dsl_dir.c revision 288549
1164856Sdds/* 2164856Sdds * CDDL HEADER START 3164856Sdds * 4164856Sdds * The contents of this file are subject to the terms of the 5164856Sdds * Common Development and Distribution License (the "License"). 6164856Sdds * You may not use this file except in compliance with the License. 7164856Sdds * 8164856Sdds * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 9164856Sdds * or http://www.opensolaris.org/os/licensing. 10164856Sdds * 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, 2014 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(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 /* 156 * The props callback list should have been cleaned up by 157 * objset_evict(). 158 */ 159 list_destroy(&dd->dd_prop_cbs); 160 mutex_destroy(&dd->dd_lock); 161 kmem_free(dd, sizeof (dsl_dir_t)); 162} 163 164int 165dsl_dir_hold_obj(dsl_pool_t *dp, uint64_t ddobj, 166 const char *tail, void *tag, dsl_dir_t **ddp) 167{ 168 dmu_buf_t *dbuf; 169 dsl_dir_t *dd; 170 int err; 171 172 ASSERT(dsl_pool_config_held(dp)); 173 174 err = dmu_bonus_hold(dp->dp_meta_objset, ddobj, tag, &dbuf); 175 if (err != 0) 176 return (err); 177 dd = dmu_buf_get_user(dbuf); 178#ifdef ZFS_DEBUG 179 { 180 dmu_object_info_t doi; 181 dmu_object_info_from_db(dbuf, &doi); 182 ASSERT3U(doi.doi_bonus_type, ==, DMU_OT_DSL_DIR); 183 ASSERT3U(doi.doi_bonus_size, >=, sizeof (dsl_dir_phys_t)); 184 } 185#endif 186 if (dd == NULL) { 187 dsl_dir_t *winner; 188 189 dd = kmem_zalloc(sizeof (dsl_dir_t), KM_SLEEP); 190 dd->dd_object = ddobj; 191 dd->dd_dbuf = dbuf; 192 dd->dd_pool = dp; 193 mutex_init(&dd->dd_lock, NULL, MUTEX_DEFAULT, NULL); 194 195 list_create(&dd->dd_prop_cbs, sizeof (dsl_prop_cb_record_t), 196 offsetof(dsl_prop_cb_record_t, cbr_node)); 197 198 dsl_dir_snap_cmtime_update(dd); 199 200 if (dsl_dir_phys(dd)->dd_parent_obj) { 201 err = dsl_dir_hold_obj(dp, 202 dsl_dir_phys(dd)->dd_parent_obj, NULL, dd, 203 &dd->dd_parent); 204 if (err != 0) 205 goto errout; 206 if (tail) { 207#ifdef ZFS_DEBUG 208 uint64_t foundobj; 209 210 err = zap_lookup(dp->dp_meta_objset, 211 dsl_dir_phys(dd->dd_parent)-> 212 dd_child_dir_zapobj, tail, 213 sizeof (foundobj), 1, &foundobj); 214 ASSERT(err || foundobj == ddobj); 215#endif 216 (void) strcpy(dd->dd_myname, tail); 217 } else { 218 err = zap_value_search(dp->dp_meta_objset, 219 dsl_dir_phys(dd->dd_parent)-> 220 dd_child_dir_zapobj, 221 ddobj, 0, dd->dd_myname); 222 } 223 if (err != 0) 224 goto errout; 225 } else { 226 (void) strcpy(dd->dd_myname, spa_name(dp->dp_spa)); 227 } 228 229 if (dsl_dir_is_clone(dd)) { 230 dmu_buf_t *origin_bonus; 231 dsl_dataset_phys_t *origin_phys; 232 233 /* 234 * We can't open the origin dataset, because 235 * that would require opening this dsl_dir. 236 * Just look at its phys directly instead. 237 */ 238 err = dmu_bonus_hold(dp->dp_meta_objset, 239 dsl_dir_phys(dd)->dd_origin_obj, FTAG, 240 &origin_bonus); 241 if (err != 0) 242 goto errout; 243 origin_phys = origin_bonus->db_data; 244 dd->dd_origin_txg = 245 origin_phys->ds_creation_txg; 246 dmu_buf_rele(origin_bonus, FTAG); 247 } 248 249 dmu_buf_init_user(&dd->dd_dbu, dsl_dir_evict, &dd->dd_dbuf); 250 winner = dmu_buf_set_user_ie(dbuf, &dd->dd_dbu); 251 if (winner != NULL) { 252 if (dd->dd_parent) 253 dsl_dir_rele(dd->dd_parent, dd); 254 mutex_destroy(&dd->dd_lock); 255 kmem_free(dd, sizeof (dsl_dir_t)); 256 dd = winner; 257 } else { 258 spa_open_ref(dp->dp_spa, dd); 259 } 260 } 261 262 /* 263 * The dsl_dir_t has both open-to-close and instantiate-to-evict 264 * holds on the spa. We need the open-to-close holds because 265 * otherwise the spa_refcnt wouldn't change when we open a 266 * dir which the spa also has open, so we could incorrectly 267 * think it was OK to unload/export/destroy the pool. We need 268 * the instantiate-to-evict hold because the dsl_dir_t has a 269 * pointer to the dd_pool, which has a pointer to the spa_t. 270 */ 271 spa_open_ref(dp->dp_spa, tag); 272 ASSERT3P(dd->dd_pool, ==, dp); 273 ASSERT3U(dd->dd_object, ==, ddobj); 274 ASSERT3P(dd->dd_dbuf, ==, dbuf); 275 *ddp = dd; 276 return (0); 277 278errout: 279 if (dd->dd_parent) 280 dsl_dir_rele(dd->dd_parent, dd); 281 mutex_destroy(&dd->dd_lock); 282 kmem_free(dd, sizeof (dsl_dir_t)); 283 dmu_buf_rele(dbuf, tag); 284 return (err); 285} 286 287void 288dsl_dir_rele(dsl_dir_t *dd, void *tag) 289{ 290 dprintf_dd(dd, "%s\n", ""); 291 spa_close(dd->dd_pool->dp_spa, tag); 292 dmu_buf_rele(dd->dd_dbuf, tag); 293} 294 295/* 296 * Remove a reference to the given dsl dir that is being asynchronously 297 * released. Async releases occur from a taskq performing eviction of 298 * dsl datasets and dirs. This process is identical to a normal release 299 * with the exception of using the async API for releasing the reference on 300 * the spa. 301 */ 302void 303dsl_dir_async_rele(dsl_dir_t *dd, void *tag) 304{ 305 dprintf_dd(dd, "%s\n", ""); 306 spa_async_close(dd->dd_pool->dp_spa, tag); 307 dmu_buf_rele(dd->dd_dbuf, tag); 308} 309 310/* buf must be long enough (MAXNAMELEN + strlen(MOS_DIR_NAME) + 1 should do) */ 311void 312dsl_dir_name(dsl_dir_t *dd, char *buf) 313{ 314 if (dd->dd_parent) { 315 dsl_dir_name(dd->dd_parent, buf); 316 (void) strcat(buf, "/"); 317 } else { 318 buf[0] = '\0'; 319 } 320 if (!MUTEX_HELD(&dd->dd_lock)) { 321 /* 322 * recursive mutex so that we can use 323 * dprintf_dd() with dd_lock held 324 */ 325 mutex_enter(&dd->dd_lock); 326 (void) strcat(buf, dd->dd_myname); 327 mutex_exit(&dd->dd_lock); 328 } else { 329 (void) strcat(buf, dd->dd_myname); 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) >= MAXNAMELEN) 379 return (SET_ERROR(ENAMETOOLONG)); 380 (void) strcpy(component, path); 381 p = NULL; 382 } else if (p[0] == '/') { 383 if (p - path >= MAXNAMELEN) 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 >= MAXNAMELEN) 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[MAXNAMELEN]; 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 937void 938dsl_dir_stats(dsl_dir_t *dd, nvlist_t *nv) 939{ 940 mutex_enter(&dd->dd_lock); 941 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USED, 942 dsl_dir_phys(dd)->dd_used_bytes); 943 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_QUOTA, 944 dsl_dir_phys(dd)->dd_quota); 945 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_RESERVATION, 946 dsl_dir_phys(dd)->dd_reserved); 947 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_COMPRESSRATIO, 948 dsl_dir_phys(dd)->dd_compressed_bytes == 0 ? 100 : 949 (dsl_dir_phys(dd)->dd_uncompressed_bytes * 100 / 950 dsl_dir_phys(dd)->dd_compressed_bytes)); 951 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_LOGICALUSED, 952 dsl_dir_phys(dd)->dd_uncompressed_bytes); 953 if (dsl_dir_phys(dd)->dd_flags & DD_FLAG_USED_BREAKDOWN) { 954 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDSNAP, 955 dsl_dir_phys(dd)->dd_used_breakdown[DD_USED_SNAP]); 956 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDDS, 957 dsl_dir_phys(dd)->dd_used_breakdown[DD_USED_HEAD]); 958 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDREFRESERV, 959 dsl_dir_phys(dd)->dd_used_breakdown[DD_USED_REFRSRV]); 960 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDCHILD, 961 dsl_dir_phys(dd)->dd_used_breakdown[DD_USED_CHILD] + 962 dsl_dir_phys(dd)->dd_used_breakdown[DD_USED_CHILD_RSRV]); 963 } 964 mutex_exit(&dd->dd_lock); 965 966 if (dsl_dir_is_zapified(dd)) { 967 uint64_t count; 968 objset_t *os = dd->dd_pool->dp_meta_objset; 969 970 if (zap_lookup(os, dd->dd_object, DD_FIELD_FILESYSTEM_COUNT, 971 sizeof (count), 1, &count) == 0) { 972 dsl_prop_nvlist_add_uint64(nv, 973 ZFS_PROP_FILESYSTEM_COUNT, count); 974 } 975 if (zap_lookup(os, dd->dd_object, DD_FIELD_SNAPSHOT_COUNT, 976 sizeof (count), 1, &count) == 0) { 977 dsl_prop_nvlist_add_uint64(nv, 978 ZFS_PROP_SNAPSHOT_COUNT, count); 979 } 980 } 981 982 if (dsl_dir_is_clone(dd)) { 983 dsl_dataset_t *ds; 984 char buf[MAXNAMELEN]; 985 986 VERIFY0(dsl_dataset_hold_obj(dd->dd_pool, 987 dsl_dir_phys(dd)->dd_origin_obj, FTAG, &ds)); 988 dsl_dataset_name(ds, buf); 989 dsl_dataset_rele(ds, FTAG); 990 dsl_prop_nvlist_add_string(nv, ZFS_PROP_ORIGIN, buf); 991 } 992} 993 994void 995dsl_dir_dirty(dsl_dir_t *dd, dmu_tx_t *tx) 996{ 997 dsl_pool_t *dp = dd->dd_pool; 998 999 ASSERT(dsl_dir_phys(dd)); 1000 1001 if (txg_list_add(&dp->dp_dirty_dirs, dd, tx->tx_txg)) { 1002 /* up the hold count until we can be written out */ 1003 dmu_buf_add_ref(dd->dd_dbuf, dd); 1004 } 1005} 1006 1007static int64_t 1008parent_delta(dsl_dir_t *dd, uint64_t used, int64_t delta) 1009{ 1010 uint64_t old_accounted = MAX(used, dsl_dir_phys(dd)->dd_reserved); 1011 uint64_t new_accounted = 1012 MAX(used + delta, dsl_dir_phys(dd)->dd_reserved); 1013 return (new_accounted - old_accounted); 1014} 1015 1016void 1017dsl_dir_sync(dsl_dir_t *dd, dmu_tx_t *tx) 1018{ 1019 ASSERT(dmu_tx_is_syncing(tx)); 1020 1021 mutex_enter(&dd->dd_lock); 1022 ASSERT0(dd->dd_tempreserved[tx->tx_txg&TXG_MASK]); 1023 dprintf_dd(dd, "txg=%llu towrite=%lluK\n", tx->tx_txg, 1024 dd->dd_space_towrite[tx->tx_txg&TXG_MASK] / 1024); 1025 dd->dd_space_towrite[tx->tx_txg&TXG_MASK] = 0; 1026 mutex_exit(&dd->dd_lock); 1027 1028 /* release the hold from dsl_dir_dirty */ 1029 dmu_buf_rele(dd->dd_dbuf, dd); 1030} 1031 1032static uint64_t 1033dsl_dir_space_towrite(dsl_dir_t *dd) 1034{ 1035 uint64_t space = 0; 1036 int i; 1037 1038 ASSERT(MUTEX_HELD(&dd->dd_lock)); 1039 1040 for (i = 0; i < TXG_SIZE; i++) { 1041 space += dd->dd_space_towrite[i&TXG_MASK]; 1042 ASSERT3U(dd->dd_space_towrite[i&TXG_MASK], >=, 0); 1043 } 1044 return (space); 1045} 1046 1047/* 1048 * How much space would dd have available if ancestor had delta applied 1049 * to it? If ondiskonly is set, we're only interested in what's 1050 * on-disk, not estimated pending changes. 1051 */ 1052uint64_t 1053dsl_dir_space_available(dsl_dir_t *dd, 1054 dsl_dir_t *ancestor, int64_t delta, int ondiskonly) 1055{ 1056 uint64_t parentspace, myspace, quota, used; 1057 1058 /* 1059 * If there are no restrictions otherwise, assume we have 1060 * unlimited space available. 1061 */ 1062 quota = UINT64_MAX; 1063 parentspace = UINT64_MAX; 1064 1065 if (dd->dd_parent != NULL) { 1066 parentspace = dsl_dir_space_available(dd->dd_parent, 1067 ancestor, delta, ondiskonly); 1068 } 1069 1070 mutex_enter(&dd->dd_lock); 1071 if (dsl_dir_phys(dd)->dd_quota != 0) 1072 quota = dsl_dir_phys(dd)->dd_quota; 1073 used = dsl_dir_phys(dd)->dd_used_bytes; 1074 if (!ondiskonly) 1075 used += dsl_dir_space_towrite(dd); 1076 1077 if (dd->dd_parent == NULL) { 1078 uint64_t poolsize = dsl_pool_adjustedsize(dd->dd_pool, FALSE); 1079 quota = MIN(quota, poolsize); 1080 } 1081 1082 if (dsl_dir_phys(dd)->dd_reserved > used && parentspace != UINT64_MAX) { 1083 /* 1084 * We have some space reserved, in addition to what our 1085 * parent gave us. 1086 */ 1087 parentspace += dsl_dir_phys(dd)->dd_reserved - used; 1088 } 1089 1090 if (dd == ancestor) { 1091 ASSERT(delta <= 0); 1092 ASSERT(used >= -delta); 1093 used += delta; 1094 if (parentspace != UINT64_MAX) 1095 parentspace -= delta; 1096 } 1097 1098 if (used > quota) { 1099 /* over quota */ 1100 myspace = 0; 1101 } else { 1102 /* 1103 * the lesser of the space provided by our parent and 1104 * the space left in our quota 1105 */ 1106 myspace = MIN(parentspace, quota - used); 1107 } 1108 1109 mutex_exit(&dd->dd_lock); 1110 1111 return (myspace); 1112} 1113 1114struct tempreserve { 1115 list_node_t tr_node; 1116 dsl_dir_t *tr_ds; 1117 uint64_t tr_size; 1118}; 1119 1120static int 1121dsl_dir_tempreserve_impl(dsl_dir_t *dd, uint64_t asize, boolean_t netfree, 1122 boolean_t ignorequota, boolean_t checkrefquota, list_t *tr_list, 1123 dmu_tx_t *tx, boolean_t first) 1124{ 1125 uint64_t txg = tx->tx_txg; 1126 uint64_t est_inflight, used_on_disk, quota, parent_rsrv; 1127 uint64_t deferred = 0; 1128 struct tempreserve *tr; 1129 int retval = EDQUOT; 1130 int txgidx = txg & TXG_MASK; 1131 int i; 1132 uint64_t ref_rsrv = 0; 1133 1134 ASSERT3U(txg, !=, 0); 1135 ASSERT3S(asize, >, 0); 1136 1137 mutex_enter(&dd->dd_lock); 1138 1139 /* 1140 * Check against the dsl_dir's quota. We don't add in the delta 1141 * when checking for over-quota because they get one free hit. 1142 */ 1143 est_inflight = dsl_dir_space_towrite(dd); 1144 for (i = 0; i < TXG_SIZE; i++) 1145 est_inflight += dd->dd_tempreserved[i]; 1146 used_on_disk = dsl_dir_phys(dd)->dd_used_bytes; 1147 1148 /* 1149 * On the first iteration, fetch the dataset's used-on-disk and 1150 * refreservation values. Also, if checkrefquota is set, test if 1151 * allocating this space would exceed the dataset's refquota. 1152 */ 1153 if (first && tx->tx_objset) { 1154 int error; 1155 dsl_dataset_t *ds = tx->tx_objset->os_dsl_dataset; 1156 1157 error = dsl_dataset_check_quota(ds, checkrefquota, 1158 asize, est_inflight, &used_on_disk, &ref_rsrv); 1159 if (error) { 1160 mutex_exit(&dd->dd_lock); 1161 return (error); 1162 } 1163 } 1164 1165 /* 1166 * If this transaction will result in a net free of space, 1167 * we want to let it through. 1168 */ 1169 if (ignorequota || netfree || dsl_dir_phys(dd)->dd_quota == 0) 1170 quota = UINT64_MAX; 1171 else 1172 quota = dsl_dir_phys(dd)->dd_quota; 1173 1174 /* 1175 * Adjust the quota against the actual pool size at the root 1176 * minus any outstanding deferred frees. 1177 * To ensure that it's possible to remove files from a full 1178 * pool without inducing transient overcommits, we throttle 1179 * netfree transactions against a quota that is slightly larger, 1180 * but still within the pool's allocation slop. In cases where 1181 * we're very close to full, this will allow a steady trickle of 1182 * removes to get through. 1183 */ 1184 if (dd->dd_parent == NULL) { 1185 spa_t *spa = dd->dd_pool->dp_spa; 1186 uint64_t poolsize = dsl_pool_adjustedsize(dd->dd_pool, netfree); 1187 deferred = metaslab_class_get_deferred(spa_normal_class(spa)); 1188 if (poolsize - deferred < quota) { 1189 quota = poolsize - deferred; 1190 retval = ENOSPC; 1191 } 1192 } 1193 1194 /* 1195 * If they are requesting more space, and our current estimate 1196 * is over quota, they get to try again unless the actual 1197 * on-disk is over quota and there are no pending changes (which 1198 * may free up space for us). 1199 */ 1200 if (used_on_disk + est_inflight >= quota) { 1201 if (est_inflight > 0 || used_on_disk < quota || 1202 (retval == ENOSPC && used_on_disk < quota + deferred)) 1203 retval = ERESTART; 1204 dprintf_dd(dd, "failing: used=%lluK inflight = %lluK " 1205 "quota=%lluK tr=%lluK err=%d\n", 1206 used_on_disk>>10, est_inflight>>10, 1207 quota>>10, asize>>10, retval); 1208 mutex_exit(&dd->dd_lock); 1209 return (SET_ERROR(retval)); 1210 } 1211 1212 /* We need to up our estimated delta before dropping dd_lock */ 1213 dd->dd_tempreserved[txgidx] += asize; 1214 1215 parent_rsrv = parent_delta(dd, used_on_disk + est_inflight, 1216 asize - ref_rsrv); 1217 mutex_exit(&dd->dd_lock); 1218 1219 tr = kmem_zalloc(sizeof (struct tempreserve), KM_SLEEP); 1220 tr->tr_ds = dd; 1221 tr->tr_size = asize; 1222 list_insert_tail(tr_list, tr); 1223 1224 /* see if it's OK with our parent */ 1225 if (dd->dd_parent && parent_rsrv) { 1226 boolean_t ismos = (dsl_dir_phys(dd)->dd_head_dataset_obj == 0); 1227 1228 return (dsl_dir_tempreserve_impl(dd->dd_parent, 1229 parent_rsrv, netfree, ismos, TRUE, tr_list, tx, FALSE)); 1230 } else { 1231 return (0); 1232 } 1233} 1234 1235/* 1236 * Reserve space in this dsl_dir, to be used in this tx's txg. 1237 * After the space has been dirtied (and dsl_dir_willuse_space() 1238 * has been called), the reservation should be canceled, using 1239 * dsl_dir_tempreserve_clear(). 1240 */ 1241int 1242dsl_dir_tempreserve_space(dsl_dir_t *dd, uint64_t lsize, uint64_t asize, 1243 uint64_t fsize, uint64_t usize, void **tr_cookiep, dmu_tx_t *tx) 1244{ 1245 int err; 1246 list_t *tr_list; 1247 1248 if (asize == 0) { 1249 *tr_cookiep = NULL; 1250 return (0); 1251 } 1252 1253 tr_list = kmem_alloc(sizeof (list_t), KM_SLEEP); 1254 list_create(tr_list, sizeof (struct tempreserve), 1255 offsetof(struct tempreserve, tr_node)); 1256 ASSERT3S(asize, >, 0); 1257 ASSERT3S(fsize, >=, 0); 1258 1259 err = arc_tempreserve_space(lsize, tx->tx_txg); 1260 if (err == 0) { 1261 struct tempreserve *tr; 1262 1263 tr = kmem_zalloc(sizeof (struct tempreserve), KM_SLEEP); 1264 tr->tr_size = lsize; 1265 list_insert_tail(tr_list, tr); 1266 } else { 1267 if (err == EAGAIN) { 1268 /* 1269 * If arc_memory_throttle() detected that pageout 1270 * is running and we are low on memory, we delay new 1271 * non-pageout transactions to give pageout an 1272 * advantage. 1273 * 1274 * It is unfortunate to be delaying while the caller's 1275 * locks are held. 1276 */ 1277 txg_delay(dd->dd_pool, tx->tx_txg, 1278 MSEC2NSEC(10), MSEC2NSEC(10)); 1279 err = SET_ERROR(ERESTART); 1280 } 1281 } 1282 1283 if (err == 0) { 1284 err = dsl_dir_tempreserve_impl(dd, asize, fsize >= asize, 1285 FALSE, asize > usize, tr_list, tx, TRUE); 1286 } 1287 1288 if (err != 0) 1289 dsl_dir_tempreserve_clear(tr_list, tx); 1290 else 1291 *tr_cookiep = tr_list; 1292 1293 return (err); 1294} 1295 1296/* 1297 * Clear a temporary reservation that we previously made with 1298 * dsl_dir_tempreserve_space(). 1299 */ 1300void 1301dsl_dir_tempreserve_clear(void *tr_cookie, dmu_tx_t *tx) 1302{ 1303 int txgidx = tx->tx_txg & TXG_MASK; 1304 list_t *tr_list = tr_cookie; 1305 struct tempreserve *tr; 1306 1307 ASSERT3U(tx->tx_txg, !=, 0); 1308 1309 if (tr_cookie == NULL) 1310 return; 1311 1312 while ((tr = list_head(tr_list)) != NULL) { 1313 if (tr->tr_ds) { 1314 mutex_enter(&tr->tr_ds->dd_lock); 1315 ASSERT3U(tr->tr_ds->dd_tempreserved[txgidx], >=, 1316 tr->tr_size); 1317 tr->tr_ds->dd_tempreserved[txgidx] -= tr->tr_size; 1318 mutex_exit(&tr->tr_ds->dd_lock); 1319 } else { 1320 arc_tempreserve_clear(tr->tr_size); 1321 } 1322 list_remove(tr_list, tr); 1323 kmem_free(tr, sizeof (struct tempreserve)); 1324 } 1325 1326 kmem_free(tr_list, sizeof (list_t)); 1327} 1328 1329/* 1330 * This should be called from open context when we think we're going to write 1331 * or free space, for example when dirtying data. Be conservative; it's okay 1332 * to write less space or free more, but we don't want to write more or free 1333 * less than the amount specified. 1334 */ 1335void 1336dsl_dir_willuse_space(dsl_dir_t *dd, int64_t space, dmu_tx_t *tx) 1337{ 1338 int64_t parent_space; 1339 uint64_t est_used; 1340 1341 mutex_enter(&dd->dd_lock); 1342 if (space > 0) 1343 dd->dd_space_towrite[tx->tx_txg & TXG_MASK] += space; 1344 1345 est_used = dsl_dir_space_towrite(dd) + dsl_dir_phys(dd)->dd_used_bytes; 1346 parent_space = parent_delta(dd, est_used, space); 1347 mutex_exit(&dd->dd_lock); 1348 1349 /* Make sure that we clean up dd_space_to* */ 1350 dsl_dir_dirty(dd, tx); 1351 1352 /* XXX this is potentially expensive and unnecessary... */ 1353 if (parent_space && dd->dd_parent) 1354 dsl_dir_willuse_space(dd->dd_parent, parent_space, tx); 1355} 1356 1357/* call from syncing context when we actually write/free space for this dd */ 1358void 1359dsl_dir_diduse_space(dsl_dir_t *dd, dd_used_t type, 1360 int64_t used, int64_t compressed, int64_t uncompressed, dmu_tx_t *tx) 1361{ 1362 int64_t accounted_delta; 1363 1364 /* 1365 * dsl_dataset_set_refreservation_sync_impl() calls this with 1366 * dd_lock held, so that it can atomically update 1367 * ds->ds_reserved and the dsl_dir accounting, so that 1368 * dsl_dataset_check_quota() can see dataset and dir accounting 1369 * consistently. 1370 */ 1371 boolean_t needlock = !MUTEX_HELD(&dd->dd_lock); 1372 1373 ASSERT(dmu_tx_is_syncing(tx)); 1374 ASSERT(type < DD_USED_NUM); 1375 1376 dmu_buf_will_dirty(dd->dd_dbuf, tx); 1377 1378 if (needlock) 1379 mutex_enter(&dd->dd_lock); 1380 accounted_delta = 1381 parent_delta(dd, dsl_dir_phys(dd)->dd_used_bytes, used); 1382 ASSERT(used >= 0 || dsl_dir_phys(dd)->dd_used_bytes >= -used); 1383 ASSERT(compressed >= 0 || 1384 dsl_dir_phys(dd)->dd_compressed_bytes >= -compressed); 1385 ASSERT(uncompressed >= 0 || 1386 dsl_dir_phys(dd)->dd_uncompressed_bytes >= -uncompressed); 1387 dsl_dir_phys(dd)->dd_used_bytes += used; 1388 dsl_dir_phys(dd)->dd_uncompressed_bytes += uncompressed; 1389 dsl_dir_phys(dd)->dd_compressed_bytes += compressed; 1390 1391 if (dsl_dir_phys(dd)->dd_flags & DD_FLAG_USED_BREAKDOWN) { 1392 ASSERT(used > 0 || 1393 dsl_dir_phys(dd)->dd_used_breakdown[type] >= -used); 1394 dsl_dir_phys(dd)->dd_used_breakdown[type] += used; 1395#ifdef DEBUG 1396 dd_used_t t; 1397 uint64_t u = 0; 1398 for (t = 0; t < DD_USED_NUM; t++) 1399 u += dsl_dir_phys(dd)->dd_used_breakdown[t]; 1400 ASSERT3U(u, ==, dsl_dir_phys(dd)->dd_used_bytes); 1401#endif 1402 } 1403 if (needlock) 1404 mutex_exit(&dd->dd_lock); 1405 1406 if (dd->dd_parent != NULL) { 1407 dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD, 1408 accounted_delta, compressed, uncompressed, tx); 1409 dsl_dir_transfer_space(dd->dd_parent, 1410 used - accounted_delta, 1411 DD_USED_CHILD_RSRV, DD_USED_CHILD, NULL); 1412 } 1413} 1414 1415void 1416dsl_dir_transfer_space(dsl_dir_t *dd, int64_t delta, 1417 dd_used_t oldtype, dd_used_t newtype, dmu_tx_t *tx) 1418{ 1419 ASSERT(tx == NULL || dmu_tx_is_syncing(tx)); 1420 ASSERT(oldtype < DD_USED_NUM); 1421 ASSERT(newtype < DD_USED_NUM); 1422 1423 if (delta == 0 || 1424 !(dsl_dir_phys(dd)->dd_flags & DD_FLAG_USED_BREAKDOWN)) 1425 return; 1426 1427 if (tx != NULL) 1428 dmu_buf_will_dirty(dd->dd_dbuf, tx); 1429 mutex_enter(&dd->dd_lock); 1430 ASSERT(delta > 0 ? 1431 dsl_dir_phys(dd)->dd_used_breakdown[oldtype] >= delta : 1432 dsl_dir_phys(dd)->dd_used_breakdown[newtype] >= -delta); 1433 ASSERT(dsl_dir_phys(dd)->dd_used_bytes >= ABS(delta)); 1434 dsl_dir_phys(dd)->dd_used_breakdown[oldtype] -= delta; 1435 dsl_dir_phys(dd)->dd_used_breakdown[newtype] += delta; 1436 mutex_exit(&dd->dd_lock); 1437} 1438 1439typedef struct dsl_dir_set_qr_arg { 1440 const char *ddsqra_name; 1441 zprop_source_t ddsqra_source; 1442 uint64_t ddsqra_value; 1443} dsl_dir_set_qr_arg_t; 1444 1445static int 1446dsl_dir_set_quota_check(void *arg, dmu_tx_t *tx) 1447{ 1448 dsl_dir_set_qr_arg_t *ddsqra = arg; 1449 dsl_pool_t *dp = dmu_tx_pool(tx); 1450 dsl_dataset_t *ds; 1451 int error; 1452 uint64_t towrite, newval; 1453 1454 error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds); 1455 if (error != 0) 1456 return (error); 1457 1458 error = dsl_prop_predict(ds->ds_dir, "quota", 1459 ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval); 1460 if (error != 0) { 1461 dsl_dataset_rele(ds, FTAG); 1462 return (error); 1463 } 1464 1465 if (newval == 0) { 1466 dsl_dataset_rele(ds, FTAG); 1467 return (0); 1468 } 1469 1470 mutex_enter(&ds->ds_dir->dd_lock); 1471 /* 1472 * If we are doing the preliminary check in open context, and 1473 * there are pending changes, then don't fail it, since the 1474 * pending changes could under-estimate the amount of space to be 1475 * freed up. 1476 */ 1477 towrite = dsl_dir_space_towrite(ds->ds_dir); 1478 if ((dmu_tx_is_syncing(tx) || towrite == 0) && 1479 (newval < dsl_dir_phys(ds->ds_dir)->dd_reserved || 1480 newval < dsl_dir_phys(ds->ds_dir)->dd_used_bytes + towrite)) { 1481 error = SET_ERROR(ENOSPC); 1482 } 1483 mutex_exit(&ds->ds_dir->dd_lock); 1484 dsl_dataset_rele(ds, FTAG); 1485 return (error); 1486} 1487 1488static void 1489dsl_dir_set_quota_sync(void *arg, dmu_tx_t *tx) 1490{ 1491 dsl_dir_set_qr_arg_t *ddsqra = arg; 1492 dsl_pool_t *dp = dmu_tx_pool(tx); 1493 dsl_dataset_t *ds; 1494 uint64_t newval; 1495 1496 VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds)); 1497 1498 if (spa_version(dp->dp_spa) >= SPA_VERSION_RECVD_PROPS) { 1499 dsl_prop_set_sync_impl(ds, zfs_prop_to_name(ZFS_PROP_QUOTA), 1500 ddsqra->ddsqra_source, sizeof (ddsqra->ddsqra_value), 1, 1501 &ddsqra->ddsqra_value, tx); 1502 1503 VERIFY0(dsl_prop_get_int_ds(ds, 1504 zfs_prop_to_name(ZFS_PROP_QUOTA), &newval)); 1505 } else { 1506 newval = ddsqra->ddsqra_value; 1507 spa_history_log_internal_ds(ds, "set", tx, "%s=%lld", 1508 zfs_prop_to_name(ZFS_PROP_QUOTA), (longlong_t)newval); 1509 } 1510 1511 dmu_buf_will_dirty(ds->ds_dir->dd_dbuf, tx); 1512 mutex_enter(&ds->ds_dir->dd_lock); 1513 dsl_dir_phys(ds->ds_dir)->dd_quota = newval; 1514 mutex_exit(&ds->ds_dir->dd_lock); 1515 dsl_dataset_rele(ds, FTAG); 1516} 1517 1518int 1519dsl_dir_set_quota(const char *ddname, zprop_source_t source, uint64_t quota) 1520{ 1521 dsl_dir_set_qr_arg_t ddsqra; 1522 1523 ddsqra.ddsqra_name = ddname; 1524 ddsqra.ddsqra_source = source; 1525 ddsqra.ddsqra_value = quota; 1526 1527 return (dsl_sync_task(ddname, dsl_dir_set_quota_check, 1528 dsl_dir_set_quota_sync, &ddsqra, 0, ZFS_SPACE_CHECK_NONE)); 1529} 1530 1531int 1532dsl_dir_set_reservation_check(void *arg, dmu_tx_t *tx) 1533{ 1534 dsl_dir_set_qr_arg_t *ddsqra = arg; 1535 dsl_pool_t *dp = dmu_tx_pool(tx); 1536 dsl_dataset_t *ds; 1537 dsl_dir_t *dd; 1538 uint64_t newval, used, avail; 1539 int error; 1540 1541 error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds); 1542 if (error != 0) 1543 return (error); 1544 dd = ds->ds_dir; 1545 1546 /* 1547 * If we are doing the preliminary check in open context, the 1548 * space estimates may be inaccurate. 1549 */ 1550 if (!dmu_tx_is_syncing(tx)) { 1551 dsl_dataset_rele(ds, FTAG); 1552 return (0); 1553 } 1554 1555 error = dsl_prop_predict(ds->ds_dir, 1556 zfs_prop_to_name(ZFS_PROP_RESERVATION), 1557 ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval); 1558 if (error != 0) { 1559 dsl_dataset_rele(ds, FTAG); 1560 return (error); 1561 } 1562 1563 mutex_enter(&dd->dd_lock); 1564 used = dsl_dir_phys(dd)->dd_used_bytes; 1565 mutex_exit(&dd->dd_lock); 1566 1567 if (dd->dd_parent) { 1568 avail = dsl_dir_space_available(dd->dd_parent, 1569 NULL, 0, FALSE); 1570 } else { 1571 avail = dsl_pool_adjustedsize(dd->dd_pool, B_FALSE) - used; 1572 } 1573 1574 if (MAX(used, newval) > MAX(used, dsl_dir_phys(dd)->dd_reserved)) { 1575 uint64_t delta = MAX(used, newval) - 1576 MAX(used, dsl_dir_phys(dd)->dd_reserved); 1577 1578 if (delta > avail || 1579 (dsl_dir_phys(dd)->dd_quota > 0 && 1580 newval > dsl_dir_phys(dd)->dd_quota)) 1581 error = SET_ERROR(ENOSPC); 1582 } 1583 1584 dsl_dataset_rele(ds, FTAG); 1585 return (error); 1586} 1587 1588void 1589dsl_dir_set_reservation_sync_impl(dsl_dir_t *dd, uint64_t value, dmu_tx_t *tx) 1590{ 1591 uint64_t used; 1592 int64_t delta; 1593 1594 dmu_buf_will_dirty(dd->dd_dbuf, tx); 1595 1596 mutex_enter(&dd->dd_lock); 1597 used = dsl_dir_phys(dd)->dd_used_bytes; 1598 delta = MAX(used, value) - MAX(used, dsl_dir_phys(dd)->dd_reserved); 1599 dsl_dir_phys(dd)->dd_reserved = value; 1600 1601 if (dd->dd_parent != NULL) { 1602 /* Roll up this additional usage into our ancestors */ 1603 dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD_RSRV, 1604 delta, 0, 0, tx); 1605 } 1606 mutex_exit(&dd->dd_lock); 1607} 1608 1609static void 1610dsl_dir_set_reservation_sync(void *arg, dmu_tx_t *tx) 1611{ 1612 dsl_dir_set_qr_arg_t *ddsqra = arg; 1613 dsl_pool_t *dp = dmu_tx_pool(tx); 1614 dsl_dataset_t *ds; 1615 uint64_t newval; 1616 1617 VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds)); 1618 1619 if (spa_version(dp->dp_spa) >= SPA_VERSION_RECVD_PROPS) { 1620 dsl_prop_set_sync_impl(ds, 1621 zfs_prop_to_name(ZFS_PROP_RESERVATION), 1622 ddsqra->ddsqra_source, sizeof (ddsqra->ddsqra_value), 1, 1623 &ddsqra->ddsqra_value, tx); 1624 1625 VERIFY0(dsl_prop_get_int_ds(ds, 1626 zfs_prop_to_name(ZFS_PROP_RESERVATION), &newval)); 1627 } else { 1628 newval = ddsqra->ddsqra_value; 1629 spa_history_log_internal_ds(ds, "set", tx, "%s=%lld", 1630 zfs_prop_to_name(ZFS_PROP_RESERVATION), 1631 (longlong_t)newval); 1632 } 1633 1634 dsl_dir_set_reservation_sync_impl(ds->ds_dir, newval, tx); 1635 dsl_dataset_rele(ds, FTAG); 1636} 1637 1638int 1639dsl_dir_set_reservation(const char *ddname, zprop_source_t source, 1640 uint64_t reservation) 1641{ 1642 dsl_dir_set_qr_arg_t ddsqra; 1643 1644 ddsqra.ddsqra_name = ddname; 1645 ddsqra.ddsqra_source = source; 1646 ddsqra.ddsqra_value = reservation; 1647 1648 return (dsl_sync_task(ddname, dsl_dir_set_reservation_check, 1649 dsl_dir_set_reservation_sync, &ddsqra, 0, ZFS_SPACE_CHECK_NONE)); 1650} 1651 1652static dsl_dir_t * 1653closest_common_ancestor(dsl_dir_t *ds1, dsl_dir_t *ds2) 1654{ 1655 for (; ds1; ds1 = ds1->dd_parent) { 1656 dsl_dir_t *dd; 1657 for (dd = ds2; dd; dd = dd->dd_parent) { 1658 if (ds1 == dd) 1659 return (dd); 1660 } 1661 } 1662 return (NULL); 1663} 1664 1665/* 1666 * If delta is applied to dd, how much of that delta would be applied to 1667 * ancestor? Syncing context only. 1668 */ 1669static int64_t 1670would_change(dsl_dir_t *dd, int64_t delta, dsl_dir_t *ancestor) 1671{ 1672 if (dd == ancestor) 1673 return (delta); 1674 1675 mutex_enter(&dd->dd_lock); 1676 delta = parent_delta(dd, dsl_dir_phys(dd)->dd_used_bytes, delta); 1677 mutex_exit(&dd->dd_lock); 1678 return (would_change(dd->dd_parent, delta, ancestor)); 1679} 1680 1681typedef struct dsl_dir_rename_arg { 1682 const char *ddra_oldname; 1683 const char *ddra_newname; 1684 cred_t *ddra_cred; 1685} dsl_dir_rename_arg_t; 1686 1687/* ARGSUSED */ 1688static int 1689dsl_valid_rename(dsl_pool_t *dp, dsl_dataset_t *ds, void *arg) 1690{ 1691 int *deltap = arg; 1692 char namebuf[MAXNAMELEN]; 1693 1694 dsl_dataset_name(ds, namebuf); 1695 1696 if (strlen(namebuf) + *deltap >= MAXNAMELEN) 1697 return (SET_ERROR(ENAMETOOLONG)); 1698 return (0); 1699} 1700 1701static int 1702dsl_dir_rename_check(void *arg, dmu_tx_t *tx) 1703{ 1704 dsl_dir_rename_arg_t *ddra = arg; 1705 dsl_pool_t *dp = dmu_tx_pool(tx); 1706 dsl_dir_t *dd, *newparent; 1707 const char *mynewname; 1708 int error; 1709 int delta = strlen(ddra->ddra_newname) - strlen(ddra->ddra_oldname); 1710 1711 /* target dir should exist */ 1712 error = dsl_dir_hold(dp, ddra->ddra_oldname, FTAG, &dd, NULL); 1713 if (error != 0) 1714 return (error); 1715 1716 /* new parent should exist */ 1717 error = dsl_dir_hold(dp, ddra->ddra_newname, FTAG, 1718 &newparent, &mynewname); 1719 if (error != 0) { 1720 dsl_dir_rele(dd, FTAG); 1721 return (error); 1722 } 1723 1724 /* can't rename to different pool */ 1725 if (dd->dd_pool != newparent->dd_pool) { 1726 dsl_dir_rele(newparent, FTAG); 1727 dsl_dir_rele(dd, FTAG); 1728 return (SET_ERROR(EXDEV)); 1729 } 1730 1731 /* new name should not already exist */ 1732 if (mynewname == NULL) { 1733 dsl_dir_rele(newparent, FTAG); 1734 dsl_dir_rele(dd, FTAG); 1735 return (SET_ERROR(EEXIST)); 1736 } 1737 1738 /* if the name length is growing, validate child name lengths */ 1739 if (delta > 0) { 1740 error = dmu_objset_find_dp(dp, dd->dd_object, dsl_valid_rename, 1741 &delta, DS_FIND_CHILDREN | DS_FIND_SNAPSHOTS); 1742 if (error != 0) { 1743 dsl_dir_rele(newparent, FTAG); 1744 dsl_dir_rele(dd, FTAG); 1745 return (error); 1746 } 1747 } 1748 1749 if (dmu_tx_is_syncing(tx)) { 1750 if (spa_feature_is_active(dp->dp_spa, 1751 SPA_FEATURE_FS_SS_LIMIT)) { 1752 /* 1753 * Although this is the check function and we don't 1754 * normally make on-disk changes in check functions, 1755 * we need to do that here. 1756 * 1757 * Ensure this portion of the tree's counts have been 1758 * initialized in case the new parent has limits set. 1759 */ 1760 dsl_dir_init_fs_ss_count(dd, tx); 1761 } 1762 } 1763 1764 if (newparent != dd->dd_parent) { 1765 /* is there enough space? */ 1766 uint64_t myspace = 1767 MAX(dsl_dir_phys(dd)->dd_used_bytes, 1768 dsl_dir_phys(dd)->dd_reserved); 1769 objset_t *os = dd->dd_pool->dp_meta_objset; 1770 uint64_t fs_cnt = 0; 1771 uint64_t ss_cnt = 0; 1772 1773 if (dsl_dir_is_zapified(dd)) { 1774 int err; 1775 1776 err = zap_lookup(os, dd->dd_object, 1777 DD_FIELD_FILESYSTEM_COUNT, sizeof (fs_cnt), 1, 1778 &fs_cnt); 1779 if (err != ENOENT && err != 0) { 1780 dsl_dir_rele(newparent, FTAG); 1781 dsl_dir_rele(dd, FTAG); 1782 return (err); 1783 } 1784 1785 /* 1786 * have to add 1 for the filesystem itself that we're 1787 * moving 1788 */ 1789 fs_cnt++; 1790 1791 err = zap_lookup(os, dd->dd_object, 1792 DD_FIELD_SNAPSHOT_COUNT, sizeof (ss_cnt), 1, 1793 &ss_cnt); 1794 if (err != ENOENT && err != 0) { 1795 dsl_dir_rele(newparent, FTAG); 1796 dsl_dir_rele(dd, FTAG); 1797 return (err); 1798 } 1799 } 1800 1801 /* no rename into our descendant */ 1802 if (closest_common_ancestor(dd, newparent) == dd) { 1803 dsl_dir_rele(newparent, FTAG); 1804 dsl_dir_rele(dd, FTAG); 1805 return (SET_ERROR(EINVAL)); 1806 } 1807 1808 error = dsl_dir_transfer_possible(dd->dd_parent, 1809 newparent, fs_cnt, ss_cnt, myspace, ddra->ddra_cred); 1810 if (error != 0) { 1811 dsl_dir_rele(newparent, FTAG); 1812 dsl_dir_rele(dd, FTAG); 1813 return (error); 1814 } 1815 } 1816 1817 dsl_dir_rele(newparent, FTAG); 1818 dsl_dir_rele(dd, FTAG); 1819 return (0); 1820} 1821 1822static void 1823dsl_dir_rename_sync(void *arg, dmu_tx_t *tx) 1824{ 1825 dsl_dir_rename_arg_t *ddra = arg; 1826 dsl_pool_t *dp = dmu_tx_pool(tx); 1827 dsl_dir_t *dd, *newparent; 1828 const char *mynewname; 1829 int error; 1830 objset_t *mos = dp->dp_meta_objset; 1831 1832 VERIFY0(dsl_dir_hold(dp, ddra->ddra_oldname, FTAG, &dd, NULL)); 1833 VERIFY0(dsl_dir_hold(dp, ddra->ddra_newname, FTAG, &newparent, 1834 &mynewname)); 1835 1836 /* Log this before we change the name. */ 1837 spa_history_log_internal_dd(dd, "rename", tx, 1838 "-> %s", ddra->ddra_newname); 1839 1840 if (newparent != dd->dd_parent) { 1841 objset_t *os = dd->dd_pool->dp_meta_objset; 1842 uint64_t fs_cnt = 0; 1843 uint64_t ss_cnt = 0; 1844 1845 /* 1846 * We already made sure the dd counts were initialized in the 1847 * check function. 1848 */ 1849 if (spa_feature_is_active(dp->dp_spa, 1850 SPA_FEATURE_FS_SS_LIMIT)) { 1851 VERIFY0(zap_lookup(os, dd->dd_object, 1852 DD_FIELD_FILESYSTEM_COUNT, sizeof (fs_cnt), 1, 1853 &fs_cnt)); 1854 /* add 1 for the filesystem itself that we're moving */ 1855 fs_cnt++; 1856 1857 VERIFY0(zap_lookup(os, dd->dd_object, 1858 DD_FIELD_SNAPSHOT_COUNT, sizeof (ss_cnt), 1, 1859 &ss_cnt)); 1860 } 1861 1862 dsl_fs_ss_count_adjust(dd->dd_parent, -fs_cnt, 1863 DD_FIELD_FILESYSTEM_COUNT, tx); 1864 dsl_fs_ss_count_adjust(newparent, fs_cnt, 1865 DD_FIELD_FILESYSTEM_COUNT, tx); 1866 1867 dsl_fs_ss_count_adjust(dd->dd_parent, -ss_cnt, 1868 DD_FIELD_SNAPSHOT_COUNT, tx); 1869 dsl_fs_ss_count_adjust(newparent, ss_cnt, 1870 DD_FIELD_SNAPSHOT_COUNT, tx); 1871 1872 dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD, 1873 -dsl_dir_phys(dd)->dd_used_bytes, 1874 -dsl_dir_phys(dd)->dd_compressed_bytes, 1875 -dsl_dir_phys(dd)->dd_uncompressed_bytes, tx); 1876 dsl_dir_diduse_space(newparent, DD_USED_CHILD, 1877 dsl_dir_phys(dd)->dd_used_bytes, 1878 dsl_dir_phys(dd)->dd_compressed_bytes, 1879 dsl_dir_phys(dd)->dd_uncompressed_bytes, tx); 1880 1881 if (dsl_dir_phys(dd)->dd_reserved > 1882 dsl_dir_phys(dd)->dd_used_bytes) { 1883 uint64_t unused_rsrv = dsl_dir_phys(dd)->dd_reserved - 1884 dsl_dir_phys(dd)->dd_used_bytes; 1885 1886 dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD_RSRV, 1887 -unused_rsrv, 0, 0, tx); 1888 dsl_dir_diduse_space(newparent, DD_USED_CHILD_RSRV, 1889 unused_rsrv, 0, 0, tx); 1890 } 1891 } 1892 1893 dmu_buf_will_dirty(dd->dd_dbuf, tx); 1894 1895 /* remove from old parent zapobj */ 1896 error = zap_remove(mos, 1897 dsl_dir_phys(dd->dd_parent)->dd_child_dir_zapobj, 1898 dd->dd_myname, tx); 1899 ASSERT0(error); 1900 1901 (void) strcpy(dd->dd_myname, mynewname); 1902 dsl_dir_rele(dd->dd_parent, dd); 1903 dsl_dir_phys(dd)->dd_parent_obj = newparent->dd_object; 1904 VERIFY0(dsl_dir_hold_obj(dp, 1905 newparent->dd_object, NULL, dd, &dd->dd_parent)); 1906 1907 /* add to new parent zapobj */ 1908 VERIFY0(zap_add(mos, dsl_dir_phys(newparent)->dd_child_dir_zapobj, 1909 dd->dd_myname, 8, 1, &dd->dd_object, tx)); 1910 1911#ifdef __FreeBSD__ 1912#ifdef _KERNEL 1913 zfsvfs_update_fromname(ddra->ddra_oldname, ddra->ddra_newname); 1914 zvol_rename_minors(ddra->ddra_oldname, ddra->ddra_newname); 1915#endif 1916#endif 1917 1918 dsl_prop_notify_all(dd); 1919 1920 dsl_dir_rele(newparent, FTAG); 1921 dsl_dir_rele(dd, FTAG); 1922} 1923 1924int 1925dsl_dir_rename(const char *oldname, const char *newname) 1926{ 1927 dsl_dir_rename_arg_t ddra; 1928 1929 ddra.ddra_oldname = oldname; 1930 ddra.ddra_newname = newname; 1931 ddra.ddra_cred = CRED(); 1932 1933 return (dsl_sync_task(oldname, 1934 dsl_dir_rename_check, dsl_dir_rename_sync, &ddra, 1935 3, ZFS_SPACE_CHECK_RESERVED)); 1936} 1937 1938int 1939dsl_dir_transfer_possible(dsl_dir_t *sdd, dsl_dir_t *tdd, 1940 uint64_t fs_cnt, uint64_t ss_cnt, uint64_t space, cred_t *cr) 1941{ 1942 dsl_dir_t *ancestor; 1943 int64_t adelta; 1944 uint64_t avail; 1945 int err; 1946 1947 ancestor = closest_common_ancestor(sdd, tdd); 1948 adelta = would_change(sdd, -space, ancestor); 1949 avail = dsl_dir_space_available(tdd, ancestor, adelta, FALSE); 1950 if (avail < space) 1951 return (SET_ERROR(ENOSPC)); 1952 1953 err = dsl_fs_ss_limit_check(tdd, fs_cnt, ZFS_PROP_FILESYSTEM_LIMIT, 1954 ancestor, cr); 1955 if (err != 0) 1956 return (err); 1957 err = dsl_fs_ss_limit_check(tdd, ss_cnt, ZFS_PROP_SNAPSHOT_LIMIT, 1958 ancestor, cr); 1959 if (err != 0) 1960 return (err); 1961 1962 return (0); 1963} 1964 1965timestruc_t 1966dsl_dir_snap_cmtime(dsl_dir_t *dd) 1967{ 1968 timestruc_t t; 1969 1970 mutex_enter(&dd->dd_lock); 1971 t = dd->dd_snap_cmtime; 1972 mutex_exit(&dd->dd_lock); 1973 1974 return (t); 1975} 1976 1977void 1978dsl_dir_snap_cmtime_update(dsl_dir_t *dd) 1979{ 1980 timestruc_t t; 1981 1982 gethrestime(&t); 1983 mutex_enter(&dd->dd_lock); 1984 dd->dd_snap_cmtime = t; 1985 mutex_exit(&dd->dd_lock); 1986} 1987 1988void 1989dsl_dir_zapify(dsl_dir_t *dd, dmu_tx_t *tx) 1990{ 1991 objset_t *mos = dd->dd_pool->dp_meta_objset; 1992 dmu_object_zapify(mos, dd->dd_object, DMU_OT_DSL_DIR, tx); 1993} 1994 1995boolean_t 1996dsl_dir_is_zapified(dsl_dir_t *dd) 1997{ 1998 dmu_object_info_t doi; 1999 2000 dmu_object_info_from_db(dd->dd_dbuf, &doi); 2001 return (doi.doi_type == DMU_OTN_ZAP_METADATA); 2002} 2003