trafgen: fix packet socket initialization with multiple CPUs
[netsniff-ng.git] / trafgen.c
blob9c5a9a672cf6c17c98902c563221ce8abef49984
1 /*
2 * netsniff-ng - the packet sniffing beast
3 * Copyright 2011 - 2013 Daniel Borkmann <dborkma@tik.ee.ethz.ch>,
4 * Swiss federal institute of technology (ETH Zurich)
5 * Subject to the GPL, version 2.
6 */
8 #define _GNU_SOURCE
10 #include <stdio.h>
11 #include <string.h>
12 #include <getopt.h>
13 #include <ctype.h>
14 #include <stdbool.h>
15 #include <sched.h>
16 #include <sys/socket.h>
17 #include <sys/types.h>
18 #include <sys/fsuid.h>
19 #include <sys/prctl.h>
20 #include <sys/stat.h>
21 #include <sys/wait.h>
22 #include <sys/mman.h>
23 #include <net/ethernet.h>
24 #include <netinet/in.h>
25 #include <netinet/ip.h>
26 #include <linux/icmp.h>
27 #include <linux/if.h>
28 #include <arpa/inet.h>
29 #include <signal.h>
30 #include <stdint.h>
31 #include <stdlib.h>
32 #include <fcntl.h>
33 #include <time.h>
34 #include <poll.h>
35 #include <netdb.h>
36 #include <math.h>
37 #include <unistd.h>
39 #include "xmalloc.h"
40 #include "die.h"
41 #include "str.h"
42 #include "sig.h"
43 #include "sock.h"
44 #include "cpus.h"
45 #include "lockme.h"
46 #include "privs.h"
47 #include "proc.h"
48 #include "ioops.h"
49 #include "irq.h"
50 #include "config.h"
51 #include "built_in.h"
52 #include "trafgen_conf.h"
53 #include "tprintf.h"
54 #include "timer.h"
55 #include "ring_tx.h"
56 #include "csum.h"
57 #include "trafgen_proto.h"
58 #include "pcap_io.h"
59 #include "trafgen_dev.h"
61 enum shaper_type {
62 SHAPER_NONE,
63 SHAPER_DELAY,
64 SHAPER_PKTS,
65 SHAPER_BYTES,
66 SHAPER_TSTAMP,
69 struct shaper {
70 enum shaper_type type;
71 unsigned long long sent;
72 unsigned long long rate;
73 struct timeval tstamp;
74 struct timespec delay;
75 struct timeval start;
76 struct timeval end;
79 struct ctx {
80 bool rand, rfraw, jumbo_support, verbose, smoke_test, enforce, qdisc_path;
81 size_t reserve_size;
82 struct dev_io *dev_out;
83 struct dev_io *dev_in;
84 unsigned long num;
85 unsigned int cpus;
86 uid_t uid; gid_t gid;
87 char *device, *rhost;
88 struct sockaddr_in dest;
89 struct shaper sh;
90 char *packet_str;
91 char *pcap_in;
94 struct cpu_stats {
95 unsigned long tv_sec, tv_usec;
96 unsigned long long tx_packets, tx_bytes;
97 unsigned long long cf_packets, cf_bytes;
98 unsigned long long cd_packets;
99 sig_atomic_t state;
102 static sig_atomic_t sigint = 0;
104 struct packet *packets = NULL;
105 size_t plen = 0;
107 struct packet_dyn *packet_dyn = NULL;
108 size_t dlen = 0;
110 static const char *short_options = "d:c:n:t:vJhS:rk:i:o:VRs:P:eE:pu:g:CHQqD:b:";
111 static const struct option long_options[] = {
112 {"dev", required_argument, NULL, 'd'},
113 {"out", required_argument, NULL, 'o'},
114 {"in", required_argument, NULL, 'i'},
115 {"conf", required_argument, NULL, 'c'},
116 {"num", required_argument, NULL, 'n'},
117 {"gap", required_argument, NULL, 't'},
118 {"rate", required_argument, NULL, 'b'},
119 {"cpus", required_argument, NULL, 'P'},
120 {"ring-size", required_argument, NULL, 'S'},
121 {"kernel-pull", required_argument, NULL, 'k'},
122 {"smoke-test", required_argument, NULL, 's'},
123 {"seed", required_argument, NULL, 'E'},
124 {"user", required_argument, NULL, 'u'},
125 {"group", required_argument, NULL, 'g'},
126 {"prio-high", no_argument, NULL, 'H'},
127 {"notouch-irq", no_argument, NULL, 'Q'},
128 {"no-sock-mem", no_argument, NULL, 'A'},
129 {"qdisc-path", no_argument, NULL, 'q'},
130 {"jumbo-support", no_argument, NULL, 'J'},
131 {"no-cpu-stats", no_argument, NULL, 'C'},
132 {"cpp", no_argument, NULL, 'p'},
133 {"define", required_argument, NULL, 'D'},
134 {"rfraw", no_argument, NULL, 'R'},
135 {"rand", no_argument, NULL, 'r'},
136 {"verbose", no_argument, NULL, 'V'},
137 {"version", no_argument, NULL, 'v'},
138 {"example", no_argument, NULL, 'e'},
139 {"help", no_argument, NULL, 'h'},
140 {NULL, 0, NULL, 0}
143 static const char *copyright = "Please report bugs to <netsniff-ng@googlegroups.com>\n"
144 "Copyright (C) 2011-2013 Daniel Borkmann <dborkma@tik.ee.ethz.ch>,\n"
145 "Swiss federal institute of technology (ETH Zurich)\n"
146 "License: GNU GPL version 2.0\n"
147 "This is free software: you are free to change and redistribute it.\n"
148 "There is NO WARRANTY, to the extent permitted by law.";
150 static struct cpu_stats *stats;
151 static unsigned int seed;
153 #define CPU_STATS_STATE_CFG 1
154 #define CPU_STATS_STATE_CHK 2
155 #define CPU_STATS_STATE_RES 4
157 #ifndef ICMP_FILTER
158 # define ICMP_FILTER 1
160 struct icmp_filter {
161 __u32 data;
163 #endif
165 #define SMOKE_N_PROBES 100
167 #define PKT_MIN_LEN 14
169 static void signal_handler(int number)
171 switch (number) {
172 case SIGINT:
173 case SIGQUIT:
174 case SIGTERM:
175 sigint = 1;
176 case SIGHUP:
177 default:
178 break;
182 static void __noreturn help(void)
184 printf("trafgen %s, multithreaded zero-copy network packet generator\n", VERSION_STRING);
185 puts("http://www.netsniff-ng.org\n\n"
186 "Usage: trafgen [options] [packet]\n"
187 "Options:\n"
188 " -i|-c|--in|--conf <cfg/-> Packet configuration file/stdin\n"
189 " -o|-d|--out|--dev <netdev|.cfg|.pcap> Networking device or configuration file i.e., eth0\n"
190 " -p|--cpp Run packet config through C preprocessor\n"
191 " -D|--define Add macro/define for C preprocessor\n"
192 " -J|--jumbo-support Support 64KB super jumbo frames (def: 2048B)\n"
193 " -R|--rfraw Inject raw 802.11 frames\n"
194 " -s|--smoke-test <ipv4> Probe if machine survived fuzz-tested packet\n"
195 " -n|--num <uint> Number of packets until exit (def: 0)\n"
196 " -r|--rand Randomize packet selection (def: round robin)\n"
197 " -P|--cpus <uint> Specify number of forks(<= CPUs) (def: #CPUs)\n"
198 " -t|--gap <time> Set approx. interpacket gap (s/ms/us/ns, def: us)\n"
199 " -b|--rate <rate> Send traffic at specified rate (pps/B/kB/MB/GB/kbit/Mbit/Gbit/KiB/MiB/GiB)\n"
200 " -S|--ring-size <size> Manually set mmap size (KiB/MiB/GiB)\n"
201 " -E|--seed <uint> Manually set srand(3) seed\n"
202 " -u|--user <userid> Drop privileges and change to userid\n"
203 " -g|--group <groupid> Drop privileges and change to groupid\n"
204 " -H|--prio-high Make this high priority process\n"
205 " -A|--no-sock-mem Don't tune core socket memory\n"
206 " -Q|--notouch-irq Do not touch IRQ CPU affinity of NIC\n"
207 " -q|--qdisc-path Enable qdisc kernel path (default off since 3.14)\n"
208 " -V|--verbose Be more verbose\n"
209 " -C|--no-cpu-stats Do not print CPU time statistics on exit\n"
210 " -v|--version Show version and exit\n"
211 " -e|--example Show built-in packet config example\n"
212 " -h|--help Guess what?!\n\n"
213 "Examples:\n"
214 " trafgen --dev eth0 --conf trafgen.cfg\n"
215 " trafgen -e | trafgen -i - -o eth0 --cpp -n 1\n"
216 " trafgen --dev eth0 --conf fuzzing.cfg --smoke-test 10.0.0.1\n"
217 " trafgen --dev wlan0 --rfraw --conf beacon-test.txf -V --cpus 2\n"
218 " trafgen --dev eth0 --conf frag_dos.cfg --rand --gap 1000us\n"
219 " trafgen --dev eth0 --conf icmp.cfg --rand --num 1400000 -k1000\n"
220 " trafgen --dev eth0 --conf tcp_syn.cfg -u `id -u bob` -g `id -g bob`\n"
221 " trafgen --dev eth0 '{ fill(0xff, 6), 0x00, 0x02, 0xb3, rnd(3), c16(0x0800), fill(0xca, 64) }'\n\n"
222 "Arbitrary packet config examples (e.g. trafgen -e > trafgen.cfg):\n"
223 " Run packet on all CPUs: { fill(0xff, 64) csum16(0, 64) }\n"
224 " Run packet only on CPU1: cpu(1): { rnd(64), 0b11001100, 0xaa }\n"
225 " Run packet only on CPU1-2: cpu(1-2): { drnd(64),'a',csum16(1, 8),'b',42 }\n\n"
226 "Generate config files from existing pcap using netsniff-ng:\n"
227 " netsniff-ng --in dump.pcap --out dump.cfg\n\n"
228 "Note:\n"
229 " Smoke/fuzz test example: machine A, 10.0.0.2 (trafgen) is directly\n"
230 " connected to machine B (test kernel), 10.0.0.1. If ICMP reply fails\n"
231 " we assume the kernel crashed, thus we print the packet and quit.\n"
232 " In case you find a ping-of-death, please mention trafgen in your\n"
233 " commit message of the fix!\n\n"
234 " For introducing bit errors, delays with random variation and more,\n"
235 " make use of tc(8) with its different disciplines, i.e. netem.\n\n"
236 " For generating different package distributions, you can use scripting\n"
237 " to generate a trafgen config file with packet ratios as:\n\n"
238 " IMIX 64:7, 570:4, 1518:1\n"
239 " Tolly 64:55, 78:5, 576:17, 1518:23\n"
240 " Cisco 64:7, 594:4, 1518:1\n"
241 " RPR Trimodal 64:60, 512:20, 1518:20\n"
242 " RPR Quadrimodal 64:50, 512:15, 1518:15, 9218:20\n");
243 puts(copyright);
244 die();
247 static void __noreturn example(void)
249 const char *e =
250 "/* Note: dynamic elements make trafgen slower! */\n"
251 "#include <stddef.h>\n\n"
252 "{\n"
253 " /* MAC Destination */\n"
254 " fill(0xff, ETH_ALEN),\n"
255 " /* MAC Source */\n"
256 " 0x00, 0x02, 0xb3, drnd(3),\n"
257 " /* IPv4 Protocol */\n"
258 " c16(ETH_P_IP),\n"
259 " /* IPv4 Version, IHL, TOS */\n"
260 " 0b01000101, 0,\n"
261 " /* IPv4 Total Len */\n"
262 " c16(59),\n"
263 " /* IPv4 Ident */\n"
264 " drnd(2),\n"
265 " /* IPv4 Flags, Frag Off */\n"
266 " 0b01000000, 0,\n"
267 " /* IPv4 TTL */\n"
268 " 64,\n"
269 " /* Proto TCP */\n"
270 " 0x06,\n"
271 " /* IPv4 Checksum (IP header from, to) */\n"
272 " csumip(14, 33),\n"
273 " /* Source IP */\n"
274 " drnd(4),\n"
275 " /* Dest IP */\n"
276 " drnd(4),\n"
277 " /* TCP Source Port */\n"
278 " drnd(2),\n"
279 " /* TCP Dest Port */\n"
280 " c16(80),\n"
281 " /* TCP Sequence Number */\n"
282 " drnd(4),\n"
283 " /* TCP Ackn. Number */\n"
284 " c32(0),\n"
285 " /* TCP Header length + TCP SYN/ECN Flag */\n"
286 " c16((8 << 12) | TCP_FLAG_SYN | TCP_FLAG_ECE)\n"
287 " /* Window Size */\n"
288 " c16(16),\n"
289 " /* TCP Checksum (offset IP, offset TCP) */\n"
290 " csumtcp(14, 34),\n"
291 " /* TCP Options */\n"
292 " 0x00, 0x00, 0x01, 0x01, 0x08, 0x0a, 0x06,\n"
293 " 0x91, 0x68, 0x7d, 0x06, 0x91, 0x68, 0x6f,\n"
294 " /* Data blob */\n"
295 " \"gotcha!\",\n"
296 "}";
297 puts(e);
298 die();
301 static void __noreturn version(void)
303 printf("trafgen %s, Git id: %s\n", VERSION_LONG, GITVERSION);
304 puts("multithreaded zero-copy network packet generator\n"
305 "http://www.netsniff-ng.org\n");
306 puts(copyright);
307 die();
310 static void apply_counter(int id)
312 size_t j, counter_max = packet_dyn[id].clen;
314 for (j = 0; j < counter_max; ++j) {
315 uint8_t val;
316 struct counter *counter = &packet_dyn[id].cnt[j];
318 val = counter->val - counter->min;
320 switch (counter->type) {
321 case TYPE_INC:
322 val = (val + counter->inc) % (counter->max - counter->min + 1);
323 break;
324 case TYPE_DEC:
325 val = (val - counter->inc) % (counter->min - counter->max + 1);
326 break;
327 default:
328 bug();
331 counter->val = val + counter->min;
332 packets[id].payload[counter->off] = val;
336 static void apply_randomizer(int id)
338 size_t j, rand_max = packet_dyn[id].rlen;
340 for (j = 0; j < rand_max; ++j) {
341 uint8_t val = (uint8_t) rand();
342 struct randomizer *randomizer = &packet_dyn[id].rnd[j];
344 packets[id].payload[randomizer->off] = val;
348 static void apply_csum16(int id)
350 size_t j, csum_max = packet_dyn[id].slen;
352 for (j = 0; j < csum_max; ++j) {
353 uint16_t sum = 0;
354 struct csum16 *csum = &packet_dyn[id].csum[j];
356 fmemset(&packets[id].payload[csum->off], 0, sizeof(sum));
357 if (unlikely((size_t) csum->to >= packets[id].len))
358 csum->to = packets[id].len - 1;
360 switch (csum->which) {
361 case CSUM_IP:
362 sum = calc_csum(packets[id].payload + csum->from,
363 csum->to - csum->from + 1);
364 break;
365 case CSUM_UDP:
366 sum = p4_csum((void *) packets[id].payload + csum->from,
367 packets[id].payload + csum->to,
368 (packets[id].len - csum->to),
369 IPPROTO_UDP);
370 break;
371 case CSUM_TCP:
372 sum = p4_csum((void *) packets[id].payload + csum->from,
373 packets[id].payload + csum->to,
374 (packets[id].len - csum->to),
375 IPPROTO_TCP);
376 break;
377 case CSUM_UDP6:
378 sum = p6_csum((void *) packets[id].payload + csum->from,
379 packets[id].payload + csum->to,
380 (packets[id].len - csum->to),
381 IPPROTO_UDP);
382 break;
383 case CSUM_TCP6:
384 sum = p6_csum((void *) packets[id].payload + csum->from,
385 packets[id].payload + csum->to,
386 (packets[id].len - csum->to),
387 IPPROTO_TCP);
388 break;
389 default:
390 bug();
391 break;
394 fmemcpy(&packets[id].payload[csum->off], &sum, sizeof(sum));
398 static void preprocess_packets(void)
400 size_t i;
402 for (i = 0; i < plen; i++) {
403 struct packet_dyn *pktd = &packet_dyn[i];
405 if (packet_dyn_has_only_csums(pktd)) {
406 apply_csum16(i);
407 pktd->slen = 0;
408 xfree(pktd->csum);
413 static struct cpu_stats *setup_shared_var(unsigned int cpus)
415 int fd;
416 size_t len = cpus * sizeof(struct cpu_stats);
417 char *zbuff, file[256];
418 struct cpu_stats *buff;
420 slprintf(file, sizeof(file), ".tmp_mmap.XXXXXX");
421 fd = mkostemp_or_die(file, O_RDWR | O_CREAT | O_TRUNC);
422 zbuff = xzmalloc(len);
423 write_or_die(fd, zbuff, len);
424 xfree(zbuff);
426 buff = mmap(NULL, len, PROT_READ | PROT_WRITE,
427 MAP_SHARED, fd, 0);
428 if (buff == MAP_FAILED)
429 panic("Cannot setup shared variable!\n");
431 close(fd);
432 unlink(file);
434 memset(buff, 0, len);
435 return buff;
438 static void destroy_shared_var(void *buff, unsigned int cpus)
440 munmap(buff, cpus * sizeof(struct cpu_stats));
443 static void dump_trafgen_snippet(uint8_t *payload, size_t len)
445 size_t i;
447 printf("{");
448 for (i = 0; i < len; ++i) {
449 if (i % 15 == 0)
450 printf("\n ");
451 printf("0x%02x, ", payload[i]);
453 printf("\n}\n");
454 fflush(stdout);
457 static int xmit_smoke_setup(struct ctx *ctx)
459 int icmp_sock, ret, ttl = 64;
460 struct icmp_filter filter;
462 icmp_sock = socket(AF_INET, SOCK_RAW, IPPROTO_ICMP);
463 if (icmp_sock < 0)
464 panic("Cannot get a ICMP socket: %s!\n", strerror(errno));
466 filter.data = ~(1 << ICMP_ECHOREPLY);
468 ret = setsockopt(icmp_sock, SOL_RAW, ICMP_FILTER, &filter, sizeof(filter));
469 if (ret < 0)
470 panic("Cannot install filter!\n");
472 ret = setsockopt(icmp_sock, SOL_IP, IP_TTL, &ttl, sizeof(ttl));
473 if (ret < 0)
474 panic("Cannot set TTL!\n");
476 memset(&ctx->dest, 0, sizeof(ctx->dest));
477 ctx->dest.sin_family = AF_INET;
478 ctx->dest.sin_port = 0;
480 ret = inet_aton(ctx->rhost, &ctx->dest.sin_addr);
481 if (ret < 0)
482 panic("Cannot resolve address!\n");
484 return icmp_sock;
487 static int xmit_smoke_probe(int icmp_sock, struct ctx *ctx)
489 int ret;
490 unsigned int i, j;
491 short ident, cnt = 1, idstore[SMOKE_N_PROBES];
492 uint8_t outpack[512], *data;
493 struct icmphdr *icmp;
494 struct iphdr *ip;
495 size_t len = sizeof(*icmp) + 56;
496 struct sockaddr_in from;
497 socklen_t from_len;
498 struct pollfd fds = {
499 .fd = icmp_sock,
500 .events = POLLIN,
503 fmemset(idstore, 0, sizeof(idstore));
504 for (j = 0; j < SMOKE_N_PROBES; j++) {
505 while ((ident = htons((short) rand())) == 0)
506 sleep(0);
507 idstore[j] = ident;
509 memset(outpack, 0, sizeof(outpack));
510 icmp = (void *) outpack;
511 icmp->type = ICMP_ECHO;
512 icmp->un.echo.id = ident;
513 icmp->un.echo.sequence = htons(cnt++);
515 data = ((uint8_t *) outpack + sizeof(*icmp));
516 for (i = 0; i < 56; ++i)
517 data[i] = (uint8_t) rand();
519 icmp->checksum = csum((unsigned short *) outpack,
520 len / sizeof(unsigned short));
522 ret = sendto(icmp_sock, outpack, len, MSG_DONTWAIT,
523 (struct sockaddr *) &ctx->dest, sizeof(ctx->dest));
524 if (unlikely(ret != (int) len))
525 panic("Cannot send out probe: %s!\n", strerror(errno));
527 ret = poll(&fds, 1, 50);
528 if (ret < 0)
529 panic("Poll failed!\n");
531 if (fds.revents & POLLIN) {
532 ret = recvfrom(icmp_sock, outpack, sizeof(outpack), 0,
533 (struct sockaddr *) &from, &from_len);
534 if (unlikely(ret <= 0))
535 panic("Probe receive failed!\n");
536 if (unlikely(from_len != sizeof(ctx->dest)))
537 continue;
538 if (unlikely(memcmp(&from, &ctx->dest, sizeof(ctx->dest))))
539 continue;
540 if (unlikely((size_t) ret < sizeof(*ip) + sizeof(*icmp)))
541 continue;
542 ip = (void *) outpack;
543 if (unlikely(ip->ihl * 4 + sizeof(*icmp) > (size_t) ret))
544 continue;
545 icmp = (void *) outpack + ip->ihl * 4;
546 for (i = 0; i < array_size(idstore); ++i) {
547 if (unlikely(icmp->un.echo.id != idstore[i]))
548 continue;
549 return 0;
554 return -1;
557 static bool shaper_is_set(struct shaper *sh)
559 return sh->type != SHAPER_NONE;
562 static void shaper_init(struct shaper *sh)
564 if (sh->type == SHAPER_NONE || sh->type == SHAPER_DELAY)
565 return;
567 memset(&sh->delay, 0, sizeof(struct timespec));
568 bug_on(gettimeofday(&sh->start, NULL));
569 sh->sent = 0;
572 static void shaper_set_delay(struct shaper *sh, time_t sec, long int ns)
574 if (!(sec | ns)) {
575 sh->type = SHAPER_NONE;
576 return;
579 sh->type = SHAPER_DELAY;
580 sh->delay.tv_sec = sec;
581 sh->delay.tv_nsec = ns;
584 static void shaper_set_rate(struct shaper *sh, unsigned long long rate,
585 enum shaper_type type)
587 memset(sh, 0, sizeof(struct shaper));
588 sh->rate = rate;
589 sh->type = type;
592 static void shaper_set_tstamp(struct shaper *sh, struct timespec *ts)
594 TIMESPEC_TO_TIMEVAL(&sh->tstamp, ts);
597 static void shaper_delay(struct shaper *sh, struct packet *pkt)
599 if (sh->type == SHAPER_BYTES || sh->type == SHAPER_PKTS) {
600 unsigned long pkt_len = pkt->len;
602 sh->sent += sh->type == SHAPER_BYTES ? pkt_len : 1;
604 if (sh->sent >= sh->rate && sh->rate > 0) {
605 struct timeval delay_us;
606 struct timeval time_sent;
607 struct timeval time_1s = { .tv_sec = 1 };
609 bug_on(gettimeofday(&sh->end, NULL));
610 timersub(&sh->end, &sh->start, &time_sent);
612 if (timercmp(&time_1s, &time_sent, > )) {
613 timersub(&time_1s, &time_sent, &delay_us);
614 TIMEVAL_TO_TIMESPEC(&delay_us, &sh->delay);
617 } else if (sh->type == SHAPER_TSTAMP) {
618 struct timeval tstamp;
619 struct timeval pkt_diff;
620 struct timeval diff;
622 bug_on(gettimeofday(&sh->end, NULL));
623 TIMESPEC_TO_TIMEVAL(&tstamp, &pkt->tstamp);
624 timersub(&sh->end, &sh->start, &diff);
625 timersub(&tstamp, &sh->tstamp, &pkt_diff);
627 if (timercmp(&diff, &pkt_diff, <)) {
628 struct timeval delay;
630 timersub(&pkt_diff, &diff, &delay);
631 TIMEVAL_TO_TIMESPEC(&delay, &sh->delay);
634 memcpy(&sh->tstamp, &tstamp, sizeof(sh->tstamp));
637 if ((sh->delay.tv_sec | sh->delay.tv_nsec) > 0) {
638 nanosleep(&sh->delay, NULL);
640 shaper_init(sh);
644 static inline void packet_apply_dyn_elements(int idx)
646 if (packet_dyn_has_elems(&packet_dyn[idx])) {
647 apply_counter(idx);
648 apply_randomizer(idx);
649 apply_csum16(idx);
652 if (packet_dyn_has_fields(&packet_dyn[idx])) {
653 uint32_t i;
655 for (i = 0; i < packet_dyn[idx].flen; i++)
656 proto_field_dyn_apply(packet_dyn[idx].fields[i]);
658 proto_packet_update(idx);
662 static void xmit_slowpath_or_die(struct ctx *ctx, unsigned int cpu, unsigned long orig_num)
664 int ret, icmp_sock = -1;
665 unsigned long num = 1, i = 0;
666 struct timeval start, end, diff;
667 unsigned long long tx_bytes = 0, tx_packets = 0;
669 if (ctx->num > 0)
670 num = ctx->num;
671 if (ctx->num == 0 && orig_num > 0)
672 num = 0;
674 if (ctx->smoke_test)
675 icmp_sock = xmit_smoke_setup(ctx);
677 drop_privileges(ctx->enforce, ctx->uid, ctx->gid);
679 bug_on(gettimeofday(&start, NULL));
681 if (shaper_is_set(&ctx->sh))
682 shaper_init(&ctx->sh);
684 while (likely(sigint == 0 && num > 0 && plen > 0)) {
685 packet_apply_dyn_elements(i);
686 retry:
687 ret = dev_io_write(ctx->dev_out, &packets[i]);
688 if (unlikely(ret < 0)) {
689 if (errno == ENOBUFS) {
690 sched_yield();
691 goto retry;
693 if (ctx->smoke_test)
694 panic("Sendto error: %s!\n", strerror(errno));
697 tx_bytes += packets[i].len;
698 tx_packets++;
700 if (ctx->smoke_test) {
701 ret = xmit_smoke_probe(icmp_sock, ctx);
702 if (unlikely(ret < 0)) {
703 printf("%sSmoke test alert:%s\n", colorize_start(bold), colorize_end());
704 printf(" Remote host seems to be unresponsive to ICMP probes!\n");
705 printf(" Last instance was packet%lu, seed:%u, trafgen snippet:\n\n",
706 i, seed);
708 dump_trafgen_snippet(packets[i].payload, packets[i].len);
709 break;
713 if (!ctx->rand) {
714 i++;
715 if (i >= plen)
716 i = 0;
717 } else
718 i = rand() % plen;
720 if (ctx->num > 0)
721 num--;
723 if (shaper_is_set(&ctx->sh))
724 shaper_delay(&ctx->sh, &packets[i]);
727 bug_on(gettimeofday(&end, NULL));
728 timersub(&end, &start, &diff);
730 if (ctx->smoke_test)
731 close(icmp_sock);
733 stats[cpu].tx_packets = tx_packets;
734 stats[cpu].tx_bytes = tx_bytes;
735 stats[cpu].tv_sec = diff.tv_sec;
736 stats[cpu].tv_usec = diff.tv_usec;
738 stats[cpu].state |= CPU_STATS_STATE_RES;
741 static void xmit_fastpath_or_die(struct ctx *ctx, unsigned int cpu, unsigned long orig_num)
743 int ifindex = dev_io_ifindex_get(ctx->dev_out);
744 uint8_t *out = NULL;
745 unsigned int it = 0, retry = 100;
746 unsigned long num = 1, i = 0;
747 size_t size = ring_size(dev_io_name_get(ctx->dev_out), ctx->reserve_size);
748 struct ring tx_ring;
749 struct frame_map *hdr;
750 struct timeval start, end, diff;
751 unsigned long long tx_bytes = 0, tx_packets = 0;
752 int sock = dev_io_fd_get(ctx->dev_out);
754 set_sock_prio(sock, 512);
756 ring_tx_setup(&tx_ring, sock, size, ifindex, ctx->jumbo_support, ctx->verbose);
758 drop_privileges(ctx->enforce, ctx->uid, ctx->gid);
760 if (ctx->num > 0)
761 num = ctx->num;
762 if (ctx->num == 0 && orig_num > 0)
763 num = 0;
765 bug_on(gettimeofday(&start, NULL));
767 while (likely(sigint == 0 && num > 0 && plen > 0)) {
768 if (!user_may_pull_from_tx(tx_ring.frames[it].iov_base)) {
769 int ret = pull_and_flush_tx_ring(sock);
770 if (unlikely(ret < 0)) {
771 /* We could hit EBADF if the socket has been closed before
772 * the timer was triggered.
774 if (errno != EBADF && errno != ENOBUFS)
775 panic("Flushing TX_RING failed: %s!\n", strerror(errno));
778 continue;
781 hdr = tx_ring.frames[it].iov_base;
782 out = ((uint8_t *) hdr) + TPACKET2_HDRLEN - sizeof(struct sockaddr_ll);
784 hdr->tp_h.tp_snaplen = packets[i].len;
785 hdr->tp_h.tp_len = packets[i].len;
787 packet_apply_dyn_elements(i);
789 fmemcpy(out, packets[i].payload, packets[i].len);
791 tx_bytes += packets[i].len;
792 tx_packets++;
794 if (!ctx->rand) {
795 i++;
796 if (i >= plen)
797 i = 0;
798 } else
799 i = rand() % plen;
801 kernel_may_pull_from_tx(&hdr->tp_h);
803 it++;
804 if (it >= tx_ring.layout.tp_frame_nr)
805 it = 0;
807 if (ctx->num > 0)
808 num--;
811 bug_on(gettimeofday(&end, NULL));
812 timersub(&end, &start, &diff);
814 while (pull_and_flush_tx_ring_wait(sock) < 0 && errno == ENOBUFS && retry-- > 0)
815 usleep(10000);
816 destroy_tx_ring(sock, &tx_ring);
818 stats[cpu].tx_packets = tx_packets;
819 stats[cpu].tx_bytes = tx_bytes;
820 stats[cpu].tv_sec = diff.tv_sec;
821 stats[cpu].tv_usec = diff.tv_usec;
823 stats[cpu].state |= CPU_STATS_STATE_RES;
826 static inline void __set_state(unsigned int cpu, sig_atomic_t s)
828 stats[cpu].state = s;
831 static inline sig_atomic_t __get_state(unsigned int cpu)
833 return stats[cpu].state;
836 static unsigned long __wait_and_sum_others(struct ctx *ctx, unsigned int cpu)
838 unsigned int i;
839 unsigned long total;
841 for (i = 0, total = plen; i < ctx->cpus; i++) {
842 if (i == cpu)
843 continue;
845 while ((__get_state(i) &
846 (CPU_STATS_STATE_CFG |
847 CPU_STATS_STATE_RES)) == 0 &&
848 sigint == 0)
849 sched_yield();
851 total += stats[i].cf_packets;
854 return total;
857 static void __correct_global_delta(struct ctx *ctx, unsigned int cpu, unsigned long orig)
859 unsigned int i;
860 unsigned long total;
861 int cpu_sel;
862 long long delta_correction = 0;
864 for (i = 0, total = ctx->num; i < ctx->cpus; i++) {
865 if (i == cpu)
866 continue;
868 while ((__get_state(i) &
869 (CPU_STATS_STATE_CHK |
870 CPU_STATS_STATE_RES)) == 0 &&
871 sigint == 0)
872 sched_yield();
874 total += stats[i].cd_packets;
877 if (total > orig)
878 delta_correction = -1 * ((long long) total - orig);
879 if (total < orig)
880 delta_correction = +1 * ((long long) orig - total);
882 for (cpu_sel = -1, i = 0; i < ctx->cpus; i++) {
883 if (stats[i].cd_packets > 0) {
884 if ((long long) stats[i].cd_packets +
885 delta_correction >= 0) {
886 cpu_sel = i;
887 break;
892 if ((int) cpu == cpu_sel)
893 ctx->num += delta_correction;
896 static void __set_state_cf(unsigned int cpu, unsigned long p, unsigned long b,
897 sig_atomic_t s)
899 stats[cpu].cf_packets = p;
900 stats[cpu].cf_bytes = b;
901 stats[cpu].state = s;
904 static void __set_state_cd(unsigned int cpu, unsigned long p, sig_atomic_t s)
906 stats[cpu].cd_packets = p;
907 stats[cpu].state = s;
910 static void xmit_packet_precheck(struct ctx *ctx, unsigned int cpu)
912 unsigned long plen_total, orig = ctx->num;
913 size_t total_len = 0;
914 unsigned int i;
916 bug_on(plen != dlen);
918 for (i = 0; i < plen; ++i)
919 total_len += packets[i].len;
921 __set_state_cf(cpu, plen, total_len, CPU_STATS_STATE_CFG);
922 plen_total = __wait_and_sum_others(ctx, cpu);
924 if (orig > 0) {
925 ctx->num = (unsigned long) round((1.0 * plen / plen_total) * orig);
927 __set_state_cd(cpu, ctx->num, CPU_STATS_STATE_CHK |
928 CPU_STATS_STATE_CFG);
929 __correct_global_delta(ctx, cpu, orig);
932 if (plen == 0) {
933 __set_state(cpu, CPU_STATS_STATE_RES);
934 return;
938 static void pcap_load_packets(struct dev_io *dev)
940 struct packet *pkt;
942 while ((pkt = dev_io_read(dev)) != 0)
943 /* nothing to do */;
946 static void main_loop(struct ctx *ctx, char *confname, bool slow,
947 unsigned int cpu, bool invoke_cpp, char **cpp_argv,
948 unsigned long orig_num)
950 if (ctx->dev_in && dev_io_is_pcap(ctx->dev_in)) {
951 pcap_load_packets(ctx->dev_in);
952 shaper_set_tstamp(&ctx->sh, &packets[0].tstamp);
953 ctx->num = plen;
954 } else {
955 if (ctx->packet_str)
956 compile_packets_str(ctx->packet_str, ctx->verbose, cpu);
957 else
958 compile_packets(confname, ctx->verbose, cpu, invoke_cpp, cpp_argv);
960 preprocess_packets();
963 xmit_packet_precheck(ctx, cpu);
965 if (cpu == 0) {
966 unsigned int i;
967 size_t total_len = 0, total_pkts = 0;
969 for (i = 0; i < ctx->cpus; ++i) {
970 total_len += stats[i].cf_bytes;
971 total_pkts += stats[i].cf_packets;
974 printf("%6zu packets to schedule\n", total_pkts);
975 printf("%6zu bytes in total\n", total_len);
976 printf("Running! Hang up with ^C!\n\n");
977 fflush(stdout);
980 dev_io_open(ctx->dev_out);
981 if (dev_io_is_netdev(ctx->dev_out) && ctx->qdisc_path == false)
982 set_sock_qdisc_bypass(dev_io_fd_get(ctx->dev_out), ctx->verbose);
984 if (slow)
985 xmit_slowpath_or_die(ctx, cpu, orig_num);
986 else
987 xmit_fastpath_or_die(ctx, cpu, orig_num);
989 cleanup_packets();
992 static unsigned int generate_srand_seed(void)
994 int fd;
995 unsigned int _seed;
997 fd = open("/dev/urandom", O_RDONLY);
998 if (fd < 0)
999 return time(NULL);
1001 read_or_die(fd, &_seed, sizeof(_seed));
1003 close(fd);
1004 return _seed;
1007 static void on_panic_del_rfmon(void *arg)
1009 dev_io_close(arg);
1012 int main(int argc, char **argv)
1014 bool slow = false, invoke_cpp = false, reseed = true, cpustats = true;
1015 bool prio_high = false, set_irq_aff = true, set_sock_mem = true;
1016 int c, vals[4] = {0}, irq;
1017 uint64_t gap = 0;
1018 unsigned int i;
1019 char *confname = NULL, *ptr;
1020 unsigned long cpus_tmp, orig_num = 0;
1021 unsigned long long tx_packets, tx_bytes;
1022 struct ctx ctx;
1023 int min_opts = 5;
1024 char **cpp_argv = NULL;
1025 size_t cpp_argc = 0;
1026 unsigned long long rate;
1027 enum shaper_type shape_type;
1028 struct timespec delay;
1030 fmemset(&ctx, 0, sizeof(ctx));
1031 ctx.cpus = get_number_cpus_online();
1032 ctx.uid = getuid();
1033 ctx.gid = getgid();
1034 ctx.qdisc_path = false;
1036 /* Keep an initial small default size to reduce cache-misses. */
1037 ctx.reserve_size = 512 * (1 << 10);
1039 while ((c = getopt_long(argc, argv, short_options, long_options,
1040 NULL)) != EOF) {
1041 switch (c) {
1042 case 'h':
1043 help();
1044 break;
1045 case 'v':
1046 version();
1047 break;
1048 case 'C':
1049 cpustats = false;
1050 break;
1051 case 'e':
1052 example();
1053 break;
1054 case 'p':
1055 invoke_cpp = true;
1056 break;
1057 case 'D':
1058 cpp_argv = argv_insert(cpp_argv, &cpp_argc, "-D");
1059 cpp_argv = argv_insert(cpp_argv, &cpp_argc, optarg);
1060 break;
1061 case 'V':
1062 ctx.verbose = true;
1063 break;
1064 case 'P':
1065 cpus_tmp = strtoul(optarg, NULL, 0);
1066 if (cpus_tmp > 0 && cpus_tmp < ctx.cpus)
1067 ctx.cpus = cpus_tmp;
1068 break;
1069 case 'd':
1070 case 'o':
1071 ctx.device = xstrdup(optarg);
1072 break;
1073 case 'H':
1074 prio_high = true;
1075 break;
1076 case 'A':
1077 set_sock_mem = false;
1078 break;
1079 case 'Q':
1080 set_irq_aff = false;
1081 break;
1082 case 'q':
1083 ctx.qdisc_path = true;
1084 break;
1085 case 'r':
1086 ctx.rand = true;
1087 break;
1088 case 's':
1089 slow = true;
1090 ctx.cpus = 1;
1091 ctx.smoke_test = true;
1092 ctx.rhost = xstrdup(optarg);
1093 break;
1094 case 'R':
1095 ctx.rfraw = true;
1096 break;
1097 case 'J':
1098 ctx.jumbo_support = true;
1099 break;
1100 case 'i':
1101 confname = xstrdup(optarg);
1102 if (strstr(confname, ".pcap")) {
1103 ctx.sh.type = SHAPER_TSTAMP;
1104 ctx.pcap_in = confname;
1105 break;
1107 case 'c':
1108 if (!strncmp("-", confname, strlen("-")))
1109 ctx.cpus = 1;
1110 break;
1111 case 'u':
1112 ctx.uid = strtoul(optarg, NULL, 0);
1113 ctx.enforce = true;
1114 break;
1115 case 'g':
1116 ctx.gid = strtoul(optarg, NULL, 0);
1117 ctx.enforce = true;
1118 break;
1119 case 'k':
1120 printf("Option -k/--kernel-pull is no longer used and "
1121 "will be removed in a future release!\n");
1122 break;
1123 case 'E':
1124 seed = strtoul(optarg, NULL, 0);
1125 reseed = false;
1126 break;
1127 case 'n':
1128 orig_num = strtoul(optarg, NULL, 0);
1129 ctx.num = orig_num;
1130 break;
1131 case 't':
1132 gap = strtoul(optarg, &ptr, 0);
1133 if (!gap && optarg == ptr)
1134 panic("Invalid gap param\n");
1136 if (!strncmp(ptr, "ns", strlen("ns"))) {
1137 delay.tv_sec = gap / 1000000000;
1138 delay.tv_nsec = gap % 1000000000;
1139 } else if (*ptr == '\0' || !strncmp(ptr, "us", strlen("us"))) {
1140 /* Default to microseconds for backwards
1141 * compatibility if no postfix is given.
1143 delay.tv_sec = gap / 1000000;
1144 delay.tv_nsec = (gap % 1000000) * 1000;
1145 } else if (!strncmp(ptr, "ms", strlen("ms"))) {
1146 delay.tv_sec = gap / 1000;
1147 delay.tv_nsec = (gap % 1000) * 1000000;
1148 } else if (!strncmp(ptr, "s", strlen("s"))) {
1149 delay.tv_sec = gap;
1150 delay.tv_nsec = 0;
1151 } else {
1152 panic("Syntax error in time param!\n");
1155 shaper_set_delay(&ctx.sh, delay.tv_sec, delay.tv_nsec);
1156 break;
1157 case 'b':
1158 rate = strtoul(optarg, &ptr, 0);
1159 if (!rate && optarg == ptr)
1160 panic("Invalid rate param\n");
1162 if (strncmp(ptr, "pps", strlen("pps")) == 0) {
1163 shape_type = SHAPER_PKTS;
1164 } else if (strncmp(ptr, "B", strlen("B")) == 0) {
1165 shape_type = SHAPER_BYTES;
1166 } else if (strncmp(ptr, "kB", strlen("kB")) == 0) {
1167 shape_type = SHAPER_BYTES;
1168 rate *= 1000;
1169 } else if (strncmp(ptr, "MB", strlen("MB")) == 0) {
1170 shape_type = SHAPER_BYTES;
1171 rate *= 1000 * 1000;
1172 } else if (strncmp(ptr, "GB", strlen("GB")) == 0) {
1173 shape_type = SHAPER_BYTES;
1174 rate *= 1000 * 1000 * 1000;
1175 } else if (strncmp(ptr, "kbit", strlen("kbit")) == 0) {
1176 shape_type = SHAPER_BYTES;
1177 rate *= 1000 / 8;
1178 } else if (strncmp(ptr, "Mbit", strlen("Mbit")) == 0) {
1179 shape_type = SHAPER_BYTES;
1180 rate *= 1000 * 1000 / 8;
1181 } else if (strncmp(ptr, "Gbit", strlen("Gbit")) == 0) {
1182 shape_type = SHAPER_BYTES;
1183 rate *= 1000 * 1000 * 1000 / 8;
1184 } else if (strncmp(ptr, "KiB", strlen("KiB")) == 0) {
1185 shape_type = SHAPER_BYTES;
1186 rate *= 1 << 10;
1187 } else if (strncmp(ptr, "MiB", strlen("MiB")) == 0) {
1188 shape_type = SHAPER_BYTES;
1189 rate *= 1 << 20;
1190 } else if (strncmp(ptr, "GiB", strlen("GiB")) == 0) {
1191 shape_type = SHAPER_BYTES;
1192 rate *= 1 << 30;
1193 } else if (!rate) {
1194 shape_type = SHAPER_NONE;
1195 } else {
1196 panic("Invalid unit type for rate\n");
1199 shaper_set_rate(&ctx.sh, rate, shape_type);
1200 break;
1201 case 'S':
1202 ctx.reserve_size = strtoul(optarg, &ptr, 0);
1203 if (ctx.reserve_size == 0 && ptr == optarg)
1204 panic("Invalid ring size param\n");
1206 if (!strncmp(ptr, "KiB", strlen("KiB")))
1207 ctx.reserve_size *= 1 << 10;
1208 else if (!strncmp(ptr, "MiB", strlen("MiB")))
1209 ctx.reserve_size = 1 << 20;
1210 else if (!strncmp(ptr, "GiB", strlen("GiB")))
1211 ctx.reserve_size *= 1 << 30;
1212 else
1213 panic("Invalid ring size unit type\n");
1215 break;
1216 case '?':
1217 switch (optopt) {
1218 case 'd':
1219 case 'c':
1220 case 'n':
1221 case 'S':
1222 case 's':
1223 case 'P':
1224 case 'o':
1225 case 'E':
1226 case 'i':
1227 case 'k':
1228 case 'u':
1229 case 'g':
1230 case 't':
1231 panic("Option -%c requires an argument!\n",
1232 optopt);
1233 default:
1234 if (isprint(optopt))
1235 printf("Unknown option character `0x%X\'!\n", optopt);
1236 die();
1238 default:
1239 break;
1243 if (argc >= optind) {
1244 min_opts = 4;
1245 ctx.packet_str = argv2str(optind, argc, argv);
1248 if (argc < min_opts)
1249 help();
1250 if (ctx.device == NULL)
1251 panic("No networking device given!\n");
1252 if (confname == NULL && !ctx.packet_str)
1253 panic("No configuration file or packet string given!\n");
1255 register_signal(SIGINT, signal_handler);
1256 register_signal(SIGQUIT, signal_handler);
1257 register_signal(SIGTERM, signal_handler);
1258 register_signal(SIGHUP, signal_handler);
1260 if (prio_high) {
1261 set_proc_prio(-20);
1262 set_sched_status(SCHED_FIFO, sched_get_priority_max(SCHED_FIFO));
1265 if (set_sock_mem)
1266 set_system_socket_memory(vals, array_size(vals));
1267 xlockme();
1269 if (ctx.pcap_in) {
1270 ctx.dev_in = dev_io_create(ctx.pcap_in, DEV_IO_IN);
1271 if (!ctx.dev_in)
1272 panic("Failed to open input device\n");
1273 dev_io_open(ctx.dev_in);
1276 ctx.dev_out = dev_io_create(ctx.device, DEV_IO_OUT);
1277 if (!ctx.dev_out)
1278 panic("Failed to open output device\n");
1280 if (ctx.rfraw) {
1281 if (dev_io_link_type_set(ctx.dev_out, LINKTYPE_IEEE802_11_RADIOTAP))
1282 panic("Failed to setup rfraw device\n");
1284 panic_handler_add(on_panic_del_rfmon, ctx.dev_out);
1287 protos_init(ctx.dev_out);
1289 if (shaper_is_set(&ctx.sh) || (ctx.dev_in && !dev_io_is_netdev(ctx.dev_in))
1290 || !dev_io_is_netdev(ctx.dev_out)) {
1292 prctl(PR_SET_TIMERSLACK, 1UL);
1293 /* Fall back to single core to not mess up correct timing.
1294 * We are slow anyway!
1296 ctx.cpus = 1;
1297 slow = true;
1301 * If number of packets is smaller than number of CPUs use only as
1302 * many CPUs as there are packets. Otherwise we end up sending more
1303 * packets than intended or none at all.
1305 if (ctx.num)
1306 ctx.cpus = min_t(unsigned int, ctx.num, ctx.cpus);
1308 if (set_irq_aff && dev_io_is_netdev(ctx.dev_out)) {
1309 irq = device_irq_number(ctx.device);
1310 device_set_irq_affinity_list(irq, 0, ctx.cpus - 1);
1313 stats = setup_shared_var(ctx.cpus);
1315 for (i = 0; i < ctx.cpus; i++) {
1316 pid_t pid = fork();
1318 switch (pid) {
1319 case 0:
1320 if (reseed)
1321 seed = generate_srand_seed();
1322 srand(seed);
1324 cpu_affinity(i);
1325 main_loop(&ctx, confname, slow, i, invoke_cpp,
1326 cpp_argv, orig_num);
1328 goto thread_out;
1329 case -1:
1330 panic("Cannot fork processes!\n");
1334 for (i = 0; i < ctx.cpus; i++) {
1335 int status;
1337 wait(&status);
1338 if (WEXITSTATUS(status) == EXIT_FAILURE)
1339 die();
1342 if (set_sock_mem)
1343 reset_system_socket_memory(vals, array_size(vals));
1345 for (i = 0, tx_packets = tx_bytes = 0; i < ctx.cpus; i++) {
1346 while ((__get_state(i) & CPU_STATS_STATE_RES) == 0)
1347 sched_yield();
1349 tx_packets += stats[i].tx_packets;
1350 tx_bytes += stats[i].tx_bytes;
1353 fflush(stdout);
1354 printf("\n");
1355 printf("\r%12llu packets outgoing\n", tx_packets);
1356 printf("\r%12llu bytes outgoing\n", tx_bytes);
1357 for (i = 0; cpustats && i < ctx.cpus; i++) {
1358 printf("\r%12lu sec, %lu usec on CPU%d (%llu packets)\n",
1359 stats[i].tv_sec, stats[i].tv_usec, i,
1360 stats[i].tx_packets);
1363 thread_out:
1364 xunlockme();
1365 destroy_shared_var(stats, ctx.cpus);
1366 if (dev_io_is_netdev(ctx.dev_out) && set_irq_aff)
1367 device_restore_irq_affinity_list();
1369 dev_io_close(ctx.dev_out);
1370 if (ctx.dev_in)
1371 dev_io_close(ctx.dev_in);
1373 argv_free(cpp_argv);
1374 free(ctx.device);
1375 free(ctx.rhost);
1376 free(confname);
1377 free(ctx.packet_str);
1379 return 0;