sys/cddl/contrib/opensolaris/uts/common/fs/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 2008 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 int
 231 dsl_dir_is_private(dsl_dir_t *dd)
 232 {
 233         int rv = FALSE;
 234
 235         if (dd->dd_parent && dsl_dir_is_private(dd->dd_parent))
 236                 rv = TRUE;
 237         if (dataset_name_hidden(dd->dd_myname))
 238                 rv = TRUE;
 239         return (rv);
 240 }
 241
 242
 243 static int
 244 getcomponent(const char *path, char *component, const char **nextp)
 245 {
 246         char *p;
 247         if (path == NULL)
 248                 return (ENOENT);
 249         /* This would be a good place to reserve some namespace... */
 250         p = strpbrk(path, "/@");
 251         if (p && (p[1] == '/' || p[1] == '@')) {
 252                 /* two separators in a row */
 253                 return (EINVAL);
 254         }
 255         if (p == NULL || p == path) {
 256                 /*
 257                  * if the first thing is an @ or /, it had better be an
 258                  * @ and it had better not have any more ats or slashes,
 259                  * and it had better have something after the @.
 260                  */
 261                 if (p != NULL &&
 262                     (p[0] != '@' || strpbrk(path+1, "/@") || p[1] == '\0'))
 263                         return (EINVAL);
 264                 if (strlen(path) >= MAXNAMELEN)
 265                         return (ENAMETOOLONG);
 266                 (void) strcpy(component, path);
 267                 p = NULL;
 268         } else if (p[0] == '/') {
 269                 if (p-path >= MAXNAMELEN)
 270                         return (ENAMETOOLONG);
 271                 (void) strncpy(component, path, p - path);
 272                 component[p-path] = '\0';
 273                 p++;
 274         } else if (p[0] == '@') {
 275                 /*
 276                  * if the next separator is an @, there better not be
 277                  * any more slashes.
 278                  */
 279                 if (strchr(path, '/'))
 280                         return (EINVAL);
 281                 if (p-path >= MAXNAMELEN)
 282                         return (ENAMETOOLONG);
 283                 (void) strncpy(component, path, p - path);
 284                 component[p-path] = '\0';
 285         } else {
 286                 ASSERT(!"invalid p");
 287         }
 288         *nextp = p;
 289         return (0);
 290 }
 291
 292 /*
 293  * same as dsl_open_dir, ignore the first component of name and use the
 294  * spa instead
 295  */
 296 int
 297 dsl_dir_open_spa(spa_t *spa, const char *name, void *tag,
 298     dsl_dir_t **ddp, const char **tailp)
 299 {
 300         char buf[MAXNAMELEN];
 301         const char *next, *nextnext = NULL;
 302         int err;
 303         dsl_dir_t *dd;
 304         dsl_pool_t *dp;
 305         uint64_t ddobj;
 306         int openedspa = FALSE;
 307
 308         dprintf("%s\n", name);
 309
 310         err = getcomponent(name, buf, &next);
 311         if (err)
 312                 return (err);
 313         if (spa == NULL) {
 314                 err = spa_open(buf, &spa, FTAG);
 315                 if (err) {
 316                         dprintf("spa_open(%s) failed\n", buf);
 317                         return (err);
 318                 }
 319                 openedspa = TRUE;
 320
 321                 /* XXX this assertion belongs in spa_open */
 322                 ASSERT(!dsl_pool_sync_context(spa_get_dsl(spa)));
 323         }
 324
 325         dp = spa_get_dsl(spa);
 326
 327         rw_enter(&dp->dp_config_rwlock, RW_READER);
 328         err = dsl_dir_open_obj(dp, dp->dp_root_dir_obj, NULL, tag, &dd);
 329         if (err) {
 330                 rw_exit(&dp->dp_config_rwlock);
 331                 if (openedspa)
 332                         spa_close(spa, FTAG);
 333                 return (err);
 334         }
 335
 336         while (next != NULL) {
 337                 dsl_dir_t *child_ds;
 338                 err = getcomponent(next, buf, &nextnext);
 339                 if (err)
 340                         break;
 341                 ASSERT(next[0] != '\0');
 342                 if (next[0] == '@')
 343                         break;
 344                 dprintf("looking up %s in obj%lld\n",
 345                     buf, dd->dd_phys->dd_child_dir_zapobj);
 346
 347                 err = zap_lookup(dp->dp_meta_objset,
 348                     dd->dd_phys->dd_child_dir_zapobj,
 349                     buf, sizeof (ddobj), 1, &ddobj);
 350                 if (err) {
 351                         if (err == ENOENT)
 352                                 err = 0;
 353                         break;
 354                 }
 355
 356                 err = dsl_dir_open_obj(dp, ddobj, buf, tag, &child_ds);
 357                 if (err)
 358                         break;
 359                 dsl_dir_close(dd, tag);
 360                 dd = child_ds;
 361                 next = nextnext;
 362         }
 363         rw_exit(&dp->dp_config_rwlock);
 364
 365         if (err) {
 366                 dsl_dir_close(dd, tag);
 367                 if (openedspa)
 368                         spa_close(spa, FTAG);
 369                 return (err);
 370         }
 371
 372         /*
 373          * It's an error if there's more than one component left, or
 374          * tailp==NULL and there's any component left.
 375          */
 376         if (next != NULL &&
 377             (tailp == NULL || (nextnext && nextnext[0] != '\0'))) {
 378                 /* bad path name */
 379                 dsl_dir_close(dd, tag);
 380                 dprintf("next=%p (%s) tail=%p\n", next, next?next:"", tailp);
 381                 err = ENOENT;
 382         }
 383         if (tailp)
 384                 *tailp = next;
 385         if (openedspa)
 386                 spa_close(spa, FTAG);
 387         *ddp = dd;
 388         return (err);
 389 }
 390
 391 /*
 392  * Return the dsl_dir_t, and possibly the last component which couldn't
 393  * be found in *tail.  Return NULL if the path is bogus, or if
 394  * tail==NULL and we couldn't parse the whole name.  (*tail)[0] == '@'
 395  * means that the last component is a snapshot.
 396  */
 397 int
 398 dsl_dir_open(const char *name, void *tag, dsl_dir_t **ddp, const char **tailp)
 399 {
 400         return (dsl_dir_open_spa(NULL, name, tag, ddp, tailp));
 401 }
 402
 403 uint64_t
 404 dsl_dir_create_sync(dsl_pool_t *dp, dsl_dir_t *pds, const char *name,
 405     dmu_tx_t *tx)
 406 {
 407         objset_t *mos = dp->dp_meta_objset;
 408         uint64_t ddobj;
 409         dsl_dir_phys_t *dsphys;
 410         dmu_buf_t *dbuf;
 411
 412         ddobj = dmu_object_alloc(mos, DMU_OT_DSL_DIR, 0,
 413             DMU_OT_DSL_DIR, sizeof (dsl_dir_phys_t), tx);
 414         if (pds) {
 415                 VERIFY(0 == zap_add(mos, pds->dd_phys->dd_child_dir_zapobj,
 416                     name, sizeof (uint64_t), 1, &ddobj, tx));
 417         } else {
 418                 /* it's the root dir */
 419                 VERIFY(0 == zap_add(mos, DMU_POOL_DIRECTORY_OBJECT,
 420                     DMU_POOL_ROOT_DATASET, sizeof (uint64_t), 1, &ddobj, tx));
 421         }
 422         VERIFY(0 == dmu_bonus_hold(mos, ddobj, FTAG, &dbuf));
 423         dmu_buf_will_dirty(dbuf, tx);
 424         dsphys = dbuf->db_data;
 425
 426         dsphys->dd_creation_time = gethrestime_sec();
 427         if (pds)
 428                 dsphys->dd_parent_obj = pds->dd_object;
 429         dsphys->dd_props_zapobj = zap_create(mos,
 430             DMU_OT_DSL_PROPS, DMU_OT_NONE, 0, tx);
 431         dsphys->dd_child_dir_zapobj = zap_create(mos,
 432             DMU_OT_DSL_DIR_CHILD_MAP, DMU_OT_NONE, 0, tx);
 433         if (spa_version(dp->dp_spa) >= SPA_VERSION_USED_BREAKDOWN)
 434                 dsphys->dd_flags |= DD_FLAG_USED_BREAKDOWN;
 435         dmu_buf_rele(dbuf, FTAG);
 436
 437         return (ddobj);
 438 }
 439
 440 /* ARGSUSED */
 441 int
 442 dsl_dir_destroy_check(void *arg1, void *arg2, dmu_tx_t *tx)
 443 {
 444         dsl_dir_t *dd = arg1;
 445         dsl_pool_t *dp = dd->dd_pool;
 446         objset_t *mos = dp->dp_meta_objset;
 447         int err;
 448         uint64_t count;
 449
 450         /*
 451          * There should be exactly two holds, both from
 452          * dsl_dataset_destroy: one on the dd directory, and one on its
 453          * head ds.  Otherwise, someone is trying to lookup something
 454          * inside this dir while we want to destroy it.  The
 455          * config_rwlock ensures that nobody else opens it after we
 456          * check.
 457          */
 458         if (dmu_buf_refcount(dd->dd_dbuf) > 2)
 459                 return (EBUSY);
 460
 461         err = zap_count(mos, dd->dd_phys->dd_child_dir_zapobj, &count);
 462         if (err)
 463                 return (err);
 464         if (count != 0)
 465                 return (EEXIST);
 466
 467         return (0);
 468 }
 469
 470 void
 471 dsl_dir_destroy_sync(void *arg1, void *tag, cred_t *cr, dmu_tx_t *tx)
 472 {
 473         dsl_dir_t *dd = arg1;
 474         objset_t *mos = dd->dd_pool->dp_meta_objset;
 475         uint64_t val, obj;
 476         dd_used_t t;
 477
 478         ASSERT(RW_WRITE_HELD(&dd->dd_pool->dp_config_rwlock));
 479         ASSERT(dd->dd_phys->dd_head_dataset_obj == 0);
 480
 481         /* Remove our reservation. */
 482         val = 0;
 483         dsl_dir_set_reservation_sync(dd, &val, cr, tx);
 484         ASSERT3U(dd->dd_phys->dd_used_bytes, ==, 0);
 485         ASSERT3U(dd->dd_phys->dd_reserved, ==, 0);
 486         for (t = 0; t < DD_USED_NUM; t++)
 487                 ASSERT3U(dd->dd_phys->dd_used_breakdown[t], ==, 0);
 488
 489         VERIFY(0 == zap_destroy(mos, dd->dd_phys->dd_child_dir_zapobj, tx));
 490         VERIFY(0 == zap_destroy(mos, dd->dd_phys->dd_props_zapobj, tx));
 491         VERIFY(0 == dsl_deleg_destroy(mos, dd->dd_phys->dd_deleg_zapobj, tx));
 492         VERIFY(0 == zap_remove(mos,
 493             dd->dd_parent->dd_phys->dd_child_dir_zapobj, dd->dd_myname, tx));
 494
 495         obj = dd->dd_object;
 496         dsl_dir_close(dd, tag);
 497         VERIFY(0 == dmu_object_free(mos, obj, tx));
 498 }
 499
 500 boolean_t
 501 dsl_dir_is_clone(dsl_dir_t *dd)
 502 {
 503         return (dd->dd_phys->dd_origin_obj &&
 504             (dd->dd_pool->dp_origin_snap == NULL ||
 505             dd->dd_phys->dd_origin_obj !=
 506             dd->dd_pool->dp_origin_snap->ds_object));
 507 }
 508
 509 void
 510 dsl_dir_stats(dsl_dir_t *dd, nvlist_t *nv)
 511 {
 512         mutex_enter(&dd->dd_lock);
 513         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USED,
 514             dd->dd_phys->dd_used_bytes);
 515         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_QUOTA, dd->dd_phys->dd_quota);
 516         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_RESERVATION,
 517             dd->dd_phys->dd_reserved);
 518         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_COMPRESSRATIO,
 519             dd->dd_phys->dd_compressed_bytes == 0 ? 100 :
 520             (dd->dd_phys->dd_uncompressed_bytes * 100 /
 521             dd->dd_phys->dd_compressed_bytes));
 522         if (dd->dd_phys->dd_flags & DD_FLAG_USED_BREAKDOWN) {
 523                 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDSNAP,
 524                     dd->dd_phys->dd_used_breakdown[DD_USED_SNAP]);
 525                 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDDS,
 526                     dd->dd_phys->dd_used_breakdown[DD_USED_HEAD]);
 527                 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDREFRESERV,
 528                     dd->dd_phys->dd_used_breakdown[DD_USED_REFRSRV]);
 529                 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDCHILD,
 530                     dd->dd_phys->dd_used_breakdown[DD_USED_CHILD] +
 531                     dd->dd_phys->dd_used_breakdown[DD_USED_CHILD_RSRV]);
 532         }
 533         mutex_exit(&dd->dd_lock);
 534
 535         rw_enter(&dd->dd_pool->dp_config_rwlock, RW_READER);
 536         if (dsl_dir_is_clone(dd)) {
 537                 dsl_dataset_t *ds;
 538                 char buf[MAXNAMELEN];
 539
 540                 VERIFY(0 == dsl_dataset_hold_obj(dd->dd_pool,
 541                     dd->dd_phys->dd_origin_obj, FTAG, &ds));
 542                 dsl_dataset_name(ds, buf);
 543                 dsl_dataset_rele(ds, FTAG);
 544                 dsl_prop_nvlist_add_string(nv, ZFS_PROP_ORIGIN, buf);
 545         }
 546         rw_exit(&dd->dd_pool->dp_config_rwlock);
 547 }
 548
 549 void
 550 dsl_dir_dirty(dsl_dir_t *dd, dmu_tx_t *tx)
 551 {
 552         dsl_pool_t *dp = dd->dd_pool;
 553
 554         ASSERT(dd->dd_phys);
 555
 556         if (txg_list_add(&dp->dp_dirty_dirs, dd, tx->tx_txg) == 0) {
 557                 /* up the hold count until we can be written out */
 558                 dmu_buf_add_ref(dd->dd_dbuf, dd);
 559         }
 560 }
 561
 562 static int64_t
 563 parent_delta(dsl_dir_t *dd, uint64_t used, int64_t delta)
 564 {
 565         uint64_t old_accounted = MAX(used, dd->dd_phys->dd_reserved);
 566         uint64_t new_accounted = MAX(used + delta, dd->dd_phys->dd_reserved);
 567         return (new_accounted - old_accounted);
 568 }
 569
 570 void
 571 dsl_dir_sync(dsl_dir_t *dd, dmu_tx_t *tx)
 572 {
 573         ASSERT(dmu_tx_is_syncing(tx));
 574
 575         dmu_buf_will_dirty(dd->dd_dbuf, tx);
 576
 577         mutex_enter(&dd->dd_lock);
 578         ASSERT3U(dd->dd_tempreserved[tx->tx_txg&TXG_MASK], ==, 0);
 579         dprintf_dd(dd, "txg=%llu towrite=%lluK\n", tx->tx_txg,
 580             dd->dd_space_towrite[tx->tx_txg&TXG_MASK] / 1024);
 581         dd->dd_space_towrite[tx->tx_txg&TXG_MASK] = 0;
 582         mutex_exit(&dd->dd_lock);
 583
 584         /* release the hold from dsl_dir_dirty */
 585         dmu_buf_rele(dd->dd_dbuf, dd);
 586 }
 587
 588 static uint64_t
 589 dsl_dir_space_towrite(dsl_dir_t *dd)
 590 {
 591         uint64_t space = 0;
 592         int i;
 593
 594         ASSERT(MUTEX_HELD(&dd->dd_lock));
 595
 596         for (i = 0; i < TXG_SIZE; i++) {
 597                 space += dd->dd_space_towrite[i&TXG_MASK];
 598                 ASSERT3U(dd->dd_space_towrite[i&TXG_MASK], >=, 0);
 599         }
 600         return (space);
 601 }
 602
 603 /*
 604  * How much space would dd have available if ancestor had delta applied
 605  * to it?  If ondiskonly is set, we're only interested in what's
 606  * on-disk, not estimated pending changes.
 607  */
 608 uint64_t
 609 dsl_dir_space_available(dsl_dir_t *dd,
 610     dsl_dir_t *ancestor, int64_t delta, int ondiskonly)
 611 {
 612         uint64_t parentspace, myspace, quota, used;
 613
 614         /*
 615          * If there are no restrictions otherwise, assume we have
 616          * unlimited space available.
 617          */
 618         quota = UINT64_MAX;
 619         parentspace = UINT64_MAX;
 620
 621         if (dd->dd_parent != NULL) {
 622                 parentspace = dsl_dir_space_available(dd->dd_parent,
 623                     ancestor, delta, ondiskonly);
 624         }
 625
 626         mutex_enter(&dd->dd_lock);
 627         if (dd->dd_phys->dd_quota != 0)
 628                 quota = dd->dd_phys->dd_quota;
 629         used = dd->dd_phys->dd_used_bytes;
 630         if (!ondiskonly)
 631                 used += dsl_dir_space_towrite(dd);
 632
 633         if (dd->dd_parent == NULL) {
 634                 uint64_t poolsize = dsl_pool_adjustedsize(dd->dd_pool, FALSE);
 635                 quota = MIN(quota, poolsize);
 636         }
 637
 638         if (dd->dd_phys->dd_reserved > used && parentspace != UINT64_MAX) {
 639                 /*
 640                  * We have some space reserved, in addition to what our
 641                  * parent gave us.
 642                  */
 643                 parentspace += dd->dd_phys->dd_reserved - used;
 644         }
 645
 646         if (dd == ancestor) {
 647                 ASSERT(delta <= 0);
 648                 ASSERT(used >= -delta);
 649                 used += delta;
 650                 if (parentspace != UINT64_MAX)
 651                         parentspace -= delta;
 652         }
 653
 654         if (used > quota) {
 655                 /* over quota */
 656                 myspace = 0;
 657
 658                 /*
 659                  * While it's OK to be a little over quota, if
 660                  * we think we are using more space than there
 661                  * is in the pool (which is already 1.6% more than
 662                  * dsl_pool_adjustedsize()), something is very
 663                  * wrong.
 664                  */
 665                 ASSERT3U(used, <=, spa_get_space(dd->dd_pool->dp_spa));
 666         } else {
 667                 /*
 668                  * the lesser of the space provided by our parent and
 669                  * the space left in our quota
 670                  */
 671                 myspace = MIN(parentspace, quota - used);
 672         }
 673
 674         mutex_exit(&dd->dd_lock);
 675
 676         return (myspace);
 677 }
 678
 679 struct tempreserve {
 680         list_node_t tr_node;
 681         dsl_pool_t *tr_dp;
 682         dsl_dir_t *tr_ds;
 683         uint64_t tr_size;
 684 };
 685
 686 static int
 687 dsl_dir_tempreserve_impl(dsl_dir_t *dd, uint64_t asize, boolean_t netfree,
 688     boolean_t ignorequota, boolean_t checkrefquota, list_t *tr_list,
 689     dmu_tx_t *tx, boolean_t first)
 690 {
 691         uint64_t txg = tx->tx_txg;
 692         uint64_t est_inflight, used_on_disk, quota, parent_rsrv;
 693         struct tempreserve *tr;
 694         int enospc = EDQUOT;
 695         int txgidx = txg & TXG_MASK;
 696         int i;
 697         uint64_t ref_rsrv = 0;
 698
 699         ASSERT3U(txg, !=, 0);
 700         ASSERT3S(asize, >, 0);
 701
 702         mutex_enter(&dd->dd_lock);
 703
 704         /*
 705          * Check against the dsl_dir's quota.  We don't add in the delta
 706          * when checking for over-quota because they get one free hit.
 707          */
 708         est_inflight = dsl_dir_space_towrite(dd);
 709         for (i = 0; i < TXG_SIZE; i++)
 710                 est_inflight += dd->dd_tempreserved[i];
 711         used_on_disk = dd->dd_phys->dd_used_bytes;
 712
 713         /*
 714          * On the first iteration, fetch the dataset's used-on-disk and
 715          * refreservation values. Also, if checkrefquota is set, test if
 716          * allocating this space would exceed the dataset's refquota.
 717          */
 718         if (first && tx->tx_objset) {
 719                 int error;
 720                 dsl_dataset_t *ds = tx->tx_objset->os->os_dsl_dataset;
 721
 722                 error = dsl_dataset_check_quota(ds, checkrefquota,
 723                     asize, est_inflight, &used_on_disk, &ref_rsrv);
 724                 if (error) {
 725                         mutex_exit(&dd->dd_lock);
 726                         return (error);
 727                 }
 728         }
 729
 730         /*
 731          * If this transaction will result in a net free of space,
 732          * we want to let it through.
 733          */
 734         if (ignorequota || netfree || dd->dd_phys->dd_quota == 0)
 735                 quota = UINT64_MAX;
 736         else
 737                 quota = dd->dd_phys->dd_quota;
 738
 739         /*
 740          * Adjust the quota against the actual pool size at the root.
 741          * To ensure that it's possible to remove files from a full
 742          * pool without inducing transient overcommits, we throttle
 743          * netfree transactions against a quota that is slightly larger,
 744          * but still within the pool's allocation slop.  In cases where
 745          * we're very close to full, this will allow a steady trickle of
 746          * removes to get through.
 747          */
 748         if (dd->dd_parent == NULL) {
 749                 uint64_t poolsize = dsl_pool_adjustedsize(dd->dd_pool, netfree);
 750                 if (poolsize < quota) {
 751                         quota = poolsize;
 752                         enospc = ENOSPC;
 753                 }
 754         }
 755
 756         /*
 757          * If they are requesting more space, and our current estimate
 758          * is over quota, they get to try again unless the actual
 759          * on-disk is over quota and there are no pending changes (which
 760          * may free up space for us).
 761          */
 762         if (used_on_disk + est_inflight > quota) {
 763                 if (est_inflight > 0 || used_on_disk < quota)
 764                         enospc = ERESTART;
 765                 dprintf_dd(dd, "failing: used=%lluK inflight = %lluK "
 766                     "quota=%lluK tr=%lluK err=%d\n",
 767                     used_on_disk>>10, est_inflight>>10,
 768                     quota>>10, asize>>10, enospc);
 769                 mutex_exit(&dd->dd_lock);
 770                 return (enospc);
 771         }
 772
 773         /* We need to up our estimated delta before dropping dd_lock */
 774         dd->dd_tempreserved[txgidx] += asize;
 775
 776         parent_rsrv = parent_delta(dd, used_on_disk + est_inflight,
 777             asize - ref_rsrv);
 778         mutex_exit(&dd->dd_lock);
 779
 780         tr = kmem_zalloc(sizeof (struct tempreserve), KM_SLEEP);
 781         tr->tr_ds = dd;
 782         tr->tr_size = asize;
 783         list_insert_tail(tr_list, tr);
 784
 785         /* see if it's OK with our parent */
 786         if (dd->dd_parent && parent_rsrv) {
 787                 boolean_t ismos = (dd->dd_phys->dd_head_dataset_obj == 0);
 788
 789                 return (dsl_dir_tempreserve_impl(dd->dd_parent,
 790                     parent_rsrv, netfree, ismos, TRUE, tr_list, tx, FALSE));
 791         } else {
 792                 return (0);
 793         }
 794 }
 795
 796 /*
 797  * Reserve space in this dsl_dir, to be used in this tx's txg.
 798  * After the space has been dirtied (and dsl_dir_willuse_space()
 799  * has been called), the reservation should be canceled, using
 800  * dsl_dir_tempreserve_clear().
 801  */
 802 int
 803 dsl_dir_tempreserve_space(dsl_dir_t *dd, uint64_t lsize, uint64_t asize,
 804     uint64_t fsize, uint64_t usize, void **tr_cookiep, dmu_tx_t *tx)
 805 {
 806         int err;
 807         list_t *tr_list;
 808
 809         if (asize == 0) {
 810                 *tr_cookiep = NULL;
 811                 return (0);
 812         }
 813
 814         tr_list = kmem_alloc(sizeof (list_t), KM_SLEEP);
 815         list_create(tr_list, sizeof (struct tempreserve),
 816             offsetof(struct tempreserve, tr_node));
 817         ASSERT3S(asize, >, 0);
 818         ASSERT3S(fsize, >=, 0);
 819
 820         err = arc_tempreserve_space(lsize, tx->tx_txg);
 821         if (err == 0) {
 822                 struct tempreserve *tr;
 823
 824                 tr = kmem_zalloc(sizeof (struct tempreserve), KM_SLEEP);
 825                 tr->tr_size = lsize;
 826                 list_insert_tail(tr_list, tr);
 827
 828                 err = dsl_pool_tempreserve_space(dd->dd_pool, asize, tx);
 829         } else {
 830                 if (err == EAGAIN) {
 831                         txg_delay(dd->dd_pool, tx->tx_txg, 1);
 832                         err = ERESTART;
 833                 }
 834                 dsl_pool_memory_pressure(dd->dd_pool);
 835         }
 836
 837         if (err == 0) {
 838                 struct tempreserve *tr;
 839
 840                 tr = kmem_zalloc(sizeof (struct tempreserve), KM_SLEEP);
 841                 tr->tr_dp = dd->dd_pool;
 842                 tr->tr_size = asize;
 843                 list_insert_tail(tr_list, tr);
 844
 845                 err = dsl_dir_tempreserve_impl(dd, asize, fsize >= asize,
 846                     FALSE, asize > usize, tr_list, tx, TRUE);
 847         }
 848
 849         if (err)
 850                 dsl_dir_tempreserve_clear(tr_list, tx);
 851         else
 852                 *tr_cookiep = tr_list;
 853
 854         return (err);
 855 }
 856
 857 /*
 858  * Clear a temporary reservation that we previously made with
 859  * dsl_dir_tempreserve_space().
 860  */
 861 void
 862 dsl_dir_tempreserve_clear(void *tr_cookie, dmu_tx_t *tx)
 863 {
 864         int txgidx = tx->tx_txg & TXG_MASK;
 865         list_t *tr_list = tr_cookie;
 866         struct tempreserve *tr;
 867
 868         ASSERT3U(tx->tx_txg, !=, 0);
 869
 870         if (tr_cookie == NULL)
 871                 return;
 872
 873         while (tr = list_head(tr_list)) {
 874                 if (tr->tr_dp) {
 875                         dsl_pool_tempreserve_clear(tr->tr_dp, tr->tr_size, tx);
 876                 } else if (tr->tr_ds) {
 877                         mutex_enter(&tr->tr_ds->dd_lock);
 878                         ASSERT3U(tr->tr_ds->dd_tempreserved[txgidx], >=,
 879                             tr->tr_size);
 880                         tr->tr_ds->dd_tempreserved[txgidx] -= tr->tr_size;
 881                         mutex_exit(&tr->tr_ds->dd_lock);
 882                 } else {
 883                         arc_tempreserve_clear(tr->tr_size);
 884                 }
 885                 list_remove(tr_list, tr);
 886                 kmem_free(tr, sizeof (struct tempreserve));
 887         }
 888
 889         kmem_free(tr_list, sizeof (list_t));
 890 }
 891
 892 static void
 893 dsl_dir_willuse_space_impl(dsl_dir_t *dd, int64_t space, dmu_tx_t *tx)
 894 {
 895         int64_t parent_space;
 896         uint64_t est_used;
 897
 898         mutex_enter(&dd->dd_lock);
 899         if (space > 0)
 900                 dd->dd_space_towrite[tx->tx_txg & TXG_MASK] += space;
 901
 902         est_used = dsl_dir_space_towrite(dd) + dd->dd_phys->dd_used_bytes;
 903         parent_space = parent_delta(dd, est_used, space);
 904         mutex_exit(&dd->dd_lock);
 905
 906         /* Make sure that we clean up dd_space_to* */
 907         dsl_dir_dirty(dd, tx);
 908
 909         /* XXX this is potentially expensive and unnecessary... */
 910         if (parent_space && dd->dd_parent)
 911                 dsl_dir_willuse_space_impl(dd->dd_parent, parent_space, tx);
 912 }
 913
 914 /*
 915  * Call in open context when we think we're going to write/free space,
 916  * eg. when dirtying data.  Be conservative (ie. OK to write less than
 917  * this or free more than this, but don't write more or free less).
 918  */
 919 void
 920 dsl_dir_willuse_space(dsl_dir_t *dd, int64_t space, dmu_tx_t *tx)
 921 {
 922         dsl_pool_willuse_space(dd->dd_pool, space, tx);
 923         dsl_dir_willuse_space_impl(dd, space, tx);
 924 }
 925
 926 /* call from syncing context when we actually write/free space for this dd */
 927 void
 928 dsl_dir_diduse_space(dsl_dir_t *dd, dd_used_t type,
 929     int64_t used, int64_t compressed, int64_t uncompressed, dmu_tx_t *tx)
 930 {
 931         int64_t accounted_delta;
 932         boolean_t needlock = !MUTEX_HELD(&dd->dd_lock);
 933
 934         ASSERT(dmu_tx_is_syncing(tx));
 935         ASSERT(type < DD_USED_NUM);
 936
 937         dsl_dir_dirty(dd, tx);
 938
 939         if (needlock)
 940                 mutex_enter(&dd->dd_lock);
 941         accounted_delta = parent_delta(dd, dd->dd_phys->dd_used_bytes, used);
 942         ASSERT(used >= 0 || dd->dd_phys->dd_used_bytes >= -used);
 943         ASSERT(compressed >= 0 ||
 944             dd->dd_phys->dd_compressed_bytes >= -compressed);
 945         ASSERT(uncompressed >= 0 ||
 946             dd->dd_phys->dd_uncompressed_bytes >= -uncompressed);
 947         dd->dd_phys->dd_used_bytes += used;
 948         dd->dd_phys->dd_uncompressed_bytes += uncompressed;
 949         dd->dd_phys->dd_compressed_bytes += compressed;
 950
 951         if (dd->dd_phys->dd_flags & DD_FLAG_USED_BREAKDOWN) {
 952                 ASSERT(used > 0 ||
 953                     dd->dd_phys->dd_used_breakdown[type] >= -used);
 954                 dd->dd_phys->dd_used_breakdown[type] += used;
 955 #ifdef DEBUG
 956                 dd_used_t t;
 957                 uint64_t u = 0;
 958                 for (t = 0; t < DD_USED_NUM; t++)
 959                         u += dd->dd_phys->dd_used_breakdown[t];
 960                 ASSERT3U(u, ==, dd->dd_phys->dd_used_bytes);
 961 #endif
 962         }
 963         if (needlock)
 964                 mutex_exit(&dd->dd_lock);
 965
 966         if (dd->dd_parent != NULL) {
 967                 dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD,
 968                     accounted_delta, compressed, uncompressed, tx);
 969                 dsl_dir_transfer_space(dd->dd_parent,
 970                     used - accounted_delta,
 971                     DD_USED_CHILD_RSRV, DD_USED_CHILD, tx);
 972         }
 973 }
 974
 975 void
 976 dsl_dir_transfer_space(dsl_dir_t *dd, int64_t delta,
 977     dd_used_t oldtype, dd_used_t newtype, dmu_tx_t *tx)
 978 {
 979         boolean_t needlock = !MUTEX_HELD(&dd->dd_lock);
 980
 981         ASSERT(dmu_tx_is_syncing(tx));
 982         ASSERT(oldtype < DD_USED_NUM);
 983         ASSERT(newtype < DD_USED_NUM);
 984
 985         if (delta == 0 || !(dd->dd_phys->dd_flags & DD_FLAG_USED_BREAKDOWN))
 986                 return;
 987
 988         dsl_dir_dirty(dd, tx);
 989         if (needlock)
 990                 mutex_enter(&dd->dd_lock);
 991         ASSERT(delta > 0 ?
 992             dd->dd_phys->dd_used_breakdown[oldtype] >= delta :
 993             dd->dd_phys->dd_used_breakdown[newtype] >= -delta);
 994         ASSERT(dd->dd_phys->dd_used_bytes >= ABS(delta));
 995         dd->dd_phys->dd_used_breakdown[oldtype] -= delta;
 996         dd->dd_phys->dd_used_breakdown[newtype] += delta;
 997         if (needlock)
 998                 mutex_exit(&dd->dd_lock);
 999 }
1000
1001 static int
1002 dsl_dir_set_quota_check(void *arg1, void *arg2, dmu_tx_t *tx)
1003 {
1004         dsl_dir_t *dd = arg1;
1005         uint64_t *quotap = arg2;
1006         uint64_t new_quota = *quotap;
1007         int err = 0;
1008         uint64_t towrite;
1009
1010         if (new_quota == 0)
1011                 return (0);
1012
1013         mutex_enter(&dd->dd_lock);
1014         /*
1015          * If we are doing the preliminary check in open context, and
1016          * there are pending changes, then don't fail it, since the
1017          * pending changes could under-estimate the amount of space to be
1018          * freed up.
1019          */
1020         towrite = dsl_dir_space_towrite(dd);
1021         if ((dmu_tx_is_syncing(tx) || towrite == 0) &&
1022             (new_quota < dd->dd_phys->dd_reserved ||
1023             new_quota < dd->dd_phys->dd_used_bytes + towrite)) {
1024                 err = ENOSPC;
1025         }
1026         mutex_exit(&dd->dd_lock);
1027         return (err);
1028 }
1029
1030 /* ARGSUSED */
1031 static void
1032 dsl_dir_set_quota_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
1033 {
1034         dsl_dir_t *dd = arg1;
1035         uint64_t *quotap = arg2;
1036         uint64_t new_quota = *quotap;
1037
1038         dmu_buf_will_dirty(dd->dd_dbuf, tx);
1039
1040         mutex_enter(&dd->dd_lock);
1041         dd->dd_phys->dd_quota = new_quota;
1042         mutex_exit(&dd->dd_lock);
1043
1044         spa_history_internal_log(LOG_DS_QUOTA, dd->dd_pool->dp_spa,
1045             tx, cr, "%lld dataset = %llu ",
1046             (longlong_t)new_quota, dd->dd_phys->dd_head_dataset_obj);
1047 }
1048
1049 int
1050 dsl_dir_set_quota(const char *ddname, uint64_t quota)
1051 {
1052         dsl_dir_t *dd;
1053         int err;
1054
1055         err = dsl_dir_open(ddname, FTAG, &dd, NULL);
1056         if (err)
1057                 return (err);
1058
1059         if (quota != dd->dd_phys->dd_quota) {
1060                 /*
1061                  * If someone removes a file, then tries to set the quota, we
1062                  * want to make sure the file freeing takes effect.
1063                  */
1064                 txg_wait_open(dd->dd_pool, 0);
1065
1066                 err = dsl_sync_task_do(dd->dd_pool, dsl_dir_set_quota_check,
1067                     dsl_dir_set_quota_sync, dd, &quota, 0);
1068         }
1069         dsl_dir_close(dd, FTAG);
1070         return (err);
1071 }
1072
1073 int
1074 dsl_dir_set_reservation_check(void *arg1, void *arg2, dmu_tx_t *tx)
1075 {
1076         dsl_dir_t *dd = arg1;
1077         uint64_t *reservationp = arg2;
1078         uint64_t new_reservation = *reservationp;
1079         uint64_t used, avail;
1080         int64_t delta;
1081
1082         if (new_reservation > INT64_MAX)
1083                 return (EOVERFLOW);
1084
1085         /*
1086          * If we are doing the preliminary check in open context, the
1087          * space estimates may be inaccurate.
1088          */
1089         if (!dmu_tx_is_syncing(tx))
1090                 return (0);
1091
1092         mutex_enter(&dd->dd_lock);
1093         used = dd->dd_phys->dd_used_bytes;
1094         delta = MAX(used, new_reservation) -
1095             MAX(used, dd->dd_phys->dd_reserved);
1096         mutex_exit(&dd->dd_lock);
1097
1098         if (dd->dd_parent) {
1099                 avail = dsl_dir_space_available(dd->dd_parent,
1100                     NULL, 0, FALSE);
1101         } else {
1102                 avail = dsl_pool_adjustedsize(dd->dd_pool, B_FALSE) - used;
1103         }
1104
1105         if (delta > 0 && delta > avail)
1106                 return (ENOSPC);
1107         if (delta > 0 && dd->dd_phys->dd_quota > 0 &&
1108             new_reservation > dd->dd_phys->dd_quota)
1109                 return (ENOSPC);
1110         return (0);
1111 }
1112
1113 /* ARGSUSED */
1114 static void
1115 dsl_dir_set_reservation_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
1116 {
1117         dsl_dir_t *dd = arg1;
1118         uint64_t *reservationp = arg2;
1119         uint64_t new_reservation = *reservationp;
1120         uint64_t used;
1121         int64_t delta;
1122
1123         dmu_buf_will_dirty(dd->dd_dbuf, tx);
1124
1125         mutex_enter(&dd->dd_lock);
1126         used = dd->dd_phys->dd_used_bytes;
1127         delta = MAX(used, new_reservation) -
1128             MAX(used, dd->dd_phys->dd_reserved);
1129         dd->dd_phys->dd_reserved = new_reservation;
1130
1131         if (dd->dd_parent != NULL) {
1132                 /* Roll up this additional usage into our ancestors */
1133                 dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD_RSRV,
1134                     delta, 0, 0, tx);
1135         }
1136         mutex_exit(&dd->dd_lock);
1137
1138         spa_history_internal_log(LOG_DS_RESERVATION, dd->dd_pool->dp_spa,
1139             tx, cr, "%lld dataset = %llu",
1140             (longlong_t)new_reservation, dd->dd_phys->dd_head_dataset_obj);
1141 }
1142
1143 int
1144 dsl_dir_set_reservation(const char *ddname, uint64_t reservation)
1145 {
1146         dsl_dir_t *dd;
1147         int err;
1148
1149         err = dsl_dir_open(ddname, FTAG, &dd, NULL);
1150         if (err)
1151                 return (err);
1152         err = dsl_sync_task_do(dd->dd_pool, dsl_dir_set_reservation_check,
1153             dsl_dir_set_reservation_sync, dd, &reservation, 0);
1154         dsl_dir_close(dd, FTAG);
1155         return (err);
1156 }
1157
1158 static dsl_dir_t *
1159 closest_common_ancestor(dsl_dir_t *ds1, dsl_dir_t *ds2)
1160 {
1161         for (; ds1; ds1 = ds1->dd_parent) {
1162                 dsl_dir_t *dd;
1163                 for (dd = ds2; dd; dd = dd->dd_parent) {
1164                         if (ds1 == dd)
1165                                 return (dd);
1166                 }
1167         }
1168         return (NULL);
1169 }
1170
1171 /*
1172  * If delta is applied to dd, how much of that delta would be applied to
1173  * ancestor?  Syncing context only.
1174  */
1175 static int64_t
1176 would_change(dsl_dir_t *dd, int64_t delta, dsl_dir_t *ancestor)
1177 {
1178         if (dd == ancestor)
1179                 return (delta);
1180
1181         mutex_enter(&dd->dd_lock);
1182         delta = parent_delta(dd, dd->dd_phys->dd_used_bytes, delta);
1183         mutex_exit(&dd->dd_lock);
1184         return (would_change(dd->dd_parent, delta, ancestor));
1185 }
1186
1187 struct renamearg {
1188         dsl_dir_t *newparent;
1189         const char *mynewname;
1190 };
1191
1192 /*ARGSUSED*/
1193 static int
1194 dsl_dir_rename_check(void *arg1, void *arg2, dmu_tx_t *tx)
1195 {
1196         dsl_dir_t *dd = arg1;
1197         struct renamearg *ra = arg2;
1198         dsl_pool_t *dp = dd->dd_pool;
1199         objset_t *mos = dp->dp_meta_objset;
1200         int err;
1201         uint64_t val;
1202
1203         /* There should be 2 references: the open and the dirty */
1204         if (dmu_buf_refcount(dd->dd_dbuf) > 2)
1205                 return (EBUSY);
1206
1207         /* check for existing name */
1208         err = zap_lookup(mos, ra->newparent->dd_phys->dd_child_dir_zapobj,
1209             ra->mynewname, 8, 1, &val);
1210         if (err == 0)
1211                 return (EEXIST);
1212         if (err != ENOENT)
1213                 return (err);
1214
1215         if (ra->newparent != dd->dd_parent) {
1216                 /* is there enough space? */
1217                 uint64_t myspace =
1218                     MAX(dd->dd_phys->dd_used_bytes, dd->dd_phys->dd_reserved);
1219
1220                 /* no rename into our descendant */
1221                 if (closest_common_ancestor(dd, ra->newparent) == dd)
1222                         return (EINVAL);
1223
1224                 if (err = dsl_dir_transfer_possible(dd->dd_parent,
1225                     ra->newparent, myspace))
1226                         return (err);
1227         }
1228
1229         return (0);
1230 }
1231
1232 static void
1233 dsl_dir_rename_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
1234 {
1235         dsl_dir_t *dd = arg1;
1236         struct renamearg *ra = arg2;
1237         dsl_pool_t *dp = dd->dd_pool;
1238         objset_t *mos = dp->dp_meta_objset;
1239         int err;
1240
1241         ASSERT(dmu_buf_refcount(dd->dd_dbuf) <= 2);
1242
1243         if (ra->newparent != dd->dd_parent) {
1244                 dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD,
1245                     -dd->dd_phys->dd_used_bytes,
1246                     -dd->dd_phys->dd_compressed_bytes,
1247                     -dd->dd_phys->dd_uncompressed_bytes, tx);
1248                 dsl_dir_diduse_space(ra->newparent, DD_USED_CHILD,
1249                     dd->dd_phys->dd_used_bytes,
1250                     dd->dd_phys->dd_compressed_bytes,
1251                     dd->dd_phys->dd_uncompressed_bytes, tx);
1252
1253                 if (dd->dd_phys->dd_reserved > dd->dd_phys->dd_used_bytes) {
1254                         uint64_t unused_rsrv = dd->dd_phys->dd_reserved -
1255                             dd->dd_phys->dd_used_bytes;
1256
1257                         dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD_RSRV,
1258                             -unused_rsrv, 0, 0, tx);
1259                         dsl_dir_diduse_space(ra->newparent, DD_USED_CHILD_RSRV,
1260                             unused_rsrv, 0, 0, tx);
1261                 }
1262         }
1263
1264         dmu_buf_will_dirty(dd->dd_dbuf, tx);
1265
1266         /* remove from old parent zapobj */
1267         err = zap_remove(mos, dd->dd_parent->dd_phys->dd_child_dir_zapobj,
1268             dd->dd_myname, tx);
1269         ASSERT3U(err, ==, 0);
1270
1271         (void) strcpy(dd->dd_myname, ra->mynewname);
1272         dsl_dir_close(dd->dd_parent, dd);
1273         dd->dd_phys->dd_parent_obj = ra->newparent->dd_object;
1274         VERIFY(0 == dsl_dir_open_obj(dd->dd_pool,
1275             ra->newparent->dd_object, NULL, dd, &dd->dd_parent));
1276
1277         /* add to new parent zapobj */
1278         err = zap_add(mos, ra->newparent->dd_phys->dd_child_dir_zapobj,
1279             dd->dd_myname, 8, 1, &dd->dd_object, tx);
1280         ASSERT3U(err, ==, 0);
1281
1282         spa_history_internal_log(LOG_DS_RENAME, dd->dd_pool->dp_spa,
1283             tx, cr, "dataset = %llu", dd->dd_phys->dd_head_dataset_obj);
1284 }
1285
1286 int
1287 dsl_dir_rename(dsl_dir_t *dd, const char *newname)
1288 {
1289         struct renamearg ra;
1290         int err;
1291
1292         /* new parent should exist */
1293         err = dsl_dir_open(newname, FTAG, &ra.newparent, &ra.mynewname);
1294         if (err)
1295                 return (err);
1296
1297         /* can't rename to different pool */
1298         if (dd->dd_pool != ra.newparent->dd_pool) {
1299                 err = ENXIO;
1300                 goto out;
1301         }
1302
1303         /* new name should not already exist */
1304         if (ra.mynewname == NULL) {
1305                 err = EEXIST;
1306                 goto out;
1307         }
1308
1309         err = dsl_sync_task_do(dd->dd_pool,
1310             dsl_dir_rename_check, dsl_dir_rename_sync, dd, &ra, 3);
1311
1312 out:
1313         dsl_dir_close(ra.newparent, FTAG);
1314         return (err);
1315 }
1316
1317 int
1318 dsl_dir_transfer_possible(dsl_dir_t *sdd, dsl_dir_t *tdd, uint64_t space)
1319 {
1320         dsl_dir_t *ancestor;
1321         int64_t adelta;
1322         uint64_t avail;
1323
1324         ancestor = closest_common_ancestor(sdd, tdd);
1325         adelta = would_change(sdd, -space, ancestor);
1326         avail = dsl_dir_space_available(tdd, ancestor, adelta, FALSE);
1327         if (avail < space)
1328                 return (ENOSPC);
1329
1330         return (0);
1331 }