sys/cddl/contrib/opensolaris/uts/common/fs/zfs/dsl_dir.c

   1 /*
   2  * CDDL HEADER START
   3  *
   4  * The contents of this file are subject to the terms of the
   5  * Common Development and Distribution License (the "License").
   6  * You may not use this file except in compliance with the License.
   7  *
   8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
   9  * or http://www.opensolaris.org/os/licensing.
  10  * See the License for the specific language governing permissions
  11  * and limitations under the License.
  12  *
  13  * When distributing Covered Code, include this CDDL HEADER in each
  14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
  15  * If applicable, add the following below this CDDL HEADER, with the
  16  * fields enclosed by brackets "[]" replaced with your own identifying
  17  * information: Portions Copyright [yyyy] [name of copyright owner]
  18  *
  19  * CDDL HEADER END
  20  */
  21 /*
  22  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
  23  * Copyright (c) 2011 Pawel Jakub Dawidek <pawel@dawidek.net>.
  24  * All rights reserved.
  25  */
  26
  27 #include <sys/dmu.h>
  28 #include <sys/dmu_objset.h>
  29 #include <sys/dmu_tx.h>
  30 #include <sys/dsl_dataset.h>
  31 #include <sys/dsl_dir.h>
  32 #include <sys/dsl_prop.h>
  33 #include <sys/dsl_synctask.h>
  34 #include <sys/dsl_deleg.h>
  35 #include <sys/spa.h>
  36 #include <sys/metaslab.h>
  37 #include <sys/zap.h>
  38 #include <sys/zio.h>
  39 #include <sys/arc.h>
  40 #include <sys/sunddi.h>
  41 #include <sys/zvol.h>
  42 #ifdef _KERNEL
  43 #include <sys/zfs_vfsops.h>
  44 #endif
  45 #include "zfs_namecheck.h"
  46
  47 static uint64_t dsl_dir_space_towrite(dsl_dir_t *dd);
  48 static void dsl_dir_set_reservation_sync(void *arg1, void *arg2, dmu_tx_t *tx);
  49
  50
  51 /* ARGSUSED */
  52 static void
  53 dsl_dir_evict(dmu_buf_t *db, void *arg)
  54 {
  55         dsl_dir_t *dd = arg;
  56         dsl_pool_t *dp = dd->dd_pool;
  57         int t;
  58
  59         for (t = 0; t < TXG_SIZE; t++) {
  60                 ASSERT(!txg_list_member(&dp->dp_dirty_dirs, dd, t));
  61                 ASSERT(dd->dd_tempreserved[t] == 0);
  62                 ASSERT(dd->dd_space_towrite[t] == 0);
  63         }
  64
  65         if (dd->dd_parent)
  66                 dsl_dir_close(dd->dd_parent, dd);
  67
  68         spa_close(dd->dd_pool->dp_spa, dd);
  69
  70         /*
  71          * The props callback list should have been cleaned up by
  72          * objset_evict().
  73          */
  74         list_destroy(&dd->dd_prop_cbs);
  75         mutex_destroy(&dd->dd_lock);
  76         kmem_free(dd, sizeof (dsl_dir_t));
  77 }
  78
  79 int
  80 dsl_dir_open_obj(dsl_pool_t *dp, uint64_t ddobj,
  81     const char *tail, void *tag, dsl_dir_t **ddp)
  82 {
  83         dmu_buf_t *dbuf;
  84         dsl_dir_t *dd;
  85         int err;
  86
  87         ASSERT(RW_LOCK_HELD(&dp->dp_config_rwlock) ||
  88             dsl_pool_sync_context(dp));
  89
  90         err = dmu_bonus_hold(dp->dp_meta_objset, ddobj, tag, &dbuf);
  91         if (err)
  92                 return (err);
  93         dd = dmu_buf_get_user(dbuf);
  94 #ifdef ZFS_DEBUG
  95         {
  96                 dmu_object_info_t doi;
  97                 dmu_object_info_from_db(dbuf, &doi);
  98                 ASSERT3U(doi.doi_type, ==, DMU_OT_DSL_DIR);
  99                 ASSERT3U(doi.doi_bonus_size, >=, sizeof (dsl_dir_phys_t));
 100         }
 101 #endif
 102         if (dd == NULL) {
 103                 dsl_dir_t *winner;
 104
 105                 dd = kmem_zalloc(sizeof (dsl_dir_t), KM_SLEEP);
 106                 dd->dd_object = ddobj;
 107                 dd->dd_dbuf = dbuf;
 108                 dd->dd_pool = dp;
 109                 dd->dd_phys = dbuf->db_data;
 110                 mutex_init(&dd->dd_lock, NULL, MUTEX_DEFAULT, NULL);
 111
 112                 list_create(&dd->dd_prop_cbs, sizeof (dsl_prop_cb_record_t),
 113                     offsetof(dsl_prop_cb_record_t, cbr_node));
 114
 115                 dsl_dir_snap_cmtime_update(dd);
 116
 117                 if (dd->dd_phys->dd_parent_obj) {
 118                         err = dsl_dir_open_obj(dp, dd->dd_phys->dd_parent_obj,
 119                             NULL, dd, &dd->dd_parent);
 120                         if (err)
 121                                 goto errout;
 122                         if (tail) {
 123 #ifdef ZFS_DEBUG
 124                                 uint64_t foundobj;
 125
 126                                 err = zap_lookup(dp->dp_meta_objset,
 127                                     dd->dd_parent->dd_phys->dd_child_dir_zapobj,
 128                                     tail, sizeof (foundobj), 1, &foundobj);
 129                                 ASSERT(err || foundobj == ddobj);
 130 #endif
 131                                 (void) strcpy(dd->dd_myname, tail);
 132                         } else {
 133                                 err = zap_value_search(dp->dp_meta_objset,
 134                                     dd->dd_parent->dd_phys->dd_child_dir_zapobj,
 135                                     ddobj, 0, dd->dd_myname);
 136                         }
 137                         if (err)
 138                                 goto errout;
 139                 } else {
 140                         (void) strcpy(dd->dd_myname, spa_name(dp->dp_spa));
 141                 }
 142
 143                 if (dsl_dir_is_clone(dd)) {
 144                         dmu_buf_t *origin_bonus;
 145                         dsl_dataset_phys_t *origin_phys;
 146
 147                         /*
 148                          * We can't open the origin dataset, because
 149                          * that would require opening this dsl_dir.
 150                          * Just look at its phys directly instead.
 151                          */
 152                         err = dmu_bonus_hold(dp->dp_meta_objset,
 153                             dd->dd_phys->dd_origin_obj, FTAG, &origin_bonus);
 154                         if (err)
 155                                 goto errout;
 156                         origin_phys = origin_bonus->db_data;
 157                         dd->dd_origin_txg =
 158                             origin_phys->ds_creation_txg;
 159                         dmu_buf_rele(origin_bonus, FTAG);
 160                 }
 161
 162                 winner = dmu_buf_set_user_ie(dbuf, dd, &dd->dd_phys,
 163                     dsl_dir_evict);
 164                 if (winner) {
 165                         if (dd->dd_parent)
 166                                 dsl_dir_close(dd->dd_parent, dd);
 167                         mutex_destroy(&dd->dd_lock);
 168                         kmem_free(dd, sizeof (dsl_dir_t));
 169                         dd = winner;
 170                 } else {
 171                         spa_open_ref(dp->dp_spa, dd);
 172                 }
 173         }
 174
 175         /*
 176          * The dsl_dir_t has both open-to-close and instantiate-to-evict
 177          * holds on the spa.  We need the open-to-close holds because
 178          * otherwise the spa_refcnt wouldn't change when we open a
 179          * dir which the spa also has open, so we could incorrectly
 180          * think it was OK to unload/export/destroy the pool.  We need
 181          * the instantiate-to-evict hold because the dsl_dir_t has a
 182          * pointer to the dd_pool, which has a pointer to the spa_t.
 183          */
 184         spa_open_ref(dp->dp_spa, tag);
 185         ASSERT3P(dd->dd_pool, ==, dp);
 186         ASSERT3U(dd->dd_object, ==, ddobj);
 187         ASSERT3P(dd->dd_dbuf, ==, dbuf);
 188         *ddp = dd;
 189         return (0);
 190
 191 errout:
 192         if (dd->dd_parent)
 193                 dsl_dir_close(dd->dd_parent, dd);
 194         mutex_destroy(&dd->dd_lock);
 195         kmem_free(dd, sizeof (dsl_dir_t));
 196         dmu_buf_rele(dbuf, tag);
 197         return (err);
 198 }
 199
 200 void
 201 dsl_dir_close(dsl_dir_t *dd, void *tag)
 202 {
 203         dprintf_dd(dd, "%s\n", "");
 204         spa_close(dd->dd_pool->dp_spa, tag);
 205         dmu_buf_rele(dd->dd_dbuf, tag);
 206 }
 207
 208 /* buf must be long enough (MAXNAMELEN + strlen(MOS_DIR_NAME) + 1 should do) */
 209 void
 210 dsl_dir_name(dsl_dir_t *dd, char *buf)
 211 {
 212         if (dd->dd_parent) {
 213                 dsl_dir_name(dd->dd_parent, buf);
 214                 (void) strcat(buf, "/");
 215         } else {
 216                 buf[0] = '\0';
 217         }
 218         if (!MUTEX_HELD(&dd->dd_lock)) {
 219                 /*
 220                  * recursive mutex so that we can use
 221                  * dprintf_dd() with dd_lock held
 222                  */
 223                 mutex_enter(&dd->dd_lock);
 224                 (void) strcat(buf, dd->dd_myname);
 225                 mutex_exit(&dd->dd_lock);
 226         } else {
 227                 (void) strcat(buf, dd->dd_myname);
 228         }
 229 }
 230
 231 /* Calculate name length, avoiding all the strcat calls of dsl_dir_name */
 232 int
 233 dsl_dir_namelen(dsl_dir_t *dd)
 234 {
 235         int result = 0;
 236
 237         if (dd->dd_parent) {
 238                 /* parent's name + 1 for the "/" */
 239                 result = dsl_dir_namelen(dd->dd_parent) + 1;
 240         }
 241
 242         if (!MUTEX_HELD(&dd->dd_lock)) {
 243                 /* see dsl_dir_name */
 244                 mutex_enter(&dd->dd_lock);
 245                 result += strlen(dd->dd_myname);
 246                 mutex_exit(&dd->dd_lock);
 247         } else {
 248                 result += strlen(dd->dd_myname);
 249         }
 250
 251         return (result);
 252 }
 253
 254 static int
 255 getcomponent(const char *path, char *component, const char **nextp)
 256 {
 257         char *p;
 258         if ((path == NULL) || (path[0] == '\0'))
 259                 return (ENOENT);
 260         /* This would be a good place to reserve some namespace... */
 261         p = strpbrk(path, "/@");
 262         if (p && (p[1] == '/' || p[1] == '@')) {
 263                 /* two separators in a row */
 264                 return (EINVAL);
 265         }
 266         if (p == NULL || p == path) {
 267                 /*
 268                  * if the first thing is an @ or /, it had better be an
 269                  * @ and it had better not have any more ats or slashes,
 270                  * and it had better have something after the @.
 271                  */
 272                 if (p != NULL &&
 273                     (p[0] != '@' || strpbrk(path+1, "/@") || p[1] == '\0'))
 274                         return (EINVAL);
 275                 if (strlen(path) >= MAXNAMELEN)
 276                         return (ENAMETOOLONG);
 277                 (void) strcpy(component, path);
 278                 p = NULL;
 279         } else if (p[0] == '/') {
 280                 if (p-path >= MAXNAMELEN)
 281                         return (ENAMETOOLONG);
 282                 (void) strncpy(component, path, p - path);
 283                 component[p-path] = '\0';
 284                 p++;
 285         } else if (p[0] == '@') {
 286                 /*
 287                  * if the next separator is an @, there better not be
 288                  * any more slashes.
 289                  */
 290                 if (strchr(path, '/'))
 291                         return (EINVAL);
 292                 if (p-path >= MAXNAMELEN)
 293                         return (ENAMETOOLONG);
 294                 (void) strncpy(component, path, p - path);
 295                 component[p-path] = '\0';
 296         } else {
 297                 ASSERT(!"invalid p");
 298         }
 299         *nextp = p;
 300         return (0);
 301 }
 302
 303 /*
 304  * same as dsl_open_dir, ignore the first component of name and use the
 305  * spa instead
 306  */
 307 int
 308 dsl_dir_open_spa(spa_t *spa, const char *name, void *tag,
 309     dsl_dir_t **ddp, const char **tailp)
 310 {
 311         char buf[MAXNAMELEN];
 312         const char *next, *nextnext = NULL;
 313         int err;
 314         dsl_dir_t *dd;
 315         dsl_pool_t *dp;
 316         uint64_t ddobj;
 317         int openedspa = FALSE;
 318
 319         dprintf("%s\n", name);
 320
 321         err = getcomponent(name, buf, &next);
 322         if (err)
 323                 return (err);
 324         if (spa == NULL) {
 325                 err = spa_open(buf, &spa, FTAG);
 326                 if (err) {
 327                         dprintf("spa_open(%s) failed\n", buf);
 328                         return (err);
 329                 }
 330                 openedspa = TRUE;
 331
 332                 /* XXX this assertion belongs in spa_open */
 333                 ASSERT(!dsl_pool_sync_context(spa_get_dsl(spa)));
 334         }
 335
 336         dp = spa_get_dsl(spa);
 337
 338         rw_enter(&dp->dp_config_rwlock, RW_READER);
 339         err = dsl_dir_open_obj(dp, dp->dp_root_dir_obj, NULL, tag, &dd);
 340         if (err) {
 341                 rw_exit(&dp->dp_config_rwlock);
 342                 if (openedspa)
 343                         spa_close(spa, FTAG);
 344                 return (err);
 345         }
 346
 347         while (next != NULL) {
 348                 dsl_dir_t *child_ds;
 349                 err = getcomponent(next, buf, &nextnext);
 350                 if (err)
 351                         break;
 352                 ASSERT(next[0] != '\0');
 353                 if (next[0] == '@')
 354                         break;
 355                 dprintf("looking up %s in obj%lld\n",
 356                     buf, dd->dd_phys->dd_child_dir_zapobj);
 357
 358                 err = zap_lookup(dp->dp_meta_objset,
 359                     dd->dd_phys->dd_child_dir_zapobj,
 360                     buf, sizeof (ddobj), 1, &ddobj);
 361                 if (err) {
 362                         if (err == ENOENT)
 363                                 err = 0;
 364                         break;
 365                 }
 366
 367                 err = dsl_dir_open_obj(dp, ddobj, buf, tag, &child_ds);
 368                 if (err)
 369                         break;
 370                 dsl_dir_close(dd, tag);
 371                 dd = child_ds;
 372                 next = nextnext;
 373         }
 374         rw_exit(&dp->dp_config_rwlock);
 375
 376         if (err) {
 377                 dsl_dir_close(dd, tag);
 378                 if (openedspa)
 379                         spa_close(spa, FTAG);
 380                 return (err);
 381         }
 382
 383         /*
 384          * It's an error if there's more than one component left, or
 385          * tailp==NULL and there's any component left.
 386          */
 387         if (next != NULL &&
 388             (tailp == NULL || (nextnext && nextnext[0] != '\0'))) {
 389                 /* bad path name */
 390                 dsl_dir_close(dd, tag);
 391                 dprintf("next=%p (%s) tail=%p\n", next, next?next:"", tailp);
 392                 err = ENOENT;
 393         }
 394         if (tailp)
 395                 *tailp = next;
 396         if (openedspa)
 397                 spa_close(spa, FTAG);
 398         *ddp = dd;
 399         return (err);
 400 }
 401
 402 /*
 403  * Return the dsl_dir_t, and possibly the last component which couldn't
 404  * be found in *tail.  Return NULL if the path is bogus, or if
 405  * tail==NULL and we couldn't parse the whole name.  (*tail)[0] == '@'
 406  * means that the last component is a snapshot.
 407  */
 408 int
 409 dsl_dir_open(const char *name, void *tag, dsl_dir_t **ddp, const char **tailp)
 410 {
 411         return (dsl_dir_open_spa(NULL, name, tag, ddp, tailp));
 412 }
 413
 414 uint64_t
 415 dsl_dir_create_sync(dsl_pool_t *dp, dsl_dir_t *pds, const char *name,
 416     dmu_tx_t *tx)
 417 {
 418         objset_t *mos = dp->dp_meta_objset;
 419         uint64_t ddobj;
 420         dsl_dir_phys_t *ddphys;
 421         dmu_buf_t *dbuf;
 422
 423         ddobj = dmu_object_alloc(mos, DMU_OT_DSL_DIR, 0,
 424             DMU_OT_DSL_DIR, sizeof (dsl_dir_phys_t), tx);
 425         if (pds) {
 426                 VERIFY(0 == zap_add(mos, pds->dd_phys->dd_child_dir_zapobj,
 427                     name, sizeof (uint64_t), 1, &ddobj, tx));
 428         } else {
 429                 /* it's the root dir */
 430                 VERIFY(0 == zap_add(mos, DMU_POOL_DIRECTORY_OBJECT,
 431                     DMU_POOL_ROOT_DATASET, sizeof (uint64_t), 1, &ddobj, tx));
 432         }
 433         VERIFY(0 == dmu_bonus_hold(mos, ddobj, FTAG, &dbuf));
 434         dmu_buf_will_dirty(dbuf, tx);
 435         ddphys = dbuf->db_data;
 436
 437         ddphys->dd_creation_time = gethrestime_sec();
 438         if (pds)
 439                 ddphys->dd_parent_obj = pds->dd_object;
 440         ddphys->dd_props_zapobj = zap_create(mos,
 441             DMU_OT_DSL_PROPS, DMU_OT_NONE, 0, tx);
 442         ddphys->dd_child_dir_zapobj = zap_create(mos,
 443             DMU_OT_DSL_DIR_CHILD_MAP, DMU_OT_NONE, 0, tx);
 444         if (spa_version(dp->dp_spa) >= SPA_VERSION_USED_BREAKDOWN)
 445                 ddphys->dd_flags |= DD_FLAG_USED_BREAKDOWN;
 446         dmu_buf_rele(dbuf, FTAG);
 447
 448         return (ddobj);
 449 }
 450
 451 /* ARGSUSED */
 452 int
 453 dsl_dir_destroy_check(void *arg1, void *arg2, dmu_tx_t *tx)
 454 {
 455         dsl_dataset_t *ds = arg1;
 456         dsl_dir_t *dd = ds->ds_dir;
 457         dsl_pool_t *dp = dd->dd_pool;
 458         objset_t *mos = dp->dp_meta_objset;
 459         int err;
 460         uint64_t count;
 461
 462         /*
 463          * There should be exactly two holds, both from
 464          * dsl_dataset_destroy: one on the dd directory, and one on its
 465          * head ds.  If there are more holds, then a concurrent thread is
 466          * performing a lookup inside this dir while we're trying to destroy
 467          * it.  To minimize this possibility, we perform this check only
 468          * in syncing context and fail the operation if we encounter
 469          * additional holds.  The dp_config_rwlock ensures that nobody else
 470          * opens it after we check.
 471          */
 472         if (dmu_tx_is_syncing(tx) && dmu_buf_refcount(dd->dd_dbuf) > 2)
 473                 return (EBUSY);
 474
 475         err = zap_count(mos, dd->dd_phys->dd_child_dir_zapobj, &count);
 476         if (err)
 477                 return (err);
 478         if (count != 0)
 479                 return (EEXIST);
 480
 481         return (0);
 482 }
 483
 484 void
 485 dsl_dir_destroy_sync(void *arg1, void *tag, dmu_tx_t *tx)
 486 {
 487         dsl_dataset_t *ds = arg1;
 488         dsl_dir_t *dd = ds->ds_dir;
 489         objset_t *mos = dd->dd_pool->dp_meta_objset;
 490         dsl_prop_setarg_t psa;
 491         uint64_t value = 0;
 492         uint64_t obj;
 493         dd_used_t t;
 494
 495         ASSERT(RW_WRITE_HELD(&dd->dd_pool->dp_config_rwlock));
 496         ASSERT(dd->dd_phys->dd_head_dataset_obj == 0);
 497
 498         /* Remove our reservation. */
 499         dsl_prop_setarg_init_uint64(&psa, "reservation",
 500             (ZPROP_SRC_NONE | ZPROP_SRC_LOCAL | ZPROP_SRC_RECEIVED),
 501             &value);
 502         psa.psa_effective_value = 0;    /* predict default value */
 503
 504         dsl_dir_set_reservation_sync(ds, &psa, tx);
 505
 506         ASSERT0(dd->dd_phys->dd_used_bytes);
 507         ASSERT0(dd->dd_phys->dd_reserved);
 508         for (t = 0; t < DD_USED_NUM; t++)
 509                 ASSERT0(dd->dd_phys->dd_used_breakdown[t]);
 510
 511         VERIFY(0 == zap_destroy(mos, dd->dd_phys->dd_child_dir_zapobj, tx));
 512         VERIFY(0 == zap_destroy(mos, dd->dd_phys->dd_props_zapobj, tx));
 513         VERIFY(0 == dsl_deleg_destroy(mos, dd->dd_phys->dd_deleg_zapobj, tx));
 514         VERIFY(0 == zap_remove(mos,
 515             dd->dd_parent->dd_phys->dd_child_dir_zapobj, dd->dd_myname, tx));
 516
 517         obj = dd->dd_object;
 518         dsl_dir_close(dd, tag);
 519         VERIFY(0 == dmu_object_free(mos, obj, tx));
 520 }
 521
 522 boolean_t
 523 dsl_dir_is_clone(dsl_dir_t *dd)
 524 {
 525         return (dd->dd_phys->dd_origin_obj &&
 526             (dd->dd_pool->dp_origin_snap == NULL ||
 527             dd->dd_phys->dd_origin_obj !=
 528             dd->dd_pool->dp_origin_snap->ds_object));
 529 }
 530
 531 void
 532 dsl_dir_stats(dsl_dir_t *dd, nvlist_t *nv)
 533 {
 534         mutex_enter(&dd->dd_lock);
 535         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USED,
 536             dd->dd_phys->dd_used_bytes);
 537         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_QUOTA, dd->dd_phys->dd_quota);
 538         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_RESERVATION,
 539             dd->dd_phys->dd_reserved);
 540         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_COMPRESSRATIO,
 541             dd->dd_phys->dd_compressed_bytes == 0 ? 100 :
 542             (dd->dd_phys->dd_uncompressed_bytes * 100 /
 543             dd->dd_phys->dd_compressed_bytes));
 544         if (dd->dd_phys->dd_flags & DD_FLAG_USED_BREAKDOWN) {
 545                 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDSNAP,
 546                     dd->dd_phys->dd_used_breakdown[DD_USED_SNAP]);
 547                 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDDS,
 548                     dd->dd_phys->dd_used_breakdown[DD_USED_HEAD]);
 549                 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDREFRESERV,
 550                     dd->dd_phys->dd_used_breakdown[DD_USED_REFRSRV]);
 551                 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDCHILD,
 552                     dd->dd_phys->dd_used_breakdown[DD_USED_CHILD] +
 553                     dd->dd_phys->dd_used_breakdown[DD_USED_CHILD_RSRV]);
 554         }
 555         mutex_exit(&dd->dd_lock);
 556
 557         rw_enter(&dd->dd_pool->dp_config_rwlock, RW_READER);
 558         if (dsl_dir_is_clone(dd)) {
 559                 dsl_dataset_t *ds;
 560                 char buf[MAXNAMELEN];
 561
 562                 VERIFY(0 == dsl_dataset_hold_obj(dd->dd_pool,
 563                     dd->dd_phys->dd_origin_obj, FTAG, &ds));
 564                 dsl_dataset_name(ds, buf);
 565                 dsl_dataset_rele(ds, FTAG);
 566                 dsl_prop_nvlist_add_string(nv, ZFS_PROP_ORIGIN, buf);
 567         }
 568         rw_exit(&dd->dd_pool->dp_config_rwlock);
 569 }
 570
 571 void
 572 dsl_dir_dirty(dsl_dir_t *dd, dmu_tx_t *tx)
 573 {
 574         dsl_pool_t *dp = dd->dd_pool;
 575
 576         ASSERT(dd->dd_phys);
 577
 578         if (txg_list_add(&dp->dp_dirty_dirs, dd, tx->tx_txg) == 0) {
 579                 /* up the hold count until we can be written out */
 580                 dmu_buf_add_ref(dd->dd_dbuf, dd);
 581         }
 582 }
 583
 584 static int64_t
 585 parent_delta(dsl_dir_t *dd, uint64_t used, int64_t delta)
 586 {
 587         uint64_t old_accounted = MAX(used, dd->dd_phys->dd_reserved);
 588         uint64_t new_accounted = MAX(used + delta, dd->dd_phys->dd_reserved);
 589         return (new_accounted - old_accounted);
 590 }
 591
 592 void
 593 dsl_dir_sync(dsl_dir_t *dd, dmu_tx_t *tx)
 594 {
 595         ASSERT(dmu_tx_is_syncing(tx));
 596
 597         mutex_enter(&dd->dd_lock);
 598         ASSERT0(dd->dd_tempreserved[tx->tx_txg&TXG_MASK]);
 599         dprintf_dd(dd, "txg=%llu towrite=%lluK\n", tx->tx_txg,
 600             dd->dd_space_towrite[tx->tx_txg&TXG_MASK] / 1024);
 601         dd->dd_space_towrite[tx->tx_txg&TXG_MASK] = 0;
 602         mutex_exit(&dd->dd_lock);
 603
 604         /* release the hold from dsl_dir_dirty */
 605         dmu_buf_rele(dd->dd_dbuf, dd);
 606 }
 607
 608 static uint64_t
 609 dsl_dir_space_towrite(dsl_dir_t *dd)
 610 {
 611         uint64_t space = 0;
 612         int i;
 613
 614         ASSERT(MUTEX_HELD(&dd->dd_lock));
 615
 616         for (i = 0; i < TXG_SIZE; i++) {
 617                 space += dd->dd_space_towrite[i&TXG_MASK];
 618                 ASSERT3U(dd->dd_space_towrite[i&TXG_MASK], >=, 0);
 619         }
 620         return (space);
 621 }
 622
 623 /*
 624  * How much space would dd have available if ancestor had delta applied
 625  * to it?  If ondiskonly is set, we're only interested in what's
 626  * on-disk, not estimated pending changes.
 627  */
 628 uint64_t
 629 dsl_dir_space_available(dsl_dir_t *dd,
 630     dsl_dir_t *ancestor, int64_t delta, int ondiskonly)
 631 {
 632         uint64_t parentspace, myspace, quota, used;
 633
 634         /*
 635          * If there are no restrictions otherwise, assume we have
 636          * unlimited space available.
 637          */
 638         quota = UINT64_MAX;
 639         parentspace = UINT64_MAX;
 640
 641         if (dd->dd_parent != NULL) {
 642                 parentspace = dsl_dir_space_available(dd->dd_parent,
 643                     ancestor, delta, ondiskonly);
 644         }
 645
 646         mutex_enter(&dd->dd_lock);
 647         if (dd->dd_phys->dd_quota != 0)
 648                 quota = dd->dd_phys->dd_quota;
 649         used = dd->dd_phys->dd_used_bytes;
 650         if (!ondiskonly)
 651                 used += dsl_dir_space_towrite(dd);
 652
 653         if (dd->dd_parent == NULL) {
 654                 uint64_t poolsize = dsl_pool_adjustedsize(dd->dd_pool, FALSE);
 655                 quota = MIN(quota, poolsize);
 656         }
 657
 658         if (dd->dd_phys->dd_reserved > used && parentspace != UINT64_MAX) {
 659                 /*
 660                  * We have some space reserved, in addition to what our
 661                  * parent gave us.
 662                  */
 663                 parentspace += dd->dd_phys->dd_reserved - used;
 664         }
 665
 666         if (dd == ancestor) {
 667                 ASSERT(delta <= 0);
 668                 ASSERT(used >= -delta);
 669                 used += delta;
 670                 if (parentspace != UINT64_MAX)
 671                         parentspace -= delta;
 672         }
 673
 674         if (used > quota) {
 675                 /* over quota */
 676                 myspace = 0;
 677         } else {
 678                 /*
 679                  * the lesser of the space provided by our parent and
 680                  * the space left in our quota
 681                  */
 682                 myspace = MIN(parentspace, quota - used);
 683         }
 684
 685         mutex_exit(&dd->dd_lock);
 686
 687         return (myspace);
 688 }
 689
 690 struct tempreserve {
 691         list_node_t tr_node;
 692         dsl_pool_t *tr_dp;
 693         dsl_dir_t *tr_ds;
 694         uint64_t tr_size;
 695 };
 696
 697 static int
 698 dsl_dir_tempreserve_impl(dsl_dir_t *dd, uint64_t asize, boolean_t netfree,
 699     boolean_t ignorequota, boolean_t checkrefquota, list_t *tr_list,
 700     dmu_tx_t *tx, boolean_t first)
 701 {
 702         uint64_t txg = tx->tx_txg;
 703         uint64_t est_inflight, used_on_disk, quota, parent_rsrv;
 704         uint64_t deferred = 0;
 705         struct tempreserve *tr;
 706         int retval = EDQUOT;
 707         int txgidx = txg & TXG_MASK;
 708         int i;
 709         uint64_t ref_rsrv = 0;
 710
 711         ASSERT3U(txg, !=, 0);
 712         ASSERT3S(asize, >, 0);
 713
 714         mutex_enter(&dd->dd_lock);
 715
 716         /*
 717          * Check against the dsl_dir's quota.  We don't add in the delta
 718          * when checking for over-quota because they get one free hit.
 719          */
 720         est_inflight = dsl_dir_space_towrite(dd);
 721         for (i = 0; i < TXG_SIZE; i++)
 722                 est_inflight += dd->dd_tempreserved[i];
 723         used_on_disk = dd->dd_phys->dd_used_bytes;
 724
 725         /*
 726          * On the first iteration, fetch the dataset's used-on-disk and
 727          * refreservation values. Also, if checkrefquota is set, test if
 728          * allocating this space would exceed the dataset's refquota.
 729          */
 730         if (first && tx->tx_objset) {
 731                 int error;
 732                 dsl_dataset_t *ds = tx->tx_objset->os_dsl_dataset;
 733
 734                 error = dsl_dataset_check_quota(ds, checkrefquota,
 735                     asize, est_inflight, &used_on_disk, &ref_rsrv);
 736                 if (error) {
 737                         mutex_exit(&dd->dd_lock);
 738                         return (error);
 739                 }
 740         }
 741
 742         /*
 743          * If this transaction will result in a net free of space,
 744          * we want to let it through.
 745          */
 746         if (ignorequota || netfree || dd->dd_phys->dd_quota == 0)
 747                 quota = UINT64_MAX;
 748         else
 749                 quota = dd->dd_phys->dd_quota;
 750
 751         /*
 752          * Adjust the quota against the actual pool size at the root
 753          * minus any outstanding deferred frees.
 754          * To ensure that it's possible to remove files from a full
 755          * pool without inducing transient overcommits, we throttle
 756          * netfree transactions against a quota that is slightly larger,
 757          * but still within the pool's allocation slop.  In cases where
 758          * we're very close to full, this will allow a steady trickle of
 759          * removes to get through.
 760          */
 761         if (dd->dd_parent == NULL) {
 762                 spa_t *spa = dd->dd_pool->dp_spa;
 763                 uint64_t poolsize = dsl_pool_adjustedsize(dd->dd_pool, netfree);
 764                 deferred = metaslab_class_get_deferred(spa_normal_class(spa));
 765                 if (poolsize - deferred < quota) {
 766                         quota = poolsize - deferred;
 767                         retval = ENOSPC;
 768                 }
 769         }
 770
 771         /*
 772          * If they are requesting more space, and our current estimate
 773          * is over quota, they get to try again unless the actual
 774          * on-disk is over quota and there are no pending changes (which
 775          * may free up space for us).
 776          */
 777         if (used_on_disk + est_inflight >= quota) {
 778                 if (est_inflight > 0 || used_on_disk < quota ||
 779                     (retval == ENOSPC && used_on_disk < quota + deferred))
 780                         retval = ERESTART;
 781                 dprintf_dd(dd, "failing: used=%lluK inflight = %lluK "
 782                     "quota=%lluK tr=%lluK err=%d\n",
 783                     used_on_disk>>10, est_inflight>>10,
 784                     quota>>10, asize>>10, retval);
 785                 mutex_exit(&dd->dd_lock);
 786                 return (retval);
 787         }
 788
 789         /* We need to up our estimated delta before dropping dd_lock */
 790         dd->dd_tempreserved[txgidx] += asize;
 791
 792         parent_rsrv = parent_delta(dd, used_on_disk + est_inflight,
 793             asize - ref_rsrv);
 794         mutex_exit(&dd->dd_lock);
 795
 796         tr = kmem_zalloc(sizeof (struct tempreserve), KM_SLEEP);
 797         tr->tr_ds = dd;
 798         tr->tr_size = asize;
 799         list_insert_tail(tr_list, tr);
 800
 801         /* see if it's OK with our parent */
 802         if (dd->dd_parent && parent_rsrv) {
 803                 boolean_t ismos = (dd->dd_phys->dd_head_dataset_obj == 0);
 804
 805                 return (dsl_dir_tempreserve_impl(dd->dd_parent,
 806                     parent_rsrv, netfree, ismos, TRUE, tr_list, tx, FALSE));
 807         } else {
 808                 return (0);
 809         }
 810 }
 811
 812 /*
 813  * Reserve space in this dsl_dir, to be used in this tx's txg.
 814  * After the space has been dirtied (and dsl_dir_willuse_space()
 815  * has been called), the reservation should be canceled, using
 816  * dsl_dir_tempreserve_clear().
 817  */
 818 int
 819 dsl_dir_tempreserve_space(dsl_dir_t *dd, uint64_t lsize, uint64_t asize,
 820     uint64_t fsize, uint64_t usize, void **tr_cookiep, dmu_tx_t *tx)
 821 {
 822         int err;
 823         list_t *tr_list;
 824
 825         if (asize == 0) {
 826                 *tr_cookiep = NULL;
 827                 return (0);
 828         }
 829
 830         tr_list = kmem_alloc(sizeof (list_t), KM_SLEEP);
 831         list_create(tr_list, sizeof (struct tempreserve),
 832             offsetof(struct tempreserve, tr_node));
 833         ASSERT3S(asize, >, 0);
 834         ASSERT3S(fsize, >=, 0);
 835
 836         err = arc_tempreserve_space(lsize, tx->tx_txg);
 837         if (err == 0) {
 838                 struct tempreserve *tr;
 839
 840                 tr = kmem_zalloc(sizeof (struct tempreserve), KM_SLEEP);
 841                 tr->tr_size = lsize;
 842                 list_insert_tail(tr_list, tr);
 843
 844                 err = dsl_pool_tempreserve_space(dd->dd_pool, asize, tx);
 845         } else {
 846                 if (err == EAGAIN) {
 847                         txg_delay(dd->dd_pool, tx->tx_txg, 1);
 848                         err = ERESTART;
 849                 }
 850                 dsl_pool_memory_pressure(dd->dd_pool);
 851         }
 852
 853         if (err == 0) {
 854                 struct tempreserve *tr;
 855
 856                 tr = kmem_zalloc(sizeof (struct tempreserve), KM_SLEEP);
 857                 tr->tr_dp = dd->dd_pool;
 858                 tr->tr_size = asize;
 859                 list_insert_tail(tr_list, tr);
 860
 861                 err = dsl_dir_tempreserve_impl(dd, asize, fsize >= asize,
 862                     FALSE, asize > usize, tr_list, tx, TRUE);
 863         }
 864
 865         if (err)
 866                 dsl_dir_tempreserve_clear(tr_list, tx);
 867         else
 868                 *tr_cookiep = tr_list;
 869
 870         return (err);
 871 }
 872
 873 /*
 874  * Clear a temporary reservation that we previously made with
 875  * dsl_dir_tempreserve_space().
 876  */
 877 void
 878 dsl_dir_tempreserve_clear(void *tr_cookie, dmu_tx_t *tx)
 879 {
 880         int txgidx = tx->tx_txg & TXG_MASK;
 881         list_t *tr_list = tr_cookie;
 882         struct tempreserve *tr;
 883
 884         ASSERT3U(tx->tx_txg, !=, 0);
 885
 886         if (tr_cookie == NULL)
 887                 return;
 888
 889         while (tr = list_head(tr_list)) {
 890                 if (tr->tr_dp) {
 891                         dsl_pool_tempreserve_clear(tr->tr_dp, tr->tr_size, tx);
 892                 } else if (tr->tr_ds) {
 893                         mutex_enter(&tr->tr_ds->dd_lock);
 894                         ASSERT3U(tr->tr_ds->dd_tempreserved[txgidx], >=,
 895                             tr->tr_size);
 896                         tr->tr_ds->dd_tempreserved[txgidx] -= tr->tr_size;
 897                         mutex_exit(&tr->tr_ds->dd_lock);
 898                 } else {
 899                         arc_tempreserve_clear(tr->tr_size);
 900                 }
 901                 list_remove(tr_list, tr);
 902                 kmem_free(tr, sizeof (struct tempreserve));
 903         }
 904
 905         kmem_free(tr_list, sizeof (list_t));
 906 }
 907
 908 static void
 909 dsl_dir_willuse_space_impl(dsl_dir_t *dd, int64_t space, dmu_tx_t *tx)
 910 {
 911         int64_t parent_space;
 912         uint64_t est_used;
 913
 914         mutex_enter(&dd->dd_lock);
 915         if (space > 0)
 916                 dd->dd_space_towrite[tx->tx_txg & TXG_MASK] += space;
 917
 918         est_used = dsl_dir_space_towrite(dd) + dd->dd_phys->dd_used_bytes;
 919         parent_space = parent_delta(dd, est_used, space);
 920         mutex_exit(&dd->dd_lock);
 921
 922         /* Make sure that we clean up dd_space_to* */
 923         dsl_dir_dirty(dd, tx);
 924
 925         /* XXX this is potentially expensive and unnecessary... */
 926         if (parent_space && dd->dd_parent)
 927                 dsl_dir_willuse_space_impl(dd->dd_parent, parent_space, tx);
 928 }
 929
 930 /*
 931  * Call in open context when we think we're going to write/free space,
 932  * eg. when dirtying data.  Be conservative (ie. OK to write less than
 933  * this or free more than this, but don't write more or free less).
 934  */
 935 void
 936 dsl_dir_willuse_space(dsl_dir_t *dd, int64_t space, dmu_tx_t *tx)
 937 {
 938         dsl_pool_willuse_space(dd->dd_pool, space, tx);
 939         dsl_dir_willuse_space_impl(dd, space, tx);
 940 }
 941
 942 /* call from syncing context when we actually write/free space for this dd */
 943 void
 944 dsl_dir_diduse_space(dsl_dir_t *dd, dd_used_t type,
 945     int64_t used, int64_t compressed, int64_t uncompressed, dmu_tx_t *tx)
 946 {
 947         int64_t accounted_delta;
 948         boolean_t needlock = !MUTEX_HELD(&dd->dd_lock);
 949
 950         ASSERT(dmu_tx_is_syncing(tx));
 951         ASSERT(type < DD_USED_NUM);
 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         dmu_buf_will_dirty(dd->dd_dbuf, tx);
 962         dd->dd_phys->dd_used_bytes += used;
 963         dd->dd_phys->dd_uncompressed_bytes += uncompressed;
 964         dd->dd_phys->dd_compressed_bytes += compressed;
 965
 966         if (dd->dd_phys->dd_flags & DD_FLAG_USED_BREAKDOWN) {
 967                 ASSERT(used > 0 ||
 968                     dd->dd_phys->dd_used_breakdown[type] >= -used);
 969                 dd->dd_phys->dd_used_breakdown[type] += used;
 970 #ifdef DEBUG
 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 #endif
 977         }
 978         if (needlock)
 979                 mutex_exit(&dd->dd_lock);
 980
 981         if (dd->dd_parent != NULL) {
 982                 dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD,
 983                     accounted_delta, compressed, uncompressed, tx);
 984                 dsl_dir_transfer_space(dd->dd_parent,
 985                     used - accounted_delta,
 986                     DD_USED_CHILD_RSRV, DD_USED_CHILD, tx);
 987         }
 988 }
 989
 990 void
 991 dsl_dir_transfer_space(dsl_dir_t *dd, int64_t delta,
 992     dd_used_t oldtype, dd_used_t newtype, dmu_tx_t *tx)
 993 {
 994         boolean_t needlock = !MUTEX_HELD(&dd->dd_lock);
 995
 996         ASSERT(dmu_tx_is_syncing(tx));
 997         ASSERT(oldtype < DD_USED_NUM);
 998         ASSERT(newtype < DD_USED_NUM);
 999
1000         if (delta == 0 || !(dd->dd_phys->dd_flags & DD_FLAG_USED_BREAKDOWN))
1001                 return;
1002
1003         if (needlock)
1004                 mutex_enter(&dd->dd_lock);
1005         ASSERT(delta > 0 ?
1006             dd->dd_phys->dd_used_breakdown[oldtype] >= delta :
1007             dd->dd_phys->dd_used_breakdown[newtype] >= -delta);
1008         ASSERT(dd->dd_phys->dd_used_bytes >= ABS(delta));
1009         dmu_buf_will_dirty(dd->dd_dbuf, tx);
1010         dd->dd_phys->dd_used_breakdown[oldtype] -= delta;
1011         dd->dd_phys->dd_used_breakdown[newtype] += delta;
1012         if (needlock)
1013                 mutex_exit(&dd->dd_lock);
1014 }
1015
1016 static int
1017 dsl_dir_set_quota_check(void *arg1, void *arg2, dmu_tx_t *tx)
1018 {
1019         dsl_dataset_t *ds = arg1;
1020         dsl_dir_t *dd = ds->ds_dir;
1021         dsl_prop_setarg_t *psa = arg2;
1022         int err;
1023         uint64_t towrite;
1024
1025         if ((err = dsl_prop_predict_sync(ds->ds_dir, psa)) != 0)
1026                 return (err);
1027
1028         if (psa->psa_effective_value == 0)
1029                 return (0);
1030
1031         mutex_enter(&dd->dd_lock);
1032         /*
1033          * If we are doing the preliminary check in open context, and
1034          * there are pending changes, then don't fail it, since the
1035          * pending changes could under-estimate the amount of space to be
1036          * freed up.
1037          */
1038         towrite = dsl_dir_space_towrite(dd);
1039         if ((dmu_tx_is_syncing(tx) || towrite == 0) &&
1040             (psa->psa_effective_value < dd->dd_phys->dd_reserved ||
1041             psa->psa_effective_value < dd->dd_phys->dd_used_bytes + towrite)) {
1042                 err = ENOSPC;
1043         }
1044         mutex_exit(&dd->dd_lock);
1045         return (err);
1046 }
1047
1048 extern dsl_syncfunc_t dsl_prop_set_sync;
1049
1050 static void
1051 dsl_dir_set_quota_sync(void *arg1, void *arg2, dmu_tx_t *tx)
1052 {
1053         dsl_dataset_t *ds = arg1;
1054         dsl_dir_t *dd = ds->ds_dir;
1055         dsl_prop_setarg_t *psa = arg2;
1056         uint64_t effective_value = psa->psa_effective_value;
1057
1058         dsl_prop_set_sync(ds, psa, tx);
1059         DSL_PROP_CHECK_PREDICTION(dd, psa);
1060
1061         dmu_buf_will_dirty(dd->dd_dbuf, tx);
1062
1063         mutex_enter(&dd->dd_lock);
1064         dd->dd_phys->dd_quota = effective_value;
1065         mutex_exit(&dd->dd_lock);
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
1180 int
1181 dsl_dir_set_reservation(const char *ddname, zprop_source_t source,
1182     uint64_t reservation)
1183 {
1184         dsl_dir_t *dd;
1185         dsl_dataset_t *ds;
1186         dsl_prop_setarg_t psa;
1187         int err;
1188
1189         dsl_prop_setarg_init_uint64(&psa, "reservation", source, &reservation);
1190
1191         err = dsl_dataset_hold(ddname, FTAG, &ds);
1192         if (err)
1193                 return (err);
1194
1195         err = dsl_dir_open(ddname, FTAG, &dd, NULL);
1196         if (err) {
1197                 dsl_dataset_rele(ds, FTAG);
1198                 return (err);
1199         }
1200
1201         ASSERT(ds->ds_dir == dd);
1202
1203         err = dsl_sync_task_do(dd->dd_pool, dsl_dir_set_reservation_check,
1204             dsl_dir_set_reservation_sync, ds, &psa, 0);
1205
1206         dsl_dir_close(dd, FTAG);
1207         dsl_dataset_rele(ds, FTAG);
1208         return (err);
1209 }
1210
1211 static dsl_dir_t *
1212 closest_common_ancestor(dsl_dir_t *ds1, dsl_dir_t *ds2)
1213 {
1214         for (; ds1; ds1 = ds1->dd_parent) {
1215                 dsl_dir_t *dd;
1216                 for (dd = ds2; dd; dd = dd->dd_parent) {
1217                         if (ds1 == dd)
1218                                 return (dd);
1219                 }
1220         }
1221         return (NULL);
1222 }
1223
1224 /*
1225  * If delta is applied to dd, how much of that delta would be applied to
1226  * ancestor?  Syncing context only.
1227  */
1228 static int64_t
1229 would_change(dsl_dir_t *dd, int64_t delta, dsl_dir_t *ancestor)
1230 {
1231         if (dd == ancestor)
1232                 return (delta);
1233
1234         mutex_enter(&dd->dd_lock);
1235         delta = parent_delta(dd, dd->dd_phys->dd_used_bytes, delta);
1236         mutex_exit(&dd->dd_lock);
1237         return (would_change(dd->dd_parent, delta, ancestor));
1238 }
1239
1240 struct renamearg {
1241         dsl_dir_t *newparent;
1242         const char *mynewname;
1243         boolean_t allowmounted;
1244 };
1245
1246 static int
1247 dsl_dir_rename_check(void *arg1, void *arg2, dmu_tx_t *tx)
1248 {
1249         dsl_dir_t *dd = arg1;
1250         struct renamearg *ra = arg2;
1251         dsl_pool_t *dp = dd->dd_pool;
1252         objset_t *mos = dp->dp_meta_objset;
1253         int err;
1254         uint64_t val;
1255
1256         /*
1257          * There should only be one reference, from dmu_objset_rename().
1258          * Fleeting holds are also possible (eg, from "zfs list" getting
1259          * stats), but any that are present in open context will likely
1260          * be gone by syncing context, so only fail from syncing
1261          * context.
1262          * Don't check if we allow renaming of busy (mounted) dataset.
1263          */
1264         if (!ra->allowmounted && dmu_tx_is_syncing(tx) &&
1265             dmu_buf_refcount(dd->dd_dbuf) > 1) {
1266                 return (EBUSY);
1267         }
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         char oldname[MAXPATHLEN], newname[MAXPATHLEN];
1298         dsl_dir_t *dd = arg1;
1299         struct renamearg *ra = arg2;
1300         dsl_pool_t *dp = dd->dd_pool;
1301         objset_t *mos = dp->dp_meta_objset;
1302         int err;
1303
1304         ASSERT(ra->allowmounted || dmu_buf_refcount(dd->dd_dbuf) <= 2);
1305
1306         if (ra->newparent != dd->dd_parent) {
1307                 dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD,
1308                     -dd->dd_phys->dd_used_bytes,
1309                     -dd->dd_phys->dd_compressed_bytes,
1310                     -dd->dd_phys->dd_uncompressed_bytes, tx);
1311                 dsl_dir_diduse_space(ra->newparent, 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
1316                 if (dd->dd_phys->dd_reserved > dd->dd_phys->dd_used_bytes) {
1317                         uint64_t unused_rsrv = dd->dd_phys->dd_reserved -
1318                             dd->dd_phys->dd_used_bytes;
1319
1320                         dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD_RSRV,
1321                             -unused_rsrv, 0, 0, tx);
1322                         dsl_dir_diduse_space(ra->newparent, DD_USED_CHILD_RSRV,
1323                             unused_rsrv, 0, 0, tx);
1324                 }
1325         }
1326
1327         dmu_buf_will_dirty(dd->dd_dbuf, tx);
1328
1329         /* remove from old parent zapobj */
1330         dsl_dir_name(dd, oldname);
1331         err = zap_remove(mos, dd->dd_parent->dd_phys->dd_child_dir_zapobj,
1332             dd->dd_myname, tx);
1333         ASSERT0(err);
1334
1335         (void) strcpy(dd->dd_myname, ra->mynewname);
1336         dsl_dir_close(dd->dd_parent, dd);
1337         dd->dd_phys->dd_parent_obj = ra->newparent->dd_object;
1338         VERIFY(0 == dsl_dir_open_obj(dd->dd_pool,
1339             ra->newparent->dd_object, NULL, dd, &dd->dd_parent));
1340
1341         /* add to new parent zapobj */
1342         err = zap_add(mos, ra->newparent->dd_phys->dd_child_dir_zapobj,
1343             dd->dd_myname, 8, 1, &dd->dd_object, tx);
1344         ASSERT0(err);
1345         dsl_dir_name(dd, newname);
1346 #ifdef _KERNEL
1347         zfsvfs_update_fromname(oldname, newname);
1348         zvol_rename_minors(oldname, newname);
1349 #endif
1350
1351         spa_history_log_internal(LOG_DS_RENAME, dd->dd_pool->dp_spa,
1352             tx, "dataset = %llu", dd->dd_phys->dd_head_dataset_obj);
1353 }
1354
1355 int
1356 dsl_dir_rename(dsl_dir_t *dd, const char *newname, int flags)
1357 {
1358         struct renamearg ra;
1359         int err;
1360
1361         /* new parent should exist */
1362         err = dsl_dir_open(newname, FTAG, &ra.newparent, &ra.mynewname);
1363         if (err)
1364                 return (err);
1365
1366         /* can't rename to different pool */
1367         if (dd->dd_pool != ra.newparent->dd_pool) {
1368                 err = ENXIO;
1369                 goto out;
1370         }
1371
1372         /* new name should not already exist */
1373         if (ra.mynewname == NULL) {
1374                 err = EEXIST;
1375                 goto out;
1376         }
1377
1378         ra.allowmounted = !!(flags & ZFS_RENAME_ALLOW_MOUNTED);
1379
1380         err = dsl_sync_task_do(dd->dd_pool,
1381             dsl_dir_rename_check, dsl_dir_rename_sync, dd, &ra, 3);
1382
1383 out:
1384         dsl_dir_close(ra.newparent, FTAG);
1385         return (err);
1386 }
1387
1388 int
1389 dsl_dir_transfer_possible(dsl_dir_t *sdd, dsl_dir_t *tdd, uint64_t space)
1390 {
1391         dsl_dir_t *ancestor;
1392         int64_t adelta;
1393         uint64_t avail;
1394
1395         ancestor = closest_common_ancestor(sdd, tdd);
1396         adelta = would_change(sdd, -space, ancestor);
1397         avail = dsl_dir_space_available(tdd, ancestor, adelta, FALSE);
1398         if (avail < space)
1399                 return (ENOSPC);
1400
1401         return (0);
1402 }
1403
1404 timestruc_t
1405 dsl_dir_snap_cmtime(dsl_dir_t *dd)
1406 {
1407         timestruc_t t;
1408
1409         mutex_enter(&dd->dd_lock);
1410         t = dd->dd_snap_cmtime;
1411         mutex_exit(&dd->dd_lock);
1412
1413         return (t);
1414 }
1415
1416 void
1417 dsl_dir_snap_cmtime_update(dsl_dir_t *dd)
1418 {
1419         timestruc_t t;
1420
1421         gethrestime(&t);
1422         mutex_enter(&dd->dd_lock);
1423         dd->dd_snap_cmtime = t;
1424         mutex_exit(&dd->dd_lock);
1425 }