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 static int eth_pcap_logtype;
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)
291 calculate_timestamp(struct timeval *ts) {
293 struct timeval cur_time;
295 cycles = rte_get_timer_cycles() - start_cycles;
296 cur_time.tv_sec = cycles / hz;
297 cur_time.tv_usec = (cycles % hz) * 1e6 / hz;
298 timeradd(&start_time, &cur_time, ts);
302 * Callback to handle writing packets to a pcap file.
305 eth_pcap_tx_dumper(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
308 struct rte_mbuf *mbuf;
309 struct pmd_process_private *pp;
310 struct pcap_tx_queue *dumper_q = queue;
312 uint32_t tx_bytes = 0;
313 struct pcap_pkthdr header;
314 pcap_dumper_t *dumper;
315 unsigned char temp_data[RTE_ETH_PCAP_SNAPLEN];
318 pp = rte_eth_devices[dumper_q->port_id].process_private;
319 dumper = pp->tx_dumper[dumper_q->queue_id];
321 if (dumper == NULL || nb_pkts == 0)
324 /* writes the nb_pkts packets to the previously opened pcap file
326 for (i = 0; i < nb_pkts; i++) {
328 len = rte_pktmbuf_pkt_len(mbuf);
329 if (unlikely(!rte_pktmbuf_is_contiguous(mbuf) &&
330 len > sizeof(temp_data))) {
332 "Dropping multi segment PCAP packet. Size (%zd) > max size (%zd).",
333 len, sizeof(temp_data));
334 rte_pktmbuf_free(mbuf);
338 calculate_timestamp(&header.ts);
340 header.caplen = header.len;
341 /* rte_pktmbuf_read() returns a pointer to the data directly
342 * in the mbuf (when the mbuf is contiguous) or, otherwise,
343 * a pointer to temp_data after copying into it.
345 pcap_dump((u_char *)dumper, &header,
346 rte_pktmbuf_read(mbuf, 0, len, temp_data));
350 rte_pktmbuf_free(mbuf);
354 * Since there's no place to hook a callback when the forwarding
355 * process stops and to make sure the pcap file is actually written,
356 * we flush the pcap dumper within each burst.
358 pcap_dump_flush(dumper);
359 dumper_q->tx_stat.pkts += num_tx;
360 dumper_q->tx_stat.bytes += tx_bytes;
361 dumper_q->tx_stat.err_pkts += nb_pkts - num_tx;
367 * Callback to handle dropping packets in the infinite rx case.
370 eth_tx_drop(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
373 uint32_t tx_bytes = 0;
374 struct pcap_tx_queue *tx_queue = queue;
376 if (unlikely(nb_pkts == 0))
379 for (i = 0; i < nb_pkts; i++) {
380 tx_bytes += bufs[i]->data_len;
381 rte_pktmbuf_free(bufs[i]);
384 tx_queue->tx_stat.pkts += nb_pkts;
385 tx_queue->tx_stat.bytes += tx_bytes;
391 * Callback to handle sending packets through a real NIC.
394 eth_pcap_tx(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
398 struct rte_mbuf *mbuf;
399 struct pmd_process_private *pp;
400 struct pcap_tx_queue *tx_queue = queue;
402 uint32_t tx_bytes = 0;
404 unsigned char temp_data[RTE_ETH_PCAP_SNAPLEN];
407 pp = rte_eth_devices[tx_queue->port_id].process_private;
408 pcap = pp->tx_pcap[tx_queue->queue_id];
410 if (unlikely(nb_pkts == 0 || pcap == NULL))
413 for (i = 0; i < nb_pkts; i++) {
415 len = rte_pktmbuf_pkt_len(mbuf);
416 if (unlikely(!rte_pktmbuf_is_contiguous(mbuf) &&
417 len > sizeof(temp_data))) {
419 "Dropping multi segment PCAP packet. Size (%zd) > max size (%zd).",
420 len, sizeof(temp_data));
421 rte_pktmbuf_free(mbuf);
425 /* rte_pktmbuf_read() returns a pointer to the data directly
426 * in the mbuf (when the mbuf is contiguous) or, otherwise,
427 * a pointer to temp_data after copying into it.
429 ret = pcap_sendpacket(pcap,
430 rte_pktmbuf_read(mbuf, 0, len, temp_data), len);
431 if (unlikely(ret != 0))
435 rte_pktmbuf_free(mbuf);
438 tx_queue->tx_stat.pkts += num_tx;
439 tx_queue->tx_stat.bytes += tx_bytes;
440 tx_queue->tx_stat.err_pkts += i - num_tx;
446 * pcap_open_live wrapper function
449 open_iface_live(const char *iface, pcap_t **pcap) {
450 *pcap = pcap_open_live(iface, RTE_ETH_PCAP_SNAPLEN,
451 RTE_ETH_PCAP_PROMISC, RTE_ETH_PCAP_TIMEOUT, errbuf);
454 PMD_LOG(ERR, "Couldn't open %s: %s", iface, errbuf);
462 open_single_iface(const char *iface, pcap_t **pcap)
464 if (open_iface_live(iface, pcap) < 0) {
465 PMD_LOG(ERR, "Couldn't open interface %s", iface);
473 open_single_tx_pcap(const char *pcap_filename, pcap_dumper_t **dumper)
478 * We need to create a dummy empty pcap_t to use it
479 * with pcap_dump_open(). We create big enough an Ethernet
482 tx_pcap = pcap_open_dead(DLT_EN10MB, RTE_ETH_PCAP_SNAPSHOT_LEN);
483 if (tx_pcap == NULL) {
484 PMD_LOG(ERR, "Couldn't create dead pcap");
488 /* The dumper is created using the previous pcap_t reference */
489 *dumper = pcap_dump_open(tx_pcap, pcap_filename);
490 if (*dumper == NULL) {
492 PMD_LOG(ERR, "Couldn't open %s for writing.",
502 open_single_rx_pcap(const char *pcap_filename, pcap_t **pcap)
504 *pcap = pcap_open_offline(pcap_filename, errbuf);
506 PMD_LOG(ERR, "Couldn't open %s: %s", pcap_filename,
515 count_packets_in_pcap(pcap_t **pcap, struct pcap_rx_queue *pcap_q)
517 const u_char *packet;
518 struct pcap_pkthdr header;
519 uint64_t pcap_pkt_count = 0;
521 while ((packet = pcap_next(*pcap, &header)))
524 /* The pcap is reopened so it can be used as normal later. */
527 open_single_rx_pcap(pcap_q->name, pcap);
529 return pcap_pkt_count;
533 eth_dev_start(struct rte_eth_dev *dev)
536 struct pmd_internals *internals = dev->data->dev_private;
537 struct pmd_process_private *pp = dev->process_private;
538 struct pcap_tx_queue *tx;
539 struct pcap_rx_queue *rx;
541 /* Special iface case. Single pcap is open and shared between tx/rx. */
542 if (internals->single_iface) {
543 tx = &internals->tx_queue[0];
544 rx = &internals->rx_queue[0];
546 if (!pp->tx_pcap[0] &&
547 strcmp(tx->type, ETH_PCAP_IFACE_ARG) == 0) {
548 if (open_single_iface(tx->name, &pp->tx_pcap[0]) < 0)
550 pp->rx_pcap[0] = pp->tx_pcap[0];
556 /* If not open already, open tx pcaps/dumpers */
557 for (i = 0; i < dev->data->nb_tx_queues; i++) {
558 tx = &internals->tx_queue[i];
560 if (!pp->tx_dumper[i] &&
561 strcmp(tx->type, ETH_PCAP_TX_PCAP_ARG) == 0) {
562 if (open_single_tx_pcap(tx->name,
563 &pp->tx_dumper[i]) < 0)
565 } else if (!pp->tx_pcap[i] &&
566 strcmp(tx->type, ETH_PCAP_TX_IFACE_ARG) == 0) {
567 if (open_single_iface(tx->name, &pp->tx_pcap[i]) < 0)
572 /* If not open already, open rx pcaps */
573 for (i = 0; i < dev->data->nb_rx_queues; i++) {
574 rx = &internals->rx_queue[i];
576 if (pp->rx_pcap[i] != NULL)
579 if (strcmp(rx->type, ETH_PCAP_RX_PCAP_ARG) == 0) {
580 if (open_single_rx_pcap(rx->name, &pp->rx_pcap[i]) < 0)
582 } else if (strcmp(rx->type, ETH_PCAP_RX_IFACE_ARG) == 0) {
583 if (open_single_iface(rx->name, &pp->rx_pcap[i]) < 0)
589 for (i = 0; i < dev->data->nb_rx_queues; i++)
590 dev->data->rx_queue_state[i] = RTE_ETH_QUEUE_STATE_STARTED;
592 for (i = 0; i < dev->data->nb_tx_queues; i++)
593 dev->data->tx_queue_state[i] = RTE_ETH_QUEUE_STATE_STARTED;
595 dev->data->dev_link.link_status = ETH_LINK_UP;
601 * This function gets called when the current port gets stopped.
602 * Is the only place for us to close all the tx streams dumpers.
603 * If not called the dumpers will be flushed within each tx burst.
606 eth_dev_stop(struct rte_eth_dev *dev)
609 struct pmd_internals *internals = dev->data->dev_private;
610 struct pmd_process_private *pp = dev->process_private;
612 /* Special iface case. Single pcap is open and shared between tx/rx. */
613 if (internals->single_iface) {
614 pcap_close(pp->tx_pcap[0]);
615 pp->tx_pcap[0] = NULL;
616 pp->rx_pcap[0] = NULL;
620 for (i = 0; i < dev->data->nb_tx_queues; i++) {
621 if (pp->tx_dumper[i] != NULL) {
622 pcap_dump_close(pp->tx_dumper[i]);
623 pp->tx_dumper[i] = NULL;
626 if (pp->tx_pcap[i] != NULL) {
627 pcap_close(pp->tx_pcap[i]);
628 pp->tx_pcap[i] = NULL;
632 for (i = 0; i < dev->data->nb_rx_queues; i++) {
633 if (pp->rx_pcap[i] != NULL) {
634 pcap_close(pp->rx_pcap[i]);
635 pp->rx_pcap[i] = NULL;
640 for (i = 0; i < dev->data->nb_rx_queues; i++)
641 dev->data->rx_queue_state[i] = RTE_ETH_QUEUE_STATE_STOPPED;
643 for (i = 0; i < dev->data->nb_tx_queues; i++)
644 dev->data->tx_queue_state[i] = RTE_ETH_QUEUE_STATE_STOPPED;
646 dev->data->dev_link.link_status = ETH_LINK_DOWN;
650 eth_dev_configure(struct rte_eth_dev *dev __rte_unused)
656 eth_dev_info(struct rte_eth_dev *dev,
657 struct rte_eth_dev_info *dev_info)
659 struct pmd_internals *internals = dev->data->dev_private;
661 dev_info->if_index = internals->if_index;
662 dev_info->max_mac_addrs = 1;
663 dev_info->max_rx_pktlen = (uint32_t) -1;
664 dev_info->max_rx_queues = dev->data->nb_rx_queues;
665 dev_info->max_tx_queues = dev->data->nb_tx_queues;
666 dev_info->min_rx_bufsize = 0;
670 eth_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
673 unsigned long rx_packets_total = 0, rx_bytes_total = 0;
674 unsigned long tx_packets_total = 0, tx_bytes_total = 0;
675 unsigned long tx_packets_err_total = 0;
676 const struct pmd_internals *internal = dev->data->dev_private;
678 for (i = 0; i < RTE_ETHDEV_QUEUE_STAT_CNTRS &&
679 i < dev->data->nb_rx_queues; i++) {
680 stats->q_ipackets[i] = internal->rx_queue[i].rx_stat.pkts;
681 stats->q_ibytes[i] = internal->rx_queue[i].rx_stat.bytes;
682 rx_packets_total += stats->q_ipackets[i];
683 rx_bytes_total += stats->q_ibytes[i];
686 for (i = 0; i < RTE_ETHDEV_QUEUE_STAT_CNTRS &&
687 i < dev->data->nb_tx_queues; i++) {
688 stats->q_opackets[i] = internal->tx_queue[i].tx_stat.pkts;
689 stats->q_obytes[i] = internal->tx_queue[i].tx_stat.bytes;
690 tx_packets_total += stats->q_opackets[i];
691 tx_bytes_total += stats->q_obytes[i];
692 tx_packets_err_total += internal->tx_queue[i].tx_stat.err_pkts;
695 stats->ipackets = rx_packets_total;
696 stats->ibytes = rx_bytes_total;
697 stats->opackets = tx_packets_total;
698 stats->obytes = tx_bytes_total;
699 stats->oerrors = tx_packets_err_total;
705 eth_stats_reset(struct rte_eth_dev *dev)
708 struct pmd_internals *internal = dev->data->dev_private;
710 for (i = 0; i < dev->data->nb_rx_queues; i++) {
711 internal->rx_queue[i].rx_stat.pkts = 0;
712 internal->rx_queue[i].rx_stat.bytes = 0;
715 for (i = 0; i < dev->data->nb_tx_queues; i++) {
716 internal->tx_queue[i].tx_stat.pkts = 0;
717 internal->tx_queue[i].tx_stat.bytes = 0;
718 internal->tx_queue[i].tx_stat.err_pkts = 0;
723 eth_dev_close(struct rte_eth_dev *dev)
726 struct pmd_internals *internals = dev->data->dev_private;
728 /* Device wide flag, but cleanup must be performed per queue. */
729 if (internals->infinite_rx) {
730 for (i = 0; i < dev->data->nb_rx_queues; i++) {
731 struct pcap_rx_queue *pcap_q = &internals->rx_queue[i];
732 struct rte_mbuf *pcap_buf;
734 while (!rte_ring_dequeue(pcap_q->pkts,
736 rte_pktmbuf_free(pcap_buf);
738 rte_ring_free(pcap_q->pkts);
745 eth_queue_release(void *q __rte_unused)
750 eth_link_update(struct rte_eth_dev *dev __rte_unused,
751 int wait_to_complete __rte_unused)
757 eth_rx_queue_setup(struct rte_eth_dev *dev,
758 uint16_t rx_queue_id,
759 uint16_t nb_rx_desc __rte_unused,
760 unsigned int socket_id __rte_unused,
761 const struct rte_eth_rxconf *rx_conf __rte_unused,
762 struct rte_mempool *mb_pool)
764 struct pmd_internals *internals = dev->data->dev_private;
765 struct pcap_rx_queue *pcap_q = &internals->rx_queue[rx_queue_id];
767 pcap_q->mb_pool = mb_pool;
768 pcap_q->port_id = dev->data->port_id;
769 pcap_q->queue_id = rx_queue_id;
770 dev->data->rx_queues[rx_queue_id] = pcap_q;
772 if (internals->infinite_rx) {
773 struct pmd_process_private *pp;
774 char ring_name[NAME_MAX];
775 static uint32_t ring_number;
776 uint64_t pcap_pkt_count = 0;
777 struct rte_mbuf *bufs[1];
780 pp = rte_eth_devices[pcap_q->port_id].process_private;
781 pcap = &pp->rx_pcap[pcap_q->queue_id];
783 if (unlikely(*pcap == NULL))
786 pcap_pkt_count = count_packets_in_pcap(pcap, pcap_q);
788 snprintf(ring_name, sizeof(ring_name), "PCAP_RING%" PRIu16,
791 pcap_q->pkts = rte_ring_create(ring_name,
792 rte_align64pow2(pcap_pkt_count + 1), 0,
793 RING_F_SP_ENQ | RING_F_SC_DEQ);
798 /* Fill ring with packets from PCAP file one by one. */
799 while (eth_pcap_rx(pcap_q, bufs, 1)) {
800 /* Check for multiseg mbufs. */
801 if (bufs[0]->nb_segs != 1) {
802 rte_pktmbuf_free(*bufs);
804 while (!rte_ring_dequeue(pcap_q->pkts,
806 rte_pktmbuf_free(*bufs);
808 rte_ring_free(pcap_q->pkts);
809 PMD_LOG(ERR, "Multiseg mbufs are not supported in infinite_rx "
814 rte_ring_enqueue_bulk(pcap_q->pkts,
815 (void * const *)bufs, 1, NULL);
818 * Reset the stats for this queue since eth_pcap_rx calls above
819 * didn't result in the application receiving packets.
821 pcap_q->rx_stat.pkts = 0;
822 pcap_q->rx_stat.bytes = 0;
829 eth_tx_queue_setup(struct rte_eth_dev *dev,
830 uint16_t tx_queue_id,
831 uint16_t nb_tx_desc __rte_unused,
832 unsigned int socket_id __rte_unused,
833 const struct rte_eth_txconf *tx_conf __rte_unused)
835 struct pmd_internals *internals = dev->data->dev_private;
836 struct pcap_tx_queue *pcap_q = &internals->tx_queue[tx_queue_id];
838 pcap_q->port_id = dev->data->port_id;
839 pcap_q->queue_id = tx_queue_id;
840 dev->data->tx_queues[tx_queue_id] = pcap_q;
846 eth_rx_queue_start(struct rte_eth_dev *dev, uint16_t rx_queue_id)
848 dev->data->rx_queue_state[rx_queue_id] = RTE_ETH_QUEUE_STATE_STARTED;
854 eth_tx_queue_start(struct rte_eth_dev *dev, uint16_t tx_queue_id)
856 dev->data->tx_queue_state[tx_queue_id] = RTE_ETH_QUEUE_STATE_STARTED;
862 eth_rx_queue_stop(struct rte_eth_dev *dev, uint16_t rx_queue_id)
864 dev->data->rx_queue_state[rx_queue_id] = RTE_ETH_QUEUE_STATE_STOPPED;
870 eth_tx_queue_stop(struct rte_eth_dev *dev, uint16_t tx_queue_id)
872 dev->data->tx_queue_state[tx_queue_id] = RTE_ETH_QUEUE_STATE_STOPPED;
877 static const struct eth_dev_ops ops = {
878 .dev_start = eth_dev_start,
879 .dev_stop = eth_dev_stop,
880 .dev_close = eth_dev_close,
881 .dev_configure = eth_dev_configure,
882 .dev_infos_get = eth_dev_info,
883 .rx_queue_setup = eth_rx_queue_setup,
884 .tx_queue_setup = eth_tx_queue_setup,
885 .rx_queue_start = eth_rx_queue_start,
886 .tx_queue_start = eth_tx_queue_start,
887 .rx_queue_stop = eth_rx_queue_stop,
888 .tx_queue_stop = eth_tx_queue_stop,
889 .rx_queue_release = eth_queue_release,
890 .tx_queue_release = eth_queue_release,
891 .link_update = eth_link_update,
892 .stats_get = eth_stats_get,
893 .stats_reset = eth_stats_reset,
897 add_queue(struct pmd_devargs *pmd, const char *name, const char *type,
898 pcap_t *pcap, pcap_dumper_t *dumper)
900 if (pmd->num_of_queue >= RTE_PMD_PCAP_MAX_QUEUES)
903 pmd->queue[pmd->num_of_queue].pcap = pcap;
905 pmd->queue[pmd->num_of_queue].dumper = dumper;
906 pmd->queue[pmd->num_of_queue].name = name;
907 pmd->queue[pmd->num_of_queue].type = type;
913 * Function handler that opens the pcap file for reading a stores a
914 * reference of it for use it later on.
917 open_rx_pcap(const char *key, const char *value, void *extra_args)
919 const char *pcap_filename = value;
920 struct pmd_devargs *rx = extra_args;
923 if (open_single_rx_pcap(pcap_filename, &pcap) < 0)
926 if (add_queue(rx, pcap_filename, key, pcap, NULL) < 0) {
935 * Opens a pcap file for writing and stores a reference to it
936 * for use it later on.
939 open_tx_pcap(const char *key, const char *value, void *extra_args)
941 const char *pcap_filename = value;
942 struct pmd_devargs *dumpers = extra_args;
943 pcap_dumper_t *dumper;
945 if (open_single_tx_pcap(pcap_filename, &dumper) < 0)
948 if (add_queue(dumpers, pcap_filename, key, NULL, dumper) < 0) {
949 pcap_dump_close(dumper);
957 * Opens an interface for reading and writing
960 open_rx_tx_iface(const char *key, const char *value, void *extra_args)
962 const char *iface = value;
963 struct pmd_devargs *tx = extra_args;
966 if (open_single_iface(iface, &pcap) < 0)
969 tx->queue[0].pcap = pcap;
970 tx->queue[0].name = iface;
971 tx->queue[0].type = key;
977 set_iface_direction(const char *iface, pcap_t *pcap,
978 pcap_direction_t direction)
980 const char *direction_str = (direction == PCAP_D_IN) ? "IN" : "OUT";
981 if (pcap_setdirection(pcap, direction) < 0) {
982 PMD_LOG(ERR, "Setting %s pcap direction %s failed - %s\n",
983 iface, direction_str, pcap_geterr(pcap));
986 PMD_LOG(INFO, "Setting %s pcap direction %s\n",
987 iface, direction_str);
992 open_iface(const char *key, const char *value, void *extra_args)
994 const char *iface = value;
995 struct pmd_devargs *pmd = extra_args;
998 if (open_single_iface(iface, &pcap) < 0)
1000 if (add_queue(pmd, iface, key, pcap, NULL) < 0) {
1009 * Opens a NIC for reading packets from it
1012 open_rx_iface(const char *key, const char *value, void *extra_args)
1014 int ret = open_iface(key, value, extra_args);
1017 if (strcmp(key, ETH_PCAP_RX_IFACE_IN_ARG) == 0) {
1018 struct pmd_devargs *pmd = extra_args;
1019 unsigned int qid = pmd->num_of_queue - 1;
1021 set_iface_direction(pmd->queue[qid].name,
1022 pmd->queue[qid].pcap,
1030 rx_iface_args_process(const char *key, const char *value, void *extra_args)
1032 if (strcmp(key, ETH_PCAP_RX_IFACE_ARG) == 0 ||
1033 strcmp(key, ETH_PCAP_RX_IFACE_IN_ARG) == 0)
1034 return open_rx_iface(key, value, extra_args);
1040 * Opens a NIC for writing packets to it
1043 open_tx_iface(const char *key, const char *value, void *extra_args)
1045 return open_iface(key, value, extra_args);
1049 select_phy_mac(const char *key __rte_unused, const char *value,
1053 const int phy_mac = atoi(value);
1054 int *enable_phy_mac = extra_args;
1057 *enable_phy_mac = 1;
1063 get_infinite_rx_arg(const char *key __rte_unused,
1064 const char *value, void *extra_args)
1067 const int infinite_rx = atoi(value);
1068 int *enable_infinite_rx = extra_args;
1070 if (infinite_rx > 0)
1071 *enable_infinite_rx = 1;
1077 pmd_init_internals(struct rte_vdev_device *vdev,
1078 const unsigned int nb_rx_queues,
1079 const unsigned int nb_tx_queues,
1080 struct pmd_internals **internals,
1081 struct rte_eth_dev **eth_dev)
1083 struct rte_eth_dev_data *data;
1084 struct pmd_process_private *pp;
1085 unsigned int numa_node = vdev->device.numa_node;
1087 PMD_LOG(INFO, "Creating pcap-backed ethdev on numa socket %d",
1090 pp = (struct pmd_process_private *)
1091 rte_zmalloc(NULL, sizeof(struct pmd_process_private),
1092 RTE_CACHE_LINE_SIZE);
1096 "Failed to allocate memory for process private");
1100 /* reserve an ethdev entry */
1101 *eth_dev = rte_eth_vdev_allocate(vdev, sizeof(**internals));
1106 (*eth_dev)->process_private = pp;
1107 /* now put it all together
1108 * - store queue data in internals,
1109 * - store numa_node info in eth_dev
1110 * - point eth_dev_data to internals
1111 * - and point eth_dev structure to new eth_dev_data structure
1113 *internals = (*eth_dev)->data->dev_private;
1115 * Interface MAC = 02:70:63:61:70:<iface_idx>
1116 * derived from: 'locally administered':'p':'c':'a':'p':'iface_idx'
1117 * where the middle 4 characters are converted to hex.
1119 (*internals)->eth_addr = (struct rte_ether_addr) {
1120 .addr_bytes = { 0x02, 0x70, 0x63, 0x61, 0x70, iface_idx++ }
1122 (*internals)->phy_mac = 0;
1123 data = (*eth_dev)->data;
1124 data->nb_rx_queues = (uint16_t)nb_rx_queues;
1125 data->nb_tx_queues = (uint16_t)nb_tx_queues;
1126 data->dev_link = pmd_link;
1127 data->mac_addrs = &(*internals)->eth_addr;
1130 * NOTE: we'll replace the data element, of originally allocated
1131 * eth_dev so the rings are local per-process
1133 (*eth_dev)->dev_ops = &ops;
1135 strlcpy((*internals)->devargs, rte_vdev_device_args(vdev),
1136 ETH_PCAP_ARG_MAXLEN);
1142 eth_pcap_update_mac(const char *if_name, struct rte_eth_dev *eth_dev,
1143 const unsigned int numa_node)
1145 #if defined(RTE_EXEC_ENV_LINUX)
1148 int if_fd = socket(AF_INET, SOCK_DGRAM, 0);
1153 rte_strscpy(ifr.ifr_name, if_name, sizeof(ifr.ifr_name));
1154 if (ioctl(if_fd, SIOCGIFHWADDR, &ifr)) {
1159 mac_addrs = rte_zmalloc_socket(NULL, RTE_ETHER_ADDR_LEN, 0, numa_node);
1165 PMD_LOG(INFO, "Setting phy MAC for %s", if_name);
1166 eth_dev->data->mac_addrs = mac_addrs;
1167 rte_memcpy(eth_dev->data->mac_addrs[0].addr_bytes,
1168 ifr.ifr_hwaddr.sa_data, RTE_ETHER_ADDR_LEN);
1174 #elif defined(RTE_EXEC_ENV_FREEBSD)
1176 struct if_msghdr *ifm;
1177 struct sockaddr_dl *sdl;
1186 mib[4] = NET_RT_IFLIST;
1187 mib[5] = if_nametoindex(if_name);
1189 if (sysctl(mib, 6, NULL, &len, NULL, 0) < 0)
1195 buf = rte_malloc(NULL, len, 0);
1199 if (sysctl(mib, 6, buf, &len, NULL, 0) < 0) {
1203 ifm = (struct if_msghdr *)buf;
1204 sdl = (struct sockaddr_dl *)(ifm + 1);
1206 mac_addrs = rte_zmalloc_socket(NULL, RTE_ETHER_ADDR_LEN, 0, numa_node);
1212 PMD_LOG(INFO, "Setting phy MAC for %s", if_name);
1213 eth_dev->data->mac_addrs = mac_addrs;
1214 rte_memcpy(eth_dev->data->mac_addrs[0].addr_bytes,
1215 LLADDR(sdl), RTE_ETHER_ADDR_LEN);
1226 eth_from_pcaps_common(struct rte_vdev_device *vdev,
1227 struct pmd_devargs_all *devargs_all,
1228 struct pmd_internals **internals, struct rte_eth_dev **eth_dev)
1230 struct pmd_process_private *pp;
1231 struct pmd_devargs *rx_queues = &devargs_all->rx_queues;
1232 struct pmd_devargs *tx_queues = &devargs_all->tx_queues;
1233 const unsigned int nb_rx_queues = rx_queues->num_of_queue;
1234 const unsigned int nb_tx_queues = tx_queues->num_of_queue;
1237 /* do some parameter checking */
1238 if (rx_queues == NULL && nb_rx_queues > 0)
1240 if (tx_queues == NULL && nb_tx_queues > 0)
1243 if (pmd_init_internals(vdev, nb_rx_queues, nb_tx_queues, internals,
1247 pp = (*eth_dev)->process_private;
1248 for (i = 0; i < nb_rx_queues; i++) {
1249 struct pcap_rx_queue *rx = &(*internals)->rx_queue[i];
1250 struct devargs_queue *queue = &rx_queues->queue[i];
1252 pp->rx_pcap[i] = queue->pcap;
1253 strlcpy(rx->name, queue->name, sizeof(rx->name));
1254 strlcpy(rx->type, queue->type, sizeof(rx->type));
1257 for (i = 0; i < nb_tx_queues; i++) {
1258 struct pcap_tx_queue *tx = &(*internals)->tx_queue[i];
1259 struct devargs_queue *queue = &tx_queues->queue[i];
1261 pp->tx_dumper[i] = queue->dumper;
1262 pp->tx_pcap[i] = queue->pcap;
1263 strlcpy(tx->name, queue->name, sizeof(tx->name));
1264 strlcpy(tx->type, queue->type, sizeof(tx->type));
1271 eth_from_pcaps(struct rte_vdev_device *vdev,
1272 struct pmd_devargs_all *devargs_all)
1274 struct pmd_internals *internals = NULL;
1275 struct rte_eth_dev *eth_dev = NULL;
1276 struct pmd_devargs *rx_queues = &devargs_all->rx_queues;
1277 int single_iface = devargs_all->single_iface;
1278 unsigned int infinite_rx = devargs_all->infinite_rx;
1281 ret = eth_from_pcaps_common(vdev, devargs_all, &internals, ð_dev);
1286 /* store weather we are using a single interface for rx/tx or not */
1287 internals->single_iface = single_iface;
1290 internals->if_index = if_nametoindex(rx_queues->queue[0].name);
1292 /* phy_mac arg is applied only only if "iface" devarg is provided */
1293 if (rx_queues->phy_mac) {
1294 int ret = eth_pcap_update_mac(rx_queues->queue[0].name,
1295 eth_dev, vdev->device.numa_node);
1297 internals->phy_mac = 1;
1301 internals->infinite_rx = infinite_rx;
1302 /* Assign rx ops. */
1304 eth_dev->rx_pkt_burst = eth_pcap_rx_infinite;
1305 else if (devargs_all->is_rx_pcap || devargs_all->is_rx_iface ||
1307 eth_dev->rx_pkt_burst = eth_pcap_rx;
1309 eth_dev->rx_pkt_burst = eth_null_rx;
1311 /* Assign tx ops. */
1312 if (devargs_all->is_tx_pcap)
1313 eth_dev->tx_pkt_burst = eth_pcap_tx_dumper;
1314 else if (devargs_all->is_tx_iface || single_iface)
1315 eth_dev->tx_pkt_burst = eth_pcap_tx;
1317 eth_dev->tx_pkt_burst = eth_tx_drop;
1319 rte_eth_dev_probing_finish(eth_dev);
1324 pmd_pcap_probe(struct rte_vdev_device *dev)
1327 struct rte_kvargs *kvlist;
1328 struct pmd_devargs pcaps = {0};
1329 struct pmd_devargs dumpers = {0};
1330 struct rte_eth_dev *eth_dev = NULL;
1331 struct pmd_internals *internal;
1334 struct pmd_devargs_all devargs_all = {
1341 name = rte_vdev_device_name(dev);
1342 PMD_LOG(INFO, "Initializing pmd_pcap for %s", name);
1344 gettimeofday(&start_time, NULL);
1345 start_cycles = rte_get_timer_cycles();
1346 hz = rte_get_timer_hz();
1348 if (rte_eal_process_type() == RTE_PROC_SECONDARY) {
1349 eth_dev = rte_eth_dev_attach_secondary(name);
1351 PMD_LOG(ERR, "Failed to probe %s", name);
1355 internal = eth_dev->data->dev_private;
1357 kvlist = rte_kvargs_parse(internal->devargs, valid_arguments);
1361 kvlist = rte_kvargs_parse(rte_vdev_device_args(dev),
1368 * If iface argument is passed we open the NICs and use them for
1371 if (rte_kvargs_count(kvlist, ETH_PCAP_IFACE_ARG) == 1) {
1373 ret = rte_kvargs_process(kvlist, ETH_PCAP_IFACE_ARG,
1374 &open_rx_tx_iface, &pcaps);
1378 dumpers.queue[0] = pcaps.queue[0];
1380 ret = rte_kvargs_process(kvlist, ETH_PCAP_PHY_MAC_ARG,
1381 &select_phy_mac, &pcaps.phy_mac);
1385 dumpers.phy_mac = pcaps.phy_mac;
1387 devargs_all.single_iface = 1;
1388 pcaps.num_of_queue = 1;
1389 dumpers.num_of_queue = 1;
1395 * We check whether we want to open a RX stream from a real NIC, a
1396 * pcap file or open a dummy RX stream
1398 devargs_all.is_rx_pcap =
1399 rte_kvargs_count(kvlist, ETH_PCAP_RX_PCAP_ARG) ? 1 : 0;
1400 devargs_all.is_rx_iface =
1401 rte_kvargs_count(kvlist, ETH_PCAP_RX_IFACE_ARG) ? 1 : 0;
1402 pcaps.num_of_queue = 0;
1404 devargs_all.is_tx_pcap =
1405 rte_kvargs_count(kvlist, ETH_PCAP_TX_PCAP_ARG) ? 1 : 0;
1406 devargs_all.is_tx_iface =
1407 rte_kvargs_count(kvlist, ETH_PCAP_TX_IFACE_ARG) ? 1 : 0;
1408 dumpers.num_of_queue = 0;
1410 if (devargs_all.is_rx_pcap) {
1412 * We check whether we want to infinitely rx the pcap file.
1414 unsigned int infinite_rx_arg_cnt = rte_kvargs_count(kvlist,
1415 ETH_PCAP_INFINITE_RX_ARG);
1417 if (infinite_rx_arg_cnt == 1) {
1418 ret = rte_kvargs_process(kvlist,
1419 ETH_PCAP_INFINITE_RX_ARG,
1420 &get_infinite_rx_arg,
1421 &devargs_all.infinite_rx);
1424 PMD_LOG(INFO, "infinite_rx has been %s for %s",
1425 devargs_all.infinite_rx ? "enabled" : "disabled",
1428 } else if (infinite_rx_arg_cnt > 1) {
1429 PMD_LOG(WARNING, "infinite_rx has not been enabled since the "
1430 "argument has been provided more than once "
1434 ret = rte_kvargs_process(kvlist, ETH_PCAP_RX_PCAP_ARG,
1435 &open_rx_pcap, &pcaps);
1436 } else if (devargs_all.is_rx_iface) {
1437 ret = rte_kvargs_process(kvlist, NULL,
1438 &rx_iface_args_process, &pcaps);
1439 } else if (devargs_all.is_tx_iface || devargs_all.is_tx_pcap) {
1442 /* Count number of tx queue args passed before dummy rx queue
1443 * creation so a dummy rx queue can be created for each tx queue
1445 unsigned int num_tx_queues =
1446 (rte_kvargs_count(kvlist, ETH_PCAP_TX_PCAP_ARG) +
1447 rte_kvargs_count(kvlist, ETH_PCAP_TX_IFACE_ARG));
1449 PMD_LOG(INFO, "Creating null rx queue since no rx queues were provided.");
1451 /* Creating a dummy rx queue for each tx queue passed */
1452 for (i = 0; i < num_tx_queues; i++)
1453 ret = add_queue(&pcaps, "dummy_rx", "rx_null", NULL,
1456 PMD_LOG(ERR, "Error - No rx or tx queues provided");
1463 * We check whether we want to open a TX stream to a real NIC,
1464 * a pcap file, or drop packets on tx
1466 if (devargs_all.is_tx_pcap) {
1467 ret = rte_kvargs_process(kvlist, ETH_PCAP_TX_PCAP_ARG,
1468 &open_tx_pcap, &dumpers);
1469 } else if (devargs_all.is_tx_iface) {
1470 ret = rte_kvargs_process(kvlist, ETH_PCAP_TX_IFACE_ARG,
1471 &open_tx_iface, &dumpers);
1475 PMD_LOG(INFO, "Dropping packets on tx since no tx queues were provided.");
1477 /* Add 1 dummy queue per rxq which counts and drops packets. */
1478 for (i = 0; i < pcaps.num_of_queue; i++)
1479 ret = add_queue(&dumpers, "dummy_tx", "tx_drop", NULL,
1487 if (rte_eal_process_type() == RTE_PROC_SECONDARY) {
1488 struct pmd_process_private *pp;
1491 internal = eth_dev->data->dev_private;
1492 pp = (struct pmd_process_private *)
1494 sizeof(struct pmd_process_private),
1495 RTE_CACHE_LINE_SIZE);
1499 "Failed to allocate memory for process private");
1504 eth_dev->dev_ops = &ops;
1505 eth_dev->device = &dev->device;
1507 /* setup process private */
1508 for (i = 0; i < pcaps.num_of_queue; i++)
1509 pp->rx_pcap[i] = pcaps.queue[i].pcap;
1511 for (i = 0; i < dumpers.num_of_queue; i++) {
1512 pp->tx_dumper[i] = dumpers.queue[i].dumper;
1513 pp->tx_pcap[i] = dumpers.queue[i].pcap;
1516 eth_dev->process_private = pp;
1517 eth_dev->rx_pkt_burst = eth_pcap_rx;
1518 if (devargs_all.is_tx_pcap)
1519 eth_dev->tx_pkt_burst = eth_pcap_tx_dumper;
1521 eth_dev->tx_pkt_burst = eth_pcap_tx;
1523 rte_eth_dev_probing_finish(eth_dev);
1527 devargs_all.rx_queues = pcaps;
1528 devargs_all.tx_queues = dumpers;
1530 ret = eth_from_pcaps(dev, &devargs_all);
1533 rte_kvargs_free(kvlist);
1539 pmd_pcap_remove(struct rte_vdev_device *dev)
1541 struct pmd_internals *internals = NULL;
1542 struct rte_eth_dev *eth_dev = NULL;
1544 PMD_LOG(INFO, "Closing pcap ethdev on numa socket %d",
1550 /* reserve an ethdev entry */
1551 eth_dev = rte_eth_dev_allocated(rte_vdev_device_name(dev));
1552 if (eth_dev == NULL)
1555 if (rte_eal_process_type() == RTE_PROC_PRIMARY) {
1556 internals = eth_dev->data->dev_private;
1557 if (internals != NULL && internals->phy_mac == 0)
1558 /* not dynamically allocated, must not be freed */
1559 eth_dev->data->mac_addrs = NULL;
1562 eth_dev_close(eth_dev);
1564 rte_free(eth_dev->process_private);
1565 rte_eth_dev_release_port(eth_dev);
1570 static struct rte_vdev_driver pmd_pcap_drv = {
1571 .probe = pmd_pcap_probe,
1572 .remove = pmd_pcap_remove,
1575 RTE_PMD_REGISTER_VDEV(net_pcap, pmd_pcap_drv);
1576 RTE_PMD_REGISTER_ALIAS(net_pcap, eth_pcap);
1577 RTE_PMD_REGISTER_PARAM_STRING(net_pcap,
1578 ETH_PCAP_RX_PCAP_ARG "=<string> "
1579 ETH_PCAP_TX_PCAP_ARG "=<string> "
1580 ETH_PCAP_RX_IFACE_ARG "=<ifc> "
1581 ETH_PCAP_RX_IFACE_IN_ARG "=<ifc> "
1582 ETH_PCAP_TX_IFACE_ARG "=<ifc> "
1583 ETH_PCAP_IFACE_ARG "=<ifc> "
1584 ETH_PCAP_PHY_MAC_ARG "=<int>"
1585 ETH_PCAP_INFINITE_RX_ARG "=<0|1>");
1587 RTE_INIT(eth_pcap_init_log)
1589 eth_pcap_logtype = rte_log_register("pmd.net.pcap");
1590 if (eth_pcap_logtype >= 0)
1591 rte_log_set_level(eth_pcap_logtype, RTE_LOG_NOTICE);