<div dir="ltr"><div class="gmail_default" style="font-family:comic sans ms,sans-serif;font-size:xx-large;color:#ff00ff"><br></div><div class="gmail_extra"><br><div class="gmail_quote">On Sat, Jul 9, 2016 at 1:25 AM, Ard Biesheuvel <span dir="ltr"><<a href="mailto:ard.biesheuvel@linaro.org" target="_blank">ard.biesheuvel@linaro.org</a>></span> wrote:<br><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex"><div><div>On 9 July 2016 at 04:22, Laura Abbott <<a href="mailto:labbott@redhat.com" target="_blank">labbott@redhat.com</a>> wrote:<br>
> On 07/06/2016 03:25 PM, Kees Cook wrote:<br>
>><br>
>> Hi,<br>
>><br>
>> This is a start of the mainline port of PAX_USERCOPY[1]. After I started<br>
>> writing tests (now in lkdtm in -next) for Casey's earlier port[2], I<br>
>> kept tweaking things further and further until I ended up with a whole<br>
>> new patch series. To that end, I took Rik's feedback and made a number<br>
>> of other changes and clean-ups as well.<br>
>><br>
>> Based on my understanding, PAX_USERCOPY was designed to catch a few<br>
>> classes of flaws around the use of copy_to_user()/copy_from_user(). These<br>
>> changes don't touch get_user() and put_user(), since these operate on<br>
>> constant sized lengths, and tend to be much less vulnerable. There<br>
>> are effectively three distinct protections in the whole series,<br>
>> each of which I've given a separate CONFIG, though this patch set is<br>
>> only the first of the three intended protections. (Generally speaking,<br>
>> PAX_USERCOPY covers what I'm calling CONFIG_HARDENED_USERCOPY (this) and<br>
>> CONFIG_HARDENED_USERCOPY_WHITELIST (future), and PAX_USERCOPY_SLABS covers<br>
>> CONFIG_HARDENED_USERCOPY_SPLIT_KMALLOC (future).)<br>
>><br>
>> This series, which adds CONFIG_HARDENED_USERCOPY, checks that objects<br>
>> being copied to/from userspace meet certain criteria:<br>
>> - if address is a heap object, the size must not exceed the object's<br>
>> allocated size. (This will catch all kinds of heap overflow flaws.)<br>
>> - if address range is in the current process stack, it must be within the<br>
>> current stack frame (if such checking is possible) or at least entirely<br>
>> within the current process's stack. (This could catch large lengths that<br>
>> would have extended beyond the current process stack, or overflows if<br>
>> their length extends back into the original stack.)<br>
>> - if the address range is part of kernel data, rodata, or bss, allow it.<br>
>> - if address range is page-allocated, that it doesn't span multiple<br>
>> allocations.<br>
>> - if address is within the kernel text, reject it.<br>
>> - everything else is accepted<br>
>><br>
>> The patches in the series are:<br>
>> - The core copy_to/from_user() checks, without the slab object checks:<br>
>> 1- mm: Hardened usercopy<br>
>> - Per-arch enablement of the protection:<br>
>> 2- x86/uaccess: Enable hardened usercopy<br>
>> 3- ARM: uaccess: Enable hardened usercopy<br>
>> 4- arm64/uaccess: Enable hardened usercopy<br>
>> 5- ia64/uaccess: Enable hardened usercopy<br>
>> 6- powerpc/uaccess: Enable hardened usercopy<br>
>> 7- sparc/uaccess: Enable hardened usercopy<br>
>> - The heap allocator implementation of object size checking:<br>
>> 8- mm: SLAB hardened usercopy support<br>
>> 9- mm: SLUB hardened usercopy support<br>
>><br>
>> Some notes:<br>
>><br>
>> - This is expected to apply on top of -next which contains fixes for the<br>
>> position of _etext on both arm and arm64.<br>
>><br>
>> - I couldn't detect a measurable performance change with these features<br>
>> enabled. Kernel build times were unchanged, hackbench was unchanged,<br>
>> etc. I think we could flip this to "on by default" at some point.<br>
>><br>
>> - The SLOB support extracted from grsecurity seems entirely broken. I<br>
>> have no idea what's going on there, I spent my time testing SLAB and<br>
>> SLUB. Having someone else look at SLOB would be nice, but this series<br>
>> doesn't depend on it.<br>
>><br>
>> Additional features that would be nice, but aren't blocking this series:<br>
>><br>
>> - Needs more architecture support for stack frame checking (only x86 now).<br>
>><br>
>><br>
><br>
> Even with the SLUB fixup I'm still seeing this blow up on my arm64 system.<br>
> This is a<br>
> Fedora rawhide kernel + the patches<br>
><br>
> [ 0.666700] usercopy: kernel memory exposure attempt detected from<br>
> fffffc0008b4dd58 (<kernel text>) (8 bytes)<br>
> [ 0.666720] CPU: 2 PID: 79 Comm: modprobe Tainted: G W<br>
> 4.7.0-0.rc6.git1.1.hardenedusercopy.fc25.aarch64 #1<br>
> [ 0.666733] Hardware name: AppliedMicro Mustang/Mustang, BIOS 1.1.0 Nov 24<br>
> 2015<br>
> [ 0.666744] Call trace:<br>
> [ 0.666756] [<fffffc0008088a20>] dump_backtrace+0x0/0x1e8<br>
> [ 0.666765] [<fffffc0008088c2c>] show_stack+0x24/0x30<br>
> [ 0.666775] [<fffffc0008455344>] dump_stack+0xa4/0xe0<br>
> [ 0.666785] [<fffffc000828d874>] __check_object_size+0x6c/0x230<br>
> [ 0.666795] [<fffffc00083a5748>] create_elf_tables+0x74/0x420<br>
> [ 0.666805] [<fffffc00082fb1f0>] load_elf_binary+0x828/0xb70<br>
> [ 0.666814] [<fffffc0008298b4c>] search_binary_handler+0xb4/0x240<br>
> [ 0.666823] [<fffffc0008299864>] do_execveat_common+0x63c/0x950<br>
> [ 0.666832] [<fffffc0008299bb4>] do_execve+0x3c/0x50<br>
> [ 0.666841] [<fffffc00080e3720>] call_usermodehelper_exec_async+0xe8/0x148<br>
> [ 0.666850] [<fffffc0008084a80>] ret_from_fork+0x10/0x50<br>
><br>
> This happens on every call to execve. This seems to be the first<br>
> copy_to_user in<br>
> create_elf_tables. I didn't get a chance to debug and I'm going out of town<br>
> all of next week so all I have is the report unfortunately. config attached.<br>
><br>
<br>
</div></div>This is a known issue, and a fix is already queued for v4.8 in the arm64 tree:<br>
<br>
9fdc14c55c arm64: mm: fix location of _etext [0]<br>
<br>
which moves _etext up in the linker script so that it does not cover .rodata<br>
<br>
ARM was suffering from the same problem, and Kees proposed a fix for<br>
it. I don't know what the status of that patch is, though.<br>
<br>
Note that on arm64, we have<br>
<br>
#define ELF_PLATFORM ("aarch64")<br>
<br>
which explains why k_platform points into .rodata in this case. On<br>
ARM, it points to a writable string (as the code quoted by Rik shows),<br>
so there it will likely explode elsewhere without the linker script<br>
fix.<br>
<br>
[0] <a href="https://git.kernel.org/cgit/linux/kernel/git/arm64/linux.git/commit/?h=for-next/core&id=9fdc14c55c" rel="noreferrer" target="_blank">https://git.kernel.org/cgit/linux/kernel/git/arm64/linux.git/commit/?h=for-next/core&id=9fdc14c55c</a><br>
<span><font color="#888888"><br>
--<br>
Ard.<br>
</font></span></blockquote></div><br></div>Ugh, I completely missed that note about the patch on arm64. Sorry for the noise.<br><br>Thanks,<br>Laura</div>