On Tue Mar 3, 2020 at 11:19 AM, Daniel Xu wrote:
> On Tue Mar 3, 2020 at 8:19 AM, Tejun Heo wrote:
> > Hello,
> >
> >
> > On Mon, Mar 02, 2020 at 05:39:00PM -0800, Daniel Xu wrote:
> > > +static int kernfs_vfs_user_xattr_set(const struct xattr_handler *handler,
> > > + struct dentry *unused, struct inode *inode,
> > > + const char *suffix, const void *value,
> > > + size_t size, int flags)
> > > +{
> > ...
> > > + if (value && atomic_inc_return(nr) > KERNFS_MAX_USER_XATTRS) {
> > > + ret = -ENOSPC;
> > > + goto dec_out;
> > > + }
> >
> >
> > So, we limit the number of user xattrs here but
> >
> >
> > > + ret = kernfs_vfs_xattr_set(handler, unused, inode, suffix, value,
> > > + size, flags);
> >
> >
> > This will call into simple_xattr_set() which doesn't put any further
> > restriction on size and just calls GFP_KERNEL kmalloc on it allowing
> > users incur high-order allocations. Maybe it'd make sense to limit
> > both the number and size?
>
>
> Ah yeah good point. Will add.
>
It looks like in fs/xattr.c:setxattr, there is already:
...
if (size) {
if (size > XATTR_SIZE_MAX)
return -E2BIG;
...
where XATTR_SIZE_MAX is defined as 64k. Do you want it even smaller?
Thanks,
Daniel
