Commit 4aabd8ef authored by David S. Miller's avatar David S. Miller

tcp: Move dynamnic metrics handling into seperate file.

Signed-off-by: default avatarDavid S. Miller <davem@davemloft.net>
parent ad7eee98
...@@ -388,6 +388,8 @@ extern void tcp_enter_frto(struct sock *sk); ...@@ -388,6 +388,8 @@ extern void tcp_enter_frto(struct sock *sk);
extern void tcp_enter_loss(struct sock *sk, int how); extern void tcp_enter_loss(struct sock *sk, int how);
extern void tcp_clear_retrans(struct tcp_sock *tp); extern void tcp_clear_retrans(struct tcp_sock *tp);
extern void tcp_update_metrics(struct sock *sk); extern void tcp_update_metrics(struct sock *sk);
extern void tcp_init_metrics(struct sock *sk);
extern void tcp_disable_fack(struct tcp_sock *tp);
extern void tcp_close(struct sock *sk, long timeout); extern void tcp_close(struct sock *sk, long timeout);
extern void tcp_init_sock(struct sock *sk); extern void tcp_init_sock(struct sock *sk);
extern unsigned int tcp_poll(struct file * file, struct socket *sock, extern unsigned int tcp_poll(struct file * file, struct socket *sock,
...@@ -556,6 +558,8 @@ static inline u32 __tcp_set_rto(const struct tcp_sock *tp) ...@@ -556,6 +558,8 @@ static inline u32 __tcp_set_rto(const struct tcp_sock *tp)
return (tp->srtt >> 3) + tp->rttvar; return (tp->srtt >> 3) + tp->rttvar;
} }
extern void tcp_set_rto(struct sock *sk);
static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd) static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd)
{ {
tp->pred_flags = htonl((tp->tcp_header_len << 26) | tp->pred_flags = htonl((tp->tcp_header_len << 26) |
......
...@@ -7,7 +7,7 @@ obj-y := route.o inetpeer.o protocol.o \ ...@@ -7,7 +7,7 @@ obj-y := route.o inetpeer.o protocol.o \
ip_output.o ip_sockglue.o inet_hashtables.o \ ip_output.o ip_sockglue.o inet_hashtables.o \
inet_timewait_sock.o inet_connection_sock.o \ inet_timewait_sock.o inet_connection_sock.o \
tcp.o tcp_input.o tcp_output.o tcp_timer.o tcp_ipv4.o \ tcp.o tcp_input.o tcp_output.o tcp_timer.o tcp_ipv4.o \
tcp_minisocks.o tcp_cong.o \ tcp_minisocks.o tcp_cong.o tcp_metrics.o \
datagram.o raw.o udp.o udplite.o \ datagram.o raw.o udp.o udplite.o \
arp.o icmp.o devinet.o af_inet.o igmp.o \ arp.o icmp.o devinet.o af_inet.o igmp.o \
fib_frontend.o fib_semantics.o fib_trie.o \ fib_frontend.o fib_semantics.o fib_trie.o \
......
...@@ -93,7 +93,6 @@ int sysctl_tcp_rfc1337 __read_mostly; ...@@ -93,7 +93,6 @@ int sysctl_tcp_rfc1337 __read_mostly;
int sysctl_tcp_max_orphans __read_mostly = NR_FILE; int sysctl_tcp_max_orphans __read_mostly = NR_FILE;
int sysctl_tcp_frto __read_mostly = 2; int sysctl_tcp_frto __read_mostly = 2;
int sysctl_tcp_frto_response __read_mostly; int sysctl_tcp_frto_response __read_mostly;
int sysctl_tcp_nometrics_save __read_mostly;
int sysctl_tcp_thin_dupack __read_mostly; int sysctl_tcp_thin_dupack __read_mostly;
...@@ -701,7 +700,7 @@ static void tcp_rtt_estimator(struct sock *sk, const __u32 mrtt) ...@@ -701,7 +700,7 @@ static void tcp_rtt_estimator(struct sock *sk, const __u32 mrtt)
/* Calculate rto without backoff. This is the second half of Van Jacobson's /* Calculate rto without backoff. This is the second half of Van Jacobson's
* routine referred to above. * routine referred to above.
*/ */
static inline void tcp_set_rto(struct sock *sk) void tcp_set_rto(struct sock *sk)
{ {
const struct tcp_sock *tp = tcp_sk(sk); const struct tcp_sock *tp = tcp_sk(sk);
/* Old crap is replaced with new one. 8) /* Old crap is replaced with new one. 8)
...@@ -728,109 +727,6 @@ static inline void tcp_set_rto(struct sock *sk) ...@@ -728,109 +727,6 @@ static inline void tcp_set_rto(struct sock *sk)
tcp_bound_rto(sk); tcp_bound_rto(sk);
} }
/* Save metrics learned by this TCP session.
This function is called only, when TCP finishes successfully
i.e. when it enters TIME-WAIT or goes from LAST-ACK to CLOSE.
*/
void tcp_update_metrics(struct sock *sk)
{
struct tcp_sock *tp = tcp_sk(sk);
struct dst_entry *dst = __sk_dst_get(sk);
if (sysctl_tcp_nometrics_save)
return;
if (dst && (dst->flags & DST_HOST)) {
const struct inet_connection_sock *icsk = inet_csk(sk);
int m;
unsigned long rtt;
dst_confirm(dst);
if (icsk->icsk_backoff || !tp->srtt) {
/* This session failed to estimate rtt. Why?
* Probably, no packets returned in time.
* Reset our results.
*/
if (!(dst_metric_locked(dst, RTAX_RTT)))
dst_metric_set(dst, RTAX_RTT, 0);
return;
}
rtt = dst_metric_rtt(dst, RTAX_RTT);
m = rtt - tp->srtt;
/* If newly calculated rtt larger than stored one,
* store new one. Otherwise, use EWMA. Remember,
* rtt overestimation is always better than underestimation.
*/
if (!(dst_metric_locked(dst, RTAX_RTT))) {
if (m <= 0)
set_dst_metric_rtt(dst, RTAX_RTT, tp->srtt);
else
set_dst_metric_rtt(dst, RTAX_RTT, rtt - (m >> 3));
}
if (!(dst_metric_locked(dst, RTAX_RTTVAR))) {
unsigned long var;
if (m < 0)
m = -m;
/* Scale deviation to rttvar fixed point */
m >>= 1;
if (m < tp->mdev)
m = tp->mdev;
var = dst_metric_rtt(dst, RTAX_RTTVAR);
if (m >= var)
var = m;
else
var -= (var - m) >> 2;
set_dst_metric_rtt(dst, RTAX_RTTVAR, var);
}
if (tcp_in_initial_slowstart(tp)) {
/* Slow start still did not finish. */
if (dst_metric(dst, RTAX_SSTHRESH) &&
!dst_metric_locked(dst, RTAX_SSTHRESH) &&
(tp->snd_cwnd >> 1) > dst_metric(dst, RTAX_SSTHRESH))
dst_metric_set(dst, RTAX_SSTHRESH, tp->snd_cwnd >> 1);
if (!dst_metric_locked(dst, RTAX_CWND) &&
tp->snd_cwnd > dst_metric(dst, RTAX_CWND))
dst_metric_set(dst, RTAX_CWND, tp->snd_cwnd);
} else if (tp->snd_cwnd > tp->snd_ssthresh &&
icsk->icsk_ca_state == TCP_CA_Open) {
/* Cong. avoidance phase, cwnd is reliable. */
if (!dst_metric_locked(dst, RTAX_SSTHRESH))
dst_metric_set(dst, RTAX_SSTHRESH,
max(tp->snd_cwnd >> 1, tp->snd_ssthresh));
if (!dst_metric_locked(dst, RTAX_CWND))
dst_metric_set(dst, RTAX_CWND,
(dst_metric(dst, RTAX_CWND) +
tp->snd_cwnd) >> 1);
} else {
/* Else slow start did not finish, cwnd is non-sense,
ssthresh may be also invalid.
*/
if (!dst_metric_locked(dst, RTAX_CWND))
dst_metric_set(dst, RTAX_CWND,
(dst_metric(dst, RTAX_CWND) +
tp->snd_ssthresh) >> 1);
if (dst_metric(dst, RTAX_SSTHRESH) &&
!dst_metric_locked(dst, RTAX_SSTHRESH) &&
tp->snd_ssthresh > dst_metric(dst, RTAX_SSTHRESH))
dst_metric_set(dst, RTAX_SSTHRESH, tp->snd_ssthresh);
}
if (!dst_metric_locked(dst, RTAX_REORDERING)) {
if (dst_metric(dst, RTAX_REORDERING) < tp->reordering &&
tp->reordering != sysctl_tcp_reordering)
dst_metric_set(dst, RTAX_REORDERING, tp->reordering);
}
}
}
__u32 tcp_init_cwnd(const struct tcp_sock *tp, const struct dst_entry *dst) __u32 tcp_init_cwnd(const struct tcp_sock *tp, const struct dst_entry *dst)
{ {
__u32 cwnd = (dst ? dst_metric(dst, RTAX_INITCWND) : 0); __u32 cwnd = (dst ? dst_metric(dst, RTAX_INITCWND) : 0);
...@@ -867,7 +763,7 @@ void tcp_enter_cwr(struct sock *sk, const int set_ssthresh) ...@@ -867,7 +763,7 @@ void tcp_enter_cwr(struct sock *sk, const int set_ssthresh)
* Packet counting of FACK is based on in-order assumptions, therefore TCP * Packet counting of FACK is based on in-order assumptions, therefore TCP
* disables it when reordering is detected * disables it when reordering is detected
*/ */
static void tcp_disable_fack(struct tcp_sock *tp) void tcp_disable_fack(struct tcp_sock *tp)
{ {
/* RFC3517 uses different metric in lost marker => reset on change */ /* RFC3517 uses different metric in lost marker => reset on change */
if (tcp_is_fack(tp)) if (tcp_is_fack(tp))
...@@ -881,86 +777,6 @@ static void tcp_dsack_seen(struct tcp_sock *tp) ...@@ -881,86 +777,6 @@ static void tcp_dsack_seen(struct tcp_sock *tp)
tp->rx_opt.sack_ok |= TCP_DSACK_SEEN; tp->rx_opt.sack_ok |= TCP_DSACK_SEEN;
} }
/* Initialize metrics on socket. */
static void tcp_init_metrics(struct sock *sk)
{
struct tcp_sock *tp = tcp_sk(sk);
struct dst_entry *dst = __sk_dst_get(sk);
if (dst == NULL)
goto reset;
dst_confirm(dst);
if (dst_metric_locked(dst, RTAX_CWND))
tp->snd_cwnd_clamp = dst_metric(dst, RTAX_CWND);
if (dst_metric(dst, RTAX_SSTHRESH)) {
tp->snd_ssthresh = dst_metric(dst, RTAX_SSTHRESH);
if (tp->snd_ssthresh > tp->snd_cwnd_clamp)
tp->snd_ssthresh = tp->snd_cwnd_clamp;
} else {
/* ssthresh may have been reduced unnecessarily during.
* 3WHS. Restore it back to its initial default.
*/
tp->snd_ssthresh = TCP_INFINITE_SSTHRESH;
}
if (dst_metric(dst, RTAX_REORDERING) &&
tp->reordering != dst_metric(dst, RTAX_REORDERING)) {
tcp_disable_fack(tp);
tcp_disable_early_retrans(tp);
tp->reordering = dst_metric(dst, RTAX_REORDERING);
}
if (dst_metric(dst, RTAX_RTT) == 0 || tp->srtt == 0)
goto reset;
/* Initial rtt is determined from SYN,SYN-ACK.
* The segment is small and rtt may appear much
* less than real one. Use per-dst memory
* to make it more realistic.
*
* A bit of theory. RTT is time passed after "normal" sized packet
* is sent until it is ACKed. In normal circumstances sending small
* packets force peer to delay ACKs and calculation is correct too.
* The algorithm is adaptive and, provided we follow specs, it
* NEVER underestimate RTT. BUT! If peer tries to make some clever
* tricks sort of "quick acks" for time long enough to decrease RTT
* to low value, and then abruptly stops to do it and starts to delay
* ACKs, wait for troubles.
*/
if (dst_metric_rtt(dst, RTAX_RTT) > tp->srtt) {
tp->srtt = dst_metric_rtt(dst, RTAX_RTT);
tp->rtt_seq = tp->snd_nxt;
}
if (dst_metric_rtt(dst, RTAX_RTTVAR) > tp->mdev) {
tp->mdev = dst_metric_rtt(dst, RTAX_RTTVAR);
tp->mdev_max = tp->rttvar = max(tp->mdev, tcp_rto_min(sk));
}
tcp_set_rto(sk);
reset:
if (tp->srtt == 0) {
/* RFC6298: 5.7 We've failed to get a valid RTT sample from
* 3WHS. This is most likely due to retransmission,
* including spurious one. Reset the RTO back to 3secs
* from the more aggressive 1sec to avoid more spurious
* retransmission.
*/
tp->mdev = tp->mdev_max = tp->rttvar = TCP_TIMEOUT_FALLBACK;
inet_csk(sk)->icsk_rto = TCP_TIMEOUT_FALLBACK;
}
/* Cut cwnd down to 1 per RFC5681 if SYN or SYN-ACK has been
* retransmitted. In light of RFC6298 more aggressive 1sec
* initRTO, we only reset cwnd when more than 1 SYN/SYN-ACK
* retransmission has occurred.
*/
if (tp->total_retrans > 1)
tp->snd_cwnd = 1;
else
tp->snd_cwnd = tcp_init_cwnd(tp, dst);
tp->snd_cwnd_stamp = tcp_time_stamp;
}
static void tcp_update_reordering(struct sock *sk, const int metric, static void tcp_update_reordering(struct sock *sk, const int metric,
const int ts) const int ts)
{ {
......
#include <linux/cache.h>
#include <linux/tcp.h>
#include <net/inet_connection_sock.h>
#include <net/sock.h>
#include <net/dst.h>
#include <net/tcp.h>
int sysctl_tcp_nometrics_save __read_mostly;
/* Save metrics learned by this TCP session. This function is called
* only, when TCP finishes successfully i.e. when it enters TIME-WAIT
* or goes from LAST-ACK to CLOSE.
*/
void tcp_update_metrics(struct sock *sk)
{
struct tcp_sock *tp = tcp_sk(sk);
struct dst_entry *dst = __sk_dst_get(sk);
if (sysctl_tcp_nometrics_save)
return;
if (dst && (dst->flags & DST_HOST)) {
const struct inet_connection_sock *icsk = inet_csk(sk);
int m;
unsigned long rtt;
dst_confirm(dst);
if (icsk->icsk_backoff || !tp->srtt) {
/* This session failed to estimate rtt. Why?
* Probably, no packets returned in time.
* Reset our results.
*/
if (!(dst_metric_locked(dst, RTAX_RTT)))
dst_metric_set(dst, RTAX_RTT, 0);
return;
}
rtt = dst_metric_rtt(dst, RTAX_RTT);
m = rtt - tp->srtt;
/* If newly calculated rtt larger than stored one,
* store new one. Otherwise, use EWMA. Remember,
* rtt overestimation is always better than underestimation.
*/
if (!(dst_metric_locked(dst, RTAX_RTT))) {
if (m <= 0)
set_dst_metric_rtt(dst, RTAX_RTT, tp->srtt);
else
set_dst_metric_rtt(dst, RTAX_RTT, rtt - (m >> 3));
}
if (!(dst_metric_locked(dst, RTAX_RTTVAR))) {
unsigned long var;
if (m < 0)
m = -m;
/* Scale deviation to rttvar fixed point */
m >>= 1;
if (m < tp->mdev)
m = tp->mdev;
var = dst_metric_rtt(dst, RTAX_RTTVAR);
if (m >= var)
var = m;
else
var -= (var - m) >> 2;
set_dst_metric_rtt(dst, RTAX_RTTVAR, var);
}
if (tcp_in_initial_slowstart(tp)) {
/* Slow start still did not finish. */
if (dst_metric(dst, RTAX_SSTHRESH) &&
!dst_metric_locked(dst, RTAX_SSTHRESH) &&
(tp->snd_cwnd >> 1) > dst_metric(dst, RTAX_SSTHRESH))
dst_metric_set(dst, RTAX_SSTHRESH, tp->snd_cwnd >> 1);
if (!dst_metric_locked(dst, RTAX_CWND) &&
tp->snd_cwnd > dst_metric(dst, RTAX_CWND))
dst_metric_set(dst, RTAX_CWND, tp->snd_cwnd);
} else if (tp->snd_cwnd > tp->snd_ssthresh &&
icsk->icsk_ca_state == TCP_CA_Open) {
/* Cong. avoidance phase, cwnd is reliable. */
if (!dst_metric_locked(dst, RTAX_SSTHRESH))
dst_metric_set(dst, RTAX_SSTHRESH,
max(tp->snd_cwnd >> 1, tp->snd_ssthresh));
if (!dst_metric_locked(dst, RTAX_CWND))
dst_metric_set(dst, RTAX_CWND,
(dst_metric(dst, RTAX_CWND) +
tp->snd_cwnd) >> 1);
} else {
/* Else slow start did not finish, cwnd is non-sense,
ssthresh may be also invalid.
*/
if (!dst_metric_locked(dst, RTAX_CWND))
dst_metric_set(dst, RTAX_CWND,
(dst_metric(dst, RTAX_CWND) +
tp->snd_ssthresh) >> 1);
if (dst_metric(dst, RTAX_SSTHRESH) &&
!dst_metric_locked(dst, RTAX_SSTHRESH) &&
tp->snd_ssthresh > dst_metric(dst, RTAX_SSTHRESH))
dst_metric_set(dst, RTAX_SSTHRESH, tp->snd_ssthresh);
}
if (!dst_metric_locked(dst, RTAX_REORDERING)) {
if (dst_metric(dst, RTAX_REORDERING) < tp->reordering &&
tp->reordering != sysctl_tcp_reordering)
dst_metric_set(dst, RTAX_REORDERING, tp->reordering);
}
}
}
/* Initialize metrics on socket. */
void tcp_init_metrics(struct sock *sk)
{
struct tcp_sock *tp = tcp_sk(sk);
struct dst_entry *dst = __sk_dst_get(sk);
if (dst == NULL)
goto reset;
dst_confirm(dst);
if (dst_metric_locked(dst, RTAX_CWND))
tp->snd_cwnd_clamp = dst_metric(dst, RTAX_CWND);
if (dst_metric(dst, RTAX_SSTHRESH)) {
tp->snd_ssthresh = dst_metric(dst, RTAX_SSTHRESH);
if (tp->snd_ssthresh > tp->snd_cwnd_clamp)
tp->snd_ssthresh = tp->snd_cwnd_clamp;
} else {
/* ssthresh may have been reduced unnecessarily during.
* 3WHS. Restore it back to its initial default.
*/
tp->snd_ssthresh = TCP_INFINITE_SSTHRESH;
}
if (dst_metric(dst, RTAX_REORDERING) &&
tp->reordering != dst_metric(dst, RTAX_REORDERING)) {
tcp_disable_fack(tp);
tcp_disable_early_retrans(tp);
tp->reordering = dst_metric(dst, RTAX_REORDERING);
}
if (dst_metric(dst, RTAX_RTT) == 0 || tp->srtt == 0)
goto reset;
/* Initial rtt is determined from SYN,SYN-ACK.
* The segment is small and rtt may appear much
* less than real one. Use per-dst memory
* to make it more realistic.
*
* A bit of theory. RTT is time passed after "normal" sized packet
* is sent until it is ACKed. In normal circumstances sending small
* packets force peer to delay ACKs and calculation is correct too.
* The algorithm is adaptive and, provided we follow specs, it
* NEVER underestimate RTT. BUT! If peer tries to make some clever
* tricks sort of "quick acks" for time long enough to decrease RTT
* to low value, and then abruptly stops to do it and starts to delay
* ACKs, wait for troubles.
*/
if (dst_metric_rtt(dst, RTAX_RTT) > tp->srtt) {
tp->srtt = dst_metric_rtt(dst, RTAX_RTT);
tp->rtt_seq = tp->snd_nxt;
}
if (dst_metric_rtt(dst, RTAX_RTTVAR) > tp->mdev) {
tp->mdev = dst_metric_rtt(dst, RTAX_RTTVAR);
tp->mdev_max = tp->rttvar = max(tp->mdev, tcp_rto_min(sk));
}
tcp_set_rto(sk);
reset:
if (tp->srtt == 0) {
/* RFC6298: 5.7 We've failed to get a valid RTT sample from
* 3WHS. This is most likely due to retransmission,
* including spurious one. Reset the RTO back to 3secs
* from the more aggressive 1sec to avoid more spurious
* retransmission.
*/
tp->mdev = tp->mdev_max = tp->rttvar = TCP_TIMEOUT_FALLBACK;
inet_csk(sk)->icsk_rto = TCP_TIMEOUT_FALLBACK;
}
/* Cut cwnd down to 1 per RFC5681 if SYN or SYN-ACK has been
* retransmitted. In light of RFC6298 more aggressive 1sec
* initRTO, we only reset cwnd when more than 1 SYN/SYN-ACK
* retransmission has occurred.
*/
if (tp->total_retrans > 1)
tp->snd_cwnd = 1;
else
tp->snd_cwnd = tcp_init_cwnd(tp, dst);
tp->snd_cwnd_stamp = tcp_time_stamp;
}
Markdown is supported
0%
or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment