Re: [PATCH] kprobes: Update ftrace_ops when clearing ftrace-based aggrprobe's post_handler

[Li Huafei]

On 2022/11/11 23:33, Masami Hiramatsu (Google) wrote:
> On Fri, 11 Nov 2022 18:10:06 +0800
> Li Huafei <lihuafei1@xxxxxxxxxx> wrote:
> 
>> In __unregister_kprobe_top(), if the currently unregistered probe has
>> post_handler but other child probes of the aggrprobe do not have
>> post_handler, the post_handler of the aggrprobe is cleared. If this is
>> a ftrace-based probe, there is a problem. In later calls to
>> disarm_kprobe(), we will use kprobe_ftrace_ops because post_handler is
>> NULL. But we're armed with kprobe_ipmodify_ops. This triggers a WARN in
>> __disarm_kprobe_ftrace() and may even cause use-after-free:
>>
>>   Failed to disarm kprobe-ftrace at kernel_clone+0x0/0x3c0 (error -2)
>>   WARNING: CPU: 5 PID: 137 at kernel/kprobes.c:1135 __disarm_kprobe_ftrace.isra.21+0xcf/0xe0
>>   Modules linked in: testKprobe_007(-)
>>   CPU: 5 PID: 137 Comm: rmmod Not tainted 6.1.0-rc4-dirty #18
>>   [...]
>>   Call Trace:
>>    <TASK>
>>    __disable_kprobe+0xcd/0xe0
>>    __unregister_kprobe_top+0x12/0x150
>>    ? mutex_lock+0xe/0x30
>>    unregister_kprobes.part.23+0x31/0xa0
>>    unregister_kprobe+0x32/0x40
>>    __x64_sys_delete_module+0x15e/0x260
>>    ? do_user_addr_fault+0x2cd/0x6b0
>>    do_syscall_64+0x3a/0x90
>>    entry_SYSCALL_64_after_hwframe+0x63/0xcd
>>    [...]
> 
> Ah, good catch! :D
> 
>>
>> For ftrace kprobe, update post_handler at the same time update
>> ftrace_ops, moving it from kprobe_ipmodify_ops to kprobe_ftrace_ops.
> 
> Hmm, but I would not like this because there can be a time
> window when it can miss an event. What about just skipping
> clearing ap->post_handler in kprobe-on-ftrace case?
> 

Agree. I hadn't considered this time window. The effects I see if I keep
ap->handler are 1) kprobe_ftrace_handler() still needs to call
aggr_post_handler() and 2) other ftrace_ops still can't set IPMODIFY on
the probe function. This doesn't seem to be a problem.

Thanks for the suggestion.

>>
>> Fixes: 0bc11ed5ab60 ("kprobes: Allow kprobes coexist with livepatch")
>> Reported-by: Zhao Gongyi <zhaogongyi@xxxxxxxxxx>
>> Signed-off-by: Li Huafei <lihuafei1@xxxxxxxxxx>
>> ---
>>  kernel/kprobes.c | 12 +++++++++++-
>>  1 file changed, 11 insertions(+), 1 deletion(-)
>>
>> diff --git a/kernel/kprobes.c b/kernel/kprobes.c
>> index cd9f5a66a690..f8bec48a9cf9 100644
>> --- a/kernel/kprobes.c
>> +++ b/kernel/kprobes.c
>> @@ -1766,7 +1766,17 @@ static int __unregister_kprobe_top(struct kprobe *p)
>>  				if ((list_p != p) && (list_p->post_handler))
>>  					goto noclean;
>>  			}
>> -			ap->post_handler = NULL;
>> +			/*
>> +			 * For ftrace kprobe, we need to update ftrace_ops
>> +			 * at the same time as we update post_handler, moving
>> +			 * it from kprobe_ipmodify_ops to kprobe_ftrace_ops.
>> +			 */
>> +			if (unlikely(kprobe_ftrace(ap))) {
>> +				disarm_kprobe(ap, false);
>> +				ap->post_handler = NULL;
>> +				arm_kprobe(ap);
>> +			} else
> 
> So here, just add; (also, don't use unlikely/likely for this case, this
> depends on where the user probes, not a systematically rare case.)
> 

Okay. Will fix it in the next version. Thanks!

> 			if (!kprobe_ftrace(ap))
> 
>> +				ap->post_handler = NULL;
> 
> Thank you!
> 
>>  		}
>>  noclean:
>>  		/*
>> -- 
>> 2.17.1
>>
> 
>