Merge git://oss.sgi.com:8090/oss/git/xfs-2.6

19 files changed, 564 insertions, 614 deletions

diff --git a/fs/xfs/linux-2.6/kmem.c b/fs/xfs/linux-2.6/kmem.c
index 364ea8c..4b18455 100644
--- a/fs/xfs/linux-2.6/kmem.c
+++ b/fs/xfs/linux-2.6/kmem.c
@@ -45,11 +45,11 @@
 
 
 void *
-kmem_alloc(size_t size, int flags)
+kmem_alloc(size_t size, unsigned int __nocast flags)
 {
-	int	retries = 0;
-	int	lflags = kmem_flags_convert(flags);
-	void	*ptr;
+	int		retries = 0;
+	unsigned int	lflags = kmem_flags_convert(flags);
+	void		*ptr;
 
 	do {
 		if (size < MAX_SLAB_SIZE || retries > MAX_VMALLOCS)
@@ -67,7 +67,7 @@ kmem_alloc(size_t size, int flags)
 }
 
 void *
-kmem_zalloc(size_t size, int flags)
+kmem_zalloc(size_t size, unsigned int __nocast flags)
 {
 	void	*ptr;
 
@@ -89,7 +89,8 @@ kmem_free(void *ptr, size_t size)
 }
 
 void *
-kmem_realloc(void *ptr, size_t newsize, size_t oldsize, int flags)
+kmem_realloc(void *ptr, size_t newsize, size_t oldsize,
+	     unsigned int __nocast flags)
 {
 	void	*new;
 
@@ -104,11 +105,11 @@ kmem_realloc(void *ptr, size_t newsize, size_t oldsize, int flags)
 }
 
 void *
-kmem_zone_alloc(kmem_zone_t *zone, int flags)
+kmem_zone_alloc(kmem_zone_t *zone, unsigned int __nocast flags)
 {
-	int	retries = 0;
-	int	lflags = kmem_flags_convert(flags);
-	void	*ptr;
+	int		retries = 0;
+	unsigned int	lflags = kmem_flags_convert(flags);
+	void		*ptr;
 
 	do {
 		ptr = kmem_cache_alloc(zone, lflags);
@@ -123,7 +124,7 @@ kmem_zone_alloc(kmem_zone_t *zone, int flags)
 }
 
 void *
-kmem_zone_zalloc(kmem_zone_t *zone, int flags)
+kmem_zone_zalloc(kmem_zone_t *zone, unsigned int __nocast flags)
 {
 	void	*ptr;
 
diff --git a/fs/xfs/linux-2.6/kmem.h b/fs/xfs/linux-2.6/kmem.h
index 1397b66..109fcf2 100644
--- a/fs/xfs/linux-2.6/kmem.h
+++ b/fs/xfs/linux-2.6/kmem.h
@@ -39,10 +39,10 @@
 /*
  * memory management routines
  */
-#define KM_SLEEP	0x0001
-#define KM_NOSLEEP	0x0002
-#define KM_NOFS		0x0004
-#define KM_MAYFAIL	0x0008
+#define KM_SLEEP	0x0001u
+#define KM_NOSLEEP	0x0002u
+#define KM_NOFS		0x0004u
+#define KM_MAYFAIL	0x0008u
 
 #define	kmem_zone	kmem_cache_s
 #define kmem_zone_t	kmem_cache_t
@@ -81,9 +81,9 @@ typedef unsigned long xfs_pflags_t;
 	*(NSTATEP) = *(OSTATEP);	\
 } while (0)
 
-static __inline unsigned int kmem_flags_convert(int flags)
+static __inline unsigned int kmem_flags_convert(unsigned int __nocast flags)
 {
-	int	lflags = __GFP_NOWARN;	/* we'll report problems, if need be */
+	unsigned int	lflags = __GFP_NOWARN;	/* we'll report problems, if need be */
 
 #ifdef DEBUG
 	if (unlikely(flags & ~(KM_SLEEP|KM_NOSLEEP|KM_NOFS|KM_MAYFAIL))) {
@@ -125,12 +125,13 @@ kmem_zone_destroy(kmem_zone_t *zone)
 		BUG();
 }
 
-extern void	    *kmem_zone_zalloc(kmem_zone_t *, int);
-extern void	    *kmem_zone_alloc(kmem_zone_t *, int);
+extern void	    *kmem_zone_zalloc(kmem_zone_t *, unsigned int __nocast);
+extern void	    *kmem_zone_alloc(kmem_zone_t *, unsigned int __nocast);
 
-extern void	    *kmem_alloc(size_t, int);
-extern void	    *kmem_realloc(void *, size_t, size_t, int);
-extern void	    *kmem_zalloc(size_t, int);
+extern void	    *kmem_alloc(size_t, unsigned int __nocast);
+extern void	    *kmem_realloc(void *, size_t, size_t,
+				  unsigned int __nocast);
+extern void	    *kmem_zalloc(size_t, unsigned int __nocast);
 extern void         kmem_free(void *, size_t);
 
 typedef struct shrinker *kmem_shaker_t;
diff --git a/fs/xfs/linux-2.6/spin.h b/fs/xfs/linux-2.6/spin.h
index bcf60a0..0039504 100644
--- a/fs/xfs/linux-2.6/spin.h
+++ b/fs/xfs/linux-2.6/spin.h
@@ -45,6 +45,9 @@
 typedef spinlock_t lock_t;
 
 #define SPLDECL(s)			unsigned long s
+#ifndef DEFINE_SPINLOCK
+#define DEFINE_SPINLOCK(s)		spinlock_t s = SPIN_LOCK_UNLOCKED
+#endif
 
 #define spinlock_init(lock, name)	spin_lock_init(lock)
 #define	spinlock_destroy(lock)
diff --git a/fs/xfs/linux-2.6/xfs_aops.c b/fs/xfs/linux-2.6/xfs_aops.c
index a3a4b5a..c6c0779 100644
--- a/fs/xfs/linux-2.6/xfs_aops.c
+++ b/fs/xfs/linux-2.6/xfs_aops.c
@@ -104,66 +104,114 @@ xfs_page_trace(
 #define xfs_page_trace(tag, inode, page, mask)
 #endif
 
-void
-linvfs_unwritten_done(
-	struct buffer_head	*bh,
-	int			uptodate)
+/*
+ * Schedule IO completion handling on a xfsdatad if this was
+ * the final hold on this ioend.
+ */
+STATIC void
+xfs_finish_ioend(
+	xfs_ioend_t		*ioend)
 {
-	xfs_buf_t		*pb = (xfs_buf_t *)bh->b_private;
+	if (atomic_dec_and_test(&ioend->io_remaining))
+		queue_work(xfsdatad_workqueue, &ioend->io_work);
+}
 
-	ASSERT(buffer_unwritten(bh));
-	bh->b_end_io = NULL;
-	clear_buffer_unwritten(bh);
-	if (!uptodate)
-		pagebuf_ioerror(pb, EIO);
-	if (atomic_dec_and_test(&pb->pb_io_remaining) == 1) {
-		pagebuf_iodone(pb, 1, 1);
-	}
-	end_buffer_async_write(bh, uptodate);
+STATIC void
+xfs_destroy_ioend(
+	xfs_ioend_t		*ioend)
+{
+	vn_iowake(ioend->io_vnode);
+	mempool_free(ioend, xfs_ioend_pool);
 }
 
 /*
  * Issue transactions to convert a buffer range from unwritten
- * to written extents (buffered IO).
+ * to written extents.
  */
 STATIC void
-linvfs_unwritten_convert(
-	xfs_buf_t	*bp)
+xfs_end_bio_unwritten(
+	void			*data)
 {
-	vnode_t		*vp = XFS_BUF_FSPRIVATE(bp, vnode_t *);
-	int		error;
+	xfs_ioend_t		*ioend = data;
+	vnode_t			*vp = ioend->io_vnode;
+	xfs_off_t		offset = ioend->io_offset;
+	size_t			size = ioend->io_size;
+	struct buffer_head	*bh, *next;
+	int			error;
+
+	if (ioend->io_uptodate)
+		VOP_BMAP(vp, offset, size, BMAPI_UNWRITTEN, NULL, NULL, error);
+
+	/* ioend->io_buffer_head is only non-NULL for buffered I/O */
+	for (bh = ioend->io_buffer_head; bh; bh = next) {
+		next = bh->b_private;
+
+		bh->b_end_io = NULL;
+		clear_buffer_unwritten(bh);
+		end_buffer_async_write(bh, ioend->io_uptodate);
+	}
 
-	BUG_ON(atomic_read(&bp->pb_hold) < 1);
-	VOP_BMAP(vp, XFS_BUF_OFFSET(bp), XFS_BUF_SIZE(bp),
-			BMAPI_UNWRITTEN, NULL, NULL, error);
-	XFS_BUF_SET_FSPRIVATE(bp, NULL);
-	XFS_BUF_CLR_IODONE_FUNC(bp);
-	XFS_BUF_UNDATAIO(bp);
-	iput(LINVFS_GET_IP(vp));
-	pagebuf_iodone(bp, 0, 0);
+	xfs_destroy_ioend(ioend);
 }
 
 /*
- * Issue transactions to convert a buffer range from unwritten
- * to written extents (direct IO).
+ * Allocate and initialise an IO completion structure.
+ * We need to track unwritten extent write completion here initially.
+ * We'll need to extend this for updating the ondisk inode size later
+ * (vs. incore size).
  */
-STATIC void
-linvfs_unwritten_convert_direct(
-	struct kiocb	*iocb,
-	loff_t		offset,
-	ssize_t		size,
-	void		*private)
+STATIC xfs_ioend_t *
+xfs_alloc_ioend(
+	struct inode		*inode)
 {
-	struct inode	*inode = iocb->ki_filp->f_dentry->d_inode;
-	ASSERT(!private || inode == (struct inode *)private);
+	xfs_ioend_t		*ioend;
 
-	/* private indicates an unwritten extent lay beneath this IO */
-	if (private && size > 0) {
-		vnode_t	*vp = LINVFS_GET_VP(inode);
-		int	error;
+	ioend = mempool_alloc(xfs_ioend_pool, GFP_NOFS);
 
-		VOP_BMAP(vp, offset, size, BMAPI_UNWRITTEN, NULL, NULL, error);
-	}
+	/*
+	 * Set the count to 1 initially, which will prevent an I/O
+	 * completion callback from happening before we have started
+	 * all the I/O from calling the completion routine too early.
+	 */
+	atomic_set(&ioend->io_remaining, 1);
+	ioend->io_uptodate = 1; /* cleared if any I/O fails */
+	ioend->io_vnode = LINVFS_GET_VP(inode);
+	ioend->io_buffer_head = NULL;
+	atomic_inc(&ioend->io_vnode->v_iocount);
+	ioend->io_offset = 0;
+	ioend->io_size = 0;
+
+	INIT_WORK(&ioend->io_work, xfs_end_bio_unwritten, ioend);
+
+	return ioend;
+}
+
+void
+linvfs_unwritten_done(
+	struct buffer_head	*bh,
+	int			uptodate)
+{
+	xfs_ioend_t		*ioend = bh->b_private;
+	static spinlock_t	unwritten_done_lock = SPIN_LOCK_UNLOCKED;
+	unsigned long		flags;
+
+	ASSERT(buffer_unwritten(bh));
+	bh->b_end_io = NULL;
+
+	if (!uptodate)
+		ioend->io_uptodate = 0;
+
+	/*
+	 * Deep magic here.  We reuse b_private in the buffer_heads to build
+	 * a chain for completing the I/O from user context after we've issued
+	 * a transaction to convert the unwritten extent.
+	 */
+	spin_lock_irqsave(&unwritten_done_lock, flags);
+	bh->b_private = ioend->io_buffer_head;
+	ioend->io_buffer_head = bh;
+	spin_unlock_irqrestore(&unwritten_done_lock, flags);
+
+	xfs_finish_ioend(ioend);
 }
 
 STATIC int
@@ -255,7 +303,7 @@ xfs_probe_unwritten_page(
 	struct address_space	*mapping,
 	pgoff_t			index,
 	xfs_iomap_t		*iomapp,
-	xfs_buf_t		*pb,
+	xfs_ioend_t		*ioend,
 	unsigned long		max_offset,
 	unsigned long		*fsbs,
 	unsigned int            bbits)
@@ -283,7 +331,7 @@ xfs_probe_unwritten_page(
 				break;
 			xfs_map_at_offset(page, bh, p_offset, bbits, iomapp);
 			set_buffer_unwritten_io(bh);
-			bh->b_private = pb;
+			bh->b_private = ioend;
 			p_offset += bh->b_size;
 			(*fsbs)++;
 		} while ((bh = bh->b_this_page) != head);
@@ -434,34 +482,15 @@ xfs_map_unwritten(
 {
 	struct buffer_head	*bh = curr;
 	xfs_iomap_t		*tmp;
-	xfs_buf_t		*pb;
-	loff_t			offset, size;
+	xfs_ioend_t		*ioend;
+	loff_t			offset;
 	unsigned long		nblocks = 0;
 
 	offset = start_page->index;
 	offset <<= PAGE_CACHE_SHIFT;
 	offset += p_offset;
 
-	/* get an "empty" pagebuf to manage IO completion
-	 * Proper values will be set before returning */
-	pb = pagebuf_lookup(iomapp->iomap_target, 0, 0, 0);
-	if (!pb)
-		return -EAGAIN;
-
-	/* Take a reference to the inode to prevent it from
-	 * being reclaimed while we have outstanding unwritten
-	 * extent IO on it.
-	 */
-	if ((igrab(inode)) != inode) {
-		pagebuf_free(pb);
-		return -EAGAIN;
-	}
-
-	/* Set the count to 1 initially, this will stop an I/O
-	 * completion callout which happens before we have started
-	 * all the I/O from calling pagebuf_iodone too early.
-	 */
-	atomic_set(&pb->pb_io_remaining, 1);
+	ioend = xfs_alloc_ioend(inode);
 
 	/* First map forwards in the page consecutive buffers
 	 * covering this unwritten extent
@@ -474,12 +503,12 @@ xfs_map_unwritten(
 			break;
 		xfs_map_at_offset(start_page, bh, p_offset, block_bits, iomapp);
 		set_buffer_unwritten_io(bh);
-		bh->b_private = pb;
+		bh->b_private = ioend;
 		p_offset += bh->b_size;
 		nblocks++;
 	} while ((bh = bh->b_this_page) != head);
 
-	atomic_add(nblocks, &pb->pb_io_remaining);
+	atomic_add(nblocks, &ioend->io_remaining);
 
 	/* If we reached the end of the page, map forwards in any
 	 * following pages which are also covered by this extent.
@@ -496,13 +525,13 @@ xfs_map_unwritten(
 		tloff = min(tlast, tloff);
 		for (tindex = start_page->index + 1; tindex < tloff; tindex++) {
 			page = xfs_probe_unwritten_page(mapping,
-						tindex, iomapp, pb,
+						tindex, iomapp, ioend,
 						PAGE_CACHE_SIZE, &bs, bbits);
 			if (!page)
 				break;
 			nblocks += bs;
-			atomic_add(bs, &pb->pb_io_remaining);
-			xfs_convert_page(inode, page, iomapp, wbc, pb,
+			atomic_add(bs, &ioend->io_remaining);
+			xfs_convert_page(inode, page, iomapp, wbc, ioend,
 							startio, all_bh);
 			/* stop if converting the next page might add
 			 * enough blocks that the corresponding byte
@@ -514,12 +543,12 @@ xfs_map_unwritten(
 		if (tindex == tlast &&
 		    (pg_offset = (i_size_read(inode) & (PAGE_CACHE_SIZE - 1)))) {
 			page = xfs_probe_unwritten_page(mapping,
-							tindex, iomapp, pb,
+							tindex, iomapp, ioend,
 							pg_offset, &bs, bbits);
 			if (page) {
 				nblocks += bs;
-				atomic_add(bs, &pb->pb_io_remaining);
-				xfs_convert_page(inode, page, iomapp, wbc, pb,
+				atomic_add(bs, &ioend->io_remaining);
+				xfs_convert_page(inode, page, iomapp, wbc, ioend,
 							startio, all_bh);
 				if (nblocks >= ((ULONG_MAX - PAGE_SIZE) >> block_bits))
 					goto enough;
@@ -528,21 +557,9 @@ xfs_map_unwritten(
 	}
 
 enough:
-	size = nblocks;		/* NB: using 64bit number here */
-	size <<= block_bits;	/* convert fsb's to byte range */
-
-	XFS_BUF_DATAIO(pb);
-	XFS_BUF_ASYNC(pb);
-	XFS_BUF_SET_SIZE(pb, size);
-	XFS_BUF_SET_COUNT(pb, size);
-	XFS_BUF_SET_OFFSET(pb, offset);
-	XFS_BUF_SET_FSPRIVATE(pb, LINVFS_GET_VP(inode));
-	XFS_BUF_SET_IODONE_FUNC(pb, linvfs_unwritten_convert);
-
-	if (atomic_dec_and_test(&pb->pb_io_remaining) == 1) {
-		pagebuf_iodone(pb, 1, 1);
-	}
-
+	ioend->io_size = (xfs_off_t)nblocks << block_bits;
+	ioend->io_offset = offset;
+	xfs_finish_ioend(ioend);
 	return 0;
 }
 
@@ -787,7 +804,7 @@ xfs_page_state_convert(
 				continue;
 			if (!iomp) {
 				err = xfs_map_blocks(inode, offset, len, &iomap,
-						BMAPI_READ|BMAPI_IGNSTATE);
+						BMAPI_WRITE|BMAPI_IGNSTATE);
 				if (err) {
 					goto error;
 				}
@@ -1028,6 +1045,44 @@ linvfs_get_blocks_direct(
 					create, 1, BMAPI_WRITE|BMAPI_DIRECT);
 }
 
+STATIC void
+linvfs_end_io_direct(
+	struct kiocb	*iocb,
+	loff_t		offset,
+	ssize_t		size,
+	void		*private)
+{
+	xfs_ioend_t	*ioend = iocb->private;
+
+	/*
+	 * Non-NULL private data means we need to issue a transaction to
+	 * convert a range from unwritten to written extents.  This needs
+	 * to happen from process contect but aio+dio I/O completion
+	 * happens from irq context so we need to defer it to a workqueue.
+	 * This is not nessecary for synchronous direct I/O, but we do
+	 * it anyway to keep the code uniform and simpler.
+	 *
+	 * The core direct I/O code might be changed to always call the
+	 * completion handler in the future, in which case all this can
+	 * go away.
+	 */
+	if (private && size > 0) {
+		ioend->io_offset = offset;
+		ioend->io_size = size;
+		xfs_finish_ioend(ioend);
+	} else {
+		ASSERT(size >= 0);
+		xfs_destroy_ioend(ioend);
+	}
+
+	/*
+	 * blockdev_direct_IO can return an error even afer the I/O
+	 * completion handler was called.  Thus we need to protect
+	 * against double-freeing.
+	 */
+	iocb->private = NULL;
+}
+
 STATIC ssize_t
 linvfs_direct_IO(
 	int			rw,
@@ -1042,16 +1097,23 @@ linvfs_direct_IO(
 	xfs_iomap_t	iomap;
 	int		maps = 1;
 	int		error;
+	ssize_t		ret;
 
 	VOP_BMAP(vp, offset, 0, BMAPI_DEVICE, &iomap, &maps, error);
 	if (error)
 		return -error;
 
-	return blockdev_direct_IO_own_locking(rw, iocb, inode,
+	iocb->private = xfs_alloc_ioend(inode);
+
+	ret = blockdev_direct_IO_own_locking(rw, iocb, inode,
 		iomap.iomap_target->pbr_bdev,
 		iov, offset, nr_segs,
 		linvfs_get_blocks_direct,
-		linvfs_unwritten_convert_direct);
+		linvfs_end_io_direct);
+
+	if (unlikely(ret <= 0 && iocb->private))
+		xfs_destroy_ioend(iocb->private);
+	return ret;
 }
 
 
@@ -1202,6 +1264,16 @@ out_unlock:
 	return error;
 }
 
+STATIC int
+linvfs_invalidate_page(
+	struct page		*page,
+	unsigned long		offset)
+{
+	xfs_page_trace(XFS_INVALIDPAGE_ENTER,
+			page->mapping->host, page, offset);
+	return block_invalidatepage(page, offset);
+}
+
 /*
  * Called to move a page into cleanable state - and from there
  * to be released. Possibly the page is already clean. We always
@@ -1279,6 +1351,7 @@ struct address_space_operations linvfs_aops = {
 	.writepage		= linvfs_writepage,
 	.sync_page		= block_sync_page,
 	.releasepage		= linvfs_release_page,
+	.invalidatepage		= linvfs_invalidate_page,
 	.prepare_write		= linvfs_prepare_write,
 	.commit_write		= generic_commit_write,
 	.bmap			= linvfs_bmap,
diff --git a/fs/xfs/linux-2.6/xfs_aops.h b/fs/xfs/linux-2.6/xfs_aops.h
new file mode 100644
index 0000000..2fa6297
--- /dev/null
+++ b/fs/xfs/linux-2.6/xfs_aops.h
@@ -0,0 +1,50 @@
+/*
+ * Copyright (c) 2005 Silicon Graphics, Inc.  All Rights Reserved.
+ *
+ * This program is free software; you can redistribute it and/or modify it
+ * under the terms of version 2 of the GNU General Public License as
+ * published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it would be useful, but
+ * WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
+ *
+ * Further, this software is distributed without any warranty that it is
+ * free of the rightful claim of any third person regarding infringement
+ * or the like.  Any license provided herein, whether implied or
+ * otherwise, applies only to this software file.  Patent licenses, if
+ * any, provided herein do not apply to combinations of this program with
+ * other software, or any other product whatsoever.
+ *
+ * You should have received a copy of the GNU General Public License along
+ * with this program; if not, write the Free Software Foundation, Inc., 59
+ * Temple Place - Suite 330, Boston MA 02111-1307, USA.
+ *
+ * Contact information: Silicon Graphics, Inc., 1600 Amphitheatre Pkwy,
+ * Mountain View, CA  94043, or:
+ *
+ * http://www.sgi.com
+ *
+ * For further information regarding this notice, see:
+ *
+ * http://oss.sgi.com/projects/GenInfo/SGIGPLNoticeExplan/
+ */
+#ifndef __XFS_AOPS_H__
+#define __XFS_AOPS_H__
+
+extern struct workqueue_struct *xfsdatad_workqueue;
+extern mempool_t *xfs_ioend_pool;
+
+typedef void (*xfs_ioend_func_t)(void *);
+
+typedef struct xfs_ioend {
+	unsigned int		io_uptodate;	/* I/O status register */
+	atomic_t		io_remaining;	/* hold count */
+	struct vnode		*io_vnode;	/* file being written to */
+	struct buffer_head	*io_buffer_head;/* buffer linked list head */
+	size_t			io_size;	/* size of the extent */
+	xfs_off_t		io_offset;	/* offset in the file */
+	struct work_struct	io_work;	/* xfsdatad work queue */
+} xfs_ioend_t;
+
+#endif /* __XFS_IOPS_H__ */
diff --git a/fs/xfs/linux-2.6/xfs_buf.c b/fs/xfs/linux-2.6/xfs_buf.c
index df0cba2..655bf4a 100644
--- a/fs/xfs/linux-2.6/xfs_buf.c
+++ b/fs/xfs/linux-2.6/xfs_buf.c
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2000-2004 Silicon Graphics, Inc.  All Rights Reserved.
+ * Copyright (c) 2000-2005 Silicon Graphics, Inc.  All Rights Reserved.
  *
  * This program is free software; you can redistribute it and/or modify it
  * under the terms of version 2 of the GNU General Public License as
@@ -54,6 +54,7 @@
 #include <linux/percpu.h>
 #include <linux/blkdev.h>
 #include <linux/hash.h>
+#include <linux/kthread.h>
 
 #include "xfs_linux.h"
 
@@ -67,7 +68,7 @@ STATIC int xfsbufd_wakeup(int, unsigned int);
 STATIC void pagebuf_delwri_queue(xfs_buf_t *, int);
 
 STATIC struct workqueue_struct *xfslogd_workqueue;
-STATIC struct workqueue_struct *xfsdatad_workqueue;
+struct workqueue_struct *xfsdatad_workqueue;
 
 /*
  * Pagebuf debugging
@@ -590,8 +591,10 @@ found:
 		PB_SET_OWNER(pb);
 	}
 
-	if (pb->pb_flags & PBF_STALE)
+	if (pb->pb_flags & PBF_STALE) {
+		ASSERT((pb->pb_flags & _PBF_DELWRI_Q) == 0);
 		pb->pb_flags &= PBF_MAPPED;
+	}
 	PB_TRACE(pb, "got_lock", 0);
 	XFS_STATS_INC(pb_get_locked);
 	return (pb);
@@ -700,25 +703,6 @@ xfs_buf_read_flags(
 }
 
 /*
- * Create a skeletal pagebuf (no pages associated with it).
- */
-xfs_buf_t *
-pagebuf_lookup(
-	xfs_buftarg_t		*target,
-	loff_t			ioff,
-	size_t			isize,
-	page_buf_flags_t	flags)
-{
-	xfs_buf_t		*pb;
-
-	pb = pagebuf_allocate(flags);
-	if (pb) {
-		_pagebuf_initialize(pb, target, ioff, isize, flags);
-	}
-	return pb;
-}
-
-/*
  * If we are not low on memory then do the readahead in a deadlock
  * safe manner.
  */
@@ -913,22 +897,23 @@ pagebuf_rele(
 			do_free = 0;
 		}
 
-		if (pb->pb_flags & PBF_DELWRI) {
-			pb->pb_flags |= PBF_ASYNC;
-			atomic_inc(&pb->pb_hold);
-			pagebuf_delwri_queue(pb, 0);
-			do_free = 0;
-		} else if (pb->pb_flags & PBF_FS_MANAGED) {
+		if (pb->pb_flags & PBF_FS_MANAGED) {
 			do_free = 0;
 		}
 
 		if (do_free) {
+			ASSERT((pb->pb_flags & (PBF_DELWRI|_PBF_DELWRI_Q)) == 0);
 			list_del_init(&pb->pb_hash_list);
 			spin_unlock(&hash->bh_lock);
 			pagebuf_free(pb);
 		} else {
 			spin_unlock(&hash->bh_lock);
 		}
+	} else {
+		/*
+		 * Catch reference count leaks
+		 */
+		ASSERT(atomic_read(&pb->pb_hold) >= 0);
 	}
 }
 
@@ -1006,13 +991,24 @@ pagebuf_lock(
  *	pagebuf_unlock
  *
  *	pagebuf_unlock releases the lock on the buffer object created by
- *	pagebuf_lock or pagebuf_cond_lock (not any
- *	pinning of underlying pages created by pagebuf_pin).
+ *	pagebuf_lock or pagebuf_cond_lock (not any pinning of underlying pages
+ *	created by pagebuf_pin).
+ *
+ *	If the buffer is marked delwri but is not queued, do so before we
+ *	unlock the buffer as we need to set flags correctly. We also need to
+ *	take a reference for the delwri queue because the unlocker is going to
+ *	drop their's and they don't know we just queued it.
  */
 void
 pagebuf_unlock(				/* unlock buffer		*/
 	xfs_buf_t		*pb)	/* buffer to unlock		*/
 {
+	if ((pb->pb_flags & (PBF_DELWRI|_PBF_DELWRI_Q)) == PBF_DELWRI) {
+		atomic_inc(&pb->pb_hold);
+		pb->pb_flags |= PBF_ASYNC;
+		pagebuf_delwri_queue(pb, 0);
+	}
+
 	PB_CLEAR_OWNER(pb);
 	up(&pb->pb_sema);
 	PB_TRACE(pb, "unlock", 0);
@@ -1249,8 +1245,8 @@ bio_end_io_pagebuf(
 	int			error)
 {
 	xfs_buf_t		*pb = (xfs_buf_t *)bio->bi_private;
-	unsigned int		i, blocksize = pb->pb_target->pbr_bsize;
-	struct bio_vec		*bvec = bio->bi_io_vec;
+	unsigned int		blocksize = pb->pb_target->pbr_bsize;
+	struct bio_vec		*bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
 
 	if (bio->bi_size)
 		return 1;
@@ -1258,10 +1254,12 @@ bio_end_io_pagebuf(
 	if (!test_bit(BIO_UPTODATE, &bio->bi_flags))
 		pb->pb_error = EIO;
 
-	for (i = 0; i < bio->bi_vcnt; i++, bvec++) {
+	do {
 		struct page	*page = bvec->bv_page;
 
-		if (pb->pb_error) {
+		if (unlikely(pb->pb_error)) {
+			if (pb->pb_flags & PBF_READ)
+				ClearPageUptodate(page);
 			SetPageError(page);
 		} else if (blocksize == PAGE_CACHE_SIZE) {
 			SetPageUptodate(page);
@@ -1270,10 +1268,13 @@ bio_end_io_pagebuf(
 			set_page_region(page, bvec->bv_offset, bvec->bv_len);
 		}
 
+		if (--bvec >= bio->bi_io_vec)
+			prefetchw(&bvec->bv_page->flags);
+
 		if (_pagebuf_iolocked(pb)) {
 			unlock_page(page);
 		}
-	}
+	} while (bvec >= bio->bi_io_vec);
 
 	_pagebuf_iodone(pb, 1);
 	bio_put(bio);
@@ -1511,6 +1512,11 @@ again:
 			ASSERT(btp == bp->pb_target);
 			if (!(bp->pb_flags & PBF_FS_MANAGED)) {
 				spin_unlock(&hash->bh_lock);
+				/*
+				 * Catch superblock reference count leaks
+				 * immediately
+				 */
+				BUG_ON(bp->pb_bn == 0);
 				delay(100);
 				goto again;
 			}
@@ -1686,17 +1692,20 @@ pagebuf_delwri_queue(
 	int			unlock)
 {
 	PB_TRACE(pb, "delwri_q", (long)unlock);
-	ASSERT(pb->pb_flags & PBF_DELWRI);
+	ASSERT((pb->pb_flags & (PBF_DELWRI|PBF_ASYNC)) ==
+					(PBF_DELWRI|PBF_ASYNC));
 
 	spin_lock(&pbd_delwrite_lock);
 	/* If already in the queue, dequeue and place at tail */
 	if (!list_empty(&pb->pb_list)) {
+		ASSERT(pb->pb_flags & _PBF_DELWRI_Q);
 		if (unlock) {
 			atomic_dec(&pb->pb_hold);
 		}
 		list_del(&pb->pb_list);
 	}
 
+	pb->pb_flags |= _PBF_DELWRI_Q;
 	list_add_tail(&pb->pb_list, &pbd_delwrite_queue);
 	pb->pb_queuetime = jiffies;
 	spin_unlock(&pbd_delwrite_lock);
@@ -1713,10 +1722,11 @@ pagebuf_delwri_dequeue(
 
 	spin_lock(&pbd_delwrite_lock);
 	if ((pb->pb_flags & PBF_DELWRI) && !list_empty(&pb->pb_list)) {
+		ASSERT(pb->pb_flags & _PBF_DELWRI_Q);
 		list_del_init(&pb->pb_list);
 		dequeued = 1;
 	}
-	pb->pb_flags &= ~PBF_DELWRI;
+	pb->pb_flags &= ~(PBF_DELWRI|_PBF_DELWRI_Q);
 	spin_unlock(&pbd_delwrite_lock);
 
 	if (dequeued)
@@ -1733,9 +1743,7 @@ pagebuf_runall_queues(
 }
 
 /* Defines for pagebuf daemon */
-STATIC DECLARE_COMPLETION(xfsbufd_done);
 STATIC struct task_struct *xfsbufd_task;
-STATIC int xfsbufd_active;
 STATIC int xfsbufd_force_flush;
 STATIC int xfsbufd_force_sleep;
 
@@ -1761,14 +1769,8 @@ xfsbufd(
 	xfs_buftarg_t		*target;
 	xfs_buf_t		*pb, *n;
 
-	/*  Set up the thread  */
-	daemonize("xfsbufd");
 	current->flags |= PF_MEMALLOC;
 
-	xfsbufd_task = current;
-	xfsbufd_active = 1;
-	barrier();
-
 	INIT_LIST_HEAD(&tmp);
 	do {
 		if (unlikely(freezing(current))) {
@@ -1795,7 +1797,7 @@ xfsbufd(
 					break;
 				}
 
-				pb->pb_flags &= ~PBF_DELWRI;
+				pb->pb_flags &= ~(PBF_DELWRI|_PBF_DELWRI_Q);
 				pb->pb_flags |= PBF_WRITE;
 				list_move(&pb->pb_list, &tmp);
 			}
@@ -1816,9 +1818,9 @@ xfsbufd(
 			purge_addresses();
 
 		xfsbufd_force_flush = 0;
-	} while (xfsbufd_active);
+	} while (!kthread_should_stop());
 
-	complete_and_exit(&xfsbufd_done, 0);
+	return 0;
 }
 
 /*
@@ -1845,15 +1847,13 @@ xfs_flush_buftarg(
 		if (pb->pb_target != target)
 			continue;
 
-		ASSERT(pb->pb_flags & PBF_DELWRI);
+		ASSERT(pb->pb_flags & (PBF_DELWRI|_PBF_DELWRI_Q));
 		PB_TRACE(pb, "walkq2", (long)pagebuf_ispin(pb));
 		if (pagebuf_ispin(pb)) {
 			pincount++;
 			continue;
 		}
 
-		pb->pb_flags &= ~PBF_DELWRI;
-		pb->pb_flags |= PBF_WRITE;
 		list_move(&pb->pb_list, &tmp);
 	}
 	spin_unlock(&pbd_delwrite_lock);
@@ -1862,12 +1862,14 @@ xfs_flush_buftarg(
 	 * Dropped the delayed write list lock, now walk the temporary list
 	 */
 	list_for_each_entry_safe(pb, n, &tmp, pb_list) {
+		pagebuf_lock(pb);
+		pb->pb_flags &= ~(PBF_DELWRI|_PBF_DELWRI_Q);
+		pb->pb_flags |= PBF_WRITE;
 		if (wait)
 			pb->pb_flags &= ~PBF_ASYNC;
 		else
 			list_del_init(&pb->pb_list);
 
-		pagebuf_lock(pb);
 		pagebuf_iostrategy(pb);
 	}
 
@@ -1901,9 +1903,11 @@ xfs_buf_daemons_start(void)
 	if (!xfsdatad_workqueue)
 		goto out_destroy_xfslogd_workqueue;
 
-	error = kernel_thread(xfsbufd, NULL, CLONE_FS|CLONE_FILES);
-	if (error < 0)
+	xfsbufd_task = kthread_run(xfsbufd, NULL, "xfsbufd");
+	if (IS_ERR(xfsbufd_task)) {
+		error = PTR_ERR(xfsbufd_task);
 		goto out_destroy_xfsdatad_workqueue;
+	}
 	return 0;
 
  out_destroy_xfsdatad_workqueue:
@@ -1920,10 +1924,7 @@ xfs_buf_daemons_start(void)
 STATIC void
 xfs_buf_daemons_stop(void)
 {
-	xfsbufd_active = 0;
-	barrier();
-	wait_for_completion(&xfsbufd_done);
-
+	kthread_stop(xfsbufd_task);
 	destroy_workqueue(xfslogd_workqueue);
 	destroy_workqueue(xfsdatad_workqueue);
 }
diff --git a/fs/xfs/linux-2.6/xfs_buf.h b/fs/xfs/linux-2.6/xfs_buf.h
index 3f8f69a..67c19f7 100644
--- a/fs/xfs/linux-2.6/xfs_buf.h
+++ b/fs/xfs/linux-2.6/xfs_buf.h
@@ -89,6 +89,7 @@ typedef enum page_buf_flags_e {		/* pb_flags values */
 	_PBF_PAGE_CACHE = (1 << 17),/* backed by pagecache		   */
 	_PBF_KMEM_ALLOC = (1 << 18),/* backed by kmem_alloc()		   */
 	_PBF_RUN_QUEUES = (1 << 19),/* run block device task queue	   */
+	_PBF_DELWRI_Q = (1 << 21),   /* buffer on delwri queue		   */
 } page_buf_flags_t;
 
 #define PBF_UPDATE (PBF_READ | PBF_WRITE)
@@ -206,13 +207,6 @@ extern xfs_buf_t *xfs_buf_read_flags(	/* allocate and read a buffer	*/
 #define xfs_buf_read(target, blkno, len, flags) \
 	xfs_buf_read_flags((target), (blkno), (len), PBF_LOCK | PBF_MAPPED)
 
-extern xfs_buf_t *pagebuf_lookup(
-		xfs_buftarg_t *,
-		loff_t,			/* starting offset of range	*/
-		size_t,			/* length of range		*/
-		page_buf_flags_t);	/* PBF_READ, PBF_WRITE,		*/
-					/* PBF_FORCEIO, 		*/
-
 extern xfs_buf_t *pagebuf_get_empty(	/* allocate pagebuf struct with	*/
 					/*  no memory or disk address	*/
 		size_t len,
@@ -344,8 +338,6 @@ extern void pagebuf_trace(
 
 
 
-
-
 /* These are just for xfs_syncsub... it sets an internal variable
  * then passes it to VOP_FLUSH_PAGES or adds the flags to a newly gotten buf_t
  */
@@ -452,7 +444,7 @@ extern void pagebuf_trace(
 
 #define XFS_BUF_PTR(bp)		(xfs_caddr_t)((bp)->pb_addr)
 
-extern inline xfs_caddr_t xfs_buf_offset(xfs_buf_t *bp, size_t offset)
+static inline xfs_caddr_t xfs_buf_offset(xfs_buf_t *bp, size_t offset)
 {
 	if (bp->pb_flags & PBF_MAPPED)
 		return XFS_BUF_PTR(bp) + offset;
diff --git a/fs/xfs/linux-2.6/xfs_file.c b/fs/xfs/linux-2.6/xfs_file.c
index f1ce432..3881622 100644
--- a/fs/xfs/linux-2.6/xfs_file.c
+++ b/fs/xfs/linux-2.6/xfs_file.c
@@ -311,6 +311,31 @@ linvfs_fsync(
 
 #define nextdp(dp)      ((struct xfs_dirent *)((char *)(dp) + (dp)->d_reclen))
 
+#ifdef CONFIG_XFS_DMAPI
+
+STATIC struct page *
+linvfs_filemap_nopage(
+	struct vm_area_struct	*area,
+	unsigned long		address,
+	int			*type)
+{
+	struct inode	*inode = area->vm_file->f_dentry->d_inode;
+	vnode_t		*vp = LINVFS_GET_VP(inode);
+	xfs_mount_t	*mp = XFS_VFSTOM(vp->v_vfsp);
+	int		error;
+
+	ASSERT_ALWAYS(vp->v_vfsp->vfs_flag & VFS_DMI);
+
+	error = XFS_SEND_MMAP(mp, area, 0);
+	if (error)
+		return NULL;
+
+	return filemap_nopage(area, address, type);
+}
+
+#endif /* CONFIG_XFS_DMAPI */
+
+
 STATIC int
 linvfs_readdir(
 	struct file	*filp,
@@ -390,14 +415,6 @@ done:
 	return -error;
 }
 
-#ifdef CONFIG_XFS_DMAPI
-STATIC void
-linvfs_mmap_close(
-	struct vm_area_struct	*vma)
-{
-	xfs_dm_mm_put(vma);
-}
-#endif /* CONFIG_XFS_DMAPI */
 
 STATIC int
 linvfs_file_mmap(
@@ -411,16 +428,11 @@ linvfs_file_mmap(
 
 	vma->vm_ops = &linvfs_file_vm_ops;
 
-	if (vp->v_vfsp->vfs_flag & VFS_DMI) {
-		xfs_mount_t	*mp = XFS_VFSTOM(vp->v_vfsp);
-
-		error = -XFS_SEND_MMAP(mp, vma, 0);
-		if (error)
-			return error;
 #ifdef CONFIG_XFS_DMAPI
+	if (vp->v_vfsp->vfs_flag & VFS_DMI) {
 		vma->vm_ops = &linvfs_dmapi_file_vm_ops;
-#endif
 	}
+#endif /* CONFIG_XFS_DMAPI */
 
 	VOP_SETATTR(vp, &va, XFS_AT_UPDATIME, NULL, error);
 	if (!error)
@@ -474,6 +486,7 @@ linvfs_ioctl_invis(
 	return error;
 }
 
+#ifdef CONFIG_XFS_DMAPI
 #ifdef HAVE_VMOP_MPROTECT
 STATIC int
 linvfs_mprotect(
@@ -494,6 +507,7 @@ linvfs_mprotect(
 	return error;
 }
 #endif /* HAVE_VMOP_MPROTECT */
+#endif /* CONFIG_XFS_DMAPI */
 
 #ifdef HAVE_FOP_OPEN_EXEC
 /* If the user is attempting to execute a file that is offline then
@@ -528,49 +542,10 @@ open_exec_out:
 }
 #endif /* HAVE_FOP_OPEN_EXEC */
 
-/*
- * Temporary workaround to the AIO direct IO write problem.
- * This code can go and we can revert to do_sync_write once
- * the writepage(s) rework is merged.
- */
-STATIC ssize_t
-linvfs_write(
-	struct file	*filp,
-	const char	__user *buf,
-	size_t		len,
-	loff_t		*ppos)
-{
-	struct kiocb	kiocb;
-	ssize_t		ret;
-
-	init_sync_kiocb(&kiocb, filp);
-	kiocb.ki_pos = *ppos;
-	ret = __linvfs_write(&kiocb, buf, 0, len, kiocb.ki_pos);
-	*ppos = kiocb.ki_pos;
-	return ret;
-}
-STATIC ssize_t
-linvfs_write_invis(
-	struct file	*filp,
-	const char	__user *buf,
-	size_t		len,
-	loff_t		*ppos)
-{
-	struct kiocb	kiocb;
-	ssize_t		ret;
-
-	init_sync_kiocb(&kiocb, filp);
-	kiocb.ki_pos = *ppos;
-	ret = __linvfs_write(&kiocb, buf, IO_INVIS, len, kiocb.ki_pos);
-	*ppos = kiocb.ki_pos;
-	return ret;
-}
-
-
 struct file_operations linvfs_file_operations = {
 	.llseek		= generic_file_llseek,
 	.read		= do_sync_read,
-	.write		= linvfs_write,
+	.write		= do_sync_write,
 	.readv		= linvfs_readv,
 	.writev		= linvfs_writev,
 	.aio_read	= linvfs_aio_read,
@@ -592,7 +567,7 @@ struct file_operations linvfs_file_operations = {
 struct file_operations linvfs_invis_file_operations = {
 	.llseek		= generic_file_llseek,
 	.read		= do_sync_read,
-	.write		= linvfs_write_invis,
+	.write		= do_sync_write,
 	.readv		= linvfs_readv_invis,
 	.writev		= linvfs_writev_invis,
 	.aio_read	= linvfs_aio_read_invis,
@@ -626,8 +601,7 @@ static struct vm_operations_struct linvfs_file_vm_ops = {
 
 #ifdef CONFIG_XFS_DMAPI
 static struct vm_operations_struct linvfs_dmapi_file_vm_ops = {
-	.close		= linvfs_mmap_close,
-	.nopage		= filemap_nopage,
+	.nopage		= linvfs_filemap_nopage,
 	.populate	= filemap_populate,
 #ifdef HAVE_VMOP_MPROTECT
 	.mprotect	= linvfs_mprotect,
diff --git a/fs/xfs/linux-2.6/xfs_ioctl.c b/fs/xfs/linux-2.6/xfs_ioctl.c
index 05a447e..6a3326b 100644
--- a/fs/xfs/linux-2.6/xfs_ioctl.c
+++ b/fs/xfs/linux-2.6/xfs_ioctl.c
@@ -141,13 +141,19 @@ xfs_find_handle(
 		return -XFS_ERROR(EINVAL);
 	}
 
-	/* we need the vnode */
-	vp = LINVFS_GET_VP(inode);
-	if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK) {
+	switch (inode->i_mode & S_IFMT) {
+	case S_IFREG:
+	case S_IFDIR:
+	case S_IFLNK:
+		break;
+	default:
 		iput(inode);
 		return -XFS_ERROR(EBADF);
 	}
 
+	/* we need the vnode */
+	vp = LINVFS_GET_VP(inode);
+
 	/* now we can grab the fsid */
 	memcpy(&handle.ha_fsid, vp->v_vfsp->vfs_altfsid, sizeof(xfs_fsid_t));
 	hsize = sizeof(xfs_fsid_t);
@@ -386,7 +392,7 @@ xfs_readlink_by_handle(
 		return -error;
 
 	/* Restrict this handle operation to symlinks only. */
-	if (vp->v_type != VLNK) {
+	if (!S_ISLNK(inode->i_mode)) {
 		VN_RELE(vp);
 		return -XFS_ERROR(EINVAL);
 	}
@@ -982,10 +988,10 @@ xfs_ioc_space(
 	if (vp->v_inode.i_flags & (S_IMMUTABLE|S_APPEND))
 		return -XFS_ERROR(EPERM);
 
-	if (!(filp->f_flags & FMODE_WRITE))
+	if (!(filp->f_mode & FMODE_WRITE))
 		return -XFS_ERROR(EBADF);
 
-	if (vp->v_type != VREG)
+	if (!VN_ISREG(vp))
 		return -XFS_ERROR(EINVAL);
 
 	if (copy_from_user(&bf, arg, sizeof(bf)))
diff --git a/fs/xfs/linux-2.6/xfs_ioctl32.c b/fs/xfs/linux-2.6/xfs_ioctl32.c
index 0f8f138..4636b7f 100644
--- a/fs/xfs/linux-2.6/xfs_ioctl32.c
+++ b/fs/xfs/linux-2.6/xfs_ioctl32.c
@@ -47,8 +47,52 @@
 #include "xfs_vnode.h"
 #include "xfs_dfrag.h"
 
+#define  _NATIVE_IOC(cmd, type) \
+	  _IOC(_IOC_DIR(cmd), _IOC_TYPE(cmd), _IOC_NR(cmd), sizeof(type))
+
 #if defined(CONFIG_IA64) || defined(CONFIG_X86_64)
 #define BROKEN_X86_ALIGNMENT
+/* on ia32 l_start is on a 32-bit boundary */
+typedef struct xfs_flock64_32 {
+	__s16		l_type;
+	__s16		l_whence;
+	__s64		l_start	__attribute__((packed));
+			/* len == 0 means until end of file */
+	__s64		l_len __attribute__((packed));
+	__s32		l_sysid;
+	__u32		l_pid;
+	__s32		l_pad[4];	/* reserve area */
+} xfs_flock64_32_t;
+
+#define XFS_IOC_ALLOCSP_32	_IOW ('X', 10, struct xfs_flock64_32)
+#define XFS_IOC_FREESP_32	_IOW ('X', 11, struct xfs_flock64_32)
+#define XFS_IOC_ALLOCSP64_32	_IOW ('X', 36, struct xfs_flock64_32)
+#define XFS_IOC_FREESP64_32	_IOW ('X', 37, struct xfs_flock64_32)
+#define XFS_IOC_RESVSP_32	_IOW ('X', 40, struct xfs_flock64_32)
+#define XFS_IOC_UNRESVSP_32	_IOW ('X', 41, struct xfs_flock64_32)
+#define XFS_IOC_RESVSP64_32	_IOW ('X', 42, struct xfs_flock64_32)
+#define XFS_IOC_UNRESVSP64_32	_IOW ('X', 43, struct xfs_flock64_32)
+
+/* just account for different alignment */
+STATIC unsigned long
+xfs_ioctl32_flock(
+	unsigned long		arg)
+{
+	xfs_flock64_32_t	__user *p32 = (void __user *)arg;
+	xfs_flock64_t		__user *p = compat_alloc_user_space(sizeof(*p));
+
+	if (copy_in_user(&p->l_type,	&p32->l_type,	sizeof(s16)) ||
+	    copy_in_user(&p->l_whence,	&p32->l_whence, sizeof(s16)) ||
+	    copy_in_user(&p->l_start,	&p32->l_start,	sizeof(s64)) ||
+	    copy_in_user(&p->l_len,	&p32->l_len,	sizeof(s64)) ||
+	    copy_in_user(&p->l_sysid,	&p32->l_sysid,	sizeof(s32)) ||
+	    copy_in_user(&p->l_pid,	&p32->l_pid,	sizeof(u32)) ||
+	    copy_in_user(&p->l_pad,	&p32->l_pad,	4*sizeof(u32)))
+		return -EFAULT;
+	
+	return (unsigned long)p;
+}
+
 #else
 
 typedef struct xfs_fsop_bulkreq32 {
@@ -103,7 +147,6 @@ __linvfs_compat_ioctl(int mode, struct file *f, unsigned cmd, unsigned long arg)
 /* not handled
 	case XFS_IOC_FD_TO_HANDLE:
 	case XFS_IOC_PATH_TO_HANDLE:
-	case XFS_IOC_PATH_TO_HANDLE:
 	case XFS_IOC_PATH_TO_FSHANDLE:
 	case XFS_IOC_OPEN_BY_HANDLE:
 	case XFS_IOC_FSSETDM_BY_HANDLE:
@@ -124,8 +167,21 @@ __linvfs_compat_ioctl(int mode, struct file *f, unsigned cmd, unsigned long arg)
 	case XFS_IOC_ERROR_CLEARALL:
 		break;
 
-#ifndef BROKEN_X86_ALIGNMENT
-	/* xfs_flock_t and xfs_bstat_t have wrong u32 vs u64 alignment */
+#ifdef BROKEN_X86_ALIGNMENT
+	/* xfs_flock_t has wrong u32 vs u64 alignment */
+	case XFS_IOC_ALLOCSP_32:
+	case XFS_IOC_FREESP_32:
+	case XFS_IOC_ALLOCSP64_32:
+	case XFS_IOC_FREESP64_32:
+	case XFS_IOC_RESVSP_32:
+	case XFS_IOC_UNRESVSP_32:
+	case XFS_IOC_RESVSP64_32:
+	case XFS_IOC_UNRESVSP64_32:
+		arg = xfs_ioctl32_flock(arg);
+		cmd = _NATIVE_IOC(cmd, struct xfs_flock64);
+		break;
+
+#else /* These are handled fine if no alignment issues */
 	case XFS_IOC_ALLOCSP:
 	case XFS_IOC_FREESP:
 	case XFS_IOC_RESVSP:
@@ -134,6 +190,9 @@ __linvfs_compat_ioctl(int mode, struct file *f, unsigned cmd, unsigned long arg)
 	case XFS_IOC_FREESP64:
 	case XFS_IOC_RESVSP64:
 	case XFS_IOC_UNRESVSP64:
+		break;
+
+	/* xfs_bstat_t still has wrong u32 vs u64 alignment */
 	case XFS_IOC_SWAPEXT:
 		break;
 
diff --git a/fs/xfs/linux-2.6/xfs_iops.c b/fs/xfs/linux-2.6/xfs_iops.c
index f252605..77708a8 100644
--- a/fs/xfs/linux-2.6/xfs_iops.c
+++ b/fs/xfs/linux-2.6/xfs_iops.c
@@ -140,7 +140,6 @@ linvfs_mknod(
 
 	memset(&va, 0, sizeof(va));
 	va.va_mask = XFS_AT_TYPE|XFS_AT_MODE;
-	va.va_type = IFTOVT(mode);
 	va.va_mode = mode;
 
 	switch (mode & S_IFMT) {
@@ -308,14 +307,13 @@ linvfs_symlink(
 	cvp = NULL;
 
 	memset(&va, 0, sizeof(va));
-	va.va_type = VLNK;
-	va.va_mode = irix_symlink_mode ? 0777 & ~current->fs->umask : S_IRWXUGO;
+	va.va_mode = S_IFLNK |
+		(irix_symlink_mode ? 0777 & ~current->fs->umask : S_IRWXUGO);
 	va.va_mask = XFS_AT_TYPE|XFS_AT_MODE;
 
 	error = 0;
 	VOP_SYMLINK(dvp, dentry, &va, (char *)symname, &cvp, NULL, error);
 	if (!error && cvp) {
-		ASSERT(cvp->v_type == VLNK);
 		ip = LINVFS_GET_IP(cvp);
 		d_instantiate(dentry, ip);
 		validate_fields(dir);
@@ -425,9 +423,14 @@ linvfs_follow_link(
 	return NULL;
 }
 
-static void linvfs_put_link(struct dentry *dentry, struct nameidata *nd, void *p)
+STATIC void
+linvfs_put_link(
+	struct dentry	*dentry,
+	struct nameidata *nd,
+	void		*p)
 {
-	char *s = nd_get_link(nd);
+	char		*s = nd_get_link(nd);
+
 	if (!IS_ERR(s))
 		kfree(s);
 }
diff --git a/fs/xfs/linux-2.6/xfs_linux.h b/fs/xfs/linux-2.6/xfs_linux.h
index 42dc5e4..68c5d88 100644
--- a/fs/xfs/linux-2.6/xfs_linux.h
+++ b/fs/xfs/linux-2.6/xfs_linux.h
@@ -64,7 +64,6 @@
 #include <sema.h>
 #include <time.h>
 
-#include <support/qsort.h>
 #include <support/ktrace.h>
 #include <support/debug.h>
 #include <support/move.h>
@@ -104,6 +103,7 @@
 #include <xfs_stats.h>
 #include <xfs_sysctl.h>
 #include <xfs_iops.h>
+#include <xfs_aops.h>
 #include <xfs_super.h>
 #include <xfs_globals.h>
 #include <xfs_fs_subr.h>
@@ -254,11 +254,18 @@ static inline void set_buffer_unwritten_io(struct buffer_head *bh)
 #define MAX(a,b)	(max(a,b))
 #define howmany(x, y)	(((x)+((y)-1))/(y))
 #define roundup(x, y)	((((x)+((y)-1))/(y))*(y))
+#define qsort(a,n,s,fn)	sort(a,n,s,fn,NULL)
 
+/*
+ * Various platform dependent calls that don't fit anywhere else
+ */
 #define xfs_stack_trace()	dump_stack()
-
 #define xfs_itruncate_data(ip, off)	\
 	(-vmtruncate(LINVFS_GET_IP(XFS_ITOV(ip)), (off)))
+#define xfs_statvfs_fsid(statp, mp)	\
+	({ u64 id = huge_encode_dev((mp)->m_dev);	\
+	   __kernel_fsid_t *fsid = &(statp)->f_fsid;	\
+	(fsid->val[0] = (u32)id, fsid->val[1] = (u32)(id >> 32)); })
 
 
 /* Move the kernel do_div definition off to one side */
@@ -371,6 +378,4 @@ static inline __uint64_t roundup_64(__uint64_t x, __uint32_t y)
 	return(x * y);
 }
 
-#define qsort(a, n, s, cmp) sort(a, n, s, cmp, NULL)
-
 #endif /* __XFS_LINUX__ */
diff --git a/fs/xfs/linux-2.6/xfs_lrw.c b/fs/xfs/linux-2.6/xfs_lrw.c
index acab58c..3b5fabe 100644
--- a/fs/xfs/linux-2.6/xfs_lrw.c
+++ b/fs/xfs/linux-2.6/xfs_lrw.c
@@ -660,9 +660,6 @@ xfs_write(
 			(xip->i_d.di_flags & XFS_DIFLAG_REALTIME) ?
 				mp->m_rtdev_targp : mp->m_ddev_targp;
 
-		if (ioflags & IO_ISAIO)
-			return XFS_ERROR(-ENOSYS);
-
 		if ((pos & target->pbr_smask) || (count & target->pbr_smask))
 			return XFS_ERROR(-EINVAL);
 
diff --git a/fs/xfs/linux-2.6/xfs_lrw.h b/fs/xfs/linux-2.6/xfs_lrw.h
index f197a72..6294dcd 100644
--- a/fs/xfs/linux-2.6/xfs_lrw.h
+++ b/fs/xfs/linux-2.6/xfs_lrw.h
@@ -70,9 +70,10 @@ struct xfs_iomap;
 #define	XFS_SENDFILE_ENTER	21
 #define	XFS_WRITEPAGE_ENTER	22
 #define	XFS_RELEASEPAGE_ENTER	23
-#define	XFS_IOMAP_ALLOC_ENTER	24
-#define	XFS_IOMAP_ALLOC_MAP	25
-#define	XFS_IOMAP_UNWRITTEN	26
+#define	XFS_INVALIDPAGE_ENTER	24
+#define	XFS_IOMAP_ALLOC_ENTER	25
+#define	XFS_IOMAP_ALLOC_MAP	26
+#define	XFS_IOMAP_UNWRITTEN	27
 extern void xfs_rw_enter_trace(int, struct xfs_iocore *,
 				void *, size_t, loff_t, int);
 extern void xfs_inval_cached_trace(struct xfs_iocore *,
diff --git a/fs/xfs/linux-2.6/xfs_super.c b/fs/xfs/linux-2.6/xfs_super.c
index f6dd7de..0da87bf 100644
--- a/fs/xfs/linux-2.6/xfs_super.c
+++ b/fs/xfs/linux-2.6/xfs_super.c
@@ -70,11 +70,15 @@
 #include <linux/namei.h>
 #include <linux/init.h>
 #include <linux/mount.h>
+#include <linux/mempool.h>
 #include <linux/writeback.h>
+#include <linux/kthread.h>
 
 STATIC struct quotactl_ops linvfs_qops;
 STATIC struct super_operations linvfs_sops;
-STATIC kmem_zone_t *linvfs_inode_zone;
+STATIC kmem_zone_t *xfs_vnode_zone;
+STATIC kmem_zone_t *xfs_ioend_zone;
+mempool_t *xfs_ioend_pool;
 
 STATIC struct xfs_mount_args *
 xfs_args_allocate(
@@ -138,24 +142,25 @@ STATIC __inline__ void
 xfs_set_inodeops(
 	struct inode		*inode)
 {
-	vnode_t			*vp = LINVFS_GET_VP(inode);
-
-	if (vp->v_type == VNON) {
-		vn_mark_bad(vp);
-	} else if (S_ISREG(inode->i_mode)) {
+	switch (inode->i_mode & S_IFMT) {
+	case S_IFREG:
 		inode->i_op = &linvfs_file_inode_operations;
 		inode->i_fop = &linvfs_file_operations;
 		inode->i_mapping->a_ops = &linvfs_aops;
-	} else if (S_ISDIR(inode->i_mode)) {
+		break;
+	case S_IFDIR:
 		inode->i_op = &linvfs_dir_inode_operations;
 		inode->i_fop = &linvfs_dir_operations;
-	} else if (S_ISLNK(inode->i_mode)) {
+		break;
+	case S_IFLNK:
 		inode->i_op = &linvfs_symlink_inode_operations;
 		if (inode->i_blocks)
 			inode->i_mapping->a_ops = &linvfs_aops;
-	} else {
+		break;
+	default:
 		inode->i_op = &linvfs_file_inode_operations;
 		init_special_inode(inode, inode->i_mode, inode->i_rdev);
+		break;
 	}
 }
 
@@ -167,16 +172,23 @@ xfs_revalidate_inode(
 {
 	struct inode		*inode = LINVFS_GET_IP(vp);
 
-	inode->i_mode	= (ip->i_d.di_mode & MODEMASK) | VTTOIF(vp->v_type);
+	inode->i_mode	= ip->i_d.di_mode;
 	inode->i_nlink	= ip->i_d.di_nlink;
 	inode->i_uid	= ip->i_d.di_uid;
 	inode->i_gid	= ip->i_d.di_gid;
-	if (((1 << vp->v_type) & ((1<<VBLK) | (1<<VCHR))) == 0) {
+
+	switch (inode->i_mode & S_IFMT) {
+	case S_IFBLK:
+	case S_IFCHR:
+		inode->i_rdev =
+			MKDEV(sysv_major(ip->i_df.if_u2.if_rdev) & 0x1ff,
+			      sysv_minor(ip->i_df.if_u2.if_rdev));
+		break;
+	default:
 		inode->i_rdev = 0;
-	} else {
-		xfs_dev_t dev = ip->i_df.if_u2.if_rdev;
-		inode->i_rdev = MKDEV(sysv_major(dev) & 0x1ff, sysv_minor(dev));
+		break;
 	}
+
 	inode->i_blksize = PAGE_CACHE_SIZE;
 	inode->i_generation = ip->i_d.di_gen;
 	i_size_write(inode, ip->i_d.di_size);
@@ -231,7 +243,6 @@ xfs_initialize_vnode(
 	 * finish our work.
 	 */
 	if (ip->i_d.di_mode != 0 && unlock && (inode->i_state & I_NEW)) {
-		vp->v_type = IFTOVT(ip->i_d.di_mode);
 		xfs_revalidate_inode(XFS_BHVTOM(bdp), vp, ip);
 		xfs_set_inodeops(inode);
 	
@@ -274,8 +285,7 @@ linvfs_alloc_inode(
 {
 	vnode_t			*vp;
 
-	vp = (vnode_t *)kmem_cache_alloc(linvfs_inode_zone, 
-                kmem_flags_convert(KM_SLEEP));
+	vp = kmem_cache_alloc(xfs_vnode_zone, kmem_flags_convert(KM_SLEEP));
 	if (!vp)
 		return NULL;
 	return LINVFS_GET_IP(vp);
@@ -285,11 +295,11 @@ STATIC void
 linvfs_destroy_inode(
 	struct inode		*inode)
 {
-	kmem_cache_free(linvfs_inode_zone, LINVFS_GET_VP(inode));
+	kmem_zone_free(xfs_vnode_zone, LINVFS_GET_VP(inode));
 }
 
 STATIC void
-init_once(
+linvfs_inode_init_once(
 	void			*data,
 	kmem_cache_t		*cachep,
 	unsigned long		flags)
@@ -302,21 +312,41 @@ init_once(
 }
 
 STATIC int
-init_inodecache( void )
+linvfs_init_zones(void)
 {
-	linvfs_inode_zone = kmem_cache_create("linvfs_icache",
+	xfs_vnode_zone = kmem_cache_create("xfs_vnode",
 				sizeof(vnode_t), 0, SLAB_RECLAIM_ACCOUNT,
-				init_once, NULL);
-	if (linvfs_inode_zone == NULL)
-		return -ENOMEM;
+				linvfs_inode_init_once, NULL);
+	if (!xfs_vnode_zone)
+		goto out;
+
+	xfs_ioend_zone = kmem_zone_init(sizeof(xfs_ioend_t), "xfs_ioend");
+	if (!xfs_ioend_zone)
+		goto out_destroy_vnode_zone;
+
+	xfs_ioend_pool = mempool_create(4 * MAX_BUF_PER_PAGE,
+			mempool_alloc_slab, mempool_free_slab,
+			xfs_ioend_zone);
+	if (!xfs_ioend_pool)
+		goto out_free_ioend_zone;
+
 	return 0;
+
+
+ out_free_ioend_zone:
+	kmem_zone_destroy(xfs_ioend_zone);
+ out_destroy_vnode_zone:
+	kmem_zone_destroy(xfs_vnode_zone);
+ out:
+	return -ENOMEM;
 }
 
 STATIC void
-destroy_inodecache( void )
+linvfs_destroy_zones(void)
 {
-	if (kmem_cache_destroy(linvfs_inode_zone))
-		printk(KERN_WARNING "%s: cache still in use!\n", __FUNCTION__);
+	mempool_destroy(xfs_ioend_pool);
+	kmem_zone_destroy(xfs_vnode_zone);
+	kmem_zone_destroy(xfs_ioend_zone);
 }
 
 /*
@@ -354,17 +384,38 @@ linvfs_clear_inode(
 	struct inode		*inode)
 {
 	vnode_t			*vp = LINVFS_GET_VP(inode);
+	int			error, cache;
 
-	if (vp) {
-		vn_rele(vp);
-		vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
-		/*
-		 * Do all our cleanup, and remove this vnode.
-		 */
-		vn_remove(vp);
+	vn_trace_entry(vp, "clear_inode", (inst_t *)__return_address);
+
+	XFS_STATS_INC(vn_rele);
+	XFS_STATS_INC(vn_remove);
+	XFS_STATS_INC(vn_reclaim);
+	XFS_STATS_DEC(vn_active);
+
+	/*
+	 * This can happen because xfs_iget_core calls xfs_idestroy if we
+	 * find an inode with di_mode == 0 but without IGET_CREATE set.
+	 */
+	if (vp->v_fbhv)
+		VOP_INACTIVE(vp, NULL, cache);
+
+	VN_LOCK(vp);
+	vp->v_flag &= ~VMODIFIED;
+	VN_UNLOCK(vp, 0);
+
+	if (vp->v_fbhv) {
+		VOP_RECLAIM(vp, error);
+		if (error)
+			panic("vn_purge: cannot reclaim");
 	}
-}
 
+	ASSERT(vp->v_fbhv == NULL);
+
+#ifdef XFS_VNODE_TRACE
+	ktrace_free(vp->v_trace);
+#endif
+}
 
 /*
  * Enqueue a work item to be picked up by the vfs xfssyncd thread.
@@ -466,25 +517,16 @@ xfssyncd(
 {
 	long			timeleft;
 	vfs_t			*vfsp = (vfs_t *) arg;
-	struct list_head	tmp;
 	struct vfs_sync_work	*work, *n;
+	LIST_HEAD		(tmp);
 
-	daemonize("xfssyncd");
-
-	vfsp->vfs_sync_work.w_vfs = vfsp;
-	vfsp->vfs_sync_work.w_syncer = vfs_sync_worker;
-	vfsp->vfs_sync_task = current;
-	wmb();
-	wake_up(&vfsp->vfs_wait_sync_task);
-
-	INIT_LIST_HEAD(&tmp);
 	timeleft = (xfs_syncd_centisecs * HZ) / 100;
 	for (;;) {
 		set_current_state(TASK_INTERRUPTIBLE);
 		timeleft = schedule_timeout(timeleft);
 		/* swsusp */
 		try_to_freeze();
-		if (vfsp->vfs_flag & VFS_UMOUNT)
+		if (kthread_should_stop())
 			break;
 
 		spin_lock(&vfsp->vfs_sync_lock);
@@ -513,10 +555,6 @@ xfssyncd(
 		}
 	}
 
-	vfsp->vfs_sync_task = NULL;
-	wmb();
-	wake_up(&vfsp->vfs_wait_sync_task);
-
 	return 0;
 }
 
@@ -524,13 +562,11 @@ STATIC int
 linvfs_start_syncd(
 	vfs_t			*vfsp)
 {
-	int			pid;
-
-	pid = kernel_thread(xfssyncd, (void *) vfsp,
-			CLONE_VM | CLONE_FS | CLONE_FILES);
-	if (pid < 0)
-		return -pid;
-	wait_event(vfsp->vfs_wait_sync_task, vfsp->vfs_sync_task);
+	vfsp->vfs_sync_work.w_syncer = vfs_sync_worker;
+	vfsp->vfs_sync_work.w_vfs = vfsp;
+	vfsp->vfs_sync_task = kthread_run(xfssyncd, vfsp, "xfssyncd");
+	if (IS_ERR(vfsp->vfs_sync_task))
+		return -PTR_ERR(vfsp->vfs_sync_task);
 	return 0;
 }
 
@@ -538,11 +574,7 @@ STATIC void
 linvfs_stop_syncd(
 	vfs_t			*vfsp)
 {
-	vfsp->vfs_flag |= VFS_UMOUNT;
-	wmb();
-
-	wake_up_process(vfsp->vfs_sync_task);
-	wait_event(vfsp->vfs_wait_sync_task, !vfsp->vfs_sync_task);
+	kthread_stop(vfsp->vfs_sync_task);
 }
 
 STATIC void
@@ -866,9 +898,9 @@ init_xfs_fs( void )
 
 	ktrace_init(64);
 
-	error = init_inodecache();
+	error = linvfs_init_zones();
 	if (error < 0)
-		goto undo_inodecache;
+		goto undo_zones;
 
 	error = pagebuf_init();
 	if (error < 0)
@@ -889,9 +921,9 @@ undo_register:
 	pagebuf_terminate();
 
 undo_pagebuf:
-	destroy_inodecache();
+	linvfs_destroy_zones();
 
-undo_inodecache:
+undo_zones:
 	return error;
 }
 
@@ -903,7 +935,7 @@ exit_xfs_fs( void )
 	unregister_filesystem(&xfs_fs_type);
 	xfs_cleanup();
 	pagebuf_terminate();
-	destroy_inodecache();
+	linvfs_destroy_zones();
 	ktrace_uninit();
 }
 
diff --git a/fs/xfs/linux-2.6/xfs_vfs.c b/fs/xfs/linux-2.6/xfs_vfs.c
index 669c616..34cc902 100644
--- a/fs/xfs/linux-2.6/xfs_vfs.c
+++ b/fs/xfs/linux-2.6/xfs_vfs.c
@@ -251,7 +251,6 @@ vfs_allocate( void )
 	bhv_head_init(VFS_BHVHEAD(vfsp), "vfs");
 	INIT_LIST_HEAD(&vfsp->vfs_sync_list);
 	spin_lock_init(&vfsp->vfs_sync_lock);
-	init_waitqueue_head(&vfsp->vfs_wait_sync_task);
 	init_waitqueue_head(&vfsp->vfs_wait_single_sync_task);
 	return vfsp;
 }
diff --git a/fs/xfs/linux-2.6/xfs_vfs.h b/fs/xfs/linux-2.6/xfs_vfs.h
index 7ee1f71..f0ab574f 100644
--- a/fs/xfs/linux-2.6/xfs_vfs.h
+++ b/fs/xfs/linux-2.6/xfs_vfs.h
@@ -65,7 +65,6 @@ typedef struct vfs {
 	spinlock_t		vfs_sync_lock;	/* work item list lock */
 	int 			vfs_sync_seq;	/* sync thread generation no. */
 	wait_queue_head_t	vfs_wait_single_sync_task;
-	wait_queue_head_t	vfs_wait_sync_task;
 } vfs_t;
 
 #define vfs_fbhv		vfs_bh.bh_first	/* 1st on vfs behavior chain */
@@ -96,7 +95,6 @@ typedef enum {
 #define VFS_RDONLY		0x0001	/* read-only vfs */
 #define VFS_GRPID		0x0002	/* group-ID assigned from directory */
 #define VFS_DMI			0x0004	/* filesystem has the DMI enabled */
-#define VFS_UMOUNT		0x0008	/* unmount in progress */
 #define VFS_END			0x0008	/* max flag */
 
 #define SYNC_ATTR		0x0001	/* sync attributes */
diff --git a/fs/xfs/linux-2.6/xfs_vnode.c b/fs/xfs/linux-2.6/xfs_vnode.c
index 250cad5..268f45b 100644
--- a/fs/xfs/linux-2.6/xfs_vnode.c
+++ b/fs/xfs/linux-2.6/xfs_vnode.c
@@ -42,93 +42,33 @@ DEFINE_SPINLOCK(vnumber_lock);
  */
 #define NVSYNC                  37
 #define vptosync(v)             (&vsync[((unsigned long)v) % NVSYNC])
-sv_t vsync[NVSYNC];
-
-/*
- * Translate stat(2) file types to vnode types and vice versa.
- * Aware of numeric order of S_IFMT and vnode type values.
- */
-enum vtype iftovt_tab[] = {
-	VNON, VFIFO, VCHR, VNON, VDIR, VNON, VBLK, VNON,
-	VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VNON
-};
-
-u_short vttoif_tab[] = {
-	0, S_IFREG, S_IFDIR, S_IFBLK, S_IFCHR, S_IFLNK, S_IFIFO, 0, S_IFSOCK
-};
+STATIC wait_queue_head_t vsync[NVSYNC];
 
 
 void
 vn_init(void)
 {
-	register sv_t *svp;
-	register int i;
+	int i;
 
-	for (svp = vsync, i = 0; i < NVSYNC; i++, svp++)
-		init_sv(svp, SV_DEFAULT, "vsy", i);
+	for (i = 0; i < NVSYNC; i++)
+		init_waitqueue_head(&vsync[i]);
 }
 
-/*
- * Clean a vnode of filesystem-specific data and prepare it for reuse.
- */
-STATIC int
-vn_reclaim(
+void
+vn_iowait(
 	struct vnode	*vp)
 {
-	int		error;
+	wait_queue_head_t *wq = vptosync(vp);
 
-	XFS_STATS_INC(vn_reclaim);
-	vn_trace_entry(vp, "vn_reclaim", (inst_t *)__return_address);
-
-	/*
-	 * Only make the VOP_RECLAIM call if there are behaviors
-	 * to call.
-	 */
-	if (vp->v_fbhv) {
-		VOP_RECLAIM(vp, error);
-		if (error)
-			return -error;
-	}
-	ASSERT(vp->v_fbhv == NULL);
-
-	VN_LOCK(vp);
-	vp->v_flag &= (VRECLM|VWAIT);
-	VN_UNLOCK(vp, 0);
-
-	vp->v_type = VNON;
-	vp->v_fbhv = NULL;
-
-#ifdef XFS_VNODE_TRACE
-	ktrace_free(vp->v_trace);
-	vp->v_trace = NULL;
-#endif
-
-	return 0;
-}
-
-STATIC void
-vn_wakeup(
-	struct vnode	*vp)
-{
-	VN_LOCK(vp);
-	if (vp->v_flag & VWAIT)
-		sv_broadcast(vptosync(vp));
-	vp->v_flag &= ~(VRECLM|VWAIT|VMODIFIED);
-	VN_UNLOCK(vp, 0);
+	wait_event(*wq, (atomic_read(&vp->v_iocount) == 0));
 }
 
-int
-vn_wait(
+void
+vn_iowake(
 	struct vnode	*vp)
 {
-	VN_LOCK(vp);
-	if (vp->v_flag & (VINACT | VRECLM)) {
-		vp->v_flag |= VWAIT;
-		sv_wait(vptosync(vp), PINOD, &vp->v_lock, 0);
-		return 1;
-	}
-	VN_UNLOCK(vp, 0);
-	return 0;
+	if (atomic_dec_and_test(&vp->v_iocount))
+		wake_up(vptosync(vp));
 }
 
 struct vnode *
@@ -154,6 +94,8 @@ vn_initialize(
 	/* Initialize the first behavior and the behavior chain head. */
 	vn_bhv_head_init(VN_BHV_HEAD(vp), "vnode");
 
+	atomic_set(&vp->v_iocount, 0);
+
 #ifdef	XFS_VNODE_TRACE
 	vp->v_trace = ktrace_alloc(VNODE_TRACE_SIZE, KM_SLEEP);
 #endif	/* XFS_VNODE_TRACE */
@@ -163,30 +105,6 @@ vn_initialize(
 }
 
 /*
- * Get a reference on a vnode.
- */
-vnode_t *
-vn_get(
-	struct vnode	*vp,
-	vmap_t		*vmap)
-{
-	struct inode	*inode;
-
-	XFS_STATS_INC(vn_get);
-	inode = LINVFS_GET_IP(vp);
-	if (inode->i_state & I_FREEING)
-		return NULL;
-
-	inode = ilookup(vmap->v_vfsp->vfs_super, vmap->v_ino);
-	if (!inode)	/* Inode not present */
-		return NULL;
-
-	vn_trace_exit(vp, "vn_get", (inst_t *)__return_address);
-
-	return vp;
-}
-
-/*
  * Revalidate the Linux inode from the vattr.
  * Note: i_size _not_ updated; we must hold the inode
  * semaphore when doing that - callers responsibility.
@@ -198,7 +116,7 @@ vn_revalidate_core(
 {
 	struct inode	*inode = LINVFS_GET_IP(vp);
 
-	inode->i_mode	    = VTTOIF(vap->va_type) | vap->va_mode;
+	inode->i_mode	    = vap->va_mode;
 	inode->i_nlink	    = vap->va_nlink;
 	inode->i_uid	    = vap->va_uid;
 	inode->i_gid	    = vap->va_gid;
@@ -247,71 +165,6 @@ vn_revalidate(
 }
 
 /*
- * purge a vnode from the cache
- * At this point the vnode is guaranteed to have no references (vn_count == 0)
- * The caller has to make sure that there are no ways someone could
- * get a handle (via vn_get) on the vnode (usually done via a mount/vfs lock).
- */
-void
-vn_purge(
-	struct vnode	*vp,
-	vmap_t		*vmap)
-{
-	vn_trace_entry(vp, "vn_purge", (inst_t *)__return_address);
-
-again:
-	/*
-	 * Check whether vp has already been reclaimed since our caller
-	 * sampled its version while holding a filesystem cache lock that
-	 * its VOP_RECLAIM function acquires.
-	 */
-	VN_LOCK(vp);
-	if (vp->v_number != vmap->v_number) {
-		VN_UNLOCK(vp, 0);
-		return;
-	}
-
-	/*
-	 * If vp is being reclaimed or inactivated, wait until it is inert,
-	 * then proceed.  Can't assume that vnode is actually reclaimed
-	 * just because the reclaimed flag is asserted -- a vn_alloc
-	 * reclaim can fail.
-	 */
-	if (vp->v_flag & (VINACT | VRECLM)) {
-		ASSERT(vn_count(vp) == 0);
-		vp->v_flag |= VWAIT;
-		sv_wait(vptosync(vp), PINOD, &vp->v_lock, 0);
-		goto again;
-	}
-
-	/*
-	 * Another process could have raced in and gotten this vnode...
-	 */
-	if (vn_count(vp) > 0) {
-		VN_UNLOCK(vp, 0);
-		return;
-	}
-
-	XFS_STATS_DEC(vn_active);
-	vp->v_flag |= VRECLM;
-	VN_UNLOCK(vp, 0);
-
-	/*
-	 * Call VOP_RECLAIM and clean vp. The FSYNC_INVAL flag tells
-	 * vp's filesystem to flush and invalidate all cached resources.
-	 * When vn_reclaim returns, vp should have no private data,
-	 * either in a system cache or attached to v_data.
-	 */
-	if (vn_reclaim(vp) != 0)
-		panic("vn_purge: cannot reclaim");
-
-	/*
-	 * Wakeup anyone waiting for vp to be reclaimed.
-	 */
-	vn_wakeup(vp);
-}
-
-/*
  * Add a reference to a referenced vnode.
  */
 struct vnode *
@@ -330,80 +183,6 @@ vn_hold(
 	return vp;
 }
 
-/*
- *  Call VOP_INACTIVE on last reference.
- */
-void
-vn_rele(
-	struct vnode	*vp)
-{
-	int		vcnt;
-	int		cache;
-
-	XFS_STATS_INC(vn_rele);
-
-	VN_LOCK(vp);
-
-	vn_trace_entry(vp, "vn_rele", (inst_t *)__return_address);
-	vcnt = vn_count(vp);
-
-	/*
-	 * Since we always get called from put_inode we know
-	 * that i_count won't be decremented after we
-	 * return.
-	 */
-	if (!vcnt) {
-		/*
-		 * As soon as we turn this on, noone can find us in vn_get
-		 * until we turn off VINACT or VRECLM
-		 */
-		vp->v_flag |= VINACT;
-		VN_UNLOCK(vp, 0);
-
-		/*
-		 * Do not make the VOP_INACTIVE call if there
-		 * are no behaviors attached to the vnode to call.
-		 */
-		if (vp->v_fbhv)
-			VOP_INACTIVE(vp, NULL, cache);
-
-		VN_LOCK(vp);
-		if (vp->v_flag & VWAIT)
-			sv_broadcast(vptosync(vp));
-
-		vp->v_flag &= ~(VINACT|VWAIT|VRECLM|VMODIFIED);
-	}
-
-	VN_UNLOCK(vp, 0);
-
-	vn_trace_exit(vp, "vn_rele", (inst_t *)__return_address);
-}
-
-/*
- * Finish the removal of a vnode.
- */
-void
-vn_remove(
-	struct vnode	*vp)
-{
-	vmap_t		vmap;
-
-	/* Make sure we don't do this to the same vnode twice */
-	if (!(vp->v_fbhv))
-		return;
-
-	XFS_STATS_INC(vn_remove);
-	vn_trace_exit(vp, "vn_remove", (inst_t *)__return_address);
-
-	/*
-	 * After the following purge the vnode
-	 * will no longer exist.
-	 */
-	VMAP(vp, vmap);
-	vn_purge(vp, &vmap);
-}
-
-
 #ifdef	XFS_VNODE_TRACE
 
 #define KTRACE_ENTER(vp, vk, s, line, ra)			\
diff --git a/fs/xfs/linux-2.6/xfs_vnode.h b/fs/xfs/linux-2.6/xfs_vnode.h
index a6e57c6..35f306c 100644
--- a/fs/xfs/linux-2.6/xfs_vnode.h
+++ b/fs/xfs/linux-2.6/xfs_vnode.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2000-2003 Silicon Graphics, Inc.  All Rights Reserved.
+ * Copyright (c) 2000-2005 Silicon Graphics, Inc.  All Rights Reserved.
  *
  * This program is free software; you can redistribute it and/or modify it
  * under the terms of version 2 of the GNU General Public License as
@@ -65,10 +65,6 @@ struct vattr;
 struct xfs_iomap;
 struct attrlist_cursor_kern;
 
-/*
- * Vnode types.  VNON means no type.
- */
-enum vtype	{ VNON, VREG, VDIR, VBLK, VCHR, VLNK, VFIFO, VBAD, VSOCK };
 
 typedef xfs_ino_t vnumber_t;
 typedef struct dentry vname_t;
@@ -77,15 +73,14 @@ typedef bhv_head_t vn_bhv_head_t;
 /*
  * MP locking protocols:
  *	v_flag, v_vfsp				VN_LOCK/VN_UNLOCK
- *	v_type					read-only or fs-dependent
  */
 typedef struct vnode {
 	__u32		v_flag;			/* vnode flags (see below) */
-	enum vtype	v_type;			/* vnode type */
 	struct vfs	*v_vfsp;		/* ptr to containing VFS */
 	vnumber_t	v_number;		/* in-core vnode number */
 	vn_bhv_head_t	v_bh;			/* behavior head */
 	spinlock_t	v_lock;			/* VN_LOCK/VN_UNLOCK */
+	atomic_t	v_iocount;		/* outstanding I/O count */
 #ifdef XFS_VNODE_TRACE
 	struct ktrace	*v_trace;		/* trace header structure    */
 #endif
@@ -93,6 +88,12 @@ typedef struct vnode {
 	/* inode MUST be last */
 } vnode_t;
 
+#define VN_ISLNK(vp)	S_ISLNK((vp)->v_inode.i_mode)
+#define VN_ISREG(vp)	S_ISREG((vp)->v_inode.i_mode)
+#define VN_ISDIR(vp)	S_ISDIR((vp)->v_inode.i_mode)
+#define VN_ISCHR(vp)	S_ISCHR((vp)->v_inode.i_mode)
+#define VN_ISBLK(vp)	S_ISBLK((vp)->v_inode.i_mode)
+
 #define v_fbhv			v_bh.bh_first	       /* first behavior */
 #define v_fops			v_bh.bh_first->bd_ops  /* first behavior ops */
 
@@ -133,22 +134,8 @@ typedef enum {
 #define LINVFS_GET_IP(vp)	(&(vp)->v_inode)
 
 /*
- * Convert between vnode types and inode formats (since POSIX.1
- * defines mode word of stat structure in terms of inode formats).
- */
-extern enum vtype	iftovt_tab[];
-extern u_short		vttoif_tab[];
-#define IFTOVT(mode)	(iftovt_tab[((mode) & S_IFMT) >> 12])
-#define VTTOIF(indx)	(vttoif_tab[(int)(indx)])
-#define MAKEIMODE(indx, mode)	(int)(VTTOIF(indx) | (mode))
-
-
-/*
  * Vnode flags.
  */
-#define VINACT		       0x1	/* vnode is being inactivated	*/
-#define VRECLM		       0x2	/* vnode is being reclaimed	*/
-#define VWAIT		       0x4	/* waiting for VINACT/VRECLM to end */
 #define VMODIFIED	       0x8	/* XFS inode state possibly differs */
 					/* to the Linux inode state.	*/
 
@@ -408,7 +395,6 @@ typedef struct vnodeops {
  */
 typedef struct vattr {
 	int		va_mask;	/* bit-mask of attributes present */
-	enum vtype	va_type;	/* vnode type (for create) */
 	mode_t		va_mode;	/* file access mode and type */
 	xfs_nlink_t	va_nlink;	/* number of references to file */
 	uid_t		va_uid;		/* owner user id */
@@ -498,27 +484,12 @@ typedef struct vattr {
  * Check whether mandatory file locking is enabled.
  */
 #define MANDLOCK(vp, mode)	\
-	((vp)->v_type == VREG && ((mode) & (VSGID|(VEXEC>>3))) == VSGID)
+	(VN_ISREG(vp) && ((mode) & (VSGID|(VEXEC>>3))) == VSGID)
 
 extern void	vn_init(void);
-extern int	vn_wait(struct vnode *);
 extern vnode_t	*vn_initialize(struct inode *);
 
 /*
- * Acquiring and invalidating vnodes:
- *
- *	if (vn_get(vp, version, 0))
- *		...;
- *	vn_purge(vp, version);
- *
- * vn_get and vn_purge must be called with vmap_t arguments, sampled
- * while a lock that the vnode's VOP_RECLAIM function acquires is
- * held, to ensure that the vnode sampled with the lock held isn't
- * recycled (VOP_RECLAIMed) or deallocated between the release of the lock
- * and the subsequent vn_get or vn_purge.
- */
-
-/*
  * vnode_map structures _must_ match vn_epoch and vnode structure sizes.
  */
 typedef struct vnode_map {
@@ -531,11 +502,11 @@ typedef struct vnode_map {
 			 (vmap).v_number = (vp)->v_number,	\
 			 (vmap).v_ino	 = (vp)->v_inode.i_ino; }
 
-extern void	vn_purge(struct vnode *, vmap_t *);
-extern vnode_t	*vn_get(struct vnode *, vmap_t *);
 extern int	vn_revalidate(struct vnode *);
 extern void	vn_revalidate_core(struct vnode *, vattr_t *);
-extern void	vn_remove(struct vnode *);
+
+extern void	vn_iowait(struct vnode *vp);
+extern void	vn_iowake(struct vnode *vp);
 
 static inline int vn_count(struct vnode *vp)
 {
@@ -546,7 +517,6 @@ static inline int vn_count(struct vnode *vp)
  * Vnode reference counting functions (and macros for compatibility).
  */
 extern vnode_t	*vn_hold(struct vnode *);
-extern void	vn_rele(struct vnode *);
 
 #if defined(XFS_VNODE_TRACE)
 #define VN_HOLD(vp)		\
@@ -560,6 +530,12 @@ extern void	vn_rele(struct vnode *);
 #define VN_RELE(vp)		(iput(LINVFS_GET_IP(vp)))
 #endif
 
+static inline struct vnode *vn_grab(struct vnode *vp)
+{
+	struct inode *inode = igrab(LINVFS_GET_IP(vp));
+	return inode ? LINVFS_GET_VP(inode) : NULL;
+}
+
 /*
  * Vname handling macros.
  */