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) 2012, 2017 by Delphix. All rights reserved.
24 */
25
26#include <sys/dmu.h>
27#include <sys/dmu_tx.h>
28#include <sys/dsl_pool.h>
29#include <sys/dsl_dir.h>
30#include <sys/dsl_synctask.h>
31#include <sys/metaslab.h>
32
33#define	DST_AVG_BLKSHIFT 14
34
35/* ARGSUSED */
36static int
37dsl_null_checkfunc(void *arg, dmu_tx_t *tx)
38{
39	return (0);
40}
41
42static int
43dsl_sync_task_common(const char *pool, dsl_checkfunc_t *checkfunc,
44    dsl_syncfunc_t *syncfunc, dsl_sigfunc_t *sigfunc, void *arg,
45    int blocks_modified, zfs_space_check_t space_check, boolean_t early)
46{
47	spa_t *spa;
48	dmu_tx_t *tx;
49	int err;
50	dsl_sync_task_t dst = { 0 };
51	dsl_pool_t *dp;
52
53	err = spa_open(pool, &spa, FTAG);
54	if (err != 0)
55		return (err);
56	dp = spa_get_dsl(spa);
57
58top:
59	tx = dmu_tx_create_dd(dp->dp_mos_dir);
60	VERIFY0(dmu_tx_assign(tx, TXG_WAIT));
61
62	dst.dst_pool = dp;
63	dst.dst_txg = dmu_tx_get_txg(tx);
64	dst.dst_space = blocks_modified << DST_AVG_BLKSHIFT;
65	dst.dst_space_check = space_check;
66	dst.dst_checkfunc = checkfunc != NULL ? checkfunc : dsl_null_checkfunc;
67	dst.dst_syncfunc = syncfunc;
68	dst.dst_arg = arg;
69	dst.dst_error = 0;
70	dst.dst_nowaiter = B_FALSE;
71
72	dsl_pool_config_enter(dp, FTAG);
73	err = dst.dst_checkfunc(arg, tx);
74	dsl_pool_config_exit(dp, FTAG);
75
76	if (err != 0) {
77		dmu_tx_commit(tx);
78		spa_close(spa, FTAG);
79		return (err);
80	}
81
82	txg_list_t *task_list = (early) ?
83	    &dp->dp_early_sync_tasks : &dp->dp_sync_tasks;
84	VERIFY(txg_list_add_tail(task_list, &dst, dst.dst_txg));
85
86	dmu_tx_commit(tx);
87
88	if (sigfunc != NULL && txg_wait_synced_sig(dp, dst.dst_txg)) {
89		/* current contract is to call func once */
90		sigfunc(arg, tx);
91		sigfunc = NULL;	/* in case of an EAGAIN retry */
92	}
93	txg_wait_synced(dp, dst.dst_txg);
94
95	if (dst.dst_error == EAGAIN) {
96		txg_wait_synced(dp, dst.dst_txg + TXG_DEFER_SIZE);
97		goto top;
98	}
99
100	spa_close(spa, FTAG);
101	return (dst.dst_error);
102}
103
104/*
105 * Called from open context to perform a callback in syncing context.  Waits
106 * for the operation to complete.
107 *
108 * The checkfunc will be called from open context as a preliminary check
109 * which can quickly fail.  If it succeeds, it will be called again from
110 * syncing context.  The checkfunc should generally be designed to work
111 * properly in either context, but if necessary it can check
112 * dmu_tx_is_syncing(tx).
113 *
114 * The synctask infrastructure enforces proper locking strategy with respect
115 * to the dp_config_rwlock -- the lock will always be held when the callbacks
116 * are called.  It will be held for read during the open-context (preliminary)
117 * call to the checkfunc, and then held for write from syncing context during
118 * the calls to the check and sync funcs.
119 *
120 * A dataset or pool name can be passed as the first argument.  Typically,
121 * the check func will hold, check the return value of the hold, and then
122 * release the dataset.  The sync func will VERIFYO(hold()) the dataset.
123 * This is safe because no changes can be made between the check and sync funcs,
124 * and the sync func will only be called if the check func successfully opened
125 * the dataset.
126 */
127int
128dsl_sync_task(const char *pool, dsl_checkfunc_t *checkfunc,
129    dsl_syncfunc_t *syncfunc, void *arg,
130    int blocks_modified, zfs_space_check_t space_check)
131{
132	return (dsl_sync_task_common(pool, checkfunc, syncfunc, NULL, arg,
133	    blocks_modified, space_check, B_FALSE));
134}
135
136/*
137 * An early synctask works exactly as a standard synctask with one important
138 * difference on the way it is handled during syncing context. Standard
139 * synctasks run after we've written out all the dirty blocks of dirty
140 * datasets. Early synctasks are executed before writing out any dirty data,
141 * and thus before standard synctasks.
142 *
143 * For that reason, early synctasks can affect the process of writing dirty
144 * changes to disk for the txg that they run and should be used with caution.
145 * In addition, early synctasks should not dirty any metaslabs as this would
146 * invalidate the precodition/invariant for subsequent early synctasks.
147 * [see dsl_pool_sync() and dsl_early_sync_task_verify()]
148 */
149int
150dsl_early_sync_task(const char *pool, dsl_checkfunc_t *checkfunc,
151    dsl_syncfunc_t *syncfunc, void *arg,
152    int blocks_modified, zfs_space_check_t space_check)
153{
154	return (dsl_sync_task_common(pool, checkfunc, syncfunc, NULL, arg,
155	    blocks_modified, space_check, B_TRUE));
156}
157
158/*
159 * A standard synctask that can be interrupted from a signal. The sigfunc
160 * is called once if a signal occurred while waiting for the task to sync.
161 */
162int
163dsl_sync_task_sig(const char *pool, dsl_checkfunc_t *checkfunc,
164    dsl_syncfunc_t *syncfunc, dsl_sigfunc_t *sigfunc, void *arg,
165    int blocks_modified, zfs_space_check_t space_check)
166{
167	return (dsl_sync_task_common(pool, checkfunc, syncfunc, sigfunc, arg,
168	    blocks_modified, space_check, B_FALSE));
169}
170
171static void
172dsl_sync_task_nowait_common(dsl_pool_t *dp, dsl_syncfunc_t *syncfunc, void *arg,
173    int blocks_modified, zfs_space_check_t space_check, dmu_tx_t *tx,
174    boolean_t early)
175{
176	dsl_sync_task_t *dst = kmem_zalloc(sizeof (*dst), KM_SLEEP);
177
178	dst->dst_pool = dp;
179	dst->dst_txg = dmu_tx_get_txg(tx);
180	dst->dst_space = blocks_modified << DST_AVG_BLKSHIFT;
181	dst->dst_space_check = space_check;
182	dst->dst_checkfunc = dsl_null_checkfunc;
183	dst->dst_syncfunc = syncfunc;
184	dst->dst_arg = arg;
185	dst->dst_error = 0;
186	dst->dst_nowaiter = B_TRUE;
187
188	txg_list_t *task_list = (early) ?
189	    &dp->dp_early_sync_tasks : &dp->dp_sync_tasks;
190	VERIFY(txg_list_add_tail(task_list, dst, dst->dst_txg));
191}
192
193void
194dsl_sync_task_nowait(dsl_pool_t *dp, dsl_syncfunc_t *syncfunc, void *arg,
195    int blocks_modified, zfs_space_check_t space_check, dmu_tx_t *tx)
196{
197	dsl_sync_task_nowait_common(dp, syncfunc, arg,
198	    blocks_modified, space_check, tx, B_FALSE);
199}
200
201void
202dsl_early_sync_task_nowait(dsl_pool_t *dp, dsl_syncfunc_t *syncfunc, void *arg,
203    int blocks_modified, zfs_space_check_t space_check, dmu_tx_t *tx)
204{
205	dsl_sync_task_nowait_common(dp, syncfunc, arg,
206	    blocks_modified, space_check, tx, B_TRUE);
207}
208
209/*
210 * Called in syncing context to execute the synctask.
211 */
212void
213dsl_sync_task_sync(dsl_sync_task_t *dst, dmu_tx_t *tx)
214{
215	dsl_pool_t *dp = dst->dst_pool;
216
217	ASSERT0(dst->dst_error);
218
219	/*
220	 * Check for sufficient space.
221	 *
222	 * When the sync task was created, the caller specified the
223	 * type of space checking required.  See the comment in
224	 * zfs_space_check_t for details on the semantics of each
225	 * type of space checking.
226	 *
227	 * We just check against what's on-disk; we don't want any
228	 * in-flight accounting to get in our way, because open context
229	 * may have already used up various in-core limits
230	 * (arc_tempreserve, dsl_pool_tempreserve).
231	 */
232	if (dst->dst_space_check != ZFS_SPACE_CHECK_NONE) {
233		uint64_t quota = dsl_pool_unreserved_space(dp,
234		    dst->dst_space_check);
235		uint64_t used = dsl_dir_phys(dp->dp_root_dir)->dd_used_bytes;
236
237		/* MOS space is triple-dittoed, so we multiply by 3. */
238		if (used + dst->dst_space * 3 > quota) {
239			dst->dst_error = SET_ERROR(ENOSPC);
240			if (dst->dst_nowaiter)
241				kmem_free(dst, sizeof (*dst));
242			return;
243		}
244	}
245
246	/*
247	 * Check for errors by calling checkfunc.
248	 */
249	rrw_enter(&dp->dp_config_rwlock, RW_WRITER, FTAG);
250	dst->dst_error = dst->dst_checkfunc(dst->dst_arg, tx);
251	if (dst->dst_error == 0)
252		dst->dst_syncfunc(dst->dst_arg, tx);
253	rrw_exit(&dp->dp_config_rwlock, FTAG);
254	if (dst->dst_nowaiter)
255		kmem_free(dst, sizeof (*dst));
256}
257