tools/lib/bpf/xsk.c

   1 // SPDX-License-Identifier: (LGPL-2.1 OR BSD-2-Clause)
   2
   3 /*
   4  * AF_XDP user-space access library.
   5  *
   6  * Copyright(c) 2018 - 2019 Intel Corporation.
   7  *
   8  * Author(s): Magnus Karlsson <magnus.karlsson@intel.com>
   9  */
  10
  11 #include <errno.h>
  12 #include <stdlib.h>
  13 #include <string.h>
  14 #include <unistd.h>
  15 #include <arpa/inet.h>
  16 #include <asm/barrier.h>
  17 #include <linux/compiler.h>
  18 #include <linux/ethtool.h>
  19 #include <linux/filter.h>
  20 #include <linux/if_ether.h>
  21 #include <linux/if_packet.h>
  22 #include <linux/if_xdp.h>
  23 #include <linux/sockios.h>
  24 #include <net/if.h>
  25 #include <sys/ioctl.h>
  26 #include <sys/mman.h>
  27 #include <sys/socket.h>
  28 #include <sys/types.h>
  29
  30 #include "bpf.h"
  31 #include "libbpf.h"
  32 #include "libbpf_internal.h"
  33 #include "xsk.h"
  34
  35 #ifndef SOL_XDP
  36  #define SOL_XDP 283
  37 #endif
  38
  39 #ifndef AF_XDP
  40  #define AF_XDP 44
  41 #endif
  42
  43 #ifndef PF_XDP
  44  #define PF_XDP AF_XDP
  45 #endif
  46
  47 struct xsk_umem {
  48         struct xsk_ring_prod *fill;
  49         struct xsk_ring_cons *comp;
  50         char *umem_area;
  51         struct xsk_umem_config config;
  52         int fd;
  53         int refcount;
  54 };
  55
  56 struct xsk_socket {
  57         struct xsk_ring_cons *rx;
  58         struct xsk_ring_prod *tx;
  59         __u64 outstanding_tx;
  60         struct xsk_umem *umem;
  61         struct xsk_socket_config config;
  62         int fd;
  63         int ifindex;
  64         int prog_fd;
  65         int xsks_map_fd;
  66         __u32 queue_id;
  67         char ifname[IFNAMSIZ];
  68 };
  69
  70 struct xsk_nl_info {
  71         bool xdp_prog_attached;
  72         int ifindex;
  73         int fd;
  74 };
  75
  76 /* Up until and including Linux 5.3 */
  77 struct xdp_ring_offset_v1 {
  78         __u64 producer;
  79         __u64 consumer;
  80         __u64 desc;
  81 };
  82
  83 /* Up until and including Linux 5.3 */
  84 struct xdp_mmap_offsets_v1 {
  85         struct xdp_ring_offset_v1 rx;
  86         struct xdp_ring_offset_v1 tx;
  87         struct xdp_ring_offset_v1 fr;
  88         struct xdp_ring_offset_v1 cr;
  89 };
  90
  91 int xsk_umem__fd(const struct xsk_umem *umem)
  92 {
  93         return umem ? umem->fd : -EINVAL;
  94 }
  95
  96 int xsk_socket__fd(const struct xsk_socket *xsk)
  97 {
  98         return xsk ? xsk->fd : -EINVAL;
  99 }
 100
 101 static bool xsk_page_aligned(void *buffer)
 102 {
 103         unsigned long addr = (unsigned long)buffer;
 104
 105         return !(addr & (getpagesize() - 1));
 106 }
 107
 108 static void xsk_set_umem_config(struct xsk_umem_config *cfg,
 109                                 const struct xsk_umem_config *usr_cfg)
 110 {
 111         if (!usr_cfg) {
 112                 cfg->fill_size = XSK_RING_PROD__DEFAULT_NUM_DESCS;
 113                 cfg->comp_size = XSK_RING_CONS__DEFAULT_NUM_DESCS;
 114                 cfg->frame_size = XSK_UMEM__DEFAULT_FRAME_SIZE;
 115                 cfg->frame_headroom = XSK_UMEM__DEFAULT_FRAME_HEADROOM;
 116                 cfg->flags = XSK_UMEM__DEFAULT_FLAGS;
 117                 return;
 118         }
 119
 120         cfg->fill_size = usr_cfg->fill_size;
 121         cfg->comp_size = usr_cfg->comp_size;
 122         cfg->frame_size = usr_cfg->frame_size;
 123         cfg->frame_headroom = usr_cfg->frame_headroom;
 124         cfg->flags = usr_cfg->flags;
 125 }
 126
 127 static int xsk_set_xdp_socket_config(struct xsk_socket_config *cfg,
 128                                      const struct xsk_socket_config *usr_cfg)
 129 {
 130         if (!usr_cfg) {
 131                 cfg->rx_size = XSK_RING_CONS__DEFAULT_NUM_DESCS;
 132                 cfg->tx_size = XSK_RING_PROD__DEFAULT_NUM_DESCS;
 133                 cfg->libbpf_flags = 0;
 134                 cfg->xdp_flags = 0;
 135                 cfg->bind_flags = 0;
 136                 return 0;
 137         }
 138
 139         if (usr_cfg->libbpf_flags & ~XSK_LIBBPF_FLAGS__INHIBIT_PROG_LOAD)
 140                 return -EINVAL;
 141
 142         cfg->rx_size = usr_cfg->rx_size;
 143         cfg->tx_size = usr_cfg->tx_size;
 144         cfg->libbpf_flags = usr_cfg->libbpf_flags;
 145         cfg->xdp_flags = usr_cfg->xdp_flags;
 146         cfg->bind_flags = usr_cfg->bind_flags;
 147
 148         return 0;
 149 }
 150
 151 static void xsk_mmap_offsets_v1(struct xdp_mmap_offsets *off)
 152 {
 153         struct xdp_mmap_offsets_v1 off_v1;
 154
 155         /* getsockopt on a kernel <= 5.3 has no flags fields.
 156          * Copy over the offsets to the correct places in the >=5.4 format
 157          * and put the flags where they would have been on that kernel.
 158          */
 159         memcpy(&off_v1, off, sizeof(off_v1));
 160
 161         off->rx.producer = off_v1.rx.producer;
 162         off->rx.consumer = off_v1.rx.consumer;
 163         off->rx.desc = off_v1.rx.desc;
 164         off->rx.flags = off_v1.rx.consumer + sizeof(__u32);
 165
 166         off->tx.producer = off_v1.tx.producer;
 167         off->tx.consumer = off_v1.tx.consumer;
 168         off->tx.desc = off_v1.tx.desc;
 169         off->tx.flags = off_v1.tx.consumer + sizeof(__u32);
 170
 171         off->fr.producer = off_v1.fr.producer;
 172         off->fr.consumer = off_v1.fr.consumer;
 173         off->fr.desc = off_v1.fr.desc;
 174         off->fr.flags = off_v1.fr.consumer + sizeof(__u32);
 175
 176         off->cr.producer = off_v1.cr.producer;
 177         off->cr.consumer = off_v1.cr.consumer;
 178         off->cr.desc = off_v1.cr.desc;
 179         off->cr.flags = off_v1.cr.consumer + sizeof(__u32);
 180 }
 181
 182 static int xsk_get_mmap_offsets(int fd, struct xdp_mmap_offsets *off)
 183 {
 184         socklen_t optlen;
 185         int err;
 186
 187         optlen = sizeof(*off);
 188         err = getsockopt(fd, SOL_XDP, XDP_MMAP_OFFSETS, off, &optlen);
 189         if (err)
 190                 return err;
 191
 192         if (optlen == sizeof(*off))
 193                 return 0;
 194
 195         if (optlen == sizeof(struct xdp_mmap_offsets_v1)) {
 196                 xsk_mmap_offsets_v1(off);
 197                 return 0;
 198         }
 199
 200         return -EINVAL;
 201 }
 202
 203 int xsk_umem__create_v0_0_4(struct xsk_umem **umem_ptr, void *umem_area,
 204                             __u64 size, struct xsk_ring_prod *fill,
 205                             struct xsk_ring_cons *comp,
 206                             const struct xsk_umem_config *usr_config)
 207 {
 208         struct xdp_mmap_offsets off;
 209         struct xdp_umem_reg mr;
 210         struct xsk_umem *umem;
 211         void *map;
 212         int err;
 213
 214         if (!umem_area || !umem_ptr || !fill || !comp)
 215                 return -EFAULT;
 216         if (!size && !xsk_page_aligned(umem_area))
 217                 return -EINVAL;
 218
 219         umem = calloc(1, sizeof(*umem));
 220         if (!umem)
 221                 return -ENOMEM;
 222
 223         umem->fd = socket(AF_XDP, SOCK_RAW, 0);
 224         if (umem->fd < 0) {
 225                 err = -errno;
 226                 goto out_umem_alloc;
 227         }
 228
 229         umem->umem_area = umem_area;
 230         xsk_set_umem_config(&umem->config, usr_config);
 231
 232         memset(&mr, 0, sizeof(mr));
 233         mr.addr = (uintptr_t)umem_area;
 234         mr.len = size;
 235         mr.chunk_size = umem->config.frame_size;
 236         mr.headroom = umem->config.frame_headroom;
 237         mr.flags = umem->config.flags;
 238
 239         err = setsockopt(umem->fd, SOL_XDP, XDP_UMEM_REG, &mr, sizeof(mr));
 240         if (err) {
 241                 err = -errno;
 242                 goto out_socket;
 243         }
 244         err = setsockopt(umem->fd, SOL_XDP, XDP_UMEM_FILL_RING,
 245                          &umem->config.fill_size,
 246                          sizeof(umem->config.fill_size));
 247         if (err) {
 248                 err = -errno;
 249                 goto out_socket;
 250         }
 251         err = setsockopt(umem->fd, SOL_XDP, XDP_UMEM_COMPLETION_RING,
 252                          &umem->config.comp_size,
 253                          sizeof(umem->config.comp_size));
 254         if (err) {
 255                 err = -errno;
 256                 goto out_socket;
 257         }
 258
 259         err = xsk_get_mmap_offsets(umem->fd, &off);
 260         if (err) {
 261                 err = -errno;
 262                 goto out_socket;
 263         }
 264
 265         map = mmap(NULL, off.fr.desc + umem->config.fill_size * sizeof(__u64),
 266                    PROT_READ | PROT_WRITE, MAP_SHARED | MAP_POPULATE, umem->fd,
 267                    XDP_UMEM_PGOFF_FILL_RING);
 268         if (map == MAP_FAILED) {
 269                 err = -errno;
 270                 goto out_socket;
 271         }
 272
 273         umem->fill = fill;
 274         fill->mask = umem->config.fill_size - 1;
 275         fill->size = umem->config.fill_size;
 276         fill->producer = map + off.fr.producer;
 277         fill->consumer = map + off.fr.consumer;
 278         fill->flags = map + off.fr.flags;
 279         fill->ring = map + off.fr.desc;
 280         fill->cached_cons = umem->config.fill_size;
 281
 282         map = mmap(NULL, off.cr.desc + umem->config.comp_size * sizeof(__u64),
 283                    PROT_READ | PROT_WRITE, MAP_SHARED | MAP_POPULATE, umem->fd,
 284                    XDP_UMEM_PGOFF_COMPLETION_RING);
 285         if (map == MAP_FAILED) {
 286                 err = -errno;
 287                 goto out_mmap;
 288         }
 289
 290         umem->comp = comp;
 291         comp->mask = umem->config.comp_size - 1;
 292         comp->size = umem->config.comp_size;
 293         comp->producer = map + off.cr.producer;
 294         comp->consumer = map + off.cr.consumer;
 295         comp->flags = map + off.cr.flags;
 296         comp->ring = map + off.cr.desc;
 297
 298         *umem_ptr = umem;
 299         return 0;
 300
 301 out_mmap:
 302         munmap(map, off.fr.desc + umem->config.fill_size * sizeof(__u64));
 303 out_socket:
 304         close(umem->fd);
 305 out_umem_alloc:
 306         free(umem);
 307         return err;
 308 }
 309
 310 struct xsk_umem_config_v1 {
 311         __u32 fill_size;
 312         __u32 comp_size;
 313         __u32 frame_size;
 314         __u32 frame_headroom;
 315 };
 316
 317 int xsk_umem__create_v0_0_2(struct xsk_umem **umem_ptr, void *umem_area,
 318                             __u64 size, struct xsk_ring_prod *fill,
 319                             struct xsk_ring_cons *comp,
 320                             const struct xsk_umem_config *usr_config)
 321 {
 322         struct xsk_umem_config config;
 323
 324         memcpy(&config, usr_config, sizeof(struct xsk_umem_config_v1));
 325         config.flags = 0;
 326
 327         return xsk_umem__create_v0_0_4(umem_ptr, umem_area, size, fill, comp,
 328                                         &config);
 329 }
 330 COMPAT_VERSION(xsk_umem__create_v0_0_2, xsk_umem__create, LIBBPF_0.0.2)
 331 DEFAULT_VERSION(xsk_umem__create_v0_0_4, xsk_umem__create, LIBBPF_0.0.4)
 332
 333 static int xsk_load_xdp_prog(struct xsk_socket *xsk)
 334 {
 335         static const int log_buf_size = 16 * 1024;
 336         char log_buf[log_buf_size];
 337         int err, prog_fd;
 338
 339         /* This is the C-program:
 340          * SEC("xdp_sock") int xdp_sock_prog(struct xdp_md *ctx)
 341          * {
 342          *     int ret, index = ctx->rx_queue_index;
 343          *
 344          *     // A set entry here means that the correspnding queue_id
 345          *     // has an active AF_XDP socket bound to it.
 346          *     ret = bpf_redirect_map(&xsks_map, index, XDP_PASS);
 347          *     if (ret > 0)
 348          *         return ret;
 349          *
 350          *     // Fallback for pre-5.3 kernels, not supporting default
 351          *     // action in the flags parameter.
 352          *     if (bpf_map_lookup_elem(&xsks_map, &index))
 353          *         return bpf_redirect_map(&xsks_map, index, 0);
 354          *     return XDP_PASS;
 355          * }
 356          */
 357         struct bpf_insn prog[] = {
 358                 /* r2 = *(u32 *)(r1 + 16) */
 359                 BPF_LDX_MEM(BPF_W, BPF_REG_2, BPF_REG_1, 16),
 360                 /* *(u32 *)(r10 - 4) = r2 */
 361                 BPF_STX_MEM(BPF_W, BPF_REG_10, BPF_REG_2, -4),
 362                 /* r1 = xskmap[] */
 363                 BPF_LD_MAP_FD(BPF_REG_1, xsk->xsks_map_fd),
 364                 /* r3 = XDP_PASS */
 365                 BPF_MOV64_IMM(BPF_REG_3, 2),
 366                 /* call bpf_redirect_map */
 367                 BPF_EMIT_CALL(BPF_FUNC_redirect_map),
 368                 /* if w0 != 0 goto pc+13 */
 369                 BPF_JMP32_IMM(BPF_JSGT, BPF_REG_0, 0, 13),
 370                 /* r2 = r10 */
 371                 BPF_MOV64_REG(BPF_REG_2, BPF_REG_10),
 372                 /* r2 += -4 */
 373                 BPF_ALU64_IMM(BPF_ADD, BPF_REG_2, -4),
 374                 /* r1 = xskmap[] */
 375                 BPF_LD_MAP_FD(BPF_REG_1, xsk->xsks_map_fd),
 376                 /* call bpf_map_lookup_elem */
 377                 BPF_EMIT_CALL(BPF_FUNC_map_lookup_elem),
 378                 /* r1 = r0 */
 379                 BPF_MOV64_REG(BPF_REG_1, BPF_REG_0),
 380                 /* r0 = XDP_PASS */
 381                 BPF_MOV64_IMM(BPF_REG_0, 2),
 382                 /* if r1 == 0 goto pc+5 */
 383                 BPF_JMP_IMM(BPF_JEQ, BPF_REG_1, 0, 5),
 384                 /* r2 = *(u32 *)(r10 - 4) */
 385                 BPF_LDX_MEM(BPF_W, BPF_REG_2, BPF_REG_10, -4),
 386                 /* r1 = xskmap[] */
 387                 BPF_LD_MAP_FD(BPF_REG_1, xsk->xsks_map_fd),
 388                 /* r3 = 0 */
 389                 BPF_MOV64_IMM(BPF_REG_3, 0),
 390                 /* call bpf_redirect_map */
 391                 BPF_EMIT_CALL(BPF_FUNC_redirect_map),
 392                 /* The jumps are to this instruction */
 393                 BPF_EXIT_INSN(),
 394         };
 395         size_t insns_cnt = sizeof(prog) / sizeof(struct bpf_insn);
 396
 397         prog_fd = bpf_load_program(BPF_PROG_TYPE_XDP, prog, insns_cnt,
 398                                    "LGPL-2.1 or BSD-2-Clause", 0, log_buf,
 399                                    log_buf_size);
 400         if (prog_fd < 0) {
 401                 pr_warn("BPF log buffer:\n%s", log_buf);
 402                 return prog_fd;
 403         }
 404
 405         err = bpf_set_link_xdp_fd(xsk->ifindex, prog_fd, xsk->config.xdp_flags);
 406         if (err) {
 407                 close(prog_fd);
 408                 return err;
 409         }
 410
 411         xsk->prog_fd = prog_fd;
 412         return 0;
 413 }
 414
 415 static int xsk_get_max_queues(struct xsk_socket *xsk)
 416 {
 417         struct ethtool_channels channels = { .cmd = ETHTOOL_GCHANNELS };
 418         struct ifreq ifr = {};
 419         int fd, err, ret;
 420
 421         fd = socket(AF_INET, SOCK_DGRAM, 0);
 422         if (fd < 0)
 423                 return -errno;
 424
 425         ifr.ifr_data = (void *)&channels;
 426         memcpy(ifr.ifr_name, xsk->ifname, IFNAMSIZ - 1);
 427         ifr.ifr_name[IFNAMSIZ - 1] = '\0';
 428         err = ioctl(fd, SIOCETHTOOL, &ifr);
 429         if (err && errno != EOPNOTSUPP) {
 430                 ret = -errno;
 431                 goto out;
 432         }
 433
 434         if (err) {
 435                 /* If the device says it has no channels, then all traffic
 436                  * is sent to a single stream, so max queues = 1.
 437                  */
 438                 ret = 1;
 439         } else {
 440                 /* Take the max of rx, tx, combined. Drivers return
 441                  * the number of channels in different ways.
 442                  */
 443                 ret = max(channels.max_rx, channels.max_tx);
 444                 ret = max(ret, (int)channels.max_combined);
 445         }
 446
 447 out:
 448         close(fd);
 449         return ret;
 450 }
 451
 452 static int xsk_create_bpf_maps(struct xsk_socket *xsk)
 453 {
 454         int max_queues;
 455         int fd;
 456
 457         max_queues = xsk_get_max_queues(xsk);
 458         if (max_queues < 0)
 459                 return max_queues;
 460
 461         fd = bpf_create_map_name(BPF_MAP_TYPE_XSKMAP, "xsks_map",
 462                                  sizeof(int), sizeof(int), max_queues, 0);
 463         if (fd < 0)
 464                 return fd;
 465
 466         xsk->xsks_map_fd = fd;
 467
 468         return 0;
 469 }
 470
 471 static void xsk_delete_bpf_maps(struct xsk_socket *xsk)
 472 {
 473         bpf_map_delete_elem(xsk->xsks_map_fd, &xsk->queue_id);
 474         close(xsk->xsks_map_fd);
 475 }
 476
 477 static int xsk_lookup_bpf_maps(struct xsk_socket *xsk)
 478 {
 479         __u32 i, *map_ids, num_maps, prog_len = sizeof(struct bpf_prog_info);
 480         __u32 map_len = sizeof(struct bpf_map_info);
 481         struct bpf_prog_info prog_info = {};
 482         struct bpf_map_info map_info;
 483         int fd, err;
 484
 485         err = bpf_obj_get_info_by_fd(xsk->prog_fd, &prog_info, &prog_len);
 486         if (err)
 487                 return err;
 488
 489         num_maps = prog_info.nr_map_ids;
 490
 491         map_ids = calloc(prog_info.nr_map_ids, sizeof(*map_ids));
 492         if (!map_ids)
 493                 return -ENOMEM;
 494
 495         memset(&prog_info, 0, prog_len);
 496         prog_info.nr_map_ids = num_maps;
 497         prog_info.map_ids = (__u64)(unsigned long)map_ids;
 498
 499         err = bpf_obj_get_info_by_fd(xsk->prog_fd, &prog_info, &prog_len);
 500         if (err)
 501                 goto out_map_ids;
 502
 503         xsk->xsks_map_fd = -1;
 504
 505         for (i = 0; i < prog_info.nr_map_ids; i++) {
 506                 fd = bpf_map_get_fd_by_id(map_ids[i]);
 507                 if (fd < 0)
 508                         continue;
 509
 510                 err = bpf_obj_get_info_by_fd(fd, &map_info, &map_len);
 511                 if (err) {
 512                         close(fd);
 513                         continue;
 514                 }
 515
 516                 if (!strcmp(map_info.name, "xsks_map")) {
 517                         xsk->xsks_map_fd = fd;
 518                         continue;
 519                 }
 520
 521                 close(fd);
 522         }
 523
 524         err = 0;
 525         if (xsk->xsks_map_fd == -1)
 526                 err = -ENOENT;
 527
 528 out_map_ids:
 529         free(map_ids);
 530         return err;
 531 }
 532
 533 static int xsk_set_bpf_maps(struct xsk_socket *xsk)
 534 {
 535         return bpf_map_update_elem(xsk->xsks_map_fd, &xsk->queue_id,
 536                                    &xsk->fd, 0);
 537 }
 538
 539 static int xsk_setup_xdp_prog(struct xsk_socket *xsk)
 540 {
 541         __u32 prog_id = 0;
 542         int err;
 543
 544         err = bpf_get_link_xdp_id(xsk->ifindex, &prog_id,
 545                                   xsk->config.xdp_flags);
 546         if (err)
 547                 return err;
 548
 549         if (!prog_id) {
 550                 err = xsk_create_bpf_maps(xsk);
 551                 if (err)
 552                         return err;
 553
 554                 err = xsk_load_xdp_prog(xsk);
 555                 if (err) {
 556                         xsk_delete_bpf_maps(xsk);
 557                         return err;
 558                 }
 559         } else {
 560                 xsk->prog_fd = bpf_prog_get_fd_by_id(prog_id);
 561                 if (xsk->prog_fd < 0)
 562                         return -errno;
 563                 err = xsk_lookup_bpf_maps(xsk);
 564                 if (err) {
 565                         close(xsk->prog_fd);
 566                         return err;
 567                 }
 568         }
 569
 570         if (xsk->rx)
 571                 err = xsk_set_bpf_maps(xsk);
 572         if (err) {
 573                 xsk_delete_bpf_maps(xsk);
 574                 close(xsk->prog_fd);
 575                 return err;
 576         }
 577
 578         return 0;
 579 }
 580
 581 int xsk_socket__create(struct xsk_socket **xsk_ptr, const char *ifname,
 582                        __u32 queue_id, struct xsk_umem *umem,
 583                        struct xsk_ring_cons *rx, struct xsk_ring_prod *tx,
 584                        const struct xsk_socket_config *usr_config)
 585 {
 586         void *rx_map = NULL, *tx_map = NULL;
 587         struct sockaddr_xdp sxdp = {};
 588         struct xdp_mmap_offsets off;
 589         struct xsk_socket *xsk;
 590         int err;
 591
 592         if (!umem || !xsk_ptr || !(rx || tx))
 593                 return -EFAULT;
 594
 595         xsk = calloc(1, sizeof(*xsk));
 596         if (!xsk)
 597                 return -ENOMEM;
 598
 599         err = xsk_set_xdp_socket_config(&xsk->config, usr_config);
 600         if (err)
 601                 goto out_xsk_alloc;
 602
 603         if (umem->refcount &&
 604             !(xsk->config.libbpf_flags & XSK_LIBBPF_FLAGS__INHIBIT_PROG_LOAD)) {
 605                 pr_warn("Error: shared umems not supported by libbpf supplied XDP program.\n");
 606                 err = -EBUSY;
 607                 goto out_xsk_alloc;
 608         }
 609
 610         if (umem->refcount++ > 0) {
 611                 xsk->fd = socket(AF_XDP, SOCK_RAW, 0);
 612                 if (xsk->fd < 0) {
 613                         err = -errno;
 614                         goto out_xsk_alloc;
 615                 }
 616         } else {
 617                 xsk->fd = umem->fd;
 618         }
 619
 620         xsk->outstanding_tx = 0;
 621         xsk->queue_id = queue_id;
 622         xsk->umem = umem;
 623         xsk->ifindex = if_nametoindex(ifname);
 624         if (!xsk->ifindex) {
 625                 err = -errno;
 626                 goto out_socket;
 627         }
 628         memcpy(xsk->ifname, ifname, IFNAMSIZ - 1);
 629         xsk->ifname[IFNAMSIZ - 1] = '\0';
 630
 631         if (rx) {
 632                 err = setsockopt(xsk->fd, SOL_XDP, XDP_RX_RING,
 633                                  &xsk->config.rx_size,
 634                                  sizeof(xsk->config.rx_size));
 635                 if (err) {
 636                         err = -errno;
 637                         goto out_socket;
 638                 }
 639         }
 640         if (tx) {
 641                 err = setsockopt(xsk->fd, SOL_XDP, XDP_TX_RING,
 642                                  &xsk->config.tx_size,
 643                                  sizeof(xsk->config.tx_size));
 644                 if (err) {
 645                         err = -errno;
 646                         goto out_socket;
 647                 }
 648         }
 649
 650         err = xsk_get_mmap_offsets(xsk->fd, &off);
 651         if (err) {
 652                 err = -errno;
 653                 goto out_socket;
 654         }
 655
 656         if (rx) {
 657                 rx_map = mmap(NULL, off.rx.desc +
 658                               xsk->config.rx_size * sizeof(struct xdp_desc),
 659                               PROT_READ | PROT_WRITE, MAP_SHARED | MAP_POPULATE,
 660                               xsk->fd, XDP_PGOFF_RX_RING);
 661                 if (rx_map == MAP_FAILED) {
 662                         err = -errno;
 663                         goto out_socket;
 664                 }
 665
 666                 rx->mask = xsk->config.rx_size - 1;
 667                 rx->size = xsk->config.rx_size;
 668                 rx->producer = rx_map + off.rx.producer;
 669                 rx->consumer = rx_map + off.rx.consumer;
 670                 rx->flags = rx_map + off.rx.flags;
 671                 rx->ring = rx_map + off.rx.desc;
 672         }
 673         xsk->rx = rx;
 674
 675         if (tx) {
 676                 tx_map = mmap(NULL, off.tx.desc +
 677                               xsk->config.tx_size * sizeof(struct xdp_desc),
 678                               PROT_READ | PROT_WRITE, MAP_SHARED | MAP_POPULATE,
 679                               xsk->fd, XDP_PGOFF_TX_RING);
 680                 if (tx_map == MAP_FAILED) {
 681                         err = -errno;
 682                         goto out_mmap_rx;
 683                 }
 684
 685                 tx->mask = xsk->config.tx_size - 1;
 686                 tx->size = xsk->config.tx_size;
 687                 tx->producer = tx_map + off.tx.producer;
 688                 tx->consumer = tx_map + off.tx.consumer;
 689                 tx->flags = tx_map + off.tx.flags;
 690                 tx->ring = tx_map + off.tx.desc;
 691                 tx->cached_cons = xsk->config.tx_size;
 692         }
 693         xsk->tx = tx;
 694
 695         sxdp.sxdp_family = PF_XDP;
 696         sxdp.sxdp_ifindex = xsk->ifindex;
 697         sxdp.sxdp_queue_id = xsk->queue_id;
 698         if (umem->refcount > 1) {
 699                 sxdp.sxdp_flags = XDP_SHARED_UMEM;
 700                 sxdp.sxdp_shared_umem_fd = umem->fd;
 701         } else {
 702                 sxdp.sxdp_flags = xsk->config.bind_flags;
 703         }
 704
 705         err = bind(xsk->fd, (struct sockaddr *)&sxdp, sizeof(sxdp));
 706         if (err) {
 707                 err = -errno;
 708                 goto out_mmap_tx;
 709         }
 710
 711         xsk->prog_fd = -1;
 712
 713         if (!(xsk->config.libbpf_flags & XSK_LIBBPF_FLAGS__INHIBIT_PROG_LOAD)) {
 714                 err = xsk_setup_xdp_prog(xsk);
 715                 if (err)
 716                         goto out_mmap_tx;
 717         }
 718
 719         *xsk_ptr = xsk;
 720         return 0;
 721
 722 out_mmap_tx:
 723         if (tx)
 724                 munmap(tx_map, off.tx.desc +
 725                        xsk->config.tx_size * sizeof(struct xdp_desc));
 726 out_mmap_rx:
 727         if (rx)
 728                 munmap(rx_map, off.rx.desc +
 729                        xsk->config.rx_size * sizeof(struct xdp_desc));
 730 out_socket:
 731         if (--umem->refcount)
 732                 close(xsk->fd);
 733 out_xsk_alloc:
 734         free(xsk);
 735         return err;
 736 }
 737
 738 int xsk_umem__delete(struct xsk_umem *umem)
 739 {
 740         struct xdp_mmap_offsets off;
 741         int err;
 742
 743         if (!umem)
 744                 return 0;
 745
 746         if (umem->refcount)
 747                 return -EBUSY;
 748
 749         err = xsk_get_mmap_offsets(umem->fd, &off);
 750         if (!err) {
 751                 munmap(umem->fill->ring - off.fr.desc,
 752                        off.fr.desc + umem->config.fill_size * sizeof(__u64));
 753                 munmap(umem->comp->ring - off.cr.desc,
 754                        off.cr.desc + umem->config.comp_size * sizeof(__u64));
 755         }
 756
 757         close(umem->fd);
 758         free(umem);
 759
 760         return 0;
 761 }
 762
 763 void xsk_socket__delete(struct xsk_socket *xsk)
 764 {
 765         size_t desc_sz = sizeof(struct xdp_desc);
 766         struct xdp_mmap_offsets off;
 767         int err;
 768
 769         if (!xsk)
 770                 return;
 771
 772         if (xsk->prog_fd != -1) {
 773                 xsk_delete_bpf_maps(xsk);
 774                 close(xsk->prog_fd);
 775         }
 776
 777         err = xsk_get_mmap_offsets(xsk->fd, &off);
 778         if (!err) {
 779                 if (xsk->rx) {
 780                         munmap(xsk->rx->ring - off.rx.desc,
 781                                off.rx.desc + xsk->config.rx_size * desc_sz);
 782                 }
 783                 if (xsk->tx) {
 784                         munmap(xsk->tx->ring - off.tx.desc,
 785                                off.tx.desc + xsk->config.tx_size * desc_sz);
 786                 }
 787
 788         }
 789
 790         xsk->umem->refcount--;
 791         /* Do not close an fd that also has an associated umem connected
 792          * to it.
 793          */
 794         if (xsk->fd != xsk->umem->fd)
 795                 close(xsk->fd);
 796         free(xsk);
 797 }