cls_u32.c revision a2aeb02a8e6a9fef397c344245a54eeae67341f6
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 <net/netlink.h> 44#include <net/act_api.h> 45#include <net/pkt_cls.h> 46 47struct tc_u_knode { 48 struct tc_u_knode __rcu *next; 49 u32 handle; 50 struct tc_u_hnode __rcu *ht_up; 51 struct tcf_exts exts; 52#ifdef CONFIG_NET_CLS_IND 53 int ifindex; 54#endif 55 u8 fshift; 56 struct tcf_result res; 57 struct tc_u_hnode __rcu *ht_down; 58#ifdef CONFIG_CLS_U32_PERF 59 struct tc_u32_pcnt __percpu *pf; 60#endif 61#ifdef CONFIG_CLS_U32_MARK 62 u32 val; 63 u32 mask; 64 u32 __percpu *pcpu_success; 65#endif 66 struct tcf_proto *tp; 67 struct rcu_head rcu; 68 /* The 'sel' field MUST be the last field in structure to allow for 69 * tc_u32_keys allocated at end of structure. 70 */ 71 struct tc_u32_sel sel; 72}; 73 74struct tc_u_hnode { 75 struct tc_u_hnode __rcu *next; 76 u32 handle; 77 u32 prio; 78 struct tc_u_common *tp_c; 79 int refcnt; 80 unsigned int divisor; 81 struct tc_u_knode __rcu *ht[1]; 82 struct rcu_head rcu; 83}; 84 85struct tc_u_common { 86 struct tc_u_hnode __rcu *hlist; 87 struct Qdisc *q; 88 int refcnt; 89 u32 hgenerator; 90 struct rcu_head rcu; 91}; 92 93static inline unsigned int u32_hash_fold(__be32 key, 94 const struct tc_u32_sel *sel, 95 u8 fshift) 96{ 97 unsigned int h = ntohl(key & sel->hmask) >> fshift; 98 99 return h; 100} 101 102static int u32_classify(struct sk_buff *skb, const struct tcf_proto *tp, struct tcf_result *res) 103{ 104 struct { 105 struct tc_u_knode *knode; 106 unsigned int off; 107 } stack[TC_U32_MAXDEPTH]; 108 109 struct tc_u_hnode *ht = rcu_dereference_bh(tp->root); 110 unsigned int off = skb_network_offset(skb); 111 struct tc_u_knode *n; 112 int sdepth = 0; 113 int off2 = 0; 114 int sel = 0; 115#ifdef CONFIG_CLS_U32_PERF 116 int j; 117#endif 118 int i, r; 119 120next_ht: 121 n = rcu_dereference_bh(ht->ht[sel]); 122 123next_knode: 124 if (n) { 125 struct tc_u32_key *key = n->sel.keys; 126 127#ifdef CONFIG_CLS_U32_PERF 128 __this_cpu_inc(n->pf->rcnt); 129 j = 0; 130#endif 131 132#ifdef CONFIG_CLS_U32_MARK 133 if ((skb->mark & n->mask) != n->val) { 134 n = rcu_dereference_bh(n->next); 135 goto next_knode; 136 } else { 137 __this_cpu_inc(*n->pcpu_success); 138 } 139#endif 140 141 for (i = n->sel.nkeys; i > 0; i--, key++) { 142 int toff = off + key->off + (off2 & key->offmask); 143 __be32 *data, hdata; 144 145 if (skb_headroom(skb) + toff > INT_MAX) 146 goto out; 147 148 data = skb_header_pointer(skb, toff, 4, &hdata); 149 if (!data) 150 goto out; 151 if ((*data ^ key->val) & key->mask) { 152 n = rcu_dereference_bh(n->next); 153 goto next_knode; 154 } 155#ifdef CONFIG_CLS_U32_PERF 156 __this_cpu_inc(n->pf->kcnts[j]); 157 j++; 158#endif 159 } 160 161 ht = rcu_dereference_bh(n->ht_down); 162 if (!ht) { 163check_terminal: 164 if (n->sel.flags & TC_U32_TERMINAL) { 165 166 *res = n->res; 167#ifdef CONFIG_NET_CLS_IND 168 if (!tcf_match_indev(skb, n->ifindex)) { 169 n = rcu_dereference_bh(n->next); 170 goto next_knode; 171 } 172#endif 173#ifdef CONFIG_CLS_U32_PERF 174 __this_cpu_inc(n->pf->rhit); 175#endif 176 r = tcf_exts_exec(skb, &n->exts, res); 177 if (r < 0) { 178 n = rcu_dereference_bh(n->next); 179 goto next_knode; 180 } 181 182 return r; 183 } 184 n = rcu_dereference_bh(n->next); 185 goto next_knode; 186 } 187 188 /* PUSH */ 189 if (sdepth >= TC_U32_MAXDEPTH) 190 goto deadloop; 191 stack[sdepth].knode = n; 192 stack[sdepth].off = off; 193 sdepth++; 194 195 ht = rcu_dereference_bh(n->ht_down); 196 sel = 0; 197 if (ht->divisor) { 198 __be32 *data, hdata; 199 200 data = skb_header_pointer(skb, off + n->sel.hoff, 4, 201 &hdata); 202 if (!data) 203 goto out; 204 sel = ht->divisor & u32_hash_fold(*data, &n->sel, 205 n->fshift); 206 } 207 if (!(n->sel.flags & (TC_U32_VAROFFSET | TC_U32_OFFSET | TC_U32_EAT))) 208 goto next_ht; 209 210 if (n->sel.flags & (TC_U32_OFFSET | TC_U32_VAROFFSET)) { 211 off2 = n->sel.off + 3; 212 if (n->sel.flags & TC_U32_VAROFFSET) { 213 __be16 *data, hdata; 214 215 data = skb_header_pointer(skb, 216 off + n->sel.offoff, 217 2, &hdata); 218 if (!data) 219 goto out; 220 off2 += ntohs(n->sel.offmask & *data) >> 221 n->sel.offshift; 222 } 223 off2 &= ~3; 224 } 225 if (n->sel.flags & TC_U32_EAT) { 226 off += off2; 227 off2 = 0; 228 } 229 230 if (off < skb->len) 231 goto next_ht; 232 } 233 234 /* POP */ 235 if (sdepth--) { 236 n = stack[sdepth].knode; 237 ht = rcu_dereference_bh(n->ht_up); 238 off = stack[sdepth].off; 239 goto check_terminal; 240 } 241out: 242 return -1; 243 244deadloop: 245 net_warn_ratelimited("cls_u32: dead loop\n"); 246 return -1; 247} 248 249static struct tc_u_hnode * 250u32_lookup_ht(struct tc_u_common *tp_c, u32 handle) 251{ 252 struct tc_u_hnode *ht; 253 254 for (ht = rtnl_dereference(tp_c->hlist); 255 ht; 256 ht = rtnl_dereference(ht->next)) 257 if (ht->handle == handle) 258 break; 259 260 return ht; 261} 262 263static struct tc_u_knode * 264u32_lookup_key(struct tc_u_hnode *ht, u32 handle) 265{ 266 unsigned int sel; 267 struct tc_u_knode *n = NULL; 268 269 sel = TC_U32_HASH(handle); 270 if (sel > ht->divisor) 271 goto out; 272 273 for (n = rtnl_dereference(ht->ht[sel]); 274 n; 275 n = rtnl_dereference(n->next)) 276 if (n->handle == handle) 277 break; 278out: 279 return n; 280} 281 282 283static unsigned long u32_get(struct tcf_proto *tp, u32 handle) 284{ 285 struct tc_u_hnode *ht; 286 struct tc_u_common *tp_c = tp->data; 287 288 if (TC_U32_HTID(handle) == TC_U32_ROOT) 289 ht = rtnl_dereference(tp->root); 290 else 291 ht = u32_lookup_ht(tp_c, TC_U32_HTID(handle)); 292 293 if (!ht) 294 return 0; 295 296 if (TC_U32_KEY(handle) == 0) 297 return (unsigned long)ht; 298 299 return (unsigned long)u32_lookup_key(ht, handle); 300} 301 302static void u32_put(struct tcf_proto *tp, unsigned long f) 303{ 304} 305 306static u32 gen_new_htid(struct tc_u_common *tp_c) 307{ 308 int i = 0x800; 309 310 /* hgenerator only used inside rtnl lock it is safe to increment 311 * without read _copy_ update semantics 312 */ 313 do { 314 if (++tp_c->hgenerator == 0x7FF) 315 tp_c->hgenerator = 1; 316 } while (--i > 0 && u32_lookup_ht(tp_c, (tp_c->hgenerator|0x800)<<20)); 317 318 return i > 0 ? (tp_c->hgenerator|0x800)<<20 : 0; 319} 320 321static int u32_init(struct tcf_proto *tp) 322{ 323 struct tc_u_hnode *root_ht; 324 struct tc_u_common *tp_c; 325 326 tp_c = tp->q->u32_node; 327 328 root_ht = kzalloc(sizeof(*root_ht), GFP_KERNEL); 329 if (root_ht == NULL) 330 return -ENOBUFS; 331 332 root_ht->divisor = 0; 333 root_ht->refcnt++; 334 root_ht->handle = tp_c ? gen_new_htid(tp_c) : 0x80000000; 335 root_ht->prio = tp->prio; 336 337 if (tp_c == NULL) { 338 tp_c = kzalloc(sizeof(*tp_c), GFP_KERNEL); 339 if (tp_c == NULL) { 340 kfree(root_ht); 341 return -ENOBUFS; 342 } 343 tp_c->q = tp->q; 344 tp->q->u32_node = tp_c; 345 } 346 347 tp_c->refcnt++; 348 RCU_INIT_POINTER(root_ht->next, tp_c->hlist); 349 rcu_assign_pointer(tp_c->hlist, root_ht); 350 root_ht->tp_c = tp_c; 351 352 rcu_assign_pointer(tp->root, root_ht); 353 tp->data = tp_c; 354 return 0; 355} 356 357static int u32_destroy_key(struct tcf_proto *tp, 358 struct tc_u_knode *n, 359 bool free_pf) 360{ 361 tcf_unbind_filter(tp, &n->res); 362 tcf_exts_destroy(tp, &n->exts); 363 if (n->ht_down) 364 n->ht_down->refcnt--; 365#ifdef CONFIG_CLS_U32_PERF 366 if (free_pf) 367 free_percpu(n->pf); 368#endif 369#ifdef CONFIG_CLS_U32_MARK 370 if (free_pf) 371 free_percpu(n->pcpu_success); 372#endif 373 kfree(n); 374 return 0; 375} 376 377/* u32_delete_key_rcu should be called when free'ing a copied 378 * version of a tc_u_knode obtained from u32_init_knode(). When 379 * copies are obtained from u32_init_knode() the statistics are 380 * shared between the old and new copies to allow readers to 381 * continue to update the statistics during the copy. To support 382 * this the u32_delete_key_rcu variant does not free the percpu 383 * statistics. 384 */ 385static void u32_delete_key_rcu(struct rcu_head *rcu) 386{ 387 struct tc_u_knode *key = container_of(rcu, struct tc_u_knode, rcu); 388 389 u32_destroy_key(key->tp, key, false); 390} 391 392/* u32_delete_key_freepf_rcu is the rcu callback variant 393 * that free's the entire structure including the statistics 394 * percpu variables. Only use this if the key is not a copy 395 * returned by u32_init_knode(). See u32_delete_key_rcu() 396 * for the variant that should be used with keys return from 397 * u32_init_knode() 398 */ 399static void u32_delete_key_freepf_rcu(struct rcu_head *rcu) 400{ 401 struct tc_u_knode *key = container_of(rcu, struct tc_u_knode, rcu); 402 403 u32_destroy_key(key->tp, key, true); 404} 405 406static int u32_delete_key(struct tcf_proto *tp, struct tc_u_knode *key) 407{ 408 struct tc_u_knode __rcu **kp; 409 struct tc_u_knode *pkp; 410 struct tc_u_hnode *ht = rtnl_dereference(key->ht_up); 411 412 if (ht) { 413 kp = &ht->ht[TC_U32_HASH(key->handle)]; 414 for (pkp = rtnl_dereference(*kp); pkp; 415 kp = &pkp->next, pkp = rtnl_dereference(*kp)) { 416 if (pkp == key) { 417 RCU_INIT_POINTER(*kp, key->next); 418 419 call_rcu(&key->rcu, u32_delete_key_freepf_rcu); 420 return 0; 421 } 422 } 423 } 424 WARN_ON(1); 425 return 0; 426} 427 428static void u32_clear_hnode(struct tc_u_hnode *ht) 429{ 430 struct tc_u_knode *n; 431 unsigned int h; 432 433 for (h = 0; h <= ht->divisor; h++) { 434 while ((n = rtnl_dereference(ht->ht[h])) != NULL) { 435 RCU_INIT_POINTER(ht->ht[h], 436 rtnl_dereference(n->next)); 437 call_rcu(&n->rcu, u32_delete_key_freepf_rcu); 438 } 439 } 440} 441 442static int u32_destroy_hnode(struct tcf_proto *tp, struct tc_u_hnode *ht) 443{ 444 struct tc_u_common *tp_c = tp->data; 445 struct tc_u_hnode __rcu **hn; 446 struct tc_u_hnode *phn; 447 448 WARN_ON(ht->refcnt); 449 450 u32_clear_hnode(ht); 451 452 hn = &tp_c->hlist; 453 for (phn = rtnl_dereference(*hn); 454 phn; 455 hn = &phn->next, phn = rtnl_dereference(*hn)) { 456 if (phn == ht) { 457 RCU_INIT_POINTER(*hn, ht->next); 458 kfree_rcu(ht, rcu); 459 return 0; 460 } 461 } 462 463 return -ENOENT; 464} 465 466static void u32_destroy(struct tcf_proto *tp) 467{ 468 struct tc_u_common *tp_c = tp->data; 469 struct tc_u_hnode *root_ht = rtnl_dereference(tp->root); 470 471 WARN_ON(root_ht == NULL); 472 473 if (root_ht && --root_ht->refcnt == 0) 474 u32_destroy_hnode(tp, root_ht); 475 476 if (--tp_c->refcnt == 0) { 477 struct tc_u_hnode *ht; 478 479 tp->q->u32_node = NULL; 480 481 for (ht = rtnl_dereference(tp_c->hlist); 482 ht; 483 ht = rtnl_dereference(ht->next)) { 484 ht->refcnt--; 485 u32_clear_hnode(ht); 486 } 487 488 while ((ht = rtnl_dereference(tp_c->hlist)) != NULL) { 489 RCU_INIT_POINTER(tp_c->hlist, ht->next); 490 kfree_rcu(ht, rcu); 491 } 492 493 kfree(tp_c); 494 } 495 496 tp->data = NULL; 497} 498 499static int u32_delete(struct tcf_proto *tp, unsigned long arg) 500{ 501 struct tc_u_hnode *ht = (struct tc_u_hnode *)arg; 502 struct tc_u_hnode *root_ht = rtnl_dereference(tp->root); 503 504 if (ht == NULL) 505 return 0; 506 507 if (TC_U32_KEY(ht->handle)) 508 return u32_delete_key(tp, (struct tc_u_knode *)ht); 509 510 if (root_ht == ht) 511 return -EINVAL; 512 513 if (ht->refcnt == 1) { 514 ht->refcnt--; 515 u32_destroy_hnode(tp, ht); 516 } else { 517 return -EBUSY; 518 } 519 520 return 0; 521} 522 523#define NR_U32_NODE (1<<12) 524static u32 gen_new_kid(struct tc_u_hnode *ht, u32 handle) 525{ 526 struct tc_u_knode *n; 527 unsigned long i; 528 unsigned long *bitmap = kzalloc(BITS_TO_LONGS(NR_U32_NODE) * sizeof(unsigned long), 529 GFP_KERNEL); 530 if (!bitmap) 531 return handle | 0xFFF; 532 533 for (n = rtnl_dereference(ht->ht[TC_U32_HASH(handle)]); 534 n; 535 n = rtnl_dereference(n->next)) 536 set_bit(TC_U32_NODE(n->handle), bitmap); 537 538 i = find_next_zero_bit(bitmap, NR_U32_NODE, 0x800); 539 if (i >= NR_U32_NODE) 540 i = find_next_zero_bit(bitmap, NR_U32_NODE, 1); 541 542 kfree(bitmap); 543 return handle | (i >= NR_U32_NODE ? 0xFFF : i); 544} 545 546static const struct nla_policy u32_policy[TCA_U32_MAX + 1] = { 547 [TCA_U32_CLASSID] = { .type = NLA_U32 }, 548 [TCA_U32_HASH] = { .type = NLA_U32 }, 549 [TCA_U32_LINK] = { .type = NLA_U32 }, 550 [TCA_U32_DIVISOR] = { .type = NLA_U32 }, 551 [TCA_U32_SEL] = { .len = sizeof(struct tc_u32_sel) }, 552 [TCA_U32_INDEV] = { .type = NLA_STRING, .len = IFNAMSIZ }, 553 [TCA_U32_MARK] = { .len = sizeof(struct tc_u32_mark) }, 554}; 555 556static int u32_set_parms(struct net *net, struct tcf_proto *tp, 557 unsigned long base, struct tc_u_hnode *ht, 558 struct tc_u_knode *n, struct nlattr **tb, 559 struct nlattr *est, bool ovr) 560{ 561 int err; 562 struct tcf_exts e; 563 564 tcf_exts_init(&e, TCA_U32_ACT, TCA_U32_POLICE); 565 err = tcf_exts_validate(net, tp, tb, est, &e, ovr); 566 if (err < 0) 567 return err; 568 569 err = -EINVAL; 570 if (tb[TCA_U32_LINK]) { 571 u32 handle = nla_get_u32(tb[TCA_U32_LINK]); 572 struct tc_u_hnode *ht_down = NULL, *ht_old; 573 574 if (TC_U32_KEY(handle)) 575 goto errout; 576 577 if (handle) { 578 ht_down = u32_lookup_ht(ht->tp_c, handle); 579 580 if (ht_down == NULL) 581 goto errout; 582 ht_down->refcnt++; 583 } 584 585 ht_old = rtnl_dereference(n->ht_down); 586 rcu_assign_pointer(n->ht_down, ht_down); 587 588 if (ht_old) 589 ht_old->refcnt--; 590 } 591 if (tb[TCA_U32_CLASSID]) { 592 n->res.classid = nla_get_u32(tb[TCA_U32_CLASSID]); 593 tcf_bind_filter(tp, &n->res, base); 594 } 595 596#ifdef CONFIG_NET_CLS_IND 597 if (tb[TCA_U32_INDEV]) { 598 int ret; 599 ret = tcf_change_indev(net, tb[TCA_U32_INDEV]); 600 if (ret < 0) 601 goto errout; 602 n->ifindex = ret; 603 } 604#endif 605 tcf_exts_change(tp, &n->exts, &e); 606 607 return 0; 608errout: 609 tcf_exts_destroy(tp, &e); 610 return err; 611} 612 613static void u32_replace_knode(struct tcf_proto *tp, 614 struct tc_u_common *tp_c, 615 struct tc_u_knode *n) 616{ 617 struct tc_u_knode __rcu **ins; 618 struct tc_u_knode *pins; 619 struct tc_u_hnode *ht; 620 621 if (TC_U32_HTID(n->handle) == TC_U32_ROOT) 622 ht = rtnl_dereference(tp->root); 623 else 624 ht = u32_lookup_ht(tp_c, TC_U32_HTID(n->handle)); 625 626 ins = &ht->ht[TC_U32_HASH(n->handle)]; 627 628 /* The node must always exist for it to be replaced if this is not the 629 * case then something went very wrong elsewhere. 630 */ 631 for (pins = rtnl_dereference(*ins); ; 632 ins = &pins->next, pins = rtnl_dereference(*ins)) 633 if (pins->handle == n->handle) 634 break; 635 636 RCU_INIT_POINTER(n->next, pins->next); 637 rcu_assign_pointer(*ins, n); 638} 639 640static struct tc_u_knode *u32_init_knode(struct tcf_proto *tp, 641 struct tc_u_knode *n) 642{ 643 struct tc_u_knode *new; 644 struct tc_u32_sel *s = &n->sel; 645 646 new = kzalloc(sizeof(*n) + s->nkeys*sizeof(struct tc_u32_key), 647 GFP_KERNEL); 648 649 if (!new) 650 return NULL; 651 652 RCU_INIT_POINTER(new->next, n->next); 653 new->handle = n->handle; 654 RCU_INIT_POINTER(new->ht_up, n->ht_up); 655 656#ifdef CONFIG_NET_CLS_IND 657 new->ifindex = n->ifindex; 658#endif 659 new->fshift = n->fshift; 660 new->res = n->res; 661 RCU_INIT_POINTER(new->ht_down, n->ht_down); 662 663 /* bump reference count as long as we hold pointer to structure */ 664 if (new->ht_down) 665 new->ht_down->refcnt++; 666 667#ifdef CONFIG_CLS_U32_PERF 668 /* Statistics may be incremented by readers during update 669 * so we must keep them in tact. When the node is later destroyed 670 * a special destroy call must be made to not free the pf memory. 671 */ 672 new->pf = n->pf; 673#endif 674 675#ifdef CONFIG_CLS_U32_MARK 676 new->val = n->val; 677 new->mask = n->mask; 678 /* Similarly success statistics must be moved as pointers */ 679 new->pcpu_success = n->pcpu_success; 680#endif 681 new->tp = tp; 682 memcpy(&new->sel, s, sizeof(*s) + s->nkeys*sizeof(struct tc_u32_key)); 683 684 tcf_exts_init(&new->exts, TCA_U32_ACT, TCA_U32_POLICE); 685 686 return new; 687} 688 689static int u32_change(struct net *net, struct sk_buff *in_skb, 690 struct tcf_proto *tp, unsigned long base, u32 handle, 691 struct nlattr **tca, 692 unsigned long *arg, bool ovr) 693{ 694 struct tc_u_common *tp_c = tp->data; 695 struct tc_u_hnode *ht; 696 struct tc_u_knode *n; 697 struct tc_u32_sel *s; 698 struct nlattr *opt = tca[TCA_OPTIONS]; 699 struct nlattr *tb[TCA_U32_MAX + 1]; 700 u32 htid; 701 int err; 702#ifdef CONFIG_CLS_U32_PERF 703 size_t size; 704#endif 705 706 if (opt == NULL) 707 return handle ? -EINVAL : 0; 708 709 err = nla_parse_nested(tb, TCA_U32_MAX, opt, u32_policy); 710 if (err < 0) 711 return err; 712 713 n = (struct tc_u_knode *)*arg; 714 if (n) { 715 struct tc_u_knode *new; 716 717 if (TC_U32_KEY(n->handle) == 0) 718 return -EINVAL; 719 720 new = u32_init_knode(tp, n); 721 if (!new) 722 return -ENOMEM; 723 724 err = u32_set_parms(net, tp, base, 725 rtnl_dereference(n->ht_up), new, tb, 726 tca[TCA_RATE], ovr); 727 728 if (err) { 729 u32_destroy_key(tp, new, false); 730 return err; 731 } 732 733 u32_replace_knode(tp, tp_c, new); 734 call_rcu(&n->rcu, u32_delete_key_rcu); 735 return 0; 736 } 737 738 if (tb[TCA_U32_DIVISOR]) { 739 unsigned int divisor = nla_get_u32(tb[TCA_U32_DIVISOR]); 740 741 if (--divisor > 0x100) 742 return -EINVAL; 743 if (TC_U32_KEY(handle)) 744 return -EINVAL; 745 if (handle == 0) { 746 handle = gen_new_htid(tp->data); 747 if (handle == 0) 748 return -ENOMEM; 749 } 750 ht = kzalloc(sizeof(*ht) + divisor*sizeof(void *), GFP_KERNEL); 751 if (ht == NULL) 752 return -ENOBUFS; 753 ht->tp_c = tp_c; 754 ht->refcnt = 1; 755 ht->divisor = divisor; 756 ht->handle = handle; 757 ht->prio = tp->prio; 758 RCU_INIT_POINTER(ht->next, tp_c->hlist); 759 rcu_assign_pointer(tp_c->hlist, ht); 760 *arg = (unsigned long)ht; 761 return 0; 762 } 763 764 if (tb[TCA_U32_HASH]) { 765 htid = nla_get_u32(tb[TCA_U32_HASH]); 766 if (TC_U32_HTID(htid) == TC_U32_ROOT) { 767 ht = rtnl_dereference(tp->root); 768 htid = ht->handle; 769 } else { 770 ht = u32_lookup_ht(tp->data, TC_U32_HTID(htid)); 771 if (ht == NULL) 772 return -EINVAL; 773 } 774 } else { 775 ht = rtnl_dereference(tp->root); 776 htid = ht->handle; 777 } 778 779 if (ht->divisor < TC_U32_HASH(htid)) 780 return -EINVAL; 781 782 if (handle) { 783 if (TC_U32_HTID(handle) && TC_U32_HTID(handle^htid)) 784 return -EINVAL; 785 handle = htid | TC_U32_NODE(handle); 786 } else 787 handle = gen_new_kid(ht, htid); 788 789 if (tb[TCA_U32_SEL] == NULL) 790 return -EINVAL; 791 792 s = nla_data(tb[TCA_U32_SEL]); 793 794 n = kzalloc(sizeof(*n) + s->nkeys*sizeof(struct tc_u32_key), GFP_KERNEL); 795 if (n == NULL) 796 return -ENOBUFS; 797 798#ifdef CONFIG_CLS_U32_PERF 799 size = sizeof(struct tc_u32_pcnt) + s->nkeys * sizeof(u64); 800 n->pf = __alloc_percpu(size, __alignof__(struct tc_u32_pcnt)); 801 if (!n->pf) { 802 kfree(n); 803 return -ENOBUFS; 804 } 805#endif 806 807 memcpy(&n->sel, s, sizeof(*s) + s->nkeys*sizeof(struct tc_u32_key)); 808 RCU_INIT_POINTER(n->ht_up, ht); 809 n->handle = handle; 810 n->fshift = s->hmask ? ffs(ntohl(s->hmask)) - 1 : 0; 811 tcf_exts_init(&n->exts, TCA_U32_ACT, TCA_U32_POLICE); 812 n->tp = tp; 813 814#ifdef CONFIG_CLS_U32_MARK 815 n->pcpu_success = alloc_percpu(u32); 816 if (!n->pcpu_success) { 817 err = -ENOMEM; 818 goto errout; 819 } 820 821 if (tb[TCA_U32_MARK]) { 822 struct tc_u32_mark *mark; 823 824 mark = nla_data(tb[TCA_U32_MARK]); 825 n->val = mark->val; 826 n->mask = mark->mask; 827 } 828#endif 829 830 err = u32_set_parms(net, tp, base, ht, n, tb, tca[TCA_RATE], ovr); 831 if (err == 0) { 832 struct tc_u_knode __rcu **ins; 833 struct tc_u_knode *pins; 834 835 ins = &ht->ht[TC_U32_HASH(handle)]; 836 for (pins = rtnl_dereference(*ins); pins; 837 ins = &pins->next, pins = rtnl_dereference(*ins)) 838 if (TC_U32_NODE(handle) < TC_U32_NODE(pins->handle)) 839 break; 840 841 RCU_INIT_POINTER(n->next, pins); 842 rcu_assign_pointer(*ins, n); 843 844 *arg = (unsigned long)n; 845 return 0; 846 } 847 848#ifdef CONFIG_CLS_U32_MARK 849 free_percpu(n->pcpu_success); 850errout: 851#endif 852 853#ifdef CONFIG_CLS_U32_PERF 854 free_percpu(n->pf); 855#endif 856 kfree(n); 857 return err; 858} 859 860static void u32_walk(struct tcf_proto *tp, struct tcf_walker *arg) 861{ 862 struct tc_u_common *tp_c = tp->data; 863 struct tc_u_hnode *ht; 864 struct tc_u_knode *n; 865 unsigned int h; 866 867 if (arg->stop) 868 return; 869 870 for (ht = rtnl_dereference(tp_c->hlist); 871 ht; 872 ht = rtnl_dereference(ht->next)) { 873 if (ht->prio != tp->prio) 874 continue; 875 if (arg->count >= arg->skip) { 876 if (arg->fn(tp, (unsigned long)ht, arg) < 0) { 877 arg->stop = 1; 878 return; 879 } 880 } 881 arg->count++; 882 for (h = 0; h <= ht->divisor; h++) { 883 for (n = rtnl_dereference(ht->ht[h]); 884 n; 885 n = rtnl_dereference(n->next)) { 886 if (arg->count < arg->skip) { 887 arg->count++; 888 continue; 889 } 890 if (arg->fn(tp, (unsigned long)n, arg) < 0) { 891 arg->stop = 1; 892 return; 893 } 894 arg->count++; 895 } 896 } 897 } 898} 899 900static int u32_dump(struct net *net, struct tcf_proto *tp, unsigned long fh, 901 struct sk_buff *skb, struct tcmsg *t) 902{ 903 struct tc_u_knode *n = (struct tc_u_knode *)fh; 904 struct tc_u_hnode *ht_up, *ht_down; 905 struct nlattr *nest; 906 907 if (n == NULL) 908 return skb->len; 909 910 t->tcm_handle = n->handle; 911 912 nest = nla_nest_start(skb, TCA_OPTIONS); 913 if (nest == NULL) 914 goto nla_put_failure; 915 916 if (TC_U32_KEY(n->handle) == 0) { 917 struct tc_u_hnode *ht = (struct tc_u_hnode *)fh; 918 u32 divisor = ht->divisor + 1; 919 920 if (nla_put_u32(skb, TCA_U32_DIVISOR, divisor)) 921 goto nla_put_failure; 922 } else { 923#ifdef CONFIG_CLS_U32_PERF 924 struct tc_u32_pcnt *gpf; 925 int cpu; 926#endif 927 928 if (nla_put(skb, TCA_U32_SEL, 929 sizeof(n->sel) + n->sel.nkeys*sizeof(struct tc_u32_key), 930 &n->sel)) 931 goto nla_put_failure; 932 933 ht_up = rtnl_dereference(n->ht_up); 934 if (ht_up) { 935 u32 htid = n->handle & 0xFFFFF000; 936 if (nla_put_u32(skb, TCA_U32_HASH, htid)) 937 goto nla_put_failure; 938 } 939 if (n->res.classid && 940 nla_put_u32(skb, TCA_U32_CLASSID, n->res.classid)) 941 goto nla_put_failure; 942 943 ht_down = rtnl_dereference(n->ht_down); 944 if (ht_down && 945 nla_put_u32(skb, TCA_U32_LINK, ht_down->handle)) 946 goto nla_put_failure; 947 948#ifdef CONFIG_CLS_U32_MARK 949 if ((n->val || n->mask)) { 950 struct tc_u32_mark mark = {.val = n->val, 951 .mask = n->mask, 952 .success = 0}; 953 int cpum; 954 955 for_each_possible_cpu(cpum) { 956 __u32 cnt = *per_cpu_ptr(n->pcpu_success, cpum); 957 958 mark.success += cnt; 959 } 960 961 if (nla_put(skb, TCA_U32_MARK, sizeof(mark), &mark)) 962 goto nla_put_failure; 963 } 964#endif 965 966 if (tcf_exts_dump(skb, &n->exts) < 0) 967 goto nla_put_failure; 968 969#ifdef CONFIG_NET_CLS_IND 970 if (n->ifindex) { 971 struct net_device *dev; 972 dev = __dev_get_by_index(net, n->ifindex); 973 if (dev && nla_put_string(skb, TCA_U32_INDEV, dev->name)) 974 goto nla_put_failure; 975 } 976#endif 977#ifdef CONFIG_CLS_U32_PERF 978 gpf = kzalloc(sizeof(struct tc_u32_pcnt) + 979 n->sel.nkeys * sizeof(u64), 980 GFP_KERNEL); 981 if (!gpf) 982 goto nla_put_failure; 983 984 for_each_possible_cpu(cpu) { 985 int i; 986 struct tc_u32_pcnt *pf = per_cpu_ptr(n->pf, cpu); 987 988 gpf->rcnt += pf->rcnt; 989 gpf->rhit += pf->rhit; 990 for (i = 0; i < n->sel.nkeys; i++) 991 gpf->kcnts[i] += pf->kcnts[i]; 992 } 993 994 if (nla_put(skb, TCA_U32_PCNT, 995 sizeof(struct tc_u32_pcnt) + n->sel.nkeys*sizeof(u64), 996 gpf)) { 997 kfree(gpf); 998 goto nla_put_failure; 999 } 1000 kfree(gpf); 1001#endif 1002 } 1003 1004 nla_nest_end(skb, nest); 1005 1006 if (TC_U32_KEY(n->handle)) 1007 if (tcf_exts_dump_stats(skb, &n->exts) < 0) 1008 goto nla_put_failure; 1009 return skb->len; 1010 1011nla_put_failure: 1012 nla_nest_cancel(skb, nest); 1013 return -1; 1014} 1015 1016static struct tcf_proto_ops cls_u32_ops __read_mostly = { 1017 .kind = "u32", 1018 .classify = u32_classify, 1019 .init = u32_init, 1020 .destroy = u32_destroy, 1021 .get = u32_get, 1022 .put = u32_put, 1023 .change = u32_change, 1024 .delete = u32_delete, 1025 .walk = u32_walk, 1026 .dump = u32_dump, 1027 .owner = THIS_MODULE, 1028}; 1029 1030static int __init init_u32(void) 1031{ 1032 pr_info("u32 classifier\n"); 1033#ifdef CONFIG_CLS_U32_PERF 1034 pr_info(" Performance counters on\n"); 1035#endif 1036#ifdef CONFIG_NET_CLS_IND 1037 pr_info(" input device check on\n"); 1038#endif 1039#ifdef CONFIG_NET_CLS_ACT 1040 pr_info(" Actions configured\n"); 1041#endif 1042 return register_tcf_proto_ops(&cls_u32_ops); 1043} 1044 1045static void __exit exit_u32(void) 1046{ 1047 unregister_tcf_proto_ops(&cls_u32_ops); 1048} 1049 1050module_init(init_u32) 1051module_exit(exit_u32) 1052MODULE_LICENSE("GPL"); 1053