[mirror_zfs.git] / module / zfs / dsl_synctask.c

/*
 * CDDL HEADER START
 *
 * The contents of this file are subject to the terms of the
 * Common Development and Distribution License (the "License").
 * You may not use this file except in compliance with the License.
 *
 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
 * or https://opensource.org/licenses/CDDL-1.0.
 * See the License for the specific language governing permissions
 * and limitations under the License.
 *
 * When distributing Covered Code, include this CDDL HEADER in each
 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
 * If applicable, add the following below this CDDL HEADER, with the
 * fields enclosed by brackets "[]" replaced with your own identifying
 * information: Portions Copyright [yyyy] [name of copyright owner]
 *
 * CDDL HEADER END
 */
/*
 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
 * Copyright (c) 2012, 2017 by Delphix. All rights reserved.
 */

#include <sys/dmu.h>
#include <sys/dmu_tx.h>
#include <sys/dsl_pool.h>
#include <sys/dsl_dir.h>
#include <sys/dsl_synctask.h>
#include <sys/metaslab.h>

#define	DST_AVG_BLKSHIFT 14

static int
dsl_null_checkfunc(void *arg, dmu_tx_t *tx)
{
	(void) arg, (void) tx;
	return (0);
}

static int
dsl_sync_task_common(const char *pool, dsl_checkfunc_t *checkfunc,
    dsl_syncfunc_t *syncfunc, dsl_sigfunc_t *sigfunc, void *arg,
    int blocks_modified, zfs_space_check_t space_check, boolean_t early)
{
	spa_t *spa;
	dmu_tx_t *tx;
	int err;
	dsl_sync_task_t dst = { { { NULL } } };
	dsl_pool_t *dp;

	err = spa_open(pool, &spa, FTAG);
	if (err != 0)
		return (err);
	dp = spa_get_dsl(spa);

top:
	tx = dmu_tx_create_dd(dp->dp_mos_dir);
	VERIFY0(dmu_tx_assign(tx, TXG_WAIT));

	dst.dst_pool = dp;
	dst.dst_txg = dmu_tx_get_txg(tx);
	dst.dst_space = blocks_modified << DST_AVG_BLKSHIFT;
	dst.dst_space_check = space_check;
	dst.dst_checkfunc = checkfunc != NULL ? checkfunc : dsl_null_checkfunc;
	dst.dst_syncfunc = syncfunc;
	dst.dst_arg = arg;
	dst.dst_error = 0;
	dst.dst_nowaiter = B_FALSE;

	dsl_pool_config_enter(dp, FTAG);
	err = dst.dst_checkfunc(arg, tx);
	dsl_pool_config_exit(dp, FTAG);

	if (err != 0) {
		dmu_tx_commit(tx);
		spa_close(spa, FTAG);
		return (err);
	}

	txg_list_t *task_list = (early) ?
	    &dp->dp_early_sync_tasks : &dp->dp_sync_tasks;
	VERIFY(txg_list_add_tail(task_list, &dst, dst.dst_txg));

	dmu_tx_commit(tx);

	if (sigfunc != NULL && txg_wait_synced_sig(dp, dst.dst_txg)) {
		/* current contract is to call func once */
		sigfunc(arg, tx);
		sigfunc = NULL;	/* in case we're performing an EAGAIN retry */
	}
	txg_wait_synced(dp, dst.dst_txg);

	if (dst.dst_error == EAGAIN) {
		txg_wait_synced(dp, dst.dst_txg + TXG_DEFER_SIZE);
		goto top;
	}

	spa_close(spa, FTAG);
	return (dst.dst_error);
}

/*
 * Called from open context to perform a callback in syncing context.  Waits
 * for the operation to complete.
 *
 * The checkfunc will be called from open context as a preliminary check
 * which can quickly fail.  If it succeeds, it will be called again from
 * syncing context.  The checkfunc should generally be designed to work
 * properly in either context, but if necessary it can check
 * dmu_tx_is_syncing(tx).
 *
 * The synctask infrastructure enforces proper locking strategy with respect
 * to the dp_config_rwlock -- the lock will always be held when the callbacks
 * are called.  It will be held for read during the open-context (preliminary)
 * call to the checkfunc, and then held for write from syncing context during
 * the calls to the check and sync funcs.
 *
 * A dataset or pool name can be passed as the first argument.  Typically,
 * the check func will hold, check the return value of the hold, and then
 * release the dataset.  The sync func will VERIFYO(hold()) the dataset.
 * This is safe because no changes can be made between the check and sync funcs,
 * and the sync func will only be called if the check func successfully opened
 * the dataset.
 */
int
dsl_sync_task(const char *pool, dsl_checkfunc_t *checkfunc,
    dsl_syncfunc_t *syncfunc, void *arg,
    int blocks_modified, zfs_space_check_t space_check)
{
	return (dsl_sync_task_common(pool, checkfunc, syncfunc, NULL, arg,
	    blocks_modified, space_check, B_FALSE));
}

/*
 * An early synctask works exactly as a standard synctask with one important
 * difference on the way it is handled during syncing context. Standard
 * synctasks run after we've written out all the dirty blocks of dirty
 * datasets. Early synctasks are executed before writing out any dirty data,
 * and thus before standard synctasks.
 *
 * For that reason, early synctasks can affect the process of writing dirty
 * changes to disk for the txg that they run and should be used with caution.
 * In addition, early synctasks should not dirty any metaslabs as this would
 * invalidate the precondition/invariant for subsequent early synctasks.
 * [see dsl_pool_sync() and dsl_early_sync_task_verify()]
 */
int
dsl_early_sync_task(const char *pool, dsl_checkfunc_t *checkfunc,
    dsl_syncfunc_t *syncfunc, void *arg,
    int blocks_modified, zfs_space_check_t space_check)
{
	return (dsl_sync_task_common(pool, checkfunc, syncfunc, NULL, arg,
	    blocks_modified, space_check, B_TRUE));
}

/*
 * A standard synctask that can be interrupted from a signal. The sigfunc
 * is called once if a signal occurred while waiting for the task to sync.
 */
int
dsl_sync_task_sig(const char *pool, dsl_checkfunc_t *checkfunc,
    dsl_syncfunc_t *syncfunc, dsl_sigfunc_t *sigfunc, void *arg,
    int blocks_modified, zfs_space_check_t space_check)
{
	return (dsl_sync_task_common(pool, checkfunc, syncfunc, sigfunc, arg,
	    blocks_modified, space_check, B_FALSE));
}

static void
dsl_sync_task_nowait_common(dsl_pool_t *dp, dsl_syncfunc_t *syncfunc, void *arg,
    dmu_tx_t *tx, boolean_t early)
{
	dsl_sync_task_t *dst = kmem_zalloc(sizeof (*dst), KM_SLEEP);

	dst->dst_pool = dp;
	dst->dst_txg = dmu_tx_get_txg(tx);
	dst->dst_space_check = ZFS_SPACE_CHECK_NONE;
	dst->dst_checkfunc = dsl_null_checkfunc;
	dst->dst_syncfunc = syncfunc;
	dst->dst_arg = arg;
	dst->dst_error = 0;
	dst->dst_nowaiter = B_TRUE;

	txg_list_t *task_list = (early) ?
	    &dp->dp_early_sync_tasks : &dp->dp_sync_tasks;
	VERIFY(txg_list_add_tail(task_list, dst, dst->dst_txg));
}

void
dsl_sync_task_nowait(dsl_pool_t *dp, dsl_syncfunc_t *syncfunc, void *arg,
    dmu_tx_t *tx)
{
	dsl_sync_task_nowait_common(dp, syncfunc, arg, tx, B_FALSE);
}

void
dsl_early_sync_task_nowait(dsl_pool_t *dp, dsl_syncfunc_t *syncfunc, void *arg,
    dmu_tx_t *tx)
{
	dsl_sync_task_nowait_common(dp, syncfunc, arg, tx, B_TRUE);
}

/*
 * Called in syncing context to execute the synctask.
 */
void
dsl_sync_task_sync(dsl_sync_task_t *dst, dmu_tx_t *tx)
{
	dsl_pool_t *dp = dst->dst_pool;

	ASSERT0(dst->dst_error);

	/*
	 * Check for sufficient space.
	 *
	 * When the sync task was created, the caller specified the
	 * type of space checking required.  See the comment in
	 * zfs_space_check_t for details on the semantics of each
	 * type of space checking.
	 *
	 * We just check against what's on-disk; we don't want any
	 * in-flight accounting to get in our way, because open context
	 * may have already used up various in-core limits
	 * (arc_tempreserve, dsl_pool_tempreserve).
	 */
	if (dst->dst_space_check != ZFS_SPACE_CHECK_NONE) {
		uint64_t quota = dsl_pool_unreserved_space(dp,
		    dst->dst_space_check);
		uint64_t used = dsl_dir_phys(dp->dp_root_dir)->dd_used_bytes;

		/* MOS space is triple-dittoed, so we multiply by 3. */
		if (used + dst->dst_space * 3 > quota) {
			dst->dst_error = SET_ERROR(ENOSPC);
			if (dst->dst_nowaiter)
				kmem_free(dst, sizeof (*dst));
			return;
		}
	}

	/*
	 * Check for errors by calling checkfunc.
	 */
	rrw_enter(&dp->dp_config_rwlock, RW_WRITER, FTAG);
	dst->dst_error = dst->dst_checkfunc(dst->dst_arg, tx);
	if (dst->dst_error == 0)
		dst->dst_syncfunc(dst->dst_arg, tx);
	rrw_exit(&dp->dp_config_rwlock, FTAG);
	if (dst->dst_nowaiter)
		kmem_free(dst, sizeof (*dst));
}

#if defined(_KERNEL)
EXPORT_SYMBOL(dsl_sync_task);
EXPORT_SYMBOL(dsl_sync_task_nowait);
#endif
Commit	Line	Data
34dc7c2f BB	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
1d3ba0bf	9	* or https://opensource.org/licenses/CDDL-1.0.
34dc7c2f BB	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	/*
428870ff	22	* Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
d2734cce	23	* Copyright (c) 2012, 2017 by Delphix. All rights reserved.
34dc7c2f BB	24	*/
34dc7c2f BB	25
34dc7c2f BB	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>
428870ff	31	#include <sys/metaslab.h>
34dc7c2f BB	32
	33	#define DST_AVG_BLKSHIFT 14
	34
34dc7c2f	35	static int
13fe0198	36	dsl_null_checkfunc(void arg, dmu_tx_t tx)
34dc7c2f	37	{
14e4e3cb	38	(void) arg, (void) tx;
34dc7c2f BB	39	return (0);
	40	}
	41
d2734cce SD	42	static int
d2734cce SD	43	dsl_sync_task_common(const char pool, dsl_checkfunc_t checkfunc,
186898bb	44	dsl_syncfunc_t syncfunc, dsl_sigfunc_t sigfunc, void *arg,
d2734cce	45	int blocks_modified, zfs_space_check_t space_check, boolean_t early)
34dc7c2f	46	{
13fe0198	47	spa_t *spa;
34dc7c2f	48	dmu_tx_t *tx;
13fe0198 MA	49	int err;
	50	dsl_sync_task_t dst = { { { NULL } } };
	51	dsl_pool_t *dp;
34dc7c2f	52
13fe0198 MA	53	err = spa_open(pool, &spa, FTAG);
	54	if (err != 0)
	55	return (err);
	56	dp = spa_get_dsl(spa);
34dc7c2f	57
13fe0198 MA	58	top:
	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;
3d45fdd6	65	dst.dst_space_check = space_check;
13fe0198 MA	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) {
34dc7c2f	77	dmu_tx_commit(tx);
13fe0198 MA	78	spa_close(spa, FTAG);
13fe0198 MA	79	return (err);
34dc7c2f BB	80	}
34dc7c2f BB	81
d2734cce SD	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));
34dc7c2f BB	85
	86	dmu_tx_commit(tx);
	87
186898bb DB	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 we're performing an EAGAIN retry */
	92	}
13fe0198	93	txg_wait_synced(dp, dst.dst_txg);
34dc7c2f	94
13fe0198 MA	95	if (dst.dst_error == EAGAIN) {
13fe0198 MA	96	txg_wait_synced(dp, dst.dst_txg + TXG_DEFER_SIZE);
34dc7c2f	97	goto top;
428870ff	98	}
34dc7c2f	99
13fe0198 MA	100	spa_close(spa, FTAG);
13fe0198 MA	101	return (dst.dst_error);
34dc7c2f BB	102	}
34dc7c2f BB	103
d2734cce SD	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	*/
	127	int
	128	dsl_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	{
186898bb	132	return (dsl_sync_task_common(pool, checkfunc, syncfunc, NULL, arg,
d2734cce SD	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
e1cfd73f	146	* invalidate the precondition/invariant for subsequent early synctasks.
d2734cce SD	147	* [see dsl_pool_sync() and dsl_early_sync_task_verify()]
	148	*/
	149	int
	150	dsl_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	{
186898bb	154	return (dsl_sync_task_common(pool, checkfunc, syncfunc, NULL, arg,
d2734cce SD	155	blocks_modified, space_check, B_TRUE));
	156	}
	157
186898bb DB	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	*/
	162	int
	163	dsl_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
d2734cce SD	171	static void
d2734cce SD	172	dsl_sync_task_nowait_common(dsl_pool_t dp, dsl_syncfunc_t syncfunc, void *arg,
38080324	173	dmu_tx_t *tx, boolean_t early)
34dc7c2f	174	{
13fe0198	175	dsl_sync_task_t dst = kmem_zalloc(sizeof (dst), KM_SLEEP);
34dc7c2f	176
13fe0198 MA	177	dst->dst_pool = dp;
13fe0198 MA	178	dst->dst_txg = dmu_tx_get_txg(tx);
38080324	179	dst->dst_space_check = ZFS_SPACE_CHECK_NONE;
13fe0198 MA	180	dst->dst_checkfunc = dsl_null_checkfunc;
	181	dst->dst_syncfunc = syncfunc;
	182	dst->dst_arg = arg;
	183	dst->dst_error = 0;
	184	dst->dst_nowaiter = B_TRUE;
34dc7c2f	185
d2734cce SD	186	txg_list_t *task_list = (early) ?
	187	&dp->dp_early_sync_tasks : &dp->dp_sync_tasks;
	188	VERIFY(txg_list_add_tail(task_list, dst, dst->dst_txg));
	189	}
	190
	191	void
	192	dsl_sync_task_nowait(dsl_pool_t dp, dsl_syncfunc_t syncfunc, void *arg,
38080324	193	dmu_tx_t *tx)
d2734cce	194	{
38080324	195	dsl_sync_task_nowait_common(dp, syncfunc, arg, tx, B_FALSE);
d2734cce SD	196	}
	197
	198	void
	199	dsl_early_sync_task_nowait(dsl_pool_t dp, dsl_syncfunc_t syncfunc, void *arg,
38080324	200	dmu_tx_t *tx)
d2734cce	201	{
38080324	202	dsl_sync_task_nowait_common(dp, syncfunc, arg, tx, B_TRUE);
34dc7c2f BB	203	}
34dc7c2f BB	204
13fe0198 MA	205	/*
	206	* Called in syncing context to execute the synctask.
	207	*/
34dc7c2f	208	void
13fe0198	209	dsl_sync_task_sync(dsl_sync_task_t dst, dmu_tx_t tx)
34dc7c2f	210	{
13fe0198	211	dsl_pool_t *dp = dst->dst_pool;
34dc7c2f	212
13fe0198	213	ASSERT0(dst->dst_error);
34dc7c2f BB	214
34dc7c2f BB	215	/*
3d45fdd6 MA	216	* Check for sufficient space.
	217	*
	218	* When the sync task was created, the caller specified the
	219	* type of space checking required. See the comment in
	220	* zfs_space_check_t for details on the semantics of each
	221	* type of space checking.
	222	*
	223	* We just check against what's on-disk; we don't want any
	224	* in-flight accounting to get in our way, because open context
	225	* may have already used up various in-core limits
	226	* (arc_tempreserve, dsl_pool_tempreserve).
34dc7c2f	227	*/
3d45fdd6	228	if (dst->dst_space_check != ZFS_SPACE_CHECK_NONE) {
d2734cce SD	229	uint64_t quota = dsl_pool_unreserved_space(dp,
d2734cce SD	230	dst->dst_space_check);
3d45fdd6	231	uint64_t used = dsl_dir_phys(dp->dp_root_dir)->dd_used_bytes;
d2734cce	232
3d45fdd6	233	/* MOS space is triple-dittoed, so we multiply by 3. */
d2734cce	234	if (used + dst->dst_space * 3 > quota) {
3d45fdd6 MA	235	dst->dst_error = SET_ERROR(ENOSPC);
	236	if (dst->dst_nowaiter)
	237	kmem_free(dst, sizeof (*dst));
	238	return;
	239	}
428870ff	240	}
34dc7c2f BB	241
34dc7c2f BB	242	/*
13fe0198	243	* Check for errors by calling checkfunc.
34dc7c2f	244	*/
13fe0198 MA	245	rrw_enter(&dp->dp_config_rwlock, RW_WRITER, FTAG);
	246	dst->dst_error = dst->dst_checkfunc(dst->dst_arg, tx);
	247	if (dst->dst_error == 0)
	248	dst->dst_syncfunc(dst->dst_arg, tx);
	249	rrw_exit(&dp->dp_config_rwlock, FTAG);
	250	if (dst->dst_nowaiter)
	251	kmem_free(dst, sizeof (*dst));
34dc7c2f	252	}
c28b2279	253
93ce2b4c	254	#if defined(_KERNEL)
e7440015 BB	255	EXPORT_SYMBOL(dsl_sync_task);
e7440015 BB	256	EXPORT_SYMBOL(dsl_sync_task_nowait);
c28b2279	257	#endif