Merge git://git.kernel.org/pub/scm/linux/kernel/git/agk/linux-2.6-dm

* git://git.kernel.org/pub/scm/linux/kernel/git/agk/linux-2.6-dm: (44 commits)
  dm raid1: report fault status
  dm raid1: handle read failures
  dm raid1: fix EIO after log failure
  dm raid1: handle recovery failures
  dm raid1: handle write failures
  dm snapshot: combine consecutive exceptions in memory
  dm: stripe enhanced status return
  dm: stripe trigger event on failure
  dm log: auto load modules
  dm: move deferred bio flushing to workqueue
  dm crypt: use async crypto
  dm crypt: prepare async callback fn
  dm crypt: add completion for async
  dm crypt: add async request mempool
  dm crypt: extract scatterlist processing
  dm crypt: tidy io ref counting
  dm crypt: introduce crypt_write_io_loop
  dm crypt: abstract crypt_write_done
  dm crypt: store sector mapping in dm_crypt_io
  dm crypt: move queue functions
  ...

drivers/md/Kconfig

@@ -204,7 +204,7 @@ config BLK_DEV_DM
 
 config DM_DEBUG
 	boolean "Device mapper debugging support"
-	depends on BLK_DEV_DM && EXPERIMENTAL
+	depends on BLK_DEV_DM
 	---help---
 	  Enable this for messages that may help debug device-mapper problems.
 
@@ -212,7 +212,7 @@ config DM_DEBUG
 
 config DM_CRYPT
 	tristate "Crypt target support"
-	depends on BLK_DEV_DM && EXPERIMENTAL
+	depends on BLK_DEV_DM
 	select CRYPTO
 	select CRYPTO_CBC
 	---help---
@@ -230,34 +230,34 @@ config DM_CRYPT
 	  If unsure, say N.
 
 config DM_SNAPSHOT
-       tristate "Snapshot target (EXPERIMENTAL)"
-       depends on BLK_DEV_DM && EXPERIMENTAL
+       tristate "Snapshot target"
+       depends on BLK_DEV_DM
        ---help---
          Allow volume managers to take writable snapshots of a device.
 
 config DM_MIRROR
-       tristate "Mirror target (EXPERIMENTAL)"
-       depends on BLK_DEV_DM && EXPERIMENTAL
+       tristate "Mirror target"
+       depends on BLK_DEV_DM
        ---help---
          Allow volume managers to mirror logical volumes, also
          needed for live data migration tools such as 'pvmove'.
 
 config DM_ZERO
-	tristate "Zero target (EXPERIMENTAL)"
-	depends on BLK_DEV_DM && EXPERIMENTAL
+	tristate "Zero target"
+	depends on BLK_DEV_DM
 	---help---
 	  A target that discards writes, and returns all zeroes for
 	  reads.  Useful in some recovery situations.
 
 config DM_MULTIPATH
-	tristate "Multipath target (EXPERIMENTAL)"
-	depends on BLK_DEV_DM && EXPERIMENTAL
+	tristate "Multipath target"
+	depends on BLK_DEV_DM
 	---help---
 	  Allow volume managers to support multipath hardware.
 
 config DM_MULTIPATH_EMC
-	tristate "EMC CX/AX multipath support (EXPERIMENTAL)"
-	depends on DM_MULTIPATH && BLK_DEV_DM && EXPERIMENTAL
+	tristate "EMC CX/AX multipath support"
+	depends on DM_MULTIPATH && BLK_DEV_DM
 	---help---
 	  Multipath support for EMC CX/AX series hardware.
 

drivers/md/dm-crypt.c

 /*
  * Copyright (C) 2003 Christophe Saout <christophe@saout.de>
  * Copyright (C) 2004 Clemens Fruhwirth <clemens@endorphin.org>
- * Copyright (C) 2006 Red Hat, Inc. All rights reserved.
+ * Copyright (C) 2006-2007 Red Hat, Inc. All rights reserved.
  *
  * This file is released under the GPL.
  */
 
+#include <linux/completion.h>
 #include <linux/err.h>
 #include <linux/module.h>
 #include <linux/init.h>
@@ -27,21 +28,11 @@
 #define DM_MSG_PREFIX "crypt"
 #define MESG_STR(x) x, sizeof(x)
 
-/*
- * per bio private data
- */
-struct dm_crypt_io {
-	struct dm_target *target;
-	struct bio *base_bio;
-	struct work_struct work;
-	atomic_t pending;
-	int error;
-};
-
 /*
  * context holding the current state of a multi-part conversion
  */
 struct convert_context {
+	struct completion restart;
 	struct bio *bio_in;
 	struct bio *bio_out;
 	unsigned int offset_in;
@@ -49,7 +40,27 @@ struct convert_context {
 	unsigned int idx_in;
 	unsigned int idx_out;
 	sector_t sector;
-	int write;
+	atomic_t pending;
+};
+
+/*
+ * per bio private data
+ */
+struct dm_crypt_io {
+	struct dm_target *target;
+	struct bio *base_bio;
+	struct work_struct work;
+
+	struct convert_context ctx;
+
+	atomic_t pending;
+	int error;
+	sector_t sector;
+};
+
+struct dm_crypt_request {
+	struct scatterlist sg_in;
+	struct scatterlist sg_out;
 };
 
 struct crypt_config;
@@ -72,10 +83,11 @@ struct crypt_config {
 	sector_t start;
 
 	/*
-	 * pool for per bio private data and
-	 * for encryption buffer pages
+	 * pool for per bio private data, crypto requests and
+	 * encryption requeusts/buffer pages
 	 */
 	mempool_t *io_pool;
+	mempool_t *req_pool;
 	mempool_t *page_pool;
 	struct bio_set *bs;
 
@@ -93,9 +105,25 @@ struct crypt_config {
 	sector_t iv_offset;
 	unsigned int iv_size;
 
+	/*
+	 * Layout of each crypto request:
+	 *
+	 *   struct ablkcipher_request
+	 *      context
+	 *      padding
+	 *   struct dm_crypt_request
+	 *      padding
+	 *   IV
+	 *
+	 * The padding is added so that dm_crypt_request and the IV are
+	 * correctly aligned.
+	 */
+	unsigned int dmreq_start;
+	struct ablkcipher_request *req;
+
 	char cipher[CRYPTO_MAX_ALG_NAME];
 	char chainmode[CRYPTO_MAX_ALG_NAME];
-	struct crypto_blkcipher *tfm;
+	struct crypto_ablkcipher *tfm;
 	unsigned long flags;
 	unsigned int key_size;
 	u8 key[0];
@@ -108,6 +136,7 @@ struct crypt_config {
 static struct kmem_cache *_crypt_io_pool;
 
 static void clone_init(struct dm_crypt_io *, struct bio *);
+static void kcryptd_queue_crypt(struct dm_crypt_io *io);
 
 /*
  * Different IV generation algorithms:
@@ -188,7 +217,7 @@ static int crypt_iv_essiv_ctr(struct crypt_config *cc, struct dm_target *ti,
 		return PTR_ERR(essiv_tfm);
 	}
 	if (crypto_cipher_blocksize(essiv_tfm) !=
-	    crypto_blkcipher_ivsize(cc->tfm)) {
+	    crypto_ablkcipher_ivsize(cc->tfm)) {
 		ti->error = "Block size of ESSIV cipher does "
 			    "not match IV size of block cipher";
 		crypto_free_cipher(essiv_tfm);
@@ -225,7 +254,7 @@ static int crypt_iv_essiv_gen(struct crypt_config *cc, u8 *iv, sector_t sector)
 static int crypt_iv_benbi_ctr(struct crypt_config *cc, struct dm_target *ti,
 			      const char *opts)
 {
-	unsigned int bs = crypto_blkcipher_blocksize(cc->tfm);
+	unsigned bs = crypto_ablkcipher_blocksize(cc->tfm);
 	int log = ilog2(bs);
 
 	/* we need to calculate how far we must shift the sector count
@@ -289,42 +318,10 @@ static struct crypt_iv_operations crypt_iv_null_ops = {
 	.generator = crypt_iv_null_gen
 };
 
-static int
-crypt_convert_scatterlist(struct crypt_config *cc, struct scatterlist *out,
-                          struct scatterlist *in, unsigned int length,
-                          int write, sector_t sector)
-{
-	u8 iv[cc->iv_size] __attribute__ ((aligned(__alignof__(u64))));
-	struct blkcipher_desc desc = {
-		.tfm = cc->tfm,
-		.info = iv,
-		.flags = CRYPTO_TFM_REQ_MAY_SLEEP,
-	};
-	int r;
-
-	if (cc->iv_gen_ops) {
-		r = cc->iv_gen_ops->generator(cc, iv, sector);
-		if (r < 0)
-			return r;
-
-		if (write)
-			r = crypto_blkcipher_encrypt_iv(&desc, out, in, length);
-		else
-			r = crypto_blkcipher_decrypt_iv(&desc, out, in, length);
-	} else {
-		if (write)
-			r = crypto_blkcipher_encrypt(&desc, out, in, length);
-		else
-			r = crypto_blkcipher_decrypt(&desc, out, in, length);
-	}
-
-	return r;
-}
-
 static void crypt_convert_init(struct crypt_config *cc,
 			       struct convert_context *ctx,
 			       struct bio *bio_out, struct bio *bio_in,
-			       sector_t sector, int write)
+			       sector_t sector)
 {
 	ctx->bio_in = bio_in;
 	ctx->bio_out = bio_out;
@@ -333,49 +330,123 @@ static void crypt_convert_init(struct crypt_config *cc,
 	ctx->idx_in = bio_in ? bio_in->bi_idx : 0;
 	ctx->idx_out = bio_out ? bio_out->bi_idx : 0;
 	ctx->sector = sector + cc->iv_offset;
-	ctx->write = write;
+	init_completion(&ctx->restart);
+	/*
+	 * Crypto operation can be asynchronous,
+	 * ctx->pending is increased after request submission.
+	 * We need to ensure that we don't call the crypt finish
+	 * operation before pending got incremented
+	 * (dependent on crypt submission return code).
+	 */
+	atomic_set(&ctx->pending, 2);
 }
 
-/*
- * Encrypt / decrypt data from one bio to another one (can be the same one)
- */
-static int crypt_convert(struct crypt_config *cc,
-			 struct convert_context *ctx)
+static int crypt_convert_block(struct crypt_config *cc,
+			       struct convert_context *ctx,
+			       struct ablkcipher_request *req)
 {
-	int r = 0;
-
-	while(ctx->idx_in < ctx->bio_in->bi_vcnt &&
-	      ctx->idx_out < ctx->bio_out->bi_vcnt) {
 	struct bio_vec *bv_in = bio_iovec_idx(ctx->bio_in, ctx->idx_in);
 	struct bio_vec *bv_out = bio_iovec_idx(ctx->bio_out, ctx->idx_out);
-		struct scatterlist sg_in, sg_out;
+	struct dm_crypt_request *dmreq;
+	u8 *iv;
+	int r = 0;
 
-		sg_init_table(&sg_in, 1);
-		sg_set_page(&sg_in, bv_in->bv_page, 1 << SECTOR_SHIFT, bv_in->bv_offset + ctx->offset_in);
+	dmreq = (struct dm_crypt_request *)((char *)req + cc->dmreq_start);
+	iv = (u8 *)ALIGN((unsigned long)(dmreq + 1),
+			 crypto_ablkcipher_alignmask(cc->tfm) + 1);
 
-		sg_init_table(&sg_out, 1);
-		sg_set_page(&sg_out, bv_out->bv_page, 1 << SECTOR_SHIFT, bv_out->bv_offset + ctx->offset_out);
+	sg_init_table(&dmreq->sg_in, 1);
+	sg_set_page(&dmreq->sg_in, bv_in->bv_page, 1 << SECTOR_SHIFT,
+		    bv_in->bv_offset + ctx->offset_in);
 
-		ctx->offset_in += sg_in.length;
+	sg_init_table(&dmreq->sg_out, 1);
+	sg_set_page(&dmreq->sg_out, bv_out->bv_page, 1 << SECTOR_SHIFT,
+		    bv_out->bv_offset + ctx->offset_out);
+
+	ctx->offset_in += 1 << SECTOR_SHIFT;
 	if (ctx->offset_in >= bv_in->bv_len) {
 		ctx->offset_in = 0;
 		ctx->idx_in++;
 	}
 
-		ctx->offset_out += sg_out.length;
+	ctx->offset_out += 1 << SECTOR_SHIFT;
 	if (ctx->offset_out >= bv_out->bv_len) {
 		ctx->offset_out = 0;
 		ctx->idx_out++;
 	}
 
-		r = crypt_convert_scatterlist(cc, &sg_out, &sg_in, sg_in.length,
-					      ctx->write, ctx->sector);
+	if (cc->iv_gen_ops) {
+		r = cc->iv_gen_ops->generator(cc, iv, ctx->sector);
 		if (r < 0)
-			break;
+			return r;
+	}
 
+	ablkcipher_request_set_crypt(req, &dmreq->sg_in, &dmreq->sg_out,
+				     1 << SECTOR_SHIFT, iv);
+
+	if (bio_data_dir(ctx->bio_in) == WRITE)
+		r = crypto_ablkcipher_encrypt(req);
+	else
+		r = crypto_ablkcipher_decrypt(req);
+
+	return r;
+}
+
+static void kcryptd_async_done(struct crypto_async_request *async_req,
+			       int error);
+static void crypt_alloc_req(struct crypt_config *cc,
+			    struct convert_context *ctx)
+{
+	if (!cc->req)
+		cc->req = mempool_alloc(cc->req_pool, GFP_NOIO);
+	ablkcipher_request_set_tfm(cc->req, cc->tfm);
+	ablkcipher_request_set_callback(cc->req, CRYPTO_TFM_REQ_MAY_BACKLOG |
+					     CRYPTO_TFM_REQ_MAY_SLEEP,
+					     kcryptd_async_done, ctx);
+}
+
+/*
+ * Encrypt / decrypt data from one bio to another one (can be the same one)
+ */
+static int crypt_convert(struct crypt_config *cc,
+			 struct convert_context *ctx)
+{
+	int r = 0;
+
+	while(ctx->idx_in < ctx->bio_in->bi_vcnt &&
+	      ctx->idx_out < ctx->bio_out->bi_vcnt) {
+
+		crypt_alloc_req(cc, ctx);
+
+		r = crypt_convert_block(cc, ctx, cc->req);
+
+		switch (r) {
+		case -EBUSY:
+			wait_for_completion(&ctx->restart);
+			INIT_COMPLETION(ctx->restart);
+			/* fall through*/
+		case -EINPROGRESS:
+			atomic_inc(&ctx->pending);
+			cc->req = NULL;
+			r = 0;
+			/* fall through*/
+		case 0:
 			ctx->sector++;
+			continue;
+		}
+
+		break;
 	}
 
+	/*
+	 * If there are pending crypto operation run async
+	 * code. Otherwise process return code synchronously.
+	 * The step of 2 ensures that async finish doesn't
+	 * call crypto finish too early.
+	 */
+	if (atomic_sub_return(2, &ctx->pending))
+		return -EINPROGRESS;
+
 	return r;
 }
 
@@ -455,18 +526,14 @@ static void crypt_free_buffer_pages(struct crypt_config *cc, struct bio *clone)
  * One of the bios was finished. Check for completion of
  * the whole request and correctly clean up the buffer.
  */
-static void crypt_dec_pending(struct dm_crypt_io *io, int error)
+static void crypt_dec_pending(struct dm_crypt_io *io)
 {
-	struct crypt_config *cc = (struct crypt_config *) io->target->private;
-
-	if (error < 0)
-		io->error = error;
+	struct crypt_config *cc = io->target->private;
 
 	if (!atomic_dec_and_test(&io->pending))
 		return;
 
 	bio_endio(io->base_bio, io->error);
-
 	mempool_free(io, cc->io_pool);
 }
 
@@ -484,30 +551,11 @@ static void crypt_dec_pending(struct dm_crypt_io *io, int error)
  * starved by new requests which can block in the first stages due
  * to memory allocation.
  */
-static void kcryptd_do_work(struct work_struct *work);
-static void kcryptd_do_crypt(struct work_struct *work);
-
-static void kcryptd_queue_io(struct dm_crypt_io *io)
-{
-	struct crypt_config *cc = io->target->private;
-
-	INIT_WORK(&io->work, kcryptd_do_work);
-	queue_work(cc->io_queue, &io->work);
-}
-
-static void kcryptd_queue_crypt(struct dm_crypt_io *io)
-{
-	struct crypt_config *cc = io->target->private;
-
-	INIT_WORK(&io->work, kcryptd_do_crypt);
-	queue_work(cc->crypt_queue, &io->work);
-}
-
 static void crypt_endio(struct bio *clone, int error)
 {
 	struct dm_crypt_io *io = clone->bi_private;
 	struct crypt_config *cc = io->target->private;
-	unsigned read_io = bio_data_dir(clone) == READ;
+	unsigned rw = bio_data_dir(clone);
 
 	if (unlikely(!bio_flagged(clone, BIO_UPTODATE) && !error))
 		error = -EIO;
@@ -515,21 +563,20 @@ static void crypt_endio(struct bio *clone, int error)
 	/*
 	 * free the processed pages
 	 */
-	if (!read_io) {
+	if (rw == WRITE)
 		crypt_free_buffer_pages(cc, clone);
-		goto out;
-	}
-
-	if (unlikely(error))
-		goto out;
 
 	bio_put(clone);
+
+	if (rw == READ && !error) {
 		kcryptd_queue_crypt(io);
 		return;
+	}
 
-out:
-	bio_put(clone);
-	crypt_dec_pending(io, error);
+	if (unlikely(error))
+		io->error = error;
+
+	crypt_dec_pending(io);
 }
 
 static void clone_init(struct dm_crypt_io *io, struct bio *clone)
@@ -543,12 +590,11 @@ static void clone_init(struct dm_crypt_io *io, struct bio *clone)
 	clone->bi_destructor = dm_crypt_bio_destructor;
 }
 
-static void process_read(struct dm_crypt_io *io)
+static void kcryptd_io_read(struct dm_crypt_io *io)
 {
 	struct crypt_config *cc = io->target->private;
 	struct bio *base_bio = io->base_bio;
 	struct bio *clone;
-	sector_t sector = base_bio->bi_sector - io->target->begin;
 
 	atomic_inc(&io->pending);
 
@@ -559,7 +605,8 @@ static void process_read(struct dm_crypt_io *io)
 	 */
 	clone = bio_alloc_bioset(GFP_NOIO, bio_segments(base_bio), cc->bs);
 	if (unlikely(!clone)) {
-		crypt_dec_pending(io, -ENOMEM);
+		io->error = -ENOMEM;
+		crypt_dec_pending(io);
 		return;
 	}
 
@@ -567,25 +614,71 @@ static void process_read(struct dm_crypt_io *io)
 	clone->bi_idx = 0;
 	clone->bi_vcnt = bio_segments(base_bio);
 	clone->bi_size = base_bio->bi_size;
-	clone->bi_sector = cc->start + sector;
+	clone->bi_sector = cc->start + io->sector;
 	memcpy(clone->bi_io_vec, bio_iovec(base_bio),
 	       sizeof(struct bio_vec) * clone->bi_vcnt);
 
 	generic_make_request(clone);
 }
 
-static void process_write(struct dm_crypt_io *io)
+static void kcryptd_io_write(struct dm_crypt_io *io)
+{
+	struct bio *clone = io->ctx.bio_out;
+
+	generic_make_request(clone);
+}
+
+static void kcryptd_io(struct work_struct *work)
+{
+	struct dm_crypt_io *io = container_of(work, struct dm_crypt_io, work);
+
+	if (bio_data_dir(io->base_bio) == READ)
+		kcryptd_io_read(io);
+	else
+		kcryptd_io_write(io);
+}
+
+static void kcryptd_queue_io(struct dm_crypt_io *io)
 {
 	struct crypt_config *cc = io->target->private;
-	struct bio *base_bio = io->base_bio;
-	struct bio *clone;
-	struct convert_context ctx;
-	unsigned remaining = base_bio->bi_size;
-	sector_t sector = base_bio->bi_sector - io->target->begin;
 
+	INIT_WORK(&io->work, kcryptd_io);
+	queue_work(cc->io_queue, &io->work);
+}
+
+static void kcryptd_crypt_write_io_submit(struct dm_crypt_io *io,
+					  int error, int async)
+{
+	struct bio *clone = io->ctx.bio_out;
+	struct crypt_config *cc = io->target->private;
+
+	if (unlikely(error < 0)) {
+		crypt_free_buffer_pages(cc, clone);
+		bio_put(clone);
+		io->error = -EIO;
+		return;
+	}
+
+	/* crypt_convert should have filled the clone bio */
+	BUG_ON(io->ctx.idx_out < clone->bi_vcnt);
+
+	clone->bi_sector = cc->start + io->sector;
+	io->sector += bio_sectors(clone);
+
+	if (async)
+		kcryptd_queue_io(io);
+	else {
 		atomic_inc(&io->pending);
+		generic_make_request(clone);
+	}
+}
 
-	crypt_convert_init(cc, &ctx, NULL, base_bio, sector, 1);
+static void kcryptd_crypt_write_convert_loop(struct dm_crypt_io *io)
+{
+	struct crypt_config *cc = io->target->private;
+	struct bio *clone;
+	unsigned remaining = io->base_bio->bi_size;
+	int r;
 
 	/*
 	 * The allocated buffers can be smaller than the whole bio,
@@ -594,70 +687,110 @@ static void process_write(struct dm_crypt_io *io)
 	while (remaining) {
 		clone = crypt_alloc_buffer(io, remaining);
 		if (unlikely(!clone)) {
-			crypt_dec_pending(io, -ENOMEM);
+			io->error = -ENOMEM;
 			return;
 		}
 
-		ctx.bio_out = clone;
-		ctx.idx_out = 0;
+		io->ctx.bio_out = clone;
+		io->ctx.idx_out = 0;
 
-		if (unlikely(crypt_convert(cc, &ctx) < 0)) {
-			crypt_free_buffer_pages(cc, clone);
-			bio_put(clone);
-			crypt_dec_pending(io, -EIO);
+		remaining -= clone->bi_size;
+
+		r = crypt_convert(cc, &io->ctx);
+
+		if (r != -EINPROGRESS) {
+			kcryptd_crypt_write_io_submit(io, r, 0);
+			if (unlikely(r < 0))
 				return;
-		}
+		} else
+			atomic_inc(&io->pending);
 
-		/* crypt_convert should have filled the clone bio */
-		BUG_ON(ctx.idx_out < clone->bi_vcnt);
+		/* out of memory -> run queues */
+		if (unlikely(remaining))
+			congestion_wait(WRITE, HZ/100);
+	}
+}
 
-		clone->bi_sector = cc->start + sector;
-		remaining -= clone->bi_size;
-		sector += bio_sectors(clone);
+static void kcryptd_crypt_write_convert(struct dm_crypt_io *io)
+{
+	struct crypt_config *cc = io->target->private;
 
-		/* Grab another reference to the io struct
-		 * before we kick off the request */
-		if (remaining)
+	/*
+	 * Prevent io from disappearing until this function completes.
+	 */
 	atomic_inc(&io->pending);
 
-		generic_make_request(clone);
+	crypt_convert_init(cc, &io->ctx, NULL, io->base_bio, io->sector);
+	kcryptd_crypt_write_convert_loop(io);
 
-		/* Do not reference clone after this - it
-		 * may be gone already. */
+	crypt_dec_pending(io);
+}
 
-		/* out of memory -> run queues */
-		if (remaining)
-			congestion_wait(WRITE, HZ/100);
-	}
+static void kcryptd_crypt_read_done(struct dm_crypt_io *io, int error)
+{
+	if (unlikely(error < 0))
+		io->error = -EIO;
+
+	crypt_dec_pending(io);
 }
 
-static void process_read_endio(struct dm_crypt_io *io)
+static void kcryptd_crypt_read_convert(struct dm_crypt_io *io)
 {
 	struct crypt_config *cc = io->target->private;
-	struct convert_context ctx;
+	int r = 0;
+
+	atomic_inc(&io->pending);
+
+	crypt_convert_init(cc, &io->ctx, io->base_bio, io->base_bio,
+			   io->sector);
+
+	r = crypt_convert(cc, &io->ctx);
 
-	crypt_convert_init(cc, &ctx, io->base_bio, io->base_bio,
-			   io->base_bio->bi_sector - io->target->begin, 0);
+	if (r != -EINPROGRESS)
+		kcryptd_crypt_read_done(io, r);
 
-	crypt_dec_pending(io, crypt_convert(cc, &ctx));
+	crypt_dec_pending(io);
 }
 
-static void kcryptd_do_work(struct work_struct *work)
+static void kcryptd_async_done(struct crypto_async_request *async_req,
+			       int error)
 {
-	struct dm_crypt_io *io = container_of(work, struct dm_crypt_io, work);
+	struct convert_context *ctx = async_req->data;
+	struct dm_crypt_io *io = container_of(ctx, struct dm_crypt_io, ctx);
+	struct crypt_config *cc = io->target->private;
+
+	if (error == -EINPROGRESS) {
+		complete(&ctx->restart);
+		return;
+	}
+
+	mempool_free(ablkcipher_request_cast(async_req), cc->req_pool);
+
+	if (!atomic_dec_and_test(&ctx->pending))
+		return;
 
 	if (bio_data_dir(io->base_bio) == READ)
-		process_read(io);
+		kcryptd_crypt_read_done(io, error);
+	else
+		kcryptd_crypt_write_io_submit(io, error, 1);
 }
 
-static void kcryptd_do_crypt(struct work_struct *work)
+static void kcryptd_crypt(struct work_struct *work)
 {
 	struct dm_crypt_io *io = container_of(work, struct dm_crypt_io, work);
 
 	if (bio_data_dir(io->base_bio) == READ)
-		process_read_endio(io);
+		kcryptd_crypt_read_convert(io);
 	else
-		process_write(io);
+		kcryptd_crypt_write_convert(io);
+}
+
+static void kcryptd_queue_crypt(struct dm_crypt_io *io)
+{
+	struct crypt_config *cc = io->target->private;
+
+	INIT_WORK(&io->work, kcryptd_crypt);
+	queue_work(cc->crypt_queue, &io->work);
 }
 
 /*
@@ -733,7 +866,7 @@ static int crypt_wipe_key(struct crypt_config *cc)
 static int crypt_ctr(struct dm_target *ti, unsigned int argc, char **argv)
 {
 	struct crypt_config *cc;
-	struct crypto_blkcipher *tfm;
+	struct crypto_ablkcipher *tfm;
 	char *tmp;
 	char *cipher;
 	char *chainmode;
@@ -787,7 +920,7 @@ static int crypt_ctr(struct dm_target *ti, unsigned int argc, char **argv)
 		goto bad_cipher;
 	}
 
-	tfm = crypto_alloc_blkcipher(cc->cipher, 0, CRYPTO_ALG_ASYNC);
+	tfm = crypto_alloc_ablkcipher(cc->cipher, 0, 0);
 	if (IS_ERR(tfm)) {
 		ti->error = "Error allocating crypto tfm";
 		goto bad_cipher;
@@ -821,7 +954,7 @@ static int crypt_ctr(struct dm_target *ti, unsigned int argc, char **argv)
 	    cc->iv_gen_ops->ctr(cc, ti, ivopts) < 0)
 		goto bad_ivmode;
 
-	cc->iv_size = crypto_blkcipher_ivsize(tfm);
+	cc->iv_size = crypto_ablkcipher_ivsize(tfm);
 	if (cc->iv_size)
 		/* at least a 64 bit sector number should fit in our buffer */
 		cc->iv_size = max(cc->iv_size,
@@ -841,6 +974,20 @@ static int crypt_ctr(struct dm_target *ti, unsigned int argc, char **argv)
 		goto bad_slab_pool;
 	}
 
+	cc->dmreq_start = sizeof(struct ablkcipher_request);
+	cc->dmreq_start += crypto_ablkcipher_reqsize(tfm);
+	cc->dmreq_start = ALIGN(cc->dmreq_start, crypto_tfm_ctx_alignment());
+	cc->dmreq_start += crypto_ablkcipher_alignmask(tfm) &
+			   ~(crypto_tfm_ctx_alignment() - 1);
+
+	cc->req_pool = mempool_create_kmalloc_pool(MIN_IOS, cc->dmreq_start +
+			sizeof(struct dm_crypt_request) + cc->iv_size);
+	if (!cc->req_pool) {
+		ti->error = "Cannot allocate crypt request mempool";
+		goto bad_req_pool;
+	}
+	cc->req = NULL;
+
 	cc->page_pool = mempool_create_page_pool(MIN_POOL_PAGES, 0);
 	if (!cc->page_pool) {
 		ti->error = "Cannot allocate page mempool";
@@ -853,7 +1000,7 @@ static int crypt_ctr(struct dm_target *ti, unsigned int argc, char **argv)
 		goto bad_bs;
 	}
 
-	if (crypto_blkcipher_setkey(tfm, cc->key, key_size) < 0) {
+	if (crypto_ablkcipher_setkey(tfm, cc->key, key_size) < 0) {
 		ti->error = "Error setting key";
 		goto bad_device;
 	}
@@ -914,12 +1061,14 @@ static int crypt_ctr(struct dm_target *ti, unsigned int argc, char **argv)
 bad_bs:
 	mempool_destroy(cc->page_pool);
 bad_page_pool:
+	mempool_destroy(cc->req_pool);
+bad_req_pool:
 	mempool_destroy(cc->io_pool);
 bad_slab_pool:
 	if (cc->iv_gen_ops && cc->iv_gen_ops->dtr)
 		cc->iv_gen_ops->dtr(cc);
 bad_ivmode:
-	crypto_free_blkcipher(tfm);
+	crypto_free_ablkcipher(tfm);
 bad_cipher:
 	/* Must zero key material before freeing */
 	memset(cc, 0, sizeof(*cc) + cc->key_size * sizeof(u8));
@@ -934,14 +1083,18 @@ static void crypt_dtr(struct dm_target *ti)
 	destroy_workqueue(cc->io_queue);
 	destroy_workqueue(cc->crypt_queue);
 
+	if (cc->req)
+		mempool_free(cc->req, cc->req_pool);
+
 	bioset_free(cc->bs);
 	mempool_destroy(cc->page_pool);
+	mempool_destroy(cc->req_pool);
 	mempool_destroy(cc->io_pool);
 
 	kfree(cc->iv_mode);
 	if (cc->iv_gen_ops && cc->iv_gen_ops->dtr)
 		cc->iv_gen_ops->dtr(cc);
-	crypto_free_blkcipher(cc->tfm);
+	crypto_free_ablkcipher(cc->tfm);
 	dm_put_device(ti, cc->dev);
 
 	/* Must zero key material before freeing */
@@ -958,6 +1111,7 @@ static int crypt_map(struct dm_target *ti, struct bio *bio,
 	io = mempool_alloc(cc->io_pool, GFP_NOIO);
 	io->target = ti;
 	io->base_bio = bio;
+	io->sector = bio->bi_sector - ti->begin;
 	io->error = 0;
 	atomic_set(&io->pending, 0);
 

drivers/md/dm-exception-store.c

@@ -449,7 +449,7 @@ static void persistent_destroy(struct exception_store *store)
 
 static int persistent_read_metadata(struct exception_store *store)
 {
-	int r, new_snapshot;
+	int r, uninitialized_var(new_snapshot);
 	struct pstore *ps = get_info(store);
 
 	/*

drivers/md/dm-ioctl.c

@@ -15,6 +15,7 @@
 #include <linux/slab.h>
 #include <linux/dm-ioctl.h>
 #include <linux/hdreg.h>
+#include <linux/compat.h>
 
 #include <asm/uaccess.h>
 
@@ -702,7 +703,7 @@ static int dev_rename(struct dm_ioctl *param, size_t param_size)
 	int r;
 	char *new_name = (char *) param + param->data_start;
 
-	if (new_name < (char *) param->data ||
+	if (new_name < param->data ||
 	    invalid_str(new_name, (void *) param + param_size)) {
 		DMWARN("Invalid new logical volume name supplied.");
 		return -EINVAL;
@@ -728,7 +729,7 @@ static int dev_set_geometry(struct dm_ioctl *param, size_t param_size)
 	if (!md)
 		return -ENXIO;
 
-	if (geostr < (char *) param->data ||
+	if (geostr < param->data ||
 	    invalid_str(geostr, (void *) param + param_size)) {
 		DMWARN("Invalid geometry supplied.");
 		goto out;
@@ -1350,10 +1351,10 @@ static int copy_params(struct dm_ioctl __user *user, struct dm_ioctl **param)
 {
 	struct dm_ioctl tmp, *dmi;
 
-	if (copy_from_user(&tmp, user, sizeof(tmp)))
+	if (copy_from_user(&tmp, user, sizeof(tmp) - sizeof(tmp.data)))
 		return -EFAULT;
 
-	if (tmp.data_size < sizeof(tmp))
+	if (tmp.data_size < (sizeof(tmp) - sizeof(tmp.data)))
 		return -EINVAL;
 
 	dmi = vmalloc(tmp.data_size);
@@ -1397,13 +1398,11 @@ static int validate_params(uint cmd, struct dm_ioctl *param)
 	return 0;
 }
 
-static int ctl_ioctl(struct inode *inode, struct file *file,
-		     uint command, ulong u)
+static int ctl_ioctl(uint command, struct dm_ioctl __user *user)
 {
 	int r = 0;
 	unsigned int cmd;
-	struct dm_ioctl *param;
-	struct dm_ioctl __user *user = (struct dm_ioctl __user *) u;
+	struct dm_ioctl *uninitialized_var(param);
 	ioctl_fn fn = NULL;
 	size_t param_size;
 
@@ -1471,8 +1470,23 @@ static int ctl_ioctl(struct inode *inode, struct file *file,
 	return r;
 }
 
+static long dm_ctl_ioctl(struct file *file, uint command, ulong u)
+{
+	return (long)ctl_ioctl(command, (struct dm_ioctl __user *)u);
+}
+
+#ifdef CONFIG_COMPAT
+static long dm_compat_ctl_ioctl(struct file *file, uint command, ulong u)
+{
+	return (long)dm_ctl_ioctl(file, command, (ulong) compat_ptr(u));
+}
+#else
+#define dm_compat_ctl_ioctl NULL
+#endif
+
 static const struct file_operations _ctl_fops = {
-	.ioctl	 = ctl_ioctl,
+	.unlocked_ioctl	 = dm_ctl_ioctl,
+	.compat_ioctl = dm_compat_ctl_ioctl,
 	.owner	 = THIS_MODULE,
 };
 

drivers/md/dm-log.c

@@ -41,7 +41,7 @@ int dm_unregister_dirty_log_type(struct dirty_log_type *type)
 	return 0;
 }
 
-static struct dirty_log_type *get_type(const char *type_name)
+static struct dirty_log_type *_get_type(const char *type_name)
 {
 	struct dirty_log_type *type;
 
@@ -61,6 +61,55 @@ static struct dirty_log_type *get_type(const char *type_name)
 	return NULL;
 }
 
+/*
+ * get_type
+ * @type_name
+ *
+ * Attempt to retrieve the dirty_log_type by name.  If not already
+ * available, attempt to load the appropriate module.
+ *
+ * Log modules are named "dm-log-" followed by the 'type_name'.
+ * Modules may contain multiple types.
+ * This function will first try the module "dm-log-<type_name>",
+ * then truncate 'type_name' on the last '-' and try again.
+ *
+ * For example, if type_name was "clustered-disk", it would search
+ * 'dm-log-clustered-disk' then 'dm-log-clustered'.
+ *
+ * Returns: dirty_log_type* on success, NULL on failure
+ */
+static struct dirty_log_type *get_type(const char *type_name)
+{
+	char *p, *type_name_dup;
+	struct dirty_log_type *type;
+
+	type = _get_type(type_name);
+	if (type)
+		return type;
+
+	type_name_dup = kstrdup(type_name, GFP_KERNEL);
+	if (!type_name_dup) {
+		DMWARN("No memory left to attempt log module load for \"%s\"",
+		       type_name);
+		return NULL;
+	}
+
+	while (request_module("dm-log-%s", type_name_dup) ||
+	       !(type = _get_type(type_name))) {
+		p = strrchr(type_name_dup, '-');
+		if (!p)
+			break;
+		p[0] = '\0';
+	}
+
+	if (!type)
+		DMWARN("Module for logging type \"%s\" not found.", type_name);
+
+	kfree(type_name_dup);
+
+	return type;
+}
+
 static void put_type(struct dirty_log_type *type)
 {
 	spin_lock(&_lock);

drivers/md/dm-mpath.c

@@ -106,7 +106,7 @@ typedef int (*action_fn) (struct pgpath *pgpath);
 
 static struct kmem_cache *_mpio_cache;
 
-struct workqueue_struct *kmultipathd;
+static struct workqueue_struct *kmultipathd;
 static void process_queued_ios(struct work_struct *work);
 static void trigger_event(struct work_struct *work);
 

drivers/md/dm-raid1.c

@@ -6,6 +6,7 @@
 
 #include "dm.h"
 #include "dm-bio-list.h"
+#include "dm-bio-record.h"
 #include "dm-io.h"
 #include "dm-log.h"
 #include "kcopyd.h"
@@ -20,6 +21,7 @@
 #include <linux/vmalloc.h>
 #include <linux/workqueue.h>
 #include <linux/log2.h>
+#include <linux/hardirq.h>
 
 #define DM_MSG_PREFIX "raid1"
 #define DM_IO_PAGES 64
@@ -113,9 +115,16 @@ struct region {
 /*-----------------------------------------------------------------
  * Mirror set structures.
  *---------------------------------------------------------------*/
+enum dm_raid1_error {
+	DM_RAID1_WRITE_ERROR,
+	DM_RAID1_SYNC_ERROR,
+	DM_RAID1_READ_ERROR
+};
+
 struct mirror {
 	struct mirror_set *ms;
 	atomic_t error_count;
+	uint32_t error_type;
 	struct dm_dev *dev;
 	sector_t offset;
 };
@@ -127,21 +136,25 @@ struct mirror_set {
 	struct kcopyd_client *kcopyd_client;
 	uint64_t features;
 
-	spinlock_t lock;	/* protects the next two lists */
+	spinlock_t lock;	/* protects the lists */
 	struct bio_list reads;
 	struct bio_list writes;
+	struct bio_list failures;
 
 	struct dm_io_client *io_client;
+	mempool_t *read_record_pool;
 
 	/* recovery */
 	region_t nr_regions;
 	int in_sync;
 	int log_failure;
+	atomic_t suspend;
 
-	struct mirror *default_mirror;	/* Default mirror */
+	atomic_t default_mirror;	/* Default mirror */
 
 	struct workqueue_struct *kmirrord_wq;
 	struct work_struct kmirrord_work;
+	struct work_struct trigger_event;
 
 	unsigned int nr_mirrors;
 	struct mirror mirror[0];
@@ -362,6 +375,16 @@ static void complete_resync_work(struct region *reg, int success)
 	struct region_hash *rh = reg->rh;
 
 	rh->log->type->set_region_sync(rh->log, reg->key, success);
+
+	/*
+	 * Dispatch the bios before we call 'wake_up_all'.
+	 * This is important because if we are suspending,
+	 * we want to know that recovery is complete and
+	 * the work queue is flushed.  If we wake_up_all
+	 * before we dispatch_bios (queue bios and call wake()),
+	 * then we risk suspending before the work queue
+	 * has been properly flushed.
+	 */
 	dispatch_bios(rh->ms, &reg->delayed_bios);
 	if (atomic_dec_and_test(&rh->recovery_in_flight))
 		wake_up_all(&_kmirrord_recovery_stopped);
@@ -626,24 +649,101 @@ static void rh_start_recovery(struct region_hash *rh)
 	wake(rh->ms);
 }
 
+#define MIN_READ_RECORDS 20
+struct dm_raid1_read_record {
+	struct mirror *m;
+	struct dm_bio_details details;
+};
+
 /*
  * 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
+ * This is yucky.  We squirrel the mirror struct away inside
+ * bi_next for read/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)
+static struct mirror *bio_get_m(struct bio *bio)
 {
-	return (struct mirror_set *) bio->bi_next;
+	return (struct mirror *) bio->bi_next;
 }
 
-static void bio_set_ms(struct bio *bio, struct mirror_set *ms)
+static void bio_set_m(struct bio *bio, struct mirror *m)
 {
-	bio->bi_next = (struct bio *) ms;
+	bio->bi_next = (struct bio *) m;
+}
+
+static struct mirror *get_default_mirror(struct mirror_set *ms)
+{
+	return &ms->mirror[atomic_read(&ms->default_mirror)];
+}
+
+static void set_default_mirror(struct mirror *m)
+{
+	struct mirror_set *ms = m->ms;
+	struct mirror *m0 = &(ms->mirror[0]);
+
+	atomic_set(&ms->default_mirror, m - m0);
+}
+
+/* fail_mirror
+ * @m: mirror device to fail
+ * @error_type: one of the enum's, DM_RAID1_*_ERROR
+ *
+ * If errors are being handled, record the type of
+ * error encountered for this device.  If this type
+ * of error has already been recorded, we can return;
+ * otherwise, we must signal userspace by triggering
+ * an event.  Additionally, if the device is the
+ * primary device, we must choose a new primary, but
+ * only if the mirror is in-sync.
+ *
+ * This function must not block.
+ */
+static void fail_mirror(struct mirror *m, enum dm_raid1_error error_type)
+{
+	struct mirror_set *ms = m->ms;
+	struct mirror *new;
+
+	if (!errors_handled(ms))
+		return;
+
+	/*
+	 * error_count is used for nothing more than a
+	 * simple way to tell if a device has encountered
+	 * errors.
+	 */
+	atomic_inc(&m->error_count);
+
+	if (test_and_set_bit(error_type, &m->error_type))
+		return;
+
+	if (m != get_default_mirror(ms))
+		goto out;
+
+	if (!ms->in_sync) {
+		/*
+		 * Better to issue requests to same failing device
+		 * than to risk returning corrupt data.
+		 */
+		DMERR("Primary mirror (%s) failed while out-of-sync: "
+		      "Reads may fail.", m->dev->name);
+		goto out;
+	}
+
+	for (new = ms->mirror; new < ms->mirror + ms->nr_mirrors; new++)
+		if (!atomic_read(&new->error_count)) {
+			set_default_mirror(new);
+			break;
+		}
+
+	if (unlikely(new == ms->mirror + ms->nr_mirrors))
+		DMWARN("All sides of mirror have failed.");
+
+out:
+	schedule_work(&ms->trigger_event);
 }
 
 /*-----------------------------------------------------------------
@@ -656,15 +756,32 @@ static void bio_set_ms(struct bio *bio, struct mirror_set *ms)
 static void recovery_complete(int read_err, unsigned int write_err,
 			      void *context)
 {
-	struct region *reg = (struct region *) context;
+	struct region *reg = (struct region *)context;
+	struct mirror_set *ms = reg->rh->ms;
+	int m, bit = 0;
 
-	if (read_err)
+	if (read_err) {
 		/* Read error means the failure of default mirror. */
 		DMERR_LIMIT("Unable to read primary mirror during recovery");
+		fail_mirror(get_default_mirror(ms), DM_RAID1_SYNC_ERROR);
+	}
 
-	if (write_err)
+	if (write_err) {
 		DMERR_LIMIT("Write error during recovery (error = 0x%x)",
 			    write_err);
+		/*
+		 * Bits correspond to devices (excluding default mirror).
+		 * The default mirror cannot change during recovery.
+		 */
+		for (m = 0; m < ms->nr_mirrors; m++) {
+			if (&ms->mirror[m] == get_default_mirror(ms))
+				continue;
+			if (test_bit(bit, &write_err))
+				fail_mirror(ms->mirror + m,
+					    DM_RAID1_SYNC_ERROR);
+			bit++;
+		}
+	}
 
 	rh_recovery_end(reg, !(read_err || write_err));
 }
@@ -678,7 +795,7 @@ static int recover(struct mirror_set *ms, struct region *reg)
 	unsigned long flags = 0;
 
 	/* fill in the source */
-	m = ms->default_mirror;
+	m = get_default_mirror(ms);
 	from.bdev = m->dev->bdev;
 	from.sector = m->offset + region_to_sector(reg->rh, reg->key);
 	if (reg->key == (ms->nr_regions - 1)) {
@@ -694,7 +811,7 @@ static int recover(struct mirror_set *ms, struct region *reg)
 
 	/* fill in the destinations */
 	for (i = 0, dest = to; i < ms->nr_mirrors; i++) {
-		if (&ms->mirror[i] == ms->default_mirror)
+		if (&ms->mirror[i] == get_default_mirror(ms))
 			continue;
 
 		m = ms->mirror + i;
@@ -748,17 +865,105 @@ static void do_recovery(struct mirror_set *ms)
  *---------------------------------------------------------------*/
 static struct mirror *choose_mirror(struct mirror_set *ms, sector_t sector)
 {
-	/* FIXME: add read balancing */
-	return ms->default_mirror;
+	struct mirror *m = get_default_mirror(ms);
+
+	do {
+		if (likely(!atomic_read(&m->error_count)))
+			return m;
+
+		if (m-- == ms->mirror)
+			m += ms->nr_mirrors;
+	} while (m != get_default_mirror(ms));
+
+	return NULL;
+}
+
+static int default_ok(struct mirror *m)
+{
+	struct mirror *default_mirror = get_default_mirror(m->ms);
+
+	return !atomic_read(&default_mirror->error_count);
+}
+
+static int mirror_available(struct mirror_set *ms, struct bio *bio)
+{
+	region_t region = bio_to_region(&ms->rh, bio);
+
+	if (ms->rh.log->type->in_sync(ms->rh.log, region, 0))
+		return choose_mirror(ms,  bio->bi_sector) ? 1 : 0;
+
+	return 0;
 }
 
 /*
  * remap a buffer to a particular mirror.
  */
-static void map_bio(struct mirror_set *ms, struct mirror *m, struct bio *bio)
+static sector_t map_sector(struct mirror *m, struct bio *bio)
+{
+	return m->offset + (bio->bi_sector - m->ms->ti->begin);
+}
+
+static void map_bio(struct mirror *m, struct bio *bio)
 {
 	bio->bi_bdev = m->dev->bdev;
-	bio->bi_sector = m->offset + (bio->bi_sector - ms->ti->begin);
+	bio->bi_sector = map_sector(m, bio);
+}
+
+static void map_region(struct io_region *io, struct mirror *m,
+		       struct bio *bio)
+{
+	io->bdev = m->dev->bdev;
+	io->sector = map_sector(m, bio);
+	io->count = bio->bi_size >> 9;
+}
+
+/*-----------------------------------------------------------------
+ * Reads
+ *---------------------------------------------------------------*/
+static void read_callback(unsigned long error, void *context)
+{
+	struct bio *bio = context;
+	struct mirror *m;
+
+	m = bio_get_m(bio);
+	bio_set_m(bio, NULL);
+
+	if (likely(!error)) {
+		bio_endio(bio, 0);
+		return;
+	}
+
+	fail_mirror(m, DM_RAID1_READ_ERROR);
+
+	if (likely(default_ok(m)) || mirror_available(m->ms, bio)) {
+		DMWARN_LIMIT("Read failure on mirror device %s.  "
+			     "Trying alternative device.",
+			     m->dev->name);
+		queue_bio(m->ms, bio, bio_rw(bio));
+		return;
+	}
+
+	DMERR_LIMIT("Read failure on mirror device %s.  Failing I/O.",
+		    m->dev->name);
+	bio_endio(bio, -EIO);
+}
+
+/* Asynchronous read. */
+static void read_async_bio(struct mirror *m, struct bio *bio)
+{
+	struct io_region io;
+	struct dm_io_request io_req = {
+		.bi_rw = READ,
+		.mem.type = DM_IO_BVEC,
+		.mem.ptr.bvec = bio->bi_io_vec + bio->bi_idx,
+		.notify.fn = read_callback,
+		.notify.context = bio,
+		.client = m->ms->io_client,
+	};
+
+	map_region(&io, m, bio);
+	bio_set_m(bio, m);
+	(void) dm_io(&io_req, 1, &io, NULL);
 }
 
 static void do_reads(struct mirror_set *ms, struct bio_list *reads)
@@ -769,17 +974,20 @@ static void do_reads(struct mirror_set *ms, struct bio_list *reads)
 
 	while ((bio = bio_list_pop(reads))) {
 		region = bio_to_region(&ms->rh, bio);
+		m = get_default_mirror(ms);
 
 		/*
 		 * We can only read balance if the region is in sync.
 		 */
-		if (rh_in_sync(&ms->rh, region, 1))
+		if (likely(rh_in_sync(&ms->rh, region, 1)))
 			m = choose_mirror(ms, bio->bi_sector);
-		else
-			m = ms->default_mirror;
+		else if (m && atomic_read(&m->error_count))
+			m = NULL;
 
-		map_bio(ms, m, bio);
-		generic_make_request(bio);
+		if (likely(m))
+			read_async_bio(m, bio);
+		else
+			bio_endio(bio, -EIO);
 	}
 }
 
@@ -793,15 +1001,70 @@ static void do_reads(struct mirror_set *ms, struct bio_list *reads)
  * RECOVERING:	delay the io until recovery completes
  * NOSYNC:	increment pending, just write to the default mirror
  *---------------------------------------------------------------*/
+
+/* __bio_mark_nosync
+ * @ms
+ * @bio
+ * @done
+ * @error
+ *
+ * The bio was written on some mirror(s) but failed on other mirror(s).
+ * We can successfully endio the bio but should avoid the region being
+ * marked clean by setting the state RH_NOSYNC.
+ *
+ * This function is _not_ safe in interrupt context!
+ */
+static void __bio_mark_nosync(struct mirror_set *ms,
+			      struct bio *bio, unsigned done, int error)
+{
+	unsigned long flags;
+	struct region_hash *rh = &ms->rh;
+	struct dirty_log *log = ms->rh.log;
+	struct region *reg;
+	region_t region = bio_to_region(rh, bio);
+	int recovering = 0;
+
+	/* We must inform the log that the sync count has changed. */
+	log->type->set_region_sync(log, region, 0);
+	ms->in_sync = 0;
+
+	read_lock(&rh->hash_lock);
+	reg = __rh_find(rh, region);
+	read_unlock(&rh->hash_lock);
+
+	/* region hash entry should exist because write was in-flight */
+	BUG_ON(!reg);
+	BUG_ON(!list_empty(&reg->list));
+
+	spin_lock_irqsave(&rh->region_lock, flags);
+	/*
+	 * Possible cases:
+	 *   1) RH_DIRTY
+	 *   2) RH_NOSYNC: was dirty, other preceeding writes failed
+	 *   3) RH_RECOVERING: flushing pending writes
+	 * Either case, the region should have not been connected to list.
+	 */
+	recovering = (reg->state == RH_RECOVERING);
+	reg->state = RH_NOSYNC;
+	BUG_ON(!list_empty(&reg->list));
+	spin_unlock_irqrestore(&rh->region_lock, flags);
+
+	bio_endio(bio, error);
+	if (recovering)
+		complete_resync_work(reg, 0);
+}
+
 static void write_callback(unsigned long error, void *context)
 {
-	unsigned int i;
-	int uptodate = 1;
+	unsigned i, ret = 0;
 	struct bio *bio = (struct bio *) context;
 	struct mirror_set *ms;
+	int uptodate = 0;
+	int should_wake = 0;
+	unsigned long flags;
 
-	ms = bio_get_ms(bio);
-	bio_set_ms(bio, NULL);
+	ms = bio_get_m(bio)->ms;
+	bio_set_m(bio, NULL);
 
 	/*
 	 * NOTE: We don't decrement the pending count here,
@@ -809,26 +1072,42 @@ static void write_callback(unsigned long error, void *context)
 	 * This way we handle both writes to SYNC and NOSYNC
 	 * regions with the same code.
 	 */
+	if (likely(!error))
+		goto out;
 
-	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)) {
+		if (test_bit(i, &error))
+			fail_mirror(ms->mirror + i, DM_RAID1_WRITE_ERROR);
+		else
 			uptodate = 1;
-				break;
-			}
+
+	if (unlikely(!uptodate)) {
+		DMERR("All replicated volumes dead, failing I/O");
+		/* None of the writes succeeded, fail the I/O. */
+		ret = -EIO;
+	} else if (errors_handled(ms)) {
+		/*
+		 * Need to raise event.  Since raising
+		 * events can block, we need to do it in
+		 * the main thread.
+		 */
+		spin_lock_irqsave(&ms->lock, flags);
+		if (!ms->failures.head)
+			should_wake = 1;
+		bio_list_add(&ms->failures, bio);
+		spin_unlock_irqrestore(&ms->lock, flags);
+		if (should_wake)
+			wake(ms);
+		return;
 	}
-	bio_endio(bio, 0);
+out:
+	bio_endio(bio, ret);
 }
 
 static void do_write(struct mirror_set *ms, struct bio *bio)
 {
 	unsigned int i;
-	struct io_region io[KCOPYD_MAX_REGIONS+1];
+	struct io_region io[ms->nr_mirrors], *dest = io;
 	struct mirror *m;
 	struct dm_io_request io_req = {
 		.bi_rw = WRITE,
@@ -839,15 +1118,14 @@ static void do_write(struct mirror_set *ms, struct bio *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;
-	}
+	for (i = 0, m = ms->mirror; i < ms->nr_mirrors; i++, m++)
+		map_region(dest++, m, bio);
 
-	bio_set_ms(bio, ms);
+	/*
+	 * Use default mirror because we only need it to retrieve the reference
+	 * to the mirror set in write_callback().
+	 */
+	bio_set_m(bio, get_default_mirror(ms));
 
 	(void) dm_io(&io_req, ms->nr_mirrors, io, NULL);
 }
@@ -900,43 +1178,125 @@ static void do_writes(struct mirror_set *ms, struct bio_list *writes)
 	/*
 	 * Dispatch io.
 	 */
-	if (unlikely(ms->log_failure))
+	if (unlikely(ms->log_failure)) {
+		spin_lock_irq(&ms->lock);
+		bio_list_merge(&ms->failures, &sync);
+		spin_unlock_irq(&ms->lock);
+	} else
 		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);
+		map_bio(get_default_mirror(ms), bio);
 		generic_make_request(bio);
 	}
 }
 
+static void do_failures(struct mirror_set *ms, struct bio_list *failures)
+{
+	struct bio *bio;
+
+	if (!failures->head)
+		return;
+
+	if (!ms->log_failure) {
+		while ((bio = bio_list_pop(failures)))
+			__bio_mark_nosync(ms, bio, bio->bi_size, 0);
+		return;
+	}
+
+	/*
+	 * If the log has failed, unattempted writes are being
+	 * put on the failures list.  We can't issue those writes
+	 * until a log has been marked, so we must store them.
+	 *
+	 * If a 'noflush' suspend is in progress, we can requeue
+	 * the I/O's to the core.  This give userspace a chance
+	 * to reconfigure the mirror, at which point the core
+	 * will reissue the writes.  If the 'noflush' flag is
+	 * not set, we have no choice but to return errors.
+	 *
+	 * Some writes on the failures list may have been
+	 * submitted before the log failure and represent a
+	 * failure to write to one of the devices.  It is ok
+	 * for us to treat them the same and requeue them
+	 * as well.
+	 */
+	if (dm_noflush_suspending(ms->ti)) {
+		while ((bio = bio_list_pop(failures)))
+			bio_endio(bio, DM_ENDIO_REQUEUE);
+		return;
+	}
+
+	if (atomic_read(&ms->suspend)) {
+		while ((bio = bio_list_pop(failures)))
+			bio_endio(bio, -EIO);
+		return;
+	}
+
+	spin_lock_irq(&ms->lock);
+	bio_list_merge(&ms->failures, failures);
+	spin_unlock_irq(&ms->lock);
+
+	wake(ms);
+}
+
+static void trigger_event(struct work_struct *work)
+{
+	struct mirror_set *ms =
+		container_of(work, struct mirror_set, trigger_event);
+
+	dm_table_event(ms->ti->table);
+}
+
 /*-----------------------------------------------------------------
  * kmirrord
  *---------------------------------------------------------------*/
-static void do_mirror(struct work_struct *work)
+static int _do_mirror(struct work_struct *work)
 {
 	struct mirror_set *ms =container_of(work, struct mirror_set,
 					    kmirrord_work);
-	struct bio_list reads, writes;
+	struct bio_list reads, writes, failures;
+	unsigned long flags;
 
-	spin_lock(&ms->lock);
+	spin_lock_irqsave(&ms->lock, flags);
 	reads = ms->reads;
 	writes = ms->writes;
+	failures = ms->failures;
 	bio_list_init(&ms->reads);
 	bio_list_init(&ms->writes);
-	spin_unlock(&ms->lock);
+	bio_list_init(&ms->failures);
+	spin_unlock_irqrestore(&ms->lock, flags);
 
 	rh_update_states(&ms->rh);
 	do_recovery(ms);
 	do_reads(ms, &reads);
 	do_writes(ms, &writes);
+	do_failures(ms, &failures);
+
+	return (ms->failures.head) ? 1 : 0;
 }
 
+static void do_mirror(struct work_struct *work)
+{
+	/*
+	 * If _do_mirror returns 1, we give it
+	 * another shot.  This helps for cases like
+	 * 'suspend' where we call flush_workqueue
+	 * and expect all work to be finished.  If
+	 * a failure happens during a suspend, we
+	 * couldn't issue a 'wake' because it would
+	 * not be honored.  Therefore, we return '1'
+	 * from _do_mirror, and retry here.
+	 */
+	while (_do_mirror(work))
+		schedule();
+}
+
+
 /*-----------------------------------------------------------------
  * Target functions
  *---------------------------------------------------------------*/
@@ -965,11 +1325,23 @@ static struct mirror_set *alloc_context(unsigned int nr_mirrors,
 	ms->nr_mirrors = nr_mirrors;
 	ms->nr_regions = dm_sector_div_up(ti->len, region_size);
 	ms->in_sync = 0;
-	ms->default_mirror = &ms->mirror[DEFAULT_MIRROR];
+	ms->log_failure = 0;
+	atomic_set(&ms->suspend, 0);
+	atomic_set(&ms->default_mirror, DEFAULT_MIRROR);
+
+	len = sizeof(struct dm_raid1_read_record);
+	ms->read_record_pool = mempool_create_kmalloc_pool(MIN_READ_RECORDS,
+							   len);
+	if (!ms->read_record_pool) {
+		ti->error = "Error creating mirror read_record_pool";
+		kfree(ms);
+		return NULL;
+	}
 
 	ms->io_client = dm_io_client_create(DM_IO_PAGES);
 	if (IS_ERR(ms->io_client)) {
 		ti->error = "Error creating dm_io client";
+		mempool_destroy(ms->read_record_pool);
 		kfree(ms);
  		return NULL;
 	}
@@ -977,6 +1349,7 @@ static struct mirror_set *alloc_context(unsigned int nr_mirrors,
 	if (rh_init(&ms->rh, ms, dl, region_size, ms->nr_regions)) {
 		ti->error = "Error creating dirty region hash";
 		dm_io_client_destroy(ms->io_client);
+		mempool_destroy(ms->read_record_pool);
 		kfree(ms);
 		return NULL;
 	}
@@ -992,6 +1365,7 @@ static void free_context(struct mirror_set *ms, struct dm_target *ti,
 
 	dm_io_client_destroy(ms->io_client);
 	rh_exit(&ms->rh);
+	mempool_destroy(ms->read_record_pool);
 	kfree(ms);
 }
 
@@ -1019,6 +1393,8 @@ static int get_mirror(struct mirror_set *ms, struct dm_target *ti,
 	}
 
 	ms->mirror[mirror].ms = ms;
+	atomic_set(&(ms->mirror[mirror].error_count), 0);
+	ms->mirror[mirror].error_type = 0;
 	ms->mirror[mirror].offset = offset;
 
 	return 0;
@@ -1171,6 +1547,7 @@ static int mirror_ctr(struct dm_target *ti, unsigned int argc, char **argv)
 		goto err_free_context;
 	}
 	INIT_WORK(&ms->kmirrord_work, do_mirror);
+	INIT_WORK(&ms->trigger_event, trigger_event);
 
 	r = parse_features(ms, argc, argv, &args_used);
 	if (r)
@@ -1220,14 +1597,15 @@ static void mirror_dtr(struct dm_target *ti)
 
 static void queue_bio(struct mirror_set *ms, struct bio *bio, int rw)
 {
+	unsigned long flags;
 	int should_wake = 0;
 	struct bio_list *bl;
 
 	bl = (rw == WRITE) ? &ms->writes : &ms->reads;
-	spin_lock(&ms->lock);
+	spin_lock_irqsave(&ms->lock, flags);
 	should_wake = !(bl->head);
 	bio_list_add(bl, bio);
-	spin_unlock(&ms->lock);
+	spin_unlock_irqrestore(&ms->lock, flags);
 
 	if (should_wake)
 		wake(ms);
@@ -1242,10 +1620,11 @@ static int mirror_map(struct dm_target *ti, struct bio *bio,
 	int r, rw = bio_rw(bio);
 	struct mirror *m;
 	struct mirror_set *ms = ti->private;
-
-	map_context->ll = bio_to_region(&ms->rh, bio);
+	struct dm_raid1_read_record *read_record = NULL;
 
 	if (rw == WRITE) {
+		/* Save region for mirror_end_io() handler */
+		map_context->ll = bio_to_region(&ms->rh, bio);
 		queue_bio(ms, bio, rw);
 		return DM_MAPIO_SUBMITTED;
 	}
@@ -1255,28 +1634,34 @@ static int mirror_map(struct dm_target *ti, struct bio *bio,
 	if (r < 0 && r != -EWOULDBLOCK)
 		return r;
 
-	if (r == -EWOULDBLOCK)	/* FIXME: ugly */
-		r = DM_MAPIO_SUBMITTED;
-
 	/*
-	 * We don't want to fast track a recovery just for a read
-	 * ahead.  So we just let it silently fail.
-	 * FIXME: get rid of this.
+	 * If region is not in-sync queue the bio.
 	 */
-	if (!r && rw == READA)
-		return -EIO;
+	if (!r || (r == -EWOULDBLOCK)) {
+		if (rw == READA)
+			return -EWOULDBLOCK;
 
-	if (!r) {
-		/* Pass this io over to the daemon */
 		queue_bio(ms, bio, rw);
 		return DM_MAPIO_SUBMITTED;
 	}
 
+	/*
+	 * The region is in-sync and we can perform reads directly.
+	 * Store enough information so we can retry if it fails.
+	 */
 	m = choose_mirror(ms, bio->bi_sector);
-	if (!m)
+	if (unlikely(!m))
 		return -EIO;
 
-	map_bio(ms, m, bio);
+	read_record = mempool_alloc(ms->read_record_pool, GFP_NOIO);
+	if (likely(read_record)) {
+		dm_bio_record(&read_record->details, bio);
+		map_context->ptr = read_record;
+		read_record->m = m;
+	}
+
+	map_bio(m, bio);
+
 	return DM_MAPIO_REMAPPED;
 }
 
@@ -1285,66 +1670,168 @@ static int mirror_end_io(struct dm_target *ti, struct bio *bio,
 {
 	int rw = bio_rw(bio);
 	struct mirror_set *ms = (struct mirror_set *) ti->private;
-	region_t region = map_context->ll;
+	struct mirror *m = NULL;
+	struct dm_bio_details *bd = NULL;
+	struct dm_raid1_read_record *read_record = map_context->ptr;
 
 	/*
 	 * We need to dec pending if this was a write.
 	 */
-	if (rw == WRITE)
-		rh_dec(&ms->rh, region);
+	if (rw == WRITE) {
+		rh_dec(&ms->rh, map_context->ll);
+		return error;
+	}
 
-	return 0;
+	if (error == -EOPNOTSUPP)
+		goto out;
+
+	if ((error == -EWOULDBLOCK) && bio_rw_ahead(bio))
+		goto out;
+
+	if (unlikely(error)) {
+		if (!read_record) {
+			/*
+			 * There wasn't enough memory to record necessary
+			 * information for a retry or there was no other
+			 * mirror in-sync.
+			 */
+			DMERR_LIMIT("Mirror read failed from %s.",
+				    m->dev->name);
+			return -EIO;
+		}
+		DMERR("Mirror read failed from %s. Trying alternative device.",
+		      m->dev->name);
+
+		m = read_record->m;
+		fail_mirror(m, DM_RAID1_READ_ERROR);
+
+		/*
+		 * A failed read is requeued for another attempt using an intact
+		 * mirror.
+		 */
+		if (default_ok(m) || mirror_available(ms, bio)) {
+			bd = &read_record->details;
+
+			dm_bio_restore(bd, bio);
+			mempool_free(read_record, ms->read_record_pool);
+			map_context->ptr = NULL;
+			queue_bio(ms, bio, rw);
+			return 1;
+		}
+		DMERR("All replicated volumes dead, failing I/O");
+	}
+
+out:
+	if (read_record) {
+		mempool_free(read_record, ms->read_record_pool);
+		map_context->ptr = NULL;
+	}
+
+	return error;
 }
 
-static void mirror_postsuspend(struct dm_target *ti)
+static void mirror_presuspend(struct dm_target *ti)
 {
 	struct mirror_set *ms = (struct mirror_set *) ti->private;
 	struct dirty_log *log = ms->rh.log;
 
+	atomic_set(&ms->suspend, 1);
+
+	/*
+	 * We must finish up all the work that we've
+	 * generated (i.e. recovery work).
+	 */
 	rh_stop_recovery(&ms->rh);
 
-	/* Wait for all I/O we generated to complete */
 	wait_event(_kmirrord_recovery_stopped,
 		   !atomic_read(&ms->rh.recovery_in_flight));
 
+	if (log->type->presuspend && log->type->presuspend(log))
+		/* FIXME: need better error handling */
+		DMWARN("log presuspend failed");
+
+	/*
+	 * Now that recovery is complete/stopped and the
+	 * delayed bios are queued, we need to wait for
+	 * the worker thread to complete.  This way,
+	 * we know that all of our I/O has been pushed.
+	 */
+	flush_workqueue(ms->kmirrord_wq);
+}
+
+static void mirror_postsuspend(struct dm_target *ti)
+{
+	struct mirror_set *ms = ti->private;
+	struct dirty_log *log = ms->rh.log;
+
 	if (log->type->postsuspend && log->type->postsuspend(log))
 		/* FIXME: need better error handling */
-		DMWARN("log suspend failed");
+		DMWARN("log postsuspend failed");
 }
 
 static void mirror_resume(struct dm_target *ti)
 {
-	struct mirror_set *ms = (struct mirror_set *) ti->private;
+	struct mirror_set *ms = ti->private;
 	struct dirty_log *log = ms->rh.log;
+
+	atomic_set(&ms->suspend, 0);
 	if (log->type->resume && log->type->resume(log))
 		/* FIXME: need better error handling */
 		DMWARN("log resume failed");
 	rh_start_recovery(&ms->rh);
 }
 
+/*
+ * device_status_char
+ * @m: mirror device/leg we want the status of
+ *
+ * We return one character representing the most severe error
+ * we have encountered.
+ *    A => Alive - No failures
+ *    D => Dead - A write failure occurred leaving mirror out-of-sync
+ *    S => Sync - A sychronization failure occurred, mirror out-of-sync
+ *    R => Read - A read failure occurred, mirror data unaffected
+ *
+ * Returns: <char>
+ */
+static char device_status_char(struct mirror *m)
+{
+	if (!atomic_read(&(m->error_count)))
+		return 'A';
+
+	return (test_bit(DM_RAID1_WRITE_ERROR, &(m->error_type))) ? 'D' :
+		(test_bit(DM_RAID1_SYNC_ERROR, &(m->error_type))) ? 'S' :
+		(test_bit(DM_RAID1_READ_ERROR, &(m->error_type))) ? 'R' : 'U';
+}
+
+
 static int mirror_status(struct dm_target *ti, status_type_t type,
 			 char *result, unsigned int maxlen)
 {
 	unsigned int m, sz = 0;
 	struct mirror_set *ms = (struct mirror_set *) ti->private;
+	struct dirty_log *log = ms->rh.log;
+	char buffer[ms->nr_mirrors + 1];
 
 	switch (type) {
 	case STATUSTYPE_INFO:
 		DMEMIT("%d ", ms->nr_mirrors);
-		for (m = 0; m < ms->nr_mirrors; m++)
+		for (m = 0; m < ms->nr_mirrors; m++) {
 			DMEMIT("%s ", ms->mirror[m].dev->name);
+			buffer[m] = device_status_char(&(ms->mirror[m]));
+		}
+		buffer[m] = '\0';
 
-		DMEMIT("%llu/%llu 0 ",
-			(unsigned long long)ms->rh.log->type->
-				get_sync_count(ms->rh.log),
-			(unsigned long long)ms->nr_regions);
+		DMEMIT("%llu/%llu 1 %s ",
+		      (unsigned long long)log->type->get_sync_count(ms->rh.log),
+		      (unsigned long long)ms->nr_regions, buffer);
 
-		sz += ms->rh.log->type->status(ms->rh.log, type, result+sz, maxlen-sz);
+		sz += log->type->status(ms->rh.log, type, result+sz, maxlen-sz);
 
 		break;
 
 	case STATUSTYPE_TABLE:
-		sz = ms->rh.log->type->status(ms->rh.log, type, result, maxlen);
+		sz = log->type->status(ms->rh.log, type, result, maxlen);
 
 		DMEMIT("%d", ms->nr_mirrors);
 		for (m = 0; m < ms->nr_mirrors; m++)
@@ -1360,12 +1847,13 @@ static int mirror_status(struct dm_target *ti, status_type_t type,
 
 static struct target_type mirror_target = {
 	.name	 = "mirror",
-	.version = {1, 0, 3},
+	.version = {1, 0, 20},
 	.module	 = THIS_MODULE,
 	.ctr	 = mirror_ctr,
 	.dtr	 = mirror_dtr,
 	.map	 = mirror_map,
 	.end_io	 = mirror_end_io,
+	.presuspend = mirror_presuspend,
 	.postsuspend = mirror_postsuspend,
 	.resume	 = mirror_resume,
 	.status	 = mirror_status,

drivers/md/dm-snap.c

@@ -213,11 +213,15 @@ static void unregister_snapshot(struct dm_snapshot *s)
 
 /*
  * Implementation of the exception hash tables.
+ * The lowest hash_shift bits of the chunk number are ignored, allowing
+ * some consecutive chunks to be grouped together.
  */
-static int init_exception_table(struct exception_table *et, uint32_t size)
+static int init_exception_table(struct exception_table *et, uint32_t size,
+				unsigned hash_shift)
 {
 	unsigned int i;
 
+	et->hash_shift = hash_shift;
 	et->hash_mask = size - 1;
 	et->table = dm_vcalloc(size, sizeof(struct list_head));
 	if (!et->table)
@@ -248,7 +252,7 @@ static void exit_exception_table(struct exception_table *et, struct kmem_cache *
 
 static uint32_t exception_hash(struct exception_table *et, chunk_t chunk)
 {
-	return chunk & et->hash_mask;
+	return (chunk >> et->hash_shift) & et->hash_mask;
 }
 
 static void insert_exception(struct exception_table *eh,
@@ -275,7 +279,8 @@ static struct dm_snap_exception *lookup_exception(struct exception_table *et,
 
 	slot = &et->table[exception_hash(et, chunk)];
 	list_for_each_entry (e, slot, hash_list)
-		if (e->old_chunk == chunk)
+		if (chunk >= e->old_chunk &&
+		    chunk <= e->old_chunk + dm_consecutive_chunk_count(e))
 			return e;
 
 	return NULL;
@@ -307,6 +312,49 @@ static void free_pending_exception(struct dm_snap_pending_exception *pe)
 	mempool_free(pe, pending_pool);
 }
 
+static void insert_completed_exception(struct dm_snapshot *s,
+				       struct dm_snap_exception *new_e)
+{
+	struct exception_table *eh = &s->complete;
+	struct list_head *l;
+	struct dm_snap_exception *e = NULL;
+
+	l = &eh->table[exception_hash(eh, new_e->old_chunk)];
+
+	/* Add immediately if this table doesn't support consecutive chunks */
+	if (!eh->hash_shift)
+		goto out;
+
+	/* List is ordered by old_chunk */
+	list_for_each_entry_reverse(e, l, hash_list) {
+		/* Insert after an existing chunk? */
+		if (new_e->old_chunk == (e->old_chunk +
+					 dm_consecutive_chunk_count(e) + 1) &&
+		    new_e->new_chunk == (dm_chunk_number(e->new_chunk) +
+					 dm_consecutive_chunk_count(e) + 1)) {
+			dm_consecutive_chunk_count_inc(e);
+			free_exception(new_e);
+			return;
+		}
+
+		/* Insert before an existing chunk? */
+		if (new_e->old_chunk == (e->old_chunk - 1) &&
+		    new_e->new_chunk == (dm_chunk_number(e->new_chunk) - 1)) {
+			dm_consecutive_chunk_count_inc(e);
+			e->old_chunk--;
+			e->new_chunk--;
+			free_exception(new_e);
+			return;
+		}
+
+		if (new_e->old_chunk > e->old_chunk)
+			break;
+	}
+
+out:
+	list_add(&new_e->hash_list, e ? &e->hash_list : l);
+}
+
 int dm_add_exception(struct dm_snapshot *s, chunk_t old, chunk_t new)
 {
 	struct dm_snap_exception *e;
@@ -316,8 +364,12 @@ int dm_add_exception(struct dm_snapshot *s, chunk_t old, chunk_t new)
 		return -ENOMEM;
 
 	e->old_chunk = old;
+
+	/* Consecutive_count is implicitly initialised to zero */
 	e->new_chunk = new;
-	insert_exception(&s->complete, e);
+
+	insert_completed_exception(s, e);
+
 	return 0;
 }
 
@@ -333,16 +385,6 @@ static int calc_max_buckets(void)
 	return mem;
 }
 
-/*
- * Rounds a number down to a power of 2.
- */
-static uint32_t round_down(uint32_t n)
-{
-	while (n & (n - 1))
-		n &= (n - 1);
-	return n;
-}
-
 /*
  * Allocate room for a suitable hash table.
  */
@@ -361,9 +403,9 @@ static int init_hash_tables(struct dm_snapshot *s)
 	hash_size = min(origin_dev_size, cow_dev_size) >> s->chunk_shift;
 	hash_size = min(hash_size, max_buckets);
 
-	/* Round it down to a power of 2 */
-	hash_size = round_down(hash_size);
-	if (init_exception_table(&s->complete, hash_size))
+	hash_size = rounddown_pow_of_two(hash_size);
+	if (init_exception_table(&s->complete, hash_size,
+				 DM_CHUNK_CONSECUTIVE_BITS))
 		return -ENOMEM;
 
 	/*
@@ -374,7 +416,7 @@ static int init_hash_tables(struct dm_snapshot *s)
 	if (hash_size < 64)
 		hash_size = 64;
 
-	if (init_exception_table(&s->pending, hash_size)) {
+	if (init_exception_table(&s->pending, hash_size, 0)) {
 		exit_exception_table(&s->complete, exception_cache);
 		return -ENOMEM;
 	}
@@ -733,7 +775,7 @@ static void pending_complete(struct dm_snap_pending_exception *pe, int success)
 	 * Add a proper exception, and remove the
 	 * in-flight exception from the list.
 	 */
-	insert_exception(&s->complete, e);
+	insert_completed_exception(s, e);
 
  out:
 	remove_exception(&pe->e);
@@ -867,10 +909,11 @@ __find_pending_exception(struct dm_snapshot *s, struct bio *bio)
 }
 
 static void remap_exception(struct dm_snapshot *s, struct dm_snap_exception *e,
-			    struct bio *bio)
+			    struct bio *bio, chunk_t chunk)
 {
 	bio->bi_bdev = s->cow->bdev;
-	bio->bi_sector = chunk_to_sector(s, e->new_chunk) +
+	bio->bi_sector = chunk_to_sector(s, dm_chunk_number(e->new_chunk) +
+			 (chunk - e->old_chunk)) +
 			 (bio->bi_sector & s->chunk_mask);
 }
 
@@ -902,7 +945,7 @@ static int snapshot_map(struct dm_target *ti, struct bio *bio,
 	/* If the block is already remapped - use that, else remap it */
 	e = lookup_exception(&s->complete, chunk);
 	if (e) {
-		remap_exception(s, e, bio);
+		remap_exception(s, e, bio, chunk);
 		goto out_unlock;
 	}
 
@@ -919,7 +962,7 @@ static int snapshot_map(struct dm_target *ti, struct bio *bio,
 			goto out_unlock;
 		}
 
-		remap_exception(s, &pe->e, bio);
+		remap_exception(s, &pe->e, bio, chunk);
 		bio_list_add(&pe->snapshot_bios, bio);
 
 		r = DM_MAPIO_SUBMITTED;
@@ -1207,7 +1250,7 @@ static int origin_status(struct dm_target *ti, status_type_t type, char *result,
 
 static struct target_type origin_target = {
 	.name    = "snapshot-origin",
-	.version = {1, 5, 0},
+	.version = {1, 6, 0},
 	.module  = THIS_MODULE,
 	.ctr     = origin_ctr,
 	.dtr     = origin_dtr,
@@ -1218,7 +1261,7 @@ static struct target_type origin_target = {
 
 static struct target_type snapshot_target = {
 	.name    = "snapshot",
-	.version = {1, 5, 0},
+	.version = {1, 6, 0},
 	.module  = THIS_MODULE,
 	.ctr     = snapshot_ctr,
 	.dtr     = snapshot_dtr,

drivers/md/dm-snap.h

@@ -16,19 +16,22 @@
 
 struct exception_table {
 	uint32_t hash_mask;
+	unsigned hash_shift;
 	struct list_head *table;
 };
 
 /*
  * The snapshot code deals with largish chunks of the disk at a
- * time. Typically 64k - 256k.
+ * time. Typically 32k - 512k.
  */
-/* FIXME: can we get away with limiting these to a uint32_t ? */
 typedef sector_t chunk_t;
 
 /*
  * An exception is used where an old chunk of data has been
  * replaced by a new one.
+ * If chunk_t is 64 bits in size, the top 8 bits of new_chunk hold the number
+ * of chunks that follow contiguously.  Remaining bits hold the number of the
+ * chunk within the device.
  */
 struct dm_snap_exception {
 	struct list_head hash_list;
@@ -37,6 +40,49 @@ struct dm_snap_exception {
 	chunk_t new_chunk;
 };
 
+/*
+ * Funtions to manipulate consecutive chunks
+ */
+#  if defined(CONFIG_LBD) || (BITS_PER_LONG == 64)
+#    define DM_CHUNK_CONSECUTIVE_BITS 8
+#    define DM_CHUNK_NUMBER_BITS 56
+
+static inline chunk_t dm_chunk_number(chunk_t chunk)
+{
+	return chunk & (chunk_t)((1ULL << DM_CHUNK_NUMBER_BITS) - 1ULL);
+}
+
+static inline unsigned dm_consecutive_chunk_count(struct dm_snap_exception *e)
+{
+	return e->new_chunk >> DM_CHUNK_NUMBER_BITS;
+}
+
+static inline void dm_consecutive_chunk_count_inc(struct dm_snap_exception *e)
+{
+	e->new_chunk += (1ULL << DM_CHUNK_NUMBER_BITS);
+
+	BUG_ON(!dm_consecutive_chunk_count(e));
+}
+
+#  else
+#    define DM_CHUNK_CONSECUTIVE_BITS 0
+
+static inline chunk_t dm_chunk_number(chunk_t chunk)
+{
+	return chunk;
+}
+
+static inline unsigned dm_consecutive_chunk_count(struct dm_snap_exception *e)
+{
+	return 0;
+}
+
+static inline void dm_consecutive_chunk_count_inc(struct dm_snap_exception *e)
+{
+}
+
+#  endif
+
 /*
  * Abstraction to handle the meta/layout of exception stores (the
  * COW device).

drivers/md/dm-stripe.c

@@ -14,10 +14,13 @@
 #include <linux/log2.h>
 
 #define DM_MSG_PREFIX "striped"
+#define DM_IO_ERROR_THRESHOLD 15
 
 struct stripe {
 	struct dm_dev *dev;
 	sector_t physical_start;
+
+	atomic_t error_count;
 };
 
 struct stripe_c {
@@ -30,9 +33,29 @@ struct stripe_c {
 	uint32_t chunk_shift;
 	sector_t chunk_mask;
 
+	/* Needed for handling events */
+	struct dm_target *ti;
+
+	/* Work struct used for triggering events*/
+	struct work_struct kstriped_ws;
+
 	struct stripe stripe[0];
 };
 
+static struct workqueue_struct *kstriped;
+
+/*
+ * An event is triggered whenever a drive
+ * drops out of a stripe volume.
+ */
+static void trigger_event(struct work_struct *work)
+{
+	struct stripe_c *sc = container_of(work, struct stripe_c, kstriped_ws);
+
+	dm_table_event(sc->ti->table);
+
+}
+
 static inline struct stripe_c *alloc_context(unsigned int stripes)
 {
 	size_t len;
@@ -63,6 +86,7 @@ static int get_stripe(struct dm_target *ti, struct stripe_c *sc,
 		return -ENXIO;
 
 	sc->stripe[stripe].physical_start = start;
+
 	return 0;
 }
 
@@ -135,6 +159,11 @@ static int stripe_ctr(struct dm_target *ti, unsigned int argc, char **argv)
 		return -ENOMEM;
 	}
 
+	INIT_WORK(&sc->kstriped_ws, trigger_event);
+
+	/* Set pointer to dm target; used in trigger_event */
+	sc->ti = ti;
+
 	sc->stripes = stripes;
 	sc->stripe_width = width;
 	ti->split_io = chunk_size;
@@ -158,9 +187,11 @@ static int stripe_ctr(struct dm_target *ti, unsigned int argc, char **argv)
 			kfree(sc);
 			return r;
 		}
+		atomic_set(&(sc->stripe[i].error_count), 0);
 	}
 
 	ti->private = sc;
+
 	return 0;
 }
 
@@ -172,6 +203,7 @@ static void stripe_dtr(struct dm_target *ti)
 	for (i = 0; i < sc->stripes; i++)
 		dm_put_device(ti, sc->stripe[i].dev);
 
+	flush_workqueue(kstriped);
 	kfree(sc);
 }
 
@@ -190,16 +222,37 @@ static int stripe_map(struct dm_target *ti, struct bio *bio,
 	return DM_MAPIO_REMAPPED;
 }
 
+/*
+ * Stripe status:
+ *
+ * INFO
+ * #stripes [stripe_name <stripe_name>] [group word count]
+ * [error count 'A|D' <error count 'A|D'>]
+ *
+ * TABLE
+ * #stripes [stripe chunk size]
+ * [stripe_name physical_start <stripe_name physical_start>]
+ *
+ */
+
 static int stripe_status(struct dm_target *ti,
 			 status_type_t type, char *result, unsigned int maxlen)
 {
 	struct stripe_c *sc = (struct stripe_c *) ti->private;
+	char buffer[sc->stripes + 1];
 	unsigned int sz = 0;
 	unsigned int i;
 
 	switch (type) {
 	case STATUSTYPE_INFO:
-		result[0] = '\0';
+		DMEMIT("%d ", sc->stripes);
+		for (i = 0; i < sc->stripes; i++)  {
+			DMEMIT("%s ", sc->stripe[i].dev->name);
+			buffer[i] = atomic_read(&(sc->stripe[i].error_count)) ?
+				'D' : 'A';
+		}
+		buffer[i] = '\0';
+		DMEMIT("1 %s", buffer);
 		break;
 
 	case STATUSTYPE_TABLE:
@@ -213,13 +266,52 @@ static int stripe_status(struct dm_target *ti,
 	return 0;
 }
 
+static int stripe_end_io(struct dm_target *ti, struct bio *bio,
+			 int error, union map_info *map_context)
+{
+	unsigned i;
+	char major_minor[16];
+	struct stripe_c *sc = ti->private;
+
+	if (!error)
+		return 0; /* I/O complete */
+
+	if ((error == -EWOULDBLOCK) && bio_rw_ahead(bio))
+		return error;
+
+	if (error == -EOPNOTSUPP)
+		return error;
+
+	memset(major_minor, 0, sizeof(major_minor));
+	sprintf(major_minor, "%d:%d",
+		bio->bi_bdev->bd_disk->major,
+		bio->bi_bdev->bd_disk->first_minor);
+
+	/*
+	 * Test to see which stripe drive triggered the event
+	 * and increment error count for all stripes on that device.
+	 * If the error count for a given device exceeds the threshold
+	 * value we will no longer trigger any further events.
+	 */
+	for (i = 0; i < sc->stripes; i++)
+		if (!strcmp(sc->stripe[i].dev->name, major_minor)) {
+			atomic_inc(&(sc->stripe[i].error_count));
+			if (atomic_read(&(sc->stripe[i].error_count)) <
+			    DM_IO_ERROR_THRESHOLD)
+				queue_work(kstriped, &sc->kstriped_ws);
+		}
+
+	return error;
+}
+
 static struct target_type stripe_target = {
 	.name   = "striped",
-	.version= {1, 0, 2},
+	.version = {1, 1, 0},
 	.module = THIS_MODULE,
 	.ctr    = stripe_ctr,
 	.dtr    = stripe_dtr,
 	.map    = stripe_map,
+	.end_io = stripe_end_io,
 	.status = stripe_status,
 };
 
@@ -231,6 +323,13 @@ int __init dm_stripe_init(void)
 	if (r < 0)
 		DMWARN("target registration failed");
 
+	kstriped = create_singlethread_workqueue("kstriped");
+	if (!kstriped) {
+		DMERR("failed to create workqueue kstriped");
+		dm_unregister_target(&stripe_target);
+		return -ENOMEM;
+	}
+
 	return r;
 }
 
@@ -239,5 +338,7 @@ void dm_stripe_exit(void)
 	if (dm_unregister_target(&stripe_target))
 		DMWARN("target unregistration failed");
 
+	destroy_workqueue(kstriped);
+
 	return;
 }

drivers/md/dm-table.c

@@ -287,9 +287,8 @@ static void free_devices(struct list_head *devices)
 {
 	struct list_head *tmp, *next;
 
-	for (tmp = devices->next; tmp != devices; tmp = next) {
+	list_for_each_safe(tmp, next, devices) {
 		struct dm_dev *dd = list_entry(tmp, struct dm_dev, list);
-		next = tmp->next;
 		kfree(dd);
 	}
 }
@@ -476,7 +475,7 @@ static int __table_get_device(struct dm_table *t, struct dm_target *ti,
 			      int mode, struct dm_dev **result)
 {
 	int r;
-	dev_t dev;
+	dev_t uninitialized_var(dev);
 	struct dm_dev *dd;
 	unsigned int major, minor;
 
@@ -805,7 +804,7 @@ static int setup_indexes(struct dm_table *t)
 		return -ENOMEM;
 
 	/* set up internal nodes, bottom-up */
-	for (i = t->depth - 2, total = 0; i >= 0; i--) {
+	for (i = t->depth - 2; i >= 0; i--) {
 		t->index[i] = indexes;
 		indexes += (KEYS_PER_NODE * t->counts[i]);
 		setup_btree_index(i, t);
@@ -993,12 +992,11 @@ int dm_table_resume_targets(struct dm_table *t)
 
 int dm_table_any_congested(struct dm_table *t, int bdi_bits)
 {
-	struct list_head *d, *devices;
+	struct dm_dev *dd;
+	struct list_head *devices = dm_table_get_devices(t);
 	int r = 0;
 
-	devices = dm_table_get_devices(t);
-	for (d = devices->next; d != devices; d = d->next) {
-		struct dm_dev *dd = list_entry(d, struct dm_dev, list);
+	list_for_each_entry(dd, devices, list) {
 		struct request_queue *q = bdev_get_queue(dd->bdev);
 		r |= bdi_congested(&q->backing_dev_info, bdi_bits);
 	}
@@ -1008,10 +1006,10 @@ int dm_table_any_congested(struct dm_table *t, int bdi_bits)
 
 void dm_table_unplug_all(struct dm_table *t)
 {
-	struct list_head *d, *devices = dm_table_get_devices(t);
+	struct dm_dev *dd;
+	struct list_head *devices = dm_table_get_devices(t);
 
-	for (d = devices->next; d != devices; d = d->next) {
-		struct dm_dev *dd = list_entry(d, struct dm_dev, list);
+	list_for_each_entry(dd, devices, list) {
 		struct request_queue *q = bdev_get_queue(dd->bdev);
 
 		blk_unplug(q);

drivers/md/dm.c

@@ -71,9 +71,22 @@ union map_info *dm_get_mapinfo(struct bio *bio)
 #define DMF_DELETING 4
 #define DMF_NOFLUSH_SUSPENDING 5
 
+/*
+ * Work processed by per-device workqueue.
+ */
+struct dm_wq_req {
+	enum {
+		DM_WQ_FLUSH_ALL,
+		DM_WQ_FLUSH_DEFERRED,
+	} type;
+	struct work_struct work;
+	struct mapped_device *md;
+	void *context;
+};
+
 struct mapped_device {
 	struct rw_semaphore io_lock;
-	struct semaphore suspend_lock;
+	struct mutex suspend_lock;
 	spinlock_t pushback_lock;
 	rwlock_t map_lock;
 	atomic_t holders;
@@ -95,6 +108,11 @@ struct mapped_device {
 	struct bio_list deferred;
 	struct bio_list pushback;
 
+	/*
+	 * Processing queue (flush/barriers)
+	 */
+	struct workqueue_struct *wq;
+
 	/*
 	 * The current mapping.
 	 */
@@ -181,7 +199,7 @@ static void local_exit(void)
 	DMINFO("cleaned up");
 }
 
-int (*_inits[])(void) __initdata = {
+static int (*_inits[])(void) __initdata = {
 	local_init,
 	dm_target_init,
 	dm_linear_init,
@@ -189,7 +207,7 @@ int (*_inits[])(void) __initdata = {
 	dm_interface_init,
 };
 
-void (*_exits[])(void) = {
+static void (*_exits[])(void) = {
 	local_exit,
 	dm_target_exit,
 	dm_linear_exit,
@@ -982,7 +1000,7 @@ static struct mapped_device *alloc_dev(int minor)
 	}
 
 	if (!try_module_get(THIS_MODULE))
-		goto bad0;
+		goto bad_module_get;
 
 	/* get a minor number for the dev */
 	if (minor == DM_ANY_MINOR)
@@ -990,11 +1008,11 @@ static struct mapped_device *alloc_dev(int minor)
 	else
 		r = specific_minor(md, minor);
 	if (r < 0)
-		goto bad1;
+		goto bad_minor;
 
 	memset(md, 0, sizeof(*md));
 	init_rwsem(&md->io_lock);
-	init_MUTEX(&md->suspend_lock);
+	mutex_init(&md->suspend_lock);
 	spin_lock_init(&md->pushback_lock);
 	rwlock_init(&md->map_lock);
 	atomic_set(&md->holders, 1);
@@ -1006,7 +1024,7 @@ static struct mapped_device *alloc_dev(int minor)
 
 	md->queue = blk_alloc_queue(GFP_KERNEL);
 	if (!md->queue)
-		goto bad1_free_minor;
+		goto bad_queue;
 
 	md->queue->queuedata = md;
 	md->queue->backing_dev_info.congested_fn = dm_any_congested;
@@ -1017,11 +1035,11 @@ static struct mapped_device *alloc_dev(int minor)
 
 	md->io_pool = mempool_create_slab_pool(MIN_IOS, _io_cache);
 	if (!md->io_pool)
-		goto bad2;
+		goto bad_io_pool;
 
 	md->tio_pool = mempool_create_slab_pool(MIN_IOS, _tio_cache);
 	if (!md->tio_pool)
-		goto bad3;
+		goto bad_tio_pool;
 
 	md->bs = bioset_create(16, 16);
 	if (!md->bs)
@@ -1029,7 +1047,7 @@ static struct mapped_device *alloc_dev(int minor)
 
 	md->disk = alloc_disk(1);
 	if (!md->disk)
-		goto bad4;
+		goto bad_disk;
 
 	atomic_set(&md->pending, 0);
 	init_waitqueue_head(&md->wait);
@@ -1044,6 +1062,10 @@ static struct mapped_device *alloc_dev(int minor)
 	add_disk(md->disk);
 	format_dev_t(md->name, MKDEV(_major, minor));
 
+	md->wq = create_singlethread_workqueue("kdmflush");
+	if (!md->wq)
+		goto bad_thread;
+
 	/* Populate the mapping, nobody knows we exist yet */
 	spin_lock(&_minor_lock);
 	old_md = idr_replace(&_minor_idr, md, minor);
@@ -1053,19 +1075,21 @@ static struct mapped_device *alloc_dev(int minor)
 
 	return md;
 
- bad4:
+bad_thread:
+	put_disk(md->disk);
+bad_disk:
 	bioset_free(md->bs);
- bad_no_bioset:
+bad_no_bioset:
 	mempool_destroy(md->tio_pool);
- bad3:
+bad_tio_pool:
 	mempool_destroy(md->io_pool);
- bad2:
+bad_io_pool:
 	blk_cleanup_queue(md->queue);
- bad1_free_minor:
+bad_queue:
 	free_minor(minor);
- bad1:
+bad_minor:
 	module_put(THIS_MODULE);
- bad0:
+bad_module_get:
 	kfree(md);
 	return NULL;
 }
@@ -1080,6 +1104,7 @@ static void free_dev(struct mapped_device *md)
 		unlock_fs(md);
 		bdput(md->suspended_bdev);
 	}
+	destroy_workqueue(md->wq);
 	mempool_destroy(md->tio_pool);
 	mempool_destroy(md->io_pool);
 	bioset_free(md->bs);
@@ -1259,20 +1284,91 @@ void dm_put(struct mapped_device *md)
 }
 EXPORT_SYMBOL_GPL(dm_put);
 
+static int dm_wait_for_completion(struct mapped_device *md)
+{
+	int r = 0;
+
+	while (1) {
+		set_current_state(TASK_INTERRUPTIBLE);
+
+		smp_mb();
+		if (!atomic_read(&md->pending))
+			break;
+
+		if (signal_pending(current)) {
+			r = -EINTR;
+			break;
+		}
+
+		io_schedule();
+	}
+	set_current_state(TASK_RUNNING);
+
+	return r;
+}
+
 /*
  * Process the deferred bios
  */
-static void __flush_deferred_io(struct mapped_device *md, struct bio *c)
+static void __flush_deferred_io(struct mapped_device *md)
 {
-	struct bio *n;
+	struct bio *c;
 
-	while (c) {
-		n = c->bi_next;
-		c->bi_next = NULL;
+	while ((c = bio_list_pop(&md->deferred))) {
 		if (__split_bio(md, c))
 			bio_io_error(c);
-		c = n;
 	}
+
+	clear_bit(DMF_BLOCK_IO, &md->flags);
+}
+
+static void __merge_pushback_list(struct mapped_device *md)
+{
+	unsigned long flags;
+
+	spin_lock_irqsave(&md->pushback_lock, flags);
+	clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
+	bio_list_merge_head(&md->deferred, &md->pushback);
+	bio_list_init(&md->pushback);
+	spin_unlock_irqrestore(&md->pushback_lock, flags);
+}
+
+static void dm_wq_work(struct work_struct *work)
+{
+	struct dm_wq_req *req = container_of(work, struct dm_wq_req, work);
+	struct mapped_device *md = req->md;
+
+	down_write(&md->io_lock);
+	switch (req->type) {
+	case DM_WQ_FLUSH_ALL:
+		__merge_pushback_list(md);
+		/* pass through */
+	case DM_WQ_FLUSH_DEFERRED:
+		__flush_deferred_io(md);
+		break;
+	default:
+		DMERR("dm_wq_work: unrecognised work type %d", req->type);
+		BUG();
+	}
+	up_write(&md->io_lock);
+}
+
+static void dm_wq_queue(struct mapped_device *md, int type, void *context,
+			struct dm_wq_req *req)
+{
+	req->type = type;
+	req->md = md;
+	req->context = context;
+	INIT_WORK(&req->work, dm_wq_work);
+	queue_work(md->wq, &req->work);
+}
+
+static void dm_queue_flush(struct mapped_device *md, int type, void *context)
+{
+	struct dm_wq_req req;
+
+	dm_wq_queue(md, type, context, &req);
+	flush_workqueue(md->wq);
 }
 
 /*
@@ -1282,7 +1378,7 @@ int dm_swap_table(struct mapped_device *md, struct dm_table *table)
 {
 	int r = -EINVAL;
 
-	down(&md->suspend_lock);
+	mutex_lock(&md->suspend_lock);
 
 	/* device must be suspended */
 	if (!dm_suspended(md))
@@ -1297,7 +1393,7 @@ int dm_swap_table(struct mapped_device *md, struct dm_table *table)
 	r = __bind(md, table);
 
 out:
-	up(&md->suspend_lock);
+	mutex_unlock(&md->suspend_lock);
 	return r;
 }
 
@@ -1346,17 +1442,17 @@ static void unlock_fs(struct mapped_device *md)
 int dm_suspend(struct mapped_device *md, unsigned suspend_flags)
 {
 	struct dm_table *map = NULL;
-	unsigned long flags;
 	DECLARE_WAITQUEUE(wait, current);
-	struct bio *def;
-	int r = -EINVAL;
+	int r = 0;
 	int do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG ? 1 : 0;
 	int noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG ? 1 : 0;
 
-	down(&md->suspend_lock);
+	mutex_lock(&md->suspend_lock);
 
-	if (dm_suspended(md))
+	if (dm_suspended(md)) {
+		r = -EINVAL;
 		goto out_unlock;
+	}
 
 	map = dm_get_table(md);
 
@@ -1378,17 +1474,17 @@ int dm_suspend(struct mapped_device *md, unsigned suspend_flags)
 			r = -ENOMEM;
 			goto flush_and_out;
 		}
-	}
 
 		/*
-	 * Flush I/O to the device.
-	 * noflush supersedes do_lockfs, because lock_fs() needs to flush I/Os.
+		 * Flush I/O to the device. noflush supersedes do_lockfs,
+		 * because lock_fs() needs to flush I/Os.
 		 */
-	if (do_lockfs && !noflush) {
+		if (do_lockfs) {
 			r = lock_fs(md);
 			if (r)
 				goto out;
 		}
+	}
 
 	/*
 	 * First we set the BLOCK_IO flag so no more ios will be mapped.
@@ -1404,66 +1500,36 @@ int dm_suspend(struct mapped_device *md, unsigned suspend_flags)
 		dm_table_unplug_all(map);
 
 	/*
-	 * Then we wait for the already mapped ios to
-	 * complete.
+	 * Wait for the already-mapped ios to complete.
 	 */
-	while (1) {
-		set_current_state(TASK_INTERRUPTIBLE);
-
-		if (!atomic_read(&md->pending) || signal_pending(current))
-			break;
-
-		io_schedule();
-	}
-	set_current_state(TASK_RUNNING);
+	r = dm_wait_for_completion(md);
 
 	down_write(&md->io_lock);
 	remove_wait_queue(&md->wait, &wait);
 
-	if (noflush) {
-		spin_lock_irqsave(&md->pushback_lock, flags);
-		clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
-		bio_list_merge_head(&md->deferred, &md->pushback);
-		bio_list_init(&md->pushback);
-		spin_unlock_irqrestore(&md->pushback_lock, flags);
-	}
+	if (noflush)
+		__merge_pushback_list(md);
+	up_write(&md->io_lock);
 
 	/* were we interrupted ? */
-	r = -EINTR;
-	if (atomic_read(&md->pending)) {
-		clear_bit(DMF_BLOCK_IO, &md->flags);
-		def = bio_list_get(&md->deferred);
-		__flush_deferred_io(md, def);
-		up_write(&md->io_lock);
+	if (r < 0) {
+		dm_queue_flush(md, DM_WQ_FLUSH_DEFERRED, NULL);
+
 		unlock_fs(md);
 		goto out; /* pushback list is already flushed, so skip flush */
 	}
-	up_write(&md->io_lock);
 
 	dm_table_postsuspend_targets(map);
 
 	set_bit(DMF_SUSPENDED, &md->flags);
 
-	r = 0;
-
 flush_and_out:
-	if (r && noflush) {
+	if (r && noflush)
 		/*
 		 * Because there may be already I/Os in the pushback list,
 		 * flush them before return.
 		 */
-		down_write(&md->io_lock);
-
-		spin_lock_irqsave(&md->pushback_lock, flags);
-		clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
-		bio_list_merge_head(&md->deferred, &md->pushback);
-		bio_list_init(&md->pushback);
-		spin_unlock_irqrestore(&md->pushback_lock, flags);
-
-		def = bio_list_get(&md->deferred);
-		__flush_deferred_io(md, def);
-		up_write(&md->io_lock);
-	}
+		dm_queue_flush(md, DM_WQ_FLUSH_ALL, NULL);
 
 out:
 	if (r && md->suspended_bdev) {
@@ -1474,17 +1540,16 @@ int dm_suspend(struct mapped_device *md, unsigned suspend_flags)
 	dm_table_put(map);
 
 out_unlock:
-	up(&md->suspend_lock);
+	mutex_unlock(&md->suspend_lock);
 	return r;
 }
 
 int dm_resume(struct mapped_device *md)
 {
 	int r = -EINVAL;
-	struct bio *def;
 	struct dm_table *map = NULL;
 
-	down(&md->suspend_lock);
+	mutex_lock(&md->suspend_lock);
 	if (!dm_suspended(md))
 		goto out;
 
@@ -1496,12 +1561,7 @@ int dm_resume(struct mapped_device *md)
 	if (r)
 		goto out;
 
-	down_write(&md->io_lock);
-	clear_bit(DMF_BLOCK_IO, &md->flags);
-
-	def = bio_list_get(&md->deferred);
-	__flush_deferred_io(md, def);
-	up_write(&md->io_lock);
+	dm_queue_flush(md, DM_WQ_FLUSH_DEFERRED, NULL);
 
 	unlock_fs(md);
 
@@ -1520,7 +1580,7 @@ int dm_resume(struct mapped_device *md)
 
 out:
 	dm_table_put(map);
-	up(&md->suspend_lock);
+	mutex_unlock(&md->suspend_lock);
 
 	return r;
 }

fs/compat_ioctl.c

@@ -78,7 +78,6 @@
 #include <linux/mii.h>
 #include <linux/if_bonding.h>
 #include <linux/watchdog.h>
-#include <linux/dm-ioctl.h>
 
 #include <linux/soundcard.h>
 #include <linux/lp.h>
@@ -1993,39 +1992,6 @@ COMPATIBLE_IOCTL(STOP_ARRAY_RO)
 COMPATIBLE_IOCTL(RESTART_ARRAY_RW)
 COMPATIBLE_IOCTL(GET_BITMAP_FILE)
 ULONG_IOCTL(SET_BITMAP_FILE)
-/* DM */
-COMPATIBLE_IOCTL(DM_VERSION_32)
-COMPATIBLE_IOCTL(DM_REMOVE_ALL_32)
-COMPATIBLE_IOCTL(DM_LIST_DEVICES_32)
-COMPATIBLE_IOCTL(DM_DEV_CREATE_32)
-COMPATIBLE_IOCTL(DM_DEV_REMOVE_32)
-COMPATIBLE_IOCTL(DM_DEV_RENAME_32)
-COMPATIBLE_IOCTL(DM_DEV_SUSPEND_32)
-COMPATIBLE_IOCTL(DM_DEV_STATUS_32)
-COMPATIBLE_IOCTL(DM_DEV_WAIT_32)
-COMPATIBLE_IOCTL(DM_TABLE_LOAD_32)
-COMPATIBLE_IOCTL(DM_TABLE_CLEAR_32)
-COMPATIBLE_IOCTL(DM_TABLE_DEPS_32)
-COMPATIBLE_IOCTL(DM_TABLE_STATUS_32)
-COMPATIBLE_IOCTL(DM_LIST_VERSIONS_32)
-COMPATIBLE_IOCTL(DM_TARGET_MSG_32)
-COMPATIBLE_IOCTL(DM_DEV_SET_GEOMETRY_32)
-COMPATIBLE_IOCTL(DM_VERSION)
-COMPATIBLE_IOCTL(DM_REMOVE_ALL)
-COMPATIBLE_IOCTL(DM_LIST_DEVICES)
-COMPATIBLE_IOCTL(DM_DEV_CREATE)
-COMPATIBLE_IOCTL(DM_DEV_REMOVE)
-COMPATIBLE_IOCTL(DM_DEV_RENAME)
-COMPATIBLE_IOCTL(DM_DEV_SUSPEND)
-COMPATIBLE_IOCTL(DM_DEV_STATUS)
-COMPATIBLE_IOCTL(DM_DEV_WAIT)
-COMPATIBLE_IOCTL(DM_TABLE_LOAD)
-COMPATIBLE_IOCTL(DM_TABLE_CLEAR)
-COMPATIBLE_IOCTL(DM_TABLE_DEPS)
-COMPATIBLE_IOCTL(DM_TABLE_STATUS)
-COMPATIBLE_IOCTL(DM_LIST_VERSIONS)
-COMPATIBLE_IOCTL(DM_TARGET_MSG)
-COMPATIBLE_IOCTL(DM_DEV_SET_GEOMETRY)
 /* Big K */
 COMPATIBLE_IOCTL(PIO_FONT)
 COMPATIBLE_IOCTL(GIO_FONT)

include/linux/device-mapper.h

@@ -110,14 +110,14 @@ struct target_type {
 };
 
 struct io_restrictions {
-	unsigned int		max_sectors;
-	unsigned short		max_phys_segments;
-	unsigned short		max_hw_segments;
-	unsigned short		hardsect_size;
-	unsigned int		max_segment_size;
-	unsigned int		max_hw_sectors;
-	unsigned long		seg_boundary_mask;
 	unsigned long bounce_pfn;
+	unsigned long seg_boundary_mask;
+	unsigned max_hw_sectors;
+	unsigned max_sectors;
+	unsigned max_segment_size;
+	unsigned short hardsect_size;
+	unsigned short max_hw_segments;
+	unsigned short max_phys_segments;
 	unsigned char no_cluster; /* inverted so that 0 is default */
 };
 

include/linux/dm-ioctl.h

@@ -232,36 +232,6 @@ enum {
 	DM_DEV_SET_GEOMETRY_CMD
 };
 
-/*
- * The dm_ioctl struct passed into the ioctl is just the header
- * on a larger chunk of memory.  On x86-64 and other
- * architectures the dm-ioctl struct will be padded to an 8 byte
- * boundary so the size will be different, which would change the
- * ioctl code - yes I really messed up.  This hack forces these
- * architectures to have the correct ioctl code.
- */
-#ifdef CONFIG_COMPAT
-typedef char ioctl_struct[308];
-#define DM_VERSION_32       _IOWR(DM_IOCTL, DM_VERSION_CMD, ioctl_struct)
-#define DM_REMOVE_ALL_32    _IOWR(DM_IOCTL, DM_REMOVE_ALL_CMD, ioctl_struct)
-#define DM_LIST_DEVICES_32  _IOWR(DM_IOCTL, DM_LIST_DEVICES_CMD, ioctl_struct)
-
-#define DM_DEV_CREATE_32    _IOWR(DM_IOCTL, DM_DEV_CREATE_CMD, ioctl_struct)
-#define DM_DEV_REMOVE_32    _IOWR(DM_IOCTL, DM_DEV_REMOVE_CMD, ioctl_struct)
-#define DM_DEV_RENAME_32    _IOWR(DM_IOCTL, DM_DEV_RENAME_CMD, ioctl_struct)
-#define DM_DEV_SUSPEND_32   _IOWR(DM_IOCTL, DM_DEV_SUSPEND_CMD, ioctl_struct)
-#define DM_DEV_STATUS_32    _IOWR(DM_IOCTL, DM_DEV_STATUS_CMD, ioctl_struct)
-#define DM_DEV_WAIT_32      _IOWR(DM_IOCTL, DM_DEV_WAIT_CMD, ioctl_struct)
-
-#define DM_TABLE_LOAD_32    _IOWR(DM_IOCTL, DM_TABLE_LOAD_CMD, ioctl_struct)
-#define DM_TABLE_CLEAR_32   _IOWR(DM_IOCTL, DM_TABLE_CLEAR_CMD, ioctl_struct)
-#define DM_TABLE_DEPS_32    _IOWR(DM_IOCTL, DM_TABLE_DEPS_CMD, ioctl_struct)
-#define DM_TABLE_STATUS_32  _IOWR(DM_IOCTL, DM_TABLE_STATUS_CMD, ioctl_struct)
-#define DM_LIST_VERSIONS_32 _IOWR(DM_IOCTL, DM_LIST_VERSIONS_CMD, ioctl_struct)
-#define DM_TARGET_MSG_32    _IOWR(DM_IOCTL, DM_TARGET_MSG_CMD, ioctl_struct)
-#define DM_DEV_SET_GEOMETRY_32	_IOWR(DM_IOCTL, DM_DEV_SET_GEOMETRY_CMD, ioctl_struct)
-#endif
-
 #define DM_IOCTL 0xfd
 
 #define DM_VERSION       _IOWR(DM_IOCTL, DM_VERSION_CMD, struct dm_ioctl)
@@ -286,9 +256,9 @@ typedef char ioctl_struct[308];
 #define DM_DEV_SET_GEOMETRY	_IOWR(DM_IOCTL, DM_DEV_SET_GEOMETRY_CMD, struct dm_ioctl)
 
 #define DM_VERSION_MAJOR	4
-#define DM_VERSION_MINOR	12
+#define DM_VERSION_MINOR	13
 #define DM_VERSION_PATCHLEVEL	0
-#define DM_VERSION_EXTRA	"-ioctl (2007-10-02)"
+#define DM_VERSION_EXTRA	"-ioctl (2007-10-18)"
 
 /* Status bits */
 #define DM_READONLY_FLAG	(1 << 0) /* In/Out */