1/* SPDX-License-Identifier: GPL-2.0
2 * Copyright (c) 2017 Jesper Dangaard Brouer, Red Hat Inc.
3 */
4static 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
29static int ifindex = -1;
30static char ifname_buf[IF_NAMESIZE];
31static char *ifname;
32static __u32 prog_id;
33
34static __u32 xdp_flags = XDP_FLAGS_UPDATE_IF_NOEXIST;
35
36static struct bpf_map *stats_global_map;
37static struct bpf_map *rx_queue_index_map;
38
39/* Exit return codes */
40#define EXIT_OK		0
41#define EXIT_FAIL		1
42#define EXIT_FAIL_OPTION	2
43#define EXIT_FAIL_XDP		3
44#define EXIT_FAIL_BPF		4
45#define EXIT_FAIL_MEM		5
46
47static const struct option long_options[] = {
48	{"help",	no_argument,		NULL, 'h' },
49	{"dev",		required_argument,	NULL, 'd' },
50	{"skb-mode",	no_argument,		NULL, 'S' },
51	{"sec",		required_argument,	NULL, 's' },
52	{"no-separators", no_argument,		NULL, 'z' },
53	{"action",	required_argument,	NULL, 'a' },
54	{"readmem",	no_argument,		NULL, 'r' },
55	{"swapmac",	no_argument,		NULL, 'm' },
56	{"force",	no_argument,		NULL, 'F' },
57	{0, 0, NULL,  0 }
58};
59
60static void int_exit(int sig)
61{
62	__u32 curr_prog_id = 0;
63
64	if (ifindex > -1) {
65		if (bpf_get_link_xdp_id(ifindex, &curr_prog_id, xdp_flags)) {
66			printf("bpf_get_link_xdp_id failed\n");
67			exit(EXIT_FAIL);
68		}
69		if (prog_id == curr_prog_id) {
70			fprintf(stderr,
71				"Interrupted: Removing XDP program on ifindex:%d device:%s\n",
72				ifindex, ifname);
73			bpf_set_link_xdp_fd(ifindex, -1, xdp_flags);
74		} else if (!curr_prog_id) {
75			printf("couldn't find a prog id on a given iface\n");
76		} else {
77			printf("program on interface changed, not removing\n");
78		}
79	}
80	exit(EXIT_OK);
81}
82
83struct config {
84	__u32 action;
85	int ifindex;
86	__u32 options;
87};
88enum cfg_options_flags {
89	NO_TOUCH = 0x0U,
90	READ_MEM = 0x1U,
91	SWAP_MAC = 0x2U,
92};
93#define XDP_ACTION_MAX (XDP_TX + 1)
94#define XDP_ACTION_MAX_STRLEN 11
95static const char *xdp_action_names[XDP_ACTION_MAX] = {
96	[XDP_ABORTED]	= "XDP_ABORTED",
97	[XDP_DROP]	= "XDP_DROP",
98	[XDP_PASS]	= "XDP_PASS",
99	[XDP_TX]	= "XDP_TX",
100};
101
102static const char *action2str(int action)
103{
104	if (action < XDP_ACTION_MAX)
105		return xdp_action_names[action];
106	return NULL;
107}
108
109static int parse_xdp_action(char *action_str)
110{
111	size_t maxlen;
112	__u64 action = -1;
113	int i;
114
115	for (i = 0; i < XDP_ACTION_MAX; i++) {
116		maxlen = XDP_ACTION_MAX_STRLEN;
117		if (strncmp(xdp_action_names[i], action_str, maxlen) == 0) {
118			action = i;
119			break;
120		}
121	}
122	return action;
123}
124
125static void list_xdp_actions(void)
126{
127	int i;
128
129	printf("Available XDP --action <options>\n");
130	for (i = 0; i < XDP_ACTION_MAX; i++)
131		printf("\t%s\n", xdp_action_names[i]);
132	printf("\n");
133}
134
135static char* options2str(enum cfg_options_flags flag)
136{
137	if (flag == NO_TOUCH)
138		return "no_touch";
139	if (flag & SWAP_MAC)
140		return "swapmac";
141	if (flag & READ_MEM)
142		return "read";
143	fprintf(stderr, "ERR: Unknown config option flags");
144	exit(EXIT_FAIL);
145}
146
147static void usage(char *argv[])
148{
149	int i;
150
151	printf("\nDOCUMENTATION:\n%s\n", __doc__);
152	printf(" Usage: %s (options-see-below)\n", argv[0]);
153	printf(" Listing options:\n");
154	for (i = 0; long_options[i].name != 0; i++) {
155		printf(" --%-12s", long_options[i].name);
156		if (long_options[i].flag != NULL)
157			printf(" flag (internal value:%d)",
158				*long_options[i].flag);
159		else
160			printf(" short-option: -%c",
161				long_options[i].val);
162		printf("\n");
163	}
164	printf("\n");
165	list_xdp_actions();
166}
167
168#define NANOSEC_PER_SEC 1000000000 /* 10^9 */
169static __u64 gettime(void)
170{
171	struct timespec t;
172	int res;
173
174	res = clock_gettime(CLOCK_MONOTONIC, &t);
175	if (res < 0) {
176		fprintf(stderr, "Error with gettimeofday! (%i)\n", res);
177		exit(EXIT_FAIL);
178	}
179	return (__u64) t.tv_sec * NANOSEC_PER_SEC + t.tv_nsec;
180}
181
182/* Common stats data record shared with _kern.c */
183struct datarec {
184	__u64 processed;
185	__u64 issue;
186};
187struct record {
188	__u64 timestamp;
189	struct datarec total;
190	struct datarec *cpu;
191};
192struct stats_record {
193	struct record stats;
194	struct record *rxq;
195};
196
197static struct datarec *alloc_record_per_cpu(void)
198{
199	unsigned int nr_cpus = bpf_num_possible_cpus();
200	struct datarec *array;
201
202	array = calloc(nr_cpus, sizeof(struct datarec));
203	if (!array) {
204		fprintf(stderr, "Mem alloc error (nr_cpus:%u)\n", nr_cpus);
205		exit(EXIT_FAIL_MEM);
206	}
207	return array;
208}
209
210static struct record *alloc_record_per_rxq(void)
211{
212	unsigned int nr_rxqs = bpf_map__def(rx_queue_index_map)->max_entries;
213	struct record *array;
214
215	array = calloc(nr_rxqs, sizeof(struct record));
216	if (!array) {
217		fprintf(stderr, "Mem alloc error (nr_rxqs:%u)\n", nr_rxqs);
218		exit(EXIT_FAIL_MEM);
219	}
220	return array;
221}
222
223static struct stats_record *alloc_stats_record(void)
224{
225	unsigned int nr_rxqs = bpf_map__def(rx_queue_index_map)->max_entries;
226	struct stats_record *rec;
227	int i;
228
229	rec = calloc(1, sizeof(struct stats_record));
230	if (!rec) {
231		fprintf(stderr, "Mem alloc error\n");
232		exit(EXIT_FAIL_MEM);
233	}
234	rec->rxq = alloc_record_per_rxq();
235	for (i = 0; i < nr_rxqs; i++)
236		rec->rxq[i].cpu = alloc_record_per_cpu();
237
238	rec->stats.cpu = alloc_record_per_cpu();
239	return rec;
240}
241
242static void free_stats_record(struct stats_record *r)
243{
244	unsigned int nr_rxqs = bpf_map__def(rx_queue_index_map)->max_entries;
245	int i;
246
247	for (i = 0; i < nr_rxqs; i++)
248		free(r->rxq[i].cpu);
249
250	free(r->rxq);
251	free(r->stats.cpu);
252	free(r);
253}
254
255static bool map_collect_percpu(int fd, __u32 key, struct record *rec)
256{
257	/* For percpu maps, userspace gets a value per possible CPU */
258	unsigned int nr_cpus = bpf_num_possible_cpus();
259	struct datarec values[nr_cpus];
260	__u64 sum_processed = 0;
261	__u64 sum_issue = 0;
262	int i;
263
264	if ((bpf_map_lookup_elem(fd, &key, values)) != 0) {
265		fprintf(stderr,
266			"ERR: bpf_map_lookup_elem failed key:0x%X\n", key);
267		return false;
268	}
269	/* Get time as close as possible to reading map contents */
270	rec->timestamp = gettime();
271
272	/* Record and sum values from each CPU */
273	for (i = 0; i < nr_cpus; i++) {
274		rec->cpu[i].processed = values[i].processed;
275		sum_processed        += values[i].processed;
276		rec->cpu[i].issue = values[i].issue;
277		sum_issue        += values[i].issue;
278	}
279	rec->total.processed = sum_processed;
280	rec->total.issue     = sum_issue;
281	return true;
282}
283
284static void stats_collect(struct stats_record *rec)
285{
286	int fd, i, max_rxqs;
287
288	fd = bpf_map__fd(stats_global_map);
289	map_collect_percpu(fd, 0, &rec->stats);
290
291	fd = bpf_map__fd(rx_queue_index_map);
292	max_rxqs = bpf_map__def(rx_queue_index_map)->max_entries;
293	for (i = 0; i < max_rxqs; i++)
294		map_collect_percpu(fd, i, &rec->rxq[i]);
295}
296
297static double calc_period(struct record *r, struct record *p)
298{
299	double period_ = 0;
300	__u64 period = 0;
301
302	period = r->timestamp - p->timestamp;
303	if (period > 0)
304		period_ = ((double) period / NANOSEC_PER_SEC);
305
306	return period_;
307}
308
309static __u64 calc_pps(struct datarec *r, struct datarec *p, double period_)
310{
311	__u64 packets = 0;
312	__u64 pps = 0;
313
314	if (period_ > 0) {
315		packets = r->processed - p->processed;
316		pps = packets / period_;
317	}
318	return pps;
319}
320
321static __u64 calc_errs_pps(struct datarec *r,
322			    struct datarec *p, double period_)
323{
324	__u64 packets = 0;
325	__u64 pps = 0;
326
327	if (period_ > 0) {
328		packets = r->issue - p->issue;
329		pps = packets / period_;
330	}
331	return pps;
332}
333
334static void stats_print(struct stats_record *stats_rec,
335			struct stats_record *stats_prev,
336			int action, __u32 cfg_opt)
337{
338	unsigned int nr_rxqs = bpf_map__def(rx_queue_index_map)->max_entries;
339	unsigned int nr_cpus = bpf_num_possible_cpus();
340	double pps = 0, err = 0;
341	struct record *rec, *prev;
342	double t;
343	int rxq;
344	int i;
345
346	/* Header */
347	printf("\nRunning XDP on dev:%s (ifindex:%d) action:%s options:%s\n",
348	       ifname, ifindex, action2str(action), options2str(cfg_opt));
349
350	/* stats_global_map */
351	{
352		char *fmt_rx = "%-15s %-7d %'-11.0f %'-10.0f %s\n";
353		char *fm2_rx = "%-15s %-7s %'-11.0f\n";
354		char *errstr = "";
355
356		printf("%-15s %-7s %-11s %-11s\n",
357		       "XDP stats", "CPU", "pps", "issue-pps");
358
359		rec  =  &stats_rec->stats;
360		prev = &stats_prev->stats;
361		t = calc_period(rec, prev);
362		for (i = 0; i < nr_cpus; i++) {
363			struct datarec *r = &rec->cpu[i];
364			struct datarec *p = &prev->cpu[i];
365
366			pps = calc_pps     (r, p, t);
367			err = calc_errs_pps(r, p, t);
368			if (err > 0)
369				errstr = "invalid-ifindex";
370			if (pps > 0)
371				printf(fmt_rx, "XDP-RX CPU",
372					i, pps, err, errstr);
373		}
374		pps  = calc_pps     (&rec->total, &prev->total, t);
375		err  = calc_errs_pps(&rec->total, &prev->total, t);
376		printf(fm2_rx, "XDP-RX CPU", "total", pps, err);
377	}
378
379	/* rx_queue_index_map */
380	printf("\n%-15s %-7s %-11s %-11s\n",
381	       "RXQ stats", "RXQ:CPU", "pps", "issue-pps");
382
383	for (rxq = 0; rxq < nr_rxqs; rxq++) {
384		char *fmt_rx = "%-15s %3d:%-3d %'-11.0f %'-10.0f %s\n";
385		char *fm2_rx = "%-15s %3d:%-3s %'-11.0f\n";
386		char *errstr = "";
387		int rxq_ = rxq;
388
389		/* Last RXQ in map catch overflows */
390		if (rxq_ == nr_rxqs - 1)
391			rxq_ = -1;
392
393		rec  =  &stats_rec->rxq[rxq];
394		prev = &stats_prev->rxq[rxq];
395		t = calc_period(rec, prev);
396		for (i = 0; i < nr_cpus; i++) {
397			struct datarec *r = &rec->cpu[i];
398			struct datarec *p = &prev->cpu[i];
399
400			pps = calc_pps     (r, p, t);
401			err = calc_errs_pps(r, p, t);
402			if (err > 0) {
403				if (rxq_ == -1)
404					errstr = "map-overflow-RXQ";
405				else
406					errstr = "err";
407			}
408			if (pps > 0)
409				printf(fmt_rx, "rx_queue_index",
410				       rxq_, i, pps, err, errstr);
411		}
412		pps  = calc_pps     (&rec->total, &prev->total, t);
413		err  = calc_errs_pps(&rec->total, &prev->total, t);
414		if (pps || err)
415			printf(fm2_rx, "rx_queue_index", rxq_, "sum", pps, err);
416	}
417}
418
419
420/* Pointer swap trick */
421static inline void swap(struct stats_record **a, struct stats_record **b)
422{
423	struct stats_record *tmp;
424
425	tmp = *a;
426	*a = *b;
427	*b = tmp;
428}
429
430static void stats_poll(int interval, int action, __u32 cfg_opt)
431{
432	struct stats_record *record, *prev;
433
434	record = alloc_stats_record();
435	prev   = alloc_stats_record();
436	stats_collect(record);
437
438	while (1) {
439		swap(&prev, &record);
440		stats_collect(record);
441		stats_print(record, prev, action, cfg_opt);
442		sleep(interval);
443	}
444
445	free_stats_record(record);
446	free_stats_record(prev);
447}
448
449
450int main(int argc, char **argv)
451{
452	__u32 cfg_options= NO_TOUCH ; /* Default: Don't touch packet memory */
453	struct rlimit r = {RLIM_INFINITY, RLIM_INFINITY};
454	struct bpf_prog_load_attr prog_load_attr = {
455		.prog_type	= BPF_PROG_TYPE_XDP,
456	};
457	struct bpf_prog_info info = {};
458	__u32 info_len = sizeof(info);
459	int prog_fd, map_fd, opt, err;
460	bool use_separators = true;
461	struct config cfg = { 0 };
462	struct bpf_object *obj;
463	struct bpf_map *map;
464	char filename[256];
465	int longindex = 0;
466	int interval = 2;
467	__u32 key = 0;
468
469
470	char action_str_buf[XDP_ACTION_MAX_STRLEN + 1 /* for \0 */] = { 0 };
471	int action = XDP_PASS; /* Default action */
472	char *action_str = NULL;
473
474	snprintf(filename, sizeof(filename), "%s_kern.o", argv[0]);
475	prog_load_attr.file = filename;
476
477	if (setrlimit(RLIMIT_MEMLOCK, &r)) {
478		perror("setrlimit(RLIMIT_MEMLOCK)");
479		return 1;
480	}
481
482	if (bpf_prog_load_xattr(&prog_load_attr, &obj, &prog_fd))
483		return EXIT_FAIL;
484
485	map =  bpf_object__find_map_by_name(obj, "config_map");
486	stats_global_map = bpf_object__find_map_by_name(obj, "stats_global_map");
487	rx_queue_index_map = bpf_object__find_map_by_name(obj, "rx_queue_index_map");
488	if (!map || !stats_global_map || !rx_queue_index_map) {
489		printf("finding a map in obj file failed\n");
490		return EXIT_FAIL;
491	}
492	map_fd = bpf_map__fd(map);
493
494	if (!prog_fd) {
495		fprintf(stderr, "ERR: bpf_prog_load_xattr: %s\n", strerror(errno));
496		return EXIT_FAIL;
497	}
498
499	/* Parse commands line args */
500	while ((opt = getopt_long(argc, argv, "FhSrmzd:s:a:",
501				  long_options, &longindex)) != -1) {
502		switch (opt) {
503		case 'd':
504			if (strlen(optarg) >= IF_NAMESIZE) {
505				fprintf(stderr, "ERR: --dev name too long\n");
506				goto error;
507			}
508			ifname = (char *)&ifname_buf;
509			strncpy(ifname, optarg, IF_NAMESIZE);
510			ifindex = if_nametoindex(ifname);
511			if (ifindex == 0) {
512				fprintf(stderr,
513					"ERR: --dev name unknown err(%d):%s\n",
514					errno, strerror(errno));
515				goto error;
516			}
517			break;
518		case 's':
519			interval = atoi(optarg);
520			break;
521		case 'S':
522			xdp_flags |= XDP_FLAGS_SKB_MODE;
523			break;
524		case 'z':
525			use_separators = false;
526			break;
527		case 'a':
528			action_str = (char *)&action_str_buf;
529			strncpy(action_str, optarg, XDP_ACTION_MAX_STRLEN);
530			break;
531		case 'r':
532			cfg_options |= READ_MEM;
533			break;
534		case 'm':
535			cfg_options |= SWAP_MAC;
536			break;
537		case 'F':
538			xdp_flags &= ~XDP_FLAGS_UPDATE_IF_NOEXIST;
539			break;
540		case 'h':
541		error:
542		default:
543			usage(argv);
544			return EXIT_FAIL_OPTION;
545		}
546	}
547
548	if (!(xdp_flags & XDP_FLAGS_SKB_MODE))
549		xdp_flags |= XDP_FLAGS_DRV_MODE;
550
551	/* Required option */
552	if (ifindex == -1) {
553		fprintf(stderr, "ERR: required option --dev missing\n");
554		usage(argv);
555		return EXIT_FAIL_OPTION;
556	}
557	cfg.ifindex = ifindex;
558
559	/* Parse action string */
560	if (action_str) {
561		action = parse_xdp_action(action_str);
562		if (action < 0) {
563			fprintf(stderr, "ERR: Invalid XDP --action: %s\n",
564				action_str);
565			list_xdp_actions();
566			return EXIT_FAIL_OPTION;
567		}
568	}
569	cfg.action = action;
570
571	/* XDP_TX requires changing MAC-addrs, else HW may drop */
572	if (action == XDP_TX)
573		cfg_options |= SWAP_MAC;
574	cfg.options = cfg_options;
575
576	/* Trick to pretty printf with thousands separators use %' */
577	if (use_separators)
578		setlocale(LC_NUMERIC, "en_US");
579
580	/* User-side setup ifindex in config_map */
581	err = bpf_map_update_elem(map_fd, &key, &cfg, 0);
582	if (err) {
583		fprintf(stderr, "Store config failed (err:%d)\n", err);
584		exit(EXIT_FAIL_BPF);
585	}
586
587	/* Remove XDP program when program is interrupted or killed */
588	signal(SIGINT, int_exit);
589	signal(SIGTERM, int_exit);
590
591	if (bpf_set_link_xdp_fd(ifindex, prog_fd, xdp_flags) < 0) {
592		fprintf(stderr, "link set xdp fd failed\n");
593		return EXIT_FAIL_XDP;
594	}
595
596	err = bpf_obj_get_info_by_fd(prog_fd, &info, &info_len);
597	if (err) {
598		printf("can't get prog info - %s\n", strerror(errno));
599		return err;
600	}
601	prog_id = info.id;
602
603	stats_poll(interval, action, cfg_options);
604	return EXIT_OK;
605}
606