xref: /titanic_41/usr/src/uts/common/fs/zfs/dsl_pool.c (revision fd9cb95cbb2f626355a60efb9d02c5f0a33c10e6)
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, Version 1.0 only
6  * (the "License").  You may not use this file except in compliance
7  * with the License.
8  *
9  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
10  * or http://www.opensolaris.org/os/licensing.
11  * See the License for the specific language governing permissions
12  * and limitations under the License.
13  *
14  * When distributing Covered Code, include this CDDL HEADER in each
15  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
16  * If applicable, add the following below this CDDL HEADER, with the
17  * fields enclosed by brackets "[]" replaced with your own identifying
18  * information: Portions Copyright [yyyy] [name of copyright owner]
19  *
20  * CDDL HEADER END
21  */
22 /*
23  * Copyright 2005 Sun Microsystems, Inc.  All rights reserved.
24  * Use is subject to license terms.
25  */
26 
27 #pragma ident	"%Z%%M%	%I%	%E% SMI"
28 
29 #include <sys/dsl_pool.h>
30 #include <sys/dsl_dataset.h>
31 #include <sys/dsl_dir.h>
32 #include <sys/dmu_tx.h>
33 #include <sys/dmu_objset.h>
34 #include <sys/arc.h>
35 #include <sys/zap.h>
36 #include <sys/zfs_context.h>
37 #include <sys/fs/zfs.h>
38 
39 /* internal reserved dir name */
40 #define	MOS_DIR_NAME "$MOS"
41 
42 static dsl_dir_t *
43 dsl_pool_open_mos_dir(dsl_pool_t *dp)
44 {
45 	uint64_t obj;
46 	int err;
47 
48 	err = zap_lookup(dp->dp_meta_objset,
49 	    dp->dp_root_dir->dd_phys->dd_child_dir_zapobj,
50 	    MOS_DIR_NAME, sizeof (obj), 1, &obj);
51 	ASSERT3U(err, ==, 0);
52 
53 	return (dsl_dir_open_obj(dp, obj, MOS_DIR_NAME, dp));
54 }
55 
56 static dsl_pool_t *
57 dsl_pool_open_impl(spa_t *spa, uint64_t txg)
58 {
59 	dsl_pool_t *dp;
60 	blkptr_t *bp = spa_get_rootblkptr(spa);
61 
62 	dp = kmem_zalloc(sizeof (dsl_pool_t), KM_SLEEP);
63 	dp->dp_spa = spa;
64 	dp->dp_meta_rootbp = *bp;
65 	txg_init(dp, txg);
66 
67 	txg_list_create(&dp->dp_dirty_datasets,
68 	    offsetof(dsl_dataset_t, ds_dirty_link));
69 	txg_list_create(&dp->dp_dirty_dirs,
70 	    offsetof(dsl_dir_t, dd_dirty_link));
71 	list_create(&dp->dp_synced_objsets, sizeof (dsl_dataset_t),
72 	    offsetof(dsl_dataset_t, ds_synced_link));
73 
74 	return (dp);
75 }
76 
77 dsl_pool_t *
78 dsl_pool_open(spa_t *spa, uint64_t txg)
79 {
80 	int err;
81 	dsl_pool_t *dp = dsl_pool_open_impl(spa, txg);
82 
83 	dp->dp_meta_objset =
84 	    &dmu_objset_open_impl(spa, NULL, &dp->dp_meta_rootbp)->os;
85 
86 	rw_enter(&dp->dp_config_rwlock, RW_READER);
87 	err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
88 	    DMU_POOL_ROOT_DATASET, sizeof (uint64_t), 1,
89 	    &dp->dp_root_dir_obj);
90 	ASSERT3U(err, ==, 0);
91 
92 	dp->dp_root_dir = dsl_dir_open_obj(dp, dp->dp_root_dir_obj,
93 	    NULL, dp);
94 	dp->dp_mos_dir = dsl_pool_open_mos_dir(dp);
95 	rw_exit(&dp->dp_config_rwlock);
96 
97 	return (dp);
98 }
99 
100 void
101 dsl_pool_close(dsl_pool_t *dp)
102 {
103 	/* drop our reference from dsl_pool_open() */
104 	dsl_dir_close(dp->dp_mos_dir, dp);
105 	dsl_dir_close(dp->dp_root_dir, dp);
106 
107 	/* undo the dmu_objset_open_impl(mos) from dsl_pool_open() */
108 	dmu_objset_evict(NULL, dp->dp_meta_objset->os);
109 
110 	txg_list_destroy(&dp->dp_dirty_datasets);
111 	txg_list_destroy(&dp->dp_dirty_dirs);
112 	list_destroy(&dp->dp_synced_objsets);
113 
114 	arc_flush();
115 	txg_fini(dp);
116 	kmem_free(dp, sizeof (dsl_pool_t));
117 }
118 
119 dsl_pool_t *
120 dsl_pool_create(spa_t *spa, uint64_t txg)
121 {
122 	int err;
123 	dsl_pool_t *dp = dsl_pool_open_impl(spa, txg);
124 	dmu_tx_t *tx = dmu_tx_create_assigned(dp, txg);
125 	dp->dp_meta_objset = &dmu_objset_create_impl(spa,
126 	    NULL, DMU_OST_META, tx)->os;
127 
128 	/* create the pool directory */
129 	err = zap_create_claim(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
130 	    DMU_OT_OBJECT_DIRECTORY, DMU_OT_NONE, 0, tx);
131 	ASSERT3U(err, ==, 0);
132 
133 	/* create and open the root dir */
134 	dsl_dataset_create_root(dp, &dp->dp_root_dir_obj, tx);
135 	dp->dp_root_dir = dsl_dir_open_obj(dp, dp->dp_root_dir_obj,
136 	    NULL, dp);
137 
138 	/* create and open the meta-objset dir */
139 	err = dsl_dir_create_sync(dp->dp_root_dir, MOS_DIR_NAME,
140 	    tx);
141 	ASSERT3U(err, ==, 0);
142 	dp->dp_mos_dir = dsl_pool_open_mos_dir(dp);
143 
144 	dmu_tx_commit(tx);
145 
146 	return (dp);
147 }
148 
149 void
150 dsl_pool_sync(dsl_pool_t *dp, uint64_t txg)
151 {
152 	dmu_tx_t *tx;
153 	objset_impl_t *mosi = dp->dp_meta_objset->os;
154 
155 	tx = dmu_tx_create_assigned(dp, txg);
156 
157 	do {
158 		dsl_dir_t *dd;
159 		dsl_dataset_t *ds;
160 
161 		while (ds = txg_list_remove(&dp->dp_dirty_datasets, txg)) {
162 			if (!list_link_active(&ds->ds_synced_link))
163 				list_insert_tail(&dp->dp_synced_objsets, ds);
164 			dsl_dataset_sync(ds, tx);
165 		}
166 		while (dd = txg_list_remove(&dp->dp_dirty_dirs, txg))
167 			dsl_dir_sync(dd, tx);
168 		/*
169 		 * We need to loop since dsl_dir_sync() could create a
170 		 * new (dirty) objset.
171 		 * XXX - isn't this taken care of by the spa's sync to
172 		 * convergence loop?
173 		 */
174 	} while (!txg_list_empty(&dp->dp_dirty_datasets, txg));
175 
176 	if (list_head(&mosi->os_dirty_dnodes[txg & TXG_MASK]) != NULL ||
177 	    list_head(&mosi->os_free_dnodes[txg & TXG_MASK]) != NULL) {
178 		dmu_objset_sync(mosi, tx);
179 		dprintf_bp(&dp->dp_meta_rootbp, "meta objset rootbp is %s", "");
180 		spa_set_rootblkptr(dp->dp_spa, &dp->dp_meta_rootbp);
181 	}
182 
183 	dmu_tx_commit(tx);
184 }
185 
186 void
187 dsl_pool_zil_clean(dsl_pool_t *dp)
188 {
189 	dsl_dataset_t *ds;
190 
191 	while (ds = list_head(&dp->dp_synced_objsets)) {
192 		list_remove(&dp->dp_synced_objsets, ds);
193 		ASSERT(ds->ds_user_ptr != NULL);
194 		zil_clean(((objset_impl_t *)ds->ds_user_ptr)->os_zil);
195 	}
196 }
197 
198 int
199 dsl_pool_sync_context(dsl_pool_t *dp)
200 {
201 	/*
202 	 * Yeah, this is cheesy.  But the SPA needs some way to let
203 	 * the sync threads invoke spa_open() and spa_close() while
204 	 * it holds the namespace lock.  I'm certainly open to better
205 	 * ideas for how to determine whether the current thread is
206 	 * operating on behalf of spa_sync().  This works for now.
207 	 */
208 	return (curthread == dp->dp_tx.tx_sync_thread ||
209 	    BP_IS_HOLE(&dp->dp_meta_rootbp));
210 }
211 
212 uint64_t
213 dsl_pool_adjustedsize(dsl_pool_t *dp, boolean_t netfree)
214 {
215 	uint64_t space, resv;
216 
217 	/*
218 	 * Reserve about 1% (1/128), or at least 16MB, for allocation
219 	 * efficiency.
220 	 * XXX The intent log is not accounted for, so it must fit
221 	 * within this slop.
222 	 *
223 	 * If we're trying to assess whether it's OK to do a free,
224 	 * cut the reservation in half to allow forward progress
225 	 * (e.g. make it possible to rm(1) files from a full pool).
226 	 */
227 	space = spa_get_space(dp->dp_spa);
228 	resv = MAX(space >> 7, SPA_MINDEVSIZE >> 2);
229 	if (netfree)
230 		resv >>= 1;
231 
232 	return (space - resv);
233 }
234