tcp: enforce receive buffer memory limits by allowing the tcp window to shrink
Subsystems (4)
PATCH v3 1/1 tcp: enforce receive buffer memory limits by allowing the tcp window to shrink
From: "[email protected]" <[email protected]>
Under certain circumstances, the tcp receive buffer memory limit set by autotuning (sk_rcvbuf) is increased due to incoming data packets as a result of the window not closing when it should be. This can result in the receive buffer growing all the way up to tcp_rmem[2], even for tcp sessions with a low BDP.
To reproduce: Connect a TCP session with the receiver doing nothing and the sender sending small packets (an infinite loop of socket send() with 4 bytes of payload with a sleep of 1 ms in between each send()). This will cause the tcp receive buffer to grow all the way up to tcp_rmem[2].
As a result, a host can have individual tcp sessions with receive buffers of size tcp_rmem[2], and the host itself can reach tcp_mem limits, causing the host to go into tcp memory pressure mode.
The fundamental issue is the relationship between the granularity of the window scaling factor and the number of byte ACKed back to the sender. This problem has previously been identified in RFC 7323, appendix F [1].
The Linux kernel currently adheres to never shrinking the window.
In addition to the overallocation of memory mentioned above, the current behavior is functionally incorrect, because once tcp_rmem[2] is reached when no remediations remain (i.e. tcp collapse fails to free up any more memory and there are no packets to prune from the out-of-order queue), the receiver will drop in-window packets resulting in retransmissions and an eventual timeout of the tcp session. A receive buffer full condition should instead result in a zero window and an indefinite wait.
In practice, this problem is largely hidden for most flows. It is not applicable to mice flows. Elephant flows can send data fast enough to "overrun" the sk_rcvbuf limit (in a single ACK), triggering a zero window.
But this problem does show up for other types of flows. Examples are websockets and other type of flows that send small amounts of data spaced apart slightly in time. In these cases, we directly encounter the problem described in [1].
RFC 7323, section 2.4 [2], says there are instances when a retracted window can be offered, and that TCP implementations MUST ensure that they handle a shrinking window, as specified in RFC 1122, section 4.2.2.16 [3]. All prior RFCs on the topic of tcp window management have made clear that sender must accept a shrunk window from the receiver, including RFC 793 [4] and RFC 1323 [5].
This patch implements the functionality to shrink the tcp window when necessary to keep the right edge within the memory limit by autotuning (sk_rcvbuf). This new functionality is enabled with the new sysctl: net.ipv4.tcp_shrink_window
Additional information can be found at:
https://blog.cloudflare.com/unbounded-memory-usage-by-tcp-for-receive-buffers-and-how-we-fixed-it/
[1] https://www.rfc-editor.org/rfc/rfc7323#appendix-F
[2] https://www.rfc-editor.org/rfc/rfc7323#section-2.4
[3] https://www.rfc-editor.org/rfc/rfc1122#page-91
[4] https://www.rfc-editor.org/rfc/rfc793
[5] https://www.rfc-editor.org/rfc/rfc1323
--- Documentation/networking/ip-sysctl.rst | 13 +++++ include/net/netns/ipv4.h | 1 + net/ipv4/sysctl_net_ipv4.c | 9 ++++ net/ipv4/tcp_ipv4.c | 2 + net/ipv4/tcp_output.c | 73 ++++++++++++++++++++++++-- 5 files changed, 93 insertions(+), 5 deletions(-)
Documentation/networking/ip-sysctl.rst +13 -0
@@ -981,6 +981,19 @@ tcp_tw_reuse - INTEGER tcp_window_scaling - BOOLEAN Enable window scaling as defined in RFC1323. +tcp_shrink_window - BOOLEAN + This changes how the TCP receive window is calculated. + + RFC 7323, section 2.4, says there are instances when a retracted + window can be offered, and that TCP implementations MUST ensure + that they handle a shrinking window, as specified in RFC 1122. + + - 0 - Disabled. The window is never shrunk. + - 1 - Enabled. The window is shrunk when necessary to remain within + the memory limit set by autotuning (sk_rcvbuf). + + Default: 0 + tcp_wmem - vector of 3 INTEGERs: min, default, max min: Amount of memory reserved for send buffers for TCP sockets. Each TCP socket has rights to use it due to fact of its birth.
include/net/netns/ipv4.h +1 -0
@@ -65,6 +65,7 @@ struct netns_ipv4 {#endif bool fib_has_custom_local_routes;bool fib_offload_disabled; +u8 sysctl_tcp_shrink_window;#ifdef CONFIG_IP_ROUTE_CLASSID atomic_t fib_num_tclassid_users;#endif
net/ipv4/sysctl_net_ipv4.c +9 -0
@@ -1480,6 +1480,15 @@ static struct ctl_table ipv4_net_table[] = { .extra1 = SYSCTL_ZERO, .extra2 = &tcp_syn_linear_timeouts_max, }, + { + .procname = "tcp_shrink_window", + .data = &init_net.ipv4.sysctl_tcp_shrink_window, + .maxlen = sizeof(u8), + .mode = 0644, + .proc_handler = proc_dou8vec_minmax, + .extra1 = SYSCTL_ZERO, + .extra2 = SYSCTL_ONE, + }, { } };
net/ipv4/tcp_ipv4.c +2 -0
@@ -3281,6 +3281,8 @@ static int __net_init tcp_sk_init(struct net *net) net->ipv4.tcp_congestion_control = &tcp_reno; net->ipv4.sysctl_tcp_syn_linear_timeouts =4 ; + net->ipv4.sysctl_tcp_shrink_window =0 ; +return 0 ; }
net/ipv4/tcp_output.c +68 -5
105 lines not on this page · raw message
Signature · 1 line
2.40.0
net/ipv4/sysctl_net_ipv4.c static struct ctl_table ipv4_net_table[] = { · line 1491
Quoted 9 lines · On Fri, 9 Jun 2023 15:47:06 -0500 Mike Freemon <[email protected]> wrote:
net/ipv4/sysctl_net_ipv4.c +9
-0+ { + .procname = "tcp_shrink_window", + .data = &init_net.ipv4.sysctl_tcp_shrink_window, + .maxlen = sizeof(u8), + .mode = 0644, + .proc_handler = proc_dou8vec_minmax, + .extra1 = SYSCTL_ZERO, + .extra2 = SYSCTL_ONE, + },
NAK on introducing another sysctl.
Quoted 14 lines · Stephen Hemminger
net/ipv4/sysctl_net_ipv4.c static struct ctl_table ipv4_net_table[] = { · line 1491
Quoted 9 lines · On Fri, 9 Jun 2023 15:47:06 -0500 Mike Freemon <[email protected]> wrote:
net/ipv4/sysctl_net_ipv4.c +9
-0+ { + .procname = "tcp_shrink_window", + .data = &init_net.ipv4.sysctl_tcp_shrink_window, + .maxlen = sizeof(u8), + .mode = 0644, + .proc_handler = proc_dou8vec_minmax, + .extra1 = SYSCTL_ZERO, + .extra2 = SYSCTL_ONE, + },
Quoted 1 line
NAK on introducing another sysctl.
We respectfully disagree.
A sysctl makes a lot of sense to us, as it allows us to not break many packetdrill tests.
Quoted 149 lines · Mike Freemon
From: "[email protected]" <[email protected]>
Under certain circumstances, the tcp receive buffer memory limit set by autotuning (
sk_rcvbuf) is increased due to incoming data packets as a result of the window not closing when it should be. This can result in the receive buffer growing all the way up to tcp_rmem[2], even for tcp sessions with a low BDP.To reproduce: Connect a TCP session with the receiver doing nothing and the sender sending small packets (an infinite loop of socket
send()with 4 bytes of payload with a sleep of 1 ms in between eachsend()). This will cause the tcp receive buffer to grow all the way up to tcp_rmem[2].As a result, a host can have individual tcp sessions with receive buffers of size tcp_rmem[2], and the host itself can reach
tcp_memlimits, causing the host to go into tcp memory pressure mode.The fundamental issue is the relationship between the granularity of the window scaling factor and the number of byte ACKed back to the sender. This problem has previously been identified in RFC 7323, appendix F [1].
The Linux kernel currently adheres to never shrinking the window.
In addition to the overallocation of memory mentioned above, the current behavior is functionally incorrect, because once tcp_rmem[2] is reached when no remediations remain (i.e. tcp collapse fails to free up any more memory and there are no packets to prune from the out-of-order queue), the receiver will drop in-window packets resulting in retransmissions and an eventual timeout of the tcp session. A receive buffer full condition should instead result in a zero window and an indefinite wait.
In practice, this problem is largely hidden for most flows. It is not applicable to mice flows. Elephant flows can send data fast enough to "overrun" the
sk_rcvbuflimit (in a single ACK), triggering a zero window.But this problem does show up for other types of flows. Examples are websockets and other type of flows that send small amounts of data spaced apart slightly in time. In these cases, we directly encounter the problem described in [1].
RFC 7323, section 2.4 [2], says there are instances when a retracted window can be offered, and that TCP implementations MUST ensure that they handle a shrinking window, as specified in RFC 1122, section 4.2.2.16 [3]. All prior RFCs on the topic of tcp window management have made clear that sender must accept a shrunk window from the receiver, including RFC 793 [4] and RFC 1323 [5].
This patch implements the functionality to shrink the tcp window when necessary to keep the right edge within the memory limit by autotuning (
sk_rcvbuf). This new functionality is enabled with the new sysctl: net.ipv4.tcp_shrink_windowAdditional information can be found at:
https://blog.cloudflare.com/unbounded-memory-usage-by-tcp-for-receive-buffers-and-how-we-fixed-it/[1] https://www.rfc-editor.org/rfc/rfc7323#appendix-F
[2] https://www.rfc-editor.org/rfc/rfc7323#section-2.4
[3] https://www.rfc-editor.org/rfc/rfc1122#page-91
[4] https://www.rfc-editor.org/rfc/rfc793
[5] https://www.rfc-editor.org/rfc/rfc1323Signed-off-by Mike Freemon <[email protected]>--- Documentation/networking/ip-sysctl.rst | 13 +++++ include/net/netns/ipv4.h | 1 + net/ipv4/sysctl_net_ipv4.c | 9 ++++ net/ipv4/tcp_ipv4.c | 2 + net/ipv4/tcp_output.c | 73 ++++++++++++++++++++++++-- 5 files changed, 93 insertions(+), 5 deletions(-)Documentation/networking/ip-sysctl.rst +13
-0@@ -981,6 +981,19 @@ tcp_tw_reuse - INTEGER tcp_window_scaling - BOOLEAN Enable window scaling as defined in RFC1323. +tcp_shrink_window - BOOLEAN + This changes how the TCP receive window is calculated. + + RFC 7323, section 2.4, says there are instances when a retracted + window can be offered, and that TCP implementations MUST ensure + that they handle a shrinking window, as specified in RFC 1122. + + - 0 - Disabled. The window is never shrunk. + - 1 - Enabled. The window is shrunk when necessary to remain within + the memory limit set by autotuning (sk_rcvbuf). + + Default: 0 + tcp_wmem - vector of 3 INTEGERs: min, default, max min: Amount of memory reserved for send buffers for TCP sockets. Each TCP socket has rights to use it due to fact of its birth.include/net/netns/ipv4.h +1
-0@@ -65,6 +65,7 @@ struct netns_ipv4 {#endif bool fib_has_custom_local_routes;bool fib_offload_disabled; +u8 sysctl_tcp_shrink_window;#ifdef CONFIG_IP_ROUTE_CLASSID atomic_t fib_num_tclassid_users;#endif net/ipv4/sysctl_net_ipv4.c +9
-0@@ -1480,6 +1480,15 @@ static struct ctl_table ipv4_net_table[] = { .extra1 = SYSCTL_ZERO, .extra2 = &tcp_syn_linear_timeouts_max, }, + { + .procname = "tcp_shrink_window", + .data = &init_net.ipv4.sysctl_tcp_shrink_window, + .maxlen = sizeof(u8), + .mode = 0644, + .proc_handler = proc_dou8vec_minmax, + .extra1 = SYSCTL_ZERO, + .extra2 = SYSCTL_ONE, + }, { } };net/ipv4/tcp_ipv4.c +2
-0@@ -3281,6 +3281,8 @@ static int __net_init tcp_sk_init(struct net *net) net->ipv4.tcp_congestion_control = &tcp_reno; net->ipv4.sysctl_tcp_syn_linear_timeouts =4 ; + net->ipv4.sysctl_tcp_shrink_window =0 ; +return 0 ; }net/ipv4/tcp_output.c +0
-0
net/ipv4/tcp_output.c static u16 tcp_select_window(struct sock *sk) · line 263
Quoted 4 lines
net/ipv4/tcp_output.c +1
-0@@ -260,8 +260,8 @@ static u16 tcp_select_window(struct sock *sk)u32 old_win = tp->rcv_wnd;u32 cur_win =tcp_receive_window (tp);u32 new_win =__tcp_select_window (sk); +struct net *net =sock_net (sk);
Here you cache sock_net() in @net variable.
Quoted 7 lines
net/ipv4/tcp_output.c +0
-1-/* Never shrink the offered window */ if (new_win < cur_win) {/* Danger Will Robinson! * Don't update rcv_wup/rcv_wnd here or else @@ -270,11 +270,15 @@ static u16 tcp_select_window(struct sock *sk)*
net/ipv4/tcp_output.c static u16 tcp_select_window(struct sock *sk) · line 276
Quoted 10 lines
net/ipv4/tcp_output.c +4
-4* Relax Will Robinson. */ -if (new_win ==0 ) -NET_INC_STATS (sock_net (sk), -LINUX_MIB_TCPWANTZEROWINDOWADV ); - new_win =ALIGN (cur_win,1 << tp->rx_opt.rcv_wscale); +if (!READ_ONCE (net->ipv4.sysctl_tcp_shrink_window)) { +/* Never shrink the offered window */ +if (new_win ==0 ) + NET_INC_STATS(sock_net(sk),
You then can replace sock_net(sk) by @net here.
Quoted 39 lines
net/ipv4/tcp_output.c +17
-0+ LINUX_MIB_TCPWANTZEROWINDOWADV); + new_win =ALIGN (cur_win,1 << tp->rx_opt.rcv_wscale); + } } + tp->rcv_wnd = new_win; tp->rcv_wup = tp->rcv_nxt;@@ -3003,6 +3007,7 @@ u32 __tcp_select_window(struct sock *sk) {struct inet_connection_sock *icsk =inet_csk (sk);struct tcp_sock *tp =tcp_sk (sk); +struct net *net =sock_net (sk);/* MSS for the peer's data. Previous versions used mss_clamp * here. I don't know if the value based on our guesses * of peer's MSS is better for the performance. It's more correct @@ -3024,6 +3029,12 @@ u32 __tcp_select_window(struct sock *sk)if (mss <=0 )return 0 ; } + +if (READ_ONCE (net->ipv4.sysctl_tcp_shrink_window)) + goto shrink_window_allowed; + +/* do not allow window to shrink */ +if (free_space < (full_space >>1 )) { icsk->icsk_ack.quick =0 ;@@ -3077,6 +3088,58 @@ u32 __tcp_select_window(struct sock *sk) window = free_space; } +return window; + +shrink_window_allowed: +/* new window should always be an exact multiple of scaling factor */ + free_space =round_down (free_space,1 << tp->rx_opt.rcv_wscale); +
net/ipv4/tcp_output.c u32 __tcp_select_window(struct sock *sk) · line 3106
Quoted 10 lines
net/ipv4/tcp_output.c +10
-0+if (free_space < (full_space >>1 )) { + icsk->icsk_ack.quick =0 ; + +if (tcp_under_memory_pressure (sk)) +tcp_adjust_rcv_ssthresh (sk); + +/* if free space is too low, return a zero window */ +if (free_space < (allowed_space >>4 ) || free_space < mss || + free_space < (1 << tp->rx_opt.rcv_wscale)) +return 0 ;
Are you sure this block can not be shared with the existing one ?
Existing one has this added part:
free_space = round_down(free_space, 1 << tp->rx_opt.rcv_wscale);
Not sure why this would break the tcp_shrink_window == 1 case.
Quoted 26 lines
net/ipv4/tcp_output.c +26
-0+ } + +if (free_space > tp->rcv_ssthresh) { + free_space = tp->rcv_ssthresh; +/* new window should always be an exact multiple of scaling factor +* +* For this case, we ALIGN "up" (increase free_space) because +* we know free_space is not zero here, it has been reduced from +* the memory-based limit, and rcv_ssthresh is not a hard limit +* (unlike sk_rcvbuf). +*/ + free_space =ALIGN (free_space, (1 << tp->rx_opt.rcv_wscale)); + } + +/* Don't do rounding if we are using window scaling, since the +* scaled window will not line up with the MSS boundary anyway. +*/ +if (tp->rx_opt.rcv_wscale) { + window = free_space; + }else { + window = tp->rcv_wnd; +/* Get the largest window that is a nice multiple of mss. +* Window clamp already applied above. +* If our current window offering is within 1 mss of the +* free space we just keep it. This prevents the divide +* and multiply from happening most of the time.
net/ipv4/tcp_output.c u32 __tcp_select_window(struct sock *sk) · line 3142
Quoted 10 lines
net/ipv4/tcp_output.c +10
-0+* We also don't do any window rounding when the free space +* is too small. +*/ +if (window <= free_space - mss || window > free_space) + window =rounddown (free_space, mss); +else if (mss == full_space && + free_space > window + (full_space >>1 )) + window = free_space; + } +
I am a bit surprised we can not come up with something simpler.
I was suggesting to look at the sysctl only if we were in a dangerous operation,
Neal, can you have a look ?
Thanks.
Quoted 266 lines · Eric Dumazet
Quoted 149 lines · Mike Freemon
From: "[email protected]" <[email protected]>
Under certain circumstances, the tcp receive buffer memory limit set by autotuning (
sk_rcvbuf) is increased due to incoming data packets as a result of the window not closing when it should be. This can result in the receive buffer growing all the way up to tcp_rmem[2], even for tcp sessions with a low BDP.To reproduce: Connect a TCP session with the receiver doing nothing and the sender sending small packets (an infinite loop of socket
send()with 4 bytes of payload with a sleep of 1 ms in between eachsend()). This will cause the tcp receive buffer to grow all the way up to tcp_rmem[2].As a result, a host can have individual tcp sessions with receive buffers of size tcp_rmem[2], and the host itself can reach
tcp_memlimits, causing the host to go into tcp memory pressure mode.The fundamental issue is the relationship between the granularity of the window scaling factor and the number of byte ACKed back to the sender. This problem has previously been identified in RFC 7323, appendix F [1].
The Linux kernel currently adheres to never shrinking the window.
In addition to the overallocation of memory mentioned above, the current behavior is functionally incorrect, because once tcp_rmem[2] is reached when no remediations remain (i.e. tcp collapse fails to free up any more memory and there are no packets to prune from the out-of-order queue), the receiver will drop in-window packets resulting in retransmissions and an eventual timeout of the tcp session. A receive buffer full condition should instead result in a zero window and an indefinite wait.
In practice, this problem is largely hidden for most flows. It is not applicable to mice flows. Elephant flows can send data fast enough to "overrun" the
sk_rcvbuflimit (in a single ACK), triggering a zero window.But this problem does show up for other types of flows. Examples are websockets and other type of flows that send small amounts of data spaced apart slightly in time. In these cases, we directly encounter the problem described in [1].
RFC 7323, section 2.4 [2], says there are instances when a retracted window can be offered, and that TCP implementations MUST ensure that they handle a shrinking window, as specified in RFC 1122, section 4.2.2.16 [3]. All prior RFCs on the topic of tcp window management have made clear that sender must accept a shrunk window from the receiver, including RFC 793 [4] and RFC 1323 [5].
This patch implements the functionality to shrink the tcp window when necessary to keep the right edge within the memory limit by autotuning (
sk_rcvbuf). This new functionality is enabled with the new sysctl: net.ipv4.tcp_shrink_windowAdditional information can be found at:
https://blog.cloudflare.com/unbounded-memory-usage-by-tcp-for-receive-buffers-and-how-we-fixed-it/[1] https://www.rfc-editor.org/rfc/rfc7323#appendix-F
[2] https://www.rfc-editor.org/rfc/rfc7323#section-2.4
[3] https://www.rfc-editor.org/rfc/rfc1122#page-91
[4] https://www.rfc-editor.org/rfc/rfc793
[5] https://www.rfc-editor.org/rfc/rfc1323Signed-off-by Mike Freemon <[email protected]>--- Documentation/networking/ip-sysctl.rst | 13 +++++ include/net/netns/ipv4.h | 1 + net/ipv4/sysctl_net_ipv4.c | 9 ++++ net/ipv4/tcp_ipv4.c | 2 + net/ipv4/tcp_output.c | 73 ++++++++++++++++++++++++-- 5 files changed, 93 insertions(+), 5 deletions(-)Documentation/networking/ip-sysctl.rst +13
-0@@ -981,6 +981,19 @@ tcp_tw_reuse - INTEGER tcp_window_scaling - BOOLEAN Enable window scaling as defined in RFC1323. +tcp_shrink_window - BOOLEAN + This changes how the TCP receive window is calculated. + + RFC 7323, section 2.4, says there are instances when a retracted + window can be offered, and that TCP implementations MUST ensure + that they handle a shrinking window, as specified in RFC 1122. + + - 0 - Disabled. The window is never shrunk. + - 1 - Enabled. The window is shrunk when necessary to remain within + the memory limit set by autotuning (sk_rcvbuf). + + Default: 0 + tcp_wmem - vector of 3 INTEGERs: min, default, max min: Amount of memory reserved for send buffers for TCP sockets. Each TCP socket has rights to use it due to fact of its birth.include/net/netns/ipv4.h +1
-0@@ -65,6 +65,7 @@ struct netns_ipv4 {#endif bool fib_has_custom_local_routes;bool fib_offload_disabled; +u8 sysctl_tcp_shrink_window;#ifdef CONFIG_IP_ROUTE_CLASSID atomic_t fib_num_tclassid_users;#endif net/ipv4/sysctl_net_ipv4.c +9
-0@@ -1480,6 +1480,15 @@ static struct ctl_table ipv4_net_table[] = { .extra1 = SYSCTL_ZERO, .extra2 = &tcp_syn_linear_timeouts_max, }, + { + .procname = "tcp_shrink_window", + .data = &init_net.ipv4.sysctl_tcp_shrink_window, + .maxlen = sizeof(u8), + .mode = 0644, + .proc_handler = proc_dou8vec_minmax, + .extra1 = SYSCTL_ZERO, + .extra2 = SYSCTL_ONE, + }, { } };net/ipv4/tcp_ipv4.c +2
-0@@ -3281,6 +3281,8 @@ static int __net_init tcp_sk_init(struct net *net) net->ipv4.tcp_congestion_control = &tcp_reno; net->ipv4.sysctl_tcp_syn_linear_timeouts =4 ; + net->ipv4.sysctl_tcp_shrink_window =0 ; +return 0 ; }net/ipv4/tcp_output.c +0
-0net/ipv4/tcp_output.c static u16 tcp_select_window(struct sock *sk) · line 263
Quoted 4 lines
net/ipv4/tcp_output.c +1
-0@@ -260,8 +260,8 @@ static u16 tcp_select_window(struct sock *sk)u32 old_win = tp->rcv_wnd;u32 cur_win =tcp_receive_window (tp);u32 new_win =__tcp_select_window (sk); +struct net *net =sock_net (sk);Here you cache
sock_net()in @net variable.Quoted 7 lines
net/ipv4/tcp_output.c +0
-1-/* Never shrink the offered window */ if (new_win < cur_win) {/* Danger Will Robinson! * Don't update rcv_wup/rcv_wnd here or else @@ -270,11 +270,15 @@ static u16 tcp_select_window(struct sock *sk)* net/ipv4/tcp_output.c static u16 tcp_select_window(struct sock *sk) · line 276
Quoted 10 lines
net/ipv4/tcp_output.c +4
-4* Relax Will Robinson. */ -if (new_win ==0 ) -NET_INC_STATS (sock_net (sk), -LINUX_MIB_TCPWANTZEROWINDOWADV ); - new_win =ALIGN (cur_win,1 << tp->rx_opt.rcv_wscale); +if (!READ_ONCE (net->ipv4.sysctl_tcp_shrink_window)) { +/* Never shrink the offered window */ +if (new_win ==0 ) + NET_INC_STATS(sock_net(sk),You then can replace
sock_net(sk)by @net here.Quoted 39 lines
net/ipv4/tcp_output.c +17
-0+ LINUX_MIB_TCPWANTZEROWINDOWADV); + new_win =ALIGN (cur_win,1 << tp->rx_opt.rcv_wscale); + } } + tp->rcv_wnd = new_win; tp->rcv_wup = tp->rcv_nxt;@@ -3003,6 +3007,7 @@ u32 __tcp_select_window(struct sock *sk) {struct inet_connection_sock *icsk =inet_csk (sk);struct tcp_sock *tp =tcp_sk (sk); +struct net *net =sock_net (sk);/* MSS for the peer's data. Previous versions used mss_clamp * here. I don't know if the value based on our guesses * of peer's MSS is better for the performance. It's more correct @@ -3024,6 +3029,12 @@ u32 __tcp_select_window(struct sock *sk)if (mss <=0 )return 0 ; } + +if (READ_ONCE (net->ipv4.sysctl_tcp_shrink_window)) + goto shrink_window_allowed; + +/* do not allow window to shrink */ +if (free_space < (full_space >>1 )) { icsk->icsk_ack.quick =0 ;@@ -3077,6 +3088,58 @@ u32 __tcp_select_window(struct sock *sk) window = free_space; } +return window; + +shrink_window_allowed: +/* new window should always be an exact multiple of scaling factor */ + free_space =round_down (free_space,1 << tp->rx_opt.rcv_wscale); +net/ipv4/tcp_output.c u32 __tcp_select_window(struct sock *sk) · line 3106
Quoted 10 lines
net/ipv4/tcp_output.c +10
-0+if (free_space < (full_space >>1 )) { + icsk->icsk_ack.quick =0 ; + +if (tcp_under_memory_pressure (sk)) +tcp_adjust_rcv_ssthresh (sk); + +/* if free space is too low, return a zero window */ +if (free_space < (allowed_space >>4 ) || free_space < mss || + free_space < (1 << tp->rx_opt.rcv_wscale)) +return 0 ;Are you sure this block can not be shared with the existing one ?
Existing one has this added part:
free_space=round_down(free_space, 1 << tp->rx_opt.rcv_wscale);Not sure why this would break the
tcp_shrink_window== 1 case.Quoted 30 lines
net/ipv4/tcp_output.c +30
-0+ } + +if (free_space > tp->rcv_ssthresh) { + free_space = tp->rcv_ssthresh; +/* new window should always be an exact multiple of scaling factor +* +* For this case, we ALIGN "up" (increase free_space) because +* we know free_space is not zero here, it has been reduced from +* the memory-based limit, and rcv_ssthresh is not a hard limit +* (unlike sk_rcvbuf). +*/ + free_space =ALIGN (free_space, (1 << tp->rx_opt.rcv_wscale)); + } + +/* Don't do rounding if we are using window scaling, since the +* scaled window will not line up with the MSS boundary anyway. +*/ +if (tp->rx_opt.rcv_wscale) { + window = free_space; + }else { + window = tp->rcv_wnd; +/* Get the largest window that is a nice multiple of mss. +* Window clamp already applied above. +* If our current window offering is within 1 mss of the +* free space we just keep it. This prevents the divide +* and multiply from happening most of the time. +* We also don't do any window rounding when the free space +* is too small. +*/ +if (window <= free_space - mss || window > free_space)
Quoted 8 lines
net/ipv4/tcp_output.c u32 __tcp_select_window(struct sock *sk) · line 3142
Quoted 6 lines
net/ipv4/tcp_output.c +6
-0+ window =rounddown (free_space, mss); + else if (mss == full_space && + free_space > window + (full_space >> 1)) + window = free_space; + } +I am a bit surprised we can not come up with something simpler.
Since sysctl_tcp_shrink_window was only intended to apply when rcv_wscale > 0, I can make that explicit, which will eliminate this whole latter block of code.
Quoted 229 lines · Eric Dumazet
Quoted 149 lines · Mike Freemon
From: "[email protected]" <[email protected]>
Under certain circumstances, the tcp receive buffer memory limit set by autotuning (
sk_rcvbuf) is increased due to incoming data packets as a result of the window not closing when it should be. This can result in the receive buffer growing all the way up to tcp_rmem[2], even for tcp sessions with a low BDP.To reproduce: Connect a TCP session with the receiver doing nothing and the sender sending small packets (an infinite loop of socket
send()with 4 bytes of payload with a sleep of 1 ms in between eachsend()). This will cause the tcp receive buffer to grow all the way up to tcp_rmem[2].As a result, a host can have individual tcp sessions with receive buffers of size tcp_rmem[2], and the host itself can reach
tcp_memlimits, causing the host to go into tcp memory pressure mode.The fundamental issue is the relationship between the granularity of the window scaling factor and the number of byte ACKed back to the sender. This problem has previously been identified in RFC 7323, appendix F [1].
The Linux kernel currently adheres to never shrinking the window.
In addition to the overallocation of memory mentioned above, the current behavior is functionally incorrect, because once tcp_rmem[2] is reached when no remediations remain (i.e. tcp collapse fails to free up any more memory and there are no packets to prune from the out-of-order queue), the receiver will drop in-window packets resulting in retransmissions and an eventual timeout of the tcp session. A receive buffer full condition should instead result in a zero window and an indefinite wait.
In practice, this problem is largely hidden for most flows. It is not applicable to mice flows. Elephant flows can send data fast enough to "overrun" the
sk_rcvbuflimit (in a single ACK), triggering a zero window.But this problem does show up for other types of flows. Examples are websockets and other type of flows that send small amounts of data spaced apart slightly in time. In these cases, we directly encounter the problem described in [1].
RFC 7323, section 2.4 [2], says there are instances when a retracted window can be offered, and that TCP implementations MUST ensure that they handle a shrinking window, as specified in RFC 1122, section 4.2.2.16 [3]. All prior RFCs on the topic of tcp window management have made clear that sender must accept a shrunk window from the receiver, including RFC 793 [4] and RFC 1323 [5].
This patch implements the functionality to shrink the tcp window when necessary to keep the right edge within the memory limit by autotuning (
sk_rcvbuf). This new functionality is enabled with the new sysctl: net.ipv4.tcp_shrink_windowAdditional information can be found at:
https://blog.cloudflare.com/unbounded-memory-usage-by-tcp-for-receive-buffers-and-how-we-fixed-it/[1] https://www.rfc-editor.org/rfc/rfc7323#appendix-F
[2] https://www.rfc-editor.org/rfc/rfc7323#section-2.4
[3] https://www.rfc-editor.org/rfc/rfc1122#page-91
[4] https://www.rfc-editor.org/rfc/rfc793
[5] https://www.rfc-editor.org/rfc/rfc1323Signed-off-by Mike Freemon <[email protected]>--- Documentation/networking/ip-sysctl.rst | 13 +++++ include/net/netns/ipv4.h | 1 + net/ipv4/sysctl_net_ipv4.c | 9 ++++ net/ipv4/tcp_ipv4.c | 2 + net/ipv4/tcp_output.c | 73 ++++++++++++++++++++++++-- 5 files changed, 93 insertions(+), 5 deletions(-)Documentation/networking/ip-sysctl.rst +13
-0@@ -981,6 +981,19 @@ tcp_tw_reuse - INTEGER tcp_window_scaling - BOOLEAN Enable window scaling as defined in RFC1323. +tcp_shrink_window - BOOLEAN + This changes how the TCP receive window is calculated. + + RFC 7323, section 2.4, says there are instances when a retracted + window can be offered, and that TCP implementations MUST ensure + that they handle a shrinking window, as specified in RFC 1122. + + - 0 - Disabled. The window is never shrunk. + - 1 - Enabled. The window is shrunk when necessary to remain within + the memory limit set by autotuning (sk_rcvbuf). + + Default: 0 + tcp_wmem - vector of 3 INTEGERs: min, default, max min: Amount of memory reserved for send buffers for TCP sockets. Each TCP socket has rights to use it due to fact of its birth.include/net/netns/ipv4.h +1
-0@@ -65,6 +65,7 @@ struct netns_ipv4 {#endif bool fib_has_custom_local_routes;bool fib_offload_disabled; +u8 sysctl_tcp_shrink_window;#ifdef CONFIG_IP_ROUTE_CLASSID atomic_t fib_num_tclassid_users;#endif net/ipv4/sysctl_net_ipv4.c +9
-0@@ -1480,6 +1480,15 @@ static struct ctl_table ipv4_net_table[] = { .extra1 = SYSCTL_ZERO, .extra2 = &tcp_syn_linear_timeouts_max, }, + { + .procname = "tcp_shrink_window", + .data = &init_net.ipv4.sysctl_tcp_shrink_window, + .maxlen = sizeof(u8), + .mode = 0644, + .proc_handler = proc_dou8vec_minmax, + .extra1 = SYSCTL_ZERO, + .extra2 = SYSCTL_ONE, + }, { } };net/ipv4/tcp_ipv4.c +2
-0@@ -3281,6 +3281,8 @@ static int __net_init tcp_sk_init(struct net *net) net->ipv4.tcp_congestion_control = &tcp_reno; net->ipv4.sysctl_tcp_syn_linear_timeouts =4 ; + net->ipv4.sysctl_tcp_shrink_window =0 ; +return 0 ; }net/ipv4/tcp_output.c +0
-0net/ipv4/tcp_output.c static u16 tcp_select_window(struct sock *sk) · line 263
Quoted 4 lines
net/ipv4/tcp_output.c +1
-0@@ -260,8 +260,8 @@ static u16 tcp_select_window(struct sock *sk)u32 old_win = tp->rcv_wnd;u32 cur_win =tcp_receive_window (tp);u32 new_win =__tcp_select_window (sk); +struct net *net =sock_net (sk);Here you cache
sock_net()in @net variable.Quoted 7 lines
net/ipv4/tcp_output.c +0
-1-/* Never shrink the offered window */ if (new_win < cur_win) {/* Danger Will Robinson! * Don't update rcv_wup/rcv_wnd here or else @@ -270,11 +270,15 @@ static u16 tcp_select_window(struct sock *sk)* net/ipv4/tcp_output.c static u16 tcp_select_window(struct sock *sk) · line 276
Quoted 10 lines
net/ipv4/tcp_output.c +4
-4* Relax Will Robinson. */ -if (new_win ==0 ) -NET_INC_STATS (sock_net (sk), -LINUX_MIB_TCPWANTZEROWINDOWADV ); - new_win =ALIGN (cur_win,1 << tp->rx_opt.rcv_wscale); +if (!READ_ONCE (net->ipv4.sysctl_tcp_shrink_window)) { +/* Never shrink the offered window */ +if (new_win ==0 ) + NET_INC_STATS(sock_net(sk),You then can replace
sock_net(sk)by @net here.Quoted 39 lines
net/ipv4/tcp_output.c +17
-0+ LINUX_MIB_TCPWANTZEROWINDOWADV); + new_win =ALIGN (cur_win,1 << tp->rx_opt.rcv_wscale); + } } + tp->rcv_wnd = new_win; tp->rcv_wup = tp->rcv_nxt;@@ -3003,6 +3007,7 @@ u32 __tcp_select_window(struct sock *sk) {struct inet_connection_sock *icsk =inet_csk (sk);struct tcp_sock *tp =tcp_sk (sk); +struct net *net =sock_net (sk);/* MSS for the peer's data. Previous versions used mss_clamp * here. I don't know if the value based on our guesses * of peer's MSS is better for the performance. It's more correct @@ -3024,6 +3029,12 @@ u32 __tcp_select_window(struct sock *sk)if (mss <=0 )return 0 ; } + +if (READ_ONCE (net->ipv4.sysctl_tcp_shrink_window)) + goto shrink_window_allowed; + +/* do not allow window to shrink */ +if (free_space < (full_space >>1 )) { icsk->icsk_ack.quick =0 ;@@ -3077,6 +3088,58 @@ u32 __tcp_select_window(struct sock *sk) window = free_space; } +return window; + +shrink_window_allowed: +/* new window should always be an exact multiple of scaling factor */ + free_space =round_down (free_space,1 << tp->rx_opt.rcv_wscale); +net/ipv4/tcp_output.c u32 __tcp_select_window(struct sock *sk) · line 3106
Quoted 10 lines
net/ipv4/tcp_output.c +10
-0+if (free_space < (full_space >>1 )) { + icsk->icsk_ack.quick =0 ; + +if (tcp_under_memory_pressure (sk)) +tcp_adjust_rcv_ssthresh (sk); + +/* if free space is too low, return a zero window */ +if (free_space < (allowed_space >>4 ) || free_space < mss || + free_space < (1 << tp->rx_opt.rcv_wscale)) +return 0 ;Are you sure this block can not be shared with the existing one ?
Quoted 5 lines
Existing one has this added part:
free_space=round_down(free_space, 1 << tp->rx_opt.rcv_wscale);Not sure why this would break the
tcp_shrink_window== 1 case.
The main issue is the additional conditional on the if that returns 0:
free_space < (1 << tp->rx_opt.rcv_wscale))
This is an important difference that I think needs to continue to be different in the two cases.
I'll post a "v4" of the patch with the cleaned up code from the other comments so we can have a clean look at it.