2021-12-29 19:08:40

by Namhyung Kim

[permalink] [raw]
Subject: Re: Lock problems in linux/tools/perf/util/dso.c

Hello,

On Wed, Dec 22, 2021 at 3:33 AM Ryan Cai <[email protected]> wrote:
>
> Hi, I found a potential lock problem in dso_data_get_fd. Because the inconsistent branch conditions of pthread_mutex_lock(&dso__data_open_lock) and pthread_mutex_unlock(&dso__data_open_lock), it is possible that the lock dso__data_open_lock is not released in the dso_data_get_fd. Also, I have a question on why the branch condition of pthread_mutex_lock(&dso__data_open_lock) is <0. I think that the branch condition should be !=0, because pthread_mutex_lock would return 0 when succeeding. Looking forward to further discussion. One this bug is confirmed, I can send a patch.

Please fix your mail client to wrap around 80 characters.

1. dso__data_get_fd() should be paired with dso__data_put_fd()
when it returns non-negative. It'd unlock the mutex.

2. I've checked the man page and it seems you're right.
It just says that it'd return an error number or zero.

https://linux.die.net/man/3/pthread_mutex_lock

Thanks,
Namhyung


>
>
>
> https://github.com/torvalds/linux/blob/e851dfae4371d3c751f1e18e8eb5eba993de1467/tools/perf/util/dso.c#L708-L722
>
>
>
> Best,
>
> Ryan
>
>
>
>


2021-12-30 02:42:36

by Ryan Cai

[permalink] [raw]
Subject: Re: Lock problems in linux/tools/perf/util/dso.c

Hi, Namhyung,

1. Indeed, I got wrong here.
2. Yes, the branch should be if (pthread_mutex_lock(&dso__data_open_lock))
instead of if (pthread_mutex_lock(&dso__data_open_lock) < 0).

Could I send a patch?

Best,
Ryan

On 30/12/2021, 3:08 AM, "Namhyung Kim" <[email protected]> wrote:

Hello,

On Wed, Dec 22, 2021 at 3:33 AM Ryan Cai <[email protected]> wrote:
>
> Hi, I found a potential lock problem in dso_data_get_fd. Because the inconsistent branch conditions of pthread_mutex_lock(&dso__data_open_lock) and pthread_mutex_unlock(&dso__data_open_lock), it is possible that the lock dso__data_open_lock is not released in the dso_data_get_fd. Also, I have a question on why the branch condition of pthread_mutex_lock(&dso__data_open_lock) is <0. I think that the branch condition should be !=0, because pthread_mutex_lock would return 0 when succeeding. Looking forward to further discussion. One this bug is confirmed, I can send a patch.

Please fix your mail client to wrap around 80 characters.

1. dso__data_get_fd() should be paired with dso__data_put_fd()
when it returns non-negative. It'd unlock the mutex.

2. I've checked the man page and it seems you're right.
It just says that it'd return an error number or zero.

https://linux.die.net/man/3/pthread_mutex_lock

Thanks,
Namhyung


>
>
>
> https://github.com/torvalds/linux/blob/e851dfae4371d3c751f1e18e8eb5eba993de1467/tools/perf/util/dso.c#L708-L722
>
>
>
> Best,
>
> Ryan
>
>
>
>



2022-01-04 19:33:53

by Namhyung Kim

[permalink] [raw]
Subject: Re: Lock problems in linux/tools/perf/util/dso.c

On Wed, Dec 29, 2021 at 6:42 PM Ryan Cai <[email protected]> wrote:
>
> Hi, Namhyung,
>
> 1. Indeed, I got wrong here.
> 2. Yes, the branch should be if (pthread_mutex_lock(&dso__data_open_lock))
> instead of if (pthread_mutex_lock(&dso__data_open_lock) < 0).
>
> Could I send a patch?

Sure!

Thanks,
Namhyung