PACKET_MMAP使用一例
时间：2010-04-04 来源：CUDev
根据上一篇文章PACKET_MMAP实现原理分析中PACKET_MMAP使用一节，写了一个简单的演示程序。
#include <stdio.h> #include <sys/types.h> #include <sys/socket.h> #include <sys/mman.h> #include <linux/if_packet.h> #include <poll.h> #include <net/ethernet.h> /* the L2 protocols */
void CallBackPacket(char *data) { printf("Recv A Packet.\n"); }
int main() { int fd = 0, ret = 0; char *buff = NULL;
//fd = socket(PF_PACKET, SOCK_RAW, htons(ETH_P_ALL)); //可以使用ARP进行一下测试 fd = socket(PF_PACKET, SOCK_DGRAM, htons (ETH_P_ARP)); if(fd<0) { perror("socket"); goto failed_2; }
//PACKET_VERSION和SO_BINDTODEVICE可以省略 #if 0 const int tpacket_version = TPACKET_V1; /* set tpacket hdr version. */ ret = setsockopt(fd, SOL_PACKET, PACKET_VERSION, &tpacket_version, sizeof (int)); if(ret<0) { perror("setsockopt"); goto failed_2; }
//#define NETDEV_NAME "wlan0" #define NETDEV_NAME "eth0" /* bind to device. */ ret = setsockopt(fd, SOL_SOCKET, SO_BINDTODEVICE, NETDEV_NAME, sizeof (NETDEV_NAME)); if(ret<0) { perror("setsockopt"); goto failed_2; } #endif
struct tpacket_req req; #define PER_PACKET_SIZE 2048 const int BUFFER_SIZE = 1024*1024*16; //16MB的缓冲区 req.tp_block_size = 4096; req.tp_block_nr = BUFFER_SIZE/req.tp_block_size; req.tp_frame_size = PER_PACKET_SIZE; req.tp_frame_nr = BUFFER_SIZE/req.tp_frame_size;
ret = setsockopt(fd, SOL_PACKET, PACKET_RX_RING, (void *)&req, sizeof(req)); if(ret<0) { perror("setsockopt"); goto failed_2; }
buff = (char *)mmap(0, BUFFER_SIZE, PROT_READ|PROT_WRITE, MAP_SHARED, fd, 0); if(buff == MAP_FAILED) { perror("mmap"); goto failed_2; }
int nIndex=0, i=0; while(1) { //这里在poll前先检查是否已经有报文被捕获了 struct tpacket_hdr* pHead = (struct tpacket_hdr*)(buff+ nIndex*PER_PACKET_SIZE); //如果frame的状态已经为TP_STATUS_USER了，说明已经在poll前已经有一个数据包被捕获了，如果poll后不再有数据包被捕获，那么这个报文不会被处理，这就是所谓的竞争情况。 if(pHead->tp_status == TP_STATUS_USER) goto process_packet;
//poll检测报文捕获 struct pollfd pfd; pfd.fd = fd; //pfd.events = POLLIN|POLLRDNORM|POLLERR; pfd.events = POLLIN; pfd.revents = 0; ret = poll(&pfd, 1, -1); if(ret<0) { perror("poll"); goto failed_1; }
process_packet: //尽力的去处理环形缓冲区中的数据frame，直到没有数据frame了 for(i=0; i<req.tp_frame_nr; i++) { struct tpacket_hdr* pHead = (struct tpacket_hdr*)(buff+ nIndex*PER_PACKET_SIZE);
//XXX: 由于frame都在一个环形缓冲区中，因此如果下一个frame中没有数据了，后面的frame也就没有frame了 if(pHead->tp_status == TP_STATUS_KERNEL) break;
//处理数据frame CallBackPacket((char*)pHead+pHead->tp_net);
//重新设置frame的状态为TP_STATUS_KERNEL pHead->tp_len = 0; pHead->tp_status = TP_STATUS_KERNEL;
//更新环形缓冲区的索引，指向下一个frame nIndex++; nIndex%=req.tp_frame_nr; }
}
success: close(fd); munmap(buff, BUFFER_SIZE); return 0;
failed_1: munmap(buff, BUFFER_SIZE); failed_2: close(fd); return -1; }
注:没有加filter，感觉直接使用bpf巨麻烦，实际上可以直接使用libpcap的关于bpf的api，就像参考1中实现的那样。
参考： http://hi.baidu.com/ah__fu/blog/item/8aadf895fad570007af48000.html http://linux.chinaunix.net/bbs/viewthread.php?tid=1134588
#include <stdio.h> #include <sys/types.h> #include <sys/socket.h> #include <sys/mman.h> #include <linux/if_packet.h> #include <poll.h> #include <net/ethernet.h> /* the L2 protocols */
void CallBackPacket(char *data) { printf("Recv A Packet.\n"); }
int main() { int fd = 0, ret = 0; char *buff = NULL;
//fd = socket(PF_PACKET, SOCK_RAW, htons(ETH_P_ALL)); //可以使用ARP进行一下测试 fd = socket(PF_PACKET, SOCK_DGRAM, htons (ETH_P_ARP)); if(fd<0) { perror("socket"); goto failed_2; }
//PACKET_VERSION和SO_BINDTODEVICE可以省略 #if 0 const int tpacket_version = TPACKET_V1; /* set tpacket hdr version. */ ret = setsockopt(fd, SOL_PACKET, PACKET_VERSION, &tpacket_version, sizeof (int)); if(ret<0) { perror("setsockopt"); goto failed_2; }
//#define NETDEV_NAME "wlan0" #define NETDEV_NAME "eth0" /* bind to device. */ ret = setsockopt(fd, SOL_SOCKET, SO_BINDTODEVICE, NETDEV_NAME, sizeof (NETDEV_NAME)); if(ret<0) { perror("setsockopt"); goto failed_2; } #endif
struct tpacket_req req; #define PER_PACKET_SIZE 2048 const int BUFFER_SIZE = 1024*1024*16; //16MB的缓冲区 req.tp_block_size = 4096; req.tp_block_nr = BUFFER_SIZE/req.tp_block_size; req.tp_frame_size = PER_PACKET_SIZE; req.tp_frame_nr = BUFFER_SIZE/req.tp_frame_size;
ret = setsockopt(fd, SOL_PACKET, PACKET_RX_RING, (void *)&req, sizeof(req)); if(ret<0) { perror("setsockopt"); goto failed_2; }
buff = (char *)mmap(0, BUFFER_SIZE, PROT_READ|PROT_WRITE, MAP_SHARED, fd, 0); if(buff == MAP_FAILED) { perror("mmap"); goto failed_2; }
int nIndex=0, i=0; while(1) { //这里在poll前先检查是否已经有报文被捕获了 struct tpacket_hdr* pHead = (struct tpacket_hdr*)(buff+ nIndex*PER_PACKET_SIZE); //如果frame的状态已经为TP_STATUS_USER了，说明已经在poll前已经有一个数据包被捕获了，如果poll后不再有数据包被捕获，那么这个报文不会被处理，这就是所谓的竞争情况。 if(pHead->tp_status == TP_STATUS_USER) goto process_packet;
//poll检测报文捕获 struct pollfd pfd; pfd.fd = fd; //pfd.events = POLLIN|POLLRDNORM|POLLERR; pfd.events = POLLIN; pfd.revents = 0; ret = poll(&pfd, 1, -1); if(ret<0) { perror("poll"); goto failed_1; }
process_packet: //尽力的去处理环形缓冲区中的数据frame，直到没有数据frame了 for(i=0; i<req.tp_frame_nr; i++) { struct tpacket_hdr* pHead = (struct tpacket_hdr*)(buff+ nIndex*PER_PACKET_SIZE);
//XXX: 由于frame都在一个环形缓冲区中，因此如果下一个frame中没有数据了，后面的frame也就没有frame了 if(pHead->tp_status == TP_STATUS_KERNEL) break;
//处理数据frame CallBackPacket((char*)pHead+pHead->tp_net);
//重新设置frame的状态为TP_STATUS_KERNEL pHead->tp_len = 0; pHead->tp_status = TP_STATUS_KERNEL;
//更新环形缓冲区的索引，指向下一个frame nIndex++; nIndex%=req.tp_frame_nr; }
}
success: close(fd); munmap(buff, BUFFER_SIZE); return 0;
failed_1: munmap(buff, BUFFER_SIZE); failed_2: close(fd); return -1; }
注:没有加filter，感觉直接使用bpf巨麻烦，实际上可以直接使用libpcap的关于bpf的api，就像参考1中实现的那样。
参考： http://hi.baidu.com/ah__fu/blog/item/8aadf895fad570007af48000.html http://linux.chinaunix.net/bbs/viewthread.php?tid=1134588
相关阅读 更多 +
