[PATCH v2 1/2] vfs: pass 'struct file *' as parameter to ->check_flags() methods

[PATCH v2 1/2] vfs: pass 'struct file *' as parameter to ->check_flags() methods

[Anand Avati]

FUSE needs to selectively permit ->check_flags() on a per-file basis as it
does not support setting/unsetting O_DIRECT flag on an open file. Hence
passing 'struct file *' as a parameter to ->check_flags() is necessary to
inspect and compare file->f_flags with the new flags.

Signed-off-by: Anand Avati <avati-+FkPdpiNhgJBDgjK7y7TUQ@public.gmane.org>
---
 Documentation/filesystems/Locking |    2 +-
 Documentation/filesystems/vfs.txt |    2 +-
 fs/bad_inode.c                    |    2 +-
 fs/fcntl.c                        |    2 +-
 fs/nfs/file.c                     |    6 +++---
 include/linux/fs.h                |    2 +-
 6 files changed, 8 insertions(+), 8 deletions(-)

diff --git a/Documentation/filesystems/Locking b/Documentation/filesystems/Locking
index 57d827d..9619841 100644
--- a/Documentation/filesystems/Locking
+++ b/Documentation/filesystems/Locking
@@ -426,7 +426,7 @@ prototypes:
 			loff_t *, int);
 	unsigned long (*get_unmapped_area)(struct file *, unsigned long,
 			unsigned long, unsigned long, unsigned long);
-	int (*check_flags)(int);
+	int (*check_flags)(struct file *, int);
 	int (*flock) (struct file *, int, struct file_lock *);
 	ssize_t (*splice_write)(struct pipe_inode_info *, struct file *, loff_t *,
 			size_t, unsigned int);
diff --git a/Documentation/filesystems/vfs.txt b/Documentation/filesystems/vfs.txt
index 88b9f55..442aefb 100644
--- a/Documentation/filesystems/vfs.txt
+++ b/Documentation/filesystems/vfs.txt
@@ -764,7 +764,7 @@ struct file_operations {
 	ssize_t (*sendfile) (struct file *, loff_t *, size_t, read_actor_t, void *);
 	ssize_t (*sendpage) (struct file *, struct page *, int, size_t, loff_t *, int);
 	unsigned long (*get_unmapped_area)(struct file *, unsigned long, unsigned long, unsigned long, unsigned long);
-	int (*check_flags)(int);
+	int (*check_flags)(struct file *, int);
 	int (*flock) (struct file *, int, struct file_lock *);
 	ssize_t (*splice_write)(struct pipe_inode_info *, struct file *, size_t, unsigned int);
 	ssize_t (*splice_read)(struct file *, struct pipe_inode_info *, size_t, unsigned int);
diff --git a/fs/bad_inode.c b/fs/bad_inode.c
index bfcb18f..c7eef18 100644
--- a/fs/bad_inode.c
+++ b/fs/bad_inode.c
@@ -120,7 +120,7 @@ static unsigned long bad_file_get_unmapped_area(struct file *file,
 	return -EIO;
 }
 
-static int bad_file_check_flags(int flags)
+static int bad_file_check_flags(struct file *filp, int flags)
 {
 	return -EIO;
 }
diff --git a/fs/fcntl.c b/fs/fcntl.c
index 22764c7..1a2a6d3 100644
--- a/fs/fcntl.c
+++ b/fs/fcntl.c
@@ -174,7 +174,7 @@ static int setfl(int fd, struct file * filp, unsigned long arg)
 	}
 
 	if (filp->f_op && filp->f_op->check_flags)
-		error = filp->f_op->check_flags(arg);
+		error = filp->f_op->check_flags(filp, arg);
 	if (error)
 		return error;
 
diff --git a/fs/nfs/file.c b/fs/nfs/file.c
index 2f093ed..9f96a8b 100644
--- a/fs/nfs/file.c
+++ b/fs/nfs/file.c
@@ -56,7 +56,7 @@ static ssize_t nfs_file_write(struct kiocb *, const struct iovec *iov,
 				unsigned long nr_segs, loff_t pos);
 static int  nfs_file_flush(struct file *, fl_owner_t id);
 static int  nfs_file_fsync(struct file *, int datasync);
-static int nfs_check_flags(int flags);
+static int nfs_check_flags(struct file *, int flags);
 static int nfs_lock(struct file *filp, int cmd, struct file_lock *fl);
 static int nfs_flock(struct file *filp, int cmd, struct file_lock *fl);
 static int nfs_setlease(struct file *file, long arg, struct file_lock **fl);
@@ -105,7 +105,7 @@ const struct inode_operations nfs3_file_inode_operations = {
 # define IS_SWAPFILE(inode)	(0)
 #endif
 
-static int nfs_check_flags(int flags)
+static int nfs_check_flags(struct file *filp, int flags)
 {
 	if ((flags & (O_APPEND | O_DIRECT)) == (O_APPEND | O_DIRECT))
 		return -EINVAL;
@@ -126,7 +126,7 @@ nfs_file_open(struct inode *inode, struct file *filp)
 			filp->f_path.dentry->d_name.name);
 
 	nfs_inc_stats(inode, NFSIOS_VFSOPEN);
-	res = nfs_check_flags(filp->f_flags);
+	res = nfs_check_flags(filp, filp->f_flags);
 	if (res)
 		return res;
 
diff --git a/include/linux/fs.h b/include/linux/fs.h
index b5b9792..98ce7c7 100644
--- a/include/linux/fs.h
+++ b/include/linux/fs.h
@@ -1564,7 +1564,7 @@ struct file_operations {
 	int (*lock) (struct file *, int, struct file_lock *);
 	ssize_t (*sendpage) (struct file *, struct page *, int, size_t, loff_t *, int);
 	unsigned long (*get_unmapped_area)(struct file *, unsigned long, unsigned long, unsigned long, unsigned long);
-	int (*check_flags)(int);
+	int (*check_flags)(struct file *, int);
 	int (*flock) (struct file *, int, struct file_lock *);
 	ssize_t (*splice_write)(struct pipe_inode_info *, struct file *, loff_t *, size_t, unsigned int);
 	ssize_t (*splice_read)(struct file *, loff_t *, struct pipe_inode_info *, size_t, unsigned int);
-- 
1.7.4.4

--
To unsubscribe from this list: send the line "unsubscribe linux-nfs" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

[PATCH v2 2/2] fuse: permit O_DIRECT flag in open()

[Anand Avati]

FUSE currently disallows O_DIRECT flag in open(). It is tricky dealing with
setting/unsetting of O_DIRECT flag on an open file. There are applications
(primarily VMs and databases) which open files with O_DIRECT flag. These
applications do not work on FUSE due to this limitation.

The approach with this patch is to permit opens with O_DIRECT, but instead
disable setting/unsetting of the O_DIRECT flag no matter how the file was
opened. This limitation is for more practical than disallowing O_DIRECT
altogether.

Signed-off-by: Anand Avati <avati-+FkPdpiNhgJBDgjK7y7TUQ@public.gmane.org>
---
 fs/fuse/file.c |   36 ++++++++++++++++++++++++++++++++----
 1 files changed, 32 insertions(+), 4 deletions(-)

diff --git a/fs/fuse/file.c b/fs/fuse/file.c
index 82a6646..355c30f 100644
--- a/fs/fuse/file.c
+++ b/fs/fuse/file.c
@@ -154,6 +154,27 @@ int fuse_do_open(struct fuse_conn *fc, u64 nodeid, struct file *file,
 		return err;
 	}
 
+	if (file->f_flags & O_DIRECT) {
+		if (!(outarg.open_flags & FOPEN_DIRECT_IO)) {
+			/* filesystem process did not acknowledge the O_DIRECT
+			   flag with FOPEN_DIRECT_IO bit in ourarg. Hence
+			   we fail the open().
+
+			   Notify the filesystem process about the failed
+			   open with fuse_sync_release right here itself as
+			   file->private_data is not yet set up for release
+			   notification.
+			*/
+			fuse_sync_release(ff, file->f_flags);
+			return -EINVAL;
+		}
+
+		/* make VFS believe we don't support O_DIRECT till we
+		   implement a_ops->direct_IO
+		*/
+		file->f_flags &= ~O_DIRECT;
+	}
+
 	if (isdir)
 		outarg.open_flags &= ~FOPEN_DIRECT_IO;
 
@@ -193,10 +214,6 @@ int fuse_open_common(struct inode *inode, struct file *file, bool isdir)
 	struct fuse_conn *fc = get_fuse_conn(inode);
 	int err;
 
-	/* VFS checks this, but only _after_ ->open() */
-	if (file->f_flags & O_DIRECT)
-		return -EINVAL;
-
 	err = generic_file_open(inode, file);
 	if (err)
 		return err;
@@ -2132,6 +2149,15 @@ int fuse_notify_poll_wakeup(struct fuse_conn *fc,
 	return 0;
 }
 
+
+static int fuse_check_flags(struct file *filp, int flags)
+{
+	if ((filp->f_flags ^ flags) & O_DIRECT)
+		return -EINVAL;
+	return 0;
+}
+
+
 static const struct file_operations fuse_file_operations = {
 	.llseek		= fuse_file_llseek,
 	.read		= do_sync_read,
@@ -2149,6 +2175,7 @@ static const struct file_operations fuse_file_operations = {
 	.unlocked_ioctl	= fuse_file_ioctl,
 	.compat_ioctl	= fuse_file_compat_ioctl,
 	.poll		= fuse_file_poll,
+	.check_flags	= fuse_check_flags,
 };
 
 static const struct file_operations fuse_direct_io_file_operations = {
@@ -2165,6 +2192,7 @@ static const struct file_operations fuse_direct_io_file_operations = {
 	.unlocked_ioctl	= fuse_file_ioctl,
 	.compat_ioctl	= fuse_file_compat_ioctl,
 	.poll		= fuse_file_poll,
+	.check_flags	= fuse_check_flags,
 	/* no splice_read */
 };
 
-- 
1.7.4.4

--
To unsubscribe from this list: send the line "unsubscribe linux-nfs" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: [PATCH v2 2/2] fuse: permit O_DIRECT flag in open()

[Christoph Hellwig]

On Wed, Jul 27, 2011 at 03:56:19AM -0700, Anand Avati wrote:
> FUSE currently disallows O_DIRECT flag in open(). It is tricky dealing with
> setting/unsetting of O_DIRECT flag on an open file. There are applications
> (primarily VMs and databases) which open files with O_DIRECT flag. These
> applications do not work on FUSE due to this limitation.
> 
> The approach with this patch is to permit opens with O_DIRECT, but instead
> disable setting/unsetting of the O_DIRECT flag no matter how the file was
> opened. This limitation is for more practical than disallowing O_DIRECT
> altogether.

But it's also entirely incorrect.  Fix your userspace to have a proper
fallback if O_DIRECT opens fail.  It's a feature only supported by a few
filesystems, and very few operating systems.

Re: [PATCH v2 2/2] fuse: permit O_DIRECT flag in open()

[Anand Avati]

On Thu, Jul 28, 2011 at 2:35 AM, Christoph Hellwig <hch@infradead.org> wrote:
> On Wed, Jul 27, 2011 at 03:56:19AM -0700, Anand Avati wrote:
>> FUSE currently disallows O_DIRECT flag in open(). It is tricky dealing with
>> setting/unsetting of O_DIRECT flag on an open file. There are applications
>> (primarily VMs and databases) which open files with O_DIRECT flag. These
>> applications do not work on FUSE due to this limitation.
>>
>> The approach with this patch is to permit opens with O_DIRECT, but instead
>> disable setting/unsetting of the O_DIRECT flag no matter how the file was
>> opened. This limitation is for more practical than disallowing O_DIRECT
>> altogether.
>
> But it's also entirely incorrect.  Fix your userspace to have a proper
> fallback if O_DIRECT opens fail.  It's a feature only supported by a few
> filesystems, and very few operating systems.

FUSE already has a direct IO implementation (struct file_operations
fuse_direct_io_file_operation) but is currently set as a file's f_ops
only based on flags returned from the filesystem server (in the reply
of an open() call). This patch just lets the filesystem server know a
userspace application's intention to open a file with O_DIRECT and
leaves it with the decision to pick direct IO mode on the file (just
the way it already does before the patch.) There is already a
framework for direct IO operations and this patch only bridges
O_DIRECT flag with that framework. Do you still think this is entirely
incorrect?

Avati
--
To unsubscribe from this list: send the line "unsubscribe linux-fsdevel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: [PATCH v2 2/2] fuse: permit O_DIRECT flag in open()

[Christoph Hellwig]

On Thu, Jul 28, 2011 at 01:44:08PM +0530, Anand Avati wrote:
> FUSE already has a direct IO implementation (struct file_operations
> fuse_direct_io_file_operation) but is currently set as a file's f_ops
> only based on flags returned from the filesystem server (in the reply
> of an open() call). This patch just lets the filesystem server know a
> userspace application's intention to open a file with O_DIRECT and
> leaves it with the decision to pick direct IO mode on the file (just
> the way it already does before the patch.) There is already a
> framework for direct IO operations and this patch only bridges
> O_DIRECT flag with that framework. Do you still think this is entirely
> incorrect?

Your clearing of the O_DIRECT flag is.  The current handling of O_DIRECT
in the VFS is a bit nasty, but if you have an issue with that fix it up
properly instead of piling more hacks on top of it.

Re: [PATCH v2 2/2] fuse: permit O_DIRECT flag in open()

[Anand Avati]

On Thu, Jul 28, 2011 at 5:19 PM, Christoph Hellwig <hch@infradead.org> wrote:
>
> Your clearing of the O_DIRECT flag is.  The current handling of O_DIRECT
> in the VFS is a bit nasty, but if you have an issue with that fix it up
> properly instead of piling more hacks on top of it.
>

Hmm, I thought that the fact VFS checked for O_DIRECT after ->open()
was intentionally giving an opportunity to the filesystem to clear it.
But looking backwards now, I think it was my misunderstanding of the
intention. I will look deeper into a cleaner implementation of
O_DIRECT support.

Avati
--
To unsubscribe from this list: send the line "unsubscribe linux-fsdevel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html