mptcp: improve ADD_ADDR timeout handling
Subsystems (4)
Geliang Tang
reviewercover mptcp: improve ADD_ADDR timeout handling
From: Geliang Tang <[email protected]>
This series optimizes MPTCP's ADD_ADDR retransmission timeout:
1. Renames sysctl to add_addr_timeout_max for clearer semantics
2. Makes timeout adaptive to subflow RTT
3. Updates selftests for the rename
The adaptive timeout improves subflow establishment responsiveness while maintaining reliability through the max timeout cap. The rename provides better configuration clarity.
Geliang Tang (3):
mptcp: rename add_addr_timeout to add_addr_timeout_max
mptcp: make ADD_ADDR timeout adaptive to subflow RTT
selftests: mptcp: update for add_addr_timeout_max rename
Documentation/networking/mptcp-sysctl.rst | 6 ++--- net/mptcp/ctrl.c | 12 ++++----- net/mptcp/pm.c | 27 ++++++++++++++++--- net/mptcp/protocol.h | 2 +- .../testing/selftests/net/mptcp/mptcp_join.sh | 8 +++--- 5 files changed, 38 insertions(+), 17 deletions(-)
Signature · 1 line
2.48.1
Geliang Tang
reviewerPATCH 1/3 mptcp: rename add_addr_timeout to add_addr_timeout_max
From: Geliang Tang <[email protected]>
This renames the MPTCP sysctl variable to better reflect its purpose as the maximum ADD_ADDR retransmission timeout.
The patch includes updates to:
- Documentation/networking/mptcp-sysctl.rst
- net/mptcp/ctrl.c
- net/mptcp/pm.c
- net/mptcp/protocol.h
All references to the old variable name have been updated to use the new naming convention.
The functionality remains unchanged.
--- Documentation/networking/mptcp-sysctl.rst | 6 +++--- net/mptcp/ctrl.c | 12 ++++++------ net/mptcp/pm.c | 6 +++--- net/mptcp/protocol.h | 2 +- 4 files changed, 13 insertions(+), 13 deletions(-)
Documentation/networking/mptcp-sysctl.rst +3 -3
@@ -7,9 +7,9 @@ MPTCP Sysfs variables /proc/sys/net/mptcp/* Variables =============================== -add_addr_timeout - INTEGER (seconds) - Set the timeout after which an ADD_ADDR control message will be - resent to an MPTCP peer that has not acknowledged a previous +add_addr_timeout_max - INTEGER (seconds) + Set the maximum value of timeout after which an ADD_ADDR control message + will be resent to an MPTCP peer that has not acknowledged a previous ADD_ADDR message. The default value matches TCP_RTO_MAX. This is a per-namespace
net/mptcp/ctrl.c +6 -6
@@ -27,7 +27,7 @@ struct mptcp_pernet {struct ctl_table_header *ctl_table_hdr;#endif -unsigned int add_addr_timeout; +unsigned int add_addr_timeout_max;unsigned int blackhole_timeout;unsigned int close_timeout;unsigned int stale_loss_cnt;
@@ -52,9 +52,9 @@ int mptcp_is_enabled(const struct net *net)return mptcp_get_pernet (net)->mptcp_enabled; } -unsigned int mptcp_get_add_addr_timeout (const struct net *net) +unsigned int mptcp_get_add_addr_timeout_max (const struct net *net) { -return mptcp_get_pernet (net)->add_addr_timeout; +return mptcp_get_pernet (net)->add_addr_timeout_max; }int mptcp_is_checksum_enabled (const struct net *net)
@@ -97,7 +97,7 @@ const char *mptcp_get_scheduler(const struct net *net)static void mptcp_pernet_set_defaults (struct mptcp_pernet *pernet) { pernet->mptcp_enabled =1 ; - pernet->add_addr_timeout =TCP_RTO_MAX ; + pernet->add_addr_timeout_max =TCP_RTO_MAX ; pernet->blackhole_timeout =3600 ; pernet->syn_retrans_before_tcp_fallback =2 ;atomic_set (&pernet->active_disable_times,0 );
@@ -284,7 +284,7 @@ static struct ctl_table mptcp_sysctl_table[] = { .extra2 = SYSCTL_ONE }, { - .procname = "add_addr_timeout", + .procname = "add_addr_timeout_max", .maxlen = sizeof(unsigned int), .mode = 0644, .proc_handler = proc_dointvec_jiffies,
@@ -377,7 +377,7 @@ static int mptcp_pernet_new_table(struct net *net, struct mptcp_pernet *pernet) } table[0 ].data = &pernet->mptcp_enabled; - table[1 ].data = &pernet->add_addr_timeout; + table[1 ].data = &pernet->add_addr_timeout_max; table[2 ].data = &pernet->checksum_enabled; table[3 ].data = &pernet->allow_join_initial_addr_port; table[4 ].data = &pernet->stale_loss_cnt;
net/mptcp/pm.c +3 -3
@@ -302,7 +302,7 @@ static void mptcp_pm_add_timer(struct timer_list *timer)if (entry->retrans_times <ADD_ADDR_RETRANS_MAX )sk_reset_timer (sk, timer, - jiffies +mptcp_get_add_addr_timeout (sock_net (sk))); + jiffies +mptcp_get_add_addr_timeout_max (sock_net (sk)));spin_unlock_bh (&msk->pm.lock);
@@ -354,7 +354,7 @@ bool mptcp_pm_alloc_anno_list(struct mptcp_sock *msk,return false;sk_reset_timer (sk, &add_entry->add_timer, - jiffies +mptcp_get_add_addr_timeout (net)); + jiffies +mptcp_get_add_addr_timeout_max (net));return true; }
@@ -370,7 +370,7 @@ bool mptcp_pm_alloc_anno_list(struct mptcp_sock *msk,timer_setup (&add_entry->add_timer, mptcp_pm_add_timer,0 );sk_reset_timer (sk, &add_entry->add_timer, - jiffies +mptcp_get_add_addr_timeout (net)); + jiffies +mptcp_get_add_addr_timeout_max (net));return true; }
net/mptcp/protocol.h +1 -1
@@ -708,7 +708,7 @@ mptcp_subflow_delegated_next(struct mptcp_delegated_action *delegated) }int mptcp_is_enabled (const struct net *net); -unsigned int mptcp_get_add_addr_timeout (const struct net *net); +unsigned int mptcp_get_add_addr_timeout_max (const struct net *net);int mptcp_is_checksum_enabled (const struct net *net);int mptcp_allow_join_id0 (const struct net *net);unsigned int mptcp_stale_loss_cnt (const struct net *net);
Signature · 1 line
2.48.1
Matthieu Baerts
maintainerRe: [PATCH mptcp-next 1/3] mptcp: rename add_addr_timeout to add_addr_timeout_max
Hi Geliang,
Quoted 4 lines · Geliang Tang
From: Geliang Tang <[email protected]>
This renames the MPTCP sysctl variable to better reflect its purpose as the maximum
ADD_ADDRretransmission timeout.
I don't think we can rename a sysctl knob: it is sort of part of the API as it is exposed. Either we deprecate it with a warning in the doc + pr_warn() somewhere, or use it as a max value without renaming it (at least not renaming the exposed variable: .procname).
I think it should be fine to use it as a max value because it was set to TCP_RTO_MAX by default, and this timeout is in second. If you do that, you can probably squash patch 1 and 2.
Cheers,
Matt
Signature · 1 line
Sponsored by the NGI0 Core fund.
Geliang Tang
reviewerPATCH 2/3 mptcp: make ADD_ADDR timeout adaptive to subflow RTT
In reply to Geliang TangFrom: Geliang Tang <[email protected]>
This patch makes the ADD_ADDR retransmission timeout adaptive by using the maximum subflow RTT, while still capping it at the configured max value (add_addr_timeout_max). This improves responsiveness when establishing new subflows.
The change:
- Adds mptcp_get_add_addr_timeout() helper
- Uses subflow RTT when available
- Falls back to max timeout otherwise
--- net/mptcp/pm.c | 27 ++++++++++++++++++++++++--- 1 file changed, 24 insertions(+), 3 deletions(-)
net/mptcp/pm.c +24 -3
@@ -268,6 +268,27 @@ int mptcp_pm_mp_prio_send_ack(struct mptcp_sock *msk,return -EINVAL ; } +static unsigned int mptcp_get_add_addr_timeout (struct mptcp_sock *msk, +const struct net *net) +{ +unsigned int timeout =mptcp_get_add_addr_timeout_max (net); +struct mptcp_subflow_context *subflow; +unsigned int srtt_us =0 ; + +mptcp_for_each_subflow (msk, subflow) { +struct sock *ssk =mptcp_subflow_tcp_sock (subflow); +struct tcp_sock *tp =tcp_sk (ssk); + +if (srtt_us < tp->srtt_us) + srtt_us = tp->srtt_us; + } + +if (srtt_us && srtt_us < timeout) + timeout = srtt_us; + +return timeout; +} +static void mptcp_pm_add_timer (struct timer_list *timer) { struct mptcp_pm_add_entry *entry = timer_container_of(entry, timer,
@@ -302,7 +323,7 @@ static void mptcp_pm_add_timer(struct timer_list *timer)if (entry->retrans_times <ADD_ADDR_RETRANS_MAX )sk_reset_timer (sk, timer, - jiffies +mptcp_get_add_addr_timeout_max (sock_net (sk))); + jiffies +mptcp_get_add_addr_timeout (msk,sock_net (sk)));spin_unlock_bh (&msk->pm.lock);
@@ -354,7 +375,7 @@ bool mptcp_pm_alloc_anno_list(struct mptcp_sock *msk,return false;sk_reset_timer (sk, &add_entry->add_timer, - jiffies +mptcp_get_add_addr_timeout_max (net)); + jiffies +mptcp_get_add_addr_timeout (msk, net));return true; }
@@ -370,7 +391,7 @@ bool mptcp_pm_alloc_anno_list(struct mptcp_sock *msk,timer_setup (&add_entry->add_timer, mptcp_pm_add_timer,0 );sk_reset_timer (sk, &add_entry->add_timer, - jiffies +mptcp_get_add_addr_timeout_max (net)); + jiffies +mptcp_get_add_addr_timeout (msk, net));return true; }
Signature · 1 line
2.48.1
Matthieu Baerts
maintainerRe: [PATCH mptcp-next 2/3] mptcp: make ADD_ADDR timeout adaptive to subflow RTT
Hi Geliang,
Quoted 7 lines · Geliang Tang
From: Geliang Tang <[email protected]>
This patch makes the
ADD_ADDRretransmission timeout adaptive by using the maximum subflow RTT, while still capping it at the configured max value (add_addr_timeout_max). This improves responsiveness when establishing new subflows.
Quoted 5 lines
The change:
- Addsmptcp_get_add_addr_timeout()helper
- Uses subflow RTT when available
- Falls back to max timeout otherwise
Can you add the Closes tag here, so GitHub will automatically close the ticket when this patch will be applied?
Quoted 15 lines
Signed-off-by Geliang Tang <[email protected]>--- net/mptcp/pm.c | 27 ++++++++++++++++++++++++--- 1 file changed, 24 insertions(+), 3 deletions(-)net/mptcp/pm.c +2
-0@@ -268,6 +268,27 @@ int mptcp_pm_mp_prio_send_ack(struct mptcp_sock *msk,return -EINVAL ; } +static unsigned int mptcp_get_add_addr_timeout (struct mptcp_sock *msk, +const struct net *net)
net/mptcp/pm.c int mptcp_pm_mp_prio_send_ack(struct mptcp_sock *msk, · line 282
Quoted 10 lines
net/mptcp/pm.c +10
-0+{ +unsigned int timeout =mptcp_get_add_addr_timeout_max (net); +struct mptcp_subflow_context *subflow; +unsigned int srtt_us =0 ; + +mptcp_for_each_subflow (msk, subflow) { +struct sock *ssk =mptcp_subflow_tcp_sock (subflow); +struct tcp_sock *tp =tcp_sk (ssk); + +if (srtt_us < tp->srtt_us)
(detail: it feels more logical do have "if (tp->srtt_us > srtt_us)" if you look for the max, but the result is the same)
net/mptcp/pm.c int mptcp_pm_mp_prio_send_ack(struct mptcp_sock *msk, · line 286
Quoted 4 lines
net/mptcp/pm.c +4
-0+ srtt_us = tp->srtt_us; + } + +if (srtt_us && srtt_us < timeout)
I don't think you can do that: timeout is in jiffies, while srtt_us is in µsec. You need to use "usecs_to_jiffies(srtt_us)"
net/mptcp/pm.c int mptcp_pm_mp_prio_send_ack(struct mptcp_sock *msk, · line 287
Quoted 1 line
net/mptcp/pm.c +1
-0+ timeout = srtt_us;
Taking just srtt_us doesn't seem to be a good idea as the ADD_ADDR echo is supposed to arrive around srtt_us: this will certainly cause too aggressive ADD_ADDR retransmissions if timeout is around the time the echo is supposed to arrive.
I think we should imitate TCP here: using srtt_us and rttvar_us. (Or start with "srtt_us << 1"?). Or maybe better to use "icsk->icsk_rto" instead, but I don't know if it will be set to a correct value when we will look, no?
net/mptcp/pm.c static void mptcp_pm_add_timer(struct timer_list *timer) · line 326
Quoted 13 lines
net/mptcp/pm.c +5
-1+ +return timeout; +} +static void mptcp_pm_add_timer (struct timer_list *timer) { struct mptcp_pm_add_entry *entry = timer_container_of(entry, timer,@@ -302,7 +323,7 @@ static void mptcp_pm_add_timer(struct timer_list *timer)if (entry->retrans_times <ADD_ADDR_RETRANS_MAX )sk_reset_timer (sk, timer, - jiffies +mptcp_get_add_addr_timeout_max (sock_net (sk))); + jiffies +mptcp_get_add_addr_timeout (msk,sock_net (sk)));
I wonder if we should increase the time after each retransmission, similar to what is done in TCP (timeout << entry->retrans_times). We could keep it linear, but because this time now depends on the measured srtt, it might be better to have an exponential backoff. It will still need to be bounded to the max. In this case, it might be easier to pass entry->retrans_times to mptcp_get_add_addr_timeout())
Quoted 13 lines
net/mptcp/pm.c +1
-1spin_unlock_bh (&msk->pm.lock);@@ -354,7 +375,7 @@ bool mptcp_pm_alloc_anno_list(struct mptcp_sock *msk,return false;sk_reset_timer (sk, &add_entry->add_timer, - jiffies +mptcp_get_add_addr_timeout_max (net)); + jiffies +mptcp_get_add_addr_timeout (msk, net));return true; }
net/mptcp/pm.c bool mptcp_pm_alloc_anno_list(struct mptcp_sock *msk, · line 397
Quoted 8 lines
net/mptcp/pm.c +1
-1@@ -370,7 +391,7 @@ bool mptcp_pm_alloc_anno_list(struct mptcp_sock *msk,timer_setup (&add_entry->add_timer, mptcp_pm_add_timer,0 );sk_reset_timer (sk, &add_entry->add_timer, - jiffies +mptcp_get_add_addr_timeout_max (net)); + jiffies +mptcp_get_add_addr_timeout (msk, net));return true; }
Cheers,
Matt
Signature · 1 line
Sponsored by the NGI0 Core fund.
Geliang Tang
reviewerPATCH 3/3 selftests: mptcp: update for add_addr_timeout_max rename
In reply to Geliang TangFrom: Geliang Tang <[email protected]>
Update MPTCP selftests to use the renamed sysctl variable add_addr_timeout_max. This aligns tests with the kernel change.
--- tools/testing/selftests/net/mptcp/mptcp_join.sh | 8 ++++---- 1 file changed, 4 insertions(+), 4 deletions(-)
tools/testing/selftests/net/mptcp/mptcp_join.sh +4 -4
@@ -347,7 +347,7 @@ reset_with_add_addr_timeout() tables="${ip6tables}" fi - ip netns exec $ns1 sysctl -q net.mptcp.add_addr_timeout=1 + ip netns exec $ns1 sysctl -q net.mptcp.add_addr_timeout_max=1 if ! ip netns exec $ns2 $tables -A OUTPUT -p tcp \ -m tcp --tcp-option 30 \
@@ -1556,7 +1556,7 @@ chk_add_nr() rx=" server" fi - timeout=$(ip netns exec ${ns_tx} sysctl -n net.mptcp.add_addr_timeout) + timeout=$(ip netns exec ${ns_tx} sysctl -n net.mptcp.add_addr_timeout_max) print_check "add addr rx${rx}" count=$(mptcp_lib_get_counter ${ns_rx} "MPTcpExtAddAddr")
@@ -1655,7 +1655,7 @@ chk_add_tx_nr() local timeout local count - timeout=$(ip netns exec $ns1 sysctl -n net.mptcp.add_addr_timeout) + timeout=$(ip netns exec $ns1 sysctl -n net.mptcp.add_addr_timeout_max) print_check "add addr tx" count=$(mptcp_lib_get_counter ${ns1} "MPTcpExtAddAddrTx")
@@ -2183,7 +2183,7 @@ signal_address_tests() pm_nl_add_endpoint $ns2 10.0.4.2 flags signal # the peer could possibly miss some addr notification, allow retransmission - ip netns exec $ns1 sysctl -q net.mptcp.add_addr_timeout=1 + ip netns exec $ns1 sysctl -q net.mptcp.add_addr_timeout_max=1 speed=slow \ run_tests $ns1 $ns2 10.0.1.1
Signature · 1 line
2.48.1
Matthieu Baerts
maintainerRe: [PATCH mptcp-next 3/3] selftests: mptcp: update for add_addr_timeout_max rename
Hi Geliang,
Quoted 4 lines · Geliang Tang
From: Geliang Tang <[email protected]>
Update MPTCP selftests to use the renamed sysctl variable
add_addr_timeout_max. This aligns tests with the kernel change.
This should not be needed, see my comment on patch 1/3.
What you will probably need to adapt is the packetdrill tests dedicated to the ADD_ADDR: mptcp/add_addr. These tests expect a retransmission every second, because net.mptcp.add_addr_timeout sysctl knob is set to 1.
Cheers,
Matt
Signature · 1 line
Sponsored by the NGI0 Core fund.
Hi Geliang,
Thank you for your modifications, that's great!
Our CI did some validations and here is its report:
- KVM Validation: normal: Unstable: 2 failed test(s): packetdrill_add_addr selftest_mptcp_connect 🔴
- KVM Validation: debug: Unstable: 3 failed test(s): packetdrill_add_addr packetdrill_sockopts selftest_mptcp_connect_checksum 🔴
- KVM Validation: btf-normal (only bpftest_all): Success! ✅
- KVM Validation: btf-debug (only bpftest_all): Success! ✅
- Task: https://github.com/multipath-tcp/mptcp_net-next/actions/runs/16591166379
Initiator: Patchew Applier
Commits: https://github.com/multipath-tcp/mptcp_net-next/commits/4e382e2fda74
Patchwork: https://patchwork.kernel.org/project/mptcp/list/?series=986638
If there are some issues, you can reproduce them using the same environment as the one used by the CI thanks to a docker image, e.g.:
$ cd [kernel source code]
$ docker run -v "${PWD}:${PWD}:rw" -w "${PWD}" --privileged --rm -it \
--pull always mptcp/mptcp-upstream-virtme-docker:latest \
auto-normalFor more details:
https://github.com/multipath-tcp/mptcp-upstream-virtme-docker
Please note that despite all the efforts that have been already done to have a stable tests suite when executed on a public CI like here, it is possible some reported issues are not due to your modifications. Still, do not hesitate to help us improve that ;-)
Cheers,
MPTCP GH Action bot
Bot operated by Matthieu Baerts (NGI0 Core)
Matthieu Baerts
maintainerHi Geliang,
Quoted 11 lines · Geliang Tang
From: Geliang Tang <[email protected]>
This series optimizes MPTCP's
ADD_ADDRretransmission timeout:1. Renames sysctl to
add_addr_timeout_maxfor clearer semantics
2. Makes timeout adaptive to subflow RTT
3. Updates selftests for the renameThe adaptive timeout improves subflow establishment responsiveness while maintaining reliability through the max timeout cap. The rename provides better configuration clarity.
Thank you for working on that. I have some comments, please see the individual patches.
Quoted 1 line
Better to (also) put that in the last patch closing the ticket, so GitHub can automatically close it when this patch will be applied.
Quoted 5 lines
Geliang Tang (3):
mptcp: renameadd_addr_timeouttoadd_addr_timeout_max
mptcp: makeADD_ADDRtimeout adaptive to subflow RTT
selftests: mptcp: update foradd_addr_timeout_maxrename
Quoted 7 lines
Documentation/networking/mptcp-sysctl.rst | 6 ++--- net/mptcp/ctrl.c | 12 ++++----- net/mptcp/pm.c | 27 ++++++++++++++++--- net/mptcp/protocol.h | 2 +- .../testing/selftests/net/mptcp/mptcp_join.sh | 8 +++--- 5 files changed, 38 insertions(+), 17 deletions(-)
Cheers,
Matt
Signature · 1 line
Sponsored by the NGI0 Core fund.