1 /* 2 * net/sched/cls_api.c Packet classifier API. 3 * 4 * This program is free software; you can redistribute it and/or 5 * modify it under the terms of the GNU General Public License 6 * as published by the Free Software Foundation; either version 7 * 2 of the License, or (at your option) any later version. 8 * 9 * Authors: Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru> 10 * 11 * Changes: 12 * 13 * Eduardo J. Blanco <ejbs@netlabs.com.uy> :990222: kmod support 14 * 15 */ 16 17 #include <linux/module.h> 18 #include <linux/types.h> 19 #include <linux/kernel.h> 20 #include <linux/string.h> 21 #include <linux/errno.h> 22 #include <linux/err.h> 23 #include <linux/skbuff.h> 24 #include <linux/init.h> 25 #include <linux/kmod.h> 26 #include <linux/err.h> 27 #include <linux/slab.h> 28 #include <net/net_namespace.h> 29 #include <net/sock.h> 30 #include <net/netlink.h> 31 #include <net/pkt_sched.h> 32 #include <net/pkt_cls.h> 33 34 /* The list of all installed classifier types */ 35 static LIST_HEAD(tcf_proto_base); 36 37 /* Protects list of registered TC modules. It is pure SMP lock. */ 38 static DEFINE_RWLOCK(cls_mod_lock); 39 40 /* Find classifier type by string name */ 41 42 static const struct tcf_proto_ops *tcf_proto_lookup_ops(const char *kind) 43 { 44 const struct tcf_proto_ops *t, *res = NULL; 45 46 if (kind) { 47 read_lock(&cls_mod_lock); 48 list_for_each_entry(t, &tcf_proto_base, head) { 49 if (strcmp(kind, t->kind) == 0) { 50 if (try_module_get(t->owner)) 51 res = t; 52 break; 53 } 54 } 55 read_unlock(&cls_mod_lock); 56 } 57 return res; 58 } 59 60 /* Register(unregister) new classifier type */ 61 62 int register_tcf_proto_ops(struct tcf_proto_ops *ops) 63 { 64 struct tcf_proto_ops *t; 65 int rc = -EEXIST; 66 67 write_lock(&cls_mod_lock); 68 list_for_each_entry(t, &tcf_proto_base, head) 69 if (!strcmp(ops->kind, t->kind)) 70 goto out; 71 72 list_add_tail(&ops->head, &tcf_proto_base); 73 rc = 0; 74 out: 75 write_unlock(&cls_mod_lock); 76 return rc; 77 } 78 EXPORT_SYMBOL(register_tcf_proto_ops); 79 80 int unregister_tcf_proto_ops(struct tcf_proto_ops *ops) 81 { 82 struct tcf_proto_ops *t; 83 int rc = -ENOENT; 84 85 /* Wait for outstanding call_rcu()s, if any, from a 86 * tcf_proto_ops's destroy() handler. 87 */ 88 rcu_barrier(); 89 90 write_lock(&cls_mod_lock); 91 list_for_each_entry(t, &tcf_proto_base, head) { 92 if (t == ops) { 93 list_del(&t->head); 94 rc = 0; 95 break; 96 } 97 } 98 write_unlock(&cls_mod_lock); 99 return rc; 100 } 101 EXPORT_SYMBOL(unregister_tcf_proto_ops); 102 103 static int tfilter_notify(struct net *net, struct sk_buff *oskb, 104 struct nlmsghdr *n, struct tcf_proto *tp, 105 unsigned long fh, int event, bool unicast); 106 107 static void tfilter_notify_chain(struct net *net, struct sk_buff *oskb, 108 struct nlmsghdr *n, 109 struct tcf_chain *chain, int event) 110 { 111 struct tcf_proto *tp; 112 113 for (tp = rtnl_dereference(chain->filter_chain); 114 tp; tp = rtnl_dereference(tp->next)) 115 tfilter_notify(net, oskb, n, tp, 0, event, false); 116 } 117 118 /* Select new prio value from the range, managed by kernel. */ 119 120 static inline u32 tcf_auto_prio(struct tcf_proto *tp) 121 { 122 u32 first = TC_H_MAKE(0xC0000000U, 0U); 123 124 if (tp) 125 first = tp->prio - 1; 126 127 return TC_H_MAJ(first); 128 } 129 130 static struct tcf_proto *tcf_proto_create(const char *kind, u32 protocol, 131 u32 prio, u32 parent, struct Qdisc *q, 132 struct tcf_chain *chain) 133 { 134 struct tcf_proto *tp; 135 int err; 136 137 tp = kzalloc(sizeof(*tp), GFP_KERNEL); 138 if (!tp) 139 return ERR_PTR(-ENOBUFS); 140 141 err = -ENOENT; 142 tp->ops = tcf_proto_lookup_ops(kind); 143 if (!tp->ops) { 144 #ifdef CONFIG_MODULES 145 rtnl_unlock(); 146 request_module("cls_%s", kind); 147 rtnl_lock(); 148 tp->ops = tcf_proto_lookup_ops(kind); 149 /* We dropped the RTNL semaphore in order to perform 150 * the module load. So, even if we succeeded in loading 151 * the module we have to replay the request. We indicate 152 * this using -EAGAIN. 153 */ 154 if (tp->ops) { 155 module_put(tp->ops->owner); 156 err = -EAGAIN; 157 } else { 158 err = -ENOENT; 159 } 160 goto errout; 161 #endif 162 } 163 tp->classify = tp->ops->classify; 164 tp->protocol = protocol; 165 tp->prio = prio; 166 tp->classid = parent; 167 tp->q = q; 168 tp->chain = chain; 169 170 err = tp->ops->init(tp); 171 if (err) { 172 module_put(tp->ops->owner); 173 goto errout; 174 } 175 return tp; 176 177 errout: 178 kfree(tp); 179 return ERR_PTR(err); 180 } 181 182 static void tcf_proto_destroy(struct tcf_proto *tp) 183 { 184 tp->ops->destroy(tp); 185 module_put(tp->ops->owner); 186 kfree_rcu(tp, rcu); 187 } 188 189 static struct tcf_chain *tcf_chain_create(struct tcf_block *block, 190 u32 chain_index) 191 { 192 struct tcf_chain *chain; 193 194 chain = kzalloc(sizeof(*chain), GFP_KERNEL); 195 if (!chain) 196 return NULL; 197 list_add_tail(&chain->list, &block->chain_list); 198 chain->block = block; 199 chain->index = chain_index; 200 chain->refcnt = 1; 201 return chain; 202 } 203 204 static void tcf_chain_flush(struct tcf_chain *chain) 205 { 206 struct tcf_proto *tp; 207 208 if (*chain->p_filter_chain) 209 RCU_INIT_POINTER(*chain->p_filter_chain, NULL); 210 while ((tp = rtnl_dereference(chain->filter_chain)) != NULL) { 211 RCU_INIT_POINTER(chain->filter_chain, tp->next); 212 tcf_proto_destroy(tp); 213 } 214 } 215 216 static void tcf_chain_destroy(struct tcf_chain *chain) 217 { 218 list_del(&chain->list); 219 tcf_chain_flush(chain); 220 kfree(chain); 221 } 222 223 struct tcf_chain *tcf_chain_get(struct tcf_block *block, u32 chain_index, 224 bool create) 225 { 226 struct tcf_chain *chain; 227 228 list_for_each_entry(chain, &block->chain_list, list) { 229 if (chain->index == chain_index) { 230 chain->refcnt++; 231 return chain; 232 } 233 } 234 if (create) 235 return tcf_chain_create(block, chain_index); 236 else 237 return NULL; 238 } 239 EXPORT_SYMBOL(tcf_chain_get); 240 241 void tcf_chain_put(struct tcf_chain *chain) 242 { 243 /* Destroy unused chain, with exception of chain 0, which is the 244 * default one and has to be always present. 245 */ 246 if (--chain->refcnt == 0 && !chain->filter_chain && chain->index != 0) 247 tcf_chain_destroy(chain); 248 } 249 EXPORT_SYMBOL(tcf_chain_put); 250 251 static void 252 tcf_chain_filter_chain_ptr_set(struct tcf_chain *chain, 253 struct tcf_proto __rcu **p_filter_chain) 254 { 255 chain->p_filter_chain = p_filter_chain; 256 } 257 258 int tcf_block_get(struct tcf_block **p_block, 259 struct tcf_proto __rcu **p_filter_chain) 260 { 261 struct tcf_block *block = kzalloc(sizeof(*block), GFP_KERNEL); 262 struct tcf_chain *chain; 263 int err; 264 265 if (!block) 266 return -ENOMEM; 267 INIT_LIST_HEAD(&block->chain_list); 268 /* Create chain 0 by default, it has to be always present. */ 269 chain = tcf_chain_create(block, 0); 270 if (!chain) { 271 err = -ENOMEM; 272 goto err_chain_create; 273 } 274 tcf_chain_filter_chain_ptr_set(chain, p_filter_chain); 275 *p_block = block; 276 return 0; 277 278 err_chain_create: 279 kfree(block); 280 return err; 281 } 282 EXPORT_SYMBOL(tcf_block_get); 283 284 void tcf_block_put(struct tcf_block *block) 285 { 286 struct tcf_chain *chain, *tmp; 287 288 if (!block) 289 return; 290 291 list_for_each_entry_safe(chain, tmp, &block->chain_list, list) 292 tcf_chain_destroy(chain); 293 kfree(block); 294 } 295 EXPORT_SYMBOL(tcf_block_put); 296 297 /* Main classifier routine: scans classifier chain attached 298 * to this qdisc, (optionally) tests for protocol and asks 299 * specific classifiers. 300 */ 301 int tcf_classify(struct sk_buff *skb, const struct tcf_proto *tp, 302 struct tcf_result *res, bool compat_mode) 303 { 304 __be16 protocol = tc_skb_protocol(skb); 305 #ifdef CONFIG_NET_CLS_ACT 306 const int max_reclassify_loop = 4; 307 const struct tcf_proto *orig_tp = tp; 308 const struct tcf_proto *first_tp; 309 int limit = 0; 310 311 reclassify: 312 #endif 313 for (; tp; tp = rcu_dereference_bh(tp->next)) { 314 int err; 315 316 if (tp->protocol != protocol && 317 tp->protocol != htons(ETH_P_ALL)) 318 continue; 319 320 err = tp->classify(skb, tp, res); 321 #ifdef CONFIG_NET_CLS_ACT 322 if (unlikely(err == TC_ACT_RECLASSIFY && !compat_mode)) { 323 first_tp = orig_tp; 324 goto reset; 325 } else if (unlikely(TC_ACT_EXT_CMP(err, TC_ACT_GOTO_CHAIN))) { 326 first_tp = res->goto_tp; 327 goto reset; 328 } 329 #endif 330 if (err >= 0) 331 return err; 332 } 333 334 return TC_ACT_UNSPEC; /* signal: continue lookup */ 335 #ifdef CONFIG_NET_CLS_ACT 336 reset: 337 if (unlikely(limit++ >= max_reclassify_loop)) { 338 net_notice_ratelimited("%s: reclassify loop, rule prio %u, protocol %02x\n", 339 tp->q->ops->id, tp->prio & 0xffff, 340 ntohs(tp->protocol)); 341 return TC_ACT_SHOT; 342 } 343 344 tp = first_tp; 345 protocol = tc_skb_protocol(skb); 346 goto reclassify; 347 #endif 348 } 349 EXPORT_SYMBOL(tcf_classify); 350 351 struct tcf_chain_info { 352 struct tcf_proto __rcu **pprev; 353 struct tcf_proto __rcu *next; 354 }; 355 356 static struct tcf_proto *tcf_chain_tp_prev(struct tcf_chain_info *chain_info) 357 { 358 return rtnl_dereference(*chain_info->pprev); 359 } 360 361 static void tcf_chain_tp_insert(struct tcf_chain *chain, 362 struct tcf_chain_info *chain_info, 363 struct tcf_proto *tp) 364 { 365 if (chain->p_filter_chain && 366 *chain_info->pprev == chain->filter_chain) 367 rcu_assign_pointer(*chain->p_filter_chain, tp); 368 RCU_INIT_POINTER(tp->next, tcf_chain_tp_prev(chain_info)); 369 rcu_assign_pointer(*chain_info->pprev, tp); 370 } 371 372 static void tcf_chain_tp_remove(struct tcf_chain *chain, 373 struct tcf_chain_info *chain_info, 374 struct tcf_proto *tp) 375 { 376 struct tcf_proto *next = rtnl_dereference(chain_info->next); 377 378 if (chain->p_filter_chain && tp == chain->filter_chain) 379 RCU_INIT_POINTER(*chain->p_filter_chain, next); 380 RCU_INIT_POINTER(*chain_info->pprev, next); 381 } 382 383 static struct tcf_proto *tcf_chain_tp_find(struct tcf_chain *chain, 384 struct tcf_chain_info *chain_info, 385 u32 protocol, u32 prio, 386 bool prio_allocate) 387 { 388 struct tcf_proto **pprev; 389 struct tcf_proto *tp; 390 391 /* Check the chain for existence of proto-tcf with this priority */ 392 for (pprev = &chain->filter_chain; 393 (tp = rtnl_dereference(*pprev)); pprev = &tp->next) { 394 if (tp->prio >= prio) { 395 if (tp->prio == prio) { 396 if (prio_allocate || 397 (tp->protocol != protocol && protocol)) 398 return ERR_PTR(-EINVAL); 399 } else { 400 tp = NULL; 401 } 402 break; 403 } 404 } 405 chain_info->pprev = pprev; 406 chain_info->next = tp ? tp->next : NULL; 407 return tp; 408 } 409 410 /* Add/change/delete/get a filter node */ 411 412 static int tc_ctl_tfilter(struct sk_buff *skb, struct nlmsghdr *n, 413 struct netlink_ext_ack *extack) 414 { 415 struct net *net = sock_net(skb->sk); 416 struct nlattr *tca[TCA_MAX + 1]; 417 struct tcmsg *t; 418 u32 protocol; 419 u32 prio; 420 bool prio_allocate; 421 u32 parent; 422 u32 chain_index; 423 struct net_device *dev; 424 struct Qdisc *q; 425 struct tcf_chain_info chain_info; 426 struct tcf_chain *chain = NULL; 427 struct tcf_block *block; 428 struct tcf_proto *tp; 429 const struct Qdisc_class_ops *cops; 430 unsigned long cl; 431 unsigned long fh; 432 int err; 433 int tp_created; 434 435 if ((n->nlmsg_type != RTM_GETTFILTER) && 436 !netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN)) 437 return -EPERM; 438 439 replay: 440 tp_created = 0; 441 442 err = nlmsg_parse(n, sizeof(*t), tca, TCA_MAX, NULL, extack); 443 if (err < 0) 444 return err; 445 446 t = nlmsg_data(n); 447 protocol = TC_H_MIN(t->tcm_info); 448 prio = TC_H_MAJ(t->tcm_info); 449 prio_allocate = false; 450 parent = t->tcm_parent; 451 cl = 0; 452 453 if (prio == 0) { 454 switch (n->nlmsg_type) { 455 case RTM_DELTFILTER: 456 if (protocol || t->tcm_handle || tca[TCA_KIND]) 457 return -ENOENT; 458 break; 459 case RTM_NEWTFILTER: 460 /* If no priority is provided by the user, 461 * we allocate one. 462 */ 463 if (n->nlmsg_flags & NLM_F_CREATE) { 464 prio = TC_H_MAKE(0x80000000U, 0U); 465 prio_allocate = true; 466 break; 467 } 468 /* fall-through */ 469 default: 470 return -ENOENT; 471 } 472 } 473 474 /* Find head of filter chain. */ 475 476 /* Find link */ 477 dev = __dev_get_by_index(net, t->tcm_ifindex); 478 if (dev == NULL) 479 return -ENODEV; 480 481 /* Find qdisc */ 482 if (!parent) { 483 q = dev->qdisc; 484 parent = q->handle; 485 } else { 486 q = qdisc_lookup(dev, TC_H_MAJ(t->tcm_parent)); 487 if (q == NULL) 488 return -EINVAL; 489 } 490 491 /* Is it classful? */ 492 cops = q->ops->cl_ops; 493 if (!cops) 494 return -EINVAL; 495 496 if (!cops->tcf_block) 497 return -EOPNOTSUPP; 498 499 /* Do we search for filter, attached to class? */ 500 if (TC_H_MIN(parent)) { 501 cl = cops->get(q, parent); 502 if (cl == 0) 503 return -ENOENT; 504 } 505 506 /* And the last stroke */ 507 block = cops->tcf_block(q, cl); 508 if (!block) { 509 err = -EINVAL; 510 goto errout; 511 } 512 513 chain_index = tca[TCA_CHAIN] ? nla_get_u32(tca[TCA_CHAIN]) : 0; 514 if (chain_index > TC_ACT_EXT_VAL_MASK) { 515 err = -EINVAL; 516 goto errout; 517 } 518 chain = tcf_chain_get(block, chain_index, 519 n->nlmsg_type == RTM_NEWTFILTER); 520 if (!chain) { 521 err = n->nlmsg_type == RTM_NEWTFILTER ? -ENOMEM : -EINVAL; 522 goto errout; 523 } 524 525 if (n->nlmsg_type == RTM_DELTFILTER && prio == 0) { 526 tfilter_notify_chain(net, skb, n, chain, RTM_DELTFILTER); 527 tcf_chain_flush(chain); 528 err = 0; 529 goto errout; 530 } 531 532 tp = tcf_chain_tp_find(chain, &chain_info, protocol, 533 prio, prio_allocate); 534 if (IS_ERR(tp)) { 535 err = PTR_ERR(tp); 536 goto errout; 537 } 538 539 if (tp == NULL) { 540 /* Proto-tcf does not exist, create new one */ 541 542 if (tca[TCA_KIND] == NULL || !protocol) { 543 err = -EINVAL; 544 goto errout; 545 } 546 547 if (n->nlmsg_type != RTM_NEWTFILTER || 548 !(n->nlmsg_flags & NLM_F_CREATE)) { 549 err = -ENOENT; 550 goto errout; 551 } 552 553 if (prio_allocate) 554 prio = tcf_auto_prio(tcf_chain_tp_prev(&chain_info)); 555 556 tp = tcf_proto_create(nla_data(tca[TCA_KIND]), 557 protocol, prio, parent, q, chain); 558 if (IS_ERR(tp)) { 559 err = PTR_ERR(tp); 560 goto errout; 561 } 562 tp_created = 1; 563 } else if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind)) { 564 err = -EINVAL; 565 goto errout; 566 } 567 568 fh = tp->ops->get(tp, t->tcm_handle); 569 570 if (fh == 0) { 571 if (n->nlmsg_type == RTM_DELTFILTER && t->tcm_handle == 0) { 572 tcf_chain_tp_remove(chain, &chain_info, tp); 573 tfilter_notify(net, skb, n, tp, fh, 574 RTM_DELTFILTER, false); 575 tcf_proto_destroy(tp); 576 err = 0; 577 goto errout; 578 } 579 580 if (n->nlmsg_type != RTM_NEWTFILTER || 581 !(n->nlmsg_flags & NLM_F_CREATE)) { 582 err = -ENOENT; 583 goto errout; 584 } 585 } else { 586 bool last; 587 588 switch (n->nlmsg_type) { 589 case RTM_NEWTFILTER: 590 if (n->nlmsg_flags & NLM_F_EXCL) { 591 if (tp_created) 592 tcf_proto_destroy(tp); 593 err = -EEXIST; 594 goto errout; 595 } 596 break; 597 case RTM_DELTFILTER: 598 err = tp->ops->delete(tp, fh, &last); 599 if (err) 600 goto errout; 601 tfilter_notify(net, skb, n, tp, t->tcm_handle, 602 RTM_DELTFILTER, false); 603 if (last) { 604 tcf_chain_tp_remove(chain, &chain_info, tp); 605 tcf_proto_destroy(tp); 606 } 607 goto errout; 608 case RTM_GETTFILTER: 609 err = tfilter_notify(net, skb, n, tp, fh, 610 RTM_NEWTFILTER, true); 611 goto errout; 612 default: 613 err = -EINVAL; 614 goto errout; 615 } 616 } 617 618 err = tp->ops->change(net, skb, tp, cl, t->tcm_handle, tca, &fh, 619 n->nlmsg_flags & NLM_F_CREATE ? TCA_ACT_NOREPLACE : TCA_ACT_REPLACE); 620 if (err == 0) { 621 if (tp_created) 622 tcf_chain_tp_insert(chain, &chain_info, tp); 623 tfilter_notify(net, skb, n, tp, fh, RTM_NEWTFILTER, false); 624 } else { 625 if (tp_created) 626 tcf_proto_destroy(tp); 627 } 628 629 errout: 630 if (chain) 631 tcf_chain_put(chain); 632 if (cl) 633 cops->put(q, cl); 634 if (err == -EAGAIN) 635 /* Replay the request. */ 636 goto replay; 637 return err; 638 } 639 640 static int tcf_fill_node(struct net *net, struct sk_buff *skb, 641 struct tcf_proto *tp, unsigned long fh, u32 portid, 642 u32 seq, u16 flags, int event) 643 { 644 struct tcmsg *tcm; 645 struct nlmsghdr *nlh; 646 unsigned char *b = skb_tail_pointer(skb); 647 648 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags); 649 if (!nlh) 650 goto out_nlmsg_trim; 651 tcm = nlmsg_data(nlh); 652 tcm->tcm_family = AF_UNSPEC; 653 tcm->tcm__pad1 = 0; 654 tcm->tcm__pad2 = 0; 655 tcm->tcm_ifindex = qdisc_dev(tp->q)->ifindex; 656 tcm->tcm_parent = tp->classid; 657 tcm->tcm_info = TC_H_MAKE(tp->prio, tp->protocol); 658 if (nla_put_string(skb, TCA_KIND, tp->ops->kind)) 659 goto nla_put_failure; 660 if (nla_put_u32(skb, TCA_CHAIN, tp->chain->index)) 661 goto nla_put_failure; 662 tcm->tcm_handle = fh; 663 if (RTM_DELTFILTER != event) { 664 tcm->tcm_handle = 0; 665 if (tp->ops->dump && tp->ops->dump(net, tp, fh, skb, tcm) < 0) 666 goto nla_put_failure; 667 } 668 nlh->nlmsg_len = skb_tail_pointer(skb) - b; 669 return skb->len; 670 671 out_nlmsg_trim: 672 nla_put_failure: 673 nlmsg_trim(skb, b); 674 return -1; 675 } 676 677 static int tfilter_notify(struct net *net, struct sk_buff *oskb, 678 struct nlmsghdr *n, struct tcf_proto *tp, 679 unsigned long fh, int event, bool unicast) 680 { 681 struct sk_buff *skb; 682 u32 portid = oskb ? NETLINK_CB(oskb).portid : 0; 683 684 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL); 685 if (!skb) 686 return -ENOBUFS; 687 688 if (tcf_fill_node(net, skb, tp, fh, portid, n->nlmsg_seq, 689 n->nlmsg_flags, event) <= 0) { 690 kfree_skb(skb); 691 return -EINVAL; 692 } 693 694 if (unicast) 695 return netlink_unicast(net->rtnl, skb, portid, MSG_DONTWAIT); 696 697 return rtnetlink_send(skb, net, portid, RTNLGRP_TC, 698 n->nlmsg_flags & NLM_F_ECHO); 699 } 700 701 struct tcf_dump_args { 702 struct tcf_walker w; 703 struct sk_buff *skb; 704 struct netlink_callback *cb; 705 }; 706 707 static int tcf_node_dump(struct tcf_proto *tp, unsigned long n, 708 struct tcf_walker *arg) 709 { 710 struct tcf_dump_args *a = (void *)arg; 711 struct net *net = sock_net(a->skb->sk); 712 713 return tcf_fill_node(net, a->skb, tp, n, NETLINK_CB(a->cb->skb).portid, 714 a->cb->nlh->nlmsg_seq, NLM_F_MULTI, 715 RTM_NEWTFILTER); 716 } 717 718 static bool tcf_chain_dump(struct tcf_chain *chain, struct sk_buff *skb, 719 struct netlink_callback *cb, 720 long index_start, long *p_index) 721 { 722 struct net *net = sock_net(skb->sk); 723 struct tcmsg *tcm = nlmsg_data(cb->nlh); 724 struct tcf_dump_args arg; 725 struct tcf_proto *tp; 726 727 for (tp = rtnl_dereference(chain->filter_chain); 728 tp; tp = rtnl_dereference(tp->next), (*p_index)++) { 729 if (*p_index < index_start) 730 continue; 731 if (TC_H_MAJ(tcm->tcm_info) && 732 TC_H_MAJ(tcm->tcm_info) != tp->prio) 733 continue; 734 if (TC_H_MIN(tcm->tcm_info) && 735 TC_H_MIN(tcm->tcm_info) != tp->protocol) 736 continue; 737 if (*p_index > index_start) 738 memset(&cb->args[1], 0, 739 sizeof(cb->args) - sizeof(cb->args[0])); 740 if (cb->args[1] == 0) { 741 if (tcf_fill_node(net, skb, tp, 0, 742 NETLINK_CB(cb->skb).portid, 743 cb->nlh->nlmsg_seq, NLM_F_MULTI, 744 RTM_NEWTFILTER) <= 0) 745 return false; 746 747 cb->args[1] = 1; 748 } 749 if (!tp->ops->walk) 750 continue; 751 arg.w.fn = tcf_node_dump; 752 arg.skb = skb; 753 arg.cb = cb; 754 arg.w.stop = 0; 755 arg.w.skip = cb->args[1] - 1; 756 arg.w.count = 0; 757 tp->ops->walk(tp, &arg.w); 758 cb->args[1] = arg.w.count + 1; 759 if (arg.w.stop) 760 return false; 761 } 762 return true; 763 } 764 765 /* called with RTNL */ 766 static int tc_dump_tfilter(struct sk_buff *skb, struct netlink_callback *cb) 767 { 768 struct net *net = sock_net(skb->sk); 769 struct nlattr *tca[TCA_MAX + 1]; 770 struct net_device *dev; 771 struct Qdisc *q; 772 struct tcf_block *block; 773 struct tcf_chain *chain; 774 struct tcmsg *tcm = nlmsg_data(cb->nlh); 775 unsigned long cl = 0; 776 const struct Qdisc_class_ops *cops; 777 long index_start; 778 long index; 779 int err; 780 781 if (nlmsg_len(cb->nlh) < sizeof(*tcm)) 782 return skb->len; 783 784 err = nlmsg_parse(cb->nlh, sizeof(*tcm), tca, TCA_MAX, NULL, NULL); 785 if (err) 786 return err; 787 788 dev = __dev_get_by_index(net, tcm->tcm_ifindex); 789 if (!dev) 790 return skb->len; 791 792 if (!tcm->tcm_parent) 793 q = dev->qdisc; 794 else 795 q = qdisc_lookup(dev, TC_H_MAJ(tcm->tcm_parent)); 796 if (!q) 797 goto out; 798 cops = q->ops->cl_ops; 799 if (!cops) 800 goto errout; 801 if (!cops->tcf_block) 802 goto errout; 803 if (TC_H_MIN(tcm->tcm_parent)) { 804 cl = cops->get(q, tcm->tcm_parent); 805 if (cl == 0) 806 goto errout; 807 } 808 block = cops->tcf_block(q, cl); 809 if (!block) 810 goto errout; 811 812 index_start = cb->args[0]; 813 index = 0; 814 815 list_for_each_entry(chain, &block->chain_list, list) { 816 if (tca[TCA_CHAIN] && 817 nla_get_u32(tca[TCA_CHAIN]) != chain->index) 818 continue; 819 if (!tcf_chain_dump(chain, skb, cb, index_start, &index)) 820 break; 821 } 822 823 cb->args[0] = index; 824 825 errout: 826 if (cl) 827 cops->put(q, cl); 828 out: 829 return skb->len; 830 } 831 832 void tcf_exts_destroy(struct tcf_exts *exts) 833 { 834 #ifdef CONFIG_NET_CLS_ACT 835 LIST_HEAD(actions); 836 837 tcf_exts_to_list(exts, &actions); 838 tcf_action_destroy(&actions, TCA_ACT_UNBIND); 839 kfree(exts->actions); 840 exts->nr_actions = 0; 841 #endif 842 } 843 EXPORT_SYMBOL(tcf_exts_destroy); 844 845 int tcf_exts_validate(struct net *net, struct tcf_proto *tp, struct nlattr **tb, 846 struct nlattr *rate_tlv, struct tcf_exts *exts, bool ovr) 847 { 848 #ifdef CONFIG_NET_CLS_ACT 849 { 850 struct tc_action *act; 851 852 if (exts->police && tb[exts->police]) { 853 act = tcf_action_init_1(net, tp, tb[exts->police], 854 rate_tlv, "police", ovr, 855 TCA_ACT_BIND); 856 if (IS_ERR(act)) 857 return PTR_ERR(act); 858 859 act->type = exts->type = TCA_OLD_COMPAT; 860 exts->actions[0] = act; 861 exts->nr_actions = 1; 862 } else if (exts->action && tb[exts->action]) { 863 LIST_HEAD(actions); 864 int err, i = 0; 865 866 err = tcf_action_init(net, tp, tb[exts->action], 867 rate_tlv, NULL, ovr, TCA_ACT_BIND, 868 &actions); 869 if (err) 870 return err; 871 list_for_each_entry(act, &actions, list) 872 exts->actions[i++] = act; 873 exts->nr_actions = i; 874 } 875 } 876 #else 877 if ((exts->action && tb[exts->action]) || 878 (exts->police && tb[exts->police])) 879 return -EOPNOTSUPP; 880 #endif 881 882 return 0; 883 } 884 EXPORT_SYMBOL(tcf_exts_validate); 885 886 void tcf_exts_change(struct tcf_proto *tp, struct tcf_exts *dst, 887 struct tcf_exts *src) 888 { 889 #ifdef CONFIG_NET_CLS_ACT 890 struct tcf_exts old = *dst; 891 892 tcf_tree_lock(tp); 893 dst->nr_actions = src->nr_actions; 894 dst->actions = src->actions; 895 dst->type = src->type; 896 tcf_tree_unlock(tp); 897 898 tcf_exts_destroy(&old); 899 #endif 900 } 901 EXPORT_SYMBOL(tcf_exts_change); 902 903 #ifdef CONFIG_NET_CLS_ACT 904 static struct tc_action *tcf_exts_first_act(struct tcf_exts *exts) 905 { 906 if (exts->nr_actions == 0) 907 return NULL; 908 else 909 return exts->actions[0]; 910 } 911 #endif 912 913 int tcf_exts_dump(struct sk_buff *skb, struct tcf_exts *exts) 914 { 915 #ifdef CONFIG_NET_CLS_ACT 916 struct nlattr *nest; 917 918 if (exts->action && exts->nr_actions) { 919 /* 920 * again for backward compatible mode - we want 921 * to work with both old and new modes of entering 922 * tc data even if iproute2 was newer - jhs 923 */ 924 if (exts->type != TCA_OLD_COMPAT) { 925 LIST_HEAD(actions); 926 927 nest = nla_nest_start(skb, exts->action); 928 if (nest == NULL) 929 goto nla_put_failure; 930 931 tcf_exts_to_list(exts, &actions); 932 if (tcf_action_dump(skb, &actions, 0, 0) < 0) 933 goto nla_put_failure; 934 nla_nest_end(skb, nest); 935 } else if (exts->police) { 936 struct tc_action *act = tcf_exts_first_act(exts); 937 nest = nla_nest_start(skb, exts->police); 938 if (nest == NULL || !act) 939 goto nla_put_failure; 940 if (tcf_action_dump_old(skb, act, 0, 0) < 0) 941 goto nla_put_failure; 942 nla_nest_end(skb, nest); 943 } 944 } 945 return 0; 946 947 nla_put_failure: 948 nla_nest_cancel(skb, nest); 949 return -1; 950 #else 951 return 0; 952 #endif 953 } 954 EXPORT_SYMBOL(tcf_exts_dump); 955 956 957 int tcf_exts_dump_stats(struct sk_buff *skb, struct tcf_exts *exts) 958 { 959 #ifdef CONFIG_NET_CLS_ACT 960 struct tc_action *a = tcf_exts_first_act(exts); 961 if (a != NULL && tcf_action_copy_stats(skb, a, 1) < 0) 962 return -1; 963 #endif 964 return 0; 965 } 966 EXPORT_SYMBOL(tcf_exts_dump_stats); 967 968 int tcf_exts_get_dev(struct net_device *dev, struct tcf_exts *exts, 969 struct net_device **hw_dev) 970 { 971 #ifdef CONFIG_NET_CLS_ACT 972 const struct tc_action *a; 973 LIST_HEAD(actions); 974 975 if (tc_no_actions(exts)) 976 return -EINVAL; 977 978 tcf_exts_to_list(exts, &actions); 979 list_for_each_entry(a, &actions, list) { 980 if (a->ops->get_dev) { 981 a->ops->get_dev(a, dev_net(dev), hw_dev); 982 break; 983 } 984 } 985 if (*hw_dev) 986 return 0; 987 #endif 988 return -EOPNOTSUPP; 989 } 990 EXPORT_SYMBOL(tcf_exts_get_dev); 991 992 static int __init tc_filter_init(void) 993 { 994 rtnl_register(PF_UNSPEC, RTM_NEWTFILTER, tc_ctl_tfilter, NULL, NULL); 995 rtnl_register(PF_UNSPEC, RTM_DELTFILTER, tc_ctl_tfilter, NULL, NULL); 996 rtnl_register(PF_UNSPEC, RTM_GETTFILTER, tc_ctl_tfilter, 997 tc_dump_tfilter, NULL); 998 999 return 0; 1000 } 1001 1002 subsys_initcall(tc_filter_init); 1003