Home | History | Annotate | Line # | Download | only in zfs
      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, 2014 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 */
     36 static int
     37 dsl_null_checkfunc(void *arg, dmu_tx_t *tx)
     38 {
     39 	return (0);
     40 }
     41 
     42 /*
     43  * Called from open context to perform a callback in syncing context.  Waits
     44  * for the operation to complete.
     45  *
     46  * The checkfunc will be called from open context as a preliminary check
     47  * which can quickly fail.  If it succeeds, it will be called again from
     48  * syncing context.  The checkfunc should generally be designed to work
     49  * properly in either context, but if necessary it can check
     50  * dmu_tx_is_syncing(tx).
     51  *
     52  * The synctask infrastructure enforces proper locking strategy with respect
     53  * to the dp_config_rwlock -- the lock will always be held when the callbacks
     54  * are called.  It will be held for read during the open-context (preliminary)
     55  * call to the checkfunc, and then held for write from syncing context during
     56  * the calls to the check and sync funcs.
     57  *
     58  * A dataset or pool name can be passed as the first argument.  Typically,
     59  * the check func will hold, check the return value of the hold, and then
     60  * release the dataset.  The sync func will VERIFYO(hold()) the dataset.
     61  * This is safe because no changes can be made between the check and sync funcs,
     62  * and the sync func will only be called if the check func successfully opened
     63  * the dataset.
     64  */
     65 int
     66 dsl_sync_task(const char *pool, dsl_checkfunc_t *checkfunc,
     67     dsl_syncfunc_t *syncfunc, void *arg,
     68     int blocks_modified, zfs_space_check_t space_check)
     69 {
     70 	spa_t *spa;
     71 	dmu_tx_t *tx;
     72 	int err;
     73 	dsl_sync_task_t dst = { 0 };
     74 	dsl_pool_t *dp;
     75 
     76 	err = spa_open(pool, &spa, FTAG);
     77 	if (err != 0)
     78 		return (err);
     79 	dp = spa_get_dsl(spa);
     80 
     81 top:
     82 	tx = dmu_tx_create_dd(dp->dp_mos_dir);
     83 	VERIFY0(dmu_tx_assign(tx, TXG_WAIT));
     84 
     85 	dst.dst_pool = dp;
     86 	dst.dst_txg = dmu_tx_get_txg(tx);
     87 	dst.dst_space = blocks_modified << DST_AVG_BLKSHIFT;
     88 	dst.dst_space_check = space_check;
     89 	dst.dst_checkfunc = checkfunc != NULL ? checkfunc : dsl_null_checkfunc;
     90 	dst.dst_syncfunc = syncfunc;
     91 	dst.dst_arg = arg;
     92 	dst.dst_error = 0;
     93 	dst.dst_nowaiter = B_FALSE;
     94 
     95 	dsl_pool_config_enter(dp, FTAG);
     96 	err = dst.dst_checkfunc(arg, tx);
     97 	dsl_pool_config_exit(dp, FTAG);
     98 
     99 	if (err != 0) {
    100 		dmu_tx_commit(tx);
    101 		spa_close(spa, FTAG);
    102 		return (err);
    103 	}
    104 
    105 	VERIFY(txg_list_add_tail(&dp->dp_sync_tasks, &dst, dst.dst_txg));
    106 
    107 	dmu_tx_commit(tx);
    108 
    109 	txg_wait_synced(dp, dst.dst_txg);
    110 
    111 	if (dst.dst_error == EAGAIN) {
    112 		txg_wait_synced(dp, dst.dst_txg + TXG_DEFER_SIZE);
    113 		goto top;
    114 	}
    115 
    116 	spa_close(spa, FTAG);
    117 	return (dst.dst_error);
    118 }
    119 
    120 void
    121 dsl_sync_task_nowait(dsl_pool_t *dp, dsl_syncfunc_t *syncfunc, void *arg,
    122     int blocks_modified, zfs_space_check_t space_check, dmu_tx_t *tx)
    123 {
    124 	dsl_sync_task_t *dst = kmem_zalloc(sizeof (*dst), KM_SLEEP);
    125 
    126 	dst->dst_pool = dp;
    127 	dst->dst_txg = dmu_tx_get_txg(tx);
    128 	dst->dst_space = blocks_modified << DST_AVG_BLKSHIFT;
    129 	dst->dst_space_check = space_check;
    130 	dst->dst_checkfunc = dsl_null_checkfunc;
    131 	dst->dst_syncfunc = syncfunc;
    132 	dst->dst_arg = arg;
    133 	dst->dst_error = 0;
    134 	dst->dst_nowaiter = B_TRUE;
    135 
    136 	VERIFY(txg_list_add_tail(&dp->dp_sync_tasks, dst, dst->dst_txg));
    137 }
    138 
    139 /*
    140  * Called in syncing context to execute the synctask.
    141  */
    142 void
    143 dsl_sync_task_sync(dsl_sync_task_t *dst, dmu_tx_t *tx)
    144 {
    145 	dsl_pool_t *dp = dst->dst_pool;
    146 
    147 	ASSERT0(dst->dst_error);
    148 
    149 	/*
    150 	 * Check for sufficient space.
    151 	 *
    152 	 * When the sync task was created, the caller specified the
    153 	 * type of space checking required.  See the comment in
    154 	 * zfs_space_check_t for details on the semantics of each
    155 	 * type of space checking.
    156 	 *
    157 	 * We just check against what's on-disk; we don't want any
    158 	 * in-flight accounting to get in our way, because open context
    159 	 * may have already used up various in-core limits
    160 	 * (arc_tempreserve, dsl_pool_tempreserve).
    161 	 */
    162 	if (dst->dst_space_check != ZFS_SPACE_CHECK_NONE) {
    163 		uint64_t quota = dsl_pool_adjustedsize(dp,
    164 		    dst->dst_space_check == ZFS_SPACE_CHECK_RESERVED) -
    165 		    metaslab_class_get_deferred(spa_normal_class(dp->dp_spa));
    166 		uint64_t used = dsl_dir_phys(dp->dp_root_dir)->dd_used_bytes;
    167 		/* MOS space is triple-dittoed, so we multiply by 3. */
    168 		if (dst->dst_space > 0 && used + dst->dst_space * 3 > quota) {
    169 			dst->dst_error = SET_ERROR(ENOSPC);
    170 			if (dst->dst_nowaiter)
    171 				kmem_free(dst, sizeof (*dst));
    172 			return;
    173 		}
    174 	}
    175 
    176 	/*
    177 	 * Check for errors by calling checkfunc.
    178 	 */
    179 	rrw_enter(&dp->dp_config_rwlock, RW_WRITER, FTAG);
    180 	dst->dst_error = dst->dst_checkfunc(dst->dst_arg, tx);
    181 	if (dst->dst_error == 0)
    182 		dst->dst_syncfunc(dst->dst_arg, tx);
    183 	rrw_exit(&dp->dp_config_rwlock, FTAG);
    184 	if (dst->dst_nowaiter)
    185 		kmem_free(dst, sizeof (*dst));
    186 }
    187