xref: /linux/tools/testing/selftests/net/lib/gro.c (revision 6a20b34fe3b31b292078bc79ec18a2ab0d9f7719)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * This testsuite provides conformance testing for GRO coalescing.
4  *
5  * Test cases:
6  *
7  * data_*:
8  *  Data packets of the same size and same header setup with correct
9  *  sequence numbers coalesce. The one exception being the last data
10  *  packet coalesced: it can be smaller than the rest and coalesced
11  *  as long as it is in the same flow.
12  *   - data_same:    same size packets coalesce
13  *   - data_lrg_sml:   large then small coalesces
14  *   - data_lrg_1byte: large then 1 byte coalesces (Ethernet padding)
15  *   - data_sml_lrg:   small then large doesn't coalesce
16  *   - data_burst:   two bursts of two, separated by 100ms
17  *
18  * ack:
19  *  Pure ACK does not coalesce.
20  *
21  * flags_*:
22  *  No packets with PSH, SYN, URG, RST, CWR set will be coalesced.
23  *   - flags_psh, flags_syn, flags_rst, flags_urg, flags_cwr
24  *
25  * tcp_*:
26  *  Packets with incorrect checksum, non-consecutive seqno and
27  *  different TCP header options shouldn't coalesce. Nit: given that
28  *  some extension headers have paddings, such as timestamp, headers
29  *  that are padded differently would not be coalesced.
30  *   - tcp_csum: incorrect checksum
31  *   - tcp_seq:  non-consecutive sequence numbers
32  *   - tcp_ts:   different timestamps
33  *   - tcp_opt:  different TCP options
34  *
35  * ip_*:
36  *  Packets with different (ECN, TTL, TOS) header, IP options or
37  *  IP fragments shouldn't coalesce.
38  *   - ip_ecn, ip_tos:            shared between IPv4/IPv6
39  *   - ip_csum:                   IPv4 only, bad IP header checksum
40  *   - ip_ttl, ip_opt, ip_frag4:  IPv4 only
41  *   - ip_id_df*:                 IPv4 IP ID field coalescing tests
42  *   - ip_frag6, ip_v6ext_*:      IPv6 only
43  *
44  * large_*:
45  *  Packets larger than GRO_MAX_SIZE packets shouldn't coalesce.
46  *   - large_max: exceeding max size
47  *   - large_rem: remainder handling
48  *
49  * single, capacity:
50  *  Boring cases used to test coalescing machinery itself and stats
51  *  more than protocol behavior.
52  *
53  * MSS is defined as 4096 - header because if it is too small
54  * (i.e. 1500 MTU - header), it will result in many packets,
55  * increasing the "large" test case's flakiness. This is because
56  * due to time sensitivity in the coalescing window, the receiver
57  * may not coalesce all of the packets.
58  *
59  * Note the timing issue applies to all of the test cases, so some
60  * flakiness is to be expected.
61  *
62  */
63 
64 #define _GNU_SOURCE
65 
66 #include <arpa/inet.h>
67 #include <errno.h>
68 #include <error.h>
69 #include <getopt.h>
70 #include <net/ethernet.h>
71 #include <net/if.h>
72 #include <linux/filter.h>
73 #include <linux/if_packet.h>
74 #include <linux/if_pppox.h>
75 #include <linux/ipv6.h>
76 #include <linux/net_tstamp.h>
77 #include <linux/ppp_defs.h>
78 #include <netinet/in.h>
79 #include <netinet/ip.h>
80 #include <netinet/ip6.h>
81 #include <netinet/tcp.h>
82 #include <stdbool.h>
83 #include <stddef.h>
84 #include <stdio.h>
85 #include <stdarg.h>
86 #include <string.h>
87 #include <time.h>
88 #include <unistd.h>
89 
90 #include "kselftest.h"
91 #include "ksft.h"
92 
93 #define DPORT 8000
94 #define SPORT 1500
95 #define PAYLOAD_LEN 100
96 #define NUM_PACKETS 4
97 #define START_SEQ 100
98 #define START_ACK 100
99 #define ETH_P_NONE 0
100 #define ASSUMED_MTU 4096
101 #define MAX_MSS (ASSUMED_MTU - sizeof(struct iphdr) - sizeof(struct tcphdr))
102 #define MAX_HDR_LEN \
103 	(ETH_HLEN + sizeof(struct ipv6hdr) * 2 + sizeof(struct tcphdr))
104 #define MAX_LARGE_PKT_CNT ((IP_MAXPACKET - (MAX_HDR_LEN - ETH_HLEN)) /	\
105 			   (ASSUMED_MTU - (MAX_HDR_LEN - ETH_HLEN)))
106 #define MIN_EXTHDR_SIZE 8
107 #define L2_HLEN_MAX	(ETH_HLEN + PPPOE_SES_HLEN)
108 #define EXT_PAYLOAD_1 "\x00\x00\x00\x00\x00\x00"
109 #define EXT_PAYLOAD_2 "\x11\x11\x11\x11\x11\x11"
110 
111 #define ipv6_optlen(p)  (((p)->hdrlen+1) << 3) /* calculate IPv6 extension header len */
112 #define BUILD_BUG_ON(condition) ((void)sizeof(char[1 - 2*!!(condition)]))
113 
114 enum flush_id_case {
115 	FLUSH_ID_DF1_INC,
116 	FLUSH_ID_DF1_FIXED,
117 	FLUSH_ID_DF0_INC,
118 	FLUSH_ID_DF0_FIXED,
119 	FLUSH_ID_DF1_INC_FIXED,
120 	FLUSH_ID_DF1_FIXED_INC,
121 };
122 
123 static const char *addr6_src = "fdaa::2";
124 static const char *addr6_dst = "fdaa::1";
125 static const char *addr4_src = "192.168.1.200";
126 static const char *addr4_dst = "192.168.1.100";
127 static int proto = -1;
128 static uint8_t src_mac[ETH_ALEN], dst_mac[ETH_ALEN];
129 static char *testname = "data";
130 static char *ifname = "eth0";
131 static char *smac = "aa:00:00:00:00:02";
132 static char *dmac = "aa:00:00:00:00:01";
133 static bool verbose;
134 static bool tx_socket = true;
135 static int tcp_offset = -1;
136 static int total_hdr_len = -1;
137 static int ethhdr_proto = -1;
138 static bool ipip;
139 static bool ip6ip6;
140 static bool pppoe;
141 static uint64_t txtime_ns;
142 static int num_flows = 4;
143 static bool order_check;
144 
145 #define CAPACITY_PAYLOAD_LEN 200
146 
147 #define TXTIME_DELAY_MS 5
148 
149 /* Max TCP payload that GRO will coalesce. The outer header overhead
150  * varies by encapsulation, reducing the effective max payload.
151  */
152 static int max_payload(void)
153 {
154 	return IP_MAXPACKET - (total_hdr_len - ETH_HLEN);
155 }
156 
157 static int calc_mss(void)
158 {
159 	return ASSUMED_MTU - (total_hdr_len - ETH_HLEN);
160 }
161 
162 static int num_large_pkt(void)
163 {
164 	return max_payload() / calc_mss();
165 }
166 
167 static void vlog(const char *fmt, ...)
168 {
169 	va_list args;
170 
171 	if (verbose) {
172 		va_start(args, fmt);
173 		vfprintf(stderr, fmt, args);
174 		va_end(args);
175 	}
176 }
177 
178 static void fill_pppoelayer(void *buf, int payload_len, uint16_t sid)
179 {
180 	struct pppoe_ppp_hdr {
181 		struct pppoe_hdr eh;
182 		__be16 proto;
183 	} *ph = buf;
184 
185 	payload_len += sizeof(struct tcphdr);
186 	ph->eh.type = 1;
187 	ph->eh.ver = 1;
188 	ph->eh.code = 0;
189 	ph->eh.sid = htons(sid);
190 	ph->eh.length = htons(payload_len + sizeof(ph->proto));
191 	ph->proto = htons(proto == PF_INET ? PPP_IP : PPP_IPV6);
192 }
193 
194 static void setup_sock_filter(int fd)
195 {
196 	const int dport_off = tcp_offset + offsetof(struct tcphdr, dest);
197 	const int ethproto_off = offsetof(struct ethhdr, h_proto);
198 	int optlen = 0;
199 	int ipproto_off, opt_ipproto_off;
200 
201 	if (proto == PF_INET)
202 		ipproto_off = tcp_offset - sizeof(struct iphdr) +
203 			      offsetof(struct iphdr, protocol);
204 	else
205 		ipproto_off = tcp_offset - sizeof(struct ipv6hdr) +
206 			      offsetof(struct ipv6hdr, nexthdr);
207 
208 	/* Overridden later if exthdrs are used: */
209 	opt_ipproto_off = ipproto_off;
210 
211 	if (strcmp(testname, "ip_opt") == 0) {
212 		optlen = sizeof(struct ip_timestamp);
213 	} else if (strcmp(testname, "ip_frag6") == 0 ||
214 		   strcmp(testname, "ip_v6ext_same") == 0 ||
215 		   strcmp(testname, "ip_v6ext_diff") == 0) {
216 		BUILD_BUG_ON(sizeof(struct ip6_hbh) > MIN_EXTHDR_SIZE);
217 		BUILD_BUG_ON(sizeof(struct ip6_dest) > MIN_EXTHDR_SIZE);
218 		BUILD_BUG_ON(sizeof(struct ip6_frag) > MIN_EXTHDR_SIZE);
219 
220 		/* same size for HBH and Fragment extension header types */
221 		optlen = MIN_EXTHDR_SIZE;
222 		opt_ipproto_off = ETH_HLEN + sizeof(struct ipv6hdr)
223 			+ offsetof(struct ip6_ext, ip6e_nxt);
224 	}
225 
226 	/* this filter validates the following:
227 	 *	- packet is IPv4/IPv6 according to the running test.
228 	 *	- packet is TCP. Also handles the case of one extension header and then TCP.
229 	 *	- checks the packet tcp dport equals to DPORT. Also handles the case of one
230 	 *	  extension header and then TCP.
231 	 */
232 	struct sock_filter filter[] = {
233 			BPF_STMT(BPF_LD  + BPF_H   + BPF_ABS, ethproto_off),
234 			BPF_JUMP(BPF_JMP + BPF_JEQ + BPF_K, ntohs(ethhdr_proto), 0, 9),
235 			BPF_STMT(BPF_LD  + BPF_B   + BPF_ABS, ipproto_off),
236 			BPF_JUMP(BPF_JMP + BPF_JEQ + BPF_K, IPPROTO_TCP, 2, 0),
237 			BPF_STMT(BPF_LD  + BPF_B   + BPF_ABS, opt_ipproto_off),
238 			BPF_JUMP(BPF_JMP + BPF_JEQ + BPF_K, IPPROTO_TCP, 0, 5),
239 			BPF_STMT(BPF_LD  + BPF_H   + BPF_ABS, dport_off),
240 			BPF_JUMP(BPF_JMP + BPF_JEQ + BPF_K, DPORT, 2, 0),
241 			BPF_STMT(BPF_LD  + BPF_H   + BPF_ABS, dport_off + optlen),
242 			BPF_JUMP(BPF_JMP + BPF_JEQ + BPF_K, DPORT, 0, 1),
243 			BPF_STMT(BPF_RET + BPF_K, 0xFFFFFFFF),
244 			BPF_STMT(BPF_RET + BPF_K, 0),
245 	};
246 
247 	struct sock_fprog bpf = {
248 		.len = ARRAY_SIZE(filter),
249 		.filter = filter,
250 	};
251 
252 	if (setsockopt(fd, SOL_SOCKET, SO_ATTACH_FILTER, &bpf, sizeof(bpf)) < 0)
253 		error(1, errno, "error setting filter");
254 }
255 
256 static uint32_t checksum_nofold(void *data, size_t len, uint32_t sum)
257 {
258 	uint16_t *words = data;
259 	int i;
260 
261 	for (i = 0; i < len / 2; i++)
262 		sum += words[i];
263 	if (len & 1)
264 		sum += ((char *)data)[len - 1];
265 	return sum;
266 }
267 
268 static uint16_t checksum_fold(void *data, size_t len, uint32_t sum)
269 {
270 	sum = checksum_nofold(data, len, sum);
271 	while (sum > 0xFFFF)
272 		sum = (sum & 0xFFFF) + (sum >> 16);
273 	return ~sum;
274 }
275 
276 static uint16_t tcp_checksum(void *buf, int payload_len)
277 {
278 	struct pseudo_header6 {
279 		struct in6_addr saddr;
280 		struct in6_addr daddr;
281 		uint16_t protocol;
282 		uint16_t payload_len;
283 	} ph6;
284 	struct pseudo_header4 {
285 		struct in_addr saddr;
286 		struct in_addr daddr;
287 		uint16_t protocol;
288 		uint16_t payload_len;
289 	} ph4;
290 	uint32_t sum = 0;
291 
292 	if (proto == PF_INET6) {
293 		if (inet_pton(AF_INET6, addr6_src, &ph6.saddr) != 1)
294 			error(1, errno, "inet_pton6 source ip pseudo");
295 		if (inet_pton(AF_INET6, addr6_dst, &ph6.daddr) != 1)
296 			error(1, errno, "inet_pton6 dest ip pseudo");
297 		ph6.protocol = htons(IPPROTO_TCP);
298 		ph6.payload_len = htons(sizeof(struct tcphdr) + payload_len);
299 
300 		sum = checksum_nofold(&ph6, sizeof(ph6), 0);
301 	} else if (proto == PF_INET) {
302 		if (inet_pton(AF_INET, addr4_src, &ph4.saddr) != 1)
303 			error(1, errno, "inet_pton source ip pseudo");
304 		if (inet_pton(AF_INET, addr4_dst, &ph4.daddr) != 1)
305 			error(1, errno, "inet_pton dest ip pseudo");
306 		ph4.protocol = htons(IPPROTO_TCP);
307 		ph4.payload_len = htons(sizeof(struct tcphdr) + payload_len);
308 
309 		sum = checksum_nofold(&ph4, sizeof(ph4), 0);
310 	}
311 
312 	return checksum_fold(buf, sizeof(struct tcphdr) + payload_len, sum);
313 }
314 
315 static void read_MAC(uint8_t *mac_addr, char *mac)
316 {
317 	if (sscanf(mac, "%hhx:%hhx:%hhx:%hhx:%hhx:%hhx",
318 		   &mac_addr[0], &mac_addr[1], &mac_addr[2],
319 		   &mac_addr[3], &mac_addr[4], &mac_addr[5]) != 6)
320 		error(1, 0, "sscanf");
321 }
322 
323 static void fill_datalinklayer(void *buf)
324 {
325 	struct ethhdr *eth = buf;
326 
327 	memcpy(eth->h_dest, dst_mac, ETH_ALEN);
328 	memcpy(eth->h_source, src_mac, ETH_ALEN);
329 	eth->h_proto = ethhdr_proto;
330 }
331 
332 static void fill_networklayer(void *buf, int payload_len, int protocol)
333 {
334 	struct ipv6hdr *ip6h = buf;
335 	struct iphdr *iph = buf;
336 
337 	if (proto == PF_INET6) {
338 		memset(ip6h, 0, sizeof(*ip6h));
339 
340 		ip6h->version = 6;
341 		ip6h->payload_len = htons(sizeof(struct tcphdr) + payload_len);
342 		ip6h->nexthdr = protocol;
343 		ip6h->hop_limit = 8;
344 		if (inet_pton(AF_INET6, addr6_src, &ip6h->saddr) != 1)
345 			error(1, errno, "inet_pton source ip6");
346 		if (inet_pton(AF_INET6, addr6_dst, &ip6h->daddr) != 1)
347 			error(1, errno, "inet_pton dest ip6");
348 	} else if (proto == PF_INET) {
349 		memset(iph, 0, sizeof(*iph));
350 
351 		iph->version = 4;
352 		iph->ihl = 5;
353 		iph->ttl = 8;
354 		iph->protocol	= protocol;
355 		iph->tot_len = htons(sizeof(struct tcphdr) +
356 				payload_len + sizeof(struct iphdr));
357 		iph->frag_off = htons(0x4000); /* DF = 1, MF = 0 */
358 		if (inet_pton(AF_INET, addr4_src, &iph->saddr) != 1)
359 			error(1, errno, "inet_pton source ip");
360 		if (inet_pton(AF_INET, addr4_dst, &iph->daddr) != 1)
361 			error(1, errno, "inet_pton dest ip");
362 		iph->check = checksum_fold(buf, sizeof(struct iphdr), 0);
363 	}
364 }
365 
366 static void fill_transportlayer(void *buf, int seq_offset, int ack_offset,
367 				int payload_len, int fin)
368 {
369 	struct tcphdr *tcph = buf;
370 
371 	memset(tcph, 0, sizeof(*tcph));
372 
373 	tcph->source = htons(SPORT);
374 	tcph->dest = htons(DPORT);
375 	tcph->seq = ntohl(START_SEQ + seq_offset);
376 	tcph->ack_seq = ntohl(START_ACK + ack_offset);
377 	tcph->ack = 1;
378 	tcph->fin = fin;
379 	tcph->doff = 5;
380 	tcph->window = htons(TCP_MAXWIN);
381 	tcph->urg_ptr = 0;
382 	tcph->check = tcp_checksum(tcph, payload_len);
383 }
384 
385 static void write_packet(int fd, char *buf, int len, struct sockaddr_ll *daddr)
386 {
387 	char control[CMSG_SPACE(sizeof(uint64_t))];
388 	struct msghdr msg = {};
389 	struct iovec iov = {};
390 	struct cmsghdr *cm;
391 	int ret = -1;
392 
393 	iov.iov_base = buf;
394 	iov.iov_len = len;
395 
396 	msg.msg_iov = &iov;
397 	msg.msg_iovlen = 1;
398 	msg.msg_name = daddr;
399 	msg.msg_namelen = sizeof(*daddr);
400 
401 	if (txtime_ns) {
402 		memset(control, 0, sizeof(control));
403 		msg.msg_control = control;
404 		msg.msg_controllen = sizeof(control);
405 
406 		cm = CMSG_FIRSTHDR(&msg);
407 		cm->cmsg_level = SOL_SOCKET;
408 		cm->cmsg_type = SCM_TXTIME;
409 		cm->cmsg_len = CMSG_LEN(sizeof(uint64_t));
410 		memcpy(CMSG_DATA(cm), &txtime_ns, sizeof(txtime_ns));
411 	}
412 
413 	ret = sendmsg(fd, &msg, 0);
414 	if (ret == -1)
415 		error(1, errno, "sendmsg failure");
416 	if (ret != len)
417 		error(1, 0, "sendmsg wrong length: %d vs %d", ret, len);
418 }
419 
420 static void create_packet(void *buf, int seq_offset, int ack_offset,
421 			  int payload_len, int fin)
422 {
423 	int ip_hdr_len = (proto == PF_INET) ?
424 			 sizeof(struct iphdr) : sizeof(struct ipv6hdr);
425 	int inner_ip_off = tcp_offset - ip_hdr_len;
426 
427 	memset(buf, 0, total_hdr_len);
428 	memset(buf + total_hdr_len, 'a', payload_len);
429 
430 	fill_transportlayer(buf + tcp_offset, seq_offset, ack_offset,
431 			    payload_len, fin);
432 
433 	fill_networklayer(buf + inner_ip_off, payload_len, IPPROTO_TCP);
434 	if (inner_ip_off > ETH_HLEN) {
435 		if (pppoe) {
436 			fill_pppoelayer(buf + ETH_HLEN, payload_len + ip_hdr_len, 0x1234);
437 		} else {
438 			int encap_proto = (proto == PF_INET) ?
439 					  IPPROTO_IPIP : IPPROTO_IPV6;
440 
441 			fill_networklayer(buf + ETH_HLEN,
442 					  payload_len + ip_hdr_len, encap_proto);
443 		}
444 	}
445 
446 	fill_datalinklayer(buf);
447 }
448 
449 static void create_capacity_packet(void *buf, int flow_id, int pkt_idx, int psh)
450 {
451 	int seq_offset = pkt_idx * CAPACITY_PAYLOAD_LEN;
452 	struct tcphdr *tcph;
453 
454 	create_packet(buf, seq_offset, 0, CAPACITY_PAYLOAD_LEN, 0);
455 
456 	/* Customize for this flow id */
457 	memset(buf + total_hdr_len, 'a' + flow_id, CAPACITY_PAYLOAD_LEN);
458 
459 	tcph = buf + tcp_offset;
460 	tcph->source = htons(SPORT + flow_id);
461 	tcph->psh = psh;
462 	tcph->check = 0;
463 	tcph->check = tcp_checksum(tcph, CAPACITY_PAYLOAD_LEN);
464 }
465 
466 /* Send a capacity test, 2 packets per flow, all first packets then all second:
467  *  A1 B1 C1 D1 ... A2 B2 C2 D2 ...
468  */
469 static void send_capacity(int fd, struct sockaddr_ll *daddr)
470 {
471 	static char buf[MAX_HDR_LEN + CAPACITY_PAYLOAD_LEN];
472 	int pkt_size = total_hdr_len + CAPACITY_PAYLOAD_LEN;
473 	int i;
474 
475 	/* Send first packet of each flow (no PSH) */
476 	for (i = 0; i < num_flows; i++) {
477 		create_capacity_packet(buf, i, 0, 0);
478 		write_packet(fd, buf, pkt_size, daddr);
479 	}
480 
481 	/* Send second packet of each flow (with PSH to flush) */
482 	for (i = 0; i < num_flows; i++) {
483 		create_capacity_packet(buf, i, 1, 1);
484 		write_packet(fd, buf, pkt_size, daddr);
485 	}
486 }
487 
488 #ifndef TH_CWR
489 #define TH_CWR 0x80
490 #endif
491 static void set_flags(struct tcphdr *tcph, int payload_len, int psh, int syn,
492 		      int rst, int urg, int cwr)
493 {
494 	tcph->psh = psh;
495 	tcph->syn = syn;
496 	tcph->rst = rst;
497 	tcph->urg = urg;
498 	if (cwr)
499 		tcph->th_flags |= TH_CWR;
500 	else
501 		tcph->th_flags &= ~TH_CWR;
502 	tcph->check = 0;
503 	tcph->check = tcp_checksum(tcph, payload_len);
504 }
505 
506 /* send extra flags of the (NUM_PACKETS / 2) and (NUM_PACKETS / 2 - 1)
507  * pkts, not first and not last pkt
508  */
509 static void send_flags(int fd, struct sockaddr_ll *daddr, int psh, int syn,
510 		       int rst, int urg, int cwr)
511 {
512 	static char flag_buf[2][MAX_HDR_LEN + PAYLOAD_LEN];
513 	static char buf[MAX_HDR_LEN + PAYLOAD_LEN];
514 	int payload_len, pkt_size, i;
515 	struct tcphdr *tcph;
516 	int flag[2];
517 
518 	payload_len = PAYLOAD_LEN * (psh || cwr);
519 	pkt_size = total_hdr_len + payload_len;
520 	flag[0] = NUM_PACKETS / 2;
521 	flag[1] = NUM_PACKETS / 2 - 1;
522 
523 	/* Create and configure packets with flags
524 	 */
525 	for (i = 0; i < 2; i++) {
526 		if (flag[i] > 0) {
527 			create_packet(flag_buf[i], flag[i] * payload_len, 0,
528 				      payload_len, 0);
529 			tcph = (struct tcphdr *)(flag_buf[i] + tcp_offset);
530 			set_flags(tcph, payload_len, psh, syn, rst, urg, cwr);
531 		}
532 	}
533 
534 	for (i = 0; i < NUM_PACKETS + 1; i++) {
535 		if (i == flag[0]) {
536 			write_packet(fd, flag_buf[0], pkt_size, daddr);
537 			continue;
538 		} else if (i == flag[1] && cwr) {
539 			write_packet(fd, flag_buf[1], pkt_size, daddr);
540 			continue;
541 		}
542 		create_packet(buf, i * PAYLOAD_LEN, 0, PAYLOAD_LEN, 0);
543 		write_packet(fd, buf, total_hdr_len + PAYLOAD_LEN, daddr);
544 	}
545 }
546 
547 /* Test for data of same length, smaller than previous
548  * and of different lengths
549  */
550 static void send_data_pkts(int fd, struct sockaddr_ll *daddr,
551 			   int payload_len1, int payload_len2)
552 {
553 	static char buf[L2_HLEN_MAX + IP_MAXPACKET];
554 
555 	create_packet(buf, 0, 0, payload_len1, 0);
556 	write_packet(fd, buf, total_hdr_len + payload_len1, daddr);
557 	create_packet(buf, payload_len1, 0, payload_len2, 0);
558 	write_packet(fd, buf, total_hdr_len + payload_len2, daddr);
559 }
560 
561 /* If incoming segments make tracked segment length exceed
562  * legal IP datagram length, do not coalesce
563  */
564 static void send_large(int fd, struct sockaddr_ll *daddr, int remainder)
565 {
566 	static char pkts[MAX_LARGE_PKT_CNT][MAX_HDR_LEN + MAX_MSS];
567 	static char new_seg[MAX_HDR_LEN + MAX_MSS];
568 	static char last[MAX_HDR_LEN + MAX_MSS];
569 	const int num_pkt = num_large_pkt();
570 	const int mss = calc_mss();
571 	int i;
572 
573 	for (i = 0; i < num_pkt; i++)
574 		create_packet(pkts[i], i * mss, 0, mss, 0);
575 	create_packet(last, num_pkt * mss, 0, remainder, 0);
576 	create_packet(new_seg, (num_pkt + 1) * mss, 0, remainder, 0);
577 
578 	for (i = 0; i < num_pkt; i++)
579 		write_packet(fd, pkts[i], total_hdr_len + mss, daddr);
580 	write_packet(fd, last, total_hdr_len + remainder, daddr);
581 	write_packet(fd, new_seg, total_hdr_len + remainder, daddr);
582 }
583 
584 /* Pure acks and dup acks don't coalesce */
585 static void send_ack(int fd, struct sockaddr_ll *daddr)
586 {
587 	static char buf[MAX_HDR_LEN];
588 
589 	create_packet(buf, 0, 0, 0, 0);
590 	write_packet(fd, buf, total_hdr_len, daddr);
591 	write_packet(fd, buf, total_hdr_len, daddr);
592 	create_packet(buf, 0, 1, 0, 0);
593 	write_packet(fd, buf, total_hdr_len, daddr);
594 }
595 
596 static void recompute_packet(char *buf, char *no_ext, int extlen)
597 {
598 	struct tcphdr *tcphdr = (struct tcphdr *)(buf + tcp_offset);
599 	int off;
600 
601 	memmove(buf, no_ext, total_hdr_len);
602 	memmove(buf + total_hdr_len + extlen,
603 		no_ext + total_hdr_len, PAYLOAD_LEN);
604 
605 	tcphdr->doff = tcphdr->doff + (extlen / 4);
606 	tcphdr->check = 0;
607 	tcphdr->check = tcp_checksum(tcphdr, PAYLOAD_LEN + extlen);
608 	if (proto == PF_INET) {
609 		for (off = ETH_HLEN; off < tcp_offset;
610 		     off += sizeof(struct iphdr)) {
611 			struct iphdr *iph = (struct iphdr *)(buf + off);
612 
613 			iph->tot_len = htons(ntohs(iph->tot_len) + extlen);
614 			iph->check = 0;
615 			iph->check = checksum_fold(iph, sizeof(struct iphdr), 0);
616 		}
617 	} else {
618 		for (off = ETH_HLEN; off < tcp_offset;
619 		     off += sizeof(struct ipv6hdr)) {
620 			struct ipv6hdr *ip6h = (struct ipv6hdr *)(buf + off);
621 
622 			ip6h->payload_len =
623 				htons(ntohs(ip6h->payload_len) + extlen);
624 		}
625 	}
626 }
627 
628 static void tcp_write_options(char *buf, int kind, int ts)
629 {
630 	struct tcp_option_ts {
631 		uint8_t kind;
632 		uint8_t len;
633 		uint32_t tsval;
634 		uint32_t tsecr;
635 	} *opt_ts = (void *)buf;
636 	struct tcp_option_window {
637 		uint8_t kind;
638 		uint8_t len;
639 		uint8_t shift;
640 	} *opt_window = (void *)buf;
641 
642 	switch (kind) {
643 	case TCPOPT_NOP:
644 		buf[0] = TCPOPT_NOP;
645 		break;
646 	case TCPOPT_WINDOW:
647 		memset(opt_window, 0, sizeof(struct tcp_option_window));
648 		opt_window->kind = TCPOPT_WINDOW;
649 		opt_window->len = TCPOLEN_WINDOW;
650 		opt_window->shift = 0;
651 		break;
652 	case TCPOPT_TIMESTAMP:
653 		memset(opt_ts, 0, sizeof(struct tcp_option_ts));
654 		opt_ts->kind = TCPOPT_TIMESTAMP;
655 		opt_ts->len = TCPOLEN_TIMESTAMP;
656 		opt_ts->tsval = ts;
657 		opt_ts->tsecr = 0;
658 		break;
659 	default:
660 		error(1, 0, "unimplemented TCP option");
661 		break;
662 	}
663 }
664 
665 /* TCP with options is always a permutation of {TS, NOP, NOP}.
666  * Implement different orders to verify coalescing stops.
667  */
668 static void add_standard_tcp_options(char *buf, char *no_ext, int ts, int order)
669 {
670 	switch (order) {
671 	case 0:
672 		tcp_write_options(buf + total_hdr_len, TCPOPT_NOP, 0);
673 		tcp_write_options(buf + total_hdr_len + 1, TCPOPT_NOP, 0);
674 		tcp_write_options(buf + total_hdr_len + 2 /* two NOP opts */,
675 				  TCPOPT_TIMESTAMP, ts);
676 		break;
677 	case 1:
678 		tcp_write_options(buf + total_hdr_len, TCPOPT_NOP, 0);
679 		tcp_write_options(buf + total_hdr_len + 1,
680 				  TCPOPT_TIMESTAMP, ts);
681 		tcp_write_options(buf + total_hdr_len + 1 + TCPOLEN_TIMESTAMP,
682 				  TCPOPT_NOP, 0);
683 		break;
684 	case 2:
685 		tcp_write_options(buf + total_hdr_len, TCPOPT_TIMESTAMP, ts);
686 		tcp_write_options(buf + total_hdr_len + TCPOLEN_TIMESTAMP + 1,
687 				  TCPOPT_NOP, 0);
688 		tcp_write_options(buf + total_hdr_len + TCPOLEN_TIMESTAMP + 2,
689 				  TCPOPT_NOP, 0);
690 		break;
691 	default:
692 		error(1, 0, "unknown order");
693 		break;
694 	}
695 	recompute_packet(buf, no_ext, TCPOLEN_TSTAMP_APPA);
696 }
697 
698 /* Packets with invalid checksum don't coalesce. */
699 static void send_changed_checksum(int fd, struct sockaddr_ll *daddr)
700 {
701 	static char buf[MAX_HDR_LEN + PAYLOAD_LEN];
702 	struct tcphdr *tcph = (struct tcphdr *)(buf + tcp_offset);
703 	int pkt_size = total_hdr_len + PAYLOAD_LEN;
704 
705 	create_packet(buf, 0, 0, PAYLOAD_LEN, 0);
706 	write_packet(fd, buf, pkt_size, daddr);
707 
708 	create_packet(buf, PAYLOAD_LEN, 0, PAYLOAD_LEN, 0);
709 	tcph->check = tcph->check - 1;
710 	write_packet(fd, buf, pkt_size, daddr);
711 }
712 
713 /* Packets with incorrect IPv4 header checksum don't coalesce. */
714 static void send_changed_ip_checksum(int fd, struct sockaddr_ll *daddr)
715 {
716 	static char buf[MAX_HDR_LEN + PAYLOAD_LEN];
717 	struct iphdr *iph = (struct iphdr *)(buf + ETH_HLEN);
718 	int pkt_size = total_hdr_len + PAYLOAD_LEN;
719 
720 	create_packet(buf, 0, 0, PAYLOAD_LEN, 0);
721 	write_packet(fd, buf, pkt_size, daddr);
722 
723 	create_packet(buf, PAYLOAD_LEN, 0, PAYLOAD_LEN, 0);
724 	iph->check = iph->check - 1;
725 	write_packet(fd, buf, pkt_size, daddr);
726 
727 	create_packet(buf, PAYLOAD_LEN * 2, 0, PAYLOAD_LEN, 0);
728 	write_packet(fd, buf, pkt_size, daddr);
729 }
730 
731  /* Packets with non-consecutive sequence number don't coalesce.*/
732 static void send_changed_seq(int fd, struct sockaddr_ll *daddr)
733 {
734 	static char buf[MAX_HDR_LEN + PAYLOAD_LEN];
735 	struct tcphdr *tcph = (struct tcphdr *)(buf + tcp_offset);
736 	int pkt_size = total_hdr_len + PAYLOAD_LEN;
737 
738 	create_packet(buf, 0, 0, PAYLOAD_LEN, 0);
739 	write_packet(fd, buf, pkt_size, daddr);
740 
741 	create_packet(buf, PAYLOAD_LEN, 0, PAYLOAD_LEN, 0);
742 	tcph->seq = ntohl(htonl(tcph->seq) + 1);
743 	tcph->check = 0;
744 	tcph->check = tcp_checksum(tcph, PAYLOAD_LEN);
745 	write_packet(fd, buf, pkt_size, daddr);
746 }
747 
748  /* Packet with different timestamp option or different timestamps
749   * don't coalesce.
750   */
751 static void send_changed_ts(int fd, struct sockaddr_ll *daddr)
752 {
753 	static char buf[MAX_HDR_LEN + PAYLOAD_LEN];
754 	static char extpkt[sizeof(buf) + TCPOLEN_TSTAMP_APPA];
755 	int pkt_size = total_hdr_len + PAYLOAD_LEN + TCPOLEN_TSTAMP_APPA;
756 
757 	create_packet(buf, 0, 0, PAYLOAD_LEN, 0);
758 	add_standard_tcp_options(extpkt, buf, 0, 0);
759 	write_packet(fd, extpkt, pkt_size, daddr);
760 
761 	create_packet(buf, PAYLOAD_LEN, 0, PAYLOAD_LEN, 0);
762 	add_standard_tcp_options(extpkt, buf, 0, 0);
763 	write_packet(fd, extpkt, pkt_size, daddr);
764 
765 	create_packet(buf, PAYLOAD_LEN * 2, 0, PAYLOAD_LEN, 0);
766 	add_standard_tcp_options(extpkt, buf, 100, 0);
767 	write_packet(fd, extpkt, pkt_size, daddr);
768 
769 	create_packet(buf, PAYLOAD_LEN * 3, 0, PAYLOAD_LEN, 0);
770 	add_standard_tcp_options(extpkt, buf, 100, 1);
771 	write_packet(fd, extpkt, pkt_size, daddr);
772 
773 	create_packet(buf, PAYLOAD_LEN * 4, 0, PAYLOAD_LEN, 0);
774 	add_standard_tcp_options(extpkt, buf, 100, 2);
775 	write_packet(fd, extpkt, pkt_size, daddr);
776 }
777 
778 /* Packet with different tcp options don't coalesce. */
779 static void send_diff_opt(int fd, struct sockaddr_ll *daddr)
780 {
781 	static char buf[MAX_HDR_LEN + PAYLOAD_LEN];
782 	static char extpkt1[sizeof(buf) + TCPOLEN_TSTAMP_APPA];
783 	static char extpkt2[sizeof(buf) + TCPOLEN_MAXSEG];
784 	int extpkt1_size = total_hdr_len + PAYLOAD_LEN + TCPOLEN_TSTAMP_APPA;
785 	int extpkt2_size = total_hdr_len + PAYLOAD_LEN + TCPOLEN_MAXSEG;
786 
787 	create_packet(buf, 0, 0, PAYLOAD_LEN, 0);
788 	add_standard_tcp_options(extpkt1, buf, 0, 0);
789 	write_packet(fd, extpkt1, extpkt1_size, daddr);
790 
791 	create_packet(buf, PAYLOAD_LEN, 0, PAYLOAD_LEN, 0);
792 	add_standard_tcp_options(extpkt1, buf, 0, 0);
793 	write_packet(fd, extpkt1, extpkt1_size, daddr);
794 
795 	create_packet(buf, PAYLOAD_LEN * 2, 0, PAYLOAD_LEN, 0);
796 	tcp_write_options(extpkt2 + MAX_HDR_LEN, TCPOPT_NOP, 0);
797 	tcp_write_options(extpkt2 + MAX_HDR_LEN + 1, TCPOPT_WINDOW, 0);
798 	recompute_packet(extpkt2, buf, TCPOLEN_WINDOW + 1);
799 	write_packet(fd, extpkt2, extpkt2_size, daddr);
800 }
801 
802 static void add_ipv4_ts_option(void *buf, void *optpkt)
803 {
804 	struct ip_timestamp *ts = (struct ip_timestamp *)(optpkt + tcp_offset);
805 	int optlen = sizeof(struct ip_timestamp);
806 	struct iphdr *iph;
807 
808 	if (optlen % 4)
809 		error(1, 0, "ipv4 timestamp length is not a multiple of 4B");
810 
811 	ts->ipt_code = IPOPT_TS;
812 	ts->ipt_len = optlen;
813 	ts->ipt_ptr = 5;
814 	ts->ipt_flg = IPOPT_TS_TSONLY;
815 
816 	memcpy(optpkt, buf, tcp_offset);
817 	memcpy(optpkt + tcp_offset + optlen, buf + tcp_offset,
818 	       sizeof(struct tcphdr) + PAYLOAD_LEN);
819 
820 	iph = (struct iphdr *)(optpkt + ETH_HLEN);
821 	iph->ihl = 5 + (optlen / 4);
822 	iph->tot_len = htons(ntohs(iph->tot_len) + optlen);
823 	iph->check = 0;
824 	iph->check = checksum_fold(iph, sizeof(struct iphdr) + optlen, 0);
825 }
826 
827 static void add_ipv6_exthdr(void *buf, void *optpkt, __u8 exthdr_type, char *ext_payload)
828 {
829 	struct ipv6_opt_hdr *exthdr = (struct ipv6_opt_hdr *)(optpkt + tcp_offset);
830 	struct ipv6hdr *iph = (struct ipv6hdr *)(optpkt + ETH_HLEN);
831 	char *exthdr_payload_start = (char *)(exthdr + 1);
832 
833 	exthdr->hdrlen = 0;
834 	exthdr->nexthdr = IPPROTO_TCP;
835 
836 	memcpy(exthdr_payload_start, ext_payload, MIN_EXTHDR_SIZE - sizeof(*exthdr));
837 
838 	memcpy(optpkt, buf, tcp_offset);
839 	memcpy(optpkt + tcp_offset + MIN_EXTHDR_SIZE, buf + tcp_offset,
840 		sizeof(struct tcphdr) + PAYLOAD_LEN);
841 
842 	iph->nexthdr = exthdr_type;
843 	iph->payload_len = htons(ntohs(iph->payload_len) + MIN_EXTHDR_SIZE);
844 }
845 
846 static void fix_ip4_checksum(struct iphdr *iph)
847 {
848 	iph->check = 0;
849 	iph->check = checksum_fold(iph, sizeof(struct iphdr), 0);
850 }
851 
852 static void send_flush_id_case(int fd, struct sockaddr_ll *daddr,
853 			       enum flush_id_case tcase)
854 {
855 	static char buf1[MAX_HDR_LEN + PAYLOAD_LEN];
856 	static char buf2[MAX_HDR_LEN + PAYLOAD_LEN];
857 	static char buf3[MAX_HDR_LEN + PAYLOAD_LEN];
858 	bool send_three = false;
859 	struct iphdr *iph1;
860 	struct iphdr *iph2;
861 	struct iphdr *iph3;
862 
863 	iph1 = (struct iphdr *)(buf1 + ETH_HLEN);
864 	iph2 = (struct iphdr *)(buf2 + ETH_HLEN);
865 	iph3 = (struct iphdr *)(buf3 + ETH_HLEN);
866 
867 	create_packet(buf1, 0, 0, PAYLOAD_LEN, 0);
868 	create_packet(buf2, PAYLOAD_LEN, 0, PAYLOAD_LEN, 0);
869 	create_packet(buf3, PAYLOAD_LEN * 2, 0, PAYLOAD_LEN, 0);
870 
871 	switch (tcase) {
872 	case FLUSH_ID_DF1_INC: /* DF=1, Incrementing - should coalesce */
873 		iph1->frag_off |= htons(IP_DF);
874 		iph1->id = htons(8);
875 
876 		iph2->frag_off |= htons(IP_DF);
877 		iph2->id = htons(9);
878 		break;
879 
880 	case FLUSH_ID_DF1_FIXED: /* DF=1, Fixed - should coalesce */
881 		iph1->frag_off |= htons(IP_DF);
882 		iph1->id = htons(8);
883 
884 		iph2->frag_off |= htons(IP_DF);
885 		iph2->id = htons(8);
886 		break;
887 
888 	case FLUSH_ID_DF0_INC: /* DF=0, Incrementing - should coalesce */
889 		iph1->frag_off &= ~htons(IP_DF);
890 		iph1->id = htons(8);
891 
892 		iph2->frag_off &= ~htons(IP_DF);
893 		iph2->id = htons(9);
894 		break;
895 
896 	case FLUSH_ID_DF0_FIXED: /* DF=0, Fixed - should coalesce */
897 		iph1->frag_off &= ~htons(IP_DF);
898 		iph1->id = htons(8);
899 
900 		iph2->frag_off &= ~htons(IP_DF);
901 		iph2->id = htons(8);
902 		break;
903 
904 	case FLUSH_ID_DF1_INC_FIXED: /* DF=1, two packets incrementing, and
905 				      * one fixed - should coalesce only the
906 				      * first two packets
907 				      */
908 		iph1->frag_off |= htons(IP_DF);
909 		iph1->id = htons(8);
910 
911 		iph2->frag_off |= htons(IP_DF);
912 		iph2->id = htons(9);
913 
914 		iph3->frag_off |= htons(IP_DF);
915 		iph3->id = htons(9);
916 		send_three = true;
917 		break;
918 
919 	case FLUSH_ID_DF1_FIXED_INC: /* DF=1, two packets fixed, and one
920 				      * incrementing - should coalesce only
921 				      * the first two packets
922 				      */
923 		iph1->frag_off |= htons(IP_DF);
924 		iph1->id = htons(8);
925 
926 		iph2->frag_off |= htons(IP_DF);
927 		iph2->id = htons(8);
928 
929 		iph3->frag_off |= htons(IP_DF);
930 		iph3->id = htons(9);
931 		send_three = true;
932 		break;
933 	}
934 
935 	fix_ip4_checksum(iph1);
936 	fix_ip4_checksum(iph2);
937 	write_packet(fd, buf1, total_hdr_len + PAYLOAD_LEN, daddr);
938 	write_packet(fd, buf2, total_hdr_len + PAYLOAD_LEN, daddr);
939 
940 	if (send_three) {
941 		fix_ip4_checksum(iph3);
942 		write_packet(fd, buf3, total_hdr_len + PAYLOAD_LEN, daddr);
943 	}
944 }
945 
946 static void send_ipv6_exthdr(int fd, struct sockaddr_ll *daddr, char *ext_data1, char *ext_data2)
947 {
948 	static char buf[MAX_HDR_LEN + PAYLOAD_LEN];
949 	static char exthdr_pck[sizeof(buf) + MIN_EXTHDR_SIZE];
950 
951 	create_packet(buf, 0, 0, PAYLOAD_LEN, 0);
952 	add_ipv6_exthdr(buf, exthdr_pck, IPPROTO_DSTOPTS, ext_data1);
953 	write_packet(fd, exthdr_pck, total_hdr_len + PAYLOAD_LEN + MIN_EXTHDR_SIZE, daddr);
954 
955 	create_packet(buf, PAYLOAD_LEN * 1, 0, PAYLOAD_LEN, 0);
956 	add_ipv6_exthdr(buf, exthdr_pck, IPPROTO_DSTOPTS, ext_data2);
957 	write_packet(fd, exthdr_pck, total_hdr_len + PAYLOAD_LEN + MIN_EXTHDR_SIZE, daddr);
958 }
959 
960 /* IPv4 options shouldn't coalesce */
961 static void send_ip_options(int fd, struct sockaddr_ll *daddr)
962 {
963 	static char buf[MAX_HDR_LEN + PAYLOAD_LEN];
964 	static char optpkt[sizeof(buf) + sizeof(struct ip_timestamp)];
965 	int optlen = sizeof(struct ip_timestamp);
966 	int pkt_size = total_hdr_len + PAYLOAD_LEN + optlen;
967 
968 	create_packet(buf, 0, 0, PAYLOAD_LEN, 0);
969 	write_packet(fd, buf, total_hdr_len + PAYLOAD_LEN, daddr);
970 
971 	create_packet(buf, PAYLOAD_LEN * 1, 0, PAYLOAD_LEN, 0);
972 	add_ipv4_ts_option(buf, optpkt);
973 	write_packet(fd, optpkt, pkt_size, daddr);
974 
975 	create_packet(buf, PAYLOAD_LEN * 2, 0, PAYLOAD_LEN, 0);
976 	write_packet(fd, buf, total_hdr_len + PAYLOAD_LEN, daddr);
977 }
978 
979 /*  IPv4 fragments shouldn't coalesce */
980 static void send_fragment4(int fd, struct sockaddr_ll *daddr)
981 {
982 	static char buf[IP_MAXPACKET];
983 	struct iphdr *iph = (struct iphdr *)(buf + ETH_HLEN);
984 	int pkt_size = total_hdr_len + PAYLOAD_LEN;
985 
986 	create_packet(buf, 0, 0, PAYLOAD_LEN, 0);
987 	write_packet(fd, buf, pkt_size, daddr);
988 
989 	/* Once fragmented, packet would retain the total_len.
990 	 * Tcp header is prepared as if rest of data is in follow-up frags,
991 	 * but follow up frags aren't actually sent.
992 	 */
993 	memset(buf + total_hdr_len, 'a', PAYLOAD_LEN * 2);
994 	fill_transportlayer(buf + tcp_offset, PAYLOAD_LEN, 0, PAYLOAD_LEN * 2, 0);
995 	fill_networklayer(buf + ETH_HLEN, PAYLOAD_LEN, IPPROTO_TCP);
996 	fill_datalinklayer(buf);
997 
998 	iph->frag_off = htons(0x6000); // DF = 1, MF = 1
999 	iph->check = 0;
1000 	iph->check = checksum_fold(iph, sizeof(struct iphdr), 0);
1001 	write_packet(fd, buf, pkt_size, daddr);
1002 }
1003 
1004 /* IPv4 packets with different ttl don't coalesce.*/
1005 static void send_changed_ttl(int fd, struct sockaddr_ll *daddr)
1006 {
1007 	int pkt_size = total_hdr_len + PAYLOAD_LEN;
1008 	static char buf[MAX_HDR_LEN + PAYLOAD_LEN];
1009 	struct iphdr *iph = (struct iphdr *)(buf + ETH_HLEN);
1010 
1011 	create_packet(buf, 0, 0, PAYLOAD_LEN, 0);
1012 	write_packet(fd, buf, pkt_size, daddr);
1013 
1014 	create_packet(buf, PAYLOAD_LEN, 0, PAYLOAD_LEN, 0);
1015 	iph->ttl = 7;
1016 	iph->check = 0;
1017 	iph->check = checksum_fold(iph, sizeof(struct iphdr), 0);
1018 	write_packet(fd, buf, pkt_size, daddr);
1019 }
1020 
1021 /* Packets with different tos don't coalesce.*/
1022 static void send_changed_tos(int fd, struct sockaddr_ll *daddr)
1023 {
1024 	int pkt_size = total_hdr_len + PAYLOAD_LEN;
1025 	static char buf[MAX_HDR_LEN + PAYLOAD_LEN];
1026 	struct iphdr *iph = (struct iphdr *)(buf + ETH_HLEN);
1027 	struct ipv6hdr *ip6h = (struct ipv6hdr *)(buf + ETH_HLEN);
1028 
1029 	create_packet(buf, 0, 0, PAYLOAD_LEN, 0);
1030 	write_packet(fd, buf, pkt_size, daddr);
1031 
1032 	create_packet(buf, PAYLOAD_LEN, 0, PAYLOAD_LEN, 0);
1033 	if (proto == PF_INET) {
1034 		iph->tos = 1;
1035 		iph->check = 0;
1036 		iph->check = checksum_fold(iph, sizeof(struct iphdr), 0);
1037 	} else if (proto == PF_INET6) {
1038 		ip6h->priority = 0xf;
1039 	}
1040 	write_packet(fd, buf, pkt_size, daddr);
1041 }
1042 
1043 /* Packets with different ECN don't coalesce.*/
1044 static void send_changed_ECN(int fd, struct sockaddr_ll *daddr)
1045 {
1046 	int pkt_size = total_hdr_len + PAYLOAD_LEN;
1047 	static char buf[MAX_HDR_LEN + PAYLOAD_LEN];
1048 	struct iphdr *iph = (struct iphdr *)(buf + ETH_HLEN);
1049 
1050 	create_packet(buf, 0, 0, PAYLOAD_LEN, 0);
1051 	write_packet(fd, buf, pkt_size, daddr);
1052 
1053 	create_packet(buf, PAYLOAD_LEN, 0, PAYLOAD_LEN, 0);
1054 	if (proto == PF_INET) {
1055 		buf[ETH_HLEN + 1] ^= 0x2; // ECN set to 10
1056 		iph->check = 0;
1057 		iph->check = checksum_fold(iph, sizeof(struct iphdr), 0);
1058 	} else {
1059 		buf[ETH_HLEN + 1] ^= 0x20; // ECN set to 10
1060 	}
1061 	write_packet(fd, buf, pkt_size, daddr);
1062 }
1063 
1064 /* IPv6 fragments and packets with extensions don't coalesce.*/
1065 static void send_fragment6(int fd, struct sockaddr_ll *daddr)
1066 {
1067 	static char buf[MAX_HDR_LEN + PAYLOAD_LEN];
1068 	static char extpkt[MAX_HDR_LEN + PAYLOAD_LEN +
1069 			   sizeof(struct ip6_frag)];
1070 	struct ipv6hdr *ip6h = (struct ipv6hdr *)(buf + ETH_HLEN);
1071 	struct ip6_frag *frag = (void *)(extpkt + tcp_offset);
1072 	int extlen = sizeof(struct ip6_frag);
1073 	int bufpkt_len = total_hdr_len + PAYLOAD_LEN;
1074 	int extpkt_len = bufpkt_len + extlen;
1075 	int i;
1076 
1077 	for (i = 0; i < 2; i++) {
1078 		create_packet(buf, PAYLOAD_LEN * i, 0, PAYLOAD_LEN, 0);
1079 		write_packet(fd, buf, bufpkt_len, daddr);
1080 	}
1081 	sleep(1);
1082 	create_packet(buf, PAYLOAD_LEN * 2, 0, PAYLOAD_LEN, 0);
1083 	memset(extpkt, 0, extpkt_len);
1084 
1085 	ip6h->nexthdr = IPPROTO_FRAGMENT;
1086 	ip6h->payload_len = htons(ntohs(ip6h->payload_len) + extlen);
1087 	frag->ip6f_nxt = IPPROTO_TCP;
1088 
1089 	memcpy(extpkt, buf, tcp_offset);
1090 	memcpy(extpkt + tcp_offset + extlen, buf + tcp_offset,
1091 	       sizeof(struct tcphdr) + PAYLOAD_LEN);
1092 	write_packet(fd, extpkt, extpkt_len, daddr);
1093 
1094 	create_packet(buf, PAYLOAD_LEN * 3, 0, PAYLOAD_LEN, 0);
1095 	write_packet(fd, buf, bufpkt_len, daddr);
1096 }
1097 
1098 static void send_changed_pppoe_sid(int fd, struct sockaddr_ll *daddr)
1099 {
1100 	static char buf[MAX_HDR_LEN + PAYLOAD_LEN];
1101 	int pkt_size = total_hdr_len + PAYLOAD_LEN;
1102 	struct pppoe_hdr *hdr = (struct pppoe_hdr *)(buf + ETH_HLEN);
1103 
1104 	create_packet(buf, 0, 0, PAYLOAD_LEN, 0);
1105 	write_packet(fd, buf, pkt_size, daddr);
1106 
1107 	create_packet(buf, PAYLOAD_LEN, 0, PAYLOAD_LEN, 0);
1108 	hdr->sid = htons(0x4321);
1109 	write_packet(fd, buf, pkt_size, daddr);
1110 }
1111 
1112 static void bind_packetsocket(int fd)
1113 {
1114 	struct sockaddr_ll daddr = {};
1115 
1116 	daddr.sll_family = AF_PACKET;
1117 	daddr.sll_protocol = ethhdr_proto;
1118 	daddr.sll_ifindex = if_nametoindex(ifname);
1119 	if (daddr.sll_ifindex == 0)
1120 		error(1, errno, "if_nametoindex");
1121 
1122 	if (bind(fd, (void *)&daddr, sizeof(daddr)) < 0)
1123 		error(1, errno, "could not bind socket");
1124 }
1125 
1126 static void set_timeout(int fd)
1127 {
1128 	struct timeval timeout;
1129 
1130 	timeout.tv_sec = 3;
1131 	timeout.tv_usec = 0;
1132 	if (setsockopt(fd, SOL_SOCKET, SO_RCVTIMEO, (char *)&timeout,
1133 		       sizeof(timeout)) < 0)
1134 		error(1, errno, "cannot set timeout, setsockopt failed");
1135 }
1136 
1137 static void set_rcvbuf(int fd)
1138 {
1139 	int bufsize = 1 * 1024 * 1024; /* 1 MB */
1140 
1141 	if (setsockopt(fd, SOL_SOCKET, SO_RCVBUF, &bufsize, sizeof(bufsize)))
1142 		error(1, errno, "cannot set rcvbuf size, setsockopt failed");
1143 }
1144 
1145 static void recv_error(int fd, int rcv_errno)
1146 {
1147 	struct tpacket_stats stats;
1148 	socklen_t len;
1149 
1150 	len = sizeof(stats);
1151 	if (getsockopt(fd, SOL_PACKET, PACKET_STATISTICS, &stats, &len))
1152 		error(1, errno, "can't get stats");
1153 
1154 	fprintf(stderr, "Socket stats: packets=%u, drops=%u\n",
1155 		stats.tp_packets, stats.tp_drops);
1156 	error(1, rcv_errno, "could not receive");
1157 }
1158 
1159 static void check_recv_pkts(int fd, int *correct_payload,
1160 			    int correct_num_pkts)
1161 {
1162 	static char buffer[IP_MAXPACKET + L2_HLEN_MAX + 1];
1163 	int nhoff = ETH_HLEN + (pppoe ? PPPOE_SES_HLEN : 0);
1164 	struct iphdr *iph = (struct iphdr *)(buffer + nhoff);
1165 	struct ipv6hdr *ip6h = (struct ipv6hdr *)(buffer + nhoff);
1166 	struct tcphdr *tcph;
1167 	bool bad_packet = false;
1168 	int tcp_ext_len = 0;
1169 	int ip_ext_len = 0;
1170 	int pkt_size = -1;
1171 	int data_len = 0;
1172 	int num_pkt = 0;
1173 	int i;
1174 
1175 	vlog("Expected {");
1176 	for (i = 0; i < correct_num_pkts; i++)
1177 		vlog("%d ", correct_payload[i]);
1178 	vlog("}, Total %d packets\nReceived {", correct_num_pkts);
1179 
1180 	while (1) {
1181 		ip_ext_len = 0;
1182 		pkt_size = recv(fd, buffer, sizeof(buffer), 0);
1183 		if (pkt_size < 0)
1184 			recv_error(fd, errno);
1185 
1186 		if (iph->version == 4)
1187 			ip_ext_len = (iph->ihl - 5) * 4;
1188 		else if (ip6h->version == 6 && !ip6ip6 &&
1189 			 ip6h->nexthdr != IPPROTO_TCP)
1190 			ip_ext_len = MIN_EXTHDR_SIZE;
1191 
1192 		tcph = (struct tcphdr *)(buffer + tcp_offset + ip_ext_len);
1193 
1194 		if (tcph->fin)
1195 			break;
1196 
1197 		tcp_ext_len = (tcph->doff - 5) * 4;
1198 		data_len = pkt_size - total_hdr_len - tcp_ext_len - ip_ext_len;
1199 		/* Min ethernet frame payload is 46(ETH_ZLEN - ETH_HLEN) by RFC 802.3.
1200 		 * Ipv4/tcp packets without at least 6 bytes of data will be padded.
1201 		 * Packet sockets are protocol agnostic, and will not trim the padding.
1202 		 */
1203 		if (pkt_size == ETH_ZLEN && iph->version == 4) {
1204 			data_len = ntohs(iph->tot_len)
1205 				- sizeof(struct tcphdr) - sizeof(struct iphdr);
1206 		}
1207 		vlog("%d ", data_len);
1208 		if (data_len != correct_payload[num_pkt]) {
1209 			vlog("[!=%d]", correct_payload[num_pkt]);
1210 			bad_packet = true;
1211 		}
1212 		num_pkt++;
1213 	}
1214 	vlog("}, Total %d packets.\n", num_pkt);
1215 	if (num_pkt != correct_num_pkts)
1216 		error(1, 0, "incorrect number of packets");
1217 	if (bad_packet)
1218 		error(1, 0, "incorrect packet geometry");
1219 
1220 	printf("Test succeeded\n\n");
1221 }
1222 
1223 static void check_capacity_pkts(int fd)
1224 {
1225 	static char buffer[IP_MAXPACKET + L2_HLEN_MAX + 1];
1226 	int nhoff = ETH_HLEN + (pppoe ? PPPOE_SES_HLEN : 0);
1227 	struct iphdr *iph = (struct iphdr *)(buffer + nhoff);
1228 	struct ipv6hdr *ip6h = (struct ipv6hdr *)(buffer + nhoff);
1229 	int num_pkt = 0, num_coal = 0, pkt_idx;
1230 	const char *fail_reason = NULL;
1231 	int flow_order[num_flows * 2];
1232 	int coalesced[num_flows];
1233 	struct tcphdr *tcph;
1234 	int ip_ext_len = 0;
1235 	int total_data = 0;
1236 	int pkt_size = -1;
1237 	int data_len = 0;
1238 	int flow_id;
1239 	int sport;
1240 
1241 	memset(coalesced, 0, sizeof(coalesced));
1242 	memset(flow_order, -1, sizeof(flow_order));
1243 
1244 	while (1) {
1245 		ip_ext_len = 0;
1246 		pkt_size = recv(fd, buffer, sizeof(buffer), 0);
1247 		if (pkt_size < 0)
1248 			recv_error(fd, errno);
1249 
1250 		if (iph->version == 4)
1251 			ip_ext_len = (iph->ihl - 5) * 4;
1252 		else if (ip6h->version == 6 && !ip6ip6 &&
1253 			 ip6h->nexthdr != IPPROTO_TCP)
1254 			ip_ext_len = MIN_EXTHDR_SIZE;
1255 
1256 		tcph = (struct tcphdr *)(buffer + tcp_offset + ip_ext_len);
1257 
1258 		if (tcph->fin)
1259 			break;
1260 
1261 		sport = ntohs(tcph->source);
1262 		flow_id = sport - SPORT;
1263 
1264 		if (flow_id < 0 || flow_id >= num_flows) {
1265 			vlog("Invalid flow_id %d from sport %d\n",
1266 			     flow_id, sport);
1267 			fail_reason = fail_reason ?: "invalid packet";
1268 			continue;
1269 		}
1270 
1271 		/* Calculate payload length */
1272 		if (pkt_size == ETH_ZLEN && iph->version == 4) {
1273 			data_len = ntohs(iph->tot_len)
1274 				- sizeof(struct tcphdr) - sizeof(struct iphdr);
1275 		} else {
1276 			data_len = pkt_size - total_hdr_len - ip_ext_len;
1277 		}
1278 
1279 		if (num_pkt < num_flows * 2) {
1280 			flow_order[num_pkt] = flow_id;
1281 		} else if (num_pkt == num_flows * 2) {
1282 			vlog("More packets than expected (%d)\n",
1283 			     num_flows * 2);
1284 			fail_reason = fail_reason ?: "too many packets";
1285 		}
1286 		coalesced[flow_id] = data_len;
1287 
1288 		if (data_len == CAPACITY_PAYLOAD_LEN * 2) {
1289 			num_coal++;
1290 		} else {
1291 			vlog("Pkt %d: flow %d, sport %d, len %d (expected %d)\n",
1292 			     num_pkt, flow_id, sport, data_len,
1293 			     CAPACITY_PAYLOAD_LEN * 2);
1294 			fail_reason = fail_reason ?: "not coalesced";
1295 		}
1296 
1297 		num_pkt++;
1298 		total_data += data_len;
1299 	}
1300 
1301 	/* Check flow ordering. We expect to see all non-coalesced first segs
1302 	 * then interleaved coalesced and non-coalesced second frames.
1303 	 */
1304 	pkt_idx = 0;
1305 	for (flow_id = 0; order_check && flow_id < num_flows; flow_id++) {
1306 		bool coaled = coalesced[flow_id] > CAPACITY_PAYLOAD_LEN;
1307 
1308 		if (coaled)
1309 			continue;
1310 
1311 		if (flow_order[pkt_idx] != flow_id) {
1312 			vlog("Flow order mismatch (non-coalesced) at position %d: expected flow %d, got flow %d\n",
1313 			     pkt_idx, flow_id, flow_order[pkt_idx]);
1314 			fail_reason = fail_reason ?: "bad packet order (1)";
1315 		}
1316 		pkt_idx++;
1317 	}
1318 	for (flow_id = 0; order_check && flow_id < num_flows; flow_id++) {
1319 		bool coaled = coalesced[flow_id] > CAPACITY_PAYLOAD_LEN;
1320 
1321 		if (flow_order[pkt_idx] != flow_id) {
1322 			vlog("Flow order mismatch at position %d: expected flow %d, got flow %d, coalesced: %d\n",
1323 			     pkt_idx, flow_id, flow_order[pkt_idx], coaled);
1324 			fail_reason = fail_reason ?: "bad packet order (2)";
1325 		}
1326 		pkt_idx++;
1327 	}
1328 
1329 	if (!fail_reason) {
1330 		vlog("All %d flows coalesced correctly\n", num_flows);
1331 		printf("Test succeeded\n\n");
1332 	} else {
1333 		printf("FAILED\n");
1334 	}
1335 
1336 	/* Always print stats for external validation */
1337 	printf("STATS: received=%d wire=%d coalesced=%d\n",
1338 	       num_pkt, num_pkt + num_coal, num_coal);
1339 
1340 	if (fail_reason)
1341 		error(1, 0, "capacity test failed %s", fail_reason);
1342 }
1343 
1344 static void gro_sender(void)
1345 {
1346 	int bufsize = 4 * 1024 * 1024; /* 4 MB */
1347 	const int fin_delay_us = 100 * 1000;
1348 	static char fin_pkt[MAX_HDR_LEN];
1349 	struct sockaddr_ll daddr = {};
1350 	int txfd = -1;
1351 
1352 	txfd = socket(PF_PACKET, SOCK_RAW, IPPROTO_RAW);
1353 	if (txfd < 0)
1354 		error(1, errno, "socket creation");
1355 
1356 	if (setsockopt(txfd, SOL_SOCKET, SO_SNDBUF, &bufsize, sizeof(bufsize)))
1357 		error(1, errno, "cannot set sndbuf size, setsockopt failed");
1358 
1359 	/* Enable SO_TXTIME unless test case generates more than one flow
1360 	 * SO_TXTIME could result in qdisc layer sorting the packets at sender.
1361 	 */
1362 	if (strcmp(testname, "single") && strcmp(testname, "capacity")) {
1363 		struct sock_txtime so_txtime = { .clockid = CLOCK_MONOTONIC, };
1364 		struct timespec ts;
1365 
1366 		if (setsockopt(txfd, SOL_SOCKET, SO_TXTIME,
1367 			       &so_txtime, sizeof(so_txtime)))
1368 			error(1, errno, "setsockopt SO_TXTIME");
1369 
1370 		if (clock_gettime(CLOCK_MONOTONIC, &ts))
1371 			error(1, errno, "clock_gettime");
1372 
1373 		txtime_ns = ts.tv_sec * 1000000000ULL + ts.tv_nsec;
1374 		txtime_ns += TXTIME_DELAY_MS * 1000000ULL;
1375 	}
1376 
1377 	memset(&daddr, 0, sizeof(daddr));
1378 	daddr.sll_ifindex = if_nametoindex(ifname);
1379 	if (daddr.sll_ifindex == 0)
1380 		error(1, errno, "if_nametoindex");
1381 	daddr.sll_family = AF_PACKET;
1382 	memcpy(daddr.sll_addr, dst_mac, ETH_ALEN);
1383 	daddr.sll_halen = ETH_ALEN;
1384 	create_packet(fin_pkt, PAYLOAD_LEN * 2, 0, 0, 1);
1385 
1386 	/* data sub-tests */
1387 	if (strcmp(testname, "data_same") == 0) {
1388 		send_data_pkts(txfd, &daddr, PAYLOAD_LEN, PAYLOAD_LEN);
1389 		write_packet(txfd, fin_pkt, total_hdr_len, &daddr);
1390 	} else if (strcmp(testname, "data_lrg_sml") == 0) {
1391 		send_data_pkts(txfd, &daddr, PAYLOAD_LEN, PAYLOAD_LEN / 2);
1392 		write_packet(txfd, fin_pkt, total_hdr_len, &daddr);
1393 	} else if (strcmp(testname, "data_lrg_1byte") == 0) {
1394 		send_data_pkts(txfd, &daddr, PAYLOAD_LEN, 1);
1395 		write_packet(txfd, fin_pkt, total_hdr_len, &daddr);
1396 	} else if (strcmp(testname, "data_sml_lrg") == 0) {
1397 		send_data_pkts(txfd, &daddr, PAYLOAD_LEN / 2, PAYLOAD_LEN);
1398 		write_packet(txfd, fin_pkt, total_hdr_len, &daddr);
1399 	} else if (strcmp(testname, "data_burst") == 0) {
1400 		static char buf[MAX_HDR_LEN + PAYLOAD_LEN];
1401 
1402 		create_packet(buf, 0, 0, PAYLOAD_LEN, 0);
1403 		write_packet(txfd, buf, total_hdr_len + PAYLOAD_LEN, &daddr);
1404 		create_packet(buf, PAYLOAD_LEN, 0, PAYLOAD_LEN, 0);
1405 		write_packet(txfd, buf, total_hdr_len + PAYLOAD_LEN, &daddr);
1406 
1407 		usleep(100 * 1000); /* 100ms */
1408 		create_packet(buf, PAYLOAD_LEN * 2, 0, PAYLOAD_LEN, 0);
1409 		write_packet(txfd, buf, total_hdr_len + PAYLOAD_LEN, &daddr);
1410 		create_packet(buf, PAYLOAD_LEN * 3, 0, PAYLOAD_LEN, 0);
1411 		write_packet(txfd, buf, total_hdr_len + PAYLOAD_LEN, &daddr);
1412 
1413 		write_packet(txfd, fin_pkt, total_hdr_len, &daddr);
1414 
1415 	/* ack test */
1416 	} else if (strcmp(testname, "ack") == 0) {
1417 		send_ack(txfd, &daddr);
1418 		write_packet(txfd, fin_pkt, total_hdr_len, &daddr);
1419 
1420 	/* flags sub-tests */
1421 	} else if (strcmp(testname, "flags_psh") == 0) {
1422 		send_flags(txfd, &daddr, 1, 0, 0, 0, 0);
1423 		write_packet(txfd, fin_pkt, total_hdr_len, &daddr);
1424 	} else if (strcmp(testname, "flags_syn") == 0) {
1425 		send_flags(txfd, &daddr, 0, 1, 0, 0, 0);
1426 		write_packet(txfd, fin_pkt, total_hdr_len, &daddr);
1427 	} else if (strcmp(testname, "flags_rst") == 0) {
1428 		send_flags(txfd, &daddr, 0, 0, 1, 0, 0);
1429 		write_packet(txfd, fin_pkt, total_hdr_len, &daddr);
1430 	} else if (strcmp(testname, "flags_urg") == 0) {
1431 		send_flags(txfd, &daddr, 0, 0, 0, 1, 0);
1432 		write_packet(txfd, fin_pkt, total_hdr_len, &daddr);
1433 	} else if (strcmp(testname, "flags_cwr") == 0) {
1434 		send_flags(txfd, &daddr, 0, 0, 0, 0, 1);
1435 		write_packet(txfd, fin_pkt, total_hdr_len, &daddr);
1436 
1437 	/* tcp sub-tests */
1438 	} else if (strcmp(testname, "tcp_csum") == 0) {
1439 		send_changed_checksum(txfd, &daddr);
1440 		usleep(fin_delay_us);
1441 		write_packet(txfd, fin_pkt, total_hdr_len, &daddr);
1442 	} else if (strcmp(testname, "tcp_seq") == 0) {
1443 		send_changed_seq(txfd, &daddr);
1444 		usleep(fin_delay_us);
1445 		write_packet(txfd, fin_pkt, total_hdr_len, &daddr);
1446 	} else if (strcmp(testname, "tcp_ts") == 0) {
1447 		send_changed_ts(txfd, &daddr);
1448 		usleep(fin_delay_us);
1449 		write_packet(txfd, fin_pkt, total_hdr_len, &daddr);
1450 	} else if (strcmp(testname, "tcp_opt") == 0) {
1451 		send_diff_opt(txfd, &daddr);
1452 		usleep(fin_delay_us);
1453 		write_packet(txfd, fin_pkt, total_hdr_len, &daddr);
1454 
1455 	/* ip sub-tests - shared between IPv4 and IPv6 */
1456 	} else if (strcmp(testname, "ip_ecn") == 0) {
1457 		send_changed_ECN(txfd, &daddr);
1458 		write_packet(txfd, fin_pkt, total_hdr_len, &daddr);
1459 	} else if (strcmp(testname, "ip_tos") == 0) {
1460 		send_changed_tos(txfd, &daddr);
1461 		write_packet(txfd, fin_pkt, total_hdr_len, &daddr);
1462 
1463 	/* ip sub-tests - IPv4 only */
1464 	} else if (strcmp(testname, "ip_csum") == 0) {
1465 		send_changed_ip_checksum(txfd, &daddr);
1466 		usleep(fin_delay_us);
1467 		write_packet(txfd, fin_pkt, total_hdr_len, &daddr);
1468 	} else if (strcmp(testname, "ip_ttl") == 0) {
1469 		send_changed_ttl(txfd, &daddr);
1470 		write_packet(txfd, fin_pkt, total_hdr_len, &daddr);
1471 	} else if (strcmp(testname, "ip_opt") == 0) {
1472 		send_ip_options(txfd, &daddr);
1473 		usleep(fin_delay_us);
1474 		write_packet(txfd, fin_pkt, total_hdr_len, &daddr);
1475 	} else if (strcmp(testname, "ip_frag4") == 0) {
1476 		send_fragment4(txfd, &daddr);
1477 		usleep(fin_delay_us);
1478 		write_packet(txfd, fin_pkt, total_hdr_len, &daddr);
1479 	} else if (strcmp(testname, "ip_id_df1_inc") == 0) {
1480 		send_flush_id_case(txfd, &daddr, FLUSH_ID_DF1_INC);
1481 		usleep(fin_delay_us);
1482 		write_packet(txfd, fin_pkt, total_hdr_len, &daddr);
1483 	} else if (strcmp(testname, "ip_id_df1_fixed") == 0) {
1484 		send_flush_id_case(txfd, &daddr, FLUSH_ID_DF1_FIXED);
1485 		usleep(fin_delay_us);
1486 		write_packet(txfd, fin_pkt, total_hdr_len, &daddr);
1487 	} else if (strcmp(testname, "ip_id_df0_inc") == 0) {
1488 		send_flush_id_case(txfd, &daddr, FLUSH_ID_DF0_INC);
1489 		usleep(fin_delay_us);
1490 		write_packet(txfd, fin_pkt, total_hdr_len, &daddr);
1491 	} else if (strcmp(testname, "ip_id_df0_fixed") == 0) {
1492 		send_flush_id_case(txfd, &daddr, FLUSH_ID_DF0_FIXED);
1493 		usleep(fin_delay_us);
1494 		write_packet(txfd, fin_pkt, total_hdr_len, &daddr);
1495 	} else if (strcmp(testname, "ip_id_df1_inc_fixed") == 0) {
1496 		send_flush_id_case(txfd, &daddr, FLUSH_ID_DF1_INC_FIXED);
1497 		usleep(fin_delay_us);
1498 		write_packet(txfd, fin_pkt, total_hdr_len, &daddr);
1499 	} else if (strcmp(testname, "ip_id_df1_fixed_inc") == 0) {
1500 		send_flush_id_case(txfd, &daddr, FLUSH_ID_DF1_FIXED_INC);
1501 		usleep(fin_delay_us);
1502 		write_packet(txfd, fin_pkt, total_hdr_len, &daddr);
1503 
1504 	/* ip sub-tests - IPv6 only */
1505 	} else if (strcmp(testname, "ip_frag6") == 0) {
1506 		send_fragment6(txfd, &daddr);
1507 		usleep(fin_delay_us);
1508 		write_packet(txfd, fin_pkt, total_hdr_len, &daddr);
1509 	} else if (strcmp(testname, "ip_v6ext_same") == 0) {
1510 		send_ipv6_exthdr(txfd, &daddr, EXT_PAYLOAD_1, EXT_PAYLOAD_1);
1511 		usleep(fin_delay_us);
1512 		write_packet(txfd, fin_pkt, total_hdr_len, &daddr);
1513 	} else if (strcmp(testname, "ip_v6ext_diff") == 0) {
1514 		send_ipv6_exthdr(txfd, &daddr, EXT_PAYLOAD_1, EXT_PAYLOAD_2);
1515 		usleep(fin_delay_us);
1516 		write_packet(txfd, fin_pkt, total_hdr_len, &daddr);
1517 
1518 	/* large sub-tests */
1519 	} else if (strcmp(testname, "large_max") == 0) {
1520 		int remainder = max_payload() % calc_mss();
1521 
1522 		send_large(txfd, &daddr, remainder);
1523 		write_packet(txfd, fin_pkt, total_hdr_len, &daddr);
1524 	} else if (strcmp(testname, "large_rem") == 0) {
1525 		int remainder = max_payload() % calc_mss();
1526 
1527 		send_large(txfd, &daddr, remainder + 1);
1528 		write_packet(txfd, fin_pkt, total_hdr_len, &daddr);
1529 
1530 	/* machinery sub-tests */
1531 	} else if (strcmp(testname, "single") == 0) {
1532 		static char buf[MAX_HDR_LEN + PAYLOAD_LEN];
1533 
1534 		create_packet(buf, 0, 0, PAYLOAD_LEN, 0);
1535 		write_packet(txfd, buf, total_hdr_len + PAYLOAD_LEN, &daddr);
1536 		write_packet(txfd, fin_pkt, total_hdr_len, &daddr);
1537 	} else if (strcmp(testname, "capacity") == 0) {
1538 		send_capacity(txfd, &daddr);
1539 		usleep(fin_delay_us);
1540 		write_packet(txfd, fin_pkt, total_hdr_len, &daddr);
1541 
1542 	/* PPPoE sub-tests */
1543 	} else if (strcmp(testname, "pppoe_sid") == 0) {
1544 		send_changed_pppoe_sid(txfd, &daddr);
1545 		usleep(fin_delay_us);
1546 		write_packet(txfd, fin_pkt, total_hdr_len, &daddr);
1547 
1548 	} else {
1549 		error(1, 0, "Unknown testcase: %s", testname);
1550 	}
1551 
1552 	if (close(txfd))
1553 		error(1, errno, "socket close");
1554 }
1555 
1556 static void gro_receiver(void)
1557 {
1558 	static int correct_payload[NUM_PACKETS];
1559 	int rxfd = -1;
1560 
1561 	rxfd = socket(PF_PACKET, SOCK_RAW, htons(ETH_P_NONE));
1562 	if (rxfd < 0)
1563 		error(1, 0, "socket creation");
1564 	setup_sock_filter(rxfd);
1565 	set_timeout(rxfd);
1566 	set_rcvbuf(rxfd);
1567 	bind_packetsocket(rxfd);
1568 
1569 	ksft_ready();
1570 
1571 	memset(correct_payload, 0, sizeof(correct_payload));
1572 
1573 	/* data sub-tests */
1574 	if (strcmp(testname, "data_same") == 0) {
1575 		printf("pure data packet of same size: ");
1576 		correct_payload[0] = PAYLOAD_LEN * 2;
1577 		check_recv_pkts(rxfd, correct_payload, 1);
1578 	} else if (strcmp(testname, "data_lrg_sml") == 0) {
1579 		printf("large data packets followed by a smaller one: ");
1580 		correct_payload[0] = PAYLOAD_LEN * 1.5;
1581 		check_recv_pkts(rxfd, correct_payload, 1);
1582 	} else if (strcmp(testname, "data_lrg_1byte") == 0) {
1583 		printf("large data packet followed by a 1 byte one: ");
1584 		correct_payload[0] = PAYLOAD_LEN + 1;
1585 		check_recv_pkts(rxfd, correct_payload, 1);
1586 	} else if (strcmp(testname, "data_sml_lrg") == 0) {
1587 		printf("small data packets followed by a larger one: ");
1588 		correct_payload[0] = PAYLOAD_LEN / 2;
1589 		correct_payload[1] = PAYLOAD_LEN;
1590 		check_recv_pkts(rxfd, correct_payload, 2);
1591 	} else if (strcmp(testname, "data_burst") == 0) {
1592 		printf("two bursts of two data packets: ");
1593 		correct_payload[0] = PAYLOAD_LEN * 2;
1594 		correct_payload[1] = PAYLOAD_LEN * 2;
1595 		check_recv_pkts(rxfd, correct_payload, 2);
1596 
1597 	/* ack test */
1598 	} else if (strcmp(testname, "ack") == 0) {
1599 		printf("duplicate ack and pure ack: ");
1600 		check_recv_pkts(rxfd, correct_payload, 3);
1601 
1602 	/* flags sub-tests */
1603 	} else if (strcmp(testname, "flags_psh") == 0) {
1604 		correct_payload[0] = PAYLOAD_LEN * 3;
1605 		correct_payload[1] = PAYLOAD_LEN * 2;
1606 		printf("psh flag ends coalescing: ");
1607 		check_recv_pkts(rxfd, correct_payload, 2);
1608 	} else if (strcmp(testname, "flags_syn") == 0) {
1609 		correct_payload[0] = PAYLOAD_LEN * 2;
1610 		correct_payload[1] = 0;
1611 		correct_payload[2] = PAYLOAD_LEN * 2;
1612 		printf("syn flag ends coalescing: ");
1613 		check_recv_pkts(rxfd, correct_payload, 3);
1614 	} else if (strcmp(testname, "flags_rst") == 0) {
1615 		correct_payload[0] = PAYLOAD_LEN * 2;
1616 		correct_payload[1] = 0;
1617 		correct_payload[2] = PAYLOAD_LEN * 2;
1618 		printf("rst flag ends coalescing: ");
1619 		check_recv_pkts(rxfd, correct_payload, 3);
1620 	} else if (strcmp(testname, "flags_urg") == 0) {
1621 		correct_payload[0] = PAYLOAD_LEN * 2;
1622 		correct_payload[1] = 0;
1623 		correct_payload[2] = PAYLOAD_LEN * 2;
1624 		printf("urg flag ends coalescing: ");
1625 		check_recv_pkts(rxfd, correct_payload, 3);
1626 	} else if (strcmp(testname, "flags_cwr") == 0) {
1627 		correct_payload[0] = PAYLOAD_LEN;
1628 		correct_payload[1] = PAYLOAD_LEN * 2;
1629 		correct_payload[2] = PAYLOAD_LEN * 2;
1630 		printf("cwr flag ends coalescing: ");
1631 		check_recv_pkts(rxfd, correct_payload, 3);
1632 
1633 	/* tcp sub-tests */
1634 	} else if (strcmp(testname, "tcp_csum") == 0) {
1635 		correct_payload[0] = PAYLOAD_LEN;
1636 		correct_payload[1] = PAYLOAD_LEN;
1637 		printf("changed checksum does not coalesce: ");
1638 		check_recv_pkts(rxfd, correct_payload, 2);
1639 	} else if (strcmp(testname, "tcp_seq") == 0) {
1640 		correct_payload[0] = PAYLOAD_LEN;
1641 		correct_payload[1] = PAYLOAD_LEN;
1642 		printf("Wrong Seq number doesn't coalesce: ");
1643 		check_recv_pkts(rxfd, correct_payload, 2);
1644 	} else if (strcmp(testname, "tcp_ts") == 0) {
1645 		correct_payload[0] = PAYLOAD_LEN * 2;
1646 		correct_payload[1] = PAYLOAD_LEN;
1647 		correct_payload[2] = PAYLOAD_LEN;
1648 		correct_payload[3] = PAYLOAD_LEN;
1649 		printf("Different timestamp doesn't coalesce: ");
1650 		check_recv_pkts(rxfd, correct_payload, 4);
1651 	} else if (strcmp(testname, "tcp_opt") == 0) {
1652 		correct_payload[0] = PAYLOAD_LEN * 2;
1653 		correct_payload[1] = PAYLOAD_LEN;
1654 		printf("Different options doesn't coalesce: ");
1655 		check_recv_pkts(rxfd, correct_payload, 2);
1656 
1657 	/* ip sub-tests - shared between IPv4 and IPv6 */
1658 	} else if (strcmp(testname, "ip_ecn") == 0) {
1659 		correct_payload[0] = PAYLOAD_LEN;
1660 		correct_payload[1] = PAYLOAD_LEN;
1661 		printf("different ECN doesn't coalesce: ");
1662 		check_recv_pkts(rxfd, correct_payload, 2);
1663 	} else if (strcmp(testname, "ip_tos") == 0) {
1664 		correct_payload[0] = PAYLOAD_LEN;
1665 		correct_payload[1] = PAYLOAD_LEN;
1666 		printf("different tos doesn't coalesce: ");
1667 		check_recv_pkts(rxfd, correct_payload, 2);
1668 
1669 	/* ip sub-tests - IPv4 only */
1670 	} else if (strcmp(testname, "ip_csum") == 0) {
1671 		correct_payload[0] = PAYLOAD_LEN;
1672 		correct_payload[1] = PAYLOAD_LEN;
1673 		correct_payload[2] = PAYLOAD_LEN;
1674 		printf("bad ip checksum doesn't coalesce: ");
1675 		check_recv_pkts(rxfd, correct_payload, 3);
1676 	} else if (strcmp(testname, "ip_ttl") == 0) {
1677 		correct_payload[0] = PAYLOAD_LEN;
1678 		correct_payload[1] = PAYLOAD_LEN;
1679 		printf("different ttl doesn't coalesce: ");
1680 		check_recv_pkts(rxfd, correct_payload, 2);
1681 	} else if (strcmp(testname, "ip_opt") == 0) {
1682 		correct_payload[0] = PAYLOAD_LEN;
1683 		correct_payload[1] = PAYLOAD_LEN;
1684 		correct_payload[2] = PAYLOAD_LEN;
1685 		printf("ip options doesn't coalesce: ");
1686 		check_recv_pkts(rxfd, correct_payload, 3);
1687 	} else if (strcmp(testname, "ip_frag4") == 0) {
1688 		correct_payload[0] = PAYLOAD_LEN;
1689 		correct_payload[1] = PAYLOAD_LEN;
1690 		printf("fragmented ip4 doesn't coalesce: ");
1691 		check_recv_pkts(rxfd, correct_payload, 2);
1692 	} else if (strcmp(testname, "ip_id_df1_inc") == 0) {
1693 		printf("DF=1, Incrementing - should coalesce: ");
1694 		correct_payload[0] = PAYLOAD_LEN * 2;
1695 		check_recv_pkts(rxfd, correct_payload, 1);
1696 	} else if (strcmp(testname, "ip_id_df1_fixed") == 0) {
1697 		printf("DF=1, Fixed - should coalesce: ");
1698 		correct_payload[0] = PAYLOAD_LEN * 2;
1699 		check_recv_pkts(rxfd, correct_payload, 1);
1700 	} else if (strcmp(testname, "ip_id_df0_inc") == 0) {
1701 		printf("DF=0, Incrementing - should coalesce: ");
1702 		correct_payload[0] = PAYLOAD_LEN * 2;
1703 		check_recv_pkts(rxfd, correct_payload, 1);
1704 	} else if (strcmp(testname, "ip_id_df0_fixed") == 0) {
1705 		printf("DF=0, Fixed - should coalesce: ");
1706 		correct_payload[0] = PAYLOAD_LEN * 2;
1707 		check_recv_pkts(rxfd, correct_payload, 1);
1708 	} else if (strcmp(testname, "ip_id_df1_inc_fixed") == 0) {
1709 		printf("DF=1, 2 Incrementing and one fixed - should coalesce only first 2 packets: ");
1710 		correct_payload[0] = PAYLOAD_LEN * 2;
1711 		correct_payload[1] = PAYLOAD_LEN;
1712 		check_recv_pkts(rxfd, correct_payload, 2);
1713 	} else if (strcmp(testname, "ip_id_df1_fixed_inc") == 0) {
1714 		printf("DF=1, 2 Fixed and one incrementing - should coalesce only first 2 packets: ");
1715 		correct_payload[0] = PAYLOAD_LEN * 2;
1716 		correct_payload[1] = PAYLOAD_LEN;
1717 		check_recv_pkts(rxfd, correct_payload, 2);
1718 
1719 	/* ip sub-tests - IPv6 only */
1720 	} else if (strcmp(testname, "ip_frag6") == 0) {
1721 		/* GRO doesn't check for ipv6 hop limit when flushing.
1722 		 * Hence no corresponding test to the ipv4 case.
1723 		 */
1724 		printf("fragmented ip6 doesn't coalesce: ");
1725 		correct_payload[0] = PAYLOAD_LEN * 2;
1726 		correct_payload[1] = PAYLOAD_LEN;
1727 		correct_payload[2] = PAYLOAD_LEN;
1728 		check_recv_pkts(rxfd, correct_payload, 3);
1729 	} else if (strcmp(testname, "ip_v6ext_same") == 0) {
1730 		printf("ipv6 with ext header does coalesce: ");
1731 		correct_payload[0] = PAYLOAD_LEN * 2;
1732 		check_recv_pkts(rxfd, correct_payload, 1);
1733 	} else if (strcmp(testname, "ip_v6ext_diff") == 0) {
1734 		printf("ipv6 with ext header with different payloads doesn't coalesce: ");
1735 		correct_payload[0] = PAYLOAD_LEN;
1736 		correct_payload[1] = PAYLOAD_LEN;
1737 		check_recv_pkts(rxfd, correct_payload, 2);
1738 
1739 	/* large sub-tests */
1740 	} else if (strcmp(testname, "large_max") == 0) {
1741 		int remainder = max_payload() % calc_mss();
1742 
1743 		correct_payload[0] = max_payload();
1744 		correct_payload[1] = remainder;
1745 		printf("Shouldn't coalesce if exceed IP max pkt size: ");
1746 		check_recv_pkts(rxfd, correct_payload, 2);
1747 	} else if (strcmp(testname, "large_rem") == 0) {
1748 		int remainder = max_payload() % calc_mss();
1749 
1750 		/* last segment sent individually, doesn't start new segment */
1751 		correct_payload[0] = max_payload() - remainder;
1752 		correct_payload[1] = remainder + 1;
1753 		correct_payload[2] = remainder + 1;
1754 		printf("last segment sent individually: ");
1755 		check_recv_pkts(rxfd, correct_payload, 3);
1756 
1757 	/* machinery sub-tests */
1758 	} else if (strcmp(testname, "single") == 0) {
1759 		printf("single data packet: ");
1760 		correct_payload[0] = PAYLOAD_LEN;
1761 		check_recv_pkts(rxfd, correct_payload, 1);
1762 	} else if (strcmp(testname, "capacity") == 0) {
1763 		check_capacity_pkts(rxfd);
1764 
1765 	} else if (strcmp(testname, "pppoe_sid") == 0) {
1766 		correct_payload[0] = PAYLOAD_LEN;
1767 		correct_payload[1] = PAYLOAD_LEN;
1768 		printf("different PPPoE session ID doesn't coalesce: ");
1769 		check_recv_pkts(rxfd, correct_payload, 2);
1770 
1771 	} else {
1772 		error(1, 0, "Test case error: unknown testname %s", testname);
1773 	}
1774 
1775 	if (close(rxfd))
1776 		error(1, 0, "socket close");
1777 }
1778 
1779 static void parse_args(int argc, char **argv)
1780 {
1781 	static const struct option opts[] = {
1782 		{ "daddr", required_argument, NULL, 'd' },
1783 		{ "dmac", required_argument, NULL, 'D' },
1784 		{ "iface", required_argument, NULL, 'i' },
1785 		{ "ipv4", no_argument, NULL, '4' },
1786 		{ "ipv6", no_argument, NULL, '6' },
1787 		{ "ipip", no_argument, NULL, 'e' },
1788 		{ "ip6ip6", no_argument, NULL, 'E' },
1789 		{ "pppoev4", no_argument, NULL, 'p' },
1790 		{ "pppoev6", no_argument, NULL, 'P' },
1791 		{ "num-flows", required_argument, NULL, 'n' },
1792 		{ "rx", no_argument, NULL, 'r' },
1793 		{ "saddr", required_argument, NULL, 's' },
1794 		{ "smac", required_argument, NULL, 'S' },
1795 		{ "test", required_argument, NULL, 't' },
1796 		{ "order-check", no_argument, NULL, 'o' },
1797 		{ "verbose", no_argument, NULL, 'v' },
1798 		{ 0, 0, 0, 0 }
1799 	};
1800 	int c;
1801 
1802 	while ((c = getopt_long(argc, argv, "46d:D:eEi:n:pPrs:S:t:ov", opts, NULL)) != -1) {
1803 		switch (c) {
1804 		case '4':
1805 			proto = PF_INET;
1806 			ethhdr_proto = htons(ETH_P_IP);
1807 			break;
1808 		case '6':
1809 			proto = PF_INET6;
1810 			ethhdr_proto = htons(ETH_P_IPV6);
1811 			break;
1812 		case 'e':
1813 			ipip = true;
1814 			proto = PF_INET;
1815 			ethhdr_proto = htons(ETH_P_IP);
1816 			break;
1817 		case 'E':
1818 			ip6ip6 = true;
1819 			proto = PF_INET6;
1820 			ethhdr_proto = htons(ETH_P_IPV6);
1821 			break;
1822 		case 'p':
1823 			pppoe = true;
1824 			proto = PF_INET;
1825 			ethhdr_proto = htons(ETH_P_PPP_SES);
1826 			break;
1827 		case 'P':
1828 			pppoe = true;
1829 			proto = PF_INET6;
1830 			ethhdr_proto = htons(ETH_P_PPP_SES);
1831 			break;
1832 		case 'd':
1833 			addr4_dst = addr6_dst = optarg;
1834 			break;
1835 		case 'D':
1836 			dmac = optarg;
1837 			break;
1838 		case 'i':
1839 			ifname = optarg;
1840 			break;
1841 		case 'n':
1842 			num_flows = atoi(optarg);
1843 			break;
1844 		case 'r':
1845 			tx_socket = false;
1846 			break;
1847 		case 's':
1848 			addr4_src = addr6_src = optarg;
1849 			break;
1850 		case 'S':
1851 			smac = optarg;
1852 			break;
1853 		case 't':
1854 			testname = optarg;
1855 			break;
1856 		case 'o':
1857 			order_check = true;
1858 			break;
1859 		case 'v':
1860 			verbose = true;
1861 			break;
1862 		default:
1863 			error(1, 0, "%s invalid option %c\n", __func__, c);
1864 			break;
1865 		}
1866 	}
1867 }
1868 
1869 int main(int argc, char **argv)
1870 {
1871 	parse_args(argc, argv);
1872 
1873 	if (ipip) {
1874 		tcp_offset = ETH_HLEN + sizeof(struct iphdr) * 2;
1875 		total_hdr_len = tcp_offset + sizeof(struct tcphdr);
1876 	} else if (ip6ip6) {
1877 		tcp_offset = ETH_HLEN + sizeof(struct ipv6hdr) * 2;
1878 		total_hdr_len = tcp_offset + sizeof(struct tcphdr);
1879 	} else if (pppoe) {
1880 		tcp_offset = ETH_HLEN + PPPOE_SES_HLEN +
1881 			(proto == PF_INET ? sizeof(struct iphdr) : sizeof(struct ipv6hdr));
1882 		total_hdr_len = tcp_offset + sizeof(struct tcphdr);
1883 	} else if (proto == PF_INET) {
1884 		tcp_offset = ETH_HLEN + sizeof(struct iphdr);
1885 		total_hdr_len = tcp_offset + sizeof(struct tcphdr);
1886 	} else if (proto == PF_INET6) {
1887 		tcp_offset = ETH_HLEN + sizeof(struct ipv6hdr);
1888 		total_hdr_len = tcp_offset + sizeof(struct tcphdr);
1889 	} else {
1890 		error(1, 0, "Protocol family is not ipv4 or ipv6");
1891 	}
1892 
1893 	read_MAC(src_mac, smac);
1894 	read_MAC(dst_mac, dmac);
1895 
1896 	if (tx_socket) {
1897 		gro_sender();
1898 	} else {
1899 		/* Only the receiver exit status determines test success. */
1900 		gro_receiver();
1901 		fprintf(stderr, "Gro::%s test passed.\n", testname);
1902 	}
1903 
1904 	return 0;
1905 }
1906