Problems with rlogin

[Rick Lindsley]

In article <332 at esl.UUCP> mac at esl.UUCP (Mike McNamara) writes:

> User     tty       login@  idle   JCPU   PCPU  what
> amp      p3board   2:13pm  1:43   1:32     16  -csh 
> amp      pty2      4:38pm238:48 123:92 123:21  - 7%4( %^
> mac      ttyp3     4:35pm         6:44   4:35  w
> 
> A reboot cleans these up.  Until a reboot, users attempting to vi get wierd
> window sizes (indeed, anything accessing window sizes over ptys gets strange
> data -- this includes people logged in over ethernet.)
> 
> Any one know a better fix than L1 A ( or shutdown -r 5 Resetting Sun :-)?

Yes, I think so. I spent the better part of a day tracking this down.
I'm running 3.0 on a 3/50. The same bug exists under 3.0 on a 3/160
too.

3.0 has kernel structures for window sizes. Of course. These are
settable via an ioctl. Termcap has been altered to look at these sizes.
If they are non-zero, they apparently REPLACE the li: and co: entries
in the termcap entry. Normally, this makes perfect sense. Unless ...
unless you aren't running in a resizeable window.

Something takes care to set these to 0 when you start.  However, this
DOESN'T happen if there is already a process running on the pty.
Judging from this, I'd say the kernel, upon final close in the tty
driver, is in charge of resetting these to 0.  I can think of many
situations where this makes perfect sense, but nevertheless it is the
cause of the bug.

The 'w' output you gave (and the amazing amount of cpu time) indicates
there is probably another process still running on that terminal. Did
you do a "ps atp1"? That will show all the processes running on ttyp1,
and might verify that this is the case.

Repeat it by making a window. Run the program below, tst, to show the
window size. Do "sleep 600 &" and close the window. Now rlogin from
another host. Check to see that you have the right tty. Run tst again,
and you'll see the "window size" hasn't changed. Unless you are running
on a terminal which happens to be the same size as your window was,
your termcap won't work. Tset does NOT reset the window size, either.
(I can see arguments in favor of this behavior also.) It might be
reasonable for stty to report (and set) these sizes. Perhaps rlogind
should do this. Doesn't matter; the fix is Sun's to ponder, not mine. I
couldn't find a Sun program which allowed me to change the window size,
so I wrote tst2. If you run tst2, below, you'll find your troubles go
away, if they are from this problem. No complaints about style please;
I admit they are quick and dirty! Embellish at will.

tst:
-----
#include <sys/ioctl.h>

main()

{
    struct ttysize size;

    if (ioctl(0,TIOCGSIZE,&size) < 0)
	perror("ioctl");
    else
	printf("Size is %d, %d\n", size.ts_lines, size.ts_cols);
}
-----

tst2:
-----
#include <sys/ioctl.h>

main()

{
    struct ttysize size;

    size.ts_lines = 0;
    size.ts_cols = 0;
    if (ioctl(0,TIOCSSIZE,&size) < 0)
	perror("ioctl");
}
-----

Rick Lindsley