netfilter: nf_conntrack: use per-conntrack locks for protocol data

Introduce per-conntrack locks and use them instead of the global protocol
locks to avoid contention. Especially tcp_lock shows up very high in
profiles on larger machines.

This will also allow to simplify the upcoming reliable event delivery patches.

Signed-off-by: Patrick McHardy <kaber@trash.net>
This commit is contained in:
Patrick McHardy 2009-06-10 14:32:47 +02:00
Родитель a31e1ffd22
Коммит 440f0d5885
8 изменённых файлов: 49 добавлений и 55 удалений

Просмотреть файл

@ -93,6 +93,8 @@ struct nf_conn {
plus 1 for any connection(s) we are `master' for */
struct nf_conntrack ct_general;
spinlock_t lock;
/* XXX should I move this to the tail ? - Y.K */
/* These are my tuples; original and reply */
struct nf_conntrack_tuple_hash tuplehash[IP_CT_DIR_MAX];

Просмотреть файл

@ -59,11 +59,11 @@ struct nf_conntrack_l4proto
const struct nf_conntrack_tuple *);
/* Print out the private part of the conntrack. */
int (*print_conntrack)(struct seq_file *s, const struct nf_conn *);
int (*print_conntrack)(struct seq_file *s, struct nf_conn *);
/* convert protoinfo to nfnetink attributes */
int (*to_nlattr)(struct sk_buff *skb, struct nlattr *nla,
const struct nf_conn *ct);
struct nf_conn *ct);
/* Calculate protoinfo nlattr size */
int (*nlattr_size)(void);

Просмотреть файл

@ -519,6 +519,7 @@ struct nf_conn *nf_conntrack_alloc(struct net *net,
return ERR_PTR(-ENOMEM);
}
spin_lock_init(&ct->lock);
atomic_set(&ct->ct_general.use, 1);
ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple = *orig;
ct->tuplehash[IP_CT_DIR_REPLY].tuple = *repl;

Просмотреть файл

@ -143,7 +143,7 @@ nla_put_failure:
}
static inline int
ctnetlink_dump_protoinfo(struct sk_buff *skb, const struct nf_conn *ct)
ctnetlink_dump_protoinfo(struct sk_buff *skb, struct nf_conn *ct)
{
struct nf_conntrack_l4proto *l4proto;
struct nlattr *nest_proto;
@ -347,7 +347,7 @@ nla_put_failure:
static int
ctnetlink_fill_info(struct sk_buff *skb, u32 pid, u32 seq,
int event, const struct nf_conn *ct)
int event, struct nf_conn *ct)
{
struct nlmsghdr *nlh;
struct nfgenmsg *nfmsg;

Просмотреть файл

@ -24,8 +24,6 @@
#include <net/netfilter/nf_conntrack_l4proto.h>
#include <net/netfilter/nf_log.h>
static DEFINE_RWLOCK(dccp_lock);
/* Timeouts are based on values from RFC4340:
*
* - REQUEST:
@ -491,7 +489,7 @@ static int dccp_packet(struct nf_conn *ct, const struct sk_buff *skb,
return NF_ACCEPT;
}
write_lock_bh(&dccp_lock);
spin_lock_bh(&ct->lock);
role = ct->proto.dccp.role[dir];
old_state = ct->proto.dccp.state;
@ -535,13 +533,13 @@ static int dccp_packet(struct nf_conn *ct, const struct sk_buff *skb,
ct->proto.dccp.last_dir = dir;
ct->proto.dccp.last_pkt = type;
write_unlock_bh(&dccp_lock);
spin_unlock_bh(&ct->lock);
if (LOG_INVALID(net, IPPROTO_DCCP))
nf_log_packet(pf, 0, skb, NULL, NULL, NULL,
"nf_ct_dccp: invalid packet ignored ");
return NF_ACCEPT;
case CT_DCCP_INVALID:
write_unlock_bh(&dccp_lock);
spin_unlock_bh(&ct->lock);
if (LOG_INVALID(net, IPPROTO_DCCP))
nf_log_packet(pf, 0, skb, NULL, NULL, NULL,
"nf_ct_dccp: invalid state transition ");
@ -551,7 +549,7 @@ static int dccp_packet(struct nf_conn *ct, const struct sk_buff *skb,
ct->proto.dccp.last_dir = dir;
ct->proto.dccp.last_pkt = type;
ct->proto.dccp.state = new_state;
write_unlock_bh(&dccp_lock);
spin_unlock_bh(&ct->lock);
dn = dccp_pernet(net);
nf_ct_refresh_acct(ct, ctinfo, skb, dn->dccp_timeout[new_state]);
@ -617,18 +615,18 @@ static int dccp_print_tuple(struct seq_file *s,
ntohs(tuple->dst.u.dccp.port));
}
static int dccp_print_conntrack(struct seq_file *s, const struct nf_conn *ct)
static int dccp_print_conntrack(struct seq_file *s, struct nf_conn *ct)
{
return seq_printf(s, "%s ", dccp_state_names[ct->proto.dccp.state]);
}
#if defined(CONFIG_NF_CT_NETLINK) || defined(CONFIG_NF_CT_NETLINK_MODULE)
static int dccp_to_nlattr(struct sk_buff *skb, struct nlattr *nla,
const struct nf_conn *ct)
struct nf_conn *ct)
{
struct nlattr *nest_parms;
read_lock_bh(&dccp_lock);
spin_lock_bh(&ct->lock);
nest_parms = nla_nest_start(skb, CTA_PROTOINFO_DCCP | NLA_F_NESTED);
if (!nest_parms)
goto nla_put_failure;
@ -638,11 +636,11 @@ static int dccp_to_nlattr(struct sk_buff *skb, struct nlattr *nla,
NLA_PUT_BE64(skb, CTA_PROTOINFO_DCCP_HANDSHAKE_SEQ,
cpu_to_be64(ct->proto.dccp.handshake_seq));
nla_nest_end(skb, nest_parms);
read_unlock_bh(&dccp_lock);
spin_unlock_bh(&ct->lock);
return 0;
nla_put_failure:
read_unlock_bh(&dccp_lock);
spin_unlock_bh(&ct->lock);
return -1;
}
@ -673,7 +671,7 @@ static int nlattr_to_dccp(struct nlattr *cda[], struct nf_conn *ct)
return -EINVAL;
}
write_lock_bh(&dccp_lock);
spin_lock_bh(&ct->lock);
ct->proto.dccp.state = nla_get_u8(tb[CTA_PROTOINFO_DCCP_STATE]);
if (nla_get_u8(tb[CTA_PROTOINFO_DCCP_ROLE]) == CT_DCCP_ROLE_CLIENT) {
ct->proto.dccp.role[IP_CT_DIR_ORIGINAL] = CT_DCCP_ROLE_CLIENT;
@ -686,7 +684,7 @@ static int nlattr_to_dccp(struct nlattr *cda[], struct nf_conn *ct)
ct->proto.dccp.handshake_seq =
be64_to_cpu(nla_get_be64(tb[CTA_PROTOINFO_DCCP_HANDSHAKE_SEQ]));
}
write_unlock_bh(&dccp_lock);
spin_unlock_bh(&ct->lock);
return 0;
}

Просмотреть файл

@ -219,8 +219,7 @@ static int gre_print_tuple(struct seq_file *s,
}
/* print private data for conntrack */
static int gre_print_conntrack(struct seq_file *s,
const struct nf_conn *ct)
static int gre_print_conntrack(struct seq_file *s, struct nf_conn *ct)
{
return seq_printf(s, "timeout=%u, stream_timeout=%u ",
(ct->proto.gre.timeout / HZ),

Просмотреть файл

@ -25,9 +25,6 @@
#include <net/netfilter/nf_conntrack_l4proto.h>
#include <net/netfilter/nf_conntrack_ecache.h>
/* Protects ct->proto.sctp */
static DEFINE_RWLOCK(sctp_lock);
/* FIXME: Examine ipfilter's timeouts and conntrack transitions more
closely. They're more complex. --RR
@ -164,13 +161,13 @@ static int sctp_print_tuple(struct seq_file *s,
}
/* Print out the private part of the conntrack. */
static int sctp_print_conntrack(struct seq_file *s, const struct nf_conn *ct)
static int sctp_print_conntrack(struct seq_file *s, struct nf_conn *ct)
{
enum sctp_conntrack state;
read_lock_bh(&sctp_lock);
spin_lock_bh(&ct->lock);
state = ct->proto.sctp.state;
read_unlock_bh(&sctp_lock);
spin_unlock_bh(&ct->lock);
return seq_printf(s, "%s ", sctp_conntrack_names[state]);
}
@ -318,7 +315,7 @@ static int sctp_packet(struct nf_conn *ct,
}
old_state = new_state = SCTP_CONNTRACK_NONE;
write_lock_bh(&sctp_lock);
spin_lock_bh(&ct->lock);
for_each_sctp_chunk (skb, sch, _sch, offset, dataoff, count) {
/* Special cases of Verification tag check (Sec 8.5.1) */
if (sch->type == SCTP_CID_INIT) {
@ -371,7 +368,7 @@ static int sctp_packet(struct nf_conn *ct,
if (old_state != new_state)
nf_conntrack_event_cache(IPCT_PROTOINFO, ct);
}
write_unlock_bh(&sctp_lock);
spin_unlock_bh(&ct->lock);
nf_ct_refresh_acct(ct, ctinfo, skb, sctp_timeouts[new_state]);
@ -386,7 +383,7 @@ static int sctp_packet(struct nf_conn *ct,
return NF_ACCEPT;
out_unlock:
write_unlock_bh(&sctp_lock);
spin_unlock_bh(&ct->lock);
out:
return -NF_ACCEPT;
}
@ -469,11 +466,11 @@ static bool sctp_new(struct nf_conn *ct, const struct sk_buff *skb,
#include <linux/netfilter/nfnetlink_conntrack.h>
static int sctp_to_nlattr(struct sk_buff *skb, struct nlattr *nla,
const struct nf_conn *ct)
struct nf_conn *ct)
{
struct nlattr *nest_parms;
read_lock_bh(&sctp_lock);
spin_lock_bh(&ct->lock);
nest_parms = nla_nest_start(skb, CTA_PROTOINFO_SCTP | NLA_F_NESTED);
if (!nest_parms)
goto nla_put_failure;
@ -488,14 +485,14 @@ static int sctp_to_nlattr(struct sk_buff *skb, struct nlattr *nla,
CTA_PROTOINFO_SCTP_VTAG_REPLY,
ct->proto.sctp.vtag[IP_CT_DIR_REPLY]);
read_unlock_bh(&sctp_lock);
spin_unlock_bh(&ct->lock);
nla_nest_end(skb, nest_parms);
return 0;
nla_put_failure:
read_unlock_bh(&sctp_lock);
spin_unlock_bh(&ct->lock);
return -1;
}
@ -527,13 +524,13 @@ static int nlattr_to_sctp(struct nlattr *cda[], struct nf_conn *ct)
!tb[CTA_PROTOINFO_SCTP_VTAG_REPLY])
return -EINVAL;
write_lock_bh(&sctp_lock);
spin_lock_bh(&ct->lock);
ct->proto.sctp.state = nla_get_u8(tb[CTA_PROTOINFO_SCTP_STATE]);
ct->proto.sctp.vtag[IP_CT_DIR_ORIGINAL] =
nla_get_be32(tb[CTA_PROTOINFO_SCTP_VTAG_ORIGINAL]);
ct->proto.sctp.vtag[IP_CT_DIR_REPLY] =
nla_get_be32(tb[CTA_PROTOINFO_SCTP_VTAG_REPLY]);
write_unlock_bh(&sctp_lock);
spin_unlock_bh(&ct->lock);
return 0;
}

Просмотреть файл

@ -29,9 +29,6 @@
#include <net/netfilter/ipv4/nf_conntrack_ipv4.h>
#include <net/netfilter/ipv6/nf_conntrack_ipv6.h>
/* Protects ct->proto.tcp */
static DEFINE_RWLOCK(tcp_lock);
/* "Be conservative in what you do,
be liberal in what you accept from others."
If it's non-zero, we mark only out of window RST segments as INVALID. */
@ -309,13 +306,13 @@ static int tcp_print_tuple(struct seq_file *s,
}
/* Print out the private part of the conntrack. */
static int tcp_print_conntrack(struct seq_file *s, const struct nf_conn *ct)
static int tcp_print_conntrack(struct seq_file *s, struct nf_conn *ct)
{
enum tcp_conntrack state;
read_lock_bh(&tcp_lock);
spin_lock_bh(&ct->lock);
state = ct->proto.tcp.state;
read_unlock_bh(&tcp_lock);
spin_unlock_bh(&ct->lock);
return seq_printf(s, "%s ", tcp_conntrack_names[state]);
}
@ -725,14 +722,14 @@ void nf_conntrack_tcp_update(const struct sk_buff *skb,
end = segment_seq_plus_len(ntohl(tcph->seq), skb->len, dataoff, tcph);
write_lock_bh(&tcp_lock);
spin_lock_bh(&ct->lock);
/*
* We have to worry for the ack in the reply packet only...
*/
if (after(end, ct->proto.tcp.seen[dir].td_end))
ct->proto.tcp.seen[dir].td_end = end;
ct->proto.tcp.last_end = end;
write_unlock_bh(&tcp_lock);
spin_unlock_bh(&ct->lock);
pr_debug("tcp_update: sender end=%u maxend=%u maxwin=%u scale=%i "
"receiver end=%u maxend=%u maxwin=%u scale=%i\n",
sender->td_end, sender->td_maxend, sender->td_maxwin,
@ -841,7 +838,7 @@ static int tcp_packet(struct nf_conn *ct,
th = skb_header_pointer(skb, dataoff, sizeof(_tcph), &_tcph);
BUG_ON(th == NULL);
write_lock_bh(&tcp_lock);
spin_lock_bh(&ct->lock);
old_state = ct->proto.tcp.state;
dir = CTINFO2DIR(ctinfo);
index = get_conntrack_index(th);
@ -871,7 +868,7 @@ static int tcp_packet(struct nf_conn *ct,
&& ct->proto.tcp.last_index == TCP_RST_SET)) {
/* Attempt to reopen a closed/aborted connection.
* Delete this connection and look up again. */
write_unlock_bh(&tcp_lock);
spin_unlock_bh(&ct->lock);
/* Only repeat if we can actually remove the timer.
* Destruction may already be in progress in process
@ -907,7 +904,7 @@ static int tcp_packet(struct nf_conn *ct,
* that the client cannot but retransmit its SYN and
* thus initiate a clean new session.
*/
write_unlock_bh(&tcp_lock);
spin_unlock_bh(&ct->lock);
if (LOG_INVALID(net, IPPROTO_TCP))
nf_log_packet(pf, 0, skb, NULL, NULL, NULL,
"nf_ct_tcp: killing out of sync session ");
@ -920,7 +917,7 @@ static int tcp_packet(struct nf_conn *ct,
ct->proto.tcp.last_end =
segment_seq_plus_len(ntohl(th->seq), skb->len, dataoff, th);
write_unlock_bh(&tcp_lock);
spin_unlock_bh(&ct->lock);
if (LOG_INVALID(net, IPPROTO_TCP))
nf_log_packet(pf, 0, skb, NULL, NULL, NULL,
"nf_ct_tcp: invalid packet ignored ");
@ -929,7 +926,7 @@ static int tcp_packet(struct nf_conn *ct,
/* Invalid packet */
pr_debug("nf_ct_tcp: Invalid dir=%i index=%u ostate=%u\n",
dir, get_conntrack_index(th), old_state);
write_unlock_bh(&tcp_lock);
spin_unlock_bh(&ct->lock);
if (LOG_INVALID(net, IPPROTO_TCP))
nf_log_packet(pf, 0, skb, NULL, NULL, NULL,
"nf_ct_tcp: invalid state ");
@ -960,7 +957,7 @@ static int tcp_packet(struct nf_conn *ct,
if (!tcp_in_window(ct, &ct->proto.tcp, dir, index,
skb, dataoff, th, pf)) {
write_unlock_bh(&tcp_lock);
spin_unlock_bh(&ct->lock);
return -NF_ACCEPT;
}
in_window:
@ -989,7 +986,7 @@ static int tcp_packet(struct nf_conn *ct,
timeout = nf_ct_tcp_timeout_unacknowledged;
else
timeout = tcp_timeouts[new_state];
write_unlock_bh(&tcp_lock);
spin_unlock_bh(&ct->lock);
if (new_state != old_state)
nf_conntrack_event_cache(IPCT_PROTOINFO, ct);
@ -1106,12 +1103,12 @@ static bool tcp_new(struct nf_conn *ct, const struct sk_buff *skb,
#include <linux/netfilter/nfnetlink_conntrack.h>
static int tcp_to_nlattr(struct sk_buff *skb, struct nlattr *nla,
const struct nf_conn *ct)
struct nf_conn *ct)
{
struct nlattr *nest_parms;
struct nf_ct_tcp_flags tmp = {};
read_lock_bh(&tcp_lock);
spin_lock_bh(&ct->lock);
nest_parms = nla_nest_start(skb, CTA_PROTOINFO_TCP | NLA_F_NESTED);
if (!nest_parms)
goto nla_put_failure;
@ -1131,14 +1128,14 @@ static int tcp_to_nlattr(struct sk_buff *skb, struct nlattr *nla,
tmp.flags = ct->proto.tcp.seen[1].flags;
NLA_PUT(skb, CTA_PROTOINFO_TCP_FLAGS_REPLY,
sizeof(struct nf_ct_tcp_flags), &tmp);
read_unlock_bh(&tcp_lock);
spin_unlock_bh(&ct->lock);
nla_nest_end(skb, nest_parms);
return 0;
nla_put_failure:
read_unlock_bh(&tcp_lock);
spin_unlock_bh(&ct->lock);
return -1;
}
@ -1169,7 +1166,7 @@ static int nlattr_to_tcp(struct nlattr *cda[], struct nf_conn *ct)
nla_get_u8(tb[CTA_PROTOINFO_TCP_STATE]) >= TCP_CONNTRACK_MAX)
return -EINVAL;
write_lock_bh(&tcp_lock);
spin_lock_bh(&ct->lock);
if (tb[CTA_PROTOINFO_TCP_STATE])
ct->proto.tcp.state = nla_get_u8(tb[CTA_PROTOINFO_TCP_STATE]);
@ -1196,7 +1193,7 @@ static int nlattr_to_tcp(struct nlattr *cda[], struct nf_conn *ct)
ct->proto.tcp.seen[1].td_scale =
nla_get_u8(tb[CTA_PROTOINFO_TCP_WSCALE_REPLY]);
}
write_unlock_bh(&tcp_lock);
spin_unlock_bh(&ct->lock);
return 0;
}