1 // SPDX-License-Identifier: GPL-2.0-or-later 2 /* 3 * net/sched/cls_api.c Packet classifier API. 4 * 5 * Authors: Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru> 6 * 7 * Changes: 8 * 9 * Eduardo J. Blanco <ejbs@netlabs.com.uy> :990222: kmod support 10 */ 11 12 #include <linux/module.h> 13 #include <linux/types.h> 14 #include <linux/kernel.h> 15 #include <linux/string.h> 16 #include <linux/errno.h> 17 #include <linux/err.h> 18 #include <linux/skbuff.h> 19 #include <linux/init.h> 20 #include <linux/kmod.h> 21 #include <linux/slab.h> 22 #include <linux/idr.h> 23 #include <linux/jhash.h> 24 #include <linux/rculist.h> 25 #include <net/net_namespace.h> 26 #include <net/sock.h> 27 #include <net/netlink.h> 28 #include <net/pkt_sched.h> 29 #include <net/pkt_cls.h> 30 #include <net/tc_act/tc_pedit.h> 31 #include <net/tc_act/tc_mirred.h> 32 #include <net/tc_act/tc_vlan.h> 33 #include <net/tc_act/tc_tunnel_key.h> 34 #include <net/tc_act/tc_csum.h> 35 #include <net/tc_act/tc_gact.h> 36 #include <net/tc_act/tc_police.h> 37 #include <net/tc_act/tc_sample.h> 38 #include <net/tc_act/tc_skbedit.h> 39 #include <net/tc_act/tc_ct.h> 40 #include <net/tc_act/tc_mpls.h> 41 #include <net/tc_act/tc_gate.h> 42 #include <net/flow_offload.h> 43 44 extern const struct nla_policy rtm_tca_policy[TCA_MAX + 1]; 45 46 /* The list of all installed classifier types */ 47 static LIST_HEAD(tcf_proto_base); 48 49 /* Protects list of registered TC modules. It is pure SMP lock. */ 50 static DEFINE_RWLOCK(cls_mod_lock); 51 52 static u32 destroy_obj_hashfn(const struct tcf_proto *tp) 53 { 54 return jhash_3words(tp->chain->index, tp->prio, 55 (__force __u32)tp->protocol, 0); 56 } 57 58 static void tcf_proto_signal_destroying(struct tcf_chain *chain, 59 struct tcf_proto *tp) 60 { 61 struct tcf_block *block = chain->block; 62 63 mutex_lock(&block->proto_destroy_lock); 64 hash_add_rcu(block->proto_destroy_ht, &tp->destroy_ht_node, 65 destroy_obj_hashfn(tp)); 66 mutex_unlock(&block->proto_destroy_lock); 67 } 68 69 static bool tcf_proto_cmp(const struct tcf_proto *tp1, 70 const struct tcf_proto *tp2) 71 { 72 return tp1->chain->index == tp2->chain->index && 73 tp1->prio == tp2->prio && 74 tp1->protocol == tp2->protocol; 75 } 76 77 static bool tcf_proto_exists_destroying(struct tcf_chain *chain, 78 struct tcf_proto *tp) 79 { 80 u32 hash = destroy_obj_hashfn(tp); 81 struct tcf_proto *iter; 82 bool found = false; 83 84 rcu_read_lock(); 85 hash_for_each_possible_rcu(chain->block->proto_destroy_ht, iter, 86 destroy_ht_node, hash) { 87 if (tcf_proto_cmp(tp, iter)) { 88 found = true; 89 break; 90 } 91 } 92 rcu_read_unlock(); 93 94 return found; 95 } 96 97 static void 98 tcf_proto_signal_destroyed(struct tcf_chain *chain, struct tcf_proto *tp) 99 { 100 struct tcf_block *block = chain->block; 101 102 mutex_lock(&block->proto_destroy_lock); 103 if (hash_hashed(&tp->destroy_ht_node)) 104 hash_del_rcu(&tp->destroy_ht_node); 105 mutex_unlock(&block->proto_destroy_lock); 106 } 107 108 /* Find classifier type by string name */ 109 110 static const struct tcf_proto_ops *__tcf_proto_lookup_ops(const char *kind) 111 { 112 const struct tcf_proto_ops *t, *res = NULL; 113 114 if (kind) { 115 read_lock(&cls_mod_lock); 116 list_for_each_entry(t, &tcf_proto_base, head) { 117 if (strcmp(kind, t->kind) == 0) { 118 if (try_module_get(t->owner)) 119 res = t; 120 break; 121 } 122 } 123 read_unlock(&cls_mod_lock); 124 } 125 return res; 126 } 127 128 static const struct tcf_proto_ops * 129 tcf_proto_lookup_ops(const char *kind, bool rtnl_held, 130 struct netlink_ext_ack *extack) 131 { 132 const struct tcf_proto_ops *ops; 133 134 ops = __tcf_proto_lookup_ops(kind); 135 if (ops) 136 return ops; 137 #ifdef CONFIG_MODULES 138 if (rtnl_held) 139 rtnl_unlock(); 140 request_module("cls_%s", kind); 141 if (rtnl_held) 142 rtnl_lock(); 143 ops = __tcf_proto_lookup_ops(kind); 144 /* We dropped the RTNL semaphore in order to perform 145 * the module load. So, even if we succeeded in loading 146 * the module we have to replay the request. We indicate 147 * this using -EAGAIN. 148 */ 149 if (ops) { 150 module_put(ops->owner); 151 return ERR_PTR(-EAGAIN); 152 } 153 #endif 154 NL_SET_ERR_MSG(extack, "TC classifier not found"); 155 return ERR_PTR(-ENOENT); 156 } 157 158 /* Register(unregister) new classifier type */ 159 160 int register_tcf_proto_ops(struct tcf_proto_ops *ops) 161 { 162 struct tcf_proto_ops *t; 163 int rc = -EEXIST; 164 165 write_lock(&cls_mod_lock); 166 list_for_each_entry(t, &tcf_proto_base, head) 167 if (!strcmp(ops->kind, t->kind)) 168 goto out; 169 170 list_add_tail(&ops->head, &tcf_proto_base); 171 rc = 0; 172 out: 173 write_unlock(&cls_mod_lock); 174 return rc; 175 } 176 EXPORT_SYMBOL(register_tcf_proto_ops); 177 178 static struct workqueue_struct *tc_filter_wq; 179 180 int unregister_tcf_proto_ops(struct tcf_proto_ops *ops) 181 { 182 struct tcf_proto_ops *t; 183 int rc = -ENOENT; 184 185 /* Wait for outstanding call_rcu()s, if any, from a 186 * tcf_proto_ops's destroy() handler. 187 */ 188 rcu_barrier(); 189 flush_workqueue(tc_filter_wq); 190 191 write_lock(&cls_mod_lock); 192 list_for_each_entry(t, &tcf_proto_base, head) { 193 if (t == ops) { 194 list_del(&t->head); 195 rc = 0; 196 break; 197 } 198 } 199 write_unlock(&cls_mod_lock); 200 return rc; 201 } 202 EXPORT_SYMBOL(unregister_tcf_proto_ops); 203 204 bool tcf_queue_work(struct rcu_work *rwork, work_func_t func) 205 { 206 INIT_RCU_WORK(rwork, func); 207 return queue_rcu_work(tc_filter_wq, rwork); 208 } 209 EXPORT_SYMBOL(tcf_queue_work); 210 211 /* Select new prio value from the range, managed by kernel. */ 212 213 static inline u32 tcf_auto_prio(struct tcf_proto *tp) 214 { 215 u32 first = TC_H_MAKE(0xC0000000U, 0U); 216 217 if (tp) 218 first = tp->prio - 1; 219 220 return TC_H_MAJ(first); 221 } 222 223 static bool tcf_proto_check_kind(struct nlattr *kind, char *name) 224 { 225 if (kind) 226 return nla_strscpy(name, kind, IFNAMSIZ) < 0; 227 memset(name, 0, IFNAMSIZ); 228 return false; 229 } 230 231 static bool tcf_proto_is_unlocked(const char *kind) 232 { 233 const struct tcf_proto_ops *ops; 234 bool ret; 235 236 if (strlen(kind) == 0) 237 return false; 238 239 ops = tcf_proto_lookup_ops(kind, false, NULL); 240 /* On error return false to take rtnl lock. Proto lookup/create 241 * functions will perform lookup again and properly handle errors. 242 */ 243 if (IS_ERR(ops)) 244 return false; 245 246 ret = !!(ops->flags & TCF_PROTO_OPS_DOIT_UNLOCKED); 247 module_put(ops->owner); 248 return ret; 249 } 250 251 static struct tcf_proto *tcf_proto_create(const char *kind, u32 protocol, 252 u32 prio, struct tcf_chain *chain, 253 bool rtnl_held, 254 struct netlink_ext_ack *extack) 255 { 256 struct tcf_proto *tp; 257 int err; 258 259 tp = kzalloc(sizeof(*tp), GFP_KERNEL); 260 if (!tp) 261 return ERR_PTR(-ENOBUFS); 262 263 tp->ops = tcf_proto_lookup_ops(kind, rtnl_held, extack); 264 if (IS_ERR(tp->ops)) { 265 err = PTR_ERR(tp->ops); 266 goto errout; 267 } 268 tp->classify = tp->ops->classify; 269 tp->protocol = protocol; 270 tp->prio = prio; 271 tp->chain = chain; 272 spin_lock_init(&tp->lock); 273 refcount_set(&tp->refcnt, 1); 274 275 err = tp->ops->init(tp); 276 if (err) { 277 module_put(tp->ops->owner); 278 goto errout; 279 } 280 return tp; 281 282 errout: 283 kfree(tp); 284 return ERR_PTR(err); 285 } 286 287 static void tcf_proto_get(struct tcf_proto *tp) 288 { 289 refcount_inc(&tp->refcnt); 290 } 291 292 static void tcf_chain_put(struct tcf_chain *chain); 293 294 static void tcf_proto_destroy(struct tcf_proto *tp, bool rtnl_held, 295 bool sig_destroy, struct netlink_ext_ack *extack) 296 { 297 tp->ops->destroy(tp, rtnl_held, extack); 298 if (sig_destroy) 299 tcf_proto_signal_destroyed(tp->chain, tp); 300 tcf_chain_put(tp->chain); 301 module_put(tp->ops->owner); 302 kfree_rcu(tp, rcu); 303 } 304 305 static void tcf_proto_put(struct tcf_proto *tp, bool rtnl_held, 306 struct netlink_ext_ack *extack) 307 { 308 if (refcount_dec_and_test(&tp->refcnt)) 309 tcf_proto_destroy(tp, rtnl_held, true, extack); 310 } 311 312 static bool tcf_proto_check_delete(struct tcf_proto *tp) 313 { 314 if (tp->ops->delete_empty) 315 return tp->ops->delete_empty(tp); 316 317 tp->deleting = true; 318 return tp->deleting; 319 } 320 321 static void tcf_proto_mark_delete(struct tcf_proto *tp) 322 { 323 spin_lock(&tp->lock); 324 tp->deleting = true; 325 spin_unlock(&tp->lock); 326 } 327 328 static bool tcf_proto_is_deleting(struct tcf_proto *tp) 329 { 330 bool deleting; 331 332 spin_lock(&tp->lock); 333 deleting = tp->deleting; 334 spin_unlock(&tp->lock); 335 336 return deleting; 337 } 338 339 #define ASSERT_BLOCK_LOCKED(block) \ 340 lockdep_assert_held(&(block)->lock) 341 342 struct tcf_filter_chain_list_item { 343 struct list_head list; 344 tcf_chain_head_change_t *chain_head_change; 345 void *chain_head_change_priv; 346 }; 347 348 static struct tcf_chain *tcf_chain_create(struct tcf_block *block, 349 u32 chain_index) 350 { 351 struct tcf_chain *chain; 352 353 ASSERT_BLOCK_LOCKED(block); 354 355 chain = kzalloc(sizeof(*chain), GFP_KERNEL); 356 if (!chain) 357 return NULL; 358 list_add_tail_rcu(&chain->list, &block->chain_list); 359 mutex_init(&chain->filter_chain_lock); 360 chain->block = block; 361 chain->index = chain_index; 362 chain->refcnt = 1; 363 if (!chain->index) 364 block->chain0.chain = chain; 365 return chain; 366 } 367 368 static void tcf_chain_head_change_item(struct tcf_filter_chain_list_item *item, 369 struct tcf_proto *tp_head) 370 { 371 if (item->chain_head_change) 372 item->chain_head_change(tp_head, item->chain_head_change_priv); 373 } 374 375 static void tcf_chain0_head_change(struct tcf_chain *chain, 376 struct tcf_proto *tp_head) 377 { 378 struct tcf_filter_chain_list_item *item; 379 struct tcf_block *block = chain->block; 380 381 if (chain->index) 382 return; 383 384 mutex_lock(&block->lock); 385 list_for_each_entry(item, &block->chain0.filter_chain_list, list) 386 tcf_chain_head_change_item(item, tp_head); 387 mutex_unlock(&block->lock); 388 } 389 390 /* Returns true if block can be safely freed. */ 391 392 static bool tcf_chain_detach(struct tcf_chain *chain) 393 { 394 struct tcf_block *block = chain->block; 395 396 ASSERT_BLOCK_LOCKED(block); 397 398 list_del_rcu(&chain->list); 399 if (!chain->index) 400 block->chain0.chain = NULL; 401 402 if (list_empty(&block->chain_list) && 403 refcount_read(&block->refcnt) == 0) 404 return true; 405 406 return false; 407 } 408 409 static void tcf_block_destroy(struct tcf_block *block) 410 { 411 mutex_destroy(&block->lock); 412 mutex_destroy(&block->proto_destroy_lock); 413 kfree_rcu(block, rcu); 414 } 415 416 static void tcf_chain_destroy(struct tcf_chain *chain, bool free_block) 417 { 418 struct tcf_block *block = chain->block; 419 420 mutex_destroy(&chain->filter_chain_lock); 421 kfree_rcu(chain, rcu); 422 if (free_block) 423 tcf_block_destroy(block); 424 } 425 426 static void tcf_chain_hold(struct tcf_chain *chain) 427 { 428 ASSERT_BLOCK_LOCKED(chain->block); 429 430 ++chain->refcnt; 431 } 432 433 static bool tcf_chain_held_by_acts_only(struct tcf_chain *chain) 434 { 435 ASSERT_BLOCK_LOCKED(chain->block); 436 437 /* In case all the references are action references, this 438 * chain should not be shown to the user. 439 */ 440 return chain->refcnt == chain->action_refcnt; 441 } 442 443 static struct tcf_chain *tcf_chain_lookup(struct tcf_block *block, 444 u32 chain_index) 445 { 446 struct tcf_chain *chain; 447 448 ASSERT_BLOCK_LOCKED(block); 449 450 list_for_each_entry(chain, &block->chain_list, list) { 451 if (chain->index == chain_index) 452 return chain; 453 } 454 return NULL; 455 } 456 457 #if IS_ENABLED(CONFIG_NET_TC_SKB_EXT) 458 static struct tcf_chain *tcf_chain_lookup_rcu(const struct tcf_block *block, 459 u32 chain_index) 460 { 461 struct tcf_chain *chain; 462 463 list_for_each_entry_rcu(chain, &block->chain_list, list) { 464 if (chain->index == chain_index) 465 return chain; 466 } 467 return NULL; 468 } 469 #endif 470 471 static int tc_chain_notify(struct tcf_chain *chain, struct sk_buff *oskb, 472 u32 seq, u16 flags, int event, bool unicast); 473 474 static struct tcf_chain *__tcf_chain_get(struct tcf_block *block, 475 u32 chain_index, bool create, 476 bool by_act) 477 { 478 struct tcf_chain *chain = NULL; 479 bool is_first_reference; 480 481 mutex_lock(&block->lock); 482 chain = tcf_chain_lookup(block, chain_index); 483 if (chain) { 484 tcf_chain_hold(chain); 485 } else { 486 if (!create) 487 goto errout; 488 chain = tcf_chain_create(block, chain_index); 489 if (!chain) 490 goto errout; 491 } 492 493 if (by_act) 494 ++chain->action_refcnt; 495 is_first_reference = chain->refcnt - chain->action_refcnt == 1; 496 mutex_unlock(&block->lock); 497 498 /* Send notification only in case we got the first 499 * non-action reference. Until then, the chain acts only as 500 * a placeholder for actions pointing to it and user ought 501 * not know about them. 502 */ 503 if (is_first_reference && !by_act) 504 tc_chain_notify(chain, NULL, 0, NLM_F_CREATE | NLM_F_EXCL, 505 RTM_NEWCHAIN, false); 506 507 return chain; 508 509 errout: 510 mutex_unlock(&block->lock); 511 return chain; 512 } 513 514 static struct tcf_chain *tcf_chain_get(struct tcf_block *block, u32 chain_index, 515 bool create) 516 { 517 return __tcf_chain_get(block, chain_index, create, false); 518 } 519 520 struct tcf_chain *tcf_chain_get_by_act(struct tcf_block *block, u32 chain_index) 521 { 522 return __tcf_chain_get(block, chain_index, true, true); 523 } 524 EXPORT_SYMBOL(tcf_chain_get_by_act); 525 526 static void tc_chain_tmplt_del(const struct tcf_proto_ops *tmplt_ops, 527 void *tmplt_priv); 528 static int tc_chain_notify_delete(const struct tcf_proto_ops *tmplt_ops, 529 void *tmplt_priv, u32 chain_index, 530 struct tcf_block *block, struct sk_buff *oskb, 531 u32 seq, u16 flags, bool unicast); 532 533 static void __tcf_chain_put(struct tcf_chain *chain, bool by_act, 534 bool explicitly_created) 535 { 536 struct tcf_block *block = chain->block; 537 const struct tcf_proto_ops *tmplt_ops; 538 bool free_block = false; 539 unsigned int refcnt; 540 void *tmplt_priv; 541 542 mutex_lock(&block->lock); 543 if (explicitly_created) { 544 if (!chain->explicitly_created) { 545 mutex_unlock(&block->lock); 546 return; 547 } 548 chain->explicitly_created = false; 549 } 550 551 if (by_act) 552 chain->action_refcnt--; 553 554 /* tc_chain_notify_delete can't be called while holding block lock. 555 * However, when block is unlocked chain can be changed concurrently, so 556 * save these to temporary variables. 557 */ 558 refcnt = --chain->refcnt; 559 tmplt_ops = chain->tmplt_ops; 560 tmplt_priv = chain->tmplt_priv; 561 562 /* The last dropped non-action reference will trigger notification. */ 563 if (refcnt - chain->action_refcnt == 0 && !by_act) { 564 tc_chain_notify_delete(tmplt_ops, tmplt_priv, chain->index, 565 block, NULL, 0, 0, false); 566 /* Last reference to chain, no need to lock. */ 567 chain->flushing = false; 568 } 569 570 if (refcnt == 0) 571 free_block = tcf_chain_detach(chain); 572 mutex_unlock(&block->lock); 573 574 if (refcnt == 0) { 575 tc_chain_tmplt_del(tmplt_ops, tmplt_priv); 576 tcf_chain_destroy(chain, free_block); 577 } 578 } 579 580 static void tcf_chain_put(struct tcf_chain *chain) 581 { 582 __tcf_chain_put(chain, false, false); 583 } 584 585 void tcf_chain_put_by_act(struct tcf_chain *chain) 586 { 587 __tcf_chain_put(chain, true, false); 588 } 589 EXPORT_SYMBOL(tcf_chain_put_by_act); 590 591 static void tcf_chain_put_explicitly_created(struct tcf_chain *chain) 592 { 593 __tcf_chain_put(chain, false, true); 594 } 595 596 static void tcf_chain_flush(struct tcf_chain *chain, bool rtnl_held) 597 { 598 struct tcf_proto *tp, *tp_next; 599 600 mutex_lock(&chain->filter_chain_lock); 601 tp = tcf_chain_dereference(chain->filter_chain, chain); 602 while (tp) { 603 tp_next = rcu_dereference_protected(tp->next, 1); 604 tcf_proto_signal_destroying(chain, tp); 605 tp = tp_next; 606 } 607 tp = tcf_chain_dereference(chain->filter_chain, chain); 608 RCU_INIT_POINTER(chain->filter_chain, NULL); 609 tcf_chain0_head_change(chain, NULL); 610 chain->flushing = true; 611 mutex_unlock(&chain->filter_chain_lock); 612 613 while (tp) { 614 tp_next = rcu_dereference_protected(tp->next, 1); 615 tcf_proto_put(tp, rtnl_held, NULL); 616 tp = tp_next; 617 } 618 } 619 620 static int tcf_block_setup(struct tcf_block *block, 621 struct flow_block_offload *bo); 622 623 static void tcf_block_offload_init(struct flow_block_offload *bo, 624 struct net_device *dev, struct Qdisc *sch, 625 enum flow_block_command command, 626 enum flow_block_binder_type binder_type, 627 struct flow_block *flow_block, 628 bool shared, struct netlink_ext_ack *extack) 629 { 630 bo->net = dev_net(dev); 631 bo->command = command; 632 bo->binder_type = binder_type; 633 bo->block = flow_block; 634 bo->block_shared = shared; 635 bo->extack = extack; 636 bo->sch = sch; 637 bo->cb_list_head = &flow_block->cb_list; 638 INIT_LIST_HEAD(&bo->cb_list); 639 } 640 641 static void tcf_block_unbind(struct tcf_block *block, 642 struct flow_block_offload *bo); 643 644 static void tc_block_indr_cleanup(struct flow_block_cb *block_cb) 645 { 646 struct tcf_block *block = block_cb->indr.data; 647 struct net_device *dev = block_cb->indr.dev; 648 struct Qdisc *sch = block_cb->indr.sch; 649 struct netlink_ext_ack extack = {}; 650 struct flow_block_offload bo = {}; 651 652 tcf_block_offload_init(&bo, dev, sch, FLOW_BLOCK_UNBIND, 653 block_cb->indr.binder_type, 654 &block->flow_block, tcf_block_shared(block), 655 &extack); 656 rtnl_lock(); 657 down_write(&block->cb_lock); 658 list_del(&block_cb->driver_list); 659 list_move(&block_cb->list, &bo.cb_list); 660 tcf_block_unbind(block, &bo); 661 up_write(&block->cb_lock); 662 rtnl_unlock(); 663 } 664 665 static bool tcf_block_offload_in_use(struct tcf_block *block) 666 { 667 return atomic_read(&block->offloadcnt); 668 } 669 670 static int tcf_block_offload_cmd(struct tcf_block *block, 671 struct net_device *dev, struct Qdisc *sch, 672 struct tcf_block_ext_info *ei, 673 enum flow_block_command command, 674 struct netlink_ext_ack *extack) 675 { 676 struct flow_block_offload bo = {}; 677 678 tcf_block_offload_init(&bo, dev, sch, command, ei->binder_type, 679 &block->flow_block, tcf_block_shared(block), 680 extack); 681 682 if (dev->netdev_ops->ndo_setup_tc) { 683 int err; 684 685 err = dev->netdev_ops->ndo_setup_tc(dev, TC_SETUP_BLOCK, &bo); 686 if (err < 0) { 687 if (err != -EOPNOTSUPP) 688 NL_SET_ERR_MSG(extack, "Driver ndo_setup_tc failed"); 689 return err; 690 } 691 692 return tcf_block_setup(block, &bo); 693 } 694 695 flow_indr_dev_setup_offload(dev, sch, TC_SETUP_BLOCK, block, &bo, 696 tc_block_indr_cleanup); 697 tcf_block_setup(block, &bo); 698 699 return -EOPNOTSUPP; 700 } 701 702 static int tcf_block_offload_bind(struct tcf_block *block, struct Qdisc *q, 703 struct tcf_block_ext_info *ei, 704 struct netlink_ext_ack *extack) 705 { 706 struct net_device *dev = q->dev_queue->dev; 707 int err; 708 709 down_write(&block->cb_lock); 710 711 /* If tc offload feature is disabled and the block we try to bind 712 * to already has some offloaded filters, forbid to bind. 713 */ 714 if (dev->netdev_ops->ndo_setup_tc && 715 !tc_can_offload(dev) && 716 tcf_block_offload_in_use(block)) { 717 NL_SET_ERR_MSG(extack, "Bind to offloaded block failed as dev has offload disabled"); 718 err = -EOPNOTSUPP; 719 goto err_unlock; 720 } 721 722 err = tcf_block_offload_cmd(block, dev, q, ei, FLOW_BLOCK_BIND, extack); 723 if (err == -EOPNOTSUPP) 724 goto no_offload_dev_inc; 725 if (err) 726 goto err_unlock; 727 728 up_write(&block->cb_lock); 729 return 0; 730 731 no_offload_dev_inc: 732 if (tcf_block_offload_in_use(block)) 733 goto err_unlock; 734 735 err = 0; 736 block->nooffloaddevcnt++; 737 err_unlock: 738 up_write(&block->cb_lock); 739 return err; 740 } 741 742 static void tcf_block_offload_unbind(struct tcf_block *block, struct Qdisc *q, 743 struct tcf_block_ext_info *ei) 744 { 745 struct net_device *dev = q->dev_queue->dev; 746 int err; 747 748 down_write(&block->cb_lock); 749 err = tcf_block_offload_cmd(block, dev, q, ei, FLOW_BLOCK_UNBIND, NULL); 750 if (err == -EOPNOTSUPP) 751 goto no_offload_dev_dec; 752 up_write(&block->cb_lock); 753 return; 754 755 no_offload_dev_dec: 756 WARN_ON(block->nooffloaddevcnt-- == 0); 757 up_write(&block->cb_lock); 758 } 759 760 static int 761 tcf_chain0_head_change_cb_add(struct tcf_block *block, 762 struct tcf_block_ext_info *ei, 763 struct netlink_ext_ack *extack) 764 { 765 struct tcf_filter_chain_list_item *item; 766 struct tcf_chain *chain0; 767 768 item = kmalloc(sizeof(*item), GFP_KERNEL); 769 if (!item) { 770 NL_SET_ERR_MSG(extack, "Memory allocation for head change callback item failed"); 771 return -ENOMEM; 772 } 773 item->chain_head_change = ei->chain_head_change; 774 item->chain_head_change_priv = ei->chain_head_change_priv; 775 776 mutex_lock(&block->lock); 777 chain0 = block->chain0.chain; 778 if (chain0) 779 tcf_chain_hold(chain0); 780 else 781 list_add(&item->list, &block->chain0.filter_chain_list); 782 mutex_unlock(&block->lock); 783 784 if (chain0) { 785 struct tcf_proto *tp_head; 786 787 mutex_lock(&chain0->filter_chain_lock); 788 789 tp_head = tcf_chain_dereference(chain0->filter_chain, chain0); 790 if (tp_head) 791 tcf_chain_head_change_item(item, tp_head); 792 793 mutex_lock(&block->lock); 794 list_add(&item->list, &block->chain0.filter_chain_list); 795 mutex_unlock(&block->lock); 796 797 mutex_unlock(&chain0->filter_chain_lock); 798 tcf_chain_put(chain0); 799 } 800 801 return 0; 802 } 803 804 static void 805 tcf_chain0_head_change_cb_del(struct tcf_block *block, 806 struct tcf_block_ext_info *ei) 807 { 808 struct tcf_filter_chain_list_item *item; 809 810 mutex_lock(&block->lock); 811 list_for_each_entry(item, &block->chain0.filter_chain_list, list) { 812 if ((!ei->chain_head_change && !ei->chain_head_change_priv) || 813 (item->chain_head_change == ei->chain_head_change && 814 item->chain_head_change_priv == ei->chain_head_change_priv)) { 815 if (block->chain0.chain) 816 tcf_chain_head_change_item(item, NULL); 817 list_del(&item->list); 818 mutex_unlock(&block->lock); 819 820 kfree(item); 821 return; 822 } 823 } 824 mutex_unlock(&block->lock); 825 WARN_ON(1); 826 } 827 828 struct tcf_net { 829 spinlock_t idr_lock; /* Protects idr */ 830 struct idr idr; 831 }; 832 833 static unsigned int tcf_net_id; 834 835 static int tcf_block_insert(struct tcf_block *block, struct net *net, 836 struct netlink_ext_ack *extack) 837 { 838 struct tcf_net *tn = net_generic(net, tcf_net_id); 839 int err; 840 841 idr_preload(GFP_KERNEL); 842 spin_lock(&tn->idr_lock); 843 err = idr_alloc_u32(&tn->idr, block, &block->index, block->index, 844 GFP_NOWAIT); 845 spin_unlock(&tn->idr_lock); 846 idr_preload_end(); 847 848 return err; 849 } 850 851 static void tcf_block_remove(struct tcf_block *block, struct net *net) 852 { 853 struct tcf_net *tn = net_generic(net, tcf_net_id); 854 855 spin_lock(&tn->idr_lock); 856 idr_remove(&tn->idr, block->index); 857 spin_unlock(&tn->idr_lock); 858 } 859 860 static struct tcf_block *tcf_block_create(struct net *net, struct Qdisc *q, 861 u32 block_index, 862 struct netlink_ext_ack *extack) 863 { 864 struct tcf_block *block; 865 866 block = kzalloc(sizeof(*block), GFP_KERNEL); 867 if (!block) { 868 NL_SET_ERR_MSG(extack, "Memory allocation for block failed"); 869 return ERR_PTR(-ENOMEM); 870 } 871 mutex_init(&block->lock); 872 mutex_init(&block->proto_destroy_lock); 873 init_rwsem(&block->cb_lock); 874 flow_block_init(&block->flow_block); 875 INIT_LIST_HEAD(&block->chain_list); 876 INIT_LIST_HEAD(&block->owner_list); 877 INIT_LIST_HEAD(&block->chain0.filter_chain_list); 878 879 refcount_set(&block->refcnt, 1); 880 block->net = net; 881 block->index = block_index; 882 883 /* Don't store q pointer for blocks which are shared */ 884 if (!tcf_block_shared(block)) 885 block->q = q; 886 return block; 887 } 888 889 static struct tcf_block *tcf_block_lookup(struct net *net, u32 block_index) 890 { 891 struct tcf_net *tn = net_generic(net, tcf_net_id); 892 893 return idr_find(&tn->idr, block_index); 894 } 895 896 static struct tcf_block *tcf_block_refcnt_get(struct net *net, u32 block_index) 897 { 898 struct tcf_block *block; 899 900 rcu_read_lock(); 901 block = tcf_block_lookup(net, block_index); 902 if (block && !refcount_inc_not_zero(&block->refcnt)) 903 block = NULL; 904 rcu_read_unlock(); 905 906 return block; 907 } 908 909 static struct tcf_chain * 910 __tcf_get_next_chain(struct tcf_block *block, struct tcf_chain *chain) 911 { 912 mutex_lock(&block->lock); 913 if (chain) 914 chain = list_is_last(&chain->list, &block->chain_list) ? 915 NULL : list_next_entry(chain, list); 916 else 917 chain = list_first_entry_or_null(&block->chain_list, 918 struct tcf_chain, list); 919 920 /* skip all action-only chains */ 921 while (chain && tcf_chain_held_by_acts_only(chain)) 922 chain = list_is_last(&chain->list, &block->chain_list) ? 923 NULL : list_next_entry(chain, list); 924 925 if (chain) 926 tcf_chain_hold(chain); 927 mutex_unlock(&block->lock); 928 929 return chain; 930 } 931 932 /* Function to be used by all clients that want to iterate over all chains on 933 * block. It properly obtains block->lock and takes reference to chain before 934 * returning it. Users of this function must be tolerant to concurrent chain 935 * insertion/deletion or ensure that no concurrent chain modification is 936 * possible. Note that all netlink dump callbacks cannot guarantee to provide 937 * consistent dump because rtnl lock is released each time skb is filled with 938 * data and sent to user-space. 939 */ 940 941 struct tcf_chain * 942 tcf_get_next_chain(struct tcf_block *block, struct tcf_chain *chain) 943 { 944 struct tcf_chain *chain_next = __tcf_get_next_chain(block, chain); 945 946 if (chain) 947 tcf_chain_put(chain); 948 949 return chain_next; 950 } 951 EXPORT_SYMBOL(tcf_get_next_chain); 952 953 static struct tcf_proto * 954 __tcf_get_next_proto(struct tcf_chain *chain, struct tcf_proto *tp) 955 { 956 u32 prio = 0; 957 958 ASSERT_RTNL(); 959 mutex_lock(&chain->filter_chain_lock); 960 961 if (!tp) { 962 tp = tcf_chain_dereference(chain->filter_chain, chain); 963 } else if (tcf_proto_is_deleting(tp)) { 964 /* 'deleting' flag is set and chain->filter_chain_lock was 965 * unlocked, which means next pointer could be invalid. Restart 966 * search. 967 */ 968 prio = tp->prio + 1; 969 tp = tcf_chain_dereference(chain->filter_chain, chain); 970 971 for (; tp; tp = tcf_chain_dereference(tp->next, chain)) 972 if (!tp->deleting && tp->prio >= prio) 973 break; 974 } else { 975 tp = tcf_chain_dereference(tp->next, chain); 976 } 977 978 if (tp) 979 tcf_proto_get(tp); 980 981 mutex_unlock(&chain->filter_chain_lock); 982 983 return tp; 984 } 985 986 /* Function to be used by all clients that want to iterate over all tp's on 987 * chain. Users of this function must be tolerant to concurrent tp 988 * insertion/deletion or ensure that no concurrent chain modification is 989 * possible. Note that all netlink dump callbacks cannot guarantee to provide 990 * consistent dump because rtnl lock is released each time skb is filled with 991 * data and sent to user-space. 992 */ 993 994 struct tcf_proto * 995 tcf_get_next_proto(struct tcf_chain *chain, struct tcf_proto *tp) 996 { 997 struct tcf_proto *tp_next = __tcf_get_next_proto(chain, tp); 998 999 if (tp) 1000 tcf_proto_put(tp, true, NULL); 1001 1002 return tp_next; 1003 } 1004 EXPORT_SYMBOL(tcf_get_next_proto); 1005 1006 static void tcf_block_flush_all_chains(struct tcf_block *block, bool rtnl_held) 1007 { 1008 struct tcf_chain *chain; 1009 1010 /* Last reference to block. At this point chains cannot be added or 1011 * removed concurrently. 1012 */ 1013 for (chain = tcf_get_next_chain(block, NULL); 1014 chain; 1015 chain = tcf_get_next_chain(block, chain)) { 1016 tcf_chain_put_explicitly_created(chain); 1017 tcf_chain_flush(chain, rtnl_held); 1018 } 1019 } 1020 1021 /* Lookup Qdisc and increments its reference counter. 1022 * Set parent, if necessary. 1023 */ 1024 1025 static int __tcf_qdisc_find(struct net *net, struct Qdisc **q, 1026 u32 *parent, int ifindex, bool rtnl_held, 1027 struct netlink_ext_ack *extack) 1028 { 1029 const struct Qdisc_class_ops *cops; 1030 struct net_device *dev; 1031 int err = 0; 1032 1033 if (ifindex == TCM_IFINDEX_MAGIC_BLOCK) 1034 return 0; 1035 1036 rcu_read_lock(); 1037 1038 /* Find link */ 1039 dev = dev_get_by_index_rcu(net, ifindex); 1040 if (!dev) { 1041 rcu_read_unlock(); 1042 return -ENODEV; 1043 } 1044 1045 /* Find qdisc */ 1046 if (!*parent) { 1047 *q = dev->qdisc; 1048 *parent = (*q)->handle; 1049 } else { 1050 *q = qdisc_lookup_rcu(dev, TC_H_MAJ(*parent)); 1051 if (!*q) { 1052 NL_SET_ERR_MSG(extack, "Parent Qdisc doesn't exists"); 1053 err = -EINVAL; 1054 goto errout_rcu; 1055 } 1056 } 1057 1058 *q = qdisc_refcount_inc_nz(*q); 1059 if (!*q) { 1060 NL_SET_ERR_MSG(extack, "Parent Qdisc doesn't exists"); 1061 err = -EINVAL; 1062 goto errout_rcu; 1063 } 1064 1065 /* Is it classful? */ 1066 cops = (*q)->ops->cl_ops; 1067 if (!cops) { 1068 NL_SET_ERR_MSG(extack, "Qdisc not classful"); 1069 err = -EINVAL; 1070 goto errout_qdisc; 1071 } 1072 1073 if (!cops->tcf_block) { 1074 NL_SET_ERR_MSG(extack, "Class doesn't support blocks"); 1075 err = -EOPNOTSUPP; 1076 goto errout_qdisc; 1077 } 1078 1079 errout_rcu: 1080 /* At this point we know that qdisc is not noop_qdisc, 1081 * which means that qdisc holds a reference to net_device 1082 * and we hold a reference to qdisc, so it is safe to release 1083 * rcu read lock. 1084 */ 1085 rcu_read_unlock(); 1086 return err; 1087 1088 errout_qdisc: 1089 rcu_read_unlock(); 1090 1091 if (rtnl_held) 1092 qdisc_put(*q); 1093 else 1094 qdisc_put_unlocked(*q); 1095 *q = NULL; 1096 1097 return err; 1098 } 1099 1100 static int __tcf_qdisc_cl_find(struct Qdisc *q, u32 parent, unsigned long *cl, 1101 int ifindex, struct netlink_ext_ack *extack) 1102 { 1103 if (ifindex == TCM_IFINDEX_MAGIC_BLOCK) 1104 return 0; 1105 1106 /* Do we search for filter, attached to class? */ 1107 if (TC_H_MIN(parent)) { 1108 const struct Qdisc_class_ops *cops = q->ops->cl_ops; 1109 1110 *cl = cops->find(q, parent); 1111 if (*cl == 0) { 1112 NL_SET_ERR_MSG(extack, "Specified class doesn't exist"); 1113 return -ENOENT; 1114 } 1115 } 1116 1117 return 0; 1118 } 1119 1120 static struct tcf_block *__tcf_block_find(struct net *net, struct Qdisc *q, 1121 unsigned long cl, int ifindex, 1122 u32 block_index, 1123 struct netlink_ext_ack *extack) 1124 { 1125 struct tcf_block *block; 1126 1127 if (ifindex == TCM_IFINDEX_MAGIC_BLOCK) { 1128 block = tcf_block_refcnt_get(net, block_index); 1129 if (!block) { 1130 NL_SET_ERR_MSG(extack, "Block of given index was not found"); 1131 return ERR_PTR(-EINVAL); 1132 } 1133 } else { 1134 const struct Qdisc_class_ops *cops = q->ops->cl_ops; 1135 1136 block = cops->tcf_block(q, cl, extack); 1137 if (!block) 1138 return ERR_PTR(-EINVAL); 1139 1140 if (tcf_block_shared(block)) { 1141 NL_SET_ERR_MSG(extack, "This filter block is shared. Please use the block index to manipulate the filters"); 1142 return ERR_PTR(-EOPNOTSUPP); 1143 } 1144 1145 /* Always take reference to block in order to support execution 1146 * of rules update path of cls API without rtnl lock. Caller 1147 * must release block when it is finished using it. 'if' block 1148 * of this conditional obtain reference to block by calling 1149 * tcf_block_refcnt_get(). 1150 */ 1151 refcount_inc(&block->refcnt); 1152 } 1153 1154 return block; 1155 } 1156 1157 static void __tcf_block_put(struct tcf_block *block, struct Qdisc *q, 1158 struct tcf_block_ext_info *ei, bool rtnl_held) 1159 { 1160 if (refcount_dec_and_mutex_lock(&block->refcnt, &block->lock)) { 1161 /* Flushing/putting all chains will cause the block to be 1162 * deallocated when last chain is freed. However, if chain_list 1163 * is empty, block has to be manually deallocated. After block 1164 * reference counter reached 0, it is no longer possible to 1165 * increment it or add new chains to block. 1166 */ 1167 bool free_block = list_empty(&block->chain_list); 1168 1169 mutex_unlock(&block->lock); 1170 if (tcf_block_shared(block)) 1171 tcf_block_remove(block, block->net); 1172 1173 if (q) 1174 tcf_block_offload_unbind(block, q, ei); 1175 1176 if (free_block) 1177 tcf_block_destroy(block); 1178 else 1179 tcf_block_flush_all_chains(block, rtnl_held); 1180 } else if (q) { 1181 tcf_block_offload_unbind(block, q, ei); 1182 } 1183 } 1184 1185 static void tcf_block_refcnt_put(struct tcf_block *block, bool rtnl_held) 1186 { 1187 __tcf_block_put(block, NULL, NULL, rtnl_held); 1188 } 1189 1190 /* Find tcf block. 1191 * Set q, parent, cl when appropriate. 1192 */ 1193 1194 static struct tcf_block *tcf_block_find(struct net *net, struct Qdisc **q, 1195 u32 *parent, unsigned long *cl, 1196 int ifindex, u32 block_index, 1197 struct netlink_ext_ack *extack) 1198 { 1199 struct tcf_block *block; 1200 int err = 0; 1201 1202 ASSERT_RTNL(); 1203 1204 err = __tcf_qdisc_find(net, q, parent, ifindex, true, extack); 1205 if (err) 1206 goto errout; 1207 1208 err = __tcf_qdisc_cl_find(*q, *parent, cl, ifindex, extack); 1209 if (err) 1210 goto errout_qdisc; 1211 1212 block = __tcf_block_find(net, *q, *cl, ifindex, block_index, extack); 1213 if (IS_ERR(block)) { 1214 err = PTR_ERR(block); 1215 goto errout_qdisc; 1216 } 1217 1218 return block; 1219 1220 errout_qdisc: 1221 if (*q) 1222 qdisc_put(*q); 1223 errout: 1224 *q = NULL; 1225 return ERR_PTR(err); 1226 } 1227 1228 static void tcf_block_release(struct Qdisc *q, struct tcf_block *block, 1229 bool rtnl_held) 1230 { 1231 if (!IS_ERR_OR_NULL(block)) 1232 tcf_block_refcnt_put(block, rtnl_held); 1233 1234 if (q) { 1235 if (rtnl_held) 1236 qdisc_put(q); 1237 else 1238 qdisc_put_unlocked(q); 1239 } 1240 } 1241 1242 struct tcf_block_owner_item { 1243 struct list_head list; 1244 struct Qdisc *q; 1245 enum flow_block_binder_type binder_type; 1246 }; 1247 1248 static void 1249 tcf_block_owner_netif_keep_dst(struct tcf_block *block, 1250 struct Qdisc *q, 1251 enum flow_block_binder_type binder_type) 1252 { 1253 if (block->keep_dst && 1254 binder_type != FLOW_BLOCK_BINDER_TYPE_CLSACT_INGRESS && 1255 binder_type != FLOW_BLOCK_BINDER_TYPE_CLSACT_EGRESS) 1256 netif_keep_dst(qdisc_dev(q)); 1257 } 1258 1259 void tcf_block_netif_keep_dst(struct tcf_block *block) 1260 { 1261 struct tcf_block_owner_item *item; 1262 1263 block->keep_dst = true; 1264 list_for_each_entry(item, &block->owner_list, list) 1265 tcf_block_owner_netif_keep_dst(block, item->q, 1266 item->binder_type); 1267 } 1268 EXPORT_SYMBOL(tcf_block_netif_keep_dst); 1269 1270 static int tcf_block_owner_add(struct tcf_block *block, 1271 struct Qdisc *q, 1272 enum flow_block_binder_type binder_type) 1273 { 1274 struct tcf_block_owner_item *item; 1275 1276 item = kmalloc(sizeof(*item), GFP_KERNEL); 1277 if (!item) 1278 return -ENOMEM; 1279 item->q = q; 1280 item->binder_type = binder_type; 1281 list_add(&item->list, &block->owner_list); 1282 return 0; 1283 } 1284 1285 static void tcf_block_owner_del(struct tcf_block *block, 1286 struct Qdisc *q, 1287 enum flow_block_binder_type binder_type) 1288 { 1289 struct tcf_block_owner_item *item; 1290 1291 list_for_each_entry(item, &block->owner_list, list) { 1292 if (item->q == q && item->binder_type == binder_type) { 1293 list_del(&item->list); 1294 kfree(item); 1295 return; 1296 } 1297 } 1298 WARN_ON(1); 1299 } 1300 1301 int tcf_block_get_ext(struct tcf_block **p_block, struct Qdisc *q, 1302 struct tcf_block_ext_info *ei, 1303 struct netlink_ext_ack *extack) 1304 { 1305 struct net *net = qdisc_net(q); 1306 struct tcf_block *block = NULL; 1307 int err; 1308 1309 if (ei->block_index) 1310 /* block_index not 0 means the shared block is requested */ 1311 block = tcf_block_refcnt_get(net, ei->block_index); 1312 1313 if (!block) { 1314 block = tcf_block_create(net, q, ei->block_index, extack); 1315 if (IS_ERR(block)) 1316 return PTR_ERR(block); 1317 if (tcf_block_shared(block)) { 1318 err = tcf_block_insert(block, net, extack); 1319 if (err) 1320 goto err_block_insert; 1321 } 1322 } 1323 1324 err = tcf_block_owner_add(block, q, ei->binder_type); 1325 if (err) 1326 goto err_block_owner_add; 1327 1328 tcf_block_owner_netif_keep_dst(block, q, ei->binder_type); 1329 1330 err = tcf_chain0_head_change_cb_add(block, ei, extack); 1331 if (err) 1332 goto err_chain0_head_change_cb_add; 1333 1334 err = tcf_block_offload_bind(block, q, ei, extack); 1335 if (err) 1336 goto err_block_offload_bind; 1337 1338 *p_block = block; 1339 return 0; 1340 1341 err_block_offload_bind: 1342 tcf_chain0_head_change_cb_del(block, ei); 1343 err_chain0_head_change_cb_add: 1344 tcf_block_owner_del(block, q, ei->binder_type); 1345 err_block_owner_add: 1346 err_block_insert: 1347 tcf_block_refcnt_put(block, true); 1348 return err; 1349 } 1350 EXPORT_SYMBOL(tcf_block_get_ext); 1351 1352 static void tcf_chain_head_change_dflt(struct tcf_proto *tp_head, void *priv) 1353 { 1354 struct tcf_proto __rcu **p_filter_chain = priv; 1355 1356 rcu_assign_pointer(*p_filter_chain, tp_head); 1357 } 1358 1359 int tcf_block_get(struct tcf_block **p_block, 1360 struct tcf_proto __rcu **p_filter_chain, struct Qdisc *q, 1361 struct netlink_ext_ack *extack) 1362 { 1363 struct tcf_block_ext_info ei = { 1364 .chain_head_change = tcf_chain_head_change_dflt, 1365 .chain_head_change_priv = p_filter_chain, 1366 }; 1367 1368 WARN_ON(!p_filter_chain); 1369 return tcf_block_get_ext(p_block, q, &ei, extack); 1370 } 1371 EXPORT_SYMBOL(tcf_block_get); 1372 1373 /* XXX: Standalone actions are not allowed to jump to any chain, and bound 1374 * actions should be all removed after flushing. 1375 */ 1376 void tcf_block_put_ext(struct tcf_block *block, struct Qdisc *q, 1377 struct tcf_block_ext_info *ei) 1378 { 1379 if (!block) 1380 return; 1381 tcf_chain0_head_change_cb_del(block, ei); 1382 tcf_block_owner_del(block, q, ei->binder_type); 1383 1384 __tcf_block_put(block, q, ei, true); 1385 } 1386 EXPORT_SYMBOL(tcf_block_put_ext); 1387 1388 void tcf_block_put(struct tcf_block *block) 1389 { 1390 struct tcf_block_ext_info ei = {0, }; 1391 1392 if (!block) 1393 return; 1394 tcf_block_put_ext(block, block->q, &ei); 1395 } 1396 1397 EXPORT_SYMBOL(tcf_block_put); 1398 1399 static int 1400 tcf_block_playback_offloads(struct tcf_block *block, flow_setup_cb_t *cb, 1401 void *cb_priv, bool add, bool offload_in_use, 1402 struct netlink_ext_ack *extack) 1403 { 1404 struct tcf_chain *chain, *chain_prev; 1405 struct tcf_proto *tp, *tp_prev; 1406 int err; 1407 1408 lockdep_assert_held(&block->cb_lock); 1409 1410 for (chain = __tcf_get_next_chain(block, NULL); 1411 chain; 1412 chain_prev = chain, 1413 chain = __tcf_get_next_chain(block, chain), 1414 tcf_chain_put(chain_prev)) { 1415 for (tp = __tcf_get_next_proto(chain, NULL); tp; 1416 tp_prev = tp, 1417 tp = __tcf_get_next_proto(chain, tp), 1418 tcf_proto_put(tp_prev, true, NULL)) { 1419 if (tp->ops->reoffload) { 1420 err = tp->ops->reoffload(tp, add, cb, cb_priv, 1421 extack); 1422 if (err && add) 1423 goto err_playback_remove; 1424 } else if (add && offload_in_use) { 1425 err = -EOPNOTSUPP; 1426 NL_SET_ERR_MSG(extack, "Filter HW offload failed - classifier without re-offloading support"); 1427 goto err_playback_remove; 1428 } 1429 } 1430 } 1431 1432 return 0; 1433 1434 err_playback_remove: 1435 tcf_proto_put(tp, true, NULL); 1436 tcf_chain_put(chain); 1437 tcf_block_playback_offloads(block, cb, cb_priv, false, offload_in_use, 1438 extack); 1439 return err; 1440 } 1441 1442 static int tcf_block_bind(struct tcf_block *block, 1443 struct flow_block_offload *bo) 1444 { 1445 struct flow_block_cb *block_cb, *next; 1446 int err, i = 0; 1447 1448 lockdep_assert_held(&block->cb_lock); 1449 1450 list_for_each_entry(block_cb, &bo->cb_list, list) { 1451 err = tcf_block_playback_offloads(block, block_cb->cb, 1452 block_cb->cb_priv, true, 1453 tcf_block_offload_in_use(block), 1454 bo->extack); 1455 if (err) 1456 goto err_unroll; 1457 if (!bo->unlocked_driver_cb) 1458 block->lockeddevcnt++; 1459 1460 i++; 1461 } 1462 list_splice(&bo->cb_list, &block->flow_block.cb_list); 1463 1464 return 0; 1465 1466 err_unroll: 1467 list_for_each_entry_safe(block_cb, next, &bo->cb_list, list) { 1468 if (i-- > 0) { 1469 list_del(&block_cb->list); 1470 tcf_block_playback_offloads(block, block_cb->cb, 1471 block_cb->cb_priv, false, 1472 tcf_block_offload_in_use(block), 1473 NULL); 1474 if (!bo->unlocked_driver_cb) 1475 block->lockeddevcnt--; 1476 } 1477 flow_block_cb_free(block_cb); 1478 } 1479 1480 return err; 1481 } 1482 1483 static void tcf_block_unbind(struct tcf_block *block, 1484 struct flow_block_offload *bo) 1485 { 1486 struct flow_block_cb *block_cb, *next; 1487 1488 lockdep_assert_held(&block->cb_lock); 1489 1490 list_for_each_entry_safe(block_cb, next, &bo->cb_list, list) { 1491 tcf_block_playback_offloads(block, block_cb->cb, 1492 block_cb->cb_priv, false, 1493 tcf_block_offload_in_use(block), 1494 NULL); 1495 list_del(&block_cb->list); 1496 flow_block_cb_free(block_cb); 1497 if (!bo->unlocked_driver_cb) 1498 block->lockeddevcnt--; 1499 } 1500 } 1501 1502 static int tcf_block_setup(struct tcf_block *block, 1503 struct flow_block_offload *bo) 1504 { 1505 int err; 1506 1507 switch (bo->command) { 1508 case FLOW_BLOCK_BIND: 1509 err = tcf_block_bind(block, bo); 1510 break; 1511 case FLOW_BLOCK_UNBIND: 1512 err = 0; 1513 tcf_block_unbind(block, bo); 1514 break; 1515 default: 1516 WARN_ON_ONCE(1); 1517 err = -EOPNOTSUPP; 1518 } 1519 1520 return err; 1521 } 1522 1523 /* Main classifier routine: scans classifier chain attached 1524 * to this qdisc, (optionally) tests for protocol and asks 1525 * specific classifiers. 1526 */ 1527 static inline int __tcf_classify(struct sk_buff *skb, 1528 const struct tcf_proto *tp, 1529 const struct tcf_proto *orig_tp, 1530 struct tcf_result *res, 1531 bool compat_mode, 1532 u32 *last_executed_chain) 1533 { 1534 #ifdef CONFIG_NET_CLS_ACT 1535 const int max_reclassify_loop = 16; 1536 const struct tcf_proto *first_tp; 1537 int limit = 0; 1538 1539 reclassify: 1540 #endif 1541 for (; tp; tp = rcu_dereference_bh(tp->next)) { 1542 __be16 protocol = skb_protocol(skb, false); 1543 int err; 1544 1545 if (tp->protocol != protocol && 1546 tp->protocol != htons(ETH_P_ALL)) 1547 continue; 1548 1549 err = tp->classify(skb, tp, res); 1550 #ifdef CONFIG_NET_CLS_ACT 1551 if (unlikely(err == TC_ACT_RECLASSIFY && !compat_mode)) { 1552 first_tp = orig_tp; 1553 *last_executed_chain = first_tp->chain->index; 1554 goto reset; 1555 } else if (unlikely(TC_ACT_EXT_CMP(err, TC_ACT_GOTO_CHAIN))) { 1556 first_tp = res->goto_tp; 1557 *last_executed_chain = err & TC_ACT_EXT_VAL_MASK; 1558 goto reset; 1559 } 1560 #endif 1561 if (err >= 0) 1562 return err; 1563 } 1564 1565 return TC_ACT_UNSPEC; /* signal: continue lookup */ 1566 #ifdef CONFIG_NET_CLS_ACT 1567 reset: 1568 if (unlikely(limit++ >= max_reclassify_loop)) { 1569 net_notice_ratelimited("%u: reclassify loop, rule prio %u, protocol %02x\n", 1570 tp->chain->block->index, 1571 tp->prio & 0xffff, 1572 ntohs(tp->protocol)); 1573 return TC_ACT_SHOT; 1574 } 1575 1576 tp = first_tp; 1577 goto reclassify; 1578 #endif 1579 } 1580 1581 int tcf_classify(struct sk_buff *skb, 1582 const struct tcf_block *block, 1583 const struct tcf_proto *tp, 1584 struct tcf_result *res, bool compat_mode) 1585 { 1586 #if !IS_ENABLED(CONFIG_NET_TC_SKB_EXT) 1587 u32 last_executed_chain = 0; 1588 1589 return __tcf_classify(skb, tp, tp, res, compat_mode, 1590 &last_executed_chain); 1591 #else 1592 u32 last_executed_chain = tp ? tp->chain->index : 0; 1593 const struct tcf_proto *orig_tp = tp; 1594 struct tc_skb_ext *ext; 1595 int ret; 1596 1597 if (block) { 1598 ext = skb_ext_find(skb, TC_SKB_EXT); 1599 1600 if (ext && ext->chain) { 1601 struct tcf_chain *fchain; 1602 1603 fchain = tcf_chain_lookup_rcu(block, ext->chain); 1604 if (!fchain) 1605 return TC_ACT_SHOT; 1606 1607 /* Consume, so cloned/redirect skbs won't inherit ext */ 1608 skb_ext_del(skb, TC_SKB_EXT); 1609 1610 tp = rcu_dereference_bh(fchain->filter_chain); 1611 last_executed_chain = fchain->index; 1612 } 1613 } 1614 1615 ret = __tcf_classify(skb, tp, orig_tp, res, compat_mode, 1616 &last_executed_chain); 1617 1618 /* If we missed on some chain */ 1619 if (ret == TC_ACT_UNSPEC && last_executed_chain) { 1620 ext = tc_skb_ext_alloc(skb); 1621 if (WARN_ON_ONCE(!ext)) 1622 return TC_ACT_SHOT; 1623 ext->chain = last_executed_chain; 1624 ext->mru = qdisc_skb_cb(skb)->mru; 1625 ext->post_ct = qdisc_skb_cb(skb)->post_ct; 1626 } 1627 1628 return ret; 1629 #endif 1630 } 1631 EXPORT_SYMBOL(tcf_classify); 1632 1633 struct tcf_chain_info { 1634 struct tcf_proto __rcu **pprev; 1635 struct tcf_proto __rcu *next; 1636 }; 1637 1638 static struct tcf_proto *tcf_chain_tp_prev(struct tcf_chain *chain, 1639 struct tcf_chain_info *chain_info) 1640 { 1641 return tcf_chain_dereference(*chain_info->pprev, chain); 1642 } 1643 1644 static int tcf_chain_tp_insert(struct tcf_chain *chain, 1645 struct tcf_chain_info *chain_info, 1646 struct tcf_proto *tp) 1647 { 1648 if (chain->flushing) 1649 return -EAGAIN; 1650 1651 if (*chain_info->pprev == chain->filter_chain) 1652 tcf_chain0_head_change(chain, tp); 1653 tcf_proto_get(tp); 1654 RCU_INIT_POINTER(tp->next, tcf_chain_tp_prev(chain, chain_info)); 1655 rcu_assign_pointer(*chain_info->pprev, tp); 1656 1657 return 0; 1658 } 1659 1660 static void tcf_chain_tp_remove(struct tcf_chain *chain, 1661 struct tcf_chain_info *chain_info, 1662 struct tcf_proto *tp) 1663 { 1664 struct tcf_proto *next = tcf_chain_dereference(chain_info->next, chain); 1665 1666 tcf_proto_mark_delete(tp); 1667 if (tp == chain->filter_chain) 1668 tcf_chain0_head_change(chain, next); 1669 RCU_INIT_POINTER(*chain_info->pprev, next); 1670 } 1671 1672 static struct tcf_proto *tcf_chain_tp_find(struct tcf_chain *chain, 1673 struct tcf_chain_info *chain_info, 1674 u32 protocol, u32 prio, 1675 bool prio_allocate); 1676 1677 /* Try to insert new proto. 1678 * If proto with specified priority already exists, free new proto 1679 * and return existing one. 1680 */ 1681 1682 static struct tcf_proto *tcf_chain_tp_insert_unique(struct tcf_chain *chain, 1683 struct tcf_proto *tp_new, 1684 u32 protocol, u32 prio, 1685 bool rtnl_held) 1686 { 1687 struct tcf_chain_info chain_info; 1688 struct tcf_proto *tp; 1689 int err = 0; 1690 1691 mutex_lock(&chain->filter_chain_lock); 1692 1693 if (tcf_proto_exists_destroying(chain, tp_new)) { 1694 mutex_unlock(&chain->filter_chain_lock); 1695 tcf_proto_destroy(tp_new, rtnl_held, false, NULL); 1696 return ERR_PTR(-EAGAIN); 1697 } 1698 1699 tp = tcf_chain_tp_find(chain, &chain_info, 1700 protocol, prio, false); 1701 if (!tp) 1702 err = tcf_chain_tp_insert(chain, &chain_info, tp_new); 1703 mutex_unlock(&chain->filter_chain_lock); 1704 1705 if (tp) { 1706 tcf_proto_destroy(tp_new, rtnl_held, false, NULL); 1707 tp_new = tp; 1708 } else if (err) { 1709 tcf_proto_destroy(tp_new, rtnl_held, false, NULL); 1710 tp_new = ERR_PTR(err); 1711 } 1712 1713 return tp_new; 1714 } 1715 1716 static void tcf_chain_tp_delete_empty(struct tcf_chain *chain, 1717 struct tcf_proto *tp, bool rtnl_held, 1718 struct netlink_ext_ack *extack) 1719 { 1720 struct tcf_chain_info chain_info; 1721 struct tcf_proto *tp_iter; 1722 struct tcf_proto **pprev; 1723 struct tcf_proto *next; 1724 1725 mutex_lock(&chain->filter_chain_lock); 1726 1727 /* Atomically find and remove tp from chain. */ 1728 for (pprev = &chain->filter_chain; 1729 (tp_iter = tcf_chain_dereference(*pprev, chain)); 1730 pprev = &tp_iter->next) { 1731 if (tp_iter == tp) { 1732 chain_info.pprev = pprev; 1733 chain_info.next = tp_iter->next; 1734 WARN_ON(tp_iter->deleting); 1735 break; 1736 } 1737 } 1738 /* Verify that tp still exists and no new filters were inserted 1739 * concurrently. 1740 * Mark tp for deletion if it is empty. 1741 */ 1742 if (!tp_iter || !tcf_proto_check_delete(tp)) { 1743 mutex_unlock(&chain->filter_chain_lock); 1744 return; 1745 } 1746 1747 tcf_proto_signal_destroying(chain, tp); 1748 next = tcf_chain_dereference(chain_info.next, chain); 1749 if (tp == chain->filter_chain) 1750 tcf_chain0_head_change(chain, next); 1751 RCU_INIT_POINTER(*chain_info.pprev, next); 1752 mutex_unlock(&chain->filter_chain_lock); 1753 1754 tcf_proto_put(tp, rtnl_held, extack); 1755 } 1756 1757 static struct tcf_proto *tcf_chain_tp_find(struct tcf_chain *chain, 1758 struct tcf_chain_info *chain_info, 1759 u32 protocol, u32 prio, 1760 bool prio_allocate) 1761 { 1762 struct tcf_proto **pprev; 1763 struct tcf_proto *tp; 1764 1765 /* Check the chain for existence of proto-tcf with this priority */ 1766 for (pprev = &chain->filter_chain; 1767 (tp = tcf_chain_dereference(*pprev, chain)); 1768 pprev = &tp->next) { 1769 if (tp->prio >= prio) { 1770 if (tp->prio == prio) { 1771 if (prio_allocate || 1772 (tp->protocol != protocol && protocol)) 1773 return ERR_PTR(-EINVAL); 1774 } else { 1775 tp = NULL; 1776 } 1777 break; 1778 } 1779 } 1780 chain_info->pprev = pprev; 1781 if (tp) { 1782 chain_info->next = tp->next; 1783 tcf_proto_get(tp); 1784 } else { 1785 chain_info->next = NULL; 1786 } 1787 return tp; 1788 } 1789 1790 static int tcf_fill_node(struct net *net, struct sk_buff *skb, 1791 struct tcf_proto *tp, struct tcf_block *block, 1792 struct Qdisc *q, u32 parent, void *fh, 1793 u32 portid, u32 seq, u16 flags, int event, 1794 bool terse_dump, bool rtnl_held) 1795 { 1796 struct tcmsg *tcm; 1797 struct nlmsghdr *nlh; 1798 unsigned char *b = skb_tail_pointer(skb); 1799 1800 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags); 1801 if (!nlh) 1802 goto out_nlmsg_trim; 1803 tcm = nlmsg_data(nlh); 1804 tcm->tcm_family = AF_UNSPEC; 1805 tcm->tcm__pad1 = 0; 1806 tcm->tcm__pad2 = 0; 1807 if (q) { 1808 tcm->tcm_ifindex = qdisc_dev(q)->ifindex; 1809 tcm->tcm_parent = parent; 1810 } else { 1811 tcm->tcm_ifindex = TCM_IFINDEX_MAGIC_BLOCK; 1812 tcm->tcm_block_index = block->index; 1813 } 1814 tcm->tcm_info = TC_H_MAKE(tp->prio, tp->protocol); 1815 if (nla_put_string(skb, TCA_KIND, tp->ops->kind)) 1816 goto nla_put_failure; 1817 if (nla_put_u32(skb, TCA_CHAIN, tp->chain->index)) 1818 goto nla_put_failure; 1819 if (!fh) { 1820 tcm->tcm_handle = 0; 1821 } else if (terse_dump) { 1822 if (tp->ops->terse_dump) { 1823 if (tp->ops->terse_dump(net, tp, fh, skb, tcm, 1824 rtnl_held) < 0) 1825 goto nla_put_failure; 1826 } else { 1827 goto cls_op_not_supp; 1828 } 1829 } else { 1830 if (tp->ops->dump && 1831 tp->ops->dump(net, tp, fh, skb, tcm, rtnl_held) < 0) 1832 goto nla_put_failure; 1833 } 1834 nlh->nlmsg_len = skb_tail_pointer(skb) - b; 1835 return skb->len; 1836 1837 out_nlmsg_trim: 1838 nla_put_failure: 1839 cls_op_not_supp: 1840 nlmsg_trim(skb, b); 1841 return -1; 1842 } 1843 1844 static int tfilter_notify(struct net *net, struct sk_buff *oskb, 1845 struct nlmsghdr *n, struct tcf_proto *tp, 1846 struct tcf_block *block, struct Qdisc *q, 1847 u32 parent, void *fh, int event, bool unicast, 1848 bool rtnl_held) 1849 { 1850 struct sk_buff *skb; 1851 u32 portid = oskb ? NETLINK_CB(oskb).portid : 0; 1852 int err = 0; 1853 1854 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL); 1855 if (!skb) 1856 return -ENOBUFS; 1857 1858 if (tcf_fill_node(net, skb, tp, block, q, parent, fh, portid, 1859 n->nlmsg_seq, n->nlmsg_flags, event, 1860 false, rtnl_held) <= 0) { 1861 kfree_skb(skb); 1862 return -EINVAL; 1863 } 1864 1865 if (unicast) 1866 err = rtnl_unicast(skb, net, portid); 1867 else 1868 err = rtnetlink_send(skb, net, portid, RTNLGRP_TC, 1869 n->nlmsg_flags & NLM_F_ECHO); 1870 return err; 1871 } 1872 1873 static int tfilter_del_notify(struct net *net, struct sk_buff *oskb, 1874 struct nlmsghdr *n, struct tcf_proto *tp, 1875 struct tcf_block *block, struct Qdisc *q, 1876 u32 parent, void *fh, bool unicast, bool *last, 1877 bool rtnl_held, struct netlink_ext_ack *extack) 1878 { 1879 struct sk_buff *skb; 1880 u32 portid = oskb ? NETLINK_CB(oskb).portid : 0; 1881 int err; 1882 1883 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL); 1884 if (!skb) 1885 return -ENOBUFS; 1886 1887 if (tcf_fill_node(net, skb, tp, block, q, parent, fh, portid, 1888 n->nlmsg_seq, n->nlmsg_flags, RTM_DELTFILTER, 1889 false, rtnl_held) <= 0) { 1890 NL_SET_ERR_MSG(extack, "Failed to build del event notification"); 1891 kfree_skb(skb); 1892 return -EINVAL; 1893 } 1894 1895 err = tp->ops->delete(tp, fh, last, rtnl_held, extack); 1896 if (err) { 1897 kfree_skb(skb); 1898 return err; 1899 } 1900 1901 if (unicast) 1902 err = rtnl_unicast(skb, net, portid); 1903 else 1904 err = rtnetlink_send(skb, net, portid, RTNLGRP_TC, 1905 n->nlmsg_flags & NLM_F_ECHO); 1906 if (err < 0) 1907 NL_SET_ERR_MSG(extack, "Failed to send filter delete notification"); 1908 1909 return err; 1910 } 1911 1912 static void tfilter_notify_chain(struct net *net, struct sk_buff *oskb, 1913 struct tcf_block *block, struct Qdisc *q, 1914 u32 parent, struct nlmsghdr *n, 1915 struct tcf_chain *chain, int event) 1916 { 1917 struct tcf_proto *tp; 1918 1919 for (tp = tcf_get_next_proto(chain, NULL); 1920 tp; tp = tcf_get_next_proto(chain, tp)) 1921 tfilter_notify(net, oskb, n, tp, block, 1922 q, parent, NULL, event, false, true); 1923 } 1924 1925 static void tfilter_put(struct tcf_proto *tp, void *fh) 1926 { 1927 if (tp->ops->put && fh) 1928 tp->ops->put(tp, fh); 1929 } 1930 1931 static int tc_new_tfilter(struct sk_buff *skb, struct nlmsghdr *n, 1932 struct netlink_ext_ack *extack) 1933 { 1934 struct net *net = sock_net(skb->sk); 1935 struct nlattr *tca[TCA_MAX + 1]; 1936 char name[IFNAMSIZ]; 1937 struct tcmsg *t; 1938 u32 protocol; 1939 u32 prio; 1940 bool prio_allocate; 1941 u32 parent; 1942 u32 chain_index; 1943 struct Qdisc *q = NULL; 1944 struct tcf_chain_info chain_info; 1945 struct tcf_chain *chain = NULL; 1946 struct tcf_block *block; 1947 struct tcf_proto *tp; 1948 unsigned long cl; 1949 void *fh; 1950 int err; 1951 int tp_created; 1952 bool rtnl_held = false; 1953 u32 flags; 1954 1955 if (!netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN)) 1956 return -EPERM; 1957 1958 replay: 1959 tp_created = 0; 1960 1961 err = nlmsg_parse_deprecated(n, sizeof(*t), tca, TCA_MAX, 1962 rtm_tca_policy, extack); 1963 if (err < 0) 1964 return err; 1965 1966 t = nlmsg_data(n); 1967 protocol = TC_H_MIN(t->tcm_info); 1968 prio = TC_H_MAJ(t->tcm_info); 1969 prio_allocate = false; 1970 parent = t->tcm_parent; 1971 tp = NULL; 1972 cl = 0; 1973 block = NULL; 1974 flags = 0; 1975 1976 if (prio == 0) { 1977 /* If no priority is provided by the user, 1978 * we allocate one. 1979 */ 1980 if (n->nlmsg_flags & NLM_F_CREATE) { 1981 prio = TC_H_MAKE(0x80000000U, 0U); 1982 prio_allocate = true; 1983 } else { 1984 NL_SET_ERR_MSG(extack, "Invalid filter command with priority of zero"); 1985 return -ENOENT; 1986 } 1987 } 1988 1989 /* Find head of filter chain. */ 1990 1991 err = __tcf_qdisc_find(net, &q, &parent, t->tcm_ifindex, false, extack); 1992 if (err) 1993 return err; 1994 1995 if (tcf_proto_check_kind(tca[TCA_KIND], name)) { 1996 NL_SET_ERR_MSG(extack, "Specified TC filter name too long"); 1997 err = -EINVAL; 1998 goto errout; 1999 } 2000 2001 /* Take rtnl mutex if rtnl_held was set to true on previous iteration, 2002 * block is shared (no qdisc found), qdisc is not unlocked, classifier 2003 * type is not specified, classifier is not unlocked. 2004 */ 2005 if (rtnl_held || 2006 (q && !(q->ops->cl_ops->flags & QDISC_CLASS_OPS_DOIT_UNLOCKED)) || 2007 !tcf_proto_is_unlocked(name)) { 2008 rtnl_held = true; 2009 rtnl_lock(); 2010 } 2011 2012 err = __tcf_qdisc_cl_find(q, parent, &cl, t->tcm_ifindex, extack); 2013 if (err) 2014 goto errout; 2015 2016 block = __tcf_block_find(net, q, cl, t->tcm_ifindex, t->tcm_block_index, 2017 extack); 2018 if (IS_ERR(block)) { 2019 err = PTR_ERR(block); 2020 goto errout; 2021 } 2022 block->classid = parent; 2023 2024 chain_index = tca[TCA_CHAIN] ? nla_get_u32(tca[TCA_CHAIN]) : 0; 2025 if (chain_index > TC_ACT_EXT_VAL_MASK) { 2026 NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit"); 2027 err = -EINVAL; 2028 goto errout; 2029 } 2030 chain = tcf_chain_get(block, chain_index, true); 2031 if (!chain) { 2032 NL_SET_ERR_MSG(extack, "Cannot create specified filter chain"); 2033 err = -ENOMEM; 2034 goto errout; 2035 } 2036 2037 mutex_lock(&chain->filter_chain_lock); 2038 tp = tcf_chain_tp_find(chain, &chain_info, protocol, 2039 prio, prio_allocate); 2040 if (IS_ERR(tp)) { 2041 NL_SET_ERR_MSG(extack, "Filter with specified priority/protocol not found"); 2042 err = PTR_ERR(tp); 2043 goto errout_locked; 2044 } 2045 2046 if (tp == NULL) { 2047 struct tcf_proto *tp_new = NULL; 2048 2049 if (chain->flushing) { 2050 err = -EAGAIN; 2051 goto errout_locked; 2052 } 2053 2054 /* Proto-tcf does not exist, create new one */ 2055 2056 if (tca[TCA_KIND] == NULL || !protocol) { 2057 NL_SET_ERR_MSG(extack, "Filter kind and protocol must be specified"); 2058 err = -EINVAL; 2059 goto errout_locked; 2060 } 2061 2062 if (!(n->nlmsg_flags & NLM_F_CREATE)) { 2063 NL_SET_ERR_MSG(extack, "Need both RTM_NEWTFILTER and NLM_F_CREATE to create a new filter"); 2064 err = -ENOENT; 2065 goto errout_locked; 2066 } 2067 2068 if (prio_allocate) 2069 prio = tcf_auto_prio(tcf_chain_tp_prev(chain, 2070 &chain_info)); 2071 2072 mutex_unlock(&chain->filter_chain_lock); 2073 tp_new = tcf_proto_create(name, protocol, prio, chain, 2074 rtnl_held, extack); 2075 if (IS_ERR(tp_new)) { 2076 err = PTR_ERR(tp_new); 2077 goto errout_tp; 2078 } 2079 2080 tp_created = 1; 2081 tp = tcf_chain_tp_insert_unique(chain, tp_new, protocol, prio, 2082 rtnl_held); 2083 if (IS_ERR(tp)) { 2084 err = PTR_ERR(tp); 2085 goto errout_tp; 2086 } 2087 } else { 2088 mutex_unlock(&chain->filter_chain_lock); 2089 } 2090 2091 if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind)) { 2092 NL_SET_ERR_MSG(extack, "Specified filter kind does not match existing one"); 2093 err = -EINVAL; 2094 goto errout; 2095 } 2096 2097 fh = tp->ops->get(tp, t->tcm_handle); 2098 2099 if (!fh) { 2100 if (!(n->nlmsg_flags & NLM_F_CREATE)) { 2101 NL_SET_ERR_MSG(extack, "Need both RTM_NEWTFILTER and NLM_F_CREATE to create a new filter"); 2102 err = -ENOENT; 2103 goto errout; 2104 } 2105 } else if (n->nlmsg_flags & NLM_F_EXCL) { 2106 tfilter_put(tp, fh); 2107 NL_SET_ERR_MSG(extack, "Filter already exists"); 2108 err = -EEXIST; 2109 goto errout; 2110 } 2111 2112 if (chain->tmplt_ops && chain->tmplt_ops != tp->ops) { 2113 NL_SET_ERR_MSG(extack, "Chain template is set to a different filter kind"); 2114 err = -EINVAL; 2115 goto errout; 2116 } 2117 2118 if (!(n->nlmsg_flags & NLM_F_CREATE)) 2119 flags |= TCA_ACT_FLAGS_REPLACE; 2120 if (!rtnl_held) 2121 flags |= TCA_ACT_FLAGS_NO_RTNL; 2122 err = tp->ops->change(net, skb, tp, cl, t->tcm_handle, tca, &fh, 2123 flags, extack); 2124 if (err == 0) { 2125 tfilter_notify(net, skb, n, tp, block, q, parent, fh, 2126 RTM_NEWTFILTER, false, rtnl_held); 2127 tfilter_put(tp, fh); 2128 /* q pointer is NULL for shared blocks */ 2129 if (q) 2130 q->flags &= ~TCQ_F_CAN_BYPASS; 2131 } 2132 2133 errout: 2134 if (err && tp_created) 2135 tcf_chain_tp_delete_empty(chain, tp, rtnl_held, NULL); 2136 errout_tp: 2137 if (chain) { 2138 if (tp && !IS_ERR(tp)) 2139 tcf_proto_put(tp, rtnl_held, NULL); 2140 if (!tp_created) 2141 tcf_chain_put(chain); 2142 } 2143 tcf_block_release(q, block, rtnl_held); 2144 2145 if (rtnl_held) 2146 rtnl_unlock(); 2147 2148 if (err == -EAGAIN) { 2149 /* Take rtnl lock in case EAGAIN is caused by concurrent flush 2150 * of target chain. 2151 */ 2152 rtnl_held = true; 2153 /* Replay the request. */ 2154 goto replay; 2155 } 2156 return err; 2157 2158 errout_locked: 2159 mutex_unlock(&chain->filter_chain_lock); 2160 goto errout; 2161 } 2162 2163 static int tc_del_tfilter(struct sk_buff *skb, struct nlmsghdr *n, 2164 struct netlink_ext_ack *extack) 2165 { 2166 struct net *net = sock_net(skb->sk); 2167 struct nlattr *tca[TCA_MAX + 1]; 2168 char name[IFNAMSIZ]; 2169 struct tcmsg *t; 2170 u32 protocol; 2171 u32 prio; 2172 u32 parent; 2173 u32 chain_index; 2174 struct Qdisc *q = NULL; 2175 struct tcf_chain_info chain_info; 2176 struct tcf_chain *chain = NULL; 2177 struct tcf_block *block = NULL; 2178 struct tcf_proto *tp = NULL; 2179 unsigned long cl = 0; 2180 void *fh = NULL; 2181 int err; 2182 bool rtnl_held = false; 2183 2184 if (!netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN)) 2185 return -EPERM; 2186 2187 err = nlmsg_parse_deprecated(n, sizeof(*t), tca, TCA_MAX, 2188 rtm_tca_policy, extack); 2189 if (err < 0) 2190 return err; 2191 2192 t = nlmsg_data(n); 2193 protocol = TC_H_MIN(t->tcm_info); 2194 prio = TC_H_MAJ(t->tcm_info); 2195 parent = t->tcm_parent; 2196 2197 if (prio == 0 && (protocol || t->tcm_handle || tca[TCA_KIND])) { 2198 NL_SET_ERR_MSG(extack, "Cannot flush filters with protocol, handle or kind set"); 2199 return -ENOENT; 2200 } 2201 2202 /* Find head of filter chain. */ 2203 2204 err = __tcf_qdisc_find(net, &q, &parent, t->tcm_ifindex, false, extack); 2205 if (err) 2206 return err; 2207 2208 if (tcf_proto_check_kind(tca[TCA_KIND], name)) { 2209 NL_SET_ERR_MSG(extack, "Specified TC filter name too long"); 2210 err = -EINVAL; 2211 goto errout; 2212 } 2213 /* Take rtnl mutex if flushing whole chain, block is shared (no qdisc 2214 * found), qdisc is not unlocked, classifier type is not specified, 2215 * classifier is not unlocked. 2216 */ 2217 if (!prio || 2218 (q && !(q->ops->cl_ops->flags & QDISC_CLASS_OPS_DOIT_UNLOCKED)) || 2219 !tcf_proto_is_unlocked(name)) { 2220 rtnl_held = true; 2221 rtnl_lock(); 2222 } 2223 2224 err = __tcf_qdisc_cl_find(q, parent, &cl, t->tcm_ifindex, extack); 2225 if (err) 2226 goto errout; 2227 2228 block = __tcf_block_find(net, q, cl, t->tcm_ifindex, t->tcm_block_index, 2229 extack); 2230 if (IS_ERR(block)) { 2231 err = PTR_ERR(block); 2232 goto errout; 2233 } 2234 2235 chain_index = tca[TCA_CHAIN] ? nla_get_u32(tca[TCA_CHAIN]) : 0; 2236 if (chain_index > TC_ACT_EXT_VAL_MASK) { 2237 NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit"); 2238 err = -EINVAL; 2239 goto errout; 2240 } 2241 chain = tcf_chain_get(block, chain_index, false); 2242 if (!chain) { 2243 /* User requested flush on non-existent chain. Nothing to do, 2244 * so just return success. 2245 */ 2246 if (prio == 0) { 2247 err = 0; 2248 goto errout; 2249 } 2250 NL_SET_ERR_MSG(extack, "Cannot find specified filter chain"); 2251 err = -ENOENT; 2252 goto errout; 2253 } 2254 2255 if (prio == 0) { 2256 tfilter_notify_chain(net, skb, block, q, parent, n, 2257 chain, RTM_DELTFILTER); 2258 tcf_chain_flush(chain, rtnl_held); 2259 err = 0; 2260 goto errout; 2261 } 2262 2263 mutex_lock(&chain->filter_chain_lock); 2264 tp = tcf_chain_tp_find(chain, &chain_info, protocol, 2265 prio, false); 2266 if (!tp || IS_ERR(tp)) { 2267 NL_SET_ERR_MSG(extack, "Filter with specified priority/protocol not found"); 2268 err = tp ? PTR_ERR(tp) : -ENOENT; 2269 goto errout_locked; 2270 } else if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind)) { 2271 NL_SET_ERR_MSG(extack, "Specified filter kind does not match existing one"); 2272 err = -EINVAL; 2273 goto errout_locked; 2274 } else if (t->tcm_handle == 0) { 2275 tcf_proto_signal_destroying(chain, tp); 2276 tcf_chain_tp_remove(chain, &chain_info, tp); 2277 mutex_unlock(&chain->filter_chain_lock); 2278 2279 tcf_proto_put(tp, rtnl_held, NULL); 2280 tfilter_notify(net, skb, n, tp, block, q, parent, fh, 2281 RTM_DELTFILTER, false, rtnl_held); 2282 err = 0; 2283 goto errout; 2284 } 2285 mutex_unlock(&chain->filter_chain_lock); 2286 2287 fh = tp->ops->get(tp, t->tcm_handle); 2288 2289 if (!fh) { 2290 NL_SET_ERR_MSG(extack, "Specified filter handle not found"); 2291 err = -ENOENT; 2292 } else { 2293 bool last; 2294 2295 err = tfilter_del_notify(net, skb, n, tp, block, 2296 q, parent, fh, false, &last, 2297 rtnl_held, extack); 2298 2299 if (err) 2300 goto errout; 2301 if (last) 2302 tcf_chain_tp_delete_empty(chain, tp, rtnl_held, extack); 2303 } 2304 2305 errout: 2306 if (chain) { 2307 if (tp && !IS_ERR(tp)) 2308 tcf_proto_put(tp, rtnl_held, NULL); 2309 tcf_chain_put(chain); 2310 } 2311 tcf_block_release(q, block, rtnl_held); 2312 2313 if (rtnl_held) 2314 rtnl_unlock(); 2315 2316 return err; 2317 2318 errout_locked: 2319 mutex_unlock(&chain->filter_chain_lock); 2320 goto errout; 2321 } 2322 2323 static int tc_get_tfilter(struct sk_buff *skb, struct nlmsghdr *n, 2324 struct netlink_ext_ack *extack) 2325 { 2326 struct net *net = sock_net(skb->sk); 2327 struct nlattr *tca[TCA_MAX + 1]; 2328 char name[IFNAMSIZ]; 2329 struct tcmsg *t; 2330 u32 protocol; 2331 u32 prio; 2332 u32 parent; 2333 u32 chain_index; 2334 struct Qdisc *q = NULL; 2335 struct tcf_chain_info chain_info; 2336 struct tcf_chain *chain = NULL; 2337 struct tcf_block *block = NULL; 2338 struct tcf_proto *tp = NULL; 2339 unsigned long cl = 0; 2340 void *fh = NULL; 2341 int err; 2342 bool rtnl_held = false; 2343 2344 err = nlmsg_parse_deprecated(n, sizeof(*t), tca, TCA_MAX, 2345 rtm_tca_policy, extack); 2346 if (err < 0) 2347 return err; 2348 2349 t = nlmsg_data(n); 2350 protocol = TC_H_MIN(t->tcm_info); 2351 prio = TC_H_MAJ(t->tcm_info); 2352 parent = t->tcm_parent; 2353 2354 if (prio == 0) { 2355 NL_SET_ERR_MSG(extack, "Invalid filter command with priority of zero"); 2356 return -ENOENT; 2357 } 2358 2359 /* Find head of filter chain. */ 2360 2361 err = __tcf_qdisc_find(net, &q, &parent, t->tcm_ifindex, false, extack); 2362 if (err) 2363 return err; 2364 2365 if (tcf_proto_check_kind(tca[TCA_KIND], name)) { 2366 NL_SET_ERR_MSG(extack, "Specified TC filter name too long"); 2367 err = -EINVAL; 2368 goto errout; 2369 } 2370 /* Take rtnl mutex if block is shared (no qdisc found), qdisc is not 2371 * unlocked, classifier type is not specified, classifier is not 2372 * unlocked. 2373 */ 2374 if ((q && !(q->ops->cl_ops->flags & QDISC_CLASS_OPS_DOIT_UNLOCKED)) || 2375 !tcf_proto_is_unlocked(name)) { 2376 rtnl_held = true; 2377 rtnl_lock(); 2378 } 2379 2380 err = __tcf_qdisc_cl_find(q, parent, &cl, t->tcm_ifindex, extack); 2381 if (err) 2382 goto errout; 2383 2384 block = __tcf_block_find(net, q, cl, t->tcm_ifindex, t->tcm_block_index, 2385 extack); 2386 if (IS_ERR(block)) { 2387 err = PTR_ERR(block); 2388 goto errout; 2389 } 2390 2391 chain_index = tca[TCA_CHAIN] ? nla_get_u32(tca[TCA_CHAIN]) : 0; 2392 if (chain_index > TC_ACT_EXT_VAL_MASK) { 2393 NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit"); 2394 err = -EINVAL; 2395 goto errout; 2396 } 2397 chain = tcf_chain_get(block, chain_index, false); 2398 if (!chain) { 2399 NL_SET_ERR_MSG(extack, "Cannot find specified filter chain"); 2400 err = -EINVAL; 2401 goto errout; 2402 } 2403 2404 mutex_lock(&chain->filter_chain_lock); 2405 tp = tcf_chain_tp_find(chain, &chain_info, protocol, 2406 prio, false); 2407 mutex_unlock(&chain->filter_chain_lock); 2408 if (!tp || IS_ERR(tp)) { 2409 NL_SET_ERR_MSG(extack, "Filter with specified priority/protocol not found"); 2410 err = tp ? PTR_ERR(tp) : -ENOENT; 2411 goto errout; 2412 } else if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind)) { 2413 NL_SET_ERR_MSG(extack, "Specified filter kind does not match existing one"); 2414 err = -EINVAL; 2415 goto errout; 2416 } 2417 2418 fh = tp->ops->get(tp, t->tcm_handle); 2419 2420 if (!fh) { 2421 NL_SET_ERR_MSG(extack, "Specified filter handle not found"); 2422 err = -ENOENT; 2423 } else { 2424 err = tfilter_notify(net, skb, n, tp, block, q, parent, 2425 fh, RTM_NEWTFILTER, true, rtnl_held); 2426 if (err < 0) 2427 NL_SET_ERR_MSG(extack, "Failed to send filter notify message"); 2428 } 2429 2430 tfilter_put(tp, fh); 2431 errout: 2432 if (chain) { 2433 if (tp && !IS_ERR(tp)) 2434 tcf_proto_put(tp, rtnl_held, NULL); 2435 tcf_chain_put(chain); 2436 } 2437 tcf_block_release(q, block, rtnl_held); 2438 2439 if (rtnl_held) 2440 rtnl_unlock(); 2441 2442 return err; 2443 } 2444 2445 struct tcf_dump_args { 2446 struct tcf_walker w; 2447 struct sk_buff *skb; 2448 struct netlink_callback *cb; 2449 struct tcf_block *block; 2450 struct Qdisc *q; 2451 u32 parent; 2452 bool terse_dump; 2453 }; 2454 2455 static int tcf_node_dump(struct tcf_proto *tp, void *n, struct tcf_walker *arg) 2456 { 2457 struct tcf_dump_args *a = (void *)arg; 2458 struct net *net = sock_net(a->skb->sk); 2459 2460 return tcf_fill_node(net, a->skb, tp, a->block, a->q, a->parent, 2461 n, NETLINK_CB(a->cb->skb).portid, 2462 a->cb->nlh->nlmsg_seq, NLM_F_MULTI, 2463 RTM_NEWTFILTER, a->terse_dump, true); 2464 } 2465 2466 static bool tcf_chain_dump(struct tcf_chain *chain, struct Qdisc *q, u32 parent, 2467 struct sk_buff *skb, struct netlink_callback *cb, 2468 long index_start, long *p_index, bool terse) 2469 { 2470 struct net *net = sock_net(skb->sk); 2471 struct tcf_block *block = chain->block; 2472 struct tcmsg *tcm = nlmsg_data(cb->nlh); 2473 struct tcf_proto *tp, *tp_prev; 2474 struct tcf_dump_args arg; 2475 2476 for (tp = __tcf_get_next_proto(chain, NULL); 2477 tp; 2478 tp_prev = tp, 2479 tp = __tcf_get_next_proto(chain, tp), 2480 tcf_proto_put(tp_prev, true, NULL), 2481 (*p_index)++) { 2482 if (*p_index < index_start) 2483 continue; 2484 if (TC_H_MAJ(tcm->tcm_info) && 2485 TC_H_MAJ(tcm->tcm_info) != tp->prio) 2486 continue; 2487 if (TC_H_MIN(tcm->tcm_info) && 2488 TC_H_MIN(tcm->tcm_info) != tp->protocol) 2489 continue; 2490 if (*p_index > index_start) 2491 memset(&cb->args[1], 0, 2492 sizeof(cb->args) - sizeof(cb->args[0])); 2493 if (cb->args[1] == 0) { 2494 if (tcf_fill_node(net, skb, tp, block, q, parent, NULL, 2495 NETLINK_CB(cb->skb).portid, 2496 cb->nlh->nlmsg_seq, NLM_F_MULTI, 2497 RTM_NEWTFILTER, false, true) <= 0) 2498 goto errout; 2499 cb->args[1] = 1; 2500 } 2501 if (!tp->ops->walk) 2502 continue; 2503 arg.w.fn = tcf_node_dump; 2504 arg.skb = skb; 2505 arg.cb = cb; 2506 arg.block = block; 2507 arg.q = q; 2508 arg.parent = parent; 2509 arg.w.stop = 0; 2510 arg.w.skip = cb->args[1] - 1; 2511 arg.w.count = 0; 2512 arg.w.cookie = cb->args[2]; 2513 arg.terse_dump = terse; 2514 tp->ops->walk(tp, &arg.w, true); 2515 cb->args[2] = arg.w.cookie; 2516 cb->args[1] = arg.w.count + 1; 2517 if (arg.w.stop) 2518 goto errout; 2519 } 2520 return true; 2521 2522 errout: 2523 tcf_proto_put(tp, true, NULL); 2524 return false; 2525 } 2526 2527 static const struct nla_policy tcf_tfilter_dump_policy[TCA_MAX + 1] = { 2528 [TCA_DUMP_FLAGS] = NLA_POLICY_BITFIELD32(TCA_DUMP_FLAGS_TERSE), 2529 }; 2530 2531 /* called with RTNL */ 2532 static int tc_dump_tfilter(struct sk_buff *skb, struct netlink_callback *cb) 2533 { 2534 struct tcf_chain *chain, *chain_prev; 2535 struct net *net = sock_net(skb->sk); 2536 struct nlattr *tca[TCA_MAX + 1]; 2537 struct Qdisc *q = NULL; 2538 struct tcf_block *block; 2539 struct tcmsg *tcm = nlmsg_data(cb->nlh); 2540 bool terse_dump = false; 2541 long index_start; 2542 long index; 2543 u32 parent; 2544 int err; 2545 2546 if (nlmsg_len(cb->nlh) < sizeof(*tcm)) 2547 return skb->len; 2548 2549 err = nlmsg_parse_deprecated(cb->nlh, sizeof(*tcm), tca, TCA_MAX, 2550 tcf_tfilter_dump_policy, cb->extack); 2551 if (err) 2552 return err; 2553 2554 if (tca[TCA_DUMP_FLAGS]) { 2555 struct nla_bitfield32 flags = 2556 nla_get_bitfield32(tca[TCA_DUMP_FLAGS]); 2557 2558 terse_dump = flags.value & TCA_DUMP_FLAGS_TERSE; 2559 } 2560 2561 if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK) { 2562 block = tcf_block_refcnt_get(net, tcm->tcm_block_index); 2563 if (!block) 2564 goto out; 2565 /* If we work with block index, q is NULL and parent value 2566 * will never be used in the following code. The check 2567 * in tcf_fill_node prevents it. However, compiler does not 2568 * see that far, so set parent to zero to silence the warning 2569 * about parent being uninitialized. 2570 */ 2571 parent = 0; 2572 } else { 2573 const struct Qdisc_class_ops *cops; 2574 struct net_device *dev; 2575 unsigned long cl = 0; 2576 2577 dev = __dev_get_by_index(net, tcm->tcm_ifindex); 2578 if (!dev) 2579 return skb->len; 2580 2581 parent = tcm->tcm_parent; 2582 if (!parent) 2583 q = dev->qdisc; 2584 else 2585 q = qdisc_lookup(dev, TC_H_MAJ(tcm->tcm_parent)); 2586 if (!q) 2587 goto out; 2588 cops = q->ops->cl_ops; 2589 if (!cops) 2590 goto out; 2591 if (!cops->tcf_block) 2592 goto out; 2593 if (TC_H_MIN(tcm->tcm_parent)) { 2594 cl = cops->find(q, tcm->tcm_parent); 2595 if (cl == 0) 2596 goto out; 2597 } 2598 block = cops->tcf_block(q, cl, NULL); 2599 if (!block) 2600 goto out; 2601 parent = block->classid; 2602 if (tcf_block_shared(block)) 2603 q = NULL; 2604 } 2605 2606 index_start = cb->args[0]; 2607 index = 0; 2608 2609 for (chain = __tcf_get_next_chain(block, NULL); 2610 chain; 2611 chain_prev = chain, 2612 chain = __tcf_get_next_chain(block, chain), 2613 tcf_chain_put(chain_prev)) { 2614 if (tca[TCA_CHAIN] && 2615 nla_get_u32(tca[TCA_CHAIN]) != chain->index) 2616 continue; 2617 if (!tcf_chain_dump(chain, q, parent, skb, cb, 2618 index_start, &index, terse_dump)) { 2619 tcf_chain_put(chain); 2620 err = -EMSGSIZE; 2621 break; 2622 } 2623 } 2624 2625 if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK) 2626 tcf_block_refcnt_put(block, true); 2627 cb->args[0] = index; 2628 2629 out: 2630 /* If we did no progress, the error (EMSGSIZE) is real */ 2631 if (skb->len == 0 && err) 2632 return err; 2633 return skb->len; 2634 } 2635 2636 static int tc_chain_fill_node(const struct tcf_proto_ops *tmplt_ops, 2637 void *tmplt_priv, u32 chain_index, 2638 struct net *net, struct sk_buff *skb, 2639 struct tcf_block *block, 2640 u32 portid, u32 seq, u16 flags, int event) 2641 { 2642 unsigned char *b = skb_tail_pointer(skb); 2643 const struct tcf_proto_ops *ops; 2644 struct nlmsghdr *nlh; 2645 struct tcmsg *tcm; 2646 void *priv; 2647 2648 ops = tmplt_ops; 2649 priv = tmplt_priv; 2650 2651 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags); 2652 if (!nlh) 2653 goto out_nlmsg_trim; 2654 tcm = nlmsg_data(nlh); 2655 tcm->tcm_family = AF_UNSPEC; 2656 tcm->tcm__pad1 = 0; 2657 tcm->tcm__pad2 = 0; 2658 tcm->tcm_handle = 0; 2659 if (block->q) { 2660 tcm->tcm_ifindex = qdisc_dev(block->q)->ifindex; 2661 tcm->tcm_parent = block->q->handle; 2662 } else { 2663 tcm->tcm_ifindex = TCM_IFINDEX_MAGIC_BLOCK; 2664 tcm->tcm_block_index = block->index; 2665 } 2666 2667 if (nla_put_u32(skb, TCA_CHAIN, chain_index)) 2668 goto nla_put_failure; 2669 2670 if (ops) { 2671 if (nla_put_string(skb, TCA_KIND, ops->kind)) 2672 goto nla_put_failure; 2673 if (ops->tmplt_dump(skb, net, priv) < 0) 2674 goto nla_put_failure; 2675 } 2676 2677 nlh->nlmsg_len = skb_tail_pointer(skb) - b; 2678 return skb->len; 2679 2680 out_nlmsg_trim: 2681 nla_put_failure: 2682 nlmsg_trim(skb, b); 2683 return -EMSGSIZE; 2684 } 2685 2686 static int tc_chain_notify(struct tcf_chain *chain, struct sk_buff *oskb, 2687 u32 seq, u16 flags, int event, bool unicast) 2688 { 2689 u32 portid = oskb ? NETLINK_CB(oskb).portid : 0; 2690 struct tcf_block *block = chain->block; 2691 struct net *net = block->net; 2692 struct sk_buff *skb; 2693 int err = 0; 2694 2695 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL); 2696 if (!skb) 2697 return -ENOBUFS; 2698 2699 if (tc_chain_fill_node(chain->tmplt_ops, chain->tmplt_priv, 2700 chain->index, net, skb, block, portid, 2701 seq, flags, event) <= 0) { 2702 kfree_skb(skb); 2703 return -EINVAL; 2704 } 2705 2706 if (unicast) 2707 err = rtnl_unicast(skb, net, portid); 2708 else 2709 err = rtnetlink_send(skb, net, portid, RTNLGRP_TC, 2710 flags & NLM_F_ECHO); 2711 2712 return err; 2713 } 2714 2715 static int tc_chain_notify_delete(const struct tcf_proto_ops *tmplt_ops, 2716 void *tmplt_priv, u32 chain_index, 2717 struct tcf_block *block, struct sk_buff *oskb, 2718 u32 seq, u16 flags, bool unicast) 2719 { 2720 u32 portid = oskb ? NETLINK_CB(oskb).portid : 0; 2721 struct net *net = block->net; 2722 struct sk_buff *skb; 2723 2724 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL); 2725 if (!skb) 2726 return -ENOBUFS; 2727 2728 if (tc_chain_fill_node(tmplt_ops, tmplt_priv, chain_index, net, skb, 2729 block, portid, seq, flags, RTM_DELCHAIN) <= 0) { 2730 kfree_skb(skb); 2731 return -EINVAL; 2732 } 2733 2734 if (unicast) 2735 return rtnl_unicast(skb, net, portid); 2736 2737 return rtnetlink_send(skb, net, portid, RTNLGRP_TC, flags & NLM_F_ECHO); 2738 } 2739 2740 static int tc_chain_tmplt_add(struct tcf_chain *chain, struct net *net, 2741 struct nlattr **tca, 2742 struct netlink_ext_ack *extack) 2743 { 2744 const struct tcf_proto_ops *ops; 2745 char name[IFNAMSIZ]; 2746 void *tmplt_priv; 2747 2748 /* If kind is not set, user did not specify template. */ 2749 if (!tca[TCA_KIND]) 2750 return 0; 2751 2752 if (tcf_proto_check_kind(tca[TCA_KIND], name)) { 2753 NL_SET_ERR_MSG(extack, "Specified TC chain template name too long"); 2754 return -EINVAL; 2755 } 2756 2757 ops = tcf_proto_lookup_ops(name, true, extack); 2758 if (IS_ERR(ops)) 2759 return PTR_ERR(ops); 2760 if (!ops->tmplt_create || !ops->tmplt_destroy || !ops->tmplt_dump) { 2761 NL_SET_ERR_MSG(extack, "Chain templates are not supported with specified classifier"); 2762 return -EOPNOTSUPP; 2763 } 2764 2765 tmplt_priv = ops->tmplt_create(net, chain, tca, extack); 2766 if (IS_ERR(tmplt_priv)) { 2767 module_put(ops->owner); 2768 return PTR_ERR(tmplt_priv); 2769 } 2770 chain->tmplt_ops = ops; 2771 chain->tmplt_priv = tmplt_priv; 2772 return 0; 2773 } 2774 2775 static void tc_chain_tmplt_del(const struct tcf_proto_ops *tmplt_ops, 2776 void *tmplt_priv) 2777 { 2778 /* If template ops are set, no work to do for us. */ 2779 if (!tmplt_ops) 2780 return; 2781 2782 tmplt_ops->tmplt_destroy(tmplt_priv); 2783 module_put(tmplt_ops->owner); 2784 } 2785 2786 /* Add/delete/get a chain */ 2787 2788 static int tc_ctl_chain(struct sk_buff *skb, struct nlmsghdr *n, 2789 struct netlink_ext_ack *extack) 2790 { 2791 struct net *net = sock_net(skb->sk); 2792 struct nlattr *tca[TCA_MAX + 1]; 2793 struct tcmsg *t; 2794 u32 parent; 2795 u32 chain_index; 2796 struct Qdisc *q = NULL; 2797 struct tcf_chain *chain = NULL; 2798 struct tcf_block *block; 2799 unsigned long cl; 2800 int err; 2801 2802 if (n->nlmsg_type != RTM_GETCHAIN && 2803 !netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN)) 2804 return -EPERM; 2805 2806 replay: 2807 err = nlmsg_parse_deprecated(n, sizeof(*t), tca, TCA_MAX, 2808 rtm_tca_policy, extack); 2809 if (err < 0) 2810 return err; 2811 2812 t = nlmsg_data(n); 2813 parent = t->tcm_parent; 2814 cl = 0; 2815 2816 block = tcf_block_find(net, &q, &parent, &cl, 2817 t->tcm_ifindex, t->tcm_block_index, extack); 2818 if (IS_ERR(block)) 2819 return PTR_ERR(block); 2820 2821 chain_index = tca[TCA_CHAIN] ? nla_get_u32(tca[TCA_CHAIN]) : 0; 2822 if (chain_index > TC_ACT_EXT_VAL_MASK) { 2823 NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit"); 2824 err = -EINVAL; 2825 goto errout_block; 2826 } 2827 2828 mutex_lock(&block->lock); 2829 chain = tcf_chain_lookup(block, chain_index); 2830 if (n->nlmsg_type == RTM_NEWCHAIN) { 2831 if (chain) { 2832 if (tcf_chain_held_by_acts_only(chain)) { 2833 /* The chain exists only because there is 2834 * some action referencing it. 2835 */ 2836 tcf_chain_hold(chain); 2837 } else { 2838 NL_SET_ERR_MSG(extack, "Filter chain already exists"); 2839 err = -EEXIST; 2840 goto errout_block_locked; 2841 } 2842 } else { 2843 if (!(n->nlmsg_flags & NLM_F_CREATE)) { 2844 NL_SET_ERR_MSG(extack, "Need both RTM_NEWCHAIN and NLM_F_CREATE to create a new chain"); 2845 err = -ENOENT; 2846 goto errout_block_locked; 2847 } 2848 chain = tcf_chain_create(block, chain_index); 2849 if (!chain) { 2850 NL_SET_ERR_MSG(extack, "Failed to create filter chain"); 2851 err = -ENOMEM; 2852 goto errout_block_locked; 2853 } 2854 } 2855 } else { 2856 if (!chain || tcf_chain_held_by_acts_only(chain)) { 2857 NL_SET_ERR_MSG(extack, "Cannot find specified filter chain"); 2858 err = -EINVAL; 2859 goto errout_block_locked; 2860 } 2861 tcf_chain_hold(chain); 2862 } 2863 2864 if (n->nlmsg_type == RTM_NEWCHAIN) { 2865 /* Modifying chain requires holding parent block lock. In case 2866 * the chain was successfully added, take a reference to the 2867 * chain. This ensures that an empty chain does not disappear at 2868 * the end of this function. 2869 */ 2870 tcf_chain_hold(chain); 2871 chain->explicitly_created = true; 2872 } 2873 mutex_unlock(&block->lock); 2874 2875 switch (n->nlmsg_type) { 2876 case RTM_NEWCHAIN: 2877 err = tc_chain_tmplt_add(chain, net, tca, extack); 2878 if (err) { 2879 tcf_chain_put_explicitly_created(chain); 2880 goto errout; 2881 } 2882 2883 tc_chain_notify(chain, NULL, 0, NLM_F_CREATE | NLM_F_EXCL, 2884 RTM_NEWCHAIN, false); 2885 break; 2886 case RTM_DELCHAIN: 2887 tfilter_notify_chain(net, skb, block, q, parent, n, 2888 chain, RTM_DELTFILTER); 2889 /* Flush the chain first as the user requested chain removal. */ 2890 tcf_chain_flush(chain, true); 2891 /* In case the chain was successfully deleted, put a reference 2892 * to the chain previously taken during addition. 2893 */ 2894 tcf_chain_put_explicitly_created(chain); 2895 break; 2896 case RTM_GETCHAIN: 2897 err = tc_chain_notify(chain, skb, n->nlmsg_seq, 2898 n->nlmsg_flags, n->nlmsg_type, true); 2899 if (err < 0) 2900 NL_SET_ERR_MSG(extack, "Failed to send chain notify message"); 2901 break; 2902 default: 2903 err = -EOPNOTSUPP; 2904 NL_SET_ERR_MSG(extack, "Unsupported message type"); 2905 goto errout; 2906 } 2907 2908 errout: 2909 tcf_chain_put(chain); 2910 errout_block: 2911 tcf_block_release(q, block, true); 2912 if (err == -EAGAIN) 2913 /* Replay the request. */ 2914 goto replay; 2915 return err; 2916 2917 errout_block_locked: 2918 mutex_unlock(&block->lock); 2919 goto errout_block; 2920 } 2921 2922 /* called with RTNL */ 2923 static int tc_dump_chain(struct sk_buff *skb, struct netlink_callback *cb) 2924 { 2925 struct net *net = sock_net(skb->sk); 2926 struct nlattr *tca[TCA_MAX + 1]; 2927 struct Qdisc *q = NULL; 2928 struct tcf_block *block; 2929 struct tcmsg *tcm = nlmsg_data(cb->nlh); 2930 struct tcf_chain *chain; 2931 long index_start; 2932 long index; 2933 int err; 2934 2935 if (nlmsg_len(cb->nlh) < sizeof(*tcm)) 2936 return skb->len; 2937 2938 err = nlmsg_parse_deprecated(cb->nlh, sizeof(*tcm), tca, TCA_MAX, 2939 rtm_tca_policy, cb->extack); 2940 if (err) 2941 return err; 2942 2943 if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK) { 2944 block = tcf_block_refcnt_get(net, tcm->tcm_block_index); 2945 if (!block) 2946 goto out; 2947 } else { 2948 const struct Qdisc_class_ops *cops; 2949 struct net_device *dev; 2950 unsigned long cl = 0; 2951 2952 dev = __dev_get_by_index(net, tcm->tcm_ifindex); 2953 if (!dev) 2954 return skb->len; 2955 2956 if (!tcm->tcm_parent) 2957 q = dev->qdisc; 2958 else 2959 q = qdisc_lookup(dev, TC_H_MAJ(tcm->tcm_parent)); 2960 2961 if (!q) 2962 goto out; 2963 cops = q->ops->cl_ops; 2964 if (!cops) 2965 goto out; 2966 if (!cops->tcf_block) 2967 goto out; 2968 if (TC_H_MIN(tcm->tcm_parent)) { 2969 cl = cops->find(q, tcm->tcm_parent); 2970 if (cl == 0) 2971 goto out; 2972 } 2973 block = cops->tcf_block(q, cl, NULL); 2974 if (!block) 2975 goto out; 2976 if (tcf_block_shared(block)) 2977 q = NULL; 2978 } 2979 2980 index_start = cb->args[0]; 2981 index = 0; 2982 2983 mutex_lock(&block->lock); 2984 list_for_each_entry(chain, &block->chain_list, list) { 2985 if ((tca[TCA_CHAIN] && 2986 nla_get_u32(tca[TCA_CHAIN]) != chain->index)) 2987 continue; 2988 if (index < index_start) { 2989 index++; 2990 continue; 2991 } 2992 if (tcf_chain_held_by_acts_only(chain)) 2993 continue; 2994 err = tc_chain_fill_node(chain->tmplt_ops, chain->tmplt_priv, 2995 chain->index, net, skb, block, 2996 NETLINK_CB(cb->skb).portid, 2997 cb->nlh->nlmsg_seq, NLM_F_MULTI, 2998 RTM_NEWCHAIN); 2999 if (err <= 0) 3000 break; 3001 index++; 3002 } 3003 mutex_unlock(&block->lock); 3004 3005 if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK) 3006 tcf_block_refcnt_put(block, true); 3007 cb->args[0] = index; 3008 3009 out: 3010 /* If we did no progress, the error (EMSGSIZE) is real */ 3011 if (skb->len == 0 && err) 3012 return err; 3013 return skb->len; 3014 } 3015 3016 void tcf_exts_destroy(struct tcf_exts *exts) 3017 { 3018 #ifdef CONFIG_NET_CLS_ACT 3019 if (exts->actions) { 3020 tcf_action_destroy(exts->actions, TCA_ACT_UNBIND); 3021 kfree(exts->actions); 3022 } 3023 exts->nr_actions = 0; 3024 #endif 3025 } 3026 EXPORT_SYMBOL(tcf_exts_destroy); 3027 3028 int tcf_exts_validate(struct net *net, struct tcf_proto *tp, struct nlattr **tb, 3029 struct nlattr *rate_tlv, struct tcf_exts *exts, 3030 u32 flags, struct netlink_ext_ack *extack) 3031 { 3032 #ifdef CONFIG_NET_CLS_ACT 3033 { 3034 int init_res[TCA_ACT_MAX_PRIO] = {}; 3035 struct tc_action *act; 3036 size_t attr_size = 0; 3037 3038 if (exts->police && tb[exts->police]) { 3039 struct tc_action_ops *a_o; 3040 3041 a_o = tc_action_load_ops(tb[exts->police], true, 3042 !(flags & TCA_ACT_FLAGS_NO_RTNL), 3043 extack); 3044 if (IS_ERR(a_o)) 3045 return PTR_ERR(a_o); 3046 flags |= TCA_ACT_FLAGS_POLICE | TCA_ACT_FLAGS_BIND; 3047 act = tcf_action_init_1(net, tp, tb[exts->police], 3048 rate_tlv, a_o, init_res, flags, 3049 extack); 3050 module_put(a_o->owner); 3051 if (IS_ERR(act)) 3052 return PTR_ERR(act); 3053 3054 act->type = exts->type = TCA_OLD_COMPAT; 3055 exts->actions[0] = act; 3056 exts->nr_actions = 1; 3057 tcf_idr_insert_many(exts->actions); 3058 } else if (exts->action && tb[exts->action]) { 3059 int err; 3060 3061 flags |= TCA_ACT_FLAGS_BIND; 3062 err = tcf_action_init(net, tp, tb[exts->action], 3063 rate_tlv, exts->actions, init_res, 3064 &attr_size, flags, extack); 3065 if (err < 0) 3066 return err; 3067 exts->nr_actions = err; 3068 } 3069 } 3070 #else 3071 if ((exts->action && tb[exts->action]) || 3072 (exts->police && tb[exts->police])) { 3073 NL_SET_ERR_MSG(extack, "Classifier actions are not supported per compile options (CONFIG_NET_CLS_ACT)"); 3074 return -EOPNOTSUPP; 3075 } 3076 #endif 3077 3078 return 0; 3079 } 3080 EXPORT_SYMBOL(tcf_exts_validate); 3081 3082 void tcf_exts_change(struct tcf_exts *dst, struct tcf_exts *src) 3083 { 3084 #ifdef CONFIG_NET_CLS_ACT 3085 struct tcf_exts old = *dst; 3086 3087 *dst = *src; 3088 tcf_exts_destroy(&old); 3089 #endif 3090 } 3091 EXPORT_SYMBOL(tcf_exts_change); 3092 3093 #ifdef CONFIG_NET_CLS_ACT 3094 static struct tc_action *tcf_exts_first_act(struct tcf_exts *exts) 3095 { 3096 if (exts->nr_actions == 0) 3097 return NULL; 3098 else 3099 return exts->actions[0]; 3100 } 3101 #endif 3102 3103 int tcf_exts_dump(struct sk_buff *skb, struct tcf_exts *exts) 3104 { 3105 #ifdef CONFIG_NET_CLS_ACT 3106 struct nlattr *nest; 3107 3108 if (exts->action && tcf_exts_has_actions(exts)) { 3109 /* 3110 * again for backward compatible mode - we want 3111 * to work with both old and new modes of entering 3112 * tc data even if iproute2 was newer - jhs 3113 */ 3114 if (exts->type != TCA_OLD_COMPAT) { 3115 nest = nla_nest_start_noflag(skb, exts->action); 3116 if (nest == NULL) 3117 goto nla_put_failure; 3118 3119 if (tcf_action_dump(skb, exts->actions, 0, 0, false) 3120 < 0) 3121 goto nla_put_failure; 3122 nla_nest_end(skb, nest); 3123 } else if (exts->police) { 3124 struct tc_action *act = tcf_exts_first_act(exts); 3125 nest = nla_nest_start_noflag(skb, exts->police); 3126 if (nest == NULL || !act) 3127 goto nla_put_failure; 3128 if (tcf_action_dump_old(skb, act, 0, 0) < 0) 3129 goto nla_put_failure; 3130 nla_nest_end(skb, nest); 3131 } 3132 } 3133 return 0; 3134 3135 nla_put_failure: 3136 nla_nest_cancel(skb, nest); 3137 return -1; 3138 #else 3139 return 0; 3140 #endif 3141 } 3142 EXPORT_SYMBOL(tcf_exts_dump); 3143 3144 int tcf_exts_terse_dump(struct sk_buff *skb, struct tcf_exts *exts) 3145 { 3146 #ifdef CONFIG_NET_CLS_ACT 3147 struct nlattr *nest; 3148 3149 if (!exts->action || !tcf_exts_has_actions(exts)) 3150 return 0; 3151 3152 nest = nla_nest_start_noflag(skb, exts->action); 3153 if (!nest) 3154 goto nla_put_failure; 3155 3156 if (tcf_action_dump(skb, exts->actions, 0, 0, true) < 0) 3157 goto nla_put_failure; 3158 nla_nest_end(skb, nest); 3159 return 0; 3160 3161 nla_put_failure: 3162 nla_nest_cancel(skb, nest); 3163 return -1; 3164 #else 3165 return 0; 3166 #endif 3167 } 3168 EXPORT_SYMBOL(tcf_exts_terse_dump); 3169 3170 int tcf_exts_dump_stats(struct sk_buff *skb, struct tcf_exts *exts) 3171 { 3172 #ifdef CONFIG_NET_CLS_ACT 3173 struct tc_action *a = tcf_exts_first_act(exts); 3174 if (a != NULL && tcf_action_copy_stats(skb, a, 1) < 0) 3175 return -1; 3176 #endif 3177 return 0; 3178 } 3179 EXPORT_SYMBOL(tcf_exts_dump_stats); 3180 3181 static void tcf_block_offload_inc(struct tcf_block *block, u32 *flags) 3182 { 3183 if (*flags & TCA_CLS_FLAGS_IN_HW) 3184 return; 3185 *flags |= TCA_CLS_FLAGS_IN_HW; 3186 atomic_inc(&block->offloadcnt); 3187 } 3188 3189 static void tcf_block_offload_dec(struct tcf_block *block, u32 *flags) 3190 { 3191 if (!(*flags & TCA_CLS_FLAGS_IN_HW)) 3192 return; 3193 *flags &= ~TCA_CLS_FLAGS_IN_HW; 3194 atomic_dec(&block->offloadcnt); 3195 } 3196 3197 static void tc_cls_offload_cnt_update(struct tcf_block *block, 3198 struct tcf_proto *tp, u32 *cnt, 3199 u32 *flags, u32 diff, bool add) 3200 { 3201 lockdep_assert_held(&block->cb_lock); 3202 3203 spin_lock(&tp->lock); 3204 if (add) { 3205 if (!*cnt) 3206 tcf_block_offload_inc(block, flags); 3207 *cnt += diff; 3208 } else { 3209 *cnt -= diff; 3210 if (!*cnt) 3211 tcf_block_offload_dec(block, flags); 3212 } 3213 spin_unlock(&tp->lock); 3214 } 3215 3216 static void 3217 tc_cls_offload_cnt_reset(struct tcf_block *block, struct tcf_proto *tp, 3218 u32 *cnt, u32 *flags) 3219 { 3220 lockdep_assert_held(&block->cb_lock); 3221 3222 spin_lock(&tp->lock); 3223 tcf_block_offload_dec(block, flags); 3224 *cnt = 0; 3225 spin_unlock(&tp->lock); 3226 } 3227 3228 static int 3229 __tc_setup_cb_call(struct tcf_block *block, enum tc_setup_type type, 3230 void *type_data, bool err_stop) 3231 { 3232 struct flow_block_cb *block_cb; 3233 int ok_count = 0; 3234 int err; 3235 3236 list_for_each_entry(block_cb, &block->flow_block.cb_list, list) { 3237 err = block_cb->cb(type, type_data, block_cb->cb_priv); 3238 if (err) { 3239 if (err_stop) 3240 return err; 3241 } else { 3242 ok_count++; 3243 } 3244 } 3245 return ok_count; 3246 } 3247 3248 int tc_setup_cb_call(struct tcf_block *block, enum tc_setup_type type, 3249 void *type_data, bool err_stop, bool rtnl_held) 3250 { 3251 bool take_rtnl = READ_ONCE(block->lockeddevcnt) && !rtnl_held; 3252 int ok_count; 3253 3254 retry: 3255 if (take_rtnl) 3256 rtnl_lock(); 3257 down_read(&block->cb_lock); 3258 /* Need to obtain rtnl lock if block is bound to devs that require it. 3259 * In block bind code cb_lock is obtained while holding rtnl, so we must 3260 * obtain the locks in same order here. 3261 */ 3262 if (!rtnl_held && !take_rtnl && block->lockeddevcnt) { 3263 up_read(&block->cb_lock); 3264 take_rtnl = true; 3265 goto retry; 3266 } 3267 3268 ok_count = __tc_setup_cb_call(block, type, type_data, err_stop); 3269 3270 up_read(&block->cb_lock); 3271 if (take_rtnl) 3272 rtnl_unlock(); 3273 return ok_count; 3274 } 3275 EXPORT_SYMBOL(tc_setup_cb_call); 3276 3277 /* Non-destructive filter add. If filter that wasn't already in hardware is 3278 * successfully offloaded, increment block offloads counter. On failure, 3279 * previously offloaded filter is considered to be intact and offloads counter 3280 * is not decremented. 3281 */ 3282 3283 int tc_setup_cb_add(struct tcf_block *block, struct tcf_proto *tp, 3284 enum tc_setup_type type, void *type_data, bool err_stop, 3285 u32 *flags, unsigned int *in_hw_count, bool rtnl_held) 3286 { 3287 bool take_rtnl = READ_ONCE(block->lockeddevcnt) && !rtnl_held; 3288 int ok_count; 3289 3290 retry: 3291 if (take_rtnl) 3292 rtnl_lock(); 3293 down_read(&block->cb_lock); 3294 /* Need to obtain rtnl lock if block is bound to devs that require it. 3295 * In block bind code cb_lock is obtained while holding rtnl, so we must 3296 * obtain the locks in same order here. 3297 */ 3298 if (!rtnl_held && !take_rtnl && block->lockeddevcnt) { 3299 up_read(&block->cb_lock); 3300 take_rtnl = true; 3301 goto retry; 3302 } 3303 3304 /* Make sure all netdevs sharing this block are offload-capable. */ 3305 if (block->nooffloaddevcnt && err_stop) { 3306 ok_count = -EOPNOTSUPP; 3307 goto err_unlock; 3308 } 3309 3310 ok_count = __tc_setup_cb_call(block, type, type_data, err_stop); 3311 if (ok_count < 0) 3312 goto err_unlock; 3313 3314 if (tp->ops->hw_add) 3315 tp->ops->hw_add(tp, type_data); 3316 if (ok_count > 0) 3317 tc_cls_offload_cnt_update(block, tp, in_hw_count, flags, 3318 ok_count, true); 3319 err_unlock: 3320 up_read(&block->cb_lock); 3321 if (take_rtnl) 3322 rtnl_unlock(); 3323 return ok_count < 0 ? ok_count : 0; 3324 } 3325 EXPORT_SYMBOL(tc_setup_cb_add); 3326 3327 /* Destructive filter replace. If filter that wasn't already in hardware is 3328 * successfully offloaded, increment block offload counter. On failure, 3329 * previously offloaded filter is considered to be destroyed and offload counter 3330 * is decremented. 3331 */ 3332 3333 int tc_setup_cb_replace(struct tcf_block *block, struct tcf_proto *tp, 3334 enum tc_setup_type type, void *type_data, bool err_stop, 3335 u32 *old_flags, unsigned int *old_in_hw_count, 3336 u32 *new_flags, unsigned int *new_in_hw_count, 3337 bool rtnl_held) 3338 { 3339 bool take_rtnl = READ_ONCE(block->lockeddevcnt) && !rtnl_held; 3340 int ok_count; 3341 3342 retry: 3343 if (take_rtnl) 3344 rtnl_lock(); 3345 down_read(&block->cb_lock); 3346 /* Need to obtain rtnl lock if block is bound to devs that require it. 3347 * In block bind code cb_lock is obtained while holding rtnl, so we must 3348 * obtain the locks in same order here. 3349 */ 3350 if (!rtnl_held && !take_rtnl && block->lockeddevcnt) { 3351 up_read(&block->cb_lock); 3352 take_rtnl = true; 3353 goto retry; 3354 } 3355 3356 /* Make sure all netdevs sharing this block are offload-capable. */ 3357 if (block->nooffloaddevcnt && err_stop) { 3358 ok_count = -EOPNOTSUPP; 3359 goto err_unlock; 3360 } 3361 3362 tc_cls_offload_cnt_reset(block, tp, old_in_hw_count, old_flags); 3363 if (tp->ops->hw_del) 3364 tp->ops->hw_del(tp, type_data); 3365 3366 ok_count = __tc_setup_cb_call(block, type, type_data, err_stop); 3367 if (ok_count < 0) 3368 goto err_unlock; 3369 3370 if (tp->ops->hw_add) 3371 tp->ops->hw_add(tp, type_data); 3372 if (ok_count > 0) 3373 tc_cls_offload_cnt_update(block, tp, new_in_hw_count, 3374 new_flags, ok_count, true); 3375 err_unlock: 3376 up_read(&block->cb_lock); 3377 if (take_rtnl) 3378 rtnl_unlock(); 3379 return ok_count < 0 ? ok_count : 0; 3380 } 3381 EXPORT_SYMBOL(tc_setup_cb_replace); 3382 3383 /* Destroy filter and decrement block offload counter, if filter was previously 3384 * offloaded. 3385 */ 3386 3387 int tc_setup_cb_destroy(struct tcf_block *block, struct tcf_proto *tp, 3388 enum tc_setup_type type, void *type_data, bool err_stop, 3389 u32 *flags, unsigned int *in_hw_count, bool rtnl_held) 3390 { 3391 bool take_rtnl = READ_ONCE(block->lockeddevcnt) && !rtnl_held; 3392 int ok_count; 3393 3394 retry: 3395 if (take_rtnl) 3396 rtnl_lock(); 3397 down_read(&block->cb_lock); 3398 /* Need to obtain rtnl lock if block is bound to devs that require it. 3399 * In block bind code cb_lock is obtained while holding rtnl, so we must 3400 * obtain the locks in same order here. 3401 */ 3402 if (!rtnl_held && !take_rtnl && block->lockeddevcnt) { 3403 up_read(&block->cb_lock); 3404 take_rtnl = true; 3405 goto retry; 3406 } 3407 3408 ok_count = __tc_setup_cb_call(block, type, type_data, err_stop); 3409 3410 tc_cls_offload_cnt_reset(block, tp, in_hw_count, flags); 3411 if (tp->ops->hw_del) 3412 tp->ops->hw_del(tp, type_data); 3413 3414 up_read(&block->cb_lock); 3415 if (take_rtnl) 3416 rtnl_unlock(); 3417 return ok_count < 0 ? ok_count : 0; 3418 } 3419 EXPORT_SYMBOL(tc_setup_cb_destroy); 3420 3421 int tc_setup_cb_reoffload(struct tcf_block *block, struct tcf_proto *tp, 3422 bool add, flow_setup_cb_t *cb, 3423 enum tc_setup_type type, void *type_data, 3424 void *cb_priv, u32 *flags, unsigned int *in_hw_count) 3425 { 3426 int err = cb(type, type_data, cb_priv); 3427 3428 if (err) { 3429 if (add && tc_skip_sw(*flags)) 3430 return err; 3431 } else { 3432 tc_cls_offload_cnt_update(block, tp, in_hw_count, flags, 1, 3433 add); 3434 } 3435 3436 return 0; 3437 } 3438 EXPORT_SYMBOL(tc_setup_cb_reoffload); 3439 3440 static int tcf_act_get_cookie(struct flow_action_entry *entry, 3441 const struct tc_action *act) 3442 { 3443 struct tc_cookie *cookie; 3444 int err = 0; 3445 3446 rcu_read_lock(); 3447 cookie = rcu_dereference(act->act_cookie); 3448 if (cookie) { 3449 entry->cookie = flow_action_cookie_create(cookie->data, 3450 cookie->len, 3451 GFP_ATOMIC); 3452 if (!entry->cookie) 3453 err = -ENOMEM; 3454 } 3455 rcu_read_unlock(); 3456 return err; 3457 } 3458 3459 static void tcf_act_put_cookie(struct flow_action_entry *entry) 3460 { 3461 flow_action_cookie_destroy(entry->cookie); 3462 } 3463 3464 void tc_cleanup_flow_action(struct flow_action *flow_action) 3465 { 3466 struct flow_action_entry *entry; 3467 int i; 3468 3469 flow_action_for_each(i, entry, flow_action) { 3470 tcf_act_put_cookie(entry); 3471 if (entry->destructor) 3472 entry->destructor(entry->destructor_priv); 3473 } 3474 } 3475 EXPORT_SYMBOL(tc_cleanup_flow_action); 3476 3477 static void tcf_mirred_get_dev(struct flow_action_entry *entry, 3478 const struct tc_action *act) 3479 { 3480 #ifdef CONFIG_NET_CLS_ACT 3481 entry->dev = act->ops->get_dev(act, &entry->destructor); 3482 if (!entry->dev) 3483 return; 3484 entry->destructor_priv = entry->dev; 3485 #endif 3486 } 3487 3488 static void tcf_tunnel_encap_put_tunnel(void *priv) 3489 { 3490 struct ip_tunnel_info *tunnel = priv; 3491 3492 kfree(tunnel); 3493 } 3494 3495 static int tcf_tunnel_encap_get_tunnel(struct flow_action_entry *entry, 3496 const struct tc_action *act) 3497 { 3498 entry->tunnel = tcf_tunnel_info_copy(act); 3499 if (!entry->tunnel) 3500 return -ENOMEM; 3501 entry->destructor = tcf_tunnel_encap_put_tunnel; 3502 entry->destructor_priv = entry->tunnel; 3503 return 0; 3504 } 3505 3506 static void tcf_sample_get_group(struct flow_action_entry *entry, 3507 const struct tc_action *act) 3508 { 3509 #ifdef CONFIG_NET_CLS_ACT 3510 entry->sample.psample_group = 3511 act->ops->get_psample_group(act, &entry->destructor); 3512 entry->destructor_priv = entry->sample.psample_group; 3513 #endif 3514 } 3515 3516 static void tcf_gate_entry_destructor(void *priv) 3517 { 3518 struct action_gate_entry *oe = priv; 3519 3520 kfree(oe); 3521 } 3522 3523 static int tcf_gate_get_entries(struct flow_action_entry *entry, 3524 const struct tc_action *act) 3525 { 3526 entry->gate.entries = tcf_gate_get_list(act); 3527 3528 if (!entry->gate.entries) 3529 return -EINVAL; 3530 3531 entry->destructor = tcf_gate_entry_destructor; 3532 entry->destructor_priv = entry->gate.entries; 3533 3534 return 0; 3535 } 3536 3537 static enum flow_action_hw_stats tc_act_hw_stats(u8 hw_stats) 3538 { 3539 if (WARN_ON_ONCE(hw_stats > TCA_ACT_HW_STATS_ANY)) 3540 return FLOW_ACTION_HW_STATS_DONT_CARE; 3541 else if (!hw_stats) 3542 return FLOW_ACTION_HW_STATS_DISABLED; 3543 3544 return hw_stats; 3545 } 3546 3547 int tc_setup_flow_action(struct flow_action *flow_action, 3548 const struct tcf_exts *exts) 3549 { 3550 struct tc_action *act; 3551 int i, j, k, err = 0; 3552 3553 BUILD_BUG_ON(TCA_ACT_HW_STATS_ANY != FLOW_ACTION_HW_STATS_ANY); 3554 BUILD_BUG_ON(TCA_ACT_HW_STATS_IMMEDIATE != FLOW_ACTION_HW_STATS_IMMEDIATE); 3555 BUILD_BUG_ON(TCA_ACT_HW_STATS_DELAYED != FLOW_ACTION_HW_STATS_DELAYED); 3556 3557 if (!exts) 3558 return 0; 3559 3560 j = 0; 3561 tcf_exts_for_each_action(i, act, exts) { 3562 struct flow_action_entry *entry; 3563 3564 entry = &flow_action->entries[j]; 3565 spin_lock_bh(&act->tcfa_lock); 3566 err = tcf_act_get_cookie(entry, act); 3567 if (err) 3568 goto err_out_locked; 3569 3570 entry->hw_stats = tc_act_hw_stats(act->hw_stats); 3571 3572 if (is_tcf_gact_ok(act)) { 3573 entry->id = FLOW_ACTION_ACCEPT; 3574 } else if (is_tcf_gact_shot(act)) { 3575 entry->id = FLOW_ACTION_DROP; 3576 } else if (is_tcf_gact_trap(act)) { 3577 entry->id = FLOW_ACTION_TRAP; 3578 } else if (is_tcf_gact_goto_chain(act)) { 3579 entry->id = FLOW_ACTION_GOTO; 3580 entry->chain_index = tcf_gact_goto_chain_index(act); 3581 } else if (is_tcf_mirred_egress_redirect(act)) { 3582 entry->id = FLOW_ACTION_REDIRECT; 3583 tcf_mirred_get_dev(entry, act); 3584 } else if (is_tcf_mirred_egress_mirror(act)) { 3585 entry->id = FLOW_ACTION_MIRRED; 3586 tcf_mirred_get_dev(entry, act); 3587 } else if (is_tcf_mirred_ingress_redirect(act)) { 3588 entry->id = FLOW_ACTION_REDIRECT_INGRESS; 3589 tcf_mirred_get_dev(entry, act); 3590 } else if (is_tcf_mirred_ingress_mirror(act)) { 3591 entry->id = FLOW_ACTION_MIRRED_INGRESS; 3592 tcf_mirred_get_dev(entry, act); 3593 } else if (is_tcf_vlan(act)) { 3594 switch (tcf_vlan_action(act)) { 3595 case TCA_VLAN_ACT_PUSH: 3596 entry->id = FLOW_ACTION_VLAN_PUSH; 3597 entry->vlan.vid = tcf_vlan_push_vid(act); 3598 entry->vlan.proto = tcf_vlan_push_proto(act); 3599 entry->vlan.prio = tcf_vlan_push_prio(act); 3600 break; 3601 case TCA_VLAN_ACT_POP: 3602 entry->id = FLOW_ACTION_VLAN_POP; 3603 break; 3604 case TCA_VLAN_ACT_MODIFY: 3605 entry->id = FLOW_ACTION_VLAN_MANGLE; 3606 entry->vlan.vid = tcf_vlan_push_vid(act); 3607 entry->vlan.proto = tcf_vlan_push_proto(act); 3608 entry->vlan.prio = tcf_vlan_push_prio(act); 3609 break; 3610 default: 3611 err = -EOPNOTSUPP; 3612 goto err_out_locked; 3613 } 3614 } else if (is_tcf_tunnel_set(act)) { 3615 entry->id = FLOW_ACTION_TUNNEL_ENCAP; 3616 err = tcf_tunnel_encap_get_tunnel(entry, act); 3617 if (err) 3618 goto err_out_locked; 3619 } else if (is_tcf_tunnel_release(act)) { 3620 entry->id = FLOW_ACTION_TUNNEL_DECAP; 3621 } else if (is_tcf_pedit(act)) { 3622 for (k = 0; k < tcf_pedit_nkeys(act); k++) { 3623 switch (tcf_pedit_cmd(act, k)) { 3624 case TCA_PEDIT_KEY_EX_CMD_SET: 3625 entry->id = FLOW_ACTION_MANGLE; 3626 break; 3627 case TCA_PEDIT_KEY_EX_CMD_ADD: 3628 entry->id = FLOW_ACTION_ADD; 3629 break; 3630 default: 3631 err = -EOPNOTSUPP; 3632 goto err_out_locked; 3633 } 3634 entry->mangle.htype = tcf_pedit_htype(act, k); 3635 entry->mangle.mask = tcf_pedit_mask(act, k); 3636 entry->mangle.val = tcf_pedit_val(act, k); 3637 entry->mangle.offset = tcf_pedit_offset(act, k); 3638 entry->hw_stats = tc_act_hw_stats(act->hw_stats); 3639 entry = &flow_action->entries[++j]; 3640 } 3641 } else if (is_tcf_csum(act)) { 3642 entry->id = FLOW_ACTION_CSUM; 3643 entry->csum_flags = tcf_csum_update_flags(act); 3644 } else if (is_tcf_skbedit_mark(act)) { 3645 entry->id = FLOW_ACTION_MARK; 3646 entry->mark = tcf_skbedit_mark(act); 3647 } else if (is_tcf_sample(act)) { 3648 entry->id = FLOW_ACTION_SAMPLE; 3649 entry->sample.trunc_size = tcf_sample_trunc_size(act); 3650 entry->sample.truncate = tcf_sample_truncate(act); 3651 entry->sample.rate = tcf_sample_rate(act); 3652 tcf_sample_get_group(entry, act); 3653 } else if (is_tcf_police(act)) { 3654 entry->id = FLOW_ACTION_POLICE; 3655 entry->police.burst = tcf_police_burst(act); 3656 entry->police.rate_bytes_ps = 3657 tcf_police_rate_bytes_ps(act); 3658 entry->police.burst_pkt = tcf_police_burst_pkt(act); 3659 entry->police.rate_pkt_ps = 3660 tcf_police_rate_pkt_ps(act); 3661 entry->police.mtu = tcf_police_tcfp_mtu(act); 3662 entry->police.index = act->tcfa_index; 3663 } else if (is_tcf_ct(act)) { 3664 entry->id = FLOW_ACTION_CT; 3665 entry->ct.action = tcf_ct_action(act); 3666 entry->ct.zone = tcf_ct_zone(act); 3667 entry->ct.flow_table = tcf_ct_ft(act); 3668 } else if (is_tcf_mpls(act)) { 3669 switch (tcf_mpls_action(act)) { 3670 case TCA_MPLS_ACT_PUSH: 3671 entry->id = FLOW_ACTION_MPLS_PUSH; 3672 entry->mpls_push.proto = tcf_mpls_proto(act); 3673 entry->mpls_push.label = tcf_mpls_label(act); 3674 entry->mpls_push.tc = tcf_mpls_tc(act); 3675 entry->mpls_push.bos = tcf_mpls_bos(act); 3676 entry->mpls_push.ttl = tcf_mpls_ttl(act); 3677 break; 3678 case TCA_MPLS_ACT_POP: 3679 entry->id = FLOW_ACTION_MPLS_POP; 3680 entry->mpls_pop.proto = tcf_mpls_proto(act); 3681 break; 3682 case TCA_MPLS_ACT_MODIFY: 3683 entry->id = FLOW_ACTION_MPLS_MANGLE; 3684 entry->mpls_mangle.label = tcf_mpls_label(act); 3685 entry->mpls_mangle.tc = tcf_mpls_tc(act); 3686 entry->mpls_mangle.bos = tcf_mpls_bos(act); 3687 entry->mpls_mangle.ttl = tcf_mpls_ttl(act); 3688 break; 3689 default: 3690 goto err_out_locked; 3691 } 3692 } else if (is_tcf_skbedit_ptype(act)) { 3693 entry->id = FLOW_ACTION_PTYPE; 3694 entry->ptype = tcf_skbedit_ptype(act); 3695 } else if (is_tcf_skbedit_priority(act)) { 3696 entry->id = FLOW_ACTION_PRIORITY; 3697 entry->priority = tcf_skbedit_priority(act); 3698 } else if (is_tcf_gate(act)) { 3699 entry->id = FLOW_ACTION_GATE; 3700 entry->gate.index = tcf_gate_index(act); 3701 entry->gate.prio = tcf_gate_prio(act); 3702 entry->gate.basetime = tcf_gate_basetime(act); 3703 entry->gate.cycletime = tcf_gate_cycletime(act); 3704 entry->gate.cycletimeext = tcf_gate_cycletimeext(act); 3705 entry->gate.num_entries = tcf_gate_num_entries(act); 3706 err = tcf_gate_get_entries(entry, act); 3707 if (err) 3708 goto err_out_locked; 3709 } else { 3710 err = -EOPNOTSUPP; 3711 goto err_out_locked; 3712 } 3713 spin_unlock_bh(&act->tcfa_lock); 3714 3715 if (!is_tcf_pedit(act)) 3716 j++; 3717 } 3718 3719 err_out: 3720 if (err) 3721 tc_cleanup_flow_action(flow_action); 3722 3723 return err; 3724 err_out_locked: 3725 spin_unlock_bh(&act->tcfa_lock); 3726 goto err_out; 3727 } 3728 EXPORT_SYMBOL(tc_setup_flow_action); 3729 3730 unsigned int tcf_exts_num_actions(struct tcf_exts *exts) 3731 { 3732 unsigned int num_acts = 0; 3733 struct tc_action *act; 3734 int i; 3735 3736 tcf_exts_for_each_action(i, act, exts) { 3737 if (is_tcf_pedit(act)) 3738 num_acts += tcf_pedit_nkeys(act); 3739 else 3740 num_acts++; 3741 } 3742 return num_acts; 3743 } 3744 EXPORT_SYMBOL(tcf_exts_num_actions); 3745 3746 #ifdef CONFIG_NET_CLS_ACT 3747 static int tcf_qevent_parse_block_index(struct nlattr *block_index_attr, 3748 u32 *p_block_index, 3749 struct netlink_ext_ack *extack) 3750 { 3751 *p_block_index = nla_get_u32(block_index_attr); 3752 if (!*p_block_index) { 3753 NL_SET_ERR_MSG(extack, "Block number may not be zero"); 3754 return -EINVAL; 3755 } 3756 3757 return 0; 3758 } 3759 3760 int tcf_qevent_init(struct tcf_qevent *qe, struct Qdisc *sch, 3761 enum flow_block_binder_type binder_type, 3762 struct nlattr *block_index_attr, 3763 struct netlink_ext_ack *extack) 3764 { 3765 u32 block_index; 3766 int err; 3767 3768 if (!block_index_attr) 3769 return 0; 3770 3771 err = tcf_qevent_parse_block_index(block_index_attr, &block_index, extack); 3772 if (err) 3773 return err; 3774 3775 if (!block_index) 3776 return 0; 3777 3778 qe->info.binder_type = binder_type; 3779 qe->info.chain_head_change = tcf_chain_head_change_dflt; 3780 qe->info.chain_head_change_priv = &qe->filter_chain; 3781 qe->info.block_index = block_index; 3782 3783 return tcf_block_get_ext(&qe->block, sch, &qe->info, extack); 3784 } 3785 EXPORT_SYMBOL(tcf_qevent_init); 3786 3787 void tcf_qevent_destroy(struct tcf_qevent *qe, struct Qdisc *sch) 3788 { 3789 if (qe->info.block_index) 3790 tcf_block_put_ext(qe->block, sch, &qe->info); 3791 } 3792 EXPORT_SYMBOL(tcf_qevent_destroy); 3793 3794 int tcf_qevent_validate_change(struct tcf_qevent *qe, struct nlattr *block_index_attr, 3795 struct netlink_ext_ack *extack) 3796 { 3797 u32 block_index; 3798 int err; 3799 3800 if (!block_index_attr) 3801 return 0; 3802 3803 err = tcf_qevent_parse_block_index(block_index_attr, &block_index, extack); 3804 if (err) 3805 return err; 3806 3807 /* Bounce newly-configured block or change in block. */ 3808 if (block_index != qe->info.block_index) { 3809 NL_SET_ERR_MSG(extack, "Change of blocks is not supported"); 3810 return -EINVAL; 3811 } 3812 3813 return 0; 3814 } 3815 EXPORT_SYMBOL(tcf_qevent_validate_change); 3816 3817 struct sk_buff *tcf_qevent_handle(struct tcf_qevent *qe, struct Qdisc *sch, struct sk_buff *skb, 3818 struct sk_buff **to_free, int *ret) 3819 { 3820 struct tcf_result cl_res; 3821 struct tcf_proto *fl; 3822 3823 if (!qe->info.block_index) 3824 return skb; 3825 3826 fl = rcu_dereference_bh(qe->filter_chain); 3827 3828 switch (tcf_classify(skb, NULL, fl, &cl_res, false)) { 3829 case TC_ACT_SHOT: 3830 qdisc_qstats_drop(sch); 3831 __qdisc_drop(skb, to_free); 3832 *ret = __NET_XMIT_BYPASS; 3833 return NULL; 3834 case TC_ACT_STOLEN: 3835 case TC_ACT_QUEUED: 3836 case TC_ACT_TRAP: 3837 __qdisc_drop(skb, to_free); 3838 *ret = __NET_XMIT_STOLEN; 3839 return NULL; 3840 case TC_ACT_REDIRECT: 3841 skb_do_redirect(skb); 3842 *ret = __NET_XMIT_STOLEN; 3843 return NULL; 3844 } 3845 3846 return skb; 3847 } 3848 EXPORT_SYMBOL(tcf_qevent_handle); 3849 3850 int tcf_qevent_dump(struct sk_buff *skb, int attr_name, struct tcf_qevent *qe) 3851 { 3852 if (!qe->info.block_index) 3853 return 0; 3854 return nla_put_u32(skb, attr_name, qe->info.block_index); 3855 } 3856 EXPORT_SYMBOL(tcf_qevent_dump); 3857 #endif 3858 3859 static __net_init int tcf_net_init(struct net *net) 3860 { 3861 struct tcf_net *tn = net_generic(net, tcf_net_id); 3862 3863 spin_lock_init(&tn->idr_lock); 3864 idr_init(&tn->idr); 3865 return 0; 3866 } 3867 3868 static void __net_exit tcf_net_exit(struct net *net) 3869 { 3870 struct tcf_net *tn = net_generic(net, tcf_net_id); 3871 3872 idr_destroy(&tn->idr); 3873 } 3874 3875 static struct pernet_operations tcf_net_ops = { 3876 .init = tcf_net_init, 3877 .exit = tcf_net_exit, 3878 .id = &tcf_net_id, 3879 .size = sizeof(struct tcf_net), 3880 }; 3881 3882 static int __init tc_filter_init(void) 3883 { 3884 int err; 3885 3886 tc_filter_wq = alloc_ordered_workqueue("tc_filter_workqueue", 0); 3887 if (!tc_filter_wq) 3888 return -ENOMEM; 3889 3890 err = register_pernet_subsys(&tcf_net_ops); 3891 if (err) 3892 goto err_register_pernet_subsys; 3893 3894 rtnl_register(PF_UNSPEC, RTM_NEWTFILTER, tc_new_tfilter, NULL, 3895 RTNL_FLAG_DOIT_UNLOCKED); 3896 rtnl_register(PF_UNSPEC, RTM_DELTFILTER, tc_del_tfilter, NULL, 3897 RTNL_FLAG_DOIT_UNLOCKED); 3898 rtnl_register(PF_UNSPEC, RTM_GETTFILTER, tc_get_tfilter, 3899 tc_dump_tfilter, RTNL_FLAG_DOIT_UNLOCKED); 3900 rtnl_register(PF_UNSPEC, RTM_NEWCHAIN, tc_ctl_chain, NULL, 0); 3901 rtnl_register(PF_UNSPEC, RTM_DELCHAIN, tc_ctl_chain, NULL, 0); 3902 rtnl_register(PF_UNSPEC, RTM_GETCHAIN, tc_ctl_chain, 3903 tc_dump_chain, 0); 3904 3905 return 0; 3906 3907 err_register_pernet_subsys: 3908 destroy_workqueue(tc_filter_wq); 3909 return err; 3910 } 3911 3912 subsys_initcall(tc_filter_init); 3913