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