Merge branch 'net-reduce-tcp_memory_allocated-inflation'

Eric Dumazet says:

====================
net: reduce tcp_memory_allocated inflation

Hosts with a lot of sockets tend to hit so called TCP memory pressure,
leading to very bad TCP performance and/or OOM.

The problem is that some TCP sockets can hold up to 2MB of 'forward
allocations' in their per-socket cache (sk->sk_forward_alloc),
and there is no mechanism to make them relinquish their share
under mem pressure.
Only under some potentially rare events their share is reclaimed,
one socket at a time.

In this series, I implemented a per-cpu cache instead of a per-socket one.

Each CPU has a +1/-1 MB (256 pages on x86) forward alloc cache, in order
to not dirty tcp_memory_allocated shared cache line too often.

We keep sk->sk_forward_alloc values as small as possible, to meet
memcg page granularity constraint.

Note that memcg already has a per-cpu cache, although MEMCG_CHARGE_BATCH
is defined to 32 pages, which seems a bit small.

Note that while this cover letter mentions TCP, this work is generic
and supports TCP, UDP, DECNET, SCTP.
====================

Link: https://lore.kernel.org/r/20220609063412.2205738-1-eric.dumazet@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

1 files changed, 1 insertions, 5 deletions

diff --git a/net/ipv4/tcp_input.c b/net/ipv4/tcp_input.c
index 2e2a9ece9af2..fdc7beb81b68 100644
--- a/net/ipv4/tcp_input.c
+++ b/net/ipv4/tcp_input.c
@@ -805,7 +805,6 @@ static void tcp_event_data_recv(struct sock *sk, struct sk_buff *skb)
 			 * restart window, so that we send ACKs quickly.
 			 */
 			tcp_incr_quickack(sk, TCP_MAX_QUICKACKS);
-			sk_mem_reclaim(sk);
 		}
 	}
 	icsk->icsk_ack.lrcvtime = now;
@@ -4390,7 +4389,6 @@ void tcp_fin(struct sock *sk)
 	skb_rbtree_purge(&tp->out_of_order_queue);
 	if (tcp_is_sack(tp))
 		tcp_sack_reset(&tp->rx_opt);
-	sk_mem_reclaim(sk);
 
 	if (!sock_flag(sk, SOCK_DEAD)) {
 		sk->sk_state_change(sk);
@@ -5287,7 +5285,7 @@ new_range:
 		    before(TCP_SKB_CB(skb)->end_seq, start)) {
 			/* Do not attempt collapsing tiny skbs */
 			if (range_truesize != head->truesize ||
-			    end - start >= SKB_WITH_OVERHEAD(SK_MEM_QUANTUM)) {
+			    end - start >= SKB_WITH_OVERHEAD(PAGE_SIZE)) {
 				tcp_collapse(sk, NULL, &tp->out_of_order_queue,
 					     head, skb, start, end);
 			} else {
@@ -5336,7 +5334,6 @@ static bool tcp_prune_ofo_queue(struct sock *sk)
 		tcp_drop_reason(sk, rb_to_skb(node),
 				SKB_DROP_REASON_TCP_OFO_QUEUE_PRUNE);
 		if (!prev || goal <= 0) {
-			sk_mem_reclaim(sk);
 			if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf &&
 			    !tcp_under_memory_pressure(sk))
 				break;
@@ -5383,7 +5380,6 @@ static int tcp_prune_queue(struct sock *sk)
 			     skb_peek(&sk->sk_receive_queue),
 			     NULL,
 			     tp->copied_seq, tp->rcv_nxt);
-	sk_mem_reclaim(sk);
 
 	if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf)
 		return 0;