On Fri, 2018-10-12 at 16:32 +0200, Jessica Yu wrote:
> +++ Dave Hansen [11/10/18 16:47 -0700]:
> > On 10/11/2018 04:31 PM, Rick Edgecombe wrote:
> > > + if (check_inc_mod_rlimit(size))
> > > + return NULL;
> > > +
> > > p = __vmalloc_node_range(size, MODULE_ALIGN, module_alloc_base,
> > > module_alloc_base + MODULES_VSIZE,
> > > gfp_mask, PAGE_KERNEL_EXEC, 0,
> > > @@ -65,6 +68,8 @@ void *module_alloc(unsigned long size)
> > > return NULL;
> > > }
> > >
> > > + update_mod_rlimit(p, size);
> >
> > Is there a reason we couldn't just rename all of the existing per-arch
> > module_alloc() calls to be called, say, "arch_module_alloc()", then put
> > this new rlimit code in a generic helper that does:
> >
> >
> > if (check_inc_mod_rlimit(size))
> > return NULL;
> >
> > p = arch_module_alloc(...);
> >
> > ...
> >
> > update_mod_rlimit(p, size);
> >
>
> I second this suggestion. Just make module_{alloc,memfree} generic,
> non-weak functions that call the rlimit helpers in addition to the
> arch-specific arch_module_{alloc,memfree} functions.
>
> Jessica
Ok, thanks. I am going to try another version of this with just a system wide
BPF JIT limit based on the problems Jann brought up. I think it would be nice to
have a module space limit, but as far as I know the only way today un-privlidged
users could fill the space is from BPF JIT. Unless you see another purpose long
term?
Rick
