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