[PATCH mptcp-next 1/2] mptcp: never fetch fwd memory from the subflow

[PATCH mptcp-next 1/2] mptcp: never fetch fwd memory from the subflow

[Paolo Abeni]

The memory accounting is broken in such exceptional code
path, and after commit 4890b686f408 ("net: keep sk->sk_forward_alloc
as small as possible") we can't find much help there.

Drop the broken code.

Signed-off-by: Paolo Abeni <pabeni@redhat.com>
---
This is possibly for net, but makes sense only on top of recent
net-next patches, so whatever ;)
---
 net/mptcp/protocol.c | 11 +++--------
 1 file changed, 3 insertions(+), 8 deletions(-)

diff --git a/net/mptcp/protocol.c b/net/mptcp/protocol.c
index 35e3060233c9..390ac5d95664 100644
--- a/net/mptcp/protocol.c
+++ b/net/mptcp/protocol.c
@@ -328,15 +328,10 @@ static bool mptcp_rmem_schedule(struct sock *sk, struct sock *ssk, int size)
 
 	amt = sk_mem_pages(size);
 	amount = amt << PAGE_SHIFT;
-	msk->rmem_fwd_alloc += amount;
-	if (!__sk_mem_raise_allocated(sk, size, amt, SK_MEM_RECV)) {
-		if (ssk->sk_forward_alloc < amount) {
-			msk->rmem_fwd_alloc -= amount;
-			return false;
-		}
+	if (!__sk_mem_raise_allocated(sk, size, amt, SK_MEM_RECV))
+		return false;
 
-		ssk->sk_forward_alloc -= amount;
-	}
+	msk->rmem_fwd_alloc += amount;
 	return true;
 }
 
-- 
2.35.3

[PATCH mptcp-next 2/2] mptcp: drop SK_RECLAIM_* macros

[Paolo Abeni]

After commit 4890b686f408 ("net: keep sk->sk_forward_alloc as small as
possible"), the MPTCP protocol is the last SK_RECLAIM_CHUNK and
SK_RECLAIM_THRESHOLD users.

Update the MPTCP reclaim schema to match the core/TCP one and drop the
mentioned macros. This additionally clean the MPTCP code a bit.

Signed-off-by: Paolo Abeni <pabeni@redhat.com>
---
 net/mptcp/protocol.c | 35 ++---------------------------------
 1 file changed, 2 insertions(+), 33 deletions(-)

diff --git a/net/mptcp/protocol.c b/net/mptcp/protocol.c
index 390ac5d95664..bc79afa7e76d 100644
--- a/net/mptcp/protocol.c
+++ b/net/mptcp/protocol.c
@@ -181,8 +181,8 @@ static void mptcp_rmem_uncharge(struct sock *sk, int size)
 	reclaimable = msk->rmem_fwd_alloc - sk_unused_reserved_mem(sk);
 
 	/* see sk_mem_uncharge() for the rationale behind the following schema */
-	if (unlikely(reclaimable >= SK_RECLAIM_THRESHOLD))
-		__mptcp_rmem_reclaim(sk, SK_RECLAIM_CHUNK);
+	if (unlikely(reclaimable >= PAGE_SIZE))
+		__mptcp_rmem_reclaim(sk, reclaimable);
 }
 
 static void mptcp_rfree(struct sk_buff *skb)
@@ -961,25 +961,6 @@ static bool mptcp_frag_can_collapse_to(const struct mptcp_sock *msk,
 		df->data_seq + df->data_len == msk->write_seq;
 }
 
-static void __mptcp_mem_reclaim_partial(struct sock *sk)
-{
-	int reclaimable = mptcp_sk(sk)->rmem_fwd_alloc - sk_unused_reserved_mem(sk);
-
-	lockdep_assert_held_once(&sk->sk_lock.slock);
-
-	if (reclaimable > (int)PAGE_SIZE)
-		__mptcp_rmem_reclaim(sk, reclaimable - 1);
-
-	sk_mem_reclaim(sk);
-}
-
-static void mptcp_mem_reclaim_partial(struct sock *sk)
-{
-	mptcp_data_lock(sk);
-	__mptcp_mem_reclaim_partial(sk);
-	mptcp_data_unlock(sk);
-}
-
 static void dfrag_uncharge(struct sock *sk, int len)
 {
 	sk_mem_uncharge(sk, len);
@@ -999,7 +980,6 @@ static void __mptcp_clean_una(struct sock *sk)
 {
 	struct mptcp_sock *msk = mptcp_sk(sk);
 	struct mptcp_data_frag *dtmp, *dfrag;
-	bool cleaned = false;
 	u64 snd_una;
 
 	/* on fallback we just need to ignore snd_una, as this is really
@@ -1022,7 +1002,6 @@ static void __mptcp_clean_una(struct sock *sk)
 		}
 
 		dfrag_clear(sk, dfrag);
-		cleaned = true;
 	}
 
 	dfrag = mptcp_rtx_head(sk);
@@ -1044,7 +1023,6 @@ static void __mptcp_clean_una(struct sock *sk)
 		dfrag->already_sent -= delta;
 
 		dfrag_uncharge(sk, delta);
-		cleaned = true;
 	}
 
 	/* all retransmitted data acked, recovery completed */
@@ -1052,9 +1030,6 @@ static void __mptcp_clean_una(struct sock *sk)
 		msk->recovery = false;
 
 out:
-	if (cleaned && tcp_under_memory_pressure(sk))
-		__mptcp_mem_reclaim_partial(sk);
-
 	if (snd_una == READ_ONCE(msk->snd_nxt) &&
 	    snd_una == READ_ONCE(msk->write_seq)) {
 		if (mptcp_timer_pending(sk) && !mptcp_data_fin_enabled(msk))
@@ -1206,12 +1181,6 @@ static struct sk_buff *mptcp_alloc_tx_skb(struct sock *sk, struct sock *ssk, boo
 {
 	gfp_t gfp = data_lock_held ? GFP_ATOMIC : sk->sk_allocation;
 
-	if (unlikely(tcp_under_memory_pressure(sk))) {
-		if (data_lock_held)
-			__mptcp_mem_reclaim_partial(sk);
-		else
-			mptcp_mem_reclaim_partial(sk);
-	}
 	return __mptcp_alloc_tx_skb(sk, ssk, gfp);
 }
 
-- 
2.35.3

Re: [PATCH mptcp-next 1/2] mptcp: never fetch fwd memory from the subflow

[Mat Martineau]

On Thu, 16 Jun 2022, Paolo Abeni wrote:

> The memory accounting is broken in such exceptional code
> path, and after commit 4890b686f408 ("net: keep sk->sk_forward_alloc
> as small as possible") we can't find much help there.
>
> Drop the broken code.
>
> Signed-off-by: Paolo Abeni <pabeni@redhat.com>
> ---
> This is possibly for net, but makes sense only on top of recent
> net-next patches, so whatever ;)

Let's go with net-next.

> ---
> net/mptcp/protocol.c | 11 +++--------
> 1 file changed, 3 insertions(+), 8 deletions(-)
>
> diff --git a/net/mptcp/protocol.c b/net/mptcp/protocol.c
> index 35e3060233c9..390ac5d95664 100644
> --- a/net/mptcp/protocol.c
> +++ b/net/mptcp/protocol.c
> @@ -328,15 +328,10 @@ static bool mptcp_rmem_schedule(struct sock *sk, struct sock *ssk, int size)
>
> 	amt = sk_mem_pages(size);
> 	amount = amt << PAGE_SHIFT;
> -	msk->rmem_fwd_alloc += amount;
> -	if (!__sk_mem_raise_allocated(sk, size, amt, SK_MEM_RECV)) {
> -		if (ssk->sk_forward_alloc < amount) {
> -			msk->rmem_fwd_alloc -= amount;
> -			return false;
> -		}
> +	if (!__sk_mem_raise_allocated(sk, size, amt, SK_MEM_RECV))
> +		return false;

Like Eric did in 7c80b038d23e, does it makes sense to try to raise the 
allocation only by the amount of incremental memory requested (size - 
msk->rmem_forward_alloc) rather than the full amount from the 'size' arg?

>
> -		ssk->sk_forward_alloc -= amount;
> -	}
> +	msk->rmem_fwd_alloc += amount;
> 	return true;
> }

--
Mat Martineau
Intel

Re: [PATCH mptcp-next 1/2] mptcp: never fetch fwd memory from the subflow

[Paolo Abeni]

On Fri, 2022-06-17 at 14:54 -0700, Mat Martineau wrote:
> On Thu, 16 Jun 2022, Paolo Abeni wrote:
> 
> > The memory accounting is broken in such exceptional code
> > path, and after commit 4890b686f408 ("net: keep sk->sk_forward_alloc
> > as small as possible") we can't find much help there.
> > 
> > Drop the broken code.
> > 
> > Signed-off-by: Paolo Abeni <pabeni@redhat.com>
> > ---
> > This is possibly for net, but makes sense only on top of recent
> > net-next patches, so whatever ;)
> 
> Let's go with net-next.
> 
> > ---
> > net/mptcp/protocol.c | 11 +++--------
> > 1 file changed, 3 insertions(+), 8 deletions(-)
> > 
> > diff --git a/net/mptcp/protocol.c b/net/mptcp/protocol.c
> > index 35e3060233c9..390ac5d95664 100644
> > --- a/net/mptcp/protocol.c
> > +++ b/net/mptcp/protocol.c
> > @@ -328,15 +328,10 @@ static bool mptcp_rmem_schedule(struct sock *sk, struct sock *ssk, int size)
> > 
> > 	amt = sk_mem_pages(size);
> > 	amount = amt << PAGE_SHIFT;
> > -	msk->rmem_fwd_alloc += amount;
> > -	if (!__sk_mem_raise_allocated(sk, size, amt, SK_MEM_RECV)) {
> > -		if (ssk->sk_forward_alloc < amount) {
> > -			msk->rmem_fwd_alloc -= amount;
> > -			return false;
> > -		}
> > +	if (!__sk_mem_raise_allocated(sk, size, amt, SK_MEM_RECV))
> > +		return false;
> 
> Like Eric did in 7c80b038d23e, does it makes sense to try to raise the 
> allocation only by the amount of incremental memory requested (size - 
> msk->rmem_forward_alloc) rather than the full amount from the 'size' arg?

Ok, but let's put such change on a different commit. Upstream just
reported a huge regression in SCTP due to the above change, and I
haven't benchmarked MPTCP yet. Keeping the change separate we can
revert/fix/backport fix more easily.

Cheers,

Paolo