Virtual interfaces don't necessarily need a qdisc attached to them. This
is signalled by setting dev->tx_queue_len to zero upon initialisation. The
problems begin when a user still adds a qdisc, as then the special value
is used as a regular one causing massive packet drops as soon as the
device sees a bit of traffic load.
This patch solves the issue by setting tx_queue_len to the global
default for physical devices if it is zero and a qdisc is added. Another
benefit of this is that veth devices can default to noqueue while still
preserving the expected behaviour when adding a qdisc.
The drawback of this patch is that the noqueue state won't be restored
after a user has removed the custom qdisc, as the information about
whether this is legitimate for the given device is lost.
Signed-off-by: Phil Sutter <p...@nwl.cc>
---
net/sched/sch_api.c | 4 ++++
1 file changed, 4 insertions(+)
diff --git a/net/sched/sch_api.c b/net/sched/sch_api.c
index f06aa01..79b8900 100644
--- a/net/sched/sch_api.c
+++ b/net/sched/sch_api.c
@@ -1315,6 +1315,10 @@ graft:
qdisc_destroy(q);
return err;
}
+ /* circumvent noqueue hack for virtual interfaces if
+ * user desires to use a qdisc on it */
+ if (!dev->tx_queue_len)
+ dev->tx_queue_len = CONFIG_DEFAULT_TX_QUEUE_LEN;
return 0;
}
--
2.1.2
--
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to majord...@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
