Skip to contentSkip to conversation

mptcp: address stall under memory pressure

PATCH Superseded Paolo Abeni 2 participants 13 messages Apr 27 – Apr 28

Subsystems (3)

v2 of 12 Superseded Latest is v12 →

lore
Share & tools
lore

Paolo Abeni

author maintainer
#0 lore raw

v2 cover mptcp: address stall under memory pressure

This an attempt to fix the data transfer stall reported by Geliang and Gang more carefully enforcing memory constraints at the MPTCP level.

Patch 1/10 moves the bound check before entering the TCP socket. Patch 2, 3, 4 and 5 are cleanups/refactors finalized to safely re-using TCP helpers on MPTCP skbs.
Patch 6 makes TCP pruning related helpers available to MPTCP and patch 7 makes use of them. Patch 8 addresses an edge scenario that could still lead to transfer stall under memory pressure.
Finally patch 9 and 10 improve the MPTCP-level retransmission schema to make recovery from memory pressure significanly faster.

Note that the diffstat is biases by the quite large patch 4/9, which contains mechanical transformation of existing code; "real" changes are noticiable smaller.

Tested successfully vs the test cases proposed by Geliang and Gang and vs the selftests.
---
Notes:
- Sashiko apparently can't understand that mptcp_data_lock() && sk not owned by the user-space is equivalent to the sk socket lock. As a consequences marks incorrectly as racy a few accesses under the data lock to the sk_rcvbuf.
- mptcp_join with "4 subflows with under pressure" may require longer than default timeout, as the transfer in such condition can be really slow.

Paolo Abeni (10):
mptcp: move checks vs rcvbuf size earlier in the RX path
mptcp: drop the mptcp_ooo_try_coalesce() helper
mptcp: drop the cant_coalesce CB field
mptcp: remove CB offset field
mptcp: sync mptcp skb cb layout with tcp one
tcp: expose the tcp_collapse_ofo_queue() helper to mptcp usage, too
mptcp: implemented OoO queue pruning
mptcp: track prune recovery status
mptcp: move the retrans loop to a separate helper
mptcp: let the retrans scheduler do its job.

 include/net/tcp.h    |   8 +
 net/ipv4/tcp_input.c |  55 +++---
 net/mptcp/fastopen.c |   9 +-
 net/mptcp/mib.c      |   3 +
 net/mptcp/mib.h      |   3 +
 net/mptcp/options.c  |  58 ++++++-
 net/mptcp/protocol.c | 403 ++++++++++++++++++++++++++++---------------
 net/mptcp/protocol.h |  18 +-
 net/mptcp/subflow.c  |  10 ++
 9 files changed, 398 insertions(+), 169 deletions(-)
Signature · 1 line
2.53.0
author maintainer
#1 lore raw

PATCH v2 1/10 mptcp: move checks vs rcvbuf size earlier in the RX path

Currently the enforcement of the rcvbuf constraint is implemented when moving the skbs into the msk receive or OoO queue.

Under significant memory pressure the above can cause permanent data transfer stalls. Move the checks early on, before landing even in the subflow queues.

---
v1 -> v2:
- deal correctly with tcp fin and zero win probe

RFC -> v1:
- limit vs actual buffer size
- use CB info instead of skb->len

Note that:
- this needs the follow-up patches to really fix the stall
- the memory comparison is intentionally very rough, as the msk socket lock is not currently held where the condition is now enforced. This should require some refinement, shared as-is to avoid more latency on my side

---
 net/mptcp/options.c  | 25 +++++++++++++++++++++++--
 net/mptcp/protocol.c | 10 ++--------
 2 files changed, 25 insertions(+), 10 deletions(-)
net/mptcp/options.c +23 -2
index 4cc583fdc7a9..ad4bb6fd86e1 100644
@@ -1158,8 +1158,23 @@ static bool add_addr_hmac_valid(struct mptcp_sock *msk,
 	return hmac == mp_opt->ahmac;
 }
 
-/* Return false in case of error (or subflow has been reset),
- * else return true.
+static bool mptcp_over_limit(struct sock *sk, const struct sock *ssk,
+			     const struct sk_buff *skb)
+{
+	if (likely(sk_rmem_alloc_get(sk) <= READ_ONCE(sk->sk_rcvbuf)))
+		return false;
+
+	/* Avoid silently dropping pure acks, fin or zero win probes. */
+	if (TCP_SKB_CB(skb)->seq == TCP_SKB_CB(skb)->end_seq ||
+	    TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN ||
+	    !after(TCP_SKB_CB(skb)->end_seq, tcp_sk(ssk)->rcv_nxt))
+		return false;
+
+	return true;
+}
+
+/* Return false when the caller must drop the packet, i.e. in case of error,
+ * subflow has been reset, or over memory limits.
  */
 bool mptcp_incoming_options(struct sock *sk, struct sk_buff *skb)
 {
@@ -1185,6 +1200,9 @@ bool mptcp_incoming_options(struct sock *sk, struct sk_buff *skb)
 
 		__mptcp_data_acked(subflow->conn);
 		mptcp_data_unlock(subflow->conn);
+
+		if (mptcp_over_limit(subflow->conn, sk, skb))
+			return false;
 		return true;
 	}
 
@@ -1263,6 +1281,9 @@ bool mptcp_incoming_options(struct sock *sk, struct sk_buff *skb)
 		return true;
 	}
 
+	if (mptcp_over_limit(subflow->conn, sk, skb))
+		return false;
+
 	mpext = skb_ext_add(skb, SKB_EXT_MPTCP);
 	if (!mpext)
 		return false;
net/mptcp/protocol.c +2 -8
index 17b9a8c13ebf..81a9b8077d6b 100644
@@ -739,7 +739,7 @@ static bool __mptcp_move_skbs_from_subflow(struct mptcp_sock *msk,
 
 			mptcp_init_skb(ssk, skb, offset, len);
 
-			if (own_msk && sk_rmem_alloc_get(sk) < sk->sk_rcvbuf) {
+			if (own_msk) {
 				mptcp_subflow_lend_fwdmem(subflow, skb);
 				ret |= __mptcp_move_skb(sk, skb);
 			} else {
@@ -2197,10 +2197,6 @@ static bool __mptcp_move_skbs(struct sock *sk, struct list_head *skbs, u32 *delt
 
 	*delta = 0;
 	while (1) {
-		/* If the msk recvbuf is full stop, don't drop */
-		if (sk_rmem_alloc_get(sk) > sk->sk_rcvbuf)
-			break;
-
 		prefetch(skb->next);
 		list_del(&skb->list);
 		*delta += skb->truesize;
@@ -2228,9 +2224,7 @@ static bool mptcp_can_spool_backlog(struct sock *sk, struct list_head *skbs)
 	DEBUG_NET_WARN_ON_ONCE(msk->backlog_unaccounted && sk->sk_socket &&
 			       mem_cgroup_from_sk(sk));
 
-	/* Don't spool the backlog if the rcvbuf is full. */
-	if (list_empty(&msk->backlog_list) ||
-	    sk_rmem_alloc_get(sk) > sk->sk_rcvbuf)
+	if (list_empty(&msk->backlog_list))
 		return false;
 
 	INIT_LIST_HEAD(skbs);
Signature · 1 line
2.53.0
author maintainer
#2 lore raw

PATCH v2 2/10 mptcp: drop the mptcp_ooo_try_coalesce() helper

In reply to Paolo Abeni

It's used to save an additional comparison for in-order skbs, but is also a barrier to remove CB offset. Remove the helper, let __mptcp_try_coalesce() always perform the sequence check and remove duplicate checks from the callers.

---
 net/mptcp/protocol.c | 21 ++++++---------------
 1 file changed, 6 insertions(+), 15 deletions(-)
net/mptcp/protocol.c +6 -15
index 81a9b8077d6b..ad0a289b544b 100644
@@ -159,7 +159,8 @@ static bool __mptcp_try_coalesce(struct sock *sk, struct sk_buff *to,
 {
 	int limit = READ_ONCE(sk->sk_rcvbuf);
 
-	if (unlikely(MPTCP_SKB_CB(to)->cant_coalesce) ||
+	if (MPTCP_SKB_CB(from)->map_seq != MPTCP_SKB_CB(to)->end_seq ||
+	    unlikely(MPTCP_SKB_CB(to)->cant_coalesce) ||
 	    MPTCP_SKB_CB(from)->offset ||
 	    ((to->len + from->len) > (limit >> 3)) ||
 	    !skb_try_coalesce(to, from, fragstolen, delta))
@@ -192,15 +193,6 @@ static bool mptcp_try_coalesce(struct sock *sk, struct sk_buff *to,
 	return true;
 }
 
-static bool mptcp_ooo_try_coalesce(struct mptcp_sock *msk, struct sk_buff *to,
-				   struct sk_buff *from)
-{
-	if (MPTCP_SKB_CB(from)->map_seq != MPTCP_SKB_CB(to)->end_seq)
-		return false;
-
-	return mptcp_try_coalesce((struct sock *)msk, to, from);
-}
-
 /* "inspired" by tcp_rcvbuf_grow(), main difference:
  * - mptcp does not maintain a msk-level window clamp
  * - returns true when  the receive buffer is actually updated
@@ -275,7 +267,7 @@ static void mptcp_data_queue_ofo(struct mptcp_sock *msk, struct sk_buff *skb)
 	/* with 2 subflows, adding at end of ooo queue is quite likely
 	 * Use of ooo_last_skb avoids the O(Log(N)) rbtree lookup.
 	 */
-	if (mptcp_ooo_try_coalesce(msk, msk->ooo_last_skb, skb)) {
+	if (mptcp_try_coalesce(sk, msk->ooo_last_skb, skb)) {
 		MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_OFOMERGE);
 		MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_OFOQUEUETAIL);
 		return;
@@ -321,7 +313,7 @@ static void mptcp_data_queue_ofo(struct mptcp_sock *msk, struct sk_buff *skb)
 				MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_DUPDATA);
 				goto merge_right;
 			}
-		} else if (mptcp_ooo_try_coalesce(msk, skb1, skb)) {
+		} else if (mptcp_try_coalesce(sk, skb1, skb)) {
 			MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_OFOMERGE);
 			return;
 		}
@@ -672,8 +664,7 @@ static void __mptcp_add_backlog(struct sock *sk,
 	if (!list_empty(&msk->backlog_list))
 		tail = list_last_entry(&msk->backlog_list, struct sk_buff, list);
 
-	if (tail && MPTCP_SKB_CB(skb)->map_seq == MPTCP_SKB_CB(tail)->end_seq &&
-	    ssk == tail->sk &&
+	if (tail && ssk == tail->sk &&
 	    __mptcp_try_coalesce(sk, tail, skb, &fragstolen, &delta)) {
 		skb->truesize -= delta;
 		kfree_skb_partial(skb, fragstolen);
@@ -797,7 +788,7 @@ static bool __mptcp_ofo_queue(struct mptcp_sock *msk)
 
 		end_seq = MPTCP_SKB_CB(skb)->end_seq;
 		tail = skb_peek_tail(&sk->sk_receive_queue);
-		if (!tail || !mptcp_ooo_try_coalesce(msk, tail, skb)) {
+		if (!tail || !mptcp_try_coalesce(sk, tail, skb)) {
 			int delta = msk->ack_seq - MPTCP_SKB_CB(skb)->map_seq;
 
 			/* skip overlapping data, if any */
Signature · 1 line
2.53.0
author maintainer
#3 lore raw

PATCH v2 3/10 mptcp: drop the cant_coalesce CB field

In reply to Paolo Abeni

Such field is used to ensure in-sequence processing in case of fastopen. Instead let's perform synchronization of the fastopen skb sequence when the IASN becomes available with the 3rd ack.

When the cant_coalesce field has been introduced, commit f03afb3aeb9d ("mptcp: drop __mptcp_fastopen_gen_msk_ackseq()") noted that updating the already queued skb for passive fastopen socket at 3rd ack time would be difficult and race prone. The main point is that such update don't need to be synchronously performed at 3rd ack time, but is sufficient to perform it before the next segment is introduced into the msk.

To such extent, add an explicit test in __mptcp_move_skb(). Performance wise this trades a conditional in the fast path - in __mptcp_try_coalesce()
- with a similar one in __mptcp_move_skb() and a couple more in slow paths.

After this change the user-space will always observe consistent sequence numbers in the receive queue, even in the TFO dummy mapping case.

---
 net/mptcp/fastopen.c |  2 +-
 net/mptcp/protocol.c | 28 ++++++++++++++++++++++++++--
 net/mptcp/protocol.h |  4 +++-
 net/mptcp/subflow.c  |  7 +++++++
 4 files changed, 37 insertions(+), 4 deletions(-)
net/mptcp/fastopen.c +1 -1
index 82ec15bcfd7f..40168adfed22 100644
@@ -46,11 +46,11 @@ void mptcp_fastopen_subflow_synack_set_params(struct mptcp_subflow_context *subf
 	MPTCP_SKB_CB(skb)->end_seq = 0;
 	MPTCP_SKB_CB(skb)->offset = 0;
 	MPTCP_SKB_CB(skb)->has_rxtstamp = TCP_SKB_CB(skb)->has_rxtstamp;
-	MPTCP_SKB_CB(skb)->cant_coalesce = 1;
 
 	mptcp_data_lock(sk);
 	DEBUG_NET_WARN_ON_ONCE(sock_owned_by_user_nocheck(sk));
 
+	mptcp_sk(sk)->rcvd_dummy_seq = true;
 	mptcp_borrow_fwdmem(sk, skb);
 	skb_set_owner_r(skb, sk);
 	__skb_queue_tail(&sk->sk_receive_queue, skb);
net/mptcp/protocol.c +26 -2
index ad0a289b544b..fd88a81f1821 100644
@@ -160,7 +160,6 @@ static bool __mptcp_try_coalesce(struct sock *sk, struct sk_buff *to,
 	int limit = READ_ONCE(sk->sk_rcvbuf);
 
 	if (MPTCP_SKB_CB(from)->map_seq != MPTCP_SKB_CB(to)->end_seq ||
-	    unlikely(MPTCP_SKB_CB(to)->cant_coalesce) ||
 	    MPTCP_SKB_CB(from)->offset ||
 	    ((to->len + from->len) > (limit >> 3)) ||
 	    !skb_try_coalesce(to, from, fragstolen, delta))
@@ -357,7 +356,6 @@ static void mptcp_init_skb(struct sock *ssk, struct sk_buff *skb, int offset,
 	MPTCP_SKB_CB(skb)->end_seq = MPTCP_SKB_CB(skb)->map_seq + copy_len;
 	MPTCP_SKB_CB(skb)->offset = offset;
 	MPTCP_SKB_CB(skb)->has_rxtstamp = has_rxtstamp;
-	MPTCP_SKB_CB(skb)->cant_coalesce = 0;
 
 	__skb_unlink(skb, &ssk->sk_receive_queue);
 
@@ -365,6 +363,24 @@ static void mptcp_init_skb(struct sock *ssk, struct sk_buff *skb, int offset,
 	skb_dst_drop(skb);
 }
 
+void __mptcp_sync_rcv_sequence(struct sock *sk)
+{
+	struct mptcp_sock *msk = mptcp_sk(sk);
+	struct sk_buff *skb;
+
+	if (likely(!msk->rcvd_dummy_seq))
+		return;
+
+	/* User space can have already received the TFO skb. */
+	msk->rcvd_dummy_seq = false;
+	skb = skb_peek_tail(&sk->sk_receive_queue);
+	if (!skb)
+		return;
+
+	MPTCP_SKB_CB(skb)->map_seq = msk->ack_seq - skb->len;
+	MPTCP_SKB_CB(skb)->end_seq = msk->ack_seq;
+}
+
 static bool __mptcp_move_skb(struct sock *sk, struct sk_buff *skb)
 {
 	u64 copy_len = MPTCP_SKB_CB(skb)->end_seq - MPTCP_SKB_CB(skb)->map_seq;
@@ -373,6 +389,12 @@ static bool __mptcp_move_skb(struct sock *sk, struct sk_buff *skb)
 
 	mptcp_borrow_fwdmem(sk, skb);
 
+	/* Be sure to sync the eventual fastopen dummy mapping before any other
+	 * skb lands into the msk.
+	 */
+	if (unlikely(msk->rcvd_dummy_seq))
+		__mptcp_sync_rcv_sequence(sk);
+
 	if (MPTCP_SKB_CB(skb)->map_seq == msk->ack_seq) {
 		/* in sequence */
 		msk->bytes_received += copy_len;
@@ -3682,6 +3704,8 @@ static void mptcp_release_cb(struct sock *sk)
 			__mptcp_error_report(sk);
 		if (__test_and_clear_bit(MPTCP_SYNC_SNDBUF, &msk->cb_flags))
 			__mptcp_sync_sndbuf(sk);
+		if (__test_and_clear_bit(MPTCP_SYNC_SEQ, &msk->cb_flags))
+			__mptcp_sync_rcv_sequence(sk);
 	}
 }
 
net/mptcp/protocol.h +3 -1
index 661600f8b573..16a1f4531dad 100644
@@ -124,13 +124,13 @@
 #define MPTCP_FLUSH_JOIN_LIST	5
 #define MPTCP_SYNC_STATE	6
 #define MPTCP_SYNC_SNDBUF	7
+#define MPTCP_SYNC_SEQ		8
 
 struct mptcp_skb_cb {
 	u64 map_seq;
 	u64 end_seq;
 	u32 offset;
 	u8  has_rxtstamp;
-	u8  cant_coalesce;
 };
 
 #define MPTCP_SKB_CB(__skb)	((struct mptcp_skb_cb *)&((__skb)->cb[0]))
@@ -310,6 +310,7 @@ struct mptcp_sock {
 	u32		token;
 	unsigned long	flags;
 	unsigned long	cb_flags;
+	bool		rcvd_dummy_seq;
 	bool		recovery;		/* closing subflow write queue reinjected */
 	bool		can_ack;
 	bool		fully_established;
@@ -1172,6 +1173,7 @@ void mptcp_event_pm_listener(const struct sock *ssk,
 			     enum mptcp_event_type event);
 bool mptcp_userspace_pm_active(const struct mptcp_sock *msk);
 
+void __mptcp_sync_rcv_sequence(struct sock *sk);
 void mptcp_fastopen_subflow_synack_set_params(struct mptcp_subflow_context *subflow,
 					      struct request_sock *req);
 int mptcp_pm_genl_fill_addr(struct sk_buff *msg,
net/mptcp/subflow.c +7 -0
index c57ed27a5fb0..b226c7cd1b79 100644
@@ -478,6 +478,8 @@ static void subflow_set_remote_key(struct mptcp_sock *msk,
 				   struct mptcp_subflow_context *subflow,
 				   const struct mptcp_options_received *mp_opt)
 {
+	struct sock *sk = (struct sock *)msk;
+
 	/* active MPC subflow will reach here multiple times:
 	 * at subflow_finish_connect() time and at 4th ack time
 	 */
@@ -496,6 +498,11 @@ static void subflow_set_remote_key(struct mptcp_sock *msk,
 	WRITE_ONCE(msk->ack_seq, subflow->iasn);
 	WRITE_ONCE(msk->can_ack, true);
 	atomic64_set(&msk->rcv_wnd_sent, subflow->iasn);
+
+	if (!sock_owned_by_user(sk))
+		__mptcp_sync_rcv_sequence(sk);
+	else
+		__set_bit(MPTCP_SYNC_SEQ, &msk->cb_flags);
 }
 
 static void mptcp_propagate_state(struct sock *sk, struct sock *ssk,
Signature · 1 line
2.53.0
author maintainer
#4 lore raw

PATCH v2 4/10 mptcp: remove CB offset field

In reply to Paolo Abeni

Instead, use a new msk-level field to track the bytes already consumed inside each skb, carring the amount of bytes already copied to user-space, alike what TCP is already doing.

copied_seq is always accessed under the msk socket lock, delegating the initialization to the msk release cb, when the socket is owned by the user-space at remote key reception time.

This simplify a bit the __mptcp_recvmsg_mskq() and mptcp_inq_hint() code and will also make possible the next patch.

---
v1 -> v2:
- deal correctly with peek, as usally "inspired" from the correspondent tcp code
- update mptcp_inq_hint(), too

Note: this has the potential to break almost everything. On the flip side the CB->offset vs copied_seq difference from TCP is quite confusing and removing it will be for the good.

Also this explicitly relays on "mptcp: do not drop partial packets" to avoid dropping partially consumed packets

---
 net/mptcp/fastopen.c |  5 +--
 net/mptcp/protocol.c | 89 +++++++++++++++++++-------------------------
 net/mptcp/protocol.h |  2 +-
 net/mptcp/subflow.c  |  6 ++-
 4 files changed, 46 insertions(+), 56 deletions(-)
net/mptcp/fastopen.c +2 -3
index 40168adfed22..cbe2a6192002 100644
@@ -42,9 +42,8 @@ void mptcp_fastopen_subflow_synack_set_params(struct mptcp_subflow_context *subf
 	subflow->ssn_offset += skb->len;
 
 	/* Only the sequence delta is relevant */
-	MPTCP_SKB_CB(skb)->map_seq = -skb->len;
-	MPTCP_SKB_CB(skb)->end_seq = 0;
-	MPTCP_SKB_CB(skb)->offset = 0;
+	MPTCP_SKB_CB(skb)->map_seq = 0;
+	MPTCP_SKB_CB(skb)->end_seq = skb->len;
 	MPTCP_SKB_CB(skb)->has_rxtstamp = TCP_SKB_CB(skb)->has_rxtstamp;
 
 	mptcp_data_lock(sk);
net/mptcp/protocol.c +39 -50

205 lines not on this page · raw message

net/mptcp/protocol.h +1 -1
index 16a1f4531dad..68bedb60871f 100644
@@ -129,7 +129,6 @@
 struct mptcp_skb_cb {
 	u64 map_seq;
 	u64 end_seq;
-	u32 offset;
 	u8  has_rxtstamp;
 };
 
@@ -289,6 +288,7 @@ struct mptcp_sock {
 	u64		bytes_sent;
 	u64		snd_nxt;
 	u64		bytes_received;
+	u64		copied_seq;
 	u64		ack_seq;
 	atomic64_t	rcv_wnd_sent;
 	u64		rcv_data_fin_seq;
net/mptcp/subflow.c +4 -2
index b226c7cd1b79..6d3d0106749f 100644
@@ -499,10 +499,12 @@ static void subflow_set_remote_key(struct mptcp_sock *msk,
 	WRITE_ONCE(msk->can_ack, true);
 	atomic64_set(&msk->rcv_wnd_sent, subflow->iasn);
 
-	if (!sock_owned_by_user(sk))
+	if (!sock_owned_by_user(sk)) {
+		msk->copied_seq = subflow->iasn;
 		__mptcp_sync_rcv_sequence(sk);
-	else
+	} else {
 		__set_bit(MPTCP_SYNC_SEQ, &msk->cb_flags);
+	}
 }
 
 static void mptcp_propagate_state(struct sock *sk, struct sock *ssk,
Signature · 1 line
2.53.0
author maintainer
#5 lore raw

PATCH v2 5/10 mptcp: sync mptcp skb cb layout with tcp one

In reply to Paolo Abeni

The MPTCP protocol uses a significantly different CB layout WRT TCP, as it includes different information and use 64 bits for the sequence numbers.

As the msk-level rcvbuf buffer size is limited by the core socket code the INT_MAX; after validating the incoming skb vs the current receive window, we can safely use 32 bits for MPTCP-level sequence number. This allow updating the MPTCP CB layout so that fields with a corresponding TCP-level data use the same area inside the CB itself.

Add build time check to ensure the latter invariant.

---
v1 -> v2:
- use u64 for admission checks

rfc -> v1:
- keep ack_seq up2date

---
 net/mptcp/fastopen.c |  2 ++
 net/mptcp/protocol.c | 78 ++++++++++++++++++++++++++++----------------
 net/mptcp/protocol.h |  7 ++--
 3 files changed, 56 insertions(+), 31 deletions(-)
net/mptcp/fastopen.c +2 -0
index cbe2a6192002..f65312b41b95 100644
@@ -42,7 +42,9 @@ void mptcp_fastopen_subflow_synack_set_params(struct mptcp_subflow_context *subf
 	subflow->ssn_offset += skb->len;
 
 	/* Only the sequence delta is relevant */
+	MPTCP_SKB_CB(skb)->map_seq64 = 0;
 	MPTCP_SKB_CB(skb)->map_seq = 0;
+	MPTCP_SKB_CB(skb)->flags = 0;
 	MPTCP_SKB_CB(skb)->end_seq = skb->len;
 	MPTCP_SKB_CB(skb)->has_rxtstamp = TCP_SKB_CB(skb)->has_rxtstamp;
 
net/mptcp/protocol.c +49 -29

200 lines not on this page · raw message

net/mptcp/protocol.h +5 -2
index 68bedb60871f..e4569b3af744 100644
@@ -127,9 +127,12 @@
 #define MPTCP_SYNC_SEQ		8
 
 struct mptcp_skb_cb {
-	u64 map_seq;
-	u64 end_seq;
+	u32 map_seq;
+	u32 end_seq;
+	u32 unused;
+	u16 flags;
 	u8  has_rxtstamp;
+	u64 map_seq64;
 };
 
 #define MPTCP_SKB_CB(__skb)	((struct mptcp_skb_cb *)&((__skb)->cb[0]))
Signature · 1 line
2.53.0
author maintainer
#6 lore raw

PATCH v2 6/10 tcp: expose the tcp_collapse_ofo_queue() helper to mptcp usage, too

In reply to Paolo Abeni

The end goal is to avoid duplicating the quite untrivial strategy at MPTCP level.

After the previous patch, the mentioned helpers could process skbs standing in MPTCP-level queues without any CB-related adaptation.

The only additional adjustment needed is explicitly providing the OoO queue reference, to cope with different sk layout.

Additionally rename the helper to clearly document its hybrid nature and let it return the number of collapsed skbs, to allow proper accounting from the future MPTCP caller.

---
rfc -> v1:
- fix arg typo

Note:
- this will need a significant amount of testing at the TCP level and explicit approval from Eric, which I can't guess if we can hope.

---
 include/net/tcp.h    |  8 +++++++
 net/ipv4/tcp_input.c | 55 ++++++++++++++++++++++++++++----------------
 2 files changed, 43 insertions(+), 20 deletions(-)
include/net/tcp.h +8 -0
index 6156d1d068e1..34a96f0bcf0a 100644
@@ -1828,6 +1828,14 @@ extern void tcp_openreq_init_rwin(struct request_sock *req,
 
 void tcp_enter_memory_pressure(struct sock *sk);
 void tcp_leave_memory_pressure(struct sock *sk);
+unsigned int xtcp_collapse(struct sock *sk, struct sk_buff_head *list,
+			   struct rb_root *root, struct sk_buff *head,
+			   struct sk_buff *tail, u32 start, u32 end,
+			   u8 scaling_ratio);
+unsigned int xtcp_collapse_ofo_queue(struct sock *sk,
+				     struct rb_root *out_of_order_queue,
+				     struct sk_buff **ooo_last_skb,
+				     u8 scaling_ratio);
 
 static inline int keepalive_intvl_when(const struct tcp_sock *tp)
 {
net/ipv4/tcp_input.c +35 -20

130 lines not on this page · raw message

Signature · 1 line
2.53.0
author maintainer
#7 lore raw

PATCH v2 7/10 mptcp: implemented OoO queue pruning

In reply to Paolo Abeni

Leverage the hybrid helpers to implement the receive queue and OoO queue collapsing at ingress time when reaching memory bounds.

If the msk is owned by the user-space at incoming skb time, perform the pruning in the release_cb. The prune check is additionally performed when the skb reaches the msk-level queues.

---
v1 -> v2:
- collapse rcv queue, too
- deal with MPC map, too
- drop left-over sentence in the commit message

RFC -> v1:
- use data_seq only when available
- avoid ack_seq lockless access
- drop limit on fallback
- collapse rcvqueue, too
- drop only when pruning is not possible and over rcvbuf * 2

Notes:
- Similarly to path 'mptcp: move checks vs rcvbuf size earlier in the RX path', some cleanup/tuning in mptcp_over_limit() will be needed
- Pruning in the release_cb() is likely not needed, should probably be removed (after more testing).

---
 net/mptcp/mib.c      |  3 ++
 net/mptcp/mib.h      |  3 ++
 net/mptcp/options.c  | 36 ++++++++++++++++++++---
 net/mptcp/protocol.c | 69 ++++++++++++++++++++++++++++++++++++++++++++
 net/mptcp/protocol.h |  2 ++
 5 files changed, 109 insertions(+), 4 deletions(-)
net/mptcp/mib.c +3 -0
index f23fda0c55a7..5128feec942c 100644
@@ -85,6 +85,9 @@ static const struct snmp_mib mptcp_snmp_list[] = {
 	SNMP_MIB_ITEM("SimultConnectFallback", MPTCP_MIB_SIMULTCONNFALLBACK),
 	SNMP_MIB_ITEM("FallbackFailed", MPTCP_MIB_FALLBACKFAILED),
 	SNMP_MIB_ITEM("WinProbe", MPTCP_MIB_WINPROBE),
+	SNMP_MIB_ITEM("OfoPruned", MPTCP_MIB_OFO_PRUNED),
+	SNMP_MIB_ITEM("RcvPruned", MPTCP_MIB_RCVPRUNED),
+	SNMP_MIB_ITEM("RcvCollapsed", MPTCP_MIB_RCVCOLLAPSED),
 };
 
 /* mptcp_mib_alloc - allocate percpu mib counters
net/mptcp/mib.h +3 -0
index 812218b5ed2b..2f8f68e33ac5 100644
@@ -88,6 +88,9 @@ enum linux_mptcp_mib_field {
 	MPTCP_MIB_SIMULTCONNFALLBACK,	/* Simultaneous connect */
 	MPTCP_MIB_FALLBACKFAILED,	/* Can't fallback due to msk status */
 	MPTCP_MIB_WINPROBE,		/* MPTCP-level zero window probe */
+	MPTCP_MIB_OFO_PRUNED,		/* MPTCP-level OoO queue pruned */
+	MPTCP_MIB_RCVPRUNED,		/* Dropped due to memory constrains */
+	MPTCP_MIB_RCVCOLLAPSED,		/* Collapsed due to memory pressure */
 	__MPTCP_MIB_MAX
 };
 
net/mptcp/options.c +32 -4
index ad4bb6fd86e1..0c1d1d4da88a 100644
@@ -1159,8 +1159,12 @@ static bool add_addr_hmac_valid(struct mptcp_sock *msk,
 }
 
 static bool mptcp_over_limit(struct sock *sk, const struct sock *ssk,
-			     const struct sk_buff *skb)
+			     const struct sk_buff *skb,
+			     const struct mptcp_options_received *mp_opt)
 {
+	struct mptcp_sock *msk = mptcp_sk(sk);
+	bool ret;
+
 	if (likely(sk_rmem_alloc_get(sk) <= READ_ONCE(sk->sk_rcvbuf)))
 		return false;
 
@@ -1170,7 +1174,27 @@ static bool mptcp_over_limit(struct sock *sk, const struct sock *ssk,
 	    !after(TCP_SKB_CB(skb)->end_seq, tcp_sk(ssk)->rcv_nxt))
 		return false;
 
-	return true;
+	mptcp_data_lock(sk);
+	if (!sock_owned_by_user(sk)) {
+		/* When the data sequence is not (yet) available for the
+		 * incoming skb, allow pruning the whole OoO queue.
+		 */
+		u32 seq = !mp_opt->use_map || mp_opt->mpc_map ? msk->ack_seq :
+			  mp_opt->data_seq;
+
+		__mptcp_check_prune(sk, seq);
+		ret = sk_rmem_alloc_get(sk) > READ_ONCE(sk->sk_rcvbuf);
+	} else {
+		u64 limit = ((u64)READ_ONCE(sk->sk_rcvbuf)) << 1;
+
+		/* Pruning will take place later in the RX path, allow
+		 * some extra slack.
+		 */
+		ret = sk_rmem_alloc_get(sk) > limit;
+		__set_bit(MPTCP_PRUNE, &msk->cb_flags);
+	}
+	mptcp_data_unlock(sk);
+	return ret;
 }
 
 /* Return false when the caller must drop the packet, i.e. in case of error,
@@ -1201,7 +1225,11 @@ bool mptcp_incoming_options(struct sock *sk, struct sk_buff *skb)
 		__mptcp_data_acked(subflow->conn);
 		mptcp_data_unlock(subflow->conn);
 
-		if (mptcp_over_limit(subflow->conn, sk, skb))
+		/* Will use ack_seq as limit for OoO pruning; any value would do
+		 * as OoO queue must be empty.
+		 */
+		mp_opt.use_map = 0;
+		if (mptcp_over_limit(subflow->conn, sk, skb, &mp_opt))
 			return false;
 		return true;
 	}
@@ -1281,7 +1309,7 @@ bool mptcp_incoming_options(struct sock *sk, struct sk_buff *skb)
 		return true;
 	}
 
-	if (mptcp_over_limit(subflow->conn, sk, skb))
+	if (mptcp_over_limit(subflow->conn, sk, skb, &mp_opt))
 		return false;
 
 	mpext = skb_ext_add(skb, SKB_EXT_MPTCP);
net/mptcp/protocol.c +69 -0

87 lines not on this page · raw message

net/mptcp/protocol.h +2 -0
index e4569b3af744..1116a402771d 100644
@@ -125,6 +125,7 @@
 #define MPTCP_SYNC_STATE	6
 #define MPTCP_SYNC_SNDBUF	7
 #define MPTCP_SYNC_SEQ		8
+#define MPTCP_PRUNE		9
 
 struct mptcp_skb_cb {
 	u32 map_seq;
@@ -831,6 +832,7 @@ bool __mptcp_close(struct sock *sk, long timeout);
 void mptcp_cancel_work(struct sock *sk);
 void __mptcp_unaccepted_force_close(struct sock *sk);
 void mptcp_set_state(struct sock *sk, int state);
+bool __mptcp_check_prune(struct sock *sk, u32 seq);
 
 bool mptcp_addresses_equal(const struct mptcp_addr_info *a,
 			   const struct mptcp_addr_info *b, bool use_port);
Signature · 1 line
2.53.0
author maintainer
#8 lore raw

PATCH v2 8/10 mptcp: track prune recovery status

In reply to Paolo Abeni

After dropping any data already acked at the TCP level, the MPTCP must avoid inducing TCP-level retransmission until the pruned data has been successfully acked at MPTCP level. Otherwise the subflows could keep retransmitting skbs carring OoO MPTCP data, preventing reinjections and stalling completely the data transfer.

Explicitly keep track of the highest pruned MPTCP-level seq number and stop dropping at TCP level until such sequence has been acked.

---
 net/mptcp/options.c  |  7 ++++++-
 net/mptcp/protocol.c | 14 +++++++++++++-
 net/mptcp/protocol.h |  3 +++
 net/mptcp/subflow.c  |  1 +
 4 files changed, 23 insertions(+), 2 deletions(-)
net/mptcp/options.c +6 -1
index 0c1d1d4da88a..2d050acad63b 100644
@@ -1194,7 +1194,12 @@ static bool mptcp_over_limit(struct sock *sk, const struct sock *ssk,
 		__set_bit(MPTCP_PRUNE, &msk->cb_flags);
 	}
 	mptcp_data_unlock(sk);
-	return ret;
+
+	/* After pruning any packets ensure that MPTCP-driven drops do not
+	 * cause TCP-level retransmission
+	 */
+	return ret &&
+	       !before(READ_ONCE(msk->ack_seq), READ_ONCE(msk->pruned_seq));
 }
 
 /* Return false when the caller must drop the packet, i.e. in case of error,
net/mptcp/protocol.c +13 -1
index 4137d587d3c5..3a6b0506d3a7 100644
@@ -394,12 +394,14 @@ static void mptcp_prune_ofo_queue(struct sock *sk, u32 seq)
 	struct mptcp_sock *msk = mptcp_sk(sk);
 	struct rb_node *node, *prev;
 	bool pruned = false;
+	u32 pruned_seq;
 
 	if (RB_EMPTY_ROOT(&msk->out_of_order_queue))
 		return;
 
 	node = &msk->ooo_last_skb->rbnode;
 
+	pruned_seq = msk->pruned_seq;
 	do {
 		struct sk_buff *skb = rb_to_skb(node);
 
@@ -410,16 +412,21 @@ static void mptcp_prune_ofo_queue(struct sock *sk, u32 seq)
 		pruned = true;
 		prev = rb_prev(node);
 		rb_erase(node, &msk->out_of_order_queue);
+		if (after(MPTCP_SKB_CB(skb)->end_seq, pruned_seq))
+			pruned_seq = MPTCP_SKB_CB(skb)->end_seq;
 		mptcp_drop(sk, skb);
 		msk->ooo_last_skb = rb_to_skb(prev);
+
 		if (atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf)
 			break;
 
 		node = prev;
 	} while (node);
 
-	if (pruned)
+	if (pruned) {
+		WRITE_ONCE(msk->pruned_seq, pruned_seq);
 		MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_OFO_PRUNED);
+	}
 }
 
 bool __mptcp_check_prune(struct sock *sk, u32 seq)
@@ -464,6 +471,8 @@ static bool __mptcp_move_skb(struct sock *sk, struct sk_buff *skb)
 		__mptcp_sync_rcv_sequence(sk);
 
 	if (__mptcp_check_prune(sk, MPTCP_SKB_CB(skb)->map_seq)) {
+		if (after(MPTCP_SKB_CB(skb)->end_seq, msk->pruned_seq))
+			WRITE_ONCE(msk->pruned_seq, MPTCP_SKB_CB(skb)->end_seq);
 		MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_RCVPRUNED);
 		mptcp_drop(sk, skb);
 		return false;
@@ -898,6 +907,8 @@ static bool __mptcp_ofo_queue(struct mptcp_sock *msk)
 		WRITE_ONCE(msk->ack_seq, msk->ack_seq + seq_delta);
 		moved = true;
 	}
+	if (after(msk->ack_seq, msk->pruned_seq))
+		WRITE_ONCE(msk->pruned_seq, (u32)msk->ack_seq);
 	return moved;
 }
 
@@ -3540,6 +3551,7 @@ static int mptcp_disconnect(struct sock *sk, int flags)
 	/* for fallback's sake */
 	WRITE_ONCE(msk->ack_seq, 0);
 	msk->copied_seq = 0;
+	WRITE_ONCE(msk->pruned_seq, 0);
 
 	WRITE_ONCE(sk->sk_shutdown, 0);
 	sk_error_report(sk);
net/mptcp/protocol.h +3 -0
index 1116a402771d..c369e0efe260 100644
@@ -307,6 +307,9 @@ struct mptcp_sock {
 	u64		bytes_acked;
 	u64		snd_una;
 	u64		wnd_end;
+	u32		pruned_seq;		/* If above ack_seq, highest
+						 * seq pruned.
+						 */
 	u32		last_data_sent;
 	u32		last_data_recv;
 	u32		last_ack_recv;
net/mptcp/subflow.c +1 -0
index 6d3d0106749f..2a719146ebce 100644
@@ -496,6 +496,7 @@ static void subflow_set_remote_key(struct mptcp_sock *msk,
 
 	WRITE_ONCE(msk->remote_key, subflow->remote_key);
 	WRITE_ONCE(msk->ack_seq, subflow->iasn);
+	WRITE_ONCE(msk->pruned_seq, subflow->iasn);
 	WRITE_ONCE(msk->can_ack, true);
 	atomic64_set(&msk->rcv_wnd_sent, subflow->iasn);
 
Signature · 1 line
2.53.0
author maintainer
#9 lore raw

PATCH v2 9/10 mptcp: move the retrans loop to a separate helper

In reply to Paolo Abeni

This is a cleanup in order to make the next patch simpler. No functional change intended.

---
 net/mptcp/protocol.c | 74 +++++++++++++++++++++++++-------------------
 1 file changed, 43 insertions(+), 31 deletions(-)
net/mptcp/protocol.c +43 -31
index 3a6b0506d3a7..64991a5ee206 100644
@@ -2840,41 +2840,14 @@ static void mptcp_check_fastclose(struct mptcp_sock *msk)
 	sk_error_report(sk);
 }
 
-static void __mptcp_retrans(struct sock *sk)
+/* Retransmit the specified data fragment on all the selected subflows. */
+static int __mptcp_push_retrans(struct sock *sk, struct mptcp_data_frag *dfrag)
 {
 	struct mptcp_sendmsg_info info = { .data_lock_held = true, };
 	struct mptcp_sock *msk = mptcp_sk(sk);
 	struct mptcp_subflow_context *subflow;
-	struct mptcp_data_frag *dfrag;
 	struct sock *ssk;
-	int ret, err;
-	u16 len = 0;
-
-	mptcp_clean_una_wakeup(sk);
-
-	/* first check ssk: need to kick "stale" logic */
-	err = mptcp_sched_get_retrans(msk);
-	dfrag = mptcp_rtx_head(sk);
-	if (!dfrag) {
-		if (mptcp_data_fin_enabled(msk)) {
-			struct inet_connection_sock *icsk = inet_csk(sk);
-
-			WRITE_ONCE(icsk->icsk_retransmits,
-				   icsk->icsk_retransmits + 1);
-			mptcp_set_datafin_timeout(sk);
-			mptcp_send_ack(msk);
-
-			goto reset_timer;
-		}
-
-		if (!mptcp_send_head(sk))
-			goto clear_scheduled;
-
-		goto reset_timer;
-	}
-
-	if (err)
-		goto reset_timer;
+	int ret, len = 0;
 
 	mptcp_for_each_subflow(msk, subflow) {
 		if (READ_ONCE(subflow->scheduled)) {
@@ -2902,7 +2875,7 @@ static void __mptcp_retrans(struct sock *sk)
 			    !msk->allow_subflows) {
 				spin_unlock_bh(&msk->fallback_lock);
 				release_sock(ssk);
-				goto clear_scheduled;
+				return -1;
 			}
 
 			while (info.sent < info.limit) {
@@ -2925,6 +2898,45 @@ static void __mptcp_retrans(struct sock *sk)
 			release_sock(ssk);
 		}
 	}
+	return len;
+}
+
+static void __mptcp_retrans(struct sock *sk)
+{
+	struct mptcp_sock *msk = mptcp_sk(sk);
+	struct mptcp_subflow_context *subflow;
+	struct mptcp_data_frag *dfrag;
+	int err, len;
+
+	mptcp_clean_una_wakeup(sk);
+
+	/* first check ssk: need to kick "stale" logic */
+	err = mptcp_sched_get_retrans(msk);
+	dfrag = mptcp_rtx_head(sk);
+	if (!dfrag) {
+		if (mptcp_data_fin_enabled(msk)) {
+			struct inet_connection_sock *icsk = inet_csk(sk);
+
+			WRITE_ONCE(icsk->icsk_retransmits,
+				   icsk->icsk_retransmits + 1);
+			mptcp_set_datafin_timeout(sk);
+			mptcp_send_ack(msk);
+
+			goto reset_timer;
+		}
+
+		if (!mptcp_send_head(sk))
+			goto clear_scheduled;
+
+		goto reset_timer;
+	}
+
+	if (err)
+		goto reset_timer;
+
+	len = __mptcp_push_retrans(sk, dfrag);
+	if (len < 0)
+		goto clear_scheduled;
 
 	msk->bytes_retrans += len;
 	dfrag->already_sent = max(dfrag->already_sent, len);
Signature · 1 line
2.53.0
author maintainer
#10 lore raw

PATCH v2 10/10 mptcp: let the retrans scheduler do its job.

In reply to Paolo Abeni

Currently the MPTCP core enforces that when MPTCP-level retrans timer fires, at most a single dfrag is retransmitted. If some corner-cases it may be necessary retransmit multiple dfrags, and the MPTCP socket will need to wait multiple retrans timeout to accomplish that.

Remove the mentioned constraint, allowing to transmit multiple dfrags per retrans period, as long as the scheduler keeps selecting subflows for retransmissions and pending data is available in the rtx queue. The default scheduler will transmit a dfrag per available subflow.

---
v1 -> v2:
- fix retrans sequence update (sashiko)

---
 net/mptcp/protocol.c | 82 ++++++++++++++++++++++++++------------------
 1 file changed, 48 insertions(+), 34 deletions(-)
net/mptcp/protocol.c +48 -34
index 64991a5ee206..a4b2d664af80 100644
@@ -1219,13 +1219,6 @@ static void __mptcp_clean_una_wakeup(struct sock *sk)
 	mptcp_write_space(sk);
 }
 
-static void mptcp_clean_una_wakeup(struct sock *sk)
-{
-	mptcp_data_lock(sk);
-	__mptcp_clean_una_wakeup(sk);
-	mptcp_data_unlock(sk);
-}
-
 static void mptcp_enter_memory_pressure(struct sock *sk)
 {
 	struct mptcp_subflow_context *subflow;
@@ -2840,7 +2833,10 @@ static void mptcp_check_fastclose(struct mptcp_sock *msk)
 	sk_error_report(sk);
 }
 
-/* Retransmit the specified data fragment on all the selected subflows. */
+/*
+ * Retransmit the specified data fragment on all the selected subflows,
+ * starting from the specified sequence
+ */
 static int __mptcp_push_retrans(struct sock *sk, struct mptcp_data_frag *dfrag)
 {
 	struct mptcp_sendmsg_info info = { .data_lock_held = true, };
@@ -2906,42 +2902,60 @@ static void __mptcp_retrans(struct sock *sk)
 	struct mptcp_sock *msk = mptcp_sk(sk);
 	struct mptcp_subflow_context *subflow;
 	struct mptcp_data_frag *dfrag;
+	u64 retrans_seq;
 	int err, len;
 
-	mptcp_clean_una_wakeup(sk);
-
-	/* first check ssk: need to kick "stale" logic */
-	err = mptcp_sched_get_retrans(msk);
-	dfrag = mptcp_rtx_head(sk);
-	if (!dfrag) {
-		if (mptcp_data_fin_enabled(msk)) {
-			struct inet_connection_sock *icsk = inet_csk(sk);
+	mptcp_data_lock(sk);
+	__mptcp_clean_una_wakeup(sk);
+	retrans_seq = msk->snd_una;
+	mptcp_data_unlock(sk);
 
-			WRITE_ONCE(icsk->icsk_retransmits,
-				   icsk->icsk_retransmits + 1);
-			mptcp_set_datafin_timeout(sk);
-			mptcp_send_ack(msk);
+	for (;;) {
+		/* first check ssk: need to kick "stale" logic */
+		err = mptcp_sched_get_retrans(msk);
+		dfrag = mptcp_rtx_head(sk);
+		if (!dfrag) {
+			if (mptcp_data_fin_enabled(msk)) {
+				struct inet_connection_sock *icsk;
+
+				icsk = inet_csk(sk);
+				WRITE_ONCE(icsk->icsk_retransmits,
+					   icsk->icsk_retransmits + 1);
+				mptcp_set_datafin_timeout(sk);
+				mptcp_send_ack(msk);
+				break;
+			}
 
-			goto reset_timer;
+			if (!mptcp_send_head(sk))
+				goto clear_scheduled;
+			break;
 		}
 
-		if (!mptcp_send_head(sk))
-			goto clear_scheduled;
-
-		goto reset_timer;
-	}
+		if (err)
+			break;
 
-	if (err)
-		goto reset_timer;
+		/* Skip the data already retransmitted in this run */
+		while (dfrag && !before64(retrans_seq, dfrag->data_seq +
+					  dfrag->data_len))
+			dfrag = list_is_last(&dfrag->list, &msk->rtx_queue) ?
+				NULL : list_next_entry(dfrag, list);
+		if (!dfrag || !dfrag->already_sent)
+			break;
 
-	len = __mptcp_push_retrans(sk, dfrag);
-	if (len < 0)
-		goto clear_scheduled;
+		len = __mptcp_push_retrans(sk, dfrag);
+		if (len < 0)
+			goto clear_scheduled;
 
-	msk->bytes_retrans += len;
-	dfrag->already_sent = max(dfrag->already_sent, len);
+		retrans_seq += len;
+		msk->bytes_retrans += len;
+		dfrag->already_sent = max(dfrag->already_sent, len);
 
-reset_timer:
+		/* Attempt the next fragment only if the current one is
+		 * completely retransmitted
+		 */
+		if (dfrag->already_sent < dfrag->data_len)
+			break;
+	}
 	mptcp_check_and_set_pending(sk);
 
 	if (!mptcp_rtx_timer_pending(sk))
Signature · 1 line
2.53.0
#11 lore raw
In reply to Paolo Abeni

Hi Paolo,

Thank you for your modifications, that's great!

Our CI did some validations and here is its report:

- KVM Validation: normal (except selftest_mptcp_join): Success! ✅
- KVM Validation: normal (only selftest_mptcp_join): Success! ✅
- KVM Validation: debug (except selftest_mptcp_join): Unstable: 2 failed test(s): packetdrill_dss selftest_mptcp_connect_checksum ⚠️
- KVM Validation: debug (only selftest_mptcp_join): Success! ✅
- 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/25017102382

Initiator: Patchew Applier
Commits: https://github.com/multipath-tcp/mptcp_net-next/commits/423dbedaff6a
Patchwork: https://patchwork.kernel.org/project/mptcp/list/?series=1086404

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-normal

For 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)

author maintainer
#12 lore raw
In reply to Paolo Abeni
Quoted 16 lines · Paolo Abeni

This an attempt to fix the data transfer stall reported by Geliang and Gang more carefully enforcing memory constraints at the MPTCP level.

Patch 1/10 moves the bound check before entering the TCP socket. Patch 2, 3, 4 and 5 are cleanups/refactors finalized to safely re-using TCP helpers on MPTCP skbs.
Patch 6 makes TCP pruning related helpers available to MPTCP and patch 7 makes use of them. Patch 8 addresses an edge scenario that could still lead to transfer stall under memory pressure.
Finally patch 9 and 10 improve the MPTCP-level retransmission schema to make recovery from memory pressure significanly faster.

Note that the diffstat is biases by the quite large patch 4/9, which contains mechanical transformation of existing code; "real" changes are noticiable smaller.

Quoted 11 lines

Tested successfully vs the test cases proposed by Geliang and Gang and vs the selftests.
---
Notes:
- Sashiko apparently can't understand that mptcp_data_lock() && sk not owned by the user-space is equivalent to the sk socket lock. As a consequences marks incorrectly as racy a few accesses under the data lock to the sk_rcvbuf.
- mptcp_join with "4 subflows with under pressure" may require longer than default timeout, as the transfer in such condition can be really slow.

Sashiko says:
---

Review

Quoted 11 lines
(quoted diff) +11 -0
+static bool mptcp_over_limit(struct sock *sk, const struct sock *ssk,
+			     const struct sk_buff *skb)
+{
+	if (likely(sk_rmem_alloc_get(sk) <= READ_ONCE(sk->sk_rcvbuf)))
+		return false;
+
+	/* Avoid silently dropping pure acks, fin or zero win probes. */
+	if (TCP_SKB_CB(skb)->seq == TCP_SKB_CB(skb)->end_seq ||
+	    TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN ||
+	    !after(TCP_SKB_CB(skb)->end_seq, tcp_sk(ssk)->rcv_nxt))
+		return false;

Will this silently drop 1-byte Zero Window Probes (ZWPs)? For a standard 1-byte ZWP where seq = rcv_nxt and end_seq = rcv_nxt + 1, it seems it would fail these conditions, returning true and dropping the packet. Also, if mptcp_incoming_options() returns false due to being over limit, does this cause tcp_data_queue() to execute __kfree_skb(skb) and return without sending or scheduling a TCP ACK? If so, could this bypass standard drop handlers (like goto out_of_window) and cause connection timeouts by not informing the sender of the zero window condition?
---

this is wrong: zero window probes use already acked sequence numbers.

---

net/mptcp/protocol.c static bool __mptcp_move_skbs_from_subflow(struct

Quoted 5 lines
net/mptcp/protocol.c +0 -0
index 718e910ff23f..77d7c7fbe8ef 100644
@@ -739,7 +739,7 @@ static bool __mptcp_move_skbs_from_subflow(struct

mptcp_sock *msk,

net/mptcp/protocol.c static bool __mptcp_move_skbs_from_subflow(struct · line 741

Quoted 5 lines
net/mptcp/protocol.c +1 -1
 			mptcp_init_skb(ssk, skb, offset, len);
 
-			if (own_msk && sk_rmem_alloc_get(sk) < sk->sk_rcvbuf) {
+			if (own_msk) {

By removing the sk_rmem_alloc_get() limit check here and moving the limit enforcement exclusively to mptcp_incoming_options() in the TCP slow path, can this bypass the MSK receive buffer limit for fallback connections taking the TCP fast path?
If an MPTCP connection falls back to plain TCP, it appears packets take the TCP fast path in tcp_rcv_established(), bypassing tcp_data_queue() and mptcp_incoming_options(). These packets would go directly through tcp_data_ready() to __mptcp_move_skbs_from_subflow(), which now unconditionally moves the packets to the MSK.
Since ownership is transferred to the MSK, the subflow's sk_rmem_alloc would remain near zero, preventing it from falling back to the slow path. Could this allow unbounded memory growth in the MSK queue if a sender transmits an infinite amount of data?
---

This is wrong, mptcp_incoming_options() is always invoked even for fallback msk, as msk explictly disables the header prediction.

---

net/mptcp/protocol.c static void mptcp_init_skb(struct sock *ssk,

Quoted 5 lines
net/mptcp/protocol.c +0 -0
index 28a7482b3887..26b9a23c63a9 100644
@@ -352,9 +350,9 @@ static void mptcp_init_skb(struct sock *ssk,

struct sk_buff *skb, int offset,

net/mptcp/protocol.c static void mptcp_init_skb(struct sock *ssk, · line 354

Quoted 8 lines
net/mptcp/protocol.c +2 -3
 	 * mptcp_subflow_get_mapped_dsn() is based on the current tp->copied_seq
 	 * value
 	 */
-	MPTCP_SKB_CB(skb)->map_seq = mptcp_subflow_get_mapped_dsn(subflow);
-	MPTCP_SKB_CB(skb)->end_seq = MPTCP_SKB_CB(skb)->map_seq + copy_len;
-	MPTCP_SKB_CB(skb)->offset = offset;
+	MPTCP_SKB_CB(skb)->map_seq = mptcp_subflow_get_mapped_dsn(subflow) -
+				     offset;

Could this cause packets with an offset greater than 0 to be incorrectly dropped?
If map_seq represents the sequence of the entire SKB rather than just the unread data, and __mptcp_move_skb() checks if MPTCP_SKB_CB(skb)->map_seq == msk->ack_seq, valid packets with an offset (due
to TCP retransmissions or overlapping segments) might fail the in-order check and be permanently dropped as old data.
---
this is wrong, the main point is that this series has a functional dependency on "mptcp: do not drop partial packets".

---

net/mptcp/protocol.c static struct sk_buff *mptcp_recv_skb(struct

Quoted 1 line
net/mptcp/protocol.c +0 -0
@@ -4364,7 +4355,7 @@ static struct sk_buff *mptcp_recv_skb(struct

sock *sk, u32 *off)

net/mptcp/protocol.c static struct sk_buff *mptcp_recv_skb(struct · line 4361

Quoted 8 lines
net/mptcp/protocol.c +1 -1
 		mptcp_move_skbs(sk);
 
 	while ((skb = skb_peek(&sk->sk_receive_queue)) != NULL) {
-		offset = MPTCP_SKB_CB(skb)->offset;
+		offset = msk->copied_seq - MPTCP_SKB_CB(skb)->map_seq;
 		if (offset < skb->len) {
 			*off = offset;
 			return skb;

Similar to the issue in mptcp_inq_hint(), if copied_seq evaluates to 0 due to
deferred initialization, could this offset calculation wrap around? Since map_seq is the crypto iasn, 0 - map_seq would wrap around to a large u32 value. This would cause offset < skb->len to evaluate to false and could lead to unconditionally dropping valid, unread SKBs via mptcp_eat_recv_skb() when called from __mptcp_read_sock().
---
This is wrong; due to __mptcp_sync_rcv_sequence() copied_seq can be zero only if map_seq is 0, too.

---
Also, what happens if __mptcp_push_retrans() successfully queues some data but returns a len less than dfrag->data_len?
The loop will continue and pick the same dfrag again since retrans_seq advanced but still falls within the fragment. However, looking at __mptcp_push_retrans(), it appears to always start from the beginning:
static int __mptcp_push_retrans(struct sock *sk, struct mptcp_data_frag *dfrag)

{
	...
	info.sent = 0;
	info.limit = READ_ONCE(msk->csum_enabled) ? dfrag->data_len :
						    dfrag->already_sent;
	while (info.sent < info.limit) {
		ret = mptcp_sendmsg_frag(sk, ssk, dfrag, &info);
	...
}

Since it doesn't take retrans_seq as an offset, will this incorrectly retransmit the first len bytes of the fragment multiple times, artificially incrementing retrans_seq until it exceeds the fragment bounds, and ultimately skipping the tail of the fragment entirely?
---
AFAICS here the LLM is allucinating a lot, as the main termination condition for that loopis dfrag->retrans_seq < than dfrag->data_len

The other comments looks valid and I'll try to address them in the next revision. I'm not sure how to guide sashiko on the above, I'll try adding more notes in the cover letter.

/P