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/skbuff.h> 23#include <linux/init.h> 24#include <linux/kmod.h> 25#include <linux/err.h> 26#include <linux/slab.h> 27#include <net/net_namespace.h> 28#include <net/sock.h> 29#include <net/netlink.h> 30#include <net/pkt_sched.h> 31#include <net/pkt_cls.h> 32 33/* The list of all installed classifier types */ 34static LIST_HEAD(tcf_proto_base); 35 36/* Protects list of registered TC modules. It is pure SMP lock. */ 37static DEFINE_RWLOCK(cls_mod_lock); 38 39/* Find classifier type by string name */ 40 41static const struct tcf_proto_ops *tcf_proto_lookup_ops(struct nlattr *kind) 42{ 43 const struct tcf_proto_ops *t, *res = NULL; 44 45 if (kind) { 46 read_lock(&cls_mod_lock); 47 list_for_each_entry(t, &tcf_proto_base, head) { 48 if (nla_strcmp(kind, t->kind) == 0) { 49 if (try_module_get(t->owner)) 50 res = t; 51 break; 52 } 53 } 54 read_unlock(&cls_mod_lock); 55 } 56 return res; 57} 58 59/* Register(unregister) new classifier type */ 60 61int register_tcf_proto_ops(struct tcf_proto_ops *ops) 62{ 63 struct tcf_proto_ops *t; 64 int rc = -EEXIST; 65 66 write_lock(&cls_mod_lock); 67 list_for_each_entry(t, &tcf_proto_base, head) 68 if (!strcmp(ops->kind, t->kind)) 69 goto out; 70 71 list_add_tail(&ops->head, &tcf_proto_base); 72 rc = 0; 73out: 74 write_unlock(&cls_mod_lock); 75 return rc; 76} 77EXPORT_SYMBOL(register_tcf_proto_ops); 78 79int unregister_tcf_proto_ops(struct tcf_proto_ops *ops) 80{ 81 struct tcf_proto_ops *t; 82 int rc = -ENOENT; 83 84 /* Wait for outstanding call_rcu()s, if any, from a 85 * tcf_proto_ops's destroy() handler. 86 */ 87 rcu_barrier(); 88 89 write_lock(&cls_mod_lock); 90 list_for_each_entry(t, &tcf_proto_base, head) { 91 if (t == ops) { 92 list_del(&t->head); 93 rc = 0; 94 break; 95 } 96 } 97 write_unlock(&cls_mod_lock); 98 return rc; 99} 100EXPORT_SYMBOL(unregister_tcf_proto_ops); 101 102static int tfilter_notify(struct net *net, struct sk_buff *oskb, 103 struct nlmsghdr *n, struct tcf_proto *tp, 104 unsigned long fh, int event); 105 106 107/* Select new prio value from the range, managed by kernel. */ 108 109static inline u32 tcf_auto_prio(struct tcf_proto *tp) 110{ 111 u32 first = TC_H_MAKE(0xC0000000U, 0U); 112 113 if (tp) 114 first = tp->prio - 1; 115 116 return first; 117} 118 119/* Add/change/delete/get a filter node */ 120 121static int tc_ctl_tfilter(struct sk_buff *skb, struct nlmsghdr *n) 122{ 123 struct net *net = sock_net(skb->sk); 124 struct nlattr *tca[TCA_MAX + 1]; 125 struct tcmsg *t; 126 u32 protocol; 127 u32 prio; 128 u32 nprio; 129 u32 parent; 130 struct net_device *dev; 131 struct Qdisc *q; 132 struct tcf_proto __rcu **back; 133 struct tcf_proto __rcu **chain; 134 struct tcf_proto *tp; 135 const struct tcf_proto_ops *tp_ops; 136 const struct Qdisc_class_ops *cops; 137 unsigned long cl; 138 unsigned long fh; 139 int err; 140 int tp_created = 0; 141 142 if ((n->nlmsg_type != RTM_GETTFILTER) && 143 !netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN)) 144 return -EPERM; 145 146replay: 147 err = nlmsg_parse(n, sizeof(*t), tca, TCA_MAX, NULL); 148 if (err < 0) 149 return err; 150 151 t = nlmsg_data(n); 152 protocol = TC_H_MIN(t->tcm_info); 153 prio = TC_H_MAJ(t->tcm_info); 154 nprio = prio; 155 parent = t->tcm_parent; 156 cl = 0; 157 158 if (prio == 0) { 159 /* If no priority is given, user wants we allocated it. */ 160 if (n->nlmsg_type != RTM_NEWTFILTER || 161 !(n->nlmsg_flags & NLM_F_CREATE)) 162 return -ENOENT; 163 prio = TC_H_MAKE(0x80000000U, 0U); 164 } 165 166 /* Find head of filter chain. */ 167 168 /* Find link */ 169 dev = __dev_get_by_index(net, t->tcm_ifindex); 170 if (dev == NULL) 171 return -ENODEV; 172 173 /* Find qdisc */ 174 if (!parent) { 175 q = dev->qdisc; 176 parent = q->handle; 177 } else { 178 q = qdisc_lookup(dev, TC_H_MAJ(t->tcm_parent)); 179 if (q == NULL) 180 return -EINVAL; 181 } 182 183 /* Is it classful? */ 184 cops = q->ops->cl_ops; 185 if (!cops) 186 return -EINVAL; 187 188 if (cops->tcf_chain == NULL) 189 return -EOPNOTSUPP; 190 191 /* Do we search for filter, attached to class? */ 192 if (TC_H_MIN(parent)) { 193 cl = cops->get(q, parent); 194 if (cl == 0) 195 return -ENOENT; 196 } 197 198 /* And the last stroke */ 199 chain = cops->tcf_chain(q, cl); 200 err = -EINVAL; 201 if (chain == NULL) 202 goto errout; 203 204 /* Check the chain for existence of proto-tcf with this priority */ 205 for (back = chain; 206 (tp = rtnl_dereference(*back)) != NULL; 207 back = &tp->next) { 208 if (tp->prio >= prio) { 209 if (tp->prio == prio) { 210 if (!nprio || 211 (tp->protocol != protocol && protocol)) 212 goto errout; 213 } else 214 tp = NULL; 215 break; 216 } 217 } 218 219 if (tp == NULL) { 220 /* Proto-tcf does not exist, create new one */ 221 222 if (tca[TCA_KIND] == NULL || !protocol) 223 goto errout; 224 225 err = -ENOENT; 226 if (n->nlmsg_type != RTM_NEWTFILTER || 227 !(n->nlmsg_flags & NLM_F_CREATE)) 228 goto errout; 229 230 231 /* Create new proto tcf */ 232 233 err = -ENOBUFS; 234 tp = kzalloc(sizeof(*tp), GFP_KERNEL); 235 if (tp == NULL) 236 goto errout; 237 err = -ENOENT; 238 tp_ops = tcf_proto_lookup_ops(tca[TCA_KIND]); 239 if (tp_ops == NULL) { 240#ifdef CONFIG_MODULES 241 struct nlattr *kind = tca[TCA_KIND]; 242 char name[IFNAMSIZ]; 243 244 if (kind != NULL && 245 nla_strlcpy(name, kind, IFNAMSIZ) < IFNAMSIZ) { 246 rtnl_unlock(); 247 request_module("cls_%s", name); 248 rtnl_lock(); 249 tp_ops = tcf_proto_lookup_ops(kind); 250 /* We dropped the RTNL semaphore in order to 251 * perform the module load. So, even if we 252 * succeeded in loading the module we have to 253 * replay the request. We indicate this using 254 * -EAGAIN. 255 */ 256 if (tp_ops != NULL) { 257 module_put(tp_ops->owner); 258 err = -EAGAIN; 259 } 260 } 261#endif 262 kfree(tp); 263 goto errout; 264 } 265 tp->ops = tp_ops; 266 tp->protocol = protocol; 267 tp->prio = nprio ? : 268 TC_H_MAJ(tcf_auto_prio(rtnl_dereference(*back))); 269 tp->q = q; 270 tp->classify = tp_ops->classify; 271 tp->classid = parent; 272 273 err = tp_ops->init(tp); 274 if (err != 0) { 275 module_put(tp_ops->owner); 276 kfree(tp); 277 goto errout; 278 } 279 280 tp_created = 1; 281 282 } else if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind)) 283 goto errout; 284 285 fh = tp->ops->get(tp, t->tcm_handle); 286 287 if (fh == 0) { 288 if (n->nlmsg_type == RTM_DELTFILTER && t->tcm_handle == 0) { 289 struct tcf_proto *next = rtnl_dereference(tp->next); 290 291 RCU_INIT_POINTER(*back, next); 292 293 tfilter_notify(net, skb, n, tp, fh, RTM_DELTFILTER); 294 tcf_destroy(tp, true); 295 err = 0; 296 goto errout; 297 } 298 299 err = -ENOENT; 300 if (n->nlmsg_type != RTM_NEWTFILTER || 301 !(n->nlmsg_flags & NLM_F_CREATE)) 302 goto errout; 303 } else { 304 switch (n->nlmsg_type) { 305 case RTM_NEWTFILTER: 306 err = -EEXIST; 307 if (n->nlmsg_flags & NLM_F_EXCL) { 308 if (tp_created) 309 tcf_destroy(tp, true); 310 goto errout; 311 } 312 break; 313 case RTM_DELTFILTER: 314 err = tp->ops->delete(tp, fh); 315 if (err == 0) { 316 struct tcf_proto *next = rtnl_dereference(tp->next); 317 318 tfilter_notify(net, skb, n, tp, fh, RTM_DELTFILTER); 319 if (tcf_destroy(tp, false)) 320 RCU_INIT_POINTER(*back, next); 321 } 322 goto errout; 323 case RTM_GETTFILTER: 324 err = tfilter_notify(net, skb, n, tp, fh, RTM_NEWTFILTER); 325 goto errout; 326 default: 327 err = -EINVAL; 328 goto errout; 329 } 330 } 331 332 err = tp->ops->change(net, skb, tp, cl, t->tcm_handle, tca, &fh, 333 n->nlmsg_flags & NLM_F_CREATE ? TCA_ACT_NOREPLACE : TCA_ACT_REPLACE); 334 if (err == 0) { 335 if (tp_created) { 336 RCU_INIT_POINTER(tp->next, rtnl_dereference(*back)); 337 rcu_assign_pointer(*back, tp); 338 } 339 tfilter_notify(net, skb, n, tp, fh, RTM_NEWTFILTER); 340 } else { 341 if (tp_created) 342 tcf_destroy(tp, true); 343 } 344 345errout: 346 if (cl) 347 cops->put(q, cl); 348 if (err == -EAGAIN) 349 /* Replay the request. */ 350 goto replay; 351 return err; 352} 353 354static int tcf_fill_node(struct net *net, struct sk_buff *skb, struct tcf_proto *tp, 355 unsigned long fh, u32 portid, u32 seq, u16 flags, int event) 356{ 357 struct tcmsg *tcm; 358 struct nlmsghdr *nlh; 359 unsigned char *b = skb_tail_pointer(skb); 360 361 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags); 362 if (!nlh) 363 goto out_nlmsg_trim; 364 tcm = nlmsg_data(nlh); 365 tcm->tcm_family = AF_UNSPEC; 366 tcm->tcm__pad1 = 0; 367 tcm->tcm__pad2 = 0; 368 tcm->tcm_ifindex = qdisc_dev(tp->q)->ifindex; 369 tcm->tcm_parent = tp->classid; 370 tcm->tcm_info = TC_H_MAKE(tp->prio, tp->protocol); 371 if (nla_put_string(skb, TCA_KIND, tp->ops->kind)) 372 goto nla_put_failure; 373 tcm->tcm_handle = fh; 374 if (RTM_DELTFILTER != event) { 375 tcm->tcm_handle = 0; 376 if (tp->ops->dump && tp->ops->dump(net, tp, fh, skb, tcm) < 0) 377 goto nla_put_failure; 378 } 379 nlh->nlmsg_len = skb_tail_pointer(skb) - b; 380 return skb->len; 381 382out_nlmsg_trim: 383nla_put_failure: 384 nlmsg_trim(skb, b); 385 return -1; 386} 387 388static int tfilter_notify(struct net *net, struct sk_buff *oskb, 389 struct nlmsghdr *n, struct tcf_proto *tp, 390 unsigned long fh, int event) 391{ 392 struct sk_buff *skb; 393 u32 portid = oskb ? NETLINK_CB(oskb).portid : 0; 394 395 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL); 396 if (!skb) 397 return -ENOBUFS; 398 399 if (tcf_fill_node(net, skb, tp, fh, portid, n->nlmsg_seq, 0, event) <= 0) { 400 kfree_skb(skb); 401 return -EINVAL; 402 } 403 404 return rtnetlink_send(skb, net, portid, RTNLGRP_TC, 405 n->nlmsg_flags & NLM_F_ECHO); 406} 407 408struct tcf_dump_args { 409 struct tcf_walker w; 410 struct sk_buff *skb; 411 struct netlink_callback *cb; 412}; 413 414static int tcf_node_dump(struct tcf_proto *tp, unsigned long n, 415 struct tcf_walker *arg) 416{ 417 struct tcf_dump_args *a = (void *)arg; 418 struct net *net = sock_net(a->skb->sk); 419 420 return tcf_fill_node(net, a->skb, tp, n, NETLINK_CB(a->cb->skb).portid, 421 a->cb->nlh->nlmsg_seq, NLM_F_MULTI, RTM_NEWTFILTER); 422} 423 424/* called with RTNL */ 425static int tc_dump_tfilter(struct sk_buff *skb, struct netlink_callback *cb) 426{ 427 struct net *net = sock_net(skb->sk); 428 int t; 429 int s_t; 430 struct net_device *dev; 431 struct Qdisc *q; 432 struct tcf_proto *tp, __rcu **chain; 433 struct tcmsg *tcm = nlmsg_data(cb->nlh); 434 unsigned long cl = 0; 435 const struct Qdisc_class_ops *cops; 436 struct tcf_dump_args arg; 437 438 if (nlmsg_len(cb->nlh) < sizeof(*tcm)) 439 return skb->len; 440 dev = __dev_get_by_index(net, tcm->tcm_ifindex); 441 if (!dev) 442 return skb->len; 443 444 if (!tcm->tcm_parent) 445 q = dev->qdisc; 446 else 447 q = qdisc_lookup(dev, TC_H_MAJ(tcm->tcm_parent)); 448 if (!q) 449 goto out; 450 cops = q->ops->cl_ops; 451 if (!cops) 452 goto errout; 453 if (cops->tcf_chain == NULL) 454 goto errout; 455 if (TC_H_MIN(tcm->tcm_parent)) { 456 cl = cops->get(q, tcm->tcm_parent); 457 if (cl == 0) 458 goto errout; 459 } 460 chain = cops->tcf_chain(q, cl); 461 if (chain == NULL) 462 goto errout; 463 464 s_t = cb->args[0]; 465 466 for (tp = rtnl_dereference(*chain), t = 0; 467 tp; tp = rtnl_dereference(tp->next), t++) { 468 if (t < s_t) 469 continue; 470 if (TC_H_MAJ(tcm->tcm_info) && 471 TC_H_MAJ(tcm->tcm_info) != tp->prio) 472 continue; 473 if (TC_H_MIN(tcm->tcm_info) && 474 TC_H_MIN(tcm->tcm_info) != tp->protocol) 475 continue; 476 if (t > s_t) 477 memset(&cb->args[1], 0, sizeof(cb->args)-sizeof(cb->args[0])); 478 if (cb->args[1] == 0) { 479 if (tcf_fill_node(net, skb, tp, 0, NETLINK_CB(cb->skb).portid, 480 cb->nlh->nlmsg_seq, NLM_F_MULTI, 481 RTM_NEWTFILTER) <= 0) 482 break; 483 484 cb->args[1] = 1; 485 } 486 if (tp->ops->walk == NULL) 487 continue; 488 arg.w.fn = tcf_node_dump; 489 arg.skb = skb; 490 arg.cb = cb; 491 arg.w.stop = 0; 492 arg.w.skip = cb->args[1] - 1; 493 arg.w.count = 0; 494 tp->ops->walk(tp, &arg.w); 495 cb->args[1] = arg.w.count + 1; 496 if (arg.w.stop) 497 break; 498 } 499 500 cb->args[0] = t; 501 502errout: 503 if (cl) 504 cops->put(q, cl); 505out: 506 return skb->len; 507} 508 509void tcf_exts_destroy(struct tcf_exts *exts) 510{ 511#ifdef CONFIG_NET_CLS_ACT 512 tcf_action_destroy(&exts->actions, TCA_ACT_UNBIND); 513 INIT_LIST_HEAD(&exts->actions); 514#endif 515} 516EXPORT_SYMBOL(tcf_exts_destroy); 517 518int tcf_exts_validate(struct net *net, struct tcf_proto *tp, struct nlattr **tb, 519 struct nlattr *rate_tlv, struct tcf_exts *exts, bool ovr) 520{ 521#ifdef CONFIG_NET_CLS_ACT 522 { 523 struct tc_action *act; 524 525 INIT_LIST_HEAD(&exts->actions); 526 if (exts->police && tb[exts->police]) { 527 act = tcf_action_init_1(net, tb[exts->police], rate_tlv, 528 "police", ovr, 529 TCA_ACT_BIND); 530 if (IS_ERR(act)) 531 return PTR_ERR(act); 532 533 act->type = exts->type = TCA_OLD_COMPAT; 534 list_add(&act->list, &exts->actions); 535 } else if (exts->action && tb[exts->action]) { 536 int err; 537 err = tcf_action_init(net, tb[exts->action], rate_tlv, 538 NULL, ovr, 539 TCA_ACT_BIND, &exts->actions); 540 if (err) 541 return err; 542 } 543 } 544#else 545 if ((exts->action && tb[exts->action]) || 546 (exts->police && tb[exts->police])) 547 return -EOPNOTSUPP; 548#endif 549 550 return 0; 551} 552EXPORT_SYMBOL(tcf_exts_validate); 553 554void tcf_exts_change(struct tcf_proto *tp, struct tcf_exts *dst, 555 struct tcf_exts *src) 556{ 557#ifdef CONFIG_NET_CLS_ACT 558 LIST_HEAD(tmp); 559 tcf_tree_lock(tp); 560 list_splice_init(&dst->actions, &tmp); 561 list_splice(&src->actions, &dst->actions); 562 dst->type = src->type; 563 tcf_tree_unlock(tp); 564 tcf_action_destroy(&tmp, TCA_ACT_UNBIND); 565#endif 566} 567EXPORT_SYMBOL(tcf_exts_change); 568 569#define tcf_exts_first_act(ext) \ 570 list_first_entry_or_null(&(exts)->actions, \ 571 struct tc_action, list) 572 573int tcf_exts_dump(struct sk_buff *skb, struct tcf_exts *exts) 574{ 575#ifdef CONFIG_NET_CLS_ACT 576 struct nlattr *nest; 577 578 if (exts->action && !list_empty(&exts->actions)) { 579 /* 580 * again for backward compatible mode - we want 581 * to work with both old and new modes of entering 582 * tc data even if iproute2 was newer - jhs 583 */ 584 if (exts->type != TCA_OLD_COMPAT) { 585 nest = nla_nest_start(skb, exts->action); 586 if (nest == NULL) 587 goto nla_put_failure; 588 if (tcf_action_dump(skb, &exts->actions, 0, 0) < 0) 589 goto nla_put_failure; 590 nla_nest_end(skb, nest); 591 } else if (exts->police) { 592 struct tc_action *act = tcf_exts_first_act(exts); 593 nest = nla_nest_start(skb, exts->police); 594 if (nest == NULL || !act) 595 goto nla_put_failure; 596 if (tcf_action_dump_old(skb, act, 0, 0) < 0) 597 goto nla_put_failure; 598 nla_nest_end(skb, nest); 599 } 600 } 601 return 0; 602 603nla_put_failure: 604 nla_nest_cancel(skb, nest); 605 return -1; 606#else 607 return 0; 608#endif 609} 610EXPORT_SYMBOL(tcf_exts_dump); 611 612 613int tcf_exts_dump_stats(struct sk_buff *skb, struct tcf_exts *exts) 614{ 615#ifdef CONFIG_NET_CLS_ACT 616 struct tc_action *a = tcf_exts_first_act(exts); 617 if (a != NULL && tcf_action_copy_stats(skb, a, 1) < 0) 618 return -1; 619#endif 620 return 0; 621} 622EXPORT_SYMBOL(tcf_exts_dump_stats); 623 624static int __init tc_filter_init(void) 625{ 626 rtnl_register(PF_UNSPEC, RTM_NEWTFILTER, tc_ctl_tfilter, NULL, NULL); 627 rtnl_register(PF_UNSPEC, RTM_DELTFILTER, tc_ctl_tfilter, NULL, NULL); 628 rtnl_register(PF_UNSPEC, RTM_GETTFILTER, tc_ctl_tfilter, 629 tc_dump_tfilter, NULL); 630 631 return 0; 632} 633 634subsys_initcall(tc_filter_init); 635