yppush going into an endless loop

[Thomas Quinot]

On a 5.2.1-REL NIS server where NIS maps are updated every hour from a
crontab, I often see yppush going into an endless loop:

0x281220dc in __vfprintf () from /lib/libc.so.5
(gdb) bt
#0  0x281220dc in __vfprintf () from /lib/libc.so.5
#1  0x28121b3f in strchr () from /lib/libc.so.5
#2  0x28122173 in __vfprintf () from /lib/libc.so.5
#3  0x281220a6 in vfprintf () from /lib/libc.so.5
#4  0x2810ed6d in fprintf () from /lib/libc.so.5
#5  0x08049bf2 in __verr (
    fmt=0x804c2c0 "warning: exiting with transfer to %s (transid = %lu) still pending", ap=0x2813d120 "\\`\f")
    at /usr/src/RELENG_5_2/usr.sbin/ypserv/yp_error.c:58
#6  0x08049c58 in yp_error (fmt=0xbbc004d0 "Ð")
    at /usr/src/RELENG_5_2/usr.sbin/ypserv/yp_error.c:71
#7  0x0804a39f in yppush_exit (now=672387024)
    at /usr/src/RELENG_5_2/usr.sbin/yppush/yppush_main.c:195
#8  0x0804a406 in handler (sig=6)
    at /usr/src/RELENG_5_2/usr.sbin/yppush/yppush_main.c:213
#9  <signal handler called>
#10 0x280c1dcf in kill () from /lib/libc.so.5
#11 0x2812ef82 in abort () from /lib/libc.so.5
#12 0x2812d6fe in tcflow () from /lib/libc.so.5
#13 0x2812d72b in tcflow () from /lib/libc.so.5
#14 0x2812e459 in free () from /lib/libc.so.5
#15 0x281090e3 in _nsdbtput () from /lib/libc.so.5
#16 0x2810909b in _nsdbtput () from /lib/libc.so.5
---Type <return> to continue, or q <return> to quit---
#17 0x28108c98 in endhostent () from /lib/libc.so.5
#18 0x2810944f in _nsdbtput () from /lib/libc.so.5
#19 0x2812f384 in exit () from /lib/libc.so.5
#20 0x0804a3c6 in yppush_exit ()
    at /usr/src/RELENG_5_2/usr.sbin/yppush/yppush_main.c:201
#21 0x0804a406 in handler (sig=6)
    at /usr/src/RELENG_5_2/usr.sbin/yppush/yppush_main.c:213
#22 <signal handler called>
#23 0x280c1dcf in kill () from /lib/libc.so.5
Script done on Tue Oct 19 15:19:12 2004

Two questions:
  1. has anyone already observed similar behaviour;
  2. from code-reading, it looks like we attempt to prevent
     yppush_exit from wrecking havoc when we call it from
     a signal handler by zeroing out yppush_jobs. Shouldn't that
     be yppush_joblist?

As far as 2 is concerned, the following change should work around the
problem (but then I'd still have to find out why yppush is exiting on a
signal in the first place).

Index: yppush_main.c
===================================================================
RCS file: /home/ncvs/src/usr.sbin/yppush/yppush_main.c,v
retrieving revision 1.18
diff -u -r1.18 yppush_main.c
--- yppush_main.c	3 May 2003 21:06:41 -0000	1.18
+++ yppush_main.c	19 Oct 2004 13:37:22 -0000
@@ -209,7 +209,7 @@
 handler(int sig)
 {
 	if (sig == SIGTERM || sig == SIGINT || sig == SIGABRT) {
-		yppush_jobs = 0;
+		yppush_joblist = NULL;
 		yppush_exit(1);
 	}
 
-- 
    Thomas.Quinot at Cuivre.FR.EU.ORG