Merge git://git.kernel.org/pub/scm/linux/kernel/git/davem/net

Pull networking fixes from David Miller:

 1) Include fixes for netrom and dsa (Fabian Frederick and Florian
    Fainelli)

 2) Fix FIXED_PHY support in stmmac, from Giuseppe CAVALLARO.

 3) Several SKB use after free fixes (vxlan, openvswitch, vxlan,
    ip_tunnel, fou), from Li ROngQing.

 4) fec driver PTP support fixes from Luwei Zhou and Nimrod Andy.

 5) Use after free in virtio_net, from Michael S Tsirkin.

 6) Fix flow mask handling for megaflows in openvswitch, from Pravin B
    Shelar.

 7) ISDN gigaset and capi bug fixes from Tilman Schmidt.

 8) Fix route leak in ip_send_unicast_reply(), from Vasily Averin.

 9) Fix two eBPF JIT bugs on x86, from Alexei Starovoitov.

10) TCP_SKB_CB() reorganization caused a few regressions, fixed by Cong
    Wang and Eric Dumazet.

11) Don't overwrite end of SKB when parsing malformed sctp ASCONF
    chunks, from Daniel Borkmann.

12) Don't call sock_kfree_s() with NULL pointers, this function also has
    the side effect of adjusting the socket memory usage.  From Cong Wang.

* git://git.kernel.org/pub/scm/linux/kernel/git/davem/net: (90 commits)
  bna: fix skb->truesize underestimation
  net: dsa: add includes for ethtool and phy_fixed definitions
  openvswitch: Set flow-key members.
  netrom: use linux/uaccess.h
  dsa: Fix conversion from host device to mii bus
  tipc: fix bug in bundled buffer reception
  ipv6: introduce tcp_v6_iif()
  sfc: add support for skb->xmit_more
  r8152: return -EBUSY for runtime suspend
  ipv4: fix a potential use after free in fou.c
  ipv4: fix a potential use after free in ip_tunnel_core.c
  hyperv: Add handling of IP header with option field in netvsc_set_hash()
  openvswitch: Create right mask with disabled megaflows
  vxlan: fix a free after use
  openvswitch: fix a use after free
  ipv4: dst_entry leak in ip_send_unicast_reply()
  ipv4: clean up cookie_v4_check()
  ipv4: share tcp_v4_save_options() with cookie_v4_check()
  ipv4: call __ip_options_echo() in cookie_v4_check()
  atm: simplify lanai.c by using module_pci_driver
  ...

1 files changed, 25 insertions, 9 deletions

diff --git a/net/ipv4/tcp_output.c b/net/ipv4/tcp_output.c
index becd98ce9a1c..3af21296d967 100644
--- a/net/ipv4/tcp_output.c
+++ b/net/ipv4/tcp_output.c
@@ -839,26 +839,38 @@ void tcp_wfree(struct sk_buff *skb)
 {
 	struct sock *sk = skb->sk;
 	struct tcp_sock *tp = tcp_sk(sk);
+	int wmem;
+
+	/* Keep one reference on sk_wmem_alloc.
+	 * Will be released by sk_free() from here or tcp_tasklet_func()
+	 */
+	wmem = atomic_sub_return(skb->truesize - 1, &sk->sk_wmem_alloc);
+
+	/* If this softirq is serviced by ksoftirqd, we are likely under stress.
+	 * Wait until our queues (qdisc + devices) are drained.
+	 * This gives :
+	 * - less callbacks to tcp_write_xmit(), reducing stress (batches)
+	 * - chance for incoming ACK (processed by another cpu maybe)
+	 *   to migrate this flow (skb->ooo_okay will be eventually set)
+	 */
+	if (wmem >= SKB_TRUESIZE(1) && this_cpu_ksoftirqd() == current)
+		goto out;
 
 	if (test_and_clear_bit(TSQ_THROTTLED, &tp->tsq_flags) &&
 	    !test_and_set_bit(TSQ_QUEUED, &tp->tsq_flags)) {
 		unsigned long flags;
 		struct tsq_tasklet *tsq;
 
-		/* Keep a ref on socket.
-		 * This last ref will be released in tcp_tasklet_func()
-		 */
-		atomic_sub(skb->truesize - 1, &sk->sk_wmem_alloc);
-
 		/* queue this socket to tasklet queue */
 		local_irq_save(flags);
 		tsq = this_cpu_ptr(&tsq_tasklet);
 		list_add(&tp->tsq_node, &tsq->head);
 		tasklet_schedule(&tsq->tasklet);
 		local_irq_restore(flags);
-	} else {
-		sock_wfree(skb);
+		return;
 	}
+out:
+	sk_free(sk);
 }
 
 /* This routine actually transmits TCP packets queued in by
@@ -914,9 +926,13 @@ static int tcp_transmit_skb(struct sock *sk, struct sk_buff *skb, int clone_it,
 		tcp_ca_event(sk, CA_EVENT_TX_START);
 
 	/* if no packet is in qdisc/device queue, then allow XPS to select
-	 * another queue.
+	 * another queue. We can be called from tcp_tsq_handler()
+	 * which holds one reference to sk_wmem_alloc.
+	 *
+	 * TODO: Ideally, in-flight pure ACK packets should not matter here.
+	 * One way to get this would be to set skb->truesize = 2 on them.
 	 */
-	skb->ooo_okay = sk_wmem_alloc_get(sk) == 0;
+	skb->ooo_okay = sk_wmem_alloc_get(sk) < SKB_TRUESIZE(1);
 
 	skb_push(skb, tcp_header_size);
 	skb_reset_transport_header(skb);