1 // SPDX-License-Identifier: GPL-2.0-only
6 #include <bpf/libbpf.h>
10 #include <linux/ethtool.h>
11 #include <linux/hashtable.h>
12 #include <linux/if_link.h>
13 #include <linux/jhash.h>
14 #include <linux/limits.h>
15 #include <linux/list.h>
16 #include <linux/sockios.h>
26 #include <sys/ioctl.h>
28 #include <sys/signalfd.h>
29 #include <sys/sysinfo.h>
30 #include <sys/timerfd.h>
31 #include <sys/utsname.h>
36 #include "xdp_sample_user.h"
38 #define __sample_print(fmt, cond, ...) \
41 printf(fmt, ##__VA_ARGS__); \
44 #define print_always(fmt, ...) __sample_print(fmt, 1, ##__VA_ARGS__)
45 #define print_default(fmt, ...) \
46 __sample_print(fmt, sample_log_level & LL_DEFAULT, ##__VA_ARGS__)
47 #define __print_err(err, fmt, ...) \
49 __sample_print(fmt, err > 0 || sample_log_level & LL_DEFAULT, \
51 sample_err_exp = sample_err_exp ? true : err > 0; \
53 #define print_err(err, fmt, ...) __print_err(err, fmt, ##__VA_ARGS__)
55 #define __COLUMN(x) "%'10" x " %-13s"
56 #define FMT_COLUMNf __COLUMN(".0f")
57 #define FMT_COLUMNd __COLUMN("d")
58 #define FMT_COLUMNl __COLUMN("llu")
59 #define RX(rx) rx, "rx/s"
60 #define PPS(pps) pps, "pkt/s"
61 #define DROP(drop) drop, "drop/s"
62 #define ERR(err) err, "error/s"
63 #define HITS(hits) hits, "hit/s"
64 #define XMIT(xmit) xmit, "xmit/s"
65 #define PASS(pass) pass, "pass/s"
66 #define REDIR(redir) redir, "redir/s"
67 #define NANOSEC_PER_SEC 1000000000 /* 10^9 */
69 #define XDP_UNKNOWN (XDP_REDIRECT + 1)
70 #define XDP_ACTION_MAX (XDP_UNKNOWN + 1)
71 #define XDP_REDIRECT_ERR_MAX 7
80 MAP_DEVMAP_XMIT_MULTI,
97 struct hlist_node node;
102 struct stats_record {
103 struct record rx_cnt;
104 struct record redir_err[XDP_REDIRECT_ERR_MAX];
105 struct record kthread;
106 struct record exception[XDP_ACTION_MAX];
107 struct record devmap_xmit;
108 DECLARE_HASHTABLE(xmit_map, 5);
112 struct sample_output {
148 } sample_xdp_progs[32];
150 struct datarec *sample_mmap[NUM_MAP];
151 struct bpf_map *sample_map[NUM_MAP];
152 size_t sample_map_count[NUM_MAP];
153 enum log_level sample_log_level;
154 struct sample_output sample_out;
155 unsigned long sample_interval;
162 static const char *xdp_redirect_err_names[XDP_REDIRECT_ERR_MAX] = {
163 /* Key=1 keeps unknown errors */
173 /* Keyed from Unknown */
174 static const char *xdp_redirect_err_help[XDP_REDIRECT_ERR_MAX - 1] = {
176 "Invalid redirection",
177 "Device being redirected to is down",
178 "Packet length too large for device",
179 "Operation not supported",
180 "No space in ptr_ring of cpumap kthread",
183 static const char *xdp_action_names[XDP_ACTION_MAX] = {
184 [XDP_ABORTED] = "XDP_ABORTED",
185 [XDP_DROP] = "XDP_DROP",
186 [XDP_PASS] = "XDP_PASS",
188 [XDP_REDIRECT] = "XDP_REDIRECT",
189 [XDP_UNKNOWN] = "XDP_UNKNOWN",
192 static __u64 gettime(void)
197 res = clock_gettime(CLOCK_MONOTONIC, &t);
199 fprintf(stderr, "Error with gettimeofday! (%i)\n", res);
202 return (__u64)t.tv_sec * NANOSEC_PER_SEC + t.tv_nsec;
205 static const char *action2str(int action)
207 if (action < XDP_ACTION_MAX)
208 return xdp_action_names[action];
212 static void sample_print_help(int mask)
214 printf("Output format description\n\n"
215 "By default, redirect success statistics are disabled, use -s to enable.\n"
216 "The terse output mode is default, verbose mode can be activated using -v\n"
217 "Use SIGQUIT (Ctrl + \\) to switch the mode dynamically at runtime\n\n"
218 "Terse mode displays at most the following fields:\n"
219 " rx/s Number of packets received per second\n"
220 " redir/s Number of packets successfully redirected per second\n"
221 " err,drop/s Aggregated count of errors per second (including dropped packets)\n"
222 " xmit/s Number of packets transmitted on the output device per second\n\n"
223 "Output description for verbose mode:\n"
224 " FIELD DESCRIPTION\n");
226 if (mask & SAMPLE_RX_CNT) {
227 printf(" receive\t\tDisplays the number of packets received & errors encountered\n"
228 " \t\t\tWhenever an error or packet drop occurs, details of per CPU error\n"
229 " \t\t\tand drop statistics will be expanded inline in terse mode.\n"
230 " \t\t\t\tpkt/s - Packets received per second\n"
231 " \t\t\t\tdrop/s - Packets dropped per second\n"
232 " \t\t\t\terror/s - Errors encountered per second\n\n");
234 if (mask & (SAMPLE_REDIRECT_CNT | SAMPLE_REDIRECT_ERR_CNT)) {
235 printf(" redirect\t\tDisplays the number of packets successfully redirected\n"
236 " \t\t\tErrors encountered are expanded under redirect_err field\n"
237 " \t\t\tNote that passing -s to enable it has a per packet overhead\n"
238 " \t\t\t\tredir/s - Packets redirected successfully per second\n\n"
239 " redirect_err\t\tDisplays the number of packets that failed redirection\n"
240 " \t\t\tThe errno is expanded under this field with per CPU count\n"
241 " \t\t\tThe recognized errors are:\n");
243 for (int i = 2; i < XDP_REDIRECT_ERR_MAX; i++)
244 printf("\t\t\t %s: %s\n", xdp_redirect_err_names[i],
245 xdp_redirect_err_help[i - 1]);
247 printf(" \n\t\t\t\terror/s - Packets that failed redirection per second\n\n");
250 if (mask & SAMPLE_CPUMAP_ENQUEUE_CNT) {
251 printf(" enqueue to cpu N\tDisplays the number of packets enqueued to bulk queue of CPU N\n"
252 " \t\t\tExpands to cpu:FROM->N to display enqueue stats for each CPU enqueuing to CPU N\n"
253 " \t\t\tReceived packets can be associated with the CPU redirect program is enqueuing \n"
254 " \t\t\tpackets to.\n"
255 " \t\t\t\tpkt/s - Packets enqueued per second from other CPU to CPU N\n"
256 " \t\t\t\tdrop/s - Packets dropped when trying to enqueue to CPU N\n"
257 " \t\t\t\tbulk-avg - Average number of packets processed for each event\n\n");
260 if (mask & SAMPLE_CPUMAP_KTHREAD_CNT) {
261 printf(" kthread\t\tDisplays the number of packets processed in CPUMAP kthread for each CPU\n"
262 " \t\t\tPackets consumed from ptr_ring in kthread, and its xdp_stats (after calling \n"
263 " \t\t\tCPUMAP bpf prog) are expanded below this. xdp_stats are expanded as a total and\n"
264 " \t\t\tthen per-CPU to associate it to each CPU's pinned CPUMAP kthread.\n"
265 " \t\t\t\tpkt/s - Packets consumed per second from ptr_ring\n"
266 " \t\t\t\tdrop/s - Packets dropped per second in kthread\n"
267 " \t\t\t\tsched - Number of times kthread called schedule()\n\n"
268 " \t\t\txdp_stats (also expands to per-CPU counts)\n"
269 " \t\t\t\tpass/s - XDP_PASS count for CPUMAP program execution\n"
270 " \t\t\t\tdrop/s - XDP_DROP count for CPUMAP program execution\n"
271 " \t\t\t\tredir/s - XDP_REDIRECT count for CPUMAP program execution\n\n");
274 if (mask & SAMPLE_EXCEPTION_CNT) {
275 printf(" xdp_exception\t\tDisplays xdp_exception tracepoint events\n"
276 " \t\t\tThis can occur due to internal driver errors, unrecognized\n"
277 " \t\t\tXDP actions and due to explicit user trigger by use of XDP_ABORTED\n"
278 " \t\t\tEach action is expanded below this field with its count\n"
279 " \t\t\t\thit/s - Number of times the tracepoint was hit per second\n\n");
282 if (mask & SAMPLE_DEVMAP_XMIT_CNT) {
283 printf(" devmap_xmit\t\tDisplays devmap_xmit tracepoint events\n"
284 " \t\t\tThis tracepoint is invoked for successful transmissions on output\n"
285 " \t\t\tdevice but these statistics are not available for generic XDP mode,\n"
286 " \t\t\thence they will be omitted from the output when using SKB mode\n"
287 " \t\t\t\txmit/s - Number of packets that were transmitted per second\n"
288 " \t\t\t\tdrop/s - Number of packets that failed transmissions per second\n"
289 " \t\t\t\tdrv_err/s - Number of internal driver errors per second\n"
290 " \t\t\t\tbulk-avg - Average number of packets processed for each event\n\n");
294 void sample_usage(char *argv[], const struct option *long_options,
295 const char *doc, int mask, bool error)
300 sample_print_help(mask);
302 printf("\n%s\nOption for %s:\n", doc, argv[0]);
303 for (i = 0; long_options[i].name != 0; i++) {
304 printf(" --%-15s", long_options[i].name);
305 if (long_options[i].flag != NULL)
306 printf(" flag (internal value: %d)",
307 *long_options[i].flag);
309 printf("\t short-option: -%c", long_options[i].val);
315 static struct datarec *alloc_record_per_cpu(void)
317 unsigned int nr_cpus = libbpf_num_possible_cpus();
318 struct datarec *array;
320 array = calloc(nr_cpus, sizeof(*array));
322 fprintf(stderr, "Failed to allocate memory (nr_cpus: %u)\n",
329 static int map_entry_init(struct map_entry *e, __u64 pair)
332 INIT_HLIST_NODE(&e->node);
333 e->val.timestamp = gettime();
334 e->val.cpu = alloc_record_per_cpu();
340 static void map_collect_percpu(struct datarec *values, struct record *rec)
342 /* For percpu maps, userspace gets a value per possible CPU */
343 unsigned int nr_cpus = libbpf_num_possible_cpus();
344 __u64 sum_xdp_redirect = 0;
345 __u64 sum_processed = 0;
346 __u64 sum_xdp_pass = 0;
347 __u64 sum_xdp_drop = 0;
348 __u64 sum_dropped = 0;
352 /* Get time as close as possible to reading map contents */
353 rec->timestamp = gettime();
355 /* Record and sum values from each CPU */
356 for (i = 0; i < nr_cpus; i++) {
357 rec->cpu[i].processed = READ_ONCE(values[i].processed);
358 rec->cpu[i].dropped = READ_ONCE(values[i].dropped);
359 rec->cpu[i].issue = READ_ONCE(values[i].issue);
360 rec->cpu[i].xdp_pass = READ_ONCE(values[i].xdp_pass);
361 rec->cpu[i].xdp_drop = READ_ONCE(values[i].xdp_drop);
362 rec->cpu[i].xdp_redirect = READ_ONCE(values[i].xdp_redirect);
364 sum_processed += rec->cpu[i].processed;
365 sum_dropped += rec->cpu[i].dropped;
366 sum_issue += rec->cpu[i].issue;
367 sum_xdp_pass += rec->cpu[i].xdp_pass;
368 sum_xdp_drop += rec->cpu[i].xdp_drop;
369 sum_xdp_redirect += rec->cpu[i].xdp_redirect;
372 rec->total.processed = sum_processed;
373 rec->total.dropped = sum_dropped;
374 rec->total.issue = sum_issue;
375 rec->total.xdp_pass = sum_xdp_pass;
376 rec->total.xdp_drop = sum_xdp_drop;
377 rec->total.xdp_redirect = sum_xdp_redirect;
380 static int map_collect_percpu_devmap(int map_fd, struct stats_record *rec)
382 unsigned int nr_cpus = bpf_num_possible_cpus();
383 __u32 batch, count = 32;
384 struct datarec *values;
389 keys = calloc(count, sizeof(__u64));
392 values = calloc(count * nr_cpus, sizeof(struct datarec));
401 ret = bpf_map_lookup_batch(map_fd, init ? &batch : NULL, &batch,
402 keys, values, &count, NULL);
403 if (ret < 0 && errno != ENOENT)
409 for (i = 0; i < count; i++) {
410 struct map_entry *e, *x = NULL;
411 __u64 pair = keys[i];
414 arr = &values[i * nr_cpus];
415 hash_for_each_possible(rec->xmit_map, e, node, pair) {
416 if (e->pair == pair) {
422 x = calloc(1, sizeof(*x));
425 if (map_entry_init(x, pair) < 0) {
429 hash_add(rec->xmit_map, &x->node, pair);
431 map_collect_percpu(arr, &x->val);
448 static struct stats_record *alloc_stats_record(void)
450 struct stats_record *rec;
453 rec = calloc(1, sizeof(*rec) + sample_n_cpus * sizeof(struct record));
455 fprintf(stderr, "Failed to allocate memory\n");
459 if (sample_mask & SAMPLE_RX_CNT) {
460 rec->rx_cnt.cpu = alloc_record_per_cpu();
461 if (!rec->rx_cnt.cpu) {
463 "Failed to allocate rx_cnt per-CPU array\n");
467 if (sample_mask & (SAMPLE_REDIRECT_CNT | SAMPLE_REDIRECT_ERR_CNT)) {
468 for (i = 0; i < XDP_REDIRECT_ERR_MAX; i++) {
469 rec->redir_err[i].cpu = alloc_record_per_cpu();
470 if (!rec->redir_err[i].cpu) {
472 "Failed to allocate redir_err per-CPU array for "
474 xdp_redirect_err_names[i]);
476 free(rec->redir_err[i].cpu);
481 if (sample_mask & SAMPLE_CPUMAP_KTHREAD_CNT) {
482 rec->kthread.cpu = alloc_record_per_cpu();
483 if (!rec->kthread.cpu) {
485 "Failed to allocate kthread per-CPU array\n");
489 if (sample_mask & SAMPLE_EXCEPTION_CNT) {
490 for (i = 0; i < XDP_ACTION_MAX; i++) {
491 rec->exception[i].cpu = alloc_record_per_cpu();
492 if (!rec->exception[i].cpu) {
494 "Failed to allocate exception per-CPU array for "
498 free(rec->exception[i].cpu);
503 if (sample_mask & SAMPLE_DEVMAP_XMIT_CNT) {
504 rec->devmap_xmit.cpu = alloc_record_per_cpu();
505 if (!rec->devmap_xmit.cpu) {
507 "Failed to allocate devmap_xmit per-CPU array\n");
511 if (sample_mask & SAMPLE_DEVMAP_XMIT_CNT_MULTI)
512 hash_init(rec->xmit_map);
513 if (sample_mask & SAMPLE_CPUMAP_ENQUEUE_CNT) {
514 for (i = 0; i < sample_n_cpus; i++) {
515 rec->enq[i].cpu = alloc_record_per_cpu();
516 if (!rec->enq[i].cpu) {
518 "Failed to allocate enqueue per-CPU array for "
522 free(rec->enq[i].cpu);
523 goto end_devmap_xmit;
531 free(rec->devmap_xmit.cpu);
533 for (i = 0; i < XDP_ACTION_MAX; i++)
534 free(rec->exception[i].cpu);
536 free(rec->kthread.cpu);
538 for (i = 0; i < XDP_REDIRECT_ERR_MAX; i++)
539 free(rec->redir_err[i].cpu);
541 free(rec->rx_cnt.cpu);
547 static void free_stats_record(struct stats_record *r)
549 struct hlist_node *tmp;
553 for (i = 0; i < sample_n_cpus; i++)
555 hash_for_each_safe(r->xmit_map, i, tmp, e, node) {
560 free(r->devmap_xmit.cpu);
561 for (i = 0; i < XDP_ACTION_MAX; i++)
562 free(r->exception[i].cpu);
563 free(r->kthread.cpu);
564 for (i = 0; i < XDP_REDIRECT_ERR_MAX; i++)
565 free(r->redir_err[i].cpu);
570 static double calc_period(struct record *r, struct record *p)
575 period = r->timestamp - p->timestamp;
577 period_ = ((double)period / NANOSEC_PER_SEC);
582 static double sample_round(double val)
584 if (val - floor(val) < 0.5)
589 static __u64 calc_pps(struct datarec *r, struct datarec *p, double period_)
595 packets = r->processed - p->processed;
596 pps = sample_round(packets / period_);
601 static __u64 calc_drop_pps(struct datarec *r, struct datarec *p, double period_)
607 packets = r->dropped - p->dropped;
608 pps = sample_round(packets / period_);
613 static __u64 calc_errs_pps(struct datarec *r, struct datarec *p, double period_)
619 packets = r->issue - p->issue;
620 pps = sample_round(packets / period_);
625 static __u64 calc_info_pps(struct datarec *r, struct datarec *p, double period_)
631 packets = r->info - p->info;
632 pps = sample_round(packets / period_);
637 static void calc_xdp_pps(struct datarec *r, struct datarec *p, double *xdp_pass,
638 double *xdp_drop, double *xdp_redirect, double period_)
640 *xdp_pass = 0, *xdp_drop = 0, *xdp_redirect = 0;
642 *xdp_redirect = (r->xdp_redirect - p->xdp_redirect) / period_;
643 *xdp_pass = (r->xdp_pass - p->xdp_pass) / period_;
644 *xdp_drop = (r->xdp_drop - p->xdp_drop) / period_;
648 static void stats_get_rx_cnt(struct stats_record *stats_rec,
649 struct stats_record *stats_prev,
650 unsigned int nr_cpus, struct sample_output *out)
652 struct record *rec, *prev;
653 double t, pps, drop, err;
656 rec = &stats_rec->rx_cnt;
657 prev = &stats_prev->rx_cnt;
658 t = calc_period(rec, prev);
660 for (i = 0; i < nr_cpus; i++) {
661 struct datarec *r = &rec->cpu[i];
662 struct datarec *p = &prev->cpu[i];
665 pps = calc_pps(r, p, t);
666 drop = calc_drop_pps(r, p, t);
667 err = calc_errs_pps(r, p, t);
668 if (!pps && !drop && !err)
671 snprintf(str, sizeof(str), "cpu:%d", i);
672 print_default(" %-18s " FMT_COLUMNf FMT_COLUMNf FMT_COLUMNf
674 str, PPS(pps), DROP(drop), ERR(err));
678 pps = calc_pps(&rec->total, &prev->total, t);
679 drop = calc_drop_pps(&rec->total, &prev->total, t);
680 err = calc_errs_pps(&rec->total, &prev->total, t);
682 out->rx_cnt.pps = pps;
683 out->rx_cnt.drop = drop;
684 out->rx_cnt.err = err;
685 out->totals.rx += pps;
686 out->totals.drop += drop;
687 out->totals.err += err;
691 static void stats_get_cpumap_enqueue(struct stats_record *stats_rec,
692 struct stats_record *stats_prev,
693 unsigned int nr_cpus)
695 struct record *rec, *prev;
696 double t, pps, drop, err;
699 /* cpumap enqueue stats */
700 for (to_cpu = 0; to_cpu < sample_n_cpus; to_cpu++) {
701 rec = &stats_rec->enq[to_cpu];
702 prev = &stats_prev->enq[to_cpu];
703 t = calc_period(rec, prev);
705 pps = calc_pps(&rec->total, &prev->total, t);
706 drop = calc_drop_pps(&rec->total, &prev->total, t);
707 err = calc_errs_pps(&rec->total, &prev->total, t);
709 if (pps > 0 || drop > 0) {
712 snprintf(str, sizeof(str), "enqueue to cpu %d", to_cpu);
715 err = pps / err; /* calc average bulk size */
718 " %-20s " FMT_COLUMNf FMT_COLUMNf __COLUMN(
720 str, PPS(pps), DROP(drop), err, "bulk-avg");
723 for (i = 0; i < nr_cpus; i++) {
724 struct datarec *r = &rec->cpu[i];
725 struct datarec *p = &prev->cpu[i];
728 pps = calc_pps(r, p, t);
729 drop = calc_drop_pps(r, p, t);
730 err = calc_errs_pps(r, p, t);
731 if (!pps && !drop && !err)
734 snprintf(str, sizeof(str), "cpu:%d->%d", i, to_cpu);
736 err = pps / err; /* calc average bulk size */
738 " %-18s " FMT_COLUMNf FMT_COLUMNf __COLUMN(
740 str, PPS(pps), DROP(drop), err, "bulk-avg");
745 static void stats_get_cpumap_remote(struct stats_record *stats_rec,
746 struct stats_record *stats_prev,
747 unsigned int nr_cpus)
749 double xdp_pass, xdp_drop, xdp_redirect;
750 struct record *rec, *prev;
754 rec = &stats_rec->kthread;
755 prev = &stats_prev->kthread;
756 t = calc_period(rec, prev);
758 calc_xdp_pps(&rec->total, &prev->total, &xdp_pass, &xdp_drop,
760 if (xdp_pass || xdp_drop || xdp_redirect) {
762 " %-18s " FMT_COLUMNf FMT_COLUMNf FMT_COLUMNf "\n",
763 "xdp_stats", PASS(xdp_pass), DROP(xdp_drop),
764 REDIR(xdp_redirect));
767 for (i = 0; i < nr_cpus; i++) {
768 struct datarec *r = &rec->cpu[i];
769 struct datarec *p = &prev->cpu[i];
772 calc_xdp_pps(r, p, &xdp_pass, &xdp_drop, &xdp_redirect, t);
773 if (!xdp_pass && !xdp_drop && !xdp_redirect)
776 snprintf(str, sizeof(str), "cpu:%d", i);
777 print_default(" %-16s " FMT_COLUMNf FMT_COLUMNf FMT_COLUMNf
779 str, PASS(xdp_pass), DROP(xdp_drop),
780 REDIR(xdp_redirect));
784 static void stats_get_cpumap_kthread(struct stats_record *stats_rec,
785 struct stats_record *stats_prev,
786 unsigned int nr_cpus)
788 struct record *rec, *prev;
789 double t, pps, drop, err;
792 rec = &stats_rec->kthread;
793 prev = &stats_prev->kthread;
794 t = calc_period(rec, prev);
796 pps = calc_pps(&rec->total, &prev->total, t);
797 drop = calc_drop_pps(&rec->total, &prev->total, t);
798 err = calc_errs_pps(&rec->total, &prev->total, t);
800 print_err(drop, " %-20s " FMT_COLUMNf FMT_COLUMNf FMT_COLUMNf "\n",
801 pps ? "kthread total" : "kthread", PPS(pps), DROP(drop), err,
804 for (i = 0; i < nr_cpus; i++) {
805 struct datarec *r = &rec->cpu[i];
806 struct datarec *p = &prev->cpu[i];
809 pps = calc_pps(r, p, t);
810 drop = calc_drop_pps(r, p, t);
811 err = calc_errs_pps(r, p, t);
812 if (!pps && !drop && !err)
815 snprintf(str, sizeof(str), "cpu:%d", i);
816 print_default(" %-18s " FMT_COLUMNf FMT_COLUMNf FMT_COLUMNf
818 str, PPS(pps), DROP(drop), err, "sched");
822 static void stats_get_redirect_cnt(struct stats_record *stats_rec,
823 struct stats_record *stats_prev,
824 unsigned int nr_cpus,
825 struct sample_output *out)
827 struct record *rec, *prev;
831 rec = &stats_rec->redir_err[0];
832 prev = &stats_prev->redir_err[0];
833 t = calc_period(rec, prev);
834 for (i = 0; i < nr_cpus; i++) {
835 struct datarec *r = &rec->cpu[i];
836 struct datarec *p = &prev->cpu[i];
839 pps = calc_pps(r, p, t);
843 snprintf(str, sizeof(str), "cpu:%d", i);
844 print_default(" %-18s " FMT_COLUMNf "\n", str, REDIR(pps));
848 pps = calc_pps(&rec->total, &prev->total, t);
849 out->redir_cnt.suc = pps;
850 out->totals.redir += pps;
854 static void stats_get_redirect_err_cnt(struct stats_record *stats_rec,
855 struct stats_record *stats_prev,
856 unsigned int nr_cpus,
857 struct sample_output *out)
859 struct record *rec, *prev;
860 double t, drop, sum = 0;
863 for (rec_i = 1; rec_i < XDP_REDIRECT_ERR_MAX; rec_i++) {
866 rec = &stats_rec->redir_err[rec_i];
867 prev = &stats_prev->redir_err[rec_i];
868 t = calc_period(rec, prev);
870 drop = calc_drop_pps(&rec->total, &prev->total, t);
871 if (drop > 0 && !out) {
872 snprintf(str, sizeof(str),
873 sample_log_level & LL_DEFAULT ? "%s total" :
875 xdp_redirect_err_names[rec_i]);
876 print_err(drop, " %-18s " FMT_COLUMNf "\n", str,
880 for (i = 0; i < nr_cpus; i++) {
881 struct datarec *r = &rec->cpu[i];
882 struct datarec *p = &prev->cpu[i];
885 drop = calc_drop_pps(r, p, t);
889 snprintf(str, sizeof(str), "cpu:%d", i);
890 print_default(" %-16s" FMT_COLUMNf "\n", str,
898 out->redir_cnt.err = sum;
899 out->totals.err += sum;
903 static void stats_get_exception_cnt(struct stats_record *stats_rec,
904 struct stats_record *stats_prev,
905 unsigned int nr_cpus,
906 struct sample_output *out)
908 double t, drop, sum = 0;
909 struct record *rec, *prev;
912 for (rec_i = 0; rec_i < XDP_ACTION_MAX; rec_i++) {
913 rec = &stats_rec->exception[rec_i];
914 prev = &stats_prev->exception[rec_i];
915 t = calc_period(rec, prev);
917 drop = calc_drop_pps(&rec->total, &prev->total, t);
918 /* Fold out errors after heading */
921 if (drop > 0 && !out) {
922 print_always(" %-18s " FMT_COLUMNf "\n",
923 action2str(rec_i), ERR(drop));
925 for (i = 0; i < nr_cpus; i++) {
926 struct datarec *r = &rec->cpu[i];
927 struct datarec *p = &prev->cpu[i];
931 drop = calc_drop_pps(r, p, t);
935 snprintf(str, sizeof(str), "cpu:%d", i);
936 print_default(" %-16s" FMT_COLUMNf "\n",
943 out->except_cnt.hits = sum;
944 out->totals.err += sum;
948 static void stats_get_devmap_xmit(struct stats_record *stats_rec,
949 struct stats_record *stats_prev,
950 unsigned int nr_cpus,
951 struct sample_output *out)
953 double pps, drop, info, err;
954 struct record *rec, *prev;
958 rec = &stats_rec->devmap_xmit;
959 prev = &stats_prev->devmap_xmit;
960 t = calc_period(rec, prev);
961 for (i = 0; i < nr_cpus; i++) {
962 struct datarec *r = &rec->cpu[i];
963 struct datarec *p = &prev->cpu[i];
966 pps = calc_pps(r, p, t);
967 drop = calc_drop_pps(r, p, t);
968 err = calc_errs_pps(r, p, t);
970 if (!pps && !drop && !err)
973 snprintf(str, sizeof(str), "cpu:%d", i);
974 info = calc_info_pps(r, p, t);
976 info = (pps + drop) / info; /* calc avg bulk */
977 print_default(" %-18s" FMT_COLUMNf FMT_COLUMNf FMT_COLUMNf
978 __COLUMN(".2f") "\n",
979 str, XMIT(pps), DROP(drop), err, "drv_err/s",
983 pps = calc_pps(&rec->total, &prev->total, t);
984 drop = calc_drop_pps(&rec->total, &prev->total, t);
985 info = calc_info_pps(&rec->total, &prev->total, t);
987 info = (pps + drop) / info; /* calc avg bulk */
988 err = calc_errs_pps(&rec->total, &prev->total, t);
990 out->xmit_cnt.pps = pps;
991 out->xmit_cnt.drop = drop;
992 out->xmit_cnt.bavg = info;
993 out->xmit_cnt.err = err;
994 out->totals.xmit += pps;
995 out->totals.drop_xmit += drop;
996 out->totals.err += err;
1000 static void stats_get_devmap_xmit_multi(struct stats_record *stats_rec,
1001 struct stats_record *stats_prev,
1002 unsigned int nr_cpus,
1003 struct sample_output *out,
1006 double pps, drop, info, err;
1007 struct map_entry *entry;
1008 struct record *r, *p;
1012 hash_for_each(stats_rec->xmit_map, bkt, entry, node) {
1013 struct map_entry *e, *x = NULL;
1014 char ifname_from[IFNAMSIZ];
1015 char ifname_to[IFNAMSIZ];
1016 const char *fstr, *tstr;
1017 unsigned long prev_time;
1018 struct record beg = {};
1019 __u32 from_idx, to_idx;
1024 prev_time = sample_interval * NANOSEC_PER_SEC;
1027 from_idx = pair >> 32;
1028 to_idx = pair & 0xFFFFFFFF;
1031 beg.timestamp = r->timestamp - prev_time;
1033 /* Find matching entry from stats_prev map */
1034 hash_for_each_possible(stats_prev->xmit_map, e, node, pair) {
1035 if (e->pair == pair) {
1044 t = calc_period(r, p);
1045 pps = calc_pps(&r->total, &p->total, t);
1046 drop = calc_drop_pps(&r->total, &p->total, t);
1047 info = calc_info_pps(&r->total, &p->total, t);
1049 info = (pps + drop) / info; /* calc avg bulk */
1050 err = calc_errs_pps(&r->total, &p->total, t);
1053 /* We are responsible for filling out totals */
1054 out->totals.xmit += pps;
1055 out->totals.drop_xmit += drop;
1056 out->totals.err += err;
1061 if (if_indextoname(from_idx, ifname_from))
1063 if (if_indextoname(to_idx, ifname_to))
1066 snprintf(str, sizeof(str), "xmit %s->%s", fstr ?: "?",
1068 /* Skip idle streams of redirection */
1069 if (pps || drop || err) {
1071 " %-20s " FMT_COLUMNf FMT_COLUMNf FMT_COLUMNf
1072 __COLUMN(".2f") "\n", str, XMIT(pps), DROP(drop),
1073 err, "drv_err/s", info, "bulk-avg");
1076 for (i = 0; i < nr_cpus; i++) {
1077 struct datarec *rc = &r->cpu[i];
1078 struct datarec *pc, p_beg = {};
1081 pc = p == &beg ? &p_beg : &p->cpu[i];
1083 pps = calc_pps(rc, pc, t);
1084 drop = calc_drop_pps(rc, pc, t);
1085 err = calc_errs_pps(rc, pc, t);
1087 if (!pps && !drop && !err)
1090 snprintf(str, sizeof(str), "cpu:%d", i);
1091 info = calc_info_pps(rc, pc, t);
1093 info = (pps + drop) / info; /* calc avg bulk */
1095 print_default(" %-18s" FMT_COLUMNf FMT_COLUMNf FMT_COLUMNf
1096 __COLUMN(".2f") "\n", str, XMIT(pps),
1097 DROP(drop), err, "drv_err/s", info, "bulk-avg");
1102 static void stats_print(const char *prefix, int mask, struct stats_record *r,
1103 struct stats_record *p, struct sample_output *out)
1105 int nr_cpus = libbpf_num_possible_cpus();
1108 print_always("%-23s", prefix ?: "Summary");
1109 if (mask & SAMPLE_RX_CNT)
1110 print_always(FMT_COLUMNl, RX(out->totals.rx));
1111 if (mask & SAMPLE_REDIRECT_CNT)
1112 print_always(FMT_COLUMNl, REDIR(out->totals.redir));
1114 out->totals.err + out->totals.drop + out->totals.drop_xmit,
1116 if (mask & SAMPLE_DEVMAP_XMIT_CNT ||
1117 mask & SAMPLE_DEVMAP_XMIT_CNT_MULTI)
1118 printf(FMT_COLUMNl, XMIT(out->totals.xmit));
1121 if (mask & SAMPLE_RX_CNT) {
1122 str = (sample_log_level & LL_DEFAULT) && out->rx_cnt.pps ?
1125 print_err((out->rx_cnt.err || out->rx_cnt.drop),
1126 " %-20s " FMT_COLUMNl FMT_COLUMNl FMT_COLUMNl "\n",
1127 str, PPS(out->rx_cnt.pps), DROP(out->rx_cnt.drop),
1128 ERR(out->rx_cnt.err));
1130 stats_get_rx_cnt(r, p, nr_cpus, NULL);
1133 if (mask & SAMPLE_CPUMAP_ENQUEUE_CNT)
1134 stats_get_cpumap_enqueue(r, p, nr_cpus);
1136 if (mask & SAMPLE_CPUMAP_KTHREAD_CNT) {
1137 stats_get_cpumap_kthread(r, p, nr_cpus);
1138 stats_get_cpumap_remote(r, p, nr_cpus);
1141 if (mask & SAMPLE_REDIRECT_CNT) {
1142 str = out->redir_cnt.suc ? "redirect total" : "redirect";
1143 print_default(" %-20s " FMT_COLUMNl "\n", str,
1144 REDIR(out->redir_cnt.suc));
1146 stats_get_redirect_cnt(r, p, nr_cpus, NULL);
1149 if (mask & SAMPLE_REDIRECT_ERR_CNT) {
1150 str = (sample_log_level & LL_DEFAULT) && out->redir_cnt.err ?
1151 "redirect_err total" :
1153 print_err(out->redir_cnt.err, " %-20s " FMT_COLUMNl "\n", str,
1154 ERR(out->redir_cnt.err));
1156 stats_get_redirect_err_cnt(r, p, nr_cpus, NULL);
1159 if (mask & SAMPLE_EXCEPTION_CNT) {
1160 str = out->except_cnt.hits ? "xdp_exception total" :
1163 print_err(out->except_cnt.hits, " %-20s " FMT_COLUMNl "\n", str,
1164 HITS(out->except_cnt.hits));
1166 stats_get_exception_cnt(r, p, nr_cpus, NULL);
1169 if (mask & SAMPLE_DEVMAP_XMIT_CNT) {
1170 str = (sample_log_level & LL_DEFAULT) && out->xmit_cnt.pps ?
1171 "devmap_xmit total" :
1174 print_err(out->xmit_cnt.err || out->xmit_cnt.drop,
1175 " %-20s " FMT_COLUMNl FMT_COLUMNl FMT_COLUMNl
1176 __COLUMN(".2f") "\n",
1177 str, XMIT(out->xmit_cnt.pps),
1178 DROP(out->xmit_cnt.drop), out->xmit_cnt.err,
1179 "drv_err/s", out->xmit_cnt.bavg, "bulk-avg");
1181 stats_get_devmap_xmit(r, p, nr_cpus, NULL);
1184 if (mask & SAMPLE_DEVMAP_XMIT_CNT_MULTI)
1185 stats_get_devmap_xmit_multi(r, p, nr_cpus, NULL,
1186 mask & SAMPLE_DEVMAP_XMIT_CNT);
1188 if (sample_log_level & LL_DEFAULT ||
1189 ((sample_log_level & LL_SIMPLE) && sample_err_exp)) {
1190 sample_err_exp = false;
1195 int sample_setup_maps(struct bpf_map **maps)
1197 sample_n_cpus = libbpf_num_possible_cpus();
1199 for (int i = 0; i < MAP_DEVMAP_XMIT_MULTI; i++) {
1200 sample_map[i] = maps[i];
1204 case MAP_CPUMAP_KTHREAD:
1205 case MAP_DEVMAP_XMIT:
1206 sample_map_count[i] = sample_n_cpus;
1208 case MAP_REDIRECT_ERR:
1209 sample_map_count[i] =
1210 XDP_REDIRECT_ERR_MAX * sample_n_cpus;
1213 sample_map_count[i] = XDP_ACTION_MAX * sample_n_cpus;
1214 case MAP_CPUMAP_ENQUEUE:
1215 sample_map_count[i] = sample_n_cpus * sample_n_cpus;
1220 if (bpf_map__set_max_entries(sample_map[i], sample_map_count[i]) < 0)
1223 sample_map[MAP_DEVMAP_XMIT_MULTI] = maps[MAP_DEVMAP_XMIT_MULTI];
1227 static int sample_setup_maps_mappings(void)
1229 for (int i = 0; i < MAP_DEVMAP_XMIT_MULTI; i++) {
1230 size_t size = sample_map_count[i] * sizeof(struct datarec);
1232 sample_mmap[i] = mmap(NULL, size, PROT_READ | PROT_WRITE,
1233 MAP_SHARED, bpf_map__fd(sample_map[i]), 0);
1234 if (sample_mmap[i] == MAP_FAILED)
1240 int __sample_init(int mask)
1245 sigaddset(&st, SIGQUIT);
1246 sigaddset(&st, SIGINT);
1247 sigaddset(&st, SIGTERM);
1249 if (sigprocmask(SIG_BLOCK, &st, NULL) < 0)
1252 sample_sig_fd = signalfd(-1, &st, SFD_CLOEXEC | SFD_NONBLOCK);
1253 if (sample_sig_fd < 0)
1258 return sample_setup_maps_mappings();
1261 static int __sample_remove_xdp(int ifindex, __u32 prog_id, int xdp_flags)
1263 __u32 cur_prog_id = 0;
1267 ret = bpf_xdp_query_id(ifindex, xdp_flags, &cur_prog_id);
1271 if (prog_id != cur_prog_id) {
1273 "Program on ifindex %d does not match installed "
1274 "program, skipping unload\n",
1280 return bpf_xdp_detach(ifindex, xdp_flags, NULL);
1283 int sample_install_xdp(struct bpf_program *xdp_prog, int ifindex, bool generic,
1286 int ret, xdp_flags = 0;
1289 if (sample_xdp_cnt == 32) {
1291 "Total limit for installed XDP programs in a sample reached\n");
1295 xdp_flags |= !force ? XDP_FLAGS_UPDATE_IF_NOEXIST : 0;
1296 xdp_flags |= generic ? XDP_FLAGS_SKB_MODE : XDP_FLAGS_DRV_MODE;
1297 ret = bpf_xdp_attach(ifindex, bpf_program__fd(xdp_prog), xdp_flags, NULL);
1301 "Failed to install program \"%s\" on ifindex %d, mode = %s, "
1303 bpf_program__name(xdp_prog), ifindex,
1304 generic ? "skb" : "native", force ? "true" : "false",
1309 ret = bpf_xdp_query_id(ifindex, xdp_flags, &prog_id);
1313 "Failed to get XDP program id for ifindex %d, removing program: %s\n",
1314 ifindex, strerror(errno));
1315 __sample_remove_xdp(ifindex, 0, xdp_flags);
1318 sample_xdp_progs[sample_xdp_cnt++] =
1319 (struct xdp_desc){ ifindex, prog_id, xdp_flags };
1324 static void sample_summary_print(void)
1326 double num = sample_out.rx_cnt.num;
1328 if (sample_out.totals.rx) {
1329 double pkts = sample_out.totals.rx;
1331 print_always(" Packets received : %'-10llu\n",
1332 sample_out.totals.rx);
1333 print_always(" Average packets/s : %'-10.0f\n",
1334 sample_round(pkts / num));
1336 if (sample_out.totals.redir) {
1337 double pkts = sample_out.totals.redir;
1339 print_always(" Packets redirected : %'-10llu\n",
1340 sample_out.totals.redir);
1341 print_always(" Average redir/s : %'-10.0f\n",
1342 sample_round(pkts / num));
1344 if (sample_out.totals.drop)
1345 print_always(" Rx dropped : %'-10llu\n",
1346 sample_out.totals.drop);
1347 if (sample_out.totals.drop_xmit)
1348 print_always(" Tx dropped : %'-10llu\n",
1349 sample_out.totals.drop_xmit);
1350 if (sample_out.totals.err)
1351 print_always(" Errors recorded : %'-10llu\n",
1352 sample_out.totals.err);
1353 if (sample_out.totals.xmit) {
1354 double pkts = sample_out.totals.xmit;
1356 print_always(" Packets transmitted : %'-10llu\n",
1357 sample_out.totals.xmit);
1358 print_always(" Average transmit/s : %'-10.0f\n",
1359 sample_round(pkts / num));
1363 void sample_exit(int status)
1367 for (int i = 0; i < NUM_MAP; i++) {
1368 size = sample_map_count[i] * sizeof(**sample_mmap);
1369 munmap(sample_mmap[i], size);
1371 while (sample_xdp_cnt--) {
1372 int i = sample_xdp_cnt, ifindex, xdp_flags;
1375 prog_id = sample_xdp_progs[i].prog_id;
1376 ifindex = sample_xdp_progs[i].ifindex;
1377 xdp_flags = sample_xdp_progs[i].flags;
1379 __sample_remove_xdp(ifindex, prog_id, xdp_flags);
1381 sample_summary_print();
1382 close(sample_sig_fd);
1386 static int sample_stats_collect(struct stats_record *rec)
1390 if (sample_mask & SAMPLE_RX_CNT)
1391 map_collect_percpu(sample_mmap[MAP_RX], &rec->rx_cnt);
1393 if (sample_mask & SAMPLE_REDIRECT_CNT)
1394 map_collect_percpu(sample_mmap[MAP_REDIRECT_ERR], &rec->redir_err[0]);
1396 if (sample_mask & SAMPLE_REDIRECT_ERR_CNT) {
1397 for (i = 1; i < XDP_REDIRECT_ERR_MAX; i++)
1398 map_collect_percpu(&sample_mmap[MAP_REDIRECT_ERR][i * sample_n_cpus],
1399 &rec->redir_err[i]);
1402 if (sample_mask & SAMPLE_CPUMAP_ENQUEUE_CNT)
1403 for (i = 0; i < sample_n_cpus; i++)
1404 map_collect_percpu(&sample_mmap[MAP_CPUMAP_ENQUEUE][i * sample_n_cpus],
1407 if (sample_mask & SAMPLE_CPUMAP_KTHREAD_CNT)
1408 map_collect_percpu(sample_mmap[MAP_CPUMAP_KTHREAD],
1411 if (sample_mask & SAMPLE_EXCEPTION_CNT)
1412 for (i = 0; i < XDP_ACTION_MAX; i++)
1413 map_collect_percpu(&sample_mmap[MAP_EXCEPTION][i * sample_n_cpus],
1414 &rec->exception[i]);
1416 if (sample_mask & SAMPLE_DEVMAP_XMIT_CNT)
1417 map_collect_percpu(sample_mmap[MAP_DEVMAP_XMIT], &rec->devmap_xmit);
1419 if (sample_mask & SAMPLE_DEVMAP_XMIT_CNT_MULTI) {
1420 if (map_collect_percpu_devmap(bpf_map__fd(sample_map[MAP_DEVMAP_XMIT_MULTI]), rec) < 0)
1426 static void sample_summary_update(struct sample_output *out)
1428 sample_out.totals.rx += out->totals.rx;
1429 sample_out.totals.redir += out->totals.redir;
1430 sample_out.totals.drop += out->totals.drop;
1431 sample_out.totals.drop_xmit += out->totals.drop_xmit;
1432 sample_out.totals.err += out->totals.err;
1433 sample_out.totals.xmit += out->totals.xmit;
1434 sample_out.rx_cnt.num++;
1437 static void sample_stats_print(int mask, struct stats_record *cur,
1438 struct stats_record *prev, char *prog_name)
1440 struct sample_output out = {};
1442 if (mask & SAMPLE_RX_CNT)
1443 stats_get_rx_cnt(cur, prev, 0, &out);
1444 if (mask & SAMPLE_REDIRECT_CNT)
1445 stats_get_redirect_cnt(cur, prev, 0, &out);
1446 if (mask & SAMPLE_REDIRECT_ERR_CNT)
1447 stats_get_redirect_err_cnt(cur, prev, 0, &out);
1448 if (mask & SAMPLE_EXCEPTION_CNT)
1449 stats_get_exception_cnt(cur, prev, 0, &out);
1450 if (mask & SAMPLE_DEVMAP_XMIT_CNT)
1451 stats_get_devmap_xmit(cur, prev, 0, &out);
1452 else if (mask & SAMPLE_DEVMAP_XMIT_CNT_MULTI)
1453 stats_get_devmap_xmit_multi(cur, prev, 0, &out,
1454 mask & SAMPLE_DEVMAP_XMIT_CNT);
1455 sample_summary_update(&out);
1457 stats_print(prog_name, mask, cur, prev, &out);
1460 void sample_switch_mode(void)
1462 sample_log_level ^= LL_DEBUG - 1;
1465 static int sample_signal_cb(void)
1467 struct signalfd_siginfo si;
1470 r = read(sample_sig_fd, &si, sizeof(si));
1474 switch (si.ssi_signo) {
1476 sample_switch_mode();
1487 /* Pointer swap trick */
1488 static void swap(struct stats_record **a, struct stats_record **b)
1490 struct stats_record *tmp;
1497 static int sample_timer_cb(int timerfd, struct stats_record **rec,
1498 struct stats_record **prev)
1500 char line[64] = "Summary";
1504 ret = read(timerfd, &t, sizeof(t));
1509 ret = sample_stats_collect(*rec);
1513 if (sample_xdp_cnt == 2 && !(sample_mask & SAMPLE_SKIP_HEADING)) {
1519 if (if_indextoname(sample_xdp_progs[0].ifindex, fi))
1521 if (if_indextoname(sample_xdp_progs[1].ifindex, to))
1524 snprintf(line, sizeof(line), "%s->%s", f ?: "?", t ?: "?");
1527 sample_stats_print(sample_mask, *rec, *prev, line);
1531 int sample_run(int interval, void (*post_cb)(void *), void *ctx)
1533 struct timespec ts = { interval, 0 };
1534 struct itimerspec its = { ts, ts };
1535 struct stats_record *rec, *prev;
1536 struct pollfd pfd[2] = {};
1540 fprintf(stderr, "Incorrect interval 0\n");
1543 sample_interval = interval;
1544 /* Pretty print numbers */
1545 setlocale(LC_NUMERIC, "en_US.UTF-8");
1547 timerfd = timerfd_create(CLOCK_MONOTONIC, TFD_CLOEXEC | TFD_NONBLOCK);
1550 timerfd_settime(timerfd, 0, &its, NULL);
1552 pfd[0].fd = sample_sig_fd;
1553 pfd[0].events = POLLIN;
1555 pfd[1].fd = timerfd;
1556 pfd[1].events = POLLIN;
1559 rec = alloc_stats_record();
1562 prev = alloc_stats_record();
1566 ret = sample_stats_collect(rec);
1571 ret = poll(pfd, 2, -1);
1579 if (pfd[0].revents & POLLIN)
1580 ret = sample_signal_cb();
1581 else if (pfd[1].revents & POLLIN)
1582 ret = sample_timer_cb(timerfd, &rec, &prev);
1592 free_stats_record(prev);
1594 free_stats_record(rec);
1601 const char *get_driver_name(int ifindex)
1603 struct ethtool_drvinfo drv = {};
1604 char ifname[IF_NAMESIZE];
1605 static char drvname[32];
1606 struct ifreq ifr = {};
1609 fd = socket(AF_INET, SOCK_DGRAM, 0);
1613 if (!if_indextoname(ifindex, ifname))
1616 drv.cmd = ETHTOOL_GDRVINFO;
1617 safe_strncpy(ifr.ifr_name, ifname, sizeof(ifr.ifr_name));
1618 ifr.ifr_data = (void *)&drv;
1620 r = ioctl(fd, SIOCETHTOOL, &ifr);
1624 safe_strncpy(drvname, drv.driver, sizeof(drvname));
1632 return r == EOPNOTSUPP ? "loopback" : "[error]";
1635 int get_mac_addr(int ifindex, void *mac_addr)
1637 char ifname[IF_NAMESIZE];
1638 struct ifreq ifr = {};
1641 fd = socket(AF_INET, SOCK_DGRAM, 0);
1645 if (!if_indextoname(ifindex, ifname)) {
1650 safe_strncpy(ifr.ifr_name, ifname, sizeof(ifr.ifr_name));
1652 r = ioctl(fd, SIOCGIFHWADDR, &ifr);
1658 memcpy(mac_addr, ifr.ifr_hwaddr.sa_data, 6 * sizeof(char));
1665 __attribute__((constructor)) static void sample_ctor(void)
1667 if (libbpf_set_strict_mode(LIBBPF_STRICT_ALL) < 0) {
1668 fprintf(stderr, "Failed to set libbpf strict mode: %s\n",
1670 /* Just exit, nothing to cleanup right now */
1671 exit(EXIT_FAIL_BPF);