-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1 Hi Sebastian,
On 11/09/2010 04:51 PM, Sebastian Nickel - Hetzner Online AG wrote: > Hello, > I tried to increase so-rcvbuf to 8m and to 16m. I changed > net.core.rmem_max values in system, too. > > I still see many of those lines in the log: Usually EAGAIN means the device is busy, but I have never seen it happen on send. Attached a small patch, could you test this? It sets SO_SNDBUF to the same value as SO_RCVBUF, perhaps the output buffer is full? If it gives permission denied, change SO_SNDBUF into SO_SNDBUFFORCE Otherwise, netstat -su? does it report send buffer overruns (or can it possibly do so?). Does a symmetric rcvbuf-sndbuf setting make sense (apart from this test)? Or do you need to be able to configure the sndbuf? (question only applicable if attachment solves problem :-) ). Best regards, Wouter -----BEGIN PGP SIGNATURE----- Version: GnuPG v1.4.11 (GNU/Linux) Comment: Using GnuPG with Fedora - http://enigmail.mozdev.org/ iEYEARECAAYFAkzZc1UACgkQkDLqNwOhpPgvxwCeK+HBCVmX+xtSHXkc4SISq+1c dXMAoJJMoVg9l2ILsRUdvQSTLNnOb4yR =mpmH -----END PGP SIGNATURE-----
Index: services/listen_dnsport.c
===================================================================
--- services/listen_dnsport.c (revision 2339)
+++ services/listen_dnsport.c (working copy)
@@ -181,6 +181,15 @@
# ifdef SO_RCVBUFFORCE
}
# endif
+ if(setsockopt(s, SOL_SOCKET, SO_SNDBUF, (void*)&rcv,
+ (socklen_t)sizeof(rcv)) < 0) {
+ log_err("setsockopt(..., SO_RCVBUF, "
+ "...) failed: %s", strerror(errno));
+ close(s);
+ *noproto = 0;
+ *inuse = 0;
+ return -1;
+ }
#endif /* SO_RCVBUF */
}
if(family == AF_INET6) {
patch_sndbuf_test.diff.sig
Description: PGP signature
_______________________________________________ Unbound-users mailing list [email protected] http://unbound.nlnetlabs.nl/mailman/listinfo/unbound-users
