[XFS] Start writeout earlier (on last close) in the case where we have a

truncate down followed by delayed allocation (buffered writes) - worst
case scenario for the notorious NULL files problem.  This reduces the
window where we are exposed to that problem significantly.

SGI-PV: 917976
SGI-Modid: xfs-linux-melb:xfs-kern:26100a

Signed-off-by: Nathan Scott <nathans@sgi.com>

5 files changed, 114 insertions, 94 deletions

diff --git a/fs/xfs/linux-2.6/xfs_aops.c b/fs/xfs/linux-2.6/xfs_aops.c
index 5835e699a7fc..c0a904316854 100644
--- a/fs/xfs/linux-2.6/xfs_aops.c
+++ b/fs/xfs/linux-2.6/xfs_aops.c
@@ -1157,6 +1157,18 @@ out_unlock:
 	return error;
 }
 
+STATIC int
+xfs_vm_writepages(
+	struct address_space	*mapping,
+	struct writeback_control *wbc)
+{
+	struct vnode		*vp = vn_from_inode(mapping->host);
+
+	if (VN_TRUNC(vp))
+		VUNTRUNCATE(vp);
+	return generic_writepages(mapping, wbc);
+}
+
 /*
  * Called to move a page into cleanable state - and from there
  * to be released. Possibly the page is already clean. We always
@@ -1451,6 +1463,7 @@ struct address_space_operations xfs_address_space_operations = {
 	.readpage		= xfs_vm_readpage,
 	.readpages		= xfs_vm_readpages,
 	.writepage		= xfs_vm_writepage,
+	.writepages		= xfs_vm_writepages,
 	.sync_page		= block_sync_page,
 	.releasepage		= xfs_vm_releasepage,
 	.invalidatepage		= xfs_vm_invalidatepage,
diff --git a/fs/xfs/linux-2.6/xfs_file.c b/fs/xfs/linux-2.6/xfs_file.c
index 7c9f7598807f..97615cc74ef5 100644
--- a/fs/xfs/linux-2.6/xfs_file.c
+++ b/fs/xfs/linux-2.6/xfs_file.c
@@ -324,6 +324,17 @@ xfs_file_open(
 }
 
 STATIC int
+xfs_file_close(
+	struct file	*filp)
+{
+	vnode_t		*vp = vn_from_inode(filp->f_dentry->d_inode);
+	int		error;
+
+	VOP_CLOSE(vp, 0, file_count(filp) > 1 ? L_FALSE : L_TRUE, NULL, error);
+	return -error;
+}
+
+STATIC int
 xfs_file_release(
 	struct inode	*inode,
 	struct file	*filp)
@@ -349,6 +360,8 @@ xfs_file_fsync(
 
 	if (datasync)
 		flags |= FSYNC_DATA;
+	if (VN_TRUNC(vp))
+		VUNTRUNCATE(vp);
 	VOP_FSYNC(vp, flags, NULL, (xfs_off_t)0, (xfs_off_t)-1, error);
 	return -error;
 }
@@ -578,6 +591,7 @@ const struct file_operations xfs_file_operations = {
 #endif
 	.mmap		= xfs_file_mmap,
 	.open		= xfs_file_open,
+	.flush		= xfs_file_close,
 	.release	= xfs_file_release,
 	.fsync		= xfs_file_fsync,
 #ifdef HAVE_FOP_OPEN_EXEC
@@ -602,6 +616,7 @@ const struct file_operations xfs_invis_file_operations = {
 #endif
 	.mmap		= xfs_file_mmap,
 	.open		= xfs_file_open,
+	.flush		= xfs_file_close,
 	.release	= xfs_file_release,
 	.fsync		= xfs_file_fsync,
 };
diff --git a/fs/xfs/linux-2.6/xfs_fs_subr.c b/fs/xfs/linux-2.6/xfs_fs_subr.c
index 575f2a790f31..f0c56daf4d6d 100644
--- a/fs/xfs/linux-2.6/xfs_fs_subr.c
+++ b/fs/xfs/linux-2.6/xfs_fs_subr.c
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2000-2002,2005 Silicon Graphics, Inc.
+ * Copyright (c) 2000-2002,2005-2006 Silicon Graphics, Inc.
  * All Rights Reserved.
  *
  * This program is free software; you can redistribute it and/or
@@ -15,40 +15,12 @@
  * along with this program; if not, write the Free Software Foundation,
  * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
  */
-
 #include "xfs.h"
 
-/*
- * Stub for no-op vnode operations that return error status.
- */
-int
-fs_noerr(void)
-{
-	return 0;
-}
+int  fs_noerr(void) { return 0; }
+int  fs_nosys(void) { return ENOSYS; }
+void fs_noval(void) { return; }
 
-/*
- * Operation unsupported under this file system.
- */
-int
-fs_nosys(void)
-{
-	return ENOSYS;
-}
-
-/*
- * Stub for inactive, strategy, and read/write lock/unlock.  Does nothing.
- */
-/* ARGSUSED */
-void
-fs_noval(void)
-{
-}
-
-/*
- * vnode pcache layer for vnode_tosspages.
- * 'last' parameter unused but left in for IRIX compatibility
- */
 void
 fs_tosspages(
 	bhv_desc_t	*bdp,
@@ -63,11 +35,6 @@ fs_tosspages(
 		truncate_inode_pages(ip->i_mapping, first);
 }
 
-
-/*
- * vnode pcache layer for vnode_flushinval_pages.
- * 'last' parameter unused but left in for IRIX compatibility
- */
 void
 fs_flushinval_pages(
 	bhv_desc_t	*bdp,
@@ -79,16 +46,13 @@ fs_flushinval_pages(
 	struct inode	*ip = vn_to_inode(vp);
 
 	if (VN_CACHED(vp)) {
+		if (VN_TRUNC(vp))
+			VUNTRUNCATE(vp);
 		filemap_write_and_wait(ip->i_mapping);
-
 		truncate_inode_pages(ip->i_mapping, first);
 	}
 }
 
-/*
- * vnode pcache layer for vnode_flush_pages.
- * 'last' parameter unused but left in for IRIX compatibility
- */
 int
 fs_flush_pages(
 	bhv_desc_t	*bdp,
@@ -100,12 +64,13 @@ fs_flush_pages(
 	vnode_t		*vp = BHV_TO_VNODE(bdp);
 	struct inode	*ip = vn_to_inode(vp);
 
-	if (VN_CACHED(vp)) {
+	if (VN_DIRTY(vp)) {
+		if (VN_TRUNC(vp))
+			VUNTRUNCATE(vp);
 		filemap_fdatawrite(ip->i_mapping);
 		if (flags & XFS_B_ASYNC)
 			return 0;
 		filemap_fdatawait(ip->i_mapping);
 	}
-
 	return 0;
 }
diff --git a/fs/xfs/linux-2.6/xfs_vnode.h b/fs/xfs/linux-2.6/xfs_vnode.h
index a64b7db67003..569a4e7b5cc1 100644
--- a/fs/xfs/linux-2.6/xfs_vnode.h
+++ b/fs/xfs/linux-2.6/xfs_vnode.h
@@ -56,12 +56,18 @@ typedef xfs_ino_t vnumber_t;
 typedef struct dentry vname_t;
 typedef bhv_head_t vn_bhv_head_t;
 
+typedef enum vflags {
+	VMODIFIED	= 0x08,	/* XFS inode state possibly differs */
+				/* to the Linux inode state. */
+	VTRUNCATED	= 0x40,	/* truncated down so flush-on-close */
+} vflags_t;
+
 /*
  * MP locking protocols:
  *	v_flag, v_vfsp				VN_LOCK/VN_UNLOCK
  */
 typedef struct vnode {
-	__u32		v_flag;			/* vnode flags (see below) */
+	vflags_t	v_flag;			/* vnode flags (see above) */
 	struct vfs	*v_vfsp;		/* ptr to containing VFS */
 	vnumber_t	v_number;		/* in-core vnode number */
 	vn_bhv_head_t	v_bh;			/* behavior head */
@@ -126,12 +132,6 @@ static inline struct inode *vn_to_inode(struct vnode *vnode)
 }
 
 /*
- * Vnode flags.
- */
-#define VMODIFIED	       0x8	/* XFS inode state possibly differs */
-					/* to the Linux inode state.	*/
-
-/*
  * Values for the VOP_RWLOCK and VOP_RWUNLOCK flags parameter.
  */
 typedef enum vrwlock {
@@ -162,8 +162,10 @@ typedef enum vchange {
 	VCHANGE_FLAGS_IOEXCL_COUNT	= 4
 } vchange_t;
 
+typedef enum { L_FALSE, L_TRUE } lastclose_t;
 
 typedef int	(*vop_open_t)(bhv_desc_t *, struct cred *);
+typedef int	(*vop_close_t)(bhv_desc_t *, int, lastclose_t, struct cred *);
 typedef ssize_t (*vop_read_t)(bhv_desc_t *, struct kiocb *,
 				const struct iovec *, unsigned int,
 				loff_t *, int, struct cred *);
@@ -234,6 +236,7 @@ typedef int	(*vop_iflush_t)(bhv_desc_t *, int);
 typedef struct vnodeops {
 	bhv_position_t  vn_position;    /* position within behavior chain */
 	vop_open_t		vop_open;
+	vop_close_t		vop_close;
 	vop_read_t		vop_read;
 	vop_write_t		vop_write;
 	vop_sendfile_t		vop_sendfile;
@@ -278,6 +281,10 @@ typedef struct vnodeops {
  */
 #define _VOP_(op, vp)	(*((vnodeops_t *)(vp)->v_fops)->op)
 
+#define VOP_OPEN(vp, cr, rv)						\
+	rv = _VOP_(vop_open, vp)((vp)->v_fbhv, cr)
+#define VOP_CLOSE(vp, f, last, cr, rv)					\
+	rv = _VOP_(vop_close, vp)((vp)->v_fbhv, f, last, cr)
 #define VOP_READ(vp,file,iov,segs,offset,ioflags,cr,rv)			\
 	rv = _VOP_(vop_read, vp)((vp)->v_fbhv,file,iov,segs,offset,ioflags,cr)
 #define VOP_WRITE(vp,file,iov,segs,offset,ioflags,cr,rv)		\
@@ -290,8 +297,6 @@ typedef struct vnodeops {
 	rv = _VOP_(vop_splice_write, vp)((vp)->v_fbhv,f,o,pipe,cnt,fl,iofl,cr)
 #define VOP_BMAP(vp,of,sz,rw,b,n,rv)					\
 	rv = _VOP_(vop_bmap, vp)((vp)->v_fbhv,of,sz,rw,b,n)
-#define VOP_OPEN(vp, cr, rv)						\
-	rv = _VOP_(vop_open, vp)((vp)->v_fbhv, cr)
 #define VOP_GETATTR(vp, vap, f, cr, rv)					\
 	rv = _VOP_(vop_getattr, vp)((vp)->v_fbhv, vap, f, cr)
 #define	VOP_SETATTR(vp, vap, f, cr, rv)					\
@@ -556,8 +561,6 @@ static inline struct vnode *vn_grab(struct vnode *vp)
  */
 #define VN_LOCK(vp)		mutex_spinlock(&(vp)->v_lock)
 #define VN_UNLOCK(vp, s)	mutex_spinunlock(&(vp)->v_lock, s)
-#define VN_FLAGSET(vp,b)	vn_flagset(vp,b)
-#define VN_FLAGCLR(vp,b)	vn_flagclr(vp,b)
 
 static __inline__ void vn_flagset(struct vnode *vp, uint flag)
 {
@@ -566,13 +569,22 @@ static __inline__ void vn_flagset(struct vnode *vp, uint flag)
 	spin_unlock(&vp->v_lock);
 }
 
-static __inline__ void vn_flagclr(struct vnode *vp, uint flag)
+static __inline__ uint vn_flagclr(struct vnode *vp, uint flag)
 {
+	uint	cleared;
+
 	spin_lock(&vp->v_lock);
+	cleared = (vp->v_flag & flag);
 	vp->v_flag &= ~flag;
 	spin_unlock(&vp->v_lock);
+	return cleared;
 }
 
+#define VMODIFY(vp)	vn_flagset(vp, VMODIFIED)
+#define VUNMODIFY(vp)	vn_flagclr(vp, VMODIFIED)
+#define VTRUNCATE(vp)	vn_flagset(vp, VTRUNCATED)
+#define VUNTRUNCATE(vp)	vn_flagclr(vp, VTRUNCATED)
+
 /*
  * Dealing with bad inodes
  */
@@ -612,8 +624,7 @@ static inline void vn_atime_to_time_t(struct vnode *vp, time_t *tt)
 #define VN_CACHED(vp)	(vn_to_inode(vp)->i_mapping->nrpages)
 #define VN_DIRTY(vp)	mapping_tagged(vn_to_inode(vp)->i_mapping, \
 					PAGECACHE_TAG_DIRTY)
-#define VMODIFY(vp)	VN_FLAGSET(vp, VMODIFIED)
-#define VUNMODIFY(vp)	VN_FLAGCLR(vp, VMODIFIED)
+#define VN_TRUNC(vp)	((vp)->v_flag & VTRUNCATED)
 
 /*
  * Flags to VOP_SETATTR/VOP_GETATTR.
diff --git a/fs/xfs/xfs_vnodeops.c b/fs/xfs/xfs_vnodeops.c
index cb36a56392e7..35906bae92e1 100644
--- a/fs/xfs/xfs_vnodeops.c
+++ b/fs/xfs/xfs_vnodeops.c
@@ -16,8 +16,6 @@
  * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
  */
 
-#include <linux/capability.h>
-
 #include "xfs.h"
 #include "xfs_fs.h"
 #include "xfs_types.h"
@@ -58,32 +56,14 @@
 #include "xfs_log_priv.h"
 #include "xfs_mac.h"
 
-
-/*
- * The maximum pathlen is 1024 bytes. Since the minimum file system
- * blocksize is 512 bytes, we can get a max of 2 extents back from
- * bmapi.
- */
-#define SYMLINK_MAPS 2
-
-/*
- * For xfs, we check that the file isn't too big to be opened by this kernel.
- * No other open action is required for regular files.  Devices are handled
- * through the specfs file system, pipes through fifofs.  Device and
- * fifo vnodes are "wrapped" by specfs and fifofs vnodes, respectively,
- * when a new vnode is first looked up or created.
- */
 STATIC int
 xfs_open(
 	bhv_desc_t	*bdp,
 	cred_t		*credp)
 {
 	int		mode;
-	vnode_t		*vp;
-	xfs_inode_t	*ip;
-
-	vp = BHV_TO_VNODE(bdp);
-	ip = XFS_BHVTOI(bdp);
+	vnode_t		*vp = BHV_TO_VNODE(bdp);
+	xfs_inode_t	*ip = XFS_BHVTOI(bdp);
 
 	if (XFS_FORCED_SHUTDOWN(ip->i_mount))
 		return XFS_ERROR(EIO);
@@ -101,6 +81,36 @@ xfs_open(
 	return 0;
 }
 
+STATIC int
+xfs_close(
+	bhv_desc_t	*bdp,
+	int		flags,
+	lastclose_t	lastclose,
+	cred_t		*credp)
+{
+	vnode_t		*vp = BHV_TO_VNODE(bdp);
+	xfs_inode_t	*ip = XFS_BHVTOI(bdp);
+	int		error = 0;
+
+	if (XFS_FORCED_SHUTDOWN(ip->i_mount))
+		return XFS_ERROR(EIO);
+
+	if (lastclose != L_TRUE || !VN_ISREG(vp))
+		return 0;
+
+	/*
+	 * If we previously truncated this file and removed old data in
+	 * the process, we want to initiate "early" writeout on the last
+	 * close.  This is an attempt to combat the notorious NULL files
+	 * problem which is particularly noticable from a truncate down,
+	 * buffered (re-)write (delalloc), followed by a crash.  What we
+	 * are effectively doing here is significantly reducing the time
+	 * window where we'd otherwise be exposed to that problem.
+	 */
+	if (VUNTRUNCATE(vp) && VN_DIRTY(vp) && ip->i_delayed_blks > 0)
+		VOP_FLUSH_PAGES(vp, 0, -1, XFS_B_ASYNC, FI_NONE, error);
+	return error;
+}
 
 /*
  * xfs_getattr
@@ -665,9 +675,17 @@ xfs_setattr(
 					    ((ip->i_d.di_nlink != 0 ||
 					      !(mp->m_flags & XFS_MOUNT_WSYNC))
 					     ? 1 : 0));
-			if (code) {
+			if (code)
 				goto abort_return;
-			}
+			/*
+			 * Truncated "down", so we're removing references
+			 * to old data here - if we now delay flushing for
+			 * a long time, we expose ourselves unduly to the
+			 * notorious NULL files problem.  So, we mark this
+			 * vnode and flush it when the file is closed, and
+			 * do not wait the usual (long) time for writeout.
+			 */
+			VTRUNCATE(vp);
 		}
 		/*
 		 * Have to do this even if the file's size doesn't change.
@@ -937,6 +955,13 @@ xfs_access(
 
 
 /*
+ * The maximum pathlen is 1024 bytes. Since the minimum file system
+ * blocksize is 512 bytes, we can get a max of 2 extents back from
+ * bmapi.
+ */
+#define SYMLINK_MAPS 2
+
+/*
  * xfs_readlink
  *
  */
@@ -1470,9 +1495,6 @@ xfs_inactive_symlink_local(
 	return 0;
 }
 
-/*
- *
- */
 STATIC int
 xfs_inactive_attrs(
 	xfs_inode_t	*ip,
@@ -1531,10 +1553,10 @@ xfs_release(
 
 	vp = BHV_TO_VNODE(bdp);
 	ip = XFS_BHVTOI(bdp);
+	mp = ip->i_mount;
 
-	if (!VN_ISREG(vp) || (ip->i_d.di_mode == 0)) {
+	if (!VN_ISREG(vp) || (ip->i_d.di_mode == 0))
 		return 0;
-	}
 
 	/* If this is a read-only mount, don't do this (would generate I/O) */
 	if (vp->v_vfsp->vfs_flag & VFS_RDONLY)
@@ -1546,8 +1568,6 @@ xfs_release(
 		return 0;
 #endif
 
-	mp = ip->i_mount;
-
 	if (ip->i_d.di_nlink != 0) {
 		if ((((ip->i_d.di_mode & S_IFMT) == S_IFREG) &&
 		     ((ip->i_d.di_size > 0) || (VN_CACHED(vp) > 0 ||
@@ -3745,7 +3765,6 @@ xfs_inode_flush(
 	return error;
 }
 
-
 int
 xfs_set_dmattrs (
 	bhv_desc_t	*bdp,
@@ -3786,10 +3805,6 @@ xfs_set_dmattrs (
 	return error;
 }
 
-
-/*
- * xfs_reclaim
- */
 STATIC int
 xfs_reclaim(
 	bhv_desc_t	*bdp)
@@ -4645,6 +4660,7 @@ xfs_change_file_space(
 vnodeops_t xfs_vnodeops = {
 	BHV_IDENTITY_INIT(VN_BHV_XFS,VNODE_POSITION_XFS),
 	.vop_open		= xfs_open,
+	.vop_close		= xfs_close,
 	.vop_read		= xfs_read,
 #ifdef HAVE_SENDFILE
 	.vop_sendfile		= xfs_sendfile,