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 (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
  23  * Copyright (c) 2011 Pawel Jakub Dawidek <pawel@dawidek.net>.
  24  * All rights reserved.
  25  */
  26
  27 #include <sys/dmu.h>
  28 #include <sys/dmu_objset.h>
  29 #include <sys/dmu_tx.h>
  30 #include <sys/dsl_dataset.h>
  31 #include <sys/dsl_dir.h>
  32 #include <sys/dsl_prop.h>
  33 #include <sys/dsl_synctask.h>
  34 #include <sys/dsl_deleg.h>
  35 #include <sys/spa.h>
  36 #include <sys/metaslab.h>
  37 #include <sys/zap.h>
  38 #include <sys/zio.h>
  39 #include <sys/arc.h>
  40 #include <sys/sunddi.h>
  41 #include <sys/zvol.h>
  42 #ifdef _KERNEL
  43 #include <sys/zfs_vfsops.h>
  44 #endif
  45 #include "zfs_namecheck.h"
  46
  47 static uint64_t dsl_dir_space_towrite(dsl_dir_t *dd);
  48 static void dsl_dir_set_reservation_sync(void *arg1, void *arg2, dmu_tx_t *tx);
  49
  50
  51 /* ARGSUSED */
  52 static void
  53 dsl_dir_evict(dmu_buf_t *db, void *arg)
  54 {
  55         dsl_dir_t *dd = arg;
  56         dsl_pool_t *dp = dd->dd_pool;
  57         int t;
  58
  59         for (t = 0; t < TXG_SIZE; t++) {
  60                 ASSERT(!txg_list_member(&dp->dp_dirty_dirs, dd, t));
  61                 ASSERT(dd->dd_tempreserved[t] == 0);
  62                 ASSERT(dd->dd_space_towrite[t] == 0);
  63         }
  64
  65         if (dd->dd_parent)
  66                 dsl_dir_close(dd->dd_parent, dd);
  67
  68         spa_close(dd->dd_pool->dp_spa, dd);
  69
  70         /*
  71          * The props callback list should have been cleaned up by
  72          * objset_evict().
  73          */
  74         list_destroy(&dd->dd_prop_cbs);
  75         mutex_destroy(&dd->dd_lock);
  76         kmem_free(dd, sizeof (dsl_dir_t));
  77 }
  78
  79 int
  80 dsl_dir_open_obj(dsl_pool_t *dp, uint64_t ddobj,
  81     const char *tail, void *tag, dsl_dir_t **ddp)
  82 {
  83         dmu_buf_t *dbuf;
  84         dsl_dir_t *dd;
  85         int err;
  86
  87         ASSERT(RW_LOCK_HELD(&dp->dp_config_rwlock) ||
  88             dsl_pool_sync_context(dp));
  89
  90         err = dmu_bonus_hold(dp->dp_meta_objset, ddobj, tag, &dbuf);
  91         if (err)
  92                 return (err);
  93         dd = dmu_buf_get_user(dbuf);
  94 #ifdef ZFS_DEBUG
  95         {
  96                 dmu_object_info_t doi;
  97                 dmu_object_info_from_db(dbuf, &doi);
  98                 ASSERT3U(doi.doi_type, ==, DMU_OT_DSL_DIR);
  99                 ASSERT3U(doi.doi_bonus_size, >=, sizeof (dsl_dir_phys_t));
 100         }
 101 #endif
 102         if (dd == NULL) {
 103                 dsl_dir_t *winner;
 104
 105                 dd = kmem_zalloc(sizeof (dsl_dir_t), KM_SLEEP);
 106                 dd->dd_object = ddobj;
 107                 dd->dd_dbuf = dbuf;
 108                 dd->dd_pool = dp;
 109                 dd->dd_phys = dbuf->db_data;
 110                 mutex_init(&dd->dd_lock, NULL, MUTEX_DEFAULT, NULL);
 111
 112                 list_create(&dd->dd_prop_cbs, sizeof (dsl_prop_cb_record_t),
 113                     offsetof(dsl_prop_cb_record_t, cbr_node));
 114
 115                 dsl_dir_snap_cmtime_update(dd);
 116
 117                 if (dd->dd_phys->dd_parent_obj) {
 118                         err = dsl_dir_open_obj(dp, dd->dd_phys->dd_parent_obj,
 119                             NULL, dd, &dd->dd_parent);
 120                         if (err)
 121                                 goto errout;
 122                         if (tail) {
 123 #ifdef ZFS_DEBUG
 124                                 uint64_t foundobj;
 125
 126                                 err = zap_lookup(dp->dp_meta_objset,
 127                                     dd->dd_parent->dd_phys->dd_child_dir_zapobj,
 128                                     tail, sizeof (foundobj), 1, &foundobj);
 129                                 ASSERT(err || foundobj == ddobj);
 130 #endif
 131                                 (void) strcpy(dd->dd_myname, tail);
 132                         } else {
 133                                 err = zap_value_search(dp->dp_meta_objset,
 134                                     dd->dd_parent->dd_phys->dd_child_dir_zapobj,
 135                                     ddobj, 0, dd->dd_myname);
 136                         }
 137                         if (err)
 138                                 goto errout;
 139                 } else {
 140                         (void) strcpy(dd->dd_myname, spa_name(dp->dp_spa));
 141                 }
 142
 143                 if (dsl_dir_is_clone(dd)) {
 144                         dmu_buf_t *origin_bonus;
 145                         dsl_dataset_phys_t *origin_phys;
 146
 147                         /*
 148                          * We can't open the origin dataset, because
 149                          * that would require opening this dsl_dir.
 150                          * Just look at its phys directly instead.
 151                          */
 152                         err = dmu_bonus_hold(dp->dp_meta_objset,
 153                             dd->dd_phys->dd_origin_obj, FTAG, &origin_bonus);
 154                         if (err)
 155                                 goto errout;
 156                         origin_phys = origin_bonus->db_data;
 157                         dd->dd_origin_txg =
 158                             origin_phys->ds_creation_txg;
 159                         dmu_buf_rele(origin_bonus, FTAG);
 160                 }
 161
 162                 winner = dmu_buf_set_user_ie(dbuf, dd, &dd->dd_phys,
 163                     dsl_dir_evict);
 164                 if (winner) {
 165                         if (dd->dd_parent)
 166                                 dsl_dir_close(dd->dd_parent, dd);
 167                         mutex_destroy(&dd->dd_lock);
 168                         kmem_free(dd, sizeof (dsl_dir_t));
 169                         dd = winner;
 170                 } else {
 171                         spa_open_ref(dp->dp_spa, dd);
 172                 }
 173         }
 174
 175         /*
 176          * The dsl_dir_t has both open-to-close and instantiate-to-evict
 177          * holds on the spa.  We need the open-to-close holds because
 178          * otherwise the spa_refcnt wouldn't change when we open a
 179          * dir which the spa also has open, so we could incorrectly
 180          * think it was OK to unload/export/destroy the pool.  We need
 181          * the instantiate-to-evict hold because the dsl_dir_t has a
 182          * pointer to the dd_pool, which has a pointer to the spa_t.
 183          */
 184         spa_open_ref(dp->dp_spa, tag);
 185         ASSERT3P(dd->dd_pool, ==, dp);
 186         ASSERT3U(dd->dd_object, ==, ddobj);
 187         ASSERT3P(dd->dd_dbuf, ==, dbuf);
 188         *ddp = dd;
 189         return (0);
 190
 191 errout:
 192         if (dd->dd_parent)
 193                 dsl_dir_close(dd->dd_parent, dd);
 194         mutex_destroy(&dd->dd_lock);
 195         kmem_free(dd, sizeof (dsl_dir_t));
 196         dmu_buf_rele(dbuf, tag);
 197         return (err);
 198
 199 }
 200
 201 void
 202 dsl_dir_close(dsl_dir_t *dd, void *tag)
 203 {
 204         dprintf_dd(dd, "%s\n", "");
 205         spa_close(dd->dd_pool->dp_spa, tag);
 206         dmu_buf_rele(dd->dd_dbuf, tag);
 207 }
 208
 209 /* buf must be long enough (MAXNAMELEN + strlen(MOS_DIR_NAME) + 1 should do) */
 210 void
 211 dsl_dir_name(dsl_dir_t *dd, char *buf)
 212 {
 213         if (dd->dd_parent) {
 214                 dsl_dir_name(dd->dd_parent, buf);
 215                 (void) strcat(buf, "/");
 216         } else {
 217                 buf[0] = '\0';
 218         }
 219         if (!MUTEX_HELD(&dd->dd_lock)) {
 220                 /*
 221                  * recursive mutex so that we can use
 222                  * dprintf_dd() with dd_lock held
 223                  */
 224                 mutex_enter(&dd->dd_lock);
 225                 (void) strcat(buf, dd->dd_myname);
 226                 mutex_exit(&dd->dd_lock);
 227         } else {
 228                 (void) strcat(buf, dd->dd_myname);
 229         }
 230 }
 231
 232 /* Calculate name legnth, avoiding all the strcat calls of dsl_dir_name */
 233 int
 234 dsl_dir_namelen(dsl_dir_t *dd)
 235 {
 236         int result = 0;
 237
 238         if (dd->dd_parent) {
 239                 /* parent's name + 1 for the "/" */
 240                 result = dsl_dir_namelen(dd->dd_parent) + 1;
 241         }
 242
 243         if (!MUTEX_HELD(&dd->dd_lock)) {
 244                 /* see dsl_dir_name */
 245                 mutex_enter(&dd->dd_lock);
 246                 result += strlen(dd->dd_myname);
 247                 mutex_exit(&dd->dd_lock);
 248         } else {
 249                 result += strlen(dd->dd_myname);
 250         }
 251
 252         return (result);
 253 }
 254
 255 static int
 256 getcomponent(const char *path, char *component, const char **nextp)
 257 {
 258         char *p;
 259         if ((path == NULL) || (path[0] == '\0'))
 260                 return (ENOENT);
 261         /* This would be a good place to reserve some namespace... */
 262         p = strpbrk(path, "/@");
 263         if (p && (p[1] == '/' || p[1] == '@')) {
 264                 /* two separators in a row */
 265                 return (EINVAL);
 266         }
 267         if (p == NULL || p == path) {
 268                 /*
 269                  * if the first thing is an @ or /, it had better be an
 270                  * @ and it had better not have any more ats or slashes,
 271                  * and it had better have something after the @.
 272                  */
 273                 if (p != NULL &&
 274                     (p[0] != '@' || strpbrk(path+1, "/@") || p[1] == '\0'))
 275                         return (EINVAL);
 276                 if (strlen(path) >= MAXNAMELEN)
 277                         return (ENAMETOOLONG);
 278                 (void) strcpy(component, path);
 279                 p = NULL;
 280         } else if (p[0] == '/') {
 281                 if (p-path >= MAXNAMELEN)
 282                         return (ENAMETOOLONG);
 283                 (void) strncpy(component, path, p - path);
 284                 component[p-path] = '\0';
 285                 p++;
 286         } else if (p[0] == '@') {
 287                 /*
 288                  * if the next separator is an @, there better not be
 289                  * any more slashes.
 290                  */
 291                 if (strchr(path, '/'))
 292                         return (EINVAL);
 293                 if (p-path >= MAXNAMELEN)
 294                         return (ENAMETOOLONG);
 295                 (void) strncpy(component, path, p - path);
 296                 component[p-path] = '\0';
 297         } else {
 298                 ASSERT(!"invalid p");
 299         }
 300         *nextp = p;
 301         return (0);
 302 }
 303
 304 /*
 305  * same as dsl_open_dir, ignore the first component of name and use the
 306  * spa instead
 307  */
 308 int
 309 dsl_dir_open_spa(spa_t *spa, const char *name, void *tag,
 310     dsl_dir_t **ddp, const char **tailp)
 311 {
 312         char buf[MAXNAMELEN];
 313         const char *next, *nextnext = NULL;
 314         int err;
 315         dsl_dir_t *dd;
 316         dsl_pool_t *dp;
 317         uint64_t ddobj;
 318         int openedspa = FALSE;
 319
 320         dprintf("%s\n", name);
 321
 322         err = getcomponent(name, buf, &next);
 323         if (err)
 324                 return (err);
 325         if (spa == NULL) {
 326                 err = spa_open(buf, &spa, FTAG);
 327                 if (err) {
 328                         dprintf("spa_open(%s) failed\n", buf);
 329                         return (err);
 330                 }
 331                 openedspa = TRUE;
 332
 333                 /* XXX this assertion belongs in spa_open */
 334                 ASSERT(!dsl_pool_sync_context(spa_get_dsl(spa)));
 335         }
 336
 337         dp = spa_get_dsl(spa);
 338
 339         rw_enter(&dp->dp_config_rwlock, RW_READER);
 340         err = dsl_dir_open_obj(dp, dp->dp_root_dir_obj, NULL, tag, &dd);
 341         if (err) {
 342                 rw_exit(&dp->dp_config_rwlock);
 343                 if (openedspa)
 344                         spa_close(spa, FTAG);
 345                 return (err);
 346         }
 347
 348         while (next != NULL) {
 349                 dsl_dir_t *child_ds;
 350                 err = getcomponent(next, buf, &nextnext);
 351                 if (err)
 352                         break;
 353                 ASSERT(next[0] != '\0');
 354                 if (next[0] == '@')
 355                         break;
 356                 dprintf("looking up %s in obj%lld\n",
 357                     buf, dd->dd_phys->dd_child_dir_zapobj);
 358
 359                 err = zap_lookup(dp->dp_meta_objset,
 360                     dd->dd_phys->dd_child_dir_zapobj,
 361                     buf, sizeof (ddobj), 1, &ddobj);
 362                 if (err) {
 363                         if (err == ENOENT)
 364                                 err = 0;
 365                         break;
 366                 }
 367
 368                 err = dsl_dir_open_obj(dp, ddobj, buf, tag, &child_ds);
 369                 if (err)
 370                         break;
 371                 dsl_dir_close(dd, tag);
 372                 dd = child_ds;
 373                 next = nextnext;
 374         }
 375         rw_exit(&dp->dp_config_rwlock);
 376
 377         if (err) {
 378                 dsl_dir_close(dd, tag);
 379                 if (openedspa)
 380                         spa_close(spa, FTAG);
 381                 return (err);
 382         }
 383
 384         /*
 385          * It's an error if there's more than one component left, or
 386          * tailp==NULL and there's any component left.
 387          */
 388         if (next != NULL &&
 389             (tailp == NULL || (nextnext && nextnext[0] != '\0'))) {
 390                 /* bad path name */
 391                 dsl_dir_close(dd, tag);
 392                 dprintf("next=%p (%s) tail=%p\n", next, next?next:"", tailp);
 393                 err = ENOENT;
 394         }
 395         if (tailp)
 396                 *tailp = next;
 397         if (openedspa)
 398                 spa_close(spa, FTAG);
 399         *ddp = dd;
 400         return (err);
 401 }
 402
 403 /*
 404  * Return the dsl_dir_t, and possibly the last component which couldn't
 405  * be found in *tail.  Return NULL if the path is bogus, or if
 406  * tail==NULL and we couldn't parse the whole name.  (*tail)[0] == '@'
 407  * means that the last component is a snapshot.
 408  */
 409 int
 410 dsl_dir_open(const char *name, void *tag, dsl_dir_t **ddp, const char **tailp)
 411 {
 412         return (dsl_dir_open_spa(NULL, name, tag, ddp, tailp));
 413 }
 414
 415 uint64_t
 416 dsl_dir_create_sync(dsl_pool_t *dp, dsl_dir_t *pds, const char *name,
 417     dmu_tx_t *tx)
 418 {
 419         objset_t *mos = dp->dp_meta_objset;
 420         uint64_t ddobj;
 421         dsl_dir_phys_t *ddphys;
 422         dmu_buf_t *dbuf;
 423
 424         ddobj = dmu_object_alloc(mos, DMU_OT_DSL_DIR, 0,
 425             DMU_OT_DSL_DIR, sizeof (dsl_dir_phys_t), tx);
 426         if (pds) {
 427                 VERIFY(0 == zap_add(mos, pds->dd_phys->dd_child_dir_zapobj,
 428                     name, sizeof (uint64_t), 1, &ddobj, tx));
 429         } else {
 430                 /* it's the root dir */
 431                 VERIFY(0 == zap_add(mos, DMU_POOL_DIRECTORY_OBJECT,
 432                     DMU_POOL_ROOT_DATASET, sizeof (uint64_t), 1, &ddobj, tx));
 433         }
 434         VERIFY(0 == dmu_bonus_hold(mos, ddobj, FTAG, &dbuf));
 435         dmu_buf_will_dirty(dbuf, tx);
 436         ddphys = dbuf->db_data;
 437
 438         ddphys->dd_creation_time = gethrestime_sec();
 439         if (pds)
 440                 ddphys->dd_parent_obj = pds->dd_object;
 441         ddphys->dd_props_zapobj = zap_create(mos,
 442             DMU_OT_DSL_PROPS, DMU_OT_NONE, 0, tx);
 443         ddphys->dd_child_dir_zapobj = zap_create(mos,
 444             DMU_OT_DSL_DIR_CHILD_MAP, DMU_OT_NONE, 0, tx);
 445         if (spa_version(dp->dp_spa) >= SPA_VERSION_USED_BREAKDOWN)
 446                 ddphys->dd_flags |= DD_FLAG_USED_BREAKDOWN;
 447         dmu_buf_rele(dbuf, FTAG);
 448
 449         return (ddobj);
 450 }
 451
 452 /* ARGSUSED */
 453 int
 454 dsl_dir_destroy_check(void *arg1, void *arg2, dmu_tx_t *tx)
 455 {
 456         dsl_dataset_t *ds = arg1;
 457         dsl_dir_t *dd = ds->ds_dir;
 458         dsl_pool_t *dp = dd->dd_pool;
 459         objset_t *mos = dp->dp_meta_objset;
 460         int err;
 461         uint64_t count;
 462
 463         /*
 464          * There should be exactly two holds, both from
 465          * dsl_dataset_destroy: one on the dd directory, and one on its
 466          * head ds.  Otherwise, someone is trying to lookup something
 467          * inside this dir while we want to destroy it.  The
 468          * config_rwlock ensures that nobody else opens it after we
 469          * check.
 470          */
 471         if (dmu_buf_refcount(dd->dd_dbuf) > 2)
 472                 return (EBUSY);
 473
 474         err = zap_count(mos, dd->dd_phys->dd_child_dir_zapobj, &count);
 475         if (err)
 476                 return (err);
 477         if (count != 0)
 478                 return (EEXIST);
 479
 480         return (0);
 481 }
 482
 483 void
 484 dsl_dir_destroy_sync(void *arg1, void *tag, dmu_tx_t *tx)
 485 {
 486         dsl_dataset_t *ds = arg1;
 487         dsl_dir_t *dd = ds->ds_dir;
 488         objset_t *mos = dd->dd_pool->dp_meta_objset;
 489         dsl_prop_setarg_t psa;
 490         uint64_t value = 0;
 491         uint64_t obj;
 492         dd_used_t t;
 493
 494         ASSERT(RW_WRITE_HELD(&dd->dd_pool->dp_config_rwlock));
 495         ASSERT(dd->dd_phys->dd_head_dataset_obj == 0);
 496
 497         /* Remove our reservation. */
 498         dsl_prop_setarg_init_uint64(&psa, "reservation",
 499             (ZPROP_SRC_NONE | ZPROP_SRC_LOCAL | ZPROP_SRC_RECEIVED),
 500             &value);
 501         psa.psa_effective_value = 0;    /* predict default value */
 502
 503         dsl_dir_set_reservation_sync(ds, &psa, tx);
 504
 505         ASSERT3U(dd->dd_phys->dd_used_bytes, ==, 0);
 506         ASSERT3U(dd->dd_phys->dd_reserved, ==, 0);
 507         for (t = 0; t < DD_USED_NUM; t++)
 508                 ASSERT3U(dd->dd_phys->dd_used_breakdown[t], ==, 0);
 509
 510         VERIFY(0 == zap_destroy(mos, dd->dd_phys->dd_child_dir_zapobj, tx));
 511         VERIFY(0 == zap_destroy(mos, dd->dd_phys->dd_props_zapobj, tx));
 512         VERIFY(0 == dsl_deleg_destroy(mos, dd->dd_phys->dd_deleg_zapobj, tx));
 513         VERIFY(0 == zap_remove(mos,
 514             dd->dd_parent->dd_phys->dd_child_dir_zapobj, dd->dd_myname, tx));
 515
 516         obj = dd->dd_object;
 517         dsl_dir_close(dd, tag);
 518         VERIFY(0 == dmu_object_free(mos, obj, tx));
 519 }
 520
 521 boolean_t
 522 dsl_dir_is_clone(dsl_dir_t *dd)
 523 {
 524         return (dd->dd_phys->dd_origin_obj &&
 525             (dd->dd_pool->dp_origin_snap == NULL ||
 526             dd->dd_phys->dd_origin_obj !=
 527             dd->dd_pool->dp_origin_snap->ds_object));
 528 }
 529
 530 void
 531 dsl_dir_stats(dsl_dir_t *dd, nvlist_t *nv)
 532 {
 533         mutex_enter(&dd->dd_lock);
 534         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USED,
 535             dd->dd_phys->dd_used_bytes);
 536         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_QUOTA, dd->dd_phys->dd_quota);
 537         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_RESERVATION,
 538             dd->dd_phys->dd_reserved);
 539         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_COMPRESSRATIO,
 540             dd->dd_phys->dd_compressed_bytes == 0 ? 100 :
 541             (dd->dd_phys->dd_uncompressed_bytes * 100 /
 542             dd->dd_phys->dd_compressed_bytes));
 543         if (dd->dd_phys->dd_flags & DD_FLAG_USED_BREAKDOWN) {
 544                 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDSNAP,
 545                     dd->dd_phys->dd_used_breakdown[DD_USED_SNAP]);
 546                 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDDS,
 547                     dd->dd_phys->dd_used_breakdown[DD_USED_HEAD]);
 548                 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDREFRESERV,
 549                     dd->dd_phys->dd_used_breakdown[DD_USED_REFRSRV]);
 550                 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDCHILD,
 551                     dd->dd_phys->dd_used_breakdown[DD_USED_CHILD] +
 552                     dd->dd_phys->dd_used_breakdown[DD_USED_CHILD_RSRV]);
 553         }
 554         mutex_exit(&dd->dd_lock);
 555
 556         rw_enter(&dd->dd_pool->dp_config_rwlock, RW_READER);
 557         if (dsl_dir_is_clone(dd)) {
 558                 dsl_dataset_t *ds;
 559                 char buf[MAXNAMELEN];
 560
 561                 VERIFY(0 == dsl_dataset_hold_obj(dd->dd_pool,
 562                     dd->dd_phys->dd_origin_obj, FTAG, &ds));
 563                 dsl_dataset_name(ds, buf);
 564                 dsl_dataset_rele(ds, FTAG);
 565                 dsl_prop_nvlist_add_string(nv, ZFS_PROP_ORIGIN, buf);
 566         }
 567         rw_exit(&dd->dd_pool->dp_config_rwlock);
 568 }
 569
 570 void
 571 dsl_dir_dirty(dsl_dir_t *dd, dmu_tx_t *tx)
 572 {
 573         dsl_pool_t *dp = dd->dd_pool;
 574
 575         ASSERT(dd->dd_phys);
 576
 577         if (txg_list_add(&dp->dp_dirty_dirs, dd, tx->tx_txg) == 0) {
 578                 /* up the hold count until we can be written out */
 579                 dmu_buf_add_ref(dd->dd_dbuf, dd);
 580         }
 581 }
 582
 583 static int64_t
 584 parent_delta(dsl_dir_t *dd, uint64_t used, int64_t delta)
 585 {
 586         uint64_t old_accounted = MAX(used, dd->dd_phys->dd_reserved);
 587         uint64_t new_accounted = MAX(used + delta, dd->dd_phys->dd_reserved);
 588         return (new_accounted - old_accounted);
 589 }
 590
 591 void
 592 dsl_dir_sync(dsl_dir_t *dd, dmu_tx_t *tx)
 593 {
 594         ASSERT(dmu_tx_is_syncing(tx));
 595
 596         dmu_buf_will_dirty(dd->dd_dbuf, tx);
 597
 598         mutex_enter(&dd->dd_lock);
 599         ASSERT3U(dd->dd_tempreserved[tx->tx_txg&TXG_MASK], ==, 0);
 600         dprintf_dd(dd, "txg=%llu towrite=%lluK\n", tx->tx_txg,
 601             dd->dd_space_towrite[tx->tx_txg&TXG_MASK] / 1024);
 602         dd->dd_space_towrite[tx->tx_txg&TXG_MASK] = 0;
 603         mutex_exit(&dd->dd_lock);
 604
 605         /* release the hold from dsl_dir_dirty */
 606         dmu_buf_rele(dd->dd_dbuf, dd);
 607 }
 608
 609 static uint64_t
 610 dsl_dir_space_towrite(dsl_dir_t *dd)
 611 {
 612         uint64_t space = 0;
 613         int i;
 614
 615         ASSERT(MUTEX_HELD(&dd->dd_lock));
 616
 617         for (i = 0; i < TXG_SIZE; i++) {
 618                 space += dd->dd_space_towrite[i&TXG_MASK];
 619                 ASSERT3U(dd->dd_space_towrite[i&TXG_MASK], >=, 0);
 620         }
 621         return (space);
 622 }
 623
 624 /*
 625  * How much space would dd have available if ancestor had delta applied
 626  * to it?  If ondiskonly is set, we're only interested in what's
 627  * on-disk, not estimated pending changes.
 628  */
 629 uint64_t
 630 dsl_dir_space_available(dsl_dir_t *dd,
 631     dsl_dir_t *ancestor, int64_t delta, int ondiskonly)
 632 {
 633         uint64_t parentspace, myspace, quota, used;
 634
 635         /*
 636          * If there are no restrictions otherwise, assume we have
 637          * unlimited space available.
 638          */
 639         quota = UINT64_MAX;
 640         parentspace = UINT64_MAX;
 641
 642         if (dd->dd_parent != NULL) {
 643                 parentspace = dsl_dir_space_available(dd->dd_parent,
 644                     ancestor, delta, ondiskonly);
 645         }
 646
 647         mutex_enter(&dd->dd_lock);
 648         if (dd->dd_phys->dd_quota != 0)
 649                 quota = dd->dd_phys->dd_quota;
 650         used = dd->dd_phys->dd_used_bytes;
 651         if (!ondiskonly)
 652                 used += dsl_dir_space_towrite(dd);
 653
 654         if (dd->dd_parent == NULL) {
 655                 uint64_t poolsize = dsl_pool_adjustedsize(dd->dd_pool, FALSE);
 656                 quota = MIN(quota, poolsize);
 657         }
 658
 659         if (dd->dd_phys->dd_reserved > used && parentspace != UINT64_MAX) {
 660                 /*
 661                  * We have some space reserved, in addition to what our
 662                  * parent gave us.
 663                  */
 664                 parentspace += dd->dd_phys->dd_reserved - used;
 665         }
 666
 667         if (dd == ancestor) {
 668                 ASSERT(delta <= 0);
 669                 ASSERT(used >= -delta);
 670                 used += delta;
 671                 if (parentspace != UINT64_MAX)
 672                         parentspace -= delta;
 673         }
 674
 675         if (used > quota) {
 676                 /* over quota */
 677                 myspace = 0;
 678         } else {
 679                 /*
 680                  * the lesser of the space provided by our parent and
 681                  * the space left in our quota
 682                  */
 683                 myspace = MIN(parentspace, quota - used);
 684         }
 685
 686         mutex_exit(&dd->dd_lock);
 687
 688         return (myspace);
 689 }
 690
 691 struct tempreserve {
 692         list_node_t tr_node;
 693         dsl_pool_t *tr_dp;
 694         dsl_dir_t *tr_ds;
 695         uint64_t tr_size;
 696 };
 697
 698 static int
 699 dsl_dir_tempreserve_impl(dsl_dir_t *dd, uint64_t asize, boolean_t netfree,
 700     boolean_t ignorequota, boolean_t checkrefquota, list_t *tr_list,
 701     dmu_tx_t *tx, boolean_t first)
 702 {
 703         uint64_t txg = tx->tx_txg;
 704         uint64_t est_inflight, used_on_disk, quota, parent_rsrv;
 705         uint64_t deferred = 0;
 706         struct tempreserve *tr;
 707         int retval = EDQUOT;
 708         int txgidx = txg & TXG_MASK;
 709         int i;
 710         uint64_t ref_rsrv = 0;
 711
 712         ASSERT3U(txg, !=, 0);
 713         ASSERT3S(asize, >, 0);
 714
 715         mutex_enter(&dd->dd_lock);
 716
 717         /*
 718          * Check against the dsl_dir's quota.  We don't add in the delta
 719          * when checking for over-quota because they get one free hit.
 720          */
 721         est_inflight = dsl_dir_space_towrite(dd);
 722         for (i = 0; i < TXG_SIZE; i++)
 723                 est_inflight += dd->dd_tempreserved[i];
 724         used_on_disk = dd->dd_phys->dd_used_bytes;
 725
 726         /*
 727          * On the first iteration, fetch the dataset's used-on-disk and
 728          * refreservation values. Also, if checkrefquota is set, test if
 729          * allocating this space would exceed the dataset's refquota.
 730          */
 731         if (first && tx->tx_objset) {
 732                 int error;
 733                 dsl_dataset_t *ds = tx->tx_objset->os_dsl_dataset;
 734
 735                 error = dsl_dataset_check_quota(ds, checkrefquota,
 736                     asize, est_inflight, &used_on_disk, &ref_rsrv);
 737                 if (error) {
 738                         mutex_exit(&dd->dd_lock);
 739                         return (error);
 740                 }
 741         }
 742
 743         /*
 744          * If this transaction will result in a net free of space,
 745          * we want to let it through.
 746          */
 747         if (ignorequota || netfree || dd->dd_phys->dd_quota == 0)
 748                 quota = UINT64_MAX;
 749         else
 750                 quota = dd->dd_phys->dd_quota;
 751
 752         /*
 753          * Adjust the quota against the actual pool size at the root
 754          * minus any outstanding deferred frees.
 755          * To ensure that it's possible to remove files from a full
 756          * pool without inducing transient overcommits, we throttle
 757          * netfree transactions against a quota that is slightly larger,
 758          * but still within the pool's allocation slop.  In cases where
 759          * we're very close to full, this will allow a steady trickle of
 760          * removes to get through.
 761          */
 762         if (dd->dd_parent == NULL) {
 763                 spa_t *spa = dd->dd_pool->dp_spa;
 764                 uint64_t poolsize = dsl_pool_adjustedsize(dd->dd_pool, netfree);
 765                 deferred = metaslab_class_get_deferred(spa_normal_class(spa));
 766                 if (poolsize - deferred < quota) {
 767                         quota = poolsize - deferred;
 768                         retval = ENOSPC;
 769                 }
 770         }
 771
 772         /*
 773          * If they are requesting more space, and our current estimate
 774          * is over quota, they get to try again unless the actual
 775          * on-disk is over quota and there are no pending changes (which
 776          * may free up space for us).
 777          */
 778         if (used_on_disk + est_inflight >= quota) {
 779                 if (est_inflight > 0 || used_on_disk < quota ||
 780                     (retval == ENOSPC && used_on_disk < quota + deferred))
 781                         retval = ERESTART;
 782                 dprintf_dd(dd, "failing: used=%lluK inflight = %lluK "
 783                     "quota=%lluK tr=%lluK err=%d\n",
 784                     used_on_disk>>10, est_inflight>>10,
 785                     quota>>10, asize>>10, retval);
 786                 mutex_exit(&dd->dd_lock);
 787                 return (retval);
 788         }
 789
 790         /* We need to up our estimated delta before dropping dd_lock */
 791         dd->dd_tempreserved[txgidx] += asize;
 792
 793         parent_rsrv = parent_delta(dd, used_on_disk + est_inflight,
 794             asize - ref_rsrv);
 795         mutex_exit(&dd->dd_lock);
 796
 797         tr = kmem_zalloc(sizeof (struct tempreserve), KM_SLEEP);
 798         tr->tr_ds = dd;
 799         tr->tr_size = asize;
 800         list_insert_tail(tr_list, tr);
 801
 802         /* see if it's OK with our parent */
 803         if (dd->dd_parent && parent_rsrv) {
 804                 boolean_t ismos = (dd->dd_phys->dd_head_dataset_obj == 0);
 805
 806                 return (dsl_dir_tempreserve_impl(dd->dd_parent,
 807                     parent_rsrv, netfree, ismos, TRUE, tr_list, tx, FALSE));
 808         } else {
 809                 return (0);
 810         }
 811 }
 812
 813 /*
 814  * Reserve space in this dsl_dir, to be used in this tx's txg.
 815  * After the space has been dirtied (and dsl_dir_willuse_space()
 816  * has been called), the reservation should be canceled, using
 817  * dsl_dir_tempreserve_clear().
 818  */
 819 int
 820 dsl_dir_tempreserve_space(dsl_dir_t *dd, uint64_t lsize, uint64_t asize,
 821     uint64_t fsize, uint64_t usize, void **tr_cookiep, dmu_tx_t *tx)
 822 {
 823         int err;
 824         list_t *tr_list;
 825
 826         if (asize == 0) {
 827                 *tr_cookiep = NULL;
 828                 return (0);
 829         }
 830
 831         tr_list = kmem_alloc(sizeof (list_t), KM_SLEEP);
 832         list_create(tr_list, sizeof (struct tempreserve),
 833             offsetof(struct tempreserve, tr_node));
 834         ASSERT3S(asize, >, 0);
 835         ASSERT3S(fsize, >=, 0);
 836
 837         err = arc_tempreserve_space(lsize, tx->tx_txg);
 838         if (err == 0) {
 839                 struct tempreserve *tr;
 840
 841                 tr = kmem_zalloc(sizeof (struct tempreserve), KM_SLEEP);
 842                 tr->tr_size = lsize;
 843                 list_insert_tail(tr_list, tr);
 844
 845                 err = dsl_pool_tempreserve_space(dd->dd_pool, asize, tx);
 846         } else {
 847                 if (err == EAGAIN) {
 848                         txg_delay(dd->dd_pool, tx->tx_txg, 1);
 849                         err = ERESTART;
 850                 }
 851                 dsl_pool_memory_pressure(dd->dd_pool);
 852         }
 853
 854         if (err == 0) {
 855                 struct tempreserve *tr;
 856
 857                 tr = kmem_zalloc(sizeof (struct tempreserve), KM_SLEEP);
 858                 tr->tr_dp = dd->dd_pool;
 859                 tr->tr_size = asize;
 860                 list_insert_tail(tr_list, tr);
 861
 862                 err = dsl_dir_tempreserve_impl(dd, asize, fsize >= asize,
 863                     FALSE, asize > usize, tr_list, tx, TRUE);
 864         }
 865
 866         if (err)
 867                 dsl_dir_tempreserve_clear(tr_list, tx);
 868         else
 869                 *tr_cookiep = tr_list;
 870
 871         return (err);
 872 }
 873
 874 /*
 875  * Clear a temporary reservation that we previously made with
 876  * dsl_dir_tempreserve_space().
 877  */
 878 void
 879 dsl_dir_tempreserve_clear(void *tr_cookie, dmu_tx_t *tx)
 880 {
 881         int txgidx = tx->tx_txg & TXG_MASK;
 882         list_t *tr_list = tr_cookie;
 883         struct tempreserve *tr;
 884
 885         ASSERT3U(tx->tx_txg, !=, 0);
 886
 887         if (tr_cookie == NULL)
 888                 return;
 889
 890         while (tr = list_head(tr_list)) {
 891                 if (tr->tr_dp) {
 892                         dsl_pool_tempreserve_clear(tr->tr_dp, tr->tr_size, tx);
 893                 } else if (tr->tr_ds) {
 894                         mutex_enter(&tr->tr_ds->dd_lock);
 895                         ASSERT3U(tr->tr_ds->dd_tempreserved[txgidx], >=,
 896                             tr->tr_size);
 897                         tr->tr_ds->dd_tempreserved[txgidx] -= tr->tr_size;
 898                         mutex_exit(&tr->tr_ds->dd_lock);
 899                 } else {
 900                         arc_tempreserve_clear(tr->tr_size);
 901                 }
 902                 list_remove(tr_list, tr);
 903                 kmem_free(tr, sizeof (struct tempreserve));
 904         }
 905
 906         kmem_free(tr_list, sizeof (list_t));
 907 }
 908
 909 static void
 910 dsl_dir_willuse_space_impl(dsl_dir_t *dd, int64_t space, dmu_tx_t *tx)
 911 {
 912         int64_t parent_space;
 913         uint64_t est_used;
 914
 915         mutex_enter(&dd->dd_lock);
 916         if (space > 0)
 917                 dd->dd_space_towrite[tx->tx_txg & TXG_MASK] += space;
 918
 919         est_used = dsl_dir_space_towrite(dd) + dd->dd_phys->dd_used_bytes;
 920         parent_space = parent_delta(dd, est_used, space);
 921         mutex_exit(&dd->dd_lock);
 922
 923         /* Make sure that we clean up dd_space_to* */
 924         dsl_dir_dirty(dd, tx);
 925
 926         /* XXX this is potentially expensive and unnecessary... */
 927         if (parent_space && dd->dd_parent)
 928                 dsl_dir_willuse_space_impl(dd->dd_parent, parent_space, tx);
 929 }
 930
 931 /*
 932  * Call in open context when we think we're going to write/free space,
 933  * eg. when dirtying data.  Be conservative (ie. OK to write less than
 934  * this or free more than this, but don't write more or free less).
 935  */
 936 void
 937 dsl_dir_willuse_space(dsl_dir_t *dd, int64_t space, dmu_tx_t *tx)
 938 {
 939         dsl_pool_willuse_space(dd->dd_pool, space, tx);
 940         dsl_dir_willuse_space_impl(dd, space, tx);
 941 }
 942
 943 /* call from syncing context when we actually write/free space for this dd */
 944 void
 945 dsl_dir_diduse_space(dsl_dir_t *dd, dd_used_t type,
 946     int64_t used, int64_t compressed, int64_t uncompressed, dmu_tx_t *tx)
 947 {
 948         int64_t accounted_delta;
 949         boolean_t needlock = !MUTEX_HELD(&dd->dd_lock);
 950
 951         ASSERT(dmu_tx_is_syncing(tx));
 952         ASSERT(type < DD_USED_NUM);
 953
 954         dsl_dir_dirty(dd, tx);
 955
 956         if (needlock)
 957                 mutex_enter(&dd->dd_lock);
 958         accounted_delta = parent_delta(dd, dd->dd_phys->dd_used_bytes, used);
 959         ASSERT(used >= 0 || dd->dd_phys->dd_used_bytes >= -used);
 960         ASSERT(compressed >= 0 ||
 961             dd->dd_phys->dd_compressed_bytes >= -compressed);
 962         ASSERT(uncompressed >= 0 ||
 963             dd->dd_phys->dd_uncompressed_bytes >= -uncompressed);
 964         dd->dd_phys->dd_used_bytes += used;
 965         dd->dd_phys->dd_uncompressed_bytes += uncompressed;
 966         dd->dd_phys->dd_compressed_bytes += compressed;
 967
 968         if (dd->dd_phys->dd_flags & DD_FLAG_USED_BREAKDOWN) {
 969                 ASSERT(used > 0 ||
 970                     dd->dd_phys->dd_used_breakdown[type] >= -used);
 971                 dd->dd_phys->dd_used_breakdown[type] += used;
 972 #ifdef DEBUG
 973                 dd_used_t t;
 974                 uint64_t u = 0;
 975                 for (t = 0; t < DD_USED_NUM; t++)
 976                         u += dd->dd_phys->dd_used_breakdown[t];
 977                 ASSERT3U(u, ==, dd->dd_phys->dd_used_bytes);
 978 #endif
 979         }
 980         if (needlock)
 981                 mutex_exit(&dd->dd_lock);
 982
 983         if (dd->dd_parent != NULL) {
 984                 dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD,
 985                     accounted_delta, compressed, uncompressed, tx);
 986                 dsl_dir_transfer_space(dd->dd_parent,
 987                     used - accounted_delta,
 988                     DD_USED_CHILD_RSRV, DD_USED_CHILD, tx);
 989         }
 990 }
 991
 992 void
 993 dsl_dir_transfer_space(dsl_dir_t *dd, int64_t delta,
 994     dd_used_t oldtype, dd_used_t newtype, dmu_tx_t *tx)
 995 {
 996         boolean_t needlock = !MUTEX_HELD(&dd->dd_lock);
 997
 998         ASSERT(dmu_tx_is_syncing(tx));
 999         ASSERT(oldtype < DD_USED_NUM);
1000         ASSERT(newtype < DD_USED_NUM);
1001
1002         if (delta == 0 || !(dd->dd_phys->dd_flags & DD_FLAG_USED_BREAKDOWN))
1003                 return;
1004
1005         dsl_dir_dirty(dd, tx);
1006         if (needlock)
1007                 mutex_enter(&dd->dd_lock);
1008         ASSERT(delta > 0 ?
1009             dd->dd_phys->dd_used_breakdown[oldtype] >= delta :
1010             dd->dd_phys->dd_used_breakdown[newtype] >= -delta);
1011         ASSERT(dd->dd_phys->dd_used_bytes >= ABS(delta));
1012         dd->dd_phys->dd_used_breakdown[oldtype] -= delta;
1013         dd->dd_phys->dd_used_breakdown[newtype] += delta;
1014         if (needlock)
1015                 mutex_exit(&dd->dd_lock);
1016 }
1017
1018 static int
1019 dsl_dir_set_quota_check(void *arg1, void *arg2, dmu_tx_t *tx)
1020 {
1021         dsl_dataset_t *ds = arg1;
1022         dsl_dir_t *dd = ds->ds_dir;
1023         dsl_prop_setarg_t *psa = arg2;
1024         int err;
1025         uint64_t towrite;
1026
1027         if ((err = dsl_prop_predict_sync(ds->ds_dir, psa)) != 0)
1028                 return (err);
1029
1030         if (psa->psa_effective_value == 0)
1031                 return (0);
1032
1033         mutex_enter(&dd->dd_lock);
1034         /*
1035          * If we are doing the preliminary check in open context, and
1036          * there are pending changes, then don't fail it, since the
1037          * pending changes could under-estimate the amount of space to be
1038          * freed up.
1039          */
1040         towrite = dsl_dir_space_towrite(dd);
1041         if ((dmu_tx_is_syncing(tx) || towrite == 0) &&
1042             (psa->psa_effective_value < dd->dd_phys->dd_reserved ||
1043             psa->psa_effective_value < dd->dd_phys->dd_used_bytes + towrite)) {
1044                 err = ENOSPC;
1045         }
1046         mutex_exit(&dd->dd_lock);
1047         return (err);
1048 }
1049
1050 extern dsl_syncfunc_t dsl_prop_set_sync;
1051
1052 static void
1053 dsl_dir_set_quota_sync(void *arg1, void *arg2, dmu_tx_t *tx)
1054 {
1055         dsl_dataset_t *ds = arg1;
1056         dsl_dir_t *dd = ds->ds_dir;
1057         dsl_prop_setarg_t *psa = arg2;
1058         uint64_t effective_value = psa->psa_effective_value;
1059
1060         dsl_prop_set_sync(ds, psa, tx);
1061         DSL_PROP_CHECK_PREDICTION(dd, psa);
1062
1063         dmu_buf_will_dirty(dd->dd_dbuf, tx);
1064
1065         mutex_enter(&dd->dd_lock);
1066         dd->dd_phys->dd_quota = effective_value;
1067         mutex_exit(&dd->dd_lock);
1068
1069         spa_history_log_internal(LOG_DS_QUOTA, dd->dd_pool->dp_spa,
1070             tx, "%lld dataset = %llu ",
1071             (longlong_t)effective_value, dd->dd_phys->dd_head_dataset_obj);
1072 }
1073
1074 int
1075 dsl_dir_set_quota(const char *ddname, zprop_source_t source, uint64_t quota)
1076 {
1077         dsl_dir_t *dd;
1078         dsl_dataset_t *ds;
1079         dsl_prop_setarg_t psa;
1080         int err;
1081
1082         dsl_prop_setarg_init_uint64(&psa, "quota", source, &quota);
1083
1084         err = dsl_dataset_hold(ddname, FTAG, &ds);
1085         if (err)
1086                 return (err);
1087
1088         err = dsl_dir_open(ddname, FTAG, &dd, NULL);
1089         if (err) {
1090                 dsl_dataset_rele(ds, FTAG);
1091                 return (err);
1092         }
1093
1094         ASSERT(ds->ds_dir == dd);
1095
1096         /*
1097          * If someone removes a file, then tries to set the quota, we want to
1098          * make sure the file freeing takes effect.
1099          */
1100         txg_wait_open(dd->dd_pool, 0);
1101
1102         err = dsl_sync_task_do(dd->dd_pool, dsl_dir_set_quota_check,
1103             dsl_dir_set_quota_sync, ds, &psa, 0);
1104
1105         dsl_dir_close(dd, FTAG);
1106         dsl_dataset_rele(ds, FTAG);
1107         return (err);
1108 }
1109
1110 int
1111 dsl_dir_set_reservation_check(void *arg1, void *arg2, dmu_tx_t *tx)
1112 {
1113         dsl_dataset_t *ds = arg1;
1114         dsl_dir_t *dd = ds->ds_dir;
1115         dsl_prop_setarg_t *psa = arg2;
1116         uint64_t effective_value;
1117         uint64_t used, avail;
1118         int err;
1119
1120         if ((err = dsl_prop_predict_sync(ds->ds_dir, psa)) != 0)
1121                 return (err);
1122
1123         effective_value = psa->psa_effective_value;
1124
1125         /*
1126          * If we are doing the preliminary check in open context, the
1127          * space estimates may be inaccurate.
1128          */
1129         if (!dmu_tx_is_syncing(tx))
1130                 return (0);
1131
1132         mutex_enter(&dd->dd_lock);
1133         used = dd->dd_phys->dd_used_bytes;
1134         mutex_exit(&dd->dd_lock);
1135
1136         if (dd->dd_parent) {
1137                 avail = dsl_dir_space_available(dd->dd_parent,
1138                     NULL, 0, FALSE);
1139         } else {
1140                 avail = dsl_pool_adjustedsize(dd->dd_pool, B_FALSE) - used;
1141         }
1142
1143         if (MAX(used, effective_value) > MAX(used, dd->dd_phys->dd_reserved)) {
1144                 uint64_t delta = MAX(used, effective_value) -
1145                     MAX(used, dd->dd_phys->dd_reserved);
1146
1147                 if (delta > avail)
1148                         return (ENOSPC);
1149                 if (dd->dd_phys->dd_quota > 0 &&
1150                     effective_value > dd->dd_phys->dd_quota)
1151                         return (ENOSPC);
1152         }
1153
1154         return (0);
1155 }
1156
1157 static void
1158 dsl_dir_set_reservation_sync(void *arg1, void *arg2, dmu_tx_t *tx)
1159 {
1160         dsl_dataset_t *ds = arg1;
1161         dsl_dir_t *dd = ds->ds_dir;
1162         dsl_prop_setarg_t *psa = arg2;
1163         uint64_t effective_value = psa->psa_effective_value;
1164         uint64_t used;
1165         int64_t delta;
1166
1167         dsl_prop_set_sync(ds, psa, tx);
1168         DSL_PROP_CHECK_PREDICTION(dd, psa);
1169
1170         dmu_buf_will_dirty(dd->dd_dbuf, tx);
1171
1172         mutex_enter(&dd->dd_lock);
1173         used = dd->dd_phys->dd_used_bytes;
1174         delta = MAX(used, effective_value) -
1175             MAX(used, dd->dd_phys->dd_reserved);
1176         dd->dd_phys->dd_reserved = effective_value;
1177
1178         if (dd->dd_parent != NULL) {
1179                 /* Roll up this additional usage into our ancestors */
1180                 dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD_RSRV,
1181                     delta, 0, 0, tx);
1182         }
1183         mutex_exit(&dd->dd_lock);
1184
1185         spa_history_log_internal(LOG_DS_RESERVATION, dd->dd_pool->dp_spa,
1186             tx, "%lld dataset = %llu",
1187             (longlong_t)effective_value, dd->dd_phys->dd_head_dataset_obj);
1188 }
1189
1190 int
1191 dsl_dir_set_reservation(const char *ddname, zprop_source_t source,
1192     uint64_t reservation)
1193 {
1194         dsl_dir_t *dd;
1195         dsl_dataset_t *ds;
1196         dsl_prop_setarg_t psa;
1197         int err;
1198
1199         dsl_prop_setarg_init_uint64(&psa, "reservation", source, &reservation);
1200
1201         err = dsl_dataset_hold(ddname, FTAG, &ds);
1202         if (err)
1203                 return (err);
1204
1205         err = dsl_dir_open(ddname, FTAG, &dd, NULL);
1206         if (err) {
1207                 dsl_dataset_rele(ds, FTAG);
1208                 return (err);
1209         }
1210
1211         ASSERT(ds->ds_dir == dd);
1212
1213         err = dsl_sync_task_do(dd->dd_pool, dsl_dir_set_reservation_check,
1214             dsl_dir_set_reservation_sync, ds, &psa, 0);
1215
1216         dsl_dir_close(dd, FTAG);
1217         dsl_dataset_rele(ds, FTAG);
1218         return (err);
1219 }
1220
1221 static dsl_dir_t *
1222 closest_common_ancestor(dsl_dir_t *ds1, dsl_dir_t *ds2)
1223 {
1224         for (; ds1; ds1 = ds1->dd_parent) {
1225                 dsl_dir_t *dd;
1226                 for (dd = ds2; dd; dd = dd->dd_parent) {
1227                         if (ds1 == dd)
1228                                 return (dd);
1229                 }
1230         }
1231         return (NULL);
1232 }
1233
1234 /*
1235  * If delta is applied to dd, how much of that delta would be applied to
1236  * ancestor?  Syncing context only.
1237  */
1238 static int64_t
1239 would_change(dsl_dir_t *dd, int64_t delta, dsl_dir_t *ancestor)
1240 {
1241         if (dd == ancestor)
1242                 return (delta);
1243
1244         mutex_enter(&dd->dd_lock);
1245         delta = parent_delta(dd, dd->dd_phys->dd_used_bytes, delta);
1246         mutex_exit(&dd->dd_lock);
1247         return (would_change(dd->dd_parent, delta, ancestor));
1248 }
1249
1250 struct renamearg {
1251         dsl_dir_t *newparent;
1252         const char *mynewname;
1253         boolean_t allowmounted;
1254 };
1255
1256 static int
1257 dsl_dir_rename_check(void *arg1, void *arg2, dmu_tx_t *tx)
1258 {
1259         dsl_dir_t *dd = arg1;
1260         struct renamearg *ra = arg2;
1261         dsl_pool_t *dp = dd->dd_pool;
1262         objset_t *mos = dp->dp_meta_objset;
1263         int err;
1264         uint64_t val;
1265
1266         /*
1267          * There should only be one reference, from dmu_objset_rename().
1268          * Fleeting holds are also possible (eg, from "zfs list" getting
1269          * stats), but any that are present in open context will likely
1270          * be gone by syncing context, so only fail from syncing
1271          * context.
1272          * Don't check if we allow renaming of busy (mounted) dataset.
1273          */
1274         if (!ra->allowmounted && dmu_tx_is_syncing(tx) &&
1275             dmu_buf_refcount(dd->dd_dbuf) > 1) {
1276                 return (EBUSY);
1277         }
1278
1279         /* check for existing name */
1280         err = zap_lookup(mos, ra->newparent->dd_phys->dd_child_dir_zapobj,
1281             ra->mynewname, 8, 1, &val);
1282         if (err == 0)
1283                 return (EEXIST);
1284         if (err != ENOENT)
1285                 return (err);
1286
1287         if (ra->newparent != dd->dd_parent) {
1288                 /* is there enough space? */
1289                 uint64_t myspace =
1290                     MAX(dd->dd_phys->dd_used_bytes, dd->dd_phys->dd_reserved);
1291
1292                 /* no rename into our descendant */
1293                 if (closest_common_ancestor(dd, ra->newparent) == dd)
1294                         return (EINVAL);
1295
1296                 if (err = dsl_dir_transfer_possible(dd->dd_parent,
1297                     ra->newparent, myspace))
1298                         return (err);
1299         }
1300
1301         return (0);
1302 }
1303
1304 static void
1305 dsl_dir_rename_sync(void *arg1, void *arg2, dmu_tx_t *tx)
1306 {
1307         char oldname[MAXPATHLEN], newname[MAXPATHLEN];
1308         dsl_dir_t *dd = arg1;
1309         struct renamearg *ra = arg2;
1310         dsl_pool_t *dp = dd->dd_pool;
1311         objset_t *mos = dp->dp_meta_objset;
1312         int err;
1313
1314         ASSERT(ra->allowmounted || dmu_buf_refcount(dd->dd_dbuf) <= 2);
1315
1316         if (ra->newparent != dd->dd_parent) {
1317                 dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD,
1318                     -dd->dd_phys->dd_used_bytes,
1319                     -dd->dd_phys->dd_compressed_bytes,
1320                     -dd->dd_phys->dd_uncompressed_bytes, tx);
1321                 dsl_dir_diduse_space(ra->newparent, DD_USED_CHILD,
1322                     dd->dd_phys->dd_used_bytes,
1323                     dd->dd_phys->dd_compressed_bytes,
1324                     dd->dd_phys->dd_uncompressed_bytes, tx);
1325
1326                 if (dd->dd_phys->dd_reserved > dd->dd_phys->dd_used_bytes) {
1327                         uint64_t unused_rsrv = dd->dd_phys->dd_reserved -
1328                             dd->dd_phys->dd_used_bytes;
1329
1330                         dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD_RSRV,
1331                             -unused_rsrv, 0, 0, tx);
1332                         dsl_dir_diduse_space(ra->newparent, DD_USED_CHILD_RSRV,
1333                             unused_rsrv, 0, 0, tx);
1334                 }
1335         }
1336
1337         dmu_buf_will_dirty(dd->dd_dbuf, tx);
1338
1339         /* remove from old parent zapobj */
1340         dsl_dir_name(dd, oldname);
1341         err = zap_remove(mos, dd->dd_parent->dd_phys->dd_child_dir_zapobj,
1342             dd->dd_myname, tx);
1343         ASSERT3U(err, ==, 0);
1344
1345         (void) strcpy(dd->dd_myname, ra->mynewname);
1346         dsl_dir_close(dd->dd_parent, dd);
1347         dd->dd_phys->dd_parent_obj = ra->newparent->dd_object;
1348         VERIFY(0 == dsl_dir_open_obj(dd->dd_pool,
1349             ra->newparent->dd_object, NULL, dd, &dd->dd_parent));
1350
1351         /* add to new parent zapobj */
1352         err = zap_add(mos, ra->newparent->dd_phys->dd_child_dir_zapobj,
1353             dd->dd_myname, 8, 1, &dd->dd_object, tx);
1354         ASSERT3U(err, ==, 0);
1355         dsl_dir_name(dd, newname);
1356 #ifdef _KERNEL
1357         zfsvfs_update_fromname(oldname, newname);
1358         zvol_rename_minors(oldname, newname);
1359 #endif
1360
1361         spa_history_log_internal(LOG_DS_RENAME, dd->dd_pool->dp_spa,
1362             tx, "dataset = %llu", dd->dd_phys->dd_head_dataset_obj);
1363 }
1364
1365 int
1366 dsl_dir_rename(dsl_dir_t *dd, const char *newname, int flags)
1367 {
1368         struct renamearg ra;
1369         int err;
1370
1371         /* new parent should exist */
1372         err = dsl_dir_open(newname, FTAG, &ra.newparent, &ra.mynewname);
1373         if (err)
1374                 return (err);
1375
1376         /* can't rename to different pool */
1377         if (dd->dd_pool != ra.newparent->dd_pool) {
1378                 err = ENXIO;
1379                 goto out;
1380         }
1381
1382         /* new name should not already exist */
1383         if (ra.mynewname == NULL) {
1384                 err = EEXIST;
1385                 goto out;
1386         }
1387
1388         ra.allowmounted = !!(flags & ZFS_RENAME_ALLOW_MOUNTED);
1389
1390         err = dsl_sync_task_do(dd->dd_pool,
1391             dsl_dir_rename_check, dsl_dir_rename_sync, dd, &ra, 3);
1392
1393 out:
1394         dsl_dir_close(ra.newparent, FTAG);
1395         return (err);
1396 }
1397
1398 int
1399 dsl_dir_transfer_possible(dsl_dir_t *sdd, dsl_dir_t *tdd, uint64_t space)
1400 {
1401         dsl_dir_t *ancestor;
1402         int64_t adelta;
1403         uint64_t avail;
1404
1405         ancestor = closest_common_ancestor(sdd, tdd);
1406         adelta = would_change(sdd, -space, ancestor);
1407         avail = dsl_dir_space_available(tdd, ancestor, adelta, FALSE);
1408         if (avail < space)
1409                 return (ENOSPC);
1410
1411         return (0);
1412 }
1413
1414 timestruc_t
1415 dsl_dir_snap_cmtime(dsl_dir_t *dd)
1416 {
1417         timestruc_t t;
1418
1419         mutex_enter(&dd->dd_lock);
1420         t = dd->dd_snap_cmtime;
1421         mutex_exit(&dd->dd_lock);
1422
1423         return (t);
1424 }
1425
1426 void
1427 dsl_dir_snap_cmtime_update(dsl_dir_t *dd)
1428 {
1429         timestruc_t t;
1430
1431         gethrestime(&t);
1432         mutex_enter(&dd->dd_lock);
1433         dd->dd_snap_cmtime = t;
1434         mutex_exit(&dd->dd_lock);
1435 }