summaryrefslogtreecommitdiffstats
diff options
context:
space:
mode:
-rw-r--r--include/linux/tcp.h2
-rw-r--r--include/net/tcp.h19
-rw-r--r--include/uapi/linux/tcp.h3
-rw-r--r--net/ipv4/tcp.c24
-rw-r--r--net/ipv4/tcp_ipv4.c10
-rw-r--r--net/ipv4/tcp_minisocks.c2
-rw-r--r--net/ipv4/tcp_output.c23
-rw-r--r--net/ipv6/tcp_ipv6.c1
8 files changed, 76 insertions, 8 deletions
diff --git a/include/linux/tcp.h b/include/linux/tcp.h
index 711361af9ce0..c23019a3b264 100644
--- a/include/linux/tcp.h
+++ b/include/linux/tcp.h
@@ -245,6 +245,7 @@ struct tcp_sock {
syn_smc:1; /* SYN includes SMC */
u32 tlp_high_seq; /* snd_nxt at the time of TLP retransmit. */
+ u32 tcp_tx_delay; /* delay (in usec) added to TX packets */
u64 tcp_wstamp_ns; /* departure time for next sent data packet */
u64 tcp_clock_cache; /* cache last tcp_clock_ns() (see tcp_mstamp_refresh()) */
@@ -436,6 +437,7 @@ struct tcp_timewait_sock {
u32 tw_last_oow_ack_time;
int tw_ts_recent_stamp;
+ u32 tw_tx_delay;
#ifdef CONFIG_TCP_MD5SIG
struct tcp_md5sig_key *tw_md5_key;
#endif
diff --git a/include/net/tcp.h b/include/net/tcp.h
index 204328b88412..49a178b8d5b2 100644
--- a/include/net/tcp.h
+++ b/include/net/tcp.h
@@ -2232,4 +2232,23 @@ void clean_acked_data_disable(struct inet_connection_sock *icsk);
void clean_acked_data_flush(void);
#endif
+DECLARE_STATIC_KEY_FALSE(tcp_tx_delay_enabled);
+static inline void tcp_add_tx_delay(struct sk_buff *skb,
+ const struct tcp_sock *tp)
+{
+ if (static_branch_unlikely(&tcp_tx_delay_enabled))
+ skb->skb_mstamp_ns += (u64)tp->tcp_tx_delay * NSEC_PER_USEC;
+}
+
+static inline void tcp_set_tx_time(struct sk_buff *skb,
+ const struct sock *sk)
+{
+ if (static_branch_unlikely(&tcp_tx_delay_enabled)) {
+ u32 delay = (sk->sk_state == TCP_TIME_WAIT) ?
+ tcp_twsk(sk)->tw_tx_delay : tcp_sk(sk)->tcp_tx_delay;
+
+ skb->skb_mstamp_ns = tcp_clock_ns() + (u64)delay * NSEC_PER_USEC;
+ }
+}
+
#endif /* _TCP_H */
diff --git a/include/uapi/linux/tcp.h b/include/uapi/linux/tcp.h
index b521464ea962..b3564f85a762 100644
--- a/include/uapi/linux/tcp.h
+++ b/include/uapi/linux/tcp.h
@@ -127,6 +127,9 @@ enum {
#define TCP_CM_INQ TCP_INQ
+#define TCP_TX_DELAY 37 /* delay outgoing packets by XX usec */
+
+
#define TCP_REPAIR_ON 1
#define TCP_REPAIR_OFF 0
#define TCP_REPAIR_OFF_NO_WP -1 /* Turn off without window probes */
diff --git a/net/ipv4/tcp.c b/net/ipv4/tcp.c
index bd0856ac680a..5542e3d778e6 100644
--- a/net/ipv4/tcp.c
+++ b/net/ipv4/tcp.c
@@ -2736,6 +2736,21 @@ static int tcp_repair_options_est(struct sock *sk,
return 0;
}
+DEFINE_STATIC_KEY_FALSE(tcp_tx_delay_enabled);
+EXPORT_SYMBOL(tcp_tx_delay_enabled);
+
+static void tcp_enable_tx_delay(void)
+{
+ if (!static_branch_unlikely(&tcp_tx_delay_enabled)) {
+ static int __tcp_tx_delay_enabled = 0;
+
+ if (cmpxchg(&__tcp_tx_delay_enabled, 0, 1) == 0) {
+ static_branch_enable(&tcp_tx_delay_enabled);
+ pr_info("TCP_TX_DELAY enabled\n");
+ }
+ }
+}
+
/*
* Socket option code for TCP.
*/
@@ -3087,6 +3102,11 @@ static int do_tcp_setsockopt(struct sock *sk, int level,
else
tp->recvmsg_inq = val;
break;
+ case TCP_TX_DELAY:
+ if (val)
+ tcp_enable_tx_delay();
+ tp->tcp_tx_delay = val;
+ break;
default:
err = -ENOPROTOOPT;
break;
@@ -3546,6 +3566,10 @@ static int do_tcp_getsockopt(struct sock *sk, int level,
val = tp->fastopen_no_cookie;
break;
+ case TCP_TX_DELAY:
+ val = tp->tcp_tx_delay;
+ break;
+
case TCP_TIMESTAMP:
val = tcp_time_stamp_raw() + tp->tsoffset;
break;
diff --git a/net/ipv4/tcp_ipv4.c b/net/ipv4/tcp_ipv4.c
index f059fbd81a84..1b7e9e1fbd3b 100644
--- a/net/ipv4/tcp_ipv4.c
+++ b/net/ipv4/tcp_ipv4.c
@@ -767,9 +767,11 @@ static void tcp_v4_send_reset(const struct sock *sk, struct sk_buff *skb)
arg.uid = sock_net_uid(net, sk && sk_fullsock(sk) ? sk : NULL);
local_bh_disable();
ctl_sk = this_cpu_read(*net->ipv4.tcp_sk);
- if (sk)
+ if (sk) {
ctl_sk->sk_mark = (sk->sk_state == TCP_TIME_WAIT) ?
inet_twsk(sk)->tw_mark : sk->sk_mark;
+ tcp_set_tx_time(skb, sk);
+ }
ip_send_unicast_reply(ctl_sk,
skb, &TCP_SKB_CB(skb)->header.h4.opt,
ip_hdr(skb)->saddr, ip_hdr(skb)->daddr,
@@ -859,9 +861,9 @@ static void tcp_v4_send_ack(const struct sock *sk,
arg.uid = sock_net_uid(net, sk_fullsock(sk) ? sk : NULL);
local_bh_disable();
ctl_sk = this_cpu_read(*net->ipv4.tcp_sk);
- if (sk)
- ctl_sk->sk_mark = (sk->sk_state == TCP_TIME_WAIT) ?
- inet_twsk(sk)->tw_mark : sk->sk_mark;
+ ctl_sk->sk_mark = (sk->sk_state == TCP_TIME_WAIT) ?
+ inet_twsk(sk)->tw_mark : sk->sk_mark;
+ tcp_set_tx_time(skb, sk);
ip_send_unicast_reply(ctl_sk,
skb, &TCP_SKB_CB(skb)->header.h4.opt,
ip_hdr(skb)->saddr, ip_hdr(skb)->daddr,
diff --git a/net/ipv4/tcp_minisocks.c b/net/ipv4/tcp_minisocks.c
index 11011e8386dc..8bcaf2586b68 100644
--- a/net/ipv4/tcp_minisocks.c
+++ b/net/ipv4/tcp_minisocks.c
@@ -274,7 +274,7 @@ void tcp_time_wait(struct sock *sk, int state, int timeo)
tcptw->tw_ts_recent_stamp = tp->rx_opt.ts_recent_stamp;
tcptw->tw_ts_offset = tp->tsoffset;
tcptw->tw_last_oow_ack_time = 0;
-
+ tcptw->tw_tx_delay = tp->tcp_tx_delay;
#if IS_ENABLED(CONFIG_IPV6)
if (tw->tw_family == PF_INET6) {
struct ipv6_pinfo *np = inet6_sk(sk);
diff --git a/net/ipv4/tcp_output.c b/net/ipv4/tcp_output.c
index f429e856e263..d954ff9069e8 100644
--- a/net/ipv4/tcp_output.c
+++ b/net/ipv4/tcp_output.c
@@ -1153,6 +1153,8 @@ static int __tcp_transmit_skb(struct sock *sk, struct sk_buff *skb,
memset(skb->cb, 0, max(sizeof(struct inet_skb_parm),
sizeof(struct inet6_skb_parm)));
+ tcp_add_tx_delay(skb, tp);
+
err = icsk->icsk_af_ops->queue_xmit(sk, skb, &inet->cork.fl);
if (unlikely(err > 0)) {
@@ -2234,6 +2236,18 @@ static bool tcp_small_queue_check(struct sock *sk, const struct sk_buff *skb,
sock_net(sk)->ipv4.sysctl_tcp_limit_output_bytes);
limit <<= factor;
+ if (static_branch_unlikely(&tcp_tx_delay_enabled) &&
+ tcp_sk(sk)->tcp_tx_delay) {
+ u64 extra_bytes = (u64)sk->sk_pacing_rate * tcp_sk(sk)->tcp_tx_delay;
+
+ /* TSQ is based on skb truesize sum (sk_wmem_alloc), so we
+ * approximate our needs assuming an ~100% skb->truesize overhead.
+ * USEC_PER_SEC is approximated by 2^20.
+ * do_div(extra_bytes, USEC_PER_SEC/2) is replaced by a right shift.
+ */
+ extra_bytes >>= (20 - 1);
+ limit += extra_bytes;
+ }
if (refcount_read(&sk->sk_wmem_alloc) > limit) {
/* Always send skb if rtx queue is empty.
* No need to wait for TX completion to call us back,
@@ -3212,6 +3226,7 @@ struct sk_buff *tcp_make_synack(const struct sock *sk, struct dst_entry *dst,
int tcp_header_size;
struct tcphdr *th;
int mss;
+ u64 now;
skb = alloc_skb(MAX_TCP_HEADER, GFP_ATOMIC);
if (unlikely(!skb)) {
@@ -3243,13 +3258,14 @@ struct sk_buff *tcp_make_synack(const struct sock *sk, struct dst_entry *dst,
mss = tcp_mss_clamp(tp, dst_metric_advmss(dst));
memset(&opts, 0, sizeof(opts));
+ now = tcp_clock_ns();
#ifdef CONFIG_SYN_COOKIES
if (unlikely(req->cookie_ts))
skb->skb_mstamp_ns = cookie_init_timestamp(req);
else
#endif
{
- skb->skb_mstamp_ns = tcp_clock_ns();
+ skb->skb_mstamp_ns = now;
if (!tcp_rsk(req)->snt_synack) /* Timestamp first SYNACK */
tcp_rsk(req)->snt_synack = tcp_skb_timestamp_us(skb);
}
@@ -3292,8 +3308,9 @@ struct sk_buff *tcp_make_synack(const struct sock *sk, struct dst_entry *dst,
rcu_read_unlock();
#endif
- /* Do not fool tcpdump (if any), clean our debris */
- skb->tstamp = 0;
+ skb->skb_mstamp_ns = now;
+ tcp_add_tx_delay(skb, tp);
+
return skb;
}
EXPORT_SYMBOL(tcp_make_synack);
diff --git a/net/ipv6/tcp_ipv6.c b/net/ipv6/tcp_ipv6.c
index ad7039137a20..5606b2131b65 100644
--- a/net/ipv6/tcp_ipv6.c
+++ b/net/ipv6/tcp_ipv6.c
@@ -892,6 +892,7 @@ static void tcp_v6_send_response(const struct sock *sk, struct sk_buff *skb, u32
} else {
mark = sk->sk_mark;
}
+ tcp_set_tx_time(buff, sk);
}
fl6.flowi6_mark = IP6_REPLY_MARK(net, skb->mark) ?: mark;
fl6.fl6_dport = t1->dest;