Method and apparatus for hardware packets reassembly in constrained networks

[0001] This application claims priority of U.S. Provisional Patent Application Ser. No. 60/622,796, filed Oct. 29, 2004, and U.S. Provisional Patent Application Ser. No. 60/668,967, filed Apr. 7, 2005. The subject matter of this earlier filed application is hereby incorporated by reference.

[0002] 1. Field of the Invention

[0003] The invention is related to a method and apparatus to enable hardware reassembly of fragmented IP data packets across traffic classes in a constrained network.

[0004] 2. Description of the Related Art

[0005] In conventional systems, large IP packets are normally fragmented using software in a transmitting host when conveyed in a constrained network to a receiving counterpart. The receiving software then reassembles a packet upon receipt of all of the relevant fragments. In a conventional system, switches or routers hardware treat fragmented packets as individual packets. Packets are normally fragmented and reassembled by software running on the network's processors on the end systems. A method and apparatus are needed to perform fragmented reassembly function directly in the switches or routers hardware to increase an overall network performance and efficiency.

[0010] In order to better describe the hardware packets reassembly apparatus and method, a switch or a router which will contain the invention has the following generic stages: ingress stage, en-queue stage, de-queue stage, and egress stage. The ingress stage is a stage where a packet is coming into a switch or a router. The ingress state contains packets parsing, look up, and forward decision logic. The en-queue stage is a stage containing logic which collects a packet's control information from the ingress stage and queues the packet into a packet buffer. The de-queue stage is a stage handling the de-queuing logic of the packet from the packet buffer. The de-queue stage also provides different Class of Service (COS) priority scheduling. The egress stage is a stage handling the packet de-queued from the packet buffer and transmitting packets out of a switch or a router. If necessary, a packet modification is made in the egress stage.

[0011] According to an embodiment of the present invention, the hardware packets reassembly apparatus and method support the following normal features in a switch system: assembly of various fragmented packet types including IP, wireless Lightweight Access Point Protocol (LWAPP) and other tunneled packets, a packet or packets fragments aging, unicast or multicast packets, different Class of Services (COS), and a Maximum Transmit Unit (MTU) check for packet reassembly decision.

[0012] The following constraints are assumed in a switching network: packets are fragmented in a finite small number, for instance, into two fragments, fragments of a packet are received in order, out of order fragments or interleaved fragments from a same source and same class will be dropped, and fragmented packets are supported by a finite number of source devices. It should be noted that although the reassembly is described hereon using two fragments, the apparatus and method, in accordance with an embodiment of the present invention, for reassembly of the two fragments can be extended to reassemble packets with more than two fragments where the fragment number can be determined from the packet header. The reassembly of the two fragments may be achieved by adding additional fields to a Packet Link RAM to store context for more than one fragment and linking the fragments.

[0013] FIG. 1 illustrates an example of a constrained network which contains both wired and wireless devices. The hardware packet reassembly apparatus resides in the WLAN switch 20. In this network, a wireless client, WC-00-0, may send packets to another wireless client, WC-50-0. The packets coming from the WC-00-0 client may be divided into a few fragments by an access point device, AP-0, after encapsulating some header information. All fragments of the packet will eventually route through the WLAN switch 20. When the WLAN switch 20 routes the packet received from WC-00-0, the packet reassembly apparatus 100 in the switch may collect all the fragments corresponding to the same packet, stripes off the header and reassemble the fragments into the packet. The entire packet is then sent through the network to the final destination of the WC-50-0 client in a full line rate speed.

[0014] According to an embodiment of the present invention, as illustrated in FIG. 2, the packet reassembly apparatus 100 is provided to ensure that all of the corresponding fragments associated with a data packet are received and processed before the data packet can be output by the WLAN switch 20 to other segments of the constrained network. To reassemble a packet from its fragments, the hardware packets reassembly apparatus of the present invention includes the following four major functionalities are provided. In the first functionality, at an ingress stage, the hardware packets reassembly apparatus parses received packets, recognizes fragments, and outputs fragments' control information. In the second functionality, at the en-queue stage, the hardware packets reassembly apparatus stores each fragment's information and data, links each related fragments and en-queue packets into a memory buffer when all fragments are available. In the third functionality, at the de-queue stage, the hardware packets reassembly apparatus de-queues packets from the packet descriptor, schedules the packets to be read out from the memory buffer according to the corresponding class of service (COS). In the fourth functionality, at the egress stage, the hardware packets reassembly apparatus assembles all fragments of a packet from the de-queue stage into a full packet and sends it out to the output port.

[0015] The following sections provide detail descriptions of each functionality of the packets reassembly apparatus in each of the stages.

[0016] When the packet reassembly apparatus 100 receives a packet in the ingress stage, it will firstly parse the packet in order to know the type of packet. A packet can be one of various Ethernet packet types, such as Ethernet II. Inside the Ethernet packet can be an IP packet or another (potentially fragmented) Layer 2 frame (Ethernet II, 802.3, 802.11 etc.). Based on each type of packet, the parsing logic inside the ingress stage can extract the control information from the header of the packet. For a fragmented packet, the header may contain the following fields for reassembly: source and destination address, fragment payload size, fragment identity (Fragment ID), fragment payload offset, source and destination ports, if applicable, Class of Services (COS) of the packets, if applicable, and/or tunnel keys, if applicable.

[0017] Based on the control information above, the hardware packets reassembly apparatus can identify if an incoming packet is a fragmented packet. If it is a fragment of a packet, the apparatus can further detect, in a hardware platform, the sequence number of a fragment corresponding to a particular data packet with respect to other fragments, that is, whether the fragment is the first fragment or the second fragment based on the fragment ID and offset. These fragment fields, once parsed from the header, includes source and destination ports/address, fragment ID, and fragment sequence number.

[0018] Further in the ingress stage, the forwarding decision for the packet is determined. All control information related to the forwarding decision of the packet is passed along with the payload to the en-queue stage logic, which includes unicast or multicast, class of services, etc. Along with this information, if the packet is a fragment, the additional fragment fields will be provided to the en-queue logic as well.

[0019] In the en-queue stage, the hardware packets reassembly apparatus includes an important piece of hardware for fragmented packets reassembly purpose in front of a normal en-queuing packet descriptor. In accordance with an aspect of the present invention, the packet reassembly apparatus 100 will not en-queue any fragment of a packet until all fragments that form the entire data packet have become available in a sequential order. Instead, each fragment's control fields from the ingress stage is maintained and stored in the fragment link buffer before en-queuing into the packet descriptor, while the payload is still kept in the memory buffer.

[0020] The fragment link buffer hardware in this packet reassembly apparatus invention stores and links all fragments of a packet to be reassembled. All of the fragment fields available from the ingress stage are kept in this link buffer before en-queuing. In additional, this link buffer performs important checking including ordering of the fragments received, missing fragments, and aging of the fragments resided in the link buffer. Packets will be dropped here due to out-of-order fragments received, missing fragments, or fragments aged out based on user programmed time. Dropped packets will be counted in a stat counter (not shown).

[0021] FIG. 3 shows the packet link buffer of hardware packets reassembly apparatus per each COS. If eight COS are supported, eight packet link buffers are provided. In accordance with an embodiment of the present invention, there are two fragments per each packet. The packet link buffer comprises two pieces of relevant memory hardware with the same numbers of entries, a content addressable memory (CAM) that contents the source address or port and fragment ID as the compare keys as well as RAMs (one per each class) that store the forwarding control information and a memory buffer pointer which contains the payload.

[0022] The first part of the relevant hardware is a CAM hardware which compares an incoming fragmented packet's field versus what have been stored in the link buffer. Every time when a fragmented packet arrives from the ingress logic, according to the COS, the source address, the destination address, the incoming port and the fragment ID are used to compare with the existing content inside the link buffer CAM. If the comparison is a miss and the fragment offset is zero, i.e., it is the first fragment, and the link buffer is not full, a new entry insertion of a new fragment will occur. An empty entry is allocated to store the source port/address and the fragment ID. The valid bit of this entry will be turned on. The aging counter of this entry will also activate. The corresponding entry number is used to index into the link buffer data RAM. In parallel with this new entry insertion, the payload of this fragmented packet is stored into the memory buffer. However, it is not yet en-queued into the packet descriptor. The forwarding information as well as the payload memory pointer are stored in the link buffer entry. If the link buffer is full, the fragmented packet will be dropped and a drop counter will keep the statistic.

[0023] If the comparison is a miss and the fragment offset is non-zero, the fragment will be dropped and the drop counter will increment. The payload will not be stored in the memory buffer and there is also no update to the link buffer RAM, which means all valid entries inside the link buffer CAM and RAM are always the first fragment of a packet. Also, fragments received out of order will be dropped. This includes packet dropped by the aging mechanism, which will be discussed shortly.

[0024] If the comparison is a hit (the entry valid is also set) and the fragment offset is zero, that means that there is an older first fragment residing in the link buffer. The older fragment, therefore, retired from the link buffer and the corresponding payload will be invalidated by returning the memory pointer back to the memory buffer. The new fragment is inserted and stored.

[0025] If the comparison is a hit and the fragment offset is non-zero, this is a normal case of fragments received in order where the hardware based packet reassembly occurs. Similarly, the payload of the second fragment will be stored into the memory buffer. The CAM hit generates an index to get the first fragment's control fields data from the RAM. At this point, the fragment offset of the second fragment checks against the first fragment offset for consistency. If the offset is not consistent, both fragments are dropped. The packet link buffer entry and together with the memory buffer entries containing the payloads will be invalidated and freed up. Otherwise, these control fields, together will the second fragment's memory pointer are sent to en-queue into the packet descriptor, as will be described shortly. The entry inside the CAM and RAM of the link buffer will be invalidated upon the read operation for en-queuing.

[0026] The hardware packets reassembly apparatus provides a user programmed aging threshold value. As described above, whenever a new fragment is inserted into the link buffer, an aging counter is activated. The aging counter will increment until it reaches the threshold where the link buffer entry will be aged and invalidated and the counter resets itself for a next insertion of that entry. Therefore, any first fragment residing in the link buffer too long is aged out, thereby clearing unwanted entries inside the link buffer due to fragments received too long between the first and second fragments as well as the first fragment received after the second fragment of the packet.

[0027] The second part of the relevant hardware structure in the en-queue stage is the packet descriptor. Each entry of the packet descriptor contains all forwarding control information, the header information, and payload pointer of the packet that is ready to be transmitted to a output port (if unicast) or multi output ports (if multicast).

[0028] As shown in FIG. 3, when both fragments of the packet are available and their offsets are consistent, the packet link buffer logic links the two fragments' information together before en-queuing into the packet descriptor. The first fragment's control information read out of a packet link buffer entry contains the destination port(s), the forwarding control, the header information, and the first fragment's payload pointer in the memory buffer. This information is combined with the second fragment's payload pointer to form a full entry and queued into the packet descriptor in one single entry. Note that an f_vld bit set denotes an entry in the packet descriptor including two fragments components. If f_vld is clear, it means that the packet is not a fragmented packet and no reassembly is required later in the egress stage.

[0029] Once the packet is en-queued into a packet descriptor entry, it is ready to be transmitted to output port(s). The transmission of the packet depends on the scheduling logic in the de-queue stage and the output port(s) availability. Upon satisfaction of the de-queuing requirement, a scheduler in the de-queue stage will read out each valid entry in the packet descriptor and transmit to the output port one at a time. When the scheduler reads out an entry with the f_vld set, it realizes that the entry is a fragment entry with two fragment payloads located in different memory buffer locations as indicated by the two different memory pointers. The scheduler has to first arrange the header and the first fragment payload to be sent to the egress port. Immediately, it should schedule the second fragment's payload to the same egress port (regardless of even other higher priority packet that needs to egress that port). The fragments' order has to be sequential and also both fragments have to be sent back to back to avoid under-run condition from happening in the egress out-going port. This also obviates the need to temporarily store the first fragment after dequeue. Essentially the scheduling logic treats the two fragments as one scheduling entity even though the buffer management logic treats each fragment as a separate packet.

[0030] The egress stage is the final stage of the hardware packets reassembly apparatus. Again, the egress logic must receive the fragments in the right sequential order and intact. In this stage, there is a packet modification hardware which physically handles the assembly of two fragments into a full packet before sending out to the output port. The modification hardware will encapsulate the proper header, align two fragmented payloads together and later regenerate cyclic redundancy check (CRC) to form a full packet. Eventually, the incoming fragmented packets are reassembled into a full packet using the hardware packets reassembly apparatus and method. The final full packet will then be transmitted out to the output port of a switch. It is possible that the reassembled packet exceeds the relevant Maximum Transmission Unit (MTU) of the outgoing logical interface and hence the reassembled packet will then be fragmented into a new set of fragments before being sent out the output port.

[0031] The principles of the present invention may similarly be applied to an 802.11 family of wireless standards and use terminology taken from these standards. Further, the principles of the present invention may similarly be applied in wireless LANs of other types, based on other standards and communication protocols. Similarly, although these embodiments are built around Ethernet LANs, the principles of the present invention may likewise be applied using LANs of other types for communication between the access points and manager node. As previously indicated, a fragmented packet reassembly apparatus and process are provided to reassemble fragmented data packets by preventing reassembling of the data packet in an improper fragment order and preventing intervening fragments not corresponding to the same data packet. In addition, in view of the foregoing, a person of ordinary skill in the art will appreciate that the packet reassembly apparatus may be applied to a wired or wireless switching network.

[0032] The many features and advantages of the invention are apparent from the detailed specification and, thus, it is intended by the appended claims to cover all such features and advantages of the invention which fall within the true spirit and scope of the invention. Further, since numerous modifications and changes will readily occur to those skilled in the art, it is not desired to limit the invention to the exact construction and operation illustrated and described, and accordingly all suitable modifications and equivalents may be resorted to, falling within the scope of the invention.