linux_dsm_epyc7002/include/net/inet_hashtables.h

403 lines
13 KiB
C
Raw Normal View History

/*
* INET An implementation of the TCP/IP protocol suite for the LINUX
* operating system. INET is implemented using the BSD Socket
* interface as the means of communication with the user level.
*
* Authors: Lotsa people, from code originally in tcp
*
* This program is free software; you can redistribute it and/or
* modify it under the terms of the GNU General Public License
* as published by the Free Software Foundation; either version
* 2 of the License, or (at your option) any later version.
*/
#ifndef _INET_HASHTABLES_H
#define _INET_HASHTABLES_H
#include <linux/interrupt.h>
#include <linux/ip.h>
#include <linux/ipv6.h>
#include <linux/list.h>
#include <linux/slab.h>
#include <linux/socket.h>
#include <linux/spinlock.h>
#include <linux/types.h>
#include <linux/wait.h>
#include <linux/vmalloc.h>
#include <net/inet_connection_sock.h>
#include <net/inet_sock.h>
#include <net/sock.h>
#include <net/route.h>
#include <net/tcp_states.h>
#include <net/netns/hash.h>
#include <linux/atomic.h>
#include <asm/byteorder.h>
/* This is for all connections with a full identity, no wildcards.
[NET]: change layout of ehash table ehash table layout is currently this one : First half of this table is used by sockets not in TIME_WAIT state Second half of it is used by sockets in TIME_WAIT state. This is non optimal because of for a given hash or socket, the two chain heads are located in separate cache lines. Moreover the locks of the second half are never used. If instead of this halving, we use two list heads in inet_ehash_bucket instead of only one, we probably can avoid one cache miss, and reduce ram usage, particularly if sizeof(rwlock_t) is big (various CONFIG_DEBUG_SPINLOCK, CONFIG_DEBUG_LOCK_ALLOC settings). So we still halves the table but we keep together related chains to speedup lookups and socket state change. In this patch I did not try to align struct inet_ehash_bucket, but a future patch could try to make this structure have a convenient size (a power of two or a multiple of L1_CACHE_SIZE). I guess rwlock will just vanish as soon as RCU is plugged into ehash :) , so maybe we dont need to scratch our heads to align the bucket... Note : In case struct inet_ehash_bucket is not a power of two, we could probably change alloc_large_system_hash() (in case it use __get_free_pages()) to free the unused space. It currently allocates a big zone, but the last quarter of it could be freed. Again, this should be a temporary 'problem'. Patch tested on ipv4 tcp only, but should be OK for IPV6 and DCCP. Signed-off-by: Eric Dumazet <dada1@cosmosbay.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2007-02-09 05:16:46 +07:00
* One chain is dedicated to TIME_WAIT sockets.
* I'll experiment with dynamic table growth later.
*/
struct inet_ehash_bucket {
struct hlist_nulls_head chain;
struct hlist_nulls_head twchain;
};
/* There are a few simple rules, which allow for local port reuse by
* an application. In essence:
*
* 1) Sockets bound to different interfaces may share a local port.
* Failing that, goto test 2.
* 2) If all sockets have sk->sk_reuse set, and none of them are in
* TCP_LISTEN state, the port may be shared.
* Failing that, goto test 3.
* 3) If all sockets are bound to a specific inet_sk(sk)->rcv_saddr local
* address, and none of them are the same, the port may be
* shared.
* Failing this, the port cannot be shared.
*
* The interesting point, is test #2. This is what an FTP server does
* all day. To optimize this case we use a specific flag bit defined
* below. As we add sockets to a bind bucket list, we perform a
* check of: (newsk->sk_reuse && (newsk->sk_state != TCP_LISTEN))
* As long as all sockets added to a bind bucket pass this test,
* the flag bit will be set.
* The resulting situation is that tcp_v[46]_verify_bind() can just check
* for this flag bit, if it is set and the socket trying to bind has
* sk->sk_reuse set, we don't even have to walk the owners list at all,
* we return that it is ok to bind this socket to the requested local port.
*
* Sounds like a lot of work, but it is worth it. In a more naive
* implementation (ie. current FreeBSD etc.) the entire list of ports
* must be walked for each data port opened by an ftp server. Needless
* to say, this does not scale at all. With a couple thousand FTP
* users logged onto your box, isn't it nice to know that new data
* ports are created in O(1) time? I thought so. ;-) -DaveM
*/
struct inet_bind_bucket {
#ifdef CONFIG_NET_NS
struct net *ib_net;
#endif
unsigned short port;
signed short fastreuse;
inet: Allowing more than 64k connections and heavily optimize bind(0) time. With simple extension to the binding mechanism, which allows to bind more than 64k sockets (or smaller amount, depending on sysctl parameters), we have to traverse the whole bind hash table to find out empty bucket. And while it is not a problem for example for 32k connections, bind() completion time grows exponentially (since after each successful binding we have to traverse one bucket more to find empty one) even if we start each time from random offset inside the hash table. So, when hash table is full, and we want to add another socket, we have to traverse the whole table no matter what, so effectivelly this will be the worst case performance and it will be constant. Attached picture shows bind() time depending on number of already bound sockets. Green area corresponds to the usual binding to zero port process, which turns on kernel port selection as described above. Red area is the bind process, when number of reuse-bound sockets is not limited by 64k (or sysctl parameters). The same exponential growth (hidden by the green area) before number of ports reaches sysctl limit. At this time bind hash table has exactly one reuse-enbaled socket in a bucket, but it is possible that they have different addresses. Actually kernel selects the first port to try randomly, so at the beginning bind will take roughly constant time, but with time number of port to check after random start will increase. And that will have exponential growth, but because of above random selection, not every next port selection will necessary take longer time than previous. So we have to consider the area below in the graph (if you could zoom it, you could find, that there are many different times placed there), so area can hide another. Blue area corresponds to the port selection optimization. This is rather simple design approach: hashtable now maintains (unprecise and racely updated) number of currently bound sockets, and when number of such sockets becomes greater than predefined value (I use maximum port range defined by sysctls), we stop traversing the whole bind hash table and just stop at first matching bucket after random start. Above limit roughly corresponds to the case, when bind hash table is full and we turned on mechanism of allowing to bind more reuse-enabled sockets, so it does not change behaviour of other sockets. Signed-off-by: Evgeniy Polyakov <zbr@ioremap.net> Tested-by: Denys Fedoryschenko <denys@visp.net.lb> Signed-off-by: David S. Miller <davem@davemloft.net>
2009-01-20 07:46:02 +07:00
int num_owners;
struct hlist_node node;
struct hlist_head owners;
};
static inline struct net *ib_net(struct inet_bind_bucket *ib)
{
return read_pnet(&ib->ib_net);
}
#define inet_bind_bucket_for_each(tb, pos, head) \
hlist_for_each_entry(tb, pos, head, node)
struct inet_bind_hashbucket {
spinlock_t lock;
struct hlist_head chain;
};
/*
* Sockets can be hashed in established or listening table
* We must use different 'nulls' end-of-chain value for listening
* hash table, or we might find a socket that was closed and
* reallocated/inserted into established hash table
*/
#define LISTENING_NULLS_BASE (1U << 29)
struct inet_listen_hashbucket {
spinlock_t lock;
struct hlist_nulls_head head;
};
/* This is for listening sockets, thus all sockets which possess wildcards. */
#define INET_LHTABLE_SIZE 32 /* Yes, really, this is all you need. */
struct inet_hashinfo {
/* This is for sockets with full identity only. Sockets here will
* always be without wildcards and will have the following invariant:
*
* TCP_ESTABLISHED <= sk->sk_state < TCP_CLOSE
*
[NET]: change layout of ehash table ehash table layout is currently this one : First half of this table is used by sockets not in TIME_WAIT state Second half of it is used by sockets in TIME_WAIT state. This is non optimal because of for a given hash or socket, the two chain heads are located in separate cache lines. Moreover the locks of the second half are never used. If instead of this halving, we use two list heads in inet_ehash_bucket instead of only one, we probably can avoid one cache miss, and reduce ram usage, particularly if sizeof(rwlock_t) is big (various CONFIG_DEBUG_SPINLOCK, CONFIG_DEBUG_LOCK_ALLOC settings). So we still halves the table but we keep together related chains to speedup lookups and socket state change. In this patch I did not try to align struct inet_ehash_bucket, but a future patch could try to make this structure have a convenient size (a power of two or a multiple of L1_CACHE_SIZE). I guess rwlock will just vanish as soon as RCU is plugged into ehash :) , so maybe we dont need to scratch our heads to align the bucket... Note : In case struct inet_ehash_bucket is not a power of two, we could probably change alloc_large_system_hash() (in case it use __get_free_pages()) to free the unused space. It currently allocates a big zone, but the last quarter of it could be freed. Again, this should be a temporary 'problem'. Patch tested on ipv4 tcp only, but should be OK for IPV6 and DCCP. Signed-off-by: Eric Dumazet <dada1@cosmosbay.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2007-02-09 05:16:46 +07:00
* TIME_WAIT sockets use a separate chain (twchain).
*/
struct inet_ehash_bucket *ehash;
spinlock_t *ehash_locks;
unsigned int ehash_mask;
unsigned int ehash_locks_mask;
/* Ok, let's try this, I give up, we do need a local binding
* TCP hash as well as the others for fast bind/connect.
*/
struct inet_bind_hashbucket *bhash;
unsigned int bhash_size;
/* 4 bytes hole on 64 bit */
struct kmem_cache *bind_bucket_cachep;
/* All the above members are written once at bootup and
* never written again _or_ are predominantly read-access.
*
* Now align to a new cache line as all the following members
* might be often dirty.
*/
/* All sockets in TCP_LISTEN state will be in here. This is the only
* table where wildcard'd TCP sockets can exist. Hash function here
* is just local port number.
*/
struct inet_listen_hashbucket listening_hash[INET_LHTABLE_SIZE]
____cacheline_aligned_in_smp;
atomic_t bsockets;
};
[INET]: speedup inet (tcp/dccp) lookups Arnaldo and I agreed it could be applied now, because I have other pending patches depending on this one (Thank you Arnaldo) (The other important patch moves skc_refcnt in a separate cache line, so that the SMP/NUMA performance doesnt suffer from cache line ping pongs) 1) First some performance data : -------------------------------- tcp_v4_rcv() wastes a *lot* of time in __inet_lookup_established() The most time critical code is : sk_for_each(sk, node, &head->chain) { if (INET_MATCH(sk, acookie, saddr, daddr, ports, dif)) goto hit; /* You sunk my battleship! */ } The sk_for_each() does use prefetch() hints but only the begining of "struct sock" is prefetched. As INET_MATCH first comparison uses inet_sk(__sk)->daddr, wich is far away from the begining of "struct sock", it has to bring into CPU cache cold cache line. Each iteration has to use at least 2 cache lines. This can be problematic if some chains are very long. 2) The goal ----------- The idea I had is to change things so that INET_MATCH() may return FALSE in 99% of cases only using the data already in the CPU cache, using one cache line per iteration. 3) Description of the patch --------------------------- Adds a new 'unsigned int skc_hash' field in 'struct sock_common', filling a 32 bits hole on 64 bits platform. struct sock_common { unsigned short skc_family; volatile unsigned char skc_state; unsigned char skc_reuse; int skc_bound_dev_if; struct hlist_node skc_node; struct hlist_node skc_bind_node; atomic_t skc_refcnt; + unsigned int skc_hash; struct proto *skc_prot; }; Store in this 32 bits field the full hash, not masked by (ehash_size - 1) Using this full hash as the first comparison done in INET_MATCH permits us immediatly skip the element without touching a second cache line in case of a miss. Suppress the sk_hashent/tw_hashent fields since skc_hash (aliased to sk_hash and tw_hash) already contains the slot number if we mask with (ehash_size - 1) File include/net/inet_hashtables.h 64 bits platforms : #define INET_MATCH(__sk, __hash, __cookie, __saddr, __daddr, __ports, __dif)\ (((__sk)->sk_hash == (__hash)) ((*((__u64 *)&(inet_sk(__sk)->daddr)))== (__cookie)) && \ ((*((__u32 *)&(inet_sk(__sk)->dport))) == (__ports)) && \ (!((__sk)->sk_bound_dev_if) || ((__sk)->sk_bound_dev_if == (__dif)))) 32bits platforms: #define TCP_IPV4_MATCH(__sk, __hash, __cookie, __saddr, __daddr, __ports, __dif)\ (((__sk)->sk_hash == (__hash)) && \ (inet_sk(__sk)->daddr == (__saddr)) && \ (inet_sk(__sk)->rcv_saddr == (__daddr)) && \ (!((__sk)->sk_bound_dev_if) || ((__sk)->sk_bound_dev_if == (__dif)))) - Adds a prefetch(head->chain.first) in __inet_lookup_established()/__tcp_v4_check_established() and __inet6_lookup_established()/__tcp_v6_check_established() and __dccp_v4_check_established() to bring into cache the first element of the list, before the {read|write}_lock(&head->lock); Signed-off-by: Eric Dumazet <dada1@cosmosbay.com> Acked-by: Arnaldo Carvalho de Melo <acme@ghostprotocols.net> Signed-off-by: David S. Miller <davem@davemloft.net>
2005-10-04 04:13:38 +07:00
static inline struct inet_ehash_bucket *inet_ehash_bucket(
struct inet_hashinfo *hashinfo,
unsigned int hash)
{
return &hashinfo->ehash[hash & hashinfo->ehash_mask];
}
static inline spinlock_t *inet_ehash_lockp(
struct inet_hashinfo *hashinfo,
unsigned int hash)
{
return &hashinfo->ehash_locks[hash & hashinfo->ehash_locks_mask];
}
static inline int inet_ehash_locks_alloc(struct inet_hashinfo *hashinfo)
{
unsigned int i, size = 256;
#if defined(CONFIG_PROVE_LOCKING)
unsigned int nr_pcpus = 2;
#else
unsigned int nr_pcpus = num_possible_cpus();
#endif
if (nr_pcpus >= 4)
size = 512;
if (nr_pcpus >= 8)
size = 1024;
if (nr_pcpus >= 16)
size = 2048;
if (nr_pcpus >= 32)
size = 4096;
if (sizeof(spinlock_t) != 0) {
#ifdef CONFIG_NUMA
if (size * sizeof(spinlock_t) > PAGE_SIZE)
hashinfo->ehash_locks = vmalloc(size * sizeof(spinlock_t));
else
#endif
hashinfo->ehash_locks = kmalloc(size * sizeof(spinlock_t),
GFP_KERNEL);
if (!hashinfo->ehash_locks)
return ENOMEM;
for (i = 0; i < size; i++)
spin_lock_init(&hashinfo->ehash_locks[i]);
}
hashinfo->ehash_locks_mask = size - 1;
return 0;
}
static inline void inet_ehash_locks_free(struct inet_hashinfo *hashinfo)
{
if (hashinfo->ehash_locks) {
#ifdef CONFIG_NUMA
unsigned int size = (hashinfo->ehash_locks_mask + 1) *
sizeof(spinlock_t);
if (size > PAGE_SIZE)
vfree(hashinfo->ehash_locks);
else
#endif
kfree(hashinfo->ehash_locks);
hashinfo->ehash_locks = NULL;
}
}
extern struct inet_bind_bucket *
inet_bind_bucket_create(struct kmem_cache *cachep,
struct net *net,
struct inet_bind_hashbucket *head,
const unsigned short snum);
extern void inet_bind_bucket_destroy(struct kmem_cache *cachep,
struct inet_bind_bucket *tb);
static inline int inet_bhashfn(struct net *net,
const __u16 lport, const int bhash_size)
{
return (lport + net_hash_mix(net)) & (bhash_size - 1);
}
extern void inet_bind_hash(struct sock *sk, struct inet_bind_bucket *tb,
const unsigned short snum);
/* These can have wildcards, don't try too hard. */
static inline int inet_lhashfn(struct net *net, const unsigned short num)
{
return (num + net_hash_mix(net)) & (INET_LHTABLE_SIZE - 1);
}
static inline int inet_sk_listen_hashfn(const struct sock *sk)
{
return inet_lhashfn(sock_net(sk), inet_sk(sk)->inet_num);
}
/* Caller must disable local BH processing. */
extern int __inet_inherit_port(struct sock *sk, struct sock *child);
[SOCK] proto: Add hashinfo member to struct proto This way we can remove TCP and DCCP specific versions of sk->sk_prot->get_port: both v4 and v6 use inet_csk_get_port sk->sk_prot->hash: inet_hash is directly used, only v6 need a specific version to deal with mapped sockets sk->sk_prot->unhash: both v4 and v6 use inet_hash directly struct inet_connection_sock_af_ops also gets a new member, bind_conflict, so that inet_csk_get_port can find the per family routine. Now only the lookup routines receive as a parameter a struct inet_hashtable. With this we further reuse code, reducing the difference among INET transport protocols. Eventually work has to be done on UDP and SCTP to make them share this infrastructure and get as a bonus inet_diag interfaces so that iproute can be used with these protocols. net-2.6/net/ipv4/inet_hashtables.c: struct proto | +8 struct inet_connection_sock_af_ops | +8 2 structs changed __inet_hash_nolisten | +18 __inet_hash | -210 inet_put_port | +8 inet_bind_bucket_create | +1 __inet_hash_connect | -8 5 functions changed, 27 bytes added, 218 bytes removed, diff: -191 net-2.6/net/core/sock.c: proto_seq_show | +3 1 function changed, 3 bytes added, diff: +3 net-2.6/net/ipv4/inet_connection_sock.c: inet_csk_get_port | +15 1 function changed, 15 bytes added, diff: +15 net-2.6/net/ipv4/tcp.c: tcp_set_state | -7 1 function changed, 7 bytes removed, diff: -7 net-2.6/net/ipv4/tcp_ipv4.c: tcp_v4_get_port | -31 tcp_v4_hash | -48 tcp_v4_destroy_sock | -7 tcp_v4_syn_recv_sock | -2 tcp_unhash | -179 5 functions changed, 267 bytes removed, diff: -267 net-2.6/net/ipv6/inet6_hashtables.c: __inet6_hash | +8 1 function changed, 8 bytes added, diff: +8 net-2.6/net/ipv4/inet_hashtables.c: inet_unhash | +190 inet_hash | +242 2 functions changed, 432 bytes added, diff: +432 vmlinux: 16 functions changed, 485 bytes added, 492 bytes removed, diff: -7 /home/acme/git/net-2.6/net/ipv6/tcp_ipv6.c: tcp_v6_get_port | -31 tcp_v6_hash | -7 tcp_v6_syn_recv_sock | -9 3 functions changed, 47 bytes removed, diff: -47 /home/acme/git/net-2.6/net/dccp/proto.c: dccp_destroy_sock | -7 dccp_unhash | -179 dccp_hash | -49 dccp_set_state | -7 dccp_done | +1 5 functions changed, 1 bytes added, 242 bytes removed, diff: -241 /home/acme/git/net-2.6/net/dccp/ipv4.c: dccp_v4_get_port | -31 dccp_v4_request_recv_sock | -2 2 functions changed, 33 bytes removed, diff: -33 /home/acme/git/net-2.6/net/dccp/ipv6.c: dccp_v6_get_port | -31 dccp_v6_hash | -7 dccp_v6_request_recv_sock | +5 3 functions changed, 5 bytes added, 38 bytes removed, diff: -33 Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2008-02-03 19:06:04 +07:00
extern void inet_put_port(struct sock *sk);
void inet_hashinfo_init(struct inet_hashinfo *h);
extern int __inet_hash_nolisten(struct sock *sk, struct inet_timewait_sock *tw);
[SOCK] proto: Add hashinfo member to struct proto This way we can remove TCP and DCCP specific versions of sk->sk_prot->get_port: both v4 and v6 use inet_csk_get_port sk->sk_prot->hash: inet_hash is directly used, only v6 need a specific version to deal with mapped sockets sk->sk_prot->unhash: both v4 and v6 use inet_hash directly struct inet_connection_sock_af_ops also gets a new member, bind_conflict, so that inet_csk_get_port can find the per family routine. Now only the lookup routines receive as a parameter a struct inet_hashtable. With this we further reuse code, reducing the difference among INET transport protocols. Eventually work has to be done on UDP and SCTP to make them share this infrastructure and get as a bonus inet_diag interfaces so that iproute can be used with these protocols. net-2.6/net/ipv4/inet_hashtables.c: struct proto | +8 struct inet_connection_sock_af_ops | +8 2 structs changed __inet_hash_nolisten | +18 __inet_hash | -210 inet_put_port | +8 inet_bind_bucket_create | +1 __inet_hash_connect | -8 5 functions changed, 27 bytes added, 218 bytes removed, diff: -191 net-2.6/net/core/sock.c: proto_seq_show | +3 1 function changed, 3 bytes added, diff: +3 net-2.6/net/ipv4/inet_connection_sock.c: inet_csk_get_port | +15 1 function changed, 15 bytes added, diff: +15 net-2.6/net/ipv4/tcp.c: tcp_set_state | -7 1 function changed, 7 bytes removed, diff: -7 net-2.6/net/ipv4/tcp_ipv4.c: tcp_v4_get_port | -31 tcp_v4_hash | -48 tcp_v4_destroy_sock | -7 tcp_v4_syn_recv_sock | -2 tcp_unhash | -179 5 functions changed, 267 bytes removed, diff: -267 net-2.6/net/ipv6/inet6_hashtables.c: __inet6_hash | +8 1 function changed, 8 bytes added, diff: +8 net-2.6/net/ipv4/inet_hashtables.c: inet_unhash | +190 inet_hash | +242 2 functions changed, 432 bytes added, diff: +432 vmlinux: 16 functions changed, 485 bytes added, 492 bytes removed, diff: -7 /home/acme/git/net-2.6/net/ipv6/tcp_ipv6.c: tcp_v6_get_port | -31 tcp_v6_hash | -7 tcp_v6_syn_recv_sock | -9 3 functions changed, 47 bytes removed, diff: -47 /home/acme/git/net-2.6/net/dccp/proto.c: dccp_destroy_sock | -7 dccp_unhash | -179 dccp_hash | -49 dccp_set_state | -7 dccp_done | +1 5 functions changed, 1 bytes added, 242 bytes removed, diff: -241 /home/acme/git/net-2.6/net/dccp/ipv4.c: dccp_v4_get_port | -31 dccp_v4_request_recv_sock | -2 2 functions changed, 33 bytes removed, diff: -33 /home/acme/git/net-2.6/net/dccp/ipv6.c: dccp_v6_get_port | -31 dccp_v6_hash | -7 dccp_v6_request_recv_sock | +5 3 functions changed, 5 bytes added, 38 bytes removed, diff: -33 Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2008-02-03 19:06:04 +07:00
extern void inet_hash(struct sock *sk);
extern void inet_unhash(struct sock *sk);
extern struct sock *__inet_lookup_listener(struct net *net,
struct inet_hashinfo *hashinfo,
const __be32 daddr,
const unsigned short hnum,
const int dif);
static inline struct sock *inet_lookup_listener(struct net *net,
struct inet_hashinfo *hashinfo,
__be32 daddr, __be16 dport, int dif)
{
return __inet_lookup_listener(net, hashinfo, daddr, ntohs(dport), dif);
}
/* Socket demux engine toys. */
/* What happens here is ugly; there's a pair of adjacent fields in
struct inet_sock; __be16 dport followed by __u16 num. We want to
search by pair, so we combine the keys into a single 32bit value
and compare with 32bit value read from &...->dport. Let's at least
make sure that it's not mixed with anything else...
On 64bit targets we combine comparisons with pair of adjacent __be32
fields in the same way.
*/
typedef __u32 __bitwise __portpair;
#ifdef __BIG_ENDIAN
#define INET_COMBINED_PORTS(__sport, __dport) \
((__force __portpair)(((__force __u32)(__be16)(__sport) << 16) | (__u32)(__dport)))
#else /* __LITTLE_ENDIAN */
#define INET_COMBINED_PORTS(__sport, __dport) \
((__force __portpair)(((__u32)(__dport) << 16) | (__force __u32)(__be16)(__sport)))
#endif
#if (BITS_PER_LONG == 64)
typedef __u64 __bitwise __addrpair;
#ifdef __BIG_ENDIAN
#define INET_ADDR_COOKIE(__name, __saddr, __daddr) \
const __addrpair __name = (__force __addrpair) ( \
(((__force __u64)(__be32)(__saddr)) << 32) | \
((__force __u64)(__be32)(__daddr)));
#else /* __LITTLE_ENDIAN */
#define INET_ADDR_COOKIE(__name, __saddr, __daddr) \
const __addrpair __name = (__force __addrpair) ( \
(((__force __u64)(__be32)(__daddr)) << 32) | \
((__force __u64)(__be32)(__saddr)));
#endif /* __BIG_ENDIAN */
#define INET_MATCH(__sk, __net, __hash, __cookie, __saddr, __daddr, __ports, __dif)\
(((__sk)->sk_hash == (__hash)) && net_eq(sock_net(__sk), (__net)) && \
((*((__addrpair *)&(inet_sk(__sk)->inet_daddr))) == (__cookie)) && \
((*((__portpair *)&(inet_sk(__sk)->inet_dport))) == (__ports)) && \
(!((__sk)->sk_bound_dev_if) || ((__sk)->sk_bound_dev_if == (__dif))))
#define INET_TW_MATCH(__sk, __net, __hash, __cookie, __saddr, __daddr, __ports, __dif)\
(((__sk)->sk_hash == (__hash)) && net_eq(sock_net(__sk), (__net)) && \
((*((__addrpair *)&(inet_twsk(__sk)->tw_daddr))) == (__cookie)) && \
((*((__portpair *)&(inet_twsk(__sk)->tw_dport))) == (__ports)) && \
(!((__sk)->sk_bound_dev_if) || ((__sk)->sk_bound_dev_if == (__dif))))
#else /* 32-bit arch */
#define INET_ADDR_COOKIE(__name, __saddr, __daddr)
#define INET_MATCH(__sk, __net, __hash, __cookie, __saddr, __daddr, __ports, __dif) \
(((__sk)->sk_hash == (__hash)) && net_eq(sock_net(__sk), (__net)) && \
(inet_sk(__sk)->inet_daddr == (__saddr)) && \
(inet_sk(__sk)->inet_rcv_saddr == (__daddr)) && \
((*((__portpair *)&(inet_sk(__sk)->inet_dport))) == (__ports)) && \
(!((__sk)->sk_bound_dev_if) || ((__sk)->sk_bound_dev_if == (__dif))))
#define INET_TW_MATCH(__sk, __net, __hash,__cookie, __saddr, __daddr, __ports, __dif) \
(((__sk)->sk_hash == (__hash)) && net_eq(sock_net(__sk), (__net)) && \
[INET]: speedup inet (tcp/dccp) lookups Arnaldo and I agreed it could be applied now, because I have other pending patches depending on this one (Thank you Arnaldo) (The other important patch moves skc_refcnt in a separate cache line, so that the SMP/NUMA performance doesnt suffer from cache line ping pongs) 1) First some performance data : -------------------------------- tcp_v4_rcv() wastes a *lot* of time in __inet_lookup_established() The most time critical code is : sk_for_each(sk, node, &head->chain) { if (INET_MATCH(sk, acookie, saddr, daddr, ports, dif)) goto hit; /* You sunk my battleship! */ } The sk_for_each() does use prefetch() hints but only the begining of "struct sock" is prefetched. As INET_MATCH first comparison uses inet_sk(__sk)->daddr, wich is far away from the begining of "struct sock", it has to bring into CPU cache cold cache line. Each iteration has to use at least 2 cache lines. This can be problematic if some chains are very long. 2) The goal ----------- The idea I had is to change things so that INET_MATCH() may return FALSE in 99% of cases only using the data already in the CPU cache, using one cache line per iteration. 3) Description of the patch --------------------------- Adds a new 'unsigned int skc_hash' field in 'struct sock_common', filling a 32 bits hole on 64 bits platform. struct sock_common { unsigned short skc_family; volatile unsigned char skc_state; unsigned char skc_reuse; int skc_bound_dev_if; struct hlist_node skc_node; struct hlist_node skc_bind_node; atomic_t skc_refcnt; + unsigned int skc_hash; struct proto *skc_prot; }; Store in this 32 bits field the full hash, not masked by (ehash_size - 1) Using this full hash as the first comparison done in INET_MATCH permits us immediatly skip the element without touching a second cache line in case of a miss. Suppress the sk_hashent/tw_hashent fields since skc_hash (aliased to sk_hash and tw_hash) already contains the slot number if we mask with (ehash_size - 1) File include/net/inet_hashtables.h 64 bits platforms : #define INET_MATCH(__sk, __hash, __cookie, __saddr, __daddr, __ports, __dif)\ (((__sk)->sk_hash == (__hash)) ((*((__u64 *)&(inet_sk(__sk)->daddr)))== (__cookie)) && \ ((*((__u32 *)&(inet_sk(__sk)->dport))) == (__ports)) && \ (!((__sk)->sk_bound_dev_if) || ((__sk)->sk_bound_dev_if == (__dif)))) 32bits platforms: #define TCP_IPV4_MATCH(__sk, __hash, __cookie, __saddr, __daddr, __ports, __dif)\ (((__sk)->sk_hash == (__hash)) && \ (inet_sk(__sk)->daddr == (__saddr)) && \ (inet_sk(__sk)->rcv_saddr == (__daddr)) && \ (!((__sk)->sk_bound_dev_if) || ((__sk)->sk_bound_dev_if == (__dif)))) - Adds a prefetch(head->chain.first) in __inet_lookup_established()/__tcp_v4_check_established() and __inet6_lookup_established()/__tcp_v6_check_established() and __dccp_v4_check_established() to bring into cache the first element of the list, before the {read|write}_lock(&head->lock); Signed-off-by: Eric Dumazet <dada1@cosmosbay.com> Acked-by: Arnaldo Carvalho de Melo <acme@ghostprotocols.net> Signed-off-by: David S. Miller <davem@davemloft.net>
2005-10-04 04:13:38 +07:00
(inet_twsk(__sk)->tw_daddr == (__saddr)) && \
(inet_twsk(__sk)->tw_rcv_saddr == (__daddr)) && \
((*((__portpair *)&(inet_twsk(__sk)->tw_dport))) == (__ports)) && \
(!((__sk)->sk_bound_dev_if) || ((__sk)->sk_bound_dev_if == (__dif))))
#endif /* 64-bit arch */
/*
* Sockets in TCP_CLOSE state are _always_ taken out of the hash, so we need
* not check it for lookups anymore, thanks Alexey. -DaveM
*
* Local BH must be disabled here.
*/
extern struct sock * __inet_lookup_established(struct net *net,
struct inet_hashinfo *hashinfo,
const __be32 saddr, const __be16 sport,
const __be32 daddr, const u16 hnum, const int dif);
static inline struct sock *
inet_lookup_established(struct net *net, struct inet_hashinfo *hashinfo,
const __be32 saddr, const __be16 sport,
const __be32 daddr, const __be16 dport,
const int dif)
{
return __inet_lookup_established(net, hashinfo, saddr, sport, daddr,
ntohs(dport), dif);
}
static inline struct sock *__inet_lookup(struct net *net,
struct inet_hashinfo *hashinfo,
const __be32 saddr, const __be16 sport,
const __be32 daddr, const __be16 dport,
const int dif)
{
u16 hnum = ntohs(dport);
struct sock *sk = __inet_lookup_established(net, hashinfo,
saddr, sport, daddr, hnum, dif);
return sk ? : __inet_lookup_listener(net, hashinfo, daddr, hnum, dif);
}
static inline struct sock *inet_lookup(struct net *net,
struct inet_hashinfo *hashinfo,
const __be32 saddr, const __be16 sport,
const __be32 daddr, const __be16 dport,
const int dif)
{
struct sock *sk;
local_bh_disable();
sk = __inet_lookup(net, hashinfo, saddr, sport, daddr, dport, dif);
local_bh_enable();
return sk;
}
static inline struct sock *__inet_lookup_skb(struct inet_hashinfo *hashinfo,
struct sk_buff *skb,
const __be16 sport,
const __be16 dport)
{
struct sock *sk;
const struct iphdr *iph = ip_hdr(skb);
if (unlikely(sk = skb_steal_sock(skb)))
return sk;
else
return __inet_lookup(dev_net(skb_dst(skb)->dev), hashinfo,
iph->saddr, sport,
iph->daddr, dport, inet_iif(skb));
}
extern int __inet_hash_connect(struct inet_timewait_death_row *death_row,
struct sock *sk,
u32 port_offset,
int (*check_established)(struct inet_timewait_death_row *,
struct sock *, __u16, struct inet_timewait_sock **),
int (*hash)(struct sock *sk, struct inet_timewait_sock *twp));
extern int inet_hash_connect(struct inet_timewait_death_row *death_row,
struct sock *sk);
#endif /* _INET_HASHTABLES_H */