OFFLOADING iSCSI WITHOUT TOE

• US 20090183057A1
• Filed: 06/12/2008
• Published: 07/16/2009
• Est. Priority Date: 01/15/2008
• Status: Active Grant

• Current Assignee: Intel Corporation
• Original Assignee: Neterion, Inc. (Maxlinear Incorporated)
• Inventors: AIZMAN, Alexander
• Granted Patent: US 8,316,276 B2
• Time in Patent Office: Days
• US Class Current: 714/807
• CPC Class Codes: H04L 69/12 Protocol engines
• CPC Class Codes: H04L 69/16 Implementation or adaptation of Internet protocol [IP], of transmission control protocol [TCP] or of user datagram protocol [UDP]
• CPC Class Codes: H04L 69/161 Implementation details of TCP/IP or UDP/IP stack architecture; Specification of modified or new header fields
• CPC Class Codes: H04L 69/22 Parsing or analysis of headers
• CPC Class Codes: H04L 69/327 in the session layer [OSI layer 5]

Abstract

A ULP offload engine system, method and associated data structure are provided for performing protocol offloads without requiring a TCP offload engine (TOE). In an embodiment, the ULP offload engine provides iSCSI offload services.

Claims

1. A method for offloading CPU and memory intensive operations from host to an advanced iSCSI capabilities network adapter, the method comprising:

receiving a protocol data unit including a TCP protocol data unit (TCP segment);

validating the protocol data unit;

identifying a control structure (receive connection context) describing a combined TCP and iSCSI state of a receive side of an iSCSI connection; and

utilizing the control structure and TCP segment information to perform iSCSI CPU and memory intensive processing, wherein the TCP state of the corresponding TCP connection is not owned by the adapter.

2. The method of claim 1, wherein the iSCSI receive side protocol offloads are executed prior to state modifying TCP processing, while maintaining RFC-specified iSCSI semantics and in accordance with the iSCSI requirements.

3. The method of claim 1, wherein a TCP offload engine (TOE) is not utilized.

4. The method of claim 2, wherein the CPU intensive iSCSI CRC calculation is offloaded from the host and performed by the advanced iSCSI capabilities network adapter.

5. The method of claim 2, wherein advanced iSCSI capabilities network adapter performs direct data placement of iSCSI data into host memory.

6. The method of claim 2, wherein the IP and TCP receive protocol processing is done only and exclusively by the host networking stack, which does not require modification to support the advanced iSCSI capabilities network adapter.

7. The method of claim 2, wherein Layer 2, 3 and 4 networking protocols processing on the receive side (including ICMP, ARP, Neighbor Discovery (IPv6)) is done only and exclusively by the host networking stack, which does not require modification to utilize the advanced iSCSI capabilities network adapter.

8. The method of claim 2, wherein TCP connection identifier including source IP, source port, destination IP, destination port from an incoming Ethernet frame, is used to locate the receive connection context that corresponds to a given iSCSI connection.

9. The method of claim 2, wherein received iSCSI protocol data units (PDUs) are delineated in an incoming TCP byte stream without TCP Offload Engine (TOE) being present in the advanced iSCSI capabilities network adapter.

10. The method of claim 5, wherein iSCSI header of a received iSCSI PDU is used to identify pre-posted application buffers and place iSCSI payload directly into those buffers in the host memory, where the data is ultimately expected by the application that uses iSCSI transport.

11. The method of claim 2, wherein the advanced iSCSI capabilities network adapter modifies certain bits of the received iSCSI protocol data unit headers, to deliver control information to the host resident iSCSI software using host networking stack.

12. The method of claim 4, wherein host based iSCSI software can unambiguously identify, based on received iSCSI headers, whether header and data digests were calculated and verified (that is, offloaded) by the advanced iSCSI capabilities network adapter.

13. The method of claim 5, wherein host based iSCSI software can unambiguously identify, based on received iSCSI headers, whether iSCSI data was directly placed into the application buffers by the advanced iSCSI capabilities network adapter.

14. The method of claim 2, wherein the advanced iSCSI capabilities network adapter delivers all protocol control information in a form of protocol headers to the host networking stack, independently of whether iSCSI data was, or was not, directly placed.

15. The method of claim 2, wherein iSCSI receive processing is performed in presence of all possible TCP/IP and iSCSI level exceptions, including out of order TCP segment arrival, bad TCP checksum, bad IP checksum, fragmented IP datagrams, bad iSCSI header digest, bad iSCSI data digest, and others, while maintaining RFC-specified iSCSI semantics and in accordance with the iSCSI requirements.

16. The method of claim 2, wherein the receive connection context acts at any point during the connection lifetime in one of the following two states:

TCP stream and iSCSI.

17. The method of claim 4, wherein iSCSI header and data digests are offloaded by an advanced iSCSI capabilities network adapter on the receive side of ULP connection when the receive connection context is in iSCSI state.

18. The method of claim 5, wherein advanced iSCSI capabilities network adapter performs direct data placement of iSCSI data into host memory when the receive connection context is in iSCSI state.

19. The method of claim 2, wherein advanced iSCSI capabilities network adapter automatically transitions the receive side of iSCSI connection into TCP stream state in presence of TCP/IP level exceptions such as out of order TCP segmented arrival, and others.

20. The method of claim 2, wherein advanced capabilities network adapter employs a plurality of algorithms to make the transition back from TCP stream state to iSCSI state.

21. The method of claim 20, wherein the transition from TCP stream state to iSCSI state is performed autonomously by the advanced iSCSI capabilities network adapter.

22. The method of claim 20, wherein the transition from TCP stream to iSCSI state is performed semi-autonomously by the advanced iSCSI capabilities network adapter, with a single message exchange between advanced iSCSI capabilities network adapter and host based iSCSI software.

23. The method of claim 20, wherein iSCSI NOP-In/NOP-Out PDUs are employed to locate iSCSI headers in the incoming TCP byte stream.

24. The method of claim 20, wherein the transition from TCP stream to iSCSI state is done at wire speed, without incurring extra latency and without extra buffering.

25. The method of claim 2, wherein pre-posted application buffers are retired by the advanced iSCSI capabilities network adapter autonomously, without the need to communicate to host based iSCSI software.

26. The method of claim 2, wherein pre-posted application buffers are retired by the host based iSCSI software autonomously, without the need to communicate to the advanced iSCSI capabilities network adapter.

27. A method for offloading CPU intensive operations from host to an advanced iSCSI capabilities network adapter, the method comprising:

calculating and inserting iSCSI CRC in outgoing iSCSI protocol data units (PDUs); and

maintaining transmit history in terms of modifications performed on the outgoing iSCSI PDUs to facilitate TCP retransmit.

28. The method of claim 27, wherein the iSCSI transmit side protocol offloads are executed after the TCP transmit processing, while at the same time maintaining RFC-specified iSCSI semantics and in accordance with the iSCSI requirements.

29. The method of claim 27, wherein a TCP offload engine (TOE) is not utilized.

30. The method of claim 28, wherein the IP and TCP transmit protocol processing is done only and exclusively by the host networking stack, which does not require modification to utilize the advanced iSCSI capabilities network adapter.

31. The method of claim 28, wherein Layer 2, 3 and 4 networking protocols processing on the transmit side (including ICMP, ARP, Neighbor Discovery (IPv6)) is done only and exclusively by the host networking stack, which does not require modification to support the advanced iSCSI capabilities network adapter.

32. The method of claim 27, wherein advanced iSCSI capabilities network adapter finishes the unfinished iSCSI PDUs submitted by host based iSCSI software for transmission.

33. The method of claim 32, wherein the finishing of iSCSI PDUs includes:

identifying a control structure (transmit connection context) describing a combined TCP and iSCSI state of a transmit side of an iSCSI connection;

utilizing the control structure and TCP segment information to perform iSCSI PDU delineation on the transmit side;

calculating iSCSI header digest and inserting it into outgoing iSCSI PDUs in accordance with iSCSI RFC 3720;

calculating iSCSI data digest and inserting it into outgoing PDUs in accordance with iSCSI RFC.

34. An Upper Layer Protocol (ULP) offload engine, the ULP offload engine comprising:

means for performing receive and transmit side iSCSI offloads as described herein.

35. The method of claim 34, wherein the ULP offload engine can be implemented by the advanced capabilities iSCSI adapter.

36. The method of claim 34, wherein the host based iSCSI software uses ULP offload engine to offload CPU and memory intensive operations.

37. The method of claim 36, wherein the host based iSCSI software can enable receive and transmit side offloads separately and independently of each other, in any combination, including:

transmit CRC offload, receive CRC offload, transmit and receive CRC offload, transmit CRC offload and receive CRC offload and direct data placement on the receive side.