]> git.proxmox.com Git - ceph.git/blob - ceph/src/spdk/dpdk/drivers/net/af_packet/rte_eth_af_packet.c
projects / ceph.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
update sources to ceph Nautilus 14.2.1
[ceph.git] / ceph / src / spdk / dpdk / drivers / net / af_packet / rte_eth_af_packet.c
1 /* SPDX-License-Identifier: BSD-3-Clause
2 * Copyright(c) 2014 John W. Linville <linville@tuxdriver.com>
3 * Originally based upon librte_pmd_pcap code:
4 * Copyright(c) 2010-2015 Intel Corporation.
5 * Copyright(c) 2014 6WIND S.A.
6 * All rights reserved.
7 */
8
9 #include <rte_mbuf.h>
10 #include <rte_ethdev_driver.h>
11 #include <rte_ethdev_vdev.h>
12 #include <rte_malloc.h>
13 #include <rte_kvargs.h>
14 #include <rte_bus_vdev.h>
15
16 #include <linux/if_ether.h>
17 #include <linux/if_packet.h>
18 #include <arpa/inet.h>
19 #include <net/if.h>
20 #include <sys/types.h>
21 #include <sys/socket.h>
22 #include <sys/ioctl.h>
23 #include <sys/mman.h>
24 #include <unistd.h>
25 #include <poll.h>
26
27 #define ETH_AF_PACKET_IFACE_ARG "iface"
28 #define ETH_AF_PACKET_NUM_Q_ARG "qpairs"
29 #define ETH_AF_PACKET_BLOCKSIZE_ARG "blocksz"
30 #define ETH_AF_PACKET_FRAMESIZE_ARG "framesz"
31 #define ETH_AF_PACKET_FRAMECOUNT_ARG "framecnt"
32 #define ETH_AF_PACKET_QDISC_BYPASS_ARG "qdisc_bypass"
33
34 #define DFLT_BLOCK_SIZE (1 << 12)
35 #define DFLT_FRAME_SIZE (1 << 11)
36 #define DFLT_FRAME_COUNT (1 << 9)
37
38 #define RTE_PMD_AF_PACKET_MAX_RINGS 16
39
40 struct pkt_rx_queue {
41 int sockfd;
42
43 struct iovec *rd;
44 uint8_t *map;
45 unsigned int framecount;
46 unsigned int framenum;
47
48 struct rte_mempool *mb_pool;
49 uint16_t in_port;
50
51 volatile unsigned long rx_pkts;
52 volatile unsigned long err_pkts;
53 volatile unsigned long rx_bytes;
54 };
55
56 struct pkt_tx_queue {
57 int sockfd;
58 unsigned int frame_data_size;
59
60 struct iovec *rd;
61 uint8_t *map;
62 unsigned int framecount;
63 unsigned int framenum;
64
65 volatile unsigned long tx_pkts;
66 volatile unsigned long err_pkts;
67 volatile unsigned long tx_bytes;
68 };
69
70 struct pmd_internals {
71 unsigned nb_queues;
72
73 int if_index;
74 char *if_name;
75 struct ether_addr eth_addr;
76
77 struct tpacket_req req;
78
79 struct pkt_rx_queue rx_queue[RTE_PMD_AF_PACKET_MAX_RINGS];
80 struct pkt_tx_queue tx_queue[RTE_PMD_AF_PACKET_MAX_RINGS];
81 };
82
83 static const char *valid_arguments[] = {
84 ETH_AF_PACKET_IFACE_ARG,
85 ETH_AF_PACKET_NUM_Q_ARG,
86 ETH_AF_PACKET_BLOCKSIZE_ARG,
87 ETH_AF_PACKET_FRAMESIZE_ARG,
88 ETH_AF_PACKET_FRAMECOUNT_ARG,
89 ETH_AF_PACKET_QDISC_BYPASS_ARG,
90 NULL
91 };
92
93 static struct rte_eth_link pmd_link = {
94 .link_speed = ETH_SPEED_NUM_10G,
95 .link_duplex = ETH_LINK_FULL_DUPLEX,
96 .link_status = ETH_LINK_DOWN,
97 .link_autoneg = ETH_LINK_FIXED,
98 };
99
100 static int af_packet_logtype;
101
102 #define PMD_LOG(level, fmt, args...) \
103 rte_log(RTE_LOG_ ## level, af_packet_logtype, \
104 "%s(): " fmt "\n", __func__, ##args)
105
106 static uint16_t
107 eth_af_packet_rx(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
108 {
109 unsigned i;
110 struct tpacket2_hdr *ppd;
111 struct rte_mbuf *mbuf;
112 uint8_t *pbuf;
113 struct pkt_rx_queue *pkt_q = queue;
114 uint16_t num_rx = 0;
115 unsigned long num_rx_bytes = 0;
116 unsigned int framecount, framenum;
117
118 if (unlikely(nb_pkts == 0))
119 return 0;
120
121 /*
122 * Reads the given number of packets from the AF_PACKET socket one by
123 * one and copies the packet data into a newly allocated mbuf.
124 */
125 framecount = pkt_q->framecount;
126 framenum = pkt_q->framenum;
127 for (i = 0; i < nb_pkts; i++) {
128 /* point at the next incoming frame */
129 ppd = (struct tpacket2_hdr *) pkt_q->rd[framenum].iov_base;
130 if ((ppd->tp_status & TP_STATUS_USER) == 0)
131 break;
132
133 /* allocate the next mbuf */
134 mbuf = rte_pktmbuf_alloc(pkt_q->mb_pool);
135 if (unlikely(mbuf == NULL))
136 break;
137
138 /* packet will fit in the mbuf, go ahead and receive it */
139 rte_pktmbuf_pkt_len(mbuf) = rte_pktmbuf_data_len(mbuf) = ppd->tp_snaplen;
140 pbuf = (uint8_t *) ppd + ppd->tp_mac;
141 memcpy(rte_pktmbuf_mtod(mbuf, void *), pbuf, rte_pktmbuf_data_len(mbuf));
142
143 /* check for vlan info */
144 if (ppd->tp_status & TP_STATUS_VLAN_VALID) {
145 mbuf->vlan_tci = ppd->tp_vlan_tci;
146 mbuf->ol_flags |= (PKT_RX_VLAN | PKT_RX_VLAN_STRIPPED);
147 }
148
149 /* release incoming frame and advance ring buffer */
150 ppd->tp_status = TP_STATUS_KERNEL;
151 if (++framenum >= framecount)
152 framenum = 0;
153 mbuf->port = pkt_q->in_port;
154
155 /* account for the receive frame */
156 bufs[i] = mbuf;
157 num_rx++;
158 num_rx_bytes += mbuf->pkt_len;
159 }
160 pkt_q->framenum = framenum;
161 pkt_q->rx_pkts += num_rx;
162 pkt_q->rx_bytes += num_rx_bytes;
163 return num_rx;
164 }
165
166 /*
167 * Callback to handle sending packets through a real NIC.
168 */
169 static uint16_t
170 eth_af_packet_tx(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
171 {
172 struct tpacket2_hdr *ppd;
173 struct rte_mbuf *mbuf;
174 uint8_t *pbuf;
175 unsigned int framecount, framenum;
176 struct pollfd pfd;
177 struct pkt_tx_queue *pkt_q = queue;
178 uint16_t num_tx = 0;
179 unsigned long num_tx_bytes = 0;
180 int i;
181
182 if (unlikely(nb_pkts == 0))
183 return 0;
184
185 memset(&pfd, 0, sizeof(pfd));
186 pfd.fd = pkt_q->sockfd;
187 pfd.events = POLLOUT;
188 pfd.revents = 0;
189
190 framecount = pkt_q->framecount;
191 framenum = pkt_q->framenum;
192 ppd = (struct tpacket2_hdr *) pkt_q->rd[framenum].iov_base;
193 for (i = 0; i < nb_pkts; i++) {
194 mbuf = *bufs++;
195
196 /* drop oversized packets */
197 if (mbuf->pkt_len > pkt_q->frame_data_size) {
198 rte_pktmbuf_free(mbuf);
199 continue;
200 }
201
202 /* insert vlan info if necessary */
203 if (mbuf->ol_flags & PKT_TX_VLAN_PKT) {
204 if (rte_vlan_insert(&mbuf)) {
205 rte_pktmbuf_free(mbuf);
206 continue;
207 }
208 }
209
210 /* point at the next incoming frame */
211 if ((ppd->tp_status != TP_STATUS_AVAILABLE) &&
212 (poll(&pfd, 1, -1) < 0))
213 break;
214
215 /* copy the tx frame data */
216 pbuf = (uint8_t *) ppd + TPACKET2_HDRLEN -
217 sizeof(struct sockaddr_ll);
218
219 struct rte_mbuf *tmp_mbuf = mbuf;
220 while (tmp_mbuf) {
221 uint16_t data_len = rte_pktmbuf_data_len(tmp_mbuf);
222 memcpy(pbuf, rte_pktmbuf_mtod(tmp_mbuf, void*), data_len);
223 pbuf += data_len;
224 tmp_mbuf = tmp_mbuf->next;
225 }
226
227 ppd->tp_len = mbuf->pkt_len;
228 ppd->tp_snaplen = mbuf->pkt_len;
229
230 /* release incoming frame and advance ring buffer */
231 ppd->tp_status = TP_STATUS_SEND_REQUEST;
232 if (++framenum >= framecount)
233 framenum = 0;
234 ppd = (struct tpacket2_hdr *) pkt_q->rd[framenum].iov_base;
235
236 num_tx++;
237 num_tx_bytes += mbuf->pkt_len;
238 rte_pktmbuf_free(mbuf);
239 }
240
241 /* kick-off transmits */
242 if (sendto(pkt_q->sockfd, NULL, 0, MSG_DONTWAIT, NULL, 0) == -1) {
243 /* error sending -- no packets transmitted */
244 num_tx = 0;
245 num_tx_bytes = 0;
246 }
247
248 pkt_q->framenum = framenum;
249 pkt_q->tx_pkts += num_tx;
250 pkt_q->err_pkts += i - num_tx;
251 pkt_q->tx_bytes += num_tx_bytes;
252 return i;
253 }
254
255 static int
256 eth_dev_start(struct rte_eth_dev *dev)
257 {
258 dev->data->dev_link.link_status = ETH_LINK_UP;
259 return 0;
260 }
261
262 /*
263 * This function gets called when the current port gets stopped.
264 */
265 static void
266 eth_dev_stop(struct rte_eth_dev *dev)
267 {
268 unsigned i;
269 int sockfd;
270 struct pmd_internals *internals = dev->data->dev_private;
271
272 for (i = 0; i < internals->nb_queues; i++) {
273 sockfd = internals->rx_queue[i].sockfd;
274 if (sockfd != -1)
275 close(sockfd);
276
277 /* Prevent use after free in case tx fd == rx fd */
278 if (sockfd != internals->tx_queue[i].sockfd) {
279 sockfd = internals->tx_queue[i].sockfd;
280 if (sockfd != -1)
281 close(sockfd);
282 }
283
284 internals->rx_queue[i].sockfd = -1;
285 internals->tx_queue[i].sockfd = -1;
286 }
287
288 dev->data->dev_link.link_status = ETH_LINK_DOWN;
289 }
290
291 static int
292 eth_dev_configure(struct rte_eth_dev *dev __rte_unused)
293 {
294 return 0;
295 }
296
297 static void
298 eth_dev_info(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
299 {
300 struct pmd_internals *internals = dev->data->dev_private;
301
302 dev_info->if_index = internals->if_index;
303 dev_info->max_mac_addrs = 1;
304 dev_info->max_rx_pktlen = (uint32_t)ETH_FRAME_LEN;
305 dev_info->max_rx_queues = (uint16_t)internals->nb_queues;
306 dev_info->max_tx_queues = (uint16_t)internals->nb_queues;
307 dev_info->min_rx_bufsize = 0;
308 dev_info->rx_offload_capa = DEV_RX_OFFLOAD_CRC_STRIP;
309 }
310
311 static int
312 eth_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *igb_stats)
313 {
314 unsigned i, imax;
315 unsigned long rx_total = 0, tx_total = 0, tx_err_total = 0;
316 unsigned long rx_bytes_total = 0, tx_bytes_total = 0;
317 const struct pmd_internals *internal = dev->data->dev_private;
318
319 imax = (internal->nb_queues < RTE_ETHDEV_QUEUE_STAT_CNTRS ?
320 internal->nb_queues : RTE_ETHDEV_QUEUE_STAT_CNTRS);
321 for (i = 0; i < imax; i++) {
322 igb_stats->q_ipackets[i] = internal->rx_queue[i].rx_pkts;
323 igb_stats->q_ibytes[i] = internal->rx_queue[i].rx_bytes;
324 rx_total += igb_stats->q_ipackets[i];
325 rx_bytes_total += igb_stats->q_ibytes[i];
326 }
327
328 imax = (internal->nb_queues < RTE_ETHDEV_QUEUE_STAT_CNTRS ?
329 internal->nb_queues : RTE_ETHDEV_QUEUE_STAT_CNTRS);
330 for (i = 0; i < imax; i++) {
331 igb_stats->q_opackets[i] = internal->tx_queue[i].tx_pkts;
332 igb_stats->q_errors[i] = internal->tx_queue[i].err_pkts;
333 igb_stats->q_obytes[i] = internal->tx_queue[i].tx_bytes;
334 tx_total += igb_stats->q_opackets[i];
335 tx_err_total += igb_stats->q_errors[i];
336 tx_bytes_total += igb_stats->q_obytes[i];
337 }
338
339 igb_stats->ipackets = rx_total;
340 igb_stats->ibytes = rx_bytes_total;
341 igb_stats->opackets = tx_total;
342 igb_stats->oerrors = tx_err_total;
343 igb_stats->obytes = tx_bytes_total;
344 return 0;
345 }
346
347 static void
348 eth_stats_reset(struct rte_eth_dev *dev)
349 {
350 unsigned i;
351 struct pmd_internals *internal = dev->data->dev_private;
352
353 for (i = 0; i < internal->nb_queues; i++) {
354 internal->rx_queue[i].rx_pkts = 0;
355 internal->rx_queue[i].rx_bytes = 0;
356 }
357
358 for (i = 0; i < internal->nb_queues; i++) {
359 internal->tx_queue[i].tx_pkts = 0;
360 internal->tx_queue[i].err_pkts = 0;
361 internal->tx_queue[i].tx_bytes = 0;
362 }
363 }
364
365 static void
366 eth_dev_close(struct rte_eth_dev *dev __rte_unused)
367 {
368 }
369
370 static void
371 eth_queue_release(void *q __rte_unused)
372 {
373 }
374
375 static int
376 eth_link_update(struct rte_eth_dev *dev __rte_unused,
377 int wait_to_complete __rte_unused)
378 {
379 return 0;
380 }
381
382 static int
383 eth_rx_queue_setup(struct rte_eth_dev *dev,
384 uint16_t rx_queue_id,
385 uint16_t nb_rx_desc __rte_unused,
386 unsigned int socket_id __rte_unused,
387 const struct rte_eth_rxconf *rx_conf __rte_unused,
388 struct rte_mempool *mb_pool)
389 {
390 struct pmd_internals *internals = dev->data->dev_private;
391 struct pkt_rx_queue *pkt_q = &internals->rx_queue[rx_queue_id];
392 unsigned int buf_size, data_size;
393
394 pkt_q->mb_pool = mb_pool;
395
396 /* Now get the space available for data in the mbuf */
397 buf_size = rte_pktmbuf_data_room_size(pkt_q->mb_pool) -
398 RTE_PKTMBUF_HEADROOM;
399 data_size = internals->req.tp_frame_size;
400 data_size -= TPACKET2_HDRLEN - sizeof(struct sockaddr_ll);
401
402 if (data_size > buf_size) {
403 PMD_LOG(ERR,
404 "%s: %d bytes will not fit in mbuf (%d bytes)",
405 dev->device->name, data_size, buf_size);
406 return -ENOMEM;
407 }
408
409 dev->data->rx_queues[rx_queue_id] = pkt_q;
410 pkt_q->in_port = dev->data->port_id;
411
412 return 0;
413 }
414
415 static int
416 eth_tx_queue_setup(struct rte_eth_dev *dev,
417 uint16_t tx_queue_id,
418 uint16_t nb_tx_desc __rte_unused,
419 unsigned int socket_id __rte_unused,
420 const struct rte_eth_txconf *tx_conf __rte_unused)
421 {
422
423 struct pmd_internals *internals = dev->data->dev_private;
424
425 dev->data->tx_queues[tx_queue_id] = &internals->tx_queue[tx_queue_id];
426 return 0;
427 }
428
429 static int
430 eth_dev_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
431 {
432 struct pmd_internals *internals = dev->data->dev_private;
433 struct ifreq ifr = { .ifr_mtu = mtu };
434 int ret;
435 int s;
436 unsigned int data_size = internals->req.tp_frame_size -
437 TPACKET2_HDRLEN -
438 sizeof(struct sockaddr_ll);
439
440 if (mtu > data_size)
441 return -EINVAL;
442
443 s = socket(PF_INET, SOCK_DGRAM, 0);
444 if (s < 0)
445 return -EINVAL;
446
447 snprintf(ifr.ifr_name, IFNAMSIZ, "%s", internals->if_name);
448 ret = ioctl(s, SIOCSIFMTU, &ifr);
449 close(s);
450
451 if (ret < 0)
452 return -EINVAL;
453
454 return 0;
455 }
456
457 static void
458 eth_dev_change_flags(char *if_name, uint32_t flags, uint32_t mask)
459 {
460 struct ifreq ifr;
461 int s;
462
463 s = socket(PF_INET, SOCK_DGRAM, 0);
464 if (s < 0)
465 return;
466
467 snprintf(ifr.ifr_name, IFNAMSIZ, "%s", if_name);
468 if (ioctl(s, SIOCGIFFLAGS, &ifr) < 0)
469 goto out;
470 ifr.ifr_flags &= mask;
471 ifr.ifr_flags |= flags;
472 if (ioctl(s, SIOCSIFFLAGS, &ifr) < 0)
473 goto out;
474 out:
475 close(s);
476 }
477
478 static void
479 eth_dev_promiscuous_enable(struct rte_eth_dev *dev)
480 {
481 struct pmd_internals *internals = dev->data->dev_private;
482
483 eth_dev_change_flags(internals->if_name, IFF_PROMISC, ~0);
484 }
485
486 static void
487 eth_dev_promiscuous_disable(struct rte_eth_dev *dev)
488 {
489 struct pmd_internals *internals = dev->data->dev_private;
490
491 eth_dev_change_flags(internals->if_name, 0, ~IFF_PROMISC);
492 }
493
494 static const struct eth_dev_ops ops = {
495 .dev_start = eth_dev_start,
496 .dev_stop = eth_dev_stop,
497 .dev_close = eth_dev_close,
498 .dev_configure = eth_dev_configure,
499 .dev_infos_get = eth_dev_info,
500 .mtu_set = eth_dev_mtu_set,
501 .promiscuous_enable = eth_dev_promiscuous_enable,
502 .promiscuous_disable = eth_dev_promiscuous_disable,
503 .rx_queue_setup = eth_rx_queue_setup,
504 .tx_queue_setup = eth_tx_queue_setup,
505 .rx_queue_release = eth_queue_release,
506 .tx_queue_release = eth_queue_release,
507 .link_update = eth_link_update,
508 .stats_get = eth_stats_get,
509 .stats_reset = eth_stats_reset,
510 };
511
512 /*
513 * Opens an AF_PACKET socket
514 */
515 static int
516 open_packet_iface(const char *key __rte_unused,
517 const char *value __rte_unused,
518 void *extra_args)
519 {
520 int *sockfd = extra_args;
521
522 /* Open an AF_PACKET socket... */
523 *sockfd = socket(AF_PACKET, SOCK_RAW, htons(ETH_P_ALL));
524 if (*sockfd == -1) {
525 PMD_LOG(ERR, "Could not open AF_PACKET socket");
526 return -1;
527 }
528
529 return 0;
530 }
531
532 static struct rte_vdev_driver pmd_af_packet_drv;
533
534 static int
535 rte_pmd_init_internals(struct rte_vdev_device *dev,
536 const int sockfd,
537 const unsigned nb_queues,
538 unsigned int blocksize,
539 unsigned int blockcnt,
540 unsigned int framesize,
541 unsigned int framecnt,
542 unsigned int qdisc_bypass,
543 struct pmd_internals **internals,
544 struct rte_eth_dev **eth_dev,
545 struct rte_kvargs *kvlist)
546 {
547 const char *name = rte_vdev_device_name(dev);
548 const unsigned int numa_node = dev->device.numa_node;
549 struct rte_eth_dev_data *data = NULL;
550 struct rte_kvargs_pair *pair = NULL;
551 struct ifreq ifr;
552 size_t ifnamelen;
553 unsigned k_idx;
554 struct sockaddr_ll sockaddr;
555 struct tpacket_req *req;
556 struct pkt_rx_queue *rx_queue;
557 struct pkt_tx_queue *tx_queue;
558 int rc, tpver, discard;
559 int qsockfd = -1;
560 unsigned int i, q, rdsize;
561 #if defined(PACKET_FANOUT)
562 int fanout_arg;
563 #endif
564
565 for (k_idx = 0; k_idx < kvlist->count; k_idx++) {
566 pair = &kvlist->pairs[k_idx];
567 if (strstr(pair->key, ETH_AF_PACKET_IFACE_ARG) != NULL)
568 break;
569 }
570 if (pair == NULL) {
571 PMD_LOG(ERR,
572 "%s: no interface specified for AF_PACKET ethdev",
573 name);
574 return -1;
575 }
576
577 PMD_LOG(INFO,
578 "%s: creating AF_PACKET-backed ethdev on numa socket %u",
579 name, numa_node);
580
581 *internals = rte_zmalloc_socket(name, sizeof(**internals),
582 0, numa_node);
583 if (*internals == NULL)
584 return -1;
585
586 for (q = 0; q < nb_queues; q++) {
587 (*internals)->rx_queue[q].map = MAP_FAILED;
588 (*internals)->tx_queue[q].map = MAP_FAILED;
589 }
590
591 req = &((*internals)->req);
592
593 req->tp_block_size = blocksize;
594 req->tp_block_nr = blockcnt;
595 req->tp_frame_size = framesize;
596 req->tp_frame_nr = framecnt;
597
598 ifnamelen = strlen(pair->value);
599 if (ifnamelen < sizeof(ifr.ifr_name)) {
600 memcpy(ifr.ifr_name, pair->value, ifnamelen);
601 ifr.ifr_name[ifnamelen] = '\0';
602 } else {
603 PMD_LOG(ERR,
604 "%s: I/F name too long (%s)",
605 name, pair->value);
606 return -1;
607 }
608 if (ioctl(sockfd, SIOCGIFINDEX, &ifr) == -1) {
609 PMD_LOG(ERR,
610 "%s: ioctl failed (SIOCGIFINDEX)",
611 name);
612 return -1;
613 }
614 (*internals)->if_name = strdup(pair->value);
615 if ((*internals)->if_name == NULL)
616 return -1;
617 (*internals)->if_index = ifr.ifr_ifindex;
618
619 if (ioctl(sockfd, SIOCGIFHWADDR, &ifr) == -1) {
620 PMD_LOG(ERR,
621 "%s: ioctl failed (SIOCGIFHWADDR)",
622 name);
623 return -1;
624 }
625 memcpy(&(*internals)->eth_addr, ifr.ifr_hwaddr.sa_data, ETH_ALEN);
626
627 memset(&sockaddr, 0, sizeof(sockaddr));
628 sockaddr.sll_family = AF_PACKET;
629 sockaddr.sll_protocol = htons(ETH_P_ALL);
630 sockaddr.sll_ifindex = (*internals)->if_index;
631
632 #if defined(PACKET_FANOUT)
633 fanout_arg = (getpid() ^ (*internals)->if_index) & 0xffff;
634 fanout_arg |= (PACKET_FANOUT_HASH | PACKET_FANOUT_FLAG_DEFRAG) << 16;
635 #if defined(PACKET_FANOUT_FLAG_ROLLOVER)
636 fanout_arg |= PACKET_FANOUT_FLAG_ROLLOVER << 16;
637 #endif
638 #endif
639
640 for (q = 0; q < nb_queues; q++) {
641 /* Open an AF_PACKET socket for this queue... */
642 qsockfd = socket(AF_PACKET, SOCK_RAW, htons(ETH_P_ALL));
643 if (qsockfd == -1) {
644 PMD_LOG(ERR,
645 "%s: could not open AF_PACKET socket",
646 name);
647 return -1;
648 }
649
650 tpver = TPACKET_V2;
651 rc = setsockopt(qsockfd, SOL_PACKET, PACKET_VERSION,
652 &tpver, sizeof(tpver));
653 if (rc == -1) {
654 PMD_LOG(ERR,
655 "%s: could not set PACKET_VERSION on AF_PACKET socket for %s",
656 name, pair->value);
657 goto error;
658 }
659
660 discard = 1;
661 rc = setsockopt(qsockfd, SOL_PACKET, PACKET_LOSS,
662 &discard, sizeof(discard));
663 if (rc == -1) {
664 PMD_LOG(ERR,
665 "%s: could not set PACKET_LOSS on AF_PACKET socket for %s",
666 name, pair->value);
667 goto error;
668 }
669
670 #if defined(PACKET_QDISC_BYPASS)
671 rc = setsockopt(qsockfd, SOL_PACKET, PACKET_QDISC_BYPASS,
672 &qdisc_bypass, sizeof(qdisc_bypass));
673 if (rc == -1) {
674 PMD_LOG(ERR,
675 "%s: could not set PACKET_QDISC_BYPASS on AF_PACKET socket for %s",
676 name, pair->value);
677 goto error;
678 }
679 #else
680 RTE_SET_USED(qdisc_bypass);
681 #endif
682
683 rc = setsockopt(qsockfd, SOL_PACKET, PACKET_RX_RING, req, sizeof(*req));
684 if (rc == -1) {
685 PMD_LOG(ERR,
686 "%s: could not set PACKET_RX_RING on AF_PACKET socket for %s",
687 name, pair->value);
688 goto error;
689 }
690
691 rc = setsockopt(qsockfd, SOL_PACKET, PACKET_TX_RING, req, sizeof(*req));
692 if (rc == -1) {
693 PMD_LOG(ERR,
694 "%s: could not set PACKET_TX_RING on AF_PACKET "
695 "socket for %s", name, pair->value);
696 goto error;
697 }
698
699 rx_queue = &((*internals)->rx_queue[q]);
700 rx_queue->framecount = req->tp_frame_nr;
701
702 rx_queue->map = mmap(NULL, 2 * req->tp_block_size * req->tp_block_nr,
703 PROT_READ | PROT_WRITE, MAP_SHARED | MAP_LOCKED,
704 qsockfd, 0);
705 if (rx_queue->map == MAP_FAILED) {
706 PMD_LOG(ERR,
707 "%s: call to mmap failed on AF_PACKET socket for %s",
708 name, pair->value);
709 goto error;
710 }
711
712 /* rdsize is same for both Tx and Rx */
713 rdsize = req->tp_frame_nr * sizeof(*(rx_queue->rd));
714
715 rx_queue->rd = rte_zmalloc_socket(name, rdsize, 0, numa_node);
716 if (rx_queue->rd == NULL)
717 goto error;
718 for (i = 0; i < req->tp_frame_nr; ++i) {
719 rx_queue->rd[i].iov_base = rx_queue->map + (i * framesize);
720 rx_queue->rd[i].iov_len = req->tp_frame_size;
721 }
722 rx_queue->sockfd = qsockfd;
723
724 tx_queue = &((*internals)->tx_queue[q]);
725 tx_queue->framecount = req->tp_frame_nr;
726 tx_queue->frame_data_size = req->tp_frame_size;
727 tx_queue->frame_data_size -= TPACKET2_HDRLEN -
728 sizeof(struct sockaddr_ll);
729
730 tx_queue->map = rx_queue->map + req->tp_block_size * req->tp_block_nr;
731
732 tx_queue->rd = rte_zmalloc_socket(name, rdsize, 0, numa_node);
733 if (tx_queue->rd == NULL)
734 goto error;
735 for (i = 0; i < req->tp_frame_nr; ++i) {
736 tx_queue->rd[i].iov_base = tx_queue->map + (i * framesize);
737 tx_queue->rd[i].iov_len = req->tp_frame_size;
738 }
739 tx_queue->sockfd = qsockfd;
740
741 rc = bind(qsockfd, (const struct sockaddr*)&sockaddr, sizeof(sockaddr));
742 if (rc == -1) {
743 PMD_LOG(ERR,
744 "%s: could not bind AF_PACKET socket to %s",
745 name, pair->value);
746 goto error;
747 }
748
749 #if defined(PACKET_FANOUT)
750 rc = setsockopt(qsockfd, SOL_PACKET, PACKET_FANOUT,
751 &fanout_arg, sizeof(fanout_arg));
752 if (rc == -1) {
753 PMD_LOG(ERR,
754 "%s: could not set PACKET_FANOUT on AF_PACKET socket "
755 "for %s", name, pair->value);
756 goto error;
757 }
758 #endif
759 }
760
761 /* reserve an ethdev entry */
762 *eth_dev = rte_eth_vdev_allocate(dev, 0);
763 if (*eth_dev == NULL)
764 goto error;
765
766 /*
767 * now put it all together
768 * - store queue data in internals,
769 * - store numa_node in eth_dev
770 * - point eth_dev_data to internals
771 * - and point eth_dev structure to new eth_dev_data structure
772 */
773
774 (*internals)->nb_queues = nb_queues;
775
776 data = (*eth_dev)->data;
777 data->dev_private = *internals;
778 data->nb_rx_queues = (uint16_t)nb_queues;
779 data->nb_tx_queues = (uint16_t)nb_queues;
780 data->dev_link = pmd_link;
781 data->mac_addrs = &(*internals)->eth_addr;
782
783 (*eth_dev)->dev_ops = &ops;
784
785 return 0;
786
787 error:
788 if (qsockfd != -1)
789 close(qsockfd);
790 for (q = 0; q < nb_queues; q++) {
791 munmap((*internals)->rx_queue[q].map,
792 2 * req->tp_block_size * req->tp_block_nr);
793
794 rte_free((*internals)->rx_queue[q].rd);
795 rte_free((*internals)->tx_queue[q].rd);
796 if (((*internals)->rx_queue[q].sockfd != 0) &&
797 ((*internals)->rx_queue[q].sockfd != qsockfd))
798 close((*internals)->rx_queue[q].sockfd);
799 }
800 free((*internals)->if_name);
801 rte_free(*internals);
802 return -1;
803 }
804
805 static int
806 rte_eth_from_packet(struct rte_vdev_device *dev,
807 int const *sockfd,
808 struct rte_kvargs *kvlist)
809 {
810 const char *name = rte_vdev_device_name(dev);
811 struct pmd_internals *internals = NULL;
812 struct rte_eth_dev *eth_dev = NULL;
813 struct rte_kvargs_pair *pair = NULL;
814 unsigned k_idx;
815 unsigned int blockcount;
816 unsigned int blocksize = DFLT_BLOCK_SIZE;
817 unsigned int framesize = DFLT_FRAME_SIZE;
818 unsigned int framecount = DFLT_FRAME_COUNT;
819 unsigned int qpairs = 1;
820 unsigned int qdisc_bypass = 1;
821
822 /* do some parameter checking */
823 if (*sockfd < 0)
824 return -1;
825
826 /*
827 * Walk arguments for configurable settings
828 */
829 for (k_idx = 0; k_idx < kvlist->count; k_idx++) {
830 pair = &kvlist->pairs[k_idx];
831 if (strstr(pair->key, ETH_AF_PACKET_NUM_Q_ARG) != NULL) {
832 qpairs = atoi(pair->value);
833 if (qpairs < 1 ||
834 qpairs > RTE_PMD_AF_PACKET_MAX_RINGS) {
835 PMD_LOG(ERR,
836 "%s: invalid qpairs value",
837 name);
838 return -1;
839 }
840 continue;
841 }
842 if (strstr(pair->key, ETH_AF_PACKET_BLOCKSIZE_ARG) != NULL) {
843 blocksize = atoi(pair->value);
844 if (!blocksize) {
845 PMD_LOG(ERR,
846 "%s: invalid blocksize value",
847 name);
848 return -1;
849 }
850 continue;
851 }
852 if (strstr(pair->key, ETH_AF_PACKET_FRAMESIZE_ARG) != NULL) {
853 framesize = atoi(pair->value);
854 if (!framesize) {
855 PMD_LOG(ERR,
856 "%s: invalid framesize value",
857 name);
858 return -1;
859 }
860 continue;
861 }
862 if (strstr(pair->key, ETH_AF_PACKET_FRAMECOUNT_ARG) != NULL) {
863 framecount = atoi(pair->value);
864 if (!framecount) {
865 PMD_LOG(ERR,
866 "%s: invalid framecount value",
867 name);
868 return -1;
869 }
870 continue;
871 }
872 if (strstr(pair->key, ETH_AF_PACKET_QDISC_BYPASS_ARG) != NULL) {
873 qdisc_bypass = atoi(pair->value);
874 if (qdisc_bypass > 1) {
875 PMD_LOG(ERR,
876 "%s: invalid bypass value",
877 name);
878 return -1;
879 }
880 continue;
881 }
882 }
883
884 if (framesize > blocksize) {
885 PMD_LOG(ERR,
886 "%s: AF_PACKET MMAP frame size exceeds block size!",
887 name);
888 return -1;
889 }
890
891 blockcount = framecount / (blocksize / framesize);
892 if (!blockcount) {
893 PMD_LOG(ERR,
894 "%s: invalid AF_PACKET MMAP parameters", name);
895 return -1;
896 }
897
898 PMD_LOG(INFO, "%s: AF_PACKET MMAP parameters:", name);
899 PMD_LOG(INFO, "%s:\tblock size %d", name, blocksize);
900 PMD_LOG(INFO, "%s:\tblock count %d", name, blockcount);
901 PMD_LOG(INFO, "%s:\tframe size %d", name, framesize);
902 PMD_LOG(INFO, "%s:\tframe count %d", name, framecount);
903
904 if (rte_pmd_init_internals(dev, *sockfd, qpairs,
905 blocksize, blockcount,
906 framesize, framecount,
907 qdisc_bypass,
908 &internals, &eth_dev,
909 kvlist) < 0)
910 return -1;
911
912 eth_dev->rx_pkt_burst = eth_af_packet_rx;
913 eth_dev->tx_pkt_burst = eth_af_packet_tx;
914
915 rte_eth_dev_probing_finish(eth_dev);
916 return 0;
917 }
918
919 static int
920 rte_pmd_af_packet_probe(struct rte_vdev_device *dev)
921 {
922 int ret = 0;
923 struct rte_kvargs *kvlist;
924 int sockfd = -1;
925 struct rte_eth_dev *eth_dev;
926 const char *name = rte_vdev_device_name(dev);
927
928 PMD_LOG(INFO, "Initializing pmd_af_packet for %s", name);
929
930 if (rte_eal_process_type() == RTE_PROC_SECONDARY &&
931 strlen(rte_vdev_device_args(dev)) == 0) {
932 eth_dev = rte_eth_dev_attach_secondary(name);
933 if (!eth_dev) {
934 PMD_LOG(ERR, "Failed to probe %s", name);
935 return -1;
936 }
937 /* TODO: request info from primary to set up Rx and Tx */
938 eth_dev->dev_ops = &ops;
939 eth_dev->device = &dev->device;
940 rte_eth_dev_probing_finish(eth_dev);
941 return 0;
942 }
943
944 kvlist = rte_kvargs_parse(rte_vdev_device_args(dev), valid_arguments);
945 if (kvlist == NULL) {
946 ret = -1;
947 goto exit;
948 }
949
950 /*
951 * If iface argument is passed we open the NICs and use them for
952 * reading / writing
953 */
954 if (rte_kvargs_count(kvlist, ETH_AF_PACKET_IFACE_ARG) == 1) {
955
956 ret = rte_kvargs_process(kvlist, ETH_AF_PACKET_IFACE_ARG,
957 &open_packet_iface, &sockfd);
958 if (ret < 0)
959 goto exit;
960 }
961
962 if (dev->device.numa_node == SOCKET_ID_ANY)
963 dev->device.numa_node = rte_socket_id();
964
965 ret = rte_eth_from_packet(dev, &sockfd, kvlist);
966 close(sockfd); /* no longer needed */
967
968 exit:
969 rte_kvargs_free(kvlist);
970 return ret;
971 }
972
973 static int
974 rte_pmd_af_packet_remove(struct rte_vdev_device *dev)
975 {
976 struct rte_eth_dev *eth_dev = NULL;
977 struct pmd_internals *internals;
978 unsigned q;
979
980 PMD_LOG(INFO, "Closing AF_PACKET ethdev on numa socket %u",
981 rte_socket_id());
982
983 if (dev == NULL)
984 return -1;
985
986 /* find the ethdev entry */
987 eth_dev = rte_eth_dev_allocated(rte_vdev_device_name(dev));
988 if (eth_dev == NULL)
989 return -1;
990
991 internals = eth_dev->data->dev_private;
992 for (q = 0; q < internals->nb_queues; q++) {
993 rte_free(internals->rx_queue[q].rd);
994 rte_free(internals->tx_queue[q].rd);
995 }
996 free(internals->if_name);
997
998 rte_free(eth_dev->data->dev_private);
999
1000 rte_eth_dev_release_port(eth_dev);
1001
1002 return 0;
1003 }
1004
1005 static struct rte_vdev_driver pmd_af_packet_drv = {
1006 .probe = rte_pmd_af_packet_probe,
1007 .remove = rte_pmd_af_packet_remove,
1008 };
1009
1010 RTE_PMD_REGISTER_VDEV(net_af_packet, pmd_af_packet_drv);
1011 RTE_PMD_REGISTER_ALIAS(net_af_packet, eth_af_packet);
1012 RTE_PMD_REGISTER_PARAM_STRING(net_af_packet,
1013 "iface=<string> "
1014 "qpairs=<int> "
1015 "blocksz=<int> "
1016 "framesz=<int> "
1017 "framecnt=<int> "
1018 "qdisc_bypass=<0|1>");
1019
1020 RTE_INIT(af_packet_init_log)
1021 {
1022 af_packet_logtype = rte_log_register("pmd.net.packet");
1023 if (af_packet_logtype >= 0)
1024 rte_log_set_level(af_packet_logtype, RTE_LOG_NOTICE);
1025 }
Ceph