iSCSI: Add support for segmentation offload for hardware offloads.

Similar to TSO, iSCSI segmentation offload permits the upper layers to
submit a "large" virtual PDU which is split up into multiple segments
(PDUs) on the wire.  Similar to how the TCP/IP headers are used as
templates for TSO, the BHS at the start of a large PDU is used as a
template to construct the specific BHS at the start of each PDU.  In
particular, the DataSN is incremented for each subsequent PDU, and the
'F' flag is only set on the last PDU.

struct icl_conn has a new 'ic_hw_isomax' field which defaults to 0,
but can be set to the largest virtual PDU a backend supports.  If this
value is non-zero, the iSCSI target and initiator use this size
instead of 'ic_max_send_data_segment_length' to determine the maximum
size for SCSI Data-In and SCSI Data-Out PDUs.  Note that since PDUs
can be constructed from multiple buffers before being dispatched, the
target and initiator must wait for the PDU to be fully constructed
before determining the number of DataSN values were consumed (and thus
updating the per-transfer DataSN value used for the start of the next
PDU).

The target generates large PDUs for SCSI Data-In PDUs in
cfiscsi_datamove_in().  The initiator generates large PDUs for SCSI
Data-Out PDUs generated in response to an R2T.

Reviewed by:	mav
Sponsored by:	Chelsio Communications
Differential Revision:	https://reviews.freebsd.org/D31222

3 files changed, 29 insertions, 16 deletions

diff --git a/sys/cam/ctl/ctl_frontend_iscsi.c b/sys/cam/ctl/ctl_frontend_iscsi.c
index b3cd8ab79d76..b8fafcea69ed 100644
--- a/sys/cam/ctl/ctl_frontend_iscsi.c
+++ b/sys/cam/ctl/ctl_frontend_iscsi.c
@@ -2447,6 +2447,7 @@ cfiscsi_datamove_in(union ctl_io *io)
 	struct iscsi_bhs_data_in *bhsdi;
 	struct ctl_sg_entry ctl_sg_entry, *ctl_sglist;
 	size_t len, expected_len, sg_len, buffer_offset;
+	size_t max_send_data_segment_length;
 	const char *sg_addr;
 	icl_pdu_cb cb;
 	int ctl_sg_count, error, i;
@@ -2504,6 +2505,11 @@ cfiscsi_datamove_in(union ctl_io *io)
 	sg_len = 0;
 	response = NULL;
 	bhsdi = NULL;
+	if (cs->cs_conn->ic_hw_isomax != 0)
+		max_send_data_segment_length = cs->cs_conn->ic_hw_isomax;
+	else
+		max_send_data_segment_length =
+		    cs->cs_conn->ic_max_send_data_segment_length;
 	for (;;) {
 		if (response == NULL) {
 			response = cfiscsi_pdu_new_response(request, M_NOWAIT);
@@ -2520,7 +2526,7 @@ cfiscsi_datamove_in(union ctl_io *io)
 			bhsdi->bhsdi_initiator_task_tag =
 			    bhssc->bhssc_initiator_task_tag;
 			bhsdi->bhsdi_target_transfer_tag = 0xffffffff;
-			bhsdi->bhsdi_datasn = htonl(PRIV_EXPDATASN(io)++);
+			bhsdi->bhsdi_datasn = htonl(PRIV_EXPDATASN(io));
 			bhsdi->bhsdi_buffer_offset = htonl(buffer_offset);
 		}
 
@@ -2536,15 +2542,11 @@ cfiscsi_datamove_in(union ctl_io *io)
 		/*
 		 * Truncate to maximum data segment length.
 		 */
-		KASSERT(response->ip_data_len <
-		    cs->cs_conn->ic_max_send_data_segment_length,
-		    ("ip_data_len %zd >= max_send_data_segment_length %d",
-		    response->ip_data_len,
-		    cs->cs_conn->ic_max_send_data_segment_length));
-		if (response->ip_data_len + len >
-		    cs->cs_conn->ic_max_send_data_segment_length) {
-			len = cs->cs_conn->ic_max_send_data_segment_length -
-			    response->ip_data_len;
+		KASSERT(response->ip_data_len < max_send_data_segment_length,
+		    ("ip_data_len %zd >= max_send_data_segment_length %zd",
+		    response->ip_data_len, max_send_data_segment_length));
+		if (response->ip_data_len + len > max_send_data_segment_length) {
+			len = max_send_data_segment_length - response->ip_data_len;
 			KASSERT(len <= sg_len, ("len %zd > sg_len %zd",
 			    len, sg_len));
 		}
@@ -2603,8 +2605,7 @@ cfiscsi_datamove_in(union ctl_io *io)
 			i++;
 		}
 
-		if (response->ip_data_len ==
-		    cs->cs_conn->ic_max_send_data_segment_length) {
+		if (response->ip_data_len == max_send_data_segment_length) {
 			/*
 			 * Can't stuff more data into the current PDU;
 			 * queue it.  Note that's not enough to check
@@ -2619,6 +2620,8 @@ cfiscsi_datamove_in(union ctl_io *io)
 				buffer_offset -= response->ip_data_len;
 				break;
 			}
+			PRIV_EXPDATASN(io) += howmany(response->ip_data_len,
+			    cs->cs_conn->ic_max_send_data_segment_length);
 			if (cb != NULL) {
 				response->ip_prv0 = io->scsiio.kern_data_ref;
 				response->ip_prv1 = io->scsiio.kern_data_arg;
@@ -2654,6 +2657,8 @@ cfiscsi_datamove_in(union ctl_io *io)
 			}
 		}
 		KASSERT(response->ip_data_len > 0, ("sending empty Data-In"));
+		PRIV_EXPDATASN(io) += howmany(response->ip_data_len,
+		    cs->cs_conn->ic_max_send_data_segment_length);
 		if (cb != NULL) {
 			response->ip_prv0 = io->scsiio.kern_data_ref;
 			response->ip_prv1 = io->scsiio.kern_data_arg;
diff --git a/sys/dev/iscsi/icl.h b/sys/dev/iscsi/icl.h
index adcd580f299b..bd12569a8957 100644
--- a/sys/dev/iscsi/icl.h
+++ b/sys/dev/iscsi/icl.h
@@ -93,6 +93,7 @@ struct icl_conn {
 #endif
 	uint32_t		ic_max_recv_data_segment_length;
 	uint32_t		ic_max_send_data_segment_length;
+	size_t			ic_hw_isomax;
 	size_t			ic_maxtags;
 	bool			ic_header_crc32c;
 	bool			ic_data_crc32c;
diff --git a/sys/dev/iscsi/iscsi.c b/sys/dev/iscsi/iscsi.c
index 9a1b539e539b..60b3e3766da7 100644
--- a/sys/dev/iscsi/iscsi.c
+++ b/sys/dev/iscsi/iscsi.c
@@ -1154,7 +1154,7 @@ iscsi_pdu_handle_r2t(struct icl_pdu *response)
 	struct iscsi_bhs_data_out *bhsdo;
 	struct iscsi_outstanding *io;
 	struct ccb_scsiio *csio;
-	size_t off, len, total_len;
+	size_t off, len, max_send_data_segment_length, total_len;
 	int error;
 	uint32_t datasn = 0;
 
@@ -1203,11 +1203,16 @@ iscsi_pdu_handle_r2t(struct icl_pdu *response)
 
 	//ISCSI_SESSION_DEBUG(is, "r2t; off %zd, len %zd", off, total_len);
 
+	if (is->is_conn->ic_hw_isomax != 0)
+		max_send_data_segment_length = is->is_conn->ic_hw_isomax;
+	else
+		max_send_data_segment_length =
+		    is->is_conn->ic_max_send_data_segment_length;
 	for (;;) {
 		len = total_len;
 
-		if (len > is->is_conn->ic_max_send_data_segment_length)
-			len = is->is_conn->ic_max_send_data_segment_length;
+		if (len > max_send_data_segment_length)
+			len = max_send_data_segment_length;
 
 		if (off + len > csio->dxfer_len) {
 			ISCSI_SESSION_WARN(is, "target requested invalid "
@@ -1232,7 +1237,7 @@ iscsi_pdu_handle_r2t(struct icl_pdu *response)
 		    bhsr2t->bhsr2t_initiator_task_tag;
 		bhsdo->bhsdo_target_transfer_tag =
 		    bhsr2t->bhsr2t_target_transfer_tag;
-		bhsdo->bhsdo_datasn = htonl(datasn++);
+		bhsdo->bhsdo_datasn = htonl(datasn);
 		bhsdo->bhsdo_buffer_offset = htonl(off);
 		error = icl_pdu_append_data(request, csio->data_ptr + off, len,
 		    M_NOWAIT);
@@ -1245,6 +1250,8 @@ iscsi_pdu_handle_r2t(struct icl_pdu *response)
 			return;
 		}
 
+		datasn += howmany(len,
+		    is->is_conn->ic_max_send_data_segment_length);
 		off += len;
 		total_len -= len;