XFS: Implement readv/writev

Modid: 2.5.x-xfs:slinx:128366a

fs/xfs/linux/xfs_file.c

@@ -40,70 +40,76 @@ static struct vm_operations_struct linvfs_file_vm_ops;
 
 
 STATIC ssize_t
-linvfs_read(
+linvfs_readv(
 	struct file		*filp,
-	char		*buf,
-	size_t		size,
-	loff_t		*offset)
+	const struct iovec	*iovp,
+	unsigned long		nr_segs,
+	loff_t			*ppos)
 {
-	vnode_t		*vp;
+	vnode_t			*vp = LINVFS_GET_VP(filp->f_dentry->d_inode);
 	int			error;
 
-	vp = LINVFS_GET_VP(filp->f_dentry->d_inode);
-	ASSERT(vp);
-
-	VOP_READ(vp, filp, buf, size, offset, NULL, error);
+	VOP_READ(vp, filp, iovp, nr_segs, ppos, NULL, error);
 
-	return(error);
+	return error;
 }
 
 
 STATIC ssize_t
-linvfs_write(
-	struct file	*file,
-	const char	*buf,
-	size_t		count,
+linvfs_writev(
+	struct file		*filp,
+	const struct iovec	*iovp,
+	unsigned long		nr_segs,
 	loff_t			*ppos)
 {
-	struct inode	*inode = file->f_dentry->d_inode;
-	loff_t		pos;
-	vnode_t		*vp;
-	int		err;	/* Use negative errors in this f'n */
-
-	if ((ssize_t) count < 0)
-		return -EINVAL;
-
-	if (!access_ok(VERIFY_READ, buf, count))
-		return -EFAULT;
-
-	pos = *ppos;
-	err = -EINVAL;
-	if (pos < 0)
-		goto out;
+	struct inode		*inode = filp->f_dentry->d_inode;
+	vnode_t			*vp = LINVFS_GET_VP(inode);
+	int			error = filp->f_error;
 
-	err = file->f_error;
-	if (err) {
-		file->f_error = 0;
-		goto out;
+	if (unlikely(error)) {
+		filp->f_error = 0;
+		return error;
 	}
 
-	vp = LINVFS_GET_VP(inode);
-	ASSERT(vp);
-
-	/* We allow multiple direct writers in, there is no
+	/*
+	 * We allow multiple direct writers in, there is no
 	 * potential call to vmtruncate in that path.
 	 */
-	if (!(file->f_flags & O_DIRECT))
+	if (filp->f_flags & O_DIRECT) {
+		VOP_WRITE(vp, filp, iovp, nr_segs, ppos, NULL, error);
+	} else {
 		down(&inode->i_sem);
+		VOP_WRITE(vp, filp, iovp, nr_segs, ppos, NULL, error);
+		up(&inode->i_sem);
+	}
 
-	VOP_WRITE(vp, file, buf, count, &pos, NULL, err);
-	*ppos = pos;
+	return error;
+}
 
-	if (!(file->f_flags & O_DIRECT))
-		up(&inode->i_sem);
-out:
 
-	return(err);
+STATIC ssize_t
+linvfs_read(
+	struct file		*filp,
+	char			*buf,
+	size_t			count,
+	loff_t			*ppos)
+{
+	struct iovec		iov = {buf, count};
+
+	return linvfs_readv(filp, &iov, 1, ppos);
+}
+
+
+STATIC ssize_t
+linvfs_write(
+	struct file		*file,
+	const char		*buf,
+	size_t			count,
+	loff_t			*ppos)
+{
+	struct iovec		iov = {(void *)buf, count};
+
+	return linvfs_writev(file, &iov, 1, ppos);
 }
 
 
@@ -312,6 +318,8 @@ struct file_operations linvfs_file_operations = {
 	.llseek		= generic_file_llseek,
 	.read		= linvfs_read,
 	.write		= linvfs_write,
+	.readv		= linvfs_readv,
+	.writev		= linvfs_writev,
 	.ioctl		= linvfs_ioctl,
 	.mmap		= linvfs_file_mmap,
 	.open		= linvfs_open,

fs/xfs/linux/xfs_lrw.c

@@ -125,35 +125,63 @@ xfs_iozero(
 ssize_t			/* bytes read, or (-)  error */
 xfs_read(
 	bhv_desc_t		*bdp,
-	struct file	*file,
-	char		*buf,
-	size_t		size,
-	loff_t		*offset,
+	struct file		*filp,
+	const struct iovec	*iovp,
+	unsigned long		segs,
+	loff_t			*offp,
 	cred_t			*credp)
 {
+	size_t			size = 0;
 	ssize_t			ret;
 	xfs_fsize_t		n;
 	xfs_inode_t		*ip;
 	xfs_mount_t		*mp;
+	unsigned long		seg;
+	int			direct = filp->f_flags & O_DIRECT;
 
 	ip = XFS_BHVTOI(bdp);
 	mp = ip->i_mount;
 
 	XFS_STATS_INC(xfsstats.xs_read_calls);
 
-	if (file->f_flags & O_DIRECT) {
-		if (((__psint_t)buf & BBMASK) ||
-		    (*offset & mp->m_blockmask) ||
+	/* START copy & waste from filemap.c */
+	for (seg = 0; seg < segs; seg++) {
+		const struct iovec *iv = &iovp[seg];
+
+		/*
+		 * If any segment has a negative length, or the cumulative
+		 * length ever wraps negative then return -EINVAL.
+		 */
+		size += iv->iov_len;
+		if (unlikely((ssize_t)(size|iv->iov_len) < 0))
+			return XFS_ERROR(-EINVAL);
+		if (direct) {	/* XFS specific check */
+			if ((__psint_t)iv->iov_base & BBMASK) {
+				if (*offp == ip->i_d.di_size)
+					return 0;
+				return XFS_ERROR(-EINVAL);
+			}
+		}
+		if (access_ok(VERIFY_WRITE, iv->iov_base, iv->iov_len))
+			continue;
+		if (seg == 0)
+			return XFS_ERROR(-EFAULT);
+		segs = seg;
+		break;
+	}
+	/* END copy & waste from filemap.c */
+
+	if (direct) {
+		if ((*offp & mp->m_blockmask) ||
 		    (size & mp->m_blockmask)) {
-			if (*offset == ip->i_d.di_size) {
+			if (*offp == ip->i_d.di_size) {
 				return (0);
 			}
 			return -XFS_ERROR(EINVAL);
 		}
 	}
 
-
-	n = XFS_MAX_FILE_OFFSET - *offset;
+	n = XFS_MAX_FILE_OFFSET - *offp;
 	if ((n <= 0) || (size == 0))
 		return 0;
 
@@ -167,26 +195,24 @@ xfs_read(
 	xfs_ilock(ip, XFS_IOLOCK_SHARED);
 
 	if (DM_EVENT_ENABLED(BHV_TO_VNODE(bdp)->v_vfsp, ip, DM_EVENT_READ) &&
-	    !(file->f_mode & FINVIS)) {
+	    !(filp->f_mode & FINVIS)) {
 		int error;
 		vrwlock_t locktype = VRWLOCK_READ;
 
-		error = xfs_dm_send_data_event(DM_EVENT_READ, bdp,
-					     *offset, size,
-					     FILP_DELAY_FLAG(file),
-					     &locktype);
+		error = xfs_dm_send_data_event(DM_EVENT_READ, bdp, *offp,
+				size, FILP_DELAY_FLAG(filp), &locktype);
 		if (error) {
 			xfs_iunlock(ip, XFS_IOLOCK_SHARED);
 			return -error;
 		}
 	}
 
-	ret = generic_file_read(file, buf, size, offset);
+	ret = generic_file_readv(filp, iovp, segs, offp);
 	xfs_iunlock(ip, XFS_IOLOCK_SHARED);
 
 	XFS_STATS_ADD(xfsstats.xs_read_bytes, ret);
 
-	if (!(file->f_mode & FINVIS))
+	if (!(filp->f_mode & FINVIS))
 		xfs_ichgtime(ip, XFS_ICHGTIME_ACC);
 
 	return ret;
@@ -420,11 +446,12 @@ ssize_t				/* bytes written, or (-) error */
 xfs_write(
 	bhv_desc_t		*bdp,
 	struct file		*file,
-	const char	*buf,
-	size_t		size,
+	const struct iovec	*iovp,
+	unsigned long		segs,
 	loff_t			*offset,
 	cred_t			*credp)
 {
+	size_t			size = 0;
 	xfs_inode_t		*xip;
 	xfs_mount_t		*mp;
 	ssize_t			ret;
@@ -433,7 +460,7 @@ xfs_write(
 	xfs_fsize_t		n, limit = XFS_MAX_FILE_OFFSET;
 	xfs_iocore_t		*io;
 	vnode_t			*vp;
-	struct iovec	iov;
+	unsigned long		seg;
 	int			iolock;
 	int			direct = file->f_flags & O_DIRECT;
 	int			eventsent = 0;
@@ -444,6 +471,30 @@ xfs_write(
 	vp = BHV_TO_VNODE(bdp);
 	xip = XFS_BHVTOI(bdp);
 
+	/* START copy & waste from filemap.c */
+	for (seg = 0; seg < segs; seg++) {
+		const struct iovec *iv = &iovp[seg];
+
+		/*
+		 * If any segment has a negative length, or the cumulative
+		 * length ever wraps negative then return -EINVAL.
+		 */
+		size += iv->iov_len;
+		if (unlikely((ssize_t)(size|iv->iov_len) < 0))
+			return XFS_ERROR(-EINVAL);
+		if (direct) {	/* XFS specific check */
+			if ((__psint_t)iv->iov_base & BBMASK)
+				return XFS_ERROR(-EINVAL);
+		}
+		if (access_ok(VERIFY_READ, iv->iov_base, iv->iov_len))
+			continue;
+		if (seg == 0)
+			return XFS_ERROR(-EFAULT);
+		segs = seg;
+		break;
+	}
+	/* END copy & waste from filemap.c */
+
 	if (size == 0)
 		return 0;
 
@@ -457,8 +508,7 @@ xfs_write(
 	}
 
 	if (direct) {
-		if (((__psint_t)buf & BBMASK) ||
-		    (*offset & mp->m_blockmask) ||
+		if ((*offset & mp->m_blockmask) ||
 		    (size & mp->m_blockmask)) {
 			return XFS_ERROR(-EINVAL);
 		}
@@ -481,6 +531,7 @@ xfs_write(
 		xfs_iunlock(xip, XFS_ILOCK_EXCL|iolock);
 		return -EFBIG;
 	}
+
 	if (n < size)
 		size = n;
 
@@ -572,10 +623,7 @@ xfs_write(
 		xfs_inval_cached_pages(vp, &xip->i_iocore, *offset, 1, 1);
 	}
 
-	iov.iov_base = (void *)buf;
-	iov.iov_len = size;
-
-	ret = generic_file_write_nolock(file, &iov, 1, offset);
+	ret = generic_file_write_nolock(file, iovp, segs, offset);
 
 	if ((ret == -ENOSPC) &&
 	    DM_EVENT_ENABLED(vp->v_vfsp, xip, DM_EVENT_NOSPACE) &&

fs/xfs/linux/xfs_lrw.h

@@ -48,18 +48,18 @@ extern int xfs_zero_eof (vnode_t *, struct xfs_iocore *, xfs_off_t,
 				xfs_fsize_t, xfs_fsize_t, struct pm *);
 extern ssize_t xfs_read (
 	struct bhv_desc		*bdp,
-	struct file		*file,
-	char			*buf,
-	size_t			size,
-	loff_t			*offset,
+	struct file		*filp,
+	const struct iovec	*iovp,
+	unsigned long		segs,
+	loff_t			*offp,
 	struct cred		*credp);
 
 extern ssize_t xfs_write (
 	struct bhv_desc		*bdp,
-	struct file		*file,
-	const char		*buf,
-	size_t			size,
-	loff_t			*offset,
+	struct file		*filp,
+	const struct iovec	*iovp,
+	unsigned long		segs,
+	loff_t			*offp,
 	struct cred		*credp);
 
 extern int xfs_recover_read_only (xlog_t *);

fs/xfs/linux/xfs_vnode.h

@@ -175,9 +175,11 @@ struct page_buf_bmap_s;
 struct attrlist_cursor_kern;
 
 typedef int	(*vop_open_t)(bhv_desc_t *, struct cred *);
-typedef ssize_t (*vop_read_t)(bhv_desc_t *, struct file *, char *, size_t,
+typedef ssize_t (*vop_read_t)(bhv_desc_t *, struct file *,
+				const struct iovec *, unsigned long,
 				loff_t *, struct cred *);
-typedef ssize_t (*vop_write_t)(bhv_desc_t *, struct file *, const char *, size_t,
+typedef ssize_t (*vop_write_t)(bhv_desc_t *, struct file *,
+				const struct iovec *, unsigned long,
 				loff_t *, struct cred *);
 typedef int	(*vop_ioctl_t)(bhv_desc_t *, struct inode *, struct file *, unsigned int, unsigned long);
 typedef int	(*vop_getattr_t)(bhv_desc_t *, struct vattr *, int,
@@ -272,16 +274,16 @@ typedef struct vnodeops {
  */
 #define _VOP_(op, vp)	(*((vnodeops_t *)(vp)->v_fops)->op)
 
-#define VOP_READ(vp,file,buf,size,offset,cr,rv)				\
+#define VOP_READ(vp,file,iov,segs,offset,cr,rv)				\
 {									\
 	VN_BHV_READ_LOCK(&(vp)->v_bh);					\
-	rv = _VOP_(vop_read, vp)((vp)->v_fbhv,file,buf,size,offset,cr); \
+	rv = _VOP_(vop_read, vp)((vp)->v_fbhv,file,iov,segs,offset,cr); \
 	VN_BHV_READ_UNLOCK(&(vp)->v_bh);				\
 }
-#define VOP_WRITE(vp,file,buf,size,offset,cr,rv)			\
+#define VOP_WRITE(vp,file,iov,segs,offset,cr,rv)			\
 {									\
 	VN_BHV_READ_LOCK(&(vp)->v_bh);					\
-	rv = _VOP_(vop_write, vp)((vp)->v_fbhv,file,buf,size,offset,cr);\
+	rv = _VOP_(vop_write, vp)((vp)->v_fbhv,file,iov,segs,offset,cr);\
 	VN_BHV_READ_UNLOCK(&(vp)->v_bh);				\
 }
 #define VOP_BMAP(vp,of,sz,rw,cr,b,n,rv)					\