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 */ 32 struct route4_fastmap { 33 struct route4_filter *filter; 34 u32 id; 35 int iif; 36 }; 37 38 struct route4_head { 39 struct route4_fastmap fastmap[16]; 40 struct route4_bucket __rcu *table[256 + 1]; 41 struct rcu_head rcu; 42 }; 43 44 struct 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 50 struct 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 union { 61 struct work_struct work; 62 struct rcu_head rcu; 63 }; 64 }; 65 66 #define ROUTE4_FAILURE ((struct route4_filter *)(-1L)) 67 68 static inline int route4_fastmap_hash(u32 id, int iif) 69 { 70 return id & 0xF; 71 } 72 73 static DEFINE_SPINLOCK(fastmap_lock); 74 static void 75 route4_reset_fastmap(struct route4_head *head) 76 { 77 spin_lock_bh(&fastmap_lock); 78 memset(head->fastmap, 0, sizeof(head->fastmap)); 79 spin_unlock_bh(&fastmap_lock); 80 } 81 82 static void 83 route4_set_fastmap(struct route4_head *head, u32 id, int iif, 84 struct route4_filter *f) 85 { 86 int h = route4_fastmap_hash(id, iif); 87 88 /* fastmap updates must look atomic to aling id, iff, filter */ 89 spin_lock_bh(&fastmap_lock); 90 head->fastmap[h].id = id; 91 head->fastmap[h].iif = iif; 92 head->fastmap[h].filter = f; 93 spin_unlock_bh(&fastmap_lock); 94 } 95 96 static inline int route4_hash_to(u32 id) 97 { 98 return id & 0xFF; 99 } 100 101 static inline int route4_hash_from(u32 id) 102 { 103 return (id >> 16) & 0xF; 104 } 105 106 static inline int route4_hash_iif(int iif) 107 { 108 return 16 + ((iif >> 16) & 0xF); 109 } 110 111 static inline int route4_hash_wild(void) 112 { 113 return 32; 114 } 115 116 #define ROUTE4_APPLY_RESULT() \ 117 { \ 118 *res = f->res; \ 119 if (tcf_exts_has_actions(&f->exts)) { \ 120 int r = tcf_exts_exec(skb, &f->exts, res); \ 121 if (r < 0) { \ 122 dont_cache = 1; \ 123 continue; \ 124 } \ 125 return r; \ 126 } else if (!dont_cache) \ 127 route4_set_fastmap(head, id, iif, f); \ 128 return 0; \ 129 } 130 131 static int route4_classify(struct sk_buff *skb, const struct tcf_proto *tp, 132 struct tcf_result *res) 133 { 134 struct route4_head *head = rcu_dereference_bh(tp->root); 135 struct dst_entry *dst; 136 struct route4_bucket *b; 137 struct route4_filter *f; 138 u32 id, h; 139 int iif, dont_cache = 0; 140 141 dst = skb_dst(skb); 142 if (!dst) 143 goto failure; 144 145 id = dst->tclassid; 146 147 iif = inet_iif(skb); 148 149 h = route4_fastmap_hash(id, iif); 150 151 spin_lock(&fastmap_lock); 152 if (id == head->fastmap[h].id && 153 iif == head->fastmap[h].iif && 154 (f = head->fastmap[h].filter) != NULL) { 155 if (f == ROUTE4_FAILURE) { 156 spin_unlock(&fastmap_lock); 157 goto failure; 158 } 159 160 *res = f->res; 161 spin_unlock(&fastmap_lock); 162 return 0; 163 } 164 spin_unlock(&fastmap_lock); 165 166 h = route4_hash_to(id); 167 168 restart: 169 b = rcu_dereference_bh(head->table[h]); 170 if (b) { 171 for (f = rcu_dereference_bh(b->ht[route4_hash_from(id)]); 172 f; 173 f = rcu_dereference_bh(f->next)) 174 if (f->id == id) 175 ROUTE4_APPLY_RESULT(); 176 177 for (f = rcu_dereference_bh(b->ht[route4_hash_iif(iif)]); 178 f; 179 f = rcu_dereference_bh(f->next)) 180 if (f->iif == iif) 181 ROUTE4_APPLY_RESULT(); 182 183 for (f = rcu_dereference_bh(b->ht[route4_hash_wild()]); 184 f; 185 f = rcu_dereference_bh(f->next)) 186 ROUTE4_APPLY_RESULT(); 187 } 188 if (h < 256) { 189 h = 256; 190 id &= ~0xFFFF; 191 goto restart; 192 } 193 194 if (!dont_cache) 195 route4_set_fastmap(head, id, iif, ROUTE4_FAILURE); 196 failure: 197 return -1; 198 } 199 200 static inline u32 to_hash(u32 id) 201 { 202 u32 h = id & 0xFF; 203 204 if (id & 0x8000) 205 h += 256; 206 return h; 207 } 208 209 static inline u32 from_hash(u32 id) 210 { 211 id &= 0xFFFF; 212 if (id == 0xFFFF) 213 return 32; 214 if (!(id & 0x8000)) { 215 if (id > 255) 216 return 256; 217 return id & 0xF; 218 } 219 return 16 + (id & 0xF); 220 } 221 222 static void *route4_get(struct tcf_proto *tp, u32 handle) 223 { 224 struct route4_head *head = rtnl_dereference(tp->root); 225 struct route4_bucket *b; 226 struct route4_filter *f; 227 unsigned int h1, h2; 228 229 h1 = to_hash(handle); 230 if (h1 > 256) 231 return NULL; 232 233 h2 = from_hash(handle >> 16); 234 if (h2 > 32) 235 return NULL; 236 237 b = rtnl_dereference(head->table[h1]); 238 if (b) { 239 for (f = rtnl_dereference(b->ht[h2]); 240 f; 241 f = rtnl_dereference(f->next)) 242 if (f->handle == handle) 243 return f; 244 } 245 return NULL; 246 } 247 248 static int route4_init(struct tcf_proto *tp) 249 { 250 struct route4_head *head; 251 252 head = kzalloc(sizeof(struct route4_head), GFP_KERNEL); 253 if (head == NULL) 254 return -ENOBUFS; 255 256 rcu_assign_pointer(tp->root, head); 257 return 0; 258 } 259 260 static void __route4_delete_filter(struct route4_filter *f) 261 { 262 tcf_exts_destroy(&f->exts); 263 tcf_exts_put_net(&f->exts); 264 kfree(f); 265 } 266 267 static void route4_delete_filter_work(struct work_struct *work) 268 { 269 struct route4_filter *f = container_of(work, struct route4_filter, work); 270 271 rtnl_lock(); 272 __route4_delete_filter(f); 273 rtnl_unlock(); 274 } 275 276 static void route4_delete_filter(struct rcu_head *head) 277 { 278 struct route4_filter *f = container_of(head, struct route4_filter, rcu); 279 280 INIT_WORK(&f->work, route4_delete_filter_work); 281 tcf_queue_work(&f->work); 282 } 283 284 static void route4_destroy(struct tcf_proto *tp) 285 { 286 struct route4_head *head = rtnl_dereference(tp->root); 287 int h1, h2; 288 289 if (head == NULL) 290 return; 291 292 for (h1 = 0; h1 <= 256; h1++) { 293 struct route4_bucket *b; 294 295 b = rtnl_dereference(head->table[h1]); 296 if (b) { 297 for (h2 = 0; h2 <= 32; h2++) { 298 struct route4_filter *f; 299 300 while ((f = rtnl_dereference(b->ht[h2])) != NULL) { 301 struct route4_filter *next; 302 303 next = rtnl_dereference(f->next); 304 RCU_INIT_POINTER(b->ht[h2], next); 305 tcf_unbind_filter(tp, &f->res); 306 if (tcf_exts_get_net(&f->exts)) 307 call_rcu(&f->rcu, route4_delete_filter); 308 else 309 __route4_delete_filter(f); 310 } 311 } 312 RCU_INIT_POINTER(head->table[h1], NULL); 313 kfree_rcu(b, rcu); 314 } 315 } 316 kfree_rcu(head, rcu); 317 } 318 319 static int route4_delete(struct tcf_proto *tp, void *arg, bool *last) 320 { 321 struct route4_head *head = rtnl_dereference(tp->root); 322 struct route4_filter *f = arg; 323 struct route4_filter __rcu **fp; 324 struct route4_filter *nf; 325 struct route4_bucket *b; 326 unsigned int h = 0; 327 int i, h1; 328 329 if (!head || !f) 330 return -EINVAL; 331 332 h = f->handle; 333 b = f->bkt; 334 335 fp = &b->ht[from_hash(h >> 16)]; 336 for (nf = rtnl_dereference(*fp); nf; 337 fp = &nf->next, nf = rtnl_dereference(*fp)) { 338 if (nf == f) { 339 /* unlink it */ 340 RCU_INIT_POINTER(*fp, rtnl_dereference(f->next)); 341 342 /* Remove any fastmap lookups that might ref filter 343 * notice we unlink'd the filter so we can't get it 344 * back in the fastmap. 345 */ 346 route4_reset_fastmap(head); 347 348 /* Delete it */ 349 tcf_unbind_filter(tp, &f->res); 350 tcf_exts_get_net(&f->exts); 351 call_rcu(&f->rcu, route4_delete_filter); 352 353 /* Strip RTNL protected tree */ 354 for (i = 0; i <= 32; i++) { 355 struct route4_filter *rt; 356 357 rt = rtnl_dereference(b->ht[i]); 358 if (rt) 359 goto out; 360 } 361 362 /* OK, session has no flows */ 363 RCU_INIT_POINTER(head->table[to_hash(h)], NULL); 364 kfree_rcu(b, rcu); 365 break; 366 } 367 } 368 369 out: 370 *last = true; 371 for (h1 = 0; h1 <= 256; h1++) { 372 if (rcu_access_pointer(head->table[h1])) { 373 *last = false; 374 break; 375 } 376 } 377 378 return 0; 379 } 380 381 static const struct nla_policy route4_policy[TCA_ROUTE4_MAX + 1] = { 382 [TCA_ROUTE4_CLASSID] = { .type = NLA_U32 }, 383 [TCA_ROUTE4_TO] = { .type = NLA_U32 }, 384 [TCA_ROUTE4_FROM] = { .type = NLA_U32 }, 385 [TCA_ROUTE4_IIF] = { .type = NLA_U32 }, 386 }; 387 388 static int route4_set_parms(struct net *net, struct tcf_proto *tp, 389 unsigned long base, struct route4_filter *f, 390 u32 handle, struct route4_head *head, 391 struct nlattr **tb, struct nlattr *est, int new, 392 bool ovr) 393 { 394 u32 id = 0, to = 0, nhandle = 0x8000; 395 struct route4_filter *fp; 396 unsigned int h1; 397 struct route4_bucket *b; 398 int err; 399 400 err = tcf_exts_validate(net, tp, tb, est, &f->exts, ovr); 401 if (err < 0) 402 return err; 403 404 if (tb[TCA_ROUTE4_TO]) { 405 if (new && handle & 0x8000) 406 return -EINVAL; 407 to = nla_get_u32(tb[TCA_ROUTE4_TO]); 408 if (to > 0xFF) 409 return -EINVAL; 410 nhandle = to; 411 } 412 413 if (tb[TCA_ROUTE4_FROM]) { 414 if (tb[TCA_ROUTE4_IIF]) 415 return -EINVAL; 416 id = nla_get_u32(tb[TCA_ROUTE4_FROM]); 417 if (id > 0xFF) 418 return -EINVAL; 419 nhandle |= id << 16; 420 } else if (tb[TCA_ROUTE4_IIF]) { 421 id = nla_get_u32(tb[TCA_ROUTE4_IIF]); 422 if (id > 0x7FFF) 423 return -EINVAL; 424 nhandle |= (id | 0x8000) << 16; 425 } else 426 nhandle |= 0xFFFF << 16; 427 428 if (handle && new) { 429 nhandle |= handle & 0x7F00; 430 if (nhandle != handle) 431 return -EINVAL; 432 } 433 434 h1 = to_hash(nhandle); 435 b = rtnl_dereference(head->table[h1]); 436 if (!b) { 437 b = kzalloc(sizeof(struct route4_bucket), GFP_KERNEL); 438 if (b == NULL) 439 return -ENOBUFS; 440 441 rcu_assign_pointer(head->table[h1], b); 442 } else { 443 unsigned int h2 = from_hash(nhandle >> 16); 444 445 for (fp = rtnl_dereference(b->ht[h2]); 446 fp; 447 fp = rtnl_dereference(fp->next)) 448 if (fp->handle == f->handle) 449 return -EEXIST; 450 } 451 452 if (tb[TCA_ROUTE4_TO]) 453 f->id = to; 454 455 if (tb[TCA_ROUTE4_FROM]) 456 f->id = to | id<<16; 457 else if (tb[TCA_ROUTE4_IIF]) 458 f->iif = id; 459 460 f->handle = nhandle; 461 f->bkt = b; 462 f->tp = tp; 463 464 if (tb[TCA_ROUTE4_CLASSID]) { 465 f->res.classid = nla_get_u32(tb[TCA_ROUTE4_CLASSID]); 466 tcf_bind_filter(tp, &f->res, base); 467 } 468 469 return 0; 470 } 471 472 static int route4_change(struct net *net, struct sk_buff *in_skb, 473 struct tcf_proto *tp, unsigned long base, u32 handle, 474 struct nlattr **tca, void **arg, bool ovr) 475 { 476 struct route4_head *head = rtnl_dereference(tp->root); 477 struct route4_filter __rcu **fp; 478 struct route4_filter *fold, *f1, *pfp, *f = NULL; 479 struct route4_bucket *b; 480 struct nlattr *opt = tca[TCA_OPTIONS]; 481 struct nlattr *tb[TCA_ROUTE4_MAX + 1]; 482 unsigned int h, th; 483 int err; 484 bool new = true; 485 486 if (opt == NULL) 487 return handle ? -EINVAL : 0; 488 489 err = nla_parse_nested(tb, TCA_ROUTE4_MAX, opt, route4_policy, NULL); 490 if (err < 0) 491 return err; 492 493 fold = *arg; 494 if (fold && handle && fold->handle != handle) 495 return -EINVAL; 496 497 err = -ENOBUFS; 498 f = kzalloc(sizeof(struct route4_filter), GFP_KERNEL); 499 if (!f) 500 goto errout; 501 502 err = tcf_exts_init(&f->exts, TCA_ROUTE4_ACT, TCA_ROUTE4_POLICE); 503 if (err < 0) 504 goto errout; 505 506 if (fold) { 507 f->id = fold->id; 508 f->iif = fold->iif; 509 f->res = fold->res; 510 f->handle = fold->handle; 511 512 f->tp = fold->tp; 513 f->bkt = fold->bkt; 514 new = false; 515 } 516 517 err = route4_set_parms(net, tp, base, f, handle, head, tb, 518 tca[TCA_RATE], new, ovr); 519 if (err < 0) 520 goto errout; 521 522 h = from_hash(f->handle >> 16); 523 fp = &f->bkt->ht[h]; 524 for (pfp = rtnl_dereference(*fp); 525 (f1 = rtnl_dereference(*fp)) != NULL; 526 fp = &f1->next) 527 if (f->handle < f1->handle) 528 break; 529 530 netif_keep_dst(qdisc_dev(tp->q)); 531 rcu_assign_pointer(f->next, f1); 532 rcu_assign_pointer(*fp, f); 533 534 if (fold && fold->handle && f->handle != fold->handle) { 535 th = to_hash(fold->handle); 536 h = from_hash(fold->handle >> 16); 537 b = rtnl_dereference(head->table[th]); 538 if (b) { 539 fp = &b->ht[h]; 540 for (pfp = rtnl_dereference(*fp); pfp; 541 fp = &pfp->next, pfp = rtnl_dereference(*fp)) { 542 if (pfp == f) { 543 *fp = f->next; 544 break; 545 } 546 } 547 } 548 } 549 550 route4_reset_fastmap(head); 551 *arg = f; 552 if (fold) { 553 tcf_unbind_filter(tp, &fold->res); 554 tcf_exts_get_net(&fold->exts); 555 call_rcu(&fold->rcu, route4_delete_filter); 556 } 557 return 0; 558 559 errout: 560 if (f) 561 tcf_exts_destroy(&f->exts); 562 kfree(f); 563 return err; 564 } 565 566 static void route4_walk(struct tcf_proto *tp, struct tcf_walker *arg) 567 { 568 struct route4_head *head = rtnl_dereference(tp->root); 569 unsigned int h, h1; 570 571 if (head == NULL) 572 arg->stop = 1; 573 574 if (arg->stop) 575 return; 576 577 for (h = 0; h <= 256; h++) { 578 struct route4_bucket *b = rtnl_dereference(head->table[h]); 579 580 if (b) { 581 for (h1 = 0; h1 <= 32; h1++) { 582 struct route4_filter *f; 583 584 for (f = rtnl_dereference(b->ht[h1]); 585 f; 586 f = rtnl_dereference(f->next)) { 587 if (arg->count < arg->skip) { 588 arg->count++; 589 continue; 590 } 591 if (arg->fn(tp, f, arg) < 0) { 592 arg->stop = 1; 593 return; 594 } 595 arg->count++; 596 } 597 } 598 } 599 } 600 } 601 602 static int route4_dump(struct net *net, struct tcf_proto *tp, void *fh, 603 struct sk_buff *skb, struct tcmsg *t) 604 { 605 struct route4_filter *f = fh; 606 struct nlattr *nest; 607 u32 id; 608 609 if (f == NULL) 610 return skb->len; 611 612 t->tcm_handle = f->handle; 613 614 nest = nla_nest_start(skb, TCA_OPTIONS); 615 if (nest == NULL) 616 goto nla_put_failure; 617 618 if (!(f->handle & 0x8000)) { 619 id = f->id & 0xFF; 620 if (nla_put_u32(skb, TCA_ROUTE4_TO, id)) 621 goto nla_put_failure; 622 } 623 if (f->handle & 0x80000000) { 624 if ((f->handle >> 16) != 0xFFFF && 625 nla_put_u32(skb, TCA_ROUTE4_IIF, f->iif)) 626 goto nla_put_failure; 627 } else { 628 id = f->id >> 16; 629 if (nla_put_u32(skb, TCA_ROUTE4_FROM, id)) 630 goto nla_put_failure; 631 } 632 if (f->res.classid && 633 nla_put_u32(skb, TCA_ROUTE4_CLASSID, f->res.classid)) 634 goto nla_put_failure; 635 636 if (tcf_exts_dump(skb, &f->exts) < 0) 637 goto nla_put_failure; 638 639 nla_nest_end(skb, nest); 640 641 if (tcf_exts_dump_stats(skb, &f->exts) < 0) 642 goto nla_put_failure; 643 644 return skb->len; 645 646 nla_put_failure: 647 nla_nest_cancel(skb, nest); 648 return -1; 649 } 650 651 static void route4_bind_class(void *fh, u32 classid, unsigned long cl) 652 { 653 struct route4_filter *f = fh; 654 655 if (f && f->res.classid == classid) 656 f->res.class = cl; 657 } 658 659 static struct tcf_proto_ops cls_route4_ops __read_mostly = { 660 .kind = "route", 661 .classify = route4_classify, 662 .init = route4_init, 663 .destroy = route4_destroy, 664 .get = route4_get, 665 .change = route4_change, 666 .delete = route4_delete, 667 .walk = route4_walk, 668 .dump = route4_dump, 669 .bind_class = route4_bind_class, 670 .owner = THIS_MODULE, 671 }; 672 673 static int __init init_route4(void) 674 { 675 return register_tcf_proto_ops(&cls_route4_ops); 676 } 677 678 static void __exit exit_route4(void) 679 { 680 unregister_tcf_proto_ops(&cls_route4_ops); 681 } 682 683 module_init(init_route4) 684 module_exit(exit_route4) 685 MODULE_LICENSE("GPL"); 686