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 struct rcu_work rwork; 72 /* The 'sel' field MUST be the last field in structure to allow for 73 * tc_u32_keys allocated at end of structure. 74 */ 75 struct tc_u32_sel sel; 76 }; 77 78 struct tc_u_hnode { 79 struct tc_u_hnode __rcu *next; 80 u32 handle; 81 u32 prio; 82 struct tc_u_common *tp_c; 83 int refcnt; 84 unsigned int divisor; 85 struct idr handle_idr; 86 struct rcu_head rcu; 87 u32 flags; 88 /* The 'ht' field MUST be the last field in structure to allow for 89 * more entries allocated at end of structure. 90 */ 91 struct tc_u_knode __rcu *ht[1]; 92 }; 93 94 struct tc_u_common { 95 struct tc_u_hnode __rcu *hlist; 96 void *ptr; 97 int refcnt; 98 struct idr handle_idr; 99 struct hlist_node hnode; 100 struct rcu_head rcu; 101 }; 102 103 static inline unsigned int u32_hash_fold(__be32 key, 104 const struct tc_u32_sel *sel, 105 u8 fshift) 106 { 107 unsigned int h = ntohl(key & sel->hmask) >> fshift; 108 109 return h; 110 } 111 112 static int u32_classify(struct sk_buff *skb, const struct tcf_proto *tp, 113 struct tcf_result *res) 114 { 115 struct { 116 struct tc_u_knode *knode; 117 unsigned int off; 118 } stack[TC_U32_MAXDEPTH]; 119 120 struct tc_u_hnode *ht = rcu_dereference_bh(tp->root); 121 unsigned int off = skb_network_offset(skb); 122 struct tc_u_knode *n; 123 int sdepth = 0; 124 int off2 = 0; 125 int sel = 0; 126 #ifdef CONFIG_CLS_U32_PERF 127 int j; 128 #endif 129 int i, r; 130 131 next_ht: 132 n = rcu_dereference_bh(ht->ht[sel]); 133 134 next_knode: 135 if (n) { 136 struct tc_u32_key *key = n->sel.keys; 137 138 #ifdef CONFIG_CLS_U32_PERF 139 __this_cpu_inc(n->pf->rcnt); 140 j = 0; 141 #endif 142 143 if (tc_skip_sw(n->flags)) { 144 n = rcu_dereference_bh(n->next); 145 goto next_knode; 146 } 147 148 #ifdef CONFIG_CLS_U32_MARK 149 if ((skb->mark & n->mask) != n->val) { 150 n = rcu_dereference_bh(n->next); 151 goto next_knode; 152 } else { 153 __this_cpu_inc(*n->pcpu_success); 154 } 155 #endif 156 157 for (i = n->sel.nkeys; i > 0; i--, key++) { 158 int toff = off + key->off + (off2 & key->offmask); 159 __be32 *data, hdata; 160 161 if (skb_headroom(skb) + toff > INT_MAX) 162 goto out; 163 164 data = skb_header_pointer(skb, toff, 4, &hdata); 165 if (!data) 166 goto out; 167 if ((*data ^ key->val) & key->mask) { 168 n = rcu_dereference_bh(n->next); 169 goto next_knode; 170 } 171 #ifdef CONFIG_CLS_U32_PERF 172 __this_cpu_inc(n->pf->kcnts[j]); 173 j++; 174 #endif 175 } 176 177 ht = rcu_dereference_bh(n->ht_down); 178 if (!ht) { 179 check_terminal: 180 if (n->sel.flags & TC_U32_TERMINAL) { 181 182 *res = n->res; 183 #ifdef CONFIG_NET_CLS_IND 184 if (!tcf_match_indev(skb, n->ifindex)) { 185 n = rcu_dereference_bh(n->next); 186 goto next_knode; 187 } 188 #endif 189 #ifdef CONFIG_CLS_U32_PERF 190 __this_cpu_inc(n->pf->rhit); 191 #endif 192 r = tcf_exts_exec(skb, &n->exts, res); 193 if (r < 0) { 194 n = rcu_dereference_bh(n->next); 195 goto next_knode; 196 } 197 198 return r; 199 } 200 n = rcu_dereference_bh(n->next); 201 goto next_knode; 202 } 203 204 /* PUSH */ 205 if (sdepth >= TC_U32_MAXDEPTH) 206 goto deadloop; 207 stack[sdepth].knode = n; 208 stack[sdepth].off = off; 209 sdepth++; 210 211 ht = rcu_dereference_bh(n->ht_down); 212 sel = 0; 213 if (ht->divisor) { 214 __be32 *data, hdata; 215 216 data = skb_header_pointer(skb, off + n->sel.hoff, 4, 217 &hdata); 218 if (!data) 219 goto out; 220 sel = ht->divisor & u32_hash_fold(*data, &n->sel, 221 n->fshift); 222 } 223 if (!(n->sel.flags & (TC_U32_VAROFFSET | TC_U32_OFFSET | TC_U32_EAT))) 224 goto next_ht; 225 226 if (n->sel.flags & (TC_U32_OFFSET | TC_U32_VAROFFSET)) { 227 off2 = n->sel.off + 3; 228 if (n->sel.flags & TC_U32_VAROFFSET) { 229 __be16 *data, hdata; 230 231 data = skb_header_pointer(skb, 232 off + n->sel.offoff, 233 2, &hdata); 234 if (!data) 235 goto out; 236 off2 += ntohs(n->sel.offmask & *data) >> 237 n->sel.offshift; 238 } 239 off2 &= ~3; 240 } 241 if (n->sel.flags & TC_U32_EAT) { 242 off += off2; 243 off2 = 0; 244 } 245 246 if (off < skb->len) 247 goto next_ht; 248 } 249 250 /* POP */ 251 if (sdepth--) { 252 n = stack[sdepth].knode; 253 ht = rcu_dereference_bh(n->ht_up); 254 off = stack[sdepth].off; 255 goto check_terminal; 256 } 257 out: 258 return -1; 259 260 deadloop: 261 net_warn_ratelimited("cls_u32: dead loop\n"); 262 return -1; 263 } 264 265 static struct tc_u_hnode *u32_lookup_ht(struct tc_u_common *tp_c, u32 handle) 266 { 267 struct tc_u_hnode *ht; 268 269 for (ht = rtnl_dereference(tp_c->hlist); 270 ht; 271 ht = rtnl_dereference(ht->next)) 272 if (ht->handle == handle) 273 break; 274 275 return ht; 276 } 277 278 static struct tc_u_knode *u32_lookup_key(struct tc_u_hnode *ht, u32 handle) 279 { 280 unsigned int sel; 281 struct tc_u_knode *n = NULL; 282 283 sel = TC_U32_HASH(handle); 284 if (sel > ht->divisor) 285 goto out; 286 287 for (n = rtnl_dereference(ht->ht[sel]); 288 n; 289 n = rtnl_dereference(n->next)) 290 if (n->handle == handle) 291 break; 292 out: 293 return n; 294 } 295 296 297 static void *u32_get(struct tcf_proto *tp, u32 handle) 298 { 299 struct tc_u_hnode *ht; 300 struct tc_u_common *tp_c = tp->data; 301 302 if (TC_U32_HTID(handle) == TC_U32_ROOT) 303 ht = rtnl_dereference(tp->root); 304 else 305 ht = u32_lookup_ht(tp_c, TC_U32_HTID(handle)); 306 307 if (!ht) 308 return NULL; 309 310 if (TC_U32_KEY(handle) == 0) 311 return ht; 312 313 return u32_lookup_key(ht, handle); 314 } 315 316 /* Protected by rtnl lock */ 317 static u32 gen_new_htid(struct tc_u_common *tp_c, struct tc_u_hnode *ptr) 318 { 319 int id = idr_alloc_cyclic(&tp_c->handle_idr, ptr, 1, 0x7FF, GFP_KERNEL); 320 if (id < 0) 321 return 0; 322 return (id | 0x800U) << 20; 323 } 324 325 static struct hlist_head *tc_u_common_hash; 326 327 #define U32_HASH_SHIFT 10 328 #define U32_HASH_SIZE (1 << U32_HASH_SHIFT) 329 330 static void *tc_u_common_ptr(const struct tcf_proto *tp) 331 { 332 struct tcf_block *block = tp->chain->block; 333 334 /* The block sharing is currently supported only 335 * for classless qdiscs. In that case we use block 336 * for tc_u_common identification. In case the 337 * block is not shared, block->q is a valid pointer 338 * and we can use that. That works for classful qdiscs. 339 */ 340 if (tcf_block_shared(block)) 341 return block; 342 else 343 return block->q; 344 } 345 346 static unsigned int tc_u_hash(const struct tcf_proto *tp) 347 { 348 return hash_ptr(tc_u_common_ptr(tp), U32_HASH_SHIFT); 349 } 350 351 static struct tc_u_common *tc_u_common_find(const struct tcf_proto *tp) 352 { 353 struct tc_u_common *tc; 354 unsigned int h; 355 356 h = tc_u_hash(tp); 357 hlist_for_each_entry(tc, &tc_u_common_hash[h], hnode) { 358 if (tc->ptr == tc_u_common_ptr(tp)) 359 return tc; 360 } 361 return NULL; 362 } 363 364 static int u32_init(struct tcf_proto *tp) 365 { 366 struct tc_u_hnode *root_ht; 367 struct tc_u_common *tp_c; 368 unsigned int h; 369 370 tp_c = tc_u_common_find(tp); 371 372 root_ht = kzalloc(sizeof(*root_ht), GFP_KERNEL); 373 if (root_ht == NULL) 374 return -ENOBUFS; 375 376 root_ht->refcnt++; 377 root_ht->handle = tp_c ? gen_new_htid(tp_c, root_ht) : 0x80000000; 378 root_ht->prio = tp->prio; 379 idr_init(&root_ht->handle_idr); 380 381 if (tp_c == NULL) { 382 tp_c = kzalloc(sizeof(*tp_c), GFP_KERNEL); 383 if (tp_c == NULL) { 384 kfree(root_ht); 385 return -ENOBUFS; 386 } 387 tp_c->ptr = tc_u_common_ptr(tp); 388 INIT_HLIST_NODE(&tp_c->hnode); 389 idr_init(&tp_c->handle_idr); 390 391 h = tc_u_hash(tp); 392 hlist_add_head(&tp_c->hnode, &tc_u_common_hash[h]); 393 } 394 395 tp_c->refcnt++; 396 RCU_INIT_POINTER(root_ht->next, tp_c->hlist); 397 rcu_assign_pointer(tp_c->hlist, root_ht); 398 root_ht->tp_c = tp_c; 399 400 rcu_assign_pointer(tp->root, root_ht); 401 tp->data = tp_c; 402 return 0; 403 } 404 405 static int u32_destroy_key(struct tcf_proto *tp, struct tc_u_knode *n, 406 bool free_pf) 407 { 408 struct tc_u_hnode *ht = rtnl_dereference(n->ht_down); 409 410 tcf_exts_destroy(&n->exts); 411 tcf_exts_put_net(&n->exts); 412 if (ht && --ht->refcnt == 0) 413 kfree(ht); 414 #ifdef CONFIG_CLS_U32_PERF 415 if (free_pf) 416 free_percpu(n->pf); 417 #endif 418 #ifdef CONFIG_CLS_U32_MARK 419 if (free_pf) 420 free_percpu(n->pcpu_success); 421 #endif 422 kfree(n); 423 return 0; 424 } 425 426 /* u32_delete_key_rcu should be called when free'ing a copied 427 * version of a tc_u_knode obtained from u32_init_knode(). When 428 * copies are obtained from u32_init_knode() the statistics are 429 * shared between the old and new copies to allow readers to 430 * continue to update the statistics during the copy. To support 431 * this the u32_delete_key_rcu variant does not free the percpu 432 * statistics. 433 */ 434 static void u32_delete_key_work(struct work_struct *work) 435 { 436 struct tc_u_knode *key = container_of(to_rcu_work(work), 437 struct tc_u_knode, 438 rwork); 439 rtnl_lock(); 440 u32_destroy_key(key->tp, key, false); 441 rtnl_unlock(); 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(to_rcu_work(work), 454 struct tc_u_knode, 455 rwork); 456 rtnl_lock(); 457 u32_destroy_key(key->tp, key, true); 458 rtnl_unlock(); 459 } 460 461 static int u32_delete_key(struct tcf_proto *tp, struct tc_u_knode *key) 462 { 463 struct tc_u_knode __rcu **kp; 464 struct tc_u_knode *pkp; 465 struct tc_u_hnode *ht = rtnl_dereference(key->ht_up); 466 467 if (ht) { 468 kp = &ht->ht[TC_U32_HASH(key->handle)]; 469 for (pkp = rtnl_dereference(*kp); pkp; 470 kp = &pkp->next, pkp = rtnl_dereference(*kp)) { 471 if (pkp == key) { 472 RCU_INIT_POINTER(*kp, key->next); 473 474 tcf_unbind_filter(tp, &key->res); 475 idr_remove(&ht->handle_idr, key->handle); 476 tcf_exts_get_net(&key->exts); 477 tcf_queue_work(&key->rwork, u32_delete_key_freepf_work); 478 return 0; 479 } 480 } 481 } 482 WARN_ON(1); 483 return 0; 484 } 485 486 static void u32_clear_hw_hnode(struct tcf_proto *tp, struct tc_u_hnode *h, 487 struct netlink_ext_ack *extack) 488 { 489 struct tcf_block *block = tp->chain->block; 490 struct tc_cls_u32_offload cls_u32 = {}; 491 492 tc_cls_common_offload_init(&cls_u32.common, tp, h->flags, extack); 493 cls_u32.command = TC_CLSU32_DELETE_HNODE; 494 cls_u32.hnode.divisor = h->divisor; 495 cls_u32.hnode.handle = h->handle; 496 cls_u32.hnode.prio = h->prio; 497 498 tc_setup_cb_call(block, NULL, TC_SETUP_CLSU32, &cls_u32, false); 499 } 500 501 static int u32_replace_hw_hnode(struct tcf_proto *tp, struct tc_u_hnode *h, 502 u32 flags, struct netlink_ext_ack *extack) 503 { 504 struct tcf_block *block = tp->chain->block; 505 struct tc_cls_u32_offload cls_u32 = {}; 506 bool skip_sw = tc_skip_sw(flags); 507 bool offloaded = false; 508 int err; 509 510 tc_cls_common_offload_init(&cls_u32.common, tp, flags, extack); 511 cls_u32.command = TC_CLSU32_NEW_HNODE; 512 cls_u32.hnode.divisor = h->divisor; 513 cls_u32.hnode.handle = h->handle; 514 cls_u32.hnode.prio = h->prio; 515 516 err = tc_setup_cb_call(block, NULL, TC_SETUP_CLSU32, &cls_u32, skip_sw); 517 if (err < 0) { 518 u32_clear_hw_hnode(tp, h, NULL); 519 return err; 520 } else if (err > 0) { 521 offloaded = true; 522 } 523 524 if (skip_sw && !offloaded) 525 return -EINVAL; 526 527 return 0; 528 } 529 530 static void u32_remove_hw_knode(struct tcf_proto *tp, struct tc_u_knode *n, 531 struct netlink_ext_ack *extack) 532 { 533 struct tcf_block *block = tp->chain->block; 534 struct tc_cls_u32_offload cls_u32 = {}; 535 536 tc_cls_common_offload_init(&cls_u32.common, tp, n->flags, extack); 537 cls_u32.command = TC_CLSU32_DELETE_KNODE; 538 cls_u32.knode.handle = n->handle; 539 540 tc_setup_cb_call(block, NULL, TC_SETUP_CLSU32, &cls_u32, false); 541 tcf_block_offload_dec(block, &n->flags); 542 } 543 544 static int u32_replace_hw_knode(struct tcf_proto *tp, struct tc_u_knode *n, 545 u32 flags, struct netlink_ext_ack *extack) 546 { 547 struct tc_u_hnode *ht = rtnl_dereference(n->ht_down); 548 struct tcf_block *block = tp->chain->block; 549 struct tc_cls_u32_offload cls_u32 = {}; 550 bool skip_sw = tc_skip_sw(flags); 551 int err; 552 553 tc_cls_common_offload_init(&cls_u32.common, tp, flags, extack); 554 cls_u32.command = TC_CLSU32_REPLACE_KNODE; 555 cls_u32.knode.handle = n->handle; 556 cls_u32.knode.fshift = n->fshift; 557 #ifdef CONFIG_CLS_U32_MARK 558 cls_u32.knode.val = n->val; 559 cls_u32.knode.mask = n->mask; 560 #else 561 cls_u32.knode.val = 0; 562 cls_u32.knode.mask = 0; 563 #endif 564 cls_u32.knode.sel = &n->sel; 565 cls_u32.knode.exts = &n->exts; 566 if (n->ht_down) 567 cls_u32.knode.link_handle = ht->handle; 568 569 err = tc_setup_cb_call(block, NULL, TC_SETUP_CLSU32, &cls_u32, skip_sw); 570 if (err < 0) { 571 u32_remove_hw_knode(tp, n, NULL); 572 return err; 573 } else if (err > 0) { 574 tcf_block_offload_inc(block, &n->flags); 575 } 576 577 if (skip_sw && !(n->flags & TCA_CLS_FLAGS_IN_HW)) 578 return -EINVAL; 579 580 return 0; 581 } 582 583 static void u32_clear_hnode(struct tcf_proto *tp, struct tc_u_hnode *ht, 584 struct netlink_ext_ack *extack) 585 { 586 struct tc_u_knode *n; 587 unsigned int h; 588 589 for (h = 0; h <= ht->divisor; h++) { 590 while ((n = rtnl_dereference(ht->ht[h])) != NULL) { 591 RCU_INIT_POINTER(ht->ht[h], 592 rtnl_dereference(n->next)); 593 tcf_unbind_filter(tp, &n->res); 594 u32_remove_hw_knode(tp, n, extack); 595 idr_remove(&ht->handle_idr, n->handle); 596 if (tcf_exts_get_net(&n->exts)) 597 tcf_queue_work(&n->rwork, u32_delete_key_freepf_work); 598 else 599 u32_destroy_key(n->tp, n, true); 600 } 601 } 602 } 603 604 static int u32_destroy_hnode(struct tcf_proto *tp, struct tc_u_hnode *ht, 605 struct netlink_ext_ack *extack) 606 { 607 struct tc_u_common *tp_c = tp->data; 608 struct tc_u_hnode __rcu **hn; 609 struct tc_u_hnode *phn; 610 611 WARN_ON(ht->refcnt); 612 613 u32_clear_hnode(tp, ht, extack); 614 615 hn = &tp_c->hlist; 616 for (phn = rtnl_dereference(*hn); 617 phn; 618 hn = &phn->next, phn = rtnl_dereference(*hn)) { 619 if (phn == ht) { 620 u32_clear_hw_hnode(tp, ht, extack); 621 idr_destroy(&ht->handle_idr); 622 idr_remove(&tp_c->handle_idr, ht->handle); 623 RCU_INIT_POINTER(*hn, ht->next); 624 kfree_rcu(ht, rcu); 625 return 0; 626 } 627 } 628 629 return -ENOENT; 630 } 631 632 static bool ht_empty(struct tc_u_hnode *ht) 633 { 634 unsigned int h; 635 636 for (h = 0; h <= ht->divisor; h++) 637 if (rcu_access_pointer(ht->ht[h])) 638 return false; 639 640 return true; 641 } 642 643 static void u32_destroy(struct tcf_proto *tp, struct netlink_ext_ack *extack) 644 { 645 struct tc_u_common *tp_c = tp->data; 646 struct tc_u_hnode *root_ht = rtnl_dereference(tp->root); 647 648 WARN_ON(root_ht == NULL); 649 650 if (root_ht && --root_ht->refcnt == 0) 651 u32_destroy_hnode(tp, root_ht, extack); 652 653 if (--tp_c->refcnt == 0) { 654 struct tc_u_hnode *ht; 655 656 hlist_del(&tp_c->hnode); 657 658 while ((ht = rtnl_dereference(tp_c->hlist)) != NULL) { 659 u32_clear_hnode(tp, ht, extack); 660 RCU_INIT_POINTER(tp_c->hlist, ht->next); 661 662 /* u32_destroy_key() will later free ht for us, if it's 663 * still referenced by some knode 664 */ 665 if (--ht->refcnt == 0) 666 kfree_rcu(ht, rcu); 667 } 668 669 idr_destroy(&tp_c->handle_idr); 670 kfree(tp_c); 671 } 672 673 tp->data = NULL; 674 } 675 676 static int u32_delete(struct tcf_proto *tp, void *arg, bool *last, 677 struct netlink_ext_ack *extack) 678 { 679 struct tc_u_hnode *ht = arg; 680 struct tc_u_hnode *root_ht = rtnl_dereference(tp->root); 681 struct tc_u_common *tp_c = tp->data; 682 int ret = 0; 683 684 if (ht == NULL) 685 goto out; 686 687 if (TC_U32_KEY(ht->handle)) { 688 u32_remove_hw_knode(tp, (struct tc_u_knode *)ht, extack); 689 ret = u32_delete_key(tp, (struct tc_u_knode *)ht); 690 goto out; 691 } 692 693 if (root_ht == ht) { 694 NL_SET_ERR_MSG_MOD(extack, "Not allowed to delete root node"); 695 return -EINVAL; 696 } 697 698 if (ht->refcnt == 1) { 699 ht->refcnt--; 700 u32_destroy_hnode(tp, ht, extack); 701 } else { 702 NL_SET_ERR_MSG_MOD(extack, "Can not delete in-use filter"); 703 return -EBUSY; 704 } 705 706 out: 707 *last = true; 708 if (root_ht) { 709 if (root_ht->refcnt > 1) { 710 *last = false; 711 goto ret; 712 } 713 if (root_ht->refcnt == 1) { 714 if (!ht_empty(root_ht)) { 715 *last = false; 716 goto ret; 717 } 718 } 719 } 720 721 if (tp_c->refcnt > 1) { 722 *last = false; 723 goto ret; 724 } 725 726 if (tp_c->refcnt == 1) { 727 struct tc_u_hnode *ht; 728 729 for (ht = rtnl_dereference(tp_c->hlist); 730 ht; 731 ht = rtnl_dereference(ht->next)) 732 if (!ht_empty(ht)) { 733 *last = false; 734 break; 735 } 736 } 737 738 ret: 739 return ret; 740 } 741 742 static u32 gen_new_kid(struct tc_u_hnode *ht, u32 htid) 743 { 744 u32 index = htid | 0x800; 745 u32 max = htid | 0xFFF; 746 747 if (idr_alloc_u32(&ht->handle_idr, NULL, &index, max, GFP_KERNEL)) { 748 index = htid + 1; 749 if (idr_alloc_u32(&ht->handle_idr, NULL, &index, max, 750 GFP_KERNEL)) 751 index = max; 752 } 753 754 return index; 755 } 756 757 static const struct nla_policy u32_policy[TCA_U32_MAX + 1] = { 758 [TCA_U32_CLASSID] = { .type = NLA_U32 }, 759 [TCA_U32_HASH] = { .type = NLA_U32 }, 760 [TCA_U32_LINK] = { .type = NLA_U32 }, 761 [TCA_U32_DIVISOR] = { .type = NLA_U32 }, 762 [TCA_U32_SEL] = { .len = sizeof(struct tc_u32_sel) }, 763 [TCA_U32_INDEV] = { .type = NLA_STRING, .len = IFNAMSIZ }, 764 [TCA_U32_MARK] = { .len = sizeof(struct tc_u32_mark) }, 765 [TCA_U32_FLAGS] = { .type = NLA_U32 }, 766 }; 767 768 static int u32_set_parms(struct net *net, struct tcf_proto *tp, 769 unsigned long base, struct tc_u_hnode *ht, 770 struct tc_u_knode *n, struct nlattr **tb, 771 struct nlattr *est, bool ovr, 772 struct netlink_ext_ack *extack) 773 { 774 int err; 775 776 err = tcf_exts_validate(net, tp, tb, est, &n->exts, ovr, extack); 777 if (err < 0) 778 return err; 779 780 if (tb[TCA_U32_LINK]) { 781 u32 handle = nla_get_u32(tb[TCA_U32_LINK]); 782 struct tc_u_hnode *ht_down = NULL, *ht_old; 783 784 if (TC_U32_KEY(handle)) { 785 NL_SET_ERR_MSG_MOD(extack, "u32 Link handle must be a hash table"); 786 return -EINVAL; 787 } 788 789 if (handle) { 790 ht_down = u32_lookup_ht(ht->tp_c, handle); 791 792 if (!ht_down) { 793 NL_SET_ERR_MSG_MOD(extack, "Link hash table not found"); 794 return -EINVAL; 795 } 796 ht_down->refcnt++; 797 } 798 799 ht_old = rtnl_dereference(n->ht_down); 800 rcu_assign_pointer(n->ht_down, ht_down); 801 802 if (ht_old) 803 ht_old->refcnt--; 804 } 805 if (tb[TCA_U32_CLASSID]) { 806 n->res.classid = nla_get_u32(tb[TCA_U32_CLASSID]); 807 tcf_bind_filter(tp, &n->res, base); 808 } 809 810 #ifdef CONFIG_NET_CLS_IND 811 if (tb[TCA_U32_INDEV]) { 812 int ret; 813 ret = tcf_change_indev(net, tb[TCA_U32_INDEV], extack); 814 if (ret < 0) 815 return -EINVAL; 816 n->ifindex = ret; 817 } 818 #endif 819 return 0; 820 } 821 822 static void u32_replace_knode(struct tcf_proto *tp, struct tc_u_common *tp_c, 823 struct tc_u_knode *n) 824 { 825 struct tc_u_knode __rcu **ins; 826 struct tc_u_knode *pins; 827 struct tc_u_hnode *ht; 828 829 if (TC_U32_HTID(n->handle) == TC_U32_ROOT) 830 ht = rtnl_dereference(tp->root); 831 else 832 ht = u32_lookup_ht(tp_c, TC_U32_HTID(n->handle)); 833 834 ins = &ht->ht[TC_U32_HASH(n->handle)]; 835 836 /* The node must always exist for it to be replaced if this is not the 837 * case then something went very wrong elsewhere. 838 */ 839 for (pins = rtnl_dereference(*ins); ; 840 ins = &pins->next, pins = rtnl_dereference(*ins)) 841 if (pins->handle == n->handle) 842 break; 843 844 idr_replace(&ht->handle_idr, n, n->handle); 845 RCU_INIT_POINTER(n->next, pins->next); 846 rcu_assign_pointer(*ins, n); 847 } 848 849 static struct tc_u_knode *u32_init_knode(struct tcf_proto *tp, 850 struct tc_u_knode *n) 851 { 852 struct tc_u_hnode *ht = rtnl_dereference(n->ht_down); 853 struct tc_u32_sel *s = &n->sel; 854 struct tc_u_knode *new; 855 856 new = kzalloc(sizeof(*n) + s->nkeys*sizeof(struct tc_u32_key), 857 GFP_KERNEL); 858 859 if (!new) 860 return NULL; 861 862 RCU_INIT_POINTER(new->next, n->next); 863 new->handle = n->handle; 864 RCU_INIT_POINTER(new->ht_up, n->ht_up); 865 866 #ifdef CONFIG_NET_CLS_IND 867 new->ifindex = n->ifindex; 868 #endif 869 new->fshift = n->fshift; 870 new->res = n->res; 871 new->flags = n->flags; 872 RCU_INIT_POINTER(new->ht_down, ht); 873 874 /* bump reference count as long as we hold pointer to structure */ 875 if (ht) 876 ht->refcnt++; 877 878 #ifdef CONFIG_CLS_U32_PERF 879 /* Statistics may be incremented by readers during update 880 * so we must keep them in tact. When the node is later destroyed 881 * a special destroy call must be made to not free the pf memory. 882 */ 883 new->pf = n->pf; 884 #endif 885 886 #ifdef CONFIG_CLS_U32_MARK 887 new->val = n->val; 888 new->mask = n->mask; 889 /* Similarly success statistics must be moved as pointers */ 890 new->pcpu_success = n->pcpu_success; 891 #endif 892 new->tp = tp; 893 memcpy(&new->sel, s, sizeof(*s) + s->nkeys*sizeof(struct tc_u32_key)); 894 895 if (tcf_exts_init(&new->exts, TCA_U32_ACT, TCA_U32_POLICE)) { 896 kfree(new); 897 return NULL; 898 } 899 900 return new; 901 } 902 903 static int u32_change(struct net *net, struct sk_buff *in_skb, 904 struct tcf_proto *tp, unsigned long base, u32 handle, 905 struct nlattr **tca, void **arg, bool ovr, 906 struct netlink_ext_ack *extack) 907 { 908 struct tc_u_common *tp_c = tp->data; 909 struct tc_u_hnode *ht; 910 struct tc_u_knode *n; 911 struct tc_u32_sel *s; 912 struct nlattr *opt = tca[TCA_OPTIONS]; 913 struct nlattr *tb[TCA_U32_MAX + 1]; 914 u32 htid, flags = 0; 915 int err; 916 #ifdef CONFIG_CLS_U32_PERF 917 size_t size; 918 #endif 919 920 if (!opt) { 921 if (handle) { 922 NL_SET_ERR_MSG_MOD(extack, "Filter handle requires options"); 923 return -EINVAL; 924 } else { 925 return 0; 926 } 927 } 928 929 err = nla_parse_nested(tb, TCA_U32_MAX, opt, u32_policy, extack); 930 if (err < 0) 931 return err; 932 933 if (tb[TCA_U32_FLAGS]) { 934 flags = nla_get_u32(tb[TCA_U32_FLAGS]); 935 if (!tc_flags_valid(flags)) { 936 NL_SET_ERR_MSG_MOD(extack, "Invalid filter flags"); 937 return -EINVAL; 938 } 939 } 940 941 n = *arg; 942 if (n) { 943 struct tc_u_knode *new; 944 945 if (TC_U32_KEY(n->handle) == 0) { 946 NL_SET_ERR_MSG_MOD(extack, "Key node id cannot be zero"); 947 return -EINVAL; 948 } 949 950 if ((n->flags ^ flags) & 951 ~(TCA_CLS_FLAGS_IN_HW | TCA_CLS_FLAGS_NOT_IN_HW)) { 952 NL_SET_ERR_MSG_MOD(extack, "Key node flags do not match passed flags"); 953 return -EINVAL; 954 } 955 956 new = u32_init_knode(tp, n); 957 if (!new) 958 return -ENOMEM; 959 960 err = u32_set_parms(net, tp, base, 961 rtnl_dereference(n->ht_up), new, tb, 962 tca[TCA_RATE], ovr, extack); 963 964 if (err) { 965 u32_destroy_key(tp, new, false); 966 return err; 967 } 968 969 err = u32_replace_hw_knode(tp, new, flags, extack); 970 if (err) { 971 u32_destroy_key(tp, new, false); 972 return err; 973 } 974 975 if (!tc_in_hw(new->flags)) 976 new->flags |= TCA_CLS_FLAGS_NOT_IN_HW; 977 978 u32_replace_knode(tp, tp_c, new); 979 tcf_unbind_filter(tp, &n->res); 980 tcf_exts_get_net(&n->exts); 981 tcf_queue_work(&n->rwork, u32_delete_key_work); 982 return 0; 983 } 984 985 if (tb[TCA_U32_DIVISOR]) { 986 unsigned int divisor = nla_get_u32(tb[TCA_U32_DIVISOR]); 987 988 if (--divisor > 0x100) { 989 NL_SET_ERR_MSG_MOD(extack, "Exceeded maximum 256 hash buckets"); 990 return -EINVAL; 991 } 992 if (TC_U32_KEY(handle)) { 993 NL_SET_ERR_MSG_MOD(extack, "Divisor can only be used on a hash table"); 994 return -EINVAL; 995 } 996 ht = kzalloc(sizeof(*ht) + divisor*sizeof(void *), GFP_KERNEL); 997 if (ht == NULL) 998 return -ENOBUFS; 999 if (handle == 0) { 1000 handle = gen_new_htid(tp->data, ht); 1001 if (handle == 0) { 1002 kfree(ht); 1003 return -ENOMEM; 1004 } 1005 } else { 1006 err = idr_alloc_u32(&tp_c->handle_idr, ht, &handle, 1007 handle, GFP_KERNEL); 1008 if (err) { 1009 kfree(ht); 1010 return err; 1011 } 1012 } 1013 ht->tp_c = tp_c; 1014 ht->refcnt = 1; 1015 ht->divisor = divisor; 1016 ht->handle = handle; 1017 ht->prio = tp->prio; 1018 idr_init(&ht->handle_idr); 1019 ht->flags = flags; 1020 1021 err = u32_replace_hw_hnode(tp, ht, flags, extack); 1022 if (err) { 1023 idr_remove(&tp_c->handle_idr, handle); 1024 kfree(ht); 1025 return err; 1026 } 1027 1028 RCU_INIT_POINTER(ht->next, tp_c->hlist); 1029 rcu_assign_pointer(tp_c->hlist, ht); 1030 *arg = ht; 1031 1032 return 0; 1033 } 1034 1035 if (tb[TCA_U32_HASH]) { 1036 htid = nla_get_u32(tb[TCA_U32_HASH]); 1037 if (TC_U32_HTID(htid) == TC_U32_ROOT) { 1038 ht = rtnl_dereference(tp->root); 1039 htid = ht->handle; 1040 } else { 1041 ht = u32_lookup_ht(tp->data, TC_U32_HTID(htid)); 1042 if (!ht) { 1043 NL_SET_ERR_MSG_MOD(extack, "Specified hash table not found"); 1044 return -EINVAL; 1045 } 1046 } 1047 } else { 1048 ht = rtnl_dereference(tp->root); 1049 htid = ht->handle; 1050 } 1051 1052 if (ht->divisor < TC_U32_HASH(htid)) { 1053 NL_SET_ERR_MSG_MOD(extack, "Specified hash table buckets exceed configured value"); 1054 return -EINVAL; 1055 } 1056 1057 if (handle) { 1058 if (TC_U32_HTID(handle) && TC_U32_HTID(handle ^ htid)) { 1059 NL_SET_ERR_MSG_MOD(extack, "Handle specified hash table address mismatch"); 1060 return -EINVAL; 1061 } 1062 handle = htid | TC_U32_NODE(handle); 1063 err = idr_alloc_u32(&ht->handle_idr, NULL, &handle, handle, 1064 GFP_KERNEL); 1065 if (err) 1066 return err; 1067 } else 1068 handle = gen_new_kid(ht, htid); 1069 1070 if (tb[TCA_U32_SEL] == NULL) { 1071 NL_SET_ERR_MSG_MOD(extack, "Selector not specified"); 1072 err = -EINVAL; 1073 goto erridr; 1074 } 1075 1076 s = nla_data(tb[TCA_U32_SEL]); 1077 1078 n = kzalloc(sizeof(*n) + s->nkeys*sizeof(struct tc_u32_key), GFP_KERNEL); 1079 if (n == NULL) { 1080 err = -ENOBUFS; 1081 goto erridr; 1082 } 1083 1084 #ifdef CONFIG_CLS_U32_PERF 1085 size = sizeof(struct tc_u32_pcnt) + s->nkeys * sizeof(u64); 1086 n->pf = __alloc_percpu(size, __alignof__(struct tc_u32_pcnt)); 1087 if (!n->pf) { 1088 err = -ENOBUFS; 1089 goto errfree; 1090 } 1091 #endif 1092 1093 memcpy(&n->sel, s, sizeof(*s) + s->nkeys*sizeof(struct tc_u32_key)); 1094 RCU_INIT_POINTER(n->ht_up, ht); 1095 n->handle = handle; 1096 n->fshift = s->hmask ? ffs(ntohl(s->hmask)) - 1 : 0; 1097 n->flags = flags; 1098 n->tp = tp; 1099 1100 err = tcf_exts_init(&n->exts, TCA_U32_ACT, TCA_U32_POLICE); 1101 if (err < 0) 1102 goto errout; 1103 1104 #ifdef CONFIG_CLS_U32_MARK 1105 n->pcpu_success = alloc_percpu(u32); 1106 if (!n->pcpu_success) { 1107 err = -ENOMEM; 1108 goto errout; 1109 } 1110 1111 if (tb[TCA_U32_MARK]) { 1112 struct tc_u32_mark *mark; 1113 1114 mark = nla_data(tb[TCA_U32_MARK]); 1115 n->val = mark->val; 1116 n->mask = mark->mask; 1117 } 1118 #endif 1119 1120 err = u32_set_parms(net, tp, base, ht, n, tb, tca[TCA_RATE], ovr, 1121 extack); 1122 if (err == 0) { 1123 struct tc_u_knode __rcu **ins; 1124 struct tc_u_knode *pins; 1125 1126 err = u32_replace_hw_knode(tp, n, flags, extack); 1127 if (err) 1128 goto errhw; 1129 1130 if (!tc_in_hw(n->flags)) 1131 n->flags |= TCA_CLS_FLAGS_NOT_IN_HW; 1132 1133 ins = &ht->ht[TC_U32_HASH(handle)]; 1134 for (pins = rtnl_dereference(*ins); pins; 1135 ins = &pins->next, pins = rtnl_dereference(*ins)) 1136 if (TC_U32_NODE(handle) < TC_U32_NODE(pins->handle)) 1137 break; 1138 1139 RCU_INIT_POINTER(n->next, pins); 1140 rcu_assign_pointer(*ins, n); 1141 *arg = n; 1142 return 0; 1143 } 1144 1145 errhw: 1146 #ifdef CONFIG_CLS_U32_MARK 1147 free_percpu(n->pcpu_success); 1148 #endif 1149 1150 errout: 1151 tcf_exts_destroy(&n->exts); 1152 #ifdef CONFIG_CLS_U32_PERF 1153 errfree: 1154 free_percpu(n->pf); 1155 #endif 1156 kfree(n); 1157 erridr: 1158 idr_remove(&ht->handle_idr, handle); 1159 return err; 1160 } 1161 1162 static void u32_walk(struct tcf_proto *tp, struct tcf_walker *arg) 1163 { 1164 struct tc_u_common *tp_c = tp->data; 1165 struct tc_u_hnode *ht; 1166 struct tc_u_knode *n; 1167 unsigned int h; 1168 1169 if (arg->stop) 1170 return; 1171 1172 for (ht = rtnl_dereference(tp_c->hlist); 1173 ht; 1174 ht = rtnl_dereference(ht->next)) { 1175 if (ht->prio != tp->prio) 1176 continue; 1177 if (arg->count >= arg->skip) { 1178 if (arg->fn(tp, ht, arg) < 0) { 1179 arg->stop = 1; 1180 return; 1181 } 1182 } 1183 arg->count++; 1184 for (h = 0; h <= ht->divisor; h++) { 1185 for (n = rtnl_dereference(ht->ht[h]); 1186 n; 1187 n = rtnl_dereference(n->next)) { 1188 if (arg->count < arg->skip) { 1189 arg->count++; 1190 continue; 1191 } 1192 if (arg->fn(tp, n, arg) < 0) { 1193 arg->stop = 1; 1194 return; 1195 } 1196 arg->count++; 1197 } 1198 } 1199 } 1200 } 1201 1202 static void u32_bind_class(void *fh, u32 classid, unsigned long cl) 1203 { 1204 struct tc_u_knode *n = fh; 1205 1206 if (n && n->res.classid == classid) 1207 n->res.class = cl; 1208 } 1209 1210 static int u32_dump(struct net *net, struct tcf_proto *tp, void *fh, 1211 struct sk_buff *skb, struct tcmsg *t) 1212 { 1213 struct tc_u_knode *n = fh; 1214 struct tc_u_hnode *ht_up, *ht_down; 1215 struct nlattr *nest; 1216 1217 if (n == NULL) 1218 return skb->len; 1219 1220 t->tcm_handle = n->handle; 1221 1222 nest = nla_nest_start(skb, TCA_OPTIONS); 1223 if (nest == NULL) 1224 goto nla_put_failure; 1225 1226 if (TC_U32_KEY(n->handle) == 0) { 1227 struct tc_u_hnode *ht = fh; 1228 u32 divisor = ht->divisor + 1; 1229 1230 if (nla_put_u32(skb, TCA_U32_DIVISOR, divisor)) 1231 goto nla_put_failure; 1232 } else { 1233 #ifdef CONFIG_CLS_U32_PERF 1234 struct tc_u32_pcnt *gpf; 1235 int cpu; 1236 #endif 1237 1238 if (nla_put(skb, TCA_U32_SEL, 1239 sizeof(n->sel) + n->sel.nkeys*sizeof(struct tc_u32_key), 1240 &n->sel)) 1241 goto nla_put_failure; 1242 1243 ht_up = rtnl_dereference(n->ht_up); 1244 if (ht_up) { 1245 u32 htid = n->handle & 0xFFFFF000; 1246 if (nla_put_u32(skb, TCA_U32_HASH, htid)) 1247 goto nla_put_failure; 1248 } 1249 if (n->res.classid && 1250 nla_put_u32(skb, TCA_U32_CLASSID, n->res.classid)) 1251 goto nla_put_failure; 1252 1253 ht_down = rtnl_dereference(n->ht_down); 1254 if (ht_down && 1255 nla_put_u32(skb, TCA_U32_LINK, ht_down->handle)) 1256 goto nla_put_failure; 1257 1258 if (n->flags && nla_put_u32(skb, TCA_U32_FLAGS, n->flags)) 1259 goto nla_put_failure; 1260 1261 #ifdef CONFIG_CLS_U32_MARK 1262 if ((n->val || n->mask)) { 1263 struct tc_u32_mark mark = {.val = n->val, 1264 .mask = n->mask, 1265 .success = 0}; 1266 int cpum; 1267 1268 for_each_possible_cpu(cpum) { 1269 __u32 cnt = *per_cpu_ptr(n->pcpu_success, cpum); 1270 1271 mark.success += cnt; 1272 } 1273 1274 if (nla_put(skb, TCA_U32_MARK, sizeof(mark), &mark)) 1275 goto nla_put_failure; 1276 } 1277 #endif 1278 1279 if (tcf_exts_dump(skb, &n->exts) < 0) 1280 goto nla_put_failure; 1281 1282 #ifdef CONFIG_NET_CLS_IND 1283 if (n->ifindex) { 1284 struct net_device *dev; 1285 dev = __dev_get_by_index(net, n->ifindex); 1286 if (dev && nla_put_string(skb, TCA_U32_INDEV, dev->name)) 1287 goto nla_put_failure; 1288 } 1289 #endif 1290 #ifdef CONFIG_CLS_U32_PERF 1291 gpf = kzalloc(sizeof(struct tc_u32_pcnt) + 1292 n->sel.nkeys * sizeof(u64), 1293 GFP_KERNEL); 1294 if (!gpf) 1295 goto nla_put_failure; 1296 1297 for_each_possible_cpu(cpu) { 1298 int i; 1299 struct tc_u32_pcnt *pf = per_cpu_ptr(n->pf, cpu); 1300 1301 gpf->rcnt += pf->rcnt; 1302 gpf->rhit += pf->rhit; 1303 for (i = 0; i < n->sel.nkeys; i++) 1304 gpf->kcnts[i] += pf->kcnts[i]; 1305 } 1306 1307 if (nla_put_64bit(skb, TCA_U32_PCNT, 1308 sizeof(struct tc_u32_pcnt) + 1309 n->sel.nkeys * sizeof(u64), 1310 gpf, TCA_U32_PAD)) { 1311 kfree(gpf); 1312 goto nla_put_failure; 1313 } 1314 kfree(gpf); 1315 #endif 1316 } 1317 1318 nla_nest_end(skb, nest); 1319 1320 if (TC_U32_KEY(n->handle)) 1321 if (tcf_exts_dump_stats(skb, &n->exts) < 0) 1322 goto nla_put_failure; 1323 return skb->len; 1324 1325 nla_put_failure: 1326 nla_nest_cancel(skb, nest); 1327 return -1; 1328 } 1329 1330 static struct tcf_proto_ops cls_u32_ops __read_mostly = { 1331 .kind = "u32", 1332 .classify = u32_classify, 1333 .init = u32_init, 1334 .destroy = u32_destroy, 1335 .get = u32_get, 1336 .change = u32_change, 1337 .delete = u32_delete, 1338 .walk = u32_walk, 1339 .dump = u32_dump, 1340 .bind_class = u32_bind_class, 1341 .owner = THIS_MODULE, 1342 }; 1343 1344 static int __init init_u32(void) 1345 { 1346 int i, ret; 1347 1348 pr_info("u32 classifier\n"); 1349 #ifdef CONFIG_CLS_U32_PERF 1350 pr_info(" Performance counters on\n"); 1351 #endif 1352 #ifdef CONFIG_NET_CLS_IND 1353 pr_info(" input device check on\n"); 1354 #endif 1355 #ifdef CONFIG_NET_CLS_ACT 1356 pr_info(" Actions configured\n"); 1357 #endif 1358 tc_u_common_hash = kvmalloc_array(U32_HASH_SIZE, 1359 sizeof(struct hlist_head), 1360 GFP_KERNEL); 1361 if (!tc_u_common_hash) 1362 return -ENOMEM; 1363 1364 for (i = 0; i < U32_HASH_SIZE; i++) 1365 INIT_HLIST_HEAD(&tc_u_common_hash[i]); 1366 1367 ret = register_tcf_proto_ops(&cls_u32_ops); 1368 if (ret) 1369 kvfree(tc_u_common_hash); 1370 return ret; 1371 } 1372 1373 static void __exit exit_u32(void) 1374 { 1375 unregister_tcf_proto_ops(&cls_u32_ops); 1376 kvfree(tc_u_common_hash); 1377 } 1378 1379 module_init(init_u32) 1380 module_exit(exit_u32) 1381 MODULE_LICENSE("GPL"); 1382