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 struct tc_skb_cb *cb = tc_skb_cb(skb); 1621 1622 ext = tc_skb_ext_alloc(skb); 1623 if (WARN_ON_ONCE(!ext)) 1624 return TC_ACT_SHOT; 1625 ext->chain = last_executed_chain; 1626 ext->mru = cb->mru; 1627 ext->post_ct = cb->post_ct; 1628 ext->zone = cb->zone; 1629 } 1630 1631 return ret; 1632 #endif 1633 } 1634 EXPORT_SYMBOL(tcf_classify); 1635 1636 struct tcf_chain_info { 1637 struct tcf_proto __rcu **pprev; 1638 struct tcf_proto __rcu *next; 1639 }; 1640 1641 static struct tcf_proto *tcf_chain_tp_prev(struct tcf_chain *chain, 1642 struct tcf_chain_info *chain_info) 1643 { 1644 return tcf_chain_dereference(*chain_info->pprev, chain); 1645 } 1646 1647 static int tcf_chain_tp_insert(struct tcf_chain *chain, 1648 struct tcf_chain_info *chain_info, 1649 struct tcf_proto *tp) 1650 { 1651 if (chain->flushing) 1652 return -EAGAIN; 1653 1654 if (*chain_info->pprev == chain->filter_chain) 1655 tcf_chain0_head_change(chain, tp); 1656 tcf_proto_get(tp); 1657 RCU_INIT_POINTER(tp->next, tcf_chain_tp_prev(chain, chain_info)); 1658 rcu_assign_pointer(*chain_info->pprev, tp); 1659 1660 return 0; 1661 } 1662 1663 static void tcf_chain_tp_remove(struct tcf_chain *chain, 1664 struct tcf_chain_info *chain_info, 1665 struct tcf_proto *tp) 1666 { 1667 struct tcf_proto *next = tcf_chain_dereference(chain_info->next, chain); 1668 1669 tcf_proto_mark_delete(tp); 1670 if (tp == chain->filter_chain) 1671 tcf_chain0_head_change(chain, next); 1672 RCU_INIT_POINTER(*chain_info->pprev, next); 1673 } 1674 1675 static struct tcf_proto *tcf_chain_tp_find(struct tcf_chain *chain, 1676 struct tcf_chain_info *chain_info, 1677 u32 protocol, u32 prio, 1678 bool prio_allocate); 1679 1680 /* Try to insert new proto. 1681 * If proto with specified priority already exists, free new proto 1682 * and return existing one. 1683 */ 1684 1685 static struct tcf_proto *tcf_chain_tp_insert_unique(struct tcf_chain *chain, 1686 struct tcf_proto *tp_new, 1687 u32 protocol, u32 prio, 1688 bool rtnl_held) 1689 { 1690 struct tcf_chain_info chain_info; 1691 struct tcf_proto *tp; 1692 int err = 0; 1693 1694 mutex_lock(&chain->filter_chain_lock); 1695 1696 if (tcf_proto_exists_destroying(chain, tp_new)) { 1697 mutex_unlock(&chain->filter_chain_lock); 1698 tcf_proto_destroy(tp_new, rtnl_held, false, NULL); 1699 return ERR_PTR(-EAGAIN); 1700 } 1701 1702 tp = tcf_chain_tp_find(chain, &chain_info, 1703 protocol, prio, false); 1704 if (!tp) 1705 err = tcf_chain_tp_insert(chain, &chain_info, tp_new); 1706 mutex_unlock(&chain->filter_chain_lock); 1707 1708 if (tp) { 1709 tcf_proto_destroy(tp_new, rtnl_held, false, NULL); 1710 tp_new = tp; 1711 } else if (err) { 1712 tcf_proto_destroy(tp_new, rtnl_held, false, NULL); 1713 tp_new = ERR_PTR(err); 1714 } 1715 1716 return tp_new; 1717 } 1718 1719 static void tcf_chain_tp_delete_empty(struct tcf_chain *chain, 1720 struct tcf_proto *tp, bool rtnl_held, 1721 struct netlink_ext_ack *extack) 1722 { 1723 struct tcf_chain_info chain_info; 1724 struct tcf_proto *tp_iter; 1725 struct tcf_proto **pprev; 1726 struct tcf_proto *next; 1727 1728 mutex_lock(&chain->filter_chain_lock); 1729 1730 /* Atomically find and remove tp from chain. */ 1731 for (pprev = &chain->filter_chain; 1732 (tp_iter = tcf_chain_dereference(*pprev, chain)); 1733 pprev = &tp_iter->next) { 1734 if (tp_iter == tp) { 1735 chain_info.pprev = pprev; 1736 chain_info.next = tp_iter->next; 1737 WARN_ON(tp_iter->deleting); 1738 break; 1739 } 1740 } 1741 /* Verify that tp still exists and no new filters were inserted 1742 * concurrently. 1743 * Mark tp for deletion if it is empty. 1744 */ 1745 if (!tp_iter || !tcf_proto_check_delete(tp)) { 1746 mutex_unlock(&chain->filter_chain_lock); 1747 return; 1748 } 1749 1750 tcf_proto_signal_destroying(chain, tp); 1751 next = tcf_chain_dereference(chain_info.next, chain); 1752 if (tp == chain->filter_chain) 1753 tcf_chain0_head_change(chain, next); 1754 RCU_INIT_POINTER(*chain_info.pprev, next); 1755 mutex_unlock(&chain->filter_chain_lock); 1756 1757 tcf_proto_put(tp, rtnl_held, extack); 1758 } 1759 1760 static struct tcf_proto *tcf_chain_tp_find(struct tcf_chain *chain, 1761 struct tcf_chain_info *chain_info, 1762 u32 protocol, u32 prio, 1763 bool prio_allocate) 1764 { 1765 struct tcf_proto **pprev; 1766 struct tcf_proto *tp; 1767 1768 /* Check the chain for existence of proto-tcf with this priority */ 1769 for (pprev = &chain->filter_chain; 1770 (tp = tcf_chain_dereference(*pprev, chain)); 1771 pprev = &tp->next) { 1772 if (tp->prio >= prio) { 1773 if (tp->prio == prio) { 1774 if (prio_allocate || 1775 (tp->protocol != protocol && protocol)) 1776 return ERR_PTR(-EINVAL); 1777 } else { 1778 tp = NULL; 1779 } 1780 break; 1781 } 1782 } 1783 chain_info->pprev = pprev; 1784 if (tp) { 1785 chain_info->next = tp->next; 1786 tcf_proto_get(tp); 1787 } else { 1788 chain_info->next = NULL; 1789 } 1790 return tp; 1791 } 1792 1793 static int tcf_fill_node(struct net *net, struct sk_buff *skb, 1794 struct tcf_proto *tp, struct tcf_block *block, 1795 struct Qdisc *q, u32 parent, void *fh, 1796 u32 portid, u32 seq, u16 flags, int event, 1797 bool terse_dump, bool rtnl_held) 1798 { 1799 struct tcmsg *tcm; 1800 struct nlmsghdr *nlh; 1801 unsigned char *b = skb_tail_pointer(skb); 1802 1803 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags); 1804 if (!nlh) 1805 goto out_nlmsg_trim; 1806 tcm = nlmsg_data(nlh); 1807 tcm->tcm_family = AF_UNSPEC; 1808 tcm->tcm__pad1 = 0; 1809 tcm->tcm__pad2 = 0; 1810 if (q) { 1811 tcm->tcm_ifindex = qdisc_dev(q)->ifindex; 1812 tcm->tcm_parent = parent; 1813 } else { 1814 tcm->tcm_ifindex = TCM_IFINDEX_MAGIC_BLOCK; 1815 tcm->tcm_block_index = block->index; 1816 } 1817 tcm->tcm_info = TC_H_MAKE(tp->prio, tp->protocol); 1818 if (nla_put_string(skb, TCA_KIND, tp->ops->kind)) 1819 goto nla_put_failure; 1820 if (nla_put_u32(skb, TCA_CHAIN, tp->chain->index)) 1821 goto nla_put_failure; 1822 if (!fh) { 1823 tcm->tcm_handle = 0; 1824 } else if (terse_dump) { 1825 if (tp->ops->terse_dump) { 1826 if (tp->ops->terse_dump(net, tp, fh, skb, tcm, 1827 rtnl_held) < 0) 1828 goto nla_put_failure; 1829 } else { 1830 goto cls_op_not_supp; 1831 } 1832 } else { 1833 if (tp->ops->dump && 1834 tp->ops->dump(net, tp, fh, skb, tcm, rtnl_held) < 0) 1835 goto nla_put_failure; 1836 } 1837 nlh->nlmsg_len = skb_tail_pointer(skb) - b; 1838 return skb->len; 1839 1840 out_nlmsg_trim: 1841 nla_put_failure: 1842 cls_op_not_supp: 1843 nlmsg_trim(skb, b); 1844 return -1; 1845 } 1846 1847 static int tfilter_notify(struct net *net, struct sk_buff *oskb, 1848 struct nlmsghdr *n, struct tcf_proto *tp, 1849 struct tcf_block *block, struct Qdisc *q, 1850 u32 parent, void *fh, int event, bool unicast, 1851 bool rtnl_held) 1852 { 1853 struct sk_buff *skb; 1854 u32 portid = oskb ? NETLINK_CB(oskb).portid : 0; 1855 int err = 0; 1856 1857 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL); 1858 if (!skb) 1859 return -ENOBUFS; 1860 1861 if (tcf_fill_node(net, skb, tp, block, q, parent, fh, portid, 1862 n->nlmsg_seq, n->nlmsg_flags, event, 1863 false, rtnl_held) <= 0) { 1864 kfree_skb(skb); 1865 return -EINVAL; 1866 } 1867 1868 if (unicast) 1869 err = rtnl_unicast(skb, net, portid); 1870 else 1871 err = rtnetlink_send(skb, net, portid, RTNLGRP_TC, 1872 n->nlmsg_flags & NLM_F_ECHO); 1873 return err; 1874 } 1875 1876 static int tfilter_del_notify(struct net *net, struct sk_buff *oskb, 1877 struct nlmsghdr *n, struct tcf_proto *tp, 1878 struct tcf_block *block, struct Qdisc *q, 1879 u32 parent, void *fh, bool unicast, bool *last, 1880 bool rtnl_held, struct netlink_ext_ack *extack) 1881 { 1882 struct sk_buff *skb; 1883 u32 portid = oskb ? NETLINK_CB(oskb).portid : 0; 1884 int err; 1885 1886 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL); 1887 if (!skb) 1888 return -ENOBUFS; 1889 1890 if (tcf_fill_node(net, skb, tp, block, q, parent, fh, portid, 1891 n->nlmsg_seq, n->nlmsg_flags, RTM_DELTFILTER, 1892 false, rtnl_held) <= 0) { 1893 NL_SET_ERR_MSG(extack, "Failed to build del event notification"); 1894 kfree_skb(skb); 1895 return -EINVAL; 1896 } 1897 1898 err = tp->ops->delete(tp, fh, last, rtnl_held, extack); 1899 if (err) { 1900 kfree_skb(skb); 1901 return err; 1902 } 1903 1904 if (unicast) 1905 err = rtnl_unicast(skb, net, portid); 1906 else 1907 err = rtnetlink_send(skb, net, portid, RTNLGRP_TC, 1908 n->nlmsg_flags & NLM_F_ECHO); 1909 if (err < 0) 1910 NL_SET_ERR_MSG(extack, "Failed to send filter delete notification"); 1911 1912 return err; 1913 } 1914 1915 static void tfilter_notify_chain(struct net *net, struct sk_buff *oskb, 1916 struct tcf_block *block, struct Qdisc *q, 1917 u32 parent, struct nlmsghdr *n, 1918 struct tcf_chain *chain, int event) 1919 { 1920 struct tcf_proto *tp; 1921 1922 for (tp = tcf_get_next_proto(chain, NULL); 1923 tp; tp = tcf_get_next_proto(chain, tp)) 1924 tfilter_notify(net, oskb, n, tp, block, 1925 q, parent, NULL, event, false, true); 1926 } 1927 1928 static void tfilter_put(struct tcf_proto *tp, void *fh) 1929 { 1930 if (tp->ops->put && fh) 1931 tp->ops->put(tp, fh); 1932 } 1933 1934 static int tc_new_tfilter(struct sk_buff *skb, struct nlmsghdr *n, 1935 struct netlink_ext_ack *extack) 1936 { 1937 struct net *net = sock_net(skb->sk); 1938 struct nlattr *tca[TCA_MAX + 1]; 1939 char name[IFNAMSIZ]; 1940 struct tcmsg *t; 1941 u32 protocol; 1942 u32 prio; 1943 bool prio_allocate; 1944 u32 parent; 1945 u32 chain_index; 1946 struct Qdisc *q = NULL; 1947 struct tcf_chain_info chain_info; 1948 struct tcf_chain *chain = NULL; 1949 struct tcf_block *block; 1950 struct tcf_proto *tp; 1951 unsigned long cl; 1952 void *fh; 1953 int err; 1954 int tp_created; 1955 bool rtnl_held = false; 1956 u32 flags; 1957 1958 if (!netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN)) 1959 return -EPERM; 1960 1961 replay: 1962 tp_created = 0; 1963 1964 err = nlmsg_parse_deprecated(n, sizeof(*t), tca, TCA_MAX, 1965 rtm_tca_policy, extack); 1966 if (err < 0) 1967 return err; 1968 1969 t = nlmsg_data(n); 1970 protocol = TC_H_MIN(t->tcm_info); 1971 prio = TC_H_MAJ(t->tcm_info); 1972 prio_allocate = false; 1973 parent = t->tcm_parent; 1974 tp = NULL; 1975 cl = 0; 1976 block = NULL; 1977 flags = 0; 1978 1979 if (prio == 0) { 1980 /* If no priority is provided by the user, 1981 * we allocate one. 1982 */ 1983 if (n->nlmsg_flags & NLM_F_CREATE) { 1984 prio = TC_H_MAKE(0x80000000U, 0U); 1985 prio_allocate = true; 1986 } else { 1987 NL_SET_ERR_MSG(extack, "Invalid filter command with priority of zero"); 1988 return -ENOENT; 1989 } 1990 } 1991 1992 /* Find head of filter chain. */ 1993 1994 err = __tcf_qdisc_find(net, &q, &parent, t->tcm_ifindex, false, extack); 1995 if (err) 1996 return err; 1997 1998 if (tcf_proto_check_kind(tca[TCA_KIND], name)) { 1999 NL_SET_ERR_MSG(extack, "Specified TC filter name too long"); 2000 err = -EINVAL; 2001 goto errout; 2002 } 2003 2004 /* Take rtnl mutex if rtnl_held was set to true on previous iteration, 2005 * block is shared (no qdisc found), qdisc is not unlocked, classifier 2006 * type is not specified, classifier is not unlocked. 2007 */ 2008 if (rtnl_held || 2009 (q && !(q->ops->cl_ops->flags & QDISC_CLASS_OPS_DOIT_UNLOCKED)) || 2010 !tcf_proto_is_unlocked(name)) { 2011 rtnl_held = true; 2012 rtnl_lock(); 2013 } 2014 2015 err = __tcf_qdisc_cl_find(q, parent, &cl, t->tcm_ifindex, extack); 2016 if (err) 2017 goto errout; 2018 2019 block = __tcf_block_find(net, q, cl, t->tcm_ifindex, t->tcm_block_index, 2020 extack); 2021 if (IS_ERR(block)) { 2022 err = PTR_ERR(block); 2023 goto errout; 2024 } 2025 block->classid = parent; 2026 2027 chain_index = tca[TCA_CHAIN] ? nla_get_u32(tca[TCA_CHAIN]) : 0; 2028 if (chain_index > TC_ACT_EXT_VAL_MASK) { 2029 NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit"); 2030 err = -EINVAL; 2031 goto errout; 2032 } 2033 chain = tcf_chain_get(block, chain_index, true); 2034 if (!chain) { 2035 NL_SET_ERR_MSG(extack, "Cannot create specified filter chain"); 2036 err = -ENOMEM; 2037 goto errout; 2038 } 2039 2040 mutex_lock(&chain->filter_chain_lock); 2041 tp = tcf_chain_tp_find(chain, &chain_info, protocol, 2042 prio, prio_allocate); 2043 if (IS_ERR(tp)) { 2044 NL_SET_ERR_MSG(extack, "Filter with specified priority/protocol not found"); 2045 err = PTR_ERR(tp); 2046 goto errout_locked; 2047 } 2048 2049 if (tp == NULL) { 2050 struct tcf_proto *tp_new = NULL; 2051 2052 if (chain->flushing) { 2053 err = -EAGAIN; 2054 goto errout_locked; 2055 } 2056 2057 /* Proto-tcf does not exist, create new one */ 2058 2059 if (tca[TCA_KIND] == NULL || !protocol) { 2060 NL_SET_ERR_MSG(extack, "Filter kind and protocol must be specified"); 2061 err = -EINVAL; 2062 goto errout_locked; 2063 } 2064 2065 if (!(n->nlmsg_flags & NLM_F_CREATE)) { 2066 NL_SET_ERR_MSG(extack, "Need both RTM_NEWTFILTER and NLM_F_CREATE to create a new filter"); 2067 err = -ENOENT; 2068 goto errout_locked; 2069 } 2070 2071 if (prio_allocate) 2072 prio = tcf_auto_prio(tcf_chain_tp_prev(chain, 2073 &chain_info)); 2074 2075 mutex_unlock(&chain->filter_chain_lock); 2076 tp_new = tcf_proto_create(name, protocol, prio, chain, 2077 rtnl_held, extack); 2078 if (IS_ERR(tp_new)) { 2079 err = PTR_ERR(tp_new); 2080 goto errout_tp; 2081 } 2082 2083 tp_created = 1; 2084 tp = tcf_chain_tp_insert_unique(chain, tp_new, protocol, prio, 2085 rtnl_held); 2086 if (IS_ERR(tp)) { 2087 err = PTR_ERR(tp); 2088 goto errout_tp; 2089 } 2090 } else { 2091 mutex_unlock(&chain->filter_chain_lock); 2092 } 2093 2094 if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind)) { 2095 NL_SET_ERR_MSG(extack, "Specified filter kind does not match existing one"); 2096 err = -EINVAL; 2097 goto errout; 2098 } 2099 2100 fh = tp->ops->get(tp, t->tcm_handle); 2101 2102 if (!fh) { 2103 if (!(n->nlmsg_flags & NLM_F_CREATE)) { 2104 NL_SET_ERR_MSG(extack, "Need both RTM_NEWTFILTER and NLM_F_CREATE to create a new filter"); 2105 err = -ENOENT; 2106 goto errout; 2107 } 2108 } else if (n->nlmsg_flags & NLM_F_EXCL) { 2109 tfilter_put(tp, fh); 2110 NL_SET_ERR_MSG(extack, "Filter already exists"); 2111 err = -EEXIST; 2112 goto errout; 2113 } 2114 2115 if (chain->tmplt_ops && chain->tmplt_ops != tp->ops) { 2116 NL_SET_ERR_MSG(extack, "Chain template is set to a different filter kind"); 2117 err = -EINVAL; 2118 goto errout; 2119 } 2120 2121 if (!(n->nlmsg_flags & NLM_F_CREATE)) 2122 flags |= TCA_ACT_FLAGS_REPLACE; 2123 if (!rtnl_held) 2124 flags |= TCA_ACT_FLAGS_NO_RTNL; 2125 err = tp->ops->change(net, skb, tp, cl, t->tcm_handle, tca, &fh, 2126 flags, extack); 2127 if (err == 0) { 2128 tfilter_notify(net, skb, n, tp, block, q, parent, fh, 2129 RTM_NEWTFILTER, false, rtnl_held); 2130 tfilter_put(tp, fh); 2131 /* q pointer is NULL for shared blocks */ 2132 if (q) 2133 q->flags &= ~TCQ_F_CAN_BYPASS; 2134 } 2135 2136 errout: 2137 if (err && tp_created) 2138 tcf_chain_tp_delete_empty(chain, tp, rtnl_held, NULL); 2139 errout_tp: 2140 if (chain) { 2141 if (tp && !IS_ERR(tp)) 2142 tcf_proto_put(tp, rtnl_held, NULL); 2143 if (!tp_created) 2144 tcf_chain_put(chain); 2145 } 2146 tcf_block_release(q, block, rtnl_held); 2147 2148 if (rtnl_held) 2149 rtnl_unlock(); 2150 2151 if (err == -EAGAIN) { 2152 /* Take rtnl lock in case EAGAIN is caused by concurrent flush 2153 * of target chain. 2154 */ 2155 rtnl_held = true; 2156 /* Replay the request. */ 2157 goto replay; 2158 } 2159 return err; 2160 2161 errout_locked: 2162 mutex_unlock(&chain->filter_chain_lock); 2163 goto errout; 2164 } 2165 2166 static int tc_del_tfilter(struct sk_buff *skb, struct nlmsghdr *n, 2167 struct netlink_ext_ack *extack) 2168 { 2169 struct net *net = sock_net(skb->sk); 2170 struct nlattr *tca[TCA_MAX + 1]; 2171 char name[IFNAMSIZ]; 2172 struct tcmsg *t; 2173 u32 protocol; 2174 u32 prio; 2175 u32 parent; 2176 u32 chain_index; 2177 struct Qdisc *q = NULL; 2178 struct tcf_chain_info chain_info; 2179 struct tcf_chain *chain = NULL; 2180 struct tcf_block *block = NULL; 2181 struct tcf_proto *tp = NULL; 2182 unsigned long cl = 0; 2183 void *fh = NULL; 2184 int err; 2185 bool rtnl_held = false; 2186 2187 if (!netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN)) 2188 return -EPERM; 2189 2190 err = nlmsg_parse_deprecated(n, sizeof(*t), tca, TCA_MAX, 2191 rtm_tca_policy, extack); 2192 if (err < 0) 2193 return err; 2194 2195 t = nlmsg_data(n); 2196 protocol = TC_H_MIN(t->tcm_info); 2197 prio = TC_H_MAJ(t->tcm_info); 2198 parent = t->tcm_parent; 2199 2200 if (prio == 0 && (protocol || t->tcm_handle || tca[TCA_KIND])) { 2201 NL_SET_ERR_MSG(extack, "Cannot flush filters with protocol, handle or kind set"); 2202 return -ENOENT; 2203 } 2204 2205 /* Find head of filter chain. */ 2206 2207 err = __tcf_qdisc_find(net, &q, &parent, t->tcm_ifindex, false, extack); 2208 if (err) 2209 return err; 2210 2211 if (tcf_proto_check_kind(tca[TCA_KIND], name)) { 2212 NL_SET_ERR_MSG(extack, "Specified TC filter name too long"); 2213 err = -EINVAL; 2214 goto errout; 2215 } 2216 /* Take rtnl mutex if flushing whole chain, block is shared (no qdisc 2217 * found), qdisc is not unlocked, classifier type is not specified, 2218 * classifier is not unlocked. 2219 */ 2220 if (!prio || 2221 (q && !(q->ops->cl_ops->flags & QDISC_CLASS_OPS_DOIT_UNLOCKED)) || 2222 !tcf_proto_is_unlocked(name)) { 2223 rtnl_held = true; 2224 rtnl_lock(); 2225 } 2226 2227 err = __tcf_qdisc_cl_find(q, parent, &cl, t->tcm_ifindex, extack); 2228 if (err) 2229 goto errout; 2230 2231 block = __tcf_block_find(net, q, cl, t->tcm_ifindex, t->tcm_block_index, 2232 extack); 2233 if (IS_ERR(block)) { 2234 err = PTR_ERR(block); 2235 goto errout; 2236 } 2237 2238 chain_index = tca[TCA_CHAIN] ? nla_get_u32(tca[TCA_CHAIN]) : 0; 2239 if (chain_index > TC_ACT_EXT_VAL_MASK) { 2240 NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit"); 2241 err = -EINVAL; 2242 goto errout; 2243 } 2244 chain = tcf_chain_get(block, chain_index, false); 2245 if (!chain) { 2246 /* User requested flush on non-existent chain. Nothing to do, 2247 * so just return success. 2248 */ 2249 if (prio == 0) { 2250 err = 0; 2251 goto errout; 2252 } 2253 NL_SET_ERR_MSG(extack, "Cannot find specified filter chain"); 2254 err = -ENOENT; 2255 goto errout; 2256 } 2257 2258 if (prio == 0) { 2259 tfilter_notify_chain(net, skb, block, q, parent, n, 2260 chain, RTM_DELTFILTER); 2261 tcf_chain_flush(chain, rtnl_held); 2262 err = 0; 2263 goto errout; 2264 } 2265 2266 mutex_lock(&chain->filter_chain_lock); 2267 tp = tcf_chain_tp_find(chain, &chain_info, protocol, 2268 prio, false); 2269 if (!tp || IS_ERR(tp)) { 2270 NL_SET_ERR_MSG(extack, "Filter with specified priority/protocol not found"); 2271 err = tp ? PTR_ERR(tp) : -ENOENT; 2272 goto errout_locked; 2273 } else if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind)) { 2274 NL_SET_ERR_MSG(extack, "Specified filter kind does not match existing one"); 2275 err = -EINVAL; 2276 goto errout_locked; 2277 } else if (t->tcm_handle == 0) { 2278 tcf_proto_signal_destroying(chain, tp); 2279 tcf_chain_tp_remove(chain, &chain_info, tp); 2280 mutex_unlock(&chain->filter_chain_lock); 2281 2282 tcf_proto_put(tp, rtnl_held, NULL); 2283 tfilter_notify(net, skb, n, tp, block, q, parent, fh, 2284 RTM_DELTFILTER, false, rtnl_held); 2285 err = 0; 2286 goto errout; 2287 } 2288 mutex_unlock(&chain->filter_chain_lock); 2289 2290 fh = tp->ops->get(tp, t->tcm_handle); 2291 2292 if (!fh) { 2293 NL_SET_ERR_MSG(extack, "Specified filter handle not found"); 2294 err = -ENOENT; 2295 } else { 2296 bool last; 2297 2298 err = tfilter_del_notify(net, skb, n, tp, block, 2299 q, parent, fh, false, &last, 2300 rtnl_held, extack); 2301 2302 if (err) 2303 goto errout; 2304 if (last) 2305 tcf_chain_tp_delete_empty(chain, tp, rtnl_held, extack); 2306 } 2307 2308 errout: 2309 if (chain) { 2310 if (tp && !IS_ERR(tp)) 2311 tcf_proto_put(tp, rtnl_held, NULL); 2312 tcf_chain_put(chain); 2313 } 2314 tcf_block_release(q, block, rtnl_held); 2315 2316 if (rtnl_held) 2317 rtnl_unlock(); 2318 2319 return err; 2320 2321 errout_locked: 2322 mutex_unlock(&chain->filter_chain_lock); 2323 goto errout; 2324 } 2325 2326 static int tc_get_tfilter(struct sk_buff *skb, struct nlmsghdr *n, 2327 struct netlink_ext_ack *extack) 2328 { 2329 struct net *net = sock_net(skb->sk); 2330 struct nlattr *tca[TCA_MAX + 1]; 2331 char name[IFNAMSIZ]; 2332 struct tcmsg *t; 2333 u32 protocol; 2334 u32 prio; 2335 u32 parent; 2336 u32 chain_index; 2337 struct Qdisc *q = NULL; 2338 struct tcf_chain_info chain_info; 2339 struct tcf_chain *chain = NULL; 2340 struct tcf_block *block = NULL; 2341 struct tcf_proto *tp = NULL; 2342 unsigned long cl = 0; 2343 void *fh = NULL; 2344 int err; 2345 bool rtnl_held = false; 2346 2347 err = nlmsg_parse_deprecated(n, sizeof(*t), tca, TCA_MAX, 2348 rtm_tca_policy, extack); 2349 if (err < 0) 2350 return err; 2351 2352 t = nlmsg_data(n); 2353 protocol = TC_H_MIN(t->tcm_info); 2354 prio = TC_H_MAJ(t->tcm_info); 2355 parent = t->tcm_parent; 2356 2357 if (prio == 0) { 2358 NL_SET_ERR_MSG(extack, "Invalid filter command with priority of zero"); 2359 return -ENOENT; 2360 } 2361 2362 /* Find head of filter chain. */ 2363 2364 err = __tcf_qdisc_find(net, &q, &parent, t->tcm_ifindex, false, extack); 2365 if (err) 2366 return err; 2367 2368 if (tcf_proto_check_kind(tca[TCA_KIND], name)) { 2369 NL_SET_ERR_MSG(extack, "Specified TC filter name too long"); 2370 err = -EINVAL; 2371 goto errout; 2372 } 2373 /* Take rtnl mutex if block is shared (no qdisc found), qdisc is not 2374 * unlocked, classifier type is not specified, classifier is not 2375 * unlocked. 2376 */ 2377 if ((q && !(q->ops->cl_ops->flags & QDISC_CLASS_OPS_DOIT_UNLOCKED)) || 2378 !tcf_proto_is_unlocked(name)) { 2379 rtnl_held = true; 2380 rtnl_lock(); 2381 } 2382 2383 err = __tcf_qdisc_cl_find(q, parent, &cl, t->tcm_ifindex, extack); 2384 if (err) 2385 goto errout; 2386 2387 block = __tcf_block_find(net, q, cl, t->tcm_ifindex, t->tcm_block_index, 2388 extack); 2389 if (IS_ERR(block)) { 2390 err = PTR_ERR(block); 2391 goto errout; 2392 } 2393 2394 chain_index = tca[TCA_CHAIN] ? nla_get_u32(tca[TCA_CHAIN]) : 0; 2395 if (chain_index > TC_ACT_EXT_VAL_MASK) { 2396 NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit"); 2397 err = -EINVAL; 2398 goto errout; 2399 } 2400 chain = tcf_chain_get(block, chain_index, false); 2401 if (!chain) { 2402 NL_SET_ERR_MSG(extack, "Cannot find specified filter chain"); 2403 err = -EINVAL; 2404 goto errout; 2405 } 2406 2407 mutex_lock(&chain->filter_chain_lock); 2408 tp = tcf_chain_tp_find(chain, &chain_info, protocol, 2409 prio, false); 2410 mutex_unlock(&chain->filter_chain_lock); 2411 if (!tp || IS_ERR(tp)) { 2412 NL_SET_ERR_MSG(extack, "Filter with specified priority/protocol not found"); 2413 err = tp ? PTR_ERR(tp) : -ENOENT; 2414 goto errout; 2415 } else if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind)) { 2416 NL_SET_ERR_MSG(extack, "Specified filter kind does not match existing one"); 2417 err = -EINVAL; 2418 goto errout; 2419 } 2420 2421 fh = tp->ops->get(tp, t->tcm_handle); 2422 2423 if (!fh) { 2424 NL_SET_ERR_MSG(extack, "Specified filter handle not found"); 2425 err = -ENOENT; 2426 } else { 2427 err = tfilter_notify(net, skb, n, tp, block, q, parent, 2428 fh, RTM_NEWTFILTER, true, rtnl_held); 2429 if (err < 0) 2430 NL_SET_ERR_MSG(extack, "Failed to send filter notify message"); 2431 } 2432 2433 tfilter_put(tp, fh); 2434 errout: 2435 if (chain) { 2436 if (tp && !IS_ERR(tp)) 2437 tcf_proto_put(tp, rtnl_held, NULL); 2438 tcf_chain_put(chain); 2439 } 2440 tcf_block_release(q, block, rtnl_held); 2441 2442 if (rtnl_held) 2443 rtnl_unlock(); 2444 2445 return err; 2446 } 2447 2448 struct tcf_dump_args { 2449 struct tcf_walker w; 2450 struct sk_buff *skb; 2451 struct netlink_callback *cb; 2452 struct tcf_block *block; 2453 struct Qdisc *q; 2454 u32 parent; 2455 bool terse_dump; 2456 }; 2457 2458 static int tcf_node_dump(struct tcf_proto *tp, void *n, struct tcf_walker *arg) 2459 { 2460 struct tcf_dump_args *a = (void *)arg; 2461 struct net *net = sock_net(a->skb->sk); 2462 2463 return tcf_fill_node(net, a->skb, tp, a->block, a->q, a->parent, 2464 n, NETLINK_CB(a->cb->skb).portid, 2465 a->cb->nlh->nlmsg_seq, NLM_F_MULTI, 2466 RTM_NEWTFILTER, a->terse_dump, true); 2467 } 2468 2469 static bool tcf_chain_dump(struct tcf_chain *chain, struct Qdisc *q, u32 parent, 2470 struct sk_buff *skb, struct netlink_callback *cb, 2471 long index_start, long *p_index, bool terse) 2472 { 2473 struct net *net = sock_net(skb->sk); 2474 struct tcf_block *block = chain->block; 2475 struct tcmsg *tcm = nlmsg_data(cb->nlh); 2476 struct tcf_proto *tp, *tp_prev; 2477 struct tcf_dump_args arg; 2478 2479 for (tp = __tcf_get_next_proto(chain, NULL); 2480 tp; 2481 tp_prev = tp, 2482 tp = __tcf_get_next_proto(chain, tp), 2483 tcf_proto_put(tp_prev, true, NULL), 2484 (*p_index)++) { 2485 if (*p_index < index_start) 2486 continue; 2487 if (TC_H_MAJ(tcm->tcm_info) && 2488 TC_H_MAJ(tcm->tcm_info) != tp->prio) 2489 continue; 2490 if (TC_H_MIN(tcm->tcm_info) && 2491 TC_H_MIN(tcm->tcm_info) != tp->protocol) 2492 continue; 2493 if (*p_index > index_start) 2494 memset(&cb->args[1], 0, 2495 sizeof(cb->args) - sizeof(cb->args[0])); 2496 if (cb->args[1] == 0) { 2497 if (tcf_fill_node(net, skb, tp, block, q, parent, NULL, 2498 NETLINK_CB(cb->skb).portid, 2499 cb->nlh->nlmsg_seq, NLM_F_MULTI, 2500 RTM_NEWTFILTER, false, true) <= 0) 2501 goto errout; 2502 cb->args[1] = 1; 2503 } 2504 if (!tp->ops->walk) 2505 continue; 2506 arg.w.fn = tcf_node_dump; 2507 arg.skb = skb; 2508 arg.cb = cb; 2509 arg.block = block; 2510 arg.q = q; 2511 arg.parent = parent; 2512 arg.w.stop = 0; 2513 arg.w.skip = cb->args[1] - 1; 2514 arg.w.count = 0; 2515 arg.w.cookie = cb->args[2]; 2516 arg.terse_dump = terse; 2517 tp->ops->walk(tp, &arg.w, true); 2518 cb->args[2] = arg.w.cookie; 2519 cb->args[1] = arg.w.count + 1; 2520 if (arg.w.stop) 2521 goto errout; 2522 } 2523 return true; 2524 2525 errout: 2526 tcf_proto_put(tp, true, NULL); 2527 return false; 2528 } 2529 2530 static const struct nla_policy tcf_tfilter_dump_policy[TCA_MAX + 1] = { 2531 [TCA_DUMP_FLAGS] = NLA_POLICY_BITFIELD32(TCA_DUMP_FLAGS_TERSE), 2532 }; 2533 2534 /* called with RTNL */ 2535 static int tc_dump_tfilter(struct sk_buff *skb, struct netlink_callback *cb) 2536 { 2537 struct tcf_chain *chain, *chain_prev; 2538 struct net *net = sock_net(skb->sk); 2539 struct nlattr *tca[TCA_MAX + 1]; 2540 struct Qdisc *q = NULL; 2541 struct tcf_block *block; 2542 struct tcmsg *tcm = nlmsg_data(cb->nlh); 2543 bool terse_dump = false; 2544 long index_start; 2545 long index; 2546 u32 parent; 2547 int err; 2548 2549 if (nlmsg_len(cb->nlh) < sizeof(*tcm)) 2550 return skb->len; 2551 2552 err = nlmsg_parse_deprecated(cb->nlh, sizeof(*tcm), tca, TCA_MAX, 2553 tcf_tfilter_dump_policy, cb->extack); 2554 if (err) 2555 return err; 2556 2557 if (tca[TCA_DUMP_FLAGS]) { 2558 struct nla_bitfield32 flags = 2559 nla_get_bitfield32(tca[TCA_DUMP_FLAGS]); 2560 2561 terse_dump = flags.value & TCA_DUMP_FLAGS_TERSE; 2562 } 2563 2564 if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK) { 2565 block = tcf_block_refcnt_get(net, tcm->tcm_block_index); 2566 if (!block) 2567 goto out; 2568 /* If we work with block index, q is NULL and parent value 2569 * will never be used in the following code. The check 2570 * in tcf_fill_node prevents it. However, compiler does not 2571 * see that far, so set parent to zero to silence the warning 2572 * about parent being uninitialized. 2573 */ 2574 parent = 0; 2575 } else { 2576 const struct Qdisc_class_ops *cops; 2577 struct net_device *dev; 2578 unsigned long cl = 0; 2579 2580 dev = __dev_get_by_index(net, tcm->tcm_ifindex); 2581 if (!dev) 2582 return skb->len; 2583 2584 parent = tcm->tcm_parent; 2585 if (!parent) 2586 q = dev->qdisc; 2587 else 2588 q = qdisc_lookup(dev, TC_H_MAJ(tcm->tcm_parent)); 2589 if (!q) 2590 goto out; 2591 cops = q->ops->cl_ops; 2592 if (!cops) 2593 goto out; 2594 if (!cops->tcf_block) 2595 goto out; 2596 if (TC_H_MIN(tcm->tcm_parent)) { 2597 cl = cops->find(q, tcm->tcm_parent); 2598 if (cl == 0) 2599 goto out; 2600 } 2601 block = cops->tcf_block(q, cl, NULL); 2602 if (!block) 2603 goto out; 2604 parent = block->classid; 2605 if (tcf_block_shared(block)) 2606 q = NULL; 2607 } 2608 2609 index_start = cb->args[0]; 2610 index = 0; 2611 2612 for (chain = __tcf_get_next_chain(block, NULL); 2613 chain; 2614 chain_prev = chain, 2615 chain = __tcf_get_next_chain(block, chain), 2616 tcf_chain_put(chain_prev)) { 2617 if (tca[TCA_CHAIN] && 2618 nla_get_u32(tca[TCA_CHAIN]) != chain->index) 2619 continue; 2620 if (!tcf_chain_dump(chain, q, parent, skb, cb, 2621 index_start, &index, terse_dump)) { 2622 tcf_chain_put(chain); 2623 err = -EMSGSIZE; 2624 break; 2625 } 2626 } 2627 2628 if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK) 2629 tcf_block_refcnt_put(block, true); 2630 cb->args[0] = index; 2631 2632 out: 2633 /* If we did no progress, the error (EMSGSIZE) is real */ 2634 if (skb->len == 0 && err) 2635 return err; 2636 return skb->len; 2637 } 2638 2639 static int tc_chain_fill_node(const struct tcf_proto_ops *tmplt_ops, 2640 void *tmplt_priv, u32 chain_index, 2641 struct net *net, struct sk_buff *skb, 2642 struct tcf_block *block, 2643 u32 portid, u32 seq, u16 flags, int event) 2644 { 2645 unsigned char *b = skb_tail_pointer(skb); 2646 const struct tcf_proto_ops *ops; 2647 struct nlmsghdr *nlh; 2648 struct tcmsg *tcm; 2649 void *priv; 2650 2651 ops = tmplt_ops; 2652 priv = tmplt_priv; 2653 2654 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags); 2655 if (!nlh) 2656 goto out_nlmsg_trim; 2657 tcm = nlmsg_data(nlh); 2658 tcm->tcm_family = AF_UNSPEC; 2659 tcm->tcm__pad1 = 0; 2660 tcm->tcm__pad2 = 0; 2661 tcm->tcm_handle = 0; 2662 if (block->q) { 2663 tcm->tcm_ifindex = qdisc_dev(block->q)->ifindex; 2664 tcm->tcm_parent = block->q->handle; 2665 } else { 2666 tcm->tcm_ifindex = TCM_IFINDEX_MAGIC_BLOCK; 2667 tcm->tcm_block_index = block->index; 2668 } 2669 2670 if (nla_put_u32(skb, TCA_CHAIN, chain_index)) 2671 goto nla_put_failure; 2672 2673 if (ops) { 2674 if (nla_put_string(skb, TCA_KIND, ops->kind)) 2675 goto nla_put_failure; 2676 if (ops->tmplt_dump(skb, net, priv) < 0) 2677 goto nla_put_failure; 2678 } 2679 2680 nlh->nlmsg_len = skb_tail_pointer(skb) - b; 2681 return skb->len; 2682 2683 out_nlmsg_trim: 2684 nla_put_failure: 2685 nlmsg_trim(skb, b); 2686 return -EMSGSIZE; 2687 } 2688 2689 static int tc_chain_notify(struct tcf_chain *chain, struct sk_buff *oskb, 2690 u32 seq, u16 flags, int event, bool unicast) 2691 { 2692 u32 portid = oskb ? NETLINK_CB(oskb).portid : 0; 2693 struct tcf_block *block = chain->block; 2694 struct net *net = block->net; 2695 struct sk_buff *skb; 2696 int err = 0; 2697 2698 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL); 2699 if (!skb) 2700 return -ENOBUFS; 2701 2702 if (tc_chain_fill_node(chain->tmplt_ops, chain->tmplt_priv, 2703 chain->index, net, skb, block, portid, 2704 seq, flags, event) <= 0) { 2705 kfree_skb(skb); 2706 return -EINVAL; 2707 } 2708 2709 if (unicast) 2710 err = rtnl_unicast(skb, net, portid); 2711 else 2712 err = rtnetlink_send(skb, net, portid, RTNLGRP_TC, 2713 flags & NLM_F_ECHO); 2714 2715 return err; 2716 } 2717 2718 static int tc_chain_notify_delete(const struct tcf_proto_ops *tmplt_ops, 2719 void *tmplt_priv, u32 chain_index, 2720 struct tcf_block *block, struct sk_buff *oskb, 2721 u32 seq, u16 flags, bool unicast) 2722 { 2723 u32 portid = oskb ? NETLINK_CB(oskb).portid : 0; 2724 struct net *net = block->net; 2725 struct sk_buff *skb; 2726 2727 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL); 2728 if (!skb) 2729 return -ENOBUFS; 2730 2731 if (tc_chain_fill_node(tmplt_ops, tmplt_priv, chain_index, net, skb, 2732 block, portid, seq, flags, RTM_DELCHAIN) <= 0) { 2733 kfree_skb(skb); 2734 return -EINVAL; 2735 } 2736 2737 if (unicast) 2738 return rtnl_unicast(skb, net, portid); 2739 2740 return rtnetlink_send(skb, net, portid, RTNLGRP_TC, flags & NLM_F_ECHO); 2741 } 2742 2743 static int tc_chain_tmplt_add(struct tcf_chain *chain, struct net *net, 2744 struct nlattr **tca, 2745 struct netlink_ext_ack *extack) 2746 { 2747 const struct tcf_proto_ops *ops; 2748 char name[IFNAMSIZ]; 2749 void *tmplt_priv; 2750 2751 /* If kind is not set, user did not specify template. */ 2752 if (!tca[TCA_KIND]) 2753 return 0; 2754 2755 if (tcf_proto_check_kind(tca[TCA_KIND], name)) { 2756 NL_SET_ERR_MSG(extack, "Specified TC chain template name too long"); 2757 return -EINVAL; 2758 } 2759 2760 ops = tcf_proto_lookup_ops(name, true, extack); 2761 if (IS_ERR(ops)) 2762 return PTR_ERR(ops); 2763 if (!ops->tmplt_create || !ops->tmplt_destroy || !ops->tmplt_dump) { 2764 NL_SET_ERR_MSG(extack, "Chain templates are not supported with specified classifier"); 2765 return -EOPNOTSUPP; 2766 } 2767 2768 tmplt_priv = ops->tmplt_create(net, chain, tca, extack); 2769 if (IS_ERR(tmplt_priv)) { 2770 module_put(ops->owner); 2771 return PTR_ERR(tmplt_priv); 2772 } 2773 chain->tmplt_ops = ops; 2774 chain->tmplt_priv = tmplt_priv; 2775 return 0; 2776 } 2777 2778 static void tc_chain_tmplt_del(const struct tcf_proto_ops *tmplt_ops, 2779 void *tmplt_priv) 2780 { 2781 /* If template ops are set, no work to do for us. */ 2782 if (!tmplt_ops) 2783 return; 2784 2785 tmplt_ops->tmplt_destroy(tmplt_priv); 2786 module_put(tmplt_ops->owner); 2787 } 2788 2789 /* Add/delete/get a chain */ 2790 2791 static int tc_ctl_chain(struct sk_buff *skb, struct nlmsghdr *n, 2792 struct netlink_ext_ack *extack) 2793 { 2794 struct net *net = sock_net(skb->sk); 2795 struct nlattr *tca[TCA_MAX + 1]; 2796 struct tcmsg *t; 2797 u32 parent; 2798 u32 chain_index; 2799 struct Qdisc *q = NULL; 2800 struct tcf_chain *chain = NULL; 2801 struct tcf_block *block; 2802 unsigned long cl; 2803 int err; 2804 2805 if (n->nlmsg_type != RTM_GETCHAIN && 2806 !netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN)) 2807 return -EPERM; 2808 2809 replay: 2810 err = nlmsg_parse_deprecated(n, sizeof(*t), tca, TCA_MAX, 2811 rtm_tca_policy, extack); 2812 if (err < 0) 2813 return err; 2814 2815 t = nlmsg_data(n); 2816 parent = t->tcm_parent; 2817 cl = 0; 2818 2819 block = tcf_block_find(net, &q, &parent, &cl, 2820 t->tcm_ifindex, t->tcm_block_index, extack); 2821 if (IS_ERR(block)) 2822 return PTR_ERR(block); 2823 2824 chain_index = tca[TCA_CHAIN] ? nla_get_u32(tca[TCA_CHAIN]) : 0; 2825 if (chain_index > TC_ACT_EXT_VAL_MASK) { 2826 NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit"); 2827 err = -EINVAL; 2828 goto errout_block; 2829 } 2830 2831 mutex_lock(&block->lock); 2832 chain = tcf_chain_lookup(block, chain_index); 2833 if (n->nlmsg_type == RTM_NEWCHAIN) { 2834 if (chain) { 2835 if (tcf_chain_held_by_acts_only(chain)) { 2836 /* The chain exists only because there is 2837 * some action referencing it. 2838 */ 2839 tcf_chain_hold(chain); 2840 } else { 2841 NL_SET_ERR_MSG(extack, "Filter chain already exists"); 2842 err = -EEXIST; 2843 goto errout_block_locked; 2844 } 2845 } else { 2846 if (!(n->nlmsg_flags & NLM_F_CREATE)) { 2847 NL_SET_ERR_MSG(extack, "Need both RTM_NEWCHAIN and NLM_F_CREATE to create a new chain"); 2848 err = -ENOENT; 2849 goto errout_block_locked; 2850 } 2851 chain = tcf_chain_create(block, chain_index); 2852 if (!chain) { 2853 NL_SET_ERR_MSG(extack, "Failed to create filter chain"); 2854 err = -ENOMEM; 2855 goto errout_block_locked; 2856 } 2857 } 2858 } else { 2859 if (!chain || tcf_chain_held_by_acts_only(chain)) { 2860 NL_SET_ERR_MSG(extack, "Cannot find specified filter chain"); 2861 err = -EINVAL; 2862 goto errout_block_locked; 2863 } 2864 tcf_chain_hold(chain); 2865 } 2866 2867 if (n->nlmsg_type == RTM_NEWCHAIN) { 2868 /* Modifying chain requires holding parent block lock. In case 2869 * the chain was successfully added, take a reference to the 2870 * chain. This ensures that an empty chain does not disappear at 2871 * the end of this function. 2872 */ 2873 tcf_chain_hold(chain); 2874 chain->explicitly_created = true; 2875 } 2876 mutex_unlock(&block->lock); 2877 2878 switch (n->nlmsg_type) { 2879 case RTM_NEWCHAIN: 2880 err = tc_chain_tmplt_add(chain, net, tca, extack); 2881 if (err) { 2882 tcf_chain_put_explicitly_created(chain); 2883 goto errout; 2884 } 2885 2886 tc_chain_notify(chain, NULL, 0, NLM_F_CREATE | NLM_F_EXCL, 2887 RTM_NEWCHAIN, false); 2888 break; 2889 case RTM_DELCHAIN: 2890 tfilter_notify_chain(net, skb, block, q, parent, n, 2891 chain, RTM_DELTFILTER); 2892 /* Flush the chain first as the user requested chain removal. */ 2893 tcf_chain_flush(chain, true); 2894 /* In case the chain was successfully deleted, put a reference 2895 * to the chain previously taken during addition. 2896 */ 2897 tcf_chain_put_explicitly_created(chain); 2898 break; 2899 case RTM_GETCHAIN: 2900 err = tc_chain_notify(chain, skb, n->nlmsg_seq, 2901 n->nlmsg_flags, n->nlmsg_type, true); 2902 if (err < 0) 2903 NL_SET_ERR_MSG(extack, "Failed to send chain notify message"); 2904 break; 2905 default: 2906 err = -EOPNOTSUPP; 2907 NL_SET_ERR_MSG(extack, "Unsupported message type"); 2908 goto errout; 2909 } 2910 2911 errout: 2912 tcf_chain_put(chain); 2913 errout_block: 2914 tcf_block_release(q, block, true); 2915 if (err == -EAGAIN) 2916 /* Replay the request. */ 2917 goto replay; 2918 return err; 2919 2920 errout_block_locked: 2921 mutex_unlock(&block->lock); 2922 goto errout_block; 2923 } 2924 2925 /* called with RTNL */ 2926 static int tc_dump_chain(struct sk_buff *skb, struct netlink_callback *cb) 2927 { 2928 struct net *net = sock_net(skb->sk); 2929 struct nlattr *tca[TCA_MAX + 1]; 2930 struct Qdisc *q = NULL; 2931 struct tcf_block *block; 2932 struct tcmsg *tcm = nlmsg_data(cb->nlh); 2933 struct tcf_chain *chain; 2934 long index_start; 2935 long index; 2936 int err; 2937 2938 if (nlmsg_len(cb->nlh) < sizeof(*tcm)) 2939 return skb->len; 2940 2941 err = nlmsg_parse_deprecated(cb->nlh, sizeof(*tcm), tca, TCA_MAX, 2942 rtm_tca_policy, cb->extack); 2943 if (err) 2944 return err; 2945 2946 if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK) { 2947 block = tcf_block_refcnt_get(net, tcm->tcm_block_index); 2948 if (!block) 2949 goto out; 2950 } else { 2951 const struct Qdisc_class_ops *cops; 2952 struct net_device *dev; 2953 unsigned long cl = 0; 2954 2955 dev = __dev_get_by_index(net, tcm->tcm_ifindex); 2956 if (!dev) 2957 return skb->len; 2958 2959 if (!tcm->tcm_parent) 2960 q = dev->qdisc; 2961 else 2962 q = qdisc_lookup(dev, TC_H_MAJ(tcm->tcm_parent)); 2963 2964 if (!q) 2965 goto out; 2966 cops = q->ops->cl_ops; 2967 if (!cops) 2968 goto out; 2969 if (!cops->tcf_block) 2970 goto out; 2971 if (TC_H_MIN(tcm->tcm_parent)) { 2972 cl = cops->find(q, tcm->tcm_parent); 2973 if (cl == 0) 2974 goto out; 2975 } 2976 block = cops->tcf_block(q, cl, NULL); 2977 if (!block) 2978 goto out; 2979 if (tcf_block_shared(block)) 2980 q = NULL; 2981 } 2982 2983 index_start = cb->args[0]; 2984 index = 0; 2985 2986 mutex_lock(&block->lock); 2987 list_for_each_entry(chain, &block->chain_list, list) { 2988 if ((tca[TCA_CHAIN] && 2989 nla_get_u32(tca[TCA_CHAIN]) != chain->index)) 2990 continue; 2991 if (index < index_start) { 2992 index++; 2993 continue; 2994 } 2995 if (tcf_chain_held_by_acts_only(chain)) 2996 continue; 2997 err = tc_chain_fill_node(chain->tmplt_ops, chain->tmplt_priv, 2998 chain->index, net, skb, block, 2999 NETLINK_CB(cb->skb).portid, 3000 cb->nlh->nlmsg_seq, NLM_F_MULTI, 3001 RTM_NEWCHAIN); 3002 if (err <= 0) 3003 break; 3004 index++; 3005 } 3006 mutex_unlock(&block->lock); 3007 3008 if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK) 3009 tcf_block_refcnt_put(block, true); 3010 cb->args[0] = index; 3011 3012 out: 3013 /* If we did no progress, the error (EMSGSIZE) is real */ 3014 if (skb->len == 0 && err) 3015 return err; 3016 return skb->len; 3017 } 3018 3019 void tcf_exts_destroy(struct tcf_exts *exts) 3020 { 3021 #ifdef CONFIG_NET_CLS_ACT 3022 if (exts->actions) { 3023 tcf_action_destroy(exts->actions, TCA_ACT_UNBIND); 3024 kfree(exts->actions); 3025 } 3026 exts->nr_actions = 0; 3027 #endif 3028 } 3029 EXPORT_SYMBOL(tcf_exts_destroy); 3030 3031 int tcf_exts_validate(struct net *net, struct tcf_proto *tp, struct nlattr **tb, 3032 struct nlattr *rate_tlv, struct tcf_exts *exts, 3033 u32 flags, struct netlink_ext_ack *extack) 3034 { 3035 #ifdef CONFIG_NET_CLS_ACT 3036 { 3037 int init_res[TCA_ACT_MAX_PRIO] = {}; 3038 struct tc_action *act; 3039 size_t attr_size = 0; 3040 3041 if (exts->police && tb[exts->police]) { 3042 struct tc_action_ops *a_o; 3043 3044 a_o = tc_action_load_ops(tb[exts->police], true, 3045 !(flags & TCA_ACT_FLAGS_NO_RTNL), 3046 extack); 3047 if (IS_ERR(a_o)) 3048 return PTR_ERR(a_o); 3049 flags |= TCA_ACT_FLAGS_POLICE | TCA_ACT_FLAGS_BIND; 3050 act = tcf_action_init_1(net, tp, tb[exts->police], 3051 rate_tlv, a_o, init_res, flags, 3052 extack); 3053 module_put(a_o->owner); 3054 if (IS_ERR(act)) 3055 return PTR_ERR(act); 3056 3057 act->type = exts->type = TCA_OLD_COMPAT; 3058 exts->actions[0] = act; 3059 exts->nr_actions = 1; 3060 tcf_idr_insert_many(exts->actions); 3061 } else if (exts->action && tb[exts->action]) { 3062 int err; 3063 3064 flags |= TCA_ACT_FLAGS_BIND; 3065 err = tcf_action_init(net, tp, tb[exts->action], 3066 rate_tlv, exts->actions, init_res, 3067 &attr_size, flags, extack); 3068 if (err < 0) 3069 return err; 3070 exts->nr_actions = err; 3071 } 3072 } 3073 #else 3074 if ((exts->action && tb[exts->action]) || 3075 (exts->police && tb[exts->police])) { 3076 NL_SET_ERR_MSG(extack, "Classifier actions are not supported per compile options (CONFIG_NET_CLS_ACT)"); 3077 return -EOPNOTSUPP; 3078 } 3079 #endif 3080 3081 return 0; 3082 } 3083 EXPORT_SYMBOL(tcf_exts_validate); 3084 3085 void tcf_exts_change(struct tcf_exts *dst, struct tcf_exts *src) 3086 { 3087 #ifdef CONFIG_NET_CLS_ACT 3088 struct tcf_exts old = *dst; 3089 3090 *dst = *src; 3091 tcf_exts_destroy(&old); 3092 #endif 3093 } 3094 EXPORT_SYMBOL(tcf_exts_change); 3095 3096 #ifdef CONFIG_NET_CLS_ACT 3097 static struct tc_action *tcf_exts_first_act(struct tcf_exts *exts) 3098 { 3099 if (exts->nr_actions == 0) 3100 return NULL; 3101 else 3102 return exts->actions[0]; 3103 } 3104 #endif 3105 3106 int tcf_exts_dump(struct sk_buff *skb, struct tcf_exts *exts) 3107 { 3108 #ifdef CONFIG_NET_CLS_ACT 3109 struct nlattr *nest; 3110 3111 if (exts->action && tcf_exts_has_actions(exts)) { 3112 /* 3113 * again for backward compatible mode - we want 3114 * to work with both old and new modes of entering 3115 * tc data even if iproute2 was newer - jhs 3116 */ 3117 if (exts->type != TCA_OLD_COMPAT) { 3118 nest = nla_nest_start_noflag(skb, exts->action); 3119 if (nest == NULL) 3120 goto nla_put_failure; 3121 3122 if (tcf_action_dump(skb, exts->actions, 0, 0, false) 3123 < 0) 3124 goto nla_put_failure; 3125 nla_nest_end(skb, nest); 3126 } else if (exts->police) { 3127 struct tc_action *act = tcf_exts_first_act(exts); 3128 nest = nla_nest_start_noflag(skb, exts->police); 3129 if (nest == NULL || !act) 3130 goto nla_put_failure; 3131 if (tcf_action_dump_old(skb, act, 0, 0) < 0) 3132 goto nla_put_failure; 3133 nla_nest_end(skb, nest); 3134 } 3135 } 3136 return 0; 3137 3138 nla_put_failure: 3139 nla_nest_cancel(skb, nest); 3140 return -1; 3141 #else 3142 return 0; 3143 #endif 3144 } 3145 EXPORT_SYMBOL(tcf_exts_dump); 3146 3147 int tcf_exts_terse_dump(struct sk_buff *skb, struct tcf_exts *exts) 3148 { 3149 #ifdef CONFIG_NET_CLS_ACT 3150 struct nlattr *nest; 3151 3152 if (!exts->action || !tcf_exts_has_actions(exts)) 3153 return 0; 3154 3155 nest = nla_nest_start_noflag(skb, exts->action); 3156 if (!nest) 3157 goto nla_put_failure; 3158 3159 if (tcf_action_dump(skb, exts->actions, 0, 0, true) < 0) 3160 goto nla_put_failure; 3161 nla_nest_end(skb, nest); 3162 return 0; 3163 3164 nla_put_failure: 3165 nla_nest_cancel(skb, nest); 3166 return -1; 3167 #else 3168 return 0; 3169 #endif 3170 } 3171 EXPORT_SYMBOL(tcf_exts_terse_dump); 3172 3173 int tcf_exts_dump_stats(struct sk_buff *skb, struct tcf_exts *exts) 3174 { 3175 #ifdef CONFIG_NET_CLS_ACT 3176 struct tc_action *a = tcf_exts_first_act(exts); 3177 if (a != NULL && tcf_action_copy_stats(skb, a, 1) < 0) 3178 return -1; 3179 #endif 3180 return 0; 3181 } 3182 EXPORT_SYMBOL(tcf_exts_dump_stats); 3183 3184 static void tcf_block_offload_inc(struct tcf_block *block, u32 *flags) 3185 { 3186 if (*flags & TCA_CLS_FLAGS_IN_HW) 3187 return; 3188 *flags |= TCA_CLS_FLAGS_IN_HW; 3189 atomic_inc(&block->offloadcnt); 3190 } 3191 3192 static void tcf_block_offload_dec(struct tcf_block *block, u32 *flags) 3193 { 3194 if (!(*flags & TCA_CLS_FLAGS_IN_HW)) 3195 return; 3196 *flags &= ~TCA_CLS_FLAGS_IN_HW; 3197 atomic_dec(&block->offloadcnt); 3198 } 3199 3200 static void tc_cls_offload_cnt_update(struct tcf_block *block, 3201 struct tcf_proto *tp, u32 *cnt, 3202 u32 *flags, u32 diff, bool add) 3203 { 3204 lockdep_assert_held(&block->cb_lock); 3205 3206 spin_lock(&tp->lock); 3207 if (add) { 3208 if (!*cnt) 3209 tcf_block_offload_inc(block, flags); 3210 *cnt += diff; 3211 } else { 3212 *cnt -= diff; 3213 if (!*cnt) 3214 tcf_block_offload_dec(block, flags); 3215 } 3216 spin_unlock(&tp->lock); 3217 } 3218 3219 static void 3220 tc_cls_offload_cnt_reset(struct tcf_block *block, struct tcf_proto *tp, 3221 u32 *cnt, u32 *flags) 3222 { 3223 lockdep_assert_held(&block->cb_lock); 3224 3225 spin_lock(&tp->lock); 3226 tcf_block_offload_dec(block, flags); 3227 *cnt = 0; 3228 spin_unlock(&tp->lock); 3229 } 3230 3231 static int 3232 __tc_setup_cb_call(struct tcf_block *block, enum tc_setup_type type, 3233 void *type_data, bool err_stop) 3234 { 3235 struct flow_block_cb *block_cb; 3236 int ok_count = 0; 3237 int err; 3238 3239 list_for_each_entry(block_cb, &block->flow_block.cb_list, list) { 3240 err = block_cb->cb(type, type_data, block_cb->cb_priv); 3241 if (err) { 3242 if (err_stop) 3243 return err; 3244 } else { 3245 ok_count++; 3246 } 3247 } 3248 return ok_count; 3249 } 3250 3251 int tc_setup_cb_call(struct tcf_block *block, enum tc_setup_type type, 3252 void *type_data, bool err_stop, bool rtnl_held) 3253 { 3254 bool take_rtnl = READ_ONCE(block->lockeddevcnt) && !rtnl_held; 3255 int ok_count; 3256 3257 retry: 3258 if (take_rtnl) 3259 rtnl_lock(); 3260 down_read(&block->cb_lock); 3261 /* Need to obtain rtnl lock if block is bound to devs that require it. 3262 * In block bind code cb_lock is obtained while holding rtnl, so we must 3263 * obtain the locks in same order here. 3264 */ 3265 if (!rtnl_held && !take_rtnl && block->lockeddevcnt) { 3266 up_read(&block->cb_lock); 3267 take_rtnl = true; 3268 goto retry; 3269 } 3270 3271 ok_count = __tc_setup_cb_call(block, type, type_data, err_stop); 3272 3273 up_read(&block->cb_lock); 3274 if (take_rtnl) 3275 rtnl_unlock(); 3276 return ok_count; 3277 } 3278 EXPORT_SYMBOL(tc_setup_cb_call); 3279 3280 /* Non-destructive filter add. If filter that wasn't already in hardware is 3281 * successfully offloaded, increment block offloads counter. On failure, 3282 * previously offloaded filter is considered to be intact and offloads counter 3283 * is not decremented. 3284 */ 3285 3286 int tc_setup_cb_add(struct tcf_block *block, struct tcf_proto *tp, 3287 enum tc_setup_type type, void *type_data, bool err_stop, 3288 u32 *flags, unsigned int *in_hw_count, bool rtnl_held) 3289 { 3290 bool take_rtnl = READ_ONCE(block->lockeddevcnt) && !rtnl_held; 3291 int ok_count; 3292 3293 retry: 3294 if (take_rtnl) 3295 rtnl_lock(); 3296 down_read(&block->cb_lock); 3297 /* Need to obtain rtnl lock if block is bound to devs that require it. 3298 * In block bind code cb_lock is obtained while holding rtnl, so we must 3299 * obtain the locks in same order here. 3300 */ 3301 if (!rtnl_held && !take_rtnl && block->lockeddevcnt) { 3302 up_read(&block->cb_lock); 3303 take_rtnl = true; 3304 goto retry; 3305 } 3306 3307 /* Make sure all netdevs sharing this block are offload-capable. */ 3308 if (block->nooffloaddevcnt && err_stop) { 3309 ok_count = -EOPNOTSUPP; 3310 goto err_unlock; 3311 } 3312 3313 ok_count = __tc_setup_cb_call(block, type, type_data, err_stop); 3314 if (ok_count < 0) 3315 goto err_unlock; 3316 3317 if (tp->ops->hw_add) 3318 tp->ops->hw_add(tp, type_data); 3319 if (ok_count > 0) 3320 tc_cls_offload_cnt_update(block, tp, in_hw_count, flags, 3321 ok_count, true); 3322 err_unlock: 3323 up_read(&block->cb_lock); 3324 if (take_rtnl) 3325 rtnl_unlock(); 3326 return ok_count < 0 ? ok_count : 0; 3327 } 3328 EXPORT_SYMBOL(tc_setup_cb_add); 3329 3330 /* Destructive filter replace. If filter that wasn't already in hardware is 3331 * successfully offloaded, increment block offload counter. On failure, 3332 * previously offloaded filter is considered to be destroyed and offload counter 3333 * is decremented. 3334 */ 3335 3336 int tc_setup_cb_replace(struct tcf_block *block, struct tcf_proto *tp, 3337 enum tc_setup_type type, void *type_data, bool err_stop, 3338 u32 *old_flags, unsigned int *old_in_hw_count, 3339 u32 *new_flags, unsigned int *new_in_hw_count, 3340 bool rtnl_held) 3341 { 3342 bool take_rtnl = READ_ONCE(block->lockeddevcnt) && !rtnl_held; 3343 int ok_count; 3344 3345 retry: 3346 if (take_rtnl) 3347 rtnl_lock(); 3348 down_read(&block->cb_lock); 3349 /* Need to obtain rtnl lock if block is bound to devs that require it. 3350 * In block bind code cb_lock is obtained while holding rtnl, so we must 3351 * obtain the locks in same order here. 3352 */ 3353 if (!rtnl_held && !take_rtnl && block->lockeddevcnt) { 3354 up_read(&block->cb_lock); 3355 take_rtnl = true; 3356 goto retry; 3357 } 3358 3359 /* Make sure all netdevs sharing this block are offload-capable. */ 3360 if (block->nooffloaddevcnt && err_stop) { 3361 ok_count = -EOPNOTSUPP; 3362 goto err_unlock; 3363 } 3364 3365 tc_cls_offload_cnt_reset(block, tp, old_in_hw_count, old_flags); 3366 if (tp->ops->hw_del) 3367 tp->ops->hw_del(tp, type_data); 3368 3369 ok_count = __tc_setup_cb_call(block, type, type_data, err_stop); 3370 if (ok_count < 0) 3371 goto err_unlock; 3372 3373 if (tp->ops->hw_add) 3374 tp->ops->hw_add(tp, type_data); 3375 if (ok_count > 0) 3376 tc_cls_offload_cnt_update(block, tp, new_in_hw_count, 3377 new_flags, ok_count, true); 3378 err_unlock: 3379 up_read(&block->cb_lock); 3380 if (take_rtnl) 3381 rtnl_unlock(); 3382 return ok_count < 0 ? ok_count : 0; 3383 } 3384 EXPORT_SYMBOL(tc_setup_cb_replace); 3385 3386 /* Destroy filter and decrement block offload counter, if filter was previously 3387 * offloaded. 3388 */ 3389 3390 int tc_setup_cb_destroy(struct tcf_block *block, struct tcf_proto *tp, 3391 enum tc_setup_type type, void *type_data, bool err_stop, 3392 u32 *flags, unsigned int *in_hw_count, bool rtnl_held) 3393 { 3394 bool take_rtnl = READ_ONCE(block->lockeddevcnt) && !rtnl_held; 3395 int ok_count; 3396 3397 retry: 3398 if (take_rtnl) 3399 rtnl_lock(); 3400 down_read(&block->cb_lock); 3401 /* Need to obtain rtnl lock if block is bound to devs that require it. 3402 * In block bind code cb_lock is obtained while holding rtnl, so we must 3403 * obtain the locks in same order here. 3404 */ 3405 if (!rtnl_held && !take_rtnl && block->lockeddevcnt) { 3406 up_read(&block->cb_lock); 3407 take_rtnl = true; 3408 goto retry; 3409 } 3410 3411 ok_count = __tc_setup_cb_call(block, type, type_data, err_stop); 3412 3413 tc_cls_offload_cnt_reset(block, tp, in_hw_count, flags); 3414 if (tp->ops->hw_del) 3415 tp->ops->hw_del(tp, type_data); 3416 3417 up_read(&block->cb_lock); 3418 if (take_rtnl) 3419 rtnl_unlock(); 3420 return ok_count < 0 ? ok_count : 0; 3421 } 3422 EXPORT_SYMBOL(tc_setup_cb_destroy); 3423 3424 int tc_setup_cb_reoffload(struct tcf_block *block, struct tcf_proto *tp, 3425 bool add, flow_setup_cb_t *cb, 3426 enum tc_setup_type type, void *type_data, 3427 void *cb_priv, u32 *flags, unsigned int *in_hw_count) 3428 { 3429 int err = cb(type, type_data, cb_priv); 3430 3431 if (err) { 3432 if (add && tc_skip_sw(*flags)) 3433 return err; 3434 } else { 3435 tc_cls_offload_cnt_update(block, tp, in_hw_count, flags, 1, 3436 add); 3437 } 3438 3439 return 0; 3440 } 3441 EXPORT_SYMBOL(tc_setup_cb_reoffload); 3442 3443 static int tcf_act_get_cookie(struct flow_action_entry *entry, 3444 const struct tc_action *act) 3445 { 3446 struct tc_cookie *cookie; 3447 int err = 0; 3448 3449 rcu_read_lock(); 3450 cookie = rcu_dereference(act->act_cookie); 3451 if (cookie) { 3452 entry->cookie = flow_action_cookie_create(cookie->data, 3453 cookie->len, 3454 GFP_ATOMIC); 3455 if (!entry->cookie) 3456 err = -ENOMEM; 3457 } 3458 rcu_read_unlock(); 3459 return err; 3460 } 3461 3462 static void tcf_act_put_cookie(struct flow_action_entry *entry) 3463 { 3464 flow_action_cookie_destroy(entry->cookie); 3465 } 3466 3467 void tc_cleanup_flow_action(struct flow_action *flow_action) 3468 { 3469 struct flow_action_entry *entry; 3470 int i; 3471 3472 flow_action_for_each(i, entry, flow_action) { 3473 tcf_act_put_cookie(entry); 3474 if (entry->destructor) 3475 entry->destructor(entry->destructor_priv); 3476 } 3477 } 3478 EXPORT_SYMBOL(tc_cleanup_flow_action); 3479 3480 static void tcf_mirred_get_dev(struct flow_action_entry *entry, 3481 const struct tc_action *act) 3482 { 3483 #ifdef CONFIG_NET_CLS_ACT 3484 entry->dev = act->ops->get_dev(act, &entry->destructor); 3485 if (!entry->dev) 3486 return; 3487 entry->destructor_priv = entry->dev; 3488 #endif 3489 } 3490 3491 static void tcf_tunnel_encap_put_tunnel(void *priv) 3492 { 3493 struct ip_tunnel_info *tunnel = priv; 3494 3495 kfree(tunnel); 3496 } 3497 3498 static int tcf_tunnel_encap_get_tunnel(struct flow_action_entry *entry, 3499 const struct tc_action *act) 3500 { 3501 entry->tunnel = tcf_tunnel_info_copy(act); 3502 if (!entry->tunnel) 3503 return -ENOMEM; 3504 entry->destructor = tcf_tunnel_encap_put_tunnel; 3505 entry->destructor_priv = entry->tunnel; 3506 return 0; 3507 } 3508 3509 static void tcf_sample_get_group(struct flow_action_entry *entry, 3510 const struct tc_action *act) 3511 { 3512 #ifdef CONFIG_NET_CLS_ACT 3513 entry->sample.psample_group = 3514 act->ops->get_psample_group(act, &entry->destructor); 3515 entry->destructor_priv = entry->sample.psample_group; 3516 #endif 3517 } 3518 3519 static void tcf_gate_entry_destructor(void *priv) 3520 { 3521 struct action_gate_entry *oe = priv; 3522 3523 kfree(oe); 3524 } 3525 3526 static int tcf_gate_get_entries(struct flow_action_entry *entry, 3527 const struct tc_action *act) 3528 { 3529 entry->gate.entries = tcf_gate_get_list(act); 3530 3531 if (!entry->gate.entries) 3532 return -EINVAL; 3533 3534 entry->destructor = tcf_gate_entry_destructor; 3535 entry->destructor_priv = entry->gate.entries; 3536 3537 return 0; 3538 } 3539 3540 static enum flow_action_hw_stats tc_act_hw_stats(u8 hw_stats) 3541 { 3542 if (WARN_ON_ONCE(hw_stats > TCA_ACT_HW_STATS_ANY)) 3543 return FLOW_ACTION_HW_STATS_DONT_CARE; 3544 else if (!hw_stats) 3545 return FLOW_ACTION_HW_STATS_DISABLED; 3546 3547 return hw_stats; 3548 } 3549 3550 int tc_setup_flow_action(struct flow_action *flow_action, 3551 const struct tcf_exts *exts) 3552 { 3553 struct tc_action *act; 3554 int i, j, k, err = 0; 3555 3556 BUILD_BUG_ON(TCA_ACT_HW_STATS_ANY != FLOW_ACTION_HW_STATS_ANY); 3557 BUILD_BUG_ON(TCA_ACT_HW_STATS_IMMEDIATE != FLOW_ACTION_HW_STATS_IMMEDIATE); 3558 BUILD_BUG_ON(TCA_ACT_HW_STATS_DELAYED != FLOW_ACTION_HW_STATS_DELAYED); 3559 3560 if (!exts) 3561 return 0; 3562 3563 j = 0; 3564 tcf_exts_for_each_action(i, act, exts) { 3565 struct flow_action_entry *entry; 3566 3567 entry = &flow_action->entries[j]; 3568 spin_lock_bh(&act->tcfa_lock); 3569 err = tcf_act_get_cookie(entry, act); 3570 if (err) 3571 goto err_out_locked; 3572 3573 entry->hw_stats = tc_act_hw_stats(act->hw_stats); 3574 3575 if (is_tcf_gact_ok(act)) { 3576 entry->id = FLOW_ACTION_ACCEPT; 3577 } else if (is_tcf_gact_shot(act)) { 3578 entry->id = FLOW_ACTION_DROP; 3579 } else if (is_tcf_gact_trap(act)) { 3580 entry->id = FLOW_ACTION_TRAP; 3581 } else if (is_tcf_gact_goto_chain(act)) { 3582 entry->id = FLOW_ACTION_GOTO; 3583 entry->chain_index = tcf_gact_goto_chain_index(act); 3584 } else if (is_tcf_mirred_egress_redirect(act)) { 3585 entry->id = FLOW_ACTION_REDIRECT; 3586 tcf_mirred_get_dev(entry, act); 3587 } else if (is_tcf_mirred_egress_mirror(act)) { 3588 entry->id = FLOW_ACTION_MIRRED; 3589 tcf_mirred_get_dev(entry, act); 3590 } else if (is_tcf_mirred_ingress_redirect(act)) { 3591 entry->id = FLOW_ACTION_REDIRECT_INGRESS; 3592 tcf_mirred_get_dev(entry, act); 3593 } else if (is_tcf_mirred_ingress_mirror(act)) { 3594 entry->id = FLOW_ACTION_MIRRED_INGRESS; 3595 tcf_mirred_get_dev(entry, act); 3596 } else if (is_tcf_vlan(act)) { 3597 switch (tcf_vlan_action(act)) { 3598 case TCA_VLAN_ACT_PUSH: 3599 entry->id = FLOW_ACTION_VLAN_PUSH; 3600 entry->vlan.vid = tcf_vlan_push_vid(act); 3601 entry->vlan.proto = tcf_vlan_push_proto(act); 3602 entry->vlan.prio = tcf_vlan_push_prio(act); 3603 break; 3604 case TCA_VLAN_ACT_POP: 3605 entry->id = FLOW_ACTION_VLAN_POP; 3606 break; 3607 case TCA_VLAN_ACT_MODIFY: 3608 entry->id = FLOW_ACTION_VLAN_MANGLE; 3609 entry->vlan.vid = tcf_vlan_push_vid(act); 3610 entry->vlan.proto = tcf_vlan_push_proto(act); 3611 entry->vlan.prio = tcf_vlan_push_prio(act); 3612 break; 3613 default: 3614 err = -EOPNOTSUPP; 3615 goto err_out_locked; 3616 } 3617 } else if (is_tcf_tunnel_set(act)) { 3618 entry->id = FLOW_ACTION_TUNNEL_ENCAP; 3619 err = tcf_tunnel_encap_get_tunnel(entry, act); 3620 if (err) 3621 goto err_out_locked; 3622 } else if (is_tcf_tunnel_release(act)) { 3623 entry->id = FLOW_ACTION_TUNNEL_DECAP; 3624 } else if (is_tcf_pedit(act)) { 3625 for (k = 0; k < tcf_pedit_nkeys(act); k++) { 3626 switch (tcf_pedit_cmd(act, k)) { 3627 case TCA_PEDIT_KEY_EX_CMD_SET: 3628 entry->id = FLOW_ACTION_MANGLE; 3629 break; 3630 case TCA_PEDIT_KEY_EX_CMD_ADD: 3631 entry->id = FLOW_ACTION_ADD; 3632 break; 3633 default: 3634 err = -EOPNOTSUPP; 3635 goto err_out_locked; 3636 } 3637 entry->mangle.htype = tcf_pedit_htype(act, k); 3638 entry->mangle.mask = tcf_pedit_mask(act, k); 3639 entry->mangle.val = tcf_pedit_val(act, k); 3640 entry->mangle.offset = tcf_pedit_offset(act, k); 3641 entry->hw_stats = tc_act_hw_stats(act->hw_stats); 3642 entry = &flow_action->entries[++j]; 3643 } 3644 } else if (is_tcf_csum(act)) { 3645 entry->id = FLOW_ACTION_CSUM; 3646 entry->csum_flags = tcf_csum_update_flags(act); 3647 } else if (is_tcf_skbedit_mark(act)) { 3648 entry->id = FLOW_ACTION_MARK; 3649 entry->mark = tcf_skbedit_mark(act); 3650 } else if (is_tcf_sample(act)) { 3651 entry->id = FLOW_ACTION_SAMPLE; 3652 entry->sample.trunc_size = tcf_sample_trunc_size(act); 3653 entry->sample.truncate = tcf_sample_truncate(act); 3654 entry->sample.rate = tcf_sample_rate(act); 3655 tcf_sample_get_group(entry, act); 3656 } else if (is_tcf_police(act)) { 3657 entry->id = FLOW_ACTION_POLICE; 3658 entry->police.burst = tcf_police_burst(act); 3659 entry->police.rate_bytes_ps = 3660 tcf_police_rate_bytes_ps(act); 3661 entry->police.burst_pkt = tcf_police_burst_pkt(act); 3662 entry->police.rate_pkt_ps = 3663 tcf_police_rate_pkt_ps(act); 3664 entry->police.mtu = tcf_police_tcfp_mtu(act); 3665 entry->police.index = act->tcfa_index; 3666 } else if (is_tcf_ct(act)) { 3667 entry->id = FLOW_ACTION_CT; 3668 entry->ct.action = tcf_ct_action(act); 3669 entry->ct.zone = tcf_ct_zone(act); 3670 entry->ct.flow_table = tcf_ct_ft(act); 3671 } else if (is_tcf_mpls(act)) { 3672 switch (tcf_mpls_action(act)) { 3673 case TCA_MPLS_ACT_PUSH: 3674 entry->id = FLOW_ACTION_MPLS_PUSH; 3675 entry->mpls_push.proto = tcf_mpls_proto(act); 3676 entry->mpls_push.label = tcf_mpls_label(act); 3677 entry->mpls_push.tc = tcf_mpls_tc(act); 3678 entry->mpls_push.bos = tcf_mpls_bos(act); 3679 entry->mpls_push.ttl = tcf_mpls_ttl(act); 3680 break; 3681 case TCA_MPLS_ACT_POP: 3682 entry->id = FLOW_ACTION_MPLS_POP; 3683 entry->mpls_pop.proto = tcf_mpls_proto(act); 3684 break; 3685 case TCA_MPLS_ACT_MODIFY: 3686 entry->id = FLOW_ACTION_MPLS_MANGLE; 3687 entry->mpls_mangle.label = tcf_mpls_label(act); 3688 entry->mpls_mangle.tc = tcf_mpls_tc(act); 3689 entry->mpls_mangle.bos = tcf_mpls_bos(act); 3690 entry->mpls_mangle.ttl = tcf_mpls_ttl(act); 3691 break; 3692 default: 3693 err = -EOPNOTSUPP; 3694 goto err_out_locked; 3695 } 3696 } else if (is_tcf_skbedit_ptype(act)) { 3697 entry->id = FLOW_ACTION_PTYPE; 3698 entry->ptype = tcf_skbedit_ptype(act); 3699 } else if (is_tcf_skbedit_priority(act)) { 3700 entry->id = FLOW_ACTION_PRIORITY; 3701 entry->priority = tcf_skbedit_priority(act); 3702 } else if (is_tcf_gate(act)) { 3703 entry->id = FLOW_ACTION_GATE; 3704 entry->gate.index = tcf_gate_index(act); 3705 entry->gate.prio = tcf_gate_prio(act); 3706 entry->gate.basetime = tcf_gate_basetime(act); 3707 entry->gate.cycletime = tcf_gate_cycletime(act); 3708 entry->gate.cycletimeext = tcf_gate_cycletimeext(act); 3709 entry->gate.num_entries = tcf_gate_num_entries(act); 3710 err = tcf_gate_get_entries(entry, act); 3711 if (err) 3712 goto err_out_locked; 3713 } else { 3714 err = -EOPNOTSUPP; 3715 goto err_out_locked; 3716 } 3717 spin_unlock_bh(&act->tcfa_lock); 3718 3719 if (!is_tcf_pedit(act)) 3720 j++; 3721 } 3722 3723 err_out: 3724 if (err) 3725 tc_cleanup_flow_action(flow_action); 3726 3727 return err; 3728 err_out_locked: 3729 spin_unlock_bh(&act->tcfa_lock); 3730 goto err_out; 3731 } 3732 EXPORT_SYMBOL(tc_setup_flow_action); 3733 3734 unsigned int tcf_exts_num_actions(struct tcf_exts *exts) 3735 { 3736 unsigned int num_acts = 0; 3737 struct tc_action *act; 3738 int i; 3739 3740 tcf_exts_for_each_action(i, act, exts) { 3741 if (is_tcf_pedit(act)) 3742 num_acts += tcf_pedit_nkeys(act); 3743 else 3744 num_acts++; 3745 } 3746 return num_acts; 3747 } 3748 EXPORT_SYMBOL(tcf_exts_num_actions); 3749 3750 #ifdef CONFIG_NET_CLS_ACT 3751 static int tcf_qevent_parse_block_index(struct nlattr *block_index_attr, 3752 u32 *p_block_index, 3753 struct netlink_ext_ack *extack) 3754 { 3755 *p_block_index = nla_get_u32(block_index_attr); 3756 if (!*p_block_index) { 3757 NL_SET_ERR_MSG(extack, "Block number may not be zero"); 3758 return -EINVAL; 3759 } 3760 3761 return 0; 3762 } 3763 3764 int tcf_qevent_init(struct tcf_qevent *qe, struct Qdisc *sch, 3765 enum flow_block_binder_type binder_type, 3766 struct nlattr *block_index_attr, 3767 struct netlink_ext_ack *extack) 3768 { 3769 u32 block_index; 3770 int err; 3771 3772 if (!block_index_attr) 3773 return 0; 3774 3775 err = tcf_qevent_parse_block_index(block_index_attr, &block_index, extack); 3776 if (err) 3777 return err; 3778 3779 if (!block_index) 3780 return 0; 3781 3782 qe->info.binder_type = binder_type; 3783 qe->info.chain_head_change = tcf_chain_head_change_dflt; 3784 qe->info.chain_head_change_priv = &qe->filter_chain; 3785 qe->info.block_index = block_index; 3786 3787 return tcf_block_get_ext(&qe->block, sch, &qe->info, extack); 3788 } 3789 EXPORT_SYMBOL(tcf_qevent_init); 3790 3791 void tcf_qevent_destroy(struct tcf_qevent *qe, struct Qdisc *sch) 3792 { 3793 if (qe->info.block_index) 3794 tcf_block_put_ext(qe->block, sch, &qe->info); 3795 } 3796 EXPORT_SYMBOL(tcf_qevent_destroy); 3797 3798 int tcf_qevent_validate_change(struct tcf_qevent *qe, struct nlattr *block_index_attr, 3799 struct netlink_ext_ack *extack) 3800 { 3801 u32 block_index; 3802 int err; 3803 3804 if (!block_index_attr) 3805 return 0; 3806 3807 err = tcf_qevent_parse_block_index(block_index_attr, &block_index, extack); 3808 if (err) 3809 return err; 3810 3811 /* Bounce newly-configured block or change in block. */ 3812 if (block_index != qe->info.block_index) { 3813 NL_SET_ERR_MSG(extack, "Change of blocks is not supported"); 3814 return -EINVAL; 3815 } 3816 3817 return 0; 3818 } 3819 EXPORT_SYMBOL(tcf_qevent_validate_change); 3820 3821 struct sk_buff *tcf_qevent_handle(struct tcf_qevent *qe, struct Qdisc *sch, struct sk_buff *skb, 3822 struct sk_buff **to_free, int *ret) 3823 { 3824 struct tcf_result cl_res; 3825 struct tcf_proto *fl; 3826 3827 if (!qe->info.block_index) 3828 return skb; 3829 3830 fl = rcu_dereference_bh(qe->filter_chain); 3831 3832 switch (tcf_classify(skb, NULL, fl, &cl_res, false)) { 3833 case TC_ACT_SHOT: 3834 qdisc_qstats_drop(sch); 3835 __qdisc_drop(skb, to_free); 3836 *ret = __NET_XMIT_BYPASS; 3837 return NULL; 3838 case TC_ACT_STOLEN: 3839 case TC_ACT_QUEUED: 3840 case TC_ACT_TRAP: 3841 __qdisc_drop(skb, to_free); 3842 *ret = __NET_XMIT_STOLEN; 3843 return NULL; 3844 case TC_ACT_REDIRECT: 3845 skb_do_redirect(skb); 3846 *ret = __NET_XMIT_STOLEN; 3847 return NULL; 3848 } 3849 3850 return skb; 3851 } 3852 EXPORT_SYMBOL(tcf_qevent_handle); 3853 3854 int tcf_qevent_dump(struct sk_buff *skb, int attr_name, struct tcf_qevent *qe) 3855 { 3856 if (!qe->info.block_index) 3857 return 0; 3858 return nla_put_u32(skb, attr_name, qe->info.block_index); 3859 } 3860 EXPORT_SYMBOL(tcf_qevent_dump); 3861 #endif 3862 3863 static __net_init int tcf_net_init(struct net *net) 3864 { 3865 struct tcf_net *tn = net_generic(net, tcf_net_id); 3866 3867 spin_lock_init(&tn->idr_lock); 3868 idr_init(&tn->idr); 3869 return 0; 3870 } 3871 3872 static void __net_exit tcf_net_exit(struct net *net) 3873 { 3874 struct tcf_net *tn = net_generic(net, tcf_net_id); 3875 3876 idr_destroy(&tn->idr); 3877 } 3878 3879 static struct pernet_operations tcf_net_ops = { 3880 .init = tcf_net_init, 3881 .exit = tcf_net_exit, 3882 .id = &tcf_net_id, 3883 .size = sizeof(struct tcf_net), 3884 }; 3885 3886 static int __init tc_filter_init(void) 3887 { 3888 int err; 3889 3890 tc_filter_wq = alloc_ordered_workqueue("tc_filter_workqueue", 0); 3891 if (!tc_filter_wq) 3892 return -ENOMEM; 3893 3894 err = register_pernet_subsys(&tcf_net_ops); 3895 if (err) 3896 goto err_register_pernet_subsys; 3897 3898 rtnl_register(PF_UNSPEC, RTM_NEWTFILTER, tc_new_tfilter, NULL, 3899 RTNL_FLAG_DOIT_UNLOCKED); 3900 rtnl_register(PF_UNSPEC, RTM_DELTFILTER, tc_del_tfilter, NULL, 3901 RTNL_FLAG_DOIT_UNLOCKED); 3902 rtnl_register(PF_UNSPEC, RTM_GETTFILTER, tc_get_tfilter, 3903 tc_dump_tfilter, RTNL_FLAG_DOIT_UNLOCKED); 3904 rtnl_register(PF_UNSPEC, RTM_NEWCHAIN, tc_ctl_chain, NULL, 0); 3905 rtnl_register(PF_UNSPEC, RTM_DELCHAIN, tc_ctl_chain, NULL, 0); 3906 rtnl_register(PF_UNSPEC, RTM_GETCHAIN, tc_ctl_chain, 3907 tc_dump_chain, 0); 3908 3909 return 0; 3910 3911 err_register_pernet_subsys: 3912 destroy_workqueue(tc_filter_wq); 3913 return err; 3914 } 3915 3916 subsys_initcall(tc_filter_init); 3917