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 static struct workqueue_struct *tc_filter_wq; 81 82 int unregister_tcf_proto_ops(struct tcf_proto_ops *ops) 83 { 84 struct tcf_proto_ops *t; 85 int rc = -ENOENT; 86 87 /* Wait for outstanding call_rcu()s, if any, from a 88 * tcf_proto_ops's destroy() handler. 89 */ 90 rcu_barrier(); 91 flush_workqueue(tc_filter_wq); 92 93 write_lock(&cls_mod_lock); 94 list_for_each_entry(t, &tcf_proto_base, head) { 95 if (t == ops) { 96 list_del(&t->head); 97 rc = 0; 98 break; 99 } 100 } 101 write_unlock(&cls_mod_lock); 102 return rc; 103 } 104 EXPORT_SYMBOL(unregister_tcf_proto_ops); 105 106 bool tcf_queue_work(struct work_struct *work) 107 { 108 return queue_work(tc_filter_wq, work); 109 } 110 EXPORT_SYMBOL(tcf_queue_work); 111 112 /* Select new prio value from the range, managed by kernel. */ 113 114 static inline u32 tcf_auto_prio(struct tcf_proto *tp) 115 { 116 u32 first = TC_H_MAKE(0xC0000000U, 0U); 117 118 if (tp) 119 first = tp->prio - 1; 120 121 return TC_H_MAJ(first); 122 } 123 124 static struct tcf_proto *tcf_proto_create(const char *kind, u32 protocol, 125 u32 prio, u32 parent, struct Qdisc *q, 126 struct tcf_chain *chain) 127 { 128 struct tcf_proto *tp; 129 int err; 130 131 tp = kzalloc(sizeof(*tp), GFP_KERNEL); 132 if (!tp) 133 return ERR_PTR(-ENOBUFS); 134 135 err = -ENOENT; 136 tp->ops = tcf_proto_lookup_ops(kind); 137 if (!tp->ops) { 138 #ifdef CONFIG_MODULES 139 rtnl_unlock(); 140 request_module("cls_%s", kind); 141 rtnl_lock(); 142 tp->ops = tcf_proto_lookup_ops(kind); 143 /* We dropped the RTNL semaphore in order to perform 144 * the module load. So, even if we succeeded in loading 145 * the module we have to replay the request. We indicate 146 * this using -EAGAIN. 147 */ 148 if (tp->ops) { 149 module_put(tp->ops->owner); 150 err = -EAGAIN; 151 } else { 152 err = -ENOENT; 153 } 154 goto errout; 155 #endif 156 } 157 tp->classify = tp->ops->classify; 158 tp->protocol = protocol; 159 tp->prio = prio; 160 tp->classid = parent; 161 tp->q = q; 162 tp->chain = chain; 163 164 err = tp->ops->init(tp); 165 if (err) { 166 module_put(tp->ops->owner); 167 goto errout; 168 } 169 return tp; 170 171 errout: 172 kfree(tp); 173 return ERR_PTR(err); 174 } 175 176 static void tcf_proto_destroy(struct tcf_proto *tp) 177 { 178 tp->ops->destroy(tp); 179 module_put(tp->ops->owner); 180 kfree_rcu(tp, rcu); 181 } 182 183 static struct tcf_chain *tcf_chain_create(struct tcf_block *block, 184 u32 chain_index) 185 { 186 struct tcf_chain *chain; 187 188 chain = kzalloc(sizeof(*chain), GFP_KERNEL); 189 if (!chain) 190 return NULL; 191 list_add_tail(&chain->list, &block->chain_list); 192 chain->block = block; 193 chain->index = chain_index; 194 chain->refcnt = 1; 195 return chain; 196 } 197 198 static void tcf_chain_head_change(struct tcf_chain *chain, 199 struct tcf_proto *tp_head) 200 { 201 if (chain->chain_head_change) 202 chain->chain_head_change(tp_head, 203 chain->chain_head_change_priv); 204 } 205 206 static void tcf_chain_flush(struct tcf_chain *chain) 207 { 208 struct tcf_proto *tp = rtnl_dereference(chain->filter_chain); 209 210 tcf_chain_head_change(chain, NULL); 211 while (tp) { 212 RCU_INIT_POINTER(chain->filter_chain, tp->next); 213 tcf_proto_destroy(tp); 214 tp = rtnl_dereference(chain->filter_chain); 215 tcf_chain_put(chain); 216 } 217 } 218 219 static void tcf_chain_destroy(struct tcf_chain *chain) 220 { 221 list_del(&chain->list); 222 kfree(chain); 223 } 224 225 static void tcf_chain_hold(struct tcf_chain *chain) 226 { 227 ++chain->refcnt; 228 } 229 230 struct tcf_chain *tcf_chain_get(struct tcf_block *block, u32 chain_index, 231 bool create) 232 { 233 struct tcf_chain *chain; 234 235 list_for_each_entry(chain, &block->chain_list, list) { 236 if (chain->index == chain_index) { 237 tcf_chain_hold(chain); 238 return chain; 239 } 240 } 241 242 return create ? tcf_chain_create(block, chain_index) : NULL; 243 } 244 EXPORT_SYMBOL(tcf_chain_get); 245 246 void tcf_chain_put(struct tcf_chain *chain) 247 { 248 if (--chain->refcnt == 0) 249 tcf_chain_destroy(chain); 250 } 251 EXPORT_SYMBOL(tcf_chain_put); 252 253 static void tcf_block_offload_cmd(struct tcf_block *block, struct Qdisc *q, 254 struct tcf_block_ext_info *ei, 255 enum tc_block_command command) 256 { 257 struct net_device *dev = q->dev_queue->dev; 258 struct tc_block_offload bo = {}; 259 260 if (!dev->netdev_ops->ndo_setup_tc) 261 return; 262 bo.command = command; 263 bo.binder_type = ei->binder_type; 264 bo.block = block; 265 dev->netdev_ops->ndo_setup_tc(dev, TC_SETUP_BLOCK, &bo); 266 } 267 268 static void tcf_block_offload_bind(struct tcf_block *block, struct Qdisc *q, 269 struct tcf_block_ext_info *ei) 270 { 271 tcf_block_offload_cmd(block, q, ei, TC_BLOCK_BIND); 272 } 273 274 static void tcf_block_offload_unbind(struct tcf_block *block, struct Qdisc *q, 275 struct tcf_block_ext_info *ei) 276 { 277 tcf_block_offload_cmd(block, q, ei, TC_BLOCK_UNBIND); 278 } 279 280 int tcf_block_get_ext(struct tcf_block **p_block, struct Qdisc *q, 281 struct tcf_block_ext_info *ei) 282 { 283 struct tcf_block *block = kzalloc(sizeof(*block), GFP_KERNEL); 284 struct tcf_chain *chain; 285 int err; 286 287 if (!block) 288 return -ENOMEM; 289 INIT_LIST_HEAD(&block->chain_list); 290 INIT_LIST_HEAD(&block->cb_list); 291 292 /* Create chain 0 by default, it has to be always present. */ 293 chain = tcf_chain_create(block, 0); 294 if (!chain) { 295 err = -ENOMEM; 296 goto err_chain_create; 297 } 298 WARN_ON(!ei->chain_head_change); 299 chain->chain_head_change = ei->chain_head_change; 300 chain->chain_head_change_priv = ei->chain_head_change_priv; 301 block->net = qdisc_net(q); 302 block->q = q; 303 tcf_block_offload_bind(block, q, ei); 304 *p_block = block; 305 return 0; 306 307 err_chain_create: 308 kfree(block); 309 return err; 310 } 311 EXPORT_SYMBOL(tcf_block_get_ext); 312 313 static void tcf_chain_head_change_dflt(struct tcf_proto *tp_head, void *priv) 314 { 315 struct tcf_proto __rcu **p_filter_chain = priv; 316 317 rcu_assign_pointer(*p_filter_chain, tp_head); 318 } 319 320 int tcf_block_get(struct tcf_block **p_block, 321 struct tcf_proto __rcu **p_filter_chain, struct Qdisc *q) 322 { 323 struct tcf_block_ext_info ei = { 324 .chain_head_change = tcf_chain_head_change_dflt, 325 .chain_head_change_priv = p_filter_chain, 326 }; 327 328 WARN_ON(!p_filter_chain); 329 return tcf_block_get_ext(p_block, q, &ei); 330 } 331 EXPORT_SYMBOL(tcf_block_get); 332 333 static void tcf_block_put_final(struct work_struct *work) 334 { 335 struct tcf_block *block = container_of(work, struct tcf_block, work); 336 struct tcf_chain *chain, *tmp; 337 338 rtnl_lock(); 339 /* Only chain 0 should be still here. */ 340 list_for_each_entry_safe(chain, tmp, &block->chain_list, list) 341 tcf_chain_put(chain); 342 rtnl_unlock(); 343 kfree(block); 344 } 345 346 /* XXX: Standalone actions are not allowed to jump to any chain, and bound 347 * actions should be all removed after flushing. However, filters are now 348 * destroyed in tc filter workqueue with RTNL lock, they can not race here. 349 */ 350 void tcf_block_put_ext(struct tcf_block *block, struct Qdisc *q, 351 struct tcf_block_ext_info *ei) 352 { 353 struct tcf_chain *chain, *tmp; 354 355 list_for_each_entry_safe(chain, tmp, &block->chain_list, list) 356 tcf_chain_flush(chain); 357 358 tcf_block_offload_unbind(block, q, ei); 359 360 INIT_WORK(&block->work, tcf_block_put_final); 361 /* Wait for existing RCU callbacks to cool down, make sure their works 362 * have been queued before this. We can not flush pending works here 363 * because we are holding the RTNL lock. 364 */ 365 rcu_barrier(); 366 tcf_queue_work(&block->work); 367 } 368 EXPORT_SYMBOL(tcf_block_put_ext); 369 370 void tcf_block_put(struct tcf_block *block) 371 { 372 struct tcf_block_ext_info ei = {0, }; 373 374 if (!block) 375 return; 376 tcf_block_put_ext(block, block->q, &ei); 377 } 378 379 EXPORT_SYMBOL(tcf_block_put); 380 381 struct tcf_block_cb { 382 struct list_head list; 383 tc_setup_cb_t *cb; 384 void *cb_ident; 385 void *cb_priv; 386 unsigned int refcnt; 387 }; 388 389 void *tcf_block_cb_priv(struct tcf_block_cb *block_cb) 390 { 391 return block_cb->cb_priv; 392 } 393 EXPORT_SYMBOL(tcf_block_cb_priv); 394 395 struct tcf_block_cb *tcf_block_cb_lookup(struct tcf_block *block, 396 tc_setup_cb_t *cb, void *cb_ident) 397 { struct tcf_block_cb *block_cb; 398 399 list_for_each_entry(block_cb, &block->cb_list, list) 400 if (block_cb->cb == cb && block_cb->cb_ident == cb_ident) 401 return block_cb; 402 return NULL; 403 } 404 EXPORT_SYMBOL(tcf_block_cb_lookup); 405 406 void tcf_block_cb_incref(struct tcf_block_cb *block_cb) 407 { 408 block_cb->refcnt++; 409 } 410 EXPORT_SYMBOL(tcf_block_cb_incref); 411 412 unsigned int tcf_block_cb_decref(struct tcf_block_cb *block_cb) 413 { 414 return --block_cb->refcnt; 415 } 416 EXPORT_SYMBOL(tcf_block_cb_decref); 417 418 struct tcf_block_cb *__tcf_block_cb_register(struct tcf_block *block, 419 tc_setup_cb_t *cb, void *cb_ident, 420 void *cb_priv) 421 { 422 struct tcf_block_cb *block_cb; 423 424 block_cb = kzalloc(sizeof(*block_cb), GFP_KERNEL); 425 if (!block_cb) 426 return NULL; 427 block_cb->cb = cb; 428 block_cb->cb_ident = cb_ident; 429 block_cb->cb_priv = cb_priv; 430 list_add(&block_cb->list, &block->cb_list); 431 return block_cb; 432 } 433 EXPORT_SYMBOL(__tcf_block_cb_register); 434 435 int tcf_block_cb_register(struct tcf_block *block, 436 tc_setup_cb_t *cb, void *cb_ident, 437 void *cb_priv) 438 { 439 struct tcf_block_cb *block_cb; 440 441 block_cb = __tcf_block_cb_register(block, cb, cb_ident, cb_priv); 442 return block_cb ? 0 : -ENOMEM; 443 } 444 EXPORT_SYMBOL(tcf_block_cb_register); 445 446 void __tcf_block_cb_unregister(struct tcf_block_cb *block_cb) 447 { 448 list_del(&block_cb->list); 449 kfree(block_cb); 450 } 451 EXPORT_SYMBOL(__tcf_block_cb_unregister); 452 453 void tcf_block_cb_unregister(struct tcf_block *block, 454 tc_setup_cb_t *cb, void *cb_ident) 455 { 456 struct tcf_block_cb *block_cb; 457 458 block_cb = tcf_block_cb_lookup(block, cb, cb_ident); 459 if (!block_cb) 460 return; 461 __tcf_block_cb_unregister(block_cb); 462 } 463 EXPORT_SYMBOL(tcf_block_cb_unregister); 464 465 static int tcf_block_cb_call(struct tcf_block *block, enum tc_setup_type type, 466 void *type_data, bool err_stop) 467 { 468 struct tcf_block_cb *block_cb; 469 int ok_count = 0; 470 int err; 471 472 list_for_each_entry(block_cb, &block->cb_list, list) { 473 err = block_cb->cb(type, type_data, block_cb->cb_priv); 474 if (err) { 475 if (err_stop) 476 return err; 477 } else { 478 ok_count++; 479 } 480 } 481 return ok_count; 482 } 483 484 /* Main classifier routine: scans classifier chain attached 485 * to this qdisc, (optionally) tests for protocol and asks 486 * specific classifiers. 487 */ 488 int tcf_classify(struct sk_buff *skb, const struct tcf_proto *tp, 489 struct tcf_result *res, bool compat_mode) 490 { 491 __be16 protocol = tc_skb_protocol(skb); 492 #ifdef CONFIG_NET_CLS_ACT 493 const int max_reclassify_loop = 4; 494 const struct tcf_proto *orig_tp = tp; 495 const struct tcf_proto *first_tp; 496 int limit = 0; 497 498 reclassify: 499 #endif 500 for (; tp; tp = rcu_dereference_bh(tp->next)) { 501 int err; 502 503 if (tp->protocol != protocol && 504 tp->protocol != htons(ETH_P_ALL)) 505 continue; 506 507 err = tp->classify(skb, tp, res); 508 #ifdef CONFIG_NET_CLS_ACT 509 if (unlikely(err == TC_ACT_RECLASSIFY && !compat_mode)) { 510 first_tp = orig_tp; 511 goto reset; 512 } else if (unlikely(TC_ACT_EXT_CMP(err, TC_ACT_GOTO_CHAIN))) { 513 first_tp = res->goto_tp; 514 goto reset; 515 } 516 #endif 517 if (err >= 0) 518 return err; 519 } 520 521 return TC_ACT_UNSPEC; /* signal: continue lookup */ 522 #ifdef CONFIG_NET_CLS_ACT 523 reset: 524 if (unlikely(limit++ >= max_reclassify_loop)) { 525 net_notice_ratelimited("%s: reclassify loop, rule prio %u, protocol %02x\n", 526 tp->q->ops->id, tp->prio & 0xffff, 527 ntohs(tp->protocol)); 528 return TC_ACT_SHOT; 529 } 530 531 tp = first_tp; 532 protocol = tc_skb_protocol(skb); 533 goto reclassify; 534 #endif 535 } 536 EXPORT_SYMBOL(tcf_classify); 537 538 struct tcf_chain_info { 539 struct tcf_proto __rcu **pprev; 540 struct tcf_proto __rcu *next; 541 }; 542 543 static struct tcf_proto *tcf_chain_tp_prev(struct tcf_chain_info *chain_info) 544 { 545 return rtnl_dereference(*chain_info->pprev); 546 } 547 548 static void tcf_chain_tp_insert(struct tcf_chain *chain, 549 struct tcf_chain_info *chain_info, 550 struct tcf_proto *tp) 551 { 552 if (*chain_info->pprev == chain->filter_chain) 553 tcf_chain_head_change(chain, tp); 554 RCU_INIT_POINTER(tp->next, tcf_chain_tp_prev(chain_info)); 555 rcu_assign_pointer(*chain_info->pprev, tp); 556 tcf_chain_hold(chain); 557 } 558 559 static void tcf_chain_tp_remove(struct tcf_chain *chain, 560 struct tcf_chain_info *chain_info, 561 struct tcf_proto *tp) 562 { 563 struct tcf_proto *next = rtnl_dereference(chain_info->next); 564 565 if (tp == chain->filter_chain) 566 tcf_chain_head_change(chain, next); 567 RCU_INIT_POINTER(*chain_info->pprev, next); 568 tcf_chain_put(chain); 569 } 570 571 static struct tcf_proto *tcf_chain_tp_find(struct tcf_chain *chain, 572 struct tcf_chain_info *chain_info, 573 u32 protocol, u32 prio, 574 bool prio_allocate) 575 { 576 struct tcf_proto **pprev; 577 struct tcf_proto *tp; 578 579 /* Check the chain for existence of proto-tcf with this priority */ 580 for (pprev = &chain->filter_chain; 581 (tp = rtnl_dereference(*pprev)); pprev = &tp->next) { 582 if (tp->prio >= prio) { 583 if (tp->prio == prio) { 584 if (prio_allocate || 585 (tp->protocol != protocol && protocol)) 586 return ERR_PTR(-EINVAL); 587 } else { 588 tp = NULL; 589 } 590 break; 591 } 592 } 593 chain_info->pprev = pprev; 594 chain_info->next = tp ? tp->next : NULL; 595 return tp; 596 } 597 598 static int tcf_fill_node(struct net *net, struct sk_buff *skb, 599 struct tcf_proto *tp, struct Qdisc *q, u32 parent, 600 void *fh, u32 portid, u32 seq, u16 flags, int event) 601 { 602 struct tcmsg *tcm; 603 struct nlmsghdr *nlh; 604 unsigned char *b = skb_tail_pointer(skb); 605 606 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags); 607 if (!nlh) 608 goto out_nlmsg_trim; 609 tcm = nlmsg_data(nlh); 610 tcm->tcm_family = AF_UNSPEC; 611 tcm->tcm__pad1 = 0; 612 tcm->tcm__pad2 = 0; 613 tcm->tcm_ifindex = qdisc_dev(q)->ifindex; 614 tcm->tcm_parent = parent; 615 tcm->tcm_info = TC_H_MAKE(tp->prio, tp->protocol); 616 if (nla_put_string(skb, TCA_KIND, tp->ops->kind)) 617 goto nla_put_failure; 618 if (nla_put_u32(skb, TCA_CHAIN, tp->chain->index)) 619 goto nla_put_failure; 620 if (!fh) { 621 tcm->tcm_handle = 0; 622 } else { 623 if (tp->ops->dump && tp->ops->dump(net, tp, fh, skb, tcm) < 0) 624 goto nla_put_failure; 625 } 626 nlh->nlmsg_len = skb_tail_pointer(skb) - b; 627 return skb->len; 628 629 out_nlmsg_trim: 630 nla_put_failure: 631 nlmsg_trim(skb, b); 632 return -1; 633 } 634 635 static int tfilter_notify(struct net *net, struct sk_buff *oskb, 636 struct nlmsghdr *n, struct tcf_proto *tp, 637 struct Qdisc *q, u32 parent, 638 void *fh, int event, bool unicast) 639 { 640 struct sk_buff *skb; 641 u32 portid = oskb ? NETLINK_CB(oskb).portid : 0; 642 643 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL); 644 if (!skb) 645 return -ENOBUFS; 646 647 if (tcf_fill_node(net, skb, tp, q, parent, fh, portid, n->nlmsg_seq, 648 n->nlmsg_flags, event) <= 0) { 649 kfree_skb(skb); 650 return -EINVAL; 651 } 652 653 if (unicast) 654 return netlink_unicast(net->rtnl, skb, portid, MSG_DONTWAIT); 655 656 return rtnetlink_send(skb, net, portid, RTNLGRP_TC, 657 n->nlmsg_flags & NLM_F_ECHO); 658 } 659 660 static int tfilter_del_notify(struct net *net, struct sk_buff *oskb, 661 struct nlmsghdr *n, struct tcf_proto *tp, 662 struct Qdisc *q, u32 parent, 663 void *fh, bool unicast, bool *last) 664 { 665 struct sk_buff *skb; 666 u32 portid = oskb ? NETLINK_CB(oskb).portid : 0; 667 int err; 668 669 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL); 670 if (!skb) 671 return -ENOBUFS; 672 673 if (tcf_fill_node(net, skb, tp, q, parent, fh, portid, n->nlmsg_seq, 674 n->nlmsg_flags, RTM_DELTFILTER) <= 0) { 675 kfree_skb(skb); 676 return -EINVAL; 677 } 678 679 err = tp->ops->delete(tp, fh, last); 680 if (err) { 681 kfree_skb(skb); 682 return err; 683 } 684 685 if (unicast) 686 return netlink_unicast(net->rtnl, skb, portid, MSG_DONTWAIT); 687 688 return rtnetlink_send(skb, net, portid, RTNLGRP_TC, 689 n->nlmsg_flags & NLM_F_ECHO); 690 } 691 692 static void tfilter_notify_chain(struct net *net, struct sk_buff *oskb, 693 struct Qdisc *q, u32 parent, 694 struct nlmsghdr *n, 695 struct tcf_chain *chain, int event) 696 { 697 struct tcf_proto *tp; 698 699 for (tp = rtnl_dereference(chain->filter_chain); 700 tp; tp = rtnl_dereference(tp->next)) 701 tfilter_notify(net, oskb, n, tp, q, parent, 0, event, false); 702 } 703 704 /* Add/change/delete/get a filter node */ 705 706 static int tc_ctl_tfilter(struct sk_buff *skb, struct nlmsghdr *n, 707 struct netlink_ext_ack *extack) 708 { 709 struct net *net = sock_net(skb->sk); 710 struct nlattr *tca[TCA_MAX + 1]; 711 struct tcmsg *t; 712 u32 protocol; 713 u32 prio; 714 bool prio_allocate; 715 u32 parent; 716 u32 chain_index; 717 struct net_device *dev; 718 struct Qdisc *q; 719 struct tcf_chain_info chain_info; 720 struct tcf_chain *chain = NULL; 721 struct tcf_block *block; 722 struct tcf_proto *tp; 723 const struct Qdisc_class_ops *cops; 724 unsigned long cl; 725 void *fh; 726 int err; 727 int tp_created; 728 729 if ((n->nlmsg_type != RTM_GETTFILTER) && 730 !netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN)) 731 return -EPERM; 732 733 replay: 734 tp_created = 0; 735 736 err = nlmsg_parse(n, sizeof(*t), tca, TCA_MAX, NULL, extack); 737 if (err < 0) 738 return err; 739 740 t = nlmsg_data(n); 741 protocol = TC_H_MIN(t->tcm_info); 742 prio = TC_H_MAJ(t->tcm_info); 743 prio_allocate = false; 744 parent = t->tcm_parent; 745 cl = 0; 746 747 if (prio == 0) { 748 switch (n->nlmsg_type) { 749 case RTM_DELTFILTER: 750 if (protocol || t->tcm_handle || tca[TCA_KIND]) 751 return -ENOENT; 752 break; 753 case RTM_NEWTFILTER: 754 /* If no priority is provided by the user, 755 * we allocate one. 756 */ 757 if (n->nlmsg_flags & NLM_F_CREATE) { 758 prio = TC_H_MAKE(0x80000000U, 0U); 759 prio_allocate = true; 760 break; 761 } 762 /* fall-through */ 763 default: 764 return -ENOENT; 765 } 766 } 767 768 /* Find head of filter chain. */ 769 770 /* Find link */ 771 dev = __dev_get_by_index(net, t->tcm_ifindex); 772 if (dev == NULL) 773 return -ENODEV; 774 775 /* Find qdisc */ 776 if (!parent) { 777 q = dev->qdisc; 778 parent = q->handle; 779 } else { 780 q = qdisc_lookup(dev, TC_H_MAJ(t->tcm_parent)); 781 if (q == NULL) 782 return -EINVAL; 783 } 784 785 /* Is it classful? */ 786 cops = q->ops->cl_ops; 787 if (!cops) 788 return -EINVAL; 789 790 if (!cops->tcf_block) 791 return -EOPNOTSUPP; 792 793 /* Do we search for filter, attached to class? */ 794 if (TC_H_MIN(parent)) { 795 cl = cops->find(q, parent); 796 if (cl == 0) 797 return -ENOENT; 798 } 799 800 /* And the last stroke */ 801 block = cops->tcf_block(q, cl); 802 if (!block) { 803 err = -EINVAL; 804 goto errout; 805 } 806 807 chain_index = tca[TCA_CHAIN] ? nla_get_u32(tca[TCA_CHAIN]) : 0; 808 if (chain_index > TC_ACT_EXT_VAL_MASK) { 809 err = -EINVAL; 810 goto errout; 811 } 812 chain = tcf_chain_get(block, chain_index, 813 n->nlmsg_type == RTM_NEWTFILTER); 814 if (!chain) { 815 err = n->nlmsg_type == RTM_NEWTFILTER ? -ENOMEM : -EINVAL; 816 goto errout; 817 } 818 819 if (n->nlmsg_type == RTM_DELTFILTER && prio == 0) { 820 tfilter_notify_chain(net, skb, q, parent, n, 821 chain, RTM_DELTFILTER); 822 tcf_chain_flush(chain); 823 err = 0; 824 goto errout; 825 } 826 827 tp = tcf_chain_tp_find(chain, &chain_info, protocol, 828 prio, prio_allocate); 829 if (IS_ERR(tp)) { 830 err = PTR_ERR(tp); 831 goto errout; 832 } 833 834 if (tp == NULL) { 835 /* Proto-tcf does not exist, create new one */ 836 837 if (tca[TCA_KIND] == NULL || !protocol) { 838 err = -EINVAL; 839 goto errout; 840 } 841 842 if (n->nlmsg_type != RTM_NEWTFILTER || 843 !(n->nlmsg_flags & NLM_F_CREATE)) { 844 err = -ENOENT; 845 goto errout; 846 } 847 848 if (prio_allocate) 849 prio = tcf_auto_prio(tcf_chain_tp_prev(&chain_info)); 850 851 tp = tcf_proto_create(nla_data(tca[TCA_KIND]), 852 protocol, prio, parent, q, chain); 853 if (IS_ERR(tp)) { 854 err = PTR_ERR(tp); 855 goto errout; 856 } 857 tp_created = 1; 858 } else if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind)) { 859 err = -EINVAL; 860 goto errout; 861 } 862 863 fh = tp->ops->get(tp, t->tcm_handle); 864 865 if (!fh) { 866 if (n->nlmsg_type == RTM_DELTFILTER && t->tcm_handle == 0) { 867 tcf_chain_tp_remove(chain, &chain_info, tp); 868 tfilter_notify(net, skb, n, tp, q, parent, fh, 869 RTM_DELTFILTER, false); 870 tcf_proto_destroy(tp); 871 err = 0; 872 goto errout; 873 } 874 875 if (n->nlmsg_type != RTM_NEWTFILTER || 876 !(n->nlmsg_flags & NLM_F_CREATE)) { 877 err = -ENOENT; 878 goto errout; 879 } 880 } else { 881 bool last; 882 883 switch (n->nlmsg_type) { 884 case RTM_NEWTFILTER: 885 if (n->nlmsg_flags & NLM_F_EXCL) { 886 if (tp_created) 887 tcf_proto_destroy(tp); 888 err = -EEXIST; 889 goto errout; 890 } 891 break; 892 case RTM_DELTFILTER: 893 err = tfilter_del_notify(net, skb, n, tp, q, parent, 894 fh, false, &last); 895 if (err) 896 goto errout; 897 if (last) { 898 tcf_chain_tp_remove(chain, &chain_info, tp); 899 tcf_proto_destroy(tp); 900 } 901 goto errout; 902 case RTM_GETTFILTER: 903 err = tfilter_notify(net, skb, n, tp, q, parent, fh, 904 RTM_NEWTFILTER, true); 905 goto errout; 906 default: 907 err = -EINVAL; 908 goto errout; 909 } 910 } 911 912 err = tp->ops->change(net, skb, tp, cl, t->tcm_handle, tca, &fh, 913 n->nlmsg_flags & NLM_F_CREATE ? TCA_ACT_NOREPLACE : TCA_ACT_REPLACE); 914 if (err == 0) { 915 if (tp_created) 916 tcf_chain_tp_insert(chain, &chain_info, tp); 917 tfilter_notify(net, skb, n, tp, q, parent, fh, 918 RTM_NEWTFILTER, false); 919 } else { 920 if (tp_created) 921 tcf_proto_destroy(tp); 922 } 923 924 errout: 925 if (chain) 926 tcf_chain_put(chain); 927 if (err == -EAGAIN) 928 /* Replay the request. */ 929 goto replay; 930 return err; 931 } 932 933 struct tcf_dump_args { 934 struct tcf_walker w; 935 struct sk_buff *skb; 936 struct netlink_callback *cb; 937 struct Qdisc *q; 938 u32 parent; 939 }; 940 941 static int tcf_node_dump(struct tcf_proto *tp, void *n, struct tcf_walker *arg) 942 { 943 struct tcf_dump_args *a = (void *)arg; 944 struct net *net = sock_net(a->skb->sk); 945 946 return tcf_fill_node(net, a->skb, tp, a->q, a->parent, 947 n, NETLINK_CB(a->cb->skb).portid, 948 a->cb->nlh->nlmsg_seq, NLM_F_MULTI, 949 RTM_NEWTFILTER); 950 } 951 952 static bool tcf_chain_dump(struct tcf_chain *chain, struct Qdisc *q, u32 parent, 953 struct sk_buff *skb, struct netlink_callback *cb, 954 long index_start, long *p_index) 955 { 956 struct net *net = sock_net(skb->sk); 957 struct tcmsg *tcm = nlmsg_data(cb->nlh); 958 struct tcf_dump_args arg; 959 struct tcf_proto *tp; 960 961 for (tp = rtnl_dereference(chain->filter_chain); 962 tp; tp = rtnl_dereference(tp->next), (*p_index)++) { 963 if (*p_index < index_start) 964 continue; 965 if (TC_H_MAJ(tcm->tcm_info) && 966 TC_H_MAJ(tcm->tcm_info) != tp->prio) 967 continue; 968 if (TC_H_MIN(tcm->tcm_info) && 969 TC_H_MIN(tcm->tcm_info) != tp->protocol) 970 continue; 971 if (*p_index > index_start) 972 memset(&cb->args[1], 0, 973 sizeof(cb->args) - sizeof(cb->args[0])); 974 if (cb->args[1] == 0) { 975 if (tcf_fill_node(net, skb, tp, q, parent, 0, 976 NETLINK_CB(cb->skb).portid, 977 cb->nlh->nlmsg_seq, NLM_F_MULTI, 978 RTM_NEWTFILTER) <= 0) 979 return false; 980 981 cb->args[1] = 1; 982 } 983 if (!tp->ops->walk) 984 continue; 985 arg.w.fn = tcf_node_dump; 986 arg.skb = skb; 987 arg.cb = cb; 988 arg.q = q; 989 arg.parent = parent; 990 arg.w.stop = 0; 991 arg.w.skip = cb->args[1] - 1; 992 arg.w.count = 0; 993 tp->ops->walk(tp, &arg.w); 994 cb->args[1] = arg.w.count + 1; 995 if (arg.w.stop) 996 return false; 997 } 998 return true; 999 } 1000 1001 /* called with RTNL */ 1002 static int tc_dump_tfilter(struct sk_buff *skb, struct netlink_callback *cb) 1003 { 1004 struct net *net = sock_net(skb->sk); 1005 struct nlattr *tca[TCA_MAX + 1]; 1006 struct net_device *dev; 1007 struct Qdisc *q; 1008 struct tcf_block *block; 1009 struct tcf_chain *chain; 1010 struct tcmsg *tcm = nlmsg_data(cb->nlh); 1011 unsigned long cl = 0; 1012 const struct Qdisc_class_ops *cops; 1013 long index_start; 1014 long index; 1015 u32 parent; 1016 int err; 1017 1018 if (nlmsg_len(cb->nlh) < sizeof(*tcm)) 1019 return skb->len; 1020 1021 err = nlmsg_parse(cb->nlh, sizeof(*tcm), tca, TCA_MAX, NULL, NULL); 1022 if (err) 1023 return err; 1024 1025 dev = __dev_get_by_index(net, tcm->tcm_ifindex); 1026 if (!dev) 1027 return skb->len; 1028 1029 parent = tcm->tcm_parent; 1030 if (!parent) { 1031 q = dev->qdisc; 1032 parent = q->handle; 1033 } else { 1034 q = qdisc_lookup(dev, TC_H_MAJ(tcm->tcm_parent)); 1035 } 1036 if (!q) 1037 goto out; 1038 cops = q->ops->cl_ops; 1039 if (!cops) 1040 goto out; 1041 if (!cops->tcf_block) 1042 goto out; 1043 if (TC_H_MIN(tcm->tcm_parent)) { 1044 cl = cops->find(q, tcm->tcm_parent); 1045 if (cl == 0) 1046 goto out; 1047 } 1048 block = cops->tcf_block(q, cl); 1049 if (!block) 1050 goto out; 1051 1052 index_start = cb->args[0]; 1053 index = 0; 1054 1055 list_for_each_entry(chain, &block->chain_list, list) { 1056 if (tca[TCA_CHAIN] && 1057 nla_get_u32(tca[TCA_CHAIN]) != chain->index) 1058 continue; 1059 if (!tcf_chain_dump(chain, q, parent, skb, cb, 1060 index_start, &index)) 1061 break; 1062 } 1063 1064 cb->args[0] = index; 1065 1066 out: 1067 return skb->len; 1068 } 1069 1070 void tcf_exts_destroy(struct tcf_exts *exts) 1071 { 1072 #ifdef CONFIG_NET_CLS_ACT 1073 LIST_HEAD(actions); 1074 1075 ASSERT_RTNL(); 1076 tcf_exts_to_list(exts, &actions); 1077 tcf_action_destroy(&actions, TCA_ACT_UNBIND); 1078 kfree(exts->actions); 1079 exts->nr_actions = 0; 1080 #endif 1081 } 1082 EXPORT_SYMBOL(tcf_exts_destroy); 1083 1084 int tcf_exts_validate(struct net *net, struct tcf_proto *tp, struct nlattr **tb, 1085 struct nlattr *rate_tlv, struct tcf_exts *exts, bool ovr) 1086 { 1087 #ifdef CONFIG_NET_CLS_ACT 1088 { 1089 struct tc_action *act; 1090 1091 if (exts->police && tb[exts->police]) { 1092 act = tcf_action_init_1(net, tp, tb[exts->police], 1093 rate_tlv, "police", ovr, 1094 TCA_ACT_BIND); 1095 if (IS_ERR(act)) 1096 return PTR_ERR(act); 1097 1098 act->type = exts->type = TCA_OLD_COMPAT; 1099 exts->actions[0] = act; 1100 exts->nr_actions = 1; 1101 } else if (exts->action && tb[exts->action]) { 1102 LIST_HEAD(actions); 1103 int err, i = 0; 1104 1105 err = tcf_action_init(net, tp, tb[exts->action], 1106 rate_tlv, NULL, ovr, TCA_ACT_BIND, 1107 &actions); 1108 if (err) 1109 return err; 1110 list_for_each_entry(act, &actions, list) 1111 exts->actions[i++] = act; 1112 exts->nr_actions = i; 1113 } 1114 exts->net = net; 1115 } 1116 #else 1117 if ((exts->action && tb[exts->action]) || 1118 (exts->police && tb[exts->police])) 1119 return -EOPNOTSUPP; 1120 #endif 1121 1122 return 0; 1123 } 1124 EXPORT_SYMBOL(tcf_exts_validate); 1125 1126 void tcf_exts_change(struct tcf_exts *dst, struct tcf_exts *src) 1127 { 1128 #ifdef CONFIG_NET_CLS_ACT 1129 struct tcf_exts old = *dst; 1130 1131 *dst = *src; 1132 tcf_exts_destroy(&old); 1133 #endif 1134 } 1135 EXPORT_SYMBOL(tcf_exts_change); 1136 1137 #ifdef CONFIG_NET_CLS_ACT 1138 static struct tc_action *tcf_exts_first_act(struct tcf_exts *exts) 1139 { 1140 if (exts->nr_actions == 0) 1141 return NULL; 1142 else 1143 return exts->actions[0]; 1144 } 1145 #endif 1146 1147 int tcf_exts_dump(struct sk_buff *skb, struct tcf_exts *exts) 1148 { 1149 #ifdef CONFIG_NET_CLS_ACT 1150 struct nlattr *nest; 1151 1152 if (exts->action && tcf_exts_has_actions(exts)) { 1153 /* 1154 * again for backward compatible mode - we want 1155 * to work with both old and new modes of entering 1156 * tc data even if iproute2 was newer - jhs 1157 */ 1158 if (exts->type != TCA_OLD_COMPAT) { 1159 LIST_HEAD(actions); 1160 1161 nest = nla_nest_start(skb, exts->action); 1162 if (nest == NULL) 1163 goto nla_put_failure; 1164 1165 tcf_exts_to_list(exts, &actions); 1166 if (tcf_action_dump(skb, &actions, 0, 0) < 0) 1167 goto nla_put_failure; 1168 nla_nest_end(skb, nest); 1169 } else if (exts->police) { 1170 struct tc_action *act = tcf_exts_first_act(exts); 1171 nest = nla_nest_start(skb, exts->police); 1172 if (nest == NULL || !act) 1173 goto nla_put_failure; 1174 if (tcf_action_dump_old(skb, act, 0, 0) < 0) 1175 goto nla_put_failure; 1176 nla_nest_end(skb, nest); 1177 } 1178 } 1179 return 0; 1180 1181 nla_put_failure: 1182 nla_nest_cancel(skb, nest); 1183 return -1; 1184 #else 1185 return 0; 1186 #endif 1187 } 1188 EXPORT_SYMBOL(tcf_exts_dump); 1189 1190 1191 int tcf_exts_dump_stats(struct sk_buff *skb, struct tcf_exts *exts) 1192 { 1193 #ifdef CONFIG_NET_CLS_ACT 1194 struct tc_action *a = tcf_exts_first_act(exts); 1195 if (a != NULL && tcf_action_copy_stats(skb, a, 1) < 0) 1196 return -1; 1197 #endif 1198 return 0; 1199 } 1200 EXPORT_SYMBOL(tcf_exts_dump_stats); 1201 1202 static int tc_exts_setup_cb_egdev_call(struct tcf_exts *exts, 1203 enum tc_setup_type type, 1204 void *type_data, bool err_stop) 1205 { 1206 int ok_count = 0; 1207 #ifdef CONFIG_NET_CLS_ACT 1208 const struct tc_action *a; 1209 struct net_device *dev; 1210 int i, ret; 1211 1212 if (!tcf_exts_has_actions(exts)) 1213 return 0; 1214 1215 for (i = 0; i < exts->nr_actions; i++) { 1216 a = exts->actions[i]; 1217 if (!a->ops->get_dev) 1218 continue; 1219 dev = a->ops->get_dev(a); 1220 if (!dev) 1221 continue; 1222 ret = tc_setup_cb_egdev_call(dev, type, type_data, err_stop); 1223 if (ret < 0) 1224 return ret; 1225 ok_count += ret; 1226 } 1227 #endif 1228 return ok_count; 1229 } 1230 1231 int tc_setup_cb_call(struct tcf_block *block, struct tcf_exts *exts, 1232 enum tc_setup_type type, void *type_data, bool err_stop) 1233 { 1234 int ok_count; 1235 int ret; 1236 1237 ret = tcf_block_cb_call(block, type, type_data, err_stop); 1238 if (ret < 0) 1239 return ret; 1240 ok_count = ret; 1241 1242 if (!exts) 1243 return ok_count; 1244 ret = tc_exts_setup_cb_egdev_call(exts, type, type_data, err_stop); 1245 if (ret < 0) 1246 return ret; 1247 ok_count += ret; 1248 1249 return ok_count; 1250 } 1251 EXPORT_SYMBOL(tc_setup_cb_call); 1252 1253 static int __init tc_filter_init(void) 1254 { 1255 tc_filter_wq = alloc_ordered_workqueue("tc_filter_workqueue", 0); 1256 if (!tc_filter_wq) 1257 return -ENOMEM; 1258 1259 rtnl_register(PF_UNSPEC, RTM_NEWTFILTER, tc_ctl_tfilter, NULL, 0); 1260 rtnl_register(PF_UNSPEC, RTM_DELTFILTER, tc_ctl_tfilter, NULL, 0); 1261 rtnl_register(PF_UNSPEC, RTM_GETTFILTER, tc_ctl_tfilter, 1262 tc_dump_tfilter, 0); 1263 1264 return 0; 1265 } 1266 1267 subsys_initcall(tc_filter_init); 1268