diff mbox series

[RFC,v2,16/45] tcp: clean ext on tx recycle

Message ID	20191002233655.24323-17-mathew.j.martineau@linux.intel.com
State	RFC
Delegated to:	David Miller
Headers	show Return-Path: <netdev-owner@vger.kernel.org> From: Mat Martineau <mathew.j.martineau@linux.intel.com> To: netdev@vger.kernel.org, edumazet@google.com Cc: Paolo Abeni <pabeni@redhat.com>, cpaasch@apple.com, fw@strlen.de, peter.krystad@linux.intel.com, dcaratti@redhat.com, matthieu.baerts@tessares.net Subject: [RFC PATCH v2 16/45] tcp: clean ext on tx recycle Date: Wed, 2 Oct 2019 16:36:26 -0700 Message-Id: <20191002233655.24323-17-mathew.j.martineau@linux.intel.com> In-Reply-To: <20191002233655.24323-1-mathew.j.martineau@linux.intel.com> References: <20191002233655.24323-1-mathew.j.martineau@linux.intel.com> MIME-Version: 1.0 Content-Transfer-Encoding: 8bit Sender: netdev-owner@vger.kernel.org Precedence: bulk
Series	Multipath TCP \| expand [RFC,v2,00/45] Multipath TCP [RFC,v2,01/45] tcp: Add MPTCP option number [RFC,v2,02/45] net: Make sock protocol value checks more specific [RFC,v2,03/45] sock: Make sk_protocol a 16-bit value [RFC,v2,04/45] tcp: Define IPPROTO_MPTCP [RFC,v2,05/45] mptcp: Add MPTCP socket stubs [RFC,v2,06/45] mptcp: Handle MPTCP TCP options [RFC,v2,07/45] mptcp: Associate MPTCP context with TCP socket [RFC,v2,08/45] tcp: Expose tcp struct and routine for MPTCP [RFC,v2,09/45] mptcp: Handle MP_CAPABLE options for outgoing connections [RFC,v2,10/45] mptcp: add mptcp_poll [RFC,v2,11/45] tcp, ulp: Add clone operation to tcp_ulp_ops [RFC,v2,12/45] mptcp: Create SUBFLOW socket for incoming connections [RFC,v2,13/45] mptcp: Add key generation and token tree [RFC,v2,14/45] mptcp: Add shutdown() socket operation [RFC,v2,15/45] mptcp: Add setsockopt()/getsockopt() socket operations [RFC,v2,16/45] tcp: clean ext on tx recycle [RFC,v2,17/45] mptcp: Add MPTCP to skb extensions [RFC,v2,18/45] tcp: Prevent coalesce/collapse when skb has MPTCP extensions [RFC,v2,19/45] tcp: Export low-level TCP functions [RFC,v2,20/45] mptcp: Write MPTCP DSS headers to outgoing data packets [RFC,v2,21/45] mptcp: Implement MPTCP receive path [RFC,v2,22/45] mptcp: use sk_page_frag() in sendmsg [RFC,v2,23/45] mptcp: sendmsg() do spool all the provided data [RFC,v2,24/45] mptcp: allow collapsing consecutive sendpages on the same substream [RFC,v2,25/45] tcp: Check for filled TCP option space before SACK [RFC,v2,26/45] mptcp: Add path manager interface [RFC,v2,27/45] mptcp: Add ADD_ADDR handling [RFC,v2,28/45] mptcp: Add handling of incoming MP_JOIN requests [RFC,v2,29/45] mptcp: harmonize locking on all socket operations. [RFC,v2,30/45] mptcp: new sysctl to control the activation per NS [RFC,v2,31/45] mptcp: add basic kselftest for mptcp [RFC,v2,32/45] mptcp: Add handling of outgoing MP_JOIN requests [RFC,v2,33/45] mptcp: Implement path manager interface commands [RFC,v2,34/45] mptcp: Make MPTCP socket block/wakeup ignore sk_receive_queue [RFC,v2,35/45] mptcp: update per unacked sequence on pkt reception [RFC,v2,36/45] mptcp: queue data for mptcp level retransmission [RFC,v2,37/45] mptcp: introduce MPTCP retransmission timer [RFC,v2,38/45] mptcp: implement memory accounting for mptcp rtx queue [RFC,v2,39/45] mptcp: rework mptcp_sendmsg_frag to accept optional dfrag [RFC,v2,40/45] mptcp: implement and use MPTCP-level retransmission [RFC,v2,41/45] selftests: mptcp: make tc delays random [RFC,v2,42/45] selftests: mptcp: extend mptcp_connect tool for ipv6 family [RFC,v2,43/45] selftests: mptcp: add accept/getpeer checks [RFC,v2,44/45] selftests: mptcp: add ipv6 connectivity [RFC,v2,45/45] selftests: mptcp: random ethtool tweaking

Commit Message

Mat Martineau Oct. 2, 2019, 11:36 p.m. UTC

From: Paolo Abeni <pabeni@redhat.com>

Otherwise we will find stray/unexpected/old extensions
value on next iteration.

On tcp_write_xmit() we can end-up splitting an already queued
skb in two parts, via tso_fragment(). The newly created skb
can be allocated via the tx cache and the mptcp stack will not
be aware of it, so nobody set properly the MPTCP ext.

End result, we transmit the skb using an hold MPTCP DSS map
and that confuses the rx side/corrupt the stream. It requires
some concurrent conditions, so it's not deterministic.

Resetting the ext on recycle fixes all the current mptcp self tests
issues.

Apparently only MPTCP has issues with this kind of stray ext,
so an alternative would be add an additional mptcp hook in
tso_fragment() or in sk_stream_alloc_skb() to always init
the ext.

Signed-off-by: Paolo Abeni <pabeni@redhat.com>
---
 include/linux/skbuff.h | 8 ++++++++
 include/net/sock.h     | 1 +
 2 files changed, 9 insertions(+)

diff mbox series

Patch

diff --git a/include/linux/skbuff.h b/include/linux/skbuff.h
index e7d3b1a513ef..e7a7abd62026 100644
--- a/include/linux/skbuff.h
+++ b/include/linux/skbuff.h
@@ -4099,6 +4099,14 @@  static inline void skb_ext_put(struct sk_buff *skb)
 		__skb_ext_put(skb->extensions);
 }
 
+static inline void skb_ext_clear(struct sk_buff *skb)
+{
+	if (skb->active_extensions) {
+		__skb_ext_put(skb->extensions);
+		skb->active_extensions = 0;
+	}
+}
+
 static inline void __skb_ext_copy(struct sk_buff *dst,
 				  const struct sk_buff *src)
 {
diff --git a/include/net/sock.h b/include/net/sock.h
index ca2071555dde..b9a085d0bb18 100644
--- a/include/net/sock.h
+++ b/include/net/sock.h
@@ -1470,6 +1470,7 @@  static inline void sk_wmem_free_skb(struct sock *sk, struct sk_buff *skb)
 	sk_mem_uncharge(sk, skb->truesize);
 	if (static_branch_unlikely(&tcp_tx_skb_cache_key) &&
 	    !sk->sk_tx_skb_cache && !skb_cloned(skb)) {
+		skb_ext_clear(skb);
 		skb_zcopy_clear(skb, true);
 		sk->sk_tx_skb_cache = skb;
 		return;