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.
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.
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]
22 * Copyright 2010 Sun Microsystems, Inc. All rights reserved.
23 * Use is subject to license terms.
24 * Copyright (c) 2012, Joyent, Inc. All rights reserved.
25 * Copyright (c) 2012, Martin Matuska <mm@FreeBSD.org>. All rights reserved.
28 #ifndef _SYS_DMU_IMPL_H
29 #define _SYS_DMU_IMPL_H
31 #include <sys/txg_impl.h>
33 #include <sys/dnode.h>
34 #include <sys/kstat.h>
35 #include <sys/zfs_context.h>
36 #include <sys/zfs_ioctl.h>
43 * This is the locking strategy for the DMU. Numbers in parenthesis are
44 * cases that use that lock order, referenced below:
46 * ARC is self-contained
47 * bplist is self-contained
48 * refcount is self-contained
49 * txg is self-contained (hopefully!)
53 * XXX try to improve evicting path?
55 * dp_config_rwlock > os_obj_lock > dn_struct_rwlock >
56 * dn_dbufs_mtx > hash_mutexes > db_mtx > dd_lock > leafs
59 * must be held before: everything
60 * protects dd namespace changes
61 * protects property changes globally
64 * dsl_dir_create_sync/w:
65 * dsl_dir_sync_destroy/w:
66 * dsl_dir_rename_sync/w:
67 * dsl_prop_changed_notify/r:
70 * must be held before:
71 * everything except dp_config_rwlock
72 * protects os_obj_next
74 * dmu_object_alloc: dn_dbufs_mtx, db_mtx, hash_mutexes, dn_struct_rwlock
77 * must be held before:
78 * everything except dp_config_rwlock and os_obj_lock
79 * protects structure of dnode (eg. nlevels)
80 * db_blkptr can change when syncing out change to nlevels
84 * phys nlevels, maxblkid, physical blkptr_t's (?)
86 * callers of dbuf_read_impl, dbuf_hold[_impl], dbuf_prefetch
87 * dmu_object_info_from_dnode: dn_dirty_mtx (dn_datablksz)
89 * dbuf_read_impl: db_mtx, dmu_zfetch()
90 * dmu_zfetch: zf_rwlock/r, zst_lock, dbuf_prefetch()
91 * dbuf_new_size: db_mtx
93 * dbuf_findbp: (callers, phys? - the real need)
94 * dbuf_create: dn_dbufs_mtx, hash_mutexes, db_mtx (phys?)
95 * dbuf_prefetch: dn_dirty_mtx, hash_mutexes, db_mtx, dn_dbufs_mtx
96 * dbuf_hold_impl: hash_mutexes, db_mtx, dn_dbufs_mtx, dbuf_findbp()
97 * dnode_sync/w (increase_indirection): db_mtx (phys)
98 * dnode_set_blksz/w: dn_dbufs_mtx (dn_*blksz*)
99 * dnode_new_blkid/w: (dn_maxblkid)
100 * dnode_free_range/w: dn_dirty_mtx (dn_maxblkid)
101 * dnode_next_offset: (phys)
104 * must be held before:
105 * db_mtx, hash_mutexes
110 * dmu_evict_user: db_mtx (dn_dbufs)
111 * dbuf_free_range: db_mtx (dn_dbufs)
112 * dbuf_remove_ref: db_mtx, callees:
113 * dbuf_hash_remove: hash_mutexes, db_mtx
114 * dbuf_create: hash_mutexes, db_mtx (dn_dbufs)
115 * dnode_set_blksz: (dn_dbufs)
117 * hash_mutexes (global)
118 * must be held before:
120 * protects dbuf_hash_table (global) and db_hash_next
123 * dbuf_hash_insert: db_mtx
124 * dbuf_hash_remove: db_mtx
127 * must be held before:
128 * dn_mtx, dn_dirty_mtx, dd_lock (leaf mutexes)
142 * dbuf_dirty: dn_mtx, dn_dirty_mtx
143 * dbuf_dirty->dsl_dir_willuse_space: dd_lock
144 * dbuf_dirty->dbuf_new_block->dsl_dataset_block_freeable: dd_lock
145 * dbuf_undirty: dn_dirty_mtx (db_d)
146 * dbuf_write_done: dn_dirty_mtx (db_state)
148 * dmu_buf_update_user: none (db_d)
149 * dmu_evict_user: none (db_d) (maybe can eliminate)
150 * dbuf_find: none (db_holds)
151 * dbuf_hash_insert: none (db_holds)
152 * dmu_buf_read_array_impl: none (db_state, db_changed)
153 * dmu_sync: none (db_dirty_node, db_d)
154 * dnode_reallocate: none (db)
167 * dn_dirtyctx_firstset
168 * (dn_phys copy fields?)
169 * (dn_phys contents?)
173 * dbuf_sync: none (phys accounting)
174 * dbuf_undirty: none (dn_ranges, dn_dirty_dbufs)
175 * dbuf_write_done: none (phys accounting)
176 * dmu_object_info_from_dnode: none (accounting)
177 * dmu_tx_commit: none
178 * dmu_tx_hold_object_impl: none
179 * dmu_tx_try_assign: dn_notxholds(cv)
180 * dmu_tx_unassign: none
183 * must be held before:
193 * dd_phys accounting?
196 * dsl_prop_changed_notify: none (dd_prop_cbs)
197 * dsl_prop_register: none (dd_prop_cbs)
198 * dsl_prop_unregister: none (dd_prop_cbs)
199 * dsl_dataset_block_freeable: none (dd_sync_*)
206 * os_downgraded_dbufs
210 * dnode_create: none (os_dnodes)
211 * dnode_destroy: none (os_dnodes)
212 * dnode_setdirty: none (dn_dirtyblksz, os_*_dnodes)
213 * dnode_free: none (dn_dirtyblksz, os_*_dnodes)
221 * ds_phys userrefs zapobj
239 typedef struct dmu_xuio {
242 struct arc_buf **bufs;
246 typedef struct xuio_stats {
247 /* loaned yet not returned arc_buf */
248 kstat_named_t xuiostat_onloan_rbuf;
249 kstat_named_t xuiostat_onloan_wbuf;
250 /* whether a copy is made when loaning out a read buffer */
251 kstat_named_t xuiostat_rbuf_copied;
252 kstat_named_t xuiostat_rbuf_nocopy;
253 /* whether a copy is made when assigning a write buffer */
254 kstat_named_t xuiostat_wbuf_copied;
255 kstat_named_t xuiostat_wbuf_nocopy;
258 static xuio_stats_t xuio_stats = {
259 { "onloan_read_buf", KSTAT_DATA_UINT64 },
260 { "onloan_write_buf", KSTAT_DATA_UINT64 },
261 { "read_buf_copied", KSTAT_DATA_UINT64 },
262 { "read_buf_nocopy", KSTAT_DATA_UINT64 },
263 { "write_buf_copied", KSTAT_DATA_UINT64 },
264 { "write_buf_nocopy", KSTAT_DATA_UINT64 }
267 #define XUIOSTAT_INCR(stat, val) \
268 atomic_add_64(&xuio_stats.stat.value.ui64, (val))
269 #define XUIOSTAT_BUMP(stat) XUIOSTAT_INCR(stat, 1)
272 * The list of data whose inclusion in a send stream can be pending from
273 * one call to backup_cb to another. Multiple calls to dump_free() and
274 * dump_freeobjects() can be aggregated into a single DRR_FREE or
275 * DRR_FREEOBJECTS replay record.
283 typedef struct dmu_sendarg {
284 list_node_t dsa_link;
285 dmu_replay_record_t *dsa_drr;
289 struct proc *dsa_proc;
295 dmu_pendop_t dsa_pending_op;
303 #endif /* _SYS_DMU_IMPL_H */