dsl_dir.c revision 247585
1/* 2 * CDDL HEADER START 3 * 4 * The contents of this file are subject to the terms of the 5 * Common Development and Distribution License (the "License"). 6 * You may not use this file except in compliance with the License. 7 * 8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 9 * or http://www.opensolaris.org/os/licensing. 10 * See the License for the specific language governing permissions 11 * and limitations under the License. 12 * 13 * When distributing Covered Code, include this CDDL HEADER in each 14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 15 * If applicable, add the following below this CDDL HEADER, with the 16 * fields enclosed by brackets "[]" replaced with your own identifying 17 * information: Portions Copyright [yyyy] [name of copyright owner] 18 * 19 * CDDL HEADER END 20 */ 21/* 22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. 23 * Copyright (c) 2011 Pawel Jakub Dawidek <pawel@dawidek.net>. 24 * All rights reserved. 25 */ 26 27#include <sys/dmu.h> 28#include <sys/dmu_objset.h> 29#include <sys/dmu_tx.h> 30#include <sys/dsl_dataset.h> 31#include <sys/dsl_dir.h> 32#include <sys/dsl_prop.h> 33#include <sys/dsl_synctask.h> 34#include <sys/dsl_deleg.h> 35#include <sys/spa.h> 36#include <sys/metaslab.h> 37#include <sys/zap.h> 38#include <sys/zio.h> 39#include <sys/arc.h> 40#include <sys/sunddi.h> 41#include <sys/zvol.h> 42#ifdef _KERNEL 43#include <sys/zfs_vfsops.h> 44#endif 45#include "zfs_namecheck.h" 46 47static uint64_t dsl_dir_space_towrite(dsl_dir_t *dd); 48static void dsl_dir_set_reservation_sync(void *arg1, void *arg2, dmu_tx_t *tx); 49 50 51/* ARGSUSED */ 52static void 53dsl_dir_evict(dmu_buf_t *db, void *arg) 54{ 55 dsl_dir_t *dd = arg; 56 dsl_pool_t *dp = dd->dd_pool; 57 int t; 58 59 for (t = 0; t < TXG_SIZE; t++) { 60 ASSERT(!txg_list_member(&dp->dp_dirty_dirs, dd, t)); 61 ASSERT(dd->dd_tempreserved[t] == 0); 62 ASSERT(dd->dd_space_towrite[t] == 0); 63 } 64 65 if (dd->dd_parent) 66 dsl_dir_close(dd->dd_parent, dd); 67 68 spa_close(dd->dd_pool->dp_spa, dd); 69 70 /* 71 * The props callback list should have been cleaned up by 72 * objset_evict(). 73 */ 74 list_destroy(&dd->dd_prop_cbs); 75 mutex_destroy(&dd->dd_lock); 76 kmem_free(dd, sizeof (dsl_dir_t)); 77} 78 79int 80dsl_dir_open_obj(dsl_pool_t *dp, uint64_t ddobj, 81 const char *tail, void *tag, dsl_dir_t **ddp) 82{ 83 dmu_buf_t *dbuf; 84 dsl_dir_t *dd; 85 int err; 86 87 ASSERT(RW_LOCK_HELD(&dp->dp_config_rwlock) || 88 dsl_pool_sync_context(dp)); 89 90 err = dmu_bonus_hold(dp->dp_meta_objset, ddobj, tag, &dbuf); 91 if (err) 92 return (err); 93 dd = dmu_buf_get_user(dbuf); 94#ifdef ZFS_DEBUG 95 { 96 dmu_object_info_t doi; 97 dmu_object_info_from_db(dbuf, &doi); 98 ASSERT3U(doi.doi_type, ==, DMU_OT_DSL_DIR); 99 ASSERT3U(doi.doi_bonus_size, >=, sizeof (dsl_dir_phys_t)); 100 } 101#endif 102 if (dd == NULL) { 103 dsl_dir_t *winner; 104 105 dd = kmem_zalloc(sizeof (dsl_dir_t), KM_SLEEP); 106 dd->dd_object = ddobj; 107 dd->dd_dbuf = dbuf; 108 dd->dd_pool = dp; 109 dd->dd_phys = dbuf->db_data; 110 mutex_init(&dd->dd_lock, NULL, MUTEX_DEFAULT, NULL); 111 112 list_create(&dd->dd_prop_cbs, sizeof (dsl_prop_cb_record_t), 113 offsetof(dsl_prop_cb_record_t, cbr_node)); 114 115 dsl_dir_snap_cmtime_update(dd); 116 117 if (dd->dd_phys->dd_parent_obj) { 118 err = dsl_dir_open_obj(dp, dd->dd_phys->dd_parent_obj, 119 NULL, dd, &dd->dd_parent); 120 if (err) 121 goto errout; 122 if (tail) { 123#ifdef ZFS_DEBUG 124 uint64_t foundobj; 125 126 err = zap_lookup(dp->dp_meta_objset, 127 dd->dd_parent->dd_phys->dd_child_dir_zapobj, 128 tail, sizeof (foundobj), 1, &foundobj); 129 ASSERT(err || foundobj == ddobj); 130#endif 131 (void) strcpy(dd->dd_myname, tail); 132 } else { 133 err = zap_value_search(dp->dp_meta_objset, 134 dd->dd_parent->dd_phys->dd_child_dir_zapobj, 135 ddobj, 0, dd->dd_myname); 136 } 137 if (err) 138 goto errout; 139 } else { 140 (void) strcpy(dd->dd_myname, spa_name(dp->dp_spa)); 141 } 142 143 if (dsl_dir_is_clone(dd)) { 144 dmu_buf_t *origin_bonus; 145 dsl_dataset_phys_t *origin_phys; 146 147 /* 148 * We can't open the origin dataset, because 149 * that would require opening this dsl_dir. 150 * Just look at its phys directly instead. 151 */ 152 err = dmu_bonus_hold(dp->dp_meta_objset, 153 dd->dd_phys->dd_origin_obj, FTAG, &origin_bonus); 154 if (err) 155 goto errout; 156 origin_phys = origin_bonus->db_data; 157 dd->dd_origin_txg = 158 origin_phys->ds_creation_txg; 159 dmu_buf_rele(origin_bonus, FTAG); 160 } 161 162 winner = dmu_buf_set_user_ie(dbuf, dd, &dd->dd_phys, 163 dsl_dir_evict); 164 if (winner) { 165 if (dd->dd_parent) 166 dsl_dir_close(dd->dd_parent, dd); 167 mutex_destroy(&dd->dd_lock); 168 kmem_free(dd, sizeof (dsl_dir_t)); 169 dd = winner; 170 } else { 171 spa_open_ref(dp->dp_spa, dd); 172 } 173 } 174 175 /* 176 * The dsl_dir_t has both open-to-close and instantiate-to-evict 177 * holds on the spa. We need the open-to-close holds because 178 * otherwise the spa_refcnt wouldn't change when we open a 179 * dir which the spa also has open, so we could incorrectly 180 * think it was OK to unload/export/destroy the pool. We need 181 * the instantiate-to-evict hold because the dsl_dir_t has a 182 * pointer to the dd_pool, which has a pointer to the spa_t. 183 */ 184 spa_open_ref(dp->dp_spa, tag); 185 ASSERT3P(dd->dd_pool, ==, dp); 186 ASSERT3U(dd->dd_object, ==, ddobj); 187 ASSERT3P(dd->dd_dbuf, ==, dbuf); 188 *ddp = dd; 189 return (0); 190 191errout: 192 if (dd->dd_parent) 193 dsl_dir_close(dd->dd_parent, dd); 194 mutex_destroy(&dd->dd_lock); 195 kmem_free(dd, sizeof (dsl_dir_t)); 196 dmu_buf_rele(dbuf, tag); 197 return (err); 198} 199 200void 201dsl_dir_close(dsl_dir_t *dd, void *tag) 202{ 203 dprintf_dd(dd, "%s\n", ""); 204 spa_close(dd->dd_pool->dp_spa, tag); 205 dmu_buf_rele(dd->dd_dbuf, tag); 206} 207 208/* buf must be long enough (MAXNAMELEN + strlen(MOS_DIR_NAME) + 1 should do) */ 209void 210dsl_dir_name(dsl_dir_t *dd, char *buf) 211{ 212 if (dd->dd_parent) { 213 dsl_dir_name(dd->dd_parent, buf); 214 (void) strcat(buf, "/"); 215 } else { 216 buf[0] = '\0'; 217 } 218 if (!MUTEX_HELD(&dd->dd_lock)) { 219 /* 220 * recursive mutex so that we can use 221 * dprintf_dd() with dd_lock held 222 */ 223 mutex_enter(&dd->dd_lock); 224 (void) strcat(buf, dd->dd_myname); 225 mutex_exit(&dd->dd_lock); 226 } else { 227 (void) strcat(buf, dd->dd_myname); 228 } 229} 230 231/* Calculate name length, avoiding all the strcat calls of dsl_dir_name */ 232int 233dsl_dir_namelen(dsl_dir_t *dd) 234{ 235 int result = 0; 236 237 if (dd->dd_parent) { 238 /* parent's name + 1 for the "/" */ 239 result = dsl_dir_namelen(dd->dd_parent) + 1; 240 } 241 242 if (!MUTEX_HELD(&dd->dd_lock)) { 243 /* see dsl_dir_name */ 244 mutex_enter(&dd->dd_lock); 245 result += strlen(dd->dd_myname); 246 mutex_exit(&dd->dd_lock); 247 } else { 248 result += strlen(dd->dd_myname); 249 } 250 251 return (result); 252} 253 254static int 255getcomponent(const char *path, char *component, const char **nextp) 256{ 257 char *p; 258 if ((path == NULL) || (path[0] == '\0')) 259 return (ENOENT); 260 /* This would be a good place to reserve some namespace... */ 261 p = strpbrk(path, "/@"); 262 if (p && (p[1] == '/' || p[1] == '@')) { 263 /* two separators in a row */ 264 return (EINVAL); 265 } 266 if (p == NULL || p == path) { 267 /* 268 * if the first thing is an @ or /, it had better be an 269 * @ and it had better not have any more ats or slashes, 270 * and it had better have something after the @. 271 */ 272 if (p != NULL && 273 (p[0] != '@' || strpbrk(path+1, "/@") || p[1] == '\0')) 274 return (EINVAL); 275 if (strlen(path) >= MAXNAMELEN) 276 return (ENAMETOOLONG); 277 (void) strcpy(component, path); 278 p = NULL; 279 } else if (p[0] == '/') { 280 if (p-path >= MAXNAMELEN) 281 return (ENAMETOOLONG); 282 (void) strncpy(component, path, p - path); 283 component[p-path] = '\0'; 284 p++; 285 } else if (p[0] == '@') { 286 /* 287 * if the next separator is an @, there better not be 288 * any more slashes. 289 */ 290 if (strchr(path, '/')) 291 return (EINVAL); 292 if (p-path >= MAXNAMELEN) 293 return (ENAMETOOLONG); 294 (void) strncpy(component, path, p - path); 295 component[p-path] = '\0'; 296 } else { 297 ASSERT(!"invalid p"); 298 } 299 *nextp = p; 300 return (0); 301} 302 303/* 304 * same as dsl_open_dir, ignore the first component of name and use the 305 * spa instead 306 */ 307int 308dsl_dir_open_spa(spa_t *spa, const char *name, void *tag, 309 dsl_dir_t **ddp, const char **tailp) 310{ 311 char buf[MAXNAMELEN]; 312 const char *next, *nextnext = NULL; 313 int err; 314 dsl_dir_t *dd; 315 dsl_pool_t *dp; 316 uint64_t ddobj; 317 int openedspa = FALSE; 318 319 dprintf("%s\n", name); 320 321 err = getcomponent(name, buf, &next); 322 if (err) 323 return (err); 324 if (spa == NULL) { 325 err = spa_open(buf, &spa, FTAG); 326 if (err) { 327 dprintf("spa_open(%s) failed\n", buf); 328 return (err); 329 } 330 openedspa = TRUE; 331 332 /* XXX this assertion belongs in spa_open */ 333 ASSERT(!dsl_pool_sync_context(spa_get_dsl(spa))); 334 } 335 336 dp = spa_get_dsl(spa); 337 338 rw_enter(&dp->dp_config_rwlock, RW_READER); 339 err = dsl_dir_open_obj(dp, dp->dp_root_dir_obj, NULL, tag, &dd); 340 if (err) { 341 rw_exit(&dp->dp_config_rwlock); 342 if (openedspa) 343 spa_close(spa, FTAG); 344 return (err); 345 } 346 347 while (next != NULL) { 348 dsl_dir_t *child_ds; 349 err = getcomponent(next, buf, &nextnext); 350 if (err) 351 break; 352 ASSERT(next[0] != '\0'); 353 if (next[0] == '@') 354 break; 355 dprintf("looking up %s in obj%lld\n", 356 buf, dd->dd_phys->dd_child_dir_zapobj); 357 358 err = zap_lookup(dp->dp_meta_objset, 359 dd->dd_phys->dd_child_dir_zapobj, 360 buf, sizeof (ddobj), 1, &ddobj); 361 if (err) { 362 if (err == ENOENT) 363 err = 0; 364 break; 365 } 366 367 err = dsl_dir_open_obj(dp, ddobj, buf, tag, &child_ds); 368 if (err) 369 break; 370 dsl_dir_close(dd, tag); 371 dd = child_ds; 372 next = nextnext; 373 } 374 rw_exit(&dp->dp_config_rwlock); 375 376 if (err) { 377 dsl_dir_close(dd, tag); 378 if (openedspa) 379 spa_close(spa, FTAG); 380 return (err); 381 } 382 383 /* 384 * It's an error if there's more than one component left, or 385 * tailp==NULL and there's any component left. 386 */ 387 if (next != NULL && 388 (tailp == NULL || (nextnext && nextnext[0] != '\0'))) { 389 /* bad path name */ 390 dsl_dir_close(dd, tag); 391 dprintf("next=%p (%s) tail=%p\n", next, next?next:"", tailp); 392 err = ENOENT; 393 } 394 if (tailp) 395 *tailp = next; 396 if (openedspa) 397 spa_close(spa, FTAG); 398 *ddp = dd; 399 return (err); 400} 401 402/* 403 * Return the dsl_dir_t, and possibly the last component which couldn't 404 * be found in *tail. Return NULL if the path is bogus, or if 405 * tail==NULL and we couldn't parse the whole name. (*tail)[0] == '@' 406 * means that the last component is a snapshot. 407 */ 408int 409dsl_dir_open(const char *name, void *tag, dsl_dir_t **ddp, const char **tailp) 410{ 411 return (dsl_dir_open_spa(NULL, name, tag, ddp, tailp)); 412} 413 414uint64_t 415dsl_dir_create_sync(dsl_pool_t *dp, dsl_dir_t *pds, const char *name, 416 dmu_tx_t *tx) 417{ 418 objset_t *mos = dp->dp_meta_objset; 419 uint64_t ddobj; 420 dsl_dir_phys_t *ddphys; 421 dmu_buf_t *dbuf; 422 423 ddobj = dmu_object_alloc(mos, DMU_OT_DSL_DIR, 0, 424 DMU_OT_DSL_DIR, sizeof (dsl_dir_phys_t), tx); 425 if (pds) { 426 VERIFY(0 == zap_add(mos, pds->dd_phys->dd_child_dir_zapobj, 427 name, sizeof (uint64_t), 1, &ddobj, tx)); 428 } else { 429 /* it's the root dir */ 430 VERIFY(0 == zap_add(mos, DMU_POOL_DIRECTORY_OBJECT, 431 DMU_POOL_ROOT_DATASET, sizeof (uint64_t), 1, &ddobj, tx)); 432 } 433 VERIFY(0 == dmu_bonus_hold(mos, ddobj, FTAG, &dbuf)); 434 dmu_buf_will_dirty(dbuf, tx); 435 ddphys = dbuf->db_data; 436 437 ddphys->dd_creation_time = gethrestime_sec(); 438 if (pds) 439 ddphys->dd_parent_obj = pds->dd_object; 440 ddphys->dd_props_zapobj = zap_create(mos, 441 DMU_OT_DSL_PROPS, DMU_OT_NONE, 0, tx); 442 ddphys->dd_child_dir_zapobj = zap_create(mos, 443 DMU_OT_DSL_DIR_CHILD_MAP, DMU_OT_NONE, 0, tx); 444 if (spa_version(dp->dp_spa) >= SPA_VERSION_USED_BREAKDOWN) 445 ddphys->dd_flags |= DD_FLAG_USED_BREAKDOWN; 446 dmu_buf_rele(dbuf, FTAG); 447 448 return (ddobj); 449} 450 451/* ARGSUSED */ 452int 453dsl_dir_destroy_check(void *arg1, void *arg2, dmu_tx_t *tx) 454{ 455 dsl_dataset_t *ds = arg1; 456 dsl_dir_t *dd = ds->ds_dir; 457 dsl_pool_t *dp = dd->dd_pool; 458 objset_t *mos = dp->dp_meta_objset; 459 int err; 460 uint64_t count; 461 462 /* 463 * There should be exactly two holds, both from 464 * dsl_dataset_destroy: one on the dd directory, and one on its 465 * head ds. If there are more holds, then a concurrent thread is 466 * performing a lookup inside this dir while we're trying to destroy 467 * it. To minimize this possibility, we perform this check only 468 * in syncing context and fail the operation if we encounter 469 * additional holds. The dp_config_rwlock ensures that nobody else 470 * opens it after we check. 471 */ 472 if (dmu_tx_is_syncing(tx) && dmu_buf_refcount(dd->dd_dbuf) > 2) 473 return (EBUSY); 474 475 err = zap_count(mos, dd->dd_phys->dd_child_dir_zapobj, &count); 476 if (err) 477 return (err); 478 if (count != 0) 479 return (EEXIST); 480 481 return (0); 482} 483 484void 485dsl_dir_destroy_sync(void *arg1, void *tag, dmu_tx_t *tx) 486{ 487 dsl_dataset_t *ds = arg1; 488 dsl_dir_t *dd = ds->ds_dir; 489 objset_t *mos = dd->dd_pool->dp_meta_objset; 490 dsl_prop_setarg_t psa; 491 uint64_t value = 0; 492 uint64_t obj; 493 dd_used_t t; 494 495 ASSERT(RW_WRITE_HELD(&dd->dd_pool->dp_config_rwlock)); 496 ASSERT(dd->dd_phys->dd_head_dataset_obj == 0); 497 498 /* Remove our reservation. */ 499 dsl_prop_setarg_init_uint64(&psa, "reservation", 500 (ZPROP_SRC_NONE | ZPROP_SRC_LOCAL | ZPROP_SRC_RECEIVED), 501 &value); 502 psa.psa_effective_value = 0; /* predict default value */ 503 504 dsl_dir_set_reservation_sync(ds, &psa, tx); 505 506 ASSERT0(dd->dd_phys->dd_used_bytes); 507 ASSERT0(dd->dd_phys->dd_reserved); 508 for (t = 0; t < DD_USED_NUM; t++) 509 ASSERT0(dd->dd_phys->dd_used_breakdown[t]); 510 511 VERIFY(0 == zap_destroy(mos, dd->dd_phys->dd_child_dir_zapobj, tx)); 512 VERIFY(0 == zap_destroy(mos, dd->dd_phys->dd_props_zapobj, tx)); 513 VERIFY(0 == dsl_deleg_destroy(mos, dd->dd_phys->dd_deleg_zapobj, tx)); 514 VERIFY(0 == zap_remove(mos, 515 dd->dd_parent->dd_phys->dd_child_dir_zapobj, dd->dd_myname, tx)); 516 517 obj = dd->dd_object; 518 dsl_dir_close(dd, tag); 519 VERIFY(0 == dmu_object_free(mos, obj, tx)); 520} 521 522boolean_t 523dsl_dir_is_clone(dsl_dir_t *dd) 524{ 525 return (dd->dd_phys->dd_origin_obj && 526 (dd->dd_pool->dp_origin_snap == NULL || 527 dd->dd_phys->dd_origin_obj != 528 dd->dd_pool->dp_origin_snap->ds_object)); 529} 530 531void 532dsl_dir_stats(dsl_dir_t *dd, nvlist_t *nv) 533{ 534 mutex_enter(&dd->dd_lock); 535 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USED, 536 dd->dd_phys->dd_used_bytes); 537 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_QUOTA, dd->dd_phys->dd_quota); 538 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_RESERVATION, 539 dd->dd_phys->dd_reserved); 540 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_COMPRESSRATIO, 541 dd->dd_phys->dd_compressed_bytes == 0 ? 100 : 542 (dd->dd_phys->dd_uncompressed_bytes * 100 / 543 dd->dd_phys->dd_compressed_bytes)); 544 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_LOGICALUSED, 545 dd->dd_phys->dd_uncompressed_bytes); 546 if (dd->dd_phys->dd_flags & DD_FLAG_USED_BREAKDOWN) { 547 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDSNAP, 548 dd->dd_phys->dd_used_breakdown[DD_USED_SNAP]); 549 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDDS, 550 dd->dd_phys->dd_used_breakdown[DD_USED_HEAD]); 551 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDREFRESERV, 552 dd->dd_phys->dd_used_breakdown[DD_USED_REFRSRV]); 553 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDCHILD, 554 dd->dd_phys->dd_used_breakdown[DD_USED_CHILD] + 555 dd->dd_phys->dd_used_breakdown[DD_USED_CHILD_RSRV]); 556 } 557 mutex_exit(&dd->dd_lock); 558 559 rw_enter(&dd->dd_pool->dp_config_rwlock, RW_READER); 560 if (dsl_dir_is_clone(dd)) { 561 dsl_dataset_t *ds; 562 char buf[MAXNAMELEN]; 563 564 VERIFY(0 == dsl_dataset_hold_obj(dd->dd_pool, 565 dd->dd_phys->dd_origin_obj, FTAG, &ds)); 566 dsl_dataset_name(ds, buf); 567 dsl_dataset_rele(ds, FTAG); 568 dsl_prop_nvlist_add_string(nv, ZFS_PROP_ORIGIN, buf); 569 } 570 rw_exit(&dd->dd_pool->dp_config_rwlock); 571} 572 573void 574dsl_dir_dirty(dsl_dir_t *dd, dmu_tx_t *tx) 575{ 576 dsl_pool_t *dp = dd->dd_pool; 577 578 ASSERT(dd->dd_phys); 579 580 if (txg_list_add(&dp->dp_dirty_dirs, dd, tx->tx_txg) == 0) { 581 /* up the hold count until we can be written out */ 582 dmu_buf_add_ref(dd->dd_dbuf, dd); 583 } 584} 585 586static int64_t 587parent_delta(dsl_dir_t *dd, uint64_t used, int64_t delta) 588{ 589 uint64_t old_accounted = MAX(used, dd->dd_phys->dd_reserved); 590 uint64_t new_accounted = MAX(used + delta, dd->dd_phys->dd_reserved); 591 return (new_accounted - old_accounted); 592} 593 594void 595dsl_dir_sync(dsl_dir_t *dd, dmu_tx_t *tx) 596{ 597 ASSERT(dmu_tx_is_syncing(tx)); 598 599 mutex_enter(&dd->dd_lock); 600 ASSERT0(dd->dd_tempreserved[tx->tx_txg&TXG_MASK]); 601 dprintf_dd(dd, "txg=%llu towrite=%lluK\n", tx->tx_txg, 602 dd->dd_space_towrite[tx->tx_txg&TXG_MASK] / 1024); 603 dd->dd_space_towrite[tx->tx_txg&TXG_MASK] = 0; 604 mutex_exit(&dd->dd_lock); 605 606 /* release the hold from dsl_dir_dirty */ 607 dmu_buf_rele(dd->dd_dbuf, dd); 608} 609 610static uint64_t 611dsl_dir_space_towrite(dsl_dir_t *dd) 612{ 613 uint64_t space = 0; 614 int i; 615 616 ASSERT(MUTEX_HELD(&dd->dd_lock)); 617 618 for (i = 0; i < TXG_SIZE; i++) { 619 space += dd->dd_space_towrite[i&TXG_MASK]; 620 ASSERT3U(dd->dd_space_towrite[i&TXG_MASK], >=, 0); 621 } 622 return (space); 623} 624 625/* 626 * How much space would dd have available if ancestor had delta applied 627 * to it? If ondiskonly is set, we're only interested in what's 628 * on-disk, not estimated pending changes. 629 */ 630uint64_t 631dsl_dir_space_available(dsl_dir_t *dd, 632 dsl_dir_t *ancestor, int64_t delta, int ondiskonly) 633{ 634 uint64_t parentspace, myspace, quota, used; 635 636 /* 637 * If there are no restrictions otherwise, assume we have 638 * unlimited space available. 639 */ 640 quota = UINT64_MAX; 641 parentspace = UINT64_MAX; 642 643 if (dd->dd_parent != NULL) { 644 parentspace = dsl_dir_space_available(dd->dd_parent, 645 ancestor, delta, ondiskonly); 646 } 647 648 mutex_enter(&dd->dd_lock); 649 if (dd->dd_phys->dd_quota != 0) 650 quota = dd->dd_phys->dd_quota; 651 used = dd->dd_phys->dd_used_bytes; 652 if (!ondiskonly) 653 used += dsl_dir_space_towrite(dd); 654 655 if (dd->dd_parent == NULL) { 656 uint64_t poolsize = dsl_pool_adjustedsize(dd->dd_pool, FALSE); 657 quota = MIN(quota, poolsize); 658 } 659 660 if (dd->dd_phys->dd_reserved > used && parentspace != UINT64_MAX) { 661 /* 662 * We have some space reserved, in addition to what our 663 * parent gave us. 664 */ 665 parentspace += dd->dd_phys->dd_reserved - used; 666 } 667 668 if (dd == ancestor) { 669 ASSERT(delta <= 0); 670 ASSERT(used >= -delta); 671 used += delta; 672 if (parentspace != UINT64_MAX) 673 parentspace -= delta; 674 } 675 676 if (used > quota) { 677 /* over quota */ 678 myspace = 0; 679 } else { 680 /* 681 * the lesser of the space provided by our parent and 682 * the space left in our quota 683 */ 684 myspace = MIN(parentspace, quota - used); 685 } 686 687 mutex_exit(&dd->dd_lock); 688 689 return (myspace); 690} 691 692struct tempreserve { 693 list_node_t tr_node; 694 dsl_pool_t *tr_dp; 695 dsl_dir_t *tr_ds; 696 uint64_t tr_size; 697}; 698 699static int 700dsl_dir_tempreserve_impl(dsl_dir_t *dd, uint64_t asize, boolean_t netfree, 701 boolean_t ignorequota, boolean_t checkrefquota, list_t *tr_list, 702 dmu_tx_t *tx, boolean_t first) 703{ 704 uint64_t txg = tx->tx_txg; 705 uint64_t est_inflight, used_on_disk, quota, parent_rsrv; 706 uint64_t deferred = 0; 707 struct tempreserve *tr; 708 int retval = EDQUOT; 709 int txgidx = txg & TXG_MASK; 710 int i; 711 uint64_t ref_rsrv = 0; 712 713 ASSERT3U(txg, !=, 0); 714 ASSERT3S(asize, >, 0); 715 716 mutex_enter(&dd->dd_lock); 717 718 /* 719 * Check against the dsl_dir's quota. We don't add in the delta 720 * when checking for over-quota because they get one free hit. 721 */ 722 est_inflight = dsl_dir_space_towrite(dd); 723 for (i = 0; i < TXG_SIZE; i++) 724 est_inflight += dd->dd_tempreserved[i]; 725 used_on_disk = dd->dd_phys->dd_used_bytes; 726 727 /* 728 * On the first iteration, fetch the dataset's used-on-disk and 729 * refreservation values. Also, if checkrefquota is set, test if 730 * allocating this space would exceed the dataset's refquota. 731 */ 732 if (first && tx->tx_objset) { 733 int error; 734 dsl_dataset_t *ds = tx->tx_objset->os_dsl_dataset; 735 736 error = dsl_dataset_check_quota(ds, checkrefquota, 737 asize, est_inflight, &used_on_disk, &ref_rsrv); 738 if (error) { 739 mutex_exit(&dd->dd_lock); 740 return (error); 741 } 742 } 743 744 /* 745 * If this transaction will result in a net free of space, 746 * we want to let it through. 747 */ 748 if (ignorequota || netfree || dd->dd_phys->dd_quota == 0) 749 quota = UINT64_MAX; 750 else 751 quota = dd->dd_phys->dd_quota; 752 753 /* 754 * Adjust the quota against the actual pool size at the root 755 * minus any outstanding deferred frees. 756 * To ensure that it's possible to remove files from a full 757 * pool without inducing transient overcommits, we throttle 758 * netfree transactions against a quota that is slightly larger, 759 * but still within the pool's allocation slop. In cases where 760 * we're very close to full, this will allow a steady trickle of 761 * removes to get through. 762 */ 763 if (dd->dd_parent == NULL) { 764 spa_t *spa = dd->dd_pool->dp_spa; 765 uint64_t poolsize = dsl_pool_adjustedsize(dd->dd_pool, netfree); 766 deferred = metaslab_class_get_deferred(spa_normal_class(spa)); 767 if (poolsize - deferred < quota) { 768 quota = poolsize - deferred; 769 retval = ENOSPC; 770 } 771 } 772 773 /* 774 * If they are requesting more space, and our current estimate 775 * is over quota, they get to try again unless the actual 776 * on-disk is over quota and there are no pending changes (which 777 * may free up space for us). 778 */ 779 if (used_on_disk + est_inflight >= quota) { 780 if (est_inflight > 0 || used_on_disk < quota || 781 (retval == ENOSPC && used_on_disk < quota + deferred)) 782 retval = ERESTART; 783 dprintf_dd(dd, "failing: used=%lluK inflight = %lluK " 784 "quota=%lluK tr=%lluK err=%d\n", 785 used_on_disk>>10, est_inflight>>10, 786 quota>>10, asize>>10, retval); 787 mutex_exit(&dd->dd_lock); 788 return (retval); 789 } 790 791 /* We need to up our estimated delta before dropping dd_lock */ 792 dd->dd_tempreserved[txgidx] += asize; 793 794 parent_rsrv = parent_delta(dd, used_on_disk + est_inflight, 795 asize - ref_rsrv); 796 mutex_exit(&dd->dd_lock); 797 798 tr = kmem_zalloc(sizeof (struct tempreserve), KM_SLEEP); 799 tr->tr_ds = dd; 800 tr->tr_size = asize; 801 list_insert_tail(tr_list, tr); 802 803 /* see if it's OK with our parent */ 804 if (dd->dd_parent && parent_rsrv) { 805 boolean_t ismos = (dd->dd_phys->dd_head_dataset_obj == 0); 806 807 return (dsl_dir_tempreserve_impl(dd->dd_parent, 808 parent_rsrv, netfree, ismos, TRUE, tr_list, tx, FALSE)); 809 } else { 810 return (0); 811 } 812} 813 814/* 815 * Reserve space in this dsl_dir, to be used in this tx's txg. 816 * After the space has been dirtied (and dsl_dir_willuse_space() 817 * has been called), the reservation should be canceled, using 818 * dsl_dir_tempreserve_clear(). 819 */ 820int 821dsl_dir_tempreserve_space(dsl_dir_t *dd, uint64_t lsize, uint64_t asize, 822 uint64_t fsize, uint64_t usize, void **tr_cookiep, dmu_tx_t *tx) 823{ 824 int err; 825 list_t *tr_list; 826 827 if (asize == 0) { 828 *tr_cookiep = NULL; 829 return (0); 830 } 831 832 tr_list = kmem_alloc(sizeof (list_t), KM_SLEEP); 833 list_create(tr_list, sizeof (struct tempreserve), 834 offsetof(struct tempreserve, tr_node)); 835 ASSERT3S(asize, >, 0); 836 ASSERT3S(fsize, >=, 0); 837 838 err = arc_tempreserve_space(lsize, tx->tx_txg); 839 if (err == 0) { 840 struct tempreserve *tr; 841 842 tr = kmem_zalloc(sizeof (struct tempreserve), KM_SLEEP); 843 tr->tr_size = lsize; 844 list_insert_tail(tr_list, tr); 845 846 err = dsl_pool_tempreserve_space(dd->dd_pool, asize, tx); 847 } else { 848 if (err == EAGAIN) { 849 txg_delay(dd->dd_pool, tx->tx_txg, 1); 850 err = ERESTART; 851 } 852 dsl_pool_memory_pressure(dd->dd_pool); 853 } 854 855 if (err == 0) { 856 struct tempreserve *tr; 857 858 tr = kmem_zalloc(sizeof (struct tempreserve), KM_SLEEP); 859 tr->tr_dp = dd->dd_pool; 860 tr->tr_size = asize; 861 list_insert_tail(tr_list, tr); 862 863 err = dsl_dir_tempreserve_impl(dd, asize, fsize >= asize, 864 FALSE, asize > usize, tr_list, tx, TRUE); 865 } 866 867 if (err) 868 dsl_dir_tempreserve_clear(tr_list, tx); 869 else 870 *tr_cookiep = tr_list; 871 872 return (err); 873} 874 875/* 876 * Clear a temporary reservation that we previously made with 877 * dsl_dir_tempreserve_space(). 878 */ 879void 880dsl_dir_tempreserve_clear(void *tr_cookie, dmu_tx_t *tx) 881{ 882 int txgidx = tx->tx_txg & TXG_MASK; 883 list_t *tr_list = tr_cookie; 884 struct tempreserve *tr; 885 886 ASSERT3U(tx->tx_txg, !=, 0); 887 888 if (tr_cookie == NULL) 889 return; 890 891 while (tr = list_head(tr_list)) { 892 if (tr->tr_dp) { 893 dsl_pool_tempreserve_clear(tr->tr_dp, tr->tr_size, tx); 894 } else if (tr->tr_ds) { 895 mutex_enter(&tr->tr_ds->dd_lock); 896 ASSERT3U(tr->tr_ds->dd_tempreserved[txgidx], >=, 897 tr->tr_size); 898 tr->tr_ds->dd_tempreserved[txgidx] -= tr->tr_size; 899 mutex_exit(&tr->tr_ds->dd_lock); 900 } else { 901 arc_tempreserve_clear(tr->tr_size); 902 } 903 list_remove(tr_list, tr); 904 kmem_free(tr, sizeof (struct tempreserve)); 905 } 906 907 kmem_free(tr_list, sizeof (list_t)); 908} 909 910static void 911dsl_dir_willuse_space_impl(dsl_dir_t *dd, int64_t space, dmu_tx_t *tx) 912{ 913 int64_t parent_space; 914 uint64_t est_used; 915 916 mutex_enter(&dd->dd_lock); 917 if (space > 0) 918 dd->dd_space_towrite[tx->tx_txg & TXG_MASK] += space; 919 920 est_used = dsl_dir_space_towrite(dd) + dd->dd_phys->dd_used_bytes; 921 parent_space = parent_delta(dd, est_used, space); 922 mutex_exit(&dd->dd_lock); 923 924 /* Make sure that we clean up dd_space_to* */ 925 dsl_dir_dirty(dd, tx); 926 927 /* XXX this is potentially expensive and unnecessary... */ 928 if (parent_space && dd->dd_parent) 929 dsl_dir_willuse_space_impl(dd->dd_parent, parent_space, tx); 930} 931 932/* 933 * Call in open context when we think we're going to write/free space, 934 * eg. when dirtying data. Be conservative (ie. OK to write less than 935 * this or free more than this, but don't write more or free less). 936 */ 937void 938dsl_dir_willuse_space(dsl_dir_t *dd, int64_t space, dmu_tx_t *tx) 939{ 940 dsl_pool_willuse_space(dd->dd_pool, space, tx); 941 dsl_dir_willuse_space_impl(dd, space, tx); 942} 943 944/* call from syncing context when we actually write/free space for this dd */ 945void 946dsl_dir_diduse_space(dsl_dir_t *dd, dd_used_t type, 947 int64_t used, int64_t compressed, int64_t uncompressed, dmu_tx_t *tx) 948{ 949 int64_t accounted_delta; 950 boolean_t needlock = !MUTEX_HELD(&dd->dd_lock); 951 952 ASSERT(dmu_tx_is_syncing(tx)); 953 ASSERT(type < DD_USED_NUM); 954 955 if (needlock) 956 mutex_enter(&dd->dd_lock); 957 accounted_delta = parent_delta(dd, dd->dd_phys->dd_used_bytes, used); 958 ASSERT(used >= 0 || dd->dd_phys->dd_used_bytes >= -used); 959 ASSERT(compressed >= 0 || 960 dd->dd_phys->dd_compressed_bytes >= -compressed); 961 ASSERT(uncompressed >= 0 || 962 dd->dd_phys->dd_uncompressed_bytes >= -uncompressed); 963 dmu_buf_will_dirty(dd->dd_dbuf, tx); 964 dd->dd_phys->dd_used_bytes += used; 965 dd->dd_phys->dd_uncompressed_bytes += uncompressed; 966 dd->dd_phys->dd_compressed_bytes += compressed; 967 968 if (dd->dd_phys->dd_flags & DD_FLAG_USED_BREAKDOWN) { 969 ASSERT(used > 0 || 970 dd->dd_phys->dd_used_breakdown[type] >= -used); 971 dd->dd_phys->dd_used_breakdown[type] += used; 972#ifdef DEBUG 973 dd_used_t t; 974 uint64_t u = 0; 975 for (t = 0; t < DD_USED_NUM; t++) 976 u += dd->dd_phys->dd_used_breakdown[t]; 977 ASSERT3U(u, ==, dd->dd_phys->dd_used_bytes); 978#endif 979 } 980 if (needlock) 981 mutex_exit(&dd->dd_lock); 982 983 if (dd->dd_parent != NULL) { 984 dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD, 985 accounted_delta, compressed, uncompressed, tx); 986 dsl_dir_transfer_space(dd->dd_parent, 987 used - accounted_delta, 988 DD_USED_CHILD_RSRV, DD_USED_CHILD, tx); 989 } 990} 991 992void 993dsl_dir_transfer_space(dsl_dir_t *dd, int64_t delta, 994 dd_used_t oldtype, dd_used_t newtype, dmu_tx_t *tx) 995{ 996 boolean_t needlock = !MUTEX_HELD(&dd->dd_lock); 997 998 ASSERT(dmu_tx_is_syncing(tx)); 999 ASSERT(oldtype < DD_USED_NUM); 1000 ASSERT(newtype < DD_USED_NUM); 1001 1002 if (delta == 0 || !(dd->dd_phys->dd_flags & DD_FLAG_USED_BREAKDOWN)) 1003 return; 1004 1005 if (needlock) 1006 mutex_enter(&dd->dd_lock); 1007 ASSERT(delta > 0 ? 1008 dd->dd_phys->dd_used_breakdown[oldtype] >= delta : 1009 dd->dd_phys->dd_used_breakdown[newtype] >= -delta); 1010 ASSERT(dd->dd_phys->dd_used_bytes >= ABS(delta)); 1011 dmu_buf_will_dirty(dd->dd_dbuf, tx); 1012 dd->dd_phys->dd_used_breakdown[oldtype] -= delta; 1013 dd->dd_phys->dd_used_breakdown[newtype] += delta; 1014 if (needlock) 1015 mutex_exit(&dd->dd_lock); 1016} 1017 1018static int 1019dsl_dir_set_quota_check(void *arg1, void *arg2, dmu_tx_t *tx) 1020{ 1021 dsl_dataset_t *ds = arg1; 1022 dsl_dir_t *dd = ds->ds_dir; 1023 dsl_prop_setarg_t *psa = arg2; 1024 int err; 1025 uint64_t towrite; 1026 1027 if ((err = dsl_prop_predict_sync(ds->ds_dir, psa)) != 0) 1028 return (err); 1029 1030 if (psa->psa_effective_value == 0) 1031 return (0); 1032 1033 mutex_enter(&dd->dd_lock); 1034 /* 1035 * If we are doing the preliminary check in open context, and 1036 * there are pending changes, then don't fail it, since the 1037 * pending changes could under-estimate the amount of space to be 1038 * freed up. 1039 */ 1040 towrite = dsl_dir_space_towrite(dd); 1041 if ((dmu_tx_is_syncing(tx) || towrite == 0) && 1042 (psa->psa_effective_value < dd->dd_phys->dd_reserved || 1043 psa->psa_effective_value < dd->dd_phys->dd_used_bytes + towrite)) { 1044 err = ENOSPC; 1045 } 1046 mutex_exit(&dd->dd_lock); 1047 return (err); 1048} 1049 1050extern dsl_syncfunc_t dsl_prop_set_sync; 1051 1052static void 1053dsl_dir_set_quota_sync(void *arg1, void *arg2, dmu_tx_t *tx) 1054{ 1055 dsl_dataset_t *ds = arg1; 1056 dsl_dir_t *dd = ds->ds_dir; 1057 dsl_prop_setarg_t *psa = arg2; 1058 uint64_t effective_value = psa->psa_effective_value; 1059 1060 dsl_prop_set_sync(ds, psa, tx); 1061 DSL_PROP_CHECK_PREDICTION(dd, psa); 1062 1063 dmu_buf_will_dirty(dd->dd_dbuf, tx); 1064 1065 mutex_enter(&dd->dd_lock); 1066 dd->dd_phys->dd_quota = effective_value; 1067 mutex_exit(&dd->dd_lock); 1068} 1069 1070int 1071dsl_dir_set_quota(const char *ddname, zprop_source_t source, uint64_t quota) 1072{ 1073 dsl_dir_t *dd; 1074 dsl_dataset_t *ds; 1075 dsl_prop_setarg_t psa; 1076 int err; 1077 1078 dsl_prop_setarg_init_uint64(&psa, "quota", source, "a); 1079 1080 err = dsl_dataset_hold(ddname, FTAG, &ds); 1081 if (err) 1082 return (err); 1083 1084 err = dsl_dir_open(ddname, FTAG, &dd, NULL); 1085 if (err) { 1086 dsl_dataset_rele(ds, FTAG); 1087 return (err); 1088 } 1089 1090 ASSERT(ds->ds_dir == dd); 1091 1092 /* 1093 * If someone removes a file, then tries to set the quota, we want to 1094 * make sure the file freeing takes effect. 1095 */ 1096 txg_wait_open(dd->dd_pool, 0); 1097 1098 err = dsl_sync_task_do(dd->dd_pool, dsl_dir_set_quota_check, 1099 dsl_dir_set_quota_sync, ds, &psa, 0); 1100 1101 dsl_dir_close(dd, FTAG); 1102 dsl_dataset_rele(ds, FTAG); 1103 return (err); 1104} 1105 1106int 1107dsl_dir_set_reservation_check(void *arg1, void *arg2, dmu_tx_t *tx) 1108{ 1109 dsl_dataset_t *ds = arg1; 1110 dsl_dir_t *dd = ds->ds_dir; 1111 dsl_prop_setarg_t *psa = arg2; 1112 uint64_t effective_value; 1113 uint64_t used, avail; 1114 int err; 1115 1116 if ((err = dsl_prop_predict_sync(ds->ds_dir, psa)) != 0) 1117 return (err); 1118 1119 effective_value = psa->psa_effective_value; 1120 1121 /* 1122 * If we are doing the preliminary check in open context, the 1123 * space estimates may be inaccurate. 1124 */ 1125 if (!dmu_tx_is_syncing(tx)) 1126 return (0); 1127 1128 mutex_enter(&dd->dd_lock); 1129 used = dd->dd_phys->dd_used_bytes; 1130 mutex_exit(&dd->dd_lock); 1131 1132 if (dd->dd_parent) { 1133 avail = dsl_dir_space_available(dd->dd_parent, 1134 NULL, 0, FALSE); 1135 } else { 1136 avail = dsl_pool_adjustedsize(dd->dd_pool, B_FALSE) - used; 1137 } 1138 1139 if (MAX(used, effective_value) > MAX(used, dd->dd_phys->dd_reserved)) { 1140 uint64_t delta = MAX(used, effective_value) - 1141 MAX(used, dd->dd_phys->dd_reserved); 1142 1143 if (delta > avail) 1144 return (ENOSPC); 1145 if (dd->dd_phys->dd_quota > 0 && 1146 effective_value > dd->dd_phys->dd_quota) 1147 return (ENOSPC); 1148 } 1149 1150 return (0); 1151} 1152 1153static void 1154dsl_dir_set_reservation_sync(void *arg1, void *arg2, dmu_tx_t *tx) 1155{ 1156 dsl_dataset_t *ds = arg1; 1157 dsl_dir_t *dd = ds->ds_dir; 1158 dsl_prop_setarg_t *psa = arg2; 1159 uint64_t effective_value = psa->psa_effective_value; 1160 uint64_t used; 1161 int64_t delta; 1162 1163 dsl_prop_set_sync(ds, psa, tx); 1164 DSL_PROP_CHECK_PREDICTION(dd, psa); 1165 1166 dmu_buf_will_dirty(dd->dd_dbuf, tx); 1167 1168 mutex_enter(&dd->dd_lock); 1169 used = dd->dd_phys->dd_used_bytes; 1170 delta = MAX(used, effective_value) - 1171 MAX(used, dd->dd_phys->dd_reserved); 1172 dd->dd_phys->dd_reserved = effective_value; 1173 1174 if (dd->dd_parent != NULL) { 1175 /* Roll up this additional usage into our ancestors */ 1176 dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD_RSRV, 1177 delta, 0, 0, tx); 1178 } 1179 mutex_exit(&dd->dd_lock); 1180} 1181 1182int 1183dsl_dir_set_reservation(const char *ddname, zprop_source_t source, 1184 uint64_t reservation) 1185{ 1186 dsl_dir_t *dd; 1187 dsl_dataset_t *ds; 1188 dsl_prop_setarg_t psa; 1189 int err; 1190 1191 dsl_prop_setarg_init_uint64(&psa, "reservation", source, &reservation); 1192 1193 err = dsl_dataset_hold(ddname, FTAG, &ds); 1194 if (err) 1195 return (err); 1196 1197 err = dsl_dir_open(ddname, FTAG, &dd, NULL); 1198 if (err) { 1199 dsl_dataset_rele(ds, FTAG); 1200 return (err); 1201 } 1202 1203 ASSERT(ds->ds_dir == dd); 1204 1205 err = dsl_sync_task_do(dd->dd_pool, dsl_dir_set_reservation_check, 1206 dsl_dir_set_reservation_sync, ds, &psa, 0); 1207 1208 dsl_dir_close(dd, FTAG); 1209 dsl_dataset_rele(ds, FTAG); 1210 return (err); 1211} 1212 1213static dsl_dir_t * 1214closest_common_ancestor(dsl_dir_t *ds1, dsl_dir_t *ds2) 1215{ 1216 for (; ds1; ds1 = ds1->dd_parent) { 1217 dsl_dir_t *dd; 1218 for (dd = ds2; dd; dd = dd->dd_parent) { 1219 if (ds1 == dd) 1220 return (dd); 1221 } 1222 } 1223 return (NULL); 1224} 1225 1226/* 1227 * If delta is applied to dd, how much of that delta would be applied to 1228 * ancestor? Syncing context only. 1229 */ 1230static int64_t 1231would_change(dsl_dir_t *dd, int64_t delta, dsl_dir_t *ancestor) 1232{ 1233 if (dd == ancestor) 1234 return (delta); 1235 1236 mutex_enter(&dd->dd_lock); 1237 delta = parent_delta(dd, dd->dd_phys->dd_used_bytes, delta); 1238 mutex_exit(&dd->dd_lock); 1239 return (would_change(dd->dd_parent, delta, ancestor)); 1240} 1241 1242struct renamearg { 1243 dsl_dir_t *newparent; 1244 const char *mynewname; 1245 boolean_t allowmounted; 1246}; 1247 1248static int 1249dsl_dir_rename_check(void *arg1, void *arg2, dmu_tx_t *tx) 1250{ 1251 dsl_dir_t *dd = arg1; 1252 struct renamearg *ra = arg2; 1253 dsl_pool_t *dp = dd->dd_pool; 1254 objset_t *mos = dp->dp_meta_objset; 1255 int err; 1256 uint64_t val; 1257 1258 /* 1259 * There should only be one reference, from dmu_objset_rename(). 1260 * Fleeting holds are also possible (eg, from "zfs list" getting 1261 * stats), but any that are present in open context will likely 1262 * be gone by syncing context, so only fail from syncing 1263 * context. 1264 * Don't check if we allow renaming of busy (mounted) dataset. 1265 */ 1266 if (!ra->allowmounted && dmu_tx_is_syncing(tx) && 1267 dmu_buf_refcount(dd->dd_dbuf) > 1) { 1268 return (EBUSY); 1269 } 1270 1271 /* check for existing name */ 1272 err = zap_lookup(mos, ra->newparent->dd_phys->dd_child_dir_zapobj, 1273 ra->mynewname, 8, 1, &val); 1274 if (err == 0) 1275 return (EEXIST); 1276 if (err != ENOENT) 1277 return (err); 1278 1279 if (ra->newparent != dd->dd_parent) { 1280 /* is there enough space? */ 1281 uint64_t myspace = 1282 MAX(dd->dd_phys->dd_used_bytes, dd->dd_phys->dd_reserved); 1283 1284 /* no rename into our descendant */ 1285 if (closest_common_ancestor(dd, ra->newparent) == dd) 1286 return (EINVAL); 1287 1288 if (err = dsl_dir_transfer_possible(dd->dd_parent, 1289 ra->newparent, myspace)) 1290 return (err); 1291 } 1292 1293 return (0); 1294} 1295 1296static void 1297dsl_dir_rename_sync(void *arg1, void *arg2, dmu_tx_t *tx) 1298{ 1299 char oldname[MAXPATHLEN], newname[MAXPATHLEN]; 1300 dsl_dir_t *dd = arg1; 1301 struct renamearg *ra = arg2; 1302 dsl_pool_t *dp = dd->dd_pool; 1303 objset_t *mos = dp->dp_meta_objset; 1304 int err; 1305 1306 ASSERT(ra->allowmounted || dmu_buf_refcount(dd->dd_dbuf) <= 2); 1307 1308 if (ra->newparent != dd->dd_parent) { 1309 dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD, 1310 -dd->dd_phys->dd_used_bytes, 1311 -dd->dd_phys->dd_compressed_bytes, 1312 -dd->dd_phys->dd_uncompressed_bytes, tx); 1313 dsl_dir_diduse_space(ra->newparent, DD_USED_CHILD, 1314 dd->dd_phys->dd_used_bytes, 1315 dd->dd_phys->dd_compressed_bytes, 1316 dd->dd_phys->dd_uncompressed_bytes, tx); 1317 1318 if (dd->dd_phys->dd_reserved > dd->dd_phys->dd_used_bytes) { 1319 uint64_t unused_rsrv = dd->dd_phys->dd_reserved - 1320 dd->dd_phys->dd_used_bytes; 1321 1322 dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD_RSRV, 1323 -unused_rsrv, 0, 0, tx); 1324 dsl_dir_diduse_space(ra->newparent, DD_USED_CHILD_RSRV, 1325 unused_rsrv, 0, 0, tx); 1326 } 1327 } 1328 1329 dmu_buf_will_dirty(dd->dd_dbuf, tx); 1330 1331 /* remove from old parent zapobj */ 1332 dsl_dir_name(dd, oldname); 1333 err = zap_remove(mos, dd->dd_parent->dd_phys->dd_child_dir_zapobj, 1334 dd->dd_myname, tx); 1335 ASSERT0(err); 1336 1337 (void) strcpy(dd->dd_myname, ra->mynewname); 1338 dsl_dir_close(dd->dd_parent, dd); 1339 dd->dd_phys->dd_parent_obj = ra->newparent->dd_object; 1340 VERIFY(0 == dsl_dir_open_obj(dd->dd_pool, 1341 ra->newparent->dd_object, NULL, dd, &dd->dd_parent)); 1342 1343 /* add to new parent zapobj */ 1344 err = zap_add(mos, ra->newparent->dd_phys->dd_child_dir_zapobj, 1345 dd->dd_myname, 8, 1, &dd->dd_object, tx); 1346 ASSERT0(err); 1347 dsl_dir_name(dd, newname); 1348#ifdef _KERNEL 1349 zfsvfs_update_fromname(oldname, newname); 1350 zvol_rename_minors(oldname, newname); 1351#endif 1352 1353 spa_history_log_internal(LOG_DS_RENAME, dd->dd_pool->dp_spa, 1354 tx, "dataset = %llu", dd->dd_phys->dd_head_dataset_obj); 1355} 1356 1357int 1358dsl_dir_rename(dsl_dir_t *dd, const char *newname, int flags) 1359{ 1360 struct renamearg ra; 1361 int err; 1362 1363 /* new parent should exist */ 1364 err = dsl_dir_open(newname, FTAG, &ra.newparent, &ra.mynewname); 1365 if (err) 1366 return (err); 1367 1368 /* can't rename to different pool */ 1369 if (dd->dd_pool != ra.newparent->dd_pool) { 1370 err = ENXIO; 1371 goto out; 1372 } 1373 1374 /* new name should not already exist */ 1375 if (ra.mynewname == NULL) { 1376 err = EEXIST; 1377 goto out; 1378 } 1379 1380 ra.allowmounted = !!(flags & ZFS_RENAME_ALLOW_MOUNTED); 1381 1382 err = dsl_sync_task_do(dd->dd_pool, 1383 dsl_dir_rename_check, dsl_dir_rename_sync, dd, &ra, 3); 1384 1385out: 1386 dsl_dir_close(ra.newparent, FTAG); 1387 return (err); 1388} 1389 1390int 1391dsl_dir_transfer_possible(dsl_dir_t *sdd, dsl_dir_t *tdd, uint64_t space) 1392{ 1393 dsl_dir_t *ancestor; 1394 int64_t adelta; 1395 uint64_t avail; 1396 1397 ancestor = closest_common_ancestor(sdd, tdd); 1398 adelta = would_change(sdd, -space, ancestor); 1399 avail = dsl_dir_space_available(tdd, ancestor, adelta, FALSE); 1400 if (avail < space) 1401 return (ENOSPC); 1402 1403 return (0); 1404} 1405 1406timestruc_t 1407dsl_dir_snap_cmtime(dsl_dir_t *dd) 1408{ 1409 timestruc_t t; 1410 1411 mutex_enter(&dd->dd_lock); 1412 t = dd->dd_snap_cmtime; 1413 mutex_exit(&dd->dd_lock); 1414 1415 return (t); 1416} 1417 1418void 1419dsl_dir_snap_cmtime_update(dsl_dir_t *dd) 1420{ 1421 timestruc_t t; 1422 1423 gethrestime(&t); 1424 mutex_enter(&dd->dd_lock); 1425 dd->dd_snap_cmtime = t; 1426 mutex_exit(&dd->dd_lock); 1427} 1428