Message-ID: <CAMj1kXHo1Wmme2eAaxjeOoWKeAkG98ZbMYWGwwja4H2dCHNiSA@mail.gmail.com>
Date:   Fri, 22 Oct 2021 10:07:51 +0200
From:   Ard Biesheuvel <ardb@...nel.org>
To:     Kees Cook <keescook@...omium.org>
Cc:     Linux ARM <linux-arm-kernel@...ts.infradead.org>,
        Nick Desaulniers <ndesaulniers@...gle.com>,
        linux-hardening@...r.kernel.org
Subject: Re: [PATCH] ARM: stackprotector: prefer compiler for TLS based
 per-task protector

On Fri, 22 Oct 2021 at 08:11, Kees Cook <keescook@...omium.org> wrote:
>
> On Thu, Oct 21, 2021 at 04:25:16PM +0200, Ard Biesheuvel wrote:
> > Currently, we implement the per-task stack protector for ARM using a GCC
> > plugin, due to lack of native compiler support. However, work is
> > underway to get this implemented in the compiler, which means we will be
> > able to deprecate the GCC plugin at some point.
> >
> > In the meantime, we will need to support both, where the native compiler
> > implementation is obviously preferred. So let's wire this up in Kconfig
> > and the Makefile.
> >
> > Cc: Kees Cook <keescook@...omium.org>
> > Cc: Nick Desaulniers <ndesaulniers@...gle.com>
> > Signed-off-by: Ard Biesheuvel <ardb@...nel.org>
>
> Awesome! I built upstream GCC with the corresponding feature[1], but my
> qemu explodes:
>

Ugh, this is what I hit with v1 and thought I fixed in v2. For some
reason, this only happens in ARM mode with SMP, and I tested Thumb2
with SMP but not ARM. The reason is that the fact that the canary
comparison clobbers the condition flags is not propagated correctly:

   0xc04923a8 <+168>: ldr r4, [sp, #20]
   0xc04923ac <+172>: ldr lr, [r5, #1256] ; 0x4e8
   0xc04923b0 <+176>: eors lr, r4, lr
   0xc04923b4 <+180>: mov r4, #0
    <gap 1>
   0xc0492400 <+256>: bne 0xc0492414 <si_mem_available+276>
    <gap 2>
   0xc0492414 <+276>: bl 0xc102c278 <__stack_chk_fail>

Loads the two values, XORs them and sets the Z flag if they are equal,
and clears the other register as well. Finally, it performs a
conditional jump to the failure path, which calls __stack_chk_fail().

The issue is that gap 1 must not touch the flags, and for some reason,
this is not being expresse correctly in the RTL.


But I will take it as a tested-by for *this* patch :-)


> smp: Bringing up secondary CPUs ...
> Kernel panic - not syncing: stack-protector: Kernel stack is corrupted in: si_mem_available+0x110/0x110
> CPU: 0 PID: 1 Comm: swapper/0 Not tainted 5.15.0-rc6-next-20211021+ #791
> Hardware name: Generic DT based system
> Backtrace:
> [<809df62c>] (dump_backtrace) from [<809dfa20>] (show_stack+0x20/0x24)
>  r7:80bc2838 r6:00000080 r5:80bd39fc r4:600000d3
> [<809dfa00>] (show_stack) from [<809e8580>] (dump_stack_lvl+0x60/0x78)
> [<809e8520>] (dump_stack_lvl) from [<809e85b0>] (dump_stack+0x18/0x1c)
>  r7:80bc2838 r6:818ef000 r5:00000000 r4:80ebea20
> [<809e8598>] (dump_stack) from [<809dff88>] (panic+0x118/0x34c)
> [<809dfe70>] (panic) from [<809edcac>] (lockdep_hardirqs_on+0x0/0x1a4)
>  r3:00000002 r2:00000005 r1:802e4e34 r0:80bc2838
>  r7:00000002
> [<809edc90>] (__stack_chk_fail) from [<802e4e34>] (__drain_all_pages+0x0/0x260)
> [<802e4d24>] (si_mem_available) from [<8022d5ec>] (__rb_allocate_pages+0x30/0x224)
>  r6:81905440 r5:81921dcc r4:00000002
> [<8022d5bc>] (__rb_allocate_pages) from [<8022f790>] (rb_allocate_cpu_buffer+0x1e4/0x2c8)
>  r10:00000002 r9:8180c300 r8:818ef000 r7:00000002 r6:81905440 r5:00000000
>  r4:818df200
> [<8022f5ac>] (rb_allocate_cpu_buffer) from [<80232fc0>] (trace_rb_cpu_prepare+0x8c/0xec)
>  r9:8180c30c r8:8180c364 r7:00000002 r6:81805c00 r5:00000001 r4:00000000
> [<80232f34>] (trace_rb_cpu_prepare) from [<801256fc>] (cpuhp_invoke_callback+0x278/0x4ec)
>  r10:80232f34 r9:ef1e43a4 r8:80e100c8 r7:0000003d r6:8180c364 r5:00000001
>  r4:00000000
> [<80125484>] (cpuhp_invoke_callback) from [<801259e4>] (cpuhp_invoke_callback_range+0x74/0xb4)
>  r10:ef1e43a4 r9:00000000 r8:00000001 r7:80e0fc04 r6:0000005a r5:00000001
>  r4:ef1e43a4
> [<80125970>] (cpuhp_invoke_callback_range) from [<8012774c>] (_cpu_up+0x128/0x2b0)
>  r8:6e470000 r7:000000e4 r6:80e08fd8 r5:00000001 r4:80d743a4
> [<80127624>] (_cpu_up) from [<80127940>] (cpu_up+0x6c/0xa0)
>  r10:818efdc4 r9:00000001 r8:80e08f14 r7:00000008 r6:80e08fd8 r5:000000e4
>  r4:00000001
> [<801278d4>] (cpu_up) from [<801280a8>] (bringup_nonboot_cpus+0x78/0x7c)
>  r5:80e08f0c r4:00000001
> [<80128030>] (bringup_nonboot_cpus) from [<80d10e94>] (smp_init+0x3c/0x84)
>  r9:00000001 r8:00000000 r7:818ef6e0 r6:80d733e4 r5:80d733e4 r4:80e2c028
> [<80d10e58>] (smp_init) from [<80d014c8>] (kernel_init_freeable+0x1c0/0x324)
>  r4:818f0880
> [<80d01308>] (kernel_init_freeable) from [<809eecbc>] (kernel_init+0x28/0x148)
>  r10:00000000 r9:00000000 r8:00000000 r7:00000000 r6:00000000 r5:809eec94
>  r4:80e08ec0
> [<809eec94>] (kernel_init) from [<80100108>] (ret_from_fork+0x14/0x2c)
> Exception stack(0x81921fb0 to 0x81921ff8)
> 1fa0:                                     00000000 00000000 00000000 00000000
> 1fc0: 00000000 00000000 00000000 00000000 00000000 00000000 00000000 00000000
> 1fe0: 00000000 00000000 00000000 00000000 00000013 00000000
>  r5:809eec94 r4:00000000
> ---[ end Kernel panic - not syncing: stack-protector: Kernel stack is corrupted in: si_mem_available+0x110/0x110 ]---
>
> Using qemu like so:
>
> qemu-system-arm \
>         -M virt \
>         -cpu cortex-a15 \
>         -smp 2 \
>         -nographic \
>         -m 2048 \
>         -kernel "$kernel" \
>         -drive file=vda.raw,id=hd0,format=raw,if=none \
>         -device virtio-blk-device,drive=hd0 \
>         -serial stdio \
>         -append "root=/dev/vda1 earlycon loglevel=8 console=ttyAMA0 $@"
>
> I built against -next, does this require the unwinder changes?
>
> (FWIW, I built with a patched 11.2 GCC.)
>
> -Kees
>
> [1] https://lore.kernel.org/linux-hardening/20211021165119.2136543-2-ardb@kernel.org/T/#u
>
> --
> Kees Cook

Powered by blists - more mailing lists