1 /* 2 * Copyright (C) 2017-2018 Netronome Systems, Inc. 3 * 4 * This software is licensed under the GNU General License Version 2, 5 * June 1991 as shown in the file COPYING in the top-level directory of this 6 * source tree. 7 * 8 * THE COPYRIGHT HOLDERS AND/OR OTHER PARTIES PROVIDE THE PROGRAM "AS IS" 9 * WITHOUT WARRANTY OF ANY KIND, EITHER EXPRESSED OR IMPLIED, INCLUDING, 10 * BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS 11 * FOR A PARTICULAR PURPOSE. THE ENTIRE RISK AS TO THE QUALITY AND PERFORMANCE 12 * OF THE PROGRAM IS WITH YOU. SHOULD THE PROGRAM PROVE DEFECTIVE, YOU ASSUME 13 * THE COST OF ALL NECESSARY SERVICING, REPAIR OR CORRECTION. 14 */ 15 16 #include <linux/bpf.h> 17 #include <linux/bpf_verifier.h> 18 #include <linux/bug.h> 19 #include <linux/kdev_t.h> 20 #include <linux/list.h> 21 #include <linux/lockdep.h> 22 #include <linux/netdevice.h> 23 #include <linux/printk.h> 24 #include <linux/proc_ns.h> 25 #include <linux/rhashtable.h> 26 #include <linux/rtnetlink.h> 27 #include <linux/rwsem.h> 28 29 /* Protects offdevs, members of bpf_offload_netdev and offload members 30 * of all progs. 31 * RTNL lock cannot be taken when holding this lock. 32 */ 33 static DECLARE_RWSEM(bpf_devs_lock); 34 35 struct bpf_offload_dev { 36 const struct bpf_prog_offload_ops *ops; 37 struct list_head netdevs; 38 void *priv; 39 }; 40 41 struct bpf_offload_netdev { 42 struct rhash_head l; 43 struct net_device *netdev; 44 struct bpf_offload_dev *offdev; 45 struct list_head progs; 46 struct list_head maps; 47 struct list_head offdev_netdevs; 48 }; 49 50 static const struct rhashtable_params offdevs_params = { 51 .nelem_hint = 4, 52 .key_len = sizeof(struct net_device *), 53 .key_offset = offsetof(struct bpf_offload_netdev, netdev), 54 .head_offset = offsetof(struct bpf_offload_netdev, l), 55 .automatic_shrinking = true, 56 }; 57 58 static struct rhashtable offdevs; 59 60 static int bpf_dev_offload_check(struct net_device *netdev) 61 { 62 if (!netdev) 63 return -EINVAL; 64 if (!netdev->netdev_ops->ndo_bpf) 65 return -EOPNOTSUPP; 66 return 0; 67 } 68 69 static struct bpf_offload_netdev * 70 bpf_offload_find_netdev(struct net_device *netdev) 71 { 72 lockdep_assert_held(&bpf_devs_lock); 73 74 return rhashtable_lookup_fast(&offdevs, &netdev, offdevs_params); 75 } 76 77 static int __bpf_offload_dev_netdev_register(struct bpf_offload_dev *offdev, 78 struct net_device *netdev) 79 { 80 struct bpf_offload_netdev *ondev; 81 int err; 82 83 ondev = kzalloc(sizeof(*ondev), GFP_KERNEL); 84 if (!ondev) 85 return -ENOMEM; 86 87 ondev->netdev = netdev; 88 ondev->offdev = offdev; 89 INIT_LIST_HEAD(&ondev->progs); 90 INIT_LIST_HEAD(&ondev->maps); 91 92 down_write(&bpf_devs_lock); 93 err = rhashtable_insert_fast(&offdevs, &ondev->l, offdevs_params); 94 if (err) { 95 netdev_warn(netdev, "failed to register for BPF offload\n"); 96 goto err_unlock_free; 97 } 98 99 list_add(&ondev->offdev_netdevs, &offdev->netdevs); 100 up_write(&bpf_devs_lock); 101 return 0; 102 103 err_unlock_free: 104 up_write(&bpf_devs_lock); 105 kfree(ondev); 106 return err; 107 } 108 109 static void __bpf_prog_offload_destroy(struct bpf_prog *prog) 110 { 111 struct bpf_prog_offload *offload = prog->aux->offload; 112 113 if (offload->dev_state) 114 offload->offdev->ops->destroy(prog); 115 116 /* Make sure BPF_PROG_GET_NEXT_ID can't find this dead program */ 117 bpf_prog_free_id(prog, true); 118 119 list_del_init(&offload->offloads); 120 kfree(offload); 121 prog->aux->offload = NULL; 122 } 123 124 static int bpf_map_offload_ndo(struct bpf_offloaded_map *offmap, 125 enum bpf_netdev_command cmd) 126 { 127 struct netdev_bpf data = {}; 128 struct net_device *netdev; 129 130 ASSERT_RTNL(); 131 132 data.command = cmd; 133 data.offmap = offmap; 134 /* Caller must make sure netdev is valid */ 135 netdev = offmap->netdev; 136 137 return netdev->netdev_ops->ndo_bpf(netdev, &data); 138 } 139 140 static void __bpf_map_offload_destroy(struct bpf_offloaded_map *offmap) 141 { 142 WARN_ON(bpf_map_offload_ndo(offmap, BPF_OFFLOAD_MAP_FREE)); 143 /* Make sure BPF_MAP_GET_NEXT_ID can't find this dead map */ 144 bpf_map_free_id(&offmap->map, true); 145 list_del_init(&offmap->offloads); 146 offmap->netdev = NULL; 147 } 148 149 static void __bpf_offload_dev_netdev_unregister(struct bpf_offload_dev *offdev, 150 struct net_device *netdev) 151 { 152 struct bpf_offload_netdev *ondev, *altdev; 153 struct bpf_offloaded_map *offmap, *mtmp; 154 struct bpf_prog_offload *offload, *ptmp; 155 156 ASSERT_RTNL(); 157 158 down_write(&bpf_devs_lock); 159 ondev = rhashtable_lookup_fast(&offdevs, &netdev, offdevs_params); 160 if (WARN_ON(!ondev)) 161 goto unlock; 162 163 WARN_ON(rhashtable_remove_fast(&offdevs, &ondev->l, offdevs_params)); 164 list_del(&ondev->offdev_netdevs); 165 166 /* Try to move the objects to another netdev of the device */ 167 altdev = list_first_entry_or_null(&offdev->netdevs, 168 struct bpf_offload_netdev, 169 offdev_netdevs); 170 if (altdev) { 171 list_for_each_entry(offload, &ondev->progs, offloads) 172 offload->netdev = altdev->netdev; 173 list_splice_init(&ondev->progs, &altdev->progs); 174 175 list_for_each_entry(offmap, &ondev->maps, offloads) 176 offmap->netdev = altdev->netdev; 177 list_splice_init(&ondev->maps, &altdev->maps); 178 } else { 179 list_for_each_entry_safe(offload, ptmp, &ondev->progs, offloads) 180 __bpf_prog_offload_destroy(offload->prog); 181 list_for_each_entry_safe(offmap, mtmp, &ondev->maps, offloads) 182 __bpf_map_offload_destroy(offmap); 183 } 184 185 WARN_ON(!list_empty(&ondev->progs)); 186 WARN_ON(!list_empty(&ondev->maps)); 187 kfree(ondev); 188 unlock: 189 up_write(&bpf_devs_lock); 190 } 191 192 int bpf_prog_offload_init(struct bpf_prog *prog, union bpf_attr *attr) 193 { 194 struct bpf_offload_netdev *ondev; 195 struct bpf_prog_offload *offload; 196 int err; 197 198 if (attr->prog_type != BPF_PROG_TYPE_SCHED_CLS && 199 attr->prog_type != BPF_PROG_TYPE_XDP) 200 return -EINVAL; 201 202 if (attr->prog_flags) 203 return -EINVAL; 204 205 offload = kzalloc(sizeof(*offload), GFP_USER); 206 if (!offload) 207 return -ENOMEM; 208 209 offload->prog = prog; 210 211 offload->netdev = dev_get_by_index(current->nsproxy->net_ns, 212 attr->prog_ifindex); 213 err = bpf_dev_offload_check(offload->netdev); 214 if (err) 215 goto err_maybe_put; 216 217 down_write(&bpf_devs_lock); 218 ondev = bpf_offload_find_netdev(offload->netdev); 219 if (!ondev) { 220 err = -EINVAL; 221 goto err_unlock; 222 } 223 offload->offdev = ondev->offdev; 224 prog->aux->offload = offload; 225 list_add_tail(&offload->offloads, &ondev->progs); 226 dev_put(offload->netdev); 227 up_write(&bpf_devs_lock); 228 229 return 0; 230 err_unlock: 231 up_write(&bpf_devs_lock); 232 err_maybe_put: 233 if (offload->netdev) 234 dev_put(offload->netdev); 235 kfree(offload); 236 return err; 237 } 238 239 int bpf_prog_offload_verifier_prep(struct bpf_prog *prog) 240 { 241 struct bpf_prog_offload *offload; 242 int ret = -ENODEV; 243 244 down_read(&bpf_devs_lock); 245 offload = prog->aux->offload; 246 if (offload) { 247 ret = offload->offdev->ops->prepare(prog); 248 offload->dev_state = !ret; 249 } 250 up_read(&bpf_devs_lock); 251 252 return ret; 253 } 254 255 int bpf_prog_offload_verify_insn(struct bpf_verifier_env *env, 256 int insn_idx, int prev_insn_idx) 257 { 258 struct bpf_prog_offload *offload; 259 int ret = -ENODEV; 260 261 down_read(&bpf_devs_lock); 262 offload = env->prog->aux->offload; 263 if (offload) 264 ret = offload->offdev->ops->insn_hook(env, insn_idx, 265 prev_insn_idx); 266 up_read(&bpf_devs_lock); 267 268 return ret; 269 } 270 271 int bpf_prog_offload_finalize(struct bpf_verifier_env *env) 272 { 273 struct bpf_prog_offload *offload; 274 int ret = -ENODEV; 275 276 down_read(&bpf_devs_lock); 277 offload = env->prog->aux->offload; 278 if (offload) { 279 if (offload->offdev->ops->finalize) 280 ret = offload->offdev->ops->finalize(env); 281 else 282 ret = 0; 283 } 284 up_read(&bpf_devs_lock); 285 286 return ret; 287 } 288 289 void 290 bpf_prog_offload_replace_insn(struct bpf_verifier_env *env, u32 off, 291 struct bpf_insn *insn) 292 { 293 const struct bpf_prog_offload_ops *ops; 294 struct bpf_prog_offload *offload; 295 int ret = -EOPNOTSUPP; 296 297 down_read(&bpf_devs_lock); 298 offload = env->prog->aux->offload; 299 if (offload) { 300 ops = offload->offdev->ops; 301 if (!offload->opt_failed && ops->replace_insn) 302 ret = ops->replace_insn(env, off, insn); 303 offload->opt_failed |= ret; 304 } 305 up_read(&bpf_devs_lock); 306 } 307 308 void 309 bpf_prog_offload_remove_insns(struct bpf_verifier_env *env, u32 off, u32 cnt) 310 { 311 struct bpf_prog_offload *offload; 312 int ret = -EOPNOTSUPP; 313 314 down_read(&bpf_devs_lock); 315 offload = env->prog->aux->offload; 316 if (offload) { 317 if (!offload->opt_failed && offload->offdev->ops->remove_insns) 318 ret = offload->offdev->ops->remove_insns(env, off, cnt); 319 offload->opt_failed |= ret; 320 } 321 up_read(&bpf_devs_lock); 322 } 323 324 void bpf_prog_offload_destroy(struct bpf_prog *prog) 325 { 326 down_write(&bpf_devs_lock); 327 if (prog->aux->offload) 328 __bpf_prog_offload_destroy(prog); 329 up_write(&bpf_devs_lock); 330 } 331 332 static int bpf_prog_offload_translate(struct bpf_prog *prog) 333 { 334 struct bpf_prog_offload *offload; 335 int ret = -ENODEV; 336 337 down_read(&bpf_devs_lock); 338 offload = prog->aux->offload; 339 if (offload) 340 ret = offload->offdev->ops->translate(prog); 341 up_read(&bpf_devs_lock); 342 343 return ret; 344 } 345 346 static unsigned int bpf_prog_warn_on_exec(const void *ctx, 347 const struct bpf_insn *insn) 348 { 349 WARN(1, "attempt to execute device eBPF program on the host!"); 350 return 0; 351 } 352 353 int bpf_prog_offload_compile(struct bpf_prog *prog) 354 { 355 prog->bpf_func = bpf_prog_warn_on_exec; 356 357 return bpf_prog_offload_translate(prog); 358 } 359 360 struct ns_get_path_bpf_prog_args { 361 struct bpf_prog *prog; 362 struct bpf_prog_info *info; 363 }; 364 365 static struct ns_common *bpf_prog_offload_info_fill_ns(void *private_data) 366 { 367 struct ns_get_path_bpf_prog_args *args = private_data; 368 struct bpf_prog_aux *aux = args->prog->aux; 369 struct ns_common *ns; 370 struct net *net; 371 372 rtnl_lock(); 373 down_read(&bpf_devs_lock); 374 375 if (aux->offload) { 376 args->info->ifindex = aux->offload->netdev->ifindex; 377 net = dev_net(aux->offload->netdev); 378 get_net(net); 379 ns = &net->ns; 380 } else { 381 args->info->ifindex = 0; 382 ns = NULL; 383 } 384 385 up_read(&bpf_devs_lock); 386 rtnl_unlock(); 387 388 return ns; 389 } 390 391 int bpf_prog_offload_info_fill(struct bpf_prog_info *info, 392 struct bpf_prog *prog) 393 { 394 struct ns_get_path_bpf_prog_args args = { 395 .prog = prog, 396 .info = info, 397 }; 398 struct bpf_prog_aux *aux = prog->aux; 399 struct inode *ns_inode; 400 struct path ns_path; 401 char __user *uinsns; 402 int res; 403 u32 ulen; 404 405 res = ns_get_path_cb(&ns_path, bpf_prog_offload_info_fill_ns, &args); 406 if (res) { 407 if (!info->ifindex) 408 return -ENODEV; 409 return res; 410 } 411 412 down_read(&bpf_devs_lock); 413 414 if (!aux->offload) { 415 up_read(&bpf_devs_lock); 416 return -ENODEV; 417 } 418 419 ulen = info->jited_prog_len; 420 info->jited_prog_len = aux->offload->jited_len; 421 if (info->jited_prog_len && ulen) { 422 uinsns = u64_to_user_ptr(info->jited_prog_insns); 423 ulen = min_t(u32, info->jited_prog_len, ulen); 424 if (copy_to_user(uinsns, aux->offload->jited_image, ulen)) { 425 up_read(&bpf_devs_lock); 426 return -EFAULT; 427 } 428 } 429 430 up_read(&bpf_devs_lock); 431 432 ns_inode = ns_path.dentry->d_inode; 433 info->netns_dev = new_encode_dev(ns_inode->i_sb->s_dev); 434 info->netns_ino = ns_inode->i_ino; 435 path_put(&ns_path); 436 437 return 0; 438 } 439 440 const struct bpf_prog_ops bpf_offload_prog_ops = { 441 }; 442 443 struct bpf_map *bpf_map_offload_map_alloc(union bpf_attr *attr) 444 { 445 struct net *net = current->nsproxy->net_ns; 446 struct bpf_offload_netdev *ondev; 447 struct bpf_offloaded_map *offmap; 448 int err; 449 450 if (!capable(CAP_SYS_ADMIN)) 451 return ERR_PTR(-EPERM); 452 if (attr->map_type != BPF_MAP_TYPE_ARRAY && 453 attr->map_type != BPF_MAP_TYPE_HASH) 454 return ERR_PTR(-EINVAL); 455 456 offmap = bpf_map_area_alloc(sizeof(*offmap), NUMA_NO_NODE); 457 if (!offmap) 458 return ERR_PTR(-ENOMEM); 459 460 bpf_map_init_from_attr(&offmap->map, attr); 461 462 rtnl_lock(); 463 down_write(&bpf_devs_lock); 464 offmap->netdev = __dev_get_by_index(net, attr->map_ifindex); 465 err = bpf_dev_offload_check(offmap->netdev); 466 if (err) 467 goto err_unlock; 468 469 ondev = bpf_offload_find_netdev(offmap->netdev); 470 if (!ondev) { 471 err = -EINVAL; 472 goto err_unlock; 473 } 474 475 err = bpf_map_offload_ndo(offmap, BPF_OFFLOAD_MAP_ALLOC); 476 if (err) 477 goto err_unlock; 478 479 list_add_tail(&offmap->offloads, &ondev->maps); 480 up_write(&bpf_devs_lock); 481 rtnl_unlock(); 482 483 return &offmap->map; 484 485 err_unlock: 486 up_write(&bpf_devs_lock); 487 rtnl_unlock(); 488 bpf_map_area_free(offmap); 489 return ERR_PTR(err); 490 } 491 492 void bpf_map_offload_map_free(struct bpf_map *map) 493 { 494 struct bpf_offloaded_map *offmap = map_to_offmap(map); 495 496 rtnl_lock(); 497 down_write(&bpf_devs_lock); 498 if (offmap->netdev) 499 __bpf_map_offload_destroy(offmap); 500 up_write(&bpf_devs_lock); 501 rtnl_unlock(); 502 503 bpf_map_area_free(offmap); 504 } 505 506 int bpf_map_offload_lookup_elem(struct bpf_map *map, void *key, void *value) 507 { 508 struct bpf_offloaded_map *offmap = map_to_offmap(map); 509 int ret = -ENODEV; 510 511 down_read(&bpf_devs_lock); 512 if (offmap->netdev) 513 ret = offmap->dev_ops->map_lookup_elem(offmap, key, value); 514 up_read(&bpf_devs_lock); 515 516 return ret; 517 } 518 519 int bpf_map_offload_update_elem(struct bpf_map *map, 520 void *key, void *value, u64 flags) 521 { 522 struct bpf_offloaded_map *offmap = map_to_offmap(map); 523 int ret = -ENODEV; 524 525 if (unlikely(flags > BPF_EXIST)) 526 return -EINVAL; 527 528 down_read(&bpf_devs_lock); 529 if (offmap->netdev) 530 ret = offmap->dev_ops->map_update_elem(offmap, key, value, 531 flags); 532 up_read(&bpf_devs_lock); 533 534 return ret; 535 } 536 537 int bpf_map_offload_delete_elem(struct bpf_map *map, void *key) 538 { 539 struct bpf_offloaded_map *offmap = map_to_offmap(map); 540 int ret = -ENODEV; 541 542 down_read(&bpf_devs_lock); 543 if (offmap->netdev) 544 ret = offmap->dev_ops->map_delete_elem(offmap, key); 545 up_read(&bpf_devs_lock); 546 547 return ret; 548 } 549 550 int bpf_map_offload_get_next_key(struct bpf_map *map, void *key, void *next_key) 551 { 552 struct bpf_offloaded_map *offmap = map_to_offmap(map); 553 int ret = -ENODEV; 554 555 down_read(&bpf_devs_lock); 556 if (offmap->netdev) 557 ret = offmap->dev_ops->map_get_next_key(offmap, key, next_key); 558 up_read(&bpf_devs_lock); 559 560 return ret; 561 } 562 563 struct ns_get_path_bpf_map_args { 564 struct bpf_offloaded_map *offmap; 565 struct bpf_map_info *info; 566 }; 567 568 static struct ns_common *bpf_map_offload_info_fill_ns(void *private_data) 569 { 570 struct ns_get_path_bpf_map_args *args = private_data; 571 struct ns_common *ns; 572 struct net *net; 573 574 rtnl_lock(); 575 down_read(&bpf_devs_lock); 576 577 if (args->offmap->netdev) { 578 args->info->ifindex = args->offmap->netdev->ifindex; 579 net = dev_net(args->offmap->netdev); 580 get_net(net); 581 ns = &net->ns; 582 } else { 583 args->info->ifindex = 0; 584 ns = NULL; 585 } 586 587 up_read(&bpf_devs_lock); 588 rtnl_unlock(); 589 590 return ns; 591 } 592 593 int bpf_map_offload_info_fill(struct bpf_map_info *info, struct bpf_map *map) 594 { 595 struct ns_get_path_bpf_map_args args = { 596 .offmap = map_to_offmap(map), 597 .info = info, 598 }; 599 struct inode *ns_inode; 600 struct path ns_path; 601 int res; 602 603 res = ns_get_path_cb(&ns_path, bpf_map_offload_info_fill_ns, &args); 604 if (res) { 605 if (!info->ifindex) 606 return -ENODEV; 607 return res; 608 } 609 610 ns_inode = ns_path.dentry->d_inode; 611 info->netns_dev = new_encode_dev(ns_inode->i_sb->s_dev); 612 info->netns_ino = ns_inode->i_ino; 613 path_put(&ns_path); 614 615 return 0; 616 } 617 618 static bool __bpf_offload_dev_match(struct bpf_prog *prog, 619 struct net_device *netdev) 620 { 621 struct bpf_offload_netdev *ondev1, *ondev2; 622 struct bpf_prog_offload *offload; 623 624 if (!bpf_prog_is_offloaded(prog->aux)) 625 return false; 626 627 offload = prog->aux->offload; 628 if (!offload) 629 return false; 630 if (offload->netdev == netdev) 631 return true; 632 633 ondev1 = bpf_offload_find_netdev(offload->netdev); 634 ondev2 = bpf_offload_find_netdev(netdev); 635 636 return ondev1 && ondev2 && ondev1->offdev == ondev2->offdev; 637 } 638 639 bool bpf_offload_dev_match(struct bpf_prog *prog, struct net_device *netdev) 640 { 641 bool ret; 642 643 down_read(&bpf_devs_lock); 644 ret = __bpf_offload_dev_match(prog, netdev); 645 up_read(&bpf_devs_lock); 646 647 return ret; 648 } 649 EXPORT_SYMBOL_GPL(bpf_offload_dev_match); 650 651 bool bpf_offload_prog_map_match(struct bpf_prog *prog, struct bpf_map *map) 652 { 653 struct bpf_offloaded_map *offmap; 654 bool ret; 655 656 if (!bpf_map_is_offloaded(map)) 657 return bpf_map_offload_neutral(map); 658 offmap = map_to_offmap(map); 659 660 down_read(&bpf_devs_lock); 661 ret = __bpf_offload_dev_match(prog, offmap->netdev); 662 up_read(&bpf_devs_lock); 663 664 return ret; 665 } 666 667 int bpf_offload_dev_netdev_register(struct bpf_offload_dev *offdev, 668 struct net_device *netdev) 669 { 670 return __bpf_offload_dev_netdev_register(offdev, netdev); 671 } 672 EXPORT_SYMBOL_GPL(bpf_offload_dev_netdev_register); 673 674 void bpf_offload_dev_netdev_unregister(struct bpf_offload_dev *offdev, 675 struct net_device *netdev) 676 { 677 __bpf_offload_dev_netdev_unregister(offdev, netdev); 678 } 679 EXPORT_SYMBOL_GPL(bpf_offload_dev_netdev_unregister); 680 681 struct bpf_offload_dev * 682 bpf_offload_dev_create(const struct bpf_prog_offload_ops *ops, void *priv) 683 { 684 struct bpf_offload_dev *offdev; 685 686 offdev = kzalloc(sizeof(*offdev), GFP_KERNEL); 687 if (!offdev) 688 return ERR_PTR(-ENOMEM); 689 690 offdev->ops = ops; 691 offdev->priv = priv; 692 INIT_LIST_HEAD(&offdev->netdevs); 693 694 return offdev; 695 } 696 EXPORT_SYMBOL_GPL(bpf_offload_dev_create); 697 698 void bpf_offload_dev_destroy(struct bpf_offload_dev *offdev) 699 { 700 WARN_ON(!list_empty(&offdev->netdevs)); 701 kfree(offdev); 702 } 703 EXPORT_SYMBOL_GPL(bpf_offload_dev_destroy); 704 705 void *bpf_offload_dev_priv(struct bpf_offload_dev *offdev) 706 { 707 return offdev->priv; 708 } 709 EXPORT_SYMBOL_GPL(bpf_offload_dev_priv); 710 711 static int __init bpf_offload_init(void) 712 { 713 return rhashtable_init(&offdevs, &offdevs_params); 714 } 715 716 late_initcall(bpf_offload_init); 717