1 /* 2 * net/sched/cls_u32.c Ugly (or Universal) 32bit key Packet 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 * The filters are packed to hash tables of key nodes 12 * with a set of 32bit key/mask pairs at every node. 13 * Nodes reference next level hash tables etc. 14 * 15 * This scheme is the best universal classifier I managed to 16 * invent; it is not super-fast, but it is not slow (provided you 17 * program it correctly), and general enough. And its relative 18 * speed grows as the number of rules becomes larger. 19 * 20 * It seems that it represents the best middle point between 21 * speed and manageability both by human and by machine. 22 * 23 * It is especially useful for link sharing combined with QoS; 24 * pure RSVP doesn't need such a general approach and can use 25 * much simpler (and faster) schemes, sort of cls_rsvp.c. 26 * 27 * JHS: We should remove the CONFIG_NET_CLS_IND from here 28 * eventually when the meta match extension is made available 29 * 30 * nfmark match added by Catalin(ux aka Dino) BOIE <catab at umbrella.ro> 31 */ 32 33 #include <linux/module.h> 34 #include <linux/slab.h> 35 #include <linux/types.h> 36 #include <linux/kernel.h> 37 #include <linux/string.h> 38 #include <linux/errno.h> 39 #include <linux/percpu.h> 40 #include <linux/rtnetlink.h> 41 #include <linux/skbuff.h> 42 #include <linux/bitmap.h> 43 #include <linux/netdevice.h> 44 #include <linux/hash.h> 45 #include <net/netlink.h> 46 #include <net/act_api.h> 47 #include <net/pkt_cls.h> 48 #include <linux/netdevice.h> 49 50 struct tc_u_knode { 51 struct tc_u_knode __rcu *next; 52 u32 handle; 53 struct tc_u_hnode __rcu *ht_up; 54 struct tcf_exts exts; 55 #ifdef CONFIG_NET_CLS_IND 56 int ifindex; 57 #endif 58 u8 fshift; 59 struct tcf_result res; 60 struct tc_u_hnode __rcu *ht_down; 61 #ifdef CONFIG_CLS_U32_PERF 62 struct tc_u32_pcnt __percpu *pf; 63 #endif 64 u32 flags; 65 #ifdef CONFIG_CLS_U32_MARK 66 u32 val; 67 u32 mask; 68 u32 __percpu *pcpu_success; 69 #endif 70 struct tcf_proto *tp; 71 union { 72 struct work_struct work; 73 struct rcu_head rcu; 74 }; 75 /* The 'sel' field MUST be the last field in structure to allow for 76 * tc_u32_keys allocated at end of structure. 77 */ 78 struct tc_u32_sel sel; 79 }; 80 81 struct tc_u_hnode { 82 struct tc_u_hnode __rcu *next; 83 u32 handle; 84 u32 prio; 85 struct tc_u_common *tp_c; 86 int refcnt; 87 unsigned int divisor; 88 struct rcu_head rcu; 89 /* The 'ht' field MUST be the last field in structure to allow for 90 * more entries allocated at end of structure. 91 */ 92 struct tc_u_knode __rcu *ht[1]; 93 }; 94 95 struct tc_u_common { 96 struct tc_u_hnode __rcu *hlist; 97 struct Qdisc *q; 98 int refcnt; 99 u32 hgenerator; 100 struct hlist_node hnode; 101 struct rcu_head rcu; 102 }; 103 104 static inline unsigned int u32_hash_fold(__be32 key, 105 const struct tc_u32_sel *sel, 106 u8 fshift) 107 { 108 unsigned int h = ntohl(key & sel->hmask) >> fshift; 109 110 return h; 111 } 112 113 static int u32_classify(struct sk_buff *skb, const struct tcf_proto *tp, 114 struct tcf_result *res) 115 { 116 struct { 117 struct tc_u_knode *knode; 118 unsigned int off; 119 } stack[TC_U32_MAXDEPTH]; 120 121 struct tc_u_hnode *ht = rcu_dereference_bh(tp->root); 122 unsigned int off = skb_network_offset(skb); 123 struct tc_u_knode *n; 124 int sdepth = 0; 125 int off2 = 0; 126 int sel = 0; 127 #ifdef CONFIG_CLS_U32_PERF 128 int j; 129 #endif 130 int i, r; 131 132 next_ht: 133 n = rcu_dereference_bh(ht->ht[sel]); 134 135 next_knode: 136 if (n) { 137 struct tc_u32_key *key = n->sel.keys; 138 139 #ifdef CONFIG_CLS_U32_PERF 140 __this_cpu_inc(n->pf->rcnt); 141 j = 0; 142 #endif 143 144 if (tc_skip_sw(n->flags)) { 145 n = rcu_dereference_bh(n->next); 146 goto next_knode; 147 } 148 149 #ifdef CONFIG_CLS_U32_MARK 150 if ((skb->mark & n->mask) != n->val) { 151 n = rcu_dereference_bh(n->next); 152 goto next_knode; 153 } else { 154 __this_cpu_inc(*n->pcpu_success); 155 } 156 #endif 157 158 for (i = n->sel.nkeys; i > 0; i--, key++) { 159 int toff = off + key->off + (off2 & key->offmask); 160 __be32 *data, hdata; 161 162 if (skb_headroom(skb) + toff > INT_MAX) 163 goto out; 164 165 data = skb_header_pointer(skb, toff, 4, &hdata); 166 if (!data) 167 goto out; 168 if ((*data ^ key->val) & key->mask) { 169 n = rcu_dereference_bh(n->next); 170 goto next_knode; 171 } 172 #ifdef CONFIG_CLS_U32_PERF 173 __this_cpu_inc(n->pf->kcnts[j]); 174 j++; 175 #endif 176 } 177 178 ht = rcu_dereference_bh(n->ht_down); 179 if (!ht) { 180 check_terminal: 181 if (n->sel.flags & TC_U32_TERMINAL) { 182 183 *res = n->res; 184 #ifdef CONFIG_NET_CLS_IND 185 if (!tcf_match_indev(skb, n->ifindex)) { 186 n = rcu_dereference_bh(n->next); 187 goto next_knode; 188 } 189 #endif 190 #ifdef CONFIG_CLS_U32_PERF 191 __this_cpu_inc(n->pf->rhit); 192 #endif 193 r = tcf_exts_exec(skb, &n->exts, res); 194 if (r < 0) { 195 n = rcu_dereference_bh(n->next); 196 goto next_knode; 197 } 198 199 return r; 200 } 201 n = rcu_dereference_bh(n->next); 202 goto next_knode; 203 } 204 205 /* PUSH */ 206 if (sdepth >= TC_U32_MAXDEPTH) 207 goto deadloop; 208 stack[sdepth].knode = n; 209 stack[sdepth].off = off; 210 sdepth++; 211 212 ht = rcu_dereference_bh(n->ht_down); 213 sel = 0; 214 if (ht->divisor) { 215 __be32 *data, hdata; 216 217 data = skb_header_pointer(skb, off + n->sel.hoff, 4, 218 &hdata); 219 if (!data) 220 goto out; 221 sel = ht->divisor & u32_hash_fold(*data, &n->sel, 222 n->fshift); 223 } 224 if (!(n->sel.flags & (TC_U32_VAROFFSET | TC_U32_OFFSET | TC_U32_EAT))) 225 goto next_ht; 226 227 if (n->sel.flags & (TC_U32_OFFSET | TC_U32_VAROFFSET)) { 228 off2 = n->sel.off + 3; 229 if (n->sel.flags & TC_U32_VAROFFSET) { 230 __be16 *data, hdata; 231 232 data = skb_header_pointer(skb, 233 off + n->sel.offoff, 234 2, &hdata); 235 if (!data) 236 goto out; 237 off2 += ntohs(n->sel.offmask & *data) >> 238 n->sel.offshift; 239 } 240 off2 &= ~3; 241 } 242 if (n->sel.flags & TC_U32_EAT) { 243 off += off2; 244 off2 = 0; 245 } 246 247 if (off < skb->len) 248 goto next_ht; 249 } 250 251 /* POP */ 252 if (sdepth--) { 253 n = stack[sdepth].knode; 254 ht = rcu_dereference_bh(n->ht_up); 255 off = stack[sdepth].off; 256 goto check_terminal; 257 } 258 out: 259 return -1; 260 261 deadloop: 262 net_warn_ratelimited("cls_u32: dead loop\n"); 263 return -1; 264 } 265 266 static struct tc_u_hnode *u32_lookup_ht(struct tc_u_common *tp_c, u32 handle) 267 { 268 struct tc_u_hnode *ht; 269 270 for (ht = rtnl_dereference(tp_c->hlist); 271 ht; 272 ht = rtnl_dereference(ht->next)) 273 if (ht->handle == handle) 274 break; 275 276 return ht; 277 } 278 279 static struct tc_u_knode *u32_lookup_key(struct tc_u_hnode *ht, u32 handle) 280 { 281 unsigned int sel; 282 struct tc_u_knode *n = NULL; 283 284 sel = TC_U32_HASH(handle); 285 if (sel > ht->divisor) 286 goto out; 287 288 for (n = rtnl_dereference(ht->ht[sel]); 289 n; 290 n = rtnl_dereference(n->next)) 291 if (n->handle == handle) 292 break; 293 out: 294 return n; 295 } 296 297 298 static void *u32_get(struct tcf_proto *tp, u32 handle) 299 { 300 struct tc_u_hnode *ht; 301 struct tc_u_common *tp_c = tp->data; 302 303 if (TC_U32_HTID(handle) == TC_U32_ROOT) 304 ht = rtnl_dereference(tp->root); 305 else 306 ht = u32_lookup_ht(tp_c, TC_U32_HTID(handle)); 307 308 if (!ht) 309 return NULL; 310 311 if (TC_U32_KEY(handle) == 0) 312 return ht; 313 314 return u32_lookup_key(ht, handle); 315 } 316 317 static u32 gen_new_htid(struct tc_u_common *tp_c) 318 { 319 int i = 0x800; 320 321 /* hgenerator only used inside rtnl lock it is safe to increment 322 * without read _copy_ update semantics 323 */ 324 do { 325 if (++tp_c->hgenerator == 0x7FF) 326 tp_c->hgenerator = 1; 327 } while (--i > 0 && u32_lookup_ht(tp_c, (tp_c->hgenerator|0x800)<<20)); 328 329 return i > 0 ? (tp_c->hgenerator|0x800)<<20 : 0; 330 } 331 332 static struct hlist_head *tc_u_common_hash; 333 334 #define U32_HASH_SHIFT 10 335 #define U32_HASH_SIZE (1 << U32_HASH_SHIFT) 336 337 static unsigned int tc_u_hash(const struct tcf_proto *tp) 338 { 339 struct net_device *dev = tp->q->dev_queue->dev; 340 u32 qhandle = tp->q->handle; 341 int ifindex = dev->ifindex; 342 343 return hash_64((u64)ifindex << 32 | qhandle, U32_HASH_SHIFT); 344 } 345 346 static struct tc_u_common *tc_u_common_find(const struct tcf_proto *tp) 347 { 348 struct tc_u_common *tc; 349 unsigned int h; 350 351 h = tc_u_hash(tp); 352 hlist_for_each_entry(tc, &tc_u_common_hash[h], hnode) { 353 if (tc->q == tp->q) 354 return tc; 355 } 356 return NULL; 357 } 358 359 static int u32_init(struct tcf_proto *tp) 360 { 361 struct tc_u_hnode *root_ht; 362 struct tc_u_common *tp_c; 363 unsigned int h; 364 365 tp_c = tc_u_common_find(tp); 366 367 root_ht = kzalloc(sizeof(*root_ht), GFP_KERNEL); 368 if (root_ht == NULL) 369 return -ENOBUFS; 370 371 root_ht->refcnt++; 372 root_ht->handle = tp_c ? gen_new_htid(tp_c) : 0x80000000; 373 root_ht->prio = tp->prio; 374 375 if (tp_c == NULL) { 376 tp_c = kzalloc(sizeof(*tp_c), GFP_KERNEL); 377 if (tp_c == NULL) { 378 kfree(root_ht); 379 return -ENOBUFS; 380 } 381 tp_c->q = tp->q; 382 INIT_HLIST_NODE(&tp_c->hnode); 383 384 h = tc_u_hash(tp); 385 hlist_add_head(&tp_c->hnode, &tc_u_common_hash[h]); 386 } 387 388 tp_c->refcnt++; 389 RCU_INIT_POINTER(root_ht->next, tp_c->hlist); 390 rcu_assign_pointer(tp_c->hlist, root_ht); 391 root_ht->tp_c = tp_c; 392 393 rcu_assign_pointer(tp->root, root_ht); 394 tp->data = tp_c; 395 return 0; 396 } 397 398 static int u32_destroy_key(struct tcf_proto *tp, struct tc_u_knode *n, 399 bool free_pf) 400 { 401 tcf_exts_destroy(&n->exts); 402 if (n->ht_down) 403 n->ht_down->refcnt--; 404 #ifdef CONFIG_CLS_U32_PERF 405 if (free_pf) 406 free_percpu(n->pf); 407 #endif 408 #ifdef CONFIG_CLS_U32_MARK 409 if (free_pf) 410 free_percpu(n->pcpu_success); 411 #endif 412 kfree(n); 413 return 0; 414 } 415 416 /* u32_delete_key_rcu should be called when free'ing a copied 417 * version of a tc_u_knode obtained from u32_init_knode(). When 418 * copies are obtained from u32_init_knode() the statistics are 419 * shared between the old and new copies to allow readers to 420 * continue to update the statistics during the copy. To support 421 * this the u32_delete_key_rcu variant does not free the percpu 422 * statistics. 423 */ 424 static void u32_delete_key_work(struct work_struct *work) 425 { 426 struct tc_u_knode *key = container_of(work, struct tc_u_knode, work); 427 428 rtnl_lock(); 429 u32_destroy_key(key->tp, key, false); 430 rtnl_unlock(); 431 } 432 433 static void u32_delete_key_rcu(struct rcu_head *rcu) 434 { 435 struct tc_u_knode *key = container_of(rcu, struct tc_u_knode, rcu); 436 437 INIT_WORK(&key->work, u32_delete_key_work); 438 tcf_queue_work(&key->work); 439 } 440 441 /* u32_delete_key_freepf_rcu is the rcu callback variant 442 * that free's the entire structure including the statistics 443 * percpu variables. Only use this if the key is not a copy 444 * returned by u32_init_knode(). See u32_delete_key_rcu() 445 * for the variant that should be used with keys return from 446 * u32_init_knode() 447 */ 448 static void u32_delete_key_freepf_work(struct work_struct *work) 449 { 450 struct tc_u_knode *key = container_of(work, struct tc_u_knode, work); 451 452 rtnl_lock(); 453 u32_destroy_key(key->tp, key, true); 454 rtnl_unlock(); 455 } 456 457 static void u32_delete_key_freepf_rcu(struct rcu_head *rcu) 458 { 459 struct tc_u_knode *key = container_of(rcu, struct tc_u_knode, rcu); 460 461 INIT_WORK(&key->work, u32_delete_key_freepf_work); 462 tcf_queue_work(&key->work); 463 } 464 465 static int u32_delete_key(struct tcf_proto *tp, struct tc_u_knode *key) 466 { 467 struct tc_u_knode __rcu **kp; 468 struct tc_u_knode *pkp; 469 struct tc_u_hnode *ht = rtnl_dereference(key->ht_up); 470 471 if (ht) { 472 kp = &ht->ht[TC_U32_HASH(key->handle)]; 473 for (pkp = rtnl_dereference(*kp); pkp; 474 kp = &pkp->next, pkp = rtnl_dereference(*kp)) { 475 if (pkp == key) { 476 RCU_INIT_POINTER(*kp, key->next); 477 478 tcf_unbind_filter(tp, &key->res); 479 call_rcu(&key->rcu, u32_delete_key_freepf_rcu); 480 return 0; 481 } 482 } 483 } 484 WARN_ON(1); 485 return 0; 486 } 487 488 static void u32_remove_hw_knode(struct tcf_proto *tp, u32 handle) 489 { 490 struct net_device *dev = tp->q->dev_queue->dev; 491 struct tc_cls_u32_offload cls_u32 = {}; 492 493 if (!tc_should_offload(dev, 0)) 494 return; 495 496 tc_cls_common_offload_init(&cls_u32.common, tp); 497 cls_u32.command = TC_CLSU32_DELETE_KNODE; 498 cls_u32.knode.handle = handle; 499 500 dev->netdev_ops->ndo_setup_tc(dev, TC_SETUP_CLSU32, &cls_u32); 501 } 502 503 static int u32_replace_hw_hnode(struct tcf_proto *tp, struct tc_u_hnode *h, 504 u32 flags) 505 { 506 struct net_device *dev = tp->q->dev_queue->dev; 507 struct tc_cls_u32_offload cls_u32 = {}; 508 int err; 509 510 if (!tc_should_offload(dev, flags)) 511 return tc_skip_sw(flags) ? -EINVAL : 0; 512 513 tc_cls_common_offload_init(&cls_u32.common, tp); 514 cls_u32.command = TC_CLSU32_NEW_HNODE; 515 cls_u32.hnode.divisor = h->divisor; 516 cls_u32.hnode.handle = h->handle; 517 cls_u32.hnode.prio = h->prio; 518 519 err = dev->netdev_ops->ndo_setup_tc(dev, TC_SETUP_CLSU32, &cls_u32); 520 if (tc_skip_sw(flags)) 521 return err; 522 523 return 0; 524 } 525 526 static void u32_clear_hw_hnode(struct tcf_proto *tp, struct tc_u_hnode *h) 527 { 528 struct net_device *dev = tp->q->dev_queue->dev; 529 struct tc_cls_u32_offload cls_u32 = {}; 530 531 if (!tc_should_offload(dev, 0)) 532 return; 533 534 tc_cls_common_offload_init(&cls_u32.common, tp); 535 cls_u32.command = TC_CLSU32_DELETE_HNODE; 536 cls_u32.hnode.divisor = h->divisor; 537 cls_u32.hnode.handle = h->handle; 538 cls_u32.hnode.prio = h->prio; 539 540 dev->netdev_ops->ndo_setup_tc(dev, TC_SETUP_CLSU32, &cls_u32); 541 } 542 543 static int u32_replace_hw_knode(struct tcf_proto *tp, struct tc_u_knode *n, 544 u32 flags) 545 { 546 struct net_device *dev = tp->q->dev_queue->dev; 547 struct tc_cls_u32_offload cls_u32 = {}; 548 int err; 549 550 if (!tc_should_offload(dev, flags)) 551 return tc_skip_sw(flags) ? -EINVAL : 0; 552 553 tc_cls_common_offload_init(&cls_u32.common, tp); 554 cls_u32.command = TC_CLSU32_REPLACE_KNODE; 555 cls_u32.knode.handle = n->handle; 556 cls_u32.knode.fshift = n->fshift; 557 #ifdef CONFIG_CLS_U32_MARK 558 cls_u32.knode.val = n->val; 559 cls_u32.knode.mask = n->mask; 560 #else 561 cls_u32.knode.val = 0; 562 cls_u32.knode.mask = 0; 563 #endif 564 cls_u32.knode.sel = &n->sel; 565 cls_u32.knode.exts = &n->exts; 566 if (n->ht_down) 567 cls_u32.knode.link_handle = n->ht_down->handle; 568 569 err = dev->netdev_ops->ndo_setup_tc(dev, TC_SETUP_CLSU32, &cls_u32); 570 571 if (!err) 572 n->flags |= TCA_CLS_FLAGS_IN_HW; 573 574 if (tc_skip_sw(flags)) 575 return err; 576 577 return 0; 578 } 579 580 static void u32_clear_hnode(struct tcf_proto *tp, struct tc_u_hnode *ht) 581 { 582 struct tc_u_knode *n; 583 unsigned int h; 584 585 for (h = 0; h <= ht->divisor; h++) { 586 while ((n = rtnl_dereference(ht->ht[h])) != NULL) { 587 RCU_INIT_POINTER(ht->ht[h], 588 rtnl_dereference(n->next)); 589 tcf_unbind_filter(tp, &n->res); 590 u32_remove_hw_knode(tp, n->handle); 591 call_rcu(&n->rcu, u32_delete_key_freepf_rcu); 592 } 593 } 594 } 595 596 static int u32_destroy_hnode(struct tcf_proto *tp, struct tc_u_hnode *ht) 597 { 598 struct tc_u_common *tp_c = tp->data; 599 struct tc_u_hnode __rcu **hn; 600 struct tc_u_hnode *phn; 601 602 WARN_ON(ht->refcnt); 603 604 u32_clear_hnode(tp, ht); 605 606 hn = &tp_c->hlist; 607 for (phn = rtnl_dereference(*hn); 608 phn; 609 hn = &phn->next, phn = rtnl_dereference(*hn)) { 610 if (phn == ht) { 611 u32_clear_hw_hnode(tp, ht); 612 RCU_INIT_POINTER(*hn, ht->next); 613 kfree_rcu(ht, rcu); 614 return 0; 615 } 616 } 617 618 return -ENOENT; 619 } 620 621 static bool ht_empty(struct tc_u_hnode *ht) 622 { 623 unsigned int h; 624 625 for (h = 0; h <= ht->divisor; h++) 626 if (rcu_access_pointer(ht->ht[h])) 627 return false; 628 629 return true; 630 } 631 632 static void u32_destroy(struct tcf_proto *tp) 633 { 634 struct tc_u_common *tp_c = tp->data; 635 struct tc_u_hnode *root_ht = rtnl_dereference(tp->root); 636 637 WARN_ON(root_ht == NULL); 638 639 if (root_ht && --root_ht->refcnt == 0) 640 u32_destroy_hnode(tp, root_ht); 641 642 if (--tp_c->refcnt == 0) { 643 struct tc_u_hnode *ht; 644 645 hlist_del(&tp_c->hnode); 646 647 for (ht = rtnl_dereference(tp_c->hlist); 648 ht; 649 ht = rtnl_dereference(ht->next)) { 650 ht->refcnt--; 651 u32_clear_hnode(tp, ht); 652 } 653 654 while ((ht = rtnl_dereference(tp_c->hlist)) != NULL) { 655 RCU_INIT_POINTER(tp_c->hlist, ht->next); 656 kfree_rcu(ht, rcu); 657 } 658 659 kfree(tp_c); 660 } 661 662 tp->data = NULL; 663 } 664 665 static int u32_delete(struct tcf_proto *tp, void *arg, bool *last) 666 { 667 struct tc_u_hnode *ht = arg; 668 struct tc_u_hnode *root_ht = rtnl_dereference(tp->root); 669 struct tc_u_common *tp_c = tp->data; 670 int ret = 0; 671 672 if (ht == NULL) 673 goto out; 674 675 if (TC_U32_KEY(ht->handle)) { 676 u32_remove_hw_knode(tp, ht->handle); 677 ret = u32_delete_key(tp, (struct tc_u_knode *)ht); 678 goto out; 679 } 680 681 if (root_ht == ht) 682 return -EINVAL; 683 684 if (ht->refcnt == 1) { 685 ht->refcnt--; 686 u32_destroy_hnode(tp, ht); 687 } else { 688 return -EBUSY; 689 } 690 691 out: 692 *last = true; 693 if (root_ht) { 694 if (root_ht->refcnt > 1) { 695 *last = false; 696 goto ret; 697 } 698 if (root_ht->refcnt == 1) { 699 if (!ht_empty(root_ht)) { 700 *last = false; 701 goto ret; 702 } 703 } 704 } 705 706 if (tp_c->refcnt > 1) { 707 *last = false; 708 goto ret; 709 } 710 711 if (tp_c->refcnt == 1) { 712 struct tc_u_hnode *ht; 713 714 for (ht = rtnl_dereference(tp_c->hlist); 715 ht; 716 ht = rtnl_dereference(ht->next)) 717 if (!ht_empty(ht)) { 718 *last = false; 719 break; 720 } 721 } 722 723 ret: 724 return ret; 725 } 726 727 #define NR_U32_NODE (1<<12) 728 static u32 gen_new_kid(struct tc_u_hnode *ht, u32 handle) 729 { 730 struct tc_u_knode *n; 731 unsigned long i; 732 unsigned long *bitmap = kzalloc(BITS_TO_LONGS(NR_U32_NODE) * sizeof(unsigned long), 733 GFP_KERNEL); 734 if (!bitmap) 735 return handle | 0xFFF; 736 737 for (n = rtnl_dereference(ht->ht[TC_U32_HASH(handle)]); 738 n; 739 n = rtnl_dereference(n->next)) 740 set_bit(TC_U32_NODE(n->handle), bitmap); 741 742 i = find_next_zero_bit(bitmap, NR_U32_NODE, 0x800); 743 if (i >= NR_U32_NODE) 744 i = find_next_zero_bit(bitmap, NR_U32_NODE, 1); 745 746 kfree(bitmap); 747 return handle | (i >= NR_U32_NODE ? 0xFFF : i); 748 } 749 750 static const struct nla_policy u32_policy[TCA_U32_MAX + 1] = { 751 [TCA_U32_CLASSID] = { .type = NLA_U32 }, 752 [TCA_U32_HASH] = { .type = NLA_U32 }, 753 [TCA_U32_LINK] = { .type = NLA_U32 }, 754 [TCA_U32_DIVISOR] = { .type = NLA_U32 }, 755 [TCA_U32_SEL] = { .len = sizeof(struct tc_u32_sel) }, 756 [TCA_U32_INDEV] = { .type = NLA_STRING, .len = IFNAMSIZ }, 757 [TCA_U32_MARK] = { .len = sizeof(struct tc_u32_mark) }, 758 [TCA_U32_FLAGS] = { .type = NLA_U32 }, 759 }; 760 761 static int u32_set_parms(struct net *net, struct tcf_proto *tp, 762 unsigned long base, struct tc_u_hnode *ht, 763 struct tc_u_knode *n, struct nlattr **tb, 764 struct nlattr *est, bool ovr) 765 { 766 int err; 767 768 err = tcf_exts_validate(net, tp, tb, est, &n->exts, ovr); 769 if (err < 0) 770 return err; 771 772 if (tb[TCA_U32_LINK]) { 773 u32 handle = nla_get_u32(tb[TCA_U32_LINK]); 774 struct tc_u_hnode *ht_down = NULL, *ht_old; 775 776 if (TC_U32_KEY(handle)) 777 return -EINVAL; 778 779 if (handle) { 780 ht_down = u32_lookup_ht(ht->tp_c, handle); 781 782 if (ht_down == NULL) 783 return -EINVAL; 784 ht_down->refcnt++; 785 } 786 787 ht_old = rtnl_dereference(n->ht_down); 788 rcu_assign_pointer(n->ht_down, ht_down); 789 790 if (ht_old) 791 ht_old->refcnt--; 792 } 793 if (tb[TCA_U32_CLASSID]) { 794 n->res.classid = nla_get_u32(tb[TCA_U32_CLASSID]); 795 tcf_bind_filter(tp, &n->res, base); 796 } 797 798 #ifdef CONFIG_NET_CLS_IND 799 if (tb[TCA_U32_INDEV]) { 800 int ret; 801 ret = tcf_change_indev(net, tb[TCA_U32_INDEV]); 802 if (ret < 0) 803 return -EINVAL; 804 n->ifindex = ret; 805 } 806 #endif 807 return 0; 808 } 809 810 static void u32_replace_knode(struct tcf_proto *tp, struct tc_u_common *tp_c, 811 struct tc_u_knode *n) 812 { 813 struct tc_u_knode __rcu **ins; 814 struct tc_u_knode *pins; 815 struct tc_u_hnode *ht; 816 817 if (TC_U32_HTID(n->handle) == TC_U32_ROOT) 818 ht = rtnl_dereference(tp->root); 819 else 820 ht = u32_lookup_ht(tp_c, TC_U32_HTID(n->handle)); 821 822 ins = &ht->ht[TC_U32_HASH(n->handle)]; 823 824 /* The node must always exist for it to be replaced if this is not the 825 * case then something went very wrong elsewhere. 826 */ 827 for (pins = rtnl_dereference(*ins); ; 828 ins = &pins->next, pins = rtnl_dereference(*ins)) 829 if (pins->handle == n->handle) 830 break; 831 832 RCU_INIT_POINTER(n->next, pins->next); 833 rcu_assign_pointer(*ins, n); 834 } 835 836 static struct tc_u_knode *u32_init_knode(struct tcf_proto *tp, 837 struct tc_u_knode *n) 838 { 839 struct tc_u_knode *new; 840 struct tc_u32_sel *s = &n->sel; 841 842 new = kzalloc(sizeof(*n) + s->nkeys*sizeof(struct tc_u32_key), 843 GFP_KERNEL); 844 845 if (!new) 846 return NULL; 847 848 RCU_INIT_POINTER(new->next, n->next); 849 new->handle = n->handle; 850 RCU_INIT_POINTER(new->ht_up, n->ht_up); 851 852 #ifdef CONFIG_NET_CLS_IND 853 new->ifindex = n->ifindex; 854 #endif 855 new->fshift = n->fshift; 856 new->res = n->res; 857 new->flags = n->flags; 858 RCU_INIT_POINTER(new->ht_down, n->ht_down); 859 860 /* bump reference count as long as we hold pointer to structure */ 861 if (new->ht_down) 862 new->ht_down->refcnt++; 863 864 #ifdef CONFIG_CLS_U32_PERF 865 /* Statistics may be incremented by readers during update 866 * so we must keep them in tact. When the node is later destroyed 867 * a special destroy call must be made to not free the pf memory. 868 */ 869 new->pf = n->pf; 870 #endif 871 872 #ifdef CONFIG_CLS_U32_MARK 873 new->val = n->val; 874 new->mask = n->mask; 875 /* Similarly success statistics must be moved as pointers */ 876 new->pcpu_success = n->pcpu_success; 877 #endif 878 new->tp = tp; 879 memcpy(&new->sel, s, sizeof(*s) + s->nkeys*sizeof(struct tc_u32_key)); 880 881 if (tcf_exts_init(&new->exts, TCA_U32_ACT, TCA_U32_POLICE)) { 882 kfree(new); 883 return NULL; 884 } 885 886 return new; 887 } 888 889 static int u32_change(struct net *net, struct sk_buff *in_skb, 890 struct tcf_proto *tp, unsigned long base, u32 handle, 891 struct nlattr **tca, void **arg, bool ovr) 892 { 893 struct tc_u_common *tp_c = tp->data; 894 struct tc_u_hnode *ht; 895 struct tc_u_knode *n; 896 struct tc_u32_sel *s; 897 struct nlattr *opt = tca[TCA_OPTIONS]; 898 struct nlattr *tb[TCA_U32_MAX + 1]; 899 u32 htid, flags = 0; 900 int err; 901 #ifdef CONFIG_CLS_U32_PERF 902 size_t size; 903 #endif 904 905 if (opt == NULL) 906 return handle ? -EINVAL : 0; 907 908 err = nla_parse_nested(tb, TCA_U32_MAX, opt, u32_policy, NULL); 909 if (err < 0) 910 return err; 911 912 if (tb[TCA_U32_FLAGS]) { 913 flags = nla_get_u32(tb[TCA_U32_FLAGS]); 914 if (!tc_flags_valid(flags)) 915 return -EINVAL; 916 } 917 918 n = *arg; 919 if (n) { 920 struct tc_u_knode *new; 921 922 if (TC_U32_KEY(n->handle) == 0) 923 return -EINVAL; 924 925 if (n->flags != flags) 926 return -EINVAL; 927 928 new = u32_init_knode(tp, n); 929 if (!new) 930 return -ENOMEM; 931 932 err = u32_set_parms(net, tp, base, 933 rtnl_dereference(n->ht_up), new, tb, 934 tca[TCA_RATE], ovr); 935 936 if (err) { 937 u32_destroy_key(tp, new, false); 938 return err; 939 } 940 941 err = u32_replace_hw_knode(tp, new, flags); 942 if (err) { 943 u32_destroy_key(tp, new, false); 944 return err; 945 } 946 947 if (!tc_in_hw(new->flags)) 948 new->flags |= TCA_CLS_FLAGS_NOT_IN_HW; 949 950 u32_replace_knode(tp, tp_c, new); 951 tcf_unbind_filter(tp, &n->res); 952 call_rcu(&n->rcu, u32_delete_key_rcu); 953 return 0; 954 } 955 956 if (tb[TCA_U32_DIVISOR]) { 957 unsigned int divisor = nla_get_u32(tb[TCA_U32_DIVISOR]); 958 959 if (--divisor > 0x100) 960 return -EINVAL; 961 if (TC_U32_KEY(handle)) 962 return -EINVAL; 963 if (handle == 0) { 964 handle = gen_new_htid(tp->data); 965 if (handle == 0) 966 return -ENOMEM; 967 } 968 ht = kzalloc(sizeof(*ht) + divisor*sizeof(void *), GFP_KERNEL); 969 if (ht == NULL) 970 return -ENOBUFS; 971 ht->tp_c = tp_c; 972 ht->refcnt = 1; 973 ht->divisor = divisor; 974 ht->handle = handle; 975 ht->prio = tp->prio; 976 977 err = u32_replace_hw_hnode(tp, ht, flags); 978 if (err) { 979 kfree(ht); 980 return err; 981 } 982 983 RCU_INIT_POINTER(ht->next, tp_c->hlist); 984 rcu_assign_pointer(tp_c->hlist, ht); 985 *arg = ht; 986 987 return 0; 988 } 989 990 if (tb[TCA_U32_HASH]) { 991 htid = nla_get_u32(tb[TCA_U32_HASH]); 992 if (TC_U32_HTID(htid) == TC_U32_ROOT) { 993 ht = rtnl_dereference(tp->root); 994 htid = ht->handle; 995 } else { 996 ht = u32_lookup_ht(tp->data, TC_U32_HTID(htid)); 997 if (ht == NULL) 998 return -EINVAL; 999 } 1000 } else { 1001 ht = rtnl_dereference(tp->root); 1002 htid = ht->handle; 1003 } 1004 1005 if (ht->divisor < TC_U32_HASH(htid)) 1006 return -EINVAL; 1007 1008 if (handle) { 1009 if (TC_U32_HTID(handle) && TC_U32_HTID(handle^htid)) 1010 return -EINVAL; 1011 handle = htid | TC_U32_NODE(handle); 1012 } else 1013 handle = gen_new_kid(ht, htid); 1014 1015 if (tb[TCA_U32_SEL] == NULL) 1016 return -EINVAL; 1017 1018 s = nla_data(tb[TCA_U32_SEL]); 1019 1020 n = kzalloc(sizeof(*n) + s->nkeys*sizeof(struct tc_u32_key), GFP_KERNEL); 1021 if (n == NULL) 1022 return -ENOBUFS; 1023 1024 #ifdef CONFIG_CLS_U32_PERF 1025 size = sizeof(struct tc_u32_pcnt) + s->nkeys * sizeof(u64); 1026 n->pf = __alloc_percpu(size, __alignof__(struct tc_u32_pcnt)); 1027 if (!n->pf) { 1028 kfree(n); 1029 return -ENOBUFS; 1030 } 1031 #endif 1032 1033 memcpy(&n->sel, s, sizeof(*s) + s->nkeys*sizeof(struct tc_u32_key)); 1034 RCU_INIT_POINTER(n->ht_up, ht); 1035 n->handle = handle; 1036 n->fshift = s->hmask ? ffs(ntohl(s->hmask)) - 1 : 0; 1037 n->flags = flags; 1038 n->tp = tp; 1039 1040 err = tcf_exts_init(&n->exts, TCA_U32_ACT, TCA_U32_POLICE); 1041 if (err < 0) 1042 goto errout; 1043 1044 #ifdef CONFIG_CLS_U32_MARK 1045 n->pcpu_success = alloc_percpu(u32); 1046 if (!n->pcpu_success) { 1047 err = -ENOMEM; 1048 goto errout; 1049 } 1050 1051 if (tb[TCA_U32_MARK]) { 1052 struct tc_u32_mark *mark; 1053 1054 mark = nla_data(tb[TCA_U32_MARK]); 1055 n->val = mark->val; 1056 n->mask = mark->mask; 1057 } 1058 #endif 1059 1060 err = u32_set_parms(net, tp, base, ht, n, tb, tca[TCA_RATE], ovr); 1061 if (err == 0) { 1062 struct tc_u_knode __rcu **ins; 1063 struct tc_u_knode *pins; 1064 1065 err = u32_replace_hw_knode(tp, n, flags); 1066 if (err) 1067 goto errhw; 1068 1069 if (!tc_in_hw(n->flags)) 1070 n->flags |= TCA_CLS_FLAGS_NOT_IN_HW; 1071 1072 ins = &ht->ht[TC_U32_HASH(handle)]; 1073 for (pins = rtnl_dereference(*ins); pins; 1074 ins = &pins->next, pins = rtnl_dereference(*ins)) 1075 if (TC_U32_NODE(handle) < TC_U32_NODE(pins->handle)) 1076 break; 1077 1078 RCU_INIT_POINTER(n->next, pins); 1079 rcu_assign_pointer(*ins, n); 1080 *arg = n; 1081 return 0; 1082 } 1083 1084 errhw: 1085 #ifdef CONFIG_CLS_U32_MARK 1086 free_percpu(n->pcpu_success); 1087 #endif 1088 1089 errout: 1090 tcf_exts_destroy(&n->exts); 1091 #ifdef CONFIG_CLS_U32_PERF 1092 free_percpu(n->pf); 1093 #endif 1094 kfree(n); 1095 return err; 1096 } 1097 1098 static void u32_walk(struct tcf_proto *tp, struct tcf_walker *arg) 1099 { 1100 struct tc_u_common *tp_c = tp->data; 1101 struct tc_u_hnode *ht; 1102 struct tc_u_knode *n; 1103 unsigned int h; 1104 1105 if (arg->stop) 1106 return; 1107 1108 for (ht = rtnl_dereference(tp_c->hlist); 1109 ht; 1110 ht = rtnl_dereference(ht->next)) { 1111 if (ht->prio != tp->prio) 1112 continue; 1113 if (arg->count >= arg->skip) { 1114 if (arg->fn(tp, ht, arg) < 0) { 1115 arg->stop = 1; 1116 return; 1117 } 1118 } 1119 arg->count++; 1120 for (h = 0; h <= ht->divisor; h++) { 1121 for (n = rtnl_dereference(ht->ht[h]); 1122 n; 1123 n = rtnl_dereference(n->next)) { 1124 if (arg->count < arg->skip) { 1125 arg->count++; 1126 continue; 1127 } 1128 if (arg->fn(tp, n, arg) < 0) { 1129 arg->stop = 1; 1130 return; 1131 } 1132 arg->count++; 1133 } 1134 } 1135 } 1136 } 1137 1138 static void u32_bind_class(void *fh, u32 classid, unsigned long cl) 1139 { 1140 struct tc_u_knode *n = fh; 1141 1142 if (n && n->res.classid == classid) 1143 n->res.class = cl; 1144 } 1145 1146 static int u32_dump(struct net *net, struct tcf_proto *tp, void *fh, 1147 struct sk_buff *skb, struct tcmsg *t) 1148 { 1149 struct tc_u_knode *n = fh; 1150 struct tc_u_hnode *ht_up, *ht_down; 1151 struct nlattr *nest; 1152 1153 if (n == NULL) 1154 return skb->len; 1155 1156 t->tcm_handle = n->handle; 1157 1158 nest = nla_nest_start(skb, TCA_OPTIONS); 1159 if (nest == NULL) 1160 goto nla_put_failure; 1161 1162 if (TC_U32_KEY(n->handle) == 0) { 1163 struct tc_u_hnode *ht = fh; 1164 u32 divisor = ht->divisor + 1; 1165 1166 if (nla_put_u32(skb, TCA_U32_DIVISOR, divisor)) 1167 goto nla_put_failure; 1168 } else { 1169 #ifdef CONFIG_CLS_U32_PERF 1170 struct tc_u32_pcnt *gpf; 1171 int cpu; 1172 #endif 1173 1174 if (nla_put(skb, TCA_U32_SEL, 1175 sizeof(n->sel) + n->sel.nkeys*sizeof(struct tc_u32_key), 1176 &n->sel)) 1177 goto nla_put_failure; 1178 1179 ht_up = rtnl_dereference(n->ht_up); 1180 if (ht_up) { 1181 u32 htid = n->handle & 0xFFFFF000; 1182 if (nla_put_u32(skb, TCA_U32_HASH, htid)) 1183 goto nla_put_failure; 1184 } 1185 if (n->res.classid && 1186 nla_put_u32(skb, TCA_U32_CLASSID, n->res.classid)) 1187 goto nla_put_failure; 1188 1189 ht_down = rtnl_dereference(n->ht_down); 1190 if (ht_down && 1191 nla_put_u32(skb, TCA_U32_LINK, ht_down->handle)) 1192 goto nla_put_failure; 1193 1194 if (n->flags && nla_put_u32(skb, TCA_U32_FLAGS, n->flags)) 1195 goto nla_put_failure; 1196 1197 #ifdef CONFIG_CLS_U32_MARK 1198 if ((n->val || n->mask)) { 1199 struct tc_u32_mark mark = {.val = n->val, 1200 .mask = n->mask, 1201 .success = 0}; 1202 int cpum; 1203 1204 for_each_possible_cpu(cpum) { 1205 __u32 cnt = *per_cpu_ptr(n->pcpu_success, cpum); 1206 1207 mark.success += cnt; 1208 } 1209 1210 if (nla_put(skb, TCA_U32_MARK, sizeof(mark), &mark)) 1211 goto nla_put_failure; 1212 } 1213 #endif 1214 1215 if (tcf_exts_dump(skb, &n->exts) < 0) 1216 goto nla_put_failure; 1217 1218 #ifdef CONFIG_NET_CLS_IND 1219 if (n->ifindex) { 1220 struct net_device *dev; 1221 dev = __dev_get_by_index(net, n->ifindex); 1222 if (dev && nla_put_string(skb, TCA_U32_INDEV, dev->name)) 1223 goto nla_put_failure; 1224 } 1225 #endif 1226 #ifdef CONFIG_CLS_U32_PERF 1227 gpf = kzalloc(sizeof(struct tc_u32_pcnt) + 1228 n->sel.nkeys * sizeof(u64), 1229 GFP_KERNEL); 1230 if (!gpf) 1231 goto nla_put_failure; 1232 1233 for_each_possible_cpu(cpu) { 1234 int i; 1235 struct tc_u32_pcnt *pf = per_cpu_ptr(n->pf, cpu); 1236 1237 gpf->rcnt += pf->rcnt; 1238 gpf->rhit += pf->rhit; 1239 for (i = 0; i < n->sel.nkeys; i++) 1240 gpf->kcnts[i] += pf->kcnts[i]; 1241 } 1242 1243 if (nla_put_64bit(skb, TCA_U32_PCNT, 1244 sizeof(struct tc_u32_pcnt) + 1245 n->sel.nkeys * sizeof(u64), 1246 gpf, TCA_U32_PAD)) { 1247 kfree(gpf); 1248 goto nla_put_failure; 1249 } 1250 kfree(gpf); 1251 #endif 1252 } 1253 1254 nla_nest_end(skb, nest); 1255 1256 if (TC_U32_KEY(n->handle)) 1257 if (tcf_exts_dump_stats(skb, &n->exts) < 0) 1258 goto nla_put_failure; 1259 return skb->len; 1260 1261 nla_put_failure: 1262 nla_nest_cancel(skb, nest); 1263 return -1; 1264 } 1265 1266 static struct tcf_proto_ops cls_u32_ops __read_mostly = { 1267 .kind = "u32", 1268 .classify = u32_classify, 1269 .init = u32_init, 1270 .destroy = u32_destroy, 1271 .get = u32_get, 1272 .change = u32_change, 1273 .delete = u32_delete, 1274 .walk = u32_walk, 1275 .dump = u32_dump, 1276 .bind_class = u32_bind_class, 1277 .owner = THIS_MODULE, 1278 }; 1279 1280 static int __init init_u32(void) 1281 { 1282 int i, ret; 1283 1284 pr_info("u32 classifier\n"); 1285 #ifdef CONFIG_CLS_U32_PERF 1286 pr_info(" Performance counters on\n"); 1287 #endif 1288 #ifdef CONFIG_NET_CLS_IND 1289 pr_info(" input device check on\n"); 1290 #endif 1291 #ifdef CONFIG_NET_CLS_ACT 1292 pr_info(" Actions configured\n"); 1293 #endif 1294 tc_u_common_hash = kvmalloc_array(U32_HASH_SIZE, 1295 sizeof(struct hlist_head), 1296 GFP_KERNEL); 1297 if (!tc_u_common_hash) 1298 return -ENOMEM; 1299 1300 for (i = 0; i < U32_HASH_SIZE; i++) 1301 INIT_HLIST_HEAD(&tc_u_common_hash[i]); 1302 1303 ret = register_tcf_proto_ops(&cls_u32_ops); 1304 if (ret) 1305 kvfree(tc_u_common_hash); 1306 return ret; 1307 } 1308 1309 static void __exit exit_u32(void) 1310 { 1311 unregister_tcf_proto_ops(&cls_u32_ops); 1312 kvfree(tc_u_common_hash); 1313 } 1314 1315 module_init(init_u32) 1316 module_exit(exit_u32) 1317 MODULE_LICENSE("GPL"); 1318