xref: /titanic_41/usr/src/uts/common/fs/zfs/dsl_pool.c (revision 058561cbaa119a6f2659bc27ef343e1b47266bb2)
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 2006 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/zfs_context.h>
37 #include <sys/fs/zfs.h>
38 
39 static int
40 dsl_pool_open_mos_dir(dsl_pool_t *dp, dsl_dir_t **ddp)
41 {
42 	uint64_t obj;
43 	int err;
44 
45 	err = zap_lookup(dp->dp_meta_objset,
46 	    dp->dp_root_dir->dd_phys->dd_child_dir_zapobj,
47 	    MOS_DIR_NAME, sizeof (obj), 1, &obj);
48 	if (err)
49 		return (err);
50 
51 	return (dsl_dir_open_obj(dp, obj, MOS_DIR_NAME, dp, ddp));
52 }
53 
54 static dsl_pool_t *
55 dsl_pool_open_impl(spa_t *spa, uint64_t txg)
56 {
57 	dsl_pool_t *dp;
58 	blkptr_t *bp = spa_get_rootblkptr(spa);
59 
60 	dp = kmem_zalloc(sizeof (dsl_pool_t), KM_SLEEP);
61 	dp->dp_spa = spa;
62 	dp->dp_meta_rootbp = *bp;
63 	rw_init(&dp->dp_config_rwlock, NULL, RW_DEFAULT, NULL);
64 	txg_init(dp, txg);
65 
66 	txg_list_create(&dp->dp_dirty_datasets,
67 	    offsetof(dsl_dataset_t, ds_dirty_link));
68 	txg_list_create(&dp->dp_dirty_dirs,
69 	    offsetof(dsl_dir_t, dd_dirty_link));
70 	txg_list_create(&dp->dp_sync_tasks,
71 	    offsetof(dsl_sync_task_group_t, dstg_node));
72 	list_create(&dp->dp_synced_objsets, sizeof (dsl_dataset_t),
73 	    offsetof(dsl_dataset_t, ds_synced_link));
74 
75 	return (dp);
76 }
77 
78 int
79 dsl_pool_open(spa_t *spa, uint64_t txg, dsl_pool_t **dpp)
80 {
81 	int err;
82 	dsl_pool_t *dp = dsl_pool_open_impl(spa, txg);
83 	objset_impl_t *osi;
84 
85 	rw_enter(&dp->dp_config_rwlock, RW_READER);
86 	err = dmu_objset_open_impl(spa, NULL, &dp->dp_meta_rootbp, &osi);
87 	if (err)
88 		goto out;
89 	dp->dp_meta_objset = &osi->os;
90 
91 	err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
92 	    DMU_POOL_ROOT_DATASET, sizeof (uint64_t), 1,
93 	    &dp->dp_root_dir_obj);
94 	if (err)
95 		goto out;
96 
97 	err = dsl_dir_open_obj(dp, dp->dp_root_dir_obj,
98 	    NULL, dp, &dp->dp_root_dir);
99 	if (err)
100 		goto out;
101 
102 	err = dsl_pool_open_mos_dir(dp, &dp->dp_mos_dir);
103 	if (err)
104 		goto out;
105 
106 out:
107 	rw_exit(&dp->dp_config_rwlock);
108 	if (err)
109 		dsl_pool_close(dp);
110 	else
111 		*dpp = dp;
112 
113 	return (err);
114 }
115 
116 void
117 dsl_pool_close(dsl_pool_t *dp)
118 {
119 	/* drop our reference from dsl_pool_open() */
120 	if (dp->dp_mos_dir)
121 		dsl_dir_close(dp->dp_mos_dir, dp);
122 	if (dp->dp_root_dir)
123 		dsl_dir_close(dp->dp_root_dir, dp);
124 
125 	/* undo the dmu_objset_open_impl(mos) from dsl_pool_open() */
126 	if (dp->dp_meta_objset)
127 		dmu_objset_evict(NULL, dp->dp_meta_objset->os);
128 
129 	txg_list_destroy(&dp->dp_dirty_datasets);
130 	txg_list_destroy(&dp->dp_dirty_dirs);
131 	list_destroy(&dp->dp_synced_objsets);
132 
133 	arc_flush();
134 	txg_fini(dp);
135 	rw_destroy(&dp->dp_config_rwlock);
136 	kmem_free(dp, sizeof (dsl_pool_t));
137 }
138 
139 dsl_pool_t *
140 dsl_pool_create(spa_t *spa, uint64_t txg)
141 {
142 	int err;
143 	dsl_pool_t *dp = dsl_pool_open_impl(spa, txg);
144 	dmu_tx_t *tx = dmu_tx_create_assigned(dp, txg);
145 	dp->dp_meta_objset = &dmu_objset_create_impl(spa,
146 	    NULL, DMU_OST_META, tx)->os;
147 
148 	/* create the pool directory */
149 	err = zap_create_claim(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
150 	    DMU_OT_OBJECT_DIRECTORY, DMU_OT_NONE, 0, tx);
151 	ASSERT3U(err, ==, 0);
152 
153 	/* create and open the root dir */
154 	dsl_dataset_create_root(dp, &dp->dp_root_dir_obj, tx);
155 	VERIFY(0 == dsl_dir_open_obj(dp, dp->dp_root_dir_obj,
156 	    NULL, dp, &dp->dp_root_dir));
157 
158 	/* create and open the meta-objset dir */
159 	(void) dsl_dir_create_sync(dp->dp_root_dir, MOS_DIR_NAME, tx);
160 	VERIFY(0 == dsl_pool_open_mos_dir(dp, &dp->dp_mos_dir));
161 
162 	dmu_tx_commit(tx);
163 
164 	return (dp);
165 }
166 
167 void
168 dsl_pool_sync(dsl_pool_t *dp, uint64_t txg)
169 {
170 	dmu_tx_t *tx;
171 	objset_impl_t *mosi = dp->dp_meta_objset->os;
172 
173 	tx = dmu_tx_create_assigned(dp, txg);
174 
175 	do {
176 		dsl_dir_t *dd;
177 		dsl_dataset_t *ds;
178 		dsl_sync_task_group_t *dstg;
179 
180 		while (ds = txg_list_remove(&dp->dp_dirty_datasets, txg)) {
181 			if (!list_link_active(&ds->ds_synced_link))
182 				list_insert_tail(&dp->dp_synced_objsets, ds);
183 			dsl_dataset_sync(ds, tx);
184 		}
185 		while (dstg = txg_list_remove(&dp->dp_sync_tasks, txg))
186 			dsl_sync_task_group_sync(dstg, tx);
187 		while (dd = txg_list_remove(&dp->dp_dirty_dirs, txg))
188 			dsl_dir_sync(dd, tx);
189 		/*
190 		 * We need to loop since dsl_sync_task_group_sync()
191 		 * could create a new (dirty) objset.
192 		 * XXX - isn't this taken care of by the spa's sync to
193 		 * convergence loop?
194 		 */
195 	} while (!txg_list_empty(&dp->dp_dirty_datasets, txg));
196 
197 	if (list_head(&mosi->os_dirty_dnodes[txg & TXG_MASK]) != NULL ||
198 	    list_head(&mosi->os_free_dnodes[txg & TXG_MASK]) != NULL) {
199 		dmu_objset_sync(mosi, tx);
200 		dprintf_bp(&dp->dp_meta_rootbp, "meta objset rootbp is %s", "");
201 		spa_set_rootblkptr(dp->dp_spa, &dp->dp_meta_rootbp);
202 	}
203 
204 	dmu_tx_commit(tx);
205 }
206 
207 void
208 dsl_pool_zil_clean(dsl_pool_t *dp)
209 {
210 	dsl_dataset_t *ds;
211 
212 	while (ds = list_head(&dp->dp_synced_objsets)) {
213 		list_remove(&dp->dp_synced_objsets, ds);
214 		ASSERT(ds->ds_user_ptr != NULL);
215 		zil_clean(((objset_impl_t *)ds->ds_user_ptr)->os_zil);
216 	}
217 }
218 
219 int
220 dsl_pool_sync_context(dsl_pool_t *dp)
221 {
222 	/*
223 	 * Yeah, this is cheesy.  But the SPA needs some way to let
224 	 * the sync threads invoke spa_open() and spa_close() while
225 	 * it holds the namespace lock.  I'm certainly open to better
226 	 * ideas for how to determine whether the current thread is
227 	 * operating on behalf of spa_sync().  This works for now.
228 	 */
229 	return (curthread == dp->dp_tx.tx_sync_thread ||
230 	    BP_IS_HOLE(&dp->dp_meta_rootbp));
231 }
232 
233 uint64_t
234 dsl_pool_adjustedsize(dsl_pool_t *dp, boolean_t netfree)
235 {
236 	uint64_t space, resv;
237 
238 	/*
239 	 * Reserve about 1.6% (1/64), or at least 32MB, for allocation
240 	 * efficiency.
241 	 * XXX The intent log is not accounted for, so it must fit
242 	 * within this slop.
243 	 *
244 	 * If we're trying to assess whether it's OK to do a free,
245 	 * cut the reservation in half to allow forward progress
246 	 * (e.g. make it possible to rm(1) files from a full pool).
247 	 */
248 	space = spa_get_dspace(dp->dp_spa);
249 	resv = MAX(space >> 6, SPA_MINDEVSIZE >> 1);
250 	if (netfree)
251 		resv >>= 1;
252 
253 	return (space - resv);
254 }
255