Re: [PATCH 00/12] Rework tty_reopen()

[Peter Hurley <peter@hurleysoftware.com>]

Hi Greg,

This series has been reported to fix a regression with Redhat's kdump
systemd service redirecting to /dev/console, when /dev/console is a
serial port.

The redirection consistently fails with EIO since
"tty: Remove tty_wait_until_sent_from_close", which is new to 4.4-rc
Prior to that patch, redirection would only occasionally fail with EIO. :)

[  The systemd repeated hangup of /dev/console also seems to be the
[  trigger for the serial driver crashes on hangup as well, which is
[  fixed by the 19-patch "Fix driver crashes on hangup" series.
[  That problem goes back to 3.10, but has only been reported recently,
[  which leads me to believe recent changes in systemd /dev/console
[  handling is a contributing factor (which I'm checking right now)

Here are what I think are the options to resolve the regression:

#1. Respin this series w/o the tty-next dependencies
#2. Split this series into the minimum necessary to fix the regression
#3. Revert from 4.4-rc (in revert order)
      "tty: Remove wait_event_interruptible_tty()"
      "tty: r3964: Replace/remove bogus tty lock use"
      "tty: r3964: Use tty->read_wait waitqueue"
      "tty: Remove tty_port::close_wait"
      "usb: gadget: gserial: Privatize close_wait"
      "tty: Remove ASYNC_CLOSING check in open()/hangup() methods"
      "tty: Remove tty_wait_until_sent_from_close()"

Let me know how you'd like me to handle this.

Sorry,
Peter Hurley


On 11/27/2015 06:25 PM, Peter Hurley wrote:
> This patch series implements two important improvements to tty open()
> behavior: interruptible open() and automatic retry when tty teardown
> has already commenced.
> 
> Interruptible open() allows signals to cancel the open wait if stalled
> waiting for tty teardown to complete.
> 
> Automatic retry of tty open() when racing a tty teardown now makes tty
> open() fully POSIX compliant. For some time, the Linux kernel has
> returned EIO from open() under certain circumstances. This happens when
> tty_open() observes a valid tty from driver lookup but the tty is
> being released (in final close) and teardown is about to commence.
> 
> The observable userspace change is that userspace will no longer need
> to retry open() on EIO error.
> 
> This series also continues the ongoing effort to cleanup and reduce the
> kernel tty interface.
> 
> Lastly, this series lays important groundwork for implementing ptmx_open()
> in tty_open(), trivially with driver lookup (still a work-in-progress).
> 
> Regards,
> 
> Peter Hurley (12):
>   tty: Fix ldisc leak in failed tty_init_dev()
>   tty: Remove !tty check from free_tty_struct()
>   tty: Fix tty_init_termios() declaration
>   tty: Re-declare tty_driver_remove_tty() file scope
>   pty: Remove pty_unix98_shutdown()
>   tty: Remove __lockfunc annotation from tty lock functions
>   tty: Wait interruptibly for tty lock on reopen
>   pty: Prepare to redefine tty driver remove() interface
>   tty: Re-define tty driver remove() interface
>   tty: Consolidate noctty checks in tty_open()
>   tty: Refactor tty_open()
>   tty: Retry failed reopen if tty teardown in-progress
> 
>  drivers/tty/pty.c            |  40 +++-------
>  drivers/tty/tty_io.c         | 180 +++++++++++++++++++++----------------------
>  drivers/tty/tty_ldisc.c      |  21 +++--
>  drivers/tty/tty_mutex.c      |  16 +++-
>  drivers/usb/serial/console.c |   6 +-
>  include/linux/tty.h          |  19 ++---
>  include/linux/tty_driver.h   |   4 +-
>  7 files changed, 128 insertions(+), 158 deletions(-)
>