1fa9e4066Sahrens /* 2fa9e4066Sahrens * CDDL HEADER START 3fa9e4066Sahrens * 4fa9e4066Sahrens * The contents of this file are subject to the terms of the 5ea8dc4b6Seschrock * Common Development and Distribution License (the "License"). 6ea8dc4b6Seschrock * You may not use this file except in compliance with the License. 7fa9e4066Sahrens * 8fa9e4066Sahrens * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 9fa9e4066Sahrens * or http://www.opensolaris.org/os/licensing. 10fa9e4066Sahrens * See the License for the specific language governing permissions 11fa9e4066Sahrens * and limitations under the License. 12fa9e4066Sahrens * 13fa9e4066Sahrens * When distributing Covered Code, include this CDDL HEADER in each 14fa9e4066Sahrens * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 15fa9e4066Sahrens * If applicable, add the following below this CDDL HEADER, with the 16fa9e4066Sahrens * fields enclosed by brackets "[]" replaced with your own identifying 17fa9e4066Sahrens * information: Portions Copyright [yyyy] [name of copyright owner] 18fa9e4066Sahrens * 19fa9e4066Sahrens * CDDL HEADER END 20fa9e4066Sahrens */ 21fa9e4066Sahrens /* 223f9d6ad7SLin Ling * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. 237d46dc6cSMatthew Ahrens * Copyright (c) 2012, 2014 by Delphix. All rights reserved. 24013023d4SMartin Matuska * Copyright (c) 2013 Martin Matuska. All rights reserved. 25a2afb611SJerry Jelinek * Copyright (c) 2014 Joyent, Inc. All rights reserved. 26fa9e4066Sahrens */ 27fa9e4066Sahrens 28fa9e4066Sahrens #include <sys/dmu.h> 29a9799022Sck #include <sys/dmu_objset.h> 30fa9e4066Sahrens #include <sys/dmu_tx.h> 31fa9e4066Sahrens #include <sys/dsl_dataset.h> 32fa9e4066Sahrens #include <sys/dsl_dir.h> 33fa9e4066Sahrens #include <sys/dsl_prop.h> 341d452cf5Sahrens #include <sys/dsl_synctask.h> 35ecd6cf80Smarks #include <sys/dsl_deleg.h> 362acef22dSMatthew Ahrens #include <sys/dmu_impl.h> 37fa9e4066Sahrens #include <sys/spa.h> 38b24ab676SJeff Bonwick #include <sys/metaslab.h> 39fa9e4066Sahrens #include <sys/zap.h> 40fa9e4066Sahrens #include <sys/zio.h> 41fa9e4066Sahrens #include <sys/arc.h> 42ecd6cf80Smarks #include <sys/sunddi.h> 43a2afb611SJerry Jelinek #include <sys/zfeature.h> 44a2afb611SJerry Jelinek #include <sys/policy.h> 45a2afb611SJerry Jelinek #include <sys/zfs_znode.h> 46fa9e4066Sahrens #include "zfs_namecheck.h" 47a2afb611SJerry Jelinek #include "zfs_prop.h" 48a2afb611SJerry Jelinek 49a2afb611SJerry Jelinek /* 50a2afb611SJerry Jelinek * Filesystem and Snapshot Limits 51a2afb611SJerry Jelinek * ------------------------------ 52a2afb611SJerry Jelinek * 53a2afb611SJerry Jelinek * These limits are used to restrict the number of filesystems and/or snapshots 54a2afb611SJerry Jelinek * that can be created at a given level in the tree or below. A typical 55a2afb611SJerry Jelinek * use-case is with a delegated dataset where the administrator wants to ensure 56a2afb611SJerry Jelinek * that a user within the zone is not creating too many additional filesystems 57a2afb611SJerry Jelinek * or snapshots, even though they're not exceeding their space quota. 58a2afb611SJerry Jelinek * 59a2afb611SJerry Jelinek * The filesystem and snapshot counts are stored as extensible properties. This 60a2afb611SJerry Jelinek * capability is controlled by a feature flag and must be enabled to be used. 61a2afb611SJerry Jelinek * Once enabled, the feature is not active until the first limit is set. At 62a2afb611SJerry Jelinek * that point, future operations to create/destroy filesystems or snapshots 63a2afb611SJerry Jelinek * will validate and update the counts. 64a2afb611SJerry Jelinek * 65a2afb611SJerry Jelinek * Because the count properties will not exist before the feature is active, 66a2afb611SJerry Jelinek * the counts are updated when a limit is first set on an uninitialized 67a2afb611SJerry Jelinek * dsl_dir node in the tree (The filesystem/snapshot count on a node includes 68a2afb611SJerry Jelinek * all of the nested filesystems/snapshots. Thus, a new leaf node has a 69a2afb611SJerry Jelinek * filesystem count of 0 and a snapshot count of 0. Non-existent filesystem and 70a2afb611SJerry Jelinek * snapshot count properties on a node indicate uninitialized counts on that 71a2afb611SJerry Jelinek * node.) When first setting a limit on an uninitialized node, the code starts 72a2afb611SJerry Jelinek * at the filesystem with the new limit and descends into all sub-filesystems 73a2afb611SJerry Jelinek * to add the count properties. 74a2afb611SJerry Jelinek * 75a2afb611SJerry Jelinek * In practice this is lightweight since a limit is typically set when the 76a2afb611SJerry Jelinek * filesystem is created and thus has no children. Once valid, changing the 77a2afb611SJerry Jelinek * limit value won't require a re-traversal since the counts are already valid. 78a2afb611SJerry Jelinek * When recursively fixing the counts, if a node with a limit is encountered 79a2afb611SJerry Jelinek * during the descent, the counts are known to be valid and there is no need to 80a2afb611SJerry Jelinek * descend into that filesystem's children. The counts on filesystems above the 81a2afb611SJerry Jelinek * one with the new limit will still be uninitialized, unless a limit is 82a2afb611SJerry Jelinek * eventually set on one of those filesystems. The counts are always recursively 83a2afb611SJerry Jelinek * updated when a limit is set on a dataset, unless there is already a limit. 84a2afb611SJerry Jelinek * When a new limit value is set on a filesystem with an existing limit, it is 85a2afb611SJerry Jelinek * possible for the new limit to be less than the current count at that level 86a2afb611SJerry Jelinek * since a user who can change the limit is also allowed to exceed the limit. 87a2afb611SJerry Jelinek * 88a2afb611SJerry Jelinek * Once the feature is active, then whenever a filesystem or snapshot is 89a2afb611SJerry Jelinek * created, the code recurses up the tree, validating the new count against the 90a2afb611SJerry Jelinek * limit at each initialized level. In practice, most levels will not have a 91a2afb611SJerry Jelinek * limit set. If there is a limit at any initialized level up the tree, the 92a2afb611SJerry Jelinek * check must pass or the creation will fail. Likewise, when a filesystem or 93a2afb611SJerry Jelinek * snapshot is destroyed, the counts are recursively adjusted all the way up 94a2afb611SJerry Jelinek * the initizized nodes in the tree. Renaming a filesystem into different point 95a2afb611SJerry Jelinek * in the tree will first validate, then update the counts on each branch up to 96a2afb611SJerry Jelinek * the common ancestor. A receive will also validate the counts and then update 97a2afb611SJerry Jelinek * them. 98a2afb611SJerry Jelinek * 99a2afb611SJerry Jelinek * An exception to the above behavior is that the limit is not enforced if the 100a2afb611SJerry Jelinek * user has permission to modify the limit. This is primarily so that 101a2afb611SJerry Jelinek * recursive snapshots in the global zone always work. We want to prevent a 102a2afb611SJerry Jelinek * denial-of-service in which a lower level delegated dataset could max out its 103a2afb611SJerry Jelinek * limit and thus block recursive snapshots from being taken in the global zone. 104a2afb611SJerry Jelinek * Because of this, it is possible for the snapshot count to be over the limit 105a2afb611SJerry Jelinek * and snapshots taken in the global zone could cause a lower level dataset to 106a2afb611SJerry Jelinek * hit or exceed its limit. The administrator taking the global zone recursive 107a2afb611SJerry Jelinek * snapshot should be aware of this side-effect and behave accordingly. 108a2afb611SJerry Jelinek * For consistency, the filesystem limit is also not enforced if the user can 109a2afb611SJerry Jelinek * modify the limit. 110a2afb611SJerry Jelinek * 111a2afb611SJerry Jelinek * The filesystem and snapshot limits are validated by dsl_fs_ss_limit_check() 112a2afb611SJerry Jelinek * and updated by dsl_fs_ss_count_adjust(). A new limit value is setup in 113a2afb611SJerry Jelinek * dsl_dir_activate_fs_ss_limit() and the counts are adjusted, if necessary, by 114a2afb611SJerry Jelinek * dsl_dir_init_fs_ss_count(). 115a2afb611SJerry Jelinek * 116a2afb611SJerry Jelinek * There is a special case when we receive a filesystem that already exists. In 117a2afb611SJerry Jelinek * this case a temporary clone name of %X is created (see dmu_recv_begin). We 118a2afb611SJerry Jelinek * never update the filesystem counts for temporary clones. 119a2afb611SJerry Jelinek * 120a2afb611SJerry Jelinek * Likewise, we do not update the snapshot counts for temporary snapshots, 121a2afb611SJerry Jelinek * such as those created by zfs diff. 122a2afb611SJerry Jelinek */ 123fa9e4066Sahrens 124*c1379625SJustin T. Gibbs extern inline dsl_dir_phys_t *dsl_dir_phys(dsl_dir_t *dd); 125*c1379625SJustin T. Gibbs 126a9799022Sck static uint64_t dsl_dir_space_towrite(dsl_dir_t *dd); 127fa9e4066Sahrens 128fa9e4066Sahrens /* ARGSUSED */ 129fa9e4066Sahrens static void 130fa9e4066Sahrens dsl_dir_evict(dmu_buf_t *db, void *arg) 131fa9e4066Sahrens { 132fa9e4066Sahrens dsl_dir_t *dd = arg; 133fa9e4066Sahrens dsl_pool_t *dp = dd->dd_pool; 134fa9e4066Sahrens int t; 135fa9e4066Sahrens 136fa9e4066Sahrens for (t = 0; t < TXG_SIZE; t++) { 137fa9e4066Sahrens ASSERT(!txg_list_member(&dp->dp_dirty_dirs, dd, t)); 138fa9e4066Sahrens ASSERT(dd->dd_tempreserved[t] == 0); 139fa9e4066Sahrens ASSERT(dd->dd_space_towrite[t] == 0); 140fa9e4066Sahrens } 141fa9e4066Sahrens 142fa9e4066Sahrens if (dd->dd_parent) 1433b2aab18SMatthew Ahrens dsl_dir_rele(dd->dd_parent, dd); 144fa9e4066Sahrens 145fa9e4066Sahrens spa_close(dd->dd_pool->dp_spa, dd); 146fa9e4066Sahrens 147fa9e4066Sahrens /* 1483f9d6ad7SLin Ling * The props callback list should have been cleaned up by 1493f9d6ad7SLin Ling * objset_evict(). 150fa9e4066Sahrens */ 151fa9e4066Sahrens list_destroy(&dd->dd_prop_cbs); 1525ad82045Snd mutex_destroy(&dd->dd_lock); 153fa9e4066Sahrens kmem_free(dd, sizeof (dsl_dir_t)); 154fa9e4066Sahrens } 155fa9e4066Sahrens 156ea8dc4b6Seschrock int 1573b2aab18SMatthew Ahrens dsl_dir_hold_obj(dsl_pool_t *dp, uint64_t ddobj, 158ea8dc4b6Seschrock const char *tail, void *tag, dsl_dir_t **ddp) 159fa9e4066Sahrens { 160fa9e4066Sahrens dmu_buf_t *dbuf; 161fa9e4066Sahrens dsl_dir_t *dd; 162ea8dc4b6Seschrock int err; 163fa9e4066Sahrens 1643b2aab18SMatthew Ahrens ASSERT(dsl_pool_config_held(dp)); 165fa9e4066Sahrens 166ea8dc4b6Seschrock err = dmu_bonus_hold(dp->dp_meta_objset, ddobj, tag, &dbuf); 1673b2aab18SMatthew Ahrens if (err != 0) 168ea8dc4b6Seschrock return (err); 169fa9e4066Sahrens dd = dmu_buf_get_user(dbuf); 170fa9e4066Sahrens #ifdef ZFS_DEBUG 171fa9e4066Sahrens { 172fa9e4066Sahrens dmu_object_info_t doi; 173fa9e4066Sahrens dmu_object_info_from_db(dbuf, &doi); 1742acef22dSMatthew Ahrens ASSERT3U(doi.doi_bonus_type, ==, DMU_OT_DSL_DIR); 17574e7dc98SMatthew Ahrens ASSERT3U(doi.doi_bonus_size, >=, sizeof (dsl_dir_phys_t)); 176fa9e4066Sahrens } 177fa9e4066Sahrens #endif 178fa9e4066Sahrens if (dd == NULL) { 179fa9e4066Sahrens dsl_dir_t *winner; 180fa9e4066Sahrens 181fa9e4066Sahrens dd = kmem_zalloc(sizeof (dsl_dir_t), KM_SLEEP); 182fa9e4066Sahrens dd->dd_object = ddobj; 183fa9e4066Sahrens dd->dd_dbuf = dbuf; 184fa9e4066Sahrens dd->dd_pool = dp; 1855ad82045Snd mutex_init(&dd->dd_lock, NULL, MUTEX_DEFAULT, NULL); 186fa9e4066Sahrens 187fa9e4066Sahrens list_create(&dd->dd_prop_cbs, sizeof (dsl_prop_cb_record_t), 188fa9e4066Sahrens offsetof(dsl_prop_cb_record_t, cbr_node)); 189fa9e4066Sahrens 19071eb0538SChris Kirby dsl_dir_snap_cmtime_update(dd); 19171eb0538SChris Kirby 192*c1379625SJustin T. Gibbs if (dsl_dir_phys(dd)->dd_parent_obj) { 193*c1379625SJustin T. Gibbs err = dsl_dir_hold_obj(dp, 194*c1379625SJustin T. Gibbs dsl_dir_phys(dd)->dd_parent_obj, NULL, dd, 195*c1379625SJustin T. Gibbs &dd->dd_parent); 1963b2aab18SMatthew Ahrens if (err != 0) 19774e7dc98SMatthew Ahrens goto errout; 198fa9e4066Sahrens if (tail) { 199fa9e4066Sahrens #ifdef ZFS_DEBUG 200fa9e4066Sahrens uint64_t foundobj; 201fa9e4066Sahrens 202fa9e4066Sahrens err = zap_lookup(dp->dp_meta_objset, 203*c1379625SJustin T. Gibbs dsl_dir_phys(dd->dd_parent)-> 204*c1379625SJustin T. Gibbs dd_child_dir_zapobj, tail, 205*c1379625SJustin T. Gibbs sizeof (foundobj), 1, &foundobj); 206ea8dc4b6Seschrock ASSERT(err || foundobj == ddobj); 207fa9e4066Sahrens #endif 208fa9e4066Sahrens (void) strcpy(dd->dd_myname, tail); 209fa9e4066Sahrens } else { 210fa9e4066Sahrens err = zap_value_search(dp->dp_meta_objset, 211*c1379625SJustin T. Gibbs dsl_dir_phys(dd->dd_parent)-> 212*c1379625SJustin T. Gibbs dd_child_dir_zapobj, 213e7437265Sahrens ddobj, 0, dd->dd_myname); 214ea8dc4b6Seschrock } 2153b2aab18SMatthew Ahrens if (err != 0) 21674e7dc98SMatthew Ahrens goto errout; 217fa9e4066Sahrens } else { 218fa9e4066Sahrens (void) strcpy(dd->dd_myname, spa_name(dp->dp_spa)); 219fa9e4066Sahrens } 220fa9e4066Sahrens 2213f9d6ad7SLin Ling if (dsl_dir_is_clone(dd)) { 2223f9d6ad7SLin Ling dmu_buf_t *origin_bonus; 2233f9d6ad7SLin Ling dsl_dataset_phys_t *origin_phys; 2243f9d6ad7SLin Ling 2253f9d6ad7SLin Ling /* 2263f9d6ad7SLin Ling * We can't open the origin dataset, because 2273f9d6ad7SLin Ling * that would require opening this dsl_dir. 2283f9d6ad7SLin Ling * Just look at its phys directly instead. 2293f9d6ad7SLin Ling */ 2303f9d6ad7SLin Ling err = dmu_bonus_hold(dp->dp_meta_objset, 231*c1379625SJustin T. Gibbs dsl_dir_phys(dd)->dd_origin_obj, FTAG, 232*c1379625SJustin T. Gibbs &origin_bonus); 2333b2aab18SMatthew Ahrens if (err != 0) 2343f9d6ad7SLin Ling goto errout; 2353f9d6ad7SLin Ling origin_phys = origin_bonus->db_data; 2363f9d6ad7SLin Ling dd->dd_origin_txg = 2373f9d6ad7SLin Ling origin_phys->ds_creation_txg; 2383f9d6ad7SLin Ling dmu_buf_rele(origin_bonus, FTAG); 2393f9d6ad7SLin Ling } 2403f9d6ad7SLin Ling 241*c1379625SJustin T. Gibbs winner = dmu_buf_set_user_ie(dbuf, dd, dsl_dir_evict); 242fa9e4066Sahrens if (winner) { 243fa9e4066Sahrens if (dd->dd_parent) 2443b2aab18SMatthew Ahrens dsl_dir_rele(dd->dd_parent, dd); 2455ad82045Snd mutex_destroy(&dd->dd_lock); 246fa9e4066Sahrens kmem_free(dd, sizeof (dsl_dir_t)); 247fa9e4066Sahrens dd = winner; 248fa9e4066Sahrens } else { 249fa9e4066Sahrens spa_open_ref(dp->dp_spa, dd); 250fa9e4066Sahrens } 251fa9e4066Sahrens } 252fa9e4066Sahrens 253fa9e4066Sahrens /* 254fa9e4066Sahrens * The dsl_dir_t has both open-to-close and instantiate-to-evict 255fa9e4066Sahrens * holds on the spa. We need the open-to-close holds because 256fa9e4066Sahrens * otherwise the spa_refcnt wouldn't change when we open a 257fa9e4066Sahrens * dir which the spa also has open, so we could incorrectly 258fa9e4066Sahrens * think it was OK to unload/export/destroy the pool. We need 259fa9e4066Sahrens * the instantiate-to-evict hold because the dsl_dir_t has a 260fa9e4066Sahrens * pointer to the dd_pool, which has a pointer to the spa_t. 261fa9e4066Sahrens */ 262fa9e4066Sahrens spa_open_ref(dp->dp_spa, tag); 263fa9e4066Sahrens ASSERT3P(dd->dd_pool, ==, dp); 264fa9e4066Sahrens ASSERT3U(dd->dd_object, ==, ddobj); 265fa9e4066Sahrens ASSERT3P(dd->dd_dbuf, ==, dbuf); 266ea8dc4b6Seschrock *ddp = dd; 267ea8dc4b6Seschrock return (0); 26874e7dc98SMatthew Ahrens 26974e7dc98SMatthew Ahrens errout: 27074e7dc98SMatthew Ahrens if (dd->dd_parent) 2713b2aab18SMatthew Ahrens dsl_dir_rele(dd->dd_parent, dd); 27274e7dc98SMatthew Ahrens mutex_destroy(&dd->dd_lock); 27374e7dc98SMatthew Ahrens kmem_free(dd, sizeof (dsl_dir_t)); 27474e7dc98SMatthew Ahrens dmu_buf_rele(dbuf, tag); 27574e7dc98SMatthew Ahrens return (err); 276fa9e4066Sahrens } 277fa9e4066Sahrens 278fa9e4066Sahrens void 2793b2aab18SMatthew Ahrens dsl_dir_rele(dsl_dir_t *dd, void *tag) 280fa9e4066Sahrens { 281fa9e4066Sahrens dprintf_dd(dd, "%s\n", ""); 282fa9e4066Sahrens spa_close(dd->dd_pool->dp_spa, tag); 283ea8dc4b6Seschrock dmu_buf_rele(dd->dd_dbuf, tag); 284fa9e4066Sahrens } 285fa9e4066Sahrens 28615f66a7fSek /* buf must be long enough (MAXNAMELEN + strlen(MOS_DIR_NAME) + 1 should do) */ 287fa9e4066Sahrens void 288fa9e4066Sahrens dsl_dir_name(dsl_dir_t *dd, char *buf) 289fa9e4066Sahrens { 290fa9e4066Sahrens if (dd->dd_parent) { 291fa9e4066Sahrens dsl_dir_name(dd->dd_parent, buf); 292fa9e4066Sahrens (void) strcat(buf, "/"); 293fa9e4066Sahrens } else { 294fa9e4066Sahrens buf[0] = '\0'; 295fa9e4066Sahrens } 296fa9e4066Sahrens if (!MUTEX_HELD(&dd->dd_lock)) { 297fa9e4066Sahrens /* 298fa9e4066Sahrens * recursive mutex so that we can use 299fa9e4066Sahrens * dprintf_dd() with dd_lock held 300fa9e4066Sahrens */ 301fa9e4066Sahrens mutex_enter(&dd->dd_lock); 302fa9e4066Sahrens (void) strcat(buf, dd->dd_myname); 303fa9e4066Sahrens mutex_exit(&dd->dd_lock); 304fa9e4066Sahrens } else { 305fa9e4066Sahrens (void) strcat(buf, dd->dd_myname); 306fa9e4066Sahrens } 307fa9e4066Sahrens } 308fa9e4066Sahrens 309ce636f8bSMatthew Ahrens /* Calculate name length, avoiding all the strcat calls of dsl_dir_name */ 310b7661cccSmmusante int 311b7661cccSmmusante dsl_dir_namelen(dsl_dir_t *dd) 312b7661cccSmmusante { 313b7661cccSmmusante int result = 0; 314b7661cccSmmusante 315b7661cccSmmusante if (dd->dd_parent) { 316b7661cccSmmusante /* parent's name + 1 for the "/" */ 317b7661cccSmmusante result = dsl_dir_namelen(dd->dd_parent) + 1; 318b7661cccSmmusante } 319b7661cccSmmusante 320b7661cccSmmusante if (!MUTEX_HELD(&dd->dd_lock)) { 321b7661cccSmmusante /* see dsl_dir_name */ 322b7661cccSmmusante mutex_enter(&dd->dd_lock); 323b7661cccSmmusante result += strlen(dd->dd_myname); 324b7661cccSmmusante mutex_exit(&dd->dd_lock); 325b7661cccSmmusante } else { 326b7661cccSmmusante result += strlen(dd->dd_myname); 327b7661cccSmmusante } 328b7661cccSmmusante 329b7661cccSmmusante return (result); 330b7661cccSmmusante } 331b7661cccSmmusante 332fa9e4066Sahrens static int 333fa9e4066Sahrens getcomponent(const char *path, char *component, const char **nextp) 334fa9e4066Sahrens { 335fa9e4066Sahrens char *p; 3363b2aab18SMatthew Ahrens 337ccba0801SRich Morris if ((path == NULL) || (path[0] == '\0')) 338be6fd75aSMatthew Ahrens return (SET_ERROR(ENOENT)); 339fa9e4066Sahrens /* This would be a good place to reserve some namespace... */ 340fa9e4066Sahrens p = strpbrk(path, "/@"); 341fa9e4066Sahrens if (p && (p[1] == '/' || p[1] == '@')) { 342fa9e4066Sahrens /* two separators in a row */ 343be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 344fa9e4066Sahrens } 345fa9e4066Sahrens if (p == NULL || p == path) { 346fa9e4066Sahrens /* 347fa9e4066Sahrens * if the first thing is an @ or /, it had better be an 348fa9e4066Sahrens * @ and it had better not have any more ats or slashes, 349fa9e4066Sahrens * and it had better have something after the @. 350fa9e4066Sahrens */ 351fa9e4066Sahrens if (p != NULL && 352fa9e4066Sahrens (p[0] != '@' || strpbrk(path+1, "/@") || p[1] == '\0')) 353be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 354fa9e4066Sahrens if (strlen(path) >= MAXNAMELEN) 355be6fd75aSMatthew Ahrens return (SET_ERROR(ENAMETOOLONG)); 356fa9e4066Sahrens (void) strcpy(component, path); 357fa9e4066Sahrens p = NULL; 358fa9e4066Sahrens } else if (p[0] == '/') { 3593b2aab18SMatthew Ahrens if (p - path >= MAXNAMELEN) 360be6fd75aSMatthew Ahrens return (SET_ERROR(ENAMETOOLONG)); 361fa9e4066Sahrens (void) strncpy(component, path, p - path); 3623b2aab18SMatthew Ahrens component[p - path] = '\0'; 363fa9e4066Sahrens p++; 364fa9e4066Sahrens } else if (p[0] == '@') { 365fa9e4066Sahrens /* 366fa9e4066Sahrens * if the next separator is an @, there better not be 367fa9e4066Sahrens * any more slashes. 368fa9e4066Sahrens */ 369fa9e4066Sahrens if (strchr(path, '/')) 370be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 3713b2aab18SMatthew Ahrens if (p - path >= MAXNAMELEN) 372be6fd75aSMatthew Ahrens return (SET_ERROR(ENAMETOOLONG)); 373fa9e4066Sahrens (void) strncpy(component, path, p - path); 3743b2aab18SMatthew Ahrens component[p - path] = '\0'; 375fa9e4066Sahrens } else { 3763b2aab18SMatthew Ahrens panic("invalid p=%p", (void *)p); 377fa9e4066Sahrens } 378fa9e4066Sahrens *nextp = p; 379fa9e4066Sahrens return (0); 380fa9e4066Sahrens } 381fa9e4066Sahrens 382fa9e4066Sahrens /* 3833b2aab18SMatthew Ahrens * Return the dsl_dir_t, and possibly the last component which couldn't 3843b2aab18SMatthew Ahrens * be found in *tail. The name must be in the specified dsl_pool_t. This 3853b2aab18SMatthew Ahrens * thread must hold the dp_config_rwlock for the pool. Returns NULL if the 3863b2aab18SMatthew Ahrens * path is bogus, or if tail==NULL and we couldn't parse the whole name. 3873b2aab18SMatthew Ahrens * (*tail)[0] == '@' means that the last component is a snapshot. 388fa9e4066Sahrens */ 389ea8dc4b6Seschrock int 3903b2aab18SMatthew Ahrens dsl_dir_hold(dsl_pool_t *dp, const char *name, void *tag, 391ea8dc4b6Seschrock dsl_dir_t **ddp, const char **tailp) 392fa9e4066Sahrens { 393fa9e4066Sahrens char buf[MAXNAMELEN]; 3943b2aab18SMatthew Ahrens const char *spaname, *next, *nextnext = NULL; 395fa9e4066Sahrens int err; 396fa9e4066Sahrens dsl_dir_t *dd; 397fa9e4066Sahrens uint64_t ddobj; 398fa9e4066Sahrens 399fa9e4066Sahrens err = getcomponent(name, buf, &next); 4003b2aab18SMatthew Ahrens if (err != 0) 401ea8dc4b6Seschrock return (err); 402fa9e4066Sahrens 4033b2aab18SMatthew Ahrens /* Make sure the name is in the specified pool. */ 4043b2aab18SMatthew Ahrens spaname = spa_name(dp->dp_spa); 4053b2aab18SMatthew Ahrens if (strcmp(buf, spaname) != 0) 406be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 407fa9e4066Sahrens 4083b2aab18SMatthew Ahrens ASSERT(dsl_pool_config_held(dp)); 409fa9e4066Sahrens 4103b2aab18SMatthew Ahrens err = dsl_dir_hold_obj(dp, dp->dp_root_dir_obj, NULL, tag, &dd); 4113b2aab18SMatthew Ahrens if (err != 0) { 412ea8dc4b6Seschrock return (err); 413ea8dc4b6Seschrock } 414ea8dc4b6Seschrock 415fa9e4066Sahrens while (next != NULL) { 416fa9e4066Sahrens dsl_dir_t *child_ds; 417fa9e4066Sahrens err = getcomponent(next, buf, &nextnext); 4183b2aab18SMatthew Ahrens if (err != 0) 419ea8dc4b6Seschrock break; 420fa9e4066Sahrens ASSERT(next[0] != '\0'); 421fa9e4066Sahrens if (next[0] == '@') 422fa9e4066Sahrens break; 423fa9e4066Sahrens dprintf("looking up %s in obj%lld\n", 424*c1379625SJustin T. Gibbs buf, dsl_dir_phys(dd)->dd_child_dir_zapobj); 425fa9e4066Sahrens 426fa9e4066Sahrens err = zap_lookup(dp->dp_meta_objset, 427*c1379625SJustin T. Gibbs dsl_dir_phys(dd)->dd_child_dir_zapobj, 428fa9e4066Sahrens buf, sizeof (ddobj), 1, &ddobj); 4293b2aab18SMatthew Ahrens if (err != 0) { 430ea8dc4b6Seschrock if (err == ENOENT) 431ea8dc4b6Seschrock err = 0; 432fa9e4066Sahrens break; 433fa9e4066Sahrens } 434fa9e4066Sahrens 4353b2aab18SMatthew Ahrens err = dsl_dir_hold_obj(dp, ddobj, buf, tag, &child_ds); 4363b2aab18SMatthew Ahrens if (err != 0) 437ea8dc4b6Seschrock break; 4383b2aab18SMatthew Ahrens dsl_dir_rele(dd, tag); 439fa9e4066Sahrens dd = child_ds; 440fa9e4066Sahrens next = nextnext; 441fa9e4066Sahrens } 442fa9e4066Sahrens 4433b2aab18SMatthew Ahrens if (err != 0) { 4443b2aab18SMatthew Ahrens dsl_dir_rele(dd, tag); 445ea8dc4b6Seschrock return (err); 446ea8dc4b6Seschrock } 447ea8dc4b6Seschrock 448fa9e4066Sahrens /* 449fa9e4066Sahrens * It's an error if there's more than one component left, or 450fa9e4066Sahrens * tailp==NULL and there's any component left. 451fa9e4066Sahrens */ 452fa9e4066Sahrens if (next != NULL && 453fa9e4066Sahrens (tailp == NULL || (nextnext && nextnext[0] != '\0'))) { 454fa9e4066Sahrens /* bad path name */ 4553b2aab18SMatthew Ahrens dsl_dir_rele(dd, tag); 456fa9e4066Sahrens dprintf("next=%p (%s) tail=%p\n", next, next?next:"", tailp); 457be6fd75aSMatthew Ahrens err = SET_ERROR(ENOENT); 458fa9e4066Sahrens } 4593b2aab18SMatthew Ahrens if (tailp != NULL) 460fa9e4066Sahrens *tailp = next; 461ea8dc4b6Seschrock *ddp = dd; 462ea8dc4b6Seschrock return (err); 463fa9e4066Sahrens } 464fa9e4066Sahrens 465a2afb611SJerry Jelinek /* 466a2afb611SJerry Jelinek * If the counts are already initialized for this filesystem and its 467a2afb611SJerry Jelinek * descendants then do nothing, otherwise initialize the counts. 468a2afb611SJerry Jelinek * 469a2afb611SJerry Jelinek * The counts on this filesystem, and those below, may be uninitialized due to 470a2afb611SJerry Jelinek * either the use of a pre-existing pool which did not support the 471a2afb611SJerry Jelinek * filesystem/snapshot limit feature, or one in which the feature had not yet 472a2afb611SJerry Jelinek * been enabled. 473a2afb611SJerry Jelinek * 474a2afb611SJerry Jelinek * Recursively descend the filesystem tree and update the filesystem/snapshot 475a2afb611SJerry Jelinek * counts on each filesystem below, then update the cumulative count on the 476a2afb611SJerry Jelinek * current filesystem. If the filesystem already has a count set on it, 477a2afb611SJerry Jelinek * then we know that its counts, and the counts on the filesystems below it, 478a2afb611SJerry Jelinek * are already correct, so we don't have to update this filesystem. 479a2afb611SJerry Jelinek */ 480a2afb611SJerry Jelinek static void 481a2afb611SJerry Jelinek dsl_dir_init_fs_ss_count(dsl_dir_t *dd, dmu_tx_t *tx) 482a2afb611SJerry Jelinek { 483a2afb611SJerry Jelinek uint64_t my_fs_cnt = 0; 484a2afb611SJerry Jelinek uint64_t my_ss_cnt = 0; 485a2afb611SJerry Jelinek dsl_pool_t *dp = dd->dd_pool; 486a2afb611SJerry Jelinek objset_t *os = dp->dp_meta_objset; 487a2afb611SJerry Jelinek zap_cursor_t *zc; 488a2afb611SJerry Jelinek zap_attribute_t *za; 489a2afb611SJerry Jelinek dsl_dataset_t *ds; 490a2afb611SJerry Jelinek 491adf34077SJerry Jelinek ASSERT(spa_feature_is_active(dp->dp_spa, SPA_FEATURE_FS_SS_LIMIT)); 492a2afb611SJerry Jelinek ASSERT(dsl_pool_config_held(dp)); 493a2afb611SJerry Jelinek ASSERT(dmu_tx_is_syncing(tx)); 494a2afb611SJerry Jelinek 495a2afb611SJerry Jelinek dsl_dir_zapify(dd, tx); 496a2afb611SJerry Jelinek 497a2afb611SJerry Jelinek /* 498a2afb611SJerry Jelinek * If the filesystem count has already been initialized then we 499a2afb611SJerry Jelinek * don't need to recurse down any further. 500a2afb611SJerry Jelinek */ 501a2afb611SJerry Jelinek if (zap_contains(os, dd->dd_object, DD_FIELD_FILESYSTEM_COUNT) == 0) 502a2afb611SJerry Jelinek return; 503a2afb611SJerry Jelinek 504a2afb611SJerry Jelinek zc = kmem_alloc(sizeof (zap_cursor_t), KM_SLEEP); 505a2afb611SJerry Jelinek za = kmem_alloc(sizeof (zap_attribute_t), KM_SLEEP); 506a2afb611SJerry Jelinek 507a2afb611SJerry Jelinek /* Iterate my child dirs */ 508*c1379625SJustin T. Gibbs for (zap_cursor_init(zc, os, dsl_dir_phys(dd)->dd_child_dir_zapobj); 509a2afb611SJerry Jelinek zap_cursor_retrieve(zc, za) == 0; zap_cursor_advance(zc)) { 510a2afb611SJerry Jelinek dsl_dir_t *chld_dd; 511a2afb611SJerry Jelinek uint64_t count; 512a2afb611SJerry Jelinek 513a2afb611SJerry Jelinek VERIFY0(dsl_dir_hold_obj(dp, za->za_first_integer, NULL, FTAG, 514a2afb611SJerry Jelinek &chld_dd)); 515a2afb611SJerry Jelinek 516a2afb611SJerry Jelinek /* 517a2afb611SJerry Jelinek * Ignore hidden ($FREE, $MOS & $ORIGIN) objsets and 518a2afb611SJerry Jelinek * temporary datasets. 519a2afb611SJerry Jelinek */ 520a2afb611SJerry Jelinek if (chld_dd->dd_myname[0] == '$' || 521a2afb611SJerry Jelinek chld_dd->dd_myname[0] == '%') { 522a2afb611SJerry Jelinek dsl_dir_rele(chld_dd, FTAG); 523a2afb611SJerry Jelinek continue; 524a2afb611SJerry Jelinek } 525a2afb611SJerry Jelinek 526a2afb611SJerry Jelinek my_fs_cnt++; /* count this child */ 527a2afb611SJerry Jelinek 528a2afb611SJerry Jelinek dsl_dir_init_fs_ss_count(chld_dd, tx); 529a2afb611SJerry Jelinek 530a2afb611SJerry Jelinek VERIFY0(zap_lookup(os, chld_dd->dd_object, 531a2afb611SJerry Jelinek DD_FIELD_FILESYSTEM_COUNT, sizeof (count), 1, &count)); 532a2afb611SJerry Jelinek my_fs_cnt += count; 533a2afb611SJerry Jelinek VERIFY0(zap_lookup(os, chld_dd->dd_object, 534a2afb611SJerry Jelinek DD_FIELD_SNAPSHOT_COUNT, sizeof (count), 1, &count)); 535a2afb611SJerry Jelinek my_ss_cnt += count; 536a2afb611SJerry Jelinek 537a2afb611SJerry Jelinek dsl_dir_rele(chld_dd, FTAG); 538a2afb611SJerry Jelinek } 539a2afb611SJerry Jelinek zap_cursor_fini(zc); 540a2afb611SJerry Jelinek /* Count my snapshots (we counted children's snapshots above) */ 541a2afb611SJerry Jelinek VERIFY0(dsl_dataset_hold_obj(dd->dd_pool, 542*c1379625SJustin T. Gibbs dsl_dir_phys(dd)->dd_head_dataset_obj, FTAG, &ds)); 543a2afb611SJerry Jelinek 544*c1379625SJustin T. Gibbs for (zap_cursor_init(zc, os, dsl_dataset_phys(ds)->ds_snapnames_zapobj); 545a2afb611SJerry Jelinek zap_cursor_retrieve(zc, za) == 0; 546a2afb611SJerry Jelinek zap_cursor_advance(zc)) { 547a2afb611SJerry Jelinek /* Don't count temporary snapshots */ 548a2afb611SJerry Jelinek if (za->za_name[0] != '%') 549a2afb611SJerry Jelinek my_ss_cnt++; 550a2afb611SJerry Jelinek } 551fb7001f1SAlex Reece zap_cursor_fini(zc); 552a2afb611SJerry Jelinek 553a2afb611SJerry Jelinek dsl_dataset_rele(ds, FTAG); 554a2afb611SJerry Jelinek 555a2afb611SJerry Jelinek kmem_free(zc, sizeof (zap_cursor_t)); 556a2afb611SJerry Jelinek kmem_free(za, sizeof (zap_attribute_t)); 557a2afb611SJerry Jelinek 558a2afb611SJerry Jelinek /* we're in a sync task, update counts */ 559a2afb611SJerry Jelinek dmu_buf_will_dirty(dd->dd_dbuf, tx); 560a2afb611SJerry Jelinek VERIFY0(zap_add(os, dd->dd_object, DD_FIELD_FILESYSTEM_COUNT, 561a2afb611SJerry Jelinek sizeof (my_fs_cnt), 1, &my_fs_cnt, tx)); 562a2afb611SJerry Jelinek VERIFY0(zap_add(os, dd->dd_object, DD_FIELD_SNAPSHOT_COUNT, 563a2afb611SJerry Jelinek sizeof (my_ss_cnt), 1, &my_ss_cnt, tx)); 564a2afb611SJerry Jelinek } 565a2afb611SJerry Jelinek 566a2afb611SJerry Jelinek static int 567a2afb611SJerry Jelinek dsl_dir_actv_fs_ss_limit_check(void *arg, dmu_tx_t *tx) 568a2afb611SJerry Jelinek { 569a2afb611SJerry Jelinek char *ddname = (char *)arg; 570a2afb611SJerry Jelinek dsl_pool_t *dp = dmu_tx_pool(tx); 571a2afb611SJerry Jelinek dsl_dataset_t *ds; 572a2afb611SJerry Jelinek dsl_dir_t *dd; 573a2afb611SJerry Jelinek int error; 574a2afb611SJerry Jelinek 575a2afb611SJerry Jelinek error = dsl_dataset_hold(dp, ddname, FTAG, &ds); 576a2afb611SJerry Jelinek if (error != 0) 577a2afb611SJerry Jelinek return (error); 578a2afb611SJerry Jelinek 579a2afb611SJerry Jelinek if (!spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_FS_SS_LIMIT)) { 580a2afb611SJerry Jelinek dsl_dataset_rele(ds, FTAG); 581a2afb611SJerry Jelinek return (SET_ERROR(ENOTSUP)); 582a2afb611SJerry Jelinek } 583a2afb611SJerry Jelinek 584a2afb611SJerry Jelinek dd = ds->ds_dir; 585a2afb611SJerry Jelinek if (spa_feature_is_active(dp->dp_spa, SPA_FEATURE_FS_SS_LIMIT) && 586a2afb611SJerry Jelinek dsl_dir_is_zapified(dd) && 587a2afb611SJerry Jelinek zap_contains(dp->dp_meta_objset, dd->dd_object, 588a2afb611SJerry Jelinek DD_FIELD_FILESYSTEM_COUNT) == 0) { 589a2afb611SJerry Jelinek dsl_dataset_rele(ds, FTAG); 590a2afb611SJerry Jelinek return (SET_ERROR(EALREADY)); 591a2afb611SJerry Jelinek } 592a2afb611SJerry Jelinek 593a2afb611SJerry Jelinek dsl_dataset_rele(ds, FTAG); 594a2afb611SJerry Jelinek return (0); 595a2afb611SJerry Jelinek } 596a2afb611SJerry Jelinek 597a2afb611SJerry Jelinek static void 598a2afb611SJerry Jelinek dsl_dir_actv_fs_ss_limit_sync(void *arg, dmu_tx_t *tx) 599a2afb611SJerry Jelinek { 600a2afb611SJerry Jelinek char *ddname = (char *)arg; 601a2afb611SJerry Jelinek dsl_pool_t *dp = dmu_tx_pool(tx); 602a2afb611SJerry Jelinek dsl_dataset_t *ds; 603a2afb611SJerry Jelinek spa_t *spa; 604a2afb611SJerry Jelinek 605a2afb611SJerry Jelinek VERIFY0(dsl_dataset_hold(dp, ddname, FTAG, &ds)); 606a2afb611SJerry Jelinek 607a2afb611SJerry Jelinek spa = dsl_dataset_get_spa(ds); 608a2afb611SJerry Jelinek 609a2afb611SJerry Jelinek if (!spa_feature_is_active(spa, SPA_FEATURE_FS_SS_LIMIT)) { 610a2afb611SJerry Jelinek /* 611a2afb611SJerry Jelinek * Since the feature was not active and we're now setting a 612a2afb611SJerry Jelinek * limit, increment the feature-active counter so that the 613a2afb611SJerry Jelinek * feature becomes active for the first time. 614a2afb611SJerry Jelinek * 615a2afb611SJerry Jelinek * We are already in a sync task so we can update the MOS. 616a2afb611SJerry Jelinek */ 617a2afb611SJerry Jelinek spa_feature_incr(spa, SPA_FEATURE_FS_SS_LIMIT, tx); 618a2afb611SJerry Jelinek } 619a2afb611SJerry Jelinek 620a2afb611SJerry Jelinek /* 621a2afb611SJerry Jelinek * Since we are now setting a non-UINT64_MAX limit on the filesystem, 622a2afb611SJerry Jelinek * we need to ensure the counts are correct. Descend down the tree from 623a2afb611SJerry Jelinek * this point and update all of the counts to be accurate. 624a2afb611SJerry Jelinek */ 625a2afb611SJerry Jelinek dsl_dir_init_fs_ss_count(ds->ds_dir, tx); 626a2afb611SJerry Jelinek 627a2afb611SJerry Jelinek dsl_dataset_rele(ds, FTAG); 628a2afb611SJerry Jelinek } 629a2afb611SJerry Jelinek 630a2afb611SJerry Jelinek /* 631a2afb611SJerry Jelinek * Make sure the feature is enabled and activate it if necessary. 632a2afb611SJerry Jelinek * Since we're setting a limit, ensure the on-disk counts are valid. 633a2afb611SJerry Jelinek * This is only called by the ioctl path when setting a limit value. 634a2afb611SJerry Jelinek * 635a2afb611SJerry Jelinek * We do not need to validate the new limit, since users who can change the 636a2afb611SJerry Jelinek * limit are also allowed to exceed the limit. 637a2afb611SJerry Jelinek */ 638a2afb611SJerry Jelinek int 639a2afb611SJerry Jelinek dsl_dir_activate_fs_ss_limit(const char *ddname) 640a2afb611SJerry Jelinek { 641a2afb611SJerry Jelinek int error; 642a2afb611SJerry Jelinek 643a2afb611SJerry Jelinek error = dsl_sync_task(ddname, dsl_dir_actv_fs_ss_limit_check, 6447d46dc6cSMatthew Ahrens dsl_dir_actv_fs_ss_limit_sync, (void *)ddname, 0, 6457d46dc6cSMatthew Ahrens ZFS_SPACE_CHECK_RESERVED); 646a2afb611SJerry Jelinek 647a2afb611SJerry Jelinek if (error == EALREADY) 648a2afb611SJerry Jelinek error = 0; 649a2afb611SJerry Jelinek 650a2afb611SJerry Jelinek return (error); 651a2afb611SJerry Jelinek } 652a2afb611SJerry Jelinek 653a2afb611SJerry Jelinek /* 654a2afb611SJerry Jelinek * Used to determine if the filesystem_limit or snapshot_limit should be 655a2afb611SJerry Jelinek * enforced. We allow the limit to be exceeded if the user has permission to 656a2afb611SJerry Jelinek * write the property value. We pass in the creds that we got in the open 657a2afb611SJerry Jelinek * context since we will always be the GZ root in syncing context. We also have 658a2afb611SJerry Jelinek * to handle the case where we are allowed to change the limit on the current 659a2afb611SJerry Jelinek * dataset, but there may be another limit in the tree above. 660a2afb611SJerry Jelinek * 661a2afb611SJerry Jelinek * We can never modify these two properties within a non-global zone. In 662a2afb611SJerry Jelinek * addition, the other checks are modeled on zfs_secpolicy_write_perms. We 663a2afb611SJerry Jelinek * can't use that function since we are already holding the dp_config_rwlock. 664a2afb611SJerry Jelinek * In addition, we already have the dd and dealing with snapshots is simplified 665a2afb611SJerry Jelinek * in this code. 666a2afb611SJerry Jelinek */ 667a2afb611SJerry Jelinek 668a2afb611SJerry Jelinek typedef enum { 669a2afb611SJerry Jelinek ENFORCE_ALWAYS, 670a2afb611SJerry Jelinek ENFORCE_NEVER, 671a2afb611SJerry Jelinek ENFORCE_ABOVE 672a2afb611SJerry Jelinek } enforce_res_t; 673a2afb611SJerry Jelinek 674a2afb611SJerry Jelinek static enforce_res_t 675a2afb611SJerry Jelinek dsl_enforce_ds_ss_limits(dsl_dir_t *dd, zfs_prop_t prop, cred_t *cr) 676a2afb611SJerry Jelinek { 677a2afb611SJerry Jelinek enforce_res_t enforce = ENFORCE_ALWAYS; 678a2afb611SJerry Jelinek uint64_t obj; 679a2afb611SJerry Jelinek dsl_dataset_t *ds; 680a2afb611SJerry Jelinek uint64_t zoned; 681a2afb611SJerry Jelinek 682a2afb611SJerry Jelinek ASSERT(prop == ZFS_PROP_FILESYSTEM_LIMIT || 683a2afb611SJerry Jelinek prop == ZFS_PROP_SNAPSHOT_LIMIT); 684a2afb611SJerry Jelinek 685a2afb611SJerry Jelinek #ifdef _KERNEL 686a2afb611SJerry Jelinek if (crgetzoneid(cr) != GLOBAL_ZONEID) 687a2afb611SJerry Jelinek return (ENFORCE_ALWAYS); 688a2afb611SJerry Jelinek 689a2afb611SJerry Jelinek if (secpolicy_zfs(cr) == 0) 690a2afb611SJerry Jelinek return (ENFORCE_NEVER); 691a2afb611SJerry Jelinek #endif 692a2afb611SJerry Jelinek 693*c1379625SJustin T. Gibbs if ((obj = dsl_dir_phys(dd)->dd_head_dataset_obj) == 0) 694a2afb611SJerry Jelinek return (ENFORCE_ALWAYS); 695a2afb611SJerry Jelinek 696a2afb611SJerry Jelinek ASSERT(dsl_pool_config_held(dd->dd_pool)); 697a2afb611SJerry Jelinek 698a2afb611SJerry Jelinek if (dsl_dataset_hold_obj(dd->dd_pool, obj, FTAG, &ds) != 0) 699a2afb611SJerry Jelinek return (ENFORCE_ALWAYS); 700a2afb611SJerry Jelinek 701a2afb611SJerry Jelinek if (dsl_prop_get_ds(ds, "zoned", 8, 1, &zoned, NULL) || zoned) { 702a2afb611SJerry Jelinek /* Only root can access zoned fs's from the GZ */ 703a2afb611SJerry Jelinek enforce = ENFORCE_ALWAYS; 704a2afb611SJerry Jelinek } else { 705a2afb611SJerry Jelinek if (dsl_deleg_access_impl(ds, zfs_prop_to_name(prop), cr) == 0) 706a2afb611SJerry Jelinek enforce = ENFORCE_ABOVE; 707a2afb611SJerry Jelinek } 708a2afb611SJerry Jelinek 709a2afb611SJerry Jelinek dsl_dataset_rele(ds, FTAG); 710a2afb611SJerry Jelinek return (enforce); 711a2afb611SJerry Jelinek } 712a2afb611SJerry Jelinek 713a2afb611SJerry Jelinek /* 714a2afb611SJerry Jelinek * Check if adding additional child filesystem(s) would exceed any filesystem 715a2afb611SJerry Jelinek * limits or adding additional snapshot(s) would exceed any snapshot limits. 716a2afb611SJerry Jelinek * The prop argument indicates which limit to check. 717a2afb611SJerry Jelinek * 718a2afb611SJerry Jelinek * Note that all filesystem limits up to the root (or the highest 719a2afb611SJerry Jelinek * initialized) filesystem or the given ancestor must be satisfied. 720a2afb611SJerry Jelinek */ 721a2afb611SJerry Jelinek int 722a2afb611SJerry Jelinek dsl_fs_ss_limit_check(dsl_dir_t *dd, uint64_t delta, zfs_prop_t prop, 723a2afb611SJerry Jelinek dsl_dir_t *ancestor, cred_t *cr) 724a2afb611SJerry Jelinek { 725a2afb611SJerry Jelinek objset_t *os = dd->dd_pool->dp_meta_objset; 726a2afb611SJerry Jelinek uint64_t limit, count; 727a2afb611SJerry Jelinek char *count_prop; 728a2afb611SJerry Jelinek enforce_res_t enforce; 729a2afb611SJerry Jelinek int err = 0; 730a2afb611SJerry Jelinek 731a2afb611SJerry Jelinek ASSERT(dsl_pool_config_held(dd->dd_pool)); 732a2afb611SJerry Jelinek ASSERT(prop == ZFS_PROP_FILESYSTEM_LIMIT || 733a2afb611SJerry Jelinek prop == ZFS_PROP_SNAPSHOT_LIMIT); 734a2afb611SJerry Jelinek 735a2afb611SJerry Jelinek /* 736a2afb611SJerry Jelinek * If we're allowed to change the limit, don't enforce the limit 737a2afb611SJerry Jelinek * e.g. this can happen if a snapshot is taken by an administrative 738a2afb611SJerry Jelinek * user in the global zone (i.e. a recursive snapshot by root). 739a2afb611SJerry Jelinek * However, we must handle the case of delegated permissions where we 740a2afb611SJerry Jelinek * are allowed to change the limit on the current dataset, but there 741a2afb611SJerry Jelinek * is another limit in the tree above. 742a2afb611SJerry Jelinek */ 743a2afb611SJerry Jelinek enforce = dsl_enforce_ds_ss_limits(dd, prop, cr); 744a2afb611SJerry Jelinek if (enforce == ENFORCE_NEVER) 745a2afb611SJerry Jelinek return (0); 746a2afb611SJerry Jelinek 747a2afb611SJerry Jelinek /* 748a2afb611SJerry Jelinek * e.g. if renaming a dataset with no snapshots, count adjustment 749a2afb611SJerry Jelinek * is 0. 750a2afb611SJerry Jelinek */ 751a2afb611SJerry Jelinek if (delta == 0) 752a2afb611SJerry Jelinek return (0); 753a2afb611SJerry Jelinek 754a2afb611SJerry Jelinek if (prop == ZFS_PROP_SNAPSHOT_LIMIT) { 755a2afb611SJerry Jelinek /* 756a2afb611SJerry Jelinek * We don't enforce the limit for temporary snapshots. This is 757a2afb611SJerry Jelinek * indicated by a NULL cred_t argument. 758a2afb611SJerry Jelinek */ 759a2afb611SJerry Jelinek if (cr == NULL) 760a2afb611SJerry Jelinek return (0); 761a2afb611SJerry Jelinek 762a2afb611SJerry Jelinek count_prop = DD_FIELD_SNAPSHOT_COUNT; 763a2afb611SJerry Jelinek } else { 764a2afb611SJerry Jelinek count_prop = DD_FIELD_FILESYSTEM_COUNT; 765a2afb611SJerry Jelinek } 766a2afb611SJerry Jelinek 767a2afb611SJerry Jelinek /* 768a2afb611SJerry Jelinek * If an ancestor has been provided, stop checking the limit once we 769a2afb611SJerry Jelinek * hit that dir. We need this during rename so that we don't overcount 770a2afb611SJerry Jelinek * the check once we recurse up to the common ancestor. 771a2afb611SJerry Jelinek */ 772a2afb611SJerry Jelinek if (ancestor == dd) 773a2afb611SJerry Jelinek return (0); 774a2afb611SJerry Jelinek 775a2afb611SJerry Jelinek /* 776a2afb611SJerry Jelinek * If we hit an uninitialized node while recursing up the tree, we can 777a2afb611SJerry Jelinek * stop since we know there is no limit here (or above). The counts are 778a2afb611SJerry Jelinek * not valid on this node and we know we won't touch this node's counts. 779a2afb611SJerry Jelinek */ 780a2afb611SJerry Jelinek if (!dsl_dir_is_zapified(dd) || zap_lookup(os, dd->dd_object, 781a2afb611SJerry Jelinek count_prop, sizeof (count), 1, &count) == ENOENT) 782a2afb611SJerry Jelinek return (0); 783a2afb611SJerry Jelinek 784a2afb611SJerry Jelinek err = dsl_prop_get_dd(dd, zfs_prop_to_name(prop), 8, 1, &limit, NULL, 785a2afb611SJerry Jelinek B_FALSE); 786a2afb611SJerry Jelinek if (err != 0) 787a2afb611SJerry Jelinek return (err); 788a2afb611SJerry Jelinek 789a2afb611SJerry Jelinek /* Is there a limit which we've hit? */ 790a2afb611SJerry Jelinek if (enforce == ENFORCE_ALWAYS && (count + delta) > limit) 791a2afb611SJerry Jelinek return (SET_ERROR(EDQUOT)); 792a2afb611SJerry Jelinek 793a2afb611SJerry Jelinek if (dd->dd_parent != NULL) 794a2afb611SJerry Jelinek err = dsl_fs_ss_limit_check(dd->dd_parent, delta, prop, 795a2afb611SJerry Jelinek ancestor, cr); 796a2afb611SJerry Jelinek 797a2afb611SJerry Jelinek return (err); 798a2afb611SJerry Jelinek } 799a2afb611SJerry Jelinek 800a2afb611SJerry Jelinek /* 801a2afb611SJerry Jelinek * Adjust the filesystem or snapshot count for the specified dsl_dir_t and all 802a2afb611SJerry Jelinek * parents. When a new filesystem/snapshot is created, increment the count on 803a2afb611SJerry Jelinek * all parents, and when a filesystem/snapshot is destroyed, decrement the 804a2afb611SJerry Jelinek * count. 805a2afb611SJerry Jelinek */ 806a2afb611SJerry Jelinek void 807a2afb611SJerry Jelinek dsl_fs_ss_count_adjust(dsl_dir_t *dd, int64_t delta, const char *prop, 808a2afb611SJerry Jelinek dmu_tx_t *tx) 809a2afb611SJerry Jelinek { 810a2afb611SJerry Jelinek int err; 811a2afb611SJerry Jelinek objset_t *os = dd->dd_pool->dp_meta_objset; 812a2afb611SJerry Jelinek uint64_t count; 813a2afb611SJerry Jelinek 814a2afb611SJerry Jelinek ASSERT(dsl_pool_config_held(dd->dd_pool)); 815a2afb611SJerry Jelinek ASSERT(dmu_tx_is_syncing(tx)); 816a2afb611SJerry Jelinek ASSERT(strcmp(prop, DD_FIELD_FILESYSTEM_COUNT) == 0 || 817a2afb611SJerry Jelinek strcmp(prop, DD_FIELD_SNAPSHOT_COUNT) == 0); 818a2afb611SJerry Jelinek 819a2afb611SJerry Jelinek /* 820a2afb611SJerry Jelinek * When we receive an incremental stream into a filesystem that already 821a2afb611SJerry Jelinek * exists, a temporary clone is created. We don't count this temporary 822a2afb611SJerry Jelinek * clone, whose name begins with a '%'. We also ignore hidden ($FREE, 823a2afb611SJerry Jelinek * $MOS & $ORIGIN) objsets. 824a2afb611SJerry Jelinek */ 825a2afb611SJerry Jelinek if ((dd->dd_myname[0] == '%' || dd->dd_myname[0] == '$') && 826a2afb611SJerry Jelinek strcmp(prop, DD_FIELD_FILESYSTEM_COUNT) == 0) 827a2afb611SJerry Jelinek return; 828a2afb611SJerry Jelinek 829a2afb611SJerry Jelinek /* 830a2afb611SJerry Jelinek * e.g. if renaming a dataset with no snapshots, count adjustment is 0 831a2afb611SJerry Jelinek */ 832a2afb611SJerry Jelinek if (delta == 0) 833a2afb611SJerry Jelinek return; 834a2afb611SJerry Jelinek 835a2afb611SJerry Jelinek /* 836a2afb611SJerry Jelinek * If we hit an uninitialized node while recursing up the tree, we can 837a2afb611SJerry Jelinek * stop since we know the counts are not valid on this node and we 838a2afb611SJerry Jelinek * know we shouldn't touch this node's counts. An uninitialized count 839a2afb611SJerry Jelinek * on the node indicates that either the feature has not yet been 840a2afb611SJerry Jelinek * activated or there are no limits on this part of the tree. 841a2afb611SJerry Jelinek */ 842a2afb611SJerry Jelinek if (!dsl_dir_is_zapified(dd) || (err = zap_lookup(os, dd->dd_object, 843a2afb611SJerry Jelinek prop, sizeof (count), 1, &count)) == ENOENT) 844a2afb611SJerry Jelinek return; 845a2afb611SJerry Jelinek VERIFY0(err); 846a2afb611SJerry Jelinek 847a2afb611SJerry Jelinek count += delta; 848a2afb611SJerry Jelinek /* Use a signed verify to make sure we're not neg. */ 849a2afb611SJerry Jelinek VERIFY3S(count, >=, 0); 850a2afb611SJerry Jelinek 851a2afb611SJerry Jelinek VERIFY0(zap_update(os, dd->dd_object, prop, sizeof (count), 1, &count, 852a2afb611SJerry Jelinek tx)); 853a2afb611SJerry Jelinek 854a2afb611SJerry Jelinek /* Roll up this additional count into our ancestors */ 855a2afb611SJerry Jelinek if (dd->dd_parent != NULL) 856a2afb611SJerry Jelinek dsl_fs_ss_count_adjust(dd->dd_parent, delta, prop, tx); 857a2afb611SJerry Jelinek } 858a2afb611SJerry Jelinek 8591d452cf5Sahrens uint64_t 860088f3894Sahrens dsl_dir_create_sync(dsl_pool_t *dp, dsl_dir_t *pds, const char *name, 861088f3894Sahrens dmu_tx_t *tx) 862fa9e4066Sahrens { 863088f3894Sahrens objset_t *mos = dp->dp_meta_objset; 864fa9e4066Sahrens uint64_t ddobj; 865cde58dbcSMatthew Ahrens dsl_dir_phys_t *ddphys; 866fa9e4066Sahrens dmu_buf_t *dbuf; 867fa9e4066Sahrens 8681649cd4bStabriz ddobj = dmu_object_alloc(mos, DMU_OT_DSL_DIR, 0, 8691649cd4bStabriz DMU_OT_DSL_DIR, sizeof (dsl_dir_phys_t), tx); 870088f3894Sahrens if (pds) { 871*c1379625SJustin T. Gibbs VERIFY(0 == zap_add(mos, dsl_dir_phys(pds)->dd_child_dir_zapobj, 872088f3894Sahrens name, sizeof (uint64_t), 1, &ddobj, tx)); 873088f3894Sahrens } else { 874088f3894Sahrens /* it's the root dir */ 875088f3894Sahrens VERIFY(0 == zap_add(mos, DMU_POOL_DIRECTORY_OBJECT, 876088f3894Sahrens DMU_POOL_ROOT_DATASET, sizeof (uint64_t), 1, &ddobj, tx)); 877088f3894Sahrens } 878ea8dc4b6Seschrock VERIFY(0 == dmu_bonus_hold(mos, ddobj, FTAG, &dbuf)); 879fa9e4066Sahrens dmu_buf_will_dirty(dbuf, tx); 880cde58dbcSMatthew Ahrens ddphys = dbuf->db_data; 881fa9e4066Sahrens 882cde58dbcSMatthew Ahrens ddphys->dd_creation_time = gethrestime_sec(); 883a2afb611SJerry Jelinek if (pds) { 884cde58dbcSMatthew Ahrens ddphys->dd_parent_obj = pds->dd_object; 885a2afb611SJerry Jelinek 886a2afb611SJerry Jelinek /* update the filesystem counts */ 887a2afb611SJerry Jelinek dsl_fs_ss_count_adjust(pds, 1, DD_FIELD_FILESYSTEM_COUNT, tx); 888a2afb611SJerry Jelinek } 889cde58dbcSMatthew Ahrens ddphys->dd_props_zapobj = zap_create(mos, 890fa9e4066Sahrens DMU_OT_DSL_PROPS, DMU_OT_NONE, 0, tx); 891cde58dbcSMatthew Ahrens ddphys->dd_child_dir_zapobj = zap_create(mos, 89287e5029aSahrens DMU_OT_DSL_DIR_CHILD_MAP, DMU_OT_NONE, 0, tx); 89374e7dc98SMatthew Ahrens if (spa_version(dp->dp_spa) >= SPA_VERSION_USED_BREAKDOWN) 894cde58dbcSMatthew Ahrens ddphys->dd_flags |= DD_FLAG_USED_BREAKDOWN; 895ea8dc4b6Seschrock dmu_buf_rele(dbuf, FTAG); 896fa9e4066Sahrens 8971d452cf5Sahrens return (ddobj); 898fa9e4066Sahrens } 899fa9e4066Sahrens 900088f3894Sahrens boolean_t 901088f3894Sahrens dsl_dir_is_clone(dsl_dir_t *dd) 902fa9e4066Sahrens { 903*c1379625SJustin T. Gibbs return (dsl_dir_phys(dd)->dd_origin_obj && 904088f3894Sahrens (dd->dd_pool->dp_origin_snap == NULL || 905*c1379625SJustin T. Gibbs dsl_dir_phys(dd)->dd_origin_obj != 906088f3894Sahrens dd->dd_pool->dp_origin_snap->ds_object)); 907fa9e4066Sahrens } 908fa9e4066Sahrens 909fa9e4066Sahrens void 910a2eea2e1Sahrens dsl_dir_stats(dsl_dir_t *dd, nvlist_t *nv) 911fa9e4066Sahrens { 912fa9e4066Sahrens mutex_enter(&dd->dd_lock); 91374e7dc98SMatthew Ahrens dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USED, 914*c1379625SJustin T. Gibbs dsl_dir_phys(dd)->dd_used_bytes); 915*c1379625SJustin T. Gibbs dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_QUOTA, 916*c1379625SJustin T. Gibbs dsl_dir_phys(dd)->dd_quota); 917a2eea2e1Sahrens dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_RESERVATION, 918*c1379625SJustin T. Gibbs dsl_dir_phys(dd)->dd_reserved); 919a2eea2e1Sahrens dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_COMPRESSRATIO, 920*c1379625SJustin T. Gibbs dsl_dir_phys(dd)->dd_compressed_bytes == 0 ? 100 : 921*c1379625SJustin T. Gibbs (dsl_dir_phys(dd)->dd_uncompressed_bytes * 100 / 922*c1379625SJustin T. Gibbs dsl_dir_phys(dd)->dd_compressed_bytes)); 92377372cb0SMatthew Ahrens dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_LOGICALUSED, 924*c1379625SJustin T. Gibbs dsl_dir_phys(dd)->dd_uncompressed_bytes); 925*c1379625SJustin T. Gibbs if (dsl_dir_phys(dd)->dd_flags & DD_FLAG_USED_BREAKDOWN) { 92674e7dc98SMatthew Ahrens dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDSNAP, 927*c1379625SJustin T. Gibbs dsl_dir_phys(dd)->dd_used_breakdown[DD_USED_SNAP]); 92874e7dc98SMatthew Ahrens dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDDS, 929*c1379625SJustin T. Gibbs dsl_dir_phys(dd)->dd_used_breakdown[DD_USED_HEAD]); 93074e7dc98SMatthew Ahrens dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDREFRESERV, 931*c1379625SJustin T. Gibbs dsl_dir_phys(dd)->dd_used_breakdown[DD_USED_REFRSRV]); 93274e7dc98SMatthew Ahrens dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDCHILD, 933*c1379625SJustin T. Gibbs dsl_dir_phys(dd)->dd_used_breakdown[DD_USED_CHILD] + 934*c1379625SJustin T. Gibbs dsl_dir_phys(dd)->dd_used_breakdown[DD_USED_CHILD_RSRV]); 93574e7dc98SMatthew Ahrens } 936fa9e4066Sahrens mutex_exit(&dd->dd_lock); 937fa9e4066Sahrens 938a2afb611SJerry Jelinek if (dsl_dir_is_zapified(dd)) { 939a2afb611SJerry Jelinek uint64_t count; 940a2afb611SJerry Jelinek objset_t *os = dd->dd_pool->dp_meta_objset; 941a2afb611SJerry Jelinek 942a2afb611SJerry Jelinek if (zap_lookup(os, dd->dd_object, DD_FIELD_FILESYSTEM_COUNT, 943a2afb611SJerry Jelinek sizeof (count), 1, &count) == 0) { 944a2afb611SJerry Jelinek dsl_prop_nvlist_add_uint64(nv, 945a2afb611SJerry Jelinek ZFS_PROP_FILESYSTEM_COUNT, count); 946a2afb611SJerry Jelinek } 947a2afb611SJerry Jelinek if (zap_lookup(os, dd->dd_object, DD_FIELD_SNAPSHOT_COUNT, 948a2afb611SJerry Jelinek sizeof (count), 1, &count) == 0) { 949a2afb611SJerry Jelinek dsl_prop_nvlist_add_uint64(nv, 950a2afb611SJerry Jelinek ZFS_PROP_SNAPSHOT_COUNT, count); 951a2afb611SJerry Jelinek } 952a2afb611SJerry Jelinek } 953a2afb611SJerry Jelinek 954088f3894Sahrens if (dsl_dir_is_clone(dd)) { 955fa9e4066Sahrens dsl_dataset_t *ds; 956a2eea2e1Sahrens char buf[MAXNAMELEN]; 957fa9e4066Sahrens 9583b2aab18SMatthew Ahrens VERIFY0(dsl_dataset_hold_obj(dd->dd_pool, 959*c1379625SJustin T. Gibbs dsl_dir_phys(dd)->dd_origin_obj, FTAG, &ds)); 960a2eea2e1Sahrens dsl_dataset_name(ds, buf); 961745cd3c5Smaybee dsl_dataset_rele(ds, FTAG); 962a2eea2e1Sahrens dsl_prop_nvlist_add_string(nv, ZFS_PROP_ORIGIN, buf); 963fa9e4066Sahrens } 964fa9e4066Sahrens } 965fa9e4066Sahrens 966fa9e4066Sahrens void 967fa9e4066Sahrens dsl_dir_dirty(dsl_dir_t *dd, dmu_tx_t *tx) 968fa9e4066Sahrens { 969fa9e4066Sahrens dsl_pool_t *dp = dd->dd_pool; 970fa9e4066Sahrens 971*c1379625SJustin T. Gibbs ASSERT(dsl_dir_phys(dd)); 972fa9e4066Sahrens 9733b2aab18SMatthew Ahrens if (txg_list_add(&dp->dp_dirty_dirs, dd, tx->tx_txg)) { 974fa9e4066Sahrens /* up the hold count until we can be written out */ 975fa9e4066Sahrens dmu_buf_add_ref(dd->dd_dbuf, dd); 976fa9e4066Sahrens } 977fa9e4066Sahrens } 978fa9e4066Sahrens 979fa9e4066Sahrens static int64_t 980fa9e4066Sahrens parent_delta(dsl_dir_t *dd, uint64_t used, int64_t delta) 981fa9e4066Sahrens { 982*c1379625SJustin T. Gibbs uint64_t old_accounted = MAX(used, dsl_dir_phys(dd)->dd_reserved); 983*c1379625SJustin T. Gibbs uint64_t new_accounted = 984*c1379625SJustin T. Gibbs MAX(used + delta, dsl_dir_phys(dd)->dd_reserved); 985fa9e4066Sahrens return (new_accounted - old_accounted); 986fa9e4066Sahrens } 987fa9e4066Sahrens 988fa9e4066Sahrens void 989fa9e4066Sahrens dsl_dir_sync(dsl_dir_t *dd, dmu_tx_t *tx) 990fa9e4066Sahrens { 991fa9e4066Sahrens ASSERT(dmu_tx_is_syncing(tx)); 992fa9e4066Sahrens 993fa9e4066Sahrens mutex_enter(&dd->dd_lock); 994fb09f5aaSMadhav Suresh ASSERT0(dd->dd_tempreserved[tx->tx_txg&TXG_MASK]); 995fa9e4066Sahrens dprintf_dd(dd, "txg=%llu towrite=%lluK\n", tx->tx_txg, 996fa9e4066Sahrens dd->dd_space_towrite[tx->tx_txg&TXG_MASK] / 1024); 997fa9e4066Sahrens dd->dd_space_towrite[tx->tx_txg&TXG_MASK] = 0; 998fa9e4066Sahrens mutex_exit(&dd->dd_lock); 999fa9e4066Sahrens 1000fa9e4066Sahrens /* release the hold from dsl_dir_dirty */ 1001ea8dc4b6Seschrock dmu_buf_rele(dd->dd_dbuf, dd); 1002fa9e4066Sahrens } 1003fa9e4066Sahrens 1004fa9e4066Sahrens static uint64_t 1005a9799022Sck dsl_dir_space_towrite(dsl_dir_t *dd) 1006fa9e4066Sahrens { 1007a9799022Sck uint64_t space = 0; 1008fa9e4066Sahrens int i; 1009fa9e4066Sahrens 1010fa9e4066Sahrens ASSERT(MUTEX_HELD(&dd->dd_lock)); 1011fa9e4066Sahrens 1012fa9e4066Sahrens for (i = 0; i < TXG_SIZE; i++) { 1013fa9e4066Sahrens space += dd->dd_space_towrite[i&TXG_MASK]; 1014fa9e4066Sahrens ASSERT3U(dd->dd_space_towrite[i&TXG_MASK], >=, 0); 1015fa9e4066Sahrens } 1016fa9e4066Sahrens return (space); 1017fa9e4066Sahrens } 1018fa9e4066Sahrens 1019fa9e4066Sahrens /* 1020fa9e4066Sahrens * How much space would dd have available if ancestor had delta applied 1021fa9e4066Sahrens * to it? If ondiskonly is set, we're only interested in what's 1022fa9e4066Sahrens * on-disk, not estimated pending changes. 1023fa9e4066Sahrens */ 1024a2eea2e1Sahrens uint64_t 1025fa9e4066Sahrens dsl_dir_space_available(dsl_dir_t *dd, 1026fa9e4066Sahrens dsl_dir_t *ancestor, int64_t delta, int ondiskonly) 1027fa9e4066Sahrens { 1028fa9e4066Sahrens uint64_t parentspace, myspace, quota, used; 1029fa9e4066Sahrens 1030fa9e4066Sahrens /* 1031fa9e4066Sahrens * If there are no restrictions otherwise, assume we have 1032fa9e4066Sahrens * unlimited space available. 1033fa9e4066Sahrens */ 1034fa9e4066Sahrens quota = UINT64_MAX; 1035fa9e4066Sahrens parentspace = UINT64_MAX; 1036fa9e4066Sahrens 1037fa9e4066Sahrens if (dd->dd_parent != NULL) { 1038fa9e4066Sahrens parentspace = dsl_dir_space_available(dd->dd_parent, 1039fa9e4066Sahrens ancestor, delta, ondiskonly); 1040fa9e4066Sahrens } 1041fa9e4066Sahrens 1042fa9e4066Sahrens mutex_enter(&dd->dd_lock); 1043*c1379625SJustin T. Gibbs if (dsl_dir_phys(dd)->dd_quota != 0) 1044*c1379625SJustin T. Gibbs quota = dsl_dir_phys(dd)->dd_quota; 1045*c1379625SJustin T. Gibbs used = dsl_dir_phys(dd)->dd_used_bytes; 1046a9799022Sck if (!ondiskonly) 1047a9799022Sck used += dsl_dir_space_towrite(dd); 1048fa9e4066Sahrens 1049fa9e4066Sahrens if (dd->dd_parent == NULL) { 105099653d4eSeschrock uint64_t poolsize = dsl_pool_adjustedsize(dd->dd_pool, FALSE); 1051fa9e4066Sahrens quota = MIN(quota, poolsize); 1052fa9e4066Sahrens } 1053fa9e4066Sahrens 1054*c1379625SJustin T. Gibbs if (dsl_dir_phys(dd)->dd_reserved > used && parentspace != UINT64_MAX) { 1055fa9e4066Sahrens /* 1056fa9e4066Sahrens * We have some space reserved, in addition to what our 1057fa9e4066Sahrens * parent gave us. 1058fa9e4066Sahrens */ 1059*c1379625SJustin T. Gibbs parentspace += dsl_dir_phys(dd)->dd_reserved - used; 1060fa9e4066Sahrens } 1061fa9e4066Sahrens 106274e7dc98SMatthew Ahrens if (dd == ancestor) { 106374e7dc98SMatthew Ahrens ASSERT(delta <= 0); 106474e7dc98SMatthew Ahrens ASSERT(used >= -delta); 106574e7dc98SMatthew Ahrens used += delta; 106674e7dc98SMatthew Ahrens if (parentspace != UINT64_MAX) 106774e7dc98SMatthew Ahrens parentspace -= delta; 106874e7dc98SMatthew Ahrens } 106974e7dc98SMatthew Ahrens 1070fa9e4066Sahrens if (used > quota) { 1071fa9e4066Sahrens /* over quota */ 1072fa9e4066Sahrens myspace = 0; 1073fa9e4066Sahrens } else { 1074fa9e4066Sahrens /* 107599653d4eSeschrock * the lesser of the space provided by our parent and 107699653d4eSeschrock * the space left in our quota 1077fa9e4066Sahrens */ 1078fa9e4066Sahrens myspace = MIN(parentspace, quota - used); 1079fa9e4066Sahrens } 1080fa9e4066Sahrens 1081fa9e4066Sahrens mutex_exit(&dd->dd_lock); 1082fa9e4066Sahrens 1083fa9e4066Sahrens return (myspace); 1084fa9e4066Sahrens } 1085fa9e4066Sahrens 1086fa9e4066Sahrens struct tempreserve { 1087fa9e4066Sahrens list_node_t tr_node; 1088fa9e4066Sahrens dsl_dir_t *tr_ds; 1089fa9e4066Sahrens uint64_t tr_size; 1090fa9e4066Sahrens }; 1091fa9e4066Sahrens 1092fa9e4066Sahrens static int 1093a9799022Sck dsl_dir_tempreserve_impl(dsl_dir_t *dd, uint64_t asize, boolean_t netfree, 1094a9799022Sck boolean_t ignorequota, boolean_t checkrefquota, list_t *tr_list, 10956df6c3bcSck dmu_tx_t *tx, boolean_t first) 1096fa9e4066Sahrens { 1097fa9e4066Sahrens uint64_t txg = tx->tx_txg; 1098a9799022Sck uint64_t est_inflight, used_on_disk, quota, parent_rsrv; 1099468c413aSTim Haley uint64_t deferred = 0; 1100a9799022Sck struct tempreserve *tr; 1101468c413aSTim Haley int retval = EDQUOT; 1102fa9e4066Sahrens int txgidx = txg & TXG_MASK; 1103fa9e4066Sahrens int i; 11049082849eSck uint64_t ref_rsrv = 0; 1105fa9e4066Sahrens 1106fa9e4066Sahrens ASSERT3U(txg, !=, 0); 1107a9799022Sck ASSERT3S(asize, >, 0); 1108fa9e4066Sahrens 1109fa9e4066Sahrens mutex_enter(&dd->dd_lock); 1110a9799022Sck 1111fa9e4066Sahrens /* 1112fa9e4066Sahrens * Check against the dsl_dir's quota. We don't add in the delta 1113fa9e4066Sahrens * when checking for over-quota because they get one free hit. 1114fa9e4066Sahrens */ 1115a9799022Sck est_inflight = dsl_dir_space_towrite(dd); 1116fa9e4066Sahrens for (i = 0; i < TXG_SIZE; i++) 1117a9799022Sck est_inflight += dd->dd_tempreserved[i]; 1118*c1379625SJustin T. Gibbs used_on_disk = dsl_dir_phys(dd)->dd_used_bytes; 1119fa9e4066Sahrens 1120f4d2e9e6Smaybee /* 11216df6c3bcSck * On the first iteration, fetch the dataset's used-on-disk and 11226df6c3bcSck * refreservation values. Also, if checkrefquota is set, test if 11236df6c3bcSck * allocating this space would exceed the dataset's refquota. 1124f4d2e9e6Smaybee */ 11256df6c3bcSck if (first && tx->tx_objset) { 1126c3fdb13aSmaybee int error; 1127503ad85cSMatthew Ahrens dsl_dataset_t *ds = tx->tx_objset->os_dsl_dataset; 11289082849eSck 1129a9799022Sck error = dsl_dataset_check_quota(ds, checkrefquota, 11309082849eSck asize, est_inflight, &used_on_disk, &ref_rsrv); 1131a9799022Sck if (error) { 1132a9799022Sck mutex_exit(&dd->dd_lock); 1133a9799022Sck return (error); 1134a9799022Sck } 1135a9799022Sck } 1136a9799022Sck 1137a9799022Sck /* 1138a9799022Sck * If this transaction will result in a net free of space, 1139a9799022Sck * we want to let it through. 1140a9799022Sck */ 1141*c1379625SJustin T. Gibbs if (ignorequota || netfree || dsl_dir_phys(dd)->dd_quota == 0) 1142f4d2e9e6Smaybee quota = UINT64_MAX; 1143f4d2e9e6Smaybee else 1144*c1379625SJustin T. Gibbs quota = dsl_dir_phys(dd)->dd_quota; 1145fa9e4066Sahrens 1146fa9e4066Sahrens /* 1147468c413aSTim Haley * Adjust the quota against the actual pool size at the root 1148468c413aSTim Haley * minus any outstanding deferred frees. 1149f4d2e9e6Smaybee * To ensure that it's possible to remove files from a full 1150f4d2e9e6Smaybee * pool without inducing transient overcommits, we throttle 1151fa9e4066Sahrens * netfree transactions against a quota that is slightly larger, 1152fa9e4066Sahrens * but still within the pool's allocation slop. In cases where 1153fa9e4066Sahrens * we're very close to full, this will allow a steady trickle of 1154fa9e4066Sahrens * removes to get through. 1155fa9e4066Sahrens */ 11561934e92fSmaybee if (dd->dd_parent == NULL) { 1157b24ab676SJeff Bonwick spa_t *spa = dd->dd_pool->dp_spa; 1158fa9e4066Sahrens uint64_t poolsize = dsl_pool_adjustedsize(dd->dd_pool, netfree); 1159b24ab676SJeff Bonwick deferred = metaslab_class_get_deferred(spa_normal_class(spa)); 1160468c413aSTim Haley if (poolsize - deferred < quota) { 1161468c413aSTim Haley quota = poolsize - deferred; 1162468c413aSTim Haley retval = ENOSPC; 1163fa9e4066Sahrens } 1164fa9e4066Sahrens } 1165fa9e4066Sahrens 1166fa9e4066Sahrens /* 1167fa9e4066Sahrens * If they are requesting more space, and our current estimate 1168a9799022Sck * is over quota, they get to try again unless the actual 1169ea8dc4b6Seschrock * on-disk is over quota and there are no pending changes (which 1170ea8dc4b6Seschrock * may free up space for us). 1171fa9e4066Sahrens */ 1172468c413aSTim Haley if (used_on_disk + est_inflight >= quota) { 1173468c413aSTim Haley if (est_inflight > 0 || used_on_disk < quota || 1174468c413aSTim Haley (retval == ENOSPC && used_on_disk < quota + deferred)) 1175468c413aSTim Haley retval = ERESTART; 1176a9799022Sck dprintf_dd(dd, "failing: used=%lluK inflight = %lluK " 1177fa9e4066Sahrens "quota=%lluK tr=%lluK err=%d\n", 1178a9799022Sck used_on_disk>>10, est_inflight>>10, 1179468c413aSTim Haley quota>>10, asize>>10, retval); 1180fa9e4066Sahrens mutex_exit(&dd->dd_lock); 1181be6fd75aSMatthew Ahrens return (SET_ERROR(retval)); 1182fa9e4066Sahrens } 1183fa9e4066Sahrens 1184fa9e4066Sahrens /* We need to up our estimated delta before dropping dd_lock */ 1185fa9e4066Sahrens dd->dd_tempreserved[txgidx] += asize; 1186fa9e4066Sahrens 11879082849eSck parent_rsrv = parent_delta(dd, used_on_disk + est_inflight, 11889082849eSck asize - ref_rsrv); 1189fa9e4066Sahrens mutex_exit(&dd->dd_lock); 1190fa9e4066Sahrens 11911ab7f2deSmaybee tr = kmem_zalloc(sizeof (struct tempreserve), KM_SLEEP); 1192fa9e4066Sahrens tr->tr_ds = dd; 1193fa9e4066Sahrens tr->tr_size = asize; 1194fa9e4066Sahrens list_insert_tail(tr_list, tr); 1195fa9e4066Sahrens 1196fa9e4066Sahrens /* see if it's OK with our parent */ 11971934e92fSmaybee if (dd->dd_parent && parent_rsrv) { 1198*c1379625SJustin T. Gibbs boolean_t ismos = (dsl_dir_phys(dd)->dd_head_dataset_obj == 0); 11991934e92fSmaybee 1200fa9e4066Sahrens return (dsl_dir_tempreserve_impl(dd->dd_parent, 12016df6c3bcSck parent_rsrv, netfree, ismos, TRUE, tr_list, tx, FALSE)); 1202fa9e4066Sahrens } else { 1203fa9e4066Sahrens return (0); 1204fa9e4066Sahrens } 1205fa9e4066Sahrens } 1206fa9e4066Sahrens 1207fa9e4066Sahrens /* 1208fa9e4066Sahrens * Reserve space in this dsl_dir, to be used in this tx's txg. 1209a9799022Sck * After the space has been dirtied (and dsl_dir_willuse_space() 1210a9799022Sck * has been called), the reservation should be canceled, using 1211a9799022Sck * dsl_dir_tempreserve_clear(). 1212fa9e4066Sahrens */ 1213fa9e4066Sahrens int 1214a9799022Sck dsl_dir_tempreserve_space(dsl_dir_t *dd, uint64_t lsize, uint64_t asize, 1215a9799022Sck uint64_t fsize, uint64_t usize, void **tr_cookiep, dmu_tx_t *tx) 1216fa9e4066Sahrens { 12171ab7f2deSmaybee int err; 1218fa9e4066Sahrens list_t *tr_list; 1219fa9e4066Sahrens 1220a9799022Sck if (asize == 0) { 1221a9799022Sck *tr_cookiep = NULL; 1222a9799022Sck return (0); 1223a9799022Sck } 1224a9799022Sck 1225fa9e4066Sahrens tr_list = kmem_alloc(sizeof (list_t), KM_SLEEP); 1226fa9e4066Sahrens list_create(tr_list, sizeof (struct tempreserve), 1227fa9e4066Sahrens offsetof(struct tempreserve, tr_node)); 1228a9799022Sck ASSERT3S(asize, >, 0); 1229ea8dc4b6Seschrock ASSERT3S(fsize, >=, 0); 1230fa9e4066Sahrens 12311ab7f2deSmaybee err = arc_tempreserve_space(lsize, tx->tx_txg); 1232fa9e4066Sahrens if (err == 0) { 1233fa9e4066Sahrens struct tempreserve *tr; 1234fa9e4066Sahrens 12351ab7f2deSmaybee tr = kmem_zalloc(sizeof (struct tempreserve), KM_SLEEP); 12361ab7f2deSmaybee tr->tr_size = lsize; 12371ab7f2deSmaybee list_insert_tail(tr_list, tr); 12381ab7f2deSmaybee } else { 12391ab7f2deSmaybee if (err == EAGAIN) { 124069962b56SMatthew Ahrens /* 124169962b56SMatthew Ahrens * If arc_memory_throttle() detected that pageout 124269962b56SMatthew Ahrens * is running and we are low on memory, we delay new 124369962b56SMatthew Ahrens * non-pageout transactions to give pageout an 124469962b56SMatthew Ahrens * advantage. 124569962b56SMatthew Ahrens * 124669962b56SMatthew Ahrens * It is unfortunate to be delaying while the caller's 124769962b56SMatthew Ahrens * locks are held. 124869962b56SMatthew Ahrens */ 12490689f76cSAdam Leventhal txg_delay(dd->dd_pool, tx->tx_txg, 12500689f76cSAdam Leventhal MSEC2NSEC(10), MSEC2NSEC(10)); 1251be6fd75aSMatthew Ahrens err = SET_ERROR(ERESTART); 1252fa9e4066Sahrens } 12531ab7f2deSmaybee } 12541ab7f2deSmaybee 12551ab7f2deSmaybee if (err == 0) { 12561ab7f2deSmaybee err = dsl_dir_tempreserve_impl(dd, asize, fsize >= asize, 12576df6c3bcSck FALSE, asize > usize, tr_list, tx, TRUE); 1258fa9e4066Sahrens } 1259fa9e4066Sahrens 12603b2aab18SMatthew Ahrens if (err != 0) 1261fa9e4066Sahrens dsl_dir_tempreserve_clear(tr_list, tx); 1262fa9e4066Sahrens else 1263fa9e4066Sahrens *tr_cookiep = tr_list; 12641ab7f2deSmaybee 1265fa9e4066Sahrens return (err); 1266fa9e4066Sahrens } 1267fa9e4066Sahrens 1268fa9e4066Sahrens /* 1269fa9e4066Sahrens * Clear a temporary reservation that we previously made with 1270fa9e4066Sahrens * dsl_dir_tempreserve_space(). 1271fa9e4066Sahrens */ 1272fa9e4066Sahrens void 1273fa9e4066Sahrens dsl_dir_tempreserve_clear(void *tr_cookie, dmu_tx_t *tx) 1274fa9e4066Sahrens { 1275fa9e4066Sahrens int txgidx = tx->tx_txg & TXG_MASK; 1276fa9e4066Sahrens list_t *tr_list = tr_cookie; 1277fa9e4066Sahrens struct tempreserve *tr; 1278fa9e4066Sahrens 1279fa9e4066Sahrens ASSERT3U(tx->tx_txg, !=, 0); 1280fa9e4066Sahrens 1281a9799022Sck if (tr_cookie == NULL) 1282a9799022Sck return; 1283a9799022Sck 128469962b56SMatthew Ahrens while ((tr = list_head(tr_list)) != NULL) { 128569962b56SMatthew Ahrens if (tr->tr_ds) { 1286fa9e4066Sahrens mutex_enter(&tr->tr_ds->dd_lock); 1287fa9e4066Sahrens ASSERT3U(tr->tr_ds->dd_tempreserved[txgidx], >=, 1288fa9e4066Sahrens tr->tr_size); 1289fa9e4066Sahrens tr->tr_ds->dd_tempreserved[txgidx] -= tr->tr_size; 1290fa9e4066Sahrens mutex_exit(&tr->tr_ds->dd_lock); 12911ab7f2deSmaybee } else { 12921ab7f2deSmaybee arc_tempreserve_clear(tr->tr_size); 1293fa9e4066Sahrens } 1294fa9e4066Sahrens list_remove(tr_list, tr); 1295fa9e4066Sahrens kmem_free(tr, sizeof (struct tempreserve)); 1296fa9e4066Sahrens } 1297fa9e4066Sahrens 1298fa9e4066Sahrens kmem_free(tr_list, sizeof (list_t)); 1299fa9e4066Sahrens } 1300fa9e4066Sahrens 130169962b56SMatthew Ahrens /* 130269962b56SMatthew Ahrens * This should be called from open context when we think we're going to write 130369962b56SMatthew Ahrens * or free space, for example when dirtying data. Be conservative; it's okay 130469962b56SMatthew Ahrens * to write less space or free more, but we don't want to write more or free 130569962b56SMatthew Ahrens * less than the amount specified. 130669962b56SMatthew Ahrens */ 130769962b56SMatthew Ahrens void 130869962b56SMatthew Ahrens dsl_dir_willuse_space(dsl_dir_t *dd, int64_t space, dmu_tx_t *tx) 1309fa9e4066Sahrens { 1310fa9e4066Sahrens int64_t parent_space; 1311fa9e4066Sahrens uint64_t est_used; 1312fa9e4066Sahrens 1313fa9e4066Sahrens mutex_enter(&dd->dd_lock); 1314fa9e4066Sahrens if (space > 0) 1315fa9e4066Sahrens dd->dd_space_towrite[tx->tx_txg & TXG_MASK] += space; 1316fa9e4066Sahrens 1317*c1379625SJustin T. Gibbs est_used = dsl_dir_space_towrite(dd) + dsl_dir_phys(dd)->dd_used_bytes; 1318fa9e4066Sahrens parent_space = parent_delta(dd, est_used, space); 1319fa9e4066Sahrens mutex_exit(&dd->dd_lock); 1320fa9e4066Sahrens 1321fa9e4066Sahrens /* Make sure that we clean up dd_space_to* */ 1322fa9e4066Sahrens dsl_dir_dirty(dd, tx); 1323fa9e4066Sahrens 1324fa9e4066Sahrens /* XXX this is potentially expensive and unnecessary... */ 1325fa9e4066Sahrens if (parent_space && dd->dd_parent) 132669962b56SMatthew Ahrens dsl_dir_willuse_space(dd->dd_parent, parent_space, tx); 1327fa9e4066Sahrens } 1328fa9e4066Sahrens 1329fa9e4066Sahrens /* call from syncing context when we actually write/free space for this dd */ 1330fa9e4066Sahrens void 133174e7dc98SMatthew Ahrens dsl_dir_diduse_space(dsl_dir_t *dd, dd_used_t type, 1332fa9e4066Sahrens int64_t used, int64_t compressed, int64_t uncompressed, dmu_tx_t *tx) 1333fa9e4066Sahrens { 1334fa9e4066Sahrens int64_t accounted_delta; 1335b62969f8SMatthew Ahrens 1336b62969f8SMatthew Ahrens /* 1337b62969f8SMatthew Ahrens * dsl_dataset_set_refreservation_sync_impl() calls this with 1338b62969f8SMatthew Ahrens * dd_lock held, so that it can atomically update 1339b62969f8SMatthew Ahrens * ds->ds_reserved and the dsl_dir accounting, so that 1340b62969f8SMatthew Ahrens * dsl_dataset_check_quota() can see dataset and dir accounting 1341b62969f8SMatthew Ahrens * consistently. 1342b62969f8SMatthew Ahrens */ 134302c8f3f0SMatthew Ahrens boolean_t needlock = !MUTEX_HELD(&dd->dd_lock); 1344fa9e4066Sahrens 1345fa9e4066Sahrens ASSERT(dmu_tx_is_syncing(tx)); 134674e7dc98SMatthew Ahrens ASSERT(type < DD_USED_NUM); 1347fa9e4066Sahrens 1348b62969f8SMatthew Ahrens dmu_buf_will_dirty(dd->dd_dbuf, tx); 1349b62969f8SMatthew Ahrens 135002c8f3f0SMatthew Ahrens if (needlock) 135102c8f3f0SMatthew Ahrens mutex_enter(&dd->dd_lock); 1352*c1379625SJustin T. Gibbs accounted_delta = 1353*c1379625SJustin T. Gibbs parent_delta(dd, dsl_dir_phys(dd)->dd_used_bytes, used); 1354*c1379625SJustin T. Gibbs ASSERT(used >= 0 || dsl_dir_phys(dd)->dd_used_bytes >= -used); 1355fa9e4066Sahrens ASSERT(compressed >= 0 || 1356*c1379625SJustin T. Gibbs dsl_dir_phys(dd)->dd_compressed_bytes >= -compressed); 1357fa9e4066Sahrens ASSERT(uncompressed >= 0 || 1358*c1379625SJustin T. Gibbs dsl_dir_phys(dd)->dd_uncompressed_bytes >= -uncompressed); 1359*c1379625SJustin T. Gibbs dsl_dir_phys(dd)->dd_used_bytes += used; 1360*c1379625SJustin T. Gibbs dsl_dir_phys(dd)->dd_uncompressed_bytes += uncompressed; 1361*c1379625SJustin T. Gibbs dsl_dir_phys(dd)->dd_compressed_bytes += compressed; 136274e7dc98SMatthew Ahrens 1363*c1379625SJustin T. Gibbs if (dsl_dir_phys(dd)->dd_flags & DD_FLAG_USED_BREAKDOWN) { 136474e7dc98SMatthew Ahrens ASSERT(used > 0 || 1365*c1379625SJustin T. Gibbs dsl_dir_phys(dd)->dd_used_breakdown[type] >= -used); 1366*c1379625SJustin T. Gibbs dsl_dir_phys(dd)->dd_used_breakdown[type] += used; 136774e7dc98SMatthew Ahrens #ifdef DEBUG 136874e7dc98SMatthew Ahrens dd_used_t t; 136974e7dc98SMatthew Ahrens uint64_t u = 0; 137074e7dc98SMatthew Ahrens for (t = 0; t < DD_USED_NUM; t++) 1371*c1379625SJustin T. Gibbs u += dsl_dir_phys(dd)->dd_used_breakdown[t]; 1372*c1379625SJustin T. Gibbs ASSERT3U(u, ==, dsl_dir_phys(dd)->dd_used_bytes); 137374e7dc98SMatthew Ahrens #endif 137474e7dc98SMatthew Ahrens } 137502c8f3f0SMatthew Ahrens if (needlock) 137602c8f3f0SMatthew Ahrens mutex_exit(&dd->dd_lock); 1377fa9e4066Sahrens 1378fa9e4066Sahrens if (dd->dd_parent != NULL) { 137974e7dc98SMatthew Ahrens dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD, 1380fa9e4066Sahrens accounted_delta, compressed, uncompressed, tx); 138174e7dc98SMatthew Ahrens dsl_dir_transfer_space(dd->dd_parent, 138274e7dc98SMatthew Ahrens used - accounted_delta, 138374e7dc98SMatthew Ahrens DD_USED_CHILD_RSRV, DD_USED_CHILD, tx); 1384fa9e4066Sahrens } 1385fa9e4066Sahrens } 1386fa9e4066Sahrens 138774e7dc98SMatthew Ahrens void 138874e7dc98SMatthew Ahrens dsl_dir_transfer_space(dsl_dir_t *dd, int64_t delta, 138974e7dc98SMatthew Ahrens dd_used_t oldtype, dd_used_t newtype, dmu_tx_t *tx) 139074e7dc98SMatthew Ahrens { 139174e7dc98SMatthew Ahrens ASSERT(dmu_tx_is_syncing(tx)); 139274e7dc98SMatthew Ahrens ASSERT(oldtype < DD_USED_NUM); 139374e7dc98SMatthew Ahrens ASSERT(newtype < DD_USED_NUM); 139474e7dc98SMatthew Ahrens 1395*c1379625SJustin T. Gibbs if (delta == 0 || 1396*c1379625SJustin T. Gibbs !(dsl_dir_phys(dd)->dd_flags & DD_FLAG_USED_BREAKDOWN)) 139774e7dc98SMatthew Ahrens return; 139874e7dc98SMatthew Ahrens 1399b62969f8SMatthew Ahrens dmu_buf_will_dirty(dd->dd_dbuf, tx); 1400b62969f8SMatthew Ahrens mutex_enter(&dd->dd_lock); 140174e7dc98SMatthew Ahrens ASSERT(delta > 0 ? 1402*c1379625SJustin T. Gibbs dsl_dir_phys(dd)->dd_used_breakdown[oldtype] >= delta : 1403*c1379625SJustin T. Gibbs dsl_dir_phys(dd)->dd_used_breakdown[newtype] >= -delta); 1404*c1379625SJustin T. Gibbs ASSERT(dsl_dir_phys(dd)->dd_used_bytes >= ABS(delta)); 1405*c1379625SJustin T. Gibbs dsl_dir_phys(dd)->dd_used_breakdown[oldtype] -= delta; 1406*c1379625SJustin T. Gibbs dsl_dir_phys(dd)->dd_used_breakdown[newtype] += delta; 1407b62969f8SMatthew Ahrens mutex_exit(&dd->dd_lock); 140874e7dc98SMatthew Ahrens } 140974e7dc98SMatthew Ahrens 14103b2aab18SMatthew Ahrens typedef struct dsl_dir_set_qr_arg { 14113b2aab18SMatthew Ahrens const char *ddsqra_name; 14123b2aab18SMatthew Ahrens zprop_source_t ddsqra_source; 14133b2aab18SMatthew Ahrens uint64_t ddsqra_value; 14143b2aab18SMatthew Ahrens } dsl_dir_set_qr_arg_t; 14153b2aab18SMatthew Ahrens 1416fa9e4066Sahrens static int 14173b2aab18SMatthew Ahrens dsl_dir_set_quota_check(void *arg, dmu_tx_t *tx) 1418fa9e4066Sahrens { 14193b2aab18SMatthew Ahrens dsl_dir_set_qr_arg_t *ddsqra = arg; 14203b2aab18SMatthew Ahrens dsl_pool_t *dp = dmu_tx_pool(tx); 14213b2aab18SMatthew Ahrens dsl_dataset_t *ds; 14223b2aab18SMatthew Ahrens int error; 14233b2aab18SMatthew Ahrens uint64_t towrite, newval; 1424fa9e4066Sahrens 14253b2aab18SMatthew Ahrens error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds); 14263b2aab18SMatthew Ahrens if (error != 0) 14273b2aab18SMatthew Ahrens return (error); 14283b2aab18SMatthew Ahrens 14293b2aab18SMatthew Ahrens error = dsl_prop_predict(ds->ds_dir, "quota", 14303b2aab18SMatthew Ahrens ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval); 14313b2aab18SMatthew Ahrens if (error != 0) { 14323b2aab18SMatthew Ahrens dsl_dataset_rele(ds, FTAG); 14333b2aab18SMatthew Ahrens return (error); 14343b2aab18SMatthew Ahrens } 143592241e0bSTom Erickson 14363b2aab18SMatthew Ahrens if (newval == 0) { 14373b2aab18SMatthew Ahrens dsl_dataset_rele(ds, FTAG); 14381d452cf5Sahrens return (0); 14393b2aab18SMatthew Ahrens } 1440fa9e4066Sahrens 14413b2aab18SMatthew Ahrens mutex_enter(&ds->ds_dir->dd_lock); 14421d452cf5Sahrens /* 14431d452cf5Sahrens * If we are doing the preliminary check in open context, and 14441d452cf5Sahrens * there are pending changes, then don't fail it, since the 1445a9799022Sck * pending changes could under-estimate the amount of space to be 14461d452cf5Sahrens * freed up. 14471d452cf5Sahrens */ 14483b2aab18SMatthew Ahrens towrite = dsl_dir_space_towrite(ds->ds_dir); 14491d452cf5Sahrens if ((dmu_tx_is_syncing(tx) || towrite == 0) && 1450*c1379625SJustin T. Gibbs (newval < dsl_dir_phys(ds->ds_dir)->dd_reserved || 1451*c1379625SJustin T. Gibbs newval < dsl_dir_phys(ds->ds_dir)->dd_used_bytes + towrite)) { 1452be6fd75aSMatthew Ahrens error = SET_ERROR(ENOSPC); 1453fa9e4066Sahrens } 14543b2aab18SMatthew Ahrens mutex_exit(&ds->ds_dir->dd_lock); 14553b2aab18SMatthew Ahrens dsl_dataset_rele(ds, FTAG); 14563b2aab18SMatthew Ahrens return (error); 1457fa9e4066Sahrens } 1458fa9e4066Sahrens 14591d452cf5Sahrens static void 14603b2aab18SMatthew Ahrens dsl_dir_set_quota_sync(void *arg, dmu_tx_t *tx) 14611d452cf5Sahrens { 14623b2aab18SMatthew Ahrens dsl_dir_set_qr_arg_t *ddsqra = arg; 14633b2aab18SMatthew Ahrens dsl_pool_t *dp = dmu_tx_pool(tx); 14643b2aab18SMatthew Ahrens dsl_dataset_t *ds; 14653b2aab18SMatthew Ahrens uint64_t newval; 146692241e0bSTom Erickson 14673b2aab18SMatthew Ahrens VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds)); 14681d452cf5Sahrens 1469013023d4SMartin Matuska if (spa_version(dp->dp_spa) >= SPA_VERSION_RECVD_PROPS) { 1470013023d4SMartin Matuska dsl_prop_set_sync_impl(ds, zfs_prop_to_name(ZFS_PROP_QUOTA), 1471013023d4SMartin Matuska ddsqra->ddsqra_source, sizeof (ddsqra->ddsqra_value), 1, 1472013023d4SMartin Matuska &ddsqra->ddsqra_value, tx); 14731d452cf5Sahrens 1474013023d4SMartin Matuska VERIFY0(dsl_prop_get_int_ds(ds, 1475013023d4SMartin Matuska zfs_prop_to_name(ZFS_PROP_QUOTA), &newval)); 1476013023d4SMartin Matuska } else { 1477013023d4SMartin Matuska newval = ddsqra->ddsqra_value; 1478013023d4SMartin Matuska spa_history_log_internal_ds(ds, "set", tx, "%s=%lld", 1479013023d4SMartin Matuska zfs_prop_to_name(ZFS_PROP_QUOTA), (longlong_t)newval); 1480013023d4SMartin Matuska } 14813b2aab18SMatthew Ahrens 14823b2aab18SMatthew Ahrens dmu_buf_will_dirty(ds->ds_dir->dd_dbuf, tx); 14833b2aab18SMatthew Ahrens mutex_enter(&ds->ds_dir->dd_lock); 1484*c1379625SJustin T. Gibbs dsl_dir_phys(ds->ds_dir)->dd_quota = newval; 14853b2aab18SMatthew Ahrens mutex_exit(&ds->ds_dir->dd_lock); 14863b2aab18SMatthew Ahrens dsl_dataset_rele(ds, FTAG); 14871d452cf5Sahrens } 14881d452cf5Sahrens 1489fa9e4066Sahrens int 149092241e0bSTom Erickson dsl_dir_set_quota(const char *ddname, zprop_source_t source, uint64_t quota) 1491fa9e4066Sahrens { 14923b2aab18SMatthew Ahrens dsl_dir_set_qr_arg_t ddsqra; 1493fa9e4066Sahrens 14943b2aab18SMatthew Ahrens ddsqra.ddsqra_name = ddname; 14953b2aab18SMatthew Ahrens ddsqra.ddsqra_source = source; 14963b2aab18SMatthew Ahrens ddsqra.ddsqra_value = quota; 149792241e0bSTom Erickson 14983b2aab18SMatthew Ahrens return (dsl_sync_task(ddname, dsl_dir_set_quota_check, 14997d46dc6cSMatthew Ahrens dsl_dir_set_quota_sync, &ddsqra, 0, ZFS_SPACE_CHECK_NONE)); 1500fa9e4066Sahrens } 1501fa9e4066Sahrens 1502a9799022Sck int 15033b2aab18SMatthew Ahrens dsl_dir_set_reservation_check(void *arg, dmu_tx_t *tx) 1504fa9e4066Sahrens { 15053b2aab18SMatthew Ahrens dsl_dir_set_qr_arg_t *ddsqra = arg; 15063b2aab18SMatthew Ahrens dsl_pool_t *dp = dmu_tx_pool(tx); 15073b2aab18SMatthew Ahrens dsl_dataset_t *ds; 15083b2aab18SMatthew Ahrens dsl_dir_t *dd; 15093b2aab18SMatthew Ahrens uint64_t newval, used, avail; 15103b2aab18SMatthew Ahrens int error; 151192241e0bSTom Erickson 15123b2aab18SMatthew Ahrens error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds); 15133b2aab18SMatthew Ahrens if (error != 0) 15143b2aab18SMatthew Ahrens return (error); 15153b2aab18SMatthew Ahrens dd = ds->ds_dir; 1516fa9e4066Sahrens 15171d452cf5Sahrens /* 15181d452cf5Sahrens * If we are doing the preliminary check in open context, the 15191d452cf5Sahrens * space estimates may be inaccurate. 15201d452cf5Sahrens */ 15213b2aab18SMatthew Ahrens if (!dmu_tx_is_syncing(tx)) { 15223b2aab18SMatthew Ahrens dsl_dataset_rele(ds, FTAG); 15231d452cf5Sahrens return (0); 15243b2aab18SMatthew Ahrens } 15253b2aab18SMatthew Ahrens 15263b2aab18SMatthew Ahrens error = dsl_prop_predict(ds->ds_dir, 15273b2aab18SMatthew Ahrens zfs_prop_to_name(ZFS_PROP_RESERVATION), 15283b2aab18SMatthew Ahrens ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval); 15293b2aab18SMatthew Ahrens if (error != 0) { 15303b2aab18SMatthew Ahrens dsl_dataset_rele(ds, FTAG); 15313b2aab18SMatthew Ahrens return (error); 15323b2aab18SMatthew Ahrens } 15331d452cf5Sahrens 1534fa9e4066Sahrens mutex_enter(&dd->dd_lock); 1535*c1379625SJustin T. Gibbs used = dsl_dir_phys(dd)->dd_used_bytes; 1536fa9e4066Sahrens mutex_exit(&dd->dd_lock); 1537fa9e4066Sahrens 1538fa9e4066Sahrens if (dd->dd_parent) { 1539fa9e4066Sahrens avail = dsl_dir_space_available(dd->dd_parent, 1540fa9e4066Sahrens NULL, 0, FALSE); 1541fa9e4066Sahrens } else { 1542fa9e4066Sahrens avail = dsl_pool_adjustedsize(dd->dd_pool, B_FALSE) - used; 1543fa9e4066Sahrens } 1544fa9e4066Sahrens 1545*c1379625SJustin T. Gibbs if (MAX(used, newval) > MAX(used, dsl_dir_phys(dd)->dd_reserved)) { 15463b2aab18SMatthew Ahrens uint64_t delta = MAX(used, newval) - 1547*c1379625SJustin T. Gibbs MAX(used, dsl_dir_phys(dd)->dd_reserved); 1548379c004dSEric Schrock 15493b2aab18SMatthew Ahrens if (delta > avail || 1550*c1379625SJustin T. Gibbs (dsl_dir_phys(dd)->dd_quota > 0 && 1551*c1379625SJustin T. Gibbs newval > dsl_dir_phys(dd)->dd_quota)) 1552be6fd75aSMatthew Ahrens error = SET_ERROR(ENOSPC); 1553379c004dSEric Schrock } 1554379c004dSEric Schrock 15553b2aab18SMatthew Ahrens dsl_dataset_rele(ds, FTAG); 15563b2aab18SMatthew Ahrens return (error); 15571d452cf5Sahrens } 15581d452cf5Sahrens 15593b2aab18SMatthew Ahrens void 15604445fffbSMatthew Ahrens dsl_dir_set_reservation_sync_impl(dsl_dir_t *dd, uint64_t value, dmu_tx_t *tx) 15611d452cf5Sahrens { 15621d452cf5Sahrens uint64_t used; 15631d452cf5Sahrens int64_t delta; 15641d452cf5Sahrens 1565a9799022Sck dmu_buf_will_dirty(dd->dd_dbuf, tx); 1566a9799022Sck 15671d452cf5Sahrens mutex_enter(&dd->dd_lock); 1568*c1379625SJustin T. Gibbs used = dsl_dir_phys(dd)->dd_used_bytes; 1569*c1379625SJustin T. Gibbs delta = MAX(used, value) - MAX(used, dsl_dir_phys(dd)->dd_reserved); 1570*c1379625SJustin T. Gibbs dsl_dir_phys(dd)->dd_reserved = value; 1571fa9e4066Sahrens 1572fa9e4066Sahrens if (dd->dd_parent != NULL) { 1573fa9e4066Sahrens /* Roll up this additional usage into our ancestors */ 157474e7dc98SMatthew Ahrens dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD_RSRV, 157574e7dc98SMatthew Ahrens delta, 0, 0, tx); 1576fa9e4066Sahrens } 157702c8f3f0SMatthew Ahrens mutex_exit(&dd->dd_lock); 15784445fffbSMatthew Ahrens } 15794445fffbSMatthew Ahrens 15804445fffbSMatthew Ahrens 15814445fffbSMatthew Ahrens static void 15823b2aab18SMatthew Ahrens dsl_dir_set_reservation_sync(void *arg, dmu_tx_t *tx) 15834445fffbSMatthew Ahrens { 15843b2aab18SMatthew Ahrens dsl_dir_set_qr_arg_t *ddsqra = arg; 15853b2aab18SMatthew Ahrens dsl_pool_t *dp = dmu_tx_pool(tx); 15863b2aab18SMatthew Ahrens dsl_dataset_t *ds; 15873b2aab18SMatthew Ahrens uint64_t newval; 15883b2aab18SMatthew Ahrens 15893b2aab18SMatthew Ahrens VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds)); 15903b2aab18SMatthew Ahrens 1591013023d4SMartin Matuska if (spa_version(dp->dp_spa) >= SPA_VERSION_RECVD_PROPS) { 1592013023d4SMartin Matuska dsl_prop_set_sync_impl(ds, 1593013023d4SMartin Matuska zfs_prop_to_name(ZFS_PROP_RESERVATION), 1594013023d4SMartin Matuska ddsqra->ddsqra_source, sizeof (ddsqra->ddsqra_value), 1, 1595013023d4SMartin Matuska &ddsqra->ddsqra_value, tx); 15964445fffbSMatthew Ahrens 1597013023d4SMartin Matuska VERIFY0(dsl_prop_get_int_ds(ds, 1598013023d4SMartin Matuska zfs_prop_to_name(ZFS_PROP_RESERVATION), &newval)); 1599013023d4SMartin Matuska } else { 1600013023d4SMartin Matuska newval = ddsqra->ddsqra_value; 1601013023d4SMartin Matuska spa_history_log_internal_ds(ds, "set", tx, "%s=%lld", 1602013023d4SMartin Matuska zfs_prop_to_name(ZFS_PROP_RESERVATION), 1603013023d4SMartin Matuska (longlong_t)newval); 1604013023d4SMartin Matuska } 1605ecd6cf80Smarks 16063b2aab18SMatthew Ahrens dsl_dir_set_reservation_sync_impl(ds->ds_dir, newval, tx); 16073b2aab18SMatthew Ahrens dsl_dataset_rele(ds, FTAG); 1608fa9e4066Sahrens } 1609fa9e4066Sahrens 1610fa9e4066Sahrens int 161192241e0bSTom Erickson dsl_dir_set_reservation(const char *ddname, zprop_source_t source, 161292241e0bSTom Erickson uint64_t reservation) 1613fa9e4066Sahrens { 16143b2aab18SMatthew Ahrens dsl_dir_set_qr_arg_t ddsqra; 161592241e0bSTom Erickson 16163b2aab18SMatthew Ahrens ddsqra.ddsqra_name = ddname; 16173b2aab18SMatthew Ahrens ddsqra.ddsqra_source = source; 16183b2aab18SMatthew Ahrens ddsqra.ddsqra_value = reservation; 161992241e0bSTom Erickson 16203b2aab18SMatthew Ahrens return (dsl_sync_task(ddname, dsl_dir_set_reservation_check, 16217d46dc6cSMatthew Ahrens dsl_dir_set_reservation_sync, &ddsqra, 0, ZFS_SPACE_CHECK_NONE)); 1622fa9e4066Sahrens } 1623fa9e4066Sahrens 1624fa9e4066Sahrens static dsl_dir_t * 1625fa9e4066Sahrens closest_common_ancestor(dsl_dir_t *ds1, dsl_dir_t *ds2) 1626fa9e4066Sahrens { 1627fa9e4066Sahrens for (; ds1; ds1 = ds1->dd_parent) { 1628fa9e4066Sahrens dsl_dir_t *dd; 1629fa9e4066Sahrens for (dd = ds2; dd; dd = dd->dd_parent) { 1630fa9e4066Sahrens if (ds1 == dd) 1631fa9e4066Sahrens return (dd); 1632fa9e4066Sahrens } 1633fa9e4066Sahrens } 1634fa9e4066Sahrens return (NULL); 1635fa9e4066Sahrens } 1636fa9e4066Sahrens 1637fa9e4066Sahrens /* 1638fa9e4066Sahrens * If delta is applied to dd, how much of that delta would be applied to 1639fa9e4066Sahrens * ancestor? Syncing context only. 1640fa9e4066Sahrens */ 1641fa9e4066Sahrens static int64_t 1642fa9e4066Sahrens would_change(dsl_dir_t *dd, int64_t delta, dsl_dir_t *ancestor) 1643fa9e4066Sahrens { 1644fa9e4066Sahrens if (dd == ancestor) 1645fa9e4066Sahrens return (delta); 1646fa9e4066Sahrens 1647fa9e4066Sahrens mutex_enter(&dd->dd_lock); 1648*c1379625SJustin T. Gibbs delta = parent_delta(dd, dsl_dir_phys(dd)->dd_used_bytes, delta); 1649fa9e4066Sahrens mutex_exit(&dd->dd_lock); 1650fa9e4066Sahrens return (would_change(dd->dd_parent, delta, ancestor)); 1651fa9e4066Sahrens } 1652fa9e4066Sahrens 16533b2aab18SMatthew Ahrens typedef struct dsl_dir_rename_arg { 16543b2aab18SMatthew Ahrens const char *ddra_oldname; 16553b2aab18SMatthew Ahrens const char *ddra_newname; 1656a2afb611SJerry Jelinek cred_t *ddra_cred; 16573b2aab18SMatthew Ahrens } dsl_dir_rename_arg_t; 16581d452cf5Sahrens 16593b2aab18SMatthew Ahrens /* ARGSUSED */ 16601d452cf5Sahrens static int 16613b2aab18SMatthew Ahrens dsl_valid_rename(dsl_pool_t *dp, dsl_dataset_t *ds, void *arg) 1662fa9e4066Sahrens { 16633b2aab18SMatthew Ahrens int *deltap = arg; 16643b2aab18SMatthew Ahrens char namebuf[MAXNAMELEN]; 1665fa9e4066Sahrens 16663b2aab18SMatthew Ahrens dsl_dataset_name(ds, namebuf); 16673b2aab18SMatthew Ahrens 16683b2aab18SMatthew Ahrens if (strlen(namebuf) + *deltap >= MAXNAMELEN) 1669be6fd75aSMatthew Ahrens return (SET_ERROR(ENAMETOOLONG)); 16703b2aab18SMatthew Ahrens return (0); 16713b2aab18SMatthew Ahrens } 16723b2aab18SMatthew Ahrens 16733b2aab18SMatthew Ahrens static int 16743b2aab18SMatthew Ahrens dsl_dir_rename_check(void *arg, dmu_tx_t *tx) 16753b2aab18SMatthew Ahrens { 16763b2aab18SMatthew Ahrens dsl_dir_rename_arg_t *ddra = arg; 16773b2aab18SMatthew Ahrens dsl_pool_t *dp = dmu_tx_pool(tx); 16783b2aab18SMatthew Ahrens dsl_dir_t *dd, *newparent; 16793b2aab18SMatthew Ahrens const char *mynewname; 16803b2aab18SMatthew Ahrens int error; 16813b2aab18SMatthew Ahrens int delta = strlen(ddra->ddra_newname) - strlen(ddra->ddra_oldname); 1682fa9e4066Sahrens 16833b2aab18SMatthew Ahrens /* target dir should exist */ 16843b2aab18SMatthew Ahrens error = dsl_dir_hold(dp, ddra->ddra_oldname, FTAG, &dd, NULL); 16853b2aab18SMatthew Ahrens if (error != 0) 16863b2aab18SMatthew Ahrens return (error); 16873b2aab18SMatthew Ahrens 16883b2aab18SMatthew Ahrens /* new parent should exist */ 16893b2aab18SMatthew Ahrens error = dsl_dir_hold(dp, ddra->ddra_newname, FTAG, 16903b2aab18SMatthew Ahrens &newparent, &mynewname); 16913b2aab18SMatthew Ahrens if (error != 0) { 16923b2aab18SMatthew Ahrens dsl_dir_rele(dd, FTAG); 16933b2aab18SMatthew Ahrens return (error); 16943b2aab18SMatthew Ahrens } 16953b2aab18SMatthew Ahrens 16963b2aab18SMatthew Ahrens /* can't rename to different pool */ 16973b2aab18SMatthew Ahrens if (dd->dd_pool != newparent->dd_pool) { 16983b2aab18SMatthew Ahrens dsl_dir_rele(newparent, FTAG); 16993b2aab18SMatthew Ahrens dsl_dir_rele(dd, FTAG); 1700be6fd75aSMatthew Ahrens return (SET_ERROR(ENXIO)); 17013b2aab18SMatthew Ahrens } 17023b2aab18SMatthew Ahrens 17033b2aab18SMatthew Ahrens /* new name should not already exist */ 17043b2aab18SMatthew Ahrens if (mynewname == NULL) { 17053b2aab18SMatthew Ahrens dsl_dir_rele(newparent, FTAG); 17063b2aab18SMatthew Ahrens dsl_dir_rele(dd, FTAG); 1707be6fd75aSMatthew Ahrens return (SET_ERROR(EEXIST)); 17083b2aab18SMatthew Ahrens } 17093b2aab18SMatthew Ahrens 17103b2aab18SMatthew Ahrens /* if the name length is growing, validate child name lengths */ 17113b2aab18SMatthew Ahrens if (delta > 0) { 17123b2aab18SMatthew Ahrens error = dmu_objset_find_dp(dp, dd->dd_object, dsl_valid_rename, 17133b2aab18SMatthew Ahrens &delta, DS_FIND_CHILDREN | DS_FIND_SNAPSHOTS); 17143b2aab18SMatthew Ahrens if (error != 0) { 17153b2aab18SMatthew Ahrens dsl_dir_rele(newparent, FTAG); 17163b2aab18SMatthew Ahrens dsl_dir_rele(dd, FTAG); 17173b2aab18SMatthew Ahrens return (error); 17183b2aab18SMatthew Ahrens } 17193b2aab18SMatthew Ahrens } 17201d452cf5Sahrens 1721a2afb611SJerry Jelinek if (dmu_tx_is_syncing(tx)) { 1722adf34077SJerry Jelinek if (spa_feature_is_active(dp->dp_spa, 1723a2afb611SJerry Jelinek SPA_FEATURE_FS_SS_LIMIT)) { 1724a2afb611SJerry Jelinek /* 1725a2afb611SJerry Jelinek * Although this is the check function and we don't 1726a2afb611SJerry Jelinek * normally make on-disk changes in check functions, 1727a2afb611SJerry Jelinek * we need to do that here. 1728a2afb611SJerry Jelinek * 1729a2afb611SJerry Jelinek * Ensure this portion of the tree's counts have been 1730a2afb611SJerry Jelinek * initialized in case the new parent has limits set. 1731a2afb611SJerry Jelinek */ 1732a2afb611SJerry Jelinek dsl_dir_init_fs_ss_count(dd, tx); 1733a2afb611SJerry Jelinek } 1734a2afb611SJerry Jelinek } 1735a2afb611SJerry Jelinek 17363b2aab18SMatthew Ahrens if (newparent != dd->dd_parent) { 173799653d4eSeschrock /* is there enough space? */ 173899653d4eSeschrock uint64_t myspace = 1739*c1379625SJustin T. Gibbs MAX(dsl_dir_phys(dd)->dd_used_bytes, 1740*c1379625SJustin T. Gibbs dsl_dir_phys(dd)->dd_reserved); 1741a2afb611SJerry Jelinek objset_t *os = dd->dd_pool->dp_meta_objset; 1742a2afb611SJerry Jelinek uint64_t fs_cnt = 0; 1743a2afb611SJerry Jelinek uint64_t ss_cnt = 0; 1744a2afb611SJerry Jelinek 1745a2afb611SJerry Jelinek if (dsl_dir_is_zapified(dd)) { 1746a2afb611SJerry Jelinek int err; 1747a2afb611SJerry Jelinek 1748a2afb611SJerry Jelinek err = zap_lookup(os, dd->dd_object, 1749a2afb611SJerry Jelinek DD_FIELD_FILESYSTEM_COUNT, sizeof (fs_cnt), 1, 1750a2afb611SJerry Jelinek &fs_cnt); 1751adf34077SJerry Jelinek if (err != ENOENT && err != 0) { 1752adf34077SJerry Jelinek dsl_dir_rele(newparent, FTAG); 1753adf34077SJerry Jelinek dsl_dir_rele(dd, FTAG); 1754a2afb611SJerry Jelinek return (err); 1755adf34077SJerry Jelinek } 1756a2afb611SJerry Jelinek 1757a2afb611SJerry Jelinek /* 1758a2afb611SJerry Jelinek * have to add 1 for the filesystem itself that we're 1759a2afb611SJerry Jelinek * moving 1760a2afb611SJerry Jelinek */ 1761a2afb611SJerry Jelinek fs_cnt++; 1762a2afb611SJerry Jelinek 1763a2afb611SJerry Jelinek err = zap_lookup(os, dd->dd_object, 1764a2afb611SJerry Jelinek DD_FIELD_SNAPSHOT_COUNT, sizeof (ss_cnt), 1, 1765a2afb611SJerry Jelinek &ss_cnt); 1766adf34077SJerry Jelinek if (err != ENOENT && err != 0) { 1767adf34077SJerry Jelinek dsl_dir_rele(newparent, FTAG); 1768adf34077SJerry Jelinek dsl_dir_rele(dd, FTAG); 1769a2afb611SJerry Jelinek return (err); 1770adf34077SJerry Jelinek } 1771a2afb611SJerry Jelinek } 1772fa9e4066Sahrens 17731d452cf5Sahrens /* no rename into our descendant */ 17743b2aab18SMatthew Ahrens if (closest_common_ancestor(dd, newparent) == dd) { 17753b2aab18SMatthew Ahrens dsl_dir_rele(newparent, FTAG); 17763b2aab18SMatthew Ahrens dsl_dir_rele(dd, FTAG); 1777be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 17783b2aab18SMatthew Ahrens } 1779fa9e4066Sahrens 17803b2aab18SMatthew Ahrens error = dsl_dir_transfer_possible(dd->dd_parent, 1781a2afb611SJerry Jelinek newparent, fs_cnt, ss_cnt, myspace, ddra->ddra_cred); 17823b2aab18SMatthew Ahrens if (error != 0) { 17833b2aab18SMatthew Ahrens dsl_dir_rele(newparent, FTAG); 17843b2aab18SMatthew Ahrens dsl_dir_rele(dd, FTAG); 17853b2aab18SMatthew Ahrens return (error); 17863b2aab18SMatthew Ahrens } 17871d452cf5Sahrens } 1788fa9e4066Sahrens 17893b2aab18SMatthew Ahrens dsl_dir_rele(newparent, FTAG); 17903b2aab18SMatthew Ahrens dsl_dir_rele(dd, FTAG); 17911d452cf5Sahrens return (0); 17921d452cf5Sahrens } 17931d452cf5Sahrens 17941d452cf5Sahrens static void 17953b2aab18SMatthew Ahrens dsl_dir_rename_sync(void *arg, dmu_tx_t *tx) 17961d452cf5Sahrens { 17973b2aab18SMatthew Ahrens dsl_dir_rename_arg_t *ddra = arg; 17983b2aab18SMatthew Ahrens dsl_pool_t *dp = dmu_tx_pool(tx); 17993b2aab18SMatthew Ahrens dsl_dir_t *dd, *newparent; 18003b2aab18SMatthew Ahrens const char *mynewname; 18013b2aab18SMatthew Ahrens int error; 18021d452cf5Sahrens objset_t *mos = dp->dp_meta_objset; 18031d452cf5Sahrens 18043b2aab18SMatthew Ahrens VERIFY0(dsl_dir_hold(dp, ddra->ddra_oldname, FTAG, &dd, NULL)); 18053b2aab18SMatthew Ahrens VERIFY0(dsl_dir_hold(dp, ddra->ddra_newname, FTAG, &newparent, 18063b2aab18SMatthew Ahrens &mynewname)); 18071d452cf5Sahrens 18084445fffbSMatthew Ahrens /* Log this before we change the name. */ 18094445fffbSMatthew Ahrens spa_history_log_internal_dd(dd, "rename", tx, 18103b2aab18SMatthew Ahrens "-> %s", ddra->ddra_newname); 18114445fffbSMatthew Ahrens 18123b2aab18SMatthew Ahrens if (newparent != dd->dd_parent) { 1813a2afb611SJerry Jelinek objset_t *os = dd->dd_pool->dp_meta_objset; 1814a2afb611SJerry Jelinek uint64_t fs_cnt = 0; 1815a2afb611SJerry Jelinek uint64_t ss_cnt = 0; 1816a2afb611SJerry Jelinek 1817a2afb611SJerry Jelinek /* 1818a2afb611SJerry Jelinek * We already made sure the dd counts were initialized in the 1819a2afb611SJerry Jelinek * check function. 1820a2afb611SJerry Jelinek */ 1821adf34077SJerry Jelinek if (spa_feature_is_active(dp->dp_spa, 1822a2afb611SJerry Jelinek SPA_FEATURE_FS_SS_LIMIT)) { 1823a2afb611SJerry Jelinek VERIFY0(zap_lookup(os, dd->dd_object, 1824a2afb611SJerry Jelinek DD_FIELD_FILESYSTEM_COUNT, sizeof (fs_cnt), 1, 1825a2afb611SJerry Jelinek &fs_cnt)); 1826a2afb611SJerry Jelinek /* add 1 for the filesystem itself that we're moving */ 1827a2afb611SJerry Jelinek fs_cnt++; 1828a2afb611SJerry Jelinek 1829a2afb611SJerry Jelinek VERIFY0(zap_lookup(os, dd->dd_object, 1830a2afb611SJerry Jelinek DD_FIELD_SNAPSHOT_COUNT, sizeof (ss_cnt), 1, 1831a2afb611SJerry Jelinek &ss_cnt)); 1832a2afb611SJerry Jelinek } 1833a2afb611SJerry Jelinek 1834a2afb611SJerry Jelinek dsl_fs_ss_count_adjust(dd->dd_parent, -fs_cnt, 1835a2afb611SJerry Jelinek DD_FIELD_FILESYSTEM_COUNT, tx); 1836a2afb611SJerry Jelinek dsl_fs_ss_count_adjust(newparent, fs_cnt, 1837a2afb611SJerry Jelinek DD_FIELD_FILESYSTEM_COUNT, tx); 1838a2afb611SJerry Jelinek 1839a2afb611SJerry Jelinek dsl_fs_ss_count_adjust(dd->dd_parent, -ss_cnt, 1840a2afb611SJerry Jelinek DD_FIELD_SNAPSHOT_COUNT, tx); 1841a2afb611SJerry Jelinek dsl_fs_ss_count_adjust(newparent, ss_cnt, 1842a2afb611SJerry Jelinek DD_FIELD_SNAPSHOT_COUNT, tx); 1843a2afb611SJerry Jelinek 184474e7dc98SMatthew Ahrens dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD, 1845*c1379625SJustin T. Gibbs -dsl_dir_phys(dd)->dd_used_bytes, 1846*c1379625SJustin T. Gibbs -dsl_dir_phys(dd)->dd_compressed_bytes, 1847*c1379625SJustin T. Gibbs -dsl_dir_phys(dd)->dd_uncompressed_bytes, tx); 18483b2aab18SMatthew Ahrens dsl_dir_diduse_space(newparent, DD_USED_CHILD, 1849*c1379625SJustin T. Gibbs dsl_dir_phys(dd)->dd_used_bytes, 1850*c1379625SJustin T. Gibbs dsl_dir_phys(dd)->dd_compressed_bytes, 1851*c1379625SJustin T. Gibbs dsl_dir_phys(dd)->dd_uncompressed_bytes, tx); 185274e7dc98SMatthew Ahrens 1853*c1379625SJustin T. Gibbs if (dsl_dir_phys(dd)->dd_reserved > 1854*c1379625SJustin T. Gibbs dsl_dir_phys(dd)->dd_used_bytes) { 1855*c1379625SJustin T. Gibbs uint64_t unused_rsrv = dsl_dir_phys(dd)->dd_reserved - 1856*c1379625SJustin T. Gibbs dsl_dir_phys(dd)->dd_used_bytes; 185774e7dc98SMatthew Ahrens 185874e7dc98SMatthew Ahrens dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD_RSRV, 185974e7dc98SMatthew Ahrens -unused_rsrv, 0, 0, tx); 18603b2aab18SMatthew Ahrens dsl_dir_diduse_space(newparent, DD_USED_CHILD_RSRV, 186174e7dc98SMatthew Ahrens unused_rsrv, 0, 0, tx); 186274e7dc98SMatthew Ahrens } 1863fa9e4066Sahrens } 1864fa9e4066Sahrens 1865fa9e4066Sahrens dmu_buf_will_dirty(dd->dd_dbuf, tx); 1866fa9e4066Sahrens 1867fa9e4066Sahrens /* remove from old parent zapobj */ 1868*c1379625SJustin T. Gibbs error = zap_remove(mos, 1869*c1379625SJustin T. Gibbs dsl_dir_phys(dd->dd_parent)->dd_child_dir_zapobj, 1870fa9e4066Sahrens dd->dd_myname, tx); 18713b2aab18SMatthew Ahrens ASSERT0(error); 1872fa9e4066Sahrens 18733b2aab18SMatthew Ahrens (void) strcpy(dd->dd_myname, mynewname); 18743b2aab18SMatthew Ahrens dsl_dir_rele(dd->dd_parent, dd); 1875*c1379625SJustin T. Gibbs dsl_dir_phys(dd)->dd_parent_obj = newparent->dd_object; 18763b2aab18SMatthew Ahrens VERIFY0(dsl_dir_hold_obj(dp, 18773b2aab18SMatthew Ahrens newparent->dd_object, NULL, dd, &dd->dd_parent)); 1878fa9e4066Sahrens 1879fa9e4066Sahrens /* add to new parent zapobj */ 1880*c1379625SJustin T. Gibbs VERIFY0(zap_add(mos, dsl_dir_phys(newparent)->dd_child_dir_zapobj, 18813b2aab18SMatthew Ahrens dd->dd_myname, 8, 1, &dd->dd_object, tx)); 18823b2aab18SMatthew Ahrens 18833b2aab18SMatthew Ahrens dsl_prop_notify_all(dd); 1884ecd6cf80Smarks 18853b2aab18SMatthew Ahrens dsl_dir_rele(newparent, FTAG); 18863b2aab18SMatthew Ahrens dsl_dir_rele(dd, FTAG); 18871d452cf5Sahrens } 1888fa9e4066Sahrens 18891d452cf5Sahrens int 18903b2aab18SMatthew Ahrens dsl_dir_rename(const char *oldname, const char *newname) 18911d452cf5Sahrens { 18923b2aab18SMatthew Ahrens dsl_dir_rename_arg_t ddra; 18931d452cf5Sahrens 18943b2aab18SMatthew Ahrens ddra.ddra_oldname = oldname; 18953b2aab18SMatthew Ahrens ddra.ddra_newname = newname; 1896a2afb611SJerry Jelinek ddra.ddra_cred = CRED(); 18971d452cf5Sahrens 18983b2aab18SMatthew Ahrens return (dsl_sync_task(oldname, 18997d46dc6cSMatthew Ahrens dsl_dir_rename_check, dsl_dir_rename_sync, &ddra, 19007d46dc6cSMatthew Ahrens 3, ZFS_SPACE_CHECK_RESERVED)); 1901fa9e4066Sahrens } 190299653d4eSeschrock 190399653d4eSeschrock int 1904a2afb611SJerry Jelinek dsl_dir_transfer_possible(dsl_dir_t *sdd, dsl_dir_t *tdd, 1905a2afb611SJerry Jelinek uint64_t fs_cnt, uint64_t ss_cnt, uint64_t space, cred_t *cr) 190699653d4eSeschrock { 190799653d4eSeschrock dsl_dir_t *ancestor; 190899653d4eSeschrock int64_t adelta; 190999653d4eSeschrock uint64_t avail; 1910a2afb611SJerry Jelinek int err; 191199653d4eSeschrock 191299653d4eSeschrock ancestor = closest_common_ancestor(sdd, tdd); 191399653d4eSeschrock adelta = would_change(sdd, -space, ancestor); 191499653d4eSeschrock avail = dsl_dir_space_available(tdd, ancestor, adelta, FALSE); 191599653d4eSeschrock if (avail < space) 1916be6fd75aSMatthew Ahrens return (SET_ERROR(ENOSPC)); 191799653d4eSeschrock 1918a2afb611SJerry Jelinek err = dsl_fs_ss_limit_check(tdd, fs_cnt, ZFS_PROP_FILESYSTEM_LIMIT, 1919a2afb611SJerry Jelinek ancestor, cr); 1920a2afb611SJerry Jelinek if (err != 0) 1921a2afb611SJerry Jelinek return (err); 1922a2afb611SJerry Jelinek err = dsl_fs_ss_limit_check(tdd, ss_cnt, ZFS_PROP_SNAPSHOT_LIMIT, 1923a2afb611SJerry Jelinek ancestor, cr); 1924a2afb611SJerry Jelinek if (err != 0) 1925a2afb611SJerry Jelinek return (err); 1926a2afb611SJerry Jelinek 192799653d4eSeschrock return (0); 192899653d4eSeschrock } 192971eb0538SChris Kirby 193071eb0538SChris Kirby timestruc_t 193171eb0538SChris Kirby dsl_dir_snap_cmtime(dsl_dir_t *dd) 193271eb0538SChris Kirby { 193371eb0538SChris Kirby timestruc_t t; 193471eb0538SChris Kirby 193571eb0538SChris Kirby mutex_enter(&dd->dd_lock); 193671eb0538SChris Kirby t = dd->dd_snap_cmtime; 193771eb0538SChris Kirby mutex_exit(&dd->dd_lock); 193871eb0538SChris Kirby 193971eb0538SChris Kirby return (t); 194071eb0538SChris Kirby } 194171eb0538SChris Kirby 194271eb0538SChris Kirby void 194371eb0538SChris Kirby dsl_dir_snap_cmtime_update(dsl_dir_t *dd) 194471eb0538SChris Kirby { 194571eb0538SChris Kirby timestruc_t t; 194671eb0538SChris Kirby 194771eb0538SChris Kirby gethrestime(&t); 194871eb0538SChris Kirby mutex_enter(&dd->dd_lock); 194971eb0538SChris Kirby dd->dd_snap_cmtime = t; 195071eb0538SChris Kirby mutex_exit(&dd->dd_lock); 195171eb0538SChris Kirby } 19522acef22dSMatthew Ahrens 19532acef22dSMatthew Ahrens void 19542acef22dSMatthew Ahrens dsl_dir_zapify(dsl_dir_t *dd, dmu_tx_t *tx) 19552acef22dSMatthew Ahrens { 19562acef22dSMatthew Ahrens objset_t *mos = dd->dd_pool->dp_meta_objset; 19572acef22dSMatthew Ahrens dmu_object_zapify(mos, dd->dd_object, DMU_OT_DSL_DIR, tx); 19582acef22dSMatthew Ahrens } 1959a2afb611SJerry Jelinek 1960a2afb611SJerry Jelinek boolean_t 1961a2afb611SJerry Jelinek dsl_dir_is_zapified(dsl_dir_t *dd) 1962a2afb611SJerry Jelinek { 1963a2afb611SJerry Jelinek dmu_object_info_t doi; 1964a2afb611SJerry Jelinek 1965a2afb611SJerry Jelinek dmu_object_info_from_db(dd->dd_dbuf, &doi); 1966a2afb611SJerry Jelinek return (doi.doi_type == DMU_OTN_ZAP_METADATA); 1967a2afb611SJerry Jelinek } 1968