tcp: set few options locklessly
Subsystems (3)
cover tcp: set few options locklessly
This series is avoiding the socket lock for six TCP options.
They are not heavily used, but this exercise can give ideas for other parts of TCP/IP stack :)
Eric Dumazet (6):
tcp: set TCP_SYNCNT locklessly
tcp: set TCP_USER_TIMEOUT locklessly
tcp: set TCP_KEEPINTVL locklessly
tcp: set TCP_KEEPCNT locklessly
tcp: set TCP_LINGER2 locklessly
tcp: set TCP_DEFER_ACCEPT locklessly
include/linux/tcp.h | 2 +- net/ipv4/tcp.c | 90 ++++++++++++++++------------------------ net/ipv4/tcp_input.c | 4 +- net/ipv4/tcp_minisocks.c | 2 +- net/ipv4/tcp_timer.c | 48 ++++++++++++--------- 5 files changed, 68 insertions(+), 78 deletions(-)
Signature · 1 line
2.41.0.640.ga95def55d0-goog
PATCH 1/6 tcp: set TCP_SYNCNT locklessly
icsk->icsk_syn_retries can safely be set without locking the socket.
We have to add READ_ONCE() annotations in tcp_fastopen_synack_timer() and tcp_write_timeout().
--- net/ipv4/tcp.c | 15 ++++++--------- net/ipv4/tcp_timer.c | 9 ++++++--- 2 files changed, 12 insertions(+), 12 deletions(-)
net/ipv4/tcp.c +6 -9
@@ -3291,9 +3291,7 @@ int tcp_sock_set_syncnt(struct sock *sk, int val)if (val <1 || val >MAX_TCP_SYNCNT )return -EINVAL ; -lock_sock (sk);WRITE_ONCE (inet_csk (sk)->icsk_syn_retries, val); -release_sock (sk);return 0 ; }EXPORT_SYMBOL (tcp_sock_set_syncnt);
@@ -3462,6 +3460,12 @@ int do_tcp_setsockopt(struct sock *sk, int level, int optname,if (copy_from_sockptr (&val, optval,sizeof (val)))return -EFAULT ; +/* Handle options that can be set without locking the socket. */ +switch (optname) { +case TCP_SYNCNT : +return tcp_sock_set_syncnt (sk, val); + } +sockopt_lock_sock (sk);switch (optname) {
@@ -3569,13 +3573,6 @@ int do_tcp_setsockopt(struct sock *sk, int level, int optname,else WRITE_ONCE (tp->keepalive_probes, val);break ; -case TCP_SYNCNT : -if (val <1 || val >MAX_TCP_SYNCNT ) - err = -EINVAL ; -else -WRITE_ONCE (icsk->icsk_syn_retries, val); -break ; -case TCP_SAVE_SYN :/* 0: disable, 1: enable, 2: start from ether_header */ if (val <0 || val >2 )
net/ipv4/tcp_timer.c +6 -3
@@ -239,7 +239,8 @@ static int tcp_write_timeout(struct sock *sk)if ((1 << sk->sk_state) & (TCPF_SYN_SENT |TCPF_SYN_RECV )) {if (icsk->icsk_retransmits)__dst_negative_advice (sk); - retry_until = icsk->icsk_syn_retries ? : +/* Paired with WRITE_ONCE() in tcp_sock_set_syncnt() */ + retry_until =READ_ONCE (icsk->icsk_syn_retries) ? :READ_ONCE (net->ipv4.sysctl_tcp_syn_retries); max_retransmits = retry_until;
@@ -421,8 +422,10 @@ static void tcp_fastopen_synack_timer(struct sock *sk, struct request_sock *req) req->rsk_ops->syn_ack_timeout (req); -/* add one more retry for fastopen */ - max_retries = icsk->icsk_syn_retries ? : +/* Add one more retry for fastopen. +* Paired with WRITE_ONCE() in tcp_sock_set_syncnt() +*/ + max_retries =READ_ONCE (icsk->icsk_syn_retries) ? :READ_ONCE (sock_net (sk)->ipv4.sysctl_tcp_synack_retries) +1 ;if (req->num_timeout >= max_retries) {
Signature · 1 line
2.41.0.640.ga95def55d0-goog
Re: [PATCH net-next 1/6] tcp: set TCP_SYNCNT locklessly
Quoted 7 lines · Eric Dumazet
icsk->icsk_syn_retriescan safely be set without locking the socket.We have to add
READ_ONCE()annotations intcp_fastopen_synack_timer()andtcp_write_timeout().Signed-off-by Eric Dumazet <[email protected]>
Acked by: Soheil Hassas Yeganeh <[email protected]>
Quoted 76 lines
--- net/ipv4/tcp.c | 15 ++++++--------- net/ipv4/tcp_timer.c | 9 ++++++--- 2 files changed, 12 insertions(+), 12 deletions(-)net/ipv4/tcp.c +6
-9@@ -3291,9 +3291,7 @@ int tcp_sock_set_syncnt(struct sock *sk, int val)if (val <1 || val >MAX_TCP_SYNCNT )return -EINVAL ; -lock_sock (sk);WRITE_ONCE (inet_csk (sk)->icsk_syn_retries, val); -release_sock (sk);return 0 ; }EXPORT_SYMBOL (tcp_sock_set_syncnt);@@ -3462,6 +3460,12 @@ int do_tcp_setsockopt(struct sock *sk, int level, int optname,if (copy_from_sockptr (&val, optval,sizeof (val)))return -EFAULT ; +/* Handle options that can be set without locking the socket. */ +switch (optname) { +case TCP_SYNCNT : +return tcp_sock_set_syncnt (sk, val); + } +sockopt_lock_sock (sk);switch (optname) {@@ -3569,13 +3573,6 @@ int do_tcp_setsockopt(struct sock *sk, int level, int optname,else WRITE_ONCE (tp->keepalive_probes, val);break ; -case TCP_SYNCNT : -if (val <1 || val >MAX_TCP_SYNCNT ) - err = -EINVAL ; -else -WRITE_ONCE (icsk->icsk_syn_retries, val); -break ; -case TCP_SAVE_SYN :/* 0: disable, 1: enable, 2: start from ether_header */ if (val <0 || val >2 )net/ipv4/tcp_timer.c +6
-4@@ -239,7 +239,8 @@ static int tcp_write_timeout(struct sock *sk)if ((1 << sk->sk_state) & (TCPF_SYN_SENT |TCPF_SYN_RECV )) {if (icsk->icsk_retransmits)__dst_negative_advice (sk); - retry_until = icsk->icsk_syn_retries ? : +/* Paired with WRITE_ONCE() in tcp_sock_set_syncnt() */ + retry_until =READ_ONCE (icsk->icsk_syn_retries) ? :READ_ONCE (net->ipv4.sysctl_tcp_syn_retries); max_retransmits = retry_until;@@ -421,8 +422,10 @@ static void tcp_fastopen_synack_timer(struct sock *sk, struct request_sock *req) req->rsk_ops->syn_ack_timeout (req); -/* add one more retry for fastopen */ - max_retries = icsk->icsk_syn_retries ? : +/* Add one more retry for fastopen. +* Paired with WRITE_ONCE() in tcp_sock_set_syncnt() +*/ + max_retries =READ_ONCE (icsk->icsk_syn_retries) ? :READ_ONCE (sock_net (sk)->ipv4.sysctl_tcp_synack_retries) +1 ;if (req->num_timeout >= max_retries) { --2.41.0.640.ga95def55d0-goog
PATCH 2/6 tcp: set TCP_USER_TIMEOUT locklessly
In reply to Eric Dumazeticsk->icsk_user_timeout can be set locklessly,
if all read sides use READ_ONCE().
--- include/linux/tcp.h | 2 +- net/ipv4/tcp.c | 23 ++++++++++------------- net/ipv4/tcp_timer.c | 37 ++++++++++++++++++++++--------------- 3 files changed, 33 insertions(+), 29 deletions(-)
include/linux/tcp.h +1 -1
@@ -564,6 +564,6 @@ void __tcp_sock_set_nodelay(struct sock *sk, bool on);void tcp_sock_set_nodelay (struct sock *sk);void tcp_sock_set_quickack (struct sock *sk,int val);int tcp_sock_set_syncnt (struct sock *sk,int val); -void tcp_sock_set_user_timeout (struct sock *sk,u32 val); +int tcp_sock_set_user_timeout (struct sock *sk,int val);#endif /* _LINUX_TCP_H */
net/ipv4/tcp.c +10 -13
@@ -3296,11 +3296,16 @@ int tcp_sock_set_syncnt(struct sock *sk, int val) }EXPORT_SYMBOL (tcp_sock_set_syncnt); -void tcp_sock_set_user_timeout (struct sock *sk,u32 val) +int tcp_sock_set_user_timeout (struct sock *sk,int val) { -lock_sock (sk); +/* Cap the max time in ms TCP will retry or probe the window +* before giving up and aborting (ETIMEDOUT) a connection. +*/ +if (val <0 ) +return -EINVAL ; +WRITE_ONCE (inet_csk (sk)->icsk_user_timeout, val); -release_sock (sk); +return 0 ; }EXPORT_SYMBOL (tcp_sock_set_user_timeout);
@@ -3464,6 +3469,8 @@ int do_tcp_setsockopt(struct sock *sk, int level, int optname,switch (optname) {case TCP_SYNCNT :return tcp_sock_set_syncnt (sk, val); +case TCP_USER_TIMEOUT : +return tcp_sock_set_user_timeout (sk, val); }sockopt_lock_sock (sk);
@@ -3611,16 +3618,6 @@ int do_tcp_setsockopt(struct sock *sk, int level, int optname, err = tp->af_specific->md5_parse (sk, optname, optval, optlen);break ;#endif -case TCP_USER_TIMEOUT : -/* Cap the max time in ms TCP will retry or probe the window -* before giving up and aborting (ETIMEDOUT) a connection. -*/ -if (val <0 ) - err = -EINVAL ; -else -WRITE_ONCE (icsk->icsk_user_timeout, val); -break ; -case TCP_FASTOPEN :if (val >=0 && ((1 << sk->sk_state) & (TCPF_CLOSE |TCPF_LISTEN ))) {
net/ipv4/tcp_timer.c +22 -15
@@ -26,14 +26,15 @@static u32 tcp_clamp_rto_to_user_timeout (const struct sock *sk) {struct inet_connection_sock *icsk =inet_csk (sk); -u32 elapsed, start_ts; +u32 elapsed, start_ts, user_timeout;s32 remaining; start_ts =tcp_sk (sk)->retrans_stamp; -if (!icsk->icsk_user_timeout) + user_timeout =READ_ONCE (icsk->icsk_user_timeout); +if (!user_timeout)return icsk->icsk_rto; elapsed =tcp_time_stamp (tcp_sk (sk)) - start_ts; - remaining = icsk->icsk_user_timeout - elapsed; + remaining = user_timeout - elapsed;if (remaining <=0 )return 1 ;/* user timeout has passed; fire ASAP */
@@ -43,16 +44,17 @@ static u32 tcp_clamp_rto_to_user_timeout(const struct sock *sk)u32 tcp_clamp_probe0_to_user_timeout (const struct sock *sk,u32 when) {struct inet_connection_sock *icsk =inet_csk (sk); -u32 remaining; +u32 remaining, user_timeout;s32 elapsed; -if (!icsk->icsk_user_timeout || !icsk->icsk_probes_tstamp) + user_timeout =READ_ONCE (icsk->icsk_user_timeout); +if (!user_timeout || !icsk->icsk_probes_tstamp)return when; elapsed = tcp_jiffies32 - icsk->icsk_probes_tstamp;if (unlikely (elapsed <0 )) elapsed =0 ; - remaining =msecs_to_jiffies (icsk->icsk_user_timeout) - elapsed; + remaining =msecs_to_jiffies (user_timeout) - elapsed; remaining =max_t (u32, remaining,TCP_TIMEOUT_MIN );return min_t (u32, remaining, when);
@@ -270,7 +272,7 @@ static int tcp_write_timeout(struct sock *sk) }if (!expired) expired =retransmits_timed_out (sk, retry_until, - icsk->icsk_user_timeout); +READ_ONCE (icsk->icsk_user_timeout));tcp_fastopen_active_detect_blackhole (sk, expired);if (BPF_SOCK_OPS_TEST_FLAG (tp,BPF_SOCK_OPS_RTO_CB_FLAG ))
@@ -384,13 +386,16 @@ static void tcp_probe_timer(struct sock *sk)* corresponding system limit. We also implement similar policy when * we use RTO to probe window in tcp_retransmit_timer(). */ -if (!icsk->icsk_probes_tstamp) +if (!icsk->icsk_probes_tstamp) { icsk->icsk_probes_tstamp = tcp_jiffies32; -else if (icsk->icsk_user_timeout && - (s32)(tcp_jiffies32 - icsk->icsk_probes_tstamp) >= -msecs_to_jiffies (icsk->icsk_user_timeout)) - goto abort; + }else { +u32 user_timeout =READ_ONCE (icsk->icsk_user_timeout); +if (user_timeout && + (s32)(tcp_jiffies32 - icsk->icsk_probes_tstamp) >= +msecs_to_jiffies (user_timeout)) + goto abort; + } max_probes =READ_ONCE (sock_net (sk)->ipv4.sysctl_tcp_retries2);if (sock_flag (sk,SOCK_DEAD )) {const bool alive =inet_csk_rto_backoff (icsk,TCP_RTO_MAX ) <TCP_RTO_MAX ;
@@ -734,13 +739,15 @@ static void tcp_keepalive_timer (struct timer_list *t) elapsed =keepalive_time_elapsed (tp);if (elapsed >=keepalive_time_when (tp)) { +u32 user_timeout =READ_ONCE (icsk->icsk_user_timeout); +/* If the TCP_USER_TIMEOUT option is enabled, use that * to determine when to timeout instead. */ -if ((icsk->icsk_user_timeout !=0 && - elapsed >=msecs_to_jiffies (icsk->icsk_user_timeout) && +if ((user_timeout !=0 && + elapsed >=msecs_to_jiffies (user_timeout) && icsk->icsk_probes_out >0 ) || - (icsk->icsk_user_timeout ==0 && + (user_timeout ==0 && icsk->icsk_probes_out >=keepalive_probes (tp))) {tcp_send_active_reset (sk,GFP_ATOMIC );tcp_write_err (sk);
Signature · 1 line
2.41.0.640.ga95def55d0-goog
Re: [PATCH net-next 2/6] tcp: set TCP_USER_TIMEOUT locklessly
Quoted 5 lines · Eric Dumazet
icsk->icsk_user_timeoutcan be set locklessly,
if all read sides useREAD_ONCE().Signed-off-by Eric Dumazet <[email protected]>
Quoted 165 lines
--- include/linux/tcp.h | 2 +- net/ipv4/tcp.c | 23 ++++++++++------------- net/ipv4/tcp_timer.c | 37 ++++++++++++++++++++++--------------- 3 files changed, 33 insertions(+), 29 deletions(-)include/linux/tcp.h +1
-1@@ -564,6 +564,6 @@ void __tcp_sock_set_nodelay(struct sock *sk, bool on);void tcp_sock_set_nodelay (struct sock *sk);void tcp_sock_set_quickack (struct sock *sk,int val);int tcp_sock_set_syncnt (struct sock *sk,int val); -void tcp_sock_set_user_timeout (struct sock *sk,u32 val); +int tcp_sock_set_user_timeout (struct sock *sk,int val);#endif /* _LINUX_TCP_H */ net/ipv4/tcp.c +10
-13@@ -3296,11 +3296,16 @@ int tcp_sock_set_syncnt(struct sock *sk, int val) }EXPORT_SYMBOL (tcp_sock_set_syncnt); -void tcp_sock_set_user_timeout (struct sock *sk,u32 val) +int tcp_sock_set_user_timeout (struct sock *sk,int val) { -lock_sock (sk); +/* Cap the max time in ms TCP will retry or probe the window +* before giving up and aborting (ETIMEDOUT) a connection. +*/ +if (val <0 ) +return -EINVAL ; +WRITE_ONCE (inet_csk (sk)->icsk_user_timeout, val); -release_sock (sk); +return 0 ; }EXPORT_SYMBOL (tcp_sock_set_user_timeout);@@ -3464,6 +3469,8 @@ int do_tcp_setsockopt(struct sock *sk, int level, int optname,switch (optname) {case TCP_SYNCNT :return tcp_sock_set_syncnt (sk, val); +case TCP_USER_TIMEOUT : +return tcp_sock_set_user_timeout (sk, val); }sockopt_lock_sock (sk);@@ -3611,16 +3618,6 @@ int do_tcp_setsockopt(struct sock *sk, int level, int optname, err = tp->af_specific->md5_parse (sk, optname, optval, optlen);break ;#endif -case TCP_USER_TIMEOUT : -/* Cap the max time in ms TCP will retry or probe the window -* before giving up and aborting (ETIMEDOUT) a connection. -*/ -if (val <0 ) - err = -EINVAL ; -else -WRITE_ONCE (icsk->icsk_user_timeout, val); -break ; -case TCP_FASTOPEN :if (val >=0 && ((1 << sk->sk_state) & (TCPF_CLOSE |TCPF_LISTEN ))) {net/ipv4/tcp_timer.c +22
-16@@ -26,14 +26,15 @@static u32 tcp_clamp_rto_to_user_timeout (const struct sock *sk) {struct inet_connection_sock *icsk =inet_csk (sk); -u32 elapsed, start_ts; +u32 elapsed, start_ts, user_timeout;s32 remaining; start_ts =tcp_sk (sk)->retrans_stamp; -if (!icsk->icsk_user_timeout) + user_timeout =READ_ONCE (icsk->icsk_user_timeout); +if (!user_timeout)return icsk->icsk_rto; elapsed =tcp_time_stamp (tcp_sk (sk)) - start_ts; - remaining = icsk->icsk_user_timeout - elapsed; + remaining = user_timeout - elapsed;if (remaining <=0 )return 1 ;/* user timeout has passed; fire ASAP */ @@ -43,16 +44,17 @@ static u32 tcp_clamp_rto_to_user_timeout(const struct sock *sk)u32 tcp_clamp_probe0_to_user_timeout (const struct sock *sk,u32 when) {struct inet_connection_sock *icsk =inet_csk (sk); -u32 remaining; +u32 remaining, user_timeout;s32 elapsed; -if (!icsk->icsk_user_timeout || !icsk->icsk_probes_tstamp) + user_timeout =READ_ONCE (icsk->icsk_user_timeout); +if (!user_timeout || !icsk->icsk_probes_tstamp)return when; elapsed = tcp_jiffies32 - icsk->icsk_probes_tstamp;if (unlikely (elapsed <0 )) elapsed =0 ; - remaining =msecs_to_jiffies (icsk->icsk_user_timeout) - elapsed; + remaining =msecs_to_jiffies (user_timeout) - elapsed; remaining =max_t (u32, remaining,TCP_TIMEOUT_MIN );return min_t (u32, remaining, when);@@ -270,7 +272,7 @@ static int tcp_write_timeout(struct sock *sk) }if (!expired) expired =retransmits_timed_out (sk, retry_until, - icsk->icsk_user_timeout); +READ_ONCE (icsk->icsk_user_timeout));tcp_fastopen_active_detect_blackhole (sk, expired);if (BPF_SOCK_OPS_TEST_FLAG (tp,BPF_SOCK_OPS_RTO_CB_FLAG ))@@ -384,13 +386,16 @@ static void tcp_probe_timer(struct sock *sk)* corresponding system limit. We also implement similar policy when * we use RTO to probe window in tcp_retransmit_timer(). */ -if (!icsk->icsk_probes_tstamp) +if (!icsk->icsk_probes_tstamp) { icsk->icsk_probes_tstamp = tcp_jiffies32; -else if (icsk->icsk_user_timeout && - (s32)(tcp_jiffies32 - icsk->icsk_probes_tstamp) >= -msecs_to_jiffies (icsk->icsk_user_timeout)) - goto abort; + }else { +u32 user_timeout =READ_ONCE (icsk->icsk_user_timeout); +if (user_timeout && + (s32)(tcp_jiffies32 - icsk->icsk_probes_tstamp) >= +msecs_to_jiffies (user_timeout)) + goto abort; + } max_probes =READ_ONCE (sock_net (sk)->ipv4.sysctl_tcp_retries2);if (sock_flag (sk,SOCK_DEAD )) {const bool alive =inet_csk_rto_backoff (icsk,TCP_RTO_MAX ) <TCP_RTO_MAX ;@@ -734,13 +739,15 @@ static void tcp_keepalive_timer (struct timer_list *t) elapsed =keepalive_time_elapsed (tp);if (elapsed >=keepalive_time_when (tp)) { +u32 user_timeout =READ_ONCE (icsk->icsk_user_timeout); +/* If the TCP_USER_TIMEOUT option is enabled, use that * to determine when to timeout instead. */ -if ((icsk->icsk_user_timeout !=0 && - elapsed >=msecs_to_jiffies (icsk->icsk_user_timeout) && +if ((user_timeout !=0 && + elapsed >=msecs_to_jiffies (user_timeout) && icsk->icsk_probes_out >0 ) || - (icsk->icsk_user_timeout ==0 && + (user_timeout ==0 && icsk->icsk_probes_out >=keepalive_probes (tp))) {tcp_send_active_reset (sk,GFP_ATOMIC );tcp_write_err (sk); --2.41.0.640.ga95def55d0-goog
PATCH 3/6 tcp: set TCP_KEEPINTVL locklessly
In reply to Eric Dumazettp->keepalive_intvl can be set locklessly, readers are already taking care of this field being potentially set by other threads.
--- net/ipv4/tcp.c | 10 ++-------- 1 file changed, 2 insertions(+), 8 deletions(-)
net/ipv4/tcp.c +2 -8
@@ -3348,9 +3348,7 @@ int tcp_sock_set_keepintvl(struct sock *sk, int val)if (val <1 || val >MAX_TCP_KEEPINTVL )return -EINVAL ; -lock_sock (sk);WRITE_ONCE (tcp_sk (sk)->keepalive_intvl, val *HZ ); -release_sock (sk);return 0 ; }EXPORT_SYMBOL (tcp_sock_set_keepintvl);
@@ -3471,6 +3469,8 @@ int do_tcp_setsockopt(struct sock *sk, int level, int optname,return tcp_sock_set_syncnt (sk, val);case TCP_USER_TIMEOUT :return tcp_sock_set_user_timeout (sk, val); +case TCP_KEEPINTVL : +return tcp_sock_set_keepintvl (sk, val); }sockopt_lock_sock (sk);
@@ -3568,12 +3568,6 @@ int do_tcp_setsockopt(struct sock *sk, int level, int optname,case TCP_KEEPIDLE : err =tcp_sock_set_keepidle_locked (sk, val);break ; -case TCP_KEEPINTVL : -if (val <1 || val >MAX_TCP_KEEPINTVL ) - err = -EINVAL ; -else -WRITE_ONCE (tp->keepalive_intvl, val *HZ ); -break ;case TCP_KEEPCNT :if (val <1 || val >MAX_TCP_KEEPCNT ) err = -EINVAL ;
Signature · 1 line
2.41.0.640.ga95def55d0-goog
Re: [PATCH net-next 3/6] tcp: set TCP_KEEPINTVL locklessly
Quoted 6 lines · Eric Dumazet
tp->keepalive_intvlcan be set locklessly, readers are already taking care of this field being potentially set by other threads.Signed-off-by Eric Dumazet <[email protected]>
Quoted 43 lines
--- net/ipv4/tcp.c | 10 ++-------- 1 file changed, 2 insertions(+), 8 deletions(-)net/ipv4/tcp.c +2
-9@@ -3348,9 +3348,7 @@ int tcp_sock_set_keepintvl(struct sock *sk, int val)if (val <1 || val >MAX_TCP_KEEPINTVL )return -EINVAL ; -lock_sock (sk);WRITE_ONCE (tcp_sk (sk)->keepalive_intvl, val *HZ ); -release_sock (sk);return 0 ; }EXPORT_SYMBOL (tcp_sock_set_keepintvl);@@ -3471,6 +3469,8 @@ int do_tcp_setsockopt(struct sock *sk, int level, int optname,return tcp_sock_set_syncnt (sk, val);case TCP_USER_TIMEOUT :return tcp_sock_set_user_timeout (sk, val); +case TCP_KEEPINTVL : +return tcp_sock_set_keepintvl (sk, val); }sockopt_lock_sock (sk);@@ -3568,12 +3568,6 @@ int do_tcp_setsockopt(struct sock *sk, int level, int optname,case TCP_KEEPIDLE : err =tcp_sock_set_keepidle_locked (sk, val);break ; -case TCP_KEEPINTVL : -if (val <1 || val >MAX_TCP_KEEPINTVL ) - err = -EINVAL ; -else -WRITE_ONCE (tp->keepalive_intvl, val *HZ ); -break ;case TCP_KEEPCNT :if (val <1 || val >MAX_TCP_KEEPCNT ) err = -EINVAL ; --2.41.0.640.ga95def55d0-goog
PATCH 4/6 tcp: set TCP_KEEPCNT locklessly
In reply to Eric Dumazettp->keepalive_probes can be set locklessly, readers are already taking care of this field being potentially set by other threads.
--- net/ipv4/tcp.c | 10 ++-------- 1 file changed, 2 insertions(+), 8 deletions(-)
net/ipv4/tcp.c +2 -8
@@ -3358,10 +3358,8 @@ int tcp_sock_set_keepcnt(struct sock *sk, int val)if (val <1 || val >MAX_TCP_KEEPCNT )return -EINVAL ; -lock_sock (sk);/* Paired with READ_ONCE() in keepalive_probes() */ WRITE_ONCE (tcp_sk (sk)->keepalive_probes, val); -release_sock (sk);return 0 ; }EXPORT_SYMBOL (tcp_sock_set_keepcnt);
@@ -3471,6 +3469,8 @@ int do_tcp_setsockopt(struct sock *sk, int level, int optname,return tcp_sock_set_user_timeout (sk, val);case TCP_KEEPINTVL :return tcp_sock_set_keepintvl (sk, val); +case TCP_KEEPCNT : +return tcp_sock_set_keepcnt (sk, val); }sockopt_lock_sock (sk);
@@ -3568,12 +3568,6 @@ int do_tcp_setsockopt(struct sock *sk, int level, int optname,case TCP_KEEPIDLE : err =tcp_sock_set_keepidle_locked (sk, val);break ; -case TCP_KEEPCNT : -if (val <1 || val >MAX_TCP_KEEPCNT ) - err = -EINVAL ; -else -WRITE_ONCE (tp->keepalive_probes, val); -break ;case TCP_SAVE_SYN :/* 0: disable, 1: enable, 2: start from ether_header */ if (val <0 || val >2 )
Signature · 1 line
2.41.0.640.ga95def55d0-goog
Re: [PATCH net-next 4/6] tcp: set TCP_KEEPCNT locklessly
Quoted 6 lines · Eric Dumazet
tp->keepalive_probescan be set locklessly, readers are already taking care of this field being potentially set by other threads.Signed-off-by Eric Dumazet <[email protected]>
Quoted 44 lines
--- net/ipv4/tcp.c | 10 ++-------- 1 file changed, 2 insertions(+), 8 deletions(-)net/ipv4/tcp.c +2
-9@@ -3358,10 +3358,8 @@ int tcp_sock_set_keepcnt(struct sock *sk, int val)if (val <1 || val >MAX_TCP_KEEPCNT )return -EINVAL ; -lock_sock (sk);/* Paired with READ_ONCE() in keepalive_probes() */ WRITE_ONCE (tcp_sk (sk)->keepalive_probes, val); -release_sock (sk);return 0 ; }EXPORT_SYMBOL (tcp_sock_set_keepcnt);@@ -3471,6 +3469,8 @@ int do_tcp_setsockopt(struct sock *sk, int level, int optname,return tcp_sock_set_user_timeout (sk, val);case TCP_KEEPINTVL :return tcp_sock_set_keepintvl (sk, val); +case TCP_KEEPCNT : +return tcp_sock_set_keepcnt (sk, val); }sockopt_lock_sock (sk);@@ -3568,12 +3568,6 @@ int do_tcp_setsockopt(struct sock *sk, int level, int optname,case TCP_KEEPIDLE : err =tcp_sock_set_keepidle_locked (sk, val);break ; -case TCP_KEEPCNT : -if (val <1 || val >MAX_TCP_KEEPCNT ) - err = -EINVAL ; -else -WRITE_ONCE (tp->keepalive_probes, val); -break ;case TCP_SAVE_SYN :/* 0: disable, 1: enable, 2: start from ether_header */ if (val <0 || val >2 ) --2.41.0.640.ga95def55d0-goog
PATCH 5/6 tcp: set TCP_LINGER2 locklessly
In reply to Eric Dumazettp->linger2 can be set locklessly as long as readers use READ_ONCE().
--- net/ipv4/tcp.c | 19 +++++++++---------- net/ipv4/tcp_input.c | 2 +- net/ipv4/tcp_timer.c | 2 +- 3 files changed, 11 insertions(+), 12 deletions(-)
net/ipv4/tcp.c +9 -10
@@ -2865,7 +2865,7 @@ void __tcp_close(struct sock *sk, long timeout)if (sk->sk_state ==TCP_FIN_WAIT2 ) {struct tcp_sock *tp =tcp_sk (sk); -if (tp->linger2 <0 ) { +if (READ_ONCE (tp->linger2) <0 ) {tcp_set_state (sk,TCP_CLOSE );tcp_send_active_reset (sk,GFP_ATOMIC ); __NET_INC_STATS(sock_net(sk),
@@ -3471,6 +3471,14 @@ int do_tcp_setsockopt(struct sock *sk, int level, int optname,return tcp_sock_set_keepintvl (sk, val);case TCP_KEEPCNT :return tcp_sock_set_keepcnt (sk, val); +case TCP_LINGER2 : +if (val <0 ) +WRITE_ONCE (tp->linger2,-1 ); +else if (val >TCP_FIN_TIMEOUT_MAX /HZ ) +WRITE_ONCE (tp->linger2,TCP_FIN_TIMEOUT_MAX ); +else +WRITE_ONCE (tp->linger2, val *HZ ); +return 0 ; }sockopt_lock_sock (sk);
@@ -3576,15 +3584,6 @@ int do_tcp_setsockopt(struct sock *sk, int level, int optname, tp->save_syn = val;break ; -case TCP_LINGER2 : -if (val <0 ) -WRITE_ONCE (tp->linger2,-1 ); -else if (val >TCP_FIN_TIMEOUT_MAX /HZ ) -WRITE_ONCE (tp->linger2,TCP_FIN_TIMEOUT_MAX ); -else -WRITE_ONCE (tp->linger2, val *HZ ); -break ; -case TCP_DEFER_ACCEPT :/* Translate value in seconds to number of retransmits */ WRITE_ONCE(icsk->icsk_accept_queue.rskq_defer_accept,
net/ipv4/tcp_input.c +1 -1
@@ -6625,7 +6625,7 @@ int tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb)break ; } -if (tp->linger2 <0 ) { +if (READ_ONCE (tp->linger2) <0 ) {tcp_done (sk);NET_INC_STATS (sock_net (sk),LINUX_MIB_TCPABORTONDATA );return 1 ;
net/ipv4/tcp_timer.c +1 -1
@@ -714,7 +714,7 @@ static void tcp_keepalive_timer (struct timer_list *t)tcp_mstamp_refresh (tp);if (sk->sk_state ==TCP_FIN_WAIT2 &&sock_flag (sk,SOCK_DEAD )) { -if (tp->linger2 >=0 ) { +if (READ_ONCE (tp->linger2) >=0 ) {const int tmo =tcp_fin_time (sk) -TCP_TIMEWAIT_LEN ;if (tmo >0 ) {
Signature · 1 line
2.41.0.640.ga95def55d0-goog
Re: [PATCH net-next 5/6] tcp: set TCP_LINGER2 locklessly
Quoted 5 lines · Eric Dumazet
tp->linger2can be set locklessly as long as readers useREAD_ONCE().Signed-off-by Eric Dumazet <[email protected]>
Quoted 79 lines
--- net/ipv4/tcp.c | 19 +++++++++---------- net/ipv4/tcp_input.c | 2 +- net/ipv4/tcp_timer.c | 2 +- 3 files changed, 11 insertions(+), 12 deletions(-)net/ipv4/tcp.c +9
-10@@ -2865,7 +2865,7 @@ void __tcp_close(struct sock *sk, long timeout)if (sk->sk_state ==TCP_FIN_WAIT2 ) {struct tcp_sock *tp =tcp_sk (sk); -if (tp->linger2 <0 ) { +if (READ_ONCE (tp->linger2) <0 ) {tcp_set_state (sk,TCP_CLOSE );tcp_send_active_reset (sk,GFP_ATOMIC ); __NET_INC_STATS(sock_net(sk),@@ -3471,6 +3471,14 @@ int do_tcp_setsockopt(struct sock *sk, int level, int optname,return tcp_sock_set_keepintvl (sk, val);case TCP_KEEPCNT :return tcp_sock_set_keepcnt (sk, val); +case TCP_LINGER2 : +if (val <0 ) +WRITE_ONCE (tp->linger2,-1 ); +else if (val >TCP_FIN_TIMEOUT_MAX /HZ ) +WRITE_ONCE (tp->linger2,TCP_FIN_TIMEOUT_MAX ); +else +WRITE_ONCE (tp->linger2, val *HZ ); +return 0 ; }sockopt_lock_sock (sk);@@ -3576,15 +3584,6 @@ int do_tcp_setsockopt(struct sock *sk, int level, int optname, tp->save_syn = val;break ; -case TCP_LINGER2 : -if (val <0 ) -WRITE_ONCE (tp->linger2,-1 ); -else if (val >TCP_FIN_TIMEOUT_MAX /HZ ) -WRITE_ONCE (tp->linger2,TCP_FIN_TIMEOUT_MAX ); -else -WRITE_ONCE (tp->linger2, val *HZ ); -break ; -case TCP_DEFER_ACCEPT :/* Translate value in seconds to number of retransmits */ WRITE_ONCE(icsk->icsk_accept_queue.rskq_defer_accept,net/ipv4/tcp_input.c +1
-1@@ -6625,7 +6625,7 @@ int tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb)break ; } -if (tp->linger2 <0 ) { +if (READ_ONCE (tp->linger2) <0 ) {tcp_done (sk);NET_INC_STATS (sock_net (sk),LINUX_MIB_TCPABORTONDATA );return 1 ;net/ipv4/tcp_timer.c +1
-2@@ -714,7 +714,7 @@ static void tcp_keepalive_timer (struct timer_list *t)tcp_mstamp_refresh (tp);if (sk->sk_state ==TCP_FIN_WAIT2 &&sock_flag (sk,SOCK_DEAD )) { -if (tp->linger2 >=0 ) { +if (READ_ONCE (tp->linger2) >=0 ) {const int tmo =tcp_fin_time (sk) -TCP_TIMEWAIT_LEN ;if (tmo >0 ) { --2.41.0.640.ga95def55d0-goog
PATCH 6/6 tcp: set TCP_DEFER_ACCEPT locklessly
In reply to Eric Dumazetrskq_defer_accept field can be read/written without the need of holding the socket lock.
--- net/ipv4/tcp.c | 13 ++++++------- net/ipv4/tcp_input.c | 2 +- net/ipv4/tcp_minisocks.c | 2 +- 3 files changed, 8 insertions(+), 9 deletions(-)
net/ipv4/tcp.c +6 -7
@@ -3479,6 +3479,12 @@ int do_tcp_setsockopt(struct sock *sk, int level, int optname,else WRITE_ONCE (tp->linger2, val *HZ );return 0 ; +case TCP_DEFER_ACCEPT : +/* Translate value in seconds to number of retransmits */ +WRITE_ONCE (icsk->icsk_accept_queue.rskq_defer_accept, +secs_to_retrans (val,TCP_TIMEOUT_INIT /HZ , +TCP_RTO_MAX /HZ )); +return 0 ; }sockopt_lock_sock (sk);
@@ -3584,13 +3590,6 @@ int do_tcp_setsockopt(struct sock *sk, int level, int optname, tp->save_syn = val;break ; -case TCP_DEFER_ACCEPT : -/* Translate value in seconds to number of retransmits */ -WRITE_ONCE (icsk->icsk_accept_queue.rskq_defer_accept, -secs_to_retrans (val,TCP_TIMEOUT_INIT /HZ , -TCP_RTO_MAX /HZ )); -break ; -case TCP_WINDOW_CLAMP : err =tcp_set_window_clamp (sk, val);break ;
net/ipv4/tcp_input.c +1 -1
@@ -6325,7 +6325,7 @@ static int tcp_rcv_synsent_state_process(struct sock *sk, struct sk_buff *skb,if (fastopen_fail)return -1 ;if (sk->sk_write_pending || - icsk->icsk_accept_queue.rskq_defer_accept || +READ_ONCE (icsk->icsk_accept_queue.rskq_defer_accept) ||inet_csk_in_pingpong_mode (sk)) {/* Save one ACK. Data will be ready after * several ticks, if write_pending is set.
net/ipv4/tcp_minisocks.c +1 -1
@@ -794,7 +794,7 @@ struct sock *tcp_check_req(struct sock *sk, struct sk_buff *skb,return sk;/* While TCP_DEFER_ACCEPT is active, drop bare ACK. */ -if (req->num_timeout <inet_csk (sk)->icsk_accept_queue.rskq_defer_accept && +if (req->num_timeout <READ_ONCE (inet_csk (sk)->icsk_accept_queue.rskq_defer_accept) &&TCP_SKB_CB (skb)->end_seq ==tcp_rsk (req)->rcv_isn +1 ) {inet_rsk (req)->acked =1 ;__NET_INC_STATS (sock_net (sk),LINUX_MIB_TCPDEFERACCEPTDROP );
Signature · 1 line
2.41.0.640.ga95def55d0-goog
Re: [PATCH net-next 6/6] tcp: set TCP_DEFER_ACCEPT locklessly
Quoted 5 lines · Eric Dumazet
rskq_defer_acceptfield can be read/written without the need of holding the socket lock.Signed-off-by Eric Dumazet <[email protected]>
Very nice series! Thank you! I doulbechecked every field and they are all READ_ONCE/WRITE_ONCE paired.
Quoted 66 lines
--- net/ipv4/tcp.c | 13 ++++++------- net/ipv4/tcp_input.c | 2 +- net/ipv4/tcp_minisocks.c | 2 +- 3 files changed, 8 insertions(+), 9 deletions(-)net/ipv4/tcp.c +6
-7@@ -3479,6 +3479,12 @@ int do_tcp_setsockopt(struct sock *sk, int level, int optname,else WRITE_ONCE (tp->linger2, val *HZ );return 0 ; +case TCP_DEFER_ACCEPT : +/* Translate value in seconds to number of retransmits */ +WRITE_ONCE (icsk->icsk_accept_queue.rskq_defer_accept, +secs_to_retrans (val,TCP_TIMEOUT_INIT /HZ , +TCP_RTO_MAX /HZ )); +return 0 ; }sockopt_lock_sock (sk);@@ -3584,13 +3590,6 @@ int do_tcp_setsockopt(struct sock *sk, int level, int optname, tp->save_syn = val;break ; -case TCP_DEFER_ACCEPT : -/* Translate value in seconds to number of retransmits */ -WRITE_ONCE (icsk->icsk_accept_queue.rskq_defer_accept, -secs_to_retrans (val,TCP_TIMEOUT_INIT /HZ , -TCP_RTO_MAX /HZ )); -break ; -case TCP_WINDOW_CLAMP : err =tcp_set_window_clamp (sk, val);break ;net/ipv4/tcp_input.c +1
-1@@ -6325,7 +6325,7 @@ static int tcp_rcv_synsent_state_process(struct sock *sk, struct sk_buff *skb,if (fastopen_fail)return -1 ;if (sk->sk_write_pending || - icsk->icsk_accept_queue.rskq_defer_accept || +READ_ONCE (icsk->icsk_accept_queue.rskq_defer_accept) ||inet_csk_in_pingpong_mode (sk)) {/* Save one ACK. Data will be ready after * several ticks, if write_pending is set. net/ipv4/tcp_minisocks.c +1
-2@@ -794,7 +794,7 @@ struct sock *tcp_check_req(struct sock *sk, struct sk_buff *skb,return sk;/* While TCP_DEFER_ACCEPT is active, drop bare ACK. */ -if (req->num_timeout <inet_csk (sk)->icsk_accept_queue.rskq_defer_accept && +if (req->num_timeout <READ_ONCE (inet_csk (sk)->icsk_accept_queue.rskq_defer_accept) &&TCP_SKB_CB (skb)->end_seq ==tcp_rsk (req)->rcv_isn +1 ) {inet_rsk (req)->acked =1 ;__NET_INC_STATS (sock_net (sk),LINUX_MIB_TCPDEFERACCEPTDROP ); --2.41.0.640.ga95def55d0-goog
This series was applied to netdev/net-next.git (main) by David S. Miller <[email protected]>:
Hello:
This series was applied to netdev/net-next.git (main) by David S. Miller <[email protected]>:
Quoted 13 lines · On Fri, 4 Aug 2023 14:46:10 +0000 you wrote:
This series is avoiding the socket lock for six TCP options.
They are not heavily used, but this exercise can give ideas for other parts of TCP/IP stack :)
Eric Dumazet (6):
tcp: setTCP_SYNCNTlocklessly
tcp: setTCP_USER_TIMEOUTlocklessly
tcp: setTCP_KEEPINTVLlocklessly
tcp: setTCP_KEEPCNTlocklessly
tcp: setTCP_LINGER2locklessly
tcp: setTCP_DEFER_ACCEPTlocklessly
Quoted 1 line
[...]
Here is the summary with links:
- [net-next,1/6] tcp: set TCP_SYNCNT locklessly
https://git.kernel.org/netdev/net-next/c/d44fd4a767b3 - [net-next,2/6] tcp: set TCP_USER_TIMEOUT locklessly https://git.kernel.org/netdev/net-next/c/d58f2e15aa0c - [net-next,3/6] tcp: set TCP_KEEPINTVL locklessly https://git.kernel.org/netdev/net-next/c/6fd70a6b4e6f - [net-next,4/6] tcp: set TCP_KEEPCNT locklessly https://git.kernel.org/netdev/net-next/c/84485080cbc1 - [net-next,5/6] tcp: set TCP_LINGER2 locklessly https://git.kernel.org/netdev/net-next/c/a81722ddd7e4 - [net-next,6/6] tcp: set TCP_DEFER_ACCEPT locklessly https://git.kernel.org/netdev/net-next/c/6e97ba552b8d
You are awesome, thank you!
Signature · 2 lines
Deet-doot-dot, I am a bot. https://korg.docs.kernel.org/patchwork/pwbot.html