From: Geliang Tang <tanggeliang@kylinos.cn>

This patch adds .established and .subflow_established interfaces for
struct mptcp_pm_ops, and calls pm->ops->established/subflow_established
in from mptcp_pm_worker(). Then get rid of the corresponding code from
__mptcp_pm_kernel_worker().

Since mptcp_pm_addr_send_ack() is a sleepable kfunc, which is invoked
by mptcp_pm_create_subflow_or_signal_addr(), .established() and
.subflow_established() interfaces of BPF PM should be invoked by
__bpf_prog_enter_sleepable(), which can't be invoked under a lock.
This patch unlocks the pm lock before invoking this interface in
mptcp_pm_worker(), while holding this lock in mptcp_pm_kernel_established()
and mptcp_pm_kernel_subflow_established().

Signed-off-by: Geliang Tang <tanggeliang@kylinos.cn>
---
 include/net/mptcp.h   |  4 ++++
 net/mptcp/pm.c        | 32 ++++++++++++++++++++++++--------
 net/mptcp/pm_kernel.c | 25 +++++++++++--------------
 3 files changed, 39 insertions(+), 22 deletions(-)

diff --git a/include/net/mptcp.h b/include/net/mptcp.h
index 9f28ef550e10..4ac936e4ce0d 100644
--- a/include/net/mptcp.h
+++ b/include/net/mptcp.h
@@ -124,6 +124,10 @@ struct mptcp_pm_ops {
 	bool (*get_priority)(struct mptcp_sock *msk,
 			     struct mptcp_addr_info *skc);
 
+	/* optional */
+	void (*established)(struct mptcp_sock *msk);
+	void (*subflow_established)(struct mptcp_sock *msk);
+
 	char			name[MPTCP_PM_NAME_MAX];
 	struct module		*owner;
 	struct list_head	list;
diff --git a/net/mptcp/pm.c b/net/mptcp/pm.c
index ba7424582ebf..e2b2c874a9f8 100644
--- a/net/mptcp/pm.c
+++ b/net/mptcp/pm.c
@@ -516,7 +516,8 @@ void mptcp_pm_fully_established(struct mptcp_sock *msk, const struct sock *ssk)
 	 * be sure to serve this event only once.
 	 */
 	if (READ_ONCE(pm->work_pending) &&
-	    !(pm->status & BIT(MPTCP_PM_ALREADY_ESTABLISHED)))
+	    !(pm->status & BIT(MPTCP_PM_ALREADY_ESTABLISHED)) &&
+	    pm->ops->established)
 		mptcp_pm_schedule_work(msk, MPTCP_PM_ESTABLISHED);
 
 	if ((pm->status & BIT(MPTCP_PM_ALREADY_ESTABLISHED)) == 0)
@@ -543,7 +544,7 @@ void mptcp_pm_subflow_established(struct mptcp_sock *msk)
 
 	pr_debug("msk=%p\n", msk);
 
-	if (!READ_ONCE(pm->work_pending))
+	if (!READ_ONCE(pm->work_pending) || !pm->ops->subflow_established)
 		return;
 
 	spin_lock_bh(&pm->lock);
@@ -570,7 +571,8 @@ void mptcp_pm_subflow_check_next(struct mptcp_sock *msk,
 		return;
 	}
 
-	if (!READ_ONCE(pm->work_pending) && !update_subflows)
+	if (!pm->ops->subflow_established ||
+	    (!READ_ONCE(pm->work_pending) && !update_subflows))
 		return;
 
 	spin_lock_bh(&pm->lock);
@@ -628,7 +630,7 @@ void mptcp_pm_add_addr_echoed(struct mptcp_sock *msk,
 
 	pr_debug("msk=%p\n", msk);
 
-	if (!READ_ONCE(pm->work_pending))
+	if (!READ_ONCE(pm->work_pending) || !pm->ops->subflow_established)
 		return;
 
 	spin_lock_bh(&pm->lock);
@@ -949,20 +951,34 @@ void mptcp_pm_worker(struct mptcp_sock *msk)
 	if (!(pm->status & MPTCP_PM_WORK_MASK))
 		return;
 
-	spin_lock_bh(&msk->pm.lock);
-
 	pr_debug("msk=%p status=%x\n", msk, pm->status);
 	if (pm->status & BIT(MPTCP_PM_ADD_ADDR_SEND_ACK)) {
+		spin_lock_bh(&pm->lock);
 		pm->status &= ~BIT(MPTCP_PM_ADD_ADDR_SEND_ACK);
 		mptcp_pm_addr_send_ack(msk);
+		spin_unlock_bh(&pm->lock);
 	}
 	if (pm->status & BIT(MPTCP_PM_RM_ADDR_RECEIVED)) {
+		spin_lock_bh(&pm->lock);
 		pm->status &= ~BIT(MPTCP_PM_RM_ADDR_RECEIVED);
 		mptcp_pm_rm_addr_recv(msk);
+		spin_unlock_bh(&pm->lock);
+	}
+	if (pm->status & BIT(MPTCP_PM_ESTABLISHED)) {
+		spin_lock_bh(&pm->lock);
+		pm->status &= ~BIT(MPTCP_PM_ESTABLISHED);
+		spin_unlock_bh(&pm->lock);
+		pm->ops->established(msk);
+	}
+	if (pm->status & BIT(MPTCP_PM_SUBFLOW_ESTABLISHED)) {
+		spin_lock_bh(&pm->lock);
+		pm->status &= ~BIT(MPTCP_PM_SUBFLOW_ESTABLISHED);
+		spin_unlock_bh(&pm->lock);
+		pm->ops->subflow_established(msk);
 	}
+	spin_lock_bh(&pm->lock);
 	__mptcp_pm_kernel_worker(msk);
-
-	spin_unlock_bh(&msk->pm.lock);
+	spin_unlock_bh(&pm->lock);
 }
 
 static void mptcp_pm_ops_init(struct mptcp_sock *msk,
diff --git a/net/mptcp/pm_kernel.c b/net/mptcp/pm_kernel.c
index 7ec81d5195d4..2e181224bccb 100644
--- a/net/mptcp/pm_kernel.c
+++ b/net/mptcp/pm_kernel.c
@@ -269,6 +269,7 @@ static void mptcp_pm_create_subflow_or_signal_addr(struct mptcp_sock *msk)
 	local_addr_max = mptcp_pm_get_local_addr_max(msk);
 	subflows_max = mptcp_pm_get_subflows_max(msk);
 
+	spin_lock_bh(&msk->pm.lock);
 	/* do lazy endpoint usage accounting for the MPC subflows */
 	if (unlikely(!(msk->pm.status & BIT(MPTCP_PM_MPC_ENDPOINT_ACCOUNTED))) && msk->first) {
 		struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(msk->first);
@@ -307,7 +308,7 @@ static void mptcp_pm_create_subflow_or_signal_addr(struct mptcp_sock *msk)
 		 * current address announce will be completed.
 		 */
 		if (msk->pm.addr_signal & BIT(MPTCP_ADD_ADDR_SIGNAL))
-			return;
+			goto out;
 
 		if (!select_signal_address(pernet, msk, &local))
 			goto subflow;
@@ -316,7 +317,7 @@ static void mptcp_pm_create_subflow_or_signal_addr(struct mptcp_sock *msk)
 		 * continuing, and trying to create subflows.
 		 */
 		if (!mptcp_pm_alloc_anno_list(msk, &local.addr))
-			return;
+			goto out;
 
 		__clear_bit(local.addr.id, msk->pm.id_avail_bitmap);
 		msk->pm.add_addr_signaled++;
@@ -365,14 +366,16 @@ static void mptcp_pm_create_subflow_or_signal_addr(struct mptcp_sock *msk)
 		spin_lock_bh(&msk->pm.lock);
 	}
 	mptcp_pm_nl_check_work_pending(msk);
+out:
+	spin_unlock_bh(&msk->pm.lock);
 }
 
-static void mptcp_pm_nl_fully_established(struct mptcp_sock *msk)
+static void mptcp_pm_kernel_established(struct mptcp_sock *msk)
 {
 	mptcp_pm_create_subflow_or_signal_addr(msk);
 }
 
-static void mptcp_pm_nl_subflow_established(struct mptcp_sock *msk)
+static void mptcp_pm_kernel_subflow_established(struct mptcp_sock *msk)
 {
 	mptcp_pm_create_subflow_or_signal_addr(msk);
 }
@@ -758,8 +761,8 @@ static int mptcp_nl_add_subflow_or_signal_addr(struct net *net,
 		spin_lock_bh(&msk->pm.lock);
 		if (mptcp_addresses_equal(addr, &mpc_addr, addr->port))
 			msk->mpc_endpoint_id = addr->id;
-		mptcp_pm_create_subflow_or_signal_addr(msk);
 		spin_unlock_bh(&msk->pm.lock);
+		mptcp_pm_create_subflow_or_signal_addr(msk);
 		release_sock(sk);
 
 next:
@@ -1243,8 +1246,8 @@ static void mptcp_pm_nl_fullmesh(struct mptcp_sock *msk,
 	spin_lock_bh(&msk->pm.lock);
 	mptcp_pm_rm_subflow(msk, &list);
 	__mark_subflow_endp_available(msk, list.ids[0]);
-	mptcp_pm_create_subflow_or_signal_addr(msk);
 	spin_unlock_bh(&msk->pm.lock);
+	mptcp_pm_create_subflow_or_signal_addr(msk);
 }
 
 static void mptcp_pm_nl_set_flags_all(struct net *net,
@@ -1348,14 +1351,6 @@ void __mptcp_pm_kernel_worker(struct mptcp_sock *msk)
 		pm->status &= ~BIT(MPTCP_PM_ADD_ADDR_RECEIVED);
 		mptcp_pm_nl_add_addr_received(msk);
 	}
-	if (pm->status & BIT(MPTCP_PM_ESTABLISHED)) {
-		pm->status &= ~BIT(MPTCP_PM_ESTABLISHED);
-		mptcp_pm_nl_fully_established(msk);
-	}
-	if (pm->status & BIT(MPTCP_PM_SUBFLOW_ESTABLISHED)) {
-		pm->status &= ~BIT(MPTCP_PM_SUBFLOW_ESTABLISHED);
-		mptcp_pm_nl_subflow_established(msk);
-	}
 }
 
 static int __net_init pm_nl_init_net(struct net *net)
@@ -1422,6 +1417,8 @@ static void mptcp_pm_kernel_init(struct mptcp_sock *msk)
 struct mptcp_pm_ops mptcp_pm_kernel = {
 	.get_local_id		= mptcp_pm_kernel_get_local_id,
 	.get_priority		= mptcp_pm_kernel_get_priority,
+	.established		= mptcp_pm_kernel_established,
+	.subflow_established	= mptcp_pm_kernel_subflow_established,
 	.init			= mptcp_pm_kernel_init,
 	.name			= "kernel",
 	.owner			= THIS_MODULE,
-- 
2.43.0

Hi Geliang,

On 24/03/2025 09:19, Geliang Tang wrote:
> From: Geliang Tang <tanggeliang@kylinos.cn>
> 
> This patch adds .established and .subflow_established interfaces for
> struct mptcp_pm_ops, and calls pm->ops->established/subflow_established
> in from mptcp_pm_worker(). Then get rid of the corresponding code from
> __mptcp_pm_kernel_worker().
> 
> Since mptcp_pm_addr_send_ack() is a sleepable kfunc, which is invoked
> by mptcp_pm_create_subflow_or_signal_addr(), .established() and
> .subflow_established() interfaces of BPF PM should be invoked by
> __bpf_prog_enter_sleepable(), which can't be invoked under a lock.
> This patch unlocks the pm lock before invoking this interface in
> mptcp_pm_worker(), while holding this lock in mptcp_pm_kernel_established()
> and mptcp_pm_kernel_subflow_established().

(...)

> @@ -949,20 +951,34 @@ void mptcp_pm_worker(struct mptcp_sock *msk)
>  	if (!(pm->status & MPTCP_PM_WORK_MASK))
>  		return;
>  
> -	spin_lock_bh(&msk->pm.lock);
> -
>  	pr_debug("msk=%p status=%x\n", msk, pm->status);
>  	if (pm->status & BIT(MPTCP_PM_ADD_ADDR_SEND_ACK)) {

This should probably be read under the lock or with a READ_ONCE. Or
manipulating pm->status before, then running the different actions. I
need to think about that.

I also think it might be good to have a dedicated patch moving the
locking mechanisms first, then introducing the new callbacks.

I will discuss that with Mat on Wednesday. Do you mind holding new
versions for this series until then please?

Cheers,
Matt
-- 
Sponsored by the NGI0 Core fund.