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 <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 36 static struct tcf_proto_ops *tcf_proto_base __read_mostly; 37 38 /* Protects list of registered TC modules. It is pure SMP lock. */ 39 static DEFINE_RWLOCK(cls_mod_lock); 40 41 /* Find classifier type by string name */ 42 43 static struct tcf_proto_ops *tcf_proto_lookup_ops(struct nlattr *kind) 44 { 45 struct tcf_proto_ops *t = NULL; 46 47 if (kind) { 48 read_lock(&cls_mod_lock); 49 for (t = tcf_proto_base; t; t = t->next) { 50 if (nla_strcmp(kind, t->kind) == 0) { 51 if (!try_module_get(t->owner)) 52 t = NULL; 53 break; 54 } 55 } 56 read_unlock(&cls_mod_lock); 57 } 58 return t; 59 } 60 61 /* Register(unregister) new classifier type */ 62 63 int register_tcf_proto_ops(struct tcf_proto_ops *ops) 64 { 65 struct tcf_proto_ops *t, **tp; 66 int rc = -EEXIST; 67 68 write_lock(&cls_mod_lock); 69 for (tp = &tcf_proto_base; (t = *tp) != NULL; tp = &t->next) 70 if (!strcmp(ops->kind, t->kind)) 71 goto out; 72 73 ops->next = NULL; 74 *tp = ops; 75 rc = 0; 76 out: 77 write_unlock(&cls_mod_lock); 78 return rc; 79 } 80 EXPORT_SYMBOL(register_tcf_proto_ops); 81 82 int unregister_tcf_proto_ops(struct tcf_proto_ops *ops) 83 { 84 struct tcf_proto_ops *t, **tp; 85 int rc = -ENOENT; 86 87 write_lock(&cls_mod_lock); 88 for (tp = &tcf_proto_base; (t=*tp) != NULL; tp = &t->next) 89 if (t == ops) 90 break; 91 92 if (!t) 93 goto out; 94 *tp = t->next; 95 rc = 0; 96 out: 97 write_unlock(&cls_mod_lock); 98 return rc; 99 } 100 EXPORT_SYMBOL(unregister_tcf_proto_ops); 101 102 static int tfilter_notify(struct sk_buff *oskb, struct nlmsghdr *n, 103 struct tcf_proto *tp, unsigned long fh, int event); 104 105 106 /* Select new prio value from the range, managed by kernel. */ 107 108 static inline u32 tcf_auto_prio(struct tcf_proto *tp) 109 { 110 u32 first = TC_H_MAKE(0xC0000000U, 0U); 111 112 if (tp) 113 first = tp->prio-1; 114 115 return first; 116 } 117 118 /* Add/change/delete/get a filter node */ 119 120 static int tc_ctl_tfilter(struct sk_buff *skb, struct nlmsghdr *n, void *arg) 121 { 122 struct net *net = sock_net(skb->sk); 123 struct nlattr *tca[TCA_MAX + 1]; 124 spinlock_t *root_lock; 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 **back, **chain; 133 struct tcf_proto *tp; 134 struct tcf_proto_ops *tp_ops; 135 const struct Qdisc_class_ops *cops; 136 unsigned long cl; 137 unsigned long fh; 138 int err; 139 int tp_created = 0; 140 141 if (!net_eq(net, &init_net)) 142 return -EINVAL; 143 144 replay: 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 || !(n->nlmsg_flags&NLM_F_CREATE)) 155 return -ENOENT; 156 prio = TC_H_MAKE(0x80000000U, 0U); 157 } 158 159 /* Find head of filter chain. */ 160 161 /* Find link */ 162 dev = __dev_get_by_index(&init_net, t->tcm_ifindex); 163 if (dev == NULL) 164 return -ENODEV; 165 166 err = nlmsg_parse(n, sizeof(*t), tca, TCA_MAX, NULL); 167 if (err < 0) 168 return err; 169 170 /* Find qdisc */ 171 if (!parent) { 172 q = dev->qdisc; 173 parent = q->handle; 174 } else { 175 q = qdisc_lookup(dev, TC_H_MAJ(t->tcm_parent)); 176 if (q == NULL) 177 return -EINVAL; 178 } 179 180 /* Is it classful? */ 181 if ((cops = q->ops->cl_ops) == NULL) 182 return -EINVAL; 183 184 if (cops->tcf_chain == NULL) 185 return -EOPNOTSUPP; 186 187 /* Do we search for filter, attached to class? */ 188 if (TC_H_MIN(parent)) { 189 cl = cops->get(q, parent); 190 if (cl == 0) 191 return -ENOENT; 192 } 193 194 /* And the last stroke */ 195 chain = cops->tcf_chain(q, cl); 196 err = -EINVAL; 197 if (chain == NULL) 198 goto errout; 199 200 /* Check the chain for existence of proto-tcf with this priority */ 201 for (back = chain; (tp=*back) != NULL; back = &tp->next) { 202 if (tp->prio >= prio) { 203 if (tp->prio == prio) { 204 if (!nprio || (tp->protocol != protocol && protocol)) 205 goto errout; 206 } else 207 tp = NULL; 208 break; 209 } 210 } 211 212 root_lock = qdisc_root_sleeping_lock(q); 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 || !(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(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(skb, n, tp, fh, RTM_DELTFILTER); 310 goto errout; 311 case RTM_GETTFILTER: 312 err = tfilter_notify(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(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(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 pid, 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_NEW(skb, pid, seq, event, sizeof(*tcm), flags); 351 tcm = NLMSG_DATA(nlh); 352 tcm->tcm_family = AF_UNSPEC; 353 tcm->tcm__pad1 = 0; 354 tcm->tcm__pad2 = 0; 355 tcm->tcm_ifindex = qdisc_dev(tp->q)->ifindex; 356 tcm->tcm_parent = tp->classid; 357 tcm->tcm_info = TC_H_MAKE(tp->prio, tp->protocol); 358 NLA_PUT_STRING(skb, TCA_KIND, tp->ops->kind); 359 tcm->tcm_handle = fh; 360 if (RTM_DELTFILTER != event) { 361 tcm->tcm_handle = 0; 362 if (tp->ops->dump && tp->ops->dump(tp, fh, skb, tcm) < 0) 363 goto nla_put_failure; 364 } 365 nlh->nlmsg_len = skb_tail_pointer(skb) - b; 366 return skb->len; 367 368 nlmsg_failure: 369 nla_put_failure: 370 nlmsg_trim(skb, b); 371 return -1; 372 } 373 374 static int tfilter_notify(struct sk_buff *oskb, struct nlmsghdr *n, 375 struct tcf_proto *tp, unsigned long fh, int event) 376 { 377 struct sk_buff *skb; 378 u32 pid = oskb ? NETLINK_CB(oskb).pid : 0; 379 380 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL); 381 if (!skb) 382 return -ENOBUFS; 383 384 if (tcf_fill_node(skb, tp, fh, pid, n->nlmsg_seq, 0, event) <= 0) { 385 kfree_skb(skb); 386 return -EINVAL; 387 } 388 389 return rtnetlink_send(skb, &init_net, pid, RTNLGRP_TC, 390 n->nlmsg_flags & NLM_F_ECHO); 391 } 392 393 struct tcf_dump_args { 394 struct tcf_walker w; 395 struct sk_buff *skb; 396 struct netlink_callback *cb; 397 }; 398 399 static int tcf_node_dump(struct tcf_proto *tp, unsigned long n, 400 struct tcf_walker *arg) 401 { 402 struct tcf_dump_args *a = (void *)arg; 403 404 return tcf_fill_node(a->skb, tp, n, NETLINK_CB(a->cb->skb).pid, 405 a->cb->nlh->nlmsg_seq, NLM_F_MULTI, RTM_NEWTFILTER); 406 } 407 408 /* called with RTNL */ 409 static int tc_dump_tfilter(struct sk_buff *skb, struct netlink_callback *cb) 410 { 411 struct net *net = sock_net(skb->sk); 412 int t; 413 int s_t; 414 struct net_device *dev; 415 struct Qdisc *q; 416 struct tcf_proto *tp, **chain; 417 struct tcmsg *tcm = (struct tcmsg *)NLMSG_DATA(cb->nlh); 418 unsigned long cl = 0; 419 const struct Qdisc_class_ops *cops; 420 struct tcf_dump_args arg; 421 422 if (!net_eq(net, &init_net)) 423 return 0; 424 425 if (cb->nlh->nlmsg_len < NLMSG_LENGTH(sizeof(*tcm))) 426 return skb->len; 427 if ((dev = __dev_get_by_index(&init_net, tcm->tcm_ifindex)) == NULL) 428 return skb->len; 429 430 if (!tcm->tcm_parent) 431 q = dev->qdisc; 432 else 433 q = qdisc_lookup(dev, TC_H_MAJ(tcm->tcm_parent)); 434 if (!q) 435 goto out; 436 if ((cops = q->ops->cl_ops) == NULL) 437 goto errout; 438 if (cops->tcf_chain == NULL) 439 goto errout; 440 if (TC_H_MIN(tcm->tcm_parent)) { 441 cl = cops->get(q, tcm->tcm_parent); 442 if (cl == 0) 443 goto errout; 444 } 445 chain = cops->tcf_chain(q, cl); 446 if (chain == NULL) 447 goto errout; 448 449 s_t = cb->args[0]; 450 451 for (tp=*chain, t=0; tp; tp = tp->next, t++) { 452 if (t < s_t) continue; 453 if (TC_H_MAJ(tcm->tcm_info) && 454 TC_H_MAJ(tcm->tcm_info) != tp->prio) 455 continue; 456 if (TC_H_MIN(tcm->tcm_info) && 457 TC_H_MIN(tcm->tcm_info) != tp->protocol) 458 continue; 459 if (t > s_t) 460 memset(&cb->args[1], 0, sizeof(cb->args)-sizeof(cb->args[0])); 461 if (cb->args[1] == 0) { 462 if (tcf_fill_node(skb, tp, 0, NETLINK_CB(cb->skb).pid, 463 cb->nlh->nlmsg_seq, NLM_F_MULTI, 464 RTM_NEWTFILTER) <= 0) 465 break; 466 467 cb->args[1] = 1; 468 } 469 if (tp->ops->walk == NULL) 470 continue; 471 arg.w.fn = tcf_node_dump; 472 arg.skb = skb; 473 arg.cb = cb; 474 arg.w.stop = 0; 475 arg.w.skip = cb->args[1]-1; 476 arg.w.count = 0; 477 tp->ops->walk(tp, &arg.w); 478 cb->args[1] = arg.w.count+1; 479 if (arg.w.stop) 480 break; 481 } 482 483 cb->args[0] = t; 484 485 errout: 486 if (cl) 487 cops->put(q, cl); 488 out: 489 return skb->len; 490 } 491 492 void tcf_exts_destroy(struct tcf_proto *tp, struct tcf_exts *exts) 493 { 494 #ifdef CONFIG_NET_CLS_ACT 495 if (exts->action) { 496 tcf_action_destroy(exts->action, TCA_ACT_UNBIND); 497 exts->action = NULL; 498 } 499 #endif 500 } 501 EXPORT_SYMBOL(tcf_exts_destroy); 502 503 int tcf_exts_validate(struct tcf_proto *tp, struct nlattr **tb, 504 struct nlattr *rate_tlv, struct tcf_exts *exts, 505 const struct tcf_ext_map *map) 506 { 507 memset(exts, 0, sizeof(*exts)); 508 509 #ifdef CONFIG_NET_CLS_ACT 510 { 511 struct tc_action *act; 512 513 if (map->police && tb[map->police]) { 514 act = tcf_action_init_1(tb[map->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 = TCA_OLD_COMPAT; 521 exts->action = act; 522 } else if (map->action && tb[map->action]) { 523 act = tcf_action_init(tb[map->action], rate_tlv, NULL, 524 TCA_ACT_NOREPLACE, TCA_ACT_BIND); 525 if (IS_ERR(act)) 526 return PTR_ERR(act); 527 528 exts->action = act; 529 } 530 } 531 #else 532 if ((map->action && tb[map->action]) || 533 (map->police && tb[map->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 (src->action) { 546 struct tc_action *act; 547 tcf_tree_lock(tp); 548 act = dst->action; 549 dst->action = src->action; 550 tcf_tree_unlock(tp); 551 if (act) 552 tcf_action_destroy(act, TCA_ACT_UNBIND); 553 } 554 #endif 555 } 556 EXPORT_SYMBOL(tcf_exts_change); 557 558 int tcf_exts_dump(struct sk_buff *skb, struct tcf_exts *exts, 559 const struct tcf_ext_map *map) 560 { 561 #ifdef CONFIG_NET_CLS_ACT 562 if (map->action && exts->action) { 563 /* 564 * again for backward compatible mode - we want 565 * to work with both old and new modes of entering 566 * tc data even if iproute2 was newer - jhs 567 */ 568 struct nlattr *nest; 569 570 if (exts->action->type != TCA_OLD_COMPAT) { 571 nest = nla_nest_start(skb, map->action); 572 if (nest == NULL) 573 goto nla_put_failure; 574 if (tcf_action_dump(skb, exts->action, 0, 0) < 0) 575 goto nla_put_failure; 576 nla_nest_end(skb, nest); 577 } else if (map->police) { 578 nest = nla_nest_start(skb, map->police); 579 if (nest == NULL) 580 goto nla_put_failure; 581 if (tcf_action_dump_old(skb, exts->action, 0, 0) < 0) 582 goto nla_put_failure; 583 nla_nest_end(skb, nest); 584 } 585 } 586 #endif 587 return 0; 588 nla_put_failure: __attribute__ ((unused)) 589 return -1; 590 } 591 EXPORT_SYMBOL(tcf_exts_dump); 592 593 594 int tcf_exts_dump_stats(struct sk_buff *skb, struct tcf_exts *exts, 595 const struct tcf_ext_map *map) 596 { 597 #ifdef CONFIG_NET_CLS_ACT 598 if (exts->action) 599 if (tcf_action_copy_stats(skb, exts->action, 1) < 0) 600 goto nla_put_failure; 601 #endif 602 return 0; 603 nla_put_failure: __attribute__ ((unused)) 604 return -1; 605 } 606 EXPORT_SYMBOL(tcf_exts_dump_stats); 607 608 static int __init tc_filter_init(void) 609 { 610 rtnl_register(PF_UNSPEC, RTM_NEWTFILTER, tc_ctl_tfilter, NULL); 611 rtnl_register(PF_UNSPEC, RTM_DELTFILTER, tc_ctl_tfilter, NULL); 612 rtnl_register(PF_UNSPEC, RTM_GETTFILTER, tc_ctl_tfilter, 613 tc_dump_tfilter); 614 615 return 0; 616 } 617 618 subsys_initcall(tc_filter_init); 619