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.
16 #include <sys/socket.h>
17 #include <sys/types.h>
18 #include <sys/fsuid.h>
19 #include <sys/prctl.h>
23 #include <net/ethernet.h>
24 #include <netinet/in.h>
25 #include <netinet/ip.h>
26 #include <linux/icmp.h>
28 #include <arpa/inet.h>
53 #include "trafgen_conf.h"
58 #include "trafgen_proto.h"
60 #include "trafgen_dev.h"
71 enum shaper_type type
;
72 unsigned long long sent
;
73 unsigned long long rate
;
74 struct timeval tstamp
;
75 struct timespec delay
;
81 bool rand
, rfraw
, jumbo_support
, verbose
, smoke_test
, enforce
, qdisc_path
;
83 struct dev_io
*dev_out
;
84 struct dev_io
*dev_in
;
88 char *device
, *device_trans
, *rhost
;
89 struct sockaddr_in dest
;
96 unsigned long tv_sec
, tv_usec
;
97 unsigned long long tx_packets
, tx_bytes
;
98 unsigned long long cf_packets
, cf_bytes
;
99 unsigned long long cd_packets
;
103 static sig_atomic_t sigint
= 0;
105 struct packet
*packets
= NULL
;
108 struct packet_dyn
*packet_dyn
= NULL
;
111 static const char *short_options
= "d:c:n:t:vJhS:rk:i:o:VRs:P:eE:pu:g:CHQqD:b:";
112 static const struct option long_options
[] = {
113 {"dev", required_argument
, NULL
, 'd'},
114 {"out", required_argument
, NULL
, 'o'},
115 {"in", required_argument
, NULL
, 'i'},
116 {"conf", required_argument
, NULL
, 'c'},
117 {"num", required_argument
, NULL
, 'n'},
118 {"gap", required_argument
, NULL
, 't'},
119 {"rate", required_argument
, NULL
, 'b'},
120 {"cpus", required_argument
, NULL
, 'P'},
121 {"ring-size", required_argument
, NULL
, 'S'},
122 {"kernel-pull", required_argument
, NULL
, 'k'},
123 {"smoke-test", required_argument
, NULL
, 's'},
124 {"seed", required_argument
, NULL
, 'E'},
125 {"user", required_argument
, NULL
, 'u'},
126 {"group", required_argument
, NULL
, 'g'},
127 {"prio-high", no_argument
, NULL
, 'H'},
128 {"notouch-irq", no_argument
, NULL
, 'Q'},
129 {"no-sock-mem", no_argument
, NULL
, 'A'},
130 {"qdisc-path", no_argument
, NULL
, 'q'},
131 {"jumbo-support", no_argument
, NULL
, 'J'},
132 {"no-cpu-stats", no_argument
, NULL
, 'C'},
133 {"cpp", no_argument
, NULL
, 'p'},
134 {"define", required_argument
, NULL
, 'D'},
135 {"rfraw", no_argument
, NULL
, 'R'},
136 {"rand", no_argument
, NULL
, 'r'},
137 {"verbose", no_argument
, NULL
, 'V'},
138 {"version", no_argument
, NULL
, 'v'},
139 {"example", no_argument
, NULL
, 'e'},
140 {"help", no_argument
, NULL
, 'h'},
144 static const char *copyright
= "Please report bugs to <netsniff-ng@googlegroups.com>\n"
145 "Copyright (C) 2011-2013 Daniel Borkmann <dborkma@tik.ee.ethz.ch>,\n"
146 "Swiss federal institute of technology (ETH Zurich)\n"
147 "License: GNU GPL version 2.0\n"
148 "This is free software: you are free to change and redistribute it.\n"
149 "There is NO WARRANTY, to the extent permitted by law.";
151 static struct cpu_stats
*stats
;
152 static unsigned int seed
;
154 #define CPU_STATS_STATE_CFG 1
155 #define CPU_STATS_STATE_CHK 2
156 #define CPU_STATS_STATE_RES 4
159 # define ICMP_FILTER 1
166 #define SMOKE_N_PROBES 100
168 #define PKT_MIN_LEN 14
170 static void signal_handler(int number
)
183 static void __noreturn
help(void)
185 printf("trafgen %s, multithreaded zero-copy network packet generator\n", VERSION_STRING
);
186 puts("http://www.netsniff-ng.org\n\n"
187 "Usage: trafgen [options] [packet]\n"
189 " -i|-c|--in|--conf <cfg/-> Packet configuration file/stdin\n"
190 " -o|-d|--out|--dev <netdev> Networking device i.e., eth0\n"
191 " -p|--cpp Run packet config through C preprocessor\n"
192 " -D|--define Add macro/define for C preprocessor\n"
193 " -J|--jumbo-support Support 64KB super jumbo frames (def: 2048B)\n"
194 " -R|--rfraw Inject raw 802.11 frames\n"
195 " -s|--smoke-test <ipv4> Probe if machine survived fuzz-tested packet\n"
196 " -n|--num <uint> Number of packets until exit (def: 0)\n"
197 " -r|--rand Randomize packet selection (def: round robin)\n"
198 " -P|--cpus <uint> Specify number of forks(<= CPUs) (def: #CPUs)\n"
199 " -t|--gap <time> Set approx. interpacket gap (s/ms/us/ns, def: us)\n"
200 " -b|--rate <rate> Send traffic at specified rate (pps/B/kB/MB/GB/kbit/Mbit/Gbit/KiB/MiB/GiB)\n"
201 " -S|--ring-size <size> Manually set mmap size (KiB/MiB/GiB)\n"
202 " -E|--seed <uint> Manually set srand(3) seed\n"
203 " -u|--user <userid> Drop privileges and change to userid\n"
204 " -g|--group <groupid> Drop privileges and change to groupid\n"
205 " -H|--prio-high Make this high priority process\n"
206 " -A|--no-sock-mem Don't tune core socket memory\n"
207 " -Q|--notouch-irq Do not touch IRQ CPU affinity of NIC\n"
208 " -q|--qdisc-path Enable qdisc kernel path (default off since 3.14)\n"
209 " -V|--verbose Be more verbose\n"
210 " -C|--no-cpu-stats Do not print CPU time statistics on exit\n"
211 " -v|--version Show version and exit\n"
212 " -e|--example Show built-in packet config example\n"
213 " -h|--help Guess what?!\n\n"
215 " trafgen --dev eth0 --conf trafgen.cfg\n"
216 " trafgen -e | trafgen -i - -o eth0 --cpp -n 1\n"
217 " trafgen --dev eth0 --conf fuzzing.cfg --smoke-test 10.0.0.1\n"
218 " trafgen --dev wlan0 --rfraw --conf beacon-test.txf -V --cpus 2\n"
219 " trafgen --dev eth0 --conf frag_dos.cfg --rand --gap 1000us\n"
220 " trafgen --dev eth0 --conf icmp.cfg --rand --num 1400000 -k1000\n"
221 " trafgen --dev eth0 --conf tcp_syn.cfg -u `id -u bob` -g `id -g bob`\n"
222 " trafgen --dev eth0 '{ fill(0xff, 6), 0x00, 0x02, 0xb3, rnd(3), c16(0x0800), fill(0xca, 64) }'\n\n"
223 "Arbitrary packet config examples (e.g. trafgen -e > trafgen.cfg):\n"
224 " Run packet on all CPUs: { fill(0xff, 64) csum16(0, 64) }\n"
225 " Run packet only on CPU1: cpu(1): { rnd(64), 0b11001100, 0xaa }\n"
226 " Run packet only on CPU1-2: cpu(1-2): { drnd(64),'a',csum16(1, 8),'b',42 }\n\n"
227 "Generate config files from existing pcap using netsniff-ng:\n"
228 " netsniff-ng --in dump.pcap --out dump.cfg\n\n"
230 " Smoke/fuzz test example: machine A, 10.0.0.2 (trafgen) is directly\n"
231 " connected to machine B (test kernel), 10.0.0.1. If ICMP reply fails\n"
232 " we assume the kernel crashed, thus we print the packet and quit.\n"
233 " In case you find a ping-of-death, please mention trafgen in your\n"
234 " commit message of the fix!\n\n"
235 " For introducing bit errors, delays with random variation and more,\n"
236 " make use of tc(8) with its different disciplines, i.e. netem.\n\n"
237 " For generating different package distributions, you can use scripting\n"
238 " to generate a trafgen config file with packet ratios as:\n\n"
239 " IMIX 64:7, 570:4, 1518:1\n"
240 " Tolly 64:55, 78:5, 576:17, 1518:23\n"
241 " Cisco 64:7, 594:4, 1518:1\n"
242 " RPR Trimodal 64:60, 512:20, 1518:20\n"
243 " RPR Quadrimodal 64:50, 512:15, 1518:15, 9218:20\n");
248 static void __noreturn
example(void)
251 "/* Note: dynamic elements make trafgen slower! */\n"
252 "#include <stddef.h>\n\n"
254 " /* MAC Destination */\n"
255 " fill(0xff, ETH_ALEN),\n"
256 " /* MAC Source */\n"
257 " 0x00, 0x02, 0xb3, drnd(3),\n"
258 " /* IPv4 Protocol */\n"
260 " /* IPv4 Version, IHL, TOS */\n"
262 " /* IPv4 Total Len */\n"
264 " /* IPv4 Ident */\n"
266 " /* IPv4 Flags, Frag Off */\n"
272 " /* IPv4 Checksum (IP header from, to) */\n"
278 " /* TCP Source Port */\n"
280 " /* TCP Dest Port */\n"
282 " /* TCP Sequence Number */\n"
284 " /* TCP Ackn. Number */\n"
286 " /* TCP Header length + TCP SYN/ECN Flag */\n"
287 " c16((8 << 12) | TCP_FLAG_SYN | TCP_FLAG_ECE)\n"
288 " /* Window Size */\n"
290 " /* TCP Checksum (offset IP, offset TCP) */\n"
291 " csumtcp(14, 34),\n"
292 " /* TCP Options */\n"
293 " 0x00, 0x00, 0x01, 0x01, 0x08, 0x0a, 0x06,\n"
294 " 0x91, 0x68, 0x7d, 0x06, 0x91, 0x68, 0x6f,\n"
302 static void __noreturn
version(void)
304 printf("trafgen %s, Git id: %s\n", VERSION_LONG
, GITVERSION
);
305 puts("multithreaded zero-copy network packet generator\n"
306 "http://www.netsniff-ng.org\n");
311 static void apply_counter(int id
)
313 size_t j
, counter_max
= packet_dyn
[id
].clen
;
315 for (j
= 0; j
< counter_max
; ++j
) {
317 struct counter
*counter
= &packet_dyn
[id
].cnt
[j
];
319 val
= counter
->val
- counter
->min
;
321 switch (counter
->type
) {
323 val
= (val
+ counter
->inc
) % (counter
->max
- counter
->min
+ 1);
326 val
= (val
- counter
->inc
) % (counter
->min
- counter
->max
+ 1);
332 counter
->val
= val
+ counter
->min
;
333 packets
[id
].payload
[counter
->off
] = val
;
337 static void apply_randomizer(int id
)
339 size_t j
, rand_max
= packet_dyn
[id
].rlen
;
341 for (j
= 0; j
< rand_max
; ++j
) {
342 uint8_t val
= (uint8_t) rand();
343 struct randomizer
*randomizer
= &packet_dyn
[id
].rnd
[j
];
345 packets
[id
].payload
[randomizer
->off
] = val
;
349 static void apply_csum16(int id
)
351 size_t j
, csum_max
= packet_dyn
[id
].slen
;
353 for (j
= 0; j
< csum_max
; ++j
) {
355 struct csum16
*csum
= &packet_dyn
[id
].csum
[j
];
357 fmemset(&packets
[id
].payload
[csum
->off
], 0, sizeof(sum
));
358 if (unlikely((size_t) csum
->to
>= packets
[id
].len
))
359 csum
->to
= packets
[id
].len
- 1;
361 switch (csum
->which
) {
363 sum
= calc_csum(packets
[id
].payload
+ csum
->from
,
364 csum
->to
- csum
->from
+ 1);
367 sum
= p4_csum((void *) packets
[id
].payload
+ csum
->from
,
368 packets
[id
].payload
+ csum
->to
,
369 (packets
[id
].len
- csum
->to
),
373 sum
= p4_csum((void *) packets
[id
].payload
+ csum
->from
,
374 packets
[id
].payload
+ csum
->to
,
375 (packets
[id
].len
- csum
->to
),
379 sum
= p6_csum((void *) packets
[id
].payload
+ csum
->from
,
380 packets
[id
].payload
+ csum
->to
,
381 (packets
[id
].len
- csum
->to
),
385 sum
= p6_csum((void *) packets
[id
].payload
+ csum
->from
,
386 packets
[id
].payload
+ csum
->to
,
387 (packets
[id
].len
- csum
->to
),
395 fmemcpy(&packets
[id
].payload
[csum
->off
], &sum
, sizeof(sum
));
399 static void preprocess_packets(void)
403 for (i
= 0; i
< plen
; i
++) {
404 struct packet_dyn
*pktd
= &packet_dyn
[i
];
406 if (packet_dyn_has_only_csums(pktd
)) {
414 static struct cpu_stats
*setup_shared_var(unsigned int cpus
)
417 size_t len
= cpus
* sizeof(struct cpu_stats
);
418 char *zbuff
, file
[256];
419 struct cpu_stats
*buff
;
421 slprintf(file
, sizeof(file
), ".tmp_mmap.XXXXXX");
422 fd
= mkostemp_or_die(file
, O_RDWR
| O_CREAT
| O_TRUNC
);
423 zbuff
= xzmalloc(len
);
424 write_or_die(fd
, zbuff
, len
);
427 buff
= mmap(NULL
, len
, PROT_READ
| PROT_WRITE
,
429 if (buff
== MAP_FAILED
)
430 panic("Cannot setup shared variable!\n");
435 memset(buff
, 0, len
);
439 static void destroy_shared_var(void *buff
, unsigned int cpus
)
441 munmap(buff
, cpus
* sizeof(struct cpu_stats
));
444 static void dump_trafgen_snippet(uint8_t *payload
, size_t len
)
449 for (i
= 0; i
< len
; ++i
) {
452 printf("0x%02x, ", payload
[i
]);
458 static int xmit_smoke_setup(struct ctx
*ctx
)
460 int icmp_sock
, ret
, ttl
= 64;
461 struct icmp_filter filter
;
463 icmp_sock
= socket(AF_INET
, SOCK_RAW
, IPPROTO_ICMP
);
465 panic("Cannot get a ICMP socket: %s!\n", strerror(errno
));
467 filter
.data
= ~(1 << ICMP_ECHOREPLY
);
469 ret
= setsockopt(icmp_sock
, SOL_RAW
, ICMP_FILTER
, &filter
, sizeof(filter
));
471 panic("Cannot install filter!\n");
473 ret
= setsockopt(icmp_sock
, SOL_IP
, IP_TTL
, &ttl
, sizeof(ttl
));
475 panic("Cannot set TTL!\n");
477 memset(&ctx
->dest
, 0, sizeof(ctx
->dest
));
478 ctx
->dest
.sin_family
= AF_INET
;
479 ctx
->dest
.sin_port
= 0;
481 ret
= inet_aton(ctx
->rhost
, &ctx
->dest
.sin_addr
);
483 panic("Cannot resolve address!\n");
488 static int xmit_smoke_probe(int icmp_sock
, struct ctx
*ctx
)
492 short ident
, cnt
= 1, idstore
[SMOKE_N_PROBES
];
493 uint8_t outpack
[512], *data
;
494 struct icmphdr
*icmp
;
496 size_t len
= sizeof(*icmp
) + 56;
497 struct sockaddr_in from
;
499 struct pollfd fds
= {
504 fmemset(idstore
, 0, sizeof(idstore
));
505 for (j
= 0; j
< SMOKE_N_PROBES
; j
++) {
506 while ((ident
= htons((short) rand())) == 0)
510 memset(outpack
, 0, sizeof(outpack
));
511 icmp
= (void *) outpack
;
512 icmp
->type
= ICMP_ECHO
;
513 icmp
->un
.echo
.id
= ident
;
514 icmp
->un
.echo
.sequence
= htons(cnt
++);
516 data
= ((uint8_t *) outpack
+ sizeof(*icmp
));
517 for (i
= 0; i
< 56; ++i
)
518 data
[i
] = (uint8_t) rand();
520 icmp
->checksum
= csum((unsigned short *) outpack
,
521 len
/ sizeof(unsigned short));
523 ret
= sendto(icmp_sock
, outpack
, len
, MSG_DONTWAIT
,
524 (struct sockaddr
*) &ctx
->dest
, sizeof(ctx
->dest
));
525 if (unlikely(ret
!= (int) len
))
526 panic("Cannot send out probe: %s!\n", strerror(errno
));
528 ret
= poll(&fds
, 1, 50);
530 panic("Poll failed!\n");
532 if (fds
.revents
& POLLIN
) {
533 ret
= recvfrom(icmp_sock
, outpack
, sizeof(outpack
), 0,
534 (struct sockaddr
*) &from
, &from_len
);
535 if (unlikely(ret
<= 0))
536 panic("Probe receive failed!\n");
537 if (unlikely(from_len
!= sizeof(ctx
->dest
)))
539 if (unlikely(memcmp(&from
, &ctx
->dest
, sizeof(ctx
->dest
))))
541 if (unlikely((size_t) ret
< sizeof(*ip
) + sizeof(*icmp
)))
543 ip
= (void *) outpack
;
544 if (unlikely(ip
->ihl
* 4 + sizeof(*icmp
) > (size_t) ret
))
546 icmp
= (void *) outpack
+ ip
->ihl
* 4;
547 for (i
= 0; i
< array_size(idstore
); ++i
) {
548 if (unlikely(icmp
->un
.echo
.id
!= idstore
[i
]))
558 static bool shaper_is_set(struct shaper
*sh
)
560 return sh
->type
!= SHAPER_NONE
;
563 static void shaper_init(struct shaper
*sh
)
565 if (sh
->type
== SHAPER_NONE
|| sh
->type
== SHAPER_DELAY
)
568 memset(&sh
->delay
, 0, sizeof(struct timespec
));
569 bug_on(gettimeofday(&sh
->start
, NULL
));
573 static void shaper_set_delay(struct shaper
*sh
, time_t sec
, long int ns
)
576 sh
->type
= SHAPER_NONE
;
580 sh
->type
= SHAPER_DELAY
;
581 sh
->delay
.tv_sec
= sec
;
582 sh
->delay
.tv_nsec
= ns
;
585 static void shaper_set_rate(struct shaper
*sh
, unsigned long long rate
,
586 enum shaper_type type
)
588 memset(sh
, 0, sizeof(struct shaper
));
593 static void shaper_set_tstamp(struct shaper
*sh
, struct timespec
*ts
)
595 TIMESPEC_TO_TIMEVAL(&sh
->tstamp
, ts
);
598 static void shaper_delay(struct shaper
*sh
, struct packet
*pkt
)
600 if (sh
->type
== SHAPER_BYTES
|| sh
->type
== SHAPER_PKTS
) {
601 unsigned long pkt_len
= pkt
->len
;
603 sh
->sent
+= sh
->type
== SHAPER_BYTES
? pkt_len
: 1;
605 if (sh
->sent
>= sh
->rate
&& sh
->rate
> 0) {
606 struct timeval delay_us
;
607 struct timeval time_sent
;
608 struct timeval time_1s
= { .tv_sec
= 1 };
610 bug_on(gettimeofday(&sh
->end
, NULL
));
611 timersub(&sh
->end
, &sh
->start
, &time_sent
);
613 if (timercmp(&time_1s
, &time_sent
, > )) {
614 timersub(&time_1s
, &time_sent
, &delay_us
);
615 TIMEVAL_TO_TIMESPEC(&delay_us
, &sh
->delay
);
618 } else if (sh
->type
== SHAPER_TSTAMP
) {
619 struct timeval tstamp
;
620 struct timeval pkt_diff
;
623 bug_on(gettimeofday(&sh
->end
, NULL
));
624 TIMESPEC_TO_TIMEVAL(&tstamp
, &pkt
->tstamp
);
625 timersub(&sh
->end
, &sh
->start
, &diff
);
626 timersub(&tstamp
, &sh
->tstamp
, &pkt_diff
);
628 if (timercmp(&diff
, &pkt_diff
, <)) {
629 struct timeval delay
;
631 timersub(&pkt_diff
, &diff
, &delay
);
632 TIMEVAL_TO_TIMESPEC(&delay
, &sh
->delay
);
635 memcpy(&sh
->tstamp
, &tstamp
, sizeof(sh
->tstamp
));
638 if ((sh
->delay
.tv_sec
| sh
->delay
.tv_nsec
) > 0) {
639 nanosleep(&sh
->delay
, NULL
);
645 static inline void packet_apply_dyn_elements(int idx
)
647 if (packet_dyn_has_elems(&packet_dyn
[idx
])) {
649 apply_randomizer(idx
);
653 if (packet_dyn_has_fields(&packet_dyn
[idx
])) {
656 for (i
= 0; i
< packet_dyn
[idx
].flen
; i
++)
657 proto_field_dyn_apply(packet_dyn
[idx
].fields
[i
]);
659 proto_packet_update(idx
);
663 static void xmit_slowpath_or_die(struct ctx
*ctx
, unsigned int cpu
, unsigned long orig_num
)
665 int ret
, icmp_sock
= -1;
666 unsigned long num
= 1, i
= 0;
667 struct timeval start
, end
, diff
;
668 unsigned long long tx_bytes
= 0, tx_packets
= 0;
672 if (ctx
->num
== 0 && orig_num
> 0)
676 icmp_sock
= xmit_smoke_setup(ctx
);
678 drop_privileges(ctx
->enforce
, ctx
->uid
, ctx
->gid
);
680 bug_on(gettimeofday(&start
, NULL
));
682 if (shaper_is_set(&ctx
->sh
))
683 shaper_init(&ctx
->sh
);
685 while (likely(sigint
== 0 && num
> 0 && plen
> 0)) {
686 packet_apply_dyn_elements(i
);
688 ret
= dev_io_write(ctx
->dev_out
, packets
[i
].payload
, packets
[i
].len
);
689 if (unlikely(ret
< 0)) {
690 if (errno
== ENOBUFS
) {
695 panic("Sendto error: %s!\n", strerror(errno
));
698 tx_bytes
+= packets
[i
].len
;
701 if (ctx
->smoke_test
) {
702 ret
= xmit_smoke_probe(icmp_sock
, ctx
);
703 if (unlikely(ret
< 0)) {
704 printf("%sSmoke test alert:%s\n", colorize_start(bold
), colorize_end());
705 printf(" Remote host seems to be unresponsive to ICMP probes!\n");
706 printf(" Last instance was packet%lu, seed:%u, trafgen snippet:\n\n",
709 dump_trafgen_snippet(packets
[i
].payload
, packets
[i
].len
);
724 if (shaper_is_set(&ctx
->sh
))
725 shaper_delay(&ctx
->sh
, &packets
[i
]);
728 bug_on(gettimeofday(&end
, NULL
));
729 timersub(&end
, &start
, &diff
);
734 stats
[cpu
].tx_packets
= tx_packets
;
735 stats
[cpu
].tx_bytes
= tx_bytes
;
736 stats
[cpu
].tv_sec
= diff
.tv_sec
;
737 stats
[cpu
].tv_usec
= diff
.tv_usec
;
739 stats
[cpu
].state
|= CPU_STATS_STATE_RES
;
742 static void xmit_fastpath_or_die(struct ctx
*ctx
, unsigned int cpu
, unsigned long orig_num
)
744 int ifindex
= dev_io_ifindex_get(ctx
->dev_out
);
746 unsigned int it
= 0, retry
= 100;
747 unsigned long num
= 1, i
= 0;
748 size_t size
= ring_size(dev_io_name_get(ctx
->dev_out
), ctx
->reserve_size
);
750 struct frame_map
*hdr
;
751 struct timeval start
, end
, diff
;
752 unsigned long long tx_bytes
= 0, tx_packets
= 0;
753 int sock
= dev_io_fd_get(ctx
->dev_out
);
755 set_sock_prio(sock
, 512);
757 ring_tx_setup(&tx_ring
, sock
, size
, ifindex
, ctx
->jumbo_support
, ctx
->verbose
);
759 drop_privileges(ctx
->enforce
, ctx
->uid
, ctx
->gid
);
763 if (ctx
->num
== 0 && orig_num
> 0)
766 bug_on(gettimeofday(&start
, NULL
));
768 while (likely(sigint
== 0 && num
> 0 && plen
> 0)) {
769 if (!user_may_pull_from_tx(tx_ring
.frames
[it
].iov_base
)) {
770 int ret
= pull_and_flush_tx_ring(sock
);
771 if (unlikely(ret
< 0)) {
772 /* We could hit EBADF if the socket has been closed before
773 * the timer was triggered.
775 if (errno
!= EBADF
&& errno
!= ENOBUFS
)
776 panic("Flushing TX_RING failed: %s!\n", strerror(errno
));
782 hdr
= tx_ring
.frames
[it
].iov_base
;
783 out
= ((uint8_t *) hdr
) + TPACKET2_HDRLEN
- sizeof(struct sockaddr_ll
);
785 hdr
->tp_h
.tp_snaplen
= packets
[i
].len
;
786 hdr
->tp_h
.tp_len
= packets
[i
].len
;
788 packet_apply_dyn_elements(i
);
790 fmemcpy(out
, packets
[i
].payload
, packets
[i
].len
);
792 tx_bytes
+= packets
[i
].len
;
802 kernel_may_pull_from_tx(&hdr
->tp_h
);
805 if (it
>= tx_ring
.layout
.tp_frame_nr
)
812 bug_on(gettimeofday(&end
, NULL
));
813 timersub(&end
, &start
, &diff
);
815 while (pull_and_flush_tx_ring_wait(sock
) < 0 && errno
== ENOBUFS
&& retry
-- > 0)
817 destroy_tx_ring(sock
, &tx_ring
);
819 stats
[cpu
].tx_packets
= tx_packets
;
820 stats
[cpu
].tx_bytes
= tx_bytes
;
821 stats
[cpu
].tv_sec
= diff
.tv_sec
;
822 stats
[cpu
].tv_usec
= diff
.tv_usec
;
824 stats
[cpu
].state
|= CPU_STATS_STATE_RES
;
827 static inline void __set_state(unsigned int cpu
, sig_atomic_t s
)
829 stats
[cpu
].state
= s
;
832 static inline sig_atomic_t __get_state(unsigned int cpu
)
834 return stats
[cpu
].state
;
837 static unsigned long __wait_and_sum_others(struct ctx
*ctx
, unsigned int cpu
)
842 for (i
= 0, total
= plen
; i
< ctx
->cpus
; i
++) {
846 while ((__get_state(i
) &
847 (CPU_STATS_STATE_CFG
|
848 CPU_STATS_STATE_RES
)) == 0 &&
852 total
+= stats
[i
].cf_packets
;
858 static void __correct_global_delta(struct ctx
*ctx
, unsigned int cpu
, unsigned long orig
)
863 long long delta_correction
= 0;
865 for (i
= 0, total
= ctx
->num
; i
< ctx
->cpus
; i
++) {
869 while ((__get_state(i
) &
870 (CPU_STATS_STATE_CHK
|
871 CPU_STATS_STATE_RES
)) == 0 &&
875 total
+= stats
[i
].cd_packets
;
879 delta_correction
= -1 * ((long long) total
- orig
);
881 delta_correction
= +1 * ((long long) orig
- total
);
883 for (cpu_sel
= -1, i
= 0; i
< ctx
->cpus
; i
++) {
884 if (stats
[i
].cd_packets
> 0) {
885 if ((long long) stats
[i
].cd_packets
+
886 delta_correction
>= 0) {
893 if ((int) cpu
== cpu_sel
)
894 ctx
->num
+= delta_correction
;
897 static void __set_state_cf(unsigned int cpu
, unsigned long p
, unsigned long b
,
900 stats
[cpu
].cf_packets
= p
;
901 stats
[cpu
].cf_bytes
= b
;
902 stats
[cpu
].state
= s
;
905 static void __set_state_cd(unsigned int cpu
, unsigned long p
, sig_atomic_t s
)
907 stats
[cpu
].cd_packets
= p
;
908 stats
[cpu
].state
= s
;
911 static void xmit_packet_precheck(struct ctx
*ctx
, unsigned int cpu
)
913 unsigned long plen_total
, orig
= ctx
->num
;
914 size_t total_len
= 0;
917 bug_on(plen
!= dlen
);
919 for (i
= 0; i
< plen
; ++i
)
920 total_len
+= packets
[i
].len
;
922 __set_state_cf(cpu
, plen
, total_len
, CPU_STATS_STATE_CFG
);
923 plen_total
= __wait_and_sum_others(ctx
, cpu
);
926 ctx
->num
= (unsigned long) round((1.0 * plen
/ plen_total
) * orig
);
928 __set_state_cd(cpu
, ctx
->num
, CPU_STATS_STATE_CHK
|
929 CPU_STATS_STATE_CFG
);
930 __correct_global_delta(ctx
, cpu
, orig
);
934 __set_state(cpu
, CPU_STATS_STATE_RES
);
939 static void pcap_load_packets(struct dev_io
*dev
)
941 struct timespec tstamp
;
946 buf_len
= round_up(1024 * 1024, RUNTIME_PAGE_SIZE
);
947 buf
= xmalloc_aligned(buf_len
, CO_CACHE_LINE_SIZE
);
949 while ((pkt_len
= dev_io_read(dev
, buf
, buf_len
, &tstamp
)) > 0) {
954 pkt
= current_packet();
956 pkt
->payload
= xzmalloc(pkt_len
);
957 memcpy(pkt
->payload
, buf
, pkt_len
);
958 memcpy(&pkt
->tstamp
, &tstamp
, sizeof(tstamp
));
964 static void main_loop(struct ctx
*ctx
, char *confname
, bool slow
,
965 unsigned int cpu
, bool invoke_cpp
, char **cpp_argv
,
966 unsigned long orig_num
)
968 if (ctx
->dev_in
&& dev_io_is_pcap(ctx
->dev_in
)) {
969 pcap_load_packets(ctx
->dev_in
);
970 shaper_set_tstamp(&ctx
->sh
, &packets
[0].tstamp
);
974 compile_packets_str(ctx
->packet_str
, ctx
->verbose
, cpu
);
976 compile_packets(confname
, ctx
->verbose
, cpu
, invoke_cpp
, cpp_argv
);
978 preprocess_packets();
981 xmit_packet_precheck(ctx
, cpu
);
985 size_t total_len
= 0, total_pkts
= 0;
987 for (i
= 0; i
< ctx
->cpus
; ++i
) {
988 total_len
+= stats
[i
].cf_bytes
;
989 total_pkts
+= stats
[i
].cf_packets
;
992 printf("%6zu packets to schedule\n", total_pkts
);
993 printf("%6zu bytes in total\n", total_len
);
994 printf("Running! Hang up with ^C!\n\n");
998 if (dev_io_is_netdev(ctx
->dev_out
) && ctx
->qdisc_path
== false)
999 set_sock_qdisc_bypass(dev_io_fd_get(ctx
->dev_out
), ctx
->verbose
);
1002 xmit_slowpath_or_die(ctx
, cpu
, orig_num
);
1004 xmit_fastpath_or_die(ctx
, cpu
, orig_num
);
1009 static unsigned int generate_srand_seed(void)
1014 fd
= open("/dev/urandom", O_RDONLY
);
1018 read_or_die(fd
, &_seed
, sizeof(_seed
));
1024 static void on_panic_del_rfmon(void *arg
)
1026 leave_rfmon_mac80211(arg
);
1029 int main(int argc
, char **argv
)
1031 bool slow
= false, invoke_cpp
= false, reseed
= true, cpustats
= true;
1032 bool prio_high
= false, set_irq_aff
= true, set_sock_mem
= true;
1033 int c
, vals
[4] = {0}, irq
;
1036 char *confname
= NULL
, *ptr
;
1037 unsigned long cpus_tmp
, orig_num
= 0;
1038 unsigned long long tx_packets
, tx_bytes
;
1041 char **cpp_argv
= NULL
;
1042 size_t cpp_argc
= 0;
1043 unsigned long long rate
;
1044 enum shaper_type shape_type
;
1045 struct timespec delay
;
1047 fmemset(&ctx
, 0, sizeof(ctx
));
1048 ctx
.cpus
= get_number_cpus_online();
1051 ctx
.qdisc_path
= false;
1053 /* Keep an initial small default size to reduce cache-misses. */
1054 ctx
.reserve_size
= 512 * (1 << 10);
1056 while ((c
= getopt_long(argc
, argv
, short_options
, long_options
,
1075 cpp_argv
= argv_insert(cpp_argv
, &cpp_argc
, "-D");
1076 cpp_argv
= argv_insert(cpp_argv
, &cpp_argc
, optarg
);
1082 cpus_tmp
= strtoul(optarg
, NULL
, 0);
1083 if (cpus_tmp
> 0 && cpus_tmp
< ctx
.cpus
)
1084 ctx
.cpus
= cpus_tmp
;
1088 ctx
.device
= xstrdup(optarg
);
1094 set_sock_mem
= false;
1097 set_irq_aff
= false;
1100 ctx
.qdisc_path
= true;
1108 ctx
.smoke_test
= true;
1109 ctx
.rhost
= xstrdup(optarg
);
1115 ctx
.jumbo_support
= true;
1118 confname
= xstrdup(optarg
);
1119 if (strstr(confname
, ".pcap")) {
1120 ctx
.sh
.type
= SHAPER_TSTAMP
;
1121 ctx
.pcap_in
= confname
;
1125 if (!strncmp("-", confname
, strlen("-")))
1129 ctx
.uid
= strtoul(optarg
, NULL
, 0);
1133 ctx
.gid
= strtoul(optarg
, NULL
, 0);
1137 printf("Option -k/--kernel-pull is no longer used and "
1138 "will be removed in a future release!\n");
1141 seed
= strtoul(optarg
, NULL
, 0);
1145 orig_num
= strtoul(optarg
, NULL
, 0);
1149 gap
= strtoul(optarg
, &ptr
, 0);
1150 if (!gap
&& optarg
== ptr
)
1151 panic("Invalid gap param\n");
1153 if (!strncmp(ptr
, "ns", strlen("ns"))) {
1154 delay
.tv_sec
= gap
/ 1000000000;
1155 delay
.tv_nsec
= gap
% 1000000000;
1156 } else if (*ptr
== '\0' || !strncmp(ptr
, "us", strlen("us"))) {
1157 /* Default to microseconds for backwards
1158 * compatibility if no postfix is given.
1160 delay
.tv_sec
= gap
/ 1000000;
1161 delay
.tv_nsec
= (gap
% 1000000) * 1000;
1162 } else if (!strncmp(ptr
, "ms", strlen("ms"))) {
1163 delay
.tv_sec
= gap
/ 1000;
1164 delay
.tv_nsec
= (gap
% 1000) * 1000000;
1165 } else if (!strncmp(ptr
, "s", strlen("s"))) {
1169 panic("Syntax error in time param!\n");
1172 shaper_set_delay(&ctx
.sh
, delay
.tv_sec
, delay
.tv_nsec
);
1175 rate
= strtoul(optarg
, &ptr
, 0);
1176 if (!rate
&& optarg
== ptr
)
1177 panic("Invalid rate param\n");
1179 if (strncmp(ptr
, "pps", strlen("pps")) == 0) {
1180 shape_type
= SHAPER_PKTS
;
1181 } else if (strncmp(ptr
, "B", strlen("B")) == 0) {
1182 shape_type
= SHAPER_BYTES
;
1183 } else if (strncmp(ptr
, "kB", strlen("kB")) == 0) {
1184 shape_type
= SHAPER_BYTES
;
1186 } else if (strncmp(ptr
, "MB", strlen("MB")) == 0) {
1187 shape_type
= SHAPER_BYTES
;
1188 rate
*= 1000 * 1000;
1189 } else if (strncmp(ptr
, "GB", strlen("GB")) == 0) {
1190 shape_type
= SHAPER_BYTES
;
1191 rate
*= 1000 * 1000 * 1000;
1192 } else if (strncmp(ptr
, "kbit", strlen("kbit")) == 0) {
1193 shape_type
= SHAPER_BYTES
;
1195 } else if (strncmp(ptr
, "Mbit", strlen("Mbit")) == 0) {
1196 shape_type
= SHAPER_BYTES
;
1197 rate
*= 1000 * 1000 / 8;
1198 } else if (strncmp(ptr
, "Gbit", strlen("Gbit")) == 0) {
1199 shape_type
= SHAPER_BYTES
;
1200 rate
*= 1000 * 1000 * 1000 / 8;
1201 } else if (strncmp(ptr
, "KiB", strlen("KiB")) == 0) {
1202 shape_type
= SHAPER_BYTES
;
1204 } else if (strncmp(ptr
, "MiB", strlen("MiB")) == 0) {
1205 shape_type
= SHAPER_BYTES
;
1207 } else if (strncmp(ptr
, "GiB", strlen("GiB")) == 0) {
1208 shape_type
= SHAPER_BYTES
;
1211 shape_type
= SHAPER_NONE
;
1213 panic("Invalid unit type for rate\n");
1216 shaper_set_rate(&ctx
.sh
, rate
, shape_type
);
1219 ctx
.reserve_size
= strtoul(optarg
, &ptr
, 0);
1220 if (ctx
.reserve_size
== 0 && ptr
== optarg
)
1221 panic("Invalid ring size param\n");
1223 if (!strncmp(ptr
, "KiB", strlen("KiB")))
1224 ctx
.reserve_size
*= 1 << 10;
1225 else if (!strncmp(ptr
, "MiB", strlen("MiB")))
1226 ctx
.reserve_size
= 1 << 20;
1227 else if (!strncmp(ptr
, "GiB", strlen("GiB")))
1228 ctx
.reserve_size
*= 1 << 30;
1230 panic("Invalid ring size unit type\n");
1248 panic("Option -%c requires an argument!\n",
1251 if (isprint(optopt
))
1252 printf("Unknown option character `0x%X\'!\n", optopt
);
1260 if (argc
>= optind
) {
1262 ctx
.packet_str
= argv2str(optind
, argc
, argv
);
1265 if (argc
< min_opts
)
1267 if (ctx
.device
== NULL
)
1268 panic("No networking device given!\n");
1269 if (confname
== NULL
&& !ctx
.packet_str
)
1270 panic("No configuration file or packet string given!\n");
1272 register_signal(SIGINT
, signal_handler
);
1273 register_signal(SIGQUIT
, signal_handler
);
1274 register_signal(SIGTERM
, signal_handler
);
1275 register_signal(SIGHUP
, signal_handler
);
1279 set_sched_status(SCHED_FIFO
, sched_get_priority_max(SCHED_FIFO
));
1283 set_system_socket_memory(vals
, array_size(vals
));
1287 ctx
.device_trans
= xstrdup(ctx
.device
);
1290 enter_rfmon_mac80211(ctx
.device_trans
, &ctx
.device
);
1291 panic_handler_add(on_panic_del_rfmon
, ctx
.device
);
1296 ctx
.dev_in
= dev_io_open(ctx
.pcap_in
, DEV_IO_IN
);
1298 panic("Failed to open input device\n");
1301 ctx
.dev_out
= dev_io_open(ctx
.device
, DEV_IO_OUT
);
1303 panic("Failed to open output device\n");
1305 protos_init(ctx
.dev_out
);
1307 if (shaper_is_set(&ctx
.sh
) || (ctx
.dev_in
&& dev_io_is_pcap(ctx
.dev_in
))
1308 || dev_io_is_pcap(ctx
.dev_out
)) {
1310 prctl(PR_SET_TIMERSLACK
, 1UL);
1311 /* Fall back to single core to not mess up correct timing.
1312 * We are slow anyway!
1319 * If number of packets is smaller than number of CPUs use only as
1320 * many CPUs as there are packets. Otherwise we end up sending more
1321 * packets than intended or none at all.
1324 ctx
.cpus
= min_t(unsigned int, ctx
.num
, ctx
.cpus
);
1326 if (set_irq_aff
&& dev_io_is_netdev(ctx
.dev_out
)) {
1327 irq
= device_irq_number(ctx
.device
);
1328 device_set_irq_affinity_list(irq
, 0, ctx
.cpus
- 1);
1331 stats
= setup_shared_var(ctx
.cpus
);
1333 for (i
= 0; i
< ctx
.cpus
; i
++) {
1339 seed
= generate_srand_seed();
1343 main_loop(&ctx
, confname
, slow
, i
, invoke_cpp
,
1344 cpp_argv
, orig_num
);
1348 panic("Cannot fork processes!\n");
1352 for (i
= 0; i
< ctx
.cpus
; i
++) {
1356 if (WEXITSTATUS(status
) == EXIT_FAILURE
)
1361 leave_rfmon_mac80211(ctx
.device
);
1364 reset_system_socket_memory(vals
, array_size(vals
));
1366 for (i
= 0, tx_packets
= tx_bytes
= 0; i
< ctx
.cpus
; i
++) {
1367 while ((__get_state(i
) & CPU_STATS_STATE_RES
) == 0)
1370 tx_packets
+= stats
[i
].tx_packets
;
1371 tx_bytes
+= stats
[i
].tx_bytes
;
1376 printf("\r%12llu packets outgoing\n", tx_packets
);
1377 printf("\r%12llu bytes outgoing\n", tx_bytes
);
1378 for (i
= 0; cpustats
&& i
< ctx
.cpus
; i
++) {
1379 printf("\r%12lu sec, %lu usec on CPU%d (%llu packets)\n",
1380 stats
[i
].tv_sec
, stats
[i
].tv_usec
, i
,
1381 stats
[i
].tx_packets
);
1386 destroy_shared_var(stats
, ctx
.cpus
);
1387 if (dev_io_is_netdev(ctx
.dev_out
) && set_irq_aff
)
1388 device_restore_irq_affinity_list();
1390 dev_io_close(ctx
.dev_out
);
1392 dev_io_close(ctx
.dev_in
);
1394 argv_free(cpp_argv
);
1396 free(ctx
.device_trans
);
1399 free(ctx
.packet_str
);