[Savannah-hackers] Re: lshd interruption problem

[Niels Möller]

Loic Dachary <address@hidden> writes:

>       I hope the lsof output provided in the previous mail (for lshd
> + the command) contains enough information. Otherwise I can provide more
> information since the command is still hanging as we speak ;-)

I'm not sure how to interpret that data. I see that the cvs server
process has some pipes (or more precisely, AF_UNIX socketpairs)
connected to its stdio fd:s:

  cvs     15429 gdenat    0u  unix 0xf7f49cc0         354931150 socket
  cvs     15429 gdenat    1u  unix 0xf7f49cc0         354931150 socket
  cvs     15429 gdenat    2u  unix 0xf38cfd00         354931152 socket

But looking at the open files for the lshd server, the only unix
socket I see is

  lshd    10234 root    3u  unix 0xde38cdc0         387347010 socket

which I guess is the connection to syslogd. It does have a few pty:s
open, though, but I hope that no pty:s were allocated for the cvs
processes (one would have to run either server or client with
--verbose to find out for sure).

To me, that indicates that lshd have closed its ends of the pipes.
Perhaps it's cvssh that doesn't notice stdin being closed?

>       I'll do that. I see that the initgroups workaround moved to
> lsh-1.4/src/lsh-execuv.c. But I don't see the
> --enable-initgroups-workaround in the configure.ac. Is it gone ?

It seems I lost that code in configure.in when moving the hack over to
lsh-1.3.x (I don't use that feature myself). I have to release a fixed
lsh-1.4.1 then. In the mean time, I would appreciate if you could add

  #define INITGROUPS_WORKAROUND 1

to config.h manually and tell me how that works.

Regards,
/Niels