1 /* SPDX-License-Identifier: GPL-2.0
2 * Copyright (c) 2017 Jesper Dangaard Brouer, Red Hat Inc.
3 */
4 static const char *__doc__ = " XDP RX-queue info extract example\n\n"
5 "Monitor how many packets per sec (pps) are received\n"
6 "per NIC RX queue index and which CPU processed the packet\n"
7 ;
8
9 #include <errno.h>
10 #include <signal.h>
11 #include <stdio.h>
12 #include <stdlib.h>
13 #include <stdbool.h>
14 #include <string.h>
15 #include <unistd.h>
16 #include <locale.h>
17 #include <sys/resource.h>
18 #include <getopt.h>
19 #include <net/if.h>
20 #include <time.h>
21
22 #include <arpa/inet.h>
23 #include <linux/if_link.h>
24
25 #include "bpf/bpf.h"
26 #include "bpf/libbpf.h"
27 #include "bpf_util.h"
28
29 static int ifindex = -1;
30 static char ifname_buf[IF_NAMESIZE];
31 static char *ifname;
32
33 static __u32 xdp_flags;
34
35 static struct bpf_map *stats_global_map;
36 static struct bpf_map *rx_queue_index_map;
37
38 /* Exit return codes */
39 #define EXIT_OK 0
40 #define EXIT_FAIL 1
41 #define EXIT_FAIL_OPTION 2
42 #define EXIT_FAIL_XDP 3
43 #define EXIT_FAIL_BPF 4
44 #define EXIT_FAIL_MEM 5
45
46 static const struct option long_options[] = {
47 {"help", no_argument, NULL, 'h' },
48 {"dev", required_argument, NULL, 'd' },
49 {"skb-mode", no_argument, NULL, 'S' },
50 {"sec", required_argument, NULL, 's' },
51 {"no-separators", no_argument, NULL, 'z' },
52 {"action", required_argument, NULL, 'a' },
53 {"readmem", no_argument, NULL, 'r' },
54 {"swapmac", no_argument, NULL, 'm' },
55 {0, 0, NULL, 0 }
56 };
57
int_exit(int sig)58 static void int_exit(int sig)
59 {
60 fprintf(stderr,
61 "Interrupted: Removing XDP program on ifindex:%d device:%s\n",
62 ifindex, ifname);
63 if (ifindex > -1)
64 bpf_set_link_xdp_fd(ifindex, -1, xdp_flags);
65 exit(EXIT_OK);
66 }
67
68 struct config {
69 __u32 action;
70 int ifindex;
71 __u32 options;
72 };
73 enum cfg_options_flags {
74 NO_TOUCH = 0x0U,
75 READ_MEM = 0x1U,
76 SWAP_MAC = 0x2U,
77 };
78 #define XDP_ACTION_MAX (XDP_TX + 1)
79 #define XDP_ACTION_MAX_STRLEN 11
80 static const char *xdp_action_names[XDP_ACTION_MAX] = {
81 [XDP_ABORTED] = "XDP_ABORTED",
82 [XDP_DROP] = "XDP_DROP",
83 [XDP_PASS] = "XDP_PASS",
84 [XDP_TX] = "XDP_TX",
85 };
86
action2str(int action)87 static const char *action2str(int action)
88 {
89 if (action < XDP_ACTION_MAX)
90 return xdp_action_names[action];
91 return NULL;
92 }
93
parse_xdp_action(char * action_str)94 static int parse_xdp_action(char *action_str)
95 {
96 size_t maxlen;
97 __u64 action = -1;
98 int i;
99
100 for (i = 0; i < XDP_ACTION_MAX; i++) {
101 maxlen = XDP_ACTION_MAX_STRLEN;
102 if (strncmp(xdp_action_names[i], action_str, maxlen) == 0) {
103 action = i;
104 break;
105 }
106 }
107 return action;
108 }
109
list_xdp_actions(void)110 static void list_xdp_actions(void)
111 {
112 int i;
113
114 printf("Available XDP --action <options>\n");
115 for (i = 0; i < XDP_ACTION_MAX; i++)
116 printf("\t%s\n", xdp_action_names[i]);
117 printf("\n");
118 }
119
options2str(enum cfg_options_flags flag)120 static char* options2str(enum cfg_options_flags flag)
121 {
122 if (flag == NO_TOUCH)
123 return "no_touch";
124 if (flag & SWAP_MAC)
125 return "swapmac";
126 if (flag & READ_MEM)
127 return "read";
128 fprintf(stderr, "ERR: Unknown config option flags");
129 exit(EXIT_FAIL);
130 }
131
usage(char * argv[])132 static void usage(char *argv[])
133 {
134 int i;
135
136 printf("\nDOCUMENTATION:\n%s\n", __doc__);
137 printf(" Usage: %s (options-see-below)\n", argv[0]);
138 printf(" Listing options:\n");
139 for (i = 0; long_options[i].name != 0; i++) {
140 printf(" --%-12s", long_options[i].name);
141 if (long_options[i].flag != NULL)
142 printf(" flag (internal value:%d)",
143 *long_options[i].flag);
144 else
145 printf(" short-option: -%c",
146 long_options[i].val);
147 printf("\n");
148 }
149 printf("\n");
150 list_xdp_actions();
151 }
152
153 #define NANOSEC_PER_SEC 1000000000 /* 10^9 */
gettime(void)154 static __u64 gettime(void)
155 {
156 struct timespec t;
157 int res;
158
159 res = clock_gettime(CLOCK_MONOTONIC, &t);
160 if (res < 0) {
161 fprintf(stderr, "Error with gettimeofday! (%i)\n", res);
162 exit(EXIT_FAIL);
163 }
164 return (__u64) t.tv_sec * NANOSEC_PER_SEC + t.tv_nsec;
165 }
166
167 /* Common stats data record shared with _kern.c */
168 struct datarec {
169 __u64 processed;
170 __u64 issue;
171 };
172 struct record {
173 __u64 timestamp;
174 struct datarec total;
175 struct datarec *cpu;
176 };
177 struct stats_record {
178 struct record stats;
179 struct record *rxq;
180 };
181
alloc_record_per_cpu(void)182 static struct datarec *alloc_record_per_cpu(void)
183 {
184 unsigned int nr_cpus = bpf_num_possible_cpus();
185 struct datarec *array;
186 size_t size;
187
188 size = sizeof(struct datarec) * nr_cpus;
189 array = malloc(size);
190 memset(array, 0, size);
191 if (!array) {
192 fprintf(stderr, "Mem alloc error (nr_cpus:%u)\n", nr_cpus);
193 exit(EXIT_FAIL_MEM);
194 }
195 return array;
196 }
197
alloc_record_per_rxq(void)198 static struct record *alloc_record_per_rxq(void)
199 {
200 unsigned int nr_rxqs = bpf_map__def(rx_queue_index_map)->max_entries;
201 struct record *array;
202 size_t size;
203
204 size = sizeof(struct record) * nr_rxqs;
205 array = malloc(size);
206 memset(array, 0, size);
207 if (!array) {
208 fprintf(stderr, "Mem alloc error (nr_rxqs:%u)\n", nr_rxqs);
209 exit(EXIT_FAIL_MEM);
210 }
211 return array;
212 }
213
alloc_stats_record(void)214 static struct stats_record *alloc_stats_record(void)
215 {
216 unsigned int nr_rxqs = bpf_map__def(rx_queue_index_map)->max_entries;
217 struct stats_record *rec;
218 int i;
219
220 rec = malloc(sizeof(*rec));
221 memset(rec, 0, sizeof(*rec));
222 if (!rec) {
223 fprintf(stderr, "Mem alloc error\n");
224 exit(EXIT_FAIL_MEM);
225 }
226 rec->rxq = alloc_record_per_rxq();
227 for (i = 0; i < nr_rxqs; i++)
228 rec->rxq[i].cpu = alloc_record_per_cpu();
229
230 rec->stats.cpu = alloc_record_per_cpu();
231 return rec;
232 }
233
free_stats_record(struct stats_record * r)234 static void free_stats_record(struct stats_record *r)
235 {
236 unsigned int nr_rxqs = bpf_map__def(rx_queue_index_map)->max_entries;
237 int i;
238
239 for (i = 0; i < nr_rxqs; i++)
240 free(r->rxq[i].cpu);
241
242 free(r->rxq);
243 free(r->stats.cpu);
244 free(r);
245 }
246
map_collect_percpu(int fd,__u32 key,struct record * rec)247 static bool map_collect_percpu(int fd, __u32 key, struct record *rec)
248 {
249 /* For percpu maps, userspace gets a value per possible CPU */
250 unsigned int nr_cpus = bpf_num_possible_cpus();
251 struct datarec values[nr_cpus];
252 __u64 sum_processed = 0;
253 __u64 sum_issue = 0;
254 int i;
255
256 if ((bpf_map_lookup_elem(fd, &key, values)) != 0) {
257 fprintf(stderr,
258 "ERR: bpf_map_lookup_elem failed key:0x%X\n", key);
259 return false;
260 }
261 /* Get time as close as possible to reading map contents */
262 rec->timestamp = gettime();
263
264 /* Record and sum values from each CPU */
265 for (i = 0; i < nr_cpus; i++) {
266 rec->cpu[i].processed = values[i].processed;
267 sum_processed += values[i].processed;
268 rec->cpu[i].issue = values[i].issue;
269 sum_issue += values[i].issue;
270 }
271 rec->total.processed = sum_processed;
272 rec->total.issue = sum_issue;
273 return true;
274 }
275
stats_collect(struct stats_record * rec)276 static void stats_collect(struct stats_record *rec)
277 {
278 int fd, i, max_rxqs;
279
280 fd = bpf_map__fd(stats_global_map);
281 map_collect_percpu(fd, 0, &rec->stats);
282
283 fd = bpf_map__fd(rx_queue_index_map);
284 max_rxqs = bpf_map__def(rx_queue_index_map)->max_entries;
285 for (i = 0; i < max_rxqs; i++)
286 map_collect_percpu(fd, i, &rec->rxq[i]);
287 }
288
calc_period(struct record * r,struct record * p)289 static double calc_period(struct record *r, struct record *p)
290 {
291 double period_ = 0;
292 __u64 period = 0;
293
294 period = r->timestamp - p->timestamp;
295 if (period > 0)
296 period_ = ((double) period / NANOSEC_PER_SEC);
297
298 return period_;
299 }
300
calc_pps(struct datarec * r,struct datarec * p,double period_)301 static __u64 calc_pps(struct datarec *r, struct datarec *p, double period_)
302 {
303 __u64 packets = 0;
304 __u64 pps = 0;
305
306 if (period_ > 0) {
307 packets = r->processed - p->processed;
308 pps = packets / period_;
309 }
310 return pps;
311 }
312
calc_errs_pps(struct datarec * r,struct datarec * p,double period_)313 static __u64 calc_errs_pps(struct datarec *r,
314 struct datarec *p, double period_)
315 {
316 __u64 packets = 0;
317 __u64 pps = 0;
318
319 if (period_ > 0) {
320 packets = r->issue - p->issue;
321 pps = packets / period_;
322 }
323 return pps;
324 }
325
stats_print(struct stats_record * stats_rec,struct stats_record * stats_prev,int action,__u32 cfg_opt)326 static void stats_print(struct stats_record *stats_rec,
327 struct stats_record *stats_prev,
328 int action, __u32 cfg_opt)
329 {
330 unsigned int nr_rxqs = bpf_map__def(rx_queue_index_map)->max_entries;
331 unsigned int nr_cpus = bpf_num_possible_cpus();
332 double pps = 0, err = 0;
333 struct record *rec, *prev;
334 double t;
335 int rxq;
336 int i;
337
338 /* Header */
339 printf("\nRunning XDP on dev:%s (ifindex:%d) action:%s options:%s\n",
340 ifname, ifindex, action2str(action), options2str(cfg_opt));
341
342 /* stats_global_map */
343 {
344 char *fmt_rx = "%-15s %-7d %'-11.0f %'-10.0f %s\n";
345 char *fm2_rx = "%-15s %-7s %'-11.0f\n";
346 char *errstr = "";
347
348 printf("%-15s %-7s %-11s %-11s\n",
349 "XDP stats", "CPU", "pps", "issue-pps");
350
351 rec = &stats_rec->stats;
352 prev = &stats_prev->stats;
353 t = calc_period(rec, prev);
354 for (i = 0; i < nr_cpus; i++) {
355 struct datarec *r = &rec->cpu[i];
356 struct datarec *p = &prev->cpu[i];
357
358 pps = calc_pps (r, p, t);
359 err = calc_errs_pps(r, p, t);
360 if (err > 0)
361 errstr = "invalid-ifindex";
362 if (pps > 0)
363 printf(fmt_rx, "XDP-RX CPU",
364 i, pps, err, errstr);
365 }
366 pps = calc_pps (&rec->total, &prev->total, t);
367 err = calc_errs_pps(&rec->total, &prev->total, t);
368 printf(fm2_rx, "XDP-RX CPU", "total", pps, err);
369 }
370
371 /* rx_queue_index_map */
372 printf("\n%-15s %-7s %-11s %-11s\n",
373 "RXQ stats", "RXQ:CPU", "pps", "issue-pps");
374
375 for (rxq = 0; rxq < nr_rxqs; rxq++) {
376 char *fmt_rx = "%-15s %3d:%-3d %'-11.0f %'-10.0f %s\n";
377 char *fm2_rx = "%-15s %3d:%-3s %'-11.0f\n";
378 char *errstr = "";
379 int rxq_ = rxq;
380
381 /* Last RXQ in map catch overflows */
382 if (rxq_ == nr_rxqs - 1)
383 rxq_ = -1;
384
385 rec = &stats_rec->rxq[rxq];
386 prev = &stats_prev->rxq[rxq];
387 t = calc_period(rec, prev);
388 for (i = 0; i < nr_cpus; i++) {
389 struct datarec *r = &rec->cpu[i];
390 struct datarec *p = &prev->cpu[i];
391
392 pps = calc_pps (r, p, t);
393 err = calc_errs_pps(r, p, t);
394 if (err > 0) {
395 if (rxq_ == -1)
396 errstr = "map-overflow-RXQ";
397 else
398 errstr = "err";
399 }
400 if (pps > 0)
401 printf(fmt_rx, "rx_queue_index",
402 rxq_, i, pps, err, errstr);
403 }
404 pps = calc_pps (&rec->total, &prev->total, t);
405 err = calc_errs_pps(&rec->total, &prev->total, t);
406 if (pps || err)
407 printf(fm2_rx, "rx_queue_index", rxq_, "sum", pps, err);
408 }
409 }
410
411
412 /* Pointer swap trick */
swap(struct stats_record ** a,struct stats_record ** b)413 static inline void swap(struct stats_record **a, struct stats_record **b)
414 {
415 struct stats_record *tmp;
416
417 tmp = *a;
418 *a = *b;
419 *b = tmp;
420 }
421
stats_poll(int interval,int action,__u32 cfg_opt)422 static void stats_poll(int interval, int action, __u32 cfg_opt)
423 {
424 struct stats_record *record, *prev;
425
426 record = alloc_stats_record();
427 prev = alloc_stats_record();
428 stats_collect(record);
429
430 while (1) {
431 swap(&prev, &record);
432 stats_collect(record);
433 stats_print(record, prev, action, cfg_opt);
434 sleep(interval);
435 }
436
437 free_stats_record(record);
438 free_stats_record(prev);
439 }
440
441
main(int argc,char ** argv)442 int main(int argc, char **argv)
443 {
444 __u32 cfg_options= NO_TOUCH ; /* Default: Don't touch packet memory */
445 struct rlimit r = {10 * 1024 * 1024, RLIM_INFINITY};
446 struct bpf_prog_load_attr prog_load_attr = {
447 .prog_type = BPF_PROG_TYPE_XDP,
448 };
449 int prog_fd, map_fd, opt, err;
450 bool use_separators = true;
451 struct config cfg = { 0 };
452 struct bpf_object *obj;
453 struct bpf_map *map;
454 char filename[256];
455 int longindex = 0;
456 int interval = 2;
457 __u32 key = 0;
458
459
460 char action_str_buf[XDP_ACTION_MAX_STRLEN + 1 /* for \0 */] = { 0 };
461 int action = XDP_PASS; /* Default action */
462 char *action_str = NULL;
463
464 snprintf(filename, sizeof(filename), "%s_kern.o", argv[0]);
465 prog_load_attr.file = filename;
466
467 if (setrlimit(RLIMIT_MEMLOCK, &r)) {
468 perror("setrlimit(RLIMIT_MEMLOCK)");
469 return 1;
470 }
471
472 if (bpf_prog_load_xattr(&prog_load_attr, &obj, &prog_fd))
473 return EXIT_FAIL;
474
475 map = bpf_map__next(NULL, obj);
476 stats_global_map = bpf_map__next(map, obj);
477 rx_queue_index_map = bpf_map__next(stats_global_map, obj);
478 if (!map || !stats_global_map || !rx_queue_index_map) {
479 printf("finding a map in obj file failed\n");
480 return EXIT_FAIL;
481 }
482 map_fd = bpf_map__fd(map);
483
484 if (!prog_fd) {
485 fprintf(stderr, "ERR: load_bpf_file: %s\n", strerror(errno));
486 return EXIT_FAIL;
487 }
488
489 /* Parse commands line args */
490 while ((opt = getopt_long(argc, argv, "hSd:",
491 long_options, &longindex)) != -1) {
492 switch (opt) {
493 case 'd':
494 if (strlen(optarg) >= IF_NAMESIZE) {
495 fprintf(stderr, "ERR: --dev name too long\n");
496 goto error;
497 }
498 ifname = (char *)&ifname_buf;
499 strncpy(ifname, optarg, IF_NAMESIZE);
500 ifindex = if_nametoindex(ifname);
501 if (ifindex == 0) {
502 fprintf(stderr,
503 "ERR: --dev name unknown err(%d):%s\n",
504 errno, strerror(errno));
505 goto error;
506 }
507 break;
508 case 's':
509 interval = atoi(optarg);
510 break;
511 case 'S':
512 xdp_flags |= XDP_FLAGS_SKB_MODE;
513 break;
514 case 'z':
515 use_separators = false;
516 break;
517 case 'a':
518 action_str = (char *)&action_str_buf;
519 strncpy(action_str, optarg, XDP_ACTION_MAX_STRLEN);
520 break;
521 case 'r':
522 cfg_options |= READ_MEM;
523 break;
524 case 'm':
525 cfg_options |= SWAP_MAC;
526 break;
527 case 'h':
528 error:
529 default:
530 usage(argv);
531 return EXIT_FAIL_OPTION;
532 }
533 }
534 /* Required option */
535 if (ifindex == -1) {
536 fprintf(stderr, "ERR: required option --dev missing\n");
537 usage(argv);
538 return EXIT_FAIL_OPTION;
539 }
540 cfg.ifindex = ifindex;
541
542 /* Parse action string */
543 if (action_str) {
544 action = parse_xdp_action(action_str);
545 if (action < 0) {
546 fprintf(stderr, "ERR: Invalid XDP --action: %s\n",
547 action_str);
548 list_xdp_actions();
549 return EXIT_FAIL_OPTION;
550 }
551 }
552 cfg.action = action;
553
554 /* XDP_TX requires changing MAC-addrs, else HW may drop */
555 if (action == XDP_TX)
556 cfg_options |= SWAP_MAC;
557 cfg.options = cfg_options;
558
559 /* Trick to pretty printf with thousands separators use %' */
560 if (use_separators)
561 setlocale(LC_NUMERIC, "en_US");
562
563 /* User-side setup ifindex in config_map */
564 err = bpf_map_update_elem(map_fd, &key, &cfg, 0);
565 if (err) {
566 fprintf(stderr, "Store config failed (err:%d)\n", err);
567 exit(EXIT_FAIL_BPF);
568 }
569
570 /* Remove XDP program when program is interrupted or killed */
571 signal(SIGINT, int_exit);
572 signal(SIGTERM, int_exit);
573
574 if (bpf_set_link_xdp_fd(ifindex, prog_fd, xdp_flags) < 0) {
575 fprintf(stderr, "link set xdp fd failed\n");
576 return EXIT_FAIL_XDP;
577 }
578
579 stats_poll(interval, action, cfg_options);
580 return EXIT_OK;
581 }
582
