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

   1 /*
   2  * CDDL HEADER START
   3  *
   4  * The contents of this file are subject to the terms of the
   5  * Common Development and Distribution License (the "License").
   6  * You may not use this file except in compliance with the License.
   7  *
   8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
   9  * or http://www.opensolaris.org/os/licensing.
  10  * See the License for the specific language governing permissions
  11  * and limitations under the License.
  12  *
  13  * When distributing Covered Code, include this CDDL HEADER in each
  14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
  15  * If applicable, add the following below this CDDL HEADER, with the
  16  * fields enclosed by brackets "[]" replaced with your own identifying
  17  * information: Portions Copyright [yyyy] [name of copyright owner]
  18  *
  19  * CDDL HEADER END
  20  */
  21 /*
  22  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
  23  * Copyright (c) 2011 Pawel Jakub Dawidek <pawel@dawidek.net>.
  24  * All rights reserved.
  25  * Copyright (c) 2012 by Delphix. All rights reserved.
  26  */
  27
  28 #include <sys/dmu.h>
  29 #include <sys/dmu_objset.h>
  30 #include <sys/dmu_tx.h>
  31 #include <sys/dsl_dataset.h>
  32 #include <sys/dsl_dir.h>
  33 #include <sys/dsl_prop.h>
  34 #include <sys/dsl_synctask.h>
  35 #include <sys/dsl_deleg.h>
  36 #include <sys/spa.h>
  37 #include <sys/metaslab.h>
  38 #include <sys/zap.h>
  39 #include <sys/zio.h>
  40 #include <sys/arc.h>
  41 #include <sys/sunddi.h>
  42 #include <sys/zvol.h>
  43 #ifdef _KERNEL
  44 #include <sys/zfs_vfsops.h>
  45 #endif
  46 #include "zfs_namecheck.h"
  47
  48 static uint64_t dsl_dir_space_towrite(dsl_dir_t *dd);
  49 static void dsl_dir_set_reservation_sync_impl(dsl_dir_t *dd,
  50     uint64_t value, dmu_tx_t *tx);
  51
  52 /* ARGSUSED */
  53 static void
  54 dsl_dir_evict(dmu_buf_t *db, void *arg)
  55 {
  56         dsl_dir_t *dd = arg;
  57         dsl_pool_t *dp = dd->dd_pool;
  58         int t;
  59
  60         for (t = 0; t < TXG_SIZE; t++) {
  61                 ASSERT(!txg_list_member(&dp->dp_dirty_dirs, dd, t));
  62                 ASSERT(dd->dd_tempreserved[t] == 0);
  63                 ASSERT(dd->dd_space_towrite[t] == 0);
  64         }
  65
  66         if (dd->dd_parent)
  67                 dsl_dir_close(dd->dd_parent, dd);
  68
  69         spa_close(dd->dd_pool->dp_spa, dd);
  70
  71         /*
  72          * The props callback list should have been cleaned up by
  73          * objset_evict().
  74          */
  75         list_destroy(&dd->dd_prop_cbs);
  76         mutex_destroy(&dd->dd_lock);
  77         kmem_free(dd, sizeof (dsl_dir_t));
  78 }
  79
  80 int
  81 dsl_dir_open_obj(dsl_pool_t *dp, uint64_t ddobj,
  82     const char *tail, void *tag, dsl_dir_t **ddp)
  83 {
  84         dmu_buf_t *dbuf;
  85         dsl_dir_t *dd;
  86         int err;
  87
  88         ASSERT(RW_LOCK_HELD(&dp->dp_config_rwlock) ||
  89             dsl_pool_sync_context(dp));
  90
  91         err = dmu_bonus_hold(dp->dp_meta_objset, ddobj, tag, &dbuf);
  92         if (err)
  93                 return (err);
  94         dd = dmu_buf_get_user(dbuf);
  95 #ifdef ZFS_DEBUG
  96         {
  97                 dmu_object_info_t doi;
  98                 dmu_object_info_from_db(dbuf, &doi);
  99                 ASSERT3U(doi.doi_type, ==, DMU_OT_DSL_DIR);
 100                 ASSERT3U(doi.doi_bonus_size, >=, sizeof (dsl_dir_phys_t));
 101         }
 102 #endif
 103         if (dd == NULL) {
 104                 dsl_dir_t *winner;
 105
 106                 dd = kmem_zalloc(sizeof (dsl_dir_t), KM_SLEEP);
 107                 dd->dd_object = ddobj;
 108                 dd->dd_dbuf = dbuf;
 109                 dd->dd_pool = dp;
 110                 dd->dd_phys = dbuf->db_data;
 111                 mutex_init(&dd->dd_lock, NULL, MUTEX_DEFAULT, NULL);
 112
 113                 list_create(&dd->dd_prop_cbs, sizeof (dsl_prop_cb_record_t),
 114                     offsetof(dsl_prop_cb_record_t, cbr_node));
 115
 116                 dsl_dir_snap_cmtime_update(dd);
 117
 118                 if (dd->dd_phys->dd_parent_obj) {
 119                         err = dsl_dir_open_obj(dp, dd->dd_phys->dd_parent_obj,
 120                             NULL, dd, &dd->dd_parent);
 121                         if (err)
 122                                 goto errout;
 123                         if (tail) {
 124 #ifdef ZFS_DEBUG
 125                                 uint64_t foundobj;
 126
 127                                 err = zap_lookup(dp->dp_meta_objset,
 128                                     dd->dd_parent->dd_phys->dd_child_dir_zapobj,
 129                                     tail, sizeof (foundobj), 1, &foundobj);
 130                                 ASSERT(err || foundobj == ddobj);
 131 #endif
 132                                 (void) strcpy(dd->dd_myname, tail);
 133                         } else {
 134                                 err = zap_value_search(dp->dp_meta_objset,
 135                                     dd->dd_parent->dd_phys->dd_child_dir_zapobj,
 136                                     ddobj, 0, dd->dd_myname);
 137                         }
 138                         if (err)
 139                                 goto errout;
 140                 } else {
 141                         (void) strcpy(dd->dd_myname, spa_name(dp->dp_spa));
 142                 }
 143
 144                 if (dsl_dir_is_clone(dd)) {
 145                         dmu_buf_t *origin_bonus;
 146                         dsl_dataset_phys_t *origin_phys;
 147
 148                         /*
 149                          * We can't open the origin dataset, because
 150                          * that would require opening this dsl_dir.
 151                          * Just look at its phys directly instead.
 152                          */
 153                         err = dmu_bonus_hold(dp->dp_meta_objset,
 154                             dd->dd_phys->dd_origin_obj, FTAG, &origin_bonus);
 155                         if (err)
 156                                 goto errout;
 157                         origin_phys = origin_bonus->db_data;
 158                         dd->dd_origin_txg =
 159                             origin_phys->ds_creation_txg;
 160                         dmu_buf_rele(origin_bonus, FTAG);
 161                 }
 162
 163                 winner = dmu_buf_set_user_ie(dbuf, dd, &dd->dd_phys,
 164                     dsl_dir_evict);
 165                 if (winner) {
 166                         if (dd->dd_parent)
 167                                 dsl_dir_close(dd->dd_parent, dd);
 168                         mutex_destroy(&dd->dd_lock);
 169                         kmem_free(dd, sizeof (dsl_dir_t));
 170                         dd = winner;
 171                 } else {
 172                         spa_open_ref(dp->dp_spa, dd);
 173                 }
 174         }
 175
 176         /*
 177          * The dsl_dir_t has both open-to-close and instantiate-to-evict
 178          * holds on the spa.  We need the open-to-close holds because
 179          * otherwise the spa_refcnt wouldn't change when we open a
 180          * dir which the spa also has open, so we could incorrectly
 181          * think it was OK to unload/export/destroy the pool.  We need
 182          * the instantiate-to-evict hold because the dsl_dir_t has a
 183          * pointer to the dd_pool, which has a pointer to the spa_t.
 184          */
 185         spa_open_ref(dp->dp_spa, tag);
 186         ASSERT3P(dd->dd_pool, ==, dp);
 187         ASSERT3U(dd->dd_object, ==, ddobj);
 188         ASSERT3P(dd->dd_dbuf, ==, dbuf);
 189         *ddp = dd;
 190         return (0);
 191
 192 errout:
 193         if (dd->dd_parent)
 194                 dsl_dir_close(dd->dd_parent, dd);
 195         mutex_destroy(&dd->dd_lock);
 196         kmem_free(dd, sizeof (dsl_dir_t));
 197         dmu_buf_rele(dbuf, tag);
 198         return (err);
 199 }
 200
 201 void
 202 dsl_dir_close(dsl_dir_t *dd, void *tag)
 203 {
 204         dprintf_dd(dd, "%s\n", "");
 205         spa_close(dd->dd_pool->dp_spa, tag);
 206         dmu_buf_rele(dd->dd_dbuf, tag);
 207 }
 208
 209 /* buf must be long enough (MAXNAMELEN + strlen(MOS_DIR_NAME) + 1 should do) */
 210 void
 211 dsl_dir_name(dsl_dir_t *dd, char *buf)
 212 {
 213         if (dd->dd_parent) {
 214                 dsl_dir_name(dd->dd_parent, buf);
 215                 (void) strcat(buf, "/");
 216         } else {
 217                 buf[0] = '\0';
 218         }
 219         if (!MUTEX_HELD(&dd->dd_lock)) {
 220                 /*
 221                  * recursive mutex so that we can use
 222                  * dprintf_dd() with dd_lock held
 223                  */
 224                 mutex_enter(&dd->dd_lock);
 225                 (void) strcat(buf, dd->dd_myname);
 226                 mutex_exit(&dd->dd_lock);
 227         } else {
 228                 (void) strcat(buf, dd->dd_myname);
 229         }
 230 }
 231
 232 /* Calculate name length, avoiding all the strcat calls of dsl_dir_name */
 233 int
 234 dsl_dir_namelen(dsl_dir_t *dd)
 235 {
 236         int result = 0;
 237
 238         if (dd->dd_parent) {
 239                 /* parent's name + 1 for the "/" */
 240                 result = dsl_dir_namelen(dd->dd_parent) + 1;
 241         }
 242
 243         if (!MUTEX_HELD(&dd->dd_lock)) {
 244                 /* see dsl_dir_name */
 245                 mutex_enter(&dd->dd_lock);
 246                 result += strlen(dd->dd_myname);
 247                 mutex_exit(&dd->dd_lock);
 248         } else {
 249                 result += strlen(dd->dd_myname);
 250         }
 251
 252         return (result);
 253 }
 254
 255 static int
 256 getcomponent(const char *path, char *component, const char **nextp)
 257 {
 258         char *p;
 259         if ((path == NULL) || (path[0] == '\0'))
 260                 return (ENOENT);
 261         /* This would be a good place to reserve some namespace... */
 262         p = strpbrk(path, "/@");
 263         if (p && (p[1] == '/' || p[1] == '@')) {
 264                 /* two separators in a row */
 265                 return (EINVAL);
 266         }
 267         if (p == NULL || p == path) {
 268                 /*
 269                  * if the first thing is an @ or /, it had better be an
 270                  * @ and it had better not have any more ats or slashes,
 271                  * and it had better have something after the @.
 272                  */
 273                 if (p != NULL &&
 274                     (p[0] != '@' || strpbrk(path+1, "/@") || p[1] == '\0'))
 275                         return (EINVAL);
 276                 if (strlen(path) >= MAXNAMELEN)
 277                         return (ENAMETOOLONG);
 278                 (void) strcpy(component, path);
 279                 p = NULL;
 280         } else if (p[0] == '/') {
 281                 if (p-path >= MAXNAMELEN)
 282                         return (ENAMETOOLONG);
 283                 (void) strncpy(component, path, p - path);
 284                 component[p-path] = '\0';
 285                 p++;
 286         } else if (p[0] == '@') {
 287                 /*
 288                  * if the next separator is an @, there better not be
 289                  * any more slashes.
 290                  */
 291                 if (strchr(path, '/'))
 292                         return (EINVAL);
 293                 if (p-path >= MAXNAMELEN)
 294                         return (ENAMETOOLONG);
 295                 (void) strncpy(component, path, p - path);
 296                 component[p-path] = '\0';
 297         } else {
 298                 ASSERT(!"invalid p");
 299         }
 300         *nextp = p;
 301         return (0);
 302 }
 303
 304 /*
 305  * same as dsl_open_dir, ignore the first component of name and use the
 306  * spa instead
 307  */
 308 int
 309 dsl_dir_open_spa(spa_t *spa, const char *name, void *tag,
 310     dsl_dir_t **ddp, const char **tailp)
 311 {
 312         char buf[MAXNAMELEN];
 313         const char *next, *nextnext = NULL;
 314         int err;
 315         dsl_dir_t *dd;
 316         dsl_pool_t *dp;
 317         uint64_t ddobj;
 318         int openedspa = FALSE;
 319
 320         dprintf("%s\n", name);
 321
 322         err = getcomponent(name, buf, &next);
 323         if (err)
 324                 return (err);
 325         if (spa == NULL) {
 326                 err = spa_open(buf, &spa, FTAG);
 327                 if (err) {
 328                         dprintf("spa_open(%s) failed\n", buf);
 329                         return (err);
 330                 }
 331                 openedspa = TRUE;
 332
 333                 /* XXX this assertion belongs in spa_open */
 334                 ASSERT(!dsl_pool_sync_context(spa_get_dsl(spa)));
 335         }
 336
 337         dp = spa_get_dsl(spa);
 338
 339         rw_enter(&dp->dp_config_rwlock, RW_READER);
 340         err = dsl_dir_open_obj(dp, dp->dp_root_dir_obj, NULL, tag, &dd);
 341         if (err) {
 342                 rw_exit(&dp->dp_config_rwlock);
 343                 if (openedspa)
 344                         spa_close(spa, FTAG);
 345                 return (err);
 346         }
 347
 348         while (next != NULL) {
 349                 dsl_dir_t *child_ds;
 350                 err = getcomponent(next, buf, &nextnext);
 351                 if (err)
 352                         break;
 353                 ASSERT(next[0] != '\0');
 354                 if (next[0] == '@')
 355                         break;
 356                 dprintf("looking up %s in obj%lld\n",
 357                     buf, dd->dd_phys->dd_child_dir_zapobj);
 358
 359                 err = zap_lookup(dp->dp_meta_objset,
 360                     dd->dd_phys->dd_child_dir_zapobj,
 361                     buf, sizeof (ddobj), 1, &ddobj);
 362                 if (err) {
 363                         if (err == ENOENT)
 364                                 err = 0;
 365                         break;
 366                 }
 367
 368                 err = dsl_dir_open_obj(dp, ddobj, buf, tag, &child_ds);
 369                 if (err)
 370                         break;
 371                 dsl_dir_close(dd, tag);
 372                 dd = child_ds;
 373                 next = nextnext;
 374         }
 375         rw_exit(&dp->dp_config_rwlock);
 376
 377         if (err) {
 378                 dsl_dir_close(dd, tag);
 379                 if (openedspa)
 380                         spa_close(spa, FTAG);
 381                 return (err);
 382         }
 383
 384         /*
 385          * It's an error if there's more than one component left, or
 386          * tailp==NULL and there's any component left.
 387          */
 388         if (next != NULL &&
 389             (tailp == NULL || (nextnext && nextnext[0] != '\0'))) {
 390                 /* bad path name */
 391                 dsl_dir_close(dd, tag);
 392                 dprintf("next=%p (%s) tail=%p\n", next, next?next:"", tailp);
 393                 err = ENOENT;
 394         }
 395         if (tailp)
 396                 *tailp = next;
 397         if (openedspa)
 398                 spa_close(spa, FTAG);
 399         *ddp = dd;
 400         return (err);
 401 }
 402
 403 /*
 404  * Return the dsl_dir_t, and possibly the last component which couldn't
 405  * be found in *tail.  Return NULL if the path is bogus, or if
 406  * tail==NULL and we couldn't parse the whole name.  (*tail)[0] == '@'
 407  * means that the last component is a snapshot.
 408  */
 409 int
 410 dsl_dir_open(const char *name, void *tag, dsl_dir_t **ddp, const char **tailp)
 411 {
 412         return (dsl_dir_open_spa(NULL, name, tag, ddp, tailp));
 413 }
 414
 415 uint64_t
 416 dsl_dir_create_sync(dsl_pool_t *dp, dsl_dir_t *pds, const char *name,
 417     dmu_tx_t *tx)
 418 {
 419         objset_t *mos = dp->dp_meta_objset;
 420         uint64_t ddobj;
 421         dsl_dir_phys_t *ddphys;
 422         dmu_buf_t *dbuf;
 423
 424         ddobj = dmu_object_alloc(mos, DMU_OT_DSL_DIR, 0,
 425             DMU_OT_DSL_DIR, sizeof (dsl_dir_phys_t), tx);
 426         if (pds) {
 427                 VERIFY(0 == zap_add(mos, pds->dd_phys->dd_child_dir_zapobj,
 428                     name, sizeof (uint64_t), 1, &ddobj, tx));
 429         } else {
 430                 /* it's the root dir */
 431                 VERIFY(0 == zap_add(mos, DMU_POOL_DIRECTORY_OBJECT,
 432                     DMU_POOL_ROOT_DATASET, sizeof (uint64_t), 1, &ddobj, tx));
 433         }
 434         VERIFY(0 == dmu_bonus_hold(mos, ddobj, FTAG, &dbuf));
 435         dmu_buf_will_dirty(dbuf, tx);
 436         ddphys = dbuf->db_data;
 437
 438         ddphys->dd_creation_time = gethrestime_sec();
 439         if (pds)
 440                 ddphys->dd_parent_obj = pds->dd_object;
 441         ddphys->dd_props_zapobj = zap_create(mos,
 442             DMU_OT_DSL_PROPS, DMU_OT_NONE, 0, tx);
 443         ddphys->dd_child_dir_zapobj = zap_create(mos,
 444             DMU_OT_DSL_DIR_CHILD_MAP, DMU_OT_NONE, 0, tx);
 445         if (spa_version(dp->dp_spa) >= SPA_VERSION_USED_BREAKDOWN)
 446                 ddphys->dd_flags |= DD_FLAG_USED_BREAKDOWN;
 447         dmu_buf_rele(dbuf, FTAG);
 448
 449         return (ddobj);
 450 }
 451
 452 /* ARGSUSED */
 453 int
 454 dsl_dir_destroy_check(void *arg1, void *arg2, dmu_tx_t *tx)
 455 {
 456         dsl_dir_t *dd = arg1;
 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_dir_t *dd = arg1;
 488         objset_t *mos = dd->dd_pool->dp_meta_objset;
 489         uint64_t obj;
 490         dd_used_t t;
 491
 492         ASSERT(RW_WRITE_HELD(&dd->dd_pool->dp_config_rwlock));
 493         ASSERT(dd->dd_phys->dd_head_dataset_obj == 0);
 494
 495         /*
 496          * Remove our reservation. The impl() routine avoids setting the
 497          * actual property, which would require the (already destroyed) ds.
 498          */
 499         dsl_dir_set_reservation_sync_impl(dd, 0, tx);
 500
 501         ASSERT0(dd->dd_phys->dd_used_bytes);
 502         ASSERT0(dd->dd_phys->dd_reserved);
 503         for (t = 0; t < DD_USED_NUM; t++)
 504                 ASSERT0(dd->dd_phys->dd_used_breakdown[t]);
 505
 506         VERIFY(0 == zap_destroy(mos, dd->dd_phys->dd_child_dir_zapobj, tx));
 507         VERIFY(0 == zap_destroy(mos, dd->dd_phys->dd_props_zapobj, tx));
 508         VERIFY(0 == dsl_deleg_destroy(mos, dd->dd_phys->dd_deleg_zapobj, tx));
 509         VERIFY(0 == zap_remove(mos,
 510             dd->dd_parent->dd_phys->dd_child_dir_zapobj, dd->dd_myname, tx));
 511
 512         obj = dd->dd_object;
 513         dsl_dir_close(dd, tag);
 514         VERIFY(0 == dmu_object_free(mos, obj, tx));
 515 }
 516
 517 boolean_t
 518 dsl_dir_is_clone(dsl_dir_t *dd)
 519 {
 520         return (dd->dd_phys->dd_origin_obj &&
 521             (dd->dd_pool->dp_origin_snap == NULL ||
 522             dd->dd_phys->dd_origin_obj !=
 523             dd->dd_pool->dp_origin_snap->ds_object));
 524 }
 525
 526 void
 527 dsl_dir_stats(dsl_dir_t *dd, nvlist_t *nv)
 528 {
 529         mutex_enter(&dd->dd_lock);
 530         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USED,
 531             dd->dd_phys->dd_used_bytes);
 532         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_QUOTA, dd->dd_phys->dd_quota);
 533         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_RESERVATION,
 534             dd->dd_phys->dd_reserved);
 535         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_COMPRESSRATIO,
 536             dd->dd_phys->dd_compressed_bytes == 0 ? 100 :
 537             (dd->dd_phys->dd_uncompressed_bytes * 100 /
 538             dd->dd_phys->dd_compressed_bytes));
 539         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_LOGICALUSED,
 540             dd->dd_phys->dd_uncompressed_bytes);
 541         if (dd->dd_phys->dd_flags & DD_FLAG_USED_BREAKDOWN) {
 542                 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDSNAP,
 543                     dd->dd_phys->dd_used_breakdown[DD_USED_SNAP]);
 544                 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDDS,
 545                     dd->dd_phys->dd_used_breakdown[DD_USED_HEAD]);
 546                 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDREFRESERV,
 547                     dd->dd_phys->dd_used_breakdown[DD_USED_REFRSRV]);
 548                 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDCHILD,
 549                     dd->dd_phys->dd_used_breakdown[DD_USED_CHILD] +
 550                     dd->dd_phys->dd_used_breakdown[DD_USED_CHILD_RSRV]);
 551         }
 552         mutex_exit(&dd->dd_lock);
 553
 554         rw_enter(&dd->dd_pool->dp_config_rwlock, RW_READER);
 555         if (dsl_dir_is_clone(dd)) {
 556                 dsl_dataset_t *ds;
 557                 char buf[MAXNAMELEN];
 558
 559                 VERIFY(0 == dsl_dataset_hold_obj(dd->dd_pool,
 560                     dd->dd_phys->dd_origin_obj, FTAG, &ds));
 561                 dsl_dataset_name(ds, buf);
 562                 dsl_dataset_rele(ds, FTAG);
 563                 dsl_prop_nvlist_add_string(nv, ZFS_PROP_ORIGIN, buf);
 564         }
 565         rw_exit(&dd->dd_pool->dp_config_rwlock);
 566 }
 567
 568 void
 569 dsl_dir_dirty(dsl_dir_t *dd, dmu_tx_t *tx)
 570 {
 571         dsl_pool_t *dp = dd->dd_pool;
 572
 573         ASSERT(dd->dd_phys);
 574
 575         if (txg_list_add(&dp->dp_dirty_dirs, dd, tx->tx_txg) == 0) {
 576                 /* up the hold count until we can be written out */
 577                 dmu_buf_add_ref(dd->dd_dbuf, dd);
 578         }
 579 }
 580
 581 static int64_t
 582 parent_delta(dsl_dir_t *dd, uint64_t used, int64_t delta)
 583 {
 584         uint64_t old_accounted = MAX(used, dd->dd_phys->dd_reserved);
 585         uint64_t new_accounted = MAX(used + delta, dd->dd_phys->dd_reserved);
 586         return (new_accounted - old_accounted);
 587 }
 588
 589 void
 590 dsl_dir_sync(dsl_dir_t *dd, dmu_tx_t *tx)
 591 {
 592         ASSERT(dmu_tx_is_syncing(tx));
 593
 594         mutex_enter(&dd->dd_lock);
 595         ASSERT0(dd->dd_tempreserved[tx->tx_txg&TXG_MASK]);
 596         dprintf_dd(dd, "txg=%llu towrite=%lluK\n", tx->tx_txg,
 597             dd->dd_space_towrite[tx->tx_txg&TXG_MASK] / 1024);
 598         dd->dd_space_towrite[tx->tx_txg&TXG_MASK] = 0;
 599         mutex_exit(&dd->dd_lock);
 600
 601         /* release the hold from dsl_dir_dirty */
 602         dmu_buf_rele(dd->dd_dbuf, dd);
 603 }
 604
 605 static uint64_t
 606 dsl_dir_space_towrite(dsl_dir_t *dd)
 607 {
 608         uint64_t space = 0;
 609         int i;
 610
 611         ASSERT(MUTEX_HELD(&dd->dd_lock));
 612
 613         for (i = 0; i < TXG_SIZE; i++) {
 614                 space += dd->dd_space_towrite[i&TXG_MASK];
 615                 ASSERT3U(dd->dd_space_towrite[i&TXG_MASK], >=, 0);
 616         }
 617         return (space);
 618 }
 619
 620 /*
 621  * How much space would dd have available if ancestor had delta applied
 622  * to it?  If ondiskonly is set, we're only interested in what's
 623  * on-disk, not estimated pending changes.
 624  */
 625 uint64_t
 626 dsl_dir_space_available(dsl_dir_t *dd,
 627     dsl_dir_t *ancestor, int64_t delta, int ondiskonly)
 628 {
 629         uint64_t parentspace, myspace, quota, used;
 630
 631         /*
 632          * If there are no restrictions otherwise, assume we have
 633          * unlimited space available.
 634          */
 635         quota = UINT64_MAX;
 636         parentspace = UINT64_MAX;
 637
 638         if (dd->dd_parent != NULL) {
 639                 parentspace = dsl_dir_space_available(dd->dd_parent,
 640                     ancestor, delta, ondiskonly);
 641         }
 642
 643         mutex_enter(&dd->dd_lock);
 644         if (dd->dd_phys->dd_quota != 0)
 645                 quota = dd->dd_phys->dd_quota;
 646         used = dd->dd_phys->dd_used_bytes;
 647         if (!ondiskonly)
 648                 used += dsl_dir_space_towrite(dd);
 649
 650         if (dd->dd_parent == NULL) {
 651                 uint64_t poolsize = dsl_pool_adjustedsize(dd->dd_pool, FALSE);
 652                 quota = MIN(quota, poolsize);
 653         }
 654
 655         if (dd->dd_phys->dd_reserved > used && parentspace != UINT64_MAX) {
 656                 /*
 657                  * We have some space reserved, in addition to what our
 658                  * parent gave us.
 659                  */
 660                 parentspace += dd->dd_phys->dd_reserved - used;
 661         }
 662
 663         if (dd == ancestor) {
 664                 ASSERT(delta <= 0);
 665                 ASSERT(used >= -delta);
 666                 used += delta;
 667                 if (parentspace != UINT64_MAX)
 668                         parentspace -= delta;
 669         }
 670
 671         if (used > quota) {
 672                 /* over quota */
 673                 myspace = 0;
 674         } else {
 675                 /*
 676                  * the lesser of the space provided by our parent and
 677                  * the space left in our quota
 678                  */
 679                 myspace = MIN(parentspace, quota - used);
 680         }
 681
 682         mutex_exit(&dd->dd_lock);
 683
 684         return (myspace);
 685 }
 686
 687 struct tempreserve {
 688         list_node_t tr_node;
 689         dsl_pool_t *tr_dp;
 690         dsl_dir_t *tr_ds;
 691         uint64_t tr_size;
 692 };
 693
 694 static int
 695 dsl_dir_tempreserve_impl(dsl_dir_t *dd, uint64_t asize, boolean_t netfree,
 696     boolean_t ignorequota, boolean_t checkrefquota, list_t *tr_list,
 697     dmu_tx_t *tx, boolean_t first)
 698 {
 699         uint64_t txg = tx->tx_txg;
 700         uint64_t est_inflight, used_on_disk, quota, parent_rsrv;
 701         uint64_t deferred = 0;
 702         struct tempreserve *tr;
 703         int retval = EDQUOT;
 704         int txgidx = txg & TXG_MASK;
 705         int i;
 706         uint64_t ref_rsrv = 0;
 707
 708         ASSERT3U(txg, !=, 0);
 709         ASSERT3S(asize, >, 0);
 710
 711         mutex_enter(&dd->dd_lock);
 712
 713         /*
 714          * Check against the dsl_dir's quota.  We don't add in the delta
 715          * when checking for over-quota because they get one free hit.
 716          */
 717         est_inflight = dsl_dir_space_towrite(dd);
 718         for (i = 0; i < TXG_SIZE; i++)
 719                 est_inflight += dd->dd_tempreserved[i];
 720         used_on_disk = dd->dd_phys->dd_used_bytes;
 721
 722         /*
 723          * On the first iteration, fetch the dataset's used-on-disk and
 724          * refreservation values. Also, if checkrefquota is set, test if
 725          * allocating this space would exceed the dataset's refquota.
 726          */
 727         if (first && tx->tx_objset) {
 728                 int error;
 729                 dsl_dataset_t *ds = tx->tx_objset->os_dsl_dataset;
 730
 731                 error = dsl_dataset_check_quota(ds, checkrefquota,
 732                     asize, est_inflight, &used_on_disk, &ref_rsrv);
 733                 if (error) {
 734                         mutex_exit(&dd->dd_lock);
 735                         return (error);
 736                 }
 737         }
 738
 739         /*
 740          * If this transaction will result in a net free of space,
 741          * we want to let it through.
 742          */
 743         if (ignorequota || netfree || dd->dd_phys->dd_quota == 0)
 744                 quota = UINT64_MAX;
 745         else
 746                 quota = dd->dd_phys->dd_quota;
 747
 748         /*
 749          * Adjust the quota against the actual pool size at the root
 750          * minus any outstanding deferred frees.
 751          * To ensure that it's possible to remove files from a full
 752          * pool without inducing transient overcommits, we throttle
 753          * netfree transactions against a quota that is slightly larger,
 754          * but still within the pool's allocation slop.  In cases where
 755          * we're very close to full, this will allow a steady trickle of
 756          * removes to get through.
 757          */
 758         if (dd->dd_parent == NULL) {
 759                 spa_t *spa = dd->dd_pool->dp_spa;
 760                 uint64_t poolsize = dsl_pool_adjustedsize(dd->dd_pool, netfree);
 761                 deferred = metaslab_class_get_deferred(spa_normal_class(spa));
 762                 if (poolsize - deferred < quota) {
 763                         quota = poolsize - deferred;
 764                         retval = ENOSPC;
 765                 }
 766         }
 767
 768         /*
 769          * If they are requesting more space, and our current estimate
 770          * is over quota, they get to try again unless the actual
 771          * on-disk is over quota and there are no pending changes (which
 772          * may free up space for us).
 773          */
 774         if (used_on_disk + est_inflight >= quota) {
 775                 if (est_inflight > 0 || used_on_disk < quota ||
 776                     (retval == ENOSPC && used_on_disk < quota + deferred))
 777                         retval = ERESTART;
 778                 dprintf_dd(dd, "failing: used=%lluK inflight = %lluK "
 779                     "quota=%lluK tr=%lluK err=%d\n",
 780                     used_on_disk>>10, est_inflight>>10,
 781                     quota>>10, asize>>10, retval);
 782                 mutex_exit(&dd->dd_lock);
 783                 return (retval);
 784         }
 785
 786         /* We need to up our estimated delta before dropping dd_lock */
 787         dd->dd_tempreserved[txgidx] += asize;
 788
 789         parent_rsrv = parent_delta(dd, used_on_disk + est_inflight,
 790             asize - ref_rsrv);
 791         mutex_exit(&dd->dd_lock);
 792
 793         tr = kmem_zalloc(sizeof (struct tempreserve), KM_SLEEP);
 794         tr->tr_ds = dd;
 795         tr->tr_size = asize;
 796         list_insert_tail(tr_list, tr);
 797
 798         /* see if it's OK with our parent */
 799         if (dd->dd_parent && parent_rsrv) {
 800                 boolean_t ismos = (dd->dd_phys->dd_head_dataset_obj == 0);
 801
 802                 return (dsl_dir_tempreserve_impl(dd->dd_parent,
 803                     parent_rsrv, netfree, ismos, TRUE, tr_list, tx, FALSE));
 804         } else {
 805                 return (0);
 806         }
 807 }
 808
 809 /*
 810  * Reserve space in this dsl_dir, to be used in this tx's txg.
 811  * After the space has been dirtied (and dsl_dir_willuse_space()
 812  * has been called), the reservation should be canceled, using
 813  * dsl_dir_tempreserve_clear().
 814  */
 815 int
 816 dsl_dir_tempreserve_space(dsl_dir_t *dd, uint64_t lsize, uint64_t asize,
 817     uint64_t fsize, uint64_t usize, void **tr_cookiep, dmu_tx_t *tx)
 818 {
 819         int err;
 820         list_t *tr_list;
 821
 822         if (asize == 0) {
 823                 *tr_cookiep = NULL;
 824                 return (0);
 825         }
 826
 827         tr_list = kmem_alloc(sizeof (list_t), KM_SLEEP);
 828         list_create(tr_list, sizeof (struct tempreserve),
 829             offsetof(struct tempreserve, tr_node));
 830         ASSERT3S(asize, >, 0);
 831         ASSERT3S(fsize, >=, 0);
 832
 833         err = arc_tempreserve_space(lsize, tx->tx_txg);
 834         if (err == 0) {
 835                 struct tempreserve *tr;
 836
 837                 tr = kmem_zalloc(sizeof (struct tempreserve), KM_SLEEP);
 838                 tr->tr_size = lsize;
 839                 list_insert_tail(tr_list, tr);
 840
 841                 err = dsl_pool_tempreserve_space(dd->dd_pool, asize, tx);
 842         } else {
 843                 if (err == EAGAIN) {
 844                         txg_delay(dd->dd_pool, tx->tx_txg, 1);
 845                         err = ERESTART;
 846                 }
 847                 dsl_pool_memory_pressure(dd->dd_pool);
 848         }
 849
 850         if (err == 0) {
 851                 struct tempreserve *tr;
 852
 853                 tr = kmem_zalloc(sizeof (struct tempreserve), KM_SLEEP);
 854                 tr->tr_dp = dd->dd_pool;
 855                 tr->tr_size = asize;
 856                 list_insert_tail(tr_list, tr);
 857
 858                 err = dsl_dir_tempreserve_impl(dd, asize, fsize >= asize,
 859                     FALSE, asize > usize, tr_list, tx, TRUE);
 860         }
 861
 862         if (err)
 863                 dsl_dir_tempreserve_clear(tr_list, tx);
 864         else
 865                 *tr_cookiep = tr_list;
 866
 867         return (err);
 868 }
 869
 870 /*
 871  * Clear a temporary reservation that we previously made with
 872  * dsl_dir_tempreserve_space().
 873  */
 874 void
 875 dsl_dir_tempreserve_clear(void *tr_cookie, dmu_tx_t *tx)
 876 {
 877         int txgidx = tx->tx_txg & TXG_MASK;
 878         list_t *tr_list = tr_cookie;
 879         struct tempreserve *tr;
 880
 881         ASSERT3U(tx->tx_txg, !=, 0);
 882
 883         if (tr_cookie == NULL)
 884                 return;
 885
 886         while (tr = list_head(tr_list)) {
 887                 if (tr->tr_dp) {
 888                         dsl_pool_tempreserve_clear(tr->tr_dp, tr->tr_size, tx);
 889                 } else if (tr->tr_ds) {
 890                         mutex_enter(&tr->tr_ds->dd_lock);
 891                         ASSERT3U(tr->tr_ds->dd_tempreserved[txgidx], >=,
 892                             tr->tr_size);
 893                         tr->tr_ds->dd_tempreserved[txgidx] -= tr->tr_size;
 894                         mutex_exit(&tr->tr_ds->dd_lock);
 895                 } else {
 896                         arc_tempreserve_clear(tr->tr_size);
 897                 }
 898                 list_remove(tr_list, tr);
 899                 kmem_free(tr, sizeof (struct tempreserve));
 900         }
 901
 902         kmem_free(tr_list, sizeof (list_t));
 903 }
 904
 905 static void
 906 dsl_dir_willuse_space_impl(dsl_dir_t *dd, int64_t space, dmu_tx_t *tx)
 907 {
 908         int64_t parent_space;
 909         uint64_t est_used;
 910
 911         mutex_enter(&dd->dd_lock);
 912         if (space > 0)
 913                 dd->dd_space_towrite[tx->tx_txg & TXG_MASK] += space;
 914
 915         est_used = dsl_dir_space_towrite(dd) + dd->dd_phys->dd_used_bytes;
 916         parent_space = parent_delta(dd, est_used, space);
 917         mutex_exit(&dd->dd_lock);
 918
 919         /* Make sure that we clean up dd_space_to* */
 920         dsl_dir_dirty(dd, tx);
 921
 922         /* XXX this is potentially expensive and unnecessary... */
 923         if (parent_space && dd->dd_parent)
 924                 dsl_dir_willuse_space_impl(dd->dd_parent, parent_space, tx);
 925 }
 926
 927 /*
 928  * Call in open context when we think we're going to write/free space,
 929  * eg. when dirtying data.  Be conservative (ie. OK to write less than
 930  * this or free more than this, but don't write more or free less).
 931  */
 932 void
 933 dsl_dir_willuse_space(dsl_dir_t *dd, int64_t space, dmu_tx_t *tx)
 934 {
 935         dsl_pool_willuse_space(dd->dd_pool, space, tx);
 936         dsl_dir_willuse_space_impl(dd, space, tx);
 937 }
 938
 939 /* call from syncing context when we actually write/free space for this dd */
 940 void
 941 dsl_dir_diduse_space(dsl_dir_t *dd, dd_used_t type,
 942     int64_t used, int64_t compressed, int64_t uncompressed, dmu_tx_t *tx)
 943 {
 944         int64_t accounted_delta;
 945         boolean_t needlock = !MUTEX_HELD(&dd->dd_lock);
 946
 947         ASSERT(dmu_tx_is_syncing(tx));
 948         ASSERT(type < DD_USED_NUM);
 949
 950         if (needlock)
 951                 mutex_enter(&dd->dd_lock);
 952         accounted_delta = parent_delta(dd, dd->dd_phys->dd_used_bytes, used);
 953         ASSERT(used >= 0 || dd->dd_phys->dd_used_bytes >= -used);
 954         ASSERT(compressed >= 0 ||
 955             dd->dd_phys->dd_compressed_bytes >= -compressed);
 956         ASSERT(uncompressed >= 0 ||
 957             dd->dd_phys->dd_uncompressed_bytes >= -uncompressed);
 958         dmu_buf_will_dirty(dd->dd_dbuf, tx);
 959         dd->dd_phys->dd_used_bytes += used;
 960         dd->dd_phys->dd_uncompressed_bytes += uncompressed;
 961         dd->dd_phys->dd_compressed_bytes += compressed;
 962
 963         if (dd->dd_phys->dd_flags & DD_FLAG_USED_BREAKDOWN) {
 964                 ASSERT(used > 0 ||
 965                     dd->dd_phys->dd_used_breakdown[type] >= -used);
 966                 dd->dd_phys->dd_used_breakdown[type] += used;
 967 #ifdef DEBUG
 968                 dd_used_t t;
 969                 uint64_t u = 0;
 970                 for (t = 0; t < DD_USED_NUM; t++)
 971                         u += dd->dd_phys->dd_used_breakdown[t];
 972                 ASSERT3U(u, ==, dd->dd_phys->dd_used_bytes);
 973 #endif
 974         }
 975         if (needlock)
 976                 mutex_exit(&dd->dd_lock);
 977
 978         if (dd->dd_parent != NULL) {
 979                 dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD,
 980                     accounted_delta, compressed, uncompressed, tx);
 981                 dsl_dir_transfer_space(dd->dd_parent,
 982                     used - accounted_delta,
 983                     DD_USED_CHILD_RSRV, DD_USED_CHILD, tx);
 984         }
 985 }
 986
 987 void
 988 dsl_dir_transfer_space(dsl_dir_t *dd, int64_t delta,
 989     dd_used_t oldtype, dd_used_t newtype, dmu_tx_t *tx)
 990 {
 991         boolean_t needlock = !MUTEX_HELD(&dd->dd_lock);
 992
 993         ASSERT(dmu_tx_is_syncing(tx));
 994         ASSERT(oldtype < DD_USED_NUM);
 995         ASSERT(newtype < DD_USED_NUM);
 996
 997         if (delta == 0 || !(dd->dd_phys->dd_flags & DD_FLAG_USED_BREAKDOWN))
 998                 return;
 999
1000         if (needlock)
1001                 mutex_enter(&dd->dd_lock);
1002         ASSERT(delta > 0 ?
1003             dd->dd_phys->dd_used_breakdown[oldtype] >= delta :
1004             dd->dd_phys->dd_used_breakdown[newtype] >= -delta);
1005         ASSERT(dd->dd_phys->dd_used_bytes >= ABS(delta));
1006         dmu_buf_will_dirty(dd->dd_dbuf, tx);
1007         dd->dd_phys->dd_used_breakdown[oldtype] -= delta;
1008         dd->dd_phys->dd_used_breakdown[newtype] += delta;
1009         if (needlock)
1010                 mutex_exit(&dd->dd_lock);
1011 }
1012
1013 static int
1014 dsl_dir_set_quota_check(void *arg1, void *arg2, dmu_tx_t *tx)
1015 {
1016         dsl_dataset_t *ds = arg1;
1017         dsl_dir_t *dd = ds->ds_dir;
1018         dsl_prop_setarg_t *psa = arg2;
1019         int err;
1020         uint64_t towrite;
1021
1022         if ((err = dsl_prop_predict_sync(ds->ds_dir, psa)) != 0)
1023                 return (err);
1024
1025         if (psa->psa_effective_value == 0)
1026                 return (0);
1027
1028         mutex_enter(&dd->dd_lock);
1029         /*
1030          * If we are doing the preliminary check in open context, and
1031          * there are pending changes, then don't fail it, since the
1032          * pending changes could under-estimate the amount of space to be
1033          * freed up.
1034          */
1035         towrite = dsl_dir_space_towrite(dd);
1036         if ((dmu_tx_is_syncing(tx) || towrite == 0) &&
1037             (psa->psa_effective_value < dd->dd_phys->dd_reserved ||
1038             psa->psa_effective_value < dd->dd_phys->dd_used_bytes + towrite)) {
1039                 err = ENOSPC;
1040         }
1041         mutex_exit(&dd->dd_lock);
1042         return (err);
1043 }
1044
1045 extern dsl_syncfunc_t dsl_prop_set_sync;
1046
1047 static void
1048 dsl_dir_set_quota_sync(void *arg1, void *arg2, dmu_tx_t *tx)
1049 {
1050         dsl_dataset_t *ds = arg1;
1051         dsl_dir_t *dd = ds->ds_dir;
1052         dsl_prop_setarg_t *psa = arg2;
1053         uint64_t effective_value = psa->psa_effective_value;
1054
1055         dsl_prop_set_sync(ds, psa, tx);
1056         DSL_PROP_CHECK_PREDICTION(dd, psa);
1057
1058         dmu_buf_will_dirty(dd->dd_dbuf, tx);
1059
1060         mutex_enter(&dd->dd_lock);
1061         dd->dd_phys->dd_quota = effective_value;
1062         mutex_exit(&dd->dd_lock);
1063 }
1064
1065 int
1066 dsl_dir_set_quota(const char *ddname, zprop_source_t source, uint64_t quota)
1067 {
1068         dsl_dir_t *dd;
1069         dsl_dataset_t *ds;
1070         dsl_prop_setarg_t psa;
1071         int err;
1072
1073         dsl_prop_setarg_init_uint64(&psa, "quota", source, &quota);
1074
1075         err = dsl_dataset_hold(ddname, FTAG, &ds);
1076         if (err)
1077                 return (err);
1078
1079         err = dsl_dir_open(ddname, FTAG, &dd, NULL);
1080         if (err) {
1081                 dsl_dataset_rele(ds, FTAG);
1082                 return (err);
1083         }
1084
1085         ASSERT(ds->ds_dir == dd);
1086
1087         /*
1088          * If someone removes a file, then tries to set the quota, we want to
1089          * make sure the file freeing takes effect.
1090          */
1091         txg_wait_open(dd->dd_pool, 0);
1092
1093         err = dsl_sync_task_do(dd->dd_pool, dsl_dir_set_quota_check,
1094             dsl_dir_set_quota_sync, ds, &psa, 0);
1095
1096         dsl_dir_close(dd, FTAG);
1097         dsl_dataset_rele(ds, FTAG);
1098         return (err);
1099 }
1100
1101 int
1102 dsl_dir_set_reservation_check(void *arg1, void *arg2, dmu_tx_t *tx)
1103 {
1104         dsl_dataset_t *ds = arg1;
1105         dsl_dir_t *dd = ds->ds_dir;
1106         dsl_prop_setarg_t *psa = arg2;
1107         uint64_t effective_value;
1108         uint64_t used, avail;
1109         int err;
1110
1111         if ((err = dsl_prop_predict_sync(ds->ds_dir, psa)) != 0)
1112                 return (err);
1113
1114         effective_value = psa->psa_effective_value;
1115
1116         /*
1117          * If we are doing the preliminary check in open context, the
1118          * space estimates may be inaccurate.
1119          */
1120         if (!dmu_tx_is_syncing(tx))
1121                 return (0);
1122
1123         mutex_enter(&dd->dd_lock);
1124         used = dd->dd_phys->dd_used_bytes;
1125         mutex_exit(&dd->dd_lock);
1126
1127         if (dd->dd_parent) {
1128                 avail = dsl_dir_space_available(dd->dd_parent,
1129                     NULL, 0, FALSE);
1130         } else {
1131                 avail = dsl_pool_adjustedsize(dd->dd_pool, B_FALSE) - used;
1132         }
1133
1134         if (MAX(used, effective_value) > MAX(used, dd->dd_phys->dd_reserved)) {
1135                 uint64_t delta = MAX(used, effective_value) -
1136                     MAX(used, dd->dd_phys->dd_reserved);
1137
1138                 if (delta > avail)
1139                         return (ENOSPC);
1140                 if (dd->dd_phys->dd_quota > 0 &&
1141                     effective_value > dd->dd_phys->dd_quota)
1142                         return (ENOSPC);
1143         }
1144
1145         return (0);
1146 }
1147
1148 static void
1149 dsl_dir_set_reservation_sync_impl(dsl_dir_t *dd, uint64_t value, dmu_tx_t *tx)
1150 {
1151         uint64_t used;
1152         int64_t delta;
1153
1154         dmu_buf_will_dirty(dd->dd_dbuf, tx);
1155
1156         mutex_enter(&dd->dd_lock);
1157         used = dd->dd_phys->dd_used_bytes;
1158         delta = MAX(used, value) - MAX(used, dd->dd_phys->dd_reserved);
1159         dd->dd_phys->dd_reserved = value;
1160
1161         if (dd->dd_parent != NULL) {
1162                 /* Roll up this additional usage into our ancestors */
1163                 dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD_RSRV,
1164                     delta, 0, 0, tx);
1165         }
1166         mutex_exit(&dd->dd_lock);
1167 }
1168
1169 static void
1170 dsl_dir_set_reservation_sync(void *arg1, void *arg2, dmu_tx_t *tx)
1171 {
1172         dsl_dataset_t *ds = arg1;
1173         dsl_dir_t *dd = ds->ds_dir;
1174         dsl_prop_setarg_t *psa = arg2;
1175         uint64_t value = psa->psa_effective_value;
1176
1177         dsl_prop_set_sync(ds, psa, tx);
1178         DSL_PROP_CHECK_PREDICTION(dd, psa);
1179
1180         dsl_dir_set_reservation_sync_impl(dd, value, tx);
1181
1182         spa_history_log_internal_dd(dd, "set reservation", tx,
1183             "reservation=%lld", (longlong_t)value);
1184 }
1185
1186 int
1187 dsl_dir_set_reservation(const char *ddname, zprop_source_t source,
1188     uint64_t reservation)
1189 {
1190         dsl_dir_t *dd;
1191         dsl_dataset_t *ds;
1192         dsl_prop_setarg_t psa;
1193         int err;
1194
1195         dsl_prop_setarg_init_uint64(&psa, "reservation", source, &reservation);
1196
1197         err = dsl_dataset_hold(ddname, FTAG, &ds);
1198         if (err)
1199                 return (err);
1200
1201         err = dsl_dir_open(ddname, FTAG, &dd, NULL);
1202         if (err) {
1203                 dsl_dataset_rele(ds, FTAG);
1204                 return (err);
1205         }
1206
1207         ASSERT(ds->ds_dir == dd);
1208
1209         err = dsl_sync_task_do(dd->dd_pool, dsl_dir_set_reservation_check,
1210             dsl_dir_set_reservation_sync, ds, &psa, 0);
1211
1212         dsl_dir_close(dd, FTAG);
1213         dsl_dataset_rele(ds, FTAG);
1214         return (err);
1215 }
1216
1217 static dsl_dir_t *
1218 closest_common_ancestor(dsl_dir_t *ds1, dsl_dir_t *ds2)
1219 {
1220         for (; ds1; ds1 = ds1->dd_parent) {
1221                 dsl_dir_t *dd;
1222                 for (dd = ds2; dd; dd = dd->dd_parent) {
1223                         if (ds1 == dd)
1224                                 return (dd);
1225                 }
1226         }
1227         return (NULL);
1228 }
1229
1230 /*
1231  * If delta is applied to dd, how much of that delta would be applied to
1232  * ancestor?  Syncing context only.
1233  */
1234 static int64_t
1235 would_change(dsl_dir_t *dd, int64_t delta, dsl_dir_t *ancestor)
1236 {
1237         if (dd == ancestor)
1238                 return (delta);
1239
1240         mutex_enter(&dd->dd_lock);
1241         delta = parent_delta(dd, dd->dd_phys->dd_used_bytes, delta);
1242         mutex_exit(&dd->dd_lock);
1243         return (would_change(dd->dd_parent, delta, ancestor));
1244 }
1245
1246 struct renamearg {
1247         dsl_dir_t *newparent;
1248         const char *mynewname;
1249         boolean_t allowmounted;
1250 };
1251
1252 static int
1253 dsl_dir_rename_check(void *arg1, void *arg2, dmu_tx_t *tx)
1254 {
1255         dsl_dir_t *dd = arg1;
1256         struct renamearg *ra = arg2;
1257         dsl_pool_t *dp = dd->dd_pool;
1258         objset_t *mos = dp->dp_meta_objset;
1259         int err;
1260         uint64_t val;
1261
1262         /*
1263          * There should only be one reference, from dmu_objset_rename().
1264          * Fleeting holds are also possible (eg, from "zfs list" getting
1265          * stats), but any that are present in open context will likely
1266          * be gone by syncing context, so only fail from syncing
1267          * context.
1268          * Don't check if we allow renaming of busy (mounted) dataset.
1269          */
1270         if (!ra->allowmounted && dmu_tx_is_syncing(tx) &&
1271             dmu_buf_refcount(dd->dd_dbuf) > 1) {
1272                 return (EBUSY);
1273         }
1274
1275         /* check for existing name */
1276         err = zap_lookup(mos, ra->newparent->dd_phys->dd_child_dir_zapobj,
1277             ra->mynewname, 8, 1, &val);
1278         if (err == 0)
1279                 return (EEXIST);
1280         if (err != ENOENT)
1281                 return (err);
1282
1283         if (ra->newparent != dd->dd_parent) {
1284                 /* is there enough space? */
1285                 uint64_t myspace =
1286                     MAX(dd->dd_phys->dd_used_bytes, dd->dd_phys->dd_reserved);
1287
1288                 /* no rename into our descendant */
1289                 if (closest_common_ancestor(dd, ra->newparent) == dd)
1290                         return (EINVAL);
1291
1292                 if (err = dsl_dir_transfer_possible(dd->dd_parent,
1293                     ra->newparent, myspace))
1294                         return (err);
1295         }
1296
1297         return (0);
1298 }
1299
1300 static void
1301 dsl_dir_rename_sync(void *arg1, void *arg2, dmu_tx_t *tx)
1302 {
1303         char oldname[MAXPATHLEN], newname[MAXPATHLEN];
1304         dsl_dir_t *dd = arg1;
1305         struct renamearg *ra = arg2;
1306         dsl_pool_t *dp = dd->dd_pool;
1307         objset_t *mos = dp->dp_meta_objset;
1308         int err;
1309         char namebuf[MAXNAMELEN];
1310
1311         ASSERT(ra->allowmounted || dmu_buf_refcount(dd->dd_dbuf) <= 2);
1312
1313         /* Log this before we change the name. */
1314         dsl_dir_name(ra->newparent, namebuf);
1315         spa_history_log_internal_dd(dd, "rename", tx,
1316             "-> %s/%s", namebuf, ra->mynewname);
1317
1318         if (ra->newparent != dd->dd_parent) {
1319                 dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD,
1320                     -dd->dd_phys->dd_used_bytes,
1321                     -dd->dd_phys->dd_compressed_bytes,
1322                     -dd->dd_phys->dd_uncompressed_bytes, tx);
1323                 dsl_dir_diduse_space(ra->newparent, DD_USED_CHILD,
1324                     dd->dd_phys->dd_used_bytes,
1325                     dd->dd_phys->dd_compressed_bytes,
1326                     dd->dd_phys->dd_uncompressed_bytes, tx);
1327
1328                 if (dd->dd_phys->dd_reserved > dd->dd_phys->dd_used_bytes) {
1329                         uint64_t unused_rsrv = dd->dd_phys->dd_reserved -
1330                             dd->dd_phys->dd_used_bytes;
1331
1332                         dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD_RSRV,
1333                             -unused_rsrv, 0, 0, tx);
1334                         dsl_dir_diduse_space(ra->newparent, DD_USED_CHILD_RSRV,
1335                             unused_rsrv, 0, 0, tx);
1336                 }
1337         }
1338
1339         dmu_buf_will_dirty(dd->dd_dbuf, tx);
1340
1341         /* remove from old parent zapobj */
1342         dsl_dir_name(dd, oldname);
1343         err = zap_remove(mos, dd->dd_parent->dd_phys->dd_child_dir_zapobj,
1344             dd->dd_myname, tx);
1345         ASSERT0(err);
1346
1347         (void) strcpy(dd->dd_myname, ra->mynewname);
1348         dsl_dir_close(dd->dd_parent, dd);
1349         dd->dd_phys->dd_parent_obj = ra->newparent->dd_object;
1350         VERIFY(0 == dsl_dir_open_obj(dd->dd_pool,
1351             ra->newparent->dd_object, NULL, dd, &dd->dd_parent));
1352
1353         /* add to new parent zapobj */
1354         err = zap_add(mos, ra->newparent->dd_phys->dd_child_dir_zapobj,
1355             dd->dd_myname, 8, 1, &dd->dd_object, tx);
1356         ASSERT0(err);
1357         dsl_dir_name(dd, newname);
1358 #ifdef _KERNEL
1359         zfsvfs_update_fromname(oldname, newname);
1360         zvol_rename_minors(oldname, newname);
1361 #endif
1362
1363 }
1364
1365 int
1366 dsl_dir_rename(dsl_dir_t *dd, const char *newname, int flags)
1367 {
1368         struct renamearg ra;
1369         int err;
1370
1371         /* new parent should exist */
1372         err = dsl_dir_open(newname, FTAG, &ra.newparent, &ra.mynewname);
1373         if (err)
1374                 return (err);
1375
1376         /* can't rename to different pool */
1377         if (dd->dd_pool != ra.newparent->dd_pool) {
1378                 err = ENXIO;
1379                 goto out;
1380         }
1381
1382         /* new name should not already exist */
1383         if (ra.mynewname == NULL) {
1384                 err = EEXIST;
1385                 goto out;
1386         }
1387
1388         ra.allowmounted = !!(flags & ZFS_RENAME_ALLOW_MOUNTED);
1389
1390         err = dsl_sync_task_do(dd->dd_pool,
1391             dsl_dir_rename_check, dsl_dir_rename_sync, dd, &ra, 3);
1392
1393 out:
1394         dsl_dir_close(ra.newparent, FTAG);
1395         return (err);
1396 }
1397
1398 int
1399 dsl_dir_transfer_possible(dsl_dir_t *sdd, dsl_dir_t *tdd, uint64_t space)
1400 {
1401         dsl_dir_t *ancestor;
1402         int64_t adelta;
1403         uint64_t avail;
1404
1405         ancestor = closest_common_ancestor(sdd, tdd);
1406         adelta = would_change(sdd, -space, ancestor);
1407         avail = dsl_dir_space_available(tdd, ancestor, adelta, FALSE);
1408         if (avail < space)
1409                 return (ENOSPC);
1410
1411         return (0);
1412 }
1413
1414 timestruc_t
1415 dsl_dir_snap_cmtime(dsl_dir_t *dd)
1416 {
1417         timestruc_t t;
1418
1419         mutex_enter(&dd->dd_lock);
1420         t = dd->dd_snap_cmtime;
1421         mutex_exit(&dd->dd_lock);
1422
1423         return (t);
1424 }
1425
1426 void
1427 dsl_dir_snap_cmtime_update(dsl_dir_t *dd)
1428 {
1429         timestruc_t t;
1430
1431         gethrestime(&t);
1432         mutex_enter(&dd->dd_lock);
1433         dd->dd_snap_cmtime = t;
1434         mutex_exit(&dd->dd_lock);
1435 }