linux/tools/lib/bpf/xsk.c
<<
>>
Prefs
   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/* make sure libbpf doesn't use kernel-only integer typedefs */
  36#pragma GCC poison u8 u16 u32 u64 s8 s16 s32 s64
  37
  38#ifndef SOL_XDP
  39 #define SOL_XDP 283
  40#endif
  41
  42#ifndef AF_XDP
  43 #define AF_XDP 44
  44#endif
  45
  46#ifndef PF_XDP
  47 #define PF_XDP AF_XDP
  48#endif
  49
  50struct xsk_umem {
  51        struct xsk_ring_prod *fill;
  52        struct xsk_ring_cons *comp;
  53        char *umem_area;
  54        struct xsk_umem_config config;
  55        int fd;
  56        int refcount;
  57};
  58
  59struct xsk_socket {
  60        struct xsk_ring_cons *rx;
  61        struct xsk_ring_prod *tx;
  62        __u64 outstanding_tx;
  63        struct xsk_umem *umem;
  64        struct xsk_socket_config config;
  65        int fd;
  66        int ifindex;
  67        int prog_fd;
  68        int xsks_map_fd;
  69        __u32 queue_id;
  70        char ifname[IFNAMSIZ];
  71};
  72
  73struct xsk_nl_info {
  74        bool xdp_prog_attached;
  75        int ifindex;
  76        int fd;
  77};
  78
  79/* Up until and including Linux 5.3 */
  80struct xdp_ring_offset_v1 {
  81        __u64 producer;
  82        __u64 consumer;
  83        __u64 desc;
  84};
  85
  86/* Up until and including Linux 5.3 */
  87struct xdp_mmap_offsets_v1 {
  88        struct xdp_ring_offset_v1 rx;
  89        struct xdp_ring_offset_v1 tx;
  90        struct xdp_ring_offset_v1 fr;
  91        struct xdp_ring_offset_v1 cr;
  92};
  93
  94int xsk_umem__fd(const struct xsk_umem *umem)
  95{
  96        return umem ? umem->fd : -EINVAL;
  97}
  98
  99int xsk_socket__fd(const struct xsk_socket *xsk)
 100{
 101        return xsk ? xsk->fd : -EINVAL;
 102}
 103
 104static bool xsk_page_aligned(void *buffer)
 105{
 106        unsigned long addr = (unsigned long)buffer;
 107
 108        return !(addr & (getpagesize() - 1));
 109}
 110
 111static void xsk_set_umem_config(struct xsk_umem_config *cfg,
 112                                const struct xsk_umem_config *usr_cfg)
 113{
 114        if (!usr_cfg) {
 115                cfg->fill_size = XSK_RING_PROD__DEFAULT_NUM_DESCS;
 116                cfg->comp_size = XSK_RING_CONS__DEFAULT_NUM_DESCS;
 117                cfg->frame_size = XSK_UMEM__DEFAULT_FRAME_SIZE;
 118                cfg->frame_headroom = XSK_UMEM__DEFAULT_FRAME_HEADROOM;
 119                cfg->flags = XSK_UMEM__DEFAULT_FLAGS;
 120                return;
 121        }
 122
 123        cfg->fill_size = usr_cfg->fill_size;
 124        cfg->comp_size = usr_cfg->comp_size;
 125        cfg->frame_size = usr_cfg->frame_size;
 126        cfg->frame_headroom = usr_cfg->frame_headroom;
 127        cfg->flags = usr_cfg->flags;
 128}
 129
 130static int xsk_set_xdp_socket_config(struct xsk_socket_config *cfg,
 131                                     const struct xsk_socket_config *usr_cfg)
 132{
 133        if (!usr_cfg) {
 134                cfg->rx_size = XSK_RING_CONS__DEFAULT_NUM_DESCS;
 135                cfg->tx_size = XSK_RING_PROD__DEFAULT_NUM_DESCS;
 136                cfg->libbpf_flags = 0;
 137                cfg->xdp_flags = 0;
 138                cfg->bind_flags = 0;
 139                return 0;
 140        }
 141
 142        if (usr_cfg->libbpf_flags & ~XSK_LIBBPF_FLAGS__INHIBIT_PROG_LOAD)
 143                return -EINVAL;
 144
 145        cfg->rx_size = usr_cfg->rx_size;
 146        cfg->tx_size = usr_cfg->tx_size;
 147        cfg->libbpf_flags = usr_cfg->libbpf_flags;
 148        cfg->xdp_flags = usr_cfg->xdp_flags;
 149        cfg->bind_flags = usr_cfg->bind_flags;
 150
 151        return 0;
 152}
 153
 154static void xsk_mmap_offsets_v1(struct xdp_mmap_offsets *off)
 155{
 156        struct xdp_mmap_offsets_v1 off_v1;
 157
 158        /* getsockopt on a kernel <= 5.3 has no flags fields.
 159         * Copy over the offsets to the correct places in the >=5.4 format
 160         * and put the flags where they would have been on that kernel.
 161         */
 162        memcpy(&off_v1, off, sizeof(off_v1));
 163
 164        off->rx.producer = off_v1.rx.producer;
 165        off->rx.consumer = off_v1.rx.consumer;
 166        off->rx.desc = off_v1.rx.desc;
 167        off->rx.flags = off_v1.rx.consumer + sizeof(__u32);
 168
 169        off->tx.producer = off_v1.tx.producer;
 170        off->tx.consumer = off_v1.tx.consumer;
 171        off->tx.desc = off_v1.tx.desc;
 172        off->tx.flags = off_v1.tx.consumer + sizeof(__u32);
 173
 174        off->fr.producer = off_v1.fr.producer;
 175        off->fr.consumer = off_v1.fr.consumer;
 176        off->fr.desc = off_v1.fr.desc;
 177        off->fr.flags = off_v1.fr.consumer + sizeof(__u32);
 178
 179        off->cr.producer = off_v1.cr.producer;
 180        off->cr.consumer = off_v1.cr.consumer;
 181        off->cr.desc = off_v1.cr.desc;
 182        off->cr.flags = off_v1.cr.consumer + sizeof(__u32);
 183}
 184
 185static int xsk_get_mmap_offsets(int fd, struct xdp_mmap_offsets *off)
 186{
 187        socklen_t optlen;
 188        int err;
 189
 190        optlen = sizeof(*off);
 191        err = getsockopt(fd, SOL_XDP, XDP_MMAP_OFFSETS, off, &optlen);
 192        if (err)
 193                return err;
 194
 195        if (optlen == sizeof(*off))
 196                return 0;
 197
 198        if (optlen == sizeof(struct xdp_mmap_offsets_v1)) {
 199                xsk_mmap_offsets_v1(off);
 200                return 0;
 201        }
 202
 203        return -EINVAL;
 204}
 205
 206int xsk_umem__create_v0_0_4(struct xsk_umem **umem_ptr, void *umem_area,
 207                            __u64 size, struct xsk_ring_prod *fill,
 208                            struct xsk_ring_cons *comp,
 209                            const struct xsk_umem_config *usr_config)
 210{
 211        struct xdp_mmap_offsets off;
 212        struct xdp_umem_reg mr;
 213        struct xsk_umem *umem;
 214        void *map;
 215        int err;
 216
 217        if (!umem_area || !umem_ptr || !fill || !comp)
 218                return -EFAULT;
 219        if (!size && !xsk_page_aligned(umem_area))
 220                return -EINVAL;
 221
 222        umem = calloc(1, sizeof(*umem));
 223        if (!umem)
 224                return -ENOMEM;
 225
 226        umem->fd = socket(AF_XDP, SOCK_RAW, 0);
 227        if (umem->fd < 0) {
 228                err = -errno;
 229                goto out_umem_alloc;
 230        }
 231
 232        umem->umem_area = umem_area;
 233        xsk_set_umem_config(&umem->config, usr_config);
 234
 235        memset(&mr, 0, sizeof(mr));
 236        mr.addr = (uintptr_t)umem_area;
 237        mr.len = size;
 238        mr.chunk_size = umem->config.frame_size;
 239        mr.headroom = umem->config.frame_headroom;
 240        mr.flags = umem->config.flags;
 241
 242        err = setsockopt(umem->fd, SOL_XDP, XDP_UMEM_REG, &mr, sizeof(mr));
 243        if (err) {
 244                err = -errno;
 245                goto out_socket;
 246        }
 247        err = setsockopt(umem->fd, SOL_XDP, XDP_UMEM_FILL_RING,
 248                         &umem->config.fill_size,
 249                         sizeof(umem->config.fill_size));
 250        if (err) {
 251                err = -errno;
 252                goto out_socket;
 253        }
 254        err = setsockopt(umem->fd, SOL_XDP, XDP_UMEM_COMPLETION_RING,
 255                         &umem->config.comp_size,
 256                         sizeof(umem->config.comp_size));
 257        if (err) {
 258                err = -errno;
 259                goto out_socket;
 260        }
 261
 262        err = xsk_get_mmap_offsets(umem->fd, &off);
 263        if (err) {
 264                err = -errno;
 265                goto out_socket;
 266        }
 267
 268        map = mmap(NULL, off.fr.desc + umem->config.fill_size * sizeof(__u64),
 269                   PROT_READ | PROT_WRITE, MAP_SHARED | MAP_POPULATE, umem->fd,
 270                   XDP_UMEM_PGOFF_FILL_RING);
 271        if (map == MAP_FAILED) {
 272                err = -errno;
 273                goto out_socket;
 274        }
 275
 276        umem->fill = fill;
 277        fill->mask = umem->config.fill_size - 1;
 278        fill->size = umem->config.fill_size;
 279        fill->producer = map + off.fr.producer;
 280        fill->consumer = map + off.fr.consumer;
 281        fill->flags = map + off.fr.flags;
 282        fill->ring = map + off.fr.desc;
 283        fill->cached_cons = umem->config.fill_size;
 284
 285        map = mmap(NULL, off.cr.desc + umem->config.comp_size * sizeof(__u64),
 286                   PROT_READ | PROT_WRITE, MAP_SHARED | MAP_POPULATE, umem->fd,
 287                   XDP_UMEM_PGOFF_COMPLETION_RING);
 288        if (map == MAP_FAILED) {
 289                err = -errno;
 290                goto out_mmap;
 291        }
 292
 293        umem->comp = comp;
 294        comp->mask = umem->config.comp_size - 1;
 295        comp->size = umem->config.comp_size;
 296        comp->producer = map + off.cr.producer;
 297        comp->consumer = map + off.cr.consumer;
 298        comp->flags = map + off.cr.flags;
 299        comp->ring = map + off.cr.desc;
 300
 301        *umem_ptr = umem;
 302        return 0;
 303
 304out_mmap:
 305        munmap(map, off.fr.desc + umem->config.fill_size * sizeof(__u64));
 306out_socket:
 307        close(umem->fd);
 308out_umem_alloc:
 309        free(umem);
 310        return err;
 311}
 312
 313struct xsk_umem_config_v1 {
 314        __u32 fill_size;
 315        __u32 comp_size;
 316        __u32 frame_size;
 317        __u32 frame_headroom;
 318};
 319
 320int xsk_umem__create_v0_0_2(struct xsk_umem **umem_ptr, void *umem_area,
 321                            __u64 size, struct xsk_ring_prod *fill,
 322                            struct xsk_ring_cons *comp,
 323                            const struct xsk_umem_config *usr_config)
 324{
 325        struct xsk_umem_config config;
 326
 327        memcpy(&config, usr_config, sizeof(struct xsk_umem_config_v1));
 328        config.flags = 0;
 329
 330        return xsk_umem__create_v0_0_4(umem_ptr, umem_area, size, fill, comp,
 331                                        &config);
 332}
 333COMPAT_VERSION(xsk_umem__create_v0_0_2, xsk_umem__create, LIBBPF_0.0.2)
 334DEFAULT_VERSION(xsk_umem__create_v0_0_4, xsk_umem__create, LIBBPF_0.0.4)
 335
 336static int xsk_load_xdp_prog(struct xsk_socket *xsk)
 337{
 338        static const int log_buf_size = 16 * 1024;
 339        char log_buf[log_buf_size];
 340        int err, prog_fd;
 341
 342        /* This is the C-program:
 343         * SEC("xdp_sock") int xdp_sock_prog(struct xdp_md *ctx)
 344         * {
 345         *     int ret, index = ctx->rx_queue_index;
 346         *
 347         *     // A set entry here means that the correspnding queue_id
 348         *     // has an active AF_XDP socket bound to it.
 349         *     ret = bpf_redirect_map(&xsks_map, index, XDP_PASS);
 350         *     if (ret > 0)
 351         *         return ret;
 352         *
 353         *     // Fallback for pre-5.3 kernels, not supporting default
 354         *     // action in the flags parameter.
 355         *     if (bpf_map_lookup_elem(&xsks_map, &index))
 356         *         return bpf_redirect_map(&xsks_map, index, 0);
 357         *     return XDP_PASS;
 358         * }
 359         */
 360        struct bpf_insn prog[] = {
 361                /* r2 = *(u32 *)(r1 + 16) */
 362                BPF_LDX_MEM(BPF_W, BPF_REG_2, BPF_REG_1, 16),
 363                /* *(u32 *)(r10 - 4) = r2 */
 364                BPF_STX_MEM(BPF_W, BPF_REG_10, BPF_REG_2, -4),
 365                /* r1 = xskmap[] */
 366                BPF_LD_MAP_FD(BPF_REG_1, xsk->xsks_map_fd),
 367                /* r3 = XDP_PASS */
 368                BPF_MOV64_IMM(BPF_REG_3, 2),
 369                /* call bpf_redirect_map */
 370                BPF_EMIT_CALL(BPF_FUNC_redirect_map),
 371                /* if w0 != 0 goto pc+13 */
 372                BPF_JMP32_IMM(BPF_JSGT, BPF_REG_0, 0, 13),
 373                /* r2 = r10 */
 374                BPF_MOV64_REG(BPF_REG_2, BPF_REG_10),
 375                /* r2 += -4 */
 376                BPF_ALU64_IMM(BPF_ADD, BPF_REG_2, -4),
 377                /* r1 = xskmap[] */
 378                BPF_LD_MAP_FD(BPF_REG_1, xsk->xsks_map_fd),
 379                /* call bpf_map_lookup_elem */
 380                BPF_EMIT_CALL(BPF_FUNC_map_lookup_elem),
 381                /* r1 = r0 */
 382                BPF_MOV64_REG(BPF_REG_1, BPF_REG_0),
 383                /* r0 = XDP_PASS */
 384                BPF_MOV64_IMM(BPF_REG_0, 2),
 385                /* if r1 == 0 goto pc+5 */
 386                BPF_JMP_IMM(BPF_JEQ, BPF_REG_1, 0, 5),
 387                /* r2 = *(u32 *)(r10 - 4) */
 388                BPF_LDX_MEM(BPF_W, BPF_REG_2, BPF_REG_10, -4),
 389                /* r1 = xskmap[] */
 390                BPF_LD_MAP_FD(BPF_REG_1, xsk->xsks_map_fd),
 391                /* r3 = 0 */
 392                BPF_MOV64_IMM(BPF_REG_3, 0),
 393                /* call bpf_redirect_map */
 394                BPF_EMIT_CALL(BPF_FUNC_redirect_map),
 395                /* The jumps are to this instruction */
 396                BPF_EXIT_INSN(),
 397        };
 398        size_t insns_cnt = sizeof(prog) / sizeof(struct bpf_insn);
 399
 400        prog_fd = bpf_load_program(BPF_PROG_TYPE_XDP, prog, insns_cnt,
 401                                   "LGPL-2.1 or BSD-2-Clause", 0, log_buf,
 402                                   log_buf_size);
 403        if (prog_fd < 0) {
 404                pr_warn("BPF log buffer:\n%s", log_buf);
 405                return prog_fd;
 406        }
 407
 408        err = bpf_set_link_xdp_fd(xsk->ifindex, prog_fd, xsk->config.xdp_flags);
 409        if (err) {
 410                close(prog_fd);
 411                return err;
 412        }
 413
 414        xsk->prog_fd = prog_fd;
 415        return 0;
 416}
 417
 418static int xsk_get_max_queues(struct xsk_socket *xsk)
 419{
 420        struct ethtool_channels channels = { .cmd = ETHTOOL_GCHANNELS };
 421        struct ifreq ifr = {};
 422        int fd, err, ret;
 423
 424        fd = socket(AF_INET, SOCK_DGRAM, 0);
 425        if (fd < 0)
 426                return -errno;
 427
 428        ifr.ifr_data = (void *)&channels;
 429        memcpy(ifr.ifr_name, xsk->ifname, IFNAMSIZ - 1);
 430        ifr.ifr_name[IFNAMSIZ - 1] = '\0';
 431        err = ioctl(fd, SIOCETHTOOL, &ifr);
 432        if (err && errno != EOPNOTSUPP) {
 433                ret = -errno;
 434                goto out;
 435        }
 436
 437        if (err) {
 438                /* If the device says it has no channels, then all traffic
 439                 * is sent to a single stream, so max queues = 1.
 440                 */
 441                ret = 1;
 442        } else {
 443                /* Take the max of rx, tx, combined. Drivers return
 444                 * the number of channels in different ways.
 445                 */
 446                ret = max(channels.max_rx, channels.max_tx);
 447                ret = max(ret, (int)channels.max_combined);
 448        }
 449
 450out:
 451        close(fd);
 452        return ret;
 453}
 454
 455static int xsk_create_bpf_maps(struct xsk_socket *xsk)
 456{
 457        int max_queues;
 458        int fd;
 459
 460        max_queues = xsk_get_max_queues(xsk);
 461        if (max_queues < 0)
 462                return max_queues;
 463
 464        fd = bpf_create_map_name(BPF_MAP_TYPE_XSKMAP, "xsks_map",
 465                                 sizeof(int), sizeof(int), max_queues, 0);
 466        if (fd < 0)
 467                return fd;
 468
 469        xsk->xsks_map_fd = fd;
 470
 471        return 0;
 472}
 473
 474static void xsk_delete_bpf_maps(struct xsk_socket *xsk)
 475{
 476        bpf_map_delete_elem(xsk->xsks_map_fd, &xsk->queue_id);
 477        close(xsk->xsks_map_fd);
 478}
 479
 480static int xsk_lookup_bpf_maps(struct xsk_socket *xsk)
 481{
 482        __u32 i, *map_ids, num_maps, prog_len = sizeof(struct bpf_prog_info);
 483        __u32 map_len = sizeof(struct bpf_map_info);
 484        struct bpf_prog_info prog_info = {};
 485        struct bpf_map_info map_info;
 486        int fd, err;
 487
 488        err = bpf_obj_get_info_by_fd(xsk->prog_fd, &prog_info, &prog_len);
 489        if (err)
 490                return err;
 491
 492        num_maps = prog_info.nr_map_ids;
 493
 494        map_ids = calloc(prog_info.nr_map_ids, sizeof(*map_ids));
 495        if (!map_ids)
 496                return -ENOMEM;
 497
 498        memset(&prog_info, 0, prog_len);
 499        prog_info.nr_map_ids = num_maps;
 500        prog_info.map_ids = (__u64)(unsigned long)map_ids;
 501
 502        err = bpf_obj_get_info_by_fd(xsk->prog_fd, &prog_info, &prog_len);
 503        if (err)
 504                goto out_map_ids;
 505
 506        xsk->xsks_map_fd = -1;
 507
 508        for (i = 0; i < prog_info.nr_map_ids; i++) {
 509                fd = bpf_map_get_fd_by_id(map_ids[i]);
 510                if (fd < 0)
 511                        continue;
 512
 513                err = bpf_obj_get_info_by_fd(fd, &map_info, &map_len);
 514                if (err) {
 515                        close(fd);
 516                        continue;
 517                }
 518
 519                if (!strcmp(map_info.name, "xsks_map")) {
 520                        xsk->xsks_map_fd = fd;
 521                        continue;
 522                }
 523
 524                close(fd);
 525        }
 526
 527        err = 0;
 528        if (xsk->xsks_map_fd == -1)
 529                err = -ENOENT;
 530
 531out_map_ids:
 532        free(map_ids);
 533        return err;
 534}
 535
 536static int xsk_set_bpf_maps(struct xsk_socket *xsk)
 537{
 538        return bpf_map_update_elem(xsk->xsks_map_fd, &xsk->queue_id,
 539                                   &xsk->fd, 0);
 540}
 541
 542static int xsk_setup_xdp_prog(struct xsk_socket *xsk)
 543{
 544        __u32 prog_id = 0;
 545        int err;
 546
 547        err = bpf_get_link_xdp_id(xsk->ifindex, &prog_id,
 548                                  xsk->config.xdp_flags);
 549        if (err)
 550                return err;
 551
 552        if (!prog_id) {
 553                err = xsk_create_bpf_maps(xsk);
 554                if (err)
 555                        return err;
 556
 557                err = xsk_load_xdp_prog(xsk);
 558                if (err) {
 559                        xsk_delete_bpf_maps(xsk);
 560                        return err;
 561                }
 562        } else {
 563                xsk->prog_fd = bpf_prog_get_fd_by_id(prog_id);
 564                if (xsk->prog_fd < 0)
 565                        return -errno;
 566                err = xsk_lookup_bpf_maps(xsk);
 567                if (err) {
 568                        close(xsk->prog_fd);
 569                        return err;
 570                }
 571        }
 572
 573        if (xsk->rx)
 574                err = xsk_set_bpf_maps(xsk);
 575        if (err) {
 576                xsk_delete_bpf_maps(xsk);
 577                close(xsk->prog_fd);
 578                return err;
 579        }
 580
 581        return 0;
 582}
 583
 584int xsk_socket__create(struct xsk_socket **xsk_ptr, const char *ifname,
 585                       __u32 queue_id, struct xsk_umem *umem,
 586                       struct xsk_ring_cons *rx, struct xsk_ring_prod *tx,
 587                       const struct xsk_socket_config *usr_config)
 588{
 589        void *rx_map = NULL, *tx_map = NULL;
 590        struct sockaddr_xdp sxdp = {};
 591        struct xdp_mmap_offsets off;
 592        struct xsk_socket *xsk;
 593        int err;
 594
 595        if (!umem || !xsk_ptr || !(rx || tx))
 596                return -EFAULT;
 597
 598        xsk = calloc(1, sizeof(*xsk));
 599        if (!xsk)
 600                return -ENOMEM;
 601
 602        err = xsk_set_xdp_socket_config(&xsk->config, usr_config);
 603        if (err)
 604                goto out_xsk_alloc;
 605
 606        if (umem->refcount &&
 607            !(xsk->config.libbpf_flags & XSK_LIBBPF_FLAGS__INHIBIT_PROG_LOAD)) {
 608                pr_warn("Error: shared umems not supported by libbpf supplied XDP program.\n");
 609                err = -EBUSY;
 610                goto out_xsk_alloc;
 611        }
 612
 613        if (umem->refcount++ > 0) {
 614                xsk->fd = socket(AF_XDP, SOCK_RAW, 0);
 615                if (xsk->fd < 0) {
 616                        err = -errno;
 617                        goto out_xsk_alloc;
 618                }
 619        } else {
 620                xsk->fd = umem->fd;
 621        }
 622
 623        xsk->outstanding_tx = 0;
 624        xsk->queue_id = queue_id;
 625        xsk->umem = umem;
 626        xsk->ifindex = if_nametoindex(ifname);
 627        if (!xsk->ifindex) {
 628                err = -errno;
 629                goto out_socket;
 630        }
 631        memcpy(xsk->ifname, ifname, IFNAMSIZ - 1);
 632        xsk->ifname[IFNAMSIZ - 1] = '\0';
 633
 634        if (rx) {
 635                err = setsockopt(xsk->fd, SOL_XDP, XDP_RX_RING,
 636                                 &xsk->config.rx_size,
 637                                 sizeof(xsk->config.rx_size));
 638                if (err) {
 639                        err = -errno;
 640                        goto out_socket;
 641                }
 642        }
 643        if (tx) {
 644                err = setsockopt(xsk->fd, SOL_XDP, XDP_TX_RING,
 645                                 &xsk->config.tx_size,
 646                                 sizeof(xsk->config.tx_size));
 647                if (err) {
 648                        err = -errno;
 649                        goto out_socket;
 650                }
 651        }
 652
 653        err = xsk_get_mmap_offsets(xsk->fd, &off);
 654        if (err) {
 655                err = -errno;
 656                goto out_socket;
 657        }
 658
 659        if (rx) {
 660                rx_map = mmap(NULL, off.rx.desc +
 661                              xsk->config.rx_size * sizeof(struct xdp_desc),
 662                              PROT_READ | PROT_WRITE, MAP_SHARED | MAP_POPULATE,
 663                              xsk->fd, XDP_PGOFF_RX_RING);
 664                if (rx_map == MAP_FAILED) {
 665                        err = -errno;
 666                        goto out_socket;
 667                }
 668
 669                rx->mask = xsk->config.rx_size - 1;
 670                rx->size = xsk->config.rx_size;
 671                rx->producer = rx_map + off.rx.producer;
 672                rx->consumer = rx_map + off.rx.consumer;
 673                rx->flags = rx_map + off.rx.flags;
 674                rx->ring = rx_map + off.rx.desc;
 675        }
 676        xsk->rx = rx;
 677
 678        if (tx) {
 679                tx_map = mmap(NULL, off.tx.desc +
 680                              xsk->config.tx_size * sizeof(struct xdp_desc),
 681                              PROT_READ | PROT_WRITE, MAP_SHARED | MAP_POPULATE,
 682                              xsk->fd, XDP_PGOFF_TX_RING);
 683                if (tx_map == MAP_FAILED) {
 684                        err = -errno;
 685                        goto out_mmap_rx;
 686                }
 687
 688                tx->mask = xsk->config.tx_size - 1;
 689                tx->size = xsk->config.tx_size;
 690                tx->producer = tx_map + off.tx.producer;
 691                tx->consumer = tx_map + off.tx.consumer;
 692                tx->flags = tx_map + off.tx.flags;
 693                tx->ring = tx_map + off.tx.desc;
 694                tx->cached_cons = xsk->config.tx_size;
 695        }
 696        xsk->tx = tx;
 697
 698        sxdp.sxdp_family = PF_XDP;
 699        sxdp.sxdp_ifindex = xsk->ifindex;
 700        sxdp.sxdp_queue_id = xsk->queue_id;
 701        if (umem->refcount > 1) {
 702                sxdp.sxdp_flags = XDP_SHARED_UMEM;
 703                sxdp.sxdp_shared_umem_fd = umem->fd;
 704        } else {
 705                sxdp.sxdp_flags = xsk->config.bind_flags;
 706        }
 707
 708        err = bind(xsk->fd, (struct sockaddr *)&sxdp, sizeof(sxdp));
 709        if (err) {
 710                err = -errno;
 711                goto out_mmap_tx;
 712        }
 713
 714        xsk->prog_fd = -1;
 715
 716        if (!(xsk->config.libbpf_flags & XSK_LIBBPF_FLAGS__INHIBIT_PROG_LOAD)) {
 717                err = xsk_setup_xdp_prog(xsk);
 718                if (err)
 719                        goto out_mmap_tx;
 720        }
 721
 722        *xsk_ptr = xsk;
 723        return 0;
 724
 725out_mmap_tx:
 726        if (tx)
 727                munmap(tx_map, off.tx.desc +
 728                       xsk->config.tx_size * sizeof(struct xdp_desc));
 729out_mmap_rx:
 730        if (rx)
 731                munmap(rx_map, off.rx.desc +
 732                       xsk->config.rx_size * sizeof(struct xdp_desc));
 733out_socket:
 734        if (--umem->refcount)
 735                close(xsk->fd);
 736out_xsk_alloc:
 737        free(xsk);
 738        return err;
 739}
 740
 741int xsk_umem__delete(struct xsk_umem *umem)
 742{
 743        struct xdp_mmap_offsets off;
 744        int err;
 745
 746        if (!umem)
 747                return 0;
 748
 749        if (umem->refcount)
 750                return -EBUSY;
 751
 752        err = xsk_get_mmap_offsets(umem->fd, &off);
 753        if (!err) {
 754                munmap(umem->fill->ring - off.fr.desc,
 755                       off.fr.desc + umem->config.fill_size * sizeof(__u64));
 756                munmap(umem->comp->ring - off.cr.desc,
 757                       off.cr.desc + umem->config.comp_size * sizeof(__u64));
 758        }
 759
 760        close(umem->fd);
 761        free(umem);
 762
 763        return 0;
 764}
 765
 766void xsk_socket__delete(struct xsk_socket *xsk)
 767{
 768        size_t desc_sz = sizeof(struct xdp_desc);
 769        struct xdp_mmap_offsets off;
 770        int err;
 771
 772        if (!xsk)
 773                return;
 774
 775        if (xsk->prog_fd != -1) {
 776                xsk_delete_bpf_maps(xsk);
 777                close(xsk->prog_fd);
 778        }
 779
 780        err = xsk_get_mmap_offsets(xsk->fd, &off);
 781        if (!err) {
 782                if (xsk->rx) {
 783                        munmap(xsk->rx->ring - off.rx.desc,
 784                               off.rx.desc + xsk->config.rx_size * desc_sz);
 785                }
 786                if (xsk->tx) {
 787                        munmap(xsk->tx->ring - off.tx.desc,
 788                               off.tx.desc + xsk->config.tx_size * desc_sz);
 789                }
 790
 791        }
 792
 793        xsk->umem->refcount--;
 794        /* Do not close an fd that also has an associated umem connected
 795         * to it.
 796         */
 797        if (xsk->fd != xsk->umem->fd)
 798                close(xsk->fd);
 799        free(xsk);
 800}
 801