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/dmu_impl.h>
  37 #include <sys/spa.h>
  38 #include <sys/metaslab.h>
  39 #include <sys/zap.h>
  40 #include <sys/zio.h>
  41 #include <sys/arc.h>
  42 #include <sys/sunddi.h>
  43 #include <sys/zvol.h>
  44 #ifdef _KERNEL
  45 #include <sys/zfs_vfsops.h>
  46 #endif
  47 #include "zfs_namecheck.h"
  48
  49 static uint64_t dsl_dir_space_towrite(dsl_dir_t *dd);
  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_rele(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_hold_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(dsl_pool_config_held(dp));
  88
  89         err = dmu_bonus_hold(dp->dp_meta_objset, ddobj, tag, &dbuf);
  90         if (err != 0)
  91                 return (err);
  92         dd = dmu_buf_get_user(dbuf);
  93 #ifdef ZFS_DEBUG
  94         {
  95                 dmu_object_info_t doi;
  96                 dmu_object_info_from_db(dbuf, &doi);
  97                 ASSERT3U(doi.doi_bonus_type, ==, DMU_OT_DSL_DIR);
  98                 ASSERT3U(doi.doi_bonus_size, >=, sizeof (dsl_dir_phys_t));
  99         }
 100 #endif
 101         if (dd == NULL) {
 102                 dsl_dir_t *winner;
 103
 104                 dd = kmem_zalloc(sizeof (dsl_dir_t), KM_SLEEP);
 105                 dd->dd_object = ddobj;
 106                 dd->dd_dbuf = dbuf;
 107                 dd->dd_pool = dp;
 108                 dd->dd_phys = dbuf->db_data;
 109                 mutex_init(&dd->dd_lock, NULL, MUTEX_DEFAULT, NULL);
 110
 111                 list_create(&dd->dd_prop_cbs, sizeof (dsl_prop_cb_record_t),
 112                     offsetof(dsl_prop_cb_record_t, cbr_node));
 113
 114                 dsl_dir_snap_cmtime_update(dd);
 115
 116                 if (dd->dd_phys->dd_parent_obj) {
 117                         err = dsl_dir_hold_obj(dp, dd->dd_phys->dd_parent_obj,
 118                             NULL, dd, &dd->dd_parent);
 119                         if (err != 0)
 120                                 goto errout;
 121                         if (tail) {
 122 #ifdef ZFS_DEBUG
 123                                 uint64_t foundobj;
 124
 125                                 err = zap_lookup(dp->dp_meta_objset,
 126                                     dd->dd_parent->dd_phys->dd_child_dir_zapobj,
 127                                     tail, sizeof (foundobj), 1, &foundobj);
 128                                 ASSERT(err || foundobj == ddobj);
 129 #endif
 130                                 (void) strcpy(dd->dd_myname, tail);
 131                         } else {
 132                                 err = zap_value_search(dp->dp_meta_objset,
 133                                     dd->dd_parent->dd_phys->dd_child_dir_zapobj,
 134                                     ddobj, 0, dd->dd_myname);
 135                         }
 136                         if (err != 0)
 137                                 goto errout;
 138                 } else {
 139                         (void) strcpy(dd->dd_myname, spa_name(dp->dp_spa));
 140                 }
 141
 142                 if (dsl_dir_is_clone(dd)) {
 143                         dmu_buf_t *origin_bonus;
 144                         dsl_dataset_phys_t *origin_phys;
 145
 146                         /*
 147                          * We can't open the origin dataset, because
 148                          * that would require opening this dsl_dir.
 149                          * Just look at its phys directly instead.
 150                          */
 151                         err = dmu_bonus_hold(dp->dp_meta_objset,
 152                             dd->dd_phys->dd_origin_obj, FTAG, &origin_bonus);
 153                         if (err != 0)
 154                                 goto errout;
 155                         origin_phys = origin_bonus->db_data;
 156                         dd->dd_origin_txg =
 157                             origin_phys->ds_creation_txg;
 158                         dmu_buf_rele(origin_bonus, FTAG);
 159                 }
 160
 161                 winner = dmu_buf_set_user_ie(dbuf, dd, &dd->dd_phys,
 162                     dsl_dir_evict);
 163                 if (winner) {
 164                         if (dd->dd_parent)
 165                                 dsl_dir_rele(dd->dd_parent, dd);
 166                         mutex_destroy(&dd->dd_lock);
 167                         kmem_free(dd, sizeof (dsl_dir_t));
 168                         dd = winner;
 169                 } else {
 170                         spa_open_ref(dp->dp_spa, dd);
 171                 }
 172         }
 173
 174         /*
 175          * The dsl_dir_t has both open-to-close and instantiate-to-evict
 176          * holds on the spa.  We need the open-to-close holds because
 177          * otherwise the spa_refcnt wouldn't change when we open a
 178          * dir which the spa also has open, so we could incorrectly
 179          * think it was OK to unload/export/destroy the pool.  We need
 180          * the instantiate-to-evict hold because the dsl_dir_t has a
 181          * pointer to the dd_pool, which has a pointer to the spa_t.
 182          */
 183         spa_open_ref(dp->dp_spa, tag);
 184         ASSERT3P(dd->dd_pool, ==, dp);
 185         ASSERT3U(dd->dd_object, ==, ddobj);
 186         ASSERT3P(dd->dd_dbuf, ==, dbuf);
 187         *ddp = dd;
 188         return (0);
 189
 190 errout:
 191         if (dd->dd_parent)
 192                 dsl_dir_rele(dd->dd_parent, dd);
 193         mutex_destroy(&dd->dd_lock);
 194         kmem_free(dd, sizeof (dsl_dir_t));
 195         dmu_buf_rele(dbuf, tag);
 196         return (err);
 197 }
 198
 199 void
 200 dsl_dir_rele(dsl_dir_t *dd, void *tag)
 201 {
 202         dprintf_dd(dd, "%s\n", "");
 203         spa_close(dd->dd_pool->dp_spa, tag);
 204         dmu_buf_rele(dd->dd_dbuf, tag);
 205 }
 206
 207 /* buf must be long enough (MAXNAMELEN + strlen(MOS_DIR_NAME) + 1 should do) */
 208 void
 209 dsl_dir_name(dsl_dir_t *dd, char *buf)
 210 {
 211         if (dd->dd_parent) {
 212                 dsl_dir_name(dd->dd_parent, buf);
 213                 (void) strcat(buf, "/");
 214         } else {
 215                 buf[0] = '\0';
 216         }
 217         if (!MUTEX_HELD(&dd->dd_lock)) {
 218                 /*
 219                  * recursive mutex so that we can use
 220                  * dprintf_dd() with dd_lock held
 221                  */
 222                 mutex_enter(&dd->dd_lock);
 223                 (void) strcat(buf, dd->dd_myname);
 224                 mutex_exit(&dd->dd_lock);
 225         } else {
 226                 (void) strcat(buf, dd->dd_myname);
 227         }
 228 }
 229
 230 /* Calculate name length, avoiding all the strcat calls of dsl_dir_name */
 231 int
 232 dsl_dir_namelen(dsl_dir_t *dd)
 233 {
 234         int result = 0;
 235
 236         if (dd->dd_parent) {
 237                 /* parent's name + 1 for the "/" */
 238                 result = dsl_dir_namelen(dd->dd_parent) + 1;
 239         }
 240
 241         if (!MUTEX_HELD(&dd->dd_lock)) {
 242                 /* see dsl_dir_name */
 243                 mutex_enter(&dd->dd_lock);
 244                 result += strlen(dd->dd_myname);
 245                 mutex_exit(&dd->dd_lock);
 246         } else {
 247                 result += strlen(dd->dd_myname);
 248         }
 249
 250         return (result);
 251 }
 252
 253 static int
 254 getcomponent(const char *path, char *component, const char **nextp)
 255 {
 256         char *p;
 257
 258         if ((path == NULL) || (path[0] == '\0'))
 259                 return (SET_ERROR(ENOENT));
 260         /* This would be a good place to reserve some namespace... */
 261         p = strpbrk(path, "/@");
 262         if (p && (p[1] == '/' || p[1] == '@')) {
 263                 /* two separators in a row */
 264                 return (SET_ERROR(EINVAL));
 265         }
 266         if (p == NULL || p == path) {
 267                 /*
 268                  * if the first thing is an @ or /, it had better be an
 269                  * @ and it had better not have any more ats or slashes,
 270                  * and it had better have something after the @.
 271                  */
 272                 if (p != NULL &&
 273                     (p[0] != '@' || strpbrk(path+1, "/@") || p[1] == '\0'))
 274                         return (SET_ERROR(EINVAL));
 275                 if (strlen(path) >= MAXNAMELEN)
 276                         return (SET_ERROR(ENAMETOOLONG));
 277                 (void) strcpy(component, path);
 278                 p = NULL;
 279         } else if (p[0] == '/') {
 280                 if (p - path >= MAXNAMELEN)
 281                         return (SET_ERROR(ENAMETOOLONG));
 282                 (void) strncpy(component, path, p - path);
 283                 component[p - path] = '\0';
 284                 p++;
 285         } else if (p[0] == '@') {
 286                 /*
 287                  * if the next separator is an @, there better not be
 288                  * any more slashes.
 289                  */
 290                 if (strchr(path, '/'))
 291                         return (SET_ERROR(EINVAL));
 292                 if (p - path >= MAXNAMELEN)
 293                         return (SET_ERROR(ENAMETOOLONG));
 294                 (void) strncpy(component, path, p - path);
 295                 component[p - path] = '\0';
 296         } else {
 297                 panic("invalid p=%p", (void *)p);
 298         }
 299         *nextp = p;
 300         return (0);
 301 }
 302
 303 /*
 304  * Return the dsl_dir_t, and possibly the last component which couldn't
 305  * be found in *tail.  The name must be in the specified dsl_pool_t.  This
 306  * thread must hold the dp_config_rwlock for the pool.  Returns NULL if the
 307  * path is bogus, or if tail==NULL and we couldn't parse the whole name.
 308  * (*tail)[0] == '@' means that the last component is a snapshot.
 309  */
 310 int
 311 dsl_dir_hold(dsl_pool_t *dp, const char *name, void *tag,
 312     dsl_dir_t **ddp, const char **tailp)
 313 {
 314         char buf[MAXNAMELEN];
 315         const char *spaname, *next, *nextnext = NULL;
 316         int err;
 317         dsl_dir_t *dd;
 318         uint64_t ddobj;
 319
 320         err = getcomponent(name, buf, &next);
 321         if (err != 0)
 322                 return (err);
 323
 324         /* Make sure the name is in the specified pool. */
 325         spaname = spa_name(dp->dp_spa);
 326         if (strcmp(buf, spaname) != 0)
 327                 return (SET_ERROR(EINVAL));
 328
 329         ASSERT(dsl_pool_config_held(dp));
 330
 331         err = dsl_dir_hold_obj(dp, dp->dp_root_dir_obj, NULL, tag, &dd);
 332         if (err != 0) {
 333                 return (err);
 334         }
 335
 336         while (next != NULL) {
 337                 dsl_dir_t *child_ds;
 338                 err = getcomponent(next, buf, &nextnext);
 339                 if (err != 0)
 340                         break;
 341                 ASSERT(next[0] != '\0');
 342                 if (next[0] == '@')
 343                         break;
 344                 dprintf("looking up %s in obj%lld\n",
 345                     buf, dd->dd_phys->dd_child_dir_zapobj);
 346
 347                 err = zap_lookup(dp->dp_meta_objset,
 348                     dd->dd_phys->dd_child_dir_zapobj,
 349                     buf, sizeof (ddobj), 1, &ddobj);
 350                 if (err != 0) {
 351                         if (err == ENOENT)
 352                                 err = 0;
 353                         break;
 354                 }
 355
 356                 err = dsl_dir_hold_obj(dp, ddobj, buf, tag, &child_ds);
 357                 if (err != 0)
 358                         break;
 359                 dsl_dir_rele(dd, tag);
 360                 dd = child_ds;
 361                 next = nextnext;
 362         }
 363
 364         if (err != 0) {
 365                 dsl_dir_rele(dd, tag);
 366                 return (err);
 367         }
 368
 369         /*
 370          * It's an error if there's more than one component left, or
 371          * tailp==NULL and there's any component left.
 372          */
 373         if (next != NULL &&
 374             (tailp == NULL || (nextnext && nextnext[0] != '\0'))) {
 375                 /* bad path name */
 376                 dsl_dir_rele(dd, tag);
 377                 dprintf("next=%p (%s) tail=%p\n", next, next?next:"", tailp);
 378                 err = SET_ERROR(ENOENT);
 379         }
 380         if (tailp != NULL)
 381                 *tailp = next;
 382         *ddp = dd;
 383         return (err);
 384 }
 385
 386 uint64_t
 387 dsl_dir_create_sync(dsl_pool_t *dp, dsl_dir_t *pds, const char *name,
 388     dmu_tx_t *tx)
 389 {
 390         objset_t *mos = dp->dp_meta_objset;
 391         uint64_t ddobj;
 392         dsl_dir_phys_t *ddphys;
 393         dmu_buf_t *dbuf;
 394
 395         ddobj = dmu_object_alloc(mos, DMU_OT_DSL_DIR, 0,
 396             DMU_OT_DSL_DIR, sizeof (dsl_dir_phys_t), tx);
 397         if (pds) {
 398                 VERIFY(0 == zap_add(mos, pds->dd_phys->dd_child_dir_zapobj,
 399                     name, sizeof (uint64_t), 1, &ddobj, tx));
 400         } else {
 401                 /* it's the root dir */
 402                 VERIFY(0 == zap_add(mos, DMU_POOL_DIRECTORY_OBJECT,
 403                     DMU_POOL_ROOT_DATASET, sizeof (uint64_t), 1, &ddobj, tx));
 404         }
 405         VERIFY(0 == dmu_bonus_hold(mos, ddobj, FTAG, &dbuf));
 406         dmu_buf_will_dirty(dbuf, tx);
 407         ddphys = dbuf->db_data;
 408
 409         ddphys->dd_creation_time = gethrestime_sec();
 410         if (pds)
 411                 ddphys->dd_parent_obj = pds->dd_object;
 412         ddphys->dd_props_zapobj = zap_create(mos,
 413             DMU_OT_DSL_PROPS, DMU_OT_NONE, 0, tx);
 414         ddphys->dd_child_dir_zapobj = zap_create(mos,
 415             DMU_OT_DSL_DIR_CHILD_MAP, DMU_OT_NONE, 0, tx);
 416         if (spa_version(dp->dp_spa) >= SPA_VERSION_USED_BREAKDOWN)
 417                 ddphys->dd_flags |= DD_FLAG_USED_BREAKDOWN;
 418         dmu_buf_rele(dbuf, FTAG);
 419
 420         return (ddobj);
 421 }
 422
 423 boolean_t
 424 dsl_dir_is_clone(dsl_dir_t *dd)
 425 {
 426         return (dd->dd_phys->dd_origin_obj &&
 427             (dd->dd_pool->dp_origin_snap == NULL ||
 428             dd->dd_phys->dd_origin_obj !=
 429             dd->dd_pool->dp_origin_snap->ds_object));
 430 }
 431
 432 void
 433 dsl_dir_stats(dsl_dir_t *dd, nvlist_t *nv)
 434 {
 435         mutex_enter(&dd->dd_lock);
 436         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USED,
 437             dd->dd_phys->dd_used_bytes);
 438         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_QUOTA, dd->dd_phys->dd_quota);
 439         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_RESERVATION,
 440             dd->dd_phys->dd_reserved);
 441         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_COMPRESSRATIO,
 442             dd->dd_phys->dd_compressed_bytes == 0 ? 100 :
 443             (dd->dd_phys->dd_uncompressed_bytes * 100 /
 444             dd->dd_phys->dd_compressed_bytes));
 445         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_LOGICALUSED,
 446             dd->dd_phys->dd_uncompressed_bytes);
 447         if (dd->dd_phys->dd_flags & DD_FLAG_USED_BREAKDOWN) {
 448                 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDSNAP,
 449                     dd->dd_phys->dd_used_breakdown[DD_USED_SNAP]);
 450                 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDDS,
 451                     dd->dd_phys->dd_used_breakdown[DD_USED_HEAD]);
 452                 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDREFRESERV,
 453                     dd->dd_phys->dd_used_breakdown[DD_USED_REFRSRV]);
 454                 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDCHILD,
 455                     dd->dd_phys->dd_used_breakdown[DD_USED_CHILD] +
 456                     dd->dd_phys->dd_used_breakdown[DD_USED_CHILD_RSRV]);
 457         }
 458         mutex_exit(&dd->dd_lock);
 459
 460         if (dsl_dir_is_clone(dd)) {
 461                 dsl_dataset_t *ds;
 462                 char buf[MAXNAMELEN];
 463
 464                 VERIFY0(dsl_dataset_hold_obj(dd->dd_pool,
 465                     dd->dd_phys->dd_origin_obj, FTAG, &ds));
 466                 dsl_dataset_name(ds, buf);
 467                 dsl_dataset_rele(ds, FTAG);
 468                 dsl_prop_nvlist_add_string(nv, ZFS_PROP_ORIGIN, buf);
 469         }
 470 }
 471
 472 void
 473 dsl_dir_dirty(dsl_dir_t *dd, dmu_tx_t *tx)
 474 {
 475         dsl_pool_t *dp = dd->dd_pool;
 476
 477         ASSERT(dd->dd_phys);
 478
 479         if (txg_list_add(&dp->dp_dirty_dirs, dd, tx->tx_txg)) {
 480                 /* up the hold count until we can be written out */
 481                 dmu_buf_add_ref(dd->dd_dbuf, dd);
 482         }
 483 }
 484
 485 static int64_t
 486 parent_delta(dsl_dir_t *dd, uint64_t used, int64_t delta)
 487 {
 488         uint64_t old_accounted = MAX(used, dd->dd_phys->dd_reserved);
 489         uint64_t new_accounted = MAX(used + delta, dd->dd_phys->dd_reserved);
 490         return (new_accounted - old_accounted);
 491 }
 492
 493 void
 494 dsl_dir_sync(dsl_dir_t *dd, dmu_tx_t *tx)
 495 {
 496         ASSERT(dmu_tx_is_syncing(tx));
 497
 498         mutex_enter(&dd->dd_lock);
 499         ASSERT0(dd->dd_tempreserved[tx->tx_txg&TXG_MASK]);
 500         dprintf_dd(dd, "txg=%llu towrite=%lluK\n", tx->tx_txg,
 501             dd->dd_space_towrite[tx->tx_txg&TXG_MASK] / 1024);
 502         dd->dd_space_towrite[tx->tx_txg&TXG_MASK] = 0;
 503         mutex_exit(&dd->dd_lock);
 504
 505         /* release the hold from dsl_dir_dirty */
 506         dmu_buf_rele(dd->dd_dbuf, dd);
 507 }
 508
 509 static uint64_t
 510 dsl_dir_space_towrite(dsl_dir_t *dd)
 511 {
 512         uint64_t space = 0;
 513         int i;
 514
 515         ASSERT(MUTEX_HELD(&dd->dd_lock));
 516
 517         for (i = 0; i < TXG_SIZE; i++) {
 518                 space += dd->dd_space_towrite[i&TXG_MASK];
 519                 ASSERT3U(dd->dd_space_towrite[i&TXG_MASK], >=, 0);
 520         }
 521         return (space);
 522 }
 523
 524 /*
 525  * How much space would dd have available if ancestor had delta applied
 526  * to it?  If ondiskonly is set, we're only interested in what's
 527  * on-disk, not estimated pending changes.
 528  */
 529 uint64_t
 530 dsl_dir_space_available(dsl_dir_t *dd,
 531     dsl_dir_t *ancestor, int64_t delta, int ondiskonly)
 532 {
 533         uint64_t parentspace, myspace, quota, used;
 534
 535         /*
 536          * If there are no restrictions otherwise, assume we have
 537          * unlimited space available.
 538          */
 539         quota = UINT64_MAX;
 540         parentspace = UINT64_MAX;
 541
 542         if (dd->dd_parent != NULL) {
 543                 parentspace = dsl_dir_space_available(dd->dd_parent,
 544                     ancestor, delta, ondiskonly);
 545         }
 546
 547         mutex_enter(&dd->dd_lock);
 548         if (dd->dd_phys->dd_quota != 0)
 549                 quota = dd->dd_phys->dd_quota;
 550         used = dd->dd_phys->dd_used_bytes;
 551         if (!ondiskonly)
 552                 used += dsl_dir_space_towrite(dd);
 553
 554         if (dd->dd_parent == NULL) {
 555                 uint64_t poolsize = dsl_pool_adjustedsize(dd->dd_pool, FALSE);
 556                 quota = MIN(quota, poolsize);
 557         }
 558
 559         if (dd->dd_phys->dd_reserved > used && parentspace != UINT64_MAX) {
 560                 /*
 561                  * We have some space reserved, in addition to what our
 562                  * parent gave us.
 563                  */
 564                 parentspace += dd->dd_phys->dd_reserved - used;
 565         }
 566
 567         if (dd == ancestor) {
 568                 ASSERT(delta <= 0);
 569                 ASSERT(used >= -delta);
 570                 used += delta;
 571                 if (parentspace != UINT64_MAX)
 572                         parentspace -= delta;
 573         }
 574
 575         if (used > quota) {
 576                 /* over quota */
 577                 myspace = 0;
 578         } else {
 579                 /*
 580                  * the lesser of the space provided by our parent and
 581                  * the space left in our quota
 582                  */
 583                 myspace = MIN(parentspace, quota - used);
 584         }
 585
 586         mutex_exit(&dd->dd_lock);
 587
 588         return (myspace);
 589 }
 590
 591 struct tempreserve {
 592         list_node_t tr_node;
 593         dsl_dir_t *tr_ds;
 594         uint64_t tr_size;
 595 };
 596
 597 static int
 598 dsl_dir_tempreserve_impl(dsl_dir_t *dd, uint64_t asize, boolean_t netfree,
 599     boolean_t ignorequota, boolean_t checkrefquota, list_t *tr_list,
 600     dmu_tx_t *tx, boolean_t first)
 601 {
 602         uint64_t txg = tx->tx_txg;
 603         uint64_t est_inflight, used_on_disk, quota, parent_rsrv;
 604         uint64_t deferred = 0;
 605         struct tempreserve *tr;
 606         int retval = EDQUOT;
 607         int txgidx = txg & TXG_MASK;
 608         int i;
 609         uint64_t ref_rsrv = 0;
 610
 611         ASSERT3U(txg, !=, 0);
 612         ASSERT3S(asize, >, 0);
 613
 614         mutex_enter(&dd->dd_lock);
 615
 616         /*
 617          * Check against the dsl_dir's quota.  We don't add in the delta
 618          * when checking for over-quota because they get one free hit.
 619          */
 620         est_inflight = dsl_dir_space_towrite(dd);
 621         for (i = 0; i < TXG_SIZE; i++)
 622                 est_inflight += dd->dd_tempreserved[i];
 623         used_on_disk = dd->dd_phys->dd_used_bytes;
 624
 625         /*
 626          * On the first iteration, fetch the dataset's used-on-disk and
 627          * refreservation values. Also, if checkrefquota is set, test if
 628          * allocating this space would exceed the dataset's refquota.
 629          */
 630         if (first && tx->tx_objset) {
 631                 int error;
 632                 dsl_dataset_t *ds = tx->tx_objset->os_dsl_dataset;
 633
 634                 error = dsl_dataset_check_quota(ds, checkrefquota,
 635                     asize, est_inflight, &used_on_disk, &ref_rsrv);
 636                 if (error) {
 637                         mutex_exit(&dd->dd_lock);
 638                         return (error);
 639                 }
 640         }
 641
 642         /*
 643          * If this transaction will result in a net free of space,
 644          * we want to let it through.
 645          */
 646         if (ignorequota || netfree || dd->dd_phys->dd_quota == 0)
 647                 quota = UINT64_MAX;
 648         else
 649                 quota = dd->dd_phys->dd_quota;
 650
 651         /*
 652          * Adjust the quota against the actual pool size at the root
 653          * minus any outstanding deferred frees.
 654          * To ensure that it's possible to remove files from a full
 655          * pool without inducing transient overcommits, we throttle
 656          * netfree transactions against a quota that is slightly larger,
 657          * but still within the pool's allocation slop.  In cases where
 658          * we're very close to full, this will allow a steady trickle of
 659          * removes to get through.
 660          */
 661         if (dd->dd_parent == NULL) {
 662                 spa_t *spa = dd->dd_pool->dp_spa;
 663                 uint64_t poolsize = dsl_pool_adjustedsize(dd->dd_pool, netfree);
 664                 deferred = metaslab_class_get_deferred(spa_normal_class(spa));
 665                 if (poolsize - deferred < quota) {
 666                         quota = poolsize - deferred;
 667                         retval = ENOSPC;
 668                 }
 669         }
 670
 671         /*
 672          * If they are requesting more space, and our current estimate
 673          * is over quota, they get to try again unless the actual
 674          * on-disk is over quota and there are no pending changes (which
 675          * may free up space for us).
 676          */
 677         if (used_on_disk + est_inflight >= quota) {
 678                 if (est_inflight > 0 || used_on_disk < quota ||
 679                     (retval == ENOSPC && used_on_disk < quota + deferred))
 680                         retval = ERESTART;
 681                 dprintf_dd(dd, "failing: used=%lluK inflight = %lluK "
 682                     "quota=%lluK tr=%lluK err=%d\n",
 683                     used_on_disk>>10, est_inflight>>10,
 684                     quota>>10, asize>>10, retval);
 685                 mutex_exit(&dd->dd_lock);
 686                 return (SET_ERROR(retval));
 687         }
 688
 689         /* We need to up our estimated delta before dropping dd_lock */
 690         dd->dd_tempreserved[txgidx] += asize;
 691
 692         parent_rsrv = parent_delta(dd, used_on_disk + est_inflight,
 693             asize - ref_rsrv);
 694         mutex_exit(&dd->dd_lock);
 695
 696         tr = kmem_zalloc(sizeof (struct tempreserve), KM_SLEEP);
 697         tr->tr_ds = dd;
 698         tr->tr_size = asize;
 699         list_insert_tail(tr_list, tr);
 700
 701         /* see if it's OK with our parent */
 702         if (dd->dd_parent && parent_rsrv) {
 703                 boolean_t ismos = (dd->dd_phys->dd_head_dataset_obj == 0);
 704
 705                 return (dsl_dir_tempreserve_impl(dd->dd_parent,
 706                     parent_rsrv, netfree, ismos, TRUE, tr_list, tx, FALSE));
 707         } else {
 708                 return (0);
 709         }
 710 }
 711
 712 /*
 713  * Reserve space in this dsl_dir, to be used in this tx's txg.
 714  * After the space has been dirtied (and dsl_dir_willuse_space()
 715  * has been called), the reservation should be canceled, using
 716  * dsl_dir_tempreserve_clear().
 717  */
 718 int
 719 dsl_dir_tempreserve_space(dsl_dir_t *dd, uint64_t lsize, uint64_t asize,
 720     uint64_t fsize, uint64_t usize, void **tr_cookiep, dmu_tx_t *tx)
 721 {
 722         int err;
 723         list_t *tr_list;
 724
 725         if (asize == 0) {
 726                 *tr_cookiep = NULL;
 727                 return (0);
 728         }
 729
 730         tr_list = kmem_alloc(sizeof (list_t), KM_SLEEP);
 731         list_create(tr_list, sizeof (struct tempreserve),
 732             offsetof(struct tempreserve, tr_node));
 733         ASSERT3S(asize, >, 0);
 734         ASSERT3S(fsize, >=, 0);
 735
 736         err = arc_tempreserve_space(lsize, tx->tx_txg);
 737         if (err == 0) {
 738                 struct tempreserve *tr;
 739
 740                 tr = kmem_zalloc(sizeof (struct tempreserve), KM_SLEEP);
 741                 tr->tr_size = lsize;
 742                 list_insert_tail(tr_list, tr);
 743         } else {
 744                 if (err == EAGAIN) {
 745                         /*
 746                          * If arc_memory_throttle() detected that pageout
 747                          * is running and we are low on memory, we delay new
 748                          * non-pageout transactions to give pageout an
 749                          * advantage.
 750                          *
 751                          * It is unfortunate to be delaying while the caller's
 752                          * locks are held.
 753                          */
 754                         txg_delay(dd->dd_pool, tx->tx_txg,
 755                             MSEC2NSEC(10), MSEC2NSEC(10));
 756                         err = SET_ERROR(ERESTART);
 757                 }
 758         }
 759
 760         if (err == 0) {
 761                 err = dsl_dir_tempreserve_impl(dd, asize, fsize >= asize,
 762                     FALSE, asize > usize, tr_list, tx, TRUE);
 763         }
 764
 765         if (err != 0)
 766                 dsl_dir_tempreserve_clear(tr_list, tx);
 767         else
 768                 *tr_cookiep = tr_list;
 769
 770         return (err);
 771 }
 772
 773 /*
 774  * Clear a temporary reservation that we previously made with
 775  * dsl_dir_tempreserve_space().
 776  */
 777 void
 778 dsl_dir_tempreserve_clear(void *tr_cookie, dmu_tx_t *tx)
 779 {
 780         int txgidx = tx->tx_txg & TXG_MASK;
 781         list_t *tr_list = tr_cookie;
 782         struct tempreserve *tr;
 783
 784         ASSERT3U(tx->tx_txg, !=, 0);
 785
 786         if (tr_cookie == NULL)
 787                 return;
 788
 789         while ((tr = list_head(tr_list)) != NULL) {
 790                 if (tr->tr_ds) {
 791                         mutex_enter(&tr->tr_ds->dd_lock);
 792                         ASSERT3U(tr->tr_ds->dd_tempreserved[txgidx], >=,
 793                             tr->tr_size);
 794                         tr->tr_ds->dd_tempreserved[txgidx] -= tr->tr_size;
 795                         mutex_exit(&tr->tr_ds->dd_lock);
 796                 } else {
 797                         arc_tempreserve_clear(tr->tr_size);
 798                 }
 799                 list_remove(tr_list, tr);
 800                 kmem_free(tr, sizeof (struct tempreserve));
 801         }
 802
 803         kmem_free(tr_list, sizeof (list_t));
 804 }
 805
 806 /*
 807  * This should be called from open context when we think we're going to write
 808  * or free space, for example when dirtying data. Be conservative; it's okay
 809  * to write less space or free more, but we don't want to write more or free
 810  * less than the amount specified.
 811  */
 812 void
 813 dsl_dir_willuse_space(dsl_dir_t *dd, int64_t space, dmu_tx_t *tx)
 814 {
 815         int64_t parent_space;
 816         uint64_t est_used;
 817
 818         mutex_enter(&dd->dd_lock);
 819         if (space > 0)
 820                 dd->dd_space_towrite[tx->tx_txg & TXG_MASK] += space;
 821
 822         est_used = dsl_dir_space_towrite(dd) + dd->dd_phys->dd_used_bytes;
 823         parent_space = parent_delta(dd, est_used, space);
 824         mutex_exit(&dd->dd_lock);
 825
 826         /* Make sure that we clean up dd_space_to* */
 827         dsl_dir_dirty(dd, tx);
 828
 829         /* XXX this is potentially expensive and unnecessary... */
 830         if (parent_space && dd->dd_parent)
 831                 dsl_dir_willuse_space(dd->dd_parent, parent_space, tx);
 832 }
 833
 834 /* call from syncing context when we actually write/free space for this dd */
 835 void
 836 dsl_dir_diduse_space(dsl_dir_t *dd, dd_used_t type,
 837     int64_t used, int64_t compressed, int64_t uncompressed, dmu_tx_t *tx)
 838 {
 839         int64_t accounted_delta;
 840
 841         /*
 842          * dsl_dataset_set_refreservation_sync_impl() calls this with
 843          * dd_lock held, so that it can atomically update
 844          * ds->ds_reserved and the dsl_dir accounting, so that
 845          * dsl_dataset_check_quota() can see dataset and dir accounting
 846          * consistently.
 847          */
 848         boolean_t needlock = !MUTEX_HELD(&dd->dd_lock);
 849
 850         ASSERT(dmu_tx_is_syncing(tx));
 851         ASSERT(type < DD_USED_NUM);
 852
 853         dmu_buf_will_dirty(dd->dd_dbuf, tx);
 854
 855         if (needlock)
 856                 mutex_enter(&dd->dd_lock);
 857         accounted_delta = parent_delta(dd, dd->dd_phys->dd_used_bytes, used);
 858         ASSERT(used >= 0 || dd->dd_phys->dd_used_bytes >= -used);
 859         ASSERT(compressed >= 0 ||
 860             dd->dd_phys->dd_compressed_bytes >= -compressed);
 861         ASSERT(uncompressed >= 0 ||
 862             dd->dd_phys->dd_uncompressed_bytes >= -uncompressed);
 863         dd->dd_phys->dd_used_bytes += used;
 864         dd->dd_phys->dd_uncompressed_bytes += uncompressed;
 865         dd->dd_phys->dd_compressed_bytes += compressed;
 866
 867         if (dd->dd_phys->dd_flags & DD_FLAG_USED_BREAKDOWN) {
 868                 ASSERT(used > 0 ||
 869                     dd->dd_phys->dd_used_breakdown[type] >= -used);
 870                 dd->dd_phys->dd_used_breakdown[type] += used;
 871 #ifdef DEBUG
 872                 dd_used_t t;
 873                 uint64_t u = 0;
 874                 for (t = 0; t < DD_USED_NUM; t++)
 875                         u += dd->dd_phys->dd_used_breakdown[t];
 876                 ASSERT3U(u, ==, dd->dd_phys->dd_used_bytes);
 877 #endif
 878         }
 879         if (needlock)
 880                 mutex_exit(&dd->dd_lock);
 881
 882         if (dd->dd_parent != NULL) {
 883                 dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD,
 884                     accounted_delta, compressed, uncompressed, tx);
 885                 dsl_dir_transfer_space(dd->dd_parent,
 886                     used - accounted_delta,
 887                     DD_USED_CHILD_RSRV, DD_USED_CHILD, tx);
 888         }
 889 }
 890
 891 void
 892 dsl_dir_transfer_space(dsl_dir_t *dd, int64_t delta,
 893     dd_used_t oldtype, dd_used_t newtype, dmu_tx_t *tx)
 894 {
 895         ASSERT(dmu_tx_is_syncing(tx));
 896         ASSERT(oldtype < DD_USED_NUM);
 897         ASSERT(newtype < DD_USED_NUM);
 898
 899         if (delta == 0 || !(dd->dd_phys->dd_flags & DD_FLAG_USED_BREAKDOWN))
 900                 return;
 901
 902         dmu_buf_will_dirty(dd->dd_dbuf, tx);
 903         mutex_enter(&dd->dd_lock);
 904         ASSERT(delta > 0 ?
 905             dd->dd_phys->dd_used_breakdown[oldtype] >= delta :
 906             dd->dd_phys->dd_used_breakdown[newtype] >= -delta);
 907         ASSERT(dd->dd_phys->dd_used_bytes >= ABS(delta));
 908         dd->dd_phys->dd_used_breakdown[oldtype] -= delta;
 909         dd->dd_phys->dd_used_breakdown[newtype] += delta;
 910         mutex_exit(&dd->dd_lock);
 911 }
 912
 913 typedef struct dsl_dir_set_qr_arg {
 914         const char *ddsqra_name;
 915         zprop_source_t ddsqra_source;
 916         uint64_t ddsqra_value;
 917 } dsl_dir_set_qr_arg_t;
 918
 919 static int
 920 dsl_dir_set_quota_check(void *arg, dmu_tx_t *tx)
 921 {
 922         dsl_dir_set_qr_arg_t *ddsqra = arg;
 923         dsl_pool_t *dp = dmu_tx_pool(tx);
 924         dsl_dataset_t *ds;
 925         int error;
 926         uint64_t towrite, newval;
 927
 928         error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds);
 929         if (error != 0)
 930                 return (error);
 931
 932         error = dsl_prop_predict(ds->ds_dir, "quota",
 933             ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval);
 934         if (error != 0) {
 935                 dsl_dataset_rele(ds, FTAG);
 936                 return (error);
 937         }
 938
 939         if (newval == 0) {
 940                 dsl_dataset_rele(ds, FTAG);
 941                 return (0);
 942         }
 943
 944         mutex_enter(&ds->ds_dir->dd_lock);
 945         /*
 946          * If we are doing the preliminary check in open context, and
 947          * there are pending changes, then don't fail it, since the
 948          * pending changes could under-estimate the amount of space to be
 949          * freed up.
 950          */
 951         towrite = dsl_dir_space_towrite(ds->ds_dir);
 952         if ((dmu_tx_is_syncing(tx) || towrite == 0) &&
 953             (newval < ds->ds_dir->dd_phys->dd_reserved ||
 954             newval < ds->ds_dir->dd_phys->dd_used_bytes + towrite)) {
 955                 error = SET_ERROR(ENOSPC);
 956         }
 957         mutex_exit(&ds->ds_dir->dd_lock);
 958         dsl_dataset_rele(ds, FTAG);
 959         return (error);
 960 }
 961
 962 static void
 963 dsl_dir_set_quota_sync(void *arg, dmu_tx_t *tx)
 964 {
 965         dsl_dir_set_qr_arg_t *ddsqra = arg;
 966         dsl_pool_t *dp = dmu_tx_pool(tx);
 967         dsl_dataset_t *ds;
 968         uint64_t newval;
 969
 970         VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds));
 971
 972         if (spa_version(dp->dp_spa) >= SPA_VERSION_RECVD_PROPS) {
 973                 dsl_prop_set_sync_impl(ds, zfs_prop_to_name(ZFS_PROP_QUOTA),
 974                     ddsqra->ddsqra_source, sizeof (ddsqra->ddsqra_value), 1,
 975                     &ddsqra->ddsqra_value, tx);
 976
 977                 VERIFY0(dsl_prop_get_int_ds(ds,
 978                     zfs_prop_to_name(ZFS_PROP_QUOTA), &newval));
 979         } else {
 980                 newval = ddsqra->ddsqra_value;
 981                 spa_history_log_internal_ds(ds, "set", tx, "%s=%lld",
 982                     zfs_prop_to_name(ZFS_PROP_QUOTA), (longlong_t)newval);
 983         }
 984
 985         dmu_buf_will_dirty(ds->ds_dir->dd_dbuf, tx);
 986         mutex_enter(&ds->ds_dir->dd_lock);
 987         ds->ds_dir->dd_phys->dd_quota = newval;
 988         mutex_exit(&ds->ds_dir->dd_lock);
 989         dsl_dataset_rele(ds, FTAG);
 990 }
 991
 992 int
 993 dsl_dir_set_quota(const char *ddname, zprop_source_t source, uint64_t quota)
 994 {
 995         dsl_dir_set_qr_arg_t ddsqra;
 996
 997         ddsqra.ddsqra_name = ddname;
 998         ddsqra.ddsqra_source = source;
 999         ddsqra.ddsqra_value = quota;
1000
1001         return (dsl_sync_task(ddname, dsl_dir_set_quota_check,
1002             dsl_dir_set_quota_sync, &ddsqra, 0));
1003 }
1004
1005 int
1006 dsl_dir_set_reservation_check(void *arg, dmu_tx_t *tx)
1007 {
1008         dsl_dir_set_qr_arg_t *ddsqra = arg;
1009         dsl_pool_t *dp = dmu_tx_pool(tx);
1010         dsl_dataset_t *ds;
1011         dsl_dir_t *dd;
1012         uint64_t newval, used, avail;
1013         int error;
1014
1015         error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds);
1016         if (error != 0)
1017                 return (error);
1018         dd = ds->ds_dir;
1019
1020         /*
1021          * If we are doing the preliminary check in open context, the
1022          * space estimates may be inaccurate.
1023          */
1024         if (!dmu_tx_is_syncing(tx)) {
1025                 dsl_dataset_rele(ds, FTAG);
1026                 return (0);
1027         }
1028
1029         error = dsl_prop_predict(ds->ds_dir,
1030             zfs_prop_to_name(ZFS_PROP_RESERVATION),
1031             ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval);
1032         if (error != 0) {
1033                 dsl_dataset_rele(ds, FTAG);
1034                 return (error);
1035         }
1036
1037         mutex_enter(&dd->dd_lock);
1038         used = dd->dd_phys->dd_used_bytes;
1039         mutex_exit(&dd->dd_lock);
1040
1041         if (dd->dd_parent) {
1042                 avail = dsl_dir_space_available(dd->dd_parent,
1043                     NULL, 0, FALSE);
1044         } else {
1045                 avail = dsl_pool_adjustedsize(dd->dd_pool, B_FALSE) - used;
1046         }
1047
1048         if (MAX(used, newval) > MAX(used, dd->dd_phys->dd_reserved)) {
1049                 uint64_t delta = MAX(used, newval) -
1050                     MAX(used, dd->dd_phys->dd_reserved);
1051
1052                 if (delta > avail ||
1053                     (dd->dd_phys->dd_quota > 0 &&
1054                     newval > dd->dd_phys->dd_quota))
1055                         error = SET_ERROR(ENOSPC);
1056         }
1057
1058         dsl_dataset_rele(ds, FTAG);
1059         return (error);
1060 }
1061
1062 void
1063 dsl_dir_set_reservation_sync_impl(dsl_dir_t *dd, uint64_t value, dmu_tx_t *tx)
1064 {
1065         uint64_t used;
1066         int64_t delta;
1067
1068         dmu_buf_will_dirty(dd->dd_dbuf, tx);
1069
1070         mutex_enter(&dd->dd_lock);
1071         used = dd->dd_phys->dd_used_bytes;
1072         delta = MAX(used, value) - MAX(used, dd->dd_phys->dd_reserved);
1073         dd->dd_phys->dd_reserved = value;
1074
1075         if (dd->dd_parent != NULL) {
1076                 /* Roll up this additional usage into our ancestors */
1077                 dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD_RSRV,
1078                     delta, 0, 0, tx);
1079         }
1080         mutex_exit(&dd->dd_lock);
1081 }
1082
1083 static void
1084 dsl_dir_set_reservation_sync(void *arg, dmu_tx_t *tx)
1085 {
1086         dsl_dir_set_qr_arg_t *ddsqra = arg;
1087         dsl_pool_t *dp = dmu_tx_pool(tx);
1088         dsl_dataset_t *ds;
1089         uint64_t newval;
1090
1091         VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds));
1092
1093         if (spa_version(dp->dp_spa) >= SPA_VERSION_RECVD_PROPS) {
1094                 dsl_prop_set_sync_impl(ds,
1095                     zfs_prop_to_name(ZFS_PROP_RESERVATION),
1096                     ddsqra->ddsqra_source, sizeof (ddsqra->ddsqra_value), 1,
1097                     &ddsqra->ddsqra_value, tx);
1098
1099                 VERIFY0(dsl_prop_get_int_ds(ds,
1100                     zfs_prop_to_name(ZFS_PROP_RESERVATION), &newval));
1101         } else {
1102                 newval = ddsqra->ddsqra_value;
1103                 spa_history_log_internal_ds(ds, "set", tx, "%s=%lld",
1104                     zfs_prop_to_name(ZFS_PROP_RESERVATION),
1105                     (longlong_t)newval);
1106         }
1107
1108         dsl_dir_set_reservation_sync_impl(ds->ds_dir, newval, tx);
1109         dsl_dataset_rele(ds, FTAG);
1110 }
1111
1112 int
1113 dsl_dir_set_reservation(const char *ddname, zprop_source_t source,
1114     uint64_t reservation)
1115 {
1116         dsl_dir_set_qr_arg_t ddsqra;
1117
1118         ddsqra.ddsqra_name = ddname;
1119         ddsqra.ddsqra_source = source;
1120         ddsqra.ddsqra_value = reservation;
1121
1122         return (dsl_sync_task(ddname, dsl_dir_set_reservation_check,
1123             dsl_dir_set_reservation_sync, &ddsqra, 0));
1124 }
1125
1126 static dsl_dir_t *
1127 closest_common_ancestor(dsl_dir_t *ds1, dsl_dir_t *ds2)
1128 {
1129         for (; ds1; ds1 = ds1->dd_parent) {
1130                 dsl_dir_t *dd;
1131                 for (dd = ds2; dd; dd = dd->dd_parent) {
1132                         if (ds1 == dd)
1133                                 return (dd);
1134                 }
1135         }
1136         return (NULL);
1137 }
1138
1139 /*
1140  * If delta is applied to dd, how much of that delta would be applied to
1141  * ancestor?  Syncing context only.
1142  */
1143 static int64_t
1144 would_change(dsl_dir_t *dd, int64_t delta, dsl_dir_t *ancestor)
1145 {
1146         if (dd == ancestor)
1147                 return (delta);
1148
1149         mutex_enter(&dd->dd_lock);
1150         delta = parent_delta(dd, dd->dd_phys->dd_used_bytes, delta);
1151         mutex_exit(&dd->dd_lock);
1152         return (would_change(dd->dd_parent, delta, ancestor));
1153 }
1154
1155 typedef struct dsl_dir_rename_arg {
1156         const char *ddra_oldname;
1157         const char *ddra_newname;
1158 } dsl_dir_rename_arg_t;
1159
1160 /* ARGSUSED */
1161 static int
1162 dsl_valid_rename(dsl_pool_t *dp, dsl_dataset_t *ds, void *arg)
1163 {
1164         int *deltap = arg;
1165         char namebuf[MAXNAMELEN];
1166
1167         dsl_dataset_name(ds, namebuf);
1168
1169         if (strlen(namebuf) + *deltap >= MAXNAMELEN)
1170                 return (SET_ERROR(ENAMETOOLONG));
1171         return (0);
1172 }
1173
1174 static int
1175 dsl_dir_rename_check(void *arg, dmu_tx_t *tx)
1176 {
1177         dsl_dir_rename_arg_t *ddra = arg;
1178         dsl_pool_t *dp = dmu_tx_pool(tx);
1179         dsl_dir_t *dd, *newparent;
1180         const char *mynewname;
1181         int error;
1182         int delta = strlen(ddra->ddra_newname) - strlen(ddra->ddra_oldname);
1183
1184         /* target dir should exist */
1185         error = dsl_dir_hold(dp, ddra->ddra_oldname, FTAG, &dd, NULL);
1186         if (error != 0)
1187                 return (error);
1188
1189         /* new parent should exist */
1190         error = dsl_dir_hold(dp, ddra->ddra_newname, FTAG,
1191             &newparent, &mynewname);
1192         if (error != 0) {
1193                 dsl_dir_rele(dd, FTAG);
1194                 return (error);
1195         }
1196
1197         /* can't rename to different pool */
1198         if (dd->dd_pool != newparent->dd_pool) {
1199                 dsl_dir_rele(newparent, FTAG);
1200                 dsl_dir_rele(dd, FTAG);
1201                 return (SET_ERROR(ENXIO));
1202         }
1203
1204         /* new name should not already exist */
1205         if (mynewname == NULL) {
1206                 dsl_dir_rele(newparent, FTAG);
1207                 dsl_dir_rele(dd, FTAG);
1208                 return (SET_ERROR(EEXIST));
1209         }
1210
1211         /* if the name length is growing, validate child name lengths */
1212         if (delta > 0) {
1213                 error = dmu_objset_find_dp(dp, dd->dd_object, dsl_valid_rename,
1214                     &delta, DS_FIND_CHILDREN | DS_FIND_SNAPSHOTS);
1215                 if (error != 0) {
1216                         dsl_dir_rele(newparent, FTAG);
1217                         dsl_dir_rele(dd, FTAG);
1218                         return (error);
1219                 }
1220         }
1221
1222         if (newparent != dd->dd_parent) {
1223                 /* is there enough space? */
1224                 uint64_t myspace =
1225                     MAX(dd->dd_phys->dd_used_bytes, dd->dd_phys->dd_reserved);
1226
1227                 /* no rename into our descendant */
1228                 if (closest_common_ancestor(dd, newparent) == dd) {
1229                         dsl_dir_rele(newparent, FTAG);
1230                         dsl_dir_rele(dd, FTAG);
1231                         return (SET_ERROR(EINVAL));
1232                 }
1233
1234                 error = dsl_dir_transfer_possible(dd->dd_parent,
1235                     newparent, myspace);
1236                 if (error != 0) {
1237                         dsl_dir_rele(newparent, FTAG);
1238                         dsl_dir_rele(dd, FTAG);
1239                         return (error);
1240                 }
1241         }
1242
1243         dsl_dir_rele(newparent, FTAG);
1244         dsl_dir_rele(dd, FTAG);
1245         return (0);
1246 }
1247
1248 static void
1249 dsl_dir_rename_sync(void *arg, dmu_tx_t *tx)
1250 {
1251         dsl_dir_rename_arg_t *ddra = arg;
1252         dsl_pool_t *dp = dmu_tx_pool(tx);
1253         dsl_dir_t *dd, *newparent;
1254         const char *mynewname;
1255         int error;
1256         objset_t *mos = dp->dp_meta_objset;
1257
1258         VERIFY0(dsl_dir_hold(dp, ddra->ddra_oldname, FTAG, &dd, NULL));
1259         VERIFY0(dsl_dir_hold(dp, ddra->ddra_newname, FTAG, &newparent,
1260             &mynewname));
1261
1262         /* Log this before we change the name. */
1263         spa_history_log_internal_dd(dd, "rename", tx,
1264             "-> %s", ddra->ddra_newname);
1265
1266         if (newparent != dd->dd_parent) {
1267                 dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD,
1268                     -dd->dd_phys->dd_used_bytes,
1269                     -dd->dd_phys->dd_compressed_bytes,
1270                     -dd->dd_phys->dd_uncompressed_bytes, tx);
1271                 dsl_dir_diduse_space(newparent, DD_USED_CHILD,
1272                     dd->dd_phys->dd_used_bytes,
1273                     dd->dd_phys->dd_compressed_bytes,
1274                     dd->dd_phys->dd_uncompressed_bytes, tx);
1275
1276                 if (dd->dd_phys->dd_reserved > dd->dd_phys->dd_used_bytes) {
1277                         uint64_t unused_rsrv = dd->dd_phys->dd_reserved -
1278                             dd->dd_phys->dd_used_bytes;
1279
1280                         dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD_RSRV,
1281                             -unused_rsrv, 0, 0, tx);
1282                         dsl_dir_diduse_space(newparent, DD_USED_CHILD_RSRV,
1283                             unused_rsrv, 0, 0, tx);
1284                 }
1285         }
1286
1287         dmu_buf_will_dirty(dd->dd_dbuf, tx);
1288
1289         /* remove from old parent zapobj */
1290         error = zap_remove(mos, dd->dd_parent->dd_phys->dd_child_dir_zapobj,
1291             dd->dd_myname, tx);
1292         ASSERT0(error);
1293
1294         (void) strcpy(dd->dd_myname, mynewname);
1295         dsl_dir_rele(dd->dd_parent, dd);
1296         dd->dd_phys->dd_parent_obj = newparent->dd_object;
1297         VERIFY0(dsl_dir_hold_obj(dp,
1298             newparent->dd_object, NULL, dd, &dd->dd_parent));
1299
1300         /* add to new parent zapobj */
1301         VERIFY0(zap_add(mos, newparent->dd_phys->dd_child_dir_zapobj,
1302             dd->dd_myname, 8, 1, &dd->dd_object, tx));
1303
1304 #ifdef __FreeBSD__
1305 #ifdef _KERNEL
1306         zfsvfs_update_fromname(ddra->ddra_oldname, ddra->ddra_newname);
1307         zvol_rename_minors(ddra->ddra_oldname, ddra->ddra_newname);
1308 #endif
1309 #endif
1310
1311         dsl_prop_notify_all(dd);
1312
1313         dsl_dir_rele(newparent, FTAG);
1314         dsl_dir_rele(dd, FTAG);
1315 }
1316
1317 int
1318 dsl_dir_rename(const char *oldname, const char *newname)
1319 {
1320         dsl_dir_rename_arg_t ddra;
1321
1322         ddra.ddra_oldname = oldname;
1323         ddra.ddra_newname = newname;
1324
1325         return (dsl_sync_task(oldname,
1326             dsl_dir_rename_check, dsl_dir_rename_sync, &ddra, 3));
1327 }
1328
1329 int
1330 dsl_dir_transfer_possible(dsl_dir_t *sdd, dsl_dir_t *tdd, uint64_t space)
1331 {
1332         dsl_dir_t *ancestor;
1333         int64_t adelta;
1334         uint64_t avail;
1335
1336         ancestor = closest_common_ancestor(sdd, tdd);
1337         adelta = would_change(sdd, -space, ancestor);
1338         avail = dsl_dir_space_available(tdd, ancestor, adelta, FALSE);
1339         if (avail < space)
1340                 return (SET_ERROR(ENOSPC));
1341
1342         return (0);
1343 }
1344
1345 timestruc_t
1346 dsl_dir_snap_cmtime(dsl_dir_t *dd)
1347 {
1348         timestruc_t t;
1349
1350         mutex_enter(&dd->dd_lock);
1351         t = dd->dd_snap_cmtime;
1352         mutex_exit(&dd->dd_lock);
1353
1354         return (t);
1355 }
1356
1357 void
1358 dsl_dir_snap_cmtime_update(dsl_dir_t *dd)
1359 {
1360         timestruc_t t;
1361
1362         gethrestime(&t);
1363         mutex_enter(&dd->dd_lock);
1364         dd->dd_snap_cmtime = t;
1365         mutex_exit(&dd->dd_lock);
1366 }
1367
1368 void
1369 dsl_dir_zapify(dsl_dir_t *dd, dmu_tx_t *tx)
1370 {
1371         objset_t *mos = dd->dd_pool->dp_meta_objset;
1372         dmu_object_zapify(mos, dd->dd_object, DMU_OT_DSL_DIR, tx);
1373 }