android_kernel_samsung_msm8976/include/net/inetpeer.h

178 lines
4.2 KiB
C
Raw Permalink Normal View History

/*
* INETPEER - A storage for permanent information about peers
*
* Authors: Andrey V. Savochkin <saw@msu.ru>
*/
#ifndef _NET_INETPEER_H
#define _NET_INETPEER_H
#include <linux/types.h>
#include <linux/init.h>
#include <linux/jiffies.h>
#include <linux/spinlock.h>
#include <linux/rtnetlink.h>
#include <net/ipv6.h>
#include <linux/atomic.h>
struct inetpeer_addr_base {
union {
__be32 a4;
__be32 a6[4];
};
};
struct inetpeer_addr {
struct inetpeer_addr_base addr;
__u16 family;
};
struct inet_peer {
/* group together avl_left,avl_right,v4daddr to speedup lookups */
struct inet_peer __rcu *avl_left, *avl_right;
struct inetpeer_addr daddr;
__u32 avl_height;
inetpeer: lower false sharing effect Profiles show false sharing in addr_compare() because refcnt/dtime changes dirty the first inet_peer cache line, where are lying the keys used at lookup time. If many cpus are calling inet_getpeer() and inet_putpeer(), or need frag ids, addr_compare() is in 2nd position in "perf top". Before patch, my udpflood bench (16 threads) on my 2x4x2 machine : 5784.00 9.7% csum_partial_copy_generic [kernel] 3356.00 5.6% addr_compare [kernel] 2638.00 4.4% fib_table_lookup [kernel] 2625.00 4.4% ip_fragment [kernel] 1934.00 3.2% neigh_lookup [kernel] 1617.00 2.7% udp_sendmsg [kernel] 1608.00 2.7% __ip_route_output_key [kernel] 1480.00 2.5% __ip_append_data [kernel] 1396.00 2.3% kfree [kernel] 1195.00 2.0% kmem_cache_free [kernel] 1157.00 1.9% inet_getpeer [kernel] 1121.00 1.9% neigh_resolve_output [kernel] 1012.00 1.7% dev_queue_xmit [kernel] # time ./udpflood.sh real 0m44.511s user 0m20.020s sys 11m22.780s # time ./udpflood.sh real 0m44.099s user 0m20.140s sys 11m15.870s After patch, no more addr_compare() in profiles : 4171.00 10.7% csum_partial_copy_generic [kernel] 1787.00 4.6% fib_table_lookup [kernel] 1756.00 4.5% ip_fragment [kernel] 1234.00 3.2% udp_sendmsg [kernel] 1191.00 3.0% neigh_lookup [kernel] 1118.00 2.9% __ip_append_data [kernel] 1022.00 2.6% kfree [kernel] 993.00 2.5% __ip_route_output_key [kernel] 841.00 2.2% neigh_resolve_output [kernel] 816.00 2.1% kmem_cache_free [kernel] 658.00 1.7% ia32_sysenter_target [kernel] 632.00 1.6% kmem_cache_alloc_node [kernel] # time ./udpflood.sh real 0m41.587s user 0m19.190s sys 10m36.370s # time ./udpflood.sh real 0m41.486s user 0m19.290s sys 10m33.650s Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2011-06-09 06:31:27 +00:00
u32 metrics[RTAX_MAX];
u32 rate_tokens; /* rate limiting for ICMP */
u32 n_redirects;
inetpeer: lower false sharing effect Profiles show false sharing in addr_compare() because refcnt/dtime changes dirty the first inet_peer cache line, where are lying the keys used at lookup time. If many cpus are calling inet_getpeer() and inet_putpeer(), or need frag ids, addr_compare() is in 2nd position in "perf top". Before patch, my udpflood bench (16 threads) on my 2x4x2 machine : 5784.00 9.7% csum_partial_copy_generic [kernel] 3356.00 5.6% addr_compare [kernel] 2638.00 4.4% fib_table_lookup [kernel] 2625.00 4.4% ip_fragment [kernel] 1934.00 3.2% neigh_lookup [kernel] 1617.00 2.7% udp_sendmsg [kernel] 1608.00 2.7% __ip_route_output_key [kernel] 1480.00 2.5% __ip_append_data [kernel] 1396.00 2.3% kfree [kernel] 1195.00 2.0% kmem_cache_free [kernel] 1157.00 1.9% inet_getpeer [kernel] 1121.00 1.9% neigh_resolve_output [kernel] 1012.00 1.7% dev_queue_xmit [kernel] # time ./udpflood.sh real 0m44.511s user 0m20.020s sys 11m22.780s # time ./udpflood.sh real 0m44.099s user 0m20.140s sys 11m15.870s After patch, no more addr_compare() in profiles : 4171.00 10.7% csum_partial_copy_generic [kernel] 1787.00 4.6% fib_table_lookup [kernel] 1756.00 4.5% ip_fragment [kernel] 1234.00 3.2% udp_sendmsg [kernel] 1191.00 3.0% neigh_lookup [kernel] 1118.00 2.9% __ip_append_data [kernel] 1022.00 2.6% kfree [kernel] 993.00 2.5% __ip_route_output_key [kernel] 841.00 2.2% neigh_resolve_output [kernel] 816.00 2.1% kmem_cache_free [kernel] 658.00 1.7% ia32_sysenter_target [kernel] 632.00 1.6% kmem_cache_alloc_node [kernel] # time ./udpflood.sh real 0m41.587s user 0m19.190s sys 10m36.370s # time ./udpflood.sh real 0m41.486s user 0m19.290s sys 10m33.650s Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2011-06-09 06:31:27 +00:00
unsigned long rate_last;
union {
struct list_head gc_list;
struct rcu_head gc_rcu;
};
/*
inetpeer: get rid of ip_id_count [ Upstream commit 73f156a6e8c1074ac6327e0abd1169e95eb66463 ] Ideally, we would need to generate IP ID using a per destination IP generator. linux kernels used inet_peer cache for this purpose, but this had a huge cost on servers disabling MTU discovery. 1) each inet_peer struct consumes 192 bytes 2) inetpeer cache uses a binary tree of inet_peer structs, with a nominal size of ~66000 elements under load. 3) lookups in this tree are hitting a lot of cache lines, as tree depth is about 20. 4) If server deals with many tcp flows, we have a high probability of not finding the inet_peer, allocating a fresh one, inserting it in the tree with same initial ip_id_count, (cf secure_ip_id()) 5) We garbage collect inet_peer aggressively. IP ID generation do not have to be 'perfect' Goal is trying to avoid duplicates in a short period of time, so that reassembly units have a chance to complete reassembly of fragments belonging to one message before receiving other fragments with a recycled ID. We simply use an array of generators, and a Jenkin hash using the dst IP as a key. ipv6_select_ident() is put back into net/ipv6/ip6_output.c where it belongs (it is only used from this file) secure_ip_id() and secure_ipv6_id() no longer are needed. Rename ip_select_ident_more() to ip_select_ident_segs() to avoid unnecessary decrement/increment of the number of segments. Signed-off-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2014-06-02 12:26:03 +00:00
* Once inet_peer is queued for deletion (refcnt == -1), following field
* is not available: rid
* We can share memory with rcu_head to help keep inet_peer small.
*/
union {
struct {
atomic_t rid; /* Frag reception counter */
};
struct rcu_head rcu;
struct inet_peer *gc_next;
};
inetpeer: lower false sharing effect Profiles show false sharing in addr_compare() because refcnt/dtime changes dirty the first inet_peer cache line, where are lying the keys used at lookup time. If many cpus are calling inet_getpeer() and inet_putpeer(), or need frag ids, addr_compare() is in 2nd position in "perf top". Before patch, my udpflood bench (16 threads) on my 2x4x2 machine : 5784.00 9.7% csum_partial_copy_generic [kernel] 3356.00 5.6% addr_compare [kernel] 2638.00 4.4% fib_table_lookup [kernel] 2625.00 4.4% ip_fragment [kernel] 1934.00 3.2% neigh_lookup [kernel] 1617.00 2.7% udp_sendmsg [kernel] 1608.00 2.7% __ip_route_output_key [kernel] 1480.00 2.5% __ip_append_data [kernel] 1396.00 2.3% kfree [kernel] 1195.00 2.0% kmem_cache_free [kernel] 1157.00 1.9% inet_getpeer [kernel] 1121.00 1.9% neigh_resolve_output [kernel] 1012.00 1.7% dev_queue_xmit [kernel] # time ./udpflood.sh real 0m44.511s user 0m20.020s sys 11m22.780s # time ./udpflood.sh real 0m44.099s user 0m20.140s sys 11m15.870s After patch, no more addr_compare() in profiles : 4171.00 10.7% csum_partial_copy_generic [kernel] 1787.00 4.6% fib_table_lookup [kernel] 1756.00 4.5% ip_fragment [kernel] 1234.00 3.2% udp_sendmsg [kernel] 1191.00 3.0% neigh_lookup [kernel] 1118.00 2.9% __ip_append_data [kernel] 1022.00 2.6% kfree [kernel] 993.00 2.5% __ip_route_output_key [kernel] 841.00 2.2% neigh_resolve_output [kernel] 816.00 2.1% kmem_cache_free [kernel] 658.00 1.7% ia32_sysenter_target [kernel] 632.00 1.6% kmem_cache_alloc_node [kernel] # time ./udpflood.sh real 0m41.587s user 0m19.190s sys 10m36.370s # time ./udpflood.sh real 0m41.486s user 0m19.290s sys 10m33.650s Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2011-06-09 06:31:27 +00:00
/* following fields might be frequently dirtied */
__u32 dtime; /* the time of last use of not referenced entries */
atomic_t refcnt;
};
struct inet_peer_base {
struct inet_peer __rcu *root;
seqlock_t lock;
u32 flush_seq;
int total;
};
#define INETPEER_BASE_BIT 0x1UL
static inline struct inet_peer *inetpeer_ptr(unsigned long val)
{
BUG_ON(val & INETPEER_BASE_BIT);
return (struct inet_peer *) val;
}
static inline struct inet_peer_base *inetpeer_base_ptr(unsigned long val)
{
if (!(val & INETPEER_BASE_BIT))
return NULL;
val &= ~INETPEER_BASE_BIT;
return (struct inet_peer_base *) val;
}
static inline bool inetpeer_ptr_is_peer(unsigned long val)
{
return !(val & INETPEER_BASE_BIT);
}
static inline void __inetpeer_ptr_set_peer(unsigned long *val, struct inet_peer *peer)
{
/* This implicitly clears INETPEER_BASE_BIT */
*val = (unsigned long) peer;
}
static inline bool inetpeer_ptr_set_peer(unsigned long *ptr, struct inet_peer *peer)
{
unsigned long val = (unsigned long) peer;
unsigned long orig = *ptr;
if (!(orig & INETPEER_BASE_BIT) ||
cmpxchg(ptr, orig, val) != orig)
return false;
return true;
}
static inline void inetpeer_init_ptr(unsigned long *ptr, struct inet_peer_base *base)
{
*ptr = (unsigned long) base | INETPEER_BASE_BIT;
}
static inline void inetpeer_transfer_peer(unsigned long *to, unsigned long *from)
{
unsigned long val = *from;
*to = val;
if (inetpeer_ptr_is_peer(val)) {
struct inet_peer *peer = inetpeer_ptr(val);
atomic_inc(&peer->refcnt);
}
}
extern void inet_peer_base_init(struct inet_peer_base *);
void inet_initpeers(void) __init;
#define INETPEER_METRICS_NEW (~(u32) 0)
static inline bool inet_metrics_new(const struct inet_peer *p)
{
return p->metrics[RTAX_LOCK-1] == INETPEER_METRICS_NEW;
}
/* can be called with or without local BH being disabled */
struct inet_peer *inet_getpeer(struct inet_peer_base *base,
const struct inetpeer_addr *daddr,
int create);
static inline struct inet_peer *inet_getpeer_v4(struct inet_peer_base *base,
__be32 v4daddr,
int create)
{
struct inetpeer_addr daddr;
daddr.addr.a4 = v4daddr;
daddr.family = AF_INET;
return inet_getpeer(base, &daddr, create);
}
static inline struct inet_peer *inet_getpeer_v6(struct inet_peer_base *base,
const struct in6_addr *v6daddr,
int create)
{
struct inetpeer_addr daddr;
*(struct in6_addr *)daddr.addr.a6 = *v6daddr;
daddr.family = AF_INET6;
return inet_getpeer(base, &daddr, create);
}
/* can be called from BH context or outside */
extern void inet_putpeer(struct inet_peer *p);
extern bool inet_peer_xrlim_allow(struct inet_peer *peer, int timeout);
extern void inetpeer_invalidate_tree(struct inet_peer_base *);
extern void inetpeer_invalidate_family(int family);
/*
inetpeer: get rid of ip_id_count [ Upstream commit 73f156a6e8c1074ac6327e0abd1169e95eb66463 ] Ideally, we would need to generate IP ID using a per destination IP generator. linux kernels used inet_peer cache for this purpose, but this had a huge cost on servers disabling MTU discovery. 1) each inet_peer struct consumes 192 bytes 2) inetpeer cache uses a binary tree of inet_peer structs, with a nominal size of ~66000 elements under load. 3) lookups in this tree are hitting a lot of cache lines, as tree depth is about 20. 4) If server deals with many tcp flows, we have a high probability of not finding the inet_peer, allocating a fresh one, inserting it in the tree with same initial ip_id_count, (cf secure_ip_id()) 5) We garbage collect inet_peer aggressively. IP ID generation do not have to be 'perfect' Goal is trying to avoid duplicates in a short period of time, so that reassembly units have a chance to complete reassembly of fragments belonging to one message before receiving other fragments with a recycled ID. We simply use an array of generators, and a Jenkin hash using the dst IP as a key. ipv6_select_ident() is put back into net/ipv6/ip6_output.c where it belongs (it is only used from this file) secure_ip_id() and secure_ipv6_id() no longer are needed. Rename ip_select_ident_more() to ip_select_ident_segs() to avoid unnecessary decrement/increment of the number of segments. Signed-off-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2014-06-02 12:26:03 +00:00
* temporary check to make sure we dont access rid, tcp_ts,
* tcp_ts_stamp if no refcount is taken on inet_peer
*/
static inline void inet_peer_refcheck(const struct inet_peer *p)
{
WARN_ON_ONCE(atomic_read(&p->refcnt) <= 0);
}
#endif /* _NET_INETPEER_H */