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/string.h> 24 #include <linux/mm.h> 25 #include <linux/socket.h> 26 #include <linux/sockios.h> 27 #include <linux/in.h> 28 #include <linux/errno.h> 29 #include <linux/interrupt.h> 30 #include <linux/netdevice.h> 31 #include <linux/skbuff.h> 32 #include <linux/rtnetlink.h> 33 #include <linux/init.h> 34 #include <linux/kmod.h> 35 #include <net/sock.h> 36 #include <net/pkt_sched.h> 37 #include <net/pkt_cls.h> 38 39 #if 0 /* control */ 40 #define DPRINTK(format,args...) printk(KERN_DEBUG format,##args) 41 #else 42 #define DPRINTK(format,args...) 43 #endif 44 45 /* The list of all installed classifier types */ 46 47 static struct tcf_proto_ops *tcf_proto_base; 48 49 /* Protects list of registered TC modules. It is pure SMP lock. */ 50 static DEFINE_RWLOCK(cls_mod_lock); 51 52 /* Find classifier type by string name */ 53 54 static struct tcf_proto_ops * tcf_proto_lookup_ops(struct rtattr *kind) 55 { 56 struct tcf_proto_ops *t = NULL; 57 58 if (kind) { 59 read_lock(&cls_mod_lock); 60 for (t = tcf_proto_base; t; t = t->next) { 61 if (rtattr_strcmp(kind, t->kind) == 0) { 62 if (!try_module_get(t->owner)) 63 t = NULL; 64 break; 65 } 66 } 67 read_unlock(&cls_mod_lock); 68 } 69 return t; 70 } 71 72 /* Register(unregister) new classifier type */ 73 74 int register_tcf_proto_ops(struct tcf_proto_ops *ops) 75 { 76 struct tcf_proto_ops *t, **tp; 77 int rc = -EEXIST; 78 79 write_lock(&cls_mod_lock); 80 for (tp = &tcf_proto_base; (t = *tp) != NULL; tp = &t->next) 81 if (!strcmp(ops->kind, t->kind)) 82 goto out; 83 84 ops->next = NULL; 85 *tp = ops; 86 rc = 0; 87 out: 88 write_unlock(&cls_mod_lock); 89 return rc; 90 } 91 92 int unregister_tcf_proto_ops(struct tcf_proto_ops *ops) 93 { 94 struct tcf_proto_ops *t, **tp; 95 int rc = -ENOENT; 96 97 write_lock(&cls_mod_lock); 98 for (tp = &tcf_proto_base; (t=*tp) != NULL; tp = &t->next) 99 if (t == ops) 100 break; 101 102 if (!t) 103 goto out; 104 *tp = t->next; 105 rc = 0; 106 out: 107 write_unlock(&cls_mod_lock); 108 return rc; 109 } 110 111 static int tfilter_notify(struct sk_buff *oskb, struct nlmsghdr *n, 112 struct tcf_proto *tp, unsigned long fh, int event); 113 114 115 /* Select new prio value from the range, managed by kernel. */ 116 117 static __inline__ u32 tcf_auto_prio(struct tcf_proto *tp) 118 { 119 u32 first = TC_H_MAKE(0xC0000000U,0U); 120 121 if (tp) 122 first = tp->prio-1; 123 124 return first; 125 } 126 127 /* Add/change/delete/get a filter node */ 128 129 static int tc_ctl_tfilter(struct sk_buff *skb, struct nlmsghdr *n, void *arg) 130 { 131 struct rtattr **tca; 132 struct tcmsg *t; 133 u32 protocol; 134 u32 prio; 135 u32 nprio; 136 u32 parent; 137 struct net_device *dev; 138 struct Qdisc *q; 139 struct tcf_proto **back, **chain; 140 struct tcf_proto *tp; 141 struct tcf_proto_ops *tp_ops; 142 struct Qdisc_class_ops *cops; 143 unsigned long cl; 144 unsigned long fh; 145 int err; 146 147 replay: 148 tca = arg; 149 t = NLMSG_DATA(n); 150 protocol = TC_H_MIN(t->tcm_info); 151 prio = TC_H_MAJ(t->tcm_info); 152 nprio = prio; 153 parent = t->tcm_parent; 154 cl = 0; 155 156 if (prio == 0) { 157 /* If no priority is given, user wants we allocated it. */ 158 if (n->nlmsg_type != RTM_NEWTFILTER || !(n->nlmsg_flags&NLM_F_CREATE)) 159 return -ENOENT; 160 prio = TC_H_MAKE(0x80000000U,0U); 161 } 162 163 /* Find head of filter chain. */ 164 165 /* Find link */ 166 if ((dev = __dev_get_by_index(t->tcm_ifindex)) == NULL) 167 return -ENODEV; 168 169 /* Find qdisc */ 170 if (!parent) { 171 q = dev->qdisc_sleeping; 172 parent = q->handle; 173 } else if ((q = qdisc_lookup(dev, TC_H_MAJ(t->tcm_parent))) == NULL) 174 return -EINVAL; 175 176 /* Is it classful? */ 177 if ((cops = q->ops->cl_ops) == NULL) 178 return -EINVAL; 179 180 /* Do we search for filter, attached to class? */ 181 if (TC_H_MIN(parent)) { 182 cl = cops->get(q, parent); 183 if (cl == 0) 184 return -ENOENT; 185 } 186 187 /* And the last stroke */ 188 chain = cops->tcf_chain(q, cl); 189 err = -EINVAL; 190 if (chain == NULL) 191 goto errout; 192 193 /* Check the chain for existence of proto-tcf with this priority */ 194 for (back = chain; (tp=*back) != NULL; back = &tp->next) { 195 if (tp->prio >= prio) { 196 if (tp->prio == prio) { 197 if (!nprio || (tp->protocol != protocol && protocol)) 198 goto errout; 199 } else 200 tp = NULL; 201 break; 202 } 203 } 204 205 if (tp == NULL) { 206 /* Proto-tcf does not exist, create new one */ 207 208 if (tca[TCA_KIND-1] == NULL || !protocol) 209 goto errout; 210 211 err = -ENOENT; 212 if (n->nlmsg_type != RTM_NEWTFILTER || !(n->nlmsg_flags&NLM_F_CREATE)) 213 goto errout; 214 215 216 /* Create new proto tcf */ 217 218 err = -ENOBUFS; 219 if ((tp = kzalloc(sizeof(*tp), GFP_KERNEL)) == NULL) 220 goto errout; 221 err = -EINVAL; 222 tp_ops = tcf_proto_lookup_ops(tca[TCA_KIND-1]); 223 if (tp_ops == NULL) { 224 #ifdef CONFIG_KMOD 225 struct rtattr *kind = tca[TCA_KIND-1]; 226 char name[IFNAMSIZ]; 227 228 if (kind != NULL && 229 rtattr_strlcpy(name, kind, IFNAMSIZ) < IFNAMSIZ) { 230 rtnl_unlock(); 231 request_module("cls_%s", name); 232 rtnl_lock(); 233 tp_ops = tcf_proto_lookup_ops(kind); 234 /* We dropped the RTNL semaphore in order to 235 * perform the module load. So, even if we 236 * succeeded in loading the module we have to 237 * replay the request. We indicate this using 238 * -EAGAIN. 239 */ 240 if (tp_ops != NULL) { 241 module_put(tp_ops->owner); 242 err = -EAGAIN; 243 } 244 } 245 #endif 246 kfree(tp); 247 goto errout; 248 } 249 tp->ops = tp_ops; 250 tp->protocol = protocol; 251 tp->prio = nprio ? : tcf_auto_prio(*back); 252 tp->q = q; 253 tp->classify = tp_ops->classify; 254 tp->classid = parent; 255 if ((err = tp_ops->init(tp)) != 0) { 256 module_put(tp_ops->owner); 257 kfree(tp); 258 goto errout; 259 } 260 261 qdisc_lock_tree(dev); 262 tp->next = *back; 263 *back = tp; 264 qdisc_unlock_tree(dev); 265 266 } else if (tca[TCA_KIND-1] && rtattr_strcmp(tca[TCA_KIND-1], tp->ops->kind)) 267 goto errout; 268 269 fh = tp->ops->get(tp, t->tcm_handle); 270 271 if (fh == 0) { 272 if (n->nlmsg_type == RTM_DELTFILTER && t->tcm_handle == 0) { 273 qdisc_lock_tree(dev); 274 *back = tp->next; 275 qdisc_unlock_tree(dev); 276 277 tfilter_notify(skb, n, tp, fh, RTM_DELTFILTER); 278 tcf_destroy(tp); 279 err = 0; 280 goto errout; 281 } 282 283 err = -ENOENT; 284 if (n->nlmsg_type != RTM_NEWTFILTER || !(n->nlmsg_flags&NLM_F_CREATE)) 285 goto errout; 286 } else { 287 switch (n->nlmsg_type) { 288 case RTM_NEWTFILTER: 289 err = -EEXIST; 290 if (n->nlmsg_flags&NLM_F_EXCL) 291 goto errout; 292 break; 293 case RTM_DELTFILTER: 294 err = tp->ops->delete(tp, fh); 295 if (err == 0) 296 tfilter_notify(skb, n, tp, fh, RTM_DELTFILTER); 297 goto errout; 298 case RTM_GETTFILTER: 299 err = tfilter_notify(skb, n, tp, fh, RTM_NEWTFILTER); 300 goto errout; 301 default: 302 err = -EINVAL; 303 goto errout; 304 } 305 } 306 307 err = tp->ops->change(tp, cl, t->tcm_handle, tca, &fh); 308 if (err == 0) 309 tfilter_notify(skb, n, tp, fh, RTM_NEWTFILTER); 310 311 errout: 312 if (cl) 313 cops->put(q, cl); 314 if (err == -EAGAIN) 315 /* Replay the request. */ 316 goto replay; 317 return err; 318 } 319 320 static int 321 tcf_fill_node(struct sk_buff *skb, struct tcf_proto *tp, unsigned long fh, 322 u32 pid, u32 seq, u16 flags, int event) 323 { 324 struct tcmsg *tcm; 325 struct nlmsghdr *nlh; 326 unsigned char *b = skb->tail; 327 328 nlh = NLMSG_NEW(skb, pid, seq, event, sizeof(*tcm), flags); 329 tcm = NLMSG_DATA(nlh); 330 tcm->tcm_family = AF_UNSPEC; 331 tcm->tcm__pad1 = 0; 332 tcm->tcm__pad1 = 0; 333 tcm->tcm_ifindex = tp->q->dev->ifindex; 334 tcm->tcm_parent = tp->classid; 335 tcm->tcm_info = TC_H_MAKE(tp->prio, tp->protocol); 336 RTA_PUT(skb, TCA_KIND, IFNAMSIZ, tp->ops->kind); 337 tcm->tcm_handle = fh; 338 if (RTM_DELTFILTER != event) { 339 tcm->tcm_handle = 0; 340 if (tp->ops->dump && tp->ops->dump(tp, fh, skb, tcm) < 0) 341 goto rtattr_failure; 342 } 343 nlh->nlmsg_len = skb->tail - b; 344 return skb->len; 345 346 nlmsg_failure: 347 rtattr_failure: 348 skb_trim(skb, b - skb->data); 349 return -1; 350 } 351 352 static int tfilter_notify(struct sk_buff *oskb, struct nlmsghdr *n, 353 struct tcf_proto *tp, unsigned long fh, int event) 354 { 355 struct sk_buff *skb; 356 u32 pid = oskb ? NETLINK_CB(oskb).pid : 0; 357 358 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL); 359 if (!skb) 360 return -ENOBUFS; 361 362 if (tcf_fill_node(skb, tp, fh, pid, n->nlmsg_seq, 0, event) <= 0) { 363 kfree_skb(skb); 364 return -EINVAL; 365 } 366 367 return rtnetlink_send(skb, pid, RTNLGRP_TC, n->nlmsg_flags&NLM_F_ECHO); 368 } 369 370 struct tcf_dump_args 371 { 372 struct tcf_walker w; 373 struct sk_buff *skb; 374 struct netlink_callback *cb; 375 }; 376 377 static int tcf_node_dump(struct tcf_proto *tp, unsigned long n, struct tcf_walker *arg) 378 { 379 struct tcf_dump_args *a = (void*)arg; 380 381 return tcf_fill_node(a->skb, tp, n, NETLINK_CB(a->cb->skb).pid, 382 a->cb->nlh->nlmsg_seq, NLM_F_MULTI, RTM_NEWTFILTER); 383 } 384 385 static int tc_dump_tfilter(struct sk_buff *skb, struct netlink_callback *cb) 386 { 387 int t; 388 int s_t; 389 struct net_device *dev; 390 struct Qdisc *q; 391 struct tcf_proto *tp, **chain; 392 struct tcmsg *tcm = (struct tcmsg*)NLMSG_DATA(cb->nlh); 393 unsigned long cl = 0; 394 struct Qdisc_class_ops *cops; 395 struct tcf_dump_args arg; 396 397 if (cb->nlh->nlmsg_len < NLMSG_LENGTH(sizeof(*tcm))) 398 return skb->len; 399 if ((dev = dev_get_by_index(tcm->tcm_ifindex)) == NULL) 400 return skb->len; 401 402 read_lock(&qdisc_tree_lock); 403 if (!tcm->tcm_parent) 404 q = dev->qdisc_sleeping; 405 else 406 q = qdisc_lookup(dev, TC_H_MAJ(tcm->tcm_parent)); 407 if (!q) 408 goto out; 409 if ((cops = q->ops->cl_ops) == NULL) 410 goto errout; 411 if (TC_H_MIN(tcm->tcm_parent)) { 412 cl = cops->get(q, tcm->tcm_parent); 413 if (cl == 0) 414 goto errout; 415 } 416 chain = cops->tcf_chain(q, cl); 417 if (chain == NULL) 418 goto errout; 419 420 s_t = cb->args[0]; 421 422 for (tp=*chain, t=0; tp; tp = tp->next, t++) { 423 if (t < s_t) continue; 424 if (TC_H_MAJ(tcm->tcm_info) && 425 TC_H_MAJ(tcm->tcm_info) != tp->prio) 426 continue; 427 if (TC_H_MIN(tcm->tcm_info) && 428 TC_H_MIN(tcm->tcm_info) != tp->protocol) 429 continue; 430 if (t > s_t) 431 memset(&cb->args[1], 0, sizeof(cb->args)-sizeof(cb->args[0])); 432 if (cb->args[1] == 0) { 433 if (tcf_fill_node(skb, tp, 0, NETLINK_CB(cb->skb).pid, 434 cb->nlh->nlmsg_seq, NLM_F_MULTI, RTM_NEWTFILTER) <= 0) { 435 break; 436 } 437 cb->args[1] = 1; 438 } 439 if (tp->ops->walk == NULL) 440 continue; 441 arg.w.fn = tcf_node_dump; 442 arg.skb = skb; 443 arg.cb = cb; 444 arg.w.stop = 0; 445 arg.w.skip = cb->args[1]-1; 446 arg.w.count = 0; 447 tp->ops->walk(tp, &arg.w); 448 cb->args[1] = arg.w.count+1; 449 if (arg.w.stop) 450 break; 451 } 452 453 cb->args[0] = t; 454 455 errout: 456 if (cl) 457 cops->put(q, cl); 458 out: 459 read_unlock(&qdisc_tree_lock); 460 dev_put(dev); 461 return skb->len; 462 } 463 464 void 465 tcf_exts_destroy(struct tcf_proto *tp, struct tcf_exts *exts) 466 { 467 #ifdef CONFIG_NET_CLS_ACT 468 if (exts->action) { 469 tcf_action_destroy(exts->action, TCA_ACT_UNBIND); 470 exts->action = NULL; 471 } 472 #elif defined CONFIG_NET_CLS_POLICE 473 if (exts->police) { 474 tcf_police_release(exts->police, TCA_ACT_UNBIND); 475 exts->police = NULL; 476 } 477 #endif 478 } 479 480 481 int 482 tcf_exts_validate(struct tcf_proto *tp, struct rtattr **tb, 483 struct rtattr *rate_tlv, struct tcf_exts *exts, 484 struct tcf_ext_map *map) 485 { 486 memset(exts, 0, sizeof(*exts)); 487 488 #ifdef CONFIG_NET_CLS_ACT 489 { 490 int err; 491 struct tc_action *act; 492 493 if (map->police && tb[map->police-1]) { 494 act = tcf_action_init_1(tb[map->police-1], rate_tlv, "police", 495 TCA_ACT_NOREPLACE, TCA_ACT_BIND, &err); 496 if (act == NULL) 497 return err; 498 499 act->type = TCA_OLD_COMPAT; 500 exts->action = act; 501 } else if (map->action && tb[map->action-1]) { 502 act = tcf_action_init(tb[map->action-1], rate_tlv, NULL, 503 TCA_ACT_NOREPLACE, TCA_ACT_BIND, &err); 504 if (act == NULL) 505 return err; 506 507 exts->action = act; 508 } 509 } 510 #elif defined CONFIG_NET_CLS_POLICE 511 if (map->police && tb[map->police-1]) { 512 struct tcf_police *p; 513 514 p = tcf_police_locate(tb[map->police-1], rate_tlv); 515 if (p == NULL) 516 return -EINVAL; 517 518 exts->police = p; 519 } else if (map->action && tb[map->action-1]) 520 return -EOPNOTSUPP; 521 #else 522 if ((map->action && tb[map->action-1]) || 523 (map->police && tb[map->police-1])) 524 return -EOPNOTSUPP; 525 #endif 526 527 return 0; 528 } 529 530 void 531 tcf_exts_change(struct tcf_proto *tp, struct tcf_exts *dst, 532 struct tcf_exts *src) 533 { 534 #ifdef CONFIG_NET_CLS_ACT 535 if (src->action) { 536 struct tc_action *act; 537 tcf_tree_lock(tp); 538 act = xchg(&dst->action, src->action); 539 tcf_tree_unlock(tp); 540 if (act) 541 tcf_action_destroy(act, TCA_ACT_UNBIND); 542 } 543 #elif defined CONFIG_NET_CLS_POLICE 544 if (src->police) { 545 struct tcf_police *p; 546 tcf_tree_lock(tp); 547 p = xchg(&dst->police, src->police); 548 tcf_tree_unlock(tp); 549 if (p) 550 tcf_police_release(p, TCA_ACT_UNBIND); 551 } 552 #endif 553 } 554 555 int 556 tcf_exts_dump(struct sk_buff *skb, struct tcf_exts *exts, 557 struct tcf_ext_map *map) 558 { 559 #ifdef CONFIG_NET_CLS_ACT 560 if (map->action && exts->action) { 561 /* 562 * again for backward compatible mode - we want 563 * to work with both old and new modes of entering 564 * tc data even if iproute2 was newer - jhs 565 */ 566 struct rtattr * p_rta = (struct rtattr*) skb->tail; 567 568 if (exts->action->type != TCA_OLD_COMPAT) { 569 RTA_PUT(skb, map->action, 0, NULL); 570 if (tcf_action_dump(skb, exts->action, 0, 0) < 0) 571 goto rtattr_failure; 572 p_rta->rta_len = skb->tail - (u8*)p_rta; 573 } else if (map->police) { 574 RTA_PUT(skb, map->police, 0, NULL); 575 if (tcf_action_dump_old(skb, exts->action, 0, 0) < 0) 576 goto rtattr_failure; 577 p_rta->rta_len = skb->tail - (u8*)p_rta; 578 } 579 } 580 #elif defined CONFIG_NET_CLS_POLICE 581 if (map->police && exts->police) { 582 struct rtattr * p_rta = (struct rtattr*) skb->tail; 583 584 RTA_PUT(skb, map->police, 0, NULL); 585 586 if (tcf_police_dump(skb, exts->police) < 0) 587 goto rtattr_failure; 588 589 p_rta->rta_len = skb->tail - (u8*)p_rta; 590 } 591 #endif 592 return 0; 593 rtattr_failure: __attribute__ ((unused)) 594 return -1; 595 } 596 597 int 598 tcf_exts_dump_stats(struct sk_buff *skb, struct tcf_exts *exts, 599 struct tcf_ext_map *map) 600 { 601 #ifdef CONFIG_NET_CLS_ACT 602 if (exts->action) 603 if (tcf_action_copy_stats(skb, exts->action, 1) < 0) 604 goto rtattr_failure; 605 #elif defined CONFIG_NET_CLS_POLICE 606 if (exts->police) 607 if (tcf_police_dump_stats(skb, exts->police) < 0) 608 goto rtattr_failure; 609 #endif 610 return 0; 611 rtattr_failure: __attribute__ ((unused)) 612 return -1; 613 } 614 615 static int __init tc_filter_init(void) 616 { 617 struct rtnetlink_link *link_p = rtnetlink_links[PF_UNSPEC]; 618 619 /* Setup rtnetlink links. It is made here to avoid 620 exporting large number of public symbols. 621 */ 622 623 if (link_p) { 624 link_p[RTM_NEWTFILTER-RTM_BASE].doit = tc_ctl_tfilter; 625 link_p[RTM_DELTFILTER-RTM_BASE].doit = tc_ctl_tfilter; 626 link_p[RTM_GETTFILTER-RTM_BASE].doit = tc_ctl_tfilter; 627 link_p[RTM_GETTFILTER-RTM_BASE].dumpit = tc_dump_tfilter; 628 } 629 return 0; 630 } 631 632 subsys_initcall(tc_filter_init); 633 634 EXPORT_SYMBOL(register_tcf_proto_ops); 635 EXPORT_SYMBOL(unregister_tcf_proto_ops); 636 EXPORT_SYMBOL(tcf_exts_validate); 637 EXPORT_SYMBOL(tcf_exts_destroy); 638 EXPORT_SYMBOL(tcf_exts_change); 639 EXPORT_SYMBOL(tcf_exts_dump); 640 EXPORT_SYMBOL(tcf_exts_dump_stats); 641