dsl_dir.c revision 325534
1/* 2 * CDDL HEADER START 3 * 4 * The contents of this file are subject to the terms of the 5 * Common Development and Distribution License (the "License"). 6 * You may not use this file except in compliance with the License. 7 * 8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 9 * or http://www.opensolaris.org/os/licensing. 10 * See the License for the specific language governing permissions 11 * and limitations under the License. 12 * 13 * When distributing Covered Code, include this CDDL HEADER in each 14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 15 * If applicable, add the following below this CDDL HEADER, with the 16 * fields enclosed by brackets "[]" replaced with your own identifying 17 * information: Portions Copyright [yyyy] [name of copyright owner] 18 * 19 * CDDL HEADER END 20 */ 21/* 22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. 23 * Copyright (c) 2011 Pawel Jakub Dawidek <pawel@dawidek.net>. 24 * All rights reserved. 25 * Copyright (c) 2012, 2016 by Delphix. All rights reserved. 26 * Copyright (c) 2014 Joyent, Inc. All rights reserved. 27 * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved. 28 * Copyright 2015 Nexenta Systems, Inc. All rights reserved. 29 */ 30 31#include <sys/dmu.h> 32#include <sys/dmu_objset.h> 33#include <sys/dmu_tx.h> 34#include <sys/dsl_dataset.h> 35#include <sys/dsl_dir.h> 36#include <sys/dsl_prop.h> 37#include <sys/dsl_synctask.h> 38#include <sys/dsl_deleg.h> 39#include <sys/dmu_impl.h> 40#include <sys/spa.h> 41#include <sys/metaslab.h> 42#include <sys/zap.h> 43#include <sys/zio.h> 44#include <sys/arc.h> 45#include <sys/sunddi.h> 46#include <sys/zvol.h> 47#ifdef _KERNEL 48#include <sys/zfs_vfsops.h> 49#endif 50#include <sys/zfeature.h> 51#include <sys/policy.h> 52#include <sys/zfs_znode.h> 53#include "zfs_namecheck.h" 54#include "zfs_prop.h" 55 56/* 57 * Filesystem and Snapshot Limits 58 * ------------------------------ 59 * 60 * These limits are used to restrict the number of filesystems and/or snapshots 61 * that can be created at a given level in the tree or below. A typical 62 * use-case is with a delegated dataset where the administrator wants to ensure 63 * that a user within the zone is not creating too many additional filesystems 64 * or snapshots, even though they're not exceeding their space quota. 65 * 66 * The filesystem and snapshot counts are stored as extensible properties. This 67 * capability is controlled by a feature flag and must be enabled to be used. 68 * Once enabled, the feature is not active until the first limit is set. At 69 * that point, future operations to create/destroy filesystems or snapshots 70 * will validate and update the counts. 71 * 72 * Because the count properties will not exist before the feature is active, 73 * the counts are updated when a limit is first set on an uninitialized 74 * dsl_dir node in the tree (The filesystem/snapshot count on a node includes 75 * all of the nested filesystems/snapshots. Thus, a new leaf node has a 76 * filesystem count of 0 and a snapshot count of 0. Non-existent filesystem and 77 * snapshot count properties on a node indicate uninitialized counts on that 78 * node.) When first setting a limit on an uninitialized node, the code starts 79 * at the filesystem with the new limit and descends into all sub-filesystems 80 * to add the count properties. 81 * 82 * In practice this is lightweight since a limit is typically set when the 83 * filesystem is created and thus has no children. Once valid, changing the 84 * limit value won't require a re-traversal since the counts are already valid. 85 * When recursively fixing the counts, if a node with a limit is encountered 86 * during the descent, the counts are known to be valid and there is no need to 87 * descend into that filesystem's children. The counts on filesystems above the 88 * one with the new limit will still be uninitialized, unless a limit is 89 * eventually set on one of those filesystems. The counts are always recursively 90 * updated when a limit is set on a dataset, unless there is already a limit. 91 * When a new limit value is set on a filesystem with an existing limit, it is 92 * possible for the new limit to be less than the current count at that level 93 * since a user who can change the limit is also allowed to exceed the limit. 94 * 95 * Once the feature is active, then whenever a filesystem or snapshot is 96 * created, the code recurses up the tree, validating the new count against the 97 * limit at each initialized level. In practice, most levels will not have a 98 * limit set. If there is a limit at any initialized level up the tree, the 99 * check must pass or the creation will fail. Likewise, when a filesystem or 100 * snapshot is destroyed, the counts are recursively adjusted all the way up 101 * the initizized nodes in the tree. Renaming a filesystem into different point 102 * in the tree will first validate, then update the counts on each branch up to 103 * the common ancestor. A receive will also validate the counts and then update 104 * them. 105 * 106 * An exception to the above behavior is that the limit is not enforced if the 107 * user has permission to modify the limit. This is primarily so that 108 * recursive snapshots in the global zone always work. We want to prevent a 109 * denial-of-service in which a lower level delegated dataset could max out its 110 * limit and thus block recursive snapshots from being taken in the global zone. 111 * Because of this, it is possible for the snapshot count to be over the limit 112 * and snapshots taken in the global zone could cause a lower level dataset to 113 * hit or exceed its limit. The administrator taking the global zone recursive 114 * snapshot should be aware of this side-effect and behave accordingly. 115 * For consistency, the filesystem limit is also not enforced if the user can 116 * modify the limit. 117 * 118 * The filesystem and snapshot limits are validated by dsl_fs_ss_limit_check() 119 * and updated by dsl_fs_ss_count_adjust(). A new limit value is setup in 120 * dsl_dir_activate_fs_ss_limit() and the counts are adjusted, if necessary, by 121 * dsl_dir_init_fs_ss_count(). 122 * 123 * There is a special case when we receive a filesystem that already exists. In 124 * this case a temporary clone name of %X is created (see dmu_recv_begin). We 125 * never update the filesystem counts for temporary clones. 126 * 127 * Likewise, we do not update the snapshot counts for temporary snapshots, 128 * such as those created by zfs diff. 129 */ 130 131extern inline dsl_dir_phys_t *dsl_dir_phys(dsl_dir_t *dd); 132 133static uint64_t dsl_dir_space_towrite(dsl_dir_t *dd); 134 135static void 136dsl_dir_evict_async(void *dbu) 137{ 138 dsl_dir_t *dd = dbu; 139 dsl_pool_t *dp = dd->dd_pool; 140 int t; 141 142 dd->dd_dbuf = NULL; 143 144 for (t = 0; t < TXG_SIZE; t++) { 145 ASSERT(!txg_list_member(&dp->dp_dirty_dirs, dd, t)); 146 ASSERT(dd->dd_tempreserved[t] == 0); 147 ASSERT(dd->dd_space_towrite[t] == 0); 148 } 149 150 if (dd->dd_parent) 151 dsl_dir_async_rele(dd->dd_parent, dd); 152 153 spa_async_close(dd->dd_pool->dp_spa, dd); 154 155 dsl_prop_fini(dd); 156 mutex_destroy(&dd->dd_lock); 157 kmem_free(dd, sizeof (dsl_dir_t)); 158} 159 160int 161dsl_dir_hold_obj(dsl_pool_t *dp, uint64_t ddobj, 162 const char *tail, void *tag, dsl_dir_t **ddp) 163{ 164 dmu_buf_t *dbuf; 165 dsl_dir_t *dd; 166 int err; 167 168 ASSERT(dsl_pool_config_held(dp)); 169 170 err = dmu_bonus_hold(dp->dp_meta_objset, ddobj, tag, &dbuf); 171 if (err != 0) 172 return (err); 173 dd = dmu_buf_get_user(dbuf); 174#ifdef ZFS_DEBUG 175 { 176 dmu_object_info_t doi; 177 dmu_object_info_from_db(dbuf, &doi); 178 ASSERT3U(doi.doi_bonus_type, ==, DMU_OT_DSL_DIR); 179 ASSERT3U(doi.doi_bonus_size, >=, sizeof (dsl_dir_phys_t)); 180 } 181#endif 182 if (dd == NULL) { 183 dsl_dir_t *winner; 184 185 dd = kmem_zalloc(sizeof (dsl_dir_t), KM_SLEEP); 186 dd->dd_object = ddobj; 187 dd->dd_dbuf = dbuf; 188 dd->dd_pool = dp; 189 mutex_init(&dd->dd_lock, NULL, MUTEX_DEFAULT, NULL); 190 dsl_prop_init(dd); 191 192 dsl_dir_snap_cmtime_update(dd); 193 194 if (dsl_dir_phys(dd)->dd_parent_obj) { 195 err = dsl_dir_hold_obj(dp, 196 dsl_dir_phys(dd)->dd_parent_obj, NULL, dd, 197 &dd->dd_parent); 198 if (err != 0) 199 goto errout; 200 if (tail) { 201#ifdef ZFS_DEBUG 202 uint64_t foundobj; 203 204 err = zap_lookup(dp->dp_meta_objset, 205 dsl_dir_phys(dd->dd_parent)-> 206 dd_child_dir_zapobj, tail, 207 sizeof (foundobj), 1, &foundobj); 208 ASSERT(err || foundobj == ddobj); 209#endif 210 (void) strcpy(dd->dd_myname, tail); 211 } else { 212 err = zap_value_search(dp->dp_meta_objset, 213 dsl_dir_phys(dd->dd_parent)-> 214 dd_child_dir_zapobj, 215 ddobj, 0, dd->dd_myname); 216 } 217 if (err != 0) 218 goto errout; 219 } else { 220 (void) strcpy(dd->dd_myname, spa_name(dp->dp_spa)); 221 } 222 223 if (dsl_dir_is_clone(dd)) { 224 dmu_buf_t *origin_bonus; 225 dsl_dataset_phys_t *origin_phys; 226 227 /* 228 * We can't open the origin dataset, because 229 * that would require opening this dsl_dir. 230 * Just look at its phys directly instead. 231 */ 232 err = dmu_bonus_hold(dp->dp_meta_objset, 233 dsl_dir_phys(dd)->dd_origin_obj, FTAG, 234 &origin_bonus); 235 if (err != 0) 236 goto errout; 237 origin_phys = origin_bonus->db_data; 238 dd->dd_origin_txg = 239 origin_phys->ds_creation_txg; 240 dmu_buf_rele(origin_bonus, FTAG); 241 } 242 243 dmu_buf_init_user(&dd->dd_dbu, NULL, dsl_dir_evict_async, 244 &dd->dd_dbuf); 245 winner = dmu_buf_set_user_ie(dbuf, &dd->dd_dbu); 246 if (winner != NULL) { 247 if (dd->dd_parent) 248 dsl_dir_rele(dd->dd_parent, dd); 249 dsl_prop_fini(dd); 250 mutex_destroy(&dd->dd_lock); 251 kmem_free(dd, sizeof (dsl_dir_t)); 252 dd = winner; 253 } else { 254 spa_open_ref(dp->dp_spa, dd); 255 } 256 } 257 258 /* 259 * The dsl_dir_t has both open-to-close and instantiate-to-evict 260 * holds on the spa. We need the open-to-close holds because 261 * otherwise the spa_refcnt wouldn't change when we open a 262 * dir which the spa also has open, so we could incorrectly 263 * think it was OK to unload/export/destroy the pool. We need 264 * the instantiate-to-evict hold because the dsl_dir_t has a 265 * pointer to the dd_pool, which has a pointer to the spa_t. 266 */ 267 spa_open_ref(dp->dp_spa, tag); 268 ASSERT3P(dd->dd_pool, ==, dp); 269 ASSERT3U(dd->dd_object, ==, ddobj); 270 ASSERT3P(dd->dd_dbuf, ==, dbuf); 271 *ddp = dd; 272 return (0); 273 274errout: 275 if (dd->dd_parent) 276 dsl_dir_rele(dd->dd_parent, dd); 277 dsl_prop_fini(dd); 278 mutex_destroy(&dd->dd_lock); 279 kmem_free(dd, sizeof (dsl_dir_t)); 280 dmu_buf_rele(dbuf, tag); 281 return (err); 282} 283 284void 285dsl_dir_rele(dsl_dir_t *dd, void *tag) 286{ 287 dprintf_dd(dd, "%s\n", ""); 288 spa_close(dd->dd_pool->dp_spa, tag); 289 dmu_buf_rele(dd->dd_dbuf, tag); 290} 291 292/* 293 * Remove a reference to the given dsl dir that is being asynchronously 294 * released. Async releases occur from a taskq performing eviction of 295 * dsl datasets and dirs. This process is identical to a normal release 296 * with the exception of using the async API for releasing the reference on 297 * the spa. 298 */ 299void 300dsl_dir_async_rele(dsl_dir_t *dd, void *tag) 301{ 302 dprintf_dd(dd, "%s\n", ""); 303 spa_async_close(dd->dd_pool->dp_spa, tag); 304 dmu_buf_rele(dd->dd_dbuf, tag); 305} 306 307/* buf must be at least ZFS_MAX_DATASET_NAME_LEN bytes */ 308void 309dsl_dir_name(dsl_dir_t *dd, char *buf) 310{ 311 if (dd->dd_parent) { 312 dsl_dir_name(dd->dd_parent, buf); 313 VERIFY3U(strlcat(buf, "/", ZFS_MAX_DATASET_NAME_LEN), <, 314 ZFS_MAX_DATASET_NAME_LEN); 315 } else { 316 buf[0] = '\0'; 317 } 318 if (!MUTEX_HELD(&dd->dd_lock)) { 319 /* 320 * recursive mutex so that we can use 321 * dprintf_dd() with dd_lock held 322 */ 323 mutex_enter(&dd->dd_lock); 324 VERIFY3U(strlcat(buf, dd->dd_myname, ZFS_MAX_DATASET_NAME_LEN), 325 <, ZFS_MAX_DATASET_NAME_LEN); 326 mutex_exit(&dd->dd_lock); 327 } else { 328 VERIFY3U(strlcat(buf, dd->dd_myname, ZFS_MAX_DATASET_NAME_LEN), 329 <, ZFS_MAX_DATASET_NAME_LEN); 330 } 331} 332 333/* Calculate name length, avoiding all the strcat calls of dsl_dir_name */ 334int 335dsl_dir_namelen(dsl_dir_t *dd) 336{ 337 int result = 0; 338 339 if (dd->dd_parent) { 340 /* parent's name + 1 for the "/" */ 341 result = dsl_dir_namelen(dd->dd_parent) + 1; 342 } 343 344 if (!MUTEX_HELD(&dd->dd_lock)) { 345 /* see dsl_dir_name */ 346 mutex_enter(&dd->dd_lock); 347 result += strlen(dd->dd_myname); 348 mutex_exit(&dd->dd_lock); 349 } else { 350 result += strlen(dd->dd_myname); 351 } 352 353 return (result); 354} 355 356static int 357getcomponent(const char *path, char *component, const char **nextp) 358{ 359 char *p; 360 361 if ((path == NULL) || (path[0] == '\0')) 362 return (SET_ERROR(ENOENT)); 363 /* This would be a good place to reserve some namespace... */ 364 p = strpbrk(path, "/@"); 365 if (p && (p[1] == '/' || p[1] == '@')) { 366 /* two separators in a row */ 367 return (SET_ERROR(EINVAL)); 368 } 369 if (p == NULL || p == path) { 370 /* 371 * if the first thing is an @ or /, it had better be an 372 * @ and it had better not have any more ats or slashes, 373 * and it had better have something after the @. 374 */ 375 if (p != NULL && 376 (p[0] != '@' || strpbrk(path+1, "/@") || p[1] == '\0')) 377 return (SET_ERROR(EINVAL)); 378 if (strlen(path) >= ZFS_MAX_DATASET_NAME_LEN) 379 return (SET_ERROR(ENAMETOOLONG)); 380 (void) strcpy(component, path); 381 p = NULL; 382 } else if (p[0] == '/') { 383 if (p - path >= ZFS_MAX_DATASET_NAME_LEN) 384 return (SET_ERROR(ENAMETOOLONG)); 385 (void) strncpy(component, path, p - path); 386 component[p - path] = '\0'; 387 p++; 388 } else if (p[0] == '@') { 389 /* 390 * if the next separator is an @, there better not be 391 * any more slashes. 392 */ 393 if (strchr(path, '/')) 394 return (SET_ERROR(EINVAL)); 395 if (p - path >= ZFS_MAX_DATASET_NAME_LEN) 396 return (SET_ERROR(ENAMETOOLONG)); 397 (void) strncpy(component, path, p - path); 398 component[p - path] = '\0'; 399 } else { 400 panic("invalid p=%p", (void *)p); 401 } 402 *nextp = p; 403 return (0); 404} 405 406/* 407 * Return the dsl_dir_t, and possibly the last component which couldn't 408 * be found in *tail. The name must be in the specified dsl_pool_t. This 409 * thread must hold the dp_config_rwlock for the pool. Returns NULL if the 410 * path is bogus, or if tail==NULL and we couldn't parse the whole name. 411 * (*tail)[0] == '@' means that the last component is a snapshot. 412 */ 413int 414dsl_dir_hold(dsl_pool_t *dp, const char *name, void *tag, 415 dsl_dir_t **ddp, const char **tailp) 416{ 417 char buf[ZFS_MAX_DATASET_NAME_LEN]; 418 const char *spaname, *next, *nextnext = NULL; 419 int err; 420 dsl_dir_t *dd; 421 uint64_t ddobj; 422 423 err = getcomponent(name, buf, &next); 424 if (err != 0) 425 return (err); 426 427 /* Make sure the name is in the specified pool. */ 428 spaname = spa_name(dp->dp_spa); 429 if (strcmp(buf, spaname) != 0) 430 return (SET_ERROR(EXDEV)); 431 432 ASSERT(dsl_pool_config_held(dp)); 433 434 err = dsl_dir_hold_obj(dp, dp->dp_root_dir_obj, NULL, tag, &dd); 435 if (err != 0) { 436 return (err); 437 } 438 439 while (next != NULL) { 440 dsl_dir_t *child_dd; 441 err = getcomponent(next, buf, &nextnext); 442 if (err != 0) 443 break; 444 ASSERT(next[0] != '\0'); 445 if (next[0] == '@') 446 break; 447 dprintf("looking up %s in obj%lld\n", 448 buf, dsl_dir_phys(dd)->dd_child_dir_zapobj); 449 450 err = zap_lookup(dp->dp_meta_objset, 451 dsl_dir_phys(dd)->dd_child_dir_zapobj, 452 buf, sizeof (ddobj), 1, &ddobj); 453 if (err != 0) { 454 if (err == ENOENT) 455 err = 0; 456 break; 457 } 458 459 err = dsl_dir_hold_obj(dp, ddobj, buf, tag, &child_dd); 460 if (err != 0) 461 break; 462 dsl_dir_rele(dd, tag); 463 dd = child_dd; 464 next = nextnext; 465 } 466 467 if (err != 0) { 468 dsl_dir_rele(dd, tag); 469 return (err); 470 } 471 472 /* 473 * It's an error if there's more than one component left, or 474 * tailp==NULL and there's any component left. 475 */ 476 if (next != NULL && 477 (tailp == NULL || (nextnext && nextnext[0] != '\0'))) { 478 /* bad path name */ 479 dsl_dir_rele(dd, tag); 480 dprintf("next=%p (%s) tail=%p\n", next, next?next:"", tailp); 481 err = SET_ERROR(ENOENT); 482 } 483 if (tailp != NULL) 484 *tailp = next; 485 *ddp = dd; 486 return (err); 487} 488 489/* 490 * If the counts are already initialized for this filesystem and its 491 * descendants then do nothing, otherwise initialize the counts. 492 * 493 * The counts on this filesystem, and those below, may be uninitialized due to 494 * either the use of a pre-existing pool which did not support the 495 * filesystem/snapshot limit feature, or one in which the feature had not yet 496 * been enabled. 497 * 498 * Recursively descend the filesystem tree and update the filesystem/snapshot 499 * counts on each filesystem below, then update the cumulative count on the 500 * current filesystem. If the filesystem already has a count set on it, 501 * then we know that its counts, and the counts on the filesystems below it, 502 * are already correct, so we don't have to update this filesystem. 503 */ 504static void 505dsl_dir_init_fs_ss_count(dsl_dir_t *dd, dmu_tx_t *tx) 506{ 507 uint64_t my_fs_cnt = 0; 508 uint64_t my_ss_cnt = 0; 509 dsl_pool_t *dp = dd->dd_pool; 510 objset_t *os = dp->dp_meta_objset; 511 zap_cursor_t *zc; 512 zap_attribute_t *za; 513 dsl_dataset_t *ds; 514 515 ASSERT(spa_feature_is_active(dp->dp_spa, SPA_FEATURE_FS_SS_LIMIT)); 516 ASSERT(dsl_pool_config_held(dp)); 517 ASSERT(dmu_tx_is_syncing(tx)); 518 519 dsl_dir_zapify(dd, tx); 520 521 /* 522 * If the filesystem count has already been initialized then we 523 * don't need to recurse down any further. 524 */ 525 if (zap_contains(os, dd->dd_object, DD_FIELD_FILESYSTEM_COUNT) == 0) 526 return; 527 528 zc = kmem_alloc(sizeof (zap_cursor_t), KM_SLEEP); 529 za = kmem_alloc(sizeof (zap_attribute_t), KM_SLEEP); 530 531 /* Iterate my child dirs */ 532 for (zap_cursor_init(zc, os, dsl_dir_phys(dd)->dd_child_dir_zapobj); 533 zap_cursor_retrieve(zc, za) == 0; zap_cursor_advance(zc)) { 534 dsl_dir_t *chld_dd; 535 uint64_t count; 536 537 VERIFY0(dsl_dir_hold_obj(dp, za->za_first_integer, NULL, FTAG, 538 &chld_dd)); 539 540 /* 541 * Ignore hidden ($FREE, $MOS & $ORIGIN) objsets and 542 * temporary datasets. 543 */ 544 if (chld_dd->dd_myname[0] == '$' || 545 chld_dd->dd_myname[0] == '%') { 546 dsl_dir_rele(chld_dd, FTAG); 547 continue; 548 } 549 550 my_fs_cnt++; /* count this child */ 551 552 dsl_dir_init_fs_ss_count(chld_dd, tx); 553 554 VERIFY0(zap_lookup(os, chld_dd->dd_object, 555 DD_FIELD_FILESYSTEM_COUNT, sizeof (count), 1, &count)); 556 my_fs_cnt += count; 557 VERIFY0(zap_lookup(os, chld_dd->dd_object, 558 DD_FIELD_SNAPSHOT_COUNT, sizeof (count), 1, &count)); 559 my_ss_cnt += count; 560 561 dsl_dir_rele(chld_dd, FTAG); 562 } 563 zap_cursor_fini(zc); 564 /* Count my snapshots (we counted children's snapshots above) */ 565 VERIFY0(dsl_dataset_hold_obj(dd->dd_pool, 566 dsl_dir_phys(dd)->dd_head_dataset_obj, FTAG, &ds)); 567 568 for (zap_cursor_init(zc, os, dsl_dataset_phys(ds)->ds_snapnames_zapobj); 569 zap_cursor_retrieve(zc, za) == 0; 570 zap_cursor_advance(zc)) { 571 /* Don't count temporary snapshots */ 572 if (za->za_name[0] != '%') 573 my_ss_cnt++; 574 } 575 zap_cursor_fini(zc); 576 577 dsl_dataset_rele(ds, FTAG); 578 579 kmem_free(zc, sizeof (zap_cursor_t)); 580 kmem_free(za, sizeof (zap_attribute_t)); 581 582 /* we're in a sync task, update counts */ 583 dmu_buf_will_dirty(dd->dd_dbuf, tx); 584 VERIFY0(zap_add(os, dd->dd_object, DD_FIELD_FILESYSTEM_COUNT, 585 sizeof (my_fs_cnt), 1, &my_fs_cnt, tx)); 586 VERIFY0(zap_add(os, dd->dd_object, DD_FIELD_SNAPSHOT_COUNT, 587 sizeof (my_ss_cnt), 1, &my_ss_cnt, tx)); 588} 589 590static int 591dsl_dir_actv_fs_ss_limit_check(void *arg, dmu_tx_t *tx) 592{ 593 char *ddname = (char *)arg; 594 dsl_pool_t *dp = dmu_tx_pool(tx); 595 dsl_dataset_t *ds; 596 dsl_dir_t *dd; 597 int error; 598 599 error = dsl_dataset_hold(dp, ddname, FTAG, &ds); 600 if (error != 0) 601 return (error); 602 603 if (!spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_FS_SS_LIMIT)) { 604 dsl_dataset_rele(ds, FTAG); 605 return (SET_ERROR(ENOTSUP)); 606 } 607 608 dd = ds->ds_dir; 609 if (spa_feature_is_active(dp->dp_spa, SPA_FEATURE_FS_SS_LIMIT) && 610 dsl_dir_is_zapified(dd) && 611 zap_contains(dp->dp_meta_objset, dd->dd_object, 612 DD_FIELD_FILESYSTEM_COUNT) == 0) { 613 dsl_dataset_rele(ds, FTAG); 614 return (SET_ERROR(EALREADY)); 615 } 616 617 dsl_dataset_rele(ds, FTAG); 618 return (0); 619} 620 621static void 622dsl_dir_actv_fs_ss_limit_sync(void *arg, dmu_tx_t *tx) 623{ 624 char *ddname = (char *)arg; 625 dsl_pool_t *dp = dmu_tx_pool(tx); 626 dsl_dataset_t *ds; 627 spa_t *spa; 628 629 VERIFY0(dsl_dataset_hold(dp, ddname, FTAG, &ds)); 630 631 spa = dsl_dataset_get_spa(ds); 632 633 if (!spa_feature_is_active(spa, SPA_FEATURE_FS_SS_LIMIT)) { 634 /* 635 * Since the feature was not active and we're now setting a 636 * limit, increment the feature-active counter so that the 637 * feature becomes active for the first time. 638 * 639 * We are already in a sync task so we can update the MOS. 640 */ 641 spa_feature_incr(spa, SPA_FEATURE_FS_SS_LIMIT, tx); 642 } 643 644 /* 645 * Since we are now setting a non-UINT64_MAX limit on the filesystem, 646 * we need to ensure the counts are correct. Descend down the tree from 647 * this point and update all of the counts to be accurate. 648 */ 649 dsl_dir_init_fs_ss_count(ds->ds_dir, tx); 650 651 dsl_dataset_rele(ds, FTAG); 652} 653 654/* 655 * Make sure the feature is enabled and activate it if necessary. 656 * Since we're setting a limit, ensure the on-disk counts are valid. 657 * This is only called by the ioctl path when setting a limit value. 658 * 659 * We do not need to validate the new limit, since users who can change the 660 * limit are also allowed to exceed the limit. 661 */ 662int 663dsl_dir_activate_fs_ss_limit(const char *ddname) 664{ 665 int error; 666 667 error = dsl_sync_task(ddname, dsl_dir_actv_fs_ss_limit_check, 668 dsl_dir_actv_fs_ss_limit_sync, (void *)ddname, 0, 669 ZFS_SPACE_CHECK_RESERVED); 670 671 if (error == EALREADY) 672 error = 0; 673 674 return (error); 675} 676 677/* 678 * Used to determine if the filesystem_limit or snapshot_limit should be 679 * enforced. We allow the limit to be exceeded if the user has permission to 680 * write the property value. We pass in the creds that we got in the open 681 * context since we will always be the GZ root in syncing context. We also have 682 * to handle the case where we are allowed to change the limit on the current 683 * dataset, but there may be another limit in the tree above. 684 * 685 * We can never modify these two properties within a non-global zone. In 686 * addition, the other checks are modeled on zfs_secpolicy_write_perms. We 687 * can't use that function since we are already holding the dp_config_rwlock. 688 * In addition, we already have the dd and dealing with snapshots is simplified 689 * in this code. 690 */ 691 692typedef enum { 693 ENFORCE_ALWAYS, 694 ENFORCE_NEVER, 695 ENFORCE_ABOVE 696} enforce_res_t; 697 698static enforce_res_t 699dsl_enforce_ds_ss_limits(dsl_dir_t *dd, zfs_prop_t prop, cred_t *cr) 700{ 701 enforce_res_t enforce = ENFORCE_ALWAYS; 702 uint64_t obj; 703 dsl_dataset_t *ds; 704 uint64_t zoned; 705 706 ASSERT(prop == ZFS_PROP_FILESYSTEM_LIMIT || 707 prop == ZFS_PROP_SNAPSHOT_LIMIT); 708 709#ifdef _KERNEL 710#ifdef __FreeBSD__ 711 if (jailed(cr)) 712#else 713 if (crgetzoneid(cr) != GLOBAL_ZONEID) 714#endif 715 return (ENFORCE_ALWAYS); 716 717 if (secpolicy_zfs(cr) == 0) 718 return (ENFORCE_NEVER); 719#endif 720 721 if ((obj = dsl_dir_phys(dd)->dd_head_dataset_obj) == 0) 722 return (ENFORCE_ALWAYS); 723 724 ASSERT(dsl_pool_config_held(dd->dd_pool)); 725 726 if (dsl_dataset_hold_obj(dd->dd_pool, obj, FTAG, &ds) != 0) 727 return (ENFORCE_ALWAYS); 728 729 if (dsl_prop_get_ds(ds, "zoned", 8, 1, &zoned, NULL) || zoned) { 730 /* Only root can access zoned fs's from the GZ */ 731 enforce = ENFORCE_ALWAYS; 732 } else { 733 if (dsl_deleg_access_impl(ds, zfs_prop_to_name(prop), cr) == 0) 734 enforce = ENFORCE_ABOVE; 735 } 736 737 dsl_dataset_rele(ds, FTAG); 738 return (enforce); 739} 740 741/* 742 * Check if adding additional child filesystem(s) would exceed any filesystem 743 * limits or adding additional snapshot(s) would exceed any snapshot limits. 744 * The prop argument indicates which limit to check. 745 * 746 * Note that all filesystem limits up to the root (or the highest 747 * initialized) filesystem or the given ancestor must be satisfied. 748 */ 749int 750dsl_fs_ss_limit_check(dsl_dir_t *dd, uint64_t delta, zfs_prop_t prop, 751 dsl_dir_t *ancestor, cred_t *cr) 752{ 753 objset_t *os = dd->dd_pool->dp_meta_objset; 754 uint64_t limit, count; 755 char *count_prop; 756 enforce_res_t enforce; 757 int err = 0; 758 759 ASSERT(dsl_pool_config_held(dd->dd_pool)); 760 ASSERT(prop == ZFS_PROP_FILESYSTEM_LIMIT || 761 prop == ZFS_PROP_SNAPSHOT_LIMIT); 762 763 /* 764 * If we're allowed to change the limit, don't enforce the limit 765 * e.g. this can happen if a snapshot is taken by an administrative 766 * user in the global zone (i.e. a recursive snapshot by root). 767 * However, we must handle the case of delegated permissions where we 768 * are allowed to change the limit on the current dataset, but there 769 * is another limit in the tree above. 770 */ 771 enforce = dsl_enforce_ds_ss_limits(dd, prop, cr); 772 if (enforce == ENFORCE_NEVER) 773 return (0); 774 775 /* 776 * e.g. if renaming a dataset with no snapshots, count adjustment 777 * is 0. 778 */ 779 if (delta == 0) 780 return (0); 781 782 if (prop == ZFS_PROP_SNAPSHOT_LIMIT) { 783 /* 784 * We don't enforce the limit for temporary snapshots. This is 785 * indicated by a NULL cred_t argument. 786 */ 787 if (cr == NULL) 788 return (0); 789 790 count_prop = DD_FIELD_SNAPSHOT_COUNT; 791 } else { 792 count_prop = DD_FIELD_FILESYSTEM_COUNT; 793 } 794 795 /* 796 * If an ancestor has been provided, stop checking the limit once we 797 * hit that dir. We need this during rename so that we don't overcount 798 * the check once we recurse up to the common ancestor. 799 */ 800 if (ancestor == dd) 801 return (0); 802 803 /* 804 * If we hit an uninitialized node while recursing up the tree, we can 805 * stop since we know there is no limit here (or above). The counts are 806 * not valid on this node and we know we won't touch this node's counts. 807 */ 808 if (!dsl_dir_is_zapified(dd) || zap_lookup(os, dd->dd_object, 809 count_prop, sizeof (count), 1, &count) == ENOENT) 810 return (0); 811 812 err = dsl_prop_get_dd(dd, zfs_prop_to_name(prop), 8, 1, &limit, NULL, 813 B_FALSE); 814 if (err != 0) 815 return (err); 816 817 /* Is there a limit which we've hit? */ 818 if (enforce == ENFORCE_ALWAYS && (count + delta) > limit) 819 return (SET_ERROR(EDQUOT)); 820 821 if (dd->dd_parent != NULL) 822 err = dsl_fs_ss_limit_check(dd->dd_parent, delta, prop, 823 ancestor, cr); 824 825 return (err); 826} 827 828/* 829 * Adjust the filesystem or snapshot count for the specified dsl_dir_t and all 830 * parents. When a new filesystem/snapshot is created, increment the count on 831 * all parents, and when a filesystem/snapshot is destroyed, decrement the 832 * count. 833 */ 834void 835dsl_fs_ss_count_adjust(dsl_dir_t *dd, int64_t delta, const char *prop, 836 dmu_tx_t *tx) 837{ 838 int err; 839 objset_t *os = dd->dd_pool->dp_meta_objset; 840 uint64_t count; 841 842 ASSERT(dsl_pool_config_held(dd->dd_pool)); 843 ASSERT(dmu_tx_is_syncing(tx)); 844 ASSERT(strcmp(prop, DD_FIELD_FILESYSTEM_COUNT) == 0 || 845 strcmp(prop, DD_FIELD_SNAPSHOT_COUNT) == 0); 846 847 /* 848 * When we receive an incremental stream into a filesystem that already 849 * exists, a temporary clone is created. We don't count this temporary 850 * clone, whose name begins with a '%'. We also ignore hidden ($FREE, 851 * $MOS & $ORIGIN) objsets. 852 */ 853 if ((dd->dd_myname[0] == '%' || dd->dd_myname[0] == '$') && 854 strcmp(prop, DD_FIELD_FILESYSTEM_COUNT) == 0) 855 return; 856 857 /* 858 * e.g. if renaming a dataset with no snapshots, count adjustment is 0 859 */ 860 if (delta == 0) 861 return; 862 863 /* 864 * If we hit an uninitialized node while recursing up the tree, we can 865 * stop since we know the counts are not valid on this node and we 866 * know we shouldn't touch this node's counts. An uninitialized count 867 * on the node indicates that either the feature has not yet been 868 * activated or there are no limits on this part of the tree. 869 */ 870 if (!dsl_dir_is_zapified(dd) || (err = zap_lookup(os, dd->dd_object, 871 prop, sizeof (count), 1, &count)) == ENOENT) 872 return; 873 VERIFY0(err); 874 875 count += delta; 876 /* Use a signed verify to make sure we're not neg. */ 877 VERIFY3S(count, >=, 0); 878 879 VERIFY0(zap_update(os, dd->dd_object, prop, sizeof (count), 1, &count, 880 tx)); 881 882 /* Roll up this additional count into our ancestors */ 883 if (dd->dd_parent != NULL) 884 dsl_fs_ss_count_adjust(dd->dd_parent, delta, prop, tx); 885} 886 887uint64_t 888dsl_dir_create_sync(dsl_pool_t *dp, dsl_dir_t *pds, const char *name, 889 dmu_tx_t *tx) 890{ 891 objset_t *mos = dp->dp_meta_objset; 892 uint64_t ddobj; 893 dsl_dir_phys_t *ddphys; 894 dmu_buf_t *dbuf; 895 896 ddobj = dmu_object_alloc(mos, DMU_OT_DSL_DIR, 0, 897 DMU_OT_DSL_DIR, sizeof (dsl_dir_phys_t), tx); 898 if (pds) { 899 VERIFY(0 == zap_add(mos, dsl_dir_phys(pds)->dd_child_dir_zapobj, 900 name, sizeof (uint64_t), 1, &ddobj, tx)); 901 } else { 902 /* it's the root dir */ 903 VERIFY(0 == zap_add(mos, DMU_POOL_DIRECTORY_OBJECT, 904 DMU_POOL_ROOT_DATASET, sizeof (uint64_t), 1, &ddobj, tx)); 905 } 906 VERIFY(0 == dmu_bonus_hold(mos, ddobj, FTAG, &dbuf)); 907 dmu_buf_will_dirty(dbuf, tx); 908 ddphys = dbuf->db_data; 909 910 ddphys->dd_creation_time = gethrestime_sec(); 911 if (pds) { 912 ddphys->dd_parent_obj = pds->dd_object; 913 914 /* update the filesystem counts */ 915 dsl_fs_ss_count_adjust(pds, 1, DD_FIELD_FILESYSTEM_COUNT, tx); 916 } 917 ddphys->dd_props_zapobj = zap_create(mos, 918 DMU_OT_DSL_PROPS, DMU_OT_NONE, 0, tx); 919 ddphys->dd_child_dir_zapobj = zap_create(mos, 920 DMU_OT_DSL_DIR_CHILD_MAP, DMU_OT_NONE, 0, tx); 921 if (spa_version(dp->dp_spa) >= SPA_VERSION_USED_BREAKDOWN) 922 ddphys->dd_flags |= DD_FLAG_USED_BREAKDOWN; 923 dmu_buf_rele(dbuf, FTAG); 924 925 return (ddobj); 926} 927 928boolean_t 929dsl_dir_is_clone(dsl_dir_t *dd) 930{ 931 return (dsl_dir_phys(dd)->dd_origin_obj && 932 (dd->dd_pool->dp_origin_snap == NULL || 933 dsl_dir_phys(dd)->dd_origin_obj != 934 dd->dd_pool->dp_origin_snap->ds_object)); 935} 936 937 938uint64_t 939dsl_dir_get_used(dsl_dir_t *dd) 940{ 941 return (dsl_dir_phys(dd)->dd_used_bytes); 942} 943 944uint64_t 945dsl_dir_get_quota(dsl_dir_t *dd) 946{ 947 return (dsl_dir_phys(dd)->dd_quota); 948} 949 950uint64_t 951dsl_dir_get_reservation(dsl_dir_t *dd) 952{ 953 return (dsl_dir_phys(dd)->dd_reserved); 954} 955 956uint64_t 957dsl_dir_get_compressratio(dsl_dir_t *dd) 958{ 959 /* a fixed point number, 100x the ratio */ 960 return (dsl_dir_phys(dd)->dd_compressed_bytes == 0 ? 100 : 961 (dsl_dir_phys(dd)->dd_uncompressed_bytes * 100 / 962 dsl_dir_phys(dd)->dd_compressed_bytes)); 963} 964 965uint64_t 966dsl_dir_get_logicalused(dsl_dir_t *dd) 967{ 968 return (dsl_dir_phys(dd)->dd_uncompressed_bytes); 969} 970 971uint64_t 972dsl_dir_get_usedsnap(dsl_dir_t *dd) 973{ 974 return (dsl_dir_phys(dd)->dd_used_breakdown[DD_USED_SNAP]); 975} 976 977uint64_t 978dsl_dir_get_usedds(dsl_dir_t *dd) 979{ 980 return (dsl_dir_phys(dd)->dd_used_breakdown[DD_USED_HEAD]); 981} 982 983uint64_t 984dsl_dir_get_usedrefreserv(dsl_dir_t *dd) 985{ 986 return (dsl_dir_phys(dd)->dd_used_breakdown[DD_USED_REFRSRV]); 987} 988 989uint64_t 990dsl_dir_get_usedchild(dsl_dir_t *dd) 991{ 992 return (dsl_dir_phys(dd)->dd_used_breakdown[DD_USED_CHILD] + 993 dsl_dir_phys(dd)->dd_used_breakdown[DD_USED_CHILD_RSRV]); 994} 995 996void 997dsl_dir_get_origin(dsl_dir_t *dd, char *buf) 998{ 999 dsl_dataset_t *ds; 1000 VERIFY0(dsl_dataset_hold_obj(dd->dd_pool, 1001 dsl_dir_phys(dd)->dd_origin_obj, FTAG, &ds)); 1002 1003 dsl_dataset_name(ds, buf); 1004 1005 dsl_dataset_rele(ds, FTAG); 1006} 1007 1008int 1009dsl_dir_get_filesystem_count(dsl_dir_t *dd, uint64_t *count) 1010{ 1011 if (dsl_dir_is_zapified(dd)) { 1012 objset_t *os = dd->dd_pool->dp_meta_objset; 1013 return (zap_lookup(os, dd->dd_object, DD_FIELD_FILESYSTEM_COUNT, 1014 sizeof (*count), 1, count)); 1015 } else { 1016 return (ENOENT); 1017 } 1018} 1019 1020int 1021dsl_dir_get_snapshot_count(dsl_dir_t *dd, uint64_t *count) 1022{ 1023 if (dsl_dir_is_zapified(dd)) { 1024 objset_t *os = dd->dd_pool->dp_meta_objset; 1025 return (zap_lookup(os, dd->dd_object, DD_FIELD_SNAPSHOT_COUNT, 1026 sizeof (*count), 1, count)); 1027 } else { 1028 return (ENOENT); 1029 } 1030} 1031 1032void 1033dsl_dir_stats(dsl_dir_t *dd, nvlist_t *nv) 1034{ 1035 mutex_enter(&dd->dd_lock); 1036 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_QUOTA, 1037 dsl_dir_get_quota(dd)); 1038 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_RESERVATION, 1039 dsl_dir_get_reservation(dd)); 1040 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_LOGICALUSED, 1041 dsl_dir_get_logicalused(dd)); 1042 if (dsl_dir_phys(dd)->dd_flags & DD_FLAG_USED_BREAKDOWN) { 1043 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDSNAP, 1044 dsl_dir_get_usedsnap(dd)); 1045 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDDS, 1046 dsl_dir_get_usedds(dd)); 1047 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDREFRESERV, 1048 dsl_dir_get_usedrefreserv(dd)); 1049 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDCHILD, 1050 dsl_dir_get_usedchild(dd)); 1051 } 1052 mutex_exit(&dd->dd_lock); 1053 1054 uint64_t count; 1055 if (dsl_dir_get_filesystem_count(dd, &count) == 0) { 1056 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_FILESYSTEM_COUNT, 1057 count); 1058 } 1059 if (dsl_dir_get_snapshot_count(dd, &count) == 0) { 1060 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_SNAPSHOT_COUNT, 1061 count); 1062 } 1063 1064 if (dsl_dir_is_clone(dd)) { 1065 char buf[ZFS_MAX_DATASET_NAME_LEN]; 1066 dsl_dir_get_origin(dd, buf); 1067 dsl_prop_nvlist_add_string(nv, ZFS_PROP_ORIGIN, buf); 1068 } 1069 1070} 1071 1072void 1073dsl_dir_dirty(dsl_dir_t *dd, dmu_tx_t *tx) 1074{ 1075 dsl_pool_t *dp = dd->dd_pool; 1076 1077 ASSERT(dsl_dir_phys(dd)); 1078 1079 if (txg_list_add(&dp->dp_dirty_dirs, dd, tx->tx_txg)) { 1080 /* up the hold count until we can be written out */ 1081 dmu_buf_add_ref(dd->dd_dbuf, dd); 1082 } 1083} 1084 1085static int64_t 1086parent_delta(dsl_dir_t *dd, uint64_t used, int64_t delta) 1087{ 1088 uint64_t old_accounted = MAX(used, dsl_dir_phys(dd)->dd_reserved); 1089 uint64_t new_accounted = 1090 MAX(used + delta, dsl_dir_phys(dd)->dd_reserved); 1091 return (new_accounted - old_accounted); 1092} 1093 1094void 1095dsl_dir_sync(dsl_dir_t *dd, dmu_tx_t *tx) 1096{ 1097 ASSERT(dmu_tx_is_syncing(tx)); 1098 1099 mutex_enter(&dd->dd_lock); 1100 ASSERT0(dd->dd_tempreserved[tx->tx_txg&TXG_MASK]); 1101 dprintf_dd(dd, "txg=%llu towrite=%lluK\n", tx->tx_txg, 1102 dd->dd_space_towrite[tx->tx_txg&TXG_MASK] / 1024); 1103 dd->dd_space_towrite[tx->tx_txg&TXG_MASK] = 0; 1104 mutex_exit(&dd->dd_lock); 1105 1106 /* release the hold from dsl_dir_dirty */ 1107 dmu_buf_rele(dd->dd_dbuf, dd); 1108} 1109 1110static uint64_t 1111dsl_dir_space_towrite(dsl_dir_t *dd) 1112{ 1113 uint64_t space = 0; 1114 1115 ASSERT(MUTEX_HELD(&dd->dd_lock)); 1116 1117 for (int i = 0; i < TXG_SIZE; i++) { 1118 space += dd->dd_space_towrite[i & TXG_MASK]; 1119 ASSERT3U(dd->dd_space_towrite[i & TXG_MASK], >=, 0); 1120 } 1121 return (space); 1122} 1123 1124/* 1125 * How much space would dd have available if ancestor had delta applied 1126 * to it? If ondiskonly is set, we're only interested in what's 1127 * on-disk, not estimated pending changes. 1128 */ 1129uint64_t 1130dsl_dir_space_available(dsl_dir_t *dd, 1131 dsl_dir_t *ancestor, int64_t delta, int ondiskonly) 1132{ 1133 uint64_t parentspace, myspace, quota, used; 1134 1135 /* 1136 * If there are no restrictions otherwise, assume we have 1137 * unlimited space available. 1138 */ 1139 quota = UINT64_MAX; 1140 parentspace = UINT64_MAX; 1141 1142 if (dd->dd_parent != NULL) { 1143 parentspace = dsl_dir_space_available(dd->dd_parent, 1144 ancestor, delta, ondiskonly); 1145 } 1146 1147 mutex_enter(&dd->dd_lock); 1148 if (dsl_dir_phys(dd)->dd_quota != 0) 1149 quota = dsl_dir_phys(dd)->dd_quota; 1150 used = dsl_dir_phys(dd)->dd_used_bytes; 1151 if (!ondiskonly) 1152 used += dsl_dir_space_towrite(dd); 1153 1154 if (dd->dd_parent == NULL) { 1155 uint64_t poolsize = dsl_pool_adjustedsize(dd->dd_pool, FALSE); 1156 quota = MIN(quota, poolsize); 1157 } 1158 1159 if (dsl_dir_phys(dd)->dd_reserved > used && parentspace != UINT64_MAX) { 1160 /* 1161 * We have some space reserved, in addition to what our 1162 * parent gave us. 1163 */ 1164 parentspace += dsl_dir_phys(dd)->dd_reserved - used; 1165 } 1166 1167 if (dd == ancestor) { 1168 ASSERT(delta <= 0); 1169 ASSERT(used >= -delta); 1170 used += delta; 1171 if (parentspace != UINT64_MAX) 1172 parentspace -= delta; 1173 } 1174 1175 if (used > quota) { 1176 /* over quota */ 1177 myspace = 0; 1178 } else { 1179 /* 1180 * the lesser of the space provided by our parent and 1181 * the space left in our quota 1182 */ 1183 myspace = MIN(parentspace, quota - used); 1184 } 1185 1186 mutex_exit(&dd->dd_lock); 1187 1188 return (myspace); 1189} 1190 1191struct tempreserve { 1192 list_node_t tr_node; 1193 dsl_dir_t *tr_ds; 1194 uint64_t tr_size; 1195}; 1196 1197static int 1198dsl_dir_tempreserve_impl(dsl_dir_t *dd, uint64_t asize, boolean_t netfree, 1199 boolean_t ignorequota, list_t *tr_list, 1200 dmu_tx_t *tx, boolean_t first) 1201{ 1202 uint64_t txg = tx->tx_txg; 1203 uint64_t quota; 1204 struct tempreserve *tr; 1205 int retval = EDQUOT; 1206 uint64_t ref_rsrv = 0; 1207 1208 ASSERT3U(txg, !=, 0); 1209 ASSERT3S(asize, >, 0); 1210 1211 mutex_enter(&dd->dd_lock); 1212 1213 /* 1214 * Check against the dsl_dir's quota. We don't add in the delta 1215 * when checking for over-quota because they get one free hit. 1216 */ 1217 uint64_t est_inflight = dsl_dir_space_towrite(dd); 1218 for (int i = 0; i < TXG_SIZE; i++) 1219 est_inflight += dd->dd_tempreserved[i]; 1220 uint64_t used_on_disk = dsl_dir_phys(dd)->dd_used_bytes; 1221 1222 /* 1223 * On the first iteration, fetch the dataset's used-on-disk and 1224 * refreservation values. Also, if checkrefquota is set, test if 1225 * allocating this space would exceed the dataset's refquota. 1226 */ 1227 if (first && tx->tx_objset) { 1228 int error; 1229 dsl_dataset_t *ds = tx->tx_objset->os_dsl_dataset; 1230 1231 error = dsl_dataset_check_quota(ds, !netfree, 1232 asize, est_inflight, &used_on_disk, &ref_rsrv); 1233 if (error != 0) { 1234 mutex_exit(&dd->dd_lock); 1235 return (error); 1236 } 1237 } 1238 1239 /* 1240 * If this transaction will result in a net free of space, 1241 * we want to let it through. 1242 */ 1243 if (ignorequota || netfree || dsl_dir_phys(dd)->dd_quota == 0) 1244 quota = UINT64_MAX; 1245 else 1246 quota = dsl_dir_phys(dd)->dd_quota; 1247 1248 /* 1249 * Adjust the quota against the actual pool size at the root 1250 * minus any outstanding deferred frees. 1251 * To ensure that it's possible to remove files from a full 1252 * pool without inducing transient overcommits, we throttle 1253 * netfree transactions against a quota that is slightly larger, 1254 * but still within the pool's allocation slop. In cases where 1255 * we're very close to full, this will allow a steady trickle of 1256 * removes to get through. 1257 */ 1258 uint64_t deferred = 0; 1259 if (dd->dd_parent == NULL) { 1260 spa_t *spa = dd->dd_pool->dp_spa; 1261 uint64_t poolsize = dsl_pool_adjustedsize(dd->dd_pool, netfree); 1262 deferred = metaslab_class_get_deferred(spa_normal_class(spa)); 1263 if (poolsize - deferred < quota) { 1264 quota = poolsize - deferred; 1265 retval = ENOSPC; 1266 } 1267 } 1268 1269 /* 1270 * If they are requesting more space, and our current estimate 1271 * is over quota, they get to try again unless the actual 1272 * on-disk is over quota and there are no pending changes (which 1273 * may free up space for us). 1274 */ 1275 if (used_on_disk + est_inflight >= quota) { 1276 if (est_inflight > 0 || used_on_disk < quota || 1277 (retval == ENOSPC && used_on_disk < quota + deferred)) 1278 retval = ERESTART; 1279 dprintf_dd(dd, "failing: used=%lluK inflight = %lluK " 1280 "quota=%lluK tr=%lluK err=%d\n", 1281 used_on_disk>>10, est_inflight>>10, 1282 quota>>10, asize>>10, retval); 1283 mutex_exit(&dd->dd_lock); 1284 return (SET_ERROR(retval)); 1285 } 1286 1287 /* We need to up our estimated delta before dropping dd_lock */ 1288 dd->dd_tempreserved[txg & TXG_MASK] += asize; 1289 1290 uint64_t parent_rsrv = parent_delta(dd, used_on_disk + est_inflight, 1291 asize - ref_rsrv); 1292 mutex_exit(&dd->dd_lock); 1293 1294 tr = kmem_zalloc(sizeof (struct tempreserve), KM_SLEEP); 1295 tr->tr_ds = dd; 1296 tr->tr_size = asize; 1297 list_insert_tail(tr_list, tr); 1298 1299 /* see if it's OK with our parent */ 1300 if (dd->dd_parent != NULL && parent_rsrv != 0) { 1301 boolean_t ismos = (dsl_dir_phys(dd)->dd_head_dataset_obj == 0); 1302 1303 return (dsl_dir_tempreserve_impl(dd->dd_parent, 1304 parent_rsrv, netfree, ismos, tr_list, tx, B_FALSE)); 1305 } else { 1306 return (0); 1307 } 1308} 1309 1310/* 1311 * Reserve space in this dsl_dir, to be used in this tx's txg. 1312 * After the space has been dirtied (and dsl_dir_willuse_space() 1313 * has been called), the reservation should be canceled, using 1314 * dsl_dir_tempreserve_clear(). 1315 */ 1316int 1317dsl_dir_tempreserve_space(dsl_dir_t *dd, uint64_t lsize, uint64_t asize, 1318 boolean_t netfree, void **tr_cookiep, dmu_tx_t *tx) 1319{ 1320 int err; 1321 list_t *tr_list; 1322 1323 if (asize == 0) { 1324 *tr_cookiep = NULL; 1325 return (0); 1326 } 1327 1328 tr_list = kmem_alloc(sizeof (list_t), KM_SLEEP); 1329 list_create(tr_list, sizeof (struct tempreserve), 1330 offsetof(struct tempreserve, tr_node)); 1331 ASSERT3S(asize, >, 0); 1332 1333 err = arc_tempreserve_space(lsize, tx->tx_txg); 1334 if (err == 0) { 1335 struct tempreserve *tr; 1336 1337 tr = kmem_zalloc(sizeof (struct tempreserve), KM_SLEEP); 1338 tr->tr_size = lsize; 1339 list_insert_tail(tr_list, tr); 1340 } else { 1341 if (err == EAGAIN) { 1342 /* 1343 * If arc_memory_throttle() detected that pageout 1344 * is running and we are low on memory, we delay new 1345 * non-pageout transactions to give pageout an 1346 * advantage. 1347 * 1348 * It is unfortunate to be delaying while the caller's 1349 * locks are held. 1350 */ 1351 txg_delay(dd->dd_pool, tx->tx_txg, 1352 MSEC2NSEC(10), MSEC2NSEC(10)); 1353 err = SET_ERROR(ERESTART); 1354 } 1355 } 1356 1357 if (err == 0) { 1358 err = dsl_dir_tempreserve_impl(dd, asize, netfree, 1359 B_FALSE, tr_list, tx, B_TRUE); 1360 } 1361 1362 if (err != 0) 1363 dsl_dir_tempreserve_clear(tr_list, tx); 1364 else 1365 *tr_cookiep = tr_list; 1366 1367 return (err); 1368} 1369 1370/* 1371 * Clear a temporary reservation that we previously made with 1372 * dsl_dir_tempreserve_space(). 1373 */ 1374void 1375dsl_dir_tempreserve_clear(void *tr_cookie, dmu_tx_t *tx) 1376{ 1377 int txgidx = tx->tx_txg & TXG_MASK; 1378 list_t *tr_list = tr_cookie; 1379 struct tempreserve *tr; 1380 1381 ASSERT3U(tx->tx_txg, !=, 0); 1382 1383 if (tr_cookie == NULL) 1384 return; 1385 1386 while ((tr = list_head(tr_list)) != NULL) { 1387 if (tr->tr_ds) { 1388 mutex_enter(&tr->tr_ds->dd_lock); 1389 ASSERT3U(tr->tr_ds->dd_tempreserved[txgidx], >=, 1390 tr->tr_size); 1391 tr->tr_ds->dd_tempreserved[txgidx] -= tr->tr_size; 1392 mutex_exit(&tr->tr_ds->dd_lock); 1393 } else { 1394 arc_tempreserve_clear(tr->tr_size); 1395 } 1396 list_remove(tr_list, tr); 1397 kmem_free(tr, sizeof (struct tempreserve)); 1398 } 1399 1400 kmem_free(tr_list, sizeof (list_t)); 1401} 1402 1403/* 1404 * This should be called from open context when we think we're going to write 1405 * or free space, for example when dirtying data. Be conservative; it's okay 1406 * to write less space or free more, but we don't want to write more or free 1407 * less than the amount specified. 1408 */ 1409void 1410dsl_dir_willuse_space(dsl_dir_t *dd, int64_t space, dmu_tx_t *tx) 1411{ 1412 int64_t parent_space; 1413 uint64_t est_used; 1414 1415 mutex_enter(&dd->dd_lock); 1416 if (space > 0) 1417 dd->dd_space_towrite[tx->tx_txg & TXG_MASK] += space; 1418 1419 est_used = dsl_dir_space_towrite(dd) + dsl_dir_phys(dd)->dd_used_bytes; 1420 parent_space = parent_delta(dd, est_used, space); 1421 mutex_exit(&dd->dd_lock); 1422 1423 /* Make sure that we clean up dd_space_to* */ 1424 dsl_dir_dirty(dd, tx); 1425 1426 /* XXX this is potentially expensive and unnecessary... */ 1427 if (parent_space && dd->dd_parent) 1428 dsl_dir_willuse_space(dd->dd_parent, parent_space, tx); 1429} 1430 1431/* call from syncing context when we actually write/free space for this dd */ 1432void 1433dsl_dir_diduse_space(dsl_dir_t *dd, dd_used_t type, 1434 int64_t used, int64_t compressed, int64_t uncompressed, dmu_tx_t *tx) 1435{ 1436 int64_t accounted_delta; 1437 1438 /* 1439 * dsl_dataset_set_refreservation_sync_impl() calls this with 1440 * dd_lock held, so that it can atomically update 1441 * ds->ds_reserved and the dsl_dir accounting, so that 1442 * dsl_dataset_check_quota() can see dataset and dir accounting 1443 * consistently. 1444 */ 1445 boolean_t needlock = !MUTEX_HELD(&dd->dd_lock); 1446 1447 ASSERT(dmu_tx_is_syncing(tx)); 1448 ASSERT(type < DD_USED_NUM); 1449 1450 dmu_buf_will_dirty(dd->dd_dbuf, tx); 1451 1452 if (needlock) 1453 mutex_enter(&dd->dd_lock); 1454 accounted_delta = 1455 parent_delta(dd, dsl_dir_phys(dd)->dd_used_bytes, used); 1456 ASSERT(used >= 0 || dsl_dir_phys(dd)->dd_used_bytes >= -used); 1457 ASSERT(compressed >= 0 || 1458 dsl_dir_phys(dd)->dd_compressed_bytes >= -compressed); 1459 ASSERT(uncompressed >= 0 || 1460 dsl_dir_phys(dd)->dd_uncompressed_bytes >= -uncompressed); 1461 dsl_dir_phys(dd)->dd_used_bytes += used; 1462 dsl_dir_phys(dd)->dd_uncompressed_bytes += uncompressed; 1463 dsl_dir_phys(dd)->dd_compressed_bytes += compressed; 1464 1465 if (dsl_dir_phys(dd)->dd_flags & DD_FLAG_USED_BREAKDOWN) { 1466 ASSERT(used > 0 || 1467 dsl_dir_phys(dd)->dd_used_breakdown[type] >= -used); 1468 dsl_dir_phys(dd)->dd_used_breakdown[type] += used; 1469#ifdef DEBUG 1470 dd_used_t t; 1471 uint64_t u = 0; 1472 for (t = 0; t < DD_USED_NUM; t++) 1473 u += dsl_dir_phys(dd)->dd_used_breakdown[t]; 1474 ASSERT3U(u, ==, dsl_dir_phys(dd)->dd_used_bytes); 1475#endif 1476 } 1477 if (needlock) 1478 mutex_exit(&dd->dd_lock); 1479 1480 if (dd->dd_parent != NULL) { 1481 dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD, 1482 accounted_delta, compressed, uncompressed, tx); 1483 dsl_dir_transfer_space(dd->dd_parent, 1484 used - accounted_delta, 1485 DD_USED_CHILD_RSRV, DD_USED_CHILD, NULL); 1486 } 1487} 1488 1489void 1490dsl_dir_transfer_space(dsl_dir_t *dd, int64_t delta, 1491 dd_used_t oldtype, dd_used_t newtype, dmu_tx_t *tx) 1492{ 1493 ASSERT(tx == NULL || dmu_tx_is_syncing(tx)); 1494 ASSERT(oldtype < DD_USED_NUM); 1495 ASSERT(newtype < DD_USED_NUM); 1496 1497 if (delta == 0 || 1498 !(dsl_dir_phys(dd)->dd_flags & DD_FLAG_USED_BREAKDOWN)) 1499 return; 1500 1501 if (tx != NULL) 1502 dmu_buf_will_dirty(dd->dd_dbuf, tx); 1503 mutex_enter(&dd->dd_lock); 1504 ASSERT(delta > 0 ? 1505 dsl_dir_phys(dd)->dd_used_breakdown[oldtype] >= delta : 1506 dsl_dir_phys(dd)->dd_used_breakdown[newtype] >= -delta); 1507 ASSERT(dsl_dir_phys(dd)->dd_used_bytes >= ABS(delta)); 1508 dsl_dir_phys(dd)->dd_used_breakdown[oldtype] -= delta; 1509 dsl_dir_phys(dd)->dd_used_breakdown[newtype] += delta; 1510 mutex_exit(&dd->dd_lock); 1511} 1512 1513typedef struct dsl_dir_set_qr_arg { 1514 const char *ddsqra_name; 1515 zprop_source_t ddsqra_source; 1516 uint64_t ddsqra_value; 1517} dsl_dir_set_qr_arg_t; 1518 1519static int 1520dsl_dir_set_quota_check(void *arg, dmu_tx_t *tx) 1521{ 1522 dsl_dir_set_qr_arg_t *ddsqra = arg; 1523 dsl_pool_t *dp = dmu_tx_pool(tx); 1524 dsl_dataset_t *ds; 1525 int error; 1526 uint64_t towrite, newval; 1527 1528 error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds); 1529 if (error != 0) 1530 return (error); 1531 1532 error = dsl_prop_predict(ds->ds_dir, "quota", 1533 ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval); 1534 if (error != 0) { 1535 dsl_dataset_rele(ds, FTAG); 1536 return (error); 1537 } 1538 1539 if (newval == 0) { 1540 dsl_dataset_rele(ds, FTAG); 1541 return (0); 1542 } 1543 1544 mutex_enter(&ds->ds_dir->dd_lock); 1545 /* 1546 * If we are doing the preliminary check in open context, and 1547 * there are pending changes, then don't fail it, since the 1548 * pending changes could under-estimate the amount of space to be 1549 * freed up. 1550 */ 1551 towrite = dsl_dir_space_towrite(ds->ds_dir); 1552 if ((dmu_tx_is_syncing(tx) || towrite == 0) && 1553 (newval < dsl_dir_phys(ds->ds_dir)->dd_reserved || 1554 newval < dsl_dir_phys(ds->ds_dir)->dd_used_bytes + towrite)) { 1555 error = SET_ERROR(ENOSPC); 1556 } 1557 mutex_exit(&ds->ds_dir->dd_lock); 1558 dsl_dataset_rele(ds, FTAG); 1559 return (error); 1560} 1561 1562static void 1563dsl_dir_set_quota_sync(void *arg, dmu_tx_t *tx) 1564{ 1565 dsl_dir_set_qr_arg_t *ddsqra = arg; 1566 dsl_pool_t *dp = dmu_tx_pool(tx); 1567 dsl_dataset_t *ds; 1568 uint64_t newval; 1569 1570 VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds)); 1571 1572 if (spa_version(dp->dp_spa) >= SPA_VERSION_RECVD_PROPS) { 1573 dsl_prop_set_sync_impl(ds, zfs_prop_to_name(ZFS_PROP_QUOTA), 1574 ddsqra->ddsqra_source, sizeof (ddsqra->ddsqra_value), 1, 1575 &ddsqra->ddsqra_value, tx); 1576 1577 VERIFY0(dsl_prop_get_int_ds(ds, 1578 zfs_prop_to_name(ZFS_PROP_QUOTA), &newval)); 1579 } else { 1580 newval = ddsqra->ddsqra_value; 1581 spa_history_log_internal_ds(ds, "set", tx, "%s=%lld", 1582 zfs_prop_to_name(ZFS_PROP_QUOTA), (longlong_t)newval); 1583 } 1584 1585 dmu_buf_will_dirty(ds->ds_dir->dd_dbuf, tx); 1586 mutex_enter(&ds->ds_dir->dd_lock); 1587 dsl_dir_phys(ds->ds_dir)->dd_quota = newval; 1588 mutex_exit(&ds->ds_dir->dd_lock); 1589 dsl_dataset_rele(ds, FTAG); 1590} 1591 1592int 1593dsl_dir_set_quota(const char *ddname, zprop_source_t source, uint64_t quota) 1594{ 1595 dsl_dir_set_qr_arg_t ddsqra; 1596 1597 ddsqra.ddsqra_name = ddname; 1598 ddsqra.ddsqra_source = source; 1599 ddsqra.ddsqra_value = quota; 1600 1601 return (dsl_sync_task(ddname, dsl_dir_set_quota_check, 1602 dsl_dir_set_quota_sync, &ddsqra, 0, ZFS_SPACE_CHECK_NONE)); 1603} 1604 1605int 1606dsl_dir_set_reservation_check(void *arg, dmu_tx_t *tx) 1607{ 1608 dsl_dir_set_qr_arg_t *ddsqra = arg; 1609 dsl_pool_t *dp = dmu_tx_pool(tx); 1610 dsl_dataset_t *ds; 1611 dsl_dir_t *dd; 1612 uint64_t newval, used, avail; 1613 int error; 1614 1615 error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds); 1616 if (error != 0) 1617 return (error); 1618 dd = ds->ds_dir; 1619 1620 /* 1621 * If we are doing the preliminary check in open context, the 1622 * space estimates may be inaccurate. 1623 */ 1624 if (!dmu_tx_is_syncing(tx)) { 1625 dsl_dataset_rele(ds, FTAG); 1626 return (0); 1627 } 1628 1629 error = dsl_prop_predict(ds->ds_dir, 1630 zfs_prop_to_name(ZFS_PROP_RESERVATION), 1631 ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval); 1632 if (error != 0) { 1633 dsl_dataset_rele(ds, FTAG); 1634 return (error); 1635 } 1636 1637 mutex_enter(&dd->dd_lock); 1638 used = dsl_dir_phys(dd)->dd_used_bytes; 1639 mutex_exit(&dd->dd_lock); 1640 1641 if (dd->dd_parent) { 1642 avail = dsl_dir_space_available(dd->dd_parent, 1643 NULL, 0, FALSE); 1644 } else { 1645 avail = dsl_pool_adjustedsize(dd->dd_pool, B_FALSE) - used; 1646 } 1647 1648 if (MAX(used, newval) > MAX(used, dsl_dir_phys(dd)->dd_reserved)) { 1649 uint64_t delta = MAX(used, newval) - 1650 MAX(used, dsl_dir_phys(dd)->dd_reserved); 1651 1652 if (delta > avail || 1653 (dsl_dir_phys(dd)->dd_quota > 0 && 1654 newval > dsl_dir_phys(dd)->dd_quota)) 1655 error = SET_ERROR(ENOSPC); 1656 } 1657 1658 dsl_dataset_rele(ds, FTAG); 1659 return (error); 1660} 1661 1662void 1663dsl_dir_set_reservation_sync_impl(dsl_dir_t *dd, uint64_t value, dmu_tx_t *tx) 1664{ 1665 uint64_t used; 1666 int64_t delta; 1667 1668 dmu_buf_will_dirty(dd->dd_dbuf, tx); 1669 1670 mutex_enter(&dd->dd_lock); 1671 used = dsl_dir_phys(dd)->dd_used_bytes; 1672 delta = MAX(used, value) - MAX(used, dsl_dir_phys(dd)->dd_reserved); 1673 dsl_dir_phys(dd)->dd_reserved = value; 1674 1675 if (dd->dd_parent != NULL) { 1676 /* Roll up this additional usage into our ancestors */ 1677 dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD_RSRV, 1678 delta, 0, 0, tx); 1679 } 1680 mutex_exit(&dd->dd_lock); 1681} 1682 1683static void 1684dsl_dir_set_reservation_sync(void *arg, dmu_tx_t *tx) 1685{ 1686 dsl_dir_set_qr_arg_t *ddsqra = arg; 1687 dsl_pool_t *dp = dmu_tx_pool(tx); 1688 dsl_dataset_t *ds; 1689 uint64_t newval; 1690 1691 VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds)); 1692 1693 if (spa_version(dp->dp_spa) >= SPA_VERSION_RECVD_PROPS) { 1694 dsl_prop_set_sync_impl(ds, 1695 zfs_prop_to_name(ZFS_PROP_RESERVATION), 1696 ddsqra->ddsqra_source, sizeof (ddsqra->ddsqra_value), 1, 1697 &ddsqra->ddsqra_value, tx); 1698 1699 VERIFY0(dsl_prop_get_int_ds(ds, 1700 zfs_prop_to_name(ZFS_PROP_RESERVATION), &newval)); 1701 } else { 1702 newval = ddsqra->ddsqra_value; 1703 spa_history_log_internal_ds(ds, "set", tx, "%s=%lld", 1704 zfs_prop_to_name(ZFS_PROP_RESERVATION), 1705 (longlong_t)newval); 1706 } 1707 1708 dsl_dir_set_reservation_sync_impl(ds->ds_dir, newval, tx); 1709 dsl_dataset_rele(ds, FTAG); 1710} 1711 1712int 1713dsl_dir_set_reservation(const char *ddname, zprop_source_t source, 1714 uint64_t reservation) 1715{ 1716 dsl_dir_set_qr_arg_t ddsqra; 1717 1718 ddsqra.ddsqra_name = ddname; 1719 ddsqra.ddsqra_source = source; 1720 ddsqra.ddsqra_value = reservation; 1721 1722 return (dsl_sync_task(ddname, dsl_dir_set_reservation_check, 1723 dsl_dir_set_reservation_sync, &ddsqra, 0, ZFS_SPACE_CHECK_NONE)); 1724} 1725 1726static dsl_dir_t * 1727closest_common_ancestor(dsl_dir_t *ds1, dsl_dir_t *ds2) 1728{ 1729 for (; ds1; ds1 = ds1->dd_parent) { 1730 dsl_dir_t *dd; 1731 for (dd = ds2; dd; dd = dd->dd_parent) { 1732 if (ds1 == dd) 1733 return (dd); 1734 } 1735 } 1736 return (NULL); 1737} 1738 1739/* 1740 * If delta is applied to dd, how much of that delta would be applied to 1741 * ancestor? Syncing context only. 1742 */ 1743static int64_t 1744would_change(dsl_dir_t *dd, int64_t delta, dsl_dir_t *ancestor) 1745{ 1746 if (dd == ancestor) 1747 return (delta); 1748 1749 mutex_enter(&dd->dd_lock); 1750 delta = parent_delta(dd, dsl_dir_phys(dd)->dd_used_bytes, delta); 1751 mutex_exit(&dd->dd_lock); 1752 return (would_change(dd->dd_parent, delta, ancestor)); 1753} 1754 1755typedef struct dsl_dir_rename_arg { 1756 const char *ddra_oldname; 1757 const char *ddra_newname; 1758 cred_t *ddra_cred; 1759} dsl_dir_rename_arg_t; 1760 1761/* ARGSUSED */ 1762static int 1763dsl_valid_rename(dsl_pool_t *dp, dsl_dataset_t *ds, void *arg) 1764{ 1765 int *deltap = arg; 1766 char namebuf[ZFS_MAX_DATASET_NAME_LEN]; 1767 1768 dsl_dataset_name(ds, namebuf); 1769 1770 if (strlen(namebuf) + *deltap >= ZFS_MAX_DATASET_NAME_LEN) 1771 return (SET_ERROR(ENAMETOOLONG)); 1772 return (0); 1773} 1774 1775static int 1776dsl_dir_rename_check(void *arg, dmu_tx_t *tx) 1777{ 1778 dsl_dir_rename_arg_t *ddra = arg; 1779 dsl_pool_t *dp = dmu_tx_pool(tx); 1780 dsl_dir_t *dd, *newparent; 1781 const char *mynewname; 1782 int error; 1783 int delta = strlen(ddra->ddra_newname) - strlen(ddra->ddra_oldname); 1784 1785 /* target dir should exist */ 1786 error = dsl_dir_hold(dp, ddra->ddra_oldname, FTAG, &dd, NULL); 1787 if (error != 0) 1788 return (error); 1789 1790 /* new parent should exist */ 1791 error = dsl_dir_hold(dp, ddra->ddra_newname, FTAG, 1792 &newparent, &mynewname); 1793 if (error != 0) { 1794 dsl_dir_rele(dd, FTAG); 1795 return (error); 1796 } 1797 1798 /* can't rename to different pool */ 1799 if (dd->dd_pool != newparent->dd_pool) { 1800 dsl_dir_rele(newparent, FTAG); 1801 dsl_dir_rele(dd, FTAG); 1802 return (SET_ERROR(EXDEV)); 1803 } 1804 1805 /* new name should not already exist */ 1806 if (mynewname == NULL) { 1807 dsl_dir_rele(newparent, FTAG); 1808 dsl_dir_rele(dd, FTAG); 1809 return (SET_ERROR(EEXIST)); 1810 } 1811 1812 /* if the name length is growing, validate child name lengths */ 1813 if (delta > 0) { 1814 error = dmu_objset_find_dp(dp, dd->dd_object, dsl_valid_rename, 1815 &delta, DS_FIND_CHILDREN | DS_FIND_SNAPSHOTS); 1816 if (error != 0) { 1817 dsl_dir_rele(newparent, FTAG); 1818 dsl_dir_rele(dd, FTAG); 1819 return (error); 1820 } 1821 } 1822 1823 if (dmu_tx_is_syncing(tx)) { 1824 if (spa_feature_is_active(dp->dp_spa, 1825 SPA_FEATURE_FS_SS_LIMIT)) { 1826 /* 1827 * Although this is the check function and we don't 1828 * normally make on-disk changes in check functions, 1829 * we need to do that here. 1830 * 1831 * Ensure this portion of the tree's counts have been 1832 * initialized in case the new parent has limits set. 1833 */ 1834 dsl_dir_init_fs_ss_count(dd, tx); 1835 } 1836 } 1837 1838 if (newparent != dd->dd_parent) { 1839 /* is there enough space? */ 1840 uint64_t myspace = 1841 MAX(dsl_dir_phys(dd)->dd_used_bytes, 1842 dsl_dir_phys(dd)->dd_reserved); 1843 objset_t *os = dd->dd_pool->dp_meta_objset; 1844 uint64_t fs_cnt = 0; 1845 uint64_t ss_cnt = 0; 1846 1847 if (dsl_dir_is_zapified(dd)) { 1848 int err; 1849 1850 err = zap_lookup(os, dd->dd_object, 1851 DD_FIELD_FILESYSTEM_COUNT, sizeof (fs_cnt), 1, 1852 &fs_cnt); 1853 if (err != ENOENT && err != 0) { 1854 dsl_dir_rele(newparent, FTAG); 1855 dsl_dir_rele(dd, FTAG); 1856 return (err); 1857 } 1858 1859 /* 1860 * have to add 1 for the filesystem itself that we're 1861 * moving 1862 */ 1863 fs_cnt++; 1864 1865 err = zap_lookup(os, dd->dd_object, 1866 DD_FIELD_SNAPSHOT_COUNT, sizeof (ss_cnt), 1, 1867 &ss_cnt); 1868 if (err != ENOENT && err != 0) { 1869 dsl_dir_rele(newparent, FTAG); 1870 dsl_dir_rele(dd, FTAG); 1871 return (err); 1872 } 1873 } 1874 1875 /* no rename into our descendant */ 1876 if (closest_common_ancestor(dd, newparent) == dd) { 1877 dsl_dir_rele(newparent, FTAG); 1878 dsl_dir_rele(dd, FTAG); 1879 return (SET_ERROR(EINVAL)); 1880 } 1881 1882 error = dsl_dir_transfer_possible(dd->dd_parent, 1883 newparent, fs_cnt, ss_cnt, myspace, ddra->ddra_cred); 1884 if (error != 0) { 1885 dsl_dir_rele(newparent, FTAG); 1886 dsl_dir_rele(dd, FTAG); 1887 return (error); 1888 } 1889 } 1890 1891 dsl_dir_rele(newparent, FTAG); 1892 dsl_dir_rele(dd, FTAG); 1893 return (0); 1894} 1895 1896static void 1897dsl_dir_rename_sync(void *arg, dmu_tx_t *tx) 1898{ 1899 dsl_dir_rename_arg_t *ddra = arg; 1900 dsl_pool_t *dp = dmu_tx_pool(tx); 1901 dsl_dir_t *dd, *newparent; 1902 const char *mynewname; 1903 int error; 1904 objset_t *mos = dp->dp_meta_objset; 1905 1906 VERIFY0(dsl_dir_hold(dp, ddra->ddra_oldname, FTAG, &dd, NULL)); 1907 VERIFY0(dsl_dir_hold(dp, ddra->ddra_newname, FTAG, &newparent, 1908 &mynewname)); 1909 1910 /* Log this before we change the name. */ 1911 spa_history_log_internal_dd(dd, "rename", tx, 1912 "-> %s", ddra->ddra_newname); 1913 1914 if (newparent != dd->dd_parent) { 1915 objset_t *os = dd->dd_pool->dp_meta_objset; 1916 uint64_t fs_cnt = 0; 1917 uint64_t ss_cnt = 0; 1918 1919 /* 1920 * We already made sure the dd counts were initialized in the 1921 * check function. 1922 */ 1923 if (spa_feature_is_active(dp->dp_spa, 1924 SPA_FEATURE_FS_SS_LIMIT)) { 1925 VERIFY0(zap_lookup(os, dd->dd_object, 1926 DD_FIELD_FILESYSTEM_COUNT, sizeof (fs_cnt), 1, 1927 &fs_cnt)); 1928 /* add 1 for the filesystem itself that we're moving */ 1929 fs_cnt++; 1930 1931 VERIFY0(zap_lookup(os, dd->dd_object, 1932 DD_FIELD_SNAPSHOT_COUNT, sizeof (ss_cnt), 1, 1933 &ss_cnt)); 1934 } 1935 1936 dsl_fs_ss_count_adjust(dd->dd_parent, -fs_cnt, 1937 DD_FIELD_FILESYSTEM_COUNT, tx); 1938 dsl_fs_ss_count_adjust(newparent, fs_cnt, 1939 DD_FIELD_FILESYSTEM_COUNT, tx); 1940 1941 dsl_fs_ss_count_adjust(dd->dd_parent, -ss_cnt, 1942 DD_FIELD_SNAPSHOT_COUNT, tx); 1943 dsl_fs_ss_count_adjust(newparent, ss_cnt, 1944 DD_FIELD_SNAPSHOT_COUNT, tx); 1945 1946 dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD, 1947 -dsl_dir_phys(dd)->dd_used_bytes, 1948 -dsl_dir_phys(dd)->dd_compressed_bytes, 1949 -dsl_dir_phys(dd)->dd_uncompressed_bytes, tx); 1950 dsl_dir_diduse_space(newparent, DD_USED_CHILD, 1951 dsl_dir_phys(dd)->dd_used_bytes, 1952 dsl_dir_phys(dd)->dd_compressed_bytes, 1953 dsl_dir_phys(dd)->dd_uncompressed_bytes, tx); 1954 1955 if (dsl_dir_phys(dd)->dd_reserved > 1956 dsl_dir_phys(dd)->dd_used_bytes) { 1957 uint64_t unused_rsrv = dsl_dir_phys(dd)->dd_reserved - 1958 dsl_dir_phys(dd)->dd_used_bytes; 1959 1960 dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD_RSRV, 1961 -unused_rsrv, 0, 0, tx); 1962 dsl_dir_diduse_space(newparent, DD_USED_CHILD_RSRV, 1963 unused_rsrv, 0, 0, tx); 1964 } 1965 } 1966 1967 dmu_buf_will_dirty(dd->dd_dbuf, tx); 1968 1969 /* remove from old parent zapobj */ 1970 error = zap_remove(mos, 1971 dsl_dir_phys(dd->dd_parent)->dd_child_dir_zapobj, 1972 dd->dd_myname, tx); 1973 ASSERT0(error); 1974 1975 (void) strcpy(dd->dd_myname, mynewname); 1976 dsl_dir_rele(dd->dd_parent, dd); 1977 dsl_dir_phys(dd)->dd_parent_obj = newparent->dd_object; 1978 VERIFY0(dsl_dir_hold_obj(dp, 1979 newparent->dd_object, NULL, dd, &dd->dd_parent)); 1980 1981 /* add to new parent zapobj */ 1982 VERIFY0(zap_add(mos, dsl_dir_phys(newparent)->dd_child_dir_zapobj, 1983 dd->dd_myname, 8, 1, &dd->dd_object, tx)); 1984 1985#ifdef __FreeBSD__ 1986#ifdef _KERNEL 1987 zfsvfs_update_fromname(ddra->ddra_oldname, ddra->ddra_newname); 1988 zvol_rename_minors(ddra->ddra_oldname, ddra->ddra_newname); 1989#endif 1990#endif 1991 1992 dsl_prop_notify_all(dd); 1993 1994 dsl_dir_rele(newparent, FTAG); 1995 dsl_dir_rele(dd, FTAG); 1996} 1997 1998int 1999dsl_dir_rename(const char *oldname, const char *newname) 2000{ 2001 dsl_dir_rename_arg_t ddra; 2002 2003 ddra.ddra_oldname = oldname; 2004 ddra.ddra_newname = newname; 2005 ddra.ddra_cred = CRED(); 2006 2007 return (dsl_sync_task(oldname, 2008 dsl_dir_rename_check, dsl_dir_rename_sync, &ddra, 2009 3, ZFS_SPACE_CHECK_RESERVED)); 2010} 2011 2012int 2013dsl_dir_transfer_possible(dsl_dir_t *sdd, dsl_dir_t *tdd, 2014 uint64_t fs_cnt, uint64_t ss_cnt, uint64_t space, cred_t *cr) 2015{ 2016 dsl_dir_t *ancestor; 2017 int64_t adelta; 2018 uint64_t avail; 2019 int err; 2020 2021 ancestor = closest_common_ancestor(sdd, tdd); 2022 adelta = would_change(sdd, -space, ancestor); 2023 avail = dsl_dir_space_available(tdd, ancestor, adelta, FALSE); 2024 if (avail < space) 2025 return (SET_ERROR(ENOSPC)); 2026 2027 err = dsl_fs_ss_limit_check(tdd, fs_cnt, ZFS_PROP_FILESYSTEM_LIMIT, 2028 ancestor, cr); 2029 if (err != 0) 2030 return (err); 2031 err = dsl_fs_ss_limit_check(tdd, ss_cnt, ZFS_PROP_SNAPSHOT_LIMIT, 2032 ancestor, cr); 2033 if (err != 0) 2034 return (err); 2035 2036 return (0); 2037} 2038 2039timestruc_t 2040dsl_dir_snap_cmtime(dsl_dir_t *dd) 2041{ 2042 timestruc_t t; 2043 2044 mutex_enter(&dd->dd_lock); 2045 t = dd->dd_snap_cmtime; 2046 mutex_exit(&dd->dd_lock); 2047 2048 return (t); 2049} 2050 2051void 2052dsl_dir_snap_cmtime_update(dsl_dir_t *dd) 2053{ 2054 timestruc_t t; 2055 2056 gethrestime(&t); 2057 mutex_enter(&dd->dd_lock); 2058 dd->dd_snap_cmtime = t; 2059 mutex_exit(&dd->dd_lock); 2060} 2061 2062void 2063dsl_dir_zapify(dsl_dir_t *dd, dmu_tx_t *tx) 2064{ 2065 objset_t *mos = dd->dd_pool->dp_meta_objset; 2066 dmu_object_zapify(mos, dd->dd_object, DMU_OT_DSL_DIR, tx); 2067} 2068 2069boolean_t 2070dsl_dir_is_zapified(dsl_dir_t *dd) 2071{ 2072 dmu_object_info_t doi; 2073 2074 dmu_object_info_from_db(dd->dd_dbuf, &doi); 2075 return (doi.doi_type == DMU_OTN_ZAP_METADATA); 2076} 2077