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