From ca065d0cf80fa547724440a8bf37f1e674d917c0 Mon Sep 17 00:00:00 2001 From: Eric Dumazet Date: Fri, 1 Apr 2016 08:52:13 -0700 Subject: udp: no longer use SLAB_DESTROY_BY_RCU Tom Herbert would like not touching UDP socket refcnt for encapsulated traffic. For this to happen, we need to use normal RCU rules, with a grace period before freeing a socket. UDP sockets are not short lived in the high usage case, so the added cost of call_rcu() should not be a concern. This actually removes a lot of complexity in UDP stack. Multicast receives no longer need to hold a bucket spinlock. Note that ip early demux still needs to take a reference on the socket. Same remark for functions used by xt_socket and xt_PROXY netfilter modules, but this might be changed later. Performance for a single UDP socket receiving flood traffic from many RX queues/cpus. Simple udp_rx using simple recvfrom() loop : 438 kpps instead of 374 kpps : 17 % increase of the peak rate. v2: Addressed Willem de Bruijn feedback in multicast handling - keep early demux break in __udp4_lib_demux_lookup() Signed-off-by: Eric Dumazet Cc: Tom Herbert Cc: Willem de Bruijn Tested-by: Tom Herbert Signed-off-by: David S. Miller --- net/ipv4/udp_diag.c | 18 +++++++++--------- 1 file changed, 9 insertions(+), 9 deletions(-) (limited to 'net/ipv4/udp_diag.c') diff --git a/net/ipv4/udp_diag.c b/net/ipv4/udp_diag.c index df1966f3b6ec..3d5ccf4b1412 100644 --- a/net/ipv4/udp_diag.c +++ b/net/ipv4/udp_diag.c @@ -36,10 +36,11 @@ static int udp_dump_one(struct udp_table *tbl, struct sk_buff *in_skb, const struct inet_diag_req_v2 *req) { int err = -EINVAL; - struct sock *sk; + struct sock *sk = NULL; struct sk_buff *rep; struct net *net = sock_net(in_skb->sk); + rcu_read_lock(); if (req->sdiag_family == AF_INET) sk = __udp4_lib_lookup(net, req->id.idiag_src[0], req->id.idiag_sport, @@ -54,9 +55,9 @@ static int udp_dump_one(struct udp_table *tbl, struct sk_buff *in_skb, req->id.idiag_dport, req->id.idiag_if, tbl, NULL); #endif - else - goto out_nosk; - + if (sk && !atomic_inc_not_zero(&sk->sk_refcnt)) + sk = NULL; + rcu_read_unlock(); err = -ENOENT; if (!sk) goto out_nosk; @@ -96,24 +97,23 @@ static void udp_dump(struct udp_table *table, struct sk_buff *skb, struct netlink_callback *cb, const struct inet_diag_req_v2 *r, struct nlattr *bc) { - int num, s_num, slot, s_slot; struct net *net = sock_net(skb->sk); + int num, s_num, slot, s_slot; s_slot = cb->args[0]; num = s_num = cb->args[1]; for (slot = s_slot; slot <= table->mask; s_num = 0, slot++) { - struct sock *sk; - struct hlist_nulls_node *node; struct udp_hslot *hslot = &table->hash[slot]; + struct sock *sk; num = 0; - if (hlist_nulls_empty(&hslot->head)) + if (hlist_empty(&hslot->head)) continue; spin_lock_bh(&hslot->lock); - sk_nulls_for_each(sk, node, &hslot->head) { + sk_for_each(sk, &hslot->head) { struct inet_sock *inet = inet_sk(sk); if (!net_eq(sock_net(sk), net)) -- cgit v1.2.3-59-g8ed1b