Re: [RFC PATCH 3/6] bpf: Save the pointer to struct perf_event to map

["Wangnan (F)" <wangnan0@huawei.com>]

On 2015/7/17 19:40, Peter Zijlstra wrote:
> On Fri, Jul 17, 2015 at 07:34:22PM +0800, Wangnan (F) wrote:
>> On 2015/7/17 19:21, Wangnan (F) wrote:
>>> On 2015/7/17 19:06, Peter Zijlstra wrote:
>>>> On Fri, Jul 17, 2015 at 06:43:33PM +0800, kaixu xia wrote:
>>>>> diff --git a/kernel/events/core.c b/kernel/events/core.c
>>>>> index e965cfa..c4e34b7 100644
>>>>> --- a/kernel/events/core.c
>>>>> +++ b/kernel/events/core.c
>>>>> @@ -8582,6 +8582,28 @@ void perf_event_delayed_put(struct task_struct
>>>>> *task)
>>>>>           WARN_ON_ONCE(task->perf_event_ctxp[ctxn]);
>>>>>   }
>>>>>   +struct perf_event *perf_event_get(unsigned int fd)
>>>>> +{
>>>>> +    struct perf_event *event;
>>>>> +    struct fd f;
>>>>> +
>>>>> +    f = fdget(fd);
>>>>> +
>>>>> +    if (!f.file)
>>>>> +        return ERR_PTR(-EBADF);
>>>>> +
>>>>> +    if (f.file->f_op != &perf_fops) {
>>>>> +        fdput(f);
>>>>> +        return ERR_PTR(-EINVAL);
>>>>> +    }
>>>>> +
>>>>> +    event = f.file->private_data;
>>>>> +
>>>>> +    fdput(f);
>>>>> +
>>>>> +    return event;
>>>>> +}
>>>> And what is stopping userspace from closing those FDs while you're using
>>>> them?
>> Shall we put atomic_long_inc_not_zero() between fdget() and fdput()?
> You pretty much _have_ to do that.

Thanks. In next version we will introduce a new function which do 
oppsite thing to
perf_event_release_kernel() in perf/event/core.c, then fetch the event 
before fdput.

Thank you.