Re: bug in procps-ng

[Corinna Vinschen <corinna-cygwin at cygwin dot com>]

On Mar 12 22:55, Brian Inglis wrote:
> On 2019-03-12 11:03, Corinna Vinschen wrote:
> > I hope you're still with us.
> > There appears to be a bug in procps-ng:
> > $ procps -e
> >   PID TTY          TIME CMD
> >  1507 ?        00:00:00 tcsh
> >  1529 ?        00:00:00 cygrunsrv
> >  1506 ?        00:00:00 mintty
> >  1531 ?        00:00:00 bash
> >  1551 pty0     00:00:00 procps
> >  1488 pty0     00:00:00 sh
> >  1487 ?        00:00:01 mintty
> >  1530 ?        00:00:00 xterm
> > As you can see, the tty info seems to be broken.  In theory
> > procps-ng should fetch the tty info from /proc/<PID>/stat.
> > As far as I can see, the tty info in this file is correct
> > for other processes.  But procps only shows info for its
> > own tty for some reason.
> > Any idea why?
> 
> Are /dev/con?? and /dev/pty? visible only in those processes?

/dev/consX, yes.  the ptys are always visible.

> Need persistent visible mappings to /dev/con?? and /dev/pty?
> Inconsistencies in /dev/con?? handling: why do con* appear under major 5 with
> different minors, and also as 3,0 consistent with /proc/PID/stat?
> 
> $ ls -l /dev/ | egrep con\|pty
> crw-rw-rw-  1 Brian  Users    5, 255 Mar 12 22:23 conin
> crw-rw-rw-  1 Brian  Users    5, 254 Mar 12 22:23 conout
> crw-rw-rw-  1 Brian  Users    5,   1 Mar 12 22:23 console
> crw--w----  1 Brian  Users  136,   0 Mar 12 22:23 pty0
> $ ls -l /dev/{con,pty}*		# /proc/PID/stat maps to these major/minors
> crw-rw-rw- 1 Brian Users   3, 0 Mar 12 22:23 /dev/conin
> crw-rw-rw- 1 Brian Users   3, 0 Mar 12 22:23 /dev/conout
> crw-rw-rw- 1 Brian Users   3, 0 Mar 12 22:23 /dev/console
> crw--w---- 1 Brian Users 136, 0 Mar 12 22:23 /dev/pty0

Try the same from within a console:

$ ls -l /dev/ | egrep con\|pty
crw-rw-rw-  4 corinna vinschen   3,   0 Mar 13 10:37 conin
crw-rw-rw-  4 corinna vinschen   3,   0 Mar 13 10:37 conout
crw-rw-rw-  4 corinna vinschen   3,   0 Mar 13 10:37 cons0
crw-rw-rw-  4 corinna vinschen   3,   0 Mar 13 10:37 console
crw--w----  1 corinna vinschen 136,   0 Mar 13 10:37 pty0
[~](64)$ ls -l /dev/{con,pty}*
crw-rw-rw- 4 corinna vinschen   3, 0 Mar 13 10:37 /dev/conin
crw-rw-rw- 4 corinna vinschen   3, 0 Mar 13 10:37 /dev/conout
crw-rw-rw- 4 corinna vinschen   3, 0 Mar 13 10:37 /dev/cons0
crw-rw-rw- 4 corinna vinschen   3, 0 Mar 13 10:37 /dev/console
crw--w---- 1 corinna vinschen 136, 0 Mar 13 10:37 /dev/pty0

The device numbers for conin/out/sole are different from a pty
because the 5,X device numbers are the offical device numbers
of these devices.  But these files have no further meaning from
inside a pty.

As soon as you are inside a console, the conin/out/sole devices
are redirected to your actual consX, which has a device number
3,X and all of them are connected now to a real device.

> [...]
> Cygwin /proc/PID/stat[$7] appears to have major in upper half, minor in lower
> half e.g. 8912896 -> 00880000 -> 136,0; 196608 -> 00030000 -> 3,0.
> Linux man 5 proc defines:
> (7) tty_nr  %d
> 	The controlling terminal of the process. (The minor device number is
> 	contained in the combination of bits 31 to 20 and 7 to 0; the major
> 	device number is in bits 15 to 8.)

Thanks.  I changed the output of /proc/<PID>/stat accordingly, but
now `procps' always shows a question mark rather than a pty number,
so that's not the problem:

https://paste.fedoraproject.org/paste/BqENX~bhFzNkYQpa-lwWMw/raw


Corinna

-- 
Corinna Vinschen
Cygwin Maintainer

Attachment: signature.asc
Description: PGP signature