From patchwork Wed Sep 24 17:28:27 2008 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Corey Minyard X-Patchwork-Id: 1329 X-Patchwork-Delegate: davem@davemloft.net Return-Path: X-Original-To: patchwork-incoming@ozlabs.org Delivered-To: patchwork-incoming@ozlabs.org Received: from vger.kernel.org (vger.kernel.org [209.132.176.167]) by ozlabs.org (Postfix) with ESMTP id 706F5DE3C5 for ; Thu, 25 Sep 2008 04:35:01 +1000 (EST) Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1752164AbYIXSew (ORCPT ); Wed, 24 Sep 2008 14:34:52 -0400 Received: (majordomo@vger.kernel.org) by vger.kernel.org id S1751685AbYIXSew (ORCPT ); Wed, 24 Sep 2008 14:34:52 -0400 Received: from vms173003pub.verizon.net ([206.46.173.3]:60707 "EHLO vms173003pub.verizon.net" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1751550AbYIXSeu (ORCPT ); Wed, 24 Sep 2008 14:34:50 -0400 X-Greylist: delayed 3684 seconds by postgrey-1.27 at vger.kernel.org; Wed, 24 Sep 2008 14:34:50 EDT Received: from wf-rch.minyard.local ([96.226.138.225]) by vms173003.mailsrvcs.net (Sun Java System Messaging Server 6.2-6.01 (built Apr 3 2006)) with ESMTPA id <0K7P005BAN7GD1UO@vms173003.mailsrvcs.net>; Wed, 24 Sep 2008 12:28:28 -0500 (CDT) Received: from i2 (i2.minyard.local [192.168.27.126]) by wf-rch.minyard.local (Postfix) with ESMTP id D58871F7F2; Wed, 24 Sep 2008 12:28:27 -0500 (CDT) Date: Wed, 24 Sep 2008 12:28:27 -0500 From: Corey Minyard Subject: [PATCH 1/1] Use RCU for the UDP hash lock To: Linux Kernel , netdev@vger.kernel.org Reply-to: minyard@acm.org Message-id: <20080924172827.GA1573@minyard.local> MIME-version: 1.0 Content-type: text/plain; charset=us-ascii Content-disposition: inline User-Agent: Mutt/1.5.18 (2008-05-17) Sender: netdev-owner@vger.kernel.org Precedence: bulk List-ID: X-Mailing-List: netdev@vger.kernel.org From: Corey Minyard Convert access to the udp_hash table to use RCU. Signed-off-by: Corey Minyard --- include/linux/rculist.h | 19 +++++++++++++++++++ include/net/sock.h | 39 +++++++++++++++++++++++++++++++++++++++ include/net/udp.h | 9 +++++---- net/ipv4/udp.c | 36 +++++++++++++++++++++--------------- net/ipv6/udp.c | 13 +++++++------ 5 files changed, 91 insertions(+), 25 deletions(-) This patch is pretty straightforward; I've tested it a while and it seems to work properly with a test program that constantly creates and destroys UDP sockets while sending and receiving large numbers of packets on an SMP box. I think I've covered all the bases, though RCU is subtle. This doesn't make much difference when using no preempt or desktop, but it makes a huge difference when used with PREEMPT_RT. More than 10 times more UDP throughput on a 16-way machine. So I'm not sure if this belongs in the RT patch or in the mainstream kernel. -- To unsubscribe from this list: send the line "unsubscribe netdev" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html diff --git a/include/linux/rculist.h b/include/linux/rculist.h index eb4443c..4d3cc58 100644 --- a/include/linux/rculist.h +++ b/include/linux/rculist.h @@ -397,5 +397,24 @@ static inline void hlist_add_after_rcu(struct hlist_node *prev, ({ tpos = hlist_entry(pos, typeof(*tpos), member); 1; }); \ pos = rcu_dereference(pos->next)) + +/** + * hlist_for_each_entry_from_rcu - iterate over rcu list starting from pos + * @tpos: the type * to use as a loop cursor. + * @pos: the &struct hlist_node to use as a loop cursor. + * @head: the head for your list. + * @member: the name of the hlist_node within the struct. + * + * This list-traversal primitive may safely run concurrently with + * the _rcu list-mutation primitives such as hlist_add_head_rcu() + * as long as the traversal is guarded by rcu_read_lock(). + */ +#define hlist_for_each_entry_from_rcu(tpos, pos, member) \ + for (; \ + rcu_dereference(pos) && ({ prefetch(pos->next); 1; }) && \ + ({ tpos = hlist_entry(pos, typeof(*tpos), member); 1; }); \ + pos = pos->next) + + #endif /* __KERNEL__ */ #endif diff --git a/include/net/sock.h b/include/net/sock.h index 06c5259..ada44ad 100644 --- a/include/net/sock.h +++ b/include/net/sock.h @@ -42,6 +42,7 @@ #include #include +#include #include #include #include @@ -361,6 +362,27 @@ static __inline__ int sk_del_node_init(struct sock *sk) return rc; } +static inline int __sk_del_node_rcu(struct sock *sk) +{ + if (sk_hashed(sk)) { + hlist_del_rcu(&sk->sk_node); + return 1; + } + return 0; +} + +static inline int sk_del_node_rcu(struct sock *sk) +{ + int rc = __sk_del_node_rcu(sk); + + if (rc) { + /* paranoid for a while -acme */ + WARN_ON(atomic_read(&sk->sk_refcnt) == 1); + __sock_put(sk); + } + return rc; +} + static __inline__ void __sk_add_node(struct sock *sk, struct hlist_head *list) { hlist_add_head(&sk->sk_node, list); @@ -372,6 +394,18 @@ static __inline__ void sk_add_node(struct sock *sk, struct hlist_head *list) __sk_add_node(sk, list); } +static inline void __sk_add_node_rcu(struct sock *sk, + struct hlist_head *list) +{ + hlist_add_head_rcu(&sk->sk_node, list); +} + +static inline void sk_add_node_rcu(struct sock *sk, struct hlist_head *list) +{ + sock_hold(sk); + __sk_add_node_rcu(sk, list); +} + static __inline__ void __sk_del_bind_node(struct sock *sk) { __hlist_del(&sk->sk_bind_node); @@ -385,9 +419,14 @@ static __inline__ void sk_add_bind_node(struct sock *sk, #define sk_for_each(__sk, node, list) \ hlist_for_each_entry(__sk, node, list, sk_node) +#define sk_for_each_rcu(__sk, node, list) \ + hlist_for_each_entry_rcu(__sk, node, list, sk_node) #define sk_for_each_from(__sk, node) \ if (__sk && ({ node = &(__sk)->sk_node; 1; })) \ hlist_for_each_entry_from(__sk, node, sk_node) +#define sk_for_each_from_rcu(__sk, node) \ + if (__sk && ({ node = &(__sk)->sk_node; 1; })) \ + hlist_for_each_entry_from_rcu(__sk, node, sk_node) #define sk_for_each_continue(__sk, node) \ if (__sk && ({ node = &(__sk)->sk_node; 1; })) \ hlist_for_each_entry_continue(__sk, node, sk_node) diff --git a/include/net/udp.h b/include/net/udp.h index addcdc6..04181f8 100644 --- a/include/net/udp.h +++ b/include/net/udp.h @@ -51,7 +51,7 @@ struct udp_skb_cb { #define UDP_SKB_CB(__skb) ((struct udp_skb_cb *)((__skb)->cb)) extern struct hlist_head udp_hash[UDP_HTABLE_SIZE]; -extern rwlock_t udp_hash_lock; +extern spinlock_t udp_hash_wlock; /* Note: this must match 'valbool' in sock_setsockopt */ @@ -112,12 +112,13 @@ static inline void udp_lib_hash(struct sock *sk) static inline void udp_lib_unhash(struct sock *sk) { - write_lock_bh(&udp_hash_lock); - if (sk_del_node_init(sk)) { + spin_lock_bh(&udp_hash_wlock); + if (sk_del_node_rcu(sk)) { inet_sk(sk)->num = 0; sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1); } - write_unlock_bh(&udp_hash_lock); + spin_unlock_bh(&udp_hash_wlock); + synchronize_sched(); } static inline void udp_lib_close(struct sock *sk, long timeout) diff --git a/net/ipv4/udp.c b/net/ipv4/udp.c index 57e26fa..3aa04da 100644 --- a/net/ipv4/udp.c +++ b/net/ipv4/udp.c @@ -112,7 +112,8 @@ DEFINE_SNMP_STAT(struct udp_mib, udp_stats_in6) __read_mostly; EXPORT_SYMBOL(udp_stats_in6); struct hlist_head udp_hash[UDP_HTABLE_SIZE]; -DEFINE_RWLOCK(udp_hash_lock); +DEFINE_SPINLOCK(udp_hash_wlock); +EXPORT_SYMBOL(udp_hash_wlock); int sysctl_udp_mem[3] __read_mostly; int sysctl_udp_rmem_min __read_mostly; @@ -155,7 +156,7 @@ int udp_lib_get_port(struct sock *sk, unsigned short snum, int error = 1; struct net *net = sock_net(sk); - write_lock_bh(&udp_hash_lock); + spin_lock_bh(&udp_hash_wlock); if (!snum) { int i, low, high, remaining; @@ -225,12 +226,12 @@ gotit: sk->sk_hash = snum; if (sk_unhashed(sk)) { head = &udptable[udp_hashfn(net, snum)]; - sk_add_node(sk, head); + sk_add_node_rcu(sk, head); sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1); } error = 0; fail: - write_unlock_bh(&udp_hash_lock); + spin_unlock_bh(&udp_hash_wlock); return error; } @@ -260,8 +261,8 @@ static struct sock *__udp4_lib_lookup(struct net *net, __be32 saddr, unsigned short hnum = ntohs(dport); int badness = -1; - read_lock(&udp_hash_lock); - sk_for_each(sk, node, &udptable[udp_hashfn(net, hnum)]) { + rcu_read_lock(); + sk_for_each_rcu(sk, node, &udptable[udp_hashfn(net, hnum)]) { struct inet_sock *inet = inet_sk(sk); if (net_eq(sock_net(sk), net) && sk->sk_hash == hnum && @@ -296,9 +297,17 @@ static struct sock *__udp4_lib_lookup(struct net *net, __be32 saddr, } } } + /* + * Note that this is safe, even with an RCU lock. + * udp_lib_unhash() is the removal function, it calls + * synchronize_sched() and the socket counter cannot go to + * zero until it returns. So if we increment it inside the + * RCU read lock, it should never go to zero and then be + * incremented again. + */ if (result) sock_hold(result); - read_unlock(&udp_hash_lock); + rcu_read_unlock(); return result; } @@ -311,7 +320,7 @@ static inline struct sock *udp_v4_mcast_next(struct sock *sk, struct sock *s = sk; unsigned short hnum = ntohs(loc_port); - sk_for_each_from(s, node) { + sk_for_each_from_rcu(s, node) { struct inet_sock *inet = inet_sk(s); if (s->sk_hash != hnum || @@ -1094,7 +1103,7 @@ static int __udp4_lib_mcast_deliver(struct net *net, struct sk_buff *skb, struct sock *sk; int dif; - read_lock(&udp_hash_lock); + rcu_read_lock(); sk = sk_head(&udptable[udp_hashfn(net, ntohs(uh->dest))]); dif = skb->dev->ifindex; sk = udp_v4_mcast_next(sk, uh->dest, daddr, uh->source, saddr, dif); @@ -1120,7 +1129,7 @@ static int __udp4_lib_mcast_deliver(struct net *net, struct sk_buff *skb, } while (sknext); } else kfree_skb(skb); - read_unlock(&udp_hash_lock); + rcu_read_unlock(); return 0; } @@ -1566,9 +1575,8 @@ static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos) } static void *udp_seq_start(struct seq_file *seq, loff_t *pos) - __acquires(udp_hash_lock) { - read_lock(&udp_hash_lock); + rcu_read_lock(); return *pos ? udp_get_idx(seq, *pos-1) : SEQ_START_TOKEN; } @@ -1586,9 +1594,8 @@ static void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos) } static void udp_seq_stop(struct seq_file *seq, void *v) - __releases(udp_hash_lock) { - read_unlock(&udp_hash_lock); + rcu_read_unlock(); } static int udp_seq_open(struct inode *inode, struct file *file) @@ -1732,7 +1739,6 @@ void __init udp_init(void) EXPORT_SYMBOL(udp_disconnect); EXPORT_SYMBOL(udp_hash); -EXPORT_SYMBOL(udp_hash_lock); EXPORT_SYMBOL(udp_ioctl); EXPORT_SYMBOL(udp_prot); EXPORT_SYMBOL(udp_sendmsg); diff --git a/net/ipv6/udp.c b/net/ipv6/udp.c index a6aecf7..d9822ac 100644 --- a/net/ipv6/udp.c +++ b/net/ipv6/udp.c @@ -64,8 +64,8 @@ static struct sock *__udp6_lib_lookup(struct net *net, unsigned short hnum = ntohs(dport); int badness = -1; - read_lock(&udp_hash_lock); - sk_for_each(sk, node, &udptable[udp_hashfn(net, hnum)]) { + rcu_read_lock(); + sk_for_each_rcu(sk, node, &udptable[udp_hashfn(net, hnum)]) { struct inet_sock *inet = inet_sk(sk); if (net_eq(sock_net(sk), net) && sk->sk_hash == hnum && @@ -101,9 +101,10 @@ static struct sock *__udp6_lib_lookup(struct net *net, } } } + /* See comment in __udp4_lib_lookup on why this is safe. */ if (result) sock_hold(result); - read_unlock(&udp_hash_lock); + rcu_read_unlock(); return result; } @@ -322,7 +323,7 @@ static struct sock *udp_v6_mcast_next(struct sock *sk, struct sock *s = sk; unsigned short num = ntohs(loc_port); - sk_for_each_from(s, node) { + sk_for_each_from_rcu(s, node) { struct inet_sock *inet = inet_sk(s); if (sock_net(s) != sock_net(sk)) @@ -365,7 +366,7 @@ static int __udp6_lib_mcast_deliver(struct net *net, struct sk_buff *skb, const struct udphdr *uh = udp_hdr(skb); int dif; - read_lock(&udp_hash_lock); + rcu_read_lock(); sk = sk_head(&udptable[udp_hashfn(net, ntohs(uh->dest))]); dif = inet6_iif(skb); sk = udp_v6_mcast_next(sk, uh->dest, daddr, uh->source, saddr, dif); @@ -394,7 +395,7 @@ static int __udp6_lib_mcast_deliver(struct net *net, struct sk_buff *skb, sk_add_backlog(sk, skb); bh_unlock_sock(sk); out: - read_unlock(&udp_hash_lock); + rcu_read_unlock(); return 0; }