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         ASSERTV(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_PUSHPAGE);
 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;
 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         buf = kmem_alloc(MAXNAMELEN, KM_SLEEP);
 317         err = getcomponent(name, buf, &next);
 318         if (err)
 319                 goto error;
 320         if (spa == NULL) {
 321                 err = spa_open(buf, &spa, FTAG);
 322                 if (err) {
 323                         dprintf("spa_open(%s) failed\n", buf);
 324                         goto error;
 325                 }
 326                 openedspa = TRUE;
 327
 328                 /* XXX this assertion belongs in spa_open */
 329                 ASSERT(!dsl_pool_sync_context(spa_get_dsl(spa)));
 330         }
 331
 332         dp = spa_get_dsl(spa);
 333
 334         rw_enter(&dp->dp_config_rwlock, RW_READER);
 335         err = dsl_dir_open_obj(dp, dp->dp_root_dir_obj, NULL, tag, &dd);
 336         if (err) {
 337                 rw_exit(&dp->dp_config_rwlock);
 338                 if (openedspa)
 339                         spa_close(spa, FTAG);
 340                 goto error;
 341         }
 342
 343         while (next != NULL) {
 344                 dsl_dir_t *child_ds;
 345                 err = getcomponent(next, buf, &nextnext);
 346                 if (err)
 347                         break;
 348                 ASSERT(next[0] != '\0');
 349                 if (next[0] == '@')
 350                         break;
 351                 dprintf("looking up %s in obj%lld\n",
 352                     buf, dd->dd_phys->dd_child_dir_zapobj);
 353
 354                 err = zap_lookup(dp->dp_meta_objset,
 355                     dd->dd_phys->dd_child_dir_zapobj,
 356                     buf, sizeof (ddobj), 1, &ddobj);
 357                 if (err) {
 358                         if (err == ENOENT)
 359                                 err = 0;
 360                         break;
 361                 }
 362
 363                 err = dsl_dir_open_obj(dp, ddobj, buf, tag, &child_ds);
 364                 if (err)
 365                         break;
 366                 dsl_dir_close(dd, tag);
 367                 dd = child_ds;
 368                 next = nextnext;
 369         }
 370         rw_exit(&dp->dp_config_rwlock);
 371
 372         if (err) {
 373                 dsl_dir_close(dd, tag);
 374                 if (openedspa)
 375                         spa_close(spa, FTAG);
 376                 goto error;
 377         }
 378
 379         /*
 380          * It's an error if there's more than one component left, or
 381          * tailp==NULL and there's any component left.
 382          */
 383         if (next != NULL &&
 384             (tailp == NULL || (nextnext && nextnext[0] != '\0'))) {
 385                 /* bad path name */
 386                 dsl_dir_close(dd, tag);
 387                 dprintf("next=%p (%s) tail=%p\n", next, next?next:"", tailp);
 388                 err = ENOENT;
 389         }
 390         if (tailp)
 391                 *tailp = next;
 392         if (openedspa)
 393                 spa_close(spa, FTAG);
 394         *ddp = dd;
 395 error:
 396         kmem_free(buf, MAXNAMELEN);
 397         return (err);
 398 }
 399
 400 /*
 401  * Return the dsl_dir_t, and possibly the last component which couldn't
 402  * be found in *tail.  Return NULL if the path is bogus, or if
 403  * tail==NULL and we couldn't parse the whole name.  (*tail)[0] == '@'
 404  * means that the last component is a snapshot.
 405  */
 406 int
 407 dsl_dir_open(const char *name, void *tag, dsl_dir_t **ddp, const char **tailp)
 408 {
 409         return (dsl_dir_open_spa(NULL, name, tag, ddp, tailp));
 410 }
 411
 412 uint64_t
 413 dsl_dir_create_sync(dsl_pool_t *dp, dsl_dir_t *pds, const char *name,
 414     dmu_tx_t *tx)
 415 {
 416         objset_t *mos = dp->dp_meta_objset;
 417         uint64_t ddobj;
 418         dsl_dir_phys_t *ddphys;
 419         dmu_buf_t *dbuf;
 420
 421         ddobj = dmu_object_alloc(mos, DMU_OT_DSL_DIR, 0,
 422             DMU_OT_DSL_DIR, sizeof (dsl_dir_phys_t), tx);
 423         if (pds) {
 424                 VERIFY(0 == zap_add(mos, pds->dd_phys->dd_child_dir_zapobj,
 425                     name, sizeof (uint64_t), 1, &ddobj, tx));
 426         } else {
 427                 /* it's the root dir */
 428                 VERIFY(0 == zap_add(mos, DMU_POOL_DIRECTORY_OBJECT,
 429                     DMU_POOL_ROOT_DATASET, sizeof (uint64_t), 1, &ddobj, tx));
 430         }
 431         VERIFY(0 == dmu_bonus_hold(mos, ddobj, FTAG, &dbuf));
 432         dmu_buf_will_dirty(dbuf, tx);
 433         ddphys = dbuf->db_data;
 434
 435         ddphys->dd_creation_time = gethrestime_sec();
 436         if (pds)
 437                 ddphys->dd_parent_obj = pds->dd_object;
 438         ddphys->dd_props_zapobj = zap_create(mos,
 439             DMU_OT_DSL_PROPS, DMU_OT_NONE, 0, tx);
 440         ddphys->dd_child_dir_zapobj = zap_create(mos,
 441             DMU_OT_DSL_DIR_CHILD_MAP, DMU_OT_NONE, 0, tx);
 442         if (spa_version(dp->dp_spa) >= SPA_VERSION_USED_BREAKDOWN)
 443                 ddphys->dd_flags |= DD_FLAG_USED_BREAKDOWN;
 444         dmu_buf_rele(dbuf, FTAG);
 445
 446         return (ddobj);
 447 }
 448
 449 /* ARGSUSED */
 450 int
 451 dsl_dir_destroy_check(void *arg1, void *arg2, dmu_tx_t *tx)
 452 {
 453         dsl_dataset_t *ds = arg1;
 454         dsl_dir_t *dd = ds->ds_dir;
 455         dsl_pool_t *dp = dd->dd_pool;
 456         objset_t *mos = dp->dp_meta_objset;
 457         int err;
 458         uint64_t count;
 459
 460         /*
 461          * There should be exactly two holds, both from
 462          * dsl_dataset_destroy: one on the dd directory, and one on its
 463          * head ds.  Otherwise, someone is trying to lookup something
 464          * inside this dir while we want to destroy it.  The
 465          * config_rwlock ensures that nobody else opens it after we
 466          * check.
 467          */
 468         if (dmu_buf_refcount(dd->dd_dbuf) > 2)
 469                 return (EBUSY);
 470
 471         err = zap_count(mos, dd->dd_phys->dd_child_dir_zapobj, &count);
 472         if (err)
 473                 return (err);
 474         if (count != 0)
 475                 return (EEXIST);
 476
 477         return (0);
 478 }
 479
 480 void
 481 dsl_dir_destroy_sync(void *arg1, void *tag, dmu_tx_t *tx)
 482 {
 483         dsl_dataset_t *ds = arg1;
 484         dsl_dir_t *dd = ds->ds_dir;
 485         objset_t *mos = dd->dd_pool->dp_meta_objset;
 486         dsl_prop_setarg_t psa;
 487         uint64_t value = 0;
 488         uint64_t obj;
 489         dd_used_t t;
 490
 491         ASSERT(RW_WRITE_HELD(&dd->dd_pool->dp_config_rwlock));
 492         ASSERT(dd->dd_phys->dd_head_dataset_obj == 0);
 493
 494         /* Remove our reservation. */
 495         dsl_prop_setarg_init_uint64(&psa, "reservation",
 496             (ZPROP_SRC_NONE | ZPROP_SRC_LOCAL | ZPROP_SRC_RECEIVED),
 497             &value);
 498         psa.psa_effective_value = 0;    /* predict default value */
 499
 500         dsl_dir_set_reservation_sync(ds, &psa, tx);
 501
 502         ASSERT3U(dd->dd_phys->dd_used_bytes, ==, 0);
 503         ASSERT3U(dd->dd_phys->dd_reserved, ==, 0);
 504         for (t = 0; t < DD_USED_NUM; t++)
 505                 ASSERT3U(dd->dd_phys->dd_used_breakdown[t], ==, 0);
 506
 507         VERIFY(0 == zap_destroy(mos, dd->dd_phys->dd_child_dir_zapobj, tx));
 508         VERIFY(0 == zap_destroy(mos, dd->dd_phys->dd_props_zapobj, tx));
 509         VERIFY(0 == dsl_deleg_destroy(mos, dd->dd_phys->dd_deleg_zapobj, tx));
 510         VERIFY(0 == zap_remove(mos,
 511             dd->dd_parent->dd_phys->dd_child_dir_zapobj, dd->dd_myname, tx));
 512
 513         obj = dd->dd_object;
 514         dsl_dir_close(dd, tag);
 515         VERIFY(0 == dmu_object_free(mos, obj, tx));
 516 }
 517
 518 boolean_t
 519 dsl_dir_is_clone(dsl_dir_t *dd)
 520 {
 521         return (dd->dd_phys->dd_origin_obj &&
 522             (dd->dd_pool->dp_origin_snap == NULL ||
 523             dd->dd_phys->dd_origin_obj !=
 524             dd->dd_pool->dp_origin_snap->ds_object));
 525 }
 526
 527 void
 528 dsl_dir_stats(dsl_dir_t *dd, nvlist_t *nv)
 529 {
 530         mutex_enter(&dd->dd_lock);
 531         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USED,
 532             dd->dd_phys->dd_used_bytes);
 533         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_QUOTA, dd->dd_phys->dd_quota);
 534         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_RESERVATION,
 535             dd->dd_phys->dd_reserved);
 536         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_COMPRESSRATIO,
 537             dd->dd_phys->dd_compressed_bytes == 0 ? 100 :
 538             (dd->dd_phys->dd_uncompressed_bytes * 100 /
 539             dd->dd_phys->dd_compressed_bytes));
 540         if (dd->dd_phys->dd_flags & DD_FLAG_USED_BREAKDOWN) {
 541                 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDSNAP,
 542                     dd->dd_phys->dd_used_breakdown[DD_USED_SNAP]);
 543                 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDDS,
 544                     dd->dd_phys->dd_used_breakdown[DD_USED_HEAD]);
 545                 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDREFRESERV,
 546                     dd->dd_phys->dd_used_breakdown[DD_USED_REFRSRV]);
 547                 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDCHILD,
 548                     dd->dd_phys->dd_used_breakdown[DD_USED_CHILD] +
 549                     dd->dd_phys->dd_used_breakdown[DD_USED_CHILD_RSRV]);
 550         }
 551         mutex_exit(&dd->dd_lock);
 552
 553         rw_enter(&dd->dd_pool->dp_config_rwlock, RW_READER);
 554         if (dsl_dir_is_clone(dd)) {
 555                 dsl_dataset_t *ds;
 556                 char buf[MAXNAMELEN];
 557
 558                 VERIFY(0 == dsl_dataset_hold_obj(dd->dd_pool,
 559                     dd->dd_phys->dd_origin_obj, FTAG, &ds));
 560                 dsl_dataset_name(ds, buf);
 561                 dsl_dataset_rele(ds, FTAG);
 562                 dsl_prop_nvlist_add_string(nv, ZFS_PROP_ORIGIN, buf);
 563         }
 564         rw_exit(&dd->dd_pool->dp_config_rwlock);
 565 }
 566
 567 void
 568 dsl_dir_dirty(dsl_dir_t *dd, dmu_tx_t *tx)
 569 {
 570         dsl_pool_t *dp = dd->dd_pool;
 571
 572         ASSERT(dd->dd_phys);
 573
 574         if (txg_list_add(&dp->dp_dirty_dirs, dd, tx->tx_txg) == 0) {
 575                 /* up the hold count until we can be written out */
 576                 dmu_buf_add_ref(dd->dd_dbuf, dd);
 577         }
 578 }
 579
 580 static int64_t
 581 parent_delta(dsl_dir_t *dd, uint64_t used, int64_t delta)
 582 {
 583         uint64_t old_accounted = MAX(used, dd->dd_phys->dd_reserved);
 584         uint64_t new_accounted = MAX(used + delta, dd->dd_phys->dd_reserved);
 585         return (new_accounted - old_accounted);
 586 }
 587
 588 void
 589 dsl_dir_sync(dsl_dir_t *dd, dmu_tx_t *tx)
 590 {
 591         ASSERT(dmu_tx_is_syncing(tx));
 592
 593         dmu_buf_will_dirty(dd->dd_dbuf, tx);
 594
 595         mutex_enter(&dd->dd_lock);
 596         ASSERT3U(dd->dd_tempreserved[tx->tx_txg&TXG_MASK], ==, 0);
 597         dprintf_dd(dd, "txg=%llu towrite=%lluK\n", tx->tx_txg,
 598             dd->dd_space_towrite[tx->tx_txg&TXG_MASK] / 1024);
 599         dd->dd_space_towrite[tx->tx_txg&TXG_MASK] = 0;
 600         mutex_exit(&dd->dd_lock);
 601
 602         /* release the hold from dsl_dir_dirty */
 603         dmu_buf_rele(dd->dd_dbuf, dd);
 604 }
 605
 606 static uint64_t
 607 dsl_dir_space_towrite(dsl_dir_t *dd)
 608 {
 609         uint64_t space = 0;
 610         int i;
 611
 612         ASSERT(MUTEX_HELD(&dd->dd_lock));
 613
 614         for (i = 0; i < TXG_SIZE; i++) {
 615                 space += dd->dd_space_towrite[i&TXG_MASK];
 616                 ASSERT3U(dd->dd_space_towrite[i&TXG_MASK], >=, 0);
 617         }
 618         return (space);
 619 }
 620
 621 /*
 622  * How much space would dd have available if ancestor had delta applied
 623  * to it?  If ondiskonly is set, we're only interested in what's
 624  * on-disk, not estimated pending changes.
 625  */
 626 uint64_t
 627 dsl_dir_space_available(dsl_dir_t *dd,
 628     dsl_dir_t *ancestor, int64_t delta, int ondiskonly)
 629 {
 630         uint64_t parentspace, myspace, quota, used;
 631
 632         /*
 633          * If there are no restrictions otherwise, assume we have
 634          * unlimited space available.
 635          */
 636         quota = UINT64_MAX;
 637         parentspace = UINT64_MAX;
 638
 639         if (dd->dd_parent != NULL) {
 640                 parentspace = dsl_dir_space_available(dd->dd_parent,
 641                     ancestor, delta, ondiskonly);
 642         }
 643
 644         mutex_enter(&dd->dd_lock);
 645         if (dd->dd_phys->dd_quota != 0)
 646                 quota = dd->dd_phys->dd_quota;
 647         used = dd->dd_phys->dd_used_bytes;
 648         if (!ondiskonly)
 649                 used += dsl_dir_space_towrite(dd);
 650
 651         if (dd->dd_parent == NULL) {
 652                 uint64_t poolsize = dsl_pool_adjustedsize(dd->dd_pool, FALSE);
 653                 quota = MIN(quota, poolsize);
 654         }
 655
 656         if (dd->dd_phys->dd_reserved > used && parentspace != UINT64_MAX) {
 657                 /*
 658                  * We have some space reserved, in addition to what our
 659                  * parent gave us.
 660                  */
 661                 parentspace += dd->dd_phys->dd_reserved - used;
 662         }
 663
 664         if (dd == ancestor) {
 665                 ASSERT(delta <= 0);
 666                 ASSERT(used >= -delta);
 667                 used += delta;
 668                 if (parentspace != UINT64_MAX)
 669                         parentspace -= delta;
 670         }
 671
 672         if (used > quota) {
 673                 /* over quota */
 674                 myspace = 0;
 675         } else {
 676                 /*
 677                  * the lesser of the space provided by our parent and
 678                  * the space left in our quota
 679                  */
 680                 myspace = MIN(parentspace, quota - used);
 681         }
 682
 683         mutex_exit(&dd->dd_lock);
 684
 685         return (myspace);
 686 }
 687
 688 struct tempreserve {
 689         list_node_t tr_node;
 690         dsl_pool_t *tr_dp;
 691         dsl_dir_t *tr_ds;
 692         uint64_t tr_size;
 693 };
 694
 695 static int
 696 dsl_dir_tempreserve_impl(dsl_dir_t *dd, uint64_t asize, boolean_t netfree,
 697     boolean_t ignorequota, boolean_t checkrefquota, list_t *tr_list,
 698     dmu_tx_t *tx, boolean_t first)
 699 {
 700         uint64_t txg = tx->tx_txg;
 701         uint64_t est_inflight, used_on_disk, quota, parent_rsrv;
 702         uint64_t deferred = 0;
 703         struct tempreserve *tr;
 704         int retval = EDQUOT;
 705         int txgidx = txg & TXG_MASK;
 706         int i;
 707         uint64_t ref_rsrv = 0;
 708
 709         ASSERT3U(txg, !=, 0);
 710         ASSERT3S(asize, >, 0);
 711
 712         mutex_enter(&dd->dd_lock);
 713
 714         /*
 715          * Check against the dsl_dir's quota.  We don't add in the delta
 716          * when checking for over-quota because they get one free hit.
 717          */
 718         est_inflight = dsl_dir_space_towrite(dd);
 719         for (i = 0; i < TXG_SIZE; i++)
 720                 est_inflight += dd->dd_tempreserved[i];
 721         used_on_disk = dd->dd_phys->dd_used_bytes;
 722
 723         /*
 724          * On the first iteration, fetch the dataset's used-on-disk and
 725          * refreservation values. Also, if checkrefquota is set, test if
 726          * allocating this space would exceed the dataset's refquota.
 727          */
 728         if (first && tx->tx_objset) {
 729                 int error;
 730                 dsl_dataset_t *ds = tx->tx_objset->os_dsl_dataset;
 731
 732                 error = dsl_dataset_check_quota(ds, checkrefquota,
 733                     asize, est_inflight, &used_on_disk, &ref_rsrv);
 734                 if (error) {
 735                         mutex_exit(&dd->dd_lock);
 736                         return (error);
 737                 }
 738         }
 739
 740         /*
 741          * If this transaction will result in a net free of space,
 742          * we want to let it through.
 743          */
 744         if (ignorequota || netfree || dd->dd_phys->dd_quota == 0)
 745                 quota = UINT64_MAX;
 746         else
 747                 quota = dd->dd_phys->dd_quota;
 748
 749         /*
 750          * Adjust the quota against the actual pool size at the root
 751          * minus any outstanding deferred frees.
 752          * To ensure that it's possible to remove files from a full
 753          * pool without inducing transient overcommits, we throttle
 754          * netfree transactions against a quota that is slightly larger,
 755          * but still within the pool's allocation slop.  In cases where
 756          * we're very close to full, this will allow a steady trickle of
 757          * removes to get through.
 758          */
 759         if (dd->dd_parent == NULL) {
 760                 spa_t *spa = dd->dd_pool->dp_spa;
 761                 uint64_t poolsize = dsl_pool_adjustedsize(dd->dd_pool, netfree);
 762                 deferred = metaslab_class_get_deferred(spa_normal_class(spa));
 763                 if (poolsize - deferred < quota) {
 764                         quota = poolsize - deferred;
 765                         retval = ENOSPC;
 766                 }
 767         }
 768
 769         /*
 770          * If they are requesting more space, and our current estimate
 771          * is over quota, they get to try again unless the actual
 772          * on-disk is over quota and there are no pending changes (which
 773          * may free up space for us).
 774          */
 775         if (used_on_disk + est_inflight >= quota) {
 776                 if (est_inflight > 0 || used_on_disk < quota ||
 777                     (retval == ENOSPC && used_on_disk < quota + deferred))
 778                         retval = ERESTART;
 779                 dprintf_dd(dd, "failing: used=%lluK inflight = %lluK "
 780                     "quota=%lluK tr=%lluK err=%d\n",
 781                     used_on_disk>>10, est_inflight>>10,
 782                     quota>>10, asize>>10, retval);
 783                 mutex_exit(&dd->dd_lock);
 784                 return (retval);
 785         }
 786
 787         /* We need to up our estimated delta before dropping dd_lock */
 788         dd->dd_tempreserved[txgidx] += asize;
 789
 790         parent_rsrv = parent_delta(dd, used_on_disk + est_inflight,
 791             asize - ref_rsrv);
 792         mutex_exit(&dd->dd_lock);
 793
 794         tr = kmem_zalloc(sizeof (struct tempreserve), KM_PUSHPAGE);
 795         tr->tr_ds = dd;
 796         tr->tr_size = asize;
 797         list_insert_tail(tr_list, tr);
 798
 799         /* see if it's OK with our parent */
 800         if (dd->dd_parent && parent_rsrv) {
 801                 boolean_t ismos = (dd->dd_phys->dd_head_dataset_obj == 0);
 802
 803                 return (dsl_dir_tempreserve_impl(dd->dd_parent,
 804                     parent_rsrv, netfree, ismos, TRUE, tr_list, tx, FALSE));
 805         } else {
 806                 return (0);
 807         }
 808 }
 809
 810 /*
 811  * Reserve space in this dsl_dir, to be used in this tx's txg.
 812  * After the space has been dirtied (and dsl_dir_willuse_space()
 813  * has been called), the reservation should be canceled, using
 814  * dsl_dir_tempreserve_clear().
 815  */
 816 int
 817 dsl_dir_tempreserve_space(dsl_dir_t *dd, uint64_t lsize, uint64_t asize,
 818     uint64_t fsize, uint64_t usize, void **tr_cookiep, dmu_tx_t *tx)
 819 {
 820         int err;
 821         list_t *tr_list;
 822
 823         if (asize == 0) {
 824                 *tr_cookiep = NULL;
 825                 return (0);
 826         }
 827
 828         tr_list = kmem_alloc(sizeof (list_t), KM_PUSHPAGE);
 829         list_create(tr_list, sizeof (struct tempreserve),
 830             offsetof(struct tempreserve, tr_node));
 831         ASSERT3S(asize, >, 0);
 832         ASSERT3S(fsize, >=, 0);
 833
 834         err = arc_tempreserve_space(lsize, tx->tx_txg);
 835         if (err == 0) {
 836                 struct tempreserve *tr;
 837
 838                 tr = kmem_zalloc(sizeof (struct tempreserve), KM_PUSHPAGE);
 839                 tr->tr_size = lsize;
 840                 list_insert_tail(tr_list, tr);
 841
 842                 err = dsl_pool_tempreserve_space(dd->dd_pool, asize, tx);
 843         } else {
 844                 if (err == EAGAIN) {
 845                         txg_delay(dd->dd_pool, tx->tx_txg, 1);
 846                         err = ERESTART;
 847                 }
 848                 dsl_pool_memory_pressure(dd->dd_pool);
 849         }
 850
 851         if (err == 0) {
 852                 struct tempreserve *tr;
 853
 854                 tr = kmem_zalloc(sizeof (struct tempreserve), KM_PUSHPAGE);
 855                 tr->tr_dp = dd->dd_pool;
 856                 tr->tr_size = asize;
 857                 list_insert_tail(tr_list, tr);
 858
 859                 err = dsl_dir_tempreserve_impl(dd, asize, fsize >= asize,
 860                     FALSE, asize > usize, tr_list, tx, TRUE);
 861         }
 862
 863         if (err)
 864                 dsl_dir_tempreserve_clear(tr_list, tx);
 865         else
 866                 *tr_cookiep = tr_list;
 867
 868         return (err);
 869 }
 870
 871 /*
 872  * Clear a temporary reservation that we previously made with
 873  * dsl_dir_tempreserve_space().
 874  */
 875 void
 876 dsl_dir_tempreserve_clear(void *tr_cookie, dmu_tx_t *tx)
 877 {
 878         int txgidx = tx->tx_txg & TXG_MASK;
 879         list_t *tr_list = tr_cookie;
 880         struct tempreserve *tr;
 881
 882         ASSERT3U(tx->tx_txg, !=, 0);
 883
 884         if (tr_cookie == NULL)
 885                 return;
 886
 887         while ((tr = list_head(tr_list))) {
 888                 if (tr->tr_dp) {
 889                         dsl_pool_tempreserve_clear(tr->tr_dp, tr->tr_size, tx);
 890                 } else if (tr->tr_ds) {
 891                         mutex_enter(&tr->tr_ds->dd_lock);
 892                         ASSERT3U(tr->tr_ds->dd_tempreserved[txgidx], >=,
 893                             tr->tr_size);
 894                         tr->tr_ds->dd_tempreserved[txgidx] -= tr->tr_size;
 895                         mutex_exit(&tr->tr_ds->dd_lock);
 896                 } else {
 897                         arc_tempreserve_clear(tr->tr_size);
 898                 }
 899                 list_remove(tr_list, tr);
 900                 kmem_free(tr, sizeof (struct tempreserve));
 901         }
 902
 903         kmem_free(tr_list, sizeof (list_t));
 904 }
 905
 906 static void
 907 dsl_dir_willuse_space_impl(dsl_dir_t *dd, int64_t space, dmu_tx_t *tx)
 908 {
 909         int64_t parent_space;
 910         uint64_t est_used;
 911
 912         mutex_enter(&dd->dd_lock);
 913         if (space > 0)
 914                 dd->dd_space_towrite[tx->tx_txg & TXG_MASK] += space;
 915
 916         est_used = dsl_dir_space_towrite(dd) + dd->dd_phys->dd_used_bytes;
 917         parent_space = parent_delta(dd, est_used, space);
 918         mutex_exit(&dd->dd_lock);
 919
 920         /* Make sure that we clean up dd_space_to* */
 921         dsl_dir_dirty(dd, tx);
 922
 923         /* XXX this is potentially expensive and unnecessary... */
 924         if (parent_space && dd->dd_parent)
 925                 dsl_dir_willuse_space_impl(dd->dd_parent, parent_space, tx);
 926 }
 927
 928 /*
 929  * Call in open context when we think we're going to write/free space,
 930  * eg. when dirtying data.  Be conservative (ie. OK to write less than
 931  * this or free more than this, but don't write more or free less).
 932  */
 933 void
 934 dsl_dir_willuse_space(dsl_dir_t *dd, int64_t space, dmu_tx_t *tx)
 935 {
 936         dsl_pool_willuse_space(dd->dd_pool, space, tx);
 937         dsl_dir_willuse_space_impl(dd, space, tx);
 938 }
 939
 940 /* call from syncing context when we actually write/free space for this dd */
 941 void
 942 dsl_dir_diduse_space(dsl_dir_t *dd, dd_used_t type,
 943     int64_t used, int64_t compressed, int64_t uncompressed, dmu_tx_t *tx)
 944 {
 945         int64_t accounted_delta;
 946         boolean_t needlock = !MUTEX_HELD(&dd->dd_lock);
 947
 948         ASSERT(dmu_tx_is_syncing(tx));
 949         ASSERT(type < DD_USED_NUM);
 950
 951         dsl_dir_dirty(dd, tx);
 952
 953         if (needlock)
 954                 mutex_enter(&dd->dd_lock);
 955         accounted_delta = parent_delta(dd, dd->dd_phys->dd_used_bytes, used);
 956         ASSERT(used >= 0 || dd->dd_phys->dd_used_bytes >= -used);
 957         ASSERT(compressed >= 0 ||
 958             dd->dd_phys->dd_compressed_bytes >= -compressed);
 959         ASSERT(uncompressed >= 0 ||
 960             dd->dd_phys->dd_uncompressed_bytes >= -uncompressed);
 961         dd->dd_phys->dd_used_bytes += used;
 962         dd->dd_phys->dd_uncompressed_bytes += uncompressed;
 963         dd->dd_phys->dd_compressed_bytes += compressed;
 964
 965         if (dd->dd_phys->dd_flags & DD_FLAG_USED_BREAKDOWN) {
 966                 ASSERT(used > 0 ||
 967                     dd->dd_phys->dd_used_breakdown[type] >= -used);
 968                 dd->dd_phys->dd_used_breakdown[type] += used;
 969 #ifdef DEBUG
 970                 {
 971                         dd_used_t t;
 972                         uint64_t u = 0;
 973                         for (t = 0; t < DD_USED_NUM; t++)
 974                                 u += dd->dd_phys->dd_used_breakdown[t];
 975                         ASSERT3U(u, ==, dd->dd_phys->dd_used_bytes);
 976                 }
 977 #endif
 978         }
 979         if (needlock)
 980                 mutex_exit(&dd->dd_lock);
 981
 982         if (dd->dd_parent != NULL) {
 983                 dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD,
 984                     accounted_delta, compressed, uncompressed, tx);
 985                 dsl_dir_transfer_space(dd->dd_parent,
 986                     used - accounted_delta,
 987                     DD_USED_CHILD_RSRV, DD_USED_CHILD, tx);
 988         }
 989 }
 990
 991 void
 992 dsl_dir_transfer_space(dsl_dir_t *dd, int64_t delta,
 993     dd_used_t oldtype, dd_used_t newtype, dmu_tx_t *tx)
 994 {
 995         boolean_t needlock = !MUTEX_HELD(&dd->dd_lock);
 996
 997         ASSERT(dmu_tx_is_syncing(tx));
 998         ASSERT(oldtype < DD_USED_NUM);
 999         ASSERT(newtype < DD_USED_NUM);
1000
1001         if (delta == 0 || !(dd->dd_phys->dd_flags & DD_FLAG_USED_BREAKDOWN))
1002                 return;
1003
1004         dsl_dir_dirty(dd, tx);
1005         if (needlock)
1006                 mutex_enter(&dd->dd_lock);
1007         ASSERT(delta > 0 ?
1008             dd->dd_phys->dd_used_breakdown[oldtype] >= delta :
1009             dd->dd_phys->dd_used_breakdown[newtype] >= -delta);
1010         ASSERT(dd->dd_phys->dd_used_bytes >= ABS(delta));
1011         dd->dd_phys->dd_used_breakdown[oldtype] -= delta;
1012         dd->dd_phys->dd_used_breakdown[newtype] += delta;
1013         if (needlock)
1014                 mutex_exit(&dd->dd_lock);
1015 }
1016
1017 static int
1018 dsl_dir_set_quota_check(void *arg1, void *arg2, dmu_tx_t *tx)
1019 {
1020         dsl_dataset_t *ds = arg1;
1021         dsl_dir_t *dd = ds->ds_dir;
1022         dsl_prop_setarg_t *psa = arg2;
1023         int err;
1024         uint64_t towrite;
1025
1026         if ((err = dsl_prop_predict_sync(ds->ds_dir, psa)) != 0)
1027                 return (err);
1028
1029         if (psa->psa_effective_value == 0)
1030                 return (0);
1031
1032         mutex_enter(&dd->dd_lock);
1033         /*
1034          * If we are doing the preliminary check in open context, and
1035          * there are pending changes, then don't fail it, since the
1036          * pending changes could under-estimate the amount of space to be
1037          * freed up.
1038          */
1039         towrite = dsl_dir_space_towrite(dd);
1040         if ((dmu_tx_is_syncing(tx) || towrite == 0) &&
1041             (psa->psa_effective_value < dd->dd_phys->dd_reserved ||
1042             psa->psa_effective_value < dd->dd_phys->dd_used_bytes + towrite)) {
1043                 err = ENOSPC;
1044         }
1045         mutex_exit(&dd->dd_lock);
1046         return (err);
1047 }
1048
1049 extern dsl_syncfunc_t dsl_prop_set_sync;
1050
1051 static void
1052 dsl_dir_set_quota_sync(void *arg1, void *arg2, dmu_tx_t *tx)
1053 {
1054         dsl_dataset_t *ds = arg1;
1055         dsl_dir_t *dd = ds->ds_dir;
1056         dsl_prop_setarg_t *psa = arg2;
1057         uint64_t effective_value = psa->psa_effective_value;
1058
1059         dsl_prop_set_sync(ds, psa, tx);
1060         DSL_PROP_CHECK_PREDICTION(dd, psa);
1061
1062         dmu_buf_will_dirty(dd->dd_dbuf, tx);
1063
1064         mutex_enter(&dd->dd_lock);
1065         dd->dd_phys->dd_quota = effective_value;
1066         mutex_exit(&dd->dd_lock);
1067
1068         spa_history_log_internal(LOG_DS_QUOTA, dd->dd_pool->dp_spa,
1069             tx, "%lld dataset = %llu ",
1070             (longlong_t)effective_value, dd->dd_phys->dd_head_dataset_obj);
1071 }
1072
1073 int
1074 dsl_dir_set_quota(const char *ddname, zprop_source_t source, uint64_t quota)
1075 {
1076         dsl_dir_t *dd;
1077         dsl_dataset_t *ds;
1078         dsl_prop_setarg_t psa;
1079         int err;
1080
1081         dsl_prop_setarg_init_uint64(&psa, "quota", source, &quota);
1082
1083         err = dsl_dataset_hold(ddname, FTAG, &ds);
1084         if (err)
1085                 return (err);
1086
1087         err = dsl_dir_open(ddname, FTAG, &dd, NULL);
1088         if (err) {
1089                 dsl_dataset_rele(ds, FTAG);
1090                 return (err);
1091         }
1092
1093         ASSERT(ds->ds_dir == dd);
1094
1095         /*
1096          * If someone removes a file, then tries to set the quota, we want to
1097          * make sure the file freeing takes effect.
1098          */
1099         txg_wait_open(dd->dd_pool, 0);
1100
1101         err = dsl_sync_task_do(dd->dd_pool, dsl_dir_set_quota_check,
1102             dsl_dir_set_quota_sync, ds, &psa, 0);
1103
1104         dsl_dir_close(dd, FTAG);
1105         dsl_dataset_rele(ds, FTAG);
1106         return (err);
1107 }
1108
1109 int
1110 dsl_dir_set_reservation_check(void *arg1, void *arg2, dmu_tx_t *tx)
1111 {
1112         dsl_dataset_t *ds = arg1;
1113         dsl_dir_t *dd = ds->ds_dir;
1114         dsl_prop_setarg_t *psa = arg2;
1115         uint64_t effective_value;
1116         uint64_t used, avail;
1117         int err;
1118
1119         if ((err = dsl_prop_predict_sync(ds->ds_dir, psa)) != 0)
1120                 return (err);
1121
1122         effective_value = psa->psa_effective_value;
1123
1124         /*
1125          * If we are doing the preliminary check in open context, the
1126          * space estimates may be inaccurate.
1127          */
1128         if (!dmu_tx_is_syncing(tx))
1129                 return (0);
1130
1131         mutex_enter(&dd->dd_lock);
1132         used = dd->dd_phys->dd_used_bytes;
1133         mutex_exit(&dd->dd_lock);
1134
1135         if (dd->dd_parent) {
1136                 avail = dsl_dir_space_available(dd->dd_parent,
1137                     NULL, 0, FALSE);
1138         } else {
1139                 avail = dsl_pool_adjustedsize(dd->dd_pool, B_FALSE) - used;
1140         }
1141
1142         if (MAX(used, effective_value) > MAX(used, dd->dd_phys->dd_reserved)) {
1143                 uint64_t delta = MAX(used, effective_value) -
1144                     MAX(used, dd->dd_phys->dd_reserved);
1145
1146                 if (delta > avail)
1147                         return (ENOSPC);
1148                 if (dd->dd_phys->dd_quota > 0 &&
1149                     effective_value > dd->dd_phys->dd_quota)
1150                         return (ENOSPC);
1151         }
1152
1153         return (0);
1154 }
1155
1156 static void
1157 dsl_dir_set_reservation_sync(void *arg1, void *arg2, dmu_tx_t *tx)
1158 {
1159         dsl_dataset_t *ds = arg1;
1160         dsl_dir_t *dd = ds->ds_dir;
1161         dsl_prop_setarg_t *psa = arg2;
1162         uint64_t effective_value = psa->psa_effective_value;
1163         uint64_t used;
1164         int64_t delta;
1165
1166         dsl_prop_set_sync(ds, psa, tx);
1167         DSL_PROP_CHECK_PREDICTION(dd, psa);
1168
1169         dmu_buf_will_dirty(dd->dd_dbuf, tx);
1170
1171         mutex_enter(&dd->dd_lock);
1172         used = dd->dd_phys->dd_used_bytes;
1173         delta = MAX(used, effective_value) -
1174             MAX(used, dd->dd_phys->dd_reserved);
1175         dd->dd_phys->dd_reserved = effective_value;
1176
1177         if (dd->dd_parent != NULL) {
1178                 /* Roll up this additional usage into our ancestors */
1179                 dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD_RSRV,
1180                     delta, 0, 0, tx);
1181         }
1182         mutex_exit(&dd->dd_lock);
1183
1184         spa_history_log_internal(LOG_DS_RESERVATION, dd->dd_pool->dp_spa,
1185             tx, "%lld dataset = %llu",
1186             (longlong_t)effective_value, dd->dd_phys->dd_head_dataset_obj);
1187 }
1188
1189 int
1190 dsl_dir_set_reservation(const char *ddname, zprop_source_t source,
1191     uint64_t reservation)
1192 {
1193         dsl_dir_t *dd;
1194         dsl_dataset_t *ds;
1195         dsl_prop_setarg_t psa;
1196         int err;
1197
1198         dsl_prop_setarg_init_uint64(&psa, "reservation", source, &reservation);
1199
1200         err = dsl_dataset_hold(ddname, FTAG, &ds);
1201         if (err)
1202                 return (err);
1203
1204         err = dsl_dir_open(ddname, FTAG, &dd, NULL);
1205         if (err) {
1206                 dsl_dataset_rele(ds, FTAG);
1207                 return (err);
1208         }
1209
1210         ASSERT(ds->ds_dir == dd);
1211
1212         err = dsl_sync_task_do(dd->dd_pool, dsl_dir_set_reservation_check,
1213             dsl_dir_set_reservation_sync, ds, &psa, 0);
1214
1215         dsl_dir_close(dd, FTAG);
1216         dsl_dataset_rele(ds, FTAG);
1217         return (err);
1218 }
1219
1220 static dsl_dir_t *
1221 closest_common_ancestor(dsl_dir_t *ds1, dsl_dir_t *ds2)
1222 {
1223         for (; ds1; ds1 = ds1->dd_parent) {
1224                 dsl_dir_t *dd;
1225                 for (dd = ds2; dd; dd = dd->dd_parent) {
1226                         if (ds1 == dd)
1227                                 return (dd);
1228                 }
1229         }
1230         return (NULL);
1231 }
1232
1233 /*
1234  * If delta is applied to dd, how much of that delta would be applied to
1235  * ancestor?  Syncing context only.
1236  */
1237 static int64_t
1238 would_change(dsl_dir_t *dd, int64_t delta, dsl_dir_t *ancestor)
1239 {
1240         if (dd == ancestor)
1241                 return (delta);
1242
1243         mutex_enter(&dd->dd_lock);
1244         delta = parent_delta(dd, dd->dd_phys->dd_used_bytes, delta);
1245         mutex_exit(&dd->dd_lock);
1246         return (would_change(dd->dd_parent, delta, ancestor));
1247 }
1248
1249 struct renamearg {
1250         dsl_dir_t *newparent;
1251         const char *mynewname;
1252 };
1253
1254 static int
1255 dsl_dir_rename_check(void *arg1, void *arg2, dmu_tx_t *tx)
1256 {
1257         dsl_dir_t *dd = arg1;
1258         struct renamearg *ra = arg2;
1259         dsl_pool_t *dp = dd->dd_pool;
1260         objset_t *mos = dp->dp_meta_objset;
1261         int err;
1262         uint64_t val;
1263
1264         /*
1265          * There should only be one reference, from dmu_objset_rename().
1266          * Fleeting holds are also possible (eg, from "zfs list" getting
1267          * stats), but any that are present in open context will likely
1268          * be gone by syncing context, so only fail from syncing
1269          * context.
1270          */
1271         if (dmu_tx_is_syncing(tx) && dmu_buf_refcount(dd->dd_dbuf) > 1)
1272                 return (EBUSY);
1273
1274         /* check for existing name */
1275         err = zap_lookup(mos, ra->newparent->dd_phys->dd_child_dir_zapobj,
1276             ra->mynewname, 8, 1, &val);
1277         if (err == 0)
1278                 return (EEXIST);
1279         if (err != ENOENT)
1280                 return (err);
1281
1282         if (ra->newparent != dd->dd_parent) {
1283                 /* is there enough space? */
1284                 uint64_t myspace =
1285                     MAX(dd->dd_phys->dd_used_bytes, dd->dd_phys->dd_reserved);
1286
1287                 /* no rename into our descendant */
1288                 if (closest_common_ancestor(dd, ra->newparent) == dd)
1289                         return (EINVAL);
1290
1291                 if ((err = dsl_dir_transfer_possible(dd->dd_parent,
1292                     ra->newparent, myspace)))
1293                         return (err);
1294         }
1295
1296         return (0);
1297 }
1298
1299 static void
1300 dsl_dir_rename_sync(void *arg1, void *arg2, dmu_tx_t *tx)
1301 {
1302         dsl_dir_t *dd = arg1;
1303         struct renamearg *ra = arg2;
1304         dsl_pool_t *dp = dd->dd_pool;
1305         objset_t *mos = dp->dp_meta_objset;
1306         int err;
1307
1308         ASSERT(dmu_buf_refcount(dd->dd_dbuf) <= 2);
1309
1310         if (ra->newparent != dd->dd_parent) {
1311                 dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD,
1312                     -dd->dd_phys->dd_used_bytes,
1313                     -dd->dd_phys->dd_compressed_bytes,
1314                     -dd->dd_phys->dd_uncompressed_bytes, tx);
1315                 dsl_dir_diduse_space(ra->newparent, DD_USED_CHILD,
1316                     dd->dd_phys->dd_used_bytes,
1317                     dd->dd_phys->dd_compressed_bytes,
1318                     dd->dd_phys->dd_uncompressed_bytes, tx);
1319
1320                 if (dd->dd_phys->dd_reserved > dd->dd_phys->dd_used_bytes) {
1321                         uint64_t unused_rsrv = dd->dd_phys->dd_reserved -
1322                             dd->dd_phys->dd_used_bytes;
1323
1324                         dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD_RSRV,
1325                             -unused_rsrv, 0, 0, tx);
1326                         dsl_dir_diduse_space(ra->newparent, DD_USED_CHILD_RSRV,
1327                             unused_rsrv, 0, 0, tx);
1328                 }
1329         }
1330
1331         dmu_buf_will_dirty(dd->dd_dbuf, tx);
1332
1333         /* remove from old parent zapobj */
1334         err = zap_remove(mos, dd->dd_parent->dd_phys->dd_child_dir_zapobj,
1335             dd->dd_myname, tx);
1336         ASSERT3U(err, ==, 0);
1337
1338         (void) strcpy(dd->dd_myname, ra->mynewname);
1339         dsl_dir_close(dd->dd_parent, dd);
1340         dd->dd_phys->dd_parent_obj = ra->newparent->dd_object;
1341         VERIFY(0 == dsl_dir_open_obj(dd->dd_pool,
1342             ra->newparent->dd_object, NULL, dd, &dd->dd_parent));
1343
1344         /* add to new parent zapobj */
1345         err = zap_add(mos, ra->newparent->dd_phys->dd_child_dir_zapobj,
1346             dd->dd_myname, 8, 1, &dd->dd_object, tx);
1347         ASSERT3U(err, ==, 0);
1348
1349         spa_history_log_internal(LOG_DS_RENAME, dd->dd_pool->dp_spa,
1350             tx, "dataset = %llu", dd->dd_phys->dd_head_dataset_obj);
1351 }
1352
1353 int
1354 dsl_dir_rename(dsl_dir_t *dd, const char *newname)
1355 {
1356         struct renamearg ra;
1357         int err;
1358
1359         /* new parent should exist */
1360         err = dsl_dir_open(newname, FTAG, &ra.newparent, &ra.mynewname);
1361         if (err)
1362                 return (err);
1363
1364         /* can't rename to different pool */
1365         if (dd->dd_pool != ra.newparent->dd_pool) {
1366                 err = ENXIO;
1367                 goto out;
1368         }
1369
1370         /* new name should not already exist */
1371         if (ra.mynewname == NULL) {
1372                 err = EEXIST;
1373                 goto out;
1374         }
1375
1376         err = dsl_sync_task_do(dd->dd_pool,
1377             dsl_dir_rename_check, dsl_dir_rename_sync, dd, &ra, 3);
1378
1379 out:
1380         dsl_dir_close(ra.newparent, FTAG);
1381         return (err);
1382 }
1383
1384 int
1385 dsl_dir_transfer_possible(dsl_dir_t *sdd, dsl_dir_t *tdd, uint64_t space)
1386 {
1387         dsl_dir_t *ancestor;
1388         int64_t adelta;
1389         uint64_t avail;
1390
1391         ancestor = closest_common_ancestor(sdd, tdd);
1392         adelta = would_change(sdd, -space, ancestor);
1393         avail = dsl_dir_space_available(tdd, ancestor, adelta, FALSE);
1394         if (avail < space)
1395                 return (ENOSPC);
1396
1397         return (0);
1398 }
1399
1400 timestruc_t
1401 dsl_dir_snap_cmtime(dsl_dir_t *dd)
1402 {
1403         timestruc_t t;
1404
1405         mutex_enter(&dd->dd_lock);
1406         t = dd->dd_snap_cmtime;
1407         mutex_exit(&dd->dd_lock);
1408
1409         return (t);
1410 }
1411
1412 void
1413 dsl_dir_snap_cmtime_update(dsl_dir_t *dd)
1414 {
1415         timestruc_t t;
1416
1417         gethrestime(&t);
1418         mutex_enter(&dd->dd_lock);
1419         dd->dd_snap_cmtime = t;
1420         mutex_exit(&dd->dd_lock);
1421 }
1422
1423 #if defined(_KERNEL) && defined(HAVE_SPL)
1424 EXPORT_SYMBOL(dsl_dir_set_quota);
1425 EXPORT_SYMBOL(dsl_dir_set_reservation);
1426 EXPORT_SYMBOL(dsl_dir_open);
1427 EXPORT_SYMBOL(dsl_dir_close);
1428 #endif