X-Git-Url: https://git.tcpdump.org/libpcap/blobdiff_plain/c821ea01280546b208744939fa6919fb1727326c..refs/heads/master:/pcap-dpdk.c diff --git a/pcap-dpdk.c b/pcap-dpdk.c index 0d4456db..76de9b76 100644 --- a/pcap-dpdk.c +++ b/pcap-dpdk.c @@ -34,7 +34,7 @@ Description: Limitations: 1. DPDK support will be on if DPDK is available. Please set DIR for --with-dpdk[=DIR] with ./configure or -DDPDK_DIR[=DIR] with cmake if DPDK is installed manually. -2. Only support link libdpdk.so dynamicly, because the libdpdk.a will not work correctly. +2. Only support link libdpdk.so dynamically, because the libdpdk.a will not work correctly. 3. Only support read operation, and packet injection has not been supported yet. Usage: @@ -50,12 +50,12 @@ sed -i 's/CONFIG_RTE_BUILD_SHARED_LIB=n/CONFIG_RTE_BUILD_SHARED_LIB=y/' $RTE_SDK You shall learn how to bind nic with DPDK-compatible driver by $RTE_SDK/usertools/dpdk-devbind.py, such as igb_uio. And enable hugepages by dpdk-setup.sh -Then launch the l2fwd with dynamic dirver support. For example: +Then launch the l2fwd with dynamic driver support. For example: $RTE_SDK/examples/l2fwd/$RTE_TARGET/l2fwd -dlibrte_pmd_e1000.so -dlibrte_pmd_ixgbe.so -dlibrte_mempool_ring.so -- -p 0x1 3. Compile libpcap with dpdk options. -If DPDK has not been found automatically, you shall export DPDK envionment variable which are used for compiling DPDK. And then pass $RTE_SDK/$RTE_TARGET to --with-dpdk or -DDPDK_DIR +If DPDK has not been found automatically, you shall export DPDK environment variable which are used for compiling DPDK. And then pass $RTE_SDK/$RTE_TARGET to --with-dpdk or -DDPDK_DIR export RTE_SDK={your DPDK base directory} export RTE_TARGET={your target name} @@ -70,22 +70,20 @@ mkdir -p build && cd build && cmake -DDPDK_DIR=$RTE_SDK/$RTE_TARGET ../ && make 4. Link your own program with libpcap, and use DPDK with the device name as dpdk:{portid}, such as dpdk:0. And you shall set DPDK configure options by environment variable DPDK_CFG -For example, the testprogs/capturetest could be lanched by: +For example, the testprogs/capturetest could be launched by: env DPDK_CFG="--log-level=debug -l0 -dlibrte_pmd_e1000.so -dlibrte_pmd_ixgbe.so -dlibrte_mempool_ring.so" ./capturetest -i dpdk:0 */ -#ifdef HAVE_CONFIG_H #include -#endif -#include #include #include #include #include #include #include +#include /* for INT_MAX */ #include #include @@ -116,6 +114,16 @@ env DPDK_CFG="--log-level=debug -l0 -dlibrte_pmd_e1000.so -dlibrte_pmd_ixgbe.so #include "pcap-int.h" #include "pcap-dpdk.h" +/* + * Deal with API changes that break source compatibility. + */ + +#ifdef HAVE_STRUCT_RTE_ETHER_ADDR +#define ETHER_ADDR_TYPE struct rte_ether_addr +#else +#define ETHER_ADDR_TYPE struct ether_addr +#endif + #define DPDK_DEF_LOG_LEV RTE_LOG_ERR // // This is set to 0 if we haven't initialized DPDK yet, 1 if we've @@ -153,7 +161,11 @@ static char dpdk_cfg_buf[DPDK_CFG_MAX_LEN]; static uint16_t nb_rxd = RTE_TEST_RX_DESC_DEFAULT; static uint16_t nb_txd = RTE_TEST_TX_DESC_DEFAULT; +#ifdef RTE_ETHER_MAX_JUMBO_FRAME_LEN +#define RTE_ETH_PCAP_SNAPLEN RTE_ETHER_MAX_JUMBO_FRAME_LEN +#else #define RTE_ETH_PCAP_SNAPLEN ETHER_MAX_JUMBO_FRAME_LEN +#endif static struct rte_eth_dev_tx_buffer *tx_buffer; @@ -177,7 +189,7 @@ struct pcap_dpdk{ uint64_t bps; struct rte_mempool * pktmbuf_pool; struct dpdk_ts_helper ts_helper; - struct ether_addr eth_addr; + ETHER_ADDR_TYPE eth_addr; char mac_addr[DPDK_MAC_ADDR_SIZE]; char pci_addr[DPDK_PCI_ADDR_SIZE]; unsigned char pcap_tmp_buf[RTE_ETH_PCAP_SNAPLEN]; @@ -208,7 +220,7 @@ static void dpdk_fmt_errmsg_for_rte_errno(char *errbuf, size_t errbuflen, size_t errbuflen_remaining; va_start(ap, fmt); - pcap_vsnprintf(errbuf, errbuflen, fmt, ap); + vsnprintf(errbuf, errbuflen, fmt, ap); va_end(ap); msglen = strlen(errbuf); @@ -236,7 +248,7 @@ static void dpdk_fmt_errmsg_for_rte_errno(char *errbuf, size_t errbuflen, * buffer (based on the "PER_LCORE" in "RTE_DEFINE_PER_LCORE" used * to declare the buffers statically) for DPDK errors. */ - pcap_snprintf(p, errbuflen_remaining, "%s", rte_strerror(errnum)); + snprintf(p, errbuflen_remaining, "%s", rte_strerror(errnum)); } static int dpdk_init_timer(struct pcap_dpdk *pd){ @@ -259,7 +271,7 @@ static inline void calculate_timestamp(struct dpdk_ts_helper *helper,struct time timeradd(&(helper->start_time), &cur_time, ts); } -static uint32_t dpdk_gather_data(unsigned char *data, int len, struct rte_mbuf *mbuf) +static uint32_t dpdk_gather_data(unsigned char *data, uint32_t len, struct rte_mbuf *mbuf) { uint32_t total_len = 0; while (mbuf && (total_len+mbuf->data_len) < len ){ @@ -271,7 +283,7 @@ static uint32_t dpdk_gather_data(unsigned char *data, int len, struct rte_mbuf * } -static int dpdk_read_with_timeout(pcap_t *p, uint16_t portid, struct rte_mbuf **pkts_burst, const uint16_t burst_cnt){ +static int dpdk_read_with_timeout(pcap_t *p, struct rte_mbuf **pkts_burst, const uint16_t burst_cnt){ struct pcap_dpdk *pd = (struct pcap_dpdk*)(p->priv); int nb_rx = 0; int timeout_ms = p->opt.timeout; @@ -280,7 +292,7 @@ static int dpdk_read_with_timeout(pcap_t *p, uint16_t portid, struct rte_mbuf ** // In non-blocking mode, just read once, no matter how many packets are captured. nb_rx = (int)rte_eth_rx_burst(pd->portid, 0, pkts_burst, burst_cnt); }else{ - // In blocking mode, read many times until packets are captured or timeout or break_loop is setted. + // In blocking mode, read many times until packets are captured or timeout or break_loop is set. // if timeout_ms == 0, it may be blocked forever. while (timeout_ms == 0 || sleep_ms < timeout_ms){ nb_rx = (int)rte_eth_rx_burst(pd->portid, 0, pkts_burst, burst_cnt); @@ -308,7 +320,6 @@ static int pcap_dpdk_dispatch(pcap_t *p, int max_cnt, pcap_handler cb, u_char *c struct rte_mbuf *pkts_burst[MAX_PKT_BURST]; struct rte_mbuf *m; struct pcap_pkthdr pcap_header; - uint16_t portid = pd->portid; // In DPDK, pkt_len is sum of lengths for all segments. And data_len is for one segment uint32_t pkt_len = 0; uint32_t caplen = 0; @@ -319,26 +330,41 @@ static int pcap_dpdk_dispatch(pcap_t *p, int max_cnt, pcap_handler cb, u_char *c u_char *large_buffer=NULL; int timeout_ms = p->opt.timeout; - if ( !PACKET_COUNT_IS_UNLIMITED(max_cnt) && max_cnt < MAX_PKT_BURST){ + /* + * This can conceivably process more than INT_MAX packets, + * which would overflow the packet count, causing it either + * to look like a negative number, and thus cause us to + * return a value that looks like an error, or overflow + * back into positive territory, and thus cause us to + * return a too-low count. + * + * Therefore, if the packet count is unlimited, we clip + * it at INT_MAX; this routine is not expected to + * process packets indefinitely, so that's not an issue. + */ + if (PACKET_COUNT_IS_UNLIMITED(max_cnt)) + max_cnt = INT_MAX; + + if (max_cnt < MAX_PKT_BURST){ burst_cnt = max_cnt; }else{ burst_cnt = MAX_PKT_BURST; } - while( PACKET_COUNT_IS_UNLIMITED(max_cnt) || pkt_cnt < max_cnt){ + while( pkt_cnt < max_cnt){ if (p->break_loop){ p->break_loop = 0; return PCAP_ERROR_BREAK; } // read once in non-blocking mode, or try many times waiting for timeout_ms. - // if timeout_ms == 0, it will be blocked until one packet arrives or break_loop is setted. - nb_rx = dpdk_read_with_timeout(p, portid, pkts_burst, burst_cnt); + // if timeout_ms == 0, it will be blocked until one packet arrives or break_loop is set. + nb_rx = dpdk_read_with_timeout(p, pkts_burst, burst_cnt); if (nb_rx == 0){ if (pd->nonblock){ RTE_LOG(DEBUG, USER1, "dpdk: no packets available in non-blocking mode.\n"); }else{ if (p->break_loop){ - RTE_LOG(DEBUG, USER1, "dpdk: no packets available and break_loop is setted in blocking mode.\n"); + RTE_LOG(DEBUG, USER1, "dpdk: no packets available and break_loop is set in blocking mode.\n"); p->break_loop = 0; return PCAP_ERROR_BREAK; @@ -366,7 +392,7 @@ static int pcap_dpdk_dispatch(pcap_t *p, int max_cnt, pcap_handler cb, u_char *c bp = rte_pktmbuf_mtod(m, u_char *); }else{ // use fast buffer pcap_tmp_buf if pkt_len is small, no need to call malloc and free - if ( pkt_len <= ETHER_MAX_JUMBO_FRAME_LEN) + if ( pkt_len <= RTE_ETH_PCAP_SNAPLEN) { gather_len = dpdk_gather_data(pd->pcap_tmp_buf, RTE_ETH_PCAP_SNAPLEN, m); bp = pd->pcap_tmp_buf; @@ -379,7 +405,7 @@ static int pcap_dpdk_dispatch(pcap_t *p, int max_cnt, pcap_handler cb, u_char *c } if (bp){ - if (p->fcode.bf_insns==NULL || pcap_filter(p->fcode.bf_insns, bp, pcap_header.len, pcap_header.caplen)){ + if (p->fcode.bf_insns==NULL || pcapint_filter(p->fcode.bf_insns, bp, pcap_header.len, pcap_header.caplen)){ cb(cb_arg, &pcap_header, bp); }else{ pd->bpf_drop++; @@ -399,7 +425,7 @@ static int pcap_dpdk_dispatch(pcap_t *p, int max_cnt, pcap_handler cb, u_char *c static int pcap_dpdk_inject(pcap_t *p, const void *buf _U_, int size _U_) { //not implemented yet - pcap_strlcpy(p->errbuf, + pcapint_strlcpy(p->errbuf, "dpdk error: Inject function has not been implemented yet", PCAP_ERRBUF_SIZE); return PCAP_ERROR; @@ -418,7 +444,7 @@ static void pcap_dpdk_close(pcap_t *p) } rte_eth_dev_stop(pd->portid); rte_eth_dev_close(pd->portid); - pcap_cleanup_live_common(p); + pcapint_cleanup_live_common(p); } static void nic_stats_display(struct pcap_dpdk *pd) @@ -473,11 +499,11 @@ static int check_link_status(uint16_t portid, struct rte_eth_link *plink) rte_eth_link_get(portid, plink); return plink->link_status == ETH_LINK_UP; } -static void eth_addr_str(struct ether_addr *addrp, char* mac_str, int len) +static void eth_addr_str(ETHER_ADDR_TYPE *addrp, char* mac_str, int len) { int offset=0; if (addrp == NULL){ - pcap_snprintf(mac_str, len-1, DPDK_DEF_MAC_ADDR); + snprintf(mac_str, len-1, DPDK_DEF_MAC_ADDR); return; } for (int i=0; i<6; i++) @@ -488,10 +514,10 @@ static void eth_addr_str(struct ether_addr *addrp, char* mac_str, int len) } if (i==0) { - pcap_snprintf(mac_str+offset, len-1-offset, "%02X",addrp->addr_bytes[i]); + snprintf(mac_str+offset, len-1-offset, "%02X",addrp->addr_bytes[i]); offset+=2; // FF }else{ - pcap_snprintf(mac_str+offset, len-1-offset, ":%02X", addrp->addr_bytes[i]); + snprintf(mac_str+offset, len-1-offset, ":%02X", addrp->addr_bytes[i]); offset+=3; // :FF } } @@ -501,8 +527,8 @@ static void eth_addr_str(struct ether_addr *addrp, char* mac_str, int len) static uint16_t portid_by_device(char * device) { uint16_t ret = DPDK_PORTID_MAX; - int len = strlen(device); - int prefix_len = strlen(DPDK_PREFIX); + size_t len = strlen(device); + size_t prefix_len = strlen(DPDK_PREFIX); unsigned long ret_ul = 0L; char *pEnd; if (len<=prefix_len || strncmp(device, DPDK_PREFIX, prefix_len)) // check prefix dpdk: @@ -542,7 +568,7 @@ static int parse_dpdk_cfg(char* dpdk_cfg,char** dargv) skip_space=!skip_space; // skip normal char dargv[cnt++] = dpdk_cfg+i; } - if (!skip_space && dpdk_cfg[i]==' '){ // fint a space + if (!skip_space && dpdk_cfg[i]==' '){ // find a space dpdk_cfg[i]=0x00; // end of this opt skip_space=!skip_space; // skip space char } @@ -572,7 +598,7 @@ static int dpdk_pre_init(char * ebuf, int eaccess_not_fatal) char *dargv[DPDK_ARGC_MAX]; char *ptr_dpdk_cfg = NULL; int ret; - // globale var + // global var if (is_dpdk_pre_inited != 0) { // already inited; did that succeed? @@ -623,7 +649,7 @@ error: if (eaccess_not_fatal) return 0; // Otherwise report a fatal error. - pcap_snprintf(ebuf, PCAP_ERRBUF_SIZE, + snprintf(ebuf, PCAP_ERRBUF_SIZE, "DPDK requires that it run as root"); return PCAP_ERROR_PERM_DENIED; @@ -633,7 +659,7 @@ error: // be attempted again." // There's no such error in pcap, so I'm // not sure what we should do here. - pcap_snprintf(ebuf, PCAP_ERRBUF_SIZE, + snprintf(ebuf, PCAP_ERRBUF_SIZE, "Bus or system resource was not available"); break; @@ -650,7 +676,7 @@ error: case EFAULT: // This "indicates the tailq configuration // name was not found in memory configuration." - pcap_snprintf(ebuf, PCAP_ERRBUF_SIZE, + snprintf(ebuf, PCAP_ERRBUF_SIZE, "The tailq configuration name was not found in the memory configuration"); return PCAP_ERROR; @@ -658,20 +684,20 @@ error: // This "indicates invalid parameters were // passed as argv/argc." Those came from // the configuration file. - pcap_snprintf(ebuf, PCAP_ERRBUF_SIZE, + snprintf(ebuf, PCAP_ERRBUF_SIZE, "The configuration file has invalid parameters"); break; case ENOMEM: // This "indicates failure likely caused by // an out-of-memory condition." - pcap_snprintf(ebuf, PCAP_ERRBUF_SIZE, + snprintf(ebuf, PCAP_ERRBUF_SIZE, "Out of memory"); break; case ENODEV: // This "indicates memory setup issues." - pcap_snprintf(ebuf, PCAP_ERRBUF_SIZE, + snprintf(ebuf, PCAP_ERRBUF_SIZE, "An error occurred setting up memory"); break; @@ -697,14 +723,14 @@ error: // as a "not available" case? If not, we // can't distinguish between the two, so // we're stuck. - pcap_snprintf(ebuf, PCAP_ERRBUF_SIZE, + snprintf(ebuf, PCAP_ERRBUF_SIZE, "PCI bus is not present or not readable by the EAL"); break; case ENOEXEC: // This "indicates that a service core // failed to launch successfully." - pcap_snprintf(ebuf, PCAP_ERRBUF_SIZE, + snprintf(ebuf, PCAP_ERRBUF_SIZE, "A service core failed to launch successfully"); break; @@ -744,7 +770,7 @@ static int pcap_dpdk_activate(pcap_t *p) if (ret < 0) { // This returns a negative value on an error. - pcap_snprintf(p->errbuf, PCAP_ERRBUF_SIZE, + snprintf(p->errbuf, PCAP_ERRBUF_SIZE, "Can't open device %s: %s", p->opt.device, dpdk_pre_init_errbuf); // ret is set to the correct error @@ -753,16 +779,16 @@ static int pcap_dpdk_activate(pcap_t *p) if (ret == 0) { // This means DPDK isn't available on this machine. - pcap_snprintf(p->errbuf, PCAP_ERRBUF_SIZE, + snprintf(p->errbuf, PCAP_ERRBUF_SIZE, "Can't open device %s: DPDK is not available on this machine", p->opt.device); return PCAP_ERROR_NO_SUCH_DEVICE; } - + ret = dpdk_init_timer(pd); if (ret<0) { - pcap_snprintf(p->errbuf, PCAP_ERRBUF_SIZE, + snprintf(p->errbuf, PCAP_ERRBUF_SIZE, "dpdk error: Init timer is zero with device %s", p->opt.device); ret = PCAP_ERROR; @@ -772,7 +798,7 @@ static int pcap_dpdk_activate(pcap_t *p) nb_ports = rte_eth_dev_count_avail(); if (nb_ports == 0) { - pcap_snprintf(p->errbuf, PCAP_ERRBUF_SIZE, + snprintf(p->errbuf, PCAP_ERRBUF_SIZE, "dpdk error: No Ethernet ports"); ret = PCAP_ERROR; break; @@ -780,7 +806,7 @@ static int pcap_dpdk_activate(pcap_t *p) portid = portid_by_device(p->opt.device); if (portid == DPDK_PORTID_MAX){ - pcap_snprintf(p->errbuf, PCAP_ERRBUF_SIZE, + snprintf(p->errbuf, PCAP_ERRBUF_SIZE, "dpdk error: portid is invalid. device %s", p->opt.device); ret = PCAP_ERROR_NO_SUCH_DEVICE; @@ -875,7 +901,7 @@ static int pcap_dpdk_activate(pcap_t *p) rte_eth_dev_socket_id(portid)); if (tx_buffer == NULL) { - pcap_snprintf(p->errbuf, PCAP_ERRBUF_SIZE, + snprintf(p->errbuf, PCAP_ERRBUF_SIZE, "dpdk error: Cannot allocate buffer for tx on port %u", portid); ret = PCAP_ERROR; break; @@ -900,7 +926,7 @@ static int pcap_dpdk_activate(pcap_t *p) // check link status is_port_up = check_link_status(portid, &link); if (!is_port_up){ - pcap_snprintf(p->errbuf, PCAP_ERRBUF_SIZE, + snprintf(p->errbuf, PCAP_ERRBUF_SIZE, "dpdk error: link is down, port=%u",portid); ret = PCAP_ERROR_IFACE_NOT_UP; break; @@ -920,15 +946,15 @@ static int pcap_dpdk_activate(pcap_t *p) p->selectable_fd = p->fd; p->read_op = pcap_dpdk_dispatch; p->inject_op = pcap_dpdk_inject; - // using pcap_filter currently, though DPDK provides their own BPF function. Because DPDK BPF needs load a ELF file as a filter. - p->setfilter_op = install_bpf_program; + // using pcapint_filter currently, though DPDK provides their own BPF function. Because DPDK BPF needs load a ELF file as a filter. + p->setfilter_op = pcapint_install_bpf_program; p->setdirection_op = NULL; p->set_datalink_op = NULL; p->getnonblock_op = pcap_dpdk_getnonblock; p->setnonblock_op = pcap_dpdk_setnonblock; p->stats_op = pcap_dpdk_stats; p->cleanup_op = pcap_dpdk_close; - p->breakloop_op = pcap_breakloop_common; + p->breakloop_op = pcapint_breakloop_common; // set default timeout pd->required_select_timeout.tv_sec = 0; pd->required_select_timeout.tv_usec = DPDK_DEF_MIN_SLEEP_MS*1000; @@ -938,7 +964,7 @@ static int pcap_dpdk_activate(pcap_t *p) if (ret <= PCAP_ERROR) // all kinds of error code { - pcap_cleanup_live_common(p); + pcapint_cleanup_live_common(p); }else{ rte_eth_dev_get_name_by_port(portid,pd->pci_addr); RTE_LOG(INFO, USER1,"Port %d device: %s, MAC:%s, PCI:%s\n", portid, p->opt.device, pd->mac_addr, pd->pci_addr); @@ -950,7 +976,7 @@ static int pcap_dpdk_activate(pcap_t *p) return ret; } -// device name for dpdk shoud be in the form as dpdk:number, such as dpdk:0 +// device name for dpdk should be in the form as dpdk:number, such as dpdk:0 pcap_t * pcap_dpdk_create(const char *device, char *ebuf, int *is_ours) { pcap_t *p=NULL; @@ -960,7 +986,7 @@ pcap_t * pcap_dpdk_create(const char *device, char *ebuf, int *is_ours) if (! *is_ours) return NULL; //memset will happen - p = pcap_create_common(ebuf, sizeof(struct pcap_dpdk)); + p = PCAP_CREATE_COMMON(ebuf, struct pcap_dpdk); if (p == NULL) return NULL; @@ -974,7 +1000,7 @@ int pcap_dpdk_findalldevs(pcap_if_list_t *devlistp, char *ebuf) unsigned int nb_ports = 0; char dpdk_name[DPDK_DEV_NAME_MAX]; char dpdk_desc[DPDK_DEV_DESC_MAX]; - struct ether_addr eth_addr; + ETHER_ADDR_TYPE eth_addr; char mac_addr[DPDK_MAC_ADDR_SIZE]; char pci_addr[DPDK_PCI_ADDR_SIZE]; do{ @@ -985,7 +1011,7 @@ int pcap_dpdk_findalldevs(pcap_if_list_t *devlistp, char *ebuf) if (ret < 0) { // This returns a negative value on an error. - pcap_snprintf(ebuf, PCAP_ERRBUF_SIZE, + snprintf(ebuf, PCAP_ERRBUF_SIZE, "Can't look for DPDK devices: %s", dpdk_pre_init_errbuf); ret = PCAP_ERROR; @@ -1005,15 +1031,15 @@ int pcap_dpdk_findalldevs(pcap_if_list_t *devlistp, char *ebuf) break; } for (unsigned int i=0; i