Home | History | Annotate | Download | only in net
      1 /*
      2  * Copyright 2013 Red Hat, Inc.
      3  * Author: Daniel Borkmann <dborkman (at) redhat.com>
      4  *         Chetan Loke <loke.chetan (at) gmail.com> (TPACKET_V3 usage example)
      5  *
      6  * A basic test of packet socket's TPACKET_V1/TPACKET_V2/TPACKET_V3 behavior.
      7  *
      8  * Control:
      9  *   Test the setup of the TPACKET socket with different patterns that are
     10  *   known to fail (TODO) resp. succeed (OK).
     11  *
     12  * Datapath:
     13  *   Open a pair of packet sockets and send resp. receive an a priori known
     14  *   packet pattern accross the sockets and check if it was received resp.
     15  *   sent correctly. Fanout in combination with RX_RING is currently not
     16  *   tested here.
     17  *
     18  *   The test currently runs for
     19  *   - TPACKET_V1: RX_RING, TX_RING
     20  *   - TPACKET_V2: RX_RING, TX_RING
     21  *   - TPACKET_V3: RX_RING
     22  *
     23  * License (GPLv2):
     24  *
     25  * This program is free software; you can redistribute it and/or modify it
     26  * under the terms and conditions of the GNU General Public License,
     27  * version 2, as published by the Free Software Foundation.
     28  *
     29  * This program is distributed in the hope it will be useful, but WITHOUT
     30  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
     31  * FITNESS FOR A PARTICULAR PURPOSE. * See the GNU General Public License for
     32  * more details.
     33  *
     34  * You should have received a copy of the GNU General Public License along with
     35  * this program; if not, write to the Free Software Foundation, Inc.,
     36  * 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
     37  */
     38 
     39 #include <stdio.h>
     40 #include <stdlib.h>
     41 #include <sys/types.h>
     42 #include <sys/stat.h>
     43 #include <sys/socket.h>
     44 #include <sys/mman.h>
     45 #include <sys/utsname.h>
     46 #include <linux/if_packet.h>
     47 #include <linux/filter.h>
     48 #include <ctype.h>
     49 #include <fcntl.h>
     50 #include <unistd.h>
     51 #ifndef __ANDROID__
     52 #include <bits/wordsize.h>
     53 #endif
     54 #include <net/ethernet.h>
     55 #include <netinet/ip.h>
     56 #include <arpa/inet.h>
     57 #include <stdint.h>
     58 #include <string.h>
     59 #include <assert.h>
     60 #include <net/if.h>
     61 #include <inttypes.h>
     62 #include <poll.h>
     63 
     64 #include "psock_lib.h"
     65 
     66 #include "../kselftest.h"
     67 
     68 #ifndef bug_on
     69 # define bug_on(cond)		assert(!(cond))
     70 #endif
     71 
     72 #ifndef __aligned_tpacket
     73 # define __aligned_tpacket	__attribute__((aligned(TPACKET_ALIGNMENT)))
     74 #endif
     75 
     76 #ifndef __align_tpacket
     77 # define __align_tpacket(x)	__attribute__((aligned(TPACKET_ALIGN(x))))
     78 #endif
     79 
     80 #define NUM_PACKETS		100
     81 #define ALIGN_8(x)		(((x) + 8 - 1) & ~(8 - 1))
     82 
     83 struct ring {
     84 	struct iovec *rd;
     85 	uint8_t *mm_space;
     86 	size_t mm_len, rd_len;
     87 	struct sockaddr_ll ll;
     88 	void (*walk)(int sock, struct ring *ring);
     89 	int type, rd_num, flen, version;
     90 	union {
     91 		struct tpacket_req  req;
     92 		struct tpacket_req3 req3;
     93 	};
     94 };
     95 
     96 struct block_desc {
     97 	uint32_t version;
     98 	uint32_t offset_to_priv;
     99 	struct tpacket_hdr_v1 h1;
    100 };
    101 
    102 union frame_map {
    103 	struct {
    104 		struct tpacket_hdr tp_h __aligned_tpacket;
    105 		struct sockaddr_ll s_ll __align_tpacket(sizeof(struct tpacket_hdr));
    106 	} *v1;
    107 	struct {
    108 		struct tpacket2_hdr tp_h __aligned_tpacket;
    109 		struct sockaddr_ll s_ll __align_tpacket(sizeof(struct tpacket2_hdr));
    110 	} *v2;
    111 	void *raw;
    112 };
    113 
    114 static unsigned int total_packets, total_bytes;
    115 
    116 static int pfsocket(int ver)
    117 {
    118 	int ret, sock = socket(PF_PACKET, SOCK_RAW, 0);
    119 	if (sock == -1) {
    120 		perror("socket");
    121 		exit(1);
    122 	}
    123 
    124 	ret = setsockopt(sock, SOL_PACKET, PACKET_VERSION, &ver, sizeof(ver));
    125 	if (ret == -1) {
    126 		perror("setsockopt");
    127 		exit(1);
    128 	}
    129 
    130 	return sock;
    131 }
    132 
    133 static void status_bar_update(void)
    134 {
    135 	if (total_packets % 10 == 0) {
    136 		fprintf(stderr, ".");
    137 		fflush(stderr);
    138 	}
    139 }
    140 
    141 static void test_payload(void *pay, size_t len)
    142 {
    143 	struct ethhdr *eth = pay;
    144 
    145 	if (len < sizeof(struct ethhdr)) {
    146 		fprintf(stderr, "test_payload: packet too "
    147 			"small: %zu bytes!\n", len);
    148 		exit(1);
    149 	}
    150 
    151 	if (eth->h_proto != htons(ETH_P_IP)) {
    152 		fprintf(stderr, "test_payload: wrong ethernet "
    153 			"type: 0x%x!\n", ntohs(eth->h_proto));
    154 		exit(1);
    155 	}
    156 }
    157 
    158 static void create_payload(void *pay, size_t *len)
    159 {
    160 	int i;
    161 	struct ethhdr *eth = pay;
    162 	struct iphdr *ip = pay + sizeof(*eth);
    163 
    164 	/* Lets create some broken crap, that still passes
    165 	 * our BPF filter.
    166 	 */
    167 
    168 	*len = DATA_LEN + 42;
    169 
    170 	memset(pay, 0xff, ETH_ALEN * 2);
    171 	eth->h_proto = htons(ETH_P_IP);
    172 
    173 	for (i = 0; i < sizeof(*ip); ++i)
    174 		((uint8_t *) pay)[i + sizeof(*eth)] = (uint8_t) rand();
    175 
    176 	ip->ihl = 5;
    177 	ip->version = 4;
    178 	ip->protocol = 0x11;
    179 	ip->frag_off = 0;
    180 	ip->ttl = 64;
    181 	ip->tot_len = htons((uint16_t) *len - sizeof(*eth));
    182 
    183 	ip->saddr = htonl(INADDR_LOOPBACK);
    184 	ip->daddr = htonl(INADDR_LOOPBACK);
    185 
    186 	memset(pay + sizeof(*eth) + sizeof(*ip),
    187 	       DATA_CHAR, DATA_LEN);
    188 }
    189 
    190 static inline int __v1_rx_kernel_ready(struct tpacket_hdr *hdr)
    191 {
    192 	return ((hdr->tp_status & TP_STATUS_USER) == TP_STATUS_USER);
    193 }
    194 
    195 static inline void __v1_rx_user_ready(struct tpacket_hdr *hdr)
    196 {
    197 	hdr->tp_status = TP_STATUS_KERNEL;
    198 	__sync_synchronize();
    199 }
    200 
    201 static inline int __v2_rx_kernel_ready(struct tpacket2_hdr *hdr)
    202 {
    203 	return ((hdr->tp_status & TP_STATUS_USER) == TP_STATUS_USER);
    204 }
    205 
    206 static inline void __v2_rx_user_ready(struct tpacket2_hdr *hdr)
    207 {
    208 	hdr->tp_status = TP_STATUS_KERNEL;
    209 	__sync_synchronize();
    210 }
    211 
    212 static inline int __v1_v2_rx_kernel_ready(void *base, int version)
    213 {
    214 	switch (version) {
    215 	case TPACKET_V1:
    216 		return __v1_rx_kernel_ready(base);
    217 	case TPACKET_V2:
    218 		return __v2_rx_kernel_ready(base);
    219 	default:
    220 		bug_on(1);
    221 		return 0;
    222 	}
    223 }
    224 
    225 static inline void __v1_v2_rx_user_ready(void *base, int version)
    226 {
    227 	switch (version) {
    228 	case TPACKET_V1:
    229 		__v1_rx_user_ready(base);
    230 		break;
    231 	case TPACKET_V2:
    232 		__v2_rx_user_ready(base);
    233 		break;
    234 	}
    235 }
    236 
    237 static void walk_v1_v2_rx(int sock, struct ring *ring)
    238 {
    239 	struct pollfd pfd;
    240 	int udp_sock[2];
    241 	union frame_map ppd;
    242 	unsigned int frame_num = 0;
    243 
    244 	bug_on(ring->type != PACKET_RX_RING);
    245 
    246 	pair_udp_open(udp_sock, PORT_BASE);
    247 
    248 	memset(&pfd, 0, sizeof(pfd));
    249 	pfd.fd = sock;
    250 	pfd.events = POLLIN | POLLERR;
    251 	pfd.revents = 0;
    252 
    253 	pair_udp_send(udp_sock, NUM_PACKETS);
    254 
    255 	while (total_packets < NUM_PACKETS * 2) {
    256 		while (__v1_v2_rx_kernel_ready(ring->rd[frame_num].iov_base,
    257 					       ring->version)) {
    258 			ppd.raw = ring->rd[frame_num].iov_base;
    259 
    260 			switch (ring->version) {
    261 			case TPACKET_V1:
    262 				test_payload((uint8_t *) ppd.raw + ppd.v1->tp_h.tp_mac,
    263 					     ppd.v1->tp_h.tp_snaplen);
    264 				total_bytes += ppd.v1->tp_h.tp_snaplen;
    265 				break;
    266 
    267 			case TPACKET_V2:
    268 				test_payload((uint8_t *) ppd.raw + ppd.v2->tp_h.tp_mac,
    269 					     ppd.v2->tp_h.tp_snaplen);
    270 				total_bytes += ppd.v2->tp_h.tp_snaplen;
    271 				break;
    272 			}
    273 
    274 			status_bar_update();
    275 			total_packets++;
    276 
    277 			__v1_v2_rx_user_ready(ppd.raw, ring->version);
    278 
    279 			frame_num = (frame_num + 1) % ring->rd_num;
    280 		}
    281 
    282 		poll(&pfd, 1, 1);
    283 	}
    284 
    285 	pair_udp_close(udp_sock);
    286 
    287 	if (total_packets != 2 * NUM_PACKETS) {
    288 		fprintf(stderr, "walk_v%d_rx: received %u out of %u pkts\n",
    289 			ring->version, total_packets, NUM_PACKETS);
    290 		exit(1);
    291 	}
    292 
    293 	fprintf(stderr, " %u pkts (%u bytes)", NUM_PACKETS, total_bytes >> 1);
    294 }
    295 
    296 static inline int __v1_tx_kernel_ready(struct tpacket_hdr *hdr)
    297 {
    298 	return !(hdr->tp_status & (TP_STATUS_SEND_REQUEST | TP_STATUS_SENDING));
    299 }
    300 
    301 static inline void __v1_tx_user_ready(struct tpacket_hdr *hdr)
    302 {
    303 	hdr->tp_status = TP_STATUS_SEND_REQUEST;
    304 	__sync_synchronize();
    305 }
    306 
    307 static inline int __v2_tx_kernel_ready(struct tpacket2_hdr *hdr)
    308 {
    309 	return !(hdr->tp_status & (TP_STATUS_SEND_REQUEST | TP_STATUS_SENDING));
    310 }
    311 
    312 static inline void __v2_tx_user_ready(struct tpacket2_hdr *hdr)
    313 {
    314 	hdr->tp_status = TP_STATUS_SEND_REQUEST;
    315 	__sync_synchronize();
    316 }
    317 
    318 static inline int __v3_tx_kernel_ready(struct tpacket3_hdr *hdr)
    319 {
    320 	return !(hdr->tp_status & (TP_STATUS_SEND_REQUEST | TP_STATUS_SENDING));
    321 }
    322 
    323 static inline void __v3_tx_user_ready(struct tpacket3_hdr *hdr)
    324 {
    325 	hdr->tp_status = TP_STATUS_SEND_REQUEST;
    326 	__sync_synchronize();
    327 }
    328 
    329 static inline int __tx_kernel_ready(void *base, int version)
    330 {
    331 	switch (version) {
    332 	case TPACKET_V1:
    333 		return __v1_tx_kernel_ready(base);
    334 	case TPACKET_V2:
    335 		return __v2_tx_kernel_ready(base);
    336 	case TPACKET_V3:
    337 		return __v3_tx_kernel_ready(base);
    338 	default:
    339 		bug_on(1);
    340 		return 0;
    341 	}
    342 }
    343 
    344 static inline void __tx_user_ready(void *base, int version)
    345 {
    346 	switch (version) {
    347 	case TPACKET_V1:
    348 		__v1_tx_user_ready(base);
    349 		break;
    350 	case TPACKET_V2:
    351 		__v2_tx_user_ready(base);
    352 		break;
    353 	case TPACKET_V3:
    354 		__v3_tx_user_ready(base);
    355 		break;
    356 	}
    357 }
    358 
    359 static void __v1_v2_set_packet_loss_discard(int sock)
    360 {
    361 	int ret, discard = 1;
    362 
    363 	ret = setsockopt(sock, SOL_PACKET, PACKET_LOSS, (void *) &discard,
    364 			 sizeof(discard));
    365 	if (ret == -1) {
    366 		perror("setsockopt");
    367 		exit(1);
    368 	}
    369 }
    370 
    371 static inline void *get_next_frame(struct ring *ring, int n)
    372 {
    373 	uint8_t *f0 = ring->rd[0].iov_base;
    374 
    375 	switch (ring->version) {
    376 	case TPACKET_V1:
    377 	case TPACKET_V2:
    378 		return ring->rd[n].iov_base;
    379 	case TPACKET_V3:
    380 		return f0 + (n * ring->req3.tp_frame_size);
    381 	default:
    382 		bug_on(1);
    383 		return NULL;
    384 	}
    385 }
    386 
    387 static void walk_tx(int sock, struct ring *ring)
    388 {
    389 	struct pollfd pfd;
    390 	int rcv_sock, ret;
    391 	size_t packet_len;
    392 	union frame_map ppd;
    393 	char packet[1024];
    394 	unsigned int frame_num = 0, got = 0;
    395 	struct sockaddr_ll ll = {
    396 		.sll_family = PF_PACKET,
    397 		.sll_halen = ETH_ALEN,
    398 	};
    399 	int nframes;
    400 
    401 	/* TPACKET_V{1,2} sets up the ring->rd* related variables based
    402 	 * on frames (e.g., rd_num is tp_frame_nr) whereas V3 sets these
    403 	 * up based on blocks (e.g, rd_num is  tp_block_nr)
    404 	 */
    405 	if (ring->version <= TPACKET_V2)
    406 		nframes = ring->rd_num;
    407 	else
    408 		nframes = ring->req3.tp_frame_nr;
    409 
    410 	bug_on(ring->type != PACKET_TX_RING);
    411 	bug_on(nframes < NUM_PACKETS);
    412 
    413 	rcv_sock = socket(PF_PACKET, SOCK_RAW, htons(ETH_P_ALL));
    414 	if (rcv_sock == -1) {
    415 		perror("socket");
    416 		exit(1);
    417 	}
    418 
    419 	pair_udp_setfilter(rcv_sock);
    420 
    421 	ll.sll_ifindex = if_nametoindex("lo");
    422 	ret = bind(rcv_sock, (struct sockaddr *) &ll, sizeof(ll));
    423 	if (ret == -1) {
    424 		perror("bind");
    425 		exit(1);
    426 	}
    427 
    428 	memset(&pfd, 0, sizeof(pfd));
    429 	pfd.fd = sock;
    430 	pfd.events = POLLOUT | POLLERR;
    431 	pfd.revents = 0;
    432 
    433 	total_packets = NUM_PACKETS;
    434 	create_payload(packet, &packet_len);
    435 
    436 	while (total_packets > 0) {
    437 		void *next = get_next_frame(ring, frame_num);
    438 
    439 		while (__tx_kernel_ready(next, ring->version) &&
    440 		       total_packets > 0) {
    441 			ppd.raw = next;
    442 
    443 			switch (ring->version) {
    444 			case TPACKET_V1:
    445 				ppd.v1->tp_h.tp_snaplen = packet_len;
    446 				ppd.v1->tp_h.tp_len = packet_len;
    447 
    448 				memcpy((uint8_t *) ppd.raw + TPACKET_HDRLEN -
    449 				       sizeof(struct sockaddr_ll), packet,
    450 				       packet_len);
    451 				total_bytes += ppd.v1->tp_h.tp_snaplen;
    452 				break;
    453 
    454 			case TPACKET_V2:
    455 				ppd.v2->tp_h.tp_snaplen = packet_len;
    456 				ppd.v2->tp_h.tp_len = packet_len;
    457 
    458 				memcpy((uint8_t *) ppd.raw + TPACKET2_HDRLEN -
    459 				       sizeof(struct sockaddr_ll), packet,
    460 				       packet_len);
    461 				total_bytes += ppd.v2->tp_h.tp_snaplen;
    462 				break;
    463 			case TPACKET_V3: {
    464 				struct tpacket3_hdr *tx = next;
    465 
    466 				tx->tp_snaplen = packet_len;
    467 				tx->tp_len = packet_len;
    468 				tx->tp_next_offset = 0;
    469 
    470 				memcpy((uint8_t *)tx + TPACKET3_HDRLEN -
    471 				       sizeof(struct sockaddr_ll), packet,
    472 				       packet_len);
    473 				total_bytes += tx->tp_snaplen;
    474 				break;
    475 			}
    476 			}
    477 
    478 			status_bar_update();
    479 			total_packets--;
    480 
    481 			__tx_user_ready(next, ring->version);
    482 
    483 			frame_num = (frame_num + 1) % nframes;
    484 		}
    485 
    486 		poll(&pfd, 1, 1);
    487 	}
    488 
    489 	bug_on(total_packets != 0);
    490 
    491 	ret = sendto(sock, NULL, 0, 0, NULL, 0);
    492 	if (ret == -1) {
    493 		perror("sendto");
    494 		exit(1);
    495 	}
    496 
    497 	while ((ret = recvfrom(rcv_sock, packet, sizeof(packet),
    498 			       0, NULL, NULL)) > 0 &&
    499 	       total_packets < NUM_PACKETS) {
    500 		got += ret;
    501 		test_payload(packet, ret);
    502 
    503 		status_bar_update();
    504 		total_packets++;
    505 	}
    506 
    507 	close(rcv_sock);
    508 
    509 	if (total_packets != NUM_PACKETS) {
    510 		fprintf(stderr, "walk_v%d_rx: received %u out of %u pkts\n",
    511 			ring->version, total_packets, NUM_PACKETS);
    512 		exit(1);
    513 	}
    514 
    515 	fprintf(stderr, " %u pkts (%u bytes)", NUM_PACKETS, got);
    516 }
    517 
    518 static void walk_v1_v2(int sock, struct ring *ring)
    519 {
    520 	if (ring->type == PACKET_RX_RING)
    521 		walk_v1_v2_rx(sock, ring);
    522 	else
    523 		walk_tx(sock, ring);
    524 }
    525 
    526 static uint64_t __v3_prev_block_seq_num = 0;
    527 
    528 void __v3_test_block_seq_num(struct block_desc *pbd)
    529 {
    530 	if (__v3_prev_block_seq_num + 1 != pbd->h1.seq_num) {
    531 		fprintf(stderr, "\nprev_block_seq_num:%"PRIu64", expected "
    532 			"seq:%"PRIu64" != actual seq:%"PRIu64"\n",
    533 			__v3_prev_block_seq_num, __v3_prev_block_seq_num + 1,
    534 			(uint64_t) pbd->h1.seq_num);
    535 		exit(1);
    536 	}
    537 
    538 	__v3_prev_block_seq_num = pbd->h1.seq_num;
    539 }
    540 
    541 static void __v3_test_block_len(struct block_desc *pbd, uint32_t bytes, int block_num)
    542 {
    543 	if (pbd->h1.num_pkts && bytes != pbd->h1.blk_len) {
    544 		fprintf(stderr, "\nblock:%u with %upackets, expected "
    545 			"len:%u != actual len:%u\n", block_num,
    546 			pbd->h1.num_pkts, bytes, pbd->h1.blk_len);
    547 		exit(1);
    548 	}
    549 }
    550 
    551 static void __v3_test_block_header(struct block_desc *pbd, const int block_num)
    552 {
    553 	if ((pbd->h1.block_status & TP_STATUS_USER) == 0) {
    554 		fprintf(stderr, "\nblock %u: not in TP_STATUS_USER\n", block_num);
    555 		exit(1);
    556 	}
    557 
    558 	__v3_test_block_seq_num(pbd);
    559 }
    560 
    561 static void __v3_walk_block(struct block_desc *pbd, const int block_num)
    562 {
    563 	int num_pkts = pbd->h1.num_pkts, i;
    564 	unsigned long bytes = 0, bytes_with_padding = ALIGN_8(sizeof(*pbd));
    565 	struct tpacket3_hdr *ppd;
    566 
    567 	__v3_test_block_header(pbd, block_num);
    568 
    569 	ppd = (struct tpacket3_hdr *) ((uint8_t *) pbd +
    570 				       pbd->h1.offset_to_first_pkt);
    571 
    572 	for (i = 0; i < num_pkts; ++i) {
    573 		bytes += ppd->tp_snaplen;
    574 
    575 		if (ppd->tp_next_offset)
    576 			bytes_with_padding += ppd->tp_next_offset;
    577 		else
    578 			bytes_with_padding += ALIGN_8(ppd->tp_snaplen + ppd->tp_mac);
    579 
    580 		test_payload((uint8_t *) ppd + ppd->tp_mac, ppd->tp_snaplen);
    581 
    582 		status_bar_update();
    583 		total_packets++;
    584 
    585 		ppd = (struct tpacket3_hdr *) ((uint8_t *) ppd + ppd->tp_next_offset);
    586 		__sync_synchronize();
    587 	}
    588 
    589 	__v3_test_block_len(pbd, bytes_with_padding, block_num);
    590 	total_bytes += bytes;
    591 }
    592 
    593 void __v3_flush_block(struct block_desc *pbd)
    594 {
    595 	pbd->h1.block_status = TP_STATUS_KERNEL;
    596 	__sync_synchronize();
    597 }
    598 
    599 static void walk_v3_rx(int sock, struct ring *ring)
    600 {
    601 	unsigned int block_num = 0;
    602 	struct pollfd pfd;
    603 	struct block_desc *pbd;
    604 	int udp_sock[2];
    605 
    606 	bug_on(ring->type != PACKET_RX_RING);
    607 
    608 	pair_udp_open(udp_sock, PORT_BASE);
    609 
    610 	memset(&pfd, 0, sizeof(pfd));
    611 	pfd.fd = sock;
    612 	pfd.events = POLLIN | POLLERR;
    613 	pfd.revents = 0;
    614 
    615 	pair_udp_send(udp_sock, NUM_PACKETS);
    616 
    617 	while (total_packets < NUM_PACKETS * 2) {
    618 		pbd = (struct block_desc *) ring->rd[block_num].iov_base;
    619 
    620 		while ((pbd->h1.block_status & TP_STATUS_USER) == 0)
    621 			poll(&pfd, 1, 1);
    622 
    623 		__v3_walk_block(pbd, block_num);
    624 		__v3_flush_block(pbd);
    625 
    626 		block_num = (block_num + 1) % ring->rd_num;
    627 	}
    628 
    629 	pair_udp_close(udp_sock);
    630 
    631 	if (total_packets != 2 * NUM_PACKETS) {
    632 		fprintf(stderr, "walk_v3_rx: received %u out of %u pkts\n",
    633 			total_packets, NUM_PACKETS);
    634 		exit(1);
    635 	}
    636 
    637 	fprintf(stderr, " %u pkts (%u bytes)", NUM_PACKETS, total_bytes >> 1);
    638 }
    639 
    640 static void walk_v3(int sock, struct ring *ring)
    641 {
    642 	if (ring->type == PACKET_RX_RING)
    643 		walk_v3_rx(sock, ring);
    644 	else
    645 		walk_tx(sock, ring);
    646 }
    647 
    648 static void __v1_v2_fill(struct ring *ring, unsigned int blocks)
    649 {
    650 	ring->req.tp_block_size = getpagesize() << 2;
    651 	ring->req.tp_frame_size = TPACKET_ALIGNMENT << 7;
    652 	ring->req.tp_block_nr = blocks;
    653 
    654 	ring->req.tp_frame_nr = ring->req.tp_block_size /
    655 				ring->req.tp_frame_size *
    656 				ring->req.tp_block_nr;
    657 
    658 	ring->mm_len = ring->req.tp_block_size * ring->req.tp_block_nr;
    659 	ring->walk = walk_v1_v2;
    660 	ring->rd_num = ring->req.tp_frame_nr;
    661 	ring->flen = ring->req.tp_frame_size;
    662 }
    663 
    664 static void __v3_fill(struct ring *ring, unsigned int blocks, int type)
    665 {
    666 	if (type == PACKET_RX_RING) {
    667 		ring->req3.tp_retire_blk_tov = 64;
    668 		ring->req3.tp_sizeof_priv = 0;
    669 		ring->req3.tp_feature_req_word = TP_FT_REQ_FILL_RXHASH;
    670 	}
    671 	ring->req3.tp_block_size = getpagesize() << 2;
    672 	ring->req3.tp_frame_size = TPACKET_ALIGNMENT << 7;
    673 	ring->req3.tp_block_nr = blocks;
    674 
    675 	ring->req3.tp_frame_nr = ring->req3.tp_block_size /
    676 				 ring->req3.tp_frame_size *
    677 				 ring->req3.tp_block_nr;
    678 
    679 	ring->mm_len = ring->req3.tp_block_size * ring->req3.tp_block_nr;
    680 	ring->walk = walk_v3;
    681 	ring->rd_num = ring->req3.tp_block_nr;
    682 	ring->flen = ring->req3.tp_block_size;
    683 }
    684 
    685 static void setup_ring(int sock, struct ring *ring, int version, int type)
    686 {
    687 	int ret = 0;
    688 	unsigned int blocks = 256;
    689 
    690 	ring->type = type;
    691 	ring->version = version;
    692 
    693 	switch (version) {
    694 	case TPACKET_V1:
    695 	case TPACKET_V2:
    696 		if (type == PACKET_TX_RING)
    697 			__v1_v2_set_packet_loss_discard(sock);
    698 		__v1_v2_fill(ring, blocks);
    699 		ret = setsockopt(sock, SOL_PACKET, type, &ring->req,
    700 				 sizeof(ring->req));
    701 		break;
    702 
    703 	case TPACKET_V3:
    704 		__v3_fill(ring, blocks, type);
    705 		ret = setsockopt(sock, SOL_PACKET, type, &ring->req3,
    706 				 sizeof(ring->req3));
    707 		break;
    708 	}
    709 
    710 	if (ret == -1) {
    711 		perror("setsockopt");
    712 		exit(1);
    713 	}
    714 
    715 	ring->rd_len = ring->rd_num * sizeof(*ring->rd);
    716 	ring->rd = malloc(ring->rd_len);
    717 	if (ring->rd == NULL) {
    718 		perror("malloc");
    719 		exit(1);
    720 	}
    721 
    722 	total_packets = 0;
    723 	total_bytes = 0;
    724 }
    725 
    726 static void mmap_ring(int sock, struct ring *ring)
    727 {
    728 	int i;
    729 
    730 	ring->mm_space = mmap(0, ring->mm_len, PROT_READ | PROT_WRITE,
    731 			      MAP_SHARED | MAP_LOCKED | MAP_POPULATE, sock, 0);
    732 	if (ring->mm_space == MAP_FAILED) {
    733 		perror("mmap");
    734 		exit(1);
    735 	}
    736 
    737 	memset(ring->rd, 0, ring->rd_len);
    738 	for (i = 0; i < ring->rd_num; ++i) {
    739 		ring->rd[i].iov_base = ring->mm_space + (i * ring->flen);
    740 		ring->rd[i].iov_len = ring->flen;
    741 	}
    742 }
    743 
    744 static void bind_ring(int sock, struct ring *ring)
    745 {
    746 	int ret;
    747 
    748 	pair_udp_setfilter(sock);
    749 
    750 	ring->ll.sll_family = PF_PACKET;
    751 	ring->ll.sll_protocol = htons(ETH_P_ALL);
    752 	ring->ll.sll_ifindex = if_nametoindex("lo");
    753 	ring->ll.sll_hatype = 0;
    754 	ring->ll.sll_pkttype = 0;
    755 	ring->ll.sll_halen = 0;
    756 
    757 	ret = bind(sock, (struct sockaddr *) &ring->ll, sizeof(ring->ll));
    758 	if (ret == -1) {
    759 		perror("bind");
    760 		exit(1);
    761 	}
    762 }
    763 
    764 static void walk_ring(int sock, struct ring *ring)
    765 {
    766 	ring->walk(sock, ring);
    767 }
    768 
    769 static void unmap_ring(int sock, struct ring *ring)
    770 {
    771 	munmap(ring->mm_space, ring->mm_len);
    772 	free(ring->rd);
    773 }
    774 
    775 static int test_kernel_bit_width(void)
    776 {
    777 	char in[512], *ptr;
    778 	int num = 0, fd;
    779 	ssize_t ret;
    780 
    781 	fd = open("/proc/kallsyms", O_RDONLY);
    782 	if (fd == -1) {
    783 		perror("open");
    784 		exit(1);
    785 	}
    786 
    787 	ret = read(fd, in, sizeof(in));
    788 	if (ret <= 0) {
    789 		perror("read");
    790 		exit(1);
    791 	}
    792 
    793 	close(fd);
    794 
    795 	ptr = in;
    796 	while(!isspace(*ptr)) {
    797 		num++;
    798 		ptr++;
    799 	}
    800 
    801 	return num * 4;
    802 }
    803 
    804 static int test_user_bit_width(void)
    805 {
    806 	return __WORDSIZE;
    807 }
    808 
    809 static const char *tpacket_str[] = {
    810 	[TPACKET_V1] = "TPACKET_V1",
    811 	[TPACKET_V2] = "TPACKET_V2",
    812 	[TPACKET_V3] = "TPACKET_V3",
    813 };
    814 
    815 static const char *type_str[] = {
    816 	[PACKET_RX_RING] = "PACKET_RX_RING",
    817 	[PACKET_TX_RING] = "PACKET_TX_RING",
    818 };
    819 
    820 static int test_tpacket(int version, int type)
    821 {
    822 	int sock;
    823 	struct ring ring;
    824 
    825 	fprintf(stderr, "test: %s with %s ", tpacket_str[version],
    826 		type_str[type]);
    827 	fflush(stderr);
    828 
    829 	if (version == TPACKET_V1 &&
    830 	    test_kernel_bit_width() != test_user_bit_width()) {
    831 		fprintf(stderr, "test: skip %s %s since user and kernel "
    832 			"space have different bit width\n",
    833 			tpacket_str[version], type_str[type]);
    834 		return KSFT_SKIP;
    835 	}
    836 
    837 	sock = pfsocket(version);
    838 	memset(&ring, 0, sizeof(ring));
    839 	setup_ring(sock, &ring, version, type);
    840 	mmap_ring(sock, &ring);
    841 	bind_ring(sock, &ring);
    842 	walk_ring(sock, &ring);
    843 	unmap_ring(sock, &ring);
    844 	close(sock);
    845 
    846 	fprintf(stderr, "\n");
    847 	return 0;
    848 }
    849 
    850 void get_kernel_version(int *version, int *patchlevel)
    851 {
    852 	int ret, sublevel;
    853 	struct utsname utsname;
    854 
    855 	ret = uname(&utsname);
    856 	if (ret) {
    857 		perror("uname");
    858 		exit(1);
    859 	}
    860 
    861 	ret = sscanf(utsname.release, "%d.%d.%d", version, patchlevel,
    862 		     &sublevel);
    863 	if (ret < 0) {
    864 		perror("sscanf");
    865 		exit(1);
    866 	} else if (ret != 3) {
    867 		printf("Malformed kernel version %s\n", &utsname.release);
    868 		exit(1);
    869 	}
    870 }
    871 
    872 int main(void)
    873 {
    874 	int ret = 0;
    875 	int version, patchlevel;
    876 
    877 	get_kernel_version(&version, &patchlevel);
    878 
    879 	ret |= test_tpacket(TPACKET_V1, PACKET_RX_RING);
    880 	ret |= test_tpacket(TPACKET_V1, PACKET_TX_RING);
    881 
    882 	ret |= test_tpacket(TPACKET_V2, PACKET_RX_RING);
    883 	ret |= test_tpacket(TPACKET_V2, PACKET_TX_RING);
    884 
    885 	ret |= test_tpacket(TPACKET_V3, PACKET_RX_RING);
    886 	if (version > 4 || (version == 4 && patchlevel >= 11))
    887 		ret |= test_tpacket(TPACKET_V3, PACKET_TX_RING);
    888 
    889 	if (ret)
    890 		return 1;
    891 
    892 	printf("OK. All tests passed\n");
    893 	return 0;
    894 }
    895