pcap-dpdk.c

/*
 * Copyright (C) 2018 jingle YANG. All rights reserved.
 *
 * Redistribution and use in source and binary forms, with or without
 * modification, are permitted provided that the following conditions
 * are met:
 *
 *   1. Redistributions of source code must retain the above copyright
 *      notice, this list of conditions and the following disclaimer.
 *   2. Redistributions in binary form must reproduce the above copyright
 *      notice, this list of conditions and the following disclaimer in the
 *      documentation and/or other materials provided with the distribution.
 *
 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS''AND
 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
 * SUCH DAMAGE.
 */

/*
Date: Dec 16, 2018

Description:
1. Pcap-dpdk provides libpcap the ability to use DPDK with the device name as dpdk:{portid}, such as dpdk:0.
2. DPDK is a set of libraries and drivers for fast packet processing. (https://www.dpdk.org/)
3. The testprogs/capturetest provides 6.4Gbps/800,000 pps on Intel 10-Gigabit X540-AT2 with DPDK 18.11.

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 dynamically, because the libdpdk.a will not work correctly.
3. Only support read operation, and packet injection has not been supported yet.

Usage:
1. Compile DPDK as shared library and install.(https://github.com/DPDK/dpdk.git)

You shall modify the file $RTE_SDK/$RTE_TARGET/.config and set:
CONFIG_RTE_BUILD_SHARED_LIB=y
By the following command:
sed -i 's/CONFIG_RTE_BUILD_SHARED_LIB=n/CONFIG_RTE_BUILD_SHARED_LIB=y/' $RTE_SDK/$RTE_TARGET/.config

2. Launch l2fwd that is one of DPDK examples correctly, and get device information.

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 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 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}

3.1 With configure

./configure --with-dpdk=$RTE_SDK/$RTE_TARGET && make -s all && make -s testprogs && make install

3.2 With cmake

mkdir -p build && cd build && cmake -DDPDK_DIR=$RTE_SDK/$RTE_TARGET ../ && make -s all && make -s testprogs && make install

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 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
*/

#include <config.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 <rte_eal.h>
#include <rte_launch.h>
#include <rte_atomic.h>
#include <rte_cycles.h>
#include <rte_lcore.h>
#include <rte_per_lcore.h>
#include <rte_branch_prediction.h>
#include <rte_interrupts.h>
#include <rte_random.h>
#include <rte_debug.h>
#include <rte_ether.h>
#include <rte_ethdev.h>
#include <rte_mempool.h>
#include <rte_mbuf.h>
#include <rte_bus.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
// successfully initialized it, a negative value, which is the negative
// of the rte_errno from rte_eal_init(), if we tried to initialize it
// and got an error.
//
static int is_dpdk_pre_inited=0;
#define DPDK_LIB_NAME "libpcap_dpdk"
#define DPDK_DESC "Data Plane Development Kit (DPDK) Interface"
#define DPDK_ERR_PERM_MSG "permission denied, DPDK needs root permission"
#define DPDK_ARGC_MAX 64
#define DPDK_CFG_MAX_LEN 1024
#define DPDK_DEV_NAME_MAX 32
#define DPDK_DEV_DESC_MAX 512
#define DPDK_CFG_ENV_NAME "DPDK_CFG"
#define DPDK_DEF_MIN_SLEEP_MS 1
static char dpdk_cfg_buf[DPDK_CFG_MAX_LEN];
#define DPDK_MAC_ADDR_SIZE 32
#define DPDK_DEF_MAC_ADDR "00:00:00:00:00:00"
#define DPDK_PCI_ADDR_SIZE 16
#define DPDK_DEF_CFG "--log-level=error -l0 -dlibrte_pmd_e1000.so -dlibrte_pmd_ixgbe.so -dlibrte_mempool_ring.so"
#define DPDK_PREFIX "dpdk:"
#define DPDK_PORTID_MAX 65535U
#define MBUF_POOL_NAME "mbuf_pool"
#define DPDK_TX_BUF_NAME "tx_buffer"
//The number of elements in the mbuf pool.
#define DPDK_NB_MBUFS 8192U
#define MEMPOOL_CACHE_SIZE 256
#define MAX_PKT_BURST 32
// Configurable number of RX/TX ring descriptors
#define RTE_TEST_RX_DESC_DEFAULT 1024
#define RTE_TEST_TX_DESC_DEFAULT 1024

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;

struct dpdk_ts_helper{
	struct timeval start_time;
	uint64_t start_cycles;
	uint64_t hz;
};
struct pcap_dpdk{
	pcap_t * orig;
	uint16_t portid; // portid of DPDK
	int must_clear_promisc;
	uint64_t bpf_drop;
	int nonblock;
	struct timeval required_select_timeout;
	struct timeval prev_ts;
	struct rte_eth_stats prev_stats;
	struct timeval curr_ts;
	struct rte_eth_stats curr_stats;
	uint64_t pps;
	uint64_t bps;
	struct rte_mempool * pktmbuf_pool;
	struct dpdk_ts_helper ts_helper;
	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];
};

static struct rte_eth_conf port_conf = {
	.rxmode = {
		.split_hdr_size = 0,
	},
	.txmode = {
		.mq_mode = ETH_MQ_TX_NONE,
	},
};

static void	dpdk_fmt_errmsg_for_rte_errno(char *, size_t, int,
    PCAP_FORMAT_STRING(const char *), ...) PCAP_PRINTFLIKE(4, 5);

/*
 * Generate an error message based on a format, arguments, and an
 * rte_errno, with a message for the rte_errno after the formatted output.
 */
static void dpdk_fmt_errmsg_for_rte_errno(char *errbuf, size_t errbuflen,
    int errnum, const char *fmt, ...)
{
	va_list ap;
	size_t msglen;
	char *p;
	size_t errbuflen_remaining;

	va_start(ap, fmt);
	vsnprintf(errbuf, errbuflen, fmt, ap);
	va_end(ap);
	msglen = strlen(errbuf);

	/*
	 * Do we have enough space to append ": "?
	 * Including the terminating '\0', that's 3 bytes.
	 */
	if (msglen + 3 > errbuflen) {
		/* No - just give them what we've produced. */
		return;
	}
	p = errbuf + msglen;
	errbuflen_remaining = errbuflen - msglen;
	*p++ = ':';
	*p++ = ' ';
	*p = '\0';
	msglen += 2;
	errbuflen_remaining -= 2;

	/*
	 * Now append the string for the error code.
	 * rte_strerror() is thread-safe, at least as of dpdk 18.11,
	 * unlike strerror() - it uses strerror_r() rather than strerror()
	 * for UN*X errno values, and prints to what I assume is a per-thread
	 * buffer (based on the "PER_LCORE" in "RTE_DEFINE_PER_LCORE" used
	 * to declare the buffers statically) for DPDK errors.
	 */
	snprintf(p, errbuflen_remaining, "%s", rte_strerror(errnum));
}

static int dpdk_init_timer(struct pcap_dpdk *pd){
	gettimeofday(&(pd->ts_helper.start_time),NULL);
	pd->ts_helper.start_cycles = rte_get_timer_cycles();
	pd->ts_helper.hz = rte_get_timer_hz();
	if (pd->ts_helper.hz == 0){
		return -1;
	}
	return 0;
}
static inline void calculate_timestamp(struct dpdk_ts_helper *helper,struct timeval *ts)
{
	uint64_t cycles;
	// delta
	struct timeval cur_time;
	cycles = rte_get_timer_cycles() - helper->start_cycles;
	cur_time.tv_sec = (time_t)(cycles/helper->hz);
	cur_time.tv_usec = (suseconds_t)((cycles%helper->hz)*1e6/helper->hz);
	timeradd(&(helper->start_time), &cur_time, ts);
}

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 ){
		rte_memcpy(data+total_len, rte_pktmbuf_mtod(mbuf,void *),mbuf->data_len);
		total_len+=mbuf->data_len;
		mbuf=mbuf->next;
	}
	return total_len;
}


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 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 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);
			if (nb_rx){ // got packets within timeout_ms
				break;
			}else{ // no packet arrives at this round.
				if (p->break_loop){
					break;
				}
				// sleep for a very short while.
				// block sleep is the only choice, since usleep() will impact performance dramatically.
				rte_delay_us_block(DPDK_DEF_MIN_SLEEP_MS*1000);
				sleep_ms += DPDK_DEF_MIN_SLEEP_MS;
			}
		}
	}
	return nb_rx;
}

static int pcap_dpdk_dispatch(pcap_t *p, int max_cnt, pcap_handler cb, u_char *cb_arg)
{
	struct pcap_dpdk *pd = (struct pcap_dpdk*)(p->priv);
	int burst_cnt = 0;
	int nb_rx = 0;
	struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
	struct rte_mbuf *m;
	struct pcap_pkthdr pcap_header;
	// 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;
	u_char *bp = NULL;
	int i=0;
	unsigned int gather_len =0;
	int pkt_cnt = 0;
	u_char *large_buffer=NULL;
	int timeout_ms = p->opt.timeout;

	/*
	 * 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( 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 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 set in blocking mode.\n");
					p->break_loop = 0;
					return PCAP_ERROR_BREAK;

				}
				RTE_LOG(DEBUG, USER1, "dpdk: no packets available for timeout %d ms in blocking mode.\n", timeout_ms);
			}
			// break if dpdk reads 0 packet, no matter in blocking(timeout) or non-blocking mode.
			break;
		}
		pkt_cnt += nb_rx;
		for ( i = 0; i < nb_rx; i++) {
			m = pkts_burst[i];
			calculate_timestamp(&(pd->ts_helper),&(pcap_header.ts));
			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 < (uint32_t)p->snapshot ? pkt_len: (uint32_t)p->snapshot;
			pcap_header.caplen = caplen;
			pcap_header.len = pkt_len;
			// volatile prefetch
			rte_prefetch0(rte_pktmbuf_mtod(m, void *));
			bp = NULL;
			if (m->nb_segs == 1)
			{
				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 <= RTE_ETH_PCAP_SNAPLEN)
				{
					gather_len = dpdk_gather_data(pd->pcap_tmp_buf, RTE_ETH_PCAP_SNAPLEN, m);
					bp = pd->pcap_tmp_buf;
				}else{
					// need call free later
					large_buffer = (u_char *)malloc(caplen*sizeof(u_char));
					gather_len = dpdk_gather_data(large_buffer, caplen, m);
					bp = large_buffer;
				}

			}
			if (bp){
				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++;
				}
			}
			//free all pktmbuf
			rte_pktmbuf_free(m);
			if (large_buffer){
				free(large_buffer);
				large_buffer=NULL;
			}
		}
	}
	return pkt_cnt;
}

static int pcap_dpdk_inject(pcap_t *p, const void *buf _U_, int size _U_)
{
	//not implemented yet
	pcapint_strlcpy(p->errbuf,
	    "dpdk error: Inject function has not been implemented yet",
	    PCAP_ERRBUF_SIZE);
	return PCAP_ERROR;
}

static void pcap_dpdk_close(pcap_t *p)
{
	struct pcap_dpdk *pd = p->priv;
	if (pd==NULL)
	{
		return;
	}
	if (pd->must_clear_promisc)
	{
		rte_eth_promiscuous_disable(pd->portid);
	}
	rte_eth_dev_stop(pd->portid);
	rte_eth_dev_close(pd->portid);
	pcapint_cleanup_live_common(p);
}

static void nic_stats_display(struct pcap_dpdk *pd)
{
	uint16_t portid = pd->portid;
	struct rte_eth_stats stats;
	rte_eth_stats_get(portid, &stats);
	RTE_LOG(INFO,USER1, "portid:%d, RX-packets: %-10"PRIu64"  RX-errors:  %-10"PRIu64
	       "  RX-bytes:  %-10"PRIu64"  RX-Imissed:  %-10"PRIu64"\n", portid, stats.ipackets, stats.ierrors,
	       stats.ibytes,stats.imissed);
	RTE_LOG(INFO,USER1, "portid:%d, RX-PPS: %-10"PRIu64" RX-Mbps: %.2lf\n", portid, pd->pps, pd->bps/1e6f );
}

static int pcap_dpdk_stats(pcap_t *p, struct pcap_stat *ps)
{
	struct pcap_dpdk *pd = p->priv;
	calculate_timestamp(&(pd->ts_helper), &(pd->curr_ts));
	rte_eth_stats_get(pd->portid,&(pd->curr_stats));
	if (ps){
		ps->ps_recv = pd->curr_stats.ipackets;
		ps->ps_drop = pd->curr_stats.ierrors;
		ps->ps_drop += pd->bpf_drop;
		ps->ps_ifdrop = pd->curr_stats.imissed;
	}
	uint64_t delta_pkt = pd->curr_stats.ipackets - pd->prev_stats.ipackets;
	struct timeval delta_tm;
	timersub(&(pd->curr_ts),&(pd->prev_ts), &delta_tm);
	uint64_t delta_usec = delta_tm.tv_sec*1e6+delta_tm.tv_usec;
	uint64_t delta_bit = (pd->curr_stats.ibytes-pd->prev_stats.ibytes)*8;
	RTE_LOG(DEBUG, USER1, "delta_usec: %-10"PRIu64" delta_pkt: %-10"PRIu64" delta_bit: %-10"PRIu64"\n", delta_usec, delta_pkt, delta_bit);
	pd->pps = (uint64_t)(delta_pkt*1e6f/delta_usec);
	pd->bps = (uint64_t)(delta_bit*1e6f/delta_usec);
	nic_stats_display(pd);
	pd->prev_stats = pd->curr_stats;
	pd->prev_ts = pd->curr_ts;
	return 0;
}

static int pcap_dpdk_setnonblock(pcap_t *p, int nonblock){
	struct pcap_dpdk *pd = (struct pcap_dpdk*)(p->priv);
	pd->nonblock = nonblock;
	return 0;
}

static int pcap_dpdk_getnonblock(pcap_t *p){
	struct pcap_dpdk *pd = (struct pcap_dpdk*)(p->priv);
	return pd->nonblock;
}
static int check_link_status(uint16_t portid, struct rte_eth_link *plink)
{
	// wait up to 9 seconds to get link status
	rte_eth_link_get(portid, plink);
	return plink->link_status == ETH_LINK_UP;
}
static void eth_addr_str(ETHER_ADDR_TYPE *addrp, char* mac_str, int len)
{
	int offset=0;
	if (addrp == NULL){
		snprintf(mac_str, len-1, DPDK_DEF_MAC_ADDR);
		return;
	}
	for (int i=0; i<6; i++)
	{
		if (offset >= len)
		{ // buffer overflow
			return;
		}
		if (i==0)
		{
			snprintf(mac_str+offset, len-1-offset, "%02X",addrp->addr_bytes[i]);
			offset+=2; // FF
		}else{
			snprintf(mac_str+offset, len-1-offset, ":%02X", addrp->addr_bytes[i]);
			offset+=3; // :FF
		}
	}
	return;
}
// return portid by device name, otherwise return -1
static uint16_t portid_by_device(char * device)
{
	uint16_t ret = DPDK_PORTID_MAX;
	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:
	{
		return ret;
	}
	//check all chars are digital
	for (int i=prefix_len; device[i]; i++){
		if (device[i]<'0' || device[i]>'9'){
			return ret;
		}
	}
	ret_ul = strtoul(&(device[prefix_len]), &pEnd, 10);
	if (pEnd == &(device[prefix_len]) || *pEnd != '\0'){
		return ret;
	}
	// too large for portid
	if (ret_ul >= DPDK_PORTID_MAX){
		return ret;
	}
	ret = (uint16_t)ret_ul;
	return ret;
}

static int parse_dpdk_cfg(char* dpdk_cfg,char** dargv)
{
	int cnt=0;
	memset(dargv,0,sizeof(dargv[0])*DPDK_ARGC_MAX);
	//current process name
	int skip_space = 1;
	int i=0;
	RTE_LOG(INFO, USER1,"dpdk cfg: %s\n",dpdk_cfg);
	// find first non space char
	// The last opt is NULL
	for (i=0;dpdk_cfg[i] && cnt<DPDK_ARGC_MAX-1;i++){
		if (skip_space && dpdk_cfg[i]!=' '){ // not space
			skip_space=!skip_space; // skip normal char
			dargv[cnt++] = dpdk_cfg+i;
		}
		if (!skip_space && dpdk_cfg[i]==' '){ // find a space
			dpdk_cfg[i]=0x00; // end of this opt
			skip_space=!skip_space; // skip space char
		}
	}
	dargv[cnt]=NULL;
	return cnt;
}

// only called once
// Returns:
//
//    1 on success;
//
//    0 if "the EAL cannot initialize on this system", which we treat as
//    meaning "DPDK isn't available";
//
//    a PCAP_ERROR_ code for other errors.
//
// 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;
	// global var
	if (is_dpdk_pre_inited != 0)
	{
		// already inited; did that succeed?
		if (is_dpdk_pre_inited < 0)
		{
			// failed
			goto error;
		}
		else
		{
			// succeeded
			return 1;
		}
	}
	// init EAL
	ptr_dpdk_cfg = getenv(DPDK_CFG_ENV_NAME);
	// set default log level to debug
	rte_log_set_global_level(DPDK_DEF_LOG_LEV);
	if (ptr_dpdk_cfg == NULL)
	{
		RTE_LOG(INFO,USER1,"env $DPDK_CFG is unset, so using default: %s\n",DPDK_DEF_CFG);
		ptr_dpdk_cfg = DPDK_DEF_CFG;
	}
	memset(dpdk_cfg_buf,0,sizeof(dpdk_cfg_buf));
	snprintf(dpdk_cfg_buf,DPDK_CFG_MAX_LEN-1,"%s %s",DPDK_LIB_NAME,ptr_dpdk_cfg);
	dargv_cnt = parse_dpdk_cfg(dpdk_cfg_buf,dargv);
	ret = rte_eal_init(dargv_cnt,dargv);
	if (ret == -1)
	{
		// Indicate that we've called rte_eal_init() by setting
		// is_dpdk_pre_inited to the negative of the error code,
		// and process the error.
		is_dpdk_pre_inited = -rte_errno;
		goto error;
	}
	// init succeeded, so we do not need to do it again later.
	is_dpdk_pre_inited = 1;
	return 1;

error:
	switch (-is_dpdk_pre_inited)
	{
		case EACCES:
			// This "indicates a permissions issue.".
			RTE_LOG(ERR, USER1, "%s\n", DPDK_ERR_PERM_MSG);
			// 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;

		case EAGAIN:
			// This "indicates either a bus or system
			// resource was not available, setup may
			// be attempted again."
			// There's no such error in pcap, so I'm
			// not sure what we should do here.
			snprintf(ebuf, PCAP_ERRBUF_SIZE,
			    "Bus or system resource was not available");
			break;

		case EALREADY:
			// This "indicates that the rte_eal_init
			// function has already been called, and
			// cannot be called again."
			// That's not an error; set the "we've
			// been here before" flag and return
			// success.
			is_dpdk_pre_inited = 1;
			return 1;

		case EFAULT:
			// This "indicates the tailq configuration
			// name was not found in memory configuration."
			snprintf(ebuf, PCAP_ERRBUF_SIZE,
			    "The tailq configuration name was not found in the memory configuration");
			return PCAP_ERROR;

		case EINVAL:
			// This "indicates invalid parameters were
			// passed as argv/argc."  Those came from
			// the configuration file.
			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."
			snprintf(ebuf, PCAP_ERRBUF_SIZE,
			    "Out of memory");
			break;

		case ENODEV:
			// This "indicates memory setup issues."
			snprintf(ebuf, PCAP_ERRBUF_SIZE,
			    "An error occurred setting up memory");
			break;

		case ENOTSUP:
			// This "indicates that the EAL cannot
			// initialize on this system."  We treat
			// that as meaning DPDK isn't available
			// on this machine, rather than as a
			// fatal error, and let our caller decide
			// whether that's a fatal error (if trying
			// to activate a DPDK device) or not (if
			// trying to enumerate devices).
			return 0;

		case EPROTO:
			// This "indicates that the PCI bus is
			// either not present, or is not readable
			// by the eal."  Does "the PCI bus is not
			// present" mean "this machine has no PCI
			// bus", which strikes me as a "not available"
			// case?  If so, should "is not readable by
			// the EAL" also something we should treat
			// as a "not available" case?  If not, we
			// can't distinguish between the two, so
			// we're stuck.
			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."
			snprintf(ebuf, PCAP_ERRBUF_SIZE,
			    "A service core failed to launch successfully");
			break;

		default:
			//
			// That's not in the list of errors in
			// the documentation; let it be reported
			// as an error.
			//
			dpdk_fmt_errmsg_for_rte_errno(ebuf,
			    PCAP_ERRBUF_SIZE, -is_dpdk_pre_inited,
			    "dpdk error: dpdk_pre_init failed");
			break;
	}
	// Error.
	return PCAP_ERROR;
}

static int pcap_dpdk_activate(pcap_t *p)
{
	struct pcap_dpdk *pd = p->priv;
	pd->orig = p;
	int ret = PCAP_ERROR;
	uint16_t nb_ports=0;
	uint16_t portid= DPDK_PORTID_MAX;
	unsigned nb_mbufs = DPDK_NB_MBUFS;
	struct rte_eth_rxconf rxq_conf;
	struct rte_eth_txconf txq_conf;
	struct rte_eth_conf local_port_conf = port_conf;
	struct rte_eth_dev_info dev_info;
	int is_port_up = 0;
	struct rte_eth_link link;
	do{
		//init EAL; fail if we have insufficient permission
		char dpdk_pre_init_errbuf[PCAP_ERRBUF_SIZE];
		ret = dpdk_pre_init(dpdk_pre_init_errbuf, 0);
		if (ret < 0)
		{
			// This returns a negative value on an error.
			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
			break;
		}
		if (ret == 0)
		{
			// This means DPDK isn't available on this machine.
			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)
		{
			snprintf(p->errbuf, PCAP_ERRBUF_SIZE,
				"dpdk error: Init timer is zero with device %s",
				p->opt.device);
			ret = PCAP_ERROR;
			break;
		}

		nb_ports = rte_eth_dev_count_avail();
		if (nb_ports == 0)
		{
			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){
			snprintf(p->errbuf, PCAP_ERRBUF_SIZE,
			    "dpdk error: portid is invalid. device %s",
			    p->opt.device);
			ret = PCAP_ERROR_NO_SUCH_DEVICE;
			break;
		}

		pd->portid = portid;

		if (p->snapshot <= 0 || p->snapshot > MAXIMUM_SNAPLEN)
		{
			p->snapshot = MAXIMUM_SNAPLEN;
		}
		// create the mbuf pool
		pd->pktmbuf_pool = rte_pktmbuf_pool_create(MBUF_POOL_NAME, nb_mbufs,
			MEMPOOL_CACHE_SIZE, 0, RTE_MBUF_DEFAULT_BUF_SIZE,
			rte_socket_id());
		if (pd->pktmbuf_pool == NULL)
		{
			dpdk_fmt_errmsg_for_rte_errno(p->errbuf,
			    PCAP_ERRBUF_SIZE, rte_errno,
			    "dpdk error: Cannot init mbuf pool");
			ret = PCAP_ERROR;
			break;
		}
		// config dev
		rte_eth_dev_info_get(portid, &dev_info);
		if (dev_info.tx_offload_capa & DEV_TX_OFFLOAD_MBUF_FAST_FREE)
		{
			local_port_conf.txmode.offloads |=DEV_TX_OFFLOAD_MBUF_FAST_FREE;
		}
		// only support 1 queue
		ret = rte_eth_dev_configure(portid, 1, 1, &local_port_conf);
		if (ret < 0)
		{
			dpdk_fmt_errmsg_for_rte_errno(p->errbuf,
			    PCAP_ERRBUF_SIZE, -ret,
			    "dpdk error: Cannot configure device: port=%u",
			    portid);
			ret = PCAP_ERROR;
			break;
		}
		// adjust rx tx
		ret = rte_eth_dev_adjust_nb_rx_tx_desc(portid, &nb_rxd, &nb_txd);
		if (ret < 0)
		{
			dpdk_fmt_errmsg_for_rte_errno(p->errbuf,
			    PCAP_ERRBUF_SIZE, -ret,
			    "dpdk error: Cannot adjust number of descriptors: port=%u",
			    portid);
			ret = PCAP_ERROR;
			break;
		}
		// get MAC addr
		rte_eth_macaddr_get(portid, &(pd->eth_addr));
		eth_addr_str(&(pd->eth_addr), pd->mac_addr, DPDK_MAC_ADDR_SIZE-1);

		// init one RX queue
		rxq_conf = dev_info.default_rxconf;
		rxq_conf.offloads = local_port_conf.rxmode.offloads;
		ret = rte_eth_rx_queue_setup(portid, 0, nb_rxd,
					     rte_eth_dev_socket_id(portid),
					     &rxq_conf,
					     pd->pktmbuf_pool);
		if (ret < 0)
		{
			dpdk_fmt_errmsg_for_rte_errno(p->errbuf,
			    PCAP_ERRBUF_SIZE, -ret,
			    "dpdk error: rte_eth_rx_queue_setup:port=%u",
			    portid);
			ret = PCAP_ERROR;
			break;
		}

		// init one TX queue
		txq_conf = dev_info.default_txconf;
		txq_conf.offloads = local_port_conf.txmode.offloads;
		ret = rte_eth_tx_queue_setup(portid, 0, nb_txd,
				rte_eth_dev_socket_id(portid),
				&txq_conf);
		if (ret < 0)
		{
			dpdk_fmt_errmsg_for_rte_errno(p->errbuf,
			    PCAP_ERRBUF_SIZE, -ret,
			    "dpdk error: rte_eth_tx_queue_setup:port=%u",
			    portid);
			ret = PCAP_ERROR;
			break;
		}
		// Initialize TX buffers
		tx_buffer = rte_zmalloc_socket(DPDK_TX_BUF_NAME,
				RTE_ETH_TX_BUFFER_SIZE(MAX_PKT_BURST), 0,
				rte_eth_dev_socket_id(portid));
		if (tx_buffer == NULL)
		{
			snprintf(p->errbuf, PCAP_ERRBUF_SIZE,
			    "dpdk error: Cannot allocate buffer for tx on port %u", portid);
			ret = PCAP_ERROR;
			break;
		}
		rte_eth_tx_buffer_init(tx_buffer, MAX_PKT_BURST);
		// Start device
		ret = rte_eth_dev_start(portid);
		if (ret < 0)
		{
			dpdk_fmt_errmsg_for_rte_errno(p->errbuf,
			    PCAP_ERRBUF_SIZE, -ret,
			    "dpdk error: rte_eth_dev_start:port=%u",
			    portid);
			ret = PCAP_ERROR;
			break;
		}
		// set promiscuous mode
		if (p->opt.promisc){
			pd->must_clear_promisc=1;
			rte_eth_promiscuous_enable(portid);
		}
		// check link status
		is_port_up = check_link_status(portid, &link);
		if (!is_port_up){
			snprintf(p->errbuf, PCAP_ERRBUF_SIZE,
			    "dpdk error: link is down, port=%u",portid);
			ret = PCAP_ERROR_IFACE_NOT_UP;
			break;
		}
		// reset statistics
		rte_eth_stats_reset(pd->portid);
		calculate_timestamp(&(pd->ts_helper), &(pd->prev_ts));
		rte_eth_stats_get(pd->portid,&(pd->prev_stats));
		// format pcap_t
		pd->portid = portid;
		p->fd = pd->portid;
		if (p->snapshot <=0 || p->snapshot> MAXIMUM_SNAPLEN)
		{
			p->snapshot = MAXIMUM_SNAPLEN;
		}
		p->linktype = DLT_EN10MB; // Ethernet, the 10MB is historical.
		p->selectable_fd = p->fd;
		p->read_op = pcap_dpdk_dispatch;
		p->inject_op = pcap_dpdk_inject;
		// 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 = 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;
		p->required_select_timeout = &pd->required_select_timeout;
		ret = 0; // OK
	}while(0);

	if (ret <= PCAP_ERROR) // all kinds of error code
	{
		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);
		RTE_LOG(INFO, USER1,"Port %d Link Up. Speed %u Mbps - %s\n",
							portid, link.link_speed,
					(link.link_duplex == ETH_LINK_FULL_DUPLEX) ?
						("full-duplex") : ("half-duplex\n"));
	}
	return ret;
}

// 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;
	*is_ours = 0;

	*is_ours = !strncmp(device, "dpdk:", 5);
	if (! *is_ours)
		return NULL;
	//memset will happen
	p = PCAP_CREATE_COMMON(ebuf, struct pcap_dpdk);

	if (p == NULL)
		return NULL;
	p->activate_op = pcap_dpdk_activate;
	return p;
}

int pcap_dpdk_findalldevs(pcap_if_list_t *devlistp, char *ebuf)
{
	int ret=0;
	unsigned int nb_ports = 0;
	char dpdk_name[DPDK_DEV_NAME_MAX];
	char dpdk_desc[DPDK_DEV_DESC_MAX];
	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, 1);
		if (ret < 0)
		{
			// This returns a negative value on an error.
			snprintf(ebuf, PCAP_ERRBUF_SIZE,
			    "Can't look for DPDK devices: %s",
			    dpdk_pre_init_errbuf);
			ret = PCAP_ERROR;
			break;
		}
		if (ret == 0)
		{
			// This means DPDK isn't available on this machine.
			// That just means "don't return any devices".
			break;
		}
		nb_ports = rte_eth_dev_count_avail();
		if (nb_ports == 0)
		{
			// That just means "don't return any devices".
			ret = 0;
			break;
		}
		for (unsigned int i=0; i<nb_ports; i++){
			snprintf(dpdk_name, DPDK_DEV_NAME_MAX-1,
			    "%s%u", DPDK_PREFIX, i);
			// mac addr
			rte_eth_macaddr_get(i, &eth_addr);
			eth_addr_str(&eth_addr,mac_addr,DPDK_MAC_ADDR_SIZE);
			// PCI addr
			rte_eth_dev_get_name_by_port(i,pci_addr);
			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;
			}
		}
	}while(0);
	return ret;
}

#ifdef DPDK_ONLY
/*
 * This libpcap build supports only DPDK, not regular network interfaces.
 */

/*
 * There are no regular interfaces, just DPDK interfaces.
 */
int
pcapint_platform_finddevs(pcap_if_list_t *devlistp _U_, char *errbuf)
{
	return (0);
}

/*
 * Attempts to open a regular interface fail.
 */
pcap_t *
pcapint_create_interface(const char *device, char *errbuf)
{
	snprintf(errbuf, PCAP_ERRBUF_SIZE, PCAP_ENODEV_MESSAGE, "DPDK");
	return NULL;
}

/*
 * Libpcap version string.
 */
const char *
pcap_lib_version(void)
{
	return (PCAP_VERSION_STRING " (DPDK-only)");
}
#endif