packet: fix packet_direct_xmit for BQL enabled drivers

Currently, in packet_direct_xmit() we test the assigned netdevice queue
for netif_xmit_frozen_or_stopped() before doing an ndo_start_xmit().

This can have the side-effect that BQL enabled drivers which make use
of netdev_tx_sent_queue() internally, set __QUEUE_STATE_STACK_XOFF from
within the stack and would not fully fill the device's TX ring from
packet sockets with PACKET_QDISC_BYPASS enabled.

Instead, use a test without BQL bit so that bursts can be absorbed
into the NICs TX ring. Fix and code suggested by Eric Dumazet, thanks!

Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: Daniel Borkmann <dborkman@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

diff --git a/include/linux/netdevice.h b/include/linux/netdevice.h
index 775cc956ff7874c5949ec707357341137c6fdb1f..7ed3a3aa6604b7b8511ea709ea2c2312a1d38bb2 100644 (file)
--- a/include/linux/netdevice.h
+++ b/include/linux/netdevice.h
@@ -519,11 +519,18 @@ enum netdev_queue_state_t {
        __QUEUE_STATE_DRV_XOFF,
        __QUEUE_STATE_STACK_XOFF,
        __QUEUE_STATE_FROZEN,
-#define QUEUE_STATE_ANY_XOFF ((1 << __QUEUE_STATE_DRV_XOFF)            | \
-                             (1 << __QUEUE_STATE_STACK_XOFF))
-#define QUEUE_STATE_ANY_XOFF_OR_FROZEN (QUEUE_STATE_ANY_XOFF           | \
-                                       (1 << __QUEUE_STATE_FROZEN))
 };
+
+#define QUEUE_STATE_DRV_XOFF   (1 << __QUEUE_STATE_DRV_XOFF)
+#define QUEUE_STATE_STACK_XOFF (1 << __QUEUE_STATE_STACK_XOFF)
+#define QUEUE_STATE_FROZEN     (1 << __QUEUE_STATE_FROZEN)
+
+#define QUEUE_STATE_ANY_XOFF   (QUEUE_STATE_DRV_XOFF | QUEUE_STATE_STACK_XOFF)
+#define QUEUE_STATE_ANY_XOFF_OR_FROZEN (QUEUE_STATE_ANY_XOFF | \
+                                       QUEUE_STATE_FROZEN)
+#define QUEUE_STATE_DRV_XOFF_OR_FROZEN (QUEUE_STATE_DRV_XOFF | \
+                                       QUEUE_STATE_FROZEN)
+
 /*
  * __QUEUE_STATE_DRV_XOFF is used by drivers to stop the transmit queue.  The
  * netif_tx_* functions below are used to manipulate this flag.  The
@@ -2252,11 +2259,18 @@ static inline bool netif_xmit_stopped(const struct netdev_queue *dev_queue)
        return dev_queue->state & QUEUE_STATE_ANY_XOFF;
 }
 
-static inline bool netif_xmit_frozen_or_stopped(const struct netdev_queue *dev_queue)
+static inline bool
+netif_xmit_frozen_or_stopped(const struct netdev_queue *dev_queue)
 {
        return dev_queue->state & QUEUE_STATE_ANY_XOFF_OR_FROZEN;
 }
 
+static inline bool
+netif_xmit_frozen_or_drv_stopped(const struct netdev_queue *dev_queue)
+{
+       return dev_queue->state & QUEUE_STATE_DRV_XOFF_OR_FROZEN;
+}
+
 static inline void netdev_tx_sent_queue(struct netdev_queue *dev_queue,
                                        unsigned int bytes)
 {

diff --git a/net/packet/af_packet.c b/net/packet/af_packet.c
index c81a9719b5b2825b943852de29a748555a38d111..72e0c71fb01dddeee530067b341601d7196b25b3 100644 (file)
--- a/net/packet/af_packet.c
+++ b/net/packet/af_packet.c
@@ -261,7 +261,7 @@ static int packet_direct_xmit(struct sk_buff *skb)
        local_bh_disable();
 
        HARD_TX_LOCK(dev, txq, smp_processor_id());
-       if (!netif_xmit_frozen_or_stopped(txq)) {
+       if (!netif_xmit_frozen_or_drv_stopped(txq)) {
                ret = ops->ndo_start_xmit(skb, dev);
                if (ret == NETDEV_TX_OK)
                        txq_trans_update(txq);