tcp: fix possible deadlock in tcp_send_fin()

Using sk_stream_alloc_skb() in tcp_send_fin() is dangerous in
case a huge process is killed by OOM, and tcp_mem[2] is hit.

To be able to free memory we need to make progress, so this
patch allows FIN packets to not care about tcp_mem[2], if
skb allocation succeeded.

In a follow-up patch, we might abort tcp_send_fin() infinite loop
in case TIF_MEMDIE is set on this thread, as memory allocator
did its best getting extra memory already.

This patch reverts d22e15371811 ("tcp: fix tcp fin memory accounting")

Fixes: d22e15371811 ("tcp: fix tcp fin memory accounting")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

1 files changed, 19 insertions, 1 deletions

diff --git a/net/ipv4/tcp_output.c b/net/ipv4/tcp_output.c
index 8c8d7e06b72f..2ade67b7cdb0 100644
--- a/net/ipv4/tcp_output.c
+++ b/net/ipv4/tcp_output.c
@@ -2812,6 +2812,21 @@ begin_fwd:
         }
 }
 
+/* We allow to exceed memory limits for FIN packets to expedite
+ * connection tear down and (memory) recovery.
+ * Otherwise tcp_send_fin() could loop forever.
+ */
+static void sk_forced_wmem_schedule(struct sock *sk, int size)
+{
+        int amt, status;
+
+        if (size <= sk->sk_forward_alloc)
+                return;
+        amt = sk_mem_pages(size);
+        sk->sk_forward_alloc += amt * SK_MEM_QUANTUM;
+        sk_memory_allocated_add(sk, amt, &status);
+}
+
 /* Send a fin.  The caller locks the socket for us.  This cannot be
  * allowed to fail queueing a FIN frame under any circumstances.
  */
@@ -2834,11 +2849,14 @@ void tcp_send_fin(struct sock *sk)
         } else {
                 /* Socket is locked, keep trying until memory is available. */
                 for (;;) {
-                        skb = sk_stream_alloc_skb(sk, 0, sk->sk_allocation);
+                        skb = alloc_skb_fclone(MAX_TCP_HEADER,
+                                               sk->sk_allocation);
                         if (skb)
                                 break;
                         yield();
                 }
+                skb_reserve(skb, MAX_TCP_HEADER);
+                sk_forced_wmem_schedule(sk, skb->truesize);
                 /* FIN eats a sequence byte, write_seq advanced by tcp_queue_skb(). */
                 tcp_init_nondata_skb(skb, tp->write_seq,
                                      TCPHDR_ACK | TCPHDR_FIN);