1 // SPDX-License-Identifier: GPL-2.0-only 2 /* Copyright (c) 2023 Isovalent */ 3 4 #include <linux/netdevice.h> 5 #include <linux/ethtool.h> 6 #include <linux/etherdevice.h> 7 #include <linux/filter.h> 8 #include <linux/netfilter_netdev.h> 9 #include <linux/bpf_mprog.h> 10 #include <linux/indirect_call_wrapper.h> 11 12 #include <net/netkit.h> 13 #include <net/dst.h> 14 #include <net/tcx.h> 15 16 #define DRV_NAME "netkit" 17 18 struct netkit { 19 /* Needed in fast-path */ 20 struct net_device __rcu *peer; 21 struct bpf_mprog_entry __rcu *active; 22 enum netkit_action policy; 23 struct bpf_mprog_bundle bundle; 24 25 /* Needed in slow-path */ 26 enum netkit_mode mode; 27 bool primary; 28 u32 headroom; 29 }; 30 31 struct netkit_link { 32 struct bpf_link link; 33 struct net_device *dev; 34 u32 location; 35 }; 36 37 static __always_inline int 38 netkit_run(const struct bpf_mprog_entry *entry, struct sk_buff *skb, 39 enum netkit_action ret) 40 { 41 const struct bpf_mprog_fp *fp; 42 const struct bpf_prog *prog; 43 44 bpf_mprog_foreach_prog(entry, fp, prog) { 45 bpf_compute_data_pointers(skb); 46 ret = bpf_prog_run(prog, skb); 47 if (ret != NETKIT_NEXT) 48 break; 49 } 50 return ret; 51 } 52 53 static void netkit_prep_forward(struct sk_buff *skb, bool xnet) 54 { 55 skb_scrub_packet(skb, xnet); 56 skb->priority = 0; 57 nf_skip_egress(skb, true); 58 skb_reset_mac_header(skb); 59 } 60 61 static struct netkit *netkit_priv(const struct net_device *dev) 62 { 63 return netdev_priv(dev); 64 } 65 66 static netdev_tx_t netkit_xmit(struct sk_buff *skb, struct net_device *dev) 67 { 68 struct netkit *nk = netkit_priv(dev); 69 enum netkit_action ret = READ_ONCE(nk->policy); 70 netdev_tx_t ret_dev = NET_XMIT_SUCCESS; 71 const struct bpf_mprog_entry *entry; 72 struct net_device *peer; 73 int len = skb->len; 74 75 rcu_read_lock(); 76 peer = rcu_dereference(nk->peer); 77 if (unlikely(!peer || !(peer->flags & IFF_UP) || 78 !pskb_may_pull(skb, ETH_HLEN) || 79 skb_orphan_frags(skb, GFP_ATOMIC))) 80 goto drop; 81 netkit_prep_forward(skb, !net_eq(dev_net(dev), dev_net(peer))); 82 eth_skb_pkt_type(skb, peer); 83 skb->dev = peer; 84 entry = rcu_dereference(nk->active); 85 if (entry) 86 ret = netkit_run(entry, skb, ret); 87 switch (ret) { 88 case NETKIT_NEXT: 89 case NETKIT_PASS: 90 eth_skb_pull_mac(skb); 91 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN); 92 if (likely(__netif_rx(skb) == NET_RX_SUCCESS)) { 93 dev_sw_netstats_tx_add(dev, 1, len); 94 dev_sw_netstats_rx_add(peer, len); 95 } else { 96 goto drop_stats; 97 } 98 break; 99 case NETKIT_REDIRECT: 100 dev_sw_netstats_tx_add(dev, 1, len); 101 skb_do_redirect(skb); 102 break; 103 case NETKIT_DROP: 104 default: 105 drop: 106 kfree_skb(skb); 107 drop_stats: 108 dev_core_stats_tx_dropped_inc(dev); 109 ret_dev = NET_XMIT_DROP; 110 break; 111 } 112 rcu_read_unlock(); 113 return ret_dev; 114 } 115 116 static int netkit_open(struct net_device *dev) 117 { 118 struct netkit *nk = netkit_priv(dev); 119 struct net_device *peer = rtnl_dereference(nk->peer); 120 121 if (!peer) 122 return -ENOTCONN; 123 if (peer->flags & IFF_UP) { 124 netif_carrier_on(dev); 125 netif_carrier_on(peer); 126 } 127 return 0; 128 } 129 130 static int netkit_close(struct net_device *dev) 131 { 132 struct netkit *nk = netkit_priv(dev); 133 struct net_device *peer = rtnl_dereference(nk->peer); 134 135 netif_carrier_off(dev); 136 if (peer) 137 netif_carrier_off(peer); 138 return 0; 139 } 140 141 static int netkit_get_iflink(const struct net_device *dev) 142 { 143 struct netkit *nk = netkit_priv(dev); 144 struct net_device *peer; 145 int iflink = 0; 146 147 rcu_read_lock(); 148 peer = rcu_dereference(nk->peer); 149 if (peer) 150 iflink = READ_ONCE(peer->ifindex); 151 rcu_read_unlock(); 152 return iflink; 153 } 154 155 static void netkit_set_multicast(struct net_device *dev) 156 { 157 /* Nothing to do, we receive whatever gets pushed to us! */ 158 } 159 160 static int netkit_set_macaddr(struct net_device *dev, void *sa) 161 { 162 struct netkit *nk = netkit_priv(dev); 163 164 if (nk->mode != NETKIT_L2) 165 return -EOPNOTSUPP; 166 167 return eth_mac_addr(dev, sa); 168 } 169 170 static void netkit_set_headroom(struct net_device *dev, int headroom) 171 { 172 struct netkit *nk = netkit_priv(dev), *nk2; 173 struct net_device *peer; 174 175 if (headroom < 0) 176 headroom = NET_SKB_PAD; 177 178 rcu_read_lock(); 179 peer = rcu_dereference(nk->peer); 180 if (unlikely(!peer)) 181 goto out; 182 183 nk2 = netkit_priv(peer); 184 nk->headroom = headroom; 185 headroom = max(nk->headroom, nk2->headroom); 186 187 peer->needed_headroom = headroom; 188 dev->needed_headroom = headroom; 189 out: 190 rcu_read_unlock(); 191 } 192 193 INDIRECT_CALLABLE_SCOPE struct net_device *netkit_peer_dev(struct net_device *dev) 194 { 195 return rcu_dereference(netkit_priv(dev)->peer); 196 } 197 198 static void netkit_get_stats(struct net_device *dev, 199 struct rtnl_link_stats64 *stats) 200 { 201 dev_fetch_sw_netstats(stats, dev->tstats); 202 stats->tx_dropped = DEV_STATS_READ(dev, tx_dropped); 203 } 204 205 static void netkit_uninit(struct net_device *dev); 206 207 static const struct net_device_ops netkit_netdev_ops = { 208 .ndo_open = netkit_open, 209 .ndo_stop = netkit_close, 210 .ndo_start_xmit = netkit_xmit, 211 .ndo_set_rx_mode = netkit_set_multicast, 212 .ndo_set_rx_headroom = netkit_set_headroom, 213 .ndo_set_mac_address = netkit_set_macaddr, 214 .ndo_get_iflink = netkit_get_iflink, 215 .ndo_get_peer_dev = netkit_peer_dev, 216 .ndo_get_stats64 = netkit_get_stats, 217 .ndo_uninit = netkit_uninit, 218 .ndo_features_check = passthru_features_check, 219 }; 220 221 static void netkit_get_drvinfo(struct net_device *dev, 222 struct ethtool_drvinfo *info) 223 { 224 strscpy(info->driver, DRV_NAME, sizeof(info->driver)); 225 } 226 227 static const struct ethtool_ops netkit_ethtool_ops = { 228 .get_drvinfo = netkit_get_drvinfo, 229 }; 230 231 static void netkit_setup(struct net_device *dev) 232 { 233 static const netdev_features_t netkit_features_hw_vlan = 234 NETIF_F_HW_VLAN_CTAG_TX | 235 NETIF_F_HW_VLAN_CTAG_RX | 236 NETIF_F_HW_VLAN_STAG_TX | 237 NETIF_F_HW_VLAN_STAG_RX; 238 static const netdev_features_t netkit_features = 239 netkit_features_hw_vlan | 240 NETIF_F_SG | 241 NETIF_F_FRAGLIST | 242 NETIF_F_HW_CSUM | 243 NETIF_F_RXCSUM | 244 NETIF_F_SCTP_CRC | 245 NETIF_F_HIGHDMA | 246 NETIF_F_GSO_SOFTWARE | 247 NETIF_F_GSO_ENCAP_ALL; 248 249 ether_setup(dev); 250 dev->max_mtu = ETH_MAX_MTU; 251 dev->pcpu_stat_type = NETDEV_PCPU_STAT_TSTATS; 252 253 dev->flags |= IFF_NOARP; 254 dev->priv_flags &= ~IFF_TX_SKB_SHARING; 255 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE; 256 dev->priv_flags |= IFF_PHONY_HEADROOM; 257 dev->priv_flags |= IFF_NO_QUEUE; 258 dev->lltx = true; 259 260 dev->ethtool_ops = &netkit_ethtool_ops; 261 dev->netdev_ops = &netkit_netdev_ops; 262 263 dev->features |= netkit_features; 264 dev->hw_features = netkit_features; 265 dev->hw_enc_features = netkit_features; 266 dev->mpls_features = NETIF_F_HW_CSUM | NETIF_F_GSO_SOFTWARE; 267 dev->vlan_features = dev->features & ~netkit_features_hw_vlan; 268 269 dev->needs_free_netdev = true; 270 271 netif_set_tso_max_size(dev, GSO_MAX_SIZE); 272 } 273 274 static struct net *netkit_get_link_net(const struct net_device *dev) 275 { 276 struct netkit *nk = netkit_priv(dev); 277 struct net_device *peer = rtnl_dereference(nk->peer); 278 279 return peer ? dev_net(peer) : dev_net(dev); 280 } 281 282 static int netkit_check_policy(int policy, struct nlattr *tb, 283 struct netlink_ext_ack *extack) 284 { 285 switch (policy) { 286 case NETKIT_PASS: 287 case NETKIT_DROP: 288 return 0; 289 default: 290 NL_SET_ERR_MSG_ATTR(extack, tb, 291 "Provided default xmit policy not supported"); 292 return -EINVAL; 293 } 294 } 295 296 static int netkit_check_mode(int mode, struct nlattr *tb, 297 struct netlink_ext_ack *extack) 298 { 299 switch (mode) { 300 case NETKIT_L2: 301 case NETKIT_L3: 302 return 0; 303 default: 304 NL_SET_ERR_MSG_ATTR(extack, tb, 305 "Provided device mode can only be L2 or L3"); 306 return -EINVAL; 307 } 308 } 309 310 static int netkit_validate(struct nlattr *tb[], struct nlattr *data[], 311 struct netlink_ext_ack *extack) 312 { 313 struct nlattr *attr = tb[IFLA_ADDRESS]; 314 315 if (!attr) 316 return 0; 317 if (nla_len(attr) != ETH_ALEN) 318 return -EINVAL; 319 if (!is_valid_ether_addr(nla_data(attr))) 320 return -EADDRNOTAVAIL; 321 return 0; 322 } 323 324 static struct rtnl_link_ops netkit_link_ops; 325 326 static int netkit_new_link(struct net *src_net, struct net_device *dev, 327 struct nlattr *tb[], struct nlattr *data[], 328 struct netlink_ext_ack *extack) 329 { 330 struct nlattr *peer_tb[IFLA_MAX + 1], **tbp = tb, *attr; 331 enum netkit_action default_prim = NETKIT_PASS; 332 enum netkit_action default_peer = NETKIT_PASS; 333 enum netkit_mode mode = NETKIT_L3; 334 unsigned char ifname_assign_type; 335 struct ifinfomsg *ifmp = NULL; 336 struct net_device *peer; 337 char ifname[IFNAMSIZ]; 338 struct netkit *nk; 339 struct net *net; 340 int err; 341 342 if (data) { 343 if (data[IFLA_NETKIT_MODE]) { 344 attr = data[IFLA_NETKIT_MODE]; 345 mode = nla_get_u32(attr); 346 err = netkit_check_mode(mode, attr, extack); 347 if (err < 0) 348 return err; 349 } 350 if (data[IFLA_NETKIT_PEER_INFO]) { 351 attr = data[IFLA_NETKIT_PEER_INFO]; 352 ifmp = nla_data(attr); 353 err = rtnl_nla_parse_ifinfomsg(peer_tb, attr, extack); 354 if (err < 0) 355 return err; 356 err = netkit_validate(peer_tb, NULL, extack); 357 if (err < 0) 358 return err; 359 tbp = peer_tb; 360 } 361 if (data[IFLA_NETKIT_POLICY]) { 362 attr = data[IFLA_NETKIT_POLICY]; 363 default_prim = nla_get_u32(attr); 364 err = netkit_check_policy(default_prim, attr, extack); 365 if (err < 0) 366 return err; 367 } 368 if (data[IFLA_NETKIT_PEER_POLICY]) { 369 attr = data[IFLA_NETKIT_PEER_POLICY]; 370 default_peer = nla_get_u32(attr); 371 err = netkit_check_policy(default_peer, attr, extack); 372 if (err < 0) 373 return err; 374 } 375 } 376 377 if (ifmp && tbp[IFLA_IFNAME]) { 378 nla_strscpy(ifname, tbp[IFLA_IFNAME], IFNAMSIZ); 379 ifname_assign_type = NET_NAME_USER; 380 } else { 381 strscpy(ifname, "nk%d", IFNAMSIZ); 382 ifname_assign_type = NET_NAME_ENUM; 383 } 384 if (mode != NETKIT_L2 && 385 (tb[IFLA_ADDRESS] || tbp[IFLA_ADDRESS])) 386 return -EOPNOTSUPP; 387 388 net = rtnl_link_get_net(src_net, tbp); 389 if (IS_ERR(net)) 390 return PTR_ERR(net); 391 392 peer = rtnl_create_link(net, ifname, ifname_assign_type, 393 &netkit_link_ops, tbp, extack); 394 if (IS_ERR(peer)) { 395 put_net(net); 396 return PTR_ERR(peer); 397 } 398 399 netif_inherit_tso_max(peer, dev); 400 401 if (mode == NETKIT_L2 && !(ifmp && tbp[IFLA_ADDRESS])) 402 eth_hw_addr_random(peer); 403 if (ifmp && dev->ifindex) 404 peer->ifindex = ifmp->ifi_index; 405 406 nk = netkit_priv(peer); 407 nk->primary = false; 408 nk->policy = default_peer; 409 nk->mode = mode; 410 bpf_mprog_bundle_init(&nk->bundle); 411 412 err = register_netdevice(peer); 413 put_net(net); 414 if (err < 0) 415 goto err_register_peer; 416 netif_carrier_off(peer); 417 if (mode == NETKIT_L2) 418 dev_change_flags(peer, peer->flags & ~IFF_NOARP, NULL); 419 420 err = rtnl_configure_link(peer, NULL, 0, NULL); 421 if (err < 0) 422 goto err_configure_peer; 423 424 if (mode == NETKIT_L2 && !tb[IFLA_ADDRESS]) 425 eth_hw_addr_random(dev); 426 if (tb[IFLA_IFNAME]) 427 nla_strscpy(dev->name, tb[IFLA_IFNAME], IFNAMSIZ); 428 else 429 strscpy(dev->name, "nk%d", IFNAMSIZ); 430 431 nk = netkit_priv(dev); 432 nk->primary = true; 433 nk->policy = default_prim; 434 nk->mode = mode; 435 bpf_mprog_bundle_init(&nk->bundle); 436 437 err = register_netdevice(dev); 438 if (err < 0) 439 goto err_configure_peer; 440 netif_carrier_off(dev); 441 if (mode == NETKIT_L2) 442 dev_change_flags(dev, dev->flags & ~IFF_NOARP, NULL); 443 444 rcu_assign_pointer(netkit_priv(dev)->peer, peer); 445 rcu_assign_pointer(netkit_priv(peer)->peer, dev); 446 return 0; 447 err_configure_peer: 448 unregister_netdevice(peer); 449 return err; 450 err_register_peer: 451 free_netdev(peer); 452 return err; 453 } 454 455 static struct bpf_mprog_entry *netkit_entry_fetch(struct net_device *dev, 456 bool bundle_fallback) 457 { 458 struct netkit *nk = netkit_priv(dev); 459 struct bpf_mprog_entry *entry; 460 461 ASSERT_RTNL(); 462 entry = rcu_dereference_rtnl(nk->active); 463 if (entry) 464 return entry; 465 if (bundle_fallback) 466 return &nk->bundle.a; 467 return NULL; 468 } 469 470 static void netkit_entry_update(struct net_device *dev, 471 struct bpf_mprog_entry *entry) 472 { 473 struct netkit *nk = netkit_priv(dev); 474 475 ASSERT_RTNL(); 476 rcu_assign_pointer(nk->active, entry); 477 } 478 479 static void netkit_entry_sync(void) 480 { 481 synchronize_rcu(); 482 } 483 484 static struct net_device *netkit_dev_fetch(struct net *net, u32 ifindex, u32 which) 485 { 486 struct net_device *dev; 487 struct netkit *nk; 488 489 ASSERT_RTNL(); 490 491 switch (which) { 492 case BPF_NETKIT_PRIMARY: 493 case BPF_NETKIT_PEER: 494 break; 495 default: 496 return ERR_PTR(-EINVAL); 497 } 498 499 dev = __dev_get_by_index(net, ifindex); 500 if (!dev) 501 return ERR_PTR(-ENODEV); 502 if (dev->netdev_ops != &netkit_netdev_ops) 503 return ERR_PTR(-ENXIO); 504 505 nk = netkit_priv(dev); 506 if (!nk->primary) 507 return ERR_PTR(-EACCES); 508 if (which == BPF_NETKIT_PEER) { 509 dev = rcu_dereference_rtnl(nk->peer); 510 if (!dev) 511 return ERR_PTR(-ENODEV); 512 } 513 return dev; 514 } 515 516 int netkit_prog_attach(const union bpf_attr *attr, struct bpf_prog *prog) 517 { 518 struct bpf_mprog_entry *entry, *entry_new; 519 struct bpf_prog *replace_prog = NULL; 520 struct net_device *dev; 521 int ret; 522 523 rtnl_lock(); 524 dev = netkit_dev_fetch(current->nsproxy->net_ns, attr->target_ifindex, 525 attr->attach_type); 526 if (IS_ERR(dev)) { 527 ret = PTR_ERR(dev); 528 goto out; 529 } 530 entry = netkit_entry_fetch(dev, true); 531 if (attr->attach_flags & BPF_F_REPLACE) { 532 replace_prog = bpf_prog_get_type(attr->replace_bpf_fd, 533 prog->type); 534 if (IS_ERR(replace_prog)) { 535 ret = PTR_ERR(replace_prog); 536 replace_prog = NULL; 537 goto out; 538 } 539 } 540 ret = bpf_mprog_attach(entry, &entry_new, prog, NULL, replace_prog, 541 attr->attach_flags, attr->relative_fd, 542 attr->expected_revision); 543 if (!ret) { 544 if (entry != entry_new) { 545 netkit_entry_update(dev, entry_new); 546 netkit_entry_sync(); 547 } 548 bpf_mprog_commit(entry); 549 } 550 out: 551 if (replace_prog) 552 bpf_prog_put(replace_prog); 553 rtnl_unlock(); 554 return ret; 555 } 556 557 int netkit_prog_detach(const union bpf_attr *attr, struct bpf_prog *prog) 558 { 559 struct bpf_mprog_entry *entry, *entry_new; 560 struct net_device *dev; 561 int ret; 562 563 rtnl_lock(); 564 dev = netkit_dev_fetch(current->nsproxy->net_ns, attr->target_ifindex, 565 attr->attach_type); 566 if (IS_ERR(dev)) { 567 ret = PTR_ERR(dev); 568 goto out; 569 } 570 entry = netkit_entry_fetch(dev, false); 571 if (!entry) { 572 ret = -ENOENT; 573 goto out; 574 } 575 ret = bpf_mprog_detach(entry, &entry_new, prog, NULL, attr->attach_flags, 576 attr->relative_fd, attr->expected_revision); 577 if (!ret) { 578 if (!bpf_mprog_total(entry_new)) 579 entry_new = NULL; 580 netkit_entry_update(dev, entry_new); 581 netkit_entry_sync(); 582 bpf_mprog_commit(entry); 583 } 584 out: 585 rtnl_unlock(); 586 return ret; 587 } 588 589 int netkit_prog_query(const union bpf_attr *attr, union bpf_attr __user *uattr) 590 { 591 struct net_device *dev; 592 int ret; 593 594 rtnl_lock(); 595 dev = netkit_dev_fetch(current->nsproxy->net_ns, 596 attr->query.target_ifindex, 597 attr->query.attach_type); 598 if (IS_ERR(dev)) { 599 ret = PTR_ERR(dev); 600 goto out; 601 } 602 ret = bpf_mprog_query(attr, uattr, netkit_entry_fetch(dev, false)); 603 out: 604 rtnl_unlock(); 605 return ret; 606 } 607 608 static struct netkit_link *netkit_link(const struct bpf_link *link) 609 { 610 return container_of(link, struct netkit_link, link); 611 } 612 613 static int netkit_link_prog_attach(struct bpf_link *link, u32 flags, 614 u32 id_or_fd, u64 revision) 615 { 616 struct netkit_link *nkl = netkit_link(link); 617 struct bpf_mprog_entry *entry, *entry_new; 618 struct net_device *dev = nkl->dev; 619 int ret; 620 621 ASSERT_RTNL(); 622 entry = netkit_entry_fetch(dev, true); 623 ret = bpf_mprog_attach(entry, &entry_new, link->prog, link, NULL, flags, 624 id_or_fd, revision); 625 if (!ret) { 626 if (entry != entry_new) { 627 netkit_entry_update(dev, entry_new); 628 netkit_entry_sync(); 629 } 630 bpf_mprog_commit(entry); 631 } 632 return ret; 633 } 634 635 static void netkit_link_release(struct bpf_link *link) 636 { 637 struct netkit_link *nkl = netkit_link(link); 638 struct bpf_mprog_entry *entry, *entry_new; 639 struct net_device *dev; 640 int ret = 0; 641 642 rtnl_lock(); 643 dev = nkl->dev; 644 if (!dev) 645 goto out; 646 entry = netkit_entry_fetch(dev, false); 647 if (!entry) { 648 ret = -ENOENT; 649 goto out; 650 } 651 ret = bpf_mprog_detach(entry, &entry_new, link->prog, link, 0, 0, 0); 652 if (!ret) { 653 if (!bpf_mprog_total(entry_new)) 654 entry_new = NULL; 655 netkit_entry_update(dev, entry_new); 656 netkit_entry_sync(); 657 bpf_mprog_commit(entry); 658 nkl->dev = NULL; 659 } 660 out: 661 WARN_ON_ONCE(ret); 662 rtnl_unlock(); 663 } 664 665 static int netkit_link_update(struct bpf_link *link, struct bpf_prog *nprog, 666 struct bpf_prog *oprog) 667 { 668 struct netkit_link *nkl = netkit_link(link); 669 struct bpf_mprog_entry *entry, *entry_new; 670 struct net_device *dev; 671 int ret = 0; 672 673 rtnl_lock(); 674 dev = nkl->dev; 675 if (!dev) { 676 ret = -ENOLINK; 677 goto out; 678 } 679 if (oprog && link->prog != oprog) { 680 ret = -EPERM; 681 goto out; 682 } 683 oprog = link->prog; 684 if (oprog == nprog) { 685 bpf_prog_put(nprog); 686 goto out; 687 } 688 entry = netkit_entry_fetch(dev, false); 689 if (!entry) { 690 ret = -ENOENT; 691 goto out; 692 } 693 ret = bpf_mprog_attach(entry, &entry_new, nprog, link, oprog, 694 BPF_F_REPLACE | BPF_F_ID, 695 link->prog->aux->id, 0); 696 if (!ret) { 697 WARN_ON_ONCE(entry != entry_new); 698 oprog = xchg(&link->prog, nprog); 699 bpf_prog_put(oprog); 700 bpf_mprog_commit(entry); 701 } 702 out: 703 rtnl_unlock(); 704 return ret; 705 } 706 707 static void netkit_link_dealloc(struct bpf_link *link) 708 { 709 kfree(netkit_link(link)); 710 } 711 712 static void netkit_link_fdinfo(const struct bpf_link *link, struct seq_file *seq) 713 { 714 const struct netkit_link *nkl = netkit_link(link); 715 u32 ifindex = 0; 716 717 rtnl_lock(); 718 if (nkl->dev) 719 ifindex = nkl->dev->ifindex; 720 rtnl_unlock(); 721 722 seq_printf(seq, "ifindex:\t%u\n", ifindex); 723 seq_printf(seq, "attach_type:\t%u (%s)\n", 724 nkl->location, 725 nkl->location == BPF_NETKIT_PRIMARY ? "primary" : "peer"); 726 } 727 728 static int netkit_link_fill_info(const struct bpf_link *link, 729 struct bpf_link_info *info) 730 { 731 const struct netkit_link *nkl = netkit_link(link); 732 u32 ifindex = 0; 733 734 rtnl_lock(); 735 if (nkl->dev) 736 ifindex = nkl->dev->ifindex; 737 rtnl_unlock(); 738 739 info->netkit.ifindex = ifindex; 740 info->netkit.attach_type = nkl->location; 741 return 0; 742 } 743 744 static int netkit_link_detach(struct bpf_link *link) 745 { 746 netkit_link_release(link); 747 return 0; 748 } 749 750 static const struct bpf_link_ops netkit_link_lops = { 751 .release = netkit_link_release, 752 .detach = netkit_link_detach, 753 .dealloc = netkit_link_dealloc, 754 .update_prog = netkit_link_update, 755 .show_fdinfo = netkit_link_fdinfo, 756 .fill_link_info = netkit_link_fill_info, 757 }; 758 759 static int netkit_link_init(struct netkit_link *nkl, 760 struct bpf_link_primer *link_primer, 761 const union bpf_attr *attr, 762 struct net_device *dev, 763 struct bpf_prog *prog) 764 { 765 bpf_link_init(&nkl->link, BPF_LINK_TYPE_NETKIT, 766 &netkit_link_lops, prog); 767 nkl->location = attr->link_create.attach_type; 768 nkl->dev = dev; 769 return bpf_link_prime(&nkl->link, link_primer); 770 } 771 772 int netkit_link_attach(const union bpf_attr *attr, struct bpf_prog *prog) 773 { 774 struct bpf_link_primer link_primer; 775 struct netkit_link *nkl; 776 struct net_device *dev; 777 int ret; 778 779 rtnl_lock(); 780 dev = netkit_dev_fetch(current->nsproxy->net_ns, 781 attr->link_create.target_ifindex, 782 attr->link_create.attach_type); 783 if (IS_ERR(dev)) { 784 ret = PTR_ERR(dev); 785 goto out; 786 } 787 nkl = kzalloc(sizeof(*nkl), GFP_KERNEL_ACCOUNT); 788 if (!nkl) { 789 ret = -ENOMEM; 790 goto out; 791 } 792 ret = netkit_link_init(nkl, &link_primer, attr, dev, prog); 793 if (ret) { 794 kfree(nkl); 795 goto out; 796 } 797 ret = netkit_link_prog_attach(&nkl->link, 798 attr->link_create.flags, 799 attr->link_create.netkit.relative_fd, 800 attr->link_create.netkit.expected_revision); 801 if (ret) { 802 nkl->dev = NULL; 803 bpf_link_cleanup(&link_primer); 804 goto out; 805 } 806 ret = bpf_link_settle(&link_primer); 807 out: 808 rtnl_unlock(); 809 return ret; 810 } 811 812 static void netkit_release_all(struct net_device *dev) 813 { 814 struct bpf_mprog_entry *entry; 815 struct bpf_tuple tuple = {}; 816 struct bpf_mprog_fp *fp; 817 struct bpf_mprog_cp *cp; 818 819 entry = netkit_entry_fetch(dev, false); 820 if (!entry) 821 return; 822 netkit_entry_update(dev, NULL); 823 netkit_entry_sync(); 824 bpf_mprog_foreach_tuple(entry, fp, cp, tuple) { 825 if (tuple.link) 826 netkit_link(tuple.link)->dev = NULL; 827 else 828 bpf_prog_put(tuple.prog); 829 } 830 } 831 832 static void netkit_uninit(struct net_device *dev) 833 { 834 netkit_release_all(dev); 835 } 836 837 static void netkit_del_link(struct net_device *dev, struct list_head *head) 838 { 839 struct netkit *nk = netkit_priv(dev); 840 struct net_device *peer = rtnl_dereference(nk->peer); 841 842 RCU_INIT_POINTER(nk->peer, NULL); 843 unregister_netdevice_queue(dev, head); 844 if (peer) { 845 nk = netkit_priv(peer); 846 RCU_INIT_POINTER(nk->peer, NULL); 847 unregister_netdevice_queue(peer, head); 848 } 849 } 850 851 static int netkit_change_link(struct net_device *dev, struct nlattr *tb[], 852 struct nlattr *data[], 853 struct netlink_ext_ack *extack) 854 { 855 struct netkit *nk = netkit_priv(dev); 856 struct net_device *peer = rtnl_dereference(nk->peer); 857 enum netkit_action policy; 858 struct nlattr *attr; 859 int err; 860 861 if (!nk->primary) { 862 NL_SET_ERR_MSG(extack, 863 "netkit link settings can be changed only through the primary device"); 864 return -EACCES; 865 } 866 867 if (data[IFLA_NETKIT_MODE]) { 868 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_NETKIT_MODE], 869 "netkit link operating mode cannot be changed after device creation"); 870 return -EACCES; 871 } 872 873 if (data[IFLA_NETKIT_PEER_INFO]) { 874 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_NETKIT_PEER_INFO], 875 "netkit peer info cannot be changed after device creation"); 876 return -EINVAL; 877 } 878 879 if (data[IFLA_NETKIT_POLICY]) { 880 attr = data[IFLA_NETKIT_POLICY]; 881 policy = nla_get_u32(attr); 882 err = netkit_check_policy(policy, attr, extack); 883 if (err) 884 return err; 885 WRITE_ONCE(nk->policy, policy); 886 } 887 888 if (data[IFLA_NETKIT_PEER_POLICY]) { 889 err = -EOPNOTSUPP; 890 attr = data[IFLA_NETKIT_PEER_POLICY]; 891 policy = nla_get_u32(attr); 892 if (peer) 893 err = netkit_check_policy(policy, attr, extack); 894 if (err) 895 return err; 896 nk = netkit_priv(peer); 897 WRITE_ONCE(nk->policy, policy); 898 } 899 900 return 0; 901 } 902 903 static size_t netkit_get_size(const struct net_device *dev) 904 { 905 return nla_total_size(sizeof(u32)) + /* IFLA_NETKIT_POLICY */ 906 nla_total_size(sizeof(u32)) + /* IFLA_NETKIT_PEER_POLICY */ 907 nla_total_size(sizeof(u8)) + /* IFLA_NETKIT_PRIMARY */ 908 nla_total_size(sizeof(u32)) + /* IFLA_NETKIT_MODE */ 909 0; 910 } 911 912 static int netkit_fill_info(struct sk_buff *skb, const struct net_device *dev) 913 { 914 struct netkit *nk = netkit_priv(dev); 915 struct net_device *peer = rtnl_dereference(nk->peer); 916 917 if (nla_put_u8(skb, IFLA_NETKIT_PRIMARY, nk->primary)) 918 return -EMSGSIZE; 919 if (nla_put_u32(skb, IFLA_NETKIT_POLICY, nk->policy)) 920 return -EMSGSIZE; 921 if (nla_put_u32(skb, IFLA_NETKIT_MODE, nk->mode)) 922 return -EMSGSIZE; 923 924 if (peer) { 925 nk = netkit_priv(peer); 926 if (nla_put_u32(skb, IFLA_NETKIT_PEER_POLICY, nk->policy)) 927 return -EMSGSIZE; 928 } 929 930 return 0; 931 } 932 933 static const struct nla_policy netkit_policy[IFLA_NETKIT_MAX + 1] = { 934 [IFLA_NETKIT_PEER_INFO] = { .len = sizeof(struct ifinfomsg) }, 935 [IFLA_NETKIT_POLICY] = { .type = NLA_U32 }, 936 [IFLA_NETKIT_MODE] = { .type = NLA_U32 }, 937 [IFLA_NETKIT_PEER_POLICY] = { .type = NLA_U32 }, 938 [IFLA_NETKIT_PRIMARY] = { .type = NLA_REJECT, 939 .reject_message = "Primary attribute is read-only" }, 940 }; 941 942 static struct rtnl_link_ops netkit_link_ops = { 943 .kind = DRV_NAME, 944 .priv_size = sizeof(struct netkit), 945 .setup = netkit_setup, 946 .newlink = netkit_new_link, 947 .dellink = netkit_del_link, 948 .changelink = netkit_change_link, 949 .get_link_net = netkit_get_link_net, 950 .get_size = netkit_get_size, 951 .fill_info = netkit_fill_info, 952 .policy = netkit_policy, 953 .validate = netkit_validate, 954 .maxtype = IFLA_NETKIT_MAX, 955 }; 956 957 static __init int netkit_init(void) 958 { 959 BUILD_BUG_ON((int)NETKIT_NEXT != (int)TCX_NEXT || 960 (int)NETKIT_PASS != (int)TCX_PASS || 961 (int)NETKIT_DROP != (int)TCX_DROP || 962 (int)NETKIT_REDIRECT != (int)TCX_REDIRECT); 963 964 return rtnl_link_register(&netkit_link_ops); 965 } 966 967 static __exit void netkit_exit(void) 968 { 969 rtnl_link_unregister(&netkit_link_ops); 970 } 971 972 module_init(netkit_init); 973 module_exit(netkit_exit); 974 975 MODULE_DESCRIPTION("BPF-programmable network device"); 976 MODULE_AUTHOR("Daniel Borkmann <daniel@iogearbox.net>"); 977 MODULE_AUTHOR("Nikolay Aleksandrov <razor@blackwall.org>"); 978 MODULE_LICENSE("GPL"); 979 MODULE_ALIAS_RTNL_LINK(DRV_NAME); 980