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_pool_t *tr_dp;
 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
 744                 err = dsl_pool_tempreserve_space(dd->dd_pool, asize, tx);
 745         } else {
 746                 if (err == EAGAIN) {
 747                         txg_delay(dd->dd_pool, tx->tx_txg, 1);
 748                         err = SET_ERROR(ERESTART);
 749                 }
 750                 dsl_pool_memory_pressure(dd->dd_pool);
 751         }
 752
 753         if (err == 0) {
 754                 struct tempreserve *tr;
 755
 756                 tr = kmem_zalloc(sizeof (struct tempreserve), KM_SLEEP);
 757                 tr->tr_dp = dd->dd_pool;
 758                 tr->tr_size = asize;
 759                 list_insert_tail(tr_list, tr);
 760
 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)) {
 790                 if (tr->tr_dp) {
 791                         dsl_pool_tempreserve_clear(tr->tr_dp, tr->tr_size, tx);
 792                 } else if (tr->tr_ds) {
 793                         mutex_enter(&tr->tr_ds->dd_lock);
 794                         ASSERT3U(tr->tr_ds->dd_tempreserved[txgidx], >=,
 795                             tr->tr_size);
 796                         tr->tr_ds->dd_tempreserved[txgidx] -= tr->tr_size;
 797                         mutex_exit(&tr->tr_ds->dd_lock);
 798                 } else {
 799                         arc_tempreserve_clear(tr->tr_size);
 800                 }
 801                 list_remove(tr_list, tr);
 802                 kmem_free(tr, sizeof (struct tempreserve));
 803         }
 804
 805         kmem_free(tr_list, sizeof (list_t));
 806 }
 807
 808 static void
 809 dsl_dir_willuse_space_impl(dsl_dir_t *dd, int64_t space, dmu_tx_t *tx)
 810 {
 811         int64_t parent_space;
 812         uint64_t est_used;
 813
 814         mutex_enter(&dd->dd_lock);
 815         if (space > 0)
 816                 dd->dd_space_towrite[tx->tx_txg & TXG_MASK] += space;
 817
 818         est_used = dsl_dir_space_towrite(dd) + dd->dd_phys->dd_used_bytes;
 819         parent_space = parent_delta(dd, est_used, space);
 820         mutex_exit(&dd->dd_lock);
 821
 822         /* Make sure that we clean up dd_space_to* */
 823         dsl_dir_dirty(dd, tx);
 824
 825         /* XXX this is potentially expensive and unnecessary... */
 826         if (parent_space && dd->dd_parent)
 827                 dsl_dir_willuse_space_impl(dd->dd_parent, parent_space, tx);
 828 }
 829
 830 /*
 831  * Call in open context when we think we're going to write/free space,
 832  * eg. when dirtying data.  Be conservative (ie. OK to write less than
 833  * this or free more than this, but don't write more or free less).
 834  */
 835 void
 836 dsl_dir_willuse_space(dsl_dir_t *dd, int64_t space, dmu_tx_t *tx)
 837 {
 838         dsl_pool_willuse_space(dd->dd_pool, space, tx);
 839         dsl_dir_willuse_space_impl(dd, space, tx);
 840 }
 841
 842 /* call from syncing context when we actually write/free space for this dd */
 843 void
 844 dsl_dir_diduse_space(dsl_dir_t *dd, dd_used_t type,
 845     int64_t used, int64_t compressed, int64_t uncompressed, dmu_tx_t *tx)
 846 {
 847         int64_t accounted_delta;
 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         if (needlock)
 854                 mutex_enter(&dd->dd_lock);
 855         accounted_delta = parent_delta(dd, dd->dd_phys->dd_used_bytes, used);
 856         ASSERT(used >= 0 || dd->dd_phys->dd_used_bytes >= -used);
 857         ASSERT(compressed >= 0 ||
 858             dd->dd_phys->dd_compressed_bytes >= -compressed);
 859         ASSERT(uncompressed >= 0 ||
 860             dd->dd_phys->dd_uncompressed_bytes >= -uncompressed);
 861         dmu_buf_will_dirty(dd->dd_dbuf, tx);
 862         dd->dd_phys->dd_used_bytes += used;
 863         dd->dd_phys->dd_uncompressed_bytes += uncompressed;
 864         dd->dd_phys->dd_compressed_bytes += compressed;
 865
 866         if (dd->dd_phys->dd_flags & DD_FLAG_USED_BREAKDOWN) {
 867                 ASSERT(used > 0 ||
 868                     dd->dd_phys->dd_used_breakdown[type] >= -used);
 869                 dd->dd_phys->dd_used_breakdown[type] += used;
 870 #ifdef DEBUG
 871                 dd_used_t t;
 872                 uint64_t u = 0;
 873                 for (t = 0; t < DD_USED_NUM; t++)
 874                         u += dd->dd_phys->dd_used_breakdown[t];
 875                 ASSERT3U(u, ==, dd->dd_phys->dd_used_bytes);
 876 #endif
 877         }
 878         if (needlock)
 879                 mutex_exit(&dd->dd_lock);
 880
 881         if (dd->dd_parent != NULL) {
 882                 dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD,
 883                     accounted_delta, compressed, uncompressed, tx);
 884                 dsl_dir_transfer_space(dd->dd_parent,
 885                     used - accounted_delta,
 886                     DD_USED_CHILD_RSRV, DD_USED_CHILD, tx);
 887         }
 888 }
 889
 890 void
 891 dsl_dir_transfer_space(dsl_dir_t *dd, int64_t delta,
 892     dd_used_t oldtype, dd_used_t newtype, dmu_tx_t *tx)
 893 {
 894         boolean_t needlock = !MUTEX_HELD(&dd->dd_lock);
 895
 896         ASSERT(dmu_tx_is_syncing(tx));
 897         ASSERT(oldtype < DD_USED_NUM);
 898         ASSERT(newtype < DD_USED_NUM);
 899
 900         if (delta == 0 || !(dd->dd_phys->dd_flags & DD_FLAG_USED_BREAKDOWN))
 901                 return;
 902
 903         if (needlock)
 904                 mutex_enter(&dd->dd_lock);
 905         ASSERT(delta > 0 ?
 906             dd->dd_phys->dd_used_breakdown[oldtype] >= delta :
 907             dd->dd_phys->dd_used_breakdown[newtype] >= -delta);
 908         ASSERT(dd->dd_phys->dd_used_bytes >= ABS(delta));
 909         dmu_buf_will_dirty(dd->dd_dbuf, tx);
 910         dd->dd_phys->dd_used_breakdown[oldtype] -= delta;
 911         dd->dd_phys->dd_used_breakdown[newtype] += delta;
 912         if (needlock)
 913                 mutex_exit(&dd->dd_lock);
 914 }
 915
 916 typedef struct dsl_dir_set_qr_arg {
 917         const char *ddsqra_name;
 918         zprop_source_t ddsqra_source;
 919         uint64_t ddsqra_value;
 920 } dsl_dir_set_qr_arg_t;
 921
 922 static int
 923 dsl_dir_set_quota_check(void *arg, dmu_tx_t *tx)
 924 {
 925         dsl_dir_set_qr_arg_t *ddsqra = arg;
 926         dsl_pool_t *dp = dmu_tx_pool(tx);
 927         dsl_dataset_t *ds;
 928         int error;
 929         uint64_t towrite, newval;
 930
 931         error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds);
 932         if (error != 0)
 933                 return (error);
 934
 935         error = dsl_prop_predict(ds->ds_dir, "quota",
 936             ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval);
 937         if (error != 0) {
 938                 dsl_dataset_rele(ds, FTAG);
 939                 return (error);
 940         }
 941
 942         if (newval == 0) {
 943                 dsl_dataset_rele(ds, FTAG);
 944                 return (0);
 945         }
 946
 947         mutex_enter(&ds->ds_dir->dd_lock);
 948         /*
 949          * If we are doing the preliminary check in open context, and
 950          * there are pending changes, then don't fail it, since the
 951          * pending changes could under-estimate the amount of space to be
 952          * freed up.
 953          */
 954         towrite = dsl_dir_space_towrite(ds->ds_dir);
 955         if ((dmu_tx_is_syncing(tx) || towrite == 0) &&
 956             (newval < ds->ds_dir->dd_phys->dd_reserved ||
 957             newval < ds->ds_dir->dd_phys->dd_used_bytes + towrite)) {
 958                 error = SET_ERROR(ENOSPC);
 959         }
 960         mutex_exit(&ds->ds_dir->dd_lock);
 961         dsl_dataset_rele(ds, FTAG);
 962         return (error);
 963 }
 964
 965 static void
 966 dsl_dir_set_quota_sync(void *arg, dmu_tx_t *tx)
 967 {
 968         dsl_dir_set_qr_arg_t *ddsqra = arg;
 969         dsl_pool_t *dp = dmu_tx_pool(tx);
 970         dsl_dataset_t *ds;
 971         uint64_t newval;
 972
 973         VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds));
 974
 975         if (spa_version(dp->dp_spa) >= SPA_VERSION_RECVD_PROPS) {
 976                 dsl_prop_set_sync_impl(ds, zfs_prop_to_name(ZFS_PROP_QUOTA),
 977                     ddsqra->ddsqra_source, sizeof (ddsqra->ddsqra_value), 1,
 978                     &ddsqra->ddsqra_value, tx);
 979
 980                 VERIFY0(dsl_prop_get_int_ds(ds,
 981                     zfs_prop_to_name(ZFS_PROP_QUOTA), &newval));
 982         } else {
 983                 newval = ddsqra->ddsqra_value;
 984                 spa_history_log_internal_ds(ds, "set", tx, "%s=%lld",
 985                     zfs_prop_to_name(ZFS_PROP_QUOTA), (longlong_t)newval);
 986         }
 987
 988         dmu_buf_will_dirty(ds->ds_dir->dd_dbuf, tx);
 989         mutex_enter(&ds->ds_dir->dd_lock);
 990         ds->ds_dir->dd_phys->dd_quota = newval;
 991         mutex_exit(&ds->ds_dir->dd_lock);
 992         dsl_dataset_rele(ds, FTAG);
 993 }
 994
 995 int
 996 dsl_dir_set_quota(const char *ddname, zprop_source_t source, uint64_t quota)
 997 {
 998         dsl_dir_set_qr_arg_t ddsqra;
 999
1000         ddsqra.ddsqra_name = ddname;
1001         ddsqra.ddsqra_source = source;
1002         ddsqra.ddsqra_value = quota;
1003
1004         return (dsl_sync_task(ddname, dsl_dir_set_quota_check,
1005             dsl_dir_set_quota_sync, &ddsqra, 0));
1006 }
1007
1008 int
1009 dsl_dir_set_reservation_check(void *arg, dmu_tx_t *tx)
1010 {
1011         dsl_dir_set_qr_arg_t *ddsqra = arg;
1012         dsl_pool_t *dp = dmu_tx_pool(tx);
1013         dsl_dataset_t *ds;
1014         dsl_dir_t *dd;
1015         uint64_t newval, used, avail;
1016         int error;
1017
1018         error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds);
1019         if (error != 0)
1020                 return (error);
1021         dd = ds->ds_dir;
1022
1023         /*
1024          * If we are doing the preliminary check in open context, the
1025          * space estimates may be inaccurate.
1026          */
1027         if (!dmu_tx_is_syncing(tx)) {
1028                 dsl_dataset_rele(ds, FTAG);
1029                 return (0);
1030         }
1031
1032         error = dsl_prop_predict(ds->ds_dir,
1033             zfs_prop_to_name(ZFS_PROP_RESERVATION),
1034             ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval);
1035         if (error != 0) {
1036                 dsl_dataset_rele(ds, FTAG);
1037                 return (error);
1038         }
1039
1040         mutex_enter(&dd->dd_lock);
1041         used = dd->dd_phys->dd_used_bytes;
1042         mutex_exit(&dd->dd_lock);
1043
1044         if (dd->dd_parent) {
1045                 avail = dsl_dir_space_available(dd->dd_parent,
1046                     NULL, 0, FALSE);
1047         } else {
1048                 avail = dsl_pool_adjustedsize(dd->dd_pool, B_FALSE) - used;
1049         }
1050
1051         if (MAX(used, newval) > MAX(used, dd->dd_phys->dd_reserved)) {
1052                 uint64_t delta = MAX(used, newval) -
1053                     MAX(used, dd->dd_phys->dd_reserved);
1054
1055                 if (delta > avail ||
1056                     (dd->dd_phys->dd_quota > 0 &&
1057                     newval > dd->dd_phys->dd_quota))
1058                         error = SET_ERROR(ENOSPC);
1059         }
1060
1061         dsl_dataset_rele(ds, FTAG);
1062         return (error);
1063 }
1064
1065 void
1066 dsl_dir_set_reservation_sync_impl(dsl_dir_t *dd, uint64_t value, dmu_tx_t *tx)
1067 {
1068         uint64_t used;
1069         int64_t delta;
1070
1071         dmu_buf_will_dirty(dd->dd_dbuf, tx);
1072
1073         mutex_enter(&dd->dd_lock);
1074         used = dd->dd_phys->dd_used_bytes;
1075         delta = MAX(used, value) - MAX(used, dd->dd_phys->dd_reserved);
1076         dd->dd_phys->dd_reserved = value;
1077
1078         if (dd->dd_parent != NULL) {
1079                 /* Roll up this additional usage into our ancestors */
1080                 dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD_RSRV,
1081                     delta, 0, 0, tx);
1082         }
1083         mutex_exit(&dd->dd_lock);
1084 }
1085
1086 static void
1087 dsl_dir_set_reservation_sync(void *arg, dmu_tx_t *tx)
1088 {
1089         dsl_dir_set_qr_arg_t *ddsqra = arg;
1090         dsl_pool_t *dp = dmu_tx_pool(tx);
1091         dsl_dataset_t *ds;
1092         uint64_t newval;
1093
1094         VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds));
1095
1096         if (spa_version(dp->dp_spa) >= SPA_VERSION_RECVD_PROPS) {
1097                 dsl_prop_set_sync_impl(ds,
1098                     zfs_prop_to_name(ZFS_PROP_RESERVATION),
1099                     ddsqra->ddsqra_source, sizeof (ddsqra->ddsqra_value), 1,
1100                     &ddsqra->ddsqra_value, tx);
1101
1102                 VERIFY0(dsl_prop_get_int_ds(ds,
1103                     zfs_prop_to_name(ZFS_PROP_RESERVATION), &newval));
1104         } else {
1105                 newval = ddsqra->ddsqra_value;
1106                 spa_history_log_internal_ds(ds, "set", tx, "%s=%lld",
1107                     zfs_prop_to_name(ZFS_PROP_RESERVATION),
1108                     (longlong_t)newval);
1109         }
1110
1111         dsl_dir_set_reservation_sync_impl(ds->ds_dir, newval, tx);
1112         dsl_dataset_rele(ds, FTAG);
1113 }
1114
1115 int
1116 dsl_dir_set_reservation(const char *ddname, zprop_source_t source,
1117     uint64_t reservation)
1118 {
1119         dsl_dir_set_qr_arg_t ddsqra;
1120
1121         ddsqra.ddsqra_name = ddname;
1122         ddsqra.ddsqra_source = source;
1123         ddsqra.ddsqra_value = reservation;
1124
1125         return (dsl_sync_task(ddname, dsl_dir_set_reservation_check,
1126             dsl_dir_set_reservation_sync, &ddsqra, 0));
1127 }
1128
1129 static dsl_dir_t *
1130 closest_common_ancestor(dsl_dir_t *ds1, dsl_dir_t *ds2)
1131 {
1132         for (; ds1; ds1 = ds1->dd_parent) {
1133                 dsl_dir_t *dd;
1134                 for (dd = ds2; dd; dd = dd->dd_parent) {
1135                         if (ds1 == dd)
1136                                 return (dd);
1137                 }
1138         }
1139         return (NULL);
1140 }
1141
1142 /*
1143  * If delta is applied to dd, how much of that delta would be applied to
1144  * ancestor?  Syncing context only.
1145  */
1146 static int64_t
1147 would_change(dsl_dir_t *dd, int64_t delta, dsl_dir_t *ancestor)
1148 {
1149         if (dd == ancestor)
1150                 return (delta);
1151
1152         mutex_enter(&dd->dd_lock);
1153         delta = parent_delta(dd, dd->dd_phys->dd_used_bytes, delta);
1154         mutex_exit(&dd->dd_lock);
1155         return (would_change(dd->dd_parent, delta, ancestor));
1156 }
1157
1158 typedef struct dsl_dir_rename_arg {
1159         const char *ddra_oldname;
1160         const char *ddra_newname;
1161 } dsl_dir_rename_arg_t;
1162
1163 /* ARGSUSED */
1164 static int
1165 dsl_valid_rename(dsl_pool_t *dp, dsl_dataset_t *ds, void *arg)
1166 {
1167         int *deltap = arg;
1168         char namebuf[MAXNAMELEN];
1169
1170         dsl_dataset_name(ds, namebuf);
1171
1172         if (strlen(namebuf) + *deltap >= MAXNAMELEN)
1173                 return (SET_ERROR(ENAMETOOLONG));
1174         return (0);
1175 }
1176
1177 static int
1178 dsl_dir_rename_check(void *arg, dmu_tx_t *tx)
1179 {
1180         dsl_dir_rename_arg_t *ddra = arg;
1181         dsl_pool_t *dp = dmu_tx_pool(tx);
1182         dsl_dir_t *dd, *newparent;
1183         const char *mynewname;
1184         int error;
1185         int delta = strlen(ddra->ddra_newname) - strlen(ddra->ddra_oldname);
1186
1187         /* target dir should exist */
1188         error = dsl_dir_hold(dp, ddra->ddra_oldname, FTAG, &dd, NULL);
1189         if (error != 0)
1190                 return (error);
1191
1192         /* new parent should exist */
1193         error = dsl_dir_hold(dp, ddra->ddra_newname, FTAG,
1194             &newparent, &mynewname);
1195         if (error != 0) {
1196                 dsl_dir_rele(dd, FTAG);
1197                 return (error);
1198         }
1199
1200         /* can't rename to different pool */
1201         if (dd->dd_pool != newparent->dd_pool) {
1202                 dsl_dir_rele(newparent, FTAG);
1203                 dsl_dir_rele(dd, FTAG);
1204                 return (SET_ERROR(ENXIO));
1205         }
1206
1207         /* new name should not already exist */
1208         if (mynewname == NULL) {
1209                 dsl_dir_rele(newparent, FTAG);
1210                 dsl_dir_rele(dd, FTAG);
1211                 return (SET_ERROR(EEXIST));
1212         }
1213
1214         /* if the name length is growing, validate child name lengths */
1215         if (delta > 0) {
1216                 error = dmu_objset_find_dp(dp, dd->dd_object, dsl_valid_rename,
1217                     &delta, DS_FIND_CHILDREN | DS_FIND_SNAPSHOTS);
1218                 if (error != 0) {
1219                         dsl_dir_rele(newparent, FTAG);
1220                         dsl_dir_rele(dd, FTAG);
1221                         return (error);
1222                 }
1223         }
1224
1225         if (newparent != dd->dd_parent) {
1226                 /* is there enough space? */
1227                 uint64_t myspace =
1228                     MAX(dd->dd_phys->dd_used_bytes, dd->dd_phys->dd_reserved);
1229
1230                 /* no rename into our descendant */
1231                 if (closest_common_ancestor(dd, newparent) == dd) {
1232                         dsl_dir_rele(newparent, FTAG);
1233                         dsl_dir_rele(dd, FTAG);
1234                         return (SET_ERROR(EINVAL));
1235                 }
1236
1237                 error = dsl_dir_transfer_possible(dd->dd_parent,
1238                     newparent, myspace);
1239                 if (error != 0) {
1240                         dsl_dir_rele(newparent, FTAG);
1241                         dsl_dir_rele(dd, FTAG);
1242                         return (error);
1243                 }
1244         }
1245
1246         dsl_dir_rele(newparent, FTAG);
1247         dsl_dir_rele(dd, FTAG);
1248         return (0);
1249 }
1250
1251 static void
1252 dsl_dir_rename_sync(void *arg, dmu_tx_t *tx)
1253 {
1254         dsl_dir_rename_arg_t *ddra = arg;
1255         dsl_pool_t *dp = dmu_tx_pool(tx);
1256         dsl_dir_t *dd, *newparent;
1257         const char *mynewname;
1258         int error;
1259         objset_t *mos = dp->dp_meta_objset;
1260
1261         VERIFY0(dsl_dir_hold(dp, ddra->ddra_oldname, FTAG, &dd, NULL));
1262         VERIFY0(dsl_dir_hold(dp, ddra->ddra_newname, FTAG, &newparent,
1263             &mynewname));
1264
1265         /* Log this before we change the name. */
1266         spa_history_log_internal_dd(dd, "rename", tx,
1267             "-> %s", ddra->ddra_newname);
1268
1269         if (newparent != dd->dd_parent) {
1270                 dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD,
1271                     -dd->dd_phys->dd_used_bytes,
1272                     -dd->dd_phys->dd_compressed_bytes,
1273                     -dd->dd_phys->dd_uncompressed_bytes, tx);
1274                 dsl_dir_diduse_space(newparent, DD_USED_CHILD,
1275                     dd->dd_phys->dd_used_bytes,
1276                     dd->dd_phys->dd_compressed_bytes,
1277                     dd->dd_phys->dd_uncompressed_bytes, tx);
1278
1279                 if (dd->dd_phys->dd_reserved > dd->dd_phys->dd_used_bytes) {
1280                         uint64_t unused_rsrv = dd->dd_phys->dd_reserved -
1281                             dd->dd_phys->dd_used_bytes;
1282
1283                         dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD_RSRV,
1284                             -unused_rsrv, 0, 0, tx);
1285                         dsl_dir_diduse_space(newparent, DD_USED_CHILD_RSRV,
1286                             unused_rsrv, 0, 0, tx);
1287                 }
1288         }
1289
1290         dmu_buf_will_dirty(dd->dd_dbuf, tx);
1291
1292         /* remove from old parent zapobj */
1293         error = zap_remove(mos, dd->dd_parent->dd_phys->dd_child_dir_zapobj,
1294             dd->dd_myname, tx);
1295         ASSERT0(error);
1296
1297         (void) strcpy(dd->dd_myname, mynewname);
1298         dsl_dir_rele(dd->dd_parent, dd);
1299         dd->dd_phys->dd_parent_obj = newparent->dd_object;
1300         VERIFY0(dsl_dir_hold_obj(dp,
1301             newparent->dd_object, NULL, dd, &dd->dd_parent));
1302
1303         /* add to new parent zapobj */
1304         VERIFY0(zap_add(mos, newparent->dd_phys->dd_child_dir_zapobj,
1305             dd->dd_myname, 8, 1, &dd->dd_object, tx));
1306
1307 #ifdef __FreeBSD__
1308 #ifdef _KERNEL
1309         zfsvfs_update_fromname(ddra->ddra_oldname, ddra->ddra_newname);
1310         zvol_rename_minors(ddra->ddra_oldname, ddra->ddra_newname);
1311 #endif
1312 #endif
1313
1314         dsl_prop_notify_all(dd);
1315
1316         dsl_dir_rele(newparent, FTAG);
1317         dsl_dir_rele(dd, FTAG);
1318 }
1319
1320 int
1321 dsl_dir_rename(const char *oldname, const char *newname)
1322 {
1323         dsl_dir_rename_arg_t ddra;
1324
1325         ddra.ddra_oldname = oldname;
1326         ddra.ddra_newname = newname;
1327
1328         return (dsl_sync_task(oldname,
1329             dsl_dir_rename_check, dsl_dir_rename_sync, &ddra, 3));
1330 }
1331
1332 int
1333 dsl_dir_transfer_possible(dsl_dir_t *sdd, dsl_dir_t *tdd, uint64_t space)
1334 {
1335         dsl_dir_t *ancestor;
1336         int64_t adelta;
1337         uint64_t avail;
1338
1339         ancestor = closest_common_ancestor(sdd, tdd);
1340         adelta = would_change(sdd, -space, ancestor);
1341         avail = dsl_dir_space_available(tdd, ancestor, adelta, FALSE);
1342         if (avail < space)
1343                 return (SET_ERROR(ENOSPC));
1344
1345         return (0);
1346 }
1347
1348 timestruc_t
1349 dsl_dir_snap_cmtime(dsl_dir_t *dd)
1350 {
1351         timestruc_t t;
1352
1353         mutex_enter(&dd->dd_lock);
1354         t = dd->dd_snap_cmtime;
1355         mutex_exit(&dd->dd_lock);
1356
1357         return (t);
1358 }
1359
1360 void
1361 dsl_dir_snap_cmtime_update(dsl_dir_t *dd)
1362 {
1363         timestruc_t t;
1364
1365         gethrestime(&t);
1366         mutex_enter(&dd->dd_lock);
1367         dd->dd_snap_cmtime = t;
1368         mutex_exit(&dd->dd_lock);
1369 }