Workaround Solaris's kernel bug

[JINMEI Tatuya / 神明達哉]

At Thu, 20 Nov 2008 06:04:39 -0600,
"J.D. Bronson" <jbronson at hanadarko.com> wrote:

> So is there a version *public release* of Bind9 that we can compile 
> right out of the box that will work correctly on Solaris 10 (10/08) and if so
> which version is it?

9.4.3 and 9.3.6.

9.5.1rc1 which will soon be released, will also have this workaround.

Notes:

- we suspect most users don't need this workaround.  we don't know the
  nature of the kernel bug, but from our experiences this happens only
  for a very very busy server.  if your server keeps running, you
  won't need it.
- we're also not sure if this kernel bug is related to threads, but,
  again, we've only seen this with threads in our experiments.  so the
  workaround code is only (able to be) enabled for named built with
  threads.
- there's a minor bug in the workaround code: with a general debug
  level of 1 or higher, named could crash when the kernel bug is
  triggered or when the daemon stops.  If you need to configure a
  higher log level please apply the patch copied below.  This bug is
  fixed in 9.5.1rc1.

---
JINMEI, Tatuya
Internet Systems Consortium, Inc.

Index: socket.c
===================================================================
RCS file: /proj/cvs/prod/bind9/lib/isc/unix/socket.c,v
retrieving revision 1.305
diff -u -r1.305 socket.c
--- socket.c	8 Nov 2008 22:35:12 -0000	1.305
+++ socket.c	19 Nov 2008 22:15:00 -0000
@@ -3331,7 +3331,6 @@
 						    ISC_LOGCATEGORY_GENERAL,
 						    ISC_LOGMODULE_SOCKET,
 						    ISC_LOG_DEBUG(1),
-						    ISC_LOG_INFO,
 						    "unexpected POLL timeout");
 				}
 				pollstate = poll_active;