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