From: Milan Broz <mbroz@redhat.com>
 
Moving all work from map dm call to workqueue.
Read is processed in two steps - generating read request and
after completion in endio second req. for decryption.
Writes are processed in one step - encryption and generating
request.
 
Previous implementation of map function in dm-crypt waits for
earlies items to complete to solve insufficient memory situations.
/* out of memory -> run queues */
        if (remaining)
                blk_congestion_wait(bio_data_dir(clone), HZ/100);
 
This will not work after reduce stack changes in generic_make_request()
which will not process IOs immediately.
(md-dm-reduce-stack-usage-with-stacked-block-devices.patch)
 
Signed-off-by: Milan Broz <mbroz@redhat.com>

Index: linux-2.6.17/drivers/md/dm-crypt.c
===================================================================
--- linux-2.6.17.orig/drivers/md/dm-crypt.c	2006-08-09 21:43:14.000000000 +0100
+++ linux-2.6.17/drivers/md/dm-crypt.c	2006-08-09 21:48:01.000000000 +0100
@@ -34,6 +34,7 @@ struct crypt_io {
 	struct work_struct work;
 	atomic_t pending;
 	int error;
+	int post_process;
 };
 
 /*
@@ -437,13 +438,6 @@ static void dec_pending(struct crypt_io 
 	mempool_free(io, cc->io_pool);
 }
 
-/*
- * kcryptd:
- *
- * Needed because it would be very unwise to do decryption in an
- * interrupt context, so bios returning from read requests get
- * queued here.
- */
 static struct workqueue_struct *_kcryptd_workqueue;
 static void kcryptd_do_work(void *data);
 
@@ -466,24 +460,27 @@ static int crypt_endio(struct bio *bio, 
 	if (!read_io)
 		crypt_free_buffer_pages(cc, bio, done);
 
+	if (unlikely(error))
+		goto out;
+
+	/* keep going - not finished yet */
 	if (unlikely(bio->bi_size))
 		return 1;
 
-	bio_put(bio);
-
-	/*
-	 * successful reads are decrypted by the worker thread
-	 */
 	if (!read_io)
 		goto out;
 
-	if (unlikely(!bio_flagged(bio, BIO_UPTODATE)))
+	if (unlikely(!bio_flagged(bio, BIO_UPTODATE))) {
+		error = -EIO;
 		goto out;
+	}
 
+	io->post_process = 1;
 	kcryptd_queue_io(io);
 	return 0;
-
 out:
+	bio_put(bio);
+
 	dec_pending(io, error);
 	return error;
 }
@@ -498,60 +495,91 @@ static void clone_init(struct crypt_io *
 	clone->bi_rw      = io->bio->bi_rw;
 }
 
-static struct bio *clone_read(struct crypt_io *io,
-			      sector_t sector)
+static void process_read(struct crypt_io *io)
 {
 	struct crypt_config *cc = io->target->private;
 	struct bio *bio = io->bio;
 	struct bio *clone;
 
+	atomic_inc(&io->pending);
+
 	/*
 	 * The block layer might modify the bvec array, so always
 	 * copy the required bvecs because we need the original
 	 * one in order to decrypt the whole bio data *afterwards*.
 	 */
 	clone = bio_alloc(GFP_NOIO, bio_segments(bio));
-	if (unlikely(!clone))
-		return NULL;
+	if (unlikely(!clone)) {
+		dec_pending(io, -ENOMEM);
+		return;
+	}
 
 	clone_init(io, clone);
 	clone->bi_idx = 0;
 	clone->bi_vcnt = bio_segments(bio);
 	clone->bi_size = bio->bi_size;
+	clone->bi_sector = cc->start + bio->bi_sector - io->target->begin;
 	memcpy(clone->bi_io_vec, bio_iovec(bio),
-	       sizeof(struct bio_vec) * clone->bi_vcnt);
-	clone->bi_sector = cc->start + sector;
+		sizeof(struct bio_vec) * clone->bi_vcnt);
 
-	return clone;
+	bio_get(clone);
+	generic_make_request(clone);
 }
 
-static struct bio *clone_write(struct crypt_io *io,
-			       sector_t sector,
-			       unsigned *bvec_idx,
-			       struct convert_context *ctx)
+static void process_write(struct crypt_io *io)
 {
 	struct crypt_config *cc = io->target->private;
 	struct bio *bio = io->bio;
 	struct bio *clone;
+	struct convert_context ctx;
+	unsigned remaining = bio->bi_size;
+	sector_t sector = bio->bi_sector - io->target->begin;
+	unsigned bvec_idx = 0;
 
-	clone = crypt_alloc_buffer(cc, bio->bi_size,
-				   io->first_clone, bvec_idx);
-	if (!clone)
-		return NULL;
+	crypt_convert_init(cc, &ctx, NULL, bio, sector, 1);
 
-	ctx->bio_out = clone;
+	/*
+	 * The allocated buffers can be smaller than the whole bio,
+	 * so repeat the whole process until all the data can be handled.
+	 */
+	do {
+		atomic_inc(&io->pending);
 
-	if (unlikely(crypt_convert(cc, ctx) < 0)) {
-		crypt_free_buffer_pages(cc, clone,
-		                        clone->bi_size);
-		bio_put(clone);
-		return NULL;
-	}
+		clone = crypt_alloc_buffer(cc, bio->bi_size,
+					   io->first_clone, &bvec_idx);
+		if (unlikely(!clone)) {
+			dec_pending(io, -ENOMEM);
+			return;
+		}
 
-	clone_init(io, clone);
-	clone->bi_sector = cc->start + sector;
+		ctx.bio_out = clone;
+
+		if (unlikely(crypt_convert(cc, &ctx) < 0)) {
+			crypt_free_buffer_pages(cc, clone, clone->bi_size);
+			bio_put(clone);
+			dec_pending(io, -EIO);
+			return;
+		}
+
+		clone_init(io, clone);
+		clone->bi_sector = cc->start + sector;
+
+		if (!io->first_clone) {
+			/*
+			 * hold a reference to the first clone, because it
+			 * holds the bio_vec array and that can't be freed
+			 * before all other clones are released
+			 */
+			bio_get(clone);
+			io->first_clone = clone;
+		}
+
+		remaining -= clone->bi_size;
+		sector += bio_sectors(clone);
+		generic_make_request(clone);
 
-	return clone;
+		/* out of memory -> wait */
+	} while (remaining && blk_congestion_wait(bio_data_dir(clone), HZ/100));
 }
 
 static void process_read_endio(struct crypt_io *io)
@@ -569,7 +597,12 @@ static void kcryptd_do_work(void *data)
 {
 	struct crypt_io *io = data;
 
-	process_read_endio(io);
+	if (io->post_process)
+		process_read_endio(io);
+	else if (bio_data_dir(io->bio) == READ)
+		process_read(io);
+	else
+		process_write(io);
 }
 
 /*
@@ -839,68 +872,19 @@ static int crypt_map(struct dm_target *t
 {
 	struct crypt_config *cc = ti->private;
 	struct crypt_io *io;
-	struct convert_context ctx;
-	struct bio *clone;
-	unsigned int remaining = bio->bi_size;
-	sector_t sector = bio->bi_sector - ti->begin;
-	unsigned int bvec_idx = 0;
 
 	io = mempool_alloc(cc->io_pool, GFP_NOIO);
+	if (unlikely(!io))
+		return -ENOMEM;
+
 	io->target = ti;
 	io->bio = bio;
 	io->first_clone = NULL;
-	io->error = 0;
-	atomic_set(&io->pending, 1); /* hold a reference */
-
-	if (bio_data_dir(bio) == WRITE)
-		crypt_convert_init(cc, &ctx, NULL, bio, sector, 1);
-
-	/*
-	 * The allocated buffers can be smaller than the whole bio,
-	 * so repeat the whole process until all the data can be handled.
-	 */
-	while (remaining) {
-		if (bio_data_dir(bio) == WRITE)
-			clone = crypt_clone_write(io, sector, &bvec_idx, &ctx);
-		else
-			clone = crypt_clone_read(io, sector);
-		if (!clone)
-			goto cleanup;
-
-		if (!io->first_clone) {
-			/*
-			 * hold a reference to the first clone, because it
-			 * holds the bio_vec array and that can't be freed
-			 * before all other clones are released
-			 */
-			bio_get(clone);
-			io->first_clone = clone;
-		}
-		atomic_inc(&io->pending);
-
-		remaining -= clone->bi_size;
-		sector += bio_sectors(clone);
-
-		generic_make_request(clone);
-
-		/* out of memory -> run queues */
-		if (remaining)
-			blk_congestion_wait(bio_data_dir(clone), HZ/100);
-	}
+	io->error = io->post_process = 0;
+	atomic_set(&io->pending, 0);	/* io created in workqueue */
+	kcryptd_queue_io(io);
 
-	/* drop reference, clones could have returned before we reach this */
-	dec_pending(io, 0);
 	return 0;
-
-cleanup:
-	if (io->first_clone) {
-		dec_pending(io, -ENOMEM);
-		return 0;
-	}
-
-	/* if no bio has been dispatched yet, we can directly return the error */
-	mempool_free(io, cc->io_pool);
-	return -ENOMEM;
 }
 
 static int crypt_status(struct dm_target *ti, status_type_t type,
@@ -1009,7 +993,7 @@ error:
 
 static struct target_type crypt_target = {
 	.name   = "crypt",
-	.version= {1, 2, 0},
+	.version= {1, 3, 0},
 	.module = THIS_MODULE,
 	.ctr    = crypt_ctr,
 	.dtr    = crypt_dtr,