[PATCH] direct-to-BIO for O_DIRECT

Here's a patch which converts O_DIRECT to go direct-to-BIO, bypassing
the kiovec layer.  It's followed by a patch which converts the raw
driver to use the O_DIRECT engine.

CPU utilisation is about the same as the kiovec-based implementation.
Read and write bandwidth are the same too, for 128k chunks.   But with
one megabyte chunks, this implementation is 20% faster at writing.

I assume this is because the kiobuf-based implementation has to stop
and wait for each 128k chunk, whereas this code streams the entire
request, regardless of its size.

This is with a single (oldish) scsi disk on aic7xxx.  I'd expect the
margin to widen on higher-end hardware which likes to have more
requests in flight.

Question is: what do we want to do with this sucker?  These are the
remaining users of kiovecs:

	drivers/md/lvm-snap.c
	drivers/media/video/video-buf.c
	drivers/mtd/devices/blkmtd.c
	drivers/scsi/sg.c

the video and mtd drivers seems to be fairly easy to de-kiobufize.
I'm aware of one proprietary driver which uses kiobufs.  XFS uses
kiobufs a little bit - just to map the pages.

So with a bit of effort and maintainer-irritation, we can extract
the kiobuf layer from the kernel.

drivers/char/raw.c

@@ -8,8 +8,8 @@
  * device are used to bind the other minor numbers to block devices.
  */
 
+#include <linux/init.h>
 #include <linux/fs.h>
-#include <linux/iobuf.h>
 #include <linux/major.h>
 #include <linux/blkdev.h>
 #include <linux/raw.h>
@@ -86,12 +86,6 @@ int raw_open(struct inode *inode, struct file *filp)
 		return 0;
 	}
 	
-	if (!filp->f_iobuf) {
-		err = alloc_kiovec(1, &filp->f_iobuf);
-		if (err)
-			return err;
-	}
-
 	down(&raw_devices[minor].mutex);
 	/*
 	 * No, it is a normal raw device.  All we need to do on open is
@@ -256,124 +250,46 @@ int raw_ctl_ioctl(struct inode *inode,
 	return err;
 }
 
-
-
-ssize_t	raw_read(struct file *filp, char * buf, 
-		 size_t size, loff_t *offp)
+ssize_t raw_read(struct file *filp, char * buf, size_t size, loff_t *offp)
 {
 	return rw_raw_dev(READ, filp, buf, size, offp);
 }
 
-ssize_t	raw_write(struct file *filp, const char *buf, 
-		  size_t size, loff_t *offp)
+ssize_t	raw_write(struct file *filp, const char *buf, size_t size, loff_t *offp)
 {
 	return rw_raw_dev(WRITE, filp, (char *) buf, size, offp);
 }
 
-#define SECTOR_BITS 9
-#define SECTOR_SIZE (1U << SECTOR_BITS)
-#define SECTOR_MASK (SECTOR_SIZE - 1)
-
-ssize_t	rw_raw_dev(int rw, struct file *filp, char *buf, 
-		   size_t size, loff_t *offp)
+ssize_t
+rw_raw_dev(int rw, struct file *filp, char *buf, size_t size, loff_t *offp)
 {
-	struct kiobuf * iobuf;
-	int		new_iobuf;
-	int		err = 0;
-	unsigned long	blocks;
-	size_t		transferred;
-	int		iosize;
-	int		minor;
-	kdev_t		dev;
-	unsigned long	limit;
-	int		sector_size, sector_bits, sector_mask;
-	sector_t	blocknr;
 	struct block_device *bdev;
-	
-	/*
-	 * First, a few checks on device size limits 
-	 */
+	struct inode *inode;
+	int minor;
+	ssize_t ret = 0;
 
 	minor = minor(filp->f_dentry->d_inode->i_rdev);
-
-	new_iobuf = 0;
-	iobuf = filp->f_iobuf;
-	if (test_and_set_bit(0, &filp->f_iobuf_lock)) {
-		/*
-		 * A parallel read/write is using the preallocated iobuf
-		 * so just run slow and allocate a new one.
-		 */
-		err = alloc_kiovec(1, &iobuf);
-		if (err)
-			goto out;
-		new_iobuf = 1;
-	}
-
 	bdev = raw_devices[minor].binding;
-	dev = to_kdev_t(bdev->bd_dev);
-	sector_size = raw_devices[minor].sector_size;
-	sector_bits = raw_devices[minor].sector_bits;
-	sector_mask = sector_size - 1;
-
-	limit = bdev->bd_inode->i_size >> sector_bits;
-	if (!limit)
-		limit = INT_MAX;
-	dprintk ("rw_raw_dev: dev %d:%d (+%d)\n",
-		 major(dev), minor(dev), limit);
-	
-	err = -EINVAL;
-	if ((*offp & sector_mask) || (size & sector_mask))
-		goto out_free;
-	err = 0;
-	if (size)
-		err = -ENXIO;
-	if ((*offp >> sector_bits) >= limit)
-		goto out_free;
-
-	transferred = 0;
-	blocknr = *offp >> sector_bits;
-	while (size > 0) {
-		blocks = size >> sector_bits;
-		if (blocks > limit - blocknr)
-			blocks = limit - blocknr;
-		if (!blocks)
-			break;
-
-		iosize = blocks << sector_bits;
+	inode = bdev->bd_inode;
 
-		err = map_user_kiobuf(rw, iobuf, (unsigned long) buf, iosize);
-		if (err)
-			break;
-
-		err = brw_kiovec(rw, 1, &iobuf, raw_devices[minor].binding, &blocknr, sector_size);
-
-		if (rw == READ && err > 0)
-			mark_dirty_kiobuf(iobuf, err);
-		
-		if (err >= 0) {
-			transferred += err;
-			size -= err;
-			buf += err;
-		}
-
-		blocknr += blocks;
-
-		unmap_kiobuf(iobuf);
-
-		if (err != iosize)
-			break;
+	if (size == 0)
+		goto out;
+	if (size < 0) {
+		ret = -EINVAL;
+		goto out;
 	}
-	
-	if (transferred) {
-		*offp += transferred;
-		err = transferred;
+	if (*offp >= inode->i_size) {
+		ret = -ENXIO;
+		goto out;
 	}
-
- out_free:
-	if (!new_iobuf)
-		clear_bit(0, &filp->f_iobuf_lock);
-	else
-		free_kiovec(1, &iobuf);
- out:	
-	return err;
+	if (size + *offp > inode->i_size)
+		size = inode->i_size - *offp;
+
+	ret = generic_file_direct_IO(rw, inode, buf, *offp, size);
+	if (ret > 0)
+		*offp += ret;
+	if (inode->i_mapping->nrpages)
+		invalidate_inode_pages2(inode->i_mapping);
+out:
+	return ret;
 }

fs/Makefile

@@ -15,7 +15,7 @@ obj-y :=	open.o read_write.o devices.o file_table.o buffer.o \
 		namei.o fcntl.o ioctl.o readdir.o select.o fifo.o locks.o \
 		dcache.o inode.o attr.o bad_inode.o file.o iobuf.o dnotify.o \
 		filesystems.o namespace.o seq_file.o xattr.o libfs.o \
-		fs-writeback.o mpage.o
+		fs-writeback.o mpage.o direct-io.o
 
 ifneq ($(CONFIG_NFSD),n)
 ifneq ($(CONFIG_NFSD),)

fs/block_dev.c

@@ -106,9 +106,12 @@ static int blkdev_get_block(struct inode * inode, sector_t iblock, struct buffer
 	return 0;
 }
 
-static int blkdev_direct_IO(int rw, struct inode * inode, struct kiobuf * iobuf, unsigned long blocknr, int blocksize)
+static int
+blkdev_direct_IO(int rw, struct inode *inode, char *buf,
+			loff_t offset, size_t count)
 {
-	return generic_direct_IO(rw, inode, iobuf, blocknr, blocksize, blkdev_get_block);
+	return generic_direct_IO(rw, inode, buf, offset,
+				count, blkdev_get_block);
 }
 
 static int blkdev_writepage(struct page * page)

fs/buffer.c

@@ -2311,6 +2311,7 @@ sector_t generic_block_bmap(struct address_space *mapping, sector_t block,
 	return tmp.b_blocknr;
 }
 
+#if 0
 int generic_direct_IO(int rw, struct inode *inode,
 			struct kiobuf *iobuf, unsigned long blocknr,
 			int blocksize, get_block_t *get_block)
@@ -2355,6 +2356,7 @@ int generic_direct_IO(int rw, struct inode *inode,
  out:
 	return retval;
 }
+#endif
 
 /*
  * Start I/O on a physical range of kernel memory, defined by a vector

fs/ext2/inode.c

@@ -607,11 +607,10 @@ static int ext2_bmap(struct address_space *mapping, long block)
 }
 
 static int
-ext2_direct_IO(int rw, struct inode *inode, struct kiobuf *iobuf,
-			unsigned long blocknr, int blocksize)
+ext2_direct_IO(int rw, struct inode *inode, char *buf,
+			loff_t offset, size_t count)
 {
-	return generic_direct_IO(rw, inode, iobuf, blocknr,
-				blocksize, ext2_get_block);
+	return generic_direct_IO(rw, inode, buf, offset, count, ext2_get_block);
 }
 
 static int

include/linux/buffer_head.h

@@ -185,8 +185,6 @@ int block_sync_page(struct page *);
 sector_t generic_block_bmap(struct address_space *, sector_t, get_block_t *);
 int generic_commit_write(struct file *, struct page *, unsigned, unsigned);
 int block_truncate_page(struct address_space *, loff_t, get_block_t *);
-int generic_direct_IO(int, struct inode *, struct kiobuf *,
-			unsigned long, int, get_block_t *);
 int file_fsync(struct file *, struct dentry *, int);
 
 #define OSYNC_METADATA	(1<<0)

include/linux/fs.h

@@ -303,8 +303,8 @@ struct address_space_operations {
 	int (*bmap)(struct address_space *, long);
 	int (*invalidatepage) (struct page *, unsigned long);
 	int (*releasepage) (struct page *, int);
-#define KERNEL_HAS_O_DIRECT /* this is for modules out of the kernel */
-	int (*direct_IO)(int, struct inode *, struct kiobuf *, unsigned long, int);
+	int (*direct_IO)(int, struct inode *, char *buf,
+				loff_t offset, size_t count);
 };
 
 struct backing_dev_info;
@@ -1128,7 +1128,7 @@ extern int check_disk_change(kdev_t);
 extern int invalidate_inodes(struct super_block *);
 extern int invalidate_device(kdev_t, int);
 extern void invalidate_inode_pages(struct inode *);
-extern void invalidate_inode_pages2(struct address_space *);
+extern void invalidate_inode_pages2(struct address_space *mapping);
 extern void write_inode_now(struct inode *, int);
 extern int filemap_fdatawrite(struct address_space *);
 extern int filemap_fdatawait(struct address_space *);
@@ -1233,6 +1233,11 @@ extern int file_read_actor(read_descriptor_t * desc, struct page *page, unsigned
 extern ssize_t generic_file_read(struct file *, char *, size_t, loff_t *);
 extern ssize_t generic_file_write(struct file *, const char *, size_t, loff_t *);
 extern void do_generic_file_read(struct file *, loff_t *, read_descriptor_t *, read_actor_t);
+ssize_t generic_file_direct_IO(int rw, struct inode *inode, char *buf,
+				loff_t offset, size_t count);
+int generic_direct_IO(int rw, struct inode *inode, char *buf,
+			loff_t offset, size_t count, get_block_t *get_block);
+
 extern loff_t no_llseek(struct file *file, loff_t offset, int origin);
 extern loff_t generic_file_llseek(struct file *file, loff_t offset, int origin);
 extern loff_t remote_llseek(struct file *file, loff_t offset, int origin);

mm/filemap.c

@@ -414,7 +414,7 @@ static int invalidate_list_pages2(struct address_space * mapping,
  * free the pages because they're mapped.
  * @mapping: the address_space which pages we want to invalidate
  */
-void invalidate_inode_pages2(struct address_space * mapping)
+void invalidate_inode_pages2(struct address_space *mapping)
 {
 	int unlocked;
 
@@ -1102,6 +1102,7 @@ void do_generic_file_read(struct file * filp, loff_t *ppos, read_descriptor_t *
 	UPDATE_ATIME(inode);
 }
 
+#if 0
 static ssize_t generic_file_direct_IO(int rw, struct file * filp, char * buf, size_t count, loff_t offset)
 {
 	ssize_t retval;
@@ -1182,6 +1183,7 @@ static ssize_t generic_file_direct_IO(int rw, struct file * filp, char * buf, si
  out:	
 	return retval;
 }
+#endif
 
 int file_read_actor(read_descriptor_t * desc, struct page *page, unsigned long offset, unsigned long size)
 {
@@ -1209,15 +1211,36 @@ int file_read_actor(read_descriptor_t * desc, struct page *page, unsigned long o
  * This is the "read()" routine for all filesystems
  * that can use the page cache directly.
  */
-ssize_t generic_file_read(struct file * filp, char * buf, size_t count, loff_t *ppos)
+ssize_t
+generic_file_read(struct file *filp, char *buf, size_t count, loff_t *ppos)
 {
 	ssize_t retval;
 
 	if ((ssize_t) count < 0)
 		return -EINVAL;
 
-	if (filp->f_flags & O_DIRECT)
-		goto o_direct;
+	if (filp->f_flags & O_DIRECT) {
+		loff_t pos = *ppos, size;
+		struct address_space *mapping;
+		struct inode *inode;
+
+		mapping = filp->f_dentry->d_inode->i_mapping;
+		inode = mapping->host;
+		retval = 0;
+		if (!count)
+			goto out; /* skip atime */
+		size = inode->i_size;
+		if (pos < size) {
+			if (pos + count > size)
+				count = size - pos;
+			retval = generic_file_direct_IO(READ, inode,
+							buf, pos, count);
+			if (retval > 0)
+				*ppos = pos + retval;
+		}
+		UPDATE_ATIME(filp->f_dentry->d_inode);
+		goto out;
+	}
 
 	retval = -EFAULT;
 	if (access_ok(VERIFY_WRITE, buf, count)) {
@@ -1230,36 +1253,14 @@ ssize_t generic_file_read(struct file * filp, char * buf, size_t count, loff_t *
 			desc.count = count;
 			desc.buf = buf;
 			desc.error = 0;
-			do_generic_file_read(filp, ppos, &desc, file_read_actor);
-
+			do_generic_file_read(filp,ppos,&desc,file_read_actor);
 			retval = desc.written;
 			if (!retval)
 				retval = desc.error;
 		}
 	}
- out:
+out:
 	return retval;
-
- o_direct:
-	{
-		loff_t pos = *ppos, size;
-		struct address_space *mapping = filp->f_dentry->d_inode->i_mapping;
-		struct inode *inode = mapping->host;
-
-		retval = 0;
-		if (!count)
-			goto out; /* skip atime */
-		size = inode->i_size;
-		if (pos < size) {
-			if (pos + count > size)
-				count = size - pos;
-			retval = generic_file_direct_IO(READ, filp, buf, count, pos);
-			if (retval > 0)
-				*ppos = pos + retval;
-		}
-		UPDATE_ATIME(filp->f_dentry->d_inode);
-		goto out;
-	}
 }
 
 static int file_send_actor(read_descriptor_t * desc, struct page *page, unsigned long offset, unsigned long size)
@@ -2186,8 +2187,8 @@ generic_file_write(struct file *file, const char *buf,
 	}
 
 	if (unlikely(file->f_flags & O_DIRECT)) {
-		written = generic_file_direct_IO(WRITE, file,
-						(char *) buf, count, pos);
+		written = generic_file_direct_IO(WRITE, inode,
+						(char *)buf, pos, count);
 		if (written > 0) {
 			loff_t end = pos + written;
 			if (end > inode->i_size && !S_ISBLK(inode->i_mode)) {
@@ -2195,7 +2196,8 @@ generic_file_write(struct file *file, const char *buf,
 				mark_inode_dirty(inode);
 			}
 			*ppos = end;
-			invalidate_inode_pages2(mapping);
+			if (mapping->nrpages)
+				invalidate_inode_pages2(mapping);
 		}
 		/*
 		 * Sync the fs metadata but not the minor inode changes and