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 <asm/uaccess.h> 18 #include <asm/system.h> 19 #include <linux/bitops.h> 20 #include <linux/module.h> 21 #include <linux/types.h> 22 #include <linux/kernel.h> 23 #include <linux/sched.h> 24 #include <linux/string.h> 25 #include <linux/mm.h> 26 #include <linux/socket.h> 27 #include <linux/sockios.h> 28 #include <linux/in.h> 29 #include <linux/errno.h> 30 #include <linux/interrupt.h> 31 #include <linux/netdevice.h> 32 #include <linux/skbuff.h> 33 #include <linux/rtnetlink.h> 34 #include <linux/init.h> 35 #include <linux/kmod.h> 36 #include <net/sock.h> 37 #include <net/pkt_sched.h> 38 #include <net/pkt_cls.h> 39 40 #if 0 /* control */ 41 #define DPRINTK(format,args...) printk(KERN_DEBUG format,##args) 42 #else 43 #define DPRINTK(format,args...) 44 #endif 45 46 /* The list of all installed classifier types */ 47 48 static struct tcf_proto_ops *tcf_proto_base; 49 50 /* Protects list of registered TC modules. It is pure SMP lock. */ 51 static DEFINE_RWLOCK(cls_mod_lock); 52 53 /* Find classifier type by string name */ 54 55 static struct tcf_proto_ops * tcf_proto_lookup_ops(struct rtattr *kind) 56 { 57 struct tcf_proto_ops *t = NULL; 58 59 if (kind) { 60 read_lock(&cls_mod_lock); 61 for (t = tcf_proto_base; t; t = t->next) { 62 if (rtattr_strcmp(kind, t->kind) == 0) { 63 if (!try_module_get(t->owner)) 64 t = NULL; 65 break; 66 } 67 } 68 read_unlock(&cls_mod_lock); 69 } 70 return t; 71 } 72 73 /* Register(unregister) new classifier type */ 74 75 int register_tcf_proto_ops(struct tcf_proto_ops *ops) 76 { 77 struct tcf_proto_ops *t, **tp; 78 int rc = -EEXIST; 79 80 write_lock(&cls_mod_lock); 81 for (tp = &tcf_proto_base; (t = *tp) != NULL; tp = &t->next) 82 if (!strcmp(ops->kind, t->kind)) 83 goto out; 84 85 ops->next = NULL; 86 *tp = ops; 87 rc = 0; 88 out: 89 write_unlock(&cls_mod_lock); 90 return rc; 91 } 92 93 int unregister_tcf_proto_ops(struct tcf_proto_ops *ops) 94 { 95 struct tcf_proto_ops *t, **tp; 96 int rc = -ENOENT; 97 98 write_lock(&cls_mod_lock); 99 for (tp = &tcf_proto_base; (t=*tp) != NULL; tp = &t->next) 100 if (t == ops) 101 break; 102 103 if (!t) 104 goto out; 105 *tp = t->next; 106 rc = 0; 107 out: 108 write_unlock(&cls_mod_lock); 109 return rc; 110 } 111 112 static int tfilter_notify(struct sk_buff *oskb, struct nlmsghdr *n, 113 struct tcf_proto *tp, unsigned long fh, int event); 114 115 116 /* Select new prio value from the range, managed by kernel. */ 117 118 static __inline__ u32 tcf_auto_prio(struct tcf_proto *tp) 119 { 120 u32 first = TC_H_MAKE(0xC0000000U,0U); 121 122 if (tp) 123 first = tp->prio-1; 124 125 return first; 126 } 127 128 /* Add/change/delete/get a filter node */ 129 130 static int tc_ctl_tfilter(struct sk_buff *skb, struct nlmsghdr *n, void *arg) 131 { 132 struct rtattr **tca; 133 struct tcmsg *t; 134 u32 protocol; 135 u32 prio; 136 u32 nprio; 137 u32 parent; 138 struct net_device *dev; 139 struct Qdisc *q; 140 struct tcf_proto **back, **chain; 141 struct tcf_proto *tp; 142 struct tcf_proto_ops *tp_ops; 143 struct Qdisc_class_ops *cops; 144 unsigned long cl; 145 unsigned long fh; 146 int err; 147 148 replay: 149 tca = arg; 150 t = NLMSG_DATA(n); 151 protocol = TC_H_MIN(t->tcm_info); 152 prio = TC_H_MAJ(t->tcm_info); 153 nprio = prio; 154 parent = t->tcm_parent; 155 cl = 0; 156 157 if (prio == 0) { 158 /* If no priority is given, user wants we allocated it. */ 159 if (n->nlmsg_type != RTM_NEWTFILTER || !(n->nlmsg_flags&NLM_F_CREATE)) 160 return -ENOENT; 161 prio = TC_H_MAKE(0x80000000U,0U); 162 } 163 164 /* Find head of filter chain. */ 165 166 /* Find link */ 167 if ((dev = __dev_get_by_index(t->tcm_ifindex)) == NULL) 168 return -ENODEV; 169 170 /* Find qdisc */ 171 if (!parent) { 172 q = dev->qdisc_sleeping; 173 parent = q->handle; 174 } else if ((q = qdisc_lookup(dev, TC_H_MAJ(t->tcm_parent))) == NULL) 175 return -EINVAL; 176 177 /* Is it classful? */ 178 if ((cops = q->ops->cl_ops) == NULL) 179 return -EINVAL; 180 181 /* Do we search for filter, attached to class? */ 182 if (TC_H_MIN(parent)) { 183 cl = cops->get(q, parent); 184 if (cl == 0) 185 return -ENOENT; 186 } 187 188 /* And the last stroke */ 189 chain = cops->tcf_chain(q, cl); 190 err = -EINVAL; 191 if (chain == NULL) 192 goto errout; 193 194 /* Check the chain for existence of proto-tcf with this priority */ 195 for (back = chain; (tp=*back) != NULL; back = &tp->next) { 196 if (tp->prio >= prio) { 197 if (tp->prio == prio) { 198 if (!nprio || (tp->protocol != protocol && protocol)) 199 goto errout; 200 } else 201 tp = NULL; 202 break; 203 } 204 } 205 206 if (tp == NULL) { 207 /* Proto-tcf does not exist, create new one */ 208 209 if (tca[TCA_KIND-1] == NULL || !protocol) 210 goto errout; 211 212 err = -ENOENT; 213 if (n->nlmsg_type != RTM_NEWTFILTER || !(n->nlmsg_flags&NLM_F_CREATE)) 214 goto errout; 215 216 217 /* Create new proto tcf */ 218 219 err = -ENOBUFS; 220 if ((tp = kmalloc(sizeof(*tp), GFP_KERNEL)) == NULL) 221 goto errout; 222 err = -EINVAL; 223 tp_ops = tcf_proto_lookup_ops(tca[TCA_KIND-1]); 224 if (tp_ops == NULL) { 225 #ifdef CONFIG_KMOD 226 struct rtattr *kind = tca[TCA_KIND-1]; 227 char name[IFNAMSIZ]; 228 229 if (kind != NULL && 230 rtattr_strlcpy(name, kind, IFNAMSIZ) < IFNAMSIZ) { 231 rtnl_unlock(); 232 request_module("cls_%s", name); 233 rtnl_lock(); 234 tp_ops = tcf_proto_lookup_ops(kind); 235 /* We dropped the RTNL semaphore in order to 236 * perform the module load. So, even if we 237 * succeeded in loading the module we have to 238 * replay the request. We indicate this using 239 * -EAGAIN. 240 */ 241 if (tp_ops != NULL) { 242 module_put(tp_ops->owner); 243 err = -EAGAIN; 244 } 245 } 246 #endif 247 kfree(tp); 248 goto errout; 249 } 250 memset(tp, 0, sizeof(*tp)); 251 tp->ops = tp_ops; 252 tp->protocol = protocol; 253 tp->prio = nprio ? : tcf_auto_prio(*back); 254 tp->q = q; 255 tp->classify = tp_ops->classify; 256 tp->classid = parent; 257 if ((err = tp_ops->init(tp)) != 0) { 258 module_put(tp_ops->owner); 259 kfree(tp); 260 goto errout; 261 } 262 263 qdisc_lock_tree(dev); 264 tp->next = *back; 265 *back = tp; 266 qdisc_unlock_tree(dev); 267 268 } else if (tca[TCA_KIND-1] && rtattr_strcmp(tca[TCA_KIND-1], tp->ops->kind)) 269 goto errout; 270 271 fh = tp->ops->get(tp, t->tcm_handle); 272 273 if (fh == 0) { 274 if (n->nlmsg_type == RTM_DELTFILTER && t->tcm_handle == 0) { 275 qdisc_lock_tree(dev); 276 *back = tp->next; 277 qdisc_unlock_tree(dev); 278 279 tfilter_notify(skb, n, tp, fh, RTM_DELTFILTER); 280 tcf_destroy(tp); 281 err = 0; 282 goto errout; 283 } 284 285 err = -ENOENT; 286 if (n->nlmsg_type != RTM_NEWTFILTER || !(n->nlmsg_flags&NLM_F_CREATE)) 287 goto errout; 288 } else { 289 switch (n->nlmsg_type) { 290 case RTM_NEWTFILTER: 291 err = -EEXIST; 292 if (n->nlmsg_flags&NLM_F_EXCL) 293 goto errout; 294 break; 295 case RTM_DELTFILTER: 296 err = tp->ops->delete(tp, fh); 297 if (err == 0) 298 tfilter_notify(skb, n, tp, fh, RTM_DELTFILTER); 299 goto errout; 300 case RTM_GETTFILTER: 301 err = tfilter_notify(skb, n, tp, fh, RTM_NEWTFILTER); 302 goto errout; 303 default: 304 err = -EINVAL; 305 goto errout; 306 } 307 } 308 309 err = tp->ops->change(tp, cl, t->tcm_handle, tca, &fh); 310 if (err == 0) 311 tfilter_notify(skb, n, tp, fh, RTM_NEWTFILTER); 312 313 errout: 314 if (cl) 315 cops->put(q, cl); 316 if (err == -EAGAIN) 317 /* Replay the request. */ 318 goto replay; 319 return err; 320 } 321 322 static int 323 tcf_fill_node(struct sk_buff *skb, struct tcf_proto *tp, unsigned long fh, 324 u32 pid, u32 seq, u16 flags, int event) 325 { 326 struct tcmsg *tcm; 327 struct nlmsghdr *nlh; 328 unsigned char *b = skb->tail; 329 330 nlh = NLMSG_NEW(skb, pid, seq, event, sizeof(*tcm), flags); 331 tcm = NLMSG_DATA(nlh); 332 tcm->tcm_family = AF_UNSPEC; 333 tcm->tcm__pad1 = 0; 334 tcm->tcm__pad1 = 0; 335 tcm->tcm_ifindex = tp->q->dev->ifindex; 336 tcm->tcm_parent = tp->classid; 337 tcm->tcm_info = TC_H_MAKE(tp->prio, tp->protocol); 338 RTA_PUT(skb, TCA_KIND, IFNAMSIZ, tp->ops->kind); 339 tcm->tcm_handle = fh; 340 if (RTM_DELTFILTER != event) { 341 tcm->tcm_handle = 0; 342 if (tp->ops->dump && tp->ops->dump(tp, fh, skb, tcm) < 0) 343 goto rtattr_failure; 344 } 345 nlh->nlmsg_len = skb->tail - b; 346 return skb->len; 347 348 nlmsg_failure: 349 rtattr_failure: 350 skb_trim(skb, b - skb->data); 351 return -1; 352 } 353 354 static int tfilter_notify(struct sk_buff *oskb, struct nlmsghdr *n, 355 struct tcf_proto *tp, unsigned long fh, int event) 356 { 357 struct sk_buff *skb; 358 u32 pid = oskb ? NETLINK_CB(oskb).pid : 0; 359 360 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL); 361 if (!skb) 362 return -ENOBUFS; 363 364 if (tcf_fill_node(skb, tp, fh, pid, n->nlmsg_seq, 0, event) <= 0) { 365 kfree_skb(skb); 366 return -EINVAL; 367 } 368 369 return rtnetlink_send(skb, pid, RTNLGRP_TC, n->nlmsg_flags&NLM_F_ECHO); 370 } 371 372 struct tcf_dump_args 373 { 374 struct tcf_walker w; 375 struct sk_buff *skb; 376 struct netlink_callback *cb; 377 }; 378 379 static int tcf_node_dump(struct tcf_proto *tp, unsigned long n, struct tcf_walker *arg) 380 { 381 struct tcf_dump_args *a = (void*)arg; 382 383 return tcf_fill_node(a->skb, tp, n, NETLINK_CB(a->cb->skb).pid, 384 a->cb->nlh->nlmsg_seq, NLM_F_MULTI, RTM_NEWTFILTER); 385 } 386 387 static int tc_dump_tfilter(struct sk_buff *skb, struct netlink_callback *cb) 388 { 389 int t; 390 int s_t; 391 struct net_device *dev; 392 struct Qdisc *q; 393 struct tcf_proto *tp, **chain; 394 struct tcmsg *tcm = (struct tcmsg*)NLMSG_DATA(cb->nlh); 395 unsigned long cl = 0; 396 struct Qdisc_class_ops *cops; 397 struct tcf_dump_args arg; 398 399 if (cb->nlh->nlmsg_len < NLMSG_LENGTH(sizeof(*tcm))) 400 return skb->len; 401 if ((dev = dev_get_by_index(tcm->tcm_ifindex)) == NULL) 402 return skb->len; 403 404 read_lock_bh(&qdisc_tree_lock); 405 if (!tcm->tcm_parent) 406 q = dev->qdisc_sleeping; 407 else 408 q = qdisc_lookup(dev, TC_H_MAJ(tcm->tcm_parent)); 409 if (!q) 410 goto out; 411 if ((cops = q->ops->cl_ops) == NULL) 412 goto errout; 413 if (TC_H_MIN(tcm->tcm_parent)) { 414 cl = cops->get(q, tcm->tcm_parent); 415 if (cl == 0) 416 goto errout; 417 } 418 chain = cops->tcf_chain(q, cl); 419 if (chain == NULL) 420 goto errout; 421 422 s_t = cb->args[0]; 423 424 for (tp=*chain, t=0; tp; tp = tp->next, t++) { 425 if (t < s_t) continue; 426 if (TC_H_MAJ(tcm->tcm_info) && 427 TC_H_MAJ(tcm->tcm_info) != tp->prio) 428 continue; 429 if (TC_H_MIN(tcm->tcm_info) && 430 TC_H_MIN(tcm->tcm_info) != tp->protocol) 431 continue; 432 if (t > s_t) 433 memset(&cb->args[1], 0, sizeof(cb->args)-sizeof(cb->args[0])); 434 if (cb->args[1] == 0) { 435 if (tcf_fill_node(skb, tp, 0, NETLINK_CB(cb->skb).pid, 436 cb->nlh->nlmsg_seq, NLM_F_MULTI, RTM_NEWTFILTER) <= 0) { 437 break; 438 } 439 cb->args[1] = 1; 440 } 441 if (tp->ops->walk == NULL) 442 continue; 443 arg.w.fn = tcf_node_dump; 444 arg.skb = skb; 445 arg.cb = cb; 446 arg.w.stop = 0; 447 arg.w.skip = cb->args[1]-1; 448 arg.w.count = 0; 449 tp->ops->walk(tp, &arg.w); 450 cb->args[1] = arg.w.count+1; 451 if (arg.w.stop) 452 break; 453 } 454 455 cb->args[0] = t; 456 457 errout: 458 if (cl) 459 cops->put(q, cl); 460 out: 461 read_unlock_bh(&qdisc_tree_lock); 462 dev_put(dev); 463 return skb->len; 464 } 465 466 void 467 tcf_exts_destroy(struct tcf_proto *tp, struct tcf_exts *exts) 468 { 469 #ifdef CONFIG_NET_CLS_ACT 470 if (exts->action) { 471 tcf_action_destroy(exts->action, TCA_ACT_UNBIND); 472 exts->action = NULL; 473 } 474 #elif defined CONFIG_NET_CLS_POLICE 475 if (exts->police) { 476 tcf_police_release(exts->police, TCA_ACT_UNBIND); 477 exts->police = NULL; 478 } 479 #endif 480 } 481 482 483 int 484 tcf_exts_validate(struct tcf_proto *tp, struct rtattr **tb, 485 struct rtattr *rate_tlv, struct tcf_exts *exts, 486 struct tcf_ext_map *map) 487 { 488 memset(exts, 0, sizeof(*exts)); 489 490 #ifdef CONFIG_NET_CLS_ACT 491 { 492 int err; 493 struct tc_action *act; 494 495 if (map->police && tb[map->police-1]) { 496 act = tcf_action_init_1(tb[map->police-1], rate_tlv, "police", 497 TCA_ACT_NOREPLACE, TCA_ACT_BIND, &err); 498 if (act == NULL) 499 return err; 500 501 act->type = TCA_OLD_COMPAT; 502 exts->action = act; 503 } else if (map->action && tb[map->action-1]) { 504 act = tcf_action_init(tb[map->action-1], rate_tlv, NULL, 505 TCA_ACT_NOREPLACE, TCA_ACT_BIND, &err); 506 if (act == NULL) 507 return err; 508 509 exts->action = act; 510 } 511 } 512 #elif defined CONFIG_NET_CLS_POLICE 513 if (map->police && tb[map->police-1]) { 514 struct tcf_police *p; 515 516 p = tcf_police_locate(tb[map->police-1], rate_tlv); 517 if (p == NULL) 518 return -EINVAL; 519 520 exts->police = p; 521 } else if (map->action && tb[map->action-1]) 522 return -EOPNOTSUPP; 523 #else 524 if ((map->action && tb[map->action-1]) || 525 (map->police && tb[map->police-1])) 526 return -EOPNOTSUPP; 527 #endif 528 529 return 0; 530 } 531 532 void 533 tcf_exts_change(struct tcf_proto *tp, struct tcf_exts *dst, 534 struct tcf_exts *src) 535 { 536 #ifdef CONFIG_NET_CLS_ACT 537 if (src->action) { 538 struct tc_action *act; 539 tcf_tree_lock(tp); 540 act = xchg(&dst->action, src->action); 541 tcf_tree_unlock(tp); 542 if (act) 543 tcf_action_destroy(act, TCA_ACT_UNBIND); 544 } 545 #elif defined CONFIG_NET_CLS_POLICE 546 if (src->police) { 547 struct tcf_police *p; 548 tcf_tree_lock(tp); 549 p = xchg(&dst->police, src->police); 550 tcf_tree_unlock(tp); 551 if (p) 552 tcf_police_release(p, TCA_ACT_UNBIND); 553 } 554 #endif 555 } 556 557 int 558 tcf_exts_dump(struct sk_buff *skb, struct tcf_exts *exts, 559 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 rtattr * p_rta = (struct rtattr*) skb->tail; 569 570 if (exts->action->type != TCA_OLD_COMPAT) { 571 RTA_PUT(skb, map->action, 0, NULL); 572 if (tcf_action_dump(skb, exts->action, 0, 0) < 0) 573 goto rtattr_failure; 574 p_rta->rta_len = skb->tail - (u8*)p_rta; 575 } else if (map->police) { 576 RTA_PUT(skb, map->police, 0, NULL); 577 if (tcf_action_dump_old(skb, exts->action, 0, 0) < 0) 578 goto rtattr_failure; 579 p_rta->rta_len = skb->tail - (u8*)p_rta; 580 } 581 } 582 #elif defined CONFIG_NET_CLS_POLICE 583 if (map->police && exts->police) { 584 struct rtattr * p_rta = (struct rtattr*) skb->tail; 585 586 RTA_PUT(skb, map->police, 0, NULL); 587 588 if (tcf_police_dump(skb, exts->police) < 0) 589 goto rtattr_failure; 590 591 p_rta->rta_len = skb->tail - (u8*)p_rta; 592 } 593 #endif 594 return 0; 595 rtattr_failure: __attribute__ ((unused)) 596 return -1; 597 } 598 599 int 600 tcf_exts_dump_stats(struct sk_buff *skb, struct tcf_exts *exts, 601 struct tcf_ext_map *map) 602 { 603 #ifdef CONFIG_NET_CLS_ACT 604 if (exts->action) 605 if (tcf_action_copy_stats(skb, exts->action, 1) < 0) 606 goto rtattr_failure; 607 #elif defined CONFIG_NET_CLS_POLICE 608 if (exts->police) 609 if (tcf_police_dump_stats(skb, exts->police) < 0) 610 goto rtattr_failure; 611 #endif 612 return 0; 613 rtattr_failure: __attribute__ ((unused)) 614 return -1; 615 } 616 617 static int __init tc_filter_init(void) 618 { 619 struct rtnetlink_link *link_p = rtnetlink_links[PF_UNSPEC]; 620 621 /* Setup rtnetlink links. It is made here to avoid 622 exporting large number of public symbols. 623 */ 624 625 if (link_p) { 626 link_p[RTM_NEWTFILTER-RTM_BASE].doit = tc_ctl_tfilter; 627 link_p[RTM_DELTFILTER-RTM_BASE].doit = tc_ctl_tfilter; 628 link_p[RTM_GETTFILTER-RTM_BASE].doit = tc_ctl_tfilter; 629 link_p[RTM_GETTFILTER-RTM_BASE].dumpit = tc_dump_tfilter; 630 } 631 return 0; 632 } 633 634 subsys_initcall(tc_filter_init); 635 636 EXPORT_SYMBOL(register_tcf_proto_ops); 637 EXPORT_SYMBOL(unregister_tcf_proto_ops); 638 EXPORT_SYMBOL(tcf_exts_validate); 639 EXPORT_SYMBOL(tcf_exts_destroy); 640 EXPORT_SYMBOL(tcf_exts_change); 641 EXPORT_SYMBOL(tcf_exts_dump); 642 EXPORT_SYMBOL(tcf_exts_dump_stats); 643