Re: [PATCH] kprobes: Forbid probing on kprobe_insn_slot

[Google]

On Wed, 27 Jul 2022 11:20:58 +0800
Chen Zhongjin <chenzhongjin@xxxxxxxxxx> wrote:

> Syzkaller reported a BUG on arm64:
> Unrecoverable kprobe detected.
> Dumping kprobe:
> Name: (null)
> Offset: 0
> Address: 0xffffa00010019000
> ------------[ cut here ]------------
> kernel BUG at arch/arm64/kernel/probes/kprobes.c:235!
> Internal error: Oops - BUG: 0 [#1] SMP
> Modules linked in:
> CPU: 1 PID: 31060 Comm: syz-executor.6 Not tainted 5.10.0 #11
> ...
> Call trace:
>  reenter_kprobe arch/arm64/kernel/probes/kprobes.c:234 [inline]
>  kprobe_handler+0x23c/0x26c arch/arm64/kernel/probes/kprobes.c:339
>  kprobe_breakpoint_handler+0x24/0x34 arch/arm64/kernel/probes/kprobes.c:406
>  call_break_hook+0xf4/0x13c arch/arm64/kernel/debug-monitors.c:322
>  brk_handler+0x2c/0xa0 arch/arm64/kernel/debug-monitors.c:329
>  do_debug_exception+0x140/0x230 arch/arm64/mm/fault.c:867
>  el1_dbg+0x38/0x50 arch/arm64/kernel/entry-common.c:182
>  el1_sync_handler+0xf4/0x150 arch/arm64/kernel/entry-common.c:219
>  el1_sync+0x74/0x100 arch/arm64/kernel/entry.S:665
>  0xffffa00010019000
>  do_futex+0x2f4/0x370 kernel/futex.c:3735
>  __do_sys_futex kernel/futex.c:3798 [inline]
>  __se_sys_futex kernel/futex.c:3764 [inline]
>  __arm64_sys_futex+0x168/0x3a0 kernel/futex.c:3764
>  __invoke_syscall arch/arm64/kernel/syscall.c:36 [inline]
>  invoke_syscall arch/arm64/kernel/syscall.c:48 [inline]
>  el0_svc_common.constprop.0+0xf4/0x414 arch/arm64/kernel/syscall.c:155
>  do_el0_svc+0x50/0x11c arch/arm64/kernel/syscall.c:217
>  el0_svc+0x20/0x30 arch/arm64/kernel/entry-common.c:353
>  el0_sync_handler+0xe4/0x1e0 arch/arm64/kernel/entry-common.c:369
>  el0_sync+0x148/0x180 arch/arm64/kernel/entry.S:683
> Code: 91018360 97ff1838 aa1703e0 97ff1fdf (d4210000)
> ---[ end trace 767503e946e01b15 ]---
> 
> Syzbot tried to porbe on a kprobe_insn_slot.
> 
> kprobe will replace instruciton with a break and store the origin one
> on kprobe_insn_slot. However these slots are not in .kprobes.text and
> exported by perf_event_ksymbol so can be probed by perf interface.
> 
> Probing these slots will triggers kprobe handler inside single step
> process and for some architectures such as arm64 this will causes a
> bug().
> 
> These slots are kprobe process so they should not be probed anyway.
> Add kprobe_insn_slot check when register_kprobe to forbid probing on
> these slots.

Oops, good catch!

Previously this was not counted as text area, but now the kernel_text_address()
returns true for these trampoline buffers.

In this case, I think kprobes should be limited to probe only
core_kernel_text and module_text.
Can you use is_module_text_address() and core_kernel_text() instead?
Below can allow kprobes probing on other trampolines like ftrace and
bpf.
Also, you may need this tag;

Fixes: 5b485629ba0d ("kprobes, extable: Identify kprobes trampolines as kernel text area")

Thank you,

> 
> Signed-off-by: Chen Zhongjin <chenzhongjin@xxxxxxxxxx>
> ---
>  kernel/kprobes.c | 2 ++
>  1 file changed, 2 insertions(+)
> 
> diff --git a/kernel/kprobes.c b/kernel/kprobes.c
> index f214f8c088ed..3e798b62db70 100644
> --- a/kernel/kprobes.c
> +++ b/kernel/kprobes.c
> @@ -1562,6 +1562,8 @@ static int check_kprobe_address_safe(struct kprobe *p,
>  	/* Ensure it is not in reserved area nor out of text */
>  	if (!kernel_text_address((unsigned long) p->addr) ||
>  	    within_kprobe_blacklist((unsigned long) p->addr) ||
> +	    is_kprobe_insn_slot((unsigned long) p->addr) ||
> +	    is_kprobe_optinsn_slot((unsigned long) p->addr) ||
>  	    jump_label_text_reserved(p->addr, p->addr) ||
>  	    static_call_text_reserved(p->addr, p->addr) ||
>  	    find_bug((unsigned long)p->addr)) {
> -- 
> 2.17.1
> 


-- 
Masami Hiramatsu (Google) <mhiramat@xxxxxxxxxx>