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/idr.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 idr handle_idr; 89 struct rcu_head rcu; 90 u32 flags; 91 /* The 'ht' field MUST be the last field in structure to allow for 92 * more entries allocated at end of structure. 93 */ 94 struct tc_u_knode __rcu *ht[1]; 95 }; 96 97 struct tc_u_common { 98 struct tc_u_hnode __rcu *hlist; 99 struct tcf_block *block; 100 int refcnt; 101 struct idr handle_idr; 102 struct hlist_node hnode; 103 struct rcu_head rcu; 104 }; 105 106 static inline unsigned int u32_hash_fold(__be32 key, 107 const struct tc_u32_sel *sel, 108 u8 fshift) 109 { 110 unsigned int h = ntohl(key & sel->hmask) >> fshift; 111 112 return h; 113 } 114 115 static int u32_classify(struct sk_buff *skb, const struct tcf_proto *tp, 116 struct tcf_result *res) 117 { 118 struct { 119 struct tc_u_knode *knode; 120 unsigned int off; 121 } stack[TC_U32_MAXDEPTH]; 122 123 struct tc_u_hnode *ht = rcu_dereference_bh(tp->root); 124 unsigned int off = skb_network_offset(skb); 125 struct tc_u_knode *n; 126 int sdepth = 0; 127 int off2 = 0; 128 int sel = 0; 129 #ifdef CONFIG_CLS_U32_PERF 130 int j; 131 #endif 132 int i, r; 133 134 next_ht: 135 n = rcu_dereference_bh(ht->ht[sel]); 136 137 next_knode: 138 if (n) { 139 struct tc_u32_key *key = n->sel.keys; 140 141 #ifdef CONFIG_CLS_U32_PERF 142 __this_cpu_inc(n->pf->rcnt); 143 j = 0; 144 #endif 145 146 if (tc_skip_sw(n->flags)) { 147 n = rcu_dereference_bh(n->next); 148 goto next_knode; 149 } 150 151 #ifdef CONFIG_CLS_U32_MARK 152 if ((skb->mark & n->mask) != n->val) { 153 n = rcu_dereference_bh(n->next); 154 goto next_knode; 155 } else { 156 __this_cpu_inc(*n->pcpu_success); 157 } 158 #endif 159 160 for (i = n->sel.nkeys; i > 0; i--, key++) { 161 int toff = off + key->off + (off2 & key->offmask); 162 __be32 *data, hdata; 163 164 if (skb_headroom(skb) + toff > INT_MAX) 165 goto out; 166 167 data = skb_header_pointer(skb, toff, 4, &hdata); 168 if (!data) 169 goto out; 170 if ((*data ^ key->val) & key->mask) { 171 n = rcu_dereference_bh(n->next); 172 goto next_knode; 173 } 174 #ifdef CONFIG_CLS_U32_PERF 175 __this_cpu_inc(n->pf->kcnts[j]); 176 j++; 177 #endif 178 } 179 180 ht = rcu_dereference_bh(n->ht_down); 181 if (!ht) { 182 check_terminal: 183 if (n->sel.flags & TC_U32_TERMINAL) { 184 185 *res = n->res; 186 #ifdef CONFIG_NET_CLS_IND 187 if (!tcf_match_indev(skb, n->ifindex)) { 188 n = rcu_dereference_bh(n->next); 189 goto next_knode; 190 } 191 #endif 192 #ifdef CONFIG_CLS_U32_PERF 193 __this_cpu_inc(n->pf->rhit); 194 #endif 195 r = tcf_exts_exec(skb, &n->exts, res); 196 if (r < 0) { 197 n = rcu_dereference_bh(n->next); 198 goto next_knode; 199 } 200 201 return r; 202 } 203 n = rcu_dereference_bh(n->next); 204 goto next_knode; 205 } 206 207 /* PUSH */ 208 if (sdepth >= TC_U32_MAXDEPTH) 209 goto deadloop; 210 stack[sdepth].knode = n; 211 stack[sdepth].off = off; 212 sdepth++; 213 214 ht = rcu_dereference_bh(n->ht_down); 215 sel = 0; 216 if (ht->divisor) { 217 __be32 *data, hdata; 218 219 data = skb_header_pointer(skb, off + n->sel.hoff, 4, 220 &hdata); 221 if (!data) 222 goto out; 223 sel = ht->divisor & u32_hash_fold(*data, &n->sel, 224 n->fshift); 225 } 226 if (!(n->sel.flags & (TC_U32_VAROFFSET | TC_U32_OFFSET | TC_U32_EAT))) 227 goto next_ht; 228 229 if (n->sel.flags & (TC_U32_OFFSET | TC_U32_VAROFFSET)) { 230 off2 = n->sel.off + 3; 231 if (n->sel.flags & TC_U32_VAROFFSET) { 232 __be16 *data, hdata; 233 234 data = skb_header_pointer(skb, 235 off + n->sel.offoff, 236 2, &hdata); 237 if (!data) 238 goto out; 239 off2 += ntohs(n->sel.offmask & *data) >> 240 n->sel.offshift; 241 } 242 off2 &= ~3; 243 } 244 if (n->sel.flags & TC_U32_EAT) { 245 off += off2; 246 off2 = 0; 247 } 248 249 if (off < skb->len) 250 goto next_ht; 251 } 252 253 /* POP */ 254 if (sdepth--) { 255 n = stack[sdepth].knode; 256 ht = rcu_dereference_bh(n->ht_up); 257 off = stack[sdepth].off; 258 goto check_terminal; 259 } 260 out: 261 return -1; 262 263 deadloop: 264 net_warn_ratelimited("cls_u32: dead loop\n"); 265 return -1; 266 } 267 268 static struct tc_u_hnode *u32_lookup_ht(struct tc_u_common *tp_c, u32 handle) 269 { 270 struct tc_u_hnode *ht; 271 272 for (ht = rtnl_dereference(tp_c->hlist); 273 ht; 274 ht = rtnl_dereference(ht->next)) 275 if (ht->handle == handle) 276 break; 277 278 return ht; 279 } 280 281 static struct tc_u_knode *u32_lookup_key(struct tc_u_hnode *ht, u32 handle) 282 { 283 unsigned int sel; 284 struct tc_u_knode *n = NULL; 285 286 sel = TC_U32_HASH(handle); 287 if (sel > ht->divisor) 288 goto out; 289 290 for (n = rtnl_dereference(ht->ht[sel]); 291 n; 292 n = rtnl_dereference(n->next)) 293 if (n->handle == handle) 294 break; 295 out: 296 return n; 297 } 298 299 300 static void *u32_get(struct tcf_proto *tp, u32 handle) 301 { 302 struct tc_u_hnode *ht; 303 struct tc_u_common *tp_c = tp->data; 304 305 if (TC_U32_HTID(handle) == TC_U32_ROOT) 306 ht = rtnl_dereference(tp->root); 307 else 308 ht = u32_lookup_ht(tp_c, TC_U32_HTID(handle)); 309 310 if (!ht) 311 return NULL; 312 313 if (TC_U32_KEY(handle) == 0) 314 return ht; 315 316 return u32_lookup_key(ht, handle); 317 } 318 319 static u32 gen_new_htid(struct tc_u_common *tp_c, struct tc_u_hnode *ptr) 320 { 321 unsigned long idr_index; 322 int err; 323 324 /* This is only used inside rtnl lock it is safe to increment 325 * without read _copy_ update semantics 326 */ 327 err = idr_alloc_ext(&tp_c->handle_idr, ptr, &idr_index, 328 1, 0x7FF, GFP_KERNEL); 329 if (err) 330 return 0; 331 return (u32)(idr_index | 0x800) << 20; 332 } 333 334 static struct hlist_head *tc_u_common_hash; 335 336 #define U32_HASH_SHIFT 10 337 #define U32_HASH_SIZE (1 << U32_HASH_SHIFT) 338 339 static unsigned int tc_u_hash(const struct tcf_proto *tp) 340 { 341 return hash_ptr(tp->chain->block, U32_HASH_SHIFT); 342 } 343 344 static struct tc_u_common *tc_u_common_find(const struct tcf_proto *tp) 345 { 346 struct tc_u_common *tc; 347 unsigned int h; 348 349 h = tc_u_hash(tp); 350 hlist_for_each_entry(tc, &tc_u_common_hash[h], hnode) { 351 if (tc->block == tp->chain->block) 352 return tc; 353 } 354 return NULL; 355 } 356 357 static int u32_init(struct tcf_proto *tp) 358 { 359 struct tc_u_hnode *root_ht; 360 struct tc_u_common *tp_c; 361 unsigned int h; 362 363 tp_c = tc_u_common_find(tp); 364 365 root_ht = kzalloc(sizeof(*root_ht), GFP_KERNEL); 366 if (root_ht == NULL) 367 return -ENOBUFS; 368 369 root_ht->refcnt++; 370 root_ht->handle = tp_c ? gen_new_htid(tp_c, root_ht) : 0x80000000; 371 root_ht->prio = tp->prio; 372 idr_init(&root_ht->handle_idr); 373 374 if (tp_c == NULL) { 375 tp_c = kzalloc(sizeof(*tp_c), GFP_KERNEL); 376 if (tp_c == NULL) { 377 kfree(root_ht); 378 return -ENOBUFS; 379 } 380 tp_c->block = tp->chain->block; 381 INIT_HLIST_NODE(&tp_c->hnode); 382 idr_init(&tp_c->handle_idr); 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 tcf_exts_put_net(&n->exts); 403 if (n->ht_down) 404 n->ht_down->refcnt--; 405 #ifdef CONFIG_CLS_U32_PERF 406 if (free_pf) 407 free_percpu(n->pf); 408 #endif 409 #ifdef CONFIG_CLS_U32_MARK 410 if (free_pf) 411 free_percpu(n->pcpu_success); 412 #endif 413 kfree(n); 414 return 0; 415 } 416 417 /* u32_delete_key_rcu should be called when free'ing a copied 418 * version of a tc_u_knode obtained from u32_init_knode(). When 419 * copies are obtained from u32_init_knode() the statistics are 420 * shared between the old and new copies to allow readers to 421 * continue to update the statistics during the copy. To support 422 * this the u32_delete_key_rcu variant does not free the percpu 423 * statistics. 424 */ 425 static void u32_delete_key_work(struct work_struct *work) 426 { 427 struct tc_u_knode *key = container_of(work, struct tc_u_knode, work); 428 429 rtnl_lock(); 430 u32_destroy_key(key->tp, key, false); 431 rtnl_unlock(); 432 } 433 434 static void u32_delete_key_rcu(struct rcu_head *rcu) 435 { 436 struct tc_u_knode *key = container_of(rcu, struct tc_u_knode, rcu); 437 438 INIT_WORK(&key->work, u32_delete_key_work); 439 tcf_queue_work(&key->work); 440 } 441 442 /* u32_delete_key_freepf_rcu is the rcu callback variant 443 * that free's the entire structure including the statistics 444 * percpu variables. Only use this if the key is not a copy 445 * returned by u32_init_knode(). See u32_delete_key_rcu() 446 * for the variant that should be used with keys return from 447 * u32_init_knode() 448 */ 449 static void u32_delete_key_freepf_work(struct work_struct *work) 450 { 451 struct tc_u_knode *key = container_of(work, struct tc_u_knode, work); 452 453 rtnl_lock(); 454 u32_destroy_key(key->tp, key, true); 455 rtnl_unlock(); 456 } 457 458 static void u32_delete_key_freepf_rcu(struct rcu_head *rcu) 459 { 460 struct tc_u_knode *key = container_of(rcu, struct tc_u_knode, rcu); 461 462 INIT_WORK(&key->work, u32_delete_key_freepf_work); 463 tcf_queue_work(&key->work); 464 } 465 466 static int u32_delete_key(struct tcf_proto *tp, struct tc_u_knode *key) 467 { 468 struct tc_u_knode __rcu **kp; 469 struct tc_u_knode *pkp; 470 struct tc_u_hnode *ht = rtnl_dereference(key->ht_up); 471 472 if (ht) { 473 kp = &ht->ht[TC_U32_HASH(key->handle)]; 474 for (pkp = rtnl_dereference(*kp); pkp; 475 kp = &pkp->next, pkp = rtnl_dereference(*kp)) { 476 if (pkp == key) { 477 RCU_INIT_POINTER(*kp, key->next); 478 479 tcf_unbind_filter(tp, &key->res); 480 tcf_exts_get_net(&key->exts); 481 call_rcu(&key->rcu, u32_delete_key_freepf_rcu); 482 return 0; 483 } 484 } 485 } 486 WARN_ON(1); 487 return 0; 488 } 489 490 static void u32_clear_hw_hnode(struct tcf_proto *tp, struct tc_u_hnode *h, 491 struct netlink_ext_ack *extack) 492 { 493 struct tcf_block *block = tp->chain->block; 494 struct tc_cls_u32_offload cls_u32 = {}; 495 496 tc_cls_common_offload_init(&cls_u32.common, tp, h->flags, extack); 497 cls_u32.command = TC_CLSU32_DELETE_HNODE; 498 cls_u32.hnode.divisor = h->divisor; 499 cls_u32.hnode.handle = h->handle; 500 cls_u32.hnode.prio = h->prio; 501 502 tc_setup_cb_call(block, NULL, TC_SETUP_CLSU32, &cls_u32, false); 503 } 504 505 static int u32_replace_hw_hnode(struct tcf_proto *tp, struct tc_u_hnode *h, 506 u32 flags, struct netlink_ext_ack *extack) 507 { 508 struct tcf_block *block = tp->chain->block; 509 struct tc_cls_u32_offload cls_u32 = {}; 510 bool skip_sw = tc_skip_sw(flags); 511 bool offloaded = false; 512 int err; 513 514 tc_cls_common_offload_init(&cls_u32.common, tp, flags, extack); 515 cls_u32.command = TC_CLSU32_NEW_HNODE; 516 cls_u32.hnode.divisor = h->divisor; 517 cls_u32.hnode.handle = h->handle; 518 cls_u32.hnode.prio = h->prio; 519 520 err = tc_setup_cb_call(block, NULL, TC_SETUP_CLSU32, &cls_u32, skip_sw); 521 if (err < 0) { 522 u32_clear_hw_hnode(tp, h, NULL); 523 return err; 524 } else if (err > 0) { 525 offloaded = true; 526 } 527 528 if (skip_sw && !offloaded) 529 return -EINVAL; 530 531 return 0; 532 } 533 534 static void u32_remove_hw_knode(struct tcf_proto *tp, struct tc_u_knode *n, 535 struct netlink_ext_ack *extack) 536 { 537 struct tcf_block *block = tp->chain->block; 538 struct tc_cls_u32_offload cls_u32 = {}; 539 540 tc_cls_common_offload_init(&cls_u32.common, tp, n->flags, extack); 541 cls_u32.command = TC_CLSU32_DELETE_KNODE; 542 cls_u32.knode.handle = n->handle; 543 544 tc_setup_cb_call(block, NULL, TC_SETUP_CLSU32, &cls_u32, false); 545 tcf_block_offload_dec(block, &n->flags); 546 } 547 548 static int u32_replace_hw_knode(struct tcf_proto *tp, struct tc_u_knode *n, 549 u32 flags, struct netlink_ext_ack *extack) 550 { 551 struct tc_u_hnode *ht = rtnl_dereference(n->ht_down); 552 struct tcf_block *block = tp->chain->block; 553 struct tc_cls_u32_offload cls_u32 = {}; 554 bool skip_sw = tc_skip_sw(flags); 555 int err; 556 557 tc_cls_common_offload_init(&cls_u32.common, tp, flags, extack); 558 cls_u32.command = TC_CLSU32_REPLACE_KNODE; 559 cls_u32.knode.handle = n->handle; 560 cls_u32.knode.fshift = n->fshift; 561 #ifdef CONFIG_CLS_U32_MARK 562 cls_u32.knode.val = n->val; 563 cls_u32.knode.mask = n->mask; 564 #else 565 cls_u32.knode.val = 0; 566 cls_u32.knode.mask = 0; 567 #endif 568 cls_u32.knode.sel = &n->sel; 569 cls_u32.knode.exts = &n->exts; 570 if (n->ht_down) 571 cls_u32.knode.link_handle = ht->handle; 572 573 err = tc_setup_cb_call(block, NULL, TC_SETUP_CLSU32, &cls_u32, skip_sw); 574 if (err < 0) { 575 u32_remove_hw_knode(tp, n, NULL); 576 return err; 577 } else if (err > 0) { 578 tcf_block_offload_inc(block, &n->flags); 579 } 580 581 if (skip_sw && !(n->flags & TCA_CLS_FLAGS_IN_HW)) 582 return -EINVAL; 583 584 return 0; 585 } 586 587 static void u32_clear_hnode(struct tcf_proto *tp, struct tc_u_hnode *ht, 588 struct netlink_ext_ack *extack) 589 { 590 struct tc_u_knode *n; 591 unsigned int h; 592 593 for (h = 0; h <= ht->divisor; h++) { 594 while ((n = rtnl_dereference(ht->ht[h])) != NULL) { 595 RCU_INIT_POINTER(ht->ht[h], 596 rtnl_dereference(n->next)); 597 tcf_unbind_filter(tp, &n->res); 598 u32_remove_hw_knode(tp, n, extack); 599 idr_remove_ext(&ht->handle_idr, n->handle); 600 if (tcf_exts_get_net(&n->exts)) 601 call_rcu(&n->rcu, u32_delete_key_freepf_rcu); 602 else 603 u32_destroy_key(n->tp, n, true); 604 } 605 } 606 } 607 608 static int u32_destroy_hnode(struct tcf_proto *tp, struct tc_u_hnode *ht, 609 struct netlink_ext_ack *extack) 610 { 611 struct tc_u_common *tp_c = tp->data; 612 struct tc_u_hnode __rcu **hn; 613 struct tc_u_hnode *phn; 614 615 WARN_ON(ht->refcnt); 616 617 u32_clear_hnode(tp, ht, extack); 618 619 hn = &tp_c->hlist; 620 for (phn = rtnl_dereference(*hn); 621 phn; 622 hn = &phn->next, phn = rtnl_dereference(*hn)) { 623 if (phn == ht) { 624 u32_clear_hw_hnode(tp, ht, extack); 625 idr_destroy(&ht->handle_idr); 626 idr_remove_ext(&tp_c->handle_idr, ht->handle); 627 RCU_INIT_POINTER(*hn, ht->next); 628 kfree_rcu(ht, rcu); 629 return 0; 630 } 631 } 632 633 return -ENOENT; 634 } 635 636 static bool ht_empty(struct tc_u_hnode *ht) 637 { 638 unsigned int h; 639 640 for (h = 0; h <= ht->divisor; h++) 641 if (rcu_access_pointer(ht->ht[h])) 642 return false; 643 644 return true; 645 } 646 647 static void u32_destroy(struct tcf_proto *tp, struct netlink_ext_ack *extack) 648 { 649 struct tc_u_common *tp_c = tp->data; 650 struct tc_u_hnode *root_ht = rtnl_dereference(tp->root); 651 652 WARN_ON(root_ht == NULL); 653 654 if (root_ht && --root_ht->refcnt == 0) 655 u32_destroy_hnode(tp, root_ht, extack); 656 657 if (--tp_c->refcnt == 0) { 658 struct tc_u_hnode *ht; 659 660 hlist_del(&tp_c->hnode); 661 662 for (ht = rtnl_dereference(tp_c->hlist); 663 ht; 664 ht = rtnl_dereference(ht->next)) { 665 ht->refcnt--; 666 u32_clear_hnode(tp, ht, extack); 667 } 668 669 while ((ht = rtnl_dereference(tp_c->hlist)) != NULL) { 670 RCU_INIT_POINTER(tp_c->hlist, ht->next); 671 kfree_rcu(ht, rcu); 672 } 673 674 idr_destroy(&tp_c->handle_idr); 675 kfree(tp_c); 676 } 677 678 tp->data = NULL; 679 } 680 681 static int u32_delete(struct tcf_proto *tp, void *arg, bool *last, 682 struct netlink_ext_ack *extack) 683 { 684 struct tc_u_hnode *ht = arg; 685 struct tc_u_hnode *root_ht = rtnl_dereference(tp->root); 686 struct tc_u_common *tp_c = tp->data; 687 int ret = 0; 688 689 if (ht == NULL) 690 goto out; 691 692 if (TC_U32_KEY(ht->handle)) { 693 u32_remove_hw_knode(tp, (struct tc_u_knode *)ht, extack); 694 ret = u32_delete_key(tp, (struct tc_u_knode *)ht); 695 goto out; 696 } 697 698 if (root_ht == ht) { 699 NL_SET_ERR_MSG_MOD(extack, "Not allowed to delete root node"); 700 return -EINVAL; 701 } 702 703 if (ht->refcnt == 1) { 704 ht->refcnt--; 705 u32_destroy_hnode(tp, ht, extack); 706 } else { 707 NL_SET_ERR_MSG_MOD(extack, "Can not delete in-use filter"); 708 return -EBUSY; 709 } 710 711 out: 712 *last = true; 713 if (root_ht) { 714 if (root_ht->refcnt > 1) { 715 *last = false; 716 goto ret; 717 } 718 if (root_ht->refcnt == 1) { 719 if (!ht_empty(root_ht)) { 720 *last = false; 721 goto ret; 722 } 723 } 724 } 725 726 if (tp_c->refcnt > 1) { 727 *last = false; 728 goto ret; 729 } 730 731 if (tp_c->refcnt == 1) { 732 struct tc_u_hnode *ht; 733 734 for (ht = rtnl_dereference(tp_c->hlist); 735 ht; 736 ht = rtnl_dereference(ht->next)) 737 if (!ht_empty(ht)) { 738 *last = false; 739 break; 740 } 741 } 742 743 ret: 744 return ret; 745 } 746 747 static u32 gen_new_kid(struct tc_u_hnode *ht, u32 htid) 748 { 749 unsigned long idr_index; 750 u32 start = htid | 0x800; 751 u32 max = htid | 0xFFF; 752 u32 min = htid; 753 754 if (idr_alloc_ext(&ht->handle_idr, NULL, &idr_index, 755 start, max + 1, GFP_KERNEL)) { 756 if (idr_alloc_ext(&ht->handle_idr, NULL, &idr_index, 757 min + 1, max + 1, GFP_KERNEL)) 758 return max; 759 } 760 761 return (u32)idr_index; 762 } 763 764 static const struct nla_policy u32_policy[TCA_U32_MAX + 1] = { 765 [TCA_U32_CLASSID] = { .type = NLA_U32 }, 766 [TCA_U32_HASH] = { .type = NLA_U32 }, 767 [TCA_U32_LINK] = { .type = NLA_U32 }, 768 [TCA_U32_DIVISOR] = { .type = NLA_U32 }, 769 [TCA_U32_SEL] = { .len = sizeof(struct tc_u32_sel) }, 770 [TCA_U32_INDEV] = { .type = NLA_STRING, .len = IFNAMSIZ }, 771 [TCA_U32_MARK] = { .len = sizeof(struct tc_u32_mark) }, 772 [TCA_U32_FLAGS] = { .type = NLA_U32 }, 773 }; 774 775 static int u32_set_parms(struct net *net, struct tcf_proto *tp, 776 unsigned long base, struct tc_u_hnode *ht, 777 struct tc_u_knode *n, struct nlattr **tb, 778 struct nlattr *est, bool ovr, 779 struct netlink_ext_ack *extack) 780 { 781 int err; 782 783 err = tcf_exts_validate(net, tp, tb, est, &n->exts, ovr, extack); 784 if (err < 0) 785 return err; 786 787 if (tb[TCA_U32_LINK]) { 788 u32 handle = nla_get_u32(tb[TCA_U32_LINK]); 789 struct tc_u_hnode *ht_down = NULL, *ht_old; 790 791 if (TC_U32_KEY(handle)) { 792 NL_SET_ERR_MSG_MOD(extack, "u32 Link handle must be a hash table"); 793 return -EINVAL; 794 } 795 796 if (handle) { 797 ht_down = u32_lookup_ht(ht->tp_c, handle); 798 799 if (!ht_down) { 800 NL_SET_ERR_MSG_MOD(extack, "Link hash table not found"); 801 return -EINVAL; 802 } 803 ht_down->refcnt++; 804 } 805 806 ht_old = rtnl_dereference(n->ht_down); 807 rcu_assign_pointer(n->ht_down, ht_down); 808 809 if (ht_old) 810 ht_old->refcnt--; 811 } 812 if (tb[TCA_U32_CLASSID]) { 813 n->res.classid = nla_get_u32(tb[TCA_U32_CLASSID]); 814 tcf_bind_filter(tp, &n->res, base); 815 } 816 817 #ifdef CONFIG_NET_CLS_IND 818 if (tb[TCA_U32_INDEV]) { 819 int ret; 820 ret = tcf_change_indev(net, tb[TCA_U32_INDEV], extack); 821 if (ret < 0) 822 return -EINVAL; 823 n->ifindex = ret; 824 } 825 #endif 826 return 0; 827 } 828 829 static void u32_replace_knode(struct tcf_proto *tp, struct tc_u_common *tp_c, 830 struct tc_u_knode *n) 831 { 832 struct tc_u_knode __rcu **ins; 833 struct tc_u_knode *pins; 834 struct tc_u_hnode *ht; 835 836 if (TC_U32_HTID(n->handle) == TC_U32_ROOT) 837 ht = rtnl_dereference(tp->root); 838 else 839 ht = u32_lookup_ht(tp_c, TC_U32_HTID(n->handle)); 840 841 ins = &ht->ht[TC_U32_HASH(n->handle)]; 842 843 /* The node must always exist for it to be replaced if this is not the 844 * case then something went very wrong elsewhere. 845 */ 846 for (pins = rtnl_dereference(*ins); ; 847 ins = &pins->next, pins = rtnl_dereference(*ins)) 848 if (pins->handle == n->handle) 849 break; 850 851 idr_replace_ext(&ht->handle_idr, n, n->handle); 852 RCU_INIT_POINTER(n->next, pins->next); 853 rcu_assign_pointer(*ins, n); 854 } 855 856 static struct tc_u_knode *u32_init_knode(struct tcf_proto *tp, 857 struct tc_u_knode *n) 858 { 859 struct tc_u_hnode *ht = rtnl_dereference(n->ht_down); 860 struct tc_u32_sel *s = &n->sel; 861 struct tc_u_knode *new; 862 863 new = kzalloc(sizeof(*n) + s->nkeys*sizeof(struct tc_u32_key), 864 GFP_KERNEL); 865 866 if (!new) 867 return NULL; 868 869 RCU_INIT_POINTER(new->next, n->next); 870 new->handle = n->handle; 871 RCU_INIT_POINTER(new->ht_up, n->ht_up); 872 873 #ifdef CONFIG_NET_CLS_IND 874 new->ifindex = n->ifindex; 875 #endif 876 new->fshift = n->fshift; 877 new->res = n->res; 878 new->flags = n->flags; 879 RCU_INIT_POINTER(new->ht_down, ht); 880 881 /* bump reference count as long as we hold pointer to structure */ 882 if (ht) 883 ht->refcnt++; 884 885 #ifdef CONFIG_CLS_U32_PERF 886 /* Statistics may be incremented by readers during update 887 * so we must keep them in tact. When the node is later destroyed 888 * a special destroy call must be made to not free the pf memory. 889 */ 890 new->pf = n->pf; 891 #endif 892 893 #ifdef CONFIG_CLS_U32_MARK 894 new->val = n->val; 895 new->mask = n->mask; 896 /* Similarly success statistics must be moved as pointers */ 897 new->pcpu_success = n->pcpu_success; 898 #endif 899 new->tp = tp; 900 memcpy(&new->sel, s, sizeof(*s) + s->nkeys*sizeof(struct tc_u32_key)); 901 902 if (tcf_exts_init(&new->exts, TCA_U32_ACT, TCA_U32_POLICE)) { 903 kfree(new); 904 return NULL; 905 } 906 907 return new; 908 } 909 910 static int u32_change(struct net *net, struct sk_buff *in_skb, 911 struct tcf_proto *tp, unsigned long base, u32 handle, 912 struct nlattr **tca, void **arg, bool ovr, 913 struct netlink_ext_ack *extack) 914 { 915 struct tc_u_common *tp_c = tp->data; 916 struct tc_u_hnode *ht; 917 struct tc_u_knode *n; 918 struct tc_u32_sel *s; 919 struct nlattr *opt = tca[TCA_OPTIONS]; 920 struct nlattr *tb[TCA_U32_MAX + 1]; 921 u32 htid, flags = 0; 922 int err; 923 #ifdef CONFIG_CLS_U32_PERF 924 size_t size; 925 #endif 926 927 if (!opt) { 928 if (handle) { 929 NL_SET_ERR_MSG_MOD(extack, "Filter handle requires options"); 930 return -EINVAL; 931 } else { 932 return 0; 933 } 934 } 935 936 err = nla_parse_nested(tb, TCA_U32_MAX, opt, u32_policy, extack); 937 if (err < 0) 938 return err; 939 940 if (tb[TCA_U32_FLAGS]) { 941 flags = nla_get_u32(tb[TCA_U32_FLAGS]); 942 if (!tc_flags_valid(flags)) { 943 NL_SET_ERR_MSG_MOD(extack, "Invalid filter flags"); 944 return -EINVAL; 945 } 946 } 947 948 n = *arg; 949 if (n) { 950 struct tc_u_knode *new; 951 952 if (TC_U32_KEY(n->handle) == 0) { 953 NL_SET_ERR_MSG_MOD(extack, "Key node id cannot be zero"); 954 return -EINVAL; 955 } 956 957 if (n->flags != flags) { 958 NL_SET_ERR_MSG_MOD(extack, "Key node flags do not match passed flags"); 959 return -EINVAL; 960 } 961 962 new = u32_init_knode(tp, n); 963 if (!new) 964 return -ENOMEM; 965 966 err = u32_set_parms(net, tp, base, 967 rtnl_dereference(n->ht_up), new, tb, 968 tca[TCA_RATE], ovr, extack); 969 970 if (err) { 971 u32_destroy_key(tp, new, false); 972 return err; 973 } 974 975 err = u32_replace_hw_knode(tp, new, flags, extack); 976 if (err) { 977 u32_destroy_key(tp, new, false); 978 return err; 979 } 980 981 if (!tc_in_hw(new->flags)) 982 new->flags |= TCA_CLS_FLAGS_NOT_IN_HW; 983 984 u32_replace_knode(tp, tp_c, new); 985 tcf_unbind_filter(tp, &n->res); 986 tcf_exts_get_net(&n->exts); 987 call_rcu(&n->rcu, u32_delete_key_rcu); 988 return 0; 989 } 990 991 if (tb[TCA_U32_DIVISOR]) { 992 unsigned int divisor = nla_get_u32(tb[TCA_U32_DIVISOR]); 993 994 if (--divisor > 0x100) { 995 NL_SET_ERR_MSG_MOD(extack, "Exceeded maximum 256 hash buckets"); 996 return -EINVAL; 997 } 998 if (TC_U32_KEY(handle)) { 999 NL_SET_ERR_MSG_MOD(extack, "Divisor can only be used on a hash table"); 1000 return -EINVAL; 1001 } 1002 ht = kzalloc(sizeof(*ht) + divisor*sizeof(void *), GFP_KERNEL); 1003 if (ht == NULL) 1004 return -ENOBUFS; 1005 if (handle == 0) { 1006 handle = gen_new_htid(tp->data, ht); 1007 if (handle == 0) { 1008 kfree(ht); 1009 return -ENOMEM; 1010 } 1011 } else { 1012 err = idr_alloc_ext(&tp_c->handle_idr, ht, NULL, 1013 handle, handle + 1, GFP_KERNEL); 1014 if (err) { 1015 kfree(ht); 1016 return err; 1017 } 1018 } 1019 ht->tp_c = tp_c; 1020 ht->refcnt = 1; 1021 ht->divisor = divisor; 1022 ht->handle = handle; 1023 ht->prio = tp->prio; 1024 idr_init(&ht->handle_idr); 1025 ht->flags = flags; 1026 1027 err = u32_replace_hw_hnode(tp, ht, flags, extack); 1028 if (err) { 1029 idr_remove_ext(&tp_c->handle_idr, handle); 1030 kfree(ht); 1031 return err; 1032 } 1033 1034 RCU_INIT_POINTER(ht->next, tp_c->hlist); 1035 rcu_assign_pointer(tp_c->hlist, ht); 1036 *arg = ht; 1037 1038 return 0; 1039 } 1040 1041 if (tb[TCA_U32_HASH]) { 1042 htid = nla_get_u32(tb[TCA_U32_HASH]); 1043 if (TC_U32_HTID(htid) == TC_U32_ROOT) { 1044 ht = rtnl_dereference(tp->root); 1045 htid = ht->handle; 1046 } else { 1047 ht = u32_lookup_ht(tp->data, TC_U32_HTID(htid)); 1048 if (!ht) { 1049 NL_SET_ERR_MSG_MOD(extack, "Specified hash table not found"); 1050 return -EINVAL; 1051 } 1052 } 1053 } else { 1054 ht = rtnl_dereference(tp->root); 1055 htid = ht->handle; 1056 } 1057 1058 if (ht->divisor < TC_U32_HASH(htid)) { 1059 NL_SET_ERR_MSG_MOD(extack, "Specified hash table buckets exceed configured value"); 1060 return -EINVAL; 1061 } 1062 1063 if (handle) { 1064 if (TC_U32_HTID(handle) && TC_U32_HTID(handle ^ htid)) { 1065 NL_SET_ERR_MSG_MOD(extack, "Handle specified hash table address mismatch"); 1066 return -EINVAL; 1067 } 1068 handle = htid | TC_U32_NODE(handle); 1069 err = idr_alloc_ext(&ht->handle_idr, NULL, NULL, 1070 handle, handle + 1, 1071 GFP_KERNEL); 1072 if (err) 1073 return err; 1074 } else 1075 handle = gen_new_kid(ht, htid); 1076 1077 if (tb[TCA_U32_SEL] == NULL) { 1078 NL_SET_ERR_MSG_MOD(extack, "Selector not specified"); 1079 err = -EINVAL; 1080 goto erridr; 1081 } 1082 1083 s = nla_data(tb[TCA_U32_SEL]); 1084 1085 n = kzalloc(sizeof(*n) + s->nkeys*sizeof(struct tc_u32_key), GFP_KERNEL); 1086 if (n == NULL) { 1087 err = -ENOBUFS; 1088 goto erridr; 1089 } 1090 1091 #ifdef CONFIG_CLS_U32_PERF 1092 size = sizeof(struct tc_u32_pcnt) + s->nkeys * sizeof(u64); 1093 n->pf = __alloc_percpu(size, __alignof__(struct tc_u32_pcnt)); 1094 if (!n->pf) { 1095 err = -ENOBUFS; 1096 goto errfree; 1097 } 1098 #endif 1099 1100 memcpy(&n->sel, s, sizeof(*s) + s->nkeys*sizeof(struct tc_u32_key)); 1101 RCU_INIT_POINTER(n->ht_up, ht); 1102 n->handle = handle; 1103 n->fshift = s->hmask ? ffs(ntohl(s->hmask)) - 1 : 0; 1104 n->flags = flags; 1105 n->tp = tp; 1106 1107 err = tcf_exts_init(&n->exts, TCA_U32_ACT, TCA_U32_POLICE); 1108 if (err < 0) 1109 goto errout; 1110 1111 #ifdef CONFIG_CLS_U32_MARK 1112 n->pcpu_success = alloc_percpu(u32); 1113 if (!n->pcpu_success) { 1114 err = -ENOMEM; 1115 goto errout; 1116 } 1117 1118 if (tb[TCA_U32_MARK]) { 1119 struct tc_u32_mark *mark; 1120 1121 mark = nla_data(tb[TCA_U32_MARK]); 1122 n->val = mark->val; 1123 n->mask = mark->mask; 1124 } 1125 #endif 1126 1127 err = u32_set_parms(net, tp, base, ht, n, tb, tca[TCA_RATE], ovr, 1128 extack); 1129 if (err == 0) { 1130 struct tc_u_knode __rcu **ins; 1131 struct tc_u_knode *pins; 1132 1133 err = u32_replace_hw_knode(tp, n, flags, extack); 1134 if (err) 1135 goto errhw; 1136 1137 if (!tc_in_hw(n->flags)) 1138 n->flags |= TCA_CLS_FLAGS_NOT_IN_HW; 1139 1140 ins = &ht->ht[TC_U32_HASH(handle)]; 1141 for (pins = rtnl_dereference(*ins); pins; 1142 ins = &pins->next, pins = rtnl_dereference(*ins)) 1143 if (TC_U32_NODE(handle) < TC_U32_NODE(pins->handle)) 1144 break; 1145 1146 RCU_INIT_POINTER(n->next, pins); 1147 rcu_assign_pointer(*ins, n); 1148 *arg = n; 1149 return 0; 1150 } 1151 1152 errhw: 1153 #ifdef CONFIG_CLS_U32_MARK 1154 free_percpu(n->pcpu_success); 1155 #endif 1156 1157 errout: 1158 tcf_exts_destroy(&n->exts); 1159 #ifdef CONFIG_CLS_U32_PERF 1160 errfree: 1161 free_percpu(n->pf); 1162 #endif 1163 kfree(n); 1164 erridr: 1165 idr_remove_ext(&ht->handle_idr, handle); 1166 return err; 1167 } 1168 1169 static void u32_walk(struct tcf_proto *tp, struct tcf_walker *arg) 1170 { 1171 struct tc_u_common *tp_c = tp->data; 1172 struct tc_u_hnode *ht; 1173 struct tc_u_knode *n; 1174 unsigned int h; 1175 1176 if (arg->stop) 1177 return; 1178 1179 for (ht = rtnl_dereference(tp_c->hlist); 1180 ht; 1181 ht = rtnl_dereference(ht->next)) { 1182 if (ht->prio != tp->prio) 1183 continue; 1184 if (arg->count >= arg->skip) { 1185 if (arg->fn(tp, ht, arg) < 0) { 1186 arg->stop = 1; 1187 return; 1188 } 1189 } 1190 arg->count++; 1191 for (h = 0; h <= ht->divisor; h++) { 1192 for (n = rtnl_dereference(ht->ht[h]); 1193 n; 1194 n = rtnl_dereference(n->next)) { 1195 if (arg->count < arg->skip) { 1196 arg->count++; 1197 continue; 1198 } 1199 if (arg->fn(tp, n, arg) < 0) { 1200 arg->stop = 1; 1201 return; 1202 } 1203 arg->count++; 1204 } 1205 } 1206 } 1207 } 1208 1209 static void u32_bind_class(void *fh, u32 classid, unsigned long cl) 1210 { 1211 struct tc_u_knode *n = fh; 1212 1213 if (n && n->res.classid == classid) 1214 n->res.class = cl; 1215 } 1216 1217 static int u32_dump(struct net *net, struct tcf_proto *tp, void *fh, 1218 struct sk_buff *skb, struct tcmsg *t) 1219 { 1220 struct tc_u_knode *n = fh; 1221 struct tc_u_hnode *ht_up, *ht_down; 1222 struct nlattr *nest; 1223 1224 if (n == NULL) 1225 return skb->len; 1226 1227 t->tcm_handle = n->handle; 1228 1229 nest = nla_nest_start(skb, TCA_OPTIONS); 1230 if (nest == NULL) 1231 goto nla_put_failure; 1232 1233 if (TC_U32_KEY(n->handle) == 0) { 1234 struct tc_u_hnode *ht = fh; 1235 u32 divisor = ht->divisor + 1; 1236 1237 if (nla_put_u32(skb, TCA_U32_DIVISOR, divisor)) 1238 goto nla_put_failure; 1239 } else { 1240 #ifdef CONFIG_CLS_U32_PERF 1241 struct tc_u32_pcnt *gpf; 1242 int cpu; 1243 #endif 1244 1245 if (nla_put(skb, TCA_U32_SEL, 1246 sizeof(n->sel) + n->sel.nkeys*sizeof(struct tc_u32_key), 1247 &n->sel)) 1248 goto nla_put_failure; 1249 1250 ht_up = rtnl_dereference(n->ht_up); 1251 if (ht_up) { 1252 u32 htid = n->handle & 0xFFFFF000; 1253 if (nla_put_u32(skb, TCA_U32_HASH, htid)) 1254 goto nla_put_failure; 1255 } 1256 if (n->res.classid && 1257 nla_put_u32(skb, TCA_U32_CLASSID, n->res.classid)) 1258 goto nla_put_failure; 1259 1260 ht_down = rtnl_dereference(n->ht_down); 1261 if (ht_down && 1262 nla_put_u32(skb, TCA_U32_LINK, ht_down->handle)) 1263 goto nla_put_failure; 1264 1265 if (n->flags && nla_put_u32(skb, TCA_U32_FLAGS, n->flags)) 1266 goto nla_put_failure; 1267 1268 #ifdef CONFIG_CLS_U32_MARK 1269 if ((n->val || n->mask)) { 1270 struct tc_u32_mark mark = {.val = n->val, 1271 .mask = n->mask, 1272 .success = 0}; 1273 int cpum; 1274 1275 for_each_possible_cpu(cpum) { 1276 __u32 cnt = *per_cpu_ptr(n->pcpu_success, cpum); 1277 1278 mark.success += cnt; 1279 } 1280 1281 if (nla_put(skb, TCA_U32_MARK, sizeof(mark), &mark)) 1282 goto nla_put_failure; 1283 } 1284 #endif 1285 1286 if (tcf_exts_dump(skb, &n->exts) < 0) 1287 goto nla_put_failure; 1288 1289 #ifdef CONFIG_NET_CLS_IND 1290 if (n->ifindex) { 1291 struct net_device *dev; 1292 dev = __dev_get_by_index(net, n->ifindex); 1293 if (dev && nla_put_string(skb, TCA_U32_INDEV, dev->name)) 1294 goto nla_put_failure; 1295 } 1296 #endif 1297 #ifdef CONFIG_CLS_U32_PERF 1298 gpf = kzalloc(sizeof(struct tc_u32_pcnt) + 1299 n->sel.nkeys * sizeof(u64), 1300 GFP_KERNEL); 1301 if (!gpf) 1302 goto nla_put_failure; 1303 1304 for_each_possible_cpu(cpu) { 1305 int i; 1306 struct tc_u32_pcnt *pf = per_cpu_ptr(n->pf, cpu); 1307 1308 gpf->rcnt += pf->rcnt; 1309 gpf->rhit += pf->rhit; 1310 for (i = 0; i < n->sel.nkeys; i++) 1311 gpf->kcnts[i] += pf->kcnts[i]; 1312 } 1313 1314 if (nla_put_64bit(skb, TCA_U32_PCNT, 1315 sizeof(struct tc_u32_pcnt) + 1316 n->sel.nkeys * sizeof(u64), 1317 gpf, TCA_U32_PAD)) { 1318 kfree(gpf); 1319 goto nla_put_failure; 1320 } 1321 kfree(gpf); 1322 #endif 1323 } 1324 1325 nla_nest_end(skb, nest); 1326 1327 if (TC_U32_KEY(n->handle)) 1328 if (tcf_exts_dump_stats(skb, &n->exts) < 0) 1329 goto nla_put_failure; 1330 return skb->len; 1331 1332 nla_put_failure: 1333 nla_nest_cancel(skb, nest); 1334 return -1; 1335 } 1336 1337 static struct tcf_proto_ops cls_u32_ops __read_mostly = { 1338 .kind = "u32", 1339 .classify = u32_classify, 1340 .init = u32_init, 1341 .destroy = u32_destroy, 1342 .get = u32_get, 1343 .change = u32_change, 1344 .delete = u32_delete, 1345 .walk = u32_walk, 1346 .dump = u32_dump, 1347 .bind_class = u32_bind_class, 1348 .owner = THIS_MODULE, 1349 }; 1350 1351 static int __init init_u32(void) 1352 { 1353 int i, ret; 1354 1355 pr_info("u32 classifier\n"); 1356 #ifdef CONFIG_CLS_U32_PERF 1357 pr_info(" Performance counters on\n"); 1358 #endif 1359 #ifdef CONFIG_NET_CLS_IND 1360 pr_info(" input device check on\n"); 1361 #endif 1362 #ifdef CONFIG_NET_CLS_ACT 1363 pr_info(" Actions configured\n"); 1364 #endif 1365 tc_u_common_hash = kvmalloc_array(U32_HASH_SIZE, 1366 sizeof(struct hlist_head), 1367 GFP_KERNEL); 1368 if (!tc_u_common_hash) 1369 return -ENOMEM; 1370 1371 for (i = 0; i < U32_HASH_SIZE; i++) 1372 INIT_HLIST_HEAD(&tc_u_common_hash[i]); 1373 1374 ret = register_tcf_proto_ops(&cls_u32_ops); 1375 if (ret) 1376 kvfree(tc_u_common_hash); 1377 return ret; 1378 } 1379 1380 static void __exit exit_u32(void) 1381 { 1382 unregister_tcf_proto_ops(&cls_u32_ops); 1383 kvfree(tc_u_common_hash); 1384 } 1385 1386 module_init(init_u32) 1387 module_exit(exit_u32) 1388 MODULE_LICENSE("GPL"); 1389