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 #include <linux/netdevice.h> 47 48 struct tc_u_knode { 49 struct tc_u_knode __rcu *next; 50 u32 handle; 51 struct tc_u_hnode __rcu *ht_up; 52 struct tcf_exts exts; 53 #ifdef CONFIG_NET_CLS_IND 54 int ifindex; 55 #endif 56 u8 fshift; 57 struct tcf_result res; 58 struct tc_u_hnode __rcu *ht_down; 59 #ifdef CONFIG_CLS_U32_PERF 60 struct tc_u32_pcnt __percpu *pf; 61 #endif 62 u32 flags; 63 #ifdef CONFIG_CLS_U32_MARK 64 u32 val; 65 u32 mask; 66 u32 __percpu *pcpu_success; 67 #endif 68 struct tcf_proto *tp; 69 struct rcu_head rcu; 70 /* The 'sel' field MUST be the last field in structure to allow for 71 * tc_u32_keys allocated at end of structure. 72 */ 73 struct tc_u32_sel sel; 74 }; 75 76 struct tc_u_hnode { 77 struct tc_u_hnode __rcu *next; 78 u32 handle; 79 u32 prio; 80 struct tc_u_common *tp_c; 81 int refcnt; 82 unsigned int divisor; 83 struct rcu_head rcu; 84 /* The 'ht' field MUST be the last field in structure to allow for 85 * more entries allocated at end of structure. 86 */ 87 struct tc_u_knode __rcu *ht[1]; 88 }; 89 90 struct tc_u_common { 91 struct tc_u_hnode __rcu *hlist; 92 struct Qdisc *q; 93 int refcnt; 94 u32 hgenerator; 95 struct rcu_head rcu; 96 }; 97 98 static inline unsigned int u32_hash_fold(__be32 key, 99 const struct tc_u32_sel *sel, 100 u8 fshift) 101 { 102 unsigned int h = ntohl(key & sel->hmask) >> fshift; 103 104 return h; 105 } 106 107 static int u32_classify(struct sk_buff *skb, const struct tcf_proto *tp, 108 struct tcf_result *res) 109 { 110 struct { 111 struct tc_u_knode *knode; 112 unsigned int off; 113 } stack[TC_U32_MAXDEPTH]; 114 115 struct tc_u_hnode *ht = rcu_dereference_bh(tp->root); 116 unsigned int off = skb_network_offset(skb); 117 struct tc_u_knode *n; 118 int sdepth = 0; 119 int off2 = 0; 120 int sel = 0; 121 #ifdef CONFIG_CLS_U32_PERF 122 int j; 123 #endif 124 int i, r; 125 126 next_ht: 127 n = rcu_dereference_bh(ht->ht[sel]); 128 129 next_knode: 130 if (n) { 131 struct tc_u32_key *key = n->sel.keys; 132 133 #ifdef CONFIG_CLS_U32_PERF 134 __this_cpu_inc(n->pf->rcnt); 135 j = 0; 136 #endif 137 138 if (tc_skip_sw(n->flags)) { 139 n = rcu_dereference_bh(n->next); 140 goto next_knode; 141 } 142 143 #ifdef CONFIG_CLS_U32_MARK 144 if ((skb->mark & n->mask) != n->val) { 145 n = rcu_dereference_bh(n->next); 146 goto next_knode; 147 } else { 148 __this_cpu_inc(*n->pcpu_success); 149 } 150 #endif 151 152 for (i = n->sel.nkeys; i > 0; i--, key++) { 153 int toff = off + key->off + (off2 & key->offmask); 154 __be32 *data, hdata; 155 156 if (skb_headroom(skb) + toff > INT_MAX) 157 goto out; 158 159 data = skb_header_pointer(skb, toff, 4, &hdata); 160 if (!data) 161 goto out; 162 if ((*data ^ key->val) & key->mask) { 163 n = rcu_dereference_bh(n->next); 164 goto next_knode; 165 } 166 #ifdef CONFIG_CLS_U32_PERF 167 __this_cpu_inc(n->pf->kcnts[j]); 168 j++; 169 #endif 170 } 171 172 ht = rcu_dereference_bh(n->ht_down); 173 if (!ht) { 174 check_terminal: 175 if (n->sel.flags & TC_U32_TERMINAL) { 176 177 *res = n->res; 178 #ifdef CONFIG_NET_CLS_IND 179 if (!tcf_match_indev(skb, n->ifindex)) { 180 n = rcu_dereference_bh(n->next); 181 goto next_knode; 182 } 183 #endif 184 #ifdef CONFIG_CLS_U32_PERF 185 __this_cpu_inc(n->pf->rhit); 186 #endif 187 r = tcf_exts_exec(skb, &n->exts, res); 188 if (r < 0) { 189 n = rcu_dereference_bh(n->next); 190 goto next_knode; 191 } 192 193 return r; 194 } 195 n = rcu_dereference_bh(n->next); 196 goto next_knode; 197 } 198 199 /* PUSH */ 200 if (sdepth >= TC_U32_MAXDEPTH) 201 goto deadloop; 202 stack[sdepth].knode = n; 203 stack[sdepth].off = off; 204 sdepth++; 205 206 ht = rcu_dereference_bh(n->ht_down); 207 sel = 0; 208 if (ht->divisor) { 209 __be32 *data, hdata; 210 211 data = skb_header_pointer(skb, off + n->sel.hoff, 4, 212 &hdata); 213 if (!data) 214 goto out; 215 sel = ht->divisor & u32_hash_fold(*data, &n->sel, 216 n->fshift); 217 } 218 if (!(n->sel.flags & (TC_U32_VAROFFSET | TC_U32_OFFSET | TC_U32_EAT))) 219 goto next_ht; 220 221 if (n->sel.flags & (TC_U32_OFFSET | TC_U32_VAROFFSET)) { 222 off2 = n->sel.off + 3; 223 if (n->sel.flags & TC_U32_VAROFFSET) { 224 __be16 *data, hdata; 225 226 data = skb_header_pointer(skb, 227 off + n->sel.offoff, 228 2, &hdata); 229 if (!data) 230 goto out; 231 off2 += ntohs(n->sel.offmask & *data) >> 232 n->sel.offshift; 233 } 234 off2 &= ~3; 235 } 236 if (n->sel.flags & TC_U32_EAT) { 237 off += off2; 238 off2 = 0; 239 } 240 241 if (off < skb->len) 242 goto next_ht; 243 } 244 245 /* POP */ 246 if (sdepth--) { 247 n = stack[sdepth].knode; 248 ht = rcu_dereference_bh(n->ht_up); 249 off = stack[sdepth].off; 250 goto check_terminal; 251 } 252 out: 253 return -1; 254 255 deadloop: 256 net_warn_ratelimited("cls_u32: dead loop\n"); 257 return -1; 258 } 259 260 static struct tc_u_hnode *u32_lookup_ht(struct tc_u_common *tp_c, u32 handle) 261 { 262 struct tc_u_hnode *ht; 263 264 for (ht = rtnl_dereference(tp_c->hlist); 265 ht; 266 ht = rtnl_dereference(ht->next)) 267 if (ht->handle == handle) 268 break; 269 270 return ht; 271 } 272 273 static struct tc_u_knode *u32_lookup_key(struct tc_u_hnode *ht, u32 handle) 274 { 275 unsigned int sel; 276 struct tc_u_knode *n = NULL; 277 278 sel = TC_U32_HASH(handle); 279 if (sel > ht->divisor) 280 goto out; 281 282 for (n = rtnl_dereference(ht->ht[sel]); 283 n; 284 n = rtnl_dereference(n->next)) 285 if (n->handle == handle) 286 break; 287 out: 288 return n; 289 } 290 291 292 static unsigned long u32_get(struct tcf_proto *tp, u32 handle) 293 { 294 struct tc_u_hnode *ht; 295 struct tc_u_common *tp_c = tp->data; 296 297 if (TC_U32_HTID(handle) == TC_U32_ROOT) 298 ht = rtnl_dereference(tp->root); 299 else 300 ht = u32_lookup_ht(tp_c, TC_U32_HTID(handle)); 301 302 if (!ht) 303 return 0; 304 305 if (TC_U32_KEY(handle) == 0) 306 return (unsigned long)ht; 307 308 return (unsigned long)u32_lookup_key(ht, handle); 309 } 310 311 static u32 gen_new_htid(struct tc_u_common *tp_c) 312 { 313 int i = 0x800; 314 315 /* hgenerator only used inside rtnl lock it is safe to increment 316 * without read _copy_ update semantics 317 */ 318 do { 319 if (++tp_c->hgenerator == 0x7FF) 320 tp_c->hgenerator = 1; 321 } while (--i > 0 && u32_lookup_ht(tp_c, (tp_c->hgenerator|0x800)<<20)); 322 323 return i > 0 ? (tp_c->hgenerator|0x800)<<20 : 0; 324 } 325 326 static int u32_init(struct tcf_proto *tp) 327 { 328 struct tc_u_hnode *root_ht; 329 struct tc_u_common *tp_c; 330 331 tp_c = tp->q->u32_node; 332 333 root_ht = kzalloc(sizeof(*root_ht), GFP_KERNEL); 334 if (root_ht == NULL) 335 return -ENOBUFS; 336 337 root_ht->refcnt++; 338 root_ht->handle = tp_c ? gen_new_htid(tp_c) : 0x80000000; 339 root_ht->prio = tp->prio; 340 341 if (tp_c == NULL) { 342 tp_c = kzalloc(sizeof(*tp_c), GFP_KERNEL); 343 if (tp_c == NULL) { 344 kfree(root_ht); 345 return -ENOBUFS; 346 } 347 tp_c->q = tp->q; 348 tp->q->u32_node = tp_c; 349 } 350 351 tp_c->refcnt++; 352 RCU_INIT_POINTER(root_ht->next, tp_c->hlist); 353 rcu_assign_pointer(tp_c->hlist, root_ht); 354 root_ht->tp_c = tp_c; 355 356 rcu_assign_pointer(tp->root, root_ht); 357 tp->data = tp_c; 358 return 0; 359 } 360 361 static int u32_destroy_key(struct tcf_proto *tp, struct tc_u_knode *n, 362 bool free_pf) 363 { 364 tcf_exts_destroy(&n->exts); 365 if (n->ht_down) 366 n->ht_down->refcnt--; 367 #ifdef CONFIG_CLS_U32_PERF 368 if (free_pf) 369 free_percpu(n->pf); 370 #endif 371 #ifdef CONFIG_CLS_U32_MARK 372 if (free_pf) 373 free_percpu(n->pcpu_success); 374 #endif 375 kfree(n); 376 return 0; 377 } 378 379 /* u32_delete_key_rcu should be called when free'ing a copied 380 * version of a tc_u_knode obtained from u32_init_knode(). When 381 * copies are obtained from u32_init_knode() the statistics are 382 * shared between the old and new copies to allow readers to 383 * continue to update the statistics during the copy. To support 384 * this the u32_delete_key_rcu variant does not free the percpu 385 * statistics. 386 */ 387 static void u32_delete_key_rcu(struct rcu_head *rcu) 388 { 389 struct tc_u_knode *key = container_of(rcu, struct tc_u_knode, rcu); 390 391 u32_destroy_key(key->tp, key, false); 392 } 393 394 /* u32_delete_key_freepf_rcu is the rcu callback variant 395 * that free's the entire structure including the statistics 396 * percpu variables. Only use this if the key is not a copy 397 * returned by u32_init_knode(). See u32_delete_key_rcu() 398 * for the variant that should be used with keys return from 399 * u32_init_knode() 400 */ 401 static void u32_delete_key_freepf_rcu(struct rcu_head *rcu) 402 { 403 struct tc_u_knode *key = container_of(rcu, struct tc_u_knode, rcu); 404 405 u32_destroy_key(key->tp, key, true); 406 } 407 408 static int u32_delete_key(struct tcf_proto *tp, struct tc_u_knode *key) 409 { 410 struct tc_u_knode __rcu **kp; 411 struct tc_u_knode *pkp; 412 struct tc_u_hnode *ht = rtnl_dereference(key->ht_up); 413 414 if (ht) { 415 kp = &ht->ht[TC_U32_HASH(key->handle)]; 416 for (pkp = rtnl_dereference(*kp); pkp; 417 kp = &pkp->next, pkp = rtnl_dereference(*kp)) { 418 if (pkp == key) { 419 RCU_INIT_POINTER(*kp, key->next); 420 421 tcf_unbind_filter(tp, &key->res); 422 call_rcu(&key->rcu, u32_delete_key_freepf_rcu); 423 return 0; 424 } 425 } 426 } 427 WARN_ON(1); 428 return 0; 429 } 430 431 static void u32_remove_hw_knode(struct tcf_proto *tp, u32 handle) 432 { 433 struct net_device *dev = tp->q->dev_queue->dev; 434 struct tc_cls_u32_offload u32_offload = {0}; 435 struct tc_to_netdev offload; 436 437 offload.type = TC_SETUP_CLSU32; 438 offload.cls_u32 = &u32_offload; 439 440 if (tc_should_offload(dev, tp, 0)) { 441 offload.cls_u32->command = TC_CLSU32_DELETE_KNODE; 442 offload.cls_u32->knode.handle = handle; 443 dev->netdev_ops->ndo_setup_tc(dev, tp->q->handle, 444 tp->protocol, &offload); 445 } 446 } 447 448 static int u32_replace_hw_hnode(struct tcf_proto *tp, struct tc_u_hnode *h, 449 u32 flags) 450 { 451 struct net_device *dev = tp->q->dev_queue->dev; 452 struct tc_cls_u32_offload u32_offload = {0}; 453 struct tc_to_netdev offload; 454 int err; 455 456 if (!tc_should_offload(dev, tp, flags)) 457 return tc_skip_sw(flags) ? -EINVAL : 0; 458 459 offload.type = TC_SETUP_CLSU32; 460 offload.cls_u32 = &u32_offload; 461 462 offload.cls_u32->command = TC_CLSU32_NEW_HNODE; 463 offload.cls_u32->hnode.divisor = h->divisor; 464 offload.cls_u32->hnode.handle = h->handle; 465 offload.cls_u32->hnode.prio = h->prio; 466 467 err = dev->netdev_ops->ndo_setup_tc(dev, tp->q->handle, 468 tp->protocol, &offload); 469 if (tc_skip_sw(flags)) 470 return err; 471 472 return 0; 473 } 474 475 static void u32_clear_hw_hnode(struct tcf_proto *tp, struct tc_u_hnode *h) 476 { 477 struct net_device *dev = tp->q->dev_queue->dev; 478 struct tc_cls_u32_offload u32_offload = {0}; 479 struct tc_to_netdev offload; 480 481 offload.type = TC_SETUP_CLSU32; 482 offload.cls_u32 = &u32_offload; 483 484 if (tc_should_offload(dev, tp, 0)) { 485 offload.cls_u32->command = TC_CLSU32_DELETE_HNODE; 486 offload.cls_u32->hnode.divisor = h->divisor; 487 offload.cls_u32->hnode.handle = h->handle; 488 offload.cls_u32->hnode.prio = h->prio; 489 490 dev->netdev_ops->ndo_setup_tc(dev, tp->q->handle, 491 tp->protocol, &offload); 492 } 493 } 494 495 static int u32_replace_hw_knode(struct tcf_proto *tp, struct tc_u_knode *n, 496 u32 flags) 497 { 498 struct net_device *dev = tp->q->dev_queue->dev; 499 struct tc_cls_u32_offload u32_offload = {0}; 500 struct tc_to_netdev offload; 501 int err; 502 503 offload.type = TC_SETUP_CLSU32; 504 offload.cls_u32 = &u32_offload; 505 506 if (!tc_should_offload(dev, tp, flags)) 507 return tc_skip_sw(flags) ? -EINVAL : 0; 508 509 offload.cls_u32->command = TC_CLSU32_REPLACE_KNODE; 510 offload.cls_u32->knode.handle = n->handle; 511 offload.cls_u32->knode.fshift = n->fshift; 512 #ifdef CONFIG_CLS_U32_MARK 513 offload.cls_u32->knode.val = n->val; 514 offload.cls_u32->knode.mask = n->mask; 515 #else 516 offload.cls_u32->knode.val = 0; 517 offload.cls_u32->knode.mask = 0; 518 #endif 519 offload.cls_u32->knode.sel = &n->sel; 520 offload.cls_u32->knode.exts = &n->exts; 521 if (n->ht_down) 522 offload.cls_u32->knode.link_handle = n->ht_down->handle; 523 524 err = dev->netdev_ops->ndo_setup_tc(dev, tp->q->handle, 525 tp->protocol, &offload); 526 527 if (!err) 528 n->flags |= TCA_CLS_FLAGS_IN_HW; 529 530 if (tc_skip_sw(flags)) 531 return err; 532 533 return 0; 534 } 535 536 static void u32_clear_hnode(struct tcf_proto *tp, struct tc_u_hnode *ht) 537 { 538 struct tc_u_knode *n; 539 unsigned int h; 540 541 for (h = 0; h <= ht->divisor; h++) { 542 while ((n = rtnl_dereference(ht->ht[h])) != NULL) { 543 RCU_INIT_POINTER(ht->ht[h], 544 rtnl_dereference(n->next)); 545 tcf_unbind_filter(tp, &n->res); 546 u32_remove_hw_knode(tp, n->handle); 547 call_rcu(&n->rcu, u32_delete_key_freepf_rcu); 548 } 549 } 550 } 551 552 static int u32_destroy_hnode(struct tcf_proto *tp, struct tc_u_hnode *ht) 553 { 554 struct tc_u_common *tp_c = tp->data; 555 struct tc_u_hnode __rcu **hn; 556 struct tc_u_hnode *phn; 557 558 WARN_ON(ht->refcnt); 559 560 u32_clear_hnode(tp, ht); 561 562 hn = &tp_c->hlist; 563 for (phn = rtnl_dereference(*hn); 564 phn; 565 hn = &phn->next, phn = rtnl_dereference(*hn)) { 566 if (phn == ht) { 567 u32_clear_hw_hnode(tp, ht); 568 RCU_INIT_POINTER(*hn, ht->next); 569 kfree_rcu(ht, rcu); 570 return 0; 571 } 572 } 573 574 return -ENOENT; 575 } 576 577 static bool ht_empty(struct tc_u_hnode *ht) 578 { 579 unsigned int h; 580 581 for (h = 0; h <= ht->divisor; h++) 582 if (rcu_access_pointer(ht->ht[h])) 583 return false; 584 585 return true; 586 } 587 588 static bool u32_destroy(struct tcf_proto *tp, bool force) 589 { 590 struct tc_u_common *tp_c = tp->data; 591 struct tc_u_hnode *root_ht = rtnl_dereference(tp->root); 592 593 WARN_ON(root_ht == NULL); 594 595 if (!force) { 596 if (root_ht) { 597 if (root_ht->refcnt > 1) 598 return false; 599 if (root_ht->refcnt == 1) { 600 if (!ht_empty(root_ht)) 601 return false; 602 } 603 } 604 605 if (tp_c->refcnt > 1) 606 return false; 607 608 if (tp_c->refcnt == 1) { 609 struct tc_u_hnode *ht; 610 611 for (ht = rtnl_dereference(tp_c->hlist); 612 ht; 613 ht = rtnl_dereference(ht->next)) 614 if (!ht_empty(ht)) 615 return false; 616 } 617 } 618 619 if (root_ht && --root_ht->refcnt == 0) 620 u32_destroy_hnode(tp, root_ht); 621 622 if (--tp_c->refcnt == 0) { 623 struct tc_u_hnode *ht; 624 625 tp->q->u32_node = NULL; 626 627 for (ht = rtnl_dereference(tp_c->hlist); 628 ht; 629 ht = rtnl_dereference(ht->next)) { 630 ht->refcnt--; 631 u32_clear_hnode(tp, ht); 632 } 633 634 while ((ht = rtnl_dereference(tp_c->hlist)) != NULL) { 635 RCU_INIT_POINTER(tp_c->hlist, ht->next); 636 kfree_rcu(ht, rcu); 637 } 638 639 kfree(tp_c); 640 } 641 642 tp->data = NULL; 643 return true; 644 } 645 646 static int u32_delete(struct tcf_proto *tp, unsigned long arg) 647 { 648 struct tc_u_hnode *ht = (struct tc_u_hnode *)arg; 649 struct tc_u_hnode *root_ht = rtnl_dereference(tp->root); 650 651 if (ht == NULL) 652 return 0; 653 654 if (TC_U32_KEY(ht->handle)) { 655 u32_remove_hw_knode(tp, ht->handle); 656 return u32_delete_key(tp, (struct tc_u_knode *)ht); 657 } 658 659 if (root_ht == ht) 660 return -EINVAL; 661 662 if (ht->refcnt == 1) { 663 ht->refcnt--; 664 u32_destroy_hnode(tp, ht); 665 } else { 666 return -EBUSY; 667 } 668 669 return 0; 670 } 671 672 #define NR_U32_NODE (1<<12) 673 static u32 gen_new_kid(struct tc_u_hnode *ht, u32 handle) 674 { 675 struct tc_u_knode *n; 676 unsigned long i; 677 unsigned long *bitmap = kzalloc(BITS_TO_LONGS(NR_U32_NODE) * sizeof(unsigned long), 678 GFP_KERNEL); 679 if (!bitmap) 680 return handle | 0xFFF; 681 682 for (n = rtnl_dereference(ht->ht[TC_U32_HASH(handle)]); 683 n; 684 n = rtnl_dereference(n->next)) 685 set_bit(TC_U32_NODE(n->handle), bitmap); 686 687 i = find_next_zero_bit(bitmap, NR_U32_NODE, 0x800); 688 if (i >= NR_U32_NODE) 689 i = find_next_zero_bit(bitmap, NR_U32_NODE, 1); 690 691 kfree(bitmap); 692 return handle | (i >= NR_U32_NODE ? 0xFFF : i); 693 } 694 695 static const struct nla_policy u32_policy[TCA_U32_MAX + 1] = { 696 [TCA_U32_CLASSID] = { .type = NLA_U32 }, 697 [TCA_U32_HASH] = { .type = NLA_U32 }, 698 [TCA_U32_LINK] = { .type = NLA_U32 }, 699 [TCA_U32_DIVISOR] = { .type = NLA_U32 }, 700 [TCA_U32_SEL] = { .len = sizeof(struct tc_u32_sel) }, 701 [TCA_U32_INDEV] = { .type = NLA_STRING, .len = IFNAMSIZ }, 702 [TCA_U32_MARK] = { .len = sizeof(struct tc_u32_mark) }, 703 [TCA_U32_FLAGS] = { .type = NLA_U32 }, 704 }; 705 706 static int u32_set_parms(struct net *net, struct tcf_proto *tp, 707 unsigned long base, struct tc_u_hnode *ht, 708 struct tc_u_knode *n, struct nlattr **tb, 709 struct nlattr *est, bool ovr) 710 { 711 struct tcf_exts e; 712 int err; 713 714 err = tcf_exts_init(&e, TCA_U32_ACT, TCA_U32_POLICE); 715 if (err < 0) 716 return err; 717 err = tcf_exts_validate(net, tp, tb, est, &e, ovr); 718 if (err < 0) 719 goto errout; 720 721 err = -EINVAL; 722 if (tb[TCA_U32_LINK]) { 723 u32 handle = nla_get_u32(tb[TCA_U32_LINK]); 724 struct tc_u_hnode *ht_down = NULL, *ht_old; 725 726 if (TC_U32_KEY(handle)) 727 goto errout; 728 729 if (handle) { 730 ht_down = u32_lookup_ht(ht->tp_c, handle); 731 732 if (ht_down == NULL) 733 goto errout; 734 ht_down->refcnt++; 735 } 736 737 ht_old = rtnl_dereference(n->ht_down); 738 rcu_assign_pointer(n->ht_down, ht_down); 739 740 if (ht_old) 741 ht_old->refcnt--; 742 } 743 if (tb[TCA_U32_CLASSID]) { 744 n->res.classid = nla_get_u32(tb[TCA_U32_CLASSID]); 745 tcf_bind_filter(tp, &n->res, base); 746 } 747 748 #ifdef CONFIG_NET_CLS_IND 749 if (tb[TCA_U32_INDEV]) { 750 int ret; 751 ret = tcf_change_indev(net, tb[TCA_U32_INDEV]); 752 if (ret < 0) 753 goto errout; 754 n->ifindex = ret; 755 } 756 #endif 757 tcf_exts_change(tp, &n->exts, &e); 758 759 return 0; 760 errout: 761 tcf_exts_destroy(&e); 762 return err; 763 } 764 765 static void u32_replace_knode(struct tcf_proto *tp, struct tc_u_common *tp_c, 766 struct tc_u_knode *n) 767 { 768 struct tc_u_knode __rcu **ins; 769 struct tc_u_knode *pins; 770 struct tc_u_hnode *ht; 771 772 if (TC_U32_HTID(n->handle) == TC_U32_ROOT) 773 ht = rtnl_dereference(tp->root); 774 else 775 ht = u32_lookup_ht(tp_c, TC_U32_HTID(n->handle)); 776 777 ins = &ht->ht[TC_U32_HASH(n->handle)]; 778 779 /* The node must always exist for it to be replaced if this is not the 780 * case then something went very wrong elsewhere. 781 */ 782 for (pins = rtnl_dereference(*ins); ; 783 ins = &pins->next, pins = rtnl_dereference(*ins)) 784 if (pins->handle == n->handle) 785 break; 786 787 RCU_INIT_POINTER(n->next, pins->next); 788 rcu_assign_pointer(*ins, n); 789 } 790 791 static struct tc_u_knode *u32_init_knode(struct tcf_proto *tp, 792 struct tc_u_knode *n) 793 { 794 struct tc_u_knode *new; 795 struct tc_u32_sel *s = &n->sel; 796 797 new = kzalloc(sizeof(*n) + s->nkeys*sizeof(struct tc_u32_key), 798 GFP_KERNEL); 799 800 if (!new) 801 return NULL; 802 803 RCU_INIT_POINTER(new->next, n->next); 804 new->handle = n->handle; 805 RCU_INIT_POINTER(new->ht_up, n->ht_up); 806 807 #ifdef CONFIG_NET_CLS_IND 808 new->ifindex = n->ifindex; 809 #endif 810 new->fshift = n->fshift; 811 new->res = n->res; 812 new->flags = n->flags; 813 RCU_INIT_POINTER(new->ht_down, n->ht_down); 814 815 /* bump reference count as long as we hold pointer to structure */ 816 if (new->ht_down) 817 new->ht_down->refcnt++; 818 819 #ifdef CONFIG_CLS_U32_PERF 820 /* Statistics may be incremented by readers during update 821 * so we must keep them in tact. When the node is later destroyed 822 * a special destroy call must be made to not free the pf memory. 823 */ 824 new->pf = n->pf; 825 #endif 826 827 #ifdef CONFIG_CLS_U32_MARK 828 new->val = n->val; 829 new->mask = n->mask; 830 /* Similarly success statistics must be moved as pointers */ 831 new->pcpu_success = n->pcpu_success; 832 #endif 833 new->tp = tp; 834 memcpy(&new->sel, s, sizeof(*s) + s->nkeys*sizeof(struct tc_u32_key)); 835 836 if (tcf_exts_init(&new->exts, TCA_U32_ACT, TCA_U32_POLICE)) { 837 kfree(new); 838 return NULL; 839 } 840 841 return new; 842 } 843 844 static int u32_change(struct net *net, struct sk_buff *in_skb, 845 struct tcf_proto *tp, unsigned long base, u32 handle, 846 struct nlattr **tca, unsigned long *arg, bool ovr) 847 { 848 struct tc_u_common *tp_c = tp->data; 849 struct tc_u_hnode *ht; 850 struct tc_u_knode *n; 851 struct tc_u32_sel *s; 852 struct nlattr *opt = tca[TCA_OPTIONS]; 853 struct nlattr *tb[TCA_U32_MAX + 1]; 854 u32 htid, flags = 0; 855 int err; 856 #ifdef CONFIG_CLS_U32_PERF 857 size_t size; 858 #endif 859 860 if (opt == NULL) 861 return handle ? -EINVAL : 0; 862 863 err = nla_parse_nested(tb, TCA_U32_MAX, opt, u32_policy); 864 if (err < 0) 865 return err; 866 867 if (tb[TCA_U32_FLAGS]) { 868 flags = nla_get_u32(tb[TCA_U32_FLAGS]); 869 if (!tc_flags_valid(flags)) 870 return -EINVAL; 871 } 872 873 n = (struct tc_u_knode *)*arg; 874 if (n) { 875 struct tc_u_knode *new; 876 877 if (TC_U32_KEY(n->handle) == 0) 878 return -EINVAL; 879 880 if (n->flags != flags) 881 return -EINVAL; 882 883 new = u32_init_knode(tp, n); 884 if (!new) 885 return -ENOMEM; 886 887 err = u32_set_parms(net, tp, base, 888 rtnl_dereference(n->ht_up), new, tb, 889 tca[TCA_RATE], ovr); 890 891 if (err) { 892 u32_destroy_key(tp, new, false); 893 return err; 894 } 895 896 err = u32_replace_hw_knode(tp, new, flags); 897 if (err) { 898 u32_destroy_key(tp, new, false); 899 return err; 900 } 901 902 if (!tc_in_hw(new->flags)) 903 new->flags |= TCA_CLS_FLAGS_NOT_IN_HW; 904 905 u32_replace_knode(tp, tp_c, new); 906 tcf_unbind_filter(tp, &n->res); 907 call_rcu(&n->rcu, u32_delete_key_rcu); 908 return 0; 909 } 910 911 if (tb[TCA_U32_DIVISOR]) { 912 unsigned int divisor = nla_get_u32(tb[TCA_U32_DIVISOR]); 913 914 if (--divisor > 0x100) 915 return -EINVAL; 916 if (TC_U32_KEY(handle)) 917 return -EINVAL; 918 if (handle == 0) { 919 handle = gen_new_htid(tp->data); 920 if (handle == 0) 921 return -ENOMEM; 922 } 923 ht = kzalloc(sizeof(*ht) + divisor*sizeof(void *), GFP_KERNEL); 924 if (ht == NULL) 925 return -ENOBUFS; 926 ht->tp_c = tp_c; 927 ht->refcnt = 1; 928 ht->divisor = divisor; 929 ht->handle = handle; 930 ht->prio = tp->prio; 931 932 err = u32_replace_hw_hnode(tp, ht, flags); 933 if (err) { 934 kfree(ht); 935 return err; 936 } 937 938 RCU_INIT_POINTER(ht->next, tp_c->hlist); 939 rcu_assign_pointer(tp_c->hlist, ht); 940 *arg = (unsigned long)ht; 941 942 return 0; 943 } 944 945 if (tb[TCA_U32_HASH]) { 946 htid = nla_get_u32(tb[TCA_U32_HASH]); 947 if (TC_U32_HTID(htid) == TC_U32_ROOT) { 948 ht = rtnl_dereference(tp->root); 949 htid = ht->handle; 950 } else { 951 ht = u32_lookup_ht(tp->data, TC_U32_HTID(htid)); 952 if (ht == NULL) 953 return -EINVAL; 954 } 955 } else { 956 ht = rtnl_dereference(tp->root); 957 htid = ht->handle; 958 } 959 960 if (ht->divisor < TC_U32_HASH(htid)) 961 return -EINVAL; 962 963 if (handle) { 964 if (TC_U32_HTID(handle) && TC_U32_HTID(handle^htid)) 965 return -EINVAL; 966 handle = htid | TC_U32_NODE(handle); 967 } else 968 handle = gen_new_kid(ht, htid); 969 970 if (tb[TCA_U32_SEL] == NULL) 971 return -EINVAL; 972 973 s = nla_data(tb[TCA_U32_SEL]); 974 975 n = kzalloc(sizeof(*n) + s->nkeys*sizeof(struct tc_u32_key), GFP_KERNEL); 976 if (n == NULL) 977 return -ENOBUFS; 978 979 #ifdef CONFIG_CLS_U32_PERF 980 size = sizeof(struct tc_u32_pcnt) + s->nkeys * sizeof(u64); 981 n->pf = __alloc_percpu(size, __alignof__(struct tc_u32_pcnt)); 982 if (!n->pf) { 983 kfree(n); 984 return -ENOBUFS; 985 } 986 #endif 987 988 memcpy(&n->sel, s, sizeof(*s) + s->nkeys*sizeof(struct tc_u32_key)); 989 RCU_INIT_POINTER(n->ht_up, ht); 990 n->handle = handle; 991 n->fshift = s->hmask ? ffs(ntohl(s->hmask)) - 1 : 0; 992 n->flags = flags; 993 n->tp = tp; 994 995 err = tcf_exts_init(&n->exts, TCA_U32_ACT, TCA_U32_POLICE); 996 if (err < 0) 997 goto errout; 998 999 #ifdef CONFIG_CLS_U32_MARK 1000 n->pcpu_success = alloc_percpu(u32); 1001 if (!n->pcpu_success) { 1002 err = -ENOMEM; 1003 goto errout; 1004 } 1005 1006 if (tb[TCA_U32_MARK]) { 1007 struct tc_u32_mark *mark; 1008 1009 mark = nla_data(tb[TCA_U32_MARK]); 1010 n->val = mark->val; 1011 n->mask = mark->mask; 1012 } 1013 #endif 1014 1015 err = u32_set_parms(net, tp, base, ht, n, tb, tca[TCA_RATE], ovr); 1016 if (err == 0) { 1017 struct tc_u_knode __rcu **ins; 1018 struct tc_u_knode *pins; 1019 1020 err = u32_replace_hw_knode(tp, n, flags); 1021 if (err) 1022 goto errhw; 1023 1024 if (!tc_in_hw(n->flags)) 1025 n->flags |= TCA_CLS_FLAGS_NOT_IN_HW; 1026 1027 ins = &ht->ht[TC_U32_HASH(handle)]; 1028 for (pins = rtnl_dereference(*ins); pins; 1029 ins = &pins->next, pins = rtnl_dereference(*ins)) 1030 if (TC_U32_NODE(handle) < TC_U32_NODE(pins->handle)) 1031 break; 1032 1033 RCU_INIT_POINTER(n->next, pins); 1034 rcu_assign_pointer(*ins, n); 1035 *arg = (unsigned long)n; 1036 return 0; 1037 } 1038 1039 errhw: 1040 #ifdef CONFIG_CLS_U32_MARK 1041 free_percpu(n->pcpu_success); 1042 #endif 1043 1044 errout: 1045 tcf_exts_destroy(&n->exts); 1046 #ifdef CONFIG_CLS_U32_PERF 1047 free_percpu(n->pf); 1048 #endif 1049 kfree(n); 1050 return err; 1051 } 1052 1053 static void u32_walk(struct tcf_proto *tp, struct tcf_walker *arg) 1054 { 1055 struct tc_u_common *tp_c = tp->data; 1056 struct tc_u_hnode *ht; 1057 struct tc_u_knode *n; 1058 unsigned int h; 1059 1060 if (arg->stop) 1061 return; 1062 1063 for (ht = rtnl_dereference(tp_c->hlist); 1064 ht; 1065 ht = rtnl_dereference(ht->next)) { 1066 if (ht->prio != tp->prio) 1067 continue; 1068 if (arg->count >= arg->skip) { 1069 if (arg->fn(tp, (unsigned long)ht, arg) < 0) { 1070 arg->stop = 1; 1071 return; 1072 } 1073 } 1074 arg->count++; 1075 for (h = 0; h <= ht->divisor; h++) { 1076 for (n = rtnl_dereference(ht->ht[h]); 1077 n; 1078 n = rtnl_dereference(n->next)) { 1079 if (arg->count < arg->skip) { 1080 arg->count++; 1081 continue; 1082 } 1083 if (arg->fn(tp, (unsigned long)n, arg) < 0) { 1084 arg->stop = 1; 1085 return; 1086 } 1087 arg->count++; 1088 } 1089 } 1090 } 1091 } 1092 1093 static int u32_dump(struct net *net, struct tcf_proto *tp, unsigned long fh, 1094 struct sk_buff *skb, struct tcmsg *t) 1095 { 1096 struct tc_u_knode *n = (struct tc_u_knode *)fh; 1097 struct tc_u_hnode *ht_up, *ht_down; 1098 struct nlattr *nest; 1099 1100 if (n == NULL) 1101 return skb->len; 1102 1103 t->tcm_handle = n->handle; 1104 1105 nest = nla_nest_start(skb, TCA_OPTIONS); 1106 if (nest == NULL) 1107 goto nla_put_failure; 1108 1109 if (TC_U32_KEY(n->handle) == 0) { 1110 struct tc_u_hnode *ht = (struct tc_u_hnode *)fh; 1111 u32 divisor = ht->divisor + 1; 1112 1113 if (nla_put_u32(skb, TCA_U32_DIVISOR, divisor)) 1114 goto nla_put_failure; 1115 } else { 1116 #ifdef CONFIG_CLS_U32_PERF 1117 struct tc_u32_pcnt *gpf; 1118 int cpu; 1119 #endif 1120 1121 if (nla_put(skb, TCA_U32_SEL, 1122 sizeof(n->sel) + n->sel.nkeys*sizeof(struct tc_u32_key), 1123 &n->sel)) 1124 goto nla_put_failure; 1125 1126 ht_up = rtnl_dereference(n->ht_up); 1127 if (ht_up) { 1128 u32 htid = n->handle & 0xFFFFF000; 1129 if (nla_put_u32(skb, TCA_U32_HASH, htid)) 1130 goto nla_put_failure; 1131 } 1132 if (n->res.classid && 1133 nla_put_u32(skb, TCA_U32_CLASSID, n->res.classid)) 1134 goto nla_put_failure; 1135 1136 ht_down = rtnl_dereference(n->ht_down); 1137 if (ht_down && 1138 nla_put_u32(skb, TCA_U32_LINK, ht_down->handle)) 1139 goto nla_put_failure; 1140 1141 if (n->flags && nla_put_u32(skb, TCA_U32_FLAGS, n->flags)) 1142 goto nla_put_failure; 1143 1144 #ifdef CONFIG_CLS_U32_MARK 1145 if ((n->val || n->mask)) { 1146 struct tc_u32_mark mark = {.val = n->val, 1147 .mask = n->mask, 1148 .success = 0}; 1149 int cpum; 1150 1151 for_each_possible_cpu(cpum) { 1152 __u32 cnt = *per_cpu_ptr(n->pcpu_success, cpum); 1153 1154 mark.success += cnt; 1155 } 1156 1157 if (nla_put(skb, TCA_U32_MARK, sizeof(mark), &mark)) 1158 goto nla_put_failure; 1159 } 1160 #endif 1161 1162 if (tcf_exts_dump(skb, &n->exts) < 0) 1163 goto nla_put_failure; 1164 1165 #ifdef CONFIG_NET_CLS_IND 1166 if (n->ifindex) { 1167 struct net_device *dev; 1168 dev = __dev_get_by_index(net, n->ifindex); 1169 if (dev && nla_put_string(skb, TCA_U32_INDEV, dev->name)) 1170 goto nla_put_failure; 1171 } 1172 #endif 1173 #ifdef CONFIG_CLS_U32_PERF 1174 gpf = kzalloc(sizeof(struct tc_u32_pcnt) + 1175 n->sel.nkeys * sizeof(u64), 1176 GFP_KERNEL); 1177 if (!gpf) 1178 goto nla_put_failure; 1179 1180 for_each_possible_cpu(cpu) { 1181 int i; 1182 struct tc_u32_pcnt *pf = per_cpu_ptr(n->pf, cpu); 1183 1184 gpf->rcnt += pf->rcnt; 1185 gpf->rhit += pf->rhit; 1186 for (i = 0; i < n->sel.nkeys; i++) 1187 gpf->kcnts[i] += pf->kcnts[i]; 1188 } 1189 1190 if (nla_put_64bit(skb, TCA_U32_PCNT, 1191 sizeof(struct tc_u32_pcnt) + 1192 n->sel.nkeys * sizeof(u64), 1193 gpf, TCA_U32_PAD)) { 1194 kfree(gpf); 1195 goto nla_put_failure; 1196 } 1197 kfree(gpf); 1198 #endif 1199 } 1200 1201 nla_nest_end(skb, nest); 1202 1203 if (TC_U32_KEY(n->handle)) 1204 if (tcf_exts_dump_stats(skb, &n->exts) < 0) 1205 goto nla_put_failure; 1206 return skb->len; 1207 1208 nla_put_failure: 1209 nla_nest_cancel(skb, nest); 1210 return -1; 1211 } 1212 1213 static struct tcf_proto_ops cls_u32_ops __read_mostly = { 1214 .kind = "u32", 1215 .classify = u32_classify, 1216 .init = u32_init, 1217 .destroy = u32_destroy, 1218 .get = u32_get, 1219 .change = u32_change, 1220 .delete = u32_delete, 1221 .walk = u32_walk, 1222 .dump = u32_dump, 1223 .owner = THIS_MODULE, 1224 }; 1225 1226 static int __init init_u32(void) 1227 { 1228 pr_info("u32 classifier\n"); 1229 #ifdef CONFIG_CLS_U32_PERF 1230 pr_info(" Performance counters on\n"); 1231 #endif 1232 #ifdef CONFIG_NET_CLS_IND 1233 pr_info(" input device check on\n"); 1234 #endif 1235 #ifdef CONFIG_NET_CLS_ACT 1236 pr_info(" Actions configured\n"); 1237 #endif 1238 return register_tcf_proto_ops(&cls_u32_ops); 1239 } 1240 1241 static void __exit exit_u32(void) 1242 { 1243 unregister_tcf_proto_ops(&cls_u32_ops); 1244 } 1245 1246 module_init(init_u32) 1247 module_exit(exit_u32) 1248 MODULE_LICENSE("GPL"); 1249