From 271b72c7fa82c2c7a795bc16896149933110672d Mon Sep 17 00:00:00 2001 From: Eric Dumazet Date: Wed, 29 Oct 2008 02:11:14 -0700 Subject: udp: RCU handling for Unicast packets. Goals are : 1) Optimizing handling of incoming Unicast UDP frames, so that no memory writes should happen in the fast path. Note: Multicasts and broadcasts still will need to take a lock, because doing a full lockless lookup in this case is difficult. 2) No expensive operations in the socket bind/unhash phases : - No expensive synchronize_rcu() calls. - No added rcu_head in socket structure, increasing memory needs, but more important, forcing us to use call_rcu() calls, that have the bad property of making sockets structure cold. (rcu grace period between socket freeing and its potential reuse make this socket being cold in CPU cache). David did a previous patch using call_rcu() and noticed a 20% impact on TCP connection rates. Quoting Cristopher Lameter : "Right. That results in cacheline cooldown. You'd want to recycle the object as they are cache hot on a per cpu basis. That is screwed up by the delayed regular rcu processing. We have seen multiple regressions due to cacheline cooldown. The only choice in cacheline hot sensitive areas is to deal with the complexity that comes with SLAB_DESTROY_BY_RCU or give up on RCU." - Because udp sockets are allocated from dedicated kmem_cache, use of SLAB_DESTROY_BY_RCU can help here. Theory of operation : --------------------- As the lookup is lockfree (using rcu_read_lock()/rcu_read_unlock()), special attention must be taken by readers and writers. Use of SLAB_DESTROY_BY_RCU is tricky too, because a socket can be freed, reused, inserted in a different chain or in worst case in the same chain while readers could do lookups in the same time. In order to avoid loops, a reader must check each socket found in a chain really belongs to the chain the reader was traversing. If it finds a mismatch, lookup must start again at the begining. This *restart* loop is the reason we had to use rdlock for the multicast case, because we dont want to send same message several times to the same socket. We use RCU only for fast path. Thus, /proc/net/udp still takes spinlocks. Signed-off-by: Eric Dumazet Signed-off-by: David S. Miller --- net/core/sock.c | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-) (limited to 'net/core/sock.c') diff --git a/net/core/sock.c b/net/core/sock.c index 5e2a3132a8c9..ded1eb5d2fd4 100644 --- a/net/core/sock.c +++ b/net/core/sock.c @@ -2042,7 +2042,8 @@ int proto_register(struct proto *prot, int alloc_slab) if (alloc_slab) { prot->slab = kmem_cache_create(prot->name, prot->obj_size, 0, - SLAB_HWCACHE_ALIGN, NULL); + SLAB_HWCACHE_ALIGN | prot->slab_flags, + NULL); if (prot->slab == NULL) { printk(KERN_CRIT "%s: Can't create sock SLAB cache!\n", -- cgit v1.2.3 From 3ab5aee7fe840b5b1b35a8d1ac11c3de5281e611 Mon Sep 17 00:00:00 2001 From: Eric Dumazet Date: Sun, 16 Nov 2008 19:40:17 -0800 Subject: net: Convert TCP & DCCP hash tables to use RCU / hlist_nulls RCU was added to UDP lookups, using a fast infrastructure : - sockets kmem_cache use SLAB_DESTROY_BY_RCU and dont pay the price of call_rcu() at freeing time. - hlist_nulls permits to use few memory barriers. This patch uses same infrastructure for TCP/DCCP established and timewait sockets. Thanks to SLAB_DESTROY_BY_RCU, no slowdown for applications using short lived TCP connections. A followup patch, converting rwlocks to spinlocks will even speedup this case. __inet_lookup_established() is pretty fast now we dont have to dirty a contended cache line (read_lock/read_unlock) Only established and timewait hashtable are converted to RCU (bind table and listen table are still using traditional locking) Signed-off-by: Eric Dumazet Signed-off-by: David S. Miller --- net/core/sock.c | 4 +++- 1 file changed, 3 insertions(+), 1 deletion(-) (limited to 'net/core/sock.c') diff --git a/net/core/sock.c b/net/core/sock.c index ded1eb5d2fd4..38de9c3f563b 100644 --- a/net/core/sock.c +++ b/net/core/sock.c @@ -2082,7 +2082,9 @@ int proto_register(struct proto *prot, int alloc_slab) prot->twsk_prot->twsk_slab = kmem_cache_create(timewait_sock_slab_name, prot->twsk_prot->twsk_obj_size, - 0, SLAB_HWCACHE_ALIGN, + 0, + SLAB_HWCACHE_ALIGN | + prot->slab_flags, NULL); if (prot->twsk_prot->twsk_slab == NULL) goto out_free_timewait_sock_slab_name; -- cgit v1.2.3 From 14e943db133489c98d426a0dcfce4a99c6e8ad97 Mon Sep 17 00:00:00 2001 From: Eric Dumazet Date: Wed, 19 Nov 2008 15:14:01 -0800 Subject: net: make /proc/net/protocols namespace aware Converting /proc/net/protocols to be namespace aware is quite easy and permits us to use sock_prot_inuse_get(). This provides seperate counters for each protocol. For example we can really count TCPv6 sockets and TCPv4 sockets, while previously, we had the same value, and this value was not namespace aware. Signed-off-by: Eric Dumazet Signed-off-by: Alexey Dobriyan Signed-off-by: David S. Miller --- net/core/sock.c | 29 ++++++++++++++++++++++++----- 1 file changed, 24 insertions(+), 5 deletions(-) (limited to 'net/core/sock.c') diff --git a/net/core/sock.c b/net/core/sock.c index 5a6fe4dfad46..a4e840e5a053 100644 --- a/net/core/sock.c +++ b/net/core/sock.c @@ -2174,7 +2174,7 @@ static void proto_seq_printf(struct seq_file *seq, struct proto *proto) "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n", proto->name, proto->obj_size, - proto->sockets_allocated != NULL ? atomic_read(proto->sockets_allocated) : -1, + sock_prot_inuse_get(seq_file_net(seq), proto), proto->memory_allocated != NULL ? atomic_read(proto->memory_allocated) : -1, proto->memory_pressure != NULL ? *proto->memory_pressure ? "yes" : "no" : "NI", proto->max_header, @@ -2228,7 +2228,8 @@ static const struct seq_operations proto_seq_ops = { static int proto_seq_open(struct inode *inode, struct file *file) { - return seq_open(file, &proto_seq_ops); + return seq_open_net(inode, file, &proto_seq_ops, + sizeof(struct seq_net_private)); } static const struct file_operations proto_seq_fops = { @@ -2236,13 +2237,31 @@ static const struct file_operations proto_seq_fops = { .open = proto_seq_open, .read = seq_read, .llseek = seq_lseek, - .release = seq_release, + .release = seq_release_net, +}; + +static __net_init int proto_init_net(struct net *net) +{ + if (!proc_net_fops_create(net, "protocols", S_IRUGO, &proto_seq_fops)) + return -ENOMEM; + + return 0; +} + +static __net_exit void proto_exit_net(struct net *net) +{ + proc_net_remove(net, "protocols"); +} + + +static __net_initdata struct pernet_operations proto_net_ops = { + .init = proto_init_net, + .exit = proto_exit_net, }; static int __init proto_init(void) { - /* register /proc/net/protocols */ - return proc_net_fops_create(&init_net, "protocols", S_IRUGO, &proto_seq_fops) == NULL ? -ENOBUFS : 0; + return register_pernet_subsys(&proto_net_ops); } subsys_initcall(proto_init); -- cgit v1.2.3 From 1748376b6626acf59c24e9592ac67b3fe2a0e026 Mon Sep 17 00:00:00 2001 From: Eric Dumazet Date: Tue, 25 Nov 2008 21:16:35 -0800 Subject: net: Use a percpu_counter for sockets_allocated Instead of using one atomic_t per protocol, use a percpu_counter for "sockets_allocated", to reduce cache line contention on heavy duty network servers. Note : We revert commit (248969ae31e1b3276fc4399d67ce29a5d81e6fd9 net: af_unix can make unix_nr_socks visbile in /proc), since it is not anymore used after sock_prot_inuse_add() addition Signed-off-by: Eric Dumazet Signed-off-by: David S. Miller --- net/core/sock.c | 10 +++++++--- 1 file changed, 7 insertions(+), 3 deletions(-) (limited to 'net/core/sock.c') diff --git a/net/core/sock.c b/net/core/sock.c index a4e840e5a053..7a081b647bf9 100644 --- a/net/core/sock.c +++ b/net/core/sock.c @@ -1071,7 +1071,7 @@ struct sock *sk_clone(const struct sock *sk, const gfp_t priority) newsk->sk_sleep = NULL; if (newsk->sk_prot->sockets_allocated) - atomic_inc(newsk->sk_prot->sockets_allocated); + percpu_counter_inc(newsk->sk_prot->sockets_allocated); } out: return newsk; @@ -1463,8 +1463,12 @@ int __sk_mem_schedule(struct sock *sk, int size, int kind) } if (prot->memory_pressure) { - if (!*prot->memory_pressure || - prot->sysctl_mem[2] > atomic_read(prot->sockets_allocated) * + int alloc; + + if (!*prot->memory_pressure) + return 1; + alloc = percpu_counter_read_positive(prot->sockets_allocated); + if (prot->sysctl_mem[2] > alloc * sk_mem_pages(sk->sk_wmem_queued + atomic_read(&sk->sk_rmem_alloc) + sk->sk_forward_alloc)) -- cgit v1.2.3 From 70355602879229c6f8bd694ec9c0814222bc4936 Mon Sep 17 00:00:00 2001 From: Eric Dumazet Date: Wed, 26 Nov 2008 01:08:18 -0800 Subject: net: release skb->dst in sock_queue_rcv_skb() When queuing a skb to sk->sk_receive_queue, we can release its dst, not anymore needed. Since current cpu did the dst_hold(), refcount is probably still hot int this cpu caches. This avoids readers to access the original dst to decrement its refcount, possibly a long time after packet reception. This should speedup UDP and RAW receive path. Signed-off-by: Eric Dumazet Signed-off-by: David S. Miller --- net/core/sock.c | 6 +++++- 1 file changed, 5 insertions(+), 1 deletion(-) (limited to 'net/core/sock.c') diff --git a/net/core/sock.c b/net/core/sock.c index 7a081b647bf9..b28764558a7d 100644 --- a/net/core/sock.c +++ b/net/core/sock.c @@ -289,7 +289,11 @@ int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) skb->dev = NULL; skb_set_owner_r(skb, sk); - + /* + * release dst right now while its hot + */ + dst_release(skb->dst); + skb->dst = NULL; /* Cache the SKB length before we tack it onto the receive * queue. Once it is added it no longer belongs to us and * may be freed by other threads of control pulling packets -- cgit v1.2.3 From 49ad9599d42da4787d5b3a19263440e0fcd4d1fc Mon Sep 17 00:00:00 2001 From: "David S. Miller" Date: Wed, 17 Dec 2008 22:11:38 -0800 Subject: Revert "net: release skb->dst in sock_queue_rcv_skb()" This reverts commit 70355602879229c6f8bd694ec9c0814222bc4936. As pointed out by Mark McLoughlin IP_PKTINFO cmsg data is one post-queueing user, so this optimization is not valid right now. Signed-off-by: David S. Miller --- net/core/sock.c | 6 +----- 1 file changed, 1 insertion(+), 5 deletions(-) (limited to 'net/core/sock.c') diff --git a/net/core/sock.c b/net/core/sock.c index ac4f0e79226b..f3a0d08cbb48 100644 --- a/net/core/sock.c +++ b/net/core/sock.c @@ -289,11 +289,7 @@ int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) skb->dev = NULL; skb_set_owner_r(skb, sk); - /* - * release dst right now while its hot - */ - dst_release(skb->dst); - skb->dst = NULL; + /* Cache the SKB length before we tack it onto the receive * queue. Once it is added it no longer belongs to us and * may be freed by other threads of control pulling packets -- cgit v1.2.3