lkml.org 
[lkml]   [2011]   [Mar]   [31]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: [PATCH 3/3] signal, ptrace: Fix delayed CONTINUED notification when ptraced
On 03/31, Tejun Heo wrote:
>
> Hello,
>
> On Thu, Mar 31, 2011 at 05:15:49PM +0200, Oleg Nesterov wrote:
> >
> > The comment says:
> >
> > * If there is a handler for SIGCONT, we must make
> > * sure that no thread returns to user mode before
> > * we post the signal
>
> I interpreted it as "when there's only single thread, it should not
> return to userland before executing the signal handler".

Yes... probably.

> > rm_from_queue(SIG_KERNEL_STOP_MASK, &signal->shared_pending);
> > t = p;
> > do {
> > - unsigned int state;
> > rm_from_queue(SIG_KERNEL_STOP_MASK, &t->pending);
> > - /*
> > - * If there is a handler for SIGCONT, we must make
> > - * sure that no thread returns to user mode before
> > - * we post the signal, in case it was the only
> > - * thread eligible to run the signal handler--then
> > - * it must not do anything between resuming and
> > - * running the handler. With the TIF_SIGPENDING
> > - * flag set, the thread will pause and acquire the
> > - * siglock that we hold now and until we've queued
> > - * the pending signal.
> > - *
> > - * Wake up the stopped thread _after_ setting
> > - * TIF_SIGPENDING
> > - */
> > - state = __TASK_STOPPED;
> > - if (sig_user_defined(t, SIGCONT) && !sigismember(&t->blocked, SIGCONT)) {
> > - set_tsk_thread_flag(t, TIF_SIGPENDING);
> > - state |= TASK_INTERRUPTIBLE;
> > - }
> > - wake_up_state(t, state);
> > + wake_up_state(t, __TASK_STOPPED);
> > } while_each_thread(p, t);
>
> This is awesome and, at the first glance, yeah, this seems to be the
> right thing to do. That part is pure signal delivery after all.

Great.

> * As wants_signal() doesn't take uninterruptible sleeps into
> consideration,

Yes! And I already thought about this before (regardless of this change).
This is not really good imho, we can improve the ->curr_target logic a
bit, this looks simple.

> the signal might get delivered later with the change
> but I don't think it's problematic in any way.

Agreed,

> * Interruptible sleeps won't be disturbed on SIGCONT generation, which
> is a visible behavior change, but, I agree, this is more of a bug
> fix.

Yes, agreed. I'll try to make the test-case which shows the difference.

> I'll mull over it a bit more but please go ahead and create a proper
> patch.

Yes, will do tomorrow (and it needs the trivial re-diff against your tree).

I spent too many time today trying to understand what was the original
reason for this code. Looks like, it could die a loooooong ago. Perhaps
the only reason was: handle_stop_signal() could drop ->siglock to notify
the parent. I am not sure, that is why I am a bit nervous and want to
recheck once again.

> I'll apply it to the ptrace branch with the previous two
> patches. (Can I add your Acked-by's there?)

Yes, thanks Tejun.

Oleg.



\
 
 \ /
  Last update: 2011-03-31 19:33    [W:0.183 / U:0.312 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site