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