xref: /illumos-gate/usr/src/uts/common/fs/zfs/dmu_object.c (revision 621be8d08fd45483b5ca1cb8e2e88239f1502b4d)
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 (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23  * Copyright (c) 2013, 2015 by Delphix. All rights reserved.
24  * Copyright 2014 HybridCluster. All rights reserved.
25  */
26 
27 #include <sys/dmu.h>
28 #include <sys/dmu_objset.h>
29 #include <sys/dmu_tx.h>
30 #include <sys/dnode.h>
31 #include <sys/zap.h>
32 #include <sys/zfeature.h>
33 
34 uint64_t
35 dmu_object_alloc(objset_t *os, dmu_object_type_t ot, int blocksize,
36     dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx)
37 {
38 	uint64_t object;
39 	uint64_t L2_dnode_count = DNODES_PER_BLOCK <<
40 	    (DMU_META_DNODE(os)->dn_indblkshift - SPA_BLKPTRSHIFT);
41 	dnode_t *dn = NULL;
42 	int restarted = B_FALSE;
43 
44 	mutex_enter(&os->os_obj_lock);
45 	for (;;) {
46 		object = os->os_obj_next;
47 		/*
48 		 * Each time we polish off an L2 bp worth of dnodes
49 		 * (2^13 objects), move to another L2 bp that's still
50 		 * reasonably sparse (at most 1/4 full).  Look from the
51 		 * beginning once, but after that keep looking from here.
52 		 * If we can't find one, just keep going from here.
53 		 *
54 		 * Note that dmu_traverse depends on the behavior that we use
55 		 * multiple blocks of the dnode object before going back to
56 		 * reuse objects.  Any change to this algorithm should preserve
57 		 * that property or find another solution to the issues
58 		 * described in traverse_visitbp.
59 		 */
60 		if (P2PHASE(object, L2_dnode_count) == 0) {
61 			uint64_t offset = restarted ? object << DNODE_SHIFT : 0;
62 			int error = dnode_next_offset(DMU_META_DNODE(os),
63 			    DNODE_FIND_HOLE,
64 			    &offset, 2, DNODES_PER_BLOCK >> 2, 0);
65 			restarted = B_TRUE;
66 			if (error == 0)
67 				object = offset >> DNODE_SHIFT;
68 		}
69 		os->os_obj_next = ++object;
70 
71 		/*
72 		 * XXX We should check for an i/o error here and return
73 		 * up to our caller.  Actually we should pre-read it in
74 		 * dmu_tx_assign(), but there is currently no mechanism
75 		 * to do so.
76 		 */
77 		(void) dnode_hold_impl(os, object, DNODE_MUST_BE_FREE,
78 		    FTAG, &dn);
79 		if (dn)
80 			break;
81 
82 		if (dmu_object_next(os, &object, B_TRUE, 0) == 0)
83 			os->os_obj_next = object - 1;
84 	}
85 
86 	dnode_allocate(dn, ot, blocksize, 0, bonustype, bonuslen, tx);
87 	dnode_rele(dn, FTAG);
88 
89 	mutex_exit(&os->os_obj_lock);
90 
91 	dmu_tx_add_new_object(tx, os, object);
92 	return (object);
93 }
94 
95 int
96 dmu_object_claim(objset_t *os, uint64_t object, dmu_object_type_t ot,
97     int blocksize, dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx)
98 {
99 	dnode_t *dn;
100 	int err;
101 
102 	if (object == DMU_META_DNODE_OBJECT && !dmu_tx_private_ok(tx))
103 		return (SET_ERROR(EBADF));
104 
105 	err = dnode_hold_impl(os, object, DNODE_MUST_BE_FREE, FTAG, &dn);
106 	if (err)
107 		return (err);
108 	dnode_allocate(dn, ot, blocksize, 0, bonustype, bonuslen, tx);
109 	dnode_rele(dn, FTAG);
110 
111 	dmu_tx_add_new_object(tx, os, object);
112 	return (0);
113 }
114 
115 int
116 dmu_object_reclaim(objset_t *os, uint64_t object, dmu_object_type_t ot,
117     int blocksize, dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx)
118 {
119 	dnode_t *dn;
120 	int err;
121 
122 	if (object == DMU_META_DNODE_OBJECT)
123 		return (SET_ERROR(EBADF));
124 
125 	err = dnode_hold_impl(os, object, DNODE_MUST_BE_ALLOCATED,
126 	    FTAG, &dn);
127 	if (err)
128 		return (err);
129 
130 	dnode_reallocate(dn, ot, blocksize, bonustype, bonuslen, tx);
131 
132 	dnode_rele(dn, FTAG);
133 	return (err);
134 }
135 
136 int
137 dmu_object_free(objset_t *os, uint64_t object, dmu_tx_t *tx)
138 {
139 	dnode_t *dn;
140 	int err;
141 
142 	ASSERT(object != DMU_META_DNODE_OBJECT || dmu_tx_private_ok(tx));
143 
144 	err = dnode_hold_impl(os, object, DNODE_MUST_BE_ALLOCATED,
145 	    FTAG, &dn);
146 	if (err)
147 		return (err);
148 
149 	ASSERT(dn->dn_type != DMU_OT_NONE);
150 	dnode_free_range(dn, 0, DMU_OBJECT_END, tx);
151 	dnode_free(dn, tx);
152 	dnode_rele(dn, FTAG);
153 
154 	return (0);
155 }
156 
157 /*
158  * Return (in *objectp) the next object which is allocated (or a hole)
159  * after *object, taking into account only objects that may have been modified
160  * after the specified txg.
161  */
162 int
163 dmu_object_next(objset_t *os, uint64_t *objectp, boolean_t hole, uint64_t txg)
164 {
165 	uint64_t offset = (*objectp + 1) << DNODE_SHIFT;
166 	int error;
167 
168 	error = dnode_next_offset(DMU_META_DNODE(os),
169 	    (hole ? DNODE_FIND_HOLE : 0), &offset, 0, DNODES_PER_BLOCK, txg);
170 
171 	*objectp = offset >> DNODE_SHIFT;
172 
173 	return (error);
174 }
175 
176 /*
177  * Turn this object from old_type into DMU_OTN_ZAP_METADATA, and bump the
178  * refcount on SPA_FEATURE_EXTENSIBLE_DATASET.
179  *
180  * Only for use from syncing context, on MOS objects.
181  */
182 void
183 dmu_object_zapify(objset_t *mos, uint64_t object, dmu_object_type_t old_type,
184     dmu_tx_t *tx)
185 {
186 	dnode_t *dn;
187 
188 	ASSERT(dmu_tx_is_syncing(tx));
189 
190 	VERIFY0(dnode_hold(mos, object, FTAG, &dn));
191 	if (dn->dn_type == DMU_OTN_ZAP_METADATA) {
192 		dnode_rele(dn, FTAG);
193 		return;
194 	}
195 	ASSERT3U(dn->dn_type, ==, old_type);
196 	ASSERT0(dn->dn_maxblkid);
197 	dn->dn_next_type[tx->tx_txg & TXG_MASK] = dn->dn_type =
198 	    DMU_OTN_ZAP_METADATA;
199 	dnode_setdirty(dn, tx);
200 	dnode_rele(dn, FTAG);
201 
202 	mzap_create_impl(mos, object, 0, 0, tx);
203 
204 	spa_feature_incr(dmu_objset_spa(mos),
205 	    SPA_FEATURE_EXTENSIBLE_DATASET, tx);
206 }
207 
208 void
209 dmu_object_free_zapified(objset_t *mos, uint64_t object, dmu_tx_t *tx)
210 {
211 	dnode_t *dn;
212 	dmu_object_type_t t;
213 
214 	ASSERT(dmu_tx_is_syncing(tx));
215 
216 	VERIFY0(dnode_hold(mos, object, FTAG, &dn));
217 	t = dn->dn_type;
218 	dnode_rele(dn, FTAG);
219 
220 	if (t == DMU_OTN_ZAP_METADATA) {
221 		spa_feature_decr(dmu_objset_spa(mos),
222 		    SPA_FEATURE_EXTENSIBLE_DATASET, tx);
223 	}
224 	VERIFY0(dmu_object_free(mos, object, tx));
225 }
226