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 /*
  23  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
  24  * Copyright (c) 2012, 2017 by Delphix. All rights reserved.
  25  * Copyright (c) 2013 by Saso Kiselkov. All rights reserved.
  26  * Copyright (c) 2013, Joyent, Inc. All rights reserved.
  27  * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
  28  * Copyright (c) 2015, STRATO AG, Inc. All rights reserved.
  29  * Copyright (c) 2014 Integros [integros.com]
  30  * Copyright 2017 Nexenta Systems, Inc.
  31  */
  32 
  33 /* Portions Copyright 2010 Robert Milkowski */
  34 
  35 #include <sys/cred.h>
  36 #include <sys/zfs_context.h>
  37 #include <sys/dmu_objset.h>
  38 #include <sys/dsl_dir.h>
  39 #include <sys/dsl_dataset.h>
  40 #include <sys/dsl_prop.h>
  41 #include <sys/dsl_pool.h>
  42 #include <sys/dsl_synctask.h>
  43 #include <sys/dsl_deleg.h>
  44 #include <sys/dnode.h>
  45 #include <sys/dbuf.h>
  46 #include <sys/zvol.h>
  47 #include <sys/dmu_tx.h>
  48 #include <sys/zap.h>
  49 #include <sys/zil.h>
  50 #include <sys/dmu_impl.h>
  51 #include <sys/zfs_ioctl.h>
  52 #include <sys/sa.h>
  53 #include <sys/zfs_onexit.h>
  54 #include <sys/dsl_destroy.h>
  55 #include <sys/vdev.h>
  56 #include <sys/wbc.h>
  57 
  58 /*
  59  * Needed to close a window in dnode_move() that allows the objset to be freed
  60  * before it can be safely accessed.
  61  */
  62 krwlock_t os_lock;
  63 
  64 extern kmem_cache_t *zfs_ds_collector_cache;
  65 
  66 /*
  67  * Tunable to overwrite the maximum number of threads for the parallization
  68  * of dmu_objset_find_dp, needed to speed up the import of pools with many
  69  * datasets.
  70  * Default is 4 times the number of leaf vdevs.
  71  */
  72 int dmu_find_threads = 0;
  73 
  74 /*
  75  * Backfill lower metadnode objects after this many have been freed.
  76  * Backfilling negatively impacts object creation rates, so only do it
  77  * if there are enough holes to fill.
  78  */
  79 int dmu_rescan_dnode_threshold = 131072;
  80 
  81 static void dmu_objset_find_dp_cb(void *arg);
  82 
  83 /* ARGSUSED */
  84 static int
  85 zfs_ds_collector_constructor(void *ds_el, void *unused, int flags)
  86 {
  87         bzero(ds_el, sizeof (zfs_ds_collector_entry_t));
  88         return (0);
  89 }
  90 
  91 void
  92 dmu_objset_init(void)
  93 {
  94         zfs_ds_collector_cache = kmem_cache_create("zfs_ds_collector_cache",
  95             sizeof (zfs_ds_collector_entry_t),
  96             8, zfs_ds_collector_constructor,
  97             NULL, NULL, NULL, NULL, 0);
  98         rw_init(&os_lock, NULL, RW_DEFAULT, NULL);
  99 }
 100 
 101 void
 102 dmu_objset_fini(void)
 103 {
 104         rw_destroy(&os_lock);
 105         kmem_cache_destroy(zfs_ds_collector_cache);
 106 }
 107 
 108 spa_t *
 109 dmu_objset_spa(objset_t *os)
 110 {
 111         return (os->os_spa);
 112 }
 113 
 114 zilog_t *
 115 dmu_objset_zil(objset_t *os)
 116 {
 117         return (os->os_zil);
 118 }
 119 
 120 dsl_pool_t *
 121 dmu_objset_pool(objset_t *os)
 122 {
 123         dsl_dataset_t *ds;
 124 
 125         if ((ds = os->os_dsl_dataset) != NULL && ds->ds_dir)
 126                 return (ds->ds_dir->dd_pool);
 127         else
 128                 return (spa_get_dsl(os->os_spa));
 129 }
 130 
 131 dsl_dataset_t *
 132 dmu_objset_ds(objset_t *os)
 133 {
 134         return (os->os_dsl_dataset);
 135 }
 136 
 137 dmu_objset_type_t
 138 dmu_objset_type(objset_t *os)
 139 {
 140         return (os->os_phys->os_type);
 141 }
 142 
 143 void
 144 dmu_objset_name(objset_t *os, char *buf)
 145 {
 146         dsl_dataset_name(os->os_dsl_dataset, buf);
 147 }
 148 
 149 uint64_t
 150 dmu_objset_id(objset_t *os)
 151 {
 152         dsl_dataset_t *ds = os->os_dsl_dataset;
 153 
 154         return (ds ? ds->ds_object : 0);
 155 }
 156 
 157 zfs_sync_type_t
 158 dmu_objset_syncprop(objset_t *os)
 159 {
 160         return (os->os_sync);
 161 }
 162 
 163 zfs_logbias_op_t
 164 dmu_objset_logbias(objset_t *os)
 165 {
 166         return (os->os_logbias);
 167 }
 168 
 169 static void
 170 checksum_changed_cb(void *arg, uint64_t newval)
 171 {
 172         objset_t *os = arg;
 173 
 174         /*
 175          * Inheritance should have been done by now.
 176          */
 177         ASSERT(newval != ZIO_CHECKSUM_INHERIT);
 178 
 179         os->os_checksum = zio_checksum_select(newval, ZIO_CHECKSUM_ON_VALUE);
 180 }
 181 
 182 static void
 183 compression_changed_cb(void *arg, uint64_t newval)
 184 {
 185         objset_t *os = arg;
 186 
 187         /*
 188          * Inheritance and range checking should have been done by now.
 189          */
 190         ASSERT(newval != ZIO_COMPRESS_INHERIT);
 191 
 192         os->os_compress = zio_compress_select(os->os_spa, newval,
 193             ZIO_COMPRESS_ON);
 194 }
 195 
 196 static void
 197 smartcomp_changed_cb(void *arg, uint64_t newval)
 198 {
 199         objset_t *os = arg;
 200 
 201         os->os_smartcomp_enabled = newval ? B_TRUE : B_FALSE;
 202 }
 203 
 204 static void
 205 copies_changed_cb(void *arg, uint64_t newval)
 206 {
 207         objset_t *os = arg;
 208 
 209         /*
 210          * Inheritance and range checking should have been done by now.
 211          */
 212         ASSERT(newval > 0);
 213         ASSERT(newval <= spa_max_replication(os->os_spa));
 214 
 215         os->os_copies = newval;
 216 }
 217 
 218 static void
 219 dedup_changed_cb(void *arg, uint64_t newval)
 220 {
 221         objset_t *os = arg;
 222         spa_t *spa = os->os_spa;
 223         enum zio_checksum checksum;
 224 
 225         /*
 226          * Inheritance should have been done by now.
 227          */
 228         ASSERT(newval != ZIO_CHECKSUM_INHERIT);
 229 
 230         checksum = zio_checksum_dedup_select(spa, newval, ZIO_CHECKSUM_OFF);
 231 
 232         os->os_dedup_checksum = checksum & ZIO_CHECKSUM_MASK;
 233         os->os_dedup_verify = !!(checksum & ZIO_CHECKSUM_VERIFY);
 234 }
 235 
 236 static void
 237 primary_cache_changed_cb(void *arg, uint64_t newval)
 238 {
 239         objset_t *os = arg;
 240 
 241         /*
 242          * Inheritance and range checking should have been done by now.
 243          */
 244         ASSERT(newval == ZFS_CACHE_ALL || newval == ZFS_CACHE_NONE ||
 245             newval == ZFS_CACHE_METADATA);
 246 
 247         os->os_primary_cache = newval;
 248 }
 249 
 250 static void
 251 secondary_cache_changed_cb(void *arg, uint64_t newval)
 252 {
 253         objset_t *os = arg;
 254 
 255         /*
 256          * Inheritance and range checking should have been done by now.
 257          */
 258         ASSERT(newval == ZFS_CACHE_ALL || newval == ZFS_CACHE_NONE ||
 259             newval == ZFS_CACHE_METADATA || newval == ZFS_CACHE_DATA);
 260 
 261         os->os_secondary_cache = newval;
 262 }
 263 
 264 static void
 265 zpl_meta_placement_changed_cb(void *arg, uint64_t newval)
 266 {
 267         objset_t *os = arg;
 268 
 269         os->os_zpl_meta_to_special = newval;
 270 }
 271 
 272 static void
 273 sync_changed_cb(void *arg, uint64_t newval)
 274 {
 275         objset_t *os = arg;
 276 
 277         /*
 278          * Inheritance and range checking should have been done by now.
 279          */
 280         ASSERT(newval == ZFS_SYNC_STANDARD || newval == ZFS_SYNC_ALWAYS ||
 281             newval == ZFS_SYNC_DISABLED);
 282 
 283         os->os_sync = newval;
 284         if (os->os_zil)
 285                 zil_set_sync(os->os_zil, newval);
 286 }
 287 
 288 static void
 289 redundant_metadata_changed_cb(void *arg, uint64_t newval)
 290 {
 291         objset_t *os = arg;
 292 
 293         /*
 294          * Inheritance and range checking should have been done by now.
 295          */
 296         ASSERT(newval == ZFS_REDUNDANT_METADATA_ALL ||
 297             newval == ZFS_REDUNDANT_METADATA_MOST);
 298 
 299         os->os_redundant_metadata = newval;
 300 }
 301 
 302 static void
 303 logbias_changed_cb(void *arg, uint64_t newval)
 304 {
 305         objset_t *os = arg;
 306 
 307         ASSERT(newval == ZFS_LOGBIAS_LATENCY ||
 308             newval == ZFS_LOGBIAS_THROUGHPUT);
 309         os->os_logbias = newval;
 310         if (os->os_zil)
 311                 zil_set_logbias(os->os_zil, newval);
 312 }
 313 
 314 static void
 315 recordsize_changed_cb(void *arg, uint64_t newval)
 316 {
 317         objset_t *os = arg;
 318 
 319         os->os_recordsize = newval;
 320 }
 321 
 322 void
 323 dmu_objset_byteswap(void *buf, size_t size)
 324 {
 325         objset_phys_t *osp = buf;
 326 
 327         ASSERT(size == OBJSET_OLD_PHYS_SIZE || size == sizeof (objset_phys_t));
 328         dnode_byteswap(&osp->os_meta_dnode);
 329         byteswap_uint64_array(&osp->os_zil_header, sizeof (zil_header_t));
 330         osp->os_type = BSWAP_64(osp->os_type);
 331         osp->os_flags = BSWAP_64(osp->os_flags);
 332         if (size == sizeof (objset_phys_t)) {
 333                 dnode_byteswap(&osp->os_userused_dnode);
 334                 dnode_byteswap(&osp->os_groupused_dnode);
 335         }
 336 }
 337 
 338 /*
 339  * The hash is a CRC-based hash of the objset_t pointer and the object number.
 340  */
 341 static uint64_t
 342 dnode_hash(const objset_t *os, uint64_t obj)
 343 {
 344         uintptr_t osv = (uintptr_t)os;
 345         uint64_t crc = -1ULL;
 346 
 347         ASSERT(zfs_crc64_table[128] == ZFS_CRC64_POLY);
 348         /*
 349          * The low 6 bits of the pointer don't have much entropy, because
 350          * the objset_t is larger than 2^6 bytes long.
 351          */
 352         crc = (crc >> 8) ^ zfs_crc64_table[(crc ^ (osv >> 6)) & 0xFF];
 353         crc = (crc >> 8) ^ zfs_crc64_table[(crc ^ (obj >> 0)) & 0xFF];
 354         crc = (crc >> 8) ^ zfs_crc64_table[(crc ^ (obj >> 8)) & 0xFF];
 355         crc = (crc >> 8) ^ zfs_crc64_table[(crc ^ (obj >> 16)) & 0xFF];
 356 
 357         crc ^= (osv>>14) ^ (obj>>24);
 358 
 359         return (crc);
 360 }
 361 
 362 unsigned int
 363 dnode_multilist_index_func(multilist_t *ml, void *obj)
 364 {
 365         dnode_t *dn = obj;
 366         return (dnode_hash(dn->dn_objset, dn->dn_object) %
 367             multilist_get_num_sublists(ml));
 368 }
 369 
 370 /*
 371  * Instantiates the objset_t in-memory structure corresponding to the
 372  * objset_phys_t that's pointed to by the specified blkptr_t.
 373  */
 374 int
 375 dmu_objset_open_impl(spa_t *spa, dsl_dataset_t *ds, blkptr_t *bp,
 376     objset_t **osp)
 377 {
 378         objset_t *os;
 379         int i, err;
 380 
 381         ASSERT(ds == NULL || MUTEX_HELD(&ds->ds_opening_lock));
 382 
 383         os = kmem_zalloc(sizeof (objset_t), KM_SLEEP);
 384         os->os_dsl_dataset = ds;
 385         os->os_spa = spa;
 386         os->os_rootbp = bp;
 387         if (!BP_IS_HOLE(os->os_rootbp)) {
 388                 arc_flags_t aflags = ARC_FLAG_WAIT;
 389                 zbookmark_phys_t zb;
 390                 SET_BOOKMARK(&zb, ds ? ds->ds_object : DMU_META_OBJSET,
 391                     ZB_ROOT_OBJECT, ZB_ROOT_LEVEL, ZB_ROOT_BLKID);
 392 
 393                 if (DMU_OS_IS_L2CACHEABLE(os))
 394                         aflags |= ARC_FLAG_L2CACHE;
 395 
 396                 dprintf_bp(os->os_rootbp, "reading %s", "");
 397                 err = arc_read(NULL, spa, os->os_rootbp,
 398                     arc_getbuf_func, &os->os_phys_buf,
 399                     ZIO_PRIORITY_SYNC_READ, ZIO_FLAG_CANFAIL, &aflags, &zb);
 400                 if (err != 0) {
 401                         kmem_free(os, sizeof (objset_t));
 402                         /* convert checksum errors into IO errors */
 403                         if (err == ECKSUM)
 404                                 err = SET_ERROR(EIO);
 405                         return (err);
 406                 }
 407 
 408                 /* Increase the blocksize if we are permitted. */
 409                 if (spa_version(spa) >= SPA_VERSION_USERSPACE &&
 410                     arc_buf_size(os->os_phys_buf) < sizeof (objset_phys_t)) {
 411                         arc_buf_t *buf = arc_alloc_buf(spa, &os->os_phys_buf,
 412                             ARC_BUFC_METADATA, sizeof (objset_phys_t));
 413                         bzero(buf->b_data, sizeof (objset_phys_t));
 414                         bcopy(os->os_phys_buf->b_data, buf->b_data,
 415                             arc_buf_size(os->os_phys_buf));
 416                         arc_buf_destroy(os->os_phys_buf, &os->os_phys_buf);
 417                         os->os_phys_buf = buf;
 418                 }
 419 
 420                 os->os_phys = os->os_phys_buf->b_data;
 421                 os->os_flags = os->os_phys->os_flags;
 422         } else {
 423                 int size = spa_version(spa) >= SPA_VERSION_USERSPACE ?
 424                     sizeof (objset_phys_t) : OBJSET_OLD_PHYS_SIZE;
 425                 os->os_phys_buf = arc_alloc_buf(spa, &os->os_phys_buf,
 426                     ARC_BUFC_METADATA, size);
 427                 os->os_phys = os->os_phys_buf->b_data;
 428                 bzero(os->os_phys, size);
 429         }
 430 
 431         /*
 432          * Note: the changed_cb will be called once before the register
 433          * func returns, thus changing the checksum/compression from the
 434          * default (fletcher2/off).  Snapshots don't need to know about
 435          * checksum/compression/copies.
 436          */
 437         if (ds != NULL) {
 438                 boolean_t needlock = B_FALSE;
 439 
 440                 /*
 441                  * Note: it's valid to open the objset if the dataset is
 442                  * long-held, in which case the pool_config lock will not
 443                  * be held.
 444                  */
 445                 if (!dsl_pool_config_held(dmu_objset_pool(os))) {
 446                         needlock = B_TRUE;
 447                         dsl_pool_config_enter(dmu_objset_pool(os), FTAG);
 448                 }
 449                 err = dsl_prop_register(ds,
 450                     zfs_prop_to_name(ZFS_PROP_PRIMARYCACHE),
 451                     primary_cache_changed_cb, os);
 452                 if (err == 0) {
 453                         err = dsl_prop_register(ds,
 454                             zfs_prop_to_name(ZFS_PROP_SECONDARYCACHE),
 455                             secondary_cache_changed_cb, os);
 456                 }
 457                 if (err == 0) {
 458                         err = dsl_prop_register(ds,
 459                             zfs_prop_to_name(ZFS_PROP_ZPL_META_TO_METADEV),
 460                             zpl_meta_placement_changed_cb, os);
 461                 }
 462                 if (!ds->ds_is_snapshot) {
 463                         if (err == 0) {
 464                                 err = dsl_prop_register(ds,
 465                                     zfs_prop_to_name(ZFS_PROP_CHECKSUM),
 466                                     checksum_changed_cb, os);
 467                         }
 468                         if (err == 0) {
 469                                 err = dsl_prop_register(ds,
 470                                     zfs_prop_to_name(ZFS_PROP_COMPRESSION),
 471                                     compression_changed_cb, os);
 472                         }
 473                         if (err == 0) {
 474                                 err = dsl_prop_register(ds,
 475                                     zfs_prop_to_name(ZFS_PROP_SMARTCOMPRESSION),
 476                                     smartcomp_changed_cb, os);
 477                         }
 478                         if (err == 0) {
 479                                 err = dsl_prop_register(ds,
 480                                     zfs_prop_to_name(ZFS_PROP_COPIES),
 481                                     copies_changed_cb, os);
 482                         }
 483                         if (err == 0) {
 484                                 err = dsl_prop_register(ds,
 485                                     zfs_prop_to_name(ZFS_PROP_DEDUP),
 486                                     dedup_changed_cb, os);
 487                         }
 488                         if (err == 0) {
 489                                 err = dsl_prop_register(ds,
 490                                     zfs_prop_to_name(ZFS_PROP_LOGBIAS),
 491                                     logbias_changed_cb, os);
 492                         }
 493                         if (err == 0) {
 494                                 err = dsl_prop_register(ds,
 495                                     zfs_prop_to_name(ZFS_PROP_SYNC),
 496                                     sync_changed_cb, os);
 497                         }
 498                         if (err == 0) {
 499                                 err = dsl_prop_register(ds,
 500                                     zfs_prop_to_name(
 501                                     ZFS_PROP_REDUNDANT_METADATA),
 502                                     redundant_metadata_changed_cb, os);
 503                         }
 504                         if (err == 0) {
 505                                 err = dsl_prop_register(ds,
 506                                     zfs_prop_to_name(ZFS_PROP_RECORDSIZE),
 507                                     recordsize_changed_cb, os);
 508                         }
 509                         if (err == 0) {
 510                                 err = dsl_prop_register(ds,
 511                                     zfs_prop_to_name(ZFS_PROP_WBC_MODE),
 512                                     wbc_mode_changed, os);
 513                         }
 514                 }
 515                 if (needlock)
 516                         dsl_pool_config_exit(dmu_objset_pool(os), FTAG);
 517                 if (err != 0) {
 518                         arc_buf_destroy(os->os_phys_buf, &os->os_phys_buf);
 519                         kmem_free(os, sizeof (objset_t));
 520                         return (err);
 521                 }
 522         } else {
 523                 /* It's the meta-objset. */
 524                 os->os_checksum = ZIO_CHECKSUM_FLETCHER_4;
 525                 os->os_compress = ZIO_COMPRESS_ON;
 526                 os->os_copies = spa_max_replication(spa);
 527                 os->os_dedup_checksum = ZIO_CHECKSUM_OFF;
 528                 os->os_dedup_verify = B_FALSE;
 529                 os->os_logbias = ZFS_LOGBIAS_LATENCY;
 530                 os->os_sync = ZFS_SYNC_STANDARD;
 531                 os->os_primary_cache = ZFS_CACHE_ALL;
 532                 os->os_secondary_cache = ZFS_CACHE_ALL;
 533                 os->os_zpl_meta_to_special = 0;
 534         }
 535         /*
 536          * These properties will be filled in by the logic in zfs_get_zplprop()
 537          * when they are queried for the first time.
 538          */
 539         os->os_version = OBJSET_PROP_UNINITIALIZED;
 540         os->os_normalization = OBJSET_PROP_UNINITIALIZED;
 541         os->os_utf8only = OBJSET_PROP_UNINITIALIZED;
 542         os->os_casesensitivity = OBJSET_PROP_UNINITIALIZED;
 543 
 544         if (ds == NULL || !ds->ds_is_snapshot)
 545                 os->os_zil_header = os->os_phys->os_zil_header;
 546         os->os_zil = zil_alloc(os, &os->os_zil_header);
 547 
 548         for (i = 0; i < TXG_SIZE; i++) {
 549                 os->os_dirty_dnodes[i] = multilist_create(sizeof (dnode_t),
 550                     offsetof(dnode_t, dn_dirty_link[i]),
 551                     dnode_multilist_index_func);
 552         }
 553         list_create(&os->os_dnodes, sizeof (dnode_t),
 554             offsetof(dnode_t, dn_link));
 555         list_create(&os->os_downgraded_dbufs, sizeof (dmu_buf_impl_t),
 556             offsetof(dmu_buf_impl_t, db_link));
 557 
 558         mutex_init(&os->os_lock, NULL, MUTEX_DEFAULT, NULL);
 559         mutex_init(&os->os_userused_lock, NULL, MUTEX_DEFAULT, NULL);
 560         mutex_init(&os->os_obj_lock, NULL, MUTEX_DEFAULT, NULL);
 561         mutex_init(&os->os_user_ptr_lock, NULL, MUTEX_DEFAULT, NULL);
 562 
 563         dnode_special_open(os, &os->os_phys->os_meta_dnode,
 564             DMU_META_DNODE_OBJECT, &os->os_meta_dnode);
 565         if (arc_buf_size(os->os_phys_buf) >= sizeof (objset_phys_t)) {
 566                 dnode_special_open(os, &os->os_phys->os_userused_dnode,
 567                     DMU_USERUSED_OBJECT, &os->os_userused_dnode);
 568                 dnode_special_open(os, &os->os_phys->os_groupused_dnode,
 569                     DMU_GROUPUSED_OBJECT, &os->os_groupused_dnode);
 570         }
 571 
 572         *osp = os;
 573         return (0);
 574 }
 575 
 576 int
 577 dmu_objset_from_ds(dsl_dataset_t *ds, objset_t **osp)
 578 {
 579         int err = 0;
 580 
 581         /*
 582          * We shouldn't be doing anything with dsl_dataset_t's unless the
 583          * pool_config lock is held, or the dataset is long-held.
 584          */
 585         ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool) ||
 586             dsl_dataset_long_held(ds));
 587 
 588         mutex_enter(&ds->ds_opening_lock);
 589         if (ds->ds_objset == NULL) {
 590                 objset_t *os;
 591                 rrw_enter(&ds->ds_bp_rwlock, RW_READER, FTAG);
 592                 err = dmu_objset_open_impl(dsl_dataset_get_spa(ds),
 593                     ds, dsl_dataset_get_blkptr(ds), &os);
 594                 rrw_exit(&ds->ds_bp_rwlock, FTAG);
 595 
 596                 if (err == 0) {
 597                         mutex_enter(&ds->ds_lock);
 598                         ASSERT(ds->ds_objset == NULL);
 599                         ds->ds_objset = os;
 600                         mutex_exit(&ds->ds_lock);
 601                 }
 602         }
 603         *osp = ds->ds_objset;
 604         mutex_exit(&ds->ds_opening_lock);
 605         return (err);
 606 }
 607 
 608 /*
 609  * Holds the pool while the objset is held.  Therefore only one objset
 610  * can be held at a time.
 611  */
 612 int
 613 dmu_objset_hold(const char *name, void *tag, objset_t **osp)
 614 {
 615         dsl_pool_t *dp;
 616         dsl_dataset_t *ds;
 617         int err;
 618 
 619         err = dsl_pool_hold(name, tag, &dp);
 620         if (err != 0)
 621                 return (err);
 622         err = dsl_dataset_hold(dp, name, tag, &ds);
 623         if (err != 0) {
 624                 dsl_pool_rele(dp, tag);
 625                 return (err);
 626         }
 627 
 628         err = dmu_objset_from_ds(ds, osp);
 629         if (err != 0) {
 630                 dsl_dataset_rele(ds, tag);
 631                 dsl_pool_rele(dp, tag);
 632         }
 633 
 634         return (err);
 635 }
 636 
 637 static int
 638 dmu_objset_own_impl(dsl_dataset_t *ds, dmu_objset_type_t type,
 639     boolean_t readonly, void *tag, objset_t **osp)
 640 {
 641         int err;
 642 
 643         err = dmu_objset_from_ds(ds, osp);
 644         if (err != 0) {
 645                 dsl_dataset_disown(ds, tag);
 646         } else if (type != DMU_OST_ANY && type != (*osp)->os_phys->os_type) {
 647                 dsl_dataset_disown(ds, tag);
 648                 return (SET_ERROR(EINVAL));
 649         } else if (!readonly && dsl_dataset_is_snapshot(ds)) {
 650                 dsl_dataset_disown(ds, tag);
 651                 return (SET_ERROR(EROFS));
 652         }
 653         return (err);
 654 }
 655 
 656 /*
 657  * dsl_pool must not be held when this is called.
 658  * Upon successful return, there will be a longhold on the dataset,
 659  * and the dsl_pool will not be held.
 660  */
 661 int
 662 dmu_objset_own(const char *name, dmu_objset_type_t type,
 663     boolean_t readonly, void *tag, objset_t **osp)
 664 {
 665         dsl_pool_t *dp;
 666         dsl_dataset_t *ds;
 667         int err;
 668 
 669         err = dsl_pool_hold(name, FTAG, &dp);
 670         if (err != 0)
 671                 return (err);
 672         err = dsl_dataset_own(dp, name, tag, &ds);
 673         if (err != 0) {
 674                 dsl_pool_rele(dp, FTAG);
 675                 return (err);
 676         }
 677         err = dmu_objset_own_impl(ds, type, readonly, tag, osp);
 678         dsl_pool_rele(dp, FTAG);
 679 
 680         return (err);
 681 }
 682 
 683 int
 684 dmu_objset_own_obj(dsl_pool_t *dp, uint64_t obj, dmu_objset_type_t type,
 685     boolean_t readonly, void *tag, objset_t **osp)
 686 {
 687         dsl_dataset_t *ds;
 688         int err;
 689 
 690         err = dsl_dataset_own_obj(dp, obj, tag, &ds);
 691         if (err != 0)
 692                 return (err);
 693 
 694         return (dmu_objset_own_impl(ds, type, readonly, tag, osp));
 695 }
 696 
 697 void
 698 dmu_objset_rele(objset_t *os, void *tag)
 699 {
 700         dsl_pool_t *dp = dmu_objset_pool(os);
 701         dsl_dataset_rele(os->os_dsl_dataset, tag);
 702         dsl_pool_rele(dp, tag);
 703 }
 704 
 705 /*
 706  * When we are called, os MUST refer to an objset associated with a dataset
 707  * that is owned by 'tag'; that is, is held and long held by 'tag' and ds_owner
 708  * == tag.  We will then release and reacquire ownership of the dataset while
 709  * holding the pool config_rwlock to avoid intervening namespace or ownership
 710  * changes may occur.
 711  *
 712  * This exists solely to accommodate zfs_ioc_userspace_upgrade()'s desire to
 713  * release the hold on its dataset and acquire a new one on the dataset of the
 714  * same name so that it can be partially torn down and reconstructed.
 715  */
 716 void
 717 dmu_objset_refresh_ownership(objset_t *os, void *tag)
 718 {
 719         dsl_pool_t *dp;
 720         dsl_dataset_t *ds, *newds;
 721         char name[ZFS_MAX_DATASET_NAME_LEN];
 722 
 723         ds = os->os_dsl_dataset;
 724         VERIFY3P(ds, !=, NULL);
 725         VERIFY3P(ds->ds_owner, ==, tag);
 726         VERIFY(dsl_dataset_long_held(ds));
 727 
 728         dsl_dataset_name(ds, name);
 729         dp = dmu_objset_pool(os);
 730         dsl_pool_config_enter(dp, FTAG);
 731         dmu_objset_disown(os, tag);
 732         VERIFY0(dsl_dataset_own(dp, name, tag, &newds));
 733         VERIFY3P(newds, ==, os->os_dsl_dataset);
 734         dsl_pool_config_exit(dp, FTAG);
 735 }
 736 
 737 void
 738 dmu_objset_disown(objset_t *os, void *tag)
 739 {
 740         dsl_dataset_disown(os->os_dsl_dataset, tag);
 741 }
 742 
 743 void
 744 dmu_objset_evict_dbufs(objset_t *os)
 745 {
 746         dnode_t dn_marker;
 747         dnode_t *dn;
 748 
 749         mutex_enter(&os->os_lock);
 750         dn = list_head(&os->os_dnodes);
 751         while (dn != NULL) {
 752                 /*
 753                  * Skip dnodes without holds.  We have to do this dance
 754                  * because dnode_add_ref() only works if there is already a
 755                  * hold.  If the dnode has no holds, then it has no dbufs.
 756                  */
 757                 if (dnode_add_ref(dn, FTAG)) {
 758                         list_insert_after(&os->os_dnodes, dn, &dn_marker);
 759                         mutex_exit(&os->os_lock);
 760 
 761                         dnode_evict_dbufs(dn, DBUF_EVICT_ALL);
 762                         dnode_rele(dn, FTAG);
 763 
 764                         mutex_enter(&os->os_lock);
 765                         dn = list_next(&os->os_dnodes, &dn_marker);
 766                         list_remove(&os->os_dnodes, &dn_marker);
 767                 } else {
 768                         dn = list_next(&os->os_dnodes, dn);
 769                 }
 770         }
 771         mutex_exit(&os->os_lock);
 772 
 773         if (DMU_USERUSED_DNODE(os) != NULL) {
 774                 dnode_evict_dbufs(DMU_GROUPUSED_DNODE(os), DBUF_EVICT_ALL);
 775                 dnode_evict_dbufs(DMU_USERUSED_DNODE(os), DBUF_EVICT_ALL);
 776         }
 777         dnode_evict_dbufs(DMU_META_DNODE(os), DBUF_EVICT_ALL);
 778 }
 779 
 780 /*
 781  * Objset eviction processing is split into into two pieces.
 782  * The first marks the objset as evicting, evicts any dbufs that
 783  * have a refcount of zero, and then queues up the objset for the
 784  * second phase of eviction.  Once os->os_dnodes has been cleared by
 785  * dnode_buf_pageout()->dnode_destroy(), the second phase is executed.
 786  * The second phase closes the special dnodes, dequeues the objset from
 787  * the list of those undergoing eviction, and finally frees the objset.
 788  *
 789  * NOTE: Due to asynchronous eviction processing (invocation of
 790  *       dnode_buf_pageout()), it is possible for the meta dnode for the
 791  *       objset to have no holds even though os->os_dnodes is not empty.
 792  */
 793 void
 794 dmu_objset_evict(objset_t *os)
 795 {
 796         dsl_dataset_t *ds = os->os_dsl_dataset;
 797 
 798         for (int t = 0; t < TXG_SIZE; t++)
 799                 ASSERT(!dmu_objset_is_dirty(os, t));
 800 
 801         if (ds)
 802                 dsl_prop_unregister_all(ds, os);
 803 
 804         if (os->os_sa)
 805                 sa_tear_down(os);
 806 
 807         dmu_objset_evict_dbufs(os);
 808 
 809         mutex_enter(&os->os_lock);
 810         spa_evicting_os_register(os->os_spa, os);
 811         if (list_is_empty(&os->os_dnodes)) {
 812                 mutex_exit(&os->os_lock);
 813                 dmu_objset_evict_done(os);
 814         } else {
 815                 mutex_exit(&os->os_lock);
 816         }
 817 }
 818 
 819 void
 820 dmu_objset_evict_done(objset_t *os)
 821 {
 822         ASSERT3P(list_head(&os->os_dnodes), ==, NULL);
 823 
 824         dnode_special_close(&os->os_meta_dnode);
 825         if (DMU_USERUSED_DNODE(os)) {
 826                 dnode_special_close(&os->os_userused_dnode);
 827                 dnode_special_close(&os->os_groupused_dnode);
 828         }
 829         zil_free(os->os_zil);
 830 
 831         arc_buf_destroy(os->os_phys_buf, &os->os_phys_buf);
 832 
 833         /*
 834          * This is a barrier to prevent the objset from going away in
 835          * dnode_move() until we can safely ensure that the objset is still in
 836          * use. We consider the objset valid before the barrier and invalid
 837          * after the barrier.
 838          */
 839         rw_enter(&os_lock, RW_READER);
 840         rw_exit(&os_lock);
 841 
 842         mutex_destroy(&os->os_lock);
 843         mutex_destroy(&os->os_userused_lock);
 844         mutex_destroy(&os->os_obj_lock);
 845         mutex_destroy(&os->os_user_ptr_lock);
 846         for (int i = 0; i < TXG_SIZE; i++) {
 847                 multilist_destroy(os->os_dirty_dnodes[i]);
 848         }
 849         spa_evicting_os_deregister(os->os_spa, os);
 850         kmem_free(os, sizeof (objset_t));
 851 }
 852 
 853 timestruc_t
 854 dmu_objset_snap_cmtime(objset_t *os)
 855 {
 856         return (dsl_dir_snap_cmtime(os->os_dsl_dataset->ds_dir));
 857 }
 858 
 859 /* called from dsl for meta-objset */
 860 objset_t *
 861 dmu_objset_create_impl(spa_t *spa, dsl_dataset_t *ds, blkptr_t *bp,
 862     dmu_objset_type_t type, dmu_tx_t *tx)
 863 {
 864         objset_t *os;
 865         dnode_t *mdn;
 866 
 867         ASSERT(dmu_tx_is_syncing(tx));
 868 
 869         if (ds != NULL)
 870                 VERIFY0(dmu_objset_from_ds(ds, &os));
 871         else
 872                 VERIFY0(dmu_objset_open_impl(spa, NULL, bp, &os));
 873 
 874         mdn = DMU_META_DNODE(os);
 875 
 876         dnode_allocate(mdn, DMU_OT_DNODE, 1 << DNODE_BLOCK_SHIFT,
 877             DN_MAX_INDBLKSHIFT, DMU_OT_NONE, 0, tx);
 878 
 879         /*
 880          * We don't want to have to increase the meta-dnode's nlevels
 881          * later, because then we could do it in quescing context while
 882          * we are also accessing it in open context.
 883          *
 884          * This precaution is not necessary for the MOS (ds == NULL),
 885          * because the MOS is only updated in syncing context.
 886          * This is most fortunate: the MOS is the only objset that
 887          * needs to be synced multiple times as spa_sync() iterates
 888          * to convergence, so minimizing its dn_nlevels matters.
 889          */
 890         if (ds != NULL) {
 891                 int levels = 1;
 892 
 893                 /*
 894                  * Determine the number of levels necessary for the meta-dnode
 895                  * to contain DN_MAX_OBJECT dnodes.  Note that in order to
 896                  * ensure that we do not overflow 64 bits, there has to be
 897                  * a nlevels that gives us a number of blocks > DN_MAX_OBJECT
 898                  * but < 2^64.  Therefore,
 899                  * (mdn->dn_indblkshift - SPA_BLKPTRSHIFT) (10) must be
 900                  * less than (64 - log2(DN_MAX_OBJECT)) (16).
 901                  */
 902                 while ((uint64_t)mdn->dn_nblkptr <<
 903                     (mdn->dn_datablkshift - DNODE_SHIFT +
 904                     (levels - 1) * (mdn->dn_indblkshift - SPA_BLKPTRSHIFT)) <
 905                     DN_MAX_OBJECT)
 906                         levels++;
 907 
 908                 mdn->dn_next_nlevels[tx->tx_txg & TXG_MASK] =
 909                     mdn->dn_nlevels = levels;
 910         }
 911 
 912         ASSERT(type != DMU_OST_NONE);
 913         ASSERT(type != DMU_OST_ANY);
 914         ASSERT(type < DMU_OST_NUMTYPES);
 915         os->os_phys->os_type = type;
 916         if (dmu_objset_userused_enabled(os)) {
 917                 os->os_phys->os_flags |= OBJSET_FLAG_USERACCOUNTING_COMPLETE;
 918                 os->os_flags = os->os_phys->os_flags;
 919         }
 920 
 921         dsl_dataset_dirty(ds, tx);
 922 
 923         return (os);
 924 }
 925 
 926 typedef struct dmu_objset_create_arg {
 927         const char *doca_name;
 928         cred_t *doca_cred;
 929         void (*doca_userfunc)(objset_t *os, void *arg,
 930             cred_t *cr, dmu_tx_t *tx);
 931         void *doca_userarg;
 932         dmu_objset_type_t doca_type;
 933         uint64_t doca_flags;
 934 } dmu_objset_create_arg_t;
 935 
 936 /*ARGSUSED*/
 937 static int
 938 dmu_objset_create_check(void *arg, dmu_tx_t *tx)
 939 {
 940         dmu_objset_create_arg_t *doca = arg;
 941         dsl_pool_t *dp = dmu_tx_pool(tx);
 942         dsl_dir_t *pdd;
 943         const char *tail;
 944         int error;
 945 
 946         if (strchr(doca->doca_name, '@') != NULL)
 947                 return (SET_ERROR(EINVAL));
 948 
 949         if (strlen(doca->doca_name) >= ZFS_MAX_DATASET_NAME_LEN)
 950                 return (SET_ERROR(ENAMETOOLONG));
 951 
 952         error = dsl_dir_hold(dp, doca->doca_name, FTAG, &pdd, &tail);
 953         if (error != 0)
 954                 return (error);
 955         if (tail == NULL) {
 956                 dsl_dir_rele(pdd, FTAG);
 957                 return (SET_ERROR(EEXIST));
 958         }
 959         error = dsl_fs_ss_limit_check(pdd, 1, ZFS_PROP_FILESYSTEM_LIMIT, NULL,
 960             doca->doca_cred);
 961         dsl_dir_rele(pdd, FTAG);
 962 
 963         return (error);
 964 }
 965 
 966 static void
 967 dmu_objset_create_sync(void *arg, dmu_tx_t *tx)
 968 {
 969         dmu_objset_create_arg_t *doca = arg;
 970         dsl_pool_t *dp = dmu_tx_pool(tx);
 971         dsl_dir_t *pdd;
 972         const char *tail;
 973         dsl_dataset_t *ds;
 974         uint64_t obj;
 975         blkptr_t *bp;
 976         objset_t *os;
 977 
 978         VERIFY0(dsl_dir_hold(dp, doca->doca_name, FTAG, &pdd, &tail));
 979 
 980         obj = dsl_dataset_create_sync(pdd, tail, NULL, doca->doca_flags,
 981             doca->doca_cred, tx);
 982 
 983         VERIFY0(dsl_dataset_hold_obj(pdd->dd_pool, obj, FTAG, &ds));
 984         rrw_enter(&ds->ds_bp_rwlock, RW_READER, FTAG);
 985         bp = dsl_dataset_get_blkptr(ds);
 986         os = dmu_objset_create_impl(pdd->dd_pool->dp_spa,
 987             ds, bp, doca->doca_type, tx);
 988         rrw_exit(&ds->ds_bp_rwlock, FTAG);
 989 
 990         if (doca->doca_userfunc != NULL) {
 991                 doca->doca_userfunc(os, doca->doca_userarg,
 992                     doca->doca_cred, tx);
 993         }
 994 
 995         spa_history_log_internal_ds(ds, "create", tx, "");
 996         dsl_dataset_rele(ds, FTAG);
 997         dsl_dir_rele(pdd, FTAG);
 998 }
 999 
1000 int
1001 dmu_objset_create(const char *name, dmu_objset_type_t type, uint64_t flags,
1002     void (*func)(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx), void *arg)
1003 {
1004         dmu_objset_create_arg_t doca;
1005 
1006         doca.doca_name = name;
1007         doca.doca_cred = CRED();
1008         doca.doca_flags = flags;
1009         doca.doca_userfunc = func;
1010         doca.doca_userarg = arg;
1011         doca.doca_type = type;
1012 
1013         return (dsl_sync_task(name,
1014             dmu_objset_create_check, dmu_objset_create_sync, &doca,
1015             5, ZFS_SPACE_CHECK_NORMAL));
1016 }
1017 
1018 typedef struct dmu_objset_clone_arg {
1019         const char *doca_clone;
1020         const char *doca_origin;
1021         cred_t *doca_cred;
1022 } dmu_objset_clone_arg_t;
1023 
1024 /*ARGSUSED*/
1025 static int
1026 dmu_objset_clone_check(void *arg, dmu_tx_t *tx)
1027 {
1028         dmu_objset_clone_arg_t *doca = arg;
1029         dsl_dir_t *pdd;
1030         const char *tail;
1031         int error;
1032         dsl_dataset_t *origin;
1033         dsl_pool_t *dp = dmu_tx_pool(tx);
1034 
1035         if (strchr(doca->doca_clone, '@') != NULL)
1036                 return (SET_ERROR(EINVAL));
1037 
1038         if (strlen(doca->doca_clone) >= ZFS_MAX_DATASET_NAME_LEN)
1039                 return (SET_ERROR(ENAMETOOLONG));
1040 
1041         error = dsl_dir_hold(dp, doca->doca_clone, FTAG, &pdd, &tail);
1042         if (error != 0)
1043                 return (error);
1044         if (tail == NULL) {
1045                 dsl_dir_rele(pdd, FTAG);
1046                 return (SET_ERROR(EEXIST));
1047         }
1048 
1049         error = dsl_fs_ss_limit_check(pdd, 1, ZFS_PROP_FILESYSTEM_LIMIT, NULL,
1050             doca->doca_cred);
1051         if (error != 0) {
1052                 dsl_dir_rele(pdd, FTAG);
1053                 return (SET_ERROR(EDQUOT));
1054         }
1055         dsl_dir_rele(pdd, FTAG);
1056 
1057         error = dsl_dataset_hold(dp, doca->doca_origin, FTAG, &origin);
1058         if (error != 0)
1059                 return (error);
1060 
1061         /* You can only clone snapshots, not the head datasets. */
1062         if (!origin->ds_is_snapshot) {
1063                 dsl_dataset_rele(origin, FTAG);
1064                 return (SET_ERROR(EINVAL));
1065         }
1066         dsl_dataset_rele(origin, FTAG);
1067 
1068         return (0);
1069 }
1070 
1071 static void
1072 dmu_objset_clone_sync(void *arg, dmu_tx_t *tx)
1073 {
1074         dmu_objset_clone_arg_t *doca = arg;
1075         dsl_pool_t *dp = dmu_tx_pool(tx);
1076         dsl_dir_t *pdd;
1077         const char *tail;
1078         dsl_dataset_t *origin, *ds;
1079         uint64_t obj;
1080         char namebuf[ZFS_MAX_DATASET_NAME_LEN];
1081 
1082         VERIFY0(dsl_dir_hold(dp, doca->doca_clone, FTAG, &pdd, &tail));
1083         VERIFY0(dsl_dataset_hold(dp, doca->doca_origin, FTAG, &origin));
1084 
1085         obj = dsl_dataset_create_sync(pdd, tail, origin, 0,
1086             doca->doca_cred, tx);
1087 
1088         VERIFY0(dsl_dataset_hold_obj(pdd->dd_pool, obj, FTAG, &ds));
1089         dsl_dataset_name(origin, namebuf);
1090         spa_history_log_internal_ds(ds, "clone", tx,
1091             "origin=%s (%llu)", namebuf, origin->ds_object);
1092         dsl_dataset_rele(ds, FTAG);
1093         dsl_dataset_rele(origin, FTAG);
1094         dsl_dir_rele(pdd, FTAG);
1095 }
1096 
1097 int
1098 dmu_objset_clone(const char *clone, const char *origin)
1099 {
1100         dmu_objset_clone_arg_t doca;
1101 
1102         doca.doca_clone = clone;
1103         doca.doca_origin = origin;
1104         doca.doca_cred = CRED();
1105 
1106         return (dsl_sync_task(clone,
1107             dmu_objset_clone_check, dmu_objset_clone_sync, &doca,
1108             5, ZFS_SPACE_CHECK_NORMAL));
1109 }
1110 
1111 int
1112 dmu_objset_snapshot_one(const char *fsname, const char *snapname)
1113 {
1114         int err;
1115         char *longsnap = kmem_asprintf("%s@%s", fsname, snapname);
1116         nvlist_t *snaps = fnvlist_alloc();
1117 
1118         fnvlist_add_boolean(snaps, longsnap);
1119         strfree(longsnap);
1120         err = dsl_dataset_snapshot(snaps, NULL, NULL);
1121         fnvlist_free(snaps);
1122         return (err);
1123 }
1124 
1125 static void
1126 dmu_objset_sync_dnodes(multilist_sublist_t *list, dmu_tx_t *tx)
1127 {
1128         dnode_t *dn;
1129 
1130         while ((dn = multilist_sublist_head(list)) != NULL) {
1131                 ASSERT(dn->dn_object != DMU_META_DNODE_OBJECT);
1132                 ASSERT(dn->dn_dbuf->db_data_pending);
1133                 /*
1134                  * Initialize dn_zio outside dnode_sync() because the
1135                  * meta-dnode needs to set it ouside dnode_sync().
1136                  */
1137                 dn->dn_zio = dn->dn_dbuf->db_data_pending->dr_zio;
1138                 ASSERT(dn->dn_zio);
1139 
1140                 ASSERT3U(dn->dn_nlevels, <=, DN_MAX_LEVELS);
1141                 multilist_sublist_remove(list, dn);
1142 
1143                 multilist_t *newlist = dn->dn_objset->os_synced_dnodes;
1144                 if (newlist != NULL) {
1145                         (void) dnode_add_ref(dn, newlist);
1146                         multilist_insert(newlist, dn);
1147                 }
1148 
1149                 dnode_sync(dn, tx);
1150         }
1151 }
1152 
1153 /* ARGSUSED */
1154 static void
1155 dmu_objset_write_ready(zio_t *zio, arc_buf_t *abuf, void *arg)
1156 {
1157         blkptr_t *bp = zio->io_bp;
1158         objset_t *os = arg;
1159         dnode_phys_t *dnp = &os->os_phys->os_meta_dnode;
1160 
1161         ASSERT(!BP_IS_EMBEDDED(bp));
1162         ASSERT3U(BP_GET_TYPE(bp), ==, DMU_OT_OBJSET);
1163         ASSERT0(BP_GET_LEVEL(bp));
1164 
1165         /*
1166          * Update rootbp fill count: it should be the number of objects
1167          * allocated in the object set (not counting the "special"
1168          * objects that are stored in the objset_phys_t -- the meta
1169          * dnode and user/group accounting objects).
1170          */
1171         bp->blk_fill = 0;
1172         for (int i = 0; i < dnp->dn_nblkptr; i++)
1173                 bp->blk_fill += BP_GET_FILL(&dnp->dn_blkptr[i]);
1174         if (os->os_dsl_dataset != NULL)
1175                 rrw_enter(&os->os_dsl_dataset->ds_bp_rwlock, RW_WRITER, FTAG);
1176         *os->os_rootbp = *bp;
1177         if (os->os_dsl_dataset != NULL)
1178                 rrw_exit(&os->os_dsl_dataset->ds_bp_rwlock, FTAG);
1179 }
1180 
1181 /* ARGSUSED */
1182 static void
1183 dmu_objset_write_done(zio_t *zio, arc_buf_t *abuf, void *arg)
1184 {
1185         blkptr_t *bp = zio->io_bp;
1186         blkptr_t *bp_orig = &zio->io_bp_orig;
1187         objset_t *os = arg;
1188 
1189         if (zio->io_flags & ZIO_FLAG_IO_REWRITE) {
1190                 ASSERT(BP_EQUAL(bp, bp_orig));
1191         } else {
1192                 dsl_dataset_t *ds = os->os_dsl_dataset;
1193                 dmu_tx_t *tx = os->os_synctx;
1194 
1195                 (void) dsl_dataset_block_kill(ds, bp_orig, tx, B_TRUE);
1196                 dsl_dataset_block_born(ds, bp, tx);
1197         }
1198         kmem_free(bp, sizeof (*bp));
1199 }
1200 
1201 typedef struct sync_dnodes_arg {
1202         multilist_t *sda_list;
1203         int sda_sublist_idx;
1204         multilist_t *sda_newlist;
1205         dmu_tx_t *sda_tx;
1206 } sync_dnodes_arg_t;
1207 
1208 static void
1209 sync_dnodes_task(void *arg)
1210 {
1211         sync_dnodes_arg_t *sda = arg;
1212 
1213         multilist_sublist_t *ms =
1214             multilist_sublist_lock(sda->sda_list, sda->sda_sublist_idx);
1215 
1216         dmu_objset_sync_dnodes(ms, sda->sda_tx);
1217 
1218         multilist_sublist_unlock(ms);
1219 
1220         kmem_free(sda, sizeof (*sda));
1221 }
1222 
1223 
1224 /* called from dsl */
1225 void
1226 dmu_objset_sync(objset_t *os, zio_t *pio, dmu_tx_t *tx)
1227 {
1228         int txgoff;
1229         zbookmark_phys_t zb;
1230         zio_prop_t zp;
1231         zio_t *zio;
1232         list_t *list;
1233         dbuf_dirty_record_t *dr;
1234         blkptr_t *blkptr_copy = kmem_alloc(sizeof (*os->os_rootbp), KM_SLEEP);
1235         *blkptr_copy = *os->os_rootbp;
1236 
1237         dprintf_ds(os->os_dsl_dataset, "txg=%llu\n", tx->tx_txg);
1238 
1239         ASSERT(dmu_tx_is_syncing(tx));
1240         /* XXX the write_done callback should really give us the tx... */
1241         os->os_synctx = tx;
1242 
1243         if (os->os_dsl_dataset == NULL) {
1244                 /*
1245                  * This is the MOS.  If we have upgraded,
1246                  * spa_max_replication() could change, so reset
1247                  * os_copies here.
1248                  */
1249                 os->os_copies = spa_max_replication(os->os_spa);
1250         }
1251 
1252         /*
1253          * Create the root block IO
1254          */
1255         SET_BOOKMARK(&zb, os->os_dsl_dataset ?
1256             os->os_dsl_dataset->ds_object : DMU_META_OBJSET,
1257             ZB_ROOT_OBJECT, ZB_ROOT_LEVEL, ZB_ROOT_BLKID);
1258         arc_release(os->os_phys_buf, &os->os_phys_buf);
1259 
1260         dmu_write_policy(os, NULL, 0, 0, &zp);
1261 
1262         zio = arc_write(pio, os->os_spa, tx->tx_txg,
1263             blkptr_copy, os->os_phys_buf, DMU_OS_IS_L2CACHEABLE(os),
1264             &zp, dmu_objset_write_ready, NULL, NULL, dmu_objset_write_done,
1265             os, ZIO_PRIORITY_ASYNC_WRITE, ZIO_FLAG_MUSTSUCCEED, &zb, NULL);
1266 
1267         /*
1268          * Sync special dnodes - the parent IO for the sync is the root block
1269          */
1270         DMU_META_DNODE(os)->dn_zio = zio;
1271         dnode_sync(DMU_META_DNODE(os), tx);
1272 
1273         os->os_phys->os_flags = os->os_flags;
1274 
1275         if (DMU_USERUSED_DNODE(os) &&
1276             DMU_USERUSED_DNODE(os)->dn_type != DMU_OT_NONE) {
1277                 DMU_USERUSED_DNODE(os)->dn_zio = zio;
1278                 dnode_sync(DMU_USERUSED_DNODE(os), tx);
1279                 DMU_GROUPUSED_DNODE(os)->dn_zio = zio;
1280                 dnode_sync(DMU_GROUPUSED_DNODE(os), tx);
1281         }
1282 
1283         txgoff = tx->tx_txg & TXG_MASK;
1284 
1285         if (dmu_objset_userused_enabled(os)) {
1286                 /*
1287                  * We must create the list here because it uses the
1288                  * dn_dirty_link[] of this txg.  But it may already
1289                  * exist because we call dsl_dataset_sync() twice per txg.
1290                  */
1291                 if (os->os_synced_dnodes == NULL) {
1292                         os->os_synced_dnodes =
1293                             multilist_create(sizeof (dnode_t),
1294                             offsetof(dnode_t, dn_dirty_link[txgoff]),
1295                             dnode_multilist_index_func);
1296                 } else {
1297                         ASSERT3U(os->os_synced_dnodes->ml_offset, ==,
1298                             offsetof(dnode_t, dn_dirty_link[txgoff]));
1299                 }
1300         }
1301 
1302         for (int i = 0;
1303             i < multilist_get_num_sublists(os->os_dirty_dnodes[txgoff]); i++) {
1304                 sync_dnodes_arg_t *sda = kmem_alloc(sizeof (*sda), KM_SLEEP);
1305                 sda->sda_list = os->os_dirty_dnodes[txgoff];
1306                 sda->sda_sublist_idx = i;
1307                 sda->sda_tx = tx;
1308                 (void) taskq_dispatch(dmu_objset_pool(os)->dp_sync_taskq,
1309                     sync_dnodes_task, sda, 0);
1310                 /* callback frees sda */
1311         }
1312         taskq_wait(dmu_objset_pool(os)->dp_sync_taskq);
1313 
1314         list = &DMU_META_DNODE(os)->dn_dirty_records[txgoff];
1315         while ((dr = list_head(list)) != NULL) {
1316                 ASSERT0(dr->dr_dbuf->db_level);
1317                 list_remove(list, dr);
1318                 if (dr->dr_zio)
1319                         zio_nowait(dr->dr_zio);
1320         }
1321 
1322         /* Enable dnode backfill if enough objects have been freed. */
1323         if (os->os_freed_dnodes >= dmu_rescan_dnode_threshold) {
1324                 os->os_rescan_dnodes = B_TRUE;
1325                 os->os_freed_dnodes = 0;
1326         }
1327 
1328         /*
1329          * Free intent log blocks up to this tx.
1330          */
1331         zil_sync(os->os_zil, tx);
1332         os->os_phys->os_zil_header = os->os_zil_header;
1333         zio_nowait(zio);
1334 }
1335 
1336 boolean_t
1337 dmu_objset_is_dirty(objset_t *os, uint64_t txg)
1338 {
1339         return (!multilist_is_empty(os->os_dirty_dnodes[txg & TXG_MASK]));
1340 }
1341 
1342 static objset_used_cb_t *used_cbs[DMU_OST_NUMTYPES];
1343 
1344 void
1345 dmu_objset_register_type(dmu_objset_type_t ost, objset_used_cb_t *cb)
1346 {
1347         used_cbs[ost] = cb;
1348 }
1349 
1350 boolean_t
1351 dmu_objset_userused_enabled(objset_t *os)
1352 {
1353         return (spa_version(os->os_spa) >= SPA_VERSION_USERSPACE &&
1354             used_cbs[os->os_phys->os_type] != NULL &&
1355             DMU_USERUSED_DNODE(os) != NULL);
1356 }
1357 
1358 typedef struct userquota_node {
1359         uint64_t uqn_id;
1360         int64_t uqn_delta;
1361         avl_node_t uqn_node;
1362 } userquota_node_t;
1363 
1364 typedef struct userquota_cache {
1365         avl_tree_t uqc_user_deltas;
1366         avl_tree_t uqc_group_deltas;
1367 } userquota_cache_t;
1368 
1369 static int
1370 userquota_compare(const void *l, const void *r)
1371 {
1372         const userquota_node_t *luqn = l;
1373         const userquota_node_t *ruqn = r;
1374 
1375         if (luqn->uqn_id < ruqn->uqn_id)
1376                 return (-1);
1377         if (luqn->uqn_id > ruqn->uqn_id)
1378                 return (1);
1379         return (0);
1380 }
1381 
1382 static void
1383 do_userquota_cacheflush(objset_t *os, userquota_cache_t *cache, dmu_tx_t *tx)
1384 {
1385         void *cookie;
1386         userquota_node_t *uqn;
1387 
1388         ASSERT(dmu_tx_is_syncing(tx));
1389 
1390         cookie = NULL;
1391         while ((uqn = avl_destroy_nodes(&cache->uqc_user_deltas,
1392             &cookie)) != NULL) {
1393                 /*
1394                  * os_userused_lock protects against concurrent calls to
1395                  * zap_increment_int().  It's needed because zap_increment_int()
1396                  * is not thread-safe (i.e. not atomic).
1397                  */
1398                 mutex_enter(&os->os_userused_lock);
1399                 VERIFY0(zap_increment_int(os, DMU_USERUSED_OBJECT,
1400                     uqn->uqn_id, uqn->uqn_delta, tx));
1401                 mutex_exit(&os->os_userused_lock);
1402                 kmem_free(uqn, sizeof (*uqn));
1403         }
1404         avl_destroy(&cache->uqc_user_deltas);
1405 
1406         cookie = NULL;
1407         while ((uqn = avl_destroy_nodes(&cache->uqc_group_deltas,
1408             &cookie)) != NULL) {
1409                 mutex_enter(&os->os_userused_lock);
1410                 VERIFY0(zap_increment_int(os, DMU_GROUPUSED_OBJECT,
1411                     uqn->uqn_id, uqn->uqn_delta, tx));
1412                 mutex_exit(&os->os_userused_lock);
1413                 kmem_free(uqn, sizeof (*uqn));
1414         }
1415         avl_destroy(&cache->uqc_group_deltas);
1416 }
1417 
1418 static void
1419 userquota_update_cache(avl_tree_t *avl, uint64_t id, int64_t delta)
1420 {
1421         userquota_node_t search = { .uqn_id = id };
1422         avl_index_t idx;
1423 
1424         userquota_node_t *uqn = avl_find(avl, &search, &idx);
1425         if (uqn == NULL) {
1426                 uqn = kmem_zalloc(sizeof (*uqn), KM_SLEEP);
1427                 uqn->uqn_id = id;
1428                 avl_insert(avl, uqn, idx);
1429         }
1430         uqn->uqn_delta += delta;
1431 }
1432 
1433 static void
1434 do_userquota_update(userquota_cache_t *cache, uint64_t used, uint64_t flags,
1435     uint64_t user, uint64_t group, boolean_t subtract)
1436 {
1437         if ((flags & DNODE_FLAG_USERUSED_ACCOUNTED)) {
1438                 int64_t delta = DNODE_SIZE + used;
1439                 if (subtract)
1440                         delta = -delta;
1441 
1442                 userquota_update_cache(&cache->uqc_user_deltas, user, delta);
1443                 userquota_update_cache(&cache->uqc_group_deltas, group, delta);
1444         }
1445 }
1446 
1447 typedef struct userquota_updates_arg {
1448         objset_t *uua_os;
1449         int uua_sublist_idx;
1450         dmu_tx_t *uua_tx;
1451 } userquota_updates_arg_t;
1452 
1453 static void
1454 userquota_updates_task(void *arg)
1455 {
1456         userquota_updates_arg_t *uua = arg;
1457         objset_t *os = uua->uua_os;
1458         dmu_tx_t *tx = uua->uua_tx;
1459         dnode_t *dn;
1460         userquota_cache_t cache = { 0 };
1461 
1462         multilist_sublist_t *list =
1463             multilist_sublist_lock(os->os_synced_dnodes, uua->uua_sublist_idx);
1464 
1465         ASSERT(multilist_sublist_head(list) == NULL ||
1466             dmu_objset_userused_enabled(os));
1467         avl_create(&cache.uqc_user_deltas, userquota_compare,
1468             sizeof (userquota_node_t), offsetof(userquota_node_t, uqn_node));
1469         avl_create(&cache.uqc_group_deltas, userquota_compare,
1470             sizeof (userquota_node_t), offsetof(userquota_node_t, uqn_node));
1471 
1472         while ((dn = multilist_sublist_head(list)) != NULL) {
1473                 int flags;
1474                 ASSERT(!DMU_OBJECT_IS_SPECIAL(dn->dn_object));
1475                 ASSERT(dn->dn_phys->dn_type == DMU_OT_NONE ||
1476                     dn->dn_phys->dn_flags &
1477                     DNODE_FLAG_USERUSED_ACCOUNTED);
1478 
1479                 flags = dn->dn_id_flags;
1480                 ASSERT(flags);
1481                 if (flags & DN_ID_OLD_EXIST)  {
1482                         do_userquota_update(&cache,
1483                             dn->dn_oldused, dn->dn_oldflags,
1484                             dn->dn_olduid, dn->dn_oldgid, B_TRUE);
1485                 }
1486                 if (flags & DN_ID_NEW_EXIST) {
1487                         do_userquota_update(&cache,
1488                             DN_USED_BYTES(dn->dn_phys),
1489                             dn->dn_phys->dn_flags,  dn->dn_newuid,
1490                             dn->dn_newgid, B_FALSE);
1491                 }
1492 
1493                 mutex_enter(&dn->dn_mtx);
1494                 dn->dn_oldused = 0;
1495                 dn->dn_oldflags = 0;
1496                 if (dn->dn_id_flags & DN_ID_NEW_EXIST) {
1497                         dn->dn_olduid = dn->dn_newuid;
1498                         dn->dn_oldgid = dn->dn_newgid;
1499                         dn->dn_id_flags |= DN_ID_OLD_EXIST;
1500                         if (dn->dn_bonuslen == 0)
1501                                 dn->dn_id_flags |= DN_ID_CHKED_SPILL;
1502                         else
1503                                 dn->dn_id_flags |= DN_ID_CHKED_BONUS;
1504                 }
1505                 dn->dn_id_flags &= ~(DN_ID_NEW_EXIST);
1506                 mutex_exit(&dn->dn_mtx);
1507 
1508                 multilist_sublist_remove(list, dn);
1509                 dnode_rele(dn, os->os_synced_dnodes);
1510         }
1511         do_userquota_cacheflush(os, &cache, tx);
1512         multilist_sublist_unlock(list);
1513         kmem_free(uua, sizeof (*uua));
1514 }
1515 
1516 void
1517 dmu_objset_do_userquota_updates(objset_t *os, dmu_tx_t *tx)
1518 {
1519         if (!dmu_objset_userused_enabled(os))
1520                 return;
1521 
1522         /* Allocate the user/groupused objects if necessary. */
1523         if (DMU_USERUSED_DNODE(os)->dn_type == DMU_OT_NONE) {
1524                 VERIFY0(zap_create_claim(os,
1525                     DMU_USERUSED_OBJECT,
1526                     DMU_OT_USERGROUP_USED, DMU_OT_NONE, 0, tx));
1527                 VERIFY0(zap_create_claim(os,
1528                     DMU_GROUPUSED_OBJECT,
1529                     DMU_OT_USERGROUP_USED, DMU_OT_NONE, 0, tx));
1530         }
1531 
1532         for (int i = 0;
1533             i < multilist_get_num_sublists(os->os_synced_dnodes); i++) {
1534                 userquota_updates_arg_t *uua =
1535                     kmem_alloc(sizeof (*uua), KM_SLEEP);
1536                 uua->uua_os = os;
1537                 uua->uua_sublist_idx = i;
1538                 uua->uua_tx = tx;
1539                 /* note: caller does taskq_wait() */
1540                 (void) taskq_dispatch(dmu_objset_pool(os)->dp_sync_taskq,
1541                     userquota_updates_task, uua, 0);
1542                 /* callback frees uua */
1543         }
1544 }
1545 
1546 /*
1547  * Returns a pointer to data to find uid/gid from
1548  *
1549  * If a dirty record for transaction group that is syncing can't
1550  * be found then NULL is returned.  In the NULL case it is assumed
1551  * the uid/gid aren't changing.
1552  */
1553 static void *
1554 dmu_objset_userquota_find_data(dmu_buf_impl_t *db, dmu_tx_t *tx)
1555 {
1556         dbuf_dirty_record_t *dr, **drp;
1557         void *data;
1558 
1559         if (db->db_dirtycnt == 0)
1560                 return (db->db.db_data);  /* Nothing is changing */
1561 
1562         for (drp = &db->db_last_dirty; (dr = *drp) != NULL; drp = &dr->dr_next)
1563                 if (dr->dr_txg == tx->tx_txg)
1564                         break;
1565 
1566         if (dr == NULL) {
1567                 data = NULL;
1568         } else {
1569                 dnode_t *dn;
1570 
1571                 DB_DNODE_ENTER(dr->dr_dbuf);
1572                 dn = DB_DNODE(dr->dr_dbuf);
1573 
1574                 if (dn->dn_bonuslen == 0 &&
1575                     dr->dr_dbuf->db_blkid == DMU_SPILL_BLKID)
1576                         data = dr->dt.dl.dr_data->b_data;
1577                 else
1578                         data = dr->dt.dl.dr_data;
1579 
1580                 DB_DNODE_EXIT(dr->dr_dbuf);
1581         }
1582 
1583         return (data);
1584 }
1585 
1586 void
1587 dmu_objset_userquota_get_ids(dnode_t *dn, boolean_t before, dmu_tx_t *tx)
1588 {
1589         objset_t *os = dn->dn_objset;
1590         void *data = NULL;
1591         dmu_buf_impl_t *db = NULL;
1592         uint64_t *user = NULL;
1593         uint64_t *group = NULL;
1594         int flags = dn->dn_id_flags;
1595         int error;
1596         boolean_t have_spill = B_FALSE;
1597 
1598         if (!dmu_objset_userused_enabled(dn->dn_objset))
1599                 return;
1600 
1601         if (before && (flags & (DN_ID_CHKED_BONUS|DN_ID_OLD_EXIST|
1602             DN_ID_CHKED_SPILL)))
1603                 return;
1604 
1605         if (before && dn->dn_bonuslen != 0)
1606                 data = DN_BONUS(dn->dn_phys);
1607         else if (!before && dn->dn_bonuslen != 0) {
1608                 if (dn->dn_bonus) {
1609                         db = dn->dn_bonus;
1610                         mutex_enter(&db->db_mtx);
1611                         data = dmu_objset_userquota_find_data(db, tx);
1612                 } else {
1613                         data = DN_BONUS(dn->dn_phys);
1614                 }
1615         } else if (dn->dn_bonuslen == 0 && dn->dn_bonustype == DMU_OT_SA) {
1616                         int rf = 0;
1617 
1618                         if (RW_WRITE_HELD(&dn->dn_struct_rwlock))
1619                                 rf |= DB_RF_HAVESTRUCT;
1620                         error = dmu_spill_hold_by_dnode(dn,
1621                             rf | DB_RF_MUST_SUCCEED,
1622                             FTAG, (dmu_buf_t **)&db);
1623                         ASSERT(error == 0);
1624                         mutex_enter(&db->db_mtx);
1625                         data = (before) ? db->db.db_data :
1626                             dmu_objset_userquota_find_data(db, tx);
1627                         have_spill = B_TRUE;
1628         } else {
1629                 mutex_enter(&dn->dn_mtx);
1630                 dn->dn_id_flags |= DN_ID_CHKED_BONUS;
1631                 mutex_exit(&dn->dn_mtx);
1632                 return;
1633         }
1634 
1635         if (before) {
1636                 ASSERT(data);
1637                 user = &dn->dn_olduid;
1638                 group = &dn->dn_oldgid;
1639         } else if (data) {
1640                 user = &dn->dn_newuid;
1641                 group = &dn->dn_newgid;
1642         }
1643 
1644         /*
1645          * Must always call the callback in case the object
1646          * type has changed and that type isn't an object type to track
1647          */
1648         error = used_cbs[os->os_phys->os_type](dn->dn_bonustype, data,
1649             user, group);
1650 
1651         /*
1652          * Preserve existing uid/gid when the callback can't determine
1653          * what the new uid/gid are and the callback returned EEXIST.
1654          * The EEXIST error tells us to just use the existing uid/gid.
1655          * If we don't know what the old values are then just assign
1656          * them to 0, since that is a new file  being created.
1657          */
1658         if (!before && data == NULL && error == EEXIST) {
1659                 if (flags & DN_ID_OLD_EXIST) {
1660                         dn->dn_newuid = dn->dn_olduid;
1661                         dn->dn_newgid = dn->dn_oldgid;
1662                 } else {
1663                         dn->dn_newuid = 0;
1664                         dn->dn_newgid = 0;
1665                 }
1666                 error = 0;
1667         }
1668 
1669         if (db)
1670                 mutex_exit(&db->db_mtx);
1671 
1672         mutex_enter(&dn->dn_mtx);
1673         if (error == 0 && before)
1674                 dn->dn_id_flags |= DN_ID_OLD_EXIST;
1675         if (error == 0 && !before)
1676                 dn->dn_id_flags |= DN_ID_NEW_EXIST;
1677 
1678         if (have_spill) {
1679                 dn->dn_id_flags |= DN_ID_CHKED_SPILL;
1680         } else {
1681                 dn->dn_id_flags |= DN_ID_CHKED_BONUS;
1682         }
1683         mutex_exit(&dn->dn_mtx);
1684         if (have_spill)
1685                 dmu_buf_rele((dmu_buf_t *)db, FTAG);
1686 }
1687 
1688 boolean_t
1689 dmu_objset_userspace_present(objset_t *os)
1690 {
1691         return (os->os_phys->os_flags &
1692             OBJSET_FLAG_USERACCOUNTING_COMPLETE);
1693 }
1694 
1695 int
1696 dmu_objset_userspace_upgrade(objset_t *os)
1697 {
1698         uint64_t obj;
1699         int err = 0;
1700 
1701         if (dmu_objset_userspace_present(os))
1702                 return (0);
1703         if (!dmu_objset_userused_enabled(os))
1704                 return (SET_ERROR(ENOTSUP));
1705         if (dmu_objset_is_snapshot(os))
1706                 return (SET_ERROR(EINVAL));
1707 
1708         /*
1709          * We simply need to mark every object dirty, so that it will be
1710          * synced out and now accounted.  If this is called
1711          * concurrently, or if we already did some work before crashing,
1712          * that's fine, since we track each object's accounted state
1713          * independently.
1714          */
1715 
1716         for (obj = 0; err == 0; err = dmu_object_next(os, &obj, FALSE, 0)) {
1717                 dmu_tx_t *tx;
1718                 dmu_buf_t *db;
1719                 int objerr;
1720 
1721                 if (issig(JUSTLOOKING) && issig(FORREAL))
1722                         return (SET_ERROR(EINTR));
1723 
1724                 objerr = dmu_bonus_hold(os, obj, FTAG, &db);
1725                 if (objerr != 0)
1726                         continue;
1727                 tx = dmu_tx_create(os);
1728                 dmu_tx_hold_bonus(tx, obj);
1729                 objerr = dmu_tx_assign(tx, TXG_WAIT);
1730                 if (objerr != 0) {
1731                         dmu_tx_abort(tx);
1732                         continue;
1733                 }
1734                 dmu_buf_will_dirty(db, tx);
1735                 dmu_buf_rele(db, FTAG);
1736                 dmu_tx_commit(tx);
1737         }
1738 
1739         os->os_flags |= OBJSET_FLAG_USERACCOUNTING_COMPLETE;
1740         txg_wait_synced(dmu_objset_pool(os), 0);
1741         return (0);
1742 }
1743 
1744 void
1745 dmu_objset_space(objset_t *os, uint64_t *refdbytesp, uint64_t *availbytesp,
1746     uint64_t *usedobjsp, uint64_t *availobjsp)
1747 {
1748         dsl_dataset_space(os->os_dsl_dataset, refdbytesp, availbytesp,
1749             usedobjsp, availobjsp);
1750 }
1751 
1752 uint64_t
1753 dmu_objset_fsid_guid(objset_t *os)
1754 {
1755         return (dsl_dataset_fsid_guid(os->os_dsl_dataset));
1756 }
1757 
1758 void
1759 dmu_objset_fast_stat(objset_t *os, dmu_objset_stats_t *stat)
1760 {
1761         stat->dds_type = os->os_phys->os_type;
1762         if (os->os_dsl_dataset)
1763                 dsl_dataset_fast_stat(os->os_dsl_dataset, stat);
1764 }
1765 
1766 void
1767 dmu_objset_stats(objset_t *os, nvlist_t *nv)
1768 {
1769         ASSERT(os->os_dsl_dataset ||
1770             os->os_phys->os_type == DMU_OST_META);
1771 
1772         if (os->os_dsl_dataset != NULL)
1773                 dsl_dataset_stats(os->os_dsl_dataset, nv);
1774 
1775         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_TYPE,
1776             os->os_phys->os_type);
1777         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USERACCOUNTING,
1778             dmu_objset_userspace_present(os));
1779 }
1780 
1781 int
1782 dmu_objset_is_snapshot(objset_t *os)
1783 {
1784         if (os->os_dsl_dataset != NULL)
1785                 return (os->os_dsl_dataset->ds_is_snapshot);
1786         else
1787                 return (B_FALSE);
1788 }
1789 
1790 int
1791 dmu_snapshot_realname(objset_t *os, char *name, char *real, int maxlen,
1792     boolean_t *conflict)
1793 {
1794         dsl_dataset_t *ds = os->os_dsl_dataset;
1795         uint64_t ignored;
1796 
1797         if (dsl_dataset_phys(ds)->ds_snapnames_zapobj == 0)
1798                 return (SET_ERROR(ENOENT));
1799 
1800         return (zap_lookup_norm(ds->ds_dir->dd_pool->dp_meta_objset,
1801             dsl_dataset_phys(ds)->ds_snapnames_zapobj, name, 8, 1, &ignored,
1802             MT_NORMALIZE, real, maxlen, conflict));
1803 }
1804 
1805 int
1806 dmu_clone_list_next(objset_t *os, int len, char *name,
1807     uint64_t *idp, uint64_t *offp)
1808 {
1809         dsl_dataset_t *ds = os->os_dsl_dataset, *clone;
1810         zap_cursor_t cursor;
1811         zap_attribute_t attr;
1812         char buf[MAXNAMELEN];
1813 
1814         ASSERT(dsl_pool_config_held(dmu_objset_pool(os)));
1815 
1816         if (dsl_dataset_phys(ds)->ds_next_clones_obj == 0)
1817                 return (SET_ERROR(ENOENT));
1818 
1819         zap_cursor_init_serialized(&cursor,
1820             ds->ds_dir->dd_pool->dp_meta_objset,
1821             dsl_dataset_phys(ds)->ds_next_clones_obj, *offp);
1822 
1823         if (zap_cursor_retrieve(&cursor, &attr) != 0) {
1824                 zap_cursor_fini(&cursor);
1825                 return (SET_ERROR(ENOENT));
1826         }
1827 
1828         VERIFY0(dsl_dataset_hold_obj(ds->ds_dir->dd_pool,
1829             attr.za_first_integer, FTAG, &clone));
1830 
1831         dsl_dir_name(clone->ds_dir, buf);
1832 
1833         dsl_dataset_rele(clone, FTAG);
1834 
1835         if (strlen(buf) >= len) {
1836                 zap_cursor_fini(&cursor);
1837                 return (SET_ERROR(ENAMETOOLONG));
1838         }
1839 
1840         (void) strcpy(name, buf);
1841         if (idp != NULL)
1842                 *idp = attr.za_first_integer;
1843 
1844         zap_cursor_advance(&cursor);
1845         *offp = zap_cursor_serialize(&cursor);
1846         zap_cursor_fini(&cursor);
1847 
1848         return (0);
1849 }
1850 
1851 int
1852 dmu_snapshot_list_next(objset_t *os, int namelen, char *name,
1853     uint64_t *idp, uint64_t *offp, boolean_t *case_conflict)
1854 {
1855         dsl_dataset_t *ds = os->os_dsl_dataset;
1856         zap_cursor_t cursor;
1857         zap_attribute_t attr;
1858 
1859         ASSERT(dsl_pool_config_held(dmu_objset_pool(os)));
1860 
1861         if (dsl_dataset_phys(ds)->ds_snapnames_zapobj == 0)
1862                 return (SET_ERROR(ENOENT));
1863 
1864         zap_cursor_init_serialized(&cursor,
1865             ds->ds_dir->dd_pool->dp_meta_objset,
1866             dsl_dataset_phys(ds)->ds_snapnames_zapobj, *offp);
1867 
1868         if (zap_cursor_retrieve(&cursor, &attr) != 0) {
1869                 zap_cursor_fini(&cursor);
1870                 return (SET_ERROR(ENOENT));
1871         }
1872 
1873         if (strlen(attr.za_name) + 1 > namelen) {
1874                 zap_cursor_fini(&cursor);
1875                 return (SET_ERROR(ENAMETOOLONG));
1876         }
1877 
1878         (void) strcpy(name, attr.za_name);
1879         if (idp)
1880                 *idp = attr.za_first_integer;
1881         if (case_conflict)
1882                 *case_conflict = attr.za_normalization_conflict;
1883         zap_cursor_advance(&cursor);
1884         *offp = zap_cursor_serialize(&cursor);
1885         zap_cursor_fini(&cursor);
1886 
1887         return (0);
1888 }
1889 
1890 int
1891 dmu_dir_list_next(objset_t *os, int namelen, char *name,
1892     uint64_t *idp, uint64_t *offp)
1893 {
1894         dsl_dir_t *dd = os->os_dsl_dataset->ds_dir;
1895         zap_cursor_t cursor;
1896         zap_attribute_t attr;
1897 
1898         /* there is no next dir on a snapshot! */
1899         if (os->os_dsl_dataset->ds_object !=
1900             dsl_dir_phys(dd)->dd_head_dataset_obj)
1901                 return (SET_ERROR(ENOENT));
1902 
1903         zap_cursor_init_serialized(&cursor,
1904             dd->dd_pool->dp_meta_objset,
1905             dsl_dir_phys(dd)->dd_child_dir_zapobj, *offp);
1906 
1907         if (zap_cursor_retrieve(&cursor, &attr) != 0) {
1908                 zap_cursor_fini(&cursor);
1909                 return (SET_ERROR(ENOENT));
1910         }
1911 
1912         if (strlen(attr.za_name) + 1 > namelen) {
1913                 zap_cursor_fini(&cursor);
1914                 return (SET_ERROR(ENAMETOOLONG));
1915         }
1916 
1917         (void) strcpy(name, attr.za_name);
1918         if (idp)
1919                 *idp = attr.za_first_integer;
1920         zap_cursor_advance(&cursor);
1921         *offp = zap_cursor_serialize(&cursor);
1922         zap_cursor_fini(&cursor);
1923 
1924         return (0);
1925 }
1926 
1927 typedef struct dmu_objset_find_ctx {
1928         taskq_t         *dc_tq;
1929         dsl_pool_t      *dc_dp;
1930         uint64_t        dc_ddobj;
1931         char            *dc_ddname; /* last component of ddobj's name */
1932         int             (*dc_func)(dsl_pool_t *, dsl_dataset_t *, void *);
1933         void            *dc_arg;
1934         int             dc_flags;
1935         kmutex_t        *dc_error_lock;
1936         int             *dc_error;
1937 } dmu_objset_find_ctx_t;
1938 
1939 static void
1940 dmu_objset_find_dp_impl(dmu_objset_find_ctx_t *dcp)
1941 {
1942         dsl_pool_t *dp = dcp->dc_dp;
1943         dsl_dir_t *dd;
1944         dsl_dataset_t *ds;
1945         zap_cursor_t zc;
1946         zap_attribute_t *attr;
1947         uint64_t thisobj;
1948         int err = 0;
1949 
1950         /* don't process if there already was an error */
1951         if (*dcp->dc_error != 0)
1952                 goto out;
1953 
1954         /*
1955          * Note: passing the name (dc_ddname) here is optional, but it
1956          * improves performance because we don't need to call
1957          * zap_value_search() to determine the name.
1958          */
1959         err = dsl_dir_hold_obj(dp, dcp->dc_ddobj, dcp->dc_ddname, FTAG, &dd);
1960         if (err != 0)
1961                 goto out;
1962 
1963         /* Don't visit hidden ($MOS & $ORIGIN) objsets. */
1964         if (dd->dd_myname[0] == '$') {
1965                 dsl_dir_rele(dd, FTAG);
1966                 goto out;
1967         }
1968 
1969         thisobj = dsl_dir_phys(dd)->dd_head_dataset_obj;
1970         attr = kmem_alloc(sizeof (zap_attribute_t), KM_SLEEP);
1971 
1972         /*
1973          * Iterate over all children.
1974          */
1975         if (dcp->dc_flags & DS_FIND_CHILDREN) {
1976                 for (zap_cursor_init(&zc, dp->dp_meta_objset,
1977                     dsl_dir_phys(dd)->dd_child_dir_zapobj);
1978                     zap_cursor_retrieve(&zc, attr) == 0;
1979                     (void) zap_cursor_advance(&zc)) {
1980                         ASSERT3U(attr->za_integer_length, ==,
1981                             sizeof (uint64_t));
1982                         ASSERT3U(attr->za_num_integers, ==, 1);
1983 
1984                         dmu_objset_find_ctx_t *child_dcp =
1985                             kmem_alloc(sizeof (*child_dcp), KM_SLEEP);
1986                         *child_dcp = *dcp;
1987                         child_dcp->dc_ddobj = attr->za_first_integer;
1988                         child_dcp->dc_ddname = spa_strdup(attr->za_name);
1989                         if (dcp->dc_tq != NULL)
1990                                 (void) taskq_dispatch(dcp->dc_tq,
1991                                     dmu_objset_find_dp_cb, child_dcp, TQ_SLEEP);
1992                         else
1993                                 dmu_objset_find_dp_impl(child_dcp);
1994                 }
1995                 zap_cursor_fini(&zc);
1996         }
1997 
1998         /*
1999          * Iterate over all snapshots.
2000          */
2001         if (dcp->dc_flags & DS_FIND_SNAPSHOTS) {
2002                 dsl_dataset_t *ds;
2003                 err = dsl_dataset_hold_obj(dp, thisobj, FTAG, &ds);
2004 
2005                 if (err == 0) {
2006                         uint64_t snapobj;
2007 
2008                         snapobj = dsl_dataset_phys(ds)->ds_snapnames_zapobj;
2009                         dsl_dataset_rele(ds, FTAG);
2010 
2011                         for (zap_cursor_init(&zc, dp->dp_meta_objset, snapobj);
2012                             zap_cursor_retrieve(&zc, attr) == 0;
2013                             (void) zap_cursor_advance(&zc)) {
2014                                 ASSERT3U(attr->za_integer_length, ==,
2015                                     sizeof (uint64_t));
2016                                 ASSERT3U(attr->za_num_integers, ==, 1);
2017 
2018                                 err = dsl_dataset_hold_obj(dp,
2019                                     attr->za_first_integer, FTAG, &ds);
2020                                 if (err != 0)
2021                                         break;
2022                                 err = dcp->dc_func(dp, ds, dcp->dc_arg);
2023                                 dsl_dataset_rele(ds, FTAG);
2024                                 if (err != 0)
2025                                         break;
2026                         }
2027                         zap_cursor_fini(&zc);
2028                 }
2029         }
2030 
2031         kmem_free(attr, sizeof (zap_attribute_t));
2032 
2033         if (err != 0) {
2034                 dsl_dir_rele(dd, FTAG);
2035                 goto out;
2036         }
2037 
2038         /*
2039          * Apply to self.
2040          */
2041         err = dsl_dataset_hold_obj(dp, thisobj, FTAG, &ds);
2042 
2043         /*
2044          * Note: we hold the dir while calling dsl_dataset_hold_obj() so
2045          * that the dir will remain cached, and we won't have to re-instantiate
2046          * it (which could be expensive due to finding its name via
2047          * zap_value_search()).
2048          */
2049         dsl_dir_rele(dd, FTAG);
2050         if (err != 0)
2051                 goto out;
2052         err = dcp->dc_func(dp, ds, dcp->dc_arg);
2053         dsl_dataset_rele(ds, FTAG);
2054 
2055 out:
2056         if (err != 0) {
2057                 mutex_enter(dcp->dc_error_lock);
2058                 /* only keep first error */
2059                 if (*dcp->dc_error == 0)
2060                         *dcp->dc_error = err;
2061                 mutex_exit(dcp->dc_error_lock);
2062         }
2063 
2064         if (dcp->dc_ddname != NULL)
2065                 spa_strfree(dcp->dc_ddname);
2066         kmem_free(dcp, sizeof (*dcp));
2067 }
2068 
2069 static void
2070 dmu_objset_find_dp_cb(void *arg)
2071 {
2072         dmu_objset_find_ctx_t *dcp = arg;
2073         dsl_pool_t *dp = dcp->dc_dp;
2074 
2075         /*
2076          * We need to get a pool_config_lock here, as there are several
2077          * asssert(pool_config_held) down the stack. Getting a lock via
2078          * dsl_pool_config_enter is risky, as it might be stalled by a
2079          * pending writer. This would deadlock, as the write lock can
2080          * only be granted when our parent thread gives up the lock.
2081          * The _prio interface gives us priority over a pending writer.
2082          */
2083         dsl_pool_config_enter_prio(dp, FTAG);
2084 
2085         dmu_objset_find_dp_impl(dcp);
2086 
2087         dsl_pool_config_exit(dp, FTAG);
2088 }
2089 
2090 /*
2091  * Find objsets under and including ddobj, call func(ds) on each.
2092  * The order for the enumeration is completely undefined.
2093  * func is called with dsl_pool_config held.
2094  */
2095 int
2096 dmu_objset_find_dp(dsl_pool_t *dp, uint64_t ddobj,
2097     int func(dsl_pool_t *, dsl_dataset_t *, void *), void *arg, int flags)
2098 {
2099         int error = 0;
2100         taskq_t *tq = NULL;
2101         int ntasks;
2102         dmu_objset_find_ctx_t *dcp;
2103         kmutex_t err_lock;
2104 
2105         mutex_init(&err_lock, NULL, MUTEX_DEFAULT, NULL);
2106         dcp = kmem_alloc(sizeof (*dcp), KM_SLEEP);
2107         dcp->dc_tq = NULL;
2108         dcp->dc_dp = dp;
2109         dcp->dc_ddobj = ddobj;
2110         dcp->dc_ddname = NULL;
2111         dcp->dc_func = func;
2112         dcp->dc_arg = arg;
2113         dcp->dc_flags = flags;
2114         dcp->dc_error_lock = &err_lock;
2115         dcp->dc_error = &error;
2116 
2117         if ((flags & DS_FIND_SERIALIZE) || dsl_pool_config_held_writer(dp)) {
2118                 /*
2119                  * In case a write lock is held we can't make use of
2120                  * parallelism, as down the stack of the worker threads
2121                  * the lock is asserted via dsl_pool_config_held.
2122                  * In case of a read lock this is solved by getting a read
2123                  * lock in each worker thread, which isn't possible in case
2124                  * of a writer lock. So we fall back to the synchronous path
2125                  * here.
2126                  * In the future it might be possible to get some magic into
2127                  * dsl_pool_config_held in a way that it returns true for
2128                  * the worker threads so that a single lock held from this
2129                  * thread suffices. For now, stay single threaded.
2130                  */
2131                 dmu_objset_find_dp_impl(dcp);
2132                 mutex_destroy(&err_lock);
2133 
2134                 return (error);
2135         }
2136 
2137         ntasks = dmu_find_threads;
2138         if (ntasks == 0)
2139                 ntasks = vdev_count_leaves(dp->dp_spa) * 4;
2140         tq = taskq_create("dmu_objset_find", ntasks, minclsyspri, ntasks,
2141             INT_MAX, 0);
2142         if (tq == NULL) {
2143                 kmem_free(dcp, sizeof (*dcp));
2144                 mutex_destroy(&err_lock);
2145 
2146                 return (SET_ERROR(ENOMEM));
2147         }
2148         dcp->dc_tq = tq;
2149 
2150         /* dcp will be freed by task */
2151         (void) taskq_dispatch(tq, dmu_objset_find_dp_cb, dcp, TQ_SLEEP);
2152 
2153         /*
2154          * PORTING: this code relies on the property of taskq_wait to wait
2155          * until no more tasks are queued and no more tasks are active. As
2156          * we always queue new tasks from within other tasks, task_wait
2157          * reliably waits for the full recursion to finish, even though we
2158          * enqueue new tasks after taskq_wait has been called.
2159          * On platforms other than illumos, taskq_wait may not have this
2160          * property.
2161          */
2162         taskq_wait(tq);
2163         taskq_destroy(tq);
2164         mutex_destroy(&err_lock);
2165 
2166         return (error);
2167 }
2168 
2169 /*
2170  * Find all objsets under name, and for each, call 'func(child_name, arg)'.
2171  * The dp_config_rwlock must not be held when this is called, and it
2172  * will not be held when the callback is called.
2173  * Therefore this function should only be used when the pool is not changing
2174  * (e.g. in syncing context), or the callback can deal with the possible races.
2175  */
2176 static int
2177 dmu_objset_find_impl(spa_t *spa, const char *name,
2178     int func(const char *, void *), void *arg, int flags)
2179 {
2180         dsl_dir_t *dd;
2181         dsl_pool_t *dp = spa_get_dsl(spa);
2182         dsl_dataset_t *ds;
2183         zap_cursor_t zc;
2184         zap_attribute_t *attr;
2185         char *child;
2186         uint64_t thisobj;
2187         int err;
2188 
2189         dsl_pool_config_enter(dp, FTAG);
2190 
2191         err = dsl_dir_hold(dp, name, FTAG, &dd, NULL);
2192         if (err != 0) {
2193                 dsl_pool_config_exit(dp, FTAG);
2194                 return (err);
2195         }
2196 
2197         /* Don't visit hidden ($MOS & $ORIGIN) objsets. */
2198         if (dd->dd_myname[0] == '$') {
2199                 dsl_dir_rele(dd, FTAG);
2200                 dsl_pool_config_exit(dp, FTAG);
2201                 return (0);
2202         }
2203 
2204         thisobj = dsl_dir_phys(dd)->dd_head_dataset_obj;
2205         attr = kmem_alloc(sizeof (zap_attribute_t), KM_SLEEP);
2206 
2207         /*
2208          * Iterate over all children.
2209          */
2210         if (flags & DS_FIND_CHILDREN) {
2211                 for (zap_cursor_init(&zc, dp->dp_meta_objset,
2212                     dsl_dir_phys(dd)->dd_child_dir_zapobj);
2213                     zap_cursor_retrieve(&zc, attr) == 0;
2214                     (void) zap_cursor_advance(&zc)) {
2215                         ASSERT3U(attr->za_integer_length, ==,
2216                             sizeof (uint64_t));
2217                         ASSERT3U(attr->za_num_integers, ==, 1);
2218 
2219                         child = kmem_asprintf("%s/%s", name, attr->za_name);
2220                         dsl_pool_config_exit(dp, FTAG);
2221                         err = dmu_objset_find_impl(spa, child,
2222                             func, arg, flags);
2223                         dsl_pool_config_enter(dp, FTAG);
2224                         strfree(child);
2225                         if (err != 0)
2226                                 break;
2227                 }
2228                 zap_cursor_fini(&zc);
2229 
2230                 if (err != 0) {
2231                         dsl_dir_rele(dd, FTAG);
2232                         dsl_pool_config_exit(dp, FTAG);
2233                         kmem_free(attr, sizeof (zap_attribute_t));
2234                         return (err);
2235                 }
2236         }
2237 
2238         /*
2239          * Iterate over all snapshots.
2240          */
2241         if (flags & DS_FIND_SNAPSHOTS) {
2242                 err = dsl_dataset_hold_obj(dp, thisobj, FTAG, &ds);
2243 
2244                 if (err == 0) {
2245                         uint64_t snapobj;
2246 
2247                         snapobj = dsl_dataset_phys(ds)->ds_snapnames_zapobj;
2248                         dsl_dataset_rele(ds, FTAG);
2249 
2250                         for (zap_cursor_init(&zc, dp->dp_meta_objset, snapobj);
2251                             zap_cursor_retrieve(&zc, attr) == 0;
2252                             (void) zap_cursor_advance(&zc)) {
2253                                 ASSERT3U(attr->za_integer_length, ==,
2254                                     sizeof (uint64_t));
2255                                 ASSERT3U(attr->za_num_integers, ==, 1);
2256 
2257                                 child = kmem_asprintf("%s@%s",
2258                                     name, attr->za_name);
2259                                 dsl_pool_config_exit(dp, FTAG);
2260                                 err = func(child, arg);
2261                                 dsl_pool_config_enter(dp, FTAG);
2262                                 strfree(child);
2263                                 if (err != 0)
2264                                         break;
2265                         }
2266                         zap_cursor_fini(&zc);
2267                 }
2268         }
2269 
2270         dsl_dir_rele(dd, FTAG);
2271         kmem_free(attr, sizeof (zap_attribute_t));
2272         dsl_pool_config_exit(dp, FTAG);
2273 
2274         if (err != 0)
2275                 return (err);
2276 
2277         /* Apply to self. */
2278         return (func(name, arg));
2279 }
2280 
2281 /*
2282  * See comment above dmu_objset_find_impl().
2283  */
2284 int
2285 dmu_objset_find(char *name, int func(const char *, void *), void *arg,
2286     int flags)
2287 {
2288         spa_t *spa;
2289         int error;
2290 
2291         error = spa_open(name, &spa, FTAG);
2292         if (error != 0)
2293                 return (error);
2294         error = dmu_objset_find_impl(spa, name, func, arg, flags);
2295         spa_close(spa, FTAG);
2296         return (error);
2297 }
2298 
2299 void
2300 dmu_objset_set_user(objset_t *os, void *user_ptr)
2301 {
2302         ASSERT(MUTEX_HELD(&os->os_user_ptr_lock));
2303         os->os_user_ptr = user_ptr;
2304 }
2305 
2306 void *
2307 dmu_objset_get_user(objset_t *os)
2308 {
2309         ASSERT(MUTEX_HELD(&os->os_user_ptr_lock));
2310         return (os->os_user_ptr);
2311 }
2312 
2313 /*
2314  * Determine name of filesystem, given name of snapshot.
2315  * buf must be at least ZFS_MAX_DATASET_NAME_LEN bytes
2316  */
2317 int
2318 dmu_fsname(const char *snapname, char *buf)
2319 {
2320         char *atp = strchr(snapname, '@');
2321         if (atp == NULL)
2322                 return (SET_ERROR(EINVAL));
2323         if (atp - snapname >= ZFS_MAX_DATASET_NAME_LEN)
2324                 return (SET_ERROR(ENAMETOOLONG));
2325         (void) strlcpy(buf, snapname, atp - snapname + 1);
2326         return (0);
2327 }
2328 
2329 /*
2330  * Call when we think we're going to write/free space in open context to track
2331  * the amount of dirty data in the open txg, which is also the amount
2332  * of memory that can not be evicted until this txg syncs.
2333  */
2334 void
2335 dmu_objset_willuse_space(objset_t *os, int64_t space, dmu_tx_t *tx)
2336 {
2337         dsl_dataset_t *ds = os->os_dsl_dataset;
2338         int64_t aspace = spa_get_worst_case_asize(os->os_spa, space);
2339 
2340         if (ds != NULL) {
2341                 dsl_dir_willuse_space(ds->ds_dir, aspace, tx);
2342                 dsl_pool_dirty_space(dmu_tx_pool(tx), space, tx);
2343         }
2344 }