[PATCH] xfs: Correctly lock inode when removing suid and security marks

[PATCH] xfs: Correctly lock inode when removing suid and security marks

[Jan Kara]

Currently XFS calls file_remove_suid() without holding i_mutex. This is
wrong because that function can end up messing with file permissions and
security xattrs for which we need i_mutex held.

Fix the problem by grabbing iolock exclusively when we will need to
change anything in permissions / xattrs.

Signed-off-by: Jan Kara <jack@suse.cz>
---
 fs/xfs/xfs_file.c | 12 ++++++++++++
 1 file changed, 12 insertions(+)

diff --git a/fs/xfs/xfs_file.c b/fs/xfs/xfs_file.c
index eb596b419942..ad6636ac4943 100644
--- a/fs/xfs/xfs_file.c
+++ b/fs/xfs/xfs_file.c
@@ -521,6 +521,18 @@ restart:
 	if (error)
 		return error;
 
+	/* For changing security info in file_remove_suid() we need i_mutex */
+	if (!IS_NOSEC(inode) && *iolock == XFS_IOLOCK_SHARED) {
+		struct dentry *dentry = file->f_path.dentry;
+
+		if (should_remove_suid(dentry) ||
+		    security_inode_need_killpriv(dentry)) {
+			xfs_rw_iunlock(ip, *iolock);
+			*iolock = XFS_IOLOCK_EXCL;
+			xfs_rw_ilock(ip, *iolock);
+			goto restart;
+		}
+	}
 	/*
 	 * If the offset is beyond the size of the file, we need to zero any
 	 * blocks that fall between the existing EOF and the start of this
-- 
1.8.1.4

_______________________________________________
xfs mailing list
xfs@oss.sgi.com
http://oss.sgi.com/mailman/listinfo/xfs

Re: [PATCH] xfs: Correctly lock inode when removing suid and security marks

[Brian Foster]

On Tue, Dec 02, 2014 at 04:01:29PM +0100, Jan Kara wrote:
> Currently XFS calls file_remove_suid() without holding i_mutex. This is
> wrong because that function can end up messing with file permissions and
> security xattrs for which we need i_mutex held.
> 
> Fix the problem by grabbing iolock exclusively when we will need to
> change anything in permissions / xattrs.
> 
> Signed-off-by: Jan Kara <jack@suse.cz>
> ---

Hi Jan,

This doesn't compile... it looks like we need to include the security.h
header. FWIW, even then I get an undefined symbol error when compiling
as a module (security_inode_need_killpriv() does not appear to be
exported).

Brian

>  fs/xfs/xfs_file.c | 12 ++++++++++++
>  1 file changed, 12 insertions(+)
> 
> diff --git a/fs/xfs/xfs_file.c b/fs/xfs/xfs_file.c
> index eb596b419942..ad6636ac4943 100644
> --- a/fs/xfs/xfs_file.c
> +++ b/fs/xfs/xfs_file.c
> @@ -521,6 +521,18 @@ restart:
>  	if (error)
>  		return error;
>  
> +	/* For changing security info in file_remove_suid() we need i_mutex */
> +	if (!IS_NOSEC(inode) && *iolock == XFS_IOLOCK_SHARED) {
> +		struct dentry *dentry = file->f_path.dentry;
> +
> +		if (should_remove_suid(dentry) ||
> +		    security_inode_need_killpriv(dentry)) {
> +			xfs_rw_iunlock(ip, *iolock);
> +			*iolock = XFS_IOLOCK_EXCL;
> +			xfs_rw_ilock(ip, *iolock);
> +			goto restart;
> +		}
> +	}
>  	/*
>  	 * If the offset is beyond the size of the file, we need to zero any
>  	 * blocks that fall between the existing EOF and the start of this
> -- 
> 1.8.1.4
> 
> _______________________________________________
> xfs mailing list
> xfs@oss.sgi.com
> http://oss.sgi.com/mailman/listinfo/xfs

_______________________________________________
xfs mailing list
xfs@oss.sgi.com
http://oss.sgi.com/mailman/listinfo/xfs

Re: [PATCH] xfs: Correctly lock inode when removing suid and security marks

[Jan Kara]

On Tue 02-12-14 11:35:48, Brian Foster wrote:
> On Tue, Dec 02, 2014 at 04:01:29PM +0100, Jan Kara wrote:
> > Currently XFS calls file_remove_suid() without holding i_mutex. This is
> > wrong because that function can end up messing with file permissions and
> > security xattrs for which we need i_mutex held.
> > 
> > Fix the problem by grabbing iolock exclusively when we will need to
> > change anything in permissions / xattrs.
> > 
> > Signed-off-by: Jan Kara <jack@suse.cz>
> > ---
> 
> Hi Jan,
> 
> This doesn't compile... it looks like we need to include the security.h
> header. FWIW, even then I get an undefined symbol error when compiling
> as a module (security_inode_need_killpriv() does not appear to be
> exported).
  Sorry, forgot to amend the include in the commit. Regarding export of
security_inode_need_killpriv() - right, I had security XFS compiled in so I
didn't notice. Before I go and fix this up in the obvious way, does anyone
have better idea how to fix this than to second guess what
file_remove_suid() does? Maybe a VFS helper like file_needs_remove_suid()
will be cleaner than what I did?

								Honza

> >  fs/xfs/xfs_file.c | 12 ++++++++++++
> >  1 file changed, 12 insertions(+)
> > 
> > diff --git a/fs/xfs/xfs_file.c b/fs/xfs/xfs_file.c
> > index eb596b419942..ad6636ac4943 100644
> > --- a/fs/xfs/xfs_file.c
> > +++ b/fs/xfs/xfs_file.c
> > @@ -521,6 +521,18 @@ restart:
> >  	if (error)
> >  		return error;
> >  
> > +	/* For changing security info in file_remove_suid() we need i_mutex */
> > +	if (!IS_NOSEC(inode) && *iolock == XFS_IOLOCK_SHARED) {
> > +		struct dentry *dentry = file->f_path.dentry;
> > +
> > +		if (should_remove_suid(dentry) ||
> > +		    security_inode_need_killpriv(dentry)) {
> > +			xfs_rw_iunlock(ip, *iolock);
> > +			*iolock = XFS_IOLOCK_EXCL;
> > +			xfs_rw_ilock(ip, *iolock);
> > +			goto restart;
> > +		}
> > +	}
> >  	/*
> >  	 * If the offset is beyond the size of the file, we need to zero any
> >  	 * blocks that fall between the existing EOF and the start of this
> > -- 
> > 1.8.1.4
> > 
> > _______________________________________________
> > xfs mailing list
> > xfs@oss.sgi.com
> > http://oss.sgi.com/mailman/listinfo/xfs
-- 
Jan Kara <jack@suse.cz>
SUSE Labs, CR

_______________________________________________
xfs mailing list
xfs@oss.sgi.com
http://oss.sgi.com/mailman/listinfo/xfs

Re: [PATCH] xfs: Correctly lock inode when removing suid and security marks

[Dave Chinner]

On Tue, Dec 02, 2014 at 09:47:56PM +0100, Jan Kara wrote:
> On Tue 02-12-14 11:35:48, Brian Foster wrote:
> > On Tue, Dec 02, 2014 at 04:01:29PM +0100, Jan Kara wrote:
> > > Currently XFS calls file_remove_suid() without holding i_mutex. This is
> > > wrong because that function can end up messing with file permissions and
> > > security xattrs for which we need i_mutex held.
> > > 
> > > Fix the problem by grabbing iolock exclusively when we will need to
> > > change anything in permissions / xattrs.
> > > 
> > > Signed-off-by: Jan Kara <jack@suse.cz>
> > > ---
> > 
> > Hi Jan,
> > 
> > This doesn't compile... it looks like we need to include the security.h
> > header. FWIW, even then I get an undefined symbol error when compiling
> > as a module (security_inode_need_killpriv() does not appear to be
> > exported).
>   Sorry, forgot to amend the include in the commit. Regarding export of
> security_inode_need_killpriv() - right, I had security XFS compiled in so I
> didn't notice. Before I go and fix this up in the obvious way, does anyone
> have better idea how to fix this than to second guess what
> file_remove_suid() does? Maybe a VFS helper like file_needs_remove_suid()
> will be cleaner than what I did?

Helper seems like a sane idea - that way the filesystems can use
them as a matched pair if need be...

Cheers,

Dave.
-- 
Dave Chinner
david@fromorbit.com

_______________________________________________
xfs mailing list
xfs@oss.sgi.com
http://oss.sgi.com/mailman/listinfo/xfs