
 855         return (spa_aux_compare(a, b));
 856 }
 857 
 858 void
 859 spa_spare_add(vdev_t *vd)
 860 {
 861         mutex_enter(&spa_spare_lock);
 862         ASSERT(!vd->vdev_isspare);
 863         spa_aux_add(vd, &spa_spare_avl);
 864         vd->vdev_isspare = B_TRUE;
 865         mutex_exit(&spa_spare_lock);
 866 }
 867 
 868 void
 869 spa_spare_remove(vdev_t *vd)
 870 {
 871         mutex_enter(&spa_spare_lock);
 872         ASSERT(vd->vdev_isspare);
 873         spa_aux_remove(vd, &spa_spare_avl);
 874         vd->vdev_isspare = B_FALSE;
 875         mutex_exit(&spa_spare_lock);
 876 }
 877 
 878 boolean_t
 879 spa_spare_exists(uint64_t guid, uint64_t *pool, int *refcnt)
 880 {
 881         boolean_t found;
 882 
 883         mutex_enter(&spa_spare_lock);
 884         found = spa_aux_exists(guid, pool, refcnt, &spa_spare_avl);
 885         mutex_exit(&spa_spare_lock);
 886 
 887         return (found);
 888 }
 889 
 890 void
 891 spa_spare_activate(vdev_t *vd)
 892 {
 893         mutex_enter(&spa_spare_lock);
 894         ASSERT(vd->vdev_isspare);
 895         spa_aux_activate(vd, &spa_spare_avl);
 896         mutex_exit(&spa_spare_lock);
 897 }
 898 
 899 /*
 900  * Level 2 ARC devices are tracked globally for the same reasons as spares.
 901  * Cache devices currently only support one pool per cache device, and so
 902  * for these devices the aux reference count is currently unused beyond 1.
 903  */
 904 
 905 static int
 906 spa_l2cache_compare(const void *a, const void *b)
 907 {
 908         return (spa_aux_compare(a, b));
 909 }
 910 
 911 void
 912 spa_l2cache_add(vdev_t *vd)
 913 {
 914         mutex_enter(&spa_l2cache_lock);
 915         ASSERT(!vd->vdev_isl2cache);
 916         spa_aux_add(vd, &spa_l2cache_avl);
 917         vd->vdev_isl2cache = B_TRUE;
 918         mutex_exit(&spa_l2cache_lock);
 919 }
 920 
 921 void
 922 spa_l2cache_remove(vdev_t *vd)
 923 {
 924         mutex_enter(&spa_l2cache_lock);
 925         ASSERT(vd->vdev_isl2cache);
 926         spa_aux_remove(vd, &spa_l2cache_avl);
 927         vd->vdev_isl2cache = B_FALSE;
 928         mutex_exit(&spa_l2cache_lock);
 929 }
 930 
 931 boolean_t
 932 spa_l2cache_exists(uint64_t guid, uint64_t *pool)
 933 {
 934         boolean_t found;
 935 
 936         mutex_enter(&spa_l2cache_lock);
 937         found = spa_aux_exists(guid, pool, NULL, &spa_l2cache_avl);
 938         mutex_exit(&spa_l2cache_lock);
 939 
 940         return (found);
 941 }
 942 
 943 void
 944 spa_l2cache_activate(vdev_t *vd)
 945 {
 946         mutex_enter(&spa_l2cache_lock);
 947         ASSERT(vd->vdev_isl2cache);
 948         spa_aux_activate(vd, &spa_l2cache_avl);
 949         mutex_exit(&spa_l2cache_lock);
 950 }
 951 
 952 /*
 953  * ==========================================================================
 954  * SPA vdev locking
 955  * ==========================================================================
 956  */
 957 
 958 /*
 959  * Lock the given spa_t for the purpose of adding or removing a vdev.
 960  * Grabs the global spa_namespace_lock plus the spa config lock for writing.
 961  * It returns the next transaction group for the spa_t.
 962  */
 963 uint64_t
 964 spa_vdev_enter(spa_t *spa)
 965 {
 966         mutex_enter(&spa->spa_vdev_top_lock);
 967         mutex_enter(&spa_namespace_lock);
 968         return (spa_vdev_config_enter(spa));
 969 }
 970 
 971 /*
 972  * Internal implementation for spa_vdev_enter().  Used when a vdev
 973  * operation requires multiple syncs (i.e. removing a device) while
 974  * keeping the spa_namespace_lock held.
 975  */
 976 uint64_t
 977 spa_vdev_config_enter(spa_t *spa)
 978 {
 979         ASSERT(MUTEX_HELD(&spa_namespace_lock));
 980 
 981         spa_config_enter(spa, SCL_ALL, spa, RW_WRITER);
 982 
 983         return (spa_last_synced_txg(spa) + 1);
 984 }
 985 
 986 /*
 987  * Used in combination with spa_vdev_config_enter() to allow the syncing
 988  * of multiple transactions without releasing the spa_namespace_lock.
 989  */
 990 void
 991 spa_vdev_config_exit(spa_t *spa, vdev_t *vd, uint64_t txg, int error, char *tag)
 992 {
 993         ASSERT(MUTEX_HELD(&spa_namespace_lock));
 994 
 995         int config_changed = B_FALSE;
 996 
 997         ASSERT(txg > spa_last_synced_txg(spa));
 998 
 999         spa->spa_pending_vdev = NULL;
1000 
1001         /*
1002          * Reassess the DTLs.
1003          */
1004         vdev_dtl_reassess(spa->spa_root_vdev, 0, 0, B_FALSE);
1005 
1006         if (error == 0 && !list_is_empty(&spa->spa_config_dirty_list)) {
1007                 config_changed = B_TRUE;
1008                 spa->spa_config_generation++;
1009         }
1010 
1011         /*
1012          * Verify the metaslab classes.
1013          */
1014         ASSERT(metaslab_class_validate(spa_normal_class(spa)) == 0);
1015         ASSERT(metaslab_class_validate(spa_log_class(spa)) == 0);
1016 
1017         spa_config_exit(spa, SCL_ALL, spa);
1018 
1019         /*
1020          * Panic the system if the specified tag requires it.  This
1021          * is useful for ensuring that configurations are updated
1022          * transactionally.
1023          */
1024         if (zio_injection_enabled)
1025                 zio_handle_panic_injection(spa, tag, 0);
1026 
1027         /*
1028          * Note: this txg_wait_synced() is important because it ensures
1029          * that there won't be more than one config change per txg.
1030          * This allows us to use the txg as the generation number.
1031          */
1032         if (error == 0)
1033                 txg_wait_synced(spa->spa_dsl_pool, txg);
1034 
1035         if (vd != NULL) {
1036                 ASSERT(!vd->vdev_detached || vd->vdev_dtl_sm == NULL);
1037                 spa_config_enter(spa, SCL_ALL, spa, RW_WRITER);
1038                 vdev_free(vd);
1039                 spa_config_exit(spa, SCL_ALL, spa);
1040         }
1041 
1042         /*
1043          * If the config changed, update the config cache.
1044          */
1045         if (config_changed)
1046                 spa_config_sync(spa, B_FALSE, B_TRUE);
1047 }
1048 
1049 /*
1050  * Unlock the spa_t after adding or removing a vdev.  Besides undoing the
1051  * locking of spa_vdev_enter(), we also want make sure the transactions have
1052  * synced to disk, and then update the global configuration cache with the new
1053  * information.
1054  */
1055 int
1056 spa_vdev_exit(spa_t *spa, vdev_t *vd, uint64_t txg, int error)
1057 {
1058         spa_vdev_config_exit(spa, vd, txg, error, FTAG);
1059         mutex_exit(&spa_namespace_lock);
1060         mutex_exit(&spa->spa_vdev_top_lock);
1061 
1062         return (error);
1063 }
1064 
1065 /*
1066  * Lock the given spa_t for the purpose of changing vdev state.
1067  */
1068 void
1069 spa_vdev_state_enter(spa_t *spa, int oplocks)
1070 {
1071         int locks = SCL_STATE_ALL | oplocks;
1072 
1073         /*
1074          * Root pools may need to read of the underlying devfs filesystem
1075          * when opening up a vdev.  Unfortunately if we're holding the
1076          * SCL_ZIO lock it will result in a deadlock when we try to issue
1077          * the read from the root filesystem.  Instead we "prefetch"
1078          * the associated vnodes that we need prior to opening the
1079          * underlying devices and cache them so that we can prevent
1080          * any I/O when we are doing the actual open.
1081          */
1082         if (spa_is_root(spa)) {
1083                 int low = locks & ~(SCL_ZIO - 1);
1084                 int high = locks & ~low;
1085 
1086                 spa_config_enter(spa, high, spa, RW_WRITER);
1087                 vdev_hold(spa->spa_root_vdev);
1088                 spa_config_enter(spa, low, spa, RW_WRITER);
1089         } else {
1090                 spa_config_enter(spa, locks, spa, RW_WRITER);
1091         }
1092         spa->spa_vdev_locks = locks;
1093 }
1094 
1095 int
1096 spa_vdev_state_exit(spa_t *spa, vdev_t *vd, int error)
1097 {
1098         boolean_t config_changed = B_FALSE;
1099 
1100         if (vd != NULL || error == 0)
1101                 vdev_dtl_reassess(vd ? vd->vdev_top : spa->spa_root_vdev,
1102                     0, 0, B_FALSE);
1103 
1104         if (vd != NULL) {
1105                 vdev_state_dirty(vd->vdev_top);
1106                 config_changed = B_TRUE;
1107                 spa->spa_config_generation++;
1108         }
1109 
1110         if (spa_is_root(spa))
1111                 vdev_rele(spa->spa_root_vdev);
1112 
1113         ASSERT3U(spa->spa_vdev_locks, >=, SCL_STATE_ALL);
1114         spa_config_exit(spa, spa->spa_vdev_locks, spa);
1115 
1116         /*
1117          * If anything changed, wait for it to sync.  This ensures that,
1118          * from the system administrator's perspective, zpool(1M) commands
1119          * are synchronous.  This is important for things like zpool offline:
1120          * when the command completes, you expect no further I/O from ZFS.
1121          */
1122         if (vd != NULL)
1123                 txg_wait_synced(spa->spa_dsl_pool, 0);
1124 
1125         /*
1126          * If the config changed, update the config cache.
1127          */
1128         if (config_changed) {
1129                 mutex_enter(&spa_namespace_lock);
1130                 spa_config_sync(spa, B_FALSE, B_TRUE);
1131                 mutex_exit(&spa_namespace_lock);
1132         }
1133 
1134         return (error);
1135 }
1136 
1137 /*
1138  * ==========================================================================
1139  * Miscellaneous functions
1140  * ==========================================================================
1141  */
1142 
1143 void
1144 spa_activate_mos_feature(spa_t *spa, const char *feature)
1145 {
1146         (void) nvlist_add_boolean(spa->spa_label_features, feature);
1147         vdev_config_dirty(spa->spa_root_vdev);
1148 }
1149 
1150 void
1151 spa_deactivate_mos_feature(spa_t *spa, const char *feature)
1152 {
1153         (void) nvlist_remove_all(spa->spa_label_features, feature);
1154         vdev_config_dirty(spa->spa_root_vdev);
1155 }
1156 
1157 /*
1158  * Rename a spa_t.
1159  */
1160 int
1161 spa_rename(const char *name, const char *newname)
1162 {
1163         spa_t *spa;
1164         int err;
1165 
1166         /*
1167          * Lookup the spa_t and grab the config lock for writing.  We need to
1168          * actually open the pool so that we can sync out the necessary labels.
1169          * It's OK to call spa_open() with the namespace lock held because we
1170          * allow recursive calls for other reasons.
1171          */
1172         mutex_enter(&spa_namespace_lock);
1173         if ((err = spa_open(name, &spa, FTAG)) != 0) {
1174                 mutex_exit(&spa_namespace_lock);
1175                 return (err);
1176         }
1177 
1178         spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1179 
1180         avl_remove(&spa_namespace_avl, spa);
1181         (void) strlcpy(spa->spa_name, newname, sizeof (spa->spa_name));
1182         avl_add(&spa_namespace_avl, spa);
1183 
1184         /*
1185          * Sync all labels to disk with the new names by marking the root vdev
1186          * dirty and waiting for it to sync.  It will pick up the new pool name
1187          * during the sync.
1188          */
1189         vdev_config_dirty(spa->spa_root_vdev);
1190 
1191         spa_config_exit(spa, SCL_ALL, FTAG);
1192 
1193         txg_wait_synced(spa->spa_dsl_pool, 0);
1194 
1195         /*
1196          * Sync the updated config cache.
1197          */
1198         spa_config_sync(spa, B_FALSE, B_TRUE);
1199 
1200         spa_close(spa, FTAG);
1201 
1202         mutex_exit(&spa_namespace_lock);
1203 
1204         return (0);
1205 }
1206 
1207 /*
1208  * Return the spa_t associated with given pool_guid, if it exists.  If
1209  * device_guid is non-zero, determine whether the pool exists *and* contains
1210  * a device with the specified device_guid.
1211  */
1212 spa_t *
1213 spa_by_guid(uint64_t pool_guid, uint64_t device_guid)
1214 {
1215         spa_t *spa;
1216         avl_tree_t *t = &spa_namespace_avl;
1217 
1218         ASSERT(MUTEX_HELD(&spa_namespace_lock));
1219 
1220         for (spa = avl_first(t); spa != NULL; spa = AVL_NEXT(t, spa)) {
1221                 if (spa->spa_state == POOL_STATE_UNINITIALIZED)
1222                         continue;
1223                 if (spa->spa_root_vdev == NULL)
1224                         continue;
1225                 if (spa_guid(spa) == pool_guid) {
1226                         if (device_guid == 0)
1227                                 break;
1228 
1229                         if (vdev_lookup_by_guid(spa->spa_root_vdev,
1230                             device_guid) != NULL)
1231                                 break;
1232 
1233                         /*
1234                          * Check any devices we may be in the process of adding.
1235                          */
1236                         if (spa->spa_pending_vdev) {
1237                                 if (vdev_lookup_by_guid(spa->spa_pending_vdev,
1238                                     device_guid) != NULL)
1239                                         break;
1240                         }
1241                 }
1242         }
1243 
1244         return (spa);
1245 }
1246 
1247 /*
1248  * Determine whether a pool with the given pool_guid exists.
1249  */
1250 boolean_t
1251 spa_guid_exists(uint64_t pool_guid, uint64_t device_guid)
1252 {
1253         return (spa_by_guid(pool_guid, device_guid) != NULL);
1254 }
1255 
1256 char *
1257 spa_strdup(const char *s)
1258 {
1259         size_t len;
1260         char *new;
1261 
1262         len = strlen(s);
1263         new = kmem_alloc(len + 1, KM_SLEEP);
1264         bcopy(s, new, len);
1265         new[len] = '\0';
1266 
1267         return (new);
1268 }
1269 
1270 void
1271 spa_strfree(char *s)
1272 {
1273         kmem_free(s, strlen(s) + 1);
1274 }
1275 
1276 uint64_t
1277 spa_get_random(uint64_t range)
1278 {
1279         uint64_t r;
1280 
1281         ASSERT(range != 0);
1282 
1283         (void) random_get_pseudo_bytes((void *)&r, sizeof (uint64_t));
1284 
1285         return (r % range);
1286 }
1287 
1288 uint64_t
1289 spa_generate_guid(spa_t *spa)
1290 {
1291         uint64_t guid = spa_get_random(-1ULL);
1292 
1293         if (spa != NULL) {
1294                 while (guid == 0 || spa_guid_exists(spa_guid(spa), guid))
1295                         guid = spa_get_random(-1ULL);
1296         } else {
1297                 while (guid == 0 || spa_guid_exists(guid, 0))
1298                         guid = spa_get_random(-1ULL);
1299         }
1300 
1301         return (guid);
1302 }
1303 
1304 void
1305 sprintf_blkptr(char *buf, const blkptr_t *bp)
1306 {
1307         char type[256];
1308         char *checksum = NULL;
1309         char *compress = NULL;
1310 
1311         if (bp != NULL) {
1312                 if (BP_GET_TYPE(bp) & DMU_OT_NEWTYPE) {
1313                         dmu_object_byteswap_t bswap =
1314                             DMU_OT_BYTESWAP(BP_GET_TYPE(bp));
1315                         (void) snprintf(type, sizeof (type), "bswap %s %s",
1316                             DMU_OT_IS_METADATA(BP_GET_TYPE(bp)) ?
1317                             "metadata" : "data",
1318                             dmu_ot_byteswap[bswap].ob_name);
1319                 } else {
1320                         (void) strlcpy(type, dmu_ot[BP_GET_TYPE(bp)].ot_name,
1321                             sizeof (type));
1322                 }
1323                 checksum = zio_checksum_table[BP_GET_CHECKSUM(bp)].ci_name;
1324                 compress = zio_compress_table[BP_GET_COMPRESS(bp)].ci_name;
1325         }
1326 
1327         SPRINTF_BLKPTR(snprintf, ' ', buf, bp, type, checksum, compress);
1328 }
1329 
1330 void
1331 spa_freeze(spa_t *spa)
1332 {
1333         uint64_t freeze_txg = 0;
1334 
1335         spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1336         if (spa->spa_freeze_txg == UINT64_MAX) {
1337                 freeze_txg = spa_last_synced_txg(spa) + TXG_SIZE;
1338                 spa->spa_freeze_txg = freeze_txg;
1339         }
1340         spa_config_exit(spa, SCL_ALL, FTAG);
1341         if (freeze_txg != 0)
1342                 txg_wait_synced(spa_get_dsl(spa), freeze_txg);
1343 }
1344 
1345 void
1346 zfs_panic_recover(const char *fmt, ...)
1347 {
1348         va_list adx;
1349 
1350         va_start(adx, fmt);
1351         vcmn_err(zfs_recover ? CE_WARN : CE_PANIC, fmt, adx);
1352         va_end(adx);
1353 }
1354 
1355 /*
1356  * This is a stripped-down version of strtoull, suitable only for converting
1357  * lowercase hexadecimal numbers that don't overflow.
1358  */
1359 uint64_t
1360 strtonum(const char *str, char **nptr)
1361 {
1362         uint64_t val = 0;
1363         char c;
1364         int digit;
1365 
1366         while ((c = *str) != '\0') {
1367                 if (c >= '0' && c <= '9')
1368                         digit = c - '0';
1369                 else if (c >= 'a' && c <= 'f')
1370                         digit = 10 + c - 'a';
1371                 else
1372                         break;
1373 
1374                 val *= 16;
1375                 val += digit;
1376 
1377                 str++;
1378         }
1379 
1380         if (nptr)
1381                 *nptr = (char *)str;
1382 
1383         return (val);
1384 }
1385 
1386 /*
1387  * ==========================================================================
1388  * Accessor functions
1389  * ==========================================================================
1390  */
1391 
1392 boolean_t
1393 spa_shutting_down(spa_t *spa)
1394 {
1395         return (spa->spa_async_suspended);
1396 }
1397 
1398 dsl_pool_t *
1399 spa_get_dsl(spa_t *spa)
1400 {
1401         return (spa->spa_dsl_pool);
1402 }
1403 
1404 boolean_t
1405 spa_is_initializing(spa_t *spa)
1406 {
1407         return (spa->spa_is_initializing);
1408 }
1409 
1410 blkptr_t *
1411 spa_get_rootblkptr(spa_t *spa)
1412 {
1413         return (&spa->spa_ubsync.ub_rootbp);
1414 }
1415 
1416 void
1417 spa_set_rootblkptr(spa_t *spa, const blkptr_t *bp)
1418 {
1419         spa->spa_uberblock.ub_rootbp = *bp;
1420 }
1421 
1422 void
1423 spa_altroot(spa_t *spa, char *buf, size_t buflen)
1424 {
1425         if (spa->spa_root == NULL)
1426                 buf[0] = '\0';
1427         else
1428                 (void) strncpy(buf, spa->spa_root, buflen);
1429 }
1430 
1431 int
1432 spa_sync_pass(spa_t *spa)
1433 {
1434         return (spa->spa_sync_pass);
1435 }
1436 
1437 char *
1438 spa_name(spa_t *spa)
1439 {
1440         return (spa->spa_name);
1441 }
1442 
1443 uint64_t
1444 spa_guid(spa_t *spa)
1445 {
1446         dsl_pool_t *dp = spa_get_dsl(spa);
1447         uint64_t guid;
1448 
1449         /*
1450          * If we fail to parse the config during spa_load(), we can go through
1451          * the error path (which posts an ereport) and end up here with no root
1452          * vdev.  We stash the original pool guid in 'spa_config_guid' to handle
1453          * this case.
1454          */
1455         if (spa->spa_root_vdev == NULL)
1456                 return (spa->spa_config_guid);
1457 
1458         guid = spa->spa_last_synced_guid != 0 ?
1459             spa->spa_last_synced_guid : spa->spa_root_vdev->vdev_guid;
1460 
1461         /*
1462          * Return the most recently synced out guid unless we're
1463          * in syncing context.
1464          */
1465         if (dp && dsl_pool_sync_context(dp))
1466                 return (spa->spa_root_vdev->vdev_guid);
1467         else
1468                 return (guid);
1469 }
1470 
1471 uint64_t
1472 spa_load_guid(spa_t *spa)
1473 {
1474         /*
1475          * This is a GUID that exists solely as a reference for the
1476          * purposes of the arc.  It is generated at load time, and
1477          * is never written to persistent storage.
1478          */
1479         return (spa->spa_load_guid);
1480 }
1481 
1482 uint64_t
1483 spa_last_synced_txg(spa_t *spa)
1484 {
1485         return (spa->spa_ubsync.ub_txg);
1486 }
1487 
1488 uint64_t
1489 spa_first_txg(spa_t *spa)
1490 {
1491         return (spa->spa_first_txg);
1492 }
1493 
1494 uint64_t
1495 spa_syncing_txg(spa_t *spa)
1496 {
1497         return (spa->spa_syncing_txg);
1498 }
1499 
1500 pool_state_t
1501 spa_state(spa_t *spa)
1502 {
1503         return (spa->spa_state);
1504 }
1505 
1506 spa_load_state_t
1507 spa_load_state(spa_t *spa)
1508 {
1509         return (spa->spa_load_state);
1510 }
1511 
1512 uint64_t
1513 spa_freeze_txg(spa_t *spa)
1514 {
1515         return (spa->spa_freeze_txg);
1516 }
1517 
1518 /* ARGSUSED */
1519 uint64_t
1520 spa_get_asize(spa_t *spa, uint64_t lsize)
1521 {
1522         return (lsize * spa_asize_inflation);
1523 }
1524 
1525 uint64_t
1526 spa_get_dspace(spa_t *spa)
1527 {
1528         return (spa->spa_dspace);
1529 }
1530 
1531 void
1532 spa_update_dspace(spa_t *spa)
1533 {
1534         spa->spa_dspace = metaslab_class_get_dspace(spa_normal_class(spa)) +
1535             ddt_get_dedup_dspace(spa);
1536 }
1537 
1538 /*
1539  * Return the failure mode that has been set to this pool. The default
1540  * behavior will be to block all I/Os when a complete failure occurs.
1541  */
1542 uint8_t
1543 spa_get_failmode(spa_t *spa)
1544 {
1545         return (spa->spa_failmode);
1546 }
1547 
1548 boolean_t
1549 spa_suspended(spa_t *spa)
1550 {
1551         return (spa->spa_suspended);
1552 }
1553 
1554 uint64_t
1555 spa_version(spa_t *spa)
1556 {
1557         return (spa->spa_ubsync.ub_version);
1558 }
1559 
1560 boolean_t
1561 spa_deflate(spa_t *spa)
1562 {
1563         return (spa->spa_deflate);
1564 }
1565 
1566 metaslab_class_t *
1567 spa_normal_class(spa_t *spa)
1568 {
1569         return (spa->spa_normal_class);
1570 }
1571 
1572 metaslab_class_t *
1573 spa_log_class(spa_t *spa)
1574 {
1575         return (spa->spa_log_class);
1576 }
1577 
1578 int
1579 spa_max_replication(spa_t *spa)
1580 {
1581         /*
1582          * As of SPA_VERSION == SPA_VERSION_DITTO_BLOCKS, we are able to
1583          * handle BPs with more than one DVA allocated.  Set our max
1584          * replication level accordingly.
1585          */
1586         if (spa_version(spa) < SPA_VERSION_DITTO_BLOCKS)
1587                 return (1);
1588         return (MIN(SPA_DVAS_PER_BP, spa_max_replication_override));
1589 }
1590 
1591 int
1592 spa_prev_software_version(spa_t *spa)
1593 {
1594         return (spa->spa_prev_software_version);
1595 }
1596 
1597 uint64_t
1598 spa_deadman_synctime(spa_t *spa)
1599 {
1600         return (spa->spa_deadman_synctime);
1601 }
1602 
1603 uint64_t
1604 dva_get_dsize_sync(spa_t *spa, const dva_t *dva)
1605 {
1606         uint64_t asize = DVA_GET_ASIZE(dva);
1607         uint64_t dsize = asize;
1608 
1609         ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
1610 
1611         if (asize != 0 && spa->spa_deflate) {
1612                 vdev_t *vd = vdev_lookup_top(spa, DVA_GET_VDEV(dva));
1613                 dsize = (asize >> SPA_MINBLOCKSHIFT) * vd->vdev_deflate_ratio;
1614         }
1615 
1616         return (dsize);
1617 }
1618 
1619 uint64_t
1620 bp_get_dsize_sync(spa_t *spa, const blkptr_t *bp)
1621 {
1622         uint64_t dsize = 0;
1623 
1624         for (int d = 0; d < SPA_DVAS_PER_BP; d++)
1625                 dsize += dva_get_dsize_sync(spa, &bp->blk_dva[d]);
1626 
1627         return (dsize);
1628 }
1629 
1630 uint64_t
1631 bp_get_dsize(spa_t *spa, const blkptr_t *bp)
1632 {
1633         uint64_t dsize = 0;
1634 
1635         spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
1636 
1637         for (int d = 0; d < SPA_DVAS_PER_BP; d++)
1638                 dsize += dva_get_dsize_sync(spa, &bp->blk_dva[d]);
1639 
1640         spa_config_exit(spa, SCL_VDEV, FTAG);
1641 
1642         return (dsize);
1643 }
1644 
1645 /*
1646  * ==========================================================================
1647  * Initialization and Termination
1648  * ==========================================================================
1649  */
1650 
1651 static int
1652 spa_name_compare(const void *a1, const void *a2)
1653 {
1654         const spa_t *s1 = a1;
1655         const spa_t *s2 = a2;
1656         int s;
1657 
1658         s = strcmp(s1->spa_name, s2->spa_name);
1659         if (s > 0)
1660                 return (1);
1661         if (s < 0)
1662                 return (-1);
1663         return (0);
1664 }
1665 
1666 int
1667 spa_busy(void)
1668 {
1669         return (spa_active_count);
1670 }
1671 
1672 void
1673 spa_boot_init()
1674 {
1675         spa_config_load();
1676 }
1677 
1678 void
1679 spa_init(int mode)
1680 {
1681         mutex_init(&spa_namespace_lock, NULL, MUTEX_DEFAULT, NULL);
1682         mutex_init(&spa_spare_lock, NULL, MUTEX_DEFAULT, NULL);
1683         mutex_init(&spa_l2cache_lock, NULL, MUTEX_DEFAULT, NULL);
1684         cv_init(&spa_namespace_cv, NULL, CV_DEFAULT, NULL);
1685 
1686         avl_create(&spa_namespace_avl, spa_name_compare, sizeof (spa_t),
1687             offsetof(spa_t, spa_avl));
1688 
1689         avl_create(&spa_spare_avl, spa_spare_compare, sizeof (spa_aux_t),
1690             offsetof(spa_aux_t, aux_avl));
1691 
1692         avl_create(&spa_l2cache_avl, spa_l2cache_compare, sizeof (spa_aux_t),
1693             offsetof(spa_aux_t, aux_avl));
1694 
1695         spa_mode_global = mode;
1696 
1697 #ifdef _KERNEL
1698         spa_arch_init();
1699 #else
1700         if (spa_mode_global != FREAD && dprintf_find_string("watch")) {
1701                 arc_procfd = open("/proc/self/ctl", O_WRONLY);
1702                 if (arc_procfd == -1) {
1703                         perror("could not enable watchpoints: "
1704                             "opening /proc/self/ctl failed: ");
1705                 } else {
1706                         arc_watch = B_TRUE;
1707                 }
1708         }
1709 #endif
1710 
1711         refcount_init();
1712         unique_init();
1713         range_tree_init();
1714         zio_init();
1715         dmu_init();
1716         zil_init();
1717         vdev_cache_stat_init();
1718         zfs_prop_init();
1719         zpool_prop_init();
1720         zpool_feature_init();
1721         spa_config_load();
1722         l2arc_start();
1723 }
1724 
1725 void
1726 spa_fini(void)
1727 {
1728         l2arc_stop();
1729 
1730         spa_evict_all();
1731 
1732         vdev_cache_stat_fini();
1733         zil_fini();
1734         dmu_fini();
1735         zio_fini();
1736         range_tree_fini();
1737         unique_fini();
1738         refcount_fini();
1739 
1740         avl_destroy(&spa_namespace_avl);
1741         avl_destroy(&spa_spare_avl);
1742         avl_destroy(&spa_l2cache_avl);
1743 
1744         cv_destroy(&spa_namespace_cv);
1745         mutex_destroy(&spa_namespace_lock);
1746         mutex_destroy(&spa_spare_lock);
1747         mutex_destroy(&spa_l2cache_lock);
1748 }
1749 
1750 /*
1751  * Return whether this pool has slogs. No locking needed.
1752  * It's not a problem if the wrong answer is returned as it's only for
1753  * performance and not correctness
1754  */
1755 boolean_t
1756 spa_has_slogs(spa_t *spa)
1757 {
1758         return (spa->spa_log_class->mc_rotor != NULL);
1759 }
1760 
1761 spa_log_state_t
1762 spa_get_log_state(spa_t *spa)
1763 {
1764         return (spa->spa_log_state);
1765 }
1766 
1767 void
1768 spa_set_log_state(spa_t *spa, spa_log_state_t state)
1769 {
1770         spa->spa_log_state = state;
1771 }
1772 
1773 boolean_t
1774 spa_is_root(spa_t *spa)
1775 {
1776         return (spa->spa_is_root);
1777 }
1778 
1779 boolean_t
1780 spa_writeable(spa_t *spa)
1781 {
1782         return (!!(spa->spa_mode & FWRITE));
1783 }
1784 
1785 int
1786 spa_mode(spa_t *spa)
1787 {
1788         return (spa->spa_mode);
1789 }
1790 
1791 uint64_t
1792 spa_bootfs(spa_t *spa)
1793 {
1794         return (spa->spa_bootfs);
1795 }
1796 
1797 uint64_t
1798 spa_delegation(spa_t *spa)
1799 {
1800         return (spa->spa_delegation);
1801 }
1802 
1803 objset_t *
1804 spa_meta_objset(spa_t *spa)
1805 {
1806         return (spa->spa_meta_objset);
1807 }
1808 
1809 enum zio_checksum
1810 spa_dedup_checksum(spa_t *spa)
1811 {
1812         return (spa->spa_dedup_checksum);
1813 }
1814 
1815 /*
1816  * Reset pool scan stat per scan pass (or reboot).
1817  */
1818 void
1819 spa_scan_stat_init(spa_t *spa)
1820 {
1821         /* data not stored on disk */
1822         spa->spa_scan_pass_start = gethrestime_sec();
1823         spa->spa_scan_pass_exam = 0;
1824         vdev_scan_stat_init(spa->spa_root_vdev);
1825 }
1826 
1827 /*
1828  * Get scan stats for zpool status reports
1829  */
1830 int
1831 spa_scan_get_stats(spa_t *spa, pool_scan_stat_t *ps)
1832 {
1833         dsl_scan_t *scn = spa->spa_dsl_pool ? spa->spa_dsl_pool->dp_scan : NULL;
1834 
1835         if (scn == NULL || scn->scn_phys.scn_func == POOL_SCAN_NONE)
1836                 return (SET_ERROR(ENOENT));
1837         bzero(ps, sizeof (pool_scan_stat_t));
1838 
1839         /* data stored on disk */
1840         ps->pss_func = scn->scn_phys.scn_func;
1841         ps->pss_start_time = scn->scn_phys.scn_start_time;
1842         ps->pss_end_time = scn->scn_phys.scn_end_time;
1843         ps->pss_to_examine = scn->scn_phys.scn_to_examine;
1844         ps->pss_examined = scn->scn_phys.scn_examined;
1845         ps->pss_to_process = scn->scn_phys.scn_to_process;
1846         ps->pss_processed = scn->scn_phys.scn_processed;
1847         ps->pss_errors = scn->scn_phys.scn_errors;
1848         ps->pss_state = scn->scn_phys.scn_state;
1849 
1850         /* data not stored on disk */
1851         ps->pss_pass_start = spa->spa_scan_pass_start;
1852         ps->pss_pass_exam = spa->spa_scan_pass_exam;
1853 
1854         return (0);
1855 }
1856 
1857 boolean_t
1858 spa_debug_enabled(spa_t *spa)
1859 {
1860         return (spa->spa_debug);
1861 }