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