On Thu, 19 Apr 2018, Andrew Morton wrote:
> On Thu, 19 Apr 2018 12:12:38 -0400 (EDT) Mikulas Patocka <mpatocka@xxxxxxxxxx> wrote:
>
> > The kvmalloc function tries to use kmalloc and falls back to vmalloc if
> > kmalloc fails.
> >
> > Unfortunatelly, some kernel code has bugs - it uses kvmalloc and then
> > uses DMA-API on the returned memory or frees it with kfree. Such bugs were
> > found in the virtio-net driver, dm-integrity or RHEL7 powerpc-specific
> > code.
> >
> > These bugs are hard to reproduce because vmalloc falls back to kmalloc
> > only if memory is fragmented.
>
> Yes, that's nasty.
>
> > In order to detect these bugs reliably I submit this patch that changes
> > kvmalloc to always use vmalloc if CONFIG_DEBUG_VM is turned on.
> >
> > ...
> >
> > --- linux-2.6.orig/mm/util.c 2018-04-18 15:46:23.000000000 +0200
> > +++ linux-2.6/mm/util.c 2018-04-18 16:00:43.000000000 +0200
> > @@ -395,6 +395,7 @@ EXPORT_SYMBOL(vm_mmap);
> > */
> > void *kvmalloc_node(size_t size, gfp_t flags, int node)
> > {
> > +#ifndef CONFIG_DEBUG_VM
> > gfp_t kmalloc_flags = flags;
> > void *ret;
> >
> > @@ -426,6 +427,7 @@ void *kvmalloc_node(size_t size, gfp_t f
> > */
> > if (ret || size <= PAGE_SIZE)
> > return ret;
> > +#endif
> >
> > return __vmalloc_node_flags_caller(size, node, flags,
> > __builtin_return_address(0));
>
> Well, it doesn't have to be done at compile-time, does it? We could
> add a knob (in debugfs, presumably) which enables this at runtime.
> That's far more user-friendly.
But who will turn it on in debugfs? It should be default for debugging
kernels, so that users using them would report the error.
Conditioning it on CONFIG_DEBUG_SG is better than CONFIG_DEBUG_VM, it will
print a stacktrace where the incorrect use happened.
Mikulas
