1 /* 2 * CDDL HEADER START 3 * 4 * The contents of this file are subject to the terms of the 5 * Common Development and Distribution License (the "License"). 6 * You may not use this file except in compliance with the License. 7 * 8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 9 * or http://www.opensolaris.org/os/licensing. 10 * See the License for the specific language governing permissions 11 * and limitations under the License. 12 * 13 * When distributing Covered Code, include this CDDL HEADER in each 14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 15 * If applicable, add the following below this CDDL HEADER, with the 16 * fields enclosed by brackets "[]" replaced with your own identifying 17 * information: Portions Copyright [yyyy] [name of copyright owner] 18 * 19 * CDDL HEADER END 20 */ 21 /* 22 * Copyright 2010 Sun Microsystems, Inc. All rights reserved. 23 * Use is subject to license terms. 24 */ 25 /* 26 * Copyright (c) 2012, Joyent, Inc. All rights reserved. 27 * Copyright (c) 2012, Martin Matuska <mm@FreeBSD.org>. All rights reserved. 28 * Copyright (c) 2013, 2015 by Delphix. All rights reserved. 29 */ 30 31 #ifndef _SYS_DMU_IMPL_H 32 #define _SYS_DMU_IMPL_H 33 34 #include <sys/txg_impl.h> 35 #include <sys/zio.h> 36 #include <sys/dnode.h> 37 #include <sys/kstat.h> 38 #include <sys/zfs_context.h> 39 #include <sys/zfs_ioctl.h> 40 41 #ifdef __cplusplus 42 extern "C" { 43 #endif 44 45 /* 46 * This is the locking strategy for the DMU. Numbers in parenthesis are 47 * cases that use that lock order, referenced below: 48 * 49 * ARC is self-contained 50 * bplist is self-contained 51 * refcount is self-contained 52 * txg is self-contained (hopefully!) 53 * zst_lock 54 * zf_rwlock 55 * 56 * XXX try to improve evicting path? 57 * 58 * dp_config_rwlock > os_obj_lock > dn_struct_rwlock > 59 * dn_dbufs_mtx > hash_mutexes > db_mtx > dd_lock > leafs 60 * 61 * dp_config_rwlock 62 * must be held before: everything 63 * protects dd namespace changes 64 * protects property changes globally 65 * held from: 66 * dsl_dir_open/r: 67 * dsl_dir_create_sync/w: 68 * dsl_dir_sync_destroy/w: 69 * dsl_dir_rename_sync/w: 70 * dsl_prop_changed_notify/r: 71 * 72 * os_obj_lock 73 * must be held before: 74 * everything except dp_config_rwlock 75 * protects os_obj_next 76 * held from: 77 * dmu_object_alloc: dn_dbufs_mtx, db_mtx, hash_mutexes, dn_struct_rwlock 78 * 79 * dn_struct_rwlock 80 * must be held before: 81 * everything except dp_config_rwlock and os_obj_lock 82 * protects structure of dnode (eg. nlevels) 83 * db_blkptr can change when syncing out change to nlevels 84 * dn_maxblkid 85 * dn_nlevels 86 * dn_*blksz* 87 * phys nlevels, maxblkid, physical blkptr_t's (?) 88 * held from: 89 * callers of dbuf_read_impl, dbuf_hold[_impl], dbuf_prefetch 90 * dmu_object_info_from_dnode: dn_dirty_mtx (dn_datablksz) 91 * dmu_tx_count_free: 92 * dbuf_read_impl: db_mtx, dmu_zfetch() 93 * dmu_zfetch: zf_rwlock/r, zst_lock, dbuf_prefetch() 94 * dbuf_new_size: db_mtx 95 * dbuf_dirty: db_mtx 96 * dbuf_findbp: (callers, phys? - the real need) 97 * dbuf_create: dn_dbufs_mtx, hash_mutexes, db_mtx (phys?) 98 * dbuf_prefetch: dn_dirty_mtx, hash_mutexes, db_mtx, dn_dbufs_mtx 99 * dbuf_hold_impl: hash_mutexes, db_mtx, dn_dbufs_mtx, dbuf_findbp() 100 * dnode_sync/w (increase_indirection): db_mtx (phys) 101 * dnode_set_blksz/w: dn_dbufs_mtx (dn_*blksz*) 102 * dnode_new_blkid/w: (dn_maxblkid) 103 * dnode_free_range/w: dn_dirty_mtx (dn_maxblkid) 104 * dnode_next_offset: (phys) 105 * 106 * dn_dbufs_mtx 107 * must be held before: 108 * db_mtx, hash_mutexes 109 * protects: 110 * dn_dbufs 111 * dn_evicted 112 * held from: 113 * dmu_evict_user: db_mtx (dn_dbufs) 114 * dbuf_free_range: db_mtx (dn_dbufs) 115 * dbuf_remove_ref: db_mtx, callees: 116 * dbuf_hash_remove: hash_mutexes, db_mtx 117 * dbuf_create: hash_mutexes, db_mtx (dn_dbufs) 118 * dnode_set_blksz: (dn_dbufs) 119 * 120 * hash_mutexes (global) 121 * must be held before: 122 * db_mtx 123 * protects dbuf_hash_table (global) and db_hash_next 124 * held from: 125 * dbuf_find: db_mtx 126 * dbuf_hash_insert: db_mtx 127 * dbuf_hash_remove: db_mtx 128 * 129 * db_mtx (meta-leaf) 130 * must be held before: 131 * dn_mtx, dn_dirty_mtx, dd_lock (leaf mutexes) 132 * protects: 133 * db_state 134 * db_holds 135 * db_buf 136 * db_changed 137 * db_data_pending 138 * db_dirtied 139 * db_link 140 * db_dirty_node (??) 141 * db_dirtycnt 142 * db_d.* 143 * db.* 144 * held from: 145 * dbuf_dirty: dn_mtx, dn_dirty_mtx 146 * dbuf_dirty->dsl_dir_willuse_space: dd_lock 147 * dbuf_dirty->dbuf_new_block->dsl_dataset_block_freeable: dd_lock 148 * dbuf_undirty: dn_dirty_mtx (db_d) 149 * dbuf_write_done: dn_dirty_mtx (db_state) 150 * dbuf_* 151 * dmu_buf_update_user: none (db_d) 152 * dmu_evict_user: none (db_d) (maybe can eliminate) 153 * dbuf_find: none (db_holds) 154 * dbuf_hash_insert: none (db_holds) 155 * dmu_buf_read_array_impl: none (db_state, db_changed) 156 * dmu_sync: none (db_dirty_node, db_d) 157 * dnode_reallocate: none (db) 158 * 159 * dn_mtx (leaf) 160 * protects: 161 * dn_dirty_dbufs 162 * dn_ranges 163 * phys accounting 164 * dn_allocated_txg 165 * dn_free_txg 166 * dn_assigned_txg 167 * dd_assigned_tx 168 * dn_notxholds 169 * dn_dirtyctx 170 * dn_dirtyctx_firstset 171 * (dn_phys copy fields?) 172 * (dn_phys contents?) 173 * held from: 174 * dnode_* 175 * dbuf_dirty: none 176 * dbuf_sync: none (phys accounting) 177 * dbuf_undirty: none (dn_ranges, dn_dirty_dbufs) 178 * dbuf_write_done: none (phys accounting) 179 * dmu_object_info_from_dnode: none (accounting) 180 * dmu_tx_commit: none 181 * dmu_tx_hold_object_impl: none 182 * dmu_tx_try_assign: dn_notxholds(cv) 183 * dmu_tx_unassign: none 184 * 185 * dd_lock 186 * must be held before: 187 * ds_lock 188 * ancestors' dd_lock 189 * protects: 190 * dd_prop_cbs 191 * dd_sync_* 192 * dd_used_bytes 193 * dd_tempreserved 194 * dd_space_towrite 195 * dd_myname 196 * dd_phys accounting? 197 * held from: 198 * dsl_dir_* 199 * dsl_prop_changed_notify: none (dd_prop_cbs) 200 * dsl_prop_register: none (dd_prop_cbs) 201 * dsl_prop_unregister: none (dd_prop_cbs) 202 * dsl_dataset_block_freeable: none (dd_sync_*) 203 * 204 * os_lock (leaf) 205 * protects: 206 * os_dirty_dnodes 207 * os_free_dnodes 208 * os_dnodes 209 * os_downgraded_dbufs 210 * dn_dirtyblksz 211 * dn_dirty_link 212 * held from: 213 * dnode_create: none (os_dnodes) 214 * dnode_destroy: none (os_dnodes) 215 * dnode_setdirty: none (dn_dirtyblksz, os_*_dnodes) 216 * dnode_free: none (dn_dirtyblksz, os_*_dnodes) 217 * 218 * ds_lock 219 * protects: 220 * ds_objset 221 * ds_open_refcount 222 * ds_snapname 223 * ds_phys accounting 224 * ds_phys userrefs zapobj 225 * ds_reserved 226 * held from: 227 * dsl_dataset_* 228 * 229 * dr_mtx (leaf) 230 * protects: 231 * dr_children 232 * held from: 233 * dbuf_dirty 234 * dbuf_undirty 235 * dbuf_sync_indirect 236 * dnode_new_blkid 237 */ 238 239 struct objset; 240 struct dmu_pool; 241 242 typedef struct dmu_xuio { 243 int next; 244 int cnt; 245 struct arc_buf **bufs; 246 iovec_t *iovp; 247 } dmu_xuio_t; 248 249 typedef struct xuio_stats { 250 /* loaned yet not returned arc_buf */ 251 kstat_named_t xuiostat_onloan_rbuf; 252 kstat_named_t xuiostat_onloan_wbuf; 253 /* whether a copy is made when loaning out a read buffer */ 254 kstat_named_t xuiostat_rbuf_copied; 255 kstat_named_t xuiostat_rbuf_nocopy; 256 /* whether a copy is made when assigning a write buffer */ 257 kstat_named_t xuiostat_wbuf_copied; 258 kstat_named_t xuiostat_wbuf_nocopy; 259 } xuio_stats_t; 260 261 static xuio_stats_t xuio_stats = { 262 { "onloan_read_buf", KSTAT_DATA_UINT64 }, 263 { "onloan_write_buf", KSTAT_DATA_UINT64 }, 264 { "read_buf_copied", KSTAT_DATA_UINT64 }, 265 { "read_buf_nocopy", KSTAT_DATA_UINT64 }, 266 { "write_buf_copied", KSTAT_DATA_UINT64 }, 267 { "write_buf_nocopy", KSTAT_DATA_UINT64 } 268 }; 269 270 #define XUIOSTAT_INCR(stat, val) \ 271 atomic_add_64(&xuio_stats.stat.value.ui64, (val)) 272 #define XUIOSTAT_BUMP(stat) XUIOSTAT_INCR(stat, 1) 273 274 /* 275 * The list of data whose inclusion in a send stream can be pending from 276 * one call to backup_cb to another. Multiple calls to dump_free() and 277 * dump_freeobjects() can be aggregated into a single DRR_FREE or 278 * DRR_FREEOBJECTS replay record. 279 */ 280 typedef enum { 281 PENDING_NONE, 282 PENDING_FREE, 283 PENDING_FREEOBJECTS 284 } dmu_pendop_t; 285 286 typedef struct dmu_sendarg { 287 list_node_t dsa_link; 288 dmu_replay_record_t *dsa_drr; 289 kthread_t *dsa_td; 290 struct file *dsa_fp; 291 int dsa_outfd; 292 struct proc *dsa_proc; 293 offset_t *dsa_off; 294 objset_t *dsa_os; 295 zio_cksum_t dsa_zc; 296 uint64_t dsa_toguid; 297 int dsa_err; 298 dmu_pendop_t dsa_pending_op; 299 uint64_t dsa_featureflags; 300 uint64_t dsa_last_data_object; 301 uint64_t dsa_last_data_offset; 302 uint64_t dsa_resume_object; 303 uint64_t dsa_resume_offset; 304 boolean_t dsa_sent_begin; 305 boolean_t dsa_sent_end; 306 } dmu_sendarg_t; 307 308 void dmu_object_zapify(objset_t *, uint64_t, dmu_object_type_t, dmu_tx_t *); 309 void dmu_object_free_zapified(objset_t *, uint64_t, dmu_tx_t *); 310 int dmu_buf_hold_noread(objset_t *, uint64_t, uint64_t, 311 void *, dmu_buf_t **); 312 313 #ifdef __cplusplus 314 } 315 #endif 316 317 #endif /* _SYS_DMU_IMPL_H */ 318