Re: [RFC][PATCH 03/20] Add vfsmount writer count

[Dave Hansen]

On Sun, 2006-06-18 at 19:33 +0100, Al Viro wrote:
> On Fri, Jun 16, 2006 at 04:12:15PM -0700, Dave Hansen wrote:
> > +/*
> > + * Must be called under write lock on mnt->mnt_sb->s_umount,
> > + * this prevents concurrent decrements which could make the
> > + * value -1, and test in mnt_want_write() wrongly succeed
> > + */
> > +static inline int mnt_make_readonly(struct vfsmount *mnt)
> > +{
> > +	if (!atomic_dec_and_test(&mnt->mnt_writers)) {
> > +		atomic_inc(&mnt->mnt_writers);
> > +		return -EBUSY;
> > +	}
> > +	return 0;
> > +}
> 
> Check in faccessat() could get screwed by that, if you just lose the
> race with final writer going away.  Then mnt_make_readonly() will 
> fail (as it should) and access(2) give -EROFS.

Very true.  How about this to fix it?

--- lxc/fs//open.c~C8.1-fix-faccesat    2006-06-19 09:59:41.000000000 -0700
+++ lxc-dave/fs//open.c 2006-06-19 10:01:25.000000000 -0700
@@ -546,8 +546,12 @@ asmlinkage long sys_faccessat(int dfd, c
           special_file(nd.dentry->d_inode->i_mode))
                goto out_path_release;

-       if(__mnt_is_readonly(nd.mnt) || IS_RDONLY(nd.dentry->d_inode))
-               res = -EROFS;
+       res = mnt_want_write(nd.mnt);
+       if (!res) {
+               mnt_drop_write(nd.mnt);
+               if(IS_RDONLY(nd.dentry->d_inode))
+                       res = -EROFS;
+       }

 out_path_release:
        path_release(&nd);


-- Dave

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/