dm-raid1.c

linux 内核源代码

	read_lock(&rh->hash_lock);
	reg = __rh_lookup(rh, region);
	read_unlock(&rh->hash_lock);

	spin_lock_irqsave(&rh->region_lock, flags);
	if (atomic_dec_and_test(&reg->pending)) {
		/*
		 * There is no pending I/O for this region.
		 * We can move the region to corresponding list for next action.
		 * At this point, the region is not yet connected to any list.
		 *
		 * If the state is RH_NOSYNC, the region should be kept off
		 * from clean list.
		 * The hash entry for RH_NOSYNC will remain in memory
		 * until the region is recovered or the map is reloaded.
		 */

		/* do nothing for RH_NOSYNC */
		if (reg->state == RH_RECOVERING) {
			list_add_tail(&reg->list, &rh->quiesced_regions);
		} else if (reg->state == RH_DIRTY) {
			reg->state = RH_CLEAN;
			list_add(&reg->list, &rh->clean_regions);
		}
		should_wake = 1;
	}
	spin_unlock_irqrestore(&rh->region_lock, flags);

	if (should_wake)
		wake(rh->ms);
}

/*
 * Starts quiescing a region in preparation for recovery.
 */
static int __rh_recovery_prepare(struct region_hash *rh)
{
	int r;
	struct region *reg;
	region_t region;

	/*
	 * Ask the dirty log what's next.
	 */
	r = rh->log->type->get_resync_work(rh->log, &region);
	if (r <= 0)
		return r;

	/*
	 * Get this region, and start it quiescing by setting the
	 * recovering flag.
	 */
	read_lock(&rh->hash_lock);
	reg = __rh_find(rh, region);
	read_unlock(&rh->hash_lock);

	spin_lock_irq(&rh->region_lock);
	reg->state = RH_RECOVERING;

	/* Already quiesced ? */
	if (atomic_read(&reg->pending))
		list_del_init(&reg->list);
	else
		list_move(&reg->list, &rh->quiesced_regions);

	spin_unlock_irq(&rh->region_lock);

	return 1;
}

static void rh_recovery_prepare(struct region_hash *rh)
{
	/* Extra reference to avoid race with rh_stop_recovery */
	atomic_inc(&rh->recovery_in_flight);

	while (!down_trylock(&rh->recovery_count)) {
		atomic_inc(&rh->recovery_in_flight);
		if (__rh_recovery_prepare(rh) <= 0) {
			atomic_dec(&rh->recovery_in_flight);
			up(&rh->recovery_count);
			break;
		}
	}

	/* Drop the extra reference */
	if (atomic_dec_and_test(&rh->recovery_in_flight))
		wake_up_all(&_kmirrord_recovery_stopped);
}

/*
 * Returns any quiesced regions.
 */
static struct region *rh_recovery_start(struct region_hash *rh)
{
	struct region *reg = NULL;

	spin_lock_irq(&rh->region_lock);
	if (!list_empty(&rh->quiesced_regions)) {
		reg = list_entry(rh->quiesced_regions.next,
				 struct region, list);
		list_del_init(&reg->list);	/* remove from the quiesced list */
	}
	spin_unlock_irq(&rh->region_lock);

	return reg;
}

static void rh_recovery_end(struct region *reg, int success)
{
	struct region_hash *rh = reg->rh;

	spin_lock_irq(&rh->region_lock);
	if (success)
		list_add(&reg->list, &reg->rh->recovered_regions);
	else {
		reg->state = RH_NOSYNC;
		list_add(&reg->list, &reg->rh->failed_recovered_regions);
	}
	spin_unlock_irq(&rh->region_lock);

	wake(rh->ms);
}

static int rh_flush(struct region_hash *rh)
{
	return rh->log->type->flush(rh->log);
}

static void rh_delay(struct region_hash *rh, struct bio *bio)
{
	struct region *reg;

	read_lock(&rh->hash_lock);
	reg = __rh_find(rh, bio_to_region(rh, bio));
	bio_list_add(&reg->delayed_bios, bio);
	read_unlock(&rh->hash_lock);
}

static void rh_stop_recovery(struct region_hash *rh)
{
	int i;

	/* wait for any recovering regions */
	for (i = 0; i < MAX_RECOVERY; i++)
		down(&rh->recovery_count);
}

static void rh_start_recovery(struct region_hash *rh)
{
	int i;

	for (i = 0; i < MAX_RECOVERY; i++)
		up(&rh->recovery_count);

	wake(rh->ms);
}

/*
 * Every mirror should look like this one.
 */
#define DEFAULT_MIRROR 0

/*
 * This is yucky.  We squirrel the mirror_set struct away inside
 * bi_next for write buffers.  This is safe since the bh
 * doesn't get submitted to the lower levels of block layer.
 */
static struct mirror_set *bio_get_ms(struct bio *bio)
{
	return (struct mirror_set *) bio->bi_next;
}

static void bio_set_ms(struct bio *bio, struct mirror_set *ms)
{
	bio->bi_next = (struct bio *) ms;
}

/*-----------------------------------------------------------------
 * Recovery.
 *
 * When a mirror is first activated we may find that some regions
 * are in the no-sync state.  We have to recover these by
 * recopying from the default mirror to all the others.
 *---------------------------------------------------------------*/
static void recovery_complete(int read_err, unsigned int write_err,
			      void *context)
{
	struct region *reg = (struct region *) context;

	if (read_err)
		/* Read error means the failure of default mirror. */
		DMERR_LIMIT("Unable to read primary mirror during recovery");

	if (write_err)
		DMERR_LIMIT("Write error during recovery (error = 0x%x)",
			    write_err);

	rh_recovery_end(reg, !(read_err || write_err));
}

static int recover(struct mirror_set *ms, struct region *reg)
{
	int r;
	unsigned int i;
	struct io_region from, to[KCOPYD_MAX_REGIONS], *dest;
	struct mirror *m;
	unsigned long flags = 0;

	/* fill in the source */
	m = ms->default_mirror;
	from.bdev = m->dev->bdev;
	from.sector = m->offset + region_to_sector(reg->rh, reg->key);
	if (reg->key == (ms->nr_regions - 1)) {
		/*
		 * The final region may be smaller than
		 * region_size.
		 */
		from.count = ms->ti->len & (reg->rh->region_size - 1);
		if (!from.count)
			from.count = reg->rh->region_size;
	} else
		from.count = reg->rh->region_size;

	/* fill in the destinations */
	for (i = 0, dest = to; i < ms->nr_mirrors; i++) {
		if (&ms->mirror[i] == ms->default_mirror)
			continue;

		m = ms->mirror + i;
		dest->bdev = m->dev->bdev;
		dest->sector = m->offset + region_to_sector(reg->rh, reg->key);
		dest->count = from.count;
		dest++;
	}

	/* hand to kcopyd */
	set_bit(KCOPYD_IGNORE_ERROR, &flags);
	r = kcopyd_copy(ms->kcopyd_client, &from, ms->nr_mirrors - 1, to, flags,
			recovery_complete, reg);

	return r;
}

static void do_recovery(struct mirror_set *ms)
{
	int r;
	struct region *reg;
	struct dirty_log *log = ms->rh.log;

	/*
	 * Start quiescing some regions.
	 */
	rh_recovery_prepare(&ms->rh);

	/*
	 * Copy any already quiesced regions.
	 */
	while ((reg = rh_recovery_start(&ms->rh))) {
		r = recover(ms, reg);
		if (r)
			rh_recovery_end(reg, 0);
	}

	/*
	 * Update the in sync flag.
	 */
	if (!ms->in_sync &&
	    (log->type->get_sync_count(log) == ms->nr_regions)) {
		/* the sync is complete */
		dm_table_event(ms->ti->table);
		ms->in_sync = 1;
	}
}

/*-----------------------------------------------------------------
 * Reads
 *---------------------------------------------------------------*/
static struct mirror *choose_mirror(struct mirror_set *ms, sector_t sector)
{
	/* FIXME: add read balancing */
	return ms->default_mirror;
}

/*
 * remap a buffer to a particular mirror.
 */
static void map_bio(struct mirror_set *ms, struct mirror *m, struct bio *bio)
{
	bio->bi_bdev = m->dev->bdev;
	bio->bi_sector = m->offset + (bio->bi_sector - ms->ti->begin);
}

static void do_reads(struct mirror_set *ms, struct bio_list *reads)
{
	region_t region;
	struct bio *bio;
	struct mirror *m;

	while ((bio = bio_list_pop(reads))) {
		region = bio_to_region(&ms->rh, bio);

		/*
		 * We can only read balance if the region is in sync.
		 */
		if (rh_in_sync(&ms->rh, region, 1))
			m = choose_mirror(ms, bio->bi_sector);
		else
			m = ms->default_mirror;

		map_bio(ms, m, bio);
		generic_make_request(bio);
	}
}

/*-----------------------------------------------------------------
 * Writes.
 *
 * We do different things with the write io depending on the
 * state of the region that it's in:
 *
 * SYNC: 	increment pending, use kcopyd to write to *all* mirrors
 * RECOVERING:	delay the io until recovery completes
 * NOSYNC:	increment pending, just write to the default mirror
 *---------------------------------------------------------------*/
static void write_callback(unsigned long error, void *context)
{
	unsigned int i;
	int uptodate = 1;
	struct bio *bio = (struct bio *) context;
	struct mirror_set *ms;

	ms = bio_get_ms(bio);
	bio_set_ms(bio, NULL);

	/*
	 * NOTE: We don't decrement the pending count here,
	 * instead it is done by the targets endio function.
	 * This way we handle both writes to SYNC and NOSYNC
	 * regions with the same code.
	 */

	if (error) {
		/*
		 * only error the io if all mirrors failed.
		 * FIXME: bogus
		 */
		uptodate = 0;
		for (i = 0; i < ms->nr_mirrors; i++)
			if (!test_bit(i, &error)) {
				uptodate = 1;
				break;
			}
	}
	bio_endio(bio, 0);
}

static void do_write(struct mirror_set *ms, struct bio *bio)
{
	unsigned int i;
	struct io_region io[KCOPYD_MAX_REGIONS+1];
	struct mirror *m;
	struct dm_io_request io_req = {
		.bi_rw = WRITE,
		.mem.type = DM_IO_BVEC,
		.mem.ptr.bvec = bio->bi_io_vec + bio->bi_idx,
		.notify.fn = write_callback,
		.notify.context = bio,
		.client = ms->io_client,
	};

	for (i = 0; i < ms->nr_mirrors; i++) {
		m = ms->mirror + i;

		io[i].bdev = m->dev->bdev;
		io[i].sector = m->offset + (bio->bi_sector - ms->ti->begin);
		io[i].count = bio->bi_size >> 9;
	}

	bio_set_ms(bio, ms);

	(void) dm_io(&io_req, ms->nr_mirrors, io, NULL);
}

static void do_writes(struct mirror_set *ms, struct bio_list *writes)
{
	int state;
	struct bio *bio;
	struct bio_list sync, nosync, recover, *this_list = NULL;

	if (!writes->head)
		return;

	/*
	 * Classify each write.
	 */
	bio_list_init(&sync);
	bio_list_init(&nosync);
	bio_list_init(&recover);

	while ((bio = bio_list_pop(writes))) {
		state = rh_state(&ms->rh, bio_to_region(&ms->rh, bio), 1);
		switch (state) {
		case RH_CLEAN:
		case RH_DIRTY:
			this_list = &sync;
			break;

		case RH_NOSYNC:
			this_list = &nosync;
			break;

		case RH_RECOVERING:
			this_list = &recover;
			break;
		}

		bio_list_add(this_list, bio);
	}

	/*
	 * Increment the pending counts for any regions that will
	 * be written to (writes to recover regions are going to
	 * be delayed).
	 */
	rh_inc_pending(&ms->rh, &sync);
	rh_inc_pending(&ms->rh, &nosync);
	ms->log_failure = rh_flush(&ms->rh) ? 1 : 0;

	/*
	 * Dispatch io.
	 */
	if (unlikely(ms->log_failure))
		while ((bio = bio_list_pop(&sync)))
			bio_endio(bio, -EIO);
	else while ((bio = bio_list_pop(&sync)))
		do_write(ms, bio);

	while ((bio = bio_list_pop(&recover)))
		rh_delay(&ms->rh, bio);

	while ((bio = bio_list_pop(&nosync))) {
		map_bio(ms, ms->default_mirror, bio);
		generic_make_request(bio);
	}
}

/*-----------------------------------------------------------------
 * kmirrord
 *---------------------------------------------------------------*/
static void do_mirror(struct work_struct *work)
{
	struct mirror_set *ms =container_of(work, struct mirror_set,
					    kmirrord_work);
	struct bio_list reads, writes;

	spin_lock(&ms->lock);
	reads = ms->reads;
	writes = ms->writes;
	bio_list_init(&ms->reads);
	bio_list_init(&ms->writes);
	spin_unlock(&ms->lock);

	rh_update_states(&ms->rh);
	do_recovery(ms);
	do_reads(ms, &reads);
	do_writes(ms, &writes);
}

/*-----------------------------------------------------------------