module/zfs/dsl_dir.c

   1 /*
   2  * CDDL HEADER START
   3  *
   4  * The contents of this file are subject to the terms of the
   5  * Common Development and Distribution License (the "License").
   6  * You may not use this file except in compliance with the License.
   7  *
   8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
   9  * or http://www.opensolaris.org/os/licensing.
  10  * See the License for the specific language governing permissions
  11  * and limitations under the License.
  12  *
  13  * When distributing Covered Code, include this CDDL HEADER in each
  14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
  15  * If applicable, add the following below this CDDL HEADER, with the
  16  * fields enclosed by brackets "[]" replaced with your own identifying
  17  * information: Portions Copyright [yyyy] [name of copyright owner]
  18  *
  19  * CDDL HEADER END
  20  */
  21 /*
  22  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
  23  * Copyright (c) 2013 by Delphix. All rights reserved.
  24  * Copyright (c) 2013 Martin Matuska. 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/dmu_impl.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 #include "zfs_namecheck.h"
  44
  45 static uint64_t dsl_dir_space_towrite(dsl_dir_t *dd);
  46
  47 /* ARGSUSED */
  48 static void
  49 dsl_dir_evict(dmu_buf_t *db, void *arg)
  50 {
  51         dsl_dir_t *dd = arg;
  52         int t;
  53         ASSERTV(dsl_pool_t *dp = dd->dd_pool);
  54
  55         for (t = 0; t < TXG_SIZE; t++) {
  56                 ASSERT(!txg_list_member(&dp->dp_dirty_dirs, dd, t));
  57                 ASSERT(dd->dd_tempreserved[t] == 0);
  58                 ASSERT(dd->dd_space_towrite[t] == 0);
  59         }
  60
  61         if (dd->dd_parent)
  62                 dsl_dir_rele(dd->dd_parent, dd);
  63
  64         spa_close(dd->dd_pool->dp_spa, dd);
  65
  66         /*
  67          * The props callback list should have been cleaned up by
  68          * objset_evict().
  69          */
  70         list_destroy(&dd->dd_prop_cbs);
  71         mutex_destroy(&dd->dd_lock);
  72         kmem_free(dd, sizeof (dsl_dir_t));
  73 }
  74
  75 int
  76 dsl_dir_hold_obj(dsl_pool_t *dp, uint64_t ddobj,
  77     const char *tail, void *tag, dsl_dir_t **ddp)
  78 {
  79         dmu_buf_t *dbuf;
  80         dsl_dir_t *dd;
  81         int err;
  82
  83         ASSERT(dsl_pool_config_held(dp));
  84
  85         err = dmu_bonus_hold(dp->dp_meta_objset, ddobj, tag, &dbuf);
  86         if (err != 0)
  87                 return (err);
  88         dd = dmu_buf_get_user(dbuf);
  89 #ifdef ZFS_DEBUG
  90         {
  91                 dmu_object_info_t doi;
  92                 dmu_object_info_from_db(dbuf, &doi);
  93                 ASSERT3U(doi.doi_bonus_type, ==, DMU_OT_DSL_DIR);
  94                 ASSERT3U(doi.doi_bonus_size, >=, sizeof (dsl_dir_phys_t));
  95         }
  96 #endif
  97         if (dd == NULL) {
  98                 dsl_dir_t *winner;
  99
 100                 dd = kmem_zalloc(sizeof (dsl_dir_t), KM_SLEEP);
 101                 dd->dd_object = ddobj;
 102                 dd->dd_dbuf = dbuf;
 103                 dd->dd_pool = dp;
 104                 dd->dd_phys = dbuf->db_data;
 105                 mutex_init(&dd->dd_lock, NULL, MUTEX_DEFAULT, NULL);
 106
 107                 list_create(&dd->dd_prop_cbs, sizeof (dsl_prop_cb_record_t),
 108                     offsetof(dsl_prop_cb_record_t, cbr_node));
 109
 110                 dsl_dir_snap_cmtime_update(dd);
 111
 112                 if (dd->dd_phys->dd_parent_obj) {
 113                         err = dsl_dir_hold_obj(dp, dd->dd_phys->dd_parent_obj,
 114                             NULL, dd, &dd->dd_parent);
 115                         if (err != 0)
 116                                 goto errout;
 117                         if (tail) {
 118 #ifdef ZFS_DEBUG
 119                                 uint64_t foundobj;
 120
 121                                 err = zap_lookup(dp->dp_meta_objset,
 122                                     dd->dd_parent->dd_phys->dd_child_dir_zapobj,
 123                                     tail, sizeof (foundobj), 1, &foundobj);
 124                                 ASSERT(err || foundobj == ddobj);
 125 #endif
 126                                 (void) strcpy(dd->dd_myname, tail);
 127                         } else {
 128                                 err = zap_value_search(dp->dp_meta_objset,
 129                                     dd->dd_parent->dd_phys->dd_child_dir_zapobj,
 130                                     ddobj, 0, dd->dd_myname);
 131                         }
 132                         if (err != 0)
 133                                 goto errout;
 134                 } else {
 135                         (void) strcpy(dd->dd_myname, spa_name(dp->dp_spa));
 136                 }
 137
 138                 if (dsl_dir_is_clone(dd)) {
 139                         dmu_buf_t *origin_bonus;
 140                         dsl_dataset_phys_t *origin_phys;
 141
 142                         /*
 143                          * We can't open the origin dataset, because
 144                          * that would require opening this dsl_dir.
 145                          * Just look at its phys directly instead.
 146                          */
 147                         err = dmu_bonus_hold(dp->dp_meta_objset,
 148                             dd->dd_phys->dd_origin_obj, FTAG, &origin_bonus);
 149                         if (err != 0)
 150                                 goto errout;
 151                         origin_phys = origin_bonus->db_data;
 152                         dd->dd_origin_txg =
 153                             origin_phys->ds_creation_txg;
 154                         dmu_buf_rele(origin_bonus, FTAG);
 155                 }
 156
 157                 winner = dmu_buf_set_user_ie(dbuf, dd, &dd->dd_phys,
 158                     dsl_dir_evict);
 159                 if (winner) {
 160                         if (dd->dd_parent)
 161                                 dsl_dir_rele(dd->dd_parent, dd);
 162                         mutex_destroy(&dd->dd_lock);
 163                         kmem_free(dd, sizeof (dsl_dir_t));
 164                         dd = winner;
 165                 } else {
 166                         spa_open_ref(dp->dp_spa, dd);
 167                 }
 168         }
 169
 170         /*
 171          * The dsl_dir_t has both open-to-close and instantiate-to-evict
 172          * holds on the spa.  We need the open-to-close holds because
 173          * otherwise the spa_refcnt wouldn't change when we open a
 174          * dir which the spa also has open, so we could incorrectly
 175          * think it was OK to unload/export/destroy the pool.  We need
 176          * the instantiate-to-evict hold because the dsl_dir_t has a
 177          * pointer to the dd_pool, which has a pointer to the spa_t.
 178          */
 179         spa_open_ref(dp->dp_spa, tag);
 180         ASSERT3P(dd->dd_pool, ==, dp);
 181         ASSERT3U(dd->dd_object, ==, ddobj);
 182         ASSERT3P(dd->dd_dbuf, ==, dbuf);
 183         *ddp = dd;
 184         return (0);
 185
 186 errout:
 187         if (dd->dd_parent)
 188                 dsl_dir_rele(dd->dd_parent, dd);
 189         mutex_destroy(&dd->dd_lock);
 190         kmem_free(dd, sizeof (dsl_dir_t));
 191         dmu_buf_rele(dbuf, tag);
 192         return (err);
 193 }
 194
 195 void
 196 dsl_dir_rele(dsl_dir_t *dd, void *tag)
 197 {
 198         dprintf_dd(dd, "%s\n", "");
 199         spa_close(dd->dd_pool->dp_spa, tag);
 200         dmu_buf_rele(dd->dd_dbuf, tag);
 201 }
 202
 203 /* buf must be long enough (MAXNAMELEN + strlen(MOS_DIR_NAME) + 1 should do) */
 204 void
 205 dsl_dir_name(dsl_dir_t *dd, char *buf)
 206 {
 207         if (dd->dd_parent) {
 208                 dsl_dir_name(dd->dd_parent, buf);
 209                 (void) strcat(buf, "/");
 210         } else {
 211                 buf[0] = '\0';
 212         }
 213         if (!MUTEX_HELD(&dd->dd_lock)) {
 214                 /*
 215                  * recursive mutex so that we can use
 216                  * dprintf_dd() with dd_lock held
 217                  */
 218                 mutex_enter(&dd->dd_lock);
 219                 (void) strcat(buf, dd->dd_myname);
 220                 mutex_exit(&dd->dd_lock);
 221         } else {
 222                 (void) strcat(buf, dd->dd_myname);
 223         }
 224 }
 225
 226 /* Calculate name length, avoiding all the strcat calls of dsl_dir_name */
 227 int
 228 dsl_dir_namelen(dsl_dir_t *dd)
 229 {
 230         int result = 0;
 231
 232         if (dd->dd_parent) {
 233                 /* parent's name + 1 for the "/" */
 234                 result = dsl_dir_namelen(dd->dd_parent) + 1;
 235         }
 236
 237         if (!MUTEX_HELD(&dd->dd_lock)) {
 238                 /* see dsl_dir_name */
 239                 mutex_enter(&dd->dd_lock);
 240                 result += strlen(dd->dd_myname);
 241                 mutex_exit(&dd->dd_lock);
 242         } else {
 243                 result += strlen(dd->dd_myname);
 244         }
 245
 246         return (result);
 247 }
 248
 249 static int
 250 getcomponent(const char *path, char *component, const char **nextp)
 251 {
 252         char *p;
 253
 254         if ((path == NULL) || (path[0] == '\0'))
 255                 return (SET_ERROR(ENOENT));
 256         /* This would be a good place to reserve some namespace... */
 257         p = strpbrk(path, "/@");
 258         if (p && (p[1] == '/' || p[1] == '@')) {
 259                 /* two separators in a row */
 260                 return (SET_ERROR(EINVAL));
 261         }
 262         if (p == NULL || p == path) {
 263                 /*
 264                  * if the first thing is an @ or /, it had better be an
 265                  * @ and it had better not have any more ats or slashes,
 266                  * and it had better have something after the @.
 267                  */
 268                 if (p != NULL &&
 269                     (p[0] != '@' || strpbrk(path+1, "/@") || p[1] == '\0'))
 270                         return (SET_ERROR(EINVAL));
 271                 if (strlen(path) >= MAXNAMELEN)
 272                         return (SET_ERROR(ENAMETOOLONG));
 273                 (void) strcpy(component, path);
 274                 p = NULL;
 275         } else if (p[0] == '/') {
 276                 if (p - path >= MAXNAMELEN)
 277                         return (SET_ERROR(ENAMETOOLONG));
 278                 (void) strncpy(component, path, p - path);
 279                 component[p - path] = '\0';
 280                 p++;
 281         } else if (p[0] == '@') {
 282                 /*
 283                  * if the next separator is an @, there better not be
 284                  * any more slashes.
 285                  */
 286                 if (strchr(path, '/'))
 287                         return (SET_ERROR(EINVAL));
 288                 if (p - path >= MAXNAMELEN)
 289                         return (SET_ERROR(ENAMETOOLONG));
 290                 (void) strncpy(component, path, p - path);
 291                 component[p - path] = '\0';
 292         } else {
 293                 panic("invalid p=%p", (void *)p);
 294         }
 295         *nextp = p;
 296         return (0);
 297 }
 298
 299 /*
 300  * Return the dsl_dir_t, and possibly the last component which couldn't
 301  * be found in *tail.  The name must be in the specified dsl_pool_t.  This
 302  * thread must hold the dp_config_rwlock for the pool.  Returns NULL if the
 303  * path is bogus, or if tail==NULL and we couldn't parse the whole name.
 304  * (*tail)[0] == '@' means that the last component is a snapshot.
 305  */
 306 int
 307 dsl_dir_hold(dsl_pool_t *dp, const char *name, void *tag,
 308     dsl_dir_t **ddp, const char **tailp)
 309 {
 310         char *buf;
 311         const char *spaname, *next, *nextnext = NULL;
 312         int err;
 313         dsl_dir_t *dd;
 314         uint64_t ddobj;
 315
 316         buf = kmem_alloc(MAXNAMELEN, KM_SLEEP);
 317         err = getcomponent(name, buf, &next);
 318         if (err != 0)
 319                 goto error;
 320
 321         /* Make sure the name is in the specified pool. */
 322         spaname = spa_name(dp->dp_spa);
 323         if (strcmp(buf, spaname) != 0) {
 324                 err = SET_ERROR(EXDEV);
 325                 goto error;
 326         }
 327
 328         ASSERT(dsl_pool_config_held(dp));
 329
 330         err = dsl_dir_hold_obj(dp, dp->dp_root_dir_obj, NULL, tag, &dd);
 331         if (err != 0) {
 332                 goto error;
 333         }
 334
 335         while (next != NULL) {
 336                 dsl_dir_t *child_ds;
 337                 err = getcomponent(next, buf, &nextnext);
 338                 if (err != 0)
 339                         break;
 340                 ASSERT(next[0] != '\0');
 341                 if (next[0] == '@')
 342                         break;
 343                 dprintf("looking up %s in obj%lld\n",
 344                     buf, dd->dd_phys->dd_child_dir_zapobj);
 345
 346                 err = zap_lookup(dp->dp_meta_objset,
 347                     dd->dd_phys->dd_child_dir_zapobj,
 348                     buf, sizeof (ddobj), 1, &ddobj);
 349                 if (err != 0) {
 350                         if (err == ENOENT)
 351                                 err = 0;
 352                         break;
 353                 }
 354
 355                 err = dsl_dir_hold_obj(dp, ddobj, buf, tag, &child_ds);
 356                 if (err != 0)
 357                         break;
 358                 dsl_dir_rele(dd, tag);
 359                 dd = child_ds;
 360                 next = nextnext;
 361         }
 362
 363         if (err != 0) {
 364                 dsl_dir_rele(dd, tag);
 365                 goto error;
 366         }
 367
 368         /*
 369          * It's an error if there's more than one component left, or
 370          * tailp==NULL and there's any component left.
 371          */
 372         if (next != NULL &&
 373             (tailp == NULL || (nextnext && nextnext[0] != '\0'))) {
 374                 /* bad path name */
 375                 dsl_dir_rele(dd, tag);
 376                 dprintf("next=%p (%s) tail=%p\n", next, next?next:"", tailp);
 377                 err = SET_ERROR(ENOENT);
 378         }
 379         if (tailp != NULL)
 380                 *tailp = next;
 381         *ddp = dd;
 382 error:
 383         kmem_free(buf, MAXNAMELEN);
 384         return (err);
 385 }
 386
 387 uint64_t
 388 dsl_dir_create_sync(dsl_pool_t *dp, dsl_dir_t *pds, const char *name,
 389     dmu_tx_t *tx)
 390 {
 391         objset_t *mos = dp->dp_meta_objset;
 392         uint64_t ddobj;
 393         dsl_dir_phys_t *ddphys;
 394         dmu_buf_t *dbuf;
 395
 396         ddobj = dmu_object_alloc(mos, DMU_OT_DSL_DIR, 0,
 397             DMU_OT_DSL_DIR, sizeof (dsl_dir_phys_t), tx);
 398         if (pds) {
 399                 VERIFY(0 == zap_add(mos, pds->dd_phys->dd_child_dir_zapobj,
 400                     name, sizeof (uint64_t), 1, &ddobj, tx));
 401         } else {
 402                 /* it's the root dir */
 403                 VERIFY(0 == zap_add(mos, DMU_POOL_DIRECTORY_OBJECT,
 404                     DMU_POOL_ROOT_DATASET, sizeof (uint64_t), 1, &ddobj, tx));
 405         }
 406         VERIFY(0 == dmu_bonus_hold(mos, ddobj, FTAG, &dbuf));
 407         dmu_buf_will_dirty(dbuf, tx);
 408         ddphys = dbuf->db_data;
 409
 410         ddphys->dd_creation_time = gethrestime_sec();
 411         if (pds)
 412                 ddphys->dd_parent_obj = pds->dd_object;
 413         ddphys->dd_props_zapobj = zap_create(mos,
 414             DMU_OT_DSL_PROPS, DMU_OT_NONE, 0, tx);
 415         ddphys->dd_child_dir_zapobj = zap_create(mos,
 416             DMU_OT_DSL_DIR_CHILD_MAP, DMU_OT_NONE, 0, tx);
 417         if (spa_version(dp->dp_spa) >= SPA_VERSION_USED_BREAKDOWN)
 418                 ddphys->dd_flags |= DD_FLAG_USED_BREAKDOWN;
 419         dmu_buf_rele(dbuf, FTAG);
 420
 421         return (ddobj);
 422 }
 423
 424 boolean_t
 425 dsl_dir_is_clone(dsl_dir_t *dd)
 426 {
 427         return (dd->dd_phys->dd_origin_obj &&
 428             (dd->dd_pool->dp_origin_snap == NULL ||
 429             dd->dd_phys->dd_origin_obj !=
 430             dd->dd_pool->dp_origin_snap->ds_object));
 431 }
 432
 433 void
 434 dsl_dir_stats(dsl_dir_t *dd, nvlist_t *nv)
 435 {
 436         mutex_enter(&dd->dd_lock);
 437         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USED,
 438             dd->dd_phys->dd_used_bytes);
 439         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_QUOTA, dd->dd_phys->dd_quota);
 440         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_RESERVATION,
 441             dd->dd_phys->dd_reserved);
 442         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_COMPRESSRATIO,
 443             dd->dd_phys->dd_compressed_bytes == 0 ? 100 :
 444             (dd->dd_phys->dd_uncompressed_bytes * 100 /
 445             dd->dd_phys->dd_compressed_bytes));
 446         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_LOGICALUSED,
 447             dd->dd_phys->dd_uncompressed_bytes);
 448         if (dd->dd_phys->dd_flags & DD_FLAG_USED_BREAKDOWN) {
 449                 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDSNAP,
 450                     dd->dd_phys->dd_used_breakdown[DD_USED_SNAP]);
 451                 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDDS,
 452                     dd->dd_phys->dd_used_breakdown[DD_USED_HEAD]);
 453                 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDREFRESERV,
 454                     dd->dd_phys->dd_used_breakdown[DD_USED_REFRSRV]);
 455                 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDCHILD,
 456                     dd->dd_phys->dd_used_breakdown[DD_USED_CHILD] +
 457                     dd->dd_phys->dd_used_breakdown[DD_USED_CHILD_RSRV]);
 458         }
 459         mutex_exit(&dd->dd_lock);
 460
 461         if (dsl_dir_is_clone(dd)) {
 462                 dsl_dataset_t *ds;
 463                 char buf[MAXNAMELEN];
 464
 465                 VERIFY0(dsl_dataset_hold_obj(dd->dd_pool,
 466                     dd->dd_phys->dd_origin_obj, FTAG, &ds));
 467                 dsl_dataset_name(ds, buf);
 468                 dsl_dataset_rele(ds, FTAG);
 469                 dsl_prop_nvlist_add_string(nv, ZFS_PROP_ORIGIN, buf);
 470         }
 471 }
 472
 473 void
 474 dsl_dir_dirty(dsl_dir_t *dd, dmu_tx_t *tx)
 475 {
 476         dsl_pool_t *dp = dd->dd_pool;
 477
 478         ASSERT(dd->dd_phys);
 479
 480         if (txg_list_add(&dp->dp_dirty_dirs, dd, tx->tx_txg)) {
 481                 /* up the hold count until we can be written out */
 482                 dmu_buf_add_ref(dd->dd_dbuf, dd);
 483         }
 484 }
 485
 486 static int64_t
 487 parent_delta(dsl_dir_t *dd, uint64_t used, int64_t delta)
 488 {
 489         uint64_t old_accounted = MAX(used, dd->dd_phys->dd_reserved);
 490         uint64_t new_accounted = MAX(used + delta, dd->dd_phys->dd_reserved);
 491         return (new_accounted - old_accounted);
 492 }
 493
 494 void
 495 dsl_dir_sync(dsl_dir_t *dd, dmu_tx_t *tx)
 496 {
 497         ASSERT(dmu_tx_is_syncing(tx));
 498
 499         mutex_enter(&dd->dd_lock);
 500         ASSERT0(dd->dd_tempreserved[tx->tx_txg&TXG_MASK]);
 501         dprintf_dd(dd, "txg=%llu towrite=%lluK\n", tx->tx_txg,
 502             dd->dd_space_towrite[tx->tx_txg&TXG_MASK] / 1024);
 503         dd->dd_space_towrite[tx->tx_txg&TXG_MASK] = 0;
 504         mutex_exit(&dd->dd_lock);
 505
 506         /* release the hold from dsl_dir_dirty */
 507         dmu_buf_rele(dd->dd_dbuf, dd);
 508 }
 509
 510 static uint64_t
 511 dsl_dir_space_towrite(dsl_dir_t *dd)
 512 {
 513         uint64_t space = 0;
 514         int i;
 515
 516         ASSERT(MUTEX_HELD(&dd->dd_lock));
 517
 518         for (i = 0; i < TXG_SIZE; i++) {
 519                 space += dd->dd_space_towrite[i&TXG_MASK];
 520                 ASSERT3U(dd->dd_space_towrite[i&TXG_MASK], >=, 0);
 521         }
 522         return (space);
 523 }
 524
 525 /*
 526  * How much space would dd have available if ancestor had delta applied
 527  * to it?  If ondiskonly is set, we're only interested in what's
 528  * on-disk, not estimated pending changes.
 529  */
 530 uint64_t
 531 dsl_dir_space_available(dsl_dir_t *dd,
 532     dsl_dir_t *ancestor, int64_t delta, int ondiskonly)
 533 {
 534         uint64_t parentspace, myspace, quota, used;
 535
 536         /*
 537          * If there are no restrictions otherwise, assume we have
 538          * unlimited space available.
 539          */
 540         quota = UINT64_MAX;
 541         parentspace = UINT64_MAX;
 542
 543         if (dd->dd_parent != NULL) {
 544                 parentspace = dsl_dir_space_available(dd->dd_parent,
 545                     ancestor, delta, ondiskonly);
 546         }
 547
 548         mutex_enter(&dd->dd_lock);
 549         if (dd->dd_phys->dd_quota != 0)
 550                 quota = dd->dd_phys->dd_quota;
 551         used = dd->dd_phys->dd_used_bytes;
 552         if (!ondiskonly)
 553                 used += dsl_dir_space_towrite(dd);
 554
 555         if (dd->dd_parent == NULL) {
 556                 uint64_t poolsize = dsl_pool_adjustedsize(dd->dd_pool, FALSE);
 557                 quota = MIN(quota, poolsize);
 558         }
 559
 560         if (dd->dd_phys->dd_reserved > used && parentspace != UINT64_MAX) {
 561                 /*
 562                  * We have some space reserved, in addition to what our
 563                  * parent gave us.
 564                  */
 565                 parentspace += dd->dd_phys->dd_reserved - used;
 566         }
 567
 568         if (dd == ancestor) {
 569                 ASSERT(delta <= 0);
 570                 ASSERT(used >= -delta);
 571                 used += delta;
 572                 if (parentspace != UINT64_MAX)
 573                         parentspace -= delta;
 574         }
 575
 576         if (used > quota) {
 577                 /* over quota */
 578                 myspace = 0;
 579         } else {
 580                 /*
 581                  * the lesser of the space provided by our parent and
 582                  * the space left in our quota
 583                  */
 584                 myspace = MIN(parentspace, quota - used);
 585         }
 586
 587         mutex_exit(&dd->dd_lock);
 588
 589         return (myspace);
 590 }
 591
 592 struct tempreserve {
 593         list_node_t tr_node;
 594         dsl_dir_t *tr_ds;
 595         uint64_t tr_size;
 596 };
 597
 598 static int
 599 dsl_dir_tempreserve_impl(dsl_dir_t *dd, uint64_t asize, boolean_t netfree,
 600     boolean_t ignorequota, boolean_t checkrefquota, list_t *tr_list,
 601     dmu_tx_t *tx, boolean_t first)
 602 {
 603         uint64_t txg = tx->tx_txg;
 604         uint64_t est_inflight, used_on_disk, quota, parent_rsrv;
 605         uint64_t deferred = 0;
 606         struct tempreserve *tr;
 607         int retval = EDQUOT;
 608         int txgidx = txg & TXG_MASK;
 609         int i;
 610         uint64_t ref_rsrv = 0;
 611
 612         ASSERT3U(txg, !=, 0);
 613         ASSERT3S(asize, >, 0);
 614
 615         mutex_enter(&dd->dd_lock);
 616
 617         /*
 618          * Check against the dsl_dir's quota.  We don't add in the delta
 619          * when checking for over-quota because they get one free hit.
 620          */
 621         est_inflight = dsl_dir_space_towrite(dd);
 622         for (i = 0; i < TXG_SIZE; i++)
 623                 est_inflight += dd->dd_tempreserved[i];
 624         used_on_disk = dd->dd_phys->dd_used_bytes;
 625
 626         /*
 627          * On the first iteration, fetch the dataset's used-on-disk and
 628          * refreservation values. Also, if checkrefquota is set, test if
 629          * allocating this space would exceed the dataset's refquota.
 630          */
 631         if (first && tx->tx_objset) {
 632                 int error;
 633                 dsl_dataset_t *ds = tx->tx_objset->os_dsl_dataset;
 634
 635                 error = dsl_dataset_check_quota(ds, checkrefquota,
 636                     asize, est_inflight, &used_on_disk, &ref_rsrv);
 637                 if (error) {
 638                         mutex_exit(&dd->dd_lock);
 639                         DMU_TX_STAT_BUMP(dmu_tx_quota);
 640                         return (error);
 641                 }
 642         }
 643
 644         /*
 645          * If this transaction will result in a net free of space,
 646          * we want to let it through.
 647          */
 648         if (ignorequota || netfree || dd->dd_phys->dd_quota == 0)
 649                 quota = UINT64_MAX;
 650         else
 651                 quota = dd->dd_phys->dd_quota;
 652
 653         /*
 654          * Adjust the quota against the actual pool size at the root
 655          * minus any outstanding deferred frees.
 656          * To ensure that it's possible to remove files from a full
 657          * pool without inducing transient overcommits, we throttle
 658          * netfree transactions against a quota that is slightly larger,
 659          * but still within the pool's allocation slop.  In cases where
 660          * we're very close to full, this will allow a steady trickle of
 661          * removes to get through.
 662          */
 663         if (dd->dd_parent == NULL) {
 664                 spa_t *spa = dd->dd_pool->dp_spa;
 665                 uint64_t poolsize = dsl_pool_adjustedsize(dd->dd_pool, netfree);
 666                 deferred = metaslab_class_get_deferred(spa_normal_class(spa));
 667                 if (poolsize - deferred < quota) {
 668                         quota = poolsize - deferred;
 669                         retval = ENOSPC;
 670                 }
 671         }
 672
 673         /*
 674          * If they are requesting more space, and our current estimate
 675          * is over quota, they get to try again unless the actual
 676          * on-disk is over quota and there are no pending changes (which
 677          * may free up space for us).
 678          */
 679         if (used_on_disk + est_inflight >= quota) {
 680                 if (est_inflight > 0 || used_on_disk < quota ||
 681                     (retval == ENOSPC && used_on_disk < quota + deferred))
 682                         retval = ERESTART;
 683                 dprintf_dd(dd, "failing: used=%lluK inflight = %lluK "
 684                     "quota=%lluK tr=%lluK err=%d\n",
 685                     used_on_disk>>10, est_inflight>>10,
 686                     quota>>10, asize>>10, retval);
 687                 mutex_exit(&dd->dd_lock);
 688                 DMU_TX_STAT_BUMP(dmu_tx_quota);
 689                 return (SET_ERROR(retval));
 690         }
 691
 692         /* We need to up our estimated delta before dropping dd_lock */
 693         dd->dd_tempreserved[txgidx] += asize;
 694
 695         parent_rsrv = parent_delta(dd, used_on_disk + est_inflight,
 696             asize - ref_rsrv);
 697         mutex_exit(&dd->dd_lock);
 698
 699         tr = kmem_zalloc(sizeof (struct tempreserve), KM_SLEEP);
 700         tr->tr_ds = dd;
 701         tr->tr_size = asize;
 702         list_insert_tail(tr_list, tr);
 703
 704         /* see if it's OK with our parent */
 705         if (dd->dd_parent && parent_rsrv) {
 706                 boolean_t ismos = (dd->dd_phys->dd_head_dataset_obj == 0);
 707
 708                 return (dsl_dir_tempreserve_impl(dd->dd_parent,
 709                     parent_rsrv, netfree, ismos, TRUE, tr_list, tx, FALSE));
 710         } else {
 711                 return (0);
 712         }
 713 }
 714
 715 /*
 716  * Reserve space in this dsl_dir, to be used in this tx's txg.
 717  * After the space has been dirtied (and dsl_dir_willuse_space()
 718  * has been called), the reservation should be canceled, using
 719  * dsl_dir_tempreserve_clear().
 720  */
 721 int
 722 dsl_dir_tempreserve_space(dsl_dir_t *dd, uint64_t lsize, uint64_t asize,
 723     uint64_t fsize, uint64_t usize, void **tr_cookiep, dmu_tx_t *tx)
 724 {
 725         int err;
 726         list_t *tr_list;
 727
 728         if (asize == 0) {
 729                 *tr_cookiep = NULL;
 730                 return (0);
 731         }
 732
 733         tr_list = kmem_alloc(sizeof (list_t), KM_SLEEP);
 734         list_create(tr_list, sizeof (struct tempreserve),
 735             offsetof(struct tempreserve, tr_node));
 736         ASSERT3S(asize, >, 0);
 737         ASSERT3S(fsize, >=, 0);
 738
 739         err = arc_tempreserve_space(lsize, tx->tx_txg);
 740         if (err == 0) {
 741                 struct tempreserve *tr;
 742
 743                 tr = kmem_zalloc(sizeof (struct tempreserve), KM_SLEEP);
 744                 tr->tr_size = lsize;
 745                 list_insert_tail(tr_list, tr);
 746         } else {
 747                 if (err == EAGAIN) {
 748                         /*
 749                          * If arc_memory_throttle() detected that pageout
 750                          * is running and we are low on memory, we delay new
 751                          * non-pageout transactions to give pageout an
 752                          * advantage.
 753                          *
 754                          * It is unfortunate to be delaying while the caller's
 755                          * locks are held.
 756                          */
 757                         txg_delay(dd->dd_pool, tx->tx_txg,
 758                             MSEC2NSEC(10), MSEC2NSEC(10));
 759                         err = SET_ERROR(ERESTART);
 760                 }
 761         }
 762
 763         if (err == 0) {
 764                 err = dsl_dir_tempreserve_impl(dd, asize, fsize >= asize,
 765                     FALSE, asize > usize, tr_list, tx, TRUE);
 766         }
 767
 768         if (err != 0)
 769                 dsl_dir_tempreserve_clear(tr_list, tx);
 770         else
 771                 *tr_cookiep = tr_list;
 772
 773         return (err);
 774 }
 775
 776 /*
 777  * Clear a temporary reservation that we previously made with
 778  * dsl_dir_tempreserve_space().
 779  */
 780 void
 781 dsl_dir_tempreserve_clear(void *tr_cookie, dmu_tx_t *tx)
 782 {
 783         int txgidx = tx->tx_txg & TXG_MASK;
 784         list_t *tr_list = tr_cookie;
 785         struct tempreserve *tr;
 786
 787         ASSERT3U(tx->tx_txg, !=, 0);
 788
 789         if (tr_cookie == NULL)
 790                 return;
 791
 792         while ((tr = list_head(tr_list)) != NULL) {
 793                 if (tr->tr_ds) {
 794                         mutex_enter(&tr->tr_ds->dd_lock);
 795                         ASSERT3U(tr->tr_ds->dd_tempreserved[txgidx], >=,
 796                             tr->tr_size);
 797                         tr->tr_ds->dd_tempreserved[txgidx] -= tr->tr_size;
 798                         mutex_exit(&tr->tr_ds->dd_lock);
 799                 } else {
 800                         arc_tempreserve_clear(tr->tr_size);
 801                 }
 802                 list_remove(tr_list, tr);
 803                 kmem_free(tr, sizeof (struct tempreserve));
 804         }
 805
 806         kmem_free(tr_list, sizeof (list_t));
 807 }
 808
 809 /*
 810  * This should be called from open context when we think we're going to write
 811  * or free space, for example when dirtying data. Be conservative; it's okay
 812  * to write less space or free more, but we don't want to write more or free
 813  * less than the amount specified.
 814  *
 815  * NOTE: The behavior of this function is identical to the Illumos / FreeBSD
 816  * version however it has been adjusted to use an iterative rather then
 817  * recursive algorithm to minimize stack usage.
 818  */
 819 void
 820 dsl_dir_willuse_space(dsl_dir_t *dd, int64_t space, dmu_tx_t *tx)
 821 {
 822         int64_t parent_space;
 823         uint64_t est_used;
 824
 825         do {
 826                 mutex_enter(&dd->dd_lock);
 827                 if (space > 0)
 828                         dd->dd_space_towrite[tx->tx_txg & TXG_MASK] += space;
 829
 830                 est_used = dsl_dir_space_towrite(dd) +
 831                     dd->dd_phys->dd_used_bytes;
 832                 parent_space = parent_delta(dd, est_used, space);
 833                 mutex_exit(&dd->dd_lock);
 834
 835                 /* Make sure that we clean up dd_space_to* */
 836                 dsl_dir_dirty(dd, tx);
 837
 838                 dd = dd->dd_parent;
 839                 space = parent_space;
 840         } while (space && dd);
 841 }
 842
 843 /* call from syncing context when we actually write/free space for this dd */
 844 void
 845 dsl_dir_diduse_space(dsl_dir_t *dd, dd_used_t type,
 846     int64_t used, int64_t compressed, int64_t uncompressed, dmu_tx_t *tx)
 847 {
 848         int64_t accounted_delta;
 849
 850         /*
 851          * dsl_dataset_set_refreservation_sync_impl() calls this with
 852          * dd_lock held, so that it can atomically update
 853          * ds->ds_reserved and the dsl_dir accounting, so that
 854          * dsl_dataset_check_quota() can see dataset and dir accounting
 855          * consistently.
 856          */
 857         boolean_t needlock = !MUTEX_HELD(&dd->dd_lock);
 858
 859         ASSERT(dmu_tx_is_syncing(tx));
 860         ASSERT(type < DD_USED_NUM);
 861
 862         dmu_buf_will_dirty(dd->dd_dbuf, tx);
 863
 864         if (needlock)
 865                 mutex_enter(&dd->dd_lock);
 866         accounted_delta = parent_delta(dd, dd->dd_phys->dd_used_bytes, used);
 867         ASSERT(used >= 0 || dd->dd_phys->dd_used_bytes >= -used);
 868         ASSERT(compressed >= 0 ||
 869             dd->dd_phys->dd_compressed_bytes >= -compressed);
 870         ASSERT(uncompressed >= 0 ||
 871             dd->dd_phys->dd_uncompressed_bytes >= -uncompressed);
 872         dd->dd_phys->dd_used_bytes += used;
 873         dd->dd_phys->dd_uncompressed_bytes += uncompressed;
 874         dd->dd_phys->dd_compressed_bytes += compressed;
 875
 876         if (dd->dd_phys->dd_flags & DD_FLAG_USED_BREAKDOWN) {
 877                 ASSERT(used > 0 ||
 878                     dd->dd_phys->dd_used_breakdown[type] >= -used);
 879                 dd->dd_phys->dd_used_breakdown[type] += used;
 880 #ifdef DEBUG
 881                 {
 882                         dd_used_t t;
 883                         uint64_t u = 0;
 884                         for (t = 0; t < DD_USED_NUM; t++)
 885                                 u += dd->dd_phys->dd_used_breakdown[t];
 886                         ASSERT3U(u, ==, dd->dd_phys->dd_used_bytes);
 887                 }
 888 #endif
 889         }
 890         if (needlock)
 891                 mutex_exit(&dd->dd_lock);
 892
 893         if (dd->dd_parent != NULL) {
 894                 dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD,
 895                     accounted_delta, compressed, uncompressed, tx);
 896                 dsl_dir_transfer_space(dd->dd_parent,
 897                     used - accounted_delta,
 898                     DD_USED_CHILD_RSRV, DD_USED_CHILD, tx);
 899         }
 900 }
 901
 902 void
 903 dsl_dir_transfer_space(dsl_dir_t *dd, int64_t delta,
 904     dd_used_t oldtype, dd_used_t newtype, dmu_tx_t *tx)
 905 {
 906         ASSERT(dmu_tx_is_syncing(tx));
 907         ASSERT(oldtype < DD_USED_NUM);
 908         ASSERT(newtype < DD_USED_NUM);
 909
 910         if (delta == 0 || !(dd->dd_phys->dd_flags & DD_FLAG_USED_BREAKDOWN))
 911                 return;
 912
 913         dmu_buf_will_dirty(dd->dd_dbuf, tx);
 914         mutex_enter(&dd->dd_lock);
 915         ASSERT(delta > 0 ?
 916             dd->dd_phys->dd_used_breakdown[oldtype] >= delta :
 917             dd->dd_phys->dd_used_breakdown[newtype] >= -delta);
 918         ASSERT(dd->dd_phys->dd_used_bytes >= ABS(delta));
 919         dd->dd_phys->dd_used_breakdown[oldtype] -= delta;
 920         dd->dd_phys->dd_used_breakdown[newtype] += delta;
 921         mutex_exit(&dd->dd_lock);
 922 }
 923
 924 typedef struct dsl_dir_set_qr_arg {
 925         const char *ddsqra_name;
 926         zprop_source_t ddsqra_source;
 927         uint64_t ddsqra_value;
 928 } dsl_dir_set_qr_arg_t;
 929
 930 static int
 931 dsl_dir_set_quota_check(void *arg, dmu_tx_t *tx)
 932 {
 933         dsl_dir_set_qr_arg_t *ddsqra = arg;
 934         dsl_pool_t *dp = dmu_tx_pool(tx);
 935         dsl_dataset_t *ds;
 936         int error;
 937         uint64_t towrite, newval;
 938
 939         error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds);
 940         if (error != 0)
 941                 return (error);
 942
 943         error = dsl_prop_predict(ds->ds_dir, "quota",
 944             ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval);
 945         if (error != 0) {
 946                 dsl_dataset_rele(ds, FTAG);
 947                 return (error);
 948         }
 949
 950         if (newval == 0) {
 951                 dsl_dataset_rele(ds, FTAG);
 952                 return (0);
 953         }
 954
 955         mutex_enter(&ds->ds_dir->dd_lock);
 956         /*
 957          * If we are doing the preliminary check in open context, and
 958          * there are pending changes, then don't fail it, since the
 959          * pending changes could under-estimate the amount of space to be
 960          * freed up.
 961          */
 962         towrite = dsl_dir_space_towrite(ds->ds_dir);
 963         if ((dmu_tx_is_syncing(tx) || towrite == 0) &&
 964             (newval < ds->ds_dir->dd_phys->dd_reserved ||
 965             newval < ds->ds_dir->dd_phys->dd_used_bytes + towrite)) {
 966                 error = SET_ERROR(ENOSPC);
 967         }
 968         mutex_exit(&ds->ds_dir->dd_lock);
 969         dsl_dataset_rele(ds, FTAG);
 970         return (error);
 971 }
 972
 973 static void
 974 dsl_dir_set_quota_sync(void *arg, dmu_tx_t *tx)
 975 {
 976         dsl_dir_set_qr_arg_t *ddsqra = arg;
 977         dsl_pool_t *dp = dmu_tx_pool(tx);
 978         dsl_dataset_t *ds;
 979         uint64_t newval;
 980
 981         VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds));
 982
 983         if (spa_version(dp->dp_spa) >= SPA_VERSION_RECVD_PROPS) {
 984                 dsl_prop_set_sync_impl(ds, zfs_prop_to_name(ZFS_PROP_QUOTA),
 985                     ddsqra->ddsqra_source, sizeof (ddsqra->ddsqra_value), 1,
 986                     &ddsqra->ddsqra_value, tx);
 987
 988                 VERIFY0(dsl_prop_get_int_ds(ds,
 989                     zfs_prop_to_name(ZFS_PROP_QUOTA), &newval));
 990         } else {
 991                 newval = ddsqra->ddsqra_value;
 992                 spa_history_log_internal_ds(ds, "set", tx, "%s=%lld",
 993                     zfs_prop_to_name(ZFS_PROP_QUOTA), (longlong_t)newval);
 994         }
 995
 996         dmu_buf_will_dirty(ds->ds_dir->dd_dbuf, tx);
 997         mutex_enter(&ds->ds_dir->dd_lock);
 998         ds->ds_dir->dd_phys->dd_quota = newval;
 999         mutex_exit(&ds->ds_dir->dd_lock);
1000         dsl_dataset_rele(ds, FTAG);
1001 }
1002
1003 int
1004 dsl_dir_set_quota(const char *ddname, zprop_source_t source, uint64_t quota)
1005 {
1006         dsl_dir_set_qr_arg_t ddsqra;
1007
1008         ddsqra.ddsqra_name = ddname;
1009         ddsqra.ddsqra_source = source;
1010         ddsqra.ddsqra_value = quota;
1011
1012         return (dsl_sync_task(ddname, dsl_dir_set_quota_check,
1013             dsl_dir_set_quota_sync, &ddsqra, 0));
1014 }
1015
1016 int
1017 dsl_dir_set_reservation_check(void *arg, dmu_tx_t *tx)
1018 {
1019         dsl_dir_set_qr_arg_t *ddsqra = arg;
1020         dsl_pool_t *dp = dmu_tx_pool(tx);
1021         dsl_dataset_t *ds;
1022         dsl_dir_t *dd;
1023         uint64_t newval, used, avail;
1024         int error;
1025
1026         error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds);
1027         if (error != 0)
1028                 return (error);
1029         dd = ds->ds_dir;
1030
1031         /*
1032          * If we are doing the preliminary check in open context, the
1033          * space estimates may be inaccurate.
1034          */
1035         if (!dmu_tx_is_syncing(tx)) {
1036                 dsl_dataset_rele(ds, FTAG);
1037                 return (0);
1038         }
1039
1040         error = dsl_prop_predict(ds->ds_dir,
1041             zfs_prop_to_name(ZFS_PROP_RESERVATION),
1042             ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval);
1043         if (error != 0) {
1044                 dsl_dataset_rele(ds, FTAG);
1045                 return (error);
1046         }
1047
1048         mutex_enter(&dd->dd_lock);
1049         used = dd->dd_phys->dd_used_bytes;
1050         mutex_exit(&dd->dd_lock);
1051
1052         if (dd->dd_parent) {
1053                 avail = dsl_dir_space_available(dd->dd_parent,
1054                     NULL, 0, FALSE);
1055         } else {
1056                 avail = dsl_pool_adjustedsize(dd->dd_pool, B_FALSE) - used;
1057         }
1058
1059         if (MAX(used, newval) > MAX(used, dd->dd_phys->dd_reserved)) {
1060                 uint64_t delta = MAX(used, newval) -
1061                     MAX(used, dd->dd_phys->dd_reserved);
1062
1063                 if (delta > avail ||
1064                     (dd->dd_phys->dd_quota > 0 &&
1065                     newval > dd->dd_phys->dd_quota))
1066                         error = SET_ERROR(ENOSPC);
1067         }
1068
1069         dsl_dataset_rele(ds, FTAG);
1070         return (error);
1071 }
1072
1073 void
1074 dsl_dir_set_reservation_sync_impl(dsl_dir_t *dd, uint64_t value, dmu_tx_t *tx)
1075 {
1076         uint64_t used;
1077         int64_t delta;
1078
1079         dmu_buf_will_dirty(dd->dd_dbuf, tx);
1080
1081         mutex_enter(&dd->dd_lock);
1082         used = dd->dd_phys->dd_used_bytes;
1083         delta = MAX(used, value) - MAX(used, dd->dd_phys->dd_reserved);
1084         dd->dd_phys->dd_reserved = value;
1085
1086         if (dd->dd_parent != NULL) {
1087                 /* Roll up this additional usage into our ancestors */
1088                 dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD_RSRV,
1089                     delta, 0, 0, tx);
1090         }
1091         mutex_exit(&dd->dd_lock);
1092 }
1093
1094 static void
1095 dsl_dir_set_reservation_sync(void *arg, dmu_tx_t *tx)
1096 {
1097         dsl_dir_set_qr_arg_t *ddsqra = arg;
1098         dsl_pool_t *dp = dmu_tx_pool(tx);
1099         dsl_dataset_t *ds;
1100         uint64_t newval;
1101
1102         VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds));
1103
1104         if (spa_version(dp->dp_spa) >= SPA_VERSION_RECVD_PROPS) {
1105                 dsl_prop_set_sync_impl(ds,
1106                     zfs_prop_to_name(ZFS_PROP_RESERVATION),
1107                     ddsqra->ddsqra_source, sizeof (ddsqra->ddsqra_value), 1,
1108                     &ddsqra->ddsqra_value, tx);
1109
1110                 VERIFY0(dsl_prop_get_int_ds(ds,
1111                     zfs_prop_to_name(ZFS_PROP_RESERVATION), &newval));
1112         } else {
1113                 newval = ddsqra->ddsqra_value;
1114                 spa_history_log_internal_ds(ds, "set", tx, "%s=%lld",
1115                     zfs_prop_to_name(ZFS_PROP_RESERVATION),
1116                     (longlong_t)newval);
1117         }
1118
1119         dsl_dir_set_reservation_sync_impl(ds->ds_dir, newval, tx);
1120         dsl_dataset_rele(ds, FTAG);
1121 }
1122
1123 int
1124 dsl_dir_set_reservation(const char *ddname, zprop_source_t source,
1125     uint64_t reservation)
1126 {
1127         dsl_dir_set_qr_arg_t ddsqra;
1128
1129         ddsqra.ddsqra_name = ddname;
1130         ddsqra.ddsqra_source = source;
1131         ddsqra.ddsqra_value = reservation;
1132
1133         return (dsl_sync_task(ddname, dsl_dir_set_reservation_check,
1134             dsl_dir_set_reservation_sync, &ddsqra, 0));
1135 }
1136
1137 static dsl_dir_t *
1138 closest_common_ancestor(dsl_dir_t *ds1, dsl_dir_t *ds2)
1139 {
1140         for (; ds1; ds1 = ds1->dd_parent) {
1141                 dsl_dir_t *dd;
1142                 for (dd = ds2; dd; dd = dd->dd_parent) {
1143                         if (ds1 == dd)
1144                                 return (dd);
1145                 }
1146         }
1147         return (NULL);
1148 }
1149
1150 /*
1151  * If delta is applied to dd, how much of that delta would be applied to
1152  * ancestor?  Syncing context only.
1153  */
1154 static int64_t
1155 would_change(dsl_dir_t *dd, int64_t delta, dsl_dir_t *ancestor)
1156 {
1157         if (dd == ancestor)
1158                 return (delta);
1159
1160         mutex_enter(&dd->dd_lock);
1161         delta = parent_delta(dd, dd->dd_phys->dd_used_bytes, delta);
1162         mutex_exit(&dd->dd_lock);
1163         return (would_change(dd->dd_parent, delta, ancestor));
1164 }
1165
1166 typedef struct dsl_dir_rename_arg {
1167         const char *ddra_oldname;
1168         const char *ddra_newname;
1169 } dsl_dir_rename_arg_t;
1170
1171 /* ARGSUSED */
1172 static int
1173 dsl_valid_rename(dsl_pool_t *dp, dsl_dataset_t *ds, void *arg)
1174 {
1175         int *deltap = arg;
1176         char namebuf[MAXNAMELEN];
1177
1178         dsl_dataset_name(ds, namebuf);
1179
1180         if (strlen(namebuf) + *deltap >= MAXNAMELEN)
1181                 return (SET_ERROR(ENAMETOOLONG));
1182         return (0);
1183 }
1184
1185 static int
1186 dsl_dir_rename_check(void *arg, dmu_tx_t *tx)
1187 {
1188         dsl_dir_rename_arg_t *ddra = arg;
1189         dsl_pool_t *dp = dmu_tx_pool(tx);
1190         dsl_dir_t *dd, *newparent;
1191         const char *mynewname;
1192         int error;
1193         int delta = strlen(ddra->ddra_newname) - strlen(ddra->ddra_oldname);
1194
1195         /* target dir should exist */
1196         error = dsl_dir_hold(dp, ddra->ddra_oldname, FTAG, &dd, NULL);
1197         if (error != 0)
1198                 return (error);
1199
1200         /* new parent should exist */
1201         error = dsl_dir_hold(dp, ddra->ddra_newname, FTAG,
1202             &newparent, &mynewname);
1203         if (error != 0) {
1204                 dsl_dir_rele(dd, FTAG);
1205                 return (error);
1206         }
1207
1208         /* can't rename to different pool */
1209         if (dd->dd_pool != newparent->dd_pool) {
1210                 dsl_dir_rele(newparent, FTAG);
1211                 dsl_dir_rele(dd, FTAG);
1212                 return (SET_ERROR(EXDEV));
1213         }
1214
1215         /* new name should not already exist */
1216         if (mynewname == NULL) {
1217                 dsl_dir_rele(newparent, FTAG);
1218                 dsl_dir_rele(dd, FTAG);
1219                 return (SET_ERROR(EEXIST));
1220         }
1221
1222         /* if the name length is growing, validate child name lengths */
1223         if (delta > 0) {
1224                 error = dmu_objset_find_dp(dp, dd->dd_object, dsl_valid_rename,
1225                     &delta, DS_FIND_CHILDREN | DS_FIND_SNAPSHOTS);
1226                 if (error != 0) {
1227                         dsl_dir_rele(newparent, FTAG);
1228                         dsl_dir_rele(dd, FTAG);
1229                         return (error);
1230                 }
1231         }
1232
1233         if (newparent != dd->dd_parent) {
1234                 /* is there enough space? */
1235                 uint64_t myspace =
1236                     MAX(dd->dd_phys->dd_used_bytes, dd->dd_phys->dd_reserved);
1237
1238                 /* no rename into our descendant */
1239                 if (closest_common_ancestor(dd, newparent) == dd) {
1240                         dsl_dir_rele(newparent, FTAG);
1241                         dsl_dir_rele(dd, FTAG);
1242                         return (SET_ERROR(EINVAL));
1243                 }
1244
1245                 error = dsl_dir_transfer_possible(dd->dd_parent,
1246                     newparent, myspace);
1247                 if (error != 0) {
1248                         dsl_dir_rele(newparent, FTAG);
1249                         dsl_dir_rele(dd, FTAG);
1250                         return (error);
1251                 }
1252         }
1253
1254         dsl_dir_rele(newparent, FTAG);
1255         dsl_dir_rele(dd, FTAG);
1256         return (0);
1257 }
1258
1259 static void
1260 dsl_dir_rename_sync(void *arg, dmu_tx_t *tx)
1261 {
1262         dsl_dir_rename_arg_t *ddra = arg;
1263         dsl_pool_t *dp = dmu_tx_pool(tx);
1264         dsl_dir_t *dd, *newparent;
1265         const char *mynewname;
1266         int error;
1267         objset_t *mos = dp->dp_meta_objset;
1268
1269         VERIFY0(dsl_dir_hold(dp, ddra->ddra_oldname, FTAG, &dd, NULL));
1270         VERIFY0(dsl_dir_hold(dp, ddra->ddra_newname, FTAG, &newparent,
1271             &mynewname));
1272
1273         /* Log this before we change the name. */
1274         spa_history_log_internal_dd(dd, "rename", tx,
1275             "-> %s", ddra->ddra_newname);
1276
1277         if (newparent != dd->dd_parent) {
1278                 dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD,
1279                     -dd->dd_phys->dd_used_bytes,
1280                     -dd->dd_phys->dd_compressed_bytes,
1281                     -dd->dd_phys->dd_uncompressed_bytes, tx);
1282                 dsl_dir_diduse_space(newparent, DD_USED_CHILD,
1283                     dd->dd_phys->dd_used_bytes,
1284                     dd->dd_phys->dd_compressed_bytes,
1285                     dd->dd_phys->dd_uncompressed_bytes, tx);
1286
1287                 if (dd->dd_phys->dd_reserved > dd->dd_phys->dd_used_bytes) {
1288                         uint64_t unused_rsrv = dd->dd_phys->dd_reserved -
1289                             dd->dd_phys->dd_used_bytes;
1290
1291                         dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD_RSRV,
1292                             -unused_rsrv, 0, 0, tx);
1293                         dsl_dir_diduse_space(newparent, DD_USED_CHILD_RSRV,
1294                             unused_rsrv, 0, 0, tx);
1295                 }
1296         }
1297
1298         dmu_buf_will_dirty(dd->dd_dbuf, tx);
1299
1300         /* remove from old parent zapobj */
1301         error = zap_remove(mos, dd->dd_parent->dd_phys->dd_child_dir_zapobj,
1302             dd->dd_myname, tx);
1303         ASSERT0(error);
1304
1305         (void) strcpy(dd->dd_myname, mynewname);
1306         dsl_dir_rele(dd->dd_parent, dd);
1307         dd->dd_phys->dd_parent_obj = newparent->dd_object;
1308         VERIFY0(dsl_dir_hold_obj(dp,
1309             newparent->dd_object, NULL, dd, &dd->dd_parent));
1310
1311         /* add to new parent zapobj */
1312         VERIFY0(zap_add(mos, newparent->dd_phys->dd_child_dir_zapobj,
1313             dd->dd_myname, 8, 1, &dd->dd_object, tx));
1314
1315 #ifdef _KERNEL
1316         zvol_rename_minors(ddra->ddra_oldname, ddra->ddra_newname);
1317 #endif
1318
1319         dsl_prop_notify_all(dd);
1320
1321         dsl_dir_rele(newparent, FTAG);
1322         dsl_dir_rele(dd, FTAG);
1323 }
1324
1325 int
1326 dsl_dir_rename(const char *oldname, const char *newname)
1327 {
1328         dsl_dir_rename_arg_t ddra;
1329
1330         ddra.ddra_oldname = oldname;
1331         ddra.ddra_newname = newname;
1332
1333         return (dsl_sync_task(oldname,
1334             dsl_dir_rename_check, dsl_dir_rename_sync, &ddra, 3));
1335 }
1336
1337 int
1338 dsl_dir_transfer_possible(dsl_dir_t *sdd, dsl_dir_t *tdd, uint64_t space)
1339 {
1340         dsl_dir_t *ancestor;
1341         int64_t adelta;
1342         uint64_t avail;
1343
1344         ancestor = closest_common_ancestor(sdd, tdd);
1345         adelta = would_change(sdd, -space, ancestor);
1346         avail = dsl_dir_space_available(tdd, ancestor, adelta, FALSE);
1347         if (avail < space)
1348                 return (SET_ERROR(ENOSPC));
1349
1350         return (0);
1351 }
1352
1353 timestruc_t
1354 dsl_dir_snap_cmtime(dsl_dir_t *dd)
1355 {
1356         timestruc_t t;
1357
1358         mutex_enter(&dd->dd_lock);
1359         t = dd->dd_snap_cmtime;
1360         mutex_exit(&dd->dd_lock);
1361
1362         return (t);
1363 }
1364
1365 void
1366 dsl_dir_snap_cmtime_update(dsl_dir_t *dd)
1367 {
1368         timestruc_t t;
1369
1370         gethrestime(&t);
1371         mutex_enter(&dd->dd_lock);
1372         dd->dd_snap_cmtime = t;
1373         mutex_exit(&dd->dd_lock);
1374 }
1375
1376 void
1377 dsl_dir_zapify(dsl_dir_t *dd, dmu_tx_t *tx)
1378 {
1379         objset_t *mos = dd->dd_pool->dp_meta_objset;
1380         dmu_object_zapify(mos, dd->dd_object, DMU_OT_DSL_DIR, tx);
1381 }
1382
1383 #if defined(_KERNEL) && defined(HAVE_SPL)
1384 EXPORT_SYMBOL(dsl_dir_set_quota);
1385 EXPORT_SYMBOL(dsl_dir_set_reservation);
1386 #endif