tcp: use a smaller percpu_counter batch size for sk_alloc

Currently, a percpu_counter with the default batch size (2*nr_cpus) is
used to record the total # of active sockets per protocol. This means
sk_sockets_allocated_read_positive() could be off by +/-2*(nr_cpus^2).
This under/over-estimation could lead to wrong memory suppression
conditions in __sk_raise_mem_allocated().
Fix this by using a more reasonable fixed batch size of 16.

See related commit cf86a086a180 ("net/dst: use a smaller percpu_counter
batch for dst entries accounting") that addresses a similar issue.

Signed-off-by: Wei Wang <weiwan@google.com>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Reviewed-by: Soheil Hassas Yeganeh <soheil@google.com>
Link: https://lore.kernel.org/r/20210202193408.1171634-1-weiwan@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

diff --git a/include/net/sock.h b/include/net/sock.h
index 129d200bccb46581f9959a8ee553acbed94466cf..690e496a0e7991d780784f663dc76dc394a843a2 100644 (file)
--- a/include/net/sock.h
+++ b/include/net/sock.h
@@ -1350,14 +1350,18 @@ sk_memory_allocated_sub(struct sock *sk, int amt)
        atomic_long_sub(amt, sk->sk_prot->memory_allocated);
 }
 
+#define SK_ALLOC_PERCPU_COUNTER_BATCH 16
+
 static inline void sk_sockets_allocated_dec(struct sock *sk)
 {
-       percpu_counter_dec(sk->sk_prot->sockets_allocated);
+       percpu_counter_add_batch(sk->sk_prot->sockets_allocated, -1,
+                                SK_ALLOC_PERCPU_COUNTER_BATCH);
 }
 
 static inline void sk_sockets_allocated_inc(struct sock *sk)
 {
-       percpu_counter_inc(sk->sk_prot->sockets_allocated);
+       percpu_counter_add_batch(sk->sk_prot->sockets_allocated, 1,
+                                SK_ALLOC_PERCPU_COUNTER_BATCH);
 }
 
 static inline u64