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 struct tc_u_hnode *ht = rtnl_dereference(n->ht_down); 402 403 tcf_exts_destroy(&n->exts); 404 tcf_exts_put_net(&n->exts); 405 if (ht && --ht->refcnt == 0) 406 kfree(ht); 407 #ifdef CONFIG_CLS_U32_PERF 408 if (free_pf) 409 free_percpu(n->pf); 410 #endif 411 #ifdef CONFIG_CLS_U32_MARK 412 if (free_pf) 413 free_percpu(n->pcpu_success); 414 #endif 415 kfree(n); 416 return 0; 417 } 418 419 /* u32_delete_key_rcu should be called when free'ing a copied 420 * version of a tc_u_knode obtained from u32_init_knode(). When 421 * copies are obtained from u32_init_knode() the statistics are 422 * shared between the old and new copies to allow readers to 423 * continue to update the statistics during the copy. To support 424 * this the u32_delete_key_rcu variant does not free the percpu 425 * statistics. 426 */ 427 static void u32_delete_key_work(struct work_struct *work) 428 { 429 struct tc_u_knode *key = container_of(work, struct tc_u_knode, work); 430 431 rtnl_lock(); 432 u32_destroy_key(key->tp, key, false); 433 rtnl_unlock(); 434 } 435 436 static void u32_delete_key_rcu(struct rcu_head *rcu) 437 { 438 struct tc_u_knode *key = container_of(rcu, struct tc_u_knode, rcu); 439 440 INIT_WORK(&key->work, u32_delete_key_work); 441 tcf_queue_work(&key->work); 442 } 443 444 /* u32_delete_key_freepf_rcu is the rcu callback variant 445 * that free's the entire structure including the statistics 446 * percpu variables. Only use this if the key is not a copy 447 * returned by u32_init_knode(). See u32_delete_key_rcu() 448 * for the variant that should be used with keys return from 449 * u32_init_knode() 450 */ 451 static void u32_delete_key_freepf_work(struct work_struct *work) 452 { 453 struct tc_u_knode *key = container_of(work, struct tc_u_knode, work); 454 455 rtnl_lock(); 456 u32_destroy_key(key->tp, key, true); 457 rtnl_unlock(); 458 } 459 460 static void u32_delete_key_freepf_rcu(struct rcu_head *rcu) 461 { 462 struct tc_u_knode *key = container_of(rcu, struct tc_u_knode, rcu); 463 464 INIT_WORK(&key->work, u32_delete_key_freepf_work); 465 tcf_queue_work(&key->work); 466 } 467 468 static int u32_delete_key(struct tcf_proto *tp, struct tc_u_knode *key) 469 { 470 struct tc_u_knode __rcu **kp; 471 struct tc_u_knode *pkp; 472 struct tc_u_hnode *ht = rtnl_dereference(key->ht_up); 473 474 if (ht) { 475 kp = &ht->ht[TC_U32_HASH(key->handle)]; 476 for (pkp = rtnl_dereference(*kp); pkp; 477 kp = &pkp->next, pkp = rtnl_dereference(*kp)) { 478 if (pkp == key) { 479 RCU_INIT_POINTER(*kp, key->next); 480 481 tcf_unbind_filter(tp, &key->res); 482 tcf_exts_get_net(&key->exts); 483 call_rcu(&key->rcu, u32_delete_key_freepf_rcu); 484 return 0; 485 } 486 } 487 } 488 WARN_ON(1); 489 return 0; 490 } 491 492 static void u32_clear_hw_hnode(struct tcf_proto *tp, struct tc_u_hnode *h, 493 struct netlink_ext_ack *extack) 494 { 495 struct tcf_block *block = tp->chain->block; 496 struct tc_cls_u32_offload cls_u32 = {}; 497 498 tc_cls_common_offload_init(&cls_u32.common, tp, h->flags, extack); 499 cls_u32.command = TC_CLSU32_DELETE_HNODE; 500 cls_u32.hnode.divisor = h->divisor; 501 cls_u32.hnode.handle = h->handle; 502 cls_u32.hnode.prio = h->prio; 503 504 tc_setup_cb_call(block, NULL, TC_SETUP_CLSU32, &cls_u32, false); 505 } 506 507 static int u32_replace_hw_hnode(struct tcf_proto *tp, struct tc_u_hnode *h, 508 u32 flags, struct netlink_ext_ack *extack) 509 { 510 struct tcf_block *block = tp->chain->block; 511 struct tc_cls_u32_offload cls_u32 = {}; 512 bool skip_sw = tc_skip_sw(flags); 513 bool offloaded = false; 514 int err; 515 516 tc_cls_common_offload_init(&cls_u32.common, tp, flags, extack); 517 cls_u32.command = TC_CLSU32_NEW_HNODE; 518 cls_u32.hnode.divisor = h->divisor; 519 cls_u32.hnode.handle = h->handle; 520 cls_u32.hnode.prio = h->prio; 521 522 err = tc_setup_cb_call(block, NULL, TC_SETUP_CLSU32, &cls_u32, skip_sw); 523 if (err < 0) { 524 u32_clear_hw_hnode(tp, h, NULL); 525 return err; 526 } else if (err > 0) { 527 offloaded = true; 528 } 529 530 if (skip_sw && !offloaded) 531 return -EINVAL; 532 533 return 0; 534 } 535 536 static void u32_remove_hw_knode(struct tcf_proto *tp, struct tc_u_knode *n, 537 struct netlink_ext_ack *extack) 538 { 539 struct tcf_block *block = tp->chain->block; 540 struct tc_cls_u32_offload cls_u32 = {}; 541 542 tc_cls_common_offload_init(&cls_u32.common, tp, n->flags, extack); 543 cls_u32.command = TC_CLSU32_DELETE_KNODE; 544 cls_u32.knode.handle = n->handle; 545 546 tc_setup_cb_call(block, NULL, TC_SETUP_CLSU32, &cls_u32, false); 547 tcf_block_offload_dec(block, &n->flags); 548 } 549 550 static int u32_replace_hw_knode(struct tcf_proto *tp, struct tc_u_knode *n, 551 u32 flags, struct netlink_ext_ack *extack) 552 { 553 struct tc_u_hnode *ht = rtnl_dereference(n->ht_down); 554 struct tcf_block *block = tp->chain->block; 555 struct tc_cls_u32_offload cls_u32 = {}; 556 bool skip_sw = tc_skip_sw(flags); 557 int err; 558 559 tc_cls_common_offload_init(&cls_u32.common, tp, flags, extack); 560 cls_u32.command = TC_CLSU32_REPLACE_KNODE; 561 cls_u32.knode.handle = n->handle; 562 cls_u32.knode.fshift = n->fshift; 563 #ifdef CONFIG_CLS_U32_MARK 564 cls_u32.knode.val = n->val; 565 cls_u32.knode.mask = n->mask; 566 #else 567 cls_u32.knode.val = 0; 568 cls_u32.knode.mask = 0; 569 #endif 570 cls_u32.knode.sel = &n->sel; 571 cls_u32.knode.exts = &n->exts; 572 if (n->ht_down) 573 cls_u32.knode.link_handle = ht->handle; 574 575 err = tc_setup_cb_call(block, NULL, TC_SETUP_CLSU32, &cls_u32, skip_sw); 576 if (err < 0) { 577 u32_remove_hw_knode(tp, n, NULL); 578 return err; 579 } else if (err > 0) { 580 tcf_block_offload_inc(block, &n->flags); 581 } 582 583 if (skip_sw && !(n->flags & TCA_CLS_FLAGS_IN_HW)) 584 return -EINVAL; 585 586 return 0; 587 } 588 589 static void u32_clear_hnode(struct tcf_proto *tp, struct tc_u_hnode *ht, 590 struct netlink_ext_ack *extack) 591 { 592 struct tc_u_knode *n; 593 unsigned int h; 594 595 for (h = 0; h <= ht->divisor; h++) { 596 while ((n = rtnl_dereference(ht->ht[h])) != NULL) { 597 RCU_INIT_POINTER(ht->ht[h], 598 rtnl_dereference(n->next)); 599 tcf_unbind_filter(tp, &n->res); 600 u32_remove_hw_knode(tp, n, extack); 601 idr_remove_ext(&ht->handle_idr, n->handle); 602 if (tcf_exts_get_net(&n->exts)) 603 call_rcu(&n->rcu, u32_delete_key_freepf_rcu); 604 else 605 u32_destroy_key(n->tp, n, true); 606 } 607 } 608 } 609 610 static int u32_destroy_hnode(struct tcf_proto *tp, struct tc_u_hnode *ht, 611 struct netlink_ext_ack *extack) 612 { 613 struct tc_u_common *tp_c = tp->data; 614 struct tc_u_hnode __rcu **hn; 615 struct tc_u_hnode *phn; 616 617 WARN_ON(ht->refcnt); 618 619 u32_clear_hnode(tp, ht, extack); 620 621 hn = &tp_c->hlist; 622 for (phn = rtnl_dereference(*hn); 623 phn; 624 hn = &phn->next, phn = rtnl_dereference(*hn)) { 625 if (phn == ht) { 626 u32_clear_hw_hnode(tp, ht, extack); 627 idr_destroy(&ht->handle_idr); 628 idr_remove_ext(&tp_c->handle_idr, ht->handle); 629 RCU_INIT_POINTER(*hn, ht->next); 630 kfree_rcu(ht, rcu); 631 return 0; 632 } 633 } 634 635 return -ENOENT; 636 } 637 638 static bool ht_empty(struct tc_u_hnode *ht) 639 { 640 unsigned int h; 641 642 for (h = 0; h <= ht->divisor; h++) 643 if (rcu_access_pointer(ht->ht[h])) 644 return false; 645 646 return true; 647 } 648 649 static void u32_destroy(struct tcf_proto *tp, struct netlink_ext_ack *extack) 650 { 651 struct tc_u_common *tp_c = tp->data; 652 struct tc_u_hnode *root_ht = rtnl_dereference(tp->root); 653 654 WARN_ON(root_ht == NULL); 655 656 if (root_ht && --root_ht->refcnt == 0) 657 u32_destroy_hnode(tp, root_ht, extack); 658 659 if (--tp_c->refcnt == 0) { 660 struct tc_u_hnode *ht; 661 662 hlist_del(&tp_c->hnode); 663 664 while ((ht = rtnl_dereference(tp_c->hlist)) != NULL) { 665 u32_clear_hnode(tp, ht, extack); 666 RCU_INIT_POINTER(tp_c->hlist, ht->next); 667 668 /* u32_destroy_key() will later free ht for us, if it's 669 * still referenced by some knode 670 */ 671 if (--ht->refcnt == 0) 672 kfree_rcu(ht, rcu); 673 } 674 675 idr_destroy(&tp_c->handle_idr); 676 kfree(tp_c); 677 } 678 679 tp->data = NULL; 680 } 681 682 static int u32_delete(struct tcf_proto *tp, void *arg, bool *last, 683 struct netlink_ext_ack *extack) 684 { 685 struct tc_u_hnode *ht = arg; 686 struct tc_u_hnode *root_ht = rtnl_dereference(tp->root); 687 struct tc_u_common *tp_c = tp->data; 688 int ret = 0; 689 690 if (ht == NULL) 691 goto out; 692 693 if (TC_U32_KEY(ht->handle)) { 694 u32_remove_hw_knode(tp, (struct tc_u_knode *)ht, extack); 695 ret = u32_delete_key(tp, (struct tc_u_knode *)ht); 696 goto out; 697 } 698 699 if (root_ht == ht) { 700 NL_SET_ERR_MSG_MOD(extack, "Not allowed to delete root node"); 701 return -EINVAL; 702 } 703 704 if (ht->refcnt == 1) { 705 ht->refcnt--; 706 u32_destroy_hnode(tp, ht, extack); 707 } else { 708 NL_SET_ERR_MSG_MOD(extack, "Can not delete in-use filter"); 709 return -EBUSY; 710 } 711 712 out: 713 *last = true; 714 if (root_ht) { 715 if (root_ht->refcnt > 1) { 716 *last = false; 717 goto ret; 718 } 719 if (root_ht->refcnt == 1) { 720 if (!ht_empty(root_ht)) { 721 *last = false; 722 goto ret; 723 } 724 } 725 } 726 727 if (tp_c->refcnt > 1) { 728 *last = false; 729 goto ret; 730 } 731 732 if (tp_c->refcnt == 1) { 733 struct tc_u_hnode *ht; 734 735 for (ht = rtnl_dereference(tp_c->hlist); 736 ht; 737 ht = rtnl_dereference(ht->next)) 738 if (!ht_empty(ht)) { 739 *last = false; 740 break; 741 } 742 } 743 744 ret: 745 return ret; 746 } 747 748 static u32 gen_new_kid(struct tc_u_hnode *ht, u32 htid) 749 { 750 unsigned long idr_index; 751 u32 start = htid | 0x800; 752 u32 max = htid | 0xFFF; 753 u32 min = htid; 754 755 if (idr_alloc_ext(&ht->handle_idr, NULL, &idr_index, 756 start, max + 1, GFP_KERNEL)) { 757 if (idr_alloc_ext(&ht->handle_idr, NULL, &idr_index, 758 min + 1, max + 1, GFP_KERNEL)) 759 return max; 760 } 761 762 return (u32)idr_index; 763 } 764 765 static const struct nla_policy u32_policy[TCA_U32_MAX + 1] = { 766 [TCA_U32_CLASSID] = { .type = NLA_U32 }, 767 [TCA_U32_HASH] = { .type = NLA_U32 }, 768 [TCA_U32_LINK] = { .type = NLA_U32 }, 769 [TCA_U32_DIVISOR] = { .type = NLA_U32 }, 770 [TCA_U32_SEL] = { .len = sizeof(struct tc_u32_sel) }, 771 [TCA_U32_INDEV] = { .type = NLA_STRING, .len = IFNAMSIZ }, 772 [TCA_U32_MARK] = { .len = sizeof(struct tc_u32_mark) }, 773 [TCA_U32_FLAGS] = { .type = NLA_U32 }, 774 }; 775 776 static int u32_set_parms(struct net *net, struct tcf_proto *tp, 777 unsigned long base, struct tc_u_hnode *ht, 778 struct tc_u_knode *n, struct nlattr **tb, 779 struct nlattr *est, bool ovr, 780 struct netlink_ext_ack *extack) 781 { 782 int err; 783 784 err = tcf_exts_validate(net, tp, tb, est, &n->exts, ovr, extack); 785 if (err < 0) 786 return err; 787 788 if (tb[TCA_U32_LINK]) { 789 u32 handle = nla_get_u32(tb[TCA_U32_LINK]); 790 struct tc_u_hnode *ht_down = NULL, *ht_old; 791 792 if (TC_U32_KEY(handle)) { 793 NL_SET_ERR_MSG_MOD(extack, "u32 Link handle must be a hash table"); 794 return -EINVAL; 795 } 796 797 if (handle) { 798 ht_down = u32_lookup_ht(ht->tp_c, handle); 799 800 if (!ht_down) { 801 NL_SET_ERR_MSG_MOD(extack, "Link hash table not found"); 802 return -EINVAL; 803 } 804 ht_down->refcnt++; 805 } 806 807 ht_old = rtnl_dereference(n->ht_down); 808 rcu_assign_pointer(n->ht_down, ht_down); 809 810 if (ht_old) 811 ht_old->refcnt--; 812 } 813 if (tb[TCA_U32_CLASSID]) { 814 n->res.classid = nla_get_u32(tb[TCA_U32_CLASSID]); 815 tcf_bind_filter(tp, &n->res, base); 816 } 817 818 #ifdef CONFIG_NET_CLS_IND 819 if (tb[TCA_U32_INDEV]) { 820 int ret; 821 ret = tcf_change_indev(net, tb[TCA_U32_INDEV], extack); 822 if (ret < 0) 823 return -EINVAL; 824 n->ifindex = ret; 825 } 826 #endif 827 return 0; 828 } 829 830 static void u32_replace_knode(struct tcf_proto *tp, struct tc_u_common *tp_c, 831 struct tc_u_knode *n) 832 { 833 struct tc_u_knode __rcu **ins; 834 struct tc_u_knode *pins; 835 struct tc_u_hnode *ht; 836 837 if (TC_U32_HTID(n->handle) == TC_U32_ROOT) 838 ht = rtnl_dereference(tp->root); 839 else 840 ht = u32_lookup_ht(tp_c, TC_U32_HTID(n->handle)); 841 842 ins = &ht->ht[TC_U32_HASH(n->handle)]; 843 844 /* The node must always exist for it to be replaced if this is not the 845 * case then something went very wrong elsewhere. 846 */ 847 for (pins = rtnl_dereference(*ins); ; 848 ins = &pins->next, pins = rtnl_dereference(*ins)) 849 if (pins->handle == n->handle) 850 break; 851 852 idr_replace_ext(&ht->handle_idr, n, n->handle); 853 RCU_INIT_POINTER(n->next, pins->next); 854 rcu_assign_pointer(*ins, n); 855 } 856 857 static struct tc_u_knode *u32_init_knode(struct tcf_proto *tp, 858 struct tc_u_knode *n) 859 { 860 struct tc_u_hnode *ht = rtnl_dereference(n->ht_down); 861 struct tc_u32_sel *s = &n->sel; 862 struct tc_u_knode *new; 863 864 new = kzalloc(sizeof(*n) + s->nkeys*sizeof(struct tc_u32_key), 865 GFP_KERNEL); 866 867 if (!new) 868 return NULL; 869 870 RCU_INIT_POINTER(new->next, n->next); 871 new->handle = n->handle; 872 RCU_INIT_POINTER(new->ht_up, n->ht_up); 873 874 #ifdef CONFIG_NET_CLS_IND 875 new->ifindex = n->ifindex; 876 #endif 877 new->fshift = n->fshift; 878 new->res = n->res; 879 new->flags = n->flags; 880 RCU_INIT_POINTER(new->ht_down, ht); 881 882 /* bump reference count as long as we hold pointer to structure */ 883 if (ht) 884 ht->refcnt++; 885 886 #ifdef CONFIG_CLS_U32_PERF 887 /* Statistics may be incremented by readers during update 888 * so we must keep them in tact. When the node is later destroyed 889 * a special destroy call must be made to not free the pf memory. 890 */ 891 new->pf = n->pf; 892 #endif 893 894 #ifdef CONFIG_CLS_U32_MARK 895 new->val = n->val; 896 new->mask = n->mask; 897 /* Similarly success statistics must be moved as pointers */ 898 new->pcpu_success = n->pcpu_success; 899 #endif 900 new->tp = tp; 901 memcpy(&new->sel, s, sizeof(*s) + s->nkeys*sizeof(struct tc_u32_key)); 902 903 if (tcf_exts_init(&new->exts, TCA_U32_ACT, TCA_U32_POLICE)) { 904 kfree(new); 905 return NULL; 906 } 907 908 return new; 909 } 910 911 static int u32_change(struct net *net, struct sk_buff *in_skb, 912 struct tcf_proto *tp, unsigned long base, u32 handle, 913 struct nlattr **tca, void **arg, bool ovr, 914 struct netlink_ext_ack *extack) 915 { 916 struct tc_u_common *tp_c = tp->data; 917 struct tc_u_hnode *ht; 918 struct tc_u_knode *n; 919 struct tc_u32_sel *s; 920 struct nlattr *opt = tca[TCA_OPTIONS]; 921 struct nlattr *tb[TCA_U32_MAX + 1]; 922 u32 htid, flags = 0; 923 int err; 924 #ifdef CONFIG_CLS_U32_PERF 925 size_t size; 926 #endif 927 928 if (!opt) { 929 if (handle) { 930 NL_SET_ERR_MSG_MOD(extack, "Filter handle requires options"); 931 return -EINVAL; 932 } else { 933 return 0; 934 } 935 } 936 937 err = nla_parse_nested(tb, TCA_U32_MAX, opt, u32_policy, extack); 938 if (err < 0) 939 return err; 940 941 if (tb[TCA_U32_FLAGS]) { 942 flags = nla_get_u32(tb[TCA_U32_FLAGS]); 943 if (!tc_flags_valid(flags)) { 944 NL_SET_ERR_MSG_MOD(extack, "Invalid filter flags"); 945 return -EINVAL; 946 } 947 } 948 949 n = *arg; 950 if (n) { 951 struct tc_u_knode *new; 952 953 if (TC_U32_KEY(n->handle) == 0) { 954 NL_SET_ERR_MSG_MOD(extack, "Key node id cannot be zero"); 955 return -EINVAL; 956 } 957 958 if (n->flags != flags) { 959 NL_SET_ERR_MSG_MOD(extack, "Key node flags do not match passed flags"); 960 return -EINVAL; 961 } 962 963 new = u32_init_knode(tp, n); 964 if (!new) 965 return -ENOMEM; 966 967 err = u32_set_parms(net, tp, base, 968 rtnl_dereference(n->ht_up), new, tb, 969 tca[TCA_RATE], ovr, extack); 970 971 if (err) { 972 u32_destroy_key(tp, new, false); 973 return err; 974 } 975 976 err = u32_replace_hw_knode(tp, new, flags, extack); 977 if (err) { 978 u32_destroy_key(tp, new, false); 979 return err; 980 } 981 982 if (!tc_in_hw(new->flags)) 983 new->flags |= TCA_CLS_FLAGS_NOT_IN_HW; 984 985 u32_replace_knode(tp, tp_c, new); 986 tcf_unbind_filter(tp, &n->res); 987 tcf_exts_get_net(&n->exts); 988 call_rcu(&n->rcu, u32_delete_key_rcu); 989 return 0; 990 } 991 992 if (tb[TCA_U32_DIVISOR]) { 993 unsigned int divisor = nla_get_u32(tb[TCA_U32_DIVISOR]); 994 995 if (--divisor > 0x100) { 996 NL_SET_ERR_MSG_MOD(extack, "Exceeded maximum 256 hash buckets"); 997 return -EINVAL; 998 } 999 if (TC_U32_KEY(handle)) { 1000 NL_SET_ERR_MSG_MOD(extack, "Divisor can only be used on a hash table"); 1001 return -EINVAL; 1002 } 1003 ht = kzalloc(sizeof(*ht) + divisor*sizeof(void *), GFP_KERNEL); 1004 if (ht == NULL) 1005 return -ENOBUFS; 1006 if (handle == 0) { 1007 handle = gen_new_htid(tp->data, ht); 1008 if (handle == 0) { 1009 kfree(ht); 1010 return -ENOMEM; 1011 } 1012 } else { 1013 err = idr_alloc_ext(&tp_c->handle_idr, ht, NULL, 1014 handle, handle + 1, GFP_KERNEL); 1015 if (err) { 1016 kfree(ht); 1017 return err; 1018 } 1019 } 1020 ht->tp_c = tp_c; 1021 ht->refcnt = 1; 1022 ht->divisor = divisor; 1023 ht->handle = handle; 1024 ht->prio = tp->prio; 1025 idr_init(&ht->handle_idr); 1026 ht->flags = flags; 1027 1028 err = u32_replace_hw_hnode(tp, ht, flags, extack); 1029 if (err) { 1030 idr_remove_ext(&tp_c->handle_idr, handle); 1031 kfree(ht); 1032 return err; 1033 } 1034 1035 RCU_INIT_POINTER(ht->next, tp_c->hlist); 1036 rcu_assign_pointer(tp_c->hlist, ht); 1037 *arg = ht; 1038 1039 return 0; 1040 } 1041 1042 if (tb[TCA_U32_HASH]) { 1043 htid = nla_get_u32(tb[TCA_U32_HASH]); 1044 if (TC_U32_HTID(htid) == TC_U32_ROOT) { 1045 ht = rtnl_dereference(tp->root); 1046 htid = ht->handle; 1047 } else { 1048 ht = u32_lookup_ht(tp->data, TC_U32_HTID(htid)); 1049 if (!ht) { 1050 NL_SET_ERR_MSG_MOD(extack, "Specified hash table not found"); 1051 return -EINVAL; 1052 } 1053 } 1054 } else { 1055 ht = rtnl_dereference(tp->root); 1056 htid = ht->handle; 1057 } 1058 1059 if (ht->divisor < TC_U32_HASH(htid)) { 1060 NL_SET_ERR_MSG_MOD(extack, "Specified hash table buckets exceed configured value"); 1061 return -EINVAL; 1062 } 1063 1064 if (handle) { 1065 if (TC_U32_HTID(handle) && TC_U32_HTID(handle ^ htid)) { 1066 NL_SET_ERR_MSG_MOD(extack, "Handle specified hash table address mismatch"); 1067 return -EINVAL; 1068 } 1069 handle = htid | TC_U32_NODE(handle); 1070 err = idr_alloc_ext(&ht->handle_idr, NULL, NULL, 1071 handle, handle + 1, 1072 GFP_KERNEL); 1073 if (err) 1074 return err; 1075 } else 1076 handle = gen_new_kid(ht, htid); 1077 1078 if (tb[TCA_U32_SEL] == NULL) { 1079 NL_SET_ERR_MSG_MOD(extack, "Selector not specified"); 1080 err = -EINVAL; 1081 goto erridr; 1082 } 1083 1084 s = nla_data(tb[TCA_U32_SEL]); 1085 1086 n = kzalloc(sizeof(*n) + s->nkeys*sizeof(struct tc_u32_key), GFP_KERNEL); 1087 if (n == NULL) { 1088 err = -ENOBUFS; 1089 goto erridr; 1090 } 1091 1092 #ifdef CONFIG_CLS_U32_PERF 1093 size = sizeof(struct tc_u32_pcnt) + s->nkeys * sizeof(u64); 1094 n->pf = __alloc_percpu(size, __alignof__(struct tc_u32_pcnt)); 1095 if (!n->pf) { 1096 err = -ENOBUFS; 1097 goto errfree; 1098 } 1099 #endif 1100 1101 memcpy(&n->sel, s, sizeof(*s) + s->nkeys*sizeof(struct tc_u32_key)); 1102 RCU_INIT_POINTER(n->ht_up, ht); 1103 n->handle = handle; 1104 n->fshift = s->hmask ? ffs(ntohl(s->hmask)) - 1 : 0; 1105 n->flags = flags; 1106 n->tp = tp; 1107 1108 err = tcf_exts_init(&n->exts, TCA_U32_ACT, TCA_U32_POLICE); 1109 if (err < 0) 1110 goto errout; 1111 1112 #ifdef CONFIG_CLS_U32_MARK 1113 n->pcpu_success = alloc_percpu(u32); 1114 if (!n->pcpu_success) { 1115 err = -ENOMEM; 1116 goto errout; 1117 } 1118 1119 if (tb[TCA_U32_MARK]) { 1120 struct tc_u32_mark *mark; 1121 1122 mark = nla_data(tb[TCA_U32_MARK]); 1123 n->val = mark->val; 1124 n->mask = mark->mask; 1125 } 1126 #endif 1127 1128 err = u32_set_parms(net, tp, base, ht, n, tb, tca[TCA_RATE], ovr, 1129 extack); 1130 if (err == 0) { 1131 struct tc_u_knode __rcu **ins; 1132 struct tc_u_knode *pins; 1133 1134 err = u32_replace_hw_knode(tp, n, flags, extack); 1135 if (err) 1136 goto errhw; 1137 1138 if (!tc_in_hw(n->flags)) 1139 n->flags |= TCA_CLS_FLAGS_NOT_IN_HW; 1140 1141 ins = &ht->ht[TC_U32_HASH(handle)]; 1142 for (pins = rtnl_dereference(*ins); pins; 1143 ins = &pins->next, pins = rtnl_dereference(*ins)) 1144 if (TC_U32_NODE(handle) < TC_U32_NODE(pins->handle)) 1145 break; 1146 1147 RCU_INIT_POINTER(n->next, pins); 1148 rcu_assign_pointer(*ins, n); 1149 *arg = n; 1150 return 0; 1151 } 1152 1153 errhw: 1154 #ifdef CONFIG_CLS_U32_MARK 1155 free_percpu(n->pcpu_success); 1156 #endif 1157 1158 errout: 1159 tcf_exts_destroy(&n->exts); 1160 #ifdef CONFIG_CLS_U32_PERF 1161 errfree: 1162 free_percpu(n->pf); 1163 #endif 1164 kfree(n); 1165 erridr: 1166 idr_remove_ext(&ht->handle_idr, handle); 1167 return err; 1168 } 1169 1170 static void u32_walk(struct tcf_proto *tp, struct tcf_walker *arg) 1171 { 1172 struct tc_u_common *tp_c = tp->data; 1173 struct tc_u_hnode *ht; 1174 struct tc_u_knode *n; 1175 unsigned int h; 1176 1177 if (arg->stop) 1178 return; 1179 1180 for (ht = rtnl_dereference(tp_c->hlist); 1181 ht; 1182 ht = rtnl_dereference(ht->next)) { 1183 if (ht->prio != tp->prio) 1184 continue; 1185 if (arg->count >= arg->skip) { 1186 if (arg->fn(tp, ht, arg) < 0) { 1187 arg->stop = 1; 1188 return; 1189 } 1190 } 1191 arg->count++; 1192 for (h = 0; h <= ht->divisor; h++) { 1193 for (n = rtnl_dereference(ht->ht[h]); 1194 n; 1195 n = rtnl_dereference(n->next)) { 1196 if (arg->count < arg->skip) { 1197 arg->count++; 1198 continue; 1199 } 1200 if (arg->fn(tp, n, arg) < 0) { 1201 arg->stop = 1; 1202 return; 1203 } 1204 arg->count++; 1205 } 1206 } 1207 } 1208 } 1209 1210 static void u32_bind_class(void *fh, u32 classid, unsigned long cl) 1211 { 1212 struct tc_u_knode *n = fh; 1213 1214 if (n && n->res.classid == classid) 1215 n->res.class = cl; 1216 } 1217 1218 static int u32_dump(struct net *net, struct tcf_proto *tp, void *fh, 1219 struct sk_buff *skb, struct tcmsg *t) 1220 { 1221 struct tc_u_knode *n = fh; 1222 struct tc_u_hnode *ht_up, *ht_down; 1223 struct nlattr *nest; 1224 1225 if (n == NULL) 1226 return skb->len; 1227 1228 t->tcm_handle = n->handle; 1229 1230 nest = nla_nest_start(skb, TCA_OPTIONS); 1231 if (nest == NULL) 1232 goto nla_put_failure; 1233 1234 if (TC_U32_KEY(n->handle) == 0) { 1235 struct tc_u_hnode *ht = fh; 1236 u32 divisor = ht->divisor + 1; 1237 1238 if (nla_put_u32(skb, TCA_U32_DIVISOR, divisor)) 1239 goto nla_put_failure; 1240 } else { 1241 #ifdef CONFIG_CLS_U32_PERF 1242 struct tc_u32_pcnt *gpf; 1243 int cpu; 1244 #endif 1245 1246 if (nla_put(skb, TCA_U32_SEL, 1247 sizeof(n->sel) + n->sel.nkeys*sizeof(struct tc_u32_key), 1248 &n->sel)) 1249 goto nla_put_failure; 1250 1251 ht_up = rtnl_dereference(n->ht_up); 1252 if (ht_up) { 1253 u32 htid = n->handle & 0xFFFFF000; 1254 if (nla_put_u32(skb, TCA_U32_HASH, htid)) 1255 goto nla_put_failure; 1256 } 1257 if (n->res.classid && 1258 nla_put_u32(skb, TCA_U32_CLASSID, n->res.classid)) 1259 goto nla_put_failure; 1260 1261 ht_down = rtnl_dereference(n->ht_down); 1262 if (ht_down && 1263 nla_put_u32(skb, TCA_U32_LINK, ht_down->handle)) 1264 goto nla_put_failure; 1265 1266 if (n->flags && nla_put_u32(skb, TCA_U32_FLAGS, n->flags)) 1267 goto nla_put_failure; 1268 1269 #ifdef CONFIG_CLS_U32_MARK 1270 if ((n->val || n->mask)) { 1271 struct tc_u32_mark mark = {.val = n->val, 1272 .mask = n->mask, 1273 .success = 0}; 1274 int cpum; 1275 1276 for_each_possible_cpu(cpum) { 1277 __u32 cnt = *per_cpu_ptr(n->pcpu_success, cpum); 1278 1279 mark.success += cnt; 1280 } 1281 1282 if (nla_put(skb, TCA_U32_MARK, sizeof(mark), &mark)) 1283 goto nla_put_failure; 1284 } 1285 #endif 1286 1287 if (tcf_exts_dump(skb, &n->exts) < 0) 1288 goto nla_put_failure; 1289 1290 #ifdef CONFIG_NET_CLS_IND 1291 if (n->ifindex) { 1292 struct net_device *dev; 1293 dev = __dev_get_by_index(net, n->ifindex); 1294 if (dev && nla_put_string(skb, TCA_U32_INDEV, dev->name)) 1295 goto nla_put_failure; 1296 } 1297 #endif 1298 #ifdef CONFIG_CLS_U32_PERF 1299 gpf = kzalloc(sizeof(struct tc_u32_pcnt) + 1300 n->sel.nkeys * sizeof(u64), 1301 GFP_KERNEL); 1302 if (!gpf) 1303 goto nla_put_failure; 1304 1305 for_each_possible_cpu(cpu) { 1306 int i; 1307 struct tc_u32_pcnt *pf = per_cpu_ptr(n->pf, cpu); 1308 1309 gpf->rcnt += pf->rcnt; 1310 gpf->rhit += pf->rhit; 1311 for (i = 0; i < n->sel.nkeys; i++) 1312 gpf->kcnts[i] += pf->kcnts[i]; 1313 } 1314 1315 if (nla_put_64bit(skb, TCA_U32_PCNT, 1316 sizeof(struct tc_u32_pcnt) + 1317 n->sel.nkeys * sizeof(u64), 1318 gpf, TCA_U32_PAD)) { 1319 kfree(gpf); 1320 goto nla_put_failure; 1321 } 1322 kfree(gpf); 1323 #endif 1324 } 1325 1326 nla_nest_end(skb, nest); 1327 1328 if (TC_U32_KEY(n->handle)) 1329 if (tcf_exts_dump_stats(skb, &n->exts) < 0) 1330 goto nla_put_failure; 1331 return skb->len; 1332 1333 nla_put_failure: 1334 nla_nest_cancel(skb, nest); 1335 return -1; 1336 } 1337 1338 static struct tcf_proto_ops cls_u32_ops __read_mostly = { 1339 .kind = "u32", 1340 .classify = u32_classify, 1341 .init = u32_init, 1342 .destroy = u32_destroy, 1343 .get = u32_get, 1344 .change = u32_change, 1345 .delete = u32_delete, 1346 .walk = u32_walk, 1347 .dump = u32_dump, 1348 .bind_class = u32_bind_class, 1349 .owner = THIS_MODULE, 1350 }; 1351 1352 static int __init init_u32(void) 1353 { 1354 int i, ret; 1355 1356 pr_info("u32 classifier\n"); 1357 #ifdef CONFIG_CLS_U32_PERF 1358 pr_info(" Performance counters on\n"); 1359 #endif 1360 #ifdef CONFIG_NET_CLS_IND 1361 pr_info(" input device check on\n"); 1362 #endif 1363 #ifdef CONFIG_NET_CLS_ACT 1364 pr_info(" Actions configured\n"); 1365 #endif 1366 tc_u_common_hash = kvmalloc_array(U32_HASH_SIZE, 1367 sizeof(struct hlist_head), 1368 GFP_KERNEL); 1369 if (!tc_u_common_hash) 1370 return -ENOMEM; 1371 1372 for (i = 0; i < U32_HASH_SIZE; i++) 1373 INIT_HLIST_HEAD(&tc_u_common_hash[i]); 1374 1375 ret = register_tcf_proto_ops(&cls_u32_ops); 1376 if (ret) 1377 kvfree(tc_u_common_hash); 1378 return ret; 1379 } 1380 1381 static void __exit exit_u32(void) 1382 { 1383 unregister_tcf_proto_ops(&cls_u32_ops); 1384 kvfree(tc_u_common_hash); 1385 } 1386 1387 module_init(init_u32) 1388 module_exit(exit_u32) 1389 MODULE_LICENSE("GPL"); 1390