solaris 2.2 & rshd & rc

solaris 2.2 & rshd & rc

[Scott Merrilees]

When doing a rsh command to a solaris 2.2 box, I find that the rc
started by rshd hangs in a cpu loop after the command has terminated.
Interestingly, SIGCLD is ingored in both rshd & rc.  Has anyone else
fixed this problem ?

Sm
--
Scott Merrilees, BHP Information Technology, Newcastle, Australia
Internet: Sm@bhpese.oz.au   Phone: +61 49 40 1687   Fax: ... 2165

Re: solaris 2.2 & rshd & rc

[noel]

    From:	Scott Merrilees <Sm@cerberus.bhpese.oz.au>
    Subject: solaris 2.2 & rshd & rc

    When doing a rsh command to a solaris 2.2 box, I find that the rc
    started by rshd hangs in a cpu loop after the command has terminated.
    Interestingly, SIGCLD is ingored in both rshd & rc.  Has anyone else
    fixed this problem ?

it's not that rc _ignores_ SIGCLD, it just doesn't touch it. when rc
is started by login and forks a program, you will notice that SIGCLD
is SIG_DFL. but when rc is started by rshd, it must be set to SIG_IGN.
bad shit happens in sysvr4 when this is done, namely, a parent is not
notified of the death of a child, so rc just sits in that wait loop
waiting to pick up the status of the child it just forked, a status
it will never get. i noticed the cpu in a tight loop when this happened.

anyway, i just hacked the code to make sure SIGCLD is _always_ SIG_DFL.
this seems to fix the problem.

Re: solaris 2.2 & rshd & rc

[Scott Merrilees]

    From:	Scott Merrilees <Sm@cerberus.bhpese.oz.au>
    When doing a rsh command to a solaris 2.2 box, I find that the rc
    started by rshd hangs in a cpu loop after the command has terminated.
    Interestingly, SIGCLD is ingored in both rshd & rc.  Has anyone else
    fixed this problem ?

>From: noel@es.su.oz.au
>it's not that rc _ignores_ SIGCLD, it just doesn't touch it. when rc
>is started by login and forks a program, you will notice that SIGCLD
>is SIG_DFL. but when rc is started by rshd, it must be set to SIG_IGN.
>bad shit happens in sysvr4 when this is done, namely, a parent is not
>notified of the death of a child, so rc just sits in that wait loop
>waiting to pick up the status of the child it just forked, a status
>it will never get. i noticed the cpu in a tight loop when this happened.
>
>anyway, i just hacked the code to make sure SIGCLD is _always_ SIG_DFL.
>this seems to fix the problem.

That was my guess, I figure that rshd must be doing a process id
specific wait, such as waitpid(), and doesn't want to get hit with
SIGCLD signals.

After looking in fn.c:setsigdefaults(), I came up with the following
patch to fn.c:inithandlers().  This seems to work.  The patch is
relative to rc-1.4.

Sm
--
*** fn.c.orig	Tue Mar 29 12:37:04 1994
--- fn.c	Tue Mar 29 12:39:16 1994
***************
*** 45,50 ****
--- 45,54 ----
  			fnrm("sigterm"); /* ditto for SIGTERM */
  		}
  	}
+ #ifdef NOSIGCLD
+ 	rc_signal(SIGCLD, SIG_DFL);
+ 	delete_fn(signals[SIGCLD].name);
+ #endif
  }
  
  /* only run this in a child process! resets signals to their default values */