I believe this will solve the previously reported problems.
With the original patch if I set net.inet.tcp.sendspace=63 and tried
to run xterm from that machine to my local workstation, I got an X error.
If I set sendspace=31 the xterm process just locked up and did nothing
until I ^C'd it.
machine A machine B
w/modified kernel w/unmodified kernel
(sysctl's on this machine)
xterm run on A --> display is on B
With this patch I can set net.inet.tcp.sendspace to anything (63, 31, 1,
whatever I want) and the xterm will still run. And yes, the xterm is
amazingly slow when I set net.inet.tcp.sendspace to 1 :-)
This patch is relative to -CURRENT but should also work with -STABLE.
-Matt
Index: uipc_socket.c
===================================================================
RCS file: /home/ncvs/src/sys/kern/uipc_socket.c,v
retrieving revision 1.60
diff -u -r1.60 uipc_socket.c
--- uipc_socket.c 1999/06/17 23:54:47 1.60
+++ uipc_socket.c 1999/07/22 23:08:38
@@ -413,7 +413,8 @@
register struct mbuf *m;
register long space, len, resid;
int clen = 0, error, s, dontroute, mlen;
- int atomic = sosendallatonce(so) || top;
+ int atomic = sosendallatonce(so) || top; /* required atomicy */
+ int try_atomic = atomic; /* requested atomicy */
if (uio)
resid = uio->uio_resid;
@@ -518,6 +519,7 @@
mlen = MCLBYTES;
len = min(min(mlen, resid), space);
} else {
+ try_atomic = 1; /* try to optimize */
nopages:
len = min(min(mlen, resid), space);
/*
@@ -541,7 +543,7 @@
top->m_flags |= M_EOR;
break;
}
- } while (space > 0 && atomic);
+ } while (space > 0 && try_atomic);
if (dontroute)
so->so_options |= SO_DONTROUTE;
s = splnet(); /* XXX */
To Unsubscribe: send mail to majord...@freebsd.org
with "unsubscribe freebsd-hackers" in the body of the message
