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 https://opensource.org/licenses/CDDL-1.0.
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, 2018 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  *   	dbuf_read_impl: db_mtx, dmu_zfetch()
90  *   	dmu_zfetch: zf_rwlock/r, zst_lock, dbuf_prefetch()
91  *   	dbuf_new_size: db_mtx
92  *   	dbuf_dirty: 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)
102  *
103  * dn_dbufs_mtx
104  *    must be held before:
105  *    	db_mtx, hash_mutexes
106  *    protects:
107  *    	dn_dbufs
108  *    	dn_evicted
109  *    held from:
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)
116  *
117  * hash_mutexes (global)
118  *   must be held before:
119  *   	db_mtx
120  *   protects dbuf_hash_table (global) and db_hash_next
121  *   held from:
122  *   	dbuf_find: db_mtx
123  *   	dbuf_hash_insert: db_mtx
124  *   	dbuf_hash_remove: db_mtx
125  *
126  * db_mtx (meta-leaf)
127  *   must be held before:
128  *   	dn_mtx, dn_dirty_mtx, dd_lock (leaf mutexes)
129  *   protects:
130  *   	db_state
131  * 	db_holds
132  * 	db_buf
133  * 	db_changed
134  * 	db_data_pending
135  * 	db_dirtied
136  * 	db_link
137  * 	db_dirty_node (??)
138  * 	db_dirtycnt
139  * 	db_d.*
140  * 	db.*
141  *   held from:
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)
147  * 	dbuf_*
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)
155  *
156  * dn_mtx (leaf)
157  *   protects:
158  *   	dn_dirty_dbufs
159  *   	dn_ranges
160  *   	phys accounting
161  * 	dn_allocated_txg
162  * 	dn_free_txg
163  * 	dn_assigned_txg
164  * 	dn_dirty_txg
165  * 	dd_assigned_tx
166  * 	dn_notxholds
167  *	dn_nodnholds
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  *
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_sendstatus {
241 	list_node_t dss_link;
242 	int dss_outfd;
243 	proc_t *dss_proc;
244 	offset_t *dss_off;
245 	uint64_t dss_blocks; /* blocks visited during the sending process */
246 } dmu_sendstatus_t;
247 
248 void dmu_object_zapify(objset_t *, uint64_t, dmu_object_type_t, dmu_tx_t *);
249 void dmu_object_free_zapified(objset_t *, uint64_t, dmu_tx_t *);
250 
251 #ifdef	__cplusplus
252 }
253 #endif
254 
255 #endif	/* _SYS_DMU_IMPL_H */
256