samples/bpf/xdp_rxq_info_user.c

   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 static __u32 prog_id;
  33
  34 static __u32 xdp_flags = XDP_FLAGS_UPDATE_IF_NOEXIST;
  35
  36 static struct bpf_map *stats_global_map;
  37 static 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
  47 static 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
  60 static 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
  83 struct config {
  84         __u32 action;
  85         int ifindex;
  86         __u32 options;
  87 };
  88 enum 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
  95 static 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
 102 static const char *action2str(int action)
 103 {
 104         if (action < XDP_ACTION_MAX)
 105                 return xdp_action_names[action];
 106         return NULL;
 107 }
 108
 109 static 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
 125 static 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
 135 static 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
 147 static 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 */
 169 static __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 */
 183 struct datarec {
 184         __u64 processed;
 185         __u64 issue;
 186 };
 187 struct record {
 188         __u64 timestamp;
 189         struct datarec total;
 190         struct datarec *cpu;
 191 };
 192 struct stats_record {
 193         struct record stats;
 194         struct record *rxq;
 195 };
 196
 197 static struct datarec *alloc_record_per_cpu(void)
 198 {
 199         unsigned int nr_cpus = bpf_num_possible_cpus();
 200         struct datarec *array;
 201         size_t size;
 202
 203         size = sizeof(struct datarec) * nr_cpus;
 204         array = malloc(size);
 205         memset(array, 0, size);
 206         if (!array) {
 207                 fprintf(stderr, "Mem alloc error (nr_cpus:%u)\n", nr_cpus);
 208                 exit(EXIT_FAIL_MEM);
 209         }
 210         return array;
 211 }
 212
 213 static struct record *alloc_record_per_rxq(void)
 214 {
 215         unsigned int nr_rxqs = bpf_map__def(rx_queue_index_map)->max_entries;
 216         struct record *array;
 217         size_t size;
 218
 219         size = sizeof(struct record) * nr_rxqs;
 220         array = malloc(size);
 221         memset(array, 0, size);
 222         if (!array) {
 223                 fprintf(stderr, "Mem alloc error (nr_rxqs:%u)\n", nr_rxqs);
 224                 exit(EXIT_FAIL_MEM);
 225         }
 226         return array;
 227 }
 228
 229 static struct stats_record *alloc_stats_record(void)
 230 {
 231         unsigned int nr_rxqs = bpf_map__def(rx_queue_index_map)->max_entries;
 232         struct stats_record *rec;
 233         int i;
 234
 235         rec = malloc(sizeof(*rec));
 236         memset(rec, 0, sizeof(*rec));
 237         if (!rec) {
 238                 fprintf(stderr, "Mem alloc error\n");
 239                 exit(EXIT_FAIL_MEM);
 240         }
 241         rec->rxq = alloc_record_per_rxq();
 242         for (i = 0; i < nr_rxqs; i++)
 243                 rec->rxq[i].cpu = alloc_record_per_cpu();
 244
 245         rec->stats.cpu = alloc_record_per_cpu();
 246         return rec;
 247 }
 248
 249 static void free_stats_record(struct stats_record *r)
 250 {
 251         unsigned int nr_rxqs = bpf_map__def(rx_queue_index_map)->max_entries;
 252         int i;
 253
 254         for (i = 0; i < nr_rxqs; i++)
 255                 free(r->rxq[i].cpu);
 256
 257         free(r->rxq);
 258         free(r->stats.cpu);
 259         free(r);
 260 }
 261
 262 static bool map_collect_percpu(int fd, __u32 key, struct record *rec)
 263 {
 264         /* For percpu maps, userspace gets a value per possible CPU */
 265         unsigned int nr_cpus = bpf_num_possible_cpus();
 266         struct datarec values[nr_cpus];
 267         __u64 sum_processed = 0;
 268         __u64 sum_issue = 0;
 269         int i;
 270
 271         if ((bpf_map_lookup_elem(fd, &key, values)) != 0) {
 272                 fprintf(stderr,
 273                         "ERR: bpf_map_lookup_elem failed key:0x%X\n", key);
 274                 return false;
 275         }
 276         /* Get time as close as possible to reading map contents */
 277         rec->timestamp = gettime();
 278
 279         /* Record and sum values from each CPU */
 280         for (i = 0; i < nr_cpus; i++) {
 281                 rec->cpu[i].processed = values[i].processed;
 282                 sum_processed        += values[i].processed;
 283                 rec->cpu[i].issue = values[i].issue;
 284                 sum_issue        += values[i].issue;
 285         }
 286         rec->total.processed = sum_processed;
 287         rec->total.issue     = sum_issue;
 288         return true;
 289 }
 290
 291 static void stats_collect(struct stats_record *rec)
 292 {
 293         int fd, i, max_rxqs;
 294
 295         fd = bpf_map__fd(stats_global_map);
 296         map_collect_percpu(fd, 0, &rec->stats);
 297
 298         fd = bpf_map__fd(rx_queue_index_map);
 299         max_rxqs = bpf_map__def(rx_queue_index_map)->max_entries;
 300         for (i = 0; i < max_rxqs; i++)
 301                 map_collect_percpu(fd, i, &rec->rxq[i]);
 302 }
 303
 304 static double calc_period(struct record *r, struct record *p)
 305 {
 306         double period_ = 0;
 307         __u64 period = 0;
 308
 309         period = r->timestamp - p->timestamp;
 310         if (period > 0)
 311                 period_ = ((double) period / NANOSEC_PER_SEC);
 312
 313         return period_;
 314 }
 315
 316 static __u64 calc_pps(struct datarec *r, struct datarec *p, double period_)
 317 {
 318         __u64 packets = 0;
 319         __u64 pps = 0;
 320
 321         if (period_ > 0) {
 322                 packets = r->processed - p->processed;
 323                 pps = packets / period_;
 324         }
 325         return pps;
 326 }
 327
 328 static __u64 calc_errs_pps(struct datarec *r,
 329                             struct datarec *p, double period_)
 330 {
 331         __u64 packets = 0;
 332         __u64 pps = 0;
 333
 334         if (period_ > 0) {
 335                 packets = r->issue - p->issue;
 336                 pps = packets / period_;
 337         }
 338         return pps;
 339 }
 340
 341 static void stats_print(struct stats_record *stats_rec,
 342                         struct stats_record *stats_prev,
 343                         int action, __u32 cfg_opt)
 344 {
 345         unsigned int nr_rxqs = bpf_map__def(rx_queue_index_map)->max_entries;
 346         unsigned int nr_cpus = bpf_num_possible_cpus();
 347         double pps = 0, err = 0;
 348         struct record *rec, *prev;
 349         double t;
 350         int rxq;
 351         int i;
 352
 353         /* Header */
 354         printf("\nRunning XDP on dev:%s (ifindex:%d) action:%s options:%s\n",
 355                ifname, ifindex, action2str(action), options2str(cfg_opt));
 356
 357         /* stats_global_map */
 358         {
 359                 char *fmt_rx = "%-15s %-7d %'-11.0f %'-10.0f %s\n";
 360                 char *fm2_rx = "%-15s %-7s %'-11.0f\n";
 361                 char *errstr = "";
 362
 363                 printf("%-15s %-7s %-11s %-11s\n",
 364                        "XDP stats", "CPU", "pps", "issue-pps");
 365
 366                 rec  =  &stats_rec->stats;
 367                 prev = &stats_prev->stats;
 368                 t = calc_period(rec, prev);
 369                 for (i = 0; i < nr_cpus; i++) {
 370                         struct datarec *r = &rec->cpu[i];
 371                         struct datarec *p = &prev->cpu[i];
 372
 373                         pps = calc_pps     (r, p, t);
 374                         err = calc_errs_pps(r, p, t);
 375                         if (err > 0)
 376                                 errstr = "invalid-ifindex";
 377                         if (pps > 0)
 378                                 printf(fmt_rx, "XDP-RX CPU",
 379                                         i, pps, err, errstr);
 380                 }
 381                 pps  = calc_pps     (&rec->total, &prev->total, t);
 382                 err  = calc_errs_pps(&rec->total, &prev->total, t);
 383                 printf(fm2_rx, "XDP-RX CPU", "total", pps, err);
 384         }
 385
 386         /* rx_queue_index_map */
 387         printf("\n%-15s %-7s %-11s %-11s\n",
 388                "RXQ stats", "RXQ:CPU", "pps", "issue-pps");
 389
 390         for (rxq = 0; rxq < nr_rxqs; rxq++) {
 391                 char *fmt_rx = "%-15s %3d:%-3d %'-11.0f %'-10.0f %s\n";
 392                 char *fm2_rx = "%-15s %3d:%-3s %'-11.0f\n";
 393                 char *errstr = "";
 394                 int rxq_ = rxq;
 395
 396                 /* Last RXQ in map catch overflows */
 397                 if (rxq_ == nr_rxqs - 1)
 398                         rxq_ = -1;
 399
 400                 rec  =  &stats_rec->rxq[rxq];
 401                 prev = &stats_prev->rxq[rxq];
 402                 t = calc_period(rec, prev);
 403                 for (i = 0; i < nr_cpus; i++) {
 404                         struct datarec *r = &rec->cpu[i];
 405                         struct datarec *p = &prev->cpu[i];
 406
 407                         pps = calc_pps     (r, p, t);
 408                         err = calc_errs_pps(r, p, t);
 409                         if (err > 0) {
 410                                 if (rxq_ == -1)
 411                                         errstr = "map-overflow-RXQ";
 412                                 else
 413                                         errstr = "err";
 414                         }
 415                         if (pps > 0)
 416                                 printf(fmt_rx, "rx_queue_index",
 417                                        rxq_, i, pps, err, errstr);
 418                 }
 419                 pps  = calc_pps     (&rec->total, &prev->total, t);
 420                 err  = calc_errs_pps(&rec->total, &prev->total, t);
 421                 if (pps || err)
 422                         printf(fm2_rx, "rx_queue_index", rxq_, "sum", pps, err);
 423         }
 424 }
 425
 426
 427 /* Pointer swap trick */
 428 static inline void swap(struct stats_record **a, struct stats_record **b)
 429 {
 430         struct stats_record *tmp;
 431
 432         tmp = *a;
 433         *a = *b;
 434         *b = tmp;
 435 }
 436
 437 static void stats_poll(int interval, int action, __u32 cfg_opt)
 438 {
 439         struct stats_record *record, *prev;
 440
 441         record = alloc_stats_record();
 442         prev   = alloc_stats_record();
 443         stats_collect(record);
 444
 445         while (1) {
 446                 swap(&prev, &record);
 447                 stats_collect(record);
 448                 stats_print(record, prev, action, cfg_opt);
 449                 sleep(interval);
 450         }
 451
 452         free_stats_record(record);
 453         free_stats_record(prev);
 454 }
 455
 456
 457 int main(int argc, char **argv)
 458 {
 459         __u32 cfg_options= NO_TOUCH ; /* Default: Don't touch packet memory */
 460         struct rlimit r = {10 * 1024 * 1024, RLIM_INFINITY};
 461         struct bpf_prog_load_attr prog_load_attr = {
 462                 .prog_type      = BPF_PROG_TYPE_XDP,
 463         };
 464         struct bpf_prog_info info = {};
 465         __u32 info_len = sizeof(info);
 466         int prog_fd, map_fd, opt, err;
 467         bool use_separators = true;
 468         struct config cfg = { 0 };
 469         struct bpf_object *obj;
 470         struct bpf_map *map;
 471         char filename[256];
 472         int longindex = 0;
 473         int interval = 2;
 474         __u32 key = 0;
 475
 476
 477         char action_str_buf[XDP_ACTION_MAX_STRLEN + 1 /* for \0 */] = { 0 };
 478         int action = XDP_PASS; /* Default action */
 479         char *action_str = NULL;
 480
 481         snprintf(filename, sizeof(filename), "%s_kern.o", argv[0]);
 482         prog_load_attr.file = filename;
 483
 484         if (setrlimit(RLIMIT_MEMLOCK, &r)) {
 485                 perror("setrlimit(RLIMIT_MEMLOCK)");
 486                 return 1;
 487         }
 488
 489         if (bpf_prog_load_xattr(&prog_load_attr, &obj, &prog_fd))
 490                 return EXIT_FAIL;
 491
 492         map =  bpf_object__find_map_by_name(obj, "config_map");
 493         stats_global_map = bpf_object__find_map_by_name(obj, "stats_global_map");
 494         rx_queue_index_map = bpf_object__find_map_by_name(obj, "rx_queue_index_map");
 495         if (!map || !stats_global_map || !rx_queue_index_map) {
 496                 printf("finding a map in obj file failed\n");
 497                 return EXIT_FAIL;
 498         }
 499         map_fd = bpf_map__fd(map);
 500
 501         if (!prog_fd) {
 502                 fprintf(stderr, "ERR: bpf_prog_load_xattr: %s\n", strerror(errno));
 503                 return EXIT_FAIL;
 504         }
 505
 506         /* Parse commands line args */
 507         while ((opt = getopt_long(argc, argv, "FhSrmzd:s:a:",
 508                                   long_options, &longindex)) != -1) {
 509                 switch (opt) {
 510                 case 'd':
 511                         if (strlen(optarg) >= IF_NAMESIZE) {
 512                                 fprintf(stderr, "ERR: --dev name too long\n");
 513                                 goto error;
 514                         }
 515                         ifname = (char *)&ifname_buf;
 516                         strncpy(ifname, optarg, IF_NAMESIZE);
 517                         ifindex = if_nametoindex(ifname);
 518                         if (ifindex == 0) {
 519                                 fprintf(stderr,
 520                                         "ERR: --dev name unknown err(%d):%s\n",
 521                                         errno, strerror(errno));
 522                                 goto error;
 523                         }
 524                         break;
 525                 case 's':
 526                         interval = atoi(optarg);
 527                         break;
 528                 case 'S':
 529                         xdp_flags |= XDP_FLAGS_SKB_MODE;
 530                         break;
 531                 case 'z':
 532                         use_separators = false;
 533                         break;
 534                 case 'a':
 535                         action_str = (char *)&action_str_buf;
 536                         strncpy(action_str, optarg, XDP_ACTION_MAX_STRLEN);
 537                         break;
 538                 case 'r':
 539                         cfg_options |= READ_MEM;
 540                         break;
 541                 case 'm':
 542                         cfg_options |= SWAP_MAC;
 543                         break;
 544                 case 'F':
 545                         xdp_flags &= ~XDP_FLAGS_UPDATE_IF_NOEXIST;
 546                         break;
 547                 case 'h':
 548                 error:
 549                 default:
 550                         usage(argv);
 551                         return EXIT_FAIL_OPTION;
 552                 }
 553         }
 554
 555         if (!(xdp_flags & XDP_FLAGS_SKB_MODE))
 556                 xdp_flags |= XDP_FLAGS_DRV_MODE;
 557
 558         /* Required option */
 559         if (ifindex == -1) {
 560                 fprintf(stderr, "ERR: required option --dev missing\n");
 561                 usage(argv);
 562                 return EXIT_FAIL_OPTION;
 563         }
 564         cfg.ifindex = ifindex;
 565
 566         /* Parse action string */
 567         if (action_str) {
 568                 action = parse_xdp_action(action_str);
 569                 if (action < 0) {
 570                         fprintf(stderr, "ERR: Invalid XDP --action: %s\n",
 571                                 action_str);
 572                         list_xdp_actions();
 573                         return EXIT_FAIL_OPTION;
 574                 }
 575         }
 576         cfg.action = action;
 577
 578         /* XDP_TX requires changing MAC-addrs, else HW may drop */
 579         if (action == XDP_TX)
 580                 cfg_options |= SWAP_MAC;
 581         cfg.options = cfg_options;
 582
 583         /* Trick to pretty printf with thousands separators use %' */
 584         if (use_separators)
 585                 setlocale(LC_NUMERIC, "en_US");
 586
 587         /* User-side setup ifindex in config_map */
 588         err = bpf_map_update_elem(map_fd, &key, &cfg, 0);
 589         if (err) {
 590                 fprintf(stderr, "Store config failed (err:%d)\n", err);
 591                 exit(EXIT_FAIL_BPF);
 592         }
 593
 594         /* Remove XDP program when program is interrupted or killed */
 595         signal(SIGINT, int_exit);
 596         signal(SIGTERM, int_exit);
 597
 598         if (bpf_set_link_xdp_fd(ifindex, prog_fd, xdp_flags) < 0) {
 599                 fprintf(stderr, "link set xdp fd failed\n");
 600                 return EXIT_FAIL_XDP;
 601         }
 602
 603         err = bpf_obj_get_info_by_fd(prog_fd, &info, &info_len);
 604         if (err) {
 605                 printf("can't get prog info - %s\n", strerror(errno));
 606                 return err;
 607         }
 608         prog_id = info.id;
 609
 610         stats_poll(interval, action, cfg_options);
 611         return EXIT_OK;
 612 }