1168404Spjd/* 2168404Spjd * CDDL HEADER START 3168404Spjd * 4168404Spjd * The contents of this file are subject to the terms of the 5168404Spjd * Common Development and Distribution License (the "License"). 6168404Spjd * You may not use this file except in compliance with the License. 7168404Spjd * 8168404Spjd * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 9168404Spjd * or http://www.opensolaris.org/os/licensing. 10168404Spjd * See the License for the specific language governing permissions 11168404Spjd * and limitations under the License. 12168404Spjd * 13168404Spjd * When distributing Covered Code, include this CDDL HEADER in each 14168404Spjd * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 15168404Spjd * If applicable, add the following below this CDDL HEADER, with the 16168404Spjd * fields enclosed by brackets "[]" replaced with your own identifying 17168404Spjd * information: Portions Copyright [yyyy] [name of copyright owner] 18168404Spjd * 19168404Spjd * CDDL HEADER END 20168404Spjd */ 21168404Spjd/* 22219089Spjd * Copyright 2010 Sun Microsystems, Inc. All rights reserved. 23168404Spjd * Use is subject to license terms. 24253821Sdelphij */ 25253821Sdelphij/* 26235222Smm * Copyright (c) 2012, Joyent, Inc. All rights reserved. 27235222Smm * Copyright (c) 2012, Martin Matuska <mm@FreeBSD.org>. All rights reserved. 28253821Sdelphij * Copyright (c) 2013 by Delphix. All rights reserved. 29168404Spjd */ 30168404Spjd 31168404Spjd#ifndef _SYS_DMU_IMPL_H 32168404Spjd#define _SYS_DMU_IMPL_H 33168404Spjd 34168404Spjd#include <sys/txg_impl.h> 35168404Spjd#include <sys/zio.h> 36168404Spjd#include <sys/dnode.h> 37219089Spjd#include <sys/kstat.h> 38168404Spjd#include <sys/zfs_context.h> 39235222Smm#include <sys/zfs_ioctl.h> 40168404Spjd 41168404Spjd#ifdef __cplusplus 42168404Spjdextern "C" { 43168404Spjd#endif 44168404Spjd 45168404Spjd/* 46168404Spjd * This is the locking strategy for the DMU. Numbers in parenthesis are 47168404Spjd * cases that use that lock order, referenced below: 48168404Spjd * 49168404Spjd * ARC is self-contained 50168404Spjd * bplist is self-contained 51168404Spjd * refcount is self-contained 52168404Spjd * txg is self-contained (hopefully!) 53168404Spjd * zst_lock 54168404Spjd * zf_rwlock 55168404Spjd * 56168404Spjd * XXX try to improve evicting path? 57168404Spjd * 58168404Spjd * dp_config_rwlock > os_obj_lock > dn_struct_rwlock > 59185029Spjd * dn_dbufs_mtx > hash_mutexes > db_mtx > dd_lock > leafs 60168404Spjd * 61168404Spjd * dp_config_rwlock 62168404Spjd * must be held before: everything 63168404Spjd * protects dd namespace changes 64168404Spjd * protects property changes globally 65168404Spjd * held from: 66168404Spjd * dsl_dir_open/r: 67168404Spjd * dsl_dir_create_sync/w: 68168404Spjd * dsl_dir_sync_destroy/w: 69168404Spjd * dsl_dir_rename_sync/w: 70168404Spjd * dsl_prop_changed_notify/r: 71168404Spjd * 72168404Spjd * os_obj_lock 73168404Spjd * must be held before: 74168404Spjd * everything except dp_config_rwlock 75168404Spjd * protects os_obj_next 76168404Spjd * held from: 77168404Spjd * dmu_object_alloc: dn_dbufs_mtx, db_mtx, hash_mutexes, dn_struct_rwlock 78168404Spjd * 79168404Spjd * dn_struct_rwlock 80168404Spjd * must be held before: 81168404Spjd * everything except dp_config_rwlock and os_obj_lock 82168404Spjd * protects structure of dnode (eg. nlevels) 83168404Spjd * db_blkptr can change when syncing out change to nlevels 84168404Spjd * dn_maxblkid 85168404Spjd * dn_nlevels 86168404Spjd * dn_*blksz* 87168404Spjd * phys nlevels, maxblkid, physical blkptr_t's (?) 88168404Spjd * held from: 89168404Spjd * callers of dbuf_read_impl, dbuf_hold[_impl], dbuf_prefetch 90168404Spjd * dmu_object_info_from_dnode: dn_dirty_mtx (dn_datablksz) 91168404Spjd * dmu_tx_count_free: 92168404Spjd * dbuf_read_impl: db_mtx, dmu_zfetch() 93168404Spjd * dmu_zfetch: zf_rwlock/r, zst_lock, dbuf_prefetch() 94168404Spjd * dbuf_new_size: db_mtx 95168404Spjd * dbuf_dirty: db_mtx 96168404Spjd * dbuf_findbp: (callers, phys? - the real need) 97168404Spjd * dbuf_create: dn_dbufs_mtx, hash_mutexes, db_mtx (phys?) 98168404Spjd * dbuf_prefetch: dn_dirty_mtx, hash_mutexes, db_mtx, dn_dbufs_mtx 99168404Spjd * dbuf_hold_impl: hash_mutexes, db_mtx, dn_dbufs_mtx, dbuf_findbp() 100168404Spjd * dnode_sync/w (increase_indirection): db_mtx (phys) 101168404Spjd * dnode_set_blksz/w: dn_dbufs_mtx (dn_*blksz*) 102168404Spjd * dnode_new_blkid/w: (dn_maxblkid) 103168404Spjd * dnode_free_range/w: dn_dirty_mtx (dn_maxblkid) 104168404Spjd * dnode_next_offset: (phys) 105168404Spjd * 106168404Spjd * dn_dbufs_mtx 107168404Spjd * must be held before: 108168404Spjd * db_mtx, hash_mutexes 109168404Spjd * protects: 110168404Spjd * dn_dbufs 111168404Spjd * dn_evicted 112168404Spjd * held from: 113168404Spjd * dmu_evict_user: db_mtx (dn_dbufs) 114168404Spjd * dbuf_free_range: db_mtx (dn_dbufs) 115168404Spjd * dbuf_remove_ref: db_mtx, callees: 116168404Spjd * dbuf_hash_remove: hash_mutexes, db_mtx 117168404Spjd * dbuf_create: hash_mutexes, db_mtx (dn_dbufs) 118168404Spjd * dnode_set_blksz: (dn_dbufs) 119168404Spjd * 120168404Spjd * hash_mutexes (global) 121168404Spjd * must be held before: 122168404Spjd * db_mtx 123168404Spjd * protects dbuf_hash_table (global) and db_hash_next 124168404Spjd * held from: 125168404Spjd * dbuf_find: db_mtx 126168404Spjd * dbuf_hash_insert: db_mtx 127168404Spjd * dbuf_hash_remove: db_mtx 128168404Spjd * 129168404Spjd * db_mtx (meta-leaf) 130168404Spjd * must be held before: 131168404Spjd * dn_mtx, dn_dirty_mtx, dd_lock (leaf mutexes) 132168404Spjd * protects: 133168404Spjd * db_state 134168404Spjd * db_holds 135168404Spjd * db_buf 136168404Spjd * db_changed 137168404Spjd * db_data_pending 138168404Spjd * db_dirtied 139168404Spjd * db_link 140168404Spjd * db_dirty_node (??) 141168404Spjd * db_dirtycnt 142168404Spjd * db_d.* 143168404Spjd * db.* 144168404Spjd * held from: 145168404Spjd * dbuf_dirty: dn_mtx, dn_dirty_mtx 146168404Spjd * dbuf_dirty->dsl_dir_willuse_space: dd_lock 147168404Spjd * dbuf_dirty->dbuf_new_block->dsl_dataset_block_freeable: dd_lock 148168404Spjd * dbuf_undirty: dn_dirty_mtx (db_d) 149168404Spjd * dbuf_write_done: dn_dirty_mtx (db_state) 150168404Spjd * dbuf_* 151168404Spjd * dmu_buf_update_user: none (db_d) 152168404Spjd * dmu_evict_user: none (db_d) (maybe can eliminate) 153168404Spjd * dbuf_find: none (db_holds) 154168404Spjd * dbuf_hash_insert: none (db_holds) 155168404Spjd * dmu_buf_read_array_impl: none (db_state, db_changed) 156168404Spjd * dmu_sync: none (db_dirty_node, db_d) 157168404Spjd * dnode_reallocate: none (db) 158168404Spjd * 159168404Spjd * dn_mtx (leaf) 160168404Spjd * protects: 161168404Spjd * dn_dirty_dbufs 162168404Spjd * dn_ranges 163168404Spjd * phys accounting 164168404Spjd * dn_allocated_txg 165168404Spjd * dn_free_txg 166168404Spjd * dn_assigned_txg 167168404Spjd * dd_assigned_tx 168168404Spjd * dn_notxholds 169168404Spjd * dn_dirtyctx 170168404Spjd * dn_dirtyctx_firstset 171168404Spjd * (dn_phys copy fields?) 172168404Spjd * (dn_phys contents?) 173168404Spjd * held from: 174168404Spjd * dnode_* 175168404Spjd * dbuf_dirty: none 176168404Spjd * dbuf_sync: none (phys accounting) 177168404Spjd * dbuf_undirty: none (dn_ranges, dn_dirty_dbufs) 178168404Spjd * dbuf_write_done: none (phys accounting) 179168404Spjd * dmu_object_info_from_dnode: none (accounting) 180168404Spjd * dmu_tx_commit: none 181168404Spjd * dmu_tx_hold_object_impl: none 182168404Spjd * dmu_tx_try_assign: dn_notxholds(cv) 183168404Spjd * dmu_tx_unassign: none 184168404Spjd * 185185029Spjd * dd_lock 186185029Spjd * must be held before: 187185029Spjd * ds_lock 188185029Spjd * ancestors' dd_lock 189168404Spjd * protects: 190168404Spjd * dd_prop_cbs 191168404Spjd * dd_sync_* 192168404Spjd * dd_used_bytes 193168404Spjd * dd_tempreserved 194168404Spjd * dd_space_towrite 195168404Spjd * dd_myname 196168404Spjd * dd_phys accounting? 197168404Spjd * held from: 198168404Spjd * dsl_dir_* 199168404Spjd * dsl_prop_changed_notify: none (dd_prop_cbs) 200168404Spjd * dsl_prop_register: none (dd_prop_cbs) 201168404Spjd * dsl_prop_unregister: none (dd_prop_cbs) 202168404Spjd * dsl_dataset_block_freeable: none (dd_sync_*) 203168404Spjd * 204168404Spjd * os_lock (leaf) 205168404Spjd * protects: 206168404Spjd * os_dirty_dnodes 207168404Spjd * os_free_dnodes 208168404Spjd * os_dnodes 209168404Spjd * os_downgraded_dbufs 210168404Spjd * dn_dirtyblksz 211168404Spjd * dn_dirty_link 212168404Spjd * held from: 213168404Spjd * dnode_create: none (os_dnodes) 214168404Spjd * dnode_destroy: none (os_dnodes) 215168404Spjd * dnode_setdirty: none (dn_dirtyblksz, os_*_dnodes) 216168404Spjd * dnode_free: none (dn_dirtyblksz, os_*_dnodes) 217168404Spjd * 218185029Spjd * ds_lock 219168404Spjd * protects: 220219089Spjd * ds_objset 221168404Spjd * ds_open_refcount 222168404Spjd * ds_snapname 223168404Spjd * ds_phys accounting 224219089Spjd * ds_phys userrefs zapobj 225185029Spjd * ds_reserved 226168404Spjd * held from: 227168404Spjd * dsl_dataset_* 228168404Spjd * 229168404Spjd * dr_mtx (leaf) 230168404Spjd * protects: 231168404Spjd * dr_children 232168404Spjd * held from: 233168404Spjd * dbuf_dirty 234168404Spjd * dbuf_undirty 235168404Spjd * dbuf_sync_indirect 236168404Spjd * dnode_new_blkid 237168404Spjd */ 238168404Spjd 239168404Spjdstruct objset; 240168404Spjdstruct dmu_pool; 241168404Spjd 242219089Spjdtypedef struct dmu_xuio { 243219089Spjd int next; 244219089Spjd int cnt; 245219089Spjd struct arc_buf **bufs; 246219089Spjd iovec_t *iovp; 247219089Spjd} dmu_xuio_t; 248219089Spjd 249219089Spjdtypedef struct xuio_stats { 250219089Spjd /* loaned yet not returned arc_buf */ 251219089Spjd kstat_named_t xuiostat_onloan_rbuf; 252219089Spjd kstat_named_t xuiostat_onloan_wbuf; 253219089Spjd /* whether a copy is made when loaning out a read buffer */ 254219089Spjd kstat_named_t xuiostat_rbuf_copied; 255219089Spjd kstat_named_t xuiostat_rbuf_nocopy; 256219089Spjd /* whether a copy is made when assigning a write buffer */ 257219089Spjd kstat_named_t xuiostat_wbuf_copied; 258219089Spjd kstat_named_t xuiostat_wbuf_nocopy; 259219089Spjd} xuio_stats_t; 260219089Spjd 261219089Spjdstatic xuio_stats_t xuio_stats = { 262219089Spjd { "onloan_read_buf", KSTAT_DATA_UINT64 }, 263219089Spjd { "onloan_write_buf", KSTAT_DATA_UINT64 }, 264219089Spjd { "read_buf_copied", KSTAT_DATA_UINT64 }, 265219089Spjd { "read_buf_nocopy", KSTAT_DATA_UINT64 }, 266219089Spjd { "write_buf_copied", KSTAT_DATA_UINT64 }, 267219089Spjd { "write_buf_nocopy", KSTAT_DATA_UINT64 } 268219089Spjd}; 269219089Spjd 270219089Spjd#define XUIOSTAT_INCR(stat, val) \ 271219089Spjd atomic_add_64(&xuio_stats.stat.value.ui64, (val)) 272219089Spjd#define XUIOSTAT_BUMP(stat) XUIOSTAT_INCR(stat, 1) 273219089Spjd 274235222Smm/* 275235222Smm * The list of data whose inclusion in a send stream can be pending from 276235222Smm * one call to backup_cb to another. Multiple calls to dump_free() and 277235222Smm * dump_freeobjects() can be aggregated into a single DRR_FREE or 278235222Smm * DRR_FREEOBJECTS replay record. 279235222Smm */ 280235222Smmtypedef enum { 281235222Smm PENDING_NONE, 282235222Smm PENDING_FREE, 283235222Smm PENDING_FREEOBJECTS 284235222Smm} dmu_pendop_t; 285219089Spjd 286235222Smmtypedef struct dmu_sendarg { 287235222Smm list_node_t dsa_link; 288235222Smm dmu_replay_record_t *dsa_drr; 289235222Smm kthread_t *dsa_td; 290235222Smm struct file *dsa_fp; 291235222Smm int dsa_outfd; 292235222Smm struct proc *dsa_proc; 293235222Smm offset_t *dsa_off; 294235222Smm objset_t *dsa_os; 295235222Smm zio_cksum_t dsa_zc; 296235222Smm uint64_t dsa_toguid; 297235222Smm int dsa_err; 298235222Smm dmu_pendop_t dsa_pending_op; 299253821Sdelphij boolean_t dsa_incremental; 300268649Sdelphij uint64_t dsa_featureflags; 301253821Sdelphij uint64_t dsa_last_data_object; 302253821Sdelphij uint64_t dsa_last_data_offset; 303235222Smm} dmu_sendarg_t; 304235222Smm 305263390Sdelphijvoid dmu_object_zapify(objset_t *, uint64_t, dmu_object_type_t, dmu_tx_t *); 306263390Sdelphijvoid dmu_object_free_zapified(objset_t *, uint64_t, dmu_tx_t *); 307268649Sdelphijint dmu_buf_hold_noread(objset_t *, uint64_t, uint64_t, 308268649Sdelphij void *, dmu_buf_t **); 309235222Smm 310168404Spjd#ifdef __cplusplus 311168404Spjd} 312168404Spjd#endif 313168404Spjd 314168404Spjd#endif /* _SYS_DMU_IMPL_H */ 315