tcp: add a force_schedule argument to sk_stream_alloc_skb()

In commit 8e4d980ac215 ("tcp: fix behavior for epoll edge trigger")
we fixed a possible hang of TCP sockets under memory pressure,
by allowing sk_stream_alloc_skb() to use sk_forced_mem_schedule()
if no packet is in socket write queue.

It turns out there are other cases where we want to force memory
schedule :

tcp_fragment() & tso_fragment() need to split a big TSO packet into
two smaller ones. If we block here because of TCP memory pressure,
we can effectively block TCP socket from sending new data.
If no further ACK is coming, this hang would be definitive, and socket
has no chance to effectively reduce its memory usage.

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

3 files changed, 18 insertions, 14 deletions

diff --git a/include/net/sock.h b/include/net/sock.h
index 4581a60636f8..26c1c3171e00 100644
--- a/include/net/sock.h
+++ b/include/net/sock.h
@@ -2025,7 +2025,8 @@ static inline void sk_stream_moderate_sndbuf(struct sock *sk)
 	}
 }
 
-struct sk_buff *sk_stream_alloc_skb(struct sock *sk, int size, gfp_t gfp);
+struct sk_buff *sk_stream_alloc_skb(struct sock *sk, int size, gfp_t gfp,
+				    bool force_schedule);
 
 /**
  * sk_page_frag - return an appropriate page_frag
diff --git a/net/ipv4/tcp.c b/net/ipv4/tcp.c
index bb9bb844204f..ca1d476c80ef 100644
--- a/net/ipv4/tcp.c
+++ b/net/ipv4/tcp.c
@@ -808,7 +808,8 @@ ssize_t tcp_splice_read(struct socket *sock, loff_t *ppos,
 }
 EXPORT_SYMBOL(tcp_splice_read);
 
-struct sk_buff *sk_stream_alloc_skb(struct sock *sk, int size, gfp_t gfp)
+struct sk_buff *sk_stream_alloc_skb(struct sock *sk, int size, gfp_t gfp,
+				    bool force_schedule)
 {
 	struct sk_buff *skb;
 
@@ -820,15 +821,15 @@ struct sk_buff *sk_stream_alloc_skb(struct sock *sk, int size, gfp_t gfp)
 
 	skb = alloc_skb_fclone(size + sk->sk_prot->max_header, gfp);
 	if (likely(skb)) {
-		bool mem_schedule;
+		bool mem_scheduled;
 
-		if (skb_queue_len(&sk->sk_write_queue) == 0) {
-			mem_schedule = true;
+		if (force_schedule) {
+			mem_scheduled = true;
 			sk_forced_mem_schedule(sk, skb->truesize);
 		} else {
-			mem_schedule = sk_wmem_schedule(sk, skb->truesize);
+			mem_scheduled = sk_wmem_schedule(sk, skb->truesize);
 		}
-		if (likely(mem_schedule)) {
+		if (likely(mem_scheduled)) {
 			skb_reserve(skb, sk->sk_prot->max_header);
 			/*
 			 * Make sure that we have exactly size bytes
@@ -918,7 +919,8 @@ new_segment:
 			if (!sk_stream_memory_free(sk))
 				goto wait_for_sndbuf;
 
-			skb = sk_stream_alloc_skb(sk, 0, sk->sk_allocation);
+			skb = sk_stream_alloc_skb(sk, 0, sk->sk_allocation,
+						  skb_queue_empty(&sk->sk_write_queue));
 			if (!skb)
 				goto wait_for_memory;
 
@@ -1154,7 +1156,8 @@ new_segment:
 
 			skb = sk_stream_alloc_skb(sk,
 						  select_size(sk, sg),
-						  sk->sk_allocation);
+						  sk->sk_allocation,
+						  skb_queue_empty(&sk->sk_write_queue));
 			if (!skb)
 				goto wait_for_memory;
 
diff --git a/net/ipv4/tcp_output.c b/net/ipv4/tcp_output.c
index e29d43b5a0bb..3aebe0157dfa 100644
--- a/net/ipv4/tcp_output.c
+++ b/net/ipv4/tcp_output.c
@@ -1172,7 +1172,7 @@ int tcp_fragment(struct sock *sk, struct sk_buff *skb, u32 len,
 		return -ENOMEM;
 
 	/* Get a new skb... force flag on. */
-	buff = sk_stream_alloc_skb(sk, nsize, gfp);
+	buff = sk_stream_alloc_skb(sk, nsize, gfp, true);
 	if (!buff)
 		return -ENOMEM; /* We'll just try again later. */
 
@@ -1731,7 +1731,7 @@ static int tso_fragment(struct sock *sk, struct sk_buff *skb, unsigned int len,
 	if (skb->len != skb->data_len)
 		return tcp_fragment(sk, skb, len, mss_now, gfp);
 
-	buff = sk_stream_alloc_skb(sk, 0, gfp);
+	buff = sk_stream_alloc_skb(sk, 0, gfp, true);
 	if (unlikely(!buff))
 		return -ENOMEM;
 
@@ -1950,7 +1950,7 @@ static int tcp_mtu_probe(struct sock *sk)
 	}
 
 	/* We're allowed to probe.  Build it now. */
-	nskb = sk_stream_alloc_skb(sk, probe_size, GFP_ATOMIC);
+	nskb = sk_stream_alloc_skb(sk, probe_size, GFP_ATOMIC, false);
 	if (!nskb)
 		return -1;
 	sk->sk_wmem_queued += nskb->truesize;
@@ -3190,7 +3190,7 @@ static int tcp_send_syn_data(struct sock *sk, struct sk_buff *syn)
 	/* limit to order-0 allocations */
 	space = min_t(size_t, space, SKB_MAX_HEAD(MAX_TCP_HEADER));
 
-	syn_data = sk_stream_alloc_skb(sk, space, sk->sk_allocation);
+	syn_data = sk_stream_alloc_skb(sk, space, sk->sk_allocation, false);
 	if (!syn_data)
 		goto fallback;
 	syn_data->ip_summed = CHECKSUM_PARTIAL;
@@ -3256,7 +3256,7 @@ int tcp_connect(struct sock *sk)
 		return 0;
 	}
 
-	buff = sk_stream_alloc_skb(sk, 0, sk->sk_allocation);
+	buff = sk_stream_alloc_skb(sk, 0, sk->sk_allocation, true);
 	if (unlikely(!buff))
 		return -ENOBUFS;