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 /* 2206e0070dSMark Shellenbaum * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. 238df0bcf0SPaul Dagnelie * Copyright (c) 2012, 2016 by Delphix. All rights reserved. 24aad02571SSaso Kiselkov * Copyright (c) 2013 by Saso Kiselkov. All rights reserved. 25a2afb611SJerry Jelinek * Copyright (c) 2013, 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. 2812380e1eSArne Jansen * Copyright (c) 2015, STRATO AG, Inc. All rights reserved. 29c3d26abcSMatthew Ahrens * Copyright (c) 2014 Integros [integros.com] 30fa9e4066Sahrens */ 31fa9e4066Sahrens 3255da60b9SMark J Musante /* Portions Copyright 2010 Robert Milkowski */ 3355da60b9SMark J Musante 34ecd6cf80Smarks #include <sys/cred.h> 35fa9e4066Sahrens #include <sys/zfs_context.h> 36fa9e4066Sahrens #include <sys/dmu_objset.h> 37fa9e4066Sahrens #include <sys/dsl_dir.h> 38fa9e4066Sahrens #include <sys/dsl_dataset.h> 39fa9e4066Sahrens #include <sys/dsl_prop.h> 40fa9e4066Sahrens #include <sys/dsl_pool.h> 411d452cf5Sahrens #include <sys/dsl_synctask.h> 42ecd6cf80Smarks #include <sys/dsl_deleg.h> 43fa9e4066Sahrens #include <sys/dnode.h> 44fa9e4066Sahrens #include <sys/dbuf.h> 45a2eea2e1Sahrens #include <sys/zvol.h> 46fa9e4066Sahrens #include <sys/dmu_tx.h> 47fa9e4066Sahrens #include <sys/zap.h> 48fa9e4066Sahrens #include <sys/zil.h> 49fa9e4066Sahrens #include <sys/dmu_impl.h> 50ecd6cf80Smarks #include <sys/zfs_ioctl.h> 510a586ceaSMark Shellenbaum #include <sys/sa.h> 5299d5e173STim Haley #include <sys/zfs_onexit.h> 533b2aab18SMatthew Ahrens #include <sys/dsl_destroy.h> 5412380e1eSArne Jansen #include <sys/vdev.h> 55fa9e4066Sahrens 56744947dcSTom Erickson /* 57744947dcSTom Erickson * Needed to close a window in dnode_move() that allows the objset to be freed 58744947dcSTom Erickson * before it can be safely accessed. 59744947dcSTom Erickson */ 60744947dcSTom Erickson krwlock_t os_lock; 61744947dcSTom Erickson 6212380e1eSArne Jansen /* 6312380e1eSArne Jansen * Tunable to overwrite the maximum number of threads for the parallization 6412380e1eSArne Jansen * of dmu_objset_find_dp, needed to speed up the import of pools with many 6512380e1eSArne Jansen * datasets. 6612380e1eSArne Jansen * Default is 4 times the number of leaf vdevs. 6712380e1eSArne Jansen */ 6812380e1eSArne Jansen int dmu_find_threads = 0; 6912380e1eSArne Jansen 70*af346df5SNed Bass /* 71*af346df5SNed Bass * Backfill lower metadnode objects after this many have been freed. 72*af346df5SNed Bass * Backfilling negatively impacts object creation rates, so only do it 73*af346df5SNed Bass * if there are enough holes to fill. 74*af346df5SNed Bass */ 75*af346df5SNed Bass int dmu_rescan_dnode_threshold = 131072; 76*af346df5SNed Bass 7712380e1eSArne Jansen static void dmu_objset_find_dp_cb(void *arg); 7812380e1eSArne Jansen 79744947dcSTom Erickson void 80744947dcSTom Erickson dmu_objset_init(void) 81744947dcSTom Erickson { 82744947dcSTom Erickson rw_init(&os_lock, NULL, RW_DEFAULT, NULL); 83744947dcSTom Erickson } 84744947dcSTom Erickson 85744947dcSTom Erickson void 86744947dcSTom Erickson dmu_objset_fini(void) 87744947dcSTom Erickson { 88744947dcSTom Erickson rw_destroy(&os_lock); 89744947dcSTom Erickson } 90744947dcSTom Erickson 91fa9e4066Sahrens spa_t * 92fa9e4066Sahrens dmu_objset_spa(objset_t *os) 93fa9e4066Sahrens { 94503ad85cSMatthew Ahrens return (os->os_spa); 95fa9e4066Sahrens } 96fa9e4066Sahrens 97fa9e4066Sahrens zilog_t * 98fa9e4066Sahrens dmu_objset_zil(objset_t *os) 99fa9e4066Sahrens { 100503ad85cSMatthew Ahrens return (os->os_zil); 101fa9e4066Sahrens } 102fa9e4066Sahrens 103fa9e4066Sahrens dsl_pool_t * 104fa9e4066Sahrens dmu_objset_pool(objset_t *os) 105fa9e4066Sahrens { 106fa9e4066Sahrens dsl_dataset_t *ds; 107fa9e4066Sahrens 108503ad85cSMatthew Ahrens if ((ds = os->os_dsl_dataset) != NULL && ds->ds_dir) 109fa9e4066Sahrens return (ds->ds_dir->dd_pool); 110fa9e4066Sahrens else 111503ad85cSMatthew Ahrens return (spa_get_dsl(os->os_spa)); 112fa9e4066Sahrens } 113fa9e4066Sahrens 114fa9e4066Sahrens dsl_dataset_t * 115fa9e4066Sahrens dmu_objset_ds(objset_t *os) 116fa9e4066Sahrens { 117503ad85cSMatthew Ahrens return (os->os_dsl_dataset); 118fa9e4066Sahrens } 119fa9e4066Sahrens 120fa9e4066Sahrens dmu_objset_type_t 121fa9e4066Sahrens dmu_objset_type(objset_t *os) 122fa9e4066Sahrens { 123503ad85cSMatthew Ahrens return (os->os_phys->os_type); 124fa9e4066Sahrens } 125fa9e4066Sahrens 126fa9e4066Sahrens void 127fa9e4066Sahrens dmu_objset_name(objset_t *os, char *buf) 128fa9e4066Sahrens { 129503ad85cSMatthew Ahrens dsl_dataset_name(os->os_dsl_dataset, buf); 130fa9e4066Sahrens } 131fa9e4066Sahrens 132fa9e4066Sahrens uint64_t 133fa9e4066Sahrens dmu_objset_id(objset_t *os) 134fa9e4066Sahrens { 135503ad85cSMatthew Ahrens dsl_dataset_t *ds = os->os_dsl_dataset; 136fa9e4066Sahrens 137fa9e4066Sahrens return (ds ? ds->ds_object : 0); 138fa9e4066Sahrens } 139fa9e4066Sahrens 140edf345e6SMatthew Ahrens zfs_sync_type_t 14155da60b9SMark J Musante dmu_objset_syncprop(objset_t *os) 14255da60b9SMark J Musante { 14355da60b9SMark J Musante return (os->os_sync); 14455da60b9SMark J Musante } 14555da60b9SMark J Musante 146edf345e6SMatthew Ahrens zfs_logbias_op_t 147e09fa4daSNeil Perrin dmu_objset_logbias(objset_t *os) 148e09fa4daSNeil Perrin { 149e09fa4daSNeil Perrin return (os->os_logbias); 150e09fa4daSNeil Perrin } 151e09fa4daSNeil Perrin 152fa9e4066Sahrens static void 153fa9e4066Sahrens checksum_changed_cb(void *arg, uint64_t newval) 154fa9e4066Sahrens { 155503ad85cSMatthew Ahrens objset_t *os = arg; 156fa9e4066Sahrens 157fa9e4066Sahrens /* 158fa9e4066Sahrens * Inheritance should have been done by now. 159fa9e4066Sahrens */ 160fa9e4066Sahrens ASSERT(newval != ZIO_CHECKSUM_INHERIT); 161fa9e4066Sahrens 162503ad85cSMatthew Ahrens os->os_checksum = zio_checksum_select(newval, ZIO_CHECKSUM_ON_VALUE); 163fa9e4066Sahrens } 164fa9e4066Sahrens 165fa9e4066Sahrens static void 166fa9e4066Sahrens compression_changed_cb(void *arg, uint64_t newval) 167fa9e4066Sahrens { 168503ad85cSMatthew Ahrens objset_t *os = arg; 169fa9e4066Sahrens 170fa9e4066Sahrens /* 171fa9e4066Sahrens * Inheritance and range checking should have been done by now. 172fa9e4066Sahrens */ 173fa9e4066Sahrens ASSERT(newval != ZIO_COMPRESS_INHERIT); 174fa9e4066Sahrens 175db1741f5SJustin T. Gibbs os->os_compress = zio_compress_select(os->os_spa, newval, 176db1741f5SJustin T. Gibbs ZIO_COMPRESS_ON); 177fa9e4066Sahrens } 178fa9e4066Sahrens 179d0ad202dSahrens static void 180d0ad202dSahrens copies_changed_cb(void *arg, uint64_t newval) 181d0ad202dSahrens { 182503ad85cSMatthew Ahrens objset_t *os = arg; 183d0ad202dSahrens 184d0ad202dSahrens /* 185d0ad202dSahrens * Inheritance and range checking should have been done by now. 186d0ad202dSahrens */ 187d0ad202dSahrens ASSERT(newval > 0); 188503ad85cSMatthew Ahrens ASSERT(newval <= spa_max_replication(os->os_spa)); 189d0ad202dSahrens 190503ad85cSMatthew Ahrens os->os_copies = newval; 191d0ad202dSahrens } 192d0ad202dSahrens 193b24ab676SJeff Bonwick static void 194b24ab676SJeff Bonwick dedup_changed_cb(void *arg, uint64_t newval) 195b24ab676SJeff Bonwick { 196b24ab676SJeff Bonwick objset_t *os = arg; 197b24ab676SJeff Bonwick spa_t *spa = os->os_spa; 198b24ab676SJeff Bonwick enum zio_checksum checksum; 199b24ab676SJeff Bonwick 200b24ab676SJeff Bonwick /* 201b24ab676SJeff Bonwick * Inheritance should have been done by now. 202b24ab676SJeff Bonwick */ 203b24ab676SJeff Bonwick ASSERT(newval != ZIO_CHECKSUM_INHERIT); 204b24ab676SJeff Bonwick 205b24ab676SJeff Bonwick checksum = zio_checksum_dedup_select(spa, newval, ZIO_CHECKSUM_OFF); 206b24ab676SJeff Bonwick 207b24ab676SJeff Bonwick os->os_dedup_checksum = checksum & ZIO_CHECKSUM_MASK; 208b24ab676SJeff Bonwick os->os_dedup_verify = !!(checksum & ZIO_CHECKSUM_VERIFY); 209b24ab676SJeff Bonwick } 210b24ab676SJeff Bonwick 2113baa08fcSek static void 2123baa08fcSek primary_cache_changed_cb(void *arg, uint64_t newval) 2133baa08fcSek { 214503ad85cSMatthew Ahrens objset_t *os = arg; 2153baa08fcSek 2163baa08fcSek /* 2173baa08fcSek * Inheritance and range checking should have been done by now. 2183baa08fcSek */ 2193baa08fcSek ASSERT(newval == ZFS_CACHE_ALL || newval == ZFS_CACHE_NONE || 2203baa08fcSek newval == ZFS_CACHE_METADATA); 2213baa08fcSek 222503ad85cSMatthew Ahrens os->os_primary_cache = newval; 2233baa08fcSek } 2243baa08fcSek 2253baa08fcSek static void 2263baa08fcSek secondary_cache_changed_cb(void *arg, uint64_t newval) 2273baa08fcSek { 228503ad85cSMatthew Ahrens objset_t *os = arg; 2293baa08fcSek 2303baa08fcSek /* 2313baa08fcSek * Inheritance and range checking should have been done by now. 2323baa08fcSek */ 2333baa08fcSek ASSERT(newval == ZFS_CACHE_ALL || newval == ZFS_CACHE_NONE || 2343baa08fcSek newval == ZFS_CACHE_METADATA); 2353baa08fcSek 236503ad85cSMatthew Ahrens os->os_secondary_cache = newval; 2373baa08fcSek } 2383baa08fcSek 23955da60b9SMark J Musante static void 24055da60b9SMark J Musante sync_changed_cb(void *arg, uint64_t newval) 24155da60b9SMark J Musante { 24255da60b9SMark J Musante objset_t *os = arg; 24355da60b9SMark J Musante 24455da60b9SMark J Musante /* 24555da60b9SMark J Musante * Inheritance and range checking should have been done by now. 24655da60b9SMark J Musante */ 24755da60b9SMark J Musante ASSERT(newval == ZFS_SYNC_STANDARD || newval == ZFS_SYNC_ALWAYS || 24855da60b9SMark J Musante newval == ZFS_SYNC_DISABLED); 24955da60b9SMark J Musante 25055da60b9SMark J Musante os->os_sync = newval; 25155da60b9SMark J Musante if (os->os_zil) 25255da60b9SMark J Musante zil_set_sync(os->os_zil, newval); 25355da60b9SMark J Musante } 25455da60b9SMark J Musante 255edf345e6SMatthew Ahrens static void 256edf345e6SMatthew Ahrens redundant_metadata_changed_cb(void *arg, uint64_t newval) 257edf345e6SMatthew Ahrens { 258edf345e6SMatthew Ahrens objset_t *os = arg; 259edf345e6SMatthew Ahrens 260edf345e6SMatthew Ahrens /* 261edf345e6SMatthew Ahrens * Inheritance and range checking should have been done by now. 262edf345e6SMatthew Ahrens */ 263edf345e6SMatthew Ahrens ASSERT(newval == ZFS_REDUNDANT_METADATA_ALL || 264edf345e6SMatthew Ahrens newval == ZFS_REDUNDANT_METADATA_MOST); 265edf345e6SMatthew Ahrens 266edf345e6SMatthew Ahrens os->os_redundant_metadata = newval; 267edf345e6SMatthew Ahrens } 268edf345e6SMatthew Ahrens 269e09fa4daSNeil Perrin static void 270e09fa4daSNeil Perrin logbias_changed_cb(void *arg, uint64_t newval) 271e09fa4daSNeil Perrin { 272e09fa4daSNeil Perrin objset_t *os = arg; 273e09fa4daSNeil Perrin 274e09fa4daSNeil Perrin ASSERT(newval == ZFS_LOGBIAS_LATENCY || 275e09fa4daSNeil Perrin newval == ZFS_LOGBIAS_THROUGHPUT); 276e09fa4daSNeil Perrin os->os_logbias = newval; 277e09fa4daSNeil Perrin if (os->os_zil) 278e09fa4daSNeil Perrin zil_set_logbias(os->os_zil, newval); 279e09fa4daSNeil Perrin } 280e09fa4daSNeil Perrin 281b5152584SMatthew Ahrens static void 282b5152584SMatthew Ahrens recordsize_changed_cb(void *arg, uint64_t newval) 283b5152584SMatthew Ahrens { 284b5152584SMatthew Ahrens objset_t *os = arg; 285b5152584SMatthew Ahrens 286b5152584SMatthew Ahrens os->os_recordsize = newval; 287b5152584SMatthew Ahrens } 288b5152584SMatthew Ahrens 289fa9e4066Sahrens void 290fa9e4066Sahrens dmu_objset_byteswap(void *buf, size_t size) 291fa9e4066Sahrens { 292fa9e4066Sahrens objset_phys_t *osp = buf; 293fa9e4066Sahrens 29414843421SMatthew Ahrens ASSERT(size == OBJSET_OLD_PHYS_SIZE || size == sizeof (objset_phys_t)); 295fa9e4066Sahrens dnode_byteswap(&osp->os_meta_dnode); 296fa9e4066Sahrens byteswap_uint64_array(&osp->os_zil_header, sizeof (zil_header_t)); 297fa9e4066Sahrens osp->os_type = BSWAP_64(osp->os_type); 29814843421SMatthew Ahrens osp->os_flags = BSWAP_64(osp->os_flags); 29914843421SMatthew Ahrens if (size == sizeof (objset_phys_t)) { 30014843421SMatthew Ahrens dnode_byteswap(&osp->os_userused_dnode); 30114843421SMatthew Ahrens dnode_byteswap(&osp->os_groupused_dnode); 30214843421SMatthew Ahrens } 303fa9e4066Sahrens } 304fa9e4066Sahrens 305ea8dc4b6Seschrock int 306ea8dc4b6Seschrock dmu_objset_open_impl(spa_t *spa, dsl_dataset_t *ds, blkptr_t *bp, 307503ad85cSMatthew Ahrens objset_t **osp) 308fa9e4066Sahrens { 309503ad85cSMatthew Ahrens objset_t *os; 310088f3894Sahrens int i, err; 311fa9e4066Sahrens 31291ebeef5Sahrens ASSERT(ds == NULL || MUTEX_HELD(&ds->ds_opening_lock)); 31391ebeef5Sahrens 314503ad85cSMatthew Ahrens os = kmem_zalloc(sizeof (objset_t), KM_SLEEP); 315503ad85cSMatthew Ahrens os->os_dsl_dataset = ds; 316503ad85cSMatthew Ahrens os->os_spa = spa; 317503ad85cSMatthew Ahrens os->os_rootbp = bp; 318503ad85cSMatthew Ahrens if (!BP_IS_HOLE(os->os_rootbp)) { 3197adb730bSGeorge Wilson arc_flags_t aflags = ARC_FLAG_WAIT; 3207802d7bfSMatthew Ahrens zbookmark_phys_t zb; 321b24ab676SJeff Bonwick SET_BOOKMARK(&zb, ds ? ds->ds_object : DMU_META_OBJSET, 322b24ab676SJeff Bonwick ZB_ROOT_OBJECT, ZB_ROOT_LEVEL, ZB_ROOT_BLKID); 323b24ab676SJeff Bonwick 324503ad85cSMatthew Ahrens if (DMU_OS_IS_L2CACHEABLE(os)) 3257adb730bSGeorge Wilson aflags |= ARC_FLAG_L2CACHE; 326ea8dc4b6Seschrock 327503ad85cSMatthew Ahrens dprintf_bp(os->os_rootbp, "reading %s", ""); 3281b912ec7SGeorge Wilson err = arc_read(NULL, spa, os->os_rootbp, 329503ad85cSMatthew Ahrens arc_getbuf_func, &os->os_phys_buf, 33013506d1eSmaybee ZIO_PRIORITY_SYNC_READ, ZIO_FLAG_CANFAIL, &aflags, &zb); 3313b2aab18SMatthew Ahrens if (err != 0) { 332503ad85cSMatthew Ahrens kmem_free(os, sizeof (objset_t)); 333b87f3af3Sperrin /* convert checksum errors into IO errors */ 334b87f3af3Sperrin if (err == ECKSUM) 335be6fd75aSMatthew Ahrens err = SET_ERROR(EIO); 336ea8dc4b6Seschrock return (err); 337ea8dc4b6Seschrock } 33814843421SMatthew Ahrens 33914843421SMatthew Ahrens /* Increase the blocksize if we are permitted. */ 34014843421SMatthew Ahrens if (spa_version(spa) >= SPA_VERSION_USERSPACE && 341503ad85cSMatthew Ahrens arc_buf_size(os->os_phys_buf) < sizeof (objset_phys_t)) { 342dcbf3bd6SGeorge Wilson arc_buf_t *buf = arc_alloc_buf(spa, 343503ad85cSMatthew Ahrens sizeof (objset_phys_t), &os->os_phys_buf, 34414843421SMatthew Ahrens ARC_BUFC_METADATA); 34514843421SMatthew Ahrens bzero(buf->b_data, sizeof (objset_phys_t)); 346503ad85cSMatthew Ahrens bcopy(os->os_phys_buf->b_data, buf->b_data, 347503ad85cSMatthew Ahrens arc_buf_size(os->os_phys_buf)); 348dcbf3bd6SGeorge Wilson arc_buf_destroy(os->os_phys_buf, &os->os_phys_buf); 349503ad85cSMatthew Ahrens os->os_phys_buf = buf; 35014843421SMatthew Ahrens } 35114843421SMatthew Ahrens 352503ad85cSMatthew Ahrens os->os_phys = os->os_phys_buf->b_data; 353503ad85cSMatthew Ahrens os->os_flags = os->os_phys->os_flags; 354fa9e4066Sahrens } else { 35514843421SMatthew Ahrens int size = spa_version(spa) >= SPA_VERSION_USERSPACE ? 35614843421SMatthew Ahrens sizeof (objset_phys_t) : OBJSET_OLD_PHYS_SIZE; 357dcbf3bd6SGeorge Wilson os->os_phys_buf = arc_alloc_buf(spa, size, 358503ad85cSMatthew Ahrens &os->os_phys_buf, ARC_BUFC_METADATA); 359503ad85cSMatthew Ahrens os->os_phys = os->os_phys_buf->b_data; 360503ad85cSMatthew Ahrens bzero(os->os_phys, size); 361fa9e4066Sahrens } 362fa9e4066Sahrens 363fa9e4066Sahrens /* 364fa9e4066Sahrens * Note: the changed_cb will be called once before the register 365fa9e4066Sahrens * func returns, thus changing the checksum/compression from the 3663baa08fcSek * default (fletcher2/off). Snapshots don't need to know about 3673baa08fcSek * checksum/compression/copies. 368fa9e4066Sahrens */ 3695d7b4d43SMatthew Ahrens if (ds != NULL) { 3709c3fd121SMatthew Ahrens boolean_t needlock = B_FALSE; 3719c3fd121SMatthew Ahrens 3729c3fd121SMatthew Ahrens /* 3739c3fd121SMatthew Ahrens * Note: it's valid to open the objset if the dataset is 3749c3fd121SMatthew Ahrens * long-held, in which case the pool_config lock will not 3759c3fd121SMatthew Ahrens * be held. 3769c3fd121SMatthew Ahrens */ 3779c3fd121SMatthew Ahrens if (!dsl_pool_config_held(dmu_objset_pool(os))) { 3789c3fd121SMatthew Ahrens needlock = B_TRUE; 3799c3fd121SMatthew Ahrens dsl_pool_config_enter(dmu_objset_pool(os), FTAG); 3809c3fd121SMatthew Ahrens } 3813b2aab18SMatthew Ahrens err = dsl_prop_register(ds, 3823b2aab18SMatthew Ahrens zfs_prop_to_name(ZFS_PROP_PRIMARYCACHE), 383503ad85cSMatthew Ahrens primary_cache_changed_cb, os); 3843b2aab18SMatthew Ahrens if (err == 0) { 3853b2aab18SMatthew Ahrens err = dsl_prop_register(ds, 3863b2aab18SMatthew Ahrens zfs_prop_to_name(ZFS_PROP_SECONDARYCACHE), 387503ad85cSMatthew Ahrens secondary_cache_changed_cb, os); 3883b2aab18SMatthew Ahrens } 389bc9014e6SJustin Gibbs if (!ds->ds_is_snapshot) { 3903b2aab18SMatthew Ahrens if (err == 0) { 3913b2aab18SMatthew Ahrens err = dsl_prop_register(ds, 3923b2aab18SMatthew Ahrens zfs_prop_to_name(ZFS_PROP_CHECKSUM), 393503ad85cSMatthew Ahrens checksum_changed_cb, os); 3943b2aab18SMatthew Ahrens } 3953b2aab18SMatthew Ahrens if (err == 0) { 3963b2aab18SMatthew Ahrens err = dsl_prop_register(ds, 3973b2aab18SMatthew Ahrens zfs_prop_to_name(ZFS_PROP_COMPRESSION), 398503ad85cSMatthew Ahrens compression_changed_cb, os); 3993b2aab18SMatthew Ahrens } 4003b2aab18SMatthew Ahrens if (err == 0) { 4013b2aab18SMatthew Ahrens err = dsl_prop_register(ds, 4023b2aab18SMatthew Ahrens zfs_prop_to_name(ZFS_PROP_COPIES), 403503ad85cSMatthew Ahrens copies_changed_cb, os); 4043b2aab18SMatthew Ahrens } 4053b2aab18SMatthew Ahrens if (err == 0) { 4063b2aab18SMatthew Ahrens err = dsl_prop_register(ds, 4073b2aab18SMatthew Ahrens zfs_prop_to_name(ZFS_PROP_DEDUP), 408b24ab676SJeff Bonwick dedup_changed_cb, os); 4093b2aab18SMatthew Ahrens } 4103b2aab18SMatthew Ahrens if (err == 0) { 4113b2aab18SMatthew Ahrens err = dsl_prop_register(ds, 4123b2aab18SMatthew Ahrens zfs_prop_to_name(ZFS_PROP_LOGBIAS), 413e09fa4daSNeil Perrin logbias_changed_cb, os); 4143b2aab18SMatthew Ahrens } 4153b2aab18SMatthew Ahrens if (err == 0) { 4163b2aab18SMatthew Ahrens err = dsl_prop_register(ds, 4173b2aab18SMatthew Ahrens zfs_prop_to_name(ZFS_PROP_SYNC), 41855da60b9SMark J Musante sync_changed_cb, os); 4193b2aab18SMatthew Ahrens } 420edf345e6SMatthew Ahrens if (err == 0) { 421edf345e6SMatthew Ahrens err = dsl_prop_register(ds, 422edf345e6SMatthew Ahrens zfs_prop_to_name( 423edf345e6SMatthew Ahrens ZFS_PROP_REDUNDANT_METADATA), 424edf345e6SMatthew Ahrens redundant_metadata_changed_cb, os); 425edf345e6SMatthew Ahrens } 426b5152584SMatthew Ahrens if (err == 0) { 427b5152584SMatthew Ahrens err = dsl_prop_register(ds, 428b5152584SMatthew Ahrens zfs_prop_to_name(ZFS_PROP_RECORDSIZE), 429b5152584SMatthew Ahrens recordsize_changed_cb, os); 430b5152584SMatthew Ahrens } 4313baa08fcSek } 4329c3fd121SMatthew Ahrens if (needlock) 4339c3fd121SMatthew Ahrens dsl_pool_config_exit(dmu_objset_pool(os), FTAG); 4343b2aab18SMatthew Ahrens if (err != 0) { 435dcbf3bd6SGeorge Wilson arc_buf_destroy(os->os_phys_buf, &os->os_phys_buf); 436503ad85cSMatthew Ahrens kmem_free(os, sizeof (objset_t)); 437ea8dc4b6Seschrock return (err); 438ea8dc4b6Seschrock } 4395d7b4d43SMatthew Ahrens } else { 440fa9e4066Sahrens /* It's the meta-objset. */ 441503ad85cSMatthew Ahrens os->os_checksum = ZIO_CHECKSUM_FLETCHER_4; 442db1741f5SJustin T. Gibbs os->os_compress = ZIO_COMPRESS_ON; 443503ad85cSMatthew Ahrens os->os_copies = spa_max_replication(spa); 444b24ab676SJeff Bonwick os->os_dedup_checksum = ZIO_CHECKSUM_OFF; 445edf345e6SMatthew Ahrens os->os_dedup_verify = B_FALSE; 446edf345e6SMatthew Ahrens os->os_logbias = ZFS_LOGBIAS_LATENCY; 447edf345e6SMatthew Ahrens os->os_sync = ZFS_SYNC_STANDARD; 448503ad85cSMatthew Ahrens os->os_primary_cache = ZFS_CACHE_ALL; 449503ad85cSMatthew Ahrens os->os_secondary_cache = ZFS_CACHE_ALL; 450fa9e4066Sahrens } 451fa9e4066Sahrens 452bc9014e6SJustin Gibbs if (ds == NULL || !ds->ds_is_snapshot) 4536e0cbcaaSMatthew Ahrens os->os_zil_header = os->os_phys->os_zil_header; 454503ad85cSMatthew Ahrens os->os_zil = zil_alloc(os, &os->os_zil_header); 455fa9e4066Sahrens 456fa9e4066Sahrens for (i = 0; i < TXG_SIZE; i++) { 457503ad85cSMatthew Ahrens list_create(&os->os_dirty_dnodes[i], sizeof (dnode_t), 458fa9e4066Sahrens offsetof(dnode_t, dn_dirty_link[i])); 459503ad85cSMatthew Ahrens list_create(&os->os_free_dnodes[i], sizeof (dnode_t), 460fa9e4066Sahrens offsetof(dnode_t, dn_dirty_link[i])); 461fa9e4066Sahrens } 462503ad85cSMatthew Ahrens list_create(&os->os_dnodes, sizeof (dnode_t), 463fa9e4066Sahrens offsetof(dnode_t, dn_link)); 464503ad85cSMatthew Ahrens list_create(&os->os_downgraded_dbufs, sizeof (dmu_buf_impl_t), 465fa9e4066Sahrens offsetof(dmu_buf_impl_t, db_link)); 466fa9e4066Sahrens 467503ad85cSMatthew Ahrens mutex_init(&os->os_lock, NULL, MUTEX_DEFAULT, NULL); 468503ad85cSMatthew Ahrens mutex_init(&os->os_obj_lock, NULL, MUTEX_DEFAULT, NULL); 469503ad85cSMatthew Ahrens mutex_init(&os->os_user_ptr_lock, NULL, MUTEX_DEFAULT, NULL); 470503ad85cSMatthew Ahrens 471bc9014e6SJustin Gibbs dnode_special_open(os, &os->os_phys->os_meta_dnode, 472bc9014e6SJustin Gibbs DMU_META_DNODE_OBJECT, &os->os_meta_dnode); 473503ad85cSMatthew Ahrens if (arc_buf_size(os->os_phys_buf) >= sizeof (objset_phys_t)) { 474bc9014e6SJustin Gibbs dnode_special_open(os, &os->os_phys->os_userused_dnode, 475bc9014e6SJustin Gibbs DMU_USERUSED_OBJECT, &os->os_userused_dnode); 476bc9014e6SJustin Gibbs dnode_special_open(os, &os->os_phys->os_groupused_dnode, 477bc9014e6SJustin Gibbs DMU_GROUPUSED_OBJECT, &os->os_groupused_dnode); 47814843421SMatthew Ahrens } 479fa9e4066Sahrens 480503ad85cSMatthew Ahrens *osp = os; 481ea8dc4b6Seschrock return (0); 482fa9e4066Sahrens } 483fa9e4066Sahrens 484503ad85cSMatthew Ahrens int 485503ad85cSMatthew Ahrens dmu_objset_from_ds(dsl_dataset_t *ds, objset_t **osp) 4863cb34c60Sahrens { 487503ad85cSMatthew Ahrens int err = 0; 4883cb34c60Sahrens 4899c3fd121SMatthew Ahrens /* 4909c3fd121SMatthew Ahrens * We shouldn't be doing anything with dsl_dataset_t's unless the 4919c3fd121SMatthew Ahrens * pool_config lock is held, or the dataset is long-held. 4929c3fd121SMatthew Ahrens */ 4939c3fd121SMatthew Ahrens ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool) || 4949c3fd121SMatthew Ahrens dsl_dataset_long_held(ds)); 4959c3fd121SMatthew Ahrens 4963cb34c60Sahrens mutex_enter(&ds->ds_opening_lock); 4975d7b4d43SMatthew Ahrens if (ds->ds_objset == NULL) { 4985d7b4d43SMatthew Ahrens objset_t *os; 499c166b69dSPaul Dagnelie rrw_enter(&ds->ds_bp_rwlock, RW_READER, FTAG); 5003cb34c60Sahrens err = dmu_objset_open_impl(dsl_dataset_get_spa(ds), 5015d7b4d43SMatthew Ahrens ds, dsl_dataset_get_blkptr(ds), &os); 502c166b69dSPaul Dagnelie rrw_exit(&ds->ds_bp_rwlock, FTAG); 5035d7b4d43SMatthew Ahrens 5045d7b4d43SMatthew Ahrens if (err == 0) { 5055d7b4d43SMatthew Ahrens mutex_enter(&ds->ds_lock); 5065d7b4d43SMatthew Ahrens ASSERT(ds->ds_objset == NULL); 5075d7b4d43SMatthew Ahrens ds->ds_objset = os; 5085d7b4d43SMatthew Ahrens mutex_exit(&ds->ds_lock); 5095d7b4d43SMatthew Ahrens } 5103cb34c60Sahrens } 5115d7b4d43SMatthew Ahrens *osp = ds->ds_objset; 5123cb34c60Sahrens mutex_exit(&ds->ds_opening_lock); 513503ad85cSMatthew Ahrens return (err); 5143cb34c60Sahrens } 5153cb34c60Sahrens 5163b2aab18SMatthew Ahrens /* 5173b2aab18SMatthew Ahrens * Holds the pool while the objset is held. Therefore only one objset 5183b2aab18SMatthew Ahrens * can be held at a time. 5193b2aab18SMatthew Ahrens */ 5203cb34c60Sahrens int 521503ad85cSMatthew Ahrens dmu_objset_hold(const char *name, void *tag, objset_t **osp) 5223cb34c60Sahrens { 5233b2aab18SMatthew Ahrens dsl_pool_t *dp; 524503ad85cSMatthew Ahrens dsl_dataset_t *ds; 5253cb34c60Sahrens int err; 5263cb34c60Sahrens 5273b2aab18SMatthew Ahrens err = dsl_pool_hold(name, tag, &dp); 5283b2aab18SMatthew Ahrens if (err != 0) 5293b2aab18SMatthew Ahrens return (err); 5303b2aab18SMatthew Ahrens err = dsl_dataset_hold(dp, name, tag, &ds); 5313b2aab18SMatthew Ahrens if (err != 0) { 5323b2aab18SMatthew Ahrens dsl_pool_rele(dp, tag); 533503ad85cSMatthew Ahrens return (err); 5343b2aab18SMatthew Ahrens } 535503ad85cSMatthew Ahrens 536503ad85cSMatthew Ahrens err = dmu_objset_from_ds(ds, osp); 5373b2aab18SMatthew Ahrens if (err != 0) { 538503ad85cSMatthew Ahrens dsl_dataset_rele(ds, tag); 5393b2aab18SMatthew Ahrens dsl_pool_rele(dp, tag); 5403b2aab18SMatthew Ahrens } 541503ad85cSMatthew Ahrens 5423cb34c60Sahrens return (err); 5433cb34c60Sahrens } 5443cb34c60Sahrens 54512380e1eSArne Jansen static int 54612380e1eSArne Jansen dmu_objset_own_impl(dsl_dataset_t *ds, dmu_objset_type_t type, 54712380e1eSArne Jansen boolean_t readonly, void *tag, objset_t **osp) 54812380e1eSArne Jansen { 54912380e1eSArne Jansen int err; 55012380e1eSArne Jansen 55112380e1eSArne Jansen err = dmu_objset_from_ds(ds, osp); 55212380e1eSArne Jansen if (err != 0) { 55312380e1eSArne Jansen dsl_dataset_disown(ds, tag); 55412380e1eSArne Jansen } else if (type != DMU_OST_ANY && type != (*osp)->os_phys->os_type) { 55512380e1eSArne Jansen dsl_dataset_disown(ds, tag); 55612380e1eSArne Jansen return (SET_ERROR(EINVAL)); 55712380e1eSArne Jansen } else if (!readonly && dsl_dataset_is_snapshot(ds)) { 55812380e1eSArne Jansen dsl_dataset_disown(ds, tag); 55912380e1eSArne Jansen return (SET_ERROR(EROFS)); 56012380e1eSArne Jansen } 56112380e1eSArne Jansen return (err); 56212380e1eSArne Jansen } 56312380e1eSArne Jansen 5643b2aab18SMatthew Ahrens /* 5653b2aab18SMatthew Ahrens * dsl_pool must not be held when this is called. 5663b2aab18SMatthew Ahrens * Upon successful return, there will be a longhold on the dataset, 5673b2aab18SMatthew Ahrens * and the dsl_pool will not be held. 5683b2aab18SMatthew Ahrens */ 569fa9e4066Sahrens int 570503ad85cSMatthew Ahrens dmu_objset_own(const char *name, dmu_objset_type_t type, 571503ad85cSMatthew Ahrens boolean_t readonly, void *tag, objset_t **osp) 572fa9e4066Sahrens { 5733b2aab18SMatthew Ahrens dsl_pool_t *dp; 574f18faf3fSek dsl_dataset_t *ds; 575f18faf3fSek int err; 576fa9e4066Sahrens 5773b2aab18SMatthew Ahrens err = dsl_pool_hold(name, FTAG, &dp); 5783b2aab18SMatthew Ahrens if (err != 0) 5793b2aab18SMatthew Ahrens return (err); 5803b2aab18SMatthew Ahrens err = dsl_dataset_own(dp, name, tag, &ds); 5813b2aab18SMatthew Ahrens if (err != 0) { 5823b2aab18SMatthew Ahrens dsl_pool_rele(dp, FTAG); 583fa9e4066Sahrens return (err); 5843b2aab18SMatthew Ahrens } 58512380e1eSArne Jansen err = dmu_objset_own_impl(ds, type, readonly, tag, osp); 5863b2aab18SMatthew Ahrens dsl_pool_rele(dp, FTAG); 58712380e1eSArne Jansen 5883cb34c60Sahrens return (err); 589fa9e4066Sahrens } 590fa9e4066Sahrens 59112380e1eSArne Jansen int 59212380e1eSArne Jansen dmu_objset_own_obj(dsl_pool_t *dp, uint64_t obj, dmu_objset_type_t type, 59312380e1eSArne Jansen boolean_t readonly, void *tag, objset_t **osp) 59412380e1eSArne Jansen { 59512380e1eSArne Jansen dsl_dataset_t *ds; 59612380e1eSArne Jansen int err; 59712380e1eSArne Jansen 59812380e1eSArne Jansen err = dsl_dataset_own_obj(dp, obj, tag, &ds); 59912380e1eSArne Jansen if (err != 0) 60012380e1eSArne Jansen return (err); 60112380e1eSArne Jansen 60212380e1eSArne Jansen return (dmu_objset_own_impl(ds, type, readonly, tag, osp)); 60312380e1eSArne Jansen } 60412380e1eSArne Jansen 605fa9e4066Sahrens void 606503ad85cSMatthew Ahrens dmu_objset_rele(objset_t *os, void *tag) 607fa9e4066Sahrens { 6083b2aab18SMatthew Ahrens dsl_pool_t *dp = dmu_objset_pool(os); 609503ad85cSMatthew Ahrens dsl_dataset_rele(os->os_dsl_dataset, tag); 6103b2aab18SMatthew Ahrens dsl_pool_rele(dp, tag); 611503ad85cSMatthew Ahrens } 612503ad85cSMatthew Ahrens 61391948b51SKeith M Wesolowski /* 61491948b51SKeith M Wesolowski * When we are called, os MUST refer to an objset associated with a dataset 61591948b51SKeith M Wesolowski * that is owned by 'tag'; that is, is held and long held by 'tag' and ds_owner 61691948b51SKeith M Wesolowski * == tag. We will then release and reacquire ownership of the dataset while 61791948b51SKeith M Wesolowski * holding the pool config_rwlock to avoid intervening namespace or ownership 61891948b51SKeith M Wesolowski * changes may occur. 61991948b51SKeith M Wesolowski * 62091948b51SKeith M Wesolowski * This exists solely to accommodate zfs_ioc_userspace_upgrade()'s desire to 62191948b51SKeith M Wesolowski * release the hold on its dataset and acquire a new one on the dataset of the 62291948b51SKeith M Wesolowski * same name so that it can be partially torn down and reconstructed. 62391948b51SKeith M Wesolowski */ 62491948b51SKeith M Wesolowski void 62591948b51SKeith M Wesolowski dmu_objset_refresh_ownership(objset_t *os, void *tag) 62691948b51SKeith M Wesolowski { 62791948b51SKeith M Wesolowski dsl_pool_t *dp; 62891948b51SKeith M Wesolowski dsl_dataset_t *ds, *newds; 6299adfa60dSMatthew Ahrens char name[ZFS_MAX_DATASET_NAME_LEN]; 63091948b51SKeith M Wesolowski 63191948b51SKeith M Wesolowski ds = os->os_dsl_dataset; 63291948b51SKeith M Wesolowski VERIFY3P(ds, !=, NULL); 63391948b51SKeith M Wesolowski VERIFY3P(ds->ds_owner, ==, tag); 63491948b51SKeith M Wesolowski VERIFY(dsl_dataset_long_held(ds)); 63591948b51SKeith M Wesolowski 63691948b51SKeith M Wesolowski dsl_dataset_name(ds, name); 63791948b51SKeith M Wesolowski dp = dmu_objset_pool(os); 63891948b51SKeith M Wesolowski dsl_pool_config_enter(dp, FTAG); 63991948b51SKeith M Wesolowski dmu_objset_disown(os, tag); 64091948b51SKeith M Wesolowski VERIFY0(dsl_dataset_own(dp, name, tag, &newds)); 64191948b51SKeith M Wesolowski VERIFY3P(newds, ==, os->os_dsl_dataset); 64291948b51SKeith M Wesolowski dsl_pool_config_exit(dp, FTAG); 64391948b51SKeith M Wesolowski } 64491948b51SKeith M Wesolowski 645503ad85cSMatthew Ahrens void 646503ad85cSMatthew Ahrens dmu_objset_disown(objset_t *os, void *tag) 647503ad85cSMatthew Ahrens { 648503ad85cSMatthew Ahrens dsl_dataset_disown(os->os_dsl_dataset, tag); 649fa9e4066Sahrens } 650fa9e4066Sahrens 6513b2aab18SMatthew Ahrens void 6521934e92fSmaybee dmu_objset_evict_dbufs(objset_t *os) 653ea8dc4b6Seschrock { 654bc9014e6SJustin Gibbs dnode_t dn_marker; 655ea8dc4b6Seschrock dnode_t *dn; 656c543ec06Sahrens 657503ad85cSMatthew Ahrens mutex_enter(&os->os_lock); 658bc9014e6SJustin Gibbs dn = list_head(&os->os_dnodes); 659bc9014e6SJustin Gibbs while (dn != NULL) { 660bc9014e6SJustin Gibbs /* 661bc9014e6SJustin Gibbs * Skip dnodes without holds. We have to do this dance 662bc9014e6SJustin Gibbs * because dnode_add_ref() only works if there is already a 663bc9014e6SJustin Gibbs * hold. If the dnode has no holds, then it has no dbufs. 664bc9014e6SJustin Gibbs */ 665bc9014e6SJustin Gibbs if (dnode_add_ref(dn, FTAG)) { 666bc9014e6SJustin Gibbs list_insert_after(&os->os_dnodes, dn, &dn_marker); 667bc9014e6SJustin Gibbs mutex_exit(&os->os_lock); 668c543ec06Sahrens 669bc9014e6SJustin Gibbs dnode_evict_dbufs(dn); 670bc9014e6SJustin Gibbs dnode_rele(dn, FTAG); 671c543ec06Sahrens 672bc9014e6SJustin Gibbs mutex_enter(&os->os_lock); 673bc9014e6SJustin Gibbs dn = list_next(&os->os_dnodes, &dn_marker); 674bc9014e6SJustin Gibbs list_remove(&os->os_dnodes, &dn_marker); 675bc9014e6SJustin Gibbs } else { 676bc9014e6SJustin Gibbs dn = list_next(&os->os_dnodes, dn); 677bc9014e6SJustin Gibbs } 678ea8dc4b6Seschrock } 679503ad85cSMatthew Ahrens mutex_exit(&os->os_lock); 680bc9014e6SJustin Gibbs 681bc9014e6SJustin Gibbs if (DMU_USERUSED_DNODE(os) != NULL) { 682bc9014e6SJustin Gibbs dnode_evict_dbufs(DMU_GROUPUSED_DNODE(os)); 683bc9014e6SJustin Gibbs dnode_evict_dbufs(DMU_USERUSED_DNODE(os)); 684bc9014e6SJustin Gibbs } 685bc9014e6SJustin Gibbs dnode_evict_dbufs(DMU_META_DNODE(os)); 686ea8dc4b6Seschrock } 687ea8dc4b6Seschrock 688bc9014e6SJustin Gibbs /* 689bc9014e6SJustin Gibbs * Objset eviction processing is split into into two pieces. 690bc9014e6SJustin Gibbs * The first marks the objset as evicting, evicts any dbufs that 691bc9014e6SJustin Gibbs * have a refcount of zero, and then queues up the objset for the 692bc9014e6SJustin Gibbs * second phase of eviction. Once os->os_dnodes has been cleared by 693bc9014e6SJustin Gibbs * dnode_buf_pageout()->dnode_destroy(), the second phase is executed. 694bc9014e6SJustin Gibbs * The second phase closes the special dnodes, dequeues the objset from 695bc9014e6SJustin Gibbs * the list of those undergoing eviction, and finally frees the objset. 696bc9014e6SJustin Gibbs * 697bc9014e6SJustin Gibbs * NOTE: Due to asynchronous eviction processing (invocation of 698bc9014e6SJustin Gibbs * dnode_buf_pageout()), it is possible for the meta dnode for the 699bc9014e6SJustin Gibbs * objset to have no holds even though os->os_dnodes is not empty. 700bc9014e6SJustin Gibbs */ 701fa9e4066Sahrens void 702503ad85cSMatthew Ahrens dmu_objset_evict(objset_t *os) 703fa9e4066Sahrens { 704503ad85cSMatthew Ahrens dsl_dataset_t *ds = os->os_dsl_dataset; 705fa9e4066Sahrens 706b24ab676SJeff Bonwick for (int t = 0; t < TXG_SIZE; t++) 707b24ab676SJeff Bonwick ASSERT(!dmu_objset_is_dirty(os, t)); 708fa9e4066Sahrens 70903bad06fSJustin Gibbs if (ds) 71003bad06fSJustin Gibbs dsl_prop_unregister_all(ds, os); 711fa9e4066Sahrens 7120a586ceaSMark Shellenbaum if (os->os_sa) 7130a586ceaSMark Shellenbaum sa_tear_down(os); 7140a586ceaSMark Shellenbaum 7153b2aab18SMatthew Ahrens dmu_objset_evict_dbufs(os); 716ea8dc4b6Seschrock 717bc9014e6SJustin Gibbs mutex_enter(&os->os_lock); 718bc9014e6SJustin Gibbs spa_evicting_os_register(os->os_spa, os); 719bc9014e6SJustin Gibbs if (list_is_empty(&os->os_dnodes)) { 720bc9014e6SJustin Gibbs mutex_exit(&os->os_lock); 721bc9014e6SJustin Gibbs dmu_objset_evict_done(os); 722bc9014e6SJustin Gibbs } else { 723bc9014e6SJustin Gibbs mutex_exit(&os->os_lock); 724bc9014e6SJustin Gibbs } 725bc9014e6SJustin Gibbs } 726bc9014e6SJustin Gibbs 727bc9014e6SJustin Gibbs void 728bc9014e6SJustin Gibbs dmu_objset_evict_done(objset_t *os) 729bc9014e6SJustin Gibbs { 730bc9014e6SJustin Gibbs ASSERT3P(list_head(&os->os_dnodes), ==, NULL); 731bc9014e6SJustin Gibbs 732744947dcSTom Erickson dnode_special_close(&os->os_meta_dnode); 733744947dcSTom Erickson if (DMU_USERUSED_DNODE(os)) { 734744947dcSTom Erickson dnode_special_close(&os->os_userused_dnode); 735744947dcSTom Erickson dnode_special_close(&os->os_groupused_dnode); 73614843421SMatthew Ahrens } 737503ad85cSMatthew Ahrens zil_free(os->os_zil); 738fa9e4066Sahrens 739dcbf3bd6SGeorge Wilson arc_buf_destroy(os->os_phys_buf, &os->os_phys_buf); 740744947dcSTom Erickson 741744947dcSTom Erickson /* 742744947dcSTom Erickson * This is a barrier to prevent the objset from going away in 743744947dcSTom Erickson * dnode_move() until we can safely ensure that the objset is still in 744744947dcSTom Erickson * use. We consider the objset valid before the barrier and invalid 745744947dcSTom Erickson * after the barrier. 746744947dcSTom Erickson */ 747744947dcSTom Erickson rw_enter(&os_lock, RW_READER); 748744947dcSTom Erickson rw_exit(&os_lock); 749744947dcSTom Erickson 750503ad85cSMatthew Ahrens mutex_destroy(&os->os_lock); 751503ad85cSMatthew Ahrens mutex_destroy(&os->os_obj_lock); 752503ad85cSMatthew Ahrens mutex_destroy(&os->os_user_ptr_lock); 753bc9014e6SJustin Gibbs spa_evicting_os_deregister(os->os_spa, os); 754503ad85cSMatthew Ahrens kmem_free(os, sizeof (objset_t)); 755fa9e4066Sahrens } 756fa9e4066Sahrens 75771eb0538SChris Kirby timestruc_t 75871eb0538SChris Kirby dmu_objset_snap_cmtime(objset_t *os) 75971eb0538SChris Kirby { 76071eb0538SChris Kirby return (dsl_dir_snap_cmtime(os->os_dsl_dataset->ds_dir)); 76171eb0538SChris Kirby } 76271eb0538SChris Kirby 763fa9e4066Sahrens /* called from dsl for meta-objset */ 764503ad85cSMatthew Ahrens objset_t * 765c717a561Smaybee dmu_objset_create_impl(spa_t *spa, dsl_dataset_t *ds, blkptr_t *bp, 766c717a561Smaybee dmu_objset_type_t type, dmu_tx_t *tx) 767fa9e4066Sahrens { 768503ad85cSMatthew Ahrens objset_t *os; 769fa9e4066Sahrens dnode_t *mdn; 770fa9e4066Sahrens 771fa9e4066Sahrens ASSERT(dmu_tx_is_syncing(tx)); 7723b2aab18SMatthew Ahrens 773feaa74e4SMark Maybee if (ds != NULL) 7743b2aab18SMatthew Ahrens VERIFY0(dmu_objset_from_ds(ds, &os)); 775feaa74e4SMark Maybee else 7763b2aab18SMatthew Ahrens VERIFY0(dmu_objset_open_impl(spa, NULL, bp, &os)); 777feaa74e4SMark Maybee 778744947dcSTom Erickson mdn = DMU_META_DNODE(os); 779fa9e4066Sahrens 780fa9e4066Sahrens dnode_allocate(mdn, DMU_OT_DNODE, 1 << DNODE_BLOCK_SHIFT, 781fa9e4066Sahrens DN_MAX_INDBLKSHIFT, DMU_OT_NONE, 0, tx); 782fa9e4066Sahrens 783fa9e4066Sahrens /* 784fa9e4066Sahrens * We don't want to have to increase the meta-dnode's nlevels 785fa9e4066Sahrens * later, because then we could do it in quescing context while 786fa9e4066Sahrens * we are also accessing it in open context. 787fa9e4066Sahrens * 788fa9e4066Sahrens * This precaution is not necessary for the MOS (ds == NULL), 789fa9e4066Sahrens * because the MOS is only updated in syncing context. 790fa9e4066Sahrens * This is most fortunate: the MOS is the only objset that 791fa9e4066Sahrens * needs to be synced multiple times as spa_sync() iterates 792fa9e4066Sahrens * to convergence, so minimizing its dn_nlevels matters. 793fa9e4066Sahrens */ 794ea8dc4b6Seschrock if (ds != NULL) { 795ea8dc4b6Seschrock int levels = 1; 796ea8dc4b6Seschrock 797ea8dc4b6Seschrock /* 798ea8dc4b6Seschrock * Determine the number of levels necessary for the meta-dnode 7994b5c8e93SMatthew Ahrens * to contain DN_MAX_OBJECT dnodes. Note that in order to 8004b5c8e93SMatthew Ahrens * ensure that we do not overflow 64 bits, there has to be 8014b5c8e93SMatthew Ahrens * a nlevels that gives us a number of blocks > DN_MAX_OBJECT 8024b5c8e93SMatthew Ahrens * but < 2^64. Therefore, 8034b5c8e93SMatthew Ahrens * (mdn->dn_indblkshift - SPA_BLKPTRSHIFT) (10) must be 8044b5c8e93SMatthew Ahrens * less than (64 - log2(DN_MAX_OBJECT)) (16). 805ea8dc4b6Seschrock */ 8064b5c8e93SMatthew Ahrens while ((uint64_t)mdn->dn_nblkptr << 8074b5c8e93SMatthew Ahrens (mdn->dn_datablkshift - DNODE_SHIFT + 808ea8dc4b6Seschrock (levels - 1) * (mdn->dn_indblkshift - SPA_BLKPTRSHIFT)) < 8094b5c8e93SMatthew Ahrens DN_MAX_OBJECT) 810ea8dc4b6Seschrock levels++; 811ea8dc4b6Seschrock 812fa9e4066Sahrens mdn->dn_next_nlevels[tx->tx_txg & TXG_MASK] = 813ea8dc4b6Seschrock mdn->dn_nlevels = levels; 814ea8dc4b6Seschrock } 815fa9e4066Sahrens 816fa9e4066Sahrens ASSERT(type != DMU_OST_NONE); 817fa9e4066Sahrens ASSERT(type != DMU_OST_ANY); 818fa9e4066Sahrens ASSERT(type < DMU_OST_NUMTYPES); 819503ad85cSMatthew Ahrens os->os_phys->os_type = type; 820503ad85cSMatthew Ahrens if (dmu_objset_userused_enabled(os)) { 821503ad85cSMatthew Ahrens os->os_phys->os_flags |= OBJSET_FLAG_USERACCOUNTING_COMPLETE; 822503ad85cSMatthew Ahrens os->os_flags = os->os_phys->os_flags; 82314843421SMatthew Ahrens } 824fa9e4066Sahrens 825fa9e4066Sahrens dsl_dataset_dirty(ds, tx); 826fa9e4066Sahrens 827503ad85cSMatthew Ahrens return (os); 828fa9e4066Sahrens } 829fa9e4066Sahrens 8303b2aab18SMatthew Ahrens typedef struct dmu_objset_create_arg { 8313b2aab18SMatthew Ahrens const char *doca_name; 8323b2aab18SMatthew Ahrens cred_t *doca_cred; 8333b2aab18SMatthew Ahrens void (*doca_userfunc)(objset_t *os, void *arg, 8343b2aab18SMatthew Ahrens cred_t *cr, dmu_tx_t *tx); 8353b2aab18SMatthew Ahrens void *doca_userarg; 8363b2aab18SMatthew Ahrens dmu_objset_type_t doca_type; 8373b2aab18SMatthew Ahrens uint64_t doca_flags; 8383b2aab18SMatthew Ahrens } dmu_objset_create_arg_t; 839fa9e4066Sahrens 840ecd6cf80Smarks /*ARGSUSED*/ 841fa9e4066Sahrens static int 8423b2aab18SMatthew Ahrens dmu_objset_create_check(void *arg, dmu_tx_t *tx) 843fa9e4066Sahrens { 8443b2aab18SMatthew Ahrens dmu_objset_create_arg_t *doca = arg; 8453b2aab18SMatthew Ahrens dsl_pool_t *dp = dmu_tx_pool(tx); 8463b2aab18SMatthew Ahrens dsl_dir_t *pdd; 8473b2aab18SMatthew Ahrens const char *tail; 8483b2aab18SMatthew Ahrens int error; 8491d452cf5Sahrens 8503b2aab18SMatthew Ahrens if (strchr(doca->doca_name, '@') != NULL) 851be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 8521d452cf5Sahrens 8539adfa60dSMatthew Ahrens if (strlen(doca->doca_name) >= ZFS_MAX_DATASET_NAME_LEN) 8549adfa60dSMatthew Ahrens return (SET_ERROR(ENAMETOOLONG)); 8559adfa60dSMatthew Ahrens 8563b2aab18SMatthew Ahrens error = dsl_dir_hold(dp, doca->doca_name, FTAG, &pdd, &tail); 8573b2aab18SMatthew Ahrens if (error != 0) 8583b2aab18SMatthew Ahrens return (error); 8593b2aab18SMatthew Ahrens if (tail == NULL) { 8603b2aab18SMatthew Ahrens dsl_dir_rele(pdd, FTAG); 861be6fd75aSMatthew Ahrens return (SET_ERROR(EEXIST)); 8621d452cf5Sahrens } 863a2afb611SJerry Jelinek error = dsl_fs_ss_limit_check(pdd, 1, ZFS_PROP_FILESYSTEM_LIMIT, NULL, 864a2afb611SJerry Jelinek doca->doca_cred); 8653b2aab18SMatthew Ahrens dsl_dir_rele(pdd, FTAG); 866ecd6cf80Smarks 867a2afb611SJerry Jelinek return (error); 8681d452cf5Sahrens } 8691d452cf5Sahrens 8701d452cf5Sahrens static void 8713b2aab18SMatthew Ahrens dmu_objset_create_sync(void *arg, dmu_tx_t *tx) 8721d452cf5Sahrens { 8733b2aab18SMatthew Ahrens dmu_objset_create_arg_t *doca = arg; 8743b2aab18SMatthew Ahrens dsl_pool_t *dp = dmu_tx_pool(tx); 8753b2aab18SMatthew Ahrens dsl_dir_t *pdd; 8763b2aab18SMatthew Ahrens const char *tail; 8774445fffbSMatthew Ahrens dsl_dataset_t *ds; 8783b2aab18SMatthew Ahrens uint64_t obj; 8794445fffbSMatthew Ahrens blkptr_t *bp; 8803b2aab18SMatthew Ahrens objset_t *os; 881fa9e4066Sahrens 8823b2aab18SMatthew Ahrens VERIFY0(dsl_dir_hold(dp, doca->doca_name, FTAG, &pdd, &tail)); 883fa9e4066Sahrens 8843b2aab18SMatthew Ahrens obj = dsl_dataset_create_sync(pdd, tail, NULL, doca->doca_flags, 8853b2aab18SMatthew Ahrens doca->doca_cred, tx); 886fa9e4066Sahrens 8873b2aab18SMatthew Ahrens VERIFY0(dsl_dataset_hold_obj(pdd->dd_pool, obj, FTAG, &ds)); 888c166b69dSPaul Dagnelie rrw_enter(&ds->ds_bp_rwlock, RW_READER, FTAG); 8894445fffbSMatthew Ahrens bp = dsl_dataset_get_blkptr(ds); 8903b2aab18SMatthew Ahrens os = dmu_objset_create_impl(pdd->dd_pool->dp_spa, 8913b2aab18SMatthew Ahrens ds, bp, doca->doca_type, tx); 892c166b69dSPaul Dagnelie rrw_exit(&ds->ds_bp_rwlock, FTAG); 893fa9e4066Sahrens 8943b2aab18SMatthew Ahrens if (doca->doca_userfunc != NULL) { 8953b2aab18SMatthew Ahrens doca->doca_userfunc(os, doca->doca_userarg, 8963b2aab18SMatthew Ahrens doca->doca_cred, tx); 897fa9e4066Sahrens } 898ecd6cf80Smarks 8993b2aab18SMatthew Ahrens spa_history_log_internal_ds(ds, "create", tx, ""); 9004445fffbSMatthew Ahrens dsl_dataset_rele(ds, FTAG); 9013b2aab18SMatthew Ahrens dsl_dir_rele(pdd, FTAG); 902fa9e4066Sahrens } 903fa9e4066Sahrens 904fa9e4066Sahrens int 905ae46e4c7SMatthew Ahrens dmu_objset_create(const char *name, dmu_objset_type_t type, uint64_t flags, 906ecd6cf80Smarks void (*func)(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx), void *arg) 907fa9e4066Sahrens { 9083b2aab18SMatthew Ahrens dmu_objset_create_arg_t doca; 909fa9e4066Sahrens 9103b2aab18SMatthew Ahrens doca.doca_name = name; 9113b2aab18SMatthew Ahrens doca.doca_cred = CRED(); 9123b2aab18SMatthew Ahrens doca.doca_flags = flags; 9133b2aab18SMatthew Ahrens doca.doca_userfunc = func; 9143b2aab18SMatthew Ahrens doca.doca_userarg = arg; 9153b2aab18SMatthew Ahrens doca.doca_type = type; 916ecd6cf80Smarks 9173b2aab18SMatthew Ahrens return (dsl_sync_task(name, 9187d46dc6cSMatthew Ahrens dmu_objset_create_check, dmu_objset_create_sync, &doca, 9197d46dc6cSMatthew Ahrens 5, ZFS_SPACE_CHECK_NORMAL)); 920ae46e4c7SMatthew Ahrens } 921ae46e4c7SMatthew Ahrens 9223b2aab18SMatthew Ahrens typedef struct dmu_objset_clone_arg { 9233b2aab18SMatthew Ahrens const char *doca_clone; 9243b2aab18SMatthew Ahrens const char *doca_origin; 9253b2aab18SMatthew Ahrens cred_t *doca_cred; 9263b2aab18SMatthew Ahrens } dmu_objset_clone_arg_t; 9273b2aab18SMatthew Ahrens 9283b2aab18SMatthew Ahrens /*ARGSUSED*/ 9293b2aab18SMatthew Ahrens static int 9303b2aab18SMatthew Ahrens dmu_objset_clone_check(void *arg, dmu_tx_t *tx) 931ae46e4c7SMatthew Ahrens { 9323b2aab18SMatthew Ahrens dmu_objset_clone_arg_t *doca = arg; 933ae46e4c7SMatthew Ahrens dsl_dir_t *pdd; 934ae46e4c7SMatthew Ahrens const char *tail; 9353b2aab18SMatthew Ahrens int error; 9363b2aab18SMatthew Ahrens dsl_dataset_t *origin; 9373b2aab18SMatthew Ahrens dsl_pool_t *dp = dmu_tx_pool(tx); 938ae46e4c7SMatthew Ahrens 9393b2aab18SMatthew Ahrens if (strchr(doca->doca_clone, '@') != NULL) 940be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 9413b2aab18SMatthew Ahrens 9429adfa60dSMatthew Ahrens if (strlen(doca->doca_clone) >= ZFS_MAX_DATASET_NAME_LEN) 9439adfa60dSMatthew Ahrens return (SET_ERROR(ENAMETOOLONG)); 9449adfa60dSMatthew Ahrens 9453b2aab18SMatthew Ahrens error = dsl_dir_hold(dp, doca->doca_clone, FTAG, &pdd, &tail); 9463b2aab18SMatthew Ahrens if (error != 0) 9473b2aab18SMatthew Ahrens return (error); 948ae46e4c7SMatthew Ahrens if (tail == NULL) { 9493b2aab18SMatthew Ahrens dsl_dir_rele(pdd, FTAG); 950be6fd75aSMatthew Ahrens return (SET_ERROR(EEXIST)); 951fa9e4066Sahrens } 95203b1c297SAlexander Eremin 953a2afb611SJerry Jelinek error = dsl_fs_ss_limit_check(pdd, 1, ZFS_PROP_FILESYSTEM_LIMIT, NULL, 954a2afb611SJerry Jelinek doca->doca_cred); 955a2afb611SJerry Jelinek if (error != 0) { 956a2afb611SJerry Jelinek dsl_dir_rele(pdd, FTAG); 957a2afb611SJerry Jelinek return (SET_ERROR(EDQUOT)); 958a2afb611SJerry Jelinek } 9593b2aab18SMatthew Ahrens dsl_dir_rele(pdd, FTAG); 960fa9e4066Sahrens 9613b2aab18SMatthew Ahrens error = dsl_dataset_hold(dp, doca->doca_origin, FTAG, &origin); 9623b2aab18SMatthew Ahrens if (error != 0) 96399d5e173STim Haley return (error); 96499d5e173STim Haley 9653b2aab18SMatthew Ahrens /* You can only clone snapshots, not the head datasets. */ 966bc9014e6SJustin Gibbs if (!origin->ds_is_snapshot) { 9673b2aab18SMatthew Ahrens dsl_dataset_rele(origin, FTAG); 968be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 96992241e0bSTom Erickson } 9703b2aab18SMatthew Ahrens dsl_dataset_rele(origin, FTAG); 97199d5e173STim Haley 9723b2aab18SMatthew Ahrens return (0); 973ea2f5b9eSMatthew Ahrens } 9741d452cf5Sahrens 9753b2aab18SMatthew Ahrens static void 9763b2aab18SMatthew Ahrens dmu_objset_clone_sync(void *arg, dmu_tx_t *tx) 9771d452cf5Sahrens { 9783b2aab18SMatthew Ahrens dmu_objset_clone_arg_t *doca = arg; 9793b2aab18SMatthew Ahrens dsl_pool_t *dp = dmu_tx_pool(tx); 9803b2aab18SMatthew Ahrens dsl_dir_t *pdd; 9813b2aab18SMatthew Ahrens const char *tail; 9823b2aab18SMatthew Ahrens dsl_dataset_t *origin, *ds; 9833b2aab18SMatthew Ahrens uint64_t obj; 9849adfa60dSMatthew Ahrens char namebuf[ZFS_MAX_DATASET_NAME_LEN]; 9854445fffbSMatthew Ahrens 9863b2aab18SMatthew Ahrens VERIFY0(dsl_dir_hold(dp, doca->doca_clone, FTAG, &pdd, &tail)); 9873b2aab18SMatthew Ahrens VERIFY0(dsl_dataset_hold(dp, doca->doca_origin, FTAG, &origin)); 9884445fffbSMatthew Ahrens 9893b2aab18SMatthew Ahrens obj = dsl_dataset_create_sync(pdd, tail, origin, 0, 9903b2aab18SMatthew Ahrens doca->doca_cred, tx); 991f2e10be3Srm 9923b2aab18SMatthew Ahrens VERIFY0(dsl_dataset_hold_obj(pdd->dd_pool, obj, FTAG, &ds)); 9933b2aab18SMatthew Ahrens dsl_dataset_name(origin, namebuf); 9943b2aab18SMatthew Ahrens spa_history_log_internal_ds(ds, "clone", tx, 9953b2aab18SMatthew Ahrens "origin=%s (%llu)", namebuf, origin->ds_object); 9963b2aab18SMatthew Ahrens dsl_dataset_rele(ds, FTAG); 9973b2aab18SMatthew Ahrens dsl_dataset_rele(origin, FTAG); 9983b2aab18SMatthew Ahrens dsl_dir_rele(pdd, FTAG); 9991d452cf5Sahrens } 10001d452cf5Sahrens 10011d452cf5Sahrens int 10023b2aab18SMatthew Ahrens dmu_objset_clone(const char *clone, const char *origin) 10031d452cf5Sahrens { 10043b2aab18SMatthew Ahrens dmu_objset_clone_arg_t doca; 10054445fffbSMatthew Ahrens 10063b2aab18SMatthew Ahrens doca.doca_clone = clone; 10073b2aab18SMatthew Ahrens doca.doca_origin = origin; 10083b2aab18SMatthew Ahrens doca.doca_cred = CRED(); 100999d5e173STim Haley 10103b2aab18SMatthew Ahrens return (dsl_sync_task(clone, 10117d46dc6cSMatthew Ahrens dmu_objset_clone_check, dmu_objset_clone_sync, &doca, 10127d46dc6cSMatthew Ahrens 5, ZFS_SPACE_CHECK_NORMAL)); 10134445fffbSMatthew Ahrens } 10144445fffbSMatthew Ahrens 10154445fffbSMatthew Ahrens int 10164445fffbSMatthew Ahrens dmu_objset_snapshot_one(const char *fsname, const char *snapname) 10174445fffbSMatthew Ahrens { 10184445fffbSMatthew Ahrens int err; 10194445fffbSMatthew Ahrens char *longsnap = kmem_asprintf("%s@%s", fsname, snapname); 10204445fffbSMatthew Ahrens nvlist_t *snaps = fnvlist_alloc(); 10214445fffbSMatthew Ahrens 10224445fffbSMatthew Ahrens fnvlist_add_boolean(snaps, longsnap); 10234445fffbSMatthew Ahrens strfree(longsnap); 10243b2aab18SMatthew Ahrens err = dsl_dataset_snapshot(snaps, NULL, NULL); 10253b2aab18SMatthew Ahrens fnvlist_free(snaps); 10264445fffbSMatthew Ahrens return (err); 10274445fffbSMatthew Ahrens } 10284445fffbSMatthew Ahrens 1029fa9e4066Sahrens static void 103014843421SMatthew Ahrens dmu_objset_sync_dnodes(list_t *list, list_t *newlist, dmu_tx_t *tx) 1031fa9e4066Sahrens { 1032c717a561Smaybee dnode_t *dn; 1033faafa6e3Sahrens 1034c717a561Smaybee while (dn = list_head(list)) { 1035c717a561Smaybee ASSERT(dn->dn_object != DMU_META_DNODE_OBJECT); 1036c717a561Smaybee ASSERT(dn->dn_dbuf->db_data_pending); 1037c717a561Smaybee /* 103814843421SMatthew Ahrens * Initialize dn_zio outside dnode_sync() because the 103914843421SMatthew Ahrens * meta-dnode needs to set it ouside dnode_sync(). 1040c717a561Smaybee */ 1041c717a561Smaybee dn->dn_zio = dn->dn_dbuf->db_data_pending->dr_zio; 1042c717a561Smaybee ASSERT(dn->dn_zio); 1043faafa6e3Sahrens 1044c717a561Smaybee ASSERT3U(dn->dn_nlevels, <=, DN_MAX_LEVELS); 1045c717a561Smaybee list_remove(list, dn); 104614843421SMatthew Ahrens 104714843421SMatthew Ahrens if (newlist) { 104814843421SMatthew Ahrens (void) dnode_add_ref(dn, newlist); 104914843421SMatthew Ahrens list_insert_tail(newlist, dn); 105014843421SMatthew Ahrens } 105114843421SMatthew Ahrens 1052c717a561Smaybee dnode_sync(dn, tx); 1053fa9e4066Sahrens } 1054fa9e4066Sahrens } 1055fa9e4066Sahrens 1056fa9e4066Sahrens /* ARGSUSED */ 1057fa9e4066Sahrens static void 1058b24ab676SJeff Bonwick dmu_objset_write_ready(zio_t *zio, arc_buf_t *abuf, void *arg) 1059fa9e4066Sahrens { 1060e14bb325SJeff Bonwick blkptr_t *bp = zio->io_bp; 1061503ad85cSMatthew Ahrens objset_t *os = arg; 1062c717a561Smaybee dnode_phys_t *dnp = &os->os_phys->os_meta_dnode; 1063fa9e4066Sahrens 10645d7b4d43SMatthew Ahrens ASSERT(!BP_IS_EMBEDDED(bp)); 10653b2aab18SMatthew Ahrens ASSERT3U(BP_GET_TYPE(bp), ==, DMU_OT_OBJSET); 10663b2aab18SMatthew Ahrens ASSERT0(BP_GET_LEVEL(bp)); 10670a4e9518Sgw 1068fa9e4066Sahrens /* 106914843421SMatthew Ahrens * Update rootbp fill count: it should be the number of objects 107014843421SMatthew Ahrens * allocated in the object set (not counting the "special" 107114843421SMatthew Ahrens * objects that are stored in the objset_phys_t -- the meta 107214843421SMatthew Ahrens * dnode and user/group accounting objects). 1073fa9e4066Sahrens */ 107414843421SMatthew Ahrens bp->blk_fill = 0; 1075e14bb325SJeff Bonwick for (int i = 0; i < dnp->dn_nblkptr; i++) 10765d7b4d43SMatthew Ahrens bp->blk_fill += BP_GET_FILL(&dnp->dn_blkptr[i]); 1077c166b69dSPaul Dagnelie if (os->os_dsl_dataset != NULL) 1078c166b69dSPaul Dagnelie rrw_enter(&os->os_dsl_dataset->ds_bp_rwlock, RW_WRITER, FTAG); 1079c166b69dSPaul Dagnelie *os->os_rootbp = *bp; 1080c166b69dSPaul Dagnelie if (os->os_dsl_dataset != NULL) 1081c166b69dSPaul Dagnelie rrw_exit(&os->os_dsl_dataset->ds_bp_rwlock, FTAG); 1082b24ab676SJeff Bonwick } 1083b24ab676SJeff Bonwick 1084b24ab676SJeff Bonwick /* ARGSUSED */ 1085b24ab676SJeff Bonwick static void 1086b24ab676SJeff Bonwick dmu_objset_write_done(zio_t *zio, arc_buf_t *abuf, void *arg) 1087b24ab676SJeff Bonwick { 1088b24ab676SJeff Bonwick blkptr_t *bp = zio->io_bp; 1089b24ab676SJeff Bonwick blkptr_t *bp_orig = &zio->io_bp_orig; 1090b24ab676SJeff Bonwick objset_t *os = arg; 10910a4e9518Sgw 1092e14bb325SJeff Bonwick if (zio->io_flags & ZIO_FLAG_IO_REWRITE) { 1093b24ab676SJeff Bonwick ASSERT(BP_EQUAL(bp, bp_orig)); 1094e14bb325SJeff Bonwick } else { 1095b24ab676SJeff Bonwick dsl_dataset_t *ds = os->os_dsl_dataset; 1096b24ab676SJeff Bonwick dmu_tx_t *tx = os->os_synctx; 1097b24ab676SJeff Bonwick 1098b24ab676SJeff Bonwick (void) dsl_dataset_block_kill(ds, bp_orig, tx, B_TRUE); 1099b24ab676SJeff Bonwick dsl_dataset_block_born(ds, bp, tx); 11000a4e9518Sgw } 1101c166b69dSPaul Dagnelie kmem_free(bp, sizeof (*bp)); 1102c717a561Smaybee } 1103c717a561Smaybee 1104fa9e4066Sahrens /* called from dsl */ 1105fa9e4066Sahrens void 1106503ad85cSMatthew Ahrens dmu_objset_sync(objset_t *os, zio_t *pio, dmu_tx_t *tx) 1107fa9e4066Sahrens { 1108fa9e4066Sahrens int txgoff; 11097802d7bfSMatthew Ahrens zbookmark_phys_t zb; 1110b24ab676SJeff Bonwick zio_prop_t zp; 1111c717a561Smaybee zio_t *zio; 1112c717a561Smaybee list_t *list; 111314843421SMatthew Ahrens list_t *newlist = NULL; 1114c717a561Smaybee dbuf_dirty_record_t *dr; 1115c166b69dSPaul Dagnelie blkptr_t *blkptr_copy = kmem_alloc(sizeof (*os->os_rootbp), KM_SLEEP); 1116c166b69dSPaul Dagnelie *blkptr_copy = *os->os_rootbp; 1117c717a561Smaybee 1118c717a561Smaybee dprintf_ds(os->os_dsl_dataset, "txg=%llu\n", tx->tx_txg); 1119fa9e4066Sahrens 1120fa9e4066Sahrens ASSERT(dmu_tx_is_syncing(tx)); 1121fa9e4066Sahrens /* XXX the write_done callback should really give us the tx... */ 1122fa9e4066Sahrens os->os_synctx = tx; 1123fa9e4066Sahrens 112487bd5c1eSahrens if (os->os_dsl_dataset == NULL) { 112587bd5c1eSahrens /* 112687bd5c1eSahrens * This is the MOS. If we have upgraded, 112787bd5c1eSahrens * spa_max_replication() could change, so reset 112887bd5c1eSahrens * os_copies here. 112987bd5c1eSahrens */ 113087bd5c1eSahrens os->os_copies = spa_max_replication(os->os_spa); 113187bd5c1eSahrens } 113287bd5c1eSahrens 1133fa9e4066Sahrens /* 1134c717a561Smaybee * Create the root block IO 1135fa9e4066Sahrens */ 1136b24ab676SJeff Bonwick SET_BOOKMARK(&zb, os->os_dsl_dataset ? 1137b24ab676SJeff Bonwick os->os_dsl_dataset->ds_object : DMU_META_OBJSET, 1138b24ab676SJeff Bonwick ZB_ROOT_OBJECT, ZB_ROOT_LEVEL, ZB_ROOT_BLKID); 11391b912ec7SGeorge Wilson arc_release(os->os_phys_buf, &os->os_phys_buf); 1140b24ab676SJeff Bonwick 1141b24ab676SJeff Bonwick dmu_write_policy(os, NULL, 0, 0, &zp); 1142b24ab676SJeff Bonwick 1143b24ab676SJeff Bonwick zio = arc_write(pio, os->os_spa, tx->tx_txg, 1144c166b69dSPaul Dagnelie blkptr_copy, os->os_phys_buf, DMU_OS_IS_L2CACHEABLE(os), 11458df0bcf0SPaul Dagnelie &zp, dmu_objset_write_ready, NULL, NULL, dmu_objset_write_done, 11468df0bcf0SPaul Dagnelie os, ZIO_PRIORITY_ASYNC_WRITE, ZIO_FLAG_MUSTSUCCEED, &zb); 1147c717a561Smaybee 1148c717a561Smaybee /* 114914843421SMatthew Ahrens * Sync special dnodes - the parent IO for the sync is the root block 1150c717a561Smaybee */ 1151744947dcSTom Erickson DMU_META_DNODE(os)->dn_zio = zio; 1152744947dcSTom Erickson dnode_sync(DMU_META_DNODE(os), tx); 1153c717a561Smaybee 115414843421SMatthew Ahrens os->os_phys->os_flags = os->os_flags; 115514843421SMatthew Ahrens 1156744947dcSTom Erickson if (DMU_USERUSED_DNODE(os) && 1157744947dcSTom Erickson DMU_USERUSED_DNODE(os)->dn_type != DMU_OT_NONE) { 1158744947dcSTom Erickson DMU_USERUSED_DNODE(os)->dn_zio = zio; 1159744947dcSTom Erickson dnode_sync(DMU_USERUSED_DNODE(os), tx); 1160744947dcSTom Erickson DMU_GROUPUSED_DNODE(os)->dn_zio = zio; 1161744947dcSTom Erickson dnode_sync(DMU_GROUPUSED_DNODE(os), tx); 116214843421SMatthew Ahrens } 116314843421SMatthew Ahrens 1164c717a561Smaybee txgoff = tx->tx_txg & TXG_MASK; 1165fa9e4066Sahrens 116614843421SMatthew Ahrens if (dmu_objset_userused_enabled(os)) { 116714843421SMatthew Ahrens newlist = &os->os_synced_dnodes; 116814843421SMatthew Ahrens /* 116914843421SMatthew Ahrens * We must create the list here because it uses the 117014843421SMatthew Ahrens * dn_dirty_link[] of this txg. 117114843421SMatthew Ahrens */ 117214843421SMatthew Ahrens list_create(newlist, sizeof (dnode_t), 117314843421SMatthew Ahrens offsetof(dnode_t, dn_dirty_link[txgoff])); 117414843421SMatthew Ahrens } 117514843421SMatthew Ahrens 117614843421SMatthew Ahrens dmu_objset_sync_dnodes(&os->os_free_dnodes[txgoff], newlist, tx); 117714843421SMatthew Ahrens dmu_objset_sync_dnodes(&os->os_dirty_dnodes[txgoff], newlist, tx); 1178fa9e4066Sahrens 1179744947dcSTom Erickson list = &DMU_META_DNODE(os)->dn_dirty_records[txgoff]; 1180c717a561Smaybee while (dr = list_head(list)) { 11813b2aab18SMatthew Ahrens ASSERT0(dr->dr_dbuf->db_level); 1182c717a561Smaybee list_remove(list, dr); 1183c717a561Smaybee if (dr->dr_zio) 1184c717a561Smaybee zio_nowait(dr->dr_zio); 1185c717a561Smaybee } 1186*af346df5SNed Bass 1187*af346df5SNed Bass /* Enable dnode backfill if enough objects have been freed. */ 1188*af346df5SNed Bass if (os->os_freed_dnodes >= dmu_rescan_dnode_threshold) { 1189*af346df5SNed Bass os->os_rescan_dnodes = B_TRUE; 1190*af346df5SNed Bass os->os_freed_dnodes = 0; 1191*af346df5SNed Bass } 1192*af346df5SNed Bass 1193c717a561Smaybee /* 1194c717a561Smaybee * Free intent log blocks up to this tx. 1195c717a561Smaybee */ 1196c717a561Smaybee zil_sync(os->os_zil, tx); 1197088f3894Sahrens os->os_phys->os_zil_header = os->os_zil_header; 1198c717a561Smaybee zio_nowait(zio); 1199fa9e4066Sahrens } 1200fa9e4066Sahrens 1201b24ab676SJeff Bonwick boolean_t 1202b24ab676SJeff Bonwick dmu_objset_is_dirty(objset_t *os, uint64_t txg) 1203b24ab676SJeff Bonwick { 1204b24ab676SJeff Bonwick return (!list_is_empty(&os->os_dirty_dnodes[txg & TXG_MASK]) || 1205b24ab676SJeff Bonwick !list_is_empty(&os->os_free_dnodes[txg & TXG_MASK])); 1206b24ab676SJeff Bonwick } 1207b24ab676SJeff Bonwick 1208744947dcSTom Erickson static objset_used_cb_t *used_cbs[DMU_OST_NUMTYPES]; 120914843421SMatthew Ahrens 121014843421SMatthew Ahrens void 121114843421SMatthew Ahrens dmu_objset_register_type(dmu_objset_type_t ost, objset_used_cb_t *cb) 121214843421SMatthew Ahrens { 121314843421SMatthew Ahrens used_cbs[ost] = cb; 121414843421SMatthew Ahrens } 121514843421SMatthew Ahrens 121614843421SMatthew Ahrens boolean_t 1217503ad85cSMatthew Ahrens dmu_objset_userused_enabled(objset_t *os) 121814843421SMatthew Ahrens { 121914843421SMatthew Ahrens return (spa_version(os->os_spa) >= SPA_VERSION_USERSPACE && 1220744947dcSTom Erickson used_cbs[os->os_phys->os_type] != NULL && 1221744947dcSTom Erickson DMU_USERUSED_DNODE(os) != NULL); 122214843421SMatthew Ahrens } 122314843421SMatthew Ahrens 12248b70546bSMatthew Ahrens typedef struct userquota_node { 12258b70546bSMatthew Ahrens uint64_t uqn_id; 12268b70546bSMatthew Ahrens int64_t uqn_delta; 12278b70546bSMatthew Ahrens avl_node_t uqn_node; 12288b70546bSMatthew Ahrens } userquota_node_t; 12298b70546bSMatthew Ahrens 12308b70546bSMatthew Ahrens typedef struct userquota_cache { 12318b70546bSMatthew Ahrens avl_tree_t uqc_user_deltas; 12328b70546bSMatthew Ahrens avl_tree_t uqc_group_deltas; 12338b70546bSMatthew Ahrens } userquota_cache_t; 12348b70546bSMatthew Ahrens 12358b70546bSMatthew Ahrens static int 12368b70546bSMatthew Ahrens userquota_compare(const void *l, const void *r) 12378b70546bSMatthew Ahrens { 12388b70546bSMatthew Ahrens const userquota_node_t *luqn = l; 12398b70546bSMatthew Ahrens const userquota_node_t *ruqn = r; 12408b70546bSMatthew Ahrens 12418b70546bSMatthew Ahrens if (luqn->uqn_id < ruqn->uqn_id) 12428b70546bSMatthew Ahrens return (-1); 12438b70546bSMatthew Ahrens if (luqn->uqn_id > ruqn->uqn_id) 12448b70546bSMatthew Ahrens return (1); 12458b70546bSMatthew Ahrens return (0); 12468b70546bSMatthew Ahrens } 12478b70546bSMatthew Ahrens 12489966ca11SMatthew Ahrens static void 12498b70546bSMatthew Ahrens do_userquota_cacheflush(objset_t *os, userquota_cache_t *cache, dmu_tx_t *tx) 12508b70546bSMatthew Ahrens { 12518b70546bSMatthew Ahrens void *cookie; 12528b70546bSMatthew Ahrens userquota_node_t *uqn; 12538b70546bSMatthew Ahrens 12548b70546bSMatthew Ahrens ASSERT(dmu_tx_is_syncing(tx)); 12558b70546bSMatthew Ahrens 12568b70546bSMatthew Ahrens cookie = NULL; 12578b70546bSMatthew Ahrens while ((uqn = avl_destroy_nodes(&cache->uqc_user_deltas, 12588b70546bSMatthew Ahrens &cookie)) != NULL) { 12598b70546bSMatthew Ahrens VERIFY0(zap_increment_int(os, DMU_USERUSED_OBJECT, 12608b70546bSMatthew Ahrens uqn->uqn_id, uqn->uqn_delta, tx)); 12618b70546bSMatthew Ahrens kmem_free(uqn, sizeof (*uqn)); 12628b70546bSMatthew Ahrens } 12638b70546bSMatthew Ahrens avl_destroy(&cache->uqc_user_deltas); 12648b70546bSMatthew Ahrens 12658b70546bSMatthew Ahrens cookie = NULL; 12668b70546bSMatthew Ahrens while ((uqn = avl_destroy_nodes(&cache->uqc_group_deltas, 12678b70546bSMatthew Ahrens &cookie)) != NULL) { 12688b70546bSMatthew Ahrens VERIFY0(zap_increment_int(os, DMU_GROUPUSED_OBJECT, 12698b70546bSMatthew Ahrens uqn->uqn_id, uqn->uqn_delta, tx)); 12708b70546bSMatthew Ahrens kmem_free(uqn, sizeof (*uqn)); 12718b70546bSMatthew Ahrens } 12728b70546bSMatthew Ahrens avl_destroy(&cache->uqc_group_deltas); 12738b70546bSMatthew Ahrens } 12748b70546bSMatthew Ahrens 12758b70546bSMatthew Ahrens static void 12768b70546bSMatthew Ahrens userquota_update_cache(avl_tree_t *avl, uint64_t id, int64_t delta) 12778b70546bSMatthew Ahrens { 12788b70546bSMatthew Ahrens userquota_node_t search = { .uqn_id = id }; 12798b70546bSMatthew Ahrens avl_index_t idx; 12808b70546bSMatthew Ahrens 12818b70546bSMatthew Ahrens userquota_node_t *uqn = avl_find(avl, &search, &idx); 12828b70546bSMatthew Ahrens if (uqn == NULL) { 12838b70546bSMatthew Ahrens uqn = kmem_zalloc(sizeof (*uqn), KM_SLEEP); 12848b70546bSMatthew Ahrens uqn->uqn_id = id; 12858b70546bSMatthew Ahrens avl_insert(avl, uqn, idx); 12868b70546bSMatthew Ahrens } 12878b70546bSMatthew Ahrens uqn->uqn_delta += delta; 12888b70546bSMatthew Ahrens } 12898b70546bSMatthew Ahrens 12908b70546bSMatthew Ahrens static void 12918b70546bSMatthew Ahrens do_userquota_update(userquota_cache_t *cache, uint64_t used, uint64_t flags, 12928b70546bSMatthew Ahrens uint64_t user, uint64_t group, boolean_t subtract) 12939966ca11SMatthew Ahrens { 12940a586ceaSMark Shellenbaum if ((flags & DNODE_FLAG_USERUSED_ACCOUNTED)) { 12950a586ceaSMark Shellenbaum int64_t delta = DNODE_SIZE + used; 12969966ca11SMatthew Ahrens if (subtract) 12979966ca11SMatthew Ahrens delta = -delta; 12988b70546bSMatthew Ahrens 12998b70546bSMatthew Ahrens userquota_update_cache(&cache->uqc_user_deltas, user, delta); 13008b70546bSMatthew Ahrens userquota_update_cache(&cache->uqc_group_deltas, group, delta); 13019966ca11SMatthew Ahrens } 13029966ca11SMatthew Ahrens } 13039966ca11SMatthew Ahrens 130414843421SMatthew Ahrens void 13050a586ceaSMark Shellenbaum dmu_objset_do_userquota_updates(objset_t *os, dmu_tx_t *tx) 130614843421SMatthew Ahrens { 130714843421SMatthew Ahrens dnode_t *dn; 130814843421SMatthew Ahrens list_t *list = &os->os_synced_dnodes; 13098b70546bSMatthew Ahrens userquota_cache_t cache = { 0 }; 131014843421SMatthew Ahrens 131114843421SMatthew Ahrens ASSERT(list_head(list) == NULL || dmu_objset_userused_enabled(os)); 131214843421SMatthew Ahrens 13138b70546bSMatthew Ahrens avl_create(&cache.uqc_user_deltas, userquota_compare, 13148b70546bSMatthew Ahrens sizeof (userquota_node_t), offsetof(userquota_node_t, uqn_node)); 13158b70546bSMatthew Ahrens avl_create(&cache.uqc_group_deltas, userquota_compare, 13168b70546bSMatthew Ahrens sizeof (userquota_node_t), offsetof(userquota_node_t, uqn_node)); 13178b70546bSMatthew Ahrens 131814843421SMatthew Ahrens while (dn = list_head(list)) { 13191d8ccc7bSMark Shellenbaum int flags; 132014843421SMatthew Ahrens ASSERT(!DMU_OBJECT_IS_SPECIAL(dn->dn_object)); 132114843421SMatthew Ahrens ASSERT(dn->dn_phys->dn_type == DMU_OT_NONE || 132214843421SMatthew Ahrens dn->dn_phys->dn_flags & 132314843421SMatthew Ahrens DNODE_FLAG_USERUSED_ACCOUNTED); 132414843421SMatthew Ahrens 132514843421SMatthew Ahrens /* Allocate the user/groupused objects if necessary. */ 1326744947dcSTom Erickson if (DMU_USERUSED_DNODE(os)->dn_type == DMU_OT_NONE) { 13278b70546bSMatthew Ahrens VERIFY0(zap_create_claim(os, 132814843421SMatthew Ahrens DMU_USERUSED_OBJECT, 132914843421SMatthew Ahrens DMU_OT_USERGROUP_USED, DMU_OT_NONE, 0, tx)); 13308b70546bSMatthew Ahrens VERIFY0(zap_create_claim(os, 133114843421SMatthew Ahrens DMU_GROUPUSED_OBJECT, 133214843421SMatthew Ahrens DMU_OT_USERGROUP_USED, DMU_OT_NONE, 0, tx)); 133314843421SMatthew Ahrens } 133414843421SMatthew Ahrens 13351d8ccc7bSMark Shellenbaum flags = dn->dn_id_flags; 13361d8ccc7bSMark Shellenbaum ASSERT(flags); 13371d8ccc7bSMark Shellenbaum if (flags & DN_ID_OLD_EXIST) { 13388b70546bSMatthew Ahrens do_userquota_update(&cache, 13398b70546bSMatthew Ahrens dn->dn_oldused, dn->dn_oldflags, 13408b70546bSMatthew Ahrens dn->dn_olduid, dn->dn_oldgid, B_TRUE); 13410a586ceaSMark Shellenbaum } 13421d8ccc7bSMark Shellenbaum if (flags & DN_ID_NEW_EXIST) { 13438b70546bSMatthew Ahrens do_userquota_update(&cache, 13448b70546bSMatthew Ahrens DN_USED_BYTES(dn->dn_phys), 13450a586ceaSMark Shellenbaum dn->dn_phys->dn_flags, dn->dn_newuid, 13468b70546bSMatthew Ahrens dn->dn_newgid, B_FALSE); 13470a586ceaSMark Shellenbaum } 13480a586ceaSMark Shellenbaum 13491d8ccc7bSMark Shellenbaum mutex_enter(&dn->dn_mtx); 13500a586ceaSMark Shellenbaum dn->dn_oldused = 0; 13510a586ceaSMark Shellenbaum dn->dn_oldflags = 0; 13520a586ceaSMark Shellenbaum if (dn->dn_id_flags & DN_ID_NEW_EXIST) { 13530a586ceaSMark Shellenbaum dn->dn_olduid = dn->dn_newuid; 13540a586ceaSMark Shellenbaum dn->dn_oldgid = dn->dn_newgid; 13550a586ceaSMark Shellenbaum dn->dn_id_flags |= DN_ID_OLD_EXIST; 13560a586ceaSMark Shellenbaum if (dn->dn_bonuslen == 0) 13570a586ceaSMark Shellenbaum dn->dn_id_flags |= DN_ID_CHKED_SPILL; 13580a586ceaSMark Shellenbaum else 13590a586ceaSMark Shellenbaum dn->dn_id_flags |= DN_ID_CHKED_BONUS; 13600a586ceaSMark Shellenbaum } 136128d97a71SMark Shellenbaum dn->dn_id_flags &= ~(DN_ID_NEW_EXIST); 136214843421SMatthew Ahrens mutex_exit(&dn->dn_mtx); 136314843421SMatthew Ahrens 136414843421SMatthew Ahrens list_remove(list, dn); 136514843421SMatthew Ahrens dnode_rele(dn, list); 136614843421SMatthew Ahrens } 13678b70546bSMatthew Ahrens do_userquota_cacheflush(os, &cache, tx); 136814843421SMatthew Ahrens } 136914843421SMatthew Ahrens 137006e0070dSMark Shellenbaum /* 137106e0070dSMark Shellenbaum * Returns a pointer to data to find uid/gid from 137206e0070dSMark Shellenbaum * 137306e0070dSMark Shellenbaum * If a dirty record for transaction group that is syncing can't 137406e0070dSMark Shellenbaum * be found then NULL is returned. In the NULL case it is assumed 137506e0070dSMark Shellenbaum * the uid/gid aren't changing. 137606e0070dSMark Shellenbaum */ 137706e0070dSMark Shellenbaum static void * 137806e0070dSMark Shellenbaum dmu_objset_userquota_find_data(dmu_buf_impl_t *db, dmu_tx_t *tx) 137906e0070dSMark Shellenbaum { 138006e0070dSMark Shellenbaum dbuf_dirty_record_t *dr, **drp; 138106e0070dSMark Shellenbaum void *data; 138206e0070dSMark Shellenbaum 138306e0070dSMark Shellenbaum if (db->db_dirtycnt == 0) 138406e0070dSMark Shellenbaum return (db->db.db_data); /* Nothing is changing */ 138506e0070dSMark Shellenbaum 138606e0070dSMark Shellenbaum for (drp = &db->db_last_dirty; (dr = *drp) != NULL; drp = &dr->dr_next) 138706e0070dSMark Shellenbaum if (dr->dr_txg == tx->tx_txg) 138806e0070dSMark Shellenbaum break; 138906e0070dSMark Shellenbaum 1390744947dcSTom Erickson if (dr == NULL) { 139106e0070dSMark Shellenbaum data = NULL; 1392744947dcSTom Erickson } else { 1393744947dcSTom Erickson dnode_t *dn; 1394744947dcSTom Erickson 1395744947dcSTom Erickson DB_DNODE_ENTER(dr->dr_dbuf); 1396744947dcSTom Erickson dn = DB_DNODE(dr->dr_dbuf); 1397744947dcSTom Erickson 1398744947dcSTom Erickson if (dn->dn_bonuslen == 0 && 1399744947dcSTom Erickson dr->dr_dbuf->db_blkid == DMU_SPILL_BLKID) 1400744947dcSTom Erickson data = dr->dt.dl.dr_data->b_data; 1401744947dcSTom Erickson else 1402744947dcSTom Erickson data = dr->dt.dl.dr_data; 1403744947dcSTom Erickson 1404744947dcSTom Erickson DB_DNODE_EXIT(dr->dr_dbuf); 1405744947dcSTom Erickson } 1406744947dcSTom Erickson 140706e0070dSMark Shellenbaum return (data); 140806e0070dSMark Shellenbaum } 140906e0070dSMark Shellenbaum 14100a586ceaSMark Shellenbaum void 141106e0070dSMark Shellenbaum dmu_objset_userquota_get_ids(dnode_t *dn, boolean_t before, dmu_tx_t *tx) 14120a586ceaSMark Shellenbaum { 14130a586ceaSMark Shellenbaum objset_t *os = dn->dn_objset; 14140a586ceaSMark Shellenbaum void *data = NULL; 141506e0070dSMark Shellenbaum dmu_buf_impl_t *db = NULL; 1416d5285caeSGeorge Wilson uint64_t *user = NULL; 1417d5285caeSGeorge Wilson uint64_t *group = NULL; 14180a586ceaSMark Shellenbaum int flags = dn->dn_id_flags; 14190a586ceaSMark Shellenbaum int error; 142006e0070dSMark Shellenbaum boolean_t have_spill = B_FALSE; 14210a586ceaSMark Shellenbaum 14220a586ceaSMark Shellenbaum if (!dmu_objset_userused_enabled(dn->dn_objset)) 14230a586ceaSMark Shellenbaum return; 14240a586ceaSMark Shellenbaum 14250a586ceaSMark Shellenbaum if (before && (flags & (DN_ID_CHKED_BONUS|DN_ID_OLD_EXIST| 14260a586ceaSMark Shellenbaum DN_ID_CHKED_SPILL))) 14270a586ceaSMark Shellenbaum return; 14280a586ceaSMark Shellenbaum 14290a586ceaSMark Shellenbaum if (before && dn->dn_bonuslen != 0) 14300a586ceaSMark Shellenbaum data = DN_BONUS(dn->dn_phys); 143106e0070dSMark Shellenbaum else if (!before && dn->dn_bonuslen != 0) { 143206e0070dSMark Shellenbaum if (dn->dn_bonus) { 143306e0070dSMark Shellenbaum db = dn->dn_bonus; 143406e0070dSMark Shellenbaum mutex_enter(&db->db_mtx); 143506e0070dSMark Shellenbaum data = dmu_objset_userquota_find_data(db, tx); 143606e0070dSMark Shellenbaum } else { 143706e0070dSMark Shellenbaum data = DN_BONUS(dn->dn_phys); 143806e0070dSMark Shellenbaum } 143906e0070dSMark Shellenbaum } else if (dn->dn_bonuslen == 0 && dn->dn_bonustype == DMU_OT_SA) { 14400a586ceaSMark Shellenbaum int rf = 0; 14410a586ceaSMark Shellenbaum 14420a586ceaSMark Shellenbaum if (RW_WRITE_HELD(&dn->dn_struct_rwlock)) 14430a586ceaSMark Shellenbaum rf |= DB_RF_HAVESTRUCT; 14441d8ccc7bSMark Shellenbaum error = dmu_spill_hold_by_dnode(dn, 14451d8ccc7bSMark Shellenbaum rf | DB_RF_MUST_SUCCEED, 144606e0070dSMark Shellenbaum FTAG, (dmu_buf_t **)&db); 14470a586ceaSMark Shellenbaum ASSERT(error == 0); 144806e0070dSMark Shellenbaum mutex_enter(&db->db_mtx); 144906e0070dSMark Shellenbaum data = (before) ? db->db.db_data : 145006e0070dSMark Shellenbaum dmu_objset_userquota_find_data(db, tx); 145106e0070dSMark Shellenbaum have_spill = B_TRUE; 14520a586ceaSMark Shellenbaum } else { 14530a586ceaSMark Shellenbaum mutex_enter(&dn->dn_mtx); 14540a586ceaSMark Shellenbaum dn->dn_id_flags |= DN_ID_CHKED_BONUS; 14550a586ceaSMark Shellenbaum mutex_exit(&dn->dn_mtx); 14560a586ceaSMark Shellenbaum return; 14570a586ceaSMark Shellenbaum } 14580a586ceaSMark Shellenbaum 14590a586ceaSMark Shellenbaum if (before) { 146006e0070dSMark Shellenbaum ASSERT(data); 14610a586ceaSMark Shellenbaum user = &dn->dn_olduid; 14620a586ceaSMark Shellenbaum group = &dn->dn_oldgid; 146306e0070dSMark Shellenbaum } else if (data) { 14640a586ceaSMark Shellenbaum user = &dn->dn_newuid; 14650a586ceaSMark Shellenbaum group = &dn->dn_newgid; 14660a586ceaSMark Shellenbaum } 14670a586ceaSMark Shellenbaum 146806e0070dSMark Shellenbaum /* 146906e0070dSMark Shellenbaum * Must always call the callback in case the object 147006e0070dSMark Shellenbaum * type has changed and that type isn't an object type to track 147106e0070dSMark Shellenbaum */ 14720a586ceaSMark Shellenbaum error = used_cbs[os->os_phys->os_type](dn->dn_bonustype, data, 14730a586ceaSMark Shellenbaum user, group); 14740a586ceaSMark Shellenbaum 147506e0070dSMark Shellenbaum /* 147606e0070dSMark Shellenbaum * Preserve existing uid/gid when the callback can't determine 147706e0070dSMark Shellenbaum * what the new uid/gid are and the callback returned EEXIST. 147806e0070dSMark Shellenbaum * The EEXIST error tells us to just use the existing uid/gid. 147906e0070dSMark Shellenbaum * If we don't know what the old values are then just assign 148006e0070dSMark Shellenbaum * them to 0, since that is a new file being created. 148106e0070dSMark Shellenbaum */ 148206e0070dSMark Shellenbaum if (!before && data == NULL && error == EEXIST) { 148306e0070dSMark Shellenbaum if (flags & DN_ID_OLD_EXIST) { 148406e0070dSMark Shellenbaum dn->dn_newuid = dn->dn_olduid; 148506e0070dSMark Shellenbaum dn->dn_newgid = dn->dn_oldgid; 148606e0070dSMark Shellenbaum } else { 148706e0070dSMark Shellenbaum dn->dn_newuid = 0; 148806e0070dSMark Shellenbaum dn->dn_newgid = 0; 148906e0070dSMark Shellenbaum } 149006e0070dSMark Shellenbaum error = 0; 149106e0070dSMark Shellenbaum } 149206e0070dSMark Shellenbaum 149306e0070dSMark Shellenbaum if (db) 149406e0070dSMark Shellenbaum mutex_exit(&db->db_mtx); 149506e0070dSMark Shellenbaum 14960a586ceaSMark Shellenbaum mutex_enter(&dn->dn_mtx); 14970a586ceaSMark Shellenbaum if (error == 0 && before) 14980a586ceaSMark Shellenbaum dn->dn_id_flags |= DN_ID_OLD_EXIST; 14990a586ceaSMark Shellenbaum if (error == 0 && !before) 15000a586ceaSMark Shellenbaum dn->dn_id_flags |= DN_ID_NEW_EXIST; 15010a586ceaSMark Shellenbaum 150206e0070dSMark Shellenbaum if (have_spill) { 15030a586ceaSMark Shellenbaum dn->dn_id_flags |= DN_ID_CHKED_SPILL; 15040a586ceaSMark Shellenbaum } else { 15050a586ceaSMark Shellenbaum dn->dn_id_flags |= DN_ID_CHKED_BONUS; 15060a586ceaSMark Shellenbaum } 15070a586ceaSMark Shellenbaum mutex_exit(&dn->dn_mtx); 150806e0070dSMark Shellenbaum if (have_spill) 150906e0070dSMark Shellenbaum dmu_buf_rele((dmu_buf_t *)db, FTAG); 15100a586ceaSMark Shellenbaum } 15110a586ceaSMark Shellenbaum 151214843421SMatthew Ahrens boolean_t 151314843421SMatthew Ahrens dmu_objset_userspace_present(objset_t *os) 151414843421SMatthew Ahrens { 1515503ad85cSMatthew Ahrens return (os->os_phys->os_flags & 151614843421SMatthew Ahrens OBJSET_FLAG_USERACCOUNTING_COMPLETE); 151714843421SMatthew Ahrens } 151814843421SMatthew Ahrens 151914843421SMatthew Ahrens int 152014843421SMatthew Ahrens dmu_objset_userspace_upgrade(objset_t *os) 152114843421SMatthew Ahrens { 152214843421SMatthew Ahrens uint64_t obj; 152314843421SMatthew Ahrens int err = 0; 152414843421SMatthew Ahrens 152514843421SMatthew Ahrens if (dmu_objset_userspace_present(os)) 152614843421SMatthew Ahrens return (0); 1527503ad85cSMatthew Ahrens if (!dmu_objset_userused_enabled(os)) 1528be6fd75aSMatthew Ahrens return (SET_ERROR(ENOTSUP)); 152914843421SMatthew Ahrens if (dmu_objset_is_snapshot(os)) 1530be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 153114843421SMatthew Ahrens 153214843421SMatthew Ahrens /* 153314843421SMatthew Ahrens * We simply need to mark every object dirty, so that it will be 153414843421SMatthew Ahrens * synced out and now accounted. If this is called 153514843421SMatthew Ahrens * concurrently, or if we already did some work before crashing, 153614843421SMatthew Ahrens * that's fine, since we track each object's accounted state 153714843421SMatthew Ahrens * independently. 153814843421SMatthew Ahrens */ 153914843421SMatthew Ahrens 154014843421SMatthew Ahrens for (obj = 0; err == 0; err = dmu_object_next(os, &obj, FALSE, 0)) { 154168857716SLin Ling dmu_tx_t *tx; 154214843421SMatthew Ahrens dmu_buf_t *db; 154314843421SMatthew Ahrens int objerr; 154414843421SMatthew Ahrens 154514843421SMatthew Ahrens if (issig(JUSTLOOKING) && issig(FORREAL)) 1546be6fd75aSMatthew Ahrens return (SET_ERROR(EINTR)); 154714843421SMatthew Ahrens 154814843421SMatthew Ahrens objerr = dmu_bonus_hold(os, obj, FTAG, &db); 15493b2aab18SMatthew Ahrens if (objerr != 0) 155014843421SMatthew Ahrens continue; 155168857716SLin Ling tx = dmu_tx_create(os); 155214843421SMatthew Ahrens dmu_tx_hold_bonus(tx, obj); 155314843421SMatthew Ahrens objerr = dmu_tx_assign(tx, TXG_WAIT); 15543b2aab18SMatthew Ahrens if (objerr != 0) { 155514843421SMatthew Ahrens dmu_tx_abort(tx); 155614843421SMatthew Ahrens continue; 155714843421SMatthew Ahrens } 155814843421SMatthew Ahrens dmu_buf_will_dirty(db, tx); 155914843421SMatthew Ahrens dmu_buf_rele(db, FTAG); 156014843421SMatthew Ahrens dmu_tx_commit(tx); 156114843421SMatthew Ahrens } 156214843421SMatthew Ahrens 1563503ad85cSMatthew Ahrens os->os_flags |= OBJSET_FLAG_USERACCOUNTING_COMPLETE; 156414843421SMatthew Ahrens txg_wait_synced(dmu_objset_pool(os), 0); 156514843421SMatthew Ahrens return (0); 156614843421SMatthew Ahrens } 156714843421SMatthew Ahrens 1568fa9e4066Sahrens void 1569a2eea2e1Sahrens dmu_objset_space(objset_t *os, uint64_t *refdbytesp, uint64_t *availbytesp, 1570a2eea2e1Sahrens uint64_t *usedobjsp, uint64_t *availobjsp) 1571fa9e4066Sahrens { 1572503ad85cSMatthew Ahrens dsl_dataset_space(os->os_dsl_dataset, refdbytesp, availbytesp, 1573a2eea2e1Sahrens usedobjsp, availobjsp); 1574a2eea2e1Sahrens } 1575a2eea2e1Sahrens 1576a2eea2e1Sahrens uint64_t 1577a2eea2e1Sahrens dmu_objset_fsid_guid(objset_t *os) 1578a2eea2e1Sahrens { 1579503ad85cSMatthew Ahrens return (dsl_dataset_fsid_guid(os->os_dsl_dataset)); 1580a2eea2e1Sahrens } 1581a2eea2e1Sahrens 1582a2eea2e1Sahrens void 1583a2eea2e1Sahrens dmu_objset_fast_stat(objset_t *os, dmu_objset_stats_t *stat) 1584a2eea2e1Sahrens { 1585503ad85cSMatthew Ahrens stat->dds_type = os->os_phys->os_type; 1586503ad85cSMatthew Ahrens if (os->os_dsl_dataset) 1587503ad85cSMatthew Ahrens dsl_dataset_fast_stat(os->os_dsl_dataset, stat); 1588a2eea2e1Sahrens } 1589a2eea2e1Sahrens 1590a2eea2e1Sahrens void 1591a2eea2e1Sahrens dmu_objset_stats(objset_t *os, nvlist_t *nv) 1592a2eea2e1Sahrens { 1593503ad85cSMatthew Ahrens ASSERT(os->os_dsl_dataset || 1594503ad85cSMatthew Ahrens os->os_phys->os_type == DMU_OST_META); 1595a2eea2e1Sahrens 1596503ad85cSMatthew Ahrens if (os->os_dsl_dataset != NULL) 1597503ad85cSMatthew Ahrens dsl_dataset_stats(os->os_dsl_dataset, nv); 1598a2eea2e1Sahrens 1599a2eea2e1Sahrens dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_TYPE, 1600503ad85cSMatthew Ahrens os->os_phys->os_type); 160114843421SMatthew Ahrens dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USERACCOUNTING, 160214843421SMatthew Ahrens dmu_objset_userspace_present(os)); 1603fa9e4066Sahrens } 1604fa9e4066Sahrens 1605fa9e4066Sahrens int 1606fa9e4066Sahrens dmu_objset_is_snapshot(objset_t *os) 1607fa9e4066Sahrens { 1608503ad85cSMatthew Ahrens if (os->os_dsl_dataset != NULL) 1609bc9014e6SJustin Gibbs return (os->os_dsl_dataset->ds_is_snapshot); 1610fa9e4066Sahrens else 1611fa9e4066Sahrens return (B_FALSE); 1612fa9e4066Sahrens } 1613fa9e4066Sahrens 1614ab04eb8eStimh int 1615ab04eb8eStimh dmu_snapshot_realname(objset_t *os, char *name, char *real, int maxlen, 1616ab04eb8eStimh boolean_t *conflict) 1617ab04eb8eStimh { 1618503ad85cSMatthew Ahrens dsl_dataset_t *ds = os->os_dsl_dataset; 1619ab04eb8eStimh uint64_t ignored; 1620ab04eb8eStimh 1621c1379625SJustin T. Gibbs if (dsl_dataset_phys(ds)->ds_snapnames_zapobj == 0) 1622be6fd75aSMatthew Ahrens return (SET_ERROR(ENOENT)); 1623ab04eb8eStimh 1624ab04eb8eStimh return (zap_lookup_norm(ds->ds_dir->dd_pool->dp_meta_objset, 1625c1379625SJustin T. Gibbs dsl_dataset_phys(ds)->ds_snapnames_zapobj, name, 8, 1, &ignored, 1626c1379625SJustin T. Gibbs MT_FIRST, real, maxlen, conflict)); 1627ab04eb8eStimh } 1628ab04eb8eStimh 1629fa9e4066Sahrens int 1630fa9e4066Sahrens dmu_snapshot_list_next(objset_t *os, int namelen, char *name, 1631b38f0970Sck uint64_t *idp, uint64_t *offp, boolean_t *case_conflict) 1632fa9e4066Sahrens { 1633503ad85cSMatthew Ahrens dsl_dataset_t *ds = os->os_dsl_dataset; 1634fa9e4066Sahrens zap_cursor_t cursor; 1635fa9e4066Sahrens zap_attribute_t attr; 1636fa9e4066Sahrens 16373b2aab18SMatthew Ahrens ASSERT(dsl_pool_config_held(dmu_objset_pool(os))); 16383b2aab18SMatthew Ahrens 1639c1379625SJustin T. Gibbs if (dsl_dataset_phys(ds)->ds_snapnames_zapobj == 0) 1640be6fd75aSMatthew Ahrens return (SET_ERROR(ENOENT)); 1641fa9e4066Sahrens 1642fa9e4066Sahrens zap_cursor_init_serialized(&cursor, 1643fa9e4066Sahrens ds->ds_dir->dd_pool->dp_meta_objset, 1644c1379625SJustin T. Gibbs dsl_dataset_phys(ds)->ds_snapnames_zapobj, *offp); 1645fa9e4066Sahrens 164687e5029aSahrens if (zap_cursor_retrieve(&cursor, &attr) != 0) { 164787e5029aSahrens zap_cursor_fini(&cursor); 1648be6fd75aSMatthew Ahrens return (SET_ERROR(ENOENT)); 164987e5029aSahrens } 165087e5029aSahrens 165187e5029aSahrens if (strlen(attr.za_name) + 1 > namelen) { 165287e5029aSahrens zap_cursor_fini(&cursor); 1653be6fd75aSMatthew Ahrens return (SET_ERROR(ENAMETOOLONG)); 165487e5029aSahrens } 165587e5029aSahrens 165687e5029aSahrens (void) strcpy(name, attr.za_name); 165787e5029aSahrens if (idp) 165887e5029aSahrens *idp = attr.za_first_integer; 1659b38f0970Sck if (case_conflict) 1660b38f0970Sck *case_conflict = attr.za_normalization_conflict; 166187e5029aSahrens zap_cursor_advance(&cursor); 166287e5029aSahrens *offp = zap_cursor_serialize(&cursor); 166387e5029aSahrens zap_cursor_fini(&cursor); 166487e5029aSahrens 166587e5029aSahrens return (0); 166687e5029aSahrens } 166787e5029aSahrens 166887e5029aSahrens int 166987e5029aSahrens dmu_dir_list_next(objset_t *os, int namelen, char *name, 167087e5029aSahrens uint64_t *idp, uint64_t *offp) 167187e5029aSahrens { 1672503ad85cSMatthew Ahrens dsl_dir_t *dd = os->os_dsl_dataset->ds_dir; 167387e5029aSahrens zap_cursor_t cursor; 167487e5029aSahrens zap_attribute_t attr; 167587e5029aSahrens 167687e5029aSahrens /* there is no next dir on a snapshot! */ 1677503ad85cSMatthew Ahrens if (os->os_dsl_dataset->ds_object != 1678c1379625SJustin T. Gibbs dsl_dir_phys(dd)->dd_head_dataset_obj) 1679be6fd75aSMatthew Ahrens return (SET_ERROR(ENOENT)); 1680fa9e4066Sahrens 168187e5029aSahrens zap_cursor_init_serialized(&cursor, 168287e5029aSahrens dd->dd_pool->dp_meta_objset, 1683c1379625SJustin T. Gibbs dsl_dir_phys(dd)->dd_child_dir_zapobj, *offp); 168487e5029aSahrens 168587e5029aSahrens if (zap_cursor_retrieve(&cursor, &attr) != 0) { 168687e5029aSahrens zap_cursor_fini(&cursor); 1687be6fd75aSMatthew Ahrens return (SET_ERROR(ENOENT)); 168887e5029aSahrens } 168987e5029aSahrens 169087e5029aSahrens if (strlen(attr.za_name) + 1 > namelen) { 169187e5029aSahrens zap_cursor_fini(&cursor); 1692be6fd75aSMatthew Ahrens return (SET_ERROR(ENAMETOOLONG)); 169387e5029aSahrens } 1694fa9e4066Sahrens 1695fa9e4066Sahrens (void) strcpy(name, attr.za_name); 169687e5029aSahrens if (idp) 169787e5029aSahrens *idp = attr.za_first_integer; 1698fa9e4066Sahrens zap_cursor_advance(&cursor); 1699fa9e4066Sahrens *offp = zap_cursor_serialize(&cursor); 170087e5029aSahrens zap_cursor_fini(&cursor); 1701fa9e4066Sahrens 1702fa9e4066Sahrens return (0); 1703fa9e4066Sahrens } 1704fa9e4066Sahrens 170512380e1eSArne Jansen typedef struct dmu_objset_find_ctx { 170612380e1eSArne Jansen taskq_t *dc_tq; 170712380e1eSArne Jansen dsl_pool_t *dc_dp; 170812380e1eSArne Jansen uint64_t dc_ddobj; 170912380e1eSArne Jansen int (*dc_func)(dsl_pool_t *, dsl_dataset_t *, void *); 171012380e1eSArne Jansen void *dc_arg; 171112380e1eSArne Jansen int dc_flags; 171212380e1eSArne Jansen kmutex_t *dc_error_lock; 171312380e1eSArne Jansen int *dc_error; 171412380e1eSArne Jansen } dmu_objset_find_ctx_t; 171512380e1eSArne Jansen 171612380e1eSArne Jansen static void 171712380e1eSArne Jansen dmu_objset_find_dp_impl(dmu_objset_find_ctx_t *dcp) 1718088f3894Sahrens { 171912380e1eSArne Jansen dsl_pool_t *dp = dcp->dc_dp; 172012380e1eSArne Jansen dmu_objset_find_ctx_t *child_dcp; 17213b2aab18SMatthew Ahrens dsl_dir_t *dd; 17223b2aab18SMatthew Ahrens dsl_dataset_t *ds; 17233b2aab18SMatthew Ahrens zap_cursor_t zc; 17243b2aab18SMatthew Ahrens zap_attribute_t *attr; 17253b2aab18SMatthew Ahrens uint64_t thisobj; 172612380e1eSArne Jansen int err = 0; 17273b2aab18SMatthew Ahrens 172812380e1eSArne Jansen /* don't process if there already was an error */ 172912380e1eSArne Jansen if (*dcp->dc_error != 0) 173012380e1eSArne Jansen goto out; 17313b2aab18SMatthew Ahrens 173212380e1eSArne Jansen err = dsl_dir_hold_obj(dp, dcp->dc_ddobj, NULL, FTAG, &dd); 17333b2aab18SMatthew Ahrens if (err != 0) 173412380e1eSArne Jansen goto out; 17353b2aab18SMatthew Ahrens 17363b2aab18SMatthew Ahrens /* Don't visit hidden ($MOS & $ORIGIN) objsets. */ 17373b2aab18SMatthew Ahrens if (dd->dd_myname[0] == '$') { 17383b2aab18SMatthew Ahrens dsl_dir_rele(dd, FTAG); 173912380e1eSArne Jansen goto out; 17403b2aab18SMatthew Ahrens } 17413b2aab18SMatthew Ahrens 1742c1379625SJustin T. Gibbs thisobj = dsl_dir_phys(dd)->dd_head_dataset_obj; 17433b2aab18SMatthew Ahrens attr = kmem_alloc(sizeof (zap_attribute_t), KM_SLEEP); 17443b2aab18SMatthew Ahrens 17453b2aab18SMatthew Ahrens /* 17463b2aab18SMatthew Ahrens * Iterate over all children. 17473b2aab18SMatthew Ahrens */ 174812380e1eSArne Jansen if (dcp->dc_flags & DS_FIND_CHILDREN) { 17493b2aab18SMatthew Ahrens for (zap_cursor_init(&zc, dp->dp_meta_objset, 1750c1379625SJustin T. Gibbs dsl_dir_phys(dd)->dd_child_dir_zapobj); 17513b2aab18SMatthew Ahrens zap_cursor_retrieve(&zc, attr) == 0; 17523b2aab18SMatthew Ahrens (void) zap_cursor_advance(&zc)) { 17533b2aab18SMatthew Ahrens ASSERT3U(attr->za_integer_length, ==, 17543b2aab18SMatthew Ahrens sizeof (uint64_t)); 17553b2aab18SMatthew Ahrens ASSERT3U(attr->za_num_integers, ==, 1); 17563b2aab18SMatthew Ahrens 175712380e1eSArne Jansen child_dcp = kmem_alloc(sizeof (*child_dcp), KM_SLEEP); 175812380e1eSArne Jansen *child_dcp = *dcp; 175912380e1eSArne Jansen child_dcp->dc_ddobj = attr->za_first_integer; 176012380e1eSArne Jansen if (dcp->dc_tq != NULL) 176112380e1eSArne Jansen (void) taskq_dispatch(dcp->dc_tq, 176212380e1eSArne Jansen dmu_objset_find_dp_cb, child_dcp, TQ_SLEEP); 176312380e1eSArne Jansen else 176412380e1eSArne Jansen dmu_objset_find_dp_impl(child_dcp); 17653b2aab18SMatthew Ahrens } 17663b2aab18SMatthew Ahrens zap_cursor_fini(&zc); 17673b2aab18SMatthew Ahrens } 17683b2aab18SMatthew Ahrens 17693b2aab18SMatthew Ahrens /* 17703b2aab18SMatthew Ahrens * Iterate over all snapshots. 17713b2aab18SMatthew Ahrens */ 177212380e1eSArne Jansen if (dcp->dc_flags & DS_FIND_SNAPSHOTS) { 17733b2aab18SMatthew Ahrens dsl_dataset_t *ds; 17743b2aab18SMatthew Ahrens err = dsl_dataset_hold_obj(dp, thisobj, FTAG, &ds); 17753b2aab18SMatthew Ahrens 17763b2aab18SMatthew Ahrens if (err == 0) { 1777c1379625SJustin T. Gibbs uint64_t snapobj; 1778c1379625SJustin T. Gibbs 1779c1379625SJustin T. Gibbs snapobj = dsl_dataset_phys(ds)->ds_snapnames_zapobj; 17803b2aab18SMatthew Ahrens dsl_dataset_rele(ds, FTAG); 17813b2aab18SMatthew Ahrens 17823b2aab18SMatthew Ahrens for (zap_cursor_init(&zc, dp->dp_meta_objset, snapobj); 17833b2aab18SMatthew Ahrens zap_cursor_retrieve(&zc, attr) == 0; 17843b2aab18SMatthew Ahrens (void) zap_cursor_advance(&zc)) { 17853b2aab18SMatthew Ahrens ASSERT3U(attr->za_integer_length, ==, 17863b2aab18SMatthew Ahrens sizeof (uint64_t)); 17873b2aab18SMatthew Ahrens ASSERT3U(attr->za_num_integers, ==, 1); 17883b2aab18SMatthew Ahrens 17893b2aab18SMatthew Ahrens err = dsl_dataset_hold_obj(dp, 17903b2aab18SMatthew Ahrens attr->za_first_integer, FTAG, &ds); 17913b2aab18SMatthew Ahrens if (err != 0) 17923b2aab18SMatthew Ahrens break; 179312380e1eSArne Jansen err = dcp->dc_func(dp, ds, dcp->dc_arg); 17943b2aab18SMatthew Ahrens dsl_dataset_rele(ds, FTAG); 17953b2aab18SMatthew Ahrens if (err != 0) 17963b2aab18SMatthew Ahrens break; 17973b2aab18SMatthew Ahrens } 17983b2aab18SMatthew Ahrens zap_cursor_fini(&zc); 17993b2aab18SMatthew Ahrens } 18003b2aab18SMatthew Ahrens } 18013b2aab18SMatthew Ahrens 18023b2aab18SMatthew Ahrens dsl_dir_rele(dd, FTAG); 18033b2aab18SMatthew Ahrens kmem_free(attr, sizeof (zap_attribute_t)); 18043b2aab18SMatthew Ahrens 18053b2aab18SMatthew Ahrens if (err != 0) 180612380e1eSArne Jansen goto out; 18073b2aab18SMatthew Ahrens 18083b2aab18SMatthew Ahrens /* 18093b2aab18SMatthew Ahrens * Apply to self. 18103b2aab18SMatthew Ahrens */ 18113b2aab18SMatthew Ahrens err = dsl_dataset_hold_obj(dp, thisobj, FTAG, &ds); 18123b2aab18SMatthew Ahrens if (err != 0) 181312380e1eSArne Jansen goto out; 181412380e1eSArne Jansen err = dcp->dc_func(dp, ds, dcp->dc_arg); 18153b2aab18SMatthew Ahrens dsl_dataset_rele(ds, FTAG); 181612380e1eSArne Jansen 181712380e1eSArne Jansen out: 181812380e1eSArne Jansen if (err != 0) { 181912380e1eSArne Jansen mutex_enter(dcp->dc_error_lock); 182012380e1eSArne Jansen /* only keep first error */ 182112380e1eSArne Jansen if (*dcp->dc_error == 0) 182212380e1eSArne Jansen *dcp->dc_error = err; 182312380e1eSArne Jansen mutex_exit(dcp->dc_error_lock); 182412380e1eSArne Jansen } 182512380e1eSArne Jansen 182612380e1eSArne Jansen kmem_free(dcp, sizeof (*dcp)); 182712380e1eSArne Jansen } 182812380e1eSArne Jansen 182912380e1eSArne Jansen static void 183012380e1eSArne Jansen dmu_objset_find_dp_cb(void *arg) 183112380e1eSArne Jansen { 183212380e1eSArne Jansen dmu_objset_find_ctx_t *dcp = arg; 183312380e1eSArne Jansen dsl_pool_t *dp = dcp->dc_dp; 183412380e1eSArne Jansen 18351d3f896fSArne Jansen /* 18361d3f896fSArne Jansen * We need to get a pool_config_lock here, as there are several 18371d3f896fSArne Jansen * asssert(pool_config_held) down the stack. Getting a lock via 18381d3f896fSArne Jansen * dsl_pool_config_enter is risky, as it might be stalled by a 18391d3f896fSArne Jansen * pending writer. This would deadlock, as the write lock can 18401d3f896fSArne Jansen * only be granted when our parent thread gives up the lock. 18411d3f896fSArne Jansen * The _prio interface gives us priority over a pending writer. 18421d3f896fSArne Jansen */ 18431d3f896fSArne Jansen dsl_pool_config_enter_prio(dp, FTAG); 184412380e1eSArne Jansen 184512380e1eSArne Jansen dmu_objset_find_dp_impl(dcp); 184612380e1eSArne Jansen 184712380e1eSArne Jansen dsl_pool_config_exit(dp, FTAG); 184812380e1eSArne Jansen } 184912380e1eSArne Jansen 185012380e1eSArne Jansen /* 185112380e1eSArne Jansen * Find objsets under and including ddobj, call func(ds) on each. 185212380e1eSArne Jansen * The order for the enumeration is completely undefined. 185312380e1eSArne Jansen * func is called with dsl_pool_config held. 185412380e1eSArne Jansen */ 185512380e1eSArne Jansen int 185612380e1eSArne Jansen dmu_objset_find_dp(dsl_pool_t *dp, uint64_t ddobj, 185712380e1eSArne Jansen int func(dsl_pool_t *, dsl_dataset_t *, void *), void *arg, int flags) 185812380e1eSArne Jansen { 185912380e1eSArne Jansen int error = 0; 186012380e1eSArne Jansen taskq_t *tq = NULL; 186112380e1eSArne Jansen int ntasks; 186212380e1eSArne Jansen dmu_objset_find_ctx_t *dcp; 186312380e1eSArne Jansen kmutex_t err_lock; 186412380e1eSArne Jansen 186512380e1eSArne Jansen mutex_init(&err_lock, NULL, MUTEX_DEFAULT, NULL); 186612380e1eSArne Jansen dcp = kmem_alloc(sizeof (*dcp), KM_SLEEP); 186712380e1eSArne Jansen dcp->dc_tq = NULL; 186812380e1eSArne Jansen dcp->dc_dp = dp; 186912380e1eSArne Jansen dcp->dc_ddobj = ddobj; 187012380e1eSArne Jansen dcp->dc_func = func; 187112380e1eSArne Jansen dcp->dc_arg = arg; 187212380e1eSArne Jansen dcp->dc_flags = flags; 187312380e1eSArne Jansen dcp->dc_error_lock = &err_lock; 187412380e1eSArne Jansen dcp->dc_error = &error; 187512380e1eSArne Jansen 187612380e1eSArne Jansen if ((flags & DS_FIND_SERIALIZE) || dsl_pool_config_held_writer(dp)) { 187712380e1eSArne Jansen /* 187812380e1eSArne Jansen * In case a write lock is held we can't make use of 187912380e1eSArne Jansen * parallelism, as down the stack of the worker threads 188012380e1eSArne Jansen * the lock is asserted via dsl_pool_config_held. 188112380e1eSArne Jansen * In case of a read lock this is solved by getting a read 188212380e1eSArne Jansen * lock in each worker thread, which isn't possible in case 188312380e1eSArne Jansen * of a writer lock. So we fall back to the synchronous path 188412380e1eSArne Jansen * here. 188512380e1eSArne Jansen * In the future it might be possible to get some magic into 188612380e1eSArne Jansen * dsl_pool_config_held in a way that it returns true for 188712380e1eSArne Jansen * the worker threads so that a single lock held from this 188812380e1eSArne Jansen * thread suffices. For now, stay single threaded. 188912380e1eSArne Jansen */ 189012380e1eSArne Jansen dmu_objset_find_dp_impl(dcp); 18912bad2258SSteven Hartland mutex_destroy(&err_lock); 189212380e1eSArne Jansen 189312380e1eSArne Jansen return (error); 189412380e1eSArne Jansen } 189512380e1eSArne Jansen 189612380e1eSArne Jansen ntasks = dmu_find_threads; 189712380e1eSArne Jansen if (ntasks == 0) 189812380e1eSArne Jansen ntasks = vdev_count_leaves(dp->dp_spa) * 4; 189912380e1eSArne Jansen tq = taskq_create("dmu_objset_find", ntasks, minclsyspri, ntasks, 190012380e1eSArne Jansen INT_MAX, 0); 190112380e1eSArne Jansen if (tq == NULL) { 190212380e1eSArne Jansen kmem_free(dcp, sizeof (*dcp)); 19032bad2258SSteven Hartland mutex_destroy(&err_lock); 19042bad2258SSteven Hartland 190512380e1eSArne Jansen return (SET_ERROR(ENOMEM)); 190612380e1eSArne Jansen } 190712380e1eSArne Jansen dcp->dc_tq = tq; 190812380e1eSArne Jansen 190912380e1eSArne Jansen /* dcp will be freed by task */ 191012380e1eSArne Jansen (void) taskq_dispatch(tq, dmu_objset_find_dp_cb, dcp, TQ_SLEEP); 191112380e1eSArne Jansen 191212380e1eSArne Jansen /* 191312380e1eSArne Jansen * PORTING: this code relies on the property of taskq_wait to wait 191412380e1eSArne Jansen * until no more tasks are queued and no more tasks are active. As 191512380e1eSArne Jansen * we always queue new tasks from within other tasks, task_wait 191612380e1eSArne Jansen * reliably waits for the full recursion to finish, even though we 191712380e1eSArne Jansen * enqueue new tasks after taskq_wait has been called. 191812380e1eSArne Jansen * On platforms other than illumos, taskq_wait may not have this 191912380e1eSArne Jansen * property. 192012380e1eSArne Jansen */ 192112380e1eSArne Jansen taskq_wait(tq); 192212380e1eSArne Jansen taskq_destroy(tq); 192312380e1eSArne Jansen mutex_destroy(&err_lock); 192412380e1eSArne Jansen 192512380e1eSArne Jansen return (error); 1926088f3894Sahrens } 1927088f3894Sahrens 1928088f3894Sahrens /* 19293b2aab18SMatthew Ahrens * Find all objsets under name, and for each, call 'func(child_name, arg)'. 19303b2aab18SMatthew Ahrens * The dp_config_rwlock must not be held when this is called, and it 19313b2aab18SMatthew Ahrens * will not be held when the callback is called. 19323b2aab18SMatthew Ahrens * Therefore this function should only be used when the pool is not changing 19333b2aab18SMatthew Ahrens * (e.g. in syncing context), or the callback can deal with the possible races. 1934088f3894Sahrens */ 19353b2aab18SMatthew Ahrens static int 19363b2aab18SMatthew Ahrens dmu_objset_find_impl(spa_t *spa, const char *name, 19373b2aab18SMatthew Ahrens int func(const char *, void *), void *arg, int flags) 1938fa9e4066Sahrens { 1939fa9e4066Sahrens dsl_dir_t *dd; 19403b2aab18SMatthew Ahrens dsl_pool_t *dp = spa_get_dsl(spa); 1941088f3894Sahrens dsl_dataset_t *ds; 1942fa9e4066Sahrens zap_cursor_t zc; 1943b7661cccSmmusante zap_attribute_t *attr; 1944fa9e4066Sahrens char *child; 1945088f3894Sahrens uint64_t thisobj; 1946088f3894Sahrens int err; 1947fa9e4066Sahrens 19483b2aab18SMatthew Ahrens dsl_pool_config_enter(dp, FTAG); 19493b2aab18SMatthew Ahrens 19503b2aab18SMatthew Ahrens err = dsl_dir_hold(dp, name, FTAG, &dd, NULL); 19513b2aab18SMatthew Ahrens if (err != 0) { 19523b2aab18SMatthew Ahrens dsl_pool_config_exit(dp, FTAG); 19531d452cf5Sahrens return (err); 19543b2aab18SMatthew Ahrens } 1955fa9e4066Sahrens 1956088f3894Sahrens /* Don't visit hidden ($MOS & $ORIGIN) objsets. */ 1957088f3894Sahrens if (dd->dd_myname[0] == '$') { 19583b2aab18SMatthew Ahrens dsl_dir_rele(dd, FTAG); 19593b2aab18SMatthew Ahrens dsl_pool_config_exit(dp, FTAG); 1960088f3894Sahrens return (0); 1961088f3894Sahrens } 1962088f3894Sahrens 1963c1379625SJustin T. Gibbs thisobj = dsl_dir_phys(dd)->dd_head_dataset_obj; 1964b7661cccSmmusante attr = kmem_alloc(sizeof (zap_attribute_t), KM_SLEEP); 1965fa9e4066Sahrens 1966fa9e4066Sahrens /* 1967fa9e4066Sahrens * Iterate over all children. 1968fa9e4066Sahrens */ 19690b69c2f0Sahrens if (flags & DS_FIND_CHILDREN) { 1970088f3894Sahrens for (zap_cursor_init(&zc, dp->dp_meta_objset, 1971c1379625SJustin T. Gibbs dsl_dir_phys(dd)->dd_child_dir_zapobj); 1972b7661cccSmmusante zap_cursor_retrieve(&zc, attr) == 0; 19730b69c2f0Sahrens (void) zap_cursor_advance(&zc)) { 19743b2aab18SMatthew Ahrens ASSERT3U(attr->za_integer_length, ==, 19753b2aab18SMatthew Ahrens sizeof (uint64_t)); 19763b2aab18SMatthew Ahrens ASSERT3U(attr->za_num_integers, ==, 1); 1977fa9e4066Sahrens 1978486ae710SMatthew Ahrens child = kmem_asprintf("%s/%s", name, attr->za_name); 19793b2aab18SMatthew Ahrens dsl_pool_config_exit(dp, FTAG); 19803b2aab18SMatthew Ahrens err = dmu_objset_find_impl(spa, child, 19813b2aab18SMatthew Ahrens func, arg, flags); 19823b2aab18SMatthew Ahrens dsl_pool_config_enter(dp, FTAG); 1983486ae710SMatthew Ahrens strfree(child); 19843b2aab18SMatthew Ahrens if (err != 0) 19850b69c2f0Sahrens break; 19860b69c2f0Sahrens } 19870b69c2f0Sahrens zap_cursor_fini(&zc); 19881d452cf5Sahrens 19893b2aab18SMatthew Ahrens if (err != 0) { 19903b2aab18SMatthew Ahrens dsl_dir_rele(dd, FTAG); 19913b2aab18SMatthew Ahrens dsl_pool_config_exit(dp, FTAG); 1992b7661cccSmmusante kmem_free(attr, sizeof (zap_attribute_t)); 19930b69c2f0Sahrens return (err); 19940b69c2f0Sahrens } 1995fa9e4066Sahrens } 1996fa9e4066Sahrens 1997fa9e4066Sahrens /* 1998fa9e4066Sahrens * Iterate over all snapshots. 1999fa9e4066Sahrens */ 2000088f3894Sahrens if (flags & DS_FIND_SNAPSHOTS) { 2001088f3894Sahrens err = dsl_dataset_hold_obj(dp, thisobj, FTAG, &ds); 2002088f3894Sahrens 2003088f3894Sahrens if (err == 0) { 2004c1379625SJustin T. Gibbs uint64_t snapobj; 2005c1379625SJustin T. Gibbs 2006c1379625SJustin T. Gibbs snapobj = dsl_dataset_phys(ds)->ds_snapnames_zapobj; 2007088f3894Sahrens dsl_dataset_rele(ds, FTAG); 2008088f3894Sahrens 2009088f3894Sahrens for (zap_cursor_init(&zc, dp->dp_meta_objset, snapobj); 2010088f3894Sahrens zap_cursor_retrieve(&zc, attr) == 0; 2011088f3894Sahrens (void) zap_cursor_advance(&zc)) { 20123b2aab18SMatthew Ahrens ASSERT3U(attr->za_integer_length, ==, 2013088f3894Sahrens sizeof (uint64_t)); 20143b2aab18SMatthew Ahrens ASSERT3U(attr->za_num_integers, ==, 1); 2015088f3894Sahrens 2016486ae710SMatthew Ahrens child = kmem_asprintf("%s@%s", 2017486ae710SMatthew Ahrens name, attr->za_name); 20183b2aab18SMatthew Ahrens dsl_pool_config_exit(dp, FTAG); 20193b2aab18SMatthew Ahrens err = func(child, arg); 20203b2aab18SMatthew Ahrens dsl_pool_config_enter(dp, FTAG); 2021486ae710SMatthew Ahrens strfree(child); 20223b2aab18SMatthew Ahrens if (err != 0) 2023088f3894Sahrens break; 2024088f3894Sahrens } 2025088f3894Sahrens zap_cursor_fini(&zc); 2026fa9e4066Sahrens } 2027fa9e4066Sahrens } 2028fa9e4066Sahrens 20293b2aab18SMatthew Ahrens dsl_dir_rele(dd, FTAG); 2030b7661cccSmmusante kmem_free(attr, sizeof (zap_attribute_t)); 20313b2aab18SMatthew Ahrens dsl_pool_config_exit(dp, FTAG); 2032fa9e4066Sahrens 20333b2aab18SMatthew Ahrens if (err != 0) 20341d452cf5Sahrens return (err); 20351d452cf5Sahrens 20363b2aab18SMatthew Ahrens /* Apply to self. */ 20373b2aab18SMatthew Ahrens return (func(name, arg)); 2038fa9e4066Sahrens } 2039f18faf3fSek 20403b2aab18SMatthew Ahrens /* 20413b2aab18SMatthew Ahrens * See comment above dmu_objset_find_impl(). 20423b2aab18SMatthew Ahrens */ 20437f73c863SRich Morris int 20443b2aab18SMatthew Ahrens dmu_objset_find(char *name, int func(const char *, void *), void *arg, 20453b2aab18SMatthew Ahrens int flags) 20467f73c863SRich Morris { 20473b2aab18SMatthew Ahrens spa_t *spa; 20483b2aab18SMatthew Ahrens int error; 20497f73c863SRich Morris 20503b2aab18SMatthew Ahrens error = spa_open(name, &spa, FTAG); 20513b2aab18SMatthew Ahrens if (error != 0) 20523b2aab18SMatthew Ahrens return (error); 20533b2aab18SMatthew Ahrens error = dmu_objset_find_impl(spa, name, func, arg, flags); 20543b2aab18SMatthew Ahrens spa_close(spa, FTAG); 20553b2aab18SMatthew Ahrens return (error); 20567f73c863SRich Morris } 20577f73c863SRich Morris 2058f18faf3fSek void 2059f18faf3fSek dmu_objset_set_user(objset_t *os, void *user_ptr) 2060f18faf3fSek { 2061503ad85cSMatthew Ahrens ASSERT(MUTEX_HELD(&os->os_user_ptr_lock)); 2062503ad85cSMatthew Ahrens os->os_user_ptr = user_ptr; 2063f18faf3fSek } 2064f18faf3fSek 2065f18faf3fSek void * 2066f18faf3fSek dmu_objset_get_user(objset_t *os) 2067f18faf3fSek { 2068503ad85cSMatthew Ahrens ASSERT(MUTEX_HELD(&os->os_user_ptr_lock)); 2069503ad85cSMatthew Ahrens return (os->os_user_ptr); 2070f18faf3fSek } 20713b2aab18SMatthew Ahrens 20723b2aab18SMatthew Ahrens /* 20733b2aab18SMatthew Ahrens * Determine name of filesystem, given name of snapshot. 20749adfa60dSMatthew Ahrens * buf must be at least ZFS_MAX_DATASET_NAME_LEN bytes 20753b2aab18SMatthew Ahrens */ 20763b2aab18SMatthew Ahrens int 20773b2aab18SMatthew Ahrens dmu_fsname(const char *snapname, char *buf) 20783b2aab18SMatthew Ahrens { 20793b2aab18SMatthew Ahrens char *atp = strchr(snapname, '@'); 20803b2aab18SMatthew Ahrens if (atp == NULL) 2081be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 20829adfa60dSMatthew Ahrens if (atp - snapname >= ZFS_MAX_DATASET_NAME_LEN) 2083be6fd75aSMatthew Ahrens return (SET_ERROR(ENAMETOOLONG)); 20843b2aab18SMatthew Ahrens (void) strlcpy(buf, snapname, atp - snapname + 1); 20853b2aab18SMatthew Ahrens return (0); 20863b2aab18SMatthew Ahrens } 2087