1168404Spjd/* 2168404Spjd * CDDL HEADER START 3168404Spjd * 4168404Spjd * The contents of this file are subject to the terms of the 5168404Spjd * Common Development and Distribution License (the "License"). 6168404Spjd * You may not use this file except in compliance with the License. 7168404Spjd * 8168404Spjd * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 9168404Spjd * or http://www.opensolaris.org/os/licensing. 10168404Spjd * See the License for the specific language governing permissions 11168404Spjd * and limitations under the License. 12168404Spjd * 13168404Spjd * When distributing Covered Code, include this CDDL HEADER in each 14168404Spjd * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 15168404Spjd * If applicable, add the following below this CDDL HEADER, with the 16168404Spjd * fields enclosed by brackets "[]" replaced with your own identifying 17168404Spjd * information: Portions Copyright [yyyy] [name of copyright owner] 18168404Spjd * 19168404Spjd * CDDL HEADER END 20168404Spjd */ 21168404Spjd/* 22219089Spjd * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. 23226707Spjd * Copyright (c) 2011 Pawel Jakub Dawidek <pawel@dawidek.net>. 24226707Spjd * All rights reserved. 25269006Sdelphij * Copyright (c) 2012, 2014 by Delphix. All rights reserved. 26265744Sdelphij * Copyright (c) 2014 Joyent, Inc. All rights reserved. 27168404Spjd */ 28168404Spjd 29168404Spjd#include <sys/dmu.h> 30185029Spjd#include <sys/dmu_objset.h> 31168404Spjd#include <sys/dmu_tx.h> 32168404Spjd#include <sys/dsl_dataset.h> 33168404Spjd#include <sys/dsl_dir.h> 34168404Spjd#include <sys/dsl_prop.h> 35168404Spjd#include <sys/dsl_synctask.h> 36185029Spjd#include <sys/dsl_deleg.h> 37263390Sdelphij#include <sys/dmu_impl.h> 38168404Spjd#include <sys/spa.h> 39219089Spjd#include <sys/metaslab.h> 40168404Spjd#include <sys/zap.h> 41168404Spjd#include <sys/zio.h> 42168404Spjd#include <sys/arc.h> 43185029Spjd#include <sys/sunddi.h> 44219317Spjd#include <sys/zvol.h> 45226678Spjd#ifdef _KERNEL 46226676Spjd#include <sys/zfs_vfsops.h> 47226678Spjd#endif 48265744Sdelphij#include <sys/zfeature.h> 49265744Sdelphij#include <sys/policy.h> 50265744Sdelphij#include <sys/zfs_znode.h> 51168404Spjd#include "zfs_namecheck.h" 52265744Sdelphij#include "zfs_prop.h" 53168404Spjd 54265744Sdelphij/* 55265744Sdelphij * Filesystem and Snapshot Limits 56265744Sdelphij * ------------------------------ 57265744Sdelphij * 58265744Sdelphij * These limits are used to restrict the number of filesystems and/or snapshots 59265744Sdelphij * that can be created at a given level in the tree or below. A typical 60265744Sdelphij * use-case is with a delegated dataset where the administrator wants to ensure 61265744Sdelphij * that a user within the zone is not creating too many additional filesystems 62265744Sdelphij * or snapshots, even though they're not exceeding their space quota. 63265744Sdelphij * 64265744Sdelphij * The filesystem and snapshot counts are stored as extensible properties. This 65265744Sdelphij * capability is controlled by a feature flag and must be enabled to be used. 66265744Sdelphij * Once enabled, the feature is not active until the first limit is set. At 67265744Sdelphij * that point, future operations to create/destroy filesystems or snapshots 68265744Sdelphij * will validate and update the counts. 69265744Sdelphij * 70265744Sdelphij * Because the count properties will not exist before the feature is active, 71265744Sdelphij * the counts are updated when a limit is first set on an uninitialized 72265744Sdelphij * dsl_dir node in the tree (The filesystem/snapshot count on a node includes 73265744Sdelphij * all of the nested filesystems/snapshots. Thus, a new leaf node has a 74265744Sdelphij * filesystem count of 0 and a snapshot count of 0. Non-existent filesystem and 75265744Sdelphij * snapshot count properties on a node indicate uninitialized counts on that 76265744Sdelphij * node.) When first setting a limit on an uninitialized node, the code starts 77265744Sdelphij * at the filesystem with the new limit and descends into all sub-filesystems 78265744Sdelphij * to add the count properties. 79265744Sdelphij * 80265744Sdelphij * In practice this is lightweight since a limit is typically set when the 81265744Sdelphij * filesystem is created and thus has no children. Once valid, changing the 82265744Sdelphij * limit value won't require a re-traversal since the counts are already valid. 83265744Sdelphij * When recursively fixing the counts, if a node with a limit is encountered 84265744Sdelphij * during the descent, the counts are known to be valid and there is no need to 85265744Sdelphij * descend into that filesystem's children. The counts on filesystems above the 86265744Sdelphij * one with the new limit will still be uninitialized, unless a limit is 87265744Sdelphij * eventually set on one of those filesystems. The counts are always recursively 88265744Sdelphij * updated when a limit is set on a dataset, unless there is already a limit. 89265744Sdelphij * When a new limit value is set on a filesystem with an existing limit, it is 90265744Sdelphij * possible for the new limit to be less than the current count at that level 91265744Sdelphij * since a user who can change the limit is also allowed to exceed the limit. 92265744Sdelphij * 93265744Sdelphij * Once the feature is active, then whenever a filesystem or snapshot is 94265744Sdelphij * created, the code recurses up the tree, validating the new count against the 95265744Sdelphij * limit at each initialized level. In practice, most levels will not have a 96265744Sdelphij * limit set. If there is a limit at any initialized level up the tree, the 97265744Sdelphij * check must pass or the creation will fail. Likewise, when a filesystem or 98265744Sdelphij * snapshot is destroyed, the counts are recursively adjusted all the way up 99265744Sdelphij * the initizized nodes in the tree. Renaming a filesystem into different point 100265744Sdelphij * in the tree will first validate, then update the counts on each branch up to 101265744Sdelphij * the common ancestor. A receive will also validate the counts and then update 102265744Sdelphij * them. 103265744Sdelphij * 104265744Sdelphij * An exception to the above behavior is that the limit is not enforced if the 105265744Sdelphij * user has permission to modify the limit. This is primarily so that 106265744Sdelphij * recursive snapshots in the global zone always work. We want to prevent a 107265744Sdelphij * denial-of-service in which a lower level delegated dataset could max out its 108265744Sdelphij * limit and thus block recursive snapshots from being taken in the global zone. 109265744Sdelphij * Because of this, it is possible for the snapshot count to be over the limit 110265744Sdelphij * and snapshots taken in the global zone could cause a lower level dataset to 111265744Sdelphij * hit or exceed its limit. The administrator taking the global zone recursive 112265744Sdelphij * snapshot should be aware of this side-effect and behave accordingly. 113265744Sdelphij * For consistency, the filesystem limit is also not enforced if the user can 114265744Sdelphij * modify the limit. 115265744Sdelphij * 116265744Sdelphij * The filesystem and snapshot limits are validated by dsl_fs_ss_limit_check() 117265744Sdelphij * and updated by dsl_fs_ss_count_adjust(). A new limit value is setup in 118265744Sdelphij * dsl_dir_activate_fs_ss_limit() and the counts are adjusted, if necessary, by 119265744Sdelphij * dsl_dir_init_fs_ss_count(). 120265744Sdelphij * 121265744Sdelphij * There is a special case when we receive a filesystem that already exists. In 122265744Sdelphij * this case a temporary clone name of %X is created (see dmu_recv_begin). We 123265744Sdelphij * never update the filesystem counts for temporary clones. 124265744Sdelphij * 125265744Sdelphij * Likewise, we do not update the snapshot counts for temporary snapshots, 126265744Sdelphij * such as those created by zfs diff. 127265744Sdelphij */ 128265744Sdelphij 129185029Spjdstatic uint64_t dsl_dir_space_towrite(dsl_dir_t *dd); 130168404Spjd 131168404Spjd/* ARGSUSED */ 132168404Spjdstatic void 133168404Spjddsl_dir_evict(dmu_buf_t *db, void *arg) 134168404Spjd{ 135168404Spjd dsl_dir_t *dd = arg; 136168404Spjd dsl_pool_t *dp = dd->dd_pool; 137168404Spjd int t; 138168404Spjd 139168404Spjd for (t = 0; t < TXG_SIZE; t++) { 140168404Spjd ASSERT(!txg_list_member(&dp->dp_dirty_dirs, dd, t)); 141168404Spjd ASSERT(dd->dd_tempreserved[t] == 0); 142168404Spjd ASSERT(dd->dd_space_towrite[t] == 0); 143168404Spjd } 144168404Spjd 145168404Spjd if (dd->dd_parent) 146248571Smm dsl_dir_rele(dd->dd_parent, dd); 147168404Spjd 148168404Spjd spa_close(dd->dd_pool->dp_spa, dd); 149168404Spjd 150168404Spjd /* 151219089Spjd * The props callback list should have been cleaned up by 152219089Spjd * objset_evict(). 153168404Spjd */ 154168404Spjd list_destroy(&dd->dd_prop_cbs); 155168404Spjd mutex_destroy(&dd->dd_lock); 156168404Spjd kmem_free(dd, sizeof (dsl_dir_t)); 157168404Spjd} 158168404Spjd 159168404Spjdint 160248571Smmdsl_dir_hold_obj(dsl_pool_t *dp, uint64_t ddobj, 161168404Spjd const char *tail, void *tag, dsl_dir_t **ddp) 162168404Spjd{ 163168404Spjd dmu_buf_t *dbuf; 164168404Spjd dsl_dir_t *dd; 165168404Spjd int err; 166168404Spjd 167248571Smm ASSERT(dsl_pool_config_held(dp)); 168168404Spjd 169168404Spjd err = dmu_bonus_hold(dp->dp_meta_objset, ddobj, tag, &dbuf); 170248571Smm if (err != 0) 171168404Spjd return (err); 172168404Spjd dd = dmu_buf_get_user(dbuf); 173168404Spjd#ifdef ZFS_DEBUG 174168404Spjd { 175168404Spjd dmu_object_info_t doi; 176168404Spjd dmu_object_info_from_db(dbuf, &doi); 177263390Sdelphij ASSERT3U(doi.doi_bonus_type, ==, DMU_OT_DSL_DIR); 178185029Spjd ASSERT3U(doi.doi_bonus_size, >=, sizeof (dsl_dir_phys_t)); 179168404Spjd } 180168404Spjd#endif 181168404Spjd if (dd == NULL) { 182168404Spjd dsl_dir_t *winner; 183168404Spjd 184168404Spjd dd = kmem_zalloc(sizeof (dsl_dir_t), KM_SLEEP); 185168404Spjd dd->dd_object = ddobj; 186168404Spjd dd->dd_dbuf = dbuf; 187168404Spjd dd->dd_pool = dp; 188168404Spjd dd->dd_phys = dbuf->db_data; 189168404Spjd mutex_init(&dd->dd_lock, NULL, MUTEX_DEFAULT, NULL); 190168404Spjd 191168404Spjd list_create(&dd->dd_prop_cbs, sizeof (dsl_prop_cb_record_t), 192168404Spjd offsetof(dsl_prop_cb_record_t, cbr_node)); 193168404Spjd 194219089Spjd dsl_dir_snap_cmtime_update(dd); 195219089Spjd 196168404Spjd if (dd->dd_phys->dd_parent_obj) { 197248571Smm err = dsl_dir_hold_obj(dp, dd->dd_phys->dd_parent_obj, 198168404Spjd NULL, dd, &dd->dd_parent); 199248571Smm if (err != 0) 200185029Spjd goto errout; 201168404Spjd if (tail) { 202168404Spjd#ifdef ZFS_DEBUG 203168404Spjd uint64_t foundobj; 204168404Spjd 205168404Spjd err = zap_lookup(dp->dp_meta_objset, 206185029Spjd dd->dd_parent->dd_phys->dd_child_dir_zapobj, 207168404Spjd tail, sizeof (foundobj), 1, &foundobj); 208168404Spjd ASSERT(err || foundobj == ddobj); 209168404Spjd#endif 210168404Spjd (void) strcpy(dd->dd_myname, tail); 211168404Spjd } else { 212168404Spjd err = zap_value_search(dp->dp_meta_objset, 213185029Spjd dd->dd_parent->dd_phys->dd_child_dir_zapobj, 214185029Spjd ddobj, 0, dd->dd_myname); 215168404Spjd } 216248571Smm if (err != 0) 217185029Spjd goto errout; 218168404Spjd } else { 219168404Spjd (void) strcpy(dd->dd_myname, spa_name(dp->dp_spa)); 220168404Spjd } 221168404Spjd 222219089Spjd if (dsl_dir_is_clone(dd)) { 223219089Spjd dmu_buf_t *origin_bonus; 224219089Spjd dsl_dataset_phys_t *origin_phys; 225219089Spjd 226219089Spjd /* 227219089Spjd * We can't open the origin dataset, because 228219089Spjd * that would require opening this dsl_dir. 229219089Spjd * Just look at its phys directly instead. 230219089Spjd */ 231219089Spjd err = dmu_bonus_hold(dp->dp_meta_objset, 232219089Spjd dd->dd_phys->dd_origin_obj, FTAG, &origin_bonus); 233248571Smm if (err != 0) 234219089Spjd goto errout; 235219089Spjd origin_phys = origin_bonus->db_data; 236219089Spjd dd->dd_origin_txg = 237219089Spjd origin_phys->ds_creation_txg; 238219089Spjd dmu_buf_rele(origin_bonus, FTAG); 239219089Spjd } 240219089Spjd 241168404Spjd winner = dmu_buf_set_user_ie(dbuf, dd, &dd->dd_phys, 242168404Spjd dsl_dir_evict); 243168404Spjd if (winner) { 244168404Spjd if (dd->dd_parent) 245248571Smm dsl_dir_rele(dd->dd_parent, dd); 246168404Spjd mutex_destroy(&dd->dd_lock); 247168404Spjd kmem_free(dd, sizeof (dsl_dir_t)); 248168404Spjd dd = winner; 249168404Spjd } else { 250168404Spjd spa_open_ref(dp->dp_spa, dd); 251168404Spjd } 252168404Spjd } 253168404Spjd 254168404Spjd /* 255168404Spjd * The dsl_dir_t has both open-to-close and instantiate-to-evict 256168404Spjd * holds on the spa. We need the open-to-close holds because 257168404Spjd * otherwise the spa_refcnt wouldn't change when we open a 258168404Spjd * dir which the spa also has open, so we could incorrectly 259168404Spjd * think it was OK to unload/export/destroy the pool. We need 260168404Spjd * the instantiate-to-evict hold because the dsl_dir_t has a 261168404Spjd * pointer to the dd_pool, which has a pointer to the spa_t. 262168404Spjd */ 263168404Spjd spa_open_ref(dp->dp_spa, tag); 264168404Spjd ASSERT3P(dd->dd_pool, ==, dp); 265168404Spjd ASSERT3U(dd->dd_object, ==, ddobj); 266168404Spjd ASSERT3P(dd->dd_dbuf, ==, dbuf); 267168404Spjd *ddp = dd; 268168404Spjd return (0); 269185029Spjd 270185029Spjderrout: 271185029Spjd if (dd->dd_parent) 272248571Smm dsl_dir_rele(dd->dd_parent, dd); 273185029Spjd mutex_destroy(&dd->dd_lock); 274185029Spjd kmem_free(dd, sizeof (dsl_dir_t)); 275185029Spjd dmu_buf_rele(dbuf, tag); 276185029Spjd return (err); 277168404Spjd} 278168404Spjd 279168404Spjdvoid 280248571Smmdsl_dir_rele(dsl_dir_t *dd, void *tag) 281168404Spjd{ 282168404Spjd dprintf_dd(dd, "%s\n", ""); 283168404Spjd spa_close(dd->dd_pool->dp_spa, tag); 284168404Spjd dmu_buf_rele(dd->dd_dbuf, tag); 285168404Spjd} 286168404Spjd 287168404Spjd/* buf must be long enough (MAXNAMELEN + strlen(MOS_DIR_NAME) + 1 should do) */ 288168404Spjdvoid 289168404Spjddsl_dir_name(dsl_dir_t *dd, char *buf) 290168404Spjd{ 291168404Spjd if (dd->dd_parent) { 292168404Spjd dsl_dir_name(dd->dd_parent, buf); 293168404Spjd (void) strcat(buf, "/"); 294168404Spjd } else { 295168404Spjd buf[0] = '\0'; 296168404Spjd } 297168404Spjd if (!MUTEX_HELD(&dd->dd_lock)) { 298168404Spjd /* 299168404Spjd * recursive mutex so that we can use 300168404Spjd * dprintf_dd() with dd_lock held 301168404Spjd */ 302168404Spjd mutex_enter(&dd->dd_lock); 303168404Spjd (void) strcat(buf, dd->dd_myname); 304168404Spjd mutex_exit(&dd->dd_lock); 305168404Spjd } else { 306168404Spjd (void) strcat(buf, dd->dd_myname); 307168404Spjd } 308168404Spjd} 309168404Spjd 310239620Smm/* Calculate name length, avoiding all the strcat calls of dsl_dir_name */ 311168404Spjdint 312168498Spjddsl_dir_namelen(dsl_dir_t *dd) 313168498Spjd{ 314168498Spjd int result = 0; 315168498Spjd 316168498Spjd if (dd->dd_parent) { 317168498Spjd /* parent's name + 1 for the "/" */ 318168498Spjd result = dsl_dir_namelen(dd->dd_parent) + 1; 319168498Spjd } 320168498Spjd 321168498Spjd if (!MUTEX_HELD(&dd->dd_lock)) { 322168498Spjd /* see dsl_dir_name */ 323168498Spjd mutex_enter(&dd->dd_lock); 324168498Spjd result += strlen(dd->dd_myname); 325168498Spjd mutex_exit(&dd->dd_lock); 326168498Spjd } else { 327168498Spjd result += strlen(dd->dd_myname); 328168498Spjd } 329168498Spjd 330168498Spjd return (result); 331168498Spjd} 332168498Spjd 333168404Spjdstatic int 334168404Spjdgetcomponent(const char *path, char *component, const char **nextp) 335168404Spjd{ 336168404Spjd char *p; 337248571Smm 338209962Smm if ((path == NULL) || (path[0] == '\0')) 339249195Smm return (SET_ERROR(ENOENT)); 340168404Spjd /* This would be a good place to reserve some namespace... */ 341168404Spjd p = strpbrk(path, "/@"); 342168404Spjd if (p && (p[1] == '/' || p[1] == '@')) { 343168404Spjd /* two separators in a row */ 344249195Smm return (SET_ERROR(EINVAL)); 345168404Spjd } 346168404Spjd if (p == NULL || p == path) { 347168404Spjd /* 348168404Spjd * if the first thing is an @ or /, it had better be an 349168404Spjd * @ and it had better not have any more ats or slashes, 350168404Spjd * and it had better have something after the @. 351168404Spjd */ 352168404Spjd if (p != NULL && 353168404Spjd (p[0] != '@' || strpbrk(path+1, "/@") || p[1] == '\0')) 354249195Smm return (SET_ERROR(EINVAL)); 355168404Spjd if (strlen(path) >= MAXNAMELEN) 356249195Smm return (SET_ERROR(ENAMETOOLONG)); 357168404Spjd (void) strcpy(component, path); 358168404Spjd p = NULL; 359168404Spjd } else if (p[0] == '/') { 360248571Smm if (p - path >= MAXNAMELEN) 361249195Smm return (SET_ERROR(ENAMETOOLONG)); 362168404Spjd (void) strncpy(component, path, p - path); 363248571Smm component[p - path] = '\0'; 364168404Spjd p++; 365168404Spjd } else if (p[0] == '@') { 366168404Spjd /* 367168404Spjd * if the next separator is an @, there better not be 368168404Spjd * any more slashes. 369168404Spjd */ 370168404Spjd if (strchr(path, '/')) 371249195Smm return (SET_ERROR(EINVAL)); 372248571Smm if (p - path >= MAXNAMELEN) 373249195Smm return (SET_ERROR(ENAMETOOLONG)); 374168404Spjd (void) strncpy(component, path, p - path); 375248571Smm component[p - path] = '\0'; 376168404Spjd } else { 377248571Smm panic("invalid p=%p", (void *)p); 378168404Spjd } 379168404Spjd *nextp = p; 380168404Spjd return (0); 381168404Spjd} 382168404Spjd 383168404Spjd/* 384248571Smm * Return the dsl_dir_t, and possibly the last component which couldn't 385248571Smm * be found in *tail. The name must be in the specified dsl_pool_t. This 386248571Smm * thread must hold the dp_config_rwlock for the pool. Returns NULL if the 387248571Smm * path is bogus, or if tail==NULL and we couldn't parse the whole name. 388248571Smm * (*tail)[0] == '@' means that the last component is a snapshot. 389168404Spjd */ 390168404Spjdint 391248571Smmdsl_dir_hold(dsl_pool_t *dp, const char *name, void *tag, 392168404Spjd dsl_dir_t **ddp, const char **tailp) 393168404Spjd{ 394168404Spjd char buf[MAXNAMELEN]; 395248571Smm const char *spaname, *next, *nextnext = NULL; 396168404Spjd int err; 397168404Spjd dsl_dir_t *dd; 398168404Spjd uint64_t ddobj; 399168404Spjd 400168404Spjd err = getcomponent(name, buf, &next); 401248571Smm if (err != 0) 402168404Spjd return (err); 403168404Spjd 404248571Smm /* Make sure the name is in the specified pool. */ 405248571Smm spaname = spa_name(dp->dp_spa); 406248571Smm if (strcmp(buf, spaname) != 0) 407249195Smm return (SET_ERROR(EINVAL)); 408168404Spjd 409248571Smm ASSERT(dsl_pool_config_held(dp)); 410168404Spjd 411248571Smm err = dsl_dir_hold_obj(dp, dp->dp_root_dir_obj, NULL, tag, &dd); 412248571Smm if (err != 0) { 413168404Spjd return (err); 414168404Spjd } 415168404Spjd 416168404Spjd while (next != NULL) { 417168404Spjd dsl_dir_t *child_ds; 418168404Spjd err = getcomponent(next, buf, &nextnext); 419248571Smm if (err != 0) 420168404Spjd break; 421168404Spjd ASSERT(next[0] != '\0'); 422168404Spjd if (next[0] == '@') 423168404Spjd break; 424168404Spjd dprintf("looking up %s in obj%lld\n", 425168404Spjd buf, dd->dd_phys->dd_child_dir_zapobj); 426168404Spjd 427168404Spjd err = zap_lookup(dp->dp_meta_objset, 428168404Spjd dd->dd_phys->dd_child_dir_zapobj, 429168404Spjd buf, sizeof (ddobj), 1, &ddobj); 430248571Smm if (err != 0) { 431168404Spjd if (err == ENOENT) 432168404Spjd err = 0; 433168404Spjd break; 434168404Spjd } 435168404Spjd 436248571Smm err = dsl_dir_hold_obj(dp, ddobj, buf, tag, &child_ds); 437248571Smm if (err != 0) 438168404Spjd break; 439248571Smm dsl_dir_rele(dd, tag); 440168404Spjd dd = child_ds; 441168404Spjd next = nextnext; 442168404Spjd } 443168404Spjd 444248571Smm if (err != 0) { 445248571Smm dsl_dir_rele(dd, tag); 446168404Spjd return (err); 447168404Spjd } 448168404Spjd 449168404Spjd /* 450168404Spjd * It's an error if there's more than one component left, or 451168404Spjd * tailp==NULL and there's any component left. 452168404Spjd */ 453168404Spjd if (next != NULL && 454168404Spjd (tailp == NULL || (nextnext && nextnext[0] != '\0'))) { 455168404Spjd /* bad path name */ 456248571Smm dsl_dir_rele(dd, tag); 457168404Spjd dprintf("next=%p (%s) tail=%p\n", next, next?next:"", tailp); 458249195Smm err = SET_ERROR(ENOENT); 459168404Spjd } 460248571Smm if (tailp != NULL) 461168404Spjd *tailp = next; 462168404Spjd *ddp = dd; 463168404Spjd return (err); 464168404Spjd} 465168404Spjd 466265744Sdelphij/* 467265744Sdelphij * If the counts are already initialized for this filesystem and its 468265744Sdelphij * descendants then do nothing, otherwise initialize the counts. 469265744Sdelphij * 470265744Sdelphij * The counts on this filesystem, and those below, may be uninitialized due to 471265744Sdelphij * either the use of a pre-existing pool which did not support the 472265744Sdelphij * filesystem/snapshot limit feature, or one in which the feature had not yet 473265744Sdelphij * been enabled. 474265744Sdelphij * 475265744Sdelphij * Recursively descend the filesystem tree and update the filesystem/snapshot 476265744Sdelphij * counts on each filesystem below, then update the cumulative count on the 477265744Sdelphij * current filesystem. If the filesystem already has a count set on it, 478265744Sdelphij * then we know that its counts, and the counts on the filesystems below it, 479265744Sdelphij * are already correct, so we don't have to update this filesystem. 480265744Sdelphij */ 481265744Sdelphijstatic void 482265744Sdelphijdsl_dir_init_fs_ss_count(dsl_dir_t *dd, dmu_tx_t *tx) 483265744Sdelphij{ 484265744Sdelphij uint64_t my_fs_cnt = 0; 485265744Sdelphij uint64_t my_ss_cnt = 0; 486265744Sdelphij dsl_pool_t *dp = dd->dd_pool; 487265744Sdelphij objset_t *os = dp->dp_meta_objset; 488265744Sdelphij zap_cursor_t *zc; 489265744Sdelphij zap_attribute_t *za; 490265744Sdelphij dsl_dataset_t *ds; 491265744Sdelphij 492267138Sdelphij ASSERT(spa_feature_is_active(dp->dp_spa, SPA_FEATURE_FS_SS_LIMIT)); 493265744Sdelphij ASSERT(dsl_pool_config_held(dp)); 494265744Sdelphij ASSERT(dmu_tx_is_syncing(tx)); 495265744Sdelphij 496265744Sdelphij dsl_dir_zapify(dd, tx); 497265744Sdelphij 498265744Sdelphij /* 499265744Sdelphij * If the filesystem count has already been initialized then we 500265744Sdelphij * don't need to recurse down any further. 501265744Sdelphij */ 502265744Sdelphij if (zap_contains(os, dd->dd_object, DD_FIELD_FILESYSTEM_COUNT) == 0) 503265744Sdelphij return; 504265744Sdelphij 505265744Sdelphij zc = kmem_alloc(sizeof (zap_cursor_t), KM_SLEEP); 506265744Sdelphij za = kmem_alloc(sizeof (zap_attribute_t), KM_SLEEP); 507265744Sdelphij 508265744Sdelphij /* Iterate my child dirs */ 509265744Sdelphij for (zap_cursor_init(zc, os, dd->dd_phys->dd_child_dir_zapobj); 510265744Sdelphij zap_cursor_retrieve(zc, za) == 0; zap_cursor_advance(zc)) { 511265744Sdelphij dsl_dir_t *chld_dd; 512265744Sdelphij uint64_t count; 513265744Sdelphij 514265744Sdelphij VERIFY0(dsl_dir_hold_obj(dp, za->za_first_integer, NULL, FTAG, 515265744Sdelphij &chld_dd)); 516265744Sdelphij 517265744Sdelphij /* 518265744Sdelphij * Ignore hidden ($FREE, $MOS & $ORIGIN) objsets and 519265744Sdelphij * temporary datasets. 520265744Sdelphij */ 521265744Sdelphij if (chld_dd->dd_myname[0] == '$' || 522265744Sdelphij chld_dd->dd_myname[0] == '%') { 523265744Sdelphij dsl_dir_rele(chld_dd, FTAG); 524265744Sdelphij continue; 525265744Sdelphij } 526265744Sdelphij 527265744Sdelphij my_fs_cnt++; /* count this child */ 528265744Sdelphij 529265744Sdelphij dsl_dir_init_fs_ss_count(chld_dd, tx); 530265744Sdelphij 531265744Sdelphij VERIFY0(zap_lookup(os, chld_dd->dd_object, 532265744Sdelphij DD_FIELD_FILESYSTEM_COUNT, sizeof (count), 1, &count)); 533265744Sdelphij my_fs_cnt += count; 534265744Sdelphij VERIFY0(zap_lookup(os, chld_dd->dd_object, 535265744Sdelphij DD_FIELD_SNAPSHOT_COUNT, sizeof (count), 1, &count)); 536265744Sdelphij my_ss_cnt += count; 537265744Sdelphij 538265744Sdelphij dsl_dir_rele(chld_dd, FTAG); 539265744Sdelphij } 540265744Sdelphij zap_cursor_fini(zc); 541265744Sdelphij /* Count my snapshots (we counted children's snapshots above) */ 542265744Sdelphij VERIFY0(dsl_dataset_hold_obj(dd->dd_pool, 543265744Sdelphij dd->dd_phys->dd_head_dataset_obj, FTAG, &ds)); 544265744Sdelphij 545265744Sdelphij for (zap_cursor_init(zc, os, ds->ds_phys->ds_snapnames_zapobj); 546265744Sdelphij zap_cursor_retrieve(zc, za) == 0; 547265744Sdelphij zap_cursor_advance(zc)) { 548265744Sdelphij /* Don't count temporary snapshots */ 549265744Sdelphij if (za->za_name[0] != '%') 550265744Sdelphij my_ss_cnt++; 551265744Sdelphij } 552267138Sdelphij zap_cursor_fini(zc); 553265744Sdelphij 554265744Sdelphij dsl_dataset_rele(ds, FTAG); 555265744Sdelphij 556265744Sdelphij kmem_free(zc, sizeof (zap_cursor_t)); 557265744Sdelphij kmem_free(za, sizeof (zap_attribute_t)); 558265744Sdelphij 559265744Sdelphij /* we're in a sync task, update counts */ 560265744Sdelphij dmu_buf_will_dirty(dd->dd_dbuf, tx); 561265744Sdelphij VERIFY0(zap_add(os, dd->dd_object, DD_FIELD_FILESYSTEM_COUNT, 562265744Sdelphij sizeof (my_fs_cnt), 1, &my_fs_cnt, tx)); 563265744Sdelphij VERIFY0(zap_add(os, dd->dd_object, DD_FIELD_SNAPSHOT_COUNT, 564265744Sdelphij sizeof (my_ss_cnt), 1, &my_ss_cnt, tx)); 565265744Sdelphij} 566265744Sdelphij 567265744Sdelphijstatic int 568265744Sdelphijdsl_dir_actv_fs_ss_limit_check(void *arg, dmu_tx_t *tx) 569265744Sdelphij{ 570265744Sdelphij char *ddname = (char *)arg; 571265744Sdelphij dsl_pool_t *dp = dmu_tx_pool(tx); 572265744Sdelphij dsl_dataset_t *ds; 573265744Sdelphij dsl_dir_t *dd; 574265744Sdelphij int error; 575265744Sdelphij 576265744Sdelphij error = dsl_dataset_hold(dp, ddname, FTAG, &ds); 577265744Sdelphij if (error != 0) 578265744Sdelphij return (error); 579265744Sdelphij 580265744Sdelphij if (!spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_FS_SS_LIMIT)) { 581265744Sdelphij dsl_dataset_rele(ds, FTAG); 582265744Sdelphij return (SET_ERROR(ENOTSUP)); 583265744Sdelphij } 584265744Sdelphij 585265744Sdelphij dd = ds->ds_dir; 586265744Sdelphij if (spa_feature_is_active(dp->dp_spa, SPA_FEATURE_FS_SS_LIMIT) && 587265744Sdelphij dsl_dir_is_zapified(dd) && 588265744Sdelphij zap_contains(dp->dp_meta_objset, dd->dd_object, 589265744Sdelphij DD_FIELD_FILESYSTEM_COUNT) == 0) { 590265744Sdelphij dsl_dataset_rele(ds, FTAG); 591265744Sdelphij return (SET_ERROR(EALREADY)); 592265744Sdelphij } 593265744Sdelphij 594265744Sdelphij dsl_dataset_rele(ds, FTAG); 595265744Sdelphij return (0); 596265744Sdelphij} 597265744Sdelphij 598265744Sdelphijstatic void 599265744Sdelphijdsl_dir_actv_fs_ss_limit_sync(void *arg, dmu_tx_t *tx) 600265744Sdelphij{ 601265744Sdelphij char *ddname = (char *)arg; 602265744Sdelphij dsl_pool_t *dp = dmu_tx_pool(tx); 603265744Sdelphij dsl_dataset_t *ds; 604265744Sdelphij spa_t *spa; 605265744Sdelphij 606265744Sdelphij VERIFY0(dsl_dataset_hold(dp, ddname, FTAG, &ds)); 607265744Sdelphij 608265744Sdelphij spa = dsl_dataset_get_spa(ds); 609265744Sdelphij 610265744Sdelphij if (!spa_feature_is_active(spa, SPA_FEATURE_FS_SS_LIMIT)) { 611265744Sdelphij /* 612265744Sdelphij * Since the feature was not active and we're now setting a 613265744Sdelphij * limit, increment the feature-active counter so that the 614265744Sdelphij * feature becomes active for the first time. 615265744Sdelphij * 616265744Sdelphij * We are already in a sync task so we can update the MOS. 617265744Sdelphij */ 618265744Sdelphij spa_feature_incr(spa, SPA_FEATURE_FS_SS_LIMIT, tx); 619265744Sdelphij } 620265744Sdelphij 621265744Sdelphij /* 622265744Sdelphij * Since we are now setting a non-UINT64_MAX limit on the filesystem, 623265744Sdelphij * we need to ensure the counts are correct. Descend down the tree from 624265744Sdelphij * this point and update all of the counts to be accurate. 625265744Sdelphij */ 626265744Sdelphij dsl_dir_init_fs_ss_count(ds->ds_dir, tx); 627265744Sdelphij 628265744Sdelphij dsl_dataset_rele(ds, FTAG); 629265744Sdelphij} 630265744Sdelphij 631265744Sdelphij/* 632265744Sdelphij * Make sure the feature is enabled and activate it if necessary. 633265744Sdelphij * Since we're setting a limit, ensure the on-disk counts are valid. 634265744Sdelphij * This is only called by the ioctl path when setting a limit value. 635265744Sdelphij * 636265744Sdelphij * We do not need to validate the new limit, since users who can change the 637265744Sdelphij * limit are also allowed to exceed the limit. 638265744Sdelphij */ 639265744Sdelphijint 640265744Sdelphijdsl_dir_activate_fs_ss_limit(const char *ddname) 641265744Sdelphij{ 642265744Sdelphij int error; 643265744Sdelphij 644265744Sdelphij error = dsl_sync_task(ddname, dsl_dir_actv_fs_ss_limit_check, 645269006Sdelphij dsl_dir_actv_fs_ss_limit_sync, (void *)ddname, 0, 646269006Sdelphij ZFS_SPACE_CHECK_RESERVED); 647265744Sdelphij 648265744Sdelphij if (error == EALREADY) 649265744Sdelphij error = 0; 650265744Sdelphij 651265744Sdelphij return (error); 652265744Sdelphij} 653265744Sdelphij 654265744Sdelphij/* 655265744Sdelphij * Used to determine if the filesystem_limit or snapshot_limit should be 656265744Sdelphij * enforced. We allow the limit to be exceeded if the user has permission to 657265744Sdelphij * write the property value. We pass in the creds that we got in the open 658265744Sdelphij * context since we will always be the GZ root in syncing context. We also have 659265744Sdelphij * to handle the case where we are allowed to change the limit on the current 660265744Sdelphij * dataset, but there may be another limit in the tree above. 661265744Sdelphij * 662265744Sdelphij * We can never modify these two properties within a non-global zone. In 663265744Sdelphij * addition, the other checks are modeled on zfs_secpolicy_write_perms. We 664265744Sdelphij * can't use that function since we are already holding the dp_config_rwlock. 665265744Sdelphij * In addition, we already have the dd and dealing with snapshots is simplified 666265744Sdelphij * in this code. 667265744Sdelphij */ 668265744Sdelphij 669265744Sdelphijtypedef enum { 670265744Sdelphij ENFORCE_ALWAYS, 671265744Sdelphij ENFORCE_NEVER, 672265744Sdelphij ENFORCE_ABOVE 673265744Sdelphij} enforce_res_t; 674265744Sdelphij 675265744Sdelphijstatic enforce_res_t 676265744Sdelphijdsl_enforce_ds_ss_limits(dsl_dir_t *dd, zfs_prop_t prop, cred_t *cr) 677265744Sdelphij{ 678265744Sdelphij enforce_res_t enforce = ENFORCE_ALWAYS; 679265744Sdelphij uint64_t obj; 680265744Sdelphij dsl_dataset_t *ds; 681265744Sdelphij uint64_t zoned; 682265744Sdelphij 683265744Sdelphij ASSERT(prop == ZFS_PROP_FILESYSTEM_LIMIT || 684265744Sdelphij prop == ZFS_PROP_SNAPSHOT_LIMIT); 685265744Sdelphij 686265744Sdelphij#ifdef _KERNEL 687265744Sdelphij#ifdef __FreeBSD__ 688265744Sdelphij if (jailed(cr)) 689265744Sdelphij#else 690265744Sdelphij if (crgetzoneid(cr) != GLOBAL_ZONEID) 691265744Sdelphij#endif 692265744Sdelphij return (ENFORCE_ALWAYS); 693265744Sdelphij 694265744Sdelphij if (secpolicy_zfs(cr) == 0) 695265744Sdelphij return (ENFORCE_NEVER); 696265744Sdelphij#endif 697265744Sdelphij 698265744Sdelphij if ((obj = dd->dd_phys->dd_head_dataset_obj) == 0) 699265744Sdelphij return (ENFORCE_ALWAYS); 700265744Sdelphij 701265744Sdelphij ASSERT(dsl_pool_config_held(dd->dd_pool)); 702265744Sdelphij 703265744Sdelphij if (dsl_dataset_hold_obj(dd->dd_pool, obj, FTAG, &ds) != 0) 704265744Sdelphij return (ENFORCE_ALWAYS); 705265744Sdelphij 706265744Sdelphij if (dsl_prop_get_ds(ds, "zoned", 8, 1, &zoned, NULL) || zoned) { 707265744Sdelphij /* Only root can access zoned fs's from the GZ */ 708265744Sdelphij enforce = ENFORCE_ALWAYS; 709265744Sdelphij } else { 710265744Sdelphij if (dsl_deleg_access_impl(ds, zfs_prop_to_name(prop), cr) == 0) 711265744Sdelphij enforce = ENFORCE_ABOVE; 712265744Sdelphij } 713265744Sdelphij 714265744Sdelphij dsl_dataset_rele(ds, FTAG); 715265744Sdelphij return (enforce); 716265744Sdelphij} 717265744Sdelphij 718265744Sdelphij/* 719265744Sdelphij * Check if adding additional child filesystem(s) would exceed any filesystem 720265744Sdelphij * limits or adding additional snapshot(s) would exceed any snapshot limits. 721265744Sdelphij * The prop argument indicates which limit to check. 722265744Sdelphij * 723265744Sdelphij * Note that all filesystem limits up to the root (or the highest 724265744Sdelphij * initialized) filesystem or the given ancestor must be satisfied. 725265744Sdelphij */ 726265744Sdelphijint 727265744Sdelphijdsl_fs_ss_limit_check(dsl_dir_t *dd, uint64_t delta, zfs_prop_t prop, 728265744Sdelphij dsl_dir_t *ancestor, cred_t *cr) 729265744Sdelphij{ 730265744Sdelphij objset_t *os = dd->dd_pool->dp_meta_objset; 731265744Sdelphij uint64_t limit, count; 732265744Sdelphij char *count_prop; 733265744Sdelphij enforce_res_t enforce; 734265744Sdelphij int err = 0; 735265744Sdelphij 736265744Sdelphij ASSERT(dsl_pool_config_held(dd->dd_pool)); 737265744Sdelphij ASSERT(prop == ZFS_PROP_FILESYSTEM_LIMIT || 738265744Sdelphij prop == ZFS_PROP_SNAPSHOT_LIMIT); 739265744Sdelphij 740265744Sdelphij /* 741265744Sdelphij * If we're allowed to change the limit, don't enforce the limit 742265744Sdelphij * e.g. this can happen if a snapshot is taken by an administrative 743265744Sdelphij * user in the global zone (i.e. a recursive snapshot by root). 744265744Sdelphij * However, we must handle the case of delegated permissions where we 745265744Sdelphij * are allowed to change the limit on the current dataset, but there 746265744Sdelphij * is another limit in the tree above. 747265744Sdelphij */ 748265744Sdelphij enforce = dsl_enforce_ds_ss_limits(dd, prop, cr); 749265744Sdelphij if (enforce == ENFORCE_NEVER) 750265744Sdelphij return (0); 751265744Sdelphij 752265744Sdelphij /* 753265744Sdelphij * e.g. if renaming a dataset with no snapshots, count adjustment 754265744Sdelphij * is 0. 755265744Sdelphij */ 756265744Sdelphij if (delta == 0) 757265744Sdelphij return (0); 758265744Sdelphij 759265744Sdelphij if (prop == ZFS_PROP_SNAPSHOT_LIMIT) { 760265744Sdelphij /* 761265744Sdelphij * We don't enforce the limit for temporary snapshots. This is 762265744Sdelphij * indicated by a NULL cred_t argument. 763265744Sdelphij */ 764265744Sdelphij if (cr == NULL) 765265744Sdelphij return (0); 766265744Sdelphij 767265744Sdelphij count_prop = DD_FIELD_SNAPSHOT_COUNT; 768265744Sdelphij } else { 769265744Sdelphij count_prop = DD_FIELD_FILESYSTEM_COUNT; 770265744Sdelphij } 771265744Sdelphij 772265744Sdelphij /* 773265744Sdelphij * If an ancestor has been provided, stop checking the limit once we 774265744Sdelphij * hit that dir. We need this during rename so that we don't overcount 775265744Sdelphij * the check once we recurse up to the common ancestor. 776265744Sdelphij */ 777265744Sdelphij if (ancestor == dd) 778265744Sdelphij return (0); 779265744Sdelphij 780265744Sdelphij /* 781265744Sdelphij * If we hit an uninitialized node while recursing up the tree, we can 782265744Sdelphij * stop since we know there is no limit here (or above). The counts are 783265744Sdelphij * not valid on this node and we know we won't touch this node's counts. 784265744Sdelphij */ 785265744Sdelphij if (!dsl_dir_is_zapified(dd) || zap_lookup(os, dd->dd_object, 786265744Sdelphij count_prop, sizeof (count), 1, &count) == ENOENT) 787265744Sdelphij return (0); 788265744Sdelphij 789265744Sdelphij err = dsl_prop_get_dd(dd, zfs_prop_to_name(prop), 8, 1, &limit, NULL, 790265744Sdelphij B_FALSE); 791265744Sdelphij if (err != 0) 792265744Sdelphij return (err); 793265744Sdelphij 794265744Sdelphij /* Is there a limit which we've hit? */ 795265744Sdelphij if (enforce == ENFORCE_ALWAYS && (count + delta) > limit) 796265744Sdelphij return (SET_ERROR(EDQUOT)); 797265744Sdelphij 798265744Sdelphij if (dd->dd_parent != NULL) 799265744Sdelphij err = dsl_fs_ss_limit_check(dd->dd_parent, delta, prop, 800265744Sdelphij ancestor, cr); 801265744Sdelphij 802265744Sdelphij return (err); 803265744Sdelphij} 804265744Sdelphij 805265744Sdelphij/* 806265744Sdelphij * Adjust the filesystem or snapshot count for the specified dsl_dir_t and all 807265744Sdelphij * parents. When a new filesystem/snapshot is created, increment the count on 808265744Sdelphij * all parents, and when a filesystem/snapshot is destroyed, decrement the 809265744Sdelphij * count. 810265744Sdelphij */ 811265744Sdelphijvoid 812265744Sdelphijdsl_fs_ss_count_adjust(dsl_dir_t *dd, int64_t delta, const char *prop, 813265744Sdelphij dmu_tx_t *tx) 814265744Sdelphij{ 815265744Sdelphij int err; 816265744Sdelphij objset_t *os = dd->dd_pool->dp_meta_objset; 817265744Sdelphij uint64_t count; 818265744Sdelphij 819265744Sdelphij ASSERT(dsl_pool_config_held(dd->dd_pool)); 820265744Sdelphij ASSERT(dmu_tx_is_syncing(tx)); 821265744Sdelphij ASSERT(strcmp(prop, DD_FIELD_FILESYSTEM_COUNT) == 0 || 822265744Sdelphij strcmp(prop, DD_FIELD_SNAPSHOT_COUNT) == 0); 823265744Sdelphij 824265744Sdelphij /* 825265744Sdelphij * When we receive an incremental stream into a filesystem that already 826265744Sdelphij * exists, a temporary clone is created. We don't count this temporary 827265744Sdelphij * clone, whose name begins with a '%'. We also ignore hidden ($FREE, 828265744Sdelphij * $MOS & $ORIGIN) objsets. 829265744Sdelphij */ 830265744Sdelphij if ((dd->dd_myname[0] == '%' || dd->dd_myname[0] == '$') && 831265744Sdelphij strcmp(prop, DD_FIELD_FILESYSTEM_COUNT) == 0) 832265744Sdelphij return; 833265744Sdelphij 834265744Sdelphij /* 835265744Sdelphij * e.g. if renaming a dataset with no snapshots, count adjustment is 0 836265744Sdelphij */ 837265744Sdelphij if (delta == 0) 838265744Sdelphij return; 839265744Sdelphij 840265744Sdelphij /* 841265744Sdelphij * If we hit an uninitialized node while recursing up the tree, we can 842265744Sdelphij * stop since we know the counts are not valid on this node and we 843265744Sdelphij * know we shouldn't touch this node's counts. An uninitialized count 844265744Sdelphij * on the node indicates that either the feature has not yet been 845265744Sdelphij * activated or there are no limits on this part of the tree. 846265744Sdelphij */ 847265744Sdelphij if (!dsl_dir_is_zapified(dd) || (err = zap_lookup(os, dd->dd_object, 848265744Sdelphij prop, sizeof (count), 1, &count)) == ENOENT) 849265744Sdelphij return; 850265744Sdelphij VERIFY0(err); 851265744Sdelphij 852265744Sdelphij count += delta; 853265744Sdelphij /* Use a signed verify to make sure we're not neg. */ 854265744Sdelphij VERIFY3S(count, >=, 0); 855265744Sdelphij 856265744Sdelphij VERIFY0(zap_update(os, dd->dd_object, prop, sizeof (count), 1, &count, 857265744Sdelphij tx)); 858265744Sdelphij 859265744Sdelphij /* Roll up this additional count into our ancestors */ 860265744Sdelphij if (dd->dd_parent != NULL) 861265744Sdelphij dsl_fs_ss_count_adjust(dd->dd_parent, delta, prop, tx); 862265744Sdelphij} 863265744Sdelphij 864168404Spjduint64_t 865185029Spjddsl_dir_create_sync(dsl_pool_t *dp, dsl_dir_t *pds, const char *name, 866185029Spjd dmu_tx_t *tx) 867168404Spjd{ 868185029Spjd objset_t *mos = dp->dp_meta_objset; 869168404Spjd uint64_t ddobj; 870219089Spjd dsl_dir_phys_t *ddphys; 871168404Spjd dmu_buf_t *dbuf; 872168404Spjd 873168404Spjd ddobj = dmu_object_alloc(mos, DMU_OT_DSL_DIR, 0, 874168404Spjd DMU_OT_DSL_DIR, sizeof (dsl_dir_phys_t), tx); 875185029Spjd if (pds) { 876185029Spjd VERIFY(0 == zap_add(mos, pds->dd_phys->dd_child_dir_zapobj, 877185029Spjd name, sizeof (uint64_t), 1, &ddobj, tx)); 878185029Spjd } else { 879185029Spjd /* it's the root dir */ 880185029Spjd VERIFY(0 == zap_add(mos, DMU_POOL_DIRECTORY_OBJECT, 881185029Spjd DMU_POOL_ROOT_DATASET, sizeof (uint64_t), 1, &ddobj, tx)); 882185029Spjd } 883168404Spjd VERIFY(0 == dmu_bonus_hold(mos, ddobj, FTAG, &dbuf)); 884168404Spjd dmu_buf_will_dirty(dbuf, tx); 885219089Spjd ddphys = dbuf->db_data; 886168404Spjd 887219089Spjd ddphys->dd_creation_time = gethrestime_sec(); 888265744Sdelphij if (pds) { 889219089Spjd ddphys->dd_parent_obj = pds->dd_object; 890265744Sdelphij 891265744Sdelphij /* update the filesystem counts */ 892265744Sdelphij dsl_fs_ss_count_adjust(pds, 1, DD_FIELD_FILESYSTEM_COUNT, tx); 893265744Sdelphij } 894219089Spjd ddphys->dd_props_zapobj = zap_create(mos, 895168404Spjd DMU_OT_DSL_PROPS, DMU_OT_NONE, 0, tx); 896219089Spjd ddphys->dd_child_dir_zapobj = zap_create(mos, 897168404Spjd DMU_OT_DSL_DIR_CHILD_MAP, DMU_OT_NONE, 0, tx); 898185029Spjd if (spa_version(dp->dp_spa) >= SPA_VERSION_USED_BREAKDOWN) 899219089Spjd ddphys->dd_flags |= DD_FLAG_USED_BREAKDOWN; 900168404Spjd dmu_buf_rele(dbuf, FTAG); 901168404Spjd 902168404Spjd return (ddobj); 903168404Spjd} 904168404Spjd 905185029Spjdboolean_t 906185029Spjddsl_dir_is_clone(dsl_dir_t *dd) 907168404Spjd{ 908185029Spjd return (dd->dd_phys->dd_origin_obj && 909185029Spjd (dd->dd_pool->dp_origin_snap == NULL || 910185029Spjd dd->dd_phys->dd_origin_obj != 911185029Spjd dd->dd_pool->dp_origin_snap->ds_object)); 912168404Spjd} 913168404Spjd 914168404Spjdvoid 915168404Spjddsl_dir_stats(dsl_dir_t *dd, nvlist_t *nv) 916168404Spjd{ 917168404Spjd mutex_enter(&dd->dd_lock); 918185029Spjd dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USED, 919185029Spjd dd->dd_phys->dd_used_bytes); 920185029Spjd dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_QUOTA, dd->dd_phys->dd_quota); 921168404Spjd dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_RESERVATION, 922168404Spjd dd->dd_phys->dd_reserved); 923168404Spjd dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_COMPRESSRATIO, 924168404Spjd dd->dd_phys->dd_compressed_bytes == 0 ? 100 : 925168404Spjd (dd->dd_phys->dd_uncompressed_bytes * 100 / 926168404Spjd dd->dd_phys->dd_compressed_bytes)); 927247585Smm dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_LOGICALUSED, 928247585Smm dd->dd_phys->dd_uncompressed_bytes); 929185029Spjd if (dd->dd_phys->dd_flags & DD_FLAG_USED_BREAKDOWN) { 930185029Spjd dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDSNAP, 931185029Spjd dd->dd_phys->dd_used_breakdown[DD_USED_SNAP]); 932185029Spjd dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDDS, 933185029Spjd dd->dd_phys->dd_used_breakdown[DD_USED_HEAD]); 934185029Spjd dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDREFRESERV, 935185029Spjd dd->dd_phys->dd_used_breakdown[DD_USED_REFRSRV]); 936185029Spjd dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDCHILD, 937185029Spjd dd->dd_phys->dd_used_breakdown[DD_USED_CHILD] + 938185029Spjd dd->dd_phys->dd_used_breakdown[DD_USED_CHILD_RSRV]); 939185029Spjd } 940168404Spjd mutex_exit(&dd->dd_lock); 941168404Spjd 942265744Sdelphij if (dsl_dir_is_zapified(dd)) { 943265744Sdelphij uint64_t count; 944265744Sdelphij objset_t *os = dd->dd_pool->dp_meta_objset; 945265744Sdelphij 946265744Sdelphij if (zap_lookup(os, dd->dd_object, DD_FIELD_FILESYSTEM_COUNT, 947265744Sdelphij sizeof (count), 1, &count) == 0) { 948265744Sdelphij dsl_prop_nvlist_add_uint64(nv, 949265744Sdelphij ZFS_PROP_FILESYSTEM_COUNT, count); 950265744Sdelphij } 951265744Sdelphij if (zap_lookup(os, dd->dd_object, DD_FIELD_SNAPSHOT_COUNT, 952265744Sdelphij sizeof (count), 1, &count) == 0) { 953265744Sdelphij dsl_prop_nvlist_add_uint64(nv, 954265744Sdelphij ZFS_PROP_SNAPSHOT_COUNT, count); 955265744Sdelphij } 956265744Sdelphij } 957265744Sdelphij 958185029Spjd if (dsl_dir_is_clone(dd)) { 959168404Spjd dsl_dataset_t *ds; 960168404Spjd char buf[MAXNAMELEN]; 961168404Spjd 962248571Smm VERIFY0(dsl_dataset_hold_obj(dd->dd_pool, 963185029Spjd dd->dd_phys->dd_origin_obj, FTAG, &ds)); 964168404Spjd dsl_dataset_name(ds, buf); 965185029Spjd dsl_dataset_rele(ds, FTAG); 966168404Spjd dsl_prop_nvlist_add_string(nv, ZFS_PROP_ORIGIN, buf); 967168404Spjd } 968168404Spjd} 969168404Spjd 970168404Spjdvoid 971168404Spjddsl_dir_dirty(dsl_dir_t *dd, dmu_tx_t *tx) 972168404Spjd{ 973168404Spjd dsl_pool_t *dp = dd->dd_pool; 974168404Spjd 975168404Spjd ASSERT(dd->dd_phys); 976168404Spjd 977248571Smm if (txg_list_add(&dp->dp_dirty_dirs, dd, tx->tx_txg)) { 978168404Spjd /* up the hold count until we can be written out */ 979168404Spjd dmu_buf_add_ref(dd->dd_dbuf, dd); 980168404Spjd } 981168404Spjd} 982168404Spjd 983168404Spjdstatic int64_t 984168404Spjdparent_delta(dsl_dir_t *dd, uint64_t used, int64_t delta) 985168404Spjd{ 986168404Spjd uint64_t old_accounted = MAX(used, dd->dd_phys->dd_reserved); 987168404Spjd uint64_t new_accounted = MAX(used + delta, dd->dd_phys->dd_reserved); 988168404Spjd return (new_accounted - old_accounted); 989168404Spjd} 990168404Spjd 991168404Spjdvoid 992168404Spjddsl_dir_sync(dsl_dir_t *dd, dmu_tx_t *tx) 993168404Spjd{ 994168404Spjd ASSERT(dmu_tx_is_syncing(tx)); 995168404Spjd 996168404Spjd mutex_enter(&dd->dd_lock); 997240415Smm ASSERT0(dd->dd_tempreserved[tx->tx_txg&TXG_MASK]); 998168404Spjd dprintf_dd(dd, "txg=%llu towrite=%lluK\n", tx->tx_txg, 999168404Spjd dd->dd_space_towrite[tx->tx_txg&TXG_MASK] / 1024); 1000168404Spjd dd->dd_space_towrite[tx->tx_txg&TXG_MASK] = 0; 1001168404Spjd mutex_exit(&dd->dd_lock); 1002168404Spjd 1003168404Spjd /* release the hold from dsl_dir_dirty */ 1004168404Spjd dmu_buf_rele(dd->dd_dbuf, dd); 1005168404Spjd} 1006168404Spjd 1007168404Spjdstatic uint64_t 1008185029Spjddsl_dir_space_towrite(dsl_dir_t *dd) 1009168404Spjd{ 1010185029Spjd uint64_t space = 0; 1011168404Spjd int i; 1012168404Spjd 1013168404Spjd ASSERT(MUTEX_HELD(&dd->dd_lock)); 1014168404Spjd 1015168404Spjd for (i = 0; i < TXG_SIZE; i++) { 1016168404Spjd space += dd->dd_space_towrite[i&TXG_MASK]; 1017168404Spjd ASSERT3U(dd->dd_space_towrite[i&TXG_MASK], >=, 0); 1018168404Spjd } 1019168404Spjd return (space); 1020168404Spjd} 1021168404Spjd 1022168404Spjd/* 1023168404Spjd * How much space would dd have available if ancestor had delta applied 1024168404Spjd * to it? If ondiskonly is set, we're only interested in what's 1025168404Spjd * on-disk, not estimated pending changes. 1026168404Spjd */ 1027168404Spjduint64_t 1028168404Spjddsl_dir_space_available(dsl_dir_t *dd, 1029168404Spjd dsl_dir_t *ancestor, int64_t delta, int ondiskonly) 1030168404Spjd{ 1031168404Spjd uint64_t parentspace, myspace, quota, used; 1032168404Spjd 1033168404Spjd /* 1034168404Spjd * If there are no restrictions otherwise, assume we have 1035168404Spjd * unlimited space available. 1036168404Spjd */ 1037168404Spjd quota = UINT64_MAX; 1038168404Spjd parentspace = UINT64_MAX; 1039168404Spjd 1040168404Spjd if (dd->dd_parent != NULL) { 1041168404Spjd parentspace = dsl_dir_space_available(dd->dd_parent, 1042168404Spjd ancestor, delta, ondiskonly); 1043168404Spjd } 1044168404Spjd 1045168404Spjd mutex_enter(&dd->dd_lock); 1046168404Spjd if (dd->dd_phys->dd_quota != 0) 1047168404Spjd quota = dd->dd_phys->dd_quota; 1048185029Spjd used = dd->dd_phys->dd_used_bytes; 1049185029Spjd if (!ondiskonly) 1050185029Spjd used += dsl_dir_space_towrite(dd); 1051168404Spjd 1052168404Spjd if (dd->dd_parent == NULL) { 1053168404Spjd uint64_t poolsize = dsl_pool_adjustedsize(dd->dd_pool, FALSE); 1054168404Spjd quota = MIN(quota, poolsize); 1055168404Spjd } 1056168404Spjd 1057168404Spjd if (dd->dd_phys->dd_reserved > used && parentspace != UINT64_MAX) { 1058168404Spjd /* 1059168404Spjd * We have some space reserved, in addition to what our 1060168404Spjd * parent gave us. 1061168404Spjd */ 1062168404Spjd parentspace += dd->dd_phys->dd_reserved - used; 1063168404Spjd } 1064168404Spjd 1065185029Spjd if (dd == ancestor) { 1066185029Spjd ASSERT(delta <= 0); 1067185029Spjd ASSERT(used >= -delta); 1068185029Spjd used += delta; 1069185029Spjd if (parentspace != UINT64_MAX) 1070185029Spjd parentspace -= delta; 1071185029Spjd } 1072185029Spjd 1073168404Spjd if (used > quota) { 1074168404Spjd /* over quota */ 1075168404Spjd myspace = 0; 1076168404Spjd } else { 1077168404Spjd /* 1078168404Spjd * the lesser of the space provided by our parent and 1079168404Spjd * the space left in our quota 1080168404Spjd */ 1081168404Spjd myspace = MIN(parentspace, quota - used); 1082168404Spjd } 1083168404Spjd 1084168404Spjd mutex_exit(&dd->dd_lock); 1085168404Spjd 1086168404Spjd return (myspace); 1087168404Spjd} 1088168404Spjd 1089168404Spjdstruct tempreserve { 1090168404Spjd list_node_t tr_node; 1091168404Spjd dsl_dir_t *tr_ds; 1092168404Spjd uint64_t tr_size; 1093168404Spjd}; 1094168404Spjd 1095168404Spjdstatic int 1096185029Spjddsl_dir_tempreserve_impl(dsl_dir_t *dd, uint64_t asize, boolean_t netfree, 1097185029Spjd boolean_t ignorequota, boolean_t checkrefquota, list_t *tr_list, 1098185029Spjd dmu_tx_t *tx, boolean_t first) 1099168404Spjd{ 1100168404Spjd uint64_t txg = tx->tx_txg; 1101185029Spjd uint64_t est_inflight, used_on_disk, quota, parent_rsrv; 1102219089Spjd uint64_t deferred = 0; 1103185029Spjd struct tempreserve *tr; 1104219089Spjd int retval = EDQUOT; 1105168404Spjd int txgidx = txg & TXG_MASK; 1106168404Spjd int i; 1107185029Spjd uint64_t ref_rsrv = 0; 1108168404Spjd 1109168404Spjd ASSERT3U(txg, !=, 0); 1110185029Spjd ASSERT3S(asize, >, 0); 1111168404Spjd 1112168404Spjd mutex_enter(&dd->dd_lock); 1113185029Spjd 1114168404Spjd /* 1115168404Spjd * Check against the dsl_dir's quota. We don't add in the delta 1116168404Spjd * when checking for over-quota because they get one free hit. 1117168404Spjd */ 1118185029Spjd est_inflight = dsl_dir_space_towrite(dd); 1119168404Spjd for (i = 0; i < TXG_SIZE; i++) 1120185029Spjd est_inflight += dd->dd_tempreserved[i]; 1121185029Spjd used_on_disk = dd->dd_phys->dd_used_bytes; 1122168404Spjd 1123185029Spjd /* 1124185029Spjd * On the first iteration, fetch the dataset's used-on-disk and 1125185029Spjd * refreservation values. Also, if checkrefquota is set, test if 1126185029Spjd * allocating this space would exceed the dataset's refquota. 1127185029Spjd */ 1128185029Spjd if (first && tx->tx_objset) { 1129185029Spjd int error; 1130219089Spjd dsl_dataset_t *ds = tx->tx_objset->os_dsl_dataset; 1131168404Spjd 1132185029Spjd error = dsl_dataset_check_quota(ds, checkrefquota, 1133185029Spjd asize, est_inflight, &used_on_disk, &ref_rsrv); 1134185029Spjd if (error) { 1135185029Spjd mutex_exit(&dd->dd_lock); 1136185029Spjd return (error); 1137185029Spjd } 1138185029Spjd } 1139185029Spjd 1140185029Spjd /* 1141185029Spjd * If this transaction will result in a net free of space, 1142185029Spjd * we want to let it through. 1143185029Spjd */ 1144185029Spjd if (ignorequota || netfree || dd->dd_phys->dd_quota == 0) 1145185029Spjd quota = UINT64_MAX; 1146185029Spjd else 1147168404Spjd quota = dd->dd_phys->dd_quota; 1148168404Spjd 1149168404Spjd /* 1150219089Spjd * Adjust the quota against the actual pool size at the root 1151219089Spjd * minus any outstanding deferred frees. 1152185029Spjd * To ensure that it's possible to remove files from a full 1153185029Spjd * pool without inducing transient overcommits, we throttle 1154168404Spjd * netfree transactions against a quota that is slightly larger, 1155168404Spjd * but still within the pool's allocation slop. In cases where 1156168404Spjd * we're very close to full, this will allow a steady trickle of 1157168404Spjd * removes to get through. 1158168404Spjd */ 1159168404Spjd if (dd->dd_parent == NULL) { 1160219089Spjd spa_t *spa = dd->dd_pool->dp_spa; 1161168404Spjd uint64_t poolsize = dsl_pool_adjustedsize(dd->dd_pool, netfree); 1162219089Spjd deferred = metaslab_class_get_deferred(spa_normal_class(spa)); 1163219089Spjd if (poolsize - deferred < quota) { 1164219089Spjd quota = poolsize - deferred; 1165219089Spjd retval = ENOSPC; 1166168404Spjd } 1167168404Spjd } 1168168404Spjd 1169168404Spjd /* 1170168404Spjd * If they are requesting more space, and our current estimate 1171185029Spjd * is over quota, they get to try again unless the actual 1172168404Spjd * on-disk is over quota and there are no pending changes (which 1173168404Spjd * may free up space for us). 1174168404Spjd */ 1175219089Spjd if (used_on_disk + est_inflight >= quota) { 1176219089Spjd if (est_inflight > 0 || used_on_disk < quota || 1177219089Spjd (retval == ENOSPC && used_on_disk < quota + deferred)) 1178219089Spjd retval = ERESTART; 1179185029Spjd dprintf_dd(dd, "failing: used=%lluK inflight = %lluK " 1180168404Spjd "quota=%lluK tr=%lluK err=%d\n", 1181185029Spjd used_on_disk>>10, est_inflight>>10, 1182219089Spjd quota>>10, asize>>10, retval); 1183168404Spjd mutex_exit(&dd->dd_lock); 1184249195Smm return (SET_ERROR(retval)); 1185168404Spjd } 1186168404Spjd 1187168404Spjd /* We need to up our estimated delta before dropping dd_lock */ 1188168404Spjd dd->dd_tempreserved[txgidx] += asize; 1189168404Spjd 1190185029Spjd parent_rsrv = parent_delta(dd, used_on_disk + est_inflight, 1191185029Spjd asize - ref_rsrv); 1192168404Spjd mutex_exit(&dd->dd_lock); 1193168404Spjd 1194185029Spjd tr = kmem_zalloc(sizeof (struct tempreserve), KM_SLEEP); 1195168404Spjd tr->tr_ds = dd; 1196168404Spjd tr->tr_size = asize; 1197168404Spjd list_insert_tail(tr_list, tr); 1198168404Spjd 1199168404Spjd /* see if it's OK with our parent */ 1200168404Spjd if (dd->dd_parent && parent_rsrv) { 1201185029Spjd boolean_t ismos = (dd->dd_phys->dd_head_dataset_obj == 0); 1202185029Spjd 1203168404Spjd return (dsl_dir_tempreserve_impl(dd->dd_parent, 1204185029Spjd parent_rsrv, netfree, ismos, TRUE, tr_list, tx, FALSE)); 1205168404Spjd } else { 1206168404Spjd return (0); 1207168404Spjd } 1208168404Spjd} 1209168404Spjd 1210168404Spjd/* 1211168404Spjd * Reserve space in this dsl_dir, to be used in this tx's txg. 1212185029Spjd * After the space has been dirtied (and dsl_dir_willuse_space() 1213185029Spjd * has been called), the reservation should be canceled, using 1214185029Spjd * dsl_dir_tempreserve_clear(). 1215168404Spjd */ 1216168404Spjdint 1217185029Spjddsl_dir_tempreserve_space(dsl_dir_t *dd, uint64_t lsize, uint64_t asize, 1218185029Spjd uint64_t fsize, uint64_t usize, void **tr_cookiep, dmu_tx_t *tx) 1219168404Spjd{ 1220185029Spjd int err; 1221168404Spjd list_t *tr_list; 1222168404Spjd 1223185029Spjd if (asize == 0) { 1224185029Spjd *tr_cookiep = NULL; 1225185029Spjd return (0); 1226185029Spjd } 1227185029Spjd 1228168404Spjd tr_list = kmem_alloc(sizeof (list_t), KM_SLEEP); 1229168404Spjd list_create(tr_list, sizeof (struct tempreserve), 1230168404Spjd offsetof(struct tempreserve, tr_node)); 1231185029Spjd ASSERT3S(asize, >, 0); 1232168404Spjd ASSERT3S(fsize, >=, 0); 1233168404Spjd 1234185029Spjd err = arc_tempreserve_space(lsize, tx->tx_txg); 1235168404Spjd if (err == 0) { 1236168404Spjd struct tempreserve *tr; 1237168404Spjd 1238185029Spjd tr = kmem_zalloc(sizeof (struct tempreserve), KM_SLEEP); 1239185029Spjd tr->tr_size = lsize; 1240185029Spjd list_insert_tail(tr_list, tr); 1241185029Spjd } else { 1242185029Spjd if (err == EAGAIN) { 1243260763Savg /* 1244260763Savg * If arc_memory_throttle() detected that pageout 1245260763Savg * is running and we are low on memory, we delay new 1246260763Savg * non-pageout transactions to give pageout an 1247260763Savg * advantage. 1248260763Savg * 1249260763Savg * It is unfortunate to be delaying while the caller's 1250260763Savg * locks are held. 1251260763Savg */ 1252255437Sdelphij txg_delay(dd->dd_pool, tx->tx_txg, 1253255437Sdelphij MSEC2NSEC(10), MSEC2NSEC(10)); 1254249195Smm err = SET_ERROR(ERESTART); 1255168404Spjd } 1256168404Spjd } 1257168404Spjd 1258185029Spjd if (err == 0) { 1259185029Spjd err = dsl_dir_tempreserve_impl(dd, asize, fsize >= asize, 1260185029Spjd FALSE, asize > usize, tr_list, tx, TRUE); 1261185029Spjd } 1262185029Spjd 1263248571Smm if (err != 0) 1264168404Spjd dsl_dir_tempreserve_clear(tr_list, tx); 1265168404Spjd else 1266168404Spjd *tr_cookiep = tr_list; 1267185029Spjd 1268168404Spjd return (err); 1269168404Spjd} 1270168404Spjd 1271168404Spjd/* 1272168404Spjd * Clear a temporary reservation that we previously made with 1273168404Spjd * dsl_dir_tempreserve_space(). 1274168404Spjd */ 1275168404Spjdvoid 1276168404Spjddsl_dir_tempreserve_clear(void *tr_cookie, dmu_tx_t *tx) 1277168404Spjd{ 1278168404Spjd int txgidx = tx->tx_txg & TXG_MASK; 1279168404Spjd list_t *tr_list = tr_cookie; 1280168404Spjd struct tempreserve *tr; 1281168404Spjd 1282168404Spjd ASSERT3U(tx->tx_txg, !=, 0); 1283168404Spjd 1284185029Spjd if (tr_cookie == NULL) 1285185029Spjd return; 1286185029Spjd 1287260763Savg while ((tr = list_head(tr_list)) != NULL) { 1288260763Savg if (tr->tr_ds) { 1289168404Spjd mutex_enter(&tr->tr_ds->dd_lock); 1290168404Spjd ASSERT3U(tr->tr_ds->dd_tempreserved[txgidx], >=, 1291168404Spjd tr->tr_size); 1292168404Spjd tr->tr_ds->dd_tempreserved[txgidx] -= tr->tr_size; 1293168404Spjd mutex_exit(&tr->tr_ds->dd_lock); 1294185029Spjd } else { 1295185029Spjd arc_tempreserve_clear(tr->tr_size); 1296168404Spjd } 1297168404Spjd list_remove(tr_list, tr); 1298168404Spjd kmem_free(tr, sizeof (struct tempreserve)); 1299168404Spjd } 1300168404Spjd 1301168404Spjd kmem_free(tr_list, sizeof (list_t)); 1302168404Spjd} 1303168404Spjd 1304260763Savg/* 1305260763Savg * This should be called from open context when we think we're going to write 1306260763Savg * or free space, for example when dirtying data. Be conservative; it's okay 1307260763Savg * to write less space or free more, but we don't want to write more or free 1308260763Savg * less than the amount specified. 1309260763Savg */ 1310260763Savgvoid 1311260763Savgdsl_dir_willuse_space(dsl_dir_t *dd, int64_t space, dmu_tx_t *tx) 1312168404Spjd{ 1313168404Spjd int64_t parent_space; 1314168404Spjd uint64_t est_used; 1315168404Spjd 1316168404Spjd mutex_enter(&dd->dd_lock); 1317168404Spjd if (space > 0) 1318168404Spjd dd->dd_space_towrite[tx->tx_txg & TXG_MASK] += space; 1319168404Spjd 1320185029Spjd est_used = dsl_dir_space_towrite(dd) + dd->dd_phys->dd_used_bytes; 1321168404Spjd parent_space = parent_delta(dd, est_used, space); 1322168404Spjd mutex_exit(&dd->dd_lock); 1323168404Spjd 1324168404Spjd /* Make sure that we clean up dd_space_to* */ 1325168404Spjd dsl_dir_dirty(dd, tx); 1326168404Spjd 1327168404Spjd /* XXX this is potentially expensive and unnecessary... */ 1328168404Spjd if (parent_space && dd->dd_parent) 1329260763Savg dsl_dir_willuse_space(dd->dd_parent, parent_space, tx); 1330168404Spjd} 1331168404Spjd 1332168404Spjd/* call from syncing context when we actually write/free space for this dd */ 1333168404Spjdvoid 1334185029Spjddsl_dir_diduse_space(dsl_dir_t *dd, dd_used_t type, 1335168404Spjd int64_t used, int64_t compressed, int64_t uncompressed, dmu_tx_t *tx) 1336168404Spjd{ 1337168404Spjd int64_t accounted_delta; 1338254757Sdelphij 1339254757Sdelphij /* 1340254757Sdelphij * dsl_dataset_set_refreservation_sync_impl() calls this with 1341254757Sdelphij * dd_lock held, so that it can atomically update 1342254757Sdelphij * ds->ds_reserved and the dsl_dir accounting, so that 1343254757Sdelphij * dsl_dataset_check_quota() can see dataset and dir accounting 1344254757Sdelphij * consistently. 1345254757Sdelphij */ 1346185029Spjd boolean_t needlock = !MUTEX_HELD(&dd->dd_lock); 1347168404Spjd 1348168404Spjd ASSERT(dmu_tx_is_syncing(tx)); 1349185029Spjd ASSERT(type < DD_USED_NUM); 1350168404Spjd 1351254757Sdelphij dmu_buf_will_dirty(dd->dd_dbuf, tx); 1352254757Sdelphij 1353185029Spjd if (needlock) 1354185029Spjd mutex_enter(&dd->dd_lock); 1355185029Spjd accounted_delta = parent_delta(dd, dd->dd_phys->dd_used_bytes, used); 1356185029Spjd ASSERT(used >= 0 || dd->dd_phys->dd_used_bytes >= -used); 1357168404Spjd ASSERT(compressed >= 0 || 1358168404Spjd dd->dd_phys->dd_compressed_bytes >= -compressed); 1359168404Spjd ASSERT(uncompressed >= 0 || 1360168404Spjd dd->dd_phys->dd_uncompressed_bytes >= -uncompressed); 1361185029Spjd dd->dd_phys->dd_used_bytes += used; 1362168404Spjd dd->dd_phys->dd_uncompressed_bytes += uncompressed; 1363168404Spjd dd->dd_phys->dd_compressed_bytes += compressed; 1364168404Spjd 1365185029Spjd if (dd->dd_phys->dd_flags & DD_FLAG_USED_BREAKDOWN) { 1366185029Spjd ASSERT(used > 0 || 1367185029Spjd dd->dd_phys->dd_used_breakdown[type] >= -used); 1368185029Spjd dd->dd_phys->dd_used_breakdown[type] += used; 1369185029Spjd#ifdef DEBUG 1370185029Spjd dd_used_t t; 1371185029Spjd uint64_t u = 0; 1372185029Spjd for (t = 0; t < DD_USED_NUM; t++) 1373185029Spjd u += dd->dd_phys->dd_used_breakdown[t]; 1374185029Spjd ASSERT3U(u, ==, dd->dd_phys->dd_used_bytes); 1375185029Spjd#endif 1376185029Spjd } 1377185029Spjd if (needlock) 1378185029Spjd mutex_exit(&dd->dd_lock); 1379185029Spjd 1380168404Spjd if (dd->dd_parent != NULL) { 1381185029Spjd dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD, 1382168404Spjd accounted_delta, compressed, uncompressed, tx); 1383185029Spjd dsl_dir_transfer_space(dd->dd_parent, 1384185029Spjd used - accounted_delta, 1385185029Spjd DD_USED_CHILD_RSRV, DD_USED_CHILD, tx); 1386168404Spjd } 1387168404Spjd} 1388168404Spjd 1389185029Spjdvoid 1390185029Spjddsl_dir_transfer_space(dsl_dir_t *dd, int64_t delta, 1391185029Spjd dd_used_t oldtype, dd_used_t newtype, dmu_tx_t *tx) 1392185029Spjd{ 1393185029Spjd ASSERT(dmu_tx_is_syncing(tx)); 1394185029Spjd ASSERT(oldtype < DD_USED_NUM); 1395185029Spjd ASSERT(newtype < DD_USED_NUM); 1396185029Spjd 1397185029Spjd if (delta == 0 || !(dd->dd_phys->dd_flags & DD_FLAG_USED_BREAKDOWN)) 1398185029Spjd return; 1399185029Spjd 1400254757Sdelphij dmu_buf_will_dirty(dd->dd_dbuf, tx); 1401254757Sdelphij mutex_enter(&dd->dd_lock); 1402185029Spjd ASSERT(delta > 0 ? 1403185029Spjd dd->dd_phys->dd_used_breakdown[oldtype] >= delta : 1404185029Spjd dd->dd_phys->dd_used_breakdown[newtype] >= -delta); 1405185029Spjd ASSERT(dd->dd_phys->dd_used_bytes >= ABS(delta)); 1406185029Spjd dd->dd_phys->dd_used_breakdown[oldtype] -= delta; 1407185029Spjd dd->dd_phys->dd_used_breakdown[newtype] += delta; 1408254757Sdelphij mutex_exit(&dd->dd_lock); 1409185029Spjd} 1410185029Spjd 1411248571Smmtypedef struct dsl_dir_set_qr_arg { 1412248571Smm const char *ddsqra_name; 1413248571Smm zprop_source_t ddsqra_source; 1414248571Smm uint64_t ddsqra_value; 1415248571Smm} dsl_dir_set_qr_arg_t; 1416248571Smm 1417168404Spjdstatic int 1418248571Smmdsl_dir_set_quota_check(void *arg, dmu_tx_t *tx) 1419168404Spjd{ 1420248571Smm dsl_dir_set_qr_arg_t *ddsqra = arg; 1421248571Smm dsl_pool_t *dp = dmu_tx_pool(tx); 1422248571Smm dsl_dataset_t *ds; 1423248571Smm int error; 1424248571Smm uint64_t towrite, newval; 1425168404Spjd 1426248571Smm error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds); 1427248571Smm if (error != 0) 1428248571Smm return (error); 1429219089Spjd 1430248571Smm error = dsl_prop_predict(ds->ds_dir, "quota", 1431248571Smm ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval); 1432248571Smm if (error != 0) { 1433248571Smm dsl_dataset_rele(ds, FTAG); 1434248571Smm return (error); 1435248571Smm } 1436248571Smm 1437248571Smm if (newval == 0) { 1438248571Smm dsl_dataset_rele(ds, FTAG); 1439168404Spjd return (0); 1440248571Smm } 1441168404Spjd 1442248571Smm mutex_enter(&ds->ds_dir->dd_lock); 1443168404Spjd /* 1444168404Spjd * If we are doing the preliminary check in open context, and 1445168404Spjd * there are pending changes, then don't fail it, since the 1446185029Spjd * pending changes could under-estimate the amount of space to be 1447168404Spjd * freed up. 1448168404Spjd */ 1449248571Smm towrite = dsl_dir_space_towrite(ds->ds_dir); 1450168404Spjd if ((dmu_tx_is_syncing(tx) || towrite == 0) && 1451248571Smm (newval < ds->ds_dir->dd_phys->dd_reserved || 1452248571Smm newval < ds->ds_dir->dd_phys->dd_used_bytes + towrite)) { 1453249195Smm error = SET_ERROR(ENOSPC); 1454168404Spjd } 1455248571Smm mutex_exit(&ds->ds_dir->dd_lock); 1456248571Smm dsl_dataset_rele(ds, FTAG); 1457248571Smm return (error); 1458168404Spjd} 1459168404Spjd 1460168404Spjdstatic void 1461248571Smmdsl_dir_set_quota_sync(void *arg, dmu_tx_t *tx) 1462168404Spjd{ 1463248571Smm dsl_dir_set_qr_arg_t *ddsqra = arg; 1464248571Smm dsl_pool_t *dp = dmu_tx_pool(tx); 1465248571Smm dsl_dataset_t *ds; 1466248571Smm uint64_t newval; 1467168404Spjd 1468248571Smm VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds)); 1469219089Spjd 1470249787Smm if (spa_version(dp->dp_spa) >= SPA_VERSION_RECVD_PROPS) { 1471249787Smm dsl_prop_set_sync_impl(ds, zfs_prop_to_name(ZFS_PROP_QUOTA), 1472249787Smm ddsqra->ddsqra_source, sizeof (ddsqra->ddsqra_value), 1, 1473249787Smm &ddsqra->ddsqra_value, tx); 1474168404Spjd 1475249787Smm VERIFY0(dsl_prop_get_int_ds(ds, 1476249787Smm zfs_prop_to_name(ZFS_PROP_QUOTA), &newval)); 1477249787Smm } else { 1478249787Smm newval = ddsqra->ddsqra_value; 1479249787Smm spa_history_log_internal_ds(ds, "set", tx, "%s=%lld", 1480249787Smm zfs_prop_to_name(ZFS_PROP_QUOTA), (longlong_t)newval); 1481249787Smm } 1482248571Smm 1483248571Smm dmu_buf_will_dirty(ds->ds_dir->dd_dbuf, tx); 1484248571Smm mutex_enter(&ds->ds_dir->dd_lock); 1485248571Smm ds->ds_dir->dd_phys->dd_quota = newval; 1486248571Smm mutex_exit(&ds->ds_dir->dd_lock); 1487248571Smm dsl_dataset_rele(ds, FTAG); 1488168404Spjd} 1489168404Spjd 1490168404Spjdint 1491219089Spjddsl_dir_set_quota(const char *ddname, zprop_source_t source, uint64_t quota) 1492168404Spjd{ 1493248571Smm dsl_dir_set_qr_arg_t ddsqra; 1494168404Spjd 1495248571Smm ddsqra.ddsqra_name = ddname; 1496248571Smm ddsqra.ddsqra_source = source; 1497248571Smm ddsqra.ddsqra_value = quota; 1498219089Spjd 1499248571Smm return (dsl_sync_task(ddname, dsl_dir_set_quota_check, 1500269006Sdelphij dsl_dir_set_quota_sync, &ddsqra, 0, ZFS_SPACE_CHECK_NONE)); 1501168404Spjd} 1502168404Spjd 1503185029Spjdint 1504248571Smmdsl_dir_set_reservation_check(void *arg, dmu_tx_t *tx) 1505168404Spjd{ 1506248571Smm dsl_dir_set_qr_arg_t *ddsqra = arg; 1507248571Smm dsl_pool_t *dp = dmu_tx_pool(tx); 1508248571Smm dsl_dataset_t *ds; 1509248571Smm dsl_dir_t *dd; 1510248571Smm uint64_t newval, used, avail; 1511248571Smm int error; 1512168404Spjd 1513248571Smm error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds); 1514248571Smm if (error != 0) 1515248571Smm return (error); 1516248571Smm dd = ds->ds_dir; 1517219089Spjd 1518168404Spjd /* 1519168404Spjd * If we are doing the preliminary check in open context, the 1520168404Spjd * space estimates may be inaccurate. 1521168404Spjd */ 1522248571Smm if (!dmu_tx_is_syncing(tx)) { 1523248571Smm dsl_dataset_rele(ds, FTAG); 1524168404Spjd return (0); 1525248571Smm } 1526168404Spjd 1527248571Smm error = dsl_prop_predict(ds->ds_dir, 1528248571Smm zfs_prop_to_name(ZFS_PROP_RESERVATION), 1529248571Smm ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval); 1530248571Smm if (error != 0) { 1531248571Smm dsl_dataset_rele(ds, FTAG); 1532248571Smm return (error); 1533248571Smm } 1534248571Smm 1535168404Spjd mutex_enter(&dd->dd_lock); 1536185029Spjd used = dd->dd_phys->dd_used_bytes; 1537168404Spjd mutex_exit(&dd->dd_lock); 1538168404Spjd 1539168404Spjd if (dd->dd_parent) { 1540168404Spjd avail = dsl_dir_space_available(dd->dd_parent, 1541168404Spjd NULL, 0, FALSE); 1542168404Spjd } else { 1543168404Spjd avail = dsl_pool_adjustedsize(dd->dd_pool, B_FALSE) - used; 1544168404Spjd } 1545168404Spjd 1546248571Smm if (MAX(used, newval) > MAX(used, dd->dd_phys->dd_reserved)) { 1547248571Smm uint64_t delta = MAX(used, newval) - 1548209962Smm MAX(used, dd->dd_phys->dd_reserved); 1549209962Smm 1550248571Smm if (delta > avail || 1551248571Smm (dd->dd_phys->dd_quota > 0 && 1552248571Smm newval > dd->dd_phys->dd_quota)) 1553249195Smm error = SET_ERROR(ENOSPC); 1554209962Smm } 1555209962Smm 1556248571Smm dsl_dataset_rele(ds, FTAG); 1557248571Smm return (error); 1558168404Spjd} 1559168404Spjd 1560248571Smmvoid 1561248571Smmdsl_dir_set_reservation_sync_impl(dsl_dir_t *dd, uint64_t value, dmu_tx_t *tx) 1562168404Spjd{ 1563168404Spjd uint64_t used; 1564168404Spjd int64_t delta; 1565168404Spjd 1566185029Spjd dmu_buf_will_dirty(dd->dd_dbuf, tx); 1567185029Spjd 1568168404Spjd mutex_enter(&dd->dd_lock); 1569185029Spjd used = dd->dd_phys->dd_used_bytes; 1570248571Smm delta = MAX(used, value) - MAX(used, dd->dd_phys->dd_reserved); 1571248571Smm dd->dd_phys->dd_reserved = value; 1572168404Spjd 1573168404Spjd if (dd->dd_parent != NULL) { 1574168404Spjd /* Roll up this additional usage into our ancestors */ 1575185029Spjd dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD_RSRV, 1576185029Spjd delta, 0, 0, tx); 1577168404Spjd } 1578185029Spjd mutex_exit(&dd->dd_lock); 1579168404Spjd} 1580168404Spjd 1581248571Smmstatic void 1582248571Smmdsl_dir_set_reservation_sync(void *arg, dmu_tx_t *tx) 1583168404Spjd{ 1584248571Smm dsl_dir_set_qr_arg_t *ddsqra = arg; 1585248571Smm dsl_pool_t *dp = dmu_tx_pool(tx); 1586219089Spjd dsl_dataset_t *ds; 1587248571Smm uint64_t newval; 1588168404Spjd 1589248571Smm VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds)); 1590219089Spjd 1591249787Smm if (spa_version(dp->dp_spa) >= SPA_VERSION_RECVD_PROPS) { 1592249787Smm dsl_prop_set_sync_impl(ds, 1593249787Smm zfs_prop_to_name(ZFS_PROP_RESERVATION), 1594249787Smm ddsqra->ddsqra_source, sizeof (ddsqra->ddsqra_value), 1, 1595249787Smm &ddsqra->ddsqra_value, tx); 1596219089Spjd 1597249787Smm VERIFY0(dsl_prop_get_int_ds(ds, 1598249787Smm zfs_prop_to_name(ZFS_PROP_RESERVATION), &newval)); 1599249787Smm } else { 1600249787Smm newval = ddsqra->ddsqra_value; 1601249787Smm spa_history_log_internal_ds(ds, "set", tx, "%s=%lld", 1602249787Smm zfs_prop_to_name(ZFS_PROP_RESERVATION), 1603249787Smm (longlong_t)newval); 1604249787Smm } 1605219089Spjd 1606248571Smm dsl_dir_set_reservation_sync_impl(ds->ds_dir, newval, tx); 1607248571Smm dsl_dataset_rele(ds, FTAG); 1608248571Smm} 1609219089Spjd 1610248571Smmint 1611248571Smmdsl_dir_set_reservation(const char *ddname, zprop_source_t source, 1612248571Smm uint64_t reservation) 1613248571Smm{ 1614248571Smm dsl_dir_set_qr_arg_t ddsqra; 1615219089Spjd 1616248571Smm ddsqra.ddsqra_name = ddname; 1617248571Smm ddsqra.ddsqra_source = source; 1618248571Smm ddsqra.ddsqra_value = reservation; 1619248571Smm 1620248571Smm return (dsl_sync_task(ddname, dsl_dir_set_reservation_check, 1621269006Sdelphij dsl_dir_set_reservation_sync, &ddsqra, 0, ZFS_SPACE_CHECK_NONE)); 1622168404Spjd} 1623168404Spjd 1624168404Spjdstatic dsl_dir_t * 1625168404Spjdclosest_common_ancestor(dsl_dir_t *ds1, dsl_dir_t *ds2) 1626168404Spjd{ 1627168404Spjd for (; ds1; ds1 = ds1->dd_parent) { 1628168404Spjd dsl_dir_t *dd; 1629168404Spjd for (dd = ds2; dd; dd = dd->dd_parent) { 1630168404Spjd if (ds1 == dd) 1631168404Spjd return (dd); 1632168404Spjd } 1633168404Spjd } 1634168404Spjd return (NULL); 1635168404Spjd} 1636168404Spjd 1637168404Spjd/* 1638168404Spjd * If delta is applied to dd, how much of that delta would be applied to 1639168404Spjd * ancestor? Syncing context only. 1640168404Spjd */ 1641168404Spjdstatic int64_t 1642168404Spjdwould_change(dsl_dir_t *dd, int64_t delta, dsl_dir_t *ancestor) 1643168404Spjd{ 1644168404Spjd if (dd == ancestor) 1645168404Spjd return (delta); 1646168404Spjd 1647168404Spjd mutex_enter(&dd->dd_lock); 1648185029Spjd delta = parent_delta(dd, dd->dd_phys->dd_used_bytes, delta); 1649168404Spjd mutex_exit(&dd->dd_lock); 1650168404Spjd return (would_change(dd->dd_parent, delta, ancestor)); 1651168404Spjd} 1652168404Spjd 1653248571Smmtypedef struct dsl_dir_rename_arg { 1654248571Smm const char *ddra_oldname; 1655248571Smm const char *ddra_newname; 1656265744Sdelphij cred_t *ddra_cred; 1657248571Smm} dsl_dir_rename_arg_t; 1658168404Spjd 1659248571Smm/* ARGSUSED */ 1660168404Spjdstatic int 1661248571Smmdsl_valid_rename(dsl_pool_t *dp, dsl_dataset_t *ds, void *arg) 1662168404Spjd{ 1663248571Smm int *deltap = arg; 1664248571Smm char namebuf[MAXNAMELEN]; 1665168404Spjd 1666248571Smm dsl_dataset_name(ds, namebuf); 1667248571Smm 1668248571Smm if (strlen(namebuf) + *deltap >= MAXNAMELEN) 1669249195Smm return (SET_ERROR(ENAMETOOLONG)); 1670248571Smm return (0); 1671248571Smm} 1672248571Smm 1673248571Smmstatic int 1674248571Smmdsl_dir_rename_check(void *arg, dmu_tx_t *tx) 1675248571Smm{ 1676248571Smm dsl_dir_rename_arg_t *ddra = arg; 1677248571Smm dsl_pool_t *dp = dmu_tx_pool(tx); 1678248571Smm dsl_dir_t *dd, *newparent; 1679248571Smm const char *mynewname; 1680248571Smm int error; 1681248571Smm int delta = strlen(ddra->ddra_newname) - strlen(ddra->ddra_oldname); 1682248571Smm 1683248571Smm /* target dir should exist */ 1684248571Smm error = dsl_dir_hold(dp, ddra->ddra_oldname, FTAG, &dd, NULL); 1685248571Smm if (error != 0) 1686248571Smm return (error); 1687248571Smm 1688248571Smm /* new parent should exist */ 1689248571Smm error = dsl_dir_hold(dp, ddra->ddra_newname, FTAG, 1690248571Smm &newparent, &mynewname); 1691248571Smm if (error != 0) { 1692248571Smm dsl_dir_rele(dd, FTAG); 1693248571Smm return (error); 1694226676Spjd } 1695168404Spjd 1696248571Smm /* can't rename to different pool */ 1697248571Smm if (dd->dd_pool != newparent->dd_pool) { 1698248571Smm dsl_dir_rele(newparent, FTAG); 1699248571Smm dsl_dir_rele(dd, FTAG); 1700249195Smm return (SET_ERROR(ENXIO)); 1701248571Smm } 1702248571Smm 1703248571Smm /* new name should not already exist */ 1704248571Smm if (mynewname == NULL) { 1705248571Smm dsl_dir_rele(newparent, FTAG); 1706248571Smm dsl_dir_rele(dd, FTAG); 1707249195Smm return (SET_ERROR(EEXIST)); 1708248571Smm } 1709168404Spjd 1710248571Smm /* if the name length is growing, validate child name lengths */ 1711248571Smm if (delta > 0) { 1712248571Smm error = dmu_objset_find_dp(dp, dd->dd_object, dsl_valid_rename, 1713248571Smm &delta, DS_FIND_CHILDREN | DS_FIND_SNAPSHOTS); 1714248571Smm if (error != 0) { 1715248571Smm dsl_dir_rele(newparent, FTAG); 1716248571Smm dsl_dir_rele(dd, FTAG); 1717248571Smm return (error); 1718248571Smm } 1719248571Smm } 1720248571Smm 1721265744Sdelphij if (dmu_tx_is_syncing(tx)) { 1722267138Sdelphij if (spa_feature_is_active(dp->dp_spa, 1723265744Sdelphij SPA_FEATURE_FS_SS_LIMIT)) { 1724265744Sdelphij /* 1725265744Sdelphij * Although this is the check function and we don't 1726265744Sdelphij * normally make on-disk changes in check functions, 1727265744Sdelphij * we need to do that here. 1728265744Sdelphij * 1729265744Sdelphij * Ensure this portion of the tree's counts have been 1730265744Sdelphij * initialized in case the new parent has limits set. 1731265744Sdelphij */ 1732265744Sdelphij dsl_dir_init_fs_ss_count(dd, tx); 1733265744Sdelphij } 1734265744Sdelphij } 1735265744Sdelphij 1736248571Smm if (newparent != dd->dd_parent) { 1737168404Spjd /* is there enough space? */ 1738168404Spjd uint64_t myspace = 1739185029Spjd MAX(dd->dd_phys->dd_used_bytes, dd->dd_phys->dd_reserved); 1740265744Sdelphij objset_t *os = dd->dd_pool->dp_meta_objset; 1741265744Sdelphij uint64_t fs_cnt = 0; 1742265744Sdelphij uint64_t ss_cnt = 0; 1743168404Spjd 1744265744Sdelphij if (dsl_dir_is_zapified(dd)) { 1745265744Sdelphij int err; 1746265744Sdelphij 1747265744Sdelphij err = zap_lookup(os, dd->dd_object, 1748265744Sdelphij DD_FIELD_FILESYSTEM_COUNT, sizeof (fs_cnt), 1, 1749265744Sdelphij &fs_cnt); 1750267138Sdelphij if (err != ENOENT && err != 0) { 1751267138Sdelphij dsl_dir_rele(newparent, FTAG); 1752267138Sdelphij dsl_dir_rele(dd, FTAG); 1753265744Sdelphij return (err); 1754267138Sdelphij } 1755265744Sdelphij 1756265744Sdelphij /* 1757265744Sdelphij * have to add 1 for the filesystem itself that we're 1758265744Sdelphij * moving 1759265744Sdelphij */ 1760265744Sdelphij fs_cnt++; 1761265744Sdelphij 1762265744Sdelphij err = zap_lookup(os, dd->dd_object, 1763265744Sdelphij DD_FIELD_SNAPSHOT_COUNT, sizeof (ss_cnt), 1, 1764265744Sdelphij &ss_cnt); 1765267138Sdelphij if (err != ENOENT && err != 0) { 1766267138Sdelphij dsl_dir_rele(newparent, FTAG); 1767267138Sdelphij dsl_dir_rele(dd, FTAG); 1768265744Sdelphij return (err); 1769267138Sdelphij } 1770265744Sdelphij } 1771265744Sdelphij 1772168404Spjd /* no rename into our descendant */ 1773248571Smm if (closest_common_ancestor(dd, newparent) == dd) { 1774248571Smm dsl_dir_rele(newparent, FTAG); 1775248571Smm dsl_dir_rele(dd, FTAG); 1776249195Smm return (SET_ERROR(EINVAL)); 1777248571Smm } 1778168404Spjd 1779248571Smm error = dsl_dir_transfer_possible(dd->dd_parent, 1780265744Sdelphij newparent, fs_cnt, ss_cnt, myspace, ddra->ddra_cred); 1781248571Smm if (error != 0) { 1782248571Smm dsl_dir_rele(newparent, FTAG); 1783248571Smm dsl_dir_rele(dd, FTAG); 1784248571Smm return (error); 1785248571Smm } 1786168404Spjd } 1787168404Spjd 1788248571Smm dsl_dir_rele(newparent, FTAG); 1789248571Smm dsl_dir_rele(dd, FTAG); 1790168404Spjd return (0); 1791168404Spjd} 1792168404Spjd 1793168404Spjdstatic void 1794248571Smmdsl_dir_rename_sync(void *arg, dmu_tx_t *tx) 1795168404Spjd{ 1796248571Smm dsl_dir_rename_arg_t *ddra = arg; 1797248571Smm dsl_pool_t *dp = dmu_tx_pool(tx); 1798248571Smm dsl_dir_t *dd, *newparent; 1799248571Smm const char *mynewname; 1800248571Smm int error; 1801168404Spjd objset_t *mos = dp->dp_meta_objset; 1802168404Spjd 1803248571Smm VERIFY0(dsl_dir_hold(dp, ddra->ddra_oldname, FTAG, &dd, NULL)); 1804248571Smm VERIFY0(dsl_dir_hold(dp, ddra->ddra_newname, FTAG, &newparent, 1805248571Smm &mynewname)); 1806248571Smm 1807248571Smm /* Log this before we change the name. */ 1808248571Smm spa_history_log_internal_dd(dd, "rename", tx, 1809248571Smm "-> %s", ddra->ddra_newname); 1810248571Smm 1811248571Smm if (newparent != dd->dd_parent) { 1812265744Sdelphij objset_t *os = dd->dd_pool->dp_meta_objset; 1813265744Sdelphij uint64_t fs_cnt = 0; 1814265744Sdelphij uint64_t ss_cnt = 0; 1815265744Sdelphij 1816265744Sdelphij /* 1817265744Sdelphij * We already made sure the dd counts were initialized in the 1818265744Sdelphij * check function. 1819265744Sdelphij */ 1820267138Sdelphij if (spa_feature_is_active(dp->dp_spa, 1821265744Sdelphij SPA_FEATURE_FS_SS_LIMIT)) { 1822265744Sdelphij VERIFY0(zap_lookup(os, dd->dd_object, 1823265744Sdelphij DD_FIELD_FILESYSTEM_COUNT, sizeof (fs_cnt), 1, 1824265744Sdelphij &fs_cnt)); 1825265744Sdelphij /* add 1 for the filesystem itself that we're moving */ 1826265744Sdelphij fs_cnt++; 1827265744Sdelphij 1828265744Sdelphij VERIFY0(zap_lookup(os, dd->dd_object, 1829265744Sdelphij DD_FIELD_SNAPSHOT_COUNT, sizeof (ss_cnt), 1, 1830265744Sdelphij &ss_cnt)); 1831265744Sdelphij } 1832265744Sdelphij 1833265744Sdelphij dsl_fs_ss_count_adjust(dd->dd_parent, -fs_cnt, 1834265744Sdelphij DD_FIELD_FILESYSTEM_COUNT, tx); 1835265744Sdelphij dsl_fs_ss_count_adjust(newparent, fs_cnt, 1836265744Sdelphij DD_FIELD_FILESYSTEM_COUNT, tx); 1837265744Sdelphij 1838265744Sdelphij dsl_fs_ss_count_adjust(dd->dd_parent, -ss_cnt, 1839265744Sdelphij DD_FIELD_SNAPSHOT_COUNT, tx); 1840265744Sdelphij dsl_fs_ss_count_adjust(newparent, ss_cnt, 1841265744Sdelphij DD_FIELD_SNAPSHOT_COUNT, tx); 1842265744Sdelphij 1843185029Spjd dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD, 1844185029Spjd -dd->dd_phys->dd_used_bytes, 1845168404Spjd -dd->dd_phys->dd_compressed_bytes, 1846168404Spjd -dd->dd_phys->dd_uncompressed_bytes, tx); 1847248571Smm dsl_dir_diduse_space(newparent, DD_USED_CHILD, 1848185029Spjd dd->dd_phys->dd_used_bytes, 1849168404Spjd dd->dd_phys->dd_compressed_bytes, 1850168404Spjd dd->dd_phys->dd_uncompressed_bytes, tx); 1851185029Spjd 1852185029Spjd if (dd->dd_phys->dd_reserved > dd->dd_phys->dd_used_bytes) { 1853185029Spjd uint64_t unused_rsrv = dd->dd_phys->dd_reserved - 1854185029Spjd dd->dd_phys->dd_used_bytes; 1855185029Spjd 1856185029Spjd dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD_RSRV, 1857185029Spjd -unused_rsrv, 0, 0, tx); 1858248571Smm dsl_dir_diduse_space(newparent, DD_USED_CHILD_RSRV, 1859185029Spjd unused_rsrv, 0, 0, tx); 1860185029Spjd } 1861168404Spjd } 1862168404Spjd 1863168404Spjd dmu_buf_will_dirty(dd->dd_dbuf, tx); 1864168404Spjd 1865168404Spjd /* remove from old parent zapobj */ 1866248571Smm error = zap_remove(mos, dd->dd_parent->dd_phys->dd_child_dir_zapobj, 1867168404Spjd dd->dd_myname, tx); 1868248571Smm ASSERT0(error); 1869168404Spjd 1870248571Smm (void) strcpy(dd->dd_myname, mynewname); 1871248571Smm dsl_dir_rele(dd->dd_parent, dd); 1872248571Smm dd->dd_phys->dd_parent_obj = newparent->dd_object; 1873248571Smm VERIFY0(dsl_dir_hold_obj(dp, 1874248571Smm newparent->dd_object, NULL, dd, &dd->dd_parent)); 1875168404Spjd 1876168404Spjd /* add to new parent zapobj */ 1877248571Smm VERIFY0(zap_add(mos, newparent->dd_phys->dd_child_dir_zapobj, 1878248571Smm dd->dd_myname, 8, 1, &dd->dd_object, tx)); 1879248571Smm 1880248571Smm#ifdef __FreeBSD__ 1881219320Spjd#ifdef _KERNEL 1882248571Smm zfsvfs_update_fromname(ddra->ddra_oldname, ddra->ddra_newname); 1883248571Smm zvol_rename_minors(ddra->ddra_oldname, ddra->ddra_newname); 1884219320Spjd#endif 1885248571Smm#endif 1886185029Spjd 1887248571Smm dsl_prop_notify_all(dd); 1888248571Smm 1889248571Smm dsl_dir_rele(newparent, FTAG); 1890248571Smm dsl_dir_rele(dd, FTAG); 1891168404Spjd} 1892168404Spjd 1893168404Spjdint 1894248571Smmdsl_dir_rename(const char *oldname, const char *newname) 1895168404Spjd{ 1896248571Smm dsl_dir_rename_arg_t ddra; 1897168404Spjd 1898248571Smm ddra.ddra_oldname = oldname; 1899248571Smm ddra.ddra_newname = newname; 1900265744Sdelphij ddra.ddra_cred = CRED(); 1901168404Spjd 1902248571Smm return (dsl_sync_task(oldname, 1903269006Sdelphij dsl_dir_rename_check, dsl_dir_rename_sync, &ddra, 1904269006Sdelphij 3, ZFS_SPACE_CHECK_RESERVED)); 1905168404Spjd} 1906168404Spjd 1907168404Spjdint 1908265744Sdelphijdsl_dir_transfer_possible(dsl_dir_t *sdd, dsl_dir_t *tdd, 1909265744Sdelphij uint64_t fs_cnt, uint64_t ss_cnt, uint64_t space, cred_t *cr) 1910168404Spjd{ 1911168404Spjd dsl_dir_t *ancestor; 1912168404Spjd int64_t adelta; 1913168404Spjd uint64_t avail; 1914265744Sdelphij int err; 1915168404Spjd 1916168404Spjd ancestor = closest_common_ancestor(sdd, tdd); 1917168404Spjd adelta = would_change(sdd, -space, ancestor); 1918168404Spjd avail = dsl_dir_space_available(tdd, ancestor, adelta, FALSE); 1919168404Spjd if (avail < space) 1920249195Smm return (SET_ERROR(ENOSPC)); 1921168404Spjd 1922265744Sdelphij err = dsl_fs_ss_limit_check(tdd, fs_cnt, ZFS_PROP_FILESYSTEM_LIMIT, 1923265744Sdelphij ancestor, cr); 1924265744Sdelphij if (err != 0) 1925265744Sdelphij return (err); 1926265744Sdelphij err = dsl_fs_ss_limit_check(tdd, ss_cnt, ZFS_PROP_SNAPSHOT_LIMIT, 1927265744Sdelphij ancestor, cr); 1928265744Sdelphij if (err != 0) 1929265744Sdelphij return (err); 1930265744Sdelphij 1931168404Spjd return (0); 1932168404Spjd} 1933219089Spjd 1934219089Spjdtimestruc_t 1935219089Spjddsl_dir_snap_cmtime(dsl_dir_t *dd) 1936219089Spjd{ 1937219089Spjd timestruc_t t; 1938219089Spjd 1939219089Spjd mutex_enter(&dd->dd_lock); 1940219089Spjd t = dd->dd_snap_cmtime; 1941219089Spjd mutex_exit(&dd->dd_lock); 1942219089Spjd 1943219089Spjd return (t); 1944219089Spjd} 1945219089Spjd 1946219089Spjdvoid 1947219089Spjddsl_dir_snap_cmtime_update(dsl_dir_t *dd) 1948219089Spjd{ 1949219089Spjd timestruc_t t; 1950219089Spjd 1951219089Spjd gethrestime(&t); 1952219089Spjd mutex_enter(&dd->dd_lock); 1953219089Spjd dd->dd_snap_cmtime = t; 1954219089Spjd mutex_exit(&dd->dd_lock); 1955219089Spjd} 1956263390Sdelphij 1957263390Sdelphijvoid 1958263390Sdelphijdsl_dir_zapify(dsl_dir_t *dd, dmu_tx_t *tx) 1959263390Sdelphij{ 1960263390Sdelphij objset_t *mos = dd->dd_pool->dp_meta_objset; 1961263390Sdelphij dmu_object_zapify(mos, dd->dd_object, DMU_OT_DSL_DIR, tx); 1962263390Sdelphij} 1963265744Sdelphij 1964265744Sdelphijboolean_t 1965265744Sdelphijdsl_dir_is_zapified(dsl_dir_t *dd) 1966265744Sdelphij{ 1967265744Sdelphij dmu_object_info_t doi; 1968265744Sdelphij 1969265744Sdelphij dmu_object_info_from_db(dd->dd_dbuf, &doi); 1970265744Sdelphij return (doi.doi_type == DMU_OTN_ZAP_METADATA); 1971265744Sdelphij} 1972