[PATCH] xfs: xfs_nfs_get_inode support zero generation

[PATCH] xfs: xfs_nfs_get_inode support zero generation

[renlei1]

From: Ren Lei <renlei1@chinatelecom.cn>

If generation is zero, bypass the verification of generation number
to avoid stale file error. (Be consistent with other fs, such as
ext4, fat, jfs, etc.)

Signed-off-by: Ren Lei<renlei1@chinatelecom.cn>
---
 fs/xfs/xfs_export.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/fs/xfs/xfs_export.c b/fs/xfs/xfs_export.c
index 1064c2342876..cbee32c5ad37 100644
--- a/fs/xfs/xfs_export.c
+++ b/fs/xfs/xfs_export.c
@@ -146,7 +146,7 @@ xfs_nfs_get_inode(
 		return ERR_PTR(error);
 	}
 
-	if (VFS_I(ip)->i_generation != generation) {
+	if (generation && VFS_I(ip)->i_generation != generation) {
 		xfs_irele(ip);
 		return ERR_PTR(-ESTALE);
 	}
-- 
2.27.0

Re: [PATCH] xfs: xfs_nfs_get_inode support zero generation

[Darrick J. Wong]

On Thu, May 11, 2023 at 06:17:21PM +0800, renlei1@chinatelecom.cn wrote:
> From: Ren Lei <renlei1@chinatelecom.cn>
> 
> If generation is zero, bypass the verification of generation number
> to avoid stale file error. (Be consistent with other fs, such as
> ext4, fat, jfs, etc.)

What code is affected by the gen==0 handles being rejected?  Is there a
user program or test case where this is required?

--D

> Signed-off-by: Ren Lei<renlei1@chinatelecom.cn>
> ---
>  fs/xfs/xfs_export.c | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/fs/xfs/xfs_export.c b/fs/xfs/xfs_export.c
> index 1064c2342876..cbee32c5ad37 100644
> --- a/fs/xfs/xfs_export.c
> +++ b/fs/xfs/xfs_export.c
> @@ -146,7 +146,7 @@ xfs_nfs_get_inode(
>  		return ERR_PTR(error);
>  	}
>  
> -	if (VFS_I(ip)->i_generation != generation) {
> +	if (generation && VFS_I(ip)->i_generation != generation) {
>  		xfs_irele(ip);
>  		return ERR_PTR(-ESTALE);
>  	}
> -- 
> 2.27.0
> 

Re: [PATCH] xfs: xfs_nfs_get_inode support zero generation

[Dave Chinner]

On Thu, May 11, 2023 at 04:22:06PM -0700, Darrick J. Wong wrote:
> On Thu, May 11, 2023 at 06:17:21PM +0800, renlei1@chinatelecom.cn wrote:
> > From: Ren Lei <renlei1@chinatelecom.cn>
> > 
> > If generation is zero, bypass the verification of generation number
> > to avoid stale file error. (Be consistent with other fs, such as
> > ext4, fat, jfs, etc.)
> 
> What code is affected by the gen==0 handles being rejected?  Is there a
> user program or test case where this is required?

A generation number of 0 is perfectly valid in XFS. We've been
creating them in XFS filesystems since 1993 and never had a problem
with filehandle verification. Indeed, every root inode in every XFS
filesystem ever made will have a generation number of 0.

Yup, a random XFS filesystem recently made from a current xfsprogs:

# xfs_db /dev/vdc
xfs_db> sb 0
xfs_db> a rootino
xfs_db> p core.gen
core.gen = 0
xfs_db> 

-Dave.
-- 
Dave Chinner
david@fromorbit.com

Re: Re: [PATCH] xfs: xfs_nfs_get_inode support zero generation

[Darrick J. Wong]

On Fri, May 12, 2023 at 10:53:29AM +0800, renlei1@chinatelecom.cn wrote:
> Yup, gen is 0 for the inodes created by libxfs, such as rootino, rbmino, rsumino.
> but those inodes will never be freed, and gen will always zero.
> so I think bypass the verification if gen==0 is still valid.

I disagree.  Handles have long encoded inode and generation to prevent
users from unintentionally modifying a file when the inode number is
recycled as a result of an unlink/create cycle.

--D

> Regards,
> Lei
>  
> From: Dave Chinner
> Date: 2023-05-12 10:00
> To: Darrick J. Wong
> CC: renlei1; linux-xfs; linux-kernel
> Subject: Re: [PATCH] xfs: xfs_nfs_get_inode support zero generation
> On Thu, May 11, 2023 at 04:22:06PM -0700, Darrick J. Wong wrote:
> > On Thu, May 11, 2023 at 06:17:21PM +0800, renlei1@chinatelecom.cn wrote:
> > > From: Ren Lei <renlei1@chinatelecom.cn>
> > > 
> > > If generation is zero, bypass the verification of generation number
> > > to avoid stale file error. (Be consistent with other fs, such as
> > > ext4, fat, jfs, etc.)
> > 
> > What code is affected by the gen==0 handles being rejected?  Is there a
> > user program or test case where this is required?
>  
> A generation number of 0 is perfectly valid in XFS. We've been
> creating them in XFS filesystems since 1993 and never had a problem
> with filehandle verification. Indeed, every root inode in every XFS
> filesystem ever made will have a generation number of 0.
>  
> Yup, a random XFS filesystem recently made from a current xfsprogs:
>  
> # xfs_db /dev/vdc
> xfs_db> sb 0
> xfs_db> a rootino
> xfs_db> p core.gen
> core.gen = 0
> xfs_db> 
>  
> -Dave.
> -- 
> Dave Chinner
> david@fromorbit.com
>  

Re: Re: [PATCH] xfs: xfs_nfs_get_inode support zero generation

[Dave Chinner]

On Fri, May 12, 2023 at 10:53:29AM +0800, renlei1@chinatelecom.cn wrote:
> Yup, gen is 0 for the inodes created by libxfs, such as rootino, rbmino, rsumino.
> but those inodes will never be freed, and gen will always zero.

Yes, but why does that even matter for file handle verification?

> so I think bypass the verification if gen==0 is still valid.

Did you think about what happens when an inode->gen
overflows from 0xffffffff to 0 as part of a normal unlink/create
cycle?

-Dave.
-- 
Dave Chinner
david@fromorbit.com