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:
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}
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 <config.h>
-#endif
-#include <ctype.h>
#include <errno.h>
#include <netdb.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <unistd.h>
+#include <limits.h> /* for INT_MAX */
#include <time.h>
#include <sys/time.h>
//header for calling dpdk
+#include <rte_config.h>
#include <rte_common.h>
+#include <rte_errno.h>
#include <rte_log.h>
#include <rte_malloc.h>
#include <rte_memory.h>
#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
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;
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];
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);
* 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){
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 ){
}
-static int dpdk_read_with_timeout(pcap_t *p, uint16_t portid, uint16_t queueid,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;
int sleep_ms = 0;
if (pd->nonblock){
- // In non-blocking mode, just read once, no mater how many packets are captured.
+ // 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);
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;
- int caplen = 0;
+ uint32_t caplen = 0;
u_char *bp = NULL;
int i=0;
unsigned int gather_len =0;
int pkt_cnt = 0;
- int is_accepted=0;
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, 0, 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;
pkt_len = rte_pktmbuf_pkt_len(m);
// caplen = min(pkt_len, p->snapshot);
// caplen will not be changed, no matter how long the rte_pktmbuf
- caplen = pkt_len < p->snapshot ? pkt_len: p->snapshot;
+ caplen = pkt_len < (uint32_t)p->snapshot ? pkt_len: (uint32_t)p->snapshot;
pcap_header.caplen = caplen;
pcap_header.len = pkt_len;
// volatile prefetch
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;
}
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++;
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;
}
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)
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++)
}
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
}
}
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:
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
}
// meaning "DPDK isn't available";
//
// a PCAP_ERROR_ code for other errors.
-static int dpdk_pre_init(char * ebuf)
+//
+// If eaccess_not_fatal is non-zero, treat "a permissions issue" the way
+// we treat "the EAL cannot initialize on this system". We use that
+// when trying to find DPDK devices, as we don't want to fail to return
+// *any* devices just because we can't support DPDK; when we're trying
+// to open a device, we need to return a permissions error in that case.
+static int dpdk_pre_init(char * ebuf, int eaccess_not_fatal)
{
int dargv_cnt=0;
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?
return 1;
}
}
- // check for root permission
- // XXX - is it sufficient to call rte_eal_init() and check for
- // EACCES?
- if( geteuid() != 0)
- {
- RTE_LOG(ERR, USER1, "%s\n", DPDK_ERR_PERM_MSG);
- pcap_snprintf(ebuf, PCAP_ERRBUF_SIZE,
- "DPDK requires that it run as root");
- return PCAP_ERROR_PERM_DENIED;
- }
// init EAL
ptr_dpdk_cfg = getenv(DPDK_CFG_ENV_NAME);
// set default log level to debug
case EACCES:
// This "indicates a permissions issue.".
RTE_LOG(ERR, USER1, "%s\n", DPDK_ERR_PERM_MSG);
- pcap_snprintf(ebuf, PCAP_ERRBUF_SIZE,
+ // If we were told to treat this as just meaning
+ // DPDK isn't available, do so.
+ if (eaccess_not_fatal)
+ return 0;
+ // Otherwise report a fatal error.
+ snprintf(ebuf, PCAP_ERRBUF_SIZE,
"DPDK requires that it run as root");
return PCAP_ERROR_PERM_DENIED;
// 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;
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;
// 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;
// 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;
int is_port_up = 0;
struct rte_eth_link link;
do{
- //init EAL
+ //init EAL; fail if we have insufficient permission
char dpdk_pre_init_errbuf[PCAP_ERRBUF_SIZE];
- ret = dpdk_pre_init(dpdk_pre_init_errbuf);
+ ret = dpdk_pre_init(dpdk_pre_init_errbuf, 0);
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
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;
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;
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;
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;
// 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;
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;
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);
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;
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;
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{
+ // init EAL; return "DPDK not available" if we
+ // have insufficient permission
char dpdk_pre_init_errbuf[PCAP_ERRBUF_SIZE];
- ret = dpdk_pre_init(dpdk_pre_init_errbuf);
+ ret = dpdk_pre_init(dpdk_pre_init_errbuf, 1);
if (ret < 0)
{
// This returns a negative value on an error.
- pcap_snprintf(p->errbuf, PCAP_ERRBUF_SIZE,
- "Can't open device %s: %s",
- p->opt.device, dpdk_pre_init_errbuf);
- ret = PCAP_ERRNO;
+ snprintf(ebuf, PCAP_ERRBUF_SIZE,
+ "Can't look for DPDK devices: %s",
+ dpdk_pre_init_errbuf);
+ ret = PCAP_ERROR;
break;
}
if (ret == 0)
break;
}
for (unsigned int i=0; i<nb_ports; i++){
- pcap_snprintf(dpdk_name, DPDK_DEV_NAME_MAX-1,
+ snprintf(dpdk_name, DPDK_DEV_NAME_MAX-1,
"%s%u", DPDK_PREFIX, i);
// mac addr
rte_eth_macaddr_get(i, ð_addr);
eth_addr_str(ð_addr,mac_addr,DPDK_MAC_ADDR_SIZE);
// PCI addr
rte_eth_dev_get_name_by_port(i,pci_addr);
- pcap_snprintf(dpdk_desc,DPDK_DEV_DESC_MAX-1,"%s %s, MAC:%s, PCI:%s", DPDK_DESC, dpdk_name, mac_addr, pci_addr);
- if (add_dev(devlistp, dpdk_name, 0, dpdk_desc, ebuf)==NULL){
+ snprintf(dpdk_desc,DPDK_DEV_DESC_MAX-1,"%s %s, MAC:%s, PCI:%s", DPDK_DESC, dpdk_name, mac_addr, pci_addr);
+ if (pcapint_add_dev(devlistp, dpdk_name, 0, dpdk_desc, ebuf)==NULL){
ret = PCAP_ERROR;
break;
}
* There are no regular interfaces, just DPDK interfaces.
*/
int
-pcap_platform_finddevs(pcap_if_list_t *devlistp _U_, char *errbuf)
+pcapint_platform_finddevs(pcap_if_list_t *devlistp _U_, char *errbuf)
{
return (0);
}
* Attempts to open a regular interface fail.
*/
pcap_t *
-pcap_create_interface(const char *device, char *errbuf)
+pcapint_create_interface(const char *device, char *errbuf)
{
- pcap_snprintf(errbuf, PCAP_ERRBUF_SIZE,
- "This version of libpcap only supports DPDK");
+ snprintf(errbuf, PCAP_ERRBUF_SIZE, PCAP_ENODEV_MESSAGE, "DPDK");
return NULL;
}
projects / libpcap / blobdiff