[PATCH] md: add write-behind support for md/raid1

If a device is flagged 'WriteMostly' and the array has a bitmap, and the
bitmap superblock indicates that write_behind is allowed, then write_behind is
enabled for WriteMostly devices.

Write requests will be acknowledges as complete to the caller (via b_end_io)
when all non-WriteMostly devices have completed the write, but will not be
cleared from the bitmap until all devices complete.

This requires memory allocation to make a local copy of the data being
written.  If there is insufficient memory, then we fall-back on normal write
semantics.

Signed-Off-By: Paul Clements <paul.clements@steeleye.com>
Signed-off-by: Neil Brown <neilb@cse.unsw.edu.au>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

3 files changed, 26 insertions, 5 deletions

diff --git a/include/linux/raid/bitmap.h b/include/linux/raid/bitmap.h
index 4bf1659f8aa..9de99198caf 100644
--- a/include/linux/raid/bitmap.h
+++ b/include/linux/raid/bitmap.h
@@ -7,7 +7,7 @@
 #define BITMAP_H 1
 
 #define BITMAP_MAJOR 3
-#define BITMAP_MINOR 38
+#define BITMAP_MINOR 39
 
 /*
  * in-memory bitmap:
@@ -147,8 +147,9 @@ typedef struct bitmap_super_s {
 	__u32 state;        /* 48  bitmap state information */
 	__u32 chunksize;    /* 52  the bitmap chunk size in bytes */
 	__u32 daemon_sleep; /* 56  seconds between disk flushes */
+	__u32 write_behind; /* 60  number of outstanding write-behind writes */
 
-	__u8  pad[256 - 60]; /* set to zero */
+	__u8  pad[256 - 64]; /* set to zero */
 } bitmap_super_t;
 
 /* notes:
@@ -226,6 +227,9 @@ struct bitmap {
 
 	unsigned long flags;
 
+	unsigned long max_write_behind; /* write-behind mode */
+	atomic_t behind_writes;
+
 	/*
 	 * the bitmap daemon - periodically wakes up and sweeps the bitmap
 	 * file, cleaning up bits and flushing out pages to disk as necessary
@@ -260,9 +264,10 @@ int  bitmap_setallbits(struct bitmap *bitmap);
 void bitmap_write_all(struct bitmap *bitmap);
 
 /* these are exported */
-int bitmap_startwrite(struct bitmap *bitmap, sector_t offset, unsigned long sectors);
-void bitmap_endwrite(struct bitmap *bitmap, sector_t offset, unsigned long sectors,
-		     int success);
+int bitmap_startwrite(struct bitmap *bitmap, sector_t offset,
+			unsigned long sectors, int behind);
+void bitmap_endwrite(struct bitmap *bitmap, sector_t offset,
+			unsigned long sectors, int success, int behind);
 int bitmap_start_sync(struct bitmap *bitmap, sector_t offset, int *blocks, int degraded);
 void bitmap_end_sync(struct bitmap *bitmap, sector_t offset, int *blocks, int aborted);
 void bitmap_close_sync(struct bitmap *bitmap);
diff --git a/include/linux/raid/md_k.h b/include/linux/raid/md_k.h
index 7ef78e15ce0..2514e5fcda7 100644
--- a/include/linux/raid/md_k.h
+++ b/include/linux/raid/md_k.h
@@ -275,6 +275,9 @@ struct mddev_s
 	atomic_t			writes_pending; 
 	request_queue_t			*queue;	/* for plugging ... */
 
+	atomic_t                        write_behind; /* outstanding async IO */
+	unsigned int                    max_write_behind; /* 0 = sync */
+
 	struct bitmap                   *bitmap; /* the bitmap for the device */
 	struct file			*bitmap_file; /* the bitmap file */
 	long				bitmap_offset; /* offset from superblock of
diff --git a/include/linux/raid/raid1.h b/include/linux/raid/raid1.h
index 9d93cf12e89..60e19b66754 100644
--- a/include/linux/raid/raid1.h
+++ b/include/linux/raid/raid1.h
@@ -80,6 +80,9 @@ struct r1bio_s {
 	atomic_t		remaining; /* 'have we finished' count,
 					    * used from IRQ handlers
 					    */
+	atomic_t		behind_remaining; /* number of write-behind ios remaining
+						 * in this BehindIO request
+						 */
 	sector_t		sector;
 	int			sectors;
 	unsigned long		state;
@@ -107,4 +110,14 @@ struct r1bio_s {
 #define	R1BIO_Uptodate	0
 #define	R1BIO_IsSync	1
 #define	R1BIO_Degraded	2
+#define	R1BIO_BehindIO   3
+/* For write-behind requests, we call bi_end_io when
+ * the last non-write-behind device completes, providing
+ * any write was successful.  Otherwise we call when
+ * any write-behind write succeeds, otherwise we call
+ * with failure when last write completes (and all failed).
+ * Record that bi_end_io was called with this flag...
+ */
+#define	R1BIO_Returned 4
+
 #endif