Skip to contentSkip to conversation

mptcp: kselftest program and a few bugs

RFC PATCH In review Florian Westphal 2 participants 6 messages Feb 7, 2019 – Feb 21, 2019

Subsystems (3)
lore
Share & tools
lore

Florian Westphal

author
#0 lore raw

cover mptcp: kselftest program and a few bugs

Hi.

First patch adds a small mptcp test program, nothing fancy, just transmits stdin via mptcp connection and back out again.

Uncovers a few issues, the one unfixed is that
read(mptcpfd, buf, buflen)

can return cause data to be discarded.
In that case, read() returns a value larger than "buflen". Did not investigate further so far.

Other issues found are handled in the remaining patches.

I'll be attending the conf call later today, so we can continue discussion there if needed.

author
#1 lore raw

RFC PATCH 1/4 mptcp: add basic kselftest program

create mptcp connection between two processes, xmit data back and forth. Data is read from stdin and written (after traversing mtcp connection twice) to stdout.

Wrapper script tests that data has passed un-altered.

Will run automatically on "make kselftest".

Remaining problem:
read(mptcpfd, buf, buflen)

can return a value larger than buflen (remaining data is discarded in kernel). Plan to look into that next week.

Signed-off-by Florian Westphal <fw(a)strlen.de>
---
 tools/testing/selftests/Makefile              |   1 +
 tools/testing/selftests/net/mptcp/Makefile    |  11 +
 tools/testing/selftests/net/mptcp/config      |   1 +
 .../selftests/net/mptcp/mptcp_connect.c       | 403 ++++++++++++++++++
 .../selftests/net/mptcp/mptcp_connect.sh      |  29 ++
 5 files changed, 445 insertions(+)
 create mode 100644 tools/testing/selftests/net/mptcp/Makefile
 create mode 100644 tools/testing/selftests/net/mptcp/config
 create mode 100644 tools/testing/selftests/net/mptcp/mptcp_connect.c
 create mode 100755 tools/testing/selftests/net/mptcp/mptcp_connect.sh
tools/testing/selftests/Makefile +1 -0
index 24b9934fb269..3f905444d8ea 100644
@@ -24,6 +24,7 @@ TARGETS += memory-hotplug
 TARGETS += mount
 TARGETS += mqueue
 TARGETS += net
+TARGETS += net/mptcp
 TARGETS += netfilter
 TARGETS += networking/timestamping
 TARGETS += nsfs
tools/testing/selftests/net/mptcp/Makefile new file +11 -0
new file mode 100644
index 000000000000..0fc5d45055ee
@@ -0,0 +1,11 @@
+# SPDX-License-Identifier: GPL-2.0
+
+top_srcdir = ../../../../..
+
+CFLAGS =  -Wall -Wl,--no-as-needed -O2 -g
+
+TEST_PROGS := mptcp_connect.sh
+
+TEST_GEN_FILES = mptcp_connect
+
+include ../../lib.mk
tools/testing/selftests/net/mptcp/config new file +1 -0
new file mode 100644
index 000000000000..3bfe60494af8
@@ -0,0 +1 @@
+CONFIG_MPTCP=y
tools/testing/selftests/net/mptcp/mptcp_connect.c new file +403 -0

403 lines not on this page · raw message

tools/testing/selftests/net/mptcp/mptcp_connect.sh new file +29 -0
new file mode 100755
index 000000000000..2714c8a3f837
@@ -0,0 +1,29 @@
+#!/bin/bash
+
+tmpin=$(mktemp)
+tmpout=$(mktemp)
+
+cleanup()
+{
+	rm -f "$tmpin" "$tmpout"
+}
+
+trap cleanup EXIT
+
+SIZE=$((RANDOM % (1024 * 1024)))
+if [ $SIZE -eq 0 ] ; then
+	SIZE=1
+fi
+
+dd if=/dev/urandom of="$tmpin" bs=1 count=$SIZE 2> /dev/null
+./mptcp_connect 127.0.0.1 43212 < "$tmpin" > "$tmpout"
+
+cmp "$tmpin" "$tmpout"
+if [ $? -ne 0 ] ;then
+	echo "FAIL: Input and output differ" 1>&2
+	ls -l "$tmpin" "$tmpout" 1>&2
+	exit 1
+fi
+
+echo "PASS: Input of $SIZE passed though mptcp connection unchanged"
+exit 0
Signature · 1 line
2.19.2
author
#2 lore raw

RFC PATCH 2/4 mptcp: don't call mptcp_finish_connect with NULL subflow->conn

In reply to Florian Westphal

else, mptcp_connect.c gives:

BUG: KASAN: null-ptr-deref in mptcp_finish_connect+0x6d/0x310
Read of size 8 at addr 00000000000007b0 by task mptcp_connect/2095
Call Trace:
 mptcp_finish_connect+0x6d/0x310
 subflow_finish_connect+0x92/0x180
 tcp_finish_connect+0x79/0x210
 tcp_rcv_state_process+0x4e1/0x1f90
 tcp_v4_do_rcv+0x2ac/0x3d0
Signed-off-by Florian Westphal <fw(a)strlen.de>
---
 net/mptcp/subflow.c | 6 ++++--
 1 file changed, 4 insertions(+), 2 deletions(-)
net/mptcp/subflow.c +4 -2
index 3c9dca4be26a..e581eaf198ec 100644
@@ -124,8 +124,10 @@ static void subflow_finish_connect(struct sock *sk, const struct sk_buff *skb)
 
 	if (!subflow->conn_finished) {
 		pr_debug("remote_key=%llu", subflow->remote_key);
-		mptcp_finish_connect(subflow->conn, subflow->mp_capable);
-		subflow->conn_finished = 1;
+		if (subflow->conn) {
+			mptcp_finish_connect(subflow->conn, subflow->mp_capable);
+			subflow->conn_finished = 1;
+		}
 
 		if (skb) {
 			pr_debug("synack seq=%u", TCP_SKB_CB(skb)->seq);
Signature · 1 line
2.19.2
author
#3 lore raw

RFC PATCH 3/4 mptcp: add and use mptcp_subflow_hold

In reply to Florian Westphal

subflow sockets already have lifetime managed by RCU, so we can switch to atimic_inc_not_zero and skip/pretend we did not find such socket in the mptcp subflow list.

This is required to get rid of synchronize_rcu() from mptcp_close().

Signed-off-by Florian Westphal <fw(a)strlen.de>
---
 net/mptcp/protocol.c | 63 ++++++++++++++++++++++++++------------------
 1 file changed, 37 insertions(+), 26 deletions(-)
net/mptcp/protocol.c +37 -26
index 12857eeded8b..91db8a0c222d 100644
@@ -31,14 +31,35 @@ static inline bool before64(__u64 seq1, __u64 seq2)
 }
 #define after64(seq2, seq1)	before64(seq1, seq2)
 
+static bool mptcp_subflow_hold(struct subflow_sock *sk)
+{
+	struct sock *_sk = (struct sock *)sk;
+
+	return refcount_inc_not_zero(&_sk->sk_refcnt);
+}
+
+static struct sock *mptcp_subflow_get_ref(const struct mptcp_sock *msk)
+{
+	struct subflow_sock *subflow;
+
+	rcu_read_lock();
+	mptcp_for_each_subflow(msk, subflow) {
+		if (mptcp_subflow_hold(subflow)) {
+			rcu_read_unlock();
+			return (struct sock *)subflow;
+		}
+	}
+
+	rcu_read_unlock();
+	return NULL;
+}
+
 static int mptcp_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
 {
 	struct mptcp_sock *msk = mptcp_sk(sk);
 	int mss_now, size_goal, poffset, ret;
 	struct mptcp_skb_cb *mcb = NULL;
-	struct subflow_sock *subflow;
 	struct page *page = NULL;
-	struct hlist_node *node;
 	struct sk_buff *skb;
 	struct sock *ssk;
 	size_t psize;
@@ -49,20 +70,17 @@ static int mptcp_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
 		return sock_sendmsg(msk->subflow, msg);
 	}
 
-	rcu_read_lock();
-	node = rcu_dereference(hlist_first_rcu(&msk->conn_list));
-	subflow = hlist_entry(node, struct subflow_sock, node);
-	ssk = sock_sk(subflow);
-	sock_hold(ssk);
-	rcu_read_unlock();
+	ssk = mptcp_subflow_get_ref(msk);
+	if (!ssk)
+		return -ENOTCONN;
 
 	if (!msg_data_left(msg)) {
 		pr_debug("empty send");
-		ret = sock_sendmsg(sock_sk(subflow)->sk_socket, msg);
+		ret = sock_sendmsg(ssk->sk_socket, msg);
 		goto put_out;
 	}
 
-	pr_debug("conn_list->subflow=%p", subflow);
+	pr_debug("conn_list->subflow=%p", ssk);
 
 	if (msg->msg_flags & ~(MSG_MORE | MSG_DONTWAIT | MSG_NOSIGNAL)) {
 		ret = -ENOTSUPP;
@@ -279,7 +297,6 @@ static int mptcp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len,
 	struct mptcp_sock *msk = mptcp_sk(sk);
 	struct subflow_sock *subflow;
 	struct mptcp_read_arg arg;
-	struct hlist_node *node;
 	read_descriptor_t desc;
 	struct tcp_sock *tp;
 	struct sock *ssk;
@@ -291,13 +308,11 @@ static int mptcp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len,
 		return sock_recvmsg(msk->subflow, msg, flags);
 	}
 
-	rcu_read_lock();
-	node = rcu_dereference(hlist_first_rcu(&msk->conn_list));
-	subflow = hlist_entry(node, struct subflow_sock, node);
-	ssk = sock_sk(subflow);
-	sock_hold(ssk);
-	rcu_read_unlock();
+	ssk = mptcp_subflow_get_ref(msk);
+	if (!ssk)
+		return -ENOTCONN;
 
+	subflow = subflow_sk(ssk);
 	tp = tcp_sk(ssk);
 
 	desc.arg.data = &arg;
@@ -715,8 +730,6 @@ int mptcp_stream_connect(struct socket *sock, struct sockaddr *uaddr,
 int mptcp_stream_getname(struct socket *sock, struct sockaddr *uaddr, int peer)
 {
 	struct mptcp_sock *msk = mptcp_sk(sock->sk);
-	struct subflow_sock *subflow;
-	struct hlist_node *node;
 	struct sock *ssk;
 	int ret;
 
@@ -731,14 +744,12 @@ int mptcp_stream_getname(struct socket *sock, struct sockaddr *uaddr, int peer)
 	 * is connected and there are multiple subflows is not defined.
 	 * For now just use the first subflow on the list.
 	 */
-	rcu_read_lock();
-	node = rcu_dereference(hlist_first_rcu(&msk->conn_list));
-	subflow = hlist_entry(node, struct subflow_sock, node);
-	ssk = sock_sk(subflow);
-	sock_hold(ssk);
-	rcu_read_unlock();
 
-	ret = inet_getname(sock_sk(subflow)->sk_socket, uaddr, peer);
+	ssk = mptcp_subflow_get_ref(msk);
+	if (!ssk)
+		return -ENOTCONN;
+
+	ret = inet_getname(ssk->sk_socket, uaddr, peer);
 	sock_put(ssk);
 	return ret;
 }
Signature · 1 line
2.19.2
author
#4 lore raw

RFC PATCH 4/4 mptcp: add mptcp_poll

In reply to Florian Westphal

Can't use tcp_poll directly:

BUG: KASAN: slab-out-of-bounds in tcp_poll+0x17f/0x540
Read of size 4 at addr ffff88806ac5e50c by task mptcp_connect/2085
Call Trace:
 tcp_poll+0x17f/0x540
 sock_poll+0x152/0x180
Signed-off-by Florian Westphal <fw(a)strlen.de>

---
Should perhaps stop walk early in case __poll_t ret has all possible events already set.

 net/mptcp/protocol.c | 46 +++++++++++++++++++++++++++++++++++++++++++-
 1 file changed, 45 insertions(+), 1 deletion(-)
net/mptcp/protocol.c +45 -1
index 91db8a0c222d..f99d0dfabe18 100644
@@ -782,6 +782,50 @@ int mptcp_stream_accept(struct socket *sock, struct socket *newsock, int flags,
 	return inet_accept(sock, newsock, flags, kern);
 }
 
+static __poll_t mptcp_poll(struct file *file, struct socket *sock,
+			   struct poll_table_struct *wait)
+{
+	struct subflow_sock *subflow, *tmp;
+	const struct mptcp_sock *msk;
+	struct sock *sk = sock->sk;
+	__poll_t ret = 0;
+	unsigned int i;
+
+	msk = mptcp_sk(sk);
+	if (msk->subflow)
+		return tcp_poll(file, msk->subflow, wait);
+
+	i = 0;
+	for (;;) {
+		int j = 0;
+		tmp = NULL;
+
+		rcu_read_lock();
+		mptcp_for_each_subflow(msk, subflow) {
+			if (j < i) {
+				j++;
+				continue;
+			}
+
+			if (!mptcp_subflow_hold(subflow))
+				continue;
+
+			tmp = subflow;
+			i++;
+			break;
+		}
+		rcu_read_unlock();
+
+		if (!tmp)
+			break;
+
+		ret |= tcp_poll(file, sock_sk(tmp)->sk_socket, wait);
+		sock_put(sock_sk(tmp));
+	}
+
+	return ret;
+}
+
 int mptcp_stream_shutdown(struct socket *sock, int how)
 {
 	struct mptcp_sock *msk = mptcp_sk(sock->sk);
@@ -833,7 +877,7 @@ const struct proto_ops mptcp_stream_ops = {
 	.socketpair	   = sock_no_socketpair,
 	.accept		   = mptcp_stream_accept,
 	.getname	   = mptcp_stream_getname,
-	.poll		   = tcp_poll,
+	.poll		   = mptcp_poll,
 	.ioctl		   = inet_ioctl,
 	.listen		   = mptcp_stream_listen,
 	.shutdown	   = mptcp_stream_shutdown,
Signature · 1 line
2.19.2
#5 lore raw
In reply to Florian Westphal
Quoted 11 lines · Florian Westphal

Hi.

First patch adds a small mptcp test program, nothing fancy, just transmits stdin via mptcp connection and back out again.

Uncovers a few issues, the one unfixed is that
read(mptcpfd, buf, buflen)

can return cause data to be discarded.
In that case, read() returns a value larger than "buflen". Did not investigate further so far.

Florian -

Thanks for finding that read bug. This appears to fix it (I will update the patch set to include this):

net/mptcp/protocol.c +4 -1
index bed3be1c380b..46603af09b25 100644
@@ -319,6 +319,7 @@ static int mptcp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len,
  	while (copied < len) {
  		enum mapping_status status;
  		size_t discard_len = 0;
+		u32 map_remaining;
  		int bytes_read;
  		u64 ack_seq;
  		u64 old_ack;
@@ -411,7 +412,9 @@ static int mptcp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len,
  		}
 
  		/* Read mapped data */
-		desc.count = ssn - subflow->map_subflow_seq + subflow->map_dll;
+		map_remaining = (ssn - subflow->map_subflow_seq +
+				 subflow->map_dll);
+		desc.count = min_t(size_t, len - copied, map_remaining);
  		arg.msg = msg;
  		bytes_read = tcp_read_sock(ssk, &desc, mptcp_read_actor);
  		if (bytes_read < 0)

--
Mat Martineau
Intel