[PATCH -mm 0/3] ptrace: detach wakeup fixes

[PATCH -mm 0/3] ptrace: detach wakeup fixes

[Oleg Nesterov]

On top of misc ptrace patches in -mm.

Q: when the tracer exits we bypass ptrace_disable() and
clear_tsk_thread_flag(TIF_SYSCALL_TRACE), is this correct?

Perhaps we should move this code into __ptrace_unlink?

Oleg.

Re: [PATCH -mm 0/3] ptrace: detach wakeup fixes

[Roland McGrath]

> Q: when the tracer exits we bypass ptrace_disable() and
> clear_tsk_thread_flag(TIF_SYSCALL_TRACE), is this correct?

It's always been that way.  It's obviously wrong in an abstract sense.  But
for me, ptrace today is purely about compatibility with how it's behaved in
the past, however stupid that was--whenever it's been fully consistent and
predictable, userland is already saddled with coping with the stupidity.

> Perhaps we should move this code into __ptrace_unlink?

Locking nightmare.  ptrace_disable can do stuff that may require the thread
to be stopped like other ptrace operations require (fiddle registers),
might include access_process_vm, etc.  Trust me, we don't want to go there
now.  (In case you were looking for reasons why I'll be soon advocating
reorganization to get ptrace entirely out of the tasklist_lock arena, this
is a fine example of the true horror that the current data structure set-up
gives us.)


Thanks,
Roland

Re: [PATCH -mm 0/3] ptrace: detach wakeup fixes

[Oleg Nesterov]

On 02/08, Roland McGrath wrote:
>
> > Q: when the tracer exits we bypass ptrace_disable() and
> > clear_tsk_thread_flag(TIF_SYSCALL_TRACE), is this correct?
>
> It's always been that way.  It's obviously wrong in an abstract sense.  But
> for me, ptrace today is purely about compatibility with how it's behaved in
> the past, however stupid that was--whenever it's been fully consistent and
> predictable, userland is already saddled with coping with the stupidity.
>
> > Perhaps we should move this code into __ptrace_unlink?
>
> Locking nightmare.  ptrace_disable can do stuff that may require the thread
> to be stopped like other ptrace operations require (fiddle registers),
> might include access_process_vm, etc.  Trust me, we don't want to go there
> now.

OK, thanks.

> (In case you were looking for reasons why I'll be soon advocating
> reorganization to get ptrace entirely out of the tasklist_lock arena, this
> is a fine example of the true horror that the current data structure set-up
> gives us.)

Heh. No, I don't need yet another reason to remove tasklist from ptrace ;)
This would be obviously great.

(btw, I do remember I promised the cleanup, will send a bit later).

Oleg.

Re: [PATCH -mm 0/3] ptrace: detach wakeup fixes

[Oleg Nesterov]

On 02/08, Oleg Nesterov wrote:
>
> On top of misc ptrace patches in -mm.

Andrew, please ignore this series.

I'll resend patches 1 and 3.

Oleg.