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  * Copyright (c) 2013 by Delphix. All rights reserved.
  26  */
  27
  28 #include <sys/dmu.h>
  29 #include <sys/dmu_objset.h>
  30 #include <sys/dmu_tx.h>
  31 #include <sys/dsl_dataset.h>
  32 #include <sys/dsl_dir.h>
  33 #include <sys/dsl_prop.h>
  34 #include <sys/dsl_synctask.h>
  35 #include <sys/dsl_deleg.h>
  36 #include <sys/spa.h>
  37 #include <sys/metaslab.h>
  38 #include <sys/zap.h>
  39 #include <sys/zio.h>
  40 #include <sys/arc.h>
  41 #include <sys/sunddi.h>
  42 #include <sys/zvol.h>
  43 #ifdef _KERNEL
  44 #include <sys/zfs_vfsops.h>
  45 #endif
  46 #include "zfs_namecheck.h"
  47
  48 static uint64_t dsl_dir_space_towrite(dsl_dir_t *dd);
  49
  50 /* ARGSUSED */
  51 static void
  52 dsl_dir_evict(dmu_buf_t *db, void *arg)
  53 {
  54         dsl_dir_t *dd = arg;
  55         dsl_pool_t *dp = dd->dd_pool;
  56         int t;
  57
  58         for (t = 0; t < TXG_SIZE; t++) {
  59                 ASSERT(!txg_list_member(&dp->dp_dirty_dirs, dd, t));
  60                 ASSERT(dd->dd_tempreserved[t] == 0);
  61                 ASSERT(dd->dd_space_towrite[t] == 0);
  62         }
  63
  64         if (dd->dd_parent)
  65                 dsl_dir_rele(dd->dd_parent, dd);
  66
  67         spa_close(dd->dd_pool->dp_spa, dd);
  68
  69         /*
  70          * The props callback list should have been cleaned up by
  71          * objset_evict().
  72          */
  73         list_destroy(&dd->dd_prop_cbs);
  74         mutex_destroy(&dd->dd_lock);
  75         kmem_free(dd, sizeof (dsl_dir_t));
  76 }
  77
  78 int
  79 dsl_dir_hold_obj(dsl_pool_t *dp, uint64_t ddobj,
  80     const char *tail, void *tag, dsl_dir_t **ddp)
  81 {
  82         dmu_buf_t *dbuf;
  83         dsl_dir_t *dd;
  84         int err;
  85
  86         ASSERT(dsl_pool_config_held(dp));
  87
  88         err = dmu_bonus_hold(dp->dp_meta_objset, ddobj, tag, &dbuf);
  89         if (err != 0)
  90                 return (err);
  91         dd = dmu_buf_get_user(dbuf);
  92 #ifdef ZFS_DEBUG
  93         {
  94                 dmu_object_info_t doi;
  95                 dmu_object_info_from_db(dbuf, &doi);
  96                 ASSERT3U(doi.doi_type, ==, DMU_OT_DSL_DIR);
  97                 ASSERT3U(doi.doi_bonus_size, >=, sizeof (dsl_dir_phys_t));
  98         }
  99 #endif
 100         if (dd == NULL) {
 101                 dsl_dir_t *winner;
 102
 103                 dd = kmem_zalloc(sizeof (dsl_dir_t), KM_SLEEP);
 104                 dd->dd_object = ddobj;
 105                 dd->dd_dbuf = dbuf;
 106                 dd->dd_pool = dp;
 107                 dd->dd_phys = dbuf->db_data;
 108                 mutex_init(&dd->dd_lock, NULL, MUTEX_DEFAULT, NULL);
 109
 110                 list_create(&dd->dd_prop_cbs, sizeof (dsl_prop_cb_record_t),
 111                     offsetof(dsl_prop_cb_record_t, cbr_node));
 112
 113                 dsl_dir_snap_cmtime_update(dd);
 114
 115                 if (dd->dd_phys->dd_parent_obj) {
 116                         err = dsl_dir_hold_obj(dp, dd->dd_phys->dd_parent_obj,
 117                             NULL, dd, &dd->dd_parent);
 118                         if (err != 0)
 119                                 goto errout;
 120                         if (tail) {
 121 #ifdef ZFS_DEBUG
 122                                 uint64_t foundobj;
 123
 124                                 err = zap_lookup(dp->dp_meta_objset,
 125                                     dd->dd_parent->dd_phys->dd_child_dir_zapobj,
 126                                     tail, sizeof (foundobj), 1, &foundobj);
 127                                 ASSERT(err || foundobj == ddobj);
 128 #endif
 129                                 (void) strcpy(dd->dd_myname, tail);
 130                         } else {
 131                                 err = zap_value_search(dp->dp_meta_objset,
 132                                     dd->dd_parent->dd_phys->dd_child_dir_zapobj,
 133                                     ddobj, 0, dd->dd_myname);
 134                         }
 135                         if (err != 0)
 136                                 goto errout;
 137                 } else {
 138                         (void) strcpy(dd->dd_myname, spa_name(dp->dp_spa));
 139                 }
 140
 141                 if (dsl_dir_is_clone(dd)) {
 142                         dmu_buf_t *origin_bonus;
 143                         dsl_dataset_phys_t *origin_phys;
 144
 145                         /*
 146                          * We can't open the origin dataset, because
 147                          * that would require opening this dsl_dir.
 148                          * Just look at its phys directly instead.
 149                          */
 150                         err = dmu_bonus_hold(dp->dp_meta_objset,
 151                             dd->dd_phys->dd_origin_obj, FTAG, &origin_bonus);
 152                         if (err != 0)
 153                                 goto errout;
 154                         origin_phys = origin_bonus->db_data;
 155                         dd->dd_origin_txg =
 156                             origin_phys->ds_creation_txg;
 157                         dmu_buf_rele(origin_bonus, FTAG);
 158                 }
 159
 160                 winner = dmu_buf_set_user_ie(dbuf, dd, &dd->dd_phys,
 161                     dsl_dir_evict);
 162                 if (winner) {
 163                         if (dd->dd_parent)
 164                                 dsl_dir_rele(dd->dd_parent, dd);
 165                         mutex_destroy(&dd->dd_lock);
 166                         kmem_free(dd, sizeof (dsl_dir_t));
 167                         dd = winner;
 168                 } else {
 169                         spa_open_ref(dp->dp_spa, dd);
 170                 }
 171         }
 172
 173         /*
 174          * The dsl_dir_t has both open-to-close and instantiate-to-evict
 175          * holds on the spa.  We need the open-to-close holds because
 176          * otherwise the spa_refcnt wouldn't change when we open a
 177          * dir which the spa also has open, so we could incorrectly
 178          * think it was OK to unload/export/destroy the pool.  We need
 179          * the instantiate-to-evict hold because the dsl_dir_t has a
 180          * pointer to the dd_pool, which has a pointer to the spa_t.
 181          */
 182         spa_open_ref(dp->dp_spa, tag);
 183         ASSERT3P(dd->dd_pool, ==, dp);
 184         ASSERT3U(dd->dd_object, ==, ddobj);
 185         ASSERT3P(dd->dd_dbuf, ==, dbuf);
 186         *ddp = dd;
 187         return (0);
 188
 189 errout:
 190         if (dd->dd_parent)
 191                 dsl_dir_rele(dd->dd_parent, dd);
 192         mutex_destroy(&dd->dd_lock);
 193         kmem_free(dd, sizeof (dsl_dir_t));
 194         dmu_buf_rele(dbuf, tag);
 195         return (err);
 196 }
 197
 198 void
 199 dsl_dir_rele(dsl_dir_t *dd, void *tag)
 200 {
 201         dprintf_dd(dd, "%s\n", "");
 202         spa_close(dd->dd_pool->dp_spa, tag);
 203         dmu_buf_rele(dd->dd_dbuf, tag);
 204 }
 205
 206 /* buf must be long enough (MAXNAMELEN + strlen(MOS_DIR_NAME) + 1 should do) */
 207 void
 208 dsl_dir_name(dsl_dir_t *dd, char *buf)
 209 {
 210         if (dd->dd_parent) {
 211                 dsl_dir_name(dd->dd_parent, buf);
 212                 (void) strcat(buf, "/");
 213         } else {
 214                 buf[0] = '\0';
 215         }
 216         if (!MUTEX_HELD(&dd->dd_lock)) {
 217                 /*
 218                  * recursive mutex so that we can use
 219                  * dprintf_dd() with dd_lock held
 220                  */
 221                 mutex_enter(&dd->dd_lock);
 222                 (void) strcat(buf, dd->dd_myname);
 223                 mutex_exit(&dd->dd_lock);
 224         } else {
 225                 (void) strcat(buf, dd->dd_myname);
 226         }
 227 }
 228
 229 /* Calculate name length, avoiding all the strcat calls of dsl_dir_name */
 230 int
 231 dsl_dir_namelen(dsl_dir_t *dd)
 232 {
 233         int result = 0;
 234
 235         if (dd->dd_parent) {
 236                 /* parent's name + 1 for the "/" */
 237                 result = dsl_dir_namelen(dd->dd_parent) + 1;
 238         }
 239
 240         if (!MUTEX_HELD(&dd->dd_lock)) {
 241                 /* see dsl_dir_name */
 242                 mutex_enter(&dd->dd_lock);
 243                 result += strlen(dd->dd_myname);
 244                 mutex_exit(&dd->dd_lock);
 245         } else {
 246                 result += strlen(dd->dd_myname);
 247         }
 248
 249         return (result);
 250 }
 251
 252 static int
 253 getcomponent(const char *path, char *component, const char **nextp)
 254 {
 255         char *p;
 256
 257         if ((path == NULL) || (path[0] == '\0'))
 258                 return (SET_ERROR(ENOENT));
 259         /* This would be a good place to reserve some namespace... */
 260         p = strpbrk(path, "/@");
 261         if (p && (p[1] == '/' || p[1] == '@')) {
 262                 /* two separators in a row */
 263                 return (SET_ERROR(EINVAL));
 264         }
 265         if (p == NULL || p == path) {
 266                 /*
 267                  * if the first thing is an @ or /, it had better be an
 268                  * @ and it had better not have any more ats or slashes,
 269                  * and it had better have something after the @.
 270                  */
 271                 if (p != NULL &&
 272                     (p[0] != '@' || strpbrk(path+1, "/@") || p[1] == '\0'))
 273                         return (SET_ERROR(EINVAL));
 274                 if (strlen(path) >= MAXNAMELEN)
 275                         return (SET_ERROR(ENAMETOOLONG));
 276                 (void) strcpy(component, path);
 277                 p = NULL;
 278         } else if (p[0] == '/') {
 279                 if (p - path >= MAXNAMELEN)
 280                         return (SET_ERROR(ENAMETOOLONG));
 281                 (void) strncpy(component, path, p - path);
 282                 component[p - path] = '\0';
 283                 p++;
 284         } else if (p[0] == '@') {
 285                 /*
 286                  * if the next separator is an @, there better not be
 287                  * any more slashes.
 288                  */
 289                 if (strchr(path, '/'))
 290                         return (SET_ERROR(EINVAL));
 291                 if (p - path >= MAXNAMELEN)
 292                         return (SET_ERROR(ENAMETOOLONG));
 293                 (void) strncpy(component, path, p - path);
 294                 component[p - path] = '\0';
 295         } else {
 296                 panic("invalid p=%p", (void *)p);
 297         }
 298         *nextp = p;
 299         return (0);
 300 }
 301
 302 /*
 303  * Return the dsl_dir_t, and possibly the last component which couldn't
 304  * be found in *tail.  The name must be in the specified dsl_pool_t.  This
 305  * thread must hold the dp_config_rwlock for the pool.  Returns NULL if the
 306  * path is bogus, or if tail==NULL and we couldn't parse the whole name.
 307  * (*tail)[0] == '@' means that the last component is a snapshot.
 308  */
 309 int
 310 dsl_dir_hold(dsl_pool_t *dp, const char *name, void *tag,
 311     dsl_dir_t **ddp, const char **tailp)
 312 {
 313         char buf[MAXNAMELEN];
 314         const char *spaname, *next, *nextnext = NULL;
 315         int err;
 316         dsl_dir_t *dd;
 317         uint64_t ddobj;
 318
 319         err = getcomponent(name, buf, &next);
 320         if (err != 0)
 321                 return (err);
 322
 323         /* Make sure the name is in the specified pool. */
 324         spaname = spa_name(dp->dp_spa);
 325         if (strcmp(buf, spaname) != 0)
 326                 return (SET_ERROR(EINVAL));
 327
 328         ASSERT(dsl_pool_config_held(dp));
 329
 330         err = dsl_dir_hold_obj(dp, dp->dp_root_dir_obj, NULL, tag, &dd);
 331         if (err != 0) {
 332                 return (err);
 333         }
 334
 335         while (next != NULL) {
 336                 dsl_dir_t *child_ds;
 337                 err = getcomponent(next, buf, &nextnext);
 338                 if (err != 0)
 339                         break;
 340                 ASSERT(next[0] != '\0');
 341                 if (next[0] == '@')
 342                         break;
 343                 dprintf("looking up %s in obj%lld\n",
 344                     buf, dd->dd_phys->dd_child_dir_zapobj);
 345
 346                 err = zap_lookup(dp->dp_meta_objset,
 347                     dd->dd_phys->dd_child_dir_zapobj,
 348                     buf, sizeof (ddobj), 1, &ddobj);
 349                 if (err != 0) {
 350                         if (err == ENOENT)
 351                                 err = 0;
 352                         break;
 353                 }
 354
 355                 err = dsl_dir_hold_obj(dp, ddobj, buf, tag, &child_ds);
 356                 if (err != 0)
 357                         break;
 358                 dsl_dir_rele(dd, tag);
 359                 dd = child_ds;
 360                 next = nextnext;
 361         }
 362
 363         if (err != 0) {
 364                 dsl_dir_rele(dd, tag);
 365                 return (err);
 366         }
 367
 368         /*
 369          * It's an error if there's more than one component left, or
 370          * tailp==NULL and there's any component left.
 371          */
 372         if (next != NULL &&
 373             (tailp == NULL || (nextnext && nextnext[0] != '\0'))) {
 374                 /* bad path name */
 375                 dsl_dir_rele(dd, tag);
 376                 dprintf("next=%p (%s) tail=%p\n", next, next?next:"", tailp);
 377                 err = SET_ERROR(ENOENT);
 378         }
 379         if (tailp != NULL)
 380                 *tailp = next;
 381         *ddp = dd;
 382         return (err);
 383 }
 384
 385 uint64_t
 386 dsl_dir_create_sync(dsl_pool_t *dp, dsl_dir_t *pds, const char *name,
 387     dmu_tx_t *tx)
 388 {
 389         objset_t *mos = dp->dp_meta_objset;
 390         uint64_t ddobj;
 391         dsl_dir_phys_t *ddphys;
 392         dmu_buf_t *dbuf;
 393
 394         ddobj = dmu_object_alloc(mos, DMU_OT_DSL_DIR, 0,
 395             DMU_OT_DSL_DIR, sizeof (dsl_dir_phys_t), tx);
 396         if (pds) {
 397                 VERIFY(0 == zap_add(mos, pds->dd_phys->dd_child_dir_zapobj,
 398                     name, sizeof (uint64_t), 1, &ddobj, tx));
 399         } else {
 400                 /* it's the root dir */
 401                 VERIFY(0 == zap_add(mos, DMU_POOL_DIRECTORY_OBJECT,
 402                     DMU_POOL_ROOT_DATASET, sizeof (uint64_t), 1, &ddobj, tx));
 403         }
 404         VERIFY(0 == dmu_bonus_hold(mos, ddobj, FTAG, &dbuf));
 405         dmu_buf_will_dirty(dbuf, tx);
 406         ddphys = dbuf->db_data;
 407
 408         ddphys->dd_creation_time = gethrestime_sec();
 409         if (pds)
 410                 ddphys->dd_parent_obj = pds->dd_object;
 411         ddphys->dd_props_zapobj = zap_create(mos,
 412             DMU_OT_DSL_PROPS, DMU_OT_NONE, 0, tx);
 413         ddphys->dd_child_dir_zapobj = zap_create(mos,
 414             DMU_OT_DSL_DIR_CHILD_MAP, DMU_OT_NONE, 0, tx);
 415         if (spa_version(dp->dp_spa) >= SPA_VERSION_USED_BREAKDOWN)
 416                 ddphys->dd_flags |= DD_FLAG_USED_BREAKDOWN;
 417         dmu_buf_rele(dbuf, FTAG);
 418
 419         return (ddobj);
 420 }
 421
 422 boolean_t
 423 dsl_dir_is_clone(dsl_dir_t *dd)
 424 {
 425         return (dd->dd_phys->dd_origin_obj &&
 426             (dd->dd_pool->dp_origin_snap == NULL ||
 427             dd->dd_phys->dd_origin_obj !=
 428             dd->dd_pool->dp_origin_snap->ds_object));
 429 }
 430
 431 void
 432 dsl_dir_stats(dsl_dir_t *dd, nvlist_t *nv)
 433 {
 434         mutex_enter(&dd->dd_lock);
 435         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USED,
 436             dd->dd_phys->dd_used_bytes);
 437         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_QUOTA, dd->dd_phys->dd_quota);
 438         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_RESERVATION,
 439             dd->dd_phys->dd_reserved);
 440         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_COMPRESSRATIO,
 441             dd->dd_phys->dd_compressed_bytes == 0 ? 100 :
 442             (dd->dd_phys->dd_uncompressed_bytes * 100 /
 443             dd->dd_phys->dd_compressed_bytes));
 444         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_LOGICALUSED,
 445             dd->dd_phys->dd_uncompressed_bytes);
 446         if (dd->dd_phys->dd_flags & DD_FLAG_USED_BREAKDOWN) {
 447                 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDSNAP,
 448                     dd->dd_phys->dd_used_breakdown[DD_USED_SNAP]);
 449                 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDDS,
 450                     dd->dd_phys->dd_used_breakdown[DD_USED_HEAD]);
 451                 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDREFRESERV,
 452                     dd->dd_phys->dd_used_breakdown[DD_USED_REFRSRV]);
 453                 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDCHILD,
 454                     dd->dd_phys->dd_used_breakdown[DD_USED_CHILD] +
 455                     dd->dd_phys->dd_used_breakdown[DD_USED_CHILD_RSRV]);
 456         }
 457         mutex_exit(&dd->dd_lock);
 458
 459         if (dsl_dir_is_clone(dd)) {
 460                 dsl_dataset_t *ds;
 461                 char buf[MAXNAMELEN];
 462
 463                 VERIFY0(dsl_dataset_hold_obj(dd->dd_pool,
 464                     dd->dd_phys->dd_origin_obj, FTAG, &ds));
 465                 dsl_dataset_name(ds, buf);
 466                 dsl_dataset_rele(ds, FTAG);
 467                 dsl_prop_nvlist_add_string(nv, ZFS_PROP_ORIGIN, buf);
 468         }
 469 }
 470
 471 void
 472 dsl_dir_dirty(dsl_dir_t *dd, dmu_tx_t *tx)
 473 {
 474         dsl_pool_t *dp = dd->dd_pool;
 475
 476         ASSERT(dd->dd_phys);
 477
 478         if (txg_list_add(&dp->dp_dirty_dirs, dd, tx->tx_txg)) {
 479                 /* up the hold count until we can be written out */
 480                 dmu_buf_add_ref(dd->dd_dbuf, dd);
 481         }
 482 }
 483
 484 static int64_t
 485 parent_delta(dsl_dir_t *dd, uint64_t used, int64_t delta)
 486 {
 487         uint64_t old_accounted = MAX(used, dd->dd_phys->dd_reserved);
 488         uint64_t new_accounted = MAX(used + delta, dd->dd_phys->dd_reserved);
 489         return (new_accounted - old_accounted);
 490 }
 491
 492 void
 493 dsl_dir_sync(dsl_dir_t *dd, dmu_tx_t *tx)
 494 {
 495         ASSERT(dmu_tx_is_syncing(tx));
 496
 497         mutex_enter(&dd->dd_lock);
 498         ASSERT0(dd->dd_tempreserved[tx->tx_txg&TXG_MASK]);
 499         dprintf_dd(dd, "txg=%llu towrite=%lluK\n", tx->tx_txg,
 500             dd->dd_space_towrite[tx->tx_txg&TXG_MASK] / 1024);
 501         dd->dd_space_towrite[tx->tx_txg&TXG_MASK] = 0;
 502         mutex_exit(&dd->dd_lock);
 503
 504         /* release the hold from dsl_dir_dirty */
 505         dmu_buf_rele(dd->dd_dbuf, dd);
 506 }
 507
 508 static uint64_t
 509 dsl_dir_space_towrite(dsl_dir_t *dd)
 510 {
 511         uint64_t space = 0;
 512         int i;
 513
 514         ASSERT(MUTEX_HELD(&dd->dd_lock));
 515
 516         for (i = 0; i < TXG_SIZE; i++) {
 517                 space += dd->dd_space_towrite[i&TXG_MASK];
 518                 ASSERT3U(dd->dd_space_towrite[i&TXG_MASK], >=, 0);
 519         }
 520         return (space);
 521 }
 522
 523 /*
 524  * How much space would dd have available if ancestor had delta applied
 525  * to it?  If ondiskonly is set, we're only interested in what's
 526  * on-disk, not estimated pending changes.
 527  */
 528 uint64_t
 529 dsl_dir_space_available(dsl_dir_t *dd,
 530     dsl_dir_t *ancestor, int64_t delta, int ondiskonly)
 531 {
 532         uint64_t parentspace, myspace, quota, used;
 533
 534         /*
 535          * If there are no restrictions otherwise, assume we have
 536          * unlimited space available.
 537          */
 538         quota = UINT64_MAX;
 539         parentspace = UINT64_MAX;
 540
 541         if (dd->dd_parent != NULL) {
 542                 parentspace = dsl_dir_space_available(dd->dd_parent,
 543                     ancestor, delta, ondiskonly);
 544         }
 545
 546         mutex_enter(&dd->dd_lock);
 547         if (dd->dd_phys->dd_quota != 0)
 548                 quota = dd->dd_phys->dd_quota;
 549         used = dd->dd_phys->dd_used_bytes;
 550         if (!ondiskonly)
 551                 used += dsl_dir_space_towrite(dd);
 552
 553         if (dd->dd_parent == NULL) {
 554                 uint64_t poolsize = dsl_pool_adjustedsize(dd->dd_pool, FALSE);
 555                 quota = MIN(quota, poolsize);
 556         }
 557
 558         if (dd->dd_phys->dd_reserved > used && parentspace != UINT64_MAX) {
 559                 /*
 560                  * We have some space reserved, in addition to what our
 561                  * parent gave us.
 562                  */
 563                 parentspace += dd->dd_phys->dd_reserved - used;
 564         }
 565
 566         if (dd == ancestor) {
 567                 ASSERT(delta <= 0);
 568                 ASSERT(used >= -delta);
 569                 used += delta;
 570                 if (parentspace != UINT64_MAX)
 571                         parentspace -= delta;
 572         }
 573
 574         if (used > quota) {
 575                 /* over quota */
 576                 myspace = 0;
 577         } else {
 578                 /*
 579                  * the lesser of the space provided by our parent and
 580                  * the space left in our quota
 581                  */
 582                 myspace = MIN(parentspace, quota - used);
 583         }
 584
 585         mutex_exit(&dd->dd_lock);
 586
 587         return (myspace);
 588 }
 589
 590 struct tempreserve {
 591         list_node_t tr_node;
 592         dsl_dir_t *tr_ds;
 593         uint64_t tr_size;
 594 };
 595
 596 static int
 597 dsl_dir_tempreserve_impl(dsl_dir_t *dd, uint64_t asize, boolean_t netfree,
 598     boolean_t ignorequota, boolean_t checkrefquota, list_t *tr_list,
 599     dmu_tx_t *tx, boolean_t first)
 600 {
 601         uint64_t txg = tx->tx_txg;
 602         uint64_t est_inflight, used_on_disk, quota, parent_rsrv;
 603         uint64_t deferred = 0;
 604         struct tempreserve *tr;
 605         int retval = EDQUOT;
 606         int txgidx = txg & TXG_MASK;
 607         int i;
 608         uint64_t ref_rsrv = 0;
 609
 610         ASSERT3U(txg, !=, 0);
 611         ASSERT3S(asize, >, 0);
 612
 613         mutex_enter(&dd->dd_lock);
 614
 615         /*
 616          * Check against the dsl_dir's quota.  We don't add in the delta
 617          * when checking for over-quota because they get one free hit.
 618          */
 619         est_inflight = dsl_dir_space_towrite(dd);
 620         for (i = 0; i < TXG_SIZE; i++)
 621                 est_inflight += dd->dd_tempreserved[i];
 622         used_on_disk = dd->dd_phys->dd_used_bytes;
 623
 624         /*
 625          * On the first iteration, fetch the dataset's used-on-disk and
 626          * refreservation values. Also, if checkrefquota is set, test if
 627          * allocating this space would exceed the dataset's refquota.
 628          */
 629         if (first && tx->tx_objset) {
 630                 int error;
 631                 dsl_dataset_t *ds = tx->tx_objset->os_dsl_dataset;
 632
 633                 error = dsl_dataset_check_quota(ds, checkrefquota,
 634                     asize, est_inflight, &used_on_disk, &ref_rsrv);
 635                 if (error) {
 636                         mutex_exit(&dd->dd_lock);
 637                         return (error);
 638                 }
 639         }
 640
 641         /*
 642          * If this transaction will result in a net free of space,
 643          * we want to let it through.
 644          */
 645         if (ignorequota || netfree || dd->dd_phys->dd_quota == 0)
 646                 quota = UINT64_MAX;
 647         else
 648                 quota = dd->dd_phys->dd_quota;
 649
 650         /*
 651          * Adjust the quota against the actual pool size at the root
 652          * minus any outstanding deferred frees.
 653          * To ensure that it's possible to remove files from a full
 654          * pool without inducing transient overcommits, we throttle
 655          * netfree transactions against a quota that is slightly larger,
 656          * but still within the pool's allocation slop.  In cases where
 657          * we're very close to full, this will allow a steady trickle of
 658          * removes to get through.
 659          */
 660         if (dd->dd_parent == NULL) {
 661                 spa_t *spa = dd->dd_pool->dp_spa;
 662                 uint64_t poolsize = dsl_pool_adjustedsize(dd->dd_pool, netfree);
 663                 deferred = metaslab_class_get_deferred(spa_normal_class(spa));
 664                 if (poolsize - deferred < quota) {
 665                         quota = poolsize - deferred;
 666                         retval = ENOSPC;
 667                 }
 668         }
 669
 670         /*
 671          * If they are requesting more space, and our current estimate
 672          * is over quota, they get to try again unless the actual
 673          * on-disk is over quota and there are no pending changes (which
 674          * may free up space for us).
 675          */
 676         if (used_on_disk + est_inflight >= quota) {
 677                 if (est_inflight > 0 || used_on_disk < quota ||
 678                     (retval == ENOSPC && used_on_disk < quota + deferred))
 679                         retval = ERESTART;
 680                 dprintf_dd(dd, "failing: used=%lluK inflight = %lluK "
 681                     "quota=%lluK tr=%lluK err=%d\n",
 682                     used_on_disk>>10, est_inflight>>10,
 683                     quota>>10, asize>>10, retval);
 684                 mutex_exit(&dd->dd_lock);
 685                 return (SET_ERROR(retval));
 686         }
 687
 688         /* We need to up our estimated delta before dropping dd_lock */
 689         dd->dd_tempreserved[txgidx] += asize;
 690
 691         parent_rsrv = parent_delta(dd, used_on_disk + est_inflight,
 692             asize - ref_rsrv);
 693         mutex_exit(&dd->dd_lock);
 694
 695         tr = kmem_zalloc(sizeof (struct tempreserve), KM_SLEEP);
 696         tr->tr_ds = dd;
 697         tr->tr_size = asize;
 698         list_insert_tail(tr_list, tr);
 699
 700         /* see if it's OK with our parent */
 701         if (dd->dd_parent && parent_rsrv) {
 702                 boolean_t ismos = (dd->dd_phys->dd_head_dataset_obj == 0);
 703
 704                 return (dsl_dir_tempreserve_impl(dd->dd_parent,
 705                     parent_rsrv, netfree, ismos, TRUE, tr_list, tx, FALSE));
 706         } else {
 707                 return (0);
 708         }
 709 }
 710
 711 /*
 712  * Reserve space in this dsl_dir, to be used in this tx's txg.
 713  * After the space has been dirtied (and dsl_dir_willuse_space()
 714  * has been called), the reservation should be canceled, using
 715  * dsl_dir_tempreserve_clear().
 716  */
 717 int
 718 dsl_dir_tempreserve_space(dsl_dir_t *dd, uint64_t lsize, uint64_t asize,
 719     uint64_t fsize, uint64_t usize, void **tr_cookiep, dmu_tx_t *tx)
 720 {
 721         int err;
 722         list_t *tr_list;
 723
 724         if (asize == 0) {
 725                 *tr_cookiep = NULL;
 726                 return (0);
 727         }
 728
 729         tr_list = kmem_alloc(sizeof (list_t), KM_SLEEP);
 730         list_create(tr_list, sizeof (struct tempreserve),
 731             offsetof(struct tempreserve, tr_node));
 732         ASSERT3S(asize, >, 0);
 733         ASSERT3S(fsize, >=, 0);
 734
 735         err = arc_tempreserve_space(lsize, tx->tx_txg);
 736         if (err == 0) {
 737                 struct tempreserve *tr;
 738
 739                 tr = kmem_zalloc(sizeof (struct tempreserve), KM_SLEEP);
 740                 tr->tr_size = lsize;
 741                 list_insert_tail(tr_list, tr);
 742         } else {
 743                 if (err == EAGAIN) {
 744                         /*
 745                          * If arc_memory_throttle() detected that pageout
 746                          * is running and we are low on memory, we delay new
 747                          * non-pageout transactions to give pageout an
 748                          * advantage.
 749                          *
 750                          * It is unfortunate to be delaying while the caller's
 751                          * locks are held.
 752                          */
 753                         txg_delay(dd->dd_pool, tx->tx_txg,
 754                             MSEC2NSEC(10), MSEC2NSEC(10));
 755                         err = SET_ERROR(ERESTART);
 756                 }
 757         }
 758
 759         if (err == 0) {
 760                 err = dsl_dir_tempreserve_impl(dd, asize, fsize >= asize,
 761                     FALSE, asize > usize, tr_list, tx, TRUE);
 762         }
 763
 764         if (err != 0)
 765                 dsl_dir_tempreserve_clear(tr_list, tx);
 766         else
 767                 *tr_cookiep = tr_list;
 768
 769         return (err);
 770 }
 771
 772 /*
 773  * Clear a temporary reservation that we previously made with
 774  * dsl_dir_tempreserve_space().
 775  */
 776 void
 777 dsl_dir_tempreserve_clear(void *tr_cookie, dmu_tx_t *tx)
 778 {
 779         int txgidx = tx->tx_txg & TXG_MASK;
 780         list_t *tr_list = tr_cookie;
 781         struct tempreserve *tr;
 782
 783         ASSERT3U(tx->tx_txg, !=, 0);
 784
 785         if (tr_cookie == NULL)
 786                 return;
 787
 788         while ((tr = list_head(tr_list)) != NULL) {
 789                 if (tr->tr_ds) {
 790                         mutex_enter(&tr->tr_ds->dd_lock);
 791                         ASSERT3U(tr->tr_ds->dd_tempreserved[txgidx], >=,
 792                             tr->tr_size);
 793                         tr->tr_ds->dd_tempreserved[txgidx] -= tr->tr_size;
 794                         mutex_exit(&tr->tr_ds->dd_lock);
 795                 } else {
 796                         arc_tempreserve_clear(tr->tr_size);
 797                 }
 798                 list_remove(tr_list, tr);
 799                 kmem_free(tr, sizeof (struct tempreserve));
 800         }
 801
 802         kmem_free(tr_list, sizeof (list_t));
 803 }
 804
 805 /*
 806  * This should be called from open context when we think we're going to write
 807  * or free space, for example when dirtying data. Be conservative; it's okay
 808  * to write less space or free more, but we don't want to write more or free
 809  * less than the amount specified.
 810  */
 811 void
 812 dsl_dir_willuse_space(dsl_dir_t *dd, int64_t space, dmu_tx_t *tx)
 813 {
 814         int64_t parent_space;
 815         uint64_t est_used;
 816
 817         mutex_enter(&dd->dd_lock);
 818         if (space > 0)
 819                 dd->dd_space_towrite[tx->tx_txg & TXG_MASK] += space;
 820
 821         est_used = dsl_dir_space_towrite(dd) + dd->dd_phys->dd_used_bytes;
 822         parent_space = parent_delta(dd, est_used, space);
 823         mutex_exit(&dd->dd_lock);
 824
 825         /* Make sure that we clean up dd_space_to* */
 826         dsl_dir_dirty(dd, tx);
 827
 828         /* XXX this is potentially expensive and unnecessary... */
 829         if (parent_space && dd->dd_parent)
 830                 dsl_dir_willuse_space(dd->dd_parent, parent_space, tx);
 831 }
 832
 833 /* call from syncing context when we actually write/free space for this dd */
 834 void
 835 dsl_dir_diduse_space(dsl_dir_t *dd, dd_used_t type,
 836     int64_t used, int64_t compressed, int64_t uncompressed, dmu_tx_t *tx)
 837 {
 838         int64_t accounted_delta;
 839         boolean_t needlock = !MUTEX_HELD(&dd->dd_lock);
 840
 841         ASSERT(dmu_tx_is_syncing(tx));
 842         ASSERT(type < DD_USED_NUM);
 843
 844         if (needlock)
 845                 mutex_enter(&dd->dd_lock);
 846         accounted_delta = parent_delta(dd, dd->dd_phys->dd_used_bytes, used);
 847         ASSERT(used >= 0 || dd->dd_phys->dd_used_bytes >= -used);
 848         ASSERT(compressed >= 0 ||
 849             dd->dd_phys->dd_compressed_bytes >= -compressed);
 850         ASSERT(uncompressed >= 0 ||
 851             dd->dd_phys->dd_uncompressed_bytes >= -uncompressed);
 852         dmu_buf_will_dirty(dd->dd_dbuf, tx);
 853         dd->dd_phys->dd_used_bytes += used;
 854         dd->dd_phys->dd_uncompressed_bytes += uncompressed;
 855         dd->dd_phys->dd_compressed_bytes += compressed;
 856
 857         if (dd->dd_phys->dd_flags & DD_FLAG_USED_BREAKDOWN) {
 858                 ASSERT(used > 0 ||
 859                     dd->dd_phys->dd_used_breakdown[type] >= -used);
 860                 dd->dd_phys->dd_used_breakdown[type] += used;
 861 #ifdef DEBUG
 862                 dd_used_t t;
 863                 uint64_t u = 0;
 864                 for (t = 0; t < DD_USED_NUM; t++)
 865                         u += dd->dd_phys->dd_used_breakdown[t];
 866                 ASSERT3U(u, ==, dd->dd_phys->dd_used_bytes);
 867 #endif
 868         }
 869         if (needlock)
 870                 mutex_exit(&dd->dd_lock);
 871
 872         if (dd->dd_parent != NULL) {
 873                 dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD,
 874                     accounted_delta, compressed, uncompressed, tx);
 875                 dsl_dir_transfer_space(dd->dd_parent,
 876                     used - accounted_delta,
 877                     DD_USED_CHILD_RSRV, DD_USED_CHILD, tx);
 878         }
 879 }
 880
 881 void
 882 dsl_dir_transfer_space(dsl_dir_t *dd, int64_t delta,
 883     dd_used_t oldtype, dd_used_t newtype, dmu_tx_t *tx)
 884 {
 885         boolean_t needlock = !MUTEX_HELD(&dd->dd_lock);
 886
 887         ASSERT(dmu_tx_is_syncing(tx));
 888         ASSERT(oldtype < DD_USED_NUM);
 889         ASSERT(newtype < DD_USED_NUM);
 890
 891         if (delta == 0 || !(dd->dd_phys->dd_flags & DD_FLAG_USED_BREAKDOWN))
 892                 return;
 893
 894         if (needlock)
 895                 mutex_enter(&dd->dd_lock);
 896         ASSERT(delta > 0 ?
 897             dd->dd_phys->dd_used_breakdown[oldtype] >= delta :
 898             dd->dd_phys->dd_used_breakdown[newtype] >= -delta);
 899         ASSERT(dd->dd_phys->dd_used_bytes >= ABS(delta));
 900         dmu_buf_will_dirty(dd->dd_dbuf, tx);
 901         dd->dd_phys->dd_used_breakdown[oldtype] -= delta;
 902         dd->dd_phys->dd_used_breakdown[newtype] += delta;
 903         if (needlock)
 904                 mutex_exit(&dd->dd_lock);
 905 }
 906
 907 typedef struct dsl_dir_set_qr_arg {
 908         const char *ddsqra_name;
 909         zprop_source_t ddsqra_source;
 910         uint64_t ddsqra_value;
 911 } dsl_dir_set_qr_arg_t;
 912
 913 static int
 914 dsl_dir_set_quota_check(void *arg, dmu_tx_t *tx)
 915 {
 916         dsl_dir_set_qr_arg_t *ddsqra = arg;
 917         dsl_pool_t *dp = dmu_tx_pool(tx);
 918         dsl_dataset_t *ds;
 919         int error;
 920         uint64_t towrite, newval;
 921
 922         error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds);
 923         if (error != 0)
 924                 return (error);
 925
 926         error = dsl_prop_predict(ds->ds_dir, "quota",
 927             ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval);
 928         if (error != 0) {
 929                 dsl_dataset_rele(ds, FTAG);
 930                 return (error);
 931         }
 932
 933         if (newval == 0) {
 934                 dsl_dataset_rele(ds, FTAG);
 935                 return (0);
 936         }
 937
 938         mutex_enter(&ds->ds_dir->dd_lock);
 939         /*
 940          * If we are doing the preliminary check in open context, and
 941          * there are pending changes, then don't fail it, since the
 942          * pending changes could under-estimate the amount of space to be
 943          * freed up.
 944          */
 945         towrite = dsl_dir_space_towrite(ds->ds_dir);
 946         if ((dmu_tx_is_syncing(tx) || towrite == 0) &&
 947             (newval < ds->ds_dir->dd_phys->dd_reserved ||
 948             newval < ds->ds_dir->dd_phys->dd_used_bytes + towrite)) {
 949                 error = SET_ERROR(ENOSPC);
 950         }
 951         mutex_exit(&ds->ds_dir->dd_lock);
 952         dsl_dataset_rele(ds, FTAG);
 953         return (error);
 954 }
 955
 956 static void
 957 dsl_dir_set_quota_sync(void *arg, dmu_tx_t *tx)
 958 {
 959         dsl_dir_set_qr_arg_t *ddsqra = arg;
 960         dsl_pool_t *dp = dmu_tx_pool(tx);
 961         dsl_dataset_t *ds;
 962         uint64_t newval;
 963
 964         VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds));
 965
 966         if (spa_version(dp->dp_spa) >= SPA_VERSION_RECVD_PROPS) {
 967                 dsl_prop_set_sync_impl(ds, zfs_prop_to_name(ZFS_PROP_QUOTA),
 968                     ddsqra->ddsqra_source, sizeof (ddsqra->ddsqra_value), 1,
 969                     &ddsqra->ddsqra_value, tx);
 970
 971                 VERIFY0(dsl_prop_get_int_ds(ds,
 972                     zfs_prop_to_name(ZFS_PROP_QUOTA), &newval));
 973         } else {
 974                 newval = ddsqra->ddsqra_value;
 975                 spa_history_log_internal_ds(ds, "set", tx, "%s=%lld",
 976                     zfs_prop_to_name(ZFS_PROP_QUOTA), (longlong_t)newval);
 977         }
 978
 979         dmu_buf_will_dirty(ds->ds_dir->dd_dbuf, tx);
 980         mutex_enter(&ds->ds_dir->dd_lock);
 981         ds->ds_dir->dd_phys->dd_quota = newval;
 982         mutex_exit(&ds->ds_dir->dd_lock);
 983         dsl_dataset_rele(ds, FTAG);
 984 }
 985
 986 int
 987 dsl_dir_set_quota(const char *ddname, zprop_source_t source, uint64_t quota)
 988 {
 989         dsl_dir_set_qr_arg_t ddsqra;
 990
 991         ddsqra.ddsqra_name = ddname;
 992         ddsqra.ddsqra_source = source;
 993         ddsqra.ddsqra_value = quota;
 994
 995         return (dsl_sync_task(ddname, dsl_dir_set_quota_check,
 996             dsl_dir_set_quota_sync, &ddsqra, 0));
 997 }
 998
 999 int
1000 dsl_dir_set_reservation_check(void *arg, dmu_tx_t *tx)
1001 {
1002         dsl_dir_set_qr_arg_t *ddsqra = arg;
1003         dsl_pool_t *dp = dmu_tx_pool(tx);
1004         dsl_dataset_t *ds;
1005         dsl_dir_t *dd;
1006         uint64_t newval, used, avail;
1007         int error;
1008
1009         error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds);
1010         if (error != 0)
1011                 return (error);
1012         dd = ds->ds_dir;
1013
1014         /*
1015          * If we are doing the preliminary check in open context, the
1016          * space estimates may be inaccurate.
1017          */
1018         if (!dmu_tx_is_syncing(tx)) {
1019                 dsl_dataset_rele(ds, FTAG);
1020                 return (0);
1021         }
1022
1023         error = dsl_prop_predict(ds->ds_dir,
1024             zfs_prop_to_name(ZFS_PROP_RESERVATION),
1025             ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval);
1026         if (error != 0) {
1027                 dsl_dataset_rele(ds, FTAG);
1028                 return (error);
1029         }
1030
1031         mutex_enter(&dd->dd_lock);
1032         used = dd->dd_phys->dd_used_bytes;
1033         mutex_exit(&dd->dd_lock);
1034
1035         if (dd->dd_parent) {
1036                 avail = dsl_dir_space_available(dd->dd_parent,
1037                     NULL, 0, FALSE);
1038         } else {
1039                 avail = dsl_pool_adjustedsize(dd->dd_pool, B_FALSE) - used;
1040         }
1041
1042         if (MAX(used, newval) > MAX(used, dd->dd_phys->dd_reserved)) {
1043                 uint64_t delta = MAX(used, newval) -
1044                     MAX(used, dd->dd_phys->dd_reserved);
1045
1046                 if (delta > avail ||
1047                     (dd->dd_phys->dd_quota > 0 &&
1048                     newval > dd->dd_phys->dd_quota))
1049                         error = SET_ERROR(ENOSPC);
1050         }
1051
1052         dsl_dataset_rele(ds, FTAG);
1053         return (error);
1054 }
1055
1056 void
1057 dsl_dir_set_reservation_sync_impl(dsl_dir_t *dd, uint64_t value, dmu_tx_t *tx)
1058 {
1059         uint64_t used;
1060         int64_t delta;
1061
1062         dmu_buf_will_dirty(dd->dd_dbuf, tx);
1063
1064         mutex_enter(&dd->dd_lock);
1065         used = dd->dd_phys->dd_used_bytes;
1066         delta = MAX(used, value) - MAX(used, dd->dd_phys->dd_reserved);
1067         dd->dd_phys->dd_reserved = value;
1068
1069         if (dd->dd_parent != NULL) {
1070                 /* Roll up this additional usage into our ancestors */
1071                 dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD_RSRV,
1072                     delta, 0, 0, tx);
1073         }
1074         mutex_exit(&dd->dd_lock);
1075 }
1076
1077 static void
1078 dsl_dir_set_reservation_sync(void *arg, dmu_tx_t *tx)
1079 {
1080         dsl_dir_set_qr_arg_t *ddsqra = arg;
1081         dsl_pool_t *dp = dmu_tx_pool(tx);
1082         dsl_dataset_t *ds;
1083         uint64_t newval;
1084
1085         VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds));
1086
1087         if (spa_version(dp->dp_spa) >= SPA_VERSION_RECVD_PROPS) {
1088                 dsl_prop_set_sync_impl(ds,
1089                     zfs_prop_to_name(ZFS_PROP_RESERVATION),
1090                     ddsqra->ddsqra_source, sizeof (ddsqra->ddsqra_value), 1,
1091                     &ddsqra->ddsqra_value, tx);
1092
1093                 VERIFY0(dsl_prop_get_int_ds(ds,
1094                     zfs_prop_to_name(ZFS_PROP_RESERVATION), &newval));
1095         } else {
1096                 newval = ddsqra->ddsqra_value;
1097                 spa_history_log_internal_ds(ds, "set", tx, "%s=%lld",
1098                     zfs_prop_to_name(ZFS_PROP_RESERVATION),
1099                     (longlong_t)newval);
1100         }
1101
1102         dsl_dir_set_reservation_sync_impl(ds->ds_dir, newval, tx);
1103         dsl_dataset_rele(ds, FTAG);
1104 }
1105
1106 int
1107 dsl_dir_set_reservation(const char *ddname, zprop_source_t source,
1108     uint64_t reservation)
1109 {
1110         dsl_dir_set_qr_arg_t ddsqra;
1111
1112         ddsqra.ddsqra_name = ddname;
1113         ddsqra.ddsqra_source = source;
1114         ddsqra.ddsqra_value = reservation;
1115
1116         return (dsl_sync_task(ddname, dsl_dir_set_reservation_check,
1117             dsl_dir_set_reservation_sync, &ddsqra, 0));
1118 }
1119
1120 static dsl_dir_t *
1121 closest_common_ancestor(dsl_dir_t *ds1, dsl_dir_t *ds2)
1122 {
1123         for (; ds1; ds1 = ds1->dd_parent) {
1124                 dsl_dir_t *dd;
1125                 for (dd = ds2; dd; dd = dd->dd_parent) {
1126                         if (ds1 == dd)
1127                                 return (dd);
1128                 }
1129         }
1130         return (NULL);
1131 }
1132
1133 /*
1134  * If delta is applied to dd, how much of that delta would be applied to
1135  * ancestor?  Syncing context only.
1136  */
1137 static int64_t
1138 would_change(dsl_dir_t *dd, int64_t delta, dsl_dir_t *ancestor)
1139 {
1140         if (dd == ancestor)
1141                 return (delta);
1142
1143         mutex_enter(&dd->dd_lock);
1144         delta = parent_delta(dd, dd->dd_phys->dd_used_bytes, delta);
1145         mutex_exit(&dd->dd_lock);
1146         return (would_change(dd->dd_parent, delta, ancestor));
1147 }
1148
1149 typedef struct dsl_dir_rename_arg {
1150         const char *ddra_oldname;
1151         const char *ddra_newname;
1152 } dsl_dir_rename_arg_t;
1153
1154 /* ARGSUSED */
1155 static int
1156 dsl_valid_rename(dsl_pool_t *dp, dsl_dataset_t *ds, void *arg)
1157 {
1158         int *deltap = arg;
1159         char namebuf[MAXNAMELEN];
1160
1161         dsl_dataset_name(ds, namebuf);
1162
1163         if (strlen(namebuf) + *deltap >= MAXNAMELEN)
1164                 return (SET_ERROR(ENAMETOOLONG));
1165         return (0);
1166 }
1167
1168 static int
1169 dsl_dir_rename_check(void *arg, dmu_tx_t *tx)
1170 {
1171         dsl_dir_rename_arg_t *ddra = arg;
1172         dsl_pool_t *dp = dmu_tx_pool(tx);
1173         dsl_dir_t *dd, *newparent;
1174         const char *mynewname;
1175         int error;
1176         int delta = strlen(ddra->ddra_newname) - strlen(ddra->ddra_oldname);
1177
1178         /* target dir should exist */
1179         error = dsl_dir_hold(dp, ddra->ddra_oldname, FTAG, &dd, NULL);
1180         if (error != 0)
1181                 return (error);
1182
1183         /* new parent should exist */
1184         error = dsl_dir_hold(dp, ddra->ddra_newname, FTAG,
1185             &newparent, &mynewname);
1186         if (error != 0) {
1187                 dsl_dir_rele(dd, FTAG);
1188                 return (error);
1189         }
1190
1191         /* can't rename to different pool */
1192         if (dd->dd_pool != newparent->dd_pool) {
1193                 dsl_dir_rele(newparent, FTAG);
1194                 dsl_dir_rele(dd, FTAG);
1195                 return (SET_ERROR(ENXIO));
1196         }
1197
1198         /* new name should not already exist */
1199         if (mynewname == NULL) {
1200                 dsl_dir_rele(newparent, FTAG);
1201                 dsl_dir_rele(dd, FTAG);
1202                 return (SET_ERROR(EEXIST));
1203         }
1204
1205         /* if the name length is growing, validate child name lengths */
1206         if (delta > 0) {
1207                 error = dmu_objset_find_dp(dp, dd->dd_object, dsl_valid_rename,
1208                     &delta, DS_FIND_CHILDREN | DS_FIND_SNAPSHOTS);
1209                 if (error != 0) {
1210                         dsl_dir_rele(newparent, FTAG);
1211                         dsl_dir_rele(dd, FTAG);
1212                         return (error);
1213                 }
1214         }
1215
1216         if (newparent != dd->dd_parent) {
1217                 /* is there enough space? */
1218                 uint64_t myspace =
1219                     MAX(dd->dd_phys->dd_used_bytes, dd->dd_phys->dd_reserved);
1220
1221                 /* no rename into our descendant */
1222                 if (closest_common_ancestor(dd, newparent) == dd) {
1223                         dsl_dir_rele(newparent, FTAG);
1224                         dsl_dir_rele(dd, FTAG);
1225                         return (SET_ERROR(EINVAL));
1226                 }
1227
1228                 error = dsl_dir_transfer_possible(dd->dd_parent,
1229                     newparent, myspace);
1230                 if (error != 0) {
1231                         dsl_dir_rele(newparent, FTAG);
1232                         dsl_dir_rele(dd, FTAG);
1233                         return (error);
1234                 }
1235         }
1236
1237         dsl_dir_rele(newparent, FTAG);
1238         dsl_dir_rele(dd, FTAG);
1239         return (0);
1240 }
1241
1242 static void
1243 dsl_dir_rename_sync(void *arg, dmu_tx_t *tx)
1244 {
1245         dsl_dir_rename_arg_t *ddra = arg;
1246         dsl_pool_t *dp = dmu_tx_pool(tx);
1247         dsl_dir_t *dd, *newparent;
1248         const char *mynewname;
1249         int error;
1250         objset_t *mos = dp->dp_meta_objset;
1251
1252         VERIFY0(dsl_dir_hold(dp, ddra->ddra_oldname, FTAG, &dd, NULL));
1253         VERIFY0(dsl_dir_hold(dp, ddra->ddra_newname, FTAG, &newparent,
1254             &mynewname));
1255
1256         /* Log this before we change the name. */
1257         spa_history_log_internal_dd(dd, "rename", tx,
1258             "-> %s", ddra->ddra_newname);
1259
1260         if (newparent != dd->dd_parent) {
1261                 dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD,
1262                     -dd->dd_phys->dd_used_bytes,
1263                     -dd->dd_phys->dd_compressed_bytes,
1264                     -dd->dd_phys->dd_uncompressed_bytes, tx);
1265                 dsl_dir_diduse_space(newparent, DD_USED_CHILD,
1266                     dd->dd_phys->dd_used_bytes,
1267                     dd->dd_phys->dd_compressed_bytes,
1268                     dd->dd_phys->dd_uncompressed_bytes, tx);
1269
1270                 if (dd->dd_phys->dd_reserved > dd->dd_phys->dd_used_bytes) {
1271                         uint64_t unused_rsrv = dd->dd_phys->dd_reserved -
1272                             dd->dd_phys->dd_used_bytes;
1273
1274                         dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD_RSRV,
1275                             -unused_rsrv, 0, 0, tx);
1276                         dsl_dir_diduse_space(newparent, DD_USED_CHILD_RSRV,
1277                             unused_rsrv, 0, 0, tx);
1278                 }
1279         }
1280
1281         dmu_buf_will_dirty(dd->dd_dbuf, tx);
1282
1283         /* remove from old parent zapobj */
1284         error = zap_remove(mos, dd->dd_parent->dd_phys->dd_child_dir_zapobj,
1285             dd->dd_myname, tx);
1286         ASSERT0(error);
1287
1288         (void) strcpy(dd->dd_myname, mynewname);
1289         dsl_dir_rele(dd->dd_parent, dd);
1290         dd->dd_phys->dd_parent_obj = newparent->dd_object;
1291         VERIFY0(dsl_dir_hold_obj(dp,
1292             newparent->dd_object, NULL, dd, &dd->dd_parent));
1293
1294         /* add to new parent zapobj */
1295         VERIFY0(zap_add(mos, newparent->dd_phys->dd_child_dir_zapobj,
1296             dd->dd_myname, 8, 1, &dd->dd_object, tx));
1297
1298 #ifdef __FreeBSD__
1299 #ifdef _KERNEL
1300         zfsvfs_update_fromname(ddra->ddra_oldname, ddra->ddra_newname);
1301         zvol_rename_minors(ddra->ddra_oldname, ddra->ddra_newname);
1302 #endif
1303 #endif
1304
1305         dsl_prop_notify_all(dd);
1306
1307         dsl_dir_rele(newparent, FTAG);
1308         dsl_dir_rele(dd, FTAG);
1309 }
1310
1311 int
1312 dsl_dir_rename(const char *oldname, const char *newname)
1313 {
1314         dsl_dir_rename_arg_t ddra;
1315
1316         ddra.ddra_oldname = oldname;
1317         ddra.ddra_newname = newname;
1318
1319         return (dsl_sync_task(oldname,
1320             dsl_dir_rename_check, dsl_dir_rename_sync, &ddra, 3));
1321 }
1322
1323 int
1324 dsl_dir_transfer_possible(dsl_dir_t *sdd, dsl_dir_t *tdd, uint64_t space)
1325 {
1326         dsl_dir_t *ancestor;
1327         int64_t adelta;
1328         uint64_t avail;
1329
1330         ancestor = closest_common_ancestor(sdd, tdd);
1331         adelta = would_change(sdd, -space, ancestor);
1332         avail = dsl_dir_space_available(tdd, ancestor, adelta, FALSE);
1333         if (avail < space)
1334                 return (SET_ERROR(ENOSPC));
1335
1336         return (0);
1337 }
1338
1339 timestruc_t
1340 dsl_dir_snap_cmtime(dsl_dir_t *dd)
1341 {
1342         timestruc_t t;
1343
1344         mutex_enter(&dd->dd_lock);
1345         t = dd->dd_snap_cmtime;
1346         mutex_exit(&dd->dd_lock);
1347
1348         return (t);
1349 }
1350
1351 void
1352 dsl_dir_snap_cmtime_update(dsl_dir_t *dd)
1353 {
1354         timestruc_t t;
1355
1356         gethrestime(&t);
1357         mutex_enter(&dd->dd_lock);
1358         dd->dd_snap_cmtime = t;
1359         mutex_exit(&dd->dd_lock);
1360 }