[PATCH net 0/3] net/tls: add a TX lock

[PATCH net 0/3] net/tls: add a TX lock

[Jakub Kicinski]

Hi!

Some time ago Pooja and Mallesham started reporting crashes with
an async accelerator. After trying to poke the existing logic into
shape I came to the conclusion that it can't be trusted, and to
preserve our sanity we should just add a lock around the TX side.

First patch removes the sk_write_pending checks from the write
space callbacks. Those don't seem to have a logical justification.

Patch 2 adds the TX lock and patch 3 associated test (which should
hang with current net).

Mallesham reports that even with these fixes applied the async
accelerator workload still occasionally hangs waiting for socket
memory. I suspect that's strictly related to the way async crypto
is integrated in TLS, so I think we should get these into net or
net-next and move from there.

Jakub Kicinski (3):
  net/tls: don't pay attention to sk_write_pending when pushing partial
    records
  net/tls: add a TX lock
  selftests/tls: add test for concurrent recv and send

 include/net/tls.h                 |   5 ++
 net/tls/tls_device.c              |  10 ++-
 net/tls/tls_main.c                |   2 +
 net/tls/tls_sw.c                  |  30 +++------
 tools/testing/selftests/net/tls.c | 108 ++++++++++++++++++++++++++++++
 5 files changed, 134 insertions(+), 21 deletions(-)

-- 
2.23.0

[PATCH net 1/3] net/tls: don't pay attention to sk_write_pending when pushing partial records

[Jakub Kicinski]

sk_write_pending being not zero does not guarantee that partial
record will be pushed. If the thread waiting for memory times out
the pending record may get stuck.

In case of tls_device there is no path where parial record is
set and writer present in the first place. Partial record is
set only in tls_push_sg() and tls_push_sg() will return an
error immediately. All tls_device callers of tls_push_sg()
will return (and not wait for memory) if it failed.

Fixes: a42055e8d2c3 ("net/tls: Add support for async encryption of records for performance")
Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Reviewed-by: Simon Horman <simon.horman@netronome.com>
---
 net/tls/tls_device.c | 4 +++-
 net/tls/tls_sw.c     | 9 +++------
 2 files changed, 6 insertions(+), 7 deletions(-)

diff --git a/net/tls/tls_device.c b/net/tls/tls_device.c
index f959487c5cd1..5a3715ddc592 100644
--- a/net/tls/tls_device.c
+++ b/net/tls/tls_device.c
@@ -623,9 +623,11 @@ static int tls_device_push_pending_record(struct sock *sk, int flags)
 
 void tls_device_write_space(struct sock *sk, struct tls_context *ctx)
 {
-	if (!sk->sk_write_pending && tls_is_partially_sent_record(ctx)) {
+	if (tls_is_partially_sent_record(ctx)) {
 		gfp_t sk_allocation = sk->sk_allocation;
 
+		WARN_ON_ONCE(sk->sk_write_pending);
+
 		sk->sk_allocation = GFP_ATOMIC;
 		tls_push_partial_record(sk, ctx,
 					MSG_DONTWAIT | MSG_NOSIGNAL |
diff --git a/net/tls/tls_sw.c b/net/tls/tls_sw.c
index c2b5e0d2ba1a..e155b792df0b 100644
--- a/net/tls/tls_sw.c
+++ b/net/tls/tls_sw.c
@@ -2180,12 +2180,9 @@ void tls_sw_write_space(struct sock *sk, struct tls_context *ctx)
 	struct tls_sw_context_tx *tx_ctx = tls_sw_ctx_tx(ctx);
 
 	/* Schedule the transmission if tx list is ready */
-	if (is_tx_ready(tx_ctx) && !sk->sk_write_pending) {
-		/* Schedule the transmission */
-		if (!test_and_set_bit(BIT_TX_SCHEDULED,
-				      &tx_ctx->tx_bitmask))
-			schedule_delayed_work(&tx_ctx->tx_work.work, 0);
-	}
+	if (is_tx_ready(tx_ctx) &&
+	    !test_and_set_bit(BIT_TX_SCHEDULED, &tx_ctx->tx_bitmask))
+		schedule_delayed_work(&tx_ctx->tx_work.work, 0);
 }
 
 void tls_sw_strparser_arm(struct sock *sk, struct tls_context *tls_ctx)
-- 
2.23.0

[PATCH net 2/3] net/tls: add a TX lock

[Jakub Kicinski]

TLS TX needs to release and re-acquire the socket lock if send buffer
fills up.

TLS SW TX path currently depends on only allowing one thread to enter
the function by the abuse of sk_write_pending. If another writer is
already waiting for memory no new ones are allowed in.

This has two problems:
 - writers don't wake other threads up when they leave the kernel;
   meaning that this scheme works for single extra thread (second
   application thread or delayed work) because memory becoming
   available will send a wake up request, but as Mallesham and
   Pooja report with larger number of threads it leads to threads
   being put to sleep indefinitely;
 - the delayed work does not get _scheduled_ but it may _run_ when
   other writers are present leading to crashes as writers don't
   expect state to change under their feet (same records get pushed
   and freed multiple times); it's hard to reliably bail from the
   work, however, because the mere presence of a writer does not
   guarantee that the writer will push pending records before exiting.

Ensuring wakeups always happen will make the code basically open
code a mutex. Just use a mutex.

The TLS HW TX path does not have any locking (not even the
sk_write_pending hack), yet it uses a per-socket sg_tx_data
array to push records.

Fixes: a42055e8d2c3 ("net/tls: Add support for async encryption of records for performance")
Reported-by: Mallesham  Jatharakonda <mallesh537@gmail.com>
Reported-by: Pooja Trivedi <poojatrivedi@gmail.com>
Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Reviewed-by: Simon Horman <simon.horman@netronome.com>
---
 include/net/tls.h    |  5 +++++
 net/tls/tls_device.c |  6 ++++++
 net/tls/tls_main.c   |  2 ++
 net/tls/tls_sw.c     | 21 +++++++--------------
 4 files changed, 20 insertions(+), 14 deletions(-)

diff --git a/include/net/tls.h b/include/net/tls.h
index c664e6dba0d1..794e297483ea 100644
--- a/include/net/tls.h
+++ b/include/net/tls.h
@@ -40,6 +40,7 @@
 #include <linux/socket.h>
 #include <linux/tcp.h>
 #include <linux/skmsg.h>
+#include <linux/mutex.h>
 #include <linux/netdevice.h>
 #include <linux/rcupdate.h>
 
@@ -269,6 +270,10 @@ struct tls_context {
 
 	bool in_tcp_sendpages;
 	bool pending_open_record_frags;
+
+	struct mutex tx_lock; /* protects partially_sent_* fields and
+			       * per-type TX fields
+			       */
 	unsigned long flags;
 
 	/* cache cold stuff */
diff --git a/net/tls/tls_device.c b/net/tls/tls_device.c
index 5a3715ddc592..683d00837693 100644
--- a/net/tls/tls_device.c
+++ b/net/tls/tls_device.c
@@ -523,8 +523,10 @@ static int tls_push_data(struct sock *sk,
 int tls_device_sendmsg(struct sock *sk, struct msghdr *msg, size_t size)
 {
 	unsigned char record_type = TLS_RECORD_TYPE_DATA;
+	struct tls_context *tls_ctx = tls_get_ctx(sk);
 	int rc;
 
+	mutex_lock(&tls_ctx->tx_lock);
 	lock_sock(sk);
 
 	if (unlikely(msg->msg_controllen)) {
@@ -538,12 +540,14 @@ int tls_device_sendmsg(struct sock *sk, struct msghdr *msg, size_t size)
 
 out:
 	release_sock(sk);
+	mutex_unlock(&tls_ctx->tx_lock);
 	return rc;
 }
 
 int tls_device_sendpage(struct sock *sk, struct page *page,
 			int offset, size_t size, int flags)
 {
+	struct tls_context *tls_ctx = tls_get_ctx(sk);
 	struct iov_iter	msg_iter;
 	char *kaddr = kmap(page);
 	struct kvec iov;
@@ -552,6 +556,7 @@ int tls_device_sendpage(struct sock *sk, struct page *page,
 	if (flags & MSG_SENDPAGE_NOTLAST)
 		flags |= MSG_MORE;
 
+	mutex_lock(&tls_ctx->tx_lock);
 	lock_sock(sk);
 
 	if (flags & MSG_OOB) {
@@ -568,6 +573,7 @@ int tls_device_sendpage(struct sock *sk, struct page *page,
 
 out:
 	release_sock(sk);
+	mutex_unlock(&tls_ctx->tx_lock);
 	return rc;
 }
 
diff --git a/net/tls/tls_main.c b/net/tls/tls_main.c
index ac88877dcade..0775ae40fcfb 100644
--- a/net/tls/tls_main.c
+++ b/net/tls/tls_main.c
@@ -267,6 +267,7 @@ void tls_ctx_free(struct sock *sk, struct tls_context *ctx)
 
 	memzero_explicit(&ctx->crypto_send, sizeof(ctx->crypto_send));
 	memzero_explicit(&ctx->crypto_recv, sizeof(ctx->crypto_recv));
+	mutex_destroy(&ctx->tx_lock);
 
 	if (sk)
 		kfree_rcu(ctx, rcu);
@@ -612,6 +613,7 @@ static struct tls_context *create_ctx(struct sock *sk)
 	if (!ctx)
 		return NULL;
 
+	mutex_init(&ctx->tx_lock);
 	rcu_assign_pointer(icsk->icsk_ulp_data, ctx);
 	ctx->sk_proto = sk->sk_prot;
 	return ctx;
diff --git a/net/tls/tls_sw.c b/net/tls/tls_sw.c
index e155b792df0b..446f23c1f3ce 100644
--- a/net/tls/tls_sw.c
+++ b/net/tls/tls_sw.c
@@ -897,15 +897,9 @@ int tls_sw_sendmsg(struct sock *sk, struct msghdr *msg, size_t size)
 	if (msg->msg_flags & ~(MSG_MORE | MSG_DONTWAIT | MSG_NOSIGNAL))
 		return -ENOTSUPP;
 
+	mutex_lock(&tls_ctx->tx_lock);
 	lock_sock(sk);
 
-	/* Wait till there is any pending write on socket */
-	if (unlikely(sk->sk_write_pending)) {
-		ret = wait_on_pending_writer(sk, &timeo);
-		if (unlikely(ret))
-			goto send_end;
-	}
-
 	if (unlikely(msg->msg_controllen)) {
 		ret = tls_proccess_cmsg(sk, msg, &record_type);
 		if (ret) {
@@ -1091,6 +1085,7 @@ int tls_sw_sendmsg(struct sock *sk, struct msghdr *msg, size_t size)
 	ret = sk_stream_error(sk, msg->msg_flags, ret);
 
 	release_sock(sk);
+	mutex_unlock(&tls_ctx->tx_lock);
 	return copied ? copied : ret;
 }
 
@@ -1114,13 +1109,6 @@ static int tls_sw_do_sendpage(struct sock *sk, struct page *page,
 	eor = !(flags & (MSG_MORE | MSG_SENDPAGE_NOTLAST));
 	sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk);
 
-	/* Wait till there is any pending write on socket */
-	if (unlikely(sk->sk_write_pending)) {
-		ret = wait_on_pending_writer(sk, &timeo);
-		if (unlikely(ret))
-			goto sendpage_end;
-	}
-
 	/* Call the sk_stream functions to manage the sndbuf mem. */
 	while (size > 0) {
 		size_t copy, required_size;
@@ -1219,15 +1207,18 @@ static int tls_sw_do_sendpage(struct sock *sk, struct page *page,
 int tls_sw_sendpage(struct sock *sk, struct page *page,
 		    int offset, size_t size, int flags)
 {
+	struct tls_context *tls_ctx = tls_get_ctx(sk);
 	int ret;
 
 	if (flags & ~(MSG_MORE | MSG_DONTWAIT | MSG_NOSIGNAL |
 		      MSG_SENDPAGE_NOTLAST | MSG_SENDPAGE_NOPOLICY))
 		return -ENOTSUPP;
 
+	mutex_lock(&tls_ctx->tx_lock);
 	lock_sock(sk);
 	ret = tls_sw_do_sendpage(sk, page, offset, size, flags);
 	release_sock(sk);
+	mutex_unlock(&tls_ctx->tx_lock);
 	return ret;
 }
 
@@ -2170,9 +2161,11 @@ static void tx_work_handler(struct work_struct *work)
 
 	if (!test_and_clear_bit(BIT_TX_SCHEDULED, &ctx->tx_bitmask))
 		return;
+	mutex_lock(&tls_ctx->tx_lock);
 	lock_sock(sk);
 	tls_tx_records(sk, -1);
 	release_sock(sk);
+	mutex_unlock(&tls_ctx->tx_lock);
 }
 
 void tls_sw_write_space(struct sock *sk, struct tls_context *ctx)
-- 
2.23.0

[PATCH net 3/3] selftests/tls: add test for concurrent recv and send

[Jakub Kicinski]

Add a test which spawns 16 threads and performs concurrent
send and recv calls on the same socket.

Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Reviewed-by: Simon Horman <simon.horman@netronome.com>
---
 tools/testing/selftests/net/tls.c | 108 ++++++++++++++++++++++++++++++
 1 file changed, 108 insertions(+)

diff --git a/tools/testing/selftests/net/tls.c b/tools/testing/selftests/net/tls.c
index 4c285b6e1db8..1c8f194d6556 100644
--- a/tools/testing/selftests/net/tls.c
+++ b/tools/testing/selftests/net/tls.c
@@ -898,6 +898,114 @@ TEST_F(tls, nonblocking)
 	}
 }
 
+static void
+test_mutliproc(struct __test_metadata *_metadata, struct _test_data_tls *self,
+	       bool sendpg, unsigned int n_readers, unsigned int n_writers)
+{
+	const unsigned int n_children = n_readers + n_writers;
+	const size_t data = 6 * 1000 * 1000;
+	const size_t file_sz = data / 100;
+	size_t read_bias, write_bias;
+	int i, fd, child_id;
+	char buf[file_sz];
+	pid_t pid;
+
+	/* Only allow multiples for simplicity */
+	ASSERT_EQ(!(n_readers % n_writers) || !(n_writers % n_readers), true);
+	read_bias = n_writers / n_readers ?: 1;
+	write_bias = n_readers / n_writers ?: 1;
+
+	/* prep a file to send */
+	fd = open("/tmp/", O_TMPFILE | O_RDWR, 0600);
+	ASSERT_GE(fd, 0);
+
+	memset(buf, 0xac, file_sz);
+	ASSERT_EQ(write(fd, buf, file_sz), file_sz);
+
+	/* spawn children */
+	for (child_id = 0; child_id < n_children; child_id++) {
+		pid = fork();
+		ASSERT_NE(pid, -1);
+		if (!pid)
+			break;
+	}
+
+	/* parent waits for all children */
+	if (pid) {
+		for (i = 0; i < n_children; i++) {
+			int status;
+
+			wait(&status);
+			EXPECT_EQ(status, 0);
+		}
+
+		return;
+	}
+
+	/* Split threads for reading and writing */
+	if (child_id < n_readers) {
+		size_t left = data * read_bias;
+		char rb[8001];
+
+		while (left) {
+			int res;
+
+			res = recv(self->cfd, rb,
+				   left > sizeof(rb) ? sizeof(rb) : left, 0);
+
+			EXPECT_GE(res, 0);
+			left -= res;
+		}
+	} else {
+		size_t left = data * write_bias;
+
+		while (left) {
+			int res;
+
+			ASSERT_EQ(lseek(fd, 0, SEEK_SET), 0);
+			if (sendpg)
+				res = sendfile(self->fd, fd, NULL,
+					       left > file_sz ? file_sz : left);
+			else
+				res = send(self->fd, buf,
+					   left > file_sz ? file_sz : left, 0);
+
+			EXPECT_GE(res, 0);
+			left -= res;
+		}
+	}
+}
+
+TEST_F(tls, mutliproc_even)
+{
+	test_mutliproc(_metadata, self, false, 6, 6);
+}
+
+TEST_F(tls, mutliproc_readers)
+{
+	test_mutliproc(_metadata, self, false, 4, 12);
+}
+
+TEST_F(tls, mutliproc_writers)
+{
+	test_mutliproc(_metadata, self, false, 10, 2);
+}
+
+TEST_F(tls, mutliproc_sendpage_even)
+{
+	test_mutliproc(_metadata, self, true, 6, 6);
+}
+
+TEST_F(tls, mutliproc_sendpage_readers)
+{
+	test_mutliproc(_metadata, self, true, 4, 12);
+}
+
+TEST_F(tls, mutliproc_sendpage_writers)
+{
+	test_mutliproc(_metadata, self, true, 10, 2);
+}
+
 TEST_F(tls, control_msg)
 {
 	if (self->notls)
-- 
2.23.0

Re: [PATCH net 0/3] net/tls: add a TX lock

[David Miller]

From: Jakub Kicinski <jakub.kicinski@netronome.com>
Date: Tue,  5 Nov 2019 14:24:33 -0800

> Some time ago Pooja and Mallesham started reporting crashes with
> an async accelerator. After trying to poke the existing logic into
> shape I came to the conclusion that it can't be trusted, and to
> preserve our sanity we should just add a lock around the TX side.
> 
> First patch removes the sk_write_pending checks from the write
> space callbacks. Those don't seem to have a logical justification.
> 
> Patch 2 adds the TX lock and patch 3 associated test (which should
> hang with current net).
> 
> Mallesham reports that even with these fixes applied the async
> accelerator workload still occasionally hangs waiting for socket
> memory. I suspect that's strictly related to the way async crypto
> is integrated in TLS, so I think we should get these into net or
> net-next and move from there.

Series applied and queued up for -stable, thanks Jakub.