[PATCH net-next 00/11] tls: rx: random refactoring part 2

[PATCH net-next 00/11] tls: rx: random refactoring part 2

[Jakub Kicinski]

TLS Rx refactoring. Part 2 of 3. This one focusing on the main loop.
A couple of features to follow.

Jakub Kicinski (11):
  tls: rx: drop unnecessary arguments from tls_setup_from_iter()
  tls: rx: don't report text length from the bowels of decrypt
  tls: rx: wrap decryption arguments in a structure
  tls: rx: simplify async wait
  tls: rx: factor out writing ContentType to cmsg
  tls: rx: don't handle async in tls_sw_advance_skb()
  tls: rx: don't track the async count
  tls: rx: pull most of zc check out of the loop
  tls: rx: inline consuming the skb at the end of the loop
  tls: rx: clear ctx->recv_pkt earlier
  tls: rx: jump out for cases which need to leave skb on list

 include/net/tls.h |   1 -
 net/tls/tls_sw.c  | 264 ++++++++++++++++++----------------------------
 2 files changed, 104 insertions(+), 161 deletions(-)

-- 
2.34.1

[PATCH net-next 01/11] tls: rx: drop unnecessary arguments from tls_setup_from_iter()

[Jakub Kicinski]

sk is unused, remove it to make it clear the function
doesn't poke at the socket.

size_used is always 0 on input and @length on success.

Signed-off-by: Jakub Kicinski <kuba@kernel.org>
---
 net/tls/tls_sw.c | 14 ++++++--------
 1 file changed, 6 insertions(+), 8 deletions(-)

diff --git a/net/tls/tls_sw.c b/net/tls/tls_sw.c
index 3a0a120f9c56..86f77f8b825e 100644
--- a/net/tls/tls_sw.c
+++ b/net/tls/tls_sw.c
@@ -1345,15 +1345,14 @@ static struct sk_buff *tls_wait_data(struct sock *sk, struct sk_psock *psock,
 	return skb;
 }
 
-static int tls_setup_from_iter(struct sock *sk, struct iov_iter *from,
+static int tls_setup_from_iter(struct iov_iter *from,
 			       int length, int *pages_used,
-			       unsigned int *size_used,
 			       struct scatterlist *to,
 			       int to_max_pages)
 {
 	int rc = 0, i = 0, num_elem = *pages_used, maxpages;
 	struct page *pages[MAX_SKB_FRAGS];
-	unsigned int size = *size_used;
+	unsigned int size = 0;
 	ssize_t copied, use;
 	size_t offset;
 
@@ -1396,8 +1395,7 @@ static int tls_setup_from_iter(struct sock *sk, struct iov_iter *from,
 		sg_mark_end(&to[num_elem - 1]);
 out:
 	if (rc)
-		iov_iter_revert(from, size - *size_used);
-	*size_used = size;
+		iov_iter_revert(from, size);
 	*pages_used = num_elem;
 
 	return rc;
@@ -1523,12 +1521,12 @@ 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);
 
-			*chunk = 0;
-			err = tls_setup_from_iter(sk, out_iov, data_len,
-						  &pages, chunk, &sgout[1],
+			err = tls_setup_from_iter(out_iov, data_len,
+						  &pages, &sgout[1],
 						  (n_sgout - 1));
 			if (err < 0)
 				goto fallback_to_reg_recv;
+			*chunk = data_len;
 		} else if (out_sg) {
 			memcpy(sgout, out_sg, n_sgout * sizeof(*sgout));
 		} else {
-- 
2.34.1

[PATCH net-next 02/11] tls: rx: don't report text length from the bowels of decrypt

[Jakub Kicinski]

We plumb pointer to chunk all the way to the decryption method.
It's set to the length of the text when decrypt_skb_update()
returns.

I think the code is written this way because original TLS
implementation passed &chunk to zerocopy_from_iter() and this
was carried forward as the code gotten more complex, without
any refactoring.

The fix for peek() introduced a new variable - to_decrypt
which for all practical purposes is what chunk is going to
get set to. Spare ourselves the pointer passing, use to_decrypt.

Use this opportunity to clean things up a little further.

Note that chunk / to_decrypt was mostly needed for the async
path, since the sync path would access rxm->full_len (decryption
transforms full_len from record size to text size). Use the
right source of truth more explicitly.

We have three cases:
 - async - it's TLS 1.2 only, so chunk == to_decrypt, but we
           need the min() because to_decrypt is a whole record
	   and we don't want to underflow len. Note that we can't
	   handle partial record by falling back to sync as it
	   would introduce reordering against records in flight.
 - zc - again, TLS 1.2 only for now, so chunk == to_decrypt,
        we don't do zc if len < to_decrypt, no need to check again.
 - normal - it already handles chunk > len, we can factor out the
            assignment to rxm->full_len and share it with zc.

Signed-off-by: Jakub Kicinski <kuba@kernel.org>
---
 net/tls/tls_sw.c | 33 ++++++++++++++-------------------
 1 file changed, 14 insertions(+), 19 deletions(-)

diff --git a/net/tls/tls_sw.c b/net/tls/tls_sw.c
index 86f77f8b825e..c321c5f85fbe 100644
--- a/net/tls/tls_sw.c
+++ b/net/tls/tls_sw.c
@@ -1412,7 +1412,7 @@ static int tls_setup_from_iter(struct iov_iter *from,
 static int decrypt_internal(struct sock *sk, struct sk_buff *skb,
 			    struct iov_iter *out_iov,
 			    struct scatterlist *out_sg,
-			    int *chunk, bool *zc, bool async)
+			    bool *zc, bool async)
 {
 	struct tls_context *tls_ctx = tls_get_ctx(sk);
 	struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
@@ -1526,7 +1526,6 @@ static int decrypt_internal(struct sock *sk, struct sk_buff *skb,
 						  (n_sgout - 1));
 			if (err < 0)
 				goto fallback_to_reg_recv;
-			*chunk = data_len;
 		} else if (out_sg) {
 			memcpy(sgout, out_sg, n_sgout * sizeof(*sgout));
 		} else {
@@ -1536,7 +1535,6 @@ static int decrypt_internal(struct sock *sk, struct sk_buff *skb,
 fallback_to_reg_recv:
 		sgout = sgin;
 		pages = 0;
-		*chunk = data_len;
 		*zc = false;
 	}
 
@@ -1555,8 +1553,7 @@ static int decrypt_internal(struct sock *sk, struct sk_buff *skb,
 }
 
 static int decrypt_skb_update(struct sock *sk, struct sk_buff *skb,
-			      struct iov_iter *dest, int *chunk, bool *zc,
-			      bool async)
+			      struct iov_iter *dest, bool *zc, bool async)
 {
 	struct tls_context *tls_ctx = tls_get_ctx(sk);
 	struct tls_prot_info *prot = &tls_ctx->prot_info;
@@ -1580,7 +1577,7 @@ static int decrypt_skb_update(struct sock *sk, struct sk_buff *skb,
 		}
 	}
 
-	err = decrypt_internal(sk, skb, dest, NULL, chunk, zc, async);
+	err = decrypt_internal(sk, skb, dest, NULL, zc, async);
 	if (err < 0) {
 		if (err == -EINPROGRESS)
 			tls_advance_record_sn(sk, prot, &tls_ctx->rx);
@@ -1607,9 +1604,8 @@ int decrypt_skb(struct sock *sk, struct sk_buff *skb,
 		struct scatterlist *sgout)
 {
 	bool zc = true;
-	int chunk;
 
-	return decrypt_internal(sk, skb, NULL, sgout, &chunk, &zc, false);
+	return decrypt_internal(sk, skb, NULL, sgout, &zc, false);
 }
 
 static bool tls_sw_advance_skb(struct sock *sk, struct sk_buff *skb,
@@ -1799,9 +1795,8 @@ int tls_sw_recvmsg(struct sock *sk,
 	num_async = 0;
 	while (len && (decrypted + copied < target || ctx->recv_pkt)) {
 		bool retain_skb = false;
+		int to_decrypt, chunk;
 		bool zc = false;
-		int to_decrypt;
-		int chunk = 0;
 		bool async_capable;
 		bool async = false;
 
@@ -1838,7 +1833,7 @@ int tls_sw_recvmsg(struct sock *sk,
 			async_capable = false;
 
 		err = decrypt_skb_update(sk, skb, &msg->msg_iter,
-					 &chunk, &zc, async_capable);
+					 &zc, async_capable);
 		if (err < 0 && err != -EINPROGRESS) {
 			tls_err_abort(sk, -EBADMSG);
 			goto recv_end;
@@ -1876,8 +1871,13 @@ int tls_sw_recvmsg(struct sock *sk,
 			}
 		}
 
-		if (async)
+		if (async) {
+			/* TLS 1.2-only, to_decrypt must be text length */
+			chunk = min_t(int, to_decrypt, len);
 			goto pick_next_record;
+		}
+		/* TLS 1.3 may have updated the length by more than overhead */
+		chunk = rxm->full_len;
 
 		if (!zc) {
 			if (bpf_strp_enabled) {
@@ -1893,11 +1893,9 @@ int tls_sw_recvmsg(struct sock *sk,
 				}
 			}
 
-			if (rxm->full_len > len) {
+			if (chunk > len) {
 				retain_skb = true;
 				chunk = len;
-			} else {
-				chunk = rxm->full_len;
 			}
 
 			err = skb_copy_datagram_msg(skb, rxm->offset,
@@ -1912,9 +1910,6 @@ int tls_sw_recvmsg(struct sock *sk,
 		}
 
 pick_next_record:
-		if (chunk > len)
-			chunk = len;
-
 		decrypted += chunk;
 		len -= chunk;
 
@@ -2016,7 +2011,7 @@ ssize_t tls_sw_splice_read(struct socket *sock,  loff_t *ppos,
 		if (!skb)
 			goto splice_read_end;
 
-		err = decrypt_skb_update(sk, skb, NULL, &chunk, &zc, false);
+		err = decrypt_skb_update(sk, skb, NULL, &zc, false);
 		if (err < 0) {
 			tls_err_abort(sk, -EBADMSG);
 			goto splice_read_end;
-- 
2.34.1

[PATCH net-next 03/11] tls: rx: wrap decryption arguments in a structure

[Jakub Kicinski]

We pass zc as a pointer to bool a few functions down as an in/out
argument. This is error prone since C will happily evalue a pointer
as a boolean (IOW forgetting *zc and writing zc leads to loss of
developer time..). Wrap the arguments into a structure.

Signed-off-by: Jakub Kicinski <kuba@kernel.org>
---
 net/tls/tls_sw.c | 49 ++++++++++++++++++++++++++----------------------
 1 file changed, 27 insertions(+), 22 deletions(-)

diff --git a/net/tls/tls_sw.c b/net/tls/tls_sw.c
index c321c5f85fbe..ecf9a3832798 100644
--- a/net/tls/tls_sw.c
+++ b/net/tls/tls_sw.c
@@ -44,6 +44,11 @@
 #include <net/strparser.h>
 #include <net/tls.h>
 
+struct tls_decrypt_arg {
+	bool zc;
+	bool async;
+};
+
 noinline void tls_err_abort(struct sock *sk, int err)
 {
 	WARN_ON_ONCE(err >= 0);
@@ -1412,7 +1417,7 @@ static int tls_setup_from_iter(struct iov_iter *from,
 static int decrypt_internal(struct sock *sk, struct sk_buff *skb,
 			    struct iov_iter *out_iov,
 			    struct scatterlist *out_sg,
-			    bool *zc, bool async)
+			    struct tls_decrypt_arg *darg)
 {
 	struct tls_context *tls_ctx = tls_get_ctx(sk);
 	struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
@@ -1429,7 +1434,7 @@ static int decrypt_internal(struct sock *sk, struct sk_buff *skb,
 			     prot->tail_size;
 	int iv_offset = 0;
 
-	if (*zc && (out_iov || out_sg)) {
+	if (darg->zc && (out_iov || out_sg)) {
 		if (out_iov)
 			n_sgout = 1 +
 				iov_iter_npages_cap(out_iov, INT_MAX, data_len);
@@ -1439,7 +1444,7 @@ static int decrypt_internal(struct sock *sk, struct sk_buff *skb,
 				 rxm->full_len - prot->prepend_size);
 	} else {
 		n_sgout = 0;
-		*zc = false;
+		darg->zc = false;
 		n_sgin = skb_cow_data(skb, 0, &unused);
 	}
 
@@ -1535,12 +1540,12 @@ static int decrypt_internal(struct sock *sk, struct sk_buff *skb,
 fallback_to_reg_recv:
 		sgout = sgin;
 		pages = 0;
-		*zc = false;
+		darg->zc = false;
 	}
 
 	/* Prepare and submit AEAD request */
 	err = tls_do_decryption(sk, skb, sgin, sgout, iv,
-				data_len, aead_req, async);
+				data_len, aead_req, darg->async);
 	if (err == -EINPROGRESS)
 		return err;
 
@@ -1553,7 +1558,8 @@ static int decrypt_internal(struct sock *sk, struct sk_buff *skb,
 }
 
 static int decrypt_skb_update(struct sock *sk, struct sk_buff *skb,
-			      struct iov_iter *dest, bool *zc, bool async)
+			      struct iov_iter *dest,
+			      struct tls_decrypt_arg *darg)
 {
 	struct tls_context *tls_ctx = tls_get_ctx(sk);
 	struct tls_prot_info *prot = &tls_ctx->prot_info;
@@ -1562,7 +1568,7 @@ static int decrypt_skb_update(struct sock *sk, struct sk_buff *skb,
 	int pad, err;
 
 	if (tlm->decrypted) {
-		*zc = false;
+		darg->zc = false;
 		return 0;
 	}
 
@@ -1572,12 +1578,12 @@ static int decrypt_skb_update(struct sock *sk, struct sk_buff *skb,
 			return err;
 		if (err > 0) {
 			tlm->decrypted = 1;
-			*zc = false;
+			darg->zc = false;
 			goto decrypt_done;
 		}
 	}
 
-	err = decrypt_internal(sk, skb, dest, NULL, zc, async);
+	err = decrypt_internal(sk, skb, dest, NULL, darg);
 	if (err < 0) {
 		if (err == -EINPROGRESS)
 			tls_advance_record_sn(sk, prot, &tls_ctx->rx);
@@ -1603,9 +1609,9 @@ static int decrypt_skb_update(struct sock *sk, struct sk_buff *skb,
 int decrypt_skb(struct sock *sk, struct sk_buff *skb,
 		struct scatterlist *sgout)
 {
-	bool zc = true;
+	struct tls_decrypt_arg darg = { .zc = true, };
 
-	return decrypt_internal(sk, skb, NULL, sgout, &zc, false);
+	return decrypt_internal(sk, skb, NULL, sgout, &darg);
 }
 
 static bool tls_sw_advance_skb(struct sock *sk, struct sk_buff *skb,
@@ -1794,11 +1800,10 @@ int tls_sw_recvmsg(struct sock *sk,
 	decrypted = 0;
 	num_async = 0;
 	while (len && (decrypted + copied < target || ctx->recv_pkt)) {
+		struct tls_decrypt_arg darg = {};
 		bool retain_skb = false;
 		int to_decrypt, chunk;
-		bool zc = false;
-		bool async_capable;
-		bool async = false;
+		bool async;
 
 		skb = tls_wait_data(sk, psock, flags & MSG_DONTWAIT, timeo, &err);
 		if (!skb) {
@@ -1824,16 +1829,15 @@ int tls_sw_recvmsg(struct sock *sk,
 		    tlm->control == TLS_RECORD_TYPE_DATA &&
 		    prot->version != TLS_1_3_VERSION &&
 		    !bpf_strp_enabled)
-			zc = true;
+			darg.zc = true;
 
 		/* Do not use async mode if record is non-data */
 		if (tlm->control == TLS_RECORD_TYPE_DATA && !bpf_strp_enabled)
-			async_capable = ctx->async_capable;
+			darg.async = ctx->async_capable;
 		else
-			async_capable = false;
+			darg.async = false;
 
-		err = decrypt_skb_update(sk, skb, &msg->msg_iter,
-					 &zc, async_capable);
+		err = decrypt_skb_update(sk, skb, &msg->msg_iter, &darg);
 		if (err < 0 && err != -EINPROGRESS) {
 			tls_err_abort(sk, -EBADMSG);
 			goto recv_end;
@@ -1879,7 +1883,7 @@ int tls_sw_recvmsg(struct sock *sk,
 		/* TLS 1.3 may have updated the length by more than overhead */
 		chunk = rxm->full_len;
 
-		if (!zc) {
+		if (!darg.zc) {
 			if (bpf_strp_enabled) {
 				err = sk_psock_tls_strp_read(psock, skb);
 				if (err != __SK_PASS) {
@@ -1996,7 +2000,6 @@ ssize_t tls_sw_splice_read(struct socket *sock,  loff_t *ppos,
 	int err = 0;
 	long timeo;
 	int chunk;
-	bool zc = false;
 
 	lock_sock(sk);
 
@@ -2006,12 +2009,14 @@ ssize_t tls_sw_splice_read(struct socket *sock,  loff_t *ppos,
 	if (from_queue) {
 		skb = __skb_dequeue(&ctx->rx_list);
 	} else {
+		struct tls_decrypt_arg darg = {};
+
 		skb = tls_wait_data(sk, NULL, flags & SPLICE_F_NONBLOCK, timeo,
 				    &err);
 		if (!skb)
 			goto splice_read_end;
 
-		err = decrypt_skb_update(sk, skb, NULL, &zc, false);
+		err = decrypt_skb_update(sk, skb, NULL, &darg);
 		if (err < 0) {
 			tls_err_abort(sk, -EBADMSG);
 			goto splice_read_end;
-- 
2.34.1

[PATCH net-next 04/11] tls: rx: simplify async wait

[Jakub Kicinski]

Since we are protected from async completions by decrypt_compl_lock
we can drop the async_notify and reinit the completion before we
start waiting.

Signed-off-by: Jakub Kicinski <kuba@kernel.org>
---
 include/net/tls.h |  1 -
 net/tls/tls_sw.c  | 14 ++------------
 2 files changed, 2 insertions(+), 13 deletions(-)

diff --git a/include/net/tls.h b/include/net/tls.h
index a01c264e5f15..6fe78361c8c8 100644
--- a/include/net/tls.h
+++ b/include/net/tls.h
@@ -152,7 +152,6 @@ struct tls_sw_context_rx {
 	atomic_t decrypt_pending;
 	/* protect crypto_wait with decrypt_pending*/
 	spinlock_t decrypt_compl_lock;
-	bool async_notify;
 };
 
 struct tls_record_info {
diff --git a/net/tls/tls_sw.c b/net/tls/tls_sw.c
index ecf9a3832798..003f7c178cde 100644
--- a/net/tls/tls_sw.c
+++ b/net/tls/tls_sw.c
@@ -173,7 +173,6 @@ static void tls_decrypt_done(struct crypto_async_request *req, int err)
 	struct scatterlist *sg;
 	struct sk_buff *skb;
 	unsigned int pages;
-	int pending;
 
 	skb = (struct sk_buff *)req->data;
 	tls_ctx = tls_get_ctx(skb->sk);
@@ -221,9 +220,7 @@ static void tls_decrypt_done(struct crypto_async_request *req, int err)
 	kfree(aead_req);
 
 	spin_lock_bh(&ctx->decrypt_compl_lock);
-	pending = atomic_dec_return(&ctx->decrypt_pending);
-
-	if (!pending && ctx->async_notify)
+	if (!atomic_dec_return(&ctx->decrypt_pending))
 		complete(&ctx->async_wait.completion);
 	spin_unlock_bh(&ctx->decrypt_compl_lock);
 }
@@ -1940,7 +1937,7 @@ int tls_sw_recvmsg(struct sock *sk,
 	if (num_async) {
 		/* Wait for all previously submitted records to be decrypted */
 		spin_lock_bh(&ctx->decrypt_compl_lock);
-		ctx->async_notify = true;
+		reinit_completion(&ctx->async_wait.completion);
 		pending = atomic_read(&ctx->decrypt_pending);
 		spin_unlock_bh(&ctx->decrypt_compl_lock);
 		if (pending) {
@@ -1952,15 +1949,8 @@ int tls_sw_recvmsg(struct sock *sk,
 				decrypted = 0;
 				goto end;
 			}
-		} else {
-			reinit_completion(&ctx->async_wait.completion);
 		}
 
-		/* There can be no concurrent accesses, since we have no
-		 * pending decrypt operations
-		 */
-		WRITE_ONCE(ctx->async_notify, false);
-
 		/* Drain records from the rx_list & copy if required */
 		if (is_peek || is_kvec)
 			err = process_rx_list(ctx, msg, &control, &cmsg, copied,
-- 
2.34.1

[PATCH net-next 05/11] tls: rx: factor out writing ContentType to cmsg

[Jakub Kicinski]

cmsg can be filled in during rx_list processing or normal
receive. Consolidate the code.

We don't need to keep the boolean to track if the cmsg was
created. 0 is an invalid content type.

Signed-off-by: Jakub Kicinski <kuba@kernel.org>
---
 net/tls/tls_sw.c | 91 +++++++++++++++++++-----------------------------
 1 file changed, 36 insertions(+), 55 deletions(-)

diff --git a/net/tls/tls_sw.c b/net/tls/tls_sw.c
index 003f7c178cde..103a1aaca934 100644
--- a/net/tls/tls_sw.c
+++ b/net/tls/tls_sw.c
@@ -1635,6 +1635,29 @@ static bool tls_sw_advance_skb(struct sock *sk, struct sk_buff *skb,
 	return true;
 }
 
+static int tls_record_content_type(struct msghdr *msg, struct tls_msg *tlm,
+				   u8 *control)
+{
+	int err;
+
+	if (!*control) {
+		*control = tlm->control;
+		if (!*control)
+			return -EBADMSG;
+
+		err = put_cmsg(msg, SOL_TLS, TLS_GET_RECORD_TYPE,
+			       sizeof(*control), control);
+		if (*control != TLS_RECORD_TYPE_DATA) {
+			if (err || msg->msg_flags & MSG_CTRUNC)
+				return -EIO;
+		}
+	} else if (*control != tlm->control) {
+		return 0;
+	}
+
+	return 1;
+}
+
 /* This function traverses the rx_list in tls receive context to copies the
  * decrypted records into the buffer provided by caller zero copy is not
  * true. Further, the records are removed from the rx_list if it is not a peek
@@ -1643,31 +1666,23 @@ static bool tls_sw_advance_skb(struct sock *sk, struct sk_buff *skb,
 static int process_rx_list(struct tls_sw_context_rx *ctx,
 			   struct msghdr *msg,
 			   u8 *control,
-			   bool *cmsg,
 			   size_t skip,
 			   size_t len,
 			   bool zc,
 			   bool is_peek)
 {
 	struct sk_buff *skb = skb_peek(&ctx->rx_list);
-	u8 ctrl = *control;
-	u8 msgc = *cmsg;
 	struct tls_msg *tlm;
 	ssize_t copied = 0;
-
-	/* Set the record type in 'control' if caller didn't pass it */
-	if (!ctrl && skb) {
-		tlm = tls_msg(skb);
-		ctrl = tlm->control;
-	}
+	int err;
 
 	while (skip && skb) {
 		struct strp_msg *rxm = strp_msg(skb);
 		tlm = tls_msg(skb);
 
-		/* Cannot process a record of different type */
-		if (ctrl != tlm->control)
-			return 0;
+		err = tls_record_content_type(msg, tlm, control);
+		if (err <= 0)
+			return err;
 
 		if (skip < rxm->full_len)
 			break;
@@ -1683,27 +1698,12 @@ static int process_rx_list(struct tls_sw_context_rx *ctx,
 
 		tlm = tls_msg(skb);
 
-		/* Cannot process a record of different type */
-		if (ctrl != tlm->control)
-			return 0;
-
-		/* Set record type if not already done. For a non-data record,
-		 * do not proceed if record type could not be copied.
-		 */
-		if (!msgc) {
-			int cerr = put_cmsg(msg, SOL_TLS, TLS_GET_RECORD_TYPE,
-					    sizeof(ctrl), &ctrl);
-			msgc = true;
-			if (ctrl != TLS_RECORD_TYPE_DATA) {
-				if (cerr || msg->msg_flags & MSG_CTRUNC)
-					return -EIO;
-
-				*cmsg = msgc;
-			}
-		}
+		err = tls_record_content_type(msg, tlm, control);
+		if (err <= 0)
+			return err;
 
 		if (!zc || (rxm->full_len - skip) > len) {
-			int err = skb_copy_datagram_msg(skb, rxm->offset + skip,
+			err = skb_copy_datagram_msg(skb, rxm->offset + skip,
 						    msg, chunk);
 			if (err < 0)
 				return err;
@@ -1740,7 +1740,6 @@ static int process_rx_list(struct tls_sw_context_rx *ctx,
 		skb = next_skb;
 	}
 
-	*control = ctrl;
 	return copied;
 }
 
@@ -1762,7 +1761,6 @@ int tls_sw_recvmsg(struct sock *sk,
 	struct tls_msg *tlm;
 	struct sk_buff *skb;
 	ssize_t copied = 0;
-	bool cmsg = false;
 	int target, err = 0;
 	long timeo;
 	bool is_kvec = iov_iter_is_kvec(&msg->msg_iter);
@@ -1779,8 +1777,7 @@ int tls_sw_recvmsg(struct sock *sk,
 	bpf_strp_enabled = sk_psock_strp_enabled(psock);
 
 	/* Process pending decrypted records. It must be non-zero-copy */
-	err = process_rx_list(ctx, msg, &control, &cmsg, 0, len, false,
-			      is_peek);
+	err = process_rx_list(ctx, msg, &control, 0, len, false, is_peek);
 	if (err < 0) {
 		tls_err_abort(sk, err);
 		goto end;
@@ -1852,26 +1849,10 @@ int tls_sw_recvmsg(struct sock *sk,
 		 * is known just after record is dequeued from stream parser.
 		 * For tls1.3, we disable async.
 		 */
-
-		if (!control)
-			control = tlm->control;
-		else if (control != tlm->control)
+		err = tls_record_content_type(msg, tlm, &control);
+		if (err <= 0)
 			goto recv_end;
 
-		if (!cmsg) {
-			int cerr;
-
-			cerr = put_cmsg(msg, SOL_TLS, TLS_GET_RECORD_TYPE,
-					sizeof(control), &control);
-			cmsg = true;
-			if (control != TLS_RECORD_TYPE_DATA) {
-				if (cerr || msg->msg_flags & MSG_CTRUNC) {
-					err = -EIO;
-					goto recv_end;
-				}
-			}
-		}
-
 		if (async) {
 			/* TLS 1.2-only, to_decrypt must be text length */
 			chunk = min_t(int, to_decrypt, len);
@@ -1953,10 +1934,10 @@ int tls_sw_recvmsg(struct sock *sk,
 
 		/* Drain records from the rx_list & copy if required */
 		if (is_peek || is_kvec)
-			err = process_rx_list(ctx, msg, &control, &cmsg, copied,
+			err = process_rx_list(ctx, msg, &control, copied,
 					      decrypted, false, is_peek);
 		else
-			err = process_rx_list(ctx, msg, &control, &cmsg, 0,
+			err = process_rx_list(ctx, msg, &control, 0,
 					      decrypted, true, is_peek);
 		if (err < 0) {
 			tls_err_abort(sk, err);
-- 
2.34.1

[PATCH net-next 06/11] tls: rx: don't handle async in tls_sw_advance_skb()

[Jakub Kicinski]

tls_sw_advance_skb() caters to the async case when skb argument
is NULL. In that case it simply unpauses the strparser.

These are surprising semantics to a person reading the code,
and result in higher LoC, so inline the __strp_unpause and
only call tls_sw_advance_skb() when we actually move past
an skb.

Signed-off-by: Jakub Kicinski <kuba@kernel.org>
---
 net/tls/tls_sw.c | 22 +++++++++-------------
 1 file changed, 9 insertions(+), 13 deletions(-)

diff --git a/net/tls/tls_sw.c b/net/tls/tls_sw.c
index 103a1aaca934..6f17f599a6d4 100644
--- a/net/tls/tls_sw.c
+++ b/net/tls/tls_sw.c
@@ -1616,17 +1616,14 @@ static bool tls_sw_advance_skb(struct sock *sk, struct sk_buff *skb,
 {
 	struct tls_context *tls_ctx = tls_get_ctx(sk);
 	struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
+	struct strp_msg *rxm = strp_msg(skb);
 
-	if (skb) {
-		struct strp_msg *rxm = strp_msg(skb);
-
-		if (len < rxm->full_len) {
-			rxm->offset += len;
-			rxm->full_len -= len;
-			return false;
-		}
-		consume_skb(skb);
+	if (len < rxm->full_len) {
+		rxm->offset += len;
+		rxm->full_len -= len;
+		return false;
 	}
+	consume_skb(skb);
 
 	/* Finished with message */
 	ctx->recv_pkt = NULL;
@@ -1898,10 +1895,9 @@ int tls_sw_recvmsg(struct sock *sk,
 		/* For async or peek case, queue the current skb */
 		if (async || is_peek || retain_skb) {
 			skb_queue_tail(&ctx->rx_list, skb);
-			skb = NULL;
-		}
-
-		if (tls_sw_advance_skb(sk, skb, chunk)) {
+			ctx->recv_pkt = NULL;
+			__strp_unpause(&ctx->strp);
+		} else if (tls_sw_advance_skb(sk, skb, chunk)) {
 			/* Return full control message to
 			 * userspace before trying to parse
 			 * another message type
-- 
2.34.1

[PATCH net-next 07/11] tls: rx: don't track the async count

[Jakub Kicinski]

We track both if the last record was handled by async crypto
and how many records were async. This is not necessary. We
implicitly assume once crypto goes async it will stay that
way, otherwise we'd reorder records. So just track if we're
in async mode, the exact number of records is not necessary.

This change also forces us into "async" mode more consistently
in case crypto ever decided to interleave async and sync.

Signed-off-by: Jakub Kicinski <kuba@kernel.org>
---
 net/tls/tls_sw.c | 12 +++++-------
 1 file changed, 5 insertions(+), 7 deletions(-)

diff --git a/net/tls/tls_sw.c b/net/tls/tls_sw.c
index 6f17f599a6d4..183e5ec292a8 100644
--- a/net/tls/tls_sw.c
+++ b/net/tls/tls_sw.c
@@ -1751,13 +1751,13 @@ int tls_sw_recvmsg(struct sock *sk,
 	struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
 	struct tls_prot_info *prot = &tls_ctx->prot_info;
 	struct sk_psock *psock;
-	int num_async, pending;
 	unsigned char control = 0;
 	ssize_t decrypted = 0;
 	struct strp_msg *rxm;
 	struct tls_msg *tlm;
 	struct sk_buff *skb;
 	ssize_t copied = 0;
+	bool async = false;
 	int target, err = 0;
 	long timeo;
 	bool is_kvec = iov_iter_is_kvec(&msg->msg_iter);
@@ -1789,12 +1789,10 @@ int tls_sw_recvmsg(struct sock *sk,
 	timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
 
 	decrypted = 0;
-	num_async = 0;
 	while (len && (decrypted + copied < target || ctx->recv_pkt)) {
 		struct tls_decrypt_arg darg = {};
 		bool retain_skb = false;
 		int to_decrypt, chunk;
-		bool async;
 
 		skb = tls_wait_data(sk, psock, flags & MSG_DONTWAIT, timeo, &err);
 		if (!skb) {
@@ -1834,10 +1832,8 @@ int tls_sw_recvmsg(struct sock *sk,
 			goto recv_end;
 		}
 
-		if (err == -EINPROGRESS) {
+		if (err == -EINPROGRESS)
 			async = true;
-			num_async++;
-		}
 
 		/* If the type of records being processed is not known yet,
 		 * set it to record type just dequeued. If it is already known,
@@ -1911,7 +1907,9 @@ int tls_sw_recvmsg(struct sock *sk,
 	}
 
 recv_end:
-	if (num_async) {
+	if (async) {
+		int pending;
+
 		/* Wait for all previously submitted records to be decrypted */
 		spin_lock_bh(&ctx->decrypt_compl_lock);
 		reinit_completion(&ctx->async_wait.completion);
-- 
2.34.1

[PATCH net-next 08/11] tls: rx: pull most of zc check out of the loop

[Jakub Kicinski]

Most of the conditions deciding if zero-copy can be used
do not change throughout the iterations, so pre-calculate
them.

Signed-off-by: Jakub Kicinski <kuba@kernel.org>
---
 net/tls/tls_sw.c | 9 +++++----
 1 file changed, 5 insertions(+), 4 deletions(-)

diff --git a/net/tls/tls_sw.c b/net/tls/tls_sw.c
index 183e5ec292a8..5ad0b2505988 100644
--- a/net/tls/tls_sw.c
+++ b/net/tls/tls_sw.c
@@ -1763,6 +1763,7 @@ int tls_sw_recvmsg(struct sock *sk,
 	bool is_kvec = iov_iter_is_kvec(&msg->msg_iter);
 	bool is_peek = flags & MSG_PEEK;
 	bool bpf_strp_enabled;
+	bool zc_capable;
 
 	flags |= nonblock;
 
@@ -1788,6 +1789,8 @@ int tls_sw_recvmsg(struct sock *sk,
 	len = len - copied;
 	timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
 
+	zc_capable = !bpf_strp_enabled && !is_kvec && !is_peek &&
+		     prot->version != TLS_1_3_VERSION;
 	decrypted = 0;
 	while (len && (decrypted + copied < target || ctx->recv_pkt)) {
 		struct tls_decrypt_arg darg = {};
@@ -1814,10 +1817,8 @@ int tls_sw_recvmsg(struct sock *sk,
 
 		to_decrypt = rxm->full_len - prot->overhead_size;
 
-		if (to_decrypt <= len && !is_kvec && !is_peek &&
-		    tlm->control == TLS_RECORD_TYPE_DATA &&
-		    prot->version != TLS_1_3_VERSION &&
-		    !bpf_strp_enabled)
+		if (zc_capable && to_decrypt <= len &&
+		    tlm->control == TLS_RECORD_TYPE_DATA)
 			darg.zc = true;
 
 		/* Do not use async mode if record is non-data */
-- 
2.34.1

[PATCH net-next 09/11] tls: rx: inline consuming the skb at the end of the loop

[Jakub Kicinski]

tls_sw_advance_skb() always consumes the skb at the end of the loop.

To fall here the following must be true:

 !async && !is_peek && !retain_skb
   retain_skb => !zc && rxm->full_len > len
     # but non-full record implies !zc, so above can be simplified as
   retain_skb => rxm->full_len > len

 !async && !is_peek && !(rxm->full_len > len)
 !async && !is_peek && rxm->full_len <= len

tls_sw_advance_skb() returns false if len < rxm->full_len
which can't be true given conditions above.

Signed-off-by: Jakub Kicinski <kuba@kernel.org>
---
 net/tls/tls_sw.c | 29 +++++------------------------
 1 file changed, 5 insertions(+), 24 deletions(-)

diff --git a/net/tls/tls_sw.c b/net/tls/tls_sw.c
index 5ad0b2505988..3aa8fe1c6e77 100644
--- a/net/tls/tls_sw.c
+++ b/net/tls/tls_sw.c
@@ -1611,27 +1611,6 @@ int decrypt_skb(struct sock *sk, struct sk_buff *skb,
 	return decrypt_internal(sk, skb, NULL, sgout, &darg);
 }
 
-static bool tls_sw_advance_skb(struct sock *sk, struct sk_buff *skb,
-			       unsigned int len)
-{
-	struct tls_context *tls_ctx = tls_get_ctx(sk);
-	struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
-	struct strp_msg *rxm = strp_msg(skb);
-
-	if (len < rxm->full_len) {
-		rxm->offset += len;
-		rxm->full_len -= len;
-		return false;
-	}
-	consume_skb(skb);
-
-	/* Finished with message */
-	ctx->recv_pkt = NULL;
-	__strp_unpause(&ctx->strp);
-
-	return true;
-}
-
 static int tls_record_content_type(struct msghdr *msg, struct tls_msg *tlm,
 				   u8 *control)
 {
@@ -1894,7 +1873,11 @@ int tls_sw_recvmsg(struct sock *sk,
 			skb_queue_tail(&ctx->rx_list, skb);
 			ctx->recv_pkt = NULL;
 			__strp_unpause(&ctx->strp);
-		} else if (tls_sw_advance_skb(sk, skb, chunk)) {
+		} else {
+			consume_skb(skb);
+			ctx->recv_pkt = NULL;
+			__strp_unpause(&ctx->strp);
+
 			/* Return full control message to
 			 * userspace before trying to parse
 			 * another message type
@@ -1902,8 +1885,6 @@ int tls_sw_recvmsg(struct sock *sk,
 			msg->msg_flags |= MSG_EOR;
 			if (control != TLS_RECORD_TYPE_DATA)
 				goto recv_end;
-		} else {
-			break;
 		}
 	}
 
-- 
2.34.1

[PATCH net-next 10/11] tls: rx: clear ctx->recv_pkt earlier

[Jakub Kicinski]

Whatever we do in the loop the skb should not remain on as
ctx->recv_pkt afterwards. We can clear that pointer and
restart strparser earlier.

This adds overhead of extra linking and unlinking to rx_list
but that's not large (upcoming change will switch to unlocked
skb list operations).

Signed-off-by: Jakub Kicinski <kuba@kernel.org>
---
 net/tls/tls_sw.c | 16 +++++++---------
 1 file changed, 7 insertions(+), 9 deletions(-)

diff --git a/net/tls/tls_sw.c b/net/tls/tls_sw.c
index 3aa8fe1c6e77..71d8082647c8 100644
--- a/net/tls/tls_sw.c
+++ b/net/tls/tls_sw.c
@@ -1826,6 +1826,10 @@ int tls_sw_recvmsg(struct sock *sk,
 		if (err <= 0)
 			goto recv_end;
 
+		ctx->recv_pkt = NULL;
+		__strp_unpause(&ctx->strp);
+		skb_queue_tail(&ctx->rx_list, skb);
+
 		if (async) {
 			/* TLS 1.2-only, to_decrypt must be text length */
 			chunk = min_t(int, to_decrypt, len);
@@ -1840,10 +1844,9 @@ int tls_sw_recvmsg(struct sock *sk,
 				if (err != __SK_PASS) {
 					rxm->offset = rxm->offset + rxm->full_len;
 					rxm->full_len = 0;
+					skb_unlink(skb, &ctx->rx_list);
 					if (err == __SK_DROP)
 						consume_skb(skb);
-					ctx->recv_pkt = NULL;
-					__strp_unpause(&ctx->strp);
 					continue;
 				}
 			}
@@ -1869,14 +1872,9 @@ int tls_sw_recvmsg(struct sock *sk,
 		len -= chunk;
 
 		/* For async or peek case, queue the current skb */
-		if (async || is_peek || retain_skb) {
-			skb_queue_tail(&ctx->rx_list, skb);
-			ctx->recv_pkt = NULL;
-			__strp_unpause(&ctx->strp);
-		} else {
+		if (!(async || is_peek || retain_skb)) {
+			skb_unlink(skb, &ctx->rx_list);
 			consume_skb(skb);
-			ctx->recv_pkt = NULL;
-			__strp_unpause(&ctx->strp);
 
 			/* Return full control message to
 			 * userspace before trying to parse
-- 
2.34.1

[PATCH net-next 11/11] tls: rx: jump out for cases which need to leave skb on list

[Jakub Kicinski]

The current invese logic is harder to follow (and adds extra
tests to the fast path). We have to enumerate all cases which
need to keep the skb before consuming it. It's simpler to
jump out of the full record flow as we detect those cases.

This makes it clear that partial consumption and peek can
only reach end of the function thru the !zc case so move
the code up there.

Signed-off-by: Jakub Kicinski <kuba@kernel.org>
---
 net/tls/tls_sw.c | 43 ++++++++++++++++++++++---------------------
 1 file changed, 22 insertions(+), 21 deletions(-)

diff --git a/net/tls/tls_sw.c b/net/tls/tls_sw.c
index 71d8082647c8..2e8a896af81a 100644
--- a/net/tls/tls_sw.c
+++ b/net/tls/tls_sw.c
@@ -1773,7 +1773,6 @@ int tls_sw_recvmsg(struct sock *sk,
 	decrypted = 0;
 	while (len && (decrypted + copied < target || ctx->recv_pkt)) {
 		struct tls_decrypt_arg darg = {};
-		bool retain_skb = false;
 		int to_decrypt, chunk;
 
 		skb = tls_wait_data(sk, psock, flags & MSG_DONTWAIT, timeo, &err);
@@ -1833,12 +1832,17 @@ int tls_sw_recvmsg(struct sock *sk,
 		if (async) {
 			/* TLS 1.2-only, to_decrypt must be text length */
 			chunk = min_t(int, to_decrypt, len);
-			goto pick_next_record;
+leave_on_list:
+			decrypted += chunk;
+			len -= chunk;
+			continue;
 		}
 		/* TLS 1.3 may have updated the length by more than overhead */
 		chunk = rxm->full_len;
 
 		if (!darg.zc) {
+			bool partially_consumed = chunk > len;
+
 			if (bpf_strp_enabled) {
 				err = sk_psock_tls_strp_read(psock, skb);
 				if (err != __SK_PASS) {
@@ -1851,39 +1855,36 @@ int tls_sw_recvmsg(struct sock *sk,
 				}
 			}
 
-			if (chunk > len) {
-				retain_skb = true;
+			if (partially_consumed)
 				chunk = len;
-			}
 
 			err = skb_copy_datagram_msg(skb, rxm->offset,
 						    msg, chunk);
 			if (err < 0)
 				goto recv_end;
 
-			if (!is_peek) {
-				rxm->offset = rxm->offset + chunk;
-				rxm->full_len = rxm->full_len - chunk;
+			if (is_peek)
+				goto leave_on_list;
+
+			if (partially_consumed) {
+				rxm->offset += chunk;
+				rxm->full_len -= chunk;
+				goto leave_on_list;
 			}
 		}
 
-pick_next_record:
 		decrypted += chunk;
 		len -= chunk;
 
-		/* For async or peek case, queue the current skb */
-		if (!(async || is_peek || retain_skb)) {
-			skb_unlink(skb, &ctx->rx_list);
-			consume_skb(skb);
+		skb_unlink(skb, &ctx->rx_list);
+		consume_skb(skb);
 
-			/* Return full control message to
-			 * userspace before trying to parse
-			 * another message type
-			 */
-			msg->msg_flags |= MSG_EOR;
-			if (control != TLS_RECORD_TYPE_DATA)
-				goto recv_end;
-		}
+		/* Return full control message to userspace before trying
+		 * to parse another message type
+		 */
+		msg->msg_flags |= MSG_EOR;
+		if (control != TLS_RECORD_TYPE_DATA)
+			break;
 	}
 
 recv_end:
-- 
2.34.1

RE: [PATCH net-next 00/11] tls: rx: random refactoring part 2

[John Fastabend]

Jakub Kicinski wrote:
> TLS Rx refactoring. Part 2 of 3. This one focusing on the main loop.
> A couple of features to follow.
> 
> Jakub Kicinski (11):
>   tls: rx: drop unnecessary arguments from tls_setup_from_iter()
>   tls: rx: don't report text length from the bowels of decrypt
>   tls: rx: wrap decryption arguments in a structure
>   tls: rx: simplify async wait
>   tls: rx: factor out writing ContentType to cmsg
>   tls: rx: don't handle async in tls_sw_advance_skb()
>   tls: rx: don't track the async count
>   tls: rx: pull most of zc check out of the loop
>   tls: rx: inline consuming the skb at the end of the loop
>   tls: rx: clear ctx->recv_pkt earlier
>   tls: rx: jump out for cases which need to leave skb on list
> 
>  include/net/tls.h |   1 -
>  net/tls/tls_sw.c  | 264 ++++++++++++++++++----------------------------
>  2 files changed, 104 insertions(+), 161 deletions(-)
> 
> -- 
> 2.34.1
> 

Thanks for doing this Jakub much appreciated on my side and been on my todo
list for way too long.

Acked-by: John Fastabend <john.fastabend@gmail.com>

Re: [PATCH net-next 00/11] tls: rx: random refactoring part 2

[patchwork-bot+netdevbpf]

Hello:

This series was applied to netdev/net-next.git (master)
by David S. Miller <davem@davemloft.net>:

On Fri,  8 Apr 2022 11:31:23 -0700 you wrote:
> TLS Rx refactoring. Part 2 of 3. This one focusing on the main loop.
> A couple of features to follow.
> 
> Jakub Kicinski (11):
>   tls: rx: drop unnecessary arguments from tls_setup_from_iter()
>   tls: rx: don't report text length from the bowels of decrypt
>   tls: rx: wrap decryption arguments in a structure
>   tls: rx: simplify async wait
>   tls: rx: factor out writing ContentType to cmsg
>   tls: rx: don't handle async in tls_sw_advance_skb()
>   tls: rx: don't track the async count
>   tls: rx: pull most of zc check out of the loop
>   tls: rx: inline consuming the skb at the end of the loop
>   tls: rx: clear ctx->recv_pkt earlier
>   tls: rx: jump out for cases which need to leave skb on list
> 
> [...]

Here is the summary with links:
  - [net-next,01/11] tls: rx: drop unnecessary arguments from tls_setup_from_iter()
    https://git.kernel.org/netdev/net-next/c/d4bd88e67666
  - [net-next,02/11] tls: rx: don't report text length from the bowels of decrypt
    https://git.kernel.org/netdev/net-next/c/9bdf75ccffa6
  - [net-next,03/11] tls: rx: wrap decryption arguments in a structure
    https://git.kernel.org/netdev/net-next/c/4175eac37123
  - [net-next,04/11] tls: rx: simplify async wait
    https://git.kernel.org/netdev/net-next/c/37943f047bfb
  - [net-next,05/11] tls: rx: factor out writing ContentType to cmsg
    https://git.kernel.org/netdev/net-next/c/06554f4ffc25
  - [net-next,06/11] tls: rx: don't handle async in tls_sw_advance_skb()
    https://git.kernel.org/netdev/net-next/c/fc8da80f9906
  - [net-next,07/11] tls: rx: don't track the async count
    https://git.kernel.org/netdev/net-next/c/7da18bcc5e4c
  - [net-next,08/11] tls: rx: pull most of zc check out of the loop
    https://git.kernel.org/netdev/net-next/c/ba13609df18d
  - [net-next,09/11] tls: rx: inline consuming the skb at the end of the loop
    https://git.kernel.org/netdev/net-next/c/465ea7353567
  - [net-next,10/11] tls: rx: clear ctx->recv_pkt earlier
    https://git.kernel.org/netdev/net-next/c/b1a2c1786330
  - [net-next,11/11] tls: rx: jump out for cases which need to leave skb on list
    https://git.kernel.org/netdev/net-next/c/f940b6efb172

You are awesome, thank you!
-- 
Deet-doot-dot, I am a bot.
https://korg.docs.kernel.org/patchwork/pwbot.html

Re: [PATCH net-next 10/11] tls: rx: clear ctx->recv_pkt earlier

[Artem Savkov]

Hi,

On Fri, Apr 08, 2022 at 11:31:33AM -0700, Jakub Kicinski wrote:
> Whatever we do in the loop the skb should not remain on as
> ctx->recv_pkt afterwards. We can clear that pointer and
> restart strparser earlier.
> 
> This adds overhead of extra linking and unlinking to rx_list
> but that's not large (upcoming change will switch to unlocked
> skb list operations).
> 
> Signed-off-by: Jakub Kicinski <kuba@kernel.org>
> ---
>  net/tls/tls_sw.c | 16 +++++++---------
>  1 file changed, 7 insertions(+), 9 deletions(-)
> 
> diff --git a/net/tls/tls_sw.c b/net/tls/tls_sw.c
> index 3aa8fe1c6e77..71d8082647c8 100644
> --- a/net/tls/tls_sw.c
> +++ b/net/tls/tls_sw.c
> @@ -1826,6 +1826,10 @@ int tls_sw_recvmsg(struct sock *sk,
>  		if (err <= 0)
>  			goto recv_end;
>  
> +		ctx->recv_pkt = NULL;
> +		__strp_unpause(&ctx->strp);
> +		skb_queue_tail(&ctx->rx_list, skb);
> +
>  		if (async) {
>  			/* TLS 1.2-only, to_decrypt must be text length */
>  			chunk = min_t(int, to_decrypt, len);
> @@ -1840,10 +1844,9 @@ int tls_sw_recvmsg(struct sock *sk,
>  				if (err != __SK_PASS) {
>  					rxm->offset = rxm->offset + rxm->full_len;
>  					rxm->full_len = 0;
> +					skb_unlink(skb, &ctx->rx_list);

I have found this patch to be breaking __SK_REDIRECT case as queuing the skb
onto psock_other->ingress_skb while already having it in rx_list breaks
the latter (at least) ultimately resulting in a null pointer dereference
in __skb_queue_purge() during tls_sw_release_resources_rx.

This is easily reproducible with selftests/bpf/test_sockmap

[ 4363.845728][  T370] BUG: kernel NULL pointer dereference, address: 0000000000000008
[ 4363.851749][  T370] #PF: supervisor write access in kernel mode
[ 4363.856364][  T370] #PF: error_code(0x0002) - not-present page
[ 4363.862678][  T370] PGD 0 P4D 0
[ 4363.866893][  T370] Oops: 0002 [#1] PREEMPT SMP PTI
[ 4363.871576][  T370] CPU: 1 PID: 370 Comm: test_sockmap Not tainted 5.18.0-rc3.bpf-00792-g98a90feedc4e-dirty #649 22ffa66a3068b39367ed1f9d77b2a5bb8f7921a2
[ 4363.882164][  T370] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.15.0-1.fc35 04/01/2014
[ 4363.890561][  T370] RIP: 0010:tls_sw_release_resources_rx+0xac/0x120
[ 4363.895582][  T370] Code: 00 48 89 ef be 01 00 00 00 83 e8 01 89 83 e8 01 00 00 48 8b 55 00 48 8b 45 08 48 c7 45 00 00 00 00 00 48 c7 45 08 00 00 00 00 <48> 89 42 08 48 89 10 e8 48 ed d4 ff 48 8b ab d8 01 00 00 49 39 ec
[ 4363.912545][  T370] RSP: 0018:ffffc90000cabb98 EFLAGS: 00010297
[ 4363.917672][  T370] RAX: 0000000000000000 RBX: ffff888113e10c00 RCX: 0000000000000000
[ 4363.922860][  T370] RDX: 0000000000000000 RSI: 0000000000000001 RDI: ffff888100810400
[ 4363.929092][  T370] RBP: ffff888100810400 R08: 0000000000000040 R09: ffff8881099bc740
[ 4363.935344][  T370] R10: 0000000000000001 R11: ffff8881099cebc8 R12: ffff888113e10dd8
[ 4363.940657][  T370] R13: ffff88810759c500 R14: ffff88810759c8a0 R15: 0000000000000001
[ 4363.944835][  T370] FS:  0000000000000000(0000) GS:ffff88813bd00000(0000) knlGS:0000000000000000
[ 4363.948247][  T370] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
[ 4363.949967][  T370] CR2: 0000000000000008 CR3: 0000000003e72005 CR4: 0000000000170ea0
[ 4363.952346][  T370] Call Trace:
[ 4363.953458][  T370]  <TASK>
[ 4363.954472][  T370]  tls_sk_proto_close+0x29e/0x3b0
[ 4363.956060][  T370]  ? __sock_release+0xf0/0xf0
[ 4363.957432][  T370]  inet_release+0x63/0xb0
[ 4363.958857][  T370]  __sock_release+0x56/0xf0
[ 4363.960319][  T370]  sock_close+0x1d/0x30
[ 4363.961668][  T370]  __fput+0xd3/0x360
[ 4363.962969][  T370]  task_work_run+0x7b/0xc0
[ 4363.964415][  T370]  do_exit+0x25f/0x610
[ 4363.965577][  T370]  do_group_exit+0x44/0xe0
[ 4363.966802][  T370]  get_signal+0xa48/0xa60
[ 4363.967993][  T370]  arch_do_signal_or_restart+0x25/0x100
[ 4363.969706][  T370]  ? lockdep_hardirqs_on+0xbf/0x130
[ 4363.971241][  T370]  exit_to_user_mode_prepare+0x123/0x190
[ 4363.972904][  T370]  syscall_exit_to_user_mode+0x19/0x50
[ 4363.974630][  T370]  do_syscall_64+0x69/0x80
[ 4363.976075][  T370]  ? do_syscall_64+0x69/0x80
[ 4363.977562][  T370]  ? lockdep_hardirqs_on+0xbf/0x130
[ 4363.979998][  T370]  ? do_syscall_64+0x69/0x80
[ 4363.981466][  T370]  ? syscall_exit_to_user_mode+0x1e/0x50
[ 4363.983414][  T370]  ? do_syscall_64+0x69/0x80
[ 4363.984805][  T370]  ? lockdep_hardirqs_on+0xbf/0x130
[ 4363.986498][  T370]  ? do_syscall_64+0x69/0x80
[ 4363.988110][  T370]  ? exc_page_fault+0xbd/0x170
[ 4363.989749][  T370]  ? asm_exc_page_fault+0x8/0x30
[ 4363.991203][  T370]  entry_SYSCALL_64_after_hwframe+0x44/0xae
[ 4363.993175][  T370] RIP: 0033:0x7ffff7e45a26
[ 4363.994479][  T370] Code: Unable to access opcode bytes at RIP 0x7ffff7e459fc.
[ 4363.996791][  T370] RSP: 002b:00007fffffffe0f8 EFLAGS: 00000246 ORIG_RAX: 000000000000003d
[ 4363.999448][  T370] RAX: fffffffffffffe00 RBX: 000000000045c720 RCX: 00007ffff7e45a26
[ 4364.001904][  T370] RDX: 0000000000000000 RSI: 00007fffffffe11c RDI: 0000000000000238
[ 4364.004265][  T370] RBP: 00007fffffffe180 R08: 0000000000000000 R09: 00007ffff7f0d0c0
[ 4364.006574][  T370] R10: 0000000000000000 R11: 0000000000000246 R12: 000000000040a970
[ 4364.009478][  T370] R13: 0000000000000000 R14: 0000000000000000 R15: 0000000000000000
[ 4364.011687][  T370]  </TASK>
[ 4364.012530][  T370] Modules linked in:
[ 4364.013470][  T370] CR2: 0000000000000008
[ 4364.014658][  T370] ---[ end trace 0000000000000000 ]---


>  					if (err == __SK_DROP)
>  						consume_skb(skb);
> -					ctx->recv_pkt = NULL;
> -					__strp_unpause(&ctx->strp);
>  					continue;
>  				}
>  			}
> @@ -1869,14 +1872,9 @@ int tls_sw_recvmsg(struct sock *sk,
>  		len -= chunk;
>  
>  		/* For async or peek case, queue the current skb */
> -		if (async || is_peek || retain_skb) {
> -			skb_queue_tail(&ctx->rx_list, skb);
> -			ctx->recv_pkt = NULL;
> -			__strp_unpause(&ctx->strp);
> -		} else {
> +		if (!(async || is_peek || retain_skb)) {
> +			skb_unlink(skb, &ctx->rx_list);
>  			consume_skb(skb);
> -			ctx->recv_pkt = NULL;
> -			__strp_unpause(&ctx->strp);
>  
>  			/* Return full control message to
>  			 * userspace before trying to parse
> -- 
> 2.34.1
> 

-- 
 Artem