From 603380f54f837a7837c30713de25cb146254a0d7 Mon Sep 17 00:00:00 2001
From: Jakub Kicinski <kuba@kernel.org>
Date: Tue, 5 Jul 2022 16:59:22 -0700
Subject: tls: rx: don't include tail size in data_len

To make future patches easier to review make data_len
contain the length of the data, without the tail.

Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
---
 net/tls/tls_sw.c | 8 ++++----
 1 file changed, 4 insertions(+), 4 deletions(-)

diff --git a/net/tls/tls_sw.c b/net/tls/tls_sw.c
index 0513f82b8537..7fcb54e43a08 100644
--- a/net/tls/tls_sw.c
+++ b/net/tls/tls_sw.c
@@ -1423,8 +1423,7 @@ static int decrypt_internal(struct sock *sk, struct sk_buff *skb,
 	u8 *aad, *iv, *mem = NULL;
 	struct scatterlist *sgin = NULL;
 	struct scatterlist *sgout = NULL;
-	const int data_len = rxm->full_len - prot->overhead_size +
-			     prot->tail_size;
+	const int data_len = rxm->full_len - prot->overhead_size;
 	int iv_offset = 0;
 
 	if (darg->zc && (out_iov || out_sg)) {
@@ -1519,7 +1518,8 @@ static int decrypt_internal(struct sock *sk, struct sk_buff *skb,
 			sg_init_table(sgout, n_sgout);
 			sg_set_buf(&sgout[0], aad, prot->aad_size);
 
-			err = tls_setup_from_iter(out_iov, data_len,
+			err = tls_setup_from_iter(out_iov,
+						  data_len + prot->tail_size,
 						  &pages, &sgout[1],
 						  (n_sgout - 1));
 			if (err < 0)
@@ -1538,7 +1538,7 @@ fallback_to_reg_recv:
 
 	/* Prepare and submit AEAD request */
 	err = tls_do_decryption(sk, skb, sgin, sgout, iv,
-				data_len, aead_req, darg);
+				data_len + prot->tail_size, aead_req, darg);
 	if (darg->async)
 		return 0;
 
-- 
cgit 


From ce61327ce989b63c0bd1cc7afee00e218ee696ac Mon Sep 17 00:00:00 2001
From: Jakub Kicinski <kuba@kernel.org>
Date: Tue, 5 Jul 2022 16:59:23 -0700
Subject: tls: rx: support optimistic decrypt to user buffer with TLS 1.3

We currently don't support decrypt to user buffer with TLS 1.3
because we don't know the record type and how much padding
record contains before decryption. In practice data records
are by far most common and padding gets used rarely so
we can assume data record, no padding, and if we find out
that wasn't the case - retry the crypto in place (decrypt
to skb).

To safeguard from user overwriting content type and padding
before we can check it attach a 1B sg entry where last byte
of the record will land.

Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
---
 net/tls/tls_sw.c | 38 +++++++++++++++++++++++++++++---------
 1 file changed, 29 insertions(+), 9 deletions(-)

diff --git a/net/tls/tls_sw.c b/net/tls/tls_sw.c
index 7fcb54e43a08..2bac57684429 100644
--- a/net/tls/tls_sw.c
+++ b/net/tls/tls_sw.c
@@ -47,6 +47,7 @@
 struct tls_decrypt_arg {
 	bool zc;
 	bool async;
+	u8 tail;
 };
 
 noinline void tls_err_abort(struct sock *sk, int err)
@@ -133,7 +134,8 @@ static int skb_nsg(struct sk_buff *skb, int offset, int len)
         return __skb_nsg(skb, offset, len, 0);
 }
 
-static int padding_length(struct tls_prot_info *prot, struct sk_buff *skb)
+static int tls_padding_length(struct tls_prot_info *prot, struct sk_buff *skb,
+			      struct tls_decrypt_arg *darg)
 {
 	struct strp_msg *rxm = strp_msg(skb);
 	struct tls_msg *tlm = tls_msg(skb);
@@ -142,7 +144,7 @@ static int padding_length(struct tls_prot_info *prot, struct sk_buff *skb)
 	/* Determine zero-padding length */
 	if (prot->version == TLS_1_3_VERSION) {
 		int offset = rxm->full_len - TLS_TAG_SIZE - 1;
-		char content_type = 0;
+		char content_type = darg->zc ? darg->tail : 0;
 		int err;
 
 		while (content_type == 0) {
@@ -1418,17 +1420,18 @@ static int decrypt_internal(struct sock *sk, struct sk_buff *skb,
 	struct strp_msg *rxm = strp_msg(skb);
 	struct tls_msg *tlm = tls_msg(skb);
 	int n_sgin, n_sgout, nsg, mem_size, aead_size, err, pages = 0;
+	u8 *aad, *iv, *tail, *mem = NULL;
 	struct aead_request *aead_req;
 	struct sk_buff *unused;
-	u8 *aad, *iv, *mem = NULL;
 	struct scatterlist *sgin = NULL;
 	struct scatterlist *sgout = NULL;
 	const int data_len = rxm->full_len - prot->overhead_size;
+	int tail_pages = !!prot->tail_size;
 	int iv_offset = 0;
 
 	if (darg->zc && (out_iov || out_sg)) {
 		if (out_iov)
-			n_sgout = 1 +
+			n_sgout = 1 + tail_pages +
 				iov_iter_npages_cap(out_iov, INT_MAX, data_len);
 		else
 			n_sgout = sg_nents(out_sg);
@@ -1452,9 +1455,10 @@ static int decrypt_internal(struct sock *sk, struct sk_buff *skb,
 	mem_size = aead_size + (nsg * sizeof(struct scatterlist));
 	mem_size = mem_size + prot->aad_size;
 	mem_size = mem_size + MAX_IV_SIZE;
+	mem_size = mem_size + prot->tail_size;
 
 	/* Allocate a single block of memory which contains
-	 * aead_req || sgin[] || sgout[] || aad || iv.
+	 * aead_req || sgin[] || sgout[] || aad || iv || tail.
 	 * This order achieves correct alignment for aead_req, sgin, sgout.
 	 */
 	mem = kmalloc(mem_size, sk->sk_allocation);
@@ -1467,6 +1471,7 @@ static int decrypt_internal(struct sock *sk, struct sk_buff *skb,
 	sgout = sgin + n_sgin;
 	aad = (u8 *)(sgout + n_sgout);
 	iv = aad + prot->aad_size;
+	tail = iv + MAX_IV_SIZE;
 
 	/* For CCM based ciphers, first byte of nonce+iv is a constant */
 	switch (prot->cipher_type) {
@@ -1518,12 +1523,18 @@ static int decrypt_internal(struct sock *sk, struct sk_buff *skb,
 			sg_init_table(sgout, n_sgout);
 			sg_set_buf(&sgout[0], aad, prot->aad_size);
 
-			err = tls_setup_from_iter(out_iov,
-						  data_len + prot->tail_size,
+			err = tls_setup_from_iter(out_iov, data_len,
 						  &pages, &sgout[1],
-						  (n_sgout - 1));
+						  (n_sgout - 1 - tail_pages));
 			if (err < 0)
 				goto fallback_to_reg_recv;
+
+			if (prot->tail_size) {
+				sg_unmark_end(&sgout[pages]);
+				sg_set_buf(&sgout[pages + 1], tail,
+					   prot->tail_size);
+				sg_mark_end(&sgout[pages + 1]);
+			}
 		} else if (out_sg) {
 			memcpy(sgout, out_sg, n_sgout * sizeof(*sgout));
 		} else {
@@ -1542,6 +1553,9 @@ fallback_to_reg_recv:
 	if (darg->async)
 		return 0;
 
+	if (prot->tail_size)
+		darg->tail = *tail;
+
 	/* Release the pages in case iov was mapped to pages */
 	for (; pages > 0; pages--)
 		put_page(sg_page(&sgout[pages]));
@@ -1583,9 +1597,15 @@ static int decrypt_skb_update(struct sock *sk, struct sk_buff *skb,
 		return err;
 	if (darg->async)
 		goto decrypt_next;
+	/* If opportunistic TLS 1.3 ZC failed retry without ZC */
+	if (unlikely(darg->zc && prot->version == TLS_1_3_VERSION &&
+		     darg->tail != TLS_RECORD_TYPE_DATA)) {
+		darg->zc = false;
+		return decrypt_skb_update(sk, skb, dest, darg);
+	}
 
 decrypt_done:
-	pad = padding_length(prot, skb);
+	pad = tls_padding_length(prot, skb, darg);
 	if (pad < 0)
 		return pad;
 
-- 
cgit 


From 88527790c079fb1ea41cbcfa4450ee37906a2fb0 Mon Sep 17 00:00:00 2001
From: Jakub Kicinski <kuba@kernel.org>
Date: Tue, 5 Jul 2022 16:59:24 -0700
Subject: tls: rx: add sockopt for enabling optimistic decrypt with TLS 1.3

Since optimisitic decrypt may add extra load in case of retries
require socket owner to explicitly opt-in.

Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
---
 Documentation/networking/tls.rst | 18 ++++++++++
 include/linux/sockptr.h          |  8 +++++
 include/net/tls.h                |  3 ++
 include/uapi/linux/snmp.h        |  1 +
 include/uapi/linux/tls.h         |  2 ++
 net/tls/tls_main.c               | 75 ++++++++++++++++++++++++++++++++++++++++
 net/tls/tls_proc.c               |  1 +
 net/tls/tls_sw.c                 | 21 +++++++----
 8 files changed, 122 insertions(+), 7 deletions(-)

diff --git a/Documentation/networking/tls.rst b/Documentation/networking/tls.rst
index be8e10c14b05..7a6643836e42 100644
--- a/Documentation/networking/tls.rst
+++ b/Documentation/networking/tls.rst
@@ -239,6 +239,19 @@ for the original TCP transmission and TCP retransmissions. To the receiver
 this will look like TLS records had been tampered with and will result
 in record authentication failures.
 
+TLS_RX_EXPECT_NO_PAD
+~~~~~~~~~~~~~~~~~~~~
+
+TLS 1.3 only. Expect the sender to not pad records. This allows the data
+to be decrypted directly into user space buffers with TLS 1.3.
+
+This optimization is safe to enable only if the remote end is trusted,
+otherwise it is an attack vector to doubling the TLS processing cost.
+
+If the record decrypted turns out to had been padded or is not a data
+record it will be decrypted again into a kernel buffer without zero copy.
+Such events are counted in the ``TlsDecryptRetry`` statistic.
+
 Statistics
 ==========
 
@@ -264,3 +277,8 @@ TLS implementation exposes the following per-namespace statistics
 
 - ``TlsDeviceRxResync`` -
   number of RX resyncs sent to NICs handling cryptography
+
+- ``TlsDecryptRetry`` -
+  number of RX records which had to be re-decrypted due to
+  ``TLS_RX_EXPECT_NO_PAD`` mis-prediction. Note that this counter will
+  also increment for non-data records.
diff --git a/include/linux/sockptr.h b/include/linux/sockptr.h
index ea193414298b..d45902fb4cad 100644
--- a/include/linux/sockptr.h
+++ b/include/linux/sockptr.h
@@ -102,4 +102,12 @@ static inline long strncpy_from_sockptr(char *dst, sockptr_t src, size_t count)
 	return strncpy_from_user(dst, src.user, count);
 }
 
+static inline int check_zeroed_sockptr(sockptr_t src, size_t offset,
+				       size_t size)
+{
+	if (!sockptr_is_kernel(src))
+		return check_zeroed_user(src.user + offset, size);
+	return memchr_inv(src.kernel + offset, 0, size) == NULL;
+}
+
 #endif /* _LINUX_SOCKPTR_H */
diff --git a/include/net/tls.h b/include/net/tls.h
index 8017f1703447..4fc16ca5f469 100644
--- a/include/net/tls.h
+++ b/include/net/tls.h
@@ -149,6 +149,7 @@ struct tls_sw_context_rx {
 
 	struct sk_buff *recv_pkt;
 	u8 async_capable:1;
+	u8 zc_capable:1;
 	atomic_t decrypt_pending;
 	/* protect crypto_wait with decrypt_pending*/
 	spinlock_t decrypt_compl_lock;
@@ -239,6 +240,7 @@ struct tls_context {
 	u8 tx_conf:3;
 	u8 rx_conf:3;
 	u8 zerocopy_sendfile:1;
+	u8 rx_no_pad:1;
 
 	int (*push_pending_record)(struct sock *sk, int flags);
 	void (*sk_write_space)(struct sock *sk);
@@ -358,6 +360,7 @@ int tls_sk_attach(struct sock *sk, int optname, char __user *optval,
 void tls_err_abort(struct sock *sk, int err);
 
 int tls_set_sw_offload(struct sock *sk, struct tls_context *ctx, int tx);
+void tls_update_rx_zc_capable(struct tls_context *tls_ctx);
 void tls_sw_strparser_arm(struct sock *sk, struct tls_context *ctx);
 void tls_sw_strparser_done(struct tls_context *tls_ctx);
 int tls_sw_sendmsg(struct sock *sk, struct msghdr *msg, size_t size);
diff --git a/include/uapi/linux/snmp.h b/include/uapi/linux/snmp.h
index 904909d020e2..1c9152add663 100644
--- a/include/uapi/linux/snmp.h
+++ b/include/uapi/linux/snmp.h
@@ -344,6 +344,7 @@ enum
 	LINUX_MIB_TLSRXDEVICE,			/* TlsRxDevice */
 	LINUX_MIB_TLSDECRYPTERROR,		/* TlsDecryptError */
 	LINUX_MIB_TLSRXDEVICERESYNC,		/* TlsRxDeviceResync */
+	LINUX_MIN_TLSDECRYPTRETRY,		/* TlsDecryptRetry */
 	__LINUX_MIB_TLSMAX
 };
 
diff --git a/include/uapi/linux/tls.h b/include/uapi/linux/tls.h
index bb8f80812b0b..f1157d8f4acd 100644
--- a/include/uapi/linux/tls.h
+++ b/include/uapi/linux/tls.h
@@ -40,6 +40,7 @@
 #define TLS_TX			1	/* Set transmit parameters */
 #define TLS_RX			2	/* Set receive parameters */
 #define TLS_TX_ZEROCOPY_RO	3	/* TX zerocopy (only sendfile now) */
+#define TLS_RX_EXPECT_NO_PAD	4	/* Attempt opportunistic zero-copy */
 
 /* Supported versions */
 #define TLS_VERSION_MINOR(ver)	((ver) & 0xFF)
@@ -162,6 +163,7 @@ enum {
 	TLS_INFO_TXCONF,
 	TLS_INFO_RXCONF,
 	TLS_INFO_ZC_RO_TX,
+	TLS_INFO_RX_NO_PAD,
 	__TLS_INFO_MAX,
 };
 #define TLS_INFO_MAX (__TLS_INFO_MAX - 1)
diff --git a/net/tls/tls_main.c b/net/tls/tls_main.c
index 2ffede463e4a..1b3efc96db0b 100644
--- a/net/tls/tls_main.c
+++ b/net/tls/tls_main.c
@@ -533,6 +533,37 @@ static int do_tls_getsockopt_tx_zc(struct sock *sk, char __user *optval,
 	return 0;
 }
 
+static int do_tls_getsockopt_no_pad(struct sock *sk, char __user *optval,
+				    int __user *optlen)
+{
+	struct tls_context *ctx = tls_get_ctx(sk);
+	unsigned int value;
+	int err, len;
+
+	if (ctx->prot_info.version != TLS_1_3_VERSION)
+		return -EINVAL;
+
+	if (get_user(len, optlen))
+		return -EFAULT;
+	if (len < sizeof(value))
+		return -EINVAL;
+
+	lock_sock(sk);
+	err = -EINVAL;
+	if (ctx->rx_conf == TLS_SW || ctx->rx_conf == TLS_HW)
+		value = ctx->rx_no_pad;
+	release_sock(sk);
+	if (err)
+		return err;
+
+	if (put_user(sizeof(value), optlen))
+		return -EFAULT;
+	if (copy_to_user(optval, &value, sizeof(value)))
+		return -EFAULT;
+
+	return 0;
+}
+
 static int do_tls_getsockopt(struct sock *sk, int optname,
 			     char __user *optval, int __user *optlen)
 {
@@ -547,6 +578,9 @@ static int do_tls_getsockopt(struct sock *sk, int optname,
 	case TLS_TX_ZEROCOPY_RO:
 		rc = do_tls_getsockopt_tx_zc(sk, optval, optlen);
 		break;
+	case TLS_RX_EXPECT_NO_PAD:
+		rc = do_tls_getsockopt_no_pad(sk, optval, optlen);
+		break;
 	default:
 		rc = -ENOPROTOOPT;
 		break;
@@ -718,6 +752,38 @@ static int do_tls_setsockopt_tx_zc(struct sock *sk, sockptr_t optval,
 	return 0;
 }
 
+static int do_tls_setsockopt_no_pad(struct sock *sk, sockptr_t optval,
+				    unsigned int optlen)
+{
+	struct tls_context *ctx = tls_get_ctx(sk);
+	u32 val;
+	int rc;
+
+	if (ctx->prot_info.version != TLS_1_3_VERSION ||
+	    sockptr_is_null(optval) || optlen < sizeof(val))
+		return -EINVAL;
+
+	rc = copy_from_sockptr(&val, optval, sizeof(val));
+	if (rc)
+		return -EFAULT;
+	if (val > 1)
+		return -EINVAL;
+	rc = check_zeroed_sockptr(optval, sizeof(val), optlen - sizeof(val));
+	if (rc < 1)
+		return rc == 0 ? -EINVAL : rc;
+
+	lock_sock(sk);
+	rc = -EINVAL;
+	if (ctx->rx_conf == TLS_SW || ctx->rx_conf == TLS_HW) {
+		ctx->rx_no_pad = val;
+		tls_update_rx_zc_capable(ctx);
+		rc = 0;
+	}
+	release_sock(sk);
+
+	return rc;
+}
+
 static int do_tls_setsockopt(struct sock *sk, int optname, sockptr_t optval,
 			     unsigned int optlen)
 {
@@ -736,6 +802,9 @@ static int do_tls_setsockopt(struct sock *sk, int optname, sockptr_t optval,
 		rc = do_tls_setsockopt_tx_zc(sk, optval, optlen);
 		release_sock(sk);
 		break;
+	case TLS_RX_EXPECT_NO_PAD:
+		rc = do_tls_setsockopt_no_pad(sk, optval, optlen);
+		break;
 	default:
 		rc = -ENOPROTOOPT;
 		break;
@@ -976,6 +1045,11 @@ static int tls_get_info(const struct sock *sk, struct sk_buff *skb)
 		if (err)
 			goto nla_failure;
 	}
+	if (ctx->rx_no_pad) {
+		err = nla_put_flag(skb, TLS_INFO_RX_NO_PAD);
+		if (err)
+			goto nla_failure;
+	}
 
 	rcu_read_unlock();
 	nla_nest_end(skb, start);
@@ -997,6 +1071,7 @@ static size_t tls_get_info_size(const struct sock *sk)
 		nla_total_size(sizeof(u16)) +	/* TLS_INFO_RXCONF */
 		nla_total_size(sizeof(u16)) +	/* TLS_INFO_TXCONF */
 		nla_total_size(0) +		/* TLS_INFO_ZC_RO_TX */
+		nla_total_size(0) +		/* TLS_INFO_RX_NO_PAD */
 		0;
 
 	return size;
diff --git a/net/tls/tls_proc.c b/net/tls/tls_proc.c
index feeceb0e4cb4..0c200000cc45 100644
--- a/net/tls/tls_proc.c
+++ b/net/tls/tls_proc.c
@@ -18,6 +18,7 @@ static const struct snmp_mib tls_mib_list[] = {
 	SNMP_MIB_ITEM("TlsRxDevice", LINUX_MIB_TLSRXDEVICE),
 	SNMP_MIB_ITEM("TlsDecryptError", LINUX_MIB_TLSDECRYPTERROR),
 	SNMP_MIB_ITEM("TlsRxDeviceResync", LINUX_MIB_TLSRXDEVICERESYNC),
+	SNMP_MIB_ITEM("TlsDecryptRetry", LINUX_MIN_TLSDECRYPTRETRY),
 	SNMP_MIB_SENTINEL
 };
 
diff --git a/net/tls/tls_sw.c b/net/tls/tls_sw.c
index 2bac57684429..7592b6519953 100644
--- a/net/tls/tls_sw.c
+++ b/net/tls/tls_sw.c
@@ -1601,6 +1601,7 @@ static int decrypt_skb_update(struct sock *sk, struct sk_buff *skb,
 	if (unlikely(darg->zc && prot->version == TLS_1_3_VERSION &&
 		     darg->tail != TLS_RECORD_TYPE_DATA)) {
 		darg->zc = false;
+		TLS_INC_STATS(sock_net(sk), LINUX_MIN_TLSDECRYPTRETRY);
 		return decrypt_skb_update(sk, skb, dest, darg);
 	}
 
@@ -1787,7 +1788,7 @@ int tls_sw_recvmsg(struct sock *sk,
 	timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
 
 	zc_capable = !bpf_strp_enabled && !is_kvec && !is_peek &&
-		     prot->version != TLS_1_3_VERSION;
+		ctx->zc_capable;
 	decrypted = 0;
 	while (len && (decrypted + copied < target || ctx->recv_pkt)) {
 		struct tls_decrypt_arg darg = {};
@@ -2269,6 +2270,14 @@ void tls_sw_strparser_arm(struct sock *sk, struct tls_context *tls_ctx)
 	strp_check_rcv(&rx_ctx->strp);
 }
 
+void tls_update_rx_zc_capable(struct tls_context *tls_ctx)
+{
+	struct tls_sw_context_rx *rx_ctx = tls_sw_ctx_rx(tls_ctx);
+
+	rx_ctx->zc_capable = tls_ctx->rx_no_pad ||
+		tls_ctx->prot_info.version != TLS_1_3_VERSION;
+}
+
 int tls_set_sw_offload(struct sock *sk, struct tls_context *ctx, int tx)
 {
 	struct tls_context *tls_ctx = tls_get_ctx(sk);
@@ -2504,12 +2513,10 @@ int tls_set_sw_offload(struct sock *sk, struct tls_context *ctx, int tx)
 	if (sw_ctx_rx) {
 		tfm = crypto_aead_tfm(sw_ctx_rx->aead_recv);
 
-		if (crypto_info->version == TLS_1_3_VERSION)
-			sw_ctx_rx->async_capable = 0;
-		else
-			sw_ctx_rx->async_capable =
-				!!(tfm->__crt_alg->cra_flags &
-				   CRYPTO_ALG_ASYNC);
+		tls_update_rx_zc_capable(ctx);
+		sw_ctx_rx->async_capable =
+			crypto_info->version != TLS_1_3_VERSION &&
+			!!(tfm->__crt_alg->cra_flags & CRYPTO_ALG_ASYNC);
 
 		/* Set up strparser */
 		memset(&cb, 0, sizeof(cb));
-- 
cgit 


From f36068a20256bad993d60e49602f02e3af336506 Mon Sep 17 00:00:00 2001
From: Jakub Kicinski <kuba@kernel.org>
Date: Tue, 5 Jul 2022 16:59:25 -0700
Subject: selftests: tls: add selftest variant for pad

Add a self-test variant with TLS 1.3 nopad set.

Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
---
 tools/testing/selftests/net/tls.c | 15 +++++++++++++++
 1 file changed, 15 insertions(+)

diff --git a/tools/testing/selftests/net/tls.c b/tools/testing/selftests/net/tls.c
index 5d70b04c482c..e71ec5846be9 100644
--- a/tools/testing/selftests/net/tls.c
+++ b/tools/testing/selftests/net/tls.c
@@ -235,6 +235,7 @@ FIXTURE_VARIANT(tls)
 {
 	uint16_t tls_version;
 	uint16_t cipher_type;
+	bool nopad;
 };
 
 FIXTURE_VARIANT_ADD(tls, 12_aes_gcm)
@@ -297,9 +298,17 @@ FIXTURE_VARIANT_ADD(tls, 13_aes_gcm_256)
 	.cipher_type = TLS_CIPHER_AES_GCM_256,
 };
 
+FIXTURE_VARIANT_ADD(tls, 13_nopad)
+{
+	.tls_version = TLS_1_3_VERSION,
+	.cipher_type = TLS_CIPHER_AES_GCM_128,
+	.nopad = true,
+};
+
 FIXTURE_SETUP(tls)
 {
 	struct tls_crypto_info_keys tls12;
+	int one = 1;
 	int ret;
 
 	tls_crypto_info_init(variant->tls_version, variant->cipher_type,
@@ -315,6 +324,12 @@ FIXTURE_SETUP(tls)
 
 	ret = setsockopt(self->cfd, SOL_TLS, TLS_RX, &tls12, tls12.len);
 	ASSERT_EQ(ret, 0);
+
+	if (variant->nopad) {
+		ret = setsockopt(self->cfd, SOL_TLS, TLS_RX_EXPECT_NO_PAD,
+				 (void *)&one, sizeof(one));
+		ASSERT_EQ(ret, 0);
+	}
 }
 
 FIXTURE_TEARDOWN(tls)
-- 
cgit 


From c46b01839f7aad5889e23505bbfbeb5f4d7fde8e Mon Sep 17 00:00:00 2001
From: Jakub Kicinski <kuba@kernel.org>
Date: Tue, 5 Jul 2022 16:59:26 -0700
Subject: tls: rx: periodically flush socket backlog

We continuously hold the socket lock during large reads and writes.
This may inflate RTT and negatively impact TCP performance.
Flush the backlog periodically. I tried to pick a flush period (128kB)
which gives significant benefit but the max Bps rate is not yet visibly
impacted.

Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
---
 net/core/sock.c  |  1 +
 net/tls/tls_sw.c | 23 +++++++++++++++++++++++
 2 files changed, 24 insertions(+)

diff --git a/net/core/sock.c b/net/core/sock.c
index 92a0296ccb18..4cb957d934a2 100644
--- a/net/core/sock.c
+++ b/net/core/sock.c
@@ -2870,6 +2870,7 @@ void __sk_flush_backlog(struct sock *sk)
 	__release_sock(sk);
 	spin_unlock_bh(&sk->sk_lock.slock);
 }
+EXPORT_SYMBOL_GPL(__sk_flush_backlog);
 
 /**
  * sk_wait_data - wait for data to arrive at sk_receive_queue
diff --git a/net/tls/tls_sw.c b/net/tls/tls_sw.c
index 7592b6519953..79043bc3da39 100644
--- a/net/tls/tls_sw.c
+++ b/net/tls/tls_sw.c
@@ -1738,6 +1738,24 @@ out:
 	return copied ? : err;
 }
 
+static void
+tls_read_flush_backlog(struct sock *sk, struct tls_prot_info *prot,
+		       size_t len_left, size_t decrypted, ssize_t done,
+		       size_t *flushed_at)
+{
+	size_t max_rec;
+
+	if (len_left <= decrypted)
+		return;
+
+	max_rec = prot->overhead_size - prot->tail_size + TLS_MAX_PAYLOAD_SIZE;
+	if (done - *flushed_at < SZ_128K && tcp_inq(sk) > max_rec)
+		return;
+
+	*flushed_at = done;
+	sk_flush_backlog(sk);
+}
+
 int tls_sw_recvmsg(struct sock *sk,
 		   struct msghdr *msg,
 		   size_t len,
@@ -1750,6 +1768,7 @@ int tls_sw_recvmsg(struct sock *sk,
 	struct sk_psock *psock;
 	unsigned char control = 0;
 	ssize_t decrypted = 0;
+	size_t flushed_at = 0;
 	struct strp_msg *rxm;
 	struct tls_msg *tlm;
 	struct sk_buff *skb;
@@ -1839,6 +1858,10 @@ int tls_sw_recvmsg(struct sock *sk,
 		if (err <= 0)
 			goto recv_end;
 
+		/* periodically flush backlog, and feed strparser */
+		tls_read_flush_backlog(sk, prot, len, to_decrypt,
+				       decrypted + copied, &flushed_at);
+
 		ctx->recv_pkt = NULL;
 		__strp_unpause(&ctx->strp);
 		__skb_queue_tail(&ctx->rx_list, skb);
-- 
cgit