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