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