refactor push pending
Subsystems (2)
v9 cover refactor push pending
v9:
- move "register default scheduler" out of this series.
- add missing "first = false" in patch 4
- patch 2 could be merged before the commit "mptcp: add struct mptcp_sched_ops" as a non-bpf cleanup.
v8:
- merge "register default scheduler" into this series.
- Move "multi-subflow scheduling" out of this series.
- This series should be merged between the commit "mptcp: add get_subflow wrappers" and "bpf: Add bpf_mptcp_sched_ops", except the squash-to patches.
v7:
- update delegate sending in patch 10
- rebase to export/20221011T063543.
v6:
- drop all msk->last_snd, add last_snd variable instead in patch 13
- fix lock_sock issue in patch 3
- merge squash-to patches
v5:
- address Mat's comments in v4.
v4:
- update __mptcp_subflow_push_pending as Mat suggested.
- add more patches from "BPF redundant scheduler" series.
v3:
- add a cleanup patch.
- remove msk->last_snd in mptcp_subflow_get_send().
- add the loop that calls the scheduler again in __mptcp_push_pending().
v2:
- add snd_burst check in dfrags loop as Mat suggested.
Refactor __mptcp_push_pending() and __mptcp_subflow_push_pending() to remove duplicate code and support redundant scheduler more easily in __mptcp_subflow_push_pending().
Geliang Tang (4):
Squash to "mptcp: add get_subflow wrappers"
mptcp: use msk instead of mptcp_sk
mptcp: change 'first' as a parameter
mptcp: refactor push_pending logic
net/mptcp/protocol.c | 183 +++++++++++++++++++------------------------ net/mptcp/protocol.h | 4 +- net/mptcp/sched.c | 61 +++++++-------- 3 files changed, 111 insertions(+), 137 deletions(-)
Signature · 1 line
2.35.3
PATCH v9 1/4 Squash to "mptcp: add get_subflow wrappers"
Please update the commit log:
'''
This patch defines two new wrappers mptcp_sched_get_send() and mptcp_sched_get_retrans(), invoke get_subflow() of msk->sched in them.
Set the subflow pointers array in struct mptcp_sched_data before invoking get_subflow(), then it can be used in get_subflow() in the BPF contexts.
Check the subflow scheduled flags to test which subflow or subflows are picked by the scheduler.
Move sock_owned_by_me() and the fallback check code from mptcp_subflow_get_send/retrans() into the wrappers. '''
--- net/mptcp/protocol.c | 8 +++--- net/mptcp/protocol.h | 4 +-- net/mptcp/sched.c | 61 +++++++++++++++++++++----------------------- 3 files changed, 35 insertions(+), 38 deletions(-)
net/mptcp/protocol.c +4 -4
@@ -1547,7 +1547,7 @@ void __mptcp_push_pending(struct sock *sk, unsigned int flags)int ret =0 ; prev_ssk = ssk; - ssk =mptcp_sched_get_send (msk); + ssk =mptcp_subflow_get_send (msk);/* First check. If the ssk has changed since * the last round, release prev_ssk
@@ -1616,7 +1616,7 @@ static void __mptcp_subflow_push_pending(struct sock *sk, struct sock *ssk)* check for a different subflow usage only after * spooling the first chunk of data */ - xmit_ssk = first ? ssk :mptcp_sched_get_send (mptcp_sk (sk)); + xmit_ssk = first ? ssk :mptcp_subflow_get_send (mptcp_sk (sk));if (!xmit_ssk) goto out;if (xmit_ssk != ssk) {
@@ -2500,7 +2500,7 @@ static void __mptcp_retrans(struct sock *sk)mptcp_clean_una_wakeup (sk);/* first check ssk: need to kick "stale" logic */ - ssk =mptcp_sched_get_retrans (msk); + ssk =mptcp_subflow_get_retrans (msk); dfrag =mptcp_rtx_head (sk);if (!dfrag) {if (mptcp_data_fin_enabled (msk)) {
@@ -3218,7 +3218,7 @@ void __mptcp_check_push(struct sock *sk, struct sock *ssk)return ;if (!sock_owned_by_user (sk)) { -struct sock *xmit_ssk =mptcp_sched_get_send (mptcp_sk (sk)); +struct sock *xmit_ssk =mptcp_subflow_get_send (mptcp_sk (sk));if (xmit_ssk == ssk)__mptcp_subflow_push_pending (sk, ssk);
net/mptcp/protocol.h +2 -2
@@ -643,8 +643,8 @@ void mptcp_subflow_set_scheduled(struct mptcp_subflow_context *subflow, bool scheduled);struct sock *mptcp_subflow_get_send (struct mptcp_sock *msk);struct sock *mptcp_subflow_get_retrans (struct mptcp_sock *msk); -struct sock *mptcp_sched_get_send (struct mptcp_sock *msk); -struct sock *mptcp_sched_get_retrans (struct mptcp_sock *msk); +int mptcp_sched_get_send (struct mptcp_sock *msk); +int mptcp_sched_get_retrans (struct mptcp_sock *msk);static inline bool __tcp_can_send (const struct sock *ssk) {
net/mptcp/sched.c +29 -32
96 lines not on this page · raw message
Signature · 1 line
2.35.3
PATCH v9 2/4 mptcp: use msk instead of mptcp_sk
In reply to Geliang TangUse msk instead of mptcp_sk(sk) in the functions where the variable "msk = mptcp_sk(sk)" has been defined.
--- net/mptcp/protocol.c | 10 +++++----- 1 file changed, 5 insertions(+), 5 deletions(-)
net/mptcp/protocol.c +5 -5
@@ -1616,7 +1616,7 @@ static void __mptcp_subflow_push_pending(struct sock *sk, struct sock *ssk)* check for a different subflow usage only after * spooling the first chunk of data */ - xmit_ssk = first ? ssk :mptcp_subflow_get_send (mptcp_sk (sk)); + xmit_ssk = first ? ssk :mptcp_subflow_get_send (msk);if (!xmit_ssk) goto out;if (xmit_ssk != ssk) {
@@ -2261,7 +2261,7 @@ bool __mptcp_retransmit_pending_data(struct sock *sk)struct mptcp_data_frag *cur, *rtx_head;struct mptcp_sock *msk =mptcp_sk (sk); -if (__mptcp_check_fallback (mptcp_sk (sk))) +if (__mptcp_check_fallback (msk))return false;if (tcp_rtx_and_write_queues_empty (sk))
@@ -2943,7 +2943,7 @@ bool __mptcp_close(struct sock *sk, long timeout)sock_hold (sk);pr_debug ("msk=%p state=%d" , sk, sk->sk_state); -if (mptcp_sk (sk)->token) +if (msk->token)mptcp_event (MPTCP_EVENT_CLOSED , msk,NULL ,GFP_KERNEL );if (sk->sk_state ==TCP_CLOSE ) {
@@ -3002,8 +3002,8 @@ static int mptcp_disconnect(struct sock *sk, int flags)mptcp_stop_timer (sk);sk_stop_timer (sk, &sk->sk_timer); -if (mptcp_sk (sk)->token) -mptcp_event (MPTCP_EVENT_CLOSED ,mptcp_sk (sk),NULL ,GFP_KERNEL ); +if (msk->token) +mptcp_event (MPTCP_EVENT_CLOSED , msk,NULL ,GFP_KERNEL );/* msk->subflow is still intact, the following will not free the first * subflow
Signature · 1 line
2.35.3
PATCH v9 3/4 mptcp: change 'first' as a parameter
In reply to Geliang TangThe function mptcp_subflow_process_delegated() uses the input ssk first, while __mptcp_check_push() invokes the packet scheduler first.
So this patch adds a new parameter named 'first' for the function __mptcp_subflow_push_pending() to deal with these two cases separately.
With this change, the code that invokes the packet scheduler in the function __mptcp_check_push() can be removed, and replaced by invoking __mptcp_subflow_push_pending() directly.
--- net/mptcp/protocol.c | 20 ++++++-------------- 1 file changed, 6 insertions(+), 14 deletions(-)
net/mptcp/protocol.c +6 -14
@@ -1593,7 +1593,7 @@ void __mptcp_push_pending(struct sock *sk, unsigned int flags)__mptcp_check_send_data_fin (sk); } -static void __mptcp_subflow_push_pending (struct sock *sk,struct sock *ssk) +static void __mptcp_subflow_push_pending (struct sock *sk,struct sock *ssk,bool first) {struct mptcp_sock *msk =mptcp_sk (sk);struct mptcp_sendmsg_info info = {
@@ -1602,7 +1602,6 @@ static void __mptcp_subflow_push_pending(struct sock *sk, struct sock *ssk)struct mptcp_data_frag *dfrag;struct sock *xmit_ssk;int len, copied =0 ; -bool first = true; info.flags =0 ;while ((dfrag =mptcp_send_head (sk))) {
@@ -1612,8 +1611,7 @@ static void __mptcp_subflow_push_pending(struct sock *sk, struct sock *ssk)while (len >0 ) {int ret =0 ; -/* the caller already invoked the packet scheduler, -* check for a different subflow usage only after +/* check for a different subflow usage only after * spooling the first chunk of data */ xmit_ssk = first ? ssk :mptcp_subflow_get_send (msk);
@@ -3217,16 +3215,10 @@ void __mptcp_check_push(struct sock *sk, struct sock *ssk)if (!mptcp_send_head (sk))return ; -if (!sock_owned_by_user (sk)) { -struct sock *xmit_ssk =mptcp_subflow_get_send (mptcp_sk (sk)); - -if (xmit_ssk == ssk) -__mptcp_subflow_push_pending (sk, ssk); -else if (xmit_ssk) -mptcp_subflow_delegate (mptcp_subflow_ctx (xmit_ssk),MPTCP_DELEGATE_SEND ); - }else { +if (!sock_owned_by_user (sk)) +__mptcp_subflow_push_pending (sk, ssk, false); +else __set_bit (MPTCP_PUSH_PENDING , &mptcp_sk (sk)->cb_flags); - } }#define MPTCP_FLAGS_PROCESS_CTX_NEED (BIT(MPTCP_PUSH_PENDING) | \
@@ -3317,7 +3309,7 @@ void mptcp_subflow_process_delegated(struct sock *ssk)if (test_bit (MPTCP_DELEGATE_SEND , &subflow->delegated_status)) {mptcp_data_lock (sk);if (!sock_owned_by_user (sk)) -__mptcp_subflow_push_pending (sk, ssk); +__mptcp_subflow_push_pending (sk, ssk, true);else __set_bit (MPTCP_PUSH_PENDING , &mptcp_sk (sk)->cb_flags);mptcp_data_unlock (sk);
Signature · 1 line
2.35.3
PATCH v9 4/4 mptcp: refactor push_pending logic
In reply to Geliang TangTo support redundant package schedulers more easily, this patch refactors __mptcp_push_pending() logic from:
For each dfrag:
While sends succeed:
Call the scheduler (selects subflow and msk->snd_burst) Update subflow locks (push/release/acquire as needed) Send the dfrag data with mptcp_sendmsg_frag() Update already_sent, snd_nxt, snd_burst
Update msk->first_pending
Push/release on final subflow
->
While the scheduler selects one subflow:
Lock the subflow
For each pending dfrag:
While sends succeed:
Send the dfrag data with mptcp_sendmsg_frag() Update already_sent, snd_nxt, snd_burst
Update msk->first_pending
Break if required by msk->snd_burst / etc
Push and release the subflow
Refactors __mptcp_subflow_push_pending logic from:
For each dfrag:
While sends succeed: Call the scheduler (selects subflow and msk->snd_burst) Send the dfrag data with mptcp_subflow_delegate(), break Send the dfrag data with mptcp_sendmsg_frag() Update dfrag->already_sent, msk->snd_nxt, msk->snd_burst Update msk->first_pending
->
While first_pending isn't empty:
Call the scheduler (selects subflow and msk->snd_burst) Send the dfrag data with mptcp_subflow_delegate(), break Send the dfrag data with mptcp_sendmsg_frag()
For each pending dfrag:
While sends succeed:
Send the dfrag data with mptcp_sendmsg_frag() Update already_sent, snd_nxt, snd_burst
Update msk->first_pending
Break if required by msk->snd_burst / etc
Move the duplicate code from __mptcp_push_pending() and __mptcp_subflow_push_pending() into a new helper function, named __subflow_push_pending(). Simplify __mptcp_push_pending() and __mptcp_subflow_push_pending() by invoking this helper.
Also move the burst check conditions out of the function mptcp_subflow_get_send(), check them in __mptcp_push_pending() and __mptcp_subflow_push_pending() in the inner "for each pending dfrag" loop.
--- net/mptcp/protocol.c | 155 +++++++++++++++++++------------------------ 1 file changed, 70 insertions(+), 85 deletions(-)
net/mptcp/protocol.c +70 -85
@@ -1417,14 +1417,6 @@ struct sock *mptcp_subflow_get_send(struct mptcp_sock *msk)u64 linger_time;long tout =0 ; -/* re-use last subflow, if the burst allow that */ -if (msk->last_snd && msk->snd_burst >0 && -sk_stream_memory_free (msk->last_snd) && -mptcp_subflow_active (mptcp_subflow_ctx (msk->last_snd))) { -mptcp_set_timeout (sk); -return msk->last_snd; - } -/* pick the subflow with the lower wmem/wspace ratio */ for (i =0 ; i <SSK_MODE_MAX ; ++i) { send_info[i].ssk =NULL ;
@@ -1491,12 +1483,6 @@ struct sock *mptcp_subflow_get_send(struct mptcp_sock *msk)return ssk; } -static void mptcp_push_release (struct sock *ssk,struct mptcp_sendmsg_info *info) -{ -tcp_push (ssk,0 , info->mss_now,tcp_sk (ssk)->nonagle, info->size_goal); -release_sock (ssk); -} -static void mptcp_update_post_push (struct mptcp_sock *msk,struct mptcp_data_frag *dfrag,u32 sent)
@@ -1528,68 +1514,80 @@ void mptcp_check_and_set_pending(struct sock *sk)mptcp_sk (sk)->push_pending |=BIT (MPTCP_PUSH_PENDING ); } -void __mptcp_push_pending (struct sock *sk,unsigned int flags) +static int __subflow_push_pending (struct sock *sk,struct sock *ssk, +struct mptcp_sendmsg_info *info) { -struct sock *prev_ssk =NULL , *ssk =NULL ;struct mptcp_sock *msk =mptcp_sk (sk); -struct mptcp_sendmsg_info info = { - .flags = flags, - }; -bool do_check_data_fin = false;struct mptcp_data_frag *dfrag; -int len; +int len, copied =0 , err =0 ;while ((dfrag =mptcp_send_head (sk))) { - info.sent = dfrag->already_sent; - info.limit = dfrag->data_len; + info->sent = dfrag->already_sent; + info->limit = dfrag->data_len; len = dfrag->data_len - dfrag->already_sent;while (len >0 ) {int ret =0 ; - prev_ssk = ssk; - ssk =mptcp_subflow_get_send (msk); - -/* First check. If the ssk has changed since -* the last round, release prev_ssk -*/ -if (ssk != prev_ssk && prev_ssk) -mptcp_push_release (prev_ssk, &info); -if (!ssk) - goto out; - -/* Need to lock the new subflow only if different -* from the previous one, otherwise we are still -* helding the relevant lock -*/ -if (ssk != prev_ssk) -lock_sock (ssk); - - ret =mptcp_sendmsg_frag (sk, ssk, dfrag, &info); + ret =mptcp_sendmsg_frag (sk, ssk, dfrag, info);if (ret <=0 ) { -if (ret == -EAGAIN ) -continue ; -mptcp_push_release (ssk, &info); + err = copied ? : ret; goto out; } - do_check_data_fin = true; - info.sent += ret; + info->sent += ret; + copied += ret; len -= ret;mptcp_update_post_push (msk, dfrag, ret); }WRITE_ONCE (msk->first_pending,mptcp_send_next (sk)); + +if (msk->snd_burst <=0 || + !sk_stream_memory_free (ssk) || + !mptcp_subflow_active (mptcp_subflow_ctx (ssk))) { + err = copied ? : -EAGAIN ; + goto out; + } +mptcp_set_timeout (sk); + } + err = copied; + +out: +if (copied) { +tcp_push (ssk,0 , info->mss_now,tcp_sk (ssk)->nonagle, + info->size_goal); } -/* at this point we held the socket lock for the last subflow we used */ -if (ssk) -mptcp_push_release (ssk, &info); +return err; +} + +void __mptcp_push_pending (struct sock *sk,unsigned int flags) +{ +struct mptcp_sock *msk =mptcp_sk (sk); +struct mptcp_sendmsg_info info = { + .flags = flags, + }; +struct sock *ssk; +int ret =0 ; + +again: +while (mptcp_send_head (sk) && (ssk =mptcp_subflow_get_send (msk))) { +lock_sock (ssk); + ret =__subflow_push_pending (sk, ssk, &info); +release_sock (ssk); + +if (ret <=0 ) { +if (ret == -EAGAIN ) + goto again; + goto out; + } + } out:/* ensure the rtx timer is running */ if (!mptcp_timer_pending (sk))mptcp_reset_timer (sk); -if (do_check_data_fin) +if (ret >0 )__mptcp_check_send_data_fin (sk); }
@@ -1599,51 +1597,38 @@ static void __mptcp_subflow_push_pending(struct sock *sk, struct sock *ssk, boolstruct mptcp_sendmsg_info info = { .data_lock_held = true, }; -struct mptcp_data_frag *dfrag;struct sock *xmit_ssk; -int len, copied =0 ; +int ret =0 ; info.flags =0 ; -while ((dfrag =mptcp_send_head (sk))) { - info.sent = dfrag->already_sent; - info.limit = dfrag->data_len; - len = dfrag->data_len - dfrag->already_sent; -while (len >0 ) { -int ret =0 ; - -/* check for a different subflow usage only after -* spooling the first chunk of data -*/ - xmit_ssk = first ? ssk :mptcp_subflow_get_send (msk); -if (!xmit_ssk) - goto out; -if (xmit_ssk != ssk) { -mptcp_subflow_delegate (mptcp_subflow_ctx (xmit_ssk), -MPTCP_DELEGATE_SEND ); - goto out; - } - - ret =mptcp_sendmsg_frag (sk, ssk, dfrag, &info); -if (ret <=0 ) - goto out; - - info.sent += ret; - copied += ret; - len -= ret; - first = false; +again: +while (mptcp_send_head (sk)) { +/* check for a different subflow usage only after +* spooling the first chunk of data +*/ + xmit_ssk = first ? ssk :mptcp_subflow_get_send (msk); +if (!xmit_ssk) + goto out; +if (xmit_ssk != ssk) { +mptcp_subflow_delegate (mptcp_subflow_ctx (xmit_ssk), +MPTCP_DELEGATE_SEND ); + goto out; + } -mptcp_update_post_push (msk, dfrag, ret); + ret =__subflow_push_pending (sk, ssk, &info); + first = false; +if (ret <=0 ) { +if (ret == -EAGAIN ) + goto again; +break ; } -WRITE_ONCE (msk->first_pending,mptcp_send_next (sk)); } out:/* __mptcp_alloc_tx_skb could have released some wmem and we are * not going to flush it via release_sock() */ -if (copied) { -tcp_push (ssk,0 , info.mss_now,tcp_sk (ssk)->nonagle, - info.size_goal); +if (ret >0 ) {if (!mptcp_timer_pending (sk))mptcp_reset_timer (sk);
Signature · 1 line
2.35.3
Re: mptcp: refactor push_pending logic: Tests Results
Hi Geliang,
Thank you for your modifications, that's great!
Our CI did some validations and here is its report:
- KVM Validation: normal:
- Unstable: 1 failed test(s): selftest_simult_flows 🔴:
- Task: https://cirrus-ci.com/task/5202885733515264
- Summary: https://api.cirrus-ci.com/v1/artifact/task/5202885733515264/summary/summary.txt
- {"code":404,"message":
- "Can't find artifacts containing file conclusion.txt"}:
- Task: https://cirrus-ci.com/task/6328785640357888
- Summary: https://api.cirrus-ci.com/v1/artifact/task/6328785640357888/summary/summary.txt
Initiator: Patchew Applier
Commits: https://github.com/multipath-tcp/mptcp_net-next/commits/ddd3c597286e
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-debugFor 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 (Tessares)
Re: [PATCH mptcp-next v9 4/4] mptcp: refactor push_pending logic
In reply to Geliang TangQuoted 185 lines · Geliang Tang
To support redundant package schedulers more easily, this patch refactors
__mptcp_push_pending()logic from:For each dfrag:
While sends succeed:
Call the scheduler (selects subflow andmsk->snd_burst) Update subflow locks (push/release/acquire as needed) Send the dfrag data withmptcp_sendmsg_frag()Updatealready_sent,snd_nxt,snd_burst
Updatemsk->first_pending
Push/release on final subflow->
While the scheduler selects one subflow:
Lock the subflow
For each pending dfrag:
While sends succeed:
Send the dfrag data withmptcp_sendmsg_frag()Updatealready_sent,snd_nxt,snd_burst
Updatemsk->first_pending
Break if required bymsk->snd_burst/ etc
Push and release the subflowRefactors
__mptcp_subflow_push_pendinglogic from:For each dfrag:
While sends succeed: Call the scheduler (selects subflow and msk->snd_burst) Send the dfrag data with mptcp_subflow_delegate(), break Send the dfrag data with mptcp_sendmsg_frag() Update dfrag->already_sent, msk->snd_nxt, msk->snd_burst Update msk->first_pending->
While
first_pendingisn't empty:
Call the scheduler (selects subflow andmsk->snd_burst) Send the dfrag data withmptcp_subflow_delegate(), break Send the dfrag data withmptcp_sendmsg_frag()
For each pending dfrag:
While sends succeed:
Send the dfrag data withmptcp_sendmsg_frag()Updatealready_sent,snd_nxt,snd_burst
Updatemsk->first_pending
Break if required bymsk->snd_burst/ etcMove the duplicate code from
__mptcp_push_pending()and__mptcp_subflow_push_pending()into a new helper function, named__subflow_push_pending(). Simplify__mptcp_push_pending()and__mptcp_subflow_push_pending()by invoking this helper.Also move the burst check conditions out of the function
mptcp_subflow_get_send(), check them in__mptcp_push_pending()and__mptcp_subflow_push_pending()in the inner "for each pending dfrag" loop.Signed-off-by Geliang Tang <[email protected]>---
net/mptcp/protocol.c| 155 +++++++++++++++++++------------------------1 file changed, 70 insertions(+), 85 deletions(-)net/mptcp/protocol.c +31
-50@@ -1417,14 +1417,6 @@ struct sock *mptcp_subflow_get_send(struct mptcp_sock *msk)u64 linger_time;long tout =0 ; -/* re-use last subflow, if the burst allow that */ -if (msk->last_snd && msk->snd_burst >0 && -sk_stream_memory_free (msk->last_snd) && -mptcp_subflow_active (mptcp_subflow_ctx (msk->last_snd))) { -mptcp_set_timeout (sk); -return msk->last_snd; - } -/* pick the subflow with the lower wmem/wspace ratio */ for (i =0 ; i <SSK_MODE_MAX ; ++i) { send_info[i].ssk =NULL ;@@ -1491,12 +1483,6 @@ struct sock *mptcp_subflow_get_send(struct mptcp_sock *msk)return ssk; } -static void mptcp_push_release (struct sock *ssk,struct mptcp_sendmsg_info *info) -{ -tcp_push (ssk,0 , info->mss_now,tcp_sk (ssk)->nonagle, info->size_goal); -release_sock (ssk); -} -static void mptcp_update_post_push (struct mptcp_sock *msk,struct mptcp_data_frag *dfrag,u32 sent)@@ -1528,68 +1514,80 @@ void mptcp_check_and_set_pending(struct sock *sk)mptcp_sk (sk)->push_pending |=BIT (MPTCP_PUSH_PENDING ); } -void __mptcp_push_pending (struct sock *sk,unsigned int flags) +static int __subflow_push_pending (struct sock *sk,struct sock *ssk, +struct mptcp_sendmsg_info *info) { -struct sock *prev_ssk =NULL , *ssk =NULL ;struct mptcp_sock *msk =mptcp_sk (sk); -struct mptcp_sendmsg_info info = { - .flags = flags, - }; -bool do_check_data_fin = false;struct mptcp_data_frag *dfrag; -int len; +int len, copied =0 , err =0 ;while ((dfrag =mptcp_send_head (sk))) { - info.sent = dfrag->already_sent; - info.limit = dfrag->data_len; + info->sent = dfrag->already_sent; + info->limit = dfrag->data_len; len = dfrag->data_len - dfrag->already_sent;while (len >0 ) {int ret =0 ; - prev_ssk = ssk; - ssk =mptcp_subflow_get_send (msk); - -/* First check. If the ssk has changed since -* the last round, release prev_ssk -*/ -if (ssk != prev_ssk && prev_ssk) -mptcp_push_release (prev_ssk, &info); -if (!ssk) - goto out; - -/* Need to lock the new subflow only if different -* from the previous one, otherwise we are still -* helding the relevant lock -*/ -if (ssk != prev_ssk) -lock_sock (ssk); - - ret =mptcp_sendmsg_frag (sk, ssk, dfrag, &info); + ret =mptcp_sendmsg_frag (sk, ssk, dfrag, info);if (ret <=0 ) { -if (ret == -EAGAIN ) -continue ; -mptcp_push_release (ssk, &info); + err = copied ? : ret; goto out; } - do_check_data_fin = true; - info.sent += ret; + info->sent += ret; + copied += ret; len -= ret;mptcp_update_post_push (msk, dfrag, ret); }WRITE_ONCE (msk->first_pending,mptcp_send_next (sk)); + +if (msk->snd_burst <=0 || + !sk_stream_memory_free (ssk) || + !mptcp_subflow_active (mptcp_subflow_ctx (ssk))) { + err = copied ? : -EAGAIN ; + goto out; + } +mptcp_set_timeout (sk); + } + err = copied; + +out: +if (copied) { +tcp_push (ssk,0 , info->mss_now,tcp_sk (ssk)->nonagle, + info->size_goal); } -/* at this point we held the socket lock for the last subflow we used */ -if (ssk) -mptcp_push_release (ssk, &info); +return err; +} + +void __mptcp_push_pending (struct sock *sk,unsigned int flags) +{ +struct mptcp_sock *msk =mptcp_sk (sk); +struct mptcp_sendmsg_info info = {
net/mptcp/protocol.c void mptcp_check_and_set_pending(struct sock *sk) · line 1577
Quoted 10 lines
net/mptcp/protocol.c +10
-0+ .flags = flags, + }; +struct sock *ssk; +int ret =0 ; + +again: +while (mptcp_send_head (sk) && (ssk =mptcp_subflow_get_send (msk))) { +lock_sock (ssk); + ret =__subflow_push_pending (sk, ssk, &info); +release_sock (ssk);
Thanks for reworking this patch set, it's easier to see what's going on now.
With this refactoring a long send will push, release, and re-lock the subflow for every "burst" (as limited by snd_burst or by the available buffer space). That seems like it could add noticeable overhead for high-throughput connections, compared to the old code that would defer the push & unlock. In earlier reviews I didn't consider this carefully enough: I thought that __subflow_push_pending() would usually be able to send all the pending dfrags, but I didn't consider sendmsg calls with lots of data (more than the burst size).
I think it's possible to keep the push/unlock optimization in __mptcp_push_pending() if the tcp_push() is handled outside __subflow_push_pending().
Would likely need more information returned from the helper to give the caller of __subflow_push_pending() enough information, like:
static int __subflow_push_pending(struct sock *sk, struct sock *ssk,
struct mptcp_sendmsg_info *info, bool *needs_push)
I'll talk to Paolo about this on Thursday too (and he's welcome to chime in here :) )
net/mptcp/protocol.c void mptcp_check_and_set_pending(struct sock *sk) · line 1581
Quoted 4 lines
net/mptcp/protocol.c +4
-0+ +if (ret <=0 ) { +if (ret == -EAGAIN ) + goto again;
Better to use a 'continue' here.
Quoted 60 lines
net/mptcp/protocol.c +3
-0+ goto out; + } + }out:
/* ensure the rtx timer is running */ if (!mptcp_timer_pending(sk)) mptcp_reset_timer(sk);net/mptcp/protocol.c +1
-1-if (do_check_data_fin) +if (ret >0 )__mptcp_check_send_data_fin(sk); }net/mptcp/protocol.c +11
-29@@ -1599,51 +1597,38 @@ static void __mptcp_subflow_push_pending(struct sock *sk, struct sock *ssk, boolstruct mptcp_sendmsg_info info = { .data_lock_held = true, }; -struct mptcp_data_frag *dfrag;struct sock *xmit_ssk; -int len, copied =0 ; +int ret =0 ; info.flags =0 ; -while ((dfrag =mptcp_send_head (sk))) { - info.sent = dfrag->already_sent; - info.limit = dfrag->data_len; - len = dfrag->data_len - dfrag->already_sent; -while (len >0 ) { -int ret =0 ; - -/* check for a different subflow usage only after -* spooling the first chunk of data -*/ - xmit_ssk = first ? ssk :mptcp_subflow_get_send (msk); -if (!xmit_ssk) - goto out; -if (xmit_ssk != ssk) { -mptcp_subflow_delegate (mptcp_subflow_ctx (xmit_ssk), -MPTCP_DELEGATE_SEND ); - goto out; - } - - ret =mptcp_sendmsg_frag (sk, ssk, dfrag, &info); -if (ret <=0 ) - goto out; - - info.sent += ret; - copied += ret; - len -= ret; - first = false; +again: +while (mptcp_send_head (sk)) { +/* check for a different subflow usage only after +* spooling the first chunk of data +*/ + xmit_ssk = first ? ssk :mptcp_subflow_get_send (msk); +if (!xmit_ssk) + goto out; +if (xmit_ssk != ssk) { + mptcp_subflow_delegate(mptcp_subflow_ctx(xmit_ssk),
net/mptcp/protocol.c static void __mptcp_subflow_push_pending(struct sock *sk, struct sock *ssk, bool · line 1622
Quoted 10 lines
net/mptcp/protocol.c +8
-1+ MPTCP_DELEGATE_SEND); + goto out; + } -mptcp_update_post_push (msk, dfrag, ret); + ret =__subflow_push_pending (sk, ssk, &info); + first = false; +if (ret <=0 ) { +if (ret == -EAGAIN ) + goto again;
Another place to use 'continue'.
Quoted 13 lines
net/mptcp/protocol.c +1
-1+break ; } -WRITE_ONCE (msk->first_pending,mptcp_send_next (sk)); }out:
/* __mptcp_alloc_tx_skb could have released some wmem and we are * not going to flush it via release_sock() */net/mptcp/protocol.c +1
-3-if (copied) { -tcp_push (ssk,0 , info.mss_now,tcp_sk (ssk)->nonagle, - info.size_goal); +if (ret >0 ) {
Quoted 8 lines
if (!mptcp_timer_pending(sk)) mptcp_reset_timer(sk);Signature · 1 line
2.35.3
--
Mat Martineau
Intel
Re: mptcp: refactor push_pending logic: Tests Results
In reply to Geliang TangHi Geliang,
Thank you for your modifications, that's great!
Our CI did some validations and here is its report:
- KVM Validation: normal:
- Unstable: 1 failed test(s): packetdrill_add_addr 🔴:
- Task: https://cirrus-ci.com/task/4741266072666112
- Summary: https://api.cirrus-ci.com/v1/artifact/task/4741266072666112/summary/summary.txt
- {"code":404,"message":
- "Can't find artifacts containing file conclusion.txt"}:
- Task: https://cirrus-ci.com/task/5867165979508736
- Summary: https://api.cirrus-ci.com/v1/artifact/task/5867165979508736/summary/summary.txt
Initiator: Patchew Applier
Commits: https://github.com/multipath-tcp/mptcp_net-next/commits/12a4c489939c
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-debugFor 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 (Tessares)
Quoted 6 lines · Geliang Tang
v9:
- move "register default scheduler" out of this series.
- add missing "first = false" in patch 4
- patch 2 could be merged before the commit "mptcp: add structmptcp_sched_ops" as a non-bpf cleanup.
The first 3 patches look good to me:
I have some other comments for patch 4.
Quoted 40 lines
v8:
- merge "register default scheduler" into this series.
- Move "multi-subflow scheduling" out of this series.
- This series should be merged between the commit "mptcp: addget_subflowwrappers" and "bpf: Addbpf_mptcp_sched_ops", except the squash-to patches.v7:
- update delegate sending in patch 10
- rebase to export/20221011T063543.v6:
- drop allmsk->last_snd, addlast_sndvariable instead in patch 13
- fixlock_sockissue in patch 3
- merge squash-to patchesv5:
- address Mat's comments in v4.v4:
- update__mptcp_subflow_push_pendingas Mat suggested.
- add more patches from "BPF redundant scheduler" series.v3:
- add a cleanup patch.
- removemsk->last_sndinmptcp_subflow_get_send().
- add the loop that calls the scheduler again in__mptcp_push_pending().v2:
- addsnd_burstcheck in dfrags loop as Mat suggested.Refactor
__mptcp_push_pending()and__mptcp_subflow_push_pending()to remove duplicate code and support redundant scheduler more easily in__mptcp_subflow_push_pending().Geliang Tang (4):
Squash to "mptcp: addget_subflowwrappers"
mptcp: use msk instead ofmptcp_sk
mptcp: change 'first' as a parameter
mptcp: refactorpush_pendinglogic
Quoted 10 lines
net/mptcp/protocol.c | 183 +++++++++++++++++++------------------------ net/mptcp/protocol.h | 4 +- net/mptcp/sched.c | 61 +++++++-------- 3 files changed, 111 insertions(+), 137 deletions(-)Signature · 1 line
2.35.3
--
Mat Martineau
Intel
Hi Geliang, Mat,
Quoted 9 lines · Mat Martineau
Quoted 6 lines · Geliang Tang
v9:
- move "register default scheduler" out of this series.
- add missing "first = false" in patch 4
- patch 2 could be merged before the commit "mptcp: add structmptcp_sched_ops" as a non-bpf cleanup.
Quoted 3 lines
The first 3 patches look good to me:
Reviewed-by Mat Martineau <[email protected]>
Thank you for the patches and the reviews!
I just added these patches in our tree (feat. with net-next) with Mat's RvB tag:
- f7b33377ef44: "squashed" patch 1/4 in "mptcp: add get_subflow wrappers"
- abc20912434f: tg:msg: update commit message
- 810d6dba28d6: mptcp: use msk instead of mptcp_sk
- 3afbefdac9ae: mptcp: change 'first' as a parameter
- Results: e4c052c32f02..1f94545f93dd (export)
Tests are now in progress:
https://cirrus-ci.com/github/multipath-tcp/mptcp_net-next/export/20221020T142332
Cheers,
Matt
Signature · 2 lines
Tessares | Belgium | Hybrid Access Solutions www.tessares.net