samples/bpf/xdp_redirect_cpu_user.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /* Copyright(c) 2017 Jesper Dangaard Brouer, Red Hat, Inc.
   3  */
   4 static const char *__doc__ =
   5         " XDP redirect with a CPU-map type \"BPF_MAP_TYPE_CPUMAP\"";
   6
   7 #include <errno.h>
   8 #include <signal.h>
   9 #include <stdio.h>
  10 #include <stdlib.h>
  11 #include <stdbool.h>
  12 #include <string.h>
  13 #include <unistd.h>
  14 #include <locale.h>
  15 #include <sys/resource.h>
  16 #include <getopt.h>
  17 #include <net/if.h>
  18 #include <time.h>
  19 #include <linux/limits.h>
  20
  21 #define __must_check
  22 #include <linux/err.h>
  23
  24 #include <arpa/inet.h>
  25 #include <linux/if_link.h>
  26
  27 #define MAX_CPUS 64 /* WARNING - sync with _kern.c */
  28
  29 /* How many xdp_progs are defined in _kern.c */
  30 #define MAX_PROG 6
  31
  32 #include <bpf/bpf.h>
  33 #include "libbpf.h"
  34
  35 #include "bpf_util.h"
  36
  37 static int ifindex = -1;
  38 static char ifname_buf[IF_NAMESIZE];
  39 static char *ifname;
  40 static __u32 prog_id;
  41
  42 static __u32 xdp_flags = XDP_FLAGS_UPDATE_IF_NOEXIST;
  43 static int cpu_map_fd;
  44 static int rx_cnt_map_fd;
  45 static int redirect_err_cnt_map_fd;
  46 static int cpumap_enqueue_cnt_map_fd;
  47 static int cpumap_kthread_cnt_map_fd;
  48 static int cpus_available_map_fd;
  49 static int cpus_count_map_fd;
  50 static int cpus_iterator_map_fd;
  51 static int exception_cnt_map_fd;
  52
  53 #define NUM_TP 5
  54 struct bpf_link *tp_links[NUM_TP] = { 0 };
  55 static int tp_cnt = 0;
  56
  57 /* Exit return codes */
  58 #define EXIT_OK         0
  59 #define EXIT_FAIL               1
  60 #define EXIT_FAIL_OPTION        2
  61 #define EXIT_FAIL_XDP           3
  62 #define EXIT_FAIL_BPF           4
  63 #define EXIT_FAIL_MEM           5
  64
  65 static const struct option long_options[] = {
  66         {"help",        no_argument,            NULL, 'h' },
  67         {"dev",         required_argument,      NULL, 'd' },
  68         {"skb-mode",    no_argument,            NULL, 'S' },
  69         {"sec",         required_argument,      NULL, 's' },
  70         {"progname",    required_argument,      NULL, 'p' },
  71         {"qsize",       required_argument,      NULL, 'q' },
  72         {"cpu",         required_argument,      NULL, 'c' },
  73         {"stress-mode", no_argument,            NULL, 'x' },
  74         {"no-separators", no_argument,          NULL, 'z' },
  75         {"force",       no_argument,            NULL, 'F' },
  76         {0, 0, NULL,  0 }
  77 };
  78
  79 static void int_exit(int sig)
  80 {
  81         __u32 curr_prog_id = 0;
  82
  83         if (ifindex > -1) {
  84                 if (bpf_get_link_xdp_id(ifindex, &curr_prog_id, xdp_flags)) {
  85                         printf("bpf_get_link_xdp_id failed\n");
  86                         exit(EXIT_FAIL);
  87                 }
  88                 if (prog_id == curr_prog_id) {
  89                         fprintf(stderr,
  90                                 "Interrupted: Removing XDP program on ifindex:%d device:%s\n",
  91                                 ifindex, ifname);
  92                         bpf_set_link_xdp_fd(ifindex, -1, xdp_flags);
  93                 } else if (!curr_prog_id) {
  94                         printf("couldn't find a prog id on a given iface\n");
  95                 } else {
  96                         printf("program on interface changed, not removing\n");
  97                 }
  98         }
  99         /* Detach tracepoints */
 100         while (tp_cnt)
 101                 bpf_link__destroy(tp_links[--tp_cnt]);
 102
 103         exit(EXIT_OK);
 104 }
 105
 106 static void print_avail_progs(struct bpf_object *obj)
 107 {
 108         struct bpf_program *pos;
 109
 110         bpf_object__for_each_program(pos, obj) {
 111                 if (bpf_program__is_xdp(pos))
 112                         printf(" %s\n", bpf_program__title(pos, false));
 113         }
 114 }
 115
 116 static void usage(char *argv[], struct bpf_object *obj)
 117 {
 118         int i;
 119
 120         printf("\nDOCUMENTATION:\n%s\n", __doc__);
 121         printf("\n");
 122         printf(" Usage: %s (options-see-below)\n", argv[0]);
 123         printf(" Listing options:\n");
 124         for (i = 0; long_options[i].name != 0; i++) {
 125                 printf(" --%-12s", long_options[i].name);
 126                 if (long_options[i].flag != NULL)
 127                         printf(" flag (internal value:%d)",
 128                                 *long_options[i].flag);
 129                 else
 130                         printf(" short-option: -%c",
 131                                 long_options[i].val);
 132                 printf("\n");
 133         }
 134         printf("\n Programs to be used for --progname:\n");
 135         print_avail_progs(obj);
 136         printf("\n");
 137 }
 138
 139 /* gettime returns the current time of day in nanoseconds.
 140  * Cost: clock_gettime (ns) => 26ns (CLOCK_MONOTONIC)
 141  *       clock_gettime (ns) =>  9ns (CLOCK_MONOTONIC_COARSE)
 142  */
 143 #define NANOSEC_PER_SEC 1000000000 /* 10^9 */
 144 static __u64 gettime(void)
 145 {
 146         struct timespec t;
 147         int res;
 148
 149         res = clock_gettime(CLOCK_MONOTONIC, &t);
 150         if (res < 0) {
 151                 fprintf(stderr, "Error with gettimeofday! (%i)\n", res);
 152                 exit(EXIT_FAIL);
 153         }
 154         return (__u64) t.tv_sec * NANOSEC_PER_SEC + t.tv_nsec;
 155 }
 156
 157 /* Common stats data record shared with _kern.c */
 158 struct datarec {
 159         __u64 processed;
 160         __u64 dropped;
 161         __u64 issue;
 162 };
 163 struct record {
 164         __u64 timestamp;
 165         struct datarec total;
 166         struct datarec *cpu;
 167 };
 168 struct stats_record {
 169         struct record rx_cnt;
 170         struct record redir_err;
 171         struct record kthread;
 172         struct record exception;
 173         struct record enq[MAX_CPUS];
 174 };
 175
 176 static bool map_collect_percpu(int fd, __u32 key, struct record *rec)
 177 {
 178         /* For percpu maps, userspace gets a value per possible CPU */
 179         unsigned int nr_cpus = bpf_num_possible_cpus();
 180         struct datarec values[nr_cpus];
 181         __u64 sum_processed = 0;
 182         __u64 sum_dropped = 0;
 183         __u64 sum_issue = 0;
 184         int i;
 185
 186         if ((bpf_map_lookup_elem(fd, &key, values)) != 0) {
 187                 fprintf(stderr,
 188                         "ERR: bpf_map_lookup_elem failed key:0x%X\n", key);
 189                 return false;
 190         }
 191         /* Get time as close as possible to reading map contents */
 192         rec->timestamp = gettime();
 193
 194         /* Record and sum values from each CPU */
 195         for (i = 0; i < nr_cpus; i++) {
 196                 rec->cpu[i].processed = values[i].processed;
 197                 sum_processed        += values[i].processed;
 198                 rec->cpu[i].dropped = values[i].dropped;
 199                 sum_dropped        += values[i].dropped;
 200                 rec->cpu[i].issue = values[i].issue;
 201                 sum_issue        += values[i].issue;
 202         }
 203         rec->total.processed = sum_processed;
 204         rec->total.dropped   = sum_dropped;
 205         rec->total.issue     = sum_issue;
 206         return true;
 207 }
 208
 209 static struct datarec *alloc_record_per_cpu(void)
 210 {
 211         unsigned int nr_cpus = bpf_num_possible_cpus();
 212         struct datarec *array;
 213         size_t size;
 214
 215         size = sizeof(struct datarec) * nr_cpus;
 216         array = malloc(size);
 217         memset(array, 0, size);
 218         if (!array) {
 219                 fprintf(stderr, "Mem alloc error (nr_cpus:%u)\n", nr_cpus);
 220                 exit(EXIT_FAIL_MEM);
 221         }
 222         return array;
 223 }
 224
 225 static struct stats_record *alloc_stats_record(void)
 226 {
 227         struct stats_record *rec;
 228         int i;
 229
 230         rec = malloc(sizeof(*rec));
 231         memset(rec, 0, sizeof(*rec));
 232         if (!rec) {
 233                 fprintf(stderr, "Mem alloc error\n");
 234                 exit(EXIT_FAIL_MEM);
 235         }
 236         rec->rx_cnt.cpu    = alloc_record_per_cpu();
 237         rec->redir_err.cpu = alloc_record_per_cpu();
 238         rec->kthread.cpu   = alloc_record_per_cpu();
 239         rec->exception.cpu = alloc_record_per_cpu();
 240         for (i = 0; i < MAX_CPUS; i++)
 241                 rec->enq[i].cpu = alloc_record_per_cpu();
 242
 243         return rec;
 244 }
 245
 246 static void free_stats_record(struct stats_record *r)
 247 {
 248         int i;
 249
 250         for (i = 0; i < MAX_CPUS; i++)
 251                 free(r->enq[i].cpu);
 252         free(r->exception.cpu);
 253         free(r->kthread.cpu);
 254         free(r->redir_err.cpu);
 255         free(r->rx_cnt.cpu);
 256         free(r);
 257 }
 258
 259 static double calc_period(struct record *r, struct record *p)
 260 {
 261         double period_ = 0;
 262         __u64 period = 0;
 263
 264         period = r->timestamp - p->timestamp;
 265         if (period > 0)
 266                 period_ = ((double) period / NANOSEC_PER_SEC);
 267
 268         return period_;
 269 }
 270
 271 static __u64 calc_pps(struct datarec *r, struct datarec *p, double period_)
 272 {
 273         __u64 packets = 0;
 274         __u64 pps = 0;
 275
 276         if (period_ > 0) {
 277                 packets = r->processed - p->processed;
 278                 pps = packets / period_;
 279         }
 280         return pps;
 281 }
 282
 283 static __u64 calc_drop_pps(struct datarec *r, struct datarec *p, double period_)
 284 {
 285         __u64 packets = 0;
 286         __u64 pps = 0;
 287
 288         if (period_ > 0) {
 289                 packets = r->dropped - p->dropped;
 290                 pps = packets / period_;
 291         }
 292         return pps;
 293 }
 294
 295 static __u64 calc_errs_pps(struct datarec *r,
 296                             struct datarec *p, double period_)
 297 {
 298         __u64 packets = 0;
 299         __u64 pps = 0;
 300
 301         if (period_ > 0) {
 302                 packets = r->issue - p->issue;
 303                 pps = packets / period_;
 304         }
 305         return pps;
 306 }
 307
 308 static void stats_print(struct stats_record *stats_rec,
 309                         struct stats_record *stats_prev,
 310                         char *prog_name)
 311 {
 312         unsigned int nr_cpus = bpf_num_possible_cpus();
 313         double pps = 0, drop = 0, err = 0;
 314         struct record *rec, *prev;
 315         int to_cpu;
 316         double t;
 317         int i;
 318
 319         /* Header */
 320         printf("Running XDP/eBPF prog_name:%s\n", prog_name);
 321         printf("%-15s %-7s %-14s %-11s %-9s\n",
 322                "XDP-cpumap", "CPU:to", "pps", "drop-pps", "extra-info");
 323
 324         /* XDP rx_cnt */
 325         {
 326                 char *fmt_rx = "%-15s %-7d %'-14.0f %'-11.0f %'-10.0f %s\n";
 327                 char *fm2_rx = "%-15s %-7s %'-14.0f %'-11.0f\n";
 328                 char *errstr = "";
 329
 330                 rec  = &stats_rec->rx_cnt;
 331                 prev = &stats_prev->rx_cnt;
 332                 t = calc_period(rec, prev);
 333                 for (i = 0; i < nr_cpus; i++) {
 334                         struct datarec *r = &rec->cpu[i];
 335                         struct datarec *p = &prev->cpu[i];
 336
 337                         pps = calc_pps(r, p, t);
 338                         drop = calc_drop_pps(r, p, t);
 339                         err  = calc_errs_pps(r, p, t);
 340                         if (err > 0)
 341                                 errstr = "cpu-dest/err";
 342                         if (pps > 0)
 343                                 printf(fmt_rx, "XDP-RX",
 344                                         i, pps, drop, err, errstr);
 345                 }
 346                 pps  = calc_pps(&rec->total, &prev->total, t);
 347                 drop = calc_drop_pps(&rec->total, &prev->total, t);
 348                 err  = calc_errs_pps(&rec->total, &prev->total, t);
 349                 printf(fm2_rx, "XDP-RX", "total", pps, drop);
 350         }
 351
 352         /* cpumap enqueue stats */
 353         for (to_cpu = 0; to_cpu < MAX_CPUS; to_cpu++) {
 354                 char *fmt = "%-15s %3d:%-3d %'-14.0f %'-11.0f %'-10.2f %s\n";
 355                 char *fm2 = "%-15s %3s:%-3d %'-14.0f %'-11.0f %'-10.2f %s\n";
 356                 char *errstr = "";
 357
 358                 rec  =  &stats_rec->enq[to_cpu];
 359                 prev = &stats_prev->enq[to_cpu];
 360                 t = calc_period(rec, prev);
 361                 for (i = 0; i < nr_cpus; i++) {
 362                         struct datarec *r = &rec->cpu[i];
 363                         struct datarec *p = &prev->cpu[i];
 364
 365                         pps  = calc_pps(r, p, t);
 366                         drop = calc_drop_pps(r, p, t);
 367                         err  = calc_errs_pps(r, p, t);
 368                         if (err > 0) {
 369                                 errstr = "bulk-average";
 370                                 err = pps / err; /* calc average bulk size */
 371                         }
 372                         if (pps > 0)
 373                                 printf(fmt, "cpumap-enqueue",
 374                                        i, to_cpu, pps, drop, err, errstr);
 375                 }
 376                 pps = calc_pps(&rec->total, &prev->total, t);
 377                 if (pps > 0) {
 378                         drop = calc_drop_pps(&rec->total, &prev->total, t);
 379                         err  = calc_errs_pps(&rec->total, &prev->total, t);
 380                         if (err > 0) {
 381                                 errstr = "bulk-average";
 382                                 err = pps / err; /* calc average bulk size */
 383                         }
 384                         printf(fm2, "cpumap-enqueue",
 385                                "sum", to_cpu, pps, drop, err, errstr);
 386                 }
 387         }
 388
 389         /* cpumap kthread stats */
 390         {
 391                 char *fmt_k = "%-15s %-7d %'-14.0f %'-11.0f %'-10.0f %s\n";
 392                 char *fm2_k = "%-15s %-7s %'-14.0f %'-11.0f %'-10.0f %s\n";
 393                 char *e_str = "";
 394
 395                 rec  = &stats_rec->kthread;
 396                 prev = &stats_prev->kthread;
 397                 t = calc_period(rec, prev);
 398                 for (i = 0; i < nr_cpus; i++) {
 399                         struct datarec *r = &rec->cpu[i];
 400                         struct datarec *p = &prev->cpu[i];
 401
 402                         pps  = calc_pps(r, p, t);
 403                         drop = calc_drop_pps(r, p, t);
 404                         err  = calc_errs_pps(r, p, t);
 405                         if (err > 0)
 406                                 e_str = "sched";
 407                         if (pps > 0)
 408                                 printf(fmt_k, "cpumap_kthread",
 409                                        i, pps, drop, err, e_str);
 410                 }
 411                 pps = calc_pps(&rec->total, &prev->total, t);
 412                 drop = calc_drop_pps(&rec->total, &prev->total, t);
 413                 err  = calc_errs_pps(&rec->total, &prev->total, t);
 414                 if (err > 0)
 415                         e_str = "sched-sum";
 416                 printf(fm2_k, "cpumap_kthread", "total", pps, drop, err, e_str);
 417         }
 418
 419         /* XDP redirect err tracepoints (very unlikely) */
 420         {
 421                 char *fmt_err = "%-15s %-7d %'-14.0f %'-11.0f\n";
 422                 char *fm2_err = "%-15s %-7s %'-14.0f %'-11.0f\n";
 423
 424                 rec  = &stats_rec->redir_err;
 425                 prev = &stats_prev->redir_err;
 426                 t = calc_period(rec, prev);
 427                 for (i = 0; i < nr_cpus; i++) {
 428                         struct datarec *r = &rec->cpu[i];
 429                         struct datarec *p = &prev->cpu[i];
 430
 431                         pps  = calc_pps(r, p, t);
 432                         drop = calc_drop_pps(r, p, t);
 433                         if (pps > 0)
 434                                 printf(fmt_err, "redirect_err", i, pps, drop);
 435                 }
 436                 pps = calc_pps(&rec->total, &prev->total, t);
 437                 drop = calc_drop_pps(&rec->total, &prev->total, t);
 438                 printf(fm2_err, "redirect_err", "total", pps, drop);
 439         }
 440
 441         /* XDP general exception tracepoints */
 442         {
 443                 char *fmt_err = "%-15s %-7d %'-14.0f %'-11.0f\n";
 444                 char *fm2_err = "%-15s %-7s %'-14.0f %'-11.0f\n";
 445
 446                 rec  = &stats_rec->exception;
 447                 prev = &stats_prev->exception;
 448                 t = calc_period(rec, prev);
 449                 for (i = 0; i < nr_cpus; i++) {
 450                         struct datarec *r = &rec->cpu[i];
 451                         struct datarec *p = &prev->cpu[i];
 452
 453                         pps  = calc_pps(r, p, t);
 454                         drop = calc_drop_pps(r, p, t);
 455                         if (pps > 0)
 456                                 printf(fmt_err, "xdp_exception", i, pps, drop);
 457                 }
 458                 pps = calc_pps(&rec->total, &prev->total, t);
 459                 drop = calc_drop_pps(&rec->total, &prev->total, t);
 460                 printf(fm2_err, "xdp_exception", "total", pps, drop);
 461         }
 462
 463         printf("\n");
 464         fflush(stdout);
 465 }
 466
 467 static void stats_collect(struct stats_record *rec)
 468 {
 469         int fd, i;
 470
 471         fd = rx_cnt_map_fd;
 472         map_collect_percpu(fd, 0, &rec->rx_cnt);
 473
 474         fd = redirect_err_cnt_map_fd;
 475         map_collect_percpu(fd, 1, &rec->redir_err);
 476
 477         fd = cpumap_enqueue_cnt_map_fd;
 478         for (i = 0; i < MAX_CPUS; i++)
 479                 map_collect_percpu(fd, i, &rec->enq[i]);
 480
 481         fd = cpumap_kthread_cnt_map_fd;
 482         map_collect_percpu(fd, 0, &rec->kthread);
 483
 484         fd = exception_cnt_map_fd;
 485         map_collect_percpu(fd, 0, &rec->exception);
 486 }
 487
 488
 489 /* Pointer swap trick */
 490 static inline void swap(struct stats_record **a, struct stats_record **b)
 491 {
 492         struct stats_record *tmp;
 493
 494         tmp = *a;
 495         *a = *b;
 496         *b = tmp;
 497 }
 498
 499 static int create_cpu_entry(__u32 cpu, __u32 queue_size,
 500                             __u32 avail_idx, bool new)
 501 {
 502         __u32 curr_cpus_count = 0;
 503         __u32 key = 0;
 504         int ret;
 505
 506         /* Add a CPU entry to cpumap, as this allocate a cpu entry in
 507          * the kernel for the cpu.
 508          */
 509         ret = bpf_map_update_elem(cpu_map_fd, &cpu, &queue_size, 0);
 510         if (ret) {
 511                 fprintf(stderr, "Create CPU entry failed (err:%d)\n", ret);
 512                 exit(EXIT_FAIL_BPF);
 513         }
 514
 515         /* Inform bpf_prog's that a new CPU is available to select
 516          * from via some control maps.
 517          */
 518         ret = bpf_map_update_elem(cpus_available_map_fd, &avail_idx, &cpu, 0);
 519         if (ret) {
 520                 fprintf(stderr, "Add to avail CPUs failed\n");
 521                 exit(EXIT_FAIL_BPF);
 522         }
 523
 524         /* When not replacing/updating existing entry, bump the count */
 525         ret = bpf_map_lookup_elem(cpus_count_map_fd, &key, &curr_cpus_count);
 526         if (ret) {
 527                 fprintf(stderr, "Failed reading curr cpus_count\n");
 528                 exit(EXIT_FAIL_BPF);
 529         }
 530         if (new) {
 531                 curr_cpus_count++;
 532                 ret = bpf_map_update_elem(cpus_count_map_fd, &key,
 533                                           &curr_cpus_count, 0);
 534                 if (ret) {
 535                         fprintf(stderr, "Failed write curr cpus_count\n");
 536                         exit(EXIT_FAIL_BPF);
 537                 }
 538         }
 539         /* map_fd[7] = cpus_iterator */
 540         printf("%s CPU:%u as idx:%u queue_size:%d (total cpus_count:%u)\n",
 541                new ? "Add-new":"Replace", cpu, avail_idx,
 542                queue_size, curr_cpus_count);
 543
 544         return 0;
 545 }
 546
 547 /* CPUs are zero-indexed. Thus, add a special sentinel default value
 548  * in map cpus_available to mark CPU index'es not configured
 549  */
 550 static void mark_cpus_unavailable(void)
 551 {
 552         __u32 invalid_cpu = MAX_CPUS;
 553         int ret, i;
 554
 555         for (i = 0; i < MAX_CPUS; i++) {
 556                 ret = bpf_map_update_elem(cpus_available_map_fd, &i,
 557                                           &invalid_cpu, 0);
 558                 if (ret) {
 559                         fprintf(stderr, "Failed marking CPU unavailable\n");
 560                         exit(EXIT_FAIL_BPF);
 561                 }
 562         }
 563 }
 564
 565 /* Stress cpumap management code by concurrently changing underlying cpumap */
 566 static void stress_cpumap(void)
 567 {
 568         /* Changing qsize will cause kernel to free and alloc a new
 569          * bpf_cpu_map_entry, with an associated/complicated tear-down
 570          * procedure.
 571          */
 572         create_cpu_entry(1,  1024, 0, false);
 573         create_cpu_entry(1,     8, 0, false);
 574         create_cpu_entry(1, 16000, 0, false);
 575 }
 576
 577 static void stats_poll(int interval, bool use_separators, char *prog_name,
 578                        bool stress_mode)
 579 {
 580         struct stats_record *record, *prev;
 581
 582         record = alloc_stats_record();
 583         prev   = alloc_stats_record();
 584         stats_collect(record);
 585
 586         /* Trick to pretty printf with thousands separators use %' */
 587         if (use_separators)
 588                 setlocale(LC_NUMERIC, "en_US");
 589
 590         while (1) {
 591                 swap(&prev, &record);
 592                 stats_collect(record);
 593                 stats_print(record, prev, prog_name);
 594                 sleep(interval);
 595                 if (stress_mode)
 596                         stress_cpumap();
 597         }
 598
 599         free_stats_record(record);
 600         free_stats_record(prev);
 601 }
 602
 603 static struct bpf_link * attach_tp(struct bpf_object *obj,
 604                                    const char *tp_category,
 605                                    const char* tp_name)
 606 {
 607         struct bpf_program *prog;
 608         struct bpf_link *link;
 609         char sec_name[PATH_MAX];
 610         int len;
 611
 612         len = snprintf(sec_name, PATH_MAX, "tracepoint/%s/%s",
 613                        tp_category, tp_name);
 614         if (len < 0)
 615                 exit(EXIT_FAIL);
 616
 617         prog = bpf_object__find_program_by_title(obj, sec_name);
 618         if (!prog) {
 619                 fprintf(stderr, "ERR: finding progsec: %s\n", sec_name);
 620                 exit(EXIT_FAIL_BPF);
 621         }
 622
 623         link = bpf_program__attach_tracepoint(prog, tp_category, tp_name);
 624         if (IS_ERR(link))
 625                 exit(EXIT_FAIL_BPF);
 626
 627         return link;
 628 }
 629
 630 static void init_tracepoints(struct bpf_object *obj) {
 631         tp_links[tp_cnt++] = attach_tp(obj, "xdp", "xdp_redirect_err");
 632         tp_links[tp_cnt++] = attach_tp(obj, "xdp", "xdp_redirect_map_err");
 633         tp_links[tp_cnt++] = attach_tp(obj, "xdp", "xdp_exception");
 634         tp_links[tp_cnt++] = attach_tp(obj, "xdp", "xdp_cpumap_enqueue");
 635         tp_links[tp_cnt++] = attach_tp(obj, "xdp", "xdp_cpumap_kthread");
 636 }
 637
 638 static int init_map_fds(struct bpf_object *obj)
 639 {
 640         /* Maps updated by tracepoints */
 641         redirect_err_cnt_map_fd =
 642                 bpf_object__find_map_fd_by_name(obj, "redirect_err_cnt");
 643         exception_cnt_map_fd =
 644                 bpf_object__find_map_fd_by_name(obj, "exception_cnt");
 645         cpumap_enqueue_cnt_map_fd =
 646                 bpf_object__find_map_fd_by_name(obj, "cpumap_enqueue_cnt");
 647         cpumap_kthread_cnt_map_fd =
 648                 bpf_object__find_map_fd_by_name(obj, "cpumap_kthread_cnt");
 649
 650         /* Maps used by XDP */
 651         rx_cnt_map_fd = bpf_object__find_map_fd_by_name(obj, "rx_cnt");
 652         cpu_map_fd = bpf_object__find_map_fd_by_name(obj, "cpu_map");
 653         cpus_available_map_fd =
 654                 bpf_object__find_map_fd_by_name(obj, "cpus_available");
 655         cpus_count_map_fd = bpf_object__find_map_fd_by_name(obj, "cpus_count");
 656         cpus_iterator_map_fd =
 657                 bpf_object__find_map_fd_by_name(obj, "cpus_iterator");
 658
 659         if (cpu_map_fd < 0 || rx_cnt_map_fd < 0 ||
 660             redirect_err_cnt_map_fd < 0 || cpumap_enqueue_cnt_map_fd < 0 ||
 661             cpumap_kthread_cnt_map_fd < 0 || cpus_available_map_fd < 0 ||
 662             cpus_count_map_fd < 0 || cpus_iterator_map_fd < 0 ||
 663             exception_cnt_map_fd < 0)
 664                 return -ENOENT;
 665
 666         return 0;
 667 }
 668
 669 int main(int argc, char **argv)
 670 {
 671         struct rlimit r = {10 * 1024 * 1024, RLIM_INFINITY};
 672         char *prog_name = "xdp_cpu_map5_lb_hash_ip_pairs";
 673         struct bpf_prog_load_attr prog_load_attr = {
 674                 .prog_type      = BPF_PROG_TYPE_UNSPEC,
 675         };
 676         struct bpf_prog_info info = {};
 677         __u32 info_len = sizeof(info);
 678         bool use_separators = true;
 679         bool stress_mode = false;
 680         struct bpf_program *prog;
 681         struct bpf_object *obj;
 682         char filename[256];
 683         int added_cpus = 0;
 684         int longindex = 0;
 685         int interval = 2;
 686         int add_cpu = -1;
 687         int opt, err;
 688         int prog_fd;
 689         __u32 qsize;
 690
 691         /* Notice: choosing he queue size is very important with the
 692          * ixgbe driver, because it's driver page recycling trick is
 693          * dependend on pages being returned quickly.  The number of
 694          * out-standing packets in the system must be less-than 2x
 695          * RX-ring size.
 696          */
 697         qsize = 128+64;
 698
 699         snprintf(filename, sizeof(filename), "%s_kern.o", argv[0]);
 700         prog_load_attr.file = filename;
 701
 702         if (setrlimit(RLIMIT_MEMLOCK, &r)) {
 703                 perror("setrlimit(RLIMIT_MEMLOCK)");
 704                 return 1;
 705         }
 706
 707         if (bpf_prog_load_xattr(&prog_load_attr, &obj, &prog_fd))
 708                 return EXIT_FAIL;
 709
 710         if (prog_fd < 0) {
 711                 fprintf(stderr, "ERR: bpf_prog_load_xattr: %s\n",
 712                         strerror(errno));
 713                 return EXIT_FAIL;
 714         }
 715         init_tracepoints(obj);
 716         if (init_map_fds(obj) < 0) {
 717                 fprintf(stderr, "bpf_object__find_map_fd_by_name failed\n");
 718                 return EXIT_FAIL;
 719         }
 720         mark_cpus_unavailable();
 721
 722         /* Parse commands line args */
 723         while ((opt = getopt_long(argc, argv, "hSd:s:p:q:c:xzF",
 724                                   long_options, &longindex)) != -1) {
 725                 switch (opt) {
 726                 case 'd':
 727                         if (strlen(optarg) >= IF_NAMESIZE) {
 728                                 fprintf(stderr, "ERR: --dev name too long\n");
 729                                 goto error;
 730                         }
 731                         ifname = (char *)&ifname_buf;
 732                         strncpy(ifname, optarg, IF_NAMESIZE);
 733                         ifindex = if_nametoindex(ifname);
 734                         if (ifindex == 0) {
 735                                 fprintf(stderr,
 736                                         "ERR: --dev name unknown err(%d):%s\n",
 737                                         errno, strerror(errno));
 738                                 goto error;
 739                         }
 740                         break;
 741                 case 's':
 742                         interval = atoi(optarg);
 743                         break;
 744                 case 'S':
 745                         xdp_flags |= XDP_FLAGS_SKB_MODE;
 746                         break;
 747                 case 'x':
 748                         stress_mode = true;
 749                         break;
 750                 case 'z':
 751                         use_separators = false;
 752                         break;
 753                 case 'p':
 754                         /* Selecting eBPF prog to load */
 755                         prog_name = optarg;
 756                         break;
 757                 case 'c':
 758                         /* Add multiple CPUs */
 759                         add_cpu = strtoul(optarg, NULL, 0);
 760                         if (add_cpu >= MAX_CPUS) {
 761                                 fprintf(stderr,
 762                                 "--cpu nr too large for cpumap err(%d):%s\n",
 763                                         errno, strerror(errno));
 764                                 goto error;
 765                         }
 766                         create_cpu_entry(add_cpu, qsize, added_cpus, true);
 767                         added_cpus++;
 768                         break;
 769                 case 'q':
 770                         qsize = atoi(optarg);
 771                         break;
 772                 case 'F':
 773                         xdp_flags &= ~XDP_FLAGS_UPDATE_IF_NOEXIST;
 774                         break;
 775                 case 'h':
 776                 error:
 777                 default:
 778                         usage(argv, obj);
 779                         return EXIT_FAIL_OPTION;
 780                 }
 781         }
 782
 783         if (!(xdp_flags & XDP_FLAGS_SKB_MODE))
 784                 xdp_flags |= XDP_FLAGS_DRV_MODE;
 785
 786         /* Required option */
 787         if (ifindex == -1) {
 788                 fprintf(stderr, "ERR: required option --dev missing\n");
 789                 usage(argv, obj);
 790                 return EXIT_FAIL_OPTION;
 791         }
 792         /* Required option */
 793         if (add_cpu == -1) {
 794                 fprintf(stderr, "ERR: required option --cpu missing\n");
 795                 fprintf(stderr, " Specify multiple --cpu option to add more\n");
 796                 usage(argv, obj);
 797                 return EXIT_FAIL_OPTION;
 798         }
 799
 800         /* Remove XDP program when program is interrupted or killed */
 801         signal(SIGINT, int_exit);
 802         signal(SIGTERM, int_exit);
 803
 804         prog = bpf_object__find_program_by_title(obj, prog_name);
 805         if (!prog) {
 806                 fprintf(stderr, "bpf_object__find_program_by_title failed\n");
 807                 return EXIT_FAIL;
 808         }
 809
 810         prog_fd = bpf_program__fd(prog);
 811         if (prog_fd < 0) {
 812                 fprintf(stderr, "bpf_program__fd failed\n");
 813                 return EXIT_FAIL;
 814         }
 815
 816         if (bpf_set_link_xdp_fd(ifindex, prog_fd, xdp_flags) < 0) {
 817                 fprintf(stderr, "link set xdp fd failed\n");
 818                 return EXIT_FAIL_XDP;
 819         }
 820
 821         err = bpf_obj_get_info_by_fd(prog_fd, &info, &info_len);
 822         if (err) {
 823                 printf("can't get prog info - %s\n", strerror(errno));
 824                 return err;
 825         }
 826         prog_id = info.id;
 827
 828         stats_poll(interval, use_separators, prog_name, stress_mode);
 829         return EXIT_OK;
 830 }