lkml.org 
[lkml]   [2010]   [Sep]   [10]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: [PATCH] move cred_guard_mutex from task_struct to signal_struct
On 09/10, KOSAKI Motohiro wrote:
>
> 1) moving cread_guard_mutex itself
> - no increase execve overhead
> -> very good
> - it also prevent parallel ptrace

No, it doesn't. Only PTRACE_ATTACH needs this mutex, and as Roland
pointed out it also needs write_lock(tasklist) which is worse. So
this change doesn't make any practical harm for ptrace.

> 2) move in_exec_mm to signal_struct too
> -> very hard. oom-killer can use very few lock because it's called
> from various place. now both ->mm and ->in_exec_mm are protected
> task_lock() and it help to avoid messy.

Yes. But, if ->in_exec_mm is only used by oom_badness(), then I think
you can use task_lock(tsk->group_leader). oom_badness() needs tasklist
anyway, this means it can't race with de_thread() changing the leader.
But up to you.

Another very minor nit (but again, up to you). Perhaps exec_mmap()
could clear ->in_exec_mm (in task_struct or signal_struct, this doesnt
matter), it takes task_lock(current) anyway (and at this point current
is always the group leader).

> Let's move ->cred_guard_mutex from task_struct to signal_struct. It
> naturally prevent multiple-threads-inside-exec.

Reviewed-by: Oleg Nesterov <oleg@redhat.com>


This is very minor, but perhaps you can also fix a couple of comments
which mention task->cred_guard_mutex,

fs/exec.c:1109 the caller must hold current->cred_guard_mutex
kernel/cred.c:328 The caller must hold current->cred_guard_mutex
include/linux/tracehook.h:153 @task->cred_guard_mutex

Oleg.



\
 
 \ /
  Last update: 2010-09-10 19:31    [W:0.073 / U:6.156 seconds]
©2003-2018 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site