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