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_strlcpy(name, kind, IFNAMSIZ) >= IFNAMSIZ; 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 INIT_LIST_HEAD(&bo->cb_list); 638 } 639 640 static void tcf_block_unbind(struct tcf_block *block, 641 struct flow_block_offload *bo); 642 643 static void tc_block_indr_cleanup(struct flow_block_cb *block_cb) 644 { 645 struct tcf_block *block = block_cb->indr.data; 646 struct net_device *dev = block_cb->indr.dev; 647 struct Qdisc *sch = block_cb->indr.sch; 648 struct netlink_ext_ack extack = {}; 649 struct flow_block_offload bo; 650 651 tcf_block_offload_init(&bo, dev, sch, FLOW_BLOCK_UNBIND, 652 block_cb->indr.binder_type, 653 &block->flow_block, tcf_block_shared(block), 654 &extack); 655 down_write(&block->cb_lock); 656 list_del(&block_cb->driver_list); 657 list_move(&block_cb->list, &bo.cb_list); 658 up_write(&block->cb_lock); 659 rtnl_lock(); 660 tcf_block_unbind(block, &bo); 661 rtnl_unlock(); 662 } 663 664 static bool tcf_block_offload_in_use(struct tcf_block *block) 665 { 666 return atomic_read(&block->offloadcnt); 667 } 668 669 static int tcf_block_offload_cmd(struct tcf_block *block, 670 struct net_device *dev, struct Qdisc *sch, 671 struct tcf_block_ext_info *ei, 672 enum flow_block_command command, 673 struct netlink_ext_ack *extack) 674 { 675 struct flow_block_offload bo = {}; 676 677 tcf_block_offload_init(&bo, dev, sch, command, ei->binder_type, 678 &block->flow_block, tcf_block_shared(block), 679 extack); 680 681 if (dev->netdev_ops->ndo_setup_tc) { 682 int err; 683 684 err = dev->netdev_ops->ndo_setup_tc(dev, TC_SETUP_BLOCK, &bo); 685 if (err < 0) { 686 if (err != -EOPNOTSUPP) 687 NL_SET_ERR_MSG(extack, "Driver ndo_setup_tc failed"); 688 return err; 689 } 690 691 return tcf_block_setup(block, &bo); 692 } 693 694 flow_indr_dev_setup_offload(dev, sch, TC_SETUP_BLOCK, block, &bo, 695 tc_block_indr_cleanup); 696 tcf_block_setup(block, &bo); 697 698 return -EOPNOTSUPP; 699 } 700 701 static int tcf_block_offload_bind(struct tcf_block *block, struct Qdisc *q, 702 struct tcf_block_ext_info *ei, 703 struct netlink_ext_ack *extack) 704 { 705 struct net_device *dev = q->dev_queue->dev; 706 int err; 707 708 down_write(&block->cb_lock); 709 710 /* If tc offload feature is disabled and the block we try to bind 711 * to already has some offloaded filters, forbid to bind. 712 */ 713 if (dev->netdev_ops->ndo_setup_tc && 714 !tc_can_offload(dev) && 715 tcf_block_offload_in_use(block)) { 716 NL_SET_ERR_MSG(extack, "Bind to offloaded block failed as dev has offload disabled"); 717 err = -EOPNOTSUPP; 718 goto err_unlock; 719 } 720 721 err = tcf_block_offload_cmd(block, dev, q, ei, FLOW_BLOCK_BIND, extack); 722 if (err == -EOPNOTSUPP) 723 goto no_offload_dev_inc; 724 if (err) 725 goto err_unlock; 726 727 up_write(&block->cb_lock); 728 return 0; 729 730 no_offload_dev_inc: 731 if (tcf_block_offload_in_use(block)) 732 goto err_unlock; 733 734 err = 0; 735 block->nooffloaddevcnt++; 736 err_unlock: 737 up_write(&block->cb_lock); 738 return err; 739 } 740 741 static void tcf_block_offload_unbind(struct tcf_block *block, struct Qdisc *q, 742 struct tcf_block_ext_info *ei) 743 { 744 struct net_device *dev = q->dev_queue->dev; 745 int err; 746 747 down_write(&block->cb_lock); 748 err = tcf_block_offload_cmd(block, dev, q, ei, FLOW_BLOCK_UNBIND, NULL); 749 if (err == -EOPNOTSUPP) 750 goto no_offload_dev_dec; 751 up_write(&block->cb_lock); 752 return; 753 754 no_offload_dev_dec: 755 WARN_ON(block->nooffloaddevcnt-- == 0); 756 up_write(&block->cb_lock); 757 } 758 759 static int 760 tcf_chain0_head_change_cb_add(struct tcf_block *block, 761 struct tcf_block_ext_info *ei, 762 struct netlink_ext_ack *extack) 763 { 764 struct tcf_filter_chain_list_item *item; 765 struct tcf_chain *chain0; 766 767 item = kmalloc(sizeof(*item), GFP_KERNEL); 768 if (!item) { 769 NL_SET_ERR_MSG(extack, "Memory allocation for head change callback item failed"); 770 return -ENOMEM; 771 } 772 item->chain_head_change = ei->chain_head_change; 773 item->chain_head_change_priv = ei->chain_head_change_priv; 774 775 mutex_lock(&block->lock); 776 chain0 = block->chain0.chain; 777 if (chain0) 778 tcf_chain_hold(chain0); 779 else 780 list_add(&item->list, &block->chain0.filter_chain_list); 781 mutex_unlock(&block->lock); 782 783 if (chain0) { 784 struct tcf_proto *tp_head; 785 786 mutex_lock(&chain0->filter_chain_lock); 787 788 tp_head = tcf_chain_dereference(chain0->filter_chain, chain0); 789 if (tp_head) 790 tcf_chain_head_change_item(item, tp_head); 791 792 mutex_lock(&block->lock); 793 list_add(&item->list, &block->chain0.filter_chain_list); 794 mutex_unlock(&block->lock); 795 796 mutex_unlock(&chain0->filter_chain_lock); 797 tcf_chain_put(chain0); 798 } 799 800 return 0; 801 } 802 803 static void 804 tcf_chain0_head_change_cb_del(struct tcf_block *block, 805 struct tcf_block_ext_info *ei) 806 { 807 struct tcf_filter_chain_list_item *item; 808 809 mutex_lock(&block->lock); 810 list_for_each_entry(item, &block->chain0.filter_chain_list, list) { 811 if ((!ei->chain_head_change && !ei->chain_head_change_priv) || 812 (item->chain_head_change == ei->chain_head_change && 813 item->chain_head_change_priv == ei->chain_head_change_priv)) { 814 if (block->chain0.chain) 815 tcf_chain_head_change_item(item, NULL); 816 list_del(&item->list); 817 mutex_unlock(&block->lock); 818 819 kfree(item); 820 return; 821 } 822 } 823 mutex_unlock(&block->lock); 824 WARN_ON(1); 825 } 826 827 struct tcf_net { 828 spinlock_t idr_lock; /* Protects idr */ 829 struct idr idr; 830 }; 831 832 static unsigned int tcf_net_id; 833 834 static int tcf_block_insert(struct tcf_block *block, struct net *net, 835 struct netlink_ext_ack *extack) 836 { 837 struct tcf_net *tn = net_generic(net, tcf_net_id); 838 int err; 839 840 idr_preload(GFP_KERNEL); 841 spin_lock(&tn->idr_lock); 842 err = idr_alloc_u32(&tn->idr, block, &block->index, block->index, 843 GFP_NOWAIT); 844 spin_unlock(&tn->idr_lock); 845 idr_preload_end(); 846 847 return err; 848 } 849 850 static void tcf_block_remove(struct tcf_block *block, struct net *net) 851 { 852 struct tcf_net *tn = net_generic(net, tcf_net_id); 853 854 spin_lock(&tn->idr_lock); 855 idr_remove(&tn->idr, block->index); 856 spin_unlock(&tn->idr_lock); 857 } 858 859 static struct tcf_block *tcf_block_create(struct net *net, struct Qdisc *q, 860 u32 block_index, 861 struct netlink_ext_ack *extack) 862 { 863 struct tcf_block *block; 864 865 block = kzalloc(sizeof(*block), GFP_KERNEL); 866 if (!block) { 867 NL_SET_ERR_MSG(extack, "Memory allocation for block failed"); 868 return ERR_PTR(-ENOMEM); 869 } 870 mutex_init(&block->lock); 871 mutex_init(&block->proto_destroy_lock); 872 init_rwsem(&block->cb_lock); 873 flow_block_init(&block->flow_block); 874 INIT_LIST_HEAD(&block->chain_list); 875 INIT_LIST_HEAD(&block->owner_list); 876 INIT_LIST_HEAD(&block->chain0.filter_chain_list); 877 878 refcount_set(&block->refcnt, 1); 879 block->net = net; 880 block->index = block_index; 881 882 /* Don't store q pointer for blocks which are shared */ 883 if (!tcf_block_shared(block)) 884 block->q = q; 885 return block; 886 } 887 888 static struct tcf_block *tcf_block_lookup(struct net *net, u32 block_index) 889 { 890 struct tcf_net *tn = net_generic(net, tcf_net_id); 891 892 return idr_find(&tn->idr, block_index); 893 } 894 895 static struct tcf_block *tcf_block_refcnt_get(struct net *net, u32 block_index) 896 { 897 struct tcf_block *block; 898 899 rcu_read_lock(); 900 block = tcf_block_lookup(net, block_index); 901 if (block && !refcount_inc_not_zero(&block->refcnt)) 902 block = NULL; 903 rcu_read_unlock(); 904 905 return block; 906 } 907 908 static struct tcf_chain * 909 __tcf_get_next_chain(struct tcf_block *block, struct tcf_chain *chain) 910 { 911 mutex_lock(&block->lock); 912 if (chain) 913 chain = list_is_last(&chain->list, &block->chain_list) ? 914 NULL : list_next_entry(chain, list); 915 else 916 chain = list_first_entry_or_null(&block->chain_list, 917 struct tcf_chain, list); 918 919 /* skip all action-only chains */ 920 while (chain && tcf_chain_held_by_acts_only(chain)) 921 chain = list_is_last(&chain->list, &block->chain_list) ? 922 NULL : list_next_entry(chain, list); 923 924 if (chain) 925 tcf_chain_hold(chain); 926 mutex_unlock(&block->lock); 927 928 return chain; 929 } 930 931 /* Function to be used by all clients that want to iterate over all chains on 932 * block. It properly obtains block->lock and takes reference to chain before 933 * returning it. Users of this function must be tolerant to concurrent chain 934 * insertion/deletion or ensure that no concurrent chain modification is 935 * possible. Note that all netlink dump callbacks cannot guarantee to provide 936 * consistent dump because rtnl lock is released each time skb is filled with 937 * data and sent to user-space. 938 */ 939 940 struct tcf_chain * 941 tcf_get_next_chain(struct tcf_block *block, struct tcf_chain *chain) 942 { 943 struct tcf_chain *chain_next = __tcf_get_next_chain(block, chain); 944 945 if (chain) 946 tcf_chain_put(chain); 947 948 return chain_next; 949 } 950 EXPORT_SYMBOL(tcf_get_next_chain); 951 952 static struct tcf_proto * 953 __tcf_get_next_proto(struct tcf_chain *chain, struct tcf_proto *tp) 954 { 955 u32 prio = 0; 956 957 ASSERT_RTNL(); 958 mutex_lock(&chain->filter_chain_lock); 959 960 if (!tp) { 961 tp = tcf_chain_dereference(chain->filter_chain, chain); 962 } else if (tcf_proto_is_deleting(tp)) { 963 /* 'deleting' flag is set and chain->filter_chain_lock was 964 * unlocked, which means next pointer could be invalid. Restart 965 * search. 966 */ 967 prio = tp->prio + 1; 968 tp = tcf_chain_dereference(chain->filter_chain, chain); 969 970 for (; tp; tp = tcf_chain_dereference(tp->next, chain)) 971 if (!tp->deleting && tp->prio >= prio) 972 break; 973 } else { 974 tp = tcf_chain_dereference(tp->next, chain); 975 } 976 977 if (tp) 978 tcf_proto_get(tp); 979 980 mutex_unlock(&chain->filter_chain_lock); 981 982 return tp; 983 } 984 985 /* Function to be used by all clients that want to iterate over all tp's on 986 * chain. Users of this function must be tolerant to concurrent tp 987 * insertion/deletion or ensure that no concurrent chain modification is 988 * possible. Note that all netlink dump callbacks cannot guarantee to provide 989 * consistent dump because rtnl lock is released each time skb is filled with 990 * data and sent to user-space. 991 */ 992 993 struct tcf_proto * 994 tcf_get_next_proto(struct tcf_chain *chain, struct tcf_proto *tp, 995 bool rtnl_held) 996 { 997 struct tcf_proto *tp_next = __tcf_get_next_proto(chain, tp); 998 999 if (tp) 1000 tcf_proto_put(tp, rtnl_held, 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 = 4; 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, const struct tcf_proto *tp, 1582 struct tcf_result *res, bool compat_mode) 1583 { 1584 u32 last_executed_chain = 0; 1585 1586 return __tcf_classify(skb, tp, tp, res, compat_mode, 1587 &last_executed_chain); 1588 } 1589 EXPORT_SYMBOL(tcf_classify); 1590 1591 int tcf_classify_ingress(struct sk_buff *skb, 1592 const struct tcf_block *ingress_block, 1593 const struct tcf_proto *tp, 1594 struct tcf_result *res, bool compat_mode) 1595 { 1596 #if !IS_ENABLED(CONFIG_NET_TC_SKB_EXT) 1597 u32 last_executed_chain = 0; 1598 1599 return __tcf_classify(skb, tp, tp, res, compat_mode, 1600 &last_executed_chain); 1601 #else 1602 u32 last_executed_chain = tp ? tp->chain->index : 0; 1603 const struct tcf_proto *orig_tp = tp; 1604 struct tc_skb_ext *ext; 1605 int ret; 1606 1607 ext = skb_ext_find(skb, TC_SKB_EXT); 1608 1609 if (ext && ext->chain) { 1610 struct tcf_chain *fchain; 1611 1612 fchain = tcf_chain_lookup_rcu(ingress_block, ext->chain); 1613 if (!fchain) 1614 return TC_ACT_SHOT; 1615 1616 /* Consume, so cloned/redirect skbs won't inherit ext */ 1617 skb_ext_del(skb, TC_SKB_EXT); 1618 1619 tp = rcu_dereference_bh(fchain->filter_chain); 1620 last_executed_chain = fchain->index; 1621 } 1622 1623 ret = __tcf_classify(skb, tp, orig_tp, res, compat_mode, 1624 &last_executed_chain); 1625 1626 /* If we missed on some chain */ 1627 if (ret == TC_ACT_UNSPEC && last_executed_chain) { 1628 ext = skb_ext_add(skb, TC_SKB_EXT); 1629 if (WARN_ON_ONCE(!ext)) 1630 return TC_ACT_SHOT; 1631 ext->chain = last_executed_chain; 1632 } 1633 1634 return ret; 1635 #endif 1636 } 1637 EXPORT_SYMBOL(tcf_classify_ingress); 1638 1639 struct tcf_chain_info { 1640 struct tcf_proto __rcu **pprev; 1641 struct tcf_proto __rcu *next; 1642 }; 1643 1644 static struct tcf_proto *tcf_chain_tp_prev(struct tcf_chain *chain, 1645 struct tcf_chain_info *chain_info) 1646 { 1647 return tcf_chain_dereference(*chain_info->pprev, chain); 1648 } 1649 1650 static int tcf_chain_tp_insert(struct tcf_chain *chain, 1651 struct tcf_chain_info *chain_info, 1652 struct tcf_proto *tp) 1653 { 1654 if (chain->flushing) 1655 return -EAGAIN; 1656 1657 if (*chain_info->pprev == chain->filter_chain) 1658 tcf_chain0_head_change(chain, tp); 1659 tcf_proto_get(tp); 1660 RCU_INIT_POINTER(tp->next, tcf_chain_tp_prev(chain, chain_info)); 1661 rcu_assign_pointer(*chain_info->pprev, tp); 1662 1663 return 0; 1664 } 1665 1666 static void tcf_chain_tp_remove(struct tcf_chain *chain, 1667 struct tcf_chain_info *chain_info, 1668 struct tcf_proto *tp) 1669 { 1670 struct tcf_proto *next = tcf_chain_dereference(chain_info->next, chain); 1671 1672 tcf_proto_mark_delete(tp); 1673 if (tp == chain->filter_chain) 1674 tcf_chain0_head_change(chain, next); 1675 RCU_INIT_POINTER(*chain_info->pprev, next); 1676 } 1677 1678 static struct tcf_proto *tcf_chain_tp_find(struct tcf_chain *chain, 1679 struct tcf_chain_info *chain_info, 1680 u32 protocol, u32 prio, 1681 bool prio_allocate); 1682 1683 /* Try to insert new proto. 1684 * If proto with specified priority already exists, free new proto 1685 * and return existing one. 1686 */ 1687 1688 static struct tcf_proto *tcf_chain_tp_insert_unique(struct tcf_chain *chain, 1689 struct tcf_proto *tp_new, 1690 u32 protocol, u32 prio, 1691 bool rtnl_held) 1692 { 1693 struct tcf_chain_info chain_info; 1694 struct tcf_proto *tp; 1695 int err = 0; 1696 1697 mutex_lock(&chain->filter_chain_lock); 1698 1699 if (tcf_proto_exists_destroying(chain, tp_new)) { 1700 mutex_unlock(&chain->filter_chain_lock); 1701 tcf_proto_destroy(tp_new, rtnl_held, false, NULL); 1702 return ERR_PTR(-EAGAIN); 1703 } 1704 1705 tp = tcf_chain_tp_find(chain, &chain_info, 1706 protocol, prio, false); 1707 if (!tp) 1708 err = tcf_chain_tp_insert(chain, &chain_info, tp_new); 1709 mutex_unlock(&chain->filter_chain_lock); 1710 1711 if (tp) { 1712 tcf_proto_destroy(tp_new, rtnl_held, false, NULL); 1713 tp_new = tp; 1714 } else if (err) { 1715 tcf_proto_destroy(tp_new, rtnl_held, false, NULL); 1716 tp_new = ERR_PTR(err); 1717 } 1718 1719 return tp_new; 1720 } 1721 1722 static void tcf_chain_tp_delete_empty(struct tcf_chain *chain, 1723 struct tcf_proto *tp, bool rtnl_held, 1724 struct netlink_ext_ack *extack) 1725 { 1726 struct tcf_chain_info chain_info; 1727 struct tcf_proto *tp_iter; 1728 struct tcf_proto **pprev; 1729 struct tcf_proto *next; 1730 1731 mutex_lock(&chain->filter_chain_lock); 1732 1733 /* Atomically find and remove tp from chain. */ 1734 for (pprev = &chain->filter_chain; 1735 (tp_iter = tcf_chain_dereference(*pprev, chain)); 1736 pprev = &tp_iter->next) { 1737 if (tp_iter == tp) { 1738 chain_info.pprev = pprev; 1739 chain_info.next = tp_iter->next; 1740 WARN_ON(tp_iter->deleting); 1741 break; 1742 } 1743 } 1744 /* Verify that tp still exists and no new filters were inserted 1745 * concurrently. 1746 * Mark tp for deletion if it is empty. 1747 */ 1748 if (!tp_iter || !tcf_proto_check_delete(tp)) { 1749 mutex_unlock(&chain->filter_chain_lock); 1750 return; 1751 } 1752 1753 tcf_proto_signal_destroying(chain, tp); 1754 next = tcf_chain_dereference(chain_info.next, chain); 1755 if (tp == chain->filter_chain) 1756 tcf_chain0_head_change(chain, next); 1757 RCU_INIT_POINTER(*chain_info.pprev, next); 1758 mutex_unlock(&chain->filter_chain_lock); 1759 1760 tcf_proto_put(tp, rtnl_held, extack); 1761 } 1762 1763 static struct tcf_proto *tcf_chain_tp_find(struct tcf_chain *chain, 1764 struct tcf_chain_info *chain_info, 1765 u32 protocol, u32 prio, 1766 bool prio_allocate) 1767 { 1768 struct tcf_proto **pprev; 1769 struct tcf_proto *tp; 1770 1771 /* Check the chain for existence of proto-tcf with this priority */ 1772 for (pprev = &chain->filter_chain; 1773 (tp = tcf_chain_dereference(*pprev, chain)); 1774 pprev = &tp->next) { 1775 if (tp->prio >= prio) { 1776 if (tp->prio == prio) { 1777 if (prio_allocate || 1778 (tp->protocol != protocol && protocol)) 1779 return ERR_PTR(-EINVAL); 1780 } else { 1781 tp = NULL; 1782 } 1783 break; 1784 } 1785 } 1786 chain_info->pprev = pprev; 1787 if (tp) { 1788 chain_info->next = tp->next; 1789 tcf_proto_get(tp); 1790 } else { 1791 chain_info->next = NULL; 1792 } 1793 return tp; 1794 } 1795 1796 static int tcf_fill_node(struct net *net, struct sk_buff *skb, 1797 struct tcf_proto *tp, struct tcf_block *block, 1798 struct Qdisc *q, u32 parent, void *fh, 1799 u32 portid, u32 seq, u16 flags, int event, 1800 bool terse_dump, bool rtnl_held) 1801 { 1802 struct tcmsg *tcm; 1803 struct nlmsghdr *nlh; 1804 unsigned char *b = skb_tail_pointer(skb); 1805 1806 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags); 1807 if (!nlh) 1808 goto out_nlmsg_trim; 1809 tcm = nlmsg_data(nlh); 1810 tcm->tcm_family = AF_UNSPEC; 1811 tcm->tcm__pad1 = 0; 1812 tcm->tcm__pad2 = 0; 1813 if (q) { 1814 tcm->tcm_ifindex = qdisc_dev(q)->ifindex; 1815 tcm->tcm_parent = parent; 1816 } else { 1817 tcm->tcm_ifindex = TCM_IFINDEX_MAGIC_BLOCK; 1818 tcm->tcm_block_index = block->index; 1819 } 1820 tcm->tcm_info = TC_H_MAKE(tp->prio, tp->protocol); 1821 if (nla_put_string(skb, TCA_KIND, tp->ops->kind)) 1822 goto nla_put_failure; 1823 if (nla_put_u32(skb, TCA_CHAIN, tp->chain->index)) 1824 goto nla_put_failure; 1825 if (!fh) { 1826 tcm->tcm_handle = 0; 1827 } else if (terse_dump) { 1828 if (tp->ops->terse_dump) { 1829 if (tp->ops->terse_dump(net, tp, fh, skb, tcm, 1830 rtnl_held) < 0) 1831 goto nla_put_failure; 1832 } else { 1833 goto cls_op_not_supp; 1834 } 1835 } else { 1836 if (tp->ops->dump && 1837 tp->ops->dump(net, tp, fh, skb, tcm, rtnl_held) < 0) 1838 goto nla_put_failure; 1839 } 1840 nlh->nlmsg_len = skb_tail_pointer(skb) - b; 1841 return skb->len; 1842 1843 out_nlmsg_trim: 1844 nla_put_failure: 1845 cls_op_not_supp: 1846 nlmsg_trim(skb, b); 1847 return -1; 1848 } 1849 1850 static int tfilter_notify(struct net *net, struct sk_buff *oskb, 1851 struct nlmsghdr *n, struct tcf_proto *tp, 1852 struct tcf_block *block, struct Qdisc *q, 1853 u32 parent, void *fh, int event, bool unicast, 1854 bool rtnl_held) 1855 { 1856 struct sk_buff *skb; 1857 u32 portid = oskb ? NETLINK_CB(oskb).portid : 0; 1858 int err = 0; 1859 1860 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL); 1861 if (!skb) 1862 return -ENOBUFS; 1863 1864 if (tcf_fill_node(net, skb, tp, block, q, parent, fh, portid, 1865 n->nlmsg_seq, n->nlmsg_flags, event, 1866 false, rtnl_held) <= 0) { 1867 kfree_skb(skb); 1868 return -EINVAL; 1869 } 1870 1871 if (unicast) 1872 err = netlink_unicast(net->rtnl, skb, portid, MSG_DONTWAIT); 1873 else 1874 err = rtnetlink_send(skb, net, portid, RTNLGRP_TC, 1875 n->nlmsg_flags & NLM_F_ECHO); 1876 1877 if (err > 0) 1878 err = 0; 1879 return err; 1880 } 1881 1882 static int tfilter_del_notify(struct net *net, struct sk_buff *oskb, 1883 struct nlmsghdr *n, struct tcf_proto *tp, 1884 struct tcf_block *block, struct Qdisc *q, 1885 u32 parent, void *fh, bool unicast, bool *last, 1886 bool rtnl_held, struct netlink_ext_ack *extack) 1887 { 1888 struct sk_buff *skb; 1889 u32 portid = oskb ? NETLINK_CB(oskb).portid : 0; 1890 int err; 1891 1892 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL); 1893 if (!skb) 1894 return -ENOBUFS; 1895 1896 if (tcf_fill_node(net, skb, tp, block, q, parent, fh, portid, 1897 n->nlmsg_seq, n->nlmsg_flags, RTM_DELTFILTER, 1898 false, rtnl_held) <= 0) { 1899 NL_SET_ERR_MSG(extack, "Failed to build del event notification"); 1900 kfree_skb(skb); 1901 return -EINVAL; 1902 } 1903 1904 err = tp->ops->delete(tp, fh, last, rtnl_held, extack); 1905 if (err) { 1906 kfree_skb(skb); 1907 return err; 1908 } 1909 1910 if (unicast) 1911 err = netlink_unicast(net->rtnl, skb, portid, MSG_DONTWAIT); 1912 else 1913 err = rtnetlink_send(skb, net, portid, RTNLGRP_TC, 1914 n->nlmsg_flags & NLM_F_ECHO); 1915 if (err < 0) 1916 NL_SET_ERR_MSG(extack, "Failed to send filter delete notification"); 1917 1918 if (err > 0) 1919 err = 0; 1920 return err; 1921 } 1922 1923 static void tfilter_notify_chain(struct net *net, struct sk_buff *oskb, 1924 struct tcf_block *block, struct Qdisc *q, 1925 u32 parent, struct nlmsghdr *n, 1926 struct tcf_chain *chain, int event, 1927 bool rtnl_held) 1928 { 1929 struct tcf_proto *tp; 1930 1931 for (tp = tcf_get_next_proto(chain, NULL, rtnl_held); 1932 tp; tp = tcf_get_next_proto(chain, tp, rtnl_held)) 1933 tfilter_notify(net, oskb, n, tp, block, 1934 q, parent, NULL, event, false, rtnl_held); 1935 } 1936 1937 static void tfilter_put(struct tcf_proto *tp, void *fh) 1938 { 1939 if (tp->ops->put && fh) 1940 tp->ops->put(tp, fh); 1941 } 1942 1943 static int tc_new_tfilter(struct sk_buff *skb, struct nlmsghdr *n, 1944 struct netlink_ext_ack *extack) 1945 { 1946 struct net *net = sock_net(skb->sk); 1947 struct nlattr *tca[TCA_MAX + 1]; 1948 char name[IFNAMSIZ]; 1949 struct tcmsg *t; 1950 u32 protocol; 1951 u32 prio; 1952 bool prio_allocate; 1953 u32 parent; 1954 u32 chain_index; 1955 struct Qdisc *q = NULL; 1956 struct tcf_chain_info chain_info; 1957 struct tcf_chain *chain = NULL; 1958 struct tcf_block *block; 1959 struct tcf_proto *tp; 1960 unsigned long cl; 1961 void *fh; 1962 int err; 1963 int tp_created; 1964 bool rtnl_held = false; 1965 1966 if (!netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN)) 1967 return -EPERM; 1968 1969 replay: 1970 tp_created = 0; 1971 1972 err = nlmsg_parse_deprecated(n, sizeof(*t), tca, TCA_MAX, 1973 rtm_tca_policy, extack); 1974 if (err < 0) 1975 return err; 1976 1977 t = nlmsg_data(n); 1978 protocol = TC_H_MIN(t->tcm_info); 1979 prio = TC_H_MAJ(t->tcm_info); 1980 prio_allocate = false; 1981 parent = t->tcm_parent; 1982 tp = NULL; 1983 cl = 0; 1984 block = NULL; 1985 1986 if (prio == 0) { 1987 /* If no priority is provided by the user, 1988 * we allocate one. 1989 */ 1990 if (n->nlmsg_flags & NLM_F_CREATE) { 1991 prio = TC_H_MAKE(0x80000000U, 0U); 1992 prio_allocate = true; 1993 } else { 1994 NL_SET_ERR_MSG(extack, "Invalid filter command with priority of zero"); 1995 return -ENOENT; 1996 } 1997 } 1998 1999 /* Find head of filter chain. */ 2000 2001 err = __tcf_qdisc_find(net, &q, &parent, t->tcm_ifindex, false, extack); 2002 if (err) 2003 return err; 2004 2005 if (tcf_proto_check_kind(tca[TCA_KIND], name)) { 2006 NL_SET_ERR_MSG(extack, "Specified TC filter name too long"); 2007 err = -EINVAL; 2008 goto errout; 2009 } 2010 2011 /* Take rtnl mutex if rtnl_held was set to true on previous iteration, 2012 * block is shared (no qdisc found), qdisc is not unlocked, classifier 2013 * type is not specified, classifier is not unlocked. 2014 */ 2015 if (rtnl_held || 2016 (q && !(q->ops->cl_ops->flags & QDISC_CLASS_OPS_DOIT_UNLOCKED)) || 2017 !tcf_proto_is_unlocked(name)) { 2018 rtnl_held = true; 2019 rtnl_lock(); 2020 } 2021 2022 err = __tcf_qdisc_cl_find(q, parent, &cl, t->tcm_ifindex, extack); 2023 if (err) 2024 goto errout; 2025 2026 block = __tcf_block_find(net, q, cl, t->tcm_ifindex, t->tcm_block_index, 2027 extack); 2028 if (IS_ERR(block)) { 2029 err = PTR_ERR(block); 2030 goto errout; 2031 } 2032 block->classid = parent; 2033 2034 chain_index = tca[TCA_CHAIN] ? nla_get_u32(tca[TCA_CHAIN]) : 0; 2035 if (chain_index > TC_ACT_EXT_VAL_MASK) { 2036 NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit"); 2037 err = -EINVAL; 2038 goto errout; 2039 } 2040 chain = tcf_chain_get(block, chain_index, true); 2041 if (!chain) { 2042 NL_SET_ERR_MSG(extack, "Cannot create specified filter chain"); 2043 err = -ENOMEM; 2044 goto errout; 2045 } 2046 2047 mutex_lock(&chain->filter_chain_lock); 2048 tp = tcf_chain_tp_find(chain, &chain_info, protocol, 2049 prio, prio_allocate); 2050 if (IS_ERR(tp)) { 2051 NL_SET_ERR_MSG(extack, "Filter with specified priority/protocol not found"); 2052 err = PTR_ERR(tp); 2053 goto errout_locked; 2054 } 2055 2056 if (tp == NULL) { 2057 struct tcf_proto *tp_new = NULL; 2058 2059 if (chain->flushing) { 2060 err = -EAGAIN; 2061 goto errout_locked; 2062 } 2063 2064 /* Proto-tcf does not exist, create new one */ 2065 2066 if (tca[TCA_KIND] == NULL || !protocol) { 2067 NL_SET_ERR_MSG(extack, "Filter kind and protocol must be specified"); 2068 err = -EINVAL; 2069 goto errout_locked; 2070 } 2071 2072 if (!(n->nlmsg_flags & NLM_F_CREATE)) { 2073 NL_SET_ERR_MSG(extack, "Need both RTM_NEWTFILTER and NLM_F_CREATE to create a new filter"); 2074 err = -ENOENT; 2075 goto errout_locked; 2076 } 2077 2078 if (prio_allocate) 2079 prio = tcf_auto_prio(tcf_chain_tp_prev(chain, 2080 &chain_info)); 2081 2082 mutex_unlock(&chain->filter_chain_lock); 2083 tp_new = tcf_proto_create(name, protocol, prio, chain, 2084 rtnl_held, extack); 2085 if (IS_ERR(tp_new)) { 2086 err = PTR_ERR(tp_new); 2087 goto errout_tp; 2088 } 2089 2090 tp_created = 1; 2091 tp = tcf_chain_tp_insert_unique(chain, tp_new, protocol, prio, 2092 rtnl_held); 2093 if (IS_ERR(tp)) { 2094 err = PTR_ERR(tp); 2095 goto errout_tp; 2096 } 2097 } else { 2098 mutex_unlock(&chain->filter_chain_lock); 2099 } 2100 2101 if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind)) { 2102 NL_SET_ERR_MSG(extack, "Specified filter kind does not match existing one"); 2103 err = -EINVAL; 2104 goto errout; 2105 } 2106 2107 fh = tp->ops->get(tp, t->tcm_handle); 2108 2109 if (!fh) { 2110 if (!(n->nlmsg_flags & NLM_F_CREATE)) { 2111 NL_SET_ERR_MSG(extack, "Need both RTM_NEWTFILTER and NLM_F_CREATE to create a new filter"); 2112 err = -ENOENT; 2113 goto errout; 2114 } 2115 } else if (n->nlmsg_flags & NLM_F_EXCL) { 2116 tfilter_put(tp, fh); 2117 NL_SET_ERR_MSG(extack, "Filter already exists"); 2118 err = -EEXIST; 2119 goto errout; 2120 } 2121 2122 if (chain->tmplt_ops && chain->tmplt_ops != tp->ops) { 2123 NL_SET_ERR_MSG(extack, "Chain template is set to a different filter kind"); 2124 err = -EINVAL; 2125 goto errout; 2126 } 2127 2128 err = tp->ops->change(net, skb, tp, cl, t->tcm_handle, tca, &fh, 2129 n->nlmsg_flags & NLM_F_CREATE ? TCA_ACT_NOREPLACE : TCA_ACT_REPLACE, 2130 rtnl_held, extack); 2131 if (err == 0) { 2132 tfilter_notify(net, skb, n, tp, block, q, parent, fh, 2133 RTM_NEWTFILTER, false, rtnl_held); 2134 tfilter_put(tp, fh); 2135 /* q pointer is NULL for shared blocks */ 2136 if (q) 2137 q->flags &= ~TCQ_F_CAN_BYPASS; 2138 } 2139 2140 errout: 2141 if (err && tp_created) 2142 tcf_chain_tp_delete_empty(chain, tp, rtnl_held, NULL); 2143 errout_tp: 2144 if (chain) { 2145 if (tp && !IS_ERR(tp)) 2146 tcf_proto_put(tp, rtnl_held, NULL); 2147 if (!tp_created) 2148 tcf_chain_put(chain); 2149 } 2150 tcf_block_release(q, block, rtnl_held); 2151 2152 if (rtnl_held) 2153 rtnl_unlock(); 2154 2155 if (err == -EAGAIN) { 2156 /* Take rtnl lock in case EAGAIN is caused by concurrent flush 2157 * of target chain. 2158 */ 2159 rtnl_held = true; 2160 /* Replay the request. */ 2161 goto replay; 2162 } 2163 return err; 2164 2165 errout_locked: 2166 mutex_unlock(&chain->filter_chain_lock); 2167 goto errout; 2168 } 2169 2170 static int tc_del_tfilter(struct sk_buff *skb, struct nlmsghdr *n, 2171 struct netlink_ext_ack *extack) 2172 { 2173 struct net *net = sock_net(skb->sk); 2174 struct nlattr *tca[TCA_MAX + 1]; 2175 char name[IFNAMSIZ]; 2176 struct tcmsg *t; 2177 u32 protocol; 2178 u32 prio; 2179 u32 parent; 2180 u32 chain_index; 2181 struct Qdisc *q = NULL; 2182 struct tcf_chain_info chain_info; 2183 struct tcf_chain *chain = NULL; 2184 struct tcf_block *block = NULL; 2185 struct tcf_proto *tp = NULL; 2186 unsigned long cl = 0; 2187 void *fh = NULL; 2188 int err; 2189 bool rtnl_held = false; 2190 2191 if (!netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN)) 2192 return -EPERM; 2193 2194 err = nlmsg_parse_deprecated(n, sizeof(*t), tca, TCA_MAX, 2195 rtm_tca_policy, extack); 2196 if (err < 0) 2197 return err; 2198 2199 t = nlmsg_data(n); 2200 protocol = TC_H_MIN(t->tcm_info); 2201 prio = TC_H_MAJ(t->tcm_info); 2202 parent = t->tcm_parent; 2203 2204 if (prio == 0 && (protocol || t->tcm_handle || tca[TCA_KIND])) { 2205 NL_SET_ERR_MSG(extack, "Cannot flush filters with protocol, handle or kind set"); 2206 return -ENOENT; 2207 } 2208 2209 /* Find head of filter chain. */ 2210 2211 err = __tcf_qdisc_find(net, &q, &parent, t->tcm_ifindex, false, extack); 2212 if (err) 2213 return err; 2214 2215 if (tcf_proto_check_kind(tca[TCA_KIND], name)) { 2216 NL_SET_ERR_MSG(extack, "Specified TC filter name too long"); 2217 err = -EINVAL; 2218 goto errout; 2219 } 2220 /* Take rtnl mutex if flushing whole chain, block is shared (no qdisc 2221 * found), qdisc is not unlocked, classifier type is not specified, 2222 * classifier is not unlocked. 2223 */ 2224 if (!prio || 2225 (q && !(q->ops->cl_ops->flags & QDISC_CLASS_OPS_DOIT_UNLOCKED)) || 2226 !tcf_proto_is_unlocked(name)) { 2227 rtnl_held = true; 2228 rtnl_lock(); 2229 } 2230 2231 err = __tcf_qdisc_cl_find(q, parent, &cl, t->tcm_ifindex, extack); 2232 if (err) 2233 goto errout; 2234 2235 block = __tcf_block_find(net, q, cl, t->tcm_ifindex, t->tcm_block_index, 2236 extack); 2237 if (IS_ERR(block)) { 2238 err = PTR_ERR(block); 2239 goto errout; 2240 } 2241 2242 chain_index = tca[TCA_CHAIN] ? nla_get_u32(tca[TCA_CHAIN]) : 0; 2243 if (chain_index > TC_ACT_EXT_VAL_MASK) { 2244 NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit"); 2245 err = -EINVAL; 2246 goto errout; 2247 } 2248 chain = tcf_chain_get(block, chain_index, false); 2249 if (!chain) { 2250 /* User requested flush on non-existent chain. Nothing to do, 2251 * so just return success. 2252 */ 2253 if (prio == 0) { 2254 err = 0; 2255 goto errout; 2256 } 2257 NL_SET_ERR_MSG(extack, "Cannot find specified filter chain"); 2258 err = -ENOENT; 2259 goto errout; 2260 } 2261 2262 if (prio == 0) { 2263 tfilter_notify_chain(net, skb, block, q, parent, n, 2264 chain, RTM_DELTFILTER, rtnl_held); 2265 tcf_chain_flush(chain, rtnl_held); 2266 err = 0; 2267 goto errout; 2268 } 2269 2270 mutex_lock(&chain->filter_chain_lock); 2271 tp = tcf_chain_tp_find(chain, &chain_info, protocol, 2272 prio, false); 2273 if (!tp || IS_ERR(tp)) { 2274 NL_SET_ERR_MSG(extack, "Filter with specified priority/protocol not found"); 2275 err = tp ? PTR_ERR(tp) : -ENOENT; 2276 goto errout_locked; 2277 } else if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind)) { 2278 NL_SET_ERR_MSG(extack, "Specified filter kind does not match existing one"); 2279 err = -EINVAL; 2280 goto errout_locked; 2281 } else if (t->tcm_handle == 0) { 2282 tcf_proto_signal_destroying(chain, tp); 2283 tcf_chain_tp_remove(chain, &chain_info, tp); 2284 mutex_unlock(&chain->filter_chain_lock); 2285 2286 tcf_proto_put(tp, rtnl_held, NULL); 2287 tfilter_notify(net, skb, n, tp, block, q, parent, fh, 2288 RTM_DELTFILTER, false, rtnl_held); 2289 err = 0; 2290 goto errout; 2291 } 2292 mutex_unlock(&chain->filter_chain_lock); 2293 2294 fh = tp->ops->get(tp, t->tcm_handle); 2295 2296 if (!fh) { 2297 NL_SET_ERR_MSG(extack, "Specified filter handle not found"); 2298 err = -ENOENT; 2299 } else { 2300 bool last; 2301 2302 err = tfilter_del_notify(net, skb, n, tp, block, 2303 q, parent, fh, false, &last, 2304 rtnl_held, extack); 2305 2306 if (err) 2307 goto errout; 2308 if (last) 2309 tcf_chain_tp_delete_empty(chain, tp, rtnl_held, extack); 2310 } 2311 2312 errout: 2313 if (chain) { 2314 if (tp && !IS_ERR(tp)) 2315 tcf_proto_put(tp, rtnl_held, NULL); 2316 tcf_chain_put(chain); 2317 } 2318 tcf_block_release(q, block, rtnl_held); 2319 2320 if (rtnl_held) 2321 rtnl_unlock(); 2322 2323 return err; 2324 2325 errout_locked: 2326 mutex_unlock(&chain->filter_chain_lock); 2327 goto errout; 2328 } 2329 2330 static int tc_get_tfilter(struct sk_buff *skb, struct nlmsghdr *n, 2331 struct netlink_ext_ack *extack) 2332 { 2333 struct net *net = sock_net(skb->sk); 2334 struct nlattr *tca[TCA_MAX + 1]; 2335 char name[IFNAMSIZ]; 2336 struct tcmsg *t; 2337 u32 protocol; 2338 u32 prio; 2339 u32 parent; 2340 u32 chain_index; 2341 struct Qdisc *q = NULL; 2342 struct tcf_chain_info chain_info; 2343 struct tcf_chain *chain = NULL; 2344 struct tcf_block *block = NULL; 2345 struct tcf_proto *tp = NULL; 2346 unsigned long cl = 0; 2347 void *fh = NULL; 2348 int err; 2349 bool rtnl_held = false; 2350 2351 err = nlmsg_parse_deprecated(n, sizeof(*t), tca, TCA_MAX, 2352 rtm_tca_policy, extack); 2353 if (err < 0) 2354 return err; 2355 2356 t = nlmsg_data(n); 2357 protocol = TC_H_MIN(t->tcm_info); 2358 prio = TC_H_MAJ(t->tcm_info); 2359 parent = t->tcm_parent; 2360 2361 if (prio == 0) { 2362 NL_SET_ERR_MSG(extack, "Invalid filter command with priority of zero"); 2363 return -ENOENT; 2364 } 2365 2366 /* Find head of filter chain. */ 2367 2368 err = __tcf_qdisc_find(net, &q, &parent, t->tcm_ifindex, false, extack); 2369 if (err) 2370 return err; 2371 2372 if (tcf_proto_check_kind(tca[TCA_KIND], name)) { 2373 NL_SET_ERR_MSG(extack, "Specified TC filter name too long"); 2374 err = -EINVAL; 2375 goto errout; 2376 } 2377 /* Take rtnl mutex if block is shared (no qdisc found), qdisc is not 2378 * unlocked, classifier type is not specified, classifier is not 2379 * unlocked. 2380 */ 2381 if ((q && !(q->ops->cl_ops->flags & QDISC_CLASS_OPS_DOIT_UNLOCKED)) || 2382 !tcf_proto_is_unlocked(name)) { 2383 rtnl_held = true; 2384 rtnl_lock(); 2385 } 2386 2387 err = __tcf_qdisc_cl_find(q, parent, &cl, t->tcm_ifindex, extack); 2388 if (err) 2389 goto errout; 2390 2391 block = __tcf_block_find(net, q, cl, t->tcm_ifindex, t->tcm_block_index, 2392 extack); 2393 if (IS_ERR(block)) { 2394 err = PTR_ERR(block); 2395 goto errout; 2396 } 2397 2398 chain_index = tca[TCA_CHAIN] ? nla_get_u32(tca[TCA_CHAIN]) : 0; 2399 if (chain_index > TC_ACT_EXT_VAL_MASK) { 2400 NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit"); 2401 err = -EINVAL; 2402 goto errout; 2403 } 2404 chain = tcf_chain_get(block, chain_index, false); 2405 if (!chain) { 2406 NL_SET_ERR_MSG(extack, "Cannot find specified filter chain"); 2407 err = -EINVAL; 2408 goto errout; 2409 } 2410 2411 mutex_lock(&chain->filter_chain_lock); 2412 tp = tcf_chain_tp_find(chain, &chain_info, protocol, 2413 prio, false); 2414 mutex_unlock(&chain->filter_chain_lock); 2415 if (!tp || IS_ERR(tp)) { 2416 NL_SET_ERR_MSG(extack, "Filter with specified priority/protocol not found"); 2417 err = tp ? PTR_ERR(tp) : -ENOENT; 2418 goto errout; 2419 } else if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind)) { 2420 NL_SET_ERR_MSG(extack, "Specified filter kind does not match existing one"); 2421 err = -EINVAL; 2422 goto errout; 2423 } 2424 2425 fh = tp->ops->get(tp, t->tcm_handle); 2426 2427 if (!fh) { 2428 NL_SET_ERR_MSG(extack, "Specified filter handle not found"); 2429 err = -ENOENT; 2430 } else { 2431 err = tfilter_notify(net, skb, n, tp, block, q, parent, 2432 fh, RTM_NEWTFILTER, true, rtnl_held); 2433 if (err < 0) 2434 NL_SET_ERR_MSG(extack, "Failed to send filter notify message"); 2435 } 2436 2437 tfilter_put(tp, fh); 2438 errout: 2439 if (chain) { 2440 if (tp && !IS_ERR(tp)) 2441 tcf_proto_put(tp, rtnl_held, NULL); 2442 tcf_chain_put(chain); 2443 } 2444 tcf_block_release(q, block, rtnl_held); 2445 2446 if (rtnl_held) 2447 rtnl_unlock(); 2448 2449 return err; 2450 } 2451 2452 struct tcf_dump_args { 2453 struct tcf_walker w; 2454 struct sk_buff *skb; 2455 struct netlink_callback *cb; 2456 struct tcf_block *block; 2457 struct Qdisc *q; 2458 u32 parent; 2459 bool terse_dump; 2460 }; 2461 2462 static int tcf_node_dump(struct tcf_proto *tp, void *n, struct tcf_walker *arg) 2463 { 2464 struct tcf_dump_args *a = (void *)arg; 2465 struct net *net = sock_net(a->skb->sk); 2466 2467 return tcf_fill_node(net, a->skb, tp, a->block, a->q, a->parent, 2468 n, NETLINK_CB(a->cb->skb).portid, 2469 a->cb->nlh->nlmsg_seq, NLM_F_MULTI, 2470 RTM_NEWTFILTER, a->terse_dump, true); 2471 } 2472 2473 static bool tcf_chain_dump(struct tcf_chain *chain, struct Qdisc *q, u32 parent, 2474 struct sk_buff *skb, struct netlink_callback *cb, 2475 long index_start, long *p_index, bool terse) 2476 { 2477 struct net *net = sock_net(skb->sk); 2478 struct tcf_block *block = chain->block; 2479 struct tcmsg *tcm = nlmsg_data(cb->nlh); 2480 struct tcf_proto *tp, *tp_prev; 2481 struct tcf_dump_args arg; 2482 2483 for (tp = __tcf_get_next_proto(chain, NULL); 2484 tp; 2485 tp_prev = tp, 2486 tp = __tcf_get_next_proto(chain, tp), 2487 tcf_proto_put(tp_prev, true, NULL), 2488 (*p_index)++) { 2489 if (*p_index < index_start) 2490 continue; 2491 if (TC_H_MAJ(tcm->tcm_info) && 2492 TC_H_MAJ(tcm->tcm_info) != tp->prio) 2493 continue; 2494 if (TC_H_MIN(tcm->tcm_info) && 2495 TC_H_MIN(tcm->tcm_info) != tp->protocol) 2496 continue; 2497 if (*p_index > index_start) 2498 memset(&cb->args[1], 0, 2499 sizeof(cb->args) - sizeof(cb->args[0])); 2500 if (cb->args[1] == 0) { 2501 if (tcf_fill_node(net, skb, tp, block, q, parent, NULL, 2502 NETLINK_CB(cb->skb).portid, 2503 cb->nlh->nlmsg_seq, NLM_F_MULTI, 2504 RTM_NEWTFILTER, false, true) <= 0) 2505 goto errout; 2506 cb->args[1] = 1; 2507 } 2508 if (!tp->ops->walk) 2509 continue; 2510 arg.w.fn = tcf_node_dump; 2511 arg.skb = skb; 2512 arg.cb = cb; 2513 arg.block = block; 2514 arg.q = q; 2515 arg.parent = parent; 2516 arg.w.stop = 0; 2517 arg.w.skip = cb->args[1] - 1; 2518 arg.w.count = 0; 2519 arg.w.cookie = cb->args[2]; 2520 arg.terse_dump = terse; 2521 tp->ops->walk(tp, &arg.w, true); 2522 cb->args[2] = arg.w.cookie; 2523 cb->args[1] = arg.w.count + 1; 2524 if (arg.w.stop) 2525 goto errout; 2526 } 2527 return true; 2528 2529 errout: 2530 tcf_proto_put(tp, true, NULL); 2531 return false; 2532 } 2533 2534 static const struct nla_policy tcf_tfilter_dump_policy[TCA_MAX + 1] = { 2535 [TCA_DUMP_FLAGS] = NLA_POLICY_BITFIELD32(TCA_DUMP_FLAGS_TERSE), 2536 }; 2537 2538 /* called with RTNL */ 2539 static int tc_dump_tfilter(struct sk_buff *skb, struct netlink_callback *cb) 2540 { 2541 struct tcf_chain *chain, *chain_prev; 2542 struct net *net = sock_net(skb->sk); 2543 struct nlattr *tca[TCA_MAX + 1]; 2544 struct Qdisc *q = NULL; 2545 struct tcf_block *block; 2546 struct tcmsg *tcm = nlmsg_data(cb->nlh); 2547 bool terse_dump = false; 2548 long index_start; 2549 long index; 2550 u32 parent; 2551 int err; 2552 2553 if (nlmsg_len(cb->nlh) < sizeof(*tcm)) 2554 return skb->len; 2555 2556 err = nlmsg_parse_deprecated(cb->nlh, sizeof(*tcm), tca, TCA_MAX, 2557 tcf_tfilter_dump_policy, cb->extack); 2558 if (err) 2559 return err; 2560 2561 if (tca[TCA_DUMP_FLAGS]) { 2562 struct nla_bitfield32 flags = 2563 nla_get_bitfield32(tca[TCA_DUMP_FLAGS]); 2564 2565 terse_dump = flags.value & TCA_DUMP_FLAGS_TERSE; 2566 } 2567 2568 if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK) { 2569 block = tcf_block_refcnt_get(net, tcm->tcm_block_index); 2570 if (!block) 2571 goto out; 2572 /* If we work with block index, q is NULL and parent value 2573 * will never be used in the following code. The check 2574 * in tcf_fill_node prevents it. However, compiler does not 2575 * see that far, so set parent to zero to silence the warning 2576 * about parent being uninitialized. 2577 */ 2578 parent = 0; 2579 } else { 2580 const struct Qdisc_class_ops *cops; 2581 struct net_device *dev; 2582 unsigned long cl = 0; 2583 2584 dev = __dev_get_by_index(net, tcm->tcm_ifindex); 2585 if (!dev) 2586 return skb->len; 2587 2588 parent = tcm->tcm_parent; 2589 if (!parent) 2590 q = dev->qdisc; 2591 else 2592 q = qdisc_lookup(dev, TC_H_MAJ(tcm->tcm_parent)); 2593 if (!q) 2594 goto out; 2595 cops = q->ops->cl_ops; 2596 if (!cops) 2597 goto out; 2598 if (!cops->tcf_block) 2599 goto out; 2600 if (TC_H_MIN(tcm->tcm_parent)) { 2601 cl = cops->find(q, tcm->tcm_parent); 2602 if (cl == 0) 2603 goto out; 2604 } 2605 block = cops->tcf_block(q, cl, NULL); 2606 if (!block) 2607 goto out; 2608 parent = block->classid; 2609 if (tcf_block_shared(block)) 2610 q = NULL; 2611 } 2612 2613 index_start = cb->args[0]; 2614 index = 0; 2615 2616 for (chain = __tcf_get_next_chain(block, NULL); 2617 chain; 2618 chain_prev = chain, 2619 chain = __tcf_get_next_chain(block, chain), 2620 tcf_chain_put(chain_prev)) { 2621 if (tca[TCA_CHAIN] && 2622 nla_get_u32(tca[TCA_CHAIN]) != chain->index) 2623 continue; 2624 if (!tcf_chain_dump(chain, q, parent, skb, cb, 2625 index_start, &index, terse_dump)) { 2626 tcf_chain_put(chain); 2627 err = -EMSGSIZE; 2628 break; 2629 } 2630 } 2631 2632 if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK) 2633 tcf_block_refcnt_put(block, true); 2634 cb->args[0] = index; 2635 2636 out: 2637 /* If we did no progress, the error (EMSGSIZE) is real */ 2638 if (skb->len == 0 && err) 2639 return err; 2640 return skb->len; 2641 } 2642 2643 static int tc_chain_fill_node(const struct tcf_proto_ops *tmplt_ops, 2644 void *tmplt_priv, u32 chain_index, 2645 struct net *net, struct sk_buff *skb, 2646 struct tcf_block *block, 2647 u32 portid, u32 seq, u16 flags, int event) 2648 { 2649 unsigned char *b = skb_tail_pointer(skb); 2650 const struct tcf_proto_ops *ops; 2651 struct nlmsghdr *nlh; 2652 struct tcmsg *tcm; 2653 void *priv; 2654 2655 ops = tmplt_ops; 2656 priv = tmplt_priv; 2657 2658 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags); 2659 if (!nlh) 2660 goto out_nlmsg_trim; 2661 tcm = nlmsg_data(nlh); 2662 tcm->tcm_family = AF_UNSPEC; 2663 tcm->tcm__pad1 = 0; 2664 tcm->tcm__pad2 = 0; 2665 tcm->tcm_handle = 0; 2666 if (block->q) { 2667 tcm->tcm_ifindex = qdisc_dev(block->q)->ifindex; 2668 tcm->tcm_parent = block->q->handle; 2669 } else { 2670 tcm->tcm_ifindex = TCM_IFINDEX_MAGIC_BLOCK; 2671 tcm->tcm_block_index = block->index; 2672 } 2673 2674 if (nla_put_u32(skb, TCA_CHAIN, chain_index)) 2675 goto nla_put_failure; 2676 2677 if (ops) { 2678 if (nla_put_string(skb, TCA_KIND, ops->kind)) 2679 goto nla_put_failure; 2680 if (ops->tmplt_dump(skb, net, priv) < 0) 2681 goto nla_put_failure; 2682 } 2683 2684 nlh->nlmsg_len = skb_tail_pointer(skb) - b; 2685 return skb->len; 2686 2687 out_nlmsg_trim: 2688 nla_put_failure: 2689 nlmsg_trim(skb, b); 2690 return -EMSGSIZE; 2691 } 2692 2693 static int tc_chain_notify(struct tcf_chain *chain, struct sk_buff *oskb, 2694 u32 seq, u16 flags, int event, bool unicast) 2695 { 2696 u32 portid = oskb ? NETLINK_CB(oskb).portid : 0; 2697 struct tcf_block *block = chain->block; 2698 struct net *net = block->net; 2699 struct sk_buff *skb; 2700 int err = 0; 2701 2702 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL); 2703 if (!skb) 2704 return -ENOBUFS; 2705 2706 if (tc_chain_fill_node(chain->tmplt_ops, chain->tmplt_priv, 2707 chain->index, net, skb, block, portid, 2708 seq, flags, event) <= 0) { 2709 kfree_skb(skb); 2710 return -EINVAL; 2711 } 2712 2713 if (unicast) 2714 err = netlink_unicast(net->rtnl, skb, portid, MSG_DONTWAIT); 2715 else 2716 err = rtnetlink_send(skb, net, portid, RTNLGRP_TC, 2717 flags & NLM_F_ECHO); 2718 2719 if (err > 0) 2720 err = 0; 2721 return err; 2722 } 2723 2724 static int tc_chain_notify_delete(const struct tcf_proto_ops *tmplt_ops, 2725 void *tmplt_priv, u32 chain_index, 2726 struct tcf_block *block, struct sk_buff *oskb, 2727 u32 seq, u16 flags, bool unicast) 2728 { 2729 u32 portid = oskb ? NETLINK_CB(oskb).portid : 0; 2730 struct net *net = block->net; 2731 struct sk_buff *skb; 2732 2733 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL); 2734 if (!skb) 2735 return -ENOBUFS; 2736 2737 if (tc_chain_fill_node(tmplt_ops, tmplt_priv, chain_index, net, skb, 2738 block, portid, seq, flags, RTM_DELCHAIN) <= 0) { 2739 kfree_skb(skb); 2740 return -EINVAL; 2741 } 2742 2743 if (unicast) 2744 return netlink_unicast(net->rtnl, skb, portid, MSG_DONTWAIT); 2745 2746 return rtnetlink_send(skb, net, portid, RTNLGRP_TC, flags & NLM_F_ECHO); 2747 } 2748 2749 static int tc_chain_tmplt_add(struct tcf_chain *chain, struct net *net, 2750 struct nlattr **tca, 2751 struct netlink_ext_ack *extack) 2752 { 2753 const struct tcf_proto_ops *ops; 2754 char name[IFNAMSIZ]; 2755 void *tmplt_priv; 2756 2757 /* If kind is not set, user did not specify template. */ 2758 if (!tca[TCA_KIND]) 2759 return 0; 2760 2761 if (tcf_proto_check_kind(tca[TCA_KIND], name)) { 2762 NL_SET_ERR_MSG(extack, "Specified TC chain template name too long"); 2763 return -EINVAL; 2764 } 2765 2766 ops = tcf_proto_lookup_ops(name, true, extack); 2767 if (IS_ERR(ops)) 2768 return PTR_ERR(ops); 2769 if (!ops->tmplt_create || !ops->tmplt_destroy || !ops->tmplt_dump) { 2770 NL_SET_ERR_MSG(extack, "Chain templates are not supported with specified classifier"); 2771 return -EOPNOTSUPP; 2772 } 2773 2774 tmplt_priv = ops->tmplt_create(net, chain, tca, extack); 2775 if (IS_ERR(tmplt_priv)) { 2776 module_put(ops->owner); 2777 return PTR_ERR(tmplt_priv); 2778 } 2779 chain->tmplt_ops = ops; 2780 chain->tmplt_priv = tmplt_priv; 2781 return 0; 2782 } 2783 2784 static void tc_chain_tmplt_del(const struct tcf_proto_ops *tmplt_ops, 2785 void *tmplt_priv) 2786 { 2787 /* If template ops are set, no work to do for us. */ 2788 if (!tmplt_ops) 2789 return; 2790 2791 tmplt_ops->tmplt_destroy(tmplt_priv); 2792 module_put(tmplt_ops->owner); 2793 } 2794 2795 /* Add/delete/get a chain */ 2796 2797 static int tc_ctl_chain(struct sk_buff *skb, struct nlmsghdr *n, 2798 struct netlink_ext_ack *extack) 2799 { 2800 struct net *net = sock_net(skb->sk); 2801 struct nlattr *tca[TCA_MAX + 1]; 2802 struct tcmsg *t; 2803 u32 parent; 2804 u32 chain_index; 2805 struct Qdisc *q = NULL; 2806 struct tcf_chain *chain = NULL; 2807 struct tcf_block *block; 2808 unsigned long cl; 2809 int err; 2810 2811 if (n->nlmsg_type != RTM_GETCHAIN && 2812 !netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN)) 2813 return -EPERM; 2814 2815 replay: 2816 err = nlmsg_parse_deprecated(n, sizeof(*t), tca, TCA_MAX, 2817 rtm_tca_policy, extack); 2818 if (err < 0) 2819 return err; 2820 2821 t = nlmsg_data(n); 2822 parent = t->tcm_parent; 2823 cl = 0; 2824 2825 block = tcf_block_find(net, &q, &parent, &cl, 2826 t->tcm_ifindex, t->tcm_block_index, extack); 2827 if (IS_ERR(block)) 2828 return PTR_ERR(block); 2829 2830 chain_index = tca[TCA_CHAIN] ? nla_get_u32(tca[TCA_CHAIN]) : 0; 2831 if (chain_index > TC_ACT_EXT_VAL_MASK) { 2832 NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit"); 2833 err = -EINVAL; 2834 goto errout_block; 2835 } 2836 2837 mutex_lock(&block->lock); 2838 chain = tcf_chain_lookup(block, chain_index); 2839 if (n->nlmsg_type == RTM_NEWCHAIN) { 2840 if (chain) { 2841 if (tcf_chain_held_by_acts_only(chain)) { 2842 /* The chain exists only because there is 2843 * some action referencing it. 2844 */ 2845 tcf_chain_hold(chain); 2846 } else { 2847 NL_SET_ERR_MSG(extack, "Filter chain already exists"); 2848 err = -EEXIST; 2849 goto errout_block_locked; 2850 } 2851 } else { 2852 if (!(n->nlmsg_flags & NLM_F_CREATE)) { 2853 NL_SET_ERR_MSG(extack, "Need both RTM_NEWCHAIN and NLM_F_CREATE to create a new chain"); 2854 err = -ENOENT; 2855 goto errout_block_locked; 2856 } 2857 chain = tcf_chain_create(block, chain_index); 2858 if (!chain) { 2859 NL_SET_ERR_MSG(extack, "Failed to create filter chain"); 2860 err = -ENOMEM; 2861 goto errout_block_locked; 2862 } 2863 } 2864 } else { 2865 if (!chain || tcf_chain_held_by_acts_only(chain)) { 2866 NL_SET_ERR_MSG(extack, "Cannot find specified filter chain"); 2867 err = -EINVAL; 2868 goto errout_block_locked; 2869 } 2870 tcf_chain_hold(chain); 2871 } 2872 2873 if (n->nlmsg_type == RTM_NEWCHAIN) { 2874 /* Modifying chain requires holding parent block lock. In case 2875 * the chain was successfully added, take a reference to the 2876 * chain. This ensures that an empty chain does not disappear at 2877 * the end of this function. 2878 */ 2879 tcf_chain_hold(chain); 2880 chain->explicitly_created = true; 2881 } 2882 mutex_unlock(&block->lock); 2883 2884 switch (n->nlmsg_type) { 2885 case RTM_NEWCHAIN: 2886 err = tc_chain_tmplt_add(chain, net, tca, extack); 2887 if (err) { 2888 tcf_chain_put_explicitly_created(chain); 2889 goto errout; 2890 } 2891 2892 tc_chain_notify(chain, NULL, 0, NLM_F_CREATE | NLM_F_EXCL, 2893 RTM_NEWCHAIN, false); 2894 break; 2895 case RTM_DELCHAIN: 2896 tfilter_notify_chain(net, skb, block, q, parent, n, 2897 chain, RTM_DELTFILTER, true); 2898 /* Flush the chain first as the user requested chain removal. */ 2899 tcf_chain_flush(chain, true); 2900 /* In case the chain was successfully deleted, put a reference 2901 * to the chain previously taken during addition. 2902 */ 2903 tcf_chain_put_explicitly_created(chain); 2904 break; 2905 case RTM_GETCHAIN: 2906 err = tc_chain_notify(chain, skb, n->nlmsg_seq, 2907 n->nlmsg_seq, n->nlmsg_type, true); 2908 if (err < 0) 2909 NL_SET_ERR_MSG(extack, "Failed to send chain notify message"); 2910 break; 2911 default: 2912 err = -EOPNOTSUPP; 2913 NL_SET_ERR_MSG(extack, "Unsupported message type"); 2914 goto errout; 2915 } 2916 2917 errout: 2918 tcf_chain_put(chain); 2919 errout_block: 2920 tcf_block_release(q, block, true); 2921 if (err == -EAGAIN) 2922 /* Replay the request. */ 2923 goto replay; 2924 return err; 2925 2926 errout_block_locked: 2927 mutex_unlock(&block->lock); 2928 goto errout_block; 2929 } 2930 2931 /* called with RTNL */ 2932 static int tc_dump_chain(struct sk_buff *skb, struct netlink_callback *cb) 2933 { 2934 struct net *net = sock_net(skb->sk); 2935 struct nlattr *tca[TCA_MAX + 1]; 2936 struct Qdisc *q = NULL; 2937 struct tcf_block *block; 2938 struct tcmsg *tcm = nlmsg_data(cb->nlh); 2939 struct tcf_chain *chain; 2940 long index_start; 2941 long index; 2942 u32 parent; 2943 int err; 2944 2945 if (nlmsg_len(cb->nlh) < sizeof(*tcm)) 2946 return skb->len; 2947 2948 err = nlmsg_parse_deprecated(cb->nlh, sizeof(*tcm), tca, TCA_MAX, 2949 rtm_tca_policy, cb->extack); 2950 if (err) 2951 return err; 2952 2953 if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK) { 2954 block = tcf_block_refcnt_get(net, tcm->tcm_block_index); 2955 if (!block) 2956 goto out; 2957 /* If we work with block index, q is NULL and parent value 2958 * will never be used in the following code. The check 2959 * in tcf_fill_node prevents it. However, compiler does not 2960 * see that far, so set parent to zero to silence the warning 2961 * about parent being uninitialized. 2962 */ 2963 parent = 0; 2964 } else { 2965 const struct Qdisc_class_ops *cops; 2966 struct net_device *dev; 2967 unsigned long cl = 0; 2968 2969 dev = __dev_get_by_index(net, tcm->tcm_ifindex); 2970 if (!dev) 2971 return skb->len; 2972 2973 parent = tcm->tcm_parent; 2974 if (!parent) { 2975 q = dev->qdisc; 2976 parent = q->handle; 2977 } else { 2978 q = qdisc_lookup(dev, TC_H_MAJ(tcm->tcm_parent)); 2979 } 2980 if (!q) 2981 goto out; 2982 cops = q->ops->cl_ops; 2983 if (!cops) 2984 goto out; 2985 if (!cops->tcf_block) 2986 goto out; 2987 if (TC_H_MIN(tcm->tcm_parent)) { 2988 cl = cops->find(q, tcm->tcm_parent); 2989 if (cl == 0) 2990 goto out; 2991 } 2992 block = cops->tcf_block(q, cl, NULL); 2993 if (!block) 2994 goto out; 2995 if (tcf_block_shared(block)) 2996 q = NULL; 2997 } 2998 2999 index_start = cb->args[0]; 3000 index = 0; 3001 3002 mutex_lock(&block->lock); 3003 list_for_each_entry(chain, &block->chain_list, list) { 3004 if ((tca[TCA_CHAIN] && 3005 nla_get_u32(tca[TCA_CHAIN]) != chain->index)) 3006 continue; 3007 if (index < index_start) { 3008 index++; 3009 continue; 3010 } 3011 if (tcf_chain_held_by_acts_only(chain)) 3012 continue; 3013 err = tc_chain_fill_node(chain->tmplt_ops, chain->tmplt_priv, 3014 chain->index, net, skb, block, 3015 NETLINK_CB(cb->skb).portid, 3016 cb->nlh->nlmsg_seq, NLM_F_MULTI, 3017 RTM_NEWCHAIN); 3018 if (err <= 0) 3019 break; 3020 index++; 3021 } 3022 mutex_unlock(&block->lock); 3023 3024 if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK) 3025 tcf_block_refcnt_put(block, true); 3026 cb->args[0] = index; 3027 3028 out: 3029 /* If we did no progress, the error (EMSGSIZE) is real */ 3030 if (skb->len == 0 && err) 3031 return err; 3032 return skb->len; 3033 } 3034 3035 void tcf_exts_destroy(struct tcf_exts *exts) 3036 { 3037 #ifdef CONFIG_NET_CLS_ACT 3038 if (exts->actions) { 3039 tcf_action_destroy(exts->actions, TCA_ACT_UNBIND); 3040 kfree(exts->actions); 3041 } 3042 exts->nr_actions = 0; 3043 #endif 3044 } 3045 EXPORT_SYMBOL(tcf_exts_destroy); 3046 3047 int tcf_exts_validate(struct net *net, struct tcf_proto *tp, struct nlattr **tb, 3048 struct nlattr *rate_tlv, struct tcf_exts *exts, bool ovr, 3049 bool rtnl_held, struct netlink_ext_ack *extack) 3050 { 3051 #ifdef CONFIG_NET_CLS_ACT 3052 { 3053 struct tc_action *act; 3054 size_t attr_size = 0; 3055 3056 if (exts->police && tb[exts->police]) { 3057 act = tcf_action_init_1(net, tp, tb[exts->police], 3058 rate_tlv, "police", ovr, 3059 TCA_ACT_BIND, rtnl_held, 3060 extack); 3061 if (IS_ERR(act)) 3062 return PTR_ERR(act); 3063 3064 act->type = exts->type = TCA_OLD_COMPAT; 3065 exts->actions[0] = act; 3066 exts->nr_actions = 1; 3067 } else if (exts->action && tb[exts->action]) { 3068 int err; 3069 3070 err = tcf_action_init(net, tp, tb[exts->action], 3071 rate_tlv, NULL, ovr, TCA_ACT_BIND, 3072 exts->actions, &attr_size, 3073 rtnl_held, extack); 3074 if (err < 0) 3075 return err; 3076 exts->nr_actions = err; 3077 } 3078 } 3079 #else 3080 if ((exts->action && tb[exts->action]) || 3081 (exts->police && tb[exts->police])) { 3082 NL_SET_ERR_MSG(extack, "Classifier actions are not supported per compile options (CONFIG_NET_CLS_ACT)"); 3083 return -EOPNOTSUPP; 3084 } 3085 #endif 3086 3087 return 0; 3088 } 3089 EXPORT_SYMBOL(tcf_exts_validate); 3090 3091 void tcf_exts_change(struct tcf_exts *dst, struct tcf_exts *src) 3092 { 3093 #ifdef CONFIG_NET_CLS_ACT 3094 struct tcf_exts old = *dst; 3095 3096 *dst = *src; 3097 tcf_exts_destroy(&old); 3098 #endif 3099 } 3100 EXPORT_SYMBOL(tcf_exts_change); 3101 3102 #ifdef CONFIG_NET_CLS_ACT 3103 static struct tc_action *tcf_exts_first_act(struct tcf_exts *exts) 3104 { 3105 if (exts->nr_actions == 0) 3106 return NULL; 3107 else 3108 return exts->actions[0]; 3109 } 3110 #endif 3111 3112 int tcf_exts_dump(struct sk_buff *skb, struct tcf_exts *exts) 3113 { 3114 #ifdef CONFIG_NET_CLS_ACT 3115 struct nlattr *nest; 3116 3117 if (exts->action && tcf_exts_has_actions(exts)) { 3118 /* 3119 * again for backward compatible mode - we want 3120 * to work with both old and new modes of entering 3121 * tc data even if iproute2 was newer - jhs 3122 */ 3123 if (exts->type != TCA_OLD_COMPAT) { 3124 nest = nla_nest_start_noflag(skb, exts->action); 3125 if (nest == NULL) 3126 goto nla_put_failure; 3127 3128 if (tcf_action_dump(skb, exts->actions, 0, 0, false) 3129 < 0) 3130 goto nla_put_failure; 3131 nla_nest_end(skb, nest); 3132 } else if (exts->police) { 3133 struct tc_action *act = tcf_exts_first_act(exts); 3134 nest = nla_nest_start_noflag(skb, exts->police); 3135 if (nest == NULL || !act) 3136 goto nla_put_failure; 3137 if (tcf_action_dump_old(skb, act, 0, 0) < 0) 3138 goto nla_put_failure; 3139 nla_nest_end(skb, nest); 3140 } 3141 } 3142 return 0; 3143 3144 nla_put_failure: 3145 nla_nest_cancel(skb, nest); 3146 return -1; 3147 #else 3148 return 0; 3149 #endif 3150 } 3151 EXPORT_SYMBOL(tcf_exts_dump); 3152 3153 int tcf_exts_terse_dump(struct sk_buff *skb, struct tcf_exts *exts) 3154 { 3155 #ifdef CONFIG_NET_CLS_ACT 3156 struct nlattr *nest; 3157 3158 if (!exts->action || !tcf_exts_has_actions(exts)) 3159 return 0; 3160 3161 nest = nla_nest_start_noflag(skb, exts->action); 3162 if (!nest) 3163 goto nla_put_failure; 3164 3165 if (tcf_action_dump(skb, exts->actions, 0, 0, true) < 0) 3166 goto nla_put_failure; 3167 nla_nest_end(skb, nest); 3168 return 0; 3169 3170 nla_put_failure: 3171 nla_nest_cancel(skb, nest); 3172 return -1; 3173 #else 3174 return 0; 3175 #endif 3176 } 3177 EXPORT_SYMBOL(tcf_exts_terse_dump); 3178 3179 int tcf_exts_dump_stats(struct sk_buff *skb, struct tcf_exts *exts) 3180 { 3181 #ifdef CONFIG_NET_CLS_ACT 3182 struct tc_action *a = tcf_exts_first_act(exts); 3183 if (a != NULL && tcf_action_copy_stats(skb, a, 1) < 0) 3184 return -1; 3185 #endif 3186 return 0; 3187 } 3188 EXPORT_SYMBOL(tcf_exts_dump_stats); 3189 3190 static void tcf_block_offload_inc(struct tcf_block *block, u32 *flags) 3191 { 3192 if (*flags & TCA_CLS_FLAGS_IN_HW) 3193 return; 3194 *flags |= TCA_CLS_FLAGS_IN_HW; 3195 atomic_inc(&block->offloadcnt); 3196 } 3197 3198 static void tcf_block_offload_dec(struct tcf_block *block, u32 *flags) 3199 { 3200 if (!(*flags & TCA_CLS_FLAGS_IN_HW)) 3201 return; 3202 *flags &= ~TCA_CLS_FLAGS_IN_HW; 3203 atomic_dec(&block->offloadcnt); 3204 } 3205 3206 static void tc_cls_offload_cnt_update(struct tcf_block *block, 3207 struct tcf_proto *tp, u32 *cnt, 3208 u32 *flags, u32 diff, bool add) 3209 { 3210 lockdep_assert_held(&block->cb_lock); 3211 3212 spin_lock(&tp->lock); 3213 if (add) { 3214 if (!*cnt) 3215 tcf_block_offload_inc(block, flags); 3216 *cnt += diff; 3217 } else { 3218 *cnt -= diff; 3219 if (!*cnt) 3220 tcf_block_offload_dec(block, flags); 3221 } 3222 spin_unlock(&tp->lock); 3223 } 3224 3225 static void 3226 tc_cls_offload_cnt_reset(struct tcf_block *block, struct tcf_proto *tp, 3227 u32 *cnt, u32 *flags) 3228 { 3229 lockdep_assert_held(&block->cb_lock); 3230 3231 spin_lock(&tp->lock); 3232 tcf_block_offload_dec(block, flags); 3233 *cnt = 0; 3234 spin_unlock(&tp->lock); 3235 } 3236 3237 static int 3238 __tc_setup_cb_call(struct tcf_block *block, enum tc_setup_type type, 3239 void *type_data, bool err_stop) 3240 { 3241 struct flow_block_cb *block_cb; 3242 int ok_count = 0; 3243 int err; 3244 3245 list_for_each_entry(block_cb, &block->flow_block.cb_list, list) { 3246 err = block_cb->cb(type, type_data, block_cb->cb_priv); 3247 if (err) { 3248 if (err_stop) 3249 return err; 3250 } else { 3251 ok_count++; 3252 } 3253 } 3254 return ok_count; 3255 } 3256 3257 int tc_setup_cb_call(struct tcf_block *block, enum tc_setup_type type, 3258 void *type_data, bool err_stop, bool rtnl_held) 3259 { 3260 bool take_rtnl = READ_ONCE(block->lockeddevcnt) && !rtnl_held; 3261 int ok_count; 3262 3263 retry: 3264 if (take_rtnl) 3265 rtnl_lock(); 3266 down_read(&block->cb_lock); 3267 /* Need to obtain rtnl lock if block is bound to devs that require it. 3268 * In block bind code cb_lock is obtained while holding rtnl, so we must 3269 * obtain the locks in same order here. 3270 */ 3271 if (!rtnl_held && !take_rtnl && block->lockeddevcnt) { 3272 up_read(&block->cb_lock); 3273 take_rtnl = true; 3274 goto retry; 3275 } 3276 3277 ok_count = __tc_setup_cb_call(block, type, type_data, err_stop); 3278 3279 up_read(&block->cb_lock); 3280 if (take_rtnl) 3281 rtnl_unlock(); 3282 return ok_count; 3283 } 3284 EXPORT_SYMBOL(tc_setup_cb_call); 3285 3286 /* Non-destructive filter add. If filter that wasn't already in hardware is 3287 * successfully offloaded, increment block offloads counter. On failure, 3288 * previously offloaded filter is considered to be intact and offloads counter 3289 * is not decremented. 3290 */ 3291 3292 int tc_setup_cb_add(struct tcf_block *block, struct tcf_proto *tp, 3293 enum tc_setup_type type, void *type_data, bool err_stop, 3294 u32 *flags, unsigned int *in_hw_count, bool rtnl_held) 3295 { 3296 bool take_rtnl = READ_ONCE(block->lockeddevcnt) && !rtnl_held; 3297 int ok_count; 3298 3299 retry: 3300 if (take_rtnl) 3301 rtnl_lock(); 3302 down_read(&block->cb_lock); 3303 /* Need to obtain rtnl lock if block is bound to devs that require it. 3304 * In block bind code cb_lock is obtained while holding rtnl, so we must 3305 * obtain the locks in same order here. 3306 */ 3307 if (!rtnl_held && !take_rtnl && block->lockeddevcnt) { 3308 up_read(&block->cb_lock); 3309 take_rtnl = true; 3310 goto retry; 3311 } 3312 3313 /* Make sure all netdevs sharing this block are offload-capable. */ 3314 if (block->nooffloaddevcnt && err_stop) { 3315 ok_count = -EOPNOTSUPP; 3316 goto err_unlock; 3317 } 3318 3319 ok_count = __tc_setup_cb_call(block, type, type_data, err_stop); 3320 if (ok_count < 0) 3321 goto err_unlock; 3322 3323 if (tp->ops->hw_add) 3324 tp->ops->hw_add(tp, type_data); 3325 if (ok_count > 0) 3326 tc_cls_offload_cnt_update(block, tp, in_hw_count, flags, 3327 ok_count, true); 3328 err_unlock: 3329 up_read(&block->cb_lock); 3330 if (take_rtnl) 3331 rtnl_unlock(); 3332 return ok_count < 0 ? ok_count : 0; 3333 } 3334 EXPORT_SYMBOL(tc_setup_cb_add); 3335 3336 /* Destructive filter replace. If filter that wasn't already in hardware is 3337 * successfully offloaded, increment block offload counter. On failure, 3338 * previously offloaded filter is considered to be destroyed and offload counter 3339 * is decremented. 3340 */ 3341 3342 int tc_setup_cb_replace(struct tcf_block *block, struct tcf_proto *tp, 3343 enum tc_setup_type type, void *type_data, bool err_stop, 3344 u32 *old_flags, unsigned int *old_in_hw_count, 3345 u32 *new_flags, unsigned int *new_in_hw_count, 3346 bool rtnl_held) 3347 { 3348 bool take_rtnl = READ_ONCE(block->lockeddevcnt) && !rtnl_held; 3349 int ok_count; 3350 3351 retry: 3352 if (take_rtnl) 3353 rtnl_lock(); 3354 down_read(&block->cb_lock); 3355 /* Need to obtain rtnl lock if block is bound to devs that require it. 3356 * In block bind code cb_lock is obtained while holding rtnl, so we must 3357 * obtain the locks in same order here. 3358 */ 3359 if (!rtnl_held && !take_rtnl && block->lockeddevcnt) { 3360 up_read(&block->cb_lock); 3361 take_rtnl = true; 3362 goto retry; 3363 } 3364 3365 /* Make sure all netdevs sharing this block are offload-capable. */ 3366 if (block->nooffloaddevcnt && err_stop) { 3367 ok_count = -EOPNOTSUPP; 3368 goto err_unlock; 3369 } 3370 3371 tc_cls_offload_cnt_reset(block, tp, old_in_hw_count, old_flags); 3372 if (tp->ops->hw_del) 3373 tp->ops->hw_del(tp, type_data); 3374 3375 ok_count = __tc_setup_cb_call(block, type, type_data, err_stop); 3376 if (ok_count < 0) 3377 goto err_unlock; 3378 3379 if (tp->ops->hw_add) 3380 tp->ops->hw_add(tp, type_data); 3381 if (ok_count > 0) 3382 tc_cls_offload_cnt_update(block, tp, new_in_hw_count, 3383 new_flags, ok_count, true); 3384 err_unlock: 3385 up_read(&block->cb_lock); 3386 if (take_rtnl) 3387 rtnl_unlock(); 3388 return ok_count < 0 ? ok_count : 0; 3389 } 3390 EXPORT_SYMBOL(tc_setup_cb_replace); 3391 3392 /* Destroy filter and decrement block offload counter, if filter was previously 3393 * offloaded. 3394 */ 3395 3396 int tc_setup_cb_destroy(struct tcf_block *block, struct tcf_proto *tp, 3397 enum tc_setup_type type, void *type_data, bool err_stop, 3398 u32 *flags, unsigned int *in_hw_count, bool rtnl_held) 3399 { 3400 bool take_rtnl = READ_ONCE(block->lockeddevcnt) && !rtnl_held; 3401 int ok_count; 3402 3403 retry: 3404 if (take_rtnl) 3405 rtnl_lock(); 3406 down_read(&block->cb_lock); 3407 /* Need to obtain rtnl lock if block is bound to devs that require it. 3408 * In block bind code cb_lock is obtained while holding rtnl, so we must 3409 * obtain the locks in same order here. 3410 */ 3411 if (!rtnl_held && !take_rtnl && block->lockeddevcnt) { 3412 up_read(&block->cb_lock); 3413 take_rtnl = true; 3414 goto retry; 3415 } 3416 3417 ok_count = __tc_setup_cb_call(block, type, type_data, err_stop); 3418 3419 tc_cls_offload_cnt_reset(block, tp, in_hw_count, flags); 3420 if (tp->ops->hw_del) 3421 tp->ops->hw_del(tp, type_data); 3422 3423 up_read(&block->cb_lock); 3424 if (take_rtnl) 3425 rtnl_unlock(); 3426 return ok_count < 0 ? ok_count : 0; 3427 } 3428 EXPORT_SYMBOL(tc_setup_cb_destroy); 3429 3430 int tc_setup_cb_reoffload(struct tcf_block *block, struct tcf_proto *tp, 3431 bool add, flow_setup_cb_t *cb, 3432 enum tc_setup_type type, void *type_data, 3433 void *cb_priv, u32 *flags, unsigned int *in_hw_count) 3434 { 3435 int err = cb(type, type_data, cb_priv); 3436 3437 if (err) { 3438 if (add && tc_skip_sw(*flags)) 3439 return err; 3440 } else { 3441 tc_cls_offload_cnt_update(block, tp, in_hw_count, flags, 1, 3442 add); 3443 } 3444 3445 return 0; 3446 } 3447 EXPORT_SYMBOL(tc_setup_cb_reoffload); 3448 3449 static int tcf_act_get_cookie(struct flow_action_entry *entry, 3450 const struct tc_action *act) 3451 { 3452 struct tc_cookie *cookie; 3453 int err = 0; 3454 3455 rcu_read_lock(); 3456 cookie = rcu_dereference(act->act_cookie); 3457 if (cookie) { 3458 entry->cookie = flow_action_cookie_create(cookie->data, 3459 cookie->len, 3460 GFP_ATOMIC); 3461 if (!entry->cookie) 3462 err = -ENOMEM; 3463 } 3464 rcu_read_unlock(); 3465 return err; 3466 } 3467 3468 static void tcf_act_put_cookie(struct flow_action_entry *entry) 3469 { 3470 flow_action_cookie_destroy(entry->cookie); 3471 } 3472 3473 void tc_cleanup_flow_action(struct flow_action *flow_action) 3474 { 3475 struct flow_action_entry *entry; 3476 int i; 3477 3478 flow_action_for_each(i, entry, flow_action) { 3479 tcf_act_put_cookie(entry); 3480 if (entry->destructor) 3481 entry->destructor(entry->destructor_priv); 3482 } 3483 } 3484 EXPORT_SYMBOL(tc_cleanup_flow_action); 3485 3486 static void tcf_mirred_get_dev(struct flow_action_entry *entry, 3487 const struct tc_action *act) 3488 { 3489 #ifdef CONFIG_NET_CLS_ACT 3490 entry->dev = act->ops->get_dev(act, &entry->destructor); 3491 if (!entry->dev) 3492 return; 3493 entry->destructor_priv = entry->dev; 3494 #endif 3495 } 3496 3497 static void tcf_tunnel_encap_put_tunnel(void *priv) 3498 { 3499 struct ip_tunnel_info *tunnel = priv; 3500 3501 kfree(tunnel); 3502 } 3503 3504 static int tcf_tunnel_encap_get_tunnel(struct flow_action_entry *entry, 3505 const struct tc_action *act) 3506 { 3507 entry->tunnel = tcf_tunnel_info_copy(act); 3508 if (!entry->tunnel) 3509 return -ENOMEM; 3510 entry->destructor = tcf_tunnel_encap_put_tunnel; 3511 entry->destructor_priv = entry->tunnel; 3512 return 0; 3513 } 3514 3515 static void tcf_sample_get_group(struct flow_action_entry *entry, 3516 const struct tc_action *act) 3517 { 3518 #ifdef CONFIG_NET_CLS_ACT 3519 entry->sample.psample_group = 3520 act->ops->get_psample_group(act, &entry->destructor); 3521 entry->destructor_priv = entry->sample.psample_group; 3522 #endif 3523 } 3524 3525 static void tcf_gate_entry_destructor(void *priv) 3526 { 3527 struct action_gate_entry *oe = priv; 3528 3529 kfree(oe); 3530 } 3531 3532 static int tcf_gate_get_entries(struct flow_action_entry *entry, 3533 const struct tc_action *act) 3534 { 3535 entry->gate.entries = tcf_gate_get_list(act); 3536 3537 if (!entry->gate.entries) 3538 return -EINVAL; 3539 3540 entry->destructor = tcf_gate_entry_destructor; 3541 entry->destructor_priv = entry->gate.entries; 3542 3543 return 0; 3544 } 3545 3546 static enum flow_action_hw_stats tc_act_hw_stats(u8 hw_stats) 3547 { 3548 if (WARN_ON_ONCE(hw_stats > TCA_ACT_HW_STATS_ANY)) 3549 return FLOW_ACTION_HW_STATS_DONT_CARE; 3550 else if (!hw_stats) 3551 return FLOW_ACTION_HW_STATS_DISABLED; 3552 3553 return hw_stats; 3554 } 3555 3556 int tc_setup_flow_action(struct flow_action *flow_action, 3557 const struct tcf_exts *exts) 3558 { 3559 struct tc_action *act; 3560 int i, j, k, err = 0; 3561 3562 BUILD_BUG_ON(TCA_ACT_HW_STATS_ANY != FLOW_ACTION_HW_STATS_ANY); 3563 BUILD_BUG_ON(TCA_ACT_HW_STATS_IMMEDIATE != FLOW_ACTION_HW_STATS_IMMEDIATE); 3564 BUILD_BUG_ON(TCA_ACT_HW_STATS_DELAYED != FLOW_ACTION_HW_STATS_DELAYED); 3565 3566 if (!exts) 3567 return 0; 3568 3569 j = 0; 3570 tcf_exts_for_each_action(i, act, exts) { 3571 struct flow_action_entry *entry; 3572 3573 entry = &flow_action->entries[j]; 3574 spin_lock_bh(&act->tcfa_lock); 3575 err = tcf_act_get_cookie(entry, act); 3576 if (err) 3577 goto err_out_locked; 3578 3579 entry->hw_stats = tc_act_hw_stats(act->hw_stats); 3580 3581 if (is_tcf_gact_ok(act)) { 3582 entry->id = FLOW_ACTION_ACCEPT; 3583 } else if (is_tcf_gact_shot(act)) { 3584 entry->id = FLOW_ACTION_DROP; 3585 } else if (is_tcf_gact_trap(act)) { 3586 entry->id = FLOW_ACTION_TRAP; 3587 } else if (is_tcf_gact_goto_chain(act)) { 3588 entry->id = FLOW_ACTION_GOTO; 3589 entry->chain_index = tcf_gact_goto_chain_index(act); 3590 } else if (is_tcf_mirred_egress_redirect(act)) { 3591 entry->id = FLOW_ACTION_REDIRECT; 3592 tcf_mirred_get_dev(entry, act); 3593 } else if (is_tcf_mirred_egress_mirror(act)) { 3594 entry->id = FLOW_ACTION_MIRRED; 3595 tcf_mirred_get_dev(entry, act); 3596 } else if (is_tcf_mirred_ingress_redirect(act)) { 3597 entry->id = FLOW_ACTION_REDIRECT_INGRESS; 3598 tcf_mirred_get_dev(entry, act); 3599 } else if (is_tcf_mirred_ingress_mirror(act)) { 3600 entry->id = FLOW_ACTION_MIRRED_INGRESS; 3601 tcf_mirred_get_dev(entry, act); 3602 } else if (is_tcf_vlan(act)) { 3603 switch (tcf_vlan_action(act)) { 3604 case TCA_VLAN_ACT_PUSH: 3605 entry->id = FLOW_ACTION_VLAN_PUSH; 3606 entry->vlan.vid = tcf_vlan_push_vid(act); 3607 entry->vlan.proto = tcf_vlan_push_proto(act); 3608 entry->vlan.prio = tcf_vlan_push_prio(act); 3609 break; 3610 case TCA_VLAN_ACT_POP: 3611 entry->id = FLOW_ACTION_VLAN_POP; 3612 break; 3613 case TCA_VLAN_ACT_MODIFY: 3614 entry->id = FLOW_ACTION_VLAN_MANGLE; 3615 entry->vlan.vid = tcf_vlan_push_vid(act); 3616 entry->vlan.proto = tcf_vlan_push_proto(act); 3617 entry->vlan.prio = tcf_vlan_push_prio(act); 3618 break; 3619 default: 3620 err = -EOPNOTSUPP; 3621 goto err_out_locked; 3622 } 3623 } else if (is_tcf_tunnel_set(act)) { 3624 entry->id = FLOW_ACTION_TUNNEL_ENCAP; 3625 err = tcf_tunnel_encap_get_tunnel(entry, act); 3626 if (err) 3627 goto err_out_locked; 3628 } else if (is_tcf_tunnel_release(act)) { 3629 entry->id = FLOW_ACTION_TUNNEL_DECAP; 3630 } else if (is_tcf_pedit(act)) { 3631 for (k = 0; k < tcf_pedit_nkeys(act); k++) { 3632 switch (tcf_pedit_cmd(act, k)) { 3633 case TCA_PEDIT_KEY_EX_CMD_SET: 3634 entry->id = FLOW_ACTION_MANGLE; 3635 break; 3636 case TCA_PEDIT_KEY_EX_CMD_ADD: 3637 entry->id = FLOW_ACTION_ADD; 3638 break; 3639 default: 3640 err = -EOPNOTSUPP; 3641 goto err_out_locked; 3642 } 3643 entry->mangle.htype = tcf_pedit_htype(act, k); 3644 entry->mangle.mask = tcf_pedit_mask(act, k); 3645 entry->mangle.val = tcf_pedit_val(act, k); 3646 entry->mangle.offset = tcf_pedit_offset(act, k); 3647 entry->hw_stats = tc_act_hw_stats(act->hw_stats); 3648 entry = &flow_action->entries[++j]; 3649 } 3650 } else if (is_tcf_csum(act)) { 3651 entry->id = FLOW_ACTION_CSUM; 3652 entry->csum_flags = tcf_csum_update_flags(act); 3653 } else if (is_tcf_skbedit_mark(act)) { 3654 entry->id = FLOW_ACTION_MARK; 3655 entry->mark = tcf_skbedit_mark(act); 3656 } else if (is_tcf_sample(act)) { 3657 entry->id = FLOW_ACTION_SAMPLE; 3658 entry->sample.trunc_size = tcf_sample_trunc_size(act); 3659 entry->sample.truncate = tcf_sample_truncate(act); 3660 entry->sample.rate = tcf_sample_rate(act); 3661 tcf_sample_get_group(entry, act); 3662 } else if (is_tcf_police(act)) { 3663 entry->id = FLOW_ACTION_POLICE; 3664 entry->police.burst = tcf_police_burst(act); 3665 entry->police.rate_bytes_ps = 3666 tcf_police_rate_bytes_ps(act); 3667 entry->police.mtu = tcf_police_tcfp_mtu(act); 3668 entry->police.index = act->tcfa_index; 3669 } else if (is_tcf_ct(act)) { 3670 entry->id = FLOW_ACTION_CT; 3671 entry->ct.action = tcf_ct_action(act); 3672 entry->ct.zone = tcf_ct_zone(act); 3673 entry->ct.flow_table = tcf_ct_ft(act); 3674 } else if (is_tcf_mpls(act)) { 3675 switch (tcf_mpls_action(act)) { 3676 case TCA_MPLS_ACT_PUSH: 3677 entry->id = FLOW_ACTION_MPLS_PUSH; 3678 entry->mpls_push.proto = tcf_mpls_proto(act); 3679 entry->mpls_push.label = tcf_mpls_label(act); 3680 entry->mpls_push.tc = tcf_mpls_tc(act); 3681 entry->mpls_push.bos = tcf_mpls_bos(act); 3682 entry->mpls_push.ttl = tcf_mpls_ttl(act); 3683 break; 3684 case TCA_MPLS_ACT_POP: 3685 entry->id = FLOW_ACTION_MPLS_POP; 3686 entry->mpls_pop.proto = tcf_mpls_proto(act); 3687 break; 3688 case TCA_MPLS_ACT_MODIFY: 3689 entry->id = FLOW_ACTION_MPLS_MANGLE; 3690 entry->mpls_mangle.label = tcf_mpls_label(act); 3691 entry->mpls_mangle.tc = tcf_mpls_tc(act); 3692 entry->mpls_mangle.bos = tcf_mpls_bos(act); 3693 entry->mpls_mangle.ttl = tcf_mpls_ttl(act); 3694 break; 3695 default: 3696 goto err_out_locked; 3697 } 3698 } else if (is_tcf_skbedit_ptype(act)) { 3699 entry->id = FLOW_ACTION_PTYPE; 3700 entry->ptype = tcf_skbedit_ptype(act); 3701 } else if (is_tcf_skbedit_priority(act)) { 3702 entry->id = FLOW_ACTION_PRIORITY; 3703 entry->priority = tcf_skbedit_priority(act); 3704 } else if (is_tcf_gate(act)) { 3705 entry->id = FLOW_ACTION_GATE; 3706 entry->gate.index = tcf_gate_index(act); 3707 entry->gate.prio = tcf_gate_prio(act); 3708 entry->gate.basetime = tcf_gate_basetime(act); 3709 entry->gate.cycletime = tcf_gate_cycletime(act); 3710 entry->gate.cycletimeext = tcf_gate_cycletimeext(act); 3711 entry->gate.num_entries = tcf_gate_num_entries(act); 3712 err = tcf_gate_get_entries(entry, act); 3713 if (err) 3714 goto err_out; 3715 } else { 3716 err = -EOPNOTSUPP; 3717 goto err_out_locked; 3718 } 3719 spin_unlock_bh(&act->tcfa_lock); 3720 3721 if (!is_tcf_pedit(act)) 3722 j++; 3723 } 3724 3725 err_out: 3726 if (err) 3727 tc_cleanup_flow_action(flow_action); 3728 3729 return err; 3730 err_out_locked: 3731 spin_unlock_bh(&act->tcfa_lock); 3732 goto err_out; 3733 } 3734 EXPORT_SYMBOL(tc_setup_flow_action); 3735 3736 unsigned int tcf_exts_num_actions(struct tcf_exts *exts) 3737 { 3738 unsigned int num_acts = 0; 3739 struct tc_action *act; 3740 int i; 3741 3742 tcf_exts_for_each_action(i, act, exts) { 3743 if (is_tcf_pedit(act)) 3744 num_acts += tcf_pedit_nkeys(act); 3745 else 3746 num_acts++; 3747 } 3748 return num_acts; 3749 } 3750 EXPORT_SYMBOL(tcf_exts_num_actions); 3751 3752 #ifdef CONFIG_NET_CLS_ACT 3753 static int tcf_qevent_parse_block_index(struct nlattr *block_index_attr, 3754 u32 *p_block_index, 3755 struct netlink_ext_ack *extack) 3756 { 3757 *p_block_index = nla_get_u32(block_index_attr); 3758 if (!*p_block_index) { 3759 NL_SET_ERR_MSG(extack, "Block number may not be zero"); 3760 return -EINVAL; 3761 } 3762 3763 return 0; 3764 } 3765 3766 int tcf_qevent_init(struct tcf_qevent *qe, struct Qdisc *sch, 3767 enum flow_block_binder_type binder_type, 3768 struct nlattr *block_index_attr, 3769 struct netlink_ext_ack *extack) 3770 { 3771 u32 block_index; 3772 int err; 3773 3774 if (!block_index_attr) 3775 return 0; 3776 3777 err = tcf_qevent_parse_block_index(block_index_attr, &block_index, extack); 3778 if (err) 3779 return err; 3780 3781 if (!block_index) 3782 return 0; 3783 3784 qe->info.binder_type = binder_type; 3785 qe->info.chain_head_change = tcf_chain_head_change_dflt; 3786 qe->info.chain_head_change_priv = &qe->filter_chain; 3787 qe->info.block_index = block_index; 3788 3789 return tcf_block_get_ext(&qe->block, sch, &qe->info, extack); 3790 } 3791 EXPORT_SYMBOL(tcf_qevent_init); 3792 3793 void tcf_qevent_destroy(struct tcf_qevent *qe, struct Qdisc *sch) 3794 { 3795 if (qe->info.block_index) 3796 tcf_block_put_ext(qe->block, sch, &qe->info); 3797 } 3798 EXPORT_SYMBOL(tcf_qevent_destroy); 3799 3800 int tcf_qevent_validate_change(struct tcf_qevent *qe, struct nlattr *block_index_attr, 3801 struct netlink_ext_ack *extack) 3802 { 3803 u32 block_index; 3804 int err; 3805 3806 if (!block_index_attr) 3807 return 0; 3808 3809 err = tcf_qevent_parse_block_index(block_index_attr, &block_index, extack); 3810 if (err) 3811 return err; 3812 3813 /* Bounce newly-configured block or change in block. */ 3814 if (block_index != qe->info.block_index) { 3815 NL_SET_ERR_MSG(extack, "Change of blocks is not supported"); 3816 return -EINVAL; 3817 } 3818 3819 return 0; 3820 } 3821 EXPORT_SYMBOL(tcf_qevent_validate_change); 3822 3823 struct sk_buff *tcf_qevent_handle(struct tcf_qevent *qe, struct Qdisc *sch, struct sk_buff *skb, 3824 struct sk_buff **to_free, int *ret) 3825 { 3826 struct tcf_result cl_res; 3827 struct tcf_proto *fl; 3828 3829 if (!qe->info.block_index) 3830 return skb; 3831 3832 fl = rcu_dereference_bh(qe->filter_chain); 3833 3834 switch (tcf_classify(skb, fl, &cl_res, false)) { 3835 case TC_ACT_SHOT: 3836 qdisc_qstats_drop(sch); 3837 __qdisc_drop(skb, to_free); 3838 *ret = __NET_XMIT_BYPASS; 3839 return NULL; 3840 case TC_ACT_STOLEN: 3841 case TC_ACT_QUEUED: 3842 case TC_ACT_TRAP: 3843 __qdisc_drop(skb, to_free); 3844 *ret = __NET_XMIT_STOLEN; 3845 return NULL; 3846 case TC_ACT_REDIRECT: 3847 skb_do_redirect(skb); 3848 *ret = __NET_XMIT_STOLEN; 3849 return NULL; 3850 } 3851 3852 return skb; 3853 } 3854 EXPORT_SYMBOL(tcf_qevent_handle); 3855 3856 int tcf_qevent_dump(struct sk_buff *skb, int attr_name, struct tcf_qevent *qe) 3857 { 3858 if (!qe->info.block_index) 3859 return 0; 3860 return nla_put_u32(skb, attr_name, qe->info.block_index); 3861 } 3862 EXPORT_SYMBOL(tcf_qevent_dump); 3863 #endif 3864 3865 static __net_init int tcf_net_init(struct net *net) 3866 { 3867 struct tcf_net *tn = net_generic(net, tcf_net_id); 3868 3869 spin_lock_init(&tn->idr_lock); 3870 idr_init(&tn->idr); 3871 return 0; 3872 } 3873 3874 static void __net_exit tcf_net_exit(struct net *net) 3875 { 3876 struct tcf_net *tn = net_generic(net, tcf_net_id); 3877 3878 idr_destroy(&tn->idr); 3879 } 3880 3881 static struct pernet_operations tcf_net_ops = { 3882 .init = tcf_net_init, 3883 .exit = tcf_net_exit, 3884 .id = &tcf_net_id, 3885 .size = sizeof(struct tcf_net), 3886 }; 3887 3888 static int __init tc_filter_init(void) 3889 { 3890 int err; 3891 3892 tc_filter_wq = alloc_ordered_workqueue("tc_filter_workqueue", 0); 3893 if (!tc_filter_wq) 3894 return -ENOMEM; 3895 3896 err = register_pernet_subsys(&tcf_net_ops); 3897 if (err) 3898 goto err_register_pernet_subsys; 3899 3900 rtnl_register(PF_UNSPEC, RTM_NEWTFILTER, tc_new_tfilter, NULL, 3901 RTNL_FLAG_DOIT_UNLOCKED); 3902 rtnl_register(PF_UNSPEC, RTM_DELTFILTER, tc_del_tfilter, NULL, 3903 RTNL_FLAG_DOIT_UNLOCKED); 3904 rtnl_register(PF_UNSPEC, RTM_GETTFILTER, tc_get_tfilter, 3905 tc_dump_tfilter, RTNL_FLAG_DOIT_UNLOCKED); 3906 rtnl_register(PF_UNSPEC, RTM_NEWCHAIN, tc_ctl_chain, NULL, 0); 3907 rtnl_register(PF_UNSPEC, RTM_DELCHAIN, tc_ctl_chain, NULL, 0); 3908 rtnl_register(PF_UNSPEC, RTM_GETCHAIN, tc_ctl_chain, 3909 tc_dump_chain, 0); 3910 3911 return 0; 3912 3913 err_register_pernet_subsys: 3914 destroy_workqueue(tc_filter_wq); 3915 return err; 3916 } 3917 3918 subsys_initcall(tc_filter_init); 3919