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. 26bc9014e6SJustin Gibbs * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved. 2703b1c297SAlexander Eremin * Copyright 2015 Nexenta Systems, Inc. All rights reserved. 28fa9e4066Sahrens */ 29fa9e4066Sahrens 30fa9e4066Sahrens #include <sys/dmu.h> 31a9799022Sck #include <sys/dmu_objset.h> 32fa9e4066Sahrens #include <sys/dmu_tx.h> 33fa9e4066Sahrens #include <sys/dsl_dataset.h> 34fa9e4066Sahrens #include <sys/dsl_dir.h> 35fa9e4066Sahrens #include <sys/dsl_prop.h> 361d452cf5Sahrens #include <sys/dsl_synctask.h> 37ecd6cf80Smarks #include <sys/dsl_deleg.h> 382acef22dSMatthew Ahrens #include <sys/dmu_impl.h> 39fa9e4066Sahrens #include <sys/spa.h> 40b24ab676SJeff Bonwick #include <sys/metaslab.h> 41fa9e4066Sahrens #include <sys/zap.h> 42fa9e4066Sahrens #include <sys/zio.h> 43fa9e4066Sahrens #include <sys/arc.h> 44ecd6cf80Smarks #include <sys/sunddi.h> 45a2afb611SJerry Jelinek #include <sys/zfeature.h> 46a2afb611SJerry Jelinek #include <sys/policy.h> 47a2afb611SJerry Jelinek #include <sys/zfs_znode.h> 48fa9e4066Sahrens #include "zfs_namecheck.h" 49a2afb611SJerry Jelinek #include "zfs_prop.h" 50a2afb611SJerry Jelinek 51a2afb611SJerry Jelinek /* 52a2afb611SJerry Jelinek * Filesystem and Snapshot Limits 53a2afb611SJerry Jelinek * ------------------------------ 54a2afb611SJerry Jelinek * 55a2afb611SJerry Jelinek * These limits are used to restrict the number of filesystems and/or snapshots 56a2afb611SJerry Jelinek * that can be created at a given level in the tree or below. A typical 57a2afb611SJerry Jelinek * use-case is with a delegated dataset where the administrator wants to ensure 58a2afb611SJerry Jelinek * that a user within the zone is not creating too many additional filesystems 59a2afb611SJerry Jelinek * or snapshots, even though they're not exceeding their space quota. 60a2afb611SJerry Jelinek * 61a2afb611SJerry Jelinek * The filesystem and snapshot counts are stored as extensible properties. This 62a2afb611SJerry Jelinek * capability is controlled by a feature flag and must be enabled to be used. 63a2afb611SJerry Jelinek * Once enabled, the feature is not active until the first limit is set. At 64a2afb611SJerry Jelinek * that point, future operations to create/destroy filesystems or snapshots 65a2afb611SJerry Jelinek * will validate and update the counts. 66a2afb611SJerry Jelinek * 67a2afb611SJerry Jelinek * Because the count properties will not exist before the feature is active, 68a2afb611SJerry Jelinek * the counts are updated when a limit is first set on an uninitialized 69a2afb611SJerry Jelinek * dsl_dir node in the tree (The filesystem/snapshot count on a node includes 70a2afb611SJerry Jelinek * all of the nested filesystems/snapshots. Thus, a new leaf node has a 71a2afb611SJerry Jelinek * filesystem count of 0 and a snapshot count of 0. Non-existent filesystem and 72a2afb611SJerry Jelinek * snapshot count properties on a node indicate uninitialized counts on that 73a2afb611SJerry Jelinek * node.) When first setting a limit on an uninitialized node, the code starts 74a2afb611SJerry Jelinek * at the filesystem with the new limit and descends into all sub-filesystems 75a2afb611SJerry Jelinek * to add the count properties. 76a2afb611SJerry Jelinek * 77a2afb611SJerry Jelinek * In practice this is lightweight since a limit is typically set when the 78a2afb611SJerry Jelinek * filesystem is created and thus has no children. Once valid, changing the 79a2afb611SJerry Jelinek * limit value won't require a re-traversal since the counts are already valid. 80a2afb611SJerry Jelinek * When recursively fixing the counts, if a node with a limit is encountered 81a2afb611SJerry Jelinek * during the descent, the counts are known to be valid and there is no need to 82a2afb611SJerry Jelinek * descend into that filesystem's children. The counts on filesystems above the 83a2afb611SJerry Jelinek * one with the new limit will still be uninitialized, unless a limit is 84a2afb611SJerry Jelinek * eventually set on one of those filesystems. The counts are always recursively 85a2afb611SJerry Jelinek * updated when a limit is set on a dataset, unless there is already a limit. 86a2afb611SJerry Jelinek * When a new limit value is set on a filesystem with an existing limit, it is 87a2afb611SJerry Jelinek * possible for the new limit to be less than the current count at that level 88a2afb611SJerry Jelinek * since a user who can change the limit is also allowed to exceed the limit. 89a2afb611SJerry Jelinek * 90a2afb611SJerry Jelinek * Once the feature is active, then whenever a filesystem or snapshot is 91a2afb611SJerry Jelinek * created, the code recurses up the tree, validating the new count against the 92a2afb611SJerry Jelinek * limit at each initialized level. In practice, most levels will not have a 93a2afb611SJerry Jelinek * limit set. If there is a limit at any initialized level up the tree, the 94a2afb611SJerry Jelinek * check must pass or the creation will fail. Likewise, when a filesystem or 95a2afb611SJerry Jelinek * snapshot is destroyed, the counts are recursively adjusted all the way up 96a2afb611SJerry Jelinek * the initizized nodes in the tree. Renaming a filesystem into different point 97a2afb611SJerry Jelinek * in the tree will first validate, then update the counts on each branch up to 98a2afb611SJerry Jelinek * the common ancestor. A receive will also validate the counts and then update 99a2afb611SJerry Jelinek * them. 100a2afb611SJerry Jelinek * 101a2afb611SJerry Jelinek * An exception to the above behavior is that the limit is not enforced if the 102a2afb611SJerry Jelinek * user has permission to modify the limit. This is primarily so that 103a2afb611SJerry Jelinek * recursive snapshots in the global zone always work. We want to prevent a 104a2afb611SJerry Jelinek * denial-of-service in which a lower level delegated dataset could max out its 105a2afb611SJerry Jelinek * limit and thus block recursive snapshots from being taken in the global zone. 106a2afb611SJerry Jelinek * Because of this, it is possible for the snapshot count to be over the limit 107a2afb611SJerry Jelinek * and snapshots taken in the global zone could cause a lower level dataset to 108a2afb611SJerry Jelinek * hit or exceed its limit. The administrator taking the global zone recursive 109a2afb611SJerry Jelinek * snapshot should be aware of this side-effect and behave accordingly. 110a2afb611SJerry Jelinek * For consistency, the filesystem limit is also not enforced if the user can 111a2afb611SJerry Jelinek * modify the limit. 112a2afb611SJerry Jelinek * 113a2afb611SJerry Jelinek * The filesystem and snapshot limits are validated by dsl_fs_ss_limit_check() 114a2afb611SJerry Jelinek * and updated by dsl_fs_ss_count_adjust(). A new limit value is setup in 115a2afb611SJerry Jelinek * dsl_dir_activate_fs_ss_limit() and the counts are adjusted, if necessary, by 116a2afb611SJerry Jelinek * dsl_dir_init_fs_ss_count(). 117a2afb611SJerry Jelinek * 118a2afb611SJerry Jelinek * There is a special case when we receive a filesystem that already exists. In 119a2afb611SJerry Jelinek * this case a temporary clone name of %X is created (see dmu_recv_begin). We 120a2afb611SJerry Jelinek * never update the filesystem counts for temporary clones. 121a2afb611SJerry Jelinek * 122a2afb611SJerry Jelinek * Likewise, we do not update the snapshot counts for temporary snapshots, 123a2afb611SJerry Jelinek * such as those created by zfs diff. 124a2afb611SJerry Jelinek */ 125fa9e4066Sahrens 126c1379625SJustin T. Gibbs extern inline dsl_dir_phys_t *dsl_dir_phys(dsl_dir_t *dd); 127c1379625SJustin T. Gibbs 128a9799022Sck static uint64_t dsl_dir_space_towrite(dsl_dir_t *dd); 129fa9e4066Sahrens 130fa9e4066Sahrens static void 131bc9014e6SJustin Gibbs dsl_dir_evict(void *dbu) 132fa9e4066Sahrens { 133bc9014e6SJustin Gibbs dsl_dir_t *dd = dbu; 134fa9e4066Sahrens dsl_pool_t *dp = dd->dd_pool; 135fa9e4066Sahrens int t; 136fa9e4066Sahrens 137bc9014e6SJustin Gibbs dd->dd_dbuf = NULL; 138bc9014e6SJustin Gibbs 139fa9e4066Sahrens for (t = 0; t < TXG_SIZE; t++) { 140fa9e4066Sahrens ASSERT(!txg_list_member(&dp->dp_dirty_dirs, dd, t)); 141fa9e4066Sahrens ASSERT(dd->dd_tempreserved[t] == 0); 142fa9e4066Sahrens ASSERT(dd->dd_space_towrite[t] == 0); 143fa9e4066Sahrens } 144fa9e4066Sahrens 145fa9e4066Sahrens if (dd->dd_parent) 146bc9014e6SJustin Gibbs dsl_dir_async_rele(dd->dd_parent, dd); 147fa9e4066Sahrens 148bc9014e6SJustin Gibbs spa_async_close(dd->dd_pool->dp_spa, dd); 149fa9e4066Sahrens 15003bad06fSJustin Gibbs dsl_prop_fini(dd); 1515ad82045Snd mutex_destroy(&dd->dd_lock); 152fa9e4066Sahrens kmem_free(dd, sizeof (dsl_dir_t)); 153fa9e4066Sahrens } 154fa9e4066Sahrens 155ea8dc4b6Seschrock int 1563b2aab18SMatthew Ahrens dsl_dir_hold_obj(dsl_pool_t *dp, uint64_t ddobj, 157ea8dc4b6Seschrock const char *tail, void *tag, dsl_dir_t **ddp) 158fa9e4066Sahrens { 159fa9e4066Sahrens dmu_buf_t *dbuf; 160fa9e4066Sahrens dsl_dir_t *dd; 161ea8dc4b6Seschrock int err; 162fa9e4066Sahrens 1633b2aab18SMatthew Ahrens ASSERT(dsl_pool_config_held(dp)); 164fa9e4066Sahrens 165ea8dc4b6Seschrock err = dmu_bonus_hold(dp->dp_meta_objset, ddobj, tag, &dbuf); 1663b2aab18SMatthew Ahrens if (err != 0) 167ea8dc4b6Seschrock return (err); 168fa9e4066Sahrens dd = dmu_buf_get_user(dbuf); 169fa9e4066Sahrens #ifdef ZFS_DEBUG 170fa9e4066Sahrens { 171fa9e4066Sahrens dmu_object_info_t doi; 172fa9e4066Sahrens dmu_object_info_from_db(dbuf, &doi); 1732acef22dSMatthew Ahrens ASSERT3U(doi.doi_bonus_type, ==, DMU_OT_DSL_DIR); 17474e7dc98SMatthew Ahrens ASSERT3U(doi.doi_bonus_size, >=, sizeof (dsl_dir_phys_t)); 175fa9e4066Sahrens } 176fa9e4066Sahrens #endif 177fa9e4066Sahrens if (dd == NULL) { 178fa9e4066Sahrens dsl_dir_t *winner; 179fa9e4066Sahrens 180fa9e4066Sahrens dd = kmem_zalloc(sizeof (dsl_dir_t), KM_SLEEP); 181fa9e4066Sahrens dd->dd_object = ddobj; 182fa9e4066Sahrens dd->dd_dbuf = dbuf; 183fa9e4066Sahrens dd->dd_pool = dp; 1845ad82045Snd mutex_init(&dd->dd_lock, NULL, MUTEX_DEFAULT, NULL); 18503bad06fSJustin Gibbs dsl_prop_init(dd); 186fa9e4066Sahrens 18771eb0538SChris Kirby dsl_dir_snap_cmtime_update(dd); 18871eb0538SChris Kirby 189c1379625SJustin T. Gibbs if (dsl_dir_phys(dd)->dd_parent_obj) { 190c1379625SJustin T. Gibbs err = dsl_dir_hold_obj(dp, 191c1379625SJustin T. Gibbs dsl_dir_phys(dd)->dd_parent_obj, NULL, dd, 192c1379625SJustin T. Gibbs &dd->dd_parent); 1933b2aab18SMatthew Ahrens if (err != 0) 19474e7dc98SMatthew Ahrens goto errout; 195fa9e4066Sahrens if (tail) { 196fa9e4066Sahrens #ifdef ZFS_DEBUG 197fa9e4066Sahrens uint64_t foundobj; 198fa9e4066Sahrens 199fa9e4066Sahrens err = zap_lookup(dp->dp_meta_objset, 200c1379625SJustin T. Gibbs dsl_dir_phys(dd->dd_parent)-> 201c1379625SJustin T. Gibbs dd_child_dir_zapobj, tail, 202c1379625SJustin T. Gibbs sizeof (foundobj), 1, &foundobj); 203ea8dc4b6Seschrock ASSERT(err || foundobj == ddobj); 204fa9e4066Sahrens #endif 205fa9e4066Sahrens (void) strcpy(dd->dd_myname, tail); 206fa9e4066Sahrens } else { 207fa9e4066Sahrens err = zap_value_search(dp->dp_meta_objset, 208c1379625SJustin T. Gibbs dsl_dir_phys(dd->dd_parent)-> 209c1379625SJustin T. Gibbs dd_child_dir_zapobj, 210e7437265Sahrens ddobj, 0, dd->dd_myname); 211ea8dc4b6Seschrock } 2123b2aab18SMatthew Ahrens if (err != 0) 21374e7dc98SMatthew Ahrens goto errout; 214fa9e4066Sahrens } else { 215fa9e4066Sahrens (void) strcpy(dd->dd_myname, spa_name(dp->dp_spa)); 216fa9e4066Sahrens } 217fa9e4066Sahrens 2183f9d6ad7SLin Ling if (dsl_dir_is_clone(dd)) { 2193f9d6ad7SLin Ling dmu_buf_t *origin_bonus; 2203f9d6ad7SLin Ling dsl_dataset_phys_t *origin_phys; 2213f9d6ad7SLin Ling 2223f9d6ad7SLin Ling /* 2233f9d6ad7SLin Ling * We can't open the origin dataset, because 2243f9d6ad7SLin Ling * that would require opening this dsl_dir. 2253f9d6ad7SLin Ling * Just look at its phys directly instead. 2263f9d6ad7SLin Ling */ 2273f9d6ad7SLin Ling err = dmu_bonus_hold(dp->dp_meta_objset, 228c1379625SJustin T. Gibbs dsl_dir_phys(dd)->dd_origin_obj, FTAG, 229c1379625SJustin T. Gibbs &origin_bonus); 2303b2aab18SMatthew Ahrens if (err != 0) 2313f9d6ad7SLin Ling goto errout; 2323f9d6ad7SLin Ling origin_phys = origin_bonus->db_data; 2333f9d6ad7SLin Ling dd->dd_origin_txg = 2343f9d6ad7SLin Ling origin_phys->ds_creation_txg; 2353f9d6ad7SLin Ling dmu_buf_rele(origin_bonus, FTAG); 2363f9d6ad7SLin Ling } 2373f9d6ad7SLin Ling 238bc9014e6SJustin Gibbs dmu_buf_init_user(&dd->dd_dbu, dsl_dir_evict, &dd->dd_dbuf); 239bc9014e6SJustin Gibbs winner = dmu_buf_set_user_ie(dbuf, &dd->dd_dbu); 240bc9014e6SJustin Gibbs if (winner != NULL) { 241fa9e4066Sahrens if (dd->dd_parent) 2423b2aab18SMatthew Ahrens dsl_dir_rele(dd->dd_parent, dd); 24303bad06fSJustin Gibbs dsl_prop_fini(dd); 2445ad82045Snd mutex_destroy(&dd->dd_lock); 245fa9e4066Sahrens kmem_free(dd, sizeof (dsl_dir_t)); 246fa9e4066Sahrens dd = winner; 247fa9e4066Sahrens } else { 248fa9e4066Sahrens spa_open_ref(dp->dp_spa, dd); 249fa9e4066Sahrens } 250fa9e4066Sahrens } 251fa9e4066Sahrens 252fa9e4066Sahrens /* 253fa9e4066Sahrens * The dsl_dir_t has both open-to-close and instantiate-to-evict 254fa9e4066Sahrens * holds on the spa. We need the open-to-close holds because 255fa9e4066Sahrens * otherwise the spa_refcnt wouldn't change when we open a 256fa9e4066Sahrens * dir which the spa also has open, so we could incorrectly 257fa9e4066Sahrens * think it was OK to unload/export/destroy the pool. We need 258fa9e4066Sahrens * the instantiate-to-evict hold because the dsl_dir_t has a 259fa9e4066Sahrens * pointer to the dd_pool, which has a pointer to the spa_t. 260fa9e4066Sahrens */ 261fa9e4066Sahrens spa_open_ref(dp->dp_spa, tag); 262fa9e4066Sahrens ASSERT3P(dd->dd_pool, ==, dp); 263fa9e4066Sahrens ASSERT3U(dd->dd_object, ==, ddobj); 264fa9e4066Sahrens ASSERT3P(dd->dd_dbuf, ==, dbuf); 265ea8dc4b6Seschrock *ddp = dd; 266ea8dc4b6Seschrock return (0); 26774e7dc98SMatthew Ahrens 26874e7dc98SMatthew Ahrens errout: 26974e7dc98SMatthew Ahrens if (dd->dd_parent) 2703b2aab18SMatthew Ahrens dsl_dir_rele(dd->dd_parent, dd); 27103bad06fSJustin Gibbs dsl_prop_fini(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 286bc9014e6SJustin Gibbs /* 287bc9014e6SJustin Gibbs * Remove a reference to the given dsl dir that is being asynchronously 288bc9014e6SJustin Gibbs * released. Async releases occur from a taskq performing eviction of 289bc9014e6SJustin Gibbs * dsl datasets and dirs. This process is identical to a normal release 290bc9014e6SJustin Gibbs * with the exception of using the async API for releasing the reference on 291bc9014e6SJustin Gibbs * the spa. 292bc9014e6SJustin Gibbs */ 293bc9014e6SJustin Gibbs void 294bc9014e6SJustin Gibbs dsl_dir_async_rele(dsl_dir_t *dd, void *tag) 295bc9014e6SJustin Gibbs { 296bc9014e6SJustin Gibbs dprintf_dd(dd, "%s\n", ""); 297bc9014e6SJustin Gibbs spa_async_close(dd->dd_pool->dp_spa, tag); 298bc9014e6SJustin Gibbs dmu_buf_rele(dd->dd_dbuf, tag); 299bc9014e6SJustin Gibbs } 300bc9014e6SJustin Gibbs 301*9adfa60dSMatthew Ahrens /* buf must be at least ZFS_MAX_DATASET_NAME_LEN bytes */ 302fa9e4066Sahrens void 303fa9e4066Sahrens dsl_dir_name(dsl_dir_t *dd, char *buf) 304fa9e4066Sahrens { 305fa9e4066Sahrens if (dd->dd_parent) { 306fa9e4066Sahrens dsl_dir_name(dd->dd_parent, buf); 307*9adfa60dSMatthew Ahrens VERIFY3U(strlcat(buf, "/", ZFS_MAX_DATASET_NAME_LEN), <, 308*9adfa60dSMatthew Ahrens ZFS_MAX_DATASET_NAME_LEN); 309fa9e4066Sahrens } else { 310fa9e4066Sahrens buf[0] = '\0'; 311fa9e4066Sahrens } 312fa9e4066Sahrens if (!MUTEX_HELD(&dd->dd_lock)) { 313fa9e4066Sahrens /* 314fa9e4066Sahrens * recursive mutex so that we can use 315fa9e4066Sahrens * dprintf_dd() with dd_lock held 316fa9e4066Sahrens */ 317fa9e4066Sahrens mutex_enter(&dd->dd_lock); 318*9adfa60dSMatthew Ahrens VERIFY3U(strlcat(buf, dd->dd_myname, ZFS_MAX_DATASET_NAME_LEN), 319*9adfa60dSMatthew Ahrens <, ZFS_MAX_DATASET_NAME_LEN); 320fa9e4066Sahrens mutex_exit(&dd->dd_lock); 321fa9e4066Sahrens } else { 322*9adfa60dSMatthew Ahrens VERIFY3U(strlcat(buf, dd->dd_myname, ZFS_MAX_DATASET_NAME_LEN), 323*9adfa60dSMatthew Ahrens <, ZFS_MAX_DATASET_NAME_LEN); 324fa9e4066Sahrens } 325fa9e4066Sahrens } 326fa9e4066Sahrens 327ce636f8bSMatthew Ahrens /* Calculate name length, avoiding all the strcat calls of dsl_dir_name */ 328b7661cccSmmusante int 329b7661cccSmmusante dsl_dir_namelen(dsl_dir_t *dd) 330b7661cccSmmusante { 331b7661cccSmmusante int result = 0; 332b7661cccSmmusante 333b7661cccSmmusante if (dd->dd_parent) { 334b7661cccSmmusante /* parent's name + 1 for the "/" */ 335b7661cccSmmusante result = dsl_dir_namelen(dd->dd_parent) + 1; 336b7661cccSmmusante } 337b7661cccSmmusante 338b7661cccSmmusante if (!MUTEX_HELD(&dd->dd_lock)) { 339b7661cccSmmusante /* see dsl_dir_name */ 340b7661cccSmmusante mutex_enter(&dd->dd_lock); 341b7661cccSmmusante result += strlen(dd->dd_myname); 342b7661cccSmmusante mutex_exit(&dd->dd_lock); 343b7661cccSmmusante } else { 344b7661cccSmmusante result += strlen(dd->dd_myname); 345b7661cccSmmusante } 346b7661cccSmmusante 347b7661cccSmmusante return (result); 348b7661cccSmmusante } 349b7661cccSmmusante 350fa9e4066Sahrens static int 351fa9e4066Sahrens getcomponent(const char *path, char *component, const char **nextp) 352fa9e4066Sahrens { 353fa9e4066Sahrens char *p; 3543b2aab18SMatthew Ahrens 355ccba0801SRich Morris if ((path == NULL) || (path[0] == '\0')) 356be6fd75aSMatthew Ahrens return (SET_ERROR(ENOENT)); 357fa9e4066Sahrens /* This would be a good place to reserve some namespace... */ 358fa9e4066Sahrens p = strpbrk(path, "/@"); 359fa9e4066Sahrens if (p && (p[1] == '/' || p[1] == '@')) { 360fa9e4066Sahrens /* two separators in a row */ 361be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 362fa9e4066Sahrens } 363fa9e4066Sahrens if (p == NULL || p == path) { 364fa9e4066Sahrens /* 365fa9e4066Sahrens * if the first thing is an @ or /, it had better be an 366fa9e4066Sahrens * @ and it had better not have any more ats or slashes, 367fa9e4066Sahrens * and it had better have something after the @. 368fa9e4066Sahrens */ 369fa9e4066Sahrens if (p != NULL && 370fa9e4066Sahrens (p[0] != '@' || strpbrk(path+1, "/@") || p[1] == '\0')) 371be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 372*9adfa60dSMatthew Ahrens if (strlen(path) >= ZFS_MAX_DATASET_NAME_LEN) 373be6fd75aSMatthew Ahrens return (SET_ERROR(ENAMETOOLONG)); 374fa9e4066Sahrens (void) strcpy(component, path); 375fa9e4066Sahrens p = NULL; 376fa9e4066Sahrens } else if (p[0] == '/') { 377*9adfa60dSMatthew Ahrens if (p - path >= ZFS_MAX_DATASET_NAME_LEN) 378be6fd75aSMatthew Ahrens return (SET_ERROR(ENAMETOOLONG)); 379fa9e4066Sahrens (void) strncpy(component, path, p - path); 3803b2aab18SMatthew Ahrens component[p - path] = '\0'; 381fa9e4066Sahrens p++; 382fa9e4066Sahrens } else if (p[0] == '@') { 383fa9e4066Sahrens /* 384fa9e4066Sahrens * if the next separator is an @, there better not be 385fa9e4066Sahrens * any more slashes. 386fa9e4066Sahrens */ 387fa9e4066Sahrens if (strchr(path, '/')) 388be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 389*9adfa60dSMatthew Ahrens if (p - path >= ZFS_MAX_DATASET_NAME_LEN) 390be6fd75aSMatthew Ahrens return (SET_ERROR(ENAMETOOLONG)); 391fa9e4066Sahrens (void) strncpy(component, path, p - path); 3923b2aab18SMatthew Ahrens component[p - path] = '\0'; 393fa9e4066Sahrens } else { 3943b2aab18SMatthew Ahrens panic("invalid p=%p", (void *)p); 395fa9e4066Sahrens } 396fa9e4066Sahrens *nextp = p; 397fa9e4066Sahrens return (0); 398fa9e4066Sahrens } 399fa9e4066Sahrens 400fa9e4066Sahrens /* 4013b2aab18SMatthew Ahrens * Return the dsl_dir_t, and possibly the last component which couldn't 4023b2aab18SMatthew Ahrens * be found in *tail. The name must be in the specified dsl_pool_t. This 4033b2aab18SMatthew Ahrens * thread must hold the dp_config_rwlock for the pool. Returns NULL if the 4043b2aab18SMatthew Ahrens * path is bogus, or if tail==NULL and we couldn't parse the whole name. 4053b2aab18SMatthew Ahrens * (*tail)[0] == '@' means that the last component is a snapshot. 406fa9e4066Sahrens */ 407ea8dc4b6Seschrock int 4083b2aab18SMatthew Ahrens dsl_dir_hold(dsl_pool_t *dp, const char *name, void *tag, 409ea8dc4b6Seschrock dsl_dir_t **ddp, const char **tailp) 410fa9e4066Sahrens { 411*9adfa60dSMatthew Ahrens char buf[ZFS_MAX_DATASET_NAME_LEN]; 4123b2aab18SMatthew Ahrens const char *spaname, *next, *nextnext = NULL; 413fa9e4066Sahrens int err; 414fa9e4066Sahrens dsl_dir_t *dd; 415fa9e4066Sahrens uint64_t ddobj; 416fa9e4066Sahrens 417fa9e4066Sahrens err = getcomponent(name, buf, &next); 4183b2aab18SMatthew Ahrens if (err != 0) 419ea8dc4b6Seschrock return (err); 420fa9e4066Sahrens 4213b2aab18SMatthew Ahrens /* Make sure the name is in the specified pool. */ 4223b2aab18SMatthew Ahrens spaname = spa_name(dp->dp_spa); 4233b2aab18SMatthew Ahrens if (strcmp(buf, spaname) != 0) 42403b1c297SAlexander Eremin return (SET_ERROR(EXDEV)); 425fa9e4066Sahrens 4263b2aab18SMatthew Ahrens ASSERT(dsl_pool_config_held(dp)); 427fa9e4066Sahrens 4283b2aab18SMatthew Ahrens err = dsl_dir_hold_obj(dp, dp->dp_root_dir_obj, NULL, tag, &dd); 4293b2aab18SMatthew Ahrens if (err != 0) { 430ea8dc4b6Seschrock return (err); 431ea8dc4b6Seschrock } 432ea8dc4b6Seschrock 433fa9e4066Sahrens while (next != NULL) { 434bc9014e6SJustin Gibbs dsl_dir_t *child_dd; 435fa9e4066Sahrens err = getcomponent(next, buf, &nextnext); 4363b2aab18SMatthew Ahrens if (err != 0) 437ea8dc4b6Seschrock break; 438fa9e4066Sahrens ASSERT(next[0] != '\0'); 439fa9e4066Sahrens if (next[0] == '@') 440fa9e4066Sahrens break; 441fa9e4066Sahrens dprintf("looking up %s in obj%lld\n", 442c1379625SJustin T. Gibbs buf, dsl_dir_phys(dd)->dd_child_dir_zapobj); 443fa9e4066Sahrens 444fa9e4066Sahrens err = zap_lookup(dp->dp_meta_objset, 445c1379625SJustin T. Gibbs dsl_dir_phys(dd)->dd_child_dir_zapobj, 446fa9e4066Sahrens buf, sizeof (ddobj), 1, &ddobj); 4473b2aab18SMatthew Ahrens if (err != 0) { 448ea8dc4b6Seschrock if (err == ENOENT) 449ea8dc4b6Seschrock err = 0; 450fa9e4066Sahrens break; 451fa9e4066Sahrens } 452fa9e4066Sahrens 453bc9014e6SJustin Gibbs err = dsl_dir_hold_obj(dp, ddobj, buf, tag, &child_dd); 4543b2aab18SMatthew Ahrens if (err != 0) 455ea8dc4b6Seschrock break; 4563b2aab18SMatthew Ahrens dsl_dir_rele(dd, tag); 457bc9014e6SJustin Gibbs dd = child_dd; 458fa9e4066Sahrens next = nextnext; 459fa9e4066Sahrens } 460fa9e4066Sahrens 4613b2aab18SMatthew Ahrens if (err != 0) { 4623b2aab18SMatthew Ahrens dsl_dir_rele(dd, tag); 463ea8dc4b6Seschrock return (err); 464ea8dc4b6Seschrock } 465ea8dc4b6Seschrock 466fa9e4066Sahrens /* 467fa9e4066Sahrens * It's an error if there's more than one component left, or 468fa9e4066Sahrens * tailp==NULL and there's any component left. 469fa9e4066Sahrens */ 470fa9e4066Sahrens if (next != NULL && 471fa9e4066Sahrens (tailp == NULL || (nextnext && nextnext[0] != '\0'))) { 472fa9e4066Sahrens /* bad path name */ 4733b2aab18SMatthew Ahrens dsl_dir_rele(dd, tag); 474fa9e4066Sahrens dprintf("next=%p (%s) tail=%p\n", next, next?next:"", tailp); 475be6fd75aSMatthew Ahrens err = SET_ERROR(ENOENT); 476fa9e4066Sahrens } 4773b2aab18SMatthew Ahrens if (tailp != NULL) 478fa9e4066Sahrens *tailp = next; 479ea8dc4b6Seschrock *ddp = dd; 480ea8dc4b6Seschrock return (err); 481fa9e4066Sahrens } 482fa9e4066Sahrens 483a2afb611SJerry Jelinek /* 484a2afb611SJerry Jelinek * If the counts are already initialized for this filesystem and its 485a2afb611SJerry Jelinek * descendants then do nothing, otherwise initialize the counts. 486a2afb611SJerry Jelinek * 487a2afb611SJerry Jelinek * The counts on this filesystem, and those below, may be uninitialized due to 488a2afb611SJerry Jelinek * either the use of a pre-existing pool which did not support the 489a2afb611SJerry Jelinek * filesystem/snapshot limit feature, or one in which the feature had not yet 490a2afb611SJerry Jelinek * been enabled. 491a2afb611SJerry Jelinek * 492a2afb611SJerry Jelinek * Recursively descend the filesystem tree and update the filesystem/snapshot 493a2afb611SJerry Jelinek * counts on each filesystem below, then update the cumulative count on the 494a2afb611SJerry Jelinek * current filesystem. If the filesystem already has a count set on it, 495a2afb611SJerry Jelinek * then we know that its counts, and the counts on the filesystems below it, 496a2afb611SJerry Jelinek * are already correct, so we don't have to update this filesystem. 497a2afb611SJerry Jelinek */ 498a2afb611SJerry Jelinek static void 499a2afb611SJerry Jelinek dsl_dir_init_fs_ss_count(dsl_dir_t *dd, dmu_tx_t *tx) 500a2afb611SJerry Jelinek { 501a2afb611SJerry Jelinek uint64_t my_fs_cnt = 0; 502a2afb611SJerry Jelinek uint64_t my_ss_cnt = 0; 503a2afb611SJerry Jelinek dsl_pool_t *dp = dd->dd_pool; 504a2afb611SJerry Jelinek objset_t *os = dp->dp_meta_objset; 505a2afb611SJerry Jelinek zap_cursor_t *zc; 506a2afb611SJerry Jelinek zap_attribute_t *za; 507a2afb611SJerry Jelinek dsl_dataset_t *ds; 508a2afb611SJerry Jelinek 509adf34077SJerry Jelinek ASSERT(spa_feature_is_active(dp->dp_spa, SPA_FEATURE_FS_SS_LIMIT)); 510a2afb611SJerry Jelinek ASSERT(dsl_pool_config_held(dp)); 511a2afb611SJerry Jelinek ASSERT(dmu_tx_is_syncing(tx)); 512a2afb611SJerry Jelinek 513a2afb611SJerry Jelinek dsl_dir_zapify(dd, tx); 514a2afb611SJerry Jelinek 515a2afb611SJerry Jelinek /* 516a2afb611SJerry Jelinek * If the filesystem count has already been initialized then we 517a2afb611SJerry Jelinek * don't need to recurse down any further. 518a2afb611SJerry Jelinek */ 519a2afb611SJerry Jelinek if (zap_contains(os, dd->dd_object, DD_FIELD_FILESYSTEM_COUNT) == 0) 520a2afb611SJerry Jelinek return; 521a2afb611SJerry Jelinek 522a2afb611SJerry Jelinek zc = kmem_alloc(sizeof (zap_cursor_t), KM_SLEEP); 523a2afb611SJerry Jelinek za = kmem_alloc(sizeof (zap_attribute_t), KM_SLEEP); 524a2afb611SJerry Jelinek 525a2afb611SJerry Jelinek /* Iterate my child dirs */ 526c1379625SJustin T. Gibbs for (zap_cursor_init(zc, os, dsl_dir_phys(dd)->dd_child_dir_zapobj); 527a2afb611SJerry Jelinek zap_cursor_retrieve(zc, za) == 0; zap_cursor_advance(zc)) { 528a2afb611SJerry Jelinek dsl_dir_t *chld_dd; 529a2afb611SJerry Jelinek uint64_t count; 530a2afb611SJerry Jelinek 531a2afb611SJerry Jelinek VERIFY0(dsl_dir_hold_obj(dp, za->za_first_integer, NULL, FTAG, 532a2afb611SJerry Jelinek &chld_dd)); 533a2afb611SJerry Jelinek 534a2afb611SJerry Jelinek /* 535a2afb611SJerry Jelinek * Ignore hidden ($FREE, $MOS & $ORIGIN) objsets and 536a2afb611SJerry Jelinek * temporary datasets. 537a2afb611SJerry Jelinek */ 538a2afb611SJerry Jelinek if (chld_dd->dd_myname[0] == '$' || 539a2afb611SJerry Jelinek chld_dd->dd_myname[0] == '%') { 540a2afb611SJerry Jelinek dsl_dir_rele(chld_dd, FTAG); 541a2afb611SJerry Jelinek continue; 542a2afb611SJerry Jelinek } 543a2afb611SJerry Jelinek 544a2afb611SJerry Jelinek my_fs_cnt++; /* count this child */ 545a2afb611SJerry Jelinek 546a2afb611SJerry Jelinek dsl_dir_init_fs_ss_count(chld_dd, tx); 547a2afb611SJerry Jelinek 548a2afb611SJerry Jelinek VERIFY0(zap_lookup(os, chld_dd->dd_object, 549a2afb611SJerry Jelinek DD_FIELD_FILESYSTEM_COUNT, sizeof (count), 1, &count)); 550a2afb611SJerry Jelinek my_fs_cnt += count; 551a2afb611SJerry Jelinek VERIFY0(zap_lookup(os, chld_dd->dd_object, 552a2afb611SJerry Jelinek DD_FIELD_SNAPSHOT_COUNT, sizeof (count), 1, &count)); 553a2afb611SJerry Jelinek my_ss_cnt += count; 554a2afb611SJerry Jelinek 555a2afb611SJerry Jelinek dsl_dir_rele(chld_dd, FTAG); 556a2afb611SJerry Jelinek } 557a2afb611SJerry Jelinek zap_cursor_fini(zc); 558a2afb611SJerry Jelinek /* Count my snapshots (we counted children's snapshots above) */ 559a2afb611SJerry Jelinek VERIFY0(dsl_dataset_hold_obj(dd->dd_pool, 560c1379625SJustin T. Gibbs dsl_dir_phys(dd)->dd_head_dataset_obj, FTAG, &ds)); 561a2afb611SJerry Jelinek 562c1379625SJustin T. Gibbs for (zap_cursor_init(zc, os, dsl_dataset_phys(ds)->ds_snapnames_zapobj); 563a2afb611SJerry Jelinek zap_cursor_retrieve(zc, za) == 0; 564a2afb611SJerry Jelinek zap_cursor_advance(zc)) { 565a2afb611SJerry Jelinek /* Don't count temporary snapshots */ 566a2afb611SJerry Jelinek if (za->za_name[0] != '%') 567a2afb611SJerry Jelinek my_ss_cnt++; 568a2afb611SJerry Jelinek } 569fb7001f1SAlex Reece zap_cursor_fini(zc); 570a2afb611SJerry Jelinek 571a2afb611SJerry Jelinek dsl_dataset_rele(ds, FTAG); 572a2afb611SJerry Jelinek 573a2afb611SJerry Jelinek kmem_free(zc, sizeof (zap_cursor_t)); 574a2afb611SJerry Jelinek kmem_free(za, sizeof (zap_attribute_t)); 575a2afb611SJerry Jelinek 576a2afb611SJerry Jelinek /* we're in a sync task, update counts */ 577a2afb611SJerry Jelinek dmu_buf_will_dirty(dd->dd_dbuf, tx); 578a2afb611SJerry Jelinek VERIFY0(zap_add(os, dd->dd_object, DD_FIELD_FILESYSTEM_COUNT, 579a2afb611SJerry Jelinek sizeof (my_fs_cnt), 1, &my_fs_cnt, tx)); 580a2afb611SJerry Jelinek VERIFY0(zap_add(os, dd->dd_object, DD_FIELD_SNAPSHOT_COUNT, 581a2afb611SJerry Jelinek sizeof (my_ss_cnt), 1, &my_ss_cnt, tx)); 582a2afb611SJerry Jelinek } 583a2afb611SJerry Jelinek 584a2afb611SJerry Jelinek static int 585a2afb611SJerry Jelinek dsl_dir_actv_fs_ss_limit_check(void *arg, dmu_tx_t *tx) 586a2afb611SJerry Jelinek { 587a2afb611SJerry Jelinek char *ddname = (char *)arg; 588a2afb611SJerry Jelinek dsl_pool_t *dp = dmu_tx_pool(tx); 589a2afb611SJerry Jelinek dsl_dataset_t *ds; 590a2afb611SJerry Jelinek dsl_dir_t *dd; 591a2afb611SJerry Jelinek int error; 592a2afb611SJerry Jelinek 593a2afb611SJerry Jelinek error = dsl_dataset_hold(dp, ddname, FTAG, &ds); 594a2afb611SJerry Jelinek if (error != 0) 595a2afb611SJerry Jelinek return (error); 596a2afb611SJerry Jelinek 597a2afb611SJerry Jelinek if (!spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_FS_SS_LIMIT)) { 598a2afb611SJerry Jelinek dsl_dataset_rele(ds, FTAG); 599a2afb611SJerry Jelinek return (SET_ERROR(ENOTSUP)); 600a2afb611SJerry Jelinek } 601a2afb611SJerry Jelinek 602a2afb611SJerry Jelinek dd = ds->ds_dir; 603a2afb611SJerry Jelinek if (spa_feature_is_active(dp->dp_spa, SPA_FEATURE_FS_SS_LIMIT) && 604a2afb611SJerry Jelinek dsl_dir_is_zapified(dd) && 605a2afb611SJerry Jelinek zap_contains(dp->dp_meta_objset, dd->dd_object, 606a2afb611SJerry Jelinek DD_FIELD_FILESYSTEM_COUNT) == 0) { 607a2afb611SJerry Jelinek dsl_dataset_rele(ds, FTAG); 608a2afb611SJerry Jelinek return (SET_ERROR(EALREADY)); 609a2afb611SJerry Jelinek } 610a2afb611SJerry Jelinek 611a2afb611SJerry Jelinek dsl_dataset_rele(ds, FTAG); 612a2afb611SJerry Jelinek return (0); 613a2afb611SJerry Jelinek } 614a2afb611SJerry Jelinek 615a2afb611SJerry Jelinek static void 616a2afb611SJerry Jelinek dsl_dir_actv_fs_ss_limit_sync(void *arg, dmu_tx_t *tx) 617a2afb611SJerry Jelinek { 618a2afb611SJerry Jelinek char *ddname = (char *)arg; 619a2afb611SJerry Jelinek dsl_pool_t *dp = dmu_tx_pool(tx); 620a2afb611SJerry Jelinek dsl_dataset_t *ds; 621a2afb611SJerry Jelinek spa_t *spa; 622a2afb611SJerry Jelinek 623a2afb611SJerry Jelinek VERIFY0(dsl_dataset_hold(dp, ddname, FTAG, &ds)); 624a2afb611SJerry Jelinek 625a2afb611SJerry Jelinek spa = dsl_dataset_get_spa(ds); 626a2afb611SJerry Jelinek 627a2afb611SJerry Jelinek if (!spa_feature_is_active(spa, SPA_FEATURE_FS_SS_LIMIT)) { 628a2afb611SJerry Jelinek /* 629a2afb611SJerry Jelinek * Since the feature was not active and we're now setting a 630a2afb611SJerry Jelinek * limit, increment the feature-active counter so that the 631a2afb611SJerry Jelinek * feature becomes active for the first time. 632a2afb611SJerry Jelinek * 633a2afb611SJerry Jelinek * We are already in a sync task so we can update the MOS. 634a2afb611SJerry Jelinek */ 635a2afb611SJerry Jelinek spa_feature_incr(spa, SPA_FEATURE_FS_SS_LIMIT, tx); 636a2afb611SJerry Jelinek } 637a2afb611SJerry Jelinek 638a2afb611SJerry Jelinek /* 639a2afb611SJerry Jelinek * Since we are now setting a non-UINT64_MAX limit on the filesystem, 640a2afb611SJerry Jelinek * we need to ensure the counts are correct. Descend down the tree from 641a2afb611SJerry Jelinek * this point and update all of the counts to be accurate. 642a2afb611SJerry Jelinek */ 643a2afb611SJerry Jelinek dsl_dir_init_fs_ss_count(ds->ds_dir, tx); 644a2afb611SJerry Jelinek 645a2afb611SJerry Jelinek dsl_dataset_rele(ds, FTAG); 646a2afb611SJerry Jelinek } 647a2afb611SJerry Jelinek 648a2afb611SJerry Jelinek /* 649a2afb611SJerry Jelinek * Make sure the feature is enabled and activate it if necessary. 650a2afb611SJerry Jelinek * Since we're setting a limit, ensure the on-disk counts are valid. 651a2afb611SJerry Jelinek * This is only called by the ioctl path when setting a limit value. 652a2afb611SJerry Jelinek * 653a2afb611SJerry Jelinek * We do not need to validate the new limit, since users who can change the 654a2afb611SJerry Jelinek * limit are also allowed to exceed the limit. 655a2afb611SJerry Jelinek */ 656a2afb611SJerry Jelinek int 657a2afb611SJerry Jelinek dsl_dir_activate_fs_ss_limit(const char *ddname) 658a2afb611SJerry Jelinek { 659a2afb611SJerry Jelinek int error; 660a2afb611SJerry Jelinek 661a2afb611SJerry Jelinek error = dsl_sync_task(ddname, dsl_dir_actv_fs_ss_limit_check, 6627d46dc6cSMatthew Ahrens dsl_dir_actv_fs_ss_limit_sync, (void *)ddname, 0, 6637d46dc6cSMatthew Ahrens ZFS_SPACE_CHECK_RESERVED); 664a2afb611SJerry Jelinek 665a2afb611SJerry Jelinek if (error == EALREADY) 666a2afb611SJerry Jelinek error = 0; 667a2afb611SJerry Jelinek 668a2afb611SJerry Jelinek return (error); 669a2afb611SJerry Jelinek } 670a2afb611SJerry Jelinek 671a2afb611SJerry Jelinek /* 672a2afb611SJerry Jelinek * Used to determine if the filesystem_limit or snapshot_limit should be 673a2afb611SJerry Jelinek * enforced. We allow the limit to be exceeded if the user has permission to 674a2afb611SJerry Jelinek * write the property value. We pass in the creds that we got in the open 675a2afb611SJerry Jelinek * context since we will always be the GZ root in syncing context. We also have 676a2afb611SJerry Jelinek * to handle the case where we are allowed to change the limit on the current 677a2afb611SJerry Jelinek * dataset, but there may be another limit in the tree above. 678a2afb611SJerry Jelinek * 679a2afb611SJerry Jelinek * We can never modify these two properties within a non-global zone. In 680a2afb611SJerry Jelinek * addition, the other checks are modeled on zfs_secpolicy_write_perms. We 681a2afb611SJerry Jelinek * can't use that function since we are already holding the dp_config_rwlock. 682a2afb611SJerry Jelinek * In addition, we already have the dd and dealing with snapshots is simplified 683a2afb611SJerry Jelinek * in this code. 684a2afb611SJerry Jelinek */ 685a2afb611SJerry Jelinek 686a2afb611SJerry Jelinek typedef enum { 687a2afb611SJerry Jelinek ENFORCE_ALWAYS, 688a2afb611SJerry Jelinek ENFORCE_NEVER, 689a2afb611SJerry Jelinek ENFORCE_ABOVE 690a2afb611SJerry Jelinek } enforce_res_t; 691a2afb611SJerry Jelinek 692a2afb611SJerry Jelinek static enforce_res_t 693a2afb611SJerry Jelinek dsl_enforce_ds_ss_limits(dsl_dir_t *dd, zfs_prop_t prop, cred_t *cr) 694a2afb611SJerry Jelinek { 695a2afb611SJerry Jelinek enforce_res_t enforce = ENFORCE_ALWAYS; 696a2afb611SJerry Jelinek uint64_t obj; 697a2afb611SJerry Jelinek dsl_dataset_t *ds; 698a2afb611SJerry Jelinek uint64_t zoned; 699a2afb611SJerry Jelinek 700a2afb611SJerry Jelinek ASSERT(prop == ZFS_PROP_FILESYSTEM_LIMIT || 701a2afb611SJerry Jelinek prop == ZFS_PROP_SNAPSHOT_LIMIT); 702a2afb611SJerry Jelinek 703a2afb611SJerry Jelinek #ifdef _KERNEL 704a2afb611SJerry Jelinek if (crgetzoneid(cr) != GLOBAL_ZONEID) 705a2afb611SJerry Jelinek return (ENFORCE_ALWAYS); 706a2afb611SJerry Jelinek 707a2afb611SJerry Jelinek if (secpolicy_zfs(cr) == 0) 708a2afb611SJerry Jelinek return (ENFORCE_NEVER); 709a2afb611SJerry Jelinek #endif 710a2afb611SJerry Jelinek 711c1379625SJustin T. Gibbs if ((obj = dsl_dir_phys(dd)->dd_head_dataset_obj) == 0) 712a2afb611SJerry Jelinek return (ENFORCE_ALWAYS); 713a2afb611SJerry Jelinek 714a2afb611SJerry Jelinek ASSERT(dsl_pool_config_held(dd->dd_pool)); 715a2afb611SJerry Jelinek 716a2afb611SJerry Jelinek if (dsl_dataset_hold_obj(dd->dd_pool, obj, FTAG, &ds) != 0) 717a2afb611SJerry Jelinek return (ENFORCE_ALWAYS); 718a2afb611SJerry Jelinek 719a2afb611SJerry Jelinek if (dsl_prop_get_ds(ds, "zoned", 8, 1, &zoned, NULL) || zoned) { 720a2afb611SJerry Jelinek /* Only root can access zoned fs's from the GZ */ 721a2afb611SJerry Jelinek enforce = ENFORCE_ALWAYS; 722a2afb611SJerry Jelinek } else { 723a2afb611SJerry Jelinek if (dsl_deleg_access_impl(ds, zfs_prop_to_name(prop), cr) == 0) 724a2afb611SJerry Jelinek enforce = ENFORCE_ABOVE; 725a2afb611SJerry Jelinek } 726a2afb611SJerry Jelinek 727a2afb611SJerry Jelinek dsl_dataset_rele(ds, FTAG); 728a2afb611SJerry Jelinek return (enforce); 729a2afb611SJerry Jelinek } 730a2afb611SJerry Jelinek 731a2afb611SJerry Jelinek /* 732a2afb611SJerry Jelinek * Check if adding additional child filesystem(s) would exceed any filesystem 733a2afb611SJerry Jelinek * limits or adding additional snapshot(s) would exceed any snapshot limits. 734a2afb611SJerry Jelinek * The prop argument indicates which limit to check. 735a2afb611SJerry Jelinek * 736a2afb611SJerry Jelinek * Note that all filesystem limits up to the root (or the highest 737a2afb611SJerry Jelinek * initialized) filesystem or the given ancestor must be satisfied. 738a2afb611SJerry Jelinek */ 739a2afb611SJerry Jelinek int 740a2afb611SJerry Jelinek dsl_fs_ss_limit_check(dsl_dir_t *dd, uint64_t delta, zfs_prop_t prop, 741a2afb611SJerry Jelinek dsl_dir_t *ancestor, cred_t *cr) 742a2afb611SJerry Jelinek { 743a2afb611SJerry Jelinek objset_t *os = dd->dd_pool->dp_meta_objset; 744a2afb611SJerry Jelinek uint64_t limit, count; 745a2afb611SJerry Jelinek char *count_prop; 746a2afb611SJerry Jelinek enforce_res_t enforce; 747a2afb611SJerry Jelinek int err = 0; 748a2afb611SJerry Jelinek 749a2afb611SJerry Jelinek ASSERT(dsl_pool_config_held(dd->dd_pool)); 750a2afb611SJerry Jelinek ASSERT(prop == ZFS_PROP_FILESYSTEM_LIMIT || 751a2afb611SJerry Jelinek prop == ZFS_PROP_SNAPSHOT_LIMIT); 752a2afb611SJerry Jelinek 753a2afb611SJerry Jelinek /* 754a2afb611SJerry Jelinek * If we're allowed to change the limit, don't enforce the limit 755a2afb611SJerry Jelinek * e.g. this can happen if a snapshot is taken by an administrative 756a2afb611SJerry Jelinek * user in the global zone (i.e. a recursive snapshot by root). 757a2afb611SJerry Jelinek * However, we must handle the case of delegated permissions where we 758a2afb611SJerry Jelinek * are allowed to change the limit on the current dataset, but there 759a2afb611SJerry Jelinek * is another limit in the tree above. 760a2afb611SJerry Jelinek */ 761a2afb611SJerry Jelinek enforce = dsl_enforce_ds_ss_limits(dd, prop, cr); 762a2afb611SJerry Jelinek if (enforce == ENFORCE_NEVER) 763a2afb611SJerry Jelinek return (0); 764a2afb611SJerry Jelinek 765a2afb611SJerry Jelinek /* 766a2afb611SJerry Jelinek * e.g. if renaming a dataset with no snapshots, count adjustment 767a2afb611SJerry Jelinek * is 0. 768a2afb611SJerry Jelinek */ 769a2afb611SJerry Jelinek if (delta == 0) 770a2afb611SJerry Jelinek return (0); 771a2afb611SJerry Jelinek 772a2afb611SJerry Jelinek if (prop == ZFS_PROP_SNAPSHOT_LIMIT) { 773a2afb611SJerry Jelinek /* 774a2afb611SJerry Jelinek * We don't enforce the limit for temporary snapshots. This is 775a2afb611SJerry Jelinek * indicated by a NULL cred_t argument. 776a2afb611SJerry Jelinek */ 777a2afb611SJerry Jelinek if (cr == NULL) 778a2afb611SJerry Jelinek return (0); 779a2afb611SJerry Jelinek 780a2afb611SJerry Jelinek count_prop = DD_FIELD_SNAPSHOT_COUNT; 781a2afb611SJerry Jelinek } else { 782a2afb611SJerry Jelinek count_prop = DD_FIELD_FILESYSTEM_COUNT; 783a2afb611SJerry Jelinek } 784a2afb611SJerry Jelinek 785a2afb611SJerry Jelinek /* 786a2afb611SJerry Jelinek * If an ancestor has been provided, stop checking the limit once we 787a2afb611SJerry Jelinek * hit that dir. We need this during rename so that we don't overcount 788a2afb611SJerry Jelinek * the check once we recurse up to the common ancestor. 789a2afb611SJerry Jelinek */ 790a2afb611SJerry Jelinek if (ancestor == dd) 791a2afb611SJerry Jelinek return (0); 792a2afb611SJerry Jelinek 793a2afb611SJerry Jelinek /* 794a2afb611SJerry Jelinek * If we hit an uninitialized node while recursing up the tree, we can 795a2afb611SJerry Jelinek * stop since we know there is no limit here (or above). The counts are 796a2afb611SJerry Jelinek * not valid on this node and we know we won't touch this node's counts. 797a2afb611SJerry Jelinek */ 798a2afb611SJerry Jelinek if (!dsl_dir_is_zapified(dd) || zap_lookup(os, dd->dd_object, 799a2afb611SJerry Jelinek count_prop, sizeof (count), 1, &count) == ENOENT) 800a2afb611SJerry Jelinek return (0); 801a2afb611SJerry Jelinek 802a2afb611SJerry Jelinek err = dsl_prop_get_dd(dd, zfs_prop_to_name(prop), 8, 1, &limit, NULL, 803a2afb611SJerry Jelinek B_FALSE); 804a2afb611SJerry Jelinek if (err != 0) 805a2afb611SJerry Jelinek return (err); 806a2afb611SJerry Jelinek 807a2afb611SJerry Jelinek /* Is there a limit which we've hit? */ 808a2afb611SJerry Jelinek if (enforce == ENFORCE_ALWAYS && (count + delta) > limit) 809a2afb611SJerry Jelinek return (SET_ERROR(EDQUOT)); 810a2afb611SJerry Jelinek 811a2afb611SJerry Jelinek if (dd->dd_parent != NULL) 812a2afb611SJerry Jelinek err = dsl_fs_ss_limit_check(dd->dd_parent, delta, prop, 813a2afb611SJerry Jelinek ancestor, cr); 814a2afb611SJerry Jelinek 815a2afb611SJerry Jelinek return (err); 816a2afb611SJerry Jelinek } 817a2afb611SJerry Jelinek 818a2afb611SJerry Jelinek /* 819a2afb611SJerry Jelinek * Adjust the filesystem or snapshot count for the specified dsl_dir_t and all 820a2afb611SJerry Jelinek * parents. When a new filesystem/snapshot is created, increment the count on 821a2afb611SJerry Jelinek * all parents, and when a filesystem/snapshot is destroyed, decrement the 822a2afb611SJerry Jelinek * count. 823a2afb611SJerry Jelinek */ 824a2afb611SJerry Jelinek void 825a2afb611SJerry Jelinek dsl_fs_ss_count_adjust(dsl_dir_t *dd, int64_t delta, const char *prop, 826a2afb611SJerry Jelinek dmu_tx_t *tx) 827a2afb611SJerry Jelinek { 828a2afb611SJerry Jelinek int err; 829a2afb611SJerry Jelinek objset_t *os = dd->dd_pool->dp_meta_objset; 830a2afb611SJerry Jelinek uint64_t count; 831a2afb611SJerry Jelinek 832a2afb611SJerry Jelinek ASSERT(dsl_pool_config_held(dd->dd_pool)); 833a2afb611SJerry Jelinek ASSERT(dmu_tx_is_syncing(tx)); 834a2afb611SJerry Jelinek ASSERT(strcmp(prop, DD_FIELD_FILESYSTEM_COUNT) == 0 || 835a2afb611SJerry Jelinek strcmp(prop, DD_FIELD_SNAPSHOT_COUNT) == 0); 836a2afb611SJerry Jelinek 837a2afb611SJerry Jelinek /* 838a2afb611SJerry Jelinek * When we receive an incremental stream into a filesystem that already 839a2afb611SJerry Jelinek * exists, a temporary clone is created. We don't count this temporary 840a2afb611SJerry Jelinek * clone, whose name begins with a '%'. We also ignore hidden ($FREE, 841a2afb611SJerry Jelinek * $MOS & $ORIGIN) objsets. 842a2afb611SJerry Jelinek */ 843a2afb611SJerry Jelinek if ((dd->dd_myname[0] == '%' || dd->dd_myname[0] == '$') && 844a2afb611SJerry Jelinek strcmp(prop, DD_FIELD_FILESYSTEM_COUNT) == 0) 845a2afb611SJerry Jelinek return; 846a2afb611SJerry Jelinek 847a2afb611SJerry Jelinek /* 848a2afb611SJerry Jelinek * e.g. if renaming a dataset with no snapshots, count adjustment is 0 849a2afb611SJerry Jelinek */ 850a2afb611SJerry Jelinek if (delta == 0) 851a2afb611SJerry Jelinek return; 852a2afb611SJerry Jelinek 853a2afb611SJerry Jelinek /* 854a2afb611SJerry Jelinek * If we hit an uninitialized node while recursing up the tree, we can 855a2afb611SJerry Jelinek * stop since we know the counts are not valid on this node and we 856a2afb611SJerry Jelinek * know we shouldn't touch this node's counts. An uninitialized count 857a2afb611SJerry Jelinek * on the node indicates that either the feature has not yet been 858a2afb611SJerry Jelinek * activated or there are no limits on this part of the tree. 859a2afb611SJerry Jelinek */ 860a2afb611SJerry Jelinek if (!dsl_dir_is_zapified(dd) || (err = zap_lookup(os, dd->dd_object, 861a2afb611SJerry Jelinek prop, sizeof (count), 1, &count)) == ENOENT) 862a2afb611SJerry Jelinek return; 863a2afb611SJerry Jelinek VERIFY0(err); 864a2afb611SJerry Jelinek 865a2afb611SJerry Jelinek count += delta; 866a2afb611SJerry Jelinek /* Use a signed verify to make sure we're not neg. */ 867a2afb611SJerry Jelinek VERIFY3S(count, >=, 0); 868a2afb611SJerry Jelinek 869a2afb611SJerry Jelinek VERIFY0(zap_update(os, dd->dd_object, prop, sizeof (count), 1, &count, 870a2afb611SJerry Jelinek tx)); 871a2afb611SJerry Jelinek 872a2afb611SJerry Jelinek /* Roll up this additional count into our ancestors */ 873a2afb611SJerry Jelinek if (dd->dd_parent != NULL) 874a2afb611SJerry Jelinek dsl_fs_ss_count_adjust(dd->dd_parent, delta, prop, tx); 875a2afb611SJerry Jelinek } 876a2afb611SJerry Jelinek 8771d452cf5Sahrens uint64_t 878088f3894Sahrens dsl_dir_create_sync(dsl_pool_t *dp, dsl_dir_t *pds, const char *name, 879088f3894Sahrens dmu_tx_t *tx) 880fa9e4066Sahrens { 881088f3894Sahrens objset_t *mos = dp->dp_meta_objset; 882fa9e4066Sahrens uint64_t ddobj; 883cde58dbcSMatthew Ahrens dsl_dir_phys_t *ddphys; 884fa9e4066Sahrens dmu_buf_t *dbuf; 885fa9e4066Sahrens 8861649cd4bStabriz ddobj = dmu_object_alloc(mos, DMU_OT_DSL_DIR, 0, 8871649cd4bStabriz DMU_OT_DSL_DIR, sizeof (dsl_dir_phys_t), tx); 888088f3894Sahrens if (pds) { 889c1379625SJustin T. Gibbs VERIFY(0 == zap_add(mos, dsl_dir_phys(pds)->dd_child_dir_zapobj, 890088f3894Sahrens name, sizeof (uint64_t), 1, &ddobj, tx)); 891088f3894Sahrens } else { 892088f3894Sahrens /* it's the root dir */ 893088f3894Sahrens VERIFY(0 == zap_add(mos, DMU_POOL_DIRECTORY_OBJECT, 894088f3894Sahrens DMU_POOL_ROOT_DATASET, sizeof (uint64_t), 1, &ddobj, tx)); 895088f3894Sahrens } 896ea8dc4b6Seschrock VERIFY(0 == dmu_bonus_hold(mos, ddobj, FTAG, &dbuf)); 897fa9e4066Sahrens dmu_buf_will_dirty(dbuf, tx); 898cde58dbcSMatthew Ahrens ddphys = dbuf->db_data; 899fa9e4066Sahrens 900cde58dbcSMatthew Ahrens ddphys->dd_creation_time = gethrestime_sec(); 901a2afb611SJerry Jelinek if (pds) { 902cde58dbcSMatthew Ahrens ddphys->dd_parent_obj = pds->dd_object; 903a2afb611SJerry Jelinek 904a2afb611SJerry Jelinek /* update the filesystem counts */ 905a2afb611SJerry Jelinek dsl_fs_ss_count_adjust(pds, 1, DD_FIELD_FILESYSTEM_COUNT, tx); 906a2afb611SJerry Jelinek } 907cde58dbcSMatthew Ahrens ddphys->dd_props_zapobj = zap_create(mos, 908fa9e4066Sahrens DMU_OT_DSL_PROPS, DMU_OT_NONE, 0, tx); 909cde58dbcSMatthew Ahrens ddphys->dd_child_dir_zapobj = zap_create(mos, 91087e5029aSahrens DMU_OT_DSL_DIR_CHILD_MAP, DMU_OT_NONE, 0, tx); 91174e7dc98SMatthew Ahrens if (spa_version(dp->dp_spa) >= SPA_VERSION_USED_BREAKDOWN) 912cde58dbcSMatthew Ahrens ddphys->dd_flags |= DD_FLAG_USED_BREAKDOWN; 913ea8dc4b6Seschrock dmu_buf_rele(dbuf, FTAG); 914fa9e4066Sahrens 9151d452cf5Sahrens return (ddobj); 916fa9e4066Sahrens } 917fa9e4066Sahrens 918088f3894Sahrens boolean_t 919088f3894Sahrens dsl_dir_is_clone(dsl_dir_t *dd) 920fa9e4066Sahrens { 921c1379625SJustin T. Gibbs return (dsl_dir_phys(dd)->dd_origin_obj && 922088f3894Sahrens (dd->dd_pool->dp_origin_snap == NULL || 923c1379625SJustin T. Gibbs dsl_dir_phys(dd)->dd_origin_obj != 924088f3894Sahrens dd->dd_pool->dp_origin_snap->ds_object)); 925fa9e4066Sahrens } 926fa9e4066Sahrens 927fa9e4066Sahrens void 928a2eea2e1Sahrens dsl_dir_stats(dsl_dir_t *dd, nvlist_t *nv) 929fa9e4066Sahrens { 930fa9e4066Sahrens mutex_enter(&dd->dd_lock); 93174e7dc98SMatthew Ahrens dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USED, 932c1379625SJustin T. Gibbs dsl_dir_phys(dd)->dd_used_bytes); 933c1379625SJustin T. Gibbs dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_QUOTA, 934c1379625SJustin T. Gibbs dsl_dir_phys(dd)->dd_quota); 935a2eea2e1Sahrens dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_RESERVATION, 936c1379625SJustin T. Gibbs dsl_dir_phys(dd)->dd_reserved); 937a2eea2e1Sahrens dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_COMPRESSRATIO, 938c1379625SJustin T. Gibbs dsl_dir_phys(dd)->dd_compressed_bytes == 0 ? 100 : 939c1379625SJustin T. Gibbs (dsl_dir_phys(dd)->dd_uncompressed_bytes * 100 / 940c1379625SJustin T. Gibbs dsl_dir_phys(dd)->dd_compressed_bytes)); 94177372cb0SMatthew Ahrens dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_LOGICALUSED, 942c1379625SJustin T. Gibbs dsl_dir_phys(dd)->dd_uncompressed_bytes); 943c1379625SJustin T. Gibbs if (dsl_dir_phys(dd)->dd_flags & DD_FLAG_USED_BREAKDOWN) { 94474e7dc98SMatthew Ahrens dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDSNAP, 945c1379625SJustin T. Gibbs dsl_dir_phys(dd)->dd_used_breakdown[DD_USED_SNAP]); 94674e7dc98SMatthew Ahrens dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDDS, 947c1379625SJustin T. Gibbs dsl_dir_phys(dd)->dd_used_breakdown[DD_USED_HEAD]); 94874e7dc98SMatthew Ahrens dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDREFRESERV, 949c1379625SJustin T. Gibbs dsl_dir_phys(dd)->dd_used_breakdown[DD_USED_REFRSRV]); 95074e7dc98SMatthew Ahrens dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDCHILD, 951c1379625SJustin T. Gibbs dsl_dir_phys(dd)->dd_used_breakdown[DD_USED_CHILD] + 952c1379625SJustin T. Gibbs dsl_dir_phys(dd)->dd_used_breakdown[DD_USED_CHILD_RSRV]); 95374e7dc98SMatthew Ahrens } 954fa9e4066Sahrens mutex_exit(&dd->dd_lock); 955fa9e4066Sahrens 956a2afb611SJerry Jelinek if (dsl_dir_is_zapified(dd)) { 957a2afb611SJerry Jelinek uint64_t count; 958a2afb611SJerry Jelinek objset_t *os = dd->dd_pool->dp_meta_objset; 959a2afb611SJerry Jelinek 960a2afb611SJerry Jelinek if (zap_lookup(os, dd->dd_object, DD_FIELD_FILESYSTEM_COUNT, 961a2afb611SJerry Jelinek sizeof (count), 1, &count) == 0) { 962a2afb611SJerry Jelinek dsl_prop_nvlist_add_uint64(nv, 963a2afb611SJerry Jelinek ZFS_PROP_FILESYSTEM_COUNT, count); 964a2afb611SJerry Jelinek } 965a2afb611SJerry Jelinek if (zap_lookup(os, dd->dd_object, DD_FIELD_SNAPSHOT_COUNT, 966a2afb611SJerry Jelinek sizeof (count), 1, &count) == 0) { 967a2afb611SJerry Jelinek dsl_prop_nvlist_add_uint64(nv, 968a2afb611SJerry Jelinek ZFS_PROP_SNAPSHOT_COUNT, count); 969a2afb611SJerry Jelinek } 970a2afb611SJerry Jelinek } 971a2afb611SJerry Jelinek 972088f3894Sahrens if (dsl_dir_is_clone(dd)) { 973fa9e4066Sahrens dsl_dataset_t *ds; 974*9adfa60dSMatthew Ahrens char buf[ZFS_MAX_DATASET_NAME_LEN]; 975fa9e4066Sahrens 9763b2aab18SMatthew Ahrens VERIFY0(dsl_dataset_hold_obj(dd->dd_pool, 977c1379625SJustin T. Gibbs dsl_dir_phys(dd)->dd_origin_obj, FTAG, &ds)); 978a2eea2e1Sahrens dsl_dataset_name(ds, buf); 979745cd3c5Smaybee dsl_dataset_rele(ds, FTAG); 980a2eea2e1Sahrens dsl_prop_nvlist_add_string(nv, ZFS_PROP_ORIGIN, buf); 981fa9e4066Sahrens } 982fa9e4066Sahrens } 983fa9e4066Sahrens 984fa9e4066Sahrens void 985fa9e4066Sahrens dsl_dir_dirty(dsl_dir_t *dd, dmu_tx_t *tx) 986fa9e4066Sahrens { 987fa9e4066Sahrens dsl_pool_t *dp = dd->dd_pool; 988fa9e4066Sahrens 989c1379625SJustin T. Gibbs ASSERT(dsl_dir_phys(dd)); 990fa9e4066Sahrens 9913b2aab18SMatthew Ahrens if (txg_list_add(&dp->dp_dirty_dirs, dd, tx->tx_txg)) { 992fa9e4066Sahrens /* up the hold count until we can be written out */ 993fa9e4066Sahrens dmu_buf_add_ref(dd->dd_dbuf, dd); 994fa9e4066Sahrens } 995fa9e4066Sahrens } 996fa9e4066Sahrens 997fa9e4066Sahrens static int64_t 998fa9e4066Sahrens parent_delta(dsl_dir_t *dd, uint64_t used, int64_t delta) 999fa9e4066Sahrens { 1000c1379625SJustin T. Gibbs uint64_t old_accounted = MAX(used, dsl_dir_phys(dd)->dd_reserved); 1001c1379625SJustin T. Gibbs uint64_t new_accounted = 1002c1379625SJustin T. Gibbs MAX(used + delta, dsl_dir_phys(dd)->dd_reserved); 1003fa9e4066Sahrens return (new_accounted - old_accounted); 1004fa9e4066Sahrens } 1005fa9e4066Sahrens 1006fa9e4066Sahrens void 1007fa9e4066Sahrens dsl_dir_sync(dsl_dir_t *dd, dmu_tx_t *tx) 1008fa9e4066Sahrens { 1009fa9e4066Sahrens ASSERT(dmu_tx_is_syncing(tx)); 1010fa9e4066Sahrens 1011fa9e4066Sahrens mutex_enter(&dd->dd_lock); 1012fb09f5aaSMadhav Suresh ASSERT0(dd->dd_tempreserved[tx->tx_txg&TXG_MASK]); 1013fa9e4066Sahrens dprintf_dd(dd, "txg=%llu towrite=%lluK\n", tx->tx_txg, 1014fa9e4066Sahrens dd->dd_space_towrite[tx->tx_txg&TXG_MASK] / 1024); 1015fa9e4066Sahrens dd->dd_space_towrite[tx->tx_txg&TXG_MASK] = 0; 1016fa9e4066Sahrens mutex_exit(&dd->dd_lock); 1017fa9e4066Sahrens 1018fa9e4066Sahrens /* release the hold from dsl_dir_dirty */ 1019ea8dc4b6Seschrock dmu_buf_rele(dd->dd_dbuf, dd); 1020fa9e4066Sahrens } 1021fa9e4066Sahrens 1022fa9e4066Sahrens static uint64_t 1023a9799022Sck dsl_dir_space_towrite(dsl_dir_t *dd) 1024fa9e4066Sahrens { 1025a9799022Sck uint64_t space = 0; 1026fa9e4066Sahrens int i; 1027fa9e4066Sahrens 1028fa9e4066Sahrens ASSERT(MUTEX_HELD(&dd->dd_lock)); 1029fa9e4066Sahrens 1030fa9e4066Sahrens for (i = 0; i < TXG_SIZE; i++) { 1031fa9e4066Sahrens space += dd->dd_space_towrite[i&TXG_MASK]; 1032fa9e4066Sahrens ASSERT3U(dd->dd_space_towrite[i&TXG_MASK], >=, 0); 1033fa9e4066Sahrens } 1034fa9e4066Sahrens return (space); 1035fa9e4066Sahrens } 1036fa9e4066Sahrens 1037fa9e4066Sahrens /* 1038fa9e4066Sahrens * How much space would dd have available if ancestor had delta applied 1039fa9e4066Sahrens * to it? If ondiskonly is set, we're only interested in what's 1040fa9e4066Sahrens * on-disk, not estimated pending changes. 1041fa9e4066Sahrens */ 1042a2eea2e1Sahrens uint64_t 1043fa9e4066Sahrens dsl_dir_space_available(dsl_dir_t *dd, 1044fa9e4066Sahrens dsl_dir_t *ancestor, int64_t delta, int ondiskonly) 1045fa9e4066Sahrens { 1046fa9e4066Sahrens uint64_t parentspace, myspace, quota, used; 1047fa9e4066Sahrens 1048fa9e4066Sahrens /* 1049fa9e4066Sahrens * If there are no restrictions otherwise, assume we have 1050fa9e4066Sahrens * unlimited space available. 1051fa9e4066Sahrens */ 1052fa9e4066Sahrens quota = UINT64_MAX; 1053fa9e4066Sahrens parentspace = UINT64_MAX; 1054fa9e4066Sahrens 1055fa9e4066Sahrens if (dd->dd_parent != NULL) { 1056fa9e4066Sahrens parentspace = dsl_dir_space_available(dd->dd_parent, 1057fa9e4066Sahrens ancestor, delta, ondiskonly); 1058fa9e4066Sahrens } 1059fa9e4066Sahrens 1060fa9e4066Sahrens mutex_enter(&dd->dd_lock); 1061c1379625SJustin T. Gibbs if (dsl_dir_phys(dd)->dd_quota != 0) 1062c1379625SJustin T. Gibbs quota = dsl_dir_phys(dd)->dd_quota; 1063c1379625SJustin T. Gibbs used = dsl_dir_phys(dd)->dd_used_bytes; 1064a9799022Sck if (!ondiskonly) 1065a9799022Sck used += dsl_dir_space_towrite(dd); 1066fa9e4066Sahrens 1067fa9e4066Sahrens if (dd->dd_parent == NULL) { 106899653d4eSeschrock uint64_t poolsize = dsl_pool_adjustedsize(dd->dd_pool, FALSE); 1069fa9e4066Sahrens quota = MIN(quota, poolsize); 1070fa9e4066Sahrens } 1071fa9e4066Sahrens 1072c1379625SJustin T. Gibbs if (dsl_dir_phys(dd)->dd_reserved > used && parentspace != UINT64_MAX) { 1073fa9e4066Sahrens /* 1074fa9e4066Sahrens * We have some space reserved, in addition to what our 1075fa9e4066Sahrens * parent gave us. 1076fa9e4066Sahrens */ 1077c1379625SJustin T. Gibbs parentspace += dsl_dir_phys(dd)->dd_reserved - used; 1078fa9e4066Sahrens } 1079fa9e4066Sahrens 108074e7dc98SMatthew Ahrens if (dd == ancestor) { 108174e7dc98SMatthew Ahrens ASSERT(delta <= 0); 108274e7dc98SMatthew Ahrens ASSERT(used >= -delta); 108374e7dc98SMatthew Ahrens used += delta; 108474e7dc98SMatthew Ahrens if (parentspace != UINT64_MAX) 108574e7dc98SMatthew Ahrens parentspace -= delta; 108674e7dc98SMatthew Ahrens } 108774e7dc98SMatthew Ahrens 1088fa9e4066Sahrens if (used > quota) { 1089fa9e4066Sahrens /* over quota */ 1090fa9e4066Sahrens myspace = 0; 1091fa9e4066Sahrens } else { 1092fa9e4066Sahrens /* 109399653d4eSeschrock * the lesser of the space provided by our parent and 109499653d4eSeschrock * the space left in our quota 1095fa9e4066Sahrens */ 1096fa9e4066Sahrens myspace = MIN(parentspace, quota - used); 1097fa9e4066Sahrens } 1098fa9e4066Sahrens 1099fa9e4066Sahrens mutex_exit(&dd->dd_lock); 1100fa9e4066Sahrens 1101fa9e4066Sahrens return (myspace); 1102fa9e4066Sahrens } 1103fa9e4066Sahrens 1104fa9e4066Sahrens struct tempreserve { 1105fa9e4066Sahrens list_node_t tr_node; 1106fa9e4066Sahrens dsl_dir_t *tr_ds; 1107fa9e4066Sahrens uint64_t tr_size; 1108fa9e4066Sahrens }; 1109fa9e4066Sahrens 1110fa9e4066Sahrens static int 1111a9799022Sck dsl_dir_tempreserve_impl(dsl_dir_t *dd, uint64_t asize, boolean_t netfree, 1112a9799022Sck boolean_t ignorequota, boolean_t checkrefquota, list_t *tr_list, 11136df6c3bcSck dmu_tx_t *tx, boolean_t first) 1114fa9e4066Sahrens { 1115fa9e4066Sahrens uint64_t txg = tx->tx_txg; 1116a9799022Sck uint64_t est_inflight, used_on_disk, quota, parent_rsrv; 1117468c413aSTim Haley uint64_t deferred = 0; 1118a9799022Sck struct tempreserve *tr; 1119468c413aSTim Haley int retval = EDQUOT; 1120fa9e4066Sahrens int txgidx = txg & TXG_MASK; 1121fa9e4066Sahrens int i; 11229082849eSck uint64_t ref_rsrv = 0; 1123fa9e4066Sahrens 1124fa9e4066Sahrens ASSERT3U(txg, !=, 0); 1125a9799022Sck ASSERT3S(asize, >, 0); 1126fa9e4066Sahrens 1127fa9e4066Sahrens mutex_enter(&dd->dd_lock); 1128a9799022Sck 1129fa9e4066Sahrens /* 1130fa9e4066Sahrens * Check against the dsl_dir's quota. We don't add in the delta 1131fa9e4066Sahrens * when checking for over-quota because they get one free hit. 1132fa9e4066Sahrens */ 1133a9799022Sck est_inflight = dsl_dir_space_towrite(dd); 1134fa9e4066Sahrens for (i = 0; i < TXG_SIZE; i++) 1135a9799022Sck est_inflight += dd->dd_tempreserved[i]; 1136c1379625SJustin T. Gibbs used_on_disk = dsl_dir_phys(dd)->dd_used_bytes; 1137fa9e4066Sahrens 1138f4d2e9e6Smaybee /* 11396df6c3bcSck * On the first iteration, fetch the dataset's used-on-disk and 11406df6c3bcSck * refreservation values. Also, if checkrefquota is set, test if 11416df6c3bcSck * allocating this space would exceed the dataset's refquota. 1142f4d2e9e6Smaybee */ 11436df6c3bcSck if (first && tx->tx_objset) { 1144c3fdb13aSmaybee int error; 1145503ad85cSMatthew Ahrens dsl_dataset_t *ds = tx->tx_objset->os_dsl_dataset; 11469082849eSck 1147a9799022Sck error = dsl_dataset_check_quota(ds, checkrefquota, 11489082849eSck asize, est_inflight, &used_on_disk, &ref_rsrv); 1149a9799022Sck if (error) { 1150a9799022Sck mutex_exit(&dd->dd_lock); 1151a9799022Sck return (error); 1152a9799022Sck } 1153a9799022Sck } 1154a9799022Sck 1155a9799022Sck /* 1156a9799022Sck * If this transaction will result in a net free of space, 1157a9799022Sck * we want to let it through. 1158a9799022Sck */ 1159c1379625SJustin T. Gibbs if (ignorequota || netfree || dsl_dir_phys(dd)->dd_quota == 0) 1160f4d2e9e6Smaybee quota = UINT64_MAX; 1161f4d2e9e6Smaybee else 1162c1379625SJustin T. Gibbs quota = dsl_dir_phys(dd)->dd_quota; 1163fa9e4066Sahrens 1164fa9e4066Sahrens /* 1165468c413aSTim Haley * Adjust the quota against the actual pool size at the root 1166468c413aSTim Haley * minus any outstanding deferred frees. 1167f4d2e9e6Smaybee * To ensure that it's possible to remove files from a full 1168f4d2e9e6Smaybee * pool without inducing transient overcommits, we throttle 1169fa9e4066Sahrens * netfree transactions against a quota that is slightly larger, 1170fa9e4066Sahrens * but still within the pool's allocation slop. In cases where 1171fa9e4066Sahrens * we're very close to full, this will allow a steady trickle of 1172fa9e4066Sahrens * removes to get through. 1173fa9e4066Sahrens */ 11741934e92fSmaybee if (dd->dd_parent == NULL) { 1175b24ab676SJeff Bonwick spa_t *spa = dd->dd_pool->dp_spa; 1176fa9e4066Sahrens uint64_t poolsize = dsl_pool_adjustedsize(dd->dd_pool, netfree); 1177b24ab676SJeff Bonwick deferred = metaslab_class_get_deferred(spa_normal_class(spa)); 1178468c413aSTim Haley if (poolsize - deferred < quota) { 1179468c413aSTim Haley quota = poolsize - deferred; 1180468c413aSTim Haley retval = ENOSPC; 1181fa9e4066Sahrens } 1182fa9e4066Sahrens } 1183fa9e4066Sahrens 1184fa9e4066Sahrens /* 1185fa9e4066Sahrens * If they are requesting more space, and our current estimate 1186a9799022Sck * is over quota, they get to try again unless the actual 1187ea8dc4b6Seschrock * on-disk is over quota and there are no pending changes (which 1188ea8dc4b6Seschrock * may free up space for us). 1189fa9e4066Sahrens */ 1190468c413aSTim Haley if (used_on_disk + est_inflight >= quota) { 1191468c413aSTim Haley if (est_inflight > 0 || used_on_disk < quota || 1192468c413aSTim Haley (retval == ENOSPC && used_on_disk < quota + deferred)) 1193468c413aSTim Haley retval = ERESTART; 1194a9799022Sck dprintf_dd(dd, "failing: used=%lluK inflight = %lluK " 1195fa9e4066Sahrens "quota=%lluK tr=%lluK err=%d\n", 1196a9799022Sck used_on_disk>>10, est_inflight>>10, 1197468c413aSTim Haley quota>>10, asize>>10, retval); 1198fa9e4066Sahrens mutex_exit(&dd->dd_lock); 1199be6fd75aSMatthew Ahrens return (SET_ERROR(retval)); 1200fa9e4066Sahrens } 1201fa9e4066Sahrens 1202fa9e4066Sahrens /* We need to up our estimated delta before dropping dd_lock */ 1203fa9e4066Sahrens dd->dd_tempreserved[txgidx] += asize; 1204fa9e4066Sahrens 12059082849eSck parent_rsrv = parent_delta(dd, used_on_disk + est_inflight, 12069082849eSck asize - ref_rsrv); 1207fa9e4066Sahrens mutex_exit(&dd->dd_lock); 1208fa9e4066Sahrens 12091ab7f2deSmaybee tr = kmem_zalloc(sizeof (struct tempreserve), KM_SLEEP); 1210fa9e4066Sahrens tr->tr_ds = dd; 1211fa9e4066Sahrens tr->tr_size = asize; 1212fa9e4066Sahrens list_insert_tail(tr_list, tr); 1213fa9e4066Sahrens 1214fa9e4066Sahrens /* see if it's OK with our parent */ 12151934e92fSmaybee if (dd->dd_parent && parent_rsrv) { 1216c1379625SJustin T. Gibbs boolean_t ismos = (dsl_dir_phys(dd)->dd_head_dataset_obj == 0); 12171934e92fSmaybee 1218fa9e4066Sahrens return (dsl_dir_tempreserve_impl(dd->dd_parent, 12196df6c3bcSck parent_rsrv, netfree, ismos, TRUE, tr_list, tx, FALSE)); 1220fa9e4066Sahrens } else { 1221fa9e4066Sahrens return (0); 1222fa9e4066Sahrens } 1223fa9e4066Sahrens } 1224fa9e4066Sahrens 1225fa9e4066Sahrens /* 1226fa9e4066Sahrens * Reserve space in this dsl_dir, to be used in this tx's txg. 1227a9799022Sck * After the space has been dirtied (and dsl_dir_willuse_space() 1228a9799022Sck * has been called), the reservation should be canceled, using 1229a9799022Sck * dsl_dir_tempreserve_clear(). 1230fa9e4066Sahrens */ 1231fa9e4066Sahrens int 1232a9799022Sck dsl_dir_tempreserve_space(dsl_dir_t *dd, uint64_t lsize, uint64_t asize, 1233a9799022Sck uint64_t fsize, uint64_t usize, void **tr_cookiep, dmu_tx_t *tx) 1234fa9e4066Sahrens { 12351ab7f2deSmaybee int err; 1236fa9e4066Sahrens list_t *tr_list; 1237fa9e4066Sahrens 1238a9799022Sck if (asize == 0) { 1239a9799022Sck *tr_cookiep = NULL; 1240a9799022Sck return (0); 1241a9799022Sck } 1242a9799022Sck 1243fa9e4066Sahrens tr_list = kmem_alloc(sizeof (list_t), KM_SLEEP); 1244fa9e4066Sahrens list_create(tr_list, sizeof (struct tempreserve), 1245fa9e4066Sahrens offsetof(struct tempreserve, tr_node)); 1246a9799022Sck ASSERT3S(asize, >, 0); 1247ea8dc4b6Seschrock ASSERT3S(fsize, >=, 0); 1248fa9e4066Sahrens 12491ab7f2deSmaybee err = arc_tempreserve_space(lsize, tx->tx_txg); 1250fa9e4066Sahrens if (err == 0) { 1251fa9e4066Sahrens struct tempreserve *tr; 1252fa9e4066Sahrens 12531ab7f2deSmaybee tr = kmem_zalloc(sizeof (struct tempreserve), KM_SLEEP); 12541ab7f2deSmaybee tr->tr_size = lsize; 12551ab7f2deSmaybee list_insert_tail(tr_list, tr); 12561ab7f2deSmaybee } else { 12571ab7f2deSmaybee if (err == EAGAIN) { 125869962b56SMatthew Ahrens /* 125969962b56SMatthew Ahrens * If arc_memory_throttle() detected that pageout 126069962b56SMatthew Ahrens * is running and we are low on memory, we delay new 126169962b56SMatthew Ahrens * non-pageout transactions to give pageout an 126269962b56SMatthew Ahrens * advantage. 126369962b56SMatthew Ahrens * 126469962b56SMatthew Ahrens * It is unfortunate to be delaying while the caller's 126569962b56SMatthew Ahrens * locks are held. 126669962b56SMatthew Ahrens */ 12670689f76cSAdam Leventhal txg_delay(dd->dd_pool, tx->tx_txg, 12680689f76cSAdam Leventhal MSEC2NSEC(10), MSEC2NSEC(10)); 1269be6fd75aSMatthew Ahrens err = SET_ERROR(ERESTART); 1270fa9e4066Sahrens } 12711ab7f2deSmaybee } 12721ab7f2deSmaybee 12731ab7f2deSmaybee if (err == 0) { 12741ab7f2deSmaybee err = dsl_dir_tempreserve_impl(dd, asize, fsize >= asize, 12756df6c3bcSck FALSE, asize > usize, tr_list, tx, TRUE); 1276fa9e4066Sahrens } 1277fa9e4066Sahrens 12783b2aab18SMatthew Ahrens if (err != 0) 1279fa9e4066Sahrens dsl_dir_tempreserve_clear(tr_list, tx); 1280fa9e4066Sahrens else 1281fa9e4066Sahrens *tr_cookiep = tr_list; 12821ab7f2deSmaybee 1283fa9e4066Sahrens return (err); 1284fa9e4066Sahrens } 1285fa9e4066Sahrens 1286fa9e4066Sahrens /* 1287fa9e4066Sahrens * Clear a temporary reservation that we previously made with 1288fa9e4066Sahrens * dsl_dir_tempreserve_space(). 1289fa9e4066Sahrens */ 1290fa9e4066Sahrens void 1291fa9e4066Sahrens dsl_dir_tempreserve_clear(void *tr_cookie, dmu_tx_t *tx) 1292fa9e4066Sahrens { 1293fa9e4066Sahrens int txgidx = tx->tx_txg & TXG_MASK; 1294fa9e4066Sahrens list_t *tr_list = tr_cookie; 1295fa9e4066Sahrens struct tempreserve *tr; 1296fa9e4066Sahrens 1297fa9e4066Sahrens ASSERT3U(tx->tx_txg, !=, 0); 1298fa9e4066Sahrens 1299a9799022Sck if (tr_cookie == NULL) 1300a9799022Sck return; 1301a9799022Sck 130269962b56SMatthew Ahrens while ((tr = list_head(tr_list)) != NULL) { 130369962b56SMatthew Ahrens if (tr->tr_ds) { 1304fa9e4066Sahrens mutex_enter(&tr->tr_ds->dd_lock); 1305fa9e4066Sahrens ASSERT3U(tr->tr_ds->dd_tempreserved[txgidx], >=, 1306fa9e4066Sahrens tr->tr_size); 1307fa9e4066Sahrens tr->tr_ds->dd_tempreserved[txgidx] -= tr->tr_size; 1308fa9e4066Sahrens mutex_exit(&tr->tr_ds->dd_lock); 13091ab7f2deSmaybee } else { 13101ab7f2deSmaybee arc_tempreserve_clear(tr->tr_size); 1311fa9e4066Sahrens } 1312fa9e4066Sahrens list_remove(tr_list, tr); 1313fa9e4066Sahrens kmem_free(tr, sizeof (struct tempreserve)); 1314fa9e4066Sahrens } 1315fa9e4066Sahrens 1316fa9e4066Sahrens kmem_free(tr_list, sizeof (list_t)); 1317fa9e4066Sahrens } 1318fa9e4066Sahrens 131969962b56SMatthew Ahrens /* 132069962b56SMatthew Ahrens * This should be called from open context when we think we're going to write 132169962b56SMatthew Ahrens * or free space, for example when dirtying data. Be conservative; it's okay 132269962b56SMatthew Ahrens * to write less space or free more, but we don't want to write more or free 132369962b56SMatthew Ahrens * less than the amount specified. 132469962b56SMatthew Ahrens */ 132569962b56SMatthew Ahrens void 132669962b56SMatthew Ahrens dsl_dir_willuse_space(dsl_dir_t *dd, int64_t space, dmu_tx_t *tx) 1327fa9e4066Sahrens { 1328fa9e4066Sahrens int64_t parent_space; 1329fa9e4066Sahrens uint64_t est_used; 1330fa9e4066Sahrens 1331fa9e4066Sahrens mutex_enter(&dd->dd_lock); 1332fa9e4066Sahrens if (space > 0) 1333fa9e4066Sahrens dd->dd_space_towrite[tx->tx_txg & TXG_MASK] += space; 1334fa9e4066Sahrens 1335c1379625SJustin T. Gibbs est_used = dsl_dir_space_towrite(dd) + dsl_dir_phys(dd)->dd_used_bytes; 1336fa9e4066Sahrens parent_space = parent_delta(dd, est_used, space); 1337fa9e4066Sahrens mutex_exit(&dd->dd_lock); 1338fa9e4066Sahrens 1339fa9e4066Sahrens /* Make sure that we clean up dd_space_to* */ 1340fa9e4066Sahrens dsl_dir_dirty(dd, tx); 1341fa9e4066Sahrens 1342fa9e4066Sahrens /* XXX this is potentially expensive and unnecessary... */ 1343fa9e4066Sahrens if (parent_space && dd->dd_parent) 134469962b56SMatthew Ahrens dsl_dir_willuse_space(dd->dd_parent, parent_space, tx); 1345fa9e4066Sahrens } 1346fa9e4066Sahrens 1347fa9e4066Sahrens /* call from syncing context when we actually write/free space for this dd */ 1348fa9e4066Sahrens void 134974e7dc98SMatthew Ahrens dsl_dir_diduse_space(dsl_dir_t *dd, dd_used_t type, 1350fa9e4066Sahrens int64_t used, int64_t compressed, int64_t uncompressed, dmu_tx_t *tx) 1351fa9e4066Sahrens { 1352fa9e4066Sahrens int64_t accounted_delta; 1353b62969f8SMatthew Ahrens 1354b62969f8SMatthew Ahrens /* 1355b62969f8SMatthew Ahrens * dsl_dataset_set_refreservation_sync_impl() calls this with 1356b62969f8SMatthew Ahrens * dd_lock held, so that it can atomically update 1357b62969f8SMatthew Ahrens * ds->ds_reserved and the dsl_dir accounting, so that 1358b62969f8SMatthew Ahrens * dsl_dataset_check_quota() can see dataset and dir accounting 1359b62969f8SMatthew Ahrens * consistently. 1360b62969f8SMatthew Ahrens */ 136102c8f3f0SMatthew Ahrens boolean_t needlock = !MUTEX_HELD(&dd->dd_lock); 1362fa9e4066Sahrens 1363fa9e4066Sahrens ASSERT(dmu_tx_is_syncing(tx)); 136474e7dc98SMatthew Ahrens ASSERT(type < DD_USED_NUM); 1365fa9e4066Sahrens 1366b62969f8SMatthew Ahrens dmu_buf_will_dirty(dd->dd_dbuf, tx); 1367b62969f8SMatthew Ahrens 136802c8f3f0SMatthew Ahrens if (needlock) 136902c8f3f0SMatthew Ahrens mutex_enter(&dd->dd_lock); 1370c1379625SJustin T. Gibbs accounted_delta = 1371c1379625SJustin T. Gibbs parent_delta(dd, dsl_dir_phys(dd)->dd_used_bytes, used); 1372c1379625SJustin T. Gibbs ASSERT(used >= 0 || dsl_dir_phys(dd)->dd_used_bytes >= -used); 1373fa9e4066Sahrens ASSERT(compressed >= 0 || 1374c1379625SJustin T. Gibbs dsl_dir_phys(dd)->dd_compressed_bytes >= -compressed); 1375fa9e4066Sahrens ASSERT(uncompressed >= 0 || 1376c1379625SJustin T. Gibbs dsl_dir_phys(dd)->dd_uncompressed_bytes >= -uncompressed); 1377c1379625SJustin T. Gibbs dsl_dir_phys(dd)->dd_used_bytes += used; 1378c1379625SJustin T. Gibbs dsl_dir_phys(dd)->dd_uncompressed_bytes += uncompressed; 1379c1379625SJustin T. Gibbs dsl_dir_phys(dd)->dd_compressed_bytes += compressed; 138074e7dc98SMatthew Ahrens 1381c1379625SJustin T. Gibbs if (dsl_dir_phys(dd)->dd_flags & DD_FLAG_USED_BREAKDOWN) { 138274e7dc98SMatthew Ahrens ASSERT(used > 0 || 1383c1379625SJustin T. Gibbs dsl_dir_phys(dd)->dd_used_breakdown[type] >= -used); 1384c1379625SJustin T. Gibbs dsl_dir_phys(dd)->dd_used_breakdown[type] += used; 138574e7dc98SMatthew Ahrens #ifdef DEBUG 138674e7dc98SMatthew Ahrens dd_used_t t; 138774e7dc98SMatthew Ahrens uint64_t u = 0; 138874e7dc98SMatthew Ahrens for (t = 0; t < DD_USED_NUM; t++) 1389c1379625SJustin T. Gibbs u += dsl_dir_phys(dd)->dd_used_breakdown[t]; 1390c1379625SJustin T. Gibbs ASSERT3U(u, ==, dsl_dir_phys(dd)->dd_used_bytes); 139174e7dc98SMatthew Ahrens #endif 139274e7dc98SMatthew Ahrens } 139302c8f3f0SMatthew Ahrens if (needlock) 139402c8f3f0SMatthew Ahrens mutex_exit(&dd->dd_lock); 1395fa9e4066Sahrens 1396fa9e4066Sahrens if (dd->dd_parent != NULL) { 139774e7dc98SMatthew Ahrens dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD, 1398fa9e4066Sahrens accounted_delta, compressed, uncompressed, tx); 139974e7dc98SMatthew Ahrens dsl_dir_transfer_space(dd->dd_parent, 140074e7dc98SMatthew Ahrens used - accounted_delta, 140174e7dc98SMatthew Ahrens DD_USED_CHILD_RSRV, DD_USED_CHILD, tx); 1402fa9e4066Sahrens } 1403fa9e4066Sahrens } 1404fa9e4066Sahrens 140574e7dc98SMatthew Ahrens void 140674e7dc98SMatthew Ahrens dsl_dir_transfer_space(dsl_dir_t *dd, int64_t delta, 140774e7dc98SMatthew Ahrens dd_used_t oldtype, dd_used_t newtype, dmu_tx_t *tx) 140874e7dc98SMatthew Ahrens { 140974e7dc98SMatthew Ahrens ASSERT(dmu_tx_is_syncing(tx)); 141074e7dc98SMatthew Ahrens ASSERT(oldtype < DD_USED_NUM); 141174e7dc98SMatthew Ahrens ASSERT(newtype < DD_USED_NUM); 141274e7dc98SMatthew Ahrens 1413c1379625SJustin T. Gibbs if (delta == 0 || 1414c1379625SJustin T. Gibbs !(dsl_dir_phys(dd)->dd_flags & DD_FLAG_USED_BREAKDOWN)) 141574e7dc98SMatthew Ahrens return; 141674e7dc98SMatthew Ahrens 1417b62969f8SMatthew Ahrens dmu_buf_will_dirty(dd->dd_dbuf, tx); 1418b62969f8SMatthew Ahrens mutex_enter(&dd->dd_lock); 141974e7dc98SMatthew Ahrens ASSERT(delta > 0 ? 1420c1379625SJustin T. Gibbs dsl_dir_phys(dd)->dd_used_breakdown[oldtype] >= delta : 1421c1379625SJustin T. Gibbs dsl_dir_phys(dd)->dd_used_breakdown[newtype] >= -delta); 1422c1379625SJustin T. Gibbs ASSERT(dsl_dir_phys(dd)->dd_used_bytes >= ABS(delta)); 1423c1379625SJustin T. Gibbs dsl_dir_phys(dd)->dd_used_breakdown[oldtype] -= delta; 1424c1379625SJustin T. Gibbs dsl_dir_phys(dd)->dd_used_breakdown[newtype] += delta; 1425b62969f8SMatthew Ahrens mutex_exit(&dd->dd_lock); 142674e7dc98SMatthew Ahrens } 142774e7dc98SMatthew Ahrens 14283b2aab18SMatthew Ahrens typedef struct dsl_dir_set_qr_arg { 14293b2aab18SMatthew Ahrens const char *ddsqra_name; 14303b2aab18SMatthew Ahrens zprop_source_t ddsqra_source; 14313b2aab18SMatthew Ahrens uint64_t ddsqra_value; 14323b2aab18SMatthew Ahrens } dsl_dir_set_qr_arg_t; 14333b2aab18SMatthew Ahrens 1434fa9e4066Sahrens static int 14353b2aab18SMatthew Ahrens dsl_dir_set_quota_check(void *arg, dmu_tx_t *tx) 1436fa9e4066Sahrens { 14373b2aab18SMatthew Ahrens dsl_dir_set_qr_arg_t *ddsqra = arg; 14383b2aab18SMatthew Ahrens dsl_pool_t *dp = dmu_tx_pool(tx); 14393b2aab18SMatthew Ahrens dsl_dataset_t *ds; 14403b2aab18SMatthew Ahrens int error; 14413b2aab18SMatthew Ahrens uint64_t towrite, newval; 1442fa9e4066Sahrens 14433b2aab18SMatthew Ahrens error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds); 14443b2aab18SMatthew Ahrens if (error != 0) 14453b2aab18SMatthew Ahrens return (error); 14463b2aab18SMatthew Ahrens 14473b2aab18SMatthew Ahrens error = dsl_prop_predict(ds->ds_dir, "quota", 14483b2aab18SMatthew Ahrens ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval); 14493b2aab18SMatthew Ahrens if (error != 0) { 14503b2aab18SMatthew Ahrens dsl_dataset_rele(ds, FTAG); 14513b2aab18SMatthew Ahrens return (error); 14523b2aab18SMatthew Ahrens } 145392241e0bSTom Erickson 14543b2aab18SMatthew Ahrens if (newval == 0) { 14553b2aab18SMatthew Ahrens dsl_dataset_rele(ds, FTAG); 14561d452cf5Sahrens return (0); 14573b2aab18SMatthew Ahrens } 1458fa9e4066Sahrens 14593b2aab18SMatthew Ahrens mutex_enter(&ds->ds_dir->dd_lock); 14601d452cf5Sahrens /* 14611d452cf5Sahrens * If we are doing the preliminary check in open context, and 14621d452cf5Sahrens * there are pending changes, then don't fail it, since the 1463a9799022Sck * pending changes could under-estimate the amount of space to be 14641d452cf5Sahrens * freed up. 14651d452cf5Sahrens */ 14663b2aab18SMatthew Ahrens towrite = dsl_dir_space_towrite(ds->ds_dir); 14671d452cf5Sahrens if ((dmu_tx_is_syncing(tx) || towrite == 0) && 1468c1379625SJustin T. Gibbs (newval < dsl_dir_phys(ds->ds_dir)->dd_reserved || 1469c1379625SJustin T. Gibbs newval < dsl_dir_phys(ds->ds_dir)->dd_used_bytes + towrite)) { 1470be6fd75aSMatthew Ahrens error = SET_ERROR(ENOSPC); 1471fa9e4066Sahrens } 14723b2aab18SMatthew Ahrens mutex_exit(&ds->ds_dir->dd_lock); 14733b2aab18SMatthew Ahrens dsl_dataset_rele(ds, FTAG); 14743b2aab18SMatthew Ahrens return (error); 1475fa9e4066Sahrens } 1476fa9e4066Sahrens 14771d452cf5Sahrens static void 14783b2aab18SMatthew Ahrens dsl_dir_set_quota_sync(void *arg, dmu_tx_t *tx) 14791d452cf5Sahrens { 14803b2aab18SMatthew Ahrens dsl_dir_set_qr_arg_t *ddsqra = arg; 14813b2aab18SMatthew Ahrens dsl_pool_t *dp = dmu_tx_pool(tx); 14823b2aab18SMatthew Ahrens dsl_dataset_t *ds; 14833b2aab18SMatthew Ahrens uint64_t newval; 148492241e0bSTom Erickson 14853b2aab18SMatthew Ahrens VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds)); 14861d452cf5Sahrens 1487013023d4SMartin Matuska if (spa_version(dp->dp_spa) >= SPA_VERSION_RECVD_PROPS) { 1488013023d4SMartin Matuska dsl_prop_set_sync_impl(ds, zfs_prop_to_name(ZFS_PROP_QUOTA), 1489013023d4SMartin Matuska ddsqra->ddsqra_source, sizeof (ddsqra->ddsqra_value), 1, 1490013023d4SMartin Matuska &ddsqra->ddsqra_value, tx); 14911d452cf5Sahrens 1492013023d4SMartin Matuska VERIFY0(dsl_prop_get_int_ds(ds, 1493013023d4SMartin Matuska zfs_prop_to_name(ZFS_PROP_QUOTA), &newval)); 1494013023d4SMartin Matuska } else { 1495013023d4SMartin Matuska newval = ddsqra->ddsqra_value; 1496013023d4SMartin Matuska spa_history_log_internal_ds(ds, "set", tx, "%s=%lld", 1497013023d4SMartin Matuska zfs_prop_to_name(ZFS_PROP_QUOTA), (longlong_t)newval); 1498013023d4SMartin Matuska } 14993b2aab18SMatthew Ahrens 15003b2aab18SMatthew Ahrens dmu_buf_will_dirty(ds->ds_dir->dd_dbuf, tx); 15013b2aab18SMatthew Ahrens mutex_enter(&ds->ds_dir->dd_lock); 1502c1379625SJustin T. Gibbs dsl_dir_phys(ds->ds_dir)->dd_quota = newval; 15033b2aab18SMatthew Ahrens mutex_exit(&ds->ds_dir->dd_lock); 15043b2aab18SMatthew Ahrens dsl_dataset_rele(ds, FTAG); 15051d452cf5Sahrens } 15061d452cf5Sahrens 1507fa9e4066Sahrens int 150892241e0bSTom Erickson dsl_dir_set_quota(const char *ddname, zprop_source_t source, uint64_t quota) 1509fa9e4066Sahrens { 15103b2aab18SMatthew Ahrens dsl_dir_set_qr_arg_t ddsqra; 1511fa9e4066Sahrens 15123b2aab18SMatthew Ahrens ddsqra.ddsqra_name = ddname; 15133b2aab18SMatthew Ahrens ddsqra.ddsqra_source = source; 15143b2aab18SMatthew Ahrens ddsqra.ddsqra_value = quota; 151592241e0bSTom Erickson 15163b2aab18SMatthew Ahrens return (dsl_sync_task(ddname, dsl_dir_set_quota_check, 15177d46dc6cSMatthew Ahrens dsl_dir_set_quota_sync, &ddsqra, 0, ZFS_SPACE_CHECK_NONE)); 1518fa9e4066Sahrens } 1519fa9e4066Sahrens 1520a9799022Sck int 15213b2aab18SMatthew Ahrens dsl_dir_set_reservation_check(void *arg, dmu_tx_t *tx) 1522fa9e4066Sahrens { 15233b2aab18SMatthew Ahrens dsl_dir_set_qr_arg_t *ddsqra = arg; 15243b2aab18SMatthew Ahrens dsl_pool_t *dp = dmu_tx_pool(tx); 15253b2aab18SMatthew Ahrens dsl_dataset_t *ds; 15263b2aab18SMatthew Ahrens dsl_dir_t *dd; 15273b2aab18SMatthew Ahrens uint64_t newval, used, avail; 15283b2aab18SMatthew Ahrens int error; 152992241e0bSTom Erickson 15303b2aab18SMatthew Ahrens error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds); 15313b2aab18SMatthew Ahrens if (error != 0) 15323b2aab18SMatthew Ahrens return (error); 15333b2aab18SMatthew Ahrens dd = ds->ds_dir; 1534fa9e4066Sahrens 15351d452cf5Sahrens /* 15361d452cf5Sahrens * If we are doing the preliminary check in open context, the 15371d452cf5Sahrens * space estimates may be inaccurate. 15381d452cf5Sahrens */ 15393b2aab18SMatthew Ahrens if (!dmu_tx_is_syncing(tx)) { 15403b2aab18SMatthew Ahrens dsl_dataset_rele(ds, FTAG); 15411d452cf5Sahrens return (0); 15423b2aab18SMatthew Ahrens } 15433b2aab18SMatthew Ahrens 15443b2aab18SMatthew Ahrens error = dsl_prop_predict(ds->ds_dir, 15453b2aab18SMatthew Ahrens zfs_prop_to_name(ZFS_PROP_RESERVATION), 15463b2aab18SMatthew Ahrens ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval); 15473b2aab18SMatthew Ahrens if (error != 0) { 15483b2aab18SMatthew Ahrens dsl_dataset_rele(ds, FTAG); 15493b2aab18SMatthew Ahrens return (error); 15503b2aab18SMatthew Ahrens } 15511d452cf5Sahrens 1552fa9e4066Sahrens mutex_enter(&dd->dd_lock); 1553c1379625SJustin T. Gibbs used = dsl_dir_phys(dd)->dd_used_bytes; 1554fa9e4066Sahrens mutex_exit(&dd->dd_lock); 1555fa9e4066Sahrens 1556fa9e4066Sahrens if (dd->dd_parent) { 1557fa9e4066Sahrens avail = dsl_dir_space_available(dd->dd_parent, 1558fa9e4066Sahrens NULL, 0, FALSE); 1559fa9e4066Sahrens } else { 1560fa9e4066Sahrens avail = dsl_pool_adjustedsize(dd->dd_pool, B_FALSE) - used; 1561fa9e4066Sahrens } 1562fa9e4066Sahrens 1563c1379625SJustin T. Gibbs if (MAX(used, newval) > MAX(used, dsl_dir_phys(dd)->dd_reserved)) { 15643b2aab18SMatthew Ahrens uint64_t delta = MAX(used, newval) - 1565c1379625SJustin T. Gibbs MAX(used, dsl_dir_phys(dd)->dd_reserved); 1566379c004dSEric Schrock 15673b2aab18SMatthew Ahrens if (delta > avail || 1568c1379625SJustin T. Gibbs (dsl_dir_phys(dd)->dd_quota > 0 && 1569c1379625SJustin T. Gibbs newval > dsl_dir_phys(dd)->dd_quota)) 1570be6fd75aSMatthew Ahrens error = SET_ERROR(ENOSPC); 1571379c004dSEric Schrock } 1572379c004dSEric Schrock 15733b2aab18SMatthew Ahrens dsl_dataset_rele(ds, FTAG); 15743b2aab18SMatthew Ahrens return (error); 15751d452cf5Sahrens } 15761d452cf5Sahrens 15773b2aab18SMatthew Ahrens void 15784445fffbSMatthew Ahrens dsl_dir_set_reservation_sync_impl(dsl_dir_t *dd, uint64_t value, dmu_tx_t *tx) 15791d452cf5Sahrens { 15801d452cf5Sahrens uint64_t used; 15811d452cf5Sahrens int64_t delta; 15821d452cf5Sahrens 1583a9799022Sck dmu_buf_will_dirty(dd->dd_dbuf, tx); 1584a9799022Sck 15851d452cf5Sahrens mutex_enter(&dd->dd_lock); 1586c1379625SJustin T. Gibbs used = dsl_dir_phys(dd)->dd_used_bytes; 1587c1379625SJustin T. Gibbs delta = MAX(used, value) - MAX(used, dsl_dir_phys(dd)->dd_reserved); 1588c1379625SJustin T. Gibbs dsl_dir_phys(dd)->dd_reserved = value; 1589fa9e4066Sahrens 1590fa9e4066Sahrens if (dd->dd_parent != NULL) { 1591fa9e4066Sahrens /* Roll up this additional usage into our ancestors */ 159274e7dc98SMatthew Ahrens dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD_RSRV, 159374e7dc98SMatthew Ahrens delta, 0, 0, tx); 1594fa9e4066Sahrens } 159502c8f3f0SMatthew Ahrens mutex_exit(&dd->dd_lock); 15964445fffbSMatthew Ahrens } 15974445fffbSMatthew Ahrens 15984445fffbSMatthew Ahrens 15994445fffbSMatthew Ahrens static void 16003b2aab18SMatthew Ahrens dsl_dir_set_reservation_sync(void *arg, dmu_tx_t *tx) 16014445fffbSMatthew Ahrens { 16023b2aab18SMatthew Ahrens dsl_dir_set_qr_arg_t *ddsqra = arg; 16033b2aab18SMatthew Ahrens dsl_pool_t *dp = dmu_tx_pool(tx); 16043b2aab18SMatthew Ahrens dsl_dataset_t *ds; 16053b2aab18SMatthew Ahrens uint64_t newval; 16063b2aab18SMatthew Ahrens 16073b2aab18SMatthew Ahrens VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds)); 16083b2aab18SMatthew Ahrens 1609013023d4SMartin Matuska if (spa_version(dp->dp_spa) >= SPA_VERSION_RECVD_PROPS) { 1610013023d4SMartin Matuska dsl_prop_set_sync_impl(ds, 1611013023d4SMartin Matuska zfs_prop_to_name(ZFS_PROP_RESERVATION), 1612013023d4SMartin Matuska ddsqra->ddsqra_source, sizeof (ddsqra->ddsqra_value), 1, 1613013023d4SMartin Matuska &ddsqra->ddsqra_value, tx); 16144445fffbSMatthew Ahrens 1615013023d4SMartin Matuska VERIFY0(dsl_prop_get_int_ds(ds, 1616013023d4SMartin Matuska zfs_prop_to_name(ZFS_PROP_RESERVATION), &newval)); 1617013023d4SMartin Matuska } else { 1618013023d4SMartin Matuska newval = ddsqra->ddsqra_value; 1619013023d4SMartin Matuska spa_history_log_internal_ds(ds, "set", tx, "%s=%lld", 1620013023d4SMartin Matuska zfs_prop_to_name(ZFS_PROP_RESERVATION), 1621013023d4SMartin Matuska (longlong_t)newval); 1622013023d4SMartin Matuska } 1623ecd6cf80Smarks 16243b2aab18SMatthew Ahrens dsl_dir_set_reservation_sync_impl(ds->ds_dir, newval, tx); 16253b2aab18SMatthew Ahrens dsl_dataset_rele(ds, FTAG); 1626fa9e4066Sahrens } 1627fa9e4066Sahrens 1628fa9e4066Sahrens int 162992241e0bSTom Erickson dsl_dir_set_reservation(const char *ddname, zprop_source_t source, 163092241e0bSTom Erickson uint64_t reservation) 1631fa9e4066Sahrens { 16323b2aab18SMatthew Ahrens dsl_dir_set_qr_arg_t ddsqra; 163392241e0bSTom Erickson 16343b2aab18SMatthew Ahrens ddsqra.ddsqra_name = ddname; 16353b2aab18SMatthew Ahrens ddsqra.ddsqra_source = source; 16363b2aab18SMatthew Ahrens ddsqra.ddsqra_value = reservation; 163792241e0bSTom Erickson 16383b2aab18SMatthew Ahrens return (dsl_sync_task(ddname, dsl_dir_set_reservation_check, 16397d46dc6cSMatthew Ahrens dsl_dir_set_reservation_sync, &ddsqra, 0, ZFS_SPACE_CHECK_NONE)); 1640fa9e4066Sahrens } 1641fa9e4066Sahrens 1642fa9e4066Sahrens static dsl_dir_t * 1643fa9e4066Sahrens closest_common_ancestor(dsl_dir_t *ds1, dsl_dir_t *ds2) 1644fa9e4066Sahrens { 1645fa9e4066Sahrens for (; ds1; ds1 = ds1->dd_parent) { 1646fa9e4066Sahrens dsl_dir_t *dd; 1647fa9e4066Sahrens for (dd = ds2; dd; dd = dd->dd_parent) { 1648fa9e4066Sahrens if (ds1 == dd) 1649fa9e4066Sahrens return (dd); 1650fa9e4066Sahrens } 1651fa9e4066Sahrens } 1652fa9e4066Sahrens return (NULL); 1653fa9e4066Sahrens } 1654fa9e4066Sahrens 1655fa9e4066Sahrens /* 1656fa9e4066Sahrens * If delta is applied to dd, how much of that delta would be applied to 1657fa9e4066Sahrens * ancestor? Syncing context only. 1658fa9e4066Sahrens */ 1659fa9e4066Sahrens static int64_t 1660fa9e4066Sahrens would_change(dsl_dir_t *dd, int64_t delta, dsl_dir_t *ancestor) 1661fa9e4066Sahrens { 1662fa9e4066Sahrens if (dd == ancestor) 1663fa9e4066Sahrens return (delta); 1664fa9e4066Sahrens 1665fa9e4066Sahrens mutex_enter(&dd->dd_lock); 1666c1379625SJustin T. Gibbs delta = parent_delta(dd, dsl_dir_phys(dd)->dd_used_bytes, delta); 1667fa9e4066Sahrens mutex_exit(&dd->dd_lock); 1668fa9e4066Sahrens return (would_change(dd->dd_parent, delta, ancestor)); 1669fa9e4066Sahrens } 1670fa9e4066Sahrens 16713b2aab18SMatthew Ahrens typedef struct dsl_dir_rename_arg { 16723b2aab18SMatthew Ahrens const char *ddra_oldname; 16733b2aab18SMatthew Ahrens const char *ddra_newname; 1674a2afb611SJerry Jelinek cred_t *ddra_cred; 16753b2aab18SMatthew Ahrens } dsl_dir_rename_arg_t; 16761d452cf5Sahrens 16773b2aab18SMatthew Ahrens /* ARGSUSED */ 16781d452cf5Sahrens static int 16793b2aab18SMatthew Ahrens dsl_valid_rename(dsl_pool_t *dp, dsl_dataset_t *ds, void *arg) 1680fa9e4066Sahrens { 16813b2aab18SMatthew Ahrens int *deltap = arg; 1682*9adfa60dSMatthew Ahrens char namebuf[ZFS_MAX_DATASET_NAME_LEN]; 1683fa9e4066Sahrens 16843b2aab18SMatthew Ahrens dsl_dataset_name(ds, namebuf); 16853b2aab18SMatthew Ahrens 1686*9adfa60dSMatthew Ahrens if (strlen(namebuf) + *deltap >= ZFS_MAX_DATASET_NAME_LEN) 1687be6fd75aSMatthew Ahrens return (SET_ERROR(ENAMETOOLONG)); 16883b2aab18SMatthew Ahrens return (0); 16893b2aab18SMatthew Ahrens } 16903b2aab18SMatthew Ahrens 16913b2aab18SMatthew Ahrens static int 16923b2aab18SMatthew Ahrens dsl_dir_rename_check(void *arg, dmu_tx_t *tx) 16933b2aab18SMatthew Ahrens { 16943b2aab18SMatthew Ahrens dsl_dir_rename_arg_t *ddra = arg; 16953b2aab18SMatthew Ahrens dsl_pool_t *dp = dmu_tx_pool(tx); 16963b2aab18SMatthew Ahrens dsl_dir_t *dd, *newparent; 16973b2aab18SMatthew Ahrens const char *mynewname; 16983b2aab18SMatthew Ahrens int error; 16993b2aab18SMatthew Ahrens int delta = strlen(ddra->ddra_newname) - strlen(ddra->ddra_oldname); 1700fa9e4066Sahrens 17013b2aab18SMatthew Ahrens /* target dir should exist */ 17023b2aab18SMatthew Ahrens error = dsl_dir_hold(dp, ddra->ddra_oldname, FTAG, &dd, NULL); 17033b2aab18SMatthew Ahrens if (error != 0) 17043b2aab18SMatthew Ahrens return (error); 17053b2aab18SMatthew Ahrens 17063b2aab18SMatthew Ahrens /* new parent should exist */ 17073b2aab18SMatthew Ahrens error = dsl_dir_hold(dp, ddra->ddra_newname, FTAG, 17083b2aab18SMatthew Ahrens &newparent, &mynewname); 17093b2aab18SMatthew Ahrens if (error != 0) { 17103b2aab18SMatthew Ahrens dsl_dir_rele(dd, FTAG); 17113b2aab18SMatthew Ahrens return (error); 17123b2aab18SMatthew Ahrens } 17133b2aab18SMatthew Ahrens 17143b2aab18SMatthew Ahrens /* can't rename to different pool */ 17153b2aab18SMatthew Ahrens if (dd->dd_pool != newparent->dd_pool) { 17163b2aab18SMatthew Ahrens dsl_dir_rele(newparent, FTAG); 17173b2aab18SMatthew Ahrens dsl_dir_rele(dd, FTAG); 1718be6fd75aSMatthew Ahrens return (SET_ERROR(ENXIO)); 17193b2aab18SMatthew Ahrens } 17203b2aab18SMatthew Ahrens 17213b2aab18SMatthew Ahrens /* new name should not already exist */ 17223b2aab18SMatthew Ahrens if (mynewname == NULL) { 17233b2aab18SMatthew Ahrens dsl_dir_rele(newparent, FTAG); 17243b2aab18SMatthew Ahrens dsl_dir_rele(dd, FTAG); 1725be6fd75aSMatthew Ahrens return (SET_ERROR(EEXIST)); 17263b2aab18SMatthew Ahrens } 17273b2aab18SMatthew Ahrens 17283b2aab18SMatthew Ahrens /* if the name length is growing, validate child name lengths */ 17293b2aab18SMatthew Ahrens if (delta > 0) { 17303b2aab18SMatthew Ahrens error = dmu_objset_find_dp(dp, dd->dd_object, dsl_valid_rename, 17313b2aab18SMatthew Ahrens &delta, DS_FIND_CHILDREN | DS_FIND_SNAPSHOTS); 17323b2aab18SMatthew Ahrens if (error != 0) { 17333b2aab18SMatthew Ahrens dsl_dir_rele(newparent, FTAG); 17343b2aab18SMatthew Ahrens dsl_dir_rele(dd, FTAG); 17353b2aab18SMatthew Ahrens return (error); 17363b2aab18SMatthew Ahrens } 17373b2aab18SMatthew Ahrens } 17381d452cf5Sahrens 1739a2afb611SJerry Jelinek if (dmu_tx_is_syncing(tx)) { 1740adf34077SJerry Jelinek if (spa_feature_is_active(dp->dp_spa, 1741a2afb611SJerry Jelinek SPA_FEATURE_FS_SS_LIMIT)) { 1742a2afb611SJerry Jelinek /* 1743a2afb611SJerry Jelinek * Although this is the check function and we don't 1744a2afb611SJerry Jelinek * normally make on-disk changes in check functions, 1745a2afb611SJerry Jelinek * we need to do that here. 1746a2afb611SJerry Jelinek * 1747a2afb611SJerry Jelinek * Ensure this portion of the tree's counts have been 1748a2afb611SJerry Jelinek * initialized in case the new parent has limits set. 1749a2afb611SJerry Jelinek */ 1750a2afb611SJerry Jelinek dsl_dir_init_fs_ss_count(dd, tx); 1751a2afb611SJerry Jelinek } 1752a2afb611SJerry Jelinek } 1753a2afb611SJerry Jelinek 17543b2aab18SMatthew Ahrens if (newparent != dd->dd_parent) { 175599653d4eSeschrock /* is there enough space? */ 175699653d4eSeschrock uint64_t myspace = 1757c1379625SJustin T. Gibbs MAX(dsl_dir_phys(dd)->dd_used_bytes, 1758c1379625SJustin T. Gibbs dsl_dir_phys(dd)->dd_reserved); 1759a2afb611SJerry Jelinek objset_t *os = dd->dd_pool->dp_meta_objset; 1760a2afb611SJerry Jelinek uint64_t fs_cnt = 0; 1761a2afb611SJerry Jelinek uint64_t ss_cnt = 0; 1762a2afb611SJerry Jelinek 1763a2afb611SJerry Jelinek if (dsl_dir_is_zapified(dd)) { 1764a2afb611SJerry Jelinek int err; 1765a2afb611SJerry Jelinek 1766a2afb611SJerry Jelinek err = zap_lookup(os, dd->dd_object, 1767a2afb611SJerry Jelinek DD_FIELD_FILESYSTEM_COUNT, sizeof (fs_cnt), 1, 1768a2afb611SJerry Jelinek &fs_cnt); 1769adf34077SJerry Jelinek if (err != ENOENT && err != 0) { 1770adf34077SJerry Jelinek dsl_dir_rele(newparent, FTAG); 1771adf34077SJerry Jelinek dsl_dir_rele(dd, FTAG); 1772a2afb611SJerry Jelinek return (err); 1773adf34077SJerry Jelinek } 1774a2afb611SJerry Jelinek 1775a2afb611SJerry Jelinek /* 1776a2afb611SJerry Jelinek * have to add 1 for the filesystem itself that we're 1777a2afb611SJerry Jelinek * moving 1778a2afb611SJerry Jelinek */ 1779a2afb611SJerry Jelinek fs_cnt++; 1780a2afb611SJerry Jelinek 1781a2afb611SJerry Jelinek err = zap_lookup(os, dd->dd_object, 1782a2afb611SJerry Jelinek DD_FIELD_SNAPSHOT_COUNT, sizeof (ss_cnt), 1, 1783a2afb611SJerry Jelinek &ss_cnt); 1784adf34077SJerry Jelinek if (err != ENOENT && err != 0) { 1785adf34077SJerry Jelinek dsl_dir_rele(newparent, FTAG); 1786adf34077SJerry Jelinek dsl_dir_rele(dd, FTAG); 1787a2afb611SJerry Jelinek return (err); 1788adf34077SJerry Jelinek } 1789a2afb611SJerry Jelinek } 1790fa9e4066Sahrens 17911d452cf5Sahrens /* no rename into our descendant */ 17923b2aab18SMatthew Ahrens if (closest_common_ancestor(dd, newparent) == dd) { 17933b2aab18SMatthew Ahrens dsl_dir_rele(newparent, FTAG); 17943b2aab18SMatthew Ahrens dsl_dir_rele(dd, FTAG); 1795be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 17963b2aab18SMatthew Ahrens } 1797fa9e4066Sahrens 17983b2aab18SMatthew Ahrens error = dsl_dir_transfer_possible(dd->dd_parent, 1799a2afb611SJerry Jelinek newparent, fs_cnt, ss_cnt, myspace, ddra->ddra_cred); 18003b2aab18SMatthew Ahrens if (error != 0) { 18013b2aab18SMatthew Ahrens dsl_dir_rele(newparent, FTAG); 18023b2aab18SMatthew Ahrens dsl_dir_rele(dd, FTAG); 18033b2aab18SMatthew Ahrens return (error); 18043b2aab18SMatthew Ahrens } 18051d452cf5Sahrens } 1806fa9e4066Sahrens 18073b2aab18SMatthew Ahrens dsl_dir_rele(newparent, FTAG); 18083b2aab18SMatthew Ahrens dsl_dir_rele(dd, FTAG); 18091d452cf5Sahrens return (0); 18101d452cf5Sahrens } 18111d452cf5Sahrens 18121d452cf5Sahrens static void 18133b2aab18SMatthew Ahrens dsl_dir_rename_sync(void *arg, dmu_tx_t *tx) 18141d452cf5Sahrens { 18153b2aab18SMatthew Ahrens dsl_dir_rename_arg_t *ddra = arg; 18163b2aab18SMatthew Ahrens dsl_pool_t *dp = dmu_tx_pool(tx); 18173b2aab18SMatthew Ahrens dsl_dir_t *dd, *newparent; 18183b2aab18SMatthew Ahrens const char *mynewname; 18193b2aab18SMatthew Ahrens int error; 18201d452cf5Sahrens objset_t *mos = dp->dp_meta_objset; 18211d452cf5Sahrens 18223b2aab18SMatthew Ahrens VERIFY0(dsl_dir_hold(dp, ddra->ddra_oldname, FTAG, &dd, NULL)); 18233b2aab18SMatthew Ahrens VERIFY0(dsl_dir_hold(dp, ddra->ddra_newname, FTAG, &newparent, 18243b2aab18SMatthew Ahrens &mynewname)); 18251d452cf5Sahrens 18264445fffbSMatthew Ahrens /* Log this before we change the name. */ 18274445fffbSMatthew Ahrens spa_history_log_internal_dd(dd, "rename", tx, 18283b2aab18SMatthew Ahrens "-> %s", ddra->ddra_newname); 18294445fffbSMatthew Ahrens 18303b2aab18SMatthew Ahrens if (newparent != dd->dd_parent) { 1831a2afb611SJerry Jelinek objset_t *os = dd->dd_pool->dp_meta_objset; 1832a2afb611SJerry Jelinek uint64_t fs_cnt = 0; 1833a2afb611SJerry Jelinek uint64_t ss_cnt = 0; 1834a2afb611SJerry Jelinek 1835a2afb611SJerry Jelinek /* 1836a2afb611SJerry Jelinek * We already made sure the dd counts were initialized in the 1837a2afb611SJerry Jelinek * check function. 1838a2afb611SJerry Jelinek */ 1839adf34077SJerry Jelinek if (spa_feature_is_active(dp->dp_spa, 1840a2afb611SJerry Jelinek SPA_FEATURE_FS_SS_LIMIT)) { 1841a2afb611SJerry Jelinek VERIFY0(zap_lookup(os, dd->dd_object, 1842a2afb611SJerry Jelinek DD_FIELD_FILESYSTEM_COUNT, sizeof (fs_cnt), 1, 1843a2afb611SJerry Jelinek &fs_cnt)); 1844a2afb611SJerry Jelinek /* add 1 for the filesystem itself that we're moving */ 1845a2afb611SJerry Jelinek fs_cnt++; 1846a2afb611SJerry Jelinek 1847a2afb611SJerry Jelinek VERIFY0(zap_lookup(os, dd->dd_object, 1848a2afb611SJerry Jelinek DD_FIELD_SNAPSHOT_COUNT, sizeof (ss_cnt), 1, 1849a2afb611SJerry Jelinek &ss_cnt)); 1850a2afb611SJerry Jelinek } 1851a2afb611SJerry Jelinek 1852a2afb611SJerry Jelinek dsl_fs_ss_count_adjust(dd->dd_parent, -fs_cnt, 1853a2afb611SJerry Jelinek DD_FIELD_FILESYSTEM_COUNT, tx); 1854a2afb611SJerry Jelinek dsl_fs_ss_count_adjust(newparent, fs_cnt, 1855a2afb611SJerry Jelinek DD_FIELD_FILESYSTEM_COUNT, tx); 1856a2afb611SJerry Jelinek 1857a2afb611SJerry Jelinek dsl_fs_ss_count_adjust(dd->dd_parent, -ss_cnt, 1858a2afb611SJerry Jelinek DD_FIELD_SNAPSHOT_COUNT, tx); 1859a2afb611SJerry Jelinek dsl_fs_ss_count_adjust(newparent, ss_cnt, 1860a2afb611SJerry Jelinek DD_FIELD_SNAPSHOT_COUNT, tx); 1861a2afb611SJerry Jelinek 186274e7dc98SMatthew Ahrens dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD, 1863c1379625SJustin T. Gibbs -dsl_dir_phys(dd)->dd_used_bytes, 1864c1379625SJustin T. Gibbs -dsl_dir_phys(dd)->dd_compressed_bytes, 1865c1379625SJustin T. Gibbs -dsl_dir_phys(dd)->dd_uncompressed_bytes, tx); 18663b2aab18SMatthew Ahrens dsl_dir_diduse_space(newparent, DD_USED_CHILD, 1867c1379625SJustin T. Gibbs dsl_dir_phys(dd)->dd_used_bytes, 1868c1379625SJustin T. Gibbs dsl_dir_phys(dd)->dd_compressed_bytes, 1869c1379625SJustin T. Gibbs dsl_dir_phys(dd)->dd_uncompressed_bytes, tx); 187074e7dc98SMatthew Ahrens 1871c1379625SJustin T. Gibbs if (dsl_dir_phys(dd)->dd_reserved > 1872c1379625SJustin T. Gibbs dsl_dir_phys(dd)->dd_used_bytes) { 1873c1379625SJustin T. Gibbs uint64_t unused_rsrv = dsl_dir_phys(dd)->dd_reserved - 1874c1379625SJustin T. Gibbs dsl_dir_phys(dd)->dd_used_bytes; 187574e7dc98SMatthew Ahrens 187674e7dc98SMatthew Ahrens dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD_RSRV, 187774e7dc98SMatthew Ahrens -unused_rsrv, 0, 0, tx); 18783b2aab18SMatthew Ahrens dsl_dir_diduse_space(newparent, DD_USED_CHILD_RSRV, 187974e7dc98SMatthew Ahrens unused_rsrv, 0, 0, tx); 188074e7dc98SMatthew Ahrens } 1881fa9e4066Sahrens } 1882fa9e4066Sahrens 1883fa9e4066Sahrens dmu_buf_will_dirty(dd->dd_dbuf, tx); 1884fa9e4066Sahrens 1885fa9e4066Sahrens /* remove from old parent zapobj */ 1886c1379625SJustin T. Gibbs error = zap_remove(mos, 1887c1379625SJustin T. Gibbs dsl_dir_phys(dd->dd_parent)->dd_child_dir_zapobj, 1888fa9e4066Sahrens dd->dd_myname, tx); 18893b2aab18SMatthew Ahrens ASSERT0(error); 1890fa9e4066Sahrens 18913b2aab18SMatthew Ahrens (void) strcpy(dd->dd_myname, mynewname); 18923b2aab18SMatthew Ahrens dsl_dir_rele(dd->dd_parent, dd); 1893c1379625SJustin T. Gibbs dsl_dir_phys(dd)->dd_parent_obj = newparent->dd_object; 18943b2aab18SMatthew Ahrens VERIFY0(dsl_dir_hold_obj(dp, 18953b2aab18SMatthew Ahrens newparent->dd_object, NULL, dd, &dd->dd_parent)); 1896fa9e4066Sahrens 1897fa9e4066Sahrens /* add to new parent zapobj */ 1898c1379625SJustin T. Gibbs VERIFY0(zap_add(mos, dsl_dir_phys(newparent)->dd_child_dir_zapobj, 18993b2aab18SMatthew Ahrens dd->dd_myname, 8, 1, &dd->dd_object, tx)); 19003b2aab18SMatthew Ahrens 19013b2aab18SMatthew Ahrens dsl_prop_notify_all(dd); 1902ecd6cf80Smarks 19033b2aab18SMatthew Ahrens dsl_dir_rele(newparent, FTAG); 19043b2aab18SMatthew Ahrens dsl_dir_rele(dd, FTAG); 19051d452cf5Sahrens } 1906fa9e4066Sahrens 19071d452cf5Sahrens int 19083b2aab18SMatthew Ahrens dsl_dir_rename(const char *oldname, const char *newname) 19091d452cf5Sahrens { 19103b2aab18SMatthew Ahrens dsl_dir_rename_arg_t ddra; 19111d452cf5Sahrens 19123b2aab18SMatthew Ahrens ddra.ddra_oldname = oldname; 19133b2aab18SMatthew Ahrens ddra.ddra_newname = newname; 1914a2afb611SJerry Jelinek ddra.ddra_cred = CRED(); 19151d452cf5Sahrens 19163b2aab18SMatthew Ahrens return (dsl_sync_task(oldname, 19177d46dc6cSMatthew Ahrens dsl_dir_rename_check, dsl_dir_rename_sync, &ddra, 19187d46dc6cSMatthew Ahrens 3, ZFS_SPACE_CHECK_RESERVED)); 1919fa9e4066Sahrens } 192099653d4eSeschrock 192199653d4eSeschrock int 1922a2afb611SJerry Jelinek dsl_dir_transfer_possible(dsl_dir_t *sdd, dsl_dir_t *tdd, 1923a2afb611SJerry Jelinek uint64_t fs_cnt, uint64_t ss_cnt, uint64_t space, cred_t *cr) 192499653d4eSeschrock { 192599653d4eSeschrock dsl_dir_t *ancestor; 192699653d4eSeschrock int64_t adelta; 192799653d4eSeschrock uint64_t avail; 1928a2afb611SJerry Jelinek int err; 192999653d4eSeschrock 193099653d4eSeschrock ancestor = closest_common_ancestor(sdd, tdd); 193199653d4eSeschrock adelta = would_change(sdd, -space, ancestor); 193299653d4eSeschrock avail = dsl_dir_space_available(tdd, ancestor, adelta, FALSE); 193399653d4eSeschrock if (avail < space) 1934be6fd75aSMatthew Ahrens return (SET_ERROR(ENOSPC)); 193599653d4eSeschrock 1936a2afb611SJerry Jelinek err = dsl_fs_ss_limit_check(tdd, fs_cnt, ZFS_PROP_FILESYSTEM_LIMIT, 1937a2afb611SJerry Jelinek ancestor, cr); 1938a2afb611SJerry Jelinek if (err != 0) 1939a2afb611SJerry Jelinek return (err); 1940a2afb611SJerry Jelinek err = dsl_fs_ss_limit_check(tdd, ss_cnt, ZFS_PROP_SNAPSHOT_LIMIT, 1941a2afb611SJerry Jelinek ancestor, cr); 1942a2afb611SJerry Jelinek if (err != 0) 1943a2afb611SJerry Jelinek return (err); 1944a2afb611SJerry Jelinek 194599653d4eSeschrock return (0); 194699653d4eSeschrock } 194771eb0538SChris Kirby 194871eb0538SChris Kirby timestruc_t 194971eb0538SChris Kirby dsl_dir_snap_cmtime(dsl_dir_t *dd) 195071eb0538SChris Kirby { 195171eb0538SChris Kirby timestruc_t t; 195271eb0538SChris Kirby 195371eb0538SChris Kirby mutex_enter(&dd->dd_lock); 195471eb0538SChris Kirby t = dd->dd_snap_cmtime; 195571eb0538SChris Kirby mutex_exit(&dd->dd_lock); 195671eb0538SChris Kirby 195771eb0538SChris Kirby return (t); 195871eb0538SChris Kirby } 195971eb0538SChris Kirby 196071eb0538SChris Kirby void 196171eb0538SChris Kirby dsl_dir_snap_cmtime_update(dsl_dir_t *dd) 196271eb0538SChris Kirby { 196371eb0538SChris Kirby timestruc_t t; 196471eb0538SChris Kirby 196571eb0538SChris Kirby gethrestime(&t); 196671eb0538SChris Kirby mutex_enter(&dd->dd_lock); 196771eb0538SChris Kirby dd->dd_snap_cmtime = t; 196871eb0538SChris Kirby mutex_exit(&dd->dd_lock); 196971eb0538SChris Kirby } 19702acef22dSMatthew Ahrens 19712acef22dSMatthew Ahrens void 19722acef22dSMatthew Ahrens dsl_dir_zapify(dsl_dir_t *dd, dmu_tx_t *tx) 19732acef22dSMatthew Ahrens { 19742acef22dSMatthew Ahrens objset_t *mos = dd->dd_pool->dp_meta_objset; 19752acef22dSMatthew Ahrens dmu_object_zapify(mos, dd->dd_object, DMU_OT_DSL_DIR, tx); 19762acef22dSMatthew Ahrens } 1977a2afb611SJerry Jelinek 1978a2afb611SJerry Jelinek boolean_t 1979a2afb611SJerry Jelinek dsl_dir_is_zapified(dsl_dir_t *dd) 1980a2afb611SJerry Jelinek { 1981a2afb611SJerry Jelinek dmu_object_info_t doi; 1982a2afb611SJerry Jelinek 1983a2afb611SJerry Jelinek dmu_object_info_from_db(dd->dd_dbuf, &doi); 1984a2afb611SJerry Jelinek return (doi.doi_type == DMU_OTN_ZAP_METADATA); 1985a2afb611SJerry Jelinek } 1986