tcp: switch orphan_count to bare per-cpu counters

Use of percpu_counter structure to track count of orphaned
sockets is causing problems on modern hosts with 256 cpus
or more.

Stefan Bach reported a serious spinlock contention in real workloads,
that I was able to reproduce with a netfilter rule dropping
incoming FIN packets.

    53.56%  server  [kernel.kallsyms]      [k] queued_spin_lock_slowpath
            |
            ---queued_spin_lock_slowpath
               |
                --53.51%--_raw_spin_lock_irqsave
                          |
                           --53.51%--__percpu_counter_sum
                                     tcp_check_oom
                                     |
                                     |--39.03%--__tcp_close
                                     |          tcp_close
                                     |          inet_release
                                     |          inet6_release
                                     |          sock_close
                                     |          __fput
                                     |          ____fput
                                     |          task_work_run
                                     |          exit_to_usermode_loop
                                     |          do_syscall_64
                                     |          entry_SYSCALL_64_after_hwframe
                                     |          __GI___libc_close
                                     |
                                      --14.48%--tcp_out_of_resources
                                                tcp_write_timeout
                                                tcp_retransmit_timer
                                                tcp_write_timer_handler
                                                tcp_write_timer
                                                call_timer_fn
                                                expire_timers
                                                __run_timers
                                                run_timer_softirq
                                                __softirqentry_text_start

As explained in commit cf86a086a180 ("net/dst: use a smaller percpu_counter
batch for dst entries accounting"), default batch size is too big
for the default value of tcp_max_orphans (262144).

But even if we reduce batch sizes, there would still be cases
where the estimated count of orphans is beyond the limit,
and where tcp_too_many_orphans() has to call the expensive
percpu_counter_sum_positive().

One solution is to use plain per-cpu counters, and have
a timer to periodically refresh this cache.

Updating this cache every 100ms seems about right, tcp pressure
state is not radically changing over shorter periods.

percpu_counter was nice 15 years ago while hosts had less
than 16 cpus, not anymore by current standards.

v2: Fix the build issue for CONFIG_CRYPTO_DEV_CHELSIO_TLS=m,
    reported by kernel test robot <lkp@intel.com>
    Remove unused socket argument from tcp_too_many_orphans()

Fixes: dd24c00191d5 ("net: Use a percpu_counter for orphan_count")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Reported-by: Stefan Bach <sfb@google.com>
Cc: Neal Cardwell <ncardwell@google.com>
Acked-by: Neal Cardwell <ncardwell@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

1 files changed, 4 insertions, 10 deletions

diff --git a/net/dccp/proto.c b/net/dccp/proto.c
index abb5c596a817..fc44dadc778b 100644
--- a/net/dccp/proto.c
+++ b/net/dccp/proto.c
@@ -42,8 +42,8 @@ DEFINE_SNMP_STAT(struct dccp_mib, dccp_statistics) __read_mostly;
 
 EXPORT_SYMBOL_GPL(dccp_statistics);
 
-struct percpu_counter dccp_orphan_count;
-EXPORT_SYMBOL_GPL(dccp_orphan_count);
+DEFINE_PER_CPU(unsigned int, dccp_orphan_count);
+EXPORT_PER_CPU_SYMBOL_GPL(dccp_orphan_count);
 
 struct inet_hashinfo dccp_hashinfo;
 EXPORT_SYMBOL_GPL(dccp_hashinfo);
@@ -1055,7 +1055,7 @@ adjudge_to_death:
 	bh_lock_sock(sk);
 	WARN_ON(sock_owned_by_user(sk));
 
-	percpu_counter_inc(sk->sk_prot->orphan_count);
+	this_cpu_inc(dccp_orphan_count);
 
 	/* Have we already been destroyed by a softirq or backlog? */
 	if (state != DCCP_CLOSED && sk->sk_state == DCCP_CLOSED)
@@ -1115,13 +1115,10 @@ static int __init dccp_init(void)
 
 	BUILD_BUG_ON(sizeof(struct dccp_skb_cb) >
 		     sizeof_field(struct sk_buff, cb));
-	rc = percpu_counter_init(&dccp_orphan_count, 0, GFP_KERNEL);
-	if (rc)
-		goto out_fail;
 	inet_hashinfo_init(&dccp_hashinfo);
 	rc = inet_hashinfo2_init_mod(&dccp_hashinfo);
 	if (rc)
-		goto out_free_percpu;
+		goto out_fail;
 	rc = -ENOBUFS;
 	dccp_hashinfo.bind_bucket_cachep =
 		kmem_cache_create("dccp_bind_bucket",
@@ -1226,8 +1223,6 @@ out_free_bind_bucket_cachep:
 	kmem_cache_destroy(dccp_hashinfo.bind_bucket_cachep);
 out_free_hashinfo2:
 	inet_hashinfo2_free_mod(&dccp_hashinfo);
-out_free_percpu:
-	percpu_counter_destroy(&dccp_orphan_count);
 out_fail:
 	dccp_hashinfo.bhash = NULL;
 	dccp_hashinfo.ehash = NULL;
@@ -1250,7 +1245,6 @@ static void __exit dccp_fini(void)
 	dccp_ackvec_exit();
 	dccp_sysctl_exit();
 	inet_hashinfo2_free_mod(&dccp_hashinfo);
-	percpu_counter_destroy(&dccp_orphan_count);
 }
 
 module_init(dccp_init);