1 /* SPDX-License-Identifier: GPL-2.0
2 * Copyright(c) 2017 Jesper Dangaard Brouer, Red Hat, Inc.
4 static const char *__doc__=
5 "XDP monitor tool, based on tracepoints\n"
8 static const char *__doc_err_only__=
9 " NOTICE: Only tracking XDP redirect errors\n"
10 " Enable TX success stats via '--stats'\n"
11 " (which comes with a per packet processing overhead)\n"
24 #include <sys/resource.h>
31 #include <bpf/libbpf.h>
42 static const char *const map_type_strings[] = {
43 [REDIRECT_ERR_CNT] = "redirect_err_cnt",
44 [EXCEPTION_CNT] = "exception_cnt",
45 [CPUMAP_ENQUEUE_CNT] = "cpumap_enqueue_cnt",
46 [CPUMAP_KTHREAD_CNT] = "cpumap_kthread_cnt",
47 [DEVMAP_XMIT_CNT] = "devmap_xmit_cnt",
55 static int verbose = 1;
56 static bool debug = false;
57 struct bpf_map *map_data[NUM_MAP] = {};
58 struct bpf_link *tp_links[NUM_TP] = {};
59 struct bpf_object *obj;
61 static const struct option long_options[] = {
62 {"help", no_argument, NULL, 'h' },
63 {"debug", no_argument, NULL, 'D' },
64 {"stats", no_argument, NULL, 'S' },
65 {"sec", required_argument, NULL, 's' },
69 static void int_exit(int sig)
71 /* Detach tracepoints */
73 bpf_link__destroy(tp_links[--tp_cnt]);
75 bpf_object__close(obj);
79 /* C standard specifies two constants, EXIT_SUCCESS(0) and EXIT_FAILURE(1) */
80 #define EXIT_FAIL_MEM 5
82 static void usage(char *argv[])
85 printf("\nDOCUMENTATION:\n%s\n", __doc__);
87 printf(" Usage: %s (options-see-below)\n",
89 printf(" Listing options:\n");
90 for (i = 0; long_options[i].name != 0; i++) {
91 printf(" --%-15s", long_options[i].name);
92 if (long_options[i].flag != NULL)
93 printf(" flag (internal value:%d)",
94 *long_options[i].flag);
96 printf("short-option: -%c",
103 #define NANOSEC_PER_SEC 1000000000 /* 10^9 */
104 static __u64 gettime(void)
109 res = clock_gettime(CLOCK_MONOTONIC, &t);
111 fprintf(stderr, "Error with gettimeofday! (%i)\n", res);
114 return (__u64) t.tv_sec * NANOSEC_PER_SEC + t.tv_nsec;
121 #define REDIR_RES_MAX 2
122 static const char *redir_names[REDIR_RES_MAX] = {
123 [REDIR_SUCCESS] = "Success",
124 [REDIR_ERROR] = "Error",
126 static const char *err2str(int err)
128 if (err < REDIR_RES_MAX)
129 return redir_names[err];
132 /* enum xdp_action */
133 #define XDP_UNKNOWN XDP_REDIRECT + 1
134 #define XDP_ACTION_MAX (XDP_UNKNOWN + 1)
135 static const char *xdp_action_names[XDP_ACTION_MAX] = {
136 [XDP_ABORTED] = "XDP_ABORTED",
137 [XDP_DROP] = "XDP_DROP",
138 [XDP_PASS] = "XDP_PASS",
140 [XDP_REDIRECT] = "XDP_REDIRECT",
141 [XDP_UNKNOWN] = "XDP_UNKNOWN",
143 static const char *action2str(int action)
145 if (action < XDP_ACTION_MAX)
146 return xdp_action_names[action];
150 /* Common stats data record shared with _kern.c */
159 /* Userspace structs for collection of stats from maps */
162 struct datarec total;
169 /* record for _kern side __u64 values */
175 struct stats_record {
176 struct record_u64 xdp_redirect[REDIR_RES_MAX];
177 struct record_u64 xdp_exception[XDP_ACTION_MAX];
178 struct record xdp_cpumap_kthread;
179 struct record xdp_cpumap_enqueue[MAX_CPUS];
180 struct record xdp_devmap_xmit;
183 static bool map_collect_record(int fd, __u32 key, struct record *rec)
185 /* For percpu maps, userspace gets a value per possible CPU */
186 unsigned int nr_cpus = bpf_num_possible_cpus();
187 struct datarec values[nr_cpus];
188 __u64 sum_processed = 0;
189 __u64 sum_dropped = 0;
194 if ((bpf_map_lookup_elem(fd, &key, values)) != 0) {
196 "ERR: bpf_map_lookup_elem failed key:0x%X\n", key);
199 /* Get time as close as possible to reading map contents */
200 rec->timestamp = gettime();
202 /* Record and sum values from each CPU */
203 for (i = 0; i < nr_cpus; i++) {
204 rec->cpu[i].processed = values[i].processed;
205 sum_processed += values[i].processed;
206 rec->cpu[i].dropped = values[i].dropped;
207 sum_dropped += values[i].dropped;
208 rec->cpu[i].info = values[i].info;
209 sum_info += values[i].info;
210 rec->cpu[i].err = values[i].err;
211 sum_err += values[i].err;
213 rec->total.processed = sum_processed;
214 rec->total.dropped = sum_dropped;
215 rec->total.info = sum_info;
216 rec->total.err = sum_err;
220 static bool map_collect_record_u64(int fd, __u32 key, struct record_u64 *rec)
222 /* For percpu maps, userspace gets a value per possible CPU */
223 unsigned int nr_cpus = bpf_num_possible_cpus();
224 struct u64rec values[nr_cpus];
228 if ((bpf_map_lookup_elem(fd, &key, values)) != 0) {
230 "ERR: bpf_map_lookup_elem failed key:0x%X\n", key);
233 /* Get time as close as possible to reading map contents */
234 rec->timestamp = gettime();
236 /* Record and sum values from each CPU */
237 for (i = 0; i < nr_cpus; i++) {
238 rec->cpu[i].processed = values[i].processed;
239 sum_total += values[i].processed;
241 rec->total.processed = sum_total;
245 static double calc_period(struct record *r, struct record *p)
250 period = r->timestamp - p->timestamp;
252 period_ = ((double) period / NANOSEC_PER_SEC);
257 static double calc_period_u64(struct record_u64 *r, struct record_u64 *p)
262 period = r->timestamp - p->timestamp;
264 period_ = ((double) period / NANOSEC_PER_SEC);
269 static double calc_pps(struct datarec *r, struct datarec *p, double period)
275 packets = r->processed - p->processed;
276 pps = packets / period;
281 static double calc_pps_u64(struct u64rec *r, struct u64rec *p, double period)
287 packets = r->processed - p->processed;
288 pps = packets / period;
293 static double calc_drop(struct datarec *r, struct datarec *p, double period)
299 packets = r->dropped - p->dropped;
300 pps = packets / period;
305 static double calc_info(struct datarec *r, struct datarec *p, double period)
311 packets = r->info - p->info;
312 pps = packets / period;
317 static double calc_err(struct datarec *r, struct datarec *p, double period)
323 packets = r->err - p->err;
324 pps = packets / period;
329 static void stats_print(struct stats_record *stats_rec,
330 struct stats_record *stats_prev,
333 unsigned int nr_cpus = bpf_num_possible_cpus();
334 int rec_i = 0, i, to_cpu;
335 double t = 0, pps = 0;
338 printf("%-15s %-7s %-12s %-12s %-9s\n",
339 "XDP-event", "CPU:to", "pps", "drop-pps", "extra-info");
341 /* tracepoint: xdp:xdp_redirect_* */
345 for (; rec_i < REDIR_RES_MAX; rec_i++) {
346 struct record_u64 *rec, *prev;
347 char *fmt1 = "%-15s %-7d %'-12.0f %'-12.0f %s\n";
348 char *fmt2 = "%-15s %-7s %'-12.0f %'-12.0f %s\n";
350 rec = &stats_rec->xdp_redirect[rec_i];
351 prev = &stats_prev->xdp_redirect[rec_i];
352 t = calc_period_u64(rec, prev);
354 for (i = 0; i < nr_cpus; i++) {
355 struct u64rec *r = &rec->cpu[i];
356 struct u64rec *p = &prev->cpu[i];
358 pps = calc_pps_u64(r, p, t);
360 printf(fmt1, "XDP_REDIRECT", i,
361 rec_i ? 0.0: pps, rec_i ? pps : 0.0,
364 pps = calc_pps_u64(&rec->total, &prev->total, t);
365 printf(fmt2, "XDP_REDIRECT", "total",
366 rec_i ? 0.0: pps, rec_i ? pps : 0.0, err2str(rec_i));
369 /* tracepoint: xdp:xdp_exception */
370 for (rec_i = 0; rec_i < XDP_ACTION_MAX; rec_i++) {
371 struct record_u64 *rec, *prev;
372 char *fmt1 = "%-15s %-7d %'-12.0f %'-12.0f %s\n";
373 char *fmt2 = "%-15s %-7s %'-12.0f %'-12.0f %s\n";
375 rec = &stats_rec->xdp_exception[rec_i];
376 prev = &stats_prev->xdp_exception[rec_i];
377 t = calc_period_u64(rec, prev);
379 for (i = 0; i < nr_cpus; i++) {
380 struct u64rec *r = &rec->cpu[i];
381 struct u64rec *p = &prev->cpu[i];
383 pps = calc_pps_u64(r, p, t);
385 printf(fmt1, "Exception", i,
386 0.0, pps, action2str(rec_i));
388 pps = calc_pps_u64(&rec->total, &prev->total, t);
390 printf(fmt2, "Exception", "total",
391 0.0, pps, action2str(rec_i));
394 /* cpumap enqueue stats */
395 for (to_cpu = 0; to_cpu < MAX_CPUS; to_cpu++) {
396 char *fmt1 = "%-15s %3d:%-3d %'-12.0f %'-12.0f %'-10.2f %s\n";
397 char *fmt2 = "%-15s %3s:%-3d %'-12.0f %'-12.0f %'-10.2f %s\n";
398 struct record *rec, *prev;
402 rec = &stats_rec->xdp_cpumap_enqueue[to_cpu];
403 prev = &stats_prev->xdp_cpumap_enqueue[to_cpu];
404 t = calc_period(rec, prev);
405 for (i = 0; i < nr_cpus; i++) {
406 struct datarec *r = &rec->cpu[i];
407 struct datarec *p = &prev->cpu[i];
409 pps = calc_pps(r, p, t);
410 drop = calc_drop(r, p, t);
411 info = calc_info(r, p, t);
413 info_str = "bulk-average";
414 info = pps / info; /* calc average bulk size */
417 printf(fmt1, "cpumap-enqueue",
418 i, to_cpu, pps, drop, info, info_str);
420 pps = calc_pps(&rec->total, &prev->total, t);
422 drop = calc_drop(&rec->total, &prev->total, t);
423 info = calc_info(&rec->total, &prev->total, t);
425 info_str = "bulk-average";
426 info = pps / info; /* calc average bulk size */
428 printf(fmt2, "cpumap-enqueue",
429 "sum", to_cpu, pps, drop, info, info_str);
433 /* cpumap kthread stats */
435 char *fmt1 = "%-15s %-7d %'-12.0f %'-12.0f %'-10.0f %s\n";
436 char *fmt2 = "%-15s %-7s %'-12.0f %'-12.0f %'-10.0f %s\n";
437 struct record *rec, *prev;
441 rec = &stats_rec->xdp_cpumap_kthread;
442 prev = &stats_prev->xdp_cpumap_kthread;
443 t = calc_period(rec, prev);
444 for (i = 0; i < nr_cpus; i++) {
445 struct datarec *r = &rec->cpu[i];
446 struct datarec *p = &prev->cpu[i];
448 pps = calc_pps(r, p, t);
449 drop = calc_drop(r, p, t);
450 info = calc_info(r, p, t);
453 if (pps > 0 || drop > 0)
454 printf(fmt1, "cpumap-kthread",
455 i, pps, drop, info, i_str);
457 pps = calc_pps(&rec->total, &prev->total, t);
458 drop = calc_drop(&rec->total, &prev->total, t);
459 info = calc_info(&rec->total, &prev->total, t);
462 printf(fmt2, "cpumap-kthread", "total", pps, drop, info, i_str);
465 /* devmap ndo_xdp_xmit stats */
467 char *fmt1 = "%-15s %-7d %'-12.0f %'-12.0f %'-10.2f %s %s\n";
468 char *fmt2 = "%-15s %-7s %'-12.0f %'-12.0f %'-10.2f %s %s\n";
469 struct record *rec, *prev;
470 double drop, info, err;
474 rec = &stats_rec->xdp_devmap_xmit;
475 prev = &stats_prev->xdp_devmap_xmit;
476 t = calc_period(rec, prev);
477 for (i = 0; i < nr_cpus; i++) {
478 struct datarec *r = &rec->cpu[i];
479 struct datarec *p = &prev->cpu[i];
481 pps = calc_pps(r, p, t);
482 drop = calc_drop(r, p, t);
483 info = calc_info(r, p, t);
484 err = calc_err(r, p, t);
486 i_str = "bulk-average";
487 info = (pps+drop) / info; /* calc avg bulk */
491 if (pps > 0 || drop > 0)
492 printf(fmt1, "devmap-xmit",
493 i, pps, drop, info, i_str, err_str);
495 pps = calc_pps(&rec->total, &prev->total, t);
496 drop = calc_drop(&rec->total, &prev->total, t);
497 info = calc_info(&rec->total, &prev->total, t);
498 err = calc_err(&rec->total, &prev->total, t);
500 i_str = "bulk-average";
501 info = (pps+drop) / info; /* calc avg bulk */
505 printf(fmt2, "devmap-xmit", "total", pps, drop,
506 info, i_str, err_str);
512 static bool stats_collect(struct stats_record *rec)
517 /* TODO: Detect if someone unloaded the perf event_fd's, as
518 * this can happen by someone running perf-record -e
521 fd = bpf_map__fd(map_data[REDIRECT_ERR_CNT]);
522 for (i = 0; i < REDIR_RES_MAX; i++)
523 map_collect_record_u64(fd, i, &rec->xdp_redirect[i]);
525 fd = bpf_map__fd(map_data[EXCEPTION_CNT]);
526 for (i = 0; i < XDP_ACTION_MAX; i++) {
527 map_collect_record_u64(fd, i, &rec->xdp_exception[i]);
530 fd = bpf_map__fd(map_data[CPUMAP_ENQUEUE_CNT]);
531 for (i = 0; i < MAX_CPUS; i++)
532 map_collect_record(fd, i, &rec->xdp_cpumap_enqueue[i]);
534 fd = bpf_map__fd(map_data[CPUMAP_KTHREAD_CNT]);
535 map_collect_record(fd, 0, &rec->xdp_cpumap_kthread);
537 fd = bpf_map__fd(map_data[DEVMAP_XMIT_CNT]);
538 map_collect_record(fd, 0, &rec->xdp_devmap_xmit);
543 static void *alloc_rec_per_cpu(int record_size)
545 unsigned int nr_cpus = bpf_num_possible_cpus();
548 array = calloc(nr_cpus, record_size);
550 fprintf(stderr, "Mem alloc error (nr_cpus:%u)\n", nr_cpus);
556 static struct stats_record *alloc_stats_record(void)
558 struct stats_record *rec;
562 /* Alloc main stats_record structure */
563 rec = calloc(1, sizeof(*rec));
565 fprintf(stderr, "Mem alloc error\n");
569 /* Alloc stats stored per CPU for each record */
570 rec_sz = sizeof(struct u64rec);
571 for (i = 0; i < REDIR_RES_MAX; i++)
572 rec->xdp_redirect[i].cpu = alloc_rec_per_cpu(rec_sz);
574 for (i = 0; i < XDP_ACTION_MAX; i++)
575 rec->xdp_exception[i].cpu = alloc_rec_per_cpu(rec_sz);
577 rec_sz = sizeof(struct datarec);
578 rec->xdp_cpumap_kthread.cpu = alloc_rec_per_cpu(rec_sz);
579 rec->xdp_devmap_xmit.cpu = alloc_rec_per_cpu(rec_sz);
581 for (i = 0; i < MAX_CPUS; i++)
582 rec->xdp_cpumap_enqueue[i].cpu = alloc_rec_per_cpu(rec_sz);
587 static void free_stats_record(struct stats_record *r)
591 for (i = 0; i < REDIR_RES_MAX; i++)
592 free(r->xdp_redirect[i].cpu);
594 for (i = 0; i < XDP_ACTION_MAX; i++)
595 free(r->xdp_exception[i].cpu);
597 free(r->xdp_cpumap_kthread.cpu);
598 free(r->xdp_devmap_xmit.cpu);
600 for (i = 0; i < MAX_CPUS; i++)
601 free(r->xdp_cpumap_enqueue[i].cpu);
606 /* Pointer swap trick */
607 static inline void swap(struct stats_record **a, struct stats_record **b)
609 struct stats_record *tmp;
616 static void stats_poll(int interval, bool err_only)
618 struct stats_record *rec, *prev;
620 rec = alloc_stats_record();
621 prev = alloc_stats_record();
625 printf("\n%s\n", __doc_err_only__);
627 /* Trick to pretty printf with thousands separators use %' */
628 setlocale(LC_NUMERIC, "en_US");
632 printf("\n%s", __doc__);
634 /* TODO Need more advanced stats on error types */
636 printf(" - Stats map0: %s\n", bpf_map__name(map_data[0]));
637 printf(" - Stats map1: %s\n", bpf_map__name(map_data[1]));
645 stats_print(rec, prev, err_only);
650 free_stats_record(rec);
651 free_stats_record(prev);
654 static void print_bpf_prog_info(void)
656 struct bpf_program *prog;
661 printf("Loaded BPF prog have %d bpf program(s)\n", tp_cnt);
662 bpf_object__for_each_program(prog, obj) {
663 printf(" - prog_fd[%d] = fd(%d)\n", i, bpf_program__fd(prog));
669 printf("Loaded BPF prog have %d map(s)\n", map_cnt);
670 bpf_object__for_each_map(map, obj) {
671 const char *name = bpf_map__name(map);
672 int fd = bpf_map__fd(map);
674 printf(" - map_data[%d] = fd(%d) name:%s\n", i, fd, name);
679 printf("Searching for (max:%d) event file descriptor(s)\n", tp_cnt);
680 for (i = 0; i < tp_cnt; i++) {
681 int fd = bpf_link__fd(tp_links[i]);
684 printf(" - event_fd[%d] = fd(%d)\n", i, fd);
688 int main(int argc, char **argv)
690 struct bpf_program *prog;
691 int longindex = 0, opt;
692 int ret = EXIT_FAILURE;
696 /* Default settings: */
697 bool errors_only = true;
700 /* Parse commands line args */
701 while ((opt = getopt_long(argc, argv, "hDSs:",
702 long_options, &longindex)) != -1) {
711 interval = atoi(optarg);
720 snprintf(filename, sizeof(filename), "%s_kern.o", argv[0]);
722 /* Remove tracepoint program when program is interrupted or killed */
723 signal(SIGINT, int_exit);
724 signal(SIGTERM, int_exit);
726 obj = bpf_object__open_file(filename, NULL);
727 if (libbpf_get_error(obj)) {
728 printf("ERROR: opening BPF object file failed\n");
733 /* load BPF program */
734 if (bpf_object__load(obj)) {
735 printf("ERROR: loading BPF object file failed\n");
739 for (type = 0; type < NUM_MAP; type++) {
741 bpf_object__find_map_by_name(obj, map_type_strings[type]);
743 if (libbpf_get_error(map_data[type])) {
744 printf("ERROR: finding a map in obj file failed\n");
750 bpf_object__for_each_program(prog, obj) {
751 tp_links[tp_cnt] = bpf_program__attach(prog);
752 if (libbpf_get_error(tp_links[tp_cnt])) {
753 printf("ERROR: bpf_program__attach failed\n");
754 tp_links[tp_cnt] = NULL;
761 print_bpf_prog_info();
764 /* Unload/stop tracepoint event by closing bpf_link's */
766 /* The bpf_link[i] depend on the order of
767 * the functions was defined in _kern.c
769 bpf_link__destroy(tp_links[2]); /* tracepoint/xdp/xdp_redirect */
772 bpf_link__destroy(tp_links[3]); /* tracepoint/xdp/xdp_redirect_map */
776 stats_poll(interval, errors_only);
781 /* Detach tracepoints */
783 bpf_link__destroy(tp_links[--tp_cnt]);
785 bpf_object__close(obj);