From: Dan Williams <dcbw@redhat.com>
Date: Wed, 4 May 2016 14:18:52 +0000 (-0500)
Subject: bridge: leave TX queue length as kernel default, not 0
X-Git-Url: https://git.halfball.org/?a=commitdiff_plain;h=899c0efd42059b773afdca80856f5eed3dc4af7a;p=cni.git

bridge: leave TX queue length as kernel default, not 0

Not using NewLinkAttrs() or not initializing TxQLen leaves
the value as 0, which tells the kernel to set a zero-length
tx_queue_len.  That messes up FIFO traffic shapers (like pfifo)
that use the device TX queue length as the default packet
limit.  This leads to a default packet limit of 0, which drops
all packets.
---

diff --git a/plugins/main/bridge/bridge.go b/plugins/main/bridge/bridge.go
index 77c5215..7bbba7e 100644
--- a/plugins/main/bridge/bridge.go
+++ b/plugins/main/bridge/bridge.go
@@ -101,6 +101,11 @@ func ensureBridge(brName string, mtu int) (*netlink.Bridge, error) {
 		LinkAttrs: netlink.LinkAttrs{
 			Name: brName,
 			MTU:  mtu,
+			// Let kernel use default txqueuelen; leaving it unset
+			// means 0, and a zero-length TX queue messes up FIFO
+			// traffic shapers which use TX queue length as the
+			// default packet limit
+			TxQLen: -1,
 		},
 	}