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