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) 2012 by Delphix. All rights reserved.
  24  * Copyright (c) 2012, Joyent, Inc. All rights reserved.
  25  */
  26 
  27 /* Portions Copyright 2010 Robert Milkowski */
  28 
  29 #include <sys/cred.h>
  30 #include <sys/zfs_context.h>
  31 #include <sys/dmu_objset.h>
  32 #include <sys/dsl_dir.h>
  33 #include <sys/dsl_dataset.h>
  34 #include <sys/dsl_prop.h>
  35 #include <sys/dsl_pool.h>
  36 #include <sys/dsl_synctask.h>
  37 #include <sys/dsl_deleg.h>
  38 #include <sys/dnode.h>
  39 #include <sys/dbuf.h>
  40 #include <sys/zvol.h>
  41 #include <sys/dmu_tx.h>
  42 #include <sys/zap.h>
  43 #include <sys/zil.h>
  44 #include <sys/dmu_impl.h>
  45 #include <sys/zfs_ioctl.h>
  46 #include <sys/sa.h>
  47 #include <sys/zfs_onexit.h>
  48 
  49 /*
  50  * Needed to close a window in dnode_move() that allows the objset to be freed
  51  * before it can be safely accessed.
  52  */
  53 krwlock_t os_lock;
  54 
  55 void
  56 dmu_objset_init(void)
  57 {
  58         rw_init(&os_lock, NULL, RW_DEFAULT, NULL);
  59 }
  60 
  61 void
  62 dmu_objset_fini(void)
  63 {
  64         rw_destroy(&os_lock);
  65 }
  66 
  67 spa_t *
  68 dmu_objset_spa(objset_t *os)
  69 {
  70         return (os->os_spa);
  71 }
  72 
  73 zilog_t *
  74 dmu_objset_zil(objset_t *os)
  75 {
  76         return (os->os_zil);
  77 }
  78 
  79 dsl_pool_t *
  80 dmu_objset_pool(objset_t *os)
  81 {
  82         dsl_dataset_t *ds;
  83 
  84         if ((ds = os->os_dsl_dataset) != NULL && ds->ds_dir)
  85                 return (ds->ds_dir->dd_pool);
  86         else
  87                 return (spa_get_dsl(os->os_spa));
  88 }
  89 
  90 dsl_dataset_t *
  91 dmu_objset_ds(objset_t *os)
  92 {
  93         return (os->os_dsl_dataset);
  94 }
  95 
  96 dmu_objset_type_t
  97 dmu_objset_type(objset_t *os)
  98 {
  99         return (os->os_phys->os_type);
 100 }
 101 
 102 void
 103 dmu_objset_name(objset_t *os, char *buf)
 104 {
 105         dsl_dataset_name(os->os_dsl_dataset, buf);
 106 }
 107 
 108 uint64_t
 109 dmu_objset_id(objset_t *os)
 110 {
 111         dsl_dataset_t *ds = os->os_dsl_dataset;
 112 
 113         return (ds ? ds->ds_object : 0);
 114 }
 115 
 116 uint64_t
 117 dmu_objset_syncprop(objset_t *os)
 118 {
 119         return (os->os_sync);
 120 }
 121 
 122 uint64_t
 123 dmu_objset_logbias(objset_t *os)
 124 {
 125         return (os->os_logbias);
 126 }
 127 
 128 static void
 129 checksum_changed_cb(void *arg, uint64_t newval)
 130 {
 131         objset_t *os = arg;
 132 
 133         /*
 134          * Inheritance should have been done by now.
 135          */
 136         ASSERT(newval != ZIO_CHECKSUM_INHERIT);
 137 
 138         os->os_checksum = zio_checksum_select(newval, ZIO_CHECKSUM_ON_VALUE);
 139 }
 140 
 141 static void
 142 compression_changed_cb(void *arg, uint64_t newval)
 143 {
 144         objset_t *os = arg;
 145 
 146         /*
 147          * Inheritance and range checking should have been done by now.
 148          */
 149         ASSERT(newval != ZIO_COMPRESS_INHERIT);
 150 
 151         os->os_compress = zio_compress_select(newval, ZIO_COMPRESS_ON_VALUE);
 152 }
 153 
 154 static void
 155 copies_changed_cb(void *arg, uint64_t newval)
 156 {
 157         objset_t *os = arg;
 158 
 159         /*
 160          * Inheritance and range checking should have been done by now.
 161          */
 162         ASSERT(newval > 0);
 163         ASSERT(newval <= spa_max_replication(os->os_spa));
 164 
 165         os->os_copies = newval;
 166 }
 167 
 168 static void
 169 dedup_changed_cb(void *arg, uint64_t newval)
 170 {
 171         objset_t *os = arg;
 172         spa_t *spa = os->os_spa;
 173         enum zio_checksum checksum;
 174 
 175         /*
 176          * Inheritance should have been done by now.
 177          */
 178         ASSERT(newval != ZIO_CHECKSUM_INHERIT);
 179 
 180         checksum = zio_checksum_dedup_select(spa, newval, ZIO_CHECKSUM_OFF);
 181 
 182         os->os_dedup_checksum = checksum & ZIO_CHECKSUM_MASK;
 183         os->os_dedup_verify = !!(checksum & ZIO_CHECKSUM_VERIFY);
 184 }
 185 
 186 static void
 187 primary_cache_changed_cb(void *arg, uint64_t newval)
 188 {
 189         objset_t *os = arg;
 190 
 191         /*
 192          * Inheritance and range checking should have been done by now.
 193          */
 194         ASSERT(newval == ZFS_CACHE_ALL || newval == ZFS_CACHE_NONE ||
 195             newval == ZFS_CACHE_METADATA);
 196 
 197         os->os_primary_cache = newval;
 198 }
 199 
 200 static void
 201 secondary_cache_changed_cb(void *arg, uint64_t newval)
 202 {
 203         objset_t *os = arg;
 204 
 205         /*
 206          * Inheritance and range checking should have been done by now.
 207          */
 208         ASSERT(newval == ZFS_CACHE_ALL || newval == ZFS_CACHE_NONE ||
 209             newval == ZFS_CACHE_METADATA);
 210 
 211         os->os_secondary_cache = newval;
 212 }
 213 
 214 static void
 215 sync_changed_cb(void *arg, uint64_t newval)
 216 {
 217         objset_t *os = arg;
 218 
 219         /*
 220          * Inheritance and range checking should have been done by now.
 221          */
 222         ASSERT(newval == ZFS_SYNC_STANDARD || newval == ZFS_SYNC_ALWAYS ||
 223             newval == ZFS_SYNC_DISABLED);
 224 
 225         os->os_sync = newval;
 226         if (os->os_zil)
 227                 zil_set_sync(os->os_zil, newval);
 228 }
 229 
 230 static void
 231 logbias_changed_cb(void *arg, uint64_t newval)
 232 {
 233         objset_t *os = arg;
 234 
 235         ASSERT(newval == ZFS_LOGBIAS_LATENCY ||
 236             newval == ZFS_LOGBIAS_THROUGHPUT);
 237         os->os_logbias = newval;
 238         if (os->os_zil)
 239                 zil_set_logbias(os->os_zil, newval);
 240 }
 241 
 242 void
 243 dmu_objset_byteswap(void *buf, size_t size)
 244 {
 245         objset_phys_t *osp = buf;
 246 
 247         ASSERT(size == OBJSET_OLD_PHYS_SIZE || size == sizeof (objset_phys_t));
 248         dnode_byteswap(&osp->os_meta_dnode);
 249         byteswap_uint64_array(&osp->os_zil_header, sizeof (zil_header_t));
 250         osp->os_type = BSWAP_64(osp->os_type);
 251         osp->os_flags = BSWAP_64(osp->os_flags);
 252         if (size == sizeof (objset_phys_t)) {
 253                 dnode_byteswap(&osp->os_userused_dnode);
 254                 dnode_byteswap(&osp->os_groupused_dnode);
 255         }
 256 }
 257 
 258 int
 259 dmu_objset_open_impl(spa_t *spa, dsl_dataset_t *ds, blkptr_t *bp,
 260     objset_t **osp)
 261 {
 262         objset_t *os;
 263         int i, err;
 264 
 265         ASSERT(ds == NULL || MUTEX_HELD(&ds->ds_opening_lock));
 266 
 267         os = kmem_zalloc(sizeof (objset_t), KM_SLEEP);
 268         os->os_dsl_dataset = ds;
 269         os->os_spa = spa;
 270         os->os_rootbp = bp;
 271         if (!BP_IS_HOLE(os->os_rootbp)) {
 272                 uint32_t aflags = ARC_WAIT;
 273                 zbookmark_t zb;
 274                 SET_BOOKMARK(&zb, ds ? ds->ds_object : DMU_META_OBJSET,
 275                     ZB_ROOT_OBJECT, ZB_ROOT_LEVEL, ZB_ROOT_BLKID);
 276 
 277                 if (DMU_OS_IS_L2CACHEABLE(os))
 278                         aflags |= ARC_L2CACHE;
 279 
 280                 dprintf_bp(os->os_rootbp, "reading %s", "");
 281                 /*
 282                  * XXX when bprewrite scrub can change the bp,
 283                  * and this is called from dmu_objset_open_ds_os, the bp
 284                  * could change, and we'll need a lock.
 285                  */
 286                 err = dsl_read_nolock(NULL, spa, os->os_rootbp,
 287                     arc_getbuf_func, &os->os_phys_buf,
 288                     ZIO_PRIORITY_SYNC_READ, ZIO_FLAG_CANFAIL, &aflags, &zb);
 289                 if (err) {
 290                         kmem_free(os, sizeof (objset_t));
 291                         /* convert checksum errors into IO errors */
 292                         if (err == ECKSUM)
 293                                 err = EIO;
 294                         return (err);
 295                 }
 296 
 297                 /* Increase the blocksize if we are permitted. */
 298                 if (spa_version(spa) >= SPA_VERSION_USERSPACE &&
 299                     arc_buf_size(os->os_phys_buf) < sizeof (objset_phys_t)) {
 300                         arc_buf_t *buf = arc_buf_alloc(spa,
 301                             sizeof (objset_phys_t), &os->os_phys_buf,
 302                             ARC_BUFC_METADATA);
 303                         bzero(buf->b_data, sizeof (objset_phys_t));
 304                         bcopy(os->os_phys_buf->b_data, buf->b_data,
 305                             arc_buf_size(os->os_phys_buf));
 306                         (void) arc_buf_remove_ref(os->os_phys_buf,
 307                             &os->os_phys_buf);
 308                         os->os_phys_buf = buf;
 309                 }
 310 
 311                 os->os_phys = os->os_phys_buf->b_data;
 312                 os->os_flags = os->os_phys->os_flags;
 313         } else {
 314                 int size = spa_version(spa) >= SPA_VERSION_USERSPACE ?
 315                     sizeof (objset_phys_t) : OBJSET_OLD_PHYS_SIZE;
 316                 os->os_phys_buf = arc_buf_alloc(spa, size,
 317                     &os->os_phys_buf, ARC_BUFC_METADATA);
 318                 os->os_phys = os->os_phys_buf->b_data;
 319                 bzero(os->os_phys, size);
 320         }
 321 
 322         /*
 323          * Note: the changed_cb will be called once before the register
 324          * func returns, thus changing the checksum/compression from the
 325          * default (fletcher2/off).  Snapshots don't need to know about
 326          * checksum/compression/copies.
 327          */
 328         if (ds) {
 329                 err = dsl_prop_register(ds, "primarycache",
 330                     primary_cache_changed_cb, os);
 331                 if (err == 0)
 332                         err = dsl_prop_register(ds, "secondarycache",
 333                             secondary_cache_changed_cb, os);
 334                 if (!dsl_dataset_is_snapshot(ds)) {
 335                         if (err == 0)
 336                                 err = dsl_prop_register(ds, "checksum",
 337                                     checksum_changed_cb, os);
 338                         if (err == 0)
 339                                 err = dsl_prop_register(ds, "compression",
 340                                     compression_changed_cb, os);
 341                         if (err == 0)
 342                                 err = dsl_prop_register(ds, "copies",
 343                                     copies_changed_cb, os);
 344                         if (err == 0)
 345                                 err = dsl_prop_register(ds, "dedup",
 346                                     dedup_changed_cb, os);
 347                         if (err == 0)
 348                                 err = dsl_prop_register(ds, "logbias",
 349                                     logbias_changed_cb, os);
 350                         if (err == 0)
 351                                 err = dsl_prop_register(ds, "sync",
 352                                     sync_changed_cb, os);
 353                 }
 354                 if (err) {
 355                         VERIFY(arc_buf_remove_ref(os->os_phys_buf,
 356                             &os->os_phys_buf) == 1);
 357                         kmem_free(os, sizeof (objset_t));
 358                         return (err);
 359                 }
 360         } else if (ds == NULL) {
 361                 /* It's the meta-objset. */
 362                 os->os_checksum = ZIO_CHECKSUM_FLETCHER_4;
 363                 os->os_compress = ZIO_COMPRESS_LZJB;
 364                 os->os_copies = spa_max_replication(spa);
 365                 os->os_dedup_checksum = ZIO_CHECKSUM_OFF;
 366                 os->os_dedup_verify = 0;
 367                 os->os_logbias = 0;
 368                 os->os_sync = 0;
 369                 os->os_primary_cache = ZFS_CACHE_ALL;
 370                 os->os_secondary_cache = ZFS_CACHE_ALL;
 371         }
 372 
 373         if (ds == NULL || !dsl_dataset_is_snapshot(ds))
 374                 os->os_zil_header = os->os_phys->os_zil_header;
 375         os->os_zil = zil_alloc(os, &os->os_zil_header);
 376 
 377         for (i = 0; i < TXG_SIZE; i++) {
 378                 list_create(&os->os_dirty_dnodes[i], sizeof (dnode_t),
 379                     offsetof(dnode_t, dn_dirty_link[i]));
 380                 list_create(&os->os_free_dnodes[i], sizeof (dnode_t),
 381                     offsetof(dnode_t, dn_dirty_link[i]));
 382         }
 383         list_create(&os->os_dnodes, sizeof (dnode_t),
 384             offsetof(dnode_t, dn_link));
 385         list_create(&os->os_downgraded_dbufs, sizeof (dmu_buf_impl_t),
 386             offsetof(dmu_buf_impl_t, db_link));
 387 
 388         mutex_init(&os->os_lock, NULL, MUTEX_DEFAULT, NULL);
 389         mutex_init(&os->os_obj_lock, NULL, MUTEX_DEFAULT, NULL);
 390         mutex_init(&os->os_user_ptr_lock, NULL, MUTEX_DEFAULT, NULL);
 391 
 392         DMU_META_DNODE(os) = dnode_special_open(os,
 393             &os->os_phys->os_meta_dnode, DMU_META_DNODE_OBJECT,
 394             &os->os_meta_dnode);
 395         if (arc_buf_size(os->os_phys_buf) >= sizeof (objset_phys_t)) {
 396                 DMU_USERUSED_DNODE(os) = dnode_special_open(os,
 397                     &os->os_phys->os_userused_dnode, DMU_USERUSED_OBJECT,
 398                     &os->os_userused_dnode);
 399                 DMU_GROUPUSED_DNODE(os) = dnode_special_open(os,
 400                     &os->os_phys->os_groupused_dnode, DMU_GROUPUSED_OBJECT,
 401                     &os->os_groupused_dnode);
 402         }
 403 
 404         /*
 405          * We should be the only thread trying to do this because we
 406          * have ds_opening_lock
 407          */
 408         if (ds) {
 409                 mutex_enter(&ds->ds_lock);
 410                 ASSERT(ds->ds_objset == NULL);
 411                 ds->ds_objset = os;
 412                 mutex_exit(&ds->ds_lock);
 413         }
 414 
 415         *osp = os;
 416         return (0);
 417 }
 418 
 419 int
 420 dmu_objset_from_ds(dsl_dataset_t *ds, objset_t **osp)
 421 {
 422         int err = 0;
 423 
 424         mutex_enter(&ds->ds_opening_lock);
 425         *osp = ds->ds_objset;
 426         if (*osp == NULL) {
 427                 err = dmu_objset_open_impl(dsl_dataset_get_spa(ds),
 428                     ds, dsl_dataset_get_blkptr(ds), osp);
 429         }
 430         mutex_exit(&ds->ds_opening_lock);
 431         return (err);
 432 }
 433 
 434 /* called from zpl */
 435 int
 436 dmu_objset_hold(const char *name, void *tag, objset_t **osp)
 437 {
 438         dsl_dataset_t *ds;
 439         int err;
 440 
 441         err = dsl_dataset_hold(name, tag, &ds);
 442         if (err)
 443                 return (err);
 444 
 445         err = dmu_objset_from_ds(ds, osp);
 446         if (err)
 447                 dsl_dataset_rele(ds, tag);
 448 
 449         return (err);
 450 }
 451 
 452 /* called from zpl */
 453 int
 454 dmu_objset_own(const char *name, dmu_objset_type_t type,
 455     boolean_t readonly, void *tag, objset_t **osp)
 456 {
 457         dsl_dataset_t *ds;
 458         int err;
 459 
 460         err = dsl_dataset_own(name, B_FALSE, tag, &ds);
 461         if (err)
 462                 return (err);
 463 
 464         err = dmu_objset_from_ds(ds, osp);
 465         if (err) {
 466                 dsl_dataset_disown(ds, tag);
 467         } else if (type != DMU_OST_ANY && type != (*osp)->os_phys->os_type) {
 468                 dmu_objset_disown(*osp, tag);
 469                 return (EINVAL);
 470         } else if (!readonly && dsl_dataset_is_snapshot(ds)) {
 471                 dmu_objset_disown(*osp, tag);
 472                 return (EROFS);
 473         }
 474         return (err);
 475 }
 476 
 477 void
 478 dmu_objset_rele(objset_t *os, void *tag)
 479 {
 480         dsl_dataset_rele(os->os_dsl_dataset, tag);
 481 }
 482 
 483 void
 484 dmu_objset_disown(objset_t *os, void *tag)
 485 {
 486         dsl_dataset_disown(os->os_dsl_dataset, tag);
 487 }
 488 
 489 int
 490 dmu_objset_evict_dbufs(objset_t *os)
 491 {
 492         dnode_t *dn;
 493 
 494         mutex_enter(&os->os_lock);
 495 
 496         /* process the mdn last, since the other dnodes have holds on it */
 497         list_remove(&os->os_dnodes, DMU_META_DNODE(os));
 498         list_insert_tail(&os->os_dnodes, DMU_META_DNODE(os));
 499 
 500         /*
 501          * Find the first dnode with holds.  We have to do this dance
 502          * because dnode_add_ref() only works if you already have a
 503          * hold.  If there are no holds then it has no dbufs so OK to
 504          * skip.
 505          */
 506         for (dn = list_head(&os->os_dnodes);
 507             dn && !dnode_add_ref(dn, FTAG);
 508             dn = list_next(&os->os_dnodes, dn))
 509                 continue;
 510 
 511         while (dn) {
 512                 dnode_t *next_dn = dn;
 513 
 514                 do {
 515                         next_dn = list_next(&os->os_dnodes, next_dn);
 516                 } while (next_dn && !dnode_add_ref(next_dn, FTAG));
 517 
 518                 mutex_exit(&os->os_lock);
 519                 dnode_evict_dbufs(dn);
 520                 dnode_rele(dn, FTAG);
 521                 mutex_enter(&os->os_lock);
 522                 dn = next_dn;
 523         }
 524         dn = list_head(&os->os_dnodes);
 525         mutex_exit(&os->os_lock);
 526         return (dn != DMU_META_DNODE(os));
 527 }
 528 
 529 void
 530 dmu_objset_evict(objset_t *os)
 531 {
 532         dsl_dataset_t *ds = os->os_dsl_dataset;
 533 
 534         for (int t = 0; t < TXG_SIZE; t++)
 535                 ASSERT(!dmu_objset_is_dirty(os, t));
 536 
 537         if (ds) {
 538                 if (!dsl_dataset_is_snapshot(ds)) {
 539                         VERIFY(0 == dsl_prop_unregister(ds, "checksum",
 540                             checksum_changed_cb, os));
 541                         VERIFY(0 == dsl_prop_unregister(ds, "compression",
 542                             compression_changed_cb, os));
 543                         VERIFY(0 == dsl_prop_unregister(ds, "copies",
 544                             copies_changed_cb, os));
 545                         VERIFY(0 == dsl_prop_unregister(ds, "dedup",
 546                             dedup_changed_cb, os));
 547                         VERIFY(0 == dsl_prop_unregister(ds, "logbias",
 548                             logbias_changed_cb, os));
 549                         VERIFY(0 == dsl_prop_unregister(ds, "sync",
 550                             sync_changed_cb, os));
 551                 }
 552                 VERIFY(0 == dsl_prop_unregister(ds, "primarycache",
 553                     primary_cache_changed_cb, os));
 554                 VERIFY(0 == dsl_prop_unregister(ds, "secondarycache",
 555                     secondary_cache_changed_cb, os));
 556         }
 557 
 558         if (os->os_sa)
 559                 sa_tear_down(os);
 560 
 561         /*
 562          * We should need only a single pass over the dnode list, since
 563          * nothing can be added to the list at this point.
 564          */
 565         (void) dmu_objset_evict_dbufs(os);
 566 
 567         dnode_special_close(&os->os_meta_dnode);
 568         if (DMU_USERUSED_DNODE(os)) {
 569                 dnode_special_close(&os->os_userused_dnode);
 570                 dnode_special_close(&os->os_groupused_dnode);
 571         }
 572         zil_free(os->os_zil);
 573 
 574         ASSERT3P(list_head(&os->os_dnodes), ==, NULL);
 575 
 576         VERIFY(arc_buf_remove_ref(os->os_phys_buf, &os->os_phys_buf) == 1);
 577 
 578         /*
 579          * This is a barrier to prevent the objset from going away in
 580          * dnode_move() until we can safely ensure that the objset is still in
 581          * use. We consider the objset valid before the barrier and invalid
 582          * after the barrier.
 583          */
 584         rw_enter(&os_lock, RW_READER);
 585         rw_exit(&os_lock);
 586 
 587         mutex_destroy(&os->os_lock);
 588         mutex_destroy(&os->os_obj_lock);
 589         mutex_destroy(&os->os_user_ptr_lock);
 590         kmem_free(os, sizeof (objset_t));
 591 }
 592 
 593 timestruc_t
 594 dmu_objset_snap_cmtime(objset_t *os)
 595 {
 596         return (dsl_dir_snap_cmtime(os->os_dsl_dataset->ds_dir));
 597 }
 598 
 599 /* called from dsl for meta-objset */
 600 objset_t *
 601 dmu_objset_create_impl(spa_t *spa, dsl_dataset_t *ds, blkptr_t *bp,
 602     dmu_objset_type_t type, dmu_tx_t *tx)
 603 {
 604         objset_t *os;
 605         dnode_t *mdn;
 606 
 607         ASSERT(dmu_tx_is_syncing(tx));
 608         if (ds != NULL)
 609                 VERIFY(0 == dmu_objset_from_ds(ds, &os));
 610         else
 611                 VERIFY(0 == dmu_objset_open_impl(spa, NULL, bp, &os));
 612 
 613         mdn = DMU_META_DNODE(os);
 614 
 615         dnode_allocate(mdn, DMU_OT_DNODE, 1 << DNODE_BLOCK_SHIFT,
 616             DN_MAX_INDBLKSHIFT, DMU_OT_NONE, 0, tx);
 617 
 618         /*
 619          * We don't want to have to increase the meta-dnode's nlevels
 620          * later, because then we could do it in quescing context while
 621          * we are also accessing it in open context.
 622          *
 623          * This precaution is not necessary for the MOS (ds == NULL),
 624          * because the MOS is only updated in syncing context.
 625          * This is most fortunate: the MOS is the only objset that
 626          * needs to be synced multiple times as spa_sync() iterates
 627          * to convergence, so minimizing its dn_nlevels matters.
 628          */
 629         if (ds != NULL) {
 630                 int levels = 1;
 631 
 632                 /*
 633                  * Determine the number of levels necessary for the meta-dnode
 634                  * to contain DN_MAX_OBJECT dnodes.
 635                  */
 636                 while ((uint64_t)mdn->dn_nblkptr << (mdn->dn_datablkshift +
 637                     (levels - 1) * (mdn->dn_indblkshift - SPA_BLKPTRSHIFT)) <
 638                     DN_MAX_OBJECT * sizeof (dnode_phys_t))
 639                         levels++;
 640 
 641                 mdn->dn_next_nlevels[tx->tx_txg & TXG_MASK] =
 642                     mdn->dn_nlevels = levels;
 643         }
 644 
 645         ASSERT(type != DMU_OST_NONE);
 646         ASSERT(type != DMU_OST_ANY);
 647         ASSERT(type < DMU_OST_NUMTYPES);
 648         os->os_phys->os_type = type;
 649         if (dmu_objset_userused_enabled(os)) {
 650                 os->os_phys->os_flags |= OBJSET_FLAG_USERACCOUNTING_COMPLETE;
 651                 os->os_flags = os->os_phys->os_flags;
 652         }
 653 
 654         dsl_dataset_dirty(ds, tx);
 655 
 656         return (os);
 657 }
 658 
 659 struct oscarg {
 660         void (*userfunc)(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx);
 661         void *userarg;
 662         dsl_dataset_t *clone_origin;
 663         const char *lastname;
 664         dmu_objset_type_t type;
 665         uint64_t flags;
 666         cred_t *cr;
 667 };
 668 
 669 /*ARGSUSED*/
 670 static int
 671 dmu_objset_create_check(void *arg1, void *arg2, dmu_tx_t *tx)
 672 {
 673         dsl_dir_t *dd = arg1;
 674         struct oscarg *oa = arg2;
 675         objset_t *mos = dd->dd_pool->dp_meta_objset;
 676         int err;
 677         uint64_t ddobj;
 678 
 679         err = zap_lookup(mos, dd->dd_phys->dd_child_dir_zapobj,
 680             oa->lastname, sizeof (uint64_t), 1, &ddobj);
 681         if (err != ENOENT)
 682                 return (err ? err : EEXIST);
 683 
 684         if (oa->clone_origin != NULL) {
 685                 /* You can't clone across pools. */
 686                 if (oa->clone_origin->ds_dir->dd_pool != dd->dd_pool)
 687                         return (EXDEV);
 688 
 689                 /* You can only clone snapshots, not the head datasets. */
 690                 if (!dsl_dataset_is_snapshot(oa->clone_origin))
 691                         return (EINVAL);
 692         }
 693 
 694         return (dsl_dir_fscount_check(dd, 1, NULL));
 695 }
 696 
 697 static void
 698 dmu_objset_create_sync(void *arg1, void *arg2, dmu_tx_t *tx)
 699 {
 700         dsl_dir_t *dd = arg1;
 701         spa_t *spa = dd->dd_pool->dp_spa;
 702         struct oscarg *oa = arg2;
 703         uint64_t obj;
 704         dsl_dataset_t *ds;
 705         blkptr_t *bp;
 706 
 707         ASSERT(dmu_tx_is_syncing(tx));
 708 
 709         dsl_dir_fscount_adjust(dd, tx, 1, B_TRUE, B_TRUE);
 710 
 711         obj = dsl_dataset_create_sync(dd, oa->lastname,
 712             oa->clone_origin, oa->flags, oa->cr, tx);
 713 
 714         VERIFY3U(0, ==, dsl_dataset_hold_obj(dd->dd_pool, obj, FTAG, &ds));
 715         bp = dsl_dataset_get_blkptr(ds);
 716         if (BP_IS_HOLE(bp)) {
 717                 objset_t *os =
 718                     dmu_objset_create_impl(spa, ds, bp, oa->type, tx);
 719 
 720                 if (oa->userfunc)
 721                         oa->userfunc(os, oa->userarg, oa->cr, tx);
 722         }
 723 
 724         if (oa->clone_origin == NULL) {
 725                 spa_history_log_internal_ds(ds, "create", tx, "");
 726         } else {
 727                 char namebuf[MAXNAMELEN];
 728                 dsl_dataset_name(oa->clone_origin, namebuf);
 729                 spa_history_log_internal_ds(ds, "clone", tx,
 730                     "origin=%s (%llu)", namebuf, oa->clone_origin->ds_object);
 731         }
 732         dsl_dataset_rele(ds, FTAG);
 733 }
 734 
 735 int
 736 dmu_objset_create(const char *name, dmu_objset_type_t type, uint64_t flags,
 737     void (*func)(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx), void *arg)
 738 {
 739         dsl_dir_t *pdd;
 740         const char *tail;
 741         int err = 0;
 742         struct oscarg oa = { 0 };
 743 
 744         ASSERT(strchr(name, '@') == NULL);
 745         err = dsl_dir_open(name, FTAG, &pdd, &tail);
 746         if (err)
 747                 return (err);
 748         if (tail == NULL) {
 749                 dsl_dir_close(pdd, FTAG);
 750                 return (EEXIST);
 751         }
 752 
 753         oa.userfunc = func;
 754         oa.userarg = arg;
 755         oa.lastname = tail;
 756         oa.type = type;
 757         oa.flags = flags;
 758         oa.cr = CRED();
 759 
 760         err = dsl_sync_task_do(pdd->dd_pool, dmu_objset_create_check,
 761             dmu_objset_create_sync, pdd, &oa, 5);
 762         dsl_dir_close(pdd, FTAG);
 763         return (err);
 764 }
 765 
 766 int
 767 dmu_objset_clone(const char *name, dsl_dataset_t *clone_origin, uint64_t flags)
 768 {
 769         dsl_dir_t *pdd;
 770         const char *tail;
 771         int err = 0;
 772         struct oscarg oa = { 0 };
 773 
 774         ASSERT(strchr(name, '@') == NULL);
 775         err = dsl_dir_open(name, FTAG, &pdd, &tail);
 776         if (err)
 777                 return (err);
 778         if (tail == NULL) {
 779                 dsl_dir_close(pdd, FTAG);
 780                 return (EEXIST);
 781         }
 782 
 783         oa.lastname = tail;
 784         oa.clone_origin = clone_origin;
 785         oa.flags = flags;
 786         oa.cr = CRED();
 787 
 788         err = dsl_sync_task_do(pdd->dd_pool, dmu_objset_create_check,
 789             dmu_objset_create_sync, pdd, &oa, 5);
 790         dsl_dir_close(pdd, FTAG);
 791         return (err);
 792 }
 793 
 794 int
 795 dmu_objset_destroy(const char *name, boolean_t defer)
 796 {
 797         dsl_dataset_t *ds;
 798         int error;
 799 
 800         error = dsl_dataset_own(name, B_TRUE, FTAG, &ds);
 801         if (error == 0) {
 802                 error = dsl_dataset_destroy(ds, FTAG, defer);
 803                 /* dsl_dataset_destroy() closes the ds. */
 804         }
 805 
 806         return (error);
 807 }
 808 
 809 typedef struct snapallarg {
 810         dsl_sync_task_group_t *saa_dstg;
 811         boolean_t saa_needsuspend;
 812         nvlist_t *saa_props;
 813 
 814         /* the following are used only if 'temporary' is set: */
 815         boolean_t saa_temporary;
 816         const char *saa_htag;
 817         struct dsl_ds_holdarg *saa_ha;
 818         dsl_dataset_t *saa_newds;
 819 } snapallarg_t;
 820 
 821 typedef struct snaponearg {
 822         const char *soa_longname; /* long snap name */
 823         const char *soa_snapname; /* short snap name */
 824         uint64_t soa_tot_cnt;
 825         snapallarg_t *soa_saa;
 826 } snaponearg_t;
 827 
 828 static int
 829 snapshot_check(void *arg1, void *arg2, dmu_tx_t *tx)
 830 {
 831         objset_t *os = arg1;
 832         snaponearg_t *soa = arg2;
 833         snapallarg_t *saa = soa->soa_saa;
 834         int error;
 835 
 836         /* The props have already been checked by zfs_check_userprops(). */
 837 
 838         error = dsl_dataset_snapshot_check(os->os_dsl_dataset,
 839             soa->soa_snapname, soa->soa_tot_cnt, tx);
 840         if (error)
 841                 return (error);
 842 
 843         if (saa->saa_temporary) {
 844                 /*
 845                  * Ideally we would just call
 846                  * dsl_dataset_user_hold_check() and
 847                  * dsl_dataset_destroy_check() here.  However the
 848                  * dataset we want to hold and destroy is the snapshot
 849                  * that we just confirmed we can create, but it won't
 850                  * exist until after these checks are run.  Do any
 851                  * checks we can here and if more checks are added to
 852                  * those routines in the future, similar checks may be
 853                  * necessary here.
 854                  */
 855                 if (spa_version(os->os_spa) < SPA_VERSION_USERREFS)
 856                         return (ENOTSUP);
 857                 /*
 858                  * Not checking number of tags because the tag will be
 859                  * unique, as it will be the only tag.
 860                  */
 861                 if (strlen(saa->saa_htag) + MAX_TAG_PREFIX_LEN >= MAXNAMELEN)
 862                         return (E2BIG);
 863 
 864                 saa->saa_ha = kmem_alloc(sizeof (struct dsl_ds_holdarg),
 865                     KM_SLEEP);
 866                 saa->saa_ha->temphold = B_TRUE;
 867                 saa->saa_ha->htag = saa->saa_htag;
 868         }
 869         return (error);
 870 }
 871 
 872 static void
 873 snapshot_sync(void *arg1, void *arg2, dmu_tx_t *tx)
 874 {
 875         objset_t *os = arg1;
 876         dsl_dataset_t *ds = os->os_dsl_dataset;
 877         snaponearg_t *soa = arg2;
 878         snapallarg_t *saa = soa->soa_saa;
 879 
 880         dsl_dataset_snapshot_sync(ds, soa->soa_snapname, tx);
 881 
 882         if (saa->saa_props != NULL) {
 883                 dsl_props_arg_t pa;
 884                 pa.pa_props = saa->saa_props;
 885                 pa.pa_source = ZPROP_SRC_LOCAL;
 886                 dsl_props_set_sync(ds->ds_prev, &pa, tx);
 887         }
 888 
 889         if (saa->saa_temporary) {
 890                 struct dsl_ds_destroyarg da;
 891 
 892                 dsl_dataset_user_hold_sync(ds->ds_prev, saa->saa_ha, tx);
 893                 kmem_free(saa->saa_ha, sizeof (struct dsl_ds_holdarg));
 894                 saa->saa_ha = NULL;
 895                 saa->saa_newds = ds->ds_prev;
 896 
 897                 da.ds = ds->ds_prev;
 898                 da.defer = B_TRUE;
 899                 dsl_dataset_destroy_sync(&da, FTAG, tx);
 900         }
 901 }
 902 
 903 static int
 904 snapshot_one_impl(const char *snapname, void *arg, uint64_t cnt)
 905 {
 906         char fsname[MAXPATHLEN];
 907         snapallarg_t *saa = arg;
 908         snaponearg_t *soa;
 909         objset_t *os;
 910         int err;
 911 
 912         (void) strlcpy(fsname, snapname, sizeof (fsname));
 913         strchr(fsname, '@')[0] = '\0';
 914 
 915         err = dmu_objset_hold(fsname, saa, &os);
 916         if (err != 0)
 917                 return (err);
 918 
 919         /*
 920          * If the objset is in an inconsistent state (eg, in the process
 921          * of being destroyed), don't snapshot it.
 922          */
 923         if (os->os_dsl_dataset->ds_phys->ds_flags & DS_FLAG_INCONSISTENT) {
 924                 dmu_objset_rele(os, saa);
 925                 return (EBUSY);
 926         }
 927 
 928         if (saa->saa_needsuspend) {
 929                 err = zil_suspend(dmu_objset_zil(os));
 930                 if (err) {
 931                         dmu_objset_rele(os, saa);
 932                         return (err);
 933                 }
 934         }
 935 
 936         soa = kmem_zalloc(sizeof (*soa), KM_SLEEP);
 937         soa->soa_saa = saa;
 938         soa->soa_longname = snapname;
 939         soa->soa_snapname = strchr(snapname, '@') + 1;
 940         soa->soa_tot_cnt = cnt;
 941 
 942         dsl_sync_task_create(saa->saa_dstg, snapshot_check, snapshot_sync,
 943             os, soa, 3);
 944 
 945         return (0);
 946 }
 947 
 948 /*
 949  * The snapshots must all be in the same pool.
 950  */
 951 int
 952 dmu_objset_snapshot(nvlist_t *snaps, nvlist_t *props, nvlist_t *errors)
 953 {
 954         dsl_sync_task_t *dst;
 955         snapallarg_t saa = { 0 };
 956         spa_t *spa;
 957         int rv = 0;
 958         int err;
 959         nvpair_t *pair;
 960         nvlist_t *cnt_track = NULL;
 961         nvpair_t *pp;
 962         char *pdelim;
 963         uint64_t *pv;
 964         char nm[MAXPATHLEN];
 965 
 966         pair = nvlist_next_nvpair(snaps, NULL);
 967         if (pair == NULL)
 968                 return (0);
 969 
 970         err = spa_open(nvpair_name(pair), &spa, FTAG);
 971         if (err)
 972                 return (err);
 973         saa.saa_dstg = dsl_sync_task_group_create(spa_get_dsl(spa));
 974         saa.saa_props = props;
 975         saa.saa_needsuspend = (spa_version(spa) < SPA_VERSION_FAST_SNAP);
 976 
 977         /*
 978          * Pre-compute how many total new snapshots will be created for each
 979          * level in the tree and below. This is needed for validating the
 980          * snapshot limit when taking a recursive snapshot.
 981          *
 982          * The problem is that the counts are not actually adjusted when
 983          * we are checking, only when we finally sync. For a single snapshot,
 984          * this is easy, the count will increase by 1 at each node up the tree,
 985          * but its more complicated for recursive snapshots. Since we are
 986          * validating each snapshot independently we need to be sure that we
 987          * are validating the complete count for the entire set of snapshots.
 988          * We do this by rolling up the counts for each component of the name
 989          * into an nvlist then we'll use that count in the validation of each
 990          * individual snapshot.
 991          *
 992          * We validated the snapshot names in zfs_ioc_snapshot so we know they
 993          * have a '@'.
 994          */
 995         VERIFY(nvlist_alloc(&cnt_track, NV_UNIQUE_NAME, KM_SLEEP) == 0);
 996 
 997         for (pair = nvlist_next_nvpair(snaps, NULL); pair != NULL;
 998             pair = nvlist_next_nvpair(snaps, pair)) {
 999                 (void) strlcpy(nm, nvpair_name(pair), sizeof (nm));
1000                 pdelim = strchr(nm, '@');
1001                 *pdelim = '\0';
1002 
1003                 do {
1004                         if (nvlist_lookup_nvpair(cnt_track, nm, &pp) != 0) {
1005                                 /* add to list */
1006                                 nvlist_add_uint64(cnt_track, nm, 1);
1007                         } else {
1008                                 /* update existing entry */
1009                                 pv = (uint64_t *)NVP_VALUE(pp);
1010                                 *pv += 1;
1011                         }
1012 
1013                         pdelim = strrchr(nm, '/');
1014                         if (pdelim != NULL)
1015                                 *pdelim = '\0';
1016                 } while (pdelim != NULL);
1017         }
1018 
1019         /*
1020          * We've calculated the counts, now validate.
1021          */
1022         for (pair = nvlist_next_nvpair(snaps, NULL); pair != NULL;
1023             pair = nvlist_next_nvpair(snaps, pair)) {
1024                 (void) strlcpy(nm, nvpair_name(pair), sizeof (nm));
1025                 pdelim = strchr(nm, '@');
1026                 *pdelim = '\0';
1027 
1028                 VERIFY(nvlist_lookup_nvpair(cnt_track, nm, &pp) == 0);
1029 
1030                 pv = (uint64_t *)NVP_VALUE(pp);
1031                 err = snapshot_one_impl(nvpair_name(pair), &saa, *pv);
1032                 if (err != 0) {
1033                         if (errors != NULL) {
1034                                 fnvlist_add_int32(errors,
1035                                     nvpair_name(pair), err);
1036                         }
1037                         rv = err;
1038                 }
1039         }
1040 
1041         nvlist_free(cnt_track);
1042 
1043         /*
1044          * If any call to snapshot_one_impl() failed, don't execute the
1045          * sync task.  The error handling code below will clean up the
1046          * snaponearg_t from any successful calls to
1047          * snapshot_one_impl().
1048          */
1049         if (rv == 0)
1050                 err = dsl_sync_task_group_wait(saa.saa_dstg);
1051         if (err != 0)
1052                 rv = err;
1053 
1054         for (dst = list_head(&saa.saa_dstg->dstg_tasks); dst;
1055             dst = list_next(&saa.saa_dstg->dstg_tasks, dst)) {
1056                 objset_t *os = dst->dst_arg1;
1057                 snaponearg_t *soa = dst->dst_arg2;
1058                 if (dst->dst_err != 0) {
1059                         if (errors != NULL) {
1060                                 fnvlist_add_int32(errors,
1061                                     soa->soa_longname, dst->dst_err);
1062                         }
1063                         rv = dst->dst_err;
1064                 }
1065 
1066                 if (saa.saa_needsuspend)
1067                         zil_resume(dmu_objset_zil(os));
1068                 dmu_objset_rele(os, &saa);
1069                 kmem_free(soa, sizeof (*soa));
1070         }
1071 
1072         dsl_sync_task_group_destroy(saa.saa_dstg);
1073         spa_close(spa, FTAG);
1074         return (rv);
1075 }
1076 
1077 int
1078 dmu_objset_snapshot_one(const char *fsname, const char *snapname)
1079 {
1080         int err;
1081         char *longsnap = kmem_asprintf("%s@%s", fsname, snapname);
1082         nvlist_t *snaps = fnvlist_alloc();
1083 
1084         fnvlist_add_boolean(snaps, longsnap);
1085         err = dmu_objset_snapshot(snaps, NULL, NULL);
1086         fnvlist_free(snaps);
1087         strfree(longsnap);
1088         return (err);
1089 }
1090 
1091 int
1092 dmu_objset_snapshot_tmp(const char *snapname, const char *tag, int cleanup_fd)
1093 {
1094         dsl_sync_task_t *dst;
1095         snapallarg_t saa = { 0 };
1096         spa_t *spa;
1097         minor_t minor;
1098         int err;
1099 
1100         err = spa_open(snapname, &spa, FTAG);
1101         if (err)
1102                 return (err);
1103         saa.saa_dstg = dsl_sync_task_group_create(spa_get_dsl(spa));
1104         saa.saa_htag = tag;
1105         saa.saa_needsuspend = (spa_version(spa) < SPA_VERSION_FAST_SNAP);
1106         saa.saa_temporary = B_TRUE;
1107 
1108         if (cleanup_fd < 0) {
1109                 spa_close(spa, FTAG);
1110                 return (EINVAL);
1111         }
1112         if ((err = zfs_onexit_fd_hold(cleanup_fd, &minor)) != 0) {
1113                 spa_close(spa, FTAG);
1114                 return (err);
1115         }
1116 
1117         err = snapshot_one_impl(snapname, &saa, 1);
1118 
1119         if (err == 0)
1120                 err = dsl_sync_task_group_wait(saa.saa_dstg);
1121 
1122         for (dst = list_head(&saa.saa_dstg->dstg_tasks); dst;
1123             dst = list_next(&saa.saa_dstg->dstg_tasks, dst)) {
1124                 objset_t *os = dst->dst_arg1;
1125                 dsl_register_onexit_hold_cleanup(saa.saa_newds, tag, minor);
1126                 if (saa.saa_needsuspend)
1127                         zil_resume(dmu_objset_zil(os));
1128                 dmu_objset_rele(os, &saa);
1129         }
1130 
1131         zfs_onexit_fd_rele(cleanup_fd);
1132         dsl_sync_task_group_destroy(saa.saa_dstg);
1133         spa_close(spa, FTAG);
1134         return (err);
1135 }
1136 
1137 
1138 static void
1139 dmu_objset_sync_dnodes(list_t *list, list_t *newlist, dmu_tx_t *tx)
1140 {
1141         dnode_t *dn;
1142 
1143         while (dn = list_head(list)) {
1144                 ASSERT(dn->dn_object != DMU_META_DNODE_OBJECT);
1145                 ASSERT(dn->dn_dbuf->db_data_pending);
1146                 /*
1147                  * Initialize dn_zio outside dnode_sync() because the
1148                  * meta-dnode needs to set it ouside dnode_sync().
1149                  */
1150                 dn->dn_zio = dn->dn_dbuf->db_data_pending->dr_zio;
1151                 ASSERT(dn->dn_zio);
1152 
1153                 ASSERT3U(dn->dn_nlevels, <=, DN_MAX_LEVELS);
1154                 list_remove(list, dn);
1155 
1156                 if (newlist) {
1157                         (void) dnode_add_ref(dn, newlist);
1158                         list_insert_tail(newlist, dn);
1159                 }
1160 
1161                 dnode_sync(dn, tx);
1162         }
1163 }
1164 
1165 /* ARGSUSED */
1166 static void
1167 dmu_objset_write_ready(zio_t *zio, arc_buf_t *abuf, void *arg)
1168 {
1169         blkptr_t *bp = zio->io_bp;
1170         objset_t *os = arg;
1171         dnode_phys_t *dnp = &os->os_phys->os_meta_dnode;
1172 
1173         ASSERT(bp == os->os_rootbp);
1174         ASSERT(BP_GET_TYPE(bp) == DMU_OT_OBJSET);
1175         ASSERT(BP_GET_LEVEL(bp) == 0);
1176 
1177         /*
1178          * Update rootbp fill count: it should be the number of objects
1179          * allocated in the object set (not counting the "special"
1180          * objects that are stored in the objset_phys_t -- the meta
1181          * dnode and user/group accounting objects).
1182          */
1183         bp->blk_fill = 0;
1184         for (int i = 0; i < dnp->dn_nblkptr; i++)
1185                 bp->blk_fill += dnp->dn_blkptr[i].blk_fill;
1186 }
1187 
1188 /* ARGSUSED */
1189 static void
1190 dmu_objset_write_done(zio_t *zio, arc_buf_t *abuf, void *arg)
1191 {
1192         blkptr_t *bp = zio->io_bp;
1193         blkptr_t *bp_orig = &zio->io_bp_orig;
1194         objset_t *os = arg;
1195 
1196         if (zio->io_flags & ZIO_FLAG_IO_REWRITE) {
1197                 ASSERT(BP_EQUAL(bp, bp_orig));
1198         } else {
1199                 dsl_dataset_t *ds = os->os_dsl_dataset;
1200                 dmu_tx_t *tx = os->os_synctx;
1201 
1202                 (void) dsl_dataset_block_kill(ds, bp_orig, tx, B_TRUE);
1203                 dsl_dataset_block_born(ds, bp, tx);
1204         }
1205 }
1206 
1207 /* called from dsl */
1208 void
1209 dmu_objset_sync(objset_t *os, zio_t *pio, dmu_tx_t *tx)
1210 {
1211         int txgoff;
1212         zbookmark_t zb;
1213         zio_prop_t zp;
1214         zio_t *zio;
1215         list_t *list;
1216         list_t *newlist = NULL;
1217         dbuf_dirty_record_t *dr;
1218 
1219         dprintf_ds(os->os_dsl_dataset, "txg=%llu\n", tx->tx_txg);
1220 
1221         ASSERT(dmu_tx_is_syncing(tx));
1222         /* XXX the write_done callback should really give us the tx... */
1223         os->os_synctx = tx;
1224 
1225         if (os->os_dsl_dataset == NULL) {
1226                 /*
1227                  * This is the MOS.  If we have upgraded,
1228                  * spa_max_replication() could change, so reset
1229                  * os_copies here.
1230                  */
1231                 os->os_copies = spa_max_replication(os->os_spa);
1232         }
1233 
1234         /*
1235          * Create the root block IO
1236          */
1237         SET_BOOKMARK(&zb, os->os_dsl_dataset ?
1238             os->os_dsl_dataset->ds_object : DMU_META_OBJSET,
1239             ZB_ROOT_OBJECT, ZB_ROOT_LEVEL, ZB_ROOT_BLKID);
1240         VERIFY3U(0, ==, arc_release_bp(os->os_phys_buf, &os->os_phys_buf,
1241             os->os_rootbp, os->os_spa, &zb));
1242 
1243         dmu_write_policy(os, NULL, 0, 0, &zp);
1244 
1245         zio = arc_write(pio, os->os_spa, tx->tx_txg,
1246             os->os_rootbp, os->os_phys_buf, DMU_OS_IS_L2CACHEABLE(os), &zp,
1247             dmu_objset_write_ready, dmu_objset_write_done, os,
1248             ZIO_PRIORITY_ASYNC_WRITE, ZIO_FLAG_MUSTSUCCEED, &zb);
1249 
1250         /*
1251          * Sync special dnodes - the parent IO for the sync is the root block
1252          */
1253         DMU_META_DNODE(os)->dn_zio = zio;
1254         dnode_sync(DMU_META_DNODE(os), tx);
1255 
1256         os->os_phys->os_flags = os->os_flags;
1257 
1258         if (DMU_USERUSED_DNODE(os) &&
1259             DMU_USERUSED_DNODE(os)->dn_type != DMU_OT_NONE) {
1260                 DMU_USERUSED_DNODE(os)->dn_zio = zio;
1261                 dnode_sync(DMU_USERUSED_DNODE(os), tx);
1262                 DMU_GROUPUSED_DNODE(os)->dn_zio = zio;
1263                 dnode_sync(DMU_GROUPUSED_DNODE(os), tx);
1264         }
1265 
1266         txgoff = tx->tx_txg & TXG_MASK;
1267 
1268         if (dmu_objset_userused_enabled(os)) {
1269                 newlist = &os->os_synced_dnodes;
1270                 /*
1271                  * We must create the list here because it uses the
1272                  * dn_dirty_link[] of this txg.
1273                  */
1274                 list_create(newlist, sizeof (dnode_t),
1275                     offsetof(dnode_t, dn_dirty_link[txgoff]));
1276         }
1277 
1278         dmu_objset_sync_dnodes(&os->os_free_dnodes[txgoff], newlist, tx);
1279         dmu_objset_sync_dnodes(&os->os_dirty_dnodes[txgoff], newlist, tx);
1280 
1281         list = &DMU_META_DNODE(os)->dn_dirty_records[txgoff];
1282         while (dr = list_head(list)) {
1283                 ASSERT(dr->dr_dbuf->db_level == 0);
1284                 list_remove(list, dr);
1285                 if (dr->dr_zio)
1286                         zio_nowait(dr->dr_zio);
1287         }
1288         /*
1289          * Free intent log blocks up to this tx.
1290          */
1291         zil_sync(os->os_zil, tx);
1292         os->os_phys->os_zil_header = os->os_zil_header;
1293         zio_nowait(zio);
1294 }
1295 
1296 boolean_t
1297 dmu_objset_is_dirty(objset_t *os, uint64_t txg)
1298 {
1299         return (!list_is_empty(&os->os_dirty_dnodes[txg & TXG_MASK]) ||
1300             !list_is_empty(&os->os_free_dnodes[txg & TXG_MASK]));
1301 }
1302 
1303 static objset_used_cb_t *used_cbs[DMU_OST_NUMTYPES];
1304 
1305 void
1306 dmu_objset_register_type(dmu_objset_type_t ost, objset_used_cb_t *cb)
1307 {
1308         used_cbs[ost] = cb;
1309 }
1310 
1311 boolean_t
1312 dmu_objset_userused_enabled(objset_t *os)
1313 {
1314         return (spa_version(os->os_spa) >= SPA_VERSION_USERSPACE &&
1315             used_cbs[os->os_phys->os_type] != NULL &&
1316             DMU_USERUSED_DNODE(os) != NULL);
1317 }
1318 
1319 static void
1320 do_userquota_update(objset_t *os, uint64_t used, uint64_t flags,
1321     uint64_t user, uint64_t group, boolean_t subtract, dmu_tx_t *tx)
1322 {
1323         if ((flags & DNODE_FLAG_USERUSED_ACCOUNTED)) {
1324                 int64_t delta = DNODE_SIZE + used;
1325                 if (subtract)
1326                         delta = -delta;
1327                 VERIFY3U(0, ==, zap_increment_int(os, DMU_USERUSED_OBJECT,
1328                     user, delta, tx));
1329                 VERIFY3U(0, ==, zap_increment_int(os, DMU_GROUPUSED_OBJECT,
1330                     group, delta, tx));
1331         }
1332 }
1333 
1334 void
1335 dmu_objset_do_userquota_updates(objset_t *os, dmu_tx_t *tx)
1336 {
1337         dnode_t *dn;
1338         list_t *list = &os->os_synced_dnodes;
1339 
1340         ASSERT(list_head(list) == NULL || dmu_objset_userused_enabled(os));
1341 
1342         while (dn = list_head(list)) {
1343                 int flags;
1344                 ASSERT(!DMU_OBJECT_IS_SPECIAL(dn->dn_object));
1345                 ASSERT(dn->dn_phys->dn_type == DMU_OT_NONE ||
1346                     dn->dn_phys->dn_flags &
1347                     DNODE_FLAG_USERUSED_ACCOUNTED);
1348 
1349                 /* Allocate the user/groupused objects if necessary. */
1350                 if (DMU_USERUSED_DNODE(os)->dn_type == DMU_OT_NONE) {
1351                         VERIFY(0 == zap_create_claim(os,
1352                             DMU_USERUSED_OBJECT,
1353                             DMU_OT_USERGROUP_USED, DMU_OT_NONE, 0, tx));
1354                         VERIFY(0 == zap_create_claim(os,
1355                             DMU_GROUPUSED_OBJECT,
1356                             DMU_OT_USERGROUP_USED, DMU_OT_NONE, 0, tx));
1357                 }
1358 
1359                 /*
1360                  * We intentionally modify the zap object even if the
1361                  * net delta is zero.  Otherwise
1362                  * the block of the zap obj could be shared between
1363                  * datasets but need to be different between them after
1364                  * a bprewrite.
1365                  */
1366 
1367                 flags = dn->dn_id_flags;
1368                 ASSERT(flags);
1369                 if (flags & DN_ID_OLD_EXIST)  {
1370                         do_userquota_update(os, dn->dn_oldused, dn->dn_oldflags,
1371                             dn->dn_olduid, dn->dn_oldgid, B_TRUE, tx);
1372                 }
1373                 if (flags & DN_ID_NEW_EXIST) {
1374                         do_userquota_update(os, DN_USED_BYTES(dn->dn_phys),
1375                             dn->dn_phys->dn_flags,  dn->dn_newuid,
1376                             dn->dn_newgid, B_FALSE, tx);
1377                 }
1378 
1379                 mutex_enter(&dn->dn_mtx);
1380                 dn->dn_oldused = 0;
1381                 dn->dn_oldflags = 0;
1382                 if (dn->dn_id_flags & DN_ID_NEW_EXIST) {
1383                         dn->dn_olduid = dn->dn_newuid;
1384                         dn->dn_oldgid = dn->dn_newgid;
1385                         dn->dn_id_flags |= DN_ID_OLD_EXIST;
1386                         if (dn->dn_bonuslen == 0)
1387                                 dn->dn_id_flags |= DN_ID_CHKED_SPILL;
1388                         else
1389                                 dn->dn_id_flags |= DN_ID_CHKED_BONUS;
1390                 }
1391                 dn->dn_id_flags &= ~(DN_ID_NEW_EXIST);
1392                 mutex_exit(&dn->dn_mtx);
1393 
1394                 list_remove(list, dn);
1395                 dnode_rele(dn, list);
1396         }
1397 }
1398 
1399 /*
1400  * Returns a pointer to data to find uid/gid from
1401  *
1402  * If a dirty record for transaction group that is syncing can't
1403  * be found then NULL is returned.  In the NULL case it is assumed
1404  * the uid/gid aren't changing.
1405  */
1406 static void *
1407 dmu_objset_userquota_find_data(dmu_buf_impl_t *db, dmu_tx_t *tx)
1408 {
1409         dbuf_dirty_record_t *dr, **drp;
1410         void *data;
1411 
1412         if (db->db_dirtycnt == 0)
1413                 return (db->db.db_data);  /* Nothing is changing */
1414 
1415         for (drp = &db->db_last_dirty; (dr = *drp) != NULL; drp = &dr->dr_next)
1416                 if (dr->dr_txg == tx->tx_txg)
1417                         break;
1418 
1419         if (dr == NULL) {
1420                 data = NULL;
1421         } else {
1422                 dnode_t *dn;
1423 
1424                 DB_DNODE_ENTER(dr->dr_dbuf);
1425                 dn = DB_DNODE(dr->dr_dbuf);
1426 
1427                 if (dn->dn_bonuslen == 0 &&
1428                     dr->dr_dbuf->db_blkid == DMU_SPILL_BLKID)
1429                         data = dr->dt.dl.dr_data->b_data;
1430                 else
1431                         data = dr->dt.dl.dr_data;
1432 
1433                 DB_DNODE_EXIT(dr->dr_dbuf);
1434         }
1435 
1436         return (data);
1437 }
1438 
1439 void
1440 dmu_objset_userquota_get_ids(dnode_t *dn, boolean_t before, dmu_tx_t *tx)
1441 {
1442         objset_t *os = dn->dn_objset;
1443         void *data = NULL;
1444         dmu_buf_impl_t *db = NULL;
1445         uint64_t *user, *group;
1446         int flags = dn->dn_id_flags;
1447         int error;
1448         boolean_t have_spill = B_FALSE;
1449 
1450         if (!dmu_objset_userused_enabled(dn->dn_objset))
1451                 return;
1452 
1453         if (before && (flags & (DN_ID_CHKED_BONUS|DN_ID_OLD_EXIST|
1454             DN_ID_CHKED_SPILL)))
1455                 return;
1456 
1457         if (before && dn->dn_bonuslen != 0)
1458                 data = DN_BONUS(dn->dn_phys);
1459         else if (!before && dn->dn_bonuslen != 0) {
1460                 if (dn->dn_bonus) {
1461                         db = dn->dn_bonus;
1462                         mutex_enter(&db->db_mtx);
1463                         data = dmu_objset_userquota_find_data(db, tx);
1464                 } else {
1465                         data = DN_BONUS(dn->dn_phys);
1466                 }
1467         } else if (dn->dn_bonuslen == 0 && dn->dn_bonustype == DMU_OT_SA) {
1468                         int rf = 0;
1469 
1470                         if (RW_WRITE_HELD(&dn->dn_struct_rwlock))
1471                                 rf |= DB_RF_HAVESTRUCT;
1472                         error = dmu_spill_hold_by_dnode(dn,
1473                             rf | DB_RF_MUST_SUCCEED,
1474                             FTAG, (dmu_buf_t **)&db);
1475                         ASSERT(error == 0);
1476                         mutex_enter(&db->db_mtx);
1477                         data = (before) ? db->db.db_data :
1478                             dmu_objset_userquota_find_data(db, tx);
1479                         have_spill = B_TRUE;
1480         } else {
1481                 mutex_enter(&dn->dn_mtx);
1482                 dn->dn_id_flags |= DN_ID_CHKED_BONUS;
1483                 mutex_exit(&dn->dn_mtx);
1484                 return;
1485         }
1486 
1487         if (before) {
1488                 ASSERT(data);
1489                 user = &dn->dn_olduid;
1490                 group = &dn->dn_oldgid;
1491         } else if (data) {
1492                 user = &dn->dn_newuid;
1493                 group = &dn->dn_newgid;
1494         }
1495 
1496         /*
1497          * Must always call the callback in case the object
1498          * type has changed and that type isn't an object type to track
1499          */
1500         error = used_cbs[os->os_phys->os_type](dn->dn_bonustype, data,
1501             user, group);
1502 
1503         /*
1504          * Preserve existing uid/gid when the callback can't determine
1505          * what the new uid/gid are and the callback returned EEXIST.
1506          * The EEXIST error tells us to just use the existing uid/gid.
1507          * If we don't know what the old values are then just assign
1508          * them to 0, since that is a new file  being created.
1509          */
1510         if (!before && data == NULL && error == EEXIST) {
1511                 if (flags & DN_ID_OLD_EXIST) {
1512                         dn->dn_newuid = dn->dn_olduid;
1513                         dn->dn_newgid = dn->dn_oldgid;
1514                 } else {
1515                         dn->dn_newuid = 0;
1516                         dn->dn_newgid = 0;
1517                 }
1518                 error = 0;
1519         }
1520 
1521         if (db)
1522                 mutex_exit(&db->db_mtx);
1523 
1524         mutex_enter(&dn->dn_mtx);
1525         if (error == 0 && before)
1526                 dn->dn_id_flags |= DN_ID_OLD_EXIST;
1527         if (error == 0 && !before)
1528                 dn->dn_id_flags |= DN_ID_NEW_EXIST;
1529 
1530         if (have_spill) {
1531                 dn->dn_id_flags |= DN_ID_CHKED_SPILL;
1532         } else {
1533                 dn->dn_id_flags |= DN_ID_CHKED_BONUS;
1534         }
1535         mutex_exit(&dn->dn_mtx);
1536         if (have_spill)
1537                 dmu_buf_rele((dmu_buf_t *)db, FTAG);
1538 }
1539 
1540 boolean_t
1541 dmu_objset_userspace_present(objset_t *os)
1542 {
1543         return (os->os_phys->os_flags &
1544             OBJSET_FLAG_USERACCOUNTING_COMPLETE);
1545 }
1546 
1547 int
1548 dmu_objset_userspace_upgrade(objset_t *os)
1549 {
1550         uint64_t obj;
1551         int err = 0;
1552 
1553         if (dmu_objset_userspace_present(os))
1554                 return (0);
1555         if (!dmu_objset_userused_enabled(os))
1556                 return (ENOTSUP);
1557         if (dmu_objset_is_snapshot(os))
1558                 return (EINVAL);
1559 
1560         /*
1561          * We simply need to mark every object dirty, so that it will be
1562          * synced out and now accounted.  If this is called
1563          * concurrently, or if we already did some work before crashing,
1564          * that's fine, since we track each object's accounted state
1565          * independently.
1566          */
1567 
1568         for (obj = 0; err == 0; err = dmu_object_next(os, &obj, FALSE, 0)) {
1569                 dmu_tx_t *tx;
1570                 dmu_buf_t *db;
1571                 int objerr;
1572 
1573                 if (issig(JUSTLOOKING) && issig(FORREAL))
1574                         return (EINTR);
1575 
1576                 objerr = dmu_bonus_hold(os, obj, FTAG, &db);
1577                 if (objerr)
1578                         continue;
1579                 tx = dmu_tx_create(os);
1580                 dmu_tx_hold_bonus(tx, obj);
1581                 objerr = dmu_tx_assign(tx, TXG_WAIT);
1582                 if (objerr) {
1583                         dmu_tx_abort(tx);
1584                         continue;
1585                 }
1586                 dmu_buf_will_dirty(db, tx);
1587                 dmu_buf_rele(db, FTAG);
1588                 dmu_tx_commit(tx);
1589         }
1590 
1591         os->os_flags |= OBJSET_FLAG_USERACCOUNTING_COMPLETE;
1592         txg_wait_synced(dmu_objset_pool(os), 0);
1593         return (0);
1594 }
1595 
1596 void
1597 dmu_objset_space(objset_t *os, uint64_t *refdbytesp, uint64_t *availbytesp,
1598     uint64_t *usedobjsp, uint64_t *availobjsp)
1599 {
1600         dsl_dataset_space(os->os_dsl_dataset, refdbytesp, availbytesp,
1601             usedobjsp, availobjsp);
1602 }
1603 
1604 uint64_t
1605 dmu_objset_fsid_guid(objset_t *os)
1606 {
1607         return (dsl_dataset_fsid_guid(os->os_dsl_dataset));
1608 }
1609 
1610 void
1611 dmu_objset_fast_stat(objset_t *os, dmu_objset_stats_t *stat)
1612 {
1613         stat->dds_type = os->os_phys->os_type;
1614         if (os->os_dsl_dataset)
1615                 dsl_dataset_fast_stat(os->os_dsl_dataset, stat);
1616 }
1617 
1618 void
1619 dmu_objset_stats(objset_t *os, nvlist_t *nv)
1620 {
1621         ASSERT(os->os_dsl_dataset ||
1622             os->os_phys->os_type == DMU_OST_META);
1623 
1624         if (os->os_dsl_dataset != NULL)
1625                 dsl_dataset_stats(os->os_dsl_dataset, nv);
1626 
1627         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_TYPE,
1628             os->os_phys->os_type);
1629         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USERACCOUNTING,
1630             dmu_objset_userspace_present(os));
1631 }
1632 
1633 int
1634 dmu_objset_is_snapshot(objset_t *os)
1635 {
1636         if (os->os_dsl_dataset != NULL)
1637                 return (dsl_dataset_is_snapshot(os->os_dsl_dataset));
1638         else
1639                 return (B_FALSE);
1640 }
1641 
1642 int
1643 dmu_snapshot_realname(objset_t *os, char *name, char *real, int maxlen,
1644     boolean_t *conflict)
1645 {
1646         dsl_dataset_t *ds = os->os_dsl_dataset;
1647         uint64_t ignored;
1648 
1649         if (ds->ds_phys->ds_snapnames_zapobj == 0)
1650                 return (ENOENT);
1651 
1652         return (zap_lookup_norm(ds->ds_dir->dd_pool->dp_meta_objset,
1653             ds->ds_phys->ds_snapnames_zapobj, name, 8, 1, &ignored, MT_FIRST,
1654             real, maxlen, conflict));
1655 }
1656 
1657 int
1658 dmu_snapshot_list_next(objset_t *os, int namelen, char *name,
1659     uint64_t *idp, uint64_t *offp, boolean_t *case_conflict)
1660 {
1661         dsl_dataset_t *ds = os->os_dsl_dataset;
1662         zap_cursor_t cursor;
1663         zap_attribute_t attr;
1664 
1665         if (ds->ds_phys->ds_snapnames_zapobj == 0)
1666                 return (ENOENT);
1667 
1668         zap_cursor_init_serialized(&cursor,
1669             ds->ds_dir->dd_pool->dp_meta_objset,
1670             ds->ds_phys->ds_snapnames_zapobj, *offp);
1671 
1672         if (zap_cursor_retrieve(&cursor, &attr) != 0) {
1673                 zap_cursor_fini(&cursor);
1674                 return (ENOENT);
1675         }
1676 
1677         if (strlen(attr.za_name) + 1 > namelen) {
1678                 zap_cursor_fini(&cursor);
1679                 return (ENAMETOOLONG);
1680         }
1681 
1682         (void) strcpy(name, attr.za_name);
1683         if (idp)
1684                 *idp = attr.za_first_integer;
1685         if (case_conflict)
1686                 *case_conflict = attr.za_normalization_conflict;
1687         zap_cursor_advance(&cursor);
1688         *offp = zap_cursor_serialize(&cursor);
1689         zap_cursor_fini(&cursor);
1690 
1691         return (0);
1692 }
1693 
1694 int
1695 dmu_dir_list_next(objset_t *os, int namelen, char *name,
1696     uint64_t *idp, uint64_t *offp)
1697 {
1698         dsl_dir_t *dd = os->os_dsl_dataset->ds_dir;
1699         zap_cursor_t cursor;
1700         zap_attribute_t attr;
1701 
1702         /* there is no next dir on a snapshot! */
1703         if (os->os_dsl_dataset->ds_object !=
1704             dd->dd_phys->dd_head_dataset_obj)
1705                 return (ENOENT);
1706 
1707         zap_cursor_init_serialized(&cursor,
1708             dd->dd_pool->dp_meta_objset,
1709             dd->dd_phys->dd_child_dir_zapobj, *offp);
1710 
1711         if (zap_cursor_retrieve(&cursor, &attr) != 0) {
1712                 zap_cursor_fini(&cursor);
1713                 return (ENOENT);
1714         }
1715 
1716         if (strlen(attr.za_name) + 1 > namelen) {
1717                 zap_cursor_fini(&cursor);
1718                 return (ENAMETOOLONG);
1719         }
1720 
1721         (void) strcpy(name, attr.za_name);
1722         if (idp)
1723                 *idp = attr.za_first_integer;
1724         zap_cursor_advance(&cursor);
1725         *offp = zap_cursor_serialize(&cursor);
1726         zap_cursor_fini(&cursor);
1727 
1728         return (0);
1729 }
1730 
1731 struct findarg {
1732         int (*func)(const char *, void *);
1733         void *arg;
1734 };
1735 
1736 /* ARGSUSED */
1737 static int
1738 findfunc(spa_t *spa, uint64_t dsobj, const char *dsname, void *arg)
1739 {
1740         struct findarg *fa = arg;
1741         return (fa->func(dsname, fa->arg));
1742 }
1743 
1744 /*
1745  * Find all objsets under name, and for each, call 'func(child_name, arg)'.
1746  * Perhaps change all callers to use dmu_objset_find_spa()?
1747  */
1748 int
1749 dmu_objset_find(char *name, int func(const char *, void *), void *arg,
1750     int flags)
1751 {
1752         struct findarg fa;
1753         fa.func = func;
1754         fa.arg = arg;
1755         return (dmu_objset_find_spa(NULL, name, findfunc, &fa, flags));
1756 }
1757 
1758 /*
1759  * Find all objsets under name, call func on each
1760  */
1761 int
1762 dmu_objset_find_spa(spa_t *spa, const char *name,
1763     int func(spa_t *, uint64_t, const char *, void *), void *arg, int flags)
1764 {
1765         dsl_dir_t *dd;
1766         dsl_pool_t *dp;
1767         dsl_dataset_t *ds;
1768         zap_cursor_t zc;
1769         zap_attribute_t *attr;
1770         char *child;
1771         uint64_t thisobj;
1772         int err;
1773 
1774         if (name == NULL)
1775                 name = spa_name(spa);
1776         err = dsl_dir_open_spa(spa, name, FTAG, &dd, NULL);
1777         if (err)
1778                 return (err);
1779 
1780         /* Don't visit hidden ($MOS & $ORIGIN) objsets. */
1781         if (dd->dd_myname[0] == '$') {
1782                 dsl_dir_close(dd, FTAG);
1783                 return (0);
1784         }
1785 
1786         thisobj = dd->dd_phys->dd_head_dataset_obj;
1787         attr = kmem_alloc(sizeof (zap_attribute_t), KM_SLEEP);
1788         dp = dd->dd_pool;
1789 
1790         /*
1791          * Iterate over all children.
1792          */
1793         if (flags & DS_FIND_CHILDREN) {
1794                 for (zap_cursor_init(&zc, dp->dp_meta_objset,
1795                     dd->dd_phys->dd_child_dir_zapobj);
1796                     zap_cursor_retrieve(&zc, attr) == 0;
1797                     (void) zap_cursor_advance(&zc)) {
1798                         ASSERT(attr->za_integer_length == sizeof (uint64_t));
1799                         ASSERT(attr->za_num_integers == 1);
1800 
1801                         child = kmem_asprintf("%s/%s", name, attr->za_name);
1802                         err = dmu_objset_find_spa(spa, child, func, arg, flags);
1803                         strfree(child);
1804                         if (err)
1805                                 break;
1806                 }
1807                 zap_cursor_fini(&zc);
1808 
1809                 if (err) {
1810                         dsl_dir_close(dd, FTAG);
1811                         kmem_free(attr, sizeof (zap_attribute_t));
1812                         return (err);
1813                 }
1814         }
1815 
1816         /*
1817          * Iterate over all snapshots.
1818          */
1819         if (flags & DS_FIND_SNAPSHOTS) {
1820                 if (!dsl_pool_sync_context(dp))
1821                         rw_enter(&dp->dp_config_rwlock, RW_READER);
1822                 err = dsl_dataset_hold_obj(dp, thisobj, FTAG, &ds);
1823                 if (!dsl_pool_sync_context(dp))
1824                         rw_exit(&dp->dp_config_rwlock);
1825 
1826                 if (err == 0) {
1827                         uint64_t snapobj = ds->ds_phys->ds_snapnames_zapobj;
1828                         dsl_dataset_rele(ds, FTAG);
1829 
1830                         for (zap_cursor_init(&zc, dp->dp_meta_objset, snapobj);
1831                             zap_cursor_retrieve(&zc, attr) == 0;
1832                             (void) zap_cursor_advance(&zc)) {
1833                                 ASSERT(attr->za_integer_length ==
1834                                     sizeof (uint64_t));
1835                                 ASSERT(attr->za_num_integers == 1);
1836 
1837                                 child = kmem_asprintf("%s@%s",
1838                                     name, attr->za_name);
1839                                 err = func(spa, attr->za_first_integer,
1840                                     child, arg);
1841                                 strfree(child);
1842                                 if (err)
1843                                         break;
1844                         }
1845                         zap_cursor_fini(&zc);
1846                 }
1847         }
1848 
1849         dsl_dir_close(dd, FTAG);
1850         kmem_free(attr, sizeof (zap_attribute_t));
1851 
1852         if (err)
1853                 return (err);
1854 
1855         /*
1856          * Apply to self if appropriate.
1857          */
1858         err = func(spa, thisobj, name, arg);
1859         return (err);
1860 }
1861 
1862 /* ARGSUSED */
1863 int
1864 dmu_objset_prefetch(const char *name, void *arg)
1865 {
1866         dsl_dataset_t *ds;
1867 
1868         if (dsl_dataset_hold(name, FTAG, &ds))
1869                 return (0);
1870 
1871         if (!BP_IS_HOLE(&ds->ds_phys->ds_bp)) {
1872                 mutex_enter(&ds->ds_opening_lock);
1873                 if (ds->ds_objset == NULL) {
1874                         uint32_t aflags = ARC_NOWAIT | ARC_PREFETCH;
1875                         zbookmark_t zb;
1876 
1877                         SET_BOOKMARK(&zb, ds->ds_object, ZB_ROOT_OBJECT,
1878                             ZB_ROOT_LEVEL, ZB_ROOT_BLKID);
1879 
1880                         (void) dsl_read_nolock(NULL, dsl_dataset_get_spa(ds),
1881                             &ds->ds_phys->ds_bp, NULL, NULL,
1882                             ZIO_PRIORITY_ASYNC_READ,
1883                             ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE,
1884                             &aflags, &zb);
1885                 }
1886                 mutex_exit(&ds->ds_opening_lock);
1887         }
1888 
1889         dsl_dataset_rele(ds, FTAG);
1890         return (0);
1891 }
1892 
1893 void
1894 dmu_objset_set_user(objset_t *os, void *user_ptr)
1895 {
1896         ASSERT(MUTEX_HELD(&os->os_user_ptr_lock));
1897         os->os_user_ptr = user_ptr;
1898 }
1899 
1900 void *
1901 dmu_objset_get_user(objset_t *os)
1902 {
1903         ASSERT(MUTEX_HELD(&os->os_user_ptr_lock));
1904         return (os->os_user_ptr);
1905 }