xref: /titanic_44/usr/src/uts/common/fs/zfs/dmu_object.c (revision 349b53dd4e695e3d833b5380540385145b2d3ae8)
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 2009 Sun Microsystems, Inc.  All rights reserved.
23  * Use is subject to license terms.
24  */
25 
26 #include <sys/dmu.h>
27 #include <sys/dmu_objset.h>
28 #include <sys/dmu_tx.h>
29 #include <sys/dnode.h>
30 
31 uint64_t
32 dmu_object_alloc(objset_t *os, dmu_object_type_t ot, int blocksize,
33     dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx)
34 {
35 	objset_impl_t *osi = os->os;
36 	uint64_t object;
37 	uint64_t L2_dnode_count = DNODES_PER_BLOCK <<
38 	    (osi->os_meta_dnode->dn_indblkshift - SPA_BLKPTRSHIFT);
39 	dnode_t *dn = NULL;
40 	int restarted = B_FALSE;
41 
42 	mutex_enter(&osi->os_obj_lock);
43 	for (;;) {
44 		object = osi->os_obj_next;
45 		/*
46 		 * Each time we polish off an L2 bp worth of dnodes
47 		 * (2^13 objects), move to another L2 bp that's still
48 		 * reasonably sparse (at most 1/4 full).  Look from the
49 		 * beginning once, but after that keep looking from here.
50 		 * If we can't find one, just keep going from here.
51 		 */
52 		if (P2PHASE(object, L2_dnode_count) == 0) {
53 			uint64_t offset = restarted ? object << DNODE_SHIFT : 0;
54 			int error = dnode_next_offset(osi->os_meta_dnode,
55 			    DNODE_FIND_HOLE,
56 			    &offset, 2, DNODES_PER_BLOCK >> 2, 0);
57 			restarted = B_TRUE;
58 			if (error == 0)
59 				object = offset >> DNODE_SHIFT;
60 		}
61 		osi->os_obj_next = ++object;
62 
63 		/*
64 		 * XXX We should check for an i/o error here and return
65 		 * up to our caller.  Actually we should pre-read it in
66 		 * dmu_tx_assign(), but there is currently no mechanism
67 		 * to do so.
68 		 */
69 		(void) dnode_hold_impl(os->os, object, DNODE_MUST_BE_FREE,
70 		    FTAG, &dn);
71 		if (dn)
72 			break;
73 
74 		if (dmu_object_next(os, &object, B_TRUE, 0) == 0)
75 			osi->os_obj_next = object - 1;
76 	}
77 
78 	dnode_allocate(dn, ot, blocksize, 0, bonustype, bonuslen, tx);
79 	dnode_rele(dn, FTAG);
80 
81 	mutex_exit(&osi->os_obj_lock);
82 
83 	dmu_tx_add_new_object(tx, os, object);
84 	return (object);
85 }
86 
87 int
88 dmu_object_claim(objset_t *os, uint64_t object, dmu_object_type_t ot,
89     int blocksize, dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx)
90 {
91 	dnode_t *dn;
92 	int err;
93 
94 	if (object == DMU_META_DNODE_OBJECT && !dmu_tx_private_ok(tx))
95 		return (EBADF);
96 
97 	err = dnode_hold_impl(os->os, object, DNODE_MUST_BE_FREE, FTAG, &dn);
98 	if (err)
99 		return (err);
100 	dnode_allocate(dn, ot, blocksize, 0, bonustype, bonuslen, tx);
101 	dnode_rele(dn, FTAG);
102 
103 	dmu_tx_add_new_object(tx, os, object);
104 	return (0);
105 }
106 
107 int
108 dmu_object_reclaim(objset_t *os, uint64_t object, dmu_object_type_t ot,
109     int blocksize, dmu_object_type_t bonustype, int bonuslen)
110 {
111 	dnode_t *dn;
112 	dmu_tx_t *tx;
113 	int nblkptr;
114 	int err;
115 
116 	if (object == DMU_META_DNODE_OBJECT)
117 		return (EBADF);
118 
119 	err = dnode_hold_impl(os->os, object, DNODE_MUST_BE_ALLOCATED,
120 	    FTAG, &dn);
121 	if (err)
122 		return (err);
123 
124 	if (dn->dn_type == ot && dn->dn_datablksz == blocksize &&
125 	    dn->dn_bonustype == bonustype && dn->dn_bonuslen == bonuslen) {
126 		/* nothing is changing, this is a noop */
127 		dnode_rele(dn, FTAG);
128 		return (0);
129 	}
130 
131 	nblkptr = 1 + ((DN_MAX_BONUSLEN - bonuslen) >> SPA_BLKPTRSHIFT);
132 
133 	/*
134 	 * If we are losing blkptrs or changing the block size this must
135 	 * be a new file instance.   We must clear out the previous file
136 	 * contents before we can change this type of metadata in the dnode.
137 	 */
138 	if (dn->dn_nblkptr > nblkptr || dn->dn_datablksz != blocksize) {
139 		err = dmu_free_long_range(os, object, 0, DMU_OBJECT_END);
140 		if (err)
141 			goto out;
142 	}
143 
144 	tx = dmu_tx_create(os);
145 	dmu_tx_hold_bonus(tx, object);
146 	err = dmu_tx_assign(tx, TXG_WAIT);
147 	if (err) {
148 		dmu_tx_abort(tx);
149 		goto out;
150 	}
151 
152 	dnode_reallocate(dn, ot, blocksize, bonustype, bonuslen, tx);
153 
154 	dmu_tx_commit(tx);
155 out:
156 	dnode_rele(dn, FTAG);
157 
158 	return (err);
159 }
160 
161 int
162 dmu_object_free(objset_t *os, uint64_t object, dmu_tx_t *tx)
163 {
164 	dnode_t *dn;
165 	int err;
166 
167 	ASSERT(object != DMU_META_DNODE_OBJECT || dmu_tx_private_ok(tx));
168 
169 	err = dnode_hold_impl(os->os, object, DNODE_MUST_BE_ALLOCATED,
170 	    FTAG, &dn);
171 	if (err)
172 		return (err);
173 
174 	ASSERT(dn->dn_type != DMU_OT_NONE);
175 	dnode_free_range(dn, 0, DMU_OBJECT_END, tx);
176 	dnode_free(dn, tx);
177 	dnode_rele(dn, FTAG);
178 
179 	return (0);
180 }
181 
182 int
183 dmu_object_next(objset_t *os, uint64_t *objectp, boolean_t hole, uint64_t txg)
184 {
185 	uint64_t offset = (*objectp + 1) << DNODE_SHIFT;
186 	int error;
187 
188 	error = dnode_next_offset(os->os->os_meta_dnode,
189 	    (hole ? DNODE_FIND_HOLE : 0), &offset, 0, DNODES_PER_BLOCK, txg);
190 
191 	*objectp = offset >> DNODE_SHIFT;
192 
193 	return (error);
194 }
195