1/* 2 * net/sched/cls_route.c ROUTE4 classifier. 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 12#include <linux/module.h> 13#include <linux/slab.h> 14#include <linux/types.h> 15#include <linux/kernel.h> 16#include <linux/string.h> 17#include <linux/errno.h> 18#include <linux/skbuff.h> 19#include <net/dst.h> 20#include <net/route.h> 21#include <net/netlink.h> 22#include <net/act_api.h> 23#include <net/pkt_cls.h> 24 25/* 26 * 1. For now we assume that route tags < 256. 27 * It allows to use direct table lookups, instead of hash tables. 28 * 2. For now we assume that "from TAG" and "fromdev DEV" statements 29 * are mutually exclusive. 30 * 3. "to TAG from ANY" has higher priority, than "to ANY from XXX" 31 */ 32struct route4_fastmap { 33 struct route4_filter *filter; 34 u32 id; 35 int iif; 36}; 37 38struct route4_head { 39 struct route4_fastmap fastmap[16]; 40 struct route4_bucket __rcu *table[256 + 1]; 41 struct rcu_head rcu; 42}; 43 44struct route4_bucket { 45 /* 16 FROM buckets + 16 IIF buckets + 1 wildcard bucket */ 46 struct route4_filter __rcu *ht[16 + 16 + 1]; 47 struct rcu_head rcu; 48}; 49 50struct route4_filter { 51 struct route4_filter __rcu *next; 52 u32 id; 53 int iif; 54 55 struct tcf_result res; 56 struct tcf_exts exts; 57 u32 handle; 58 struct route4_bucket *bkt; 59 struct tcf_proto *tp; 60 struct rcu_head rcu; 61}; 62 63#define ROUTE4_FAILURE ((struct route4_filter *)(-1L)) 64 65static inline int route4_fastmap_hash(u32 id, int iif) 66{ 67 return id & 0xF; 68} 69 70static DEFINE_SPINLOCK(fastmap_lock); 71static void 72route4_reset_fastmap(struct route4_head *head) 73{ 74 spin_lock_bh(&fastmap_lock); 75 memset(head->fastmap, 0, sizeof(head->fastmap)); 76 spin_unlock_bh(&fastmap_lock); 77} 78 79static void 80route4_set_fastmap(struct route4_head *head, u32 id, int iif, 81 struct route4_filter *f) 82{ 83 int h = route4_fastmap_hash(id, iif); 84 85 /* fastmap updates must look atomic to aling id, iff, filter */ 86 spin_lock_bh(&fastmap_lock); 87 head->fastmap[h].id = id; 88 head->fastmap[h].iif = iif; 89 head->fastmap[h].filter = f; 90 spin_unlock_bh(&fastmap_lock); 91} 92 93static inline int route4_hash_to(u32 id) 94{ 95 return id & 0xFF; 96} 97 98static inline int route4_hash_from(u32 id) 99{ 100 return (id >> 16) & 0xF; 101} 102 103static inline int route4_hash_iif(int iif) 104{ 105 return 16 + ((iif >> 16) & 0xF); 106} 107 108static inline int route4_hash_wild(void) 109{ 110 return 32; 111} 112 113#define ROUTE4_APPLY_RESULT() \ 114{ \ 115 *res = f->res; \ 116 if (tcf_exts_is_available(&f->exts)) { \ 117 int r = tcf_exts_exec(skb, &f->exts, res); \ 118 if (r < 0) { \ 119 dont_cache = 1; \ 120 continue; \ 121 } \ 122 return r; \ 123 } else if (!dont_cache) \ 124 route4_set_fastmap(head, id, iif, f); \ 125 return 0; \ 126} 127 128static int route4_classify(struct sk_buff *skb, const struct tcf_proto *tp, 129 struct tcf_result *res) 130{ 131 struct route4_head *head = rcu_dereference_bh(tp->root); 132 struct dst_entry *dst; 133 struct route4_bucket *b; 134 struct route4_filter *f; 135 u32 id, h; 136 int iif, dont_cache = 0; 137 138 dst = skb_dst(skb); 139 if (!dst) 140 goto failure; 141 142 id = dst->tclassid; 143 if (head == NULL) 144 goto old_method; 145 146 iif = inet_iif(skb); 147 148 h = route4_fastmap_hash(id, iif); 149 150 spin_lock(&fastmap_lock); 151 if (id == head->fastmap[h].id && 152 iif == head->fastmap[h].iif && 153 (f = head->fastmap[h].filter) != NULL) { 154 if (f == ROUTE4_FAILURE) { 155 spin_unlock(&fastmap_lock); 156 goto failure; 157 } 158 159 *res = f->res; 160 spin_unlock(&fastmap_lock); 161 return 0; 162 } 163 spin_unlock(&fastmap_lock); 164 165 h = route4_hash_to(id); 166 167restart: 168 b = rcu_dereference_bh(head->table[h]); 169 if (b) { 170 for (f = rcu_dereference_bh(b->ht[route4_hash_from(id)]); 171 f; 172 f = rcu_dereference_bh(f->next)) 173 if (f->id == id) 174 ROUTE4_APPLY_RESULT(); 175 176 for (f = rcu_dereference_bh(b->ht[route4_hash_iif(iif)]); 177 f; 178 f = rcu_dereference_bh(f->next)) 179 if (f->iif == iif) 180 ROUTE4_APPLY_RESULT(); 181 182 for (f = rcu_dereference_bh(b->ht[route4_hash_wild()]); 183 f; 184 f = rcu_dereference_bh(f->next)) 185 ROUTE4_APPLY_RESULT(); 186 } 187 if (h < 256) { 188 h = 256; 189 id &= ~0xFFFF; 190 goto restart; 191 } 192 193 if (!dont_cache) 194 route4_set_fastmap(head, id, iif, ROUTE4_FAILURE); 195failure: 196 return -1; 197 198old_method: 199 if (id && (TC_H_MAJ(id) == 0 || 200 !(TC_H_MAJ(id^tp->q->handle)))) { 201 res->classid = id; 202 res->class = 0; 203 return 0; 204 } 205 return -1; 206} 207 208static inline u32 to_hash(u32 id) 209{ 210 u32 h = id & 0xFF; 211 212 if (id & 0x8000) 213 h += 256; 214 return h; 215} 216 217static inline u32 from_hash(u32 id) 218{ 219 id &= 0xFFFF; 220 if (id == 0xFFFF) 221 return 32; 222 if (!(id & 0x8000)) { 223 if (id > 255) 224 return 256; 225 return id & 0xF; 226 } 227 return 16 + (id & 0xF); 228} 229 230static unsigned long route4_get(struct tcf_proto *tp, u32 handle) 231{ 232 struct route4_head *head = rtnl_dereference(tp->root); 233 struct route4_bucket *b; 234 struct route4_filter *f; 235 unsigned int h1, h2; 236 237 if (!head) 238 return 0; 239 240 h1 = to_hash(handle); 241 if (h1 > 256) 242 return 0; 243 244 h2 = from_hash(handle >> 16); 245 if (h2 > 32) 246 return 0; 247 248 b = rtnl_dereference(head->table[h1]); 249 if (b) { 250 for (f = rtnl_dereference(b->ht[h2]); 251 f; 252 f = rtnl_dereference(f->next)) 253 if (f->handle == handle) 254 return (unsigned long)f; 255 } 256 return 0; 257} 258 259static int route4_init(struct tcf_proto *tp) 260{ 261 struct route4_head *head; 262 263 head = kzalloc(sizeof(struct route4_head), GFP_KERNEL); 264 if (head == NULL) 265 return -ENOBUFS; 266 267 rcu_assign_pointer(tp->root, head); 268 return 0; 269} 270 271static void 272route4_delete_filter(struct rcu_head *head) 273{ 274 struct route4_filter *f = container_of(head, struct route4_filter, rcu); 275 276 tcf_exts_destroy(&f->exts); 277 kfree(f); 278} 279 280static bool route4_destroy(struct tcf_proto *tp, bool force) 281{ 282 struct route4_head *head = rtnl_dereference(tp->root); 283 int h1, h2; 284 285 if (head == NULL) 286 return true; 287 288 if (!force) { 289 for (h1 = 0; h1 <= 256; h1++) { 290 if (rcu_access_pointer(head->table[h1])) 291 return false; 292 } 293 } 294 295 for (h1 = 0; h1 <= 256; h1++) { 296 struct route4_bucket *b; 297 298 b = rtnl_dereference(head->table[h1]); 299 if (b) { 300 for (h2 = 0; h2 <= 32; h2++) { 301 struct route4_filter *f; 302 303 while ((f = rtnl_dereference(b->ht[h2])) != NULL) { 304 struct route4_filter *next; 305 306 next = rtnl_dereference(f->next); 307 RCU_INIT_POINTER(b->ht[h2], next); 308 tcf_unbind_filter(tp, &f->res); 309 call_rcu(&f->rcu, route4_delete_filter); 310 } 311 } 312 RCU_INIT_POINTER(head->table[h1], NULL); 313 kfree_rcu(b, rcu); 314 } 315 } 316 RCU_INIT_POINTER(tp->root, NULL); 317 kfree_rcu(head, rcu); 318 return true; 319} 320 321static int route4_delete(struct tcf_proto *tp, unsigned long arg) 322{ 323 struct route4_head *head = rtnl_dereference(tp->root); 324 struct route4_filter *f = (struct route4_filter *)arg; 325 struct route4_filter __rcu **fp; 326 struct route4_filter *nf; 327 struct route4_bucket *b; 328 unsigned int h = 0; 329 int i; 330 331 if (!head || !f) 332 return -EINVAL; 333 334 h = f->handle; 335 b = f->bkt; 336 337 fp = &b->ht[from_hash(h >> 16)]; 338 for (nf = rtnl_dereference(*fp); nf; 339 fp = &nf->next, nf = rtnl_dereference(*fp)) { 340 if (nf == f) { 341 /* unlink it */ 342 RCU_INIT_POINTER(*fp, rtnl_dereference(f->next)); 343 344 /* Remove any fastmap lookups that might ref filter 345 * notice we unlink'd the filter so we can't get it 346 * back in the fastmap. 347 */ 348 route4_reset_fastmap(head); 349 350 /* Delete it */ 351 tcf_unbind_filter(tp, &f->res); 352 call_rcu(&f->rcu, route4_delete_filter); 353 354 /* Strip RTNL protected tree */ 355 for (i = 0; i <= 32; i++) { 356 struct route4_filter *rt; 357 358 rt = rtnl_dereference(b->ht[i]); 359 if (rt) 360 return 0; 361 } 362 363 /* OK, session has no flows */ 364 RCU_INIT_POINTER(head->table[to_hash(h)], NULL); 365 kfree_rcu(b, rcu); 366 367 return 0; 368 } 369 } 370 return 0; 371} 372 373static const struct nla_policy route4_policy[TCA_ROUTE4_MAX + 1] = { 374 [TCA_ROUTE4_CLASSID] = { .type = NLA_U32 }, 375 [TCA_ROUTE4_TO] = { .type = NLA_U32 }, 376 [TCA_ROUTE4_FROM] = { .type = NLA_U32 }, 377 [TCA_ROUTE4_IIF] = { .type = NLA_U32 }, 378}; 379 380static int route4_set_parms(struct net *net, struct tcf_proto *tp, 381 unsigned long base, struct route4_filter *f, 382 u32 handle, struct route4_head *head, 383 struct nlattr **tb, struct nlattr *est, int new, 384 bool ovr) 385{ 386 int err; 387 u32 id = 0, to = 0, nhandle = 0x8000; 388 struct route4_filter *fp; 389 unsigned int h1; 390 struct route4_bucket *b; 391 struct tcf_exts e; 392 393 tcf_exts_init(&e, TCA_ROUTE4_ACT, TCA_ROUTE4_POLICE); 394 err = tcf_exts_validate(net, tp, tb, est, &e, ovr); 395 if (err < 0) 396 return err; 397 398 err = -EINVAL; 399 if (tb[TCA_ROUTE4_TO]) { 400 if (new && handle & 0x8000) 401 goto errout; 402 to = nla_get_u32(tb[TCA_ROUTE4_TO]); 403 if (to > 0xFF) 404 goto errout; 405 nhandle = to; 406 } 407 408 if (tb[TCA_ROUTE4_FROM]) { 409 if (tb[TCA_ROUTE4_IIF]) 410 goto errout; 411 id = nla_get_u32(tb[TCA_ROUTE4_FROM]); 412 if (id > 0xFF) 413 goto errout; 414 nhandle |= id << 16; 415 } else if (tb[TCA_ROUTE4_IIF]) { 416 id = nla_get_u32(tb[TCA_ROUTE4_IIF]); 417 if (id > 0x7FFF) 418 goto errout; 419 nhandle |= (id | 0x8000) << 16; 420 } else 421 nhandle |= 0xFFFF << 16; 422 423 if (handle && new) { 424 nhandle |= handle & 0x7F00; 425 if (nhandle != handle) 426 goto errout; 427 } 428 429 h1 = to_hash(nhandle); 430 b = rtnl_dereference(head->table[h1]); 431 if (!b) { 432 err = -ENOBUFS; 433 b = kzalloc(sizeof(struct route4_bucket), GFP_KERNEL); 434 if (b == NULL) 435 goto errout; 436 437 rcu_assign_pointer(head->table[h1], b); 438 } else { 439 unsigned int h2 = from_hash(nhandle >> 16); 440 441 err = -EEXIST; 442 for (fp = rtnl_dereference(b->ht[h2]); 443 fp; 444 fp = rtnl_dereference(fp->next)) 445 if (fp->handle == f->handle) 446 goto errout; 447 } 448 449 if (tb[TCA_ROUTE4_TO]) 450 f->id = to; 451 452 if (tb[TCA_ROUTE4_FROM]) 453 f->id = to | id<<16; 454 else if (tb[TCA_ROUTE4_IIF]) 455 f->iif = id; 456 457 f->handle = nhandle; 458 f->bkt = b; 459 f->tp = tp; 460 461 if (tb[TCA_ROUTE4_CLASSID]) { 462 f->res.classid = nla_get_u32(tb[TCA_ROUTE4_CLASSID]); 463 tcf_bind_filter(tp, &f->res, base); 464 } 465 466 tcf_exts_change(tp, &f->exts, &e); 467 468 return 0; 469errout: 470 tcf_exts_destroy(&e); 471 return err; 472} 473 474static int route4_change(struct net *net, struct sk_buff *in_skb, 475 struct tcf_proto *tp, unsigned long base, 476 u32 handle, 477 struct nlattr **tca, 478 unsigned long *arg, bool ovr) 479{ 480 struct route4_head *head = rtnl_dereference(tp->root); 481 struct route4_filter __rcu **fp; 482 struct route4_filter *fold, *f1, *pfp, *f = NULL; 483 struct route4_bucket *b; 484 struct nlattr *opt = tca[TCA_OPTIONS]; 485 struct nlattr *tb[TCA_ROUTE4_MAX + 1]; 486 unsigned int h, th; 487 int err; 488 bool new = true; 489 490 if (opt == NULL) 491 return handle ? -EINVAL : 0; 492 493 err = nla_parse_nested(tb, TCA_ROUTE4_MAX, opt, route4_policy); 494 if (err < 0) 495 return err; 496 497 fold = (struct route4_filter *)*arg; 498 if (fold && handle && fold->handle != handle) 499 return -EINVAL; 500 501 err = -ENOBUFS; 502 f = kzalloc(sizeof(struct route4_filter), GFP_KERNEL); 503 if (!f) 504 goto errout; 505 506 tcf_exts_init(&f->exts, TCA_ROUTE4_ACT, TCA_ROUTE4_POLICE); 507 if (fold) { 508 f->id = fold->id; 509 f->iif = fold->iif; 510 f->res = fold->res; 511 f->handle = fold->handle; 512 513 f->tp = fold->tp; 514 f->bkt = fold->bkt; 515 new = false; 516 } 517 518 err = route4_set_parms(net, tp, base, f, handle, head, tb, 519 tca[TCA_RATE], new, ovr); 520 if (err < 0) 521 goto errout; 522 523 h = from_hash(f->handle >> 16); 524 fp = &f->bkt->ht[h]; 525 for (pfp = rtnl_dereference(*fp); 526 (f1 = rtnl_dereference(*fp)) != NULL; 527 fp = &f1->next) 528 if (f->handle < f1->handle) 529 break; 530 531 netif_keep_dst(qdisc_dev(tp->q)); 532 rcu_assign_pointer(f->next, f1); 533 rcu_assign_pointer(*fp, f); 534 535 if (fold && fold->handle && f->handle != fold->handle) { 536 th = to_hash(fold->handle); 537 h = from_hash(fold->handle >> 16); 538 b = rtnl_dereference(head->table[th]); 539 if (b) { 540 fp = &b->ht[h]; 541 for (pfp = rtnl_dereference(*fp); pfp; 542 fp = &pfp->next, pfp = rtnl_dereference(*fp)) { 543 if (pfp == f) { 544 *fp = f->next; 545 break; 546 } 547 } 548 } 549 } 550 551 route4_reset_fastmap(head); 552 *arg = (unsigned long)f; 553 if (fold) { 554 tcf_unbind_filter(tp, &fold->res); 555 call_rcu(&fold->rcu, route4_delete_filter); 556 } 557 return 0; 558 559errout: 560 kfree(f); 561 return err; 562} 563 564static void route4_walk(struct tcf_proto *tp, struct tcf_walker *arg) 565{ 566 struct route4_head *head = rtnl_dereference(tp->root); 567 unsigned int h, h1; 568 569 if (head == NULL) 570 arg->stop = 1; 571 572 if (arg->stop) 573 return; 574 575 for (h = 0; h <= 256; h++) { 576 struct route4_bucket *b = rtnl_dereference(head->table[h]); 577 578 if (b) { 579 for (h1 = 0; h1 <= 32; h1++) { 580 struct route4_filter *f; 581 582 for (f = rtnl_dereference(b->ht[h1]); 583 f; 584 f = rtnl_dereference(f->next)) { 585 if (arg->count < arg->skip) { 586 arg->count++; 587 continue; 588 } 589 if (arg->fn(tp, (unsigned long)f, arg) < 0) { 590 arg->stop = 1; 591 return; 592 } 593 arg->count++; 594 } 595 } 596 } 597 } 598} 599 600static int route4_dump(struct net *net, struct tcf_proto *tp, unsigned long fh, 601 struct sk_buff *skb, struct tcmsg *t) 602{ 603 struct route4_filter *f = (struct route4_filter *)fh; 604 struct nlattr *nest; 605 u32 id; 606 607 if (f == NULL) 608 return skb->len; 609 610 t->tcm_handle = f->handle; 611 612 nest = nla_nest_start(skb, TCA_OPTIONS); 613 if (nest == NULL) 614 goto nla_put_failure; 615 616 if (!(f->handle & 0x8000)) { 617 id = f->id & 0xFF; 618 if (nla_put_u32(skb, TCA_ROUTE4_TO, id)) 619 goto nla_put_failure; 620 } 621 if (f->handle & 0x80000000) { 622 if ((f->handle >> 16) != 0xFFFF && 623 nla_put_u32(skb, TCA_ROUTE4_IIF, f->iif)) 624 goto nla_put_failure; 625 } else { 626 id = f->id >> 16; 627 if (nla_put_u32(skb, TCA_ROUTE4_FROM, id)) 628 goto nla_put_failure; 629 } 630 if (f->res.classid && 631 nla_put_u32(skb, TCA_ROUTE4_CLASSID, f->res.classid)) 632 goto nla_put_failure; 633 634 if (tcf_exts_dump(skb, &f->exts) < 0) 635 goto nla_put_failure; 636 637 nla_nest_end(skb, nest); 638 639 if (tcf_exts_dump_stats(skb, &f->exts) < 0) 640 goto nla_put_failure; 641 642 return skb->len; 643 644nla_put_failure: 645 nla_nest_cancel(skb, nest); 646 return -1; 647} 648 649static struct tcf_proto_ops cls_route4_ops __read_mostly = { 650 .kind = "route", 651 .classify = route4_classify, 652 .init = route4_init, 653 .destroy = route4_destroy, 654 .get = route4_get, 655 .change = route4_change, 656 .delete = route4_delete, 657 .walk = route4_walk, 658 .dump = route4_dump, 659 .owner = THIS_MODULE, 660}; 661 662static int __init init_route4(void) 663{ 664 return register_tcf_proto_ops(&cls_route4_ops); 665} 666 667static void __exit exit_route4(void) 668{ 669 unregister_tcf_proto_ops(&cls_route4_ops); 670} 671 672module_init(init_route4) 673module_exit(exit_route4) 674MODULE_LICENSE("GPL"); 675