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 */ 211c17160aSKevin Crowe 22fa9e4066Sahrens /* 2306e0070dSMark Shellenbaum * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. 2494c2d0ebSMatthew Ahrens * Copyright (c) 2012, 2017 by Delphix. All rights reserved. 25aad02571SSaso Kiselkov * Copyright (c) 2013 by Saso Kiselkov. All rights reserved. 26a2afb611SJerry Jelinek * Copyright (c) 2013, Joyent, Inc. All rights reserved. 27bc9014e6SJustin Gibbs * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved. 2812380e1eSArne Jansen * Copyright (c) 2015, STRATO AG, Inc. All rights reserved. 29c3d26abcSMatthew Ahrens * Copyright (c) 2014 Integros [integros.com] 301c17160aSKevin Crowe * Copyright 2017 Nexenta Systems, Inc. 31fa9e4066Sahrens */ 32fa9e4066Sahrens 3355da60b9SMark J Musante /* Portions Copyright 2010 Robert Milkowski */ 3455da60b9SMark J Musante 35ecd6cf80Smarks #include <sys/cred.h> 36fa9e4066Sahrens #include <sys/zfs_context.h> 37fa9e4066Sahrens #include <sys/dmu_objset.h> 38fa9e4066Sahrens #include <sys/dsl_dir.h> 39fa9e4066Sahrens #include <sys/dsl_dataset.h> 40fa9e4066Sahrens #include <sys/dsl_prop.h> 41fa9e4066Sahrens #include <sys/dsl_pool.h> 421d452cf5Sahrens #include <sys/dsl_synctask.h> 43ecd6cf80Smarks #include <sys/dsl_deleg.h> 44fa9e4066Sahrens #include <sys/dnode.h> 45fa9e4066Sahrens #include <sys/dbuf.h> 46a2eea2e1Sahrens #include <sys/zvol.h> 47fa9e4066Sahrens #include <sys/dmu_tx.h> 48fa9e4066Sahrens #include <sys/zap.h> 49fa9e4066Sahrens #include <sys/zil.h> 50fa9e4066Sahrens #include <sys/dmu_impl.h> 51ecd6cf80Smarks #include <sys/zfs_ioctl.h> 520a586ceaSMark Shellenbaum #include <sys/sa.h> 5399d5e173STim Haley #include <sys/zfs_onexit.h> 543b2aab18SMatthew Ahrens #include <sys/dsl_destroy.h> 5512380e1eSArne Jansen #include <sys/vdev.h> 565cabbc6bSPrashanth Sreenivasa #include <sys/zfeature.h> 575ac95da7SSerapheim Dimitropoulos #include "zfs_namecheck.h" 58fa9e4066Sahrens 59744947dcSTom Erickson /* 60744947dcSTom Erickson * Needed to close a window in dnode_move() that allows the objset to be freed 61744947dcSTom Erickson * before it can be safely accessed. 62744947dcSTom Erickson */ 63744947dcSTom Erickson krwlock_t os_lock; 64744947dcSTom Erickson 6512380e1eSArne Jansen /* 6612380e1eSArne Jansen * Tunable to overwrite the maximum number of threads for the parallization 6712380e1eSArne Jansen * of dmu_objset_find_dp, needed to speed up the import of pools with many 6812380e1eSArne Jansen * datasets. 6912380e1eSArne Jansen * Default is 4 times the number of leaf vdevs. 7012380e1eSArne Jansen */ 7112380e1eSArne Jansen int dmu_find_threads = 0; 7212380e1eSArne Jansen 73af346df5SNed Bass /* 74af346df5SNed Bass * Backfill lower metadnode objects after this many have been freed. 75af346df5SNed Bass * Backfilling negatively impacts object creation rates, so only do it 76af346df5SNed Bass * if there are enough holes to fill. 77af346df5SNed Bass */ 78af346df5SNed Bass int dmu_rescan_dnode_threshold = 131072; 79af346df5SNed Bass 8012380e1eSArne Jansen static void dmu_objset_find_dp_cb(void *arg); 8112380e1eSArne Jansen 82744947dcSTom Erickson void 83744947dcSTom Erickson dmu_objset_init(void) 84744947dcSTom Erickson { 85744947dcSTom Erickson rw_init(&os_lock, NULL, RW_DEFAULT, NULL); 86744947dcSTom Erickson } 87744947dcSTom Erickson 88744947dcSTom Erickson void 89744947dcSTom Erickson dmu_objset_fini(void) 90744947dcSTom Erickson { 91744947dcSTom Erickson rw_destroy(&os_lock); 92744947dcSTom Erickson } 93744947dcSTom Erickson 94fa9e4066Sahrens spa_t * 95fa9e4066Sahrens dmu_objset_spa(objset_t *os) 96fa9e4066Sahrens { 97503ad85cSMatthew Ahrens return (os->os_spa); 98fa9e4066Sahrens } 99fa9e4066Sahrens 100fa9e4066Sahrens zilog_t * 101fa9e4066Sahrens dmu_objset_zil(objset_t *os) 102fa9e4066Sahrens { 103503ad85cSMatthew Ahrens return (os->os_zil); 104fa9e4066Sahrens } 105fa9e4066Sahrens 106fa9e4066Sahrens dsl_pool_t * 107fa9e4066Sahrens dmu_objset_pool(objset_t *os) 108fa9e4066Sahrens { 109fa9e4066Sahrens dsl_dataset_t *ds; 110fa9e4066Sahrens 111503ad85cSMatthew Ahrens if ((ds = os->os_dsl_dataset) != NULL && ds->ds_dir) 112fa9e4066Sahrens return (ds->ds_dir->dd_pool); 113fa9e4066Sahrens else 114503ad85cSMatthew Ahrens return (spa_get_dsl(os->os_spa)); 115fa9e4066Sahrens } 116fa9e4066Sahrens 117fa9e4066Sahrens dsl_dataset_t * 118fa9e4066Sahrens dmu_objset_ds(objset_t *os) 119fa9e4066Sahrens { 120503ad85cSMatthew Ahrens return (os->os_dsl_dataset); 121fa9e4066Sahrens } 122fa9e4066Sahrens 123fa9e4066Sahrens dmu_objset_type_t 124fa9e4066Sahrens dmu_objset_type(objset_t *os) 125fa9e4066Sahrens { 126503ad85cSMatthew Ahrens return (os->os_phys->os_type); 127fa9e4066Sahrens } 128fa9e4066Sahrens 129fa9e4066Sahrens void 130fa9e4066Sahrens dmu_objset_name(objset_t *os, char *buf) 131fa9e4066Sahrens { 132503ad85cSMatthew Ahrens dsl_dataset_name(os->os_dsl_dataset, buf); 133fa9e4066Sahrens } 134fa9e4066Sahrens 135fa9e4066Sahrens uint64_t 136fa9e4066Sahrens dmu_objset_id(objset_t *os) 137fa9e4066Sahrens { 138503ad85cSMatthew Ahrens dsl_dataset_t *ds = os->os_dsl_dataset; 139fa9e4066Sahrens 140fa9e4066Sahrens return (ds ? ds->ds_object : 0); 141fa9e4066Sahrens } 142fa9e4066Sahrens 14354811da5SToomas Soome uint64_t 14454811da5SToomas Soome dmu_objset_dnodesize(objset_t *os) 14554811da5SToomas Soome { 14654811da5SToomas Soome return (os->os_dnodesize); 14754811da5SToomas Soome } 14854811da5SToomas Soome 149edf345e6SMatthew Ahrens zfs_sync_type_t 15055da60b9SMark J Musante dmu_objset_syncprop(objset_t *os) 15155da60b9SMark J Musante { 15255da60b9SMark J Musante return (os->os_sync); 15355da60b9SMark J Musante } 15455da60b9SMark J Musante 155edf345e6SMatthew Ahrens zfs_logbias_op_t 156e09fa4daSNeil Perrin dmu_objset_logbias(objset_t *os) 157e09fa4daSNeil Perrin { 158e09fa4daSNeil Perrin return (os->os_logbias); 159e09fa4daSNeil Perrin } 160e09fa4daSNeil Perrin 161fa9e4066Sahrens static void 162fa9e4066Sahrens checksum_changed_cb(void *arg, uint64_t newval) 163fa9e4066Sahrens { 164503ad85cSMatthew Ahrens objset_t *os = arg; 165fa9e4066Sahrens 166fa9e4066Sahrens /* 167fa9e4066Sahrens * Inheritance should have been done by now. 168fa9e4066Sahrens */ 169fa9e4066Sahrens ASSERT(newval != ZIO_CHECKSUM_INHERIT); 170fa9e4066Sahrens 171503ad85cSMatthew Ahrens os->os_checksum = zio_checksum_select(newval, ZIO_CHECKSUM_ON_VALUE); 172fa9e4066Sahrens } 173fa9e4066Sahrens 174fa9e4066Sahrens static void 175fa9e4066Sahrens compression_changed_cb(void *arg, uint64_t newval) 176fa9e4066Sahrens { 177503ad85cSMatthew Ahrens objset_t *os = arg; 178fa9e4066Sahrens 179fa9e4066Sahrens /* 180fa9e4066Sahrens * Inheritance and range checking should have been done by now. 181fa9e4066Sahrens */ 182fa9e4066Sahrens ASSERT(newval != ZIO_COMPRESS_INHERIT); 183fa9e4066Sahrens 184db1741f5SJustin T. Gibbs os->os_compress = zio_compress_select(os->os_spa, newval, 185db1741f5SJustin T. Gibbs ZIO_COMPRESS_ON); 186fa9e4066Sahrens } 187fa9e4066Sahrens 188d0ad202dSahrens static void 189d0ad202dSahrens copies_changed_cb(void *arg, uint64_t newval) 190d0ad202dSahrens { 191503ad85cSMatthew Ahrens objset_t *os = arg; 192d0ad202dSahrens 193d0ad202dSahrens /* 194d0ad202dSahrens * Inheritance and range checking should have been done by now. 195d0ad202dSahrens */ 196d0ad202dSahrens ASSERT(newval > 0); 197503ad85cSMatthew Ahrens ASSERT(newval <= spa_max_replication(os->os_spa)); 198d0ad202dSahrens 199503ad85cSMatthew Ahrens os->os_copies = newval; 200d0ad202dSahrens } 201d0ad202dSahrens 202b24ab676SJeff Bonwick static void 203b24ab676SJeff Bonwick dedup_changed_cb(void *arg, uint64_t newval) 204b24ab676SJeff Bonwick { 205b24ab676SJeff Bonwick objset_t *os = arg; 206b24ab676SJeff Bonwick spa_t *spa = os->os_spa; 207b24ab676SJeff Bonwick enum zio_checksum checksum; 208b24ab676SJeff Bonwick 209b24ab676SJeff Bonwick /* 210b24ab676SJeff Bonwick * Inheritance should have been done by now. 211b24ab676SJeff Bonwick */ 212b24ab676SJeff Bonwick ASSERT(newval != ZIO_CHECKSUM_INHERIT); 213b24ab676SJeff Bonwick 214b24ab676SJeff Bonwick checksum = zio_checksum_dedup_select(spa, newval, ZIO_CHECKSUM_OFF); 215b24ab676SJeff Bonwick 216b24ab676SJeff Bonwick os->os_dedup_checksum = checksum & ZIO_CHECKSUM_MASK; 217b24ab676SJeff Bonwick os->os_dedup_verify = !!(checksum & ZIO_CHECKSUM_VERIFY); 218b24ab676SJeff Bonwick } 219b24ab676SJeff Bonwick 2203baa08fcSek static void 2213baa08fcSek primary_cache_changed_cb(void *arg, uint64_t newval) 2223baa08fcSek { 223503ad85cSMatthew Ahrens objset_t *os = arg; 2243baa08fcSek 2253baa08fcSek /* 2263baa08fcSek * Inheritance and range checking should have been done by now. 2273baa08fcSek */ 2283baa08fcSek ASSERT(newval == ZFS_CACHE_ALL || newval == ZFS_CACHE_NONE || 2293baa08fcSek newval == ZFS_CACHE_METADATA); 2303baa08fcSek 231503ad85cSMatthew Ahrens os->os_primary_cache = newval; 2323baa08fcSek } 2333baa08fcSek 2343baa08fcSek static void 2353baa08fcSek secondary_cache_changed_cb(void *arg, uint64_t newval) 2363baa08fcSek { 237503ad85cSMatthew Ahrens objset_t *os = arg; 2383baa08fcSek 2393baa08fcSek /* 2403baa08fcSek * Inheritance and range checking should have been done by now. 2413baa08fcSek */ 2423baa08fcSek ASSERT(newval == ZFS_CACHE_ALL || newval == ZFS_CACHE_NONE || 2433baa08fcSek newval == ZFS_CACHE_METADATA); 2443baa08fcSek 245503ad85cSMatthew Ahrens os->os_secondary_cache = newval; 2463baa08fcSek } 2473baa08fcSek 24855da60b9SMark J Musante static void 24955da60b9SMark J Musante sync_changed_cb(void *arg, uint64_t newval) 25055da60b9SMark J Musante { 25155da60b9SMark J Musante objset_t *os = arg; 25255da60b9SMark J Musante 25355da60b9SMark J Musante /* 25455da60b9SMark J Musante * Inheritance and range checking should have been done by now. 25555da60b9SMark J Musante */ 25655da60b9SMark J Musante ASSERT(newval == ZFS_SYNC_STANDARD || newval == ZFS_SYNC_ALWAYS || 25755da60b9SMark J Musante newval == ZFS_SYNC_DISABLED); 25855da60b9SMark J Musante 25955da60b9SMark J Musante os->os_sync = newval; 26055da60b9SMark J Musante if (os->os_zil) 26155da60b9SMark J Musante zil_set_sync(os->os_zil, newval); 26255da60b9SMark J Musante } 26355da60b9SMark J Musante 264edf345e6SMatthew Ahrens static void 265edf345e6SMatthew Ahrens redundant_metadata_changed_cb(void *arg, uint64_t newval) 266edf345e6SMatthew Ahrens { 267edf345e6SMatthew Ahrens objset_t *os = arg; 268edf345e6SMatthew Ahrens 269edf345e6SMatthew Ahrens /* 270edf345e6SMatthew Ahrens * Inheritance and range checking should have been done by now. 271edf345e6SMatthew Ahrens */ 272edf345e6SMatthew Ahrens ASSERT(newval == ZFS_REDUNDANT_METADATA_ALL || 273edf345e6SMatthew Ahrens newval == ZFS_REDUNDANT_METADATA_MOST); 274edf345e6SMatthew Ahrens 275edf345e6SMatthew Ahrens os->os_redundant_metadata = newval; 276edf345e6SMatthew Ahrens } 277edf345e6SMatthew Ahrens 27854811da5SToomas Soome static void 27954811da5SToomas Soome dnodesize_changed_cb(void *arg, uint64_t newval) 28054811da5SToomas Soome { 28154811da5SToomas Soome objset_t *os = arg; 28254811da5SToomas Soome 28354811da5SToomas Soome switch (newval) { 28454811da5SToomas Soome case ZFS_DNSIZE_LEGACY: 28554811da5SToomas Soome os->os_dnodesize = DNODE_MIN_SIZE; 28654811da5SToomas Soome break; 28754811da5SToomas Soome case ZFS_DNSIZE_AUTO: 28854811da5SToomas Soome /* 28954811da5SToomas Soome * Choose a dnode size that will work well for most 29054811da5SToomas Soome * workloads if the user specified "auto". Future code 29154811da5SToomas Soome * improvements could dynamically select a dnode size 29254811da5SToomas Soome * based on observed workload patterns. 29354811da5SToomas Soome */ 29454811da5SToomas Soome os->os_dnodesize = DNODE_MIN_SIZE * 2; 29554811da5SToomas Soome break; 29654811da5SToomas Soome case ZFS_DNSIZE_1K: 29754811da5SToomas Soome case ZFS_DNSIZE_2K: 29854811da5SToomas Soome case ZFS_DNSIZE_4K: 29954811da5SToomas Soome case ZFS_DNSIZE_8K: 30054811da5SToomas Soome case ZFS_DNSIZE_16K: 30154811da5SToomas Soome os->os_dnodesize = newval; 30254811da5SToomas Soome break; 30354811da5SToomas Soome } 30454811da5SToomas Soome } 30554811da5SToomas Soome 306663207adSDon Brady static void 307663207adSDon Brady smallblk_changed_cb(void *arg, uint64_t newval) 308663207adSDon Brady { 309663207adSDon Brady objset_t *os = arg; 310663207adSDon Brady 311663207adSDon Brady /* 312663207adSDon Brady * Inheritance and range checking should have been done by now. 313663207adSDon Brady */ 314663207adSDon Brady ASSERT(newval <= SPA_OLD_MAXBLOCKSIZE); 315663207adSDon Brady ASSERT(ISP2(newval)); 316663207adSDon Brady 317663207adSDon Brady os->os_zpl_special_smallblock = newval; 318663207adSDon Brady } 319663207adSDon Brady 320e09fa4daSNeil Perrin static void 321e09fa4daSNeil Perrin logbias_changed_cb(void *arg, uint64_t newval) 322e09fa4daSNeil Perrin { 323e09fa4daSNeil Perrin objset_t *os = arg; 324e09fa4daSNeil Perrin 325e09fa4daSNeil Perrin ASSERT(newval == ZFS_LOGBIAS_LATENCY || 326e09fa4daSNeil Perrin newval == ZFS_LOGBIAS_THROUGHPUT); 327e09fa4daSNeil Perrin os->os_logbias = newval; 328e09fa4daSNeil Perrin if (os->os_zil) 329e09fa4daSNeil Perrin zil_set_logbias(os->os_zil, newval); 330e09fa4daSNeil Perrin } 331e09fa4daSNeil Perrin 332b5152584SMatthew Ahrens static void 333b5152584SMatthew Ahrens recordsize_changed_cb(void *arg, uint64_t newval) 334b5152584SMatthew Ahrens { 335b5152584SMatthew Ahrens objset_t *os = arg; 336b5152584SMatthew Ahrens 337b5152584SMatthew Ahrens os->os_recordsize = newval; 338b5152584SMatthew Ahrens } 339b5152584SMatthew Ahrens 340fa9e4066Sahrens void 341fa9e4066Sahrens dmu_objset_byteswap(void *buf, size_t size) 342fa9e4066Sahrens { 343fa9e4066Sahrens objset_phys_t *osp = buf; 344fa9e4066Sahrens 34514843421SMatthew Ahrens ASSERT(size == OBJSET_OLD_PHYS_SIZE || size == sizeof (objset_phys_t)); 346fa9e4066Sahrens dnode_byteswap(&osp->os_meta_dnode); 347fa9e4066Sahrens byteswap_uint64_array(&osp->os_zil_header, sizeof (zil_header_t)); 348fa9e4066Sahrens osp->os_type = BSWAP_64(osp->os_type); 34914843421SMatthew Ahrens osp->os_flags = BSWAP_64(osp->os_flags); 35014843421SMatthew Ahrens if (size == sizeof (objset_phys_t)) { 35114843421SMatthew Ahrens dnode_byteswap(&osp->os_userused_dnode); 35214843421SMatthew Ahrens dnode_byteswap(&osp->os_groupused_dnode); 35314843421SMatthew Ahrens } 354fa9e4066Sahrens } 355fa9e4066Sahrens 35694c2d0ebSMatthew Ahrens /* 35794c2d0ebSMatthew Ahrens * The hash is a CRC-based hash of the objset_t pointer and the object number. 35894c2d0ebSMatthew Ahrens */ 35994c2d0ebSMatthew Ahrens static uint64_t 36094c2d0ebSMatthew Ahrens dnode_hash(const objset_t *os, uint64_t obj) 36194c2d0ebSMatthew Ahrens { 36294c2d0ebSMatthew Ahrens uintptr_t osv = (uintptr_t)os; 36394c2d0ebSMatthew Ahrens uint64_t crc = -1ULL; 36494c2d0ebSMatthew Ahrens 36594c2d0ebSMatthew Ahrens ASSERT(zfs_crc64_table[128] == ZFS_CRC64_POLY); 36694c2d0ebSMatthew Ahrens /* 36794c2d0ebSMatthew Ahrens * The low 6 bits of the pointer don't have much entropy, because 36894c2d0ebSMatthew Ahrens * the objset_t is larger than 2^6 bytes long. 36994c2d0ebSMatthew Ahrens */ 37094c2d0ebSMatthew Ahrens crc = (crc >> 8) ^ zfs_crc64_table[(crc ^ (osv >> 6)) & 0xFF]; 37194c2d0ebSMatthew Ahrens crc = (crc >> 8) ^ zfs_crc64_table[(crc ^ (obj >> 0)) & 0xFF]; 37294c2d0ebSMatthew Ahrens crc = (crc >> 8) ^ zfs_crc64_table[(crc ^ (obj >> 8)) & 0xFF]; 37394c2d0ebSMatthew Ahrens crc = (crc >> 8) ^ zfs_crc64_table[(crc ^ (obj >> 16)) & 0xFF]; 37494c2d0ebSMatthew Ahrens 37594c2d0ebSMatthew Ahrens crc ^= (osv>>14) ^ (obj>>24); 37694c2d0ebSMatthew Ahrens 37794c2d0ebSMatthew Ahrens return (crc); 37894c2d0ebSMatthew Ahrens } 37994c2d0ebSMatthew Ahrens 38094c2d0ebSMatthew Ahrens unsigned int 38194c2d0ebSMatthew Ahrens dnode_multilist_index_func(multilist_t *ml, void *obj) 38294c2d0ebSMatthew Ahrens { 38394c2d0ebSMatthew Ahrens dnode_t *dn = obj; 38494c2d0ebSMatthew Ahrens return (dnode_hash(dn->dn_objset, dn->dn_object) % 38594c2d0ebSMatthew Ahrens multilist_get_num_sublists(ml)); 38694c2d0ebSMatthew Ahrens } 38794c2d0ebSMatthew Ahrens 38894c2d0ebSMatthew Ahrens /* 38994c2d0ebSMatthew Ahrens * Instantiates the objset_t in-memory structure corresponding to the 39094c2d0ebSMatthew Ahrens * objset_phys_t that's pointed to by the specified blkptr_t. 39194c2d0ebSMatthew Ahrens */ 392ea8dc4b6Seschrock int 393ea8dc4b6Seschrock dmu_objset_open_impl(spa_t *spa, dsl_dataset_t *ds, blkptr_t *bp, 394503ad85cSMatthew Ahrens objset_t **osp) 395fa9e4066Sahrens { 396503ad85cSMatthew Ahrens objset_t *os; 397088f3894Sahrens int i, err; 398fa9e4066Sahrens 39991ebeef5Sahrens ASSERT(ds == NULL || MUTEX_HELD(&ds->ds_opening_lock)); 40091ebeef5Sahrens 401*a3874b8bSToomas Soome #if 0 4025cabbc6bSPrashanth Sreenivasa /* 4035cabbc6bSPrashanth Sreenivasa * The $ORIGIN dataset (if it exists) doesn't have an associated 4045cabbc6bSPrashanth Sreenivasa * objset, so there's no reason to open it. The $ORIGIN dataset 4055cabbc6bSPrashanth Sreenivasa * will not exist on pools older than SPA_VERSION_ORIGIN. 4065cabbc6bSPrashanth Sreenivasa */ 4075cabbc6bSPrashanth Sreenivasa if (ds != NULL && spa_get_dsl(spa) != NULL && 4085cabbc6bSPrashanth Sreenivasa spa_get_dsl(spa)->dp_origin_snap != NULL) { 4095cabbc6bSPrashanth Sreenivasa ASSERT3P(ds->ds_dir, !=, 4105cabbc6bSPrashanth Sreenivasa spa_get_dsl(spa)->dp_origin_snap->ds_dir); 4115cabbc6bSPrashanth Sreenivasa } 412*a3874b8bSToomas Soome #endif 4135cabbc6bSPrashanth Sreenivasa 414503ad85cSMatthew Ahrens os = kmem_zalloc(sizeof (objset_t), KM_SLEEP); 415503ad85cSMatthew Ahrens os->os_dsl_dataset = ds; 416503ad85cSMatthew Ahrens os->os_spa = spa; 417503ad85cSMatthew Ahrens os->os_rootbp = bp; 418503ad85cSMatthew Ahrens if (!BP_IS_HOLE(os->os_rootbp)) { 4197adb730bSGeorge Wilson arc_flags_t aflags = ARC_FLAG_WAIT; 4207802d7bfSMatthew Ahrens zbookmark_phys_t zb; 421b24ab676SJeff Bonwick SET_BOOKMARK(&zb, ds ? ds->ds_object : DMU_META_OBJSET, 422b24ab676SJeff Bonwick ZB_ROOT_OBJECT, ZB_ROOT_LEVEL, ZB_ROOT_BLKID); 423b24ab676SJeff Bonwick 424503ad85cSMatthew Ahrens if (DMU_OS_IS_L2CACHEABLE(os)) 4257adb730bSGeorge Wilson aflags |= ARC_FLAG_L2CACHE; 426ea8dc4b6Seschrock 427503ad85cSMatthew Ahrens dprintf_bp(os->os_rootbp, "reading %s", ""); 4281b912ec7SGeorge Wilson err = arc_read(NULL, spa, os->os_rootbp, 429503ad85cSMatthew Ahrens arc_getbuf_func, &os->os_phys_buf, 43013506d1eSmaybee ZIO_PRIORITY_SYNC_READ, ZIO_FLAG_CANFAIL, &aflags, &zb); 4313b2aab18SMatthew Ahrens if (err != 0) { 432503ad85cSMatthew Ahrens kmem_free(os, sizeof (objset_t)); 433b87f3af3Sperrin /* convert checksum errors into IO errors */ 434b87f3af3Sperrin if (err == ECKSUM) 435be6fd75aSMatthew Ahrens err = SET_ERROR(EIO); 436ea8dc4b6Seschrock return (err); 437ea8dc4b6Seschrock } 43814843421SMatthew Ahrens 43914843421SMatthew Ahrens /* Increase the blocksize if we are permitted. */ 44014843421SMatthew Ahrens if (spa_version(spa) >= SPA_VERSION_USERSPACE && 441503ad85cSMatthew Ahrens arc_buf_size(os->os_phys_buf) < sizeof (objset_phys_t)) { 4425602294fSDan Kimmel arc_buf_t *buf = arc_alloc_buf(spa, &os->os_phys_buf, 4435602294fSDan Kimmel ARC_BUFC_METADATA, sizeof (objset_phys_t)); 44414843421SMatthew Ahrens bzero(buf->b_data, sizeof (objset_phys_t)); 445503ad85cSMatthew Ahrens bcopy(os->os_phys_buf->b_data, buf->b_data, 446503ad85cSMatthew Ahrens arc_buf_size(os->os_phys_buf)); 447dcbf3bd6SGeorge Wilson arc_buf_destroy(os->os_phys_buf, &os->os_phys_buf); 448503ad85cSMatthew Ahrens os->os_phys_buf = buf; 44914843421SMatthew Ahrens } 45014843421SMatthew Ahrens 451503ad85cSMatthew Ahrens os->os_phys = os->os_phys_buf->b_data; 452503ad85cSMatthew Ahrens os->os_flags = os->os_phys->os_flags; 453fa9e4066Sahrens } else { 45414843421SMatthew Ahrens int size = spa_version(spa) >= SPA_VERSION_USERSPACE ? 45514843421SMatthew Ahrens sizeof (objset_phys_t) : OBJSET_OLD_PHYS_SIZE; 4565602294fSDan Kimmel os->os_phys_buf = arc_alloc_buf(spa, &os->os_phys_buf, 4575602294fSDan Kimmel ARC_BUFC_METADATA, size); 458503ad85cSMatthew Ahrens os->os_phys = os->os_phys_buf->b_data; 459503ad85cSMatthew Ahrens bzero(os->os_phys, size); 460fa9e4066Sahrens } 461fa9e4066Sahrens 462fa9e4066Sahrens /* 463fa9e4066Sahrens * Note: the changed_cb will be called once before the register 464fa9e4066Sahrens * func returns, thus changing the checksum/compression from the 4653baa08fcSek * default (fletcher2/off). Snapshots don't need to know about 4663baa08fcSek * checksum/compression/copies. 467fa9e4066Sahrens */ 4685d7b4d43SMatthew Ahrens if (ds != NULL) { 4699c3fd121SMatthew Ahrens boolean_t needlock = B_FALSE; 4709c3fd121SMatthew Ahrens 4719c3fd121SMatthew Ahrens /* 4729c3fd121SMatthew Ahrens * Note: it's valid to open the objset if the dataset is 4739c3fd121SMatthew Ahrens * long-held, in which case the pool_config lock will not 4749c3fd121SMatthew Ahrens * be held. 4759c3fd121SMatthew Ahrens */ 4769c3fd121SMatthew Ahrens if (!dsl_pool_config_held(dmu_objset_pool(os))) { 4779c3fd121SMatthew Ahrens needlock = B_TRUE; 4789c3fd121SMatthew Ahrens dsl_pool_config_enter(dmu_objset_pool(os), FTAG); 4799c3fd121SMatthew Ahrens } 4803b2aab18SMatthew Ahrens err = dsl_prop_register(ds, 4813b2aab18SMatthew Ahrens zfs_prop_to_name(ZFS_PROP_PRIMARYCACHE), 482503ad85cSMatthew Ahrens primary_cache_changed_cb, os); 4833b2aab18SMatthew Ahrens if (err == 0) { 4843b2aab18SMatthew Ahrens err = dsl_prop_register(ds, 4853b2aab18SMatthew Ahrens zfs_prop_to_name(ZFS_PROP_SECONDARYCACHE), 486503ad85cSMatthew Ahrens secondary_cache_changed_cb, os); 4873b2aab18SMatthew Ahrens } 488bc9014e6SJustin Gibbs if (!ds->ds_is_snapshot) { 4893b2aab18SMatthew Ahrens if (err == 0) { 4903b2aab18SMatthew Ahrens err = dsl_prop_register(ds, 4913b2aab18SMatthew Ahrens zfs_prop_to_name(ZFS_PROP_CHECKSUM), 492503ad85cSMatthew Ahrens checksum_changed_cb, os); 4933b2aab18SMatthew Ahrens } 4943b2aab18SMatthew Ahrens if (err == 0) { 4953b2aab18SMatthew Ahrens err = dsl_prop_register(ds, 4963b2aab18SMatthew Ahrens zfs_prop_to_name(ZFS_PROP_COMPRESSION), 497503ad85cSMatthew Ahrens compression_changed_cb, os); 4983b2aab18SMatthew Ahrens } 4993b2aab18SMatthew Ahrens if (err == 0) { 5003b2aab18SMatthew Ahrens err = dsl_prop_register(ds, 5013b2aab18SMatthew Ahrens zfs_prop_to_name(ZFS_PROP_COPIES), 502503ad85cSMatthew Ahrens copies_changed_cb, os); 5033b2aab18SMatthew Ahrens } 5043b2aab18SMatthew Ahrens if (err == 0) { 5053b2aab18SMatthew Ahrens err = dsl_prop_register(ds, 5063b2aab18SMatthew Ahrens zfs_prop_to_name(ZFS_PROP_DEDUP), 507b24ab676SJeff Bonwick dedup_changed_cb, os); 5083b2aab18SMatthew Ahrens } 5093b2aab18SMatthew Ahrens if (err == 0) { 5103b2aab18SMatthew Ahrens err = dsl_prop_register(ds, 5113b2aab18SMatthew Ahrens zfs_prop_to_name(ZFS_PROP_LOGBIAS), 512e09fa4daSNeil Perrin logbias_changed_cb, os); 5133b2aab18SMatthew Ahrens } 5143b2aab18SMatthew Ahrens if (err == 0) { 5153b2aab18SMatthew Ahrens err = dsl_prop_register(ds, 5163b2aab18SMatthew Ahrens zfs_prop_to_name(ZFS_PROP_SYNC), 51755da60b9SMark J Musante sync_changed_cb, os); 5183b2aab18SMatthew Ahrens } 519edf345e6SMatthew Ahrens if (err == 0) { 520edf345e6SMatthew Ahrens err = dsl_prop_register(ds, 521edf345e6SMatthew Ahrens zfs_prop_to_name( 522edf345e6SMatthew Ahrens ZFS_PROP_REDUNDANT_METADATA), 523edf345e6SMatthew Ahrens redundant_metadata_changed_cb, os); 524edf345e6SMatthew Ahrens } 525b5152584SMatthew Ahrens if (err == 0) { 526b5152584SMatthew Ahrens err = dsl_prop_register(ds, 527b5152584SMatthew Ahrens zfs_prop_to_name(ZFS_PROP_RECORDSIZE), 528b5152584SMatthew Ahrens recordsize_changed_cb, os); 529b5152584SMatthew Ahrens } 53054811da5SToomas Soome if (err == 0) { 53154811da5SToomas Soome err = dsl_prop_register(ds, 53254811da5SToomas Soome zfs_prop_to_name(ZFS_PROP_DNODESIZE), 53354811da5SToomas Soome dnodesize_changed_cb, os); 53454811da5SToomas Soome } 535663207adSDon Brady if (err == 0) { 536663207adSDon Brady err = dsl_prop_register(ds, 537663207adSDon Brady zfs_prop_to_name( 538663207adSDon Brady ZFS_PROP_SPECIAL_SMALL_BLOCKS), 539663207adSDon Brady smallblk_changed_cb, os); 540663207adSDon Brady } 5413baa08fcSek } 5429c3fd121SMatthew Ahrens if (needlock) 5439c3fd121SMatthew Ahrens dsl_pool_config_exit(dmu_objset_pool(os), FTAG); 5443b2aab18SMatthew Ahrens if (err != 0) { 545dcbf3bd6SGeorge Wilson arc_buf_destroy(os->os_phys_buf, &os->os_phys_buf); 546503ad85cSMatthew Ahrens kmem_free(os, sizeof (objset_t)); 547ea8dc4b6Seschrock return (err); 548ea8dc4b6Seschrock } 5495d7b4d43SMatthew Ahrens } else { 550fa9e4066Sahrens /* It's the meta-objset. */ 551503ad85cSMatthew Ahrens os->os_checksum = ZIO_CHECKSUM_FLETCHER_4; 552db1741f5SJustin T. Gibbs os->os_compress = ZIO_COMPRESS_ON; 553503ad85cSMatthew Ahrens os->os_copies = spa_max_replication(spa); 554b24ab676SJeff Bonwick os->os_dedup_checksum = ZIO_CHECKSUM_OFF; 555edf345e6SMatthew Ahrens os->os_dedup_verify = B_FALSE; 556edf345e6SMatthew Ahrens os->os_logbias = ZFS_LOGBIAS_LATENCY; 557edf345e6SMatthew Ahrens os->os_sync = ZFS_SYNC_STANDARD; 558503ad85cSMatthew Ahrens os->os_primary_cache = ZFS_CACHE_ALL; 559503ad85cSMatthew Ahrens os->os_secondary_cache = ZFS_CACHE_ALL; 56054811da5SToomas Soome os->os_dnodesize = DNODE_MIN_SIZE; 561fa9e4066Sahrens } 562adb52d92SMatthew Ahrens /* 563adb52d92SMatthew Ahrens * These properties will be filled in by the logic in zfs_get_zplprop() 564adb52d92SMatthew Ahrens * when they are queried for the first time. 565adb52d92SMatthew Ahrens */ 566adb52d92SMatthew Ahrens os->os_version = OBJSET_PROP_UNINITIALIZED; 567adb52d92SMatthew Ahrens os->os_normalization = OBJSET_PROP_UNINITIALIZED; 568adb52d92SMatthew Ahrens os->os_utf8only = OBJSET_PROP_UNINITIALIZED; 569adb52d92SMatthew Ahrens os->os_casesensitivity = OBJSET_PROP_UNINITIALIZED; 570fa9e4066Sahrens 571bc9014e6SJustin Gibbs if (ds == NULL || !ds->ds_is_snapshot) 5726e0cbcaaSMatthew Ahrens os->os_zil_header = os->os_phys->os_zil_header; 573503ad85cSMatthew Ahrens os->os_zil = zil_alloc(os, &os->os_zil_header); 574fa9e4066Sahrens 575fa9e4066Sahrens for (i = 0; i < TXG_SIZE; i++) { 57694c2d0ebSMatthew Ahrens os->os_dirty_dnodes[i] = multilist_create(sizeof (dnode_t), 57794c2d0ebSMatthew Ahrens offsetof(dnode_t, dn_dirty_link[i]), 57894c2d0ebSMatthew Ahrens dnode_multilist_index_func); 579fa9e4066Sahrens } 580503ad85cSMatthew Ahrens list_create(&os->os_dnodes, sizeof (dnode_t), 581fa9e4066Sahrens offsetof(dnode_t, dn_link)); 582503ad85cSMatthew Ahrens list_create(&os->os_downgraded_dbufs, sizeof (dmu_buf_impl_t), 583fa9e4066Sahrens offsetof(dmu_buf_impl_t, db_link)); 584fa9e4066Sahrens 585503ad85cSMatthew Ahrens mutex_init(&os->os_lock, NULL, MUTEX_DEFAULT, NULL); 58694c2d0ebSMatthew Ahrens mutex_init(&os->os_userused_lock, NULL, MUTEX_DEFAULT, NULL); 587503ad85cSMatthew Ahrens mutex_init(&os->os_obj_lock, NULL, MUTEX_DEFAULT, NULL); 588503ad85cSMatthew Ahrens mutex_init(&os->os_user_ptr_lock, NULL, MUTEX_DEFAULT, NULL); 58954811da5SToomas Soome os->os_obj_next_percpu_len = boot_ncpus; 59054811da5SToomas Soome os->os_obj_next_percpu = kmem_zalloc(os->os_obj_next_percpu_len * 59154811da5SToomas Soome sizeof (os->os_obj_next_percpu[0]), KM_SLEEP); 592503ad85cSMatthew Ahrens 593bc9014e6SJustin Gibbs dnode_special_open(os, &os->os_phys->os_meta_dnode, 594bc9014e6SJustin Gibbs DMU_META_DNODE_OBJECT, &os->os_meta_dnode); 595503ad85cSMatthew Ahrens if (arc_buf_size(os->os_phys_buf) >= sizeof (objset_phys_t)) { 596bc9014e6SJustin Gibbs dnode_special_open(os, &os->os_phys->os_userused_dnode, 597bc9014e6SJustin Gibbs DMU_USERUSED_OBJECT, &os->os_userused_dnode); 598bc9014e6SJustin Gibbs dnode_special_open(os, &os->os_phys->os_groupused_dnode, 599bc9014e6SJustin Gibbs DMU_GROUPUSED_OBJECT, &os->os_groupused_dnode); 60014843421SMatthew Ahrens } 601fa9e4066Sahrens 602503ad85cSMatthew Ahrens *osp = os; 603ea8dc4b6Seschrock return (0); 604fa9e4066Sahrens } 605fa9e4066Sahrens 606503ad85cSMatthew Ahrens int 607503ad85cSMatthew Ahrens dmu_objset_from_ds(dsl_dataset_t *ds, objset_t **osp) 6083cb34c60Sahrens { 609503ad85cSMatthew Ahrens int err = 0; 6103cb34c60Sahrens 6119c3fd121SMatthew Ahrens /* 6129c3fd121SMatthew Ahrens * We shouldn't be doing anything with dsl_dataset_t's unless the 6139c3fd121SMatthew Ahrens * pool_config lock is held, or the dataset is long-held. 6149c3fd121SMatthew Ahrens */ 6159c3fd121SMatthew Ahrens ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool) || 6169c3fd121SMatthew Ahrens dsl_dataset_long_held(ds)); 6179c3fd121SMatthew Ahrens 6183cb34c60Sahrens mutex_enter(&ds->ds_opening_lock); 6195d7b4d43SMatthew Ahrens if (ds->ds_objset == NULL) { 6205d7b4d43SMatthew Ahrens objset_t *os; 621c166b69dSPaul Dagnelie rrw_enter(&ds->ds_bp_rwlock, RW_READER, FTAG); 6223cb34c60Sahrens err = dmu_objset_open_impl(dsl_dataset_get_spa(ds), 6235d7b4d43SMatthew Ahrens ds, dsl_dataset_get_blkptr(ds), &os); 624c166b69dSPaul Dagnelie rrw_exit(&ds->ds_bp_rwlock, FTAG); 6255d7b4d43SMatthew Ahrens 6265d7b4d43SMatthew Ahrens if (err == 0) { 6275d7b4d43SMatthew Ahrens mutex_enter(&ds->ds_lock); 6285d7b4d43SMatthew Ahrens ASSERT(ds->ds_objset == NULL); 6295d7b4d43SMatthew Ahrens ds->ds_objset = os; 6305d7b4d43SMatthew Ahrens mutex_exit(&ds->ds_lock); 6315d7b4d43SMatthew Ahrens } 6323cb34c60Sahrens } 6335d7b4d43SMatthew Ahrens *osp = ds->ds_objset; 6343cb34c60Sahrens mutex_exit(&ds->ds_opening_lock); 635503ad85cSMatthew Ahrens return (err); 6363cb34c60Sahrens } 6373cb34c60Sahrens 6383b2aab18SMatthew Ahrens /* 6393b2aab18SMatthew Ahrens * Holds the pool while the objset is held. Therefore only one objset 6403b2aab18SMatthew Ahrens * can be held at a time. 6413b2aab18SMatthew Ahrens */ 6423cb34c60Sahrens int 643503ad85cSMatthew Ahrens dmu_objset_hold(const char *name, void *tag, objset_t **osp) 6443cb34c60Sahrens { 6453b2aab18SMatthew Ahrens dsl_pool_t *dp; 646503ad85cSMatthew Ahrens dsl_dataset_t *ds; 6473cb34c60Sahrens int err; 6483cb34c60Sahrens 6493b2aab18SMatthew Ahrens err = dsl_pool_hold(name, tag, &dp); 6503b2aab18SMatthew Ahrens if (err != 0) 6513b2aab18SMatthew Ahrens return (err); 6523b2aab18SMatthew Ahrens err = dsl_dataset_hold(dp, name, tag, &ds); 6533b2aab18SMatthew Ahrens if (err != 0) { 6543b2aab18SMatthew Ahrens dsl_pool_rele(dp, tag); 655503ad85cSMatthew Ahrens return (err); 6563b2aab18SMatthew Ahrens } 657503ad85cSMatthew Ahrens 658503ad85cSMatthew Ahrens err = dmu_objset_from_ds(ds, osp); 6593b2aab18SMatthew Ahrens if (err != 0) { 660503ad85cSMatthew Ahrens dsl_dataset_rele(ds, tag); 6613b2aab18SMatthew Ahrens dsl_pool_rele(dp, tag); 6623b2aab18SMatthew Ahrens } 663503ad85cSMatthew Ahrens 6643cb34c60Sahrens return (err); 6653cb34c60Sahrens } 6663cb34c60Sahrens 66712380e1eSArne Jansen static int 66812380e1eSArne Jansen dmu_objset_own_impl(dsl_dataset_t *ds, dmu_objset_type_t type, 66912380e1eSArne Jansen boolean_t readonly, void *tag, objset_t **osp) 67012380e1eSArne Jansen { 67112380e1eSArne Jansen int err; 67212380e1eSArne Jansen 67312380e1eSArne Jansen err = dmu_objset_from_ds(ds, osp); 67412380e1eSArne Jansen if (err != 0) { 67512380e1eSArne Jansen dsl_dataset_disown(ds, tag); 67612380e1eSArne Jansen } else if (type != DMU_OST_ANY && type != (*osp)->os_phys->os_type) { 67712380e1eSArne Jansen dsl_dataset_disown(ds, tag); 67812380e1eSArne Jansen return (SET_ERROR(EINVAL)); 67912380e1eSArne Jansen } else if (!readonly && dsl_dataset_is_snapshot(ds)) { 68012380e1eSArne Jansen dsl_dataset_disown(ds, tag); 68112380e1eSArne Jansen return (SET_ERROR(EROFS)); 68212380e1eSArne Jansen } 68312380e1eSArne Jansen return (err); 68412380e1eSArne Jansen } 68512380e1eSArne Jansen 6863b2aab18SMatthew Ahrens /* 6873b2aab18SMatthew Ahrens * dsl_pool must not be held when this is called. 6883b2aab18SMatthew Ahrens * Upon successful return, there will be a longhold on the dataset, 6893b2aab18SMatthew Ahrens * and the dsl_pool will not be held. 6903b2aab18SMatthew Ahrens */ 691fa9e4066Sahrens int 692503ad85cSMatthew Ahrens dmu_objset_own(const char *name, dmu_objset_type_t type, 693503ad85cSMatthew Ahrens boolean_t readonly, void *tag, objset_t **osp) 694fa9e4066Sahrens { 6953b2aab18SMatthew Ahrens dsl_pool_t *dp; 696f18faf3fSek dsl_dataset_t *ds; 697f18faf3fSek int err; 698fa9e4066Sahrens 6993b2aab18SMatthew Ahrens err = dsl_pool_hold(name, FTAG, &dp); 7003b2aab18SMatthew Ahrens if (err != 0) 7013b2aab18SMatthew Ahrens return (err); 7023b2aab18SMatthew Ahrens err = dsl_dataset_own(dp, name, tag, &ds); 7033b2aab18SMatthew Ahrens if (err != 0) { 7043b2aab18SMatthew Ahrens dsl_pool_rele(dp, FTAG); 705fa9e4066Sahrens return (err); 7063b2aab18SMatthew Ahrens } 70712380e1eSArne Jansen err = dmu_objset_own_impl(ds, type, readonly, tag, osp); 7083b2aab18SMatthew Ahrens dsl_pool_rele(dp, FTAG); 70912380e1eSArne Jansen 7103cb34c60Sahrens return (err); 711fa9e4066Sahrens } 712fa9e4066Sahrens 71312380e1eSArne Jansen int 71412380e1eSArne Jansen dmu_objset_own_obj(dsl_pool_t *dp, uint64_t obj, dmu_objset_type_t type, 71512380e1eSArne Jansen boolean_t readonly, void *tag, objset_t **osp) 71612380e1eSArne Jansen { 71712380e1eSArne Jansen dsl_dataset_t *ds; 71812380e1eSArne Jansen int err; 71912380e1eSArne Jansen 72012380e1eSArne Jansen err = dsl_dataset_own_obj(dp, obj, tag, &ds); 72112380e1eSArne Jansen if (err != 0) 72212380e1eSArne Jansen return (err); 72312380e1eSArne Jansen 72412380e1eSArne Jansen return (dmu_objset_own_impl(ds, type, readonly, tag, osp)); 72512380e1eSArne Jansen } 72612380e1eSArne Jansen 727fa9e4066Sahrens void 728503ad85cSMatthew Ahrens dmu_objset_rele(objset_t *os, void *tag) 729fa9e4066Sahrens { 7303b2aab18SMatthew Ahrens dsl_pool_t *dp = dmu_objset_pool(os); 731503ad85cSMatthew Ahrens dsl_dataset_rele(os->os_dsl_dataset, tag); 7323b2aab18SMatthew Ahrens dsl_pool_rele(dp, tag); 733503ad85cSMatthew Ahrens } 734503ad85cSMatthew Ahrens 73591948b51SKeith M Wesolowski /* 73691948b51SKeith M Wesolowski * When we are called, os MUST refer to an objset associated with a dataset 73791948b51SKeith M Wesolowski * that is owned by 'tag'; that is, is held and long held by 'tag' and ds_owner 73891948b51SKeith M Wesolowski * == tag. We will then release and reacquire ownership of the dataset while 73991948b51SKeith M Wesolowski * holding the pool config_rwlock to avoid intervening namespace or ownership 74091948b51SKeith M Wesolowski * changes may occur. 74191948b51SKeith M Wesolowski * 74291948b51SKeith M Wesolowski * This exists solely to accommodate zfs_ioc_userspace_upgrade()'s desire to 74391948b51SKeith M Wesolowski * release the hold on its dataset and acquire a new one on the dataset of the 74491948b51SKeith M Wesolowski * same name so that it can be partially torn down and reconstructed. 74591948b51SKeith M Wesolowski */ 74691948b51SKeith M Wesolowski void 7475f5913bbSAndriy Gapon dmu_objset_refresh_ownership(dsl_dataset_t *ds, dsl_dataset_t **newds, 7485f5913bbSAndriy Gapon void *tag) 74991948b51SKeith M Wesolowski { 75091948b51SKeith M Wesolowski dsl_pool_t *dp; 7519adfa60dSMatthew Ahrens char name[ZFS_MAX_DATASET_NAME_LEN]; 75291948b51SKeith M Wesolowski 75391948b51SKeith M Wesolowski VERIFY3P(ds, !=, NULL); 75491948b51SKeith M Wesolowski VERIFY3P(ds->ds_owner, ==, tag); 75591948b51SKeith M Wesolowski VERIFY(dsl_dataset_long_held(ds)); 75691948b51SKeith M Wesolowski 75791948b51SKeith M Wesolowski dsl_dataset_name(ds, name); 7585f5913bbSAndriy Gapon dp = ds->ds_dir->dd_pool; 75991948b51SKeith M Wesolowski dsl_pool_config_enter(dp, FTAG); 7605f5913bbSAndriy Gapon dsl_dataset_disown(ds, tag); 7615f5913bbSAndriy Gapon VERIFY0(dsl_dataset_own(dp, name, tag, newds)); 76291948b51SKeith M Wesolowski dsl_pool_config_exit(dp, FTAG); 76391948b51SKeith M Wesolowski } 76491948b51SKeith M Wesolowski 765503ad85cSMatthew Ahrens void 766503ad85cSMatthew Ahrens dmu_objset_disown(objset_t *os, void *tag) 767503ad85cSMatthew Ahrens { 768503ad85cSMatthew Ahrens dsl_dataset_disown(os->os_dsl_dataset, tag); 769fa9e4066Sahrens } 770fa9e4066Sahrens 7713b2aab18SMatthew Ahrens void 7721934e92fSmaybee dmu_objset_evict_dbufs(objset_t *os) 773ea8dc4b6Seschrock { 774bc9014e6SJustin Gibbs dnode_t dn_marker; 775ea8dc4b6Seschrock dnode_t *dn; 776c543ec06Sahrens 777503ad85cSMatthew Ahrens mutex_enter(&os->os_lock); 778bc9014e6SJustin Gibbs dn = list_head(&os->os_dnodes); 779bc9014e6SJustin Gibbs while (dn != NULL) { 780bc9014e6SJustin Gibbs /* 781bc9014e6SJustin Gibbs * Skip dnodes without holds. We have to do this dance 782bc9014e6SJustin Gibbs * because dnode_add_ref() only works if there is already a 783bc9014e6SJustin Gibbs * hold. If the dnode has no holds, then it has no dbufs. 784bc9014e6SJustin Gibbs */ 785bc9014e6SJustin Gibbs if (dnode_add_ref(dn, FTAG)) { 786bc9014e6SJustin Gibbs list_insert_after(&os->os_dnodes, dn, &dn_marker); 787bc9014e6SJustin Gibbs mutex_exit(&os->os_lock); 788c543ec06Sahrens 789bc9014e6SJustin Gibbs dnode_evict_dbufs(dn); 790bc9014e6SJustin Gibbs dnode_rele(dn, FTAG); 791c543ec06Sahrens 792bc9014e6SJustin Gibbs mutex_enter(&os->os_lock); 793bc9014e6SJustin Gibbs dn = list_next(&os->os_dnodes, &dn_marker); 794bc9014e6SJustin Gibbs list_remove(&os->os_dnodes, &dn_marker); 795bc9014e6SJustin Gibbs } else { 796bc9014e6SJustin Gibbs dn = list_next(&os->os_dnodes, dn); 797bc9014e6SJustin Gibbs } 798ea8dc4b6Seschrock } 799503ad85cSMatthew Ahrens mutex_exit(&os->os_lock); 800bc9014e6SJustin Gibbs 801bc9014e6SJustin Gibbs if (DMU_USERUSED_DNODE(os) != NULL) { 802bc9014e6SJustin Gibbs dnode_evict_dbufs(DMU_GROUPUSED_DNODE(os)); 803bc9014e6SJustin Gibbs dnode_evict_dbufs(DMU_USERUSED_DNODE(os)); 804bc9014e6SJustin Gibbs } 805bc9014e6SJustin Gibbs dnode_evict_dbufs(DMU_META_DNODE(os)); 806ea8dc4b6Seschrock } 807ea8dc4b6Seschrock 808bc9014e6SJustin Gibbs /* 809bc9014e6SJustin Gibbs * Objset eviction processing is split into into two pieces. 810bc9014e6SJustin Gibbs * The first marks the objset as evicting, evicts any dbufs that 811bc9014e6SJustin Gibbs * have a refcount of zero, and then queues up the objset for the 812bc9014e6SJustin Gibbs * second phase of eviction. Once os->os_dnodes has been cleared by 813bc9014e6SJustin Gibbs * dnode_buf_pageout()->dnode_destroy(), the second phase is executed. 814bc9014e6SJustin Gibbs * The second phase closes the special dnodes, dequeues the objset from 815bc9014e6SJustin Gibbs * the list of those undergoing eviction, and finally frees the objset. 816bc9014e6SJustin Gibbs * 817bc9014e6SJustin Gibbs * NOTE: Due to asynchronous eviction processing (invocation of 818bc9014e6SJustin Gibbs * dnode_buf_pageout()), it is possible for the meta dnode for the 819bc9014e6SJustin Gibbs * objset to have no holds even though os->os_dnodes is not empty. 820bc9014e6SJustin Gibbs */ 821fa9e4066Sahrens void 822503ad85cSMatthew Ahrens dmu_objset_evict(objset_t *os) 823fa9e4066Sahrens { 824503ad85cSMatthew Ahrens dsl_dataset_t *ds = os->os_dsl_dataset; 825fa9e4066Sahrens 826b24ab676SJeff Bonwick for (int t = 0; t < TXG_SIZE; t++) 827b24ab676SJeff Bonwick ASSERT(!dmu_objset_is_dirty(os, t)); 828fa9e4066Sahrens 82903bad06fSJustin Gibbs if (ds) 83003bad06fSJustin Gibbs dsl_prop_unregister_all(ds, os); 831fa9e4066Sahrens 8320a586ceaSMark Shellenbaum if (os->os_sa) 8330a586ceaSMark Shellenbaum sa_tear_down(os); 8340a586ceaSMark Shellenbaum 8353b2aab18SMatthew Ahrens dmu_objset_evict_dbufs(os); 836ea8dc4b6Seschrock 837bc9014e6SJustin Gibbs mutex_enter(&os->os_lock); 838bc9014e6SJustin Gibbs spa_evicting_os_register(os->os_spa, os); 839bc9014e6SJustin Gibbs if (list_is_empty(&os->os_dnodes)) { 840bc9014e6SJustin Gibbs mutex_exit(&os->os_lock); 841bc9014e6SJustin Gibbs dmu_objset_evict_done(os); 842bc9014e6SJustin Gibbs } else { 843bc9014e6SJustin Gibbs mutex_exit(&os->os_lock); 844bc9014e6SJustin Gibbs } 845bc9014e6SJustin Gibbs } 846bc9014e6SJustin Gibbs 847bc9014e6SJustin Gibbs void 848bc9014e6SJustin Gibbs dmu_objset_evict_done(objset_t *os) 849bc9014e6SJustin Gibbs { 850bc9014e6SJustin Gibbs ASSERT3P(list_head(&os->os_dnodes), ==, NULL); 851bc9014e6SJustin Gibbs 852744947dcSTom Erickson dnode_special_close(&os->os_meta_dnode); 853744947dcSTom Erickson if (DMU_USERUSED_DNODE(os)) { 854744947dcSTom Erickson dnode_special_close(&os->os_userused_dnode); 855744947dcSTom Erickson dnode_special_close(&os->os_groupused_dnode); 85614843421SMatthew Ahrens } 857503ad85cSMatthew Ahrens zil_free(os->os_zil); 858fa9e4066Sahrens 859dcbf3bd6SGeorge Wilson arc_buf_destroy(os->os_phys_buf, &os->os_phys_buf); 860744947dcSTom Erickson 861744947dcSTom Erickson /* 862744947dcSTom Erickson * This is a barrier to prevent the objset from going away in 863744947dcSTom Erickson * dnode_move() until we can safely ensure that the objset is still in 864744947dcSTom Erickson * use. We consider the objset valid before the barrier and invalid 865744947dcSTom Erickson * after the barrier. 866744947dcSTom Erickson */ 867744947dcSTom Erickson rw_enter(&os_lock, RW_READER); 868744947dcSTom Erickson rw_exit(&os_lock); 869744947dcSTom Erickson 87054811da5SToomas Soome kmem_free(os->os_obj_next_percpu, 87154811da5SToomas Soome os->os_obj_next_percpu_len * sizeof (os->os_obj_next_percpu[0])); 87254811da5SToomas Soome 873503ad85cSMatthew Ahrens mutex_destroy(&os->os_lock); 87494c2d0ebSMatthew Ahrens mutex_destroy(&os->os_userused_lock); 875503ad85cSMatthew Ahrens mutex_destroy(&os->os_obj_lock); 876503ad85cSMatthew Ahrens mutex_destroy(&os->os_user_ptr_lock); 87794c2d0ebSMatthew Ahrens for (int i = 0; i < TXG_SIZE; i++) { 87894c2d0ebSMatthew Ahrens multilist_destroy(os->os_dirty_dnodes[i]); 87994c2d0ebSMatthew Ahrens } 880bc9014e6SJustin Gibbs spa_evicting_os_deregister(os->os_spa, os); 881503ad85cSMatthew Ahrens kmem_free(os, sizeof (objset_t)); 882fa9e4066Sahrens } 883fa9e4066Sahrens 88471eb0538SChris Kirby timestruc_t 88571eb0538SChris Kirby dmu_objset_snap_cmtime(objset_t *os) 88671eb0538SChris Kirby { 88771eb0538SChris Kirby return (dsl_dir_snap_cmtime(os->os_dsl_dataset->ds_dir)); 88871eb0538SChris Kirby } 88971eb0538SChris Kirby 890fa9e4066Sahrens /* called from dsl for meta-objset */ 891503ad85cSMatthew Ahrens objset_t * 892c717a561Smaybee dmu_objset_create_impl(spa_t *spa, dsl_dataset_t *ds, blkptr_t *bp, 893c717a561Smaybee dmu_objset_type_t type, dmu_tx_t *tx) 894fa9e4066Sahrens { 895503ad85cSMatthew Ahrens objset_t *os; 896fa9e4066Sahrens dnode_t *mdn; 897fa9e4066Sahrens 898fa9e4066Sahrens ASSERT(dmu_tx_is_syncing(tx)); 8993b2aab18SMatthew Ahrens 900feaa74e4SMark Maybee if (ds != NULL) 9013b2aab18SMatthew Ahrens VERIFY0(dmu_objset_from_ds(ds, &os)); 902feaa74e4SMark Maybee else 9033b2aab18SMatthew Ahrens VERIFY0(dmu_objset_open_impl(spa, NULL, bp, &os)); 904feaa74e4SMark Maybee 905744947dcSTom Erickson mdn = DMU_META_DNODE(os); 906fa9e4066Sahrens 90754811da5SToomas Soome dnode_allocate(mdn, DMU_OT_DNODE, DNODE_BLOCK_SIZE, DN_MAX_INDBLKSHIFT, 90854811da5SToomas Soome DMU_OT_NONE, 0, DNODE_MIN_SLOTS, tx); 909fa9e4066Sahrens 910fa9e4066Sahrens /* 911fa9e4066Sahrens * We don't want to have to increase the meta-dnode's nlevels 912fa9e4066Sahrens * later, because then we could do it in quescing context while 913fa9e4066Sahrens * we are also accessing it in open context. 914fa9e4066Sahrens * 915fa9e4066Sahrens * This precaution is not necessary for the MOS (ds == NULL), 916fa9e4066Sahrens * because the MOS is only updated in syncing context. 917fa9e4066Sahrens * This is most fortunate: the MOS is the only objset that 918fa9e4066Sahrens * needs to be synced multiple times as spa_sync() iterates 919fa9e4066Sahrens * to convergence, so minimizing its dn_nlevels matters. 920fa9e4066Sahrens */ 921ea8dc4b6Seschrock if (ds != NULL) { 922ea8dc4b6Seschrock int levels = 1; 923ea8dc4b6Seschrock 924ea8dc4b6Seschrock /* 925ea8dc4b6Seschrock * Determine the number of levels necessary for the meta-dnode 9264b5c8e93SMatthew Ahrens * to contain DN_MAX_OBJECT dnodes. Note that in order to 9274b5c8e93SMatthew Ahrens * ensure that we do not overflow 64 bits, there has to be 9284b5c8e93SMatthew Ahrens * a nlevels that gives us a number of blocks > DN_MAX_OBJECT 9294b5c8e93SMatthew Ahrens * but < 2^64. Therefore, 9304b5c8e93SMatthew Ahrens * (mdn->dn_indblkshift - SPA_BLKPTRSHIFT) (10) must be 9314b5c8e93SMatthew Ahrens * less than (64 - log2(DN_MAX_OBJECT)) (16). 932ea8dc4b6Seschrock */ 9334b5c8e93SMatthew Ahrens while ((uint64_t)mdn->dn_nblkptr << 9344b5c8e93SMatthew Ahrens (mdn->dn_datablkshift - DNODE_SHIFT + 935ea8dc4b6Seschrock (levels - 1) * (mdn->dn_indblkshift - SPA_BLKPTRSHIFT)) < 9364b5c8e93SMatthew Ahrens DN_MAX_OBJECT) 937ea8dc4b6Seschrock levels++; 938ea8dc4b6Seschrock 939fa9e4066Sahrens mdn->dn_next_nlevels[tx->tx_txg & TXG_MASK] = 940ea8dc4b6Seschrock mdn->dn_nlevels = levels; 941ea8dc4b6Seschrock } 942fa9e4066Sahrens 943fa9e4066Sahrens ASSERT(type != DMU_OST_NONE); 944fa9e4066Sahrens ASSERT(type != DMU_OST_ANY); 945fa9e4066Sahrens ASSERT(type < DMU_OST_NUMTYPES); 946503ad85cSMatthew Ahrens os->os_phys->os_type = type; 947503ad85cSMatthew Ahrens if (dmu_objset_userused_enabled(os)) { 948503ad85cSMatthew Ahrens os->os_phys->os_flags |= OBJSET_FLAG_USERACCOUNTING_COMPLETE; 949503ad85cSMatthew Ahrens os->os_flags = os->os_phys->os_flags; 95014843421SMatthew Ahrens } 951fa9e4066Sahrens 952fa9e4066Sahrens dsl_dataset_dirty(ds, tx); 953fa9e4066Sahrens 954503ad85cSMatthew Ahrens return (os); 955fa9e4066Sahrens } 956fa9e4066Sahrens 9573b2aab18SMatthew Ahrens typedef struct dmu_objset_create_arg { 9583b2aab18SMatthew Ahrens const char *doca_name; 9593b2aab18SMatthew Ahrens cred_t *doca_cred; 9603b2aab18SMatthew Ahrens void (*doca_userfunc)(objset_t *os, void *arg, 9613b2aab18SMatthew Ahrens cred_t *cr, dmu_tx_t *tx); 9623b2aab18SMatthew Ahrens void *doca_userarg; 9633b2aab18SMatthew Ahrens dmu_objset_type_t doca_type; 9643b2aab18SMatthew Ahrens uint64_t doca_flags; 9653b2aab18SMatthew Ahrens } dmu_objset_create_arg_t; 966fa9e4066Sahrens 967ecd6cf80Smarks /*ARGSUSED*/ 968fa9e4066Sahrens static int 9693b2aab18SMatthew Ahrens dmu_objset_create_check(void *arg, dmu_tx_t *tx) 970fa9e4066Sahrens { 9713b2aab18SMatthew Ahrens dmu_objset_create_arg_t *doca = arg; 9723b2aab18SMatthew Ahrens dsl_pool_t *dp = dmu_tx_pool(tx); 9733b2aab18SMatthew Ahrens dsl_dir_t *pdd; 9743b2aab18SMatthew Ahrens const char *tail; 9753b2aab18SMatthew Ahrens int error; 9761d452cf5Sahrens 9773b2aab18SMatthew Ahrens if (strchr(doca->doca_name, '@') != NULL) 978be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 9791d452cf5Sahrens 9809adfa60dSMatthew Ahrens if (strlen(doca->doca_name) >= ZFS_MAX_DATASET_NAME_LEN) 9819adfa60dSMatthew Ahrens return (SET_ERROR(ENAMETOOLONG)); 9829adfa60dSMatthew Ahrens 9835ac95da7SSerapheim Dimitropoulos if (dataset_nestcheck(doca->doca_name) != 0) 9845ac95da7SSerapheim Dimitropoulos return (SET_ERROR(ENAMETOOLONG)); 9855ac95da7SSerapheim Dimitropoulos 9863b2aab18SMatthew Ahrens error = dsl_dir_hold(dp, doca->doca_name, FTAG, &pdd, &tail); 9873b2aab18SMatthew Ahrens if (error != 0) 9883b2aab18SMatthew Ahrens return (error); 9893b2aab18SMatthew Ahrens if (tail == NULL) { 9903b2aab18SMatthew Ahrens dsl_dir_rele(pdd, FTAG); 991be6fd75aSMatthew Ahrens return (SET_ERROR(EEXIST)); 9921d452cf5Sahrens } 993a2afb611SJerry Jelinek error = dsl_fs_ss_limit_check(pdd, 1, ZFS_PROP_FILESYSTEM_LIMIT, NULL, 994a2afb611SJerry Jelinek doca->doca_cred); 9953b2aab18SMatthew Ahrens dsl_dir_rele(pdd, FTAG); 996ecd6cf80Smarks 997a2afb611SJerry Jelinek return (error); 9981d452cf5Sahrens } 9991d452cf5Sahrens 10001d452cf5Sahrens static void 10013b2aab18SMatthew Ahrens dmu_objset_create_sync(void *arg, dmu_tx_t *tx) 10021d452cf5Sahrens { 10033b2aab18SMatthew Ahrens dmu_objset_create_arg_t *doca = arg; 10043b2aab18SMatthew Ahrens dsl_pool_t *dp = dmu_tx_pool(tx); 10053b2aab18SMatthew Ahrens dsl_dir_t *pdd; 10063b2aab18SMatthew Ahrens const char *tail; 10074445fffbSMatthew Ahrens dsl_dataset_t *ds; 10083b2aab18SMatthew Ahrens uint64_t obj; 10094445fffbSMatthew Ahrens blkptr_t *bp; 10103b2aab18SMatthew Ahrens objset_t *os; 1011fa9e4066Sahrens 10123b2aab18SMatthew Ahrens VERIFY0(dsl_dir_hold(dp, doca->doca_name, FTAG, &pdd, &tail)); 1013fa9e4066Sahrens 10143b2aab18SMatthew Ahrens obj = dsl_dataset_create_sync(pdd, tail, NULL, doca->doca_flags, 10153b2aab18SMatthew Ahrens doca->doca_cred, tx); 1016fa9e4066Sahrens 10173b2aab18SMatthew Ahrens VERIFY0(dsl_dataset_hold_obj(pdd->dd_pool, obj, FTAG, &ds)); 1018c166b69dSPaul Dagnelie rrw_enter(&ds->ds_bp_rwlock, RW_READER, FTAG); 10194445fffbSMatthew Ahrens bp = dsl_dataset_get_blkptr(ds); 10203b2aab18SMatthew Ahrens os = dmu_objset_create_impl(pdd->dd_pool->dp_spa, 10213b2aab18SMatthew Ahrens ds, bp, doca->doca_type, tx); 1022c166b69dSPaul Dagnelie rrw_exit(&ds->ds_bp_rwlock, FTAG); 1023fa9e4066Sahrens 10243b2aab18SMatthew Ahrens if (doca->doca_userfunc != NULL) { 10253b2aab18SMatthew Ahrens doca->doca_userfunc(os, doca->doca_userarg, 10263b2aab18SMatthew Ahrens doca->doca_cred, tx); 1027fa9e4066Sahrens } 1028ecd6cf80Smarks 10293b2aab18SMatthew Ahrens spa_history_log_internal_ds(ds, "create", tx, ""); 10304445fffbSMatthew Ahrens dsl_dataset_rele(ds, FTAG); 10313b2aab18SMatthew Ahrens dsl_dir_rele(pdd, FTAG); 1032fa9e4066Sahrens } 1033fa9e4066Sahrens 1034fa9e4066Sahrens int 1035ae46e4c7SMatthew Ahrens dmu_objset_create(const char *name, dmu_objset_type_t type, uint64_t flags, 1036ecd6cf80Smarks void (*func)(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx), void *arg) 1037fa9e4066Sahrens { 10383b2aab18SMatthew Ahrens dmu_objset_create_arg_t doca; 1039fa9e4066Sahrens 10403b2aab18SMatthew Ahrens doca.doca_name = name; 10413b2aab18SMatthew Ahrens doca.doca_cred = CRED(); 10423b2aab18SMatthew Ahrens doca.doca_flags = flags; 10433b2aab18SMatthew Ahrens doca.doca_userfunc = func; 10443b2aab18SMatthew Ahrens doca.doca_userarg = arg; 10453b2aab18SMatthew Ahrens doca.doca_type = type; 1046ecd6cf80Smarks 10473b2aab18SMatthew Ahrens return (dsl_sync_task(name, 10487d46dc6cSMatthew Ahrens dmu_objset_create_check, dmu_objset_create_sync, &doca, 10497d46dc6cSMatthew Ahrens 5, ZFS_SPACE_CHECK_NORMAL)); 1050ae46e4c7SMatthew Ahrens } 1051ae46e4c7SMatthew Ahrens 10523b2aab18SMatthew Ahrens typedef struct dmu_objset_clone_arg { 10533b2aab18SMatthew Ahrens const char *doca_clone; 10543b2aab18SMatthew Ahrens const char *doca_origin; 10553b2aab18SMatthew Ahrens cred_t *doca_cred; 10563b2aab18SMatthew Ahrens } dmu_objset_clone_arg_t; 10573b2aab18SMatthew Ahrens 10583b2aab18SMatthew Ahrens /*ARGSUSED*/ 10593b2aab18SMatthew Ahrens static int 10603b2aab18SMatthew Ahrens dmu_objset_clone_check(void *arg, dmu_tx_t *tx) 1061ae46e4c7SMatthew Ahrens { 10623b2aab18SMatthew Ahrens dmu_objset_clone_arg_t *doca = arg; 1063ae46e4c7SMatthew Ahrens dsl_dir_t *pdd; 1064ae46e4c7SMatthew Ahrens const char *tail; 10653b2aab18SMatthew Ahrens int error; 10663b2aab18SMatthew Ahrens dsl_dataset_t *origin; 10673b2aab18SMatthew Ahrens dsl_pool_t *dp = dmu_tx_pool(tx); 1068ae46e4c7SMatthew Ahrens 10693b2aab18SMatthew Ahrens if (strchr(doca->doca_clone, '@') != NULL) 1070be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 10713b2aab18SMatthew Ahrens 10729adfa60dSMatthew Ahrens if (strlen(doca->doca_clone) >= ZFS_MAX_DATASET_NAME_LEN) 10739adfa60dSMatthew Ahrens return (SET_ERROR(ENAMETOOLONG)); 10749adfa60dSMatthew Ahrens 10753b2aab18SMatthew Ahrens error = dsl_dir_hold(dp, doca->doca_clone, FTAG, &pdd, &tail); 10763b2aab18SMatthew Ahrens if (error != 0) 10773b2aab18SMatthew Ahrens return (error); 1078ae46e4c7SMatthew Ahrens if (tail == NULL) { 10793b2aab18SMatthew Ahrens dsl_dir_rele(pdd, FTAG); 1080be6fd75aSMatthew Ahrens return (SET_ERROR(EEXIST)); 1081fa9e4066Sahrens } 108203b1c297SAlexander Eremin 1083a2afb611SJerry Jelinek error = dsl_fs_ss_limit_check(pdd, 1, ZFS_PROP_FILESYSTEM_LIMIT, NULL, 1084a2afb611SJerry Jelinek doca->doca_cred); 1085a2afb611SJerry Jelinek if (error != 0) { 1086a2afb611SJerry Jelinek dsl_dir_rele(pdd, FTAG); 1087a2afb611SJerry Jelinek return (SET_ERROR(EDQUOT)); 1088a2afb611SJerry Jelinek } 10893b2aab18SMatthew Ahrens dsl_dir_rele(pdd, FTAG); 1090fa9e4066Sahrens 10913b2aab18SMatthew Ahrens error = dsl_dataset_hold(dp, doca->doca_origin, FTAG, &origin); 10923b2aab18SMatthew Ahrens if (error != 0) 109399d5e173STim Haley return (error); 109499d5e173STim Haley 10953b2aab18SMatthew Ahrens /* You can only clone snapshots, not the head datasets. */ 1096bc9014e6SJustin Gibbs if (!origin->ds_is_snapshot) { 10973b2aab18SMatthew Ahrens dsl_dataset_rele(origin, FTAG); 1098be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 109992241e0bSTom Erickson } 11003b2aab18SMatthew Ahrens dsl_dataset_rele(origin, FTAG); 110199d5e173STim Haley 11023b2aab18SMatthew Ahrens return (0); 1103ea2f5b9eSMatthew Ahrens } 11041d452cf5Sahrens 11053b2aab18SMatthew Ahrens static void 11063b2aab18SMatthew Ahrens dmu_objset_clone_sync(void *arg, dmu_tx_t *tx) 11071d452cf5Sahrens { 11083b2aab18SMatthew Ahrens dmu_objset_clone_arg_t *doca = arg; 11093b2aab18SMatthew Ahrens dsl_pool_t *dp = dmu_tx_pool(tx); 11103b2aab18SMatthew Ahrens dsl_dir_t *pdd; 11113b2aab18SMatthew Ahrens const char *tail; 11123b2aab18SMatthew Ahrens dsl_dataset_t *origin, *ds; 11133b2aab18SMatthew Ahrens uint64_t obj; 11149adfa60dSMatthew Ahrens char namebuf[ZFS_MAX_DATASET_NAME_LEN]; 11154445fffbSMatthew Ahrens 11163b2aab18SMatthew Ahrens VERIFY0(dsl_dir_hold(dp, doca->doca_clone, FTAG, &pdd, &tail)); 11173b2aab18SMatthew Ahrens VERIFY0(dsl_dataset_hold(dp, doca->doca_origin, FTAG, &origin)); 11184445fffbSMatthew Ahrens 11193b2aab18SMatthew Ahrens obj = dsl_dataset_create_sync(pdd, tail, origin, 0, 11203b2aab18SMatthew Ahrens doca->doca_cred, tx); 1121f2e10be3Srm 11223b2aab18SMatthew Ahrens VERIFY0(dsl_dataset_hold_obj(pdd->dd_pool, obj, FTAG, &ds)); 11233b2aab18SMatthew Ahrens dsl_dataset_name(origin, namebuf); 11243b2aab18SMatthew Ahrens spa_history_log_internal_ds(ds, "clone", tx, 11253b2aab18SMatthew Ahrens "origin=%s (%llu)", namebuf, origin->ds_object); 11263b2aab18SMatthew Ahrens dsl_dataset_rele(ds, FTAG); 11273b2aab18SMatthew Ahrens dsl_dataset_rele(origin, FTAG); 11283b2aab18SMatthew Ahrens dsl_dir_rele(pdd, FTAG); 11291d452cf5Sahrens } 11301d452cf5Sahrens 11311d452cf5Sahrens int 11323b2aab18SMatthew Ahrens dmu_objset_clone(const char *clone, const char *origin) 11331d452cf5Sahrens { 11343b2aab18SMatthew Ahrens dmu_objset_clone_arg_t doca; 11354445fffbSMatthew Ahrens 11363b2aab18SMatthew Ahrens doca.doca_clone = clone; 11373b2aab18SMatthew Ahrens doca.doca_origin = origin; 11383b2aab18SMatthew Ahrens doca.doca_cred = CRED(); 113999d5e173STim Haley 11403b2aab18SMatthew Ahrens return (dsl_sync_task(clone, 11417d46dc6cSMatthew Ahrens dmu_objset_clone_check, dmu_objset_clone_sync, &doca, 11427d46dc6cSMatthew Ahrens 5, ZFS_SPACE_CHECK_NORMAL)); 11434445fffbSMatthew Ahrens } 11444445fffbSMatthew Ahrens 11455cabbc6bSPrashanth Sreenivasa static int 11465cabbc6bSPrashanth Sreenivasa dmu_objset_remap_indirects_impl(objset_t *os, uint64_t last_removed_txg) 11475cabbc6bSPrashanth Sreenivasa { 11485cabbc6bSPrashanth Sreenivasa int error = 0; 11495cabbc6bSPrashanth Sreenivasa uint64_t object = 0; 11505cabbc6bSPrashanth Sreenivasa while ((error = dmu_object_next(os, &object, B_FALSE, 0)) == 0) { 11515cabbc6bSPrashanth Sreenivasa error = dmu_object_remap_indirects(os, object, 11525cabbc6bSPrashanth Sreenivasa last_removed_txg); 11535cabbc6bSPrashanth Sreenivasa /* 11545cabbc6bSPrashanth Sreenivasa * If the ZPL removed the object before we managed to dnode_hold 11555cabbc6bSPrashanth Sreenivasa * it, we would get an ENOENT. If the ZPL declares its intent 11565cabbc6bSPrashanth Sreenivasa * to remove the object (dnode_free) before we manage to 11575cabbc6bSPrashanth Sreenivasa * dnode_hold it, we would get an EEXIST. In either case, we 11585cabbc6bSPrashanth Sreenivasa * want to continue remapping the other objects in the objset; 11595cabbc6bSPrashanth Sreenivasa * in all other cases, we want to break early. 11605cabbc6bSPrashanth Sreenivasa */ 11615cabbc6bSPrashanth Sreenivasa if (error != 0 && error != ENOENT && error != EEXIST) { 11625cabbc6bSPrashanth Sreenivasa break; 11635cabbc6bSPrashanth Sreenivasa } 11645cabbc6bSPrashanth Sreenivasa } 11655cabbc6bSPrashanth Sreenivasa if (error == ESRCH) { 11665cabbc6bSPrashanth Sreenivasa error = 0; 11675cabbc6bSPrashanth Sreenivasa } 11685cabbc6bSPrashanth Sreenivasa return (error); 11695cabbc6bSPrashanth Sreenivasa } 11705cabbc6bSPrashanth Sreenivasa 11715cabbc6bSPrashanth Sreenivasa int 11725cabbc6bSPrashanth Sreenivasa dmu_objset_remap_indirects(const char *fsname) 11735cabbc6bSPrashanth Sreenivasa { 11745cabbc6bSPrashanth Sreenivasa int error = 0; 11755cabbc6bSPrashanth Sreenivasa objset_t *os = NULL; 11765cabbc6bSPrashanth Sreenivasa uint64_t last_removed_txg; 11775cabbc6bSPrashanth Sreenivasa uint64_t remap_start_txg; 11785cabbc6bSPrashanth Sreenivasa dsl_dir_t *dd; 11795cabbc6bSPrashanth Sreenivasa 11805cabbc6bSPrashanth Sreenivasa error = dmu_objset_hold(fsname, FTAG, &os); 11815cabbc6bSPrashanth Sreenivasa if (error != 0) { 11825cabbc6bSPrashanth Sreenivasa return (error); 11835cabbc6bSPrashanth Sreenivasa } 11845cabbc6bSPrashanth Sreenivasa dd = dmu_objset_ds(os)->ds_dir; 11855cabbc6bSPrashanth Sreenivasa 11865cabbc6bSPrashanth Sreenivasa if (!spa_feature_is_enabled(dmu_objset_spa(os), 11875cabbc6bSPrashanth Sreenivasa SPA_FEATURE_OBSOLETE_COUNTS)) { 11885cabbc6bSPrashanth Sreenivasa dmu_objset_rele(os, FTAG); 11895cabbc6bSPrashanth Sreenivasa return (SET_ERROR(ENOTSUP)); 11905cabbc6bSPrashanth Sreenivasa } 11915cabbc6bSPrashanth Sreenivasa 11925cabbc6bSPrashanth Sreenivasa if (dsl_dataset_is_snapshot(dmu_objset_ds(os))) { 11935cabbc6bSPrashanth Sreenivasa dmu_objset_rele(os, FTAG); 11945cabbc6bSPrashanth Sreenivasa return (SET_ERROR(EINVAL)); 11955cabbc6bSPrashanth Sreenivasa } 11965cabbc6bSPrashanth Sreenivasa 11975cabbc6bSPrashanth Sreenivasa /* 11985cabbc6bSPrashanth Sreenivasa * If there has not been a removal, we're done. 11995cabbc6bSPrashanth Sreenivasa */ 12005cabbc6bSPrashanth Sreenivasa last_removed_txg = spa_get_last_removal_txg(dmu_objset_spa(os)); 12015cabbc6bSPrashanth Sreenivasa if (last_removed_txg == -1ULL) { 12025cabbc6bSPrashanth Sreenivasa dmu_objset_rele(os, FTAG); 12035cabbc6bSPrashanth Sreenivasa return (0); 12045cabbc6bSPrashanth Sreenivasa } 12055cabbc6bSPrashanth Sreenivasa 12065cabbc6bSPrashanth Sreenivasa /* 12075cabbc6bSPrashanth Sreenivasa * If we have remapped since the last removal, we're done. 12085cabbc6bSPrashanth Sreenivasa */ 12095cabbc6bSPrashanth Sreenivasa if (dsl_dir_is_zapified(dd)) { 12105cabbc6bSPrashanth Sreenivasa uint64_t last_remap_txg; 12115cabbc6bSPrashanth Sreenivasa if (zap_lookup(spa_meta_objset(dmu_objset_spa(os)), 12125cabbc6bSPrashanth Sreenivasa dd->dd_object, DD_FIELD_LAST_REMAP_TXG, 12135cabbc6bSPrashanth Sreenivasa sizeof (last_remap_txg), 1, &last_remap_txg) == 0 && 12145cabbc6bSPrashanth Sreenivasa last_remap_txg > last_removed_txg) { 12155cabbc6bSPrashanth Sreenivasa dmu_objset_rele(os, FTAG); 12165cabbc6bSPrashanth Sreenivasa return (0); 12175cabbc6bSPrashanth Sreenivasa } 12185cabbc6bSPrashanth Sreenivasa } 12195cabbc6bSPrashanth Sreenivasa 12205cabbc6bSPrashanth Sreenivasa dsl_dataset_long_hold(dmu_objset_ds(os), FTAG); 12215cabbc6bSPrashanth Sreenivasa dsl_pool_rele(dmu_objset_pool(os), FTAG); 12225cabbc6bSPrashanth Sreenivasa 12235cabbc6bSPrashanth Sreenivasa remap_start_txg = spa_last_synced_txg(dmu_objset_spa(os)); 12245cabbc6bSPrashanth Sreenivasa error = dmu_objset_remap_indirects_impl(os, last_removed_txg); 12255cabbc6bSPrashanth Sreenivasa if (error == 0) { 12265cabbc6bSPrashanth Sreenivasa /* 12275cabbc6bSPrashanth Sreenivasa * We update the last_remap_txg to be the start txg so that 12285cabbc6bSPrashanth Sreenivasa * we can guarantee that every block older than last_remap_txg 12295cabbc6bSPrashanth Sreenivasa * that can be remapped has been remapped. 12305cabbc6bSPrashanth Sreenivasa */ 12315cabbc6bSPrashanth Sreenivasa error = dsl_dir_update_last_remap_txg(dd, remap_start_txg); 12325cabbc6bSPrashanth Sreenivasa } 12335cabbc6bSPrashanth Sreenivasa 12345cabbc6bSPrashanth Sreenivasa dsl_dataset_long_rele(dmu_objset_ds(os), FTAG); 12355cabbc6bSPrashanth Sreenivasa dsl_dataset_rele(dmu_objset_ds(os), FTAG); 12365cabbc6bSPrashanth Sreenivasa 12375cabbc6bSPrashanth Sreenivasa return (error); 12385cabbc6bSPrashanth Sreenivasa } 12395cabbc6bSPrashanth Sreenivasa 12404445fffbSMatthew Ahrens int 12414445fffbSMatthew Ahrens dmu_objset_snapshot_one(const char *fsname, const char *snapname) 12424445fffbSMatthew Ahrens { 12434445fffbSMatthew Ahrens int err; 12444445fffbSMatthew Ahrens char *longsnap = kmem_asprintf("%s@%s", fsname, snapname); 12454445fffbSMatthew Ahrens nvlist_t *snaps = fnvlist_alloc(); 12464445fffbSMatthew Ahrens 12474445fffbSMatthew Ahrens fnvlist_add_boolean(snaps, longsnap); 12484445fffbSMatthew Ahrens strfree(longsnap); 12493b2aab18SMatthew Ahrens err = dsl_dataset_snapshot(snaps, NULL, NULL); 12503b2aab18SMatthew Ahrens fnvlist_free(snaps); 12514445fffbSMatthew Ahrens return (err); 12524445fffbSMatthew Ahrens } 12534445fffbSMatthew Ahrens 1254fa9e4066Sahrens static void 125594c2d0ebSMatthew Ahrens dmu_objset_sync_dnodes(multilist_sublist_t *list, dmu_tx_t *tx) 1256fa9e4066Sahrens { 1257c717a561Smaybee dnode_t *dn; 1258faafa6e3Sahrens 125994c2d0ebSMatthew Ahrens while ((dn = multilist_sublist_head(list)) != NULL) { 1260c717a561Smaybee ASSERT(dn->dn_object != DMU_META_DNODE_OBJECT); 1261c717a561Smaybee ASSERT(dn->dn_dbuf->db_data_pending); 1262c717a561Smaybee /* 126314843421SMatthew Ahrens * Initialize dn_zio outside dnode_sync() because the 126414843421SMatthew Ahrens * meta-dnode needs to set it ouside dnode_sync(). 1265c717a561Smaybee */ 1266c717a561Smaybee dn->dn_zio = dn->dn_dbuf->db_data_pending->dr_zio; 1267c717a561Smaybee ASSERT(dn->dn_zio); 1268faafa6e3Sahrens 1269c717a561Smaybee ASSERT3U(dn->dn_nlevels, <=, DN_MAX_LEVELS); 127094c2d0ebSMatthew Ahrens multilist_sublist_remove(list, dn); 127114843421SMatthew Ahrens 1272aa02ea01STom Caputi /* 1273aa02ea01STom Caputi * If we are not doing useraccounting (os_synced_dnodes == NULL) 1274aa02ea01STom Caputi * we are done with this dnode for this txg. Unset dn_dirty_txg 1275aa02ea01STom Caputi * if later txgs aren't dirtying it so that future holders do 1276aa02ea01STom Caputi * not get a stale value. Otherwise, we will do this in 1277aa02ea01STom Caputi * userquota_updates_task() when processing has completely 1278aa02ea01STom Caputi * finished for this txg. 1279aa02ea01STom Caputi */ 128094c2d0ebSMatthew Ahrens multilist_t *newlist = dn->dn_objset->os_synced_dnodes; 128194c2d0ebSMatthew Ahrens if (newlist != NULL) { 128214843421SMatthew Ahrens (void) dnode_add_ref(dn, newlist); 128394c2d0ebSMatthew Ahrens multilist_insert(newlist, dn); 1284aa02ea01STom Caputi } else { 1285aa02ea01STom Caputi mutex_enter(&dn->dn_mtx); 1286aa02ea01STom Caputi if (dn->dn_dirty_txg == tx->tx_txg) 1287aa02ea01STom Caputi dn->dn_dirty_txg = 0; 1288aa02ea01STom Caputi mutex_exit(&dn->dn_mtx); 128914843421SMatthew Ahrens } 129014843421SMatthew Ahrens 1291c717a561Smaybee dnode_sync(dn, tx); 1292fa9e4066Sahrens } 1293fa9e4066Sahrens } 1294fa9e4066Sahrens 1295fa9e4066Sahrens /* ARGSUSED */ 1296fa9e4066Sahrens static void 1297b24ab676SJeff Bonwick dmu_objset_write_ready(zio_t *zio, arc_buf_t *abuf, void *arg) 1298fa9e4066Sahrens { 1299e14bb325SJeff Bonwick blkptr_t *bp = zio->io_bp; 1300503ad85cSMatthew Ahrens objset_t *os = arg; 1301c717a561Smaybee dnode_phys_t *dnp = &os->os_phys->os_meta_dnode; 1302fa9e4066Sahrens 13035d7b4d43SMatthew Ahrens ASSERT(!BP_IS_EMBEDDED(bp)); 13043b2aab18SMatthew Ahrens ASSERT3U(BP_GET_TYPE(bp), ==, DMU_OT_OBJSET); 13053b2aab18SMatthew Ahrens ASSERT0(BP_GET_LEVEL(bp)); 13060a4e9518Sgw 1307fa9e4066Sahrens /* 130814843421SMatthew Ahrens * Update rootbp fill count: it should be the number of objects 130914843421SMatthew Ahrens * allocated in the object set (not counting the "special" 131014843421SMatthew Ahrens * objects that are stored in the objset_phys_t -- the meta 131114843421SMatthew Ahrens * dnode and user/group accounting objects). 1312fa9e4066Sahrens */ 131314843421SMatthew Ahrens bp->blk_fill = 0; 1314e14bb325SJeff Bonwick for (int i = 0; i < dnp->dn_nblkptr; i++) 13155d7b4d43SMatthew Ahrens bp->blk_fill += BP_GET_FILL(&dnp->dn_blkptr[i]); 1316c166b69dSPaul Dagnelie if (os->os_dsl_dataset != NULL) 1317c166b69dSPaul Dagnelie rrw_enter(&os->os_dsl_dataset->ds_bp_rwlock, RW_WRITER, FTAG); 1318c166b69dSPaul Dagnelie *os->os_rootbp = *bp; 1319c166b69dSPaul Dagnelie if (os->os_dsl_dataset != NULL) 1320c166b69dSPaul Dagnelie rrw_exit(&os->os_dsl_dataset->ds_bp_rwlock, FTAG); 1321b24ab676SJeff Bonwick } 1322b24ab676SJeff Bonwick 1323b24ab676SJeff Bonwick /* ARGSUSED */ 1324b24ab676SJeff Bonwick static void 1325b24ab676SJeff Bonwick dmu_objset_write_done(zio_t *zio, arc_buf_t *abuf, void *arg) 1326b24ab676SJeff Bonwick { 1327b24ab676SJeff Bonwick blkptr_t *bp = zio->io_bp; 1328b24ab676SJeff Bonwick blkptr_t *bp_orig = &zio->io_bp_orig; 1329b24ab676SJeff Bonwick objset_t *os = arg; 13300a4e9518Sgw 1331e14bb325SJeff Bonwick if (zio->io_flags & ZIO_FLAG_IO_REWRITE) { 1332b24ab676SJeff Bonwick ASSERT(BP_EQUAL(bp, bp_orig)); 1333e14bb325SJeff Bonwick } else { 1334b24ab676SJeff Bonwick dsl_dataset_t *ds = os->os_dsl_dataset; 1335b24ab676SJeff Bonwick dmu_tx_t *tx = os->os_synctx; 1336b24ab676SJeff Bonwick 1337b24ab676SJeff Bonwick (void) dsl_dataset_block_kill(ds, bp_orig, tx, B_TRUE); 1338b24ab676SJeff Bonwick dsl_dataset_block_born(ds, bp, tx); 13390a4e9518Sgw } 1340c166b69dSPaul Dagnelie kmem_free(bp, sizeof (*bp)); 1341c717a561Smaybee } 1342c717a561Smaybee 134394c2d0ebSMatthew Ahrens typedef struct sync_dnodes_arg { 134494c2d0ebSMatthew Ahrens multilist_t *sda_list; 134594c2d0ebSMatthew Ahrens int sda_sublist_idx; 134694c2d0ebSMatthew Ahrens multilist_t *sda_newlist; 134794c2d0ebSMatthew Ahrens dmu_tx_t *sda_tx; 134894c2d0ebSMatthew Ahrens } sync_dnodes_arg_t; 134994c2d0ebSMatthew Ahrens 135094c2d0ebSMatthew Ahrens static void 135194c2d0ebSMatthew Ahrens sync_dnodes_task(void *arg) 135294c2d0ebSMatthew Ahrens { 135394c2d0ebSMatthew Ahrens sync_dnodes_arg_t *sda = arg; 135494c2d0ebSMatthew Ahrens 135594c2d0ebSMatthew Ahrens multilist_sublist_t *ms = 135694c2d0ebSMatthew Ahrens multilist_sublist_lock(sda->sda_list, sda->sda_sublist_idx); 135794c2d0ebSMatthew Ahrens 135894c2d0ebSMatthew Ahrens dmu_objset_sync_dnodes(ms, sda->sda_tx); 135994c2d0ebSMatthew Ahrens 136094c2d0ebSMatthew Ahrens multilist_sublist_unlock(ms); 136194c2d0ebSMatthew Ahrens 136294c2d0ebSMatthew Ahrens kmem_free(sda, sizeof (*sda)); 136394c2d0ebSMatthew Ahrens } 136494c2d0ebSMatthew Ahrens 136594c2d0ebSMatthew Ahrens 1366fa9e4066Sahrens /* called from dsl */ 1367fa9e4066Sahrens void 1368503ad85cSMatthew Ahrens dmu_objset_sync(objset_t *os, zio_t *pio, dmu_tx_t *tx) 1369fa9e4066Sahrens { 1370fa9e4066Sahrens int txgoff; 13717802d7bfSMatthew Ahrens zbookmark_phys_t zb; 1372b24ab676SJeff Bonwick zio_prop_t zp; 1373c717a561Smaybee zio_t *zio; 1374c717a561Smaybee list_t *list; 1375c717a561Smaybee dbuf_dirty_record_t *dr; 1376c166b69dSPaul Dagnelie blkptr_t *blkptr_copy = kmem_alloc(sizeof (*os->os_rootbp), KM_SLEEP); 1377c166b69dSPaul Dagnelie *blkptr_copy = *os->os_rootbp; 1378c717a561Smaybee 1379c717a561Smaybee dprintf_ds(os->os_dsl_dataset, "txg=%llu\n", tx->tx_txg); 1380fa9e4066Sahrens 1381fa9e4066Sahrens ASSERT(dmu_tx_is_syncing(tx)); 1382fa9e4066Sahrens /* XXX the write_done callback should really give us the tx... */ 1383fa9e4066Sahrens os->os_synctx = tx; 1384fa9e4066Sahrens 138587bd5c1eSahrens if (os->os_dsl_dataset == NULL) { 138687bd5c1eSahrens /* 138787bd5c1eSahrens * This is the MOS. If we have upgraded, 138887bd5c1eSahrens * spa_max_replication() could change, so reset 138987bd5c1eSahrens * os_copies here. 139087bd5c1eSahrens */ 139187bd5c1eSahrens os->os_copies = spa_max_replication(os->os_spa); 139287bd5c1eSahrens } 139387bd5c1eSahrens 1394fa9e4066Sahrens /* 1395c717a561Smaybee * Create the root block IO 1396fa9e4066Sahrens */ 1397b24ab676SJeff Bonwick SET_BOOKMARK(&zb, os->os_dsl_dataset ? 1398b24ab676SJeff Bonwick os->os_dsl_dataset->ds_object : DMU_META_OBJSET, 1399b24ab676SJeff Bonwick ZB_ROOT_OBJECT, ZB_ROOT_LEVEL, ZB_ROOT_BLKID); 14001b912ec7SGeorge Wilson arc_release(os->os_phys_buf, &os->os_phys_buf); 1401b24ab676SJeff Bonwick 1402adaec86aSMatthew Ahrens dmu_write_policy(os, NULL, 0, 0, &zp); 1403b24ab676SJeff Bonwick 1404b24ab676SJeff Bonwick zio = arc_write(pio, os->os_spa, tx->tx_txg, 1405c166b69dSPaul Dagnelie blkptr_copy, os->os_phys_buf, DMU_OS_IS_L2CACHEABLE(os), 14068df0bcf0SPaul Dagnelie &zp, dmu_objset_write_ready, NULL, NULL, dmu_objset_write_done, 14078df0bcf0SPaul Dagnelie os, ZIO_PRIORITY_ASYNC_WRITE, ZIO_FLAG_MUSTSUCCEED, &zb); 1408c717a561Smaybee 1409c717a561Smaybee /* 141014843421SMatthew Ahrens * Sync special dnodes - the parent IO for the sync is the root block 1411c717a561Smaybee */ 1412744947dcSTom Erickson DMU_META_DNODE(os)->dn_zio = zio; 1413744947dcSTom Erickson dnode_sync(DMU_META_DNODE(os), tx); 1414c717a561Smaybee 141514843421SMatthew Ahrens os->os_phys->os_flags = os->os_flags; 141614843421SMatthew Ahrens 1417744947dcSTom Erickson if (DMU_USERUSED_DNODE(os) && 1418744947dcSTom Erickson DMU_USERUSED_DNODE(os)->dn_type != DMU_OT_NONE) { 1419744947dcSTom Erickson DMU_USERUSED_DNODE(os)->dn_zio = zio; 1420744947dcSTom Erickson dnode_sync(DMU_USERUSED_DNODE(os), tx); 1421744947dcSTom Erickson DMU_GROUPUSED_DNODE(os)->dn_zio = zio; 1422744947dcSTom Erickson dnode_sync(DMU_GROUPUSED_DNODE(os), tx); 142314843421SMatthew Ahrens } 142414843421SMatthew Ahrens 1425c717a561Smaybee txgoff = tx->tx_txg & TXG_MASK; 1426fa9e4066Sahrens 142714843421SMatthew Ahrens if (dmu_objset_userused_enabled(os)) { 142814843421SMatthew Ahrens /* 142914843421SMatthew Ahrens * We must create the list here because it uses the 143094c2d0ebSMatthew Ahrens * dn_dirty_link[] of this txg. But it may already 143194c2d0ebSMatthew Ahrens * exist because we call dsl_dataset_sync() twice per txg. 143214843421SMatthew Ahrens */ 143394c2d0ebSMatthew Ahrens if (os->os_synced_dnodes == NULL) { 143494c2d0ebSMatthew Ahrens os->os_synced_dnodes = 143594c2d0ebSMatthew Ahrens multilist_create(sizeof (dnode_t), 143694c2d0ebSMatthew Ahrens offsetof(dnode_t, dn_dirty_link[txgoff]), 143794c2d0ebSMatthew Ahrens dnode_multilist_index_func); 143894c2d0ebSMatthew Ahrens } else { 143994c2d0ebSMatthew Ahrens ASSERT3U(os->os_synced_dnodes->ml_offset, ==, 144094c2d0ebSMatthew Ahrens offsetof(dnode_t, dn_dirty_link[txgoff])); 144194c2d0ebSMatthew Ahrens } 144214843421SMatthew Ahrens } 144314843421SMatthew Ahrens 144494c2d0ebSMatthew Ahrens for (int i = 0; 144594c2d0ebSMatthew Ahrens i < multilist_get_num_sublists(os->os_dirty_dnodes[txgoff]); i++) { 144694c2d0ebSMatthew Ahrens sync_dnodes_arg_t *sda = kmem_alloc(sizeof (*sda), KM_SLEEP); 144794c2d0ebSMatthew Ahrens sda->sda_list = os->os_dirty_dnodes[txgoff]; 144894c2d0ebSMatthew Ahrens sda->sda_sublist_idx = i; 144994c2d0ebSMatthew Ahrens sda->sda_tx = tx; 145094c2d0ebSMatthew Ahrens (void) taskq_dispatch(dmu_objset_pool(os)->dp_sync_taskq, 145194c2d0ebSMatthew Ahrens sync_dnodes_task, sda, 0); 145294c2d0ebSMatthew Ahrens /* callback frees sda */ 145394c2d0ebSMatthew Ahrens } 145494c2d0ebSMatthew Ahrens taskq_wait(dmu_objset_pool(os)->dp_sync_taskq); 1455fa9e4066Sahrens 1456744947dcSTom Erickson list = &DMU_META_DNODE(os)->dn_dirty_records[txgoff]; 145794c2d0ebSMatthew Ahrens while ((dr = list_head(list)) != NULL) { 14583b2aab18SMatthew Ahrens ASSERT0(dr->dr_dbuf->db_level); 1459c717a561Smaybee list_remove(list, dr); 1460c717a561Smaybee if (dr->dr_zio) 1461c717a561Smaybee zio_nowait(dr->dr_zio); 1462c717a561Smaybee } 1463af346df5SNed Bass 1464af346df5SNed Bass /* Enable dnode backfill if enough objects have been freed. */ 1465af346df5SNed Bass if (os->os_freed_dnodes >= dmu_rescan_dnode_threshold) { 1466af346df5SNed Bass os->os_rescan_dnodes = B_TRUE; 1467af346df5SNed Bass os->os_freed_dnodes = 0; 1468af346df5SNed Bass } 1469af346df5SNed Bass 1470c717a561Smaybee /* 1471c717a561Smaybee * Free intent log blocks up to this tx. 1472c717a561Smaybee */ 1473c717a561Smaybee zil_sync(os->os_zil, tx); 1474088f3894Sahrens os->os_phys->os_zil_header = os->os_zil_header; 1475c717a561Smaybee zio_nowait(zio); 1476fa9e4066Sahrens } 1477fa9e4066Sahrens 1478b24ab676SJeff Bonwick boolean_t 1479b24ab676SJeff Bonwick dmu_objset_is_dirty(objset_t *os, uint64_t txg) 1480b24ab676SJeff Bonwick { 148194c2d0ebSMatthew Ahrens return (!multilist_is_empty(os->os_dirty_dnodes[txg & TXG_MASK])); 1482b24ab676SJeff Bonwick } 1483b24ab676SJeff Bonwick 1484744947dcSTom Erickson static objset_used_cb_t *used_cbs[DMU_OST_NUMTYPES]; 148514843421SMatthew Ahrens 148614843421SMatthew Ahrens void 148714843421SMatthew Ahrens dmu_objset_register_type(dmu_objset_type_t ost, objset_used_cb_t *cb) 148814843421SMatthew Ahrens { 148914843421SMatthew Ahrens used_cbs[ost] = cb; 149014843421SMatthew Ahrens } 149114843421SMatthew Ahrens 149214843421SMatthew Ahrens boolean_t 1493503ad85cSMatthew Ahrens dmu_objset_userused_enabled(objset_t *os) 149414843421SMatthew Ahrens { 149514843421SMatthew Ahrens return (spa_version(os->os_spa) >= SPA_VERSION_USERSPACE && 1496744947dcSTom Erickson used_cbs[os->os_phys->os_type] != NULL && 1497744947dcSTom Erickson DMU_USERUSED_DNODE(os) != NULL); 149814843421SMatthew Ahrens } 149914843421SMatthew Ahrens 15008b70546bSMatthew Ahrens typedef struct userquota_node { 15018b70546bSMatthew Ahrens uint64_t uqn_id; 15028b70546bSMatthew Ahrens int64_t uqn_delta; 15038b70546bSMatthew Ahrens avl_node_t uqn_node; 15048b70546bSMatthew Ahrens } userquota_node_t; 15058b70546bSMatthew Ahrens 15068b70546bSMatthew Ahrens typedef struct userquota_cache { 15078b70546bSMatthew Ahrens avl_tree_t uqc_user_deltas; 15088b70546bSMatthew Ahrens avl_tree_t uqc_group_deltas; 15098b70546bSMatthew Ahrens } userquota_cache_t; 15108b70546bSMatthew Ahrens 15118b70546bSMatthew Ahrens static int 15128b70546bSMatthew Ahrens userquota_compare(const void *l, const void *r) 15138b70546bSMatthew Ahrens { 15148b70546bSMatthew Ahrens const userquota_node_t *luqn = l; 15158b70546bSMatthew Ahrens const userquota_node_t *ruqn = r; 15168b70546bSMatthew Ahrens 15178b70546bSMatthew Ahrens if (luqn->uqn_id < ruqn->uqn_id) 15188b70546bSMatthew Ahrens return (-1); 15198b70546bSMatthew Ahrens if (luqn->uqn_id > ruqn->uqn_id) 15208b70546bSMatthew Ahrens return (1); 15218b70546bSMatthew Ahrens return (0); 15228b70546bSMatthew Ahrens } 15238b70546bSMatthew Ahrens 15249966ca11SMatthew Ahrens static void 15258b70546bSMatthew Ahrens do_userquota_cacheflush(objset_t *os, userquota_cache_t *cache, dmu_tx_t *tx) 15268b70546bSMatthew Ahrens { 15278b70546bSMatthew Ahrens void *cookie; 15288b70546bSMatthew Ahrens userquota_node_t *uqn; 15298b70546bSMatthew Ahrens 15308b70546bSMatthew Ahrens ASSERT(dmu_tx_is_syncing(tx)); 15318b70546bSMatthew Ahrens 15328b70546bSMatthew Ahrens cookie = NULL; 15338b70546bSMatthew Ahrens while ((uqn = avl_destroy_nodes(&cache->uqc_user_deltas, 15348b70546bSMatthew Ahrens &cookie)) != NULL) { 153594c2d0ebSMatthew Ahrens /* 153694c2d0ebSMatthew Ahrens * os_userused_lock protects against concurrent calls to 153794c2d0ebSMatthew Ahrens * zap_increment_int(). It's needed because zap_increment_int() 153894c2d0ebSMatthew Ahrens * is not thread-safe (i.e. not atomic). 153994c2d0ebSMatthew Ahrens */ 154094c2d0ebSMatthew Ahrens mutex_enter(&os->os_userused_lock); 15418b70546bSMatthew Ahrens VERIFY0(zap_increment_int(os, DMU_USERUSED_OBJECT, 15428b70546bSMatthew Ahrens uqn->uqn_id, uqn->uqn_delta, tx)); 154394c2d0ebSMatthew Ahrens mutex_exit(&os->os_userused_lock); 15448b70546bSMatthew Ahrens kmem_free(uqn, sizeof (*uqn)); 15458b70546bSMatthew Ahrens } 15468b70546bSMatthew Ahrens avl_destroy(&cache->uqc_user_deltas); 15478b70546bSMatthew Ahrens 15488b70546bSMatthew Ahrens cookie = NULL; 15498b70546bSMatthew Ahrens while ((uqn = avl_destroy_nodes(&cache->uqc_group_deltas, 15508b70546bSMatthew Ahrens &cookie)) != NULL) { 155194c2d0ebSMatthew Ahrens mutex_enter(&os->os_userused_lock); 15528b70546bSMatthew Ahrens VERIFY0(zap_increment_int(os, DMU_GROUPUSED_OBJECT, 15538b70546bSMatthew Ahrens uqn->uqn_id, uqn->uqn_delta, tx)); 155494c2d0ebSMatthew Ahrens mutex_exit(&os->os_userused_lock); 15558b70546bSMatthew Ahrens kmem_free(uqn, sizeof (*uqn)); 15568b70546bSMatthew Ahrens } 15578b70546bSMatthew Ahrens avl_destroy(&cache->uqc_group_deltas); 15588b70546bSMatthew Ahrens } 15598b70546bSMatthew Ahrens 15608b70546bSMatthew Ahrens static void 15618b70546bSMatthew Ahrens userquota_update_cache(avl_tree_t *avl, uint64_t id, int64_t delta) 15628b70546bSMatthew Ahrens { 15638b70546bSMatthew Ahrens userquota_node_t search = { .uqn_id = id }; 15648b70546bSMatthew Ahrens avl_index_t idx; 15658b70546bSMatthew Ahrens 15668b70546bSMatthew Ahrens userquota_node_t *uqn = avl_find(avl, &search, &idx); 15678b70546bSMatthew Ahrens if (uqn == NULL) { 15688b70546bSMatthew Ahrens uqn = kmem_zalloc(sizeof (*uqn), KM_SLEEP); 15698b70546bSMatthew Ahrens uqn->uqn_id = id; 15708b70546bSMatthew Ahrens avl_insert(avl, uqn, idx); 15718b70546bSMatthew Ahrens } 15728b70546bSMatthew Ahrens uqn->uqn_delta += delta; 15738b70546bSMatthew Ahrens } 15748b70546bSMatthew Ahrens 15758b70546bSMatthew Ahrens static void 15768b70546bSMatthew Ahrens do_userquota_update(userquota_cache_t *cache, uint64_t used, uint64_t flags, 15778b70546bSMatthew Ahrens uint64_t user, uint64_t group, boolean_t subtract) 15789966ca11SMatthew Ahrens { 15790a586ceaSMark Shellenbaum if ((flags & DNODE_FLAG_USERUSED_ACCOUNTED)) { 158054811da5SToomas Soome int64_t delta = DNODE_MIN_SIZE + used; 15819966ca11SMatthew Ahrens if (subtract) 15829966ca11SMatthew Ahrens delta = -delta; 15838b70546bSMatthew Ahrens 15848b70546bSMatthew Ahrens userquota_update_cache(&cache->uqc_user_deltas, user, delta); 15858b70546bSMatthew Ahrens userquota_update_cache(&cache->uqc_group_deltas, group, delta); 15869966ca11SMatthew Ahrens } 15879966ca11SMatthew Ahrens } 15889966ca11SMatthew Ahrens 158994c2d0ebSMatthew Ahrens typedef struct userquota_updates_arg { 159094c2d0ebSMatthew Ahrens objset_t *uua_os; 159194c2d0ebSMatthew Ahrens int uua_sublist_idx; 159294c2d0ebSMatthew Ahrens dmu_tx_t *uua_tx; 159394c2d0ebSMatthew Ahrens } userquota_updates_arg_t; 159494c2d0ebSMatthew Ahrens 159594c2d0ebSMatthew Ahrens static void 159694c2d0ebSMatthew Ahrens userquota_updates_task(void *arg) 159714843421SMatthew Ahrens { 159894c2d0ebSMatthew Ahrens userquota_updates_arg_t *uua = arg; 159994c2d0ebSMatthew Ahrens objset_t *os = uua->uua_os; 160094c2d0ebSMatthew Ahrens dmu_tx_t *tx = uua->uua_tx; 160114843421SMatthew Ahrens dnode_t *dn; 16028b70546bSMatthew Ahrens userquota_cache_t cache = { 0 }; 160314843421SMatthew Ahrens 160494c2d0ebSMatthew Ahrens multilist_sublist_t *list = 160594c2d0ebSMatthew Ahrens multilist_sublist_lock(os->os_synced_dnodes, uua->uua_sublist_idx); 160614843421SMatthew Ahrens 160794c2d0ebSMatthew Ahrens ASSERT(multilist_sublist_head(list) == NULL || 160894c2d0ebSMatthew Ahrens dmu_objset_userused_enabled(os)); 16098b70546bSMatthew Ahrens avl_create(&cache.uqc_user_deltas, userquota_compare, 16108b70546bSMatthew Ahrens sizeof (userquota_node_t), offsetof(userquota_node_t, uqn_node)); 16118b70546bSMatthew Ahrens avl_create(&cache.uqc_group_deltas, userquota_compare, 16128b70546bSMatthew Ahrens sizeof (userquota_node_t), offsetof(userquota_node_t, uqn_node)); 16138b70546bSMatthew Ahrens 161494c2d0ebSMatthew Ahrens while ((dn = multilist_sublist_head(list)) != NULL) { 16151d8ccc7bSMark Shellenbaum int flags; 161614843421SMatthew Ahrens ASSERT(!DMU_OBJECT_IS_SPECIAL(dn->dn_object)); 161714843421SMatthew Ahrens ASSERT(dn->dn_phys->dn_type == DMU_OT_NONE || 161814843421SMatthew Ahrens dn->dn_phys->dn_flags & 161914843421SMatthew Ahrens DNODE_FLAG_USERUSED_ACCOUNTED); 162014843421SMatthew Ahrens 16211d8ccc7bSMark Shellenbaum flags = dn->dn_id_flags; 16221d8ccc7bSMark Shellenbaum ASSERT(flags); 16231d8ccc7bSMark Shellenbaum if (flags & DN_ID_OLD_EXIST) { 16248b70546bSMatthew Ahrens do_userquota_update(&cache, 16258b70546bSMatthew Ahrens dn->dn_oldused, dn->dn_oldflags, 16268b70546bSMatthew Ahrens dn->dn_olduid, dn->dn_oldgid, B_TRUE); 16270a586ceaSMark Shellenbaum } 16281d8ccc7bSMark Shellenbaum if (flags & DN_ID_NEW_EXIST) { 16298b70546bSMatthew Ahrens do_userquota_update(&cache, 16308b70546bSMatthew Ahrens DN_USED_BYTES(dn->dn_phys), 16310a586ceaSMark Shellenbaum dn->dn_phys->dn_flags, dn->dn_newuid, 16328b70546bSMatthew Ahrens dn->dn_newgid, B_FALSE); 16330a586ceaSMark Shellenbaum } 16340a586ceaSMark Shellenbaum 16351d8ccc7bSMark Shellenbaum mutex_enter(&dn->dn_mtx); 16360a586ceaSMark Shellenbaum dn->dn_oldused = 0; 16370a586ceaSMark Shellenbaum dn->dn_oldflags = 0; 16380a586ceaSMark Shellenbaum if (dn->dn_id_flags & DN_ID_NEW_EXIST) { 16390a586ceaSMark Shellenbaum dn->dn_olduid = dn->dn_newuid; 16400a586ceaSMark Shellenbaum dn->dn_oldgid = dn->dn_newgid; 16410a586ceaSMark Shellenbaum dn->dn_id_flags |= DN_ID_OLD_EXIST; 16420a586ceaSMark Shellenbaum if (dn->dn_bonuslen == 0) 16430a586ceaSMark Shellenbaum dn->dn_id_flags |= DN_ID_CHKED_SPILL; 16440a586ceaSMark Shellenbaum else 16450a586ceaSMark Shellenbaum dn->dn_id_flags |= DN_ID_CHKED_BONUS; 16460a586ceaSMark Shellenbaum } 164728d97a71SMark Shellenbaum dn->dn_id_flags &= ~(DN_ID_NEW_EXIST); 1648aa02ea01STom Caputi if (dn->dn_dirty_txg == spa_syncing_txg(os->os_spa)) 1649aa02ea01STom Caputi dn->dn_dirty_txg = 0; 165014843421SMatthew Ahrens mutex_exit(&dn->dn_mtx); 165114843421SMatthew Ahrens 165294c2d0ebSMatthew Ahrens multilist_sublist_remove(list, dn); 165394c2d0ebSMatthew Ahrens dnode_rele(dn, os->os_synced_dnodes); 165414843421SMatthew Ahrens } 16558b70546bSMatthew Ahrens do_userquota_cacheflush(os, &cache, tx); 165694c2d0ebSMatthew Ahrens multilist_sublist_unlock(list); 165794c2d0ebSMatthew Ahrens kmem_free(uua, sizeof (*uua)); 165894c2d0ebSMatthew Ahrens } 165994c2d0ebSMatthew Ahrens 166094c2d0ebSMatthew Ahrens void 166194c2d0ebSMatthew Ahrens dmu_objset_do_userquota_updates(objset_t *os, dmu_tx_t *tx) 166294c2d0ebSMatthew Ahrens { 166394c2d0ebSMatthew Ahrens if (!dmu_objset_userused_enabled(os)) 166494c2d0ebSMatthew Ahrens return; 166594c2d0ebSMatthew Ahrens 166694c2d0ebSMatthew Ahrens /* Allocate the user/groupused objects if necessary. */ 166794c2d0ebSMatthew Ahrens if (DMU_USERUSED_DNODE(os)->dn_type == DMU_OT_NONE) { 166894c2d0ebSMatthew Ahrens VERIFY0(zap_create_claim(os, 166994c2d0ebSMatthew Ahrens DMU_USERUSED_OBJECT, 167094c2d0ebSMatthew Ahrens DMU_OT_USERGROUP_USED, DMU_OT_NONE, 0, tx)); 167194c2d0ebSMatthew Ahrens VERIFY0(zap_create_claim(os, 167294c2d0ebSMatthew Ahrens DMU_GROUPUSED_OBJECT, 167394c2d0ebSMatthew Ahrens DMU_OT_USERGROUP_USED, DMU_OT_NONE, 0, tx)); 167494c2d0ebSMatthew Ahrens } 167594c2d0ebSMatthew Ahrens 167694c2d0ebSMatthew Ahrens for (int i = 0; 167794c2d0ebSMatthew Ahrens i < multilist_get_num_sublists(os->os_synced_dnodes); i++) { 167894c2d0ebSMatthew Ahrens userquota_updates_arg_t *uua = 167994c2d0ebSMatthew Ahrens kmem_alloc(sizeof (*uua), KM_SLEEP); 168094c2d0ebSMatthew Ahrens uua->uua_os = os; 168194c2d0ebSMatthew Ahrens uua->uua_sublist_idx = i; 168294c2d0ebSMatthew Ahrens uua->uua_tx = tx; 168394c2d0ebSMatthew Ahrens /* note: caller does taskq_wait() */ 168494c2d0ebSMatthew Ahrens (void) taskq_dispatch(dmu_objset_pool(os)->dp_sync_taskq, 168594c2d0ebSMatthew Ahrens userquota_updates_task, uua, 0); 168694c2d0ebSMatthew Ahrens /* callback frees uua */ 168794c2d0ebSMatthew Ahrens } 168814843421SMatthew Ahrens } 168914843421SMatthew Ahrens 169006e0070dSMark Shellenbaum /* 169106e0070dSMark Shellenbaum * Returns a pointer to data to find uid/gid from 169206e0070dSMark Shellenbaum * 169306e0070dSMark Shellenbaum * If a dirty record for transaction group that is syncing can't 169406e0070dSMark Shellenbaum * be found then NULL is returned. In the NULL case it is assumed 169506e0070dSMark Shellenbaum * the uid/gid aren't changing. 169606e0070dSMark Shellenbaum */ 169706e0070dSMark Shellenbaum static void * 169806e0070dSMark Shellenbaum dmu_objset_userquota_find_data(dmu_buf_impl_t *db, dmu_tx_t *tx) 169906e0070dSMark Shellenbaum { 170006e0070dSMark Shellenbaum dbuf_dirty_record_t *dr, **drp; 170106e0070dSMark Shellenbaum void *data; 170206e0070dSMark Shellenbaum 170306e0070dSMark Shellenbaum if (db->db_dirtycnt == 0) 170406e0070dSMark Shellenbaum return (db->db.db_data); /* Nothing is changing */ 170506e0070dSMark Shellenbaum 170606e0070dSMark Shellenbaum for (drp = &db->db_last_dirty; (dr = *drp) != NULL; drp = &dr->dr_next) 170706e0070dSMark Shellenbaum if (dr->dr_txg == tx->tx_txg) 170806e0070dSMark Shellenbaum break; 170906e0070dSMark Shellenbaum 1710744947dcSTom Erickson if (dr == NULL) { 171106e0070dSMark Shellenbaum data = NULL; 1712744947dcSTom Erickson } else { 1713744947dcSTom Erickson dnode_t *dn; 1714744947dcSTom Erickson 1715744947dcSTom Erickson DB_DNODE_ENTER(dr->dr_dbuf); 1716744947dcSTom Erickson dn = DB_DNODE(dr->dr_dbuf); 1717744947dcSTom Erickson 1718744947dcSTom Erickson if (dn->dn_bonuslen == 0 && 1719744947dcSTom Erickson dr->dr_dbuf->db_blkid == DMU_SPILL_BLKID) 1720744947dcSTom Erickson data = dr->dt.dl.dr_data->b_data; 1721744947dcSTom Erickson else 1722744947dcSTom Erickson data = dr->dt.dl.dr_data; 1723744947dcSTom Erickson 1724744947dcSTom Erickson DB_DNODE_EXIT(dr->dr_dbuf); 1725744947dcSTom Erickson } 1726744947dcSTom Erickson 172706e0070dSMark Shellenbaum return (data); 172806e0070dSMark Shellenbaum } 172906e0070dSMark Shellenbaum 17300a586ceaSMark Shellenbaum void 173106e0070dSMark Shellenbaum dmu_objset_userquota_get_ids(dnode_t *dn, boolean_t before, dmu_tx_t *tx) 17320a586ceaSMark Shellenbaum { 17330a586ceaSMark Shellenbaum objset_t *os = dn->dn_objset; 17340a586ceaSMark Shellenbaum void *data = NULL; 173506e0070dSMark Shellenbaum dmu_buf_impl_t *db = NULL; 1736d5285caeSGeorge Wilson uint64_t *user = NULL; 1737d5285caeSGeorge Wilson uint64_t *group = NULL; 17380a586ceaSMark Shellenbaum int flags = dn->dn_id_flags; 17390a586ceaSMark Shellenbaum int error; 174006e0070dSMark Shellenbaum boolean_t have_spill = B_FALSE; 17410a586ceaSMark Shellenbaum 17420a586ceaSMark Shellenbaum if (!dmu_objset_userused_enabled(dn->dn_objset)) 17430a586ceaSMark Shellenbaum return; 17440a586ceaSMark Shellenbaum 17450a586ceaSMark Shellenbaum if (before && (flags & (DN_ID_CHKED_BONUS|DN_ID_OLD_EXIST| 17460a586ceaSMark Shellenbaum DN_ID_CHKED_SPILL))) 17470a586ceaSMark Shellenbaum return; 17480a586ceaSMark Shellenbaum 17490a586ceaSMark Shellenbaum if (before && dn->dn_bonuslen != 0) 17500a586ceaSMark Shellenbaum data = DN_BONUS(dn->dn_phys); 175106e0070dSMark Shellenbaum else if (!before && dn->dn_bonuslen != 0) { 175206e0070dSMark Shellenbaum if (dn->dn_bonus) { 175306e0070dSMark Shellenbaum db = dn->dn_bonus; 175406e0070dSMark Shellenbaum mutex_enter(&db->db_mtx); 175506e0070dSMark Shellenbaum data = dmu_objset_userquota_find_data(db, tx); 175606e0070dSMark Shellenbaum } else { 175706e0070dSMark Shellenbaum data = DN_BONUS(dn->dn_phys); 175806e0070dSMark Shellenbaum } 175906e0070dSMark Shellenbaum } else if (dn->dn_bonuslen == 0 && dn->dn_bonustype == DMU_OT_SA) { 17600a586ceaSMark Shellenbaum int rf = 0; 17610a586ceaSMark Shellenbaum 17620a586ceaSMark Shellenbaum if (RW_WRITE_HELD(&dn->dn_struct_rwlock)) 17630a586ceaSMark Shellenbaum rf |= DB_RF_HAVESTRUCT; 17641d8ccc7bSMark Shellenbaum error = dmu_spill_hold_by_dnode(dn, 17651d8ccc7bSMark Shellenbaum rf | DB_RF_MUST_SUCCEED, 176606e0070dSMark Shellenbaum FTAG, (dmu_buf_t **)&db); 17670a586ceaSMark Shellenbaum ASSERT(error == 0); 176806e0070dSMark Shellenbaum mutex_enter(&db->db_mtx); 176906e0070dSMark Shellenbaum data = (before) ? db->db.db_data : 177006e0070dSMark Shellenbaum dmu_objset_userquota_find_data(db, tx); 177106e0070dSMark Shellenbaum have_spill = B_TRUE; 17720a586ceaSMark Shellenbaum } else { 17730a586ceaSMark Shellenbaum mutex_enter(&dn->dn_mtx); 17740a586ceaSMark Shellenbaum dn->dn_id_flags |= DN_ID_CHKED_BONUS; 17750a586ceaSMark Shellenbaum mutex_exit(&dn->dn_mtx); 17760a586ceaSMark Shellenbaum return; 17770a586ceaSMark Shellenbaum } 17780a586ceaSMark Shellenbaum 17790a586ceaSMark Shellenbaum if (before) { 178006e0070dSMark Shellenbaum ASSERT(data); 17810a586ceaSMark Shellenbaum user = &dn->dn_olduid; 17820a586ceaSMark Shellenbaum group = &dn->dn_oldgid; 178306e0070dSMark Shellenbaum } else if (data) { 17840a586ceaSMark Shellenbaum user = &dn->dn_newuid; 17850a586ceaSMark Shellenbaum group = &dn->dn_newgid; 17860a586ceaSMark Shellenbaum } 17870a586ceaSMark Shellenbaum 178806e0070dSMark Shellenbaum /* 178906e0070dSMark Shellenbaum * Must always call the callback in case the object 179006e0070dSMark Shellenbaum * type has changed and that type isn't an object type to track 179106e0070dSMark Shellenbaum */ 17920a586ceaSMark Shellenbaum error = used_cbs[os->os_phys->os_type](dn->dn_bonustype, data, 17930a586ceaSMark Shellenbaum user, group); 17940a586ceaSMark Shellenbaum 179506e0070dSMark Shellenbaum /* 179606e0070dSMark Shellenbaum * Preserve existing uid/gid when the callback can't determine 179706e0070dSMark Shellenbaum * what the new uid/gid are and the callback returned EEXIST. 179806e0070dSMark Shellenbaum * The EEXIST error tells us to just use the existing uid/gid. 179906e0070dSMark Shellenbaum * If we don't know what the old values are then just assign 180006e0070dSMark Shellenbaum * them to 0, since that is a new file being created. 180106e0070dSMark Shellenbaum */ 180206e0070dSMark Shellenbaum if (!before && data == NULL && error == EEXIST) { 180306e0070dSMark Shellenbaum if (flags & DN_ID_OLD_EXIST) { 180406e0070dSMark Shellenbaum dn->dn_newuid = dn->dn_olduid; 180506e0070dSMark Shellenbaum dn->dn_newgid = dn->dn_oldgid; 180606e0070dSMark Shellenbaum } else { 180706e0070dSMark Shellenbaum dn->dn_newuid = 0; 180806e0070dSMark Shellenbaum dn->dn_newgid = 0; 180906e0070dSMark Shellenbaum } 181006e0070dSMark Shellenbaum error = 0; 181106e0070dSMark Shellenbaum } 181206e0070dSMark Shellenbaum 181306e0070dSMark Shellenbaum if (db) 181406e0070dSMark Shellenbaum mutex_exit(&db->db_mtx); 181506e0070dSMark Shellenbaum 18160a586ceaSMark Shellenbaum mutex_enter(&dn->dn_mtx); 18170a586ceaSMark Shellenbaum if (error == 0 && before) 18180a586ceaSMark Shellenbaum dn->dn_id_flags |= DN_ID_OLD_EXIST; 18190a586ceaSMark Shellenbaum if (error == 0 && !before) 18200a586ceaSMark Shellenbaum dn->dn_id_flags |= DN_ID_NEW_EXIST; 18210a586ceaSMark Shellenbaum 182206e0070dSMark Shellenbaum if (have_spill) { 18230a586ceaSMark Shellenbaum dn->dn_id_flags |= DN_ID_CHKED_SPILL; 18240a586ceaSMark Shellenbaum } else { 18250a586ceaSMark Shellenbaum dn->dn_id_flags |= DN_ID_CHKED_BONUS; 18260a586ceaSMark Shellenbaum } 18270a586ceaSMark Shellenbaum mutex_exit(&dn->dn_mtx); 182806e0070dSMark Shellenbaum if (have_spill) 182906e0070dSMark Shellenbaum dmu_buf_rele((dmu_buf_t *)db, FTAG); 18300a586ceaSMark Shellenbaum } 18310a586ceaSMark Shellenbaum 183214843421SMatthew Ahrens boolean_t 183314843421SMatthew Ahrens dmu_objset_userspace_present(objset_t *os) 183414843421SMatthew Ahrens { 1835503ad85cSMatthew Ahrens return (os->os_phys->os_flags & 183614843421SMatthew Ahrens OBJSET_FLAG_USERACCOUNTING_COMPLETE); 183714843421SMatthew Ahrens } 183814843421SMatthew Ahrens 183914843421SMatthew Ahrens int 184014843421SMatthew Ahrens dmu_objset_userspace_upgrade(objset_t *os) 184114843421SMatthew Ahrens { 184214843421SMatthew Ahrens uint64_t obj; 184314843421SMatthew Ahrens int err = 0; 184414843421SMatthew Ahrens 184514843421SMatthew Ahrens if (dmu_objset_userspace_present(os)) 184614843421SMatthew Ahrens return (0); 1847503ad85cSMatthew Ahrens if (!dmu_objset_userused_enabled(os)) 1848be6fd75aSMatthew Ahrens return (SET_ERROR(ENOTSUP)); 184914843421SMatthew Ahrens if (dmu_objset_is_snapshot(os)) 1850be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 185114843421SMatthew Ahrens 185214843421SMatthew Ahrens /* 185314843421SMatthew Ahrens * We simply need to mark every object dirty, so that it will be 185414843421SMatthew Ahrens * synced out and now accounted. If this is called 185514843421SMatthew Ahrens * concurrently, or if we already did some work before crashing, 185614843421SMatthew Ahrens * that's fine, since we track each object's accounted state 185714843421SMatthew Ahrens * independently. 185814843421SMatthew Ahrens */ 185914843421SMatthew Ahrens 186014843421SMatthew Ahrens for (obj = 0; err == 0; err = dmu_object_next(os, &obj, FALSE, 0)) { 186168857716SLin Ling dmu_tx_t *tx; 186214843421SMatthew Ahrens dmu_buf_t *db; 186314843421SMatthew Ahrens int objerr; 186414843421SMatthew Ahrens 186514843421SMatthew Ahrens if (issig(JUSTLOOKING) && issig(FORREAL)) 1866be6fd75aSMatthew Ahrens return (SET_ERROR(EINTR)); 186714843421SMatthew Ahrens 186814843421SMatthew Ahrens objerr = dmu_bonus_hold(os, obj, FTAG, &db); 18693b2aab18SMatthew Ahrens if (objerr != 0) 187014843421SMatthew Ahrens continue; 187168857716SLin Ling tx = dmu_tx_create(os); 187214843421SMatthew Ahrens dmu_tx_hold_bonus(tx, obj); 187314843421SMatthew Ahrens objerr = dmu_tx_assign(tx, TXG_WAIT); 18743b2aab18SMatthew Ahrens if (objerr != 0) { 187514843421SMatthew Ahrens dmu_tx_abort(tx); 187614843421SMatthew Ahrens continue; 187714843421SMatthew Ahrens } 187814843421SMatthew Ahrens dmu_buf_will_dirty(db, tx); 187914843421SMatthew Ahrens dmu_buf_rele(db, FTAG); 188014843421SMatthew Ahrens dmu_tx_commit(tx); 188114843421SMatthew Ahrens } 188214843421SMatthew Ahrens 1883503ad85cSMatthew Ahrens os->os_flags |= OBJSET_FLAG_USERACCOUNTING_COMPLETE; 188414843421SMatthew Ahrens txg_wait_synced(dmu_objset_pool(os), 0); 188514843421SMatthew Ahrens return (0); 188614843421SMatthew Ahrens } 188714843421SMatthew Ahrens 1888fa9e4066Sahrens void 1889a2eea2e1Sahrens dmu_objset_space(objset_t *os, uint64_t *refdbytesp, uint64_t *availbytesp, 1890a2eea2e1Sahrens uint64_t *usedobjsp, uint64_t *availobjsp) 1891fa9e4066Sahrens { 1892503ad85cSMatthew Ahrens dsl_dataset_space(os->os_dsl_dataset, refdbytesp, availbytesp, 1893a2eea2e1Sahrens usedobjsp, availobjsp); 1894a2eea2e1Sahrens } 1895a2eea2e1Sahrens 1896a2eea2e1Sahrens uint64_t 1897a2eea2e1Sahrens dmu_objset_fsid_guid(objset_t *os) 1898a2eea2e1Sahrens { 1899503ad85cSMatthew Ahrens return (dsl_dataset_fsid_guid(os->os_dsl_dataset)); 1900a2eea2e1Sahrens } 1901a2eea2e1Sahrens 1902a2eea2e1Sahrens void 1903a2eea2e1Sahrens dmu_objset_fast_stat(objset_t *os, dmu_objset_stats_t *stat) 1904a2eea2e1Sahrens { 1905503ad85cSMatthew Ahrens stat->dds_type = os->os_phys->os_type; 1906503ad85cSMatthew Ahrens if (os->os_dsl_dataset) 1907503ad85cSMatthew Ahrens dsl_dataset_fast_stat(os->os_dsl_dataset, stat); 1908a2eea2e1Sahrens } 1909a2eea2e1Sahrens 1910a2eea2e1Sahrens void 1911a2eea2e1Sahrens dmu_objset_stats(objset_t *os, nvlist_t *nv) 1912a2eea2e1Sahrens { 1913503ad85cSMatthew Ahrens ASSERT(os->os_dsl_dataset || 1914503ad85cSMatthew Ahrens os->os_phys->os_type == DMU_OST_META); 1915a2eea2e1Sahrens 1916503ad85cSMatthew Ahrens if (os->os_dsl_dataset != NULL) 1917503ad85cSMatthew Ahrens dsl_dataset_stats(os->os_dsl_dataset, nv); 1918a2eea2e1Sahrens 1919a2eea2e1Sahrens dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_TYPE, 1920503ad85cSMatthew Ahrens os->os_phys->os_type); 192114843421SMatthew Ahrens dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USERACCOUNTING, 192214843421SMatthew Ahrens dmu_objset_userspace_present(os)); 1923fa9e4066Sahrens } 1924fa9e4066Sahrens 1925fa9e4066Sahrens int 1926fa9e4066Sahrens dmu_objset_is_snapshot(objset_t *os) 1927fa9e4066Sahrens { 1928503ad85cSMatthew Ahrens if (os->os_dsl_dataset != NULL) 1929bc9014e6SJustin Gibbs return (os->os_dsl_dataset->ds_is_snapshot); 1930fa9e4066Sahrens else 1931fa9e4066Sahrens return (B_FALSE); 1932fa9e4066Sahrens } 1933fa9e4066Sahrens 1934ab04eb8eStimh int 1935ab04eb8eStimh dmu_snapshot_realname(objset_t *os, char *name, char *real, int maxlen, 1936ab04eb8eStimh boolean_t *conflict) 1937ab04eb8eStimh { 1938503ad85cSMatthew Ahrens dsl_dataset_t *ds = os->os_dsl_dataset; 1939ab04eb8eStimh uint64_t ignored; 1940ab04eb8eStimh 1941c1379625SJustin T. Gibbs if (dsl_dataset_phys(ds)->ds_snapnames_zapobj == 0) 1942be6fd75aSMatthew Ahrens return (SET_ERROR(ENOENT)); 1943ab04eb8eStimh 1944ab04eb8eStimh return (zap_lookup_norm(ds->ds_dir->dd_pool->dp_meta_objset, 1945c1379625SJustin T. Gibbs dsl_dataset_phys(ds)->ds_snapnames_zapobj, name, 8, 1, &ignored, 19461c17160aSKevin Crowe MT_NORMALIZE, real, maxlen, conflict)); 1947ab04eb8eStimh } 1948ab04eb8eStimh 1949fa9e4066Sahrens int 1950fa9e4066Sahrens dmu_snapshot_list_next(objset_t *os, int namelen, char *name, 1951b38f0970Sck uint64_t *idp, uint64_t *offp, boolean_t *case_conflict) 1952fa9e4066Sahrens { 1953503ad85cSMatthew Ahrens dsl_dataset_t *ds = os->os_dsl_dataset; 1954fa9e4066Sahrens zap_cursor_t cursor; 1955fa9e4066Sahrens zap_attribute_t attr; 1956fa9e4066Sahrens 19573b2aab18SMatthew Ahrens ASSERT(dsl_pool_config_held(dmu_objset_pool(os))); 19583b2aab18SMatthew Ahrens 1959c1379625SJustin T. Gibbs if (dsl_dataset_phys(ds)->ds_snapnames_zapobj == 0) 1960be6fd75aSMatthew Ahrens return (SET_ERROR(ENOENT)); 1961fa9e4066Sahrens 1962fa9e4066Sahrens zap_cursor_init_serialized(&cursor, 1963fa9e4066Sahrens ds->ds_dir->dd_pool->dp_meta_objset, 1964c1379625SJustin T. Gibbs dsl_dataset_phys(ds)->ds_snapnames_zapobj, *offp); 1965fa9e4066Sahrens 196687e5029aSahrens if (zap_cursor_retrieve(&cursor, &attr) != 0) { 196787e5029aSahrens zap_cursor_fini(&cursor); 1968be6fd75aSMatthew Ahrens return (SET_ERROR(ENOENT)); 196987e5029aSahrens } 197087e5029aSahrens 197187e5029aSahrens if (strlen(attr.za_name) + 1 > namelen) { 197287e5029aSahrens zap_cursor_fini(&cursor); 1973be6fd75aSMatthew Ahrens return (SET_ERROR(ENAMETOOLONG)); 197487e5029aSahrens } 197587e5029aSahrens 197687e5029aSahrens (void) strcpy(name, attr.za_name); 197787e5029aSahrens if (idp) 197887e5029aSahrens *idp = attr.za_first_integer; 1979b38f0970Sck if (case_conflict) 1980b38f0970Sck *case_conflict = attr.za_normalization_conflict; 198187e5029aSahrens zap_cursor_advance(&cursor); 198287e5029aSahrens *offp = zap_cursor_serialize(&cursor); 198387e5029aSahrens zap_cursor_fini(&cursor); 198487e5029aSahrens 198587e5029aSahrens return (0); 198687e5029aSahrens } 198787e5029aSahrens 198887e5029aSahrens int 198987e5029aSahrens dmu_dir_list_next(objset_t *os, int namelen, char *name, 199087e5029aSahrens uint64_t *idp, uint64_t *offp) 199187e5029aSahrens { 1992503ad85cSMatthew Ahrens dsl_dir_t *dd = os->os_dsl_dataset->ds_dir; 199387e5029aSahrens zap_cursor_t cursor; 199487e5029aSahrens zap_attribute_t attr; 199587e5029aSahrens 199687e5029aSahrens /* there is no next dir on a snapshot! */ 1997503ad85cSMatthew Ahrens if (os->os_dsl_dataset->ds_object != 1998c1379625SJustin T. Gibbs dsl_dir_phys(dd)->dd_head_dataset_obj) 1999be6fd75aSMatthew Ahrens return (SET_ERROR(ENOENT)); 2000fa9e4066Sahrens 200187e5029aSahrens zap_cursor_init_serialized(&cursor, 200287e5029aSahrens dd->dd_pool->dp_meta_objset, 2003c1379625SJustin T. Gibbs dsl_dir_phys(dd)->dd_child_dir_zapobj, *offp); 200487e5029aSahrens 200587e5029aSahrens if (zap_cursor_retrieve(&cursor, &attr) != 0) { 200687e5029aSahrens zap_cursor_fini(&cursor); 2007be6fd75aSMatthew Ahrens return (SET_ERROR(ENOENT)); 200887e5029aSahrens } 200987e5029aSahrens 201087e5029aSahrens if (strlen(attr.za_name) + 1 > namelen) { 201187e5029aSahrens zap_cursor_fini(&cursor); 2012be6fd75aSMatthew Ahrens return (SET_ERROR(ENAMETOOLONG)); 201387e5029aSahrens } 2014fa9e4066Sahrens 2015fa9e4066Sahrens (void) strcpy(name, attr.za_name); 201687e5029aSahrens if (idp) 201787e5029aSahrens *idp = attr.za_first_integer; 2018fa9e4066Sahrens zap_cursor_advance(&cursor); 2019fa9e4066Sahrens *offp = zap_cursor_serialize(&cursor); 202087e5029aSahrens zap_cursor_fini(&cursor); 2021fa9e4066Sahrens 2022fa9e4066Sahrens return (0); 2023fa9e4066Sahrens } 2024fa9e4066Sahrens 202512380e1eSArne Jansen typedef struct dmu_objset_find_ctx { 202612380e1eSArne Jansen taskq_t *dc_tq; 202712380e1eSArne Jansen dsl_pool_t *dc_dp; 202812380e1eSArne Jansen uint64_t dc_ddobj; 20297588687eSMatthew Ahrens char *dc_ddname; /* last component of ddobj's name */ 203012380e1eSArne Jansen int (*dc_func)(dsl_pool_t *, dsl_dataset_t *, void *); 203112380e1eSArne Jansen void *dc_arg; 203212380e1eSArne Jansen int dc_flags; 203312380e1eSArne Jansen kmutex_t *dc_error_lock; 203412380e1eSArne Jansen int *dc_error; 203512380e1eSArne Jansen } dmu_objset_find_ctx_t; 203612380e1eSArne Jansen 203712380e1eSArne Jansen static void 203812380e1eSArne Jansen dmu_objset_find_dp_impl(dmu_objset_find_ctx_t *dcp) 2039088f3894Sahrens { 204012380e1eSArne Jansen dsl_pool_t *dp = dcp->dc_dp; 20413b2aab18SMatthew Ahrens dsl_dir_t *dd; 20423b2aab18SMatthew Ahrens dsl_dataset_t *ds; 20433b2aab18SMatthew Ahrens zap_cursor_t zc; 20443b2aab18SMatthew Ahrens zap_attribute_t *attr; 20453b2aab18SMatthew Ahrens uint64_t thisobj; 204612380e1eSArne Jansen int err = 0; 20473b2aab18SMatthew Ahrens 204812380e1eSArne Jansen /* don't process if there already was an error */ 204912380e1eSArne Jansen if (*dcp->dc_error != 0) 205012380e1eSArne Jansen goto out; 20513b2aab18SMatthew Ahrens 20527588687eSMatthew Ahrens /* 20537588687eSMatthew Ahrens * Note: passing the name (dc_ddname) here is optional, but it 20547588687eSMatthew Ahrens * improves performance because we don't need to call 20557588687eSMatthew Ahrens * zap_value_search() to determine the name. 20567588687eSMatthew Ahrens */ 20577588687eSMatthew Ahrens err = dsl_dir_hold_obj(dp, dcp->dc_ddobj, dcp->dc_ddname, FTAG, &dd); 20583b2aab18SMatthew Ahrens if (err != 0) 205912380e1eSArne Jansen goto out; 20603b2aab18SMatthew Ahrens 20613b2aab18SMatthew Ahrens /* Don't visit hidden ($MOS & $ORIGIN) objsets. */ 20623b2aab18SMatthew Ahrens if (dd->dd_myname[0] == '$') { 20633b2aab18SMatthew Ahrens dsl_dir_rele(dd, FTAG); 206412380e1eSArne Jansen goto out; 20653b2aab18SMatthew Ahrens } 20663b2aab18SMatthew Ahrens 2067c1379625SJustin T. Gibbs thisobj = dsl_dir_phys(dd)->dd_head_dataset_obj; 20683b2aab18SMatthew Ahrens attr = kmem_alloc(sizeof (zap_attribute_t), KM_SLEEP); 20693b2aab18SMatthew Ahrens 20703b2aab18SMatthew Ahrens /* 20713b2aab18SMatthew Ahrens * Iterate over all children. 20723b2aab18SMatthew Ahrens */ 207312380e1eSArne Jansen if (dcp->dc_flags & DS_FIND_CHILDREN) { 20743b2aab18SMatthew Ahrens for (zap_cursor_init(&zc, dp->dp_meta_objset, 2075c1379625SJustin T. Gibbs dsl_dir_phys(dd)->dd_child_dir_zapobj); 20763b2aab18SMatthew Ahrens zap_cursor_retrieve(&zc, attr) == 0; 20773b2aab18SMatthew Ahrens (void) zap_cursor_advance(&zc)) { 20783b2aab18SMatthew Ahrens ASSERT3U(attr->za_integer_length, ==, 20793b2aab18SMatthew Ahrens sizeof (uint64_t)); 20803b2aab18SMatthew Ahrens ASSERT3U(attr->za_num_integers, ==, 1); 20813b2aab18SMatthew Ahrens 20827588687eSMatthew Ahrens dmu_objset_find_ctx_t *child_dcp = 20837588687eSMatthew Ahrens kmem_alloc(sizeof (*child_dcp), KM_SLEEP); 208412380e1eSArne Jansen *child_dcp = *dcp; 208512380e1eSArne Jansen child_dcp->dc_ddobj = attr->za_first_integer; 20867588687eSMatthew Ahrens child_dcp->dc_ddname = spa_strdup(attr->za_name); 208712380e1eSArne Jansen if (dcp->dc_tq != NULL) 208812380e1eSArne Jansen (void) taskq_dispatch(dcp->dc_tq, 208912380e1eSArne Jansen dmu_objset_find_dp_cb, child_dcp, TQ_SLEEP); 209012380e1eSArne Jansen else 209112380e1eSArne Jansen dmu_objset_find_dp_impl(child_dcp); 20923b2aab18SMatthew Ahrens } 20933b2aab18SMatthew Ahrens zap_cursor_fini(&zc); 20943b2aab18SMatthew Ahrens } 20953b2aab18SMatthew Ahrens 20963b2aab18SMatthew Ahrens /* 20973b2aab18SMatthew Ahrens * Iterate over all snapshots. 20983b2aab18SMatthew Ahrens */ 209912380e1eSArne Jansen if (dcp->dc_flags & DS_FIND_SNAPSHOTS) { 21003b2aab18SMatthew Ahrens dsl_dataset_t *ds; 21013b2aab18SMatthew Ahrens err = dsl_dataset_hold_obj(dp, thisobj, FTAG, &ds); 21023b2aab18SMatthew Ahrens 21033b2aab18SMatthew Ahrens if (err == 0) { 2104c1379625SJustin T. Gibbs uint64_t snapobj; 2105c1379625SJustin T. Gibbs 2106c1379625SJustin T. Gibbs snapobj = dsl_dataset_phys(ds)->ds_snapnames_zapobj; 21073b2aab18SMatthew Ahrens dsl_dataset_rele(ds, FTAG); 21083b2aab18SMatthew Ahrens 21093b2aab18SMatthew Ahrens for (zap_cursor_init(&zc, dp->dp_meta_objset, snapobj); 21103b2aab18SMatthew Ahrens zap_cursor_retrieve(&zc, attr) == 0; 21113b2aab18SMatthew Ahrens (void) zap_cursor_advance(&zc)) { 21123b2aab18SMatthew Ahrens ASSERT3U(attr->za_integer_length, ==, 21133b2aab18SMatthew Ahrens sizeof (uint64_t)); 21143b2aab18SMatthew Ahrens ASSERT3U(attr->za_num_integers, ==, 1); 21153b2aab18SMatthew Ahrens 21163b2aab18SMatthew Ahrens err = dsl_dataset_hold_obj(dp, 21173b2aab18SMatthew Ahrens attr->za_first_integer, FTAG, &ds); 21183b2aab18SMatthew Ahrens if (err != 0) 21193b2aab18SMatthew Ahrens break; 212012380e1eSArne Jansen err = dcp->dc_func(dp, ds, dcp->dc_arg); 21213b2aab18SMatthew Ahrens dsl_dataset_rele(ds, FTAG); 21223b2aab18SMatthew Ahrens if (err != 0) 21233b2aab18SMatthew Ahrens break; 21243b2aab18SMatthew Ahrens } 21253b2aab18SMatthew Ahrens zap_cursor_fini(&zc); 21263b2aab18SMatthew Ahrens } 21273b2aab18SMatthew Ahrens } 21283b2aab18SMatthew Ahrens 21293b2aab18SMatthew Ahrens kmem_free(attr, sizeof (zap_attribute_t)); 21303b2aab18SMatthew Ahrens 21317588687eSMatthew Ahrens if (err != 0) { 21327588687eSMatthew Ahrens dsl_dir_rele(dd, FTAG); 213312380e1eSArne Jansen goto out; 21347588687eSMatthew Ahrens } 21353b2aab18SMatthew Ahrens 21363b2aab18SMatthew Ahrens /* 21373b2aab18SMatthew Ahrens * Apply to self. 21383b2aab18SMatthew Ahrens */ 21393b2aab18SMatthew Ahrens err = dsl_dataset_hold_obj(dp, thisobj, FTAG, &ds); 21407588687eSMatthew Ahrens 21417588687eSMatthew Ahrens /* 21427588687eSMatthew Ahrens * Note: we hold the dir while calling dsl_dataset_hold_obj() so 21437588687eSMatthew Ahrens * that the dir will remain cached, and we won't have to re-instantiate 21447588687eSMatthew Ahrens * it (which could be expensive due to finding its name via 21457588687eSMatthew Ahrens * zap_value_search()). 21467588687eSMatthew Ahrens */ 21477588687eSMatthew Ahrens dsl_dir_rele(dd, FTAG); 21483b2aab18SMatthew Ahrens if (err != 0) 214912380e1eSArne Jansen goto out; 215012380e1eSArne Jansen err = dcp->dc_func(dp, ds, dcp->dc_arg); 21513b2aab18SMatthew Ahrens dsl_dataset_rele(ds, FTAG); 215212380e1eSArne Jansen 215312380e1eSArne Jansen out: 215412380e1eSArne Jansen if (err != 0) { 215512380e1eSArne Jansen mutex_enter(dcp->dc_error_lock); 215612380e1eSArne Jansen /* only keep first error */ 215712380e1eSArne Jansen if (*dcp->dc_error == 0) 215812380e1eSArne Jansen *dcp->dc_error = err; 215912380e1eSArne Jansen mutex_exit(dcp->dc_error_lock); 216012380e1eSArne Jansen } 216112380e1eSArne Jansen 21627588687eSMatthew Ahrens if (dcp->dc_ddname != NULL) 21637588687eSMatthew Ahrens spa_strfree(dcp->dc_ddname); 216412380e1eSArne Jansen kmem_free(dcp, sizeof (*dcp)); 216512380e1eSArne Jansen } 216612380e1eSArne Jansen 216712380e1eSArne Jansen static void 216812380e1eSArne Jansen dmu_objset_find_dp_cb(void *arg) 216912380e1eSArne Jansen { 217012380e1eSArne Jansen dmu_objset_find_ctx_t *dcp = arg; 217112380e1eSArne Jansen dsl_pool_t *dp = dcp->dc_dp; 217212380e1eSArne Jansen 21731d3f896fSArne Jansen /* 21741d3f896fSArne Jansen * We need to get a pool_config_lock here, as there are several 21751d3f896fSArne Jansen * asssert(pool_config_held) down the stack. Getting a lock via 21761d3f896fSArne Jansen * dsl_pool_config_enter is risky, as it might be stalled by a 21771d3f896fSArne Jansen * pending writer. This would deadlock, as the write lock can 21781d3f896fSArne Jansen * only be granted when our parent thread gives up the lock. 21791d3f896fSArne Jansen * The _prio interface gives us priority over a pending writer. 21801d3f896fSArne Jansen */ 21811d3f896fSArne Jansen dsl_pool_config_enter_prio(dp, FTAG); 218212380e1eSArne Jansen 218312380e1eSArne Jansen dmu_objset_find_dp_impl(dcp); 218412380e1eSArne Jansen 218512380e1eSArne Jansen dsl_pool_config_exit(dp, FTAG); 218612380e1eSArne Jansen } 218712380e1eSArne Jansen 218812380e1eSArne Jansen /* 218912380e1eSArne Jansen * Find objsets under and including ddobj, call func(ds) on each. 219012380e1eSArne Jansen * The order for the enumeration is completely undefined. 219112380e1eSArne Jansen * func is called with dsl_pool_config held. 219212380e1eSArne Jansen */ 219312380e1eSArne Jansen int 219412380e1eSArne Jansen dmu_objset_find_dp(dsl_pool_t *dp, uint64_t ddobj, 219512380e1eSArne Jansen int func(dsl_pool_t *, dsl_dataset_t *, void *), void *arg, int flags) 219612380e1eSArne Jansen { 219712380e1eSArne Jansen int error = 0; 219812380e1eSArne Jansen taskq_t *tq = NULL; 219912380e1eSArne Jansen int ntasks; 220012380e1eSArne Jansen dmu_objset_find_ctx_t *dcp; 220112380e1eSArne Jansen kmutex_t err_lock; 220212380e1eSArne Jansen 220312380e1eSArne Jansen mutex_init(&err_lock, NULL, MUTEX_DEFAULT, NULL); 220412380e1eSArne Jansen dcp = kmem_alloc(sizeof (*dcp), KM_SLEEP); 220512380e1eSArne Jansen dcp->dc_tq = NULL; 220612380e1eSArne Jansen dcp->dc_dp = dp; 220712380e1eSArne Jansen dcp->dc_ddobj = ddobj; 22087588687eSMatthew Ahrens dcp->dc_ddname = NULL; 220912380e1eSArne Jansen dcp->dc_func = func; 221012380e1eSArne Jansen dcp->dc_arg = arg; 221112380e1eSArne Jansen dcp->dc_flags = flags; 221212380e1eSArne Jansen dcp->dc_error_lock = &err_lock; 221312380e1eSArne Jansen dcp->dc_error = &error; 221412380e1eSArne Jansen 221512380e1eSArne Jansen if ((flags & DS_FIND_SERIALIZE) || dsl_pool_config_held_writer(dp)) { 221612380e1eSArne Jansen /* 221712380e1eSArne Jansen * In case a write lock is held we can't make use of 221812380e1eSArne Jansen * parallelism, as down the stack of the worker threads 221912380e1eSArne Jansen * the lock is asserted via dsl_pool_config_held. 222012380e1eSArne Jansen * In case of a read lock this is solved by getting a read 222112380e1eSArne Jansen * lock in each worker thread, which isn't possible in case 222212380e1eSArne Jansen * of a writer lock. So we fall back to the synchronous path 222312380e1eSArne Jansen * here. 222412380e1eSArne Jansen * In the future it might be possible to get some magic into 222512380e1eSArne Jansen * dsl_pool_config_held in a way that it returns true for 222612380e1eSArne Jansen * the worker threads so that a single lock held from this 222712380e1eSArne Jansen * thread suffices. For now, stay single threaded. 222812380e1eSArne Jansen */ 222912380e1eSArne Jansen dmu_objset_find_dp_impl(dcp); 22302bad2258SSteven Hartland mutex_destroy(&err_lock); 223112380e1eSArne Jansen 223212380e1eSArne Jansen return (error); 223312380e1eSArne Jansen } 223412380e1eSArne Jansen 223512380e1eSArne Jansen ntasks = dmu_find_threads; 223612380e1eSArne Jansen if (ntasks == 0) 223712380e1eSArne Jansen ntasks = vdev_count_leaves(dp->dp_spa) * 4; 223812380e1eSArne Jansen tq = taskq_create("dmu_objset_find", ntasks, minclsyspri, ntasks, 223912380e1eSArne Jansen INT_MAX, 0); 224012380e1eSArne Jansen if (tq == NULL) { 224112380e1eSArne Jansen kmem_free(dcp, sizeof (*dcp)); 22422bad2258SSteven Hartland mutex_destroy(&err_lock); 22432bad2258SSteven Hartland 224412380e1eSArne Jansen return (SET_ERROR(ENOMEM)); 224512380e1eSArne Jansen } 224612380e1eSArne Jansen dcp->dc_tq = tq; 224712380e1eSArne Jansen 224812380e1eSArne Jansen /* dcp will be freed by task */ 224912380e1eSArne Jansen (void) taskq_dispatch(tq, dmu_objset_find_dp_cb, dcp, TQ_SLEEP); 225012380e1eSArne Jansen 225112380e1eSArne Jansen /* 225212380e1eSArne Jansen * PORTING: this code relies on the property of taskq_wait to wait 225312380e1eSArne Jansen * until no more tasks are queued and no more tasks are active. As 225412380e1eSArne Jansen * we always queue new tasks from within other tasks, task_wait 225512380e1eSArne Jansen * reliably waits for the full recursion to finish, even though we 225612380e1eSArne Jansen * enqueue new tasks after taskq_wait has been called. 225712380e1eSArne Jansen * On platforms other than illumos, taskq_wait may not have this 225812380e1eSArne Jansen * property. 225912380e1eSArne Jansen */ 226012380e1eSArne Jansen taskq_wait(tq); 226112380e1eSArne Jansen taskq_destroy(tq); 226212380e1eSArne Jansen mutex_destroy(&err_lock); 226312380e1eSArne Jansen 226412380e1eSArne Jansen return (error); 2265088f3894Sahrens } 2266088f3894Sahrens 2267088f3894Sahrens /* 22683b2aab18SMatthew Ahrens * Find all objsets under name, and for each, call 'func(child_name, arg)'. 22693b2aab18SMatthew Ahrens * The dp_config_rwlock must not be held when this is called, and it 22703b2aab18SMatthew Ahrens * will not be held when the callback is called. 22713b2aab18SMatthew Ahrens * Therefore this function should only be used when the pool is not changing 22723b2aab18SMatthew Ahrens * (e.g. in syncing context), or the callback can deal with the possible races. 2273088f3894Sahrens */ 22743b2aab18SMatthew Ahrens static int 22753b2aab18SMatthew Ahrens dmu_objset_find_impl(spa_t *spa, const char *name, 22763b2aab18SMatthew Ahrens int func(const char *, void *), void *arg, int flags) 2277fa9e4066Sahrens { 2278fa9e4066Sahrens dsl_dir_t *dd; 22793b2aab18SMatthew Ahrens dsl_pool_t *dp = spa_get_dsl(spa); 2280088f3894Sahrens dsl_dataset_t *ds; 2281fa9e4066Sahrens zap_cursor_t zc; 2282b7661cccSmmusante zap_attribute_t *attr; 2283fa9e4066Sahrens char *child; 2284088f3894Sahrens uint64_t thisobj; 2285088f3894Sahrens int err; 2286fa9e4066Sahrens 22873b2aab18SMatthew Ahrens dsl_pool_config_enter(dp, FTAG); 22883b2aab18SMatthew Ahrens 22893b2aab18SMatthew Ahrens err = dsl_dir_hold(dp, name, FTAG, &dd, NULL); 22903b2aab18SMatthew Ahrens if (err != 0) { 22913b2aab18SMatthew Ahrens dsl_pool_config_exit(dp, FTAG); 22921d452cf5Sahrens return (err); 22933b2aab18SMatthew Ahrens } 2294fa9e4066Sahrens 2295088f3894Sahrens /* Don't visit hidden ($MOS & $ORIGIN) objsets. */ 2296088f3894Sahrens if (dd->dd_myname[0] == '$') { 22973b2aab18SMatthew Ahrens dsl_dir_rele(dd, FTAG); 22983b2aab18SMatthew Ahrens dsl_pool_config_exit(dp, FTAG); 2299088f3894Sahrens return (0); 2300088f3894Sahrens } 2301088f3894Sahrens 2302c1379625SJustin T. Gibbs thisobj = dsl_dir_phys(dd)->dd_head_dataset_obj; 2303b7661cccSmmusante attr = kmem_alloc(sizeof (zap_attribute_t), KM_SLEEP); 2304fa9e4066Sahrens 2305fa9e4066Sahrens /* 2306fa9e4066Sahrens * Iterate over all children. 2307fa9e4066Sahrens */ 23080b69c2f0Sahrens if (flags & DS_FIND_CHILDREN) { 2309088f3894Sahrens for (zap_cursor_init(&zc, dp->dp_meta_objset, 2310c1379625SJustin T. Gibbs dsl_dir_phys(dd)->dd_child_dir_zapobj); 2311b7661cccSmmusante zap_cursor_retrieve(&zc, attr) == 0; 23120b69c2f0Sahrens (void) zap_cursor_advance(&zc)) { 23133b2aab18SMatthew Ahrens ASSERT3U(attr->za_integer_length, ==, 23143b2aab18SMatthew Ahrens sizeof (uint64_t)); 23153b2aab18SMatthew Ahrens ASSERT3U(attr->za_num_integers, ==, 1); 2316fa9e4066Sahrens 2317486ae710SMatthew Ahrens child = kmem_asprintf("%s/%s", name, attr->za_name); 23183b2aab18SMatthew Ahrens dsl_pool_config_exit(dp, FTAG); 23193b2aab18SMatthew Ahrens err = dmu_objset_find_impl(spa, child, 23203b2aab18SMatthew Ahrens func, arg, flags); 23213b2aab18SMatthew Ahrens dsl_pool_config_enter(dp, FTAG); 2322486ae710SMatthew Ahrens strfree(child); 23233b2aab18SMatthew Ahrens if (err != 0) 23240b69c2f0Sahrens break; 23250b69c2f0Sahrens } 23260b69c2f0Sahrens zap_cursor_fini(&zc); 23271d452cf5Sahrens 23283b2aab18SMatthew Ahrens if (err != 0) { 23293b2aab18SMatthew Ahrens dsl_dir_rele(dd, FTAG); 23303b2aab18SMatthew Ahrens dsl_pool_config_exit(dp, FTAG); 2331b7661cccSmmusante kmem_free(attr, sizeof (zap_attribute_t)); 23320b69c2f0Sahrens return (err); 23330b69c2f0Sahrens } 2334fa9e4066Sahrens } 2335fa9e4066Sahrens 2336fa9e4066Sahrens /* 2337fa9e4066Sahrens * Iterate over all snapshots. 2338fa9e4066Sahrens */ 2339088f3894Sahrens if (flags & DS_FIND_SNAPSHOTS) { 2340088f3894Sahrens err = dsl_dataset_hold_obj(dp, thisobj, FTAG, &ds); 2341088f3894Sahrens 2342088f3894Sahrens if (err == 0) { 2343c1379625SJustin T. Gibbs uint64_t snapobj; 2344c1379625SJustin T. Gibbs 2345c1379625SJustin T. Gibbs snapobj = dsl_dataset_phys(ds)->ds_snapnames_zapobj; 2346088f3894Sahrens dsl_dataset_rele(ds, FTAG); 2347088f3894Sahrens 2348088f3894Sahrens for (zap_cursor_init(&zc, dp->dp_meta_objset, snapobj); 2349088f3894Sahrens zap_cursor_retrieve(&zc, attr) == 0; 2350088f3894Sahrens (void) zap_cursor_advance(&zc)) { 23513b2aab18SMatthew Ahrens ASSERT3U(attr->za_integer_length, ==, 2352088f3894Sahrens sizeof (uint64_t)); 23533b2aab18SMatthew Ahrens ASSERT3U(attr->za_num_integers, ==, 1); 2354088f3894Sahrens 2355486ae710SMatthew Ahrens child = kmem_asprintf("%s@%s", 2356486ae710SMatthew Ahrens name, attr->za_name); 23573b2aab18SMatthew Ahrens dsl_pool_config_exit(dp, FTAG); 23583b2aab18SMatthew Ahrens err = func(child, arg); 23593b2aab18SMatthew Ahrens dsl_pool_config_enter(dp, FTAG); 2360486ae710SMatthew Ahrens strfree(child); 23613b2aab18SMatthew Ahrens if (err != 0) 2362088f3894Sahrens break; 2363088f3894Sahrens } 2364088f3894Sahrens zap_cursor_fini(&zc); 2365fa9e4066Sahrens } 2366fa9e4066Sahrens } 2367fa9e4066Sahrens 23683b2aab18SMatthew Ahrens dsl_dir_rele(dd, FTAG); 2369b7661cccSmmusante kmem_free(attr, sizeof (zap_attribute_t)); 23703b2aab18SMatthew Ahrens dsl_pool_config_exit(dp, FTAG); 2371fa9e4066Sahrens 23723b2aab18SMatthew Ahrens if (err != 0) 23731d452cf5Sahrens return (err); 23741d452cf5Sahrens 23753b2aab18SMatthew Ahrens /* Apply to self. */ 23763b2aab18SMatthew Ahrens return (func(name, arg)); 2377fa9e4066Sahrens } 2378f18faf3fSek 23793b2aab18SMatthew Ahrens /* 23803b2aab18SMatthew Ahrens * See comment above dmu_objset_find_impl(). 23813b2aab18SMatthew Ahrens */ 23827f73c863SRich Morris int 23833b2aab18SMatthew Ahrens dmu_objset_find(char *name, int func(const char *, void *), void *arg, 23843b2aab18SMatthew Ahrens int flags) 23857f73c863SRich Morris { 23863b2aab18SMatthew Ahrens spa_t *spa; 23873b2aab18SMatthew Ahrens int error; 23887f73c863SRich Morris 23893b2aab18SMatthew Ahrens error = spa_open(name, &spa, FTAG); 23903b2aab18SMatthew Ahrens if (error != 0) 23913b2aab18SMatthew Ahrens return (error); 23923b2aab18SMatthew Ahrens error = dmu_objset_find_impl(spa, name, func, arg, flags); 23933b2aab18SMatthew Ahrens spa_close(spa, FTAG); 23943b2aab18SMatthew Ahrens return (error); 23957f73c863SRich Morris } 23967f73c863SRich Morris 2397f18faf3fSek void 2398f18faf3fSek dmu_objset_set_user(objset_t *os, void *user_ptr) 2399f18faf3fSek { 2400503ad85cSMatthew Ahrens ASSERT(MUTEX_HELD(&os->os_user_ptr_lock)); 2401503ad85cSMatthew Ahrens os->os_user_ptr = user_ptr; 2402f18faf3fSek } 2403f18faf3fSek 2404f18faf3fSek void * 2405f18faf3fSek dmu_objset_get_user(objset_t *os) 2406f18faf3fSek { 2407503ad85cSMatthew Ahrens ASSERT(MUTEX_HELD(&os->os_user_ptr_lock)); 2408503ad85cSMatthew Ahrens return (os->os_user_ptr); 2409f18faf3fSek } 24103b2aab18SMatthew Ahrens 24113b2aab18SMatthew Ahrens /* 24123b2aab18SMatthew Ahrens * Determine name of filesystem, given name of snapshot. 24139adfa60dSMatthew Ahrens * buf must be at least ZFS_MAX_DATASET_NAME_LEN bytes 24143b2aab18SMatthew Ahrens */ 24153b2aab18SMatthew Ahrens int 24163b2aab18SMatthew Ahrens dmu_fsname(const char *snapname, char *buf) 24173b2aab18SMatthew Ahrens { 24183b2aab18SMatthew Ahrens char *atp = strchr(snapname, '@'); 24193b2aab18SMatthew Ahrens if (atp == NULL) 2420be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 24219adfa60dSMatthew Ahrens if (atp - snapname >= ZFS_MAX_DATASET_NAME_LEN) 2422be6fd75aSMatthew Ahrens return (SET_ERROR(ENAMETOOLONG)); 24233b2aab18SMatthew Ahrens (void) strlcpy(buf, snapname, atp - snapname + 1); 24243b2aab18SMatthew Ahrens return (0); 24253b2aab18SMatthew Ahrens } 242661e255ceSMatthew Ahrens 242761e255ceSMatthew Ahrens /* 242861e255ceSMatthew Ahrens * Call when we think we're going to write/free space in open context to track 242961e255ceSMatthew Ahrens * the amount of dirty data in the open txg, which is also the amount 243061e255ceSMatthew Ahrens * of memory that can not be evicted until this txg syncs. 243161e255ceSMatthew Ahrens */ 243261e255ceSMatthew Ahrens void 243361e255ceSMatthew Ahrens dmu_objset_willuse_space(objset_t *os, int64_t space, dmu_tx_t *tx) 243461e255ceSMatthew Ahrens { 243561e255ceSMatthew Ahrens dsl_dataset_t *ds = os->os_dsl_dataset; 243661e255ceSMatthew Ahrens int64_t aspace = spa_get_worst_case_asize(os->os_spa, space); 243761e255ceSMatthew Ahrens 243861e255ceSMatthew Ahrens if (ds != NULL) { 243961e255ceSMatthew Ahrens dsl_dir_willuse_space(ds->ds_dir, aspace, tx); 244061e255ceSMatthew Ahrens dsl_pool_dirty_space(dmu_tx_pool(tx), space, tx); 244161e255ceSMatthew Ahrens } 244261e255ceSMatthew Ahrens } 2443