Re: Question: livepatch failed for new fork() task stack unreliable

[Wangshaobo (bobo)]

å 2020/6/2 21:14, Josh Poimboeuf åé:
> On Tue, Jun 02, 2020 at 09:22:30AM +0800, Wangshaobo (bobo) wrote:
> > so i think this question is related to ORC unwinder, could i ask if you have
> > strategy or plan to avoid this problem ?
> I suspect something like this would fix it (untested):
>
> diff --git a/arch/x86/kernel/stacktrace.c b/arch/x86/kernel/stacktrace.c
> index 6ad43fc44556..8cf95ded1410 100644
> --- a/arch/x86/kernel/stacktrace.c
> +++ b/arch/x86/kernel/stacktrace.c
> @@ -50,7 +50,7 @@ int arch_stack_walk_reliable(stack_trace_consume_fn consume_entry,
>   		if (regs) {
>   			/* Success path for user tasks */
>   			if (user_mode(regs))
> -				return 0;
> +				break;
>   
>   			/*
>   			 * Kernel mode registers on the stack indicate an
> @@ -81,10 +81,6 @@ int arch_stack_walk_reliable(stack_trace_consume_fn consume_entry,
>   	if (unwind_error(&state))
>   		return -EINVAL;
>   
> -	/* Success path for non-user tasks, i.e. kthreads and idle tasks */
> -	if (!(task->flags & (PF_KTHREAD | PF_IDLE)))
> -		return -EINVAL;
> -
>   	return 0;
>   }
>   
> diff --git a/arch/x86/kernel/unwind_orc.c b/arch/x86/kernel/unwind_orc.c
> index 7f969b2d240f..d7396431261a 100644
> --- a/arch/x86/kernel/unwind_orc.c
> +++ b/arch/x86/kernel/unwind_orc.c
> @@ -540,7 +540,7 @@ bool unwind_next_frame(struct unwind_state *state)
>   		state->sp = sp;
>   		state->regs = NULL;
>   		state->prev_regs = NULL;
> -		state->signal = false;
> +		state->signal = ((void *)state->ip == ret_from_fork);
>   		break;
>   
>   	case ORC_TYPE_REGS:

what a awesome job, thanks a lot, Josh

Today I test your fix, but arch_stack_walk_reliable() still return 
failed sometimes, I

found one of three scenarios mentioned failed:


1. user task (just fork) but not been scheduled

ÂÂÂ test FAILED

ÂÂÂ it is because unwind_next_frame() get the first frame, this time 
state->signal is false, and then return

ÂÂÂ failed in the same place for ret_from_fork has not executed at all.


2. user task (just fork) start excuting ret_from_fork() till 
schedule_tail but not UNWIND_HINT_REGS

 test condition :loop fork() in current system

ÂÂÂ result : SUCCESS,

ÂÂÂ it looks like this modification works for my perspective :

	-	/* Success path for non-user tasks, i.e. kthreads and idle tasks */
	-	if (!(task->flags & (PF_KTHREAD | PF_IDLE)))
	-		return -EINVAL;
  but is this possible to miss one invalid judgement condition ? (1)

3. call_usermodehelper_exec_async

ÂÂÂ test condition :loop call call_usermodehelper() in a module selfmade.

ÂÂÂ result : SUCCESS,

ÂÂ it looks state->signal==true works when unwind_next_frame() gets the 
end ret_from_fork() frame,

ÂÂ but i don't know how does it work, i am confused by this sentences, 
how does the comment means sibling calls and

ÂÂÂ calls to noreturn functions? (2)

ÂÂÂ ÂÂÂ ÂÂÂ /*
ÂÂÂÂ ÂÂÂ ÂÂÂ * Find the orc_entry associated with the text address.
ÂÂÂÂ ÂÂÂ ÂÂÂ *
 Â Â Â * Decrement call return addresses by one so they work for 
sibling
ÂÂÂÂ ÂÂÂ ÂÂÂ * calls and calls to noreturn functions.
ÂÂÂÂ ÂÂÂ ÂÂÂ */
ÂÂ ÂÂ ÂÂÂ Â orc = orc_find(state->signal ? state->ip : state->ip - 1);
ÂÂ ÂÂ ÂÂÂ Â if (!orc) {


So i slightly modify your code, i move state->signal = ((void 
*)state->ip == ret_from_fork) to unwind_start()

and render unwind_next_frame() remain the same as before:

diff --git a/arch/x86/kernel/unwind_orc.c b/arch/x86/kernel/unwind_orc.c
index e9cc182aa97e..ecce5051e8fd 100644
--- a/arch/x86/kernel/unwind_orc.c
+++ b/arch/x86/kernel/unwind_orc.c
@@ -620,6 +620,7 @@ void __unwind_start(struct unwind_state *state, 
struct task_struct *task,
ÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂ state->sp = task->thread.sp;
ÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂ state->bp = READ_ONCE_NOCHECK(frame->bp);
ÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂ state->ip = READ_ONCE_NOCHECK(frame->ret_addr);
+ÂÂÂÂÂÂÂÂÂÂÂÂÂ state->signal = ((void *)state->ip == ret_from_fork);
ÂÂÂÂÂÂÂ }

diff --git a/arch/x86/kernel/unwind_orc.c b/arch/x86/kernel/unwind_orc.c
index 7f969b2d240f..d7396431261a 100644
--- a/arch/x86/kernel/unwind_orc.c
+++ b/arch/x86/kernel/unwind_orc.c
@@ -540,7 +540,7 @@ bool unwind_next_frame(struct unwind_state *state)
 		state->sp = sp;
 		state->regs = NULL;
 		state->prev_regs = NULL;
-		state->signal = ((void *)state->ip == ret_from_fork);
+		state->signal = false;
Â		break;


After modification all the three scenarios are captured and no longer 
return failed, but i don't know

how does it affect the scenarios 3, because current frame->ret_addr(the 
first frame) is not ret_from_fork,

it should return failed as scenarios1, but it didn't , i really want to 
know the reason. (3)


thanks again

Wang ShaoBo