diff mbox series

[net,v2,3/7] cxgb4/ch_ktls: creating skbs causes panic

Message ID 20201023053134.26021-4-rohitm@chelsio.com
State Changes Requested
Delegated to: David Miller
Headers show
Series cxgb4/ch_ktls: Fixes in nic tls code | expand

Checks

Context Check Description
jkicinski/stable success Stable not CCed
jkicinski/header_inline success Link
jkicinski/build_allmodconfig_warn success Errors and warnings before: 42 this patch: 42
jkicinski/checkpatch success total: 0 errors, 0 warnings, 0 checks, 954 lines checked
jkicinski/verify_fixes success Link
jkicinski/kdoc success Errors and warnings before: 0 this patch: 0
jkicinski/build_32bit success Errors and warnings before: 42 this patch: 42
jkicinski/module_param success Was 0 now: 0
jkicinski/verify_signedoff success Link
jkicinski/source_inline success Was 0 now: 0
jkicinski/subject_prefix success Link
jkicinski/tree_selection success Clearly marked for net
jkicinski/patch_count success Link
jkicinski/fixes_present success Link
jkicinski/cover_letter success Link

Commit Message

Rohit Maheshwari Oct. 23, 2020, 5:31 a.m. UTC
Creating SKB per tls record and freeing the original one causes
panic. There will be race if connection reset is requested. By
freeing original skb, refcnt will be decremented and that means,
there is no pending record to send, and so tls_dev_del will be
requested in control path while SKB of related connection is in
queue.
 Better approach is to use same SKB to send one record (partial
data) at a time. We still have to create a new SKB when partial
last part of a record is requested.
 This fix introduces new API cxgb4_write_partial_sgl() to send
partial part of skb. Present cxgb4_write_sgl can only provide
feasibility to start from an offset which limits to header only
and it can write sgls for the whole skb len. But this new API
will help in both. It can start from any offset and can end
writing in middle of the skb.

Fixes: 429765a149f1 ("chcr: handle partial end part of a record")
Signed-off-by: Rohit Maheshwari <rohitm@chelsio.com>
---
 drivers/net/ethernet/chelsio/cxgb4/cxgb4.h    |   3 +
 drivers/net/ethernet/chelsio/cxgb4/sge.c      | 108 ++++
 .../chelsio/inline_crypto/ch_ktls/chcr_ktls.c | 543 +++++++-----------
 3 files changed, 327 insertions(+), 327 deletions(-)

Comments

Jakub Kicinski Oct. 24, 2020, 2:05 a.m. UTC | #1
On Fri, 23 Oct 2020 11:01:30 +0530 Rohit Maheshwari wrote:
> Creating SKB per tls record and freeing the original one causes
> panic. There will be race if connection reset is requested. By
> freeing original skb, refcnt will be decremented and that means,
> there is no pending record to send, and so tls_dev_del will be
> requested in control path while SKB of related connection is in
> queue.

Can't you make your new skbs take a reference on the socket, then?

> @@ -1672,42 +1572,60 @@ static int chcr_end_part_handler(struct chcr_ktls_info *tx_info,
>  				 struct tls_record_info *record,
>  				 u32 tcp_seq, int mss, bool tcp_push_no_fin,
>  				 struct sge_eth_txq *q,
> -				 u32 tls_end_offset, bool last_wr)
> +				 u32 tls_end_offset, u32 skb_offset)
>  {
>  	struct sk_buff *nskb = NULL;
> +	bool is_last_wr = false;
> +	int ret;
> +
> +	if (skb_offset + tls_end_offset == skb->len)
> +		is_last_wr = true;
> +
>  	/* check if it is a complete record */
>  	if (tls_end_offset == record->len) {
>  		nskb = skb;
>  		atomic64_inc(&tx_info->adap->ch_ktls_stats.ktls_tx_complete_pkts);
>  	} else {
> -		dev_kfree_skb_any(skb);
> -
> +		/* TAG needs to be calculated so, need to send complete record,
> +		 * free the original skb and send a new one.
> +		 */
>  		nskb = alloc_skb(0, GFP_KERNEL);
> -		if (!nskb)

Please add a patch that fixes the misuses of GFP_KERNEL as pointed out
by Julia. Make it a separate patch in this series, before this one.
diff mbox series

Patch

diff --git a/drivers/net/ethernet/chelsio/cxgb4/cxgb4.h b/drivers/net/ethernet/chelsio/cxgb4/cxgb4.h
index 3352dad6ca99..27308600da15 100644
--- a/drivers/net/ethernet/chelsio/cxgb4/cxgb4.h
+++ b/drivers/net/ethernet/chelsio/cxgb4/cxgb4.h
@@ -2124,6 +2124,9 @@  void cxgb4_inline_tx_skb(const struct sk_buff *skb, const struct sge_txq *q,
 void cxgb4_write_sgl(const struct sk_buff *skb, struct sge_txq *q,
 		     struct ulptx_sgl *sgl, u64 *end, unsigned int start,
 		     const dma_addr_t *addr);
+void cxgb4_write_partial_sgl(const struct sk_buff *skb, struct sge_txq *q,
+			     struct ulptx_sgl *sgl, u64 *end,
+			     const dma_addr_t *addr, u32 start, u32 send_len);
 void cxgb4_ring_tx_db(struct adapter *adap, struct sge_txq *q, int n);
 int t4_set_vlan_acl(struct adapter *adap, unsigned int mbox, unsigned int vf,
 		    u16 vlan);
diff --git a/drivers/net/ethernet/chelsio/cxgb4/sge.c b/drivers/net/ethernet/chelsio/cxgb4/sge.c
index 01bd9c0dfe4e..196652a114c5 100644
--- a/drivers/net/ethernet/chelsio/cxgb4/sge.c
+++ b/drivers/net/ethernet/chelsio/cxgb4/sge.c
@@ -890,6 +890,114 @@  void cxgb4_write_sgl(const struct sk_buff *skb, struct sge_txq *q,
 }
 EXPORT_SYMBOL(cxgb4_write_sgl);
 
+/*	cxgb4_write_partial_sgl - populate SGL for partial packet
+ *	@skb: the packet
+ *	@q: the Tx queue we are writing into
+ *	@sgl: starting location for writing the SGL
+ *	@end: points right after the end of the SGL
+ *	@addr: the list of bus addresses for the SGL elements
+ *	@start: start offset in the SKB where partial data starts
+ *	@len: length of data from @start to send out
+ *
+ *	This API will handle sending out partial data of a skb if required.
+ *	Unlike cxgb4_write_sgl, @start can be any offset into the skb data,
+ *	and @len will decide how much data after @start offset to send out.
+ */
+void cxgb4_write_partial_sgl(const struct sk_buff *skb, struct sge_txq *q,
+			     struct ulptx_sgl *sgl, u64 *end,
+			     const dma_addr_t *addr, u32 start, u32 len)
+{
+	struct ulptx_sge_pair buf[MAX_SKB_FRAGS / 2 + 1] = {0}, *to;
+	u32 frag_size, skb_linear_data_len = skb_headlen(skb);
+	struct skb_shared_info *si = skb_shinfo(skb);
+	u8 i = 0, frag_idx = 0, nfrags = 0;
+	skb_frag_t *frag;
+
+	/* Fill the first SGL either from linear data or from partial
+	 * frag based on @start.
+	 */
+	if (unlikely(start < skb_linear_data_len)) {
+		frag_size = min(len, skb_linear_data_len - start);
+		sgl->len0 = htonl(frag_size);
+		sgl->addr0 = cpu_to_be64(addr[0] + start);
+		len -= frag_size;
+		nfrags++;
+	} else {
+		start -= skb_linear_data_len;
+		frag = &si->frags[frag_idx];
+		frag_size = skb_frag_size(frag);
+		/* find the first frag */
+		while (start >= frag_size) {
+			start -= frag_size;
+			frag_idx++;
+			frag = &si->frags[frag_idx];
+			frag_size = skb_frag_size(frag);
+		}
+
+		frag_size = min(len, skb_frag_size(frag) - start);
+		sgl->len0 = cpu_to_be32(frag_size);
+		sgl->addr0 = cpu_to_be64(addr[frag_idx + 1] + start);
+		len -= frag_size;
+		nfrags++;
+		frag_idx++;
+	}
+
+	/* If the entire partial data fit in one SGL, then send it out
+	 * now.
+	 */
+	if (!len)
+		goto done;
+
+	/* Most of the complexity below deals with the possibility we hit the
+	 * end of the queue in the middle of writing the SGL.  For this case
+	 * only we create the SGL in a temporary buffer and then copy it.
+	 */
+	to = (u8 *)end > (u8 *)q->stat ? buf : sgl->sge;
+
+	/* If the skb couldn't fit in first SGL completely, fill the
+	 * rest of the frags in subsequent SGLs. Note that each SGL
+	 * pair can store 2 frags.
+	 */
+	while (len) {
+		frag_size = min(len, skb_frag_size(&si->frags[frag_idx]));
+		to->len[i & 1] = cpu_to_be32(frag_size);
+		to->addr[i & 1] = cpu_to_be64(addr[frag_idx + 1]);
+		if (i && (i & 1))
+			to++;
+		nfrags++;
+		frag_idx++;
+		i++;
+		len -= frag_size;
+	}
+
+	/* If we ended in an odd boundary, then set the second SGL's
+	 * length in the pair to 0.
+	 */
+	if (i & 1)
+		to->len[1] = cpu_to_be32(0);
+
+	/* Copy from temporary buffer to Tx ring, in case we hit the
+	 * end of the queue in the middle of writing the SGL.
+	 */
+	if (unlikely((u8 *)end > (u8 *)q->stat)) {
+		u32 part0 = (u8 *)q->stat - (u8 *)sgl->sge, part1;
+
+		if (likely(part0))
+			memcpy(sgl->sge, buf, part0);
+		part1 = (u8 *)end - (u8 *)q->stat;
+		memcpy(q->desc, (u8 *)buf + part0, part1);
+		end = (void *)q->desc + part1;
+	}
+
+	/* 0-pad to multiple of 16 */
+	if ((uintptr_t)end & 8)
+		*end = 0;
+done:
+	sgl->cmd_nsge = htonl(ULPTX_CMD_V(ULP_TX_SC_DSGL) |
+			ULPTX_NSGE_V(nfrags));
+}
+EXPORT_SYMBOL(cxgb4_write_partial_sgl);
+
 /* This function copies 64 byte coalesced work request to
  * memory mapped BAR2 space. For coalesced WR SGE fetches
  * data from the FIFO instead of from Host.
diff --git a/drivers/net/ethernet/chelsio/inline_crypto/ch_ktls/chcr_ktls.c b/drivers/net/ethernet/chelsio/inline_crypto/ch_ktls/chcr_ktls.c
index c69c3901a707..9fa8e40ca0bf 100644
--- a/drivers/net/ethernet/chelsio/inline_crypto/ch_ktls/chcr_ktls.c
+++ b/drivers/net/ethernet/chelsio/inline_crypto/ch_ktls/chcr_ktls.c
@@ -14,6 +14,50 @@ 
 static LIST_HEAD(uld_ctx_list);
 static DEFINE_MUTEX(dev_mutex);
 
+/* chcr_get_nfrags_to_send: get the remaining nfrags after start offset
+ * @skb: skb
+ * @start: start offset.
+ * @len: how much data to send after @start
+ */
+static int chcr_get_nfrags_to_send(struct sk_buff *skb, u32 start, u32 len)
+{
+	struct skb_shared_info *si = skb_shinfo(skb);
+	u32 frag_size, skb_linear_data_len = skb_headlen(skb);
+	u8 nfrags = 0, frag_idx = 0;
+	skb_frag_t *frag;
+
+	/* if its a linear skb then return 1 */
+	if (!skb_is_nonlinear(skb))
+		return 1;
+
+	if (unlikely(start < skb_linear_data_len)) {
+		frag_size = min(len, skb_linear_data_len - start);
+		start = 0;
+	} else {
+		start -= skb_linear_data_len;
+
+		frag = &si->frags[frag_idx];
+		frag_size = skb_frag_size(frag);
+		while (start >= frag_size) {
+			start -= frag_size;
+			frag_idx++;
+			frag = &si->frags[frag_idx];
+			frag_size = skb_frag_size(frag);
+		}
+		frag_size = min(len, skb_frag_size(frag) - start);
+	}
+	len -= frag_size;
+	nfrags++;
+
+	while (len) {
+		frag_size = min(len, skb_frag_size(&si->frags[frag_idx]));
+		len -= frag_size;
+		nfrags++;
+		frag_idx++;
+	}
+	return nfrags;
+}
+
 static int chcr_init_tcb_fields(struct chcr_ktls_info *tx_info);
 /*
  * chcr_ktls_save_keys: calculate and save crypto keys.
@@ -865,38 +909,6 @@  static int chcr_ktls_xmit_tcb_cpls(struct chcr_ktls_info *tx_info,
 	return 0;
 }
 
-/*
- * chcr_ktls_skb_copy
- * @nskb - new skb where the frags to be added.
- * @skb - old skb from which frags will be copied.
- */
-static void chcr_ktls_skb_copy(struct sk_buff *skb, struct sk_buff *nskb)
-{
-	int i;
-
-	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
-		skb_shinfo(nskb)->frags[i] = skb_shinfo(skb)->frags[i];
-		__skb_frag_ref(&skb_shinfo(nskb)->frags[i]);
-	}
-
-	skb_shinfo(nskb)->nr_frags = skb_shinfo(skb)->nr_frags;
-	nskb->len += skb->data_len;
-	nskb->data_len = skb->data_len;
-	nskb->truesize += skb->data_len;
-}
-
-/*
- * chcr_ktls_get_tx_flits
- * returns number of flits to be sent out, it includes key context length, WR
- * size and skb fragments.
- */
-static unsigned int
-chcr_ktls_get_tx_flits(const struct sk_buff *skb, unsigned int key_ctx_len)
-{
-	return chcr_sgl_len(skb_shinfo(skb)->nr_frags) +
-	       DIV_ROUND_UP(key_ctx_len + CHCR_KTLS_WR_SIZE, 8);
-}
-
 /*
  * chcr_ktls_check_tcp_options: To check if there is any TCP option availbale
  * other than timestamp.
@@ -940,7 +952,6 @@  chcr_ktls_check_tcp_options(struct tcphdr *tcp)
  * @tx_info - driver specific tls info.
  * @skb - skb contains partial record..
  * @q - TX queue.
- * @tx_chan - channel number.
  * return: NETDEV_TX_OK/NETDEV_TX_BUSY.
  */
 static int chcr_ktls_write_tcp_options(struct chcr_ktls_info *tx_info,
@@ -1043,71 +1054,6 @@  static int chcr_ktls_write_tcp_options(struct chcr_ktls_info *tx_info,
 	return 0;
 }
 
-/* chcr_ktls_skb_shift - Shifts request length paged data from skb to another.
- * @tgt- buffer into which tail data gets added
- * @skb- buffer from which the paged data comes from
- * @shiftlen- shift up to this many bytes
- */
-static int chcr_ktls_skb_shift(struct sk_buff *tgt, struct sk_buff *skb,
-			       int shiftlen)
-{
-	skb_frag_t *fragfrom, *fragto;
-	int from, to, todo;
-
-	WARN_ON(shiftlen > skb->data_len);
-
-	todo = shiftlen;
-	from = 0;
-	to = 0;
-	fragfrom = &skb_shinfo(skb)->frags[from];
-
-	while ((todo > 0) && (from < skb_shinfo(skb)->nr_frags)) {
-		fragfrom = &skb_shinfo(skb)->frags[from];
-		fragto = &skb_shinfo(tgt)->frags[to];
-
-		if (todo >= skb_frag_size(fragfrom)) {
-			*fragto = *fragfrom;
-			todo -= skb_frag_size(fragfrom);
-			from++;
-			to++;
-
-		} else {
-			__skb_frag_ref(fragfrom);
-			skb_frag_page_copy(fragto, fragfrom);
-			skb_frag_off_copy(fragto, fragfrom);
-			skb_frag_size_set(fragto, todo);
-
-			skb_frag_off_add(fragfrom, todo);
-			skb_frag_size_sub(fragfrom, todo);
-			todo = 0;
-
-			to++;
-			break;
-		}
-	}
-
-	/* Ready to "commit" this state change to tgt */
-	skb_shinfo(tgt)->nr_frags = to;
-
-	/* Reposition in the original skb */
-	to = 0;
-	while (from < skb_shinfo(skb)->nr_frags)
-		skb_shinfo(skb)->frags[to++] = skb_shinfo(skb)->frags[from++];
-
-	skb_shinfo(skb)->nr_frags = to;
-
-	WARN_ON(todo > 0 && !skb_shinfo(skb)->nr_frags);
-
-	skb->len -= shiftlen;
-	skb->data_len -= shiftlen;
-	skb->truesize -= shiftlen;
-	tgt->len += shiftlen;
-	tgt->data_len += shiftlen;
-	tgt->truesize += shiftlen;
-
-	return shiftlen;
-}
-
 /*
  * chcr_ktls_xmit_wr_complete: This sends out the complete record. If an skb
  * received has partial end part of the record, send out the complete record, so
@@ -1121,9 +1067,11 @@  static int chcr_ktls_skb_shift(struct sk_buff *tgt, struct sk_buff *skb,
  * return: NETDEV_TX_BUSY/NET_TX_OK.
  */
 static int chcr_ktls_xmit_wr_complete(struct sk_buff *skb,
+				      bool is_last_wr,
 				      struct chcr_ktls_info *tx_info,
 				      struct sge_eth_txq *q, u32 tcp_seq,
-				      bool tcp_push, u32 mss)
+				      bool tcp_push, u32 mss, u32 data_len,
+				      u32 skb_offset, u32 nfrags)
 {
 	u32 len16, wr_mid = 0, flits = 0, ndesc, cipher_start;
 	struct adapter *adap = tx_info->adap;
@@ -1138,7 +1086,8 @@  static int chcr_ktls_xmit_wr_complete(struct sk_buff *skb,
 	u64 *end;
 
 	/* get the number of flits required */
-	flits = chcr_ktls_get_tx_flits(skb, tx_info->key_ctx_len);
+	flits = chcr_sgl_len(nfrags) +
+		DIV_ROUND_UP(tx_info->key_ctx_len + CHCR_KTLS_WR_SIZE, 8);
 	/* number of descriptors */
 	ndesc = chcr_flits_to_desc(flits);
 	/* check if enough credits available */
@@ -1167,6 +1116,9 @@  static int chcr_ktls_xmit_wr_complete(struct sk_buff *skb,
 		return NETDEV_TX_BUSY;
 	}
 
+	if (!is_last_wr)
+		skb_get(skb);
+
 	pos = &q->q.desc[q->q.pidx];
 	end = (u64 *)pos + flits;
 	/* FW_ULPTX_WR */
@@ -1199,7 +1151,7 @@  static int chcr_ktls_xmit_wr_complete(struct sk_buff *skb,
 		      CPL_TX_SEC_PDU_CPLLEN_V(CHCR_CPL_TX_SEC_PDU_LEN_64BIT) |
 		      CPL_TX_SEC_PDU_PLACEHOLDER_V(1) |
 		      CPL_TX_SEC_PDU_IVINSRTOFST_V(TLS_HEADER_SIZE + 1));
-	cpl->pldlen = htonl(skb->data_len);
+	cpl->pldlen = htonl(data_len);
 
 	/* encryption should start after tls header size + iv size */
 	cipher_start = TLS_HEADER_SIZE + tx_info->iv_size + 1;
@@ -1241,13 +1193,12 @@  static int chcr_ktls_xmit_wr_complete(struct sk_buff *skb,
 	/* CPL_TX_DATA */
 	tx_data = (void *)pos;
 	OPCODE_TID(tx_data) = htonl(MK_OPCODE_TID(CPL_TX_DATA, tx_info->tid));
-	tx_data->len = htonl(TX_DATA_MSS_V(mss) | TX_LENGTH_V(skb->data_len));
+	tx_data->len = htonl(TX_DATA_MSS_V(mss) | TX_LENGTH_V(data_len));
 
 	tx_data->rsvd = htonl(tcp_seq);
 
-	tx_data->flags = htonl(TX_BYPASS_F);
-	if (tcp_push)
-		tx_data->flags |= htonl(TX_PUSH_F | TX_SHOVE_F);
+	tx_data->flags = htonl(TX_BYPASS_F |
+			       (tcp_push ? TX_PUSH_F | TX_SHOVE_F : 0));
 
 	/* check left again, it might go beyond queue limit */
 	pos = tx_data + 1;
@@ -1261,8 +1212,8 @@  static int chcr_ktls_xmit_wr_complete(struct sk_buff *skb,
 	}
 
 	/* send the complete packet except the header */
-	cxgb4_write_sgl(skb, &q->q, pos, end, skb->len - skb->data_len,
-			sgl_sdesc->addr);
+	cxgb4_write_partial_sgl(skb, &q->q, pos, end, sgl_sdesc->addr,
+				skb_offset, data_len);
 	sgl_sdesc->skb = skb;
 
 	chcr_txq_advance(&q->q, ndesc);
@@ -1294,11 +1245,11 @@  static int chcr_ktls_xmit_wr_short(struct sk_buff *skb,
 				   struct sge_eth_txq *q,
 				   u32 tcp_seq, bool tcp_push, u32 mss,
 				   u32 tls_rec_offset, u8 *prior_data,
-				   u32 prior_data_len)
+				   u32 prior_data_len, u32 data_len,
+				   u32 skb_offset)
 {
+	u32 len16, wr_mid = 0, cipher_start, nfrags, flits = 0, ndesc;
 	struct adapter *adap = tx_info->adap;
-	u32 len16, wr_mid = 0, cipher_start;
-	unsigned int flits = 0, ndesc;
 	int credits, left, last_desc;
 	struct tx_sw_desc *sgl_sdesc;
 	struct cpl_tx_data *tx_data;
@@ -1310,14 +1261,17 @@  static int chcr_ktls_xmit_wr_short(struct sk_buff *skb,
 	void *pos;
 	u64 *end;
 
-	/* get the number of flits required, it's a partial record so 2 flits
-	 * (AES_BLOCK_SIZE) will be added.
+	nfrags = chcr_get_nfrags_to_send(skb, skb_offset, data_len);
+	flits = chcr_sgl_len(nfrags) +
+		DIV_ROUND_UP(tx_info->key_ctx_len + CHCR_KTLS_WR_SIZE, 8);
+	/* it's a partial record so 2 flits (AES_BLOCK_SIZE) will also be
+	 * added.
 	 */
-	flits = chcr_ktls_get_tx_flits(skb, tx_info->key_ctx_len) + 2;
+	flits += 2;
 	/* get the correct 8 byte IV of this record */
 	iv_record = cpu_to_be64(tx_info->iv + tx_info->record_no);
 	/* If it's a middle record and not 16 byte aligned to run AES CTR, need
-	 * to make it 16 byte aligned. So atleadt 2 extra flits of immediate
+	 * to make it 16 byte aligned. So atleast 2 extra flits of immediate
 	 * data will be added.
 	 */
 	if (prior_data_len)
@@ -1385,7 +1339,7 @@  static int chcr_ktls_xmit_wr_short(struct sk_buff *skb,
 		htonl(CPL_TX_SEC_PDU_OPCODE_V(CPL_TX_SEC_PDU) |
 		      CPL_TX_SEC_PDU_CPLLEN_V(CHCR_CPL_TX_SEC_PDU_LEN_64BIT) |
 		      CPL_TX_SEC_PDU_IVINSRTOFST_V(1));
-	cpl->pldlen = htonl(skb->data_len + AES_BLOCK_LEN + prior_data_len);
+	cpl->pldlen = htonl(data_len + AES_BLOCK_LEN + prior_data_len);
 	cpl->aadstart_cipherstop_hi =
 		htonl(CPL_TX_SEC_PDU_CIPHERSTART_V(cipher_start));
 	cpl->cipherstop_lo_authinsert = 0;
@@ -1416,11 +1370,10 @@  static int chcr_ktls_xmit_wr_short(struct sk_buff *skb,
 	tx_data = (void *)pos;
 	OPCODE_TID(tx_data) = htonl(MK_OPCODE_TID(CPL_TX_DATA, tx_info->tid));
 	tx_data->len = htonl(TX_DATA_MSS_V(mss) |
-			TX_LENGTH_V(skb->data_len + prior_data_len));
+			     TX_LENGTH_V(data_len + prior_data_len));
 	tx_data->rsvd = htonl(tcp_seq);
-	tx_data->flags = htonl(TX_BYPASS_F);
-	if (tcp_push)
-		tx_data->flags |= htonl(TX_PUSH_F | TX_SHOVE_F);
+	tx_data->flags = htonl(TX_BYPASS_F |
+			       (tcp_push ? TX_PUSH_F | TX_SHOVE_F : 0));
 
 	/* check left again, it might go beyond queue limit */
 	pos = tx_data + 1;
@@ -1449,8 +1402,8 @@  static int chcr_ktls_xmit_wr_short(struct sk_buff *skb,
 	if (prior_data_len)
 		pos = chcr_copy_to_txd(prior_data, &q->q, pos, 16);
 	/* send the complete packet except the header */
-	cxgb4_write_sgl(skb, &q->q, pos, end, skb->len - skb->data_len,
-			sgl_sdesc->addr);
+	cxgb4_write_partial_sgl(skb, &q->q, pos, end, sgl_sdesc->addr,
+				skb_offset, data_len);
 	sgl_sdesc->skb = skb;
 
 	chcr_txq_advance(&q->q, ndesc);
@@ -1477,24 +1430,22 @@  static int chcr_ktls_xmit_wr_short(struct sk_buff *skb,
 static int chcr_ktls_tx_plaintxt(struct chcr_ktls_info *tx_info,
 				 struct sk_buff *skb, u32 tcp_seq, u32 mss,
 				 bool tcp_push, struct sge_eth_txq *q,
-				 u32 port_id, u8 *prior_data,
-				 u32 prior_data_len)
+				 u32 data_len, u32 skb_offset)
 {
+	u32 flits = 0, ndesc, nfrags, wr_mid = 0;
 	int credits, left, len16, last_desc;
-	unsigned int flits = 0, ndesc;
 	struct tx_sw_desc *sgl_sdesc;
 	struct cpl_tx_data *tx_data;
 	struct ulptx_idata *idata;
 	struct ulp_txpkt *ulptx;
 	struct fw_ulptx_wr *wr;
-	u32 wr_mid = 0;
 	void *pos;
 	u64 *end;
 
 	flits = DIV_ROUND_UP(CHCR_PLAIN_TX_DATA_LEN, 8);
-	flits += chcr_sgl_len(skb_shinfo(skb)->nr_frags);
-	if (prior_data_len)
-		flits += 2;
+	nfrags = chcr_get_nfrags_to_send(skb, skb_offset, data_len);
+
+	flits += chcr_sgl_len(nfrags);
 	/* WR will need len16 */
 	len16 = DIV_ROUND_UP(flits, 2);
 	/* check how many descriptors needed */
@@ -1534,20 +1485,18 @@  static int chcr_ktls_tx_plaintxt(struct chcr_ktls_info *tx_info,
 	/* ULP_TXPKT */
 	ulptx = (struct ulp_txpkt *)(wr + 1);
 	ulptx->cmd_dest = htonl(ULPTX_CMD_V(ULP_TX_PKT) |
-			ULP_TXPKT_DATAMODIFY_V(0) |
 			ULP_TXPKT_CHANNELID_V(tx_info->port_id) |
-			ULP_TXPKT_DEST_V(0) |
-			ULP_TXPKT_FID_V(q->q.cntxt_id) | ULP_TXPKT_RO_V(1));
+			ULP_TXPKT_FID_V(q->q.cntxt_id) |
+			ULP_TXPKT_RO_F);
 	ulptx->len = htonl(len16 - 1);
 	/* ULPTX_IDATA sub-command */
 	idata = (struct ulptx_idata *)(ulptx + 1);
 	idata->cmd_more = htonl(ULPTX_CMD_V(ULP_TX_SC_IMM) | ULP_TX_SC_MORE_F);
-	idata->len = htonl(sizeof(*tx_data) + prior_data_len);
+	idata->len = htonl(sizeof(*tx_data));
 	/* CPL_TX_DATA */
 	tx_data = (struct cpl_tx_data *)(idata + 1);
 	OPCODE_TID(tx_data) = htonl(MK_OPCODE_TID(CPL_TX_DATA, tx_info->tid));
-	tx_data->len = htonl(TX_DATA_MSS_V(mss) |
-			TX_LENGTH_V(skb->data_len + prior_data_len));
+	tx_data->len = htonl(TX_DATA_MSS_V(mss) | TX_LENGTH_V(data_len));
 	/* set tcp seq number */
 	tx_data->rsvd = htonl(tcp_seq);
 	tx_data->flags = htonl(TX_BYPASS_F);
@@ -1555,12 +1504,6 @@  static int chcr_ktls_tx_plaintxt(struct chcr_ktls_info *tx_info,
 		tx_data->flags |= htonl(TX_PUSH_F | TX_SHOVE_F);
 
 	pos = tx_data + 1;
-	/* apart from prior_data_len, we should set remaining part of 16 bytes
-	 * to be zero.
-	 */
-	if (prior_data_len)
-		pos = chcr_copy_to_txd(prior_data, &q->q, pos, 16);
-
 	/* check left again, it might go beyond queue limit */
 	left = (void *)q->q.stat - pos;
 
@@ -1571,8 +1514,8 @@  static int chcr_ktls_tx_plaintxt(struct chcr_ktls_info *tx_info,
 		end = pos + left;
 	}
 	/* send the complete packet including the header */
-	cxgb4_write_sgl(skb, &q->q, pos, end, skb->len - skb->data_len,
-			sgl_sdesc->addr);
+	cxgb4_write_partial_sgl(skb, &q->q, pos, end, sgl_sdesc->addr,
+				skb_offset, data_len);
 	sgl_sdesc->skb = skb;
 
 	chcr_txq_advance(&q->q, ndesc);
@@ -1584,9 +1527,11 @@  static int chcr_ktls_tx_plaintxt(struct chcr_ktls_info *tx_info,
  * chcr_ktls_copy_record_in_skb
  * @nskb - new skb where the frags to be added.
  * @record - specific record which has complete 16k record in frags.
+ * @skb - old skb, to copy socket and destructor details.
  */
 static void chcr_ktls_copy_record_in_skb(struct sk_buff *nskb,
-					 struct tls_record_info *record)
+					 struct tls_record_info *record,
+					 struct sk_buff *skb)
 {
 	int i = 0;
 
@@ -1597,57 +1542,12 @@  static void chcr_ktls_copy_record_in_skb(struct sk_buff *nskb,
 	}
 
 	skb_shinfo(nskb)->nr_frags = record->num_frags;
-	nskb->data_len = record->len;
+	nskb->data_len += record->len;
 	nskb->len += record->len;
 	nskb->truesize += record->len;
-}
-
-/*
- * chcr_ktls_update_snd_una:  Reset the SEND_UNA. It will be done to avoid
- * sending the same segment again. It will discard the segment which is before
- * the current tx max.
- * @tx_info - driver specific tls info.
- * @q - TX queue.
- * return: NET_TX_OK/NET_XMIT_DROP.
- */
-static int chcr_ktls_update_snd_una(struct chcr_ktls_info *tx_info,
-				    struct sge_eth_txq *q)
-{
-	struct fw_ulptx_wr *wr;
-	unsigned int ndesc;
-	int credits;
-	void *pos;
-	u32 len;
-
-	len = sizeof(*wr) + roundup(CHCR_SET_TCB_FIELD_LEN, 16);
-	ndesc = DIV_ROUND_UP(len, 64);
-
-	credits = chcr_txq_avail(&q->q) - ndesc;
-	if (unlikely(credits < 0)) {
-		chcr_eth_txq_stop(q);
-		return NETDEV_TX_BUSY;
-	}
-
-	pos = &q->q.desc[q->q.pidx];
-
-	wr = pos;
-	/* ULPTX wr */
-	wr->op_to_compl = htonl(FW_WR_OP_V(FW_ULPTX_WR));
-	wr->cookie = 0;
-	/* fill len in wr field */
-	wr->flowid_len16 = htonl(FW_WR_LEN16_V(DIV_ROUND_UP(len, 16)));
-
-	pos += sizeof(*wr);
-
-	pos = chcr_write_cpl_set_tcb_ulp(tx_info, q, tx_info->tid, pos,
-					 TCB_SND_UNA_RAW_W,
-					 TCB_SND_UNA_RAW_V(TCB_SND_UNA_RAW_M),
-					 TCB_SND_UNA_RAW_V(0), 0);
-
-	chcr_txq_advance(&q->q, ndesc);
-	cxgb4_ring_tx_db(tx_info->adap, &q->q, ndesc);
-
-	return 0;
+	nskb->sk = skb->sk;
+	nskb->destructor = skb->destructor;
+	refcount_add(nskb->truesize, &nskb->sk->sk_wmem_alloc);
 }
 
 /*
@@ -1672,42 +1572,60 @@  static int chcr_end_part_handler(struct chcr_ktls_info *tx_info,
 				 struct tls_record_info *record,
 				 u32 tcp_seq, int mss, bool tcp_push_no_fin,
 				 struct sge_eth_txq *q,
-				 u32 tls_end_offset, bool last_wr)
+				 u32 tls_end_offset, u32 skb_offset)
 {
 	struct sk_buff *nskb = NULL;
+	bool is_last_wr = false;
+	int ret;
+
+	if (skb_offset + tls_end_offset == skb->len)
+		is_last_wr = true;
+
 	/* check if it is a complete record */
 	if (tls_end_offset == record->len) {
 		nskb = skb;
 		atomic64_inc(&tx_info->adap->ch_ktls_stats.ktls_tx_complete_pkts);
 	} else {
-		dev_kfree_skb_any(skb);
-
+		/* TAG needs to be calculated so, need to send complete record,
+		 * free the original skb and send a new one.
+		 */
 		nskb = alloc_skb(0, GFP_KERNEL);
-		if (!nskb)
+		if (!nskb) {
+			dev_kfree_skb_any(skb);
 			return NETDEV_TX_BUSY;
+		}
+
 		/* copy complete record in skb */
-		chcr_ktls_copy_record_in_skb(nskb, record);
+		chcr_ktls_copy_record_in_skb(nskb, record, skb);
 		/* packet is being sent from the beginning, update the tcp_seq
 		 * accordingly.
 		 */
 		tcp_seq = tls_record_start_seq(record);
-		/* reset snd una, so the middle record won't send the already
-		 * sent part.
-		 */
-		if (chcr_ktls_update_snd_una(tx_info, q))
-			goto out;
+		/* reset skb offset */
+		skb_offset = 0;
+
+		if (is_last_wr)
+			dev_kfree_skb_any(skb);
+
+		is_last_wr = true;
+
 		atomic64_inc(&tx_info->adap->ch_ktls_stats.ktls_tx_end_pkts);
 	}
 
-	if (chcr_ktls_xmit_wr_complete(nskb, tx_info, q, tcp_seq,
-				       (last_wr && tcp_push_no_fin),
-				       mss)) {
+	ret = chcr_ktls_xmit_wr_complete(nskb, is_last_wr,
+					 tx_info, q, tcp_seq,
+					 (is_last_wr && tcp_push_no_fin),
+					 mss, record->len, skb_offset,
+					 record->num_frags);
+	if (ret)
 		goto out;
-	}
+
+	tx_info->prev_seq = record->end_seq;
+
 	return 0;
 out:
 	dev_kfree_skb_any(nskb);
-	return NETDEV_TX_BUSY;
+	return ret;
 }
 
 /*
@@ -1735,41 +1653,46 @@  static int chcr_short_record_handler(struct chcr_ktls_info *tx_info,
 				     struct sk_buff *skb,
 				     struct tls_record_info *record,
 				     u32 tcp_seq, int mss, bool tcp_push_no_fin,
-				     struct sge_eth_txq *q, u32 tls_end_offset)
+				     struct sge_eth_txq *q, u32 tls_end_offset,
+				     u32 data_len, u32 skb_offset)
 {
-	u32 tls_rec_offset = tcp_seq - tls_record_start_seq(record);
+	u32 tls_rec_offset = record->len - tls_end_offset;
 	u8 prior_data[16] = {0};
 	u32 prior_data_len = 0;
-	u32 data_len;
 
 	/* check if the skb is ending in middle of tag/HASH, its a big
 	 * trouble, send the packet before the HASH.
 	 */
-	int remaining_record = tls_end_offset - skb->data_len;
+	int remaining_record = tls_end_offset - data_len;
 
 	if (remaining_record > 0 &&
 	    remaining_record < TLS_CIPHER_AES_GCM_128_TAG_SIZE) {
-		int trimmed_len = skb->data_len -
-			(TLS_CIPHER_AES_GCM_128_TAG_SIZE - remaining_record);
-		struct sk_buff *tmp_skb = NULL;
-		/* don't process the pkt if it is only a partial tag */
-		if (skb->data_len < TLS_CIPHER_AES_GCM_128_TAG_SIZE)
+		int trimmed_len = 0;
+
+		if (tls_end_offset > TLS_CIPHER_AES_GCM_128_TAG_SIZE)
+			trimmed_len = data_len -
+				      (TLS_CIPHER_AES_GCM_128_TAG_SIZE -
+				       remaining_record);
+		if (!trimmed_len)
 			goto out;
 
-		WARN_ON(trimmed_len > skb->data_len);
+		WARN_ON(trimmed_len > data_len);
+
+		data_len = trimmed_len;
+		atomic64_inc(&tx_info->adap->ch_ktls_stats.ktls_tx_trimmed_pkts);
+	}
 
-		/* shift to those many bytes */
-		tmp_skb = alloc_skb(0, GFP_KERNEL);
-		if (unlikely(!tmp_skb))
+	/* check if it is only the header part. */
+	if (tls_rec_offset + data_len <= (TLS_HEADER_SIZE + tx_info->iv_size)) {
+		if (chcr_ktls_tx_plaintxt(tx_info, skb, tcp_seq, mss,
+					  tcp_push_no_fin, q, data_len,
+					  skb_offset))
 			goto out;
 
-		chcr_ktls_skb_shift(tmp_skb, skb, trimmed_len);
-		/* free the last trimmed portion */
-		dev_kfree_skb_any(skb);
-		skb = tmp_skb;
-		atomic64_inc(&tx_info->adap->ch_ktls_stats.ktls_tx_trimmed_pkts);
+		tx_info->prev_seq = tcp_seq + data_len;
+		return 0;
+
 	}
-	data_len = skb->data_len;
 	/* check if the middle record's start point is 16 byte aligned. CTR
 	 * needs 16 byte aligned start point to start encryption.
 	 */
@@ -1830,39 +1753,19 @@  static int chcr_short_record_handler(struct chcr_ktls_info *tx_info,
 			}
 			/* reset tcp_seq as per the prior_data_required len */
 			tcp_seq -= prior_data_len;
-			/* include prio_data_len for  further calculation.
-			 */
-			data_len += prior_data_len;
 		}
-		/* reset snd una, so the middle record won't send the already
-		 * sent part.
-		 */
-		if (chcr_ktls_update_snd_una(tx_info, q))
-			goto out;
 		atomic64_inc(&tx_info->adap->ch_ktls_stats.ktls_tx_middle_pkts);
 	} else {
-		/* Else means, its a partial first part of the record. Check if
-		 * its only the header, don't need to send for encryption then.
-		 */
-		if (data_len <= TLS_HEADER_SIZE + tx_info->iv_size) {
-			if (chcr_ktls_tx_plaintxt(tx_info, skb, tcp_seq, mss,
-						  tcp_push_no_fin, q,
-						  tx_info->port_id,
-						  prior_data,
-						  prior_data_len)) {
-				goto out;
-			}
-			return 0;
-		}
 		atomic64_inc(&tx_info->adap->ch_ktls_stats.ktls_tx_start_pkts);
 	}
 
 	if (chcr_ktls_xmit_wr_short(skb, tx_info, q, tcp_seq, tcp_push_no_fin,
 				    mss, tls_rec_offset, prior_data,
-				    prior_data_len)) {
+				    prior_data_len, data_len, skb_offset)) {
 		goto out;
 	}
 
+	tx_info->prev_seq = tcp_seq + data_len + prior_data_len;
 	return 0;
 out:
 	dev_kfree_skb_any(skb);
@@ -1881,7 +1784,6 @@  static int chcr_ktls_xmit(struct sk_buff *skb, struct net_device *dev)
 	struct tls_record_info *record;
 	struct chcr_ktls_info *tx_info;
 	struct tls_context *tls_ctx;
-	struct sk_buff *local_skb;
 	struct sge_eth_txq *q;
 	struct adapter *adap;
 	unsigned long flags;
@@ -1903,51 +1805,23 @@  static int chcr_ktls_xmit(struct sk_buff *skb, struct net_device *dev)
 	if (unlikely(!tx_info))
 		goto out;
 
-	/* don't touch the original skb, make a new skb to extract each records
-	 * and send them separately.
-	 */
-	local_skb = alloc_skb(0, GFP_KERNEL);
-
-	if (unlikely(!local_skb))
-		return NETDEV_TX_BUSY;
-
 	adap = tx_info->adap;
 	stats = &adap->ch_ktls_stats;
 	port_stats = &stats->ktls_port[tx_info->port_id];
 
 	qidx = skb->queue_mapping;
 	q = &adap->sge.ethtxq[qidx + tx_info->first_qset];
-	cxgb4_reclaim_completed_tx(adap, &q->q, true);
-	/* if tcp options are set but finish is not send the options first */
-	if (!th->fin && chcr_ktls_check_tcp_options(th)) {
-		ret = chcr_ktls_write_tcp_options(tx_info, skb, q, false);
-		if (ret)
-			return NETDEV_TX_BUSY;
-	}
-	/* update tcb */
-	ret = chcr_ktls_xmit_tcb_cpls(tx_info, q, ntohl(th->seq),
-				      ntohl(th->ack_seq),
-				      ntohs(th->window));
-	if (ret) {
-		dev_kfree_skb_any(local_skb);
-		return NETDEV_TX_BUSY;
-	}
 
-	/* copy skb contents into local skb */
-	chcr_ktls_skb_copy(skb, local_skb);
+	cxgb4_reclaim_completed_tx(adap, &q->q, true);
+	skb_tx_timestamp(skb);
 
-	/* TCP segments can be in received either complete or partial.
-	 * chcr_end_part_handler will handle cases if complete record or end
-	 * part of the record is received. Incase of partial end part of record,
-	 * we will send the complete record again.
+	/* TCP segments can be received either complete or partial. Incase of
+	 * partial end part of record, we will send the complete record again.
 	 */
+	/* lock taken */
+	spin_lock_irqsave(&tx_ctx->base.lock, flags);
 
 	do {
-		int i;
-
-		cxgb4_reclaim_completed_tx(adap, &q->q, true);
-		/* lock taken */
-		spin_lock_irqsave(&tx_ctx->base.lock, flags);
 		/* fetch the tls record */
 		record = tls_get_record(&tx_ctx->base, tcp_seq,
 					&tx_info->record_no);
@@ -1966,82 +1840,97 @@  static int chcr_ktls_xmit(struct sk_buff *skb, struct net_device *dev)
 			goto out;
 		}
 
-		/* increase page reference count of the record, so that there
-		 * won't be any chance of page free in middle if in case stack
-		 * receives ACK and try to delete the record.
-		 */
-		for (i = 0; i < record->num_frags; i++)
-			__skb_frag_ref(&record->frags[i]);
-		/* lock cleared */
-		spin_unlock_irqrestore(&tx_ctx->base.lock, flags);
-
 		tls_end_offset = record->end_seq - tcp_seq;
 
-		pr_debug("seq 0x%x, end_seq 0x%x prev_seq 0x%x, datalen 0x%x\n",
-			 tcp_seq, record->end_seq, tx_info->prev_seq, data_len);
-		/* if a tls record is finishing in this SKB */
-		if (tls_end_offset <= data_len) {
-			struct sk_buff *nskb = NULL;
+		pr_debug("seq %#x, start %#x end %#x prev %#x, datalen %d offset %d\n",
+			 tcp_seq, tls_record_start_seq(record), record->end_seq,
+			 tx_info->prev_seq, data_len, tls_end_offset);
 
-			if (tls_end_offset < data_len) {
-				nskb = alloc_skb(0, GFP_KERNEL);
-				if (unlikely(!nskb)) {
-					ret = -ENOMEM;
-					goto clear_ref;
+		/* update tcb for the skb */
+		if (skb_data_len == data_len) {
+			u32 tx_max = tcp_seq;
+
+			if (!tls_record_is_start_marker(record) &&
+			    tls_end_offset < TLS_CIPHER_AES_GCM_128_TAG_SIZE)
+				tx_max = record->end_seq -
+					 TLS_CIPHER_AES_GCM_128_TAG_SIZE;
+			/* if tcp options are set but finish is not send the
+			 * options first
+			 */
+			if (!th->fin && chcr_ktls_check_tcp_options(th)) {
+				ret = chcr_ktls_write_tcp_options(tx_info, skb,
+								  q, false);
+				if (ret) {
+					spin_unlock_irqrestore(&tx_ctx->base.lock,
+							       flags);
+					goto out;
 				}
+			}
 
-				chcr_ktls_skb_shift(nskb, local_skb,
-						    tls_end_offset);
-			} else {
-				/* its the only record in this skb, directly
-				 * point it.
-				 */
-				nskb = local_skb;
+			ret = chcr_ktls_xmit_tcb_cpls(tx_info, q, tx_max,
+						      ntohl(th->ack_seq),
+						      ntohs(th->window));
+			if (ret) {
+				spin_unlock_irqrestore(&tx_ctx->base.lock,
+						       flags);
+				goto out;
 			}
-			ret = chcr_end_part_handler(tx_info, nskb, record,
+
+			if (th->fin)
+				skb_get(skb);
+		}
+
+		/* if a tls record is finishing in this SKB */
+		if (tls_end_offset <= data_len) {
+			ret = chcr_end_part_handler(tx_info, skb, record,
 						    tcp_seq, mss,
 						    (!th->fin && th->psh), q,
 						    tls_end_offset,
-						    (nskb == local_skb));
-
-			if (ret && nskb != local_skb)
-				dev_kfree_skb_any(local_skb);
+						    skb_offset);
 
 			data_len -= tls_end_offset;
 			/* tcp_seq increment is required to handle next record.
 			 */
-			tcp_seq += tls_end_offset;
+			tcp_seq = record->end_seq;
+			skb_offset += tls_end_offset;
 		} else {
-			ret = chcr_short_record_handler(tx_info, local_skb,
+			ret = chcr_short_record_handler(tx_info, skb,
 							record, tcp_seq, mss,
 							(!th->fin && th->psh),
-							q, tls_end_offset);
+							q, tls_end_offset,
+							data_len,
+							skb_offset);
 			data_len = 0;
 		}
-clear_ref:
-		/* clear the frag ref count which increased locally before */
-		for (i = 0; i < record->num_frags; i++) {
-			/* clear the frag ref count */
-			__skb_frag_unref(&record->frags[i]);
-		}
+
 		/* if any failure, come out from the loop. */
-		if (ret)
-			goto out;
+		if (ret) {
+			spin_unlock_irqrestore(&tx_ctx->base.lock, flags);
+			/* clear the extra ref count taken if fin is set. */
+			if (th->fin)
+				dev_kfree_skb_any(skb);
+
+			return NETDEV_TX_OK;
+		}
+
 		/* length should never be less than 0 */
 		WARN_ON(data_len < 0);
 
 	} while (data_len > 0);
 
-	tx_info->prev_seq = ntohl(th->seq) + skb_data_len;
+	spin_unlock_irqrestore(&tx_ctx->base.lock, flags);
 	atomic64_inc(&port_stats->ktls_tx_encrypted_packets);
 	atomic64_add(skb_data_len, &port_stats->ktls_tx_encrypted_bytes);
 
 	/* tcp finish is set, send a separate tcp msg including all the options
 	 * as well.
 	 */
-	if (th->fin)
+	if (th->fin) {
 		chcr_ktls_write_tcp_options(tx_info, skb, q, true);
+		dev_kfree_skb_any(skb);
+	}
 
+	return NETDEV_TX_OK;
 out:
 	dev_kfree_skb_any(skb);
 	return NETDEV_TX_OK;