module/zfs/dsl_dir.c

   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 2009 Sun Microsystems, Inc.  All rights reserved.
  23  * Use is subject to license terms.
  24  */
  25
  26 #include <sys/dmu.h>
  27 #include <sys/dmu_objset.h>
  28 #include <sys/dmu_tx.h>
  29 #include <sys/dsl_dataset.h>
  30 #include <sys/dsl_dir.h>
  31 #include <sys/dsl_prop.h>
  32 #include <sys/dsl_synctask.h>
  33 #include <sys/dsl_deleg.h>
  34 #include <sys/spa.h>
  35 #include <sys/zap.h>
  36 #include <sys/zio.h>
  37 #include <sys/arc.h>
  38 #include <sys/sunddi.h>
  39 #include "zfs_namecheck.h"
  40
  41 static uint64_t dsl_dir_space_towrite(dsl_dir_t *dd);
  42 static void dsl_dir_set_reservation_sync(void *arg1, void *arg2,
  43     cred_t *cr, dmu_tx_t *tx);
  44
  45
  46 /* ARGSUSED */
  47 static void
  48 dsl_dir_evict(dmu_buf_t *db, void *arg)
  49 {
  50         dsl_dir_t *dd = arg;
  51         dsl_pool_t *dp = dd->dd_pool;
  52         int t;
  53
  54         for (t = 0; t < TXG_SIZE; t++) {
  55                 ASSERT(!txg_list_member(&dp->dp_dirty_dirs, dd, t));
  56                 ASSERT(dd->dd_tempreserved[t] == 0);
  57                 ASSERT(dd->dd_space_towrite[t] == 0);
  58         }
  59
  60         if (dd->dd_parent)
  61                 dsl_dir_close(dd->dd_parent, dd);
  62
  63         spa_close(dd->dd_pool->dp_spa, dd);
  64
  65         /*
  66          * The props callback list should be empty since they hold the
  67          * dir open.
  68          */
  69         list_destroy(&dd->dd_prop_cbs);
  70         mutex_destroy(&dd->dd_lock);
  71         kmem_free(dd, sizeof (dsl_dir_t));
  72 }
  73
  74 int
  75 dsl_dir_open_obj(dsl_pool_t *dp, uint64_t ddobj,
  76     const char *tail, void *tag, dsl_dir_t **ddp)
  77 {
  78         dmu_buf_t *dbuf;
  79         dsl_dir_t *dd;
  80         int err;
  81
  82         ASSERT(RW_LOCK_HELD(&dp->dp_config_rwlock) ||
  83             dsl_pool_sync_context(dp));
  84
  85         err = dmu_bonus_hold(dp->dp_meta_objset, ddobj, tag, &dbuf);
  86         if (err)
  87                 return (err);
  88         dd = dmu_buf_get_user(dbuf);
  89 #ifdef ZFS_DEBUG
  90         {
  91                 dmu_object_info_t doi;
  92                 dmu_object_info_from_db(dbuf, &doi);
  93                 ASSERT3U(doi.doi_type, ==, DMU_OT_DSL_DIR);
  94                 ASSERT3U(doi.doi_bonus_size, >=, sizeof (dsl_dir_phys_t));
  95         }
  96 #endif
  97         if (dd == NULL) {
  98                 dsl_dir_t *winner;
  99                 int err;
 100
 101                 dd = kmem_zalloc(sizeof (dsl_dir_t), KM_SLEEP);
 102                 dd->dd_object = ddobj;
 103                 dd->dd_dbuf = dbuf;
 104                 dd->dd_pool = dp;
 105                 dd->dd_phys = dbuf->db_data;
 106                 mutex_init(&dd->dd_lock, NULL, MUTEX_DEFAULT, NULL);
 107
 108                 list_create(&dd->dd_prop_cbs, sizeof (dsl_prop_cb_record_t),
 109                     offsetof(dsl_prop_cb_record_t, cbr_node));
 110
 111                 if (dd->dd_phys->dd_parent_obj) {
 112                         err = dsl_dir_open_obj(dp, dd->dd_phys->dd_parent_obj,
 113                             NULL, dd, &dd->dd_parent);
 114                         if (err)
 115                                 goto errout;
 116                         if (tail) {
 117 #ifdef ZFS_DEBUG
 118                                 uint64_t foundobj;
 119
 120                                 err = zap_lookup(dp->dp_meta_objset,
 121                                     dd->dd_parent->dd_phys->dd_child_dir_zapobj,
 122                                     tail, sizeof (foundobj), 1, &foundobj);
 123                                 ASSERT(err || foundobj == ddobj);
 124 #endif
 125                                 (void) strcpy(dd->dd_myname, tail);
 126                         } else {
 127                                 err = zap_value_search(dp->dp_meta_objset,
 128                                     dd->dd_parent->dd_phys->dd_child_dir_zapobj,
 129                                     ddobj, 0, dd->dd_myname);
 130                         }
 131                         if (err)
 132                                 goto errout;
 133                 } else {
 134                         (void) strcpy(dd->dd_myname, spa_name(dp->dp_spa));
 135                 }
 136
 137                 winner = dmu_buf_set_user_ie(dbuf, dd, &dd->dd_phys,
 138                     dsl_dir_evict);
 139                 if (winner) {
 140                         if (dd->dd_parent)
 141                                 dsl_dir_close(dd->dd_parent, dd);
 142                         mutex_destroy(&dd->dd_lock);
 143                         kmem_free(dd, sizeof (dsl_dir_t));
 144                         dd = winner;
 145                 } else {
 146                         spa_open_ref(dp->dp_spa, dd);
 147                 }
 148         }
 149
 150         /*
 151          * The dsl_dir_t has both open-to-close and instantiate-to-evict
 152          * holds on the spa.  We need the open-to-close holds because
 153          * otherwise the spa_refcnt wouldn't change when we open a
 154          * dir which the spa also has open, so we could incorrectly
 155          * think it was OK to unload/export/destroy the pool.  We need
 156          * the instantiate-to-evict hold because the dsl_dir_t has a
 157          * pointer to the dd_pool, which has a pointer to the spa_t.
 158          */
 159         spa_open_ref(dp->dp_spa, tag);
 160         ASSERT3P(dd->dd_pool, ==, dp);
 161         ASSERT3U(dd->dd_object, ==, ddobj);
 162         ASSERT3P(dd->dd_dbuf, ==, dbuf);
 163         *ddp = dd;
 164         return (0);
 165
 166 errout:
 167         if (dd->dd_parent)
 168                 dsl_dir_close(dd->dd_parent, dd);
 169         mutex_destroy(&dd->dd_lock);
 170         kmem_free(dd, sizeof (dsl_dir_t));
 171         dmu_buf_rele(dbuf, tag);
 172         return (err);
 173
 174 }
 175
 176 void
 177 dsl_dir_close(dsl_dir_t *dd, void *tag)
 178 {
 179         dprintf_dd(dd, "%s\n", "");
 180         spa_close(dd->dd_pool->dp_spa, tag);
 181         dmu_buf_rele(dd->dd_dbuf, tag);
 182 }
 183
 184 /* buf must be long enough (MAXNAMELEN + strlen(MOS_DIR_NAME) + 1 should do) */
 185 void
 186 dsl_dir_name(dsl_dir_t *dd, char *buf)
 187 {
 188         if (dd->dd_parent) {
 189                 dsl_dir_name(dd->dd_parent, buf);
 190                 (void) strcat(buf, "/");
 191         } else {
 192                 buf[0] = '\0';
 193         }
 194         if (!MUTEX_HELD(&dd->dd_lock)) {
 195                 /*
 196                  * recursive mutex so that we can use
 197                  * dprintf_dd() with dd_lock held
 198                  */
 199                 mutex_enter(&dd->dd_lock);
 200                 (void) strcat(buf, dd->dd_myname);
 201                 mutex_exit(&dd->dd_lock);
 202         } else {
 203                 (void) strcat(buf, dd->dd_myname);
 204         }
 205 }
 206
 207 /* Calculate name legnth, avoiding all the strcat calls of dsl_dir_name */
 208 int
 209 dsl_dir_namelen(dsl_dir_t *dd)
 210 {
 211         int result = 0;
 212
 213         if (dd->dd_parent) {
 214                 /* parent's name + 1 for the "/" */
 215                 result = dsl_dir_namelen(dd->dd_parent) + 1;
 216         }
 217
 218         if (!MUTEX_HELD(&dd->dd_lock)) {
 219                 /* see dsl_dir_name */
 220                 mutex_enter(&dd->dd_lock);
 221                 result += strlen(dd->dd_myname);
 222                 mutex_exit(&dd->dd_lock);
 223         } else {
 224                 result += strlen(dd->dd_myname);
 225         }
 226
 227         return (result);
 228 }
 229
 230 static int
 231 getcomponent(const char *path, char *component, const char **nextp)
 232 {
 233         char *p;
 234         if ((path == NULL) || (path[0] == '\0'))
 235                 return (ENOENT);
 236         /* This would be a good place to reserve some namespace... */
 237         p = strpbrk(path, "/@");
 238         if (p && (p[1] == '/' || p[1] == '@')) {
 239                 /* two separators in a row */
 240                 return (EINVAL);
 241         }
 242         if (p == NULL || p == path) {
 243                 /*
 244                  * if the first thing is an @ or /, it had better be an
 245                  * @ and it had better not have any more ats or slashes,
 246                  * and it had better have something after the @.
 247                  */
 248                 if (p != NULL &&
 249                     (p[0] != '@' || strpbrk(path+1, "/@") || p[1] == '\0'))
 250                         return (EINVAL);
 251                 if (strlen(path) >= MAXNAMELEN)
 252                         return (ENAMETOOLONG);
 253                 (void) strcpy(component, path);
 254                 p = NULL;
 255         } else if (p[0] == '/') {
 256                 if (p-path >= MAXNAMELEN)
 257                         return (ENAMETOOLONG);
 258                 (void) strncpy(component, path, p - path);
 259                 component[p-path] = '\0';
 260                 p++;
 261         } else if (p[0] == '@') {
 262                 /*
 263                  * if the next separator is an @, there better not be
 264                  * any more slashes.
 265                  */
 266                 if (strchr(path, '/'))
 267                         return (EINVAL);
 268                 if (p-path >= MAXNAMELEN)
 269                         return (ENAMETOOLONG);
 270                 (void) strncpy(component, path, p - path);
 271                 component[p-path] = '\0';
 272         } else {
 273                 ASSERT(!"invalid p");
 274         }
 275         *nextp = p;
 276         return (0);
 277 }
 278
 279 /*
 280  * same as dsl_open_dir, ignore the first component of name and use the
 281  * spa instead
 282  */
 283 int
 284 dsl_dir_open_spa(spa_t *spa, const char *name, void *tag,
 285     dsl_dir_t **ddp, const char **tailp)
 286 {
 287         char buf[MAXNAMELEN];
 288         const char *next, *nextnext = NULL;
 289         int err;
 290         dsl_dir_t *dd;
 291         dsl_pool_t *dp;
 292         uint64_t ddobj;
 293         int openedspa = FALSE;
 294
 295         dprintf("%s\n", name);
 296
 297         err = getcomponent(name, buf, &next);
 298         if (err)
 299                 return (err);
 300         if (spa == NULL) {
 301                 err = spa_open(buf, &spa, FTAG);
 302                 if (err) {
 303                         dprintf("spa_open(%s) failed\n", buf);
 304                         return (err);
 305                 }
 306                 openedspa = TRUE;
 307
 308                 /* XXX this assertion belongs in spa_open */
 309                 ASSERT(!dsl_pool_sync_context(spa_get_dsl(spa)));
 310         }
 311
 312         dp = spa_get_dsl(spa);
 313
 314         rw_enter(&dp->dp_config_rwlock, RW_READER);
 315         err = dsl_dir_open_obj(dp, dp->dp_root_dir_obj, NULL, tag, &dd);
 316         if (err) {
 317                 rw_exit(&dp->dp_config_rwlock);
 318                 if (openedspa)
 319                         spa_close(spa, FTAG);
 320                 return (err);
 321         }
 322
 323         while (next != NULL) {
 324                 dsl_dir_t *child_ds;
 325                 err = getcomponent(next, buf, &nextnext);
 326                 if (err)
 327                         break;
 328                 ASSERT(next[0] != '\0');
 329                 if (next[0] == '@')
 330                         break;
 331                 dprintf("looking up %s in obj%lld\n",
 332                     buf, dd->dd_phys->dd_child_dir_zapobj);
 333
 334                 err = zap_lookup(dp->dp_meta_objset,
 335                     dd->dd_phys->dd_child_dir_zapobj,
 336                     buf, sizeof (ddobj), 1, &ddobj);
 337                 if (err) {
 338                         if (err == ENOENT)
 339                                 err = 0;
 340                         break;
 341                 }
 342
 343                 err = dsl_dir_open_obj(dp, ddobj, buf, tag, &child_ds);
 344                 if (err)
 345                         break;
 346                 dsl_dir_close(dd, tag);
 347                 dd = child_ds;
 348                 next = nextnext;
 349         }
 350         rw_exit(&dp->dp_config_rwlock);
 351
 352         if (err) {
 353                 dsl_dir_close(dd, tag);
 354                 if (openedspa)
 355                         spa_close(spa, FTAG);
 356                 return (err);
 357         }
 358
 359         /*
 360          * It's an error if there's more than one component left, or
 361          * tailp==NULL and there's any component left.
 362          */
 363         if (next != NULL &&
 364             (tailp == NULL || (nextnext && nextnext[0] != '\0'))) {
 365                 /* bad path name */
 366                 dsl_dir_close(dd, tag);
 367                 dprintf("next=%p (%s) tail=%p\n", next, next?next:"", tailp);
 368                 err = ENOENT;
 369         }
 370         if (tailp)
 371                 *tailp = next;
 372         if (openedspa)
 373                 spa_close(spa, FTAG);
 374         *ddp = dd;
 375         return (err);
 376 }
 377
 378 /*
 379  * Return the dsl_dir_t, and possibly the last component which couldn't
 380  * be found in *tail.  Return NULL if the path is bogus, or if
 381  * tail==NULL and we couldn't parse the whole name.  (*tail)[0] == '@'
 382  * means that the last component is a snapshot.
 383  */
 384 int
 385 dsl_dir_open(const char *name, void *tag, dsl_dir_t **ddp, const char **tailp)
 386 {
 387         return (dsl_dir_open_spa(NULL, name, tag, ddp, tailp));
 388 }
 389
 390 uint64_t
 391 dsl_dir_create_sync(dsl_pool_t *dp, dsl_dir_t *pds, const char *name,
 392     dmu_tx_t *tx)
 393 {
 394         objset_t *mos = dp->dp_meta_objset;
 395         uint64_t ddobj;
 396         dsl_dir_phys_t *dsphys;
 397         dmu_buf_t *dbuf;
 398
 399         ddobj = dmu_object_alloc(mos, DMU_OT_DSL_DIR, 0,
 400             DMU_OT_DSL_DIR, sizeof (dsl_dir_phys_t), tx);
 401         if (pds) {
 402                 VERIFY(0 == zap_add(mos, pds->dd_phys->dd_child_dir_zapobj,
 403                     name, sizeof (uint64_t), 1, &ddobj, tx));
 404         } else {
 405                 /* it's the root dir */
 406                 VERIFY(0 == zap_add(mos, DMU_POOL_DIRECTORY_OBJECT,
 407                     DMU_POOL_ROOT_DATASET, sizeof (uint64_t), 1, &ddobj, tx));
 408         }
 409         VERIFY(0 == dmu_bonus_hold(mos, ddobj, FTAG, &dbuf));
 410         dmu_buf_will_dirty(dbuf, tx);
 411         dsphys = dbuf->db_data;
 412
 413         dsphys->dd_creation_time = gethrestime_sec();
 414         if (pds)
 415                 dsphys->dd_parent_obj = pds->dd_object;
 416         dsphys->dd_props_zapobj = zap_create(mos,
 417             DMU_OT_DSL_PROPS, DMU_OT_NONE, 0, tx);
 418         dsphys->dd_child_dir_zapobj = zap_create(mos,
 419             DMU_OT_DSL_DIR_CHILD_MAP, DMU_OT_NONE, 0, tx);
 420         if (spa_version(dp->dp_spa) >= SPA_VERSION_USED_BREAKDOWN)
 421                 dsphys->dd_flags |= DD_FLAG_USED_BREAKDOWN;
 422         dmu_buf_rele(dbuf, FTAG);
 423
 424         return (ddobj);
 425 }
 426
 427 /* ARGSUSED */
 428 int
 429 dsl_dir_destroy_check(void *arg1, void *arg2, dmu_tx_t *tx)
 430 {
 431         dsl_dir_t *dd = arg1;
 432         dsl_pool_t *dp = dd->dd_pool;
 433         objset_t *mos = dp->dp_meta_objset;
 434         int err;
 435         uint64_t count;
 436
 437         /*
 438          * There should be exactly two holds, both from
 439          * dsl_dataset_destroy: one on the dd directory, and one on its
 440          * head ds.  Otherwise, someone is trying to lookup something
 441          * inside this dir while we want to destroy it.  The
 442          * config_rwlock ensures that nobody else opens it after we
 443          * check.
 444          */
 445         if (dmu_buf_refcount(dd->dd_dbuf) > 2)
 446                 return (EBUSY);
 447
 448         err = zap_count(mos, dd->dd_phys->dd_child_dir_zapobj, &count);
 449         if (err)
 450                 return (err);
 451         if (count != 0)
 452                 return (EEXIST);
 453
 454         return (0);
 455 }
 456
 457 void
 458 dsl_dir_destroy_sync(void *arg1, void *tag, cred_t *cr, dmu_tx_t *tx)
 459 {
 460         dsl_dir_t *dd = arg1;
 461         objset_t *mos = dd->dd_pool->dp_meta_objset;
 462         uint64_t val, obj;
 463         dd_used_t t;
 464
 465         ASSERT(RW_WRITE_HELD(&dd->dd_pool->dp_config_rwlock));
 466         ASSERT(dd->dd_phys->dd_head_dataset_obj == 0);
 467
 468         /* Remove our reservation. */
 469         val = 0;
 470         dsl_dir_set_reservation_sync(dd, &val, cr, tx);
 471         ASSERT3U(dd->dd_phys->dd_used_bytes, ==, 0);
 472         ASSERT3U(dd->dd_phys->dd_reserved, ==, 0);
 473         for (t = 0; t < DD_USED_NUM; t++)
 474                 ASSERT3U(dd->dd_phys->dd_used_breakdown[t], ==, 0);
 475
 476         VERIFY(0 == zap_destroy(mos, dd->dd_phys->dd_child_dir_zapobj, tx));
 477         VERIFY(0 == zap_destroy(mos, dd->dd_phys->dd_props_zapobj, tx));
 478         VERIFY(0 == dsl_deleg_destroy(mos, dd->dd_phys->dd_deleg_zapobj, tx));
 479         VERIFY(0 == zap_remove(mos,
 480             dd->dd_parent->dd_phys->dd_child_dir_zapobj, dd->dd_myname, tx));
 481
 482         obj = dd->dd_object;
 483         dsl_dir_close(dd, tag);
 484         VERIFY(0 == dmu_object_free(mos, obj, tx));
 485 }
 486
 487 boolean_t
 488 dsl_dir_is_clone(dsl_dir_t *dd)
 489 {
 490         return (dd->dd_phys->dd_origin_obj &&
 491             (dd->dd_pool->dp_origin_snap == NULL ||
 492             dd->dd_phys->dd_origin_obj !=
 493             dd->dd_pool->dp_origin_snap->ds_object));
 494 }
 495
 496 void
 497 dsl_dir_stats(dsl_dir_t *dd, nvlist_t *nv)
 498 {
 499         mutex_enter(&dd->dd_lock);
 500         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USED,
 501             dd->dd_phys->dd_used_bytes);
 502         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_QUOTA, dd->dd_phys->dd_quota);
 503         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_RESERVATION,
 504             dd->dd_phys->dd_reserved);
 505         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_COMPRESSRATIO,
 506             dd->dd_phys->dd_compressed_bytes == 0 ? 100 :
 507             (dd->dd_phys->dd_uncompressed_bytes * 100 /
 508             dd->dd_phys->dd_compressed_bytes));
 509         if (dd->dd_phys->dd_flags & DD_FLAG_USED_BREAKDOWN) {
 510                 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDSNAP,
 511                     dd->dd_phys->dd_used_breakdown[DD_USED_SNAP]);
 512                 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDDS,
 513                     dd->dd_phys->dd_used_breakdown[DD_USED_HEAD]);
 514                 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDREFRESERV,
 515                     dd->dd_phys->dd_used_breakdown[DD_USED_REFRSRV]);
 516                 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDCHILD,
 517                     dd->dd_phys->dd_used_breakdown[DD_USED_CHILD] +
 518                     dd->dd_phys->dd_used_breakdown[DD_USED_CHILD_RSRV]);
 519         }
 520         mutex_exit(&dd->dd_lock);
 521
 522         rw_enter(&dd->dd_pool->dp_config_rwlock, RW_READER);
 523         if (dsl_dir_is_clone(dd)) {
 524                 dsl_dataset_t *ds;
 525                 char buf[MAXNAMELEN];
 526
 527                 VERIFY(0 == dsl_dataset_hold_obj(dd->dd_pool,
 528                     dd->dd_phys->dd_origin_obj, FTAG, &ds));
 529                 dsl_dataset_name(ds, buf);
 530                 dsl_dataset_rele(ds, FTAG);
 531                 dsl_prop_nvlist_add_string(nv, ZFS_PROP_ORIGIN, buf);
 532         }
 533         rw_exit(&dd->dd_pool->dp_config_rwlock);
 534 }
 535
 536 void
 537 dsl_dir_dirty(dsl_dir_t *dd, dmu_tx_t *tx)
 538 {
 539         dsl_pool_t *dp = dd->dd_pool;
 540
 541         ASSERT(dd->dd_phys);
 542
 543         if (txg_list_add(&dp->dp_dirty_dirs, dd, tx->tx_txg) == 0) {
 544                 /* up the hold count until we can be written out */
 545                 dmu_buf_add_ref(dd->dd_dbuf, dd);
 546         }
 547 }
 548
 549 static int64_t
 550 parent_delta(dsl_dir_t *dd, uint64_t used, int64_t delta)
 551 {
 552         uint64_t old_accounted = MAX(used, dd->dd_phys->dd_reserved);
 553         uint64_t new_accounted = MAX(used + delta, dd->dd_phys->dd_reserved);
 554         return (new_accounted - old_accounted);
 555 }
 556
 557 void
 558 dsl_dir_sync(dsl_dir_t *dd, dmu_tx_t *tx)
 559 {
 560         ASSERT(dmu_tx_is_syncing(tx));
 561
 562         dmu_buf_will_dirty(dd->dd_dbuf, tx);
 563
 564         mutex_enter(&dd->dd_lock);
 565         ASSERT3U(dd->dd_tempreserved[tx->tx_txg&TXG_MASK], ==, 0);
 566         dprintf_dd(dd, "txg=%llu towrite=%lluK\n", tx->tx_txg,
 567             dd->dd_space_towrite[tx->tx_txg&TXG_MASK] / 1024);
 568         dd->dd_space_towrite[tx->tx_txg&TXG_MASK] = 0;
 569         mutex_exit(&dd->dd_lock);
 570
 571         /* release the hold from dsl_dir_dirty */
 572         dmu_buf_rele(dd->dd_dbuf, dd);
 573 }
 574
 575 static uint64_t
 576 dsl_dir_space_towrite(dsl_dir_t *dd)
 577 {
 578         uint64_t space = 0;
 579         int i;
 580
 581         ASSERT(MUTEX_HELD(&dd->dd_lock));
 582
 583         for (i = 0; i < TXG_SIZE; i++) {
 584                 space += dd->dd_space_towrite[i&TXG_MASK];
 585                 ASSERT3U(dd->dd_space_towrite[i&TXG_MASK], >=, 0);
 586         }
 587         return (space);
 588 }
 589
 590 /*
 591  * How much space would dd have available if ancestor had delta applied
 592  * to it?  If ondiskonly is set, we're only interested in what's
 593  * on-disk, not estimated pending changes.
 594  */
 595 uint64_t
 596 dsl_dir_space_available(dsl_dir_t *dd,
 597     dsl_dir_t *ancestor, int64_t delta, int ondiskonly)
 598 {
 599         uint64_t parentspace, myspace, quota, used;
 600
 601         /*
 602          * If there are no restrictions otherwise, assume we have
 603          * unlimited space available.
 604          */
 605         quota = UINT64_MAX;
 606         parentspace = UINT64_MAX;
 607
 608         if (dd->dd_parent != NULL) {
 609                 parentspace = dsl_dir_space_available(dd->dd_parent,
 610                     ancestor, delta, ondiskonly);
 611         }
 612
 613         mutex_enter(&dd->dd_lock);
 614         if (dd->dd_phys->dd_quota != 0)
 615                 quota = dd->dd_phys->dd_quota;
 616         used = dd->dd_phys->dd_used_bytes;
 617         if (!ondiskonly)
 618                 used += dsl_dir_space_towrite(dd);
 619
 620         if (dd->dd_parent == NULL) {
 621                 uint64_t poolsize = dsl_pool_adjustedsize(dd->dd_pool, FALSE);
 622                 quota = MIN(quota, poolsize);
 623         }
 624
 625         if (dd->dd_phys->dd_reserved > used && parentspace != UINT64_MAX) {
 626                 /*
 627                  * We have some space reserved, in addition to what our
 628                  * parent gave us.
 629                  */
 630                 parentspace += dd->dd_phys->dd_reserved - used;
 631         }
 632
 633         if (dd == ancestor) {
 634                 ASSERT(delta <= 0);
 635                 ASSERT(used >= -delta);
 636                 used += delta;
 637                 if (parentspace != UINT64_MAX)
 638                         parentspace -= delta;
 639         }
 640
 641         if (used > quota) {
 642                 /* over quota */
 643                 myspace = 0;
 644
 645                 /*
 646                  * While it's OK to be a little over quota, if
 647                  * we think we are using more space than there
 648                  * is in the pool (which is already 1.6% more than
 649                  * dsl_pool_adjustedsize()), something is very
 650                  * wrong.
 651                  */
 652                 ASSERT3U(used, <=, spa_get_space(dd->dd_pool->dp_spa));
 653         } else {
 654                 /*
 655                  * the lesser of the space provided by our parent and
 656                  * the space left in our quota
 657                  */
 658                 myspace = MIN(parentspace, quota - used);
 659         }
 660
 661         mutex_exit(&dd->dd_lock);
 662
 663         return (myspace);
 664 }
 665
 666 struct tempreserve {
 667         list_node_t tr_node;
 668         dsl_pool_t *tr_dp;
 669         dsl_dir_t *tr_ds;
 670         uint64_t tr_size;
 671 };
 672
 673 static int
 674 dsl_dir_tempreserve_impl(dsl_dir_t *dd, uint64_t asize, boolean_t netfree,
 675     boolean_t ignorequota, boolean_t checkrefquota, list_t *tr_list,
 676     dmu_tx_t *tx, boolean_t first)
 677 {
 678         uint64_t txg = tx->tx_txg;
 679         uint64_t est_inflight, used_on_disk, quota, parent_rsrv;
 680         struct tempreserve *tr;
 681         int enospc = EDQUOT;
 682         int txgidx = txg & TXG_MASK;
 683         int i;
 684         uint64_t ref_rsrv = 0;
 685
 686         ASSERT3U(txg, !=, 0);
 687         ASSERT3S(asize, >, 0);
 688
 689         mutex_enter(&dd->dd_lock);
 690
 691         /*
 692          * Check against the dsl_dir's quota.  We don't add in the delta
 693          * when checking for over-quota because they get one free hit.
 694          */
 695         est_inflight = dsl_dir_space_towrite(dd);
 696         for (i = 0; i < TXG_SIZE; i++)
 697                 est_inflight += dd->dd_tempreserved[i];
 698         used_on_disk = dd->dd_phys->dd_used_bytes;
 699
 700         /*
 701          * On the first iteration, fetch the dataset's used-on-disk and
 702          * refreservation values. Also, if checkrefquota is set, test if
 703          * allocating this space would exceed the dataset's refquota.
 704          */
 705         if (first && tx->tx_objset) {
 706                 int error;
 707                 dsl_dataset_t *ds = tx->tx_objset->os->os_dsl_dataset;
 708
 709                 error = dsl_dataset_check_quota(ds, checkrefquota,
 710                     asize, est_inflight, &used_on_disk, &ref_rsrv);
 711                 if (error) {
 712                         mutex_exit(&dd->dd_lock);
 713                         return (error);
 714                 }
 715         }
 716
 717         /*
 718          * If this transaction will result in a net free of space,
 719          * we want to let it through.
 720          */
 721         if (ignorequota || netfree || dd->dd_phys->dd_quota == 0)
 722                 quota = UINT64_MAX;
 723         else
 724                 quota = dd->dd_phys->dd_quota;
 725
 726         /*
 727          * Adjust the quota against the actual pool size at the root.
 728          * To ensure that it's possible to remove files from a full
 729          * pool without inducing transient overcommits, we throttle
 730          * netfree transactions against a quota that is slightly larger,
 731          * but still within the pool's allocation slop.  In cases where
 732          * we're very close to full, this will allow a steady trickle of
 733          * removes to get through.
 734          */
 735         if (dd->dd_parent == NULL) {
 736                 uint64_t poolsize = dsl_pool_adjustedsize(dd->dd_pool, netfree);
 737                 if (poolsize < quota) {
 738                         quota = poolsize;
 739                         enospc = ENOSPC;
 740                 }
 741         }
 742
 743         /*
 744          * If they are requesting more space, and our current estimate
 745          * is over quota, they get to try again unless the actual
 746          * on-disk is over quota and there are no pending changes (which
 747          * may free up space for us).
 748          */
 749         if (used_on_disk + est_inflight > quota) {
 750                 if (est_inflight > 0 || used_on_disk < quota)
 751                         enospc = ERESTART;
 752                 dprintf_dd(dd, "failing: used=%lluK inflight = %lluK "
 753                     "quota=%lluK tr=%lluK err=%d\n",
 754                     used_on_disk>>10, est_inflight>>10,
 755                     quota>>10, asize>>10, enospc);
 756                 mutex_exit(&dd->dd_lock);
 757                 return (enospc);
 758         }
 759
 760         /* We need to up our estimated delta before dropping dd_lock */
 761         dd->dd_tempreserved[txgidx] += asize;
 762
 763         parent_rsrv = parent_delta(dd, used_on_disk + est_inflight,
 764             asize - ref_rsrv);
 765         mutex_exit(&dd->dd_lock);
 766
 767         tr = kmem_zalloc(sizeof (struct tempreserve), KM_SLEEP);
 768         tr->tr_ds = dd;
 769         tr->tr_size = asize;
 770         list_insert_tail(tr_list, tr);
 771
 772         /* see if it's OK with our parent */
 773         if (dd->dd_parent && parent_rsrv) {
 774                 boolean_t ismos = (dd->dd_phys->dd_head_dataset_obj == 0);
 775
 776                 return (dsl_dir_tempreserve_impl(dd->dd_parent,
 777                     parent_rsrv, netfree, ismos, TRUE, tr_list, tx, FALSE));
 778         } else {
 779                 return (0);
 780         }
 781 }
 782
 783 /*
 784  * Reserve space in this dsl_dir, to be used in this tx's txg.
 785  * After the space has been dirtied (and dsl_dir_willuse_space()
 786  * has been called), the reservation should be canceled, using
 787  * dsl_dir_tempreserve_clear().
 788  */
 789 int
 790 dsl_dir_tempreserve_space(dsl_dir_t *dd, uint64_t lsize, uint64_t asize,
 791     uint64_t fsize, uint64_t usize, void **tr_cookiep, dmu_tx_t *tx)
 792 {
 793         int err;
 794         list_t *tr_list;
 795
 796         if (asize == 0) {
 797                 *tr_cookiep = NULL;
 798                 return (0);
 799         }
 800
 801         tr_list = kmem_alloc(sizeof (list_t), KM_SLEEP);
 802         list_create(tr_list, sizeof (struct tempreserve),
 803             offsetof(struct tempreserve, tr_node));
 804         ASSERT3S(asize, >, 0);
 805         ASSERT3S(fsize, >=, 0);
 806
 807         err = arc_tempreserve_space(lsize, tx->tx_txg);
 808         if (err == 0) {
 809                 struct tempreserve *tr;
 810
 811                 tr = kmem_zalloc(sizeof (struct tempreserve), KM_SLEEP);
 812                 tr->tr_size = lsize;
 813                 list_insert_tail(tr_list, tr);
 814
 815                 err = dsl_pool_tempreserve_space(dd->dd_pool, asize, tx);
 816         } else {
 817                 if (err == EAGAIN) {
 818                         txg_delay(dd->dd_pool, tx->tx_txg, 1);
 819                         err = ERESTART;
 820                 }
 821                 dsl_pool_memory_pressure(dd->dd_pool);
 822         }
 823
 824         if (err == 0) {
 825                 struct tempreserve *tr;
 826
 827                 tr = kmem_zalloc(sizeof (struct tempreserve), KM_SLEEP);
 828                 tr->tr_dp = dd->dd_pool;
 829                 tr->tr_size = asize;
 830                 list_insert_tail(tr_list, tr);
 831
 832                 err = dsl_dir_tempreserve_impl(dd, asize, fsize >= asize,
 833                     FALSE, asize > usize, tr_list, tx, TRUE);
 834         }
 835
 836         if (err)
 837                 dsl_dir_tempreserve_clear(tr_list, tx);
 838         else
 839                 *tr_cookiep = tr_list;
 840
 841         return (err);
 842 }
 843
 844 /*
 845  * Clear a temporary reservation that we previously made with
 846  * dsl_dir_tempreserve_space().
 847  */
 848 void
 849 dsl_dir_tempreserve_clear(void *tr_cookie, dmu_tx_t *tx)
 850 {
 851         int txgidx = tx->tx_txg & TXG_MASK;
 852         list_t *tr_list = tr_cookie;
 853         struct tempreserve *tr;
 854
 855         ASSERT3U(tx->tx_txg, !=, 0);
 856
 857         if (tr_cookie == NULL)
 858                 return;
 859
 860         while (tr = list_head(tr_list)) {
 861                 if (tr->tr_dp) {
 862                         dsl_pool_tempreserve_clear(tr->tr_dp, tr->tr_size, tx);
 863                 } else if (tr->tr_ds) {
 864                         mutex_enter(&tr->tr_ds->dd_lock);
 865                         ASSERT3U(tr->tr_ds->dd_tempreserved[txgidx], >=,
 866                             tr->tr_size);
 867                         tr->tr_ds->dd_tempreserved[txgidx] -= tr->tr_size;
 868                         mutex_exit(&tr->tr_ds->dd_lock);
 869                 } else {
 870                         arc_tempreserve_clear(tr->tr_size);
 871                 }
 872                 list_remove(tr_list, tr);
 873                 kmem_free(tr, sizeof (struct tempreserve));
 874         }
 875
 876         kmem_free(tr_list, sizeof (list_t));
 877 }
 878
 879 static void
 880 dsl_dir_willuse_space_impl(dsl_dir_t *dd, int64_t space, dmu_tx_t *tx)
 881 {
 882         int64_t parent_space;
 883         uint64_t est_used;
 884
 885         mutex_enter(&dd->dd_lock);
 886         if (space > 0)
 887                 dd->dd_space_towrite[tx->tx_txg & TXG_MASK] += space;
 888
 889         est_used = dsl_dir_space_towrite(dd) + dd->dd_phys->dd_used_bytes;
 890         parent_space = parent_delta(dd, est_used, space);
 891         mutex_exit(&dd->dd_lock);
 892
 893         /* Make sure that we clean up dd_space_to* */
 894         dsl_dir_dirty(dd, tx);
 895
 896         /* XXX this is potentially expensive and unnecessary... */
 897         if (parent_space && dd->dd_parent)
 898                 dsl_dir_willuse_space_impl(dd->dd_parent, parent_space, tx);
 899 }
 900
 901 /*
 902  * Call in open context when we think we're going to write/free space,
 903  * eg. when dirtying data.  Be conservative (ie. OK to write less than
 904  * this or free more than this, but don't write more or free less).
 905  */
 906 void
 907 dsl_dir_willuse_space(dsl_dir_t *dd, int64_t space, dmu_tx_t *tx)
 908 {
 909         dsl_pool_willuse_space(dd->dd_pool, space, tx);
 910         dsl_dir_willuse_space_impl(dd, space, tx);
 911 }
 912
 913 /* call from syncing context when we actually write/free space for this dd */
 914 void
 915 dsl_dir_diduse_space(dsl_dir_t *dd, dd_used_t type,
 916     int64_t used, int64_t compressed, int64_t uncompressed, dmu_tx_t *tx)
 917 {
 918         int64_t accounted_delta;
 919         boolean_t needlock = !MUTEX_HELD(&dd->dd_lock);
 920
 921         ASSERT(dmu_tx_is_syncing(tx));
 922         ASSERT(type < DD_USED_NUM);
 923
 924         dsl_dir_dirty(dd, tx);
 925
 926         if (needlock)
 927                 mutex_enter(&dd->dd_lock);
 928         accounted_delta = parent_delta(dd, dd->dd_phys->dd_used_bytes, used);
 929         ASSERT(used >= 0 || dd->dd_phys->dd_used_bytes >= -used);
 930         ASSERT(compressed >= 0 ||
 931             dd->dd_phys->dd_compressed_bytes >= -compressed);
 932         ASSERT(uncompressed >= 0 ||
 933             dd->dd_phys->dd_uncompressed_bytes >= -uncompressed);
 934         dd->dd_phys->dd_used_bytes += used;
 935         dd->dd_phys->dd_uncompressed_bytes += uncompressed;
 936         dd->dd_phys->dd_compressed_bytes += compressed;
 937
 938         if (dd->dd_phys->dd_flags & DD_FLAG_USED_BREAKDOWN) {
 939                 ASSERT(used > 0 ||
 940                     dd->dd_phys->dd_used_breakdown[type] >= -used);
 941                 dd->dd_phys->dd_used_breakdown[type] += used;
 942 #ifdef DEBUG
 943                 dd_used_t t;
 944                 uint64_t u = 0;
 945                 for (t = 0; t < DD_USED_NUM; t++)
 946                         u += dd->dd_phys->dd_used_breakdown[t];
 947                 ASSERT3U(u, ==, dd->dd_phys->dd_used_bytes);
 948 #endif
 949         }
 950         if (needlock)
 951                 mutex_exit(&dd->dd_lock);
 952
 953         if (dd->dd_parent != NULL) {
 954                 dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD,
 955                     accounted_delta, compressed, uncompressed, tx);
 956                 dsl_dir_transfer_space(dd->dd_parent,
 957                     used - accounted_delta,
 958                     DD_USED_CHILD_RSRV, DD_USED_CHILD, tx);
 959         }
 960 }
 961
 962 void
 963 dsl_dir_transfer_space(dsl_dir_t *dd, int64_t delta,
 964     dd_used_t oldtype, dd_used_t newtype, dmu_tx_t *tx)
 965 {
 966         boolean_t needlock = !MUTEX_HELD(&dd->dd_lock);
 967
 968         ASSERT(dmu_tx_is_syncing(tx));
 969         ASSERT(oldtype < DD_USED_NUM);
 970         ASSERT(newtype < DD_USED_NUM);
 971
 972         if (delta == 0 || !(dd->dd_phys->dd_flags & DD_FLAG_USED_BREAKDOWN))
 973                 return;
 974
 975         dsl_dir_dirty(dd, tx);
 976         if (needlock)
 977                 mutex_enter(&dd->dd_lock);
 978         ASSERT(delta > 0 ?
 979             dd->dd_phys->dd_used_breakdown[oldtype] >= delta :
 980             dd->dd_phys->dd_used_breakdown[newtype] >= -delta);
 981         ASSERT(dd->dd_phys->dd_used_bytes >= ABS(delta));
 982         dd->dd_phys->dd_used_breakdown[oldtype] -= delta;
 983         dd->dd_phys->dd_used_breakdown[newtype] += delta;
 984         if (needlock)
 985                 mutex_exit(&dd->dd_lock);
 986 }
 987
 988 static int
 989 dsl_dir_set_quota_check(void *arg1, void *arg2, dmu_tx_t *tx)
 990 {
 991         dsl_dir_t *dd = arg1;
 992         uint64_t *quotap = arg2;
 993         uint64_t new_quota = *quotap;
 994         int err = 0;
 995         uint64_t towrite;
 996
 997         if (new_quota == 0)
 998                 return (0);
 999
1000         mutex_enter(&dd->dd_lock);
1001         /*
1002          * If we are doing the preliminary check in open context, and
1003          * there are pending changes, then don't fail it, since the
1004          * pending changes could under-estimate the amount of space to be
1005          * freed up.
1006          */
1007         towrite = dsl_dir_space_towrite(dd);
1008         if ((dmu_tx_is_syncing(tx) || towrite == 0) &&
1009             (new_quota < dd->dd_phys->dd_reserved ||
1010             new_quota < dd->dd_phys->dd_used_bytes + towrite)) {
1011                 err = ENOSPC;
1012         }
1013         mutex_exit(&dd->dd_lock);
1014         return (err);
1015 }
1016
1017 /* ARGSUSED */
1018 static void
1019 dsl_dir_set_quota_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
1020 {
1021         dsl_dir_t *dd = arg1;
1022         uint64_t *quotap = arg2;
1023         uint64_t new_quota = *quotap;
1024
1025         dmu_buf_will_dirty(dd->dd_dbuf, tx);
1026
1027         mutex_enter(&dd->dd_lock);
1028         dd->dd_phys->dd_quota = new_quota;
1029         mutex_exit(&dd->dd_lock);
1030
1031         spa_history_internal_log(LOG_DS_QUOTA, dd->dd_pool->dp_spa,
1032             tx, cr, "%lld dataset = %llu ",
1033             (longlong_t)new_quota, dd->dd_phys->dd_head_dataset_obj);
1034 }
1035
1036 int
1037 dsl_dir_set_quota(const char *ddname, uint64_t quota)
1038 {
1039         dsl_dir_t *dd;
1040         int err;
1041
1042         err = dsl_dir_open(ddname, FTAG, &dd, NULL);
1043         if (err)
1044                 return (err);
1045
1046         if (quota != dd->dd_phys->dd_quota) {
1047                 /*
1048                  * If someone removes a file, then tries to set the quota, we
1049                  * want to make sure the file freeing takes effect.
1050                  */
1051                 txg_wait_open(dd->dd_pool, 0);
1052
1053                 err = dsl_sync_task_do(dd->dd_pool, dsl_dir_set_quota_check,
1054                     dsl_dir_set_quota_sync, dd, &quota, 0);
1055         }
1056         dsl_dir_close(dd, FTAG);
1057         return (err);
1058 }
1059
1060 int
1061 dsl_dir_set_reservation_check(void *arg1, void *arg2, dmu_tx_t *tx)
1062 {
1063         dsl_dir_t *dd = arg1;
1064         uint64_t *reservationp = arg2;
1065         uint64_t new_reservation = *reservationp;
1066         uint64_t used, avail;
1067
1068         /*
1069          * If we are doing the preliminary check in open context, the
1070          * space estimates may be inaccurate.
1071          */
1072         if (!dmu_tx_is_syncing(tx))
1073                 return (0);
1074
1075         mutex_enter(&dd->dd_lock);
1076         used = dd->dd_phys->dd_used_bytes;
1077         mutex_exit(&dd->dd_lock);
1078
1079         if (dd->dd_parent) {
1080                 avail = dsl_dir_space_available(dd->dd_parent,
1081                     NULL, 0, FALSE);
1082         } else {
1083                 avail = dsl_pool_adjustedsize(dd->dd_pool, B_FALSE) - used;
1084         }
1085
1086         if (MAX(used, new_reservation) > MAX(used, dd->dd_phys->dd_reserved)) {
1087                 uint64_t delta = MAX(used, new_reservation) -
1088                     MAX(used, dd->dd_phys->dd_reserved);
1089
1090                 if (delta > avail)
1091                         return (ENOSPC);
1092                 if (dd->dd_phys->dd_quota > 0 &&
1093                     new_reservation > dd->dd_phys->dd_quota)
1094                         return (ENOSPC);
1095         }
1096
1097         return (0);
1098 }
1099
1100 /* ARGSUSED */
1101 static void
1102 dsl_dir_set_reservation_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
1103 {
1104         dsl_dir_t *dd = arg1;
1105         uint64_t *reservationp = arg2;
1106         uint64_t new_reservation = *reservationp;
1107         uint64_t used;
1108         int64_t delta;
1109
1110         dmu_buf_will_dirty(dd->dd_dbuf, tx);
1111
1112         mutex_enter(&dd->dd_lock);
1113         used = dd->dd_phys->dd_used_bytes;
1114         delta = MAX(used, new_reservation) -
1115             MAX(used, dd->dd_phys->dd_reserved);
1116         dd->dd_phys->dd_reserved = new_reservation;
1117
1118         if (dd->dd_parent != NULL) {
1119                 /* Roll up this additional usage into our ancestors */
1120                 dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD_RSRV,
1121                     delta, 0, 0, tx);
1122         }
1123         mutex_exit(&dd->dd_lock);
1124
1125         spa_history_internal_log(LOG_DS_RESERVATION, dd->dd_pool->dp_spa,
1126             tx, cr, "%lld dataset = %llu",
1127             (longlong_t)new_reservation, dd->dd_phys->dd_head_dataset_obj);
1128 }
1129
1130 int
1131 dsl_dir_set_reservation(const char *ddname, uint64_t reservation)
1132 {
1133         dsl_dir_t *dd;
1134         int err;
1135
1136         err = dsl_dir_open(ddname, FTAG, &dd, NULL);
1137         if (err)
1138                 return (err);
1139         err = dsl_sync_task_do(dd->dd_pool, dsl_dir_set_reservation_check,
1140             dsl_dir_set_reservation_sync, dd, &reservation, 0);
1141         dsl_dir_close(dd, FTAG);
1142         return (err);
1143 }
1144
1145 static dsl_dir_t *
1146 closest_common_ancestor(dsl_dir_t *ds1, dsl_dir_t *ds2)
1147 {
1148         for (; ds1; ds1 = ds1->dd_parent) {
1149                 dsl_dir_t *dd;
1150                 for (dd = ds2; dd; dd = dd->dd_parent) {
1151                         if (ds1 == dd)
1152                                 return (dd);
1153                 }
1154         }
1155         return (NULL);
1156 }
1157
1158 /*
1159  * If delta is applied to dd, how much of that delta would be applied to
1160  * ancestor?  Syncing context only.
1161  */
1162 static int64_t
1163 would_change(dsl_dir_t *dd, int64_t delta, dsl_dir_t *ancestor)
1164 {
1165         if (dd == ancestor)
1166                 return (delta);
1167
1168         mutex_enter(&dd->dd_lock);
1169         delta = parent_delta(dd, dd->dd_phys->dd_used_bytes, delta);
1170         mutex_exit(&dd->dd_lock);
1171         return (would_change(dd->dd_parent, delta, ancestor));
1172 }
1173
1174 struct renamearg {
1175         dsl_dir_t *newparent;
1176         const char *mynewname;
1177 };
1178
1179 /*ARGSUSED*/
1180 static int
1181 dsl_dir_rename_check(void *arg1, void *arg2, dmu_tx_t *tx)
1182 {
1183         dsl_dir_t *dd = arg1;
1184         struct renamearg *ra = arg2;
1185         dsl_pool_t *dp = dd->dd_pool;
1186         objset_t *mos = dp->dp_meta_objset;
1187         int err;
1188         uint64_t val;
1189
1190         /* There should be 2 references: the open and the dirty */
1191         if (dmu_buf_refcount(dd->dd_dbuf) > 2)
1192                 return (EBUSY);
1193
1194         /* check for existing name */
1195         err = zap_lookup(mos, ra->newparent->dd_phys->dd_child_dir_zapobj,
1196             ra->mynewname, 8, 1, &val);
1197         if (err == 0)
1198                 return (EEXIST);
1199         if (err != ENOENT)
1200                 return (err);
1201
1202         if (ra->newparent != dd->dd_parent) {
1203                 /* is there enough space? */
1204                 uint64_t myspace =
1205                     MAX(dd->dd_phys->dd_used_bytes, dd->dd_phys->dd_reserved);
1206
1207                 /* no rename into our descendant */
1208                 if (closest_common_ancestor(dd, ra->newparent) == dd)
1209                         return (EINVAL);
1210
1211                 if (err = dsl_dir_transfer_possible(dd->dd_parent,
1212                     ra->newparent, myspace))
1213                         return (err);
1214         }
1215
1216         return (0);
1217 }
1218
1219 static void
1220 dsl_dir_rename_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
1221 {
1222         dsl_dir_t *dd = arg1;
1223         struct renamearg *ra = arg2;
1224         dsl_pool_t *dp = dd->dd_pool;
1225         objset_t *mos = dp->dp_meta_objset;
1226         int err;
1227
1228         ASSERT(dmu_buf_refcount(dd->dd_dbuf) <= 2);
1229
1230         if (ra->newparent != dd->dd_parent) {
1231                 dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD,
1232                     -dd->dd_phys->dd_used_bytes,
1233                     -dd->dd_phys->dd_compressed_bytes,
1234                     -dd->dd_phys->dd_uncompressed_bytes, tx);
1235                 dsl_dir_diduse_space(ra->newparent, DD_USED_CHILD,
1236                     dd->dd_phys->dd_used_bytes,
1237                     dd->dd_phys->dd_compressed_bytes,
1238                     dd->dd_phys->dd_uncompressed_bytes, tx);
1239
1240                 if (dd->dd_phys->dd_reserved > dd->dd_phys->dd_used_bytes) {
1241                         uint64_t unused_rsrv = dd->dd_phys->dd_reserved -
1242                             dd->dd_phys->dd_used_bytes;
1243
1244                         dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD_RSRV,
1245                             -unused_rsrv, 0, 0, tx);
1246                         dsl_dir_diduse_space(ra->newparent, DD_USED_CHILD_RSRV,
1247                             unused_rsrv, 0, 0, tx);
1248                 }
1249         }
1250
1251         dmu_buf_will_dirty(dd->dd_dbuf, tx);
1252
1253         /* remove from old parent zapobj */
1254         err = zap_remove(mos, dd->dd_parent->dd_phys->dd_child_dir_zapobj,
1255             dd->dd_myname, tx);
1256         ASSERT3U(err, ==, 0);
1257
1258         (void) strcpy(dd->dd_myname, ra->mynewname);
1259         dsl_dir_close(dd->dd_parent, dd);
1260         dd->dd_phys->dd_parent_obj = ra->newparent->dd_object;
1261         VERIFY(0 == dsl_dir_open_obj(dd->dd_pool,
1262             ra->newparent->dd_object, NULL, dd, &dd->dd_parent));
1263
1264         /* add to new parent zapobj */
1265         err = zap_add(mos, ra->newparent->dd_phys->dd_child_dir_zapobj,
1266             dd->dd_myname, 8, 1, &dd->dd_object, tx);
1267         ASSERT3U(err, ==, 0);
1268
1269         spa_history_internal_log(LOG_DS_RENAME, dd->dd_pool->dp_spa,
1270             tx, cr, "dataset = %llu", dd->dd_phys->dd_head_dataset_obj);
1271 }
1272
1273 int
1274 dsl_dir_rename(dsl_dir_t *dd, const char *newname)
1275 {
1276         struct renamearg ra;
1277         int err;
1278
1279         /* new parent should exist */
1280         err = dsl_dir_open(newname, FTAG, &ra.newparent, &ra.mynewname);
1281         if (err)
1282                 return (err);
1283
1284         /* can't rename to different pool */
1285         if (dd->dd_pool != ra.newparent->dd_pool) {
1286                 err = ENXIO;
1287                 goto out;
1288         }
1289
1290         /* new name should not already exist */
1291         if (ra.mynewname == NULL) {
1292                 err = EEXIST;
1293                 goto out;
1294         }
1295
1296         err = dsl_sync_task_do(dd->dd_pool,
1297             dsl_dir_rename_check, dsl_dir_rename_sync, dd, &ra, 3);
1298
1299 out:
1300         dsl_dir_close(ra.newparent, FTAG);
1301         return (err);
1302 }
1303
1304 int
1305 dsl_dir_transfer_possible(dsl_dir_t *sdd, dsl_dir_t *tdd, uint64_t space)
1306 {
1307         dsl_dir_t *ancestor;
1308         int64_t adelta;
1309         uint64_t avail;
1310
1311         ancestor = closest_common_ancestor(sdd, tdd);
1312         adelta = would_change(sdd, -space, ancestor);
1313         avail = dsl_dir_space_available(tdd, ancestor, adelta, FALSE);
1314         if (avail < space)
1315                 return (ENOSPC);
1316
1317         return (0);
1318 }