[PATCH 1/1] fs: xattr: Don't display attributes without read access

[PATCH 1/1] fs: xattr: Don't display attributes without read access

[Fabian Frederick]

Any user can display extented attribute names without read
access.

eg: attr -l <filename>

This patch checks inode_permission in listxattr common
function before executing vfs_listxattr.

Signed-off-by: Fabian Frederick <fabf@skynet.be>
---
 fs/xattr.c | 4 ++++
 1 file changed, 4 insertions(+)

diff --git a/fs/xattr.c b/fs/xattr.c
index 3377dff..d26b280 100644
--- a/fs/xattr.c
+++ b/fs/xattr.c
@@ -543,6 +543,10 @@ listxattr(struct dentry *d, char __user *list, size_t size)
 	char *klist = NULL;
 	char *vlist = NULL;	/* If non-NULL, we used vmalloc() */
 
+	error = inode_permission(d->d_inode, MAY_READ);
+	if (error)
+		return error;
+
 	if (size) {
 		if (size > XATTR_LIST_MAX)
 			size = XATTR_LIST_MAX;
-- 
1.8.1.4

Re: [PATCH 1/1] fs: xattr: Don't display attributes without read access

[Fabian Frederick]

On Tue, 18 Feb 2014 15:16:50 -0800
Andrew Morton <akpm@linux-foundation.org> wrote:

> On Sun, 16 Feb 2014 20:31:01 +0800 Fabian Frederick <fabf@skynet.be> wrote:
> 
> > Any user can display extented attribute names without read
> > access.
> > 
> > eg: attr -l <filename>
> > 
> > This patch checks inode_permission in listxattr common
> > function before executing vfs_listxattr.
> > 
> > ...
> >
> > --- a/fs/xattr.c
> > +++ b/fs/xattr.c
> > @@ -543,6 +543,10 @@ listxattr(struct dentry *d, char __user *list, size_t size)
> >  	char *klist = NULL;
> >  	char *vlist = NULL;	/* If non-NULL, we used vmalloc() */
> >  
> > +	error = inode_permission(d->d_inode, MAY_READ);
> > +	if (error)
> > +		return error;
> > +
> >  	if (size) {
> >  		if (size > XATTR_LIST_MAX)
> >  			size = XATTR_LIST_MAX;
> 
> erk.  Doesn't this mean that if existing userspace is relying on the
> current behaviour, this patch will cause breakage?
> 
IMHO userspace applications already receive weird results in that case.

Without read permission, attr -l receives attribute names which means
it tries lgetxattr on those attributes where result is EACCESS :
"Attribute <attribute name> has -1 byte" !!!

Besides, is it semantically correct for a user to have access to
 a part of "file content" without read access ?
 
With that patch, attr -l <filename> displays

attr_list: Permission denied
Could not list "(null)" for <filename>

On the other hand, when stracing that situation, 
I do see attr making more noise than usual ie it's opening all attr.mo 
twice so I guess I should return something else than "permission denied" to avoid problems in userspace ...

Fabian

Re: [PATCH 1/1] fs: xattr: Don't display attributes without read access

[Andrew Morton]

On Sun, 16 Feb 2014 20:31:01 +0800 Fabian Frederick <fabf@skynet.be> wrote:

> Any user can display extented attribute names without read
> access.
> 
> eg: attr -l <filename>
> 
> This patch checks inode_permission in listxattr common
> function before executing vfs_listxattr.
> 
> ...
>
> --- a/fs/xattr.c
> +++ b/fs/xattr.c
> @@ -543,6 +543,10 @@ listxattr(struct dentry *d, char __user *list, size_t size)
>  	char *klist = NULL;
>  	char *vlist = NULL;	/* If non-NULL, we used vmalloc() */
>  
> +	error = inode_permission(d->d_inode, MAY_READ);
> +	if (error)
> +		return error;
> +
>  	if (size) {
>  		if (size > XATTR_LIST_MAX)
>  			size = XATTR_LIST_MAX;

erk.  Doesn't this mean that if existing userspace is relying on the
current behaviour, this patch will cause breakage?

Re: [PATCH 1/1] fs: xattr: Don't display attributes without read access

[Fabian Frederick]

On Tue, 18 Feb 2014 15:16:50 -0800
Andrew Morton <akpm@linux-foundation.org> wrote:

> On Sun, 16 Feb 2014 20:31:01 +0800 Fabian Frederick <fabf@skynet.be> wrote:
> 
> > Any user can display extented attribute names without read
> > access.
> > 
> > eg: attr -l <filename>
> > 
> > This patch checks inode_permission in listxattr common
> > function before executing vfs_listxattr.
> > 
> > ...
> >
> > --- a/fs/xattr.c
> > +++ b/fs/xattr.c
> > @@ -543,6 +543,10 @@ listxattr(struct dentry *d, char __user *list, size_t size)
> >  	char *klist = NULL;
> >  	char *vlist = NULL;	/* If non-NULL, we used vmalloc() */
> >  
> > +	error = inode_permission(d->d_inode, MAY_READ);
> > +	if (error)
> > +		return error;
> > +
> >  	if (size) {
> >  		if (size > XATTR_LIST_MAX)
> >  			size = XATTR_LIST_MAX;
> 
> erk.  Doesn't this mean that if existing userspace is relying on the
> current behaviour, this patch will cause breakage?
> 
FYI, I just noticed FreeBSD was giving the expected result : no attribute name revealed
 without read access ....