xref: /titanic_41/usr/src/uts/common/fs/zfs/dsl_pool.c (revision 6f45ec7b0b964c3be967c4880e8867ac1e7763a5)
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 2007 Sun Microsystems, Inc.  All rights reserved.
23  * Use is subject to license terms.
24  */
25 
26 #pragma ident	"%Z%%M%	%I%	%E% SMI"
27 
28 #include <sys/dsl_pool.h>
29 #include <sys/dsl_dataset.h>
30 #include <sys/dsl_dir.h>
31 #include <sys/dsl_synctask.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/zio.h>
37 #include <sys/zfs_context.h>
38 #include <sys/fs/zfs.h>
39 
40 static int
41 dsl_pool_open_mos_dir(dsl_pool_t *dp, dsl_dir_t **ddp)
42 {
43 	uint64_t obj;
44 	int err;
45 
46 	err = zap_lookup(dp->dp_meta_objset,
47 	    dp->dp_root_dir->dd_phys->dd_child_dir_zapobj,
48 	    MOS_DIR_NAME, sizeof (obj), 1, &obj);
49 	if (err)
50 		return (err);
51 
52 	return (dsl_dir_open_obj(dp, obj, MOS_DIR_NAME, dp, ddp));
53 }
54 
55 static dsl_pool_t *
56 dsl_pool_open_impl(spa_t *spa, uint64_t txg)
57 {
58 	dsl_pool_t *dp;
59 	blkptr_t *bp = spa_get_rootblkptr(spa);
60 
61 	dp = kmem_zalloc(sizeof (dsl_pool_t), KM_SLEEP);
62 	dp->dp_spa = spa;
63 	dp->dp_meta_rootbp = *bp;
64 	rw_init(&dp->dp_config_rwlock, NULL, RW_DEFAULT, NULL);
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 	txg_list_create(&dp->dp_sync_tasks,
72 	    offsetof(dsl_sync_task_group_t, dstg_node));
73 	list_create(&dp->dp_synced_objsets, sizeof (dsl_dataset_t),
74 	    offsetof(dsl_dataset_t, ds_synced_link));
75 
76 	return (dp);
77 }
78 
79 int
80 dsl_pool_open(spa_t *spa, uint64_t txg, dsl_pool_t **dpp)
81 {
82 	int err;
83 	dsl_pool_t *dp = dsl_pool_open_impl(spa, txg);
84 	objset_impl_t *osi;
85 
86 	rw_enter(&dp->dp_config_rwlock, RW_READER);
87 	err = dmu_objset_open_impl(spa, NULL, &dp->dp_meta_rootbp, &osi);
88 	if (err)
89 		goto out;
90 	dp->dp_meta_objset = &osi->os;
91 
92 	err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
93 	    DMU_POOL_ROOT_DATASET, sizeof (uint64_t), 1,
94 	    &dp->dp_root_dir_obj);
95 	if (err)
96 		goto out;
97 
98 	err = dsl_dir_open_obj(dp, dp->dp_root_dir_obj,
99 	    NULL, dp, &dp->dp_root_dir);
100 	if (err)
101 		goto out;
102 
103 	err = dsl_pool_open_mos_dir(dp, &dp->dp_mos_dir);
104 	if (err)
105 		goto out;
106 
107 out:
108 	rw_exit(&dp->dp_config_rwlock);
109 	if (err)
110 		dsl_pool_close(dp);
111 	else
112 		*dpp = dp;
113 
114 	return (err);
115 }
116 
117 void
118 dsl_pool_close(dsl_pool_t *dp)
119 {
120 	/* drop our reference from dsl_pool_open() */
121 	if (dp->dp_mos_dir)
122 		dsl_dir_close(dp->dp_mos_dir, dp);
123 	if (dp->dp_root_dir)
124 		dsl_dir_close(dp->dp_root_dir, dp);
125 
126 	/* undo the dmu_objset_open_impl(mos) from dsl_pool_open() */
127 	if (dp->dp_meta_objset)
128 		dmu_objset_evict(NULL, dp->dp_meta_objset->os);
129 
130 	txg_list_destroy(&dp->dp_dirty_datasets);
131 	txg_list_destroy(&dp->dp_dirty_dirs);
132 	list_destroy(&dp->dp_synced_objsets);
133 
134 	arc_flush();
135 	txg_fini(dp);
136 	rw_destroy(&dp->dp_config_rwlock);
137 	kmem_free(dp, sizeof (dsl_pool_t));
138 }
139 
140 dsl_pool_t *
141 dsl_pool_create(spa_t *spa, uint64_t txg)
142 {
143 	int err;
144 	dsl_pool_t *dp = dsl_pool_open_impl(spa, txg);
145 	dmu_tx_t *tx = dmu_tx_create_assigned(dp, txg);
146 	dp->dp_meta_objset = &dmu_objset_create_impl(spa,
147 	    NULL, &dp->dp_meta_rootbp, DMU_OST_META, tx)->os;
148 
149 	/* create the pool directory */
150 	err = zap_create_claim(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
151 	    DMU_OT_OBJECT_DIRECTORY, DMU_OT_NONE, 0, tx);
152 	ASSERT3U(err, ==, 0);
153 
154 	/* create and open the root dir */
155 	dsl_dataset_create_root(dp, &dp->dp_root_dir_obj, tx);
156 	VERIFY(0 == dsl_dir_open_obj(dp, dp->dp_root_dir_obj,
157 	    NULL, dp, &dp->dp_root_dir));
158 
159 	/* create and open the meta-objset dir */
160 	(void) dsl_dir_create_sync(dp->dp_root_dir, MOS_DIR_NAME, tx);
161 	VERIFY(0 == dsl_pool_open_mos_dir(dp, &dp->dp_mos_dir));
162 
163 	dmu_tx_commit(tx);
164 
165 	return (dp);
166 }
167 
168 void
169 dsl_pool_sync(dsl_pool_t *dp, uint64_t txg)
170 {
171 	zio_t *zio;
172 	dmu_tx_t *tx;
173 	dsl_dir_t *dd;
174 	dsl_dataset_t *ds;
175 	dsl_sync_task_group_t *dstg;
176 	objset_impl_t *mosi = dp->dp_meta_objset->os;
177 	int err;
178 
179 	tx = dmu_tx_create_assigned(dp, txg);
180 
181 	zio = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED);
182 	while (ds = txg_list_remove(&dp->dp_dirty_datasets, txg)) {
183 		if (!list_link_active(&ds->ds_synced_link))
184 			list_insert_tail(&dp->dp_synced_objsets, ds);
185 		dsl_dataset_sync(ds, zio, tx);
186 	}
187 	err = zio_wait(zio);
188 	ASSERT(err == 0);
189 
190 	while (dstg = txg_list_remove(&dp->dp_sync_tasks, txg))
191 		dsl_sync_task_group_sync(dstg, tx);
192 	while (dd = txg_list_remove(&dp->dp_dirty_dirs, txg))
193 		dsl_dir_sync(dd, tx);
194 
195 	if (list_head(&mosi->os_dirty_dnodes[txg & TXG_MASK]) != NULL ||
196 	    list_head(&mosi->os_free_dnodes[txg & TXG_MASK]) != NULL) {
197 		zio = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED);
198 		dmu_objset_sync(mosi, zio, tx);
199 		err = zio_wait(zio);
200 		ASSERT(err == 0);
201 		dprintf_bp(&dp->dp_meta_rootbp, "meta objset rootbp is %s", "");
202 		spa_set_rootblkptr(dp->dp_spa, &dp->dp_meta_rootbp);
203 	}
204 
205 	dmu_tx_commit(tx);
206 }
207 
208 void
209 dsl_pool_zil_clean(dsl_pool_t *dp)
210 {
211 	dsl_dataset_t *ds;
212 
213 	while (ds = list_head(&dp->dp_synced_objsets)) {
214 		list_remove(&dp->dp_synced_objsets, ds);
215 		ASSERT(ds->ds_user_ptr != NULL);
216 		zil_clean(((objset_impl_t *)ds->ds_user_ptr)->os_zil);
217 	}
218 }
219 
220 /*
221  * TRUE if the current thread is the tx_sync_thread or if we
222  * are being called from SPA context during pool initialization.
223  */
224 int
225 dsl_pool_sync_context(dsl_pool_t *dp)
226 {
227 	return (curthread == dp->dp_tx.tx_sync_thread ||
228 	    spa_get_dsl(dp->dp_spa) == NULL);
229 }
230 
231 uint64_t
232 dsl_pool_adjustedsize(dsl_pool_t *dp, boolean_t netfree)
233 {
234 	uint64_t space, resv;
235 
236 	/*
237 	 * Reserve about 1.6% (1/64), or at least 32MB, for allocation
238 	 * efficiency.
239 	 * XXX The intent log is not accounted for, so it must fit
240 	 * within this slop.
241 	 *
242 	 * If we're trying to assess whether it's OK to do a free,
243 	 * cut the reservation in half to allow forward progress
244 	 * (e.g. make it possible to rm(1) files from a full pool).
245 	 */
246 	space = spa_get_dspace(dp->dp_spa);
247 	resv = MAX(space >> 6, SPA_MINDEVSIZE >> 1);
248 	if (netfree)
249 		resv >>= 1;
250 
251 	return (space - resv);
252 }
253