Messages in this thread | | | Date | Tue, 17 May 2011 15:56:24 -0700 (PDT) | From | David Rientjes <> | Subject | Re: [PATCH 1/3] comm: Introduce comm_lock spinlock to protect task->comm access |
| |
On Tue, 17 May 2011, Peter Zijlstra wrote:
> The changelog also fails to mention _WHY_ this is no longer true. Nor > does it treat why making it true again isn't an option. >
It's been true since:
4614a696bd1c3a9af3a08f0e5874830a85b889d4 Author: john stultz <johnstul@us.ibm.com> Date: Mon Dec 14 18:00:05 2009 -0800
procfs: allow threads to rename siblings via /proc/pid/tasks/tid/comm
Although at the time it appears that nobody was concerned about races so proper syncronization was never implemented. We always had the prctl(PR_SET_NAME) so the majority of comm reads, those to current, required no locking, but this commit changed that. The remainder of comm dereferences always required task_lock() and the helper get_task_comm() to read the string into a (usually stack-allocated) buffer.
> Who is changing another task's comm? That's just silly. >
I agree, and I suggested taking write privileges away from /proc/pid/comm, but others find that it is useful to be able to differentiate between threads in the same thread group without using the prctl() for debugging?
| |