/drivers/md/raid10.c

/*
 * raid10.c : Multiple Devices driver for Linux
 *
 * Copyright (C) 2000-2004 Neil Brown
 *
 * RAID-10 support for md.
 *
 * Base on code in raid1.c.  See raid1.c for further copyright information.
 *
 *
 * This program is free software; you can redistribute it and/or modify
 * it under the terms of the GNU General Public License as published by
 * the Free Software Foundation; either version 2, or (at your option)
 * any later version.
 *
 * You should have received a copy of the GNU General Public License
 * (for example /usr/src/linux/COPYING); if not, write to the Free
 * Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
 */

#include <linux/slab.h>
#include <linux/delay.h>
#include <linux/blkdev.h>
#include <linux/seq_file.h>
#include <linux/ratelimit.h>
#include "md.h"
#include "raid10.h"
#include "raid0.h"
#include "bitmap.h"

/*
 * RAID10 provides a combination of RAID0 and RAID1 functionality.
 * The layout of data is defined by
 *    chunk_size
 *    raid_disks
 *    near_copies (stored in low byte of layout)
 *    far_copies (stored in second byte of layout)
 *    far_offset (stored in bit 16 of layout )
 *
 * The data to be stored is divided into chunks using chunksize.
 * Each device is divided into far_copies sections.
 * In each section, chunks are laid out in a style similar to raid0, but
 * near_copies copies of each chunk is stored (each on a different drive).
 * The starting device for each section is offset near_copies from the starting
 * device of the previous section.
 * Thus they are (near_copies*far_copies) of each chunk, and each is on a different
 * drive.
 * near_copies and far_copies must be at least one, and their product is at most
 * raid_disks.
 *
 * If far_offset is true, then the far_copies are handled a bit differently.
 * The copies are still in different stripes, but instead of be very far apart
 * on disk, there are adjacent stripes.
 */

/*
 * Number of guaranteed r10bios in case of extreme VM load:
 */
#define	NR_RAID10_BIOS 256

static void allow_barrier(conf_t *conf);
static void lower_barrier(conf_t *conf);

static void * r10bio_pool_alloc(gfp_t gfp_flags, void *data)
{
	conf_t *conf = data;
	int size = offsetof(struct r10bio_s, devs[conf->copies]);

	/* allocate a r10bio with room for raid_disks entries in the bios array */
	return kzalloc(size, gfp_flags);
}

static void r10bio_pool_free(void *r10_bio, void *data)
{
	kfree(r10_bio);
}

/* Maximum size of each resync request */
#define RESYNC_BLOCK_SIZE (64*1024)
#define RESYNC_PAGES ((RESYNC_BLOCK_SIZE + PAGE_SIZE-1) / PAGE_SIZE)
/* amount of memory to reserve for resync requests */
#define RESYNC_WINDOW (1024*1024)
/* maximum number of concurrent requests, memory permitting */
#define RESYNC_DEPTH (32*1024*1024/RESYNC_BLOCK_SIZE)

/*
 * When performing a resync, we need to read and compare, so
 * we need as many pages are there are copies.
 * When performing a recovery, we need 2 bios, one for read,
 * one for write (we recover only one drive per r10buf)
 *
 */
static void * r10buf_pool_alloc(gfp_t gfp_flags, void *data)
{
	conf_t *conf = data;
	struct page *page;
	r10bio_t *r10_bio;
	struct bio *bio;
	int i, j;
	int nalloc;

	r10_bio = r10bio_pool_alloc(gfp_flags, conf);
	if (!r10_bio)
		return NULL;

	if (test_bit(MD_RECOVERY_SYNC, &conf->mddev->recovery))
		nalloc = conf->copies; /* resync */
	else
		nalloc = 2; /* recovery */

	/*
	 * Allocate bios.
	 */
	for (j = nalloc ; j-- ; ) {
		bio = bio_kmalloc(gfp_flags, RESYNC_PAGES);
		if (!bio)
			goto out_free_bio;
		r10_bio->devs[j].bio = bio;
	}
	/*
	 * Allocate RESYNC_PAGES data pages and attach them
	 * where needed.
	 */
	for (j = 0 ; j < nalloc; j++) {
		bio = r10_bio->devs[j].bio;
		for (i = 0; i < RESYNC_PAGES; i++) {
			if (j == 1 && !test_bit(MD_RECOVERY_SYNC,
						&conf->mddev->recovery)) {
				/* we can share bv_page's during recovery */
				struct bio *rbio = r10_bio->devs[0].bio;
				page = rbio->bi_io_vec[i].bv_page;
				get_page(page);
			} else
				page = alloc_page(gfp_flags);
			if (unlikely(!page))
				goto out_free_pages;

			bio->bi_io_vec[i].bv_page = page;
		}
	}

	return r10_bio;

out_free_pages:
	for ( ; i > 0 ; i--)
		safe_put_page(bio->bi_io_vec[i-1].bv_page);
	while (j--)
		for (i = 0; i < RESYNC_PAGES ; i++)
			safe_put_page(r10_bio->devs[j].bio->bi_io_vec[i].bv_page);
	j = -1;
out_free_bio:
	while ( ++j < nalloc )
		bio_put(r10_bio->devs[j].bio);
	r10bio_pool_free(r10_bio, conf);
	return NULL;
}

static void r10buf_pool_free(void *__r10_bio, void *data)
{
	int i;
	conf_t *conf = data;
	r10bio_t *r10bio = __r10_bio;
	int j;

	for (j=0; j < conf->copies; j++) {
		struct bio *bio = r10bio->devs[j].bio;
		if (bio) {
			for (i = 0; i < RESYNC_PAGES; i++) {
				safe_put_page(bio->bi_io_vec[i].bv_page);
				bio->bi_io_vec[i].bv_page = NULL;
			}
			bio_put(bio);
		}
	}
	r10bio_pool_free(r10bio, conf);
}

static void put_all_bios(conf_t *conf, r10bio_t *r10_bio)
{
	int i;

	for (i = 0; i < conf->copies; i++) {
		struct bio **bio = & r10_bio->devs[i].bio;
		if (!BIO_SPECIAL(*bio))
			bio_put(*bio);
		*bio = NULL;
	}
}

static void free_r10bio(r10bio_t *r10_bio)
{
	conf_t *conf = r10_bio->mddev->private;

	put_all_bios(conf, r10_bio);
	mempool_free(r10_bio, conf->r10bio_pool);
}

static void put_buf(r10bio_t *r10_bio)
{
	conf_t *conf = r10_bio->mddev->private;

	mempool_free(r10_bio, conf->r10buf_pool);

	lower_barrier(conf);
}

static void reschedule_retry(r10bio_t *r10_bio)
{
	unsigned long flags;
	mddev_t *mddev = r10_bio->mddev;
	conf_t *conf = mddev->private;

	spin_lock_irqsave(&conf->device_lock, flags);
	list_add(&r10_bio->retry_list, &conf->retry_list);
	conf->nr_queued ++;
	spin_unlock_irqrestore(&conf->device_lock, flags);

	/* wake up frozen array... */
	wake_up(&conf->wait_barrier);

	md_wakeup_thread(mddev->thread);
}

/*
 * raid_end_bio_io() is called when we have finished servicing a mirrored
 * operation and are ready to return a success/failure code to the buffer
 * cache layer.
 */
static void raid_end_bio_io(r10bio_t *r10_bio)
{
	struct bio *bio = r10_bio->master_bio;
	int done;
	conf_t *conf = r10_bio->mddev->private;

	if (bio->bi_phys_segments) {
		unsigned long flags;
		spin_lock_irqsave(&conf->device_lock, flags);
		bio->bi_phys_segments--;
		done = (bio->bi_phys_segments == 0);
		spin_unlock_irqrestore(&conf->device_lock, flags);
	} else
		done = 1;
	if (!test_bit(R10BIO_Uptodate, &r10_bio->state))
		clear_bit(BIO_UPTODATE, &bio->bi_flags);
	if (done) {
		bio_endio(bio, 0);
		/*
		 * Wake up any possible resync thread that waits for the device
		 * to go idle.
		 */
		allow_barrier(conf);
	}
	free_r10bio(r10_bio);
}

/*
 * Update disk head position estimator based on IRQ completion info.
 */
static inline void update_head_pos(int slot, r10bio_t *r10_bio)
{
	conf_t *conf = r10_bio->mddev->private;

	conf->mirrors[r10_bio->devs[slot].devnum].head_position =
		r10_bio->devs[slot].addr + (r10_bio->sectors);
}

/*
 * Find the disk number which triggered given bio
 */
static int find_bio_disk(conf_t *conf, r10bio_t *r10_bio,
			 struct bio *bio, int *slotp)
{
	int slot;

	for (slot = 0; slot < conf->copies; slot++)
		if (r10_bio->devs[slot].bio == bio)
			break;

	BUG_ON(slot == conf->copies);
	update_head_pos(slot, r10_bio);

	if (slotp)
		*slotp = slot;
	return r10_bio->devs[slot].devnum;
}

static void raid10_end_read_request(struct bio *bio, int error)
{
	int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
	r10bio_t *r10_bio = bio->bi_private;
	int slot, dev;
	conf_t *conf = r10_bio->mddev->private;


	slot = r10_bio->read_slot;
	dev = r10_bio->devs[slot].devnum;
	/*
	 * this branch is our 'one mirror IO has finished' event handler:
	 */
	update_head_pos(slot, r10_bio);

	if (uptodate) {
		/*
		 * Set R10BIO_Uptodate in our master bio, so that
		 * we will return a good error code to the higher
		 * levels even if IO on some other mirrored buffer fails.
		 *
		 * The 'master' represents the composite IO operation to
		 * user-side. So if something waits for IO, then it will
		 * wait for the 'master' bio.
		 */
		set_bit(R10BIO_Uptodate, &r10_bio->state);
		raid_end_bio_io(r10_bio);
		rdev_dec_pending(conf->mirrors[dev].rdev, conf->mddev);
	} else {
		/*
		 * oops, read error - keep the refcount on the rdev
		 */
		char b[BDEVNAME_SIZE];
		printk_ratelimited(KERN_ERR
				   "md/raid10:%s: %s: rescheduling sector %llu\n",
				   mdname(conf->mddev),
				   bdevname(conf->mirrors[dev].rdev->bdev, b),
				   (unsigned long long)r10_bio->sector);
		set_bit(R10BIO_ReadError, &r10_bio->state);
		reschedule_retry(r10_bio);
	}
}

static void close_write(r10bio_t *r10_bio)
{
	/* clear the bitmap if all writes complete successfully */
	bitmap_endwrite(r10_bio->mddev->bitmap, r10_bio->sector,
			r10_bio->sectors,
			!test_bit(R10BIO_Degraded, &r10_bio->state),
			0);
	md_write_end(r10_bio->mddev);
}

static void raid10_end_write_request(struct bio *bio, int error)
{
	int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
	r10bio_t *r10_bio = bio->bi_private;
	int dev;
	int dec_rdev = 1;
	conf_t *conf = r10_bio->mddev->private;
	int slot;

	dev = find_bio_disk(conf, r10_bio, bio, &slot);

	/*
	 * this branch is our 'one mirror IO has finished' event handler:
	 */
	if (!uptodate) {
		set_bit(WriteErrorSeen,	&conf->mirrors[dev].rdev->flags);
		set_bit(R10BIO_WriteError, &r10_bio->state);
		dec_rdev = 0;
	} else {
		/*
		 * Set R10BIO_Uptodate in our master bio, so that
		 * we will return a good error code for to the higher
		 * levels even if IO on some other mirrored buffer fails.
		 *
		 * The 'master' represents the composite IO operation to
		 * user-side. So if something waits for IO, then it will
		 * wait for the 'master' bio.
		 */
		sector_t first_bad;
		int bad_sectors;

		set_bit(R10BIO_Uptodate, &r10_bio->state);

		/* Maybe we can clear some bad blocks. */
		if (is_badblock(conf->mirrors[dev].rdev,
				r10_bio->devs[slot].addr,
				r10_bio->sectors,
				&first_bad, &bad_sectors)) {
			bio_put(bio);
			r10_bio->devs[slot].bio = IO_MADE_GOOD;
			dec_rdev = 0;
			set_bit(R10BIO_MadeGood, &r10_bio->state);
		}
	}

	/*
	 *
	 * Let's see if all mirrored write operations have finished
	 * already.
	 */
	if (atomic_dec_and_test(&r10_bio->remaining)) {
		if (test_bit(R10BIO_WriteError, &r10_bio->state))
			reschedule_retry(r10_bio);
		else {
			close_write(r10_bio);
			if (test_bit(R10BIO_MadeGood, &r10_bio->state))
				reschedule_retry(r10_bio);
			else
				raid_end_bio_io(r10_bio);
		}
	}
	if (dec_rdev)
		rdev_dec_pending(conf->mirrors[dev].rdev, conf->mddev);
}


/*
 * RAID10 layout manager
 * As well as the chunksize and raid_disks count, there are two
 * parameters: near_copies and far_copies.
 * near_copies * far_copies must be <= raid_disks.
 * Normally one of these will be 1.
 * If both are 1, we get raid0.
 * If near_copies == raid_disks, we get raid1.
 *
 * Chunks are laid out in raid0 style with near_copies copies of the
 * first chunk, followed by near_copies copies of the next chunk and
 * so on.
 * If far_copies > 1, then after 1/far_copies of the array has been assigned
 * as described above, we start again with a device offset of near_copies.
 * So we effectively have another copy of the whole array further down all
 * the drives, but with blocks on different drives.
 * With this layout, and block is never stored twice on the one device.
 *
 * raid10_find_phys finds the sector offset of a given virtual sector
 * on each device that it is on.
 *
 * raid10_find_virt does the reverse mapping, from a device and a
 * sector offset to a virtual address
 */

static void raid10_find_phys(conf_t *conf, r10bio_t *r10bio)
{
	int n,f;
	sector_t sector;
	sector_t chunk;
	sector_t stripe;
	int dev;

	int slot = 0;

	/* now calculate first sector/dev */
	chunk = r10bio->sector >> conf->chunk_shift;
	sector = r10bio->sector & conf->chunk_mask;

	chunk *= conf->near_copies;
	stripe = chunk;
	dev = sector_div(stripe, conf->raid_disks);
	if (conf->far_offset)
		stripe *= conf->far_copies;

	sector += stripe << conf->chunk_shift;

	/* and calculate all the others */
	for (n=0; n < conf->near_copies; n++) {
		int d = dev;
		sector_t s = sector;
		r10bio->devs[slot].addr = sector;
		r10bio->devs[slot].devnum = d;
		slot++;

		for (f = 1; f < conf->far_copies; f++) {
			d += conf->near_copies;
			if (d >= conf->raid_disks)
				d -= conf->raid_disks;
			s += conf->stride;
			r10bio->devs[slot].devnum = d;
			r10bio->devs[slot].addr = s;
			slot++;
		}
		dev++;
		if (dev >= conf->raid_disks) {
			dev = 0;
			sector += (conf->chunk_mask + 1);
		}
	}
	BUG_ON(slot != conf->copies);
}

static sector_t raid10_find_virt(conf_t *conf, sector_t sector, int dev)
{
	sector_t offset, chunk, vchunk;

	offset = sector & conf->chunk_mask;
	if (conf->far_offset) {
		int fc;
		chunk = sector >> conf->chunk_shift;
		fc = sector_div(chunk, conf->far_copies);
		dev -= fc * conf->near_copies;
		if (dev < 0)
			dev += conf->raid_disks;
	} else {
		while (sector >= conf->stride) {
			sector -= conf->stride;
			if (dev < conf->near_copies)
				dev += conf->raid_disks - conf->near_copies;
			else
				dev -= conf->near_copies;
		}
		chunk = sector >> conf->chunk_shift;
	}
	vchunk = chunk * conf->raid_disks + dev;
	sector_div(vchunk, conf->near_copies);
	return (vchunk << conf->chunk_shift) + offset;
}

/**
 *	raid10_mergeable_bvec -- tell bio layer if a two requests can be merged
 *	@q: request queue
 *	@bvm: properties of new bio
 *	@biovec: the request that could be merged to it.
 *
 *	Return amount of bytes we can accept at this offset
 *      If near_copies == raid_disk, there are no striping issues,
 *      but in that case, the function isn't called at all.
 */
static int raid10_mergeable_bvec(struct request_queue *q,
				 struct bvec_merge_data *bvm,
				 struct bio_vec *biovec)
{
	mddev_t *mddev = q->queuedata;
	sector_t sector = bvm->bi_sector + get_start_sect(bvm->bi_bdev);
	int max;
	unsigned int chunk_sectors = mddev->chunk_sectors;
	unsigned int bio_sectors = bvm->bi_size >> 9;

	max =  (chunk_sectors - ((sector & (chunk_sectors - 1)) + bio_sectors)) << 9;
	if (max < 0) max = 0; /* bio_add cannot handle a negative return */
	if (max <= biovec->bv_len && bio_sectors == 0)
		return biovec->bv_len;
	else
		return max;
}

/*
 * This routine returns the disk from which the requested read should
 * be done. There is a per-array 'next expected sequential IO' sector
 * number - if this matches on the next IO then we use the last disk.
 * There is also a per-disk 'last know head position' sector that is
 * maintained from IRQ contexts, both the normal and the resync IO
 * completion handlers update this position correctly. If there is no
 * perfect sequential match then we pick the disk whose head is closest.
 *
 * If there are 2 mirrors in the same 2 devices, performance degrades
 * because position is mirror, not device based.
 *
 * The rdev for the device selected will have nr_pending incremented.
 */

/*
 * FIXME: possibly should rethink readbalancing and do it differently
 * depending on near_copies / far_copies geometry.
 */
static int read_balance(conf_t *conf, r10bio_t *r10_bio, int *max_sectors)
{
	const sector_t this_sector = r10_bio->sector;
	int disk, slot;
	int sectors = r10_bio->sectors;
	int best_good_sectors;
	sector_t new_distance, best_dist;
	mdk_rdev_t *rdev;
	int do_balance;
	int best_slot;

	raid10_find_phys(conf, r10_bio);
	rcu_read_lock();
retry:
	sectors = r10_bio->sectors;
	best_slot = -1;
	best_dist = MaxSector;
	best_good_sectors = 0;
	do_balance = 1;
	/*
	 * Check if we can balance. We can balance on the whole
	 * device if no resync is going on (recovery is ok), or below
	 * the resync window. We take the first readable disk when
	 * above the resync window.
	 */
	if (conf->mddev->recovery_cp < MaxSector
	    && (this_sector + sectors >= conf->next_resync))
		do_balance = 0;

	for (slot = 0; slot < conf->copies ; slot++) {
		sector_t first_bad;
		int bad_sectors;
		sector_t dev_sector;

		if (r10_bio->devs[slot].bio == IO_BLOCKED)
			continue;
		disk = r10_bio->devs[slot].devnum;
		rdev = rcu_dereference(conf->mirrors[disk].rdev);
		if (rdev == NULL)
			continue;
		if (!test_bit(In_sync, &rdev->flags))
			continue;

		dev_sector = r10_bio->devs[slot].addr;
		if (is_badblock(rdev, dev_sector, sectors,
				&first_bad, &bad_sectors)) {
			if (best_dist < MaxSector)
				/* Already have a better slot */
				continue;
			if (first_bad <= dev_sector) {
				/* Cannot read here.  If this is the
				 * 'primary' device, then we must not read
				 * beyond 'bad_sectors' from another device.
				 */
				bad_sectors -= (dev_sector - first_bad);
				if (!do_balance && sectors > bad_sectors)
					sectors = bad_sectors;
				if (best_good_sectors > sectors)
					best_good_sectors = sectors;
			} else {
				sector_t good_sectors =
					first_bad - dev_sector;
				if (good_sectors > best_good_sectors) {
					best_good_sectors = good_sectors;
					best_slot = slot;
				}
				if (!do_balance)
					/* Must read from here */
					break;
			}
			continue;
		} else
			best_good_sectors = sectors;

		if (!do_balance)
			break;

		/* This optimisation is debatable, and completely destroys
		 * sequential read speed for 'far copies' arrays.  So only
		 * keep it for 'near' arrays, and review those later.
		 */
		if (conf->near_copies > 1 && !atomic_read(&rdev->nr_pending))
			break;

		/* for far > 1 always use the lowest address */
		if (conf->far_copies > 1)
			new_distance = r10_bio->devs[slot].addr;
		else
			new_distance = abs(r10_bio->devs[slot].addr -
					   conf->mirrors[disk].head_position);
		if (new_distance < best_dist) {
			best_dist = new_distance;
			best_slot = slot;
		}
	}
	if (slot == conf->copies)
		slot = best_slot;

	if (slot >= 0) {
		disk = r10_bio->devs[slot].devnum;
		rdev = rcu_dereference(conf->mirrors[disk].rdev);
		if (!rdev)
			goto retry;
		atomic_inc(&rdev->nr_pending);
		if (test_bit(Faulty, &rdev->flags)) {
			/* Cannot risk returning a device that failed
			 * before we inc'ed nr_pending
			 */
			rdev_dec_pending(rdev, conf->mddev);
			goto retry;
		}
		r10_bio->read_slot = slot;
	} else
		disk = -1;
	rcu_read_unlock();
	*max_sectors = best_good_sectors;

	return disk;
}

static int raid10_congested(void *data, int bits)
{
	mddev_t *mddev = data;
	conf_t *conf = mddev->private;
	int i, ret = 0;

	if (mddev_congested(mddev, bits))
		return 1;
	rcu_read_lock();
	for (i = 0; i < conf->raid_disks && ret == 0; i++) {
		mdk_rdev_t *rdev = rcu_dereference(conf->mirrors[i].rdev);
		if (rdev && !test_bit(Faulty, &rdev->flags)) {
			struct request_queue *q = bdev_get_queue(rdev->bdev);

			ret |= bdi_congested(&q->backing_dev_info, bits);
		}
	}
	rcu_read_unlock();
	return ret;
}

static void flush_pending_writes(conf_t *conf)
{
	/* Any writes that have been queued but are awaiting
	 * bitmap updates get flushed here.
	 */
	spin_lock_irq(&conf->device_lock);

	if (conf->pending_bio_list.head) {
		struct bio *bio;
		bio = bio_list_get(&conf->pending_bio_list);
		spin_unlock_irq(&conf->device_lock);
		/* flush any pending bitmap writes to disk
		 * before proceeding w/ I/O */
		bitmap_unplug(conf->mddev->bitmap);

		while (bio) { /* submit pending writes */
			struct bio *next = bio->bi_next;
			bio->bi_next = NULL;
			generic_make_request(bio);
			bio = next;
		}
	} else
		spin_unlock_irq(&conf->device_lock);
}

/* Barriers....
 * Sometimes we need to suspend IO while we do something else,
 * either some resync/recovery, or reconfigure the array.
 * To do this we raise a 'barrier'.
 * The 'barrier' is a counter that can be raised multiple times
 * to count how many activities are happening which preclude
 * normal IO.
 * We can only raise the barrier if there is no pending IO.
 * i.e. if nr_pending == 0.
 * We choose only to raise the barrier if no-one is waiting for the
 * barrier to go down.  This means that as soon as an IO request
 * is ready, no other operations which require a barrier will start
 * until the IO request has had a chance.
 *
 * So: regular IO calls 'wait_barrier'.  When that returns there
 *    is no backgroup IO happening,  It must arrange to call
 *    allow_barrier when it has finished its IO.
 * backgroup IO calls must call raise_barrier.  Once that returns
 *    there is no normal IO happeing.  It must arrange to call
 *    lower_barrier when the particular background IO completes.
 */

static void raise_barrier(conf_t *conf, int force)
{
	BUG_ON(force && !conf->barrier);
	spin_lock_irq(&conf->resync_lock);

	/* Wait until no block IO is waiting (unless 'force') */
	wait_event_lock_irq(conf->wait_barrier, force || !conf->nr_waiting,
			    conf->resync_lock, );

	/* block any new IO from starting */
	conf->barrier++;

	/* Now wait for all pending IO to complete */
	wait_event_lock_irq(conf->wait_barrier,
			    !conf->nr_pending && conf->barrier < RESYNC_DEPTH,
			    conf->resync_lock, );

	spin_unlock_irq(&conf->resync_lock);
}

static void lower_barrier(conf_t *conf)
{
	unsigned long flags;
	spin_lock_irqsave(&conf->resync_lock, flags);
	conf->barrier--;
	spin_unlock_irqrestore(&conf->resync_lock, flags);
	wake_up(&conf->wait_barrier);
}

static void wait_barrier(conf_t *conf)
{
	spin_lock_irq(&conf->resync_lock);
	if (conf->barrier) {
		conf->nr_waiting++;
		wait_event_lock_irq(conf->wait_barrier, !conf->barrier,
				    conf->resync_lock,
				    );
		conf->nr_waiting--;
	}
	conf->nr_pending++;
	spin_unlock_irq(&conf->resync_lock);
}

static void allow_barrier(conf_t *conf)
{
	unsigned long flags;
	spin_lock_irqsave(&conf->resync_lock, flags);
	conf->nr_pending--;
	spin_unlock_irqrestore(&conf->resync_lock, flags);
	wake_up(&conf->wait_barrier);
}

static void freeze_array(conf_t *conf)
{
	/* stop syncio and normal IO and wait for everything to
	 * go quiet.
	 * We increment barrier and nr_waiting, and then
	 * wait until nr_pending match nr_queued+1
	 * This is called in the context of one normal IO request
	 * that has failed. Thus any sync request that might be pending
	 * will be blocked by nr_pending, and we need to wait for
	 * pending IO requests to complete or be queued for re-try.
	 * Thus the number queued (nr_queued) plus this request (1)
	 * must match the number of pending IOs (nr_pending) before
	 * we continue.
	 */
	spin_lock_irq(&conf->resync_lock);
	conf->barrier++;
	conf->nr_waiting++;
	wait_event_lock_irq(conf->wait_barrier,
			    conf->nr_pending == conf->nr_queued+1,
			    conf->resync_lock,
			    flush_pending_writes(conf));

	spin_unlock_irq(&conf->resync_lock);
}

static void unfreeze_array(conf_t *conf)
{
	/* reverse the effect of the freeze */
	spin_lock_irq(&conf->resync_lock);
	conf->barrier--;
	conf->nr_waiting--;
	wake_up(&conf->wait_barrier);
	spin_unlock_irq(&conf->resync_lock);
}

static int make_request(mddev_t *mddev, struct bio * bio)
{
	conf_t *conf = mddev->private;
	mirror_info_t *mirror;
	r10bio_t *r10_bio;
	struct bio *read_bio;
	int i;
	int chunk_sects = conf->chunk_mask + 1;
	const int rw = bio_data_dir(bio);
	const unsigned long do_sync = (bio->bi_rw & REQ_SYNC);
	const unsigned long do_fua = (bio->bi_rw & REQ_FUA);
	unsigned long flags;
	mdk_rdev_t *blocked_rdev;
	int plugged;
	int sectors_handled;
	int max_sectors;

	if (unlikely(bio->bi_rw & REQ_FLUSH)) {
		md_flush_request(mddev, bio);
		return 0;
	}

	/* If this request crosses a chunk boundary, we need to
	 * split it.  This will only happen for 1 PAGE (or less) requests.
	 */
	if (unlikely( (bio->bi_sector & conf->chunk_mask) + (bio->bi_size >> 9)
		      > chunk_sects &&
		    conf->near_copies < conf->raid_disks)) {
		struct bio_pair *bp;
		/* Sanity check -- queue functions should prevent this happening */
		if (bio->bi_vcnt != 1 ||
		    bio->bi_idx != 0)
			goto bad_map;
		/* This is a one page bio that upper layers
		 * refuse to split for us, so we need to split it.
		 */
		bp = bio_split(bio,
			       chunk_sects - (bio->bi_sector & (chunk_sects - 1)) );

		/* Each of these 'make_request' calls will call 'wait_barrier'.
		 * If the first succeeds but the second blocks due to the resync
		 * thread raising the barrier, we will deadlock because the
		 * IO to the underlying device will be queued in generic_make_request
		 * and will never complete, so will never reduce nr_pending.
		 * So increment nr_waiting here so no new raise_barriers will
		 * succeed, and so the second wait_barrier cannot block.
		 */
		spin_lock_irq(&conf->resync_lock);
		conf->nr_waiting++;
		spin_unlock_irq(&conf->resync_lock);

		if (make_request(mddev, &bp->bio1))
			generic_make_request(&bp->bio1);
		if (make_request(mddev, &bp->bio2))
			generic_make_request(&bp->bio2);

		spin_lock_irq(&conf->resync_lock);
		conf->nr_waiting--;
		wake_up(&conf->wait_barrier);
		spin_unlock_irq(&conf->resync_lock);

		bio_pair_release(bp);
		return 0;
	bad_map:
		printk("md/raid10:%s: make_request bug: can't convert block across chunks"
		       " or bigger than %dk %llu %d\n", mdname(mddev), chunk_sects/2,
		       (unsigned long long)bio->bi_sector, bio->bi_size >> 10);

		bio_io_error(bio);
		return 0;
	}

	md_write_start(mddev, bio);

	/*
	 * Register the new request and wait if the reconstruction
	 * thread has put up a bar for new requests.
	 * Continue immediately if no resync is active currently.
	 */
	wait_barrier(conf);

	r10_bio = mempool_alloc(conf->r10bio_pool, GFP_NOIO);

	r10_bio->master_bio = bio;
	r10_bio->sectors = bio->bi_size >> 9;

	r10_bio->mddev = mddev;
	r10_bio->sector = bio->bi_sector;
	r10_bio->state = 0;

	/* We might need to issue multiple reads to different
	 * devices if there are bad blocks around, so we keep
	 * track of the number of reads in bio->bi_phys_segments.
	 * If this is 0, there is only one r10_bio and no locking
	 * will be needed when the request completes.  If it is
	 * non-zero, then it is the number of not-completed requests.
	 */
	bio->bi_phys_segments = 0;
	clear_bit(BIO_SEG_VALID, &bio->bi_flags);

	if (rw == READ) {
		/*
		 * read balancing logic:
		 */
		int disk;
		int slot;

read_again:
		disk = read_balance(conf, r10_bio, &max_sectors);
		slot = r10_bio->read_slot;
		if (disk < 0) {
			raid_end_bio_io(r10_bio);
			return 0;
		}
		mirror = conf->mirrors + disk;

		read_bio = bio_clone_mddev(bio, GFP_NOIO, mddev);
		md_trim_bio(read_bio, r10_bio->sector - bio->bi_sector,
			    max_sectors);

		r10_bio->devs[slot].bio = read_bio;

		read_bio->bi_sector = r10_bio->devs[slot].addr +
			mirror->rdev->data_offset;
		read_bio->bi_bdev = mirror->rdev->bdev;
		read_bio->bi_end_io = raid10_end_read_request;
		read_bio->bi_rw = READ | do_sync;
		read_bio->bi_private = r10_bio;

		if (max_sectors < r10_bio->sectors) {
			/* Could not read all from this device, so we will
			 * need another r10_bio.
			 */
			sectors_handled = (r10_bio->sectors + max_sectors
					   - bio->bi_sector);
			r10_bio->sectors = max_sectors;
			spin_lock_irq(&conf->device_lock);
			if (bio->bi_phys_segments == 0)
				bio->bi_phys_segments = 2;
			else
				bio->bi_phys_segments++;
			spin_unlock(&conf->device_lock);
			/* Cannot call generic_make_request directly
			 * as that will be queued in __generic_make_request
			 * and subsequent mempool_alloc might block
			 * waiting for it.  so hand bio over to raid10d.
			 */
			reschedule_retry(r10_bio);

			r10_bio = mempool_alloc(conf->r10bio_pool, GFP_NOIO);

			r10_bio->master_bio = bio;
			r10_bio->sectors = ((bio->bi_size >> 9)
					    - sectors_handled);
			r10_bio->state = 0;
			r10_bio->mddev = mddev;
			r10_bio->sector = bio->bi_sector + sectors_handled;
			goto read_again;
		} else
			generic_make_request(read_bio);
		return 0;
	}

	/*
	 * WRITE:
	 */
	/* first select target devices under rcu_lock and
	 * inc refcount on their rdev.  Record them by setting
	 * bios[x] to bio
	 * If there are known/acknowledged bad blocks on any device
	 * on which we have seen a write error, we want to avoid
	 * writing to those blocks.  This potentially requires several
	 * writes to write around the bad blocks.  Each set of writes
	 * gets its own r10_bio with a set of bios attached.  The number
	 * of r10_bios is recored in bio->bi_phys_segments just as with
	 * the read case.
	 */
	plugged = mddev_check_plugged(mddev);

	raid10_find_phys(conf, r10_bio);
retry_write:
	blocked_rdev = NULL;
	rcu_read_lock();
	max_sectors = r10_bio->sectors;

	for (i = 0;  i < conf->copies; i++) {
		int d = r10_bio->devs[i].devnum;
		mdk_rdev_t *rdev = rcu_dereference(conf->mirrors[d].rdev);
		if (rdev && unlikely(test_bit(Blocked, &rdev->flags))) {
			atomic_inc(&rdev->nr_pending);
			blocked_rdev = rdev;
			break;
		}
		r10_bio->devs[i].bio = NULL;
		if (!rdev || test_bit(Faulty, &rdev->flags)) {
			set_bit(R10BIO_Degraded, &r10_bio->state);
			continue;
		}
		if (test_bit(WriteErrorSeen, &rdev->flags)) {
			sector_t first_bad;
			sector_t dev_sector = r10_bio->devs[i].addr;
			int bad_sectors;
			int is_bad;

			is_bad = is_badblock(rdev, dev_sector,
					     max_sectors,
					     &first_bad, &bad_sectors);
			if (is_bad < 0) {
				/* Mustn't write here until the bad block
				 * is acknowledged
				 */
				atomic_inc(&rdev->nr_pending);
				set_bit(BlockedBadBlocks, &rdev->flags);
				blocked_rdev = rdev;
				break;
			}
			if (is_bad && first_bad <= dev_sector) {
				/* Cannot write here at all */
				bad_sectors -= (dev_sector - first_bad);
				if (bad_sectors < max_sectors)
					/* Mustn't write more than bad_sectors
					 * to other devices yet
					 */
					max_sectors = bad_sectors;
				/* We don't set R10BIO_Degraded as that
				 * only applies if the disk is missing,
				 * so it might be re-added, and we want to
				 * know to recover this chunk.
				 * In this case the device is here, and the
				 * fact that this chunk is not in-sync is
				 * recorded in the bad block log.
				 */
				continue;
			}
			if (is_bad) {
				int good_sectors = first_bad - dev_sector;
				if (good_sectors < max_sectors)
					max_sectors = good_sectors;
			}
		}
		r10_bio->devs[i].bio = bio;
		atomic_inc(&rdev->nr_pending);
	}
	rcu_read_unlock();

	if (unlikely(blocked_rdev)) {
		/* Have to wait for this device to get unblocked, then retry */
		int j;
		int d;

		for (j = 0; j < i; j++)
			if (r10_bio->devs[j].bio) {
				d = r10_bio->devs[j].devnum;
				rdev_dec_pending(conf->mirrors[d].rdev, mddev);
			}
		allow_barrier(conf);
		md_wait_for_blocked_rdev(blocked_rdev, mddev);
		wait_barrier(conf);
		goto retry_write;
	}

	if (max_sectors < r10_bio->sectors) {
		/* We are splitting this into multiple parts, so
		 * we need to prepare for allocating another r10_bio.
		 */
		r10_bio->sectors = max_sectors;
		spin_lock_irq(&conf->device_lock);
		if (bio->bi_phys_segments == 0)
			bio->bi_phys_segments = 2;
		else
			bio->bi_phys_segments++;
		spin_unlock_irq(&conf->device_lock);
	}
	sectors_handled = r10_bio->sector + max_sectors - bio->bi_sector;

	atomic_set(&r10_bio->remaining, 1);
	bitmap_startwrite(mddev->bitmap, r10_bio->sector, r10_bio->sectors, 0);

	for (i = 0; i < conf->copies; i++) {
		struct bio *mbio;
		int d = r10_bio->devs[i].devnum;
		if (!r10_bio->devs[i].bio)
			continue;

		mbio = bio_clone_mddev(bio, GFP_NOIO, mddev);
		md_trim_bio(mbio, r10_bio->sector - bio->bi_sector,
			    max_sectors);
		r10_bio->devs[i].bio = mbio;

		mbio->bi_sector	= (r10_bio->devs[i].addr+
				   conf->mirrors[d].rdev->data_offset);
		mbio->bi_bdev = conf->mirrors[d].rdev->bdev;
		mbio->bi_end_io	= raid10_end_write_request;
		mbio->bi_rw = WRITE | do_sync | do_fua;
		mbio->bi_private = r10_bio;

		atomic_inc(&r10_bio->remaining);
		spin_lock_irqsave(&conf->device_lock, flags);
		bio_list_add(&conf->pending_bio_list, mbio);
		spin_unlock_irqrestore(&conf->device_lock, flags);
	}

	if (atomic_dec_and_test(&r10_bio->remaining)) {
		/* This matches the end of raid10_end_write_request() */
		bitmap_endwrite(r10_bio->mddev->bitmap, r10_bio->sector,
				r10_bio->sectors,
				!test_bit(R10BIO_Degraded, &r10_bio->state),
				0);
		md_write_end(mddev);
		raid_end_bio_io(r10_bio);
	}

	/* In case raid10d snuck in to freeze_array */
	wake_up(&conf->wait_barrier);

	if (sectors_handled < (bio->bi_size >> 9)) {
		/* We need another r10_bio.  It has already been counted
		 * in bio->bi_phys_segments.
		 */
		r10_bio = mempool_alloc(conf->r10bio_pool, GFP_NOIO);

		r10_bio->master_bio = bio;
		r10_bio->sectors = (bio->bi_size >> 9) - sectors_handled;

		r10_bio->mddev = mddev;
		r10_bio->sector = bio->bi_sector + sectors_handled;
		r10_bio->state = 0;
		goto retry_write;
	}

	if (do_sync || !mddev->bitmap || !plugged)
		md_wakeup_thread(mddev->thread);
	return 0;
}

static void status(struct seq_file *seq, mddev_t *mddev)
{
	conf_t *conf = mddev->private;
	int i;

	if (conf->near_copies < conf->raid_disks)
		seq_printf(seq, " %dK chunks", mddev->chunk_sectors / 2);
	if (conf->near_copies > 1)
		seq_printf(seq, " %d near-copies", conf->near_copies);
	if (conf->far_copies > 1) {
		if (conf->far_offset)
			seq_printf(seq, " %d offset-copies", conf->far_copies);
		else
			seq_printf(seq, " %d far-copies", conf->far_copies);
	}
	seq_printf(seq, " [%d/%d] [", conf->raid_disks,
					conf->raid_disks - mddev->degraded);
	for (i = 0; i < conf->raid_disks; i++)
		seq_printf(seq, "%s",
			      conf->mirrors[i].rdev &&
			      test_bit(In_sync, &conf->mirrors[i].rdev->flags) ? "U" : "_");
	seq_printf(seq, "]");
}

/* check if there are enough drives for
 * every block to appear on atleast one.
 * Don't consider the device numbered 'ignore'
 * as we might be about to remove it.
 */
static int enough(conf_t *conf, int ignore)
{
	int first = 0;

	do {
		int n = conf->copies;
		int cnt = 0;
		while (n--) {
			if (conf->mirrors[first].rdev &&
			    first != ignore)
				cnt++;
			first = (first+1) % conf->raid_disks;
		}
		if (cnt == 0)
			return 0;
	} while (first != 0);
	return 1;
}

static void error(mddev_t *mddev, mdk_rdev_t *rdev)
{
	char b[BDEVNAME_SIZE];
	conf_t *conf = mddev->private;

	/*
	 * If it is not operational, then we have already marked it as dead
	 * else if it is the last working disks, ignore the error, let the
	 * next level up know.
	 * else mark the drive as failed
	 */
	if (test_bit(In_sync, &rdev->flags)
	    && !enough(conf, rdev->raid_disk))
		/*
		 * Don't fail the drive, just return an IO error.
		 */
		return;
	if (test_and_clear_bit(In_sync, &rdev->flags)) {
		unsigned long flags;
		spin_lock_irqsave(&conf->device_lock, flags);
		mddev->degraded++;
		spin_unlock_irqrestore(&conf->device_lock, flags);
		/*
		 * if recovery is running, make sure it aborts.
		 */
		set_bit(MD_RECOVERY_INTR, &mddev->recovery);
	}
	set_bit(Blocked, &rdev->flags);
	set_bit(Faulty, &rdev->flags);
	set_bit(MD_CHANGE_DEVS, &mddev->flags);
	printk(KERN_ALERT
	       "md/raid10:%s: Disk failure on %s, disabling device.\n"
	       "md/raid10:%s: Operation continuing on %d devices.\n",
	       mdname(mddev), bdevname(rdev->bdev, b),
	       mdname(mddev), conf->raid_disks - mddev->degraded);
}

static void print_conf(conf_t *conf)
{
	int i;
	mirror_info_t *tmp;

	printk(KERN_DEBUG "RAID10 conf printout:\n");
	if (!conf) {
		printk(KERN_DEBUG "(!conf)\n");
		return;
	}
	printk(KERN_DEBUG " --- wd:%d rd:%d\n", conf->raid_disks - conf->mddev->degraded,
		conf->raid_disks);

	for (i = 0; i < conf->raid_disks; i++) {
		char b[BDEVNAME_SIZE];
		tmp = conf->mirrors + i;
		if (tmp->rdev)
			printk(KERN_DEBUG " disk %d, wo:%d, o:%d, dev:%s\n",
				i, !test_bit(In_sync, &tmp->rdev->flags),
			        !test_bit(Faulty, &tmp->rdev->flags),
				bdevname(tmp->rdev->bdev,b));
	}
}

static void close_sync(conf_t *conf)
{
	wait_barrier(conf);
	allow_barrier(conf);

	mempool_destroy(conf->r10buf_pool);
	conf->r10buf_pool = NULL;
}

static int raid10_spare_active(mddev_t *mddev)
{
	int i;
	conf_t *conf = mddev->private;
	mirror_info_t *tmp;
	int count = 0;
	unsigned long flags;

	/*
	 * Find all non-in_sync disks within the RAID10 configuration
	 * and mark them in_sync
	 */
	for (i = 0; i < conf->raid_disks; i++) {
		tmp = conf->mirrors + i;
		if (tmp->rdev
		    && !test_bit(Faulty, &tmp->rdev->flags)
		    && !test_and_set_bit(In_sync, &tmp->rdev->flags)) {
			count++;
			sysfs_notify_dirent(tmp->rdev->sysfs_state);
		}
	}
	spin_lock_irqsave(&conf->device_lock, flags);
	mddev->degraded -= count;
	spin_unlock_irqrestore(&conf->device_lock, flags);

	print_conf(conf);
	return count;
}


static int raid10_add_disk(mddev_t *mddev, mdk_rdev_t *rdev)
{
	conf_t *conf = mddev->private;
	int err = -EEXIST;
	int mirror;
	int first = 0;
	int last = conf->raid_disks - 1;

	if (mddev->recovery_cp < MaxSector)
		/* only hot-add to in-sync arrays, as recovery is
		 * very different from resync
		 */
		return -EBUSY;
	if (!enough(conf, -1))
		return -EINVAL;

	if (rdev->raid_disk >= 0)
		first = last = rdev->raid_disk;

	if (rdev->saved_raid_disk >= first &&
	    conf->mirrors[rdev->saved_raid_disk].rdev == NULL)
		mirror = rdev->saved_raid_disk;
	else
		mirror = first;
	for ( ; mirror <= last ; mirror++) {
		mirror_info_t *p = &conf->mirrors[mirror];
		if (p->recovery_disabled == mddev->recovery_disabled)
			continue;
		if (!p->rdev)
			continue;

		disk_stack_limits(mddev->gendisk, rdev->bdev,
				  rdev->data_offset << 9);
		/* as we don't honour merge_bvec_fn, we must
		 * never risk violating it, so limit
		 * ->max_segments to one lying with a single
		 * page, as a one page request is never in
		 * violation.
		 */
		if (rdev->bdev->bd_disk->queue->merge_bvec_fn) {
			blk_queue_max_segments(mddev->queue, 1);
			blk_queue_segment_boundary(mddev->queue,
						   PAGE_CACHE_SIZE - 1);
		}

		p->head_position = 0;
		rdev->raid_disk = mirror;
		err = 0;
		if (rdev->saved_raid_disk != mirror)
			conf->fullsync = 1;
		rcu_assign_pointer(p->rdev, rdev);
		break;
	}

	md_integrity_add_rdev(rdev, mddev);
	print_conf(conf);
	return err;
}

static int raid10_remove_disk(mddev_t *mddev, int number)
{
	conf_t *conf = mddev->private;
	int err = 0;
	mdk_rdev_t *rdev;
	mirror_info_t *p = conf->mirrors+ number;

	print_conf(conf);
	rdev = p->rdev;
	if (rdev) {
		if (test_bit(In_sync, &rdev->flags) ||
		    atomic_read(&rdev->nr_pending)) {
			err = -EBUSY;
			goto abort;
		}
		/* Only remove faulty devices in recovery
		 * is not possible.
		 */
		if (!test_bit(Faulty, &rdev->flags) &&
		    mddev->recovery_disabled != p->recovery_disabled &&
		    enough(conf, -1)) {
			err = -EBUSY;
			goto abort;
		}
		p->rdev = NULL;
		synchronize_rcu();
		if (atomic_read(&rdev->nr_pending)) {
			/* lost the race, try later */
			err = -EBUSY;
			p->rdev = rdev;
			goto abort;
		}
		err = md_integrity_register(mddev);
	}
abort:

	print_conf(conf);
	return err;
}


static void end_sync_read(struct bio *bio, int error)
{
	r10bio_t *r10_bio = bio->bi_private;
	conf_t *conf = r10_bio->mddev->private;
	int d;

	d = find_bio_disk(conf, r10_bio, bio, NULL);

	if (test_bit(BIO_UPTODATE, &bio->bi_flags))
		set_bit(R10BIO_Uptodate, &r10_bio->state);
	else
		/* The write handler will notice the lack of
		 * R10BIO_Uptodate and record any errors etc
		 */
		atomic_add(r10_bio->sectors,
			   &conf->mirrors[d].rdev->corrected_errors);

	/* for reconstruct, we always reschedule after a read.
	 * for resync, only after all reads
	 */
	rdev_dec_pending(conf->mirrors[d].rdev, conf->mddev);
	if (test_bit(R10BIO_IsRecover, &r10_bio->state) ||
	    atomic_dec_and_test(&r10_bio->remaining)) {
		/* we have read all the blocks,
		 * do the comparison in process context in raid10d
		 */
		reschedule_retry(r10_bio);
	}
}

static void end_sync_request(r10bio_t *r10_bio)
{
	mddev_t *mddev = r10_bio->mddev;

	while (atomic_dec_and_test(&r10_bio->remaining)) {
		if (r10_bio->master_bio == NULL) {
			/* the primary of several recovery bios */
			sector_t s = r10_bio->sectors;
			if (test_bit(R10BIO_MadeGood, &r10_bio->state) ||
			    test_bit(R10BIO_WriteError, &r10_bio->state))
				reschedule_retry(r10_bio);
			else
				put_buf(r10_bio);
			md_done_sync(mddev, s, 1);
			break;
		} else {
			r10bio_t *r10_bio2 = (r10bio_t *)r10_bio->master_bio;
			if (test_bit(R10BIO_MadeGood, &r10_bio->state) ||
			    test_bit(R10BIO_WriteError, &r10_bio->state))
				reschedule_retry(r10_bio);
			else
				put_buf(r10_bio);
			r10_bio = r10_bio2;
		}
	}
}

static void end_sync_write(struct bio *bio, int error)
{
	int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
	r10bio_t *r10_bio = bio->bi_private;
	mddev_t *mddev = r10_bio->mddev;
	conf_t *conf = mddev->private;
	int d;
	sector_t first_bad;
	int bad_sectors;
	int slot;

	d = find_bio_disk(conf, r10_bio, bio, &slot);

	if (!uptodate) {
		set_bit(WriteErrorSeen, &conf->mirrors[d].rdev->flags);
		set_bit(R10BIO_WriteError, &r10_bio->state);
	} else if (is_badblock(conf->mirrors[d].rdev,
			     r10_bio->devs[slot].addr,
			     r10_bio->sectors,
			     &first_bad, &bad_sectors))
		set_bit(R10BIO_MadeGood, &r10_bio->state);

	rdev_dec_pending(conf->mirrors[d].rdev, mddev);

	end_sync_request(r10_bio);
}

/*
 * Note: sync and recover and handled very differently for raid10
 * This code is for resync.
 * For resync, we read through virtual addresses and read all blocks.
 * If there is any error, we schedule a write.  The lowest numbered
 * drive is authoritative.
 * However requests come for physical address, so we need to map.
 * For every physical address there are raid_disks/copies virtual addresses,
 * which is always are least one, but is not necessarly an integer.
 * This means that a physical address can span multiple chunks, so we may
 * have to submit multiple io requests for a single sync request.
 */
/*
 * We check if all blocks are in-sync and only write to blocks that
 * aren't in sync
 */
static void sync_request_write(mddev_t *mddev, r10bio_t *r10_bio)
{
	conf_t *conf = mddev->private;
	int i, first;
	struct bio *tbio, *fbio;

	atomic_set(&r10_bio->remaining, 1);

	/* find the first device with a block */
	for (i=0; i<conf->copies; i++)
		if (test_bit(BIO_UPTODATE, &r10_bio->devs[i].bio->bi_flags))
			break;

	if (i == conf->copies)
		goto done;

	first = i;
	fbio = r10_bio->devs[i].bio;

	/* now find blocks with errors */
	for (i=0 ; i < conf->copies ; i++) {
		int  j, d;
		int vcnt = r10_bio->sectors >> (PAGE_SHIFT-9);

		tbio = r10_bio->devs[i].bio;

		if (tbio->bi_end_io != end_sync_read)
			continue;
		if (i == first)
			continue;
		if (test_bit(BIO_UPTODATE, &r10_bio->devs[i].bio->bi_flags)) {
			/* We know that the bi_io_vec layout is the same for
			 * both 'first' and 'i', so we just compare them.
			 * All vec entries are PAGE_SIZE;
			 */
			for (j = 0; j < vcnt; j++)
				if (memcmp(page_address(fbio->bi_io_vec[j].bv_page),
					   page_address(tbio->bi_io_vec[j].bv_page),
					   PAGE_SIZE))
					break;
			if (j == vcnt)
				continue;
			mddev->resync_mismatches += r10_bio->sectors;
			if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
				/* Don't fix anything. */
				continue;
		}
		/* Ok, we need to write this bio, either to correct an
		 * inconsistency or to correct an unreadable block.
		 * First we need to fixup bv_offset, bv_len and
		 * bi_vecs, as the read request might have corrupted these
		 */
		tbio->bi_vcnt = vcnt;
		tbio->bi_size = r10_bio->sectors << 9;
		tbio->bi_idx = 0;
		tbio->bi_phys_segments = 0;
		tbio->bi_flags &= ~(BIO_POOL_MASK - 1);
		tbio->bi_flags |= 1 << BIO_UPTODATE;
		tbio->bi_next = NULL;
		tbio->bi_rw = WRITE;
		tbio->bi_private = r10_bio;
		tbio->bi_sector = r10_bio->devs[i].addr;

		for (j=0; j < vcnt ; j++) {
			tbio->bi_io_vec[j].bv_offset = 0;
			tbio->bi_io_vec[j].bv_len = PAGE_SIZE;

			memcpy(page_address(tbio->bi_io_vec[j].bv_page),
			       page_address(fbio->bi_io_vec[j].bv_page),
			       PAGE_SIZE);
		}
		tbio->bi_end_io = end_sync_write;

		d = r10_bio->devs[i].devnum;
		atomic_inc(&conf->mirrors[d].rdev->nr_pending);
		atomic_inc(&r10_bio->remaining);
		md_sync_acct(conf->mirrors[d].rdev->bdev, tbio->bi_size >> 9);

		tbio->bi_sector += conf->mirrors[d].rdev->data_offset;
		tbio->bi_bdev = conf->mirrors[d].rdev->bdev;
		generic_make_request(tbio);
	}

done:
	if (atomic_dec_and_test(&r10_bio->remaining)) {
		md_done_sync(mddev, r10_bio->sectors, 1);
		put_buf(r10_bio);
	}
}

/*
 * Now for the recovery code.
 * Recovery happens across physical sectors.
 * We recover all non-is_sync drives by finding the virtual address of
 * each, and then choose a working drive that also has that virt address.
 * There is a separate r10_bio for each non-in_sync drive.
 * Only the first two slots are in use. The first for reading,
 * The second for writing.
 *
 */
static void fix_recovery_read_error(r10bio_t *r10_bio)
{
	/* We got a read error during recovery.
	 * We repeat the read in smaller page-sized sections.
	 * If a read succeeds, write it to the new device or record
	 * a bad block if we cannot.
	 * If a read fails, record a bad block on both old and
	 * new devices.
	 */
	mddev_t *mddev = r10_bio->mddev;
	conf_t *conf = mddev->private;
	struct bio *bio = r10_bio->devs[0].bio;
	sector_t sect = 0;
	int sectors = r10_bio->sectors;
	int idx = 0;
	int dr = r10_bio->devs[0].devnum;
	int dw = r10_bio->devs[1].devnum;

	while (sectors) {
		int s = sectors;
		mdk_rdev_t *rdev;
		sector_t addr;
		int ok;

		if (s > (PAGE_SIZE>>9))
			s = PAGE_SIZE >> 9;

		rdev = conf->mirrors[dr].rdev;
		addr = r10_bio->devs[0].addr + sect,
		ok = sync_page_io(rdev,
				  addr,
				  s << 9,
				  bio->bi_io_vec[idx].bv_page,
				  READ, false);
		if (ok) {
			rdev = conf->mirrors[dw].rdev;
			addr = r10_bio->devs[1].addr + sect;
			ok = sync_page_io(rdev,
					  addr,
					  s << 9,
					  bio->bi_io_vec[idx].bv_page,
					  WRITE, false);
			if (!ok)
				set_bit(WriteErrorSeen, &rdev->flags);
		}
		if (!ok) {
			/* We don't worry if we cannot set a bad block -
			 * it really is bad so there is no loss in not
			 * recording it yet
			 */
			rdev_set_badblocks(rdev, addr, s, 0);

			if (rdev != conf->mirrors[dw].rdev) {
				/* need bad block on destination too */
				mdk_rdev_t *rdev2 = conf->mirrors[dw].rdev;
				addr = r10_bio->devs[1].addr + sect;
				ok = rdev_set_badblocks(rdev2, addr, s, 0);
				if (!ok) {
					/* just abort the recovery */
					printk(KERN_NOTICE
					       "md/raid10:%s: recovery aborted"
					       " due to read error\n",
					       mdname(mddev));

					conf->mirrors[dw].recovery_disabled
						= mddev->recovery_disabled;
					set_bit(MD_RECOVERY_INTR,
						&mddev->recovery);
					break;
				}
			}
		}

		sectors -= s;
		sect += s;
		idx++;
	}
}

static void recovery_request_write(mddev_t *mddev, r10bio_t *r10_bio)
{
	conf_t *conf = mddev->private;
	int d;
	struct bio *wbio;

	if (!test_bit(R10BIO_Uptodate, &r10_bio->state)) {
		fix_recovery_read_error(r10_bio);
		end_sync_request(r10_bio);
		return;
	}

	/*
	 * share the pages with the first bio
	 * and submit the write request
	 */
	wbio = r10_bio->devs[1].bio;
	d = r10_bio->devs[1].devnum;

	atomic_inc(&conf->mirrors[d].rdev->nr_pending);
	md_sync_acct(conf->mirrors[d].rdev->bdev, wbio->bi_size >> 9);
	generic_make_request(wbio);
}


/*
 * Used by fix_read_error() to decay the per rdev read_errors.
 * We halve the read error count for every hour that has elapsed
 * since the last recorded read error.
 *
 */
static void check_decay_read_errors(mddev_t *mddev, mdk_rdev_t *rdev)
{
	struct timespec cur_time_mon;
	unsigned long hours_since_last;
	unsigned int read_errors = atomic_read(&rdev->read_errors);

	ktime_get_ts(&cur_time_mon);

	if (rdev->last_read_error.tv_sec == 0 &&
	    rdev->last_read_error.tv_nsec == 0) {
		/* first time we've seen a read error */
		rdev->last_read_error = cur_time_mon;
		return;
	}

	hours_since_last = (cur_time_mon.tv_sec -
			    rdev->last_read_error.tv_sec) / 3600;

	rdev->last_read_error = cur_time_mon;

	/*
	 * if hours_since_last is > the number of bits in read_errors
	 * just set read errors to 0. We do this to avoid
	 * overflowing the shift of read_errors by hours_since_last.
	 */
	if (hours_since_last >= 8 * sizeof(read_errors))
		atomic_set(&rdev->read_errors, 0);
	else
		atomic_set(&rdev->read_errors, read_errors >> hours_since_last);
}

static int r10_sync_page_io(mdk_rdev_t *rdev, sector_t sector,
			    int sectors, struct page *page, int rw)
{
	sector_t first_bad;
	int bad_sectors;

	if (is_badblock(rdev, sector, sectors, &first_bad, &bad_sectors)
	    && (rw == READ || test_bit(WriteErrorSeen, &rdev->flags)))
		return -1;
	if (sync_page_io(rdev, sector, sectors << 9, page, rw, false))
		/* success */
		return 1;
	if (rw == WRITE)
		set_bit(WriteErrorSeen, &rdev->flags);
	/* need to record an error - either for the block or the device */
	if (!rdev_set_badblocks(rdev, sector, sectors, 0))
		md_error(rdev->mddev, rdev);
	return 0;
}

/*
 * This is a kernel thread which:
 *
 *	1.	Retries failed read operations on working mirrors.
 *	2.	Updates the raid superblock when problems encounter.
 *	3.	Performs writes following reads for array synchronising.
 */

static void fix_read_error(conf_t *conf, mddev_t *mddev, r10bio_t *r10_bio)
{
	int sect = 0; /* Offset from r10_bio->sector */
	int sectors = r10_bio->sectors;
	mdk_rdev_t*rdev;
	int max_read_errors = atomic_read(&mddev->max_corr_read_errors);
	int d = r10_bio->devs[r10_bio->read_slot].devnum;

	/* still own a reference to this rdev, so it cannot
	 * have been cleared recently.
	 */
	rdev = conf->mirrors[d].rdev;

	if (test_bit(Faulty, &rdev->flags))
		/* drive has already been failed, just ignore any
		   more fix_read_error() attempts */
		return;

	check_decay_read_errors(mddev, rdev);
	atomic_inc(&rdev->read_errors);
	if (atomic_read(&rdev->read_errors) > max_read_errors) {
		char b[BDEVNAME_SIZE];
		bdevname(rdev->bdev, b);

		printk(KERN_NOTICE
		       "md/raid10:%s: %s: Raid device exceeded "
		       "read_error threshold [cur %d:max %d]\n",
		       mdname(mddev), b,
		       atomic_read(&rdev->read_errors), max_read_errors);
		printk(KERN_NOTICE
		       "md/raid10:%s: %s: Failing raid device\n",
		       mdname(mddev), b);
		md_err