Bug#35704: libc6: ssh fails to allocate a pseudo tty
On Wed, 7 Apr 1999 12:34:19 -0600, Erik Andersen wrote:
>>
>> This sounds like the problem where openpty() sees /dev/ptmx, tries to
>> use it, fails because /dev/pts isn't populated, and gives up. I bet
>> your problem will go away if you either remove /dev/ptmx, or mount
>> devptsfs on /dev/pts.
>>
>> Libc should probably be tweaked to fall back to the old pty devices if
>> /dev/ptmx exists but /dev/pts is empty. I'll look into it.
>>
>> zw
>
>Yes, for now a "mkdir /dev/pts" and then adding:
>
> "devpts /dev/pts devpts defaults 0 0"
>
>to my /etc/fstab, and ensuring that my 2.2.5 kernel was compiled with
>"CONFIG_UNIX98_PTYS=y", "CONFIG_UNIX98_PTY_COUNT=256" and "CONFIG_DEVPTS_FS=y"
>in the .config allows ssh to work again.
>
>Is the plan to have the 2.2 kernel be default in potato? If so, then we
>need to get kernel 2.2.x into potato with the PTY config options turned
>on, and then have libc6 include the /dev/pts directory in the .deb package
>and install the /etc/fstab entry as well to ensure that installing libc
>doesn't break ssh, telnet, and friends.
I believe 2.2.x is supposed to be the default in potato. I assume
we're waiting for it to get past its teething problems.
Unfortunately, because of the way devptsfs works, I can't make an
ironclad test for the presence of the pty slaves in libc. Best I can
do is make sure that /dev/ptmx exists and /dev/pts is a directory.
That means, if /dev/pts is provided by libc, we'll *still* get it
wrong on a 2.0.x system.
...Therefore I think /dev/pts and the fstab entry should be created by
the 2.2.x kernel package, not by libc. We do have an update-fstab
thingy, right?
zw
Reply to: