[lkml]   [2013]   [Nov]   [18]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
Messages in this thread
SubjectRe: [PATCH] tty: Only hangup once
On 11/18/2013 08:42 AM, One Thousand Gnomes wrote:
>> After upgrading to kernel 3.12 I noticed one issue with tmux software.
>> The easiest way to reproduce will be:
>> 1. Start tmux session as root.
>> 2. Connect via ssh and use "tmux attach" to attach to the running
>> session.
>> 3. Kill ssh client.


Thanks for the report.

>> You can notice that shell (zsh in my case) and "tmux attach" are still
>> remains in process' list. That didn't happen in previous kernels.

This may have been a bug in previous kernels.
The tmux(1) man page has this to say:

Each session is persistent and will survive accidental disconnection
(such as ssh(1) connection timeout) or intentional detaching (with the
`C-b d' key strokes). tmux may be reattached using:

$ tmux attach

I'll confirm with tmux author(s) what the intended behavior is.

>> I've tried to bisect this in kernel sources and found commit
>> cb50e5235b8ae5aa0fe422eaaa8e444024c5bd98 which contains this exact
>> patch. I have not enough experience to investigate more so most likely
>> I will not find anything more. But it will be good if someone more
>> experienced will have a look at it.
> The patch should be reverted. The submission gives no reason that the
> patch was required - it just adds code and optimises a path that doesn't
> need optimising anyway.


This patch isn't about optimizing; it's about guaranteeing the line discipline
and a tty driver that ops->hangup() will only occur once for any given tty.

> It's theoretically true you only need one hangup, unfortunately however
> I think it has to be the *last* hangup not the first or there are races
> between the tty code and the process group handling.

I doubt this is caused by a race condition; the first hangup would do most
of the destruction regardless, and a second hangup can't really race with
the first because of the tty_lock() held for most of the hangup.

In any event, it's worth discovering what state a subsequent hangup can
effect that the first hangup left incomplete. I'll look into it.

Peter Hurley

 \ /
  Last update: 2013-11-18 18:41    [W:0.130 / U:0.720 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site