Re: [External] Re: [PATCH 2/2] ftrace: setup correct flags before replace code of module rec

[Chengming Zhou <zhouchengming@bytedance.com>]

在 2020/7/28 下午9:02, Steven Rostedt 写道:
> On Tue, 28 Jul 2020 18:27:20 +0800
> Chengming Zhou <zhouchengming@bytedance.com> wrote:
>
>> When module loaded and enabled, we will use __ftrace_replace_code
>> for module if any ftrace_ops referenced it found. But we will get
>> wrong ftrace_addr for module rec in ftrace_get_addr_new, because
>> rec->flags has not been setup correctly.
>> So setup correct rec->flags when we call referenced_filters to find
>> ftrace_ops references it.
> This is somewhat correct ;-)
>
>> Signed-off-by: Chengming Zhou <zhouchengming@bytedance.com>
>> Signed-off-by: Muchun Song <songmuchun@bytedance.com>
>> ---
>>  kernel/trace/ftrace.c | 16 +++++++++++++---
>>  1 file changed, 13 insertions(+), 3 deletions(-)
>>
>> diff --git a/kernel/trace/ftrace.c b/kernel/trace/ftrace.c
>> index fca01a168ae5..00087dea0174 100644
>> --- a/kernel/trace/ftrace.c
>> +++ b/kernel/trace/ftrace.c
>> @@ -6190,8 +6190,17 @@ static int referenced_filters(struct dyn_ftrace *rec)
>>  	int cnt = 0;
>>  
>>  	for (ops = ftrace_ops_list; ops != &ftrace_list_end; ops = ops->next) {
>> -		if (ops_references_rec(ops, rec))
>> -		    cnt++;
>> +		if (ops_references_rec(ops, rec)) {
>> +			cnt++;
>> +			if (ops->flags & FTRACE_OPS_FL_DIRECT)
>> +				rec->flags |= FTRACE_FL_DIRECT;
> The above should be:
>
> 			if (WARN_ON_ONCE(ops->flags & FTRACE_OPS_FL_DIRECT))
> 				continue;
> 			cnt++;
>
> The direct flag is *very* special, and should not be set automatically
> like this.
>
> Probably should add the same kind of warning and skip for
> FTRACE_OPS_FL_IPMODIFY.
Ok, I think it's fine to warn and skip these ops.
>> +			if (ops->flags & FTRACE_OPS_FL_SAVE_REGS)
>> +				rec->flags |= FTRACE_FL_REGS;
> The above is definitely a bug fix. I'm thinking this patch should be
> broken up into two. One with just this update (and the clear below),
> and the rest later. As this should be backported to stable.

Yes, this bug cause a kernel crash on our server...

So I will send a bugfix patch just including this update and the clear
below.

>> +			if (cnt == 1 && ops->trampoline)
>> +				rec->flags |= FTRACE_FL_TRAMP;
>> +			else
>> +				rec->flags &= ~FTRACE_FL_TRAMP;
> The above is correct, but not critical that it would need to be
> backported.

I will put the rest in the second patch later.

Thanks!

>
> Thanks!
>
> -- Steve
>
>> +		}
>>  	}
>>  
>>  	return cnt;
>> @@ -6373,7 +6382,8 @@ void ftrace_module_enable(struct module *mod)
>>  			cnt += referenced_filters(rec);
>>  
>>  		/* This clears FTRACE_FL_DISABLED */
>> -		rec->flags = cnt;
>> +		rec->flags &= ~FTRACE_FL_DISABLED;
>> +		rec->flags += cnt;
>>  
>>  		if (ftrace_start_up && cnt) {
>>  			int failed = __ftrace_replace_code(rec, 1);