diff mbox series

[v2,3/4] exportfs: define FILEID_INO64_GEN* file handle types

Message ID 20231023180801.2953446-4-amir73il@gmail.com (mailing list archive)
State New, archived
Headers show
Series Support more filesystems with FAN_REPORT_FID | expand

Commit Message

Amir Goldstein Oct. 23, 2023, 6:08 p.m. UTC 
Similar to the common FILEID_INO32* file handle types, define common
FILEID_INO64* file handle types.

The type values of FILEID_INO64_GEN and FILEID_INO64_GEN_PARENT are the
values returned by fuse and xfs for 64bit ino encoded file handle types.

Note that these type value are filesystem specific and they do not define
a universal file handle format, for example:
fuse encodes FILEID_INO64_GEN as [ino-hi32,ino-lo32,gen] and xfs encodes
FILEID_INO64_GEN as [hostr-order-ino64,gen] (a.k.a xfs_fid64).

The FILEID_INO64_GEN fhandle type is going to be used for file ids for
fanotify from filesystems that do not support NFS export.

Reviewed-by: Jan Kara <jack@suse.cz>
Reviewed-by: Jeff Layton <jlayton@kernel.org>
Signed-off-by: Amir Goldstein <amir73il@gmail.com>
---
 fs/fuse/inode.c          |  7 ++++---
 include/linux/exportfs.h | 11 +++++++++++
 2 files changed, 15 insertions(+), 3 deletions(-)

Comments

Christoph Hellwig Oct. 27, 2023, 6:05 a.m. UTC | #1 
On Mon, Oct 23, 2023 at 09:08:00PM +0300, Amir Goldstein wrote:
> Similar to the common FILEID_INO32* file handle types, define common
> FILEID_INO64* file handle types.
> 
> The type values of FILEID_INO64_GEN and FILEID_INO64_GEN_PARENT are the
> values returned by fuse and xfs for 64bit ino encoded file handle types.

Please actually switch xfs to fully use the helpers instead of
duplicating the logic.  Presumable the same for fuse, but for that
I'd need to look at how it works for fuse right now and if there's not
some subtle differences.
Amir Goldstein Oct. 27, 2023, 6:43 a.m. UTC | #2 
On Fri, Oct 27, 2023 at 9:05 AM Christoph Hellwig <hch@infradead.org> wrote:
>
> On Mon, Oct 23, 2023 at 09:08:00PM +0300, Amir Goldstein wrote:
> > Similar to the common FILEID_INO32* file handle types, define common
> > FILEID_INO64* file handle types.
> >
> > The type values of FILEID_INO64_GEN and FILEID_INO64_GEN_PARENT are the
> > values returned by fuse and xfs for 64bit ino encoded file handle types.
>
> Please actually switch xfs to fully use the helpers instead of
> duplicating the logic.

I will follow up with another patch.

> Presumable the same for fuse, but for that
> I'd need to look at how it works for fuse right now and if there's not
> some subtle differences.
>

There are subtle differences:
1. fuse encodes an internal nodeid - not i_ino
2. fuse encodes the inode number as [low32,high32]

It cannot use the generic helper.

Thanks,
Amir.
Christoph Hellwig Oct. 27, 2023, 7:32 a.m. UTC | #3 
On Fri, Oct 27, 2023 at 09:43:01AM +0300, Amir Goldstein wrote:
> > Presumable the same for fuse, but for that
> > I'd need to look at how it works for fuse right now and if there's not
> > some subtle differences.
> >
> 
> There are subtle differences:
> 1. fuse encodes an internal nodeid - not i_ino
> 2. fuse encodes the inode number as [low32,high32]
> 
> It cannot use the generic helper.

That's what I almost feared.  It still should use the common symbolic
name for the format just to make everyones life simpler.
Amir Goldstein Oct. 27, 2023, 2:27 p.m. UTC | #4 
On Fri, Oct 27, 2023 at 10:32 AM Christoph Hellwig <hch@infradead.org> wrote:
>
> On Fri, Oct 27, 2023 at 09:43:01AM +0300, Amir Goldstein wrote:
> > > Presumable the same for fuse, but for that
> > > I'd need to look at how it works for fuse right now and if there's not
> > > some subtle differences.
> > >
> >
> > There are subtle differences:
> > 1. fuse encodes an internal nodeid - not i_ino
> > 2. fuse encodes the inode number as [low32,high32]

Sorry, that's [hi32,lo32] as written in commit message.

> >
> > It cannot use the generic helper.
>
> That's what I almost feared.  It still should use the common symbolic
> name for the format just to make everyones life simpler.

That's what I thought.
This patch converts fuse to use the new defined FILEID_INO64*
constants.

I plan to send a followup patch to xfs to use the symbolic name
after this constant has landed in vfs.
It is going to be easier than collaborating the merges of xfs and vfs
and there is no reason to rush it.

Thanks,
Amir.
diff mbox series

Patch

diff --git a/fs/fuse/inode.c b/fs/fuse/inode.c
index 2e4eb7cf26fb..e63f966698a5 100644
--- a/fs/fuse/inode.c
+++ b/fs/fuse/inode.c
@@ -1002,7 +1002,7 @@  static int fuse_encode_fh(struct inode *inode, u32 *fh, int *max_len,
 	}
 
 	*max_len = len;
-	return parent ? 0x82 : 0x81;
+	return parent ? FILEID_INO64_GEN_PARENT : FILEID_INO64_GEN;
 }
 
 static struct dentry *fuse_fh_to_dentry(struct super_block *sb,
@@ -1010,7 +1010,8 @@  static struct dentry *fuse_fh_to_dentry(struct super_block *sb,
 {
 	struct fuse_inode_handle handle;
 
-	if ((fh_type != 0x81 && fh_type != 0x82) || fh_len < 3)
+	if ((fh_type != FILEID_INO64_GEN &&
+	     fh_type != FILEID_INO64_GEN_PARENT) || fh_len < 3)
 		return NULL;
 
 	handle.nodeid = (u64) fid->raw[0] << 32;
@@ -1024,7 +1025,7 @@  static struct dentry *fuse_fh_to_parent(struct super_block *sb,
 {
 	struct fuse_inode_handle parent;
 
-	if (fh_type != 0x82 || fh_len < 6)
+	if (fh_type != FILEID_INO64_GEN_PARENT || fh_len < 6)
 		return NULL;
 
 	parent.nodeid = (u64) fid->raw[3] << 32;
diff --git a/include/linux/exportfs.h b/include/linux/exportfs.h
index 85bd027494e5..4119d3ee72eb 100644
--- a/include/linux/exportfs.h
+++ b/include/linux/exportfs.h
@@ -98,6 +98,17 @@  enum fid_type {
 	 */
 	FILEID_FAT_WITH_PARENT = 0x72,
 
+	/*
+	 * 64 bit inode number, 32 bit generation number.
+	 */
+	FILEID_INO64_GEN = 0x81,
+
+	/*
+	 * 64 bit inode number, 32 bit generation number,
+	 * 64 bit parent inode number, 32 bit parent generation.
+	 */
+	FILEID_INO64_GEN_PARENT = 0x82,
+
 	/*
 	 * 128 bit child FID (struct lu_fid)
 	 * 128 bit parent FID (struct lu_fid)