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, 2014 by Delphix. All rights reserved.
  24  * Copyright (c) 2013 by Saso Kiselkov. All rights reserved.
  25  * Copyright (c) 2013, Joyent, Inc. All rights reserved.
  26  * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
  27  * Copyright 2015 Nexenta Systems, Inc. All rights reserved.
  28  * Copyright (c) 2015, STRATO AG, Inc. All rights reserved.
  29  */
  30 
  31 /* Portions Copyright 2010 Robert Milkowski */
  32 
  33 #include <sys/cred.h>
  34 #include <sys/zfs_context.h>
  35 #include <sys/dmu_objset.h>
  36 #include <sys/dsl_dir.h>
  37 #include <sys/dsl_dataset.h>
  38 #include <sys/dsl_prop.h>
  39 #include <sys/dsl_pool.h>
  40 #include <sys/dsl_synctask.h>
  41 #include <sys/dsl_deleg.h>
  42 #include <sys/dnode.h>
  43 #include <sys/dbuf.h>
  44 #include <sys/zvol.h>
  45 #include <sys/dmu_tx.h>
  46 #include <sys/zap.h>
  47 #include <sys/zil.h>
  48 #include <sys/dmu_impl.h>
  49 #include <sys/zfs_ioctl.h>
  50 #include <sys/sa.h>
  51 #include <sys/zfs_onexit.h>
  52 #include <sys/dsl_destroy.h>
  53 #include <sys/vdev.h>
  54 
  55 /*
  56  * Needed to close a window in dnode_move() that allows the objset to be freed
  57  * before it can be safely accessed.
  58  */
  59 krwlock_t os_lock;
  60 
  61 /*
  62  * Tunable to overwrite the maximum number of threads for the parallization
  63  * of dmu_objset_find_dp, needed to speed up the import of pools with many
  64  * datasets.
  65  * Default is 4 times the number of leaf vdevs.
  66  */
  67 int dmu_find_threads = 0;
  68 
  69 static void dmu_objset_find_dp_cb(void *arg);
  70 
  71 void
  72 dmu_objset_init(void)
  73 {
  74         rw_init(&os_lock, NULL, RW_DEFAULT, NULL);
  75 }
  76 
  77 void
  78 dmu_objset_fini(void)
  79 {
  80         rw_destroy(&os_lock);
  81 }
  82 
  83 spa_t *
  84 dmu_objset_spa(objset_t *os)
  85 {
  86         return (os->os_spa);
  87 }
  88 
  89 zilog_t *
  90 dmu_objset_zil(objset_t *os)
  91 {
  92         return (os->os_zil);
  93 }
  94 
  95 dsl_pool_t *
  96 dmu_objset_pool(objset_t *os)
  97 {
  98         dsl_dataset_t *ds;
  99 
 100         if ((ds = os->os_dsl_dataset) != NULL && ds->ds_dir)
 101                 return (ds->ds_dir->dd_pool);
 102         else
 103                 return (spa_get_dsl(os->os_spa));
 104 }
 105 
 106 dsl_dataset_t *
 107 dmu_objset_ds(objset_t *os)
 108 {
 109         return (os->os_dsl_dataset);
 110 }
 111 
 112 dmu_objset_type_t
 113 dmu_objset_type(objset_t *os)
 114 {
 115         return (os->os_phys->os_type);
 116 }
 117 
 118 void
 119 dmu_objset_name(objset_t *os, char *buf)
 120 {
 121         dsl_dataset_name(os->os_dsl_dataset, buf);
 122 }
 123 
 124 uint64_t
 125 dmu_objset_id(objset_t *os)
 126 {
 127         dsl_dataset_t *ds = os->os_dsl_dataset;
 128 
 129         return (ds ? ds->ds_object : 0);
 130 }
 131 
 132 zfs_sync_type_t
 133 dmu_objset_syncprop(objset_t *os)
 134 {
 135         return (os->os_sync);
 136 }
 137 
 138 zfs_logbias_op_t
 139 dmu_objset_logbias(objset_t *os)
 140 {
 141         return (os->os_logbias);
 142 }
 143 
 144 static void
 145 checksum_changed_cb(void *arg, uint64_t newval)
 146 {
 147         objset_t *os = arg;
 148 
 149         /*
 150          * Inheritance should have been done by now.
 151          */
 152         ASSERT(newval != ZIO_CHECKSUM_INHERIT);
 153 
 154         os->os_checksum = zio_checksum_select(newval, ZIO_CHECKSUM_ON_VALUE);
 155 }
 156 
 157 static void
 158 compression_changed_cb(void *arg, uint64_t newval)
 159 {
 160         objset_t *os = arg;
 161 
 162         /*
 163          * Inheritance and range checking should have been done by now.
 164          */
 165         ASSERT(newval != ZIO_COMPRESS_INHERIT);
 166 
 167         os->os_compress = zio_compress_select(os->os_spa, newval,
 168             ZIO_COMPRESS_ON);
 169 }
 170 
 171 static void
 172 copies_changed_cb(void *arg, uint64_t newval)
 173 {
 174         objset_t *os = arg;
 175 
 176         /*
 177          * Inheritance and range checking should have been done by now.
 178          */
 179         ASSERT(newval > 0);
 180         ASSERT(newval <= spa_max_replication(os->os_spa));
 181 
 182         os->os_copies = newval;
 183 }
 184 
 185 static void
 186 dedup_changed_cb(void *arg, uint64_t newval)
 187 {
 188         objset_t *os = arg;
 189         spa_t *spa = os->os_spa;
 190         enum zio_checksum checksum;
 191 
 192         /*
 193          * Inheritance should have been done by now.
 194          */
 195         ASSERT(newval != ZIO_CHECKSUM_INHERIT);
 196 
 197         checksum = zio_checksum_dedup_select(spa, newval, ZIO_CHECKSUM_OFF);
 198 
 199         os->os_dedup_checksum = checksum & ZIO_CHECKSUM_MASK;
 200         os->os_dedup_verify = !!(checksum & ZIO_CHECKSUM_VERIFY);
 201 }
 202 
 203 static void
 204 primary_cache_changed_cb(void *arg, uint64_t newval)
 205 {
 206         objset_t *os = arg;
 207 
 208         /*
 209          * Inheritance and range checking should have been done by now.
 210          */
 211         ASSERT(newval == ZFS_CACHE_ALL || newval == ZFS_CACHE_NONE ||
 212             newval == ZFS_CACHE_METADATA);
 213 
 214         os->os_primary_cache = newval;
 215 }
 216 
 217 static void
 218 secondary_cache_changed_cb(void *arg, uint64_t newval)
 219 {
 220         objset_t *os = arg;
 221 
 222         /*
 223          * Inheritance and range checking should have been done by now.
 224          */
 225         ASSERT(newval == ZFS_CACHE_ALL || newval == ZFS_CACHE_NONE ||
 226             newval == ZFS_CACHE_METADATA);
 227 
 228         os->os_secondary_cache = newval;
 229 }
 230 
 231 static void
 232 sync_changed_cb(void *arg, uint64_t newval)
 233 {
 234         objset_t *os = arg;
 235 
 236         /*
 237          * Inheritance and range checking should have been done by now.
 238          */
 239         ASSERT(newval == ZFS_SYNC_STANDARD || newval == ZFS_SYNC_ALWAYS ||
 240             newval == ZFS_SYNC_DISABLED);
 241 
 242         os->os_sync = newval;
 243         if (os->os_zil)
 244                 zil_set_sync(os->os_zil, newval);
 245 }
 246 
 247 static void
 248 redundant_metadata_changed_cb(void *arg, uint64_t newval)
 249 {
 250         objset_t *os = arg;
 251 
 252         /*
 253          * Inheritance and range checking should have been done by now.
 254          */
 255         ASSERT(newval == ZFS_REDUNDANT_METADATA_ALL ||
 256             newval == ZFS_REDUNDANT_METADATA_MOST);
 257 
 258         os->os_redundant_metadata = newval;
 259 }
 260 
 261 static void
 262 logbias_changed_cb(void *arg, uint64_t newval)
 263 {
 264         objset_t *os = arg;
 265 
 266         ASSERT(newval == ZFS_LOGBIAS_LATENCY ||
 267             newval == ZFS_LOGBIAS_THROUGHPUT);
 268         os->os_logbias = newval;
 269         if (os->os_zil)
 270                 zil_set_logbias(os->os_zil, newval);
 271 }
 272 
 273 static void
 274 recordsize_changed_cb(void *arg, uint64_t newval)
 275 {
 276         objset_t *os = arg;
 277 
 278         os->os_recordsize = newval;
 279 }
 280 
 281 void
 282 dmu_objset_byteswap(void *buf, size_t size)
 283 {
 284         objset_phys_t *osp = buf;
 285 
 286         ASSERT(size == OBJSET_OLD_PHYS_SIZE || size == sizeof (objset_phys_t));
 287         dnode_byteswap(&osp->os_meta_dnode);
 288         byteswap_uint64_array(&osp->os_zil_header, sizeof (zil_header_t));
 289         osp->os_type = BSWAP_64(osp->os_type);
 290         osp->os_flags = BSWAP_64(osp->os_flags);
 291         if (size == sizeof (objset_phys_t)) {
 292                 dnode_byteswap(&osp->os_userused_dnode);
 293                 dnode_byteswap(&osp->os_groupused_dnode);
 294         }
 295 }
 296 
 297 int
 298 dmu_objset_open_impl(spa_t *spa, dsl_dataset_t *ds, blkptr_t *bp,
 299     objset_t **osp)
 300 {
 301         objset_t *os;
 302         int i, err;
 303 
 304         ASSERT(ds == NULL || MUTEX_HELD(&ds->ds_opening_lock));
 305 
 306         os = kmem_zalloc(sizeof (objset_t), KM_SLEEP);
 307         os->os_dsl_dataset = ds;
 308         os->os_spa = spa;
 309         os->os_rootbp = bp;
 310         if (!BP_IS_HOLE(os->os_rootbp)) {
 311                 arc_flags_t aflags = ARC_FLAG_WAIT;
 312                 zbookmark_phys_t zb;
 313                 SET_BOOKMARK(&zb, ds ? ds->ds_object : DMU_META_OBJSET,
 314                     ZB_ROOT_OBJECT, ZB_ROOT_LEVEL, ZB_ROOT_BLKID);
 315 
 316                 if (DMU_OS_IS_L2CACHEABLE(os))
 317                         aflags |= ARC_FLAG_L2CACHE;
 318                 if (DMU_OS_IS_L2COMPRESSIBLE(os))
 319                         aflags |= ARC_FLAG_L2COMPRESS;
 320 
 321                 dprintf_bp(os->os_rootbp, "reading %s", "");
 322                 err = arc_read(NULL, spa, os->os_rootbp,
 323                     arc_getbuf_func, &os->os_phys_buf,
 324                     ZIO_PRIORITY_SYNC_READ, ZIO_FLAG_CANFAIL, &aflags, &zb);
 325                 if (err != 0) {
 326                         kmem_free(os, sizeof (objset_t));
 327                         /* convert checksum errors into IO errors */
 328                         if (err == ECKSUM)
 329                                 err = SET_ERROR(EIO);
 330                         return (err);
 331                 }
 332 
 333                 /* Increase the blocksize if we are permitted. */
 334                 if (spa_version(spa) >= SPA_VERSION_USERSPACE &&
 335                     arc_buf_size(os->os_phys_buf) < sizeof (objset_phys_t)) {
 336                         arc_buf_t *buf = arc_buf_alloc(spa,
 337                             sizeof (objset_phys_t), &os->os_phys_buf,
 338                             ARC_BUFC_METADATA);
 339                         bzero(buf->b_data, sizeof (objset_phys_t));
 340                         bcopy(os->os_phys_buf->b_data, buf->b_data,
 341                             arc_buf_size(os->os_phys_buf));
 342                         (void) arc_buf_remove_ref(os->os_phys_buf,
 343                             &os->os_phys_buf);
 344                         os->os_phys_buf = buf;
 345                 }
 346 
 347                 os->os_phys = os->os_phys_buf->b_data;
 348                 os->os_flags = os->os_phys->os_flags;
 349         } else {
 350                 int size = spa_version(spa) >= SPA_VERSION_USERSPACE ?
 351                     sizeof (objset_phys_t) : OBJSET_OLD_PHYS_SIZE;
 352                 os->os_phys_buf = arc_buf_alloc(spa, size,
 353                     &os->os_phys_buf, ARC_BUFC_METADATA);
 354                 os->os_phys = os->os_phys_buf->b_data;
 355                 bzero(os->os_phys, size);
 356         }
 357 
 358         /*
 359          * Note: the changed_cb will be called once before the register
 360          * func returns, thus changing the checksum/compression from the
 361          * default (fletcher2/off).  Snapshots don't need to know about
 362          * checksum/compression/copies.
 363          */
 364         if (ds != NULL) {
 365                 err = dsl_prop_register(ds,
 366                     zfs_prop_to_name(ZFS_PROP_PRIMARYCACHE),
 367                     primary_cache_changed_cb, os);
 368                 if (err == 0) {
 369                         err = dsl_prop_register(ds,
 370                             zfs_prop_to_name(ZFS_PROP_SECONDARYCACHE),
 371                             secondary_cache_changed_cb, os);
 372                 }
 373                 if (!ds->ds_is_snapshot) {
 374                         if (err == 0) {
 375                                 err = dsl_prop_register(ds,
 376                                     zfs_prop_to_name(ZFS_PROP_CHECKSUM),
 377                                     checksum_changed_cb, os);
 378                         }
 379                         if (err == 0) {
 380                                 err = dsl_prop_register(ds,
 381                                     zfs_prop_to_name(ZFS_PROP_COMPRESSION),
 382                                     compression_changed_cb, os);
 383                         }
 384                         if (err == 0) {
 385                                 err = dsl_prop_register(ds,
 386                                     zfs_prop_to_name(ZFS_PROP_COPIES),
 387                                     copies_changed_cb, os);
 388                         }
 389                         if (err == 0) {
 390                                 err = dsl_prop_register(ds,
 391                                     zfs_prop_to_name(ZFS_PROP_DEDUP),
 392                                     dedup_changed_cb, os);
 393                         }
 394                         if (err == 0) {
 395                                 err = dsl_prop_register(ds,
 396                                     zfs_prop_to_name(ZFS_PROP_LOGBIAS),
 397                                     logbias_changed_cb, os);
 398                         }
 399                         if (err == 0) {
 400                                 err = dsl_prop_register(ds,
 401                                     zfs_prop_to_name(ZFS_PROP_SYNC),
 402                                     sync_changed_cb, os);
 403                         }
 404                         if (err == 0) {
 405                                 err = dsl_prop_register(ds,
 406                                     zfs_prop_to_name(
 407                                     ZFS_PROP_REDUNDANT_METADATA),
 408                                     redundant_metadata_changed_cb, os);
 409                         }
 410                         if (err == 0) {
 411                                 err = dsl_prop_register(ds,
 412                                     zfs_prop_to_name(ZFS_PROP_RECORDSIZE),
 413                                     recordsize_changed_cb, os);
 414                         }
 415                 }
 416                 if (err != 0) {
 417                         VERIFY(arc_buf_remove_ref(os->os_phys_buf,
 418                             &os->os_phys_buf));
 419                         kmem_free(os, sizeof (objset_t));
 420                         return (err);
 421                 }
 422         } else {
 423                 /* It's the meta-objset. */
 424                 os->os_checksum = ZIO_CHECKSUM_FLETCHER_4;
 425                 os->os_compress = ZIO_COMPRESS_ON;
 426                 os->os_copies = spa_max_replication(spa);
 427                 os->os_dedup_checksum = ZIO_CHECKSUM_OFF;
 428                 os->os_dedup_verify = B_FALSE;
 429                 os->os_logbias = ZFS_LOGBIAS_LATENCY;
 430                 os->os_sync = ZFS_SYNC_STANDARD;
 431                 os->os_primary_cache = ZFS_CACHE_ALL;
 432                 os->os_secondary_cache = ZFS_CACHE_ALL;
 433         }
 434 
 435         if (ds == NULL || !ds->ds_is_snapshot)
 436                 os->os_zil_header = os->os_phys->os_zil_header;
 437         os->os_zil = zil_alloc(os, &os->os_zil_header);
 438 
 439         for (i = 0; i < TXG_SIZE; i++) {
 440                 list_create(&os->os_dirty_dnodes[i], sizeof (dnode_t),
 441                     offsetof(dnode_t, dn_dirty_link[i]));
 442                 list_create(&os->os_free_dnodes[i], sizeof (dnode_t),
 443                     offsetof(dnode_t, dn_dirty_link[i]));
 444         }
 445         list_create(&os->os_dnodes, sizeof (dnode_t),
 446             offsetof(dnode_t, dn_link));
 447         list_create(&os->os_downgraded_dbufs, sizeof (dmu_buf_impl_t),
 448             offsetof(dmu_buf_impl_t, db_link));
 449 
 450         mutex_init(&os->os_lock, NULL, MUTEX_DEFAULT, NULL);
 451         mutex_init(&os->os_obj_lock, NULL, MUTEX_DEFAULT, NULL);
 452         mutex_init(&os->os_user_ptr_lock, NULL, MUTEX_DEFAULT, NULL);
 453 
 454         dnode_special_open(os, &os->os_phys->os_meta_dnode,
 455             DMU_META_DNODE_OBJECT, &os->os_meta_dnode);
 456         if (arc_buf_size(os->os_phys_buf) >= sizeof (objset_phys_t)) {
 457                 dnode_special_open(os, &os->os_phys->os_userused_dnode,
 458                     DMU_USERUSED_OBJECT, &os->os_userused_dnode);
 459                 dnode_special_open(os, &os->os_phys->os_groupused_dnode,
 460                     DMU_GROUPUSED_OBJECT, &os->os_groupused_dnode);
 461         }
 462 
 463         *osp = os;
 464         return (0);
 465 }
 466 
 467 int
 468 dmu_objset_from_ds(dsl_dataset_t *ds, objset_t **osp)
 469 {
 470         int err = 0;
 471 
 472         mutex_enter(&ds->ds_opening_lock);
 473         if (ds->ds_objset == NULL) {
 474                 objset_t *os;
 475                 err = dmu_objset_open_impl(dsl_dataset_get_spa(ds),
 476                     ds, dsl_dataset_get_blkptr(ds), &os);
 477 
 478                 if (err == 0) {
 479                         mutex_enter(&ds->ds_lock);
 480                         ASSERT(ds->ds_objset == NULL);
 481                         ds->ds_objset = os;
 482                         mutex_exit(&ds->ds_lock);
 483                 }
 484         }
 485         *osp = ds->ds_objset;
 486         mutex_exit(&ds->ds_opening_lock);
 487         return (err);
 488 }
 489 
 490 /*
 491  * Holds the pool while the objset is held.  Therefore only one objset
 492  * can be held at a time.
 493  */
 494 int
 495 dmu_objset_hold(const char *name, void *tag, objset_t **osp)
 496 {
 497         dsl_pool_t *dp;
 498         dsl_dataset_t *ds;
 499         int err;
 500 
 501         err = dsl_pool_hold(name, tag, &dp);
 502         if (err != 0)
 503                 return (err);
 504         err = dsl_dataset_hold(dp, name, tag, &ds);
 505         if (err != 0) {
 506                 dsl_pool_rele(dp, tag);
 507                 return (err);
 508         }
 509 
 510         err = dmu_objset_from_ds(ds, osp);
 511         if (err != 0) {
 512                 dsl_dataset_rele(ds, tag);
 513                 dsl_pool_rele(dp, tag);
 514         }
 515 
 516         return (err);
 517 }
 518 
 519 static int
 520 dmu_objset_own_impl(dsl_dataset_t *ds, dmu_objset_type_t type,
 521     boolean_t readonly, void *tag, objset_t **osp)
 522 {
 523         int err;
 524 
 525         err = dmu_objset_from_ds(ds, osp);
 526         if (err != 0) {
 527                 dsl_dataset_disown(ds, tag);
 528         } else if (type != DMU_OST_ANY && type != (*osp)->os_phys->os_type) {
 529                 dsl_dataset_disown(ds, tag);
 530                 return (SET_ERROR(EINVAL));
 531         } else if (!readonly && dsl_dataset_is_snapshot(ds)) {
 532                 dsl_dataset_disown(ds, tag);
 533                 return (SET_ERROR(EROFS));
 534         }
 535         return (err);
 536 }
 537 
 538 /*
 539  * dsl_pool must not be held when this is called.
 540  * Upon successful return, there will be a longhold on the dataset,
 541  * and the dsl_pool will not be held.
 542  */
 543 int
 544 dmu_objset_own(const char *name, dmu_objset_type_t type,
 545     boolean_t readonly, void *tag, objset_t **osp)
 546 {
 547         dsl_pool_t *dp;
 548         dsl_dataset_t *ds;
 549         int err;
 550 
 551         err = dsl_pool_hold(name, FTAG, &dp);
 552         if (err != 0)
 553                 return (err);
 554         err = dsl_dataset_own(dp, name, tag, &ds);
 555         if (err != 0) {
 556                 dsl_pool_rele(dp, FTAG);
 557                 return (err);
 558         }
 559         err = dmu_objset_own_impl(ds, type, readonly, tag, osp);
 560         dsl_pool_rele(dp, FTAG);
 561 
 562         return (err);
 563 }
 564 
 565 int
 566 dmu_objset_own_obj(dsl_pool_t *dp, uint64_t obj, dmu_objset_type_t type,
 567     boolean_t readonly, void *tag, objset_t **osp)
 568 {
 569         dsl_dataset_t *ds;
 570         int err;
 571 
 572         err = dsl_dataset_own_obj(dp, obj, tag, &ds);
 573         if (err != 0)
 574                 return (err);
 575 
 576         return (dmu_objset_own_impl(ds, type, readonly, tag, osp));
 577 }
 578 
 579 void
 580 dmu_objset_rele(objset_t *os, void *tag)
 581 {
 582         dsl_pool_t *dp = dmu_objset_pool(os);
 583         dsl_dataset_rele(os->os_dsl_dataset, tag);
 584         dsl_pool_rele(dp, tag);
 585 }
 586 
 587 /*
 588  * When we are called, os MUST refer to an objset associated with a dataset
 589  * that is owned by 'tag'; that is, is held and long held by 'tag' and ds_owner
 590  * == tag.  We will then release and reacquire ownership of the dataset while
 591  * holding the pool config_rwlock to avoid intervening namespace or ownership
 592  * changes may occur.
 593  *
 594  * This exists solely to accommodate zfs_ioc_userspace_upgrade()'s desire to
 595  * release the hold on its dataset and acquire a new one on the dataset of the
 596  * same name so that it can be partially torn down and reconstructed.
 597  */
 598 void
 599 dmu_objset_refresh_ownership(objset_t *os, void *tag)
 600 {
 601         dsl_pool_t *dp;
 602         dsl_dataset_t *ds, *newds;
 603         char name[MAXNAMELEN];
 604 
 605         ds = os->os_dsl_dataset;
 606         VERIFY3P(ds, !=, NULL);
 607         VERIFY3P(ds->ds_owner, ==, tag);
 608         VERIFY(dsl_dataset_long_held(ds));
 609 
 610         dsl_dataset_name(ds, name);
 611         dp = dmu_objset_pool(os);
 612         dsl_pool_config_enter(dp, FTAG);
 613         dmu_objset_disown(os, tag);
 614         VERIFY0(dsl_dataset_own(dp, name, tag, &newds));
 615         VERIFY3P(newds, ==, os->os_dsl_dataset);
 616         dsl_pool_config_exit(dp, FTAG);
 617 }
 618 
 619 void
 620 dmu_objset_disown(objset_t *os, void *tag)
 621 {
 622         dsl_dataset_disown(os->os_dsl_dataset, tag);
 623 }
 624 
 625 void
 626 dmu_objset_evict_dbufs(objset_t *os)
 627 {
 628         dnode_t dn_marker;
 629         dnode_t *dn;
 630 
 631         mutex_enter(&os->os_lock);
 632         dn = list_head(&os->os_dnodes);
 633         while (dn != NULL) {
 634                 /*
 635                  * Skip dnodes without holds.  We have to do this dance
 636                  * because dnode_add_ref() only works if there is already a
 637                  * hold.  If the dnode has no holds, then it has no dbufs.
 638                  */
 639                 if (dnode_add_ref(dn, FTAG)) {
 640                         list_insert_after(&os->os_dnodes, dn, &dn_marker);
 641                         mutex_exit(&os->os_lock);
 642 
 643                         dnode_evict_dbufs(dn);
 644                         dnode_rele(dn, FTAG);
 645 
 646                         mutex_enter(&os->os_lock);
 647                         dn = list_next(&os->os_dnodes, &dn_marker);
 648                         list_remove(&os->os_dnodes, &dn_marker);
 649                 } else {
 650                         dn = list_next(&os->os_dnodes, dn);
 651                 }
 652         }
 653         mutex_exit(&os->os_lock);
 654 
 655         if (DMU_USERUSED_DNODE(os) != NULL) {
 656                 dnode_evict_dbufs(DMU_GROUPUSED_DNODE(os));
 657                 dnode_evict_dbufs(DMU_USERUSED_DNODE(os));
 658         }
 659         dnode_evict_dbufs(DMU_META_DNODE(os));
 660 }
 661 
 662 /*
 663  * Objset eviction processing is split into into two pieces.
 664  * The first marks the objset as evicting, evicts any dbufs that
 665  * have a refcount of zero, and then queues up the objset for the
 666  * second phase of eviction.  Once os->os_dnodes has been cleared by
 667  * dnode_buf_pageout()->dnode_destroy(), the second phase is executed.
 668  * The second phase closes the special dnodes, dequeues the objset from
 669  * the list of those undergoing eviction, and finally frees the objset.
 670  *
 671  * NOTE: Due to asynchronous eviction processing (invocation of
 672  *       dnode_buf_pageout()), it is possible for the meta dnode for the
 673  *       objset to have no holds even though os->os_dnodes is not empty.
 674  */
 675 void
 676 dmu_objset_evict(objset_t *os)
 677 {
 678         dsl_dataset_t *ds = os->os_dsl_dataset;
 679 
 680         for (int t = 0; t < TXG_SIZE; t++)
 681                 ASSERT(!dmu_objset_is_dirty(os, t));
 682 
 683         if (ds) {
 684                 if (!ds->ds_is_snapshot) {
 685                         VERIFY0(dsl_prop_unregister(ds,
 686                             zfs_prop_to_name(ZFS_PROP_CHECKSUM),
 687                             checksum_changed_cb, os));
 688                         VERIFY0(dsl_prop_unregister(ds,
 689                             zfs_prop_to_name(ZFS_PROP_COMPRESSION),
 690                             compression_changed_cb, os));
 691                         VERIFY0(dsl_prop_unregister(ds,
 692                             zfs_prop_to_name(ZFS_PROP_COPIES),
 693                             copies_changed_cb, os));
 694                         VERIFY0(dsl_prop_unregister(ds,
 695                             zfs_prop_to_name(ZFS_PROP_DEDUP),
 696                             dedup_changed_cb, os));
 697                         VERIFY0(dsl_prop_unregister(ds,
 698                             zfs_prop_to_name(ZFS_PROP_LOGBIAS),
 699                             logbias_changed_cb, os));
 700                         VERIFY0(dsl_prop_unregister(ds,
 701                             zfs_prop_to_name(ZFS_PROP_SYNC),
 702                             sync_changed_cb, os));
 703                         VERIFY0(dsl_prop_unregister(ds,
 704                             zfs_prop_to_name(ZFS_PROP_REDUNDANT_METADATA),
 705                             redundant_metadata_changed_cb, os));
 706                         VERIFY0(dsl_prop_unregister(ds,
 707                             zfs_prop_to_name(ZFS_PROP_RECORDSIZE),
 708                             recordsize_changed_cb, os));
 709                 }
 710                 VERIFY0(dsl_prop_unregister(ds,
 711                     zfs_prop_to_name(ZFS_PROP_PRIMARYCACHE),
 712                     primary_cache_changed_cb, os));
 713                 VERIFY0(dsl_prop_unregister(ds,
 714                     zfs_prop_to_name(ZFS_PROP_SECONDARYCACHE),
 715                     secondary_cache_changed_cb, os));
 716         }
 717 
 718         if (os->os_sa)
 719                 sa_tear_down(os);
 720 
 721         os->os_evicting = B_TRUE;
 722         dmu_objset_evict_dbufs(os);
 723 
 724         mutex_enter(&os->os_lock);
 725         spa_evicting_os_register(os->os_spa, os);
 726         if (list_is_empty(&os->os_dnodes)) {
 727                 mutex_exit(&os->os_lock);
 728                 dmu_objset_evict_done(os);
 729         } else {
 730                 mutex_exit(&os->os_lock);
 731         }
 732 }
 733 
 734 void
 735 dmu_objset_evict_done(objset_t *os)
 736 {
 737         ASSERT3P(list_head(&os->os_dnodes), ==, NULL);
 738 
 739         dnode_special_close(&os->os_meta_dnode);
 740         if (DMU_USERUSED_DNODE(os)) {
 741                 dnode_special_close(&os->os_userused_dnode);
 742                 dnode_special_close(&os->os_groupused_dnode);
 743         }
 744         zil_free(os->os_zil);
 745 
 746         VERIFY(arc_buf_remove_ref(os->os_phys_buf, &os->os_phys_buf));
 747 
 748         /*
 749          * This is a barrier to prevent the objset from going away in
 750          * dnode_move() until we can safely ensure that the objset is still in
 751          * use. We consider the objset valid before the barrier and invalid
 752          * after the barrier.
 753          */
 754         rw_enter(&os_lock, RW_READER);
 755         rw_exit(&os_lock);
 756 
 757         mutex_destroy(&os->os_lock);
 758         mutex_destroy(&os->os_obj_lock);
 759         mutex_destroy(&os->os_user_ptr_lock);
 760         spa_evicting_os_deregister(os->os_spa, os);
 761         kmem_free(os, sizeof (objset_t));
 762 }
 763 
 764 timestruc_t
 765 dmu_objset_snap_cmtime(objset_t *os)
 766 {
 767         return (dsl_dir_snap_cmtime(os->os_dsl_dataset->ds_dir));
 768 }
 769 
 770 /* called from dsl for meta-objset */
 771 objset_t *
 772 dmu_objset_create_impl(spa_t *spa, dsl_dataset_t *ds, blkptr_t *bp,
 773     dmu_objset_type_t type, dmu_tx_t *tx)
 774 {
 775         objset_t *os;
 776         dnode_t *mdn;
 777 
 778         ASSERT(dmu_tx_is_syncing(tx));
 779 
 780         if (ds != NULL)
 781                 VERIFY0(dmu_objset_from_ds(ds, &os));
 782         else
 783                 VERIFY0(dmu_objset_open_impl(spa, NULL, bp, &os));
 784 
 785         mdn = DMU_META_DNODE(os);
 786 
 787         dnode_allocate(mdn, DMU_OT_DNODE, 1 << DNODE_BLOCK_SHIFT,
 788             DN_MAX_INDBLKSHIFT, DMU_OT_NONE, 0, tx);
 789 
 790         /*
 791          * We don't want to have to increase the meta-dnode's nlevels
 792          * later, because then we could do it in quescing context while
 793          * we are also accessing it in open context.
 794          *
 795          * This precaution is not necessary for the MOS (ds == NULL),
 796          * because the MOS is only updated in syncing context.
 797          * This is most fortunate: the MOS is the only objset that
 798          * needs to be synced multiple times as spa_sync() iterates
 799          * to convergence, so minimizing its dn_nlevels matters.
 800          */
 801         if (ds != NULL) {
 802                 int levels = 1;
 803 
 804                 /*
 805                  * Determine the number of levels necessary for the meta-dnode
 806                  * to contain DN_MAX_OBJECT dnodes.
 807                  */
 808                 while ((uint64_t)mdn->dn_nblkptr << (mdn->dn_datablkshift +
 809                     (levels - 1) * (mdn->dn_indblkshift - SPA_BLKPTRSHIFT)) <
 810                     DN_MAX_OBJECT * sizeof (dnode_phys_t))
 811                         levels++;
 812 
 813                 mdn->dn_next_nlevels[tx->tx_txg & TXG_MASK] =
 814                     mdn->dn_nlevels = levels;
 815         }
 816 
 817         ASSERT(type != DMU_OST_NONE);
 818         ASSERT(type != DMU_OST_ANY);
 819         ASSERT(type < DMU_OST_NUMTYPES);
 820         os->os_phys->os_type = type;
 821         if (dmu_objset_userused_enabled(os)) {
 822                 os->os_phys->os_flags |= OBJSET_FLAG_USERACCOUNTING_COMPLETE;
 823                 os->os_flags = os->os_phys->os_flags;
 824         }
 825 
 826         dsl_dataset_dirty(ds, tx);
 827 
 828         return (os);
 829 }
 830 
 831 typedef struct dmu_objset_create_arg {
 832         const char *doca_name;
 833         cred_t *doca_cred;
 834         void (*doca_userfunc)(objset_t *os, void *arg,
 835             cred_t *cr, dmu_tx_t *tx);
 836         void *doca_userarg;
 837         dmu_objset_type_t doca_type;
 838         uint64_t doca_flags;
 839 } dmu_objset_create_arg_t;
 840 
 841 /*ARGSUSED*/
 842 static int
 843 dmu_objset_create_check(void *arg, dmu_tx_t *tx)
 844 {
 845         dmu_objset_create_arg_t *doca = arg;
 846         dsl_pool_t *dp = dmu_tx_pool(tx);
 847         dsl_dir_t *pdd;
 848         const char *tail;
 849         int error;
 850 
 851         if (strchr(doca->doca_name, '@') != NULL)
 852                 return (SET_ERROR(EINVAL));
 853 
 854         error = dsl_dir_hold(dp, doca->doca_name, FTAG, &pdd, &tail);
 855         if (error != 0)
 856                 return (error);
 857         if (tail == NULL) {
 858                 dsl_dir_rele(pdd, FTAG);
 859                 return (SET_ERROR(EEXIST));
 860         }
 861         error = dsl_fs_ss_limit_check(pdd, 1, ZFS_PROP_FILESYSTEM_LIMIT, NULL,
 862             doca->doca_cred);
 863         dsl_dir_rele(pdd, FTAG);
 864 
 865         return (error);
 866 }
 867 
 868 static void
 869 dmu_objset_create_sync(void *arg, dmu_tx_t *tx)
 870 {
 871         dmu_objset_create_arg_t *doca = arg;
 872         dsl_pool_t *dp = dmu_tx_pool(tx);
 873         dsl_dir_t *pdd;
 874         const char *tail;
 875         dsl_dataset_t *ds;
 876         uint64_t obj;
 877         blkptr_t *bp;
 878         objset_t *os;
 879 
 880         VERIFY0(dsl_dir_hold(dp, doca->doca_name, FTAG, &pdd, &tail));
 881 
 882         obj = dsl_dataset_create_sync(pdd, tail, NULL, doca->doca_flags,
 883             doca->doca_cred, tx);
 884 
 885         VERIFY0(dsl_dataset_hold_obj(pdd->dd_pool, obj, FTAG, &ds));
 886         bp = dsl_dataset_get_blkptr(ds);
 887         os = dmu_objset_create_impl(pdd->dd_pool->dp_spa,
 888             ds, bp, doca->doca_type, tx);
 889 
 890         if (doca->doca_userfunc != NULL) {
 891                 doca->doca_userfunc(os, doca->doca_userarg,
 892                     doca->doca_cred, tx);
 893         }
 894 
 895         spa_history_log_internal_ds(ds, "create", tx, "");
 896         dsl_dataset_rele(ds, FTAG);
 897         dsl_dir_rele(pdd, FTAG);
 898 }
 899 
 900 int
 901 dmu_objset_create(const char *name, dmu_objset_type_t type, uint64_t flags,
 902     void (*func)(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx), void *arg)
 903 {
 904         dmu_objset_create_arg_t doca;
 905 
 906         doca.doca_name = name;
 907         doca.doca_cred = CRED();
 908         doca.doca_flags = flags;
 909         doca.doca_userfunc = func;
 910         doca.doca_userarg = arg;
 911         doca.doca_type = type;
 912 
 913         return (dsl_sync_task(name,
 914             dmu_objset_create_check, dmu_objset_create_sync, &doca,
 915             5, ZFS_SPACE_CHECK_NORMAL));
 916 }
 917 
 918 typedef struct dmu_objset_clone_arg {
 919         const char *doca_clone;
 920         const char *doca_origin;
 921         cred_t *doca_cred;
 922 } dmu_objset_clone_arg_t;
 923 
 924 /*ARGSUSED*/
 925 static int
 926 dmu_objset_clone_check(void *arg, dmu_tx_t *tx)
 927 {
 928         dmu_objset_clone_arg_t *doca = arg;
 929         dsl_dir_t *pdd;
 930         const char *tail;
 931         int error;
 932         dsl_dataset_t *origin;
 933         dsl_pool_t *dp = dmu_tx_pool(tx);
 934 
 935         if (strchr(doca->doca_clone, '@') != NULL)
 936                 return (SET_ERROR(EINVAL));
 937 
 938         error = dsl_dir_hold(dp, doca->doca_clone, FTAG, &pdd, &tail);
 939         if (error != 0)
 940                 return (error);
 941         if (tail == NULL) {
 942                 dsl_dir_rele(pdd, FTAG);
 943                 return (SET_ERROR(EEXIST));
 944         }
 945 
 946         error = dsl_fs_ss_limit_check(pdd, 1, ZFS_PROP_FILESYSTEM_LIMIT, NULL,
 947             doca->doca_cred);
 948         if (error != 0) {
 949                 dsl_dir_rele(pdd, FTAG);
 950                 return (SET_ERROR(EDQUOT));
 951         }
 952         dsl_dir_rele(pdd, FTAG);
 953 
 954         error = dsl_dataset_hold(dp, doca->doca_origin, FTAG, &origin);
 955         if (error != 0)
 956                 return (error);
 957 
 958         /* You can only clone snapshots, not the head datasets. */
 959         if (!origin->ds_is_snapshot) {
 960                 dsl_dataset_rele(origin, FTAG);
 961                 return (SET_ERROR(EINVAL));
 962         }
 963         dsl_dataset_rele(origin, FTAG);
 964 
 965         return (0);
 966 }
 967 
 968 static void
 969 dmu_objset_clone_sync(void *arg, dmu_tx_t *tx)
 970 {
 971         dmu_objset_clone_arg_t *doca = arg;
 972         dsl_pool_t *dp = dmu_tx_pool(tx);
 973         dsl_dir_t *pdd;
 974         const char *tail;
 975         dsl_dataset_t *origin, *ds;
 976         uint64_t obj;
 977         char namebuf[MAXNAMELEN];
 978 
 979         VERIFY0(dsl_dir_hold(dp, doca->doca_clone, FTAG, &pdd, &tail));
 980         VERIFY0(dsl_dataset_hold(dp, doca->doca_origin, FTAG, &origin));
 981 
 982         obj = dsl_dataset_create_sync(pdd, tail, origin, 0,
 983             doca->doca_cred, tx);
 984 
 985         VERIFY0(dsl_dataset_hold_obj(pdd->dd_pool, obj, FTAG, &ds));
 986         dsl_dataset_name(origin, namebuf);
 987         spa_history_log_internal_ds(ds, "clone", tx,
 988             "origin=%s (%llu)", namebuf, origin->ds_object);
 989         dsl_dataset_rele(ds, FTAG);
 990         dsl_dataset_rele(origin, FTAG);
 991         dsl_dir_rele(pdd, FTAG);
 992 }
 993 
 994 int
 995 dmu_objset_clone(const char *clone, const char *origin)
 996 {
 997         dmu_objset_clone_arg_t doca;
 998 
 999         doca.doca_clone = clone;
1000         doca.doca_origin = origin;
1001         doca.doca_cred = CRED();
1002 
1003         return (dsl_sync_task(clone,
1004             dmu_objset_clone_check, dmu_objset_clone_sync, &doca,
1005             5, ZFS_SPACE_CHECK_NORMAL));
1006 }
1007 
1008 int
1009 dmu_objset_snapshot_one(const char *fsname, const char *snapname)
1010 {
1011         int err;
1012         char *longsnap = kmem_asprintf("%s@%s", fsname, snapname);
1013         nvlist_t *snaps = fnvlist_alloc();
1014 
1015         fnvlist_add_boolean(snaps, longsnap);
1016         strfree(longsnap);
1017         err = dsl_dataset_snapshot(snaps, NULL, NULL);
1018         fnvlist_free(snaps);
1019         return (err);
1020 }
1021 
1022 static void
1023 dmu_objset_sync_dnodes(list_t *list, list_t *newlist, dmu_tx_t *tx)
1024 {
1025         dnode_t *dn;
1026 
1027         while (dn = list_head(list)) {
1028                 ASSERT(dn->dn_object != DMU_META_DNODE_OBJECT);
1029                 ASSERT(dn->dn_dbuf->db_data_pending);
1030                 /*
1031                  * Initialize dn_zio outside dnode_sync() because the
1032                  * meta-dnode needs to set it ouside dnode_sync().
1033                  */
1034                 dn->dn_zio = dn->dn_dbuf->db_data_pending->dr_zio;
1035                 ASSERT(dn->dn_zio);
1036 
1037                 ASSERT3U(dn->dn_nlevels, <=, DN_MAX_LEVELS);
1038                 list_remove(list, dn);
1039 
1040                 if (newlist) {
1041                         (void) dnode_add_ref(dn, newlist);
1042                         list_insert_tail(newlist, dn);
1043                 }
1044 
1045                 dnode_sync(dn, tx);
1046         }
1047 }
1048 
1049 /* ARGSUSED */
1050 static void
1051 dmu_objset_write_ready(zio_t *zio, arc_buf_t *abuf, void *arg)
1052 {
1053         blkptr_t *bp = zio->io_bp;
1054         objset_t *os = arg;
1055         dnode_phys_t *dnp = &os->os_phys->os_meta_dnode;
1056 
1057         ASSERT(!BP_IS_EMBEDDED(bp));
1058         ASSERT3P(bp, ==, os->os_rootbp);
1059         ASSERT3U(BP_GET_TYPE(bp), ==, DMU_OT_OBJSET);
1060         ASSERT0(BP_GET_LEVEL(bp));
1061 
1062         /*
1063          * Update rootbp fill count: it should be the number of objects
1064          * allocated in the object set (not counting the "special"
1065          * objects that are stored in the objset_phys_t -- the meta
1066          * dnode and user/group accounting objects).
1067          */
1068         bp->blk_fill = 0;
1069         for (int i = 0; i < dnp->dn_nblkptr; i++)
1070                 bp->blk_fill += BP_GET_FILL(&dnp->dn_blkptr[i]);
1071 }
1072 
1073 /* ARGSUSED */
1074 static void
1075 dmu_objset_write_done(zio_t *zio, arc_buf_t *abuf, void *arg)
1076 {
1077         blkptr_t *bp = zio->io_bp;
1078         blkptr_t *bp_orig = &zio->io_bp_orig;
1079         objset_t *os = arg;
1080 
1081         if (zio->io_flags & ZIO_FLAG_IO_REWRITE) {
1082                 ASSERT(BP_EQUAL(bp, bp_orig));
1083         } else {
1084                 dsl_dataset_t *ds = os->os_dsl_dataset;
1085                 dmu_tx_t *tx = os->os_synctx;
1086 
1087                 (void) dsl_dataset_block_kill(ds, bp_orig, tx, B_TRUE);
1088                 dsl_dataset_block_born(ds, bp, tx);
1089         }
1090 }
1091 
1092 /* called from dsl */
1093 void
1094 dmu_objset_sync(objset_t *os, zio_t *pio, dmu_tx_t *tx)
1095 {
1096         int txgoff;
1097         zbookmark_phys_t zb;
1098         zio_prop_t zp;
1099         zio_t *zio;
1100         list_t *list;
1101         list_t *newlist = NULL;
1102         dbuf_dirty_record_t *dr;
1103 
1104         dprintf_ds(os->os_dsl_dataset, "txg=%llu\n", tx->tx_txg);
1105 
1106         ASSERT(dmu_tx_is_syncing(tx));
1107         /* XXX the write_done callback should really give us the tx... */
1108         os->os_synctx = tx;
1109 
1110         if (os->os_dsl_dataset == NULL) {
1111                 /*
1112                  * This is the MOS.  If we have upgraded,
1113                  * spa_max_replication() could change, so reset
1114                  * os_copies here.
1115                  */
1116                 os->os_copies = spa_max_replication(os->os_spa);
1117         }
1118 
1119         /*
1120          * Create the root block IO
1121          */
1122         SET_BOOKMARK(&zb, os->os_dsl_dataset ?
1123             os->os_dsl_dataset->ds_object : DMU_META_OBJSET,
1124             ZB_ROOT_OBJECT, ZB_ROOT_LEVEL, ZB_ROOT_BLKID);
1125         arc_release(os->os_phys_buf, &os->os_phys_buf);
1126 
1127         dmu_write_policy(os, NULL, 0, 0, &zp);
1128 
1129         zio = arc_write(pio, os->os_spa, tx->tx_txg,
1130             os->os_rootbp, os->os_phys_buf, DMU_OS_IS_L2CACHEABLE(os),
1131             DMU_OS_IS_L2COMPRESSIBLE(os), &zp, dmu_objset_write_ready,
1132             NULL, dmu_objset_write_done, os, ZIO_PRIORITY_ASYNC_WRITE,
1133             ZIO_FLAG_MUSTSUCCEED, &zb);
1134 
1135         /*
1136          * Sync special dnodes - the parent IO for the sync is the root block
1137          */
1138         DMU_META_DNODE(os)->dn_zio = zio;
1139         dnode_sync(DMU_META_DNODE(os), tx);
1140 
1141         os->os_phys->os_flags = os->os_flags;
1142 
1143         if (DMU_USERUSED_DNODE(os) &&
1144             DMU_USERUSED_DNODE(os)->dn_type != DMU_OT_NONE) {
1145                 DMU_USERUSED_DNODE(os)->dn_zio = zio;
1146                 dnode_sync(DMU_USERUSED_DNODE(os), tx);
1147                 DMU_GROUPUSED_DNODE(os)->dn_zio = zio;
1148                 dnode_sync(DMU_GROUPUSED_DNODE(os), tx);
1149         }
1150 
1151         txgoff = tx->tx_txg & TXG_MASK;
1152 
1153         if (dmu_objset_userused_enabled(os)) {
1154                 newlist = &os->os_synced_dnodes;
1155                 /*
1156                  * We must create the list here because it uses the
1157                  * dn_dirty_link[] of this txg.
1158                  */
1159                 list_create(newlist, sizeof (dnode_t),
1160                     offsetof(dnode_t, dn_dirty_link[txgoff]));
1161         }
1162 
1163         dmu_objset_sync_dnodes(&os->os_free_dnodes[txgoff], newlist, tx);
1164         dmu_objset_sync_dnodes(&os->os_dirty_dnodes[txgoff], newlist, tx);
1165 
1166         list = &DMU_META_DNODE(os)->dn_dirty_records[txgoff];
1167         while (dr = list_head(list)) {
1168                 ASSERT0(dr->dr_dbuf->db_level);
1169                 list_remove(list, dr);
1170                 if (dr->dr_zio)
1171                         zio_nowait(dr->dr_zio);
1172         }
1173         /*
1174          * Free intent log blocks up to this tx.
1175          */
1176         zil_sync(os->os_zil, tx);
1177         os->os_phys->os_zil_header = os->os_zil_header;
1178         zio_nowait(zio);
1179 }
1180 
1181 boolean_t
1182 dmu_objset_is_dirty(objset_t *os, uint64_t txg)
1183 {
1184         return (!list_is_empty(&os->os_dirty_dnodes[txg & TXG_MASK]) ||
1185             !list_is_empty(&os->os_free_dnodes[txg & TXG_MASK]));
1186 }
1187 
1188 static objset_used_cb_t *used_cbs[DMU_OST_NUMTYPES];
1189 
1190 void
1191 dmu_objset_register_type(dmu_objset_type_t ost, objset_used_cb_t *cb)
1192 {
1193         used_cbs[ost] = cb;
1194 }
1195 
1196 boolean_t
1197 dmu_objset_userused_enabled(objset_t *os)
1198 {
1199         return (spa_version(os->os_spa) >= SPA_VERSION_USERSPACE &&
1200             used_cbs[os->os_phys->os_type] != NULL &&
1201             DMU_USERUSED_DNODE(os) != NULL);
1202 }
1203 
1204 static void
1205 do_userquota_update(objset_t *os, uint64_t used, uint64_t flags,
1206     uint64_t user, uint64_t group, boolean_t subtract, dmu_tx_t *tx)
1207 {
1208         if ((flags & DNODE_FLAG_USERUSED_ACCOUNTED)) {
1209                 int64_t delta = DNODE_SIZE + used;
1210                 if (subtract)
1211                         delta = -delta;
1212                 VERIFY3U(0, ==, zap_increment_int(os, DMU_USERUSED_OBJECT,
1213                     user, delta, tx));
1214                 VERIFY3U(0, ==, zap_increment_int(os, DMU_GROUPUSED_OBJECT,
1215                     group, delta, tx));
1216         }
1217 }
1218 
1219 void
1220 dmu_objset_do_userquota_updates(objset_t *os, dmu_tx_t *tx)
1221 {
1222         dnode_t *dn;
1223         list_t *list = &os->os_synced_dnodes;
1224 
1225         ASSERT(list_head(list) == NULL || dmu_objset_userused_enabled(os));
1226 
1227         while (dn = list_head(list)) {
1228                 int flags;
1229                 ASSERT(!DMU_OBJECT_IS_SPECIAL(dn->dn_object));
1230                 ASSERT(dn->dn_phys->dn_type == DMU_OT_NONE ||
1231                     dn->dn_phys->dn_flags &
1232                     DNODE_FLAG_USERUSED_ACCOUNTED);
1233 
1234                 /* Allocate the user/groupused objects if necessary. */
1235                 if (DMU_USERUSED_DNODE(os)->dn_type == DMU_OT_NONE) {
1236                         VERIFY(0 == zap_create_claim(os,
1237                             DMU_USERUSED_OBJECT,
1238                             DMU_OT_USERGROUP_USED, DMU_OT_NONE, 0, tx));
1239                         VERIFY(0 == zap_create_claim(os,
1240                             DMU_GROUPUSED_OBJECT,
1241                             DMU_OT_USERGROUP_USED, DMU_OT_NONE, 0, tx));
1242                 }
1243 
1244                 /*
1245                  * We intentionally modify the zap object even if the
1246                  * net delta is zero.  Otherwise
1247                  * the block of the zap obj could be shared between
1248                  * datasets but need to be different between them after
1249                  * a bprewrite.
1250                  */
1251 
1252                 flags = dn->dn_id_flags;
1253                 ASSERT(flags);
1254                 if (flags & DN_ID_OLD_EXIST)  {
1255                         do_userquota_update(os, dn->dn_oldused, dn->dn_oldflags,
1256                             dn->dn_olduid, dn->dn_oldgid, B_TRUE, tx);
1257                 }
1258                 if (flags & DN_ID_NEW_EXIST) {
1259                         do_userquota_update(os, DN_USED_BYTES(dn->dn_phys),
1260                             dn->dn_phys->dn_flags,  dn->dn_newuid,
1261                             dn->dn_newgid, B_FALSE, tx);
1262                 }
1263 
1264                 mutex_enter(&dn->dn_mtx);
1265                 dn->dn_oldused = 0;
1266                 dn->dn_oldflags = 0;
1267                 if (dn->dn_id_flags & DN_ID_NEW_EXIST) {
1268                         dn->dn_olduid = dn->dn_newuid;
1269                         dn->dn_oldgid = dn->dn_newgid;
1270                         dn->dn_id_flags |= DN_ID_OLD_EXIST;
1271                         if (dn->dn_bonuslen == 0)
1272                                 dn->dn_id_flags |= DN_ID_CHKED_SPILL;
1273                         else
1274                                 dn->dn_id_flags |= DN_ID_CHKED_BONUS;
1275                 }
1276                 dn->dn_id_flags &= ~(DN_ID_NEW_EXIST);
1277                 mutex_exit(&dn->dn_mtx);
1278 
1279                 list_remove(list, dn);
1280                 dnode_rele(dn, list);
1281         }
1282 }
1283 
1284 /*
1285  * Returns a pointer to data to find uid/gid from
1286  *
1287  * If a dirty record for transaction group that is syncing can't
1288  * be found then NULL is returned.  In the NULL case it is assumed
1289  * the uid/gid aren't changing.
1290  */
1291 static void *
1292 dmu_objset_userquota_find_data(dmu_buf_impl_t *db, dmu_tx_t *tx)
1293 {
1294         dbuf_dirty_record_t *dr, **drp;
1295         void *data;
1296 
1297         if (db->db_dirtycnt == 0)
1298                 return (db->db.db_data);  /* Nothing is changing */
1299 
1300         for (drp = &db->db_last_dirty; (dr = *drp) != NULL; drp = &dr->dr_next)
1301                 if (dr->dr_txg == tx->tx_txg)
1302                         break;
1303 
1304         if (dr == NULL) {
1305                 data = NULL;
1306         } else {
1307                 dnode_t *dn;
1308 
1309                 DB_DNODE_ENTER(dr->dr_dbuf);
1310                 dn = DB_DNODE(dr->dr_dbuf);
1311 
1312                 if (dn->dn_bonuslen == 0 &&
1313                     dr->dr_dbuf->db_blkid == DMU_SPILL_BLKID)
1314                         data = dr->dt.dl.dr_data->b_data;
1315                 else
1316                         data = dr->dt.dl.dr_data;
1317 
1318                 DB_DNODE_EXIT(dr->dr_dbuf);
1319         }
1320 
1321         return (data);
1322 }
1323 
1324 void
1325 dmu_objset_userquota_get_ids(dnode_t *dn, boolean_t before, dmu_tx_t *tx)
1326 {
1327         objset_t *os = dn->dn_objset;
1328         void *data = NULL;
1329         dmu_buf_impl_t *db = NULL;
1330         uint64_t *user = NULL;
1331         uint64_t *group = NULL;
1332         int flags = dn->dn_id_flags;
1333         int error;
1334         boolean_t have_spill = B_FALSE;
1335 
1336         if (!dmu_objset_userused_enabled(dn->dn_objset))
1337                 return;
1338 
1339         if (before && (flags & (DN_ID_CHKED_BONUS|DN_ID_OLD_EXIST|
1340             DN_ID_CHKED_SPILL)))
1341                 return;
1342 
1343         if (before && dn->dn_bonuslen != 0)
1344                 data = DN_BONUS(dn->dn_phys);
1345         else if (!before && dn->dn_bonuslen != 0) {
1346                 if (dn->dn_bonus) {
1347                         db = dn->dn_bonus;
1348                         mutex_enter(&db->db_mtx);
1349                         data = dmu_objset_userquota_find_data(db, tx);
1350                 } else {
1351                         data = DN_BONUS(dn->dn_phys);
1352                 }
1353         } else if (dn->dn_bonuslen == 0 && dn->dn_bonustype == DMU_OT_SA) {
1354                         int rf = 0;
1355 
1356                         if (RW_WRITE_HELD(&dn->dn_struct_rwlock))
1357                                 rf |= DB_RF_HAVESTRUCT;
1358                         error = dmu_spill_hold_by_dnode(dn,
1359                             rf | DB_RF_MUST_SUCCEED,
1360                             FTAG, (dmu_buf_t **)&db);
1361                         ASSERT(error == 0);
1362                         mutex_enter(&db->db_mtx);
1363                         data = (before) ? db->db.db_data :
1364                             dmu_objset_userquota_find_data(db, tx);
1365                         have_spill = B_TRUE;
1366         } else {
1367                 mutex_enter(&dn->dn_mtx);
1368                 dn->dn_id_flags |= DN_ID_CHKED_BONUS;
1369                 mutex_exit(&dn->dn_mtx);
1370                 return;
1371         }
1372 
1373         if (before) {
1374                 ASSERT(data);
1375                 user = &dn->dn_olduid;
1376                 group = &dn->dn_oldgid;
1377         } else if (data) {
1378                 user = &dn->dn_newuid;
1379                 group = &dn->dn_newgid;
1380         }
1381 
1382         /*
1383          * Must always call the callback in case the object
1384          * type has changed and that type isn't an object type to track
1385          */
1386         error = used_cbs[os->os_phys->os_type](dn->dn_bonustype, data,
1387             user, group);
1388 
1389         /*
1390          * Preserve existing uid/gid when the callback can't determine
1391          * what the new uid/gid are and the callback returned EEXIST.
1392          * The EEXIST error tells us to just use the existing uid/gid.
1393          * If we don't know what the old values are then just assign
1394          * them to 0, since that is a new file  being created.
1395          */
1396         if (!before && data == NULL && error == EEXIST) {
1397                 if (flags & DN_ID_OLD_EXIST) {
1398                         dn->dn_newuid = dn->dn_olduid;
1399                         dn->dn_newgid = dn->dn_oldgid;
1400                 } else {
1401                         dn->dn_newuid = 0;
1402                         dn->dn_newgid = 0;
1403                 }
1404                 error = 0;
1405         }
1406 
1407         if (db)
1408                 mutex_exit(&db->db_mtx);
1409 
1410         mutex_enter(&dn->dn_mtx);
1411         if (error == 0 && before)
1412                 dn->dn_id_flags |= DN_ID_OLD_EXIST;
1413         if (error == 0 && !before)
1414                 dn->dn_id_flags |= DN_ID_NEW_EXIST;
1415 
1416         if (have_spill) {
1417                 dn->dn_id_flags |= DN_ID_CHKED_SPILL;
1418         } else {
1419                 dn->dn_id_flags |= DN_ID_CHKED_BONUS;
1420         }
1421         mutex_exit(&dn->dn_mtx);
1422         if (have_spill)
1423                 dmu_buf_rele((dmu_buf_t *)db, FTAG);
1424 }
1425 
1426 boolean_t
1427 dmu_objset_userspace_present(objset_t *os)
1428 {
1429         return (os->os_phys->os_flags &
1430             OBJSET_FLAG_USERACCOUNTING_COMPLETE);
1431 }
1432 
1433 int
1434 dmu_objset_userspace_upgrade(objset_t *os)
1435 {
1436         uint64_t obj;
1437         int err = 0;
1438 
1439         if (dmu_objset_userspace_present(os))
1440                 return (0);
1441         if (!dmu_objset_userused_enabled(os))
1442                 return (SET_ERROR(ENOTSUP));
1443         if (dmu_objset_is_snapshot(os))
1444                 return (SET_ERROR(EINVAL));
1445 
1446         /*
1447          * We simply need to mark every object dirty, so that it will be
1448          * synced out and now accounted.  If this is called
1449          * concurrently, or if we already did some work before crashing,
1450          * that's fine, since we track each object's accounted state
1451          * independently.
1452          */
1453 
1454         for (obj = 0; err == 0; err = dmu_object_next(os, &obj, FALSE, 0)) {
1455                 dmu_tx_t *tx;
1456                 dmu_buf_t *db;
1457                 int objerr;
1458 
1459                 if (issig(JUSTLOOKING) && issig(FORREAL))
1460                         return (SET_ERROR(EINTR));
1461 
1462                 objerr = dmu_bonus_hold(os, obj, FTAG, &db);
1463                 if (objerr != 0)
1464                         continue;
1465                 tx = dmu_tx_create(os);
1466                 dmu_tx_hold_bonus(tx, obj);
1467                 objerr = dmu_tx_assign(tx, TXG_WAIT);
1468                 if (objerr != 0) {
1469                         dmu_tx_abort(tx);
1470                         continue;
1471                 }
1472                 dmu_buf_will_dirty(db, tx);
1473                 dmu_buf_rele(db, FTAG);
1474                 dmu_tx_commit(tx);
1475         }
1476 
1477         os->os_flags |= OBJSET_FLAG_USERACCOUNTING_COMPLETE;
1478         txg_wait_synced(dmu_objset_pool(os), 0);
1479         return (0);
1480 }
1481 
1482 void
1483 dmu_objset_space(objset_t *os, uint64_t *refdbytesp, uint64_t *availbytesp,
1484     uint64_t *usedobjsp, uint64_t *availobjsp)
1485 {
1486         dsl_dataset_space(os->os_dsl_dataset, refdbytesp, availbytesp,
1487             usedobjsp, availobjsp);
1488 }
1489 
1490 uint64_t
1491 dmu_objset_fsid_guid(objset_t *os)
1492 {
1493         return (dsl_dataset_fsid_guid(os->os_dsl_dataset));
1494 }
1495 
1496 void
1497 dmu_objset_fast_stat(objset_t *os, dmu_objset_stats_t *stat)
1498 {
1499         stat->dds_type = os->os_phys->os_type;
1500         if (os->os_dsl_dataset)
1501                 dsl_dataset_fast_stat(os->os_dsl_dataset, stat);
1502 }
1503 
1504 void
1505 dmu_objset_stats(objset_t *os, nvlist_t *nv)
1506 {
1507         ASSERT(os->os_dsl_dataset ||
1508             os->os_phys->os_type == DMU_OST_META);
1509 
1510         if (os->os_dsl_dataset != NULL)
1511                 dsl_dataset_stats(os->os_dsl_dataset, nv);
1512 
1513         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_TYPE,
1514             os->os_phys->os_type);
1515         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USERACCOUNTING,
1516             dmu_objset_userspace_present(os));
1517 }
1518 
1519 int
1520 dmu_objset_is_snapshot(objset_t *os)
1521 {
1522         if (os->os_dsl_dataset != NULL)
1523                 return (os->os_dsl_dataset->ds_is_snapshot);
1524         else
1525                 return (B_FALSE);
1526 }
1527 
1528 int
1529 dmu_snapshot_realname(objset_t *os, char *name, char *real, int maxlen,
1530     boolean_t *conflict)
1531 {
1532         dsl_dataset_t *ds = os->os_dsl_dataset;
1533         uint64_t ignored;
1534 
1535         if (dsl_dataset_phys(ds)->ds_snapnames_zapobj == 0)
1536                 return (SET_ERROR(ENOENT));
1537 
1538         return (zap_lookup_norm(ds->ds_dir->dd_pool->dp_meta_objset,
1539             dsl_dataset_phys(ds)->ds_snapnames_zapobj, name, 8, 1, &ignored,
1540             MT_FIRST, real, maxlen, conflict));
1541 }
1542 
1543 int
1544 dmu_snapshot_list_next(objset_t *os, int namelen, char *name,
1545     uint64_t *idp, uint64_t *offp, boolean_t *case_conflict)
1546 {
1547         dsl_dataset_t *ds = os->os_dsl_dataset;
1548         zap_cursor_t cursor;
1549         zap_attribute_t attr;
1550 
1551         ASSERT(dsl_pool_config_held(dmu_objset_pool(os)));
1552 
1553         if (dsl_dataset_phys(ds)->ds_snapnames_zapobj == 0)
1554                 return (SET_ERROR(ENOENT));
1555 
1556         zap_cursor_init_serialized(&cursor,
1557             ds->ds_dir->dd_pool->dp_meta_objset,
1558             dsl_dataset_phys(ds)->ds_snapnames_zapobj, *offp);
1559 
1560         if (zap_cursor_retrieve(&cursor, &attr) != 0) {
1561                 zap_cursor_fini(&cursor);
1562                 return (SET_ERROR(ENOENT));
1563         }
1564 
1565         if (strlen(attr.za_name) + 1 > namelen) {
1566                 zap_cursor_fini(&cursor);
1567                 return (SET_ERROR(ENAMETOOLONG));
1568         }
1569 
1570         (void) strcpy(name, attr.za_name);
1571         if (idp)
1572                 *idp = attr.za_first_integer;
1573         if (case_conflict)
1574                 *case_conflict = attr.za_normalization_conflict;
1575         zap_cursor_advance(&cursor);
1576         *offp = zap_cursor_serialize(&cursor);
1577         zap_cursor_fini(&cursor);
1578 
1579         return (0);
1580 }
1581 
1582 int
1583 dmu_dir_list_next(objset_t *os, int namelen, char *name,
1584     uint64_t *idp, uint64_t *offp)
1585 {
1586         dsl_dir_t *dd = os->os_dsl_dataset->ds_dir;
1587         zap_cursor_t cursor;
1588         zap_attribute_t attr;
1589 
1590         /* there is no next dir on a snapshot! */
1591         if (os->os_dsl_dataset->ds_object !=
1592             dsl_dir_phys(dd)->dd_head_dataset_obj)
1593                 return (SET_ERROR(ENOENT));
1594 
1595         zap_cursor_init_serialized(&cursor,
1596             dd->dd_pool->dp_meta_objset,
1597             dsl_dir_phys(dd)->dd_child_dir_zapobj, *offp);
1598 
1599         if (zap_cursor_retrieve(&cursor, &attr) != 0) {
1600                 zap_cursor_fini(&cursor);
1601                 return (SET_ERROR(ENOENT));
1602         }
1603 
1604         if (strlen(attr.za_name) + 1 > namelen) {
1605                 zap_cursor_fini(&cursor);
1606                 return (SET_ERROR(ENAMETOOLONG));
1607         }
1608 
1609         (void) strcpy(name, attr.za_name);
1610         if (idp)
1611                 *idp = attr.za_first_integer;
1612         zap_cursor_advance(&cursor);
1613         *offp = zap_cursor_serialize(&cursor);
1614         zap_cursor_fini(&cursor);
1615 
1616         return (0);
1617 }
1618 
1619 typedef struct dmu_objset_find_ctx {
1620         taskq_t         *dc_tq;
1621         dsl_pool_t      *dc_dp;
1622         uint64_t        dc_ddobj;
1623         int             (*dc_func)(dsl_pool_t *, dsl_dataset_t *, void *);
1624         void            *dc_arg;
1625         int             dc_flags;
1626         kmutex_t        *dc_error_lock;
1627         int             *dc_error;
1628 } dmu_objset_find_ctx_t;
1629 
1630 static void
1631 dmu_objset_find_dp_impl(dmu_objset_find_ctx_t *dcp)
1632 {
1633         dsl_pool_t *dp = dcp->dc_dp;
1634         dmu_objset_find_ctx_t *child_dcp;
1635         dsl_dir_t *dd;
1636         dsl_dataset_t *ds;
1637         zap_cursor_t zc;
1638         zap_attribute_t *attr;
1639         uint64_t thisobj;
1640         int err = 0;
1641 
1642         /* don't process if there already was an error */
1643         if (*dcp->dc_error != 0)
1644                 goto out;
1645 
1646         err = dsl_dir_hold_obj(dp, dcp->dc_ddobj, NULL, FTAG, &dd);
1647         if (err != 0)
1648                 goto out;
1649 
1650         /* Don't visit hidden ($MOS & $ORIGIN) objsets. */
1651         if (dd->dd_myname[0] == '$') {
1652                 dsl_dir_rele(dd, FTAG);
1653                 goto out;
1654         }
1655 
1656         thisobj = dsl_dir_phys(dd)->dd_head_dataset_obj;
1657         attr = kmem_alloc(sizeof (zap_attribute_t), KM_SLEEP);
1658 
1659         /*
1660          * Iterate over all children.
1661          */
1662         if (dcp->dc_flags & DS_FIND_CHILDREN) {
1663                 for (zap_cursor_init(&zc, dp->dp_meta_objset,
1664                     dsl_dir_phys(dd)->dd_child_dir_zapobj);
1665                     zap_cursor_retrieve(&zc, attr) == 0;
1666                     (void) zap_cursor_advance(&zc)) {
1667                         ASSERT3U(attr->za_integer_length, ==,
1668                             sizeof (uint64_t));
1669                         ASSERT3U(attr->za_num_integers, ==, 1);
1670 
1671                         child_dcp = kmem_alloc(sizeof(*child_dcp), KM_SLEEP);
1672                         *child_dcp = *dcp;
1673                         child_dcp->dc_ddobj = attr->za_first_integer;
1674                         if (dcp->dc_tq != NULL)
1675                                 (void) taskq_dispatch(dcp->dc_tq,
1676                                     dmu_objset_find_dp_cb, child_dcp, TQ_SLEEP);
1677                         else
1678                                 dmu_objset_find_dp_impl(child_dcp);
1679                 }
1680                 zap_cursor_fini(&zc);
1681         }
1682 
1683         /*
1684          * Iterate over all snapshots.
1685          */
1686         if (dcp->dc_flags & DS_FIND_SNAPSHOTS) {
1687                 dsl_dataset_t *ds;
1688                 err = dsl_dataset_hold_obj(dp, thisobj, FTAG, &ds);
1689 
1690                 if (err == 0) {
1691                         uint64_t snapobj;
1692 
1693                         snapobj = dsl_dataset_phys(ds)->ds_snapnames_zapobj;
1694                         dsl_dataset_rele(ds, FTAG);
1695 
1696                         for (zap_cursor_init(&zc, dp->dp_meta_objset, snapobj);
1697                             zap_cursor_retrieve(&zc, attr) == 0;
1698                             (void) zap_cursor_advance(&zc)) {
1699                                 ASSERT3U(attr->za_integer_length, ==,
1700                                     sizeof (uint64_t));
1701                                 ASSERT3U(attr->za_num_integers, ==, 1);
1702 
1703                                 err = dsl_dataset_hold_obj(dp,
1704                                     attr->za_first_integer, FTAG, &ds);
1705                                 if (err != 0)
1706                                         break;
1707                                 err = dcp->dc_func(dp, ds, dcp->dc_arg);
1708                                 dsl_dataset_rele(ds, FTAG);
1709                                 if (err != 0)
1710                                         break;
1711                         }
1712                         zap_cursor_fini(&zc);
1713                 }
1714         }
1715 
1716         dsl_dir_rele(dd, FTAG);
1717         kmem_free(attr, sizeof (zap_attribute_t));
1718 
1719         if (err != 0)
1720                 goto out;
1721 
1722         /*
1723          * Apply to self.
1724          */
1725         err = dsl_dataset_hold_obj(dp, thisobj, FTAG, &ds);
1726         if (err != 0)
1727                 goto out;
1728         err = dcp->dc_func(dp, ds, dcp->dc_arg);
1729         dsl_dataset_rele(ds, FTAG);
1730 
1731 out:
1732         if (err != 0) {
1733                 mutex_enter(dcp->dc_error_lock);
1734                 /* only keep first error */
1735                 if (*dcp->dc_error == 0)
1736                         *dcp->dc_error = err;
1737                 mutex_exit(dcp->dc_error_lock);
1738         }
1739 
1740         kmem_free(dcp, sizeof(*dcp));
1741 }
1742 
1743 static void
1744 dmu_objset_find_dp_cb(void *arg)
1745 {
1746         dmu_objset_find_ctx_t *dcp = arg;
1747         dsl_pool_t *dp = dcp->dc_dp;
1748 
1749         /*
1750          * We need to get a pool_config_lock here, as there are several
1751          * asssert(pool_config_held) down the stack. Getting a lock via
1752          * dsl_pool_config_enter is risky, as it might be stalled by a
1753          * pending writer. This would deadlock, as the write lock can
1754          * only be granted when our parent thread gives up the lock.
1755          * The _prio interface gives us priority over a pending writer.
1756          * On the other hand, we don't risk to stall any pending writers,
1757          * as the parent thread already holds a config lock. We give up
1758          * our lock before the parent does, so in effect we do not prolong
1759          * the waiting time for the writer.
1760          */
1761         dsl_pool_config_enter_prio(dp, FTAG);
1762 
1763         dmu_objset_find_dp_impl(dcp);
1764 
1765         dsl_pool_config_exit(dp, FTAG);
1766 }
1767 
1768 /*
1769  * Find objsets under and including ddobj, call func(ds) on each.
1770  * The order for the enumeration is completely undefined.
1771  * func is called with dsl_pool_config held.
1772  */
1773 int
1774 dmu_objset_find_dp(dsl_pool_t *dp, uint64_t ddobj,
1775     int func(dsl_pool_t *, dsl_dataset_t *, void *), void *arg, int flags)
1776 {
1777         int error = 0;
1778         taskq_t *tq = NULL;
1779         int ntasks;
1780         dmu_objset_find_ctx_t *dcp;
1781         kmutex_t err_lock;
1782 
1783         mutex_init(&err_lock, NULL, MUTEX_DEFAULT, NULL);
1784         dcp = kmem_alloc(sizeof(*dcp), KM_SLEEP);
1785         dcp->dc_tq = NULL;
1786         dcp->dc_dp = dp;
1787         dcp->dc_ddobj = ddobj;
1788         dcp->dc_func = func;
1789         dcp->dc_arg = arg;
1790         dcp->dc_flags = flags;
1791         dcp->dc_error_lock = &err_lock;
1792         dcp->dc_error = &error;
1793 
1794         if ((flags & DS_FIND_SERIALIZE) || dsl_pool_config_held_writer(dp)) {
1795                 /*
1796                  * In case a write lock is held we can't make use of
1797                  * parallelism, as down the stack of the worker threads
1798                  * the lock is asserted via dsl_pool_config_held.
1799                  * In case of a read lock this is solved by getting a read
1800                  * lock in each worker thread, which isn't possible in case
1801                  * of a writer lock. So we fall back to the synchronous path
1802                  * here.
1803                  * In the future it might be possible to get some magic into
1804                  * dsl_pool_config_held in a way that it returns true for
1805                  * the worker threads so that a single lock held from this
1806                  * thread suffices. For now, stay single threaded.
1807                  */
1808                 dmu_objset_find_dp_impl(dcp);
1809 
1810                 return (error);
1811         }
1812 
1813         ntasks = dmu_find_threads;
1814         if (ntasks == 0)
1815                 ntasks = vdev_count_leaves(dp->dp_spa) * 4;
1816         tq = taskq_create("dmu_objset_find", ntasks, minclsyspri, ntasks,
1817             INT_MAX, 0);
1818         if (tq == NULL) {
1819                 kmem_free(dcp, sizeof(*dcp));
1820                 return (SET_ERROR(ENOMEM));
1821         }
1822         dcp->dc_tq = tq;
1823 
1824         /* dcp will be freed by task */
1825         (void) taskq_dispatch(tq, dmu_objset_find_dp_cb, dcp, TQ_SLEEP);
1826 
1827         /*
1828          * PORTING: this code relies on the property of taskq_wait to wait
1829          * until no more tasks are queued and no more tasks are active. As
1830          * we always queue new tasks from within other tasks, task_wait
1831          * reliably waits for the full recursion to finish, even though we
1832          * enqueue new tasks after taskq_wait has been called.
1833          * On platforms other than illumos, taskq_wait may not have this
1834          * property.
1835          */
1836         taskq_wait(tq);
1837         taskq_destroy(tq);
1838         mutex_destroy(&err_lock);
1839 
1840         return (error);
1841 }
1842 
1843 /*
1844  * Find all objsets under name, and for each, call 'func(child_name, arg)'.
1845  * The dp_config_rwlock must not be held when this is called, and it
1846  * will not be held when the callback is called.
1847  * Therefore this function should only be used when the pool is not changing
1848  * (e.g. in syncing context), or the callback can deal with the possible races.
1849  */
1850 static int
1851 dmu_objset_find_impl(spa_t *spa, const char *name,
1852     int func(const char *, void *), void *arg, int flags)
1853 {
1854         dsl_dir_t *dd;
1855         dsl_pool_t *dp = spa_get_dsl(spa);
1856         dsl_dataset_t *ds;
1857         zap_cursor_t zc;
1858         zap_attribute_t *attr;
1859         char *child;
1860         uint64_t thisobj;
1861         int err;
1862 
1863         dsl_pool_config_enter(dp, FTAG);
1864 
1865         err = dsl_dir_hold(dp, name, FTAG, &dd, NULL);
1866         if (err != 0) {
1867                 dsl_pool_config_exit(dp, FTAG);
1868                 return (err);
1869         }
1870 
1871         /* Don't visit hidden ($MOS & $ORIGIN) objsets. */
1872         if (dd->dd_myname[0] == '$') {
1873                 dsl_dir_rele(dd, FTAG);
1874                 dsl_pool_config_exit(dp, FTAG);
1875                 return (0);
1876         }
1877 
1878         thisobj = dsl_dir_phys(dd)->dd_head_dataset_obj;
1879         attr = kmem_alloc(sizeof (zap_attribute_t), KM_SLEEP);
1880 
1881         /*
1882          * Iterate over all children.
1883          */
1884         if (flags & DS_FIND_CHILDREN) {
1885                 for (zap_cursor_init(&zc, dp->dp_meta_objset,
1886                     dsl_dir_phys(dd)->dd_child_dir_zapobj);
1887                     zap_cursor_retrieve(&zc, attr) == 0;
1888                     (void) zap_cursor_advance(&zc)) {
1889                         ASSERT3U(attr->za_integer_length, ==,
1890                             sizeof (uint64_t));
1891                         ASSERT3U(attr->za_num_integers, ==, 1);
1892 
1893                         child = kmem_asprintf("%s/%s", name, attr->za_name);
1894                         dsl_pool_config_exit(dp, FTAG);
1895                         err = dmu_objset_find_impl(spa, child,
1896                             func, arg, flags);
1897                         dsl_pool_config_enter(dp, FTAG);
1898                         strfree(child);
1899                         if (err != 0)
1900                                 break;
1901                 }
1902                 zap_cursor_fini(&zc);
1903 
1904                 if (err != 0) {
1905                         dsl_dir_rele(dd, FTAG);
1906                         dsl_pool_config_exit(dp, FTAG);
1907                         kmem_free(attr, sizeof (zap_attribute_t));
1908                         return (err);
1909                 }
1910         }
1911 
1912         /*
1913          * Iterate over all snapshots.
1914          */
1915         if (flags & DS_FIND_SNAPSHOTS) {
1916                 err = dsl_dataset_hold_obj(dp, thisobj, FTAG, &ds);
1917 
1918                 if (err == 0) {
1919                         uint64_t snapobj;
1920 
1921                         snapobj = dsl_dataset_phys(ds)->ds_snapnames_zapobj;
1922                         dsl_dataset_rele(ds, FTAG);
1923 
1924                         for (zap_cursor_init(&zc, dp->dp_meta_objset, snapobj);
1925                             zap_cursor_retrieve(&zc, attr) == 0;
1926                             (void) zap_cursor_advance(&zc)) {
1927                                 ASSERT3U(attr->za_integer_length, ==,
1928                                     sizeof (uint64_t));
1929                                 ASSERT3U(attr->za_num_integers, ==, 1);
1930 
1931                                 child = kmem_asprintf("%s@%s",
1932                                     name, attr->za_name);
1933                                 dsl_pool_config_exit(dp, FTAG);
1934                                 err = func(child, arg);
1935                                 dsl_pool_config_enter(dp, FTAG);
1936                                 strfree(child);
1937                                 if (err != 0)
1938                                         break;
1939                         }
1940                         zap_cursor_fini(&zc);
1941                 }
1942         }
1943 
1944         dsl_dir_rele(dd, FTAG);
1945         kmem_free(attr, sizeof (zap_attribute_t));
1946         dsl_pool_config_exit(dp, FTAG);
1947 
1948         if (err != 0)
1949                 return (err);
1950 
1951         /* Apply to self. */
1952         return (func(name, arg));
1953 }
1954 
1955 /*
1956  * See comment above dmu_objset_find_impl().
1957  */
1958 int
1959 dmu_objset_find(char *name, int func(const char *, void *), void *arg,
1960     int flags)
1961 {
1962         spa_t *spa;
1963         int error;
1964 
1965         error = spa_open(name, &spa, FTAG);
1966         if (error != 0)
1967                 return (error);
1968         error = dmu_objset_find_impl(spa, name, func, arg, flags);
1969         spa_close(spa, FTAG);
1970         return (error);
1971 }
1972 
1973 void
1974 dmu_objset_set_user(objset_t *os, void *user_ptr)
1975 {
1976         ASSERT(MUTEX_HELD(&os->os_user_ptr_lock));
1977         os->os_user_ptr = user_ptr;
1978 }
1979 
1980 void *
1981 dmu_objset_get_user(objset_t *os)
1982 {
1983         ASSERT(MUTEX_HELD(&os->os_user_ptr_lock));
1984         return (os->os_user_ptr);
1985 }
1986 
1987 /*
1988  * Determine name of filesystem, given name of snapshot.
1989  * buf must be at least MAXNAMELEN bytes
1990  */
1991 int
1992 dmu_fsname(const char *snapname, char *buf)
1993 {
1994         char *atp = strchr(snapname, '@');
1995         if (atp == NULL)
1996                 return (SET_ERROR(EINVAL));
1997         if (atp - snapname >= MAXNAMELEN)
1998                 return (SET_ERROR(ENAMETOOLONG));
1999         (void) strlcpy(buf, snapname, atp - snapname + 1);
2000         return (0);
2001 }