[mirror_zfs.git] / module / zfs / vdev_mirror.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 http://www.opensolaris.org/os/licensing.
 * 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 2010 Sun Microsystems, Inc.  All rights reserved.
 * Use is subject to license terms.
 */

/*
 * Copyright (c) 2012 by Delphix. All rights reserved.
 */

#include <sys/zfs_context.h>
#include <sys/spa.h>
#include <sys/vdev_impl.h>
#include <sys/zio.h>
#include <sys/fs/zfs.h>

/*
 * Virtual device vector for mirroring.
 */

typedef struct mirror_child {
	vdev_t		*mc_vd;
	uint64_t	mc_offset;
	int		mc_error;
	uint8_t		mc_tried;
	uint8_t		mc_skipped;
	uint8_t		mc_speculative;
} mirror_child_t;

typedef struct mirror_map {
	int		mm_children;
	int		mm_replacing;
	int		mm_preferred;
	int		mm_root;
	mirror_child_t	mm_child[1];
} mirror_map_t;

int vdev_mirror_shift = 21;

static void
vdev_mirror_map_free(zio_t *zio)
{
	mirror_map_t *mm = zio->io_vsd;

	kmem_free(mm, offsetof(mirror_map_t, mm_child[mm->mm_children]));
}

static const zio_vsd_ops_t vdev_mirror_vsd_ops = {
	vdev_mirror_map_free,
	zio_vsd_default_cksum_report
};

static mirror_map_t *
vdev_mirror_map_alloc(zio_t *zio)
{
	mirror_map_t *mm = NULL;
	mirror_child_t *mc;
	vdev_t *vd = zio->io_vd;
	int c, d;

	if (vd == NULL) {
		dva_t *dva = zio->io_bp->blk_dva;
		spa_t *spa = zio->io_spa;

		c = BP_GET_NDVAS(zio->io_bp);

		mm = kmem_zalloc(offsetof(mirror_map_t, mm_child[c]), KM_PUSHPAGE);
		mm->mm_children = c;
		mm->mm_replacing = B_FALSE;
		mm->mm_preferred = spa_get_random(c);
		mm->mm_root = B_TRUE;

		/*
		 * Check the other, lower-index DVAs to see if they're on
		 * the same vdev as the child we picked.  If they are, use
		 * them since they are likely to have been allocated from
		 * the primary metaslab in use at the time, and hence are
		 * more likely to have locality with single-copy data.
		 */
		for (c = mm->mm_preferred, d = c - 1; d >= 0; d--) {
			if (DVA_GET_VDEV(&dva[d]) == DVA_GET_VDEV(&dva[c]))
				mm->mm_preferred = d;
		}

		for (c = 0; c < mm->mm_children; c++) {
			mc = &mm->mm_child[c];

			mc->mc_vd = vdev_lookup_top(spa, DVA_GET_VDEV(&dva[c]));
			mc->mc_offset = DVA_GET_OFFSET(&dva[c]);
		}
	} else {
		c = vd->vdev_children;

		mm = kmem_zalloc(offsetof(mirror_map_t, mm_child[c]), KM_PUSHPAGE);
		mm->mm_children = c;
		mm->mm_replacing = (vd->vdev_ops == &vdev_replacing_ops ||
		    vd->vdev_ops == &vdev_spare_ops);
		mm->mm_preferred = mm->mm_replacing ? 0 :
		    (zio->io_offset >> vdev_mirror_shift) % c;
		mm->mm_root = B_FALSE;

		for (c = 0; c < mm->mm_children; c++) {
			mc = &mm->mm_child[c];
			mc->mc_vd = vd->vdev_child[c];
			mc->mc_offset = zio->io_offset;
		}
	}

	zio->io_vsd = mm;
	zio->io_vsd_ops = &vdev_mirror_vsd_ops;
	return (mm);
}

static int
vdev_mirror_open(vdev_t *vd, uint64_t *asize, uint64_t *max_asize,
    uint64_t *ashift)
{
	int numerrors = 0;
	int lasterror = 0;
	int c;

	if (vd->vdev_children == 0) {
		vd->vdev_stat.vs_aux = VDEV_AUX_BAD_LABEL;
		return (EINVAL);
	}

	vdev_open_children(vd);

	for (c = 0; c < vd->vdev_children; c++) {
		vdev_t *cvd = vd->vdev_child[c];

		if (cvd->vdev_open_error) {
			lasterror = cvd->vdev_open_error;
			numerrors++;
			continue;
		}

		*asize = MIN(*asize - 1, cvd->vdev_asize - 1) + 1;
		*max_asize = MIN(*max_asize - 1, cvd->vdev_max_asize - 1) + 1;
		*ashift = MAX(*ashift, cvd->vdev_ashift);
	}

	if (numerrors == vd->vdev_children) {
		vd->vdev_stat.vs_aux = VDEV_AUX_NO_REPLICAS;
		return (lasterror);
	}

	return (0);
}

static void
vdev_mirror_close(vdev_t *vd)
{
	int c;

	for (c = 0; c < vd->vdev_children; c++)
		vdev_close(vd->vdev_child[c]);
}

static void
vdev_mirror_child_done(zio_t *zio)
{
	mirror_child_t *mc = zio->io_private;

	mc->mc_error = zio->io_error;
	mc->mc_tried = 1;
	mc->mc_skipped = 0;
}

static void
vdev_mirror_scrub_done(zio_t *zio)
{
	mirror_child_t *mc = zio->io_private;

	if (zio->io_error == 0) {
		zio_t *pio;

		mutex_enter(&zio->io_lock);
		while ((pio = zio_walk_parents(zio)) != NULL) {
			mutex_enter(&pio->io_lock);
			ASSERT3U(zio->io_size, >=, pio->io_size);
			bcopy(zio->io_data, pio->io_data, pio->io_size);
			mutex_exit(&pio->io_lock);
		}
		mutex_exit(&zio->io_lock);
	}

	zio_buf_free(zio->io_data, zio->io_size);

	mc->mc_error = zio->io_error;
	mc->mc_tried = 1;
	mc->mc_skipped = 0;
}

/*
 * Try to find a child whose DTL doesn't contain the block we want to read.
 * If we can't, try the read on any vdev we haven't already tried.
 */
static int
vdev_mirror_child_select(zio_t *zio)
{
	mirror_map_t *mm = zio->io_vsd;
	mirror_child_t *mc;
	uint64_t txg = zio->io_txg;
	int i, c;

	ASSERT(zio->io_bp == NULL || BP_PHYSICAL_BIRTH(zio->io_bp) == txg);

	/*
	 * Try to find a child whose DTL doesn't contain the block to read.
	 * If a child is known to be completely inaccessible (indicated by
	 * vdev_readable() returning B_FALSE), don't even try.
	 */
	for (i = 0, c = mm->mm_preferred; i < mm->mm_children; i++, c++) {
		if (c >= mm->mm_children)
			c = 0;
		mc = &mm->mm_child[c];
		if (mc->mc_tried || mc->mc_skipped)
			continue;
		if (!vdev_readable(mc->mc_vd)) {
			mc->mc_error = ENXIO;
			mc->mc_tried = 1;	/* don't even try */
			mc->mc_skipped = 1;
			continue;
		}
		if (!vdev_dtl_contains(mc->mc_vd, DTL_MISSING, txg, 1))
			return (c);
		mc->mc_error = ESTALE;
		mc->mc_skipped = 1;
		mc->mc_speculative = 1;
	}

	/*
	 * Every device is either missing or has this txg in its DTL.
	 * Look for any child we haven't already tried before giving up.
	 */
	for (c = 0; c < mm->mm_children; c++)
		if (!mm->mm_child[c].mc_tried)
			return (c);

	/*
	 * Every child failed.  There's no place left to look.
	 */
	return (-1);
}

static int
vdev_mirror_io_start(zio_t *zio)
{
	mirror_map_t *mm;
	mirror_child_t *mc;
	int c, children;

	mm = vdev_mirror_map_alloc(zio);

	if (zio->io_type == ZIO_TYPE_READ) {
		if ((zio->io_flags & ZIO_FLAG_SCRUB) && !mm->mm_replacing) {
			/*
			 * For scrubbing reads we need to allocate a read
			 * buffer for each child and issue reads to all
			 * children.  If any child succeeds, it will copy its
			 * data into zio->io_data in vdev_mirror_scrub_done.
			 */
			for (c = 0; c < mm->mm_children; c++) {
				mc = &mm->mm_child[c];
				zio_nowait(zio_vdev_child_io(zio, zio->io_bp,
				    mc->mc_vd, mc->mc_offset,
				    zio_buf_alloc(zio->io_size), zio->io_size,
				    zio->io_type, zio->io_priority, 0,
				    vdev_mirror_scrub_done, mc));
			}
			return (ZIO_PIPELINE_CONTINUE);
		}
		/*
		 * For normal reads just pick one child.
		 */
		c = vdev_mirror_child_select(zio);
		children = (c >= 0);
	} else {
		ASSERT(zio->io_type == ZIO_TYPE_WRITE);

		/*
		 * Writes go to all children.
		 */
		c = 0;
		children = mm->mm_children;
	}

	while (children--) {
		mc = &mm->mm_child[c];
		zio_nowait(zio_vdev_child_io(zio, zio->io_bp,
		    mc->mc_vd, mc->mc_offset, zio->io_data, zio->io_size,
		    zio->io_type, zio->io_priority, 0,
		    vdev_mirror_child_done, mc));
		c++;
	}

	return (ZIO_PIPELINE_CONTINUE);
}

static int
vdev_mirror_worst_error(mirror_map_t *mm)
{
	int c, error[2] = { 0, 0 };

	for (c = 0; c < mm->mm_children; c++) {
		mirror_child_t *mc = &mm->mm_child[c];
		int s = mc->mc_speculative;
		error[s] = zio_worst_error(error[s], mc->mc_error);
	}

	return (error[0] ? error[0] : error[1]);
}

static void
vdev_mirror_io_done(zio_t *zio)
{
	mirror_map_t *mm = zio->io_vsd;
	mirror_child_t *mc;
	int c;
	int good_copies = 0;
	int unexpected_errors = 0;

	for (c = 0; c < mm->mm_children; c++) {
		mc = &mm->mm_child[c];

		if (mc->mc_error) {
			if (!mc->mc_skipped)
				unexpected_errors++;
		} else if (mc->mc_tried) {
			good_copies++;
		}
	}

	if (zio->io_type == ZIO_TYPE_WRITE) {
		/*
		 * XXX -- for now, treat partial writes as success.
		 *
		 * Now that we support write reallocation, it would be better
		 * to treat partial failure as real failure unless there are
		 * no non-degraded top-level vdevs left, and not update DTLs
		 * if we intend to reallocate.
		 */
		/* XXPOLICY */
		if (good_copies != mm->mm_children) {
			/*
			 * Always require at least one good copy.
			 *
			 * For ditto blocks (io_vd == NULL), require
			 * all copies to be good.
			 *
			 * XXX -- for replacing vdevs, there's no great answer.
			 * If the old device is really dead, we may not even
			 * be able to access it -- so we only want to
			 * require good writes to the new device.  But if
			 * the new device turns out to be flaky, we want
			 * to be able to detach it -- which requires all
			 * writes to the old device to have succeeded.
			 */
			if (good_copies == 0 || zio->io_vd == NULL)
				zio->io_error = vdev_mirror_worst_error(mm);
		}
		return;
	}

	ASSERT(zio->io_type == ZIO_TYPE_READ);

	/*
	 * If we don't have a good copy yet, keep trying other children.
	 */
	/* XXPOLICY */
	if (good_copies == 0 && (c = vdev_mirror_child_select(zio)) != -1) {
		ASSERT(c >= 0 && c < mm->mm_children);
		mc = &mm->mm_child[c];
		zio_vdev_io_redone(zio);
		zio_nowait(zio_vdev_child_io(zio, zio->io_bp,
		    mc->mc_vd, mc->mc_offset, zio->io_data, zio->io_size,
		    ZIO_TYPE_READ, zio->io_priority, 0,
		    vdev_mirror_child_done, mc));
		return;
	}

	/* XXPOLICY */
	if (good_copies == 0) {
		zio->io_error = vdev_mirror_worst_error(mm);
		ASSERT(zio->io_error != 0);
	}

	if (good_copies && spa_writeable(zio->io_spa) &&
	    (unexpected_errors ||
	    (zio->io_flags & ZIO_FLAG_RESILVER) ||
	    ((zio->io_flags & ZIO_FLAG_SCRUB) && mm->mm_replacing))) {
		/*
		 * Use the good data we have in hand to repair damaged children.
		 */
		for (c = 0; c < mm->mm_children; c++) {
			/*
			 * Don't rewrite known good children.
			 * Not only is it unnecessary, it could
			 * actually be harmful: if the system lost
			 * power while rewriting the only good copy,
			 * there would be no good copies left!
			 */
			mc = &mm->mm_child[c];

			if (mc->mc_error == 0) {
				if (mc->mc_tried)
					continue;
				if (!(zio->io_flags & ZIO_FLAG_SCRUB) &&
				    !vdev_dtl_contains(mc->mc_vd, DTL_PARTIAL,
				    zio->io_txg, 1))
					continue;
				mc->mc_error = ESTALE;
			}

			zio_nowait(zio_vdev_child_io(zio, zio->io_bp,
			    mc->mc_vd, mc->mc_offset,
			    zio->io_data, zio->io_size,
			    ZIO_TYPE_WRITE, zio->io_priority,
			    ZIO_FLAG_IO_REPAIR | (unexpected_errors ?
			    ZIO_FLAG_SELF_HEAL : 0), NULL, NULL));
		}
	}
}

static void
vdev_mirror_state_change(vdev_t *vd, int faulted, int degraded)
{
	if (faulted == vd->vdev_children)
		vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
		    VDEV_AUX_NO_REPLICAS);
	else if (degraded + faulted != 0)
		vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED, VDEV_AUX_NONE);
	else
		vdev_set_state(vd, B_FALSE, VDEV_STATE_HEALTHY, VDEV_AUX_NONE);
}

vdev_ops_t vdev_mirror_ops = {
	vdev_mirror_open,
	vdev_mirror_close,
	vdev_default_asize,
	vdev_mirror_io_start,
	vdev_mirror_io_done,
	vdev_mirror_state_change,
	NULL,
	NULL,
	VDEV_TYPE_MIRROR,	/* name of this vdev type */
	B_FALSE			/* not a leaf vdev */
};

vdev_ops_t vdev_replacing_ops = {
	vdev_mirror_open,
	vdev_mirror_close,
	vdev_default_asize,
	vdev_mirror_io_start,
	vdev_mirror_io_done,
	vdev_mirror_state_change,
	NULL,
	NULL,
	VDEV_TYPE_REPLACING,	/* name of this vdev type */
	B_FALSE			/* not a leaf vdev */
};

vdev_ops_t vdev_spare_ops = {
	vdev_mirror_open,
	vdev_mirror_close,
	vdev_default_asize,
	vdev_mirror_io_start,
	vdev_mirror_io_done,
	vdev_mirror_state_change,
	NULL,
	NULL,
	VDEV_TYPE_SPARE,	/* name of this vdev type */
	B_FALSE			/* not a leaf vdev */
};
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
	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	/*
428870ff	22	* Copyright 2010 Sun Microsystems, Inc. All rights reserved.
34dc7c2f BB	23	* Use is subject to license terms.
	24	*/
	25
1bd201e7 CS	26	/*
	27	* Copyright (c) 2012 by Delphix. All rights reserved.
	28	*/
	29
34dc7c2f BB	30	#include <sys/zfs_context.h>
	31	#include <sys/spa.h>
	32	#include <sys/vdev_impl.h>
	33	#include <sys/zio.h>
	34	#include <sys/fs/zfs.h>
	35
	36	/*
	37	* Virtual device vector for mirroring.
	38	*/
	39
	40	typedef struct mirror_child {
	41	vdev_t *mc_vd;
	42	uint64_t mc_offset;
	43	int mc_error;
b128c09f BB	44	uint8_t mc_tried;
	45	uint8_t mc_skipped;
	46	uint8_t mc_speculative;
34dc7c2f BB	47	} mirror_child_t;
	48
	49	typedef struct mirror_map {
	50	int mm_children;
	51	int mm_replacing;
	52	int mm_preferred;
	53	int mm_root;
	54	mirror_child_t mm_child[1];
	55	} mirror_map_t;
	56
	57	int vdev_mirror_shift = 21;
	58
b128c09f BB	59	static void
	60	vdev_mirror_map_free(zio_t *zio)
	61	{
	62	mirror_map_t *mm = zio->io_vsd;
	63
	64	kmem_free(mm, offsetof(mirror_map_t, mm_child[mm->mm_children]));
	65	}
	66
428870ff BB	67	static const zio_vsd_ops_t vdev_mirror_vsd_ops = {
	68	vdev_mirror_map_free,
	69	zio_vsd_default_cksum_report
	70	};
	71
34dc7c2f BB	72	static mirror_map_t *
	73	vdev_mirror_map_alloc(zio_t *zio)
	74	{
	75	mirror_map_t *mm = NULL;
	76	mirror_child_t *mc;
	77	vdev_t *vd = zio->io_vd;
	78	int c, d;
	79
	80	if (vd == NULL) {
	81	dva_t *dva = zio->io_bp->blk_dva;
	82	spa_t *spa = zio->io_spa;
	83
	84	c = BP_GET_NDVAS(zio->io_bp);
	85
b8d06fca	86	mm = kmem_zalloc(offsetof(mirror_map_t, mm_child[c]), KM_PUSHPAGE);
34dc7c2f BB	87	mm->mm_children = c;
	88	mm->mm_replacing = B_FALSE;
	89	mm->mm_preferred = spa_get_random(c);
	90	mm->mm_root = B_TRUE;
	91
	92	/*
	93	* Check the other, lower-index DVAs to see if they're on
	94	* the same vdev as the child we picked. If they are, use
	95	* them since they are likely to have been allocated from
	96	* the primary metaslab in use at the time, and hence are
	97	* more likely to have locality with single-copy data.
	98	*/
	99	for (c = mm->mm_preferred, d = c - 1; d >= 0; d--) {
	100	if (DVA_GET_VDEV(&dva[d]) == DVA_GET_VDEV(&dva[c]))
	101	mm->mm_preferred = d;
	102	}
	103
	104	for (c = 0; c < mm->mm_children; c++) {
	105	mc = &mm->mm_child[c];
	106
	107	mc->mc_vd = vdev_lookup_top(spa, DVA_GET_VDEV(&dva[c]));
	108	mc->mc_offset = DVA_GET_OFFSET(&dva[c]);
	109	}
	110	} else {
	111	c = vd->vdev_children;
	112
b8d06fca	113	mm = kmem_zalloc(offsetof(mirror_map_t, mm_child[c]), KM_PUSHPAGE);
34dc7c2f BB	114	mm->mm_children = c;
	115	mm->mm_replacing = (vd->vdev_ops == &vdev_replacing_ops \|\|
	116	vd->vdev_ops == &vdev_spare_ops);
	117	mm->mm_preferred = mm->mm_replacing ? 0 :
	118	(zio->io_offset >> vdev_mirror_shift) % c;
	119	mm->mm_root = B_FALSE;
	120
	121	for (c = 0; c < mm->mm_children; c++) {
	122	mc = &mm->mm_child[c];
	123	mc->mc_vd = vd->vdev_child[c];
	124	mc->mc_offset = zio->io_offset;
	125	}
	126	}
	127
	128	zio->io_vsd = mm;
428870ff	129	zio->io_vsd_ops = &vdev_mirror_vsd_ops;
34dc7c2f BB	130	return (mm);
	131	}
	132
34dc7c2f	133	static int
1bd201e7 CS	134	vdev_mirror_open(vdev_t vd, uint64_t asize, uint64_t *max_asize,
1bd201e7 CS	135	uint64_t *ashift)
34dc7c2f	136	{
34dc7c2f	137	int numerrors = 0;
45d1cae3	138	int lasterror = 0;
d6320ddb	139	int c;
34dc7c2f BB	140
	141	if (vd->vdev_children == 0) {
	142	vd->vdev_stat.vs_aux = VDEV_AUX_BAD_LABEL;
	143	return (EINVAL);
	144	}
	145
45d1cae3	146	vdev_open_children(vd);
34dc7c2f	147
d6320ddb	148	for (c = 0; c < vd->vdev_children; c++) {
45d1cae3 BB	149	vdev_t *cvd = vd->vdev_child[c];
	150
	151	if (cvd->vdev_open_error) {
	152	lasterror = cvd->vdev_open_error;
34dc7c2f BB	153	numerrors++;
	154	continue;
	155	}
	156
	157	asize = MIN(asize - 1, cvd->vdev_asize - 1) + 1;
1bd201e7	158	max_asize = MIN(max_asize - 1, cvd->vdev_max_asize - 1) + 1;
34dc7c2f BB	159	ashift = MAX(ashift, cvd->vdev_ashift);
	160	}
	161
	162	if (numerrors == vd->vdev_children) {
	163	vd->vdev_stat.vs_aux = VDEV_AUX_NO_REPLICAS;
	164	return (lasterror);
	165	}
	166
	167	return (0);
	168	}
	169
	170	static void
	171	vdev_mirror_close(vdev_t *vd)
	172	{
d6320ddb BB	173	int c;
	174
	175	for (c = 0; c < vd->vdev_children; c++)
34dc7c2f BB	176	vdev_close(vd->vdev_child[c]);
	177	}
	178
	179	static void
	180	vdev_mirror_child_done(zio_t *zio)
	181	{
	182	mirror_child_t *mc = zio->io_private;
	183
	184	mc->mc_error = zio->io_error;
	185	mc->mc_tried = 1;
	186	mc->mc_skipped = 0;
	187	}
	188
	189	static void
	190	vdev_mirror_scrub_done(zio_t *zio)
	191	{
	192	mirror_child_t *mc = zio->io_private;
	193
	194	if (zio->io_error == 0) {
d164b209 BB	195	zio_t *pio;
	196
	197	mutex_enter(&zio->io_lock);
	198	while ((pio = zio_walk_parents(zio)) != NULL) {
	199	mutex_enter(&pio->io_lock);
	200	ASSERT3U(zio->io_size, >=, pio->io_size);
	201	bcopy(zio->io_data, pio->io_data, pio->io_size);
	202	mutex_exit(&pio->io_lock);
	203	}
	204	mutex_exit(&zio->io_lock);
34dc7c2f BB	205	}
	206
	207	zio_buf_free(zio->io_data, zio->io_size);
	208
	209	mc->mc_error = zio->io_error;
	210	mc->mc_tried = 1;
	211	mc->mc_skipped = 0;
	212	}
	213
34dc7c2f BB	214	/*
	215	* Try to find a child whose DTL doesn't contain the block we want to read.
	216	* If we can't, try the read on any vdev we haven't already tried.
	217	*/
	218	static int
	219	vdev_mirror_child_select(zio_t *zio)
	220	{
	221	mirror_map_t *mm = zio->io_vsd;
	222	mirror_child_t *mc;
	223	uint64_t txg = zio->io_txg;
	224	int i, c;
	225
428870ff	226	ASSERT(zio->io_bp == NULL \|\| BP_PHYSICAL_BIRTH(zio->io_bp) == txg);
34dc7c2f BB	227
	228	/*
	229	* Try to find a child whose DTL doesn't contain the block to read.
	230	* If a child is known to be completely inaccessible (indicated by
	231	* vdev_readable() returning B_FALSE), don't even try.
	232	*/
	233	for (i = 0, c = mm->mm_preferred; i < mm->mm_children; i++, c++) {
	234	if (c >= mm->mm_children)
	235	c = 0;
	236	mc = &mm->mm_child[c];
	237	if (mc->mc_tried \|\| mc->mc_skipped)
	238	continue;
b128c09f	239	if (!vdev_readable(mc->mc_vd)) {
34dc7c2f BB	240	mc->mc_error = ENXIO;
	241	mc->mc_tried = 1; /* don't even try */
	242	mc->mc_skipped = 1;
	243	continue;
	244	}
fb5f0bc8	245	if (!vdev_dtl_contains(mc->mc_vd, DTL_MISSING, txg, 1))
34dc7c2f BB	246	return (c);
	247	mc->mc_error = ESTALE;
	248	mc->mc_skipped = 1;
b128c09f	249	mc->mc_speculative = 1;
34dc7c2f BB	250	}
	251
	252	/*
	253	* Every device is either missing or has this txg in its DTL.
	254	* Look for any child we haven't already tried before giving up.
	255	*/
	256	for (c = 0; c < mm->mm_children; c++)
	257	if (!mm->mm_child[c].mc_tried)
	258	return (c);
	259
	260	/*
	261	* Every child failed. There's no place left to look.
	262	*/
	263	return (-1);
	264	}
	265
	266	static int
	267	vdev_mirror_io_start(zio_t *zio)
	268	{
	269	mirror_map_t *mm;
	270	mirror_child_t *mc;
	271	int c, children;
	272
	273	mm = vdev_mirror_map_alloc(zio);
	274
	275	if (zio->io_type == ZIO_TYPE_READ) {
	276	if ((zio->io_flags & ZIO_FLAG_SCRUB) && !mm->mm_replacing) {
	277	/*
	278	* For scrubbing reads we need to allocate a read
	279	* buffer for each child and issue reads to all
	280	* children. If any child succeeds, it will copy its
	281	* data into zio->io_data in vdev_mirror_scrub_done.
	282	*/
	283	for (c = 0; c < mm->mm_children; c++) {
	284	mc = &mm->mm_child[c];
	285	zio_nowait(zio_vdev_child_io(zio, zio->io_bp,
	286	mc->mc_vd, mc->mc_offset,
	287	zio_buf_alloc(zio->io_size), zio->io_size,
b128c09f	288	zio->io_type, zio->io_priority, 0,
34dc7c2f BB	289	vdev_mirror_scrub_done, mc));
34dc7c2f BB	290	}
b128c09f	291	return (ZIO_PIPELINE_CONTINUE);
34dc7c2f BB	292	}
	293	/*
	294	* For normal reads just pick one child.
	295	*/
	296	c = vdev_mirror_child_select(zio);
	297	children = (c >= 0);
	298	} else {
	299	ASSERT(zio->io_type == ZIO_TYPE_WRITE);
	300
	301	/*
fb5f0bc8	302	* Writes go to all children.
34dc7c2f	303	*/
fb5f0bc8 BB	304	c = 0;
fb5f0bc8 BB	305	children = mm->mm_children;
34dc7c2f BB	306	}
	307
	308	while (children--) {
	309	mc = &mm->mm_child[c];
	310	zio_nowait(zio_vdev_child_io(zio, zio->io_bp,
b128c09f BB	311	mc->mc_vd, mc->mc_offset, zio->io_data, zio->io_size,
	312	zio->io_type, zio->io_priority, 0,
	313	vdev_mirror_child_done, mc));
34dc7c2f BB	314	c++;
	315	}
	316
b128c09f	317	return (ZIO_PIPELINE_CONTINUE);
34dc7c2f BB	318	}
	319
	320	static int
b128c09f BB	321	vdev_mirror_worst_error(mirror_map_t *mm)
b128c09f BB	322	{
d6320ddb	323	int c, error[2] = { 0, 0 };
b128c09f	324
d6320ddb	325	for (c = 0; c < mm->mm_children; c++) {
b128c09f BB	326	mirror_child_t *mc = &mm->mm_child[c];
	327	int s = mc->mc_speculative;
	328	error[s] = zio_worst_error(error[s], mc->mc_error);
	329	}
	330
	331	return (error[0] ? error[0] : error[1]);
	332	}
	333
	334	static void
34dc7c2f BB	335	vdev_mirror_io_done(zio_t *zio)
	336	{
	337	mirror_map_t *mm = zio->io_vsd;
	338	mirror_child_t *mc;
	339	int c;
	340	int good_copies = 0;
	341	int unexpected_errors = 0;
	342
34dc7c2f BB	343	for (c = 0; c < mm->mm_children; c++) {
	344	mc = &mm->mm_child[c];
	345
34dc7c2f	346	if (mc->mc_error) {
34dc7c2f BB	347	if (!mc->mc_skipped)
34dc7c2f BB	348	unexpected_errors++;
b128c09f BB	349	} else if (mc->mc_tried) {
b128c09f BB	350	good_copies++;
34dc7c2f BB	351	}
	352	}
	353
	354	if (zio->io_type == ZIO_TYPE_WRITE) {
	355	/*
	356	* XXX -- for now, treat partial writes as success.
b128c09f BB	357	*
	358	* Now that we support write reallocation, it would be better
	359	* to treat partial failure as real failure unless there are
	360	* no non-degraded top-level vdevs left, and not update DTLs
	361	* if we intend to reallocate.
34dc7c2f BB	362	*/
34dc7c2f BB	363	/* XXPOLICY */
b128c09f BB	364	if (good_copies != mm->mm_children) {
	365	/*
	366	* Always require at least one good copy.
	367	*
	368	* For ditto blocks (io_vd == NULL), require
	369	* all copies to be good.
	370	*
	371	* XXX -- for replacing vdevs, there's no great answer.
	372	* If the old device is really dead, we may not even
	373	* be able to access it -- so we only want to
	374	* require good writes to the new device. But if
	375	* the new device turns out to be flaky, we want
	376	* to be able to detach it -- which requires all
	377	* writes to the old device to have succeeded.
	378	*/
	379	if (good_copies == 0 \|\| zio->io_vd == NULL)
	380	zio->io_error = vdev_mirror_worst_error(mm);
	381	}
	382	return;
34dc7c2f BB	383	}
	384
	385	ASSERT(zio->io_type == ZIO_TYPE_READ);
	386
	387	/*
	388	* If we don't have a good copy yet, keep trying other children.
	389	*/
	390	/* XXPOLICY */
	391	if (good_copies == 0 && (c = vdev_mirror_child_select(zio)) != -1) {
	392	ASSERT(c >= 0 && c < mm->mm_children);
	393	mc = &mm->mm_child[c];
34dc7c2f BB	394	zio_vdev_io_redone(zio);
	395	zio_nowait(zio_vdev_child_io(zio, zio->io_bp,
	396	mc->mc_vd, mc->mc_offset, zio->io_data, zio->io_size,
b128c09f	397	ZIO_TYPE_READ, zio->io_priority, 0,
34dc7c2f	398	vdev_mirror_child_done, mc));
b128c09f	399	return;
34dc7c2f BB	400	}
	401
	402	/* XXPOLICY */
b128c09f BB	403	if (good_copies == 0) {
b128c09f BB	404	zio->io_error = vdev_mirror_worst_error(mm);
34dc7c2f	405	ASSERT(zio->io_error != 0);
b128c09f	406	}
34dc7c2f	407
fb5f0bc8	408	if (good_copies && spa_writeable(zio->io_spa) &&
34dc7c2f BB	409	(unexpected_errors \|\|
	410	(zio->io_flags & ZIO_FLAG_RESILVER) \|\|
	411	((zio->io_flags & ZIO_FLAG_SCRUB) && mm->mm_replacing))) {
34dc7c2f BB	412	/*
34dc7c2f BB	413	* Use the good data we have in hand to repair damaged children.
34dc7c2f	414	*/
34dc7c2f BB	415	for (c = 0; c < mm->mm_children; c++) {
	416	/*
	417	* Don't rewrite known good children.
	418	* Not only is it unnecessary, it could
	419	* actually be harmful: if the system lost
	420	* power while rewriting the only good copy,
	421	* there would be no good copies left!
	422	*/
	423	mc = &mm->mm_child[c];
	424
	425	if (mc->mc_error == 0) {
	426	if (mc->mc_tried)
	427	continue;
	428	if (!(zio->io_flags & ZIO_FLAG_SCRUB) &&
fb5f0bc8	429	!vdev_dtl_contains(mc->mc_vd, DTL_PARTIAL,
34dc7c2f BB	430	zio->io_txg, 1))
	431	continue;
	432	mc->mc_error = ESTALE;
	433	}
	434
b128c09f BB	435	zio_nowait(zio_vdev_child_io(zio, zio->io_bp,
	436	mc->mc_vd, mc->mc_offset,
	437	zio->io_data, zio->io_size,
34dc7c2f	438	ZIO_TYPE_WRITE, zio->io_priority,
fb5f0bc8 BB	439	ZIO_FLAG_IO_REPAIR \| (unexpected_errors ?
fb5f0bc8 BB	440	ZIO_FLAG_SELF_HEAL : 0), NULL, NULL));
34dc7c2f	441	}
34dc7c2f	442	}
34dc7c2f BB	443	}
	444
	445	static void
	446	vdev_mirror_state_change(vdev_t *vd, int faulted, int degraded)
	447	{
	448	if (faulted == vd->vdev_children)
	449	vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
	450	VDEV_AUX_NO_REPLICAS);
	451	else if (degraded + faulted != 0)
	452	vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED, VDEV_AUX_NONE);
	453	else
	454	vdev_set_state(vd, B_FALSE, VDEV_STATE_HEALTHY, VDEV_AUX_NONE);
	455	}
	456
	457	vdev_ops_t vdev_mirror_ops = {
	458	vdev_mirror_open,
	459	vdev_mirror_close,
34dc7c2f BB	460	vdev_default_asize,
	461	vdev_mirror_io_start,
	462	vdev_mirror_io_done,
	463	vdev_mirror_state_change,
428870ff BB	464	NULL,
428870ff BB	465	NULL,
34dc7c2f BB	466	VDEV_TYPE_MIRROR, /* name of this vdev type */
	467	B_FALSE /* not a leaf vdev */
	468	};
	469
	470	vdev_ops_t vdev_replacing_ops = {
	471	vdev_mirror_open,
	472	vdev_mirror_close,
34dc7c2f BB	473	vdev_default_asize,
	474	vdev_mirror_io_start,
	475	vdev_mirror_io_done,
	476	vdev_mirror_state_change,
428870ff BB	477	NULL,
428870ff BB	478	NULL,
34dc7c2f BB	479	VDEV_TYPE_REPLACING, /* name of this vdev type */
	480	B_FALSE /* not a leaf vdev */
	481	};
	482
	483	vdev_ops_t vdev_spare_ops = {
	484	vdev_mirror_open,
	485	vdev_mirror_close,
34dc7c2f BB	486	vdev_default_asize,
	487	vdev_mirror_io_start,
	488	vdev_mirror_io_done,
	489	vdev_mirror_state_change,
428870ff BB	490	NULL,
428870ff BB	491	NULL,
34dc7c2f BB	492	VDEV_TYPE_SPARE, /* name of this vdev type */
	493	B_FALSE /* not a leaf vdev */
	494	};