Messages in this thread | | | Date | Tue, 9 Sep 2014 15:41:18 -0400 | From | Vivek Goyal <> | Subject | Re: [PATCH 2/4] kaslr: check if the random addr is available |
| |
On Sat, Sep 06, 2014 at 06:16:57AM +0800, Baoquan He wrote: > On 09/05/14 at 10:16am, Kees Cook wrote: > > On Fri, Sep 5, 2014 at 7:08 AM, Baoquan He <bhe@redhat.com> wrote: > > > diff --git a/arch/x86/boot/compressed/misc.c b/arch/x86/boot/compressed/misc.c > > > index 7780a5b..d2a0eaa 100644 > > > --- a/arch/x86/boot/compressed/misc.c > > > +++ b/arch/x86/boot/compressed/misc.c > > > @@ -250,6 +250,11 @@ static void handle_relocations(void *output, unsigned long output_len) > > > } > > > #endif > > > > > > + if (max_addr > CONFIG_RANDOMIZE_BASE_MAX_OFFSET) { > > > + debug_putstr("Random addr is not allowed. No relocation needed... \n"); > > > + return; > > > + } > > > + > > > > It's not clear to me what this is fixing. In aslr.c, > > process_e820_entry() should already make it impossible to select > > max_addr > CONFIG_RANDOMIZE_BASE_MAX_OFFSET. If you're trying to > > detect a non-kaslr boot, I think this is better handled in 1/4 where I > > suggest examining the "output" location before/after > > choose_kernel_location. > > > > -Kees >
[CC hpa ]
> Hi Kees, > > Yes, process_e820_entry() can make sure the choice+output_len < > CONFIG_RANDOMIZE_BASE_MAX_OFFSET, but that can't stop other bootloaders > to put kernel in region above CONFIG_RANDOMIZE_BASE_MAX_OFFSET. > > E.g in kdump, we can set crashkernel=256M@1024M in cmdline. Then the 1st > kernel will reserve 256M memory just at 1024M place. So if load kdump > kernel now, the output will be 1024M before choose_kernel_location(). > With this value, output won't be changed in choose_kernel_location(), > then it will do decompress(), then call handle_relocations(). Then since > 1024 is not equal to LOAD_PHYSICAL_ADDR, it will start relocatoins > handling. And this cause _text stamping into MODULES vaddr range. System > will be exceptional.
Bao,
If you apply your first patch where output_orig == output, then handle_relocations() will not do anything for x86_64 case and bail out. That should take care of this issue. Isn't it? And we should not require this patch.
Thanks Vivek
| |