Bug 3075 - Causes terminal corruption by disabling XON/XOFF unconditionally
Summary: Causes terminal corruption by disabling XON/XOFF unconditionally
Status: NEW
Alias: None
Product: Portable OpenSSH
Classification: Unclassified
Component: ssh (show other bugs)
Version: 7.9p1
Hardware: amd64 Linux
: P5 normal
Assignee: Assigned to nobody
URL:
Keywords:
Depends on:
Blocks:
 
Reported: 2019-10-01 03:00 AEST by John Goerzen
Modified: 2020-01-25 17:31 AEDT (History)
2 users (show)

See Also:


Attachments

Note You need to log in before you can comment on or make changes to this bug.
Description John Goerzen 2019-10-01 03:00:27 AEST
Hello,

I am using an original DEC vt420 serial terminal connected to a Debian
box.  As with many such terminals, it:

1) Is incapable of consistently processing incoming data, particularly
if it contains escape sequences, at line rate if the line rate is
above 4800bps (though it supports line rates up to 38400bps);

2) Supports only XON/XOFF flow control.

I observed issues that were clearly related to flow control with
sshing from it to other systems, although it was working fine
locally.  I narrowed it down to this code in
sshtty.c:enter_raw_mode():

        tio.c_iflag &= ~(ISTRIP | INLCR | IGNCR | ICRNL | IXON | IXANY | IXOFF);

in other words, it unconditionally disables XON/XOFF handling on the
local machine.

You might be asking at this point: well, couldn't the remote handle
it?  And the answer is no, for several reasons:

1) The internal buffer on this terminal is 254 bytes, far less than even
the TCP packet size.  It is quite possible -- likely, even -- that a
large screen update already caused a packet to be transmitted from the
remote end that will overflow the local buffer.

2) The latency in processing could mean that the remote sends enough
text to overflow the terminal's buffer after the terminal sends XOFF.

Other people from around the internet also report this issue.  For
instance:

https://superuser.com/questions/1096862/ssh-and-xon-xoff-software-flow-control

I recognize that there may be reasons to drop IXON and IXOFF by
default, but they are not appropriate for situations in which IXON and
IXOFF are legitimately needed.  Perhaps a configuration option here?

I'm happy to test patches, etc.

Thanks,

John
Comment 1 Damien Miller 2020-01-25 17:31:23 AEDT
Adding an option to control the terminal modes at the client seem pretty niche unfortunately. Could you do it via stty(1) manually? It might be possible to script it via the LocalCommand in ssh_config with something like:

Match all
  LocalCommand sleep 1 && stty ixon ioff

(untested)