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