1 /* SPDX-License-Identifier: BSD-3-Clause
2 * Copyright(c) 2010-2016 Intel Corporation.
3 * Copyright(c) 2014 6WIND S.A.
10 #include <sys/socket.h>
11 #include <sys/ioctl.h>
14 #if defined(RTE_EXEC_ENV_FREEBSD)
15 #include <sys/sysctl.h>
16 #include <net/if_dl.h>
21 #include <rte_cycles.h>
22 #include <rte_ethdev_driver.h>
23 #include <rte_ethdev_vdev.h>
24 #include <rte_kvargs.h>
25 #include <rte_malloc.h>
27 #include <rte_bus_vdev.h>
28 #include <rte_string_fns.h>
30 #define RTE_ETH_PCAP_SNAPSHOT_LEN 65535
31 #define RTE_ETH_PCAP_SNAPLEN RTE_ETHER_MAX_JUMBO_FRAME_LEN
32 #define RTE_ETH_PCAP_PROMISC 1
33 #define RTE_ETH_PCAP_TIMEOUT -1
35 #define ETH_PCAP_RX_PCAP_ARG "rx_pcap"
36 #define ETH_PCAP_TX_PCAP_ARG "tx_pcap"
37 #define ETH_PCAP_RX_IFACE_ARG "rx_iface"
38 #define ETH_PCAP_RX_IFACE_IN_ARG "rx_iface_in"
39 #define ETH_PCAP_TX_IFACE_ARG "tx_iface"
40 #define ETH_PCAP_IFACE_ARG "iface"
41 #define ETH_PCAP_PHY_MAC_ARG "phy_mac"
42 #define ETH_PCAP_INFINITE_RX_ARG "infinite_rx"
44 #define ETH_PCAP_ARG_MAXLEN 64
46 #define RTE_PMD_PCAP_MAX_QUEUES 16
48 static char errbuf[PCAP_ERRBUF_SIZE];
49 static struct timeval start_time;
50 static uint64_t start_cycles;
52 static uint8_t iface_idx;
55 volatile unsigned long pkts;
56 volatile unsigned long bytes;
57 volatile unsigned long err_pkts;
60 struct pcap_rx_queue {
63 struct rte_mempool *mb_pool;
64 struct queue_stat rx_stat;
66 char type[ETH_PCAP_ARG_MAXLEN];
68 /* Contains pre-generated packets to be looped through */
69 struct rte_ring *pkts;
72 struct pcap_tx_queue {
75 struct queue_stat tx_stat;
77 char type[ETH_PCAP_ARG_MAXLEN];
80 struct pmd_internals {
81 struct pcap_rx_queue rx_queue[RTE_PMD_PCAP_MAX_QUEUES];
82 struct pcap_tx_queue tx_queue[RTE_PMD_PCAP_MAX_QUEUES];
83 char devargs[ETH_PCAP_ARG_MAXLEN];
84 struct rte_ether_addr eth_addr;
88 unsigned int infinite_rx;
91 struct pmd_process_private {
92 pcap_t *rx_pcap[RTE_PMD_PCAP_MAX_QUEUES];
93 pcap_t *tx_pcap[RTE_PMD_PCAP_MAX_QUEUES];
94 pcap_dumper_t *tx_dumper[RTE_PMD_PCAP_MAX_QUEUES];
98 unsigned int num_of_queue;
99 struct devargs_queue {
100 pcap_dumper_t *dumper;
104 } queue[RTE_PMD_PCAP_MAX_QUEUES];
108 struct pmd_devargs_all {
109 struct pmd_devargs rx_queues;
110 struct pmd_devargs tx_queues;
112 unsigned int is_tx_pcap;
113 unsigned int is_tx_iface;
114 unsigned int is_rx_pcap;
115 unsigned int is_rx_iface;
116 unsigned int infinite_rx;
119 static const char *valid_arguments[] = {
120 ETH_PCAP_RX_PCAP_ARG,
121 ETH_PCAP_TX_PCAP_ARG,
122 ETH_PCAP_RX_IFACE_ARG,
123 ETH_PCAP_RX_IFACE_IN_ARG,
124 ETH_PCAP_TX_IFACE_ARG,
126 ETH_PCAP_PHY_MAC_ARG,
127 ETH_PCAP_INFINITE_RX_ARG,
131 static struct rte_eth_link pmd_link = {
132 .link_speed = ETH_SPEED_NUM_10G,
133 .link_duplex = ETH_LINK_FULL_DUPLEX,
134 .link_status = ETH_LINK_DOWN,
135 .link_autoneg = ETH_LINK_FIXED,
138 RTE_LOG_REGISTER(eth_pcap_logtype, pmd.net.pcap, NOTICE);
140 #define PMD_LOG(level, fmt, args...) \
141 rte_log(RTE_LOG_ ## level, eth_pcap_logtype, \
142 "%s(): " fmt "\n", __func__, ##args)
145 eth_pcap_rx_jumbo(struct rte_mempool *mb_pool, struct rte_mbuf *mbuf,
146 const u_char *data, uint16_t data_len)
148 /* Copy the first segment. */
149 uint16_t len = rte_pktmbuf_tailroom(mbuf);
150 struct rte_mbuf *m = mbuf;
152 rte_memcpy(rte_pktmbuf_append(mbuf, len), data, len);
156 while (data_len > 0) {
157 /* Allocate next mbuf and point to that. */
158 m->next = rte_pktmbuf_alloc(mb_pool);
160 if (unlikely(!m->next))
165 /* Headroom is not needed in chained mbufs. */
166 rte_pktmbuf_prepend(m, rte_pktmbuf_headroom(m));
170 /* Copy next segment. */
171 len = RTE_MIN(rte_pktmbuf_tailroom(m), data_len);
172 rte_memcpy(rte_pktmbuf_append(m, len), data, len);
179 return mbuf->nb_segs;
183 eth_pcap_rx_infinite(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
186 struct pcap_rx_queue *pcap_q = queue;
187 uint32_t rx_bytes = 0;
189 if (unlikely(nb_pkts == 0))
192 if (rte_pktmbuf_alloc_bulk(pcap_q->mb_pool, bufs, nb_pkts) != 0)
195 for (i = 0; i < nb_pkts; i++) {
196 struct rte_mbuf *pcap_buf;
197 int err = rte_ring_dequeue(pcap_q->pkts, (void **)&pcap_buf);
201 rte_memcpy(rte_pktmbuf_mtod(bufs[i], void *),
202 rte_pktmbuf_mtod(pcap_buf, void *),
204 bufs[i]->data_len = pcap_buf->data_len;
205 bufs[i]->pkt_len = pcap_buf->pkt_len;
206 bufs[i]->port = pcap_q->port_id;
207 rx_bytes += pcap_buf->data_len;
209 /* Enqueue packet back on ring to allow infinite rx. */
210 rte_ring_enqueue(pcap_q->pkts, pcap_buf);
213 pcap_q->rx_stat.pkts += i;
214 pcap_q->rx_stat.bytes += rx_bytes;
220 eth_pcap_rx(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
223 struct pcap_pkthdr header;
224 struct pmd_process_private *pp;
225 const u_char *packet;
226 struct rte_mbuf *mbuf;
227 struct pcap_rx_queue *pcap_q = queue;
229 uint32_t rx_bytes = 0;
232 pp = rte_eth_devices[pcap_q->port_id].process_private;
233 pcap = pp->rx_pcap[pcap_q->queue_id];
235 if (unlikely(pcap == NULL || nb_pkts == 0))
238 /* Reads the given number of packets from the pcap file one by one
239 * and copies the packet data into a newly allocated mbuf to return.
241 for (i = 0; i < nb_pkts; i++) {
242 /* Get the next PCAP packet */
243 packet = pcap_next(pcap, &header);
244 if (unlikely(packet == NULL))
247 mbuf = rte_pktmbuf_alloc(pcap_q->mb_pool);
248 if (unlikely(mbuf == NULL))
251 if (header.caplen <= rte_pktmbuf_tailroom(mbuf)) {
252 /* pcap packet will fit in the mbuf, can copy it */
253 rte_memcpy(rte_pktmbuf_mtod(mbuf, void *), packet,
255 mbuf->data_len = (uint16_t)header.caplen;
257 /* Try read jumbo frame into multi mbufs. */
258 if (unlikely(eth_pcap_rx_jumbo(pcap_q->mb_pool,
261 header.caplen) == -1)) {
262 rte_pktmbuf_free(mbuf);
267 mbuf->pkt_len = (uint16_t)header.caplen;
268 mbuf->timestamp = (uint64_t)header.ts.tv_sec * 1000000
270 mbuf->ol_flags |= PKT_RX_TIMESTAMP;
271 mbuf->port = pcap_q->port_id;
274 rx_bytes += header.caplen;
276 pcap_q->rx_stat.pkts += num_rx;
277 pcap_q->rx_stat.bytes += rx_bytes;
283 eth_null_rx(void *queue __rte_unused,
284 struct rte_mbuf **bufs __rte_unused,
285 uint16_t nb_pkts __rte_unused)
290 #define NSEC_PER_SEC 1000000000L
293 calculate_timestamp(struct timeval *ts) {
295 struct timeval cur_time;
297 cycles = rte_get_timer_cycles() - start_cycles;
298 cur_time.tv_sec = cycles / hz;
299 cur_time.tv_usec = (cycles % hz) * NSEC_PER_SEC / hz;
301 ts->tv_sec = start_time.tv_sec + cur_time.tv_sec;
302 ts->tv_usec = start_time.tv_usec + cur_time.tv_usec;
303 if (ts->tv_usec >= NSEC_PER_SEC) {
304 ts->tv_usec -= NSEC_PER_SEC;
310 * Callback to handle writing packets to a pcap file.
313 eth_pcap_tx_dumper(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
316 struct rte_mbuf *mbuf;
317 struct pmd_process_private *pp;
318 struct pcap_tx_queue *dumper_q = queue;
320 uint32_t tx_bytes = 0;
321 struct pcap_pkthdr header;
322 pcap_dumper_t *dumper;
323 unsigned char temp_data[RTE_ETH_PCAP_SNAPLEN];
326 pp = rte_eth_devices[dumper_q->port_id].process_private;
327 dumper = pp->tx_dumper[dumper_q->queue_id];
329 if (dumper == NULL || nb_pkts == 0)
332 /* writes the nb_pkts packets to the previously opened pcap file
334 for (i = 0; i < nb_pkts; i++) {
336 len = caplen = rte_pktmbuf_pkt_len(mbuf);
337 if (unlikely(!rte_pktmbuf_is_contiguous(mbuf) &&
338 len > sizeof(temp_data))) {
339 caplen = sizeof(temp_data);
342 calculate_timestamp(&header.ts);
344 header.caplen = caplen;
345 /* rte_pktmbuf_read() returns a pointer to the data directly
346 * in the mbuf (when the mbuf is contiguous) or, otherwise,
347 * a pointer to temp_data after copying into it.
349 pcap_dump((u_char *)dumper, &header,
350 rte_pktmbuf_read(mbuf, 0, caplen, temp_data));
354 rte_pktmbuf_free(mbuf);
358 * Since there's no place to hook a callback when the forwarding
359 * process stops and to make sure the pcap file is actually written,
360 * we flush the pcap dumper within each burst.
362 pcap_dump_flush(dumper);
363 dumper_q->tx_stat.pkts += num_tx;
364 dumper_q->tx_stat.bytes += tx_bytes;
365 dumper_q->tx_stat.err_pkts += nb_pkts - num_tx;
371 * Callback to handle dropping packets in the infinite rx case.
374 eth_tx_drop(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
377 uint32_t tx_bytes = 0;
378 struct pcap_tx_queue *tx_queue = queue;
380 if (unlikely(nb_pkts == 0))
383 for (i = 0; i < nb_pkts; i++) {
384 tx_bytes += bufs[i]->data_len;
385 rte_pktmbuf_free(bufs[i]);
388 tx_queue->tx_stat.pkts += nb_pkts;
389 tx_queue->tx_stat.bytes += tx_bytes;
395 * Callback to handle sending packets through a real NIC.
398 eth_pcap_tx(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
402 struct rte_mbuf *mbuf;
403 struct pmd_process_private *pp;
404 struct pcap_tx_queue *tx_queue = queue;
406 uint32_t tx_bytes = 0;
408 unsigned char temp_data[RTE_ETH_PCAP_SNAPLEN];
411 pp = rte_eth_devices[tx_queue->port_id].process_private;
412 pcap = pp->tx_pcap[tx_queue->queue_id];
414 if (unlikely(nb_pkts == 0 || pcap == NULL))
417 for (i = 0; i < nb_pkts; i++) {
419 len = rte_pktmbuf_pkt_len(mbuf);
420 if (unlikely(!rte_pktmbuf_is_contiguous(mbuf) &&
421 len > sizeof(temp_data))) {
423 "Dropping multi segment PCAP packet. Size (%zd) > max size (%zd).",
424 len, sizeof(temp_data));
425 rte_pktmbuf_free(mbuf);
429 /* rte_pktmbuf_read() returns a pointer to the data directly
430 * in the mbuf (when the mbuf is contiguous) or, otherwise,
431 * a pointer to temp_data after copying into it.
433 ret = pcap_sendpacket(pcap,
434 rte_pktmbuf_read(mbuf, 0, len, temp_data), len);
435 if (unlikely(ret != 0))
439 rte_pktmbuf_free(mbuf);
442 tx_queue->tx_stat.pkts += num_tx;
443 tx_queue->tx_stat.bytes += tx_bytes;
444 tx_queue->tx_stat.err_pkts += i - num_tx;
450 * pcap_open_live wrapper function
453 open_iface_live(const char *iface, pcap_t **pcap) {
454 *pcap = pcap_open_live(iface, RTE_ETH_PCAP_SNAPLEN,
455 RTE_ETH_PCAP_PROMISC, RTE_ETH_PCAP_TIMEOUT, errbuf);
458 PMD_LOG(ERR, "Couldn't open %s: %s", iface, errbuf);
466 open_single_iface(const char *iface, pcap_t **pcap)
468 if (open_iface_live(iface, pcap) < 0) {
469 PMD_LOG(ERR, "Couldn't open interface %s", iface);
477 open_single_tx_pcap(const char *pcap_filename, pcap_dumper_t **dumper)
482 * We need to create a dummy empty pcap_t to use it
483 * with pcap_dump_open(). We create big enough an Ethernet
486 tx_pcap = pcap_open_dead_with_tstamp_precision(DLT_EN10MB,
487 RTE_ETH_PCAP_SNAPSHOT_LEN, PCAP_TSTAMP_PRECISION_NANO);
488 if (tx_pcap == NULL) {
489 PMD_LOG(ERR, "Couldn't create dead pcap");
493 /* The dumper is created using the previous pcap_t reference */
494 *dumper = pcap_dump_open(tx_pcap, pcap_filename);
495 if (*dumper == NULL) {
497 PMD_LOG(ERR, "Couldn't open %s for writing.",
507 open_single_rx_pcap(const char *pcap_filename, pcap_t **pcap)
509 *pcap = pcap_open_offline(pcap_filename, errbuf);
511 PMD_LOG(ERR, "Couldn't open %s: %s", pcap_filename,
520 count_packets_in_pcap(pcap_t **pcap, struct pcap_rx_queue *pcap_q)
522 const u_char *packet;
523 struct pcap_pkthdr header;
524 uint64_t pcap_pkt_count = 0;
526 while ((packet = pcap_next(*pcap, &header)))
529 /* The pcap is reopened so it can be used as normal later. */
532 open_single_rx_pcap(pcap_q->name, pcap);
534 return pcap_pkt_count;
538 eth_dev_start(struct rte_eth_dev *dev)
541 struct pmd_internals *internals = dev->data->dev_private;
542 struct pmd_process_private *pp = dev->process_private;
543 struct pcap_tx_queue *tx;
544 struct pcap_rx_queue *rx;
546 /* Special iface case. Single pcap is open and shared between tx/rx. */
547 if (internals->single_iface) {
548 tx = &internals->tx_queue[0];
549 rx = &internals->rx_queue[0];
551 if (!pp->tx_pcap[0] &&
552 strcmp(tx->type, ETH_PCAP_IFACE_ARG) == 0) {
553 if (open_single_iface(tx->name, &pp->tx_pcap[0]) < 0)
555 pp->rx_pcap[0] = pp->tx_pcap[0];
561 /* If not open already, open tx pcaps/dumpers */
562 for (i = 0; i < dev->data->nb_tx_queues; i++) {
563 tx = &internals->tx_queue[i];
565 if (!pp->tx_dumper[i] &&
566 strcmp(tx->type, ETH_PCAP_TX_PCAP_ARG) == 0) {
567 if (open_single_tx_pcap(tx->name,
568 &pp->tx_dumper[i]) < 0)
570 } else if (!pp->tx_pcap[i] &&
571 strcmp(tx->type, ETH_PCAP_TX_IFACE_ARG) == 0) {
572 if (open_single_iface(tx->name, &pp->tx_pcap[i]) < 0)
577 /* If not open already, open rx pcaps */
578 for (i = 0; i < dev->data->nb_rx_queues; i++) {
579 rx = &internals->rx_queue[i];
581 if (pp->rx_pcap[i] != NULL)
584 if (strcmp(rx->type, ETH_PCAP_RX_PCAP_ARG) == 0) {
585 if (open_single_rx_pcap(rx->name, &pp->rx_pcap[i]) < 0)
587 } else if (strcmp(rx->type, ETH_PCAP_RX_IFACE_ARG) == 0) {
588 if (open_single_iface(rx->name, &pp->rx_pcap[i]) < 0)
594 for (i = 0; i < dev->data->nb_rx_queues; i++)
595 dev->data->rx_queue_state[i] = RTE_ETH_QUEUE_STATE_STARTED;
597 for (i = 0; i < dev->data->nb_tx_queues; i++)
598 dev->data->tx_queue_state[i] = RTE_ETH_QUEUE_STATE_STARTED;
600 dev->data->dev_link.link_status = ETH_LINK_UP;
606 * This function gets called when the current port gets stopped.
607 * Is the only place for us to close all the tx streams dumpers.
608 * If not called the dumpers will be flushed within each tx burst.
611 eth_dev_stop(struct rte_eth_dev *dev)
614 struct pmd_internals *internals = dev->data->dev_private;
615 struct pmd_process_private *pp = dev->process_private;
617 /* Special iface case. Single pcap is open and shared between tx/rx. */
618 if (internals->single_iface) {
619 pcap_close(pp->tx_pcap[0]);
620 pp->tx_pcap[0] = NULL;
621 pp->rx_pcap[0] = NULL;
625 for (i = 0; i < dev->data->nb_tx_queues; i++) {
626 if (pp->tx_dumper[i] != NULL) {
627 pcap_dump_close(pp->tx_dumper[i]);
628 pp->tx_dumper[i] = NULL;
631 if (pp->tx_pcap[i] != NULL) {
632 pcap_close(pp->tx_pcap[i]);
633 pp->tx_pcap[i] = NULL;
637 for (i = 0; i < dev->data->nb_rx_queues; i++) {
638 if (pp->rx_pcap[i] != NULL) {
639 pcap_close(pp->rx_pcap[i]);
640 pp->rx_pcap[i] = NULL;
645 for (i = 0; i < dev->data->nb_rx_queues; i++)
646 dev->data->rx_queue_state[i] = RTE_ETH_QUEUE_STATE_STOPPED;
648 for (i = 0; i < dev->data->nb_tx_queues; i++)
649 dev->data->tx_queue_state[i] = RTE_ETH_QUEUE_STATE_STOPPED;
651 dev->data->dev_link.link_status = ETH_LINK_DOWN;
655 eth_dev_configure(struct rte_eth_dev *dev __rte_unused)
661 eth_dev_info(struct rte_eth_dev *dev,
662 struct rte_eth_dev_info *dev_info)
664 struct pmd_internals *internals = dev->data->dev_private;
666 dev_info->if_index = internals->if_index;
667 dev_info->max_mac_addrs = 1;
668 dev_info->max_rx_pktlen = (uint32_t) -1;
669 dev_info->max_rx_queues = dev->data->nb_rx_queues;
670 dev_info->max_tx_queues = dev->data->nb_tx_queues;
671 dev_info->min_rx_bufsize = 0;
677 eth_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
680 unsigned long rx_packets_total = 0, rx_bytes_total = 0;
681 unsigned long tx_packets_total = 0, tx_bytes_total = 0;
682 unsigned long tx_packets_err_total = 0;
683 const struct pmd_internals *internal = dev->data->dev_private;
685 for (i = 0; i < RTE_ETHDEV_QUEUE_STAT_CNTRS &&
686 i < dev->data->nb_rx_queues; i++) {
687 stats->q_ipackets[i] = internal->rx_queue[i].rx_stat.pkts;
688 stats->q_ibytes[i] = internal->rx_queue[i].rx_stat.bytes;
689 rx_packets_total += stats->q_ipackets[i];
690 rx_bytes_total += stats->q_ibytes[i];
693 for (i = 0; i < RTE_ETHDEV_QUEUE_STAT_CNTRS &&
694 i < dev->data->nb_tx_queues; i++) {
695 stats->q_opackets[i] = internal->tx_queue[i].tx_stat.pkts;
696 stats->q_obytes[i] = internal->tx_queue[i].tx_stat.bytes;
697 tx_packets_total += stats->q_opackets[i];
698 tx_bytes_total += stats->q_obytes[i];
699 tx_packets_err_total += internal->tx_queue[i].tx_stat.err_pkts;
702 stats->ipackets = rx_packets_total;
703 stats->ibytes = rx_bytes_total;
704 stats->opackets = tx_packets_total;
705 stats->obytes = tx_bytes_total;
706 stats->oerrors = tx_packets_err_total;
712 eth_stats_reset(struct rte_eth_dev *dev)
715 struct pmd_internals *internal = dev->data->dev_private;
717 for (i = 0; i < dev->data->nb_rx_queues; i++) {
718 internal->rx_queue[i].rx_stat.pkts = 0;
719 internal->rx_queue[i].rx_stat.bytes = 0;
722 for (i = 0; i < dev->data->nb_tx_queues; i++) {
723 internal->tx_queue[i].tx_stat.pkts = 0;
724 internal->tx_queue[i].tx_stat.bytes = 0;
725 internal->tx_queue[i].tx_stat.err_pkts = 0;
732 eth_dev_close(struct rte_eth_dev *dev)
735 struct pmd_internals *internals = dev->data->dev_private;
737 /* Device wide flag, but cleanup must be performed per queue. */
738 if (internals->infinite_rx) {
739 for (i = 0; i < dev->data->nb_rx_queues; i++) {
740 struct pcap_rx_queue *pcap_q = &internals->rx_queue[i];
741 struct rte_mbuf *pcap_buf;
743 while (!rte_ring_dequeue(pcap_q->pkts,
745 rte_pktmbuf_free(pcap_buf);
747 rte_ring_free(pcap_q->pkts);
755 eth_queue_release(void *q __rte_unused)
760 eth_link_update(struct rte_eth_dev *dev __rte_unused,
761 int wait_to_complete __rte_unused)
767 eth_rx_queue_setup(struct rte_eth_dev *dev,
768 uint16_t rx_queue_id,
769 uint16_t nb_rx_desc __rte_unused,
770 unsigned int socket_id __rte_unused,
771 const struct rte_eth_rxconf *rx_conf __rte_unused,
772 struct rte_mempool *mb_pool)
774 struct pmd_internals *internals = dev->data->dev_private;
775 struct pcap_rx_queue *pcap_q = &internals->rx_queue[rx_queue_id];
777 pcap_q->mb_pool = mb_pool;
778 pcap_q->port_id = dev->data->port_id;
779 pcap_q->queue_id = rx_queue_id;
780 dev->data->rx_queues[rx_queue_id] = pcap_q;
782 if (internals->infinite_rx) {
783 struct pmd_process_private *pp;
784 char ring_name[NAME_MAX];
785 static uint32_t ring_number;
786 uint64_t pcap_pkt_count = 0;
787 struct rte_mbuf *bufs[1];
790 pp = rte_eth_devices[pcap_q->port_id].process_private;
791 pcap = &pp->rx_pcap[pcap_q->queue_id];
793 if (unlikely(*pcap == NULL))
796 pcap_pkt_count = count_packets_in_pcap(pcap, pcap_q);
798 snprintf(ring_name, sizeof(ring_name), "PCAP_RING%" PRIu16,
801 pcap_q->pkts = rte_ring_create(ring_name,
802 rte_align64pow2(pcap_pkt_count + 1), 0,
803 RING_F_SP_ENQ | RING_F_SC_DEQ);
808 /* Fill ring with packets from PCAP file one by one. */
809 while (eth_pcap_rx(pcap_q, bufs, 1)) {
810 /* Check for multiseg mbufs. */
811 if (bufs[0]->nb_segs != 1) {
812 rte_pktmbuf_free(*bufs);
814 while (!rte_ring_dequeue(pcap_q->pkts,
816 rte_pktmbuf_free(*bufs);
818 rte_ring_free(pcap_q->pkts);
819 PMD_LOG(ERR, "Multiseg mbufs are not supported in infinite_rx "
824 rte_ring_enqueue_bulk(pcap_q->pkts,
825 (void * const *)bufs, 1, NULL);
828 * Reset the stats for this queue since eth_pcap_rx calls above
829 * didn't result in the application receiving packets.
831 pcap_q->rx_stat.pkts = 0;
832 pcap_q->rx_stat.bytes = 0;
839 eth_tx_queue_setup(struct rte_eth_dev *dev,
840 uint16_t tx_queue_id,
841 uint16_t nb_tx_desc __rte_unused,
842 unsigned int socket_id __rte_unused,
843 const struct rte_eth_txconf *tx_conf __rte_unused)
845 struct pmd_internals *internals = dev->data->dev_private;
846 struct pcap_tx_queue *pcap_q = &internals->tx_queue[tx_queue_id];
848 pcap_q->port_id = dev->data->port_id;
849 pcap_q->queue_id = tx_queue_id;
850 dev->data->tx_queues[tx_queue_id] = pcap_q;
856 eth_rx_queue_start(struct rte_eth_dev *dev, uint16_t rx_queue_id)
858 dev->data->rx_queue_state[rx_queue_id] = RTE_ETH_QUEUE_STATE_STARTED;
864 eth_tx_queue_start(struct rte_eth_dev *dev, uint16_t tx_queue_id)
866 dev->data->tx_queue_state[tx_queue_id] = RTE_ETH_QUEUE_STATE_STARTED;
872 eth_rx_queue_stop(struct rte_eth_dev *dev, uint16_t rx_queue_id)
874 dev->data->rx_queue_state[rx_queue_id] = RTE_ETH_QUEUE_STATE_STOPPED;
880 eth_tx_queue_stop(struct rte_eth_dev *dev, uint16_t tx_queue_id)
882 dev->data->tx_queue_state[tx_queue_id] = RTE_ETH_QUEUE_STATE_STOPPED;
887 static const struct eth_dev_ops ops = {
888 .dev_start = eth_dev_start,
889 .dev_stop = eth_dev_stop,
890 .dev_close = eth_dev_close,
891 .dev_configure = eth_dev_configure,
892 .dev_infos_get = eth_dev_info,
893 .rx_queue_setup = eth_rx_queue_setup,
894 .tx_queue_setup = eth_tx_queue_setup,
895 .rx_queue_start = eth_rx_queue_start,
896 .tx_queue_start = eth_tx_queue_start,
897 .rx_queue_stop = eth_rx_queue_stop,
898 .tx_queue_stop = eth_tx_queue_stop,
899 .rx_queue_release = eth_queue_release,
900 .tx_queue_release = eth_queue_release,
901 .link_update = eth_link_update,
902 .stats_get = eth_stats_get,
903 .stats_reset = eth_stats_reset,
907 add_queue(struct pmd_devargs *pmd, const char *name, const char *type,
908 pcap_t *pcap, pcap_dumper_t *dumper)
910 if (pmd->num_of_queue >= RTE_PMD_PCAP_MAX_QUEUES)
913 pmd->queue[pmd->num_of_queue].pcap = pcap;
915 pmd->queue[pmd->num_of_queue].dumper = dumper;
916 pmd->queue[pmd->num_of_queue].name = name;
917 pmd->queue[pmd->num_of_queue].type = type;
923 * Function handler that opens the pcap file for reading a stores a
924 * reference of it for use it later on.
927 open_rx_pcap(const char *key, const char *value, void *extra_args)
929 const char *pcap_filename = value;
930 struct pmd_devargs *rx = extra_args;
933 if (open_single_rx_pcap(pcap_filename, &pcap) < 0)
936 if (add_queue(rx, pcap_filename, key, pcap, NULL) < 0) {
945 * Opens a pcap file for writing and stores a reference to it
946 * for use it later on.
949 open_tx_pcap(const char *key, const char *value, void *extra_args)
951 const char *pcap_filename = value;
952 struct pmd_devargs *dumpers = extra_args;
953 pcap_dumper_t *dumper;
955 if (open_single_tx_pcap(pcap_filename, &dumper) < 0)
958 if (add_queue(dumpers, pcap_filename, key, NULL, dumper) < 0) {
959 pcap_dump_close(dumper);
967 * Opens an interface for reading and writing
970 open_rx_tx_iface(const char *key, const char *value, void *extra_args)
972 const char *iface = value;
973 struct pmd_devargs *tx = extra_args;
976 if (open_single_iface(iface, &pcap) < 0)
979 tx->queue[0].pcap = pcap;
980 tx->queue[0].name = iface;
981 tx->queue[0].type = key;
987 set_iface_direction(const char *iface, pcap_t *pcap,
988 pcap_direction_t direction)
990 const char *direction_str = (direction == PCAP_D_IN) ? "IN" : "OUT";
991 if (pcap_setdirection(pcap, direction) < 0) {
992 PMD_LOG(ERR, "Setting %s pcap direction %s failed - %s\n",
993 iface, direction_str, pcap_geterr(pcap));
996 PMD_LOG(INFO, "Setting %s pcap direction %s\n",
997 iface, direction_str);
1002 open_iface(const char *key, const char *value, void *extra_args)
1004 const char *iface = value;
1005 struct pmd_devargs *pmd = extra_args;
1006 pcap_t *pcap = NULL;
1008 if (open_single_iface(iface, &pcap) < 0)
1010 if (add_queue(pmd, iface, key, pcap, NULL) < 0) {
1019 * Opens a NIC for reading packets from it
1022 open_rx_iface(const char *key, const char *value, void *extra_args)
1024 int ret = open_iface(key, value, extra_args);
1027 if (strcmp(key, ETH_PCAP_RX_IFACE_IN_ARG) == 0) {
1028 struct pmd_devargs *pmd = extra_args;
1029 unsigned int qid = pmd->num_of_queue - 1;
1031 set_iface_direction(pmd->queue[qid].name,
1032 pmd->queue[qid].pcap,
1040 rx_iface_args_process(const char *key, const char *value, void *extra_args)
1042 if (strcmp(key, ETH_PCAP_RX_IFACE_ARG) == 0 ||
1043 strcmp(key, ETH_PCAP_RX_IFACE_IN_ARG) == 0)
1044 return open_rx_iface(key, value, extra_args);
1050 * Opens a NIC for writing packets to it
1053 open_tx_iface(const char *key, const char *value, void *extra_args)
1055 return open_iface(key, value, extra_args);
1059 select_phy_mac(const char *key __rte_unused, const char *value,
1063 const int phy_mac = atoi(value);
1064 int *enable_phy_mac = extra_args;
1067 *enable_phy_mac = 1;
1073 get_infinite_rx_arg(const char *key __rte_unused,
1074 const char *value, void *extra_args)
1077 const int infinite_rx = atoi(value);
1078 int *enable_infinite_rx = extra_args;
1080 if (infinite_rx > 0)
1081 *enable_infinite_rx = 1;
1087 pmd_init_internals(struct rte_vdev_device *vdev,
1088 const unsigned int nb_rx_queues,
1089 const unsigned int nb_tx_queues,
1090 struct pmd_internals **internals,
1091 struct rte_eth_dev **eth_dev)
1093 struct rte_eth_dev_data *data;
1094 struct pmd_process_private *pp;
1095 unsigned int numa_node = vdev->device.numa_node;
1097 PMD_LOG(INFO, "Creating pcap-backed ethdev on numa socket %d",
1100 pp = (struct pmd_process_private *)
1101 rte_zmalloc(NULL, sizeof(struct pmd_process_private),
1102 RTE_CACHE_LINE_SIZE);
1106 "Failed to allocate memory for process private");
1110 /* reserve an ethdev entry */
1111 *eth_dev = rte_eth_vdev_allocate(vdev, sizeof(**internals));
1116 (*eth_dev)->process_private = pp;
1117 /* now put it all together
1118 * - store queue data in internals,
1119 * - store numa_node info in eth_dev
1120 * - point eth_dev_data to internals
1121 * - and point eth_dev structure to new eth_dev_data structure
1123 *internals = (*eth_dev)->data->dev_private;
1125 * Interface MAC = 02:70:63:61:70:<iface_idx>
1126 * derived from: 'locally administered':'p':'c':'a':'p':'iface_idx'
1127 * where the middle 4 characters are converted to hex.
1129 (*internals)->eth_addr = (struct rte_ether_addr) {
1130 .addr_bytes = { 0x02, 0x70, 0x63, 0x61, 0x70, iface_idx++ }
1132 (*internals)->phy_mac = 0;
1133 data = (*eth_dev)->data;
1134 data->nb_rx_queues = (uint16_t)nb_rx_queues;
1135 data->nb_tx_queues = (uint16_t)nb_tx_queues;
1136 data->dev_link = pmd_link;
1137 data->mac_addrs = &(*internals)->eth_addr;
1138 data->promiscuous = 1;
1139 data->all_multicast = 1;
1142 * NOTE: we'll replace the data element, of originally allocated
1143 * eth_dev so the rings are local per-process
1145 (*eth_dev)->dev_ops = &ops;
1147 strlcpy((*internals)->devargs, rte_vdev_device_args(vdev),
1148 ETH_PCAP_ARG_MAXLEN);
1154 eth_pcap_update_mac(const char *if_name, struct rte_eth_dev *eth_dev,
1155 const unsigned int numa_node)
1157 #if defined(RTE_EXEC_ENV_LINUX)
1160 int if_fd = socket(AF_INET, SOCK_DGRAM, 0);
1165 rte_strscpy(ifr.ifr_name, if_name, sizeof(ifr.ifr_name));
1166 if (ioctl(if_fd, SIOCGIFHWADDR, &ifr)) {
1171 mac_addrs = rte_zmalloc_socket(NULL, RTE_ETHER_ADDR_LEN, 0, numa_node);
1177 PMD_LOG(INFO, "Setting phy MAC for %s", if_name);
1178 eth_dev->data->mac_addrs = mac_addrs;
1179 rte_memcpy(eth_dev->data->mac_addrs[0].addr_bytes,
1180 ifr.ifr_hwaddr.sa_data, RTE_ETHER_ADDR_LEN);
1186 #elif defined(RTE_EXEC_ENV_FREEBSD)
1188 struct if_msghdr *ifm;
1189 struct sockaddr_dl *sdl;
1198 mib[4] = NET_RT_IFLIST;
1199 mib[5] = if_nametoindex(if_name);
1201 if (sysctl(mib, 6, NULL, &len, NULL, 0) < 0)
1207 buf = rte_malloc(NULL, len, 0);
1211 if (sysctl(mib, 6, buf, &len, NULL, 0) < 0) {
1215 ifm = (struct if_msghdr *)buf;
1216 sdl = (struct sockaddr_dl *)(ifm + 1);
1218 mac_addrs = rte_zmalloc_socket(NULL, RTE_ETHER_ADDR_LEN, 0, numa_node);
1224 PMD_LOG(INFO, "Setting phy MAC for %s", if_name);
1225 eth_dev->data->mac_addrs = mac_addrs;
1226 rte_memcpy(eth_dev->data->mac_addrs[0].addr_bytes,
1227 LLADDR(sdl), RTE_ETHER_ADDR_LEN);
1238 eth_from_pcaps_common(struct rte_vdev_device *vdev,
1239 struct pmd_devargs_all *devargs_all,
1240 struct pmd_internals **internals, struct rte_eth_dev **eth_dev)
1242 struct pmd_process_private *pp;
1243 struct pmd_devargs *rx_queues = &devargs_all->rx_queues;
1244 struct pmd_devargs *tx_queues = &devargs_all->tx_queues;
1245 const unsigned int nb_rx_queues = rx_queues->num_of_queue;
1246 const unsigned int nb_tx_queues = tx_queues->num_of_queue;
1249 if (pmd_init_internals(vdev, nb_rx_queues, nb_tx_queues, internals,
1253 pp = (*eth_dev)->process_private;
1254 for (i = 0; i < nb_rx_queues; i++) {
1255 struct pcap_rx_queue *rx = &(*internals)->rx_queue[i];
1256 struct devargs_queue *queue = &rx_queues->queue[i];
1258 pp->rx_pcap[i] = queue->pcap;
1259 strlcpy(rx->name, queue->name, sizeof(rx->name));
1260 strlcpy(rx->type, queue->type, sizeof(rx->type));
1263 for (i = 0; i < nb_tx_queues; i++) {
1264 struct pcap_tx_queue *tx = &(*internals)->tx_queue[i];
1265 struct devargs_queue *queue = &tx_queues->queue[i];
1267 pp->tx_dumper[i] = queue->dumper;
1268 pp->tx_pcap[i] = queue->pcap;
1269 strlcpy(tx->name, queue->name, sizeof(tx->name));
1270 strlcpy(tx->type, queue->type, sizeof(tx->type));
1277 eth_from_pcaps(struct rte_vdev_device *vdev,
1278 struct pmd_devargs_all *devargs_all)
1280 struct pmd_internals *internals = NULL;
1281 struct rte_eth_dev *eth_dev = NULL;
1282 struct pmd_devargs *rx_queues = &devargs_all->rx_queues;
1283 int single_iface = devargs_all->single_iface;
1284 unsigned int infinite_rx = devargs_all->infinite_rx;
1287 ret = eth_from_pcaps_common(vdev, devargs_all, &internals, ð_dev);
1292 /* store weather we are using a single interface for rx/tx or not */
1293 internals->single_iface = single_iface;
1296 internals->if_index = if_nametoindex(rx_queues->queue[0].name);
1298 /* phy_mac arg is applied only only if "iface" devarg is provided */
1299 if (rx_queues->phy_mac) {
1300 int ret = eth_pcap_update_mac(rx_queues->queue[0].name,
1301 eth_dev, vdev->device.numa_node);
1303 internals->phy_mac = 1;
1307 internals->infinite_rx = infinite_rx;
1308 /* Assign rx ops. */
1310 eth_dev->rx_pkt_burst = eth_pcap_rx_infinite;
1311 else if (devargs_all->is_rx_pcap || devargs_all->is_rx_iface ||
1313 eth_dev->rx_pkt_burst = eth_pcap_rx;
1315 eth_dev->rx_pkt_burst = eth_null_rx;
1317 /* Assign tx ops. */
1318 if (devargs_all->is_tx_pcap)
1319 eth_dev->tx_pkt_burst = eth_pcap_tx_dumper;
1320 else if (devargs_all->is_tx_iface || single_iface)
1321 eth_dev->tx_pkt_burst = eth_pcap_tx;
1323 eth_dev->tx_pkt_burst = eth_tx_drop;
1325 rte_eth_dev_probing_finish(eth_dev);
1330 pmd_pcap_probe(struct rte_vdev_device *dev)
1333 struct rte_kvargs *kvlist;
1334 struct pmd_devargs pcaps = {0};
1335 struct pmd_devargs dumpers = {0};
1336 struct rte_eth_dev *eth_dev = NULL;
1337 struct pmd_internals *internal;
1340 struct pmd_devargs_all devargs_all = {
1347 name = rte_vdev_device_name(dev);
1348 PMD_LOG(INFO, "Initializing pmd_pcap for %s", name);
1350 gettimeofday(&start_time, NULL);
1351 start_cycles = rte_get_timer_cycles();
1352 hz = rte_get_timer_hz();
1354 if (rte_eal_process_type() == RTE_PROC_SECONDARY) {
1355 eth_dev = rte_eth_dev_attach_secondary(name);
1357 PMD_LOG(ERR, "Failed to probe %s", name);
1361 internal = eth_dev->data->dev_private;
1363 kvlist = rte_kvargs_parse(internal->devargs, valid_arguments);
1367 kvlist = rte_kvargs_parse(rte_vdev_device_args(dev),
1374 * If iface argument is passed we open the NICs and use them for
1377 if (rte_kvargs_count(kvlist, ETH_PCAP_IFACE_ARG) == 1) {
1379 ret = rte_kvargs_process(kvlist, ETH_PCAP_IFACE_ARG,
1380 &open_rx_tx_iface, &pcaps);
1384 dumpers.queue[0] = pcaps.queue[0];
1386 ret = rte_kvargs_process(kvlist, ETH_PCAP_PHY_MAC_ARG,
1387 &select_phy_mac, &pcaps.phy_mac);
1391 dumpers.phy_mac = pcaps.phy_mac;
1393 devargs_all.single_iface = 1;
1394 pcaps.num_of_queue = 1;
1395 dumpers.num_of_queue = 1;
1401 * We check whether we want to open a RX stream from a real NIC, a
1402 * pcap file or open a dummy RX stream
1404 devargs_all.is_rx_pcap =
1405 rte_kvargs_count(kvlist, ETH_PCAP_RX_PCAP_ARG) ? 1 : 0;
1406 devargs_all.is_rx_iface =
1407 rte_kvargs_count(kvlist, ETH_PCAP_RX_IFACE_ARG) ? 1 : 0;
1408 pcaps.num_of_queue = 0;
1410 devargs_all.is_tx_pcap =
1411 rte_kvargs_count(kvlist, ETH_PCAP_TX_PCAP_ARG) ? 1 : 0;
1412 devargs_all.is_tx_iface =
1413 rte_kvargs_count(kvlist, ETH_PCAP_TX_IFACE_ARG) ? 1 : 0;
1414 dumpers.num_of_queue = 0;
1416 if (devargs_all.is_rx_pcap) {
1418 * We check whether we want to infinitely rx the pcap file.
1420 unsigned int infinite_rx_arg_cnt = rte_kvargs_count(kvlist,
1421 ETH_PCAP_INFINITE_RX_ARG);
1423 if (infinite_rx_arg_cnt == 1) {
1424 ret = rte_kvargs_process(kvlist,
1425 ETH_PCAP_INFINITE_RX_ARG,
1426 &get_infinite_rx_arg,
1427 &devargs_all.infinite_rx);
1430 PMD_LOG(INFO, "infinite_rx has been %s for %s",
1431 devargs_all.infinite_rx ? "enabled" : "disabled",
1434 } else if (infinite_rx_arg_cnt > 1) {
1435 PMD_LOG(WARNING, "infinite_rx has not been enabled since the "
1436 "argument has been provided more than once "
1440 ret = rte_kvargs_process(kvlist, ETH_PCAP_RX_PCAP_ARG,
1441 &open_rx_pcap, &pcaps);
1442 } else if (devargs_all.is_rx_iface) {
1443 ret = rte_kvargs_process(kvlist, NULL,
1444 &rx_iface_args_process, &pcaps);
1445 } else if (devargs_all.is_tx_iface || devargs_all.is_tx_pcap) {
1448 /* Count number of tx queue args passed before dummy rx queue
1449 * creation so a dummy rx queue can be created for each tx queue
1451 unsigned int num_tx_queues =
1452 (rte_kvargs_count(kvlist, ETH_PCAP_TX_PCAP_ARG) +
1453 rte_kvargs_count(kvlist, ETH_PCAP_TX_IFACE_ARG));
1455 PMD_LOG(INFO, "Creating null rx queue since no rx queues were provided.");
1457 /* Creating a dummy rx queue for each tx queue passed */
1458 for (i = 0; i < num_tx_queues; i++)
1459 ret = add_queue(&pcaps, "dummy_rx", "rx_null", NULL,
1462 PMD_LOG(ERR, "Error - No rx or tx queues provided");
1469 * We check whether we want to open a TX stream to a real NIC,
1470 * a pcap file, or drop packets on tx
1472 if (devargs_all.is_tx_pcap) {
1473 ret = rte_kvargs_process(kvlist, ETH_PCAP_TX_PCAP_ARG,
1474 &open_tx_pcap, &dumpers);
1475 } else if (devargs_all.is_tx_iface) {
1476 ret = rte_kvargs_process(kvlist, ETH_PCAP_TX_IFACE_ARG,
1477 &open_tx_iface, &dumpers);
1481 PMD_LOG(INFO, "Dropping packets on tx since no tx queues were provided.");
1483 /* Add 1 dummy queue per rxq which counts and drops packets. */
1484 for (i = 0; i < pcaps.num_of_queue; i++)
1485 ret = add_queue(&dumpers, "dummy_tx", "tx_drop", NULL,
1493 if (rte_eal_process_type() == RTE_PROC_SECONDARY) {
1494 struct pmd_process_private *pp;
1497 internal = eth_dev->data->dev_private;
1498 pp = (struct pmd_process_private *)
1500 sizeof(struct pmd_process_private),
1501 RTE_CACHE_LINE_SIZE);
1505 "Failed to allocate memory for process private");
1510 eth_dev->dev_ops = &ops;
1511 eth_dev->device = &dev->device;
1513 /* setup process private */
1514 for (i = 0; i < pcaps.num_of_queue; i++)
1515 pp->rx_pcap[i] = pcaps.queue[i].pcap;
1517 for (i = 0; i < dumpers.num_of_queue; i++) {
1518 pp->tx_dumper[i] = dumpers.queue[i].dumper;
1519 pp->tx_pcap[i] = dumpers.queue[i].pcap;
1522 eth_dev->process_private = pp;
1523 eth_dev->rx_pkt_burst = eth_pcap_rx;
1524 if (devargs_all.is_tx_pcap)
1525 eth_dev->tx_pkt_burst = eth_pcap_tx_dumper;
1527 eth_dev->tx_pkt_burst = eth_pcap_tx;
1529 rte_eth_dev_probing_finish(eth_dev);
1533 devargs_all.rx_queues = pcaps;
1534 devargs_all.tx_queues = dumpers;
1536 ret = eth_from_pcaps(dev, &devargs_all);
1539 rte_kvargs_free(kvlist);
1545 pmd_pcap_remove(struct rte_vdev_device *dev)
1547 struct pmd_internals *internals = NULL;
1548 struct rte_eth_dev *eth_dev = NULL;
1550 PMD_LOG(INFO, "Closing pcap ethdev on numa socket %d",
1556 /* reserve an ethdev entry */
1557 eth_dev = rte_eth_dev_allocated(rte_vdev_device_name(dev));
1558 if (eth_dev == NULL)
1561 if (rte_eal_process_type() == RTE_PROC_PRIMARY) {
1562 internals = eth_dev->data->dev_private;
1563 if (internals != NULL && internals->phy_mac == 0)
1564 /* not dynamically allocated, must not be freed */
1565 eth_dev->data->mac_addrs = NULL;
1568 eth_dev_close(eth_dev);
1570 rte_free(eth_dev->process_private);
1571 rte_eth_dev_release_port(eth_dev);
1576 static struct rte_vdev_driver pmd_pcap_drv = {
1577 .probe = pmd_pcap_probe,
1578 .remove = pmd_pcap_remove,
1581 RTE_PMD_REGISTER_VDEV(net_pcap, pmd_pcap_drv);
1582 RTE_PMD_REGISTER_ALIAS(net_pcap, eth_pcap);
1583 RTE_PMD_REGISTER_PARAM_STRING(net_pcap,
1584 ETH_PCAP_RX_PCAP_ARG "=<string> "
1585 ETH_PCAP_TX_PCAP_ARG "=<string> "
1586 ETH_PCAP_RX_IFACE_ARG "=<ifc> "
1587 ETH_PCAP_RX_IFACE_IN_ARG "=<ifc> "
1588 ETH_PCAP_TX_IFACE_ARG "=<ifc> "
1589 ETH_PCAP_IFACE_ARG "=<ifc> "
1590 ETH_PCAP_PHY_MAC_ARG "=<int>"
1591 ETH_PCAP_INFINITE_RX_ARG "=<0|1>");