提交 b3d05147 编写于 作者: E Eric Dumazet 提交者: David S. Miller

tcp: do not mess with listener sk_wmem_alloc

When removing sk_refcnt manipulation on synflood, I missed that
using skb_set_owner_w() was racy, if sk->sk_wmem_alloc had already
transitioned to 0.

We should hold sk_refcnt instead, but this is a big deal under attack.
(Doing so increase performance from 3.2 Mpps to 3.8 Mpps only)

In this patch, I chose to not attach a socket to syncookies skb.

Performance is now 5 Mpps instead of 3.2 Mpps.

Following patch will remove last known false sharing in
tcp_rcv_state_process()

Fixes: 3b24d854 ("tcp/dccp: do not touch listener sk_refcnt under synflood")
Signed-off-by: NEric Dumazet <edumazet@google.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>
上级 ac18dd9e
...@@ -452,10 +452,15 @@ struct sock *tcp_v4_syn_recv_sock(const struct sock *sk, struct sk_buff *skb, ...@@ -452,10 +452,15 @@ struct sock *tcp_v4_syn_recv_sock(const struct sock *sk, struct sk_buff *skb,
int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb); int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb);
int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len); int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len);
int tcp_connect(struct sock *sk); int tcp_connect(struct sock *sk);
enum tcp_synack_type {
TCP_SYNACK_NORMAL,
TCP_SYNACK_FASTOPEN,
TCP_SYNACK_COOKIE,
};
struct sk_buff *tcp_make_synack(const struct sock *sk, struct dst_entry *dst, struct sk_buff *tcp_make_synack(const struct sock *sk, struct dst_entry *dst,
struct request_sock *req, struct request_sock *req,
struct tcp_fastopen_cookie *foc, struct tcp_fastopen_cookie *foc,
bool attach_req); enum tcp_synack_type synack_type);
int tcp_disconnect(struct sock *sk, int flags); int tcp_disconnect(struct sock *sk, int flags);
void tcp_finish_connect(struct sock *sk, struct sk_buff *skb); void tcp_finish_connect(struct sock *sk, struct sk_buff *skb);
...@@ -1728,7 +1733,7 @@ struct tcp_request_sock_ops { ...@@ -1728,7 +1733,7 @@ struct tcp_request_sock_ops {
int (*send_synack)(const struct sock *sk, struct dst_entry *dst, int (*send_synack)(const struct sock *sk, struct dst_entry *dst,
struct flowi *fl, struct request_sock *req, struct flowi *fl, struct request_sock *req,
struct tcp_fastopen_cookie *foc, struct tcp_fastopen_cookie *foc,
bool attach_req); enum tcp_synack_type synack_type);
}; };
#ifdef CONFIG_SYN_COOKIES #ifdef CONFIG_SYN_COOKIES
......
...@@ -6327,7 +6327,7 @@ int tcp_conn_request(struct request_sock_ops *rsk_ops, ...@@ -6327,7 +6327,7 @@ int tcp_conn_request(struct request_sock_ops *rsk_ops,
} }
if (fastopen_sk) { if (fastopen_sk) {
af_ops->send_synack(fastopen_sk, dst, &fl, req, af_ops->send_synack(fastopen_sk, dst, &fl, req,
&foc, false); &foc, TCP_SYNACK_FASTOPEN);
/* Add the child socket directly into the accept queue */ /* Add the child socket directly into the accept queue */
inet_csk_reqsk_queue_add(sk, req, fastopen_sk); inet_csk_reqsk_queue_add(sk, req, fastopen_sk);
sk->sk_data_ready(sk); sk->sk_data_ready(sk);
...@@ -6337,8 +6337,9 @@ int tcp_conn_request(struct request_sock_ops *rsk_ops, ...@@ -6337,8 +6337,9 @@ int tcp_conn_request(struct request_sock_ops *rsk_ops,
tcp_rsk(req)->tfo_listener = false; tcp_rsk(req)->tfo_listener = false;
if (!want_cookie) if (!want_cookie)
inet_csk_reqsk_queue_hash_add(sk, req, TCP_TIMEOUT_INIT); inet_csk_reqsk_queue_hash_add(sk, req, TCP_TIMEOUT_INIT);
af_ops->send_synack(sk, dst, &fl, req, af_ops->send_synack(sk, dst, &fl, req, &foc,
&foc, !want_cookie); !want_cookie ? TCP_SYNACK_NORMAL :
TCP_SYNACK_COOKIE);
if (want_cookie) { if (want_cookie) {
reqsk_free(req); reqsk_free(req);
return 0; return 0;
......
...@@ -830,7 +830,7 @@ static int tcp_v4_send_synack(const struct sock *sk, struct dst_entry *dst, ...@@ -830,7 +830,7 @@ static int tcp_v4_send_synack(const struct sock *sk, struct dst_entry *dst,
struct flowi *fl, struct flowi *fl,
struct request_sock *req, struct request_sock *req,
struct tcp_fastopen_cookie *foc, struct tcp_fastopen_cookie *foc,
bool attach_req) enum tcp_synack_type synack_type)
{ {
const struct inet_request_sock *ireq = inet_rsk(req); const struct inet_request_sock *ireq = inet_rsk(req);
struct flowi4 fl4; struct flowi4 fl4;
...@@ -841,7 +841,7 @@ static int tcp_v4_send_synack(const struct sock *sk, struct dst_entry *dst, ...@@ -841,7 +841,7 @@ static int tcp_v4_send_synack(const struct sock *sk, struct dst_entry *dst,
if (!dst && (dst = inet_csk_route_req(sk, &fl4, req)) == NULL) if (!dst && (dst = inet_csk_route_req(sk, &fl4, req)) == NULL)
return -1; return -1;
skb = tcp_make_synack(sk, dst, req, foc, attach_req); skb = tcp_make_synack(sk, dst, req, foc, synack_type);
if (skb) { if (skb) {
__tcp_v4_send_check(skb, ireq->ir_loc_addr, ireq->ir_rmt_addr); __tcp_v4_send_check(skb, ireq->ir_loc_addr, ireq->ir_rmt_addr);
......
...@@ -2944,7 +2944,7 @@ int tcp_send_synack(struct sock *sk) ...@@ -2944,7 +2944,7 @@ int tcp_send_synack(struct sock *sk)
struct sk_buff *tcp_make_synack(const struct sock *sk, struct dst_entry *dst, struct sk_buff *tcp_make_synack(const struct sock *sk, struct dst_entry *dst,
struct request_sock *req, struct request_sock *req,
struct tcp_fastopen_cookie *foc, struct tcp_fastopen_cookie *foc,
bool attach_req) enum tcp_synack_type synack_type)
{ {
struct inet_request_sock *ireq = inet_rsk(req); struct inet_request_sock *ireq = inet_rsk(req);
const struct tcp_sock *tp = tcp_sk(sk); const struct tcp_sock *tp = tcp_sk(sk);
...@@ -2964,14 +2964,22 @@ struct sk_buff *tcp_make_synack(const struct sock *sk, struct dst_entry *dst, ...@@ -2964,14 +2964,22 @@ struct sk_buff *tcp_make_synack(const struct sock *sk, struct dst_entry *dst,
/* Reserve space for headers. */ /* Reserve space for headers. */
skb_reserve(skb, MAX_TCP_HEADER); skb_reserve(skb, MAX_TCP_HEADER);
if (attach_req) { switch (synack_type) {
case TCP_SYNACK_NORMAL:
skb_set_owner_w(skb, req_to_sk(req)); skb_set_owner_w(skb, req_to_sk(req));
} else { break;
case TCP_SYNACK_COOKIE:
/* Under synflood, we do not attach skb to a socket,
* to avoid false sharing.
*/
break;
case TCP_SYNACK_FASTOPEN:
/* sk is a const pointer, because we want to express multiple /* sk is a const pointer, because we want to express multiple
* cpu might call us concurrently. * cpu might call us concurrently.
* sk->sk_wmem_alloc in an atomic, we can promote to rw. * sk->sk_wmem_alloc in an atomic, we can promote to rw.
*/ */
skb_set_owner_w(skb, (struct sock *)sk); skb_set_owner_w(skb, (struct sock *)sk);
break;
} }
skb_dst_set(skb, dst); skb_dst_set(skb, dst);
...@@ -3516,7 +3524,7 @@ int tcp_rtx_synack(const struct sock *sk, struct request_sock *req) ...@@ -3516,7 +3524,7 @@ int tcp_rtx_synack(const struct sock *sk, struct request_sock *req)
int res; int res;
tcp_rsk(req)->txhash = net_tx_rndhash(); tcp_rsk(req)->txhash = net_tx_rndhash();
res = af_ops->send_synack(sk, NULL, &fl, req, NULL, true); res = af_ops->send_synack(sk, NULL, &fl, req, NULL, TCP_SYNACK_NORMAL);
if (!res) { if (!res) {
TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_RETRANSSEGS); TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_RETRANSSEGS);
NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPSYNRETRANS); NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPSYNRETRANS);
......
...@@ -439,7 +439,7 @@ static int tcp_v6_send_synack(const struct sock *sk, struct dst_entry *dst, ...@@ -439,7 +439,7 @@ static int tcp_v6_send_synack(const struct sock *sk, struct dst_entry *dst,
struct flowi *fl, struct flowi *fl,
struct request_sock *req, struct request_sock *req,
struct tcp_fastopen_cookie *foc, struct tcp_fastopen_cookie *foc,
bool attach_req) enum tcp_synack_type synack_type)
{ {
struct inet_request_sock *ireq = inet_rsk(req); struct inet_request_sock *ireq = inet_rsk(req);
struct ipv6_pinfo *np = inet6_sk(sk); struct ipv6_pinfo *np = inet6_sk(sk);
...@@ -452,7 +452,7 @@ static int tcp_v6_send_synack(const struct sock *sk, struct dst_entry *dst, ...@@ -452,7 +452,7 @@ static int tcp_v6_send_synack(const struct sock *sk, struct dst_entry *dst,
IPPROTO_TCP)) == NULL) IPPROTO_TCP)) == NULL)
goto done; goto done;
skb = tcp_make_synack(sk, dst, req, foc, attach_req); skb = tcp_make_synack(sk, dst, req, foc, synack_type);
if (skb) { if (skb) {
__tcp_v6_send_check(skb, &ireq->ir_v6_loc_addr, __tcp_v6_send_check(skb, &ireq->ir_v6_loc_addr,
......
Markdown is supported
0% .
You are about to add 0 people to the discussion. Proceed with caution.
先完成此消息的编辑!
想要评论请 注册