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