JP5039292B2 - Network adapter, communication system, and communication method - Google Patents

Description

  The present invention relates to a technique for transferring a packet received via a network to a host computer.

  As the network speeds up, how to reduce the load on the host computer for communication has become a problem in recent years. The combination of TCP (Transmission Control Protocol) / IP (Internet Protocol) and Socket API (Socket Application Interface), which are currently widely used, reduces the load on protocol processing and inter-memory copying for each packet in the host computer. Has become a particular challenge.

  In packet transmission / reception, processing such as header generation and analysis is performed for each packet transmitted / received. In order to increase the speed of communication, it is preferable to increase the size per packet to be transmitted as much as possible and to shorten the time required for these processes. However, in a network such as Ethernet (registered trademark), a maximum length (MTU, Maximum Transmit Unit) of one packet is defined. For example, in general Ethernet (registered trademark), 1500 bytes is generally used, and even if an extended function called a jumbo frame is used, it is generally about 9 kilobytes. In recent high-speed networks, protocol processing for each packet may become a bottleneck.

  As a technique for reducing the processing for each packet, for example, a technique for reducing the load on the CPU of the host computer by collecting packet information such as headers when receiving packets from a communication network is disclosed. (For example, Patent Document 1). As another example, a technique is disclosed in which a plurality of packets having a common route are extracted from the destination address of the packet, the plurality of packets are aggregated, and an aggregate packet including information on each packet is transmitted (for example, Patent Document 2). As yet another example, a technique for dividing a packet on the transmission side and a technique for assembling the reception side at the TCP level are disclosed (for example, Patent Document 3).

As another technique for speeding up communication, there is a technique for reducing the time required for copying between memories in a host computer, as in International Application No. PCTJP04 / 12006. According to this technique, on the packet transmission side, the packet main data part is directly DMA (Direct Memory Access) transferred from the user space of the host computer and transferred to the buffer (transmission buffer) of the network adapter. Data stored in the buffer of the network adapter is transmitted to the network using the generated header. In the network adapter on the packet receiving side, when a packet received from the network side is transferred to the host computer, the header and the main body data are separated, and the generated header is first transferred to the host computer. The main body data is transferred after the host computer software executes protocol processing to determine the user space of the data delivery destination.
JP 2003-87255 A (summary, FIG. 1 and FIG. 2, paragraph 0035) JP 2003-003673 A (summary, claim 1) Japanese Patent Laid-Open No. 2002-84289 (summary, FIG. 1, paragraph 0008)

  When the network adapter receives from the network a packet to be transferred to the host computer, it is difficult to receive each packet so that it is stored as one continuous packet in the memory of its own device. In other words, if a packet arrived from the network side is immediately analyzed, and it is determined which part of the entire packet the main data hits and an attempt is made to store it in an appropriate location, the memory-to-memory copy in the network adapter becomes overhead, It is difficult to realize. Therefore, non-consecutive packets are transferred to the memory of the network adapter so as to be a continuous area when transferred to the memory of the host computer. Since this technique is a general technique called Scatter Gather DMA, its description is omitted here.

  The host computer issues an instruction to transfer the main body data stored in the network adapter to the host computer. In order to realize this, it is necessary for the host computer to recognize where each main body data exists in the memory of the network adapter. In order for the host computer to realize transfer processing without recognizing the position of the main unit data, a handle is issued to the packet in which the main unit data is collected in the network adapter, and the correspondence relationship between this handle and each main unit data is determined in the network adapter. Must be managed. This requires management costs on the network adapter side.

  Alternatively, when transferring the main body data of the packets stored discontinuously in the memory of the network adapter to the host computer, the start address and data length information on the memory of each stored main body data is transferred from the network adapter to the host computer. Need to be notified. In general DMA transfer that has been used conventionally, these pieces of information are included in the DMA descriptor and notified to the host computer. However, since the DMA descriptor has a fixed length, the larger the data amount of the received packet is, the larger the number of pairs of the head address and the data length is. For example, if packets received from Ethernet (registered trademark) with an MTU of 1500 bytes are collected every 64 kilobytes, the number of pairs of the top address and the data length is 44. An area for storing such a large amount of information must be secured for one DMA descriptor.

  According to the present invention, when transferring a packet received from a network to a host computer, the cost of the network adapter is suppressed and the protocol processing of the host computer is kept simple, and the memory copy in the host computer is abolished. It aims at providing the technique which can improve the use efficiency of.

  A network adapter according to the present invention is for transferring a plurality of packets received via a network to a host computer, separating means for separating each received packet into a header part and a body data part, and the separation Storage means for storing each packet separated by the means; creation means for creating a new header based on each header portion of the plurality of packets; and a plurality of new headers and a plurality of storages stored in the storage means Notification means for notifying the host computer of a plurality of pieces of position specifying information indicating the respective storage positions of the main body data portion, and from the storage means based on a transfer instruction created using the position specifying information in the host computer. Transfer means for reading a plurality of main body data portions and transferring them to a host computer.

  In the network adapter, a plurality of received packets are logically combined into one packet composed of a new header and a plurality of body data parts. The new header and position specifying information are transferred to the host computer separately from the main body data portion. The main body data portion is transferred to a predetermined area of the memory of the host computer based on a transfer instruction created using the position specifying information in the host computer. The host computer performs protocol processing based on the new header, determines the transfer destination of the main body data portion, and creates a transfer instruction so that the plurality of main body data portions are continuous in the host computer memory of the transfer destination.

  In the network adapter, the notification unit transmits the new header and the position specifying information to a kernel space of a memory in the host computer, and the transfer unit transmits the plurality of main body data units in the host computer. You may make it transfer with respect to the user space of memory. In this configuration, it is not necessary to transfer the main body data part from the kernel space to the user space in the host computer.

  In addition to the new header and the plurality of position specifying information, the notification means may notify the host computer of the number of the plurality of position specifying information. In this configuration, the host computer can recognize the storage location of each location specification information or the storage location of a new header based on the number of location specification information.

  According to the present invention, a plurality of packets can be logically combined into one packet and transferred to the host computer without providing a complicated buffer management function in the network adapter. Further, the memory utilization efficiency is improved in the host computer.

DESCRIPTION OF EXEMPLARY EMBODIMENTS Hereinafter, preferred embodiments of the invention will be described in detail with reference to the drawings.
FIG. 1 is a diagram schematically showing the flow of packets transmitted and received in the communication system 10 according to the present invention. The packet transmission side host computer 2S and the reception side host computer 2R are connected to each other via a transmission side network adapter 1 (1S) and a reception side network adapter 1 (1R). The memory 3 of the host computer 2 includes a user space 4 and a kernel space 5. Here, transmission / reception of data (packets) in the user space 4, that is, data in the application buffer 6 stored in the user space 4 is stored. Consider a case where transmission / reception is performed by the COE (Copy Offload Engine) technique described in the above-mentioned International Application No. PCTJP04 / 12006.

  Data in the application buffer 6 in the transmission side host computer 2S is transferred to the transmission buffer 6 on the transmission side without passing through the kernel space 5 by the COE technique, and is divided for each MTU size by the transmission side network adapter 1 (1S). The packet (fragment) is transmitted to the network adapter 1 (1R). Thereafter, the data received by the network adapter 1 (1R) is transferred to the user space 4 of the receiving host computer 2R without passing through the kernel space 5. Transfer processing of the packet to the user space 4 on the reception side is performed by the host computer 2 on the data in the reception buffer in the network adapter 1. As described in the background art above, the protocol processing is executed by the software of the host computer 2.

  FIG. 2 is a diagram illustrating a DMA interface for executing a general DMA transfer. Before describing the operation of the network adapter 1 according to the present embodiment, first, the operation of the network adapter in general DMA transfer will be described with reference to FIG.

  Packets received via the network are stored discontinuously in the adapter memory (reception buffer) 101 in the order of arrival at the network adapter 100. Packets stored in the adapter memory 101 of the network adapter 100 are DMA-transferred to the memory (host memory) 201 of the host computer 200 using a DMA descriptor.

  The memory 201 of the host computer 200 that receives a packet from the network adapter 100 includes a descriptor area and a reception buffer. The descriptor area is an area for storing descriptors used for DMA transfer, and the reception buffer is an area for temporarily storing packets DMA-transferred from the network adapter 100.

  The descriptor arranged in the descriptor area is composed of an array or a list, and includes a physical address for storing information necessary for reading packet data, a command from the host computer 200 to the network adapter 100, a length of the received packet, and It holds control information such as status. Whether or not all of the plurality of packets transferred from the network side have been received is indicated by the status which is one of the control information. Note that control information such as the length and status of the received packet is updated by the network adapter 100 when reception of the packet is completed.

  FIG. 3 is a diagram for explaining an outline of an operation of transferring a received packet to the host computer 2 using the network adapter 1 according to the present embodiment. In the present embodiment, when packets received from the network side are stored discontinuously in the adapter memory 12 of the network adapter 1, the network adapter 1 side does not perform the process of reconstructing continuous packets, and the received multiple Based on the header of the packet, a packet that is logically combined is created. Then, a header of the logically created packet and a plurality of body data portions extracted from the plurality of received packets are transferred to the host computer 2, and the host computer 2 performs a protocol based on the header of the logically created packet. Execute the process. Hereinafter, a new header that is logically created from the received headers of a plurality of packets and is transferred to the host computer 2 is referred to as a reconfiguration header 30.

  The reconfiguration header 30 transferred from the network adapter 1 is stored in the kernel space 5 of the host memory 3 in the host computer 2. The reconfiguration header 30 is processed by software that executes the protocol processing of the host computer 2, and the transfer destination of the main data of the packet is determined. Here, the transfer destination of the packet body data is an area on the user space 4 of the host memory 3. In this embodiment, the host computer 2 does not perform inter-memory copying, and the adapter memory 12 of the network adapter 1. The main data of the upper packet is DMA transferred directly to the user space 4 of the host computer 2. The storage location of the main body data in the adapter memory 12 is transmitted from the network adapter 1 to the host computer 2 as position information (position specifying information) of the main body data prior to the transfer of the main body data.

  FIG. 4 is a block diagram of the communication system 10 configured using the network adapter 1 according to the present embodiment. In the present embodiment, as described above, the communication system 10 includes the network adapter 1 and the host computer 2. With reference to FIG. 4, functions in each device when the operation schematically illustrated in FIG. 3 is specifically executed will be described.

  The network adapter 1 includes a MAC (Media Access Control) 11, a reception buffer (adapter memory) 12, a DMA control unit 13, and a control unit 14. The MAC 11 is positioned below the data link layer, and defines whether data is transmitted / received by a predetermined method, such as a frame format and an error detection method, regarding transmission / reception of a frame which is a data transmission / reception unit. As described above, the reception buffer 12 stores packets received from the network side and packets transmitted from the host computer 2 toward the network. The DMA control unit 13 performs DMA transfer according to a command (DMA descriptor) created by the host computer 2. The control unit 14 controls packets transmitted and received.

  The control unit 14 of the network adapter 1 includes a checksum verification mechanism unit 41, a COE operation determination processing unit 42, a header processing unit 43, and a packet reconfiguration processing unit 44. The checksum verification mechanism unit 41 calculates a checksum used when checking whether a packet to be transmitted / received is broken, or performs an inspection using the checksum. The COE operation determination processing unit 42 determines whether or not to transfer the packet using the reconfiguration header 30 based on the size of the divided packet (fragment). The header processing unit 43 creates the reconfiguration header 30. The packet reconstruction processing unit 44 creates information necessary for arranging the main data of the fragments in the order before the division in the host computer 2.

  The host computer 2 includes a memory 3, a driver 22 and a control unit 23. As described above, the memory 3 is an area for storing the reconfiguration header 30 and the packet main data received from the network adapter 1, and includes the user space 4 and the kernel space 5.

  The driver 22 controls transmission / reception of packets to / from the network adapter 1. The driver 22 includes a normal packet transfer unit 221 and a COE data transfer unit 222. The normal packet transfer unit 221 performs normal packet transfer. Based on the reconfiguration header 30, the COE data transfer unit 222 transmits and receives the main data of the packet stored in the reception buffer 12 of the network adapter 1.

  The memory 3 includes the user space 4 and the kernel space 5 as described above. Since the user space 4 and the kernel space 5 are as described above, description thereof is omitted here.

  The control unit 23 of the host computer 2 performs control related to packet transmission / reception in the host computer 2 as an operation according to the present invention. The control unit 23 includes a protocol processing unit 231. The protocol processing unit 231 activates software for executing the protocol processing based on the reconfiguration header 30 received from the network adapter 1.

  FIG. 5 is a diagram for explaining that the position information 40 of the packet main data and the reconfiguration header 30 are stored in the reception buffer of the host computer 2 in the communication system 10 according to the present embodiment. In the kernel space 5 of the host computer 2 shown in FIG. 5, the upper part represents the descriptor area, and the lower part represents the reception buffer.

  In FIG. 5, the reconfiguration header 30 and the position information 40 sent from the network adapter 1 are stored in the reception buffer area. Here, the position information 40 includes, for example, the head address of each main body data and the data length of each main body data in the adapter memory 12 of the network adapter 1. The transfer of the reconfiguration header 30 and the position information 40 from the network adapter 1 to the host computer 2 is performed according to a header transfer DMA descriptor described later.

  A descriptor for data transfer is created by the host computer 2 using the position information 40. Here, as described above with reference to FIG. 2, the header transfer descriptor of FIG. 5 includes information such as the physical address of the reception buffer, the length of the reception packet, and the status.

  In FIG. 5, the status in the descriptor area is as described above with reference to FIG. The buffer address stores the start address of the reception buffer. The number of position information stores information on the number of position information 40 in the adapter memory 12, that is, the number of main body data accumulated in the adapter memory 12. Then, the host computer 2 creates a DMA descriptor for data transfer based on the position information 40, and the network adapter 1 reads the main body data of the packet from the adapter memory 12 according to the DMA descriptor and transfers it to the host computer 2. The transfer byte count stores the number of bytes of the reconstruction header 30 and the position information 40 transferred to the reception buffer.

  In the example shown in FIG. 5, the information stored in the reception buffer is set with the position information 40 from the beginning, and the reconstruction header 30 is set in the subsequent area, but is not limited thereto. For example, the host computer 2 may set the position information 40 and the start address of the reconfiguration header 30 and then notify the network adapter 1 of the set values. Alternatively, for example, a storage area for the position information 40 and the reconstruction header 30 is determined, the reconstruction header 30 is stored at the head of the area, and the position information 40 is stored at a predetermined position such as the end of the area. Also good.

  As described above, in the communication system according to the embodiment, the number of pieces of position information 40 corresponding to a plurality of received packets is created in the network adapter 1 and the position information 40 is notified from the network adapter 1 to the host computer 2. In addition, the number of the position information 40 is also notified to the host computer 2. Therefore, assuming that the data lengths of the position information 40 are the same as each other, the host computer 2 does not recognize the storage address of the position information 40 in the reception buffer, but stores the storage address of the head position information 40. If it is recognized, the storage address of other arbitrary position information 40 (and the storage address of the reconstruction header 30 in some cases) can be obtained by calculation.

  FIG. 6 is an example of the structure of the reconfiguration header 30. The reconfiguration header 30 created in this embodiment is based on the IP header of a divided packet (fragment) received from the network side. The network adapter 1 refers to the IP header of the received fragment, and creates a reconfiguration header 30 for collectively transferring the fragments having the same source, destination, identifier, etc. to the host computer 2 as one packet. Each body data of the fragment is protocol-processed as one packet according to the reconfiguration header 30 and then directly transferred to the user space 4. In the present embodiment, information such as a packet transmission source for distinguishing fragments to be collected from other fragments is used as identification information. In FIG. 6, “identification”, “protocol”, “source address”, and “destination address” correspond to this identification information.

  In the identifier, the same value is stored in all the divided packets. The protocol is a protocol of data stored in the IP packet. The transmission source address and destination address store the IP address of the transmission source and transmission destination of the packet, respectively.

  Each value of the identification information is set based on the header of the packet received from the network side. The header used is, for example, the IP header of the top fragment. Whether or not the received packet is the first fragment is determined by referring to the information of “Fragment Offset (position in original data)” of the header. If the received packet is the first fragment, “0” is stored as the value, so that it can be distinguished from fragments other than the first fragment.

  Among the information constituting the reconfiguration header 30, all fragments are stored in the adapter memory 12 for “Total Length”, “MF (More Fragment Flag)”, and “Header Checksum”. Updated when accumulated. The total length of the packet (Total Length) is calculated based on the header length (IHL) of the first fragment, the header length (IHL) of the last fragment, the total length of the packet (Total Length), and “Fragment Offset”. . The fragment offset stores an offset value from the beginning of the data before one packet is divided into a plurality of fragments.

  Regarding the MF, in this embodiment, “0” is stored as a value indicating that the divided packets are collected in the network adapter 1 and transferred to the host computer 2 as one packet. The header checksum is used to check whether the reassembly header 30 of the combined packet is not broken, and is updated and set based on a value changed by the reconfiguration such as the total length of the packet. The Or you may set what was computed by normal IP header checksum calculation after updating the full length etc. of a packet.

  Of the fields of the reconfiguration header 30 shown in FIG. 6, “Option” and “Padding” do not necessarily have to be set as fields constituting the header. The other fields shown in FIG. 6 are used when confirming the validity of the packet. For example, when the IP version is “IPv4”, “4” is stored as “Version”. In addition, “DF (Do n’t Fragment Flag)” stores “0” when the packet may be divided, and “1” when the packet should not be divided. Therefore, if the packet is fragmented but 1 is stored in the DF, it is determined that the packet is not valid. Similarly, when 0 is stored in “Time to Live”, it is determined that the time is not valid.

  The identification information of the reconstruction header 30 shown in FIG. 6 is created based on the identification information of the head fragment. Similarly, the identification information is stored in the header portion of each fragment for the fragments other than the head. In the present embodiment, a plurality of main body data portions to be combined into one packet from the network adapter 1 to the host computer 2 are determined based on identification information included in the fragment header. Is transmitted to the host computer 2. Information necessary for transferring body data, which is managed based on the fragment header, is used as management information. Hereinafter, a method of managing the fragment header information in the network adapter 1 will be described.

  FIG. 7 shows an example of the data structure of management information. By searching using fragment identification information as a key, information such as the storage destination address and data length of the main body data portion in each fragment in the adapter memory 12 is obtained.

  In FIG. 7, the data structure shown at the top is packet management information P for one packet to be collected. The data structure shown in the middle is the fragment management information F1 for the fragment first stored in the adapter memory 12, and the data structure shown in the lower is the fragment management information F2 for the second fragment, hereinafter the Nth fragment. Is fragment management information FN. In this embodiment, fragment management information for received fragments is managed by a list structure.

  Information about the fragments stored in the adapter memory 12 is sequentially added to the “received fragment list” of the packet management information P, and the “number of received fragments” is added. The “timer” of the packet management information P is initialized when the fragment management information F1 is created for the first accumulated fragment, and is stored in the adapter memory 12 for all fragments generated by dividing one packet. Measure the accumulation period.

  As an example, fragment management information F1 for fragments initially stored in the adapter memory 12 will be described. The “end fragment flag” is set so that the last fragment of the divided data can be distinguished from other fragments. In the “datagram offset”, an offset value from the head in the packet (datagram) before division, that is, the value of the “fragment offset” of the header is set. The “data portion size” is set to the data length of the main data portion excluding the header of the fragment. From the “offset in datagram” and “size of data portion”, the position occupied when the fragments are combined into one packet is calculated. In the present embodiment, it is determined whether or not all the fragments constituting the packet have been accumulated using these three data.

  In “fragment size”, the data length of the entire fragment is set. In the “accumulated buffer start address”, the address of the position accumulated in the adapter memory 12 is set. In the “data portion offset”, an offset value from the head, that is, the head address of the accumulated buffer in the adapter memory 12 is set. The “partial checksum of the data part” is set to a value for checking whether or not the packet (datagram) is broken, calculated based on the data of the fragment itself.

  The adapter memory 12 stores the header and main body data portion of each fragment. A value obtained by adding “data portion offset” of each fragment to “accumulated buffer start address” is the start address of the main data portion of the fragment to be transmitted to the host computer 2. The start address of the main body data portion and the “data portion size” which is the data length of the main body data portion are notified to the host computer 2 as the position information.

  When the period measured by the “timer” of the packet management information P before storing all the fragments in the adapter memory 12 exceeds a predetermined period, for example, “the head address of the accumulated buffer” and “the fragment size” Using the data, the packet is restored from the header and the main body data part stored in the adapter memory 12 at that time, and these packets are transferred to the host computer 2.

  The partial checksum for the data part of each fragment is stored for each fragment by reconfiguration and stored as fragment management information. When all fragments are accumulated, all the values are added together. The checksum can be obtained for the received packet. Based on this checksum, a TCP (Transmission Control Protocol) or UDP (User Datagram Protocol) checksum can be calculated.

  FIG. 8 is a diagram for explaining the outline of the operation for transferring the main body data portion of the fragment using the descriptor. As described with reference to FIG. 7, when the information necessary for the DMA transfer is notified from the management information created by the network adapter 1, the host computer 2 selects the descriptor used for the DMA transfer based on the notified information. Generate. When the DMA descriptor is created by the host computer 2, the DMA engine of the network adapter 1 executes DMA transfer processing of the main data of the packet according to the descriptor.

  The descriptor for transferring the main body data part of one fragment includes a status, a write destination host address of the main body data part (start address of the storage destination in the host computer 2), a transmission source storage address of the main body data part (of the network adapter 1). Storage address of the adapter memory 12) and the data length of the main body data portion. In the present embodiment, the host computer 2 sets the write destination host address, the transmission source address, and the data length based on the management information notified from the network adapter 1.

  The start address of the storage destination in the host computer 2 is the start address of the main body data in the user space 4 of the host computer 2. The transmission source address is the head address of the main body data in the adapter memory 12, and the data length is the size of the main body data. In the example of FIG. 8, descriptors 1 and 2 generated for fragments 1 and 2 are shown. The start address of the fragment 2 in the host computer 2 stores a value obtained by adding the data length of the fragment 1 to the start address of the fragment 1 (“Host Address” in FIG. 8). The main body data transferred to the user space 4 is sequentially stored in accordance with the packet configuration. The network adapter 1 sets the status value to a value indicating the “completed” status every time the DMA transfer process designated by each descriptor is completed.

  9 to 11 are flowcharts relating to packet transfer according to the present embodiment. Hereinafter, a processing method for realizing packet transfer according to the present embodiment will be described with reference to the drawings.

FIG. 9 is a flowchart for explaining the operation of the network adapter 1 that has received a packet.
First, in step S1, a packet is received. Next, in step S2, it is determined whether the IP header of the received packet is valid. Only those whose destination IP address is that of the host computer 2 are considered valid. If it is determined to be valid, identification information is extracted from the header portion of the received packet in step S3. Since the identification information has been described with reference to FIG. 6, it is omitted here.

  In step S4, it is determined whether the received packet is an IP fragment. If it is an IP fragment, it is determined in step S5 whether the fragment is a target for reconstruction. For example, in this embodiment, it is determined based on whether or not the protocol is a TCP protocol, and whether or not the data size to be transferred to the host computer is suitable for executing the transfer method of the present invention.

  If it is determined that the received packet is to be reconfigured, it is determined in step S6 whether or not the identification information matches that of a previously received packet. If they do not coincide with each other, in step S7, management information for the received packet is newly created, and further, a “timer” of the management information is initialized, and then the process proceeds to step S9. If the identification information matches the identification information of the packet received in the past in step S6, necessary information is extracted from the received packet to update the management information in step S8, and the process proceeds to step S9. When the management information is created or updated in step S7 or S8, position information (storage address and data length in the adapter memory 12) about the main body data portion of the received packet is created.

  In step S 9, the received packet is stored in the adapter memory 12. In step S10, it is determined whether or not it is the first fragment stored in the packet. Note that the IP header of the first fragment stores “0” as the fragment offset value, which is the same as the reconstruction header described above with reference to FIG. If it is determined that the received packet is the first fragment, the process proceeds to step S11, and the information stored in the header of the fragment is copied to the reconfiguration header 30, and the process proceeds to step S12. If it is not the first fragment, step S11 is skipped and the process proceeds to step S12.

  In step S12, it is determined whether or not all fragments have been accumulated in the adapter memory 12. If the accumulation of all fragments has not been completed, it is further determined in step S13 whether or not the timer has been completed. When the timer is not completed, the process is terminated without performing any particular processing. On the other hand, if the timer has been completed, in step S14, all fragments having the same identification information stored in the adapter memory 12 are transferred to the host computer 2 including the header, and the process is terminated.

  If it is determined in step S12 that all fragments have been accumulated, “MF” of the reconfiguration header 30 is set to 0 in step S15, and “full length of packet” and “header (IP) check are checked from the fragment management information. Calculate and set “thumb”. In step 16, the position information of each fragment and the reconstruction header 30 are transferred to the host computer 2. Note that the position information is created for each received packet in step S7 or S8 as described above. In step S17, the area holding the management information and the reconstruction header 30 is released, the timer is cleared, and the process is terminated.

  If the packet received in the previous determination in step S4 is not an IP fragment, and if it is determined in step S5 that the received packet is not a target for reconstruction, the process proceeds to step S18 to determine whether or not the identification information matches. judge. If they match, it is determined that the management information or the reconfiguration header 30 has been generated by mistake due to data corruption or the like for the packet for which the management information has been previously generated. In step S19, the management information and the reconfiguration header 30 are determined. And clear the timer. In step S20, the received packet is transferred to the host computer 2 as it is, and the process ends. Even when it is determined in step S2 that the IP header is not valid, the process proceeds to step S20, and packet transfer by the reconfiguration header 30 is not performed.

  In the present embodiment, a packet that is not considered valid in step S2 is transferred to the host computer 2 as it is in the process of step S20, and when the timer is completed in step S13. Is also transferred to the host computer 2 in step S14, but is not limited thereto. For example, for a packet in which the IP header is not valid in step S2 (except that the destination IP address is not considered valid because the destination IP address is not that of the host computer 2), or all fragments for which the timer has been completed in step S13, The network adapter 1 may be discarded.

  FIG. 10 is a flowchart of processing for notifying the reconfiguration header 30 and the position information 40 to the host computer 2. First, in step S31, the position information 40 is written in a predetermined buffer area in the network adapter 1. In step S32, it is determined whether or not padding is necessary when notifying the host computer 2. If padding is not required, the process proceeds to step S34 without performing any particular process. If padding is necessary, the process proceeds to step S34 after padding the buffer in step S33.

  In step S34, the reconfiguration header 30 is written in a predetermined buffer area in the network adapter 1. Here, the buffer in which the reconfiguration header 30 is written may be a buffer different from the buffer in which the position information 40 has been previously written. In step S35, the buffer contents, that is, the position information 40 and the reconstruction header 30 are transferred to the reception buffer in the kernel space of the host computer 2. The process of transferring the position information 40 and the reconfiguration header 30 to the host computer 2 follows, for example, a header transfer DMA descriptor created by the host computer 2.

  In step S 36, the number of bytes (header size) of the reconfiguration header 30 and the entire position information 40 and the number of position information 40 are set in the data transfer DMA descriptor created in the host computer 2. In step S37, a “complete” status is set as the descriptor status, and the process ends. Note that the processing from steps S31 to S37 is performed by the network adapter 1.

  FIG. 11 is a flowchart of a process in which the host computer 2 creates a data transfer DMA descriptor for DMA transfer of fragment main data. When the setting of the number of bytes of the transfer data of the descriptor, the position information 40 and the status is completed by executing the processing of FIG. 10, the host computer 2 further transfers the transfer source address, the transfer destination address, and the number of bytes of the main body data portion to be transferred. The command is set, and the DMA transfer of the fragment main data is executed.

  First, in step S41, a pointer for reading the position information 40 is initialized. The position information 40 has been transmitted from the network adapter 1 by the processing of the flowchart shown in FIG. At this time, the head address of the position information 40 included in the reception buffer in the kernel space 5 is set. Next, in step S42, the transfer destination pointer is initialized. At this time, the start address of the application buffer (user space 4) of the host computer 2 is set.

  In step S43, whether or not untransferred data exists in the adapter memory 12 is determined based on whether or not the pointer for reading the position information 40 has reached the end of the position information 40. If untransferred data exists, the process proceeds to step S44, where the descriptor transfer source (the transfer source address in FIG. 8 corresponds to this) is set to the position information 40 pointed to by the pointer for reading the position information 40. Set the head address of the main data. In step S45, the address indicated by the transfer destination pointer is set as the transfer destination of the descriptor (the host address in FIG. 8 corresponds to this).

  In step S46, the data length of the main body data set in the position information 40 pointed to by the pointer for reading the position information 40 is set to the descriptor transfer size (the data length in FIG. 8 corresponds to this). In step S47, a descriptor command is set. The network adapter 1 drives the DMA engine based on the received command. In step S48, the pointer is updated so that the pointer for reading the position information 40 points to the next position information 40. In step S49, the result of adding the transfer size set in step S46 to the transfer destination pointer is reset to the transfer destination pointer, and the process returns to step S43.

  Thereafter, the processing from step S43 to step S49 is repeated, and if the fragment body data accumulated in the adapter memory 12 no longer exists, the process proceeds to step S50. In step S50, assuming that descriptors have been set for all fragments to be transferred, the network adapter 1 is instructed to start DMA, and the process ends.

  As described above, according to the communication system including the network adapter 1 according to the present embodiment, the packet is reorganized by logically grouping the fragments in the network adapter 1, and the header (reconstruction header 30) is replaced with the position information 40 (each The information is sent to the host computer 2 together with information indicating the head address and data length of the adapter memory 12 of the fragment. In the host computer 2, DMA transfer is possible based on the information on the head address and data length of each fragment. At this time, the main data part of the received packet is directly transferred from the network adapter 1 to the user space 4 of the memory of the host computer 2, so that the host computer 2 does not need to copy data between the memories. Therefore, the memory utilization efficiency is improved. In the host computer 2, management of the reception buffer by protocol processing software is facilitated.

  Further, in the communication system 10 of the embodiment, the header and the main body data part of each received packet are separated in the network adapter 1, and the reconfiguration header 30 created based on these header parts and the position information of each main body data part Only 40 is transferred to the reception buffer area of the kernel space 5 of the host computer 2 separately from the main body data part. Here, the amount of data of the reconfiguration header 30 and the position information 40 of each main body data portion written in the reception buffer area is sufficiently small compared with a general MTU (for example, 1500 bytes). There is no need to prepare a special data area in order to realize the method.

  In the above embodiment, the case where the received packet is an IP fragment has been described, but the present invention is not limited to this. For example, a plurality of received TCP packets (segments) are logically grouped and reconfigured as a large TCP packet (segment), and the start address and data size on the adapter memory of each segment are sent to the host computer together with the reconfigured header. By providing the notifying means, it is possible to realize packet transfer as in the above embodiment.

(Appendix 1)
A network adapter for transferring a plurality of packets received via a network to a host computer,
Separating means for separating each received packet into a header part and a body data part,
Storage means for storing each packet separated by the separation means;
Creating means for creating a new header based on each header portion of the plurality of packets;
Notification means for notifying the host computer of a plurality of pieces of position specifying information indicating the storage positions of the new header and the plurality of main body data parts stored in the storage means;
A network adapter comprising: transfer means for reading the plurality of main body data parts from the storage means and transferring them to the host computer based on a transfer instruction created using the location specifying information in the host computer .
(Appendix 2)
The notifying means transmits the new header and the position specifying information to a kernel space of a memory in the host computer;
The network adapter according to appendix 1, wherein the transfer means transfers the plurality of main data portions to a user space of a memory in the host computer.
(Appendix 3)
The network according to claim 2, wherein the notifying unit transmits the new header and the position specifying information to a predetermined area of the kernel space according to the DMA descriptor written in the kernel space. adapter.
(Appendix 4)
The network adapter according to claim 1, wherein the notifying unit notifies the host computer of the number of the plurality of pieces of position specifying information in addition to the new header and the plurality of pieces of position specifying information.
(Appendix 5)
The notifying means transmits the new header to the head of a buffer area prepared in advance in the memory of the host computer, transmits the plurality of position specifying information to a predetermined position in the buffer area, and specifies the position. The network adapter according to appendix 4, wherein a destination address of information is notified to the host computer.
(Appendix 6)
The notifying means transmits the new header to the beginning of a buffer area prepared in advance in the memory of the host computer, and transmits the plurality of position specifying information to the end of the buffer area. The network adapter according to appendix 4.
(Appendix 7)
The notifying means writes the plurality of position control information at the head of a buffer area prepared in advance in the memory of the host computer, and writes the new header in an area following the position specifying information. The network adapter according to appendix 4.
(Appendix 8)
The network adapter according to appendix 1, wherein the plurality of packets are obtained by dividing one packet in a transmission device.
(Appendix 9)
The network adapter according to appendix 1, wherein the position specifying information includes information indicating a data length of a corresponding main body data part.
(Appendix 10)
The network adapter according to claim 1, wherein the storage unit discards all corresponding header units when all of the plurality of main body data units are transferred to the host computer by the transfer unit.
(Appendix 11)
The new header includes a flag indicating whether or not all of the plurality of main body data parts are accumulated in the accumulation unit,
When the notifying unit recognizes that all of the plurality of main data portions are stored in the storing unit based on the flag, starts the process of transferring the new header and the position specifying information to the host computer. The network adapter according to appendix 1, wherein:
(Appendix 12)
Timer means for measuring the time from when the first packet of the plurality of packets is received until the last packet is received;
The network adapter according to appendix 1, wherein when the time measured by the timer means exceeds a predetermined time, the transfer means transfers the received packets as they are to the host computer.
(Appendix 13)
A communication system including a host computer and a network adapter for transferring a plurality of packets received via the network to the host computer,
The network adapter is
Separating means for separating each received packet into a header part and a body data part,
Storage means for storing each packet separated by the separation means;
Creating means for creating a new header based on each header portion of the plurality of packets;
Notification means for notifying the host computer of a plurality of pieces of position specifying information indicating the storage positions of the new header and the plurality of main body data parts stored in the storage means;
Transfer means for reading the plurality of main body data sections from the storage means and transferring them to the host computer based on a transfer instruction created using the position specifying information in the host computer,
The host computer assembles the plurality of main body data sections based on the new header.
(Appendix 14)
14. The communication system according to appendix 13, wherein the host computer creates a DMA descriptor to be executed by the network adapter as the transfer instruction based on a plurality of pieces of position specifying information notified from the network adapter.
(Appendix 15)
A communication method for transferring a plurality of packets received via a network to a host computer,
Each received packet is separated into a header part and body data part,
Storing each separated packet in a storage means;
Create a new header based on each header portion of the plurality of packets,
Notifying the host computer of a plurality of position specifying information indicating the storage positions of the new header and the plurality of main body data parts stored in the storage means,
A communication method comprising: reading out the plurality of main body data parts from the storage means and transferring them to the host computer based on a transfer instruction created using the position specifying information in the host computer.
(Appendix 16)
A network transfer program for causing a network adapter to execute a transfer process of transferring a plurality of packets received via a network to a host computer,
In the network adapter,
Each received packet is separated into a header part and body data part,
Storing each separated packet in a storage means;
Create a new header based on each header portion of the plurality of packets,
Notifying the host computer of a plurality of position specifying information indicating the storage positions of the new header and the plurality of main body data parts stored in the storage means,
A network transfer program for executing a process of reading the plurality of main body data parts from the storage means and transferring them to the host computer based on a transfer instruction created using the position specifying information in the host computer . (7)

It is the figure which showed roughly the flow of the packet transmitted / received in the communication system which concerns on this invention. It is a figure explaining the DMA interface for performing general DMA transfer. It is a figure explaining the outline of the operation | movement which transfers a received packet to a host computer using the network adapter which concerns on this embodiment. It is a block diagram of the communication system comprised using the network adapter which concerns on this embodiment. It is a figure explaining storing the positional information on the main body data of a packet, and the reconstruction header in the reception buffer of the host computer in the communication system which concerns on this embodiment. It is an example of the structure of a reconstruction header. It is an example of the data structure of management information. It is a figure explaining the outline of the operation | movement which transfers the main-body data part of a fragment using a descriptor. It is a flowchart explaining operation | movement of the network adapter which received the packet. It is a flowchart about the process which notifies a reconfiguration header and position information to a host computer. It is a flowchart about the process which produces the descriptor for data transfer.

Explanation of symbols

1 Network adapter 2 Host computer 3 Host memory 4 User space 5 Kernel space 12 Adapter memory (receive buffer)
13 DMA control unit 14 Control unit 41 Checksum verification mechanism unit 42 COE operation determination processing unit 43 Header processing unit 44 Packet reconfiguration processing unit

Claims (10)

Received via the network, a plurality of packets obtained by dividing one data, a Rene Tsu network adapter forwards to the host computer,
Separating means for separating each received packet into a header part and a body data part,
And accumulation means you accumulate each packet separated by the separating means,
Creating means for creating a new header common to the plurality of packets, having an identifier generated based on the header portion of the first packet of the plurality of packets based on each header portion of the plurality of packets ;
Notification means for notifying the host computer of a plurality of pieces of position specifying information indicating the storage positions of the new header and the plurality of main body data parts stored in the storage means;
And characterized in that it comprises transfer means said the host computer based on the transfer instruction generated using a plurality of position specifying information, and transfers from said storage means reads said plurality of main data portion to the host computer Network adapter to use. The notification means transmits the new header and the plurality of position specifying information to a kernel space of a memory in the host computer,
The network adapter according to claim 1, wherein the transfer unit transfers the plurality of main body data units to a user space of a memory in the host computer. The notification unit transmits the new header and the plurality of pieces of position specifying information to a predetermined area of the kernel space according to the DMA descriptor written in the kernel space. The described network adapter. The network adapter according to claim 1, wherein the notifying unit notifies the host computer of the number of the plurality of pieces of position specifying information in addition to the new header and the plurality of pieces of position specifying information. The notifying means transmits the new header to the head of a buffer area prepared in advance in the memory of the host computer, transmits the plurality of position specifying information to a predetermined position in the buffer area, and specifies the position. The network adapter according to claim 4, wherein the host computer is notified of an information transmission destination address. The notifying means transmits the new header to the beginning of a buffer area prepared in advance in the memory of the host computer, and transmits the plurality of position specifying information to the end of the buffer area. The network adapter according to claim 4. A network transfer program for causing a network adapter to execute a transfer process of transferring a plurality of packets obtained by dividing one data received via a network to a host computer,
In the network adapter,
Each received packet is separated into a header part and body data part,
Storing each separated packet in a storage means;
Based on each header portion of the plurality of packets, the identifier generated based on the header portion of the first packet of the plurality of packets, creating a new header common to the plurality of packets,
Notifying the host computer of a plurality of position specifying information indicating the storage positions of the new header and the plurality of main body data parts stored in the storage means,
Wherein based on the plurality of position specifying information transfer instruction that has been generated using the host computer, characterized in that to execute the process of transmitting from said storage means reads said plurality of main data portion to the host computer Network transfer program. Received via the host computer, and a network, comprising a plurality of packets obtained by dividing one data communications system including a Rene Tsu network adapter forwards to the host computer,
The network adapter is
Separating means for separating each received packet into a header part and a body data part,
And accumulation means you accumulate each packet separated by the separating means,
Creating means for creating a new header common to the plurality of packets, having an identifier generated based on the header portion of the first packet of the plurality of packets based on each header portion of the plurality of packets ;
Notification means for notifying the host computer of a plurality of pieces of position specifying information indicating the storage positions of the new header and the plurality of main body data parts stored in the storage means;
On the basis of the transfer instruction created using the plurality of position specifying information in the host computer, and a transfer means for transferring to the host computer reads out the plurality of main data portion from the storage means,
The host computer assembles the plurality of main body data sections based on the new header. 9. The communication system according to claim 8, wherein the host computer creates a DMA descriptor to be executed by the network adapter as the transfer instruction based on a plurality of pieces of position specifying information notified from the network adapter. . A network adapter executes a process of transferring a plurality of packets obtained by dividing one data received via a network to a host computer,
The network adapter is
Each received packet is separated into a header part and body data part,
Storing each separated packet in a storage means;
Based on each header portion of the plurality of packets, the identifier generated based on the header portion of the first packet of the plurality of packets, creating a new header common to the plurality of packets,
Notifying the host computer of a plurality of position specifying information indicating the storage positions of the new header and the plurality of main body data parts stored in the storage means,
Communication on the basis of the transfer instruction created using the plurality of position specifying information in the host computer, characterized in that the said storage means comprises a process of transferring to the host computer reads out the plurality of main data portion Method.