net/sched/cls_u32.c

   1 // SPDX-License-Identifier: GPL-2.0-or-later
   2 /*
   3  * net/sched/cls_u32.c  Ugly (or Universal) 32bit key Packet Classifier.
   4  *
   5  * Authors:     Alexey Kuznetsov, <[email protected]>
   6  *
   7  *      The filters are packed to hash tables of key nodes
   8  *      with a set of 32bit key/mask pairs at every node.
   9  *      Nodes reference next level hash tables etc.
  10  *
  11  *      This scheme is the best universal classifier I managed to
  12  *      invent; it is not super-fast, but it is not slow (provided you
  13  *      program it correctly), and general enough.  And its relative
  14  *      speed grows as the number of rules becomes larger.
  15  *
  16  *      It seems that it represents the best middle point between
  17  *      speed and manageability both by human and by machine.
  18  *
  19  *      It is especially useful for link sharing combined with QoS;
  20  *      pure RSVP doesn't need such a general approach and can use
  21  *      much simpler (and faster) schemes, sort of cls_rsvp.c.
  22  *
  23  *      nfmark match added by Catalin(ux aka Dino) BOIE <catab at umbrella.ro>
  24  */
  25
  26 #include <linux/module.h>
  27 #include <linux/slab.h>
  28 #include <linux/types.h>
  29 #include <linux/kernel.h>
  30 #include <linux/string.h>
  31 #include <linux/errno.h>
  32 #include <linux/percpu.h>
  33 #include <linux/rtnetlink.h>
  34 #include <linux/skbuff.h>
  35 #include <linux/bitmap.h>
  36 #include <linux/netdevice.h>
  37 #include <linux/hash.h>
  38 #include <net/netlink.h>
  39 #include <net/act_api.h>
  40 #include <net/pkt_cls.h>
  41 #include <linux/idr.h>
  42 #include <net/tc_wrapper.h>
  43
  44 struct tc_u_knode {
  45         struct tc_u_knode __rcu *next;
  46         u32                     handle;
  47         struct tc_u_hnode __rcu *ht_up;
  48         struct tcf_exts         exts;
  49         int                     ifindex;
  50         u8                      fshift;
  51         struct tcf_result       res;
  52         struct tc_u_hnode __rcu *ht_down;
  53 #ifdef CONFIG_CLS_U32_PERF
  54         struct tc_u32_pcnt __percpu *pf;
  55 #endif
  56         u32                     flags;
  57         unsigned int            in_hw_count;
  58 #ifdef CONFIG_CLS_U32_MARK
  59         u32                     val;
  60         u32                     mask;
  61         u32 __percpu            *pcpu_success;
  62 #endif
  63         struct rcu_work         rwork;
  64         /* The 'sel' field MUST be the last field in structure to allow for
  65          * tc_u32_keys allocated at end of structure.
  66          */
  67         struct tc_u32_sel       sel;
  68 };
  69
  70 struct tc_u_hnode {
  71         struct tc_u_hnode __rcu *next;
  72         u32                     handle;
  73         u32                     prio;
  74         refcount_t              refcnt;
  75         unsigned int            divisor;
  76         struct idr              handle_idr;
  77         bool                    is_root;
  78         struct rcu_head         rcu;
  79         u32                     flags;
  80         /* The 'ht' field MUST be the last field in structure to allow for
  81          * more entries allocated at end of structure.
  82          */
  83         struct tc_u_knode __rcu *ht[];
  84 };
  85
  86 struct tc_u_common {
  87         struct tc_u_hnode __rcu *hlist;
  88         void                    *ptr;
  89         refcount_t              refcnt;
  90         struct idr              handle_idr;
  91         struct hlist_node       hnode;
  92         long                    knodes;
  93 };
  94
  95 static u32 handle2id(u32 h)
  96 {
  97         return ((h & 0x80000000) ? ((h >> 20) & 0x7FF) : h);
  98 }
  99
 100 static u32 id2handle(u32 id)
 101 {
 102         return (id | 0x800U) << 20;
 103 }
 104
 105 static inline unsigned int u32_hash_fold(__be32 key,
 106                                          const struct tc_u32_sel *sel,
 107                                          u8 fshift)
 108 {
 109         unsigned int h = ntohl(key & sel->hmask) >> fshift;
 110
 111         return h;
 112 }
 113
 114 TC_INDIRECT_SCOPE int u32_classify(struct sk_buff *skb,
 115                                    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                                 if (!tcf_match_indev(skb, n->ifindex)) {
 187                                         n = rcu_dereference_bh(n->next);
 188                                         goto next_knode;
 189                                 }
 190 #ifdef CONFIG_CLS_U32_PERF
 191                                 __this_cpu_inc(n->pf->rhit);
 192 #endif
 193                                 r = tcf_exts_exec(skb, &n->exts, res);
 194                                 if (r < 0) {
 195                                         n = rcu_dereference_bh(n->next);
 196                                         goto next_knode;
 197                                 }
 198
 199                                 return r;
 200                         }
 201                         n = rcu_dereference_bh(n->next);
 202                         goto next_knode;
 203                 }
 204
 205                 /* PUSH */
 206                 if (sdepth >= TC_U32_MAXDEPTH)
 207                         goto deadloop;
 208                 stack[sdepth].knode = n;
 209                 stack[sdepth].off = off;
 210                 sdepth++;
 211
 212                 ht = rcu_dereference_bh(n->ht_down);
 213                 sel = 0;
 214                 if (ht->divisor) {
 215                         __be32 *data, hdata;
 216
 217                         data = skb_header_pointer(skb, off + n->sel.hoff, 4,
 218                                                   &hdata);
 219                         if (!data)
 220                                 goto out;
 221                         sel = ht->divisor & u32_hash_fold(*data, &n->sel,
 222                                                           n->fshift);
 223                 }
 224                 if (!(n->sel.flags & (TC_U32_VAROFFSET | TC_U32_OFFSET | TC_U32_EAT)))
 225                         goto next_ht;
 226
 227                 if (n->sel.flags & (TC_U32_OFFSET | TC_U32_VAROFFSET)) {
 228                         off2 = n->sel.off + 3;
 229                         if (n->sel.flags & TC_U32_VAROFFSET) {
 230                                 __be16 *data, hdata;
 231
 232                                 data = skb_header_pointer(skb,
 233                                                           off + n->sel.offoff,
 234                                                           2, &hdata);
 235                                 if (!data)
 236                                         goto out;
 237                                 off2 += ntohs(n->sel.offmask & *data) >>
 238                                         n->sel.offshift;
 239                         }
 240                         off2 &= ~3;
 241                 }
 242                 if (n->sel.flags & TC_U32_EAT) {
 243                         off += off2;
 244                         off2 = 0;
 245                 }
 246
 247                 if (off < skb->len)
 248                         goto next_ht;
 249         }
 250
 251         /* POP */
 252         if (sdepth--) {
 253                 n = stack[sdepth].knode;
 254                 ht = rcu_dereference_bh(n->ht_up);
 255                 off = stack[sdepth].off;
 256                 goto check_terminal;
 257         }
 258 out:
 259         return -1;
 260
 261 deadloop:
 262         net_warn_ratelimited("cls_u32: dead loop\n");
 263         return -1;
 264 }
 265
 266 static struct tc_u_hnode *u32_lookup_ht(struct tc_u_common *tp_c, u32 handle)
 267 {
 268         struct tc_u_hnode *ht;
 269
 270         for (ht = rtnl_dereference(tp_c->hlist);
 271              ht;
 272              ht = rtnl_dereference(ht->next))
 273                 if (ht->handle == handle)
 274                         break;
 275
 276         return ht;
 277 }
 278
 279 static struct tc_u_knode *u32_lookup_key(struct tc_u_hnode *ht, u32 handle)
 280 {
 281         unsigned int sel;
 282         struct tc_u_knode *n = NULL;
 283
 284         sel = TC_U32_HASH(handle);
 285         if (sel > ht->divisor)
 286                 goto out;
 287
 288         for (n = rtnl_dereference(ht->ht[sel]);
 289              n;
 290              n = rtnl_dereference(n->next))
 291                 if (n->handle == handle)
 292                         break;
 293 out:
 294         return n;
 295 }
 296
 297
 298 static void *u32_get(struct tcf_proto *tp, u32 handle)
 299 {
 300         struct tc_u_hnode *ht;
 301         struct tc_u_common *tp_c = tp->data;
 302
 303         if (TC_U32_HTID(handle) == TC_U32_ROOT)
 304                 ht = rtnl_dereference(tp->root);
 305         else
 306                 ht = u32_lookup_ht(tp_c, TC_U32_HTID(handle));
 307
 308         if (!ht)
 309                 return NULL;
 310
 311         if (TC_U32_KEY(handle) == 0)
 312                 return ht;
 313
 314         return u32_lookup_key(ht, handle);
 315 }
 316
 317 /* Protected by rtnl lock */
 318 static u32 gen_new_htid(struct tc_u_common *tp_c, struct tc_u_hnode *ptr)
 319 {
 320         int id = idr_alloc_cyclic(&tp_c->handle_idr, ptr, 1, 0x7FF, GFP_KERNEL);
 321         if (id < 0)
 322                 return 0;
 323         return id2handle(id);
 324 }
 325
 326 static struct hlist_head *tc_u_common_hash;
 327
 328 #define U32_HASH_SHIFT 10
 329 #define U32_HASH_SIZE (1 << U32_HASH_SHIFT)
 330
 331 static void *tc_u_common_ptr(const struct tcf_proto *tp)
 332 {
 333         struct tcf_block *block = tp->chain->block;
 334
 335         /* The block sharing is currently supported only
 336          * for classless qdiscs. In that case we use block
 337          * for tc_u_common identification. In case the
 338          * block is not shared, block->q is a valid pointer
 339          * and we can use that. That works for classful qdiscs.
 340          */
 341         if (tcf_block_shared(block))
 342                 return block;
 343         else
 344                 return block->q;
 345 }
 346
 347 static struct hlist_head *tc_u_hash(void *key)
 348 {
 349         return tc_u_common_hash + hash_ptr(key, U32_HASH_SHIFT);
 350 }
 351
 352 static struct tc_u_common *tc_u_common_find(void *key)
 353 {
 354         struct tc_u_common *tc;
 355         hlist_for_each_entry(tc, tc_u_hash(key), hnode) {
 356                 if (tc->ptr == key)
 357                         return tc;
 358         }
 359         return NULL;
 360 }
 361
 362 static int u32_init(struct tcf_proto *tp)
 363 {
 364         struct tc_u_hnode *root_ht;
 365         void *key = tc_u_common_ptr(tp);
 366         struct tc_u_common *tp_c = tc_u_common_find(key);
 367
 368         root_ht = kzalloc(struct_size(root_ht, ht, 1), GFP_KERNEL);
 369         if (root_ht == NULL)
 370                 return -ENOBUFS;
 371
 372         refcount_set(&root_ht->refcnt, 1);
 373         root_ht->handle = tp_c ? gen_new_htid(tp_c, root_ht) : id2handle(0);
 374         root_ht->prio = tp->prio;
 375         root_ht->is_root = true;
 376         idr_init(&root_ht->handle_idr);
 377
 378         if (tp_c == NULL) {
 379                 tp_c = kzalloc(sizeof(*tp_c), GFP_KERNEL);
 380                 if (tp_c == NULL) {
 381                         kfree(root_ht);
 382                         return -ENOBUFS;
 383                 }
 384                 refcount_set(&tp_c->refcnt, 1);
 385                 tp_c->ptr = key;
 386                 INIT_HLIST_NODE(&tp_c->hnode);
 387                 idr_init(&tp_c->handle_idr);
 388
 389                 hlist_add_head(&tp_c->hnode, tc_u_hash(key));
 390         } else {
 391                 refcount_inc(&tp_c->refcnt);
 392         }
 393
 394         RCU_INIT_POINTER(root_ht->next, tp_c->hlist);
 395         rcu_assign_pointer(tp_c->hlist, root_ht);
 396
 397         /* root_ht must be destroyed when tcf_proto is destroyed */
 398         rcu_assign_pointer(tp->root, root_ht);
 399         tp->data = tp_c;
 400         return 0;
 401 }
 402
 403 static void __u32_destroy_key(struct tc_u_knode *n)
 404 {
 405         struct tc_u_hnode *ht = rtnl_dereference(n->ht_down);
 406
 407         tcf_exts_destroy(&n->exts);
 408         if (ht && refcount_dec_and_test(&ht->refcnt))
 409                 kfree(ht);
 410         kfree(n);
 411 }
 412
 413 static void u32_destroy_key(struct tc_u_knode *n, bool free_pf)
 414 {
 415         tcf_exts_put_net(&n->exts);
 416 #ifdef CONFIG_CLS_U32_PERF
 417         if (free_pf)
 418                 free_percpu(n->pf);
 419 #endif
 420 #ifdef CONFIG_CLS_U32_MARK
 421         if (free_pf)
 422                 free_percpu(n->pcpu_success);
 423 #endif
 424         __u32_destroy_key(n);
 425 }
 426
 427 /* u32_delete_key_rcu should be called when free'ing a copied
 428  * version of a tc_u_knode obtained from u32_init_knode(). When
 429  * copies are obtained from u32_init_knode() the statistics are
 430  * shared between the old and new copies to allow readers to
 431  * continue to update the statistics during the copy. To support
 432  * this the u32_delete_key_rcu variant does not free the percpu
 433  * statistics.
 434  */
 435 static void u32_delete_key_work(struct work_struct *work)
 436 {
 437         struct tc_u_knode *key = container_of(to_rcu_work(work),
 438                                               struct tc_u_knode,
 439                                               rwork);
 440         rtnl_lock();
 441         u32_destroy_key(key, false);
 442         rtnl_unlock();
 443 }
 444
 445 /* u32_delete_key_freepf_rcu is the rcu callback variant
 446  * that free's the entire structure including the statistics
 447  * percpu variables. Only use this if the key is not a copy
 448  * returned by u32_init_knode(). See u32_delete_key_rcu()
 449  * for the variant that should be used with keys return from
 450  * u32_init_knode()
 451  */
 452 static void u32_delete_key_freepf_work(struct work_struct *work)
 453 {
 454         struct tc_u_knode *key = container_of(to_rcu_work(work),
 455                                               struct tc_u_knode,
 456                                               rwork);
 457         rtnl_lock();
 458         u32_destroy_key(key, true);
 459         rtnl_unlock();
 460 }
 461
 462 static int u32_delete_key(struct tcf_proto *tp, struct tc_u_knode *key)
 463 {
 464         struct tc_u_common *tp_c = tp->data;
 465         struct tc_u_knode __rcu **kp;
 466         struct tc_u_knode *pkp;
 467         struct tc_u_hnode *ht = rtnl_dereference(key->ht_up);
 468
 469         if (ht) {
 470                 kp = &ht->ht[TC_U32_HASH(key->handle)];
 471                 for (pkp = rtnl_dereference(*kp); pkp;
 472                      kp = &pkp->next, pkp = rtnl_dereference(*kp)) {
 473                         if (pkp == key) {
 474                                 RCU_INIT_POINTER(*kp, key->next);
 475                                 tp_c->knodes--;
 476
 477                                 tcf_unbind_filter(tp, &key->res);
 478                                 idr_remove(&ht->handle_idr, key->handle);
 479                                 tcf_exts_get_net(&key->exts);
 480                                 tcf_queue_work(&key->rwork, u32_delete_key_freepf_work);
 481                                 return 0;
 482                         }
 483                 }
 484         }
 485         WARN_ON(1);
 486         return 0;
 487 }
 488
 489 static void u32_clear_hw_hnode(struct tcf_proto *tp, struct tc_u_hnode *h,
 490                                struct netlink_ext_ack *extack)
 491 {
 492         struct tcf_block *block = tp->chain->block;
 493         struct tc_cls_u32_offload cls_u32 = {};
 494
 495         tc_cls_common_offload_init(&cls_u32.common, tp, h->flags, extack);
 496         cls_u32.command = TC_CLSU32_DELETE_HNODE;
 497         cls_u32.hnode.divisor = h->divisor;
 498         cls_u32.hnode.handle = h->handle;
 499         cls_u32.hnode.prio = h->prio;
 500
 501         tc_setup_cb_call(block, TC_SETUP_CLSU32, &cls_u32, false, true);
 502 }
 503
 504 static int u32_replace_hw_hnode(struct tcf_proto *tp, struct tc_u_hnode *h,
 505                                 u32 flags, struct netlink_ext_ack *extack)
 506 {
 507         struct tcf_block *block = tp->chain->block;
 508         struct tc_cls_u32_offload cls_u32 = {};
 509         bool skip_sw = tc_skip_sw(flags);
 510         bool offloaded = false;
 511         int err;
 512
 513         tc_cls_common_offload_init(&cls_u32.common, tp, flags, extack);
 514         cls_u32.command = TC_CLSU32_NEW_HNODE;
 515         cls_u32.hnode.divisor = h->divisor;
 516         cls_u32.hnode.handle = h->handle;
 517         cls_u32.hnode.prio = h->prio;
 518
 519         err = tc_setup_cb_call(block, TC_SETUP_CLSU32, &cls_u32, skip_sw, true);
 520         if (err < 0) {
 521                 u32_clear_hw_hnode(tp, h, NULL);
 522                 return err;
 523         } else if (err > 0) {
 524                 offloaded = true;
 525         }
 526
 527         if (skip_sw && !offloaded)
 528                 return -EINVAL;
 529
 530         return 0;
 531 }
 532
 533 static void u32_remove_hw_knode(struct tcf_proto *tp, struct tc_u_knode *n,
 534                                 struct netlink_ext_ack *extack)
 535 {
 536         struct tcf_block *block = tp->chain->block;
 537         struct tc_cls_u32_offload cls_u32 = {};
 538
 539         tc_cls_common_offload_init(&cls_u32.common, tp, n->flags, extack);
 540         cls_u32.command = TC_CLSU32_DELETE_KNODE;
 541         cls_u32.knode.handle = n->handle;
 542
 543         tc_setup_cb_destroy(block, tp, TC_SETUP_CLSU32, &cls_u32, false,
 544                             &n->flags, &n->in_hw_count, true);
 545 }
 546
 547 static int u32_replace_hw_knode(struct tcf_proto *tp, struct tc_u_knode *n,
 548                                 u32 flags, struct netlink_ext_ack *extack)
 549 {
 550         struct tc_u_hnode *ht = rtnl_dereference(n->ht_down);
 551         struct tcf_block *block = tp->chain->block;
 552         struct tc_cls_u32_offload cls_u32 = {};
 553         bool skip_sw = tc_skip_sw(flags);
 554         int err;
 555
 556         tc_cls_common_offload_init(&cls_u32.common, tp, flags, extack);
 557         cls_u32.command = TC_CLSU32_REPLACE_KNODE;
 558         cls_u32.knode.handle = n->handle;
 559         cls_u32.knode.fshift = n->fshift;
 560 #ifdef CONFIG_CLS_U32_MARK
 561         cls_u32.knode.val = n->val;
 562         cls_u32.knode.mask = n->mask;
 563 #else
 564         cls_u32.knode.val = 0;
 565         cls_u32.knode.mask = 0;
 566 #endif
 567         cls_u32.knode.sel = &n->sel;
 568         cls_u32.knode.res = &n->res;
 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_add(block, tp, TC_SETUP_CLSU32, &cls_u32, skip_sw,
 574                               &n->flags, &n->in_hw_count, true);
 575         if (err) {
 576                 u32_remove_hw_knode(tp, n, NULL);
 577                 return err;
 578         }
 579
 580         if (skip_sw && !(n->flags & TCA_CLS_FLAGS_IN_HW))
 581                 return -EINVAL;
 582
 583         return 0;
 584 }
 585
 586 static void u32_clear_hnode(struct tcf_proto *tp, struct tc_u_hnode *ht,
 587                             struct netlink_ext_ack *extack)
 588 {
 589         struct tc_u_common *tp_c = tp->data;
 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                         tp_c->knodes--;
 598                         tcf_unbind_filter(tp, &n->res);
 599                         u32_remove_hw_knode(tp, n, extack);
 600                         idr_remove(&ht->handle_idr, n->handle);
 601                         if (tcf_exts_get_net(&n->exts))
 602                                 tcf_queue_work(&n->rwork, u32_delete_key_freepf_work);
 603                         else
 604                                 u32_destroy_key(n, true);
 605                 }
 606         }
 607 }
 608
 609 static int u32_destroy_hnode(struct tcf_proto *tp, struct tc_u_hnode *ht,
 610                              struct netlink_ext_ack *extack)
 611 {
 612         struct tc_u_common *tp_c = tp->data;
 613         struct tc_u_hnode __rcu **hn;
 614         struct tc_u_hnode *phn;
 615
 616         u32_clear_hnode(tp, ht, extack);
 617
 618         hn = &tp_c->hlist;
 619         for (phn = rtnl_dereference(*hn);
 620              phn;
 621              hn = &phn->next, phn = rtnl_dereference(*hn)) {
 622                 if (phn == ht) {
 623                         u32_clear_hw_hnode(tp, ht, extack);
 624                         idr_destroy(&ht->handle_idr);
 625                         idr_remove(&tp_c->handle_idr, handle2id(ht->handle));
 626                         RCU_INIT_POINTER(*hn, ht->next);
 627                         kfree_rcu(ht, rcu);
 628                         return 0;
 629                 }
 630         }
 631
 632         return -ENOENT;
 633 }
 634
 635 static void u32_destroy(struct tcf_proto *tp, bool rtnl_held,
 636                         struct netlink_ext_ack *extack)
 637 {
 638         struct tc_u_common *tp_c = tp->data;
 639         struct tc_u_hnode *root_ht = rtnl_dereference(tp->root);
 640
 641         WARN_ON(root_ht == NULL);
 642
 643         if (root_ht && refcount_dec_and_test(&root_ht->refcnt))
 644                 u32_destroy_hnode(tp, root_ht, extack);
 645
 646         if (refcount_dec_and_test(&tp_c->refcnt)) {
 647                 struct tc_u_hnode *ht;
 648
 649                 hlist_del(&tp_c->hnode);
 650
 651                 while ((ht = rtnl_dereference(tp_c->hlist)) != NULL) {
 652                         u32_clear_hnode(tp, ht, extack);
 653                         RCU_INIT_POINTER(tp_c->hlist, ht->next);
 654
 655                         /* u32_destroy_key() will later free ht for us, if it's
 656                          * still referenced by some knode
 657                          */
 658                         if (refcount_dec_and_test(&ht->refcnt))
 659                                 kfree_rcu(ht, rcu);
 660                 }
 661
 662                 idr_destroy(&tp_c->handle_idr);
 663                 kfree(tp_c);
 664         }
 665
 666         tp->data = NULL;
 667 }
 668
 669 static int u32_delete(struct tcf_proto *tp, void *arg, bool *last,
 670                       bool rtnl_held, struct netlink_ext_ack *extack)
 671 {
 672         struct tc_u_hnode *ht = arg;
 673         struct tc_u_common *tp_c = tp->data;
 674         int ret = 0;
 675
 676         if (TC_U32_KEY(ht->handle)) {
 677                 u32_remove_hw_knode(tp, (struct tc_u_knode *)ht, extack);
 678                 ret = u32_delete_key(tp, (struct tc_u_knode *)ht);
 679                 goto out;
 680         }
 681
 682         if (ht->is_root) {
 683                 NL_SET_ERR_MSG_MOD(extack, "Not allowed to delete root node");
 684                 return -EINVAL;
 685         }
 686
 687         if (refcount_dec_if_one(&ht->refcnt)) {
 688                 u32_destroy_hnode(tp, ht, extack);
 689         } else {
 690                 NL_SET_ERR_MSG_MOD(extack, "Can not delete in-use filter");
 691                 return -EBUSY;
 692         }
 693
 694 out:
 695         *last = refcount_read(&tp_c->refcnt) == 1 && tp_c->knodes == 0;
 696         return ret;
 697 }
 698
 699 static u32 gen_new_kid(struct tc_u_hnode *ht, u32 htid)
 700 {
 701         u32 index = htid | 0x800;
 702         u32 max = htid | 0xFFF;
 703
 704         if (idr_alloc_u32(&ht->handle_idr, NULL, &index, max, GFP_KERNEL)) {
 705                 index = htid + 1;
 706                 if (idr_alloc_u32(&ht->handle_idr, NULL, &index, max,
 707                                  GFP_KERNEL))
 708                         index = max;
 709         }
 710
 711         return index;
 712 }
 713
 714 static const struct nla_policy u32_policy[TCA_U32_MAX + 1] = {
 715         [TCA_U32_CLASSID]       = { .type = NLA_U32 },
 716         [TCA_U32_HASH]          = { .type = NLA_U32 },
 717         [TCA_U32_LINK]          = { .type = NLA_U32 },
 718         [TCA_U32_DIVISOR]       = { .type = NLA_U32 },
 719         [TCA_U32_SEL]           = { .len = sizeof(struct tc_u32_sel) },
 720         [TCA_U32_INDEV]         = { .type = NLA_STRING, .len = IFNAMSIZ },
 721         [TCA_U32_MARK]          = { .len = sizeof(struct tc_u32_mark) },
 722         [TCA_U32_FLAGS]         = { .type = NLA_U32 },
 723 };
 724
 725 static void u32_unbind_filter(struct tcf_proto *tp, struct tc_u_knode *n,
 726                               struct nlattr **tb)
 727 {
 728         if (tb[TCA_U32_CLASSID])
 729                 tcf_unbind_filter(tp, &n->res);
 730 }
 731
 732 static void u32_bind_filter(struct tcf_proto *tp, struct tc_u_knode *n,
 733                             unsigned long base, struct nlattr **tb)
 734 {
 735         if (tb[TCA_U32_CLASSID]) {
 736                 n->res.classid = nla_get_u32(tb[TCA_U32_CLASSID]);
 737                 tcf_bind_filter(tp, &n->res, base);
 738         }
 739 }
 740
 741 static int u32_set_parms(struct net *net, struct tcf_proto *tp,
 742                          struct tc_u_knode *n, struct nlattr **tb,
 743                          struct nlattr *est, u32 flags, u32 fl_flags,
 744                          struct netlink_ext_ack *extack)
 745 {
 746         int err, ifindex = -1;
 747
 748         err = tcf_exts_validate_ex(net, tp, tb, est, &n->exts, flags,
 749                                    fl_flags, extack);
 750         if (err < 0)
 751                 return err;
 752
 753         if (tb[TCA_U32_INDEV]) {
 754                 ifindex = tcf_change_indev(net, tb[TCA_U32_INDEV], extack);
 755                 if (ifindex < 0)
 756                         return -EINVAL;
 757         }
 758
 759         if (tb[TCA_U32_LINK]) {
 760                 u32 handle = nla_get_u32(tb[TCA_U32_LINK]);
 761                 struct tc_u_hnode *ht_down = NULL, *ht_old;
 762
 763                 if (TC_U32_KEY(handle)) {
 764                         NL_SET_ERR_MSG_MOD(extack, "u32 Link handle must be a hash table");
 765                         return -EINVAL;
 766                 }
 767
 768                 if (handle) {
 769                         ht_down = u32_lookup_ht(tp->data, handle);
 770
 771                         if (!ht_down) {
 772                                 NL_SET_ERR_MSG_MOD(extack, "Link hash table not found");
 773                                 return -EINVAL;
 774                         }
 775                         if (ht_down->is_root) {
 776                                 NL_SET_ERR_MSG_MOD(extack, "Not linking to root node");
 777                                 return -EINVAL;
 778                         }
 779                         refcount_inc(&ht_down->refcnt);
 780                 }
 781
 782                 ht_old = rtnl_dereference(n->ht_down);
 783                 rcu_assign_pointer(n->ht_down, ht_down);
 784
 785                 if (ht_old)
 786                         refcount_dec(&ht_old->refcnt);
 787         }
 788
 789         if (ifindex >= 0)
 790                 n->ifindex = ifindex;
 791
 792         return 0;
 793 }
 794
 795 static void u32_replace_knode(struct tcf_proto *tp, struct tc_u_common *tp_c,
 796                               struct tc_u_knode *n)
 797 {
 798         struct tc_u_knode __rcu **ins;
 799         struct tc_u_knode *pins;
 800         struct tc_u_hnode *ht;
 801
 802         if (TC_U32_HTID(n->handle) == TC_U32_ROOT)
 803                 ht = rtnl_dereference(tp->root);
 804         else
 805                 ht = u32_lookup_ht(tp_c, TC_U32_HTID(n->handle));
 806
 807         ins = &ht->ht[TC_U32_HASH(n->handle)];
 808
 809         /* The node must always exist for it to be replaced if this is not the
 810          * case then something went very wrong elsewhere.
 811          */
 812         for (pins = rtnl_dereference(*ins); ;
 813              ins = &pins->next, pins = rtnl_dereference(*ins))
 814                 if (pins->handle == n->handle)
 815                         break;
 816
 817         idr_replace(&ht->handle_idr, n, n->handle);
 818         RCU_INIT_POINTER(n->next, pins->next);
 819         rcu_assign_pointer(*ins, n);
 820 }
 821
 822 static struct tc_u_knode *u32_init_knode(struct net *net, struct tcf_proto *tp,
 823                                          struct tc_u_knode *n)
 824 {
 825         struct tc_u_hnode *ht = rtnl_dereference(n->ht_down);
 826         struct tc_u32_sel *s = &n->sel;
 827         struct tc_u_knode *new;
 828
 829         new = kzalloc(struct_size(new, sel.keys, s->nkeys), GFP_KERNEL);
 830         if (!new)
 831                 return NULL;
 832
 833         RCU_INIT_POINTER(new->next, n->next);
 834         new->handle = n->handle;
 835         RCU_INIT_POINTER(new->ht_up, n->ht_up);
 836
 837         new->ifindex = n->ifindex;
 838         new->fshift = n->fshift;
 839         new->flags = n->flags;
 840         RCU_INIT_POINTER(new->ht_down, ht);
 841
 842 #ifdef CONFIG_CLS_U32_PERF
 843         /* Statistics may be incremented by readers during update
 844          * so we must keep them in tact. When the node is later destroyed
 845          * a special destroy call must be made to not free the pf memory.
 846          */
 847         new->pf = n->pf;
 848 #endif
 849
 850 #ifdef CONFIG_CLS_U32_MARK
 851         new->val = n->val;
 852         new->mask = n->mask;
 853         /* Similarly success statistics must be moved as pointers */
 854         new->pcpu_success = n->pcpu_success;
 855 #endif
 856         memcpy(&new->sel, s, struct_size(s, keys, s->nkeys));
 857
 858         if (tcf_exts_init(&new->exts, net, TCA_U32_ACT, TCA_U32_POLICE)) {
 859                 kfree(new);
 860                 return NULL;
 861         }
 862
 863         /* bump reference count as long as we hold pointer to structure */
 864         if (ht)
 865                 refcount_inc(&ht->refcnt);
 866
 867         return new;
 868 }
 869
 870 static int u32_change(struct net *net, struct sk_buff *in_skb,
 871                       struct tcf_proto *tp, unsigned long base, u32 handle,
 872                       struct nlattr **tca, void **arg, u32 flags,
 873                       struct netlink_ext_ack *extack)
 874 {
 875         struct tc_u_common *tp_c = tp->data;
 876         struct tc_u_hnode *ht;
 877         struct tc_u_knode *n;
 878         struct tc_u32_sel *s;
 879         struct nlattr *opt = tca[TCA_OPTIONS];
 880         struct nlattr *tb[TCA_U32_MAX + 1];
 881         u32 htid, userflags = 0;
 882         size_t sel_size;
 883         int err;
 884
 885         if (!opt) {
 886                 if (handle) {
 887                         NL_SET_ERR_MSG_MOD(extack, "Filter handle requires options");
 888                         return -EINVAL;
 889                 } else {
 890                         return 0;
 891                 }
 892         }
 893
 894         err = nla_parse_nested_deprecated(tb, TCA_U32_MAX, opt, u32_policy,
 895                                           extack);
 896         if (err < 0)
 897                 return err;
 898
 899         if (tb[TCA_U32_FLAGS]) {
 900                 userflags = nla_get_u32(tb[TCA_U32_FLAGS]);
 901                 if (!tc_flags_valid(userflags)) {
 902                         NL_SET_ERR_MSG_MOD(extack, "Invalid filter flags");
 903                         return -EINVAL;
 904                 }
 905         }
 906
 907         n = *arg;
 908         if (n) {
 909                 struct tc_u_knode *new;
 910
 911                 if (TC_U32_KEY(n->handle) == 0) {
 912                         NL_SET_ERR_MSG_MOD(extack, "Key node id cannot be zero");
 913                         return -EINVAL;
 914                 }
 915
 916                 if ((n->flags ^ userflags) &
 917                     ~(TCA_CLS_FLAGS_IN_HW | TCA_CLS_FLAGS_NOT_IN_HW)) {
 918                         NL_SET_ERR_MSG_MOD(extack, "Key node flags do not match passed flags");
 919                         return -EINVAL;
 920                 }
 921
 922                 new = u32_init_knode(net, tp, n);
 923                 if (!new)
 924                         return -ENOMEM;
 925
 926                 err = u32_set_parms(net, tp, new, tb, tca[TCA_RATE],
 927                                     flags, new->flags, extack);
 928
 929                 if (err) {
 930                         __u32_destroy_key(new);
 931                         return err;
 932                 }
 933
 934                 u32_bind_filter(tp, new, base, tb);
 935
 936                 err = u32_replace_hw_knode(tp, new, flags, extack);
 937                 if (err) {
 938                         u32_unbind_filter(tp, new, tb);
 939
 940                         if (tb[TCA_U32_LINK]) {
 941                                 struct tc_u_hnode *ht_old;
 942
 943                                 ht_old = rtnl_dereference(n->ht_down);
 944                                 if (ht_old)
 945                                         refcount_inc(&ht_old->refcnt);
 946                         }
 947                         __u32_destroy_key(new);
 948                         return err;
 949                 }
 950
 951                 if (!tc_in_hw(new->flags))
 952                         new->flags |= TCA_CLS_FLAGS_NOT_IN_HW;
 953
 954                 u32_replace_knode(tp, tp_c, new);
 955                 tcf_unbind_filter(tp, &n->res);
 956                 tcf_exts_get_net(&n->exts);
 957                 tcf_queue_work(&n->rwork, u32_delete_key_work);
 958                 return 0;
 959         }
 960
 961         if (tb[TCA_U32_DIVISOR]) {
 962                 unsigned int divisor = nla_get_u32(tb[TCA_U32_DIVISOR]);
 963
 964                 if (!is_power_of_2(divisor)) {
 965                         NL_SET_ERR_MSG_MOD(extack, "Divisor is not a power of 2");
 966                         return -EINVAL;
 967                 }
 968                 if (divisor-- > 0x100) {
 969                         NL_SET_ERR_MSG_MOD(extack, "Exceeded maximum 256 hash buckets");
 970                         return -EINVAL;
 971                 }
 972                 if (TC_U32_KEY(handle)) {
 973                         NL_SET_ERR_MSG_MOD(extack, "Divisor can only be used on a hash table");
 974                         return -EINVAL;
 975                 }
 976                 ht = kzalloc(struct_size(ht, ht, divisor + 1), GFP_KERNEL);
 977                 if (ht == NULL)
 978                         return -ENOBUFS;
 979                 if (handle == 0) {
 980                         handle = gen_new_htid(tp->data, ht);
 981                         if (handle == 0) {
 982                                 kfree(ht);
 983                                 return -ENOMEM;
 984                         }
 985                 } else {
 986                         err = idr_alloc_u32(&tp_c->handle_idr, ht, &handle,
 987                                             handle, GFP_KERNEL);
 988                         if (err) {
 989                                 kfree(ht);
 990                                 return err;
 991                         }
 992                 }
 993                 refcount_set(&ht->refcnt, 1);
 994                 ht->divisor = divisor;
 995                 ht->handle = handle;
 996                 ht->prio = tp->prio;
 997                 idr_init(&ht->handle_idr);
 998                 ht->flags = userflags;
 999
1000                 err = u32_replace_hw_hnode(tp, ht, userflags, extack);
1001                 if (err) {
1002                         idr_remove(&tp_c->handle_idr, handle2id(handle));
1003                         kfree(ht);
1004                         return err;
1005                 }
1006
1007                 RCU_INIT_POINTER(ht->next, tp_c->hlist);
1008                 rcu_assign_pointer(tp_c->hlist, ht);
1009                 *arg = ht;
1010
1011                 return 0;
1012         }
1013
1014         if (tb[TCA_U32_HASH]) {
1015                 htid = nla_get_u32(tb[TCA_U32_HASH]);
1016                 if (TC_U32_HTID(htid) == TC_U32_ROOT) {
1017                         ht = rtnl_dereference(tp->root);
1018                         htid = ht->handle;
1019                 } else {
1020                         ht = u32_lookup_ht(tp->data, TC_U32_HTID(htid));
1021                         if (!ht) {
1022                                 NL_SET_ERR_MSG_MOD(extack, "Specified hash table not found");
1023                                 return -EINVAL;
1024                         }
1025                 }
1026         } else {
1027                 ht = rtnl_dereference(tp->root);
1028                 htid = ht->handle;
1029         }
1030
1031         if (ht->divisor < TC_U32_HASH(htid)) {
1032                 NL_SET_ERR_MSG_MOD(extack, "Specified hash table buckets exceed configured value");
1033                 return -EINVAL;
1034         }
1035
1036         /* At this point, we need to derive the new handle that will be used to
1037          * uniquely map the identity of this table match entry. The
1038          * identity of the entry that we need to construct is 32 bits made of:
1039          *     htid(12b):bucketid(8b):node/entryid(12b)
1040          *
1041          * At this point _we have the table(ht)_ in which we will insert this
1042          * entry. We carry the table's id in variable "htid".
1043          * Note that earlier code picked the ht selection either by a) the user
1044          * providing the htid specified via TCA_U32_HASH attribute or b) when
1045          * no such attribute is passed then the root ht, is default to at ID
1046          * 0x[800][00][000]. Rule: the root table has a single bucket with ID 0.
1047          * If OTOH the user passed us the htid, they may also pass a bucketid of
1048          * choice. 0 is fine. For example a user htid is 0x[600][01][000] it is
1049          * indicating hash bucketid of 1. Rule: the entry/node ID _cannot_ be
1050          * passed via the htid, so even if it was non-zero it will be ignored.
1051          *
1052          * We may also have a handle, if the user passed one. The handle also
1053          * carries the same addressing of htid(12b):bucketid(8b):node/entryid(12b).
1054          * Rule: the bucketid on the handle is ignored even if one was passed;
1055          * rather the value on "htid" is always assumed to be the bucketid.
1056          */
1057         if (handle) {
1058                 /* Rule: The htid from handle and tableid from htid must match */
1059                 if (TC_U32_HTID(handle) && TC_U32_HTID(handle ^ htid)) {
1060                         NL_SET_ERR_MSG_MOD(extack, "Handle specified hash table address mismatch");
1061                         return -EINVAL;
1062                 }
1063                 /* Ok, so far we have a valid htid(12b):bucketid(8b) but we
1064                  * need to finalize the table entry identification with the last
1065                  * part - the node/entryid(12b)). Rule: Nodeid _cannot be 0_ for
1066                  * entries. Rule: nodeid of 0 is reserved only for tables(see
1067                  * earlier code which processes TC_U32_DIVISOR attribute).
1068                  * Rule: The nodeid can only be derived from the handle (and not
1069                  * htid).
1070                  * Rule: if the handle specified zero for the node id example
1071                  * 0x60000000, then pick a new nodeid from the pool of IDs
1072                  * this hash table has been allocating from.
1073                  * If OTOH it is specified (i.e for example the user passed a
1074                  * handle such as 0x60000123), then we use it generate our final
1075                  * handle which is used to uniquely identify the match entry.
1076                  */
1077                 if (!TC_U32_NODE(handle)) {
1078                         handle = gen_new_kid(ht, htid);
1079                 } else {
1080                         handle = htid | TC_U32_NODE(handle);
1081                         err = idr_alloc_u32(&ht->handle_idr, NULL, &handle,
1082                                             handle, GFP_KERNEL);
1083                         if (err)
1084                                 return err;
1085                 }
1086         } else {
1087                 /* The user did not give us a handle; lets just generate one
1088                  * from the table's pool of nodeids.
1089                  */
1090                 handle = gen_new_kid(ht, htid);
1091         }
1092
1093         if (tb[TCA_U32_SEL] == NULL) {
1094                 NL_SET_ERR_MSG_MOD(extack, "Selector not specified");
1095                 err = -EINVAL;
1096                 goto erridr;
1097         }
1098
1099         s = nla_data(tb[TCA_U32_SEL]);
1100         sel_size = struct_size(s, keys, s->nkeys);
1101         if (nla_len(tb[TCA_U32_SEL]) < sel_size) {
1102                 err = -EINVAL;
1103                 goto erridr;
1104         }
1105
1106         n = kzalloc(struct_size(n, sel.keys, s->nkeys), GFP_KERNEL);
1107         if (n == NULL) {
1108                 err = -ENOBUFS;
1109                 goto erridr;
1110         }
1111
1112 #ifdef CONFIG_CLS_U32_PERF
1113         n->pf = __alloc_percpu(struct_size(n->pf, kcnts, s->nkeys),
1114                                __alignof__(struct tc_u32_pcnt));
1115         if (!n->pf) {
1116                 err = -ENOBUFS;
1117                 goto errfree;
1118         }
1119 #endif
1120
1121         unsafe_memcpy(&n->sel, s, sel_size,
1122                       /* A composite flex-array structure destination,
1123                        * which was correctly sized with struct_size(),
1124                        * bounds-checked against nla_len(), and allocated
1125                        * above. */);
1126         RCU_INIT_POINTER(n->ht_up, ht);
1127         n->handle = handle;
1128         n->fshift = s->hmask ? ffs(ntohl(s->hmask)) - 1 : 0;
1129         n->flags = userflags;
1130
1131         err = tcf_exts_init(&n->exts, net, TCA_U32_ACT, TCA_U32_POLICE);
1132         if (err < 0)
1133                 goto errout;
1134
1135 #ifdef CONFIG_CLS_U32_MARK
1136         n->pcpu_success = alloc_percpu(u32);
1137         if (!n->pcpu_success) {
1138                 err = -ENOMEM;
1139                 goto errout;
1140         }
1141
1142         if (tb[TCA_U32_MARK]) {
1143                 struct tc_u32_mark *mark;
1144
1145                 mark = nla_data(tb[TCA_U32_MARK]);
1146                 n->val = mark->val;
1147                 n->mask = mark->mask;
1148         }
1149 #endif
1150
1151         err = u32_set_parms(net, tp, n, tb, tca[TCA_RATE],
1152                             flags, n->flags, extack);
1153
1154         u32_bind_filter(tp, n, base, tb);
1155
1156         if (err == 0) {
1157                 struct tc_u_knode __rcu **ins;
1158                 struct tc_u_knode *pins;
1159
1160                 err = u32_replace_hw_knode(tp, n, flags, extack);
1161                 if (err)
1162                         goto errunbind;
1163
1164                 if (!tc_in_hw(n->flags))
1165                         n->flags |= TCA_CLS_FLAGS_NOT_IN_HW;
1166
1167                 ins = &ht->ht[TC_U32_HASH(handle)];
1168                 for (pins = rtnl_dereference(*ins); pins;
1169                      ins = &pins->next, pins = rtnl_dereference(*ins))
1170                         if (TC_U32_NODE(handle) < TC_U32_NODE(pins->handle))
1171                                 break;
1172
1173                 RCU_INIT_POINTER(n->next, pins);
1174                 rcu_assign_pointer(*ins, n);
1175                 tp_c->knodes++;
1176                 *arg = n;
1177                 return 0;
1178         }
1179
1180 errunbind:
1181         u32_unbind_filter(tp, n, tb);
1182
1183 #ifdef CONFIG_CLS_U32_MARK
1184         free_percpu(n->pcpu_success);
1185 #endif
1186
1187 errout:
1188         tcf_exts_destroy(&n->exts);
1189 #ifdef CONFIG_CLS_U32_PERF
1190 errfree:
1191         free_percpu(n->pf);
1192 #endif
1193         kfree(n);
1194 erridr:
1195         idr_remove(&ht->handle_idr, handle);
1196         return err;
1197 }
1198
1199 static void u32_walk(struct tcf_proto *tp, struct tcf_walker *arg,
1200                      bool rtnl_held)
1201 {
1202         struct tc_u_common *tp_c = tp->data;
1203         struct tc_u_hnode *ht;
1204         struct tc_u_knode *n;
1205         unsigned int h;
1206
1207         if (arg->stop)
1208                 return;
1209
1210         for (ht = rtnl_dereference(tp_c->hlist);
1211              ht;
1212              ht = rtnl_dereference(ht->next)) {
1213                 if (ht->prio != tp->prio)
1214                         continue;
1215
1216                 if (!tc_cls_stats_dump(tp, arg, ht))
1217                         return;
1218
1219                 for (h = 0; h <= ht->divisor; h++) {
1220                         for (n = rtnl_dereference(ht->ht[h]);
1221                              n;
1222                              n = rtnl_dereference(n->next)) {
1223                                 if (!tc_cls_stats_dump(tp, arg, n))
1224                                         return;
1225                         }
1226                 }
1227         }
1228 }
1229
1230 static int u32_reoffload_hnode(struct tcf_proto *tp, struct tc_u_hnode *ht,
1231                                bool add, flow_setup_cb_t *cb, void *cb_priv,
1232                                struct netlink_ext_ack *extack)
1233 {
1234         struct tc_cls_u32_offload cls_u32 = {};
1235         int err;
1236
1237         tc_cls_common_offload_init(&cls_u32.common, tp, ht->flags, extack);
1238         cls_u32.command = add ? TC_CLSU32_NEW_HNODE : TC_CLSU32_DELETE_HNODE;
1239         cls_u32.hnode.divisor = ht->divisor;
1240         cls_u32.hnode.handle = ht->handle;
1241         cls_u32.hnode.prio = ht->prio;
1242
1243         err = cb(TC_SETUP_CLSU32, &cls_u32, cb_priv);
1244         if (err && add && tc_skip_sw(ht->flags))
1245                 return err;
1246
1247         return 0;
1248 }
1249
1250 static int u32_reoffload_knode(struct tcf_proto *tp, struct tc_u_knode *n,
1251                                bool add, flow_setup_cb_t *cb, void *cb_priv,
1252                                struct netlink_ext_ack *extack)
1253 {
1254         struct tc_u_hnode *ht = rtnl_dereference(n->ht_down);
1255         struct tcf_block *block = tp->chain->block;
1256         struct tc_cls_u32_offload cls_u32 = {};
1257
1258         tc_cls_common_offload_init(&cls_u32.common, tp, n->flags, extack);
1259         cls_u32.command = add ?
1260                 TC_CLSU32_REPLACE_KNODE : TC_CLSU32_DELETE_KNODE;
1261         cls_u32.knode.handle = n->handle;
1262
1263         if (add) {
1264                 cls_u32.knode.fshift = n->fshift;
1265 #ifdef CONFIG_CLS_U32_MARK
1266                 cls_u32.knode.val = n->val;
1267                 cls_u32.knode.mask = n->mask;
1268 #else
1269                 cls_u32.knode.val = 0;
1270                 cls_u32.knode.mask = 0;
1271 #endif
1272                 cls_u32.knode.sel = &n->sel;
1273                 cls_u32.knode.res = &n->res;
1274                 cls_u32.knode.exts = &n->exts;
1275                 if (n->ht_down)
1276                         cls_u32.knode.link_handle = ht->handle;
1277         }
1278
1279         return tc_setup_cb_reoffload(block, tp, add, cb, TC_SETUP_CLSU32,
1280                                      &cls_u32, cb_priv, &n->flags,
1281                                      &n->in_hw_count);
1282 }
1283
1284 static int u32_reoffload(struct tcf_proto *tp, bool add, flow_setup_cb_t *cb,
1285                          void *cb_priv, struct netlink_ext_ack *extack)
1286 {
1287         struct tc_u_common *tp_c = tp->data;
1288         struct tc_u_hnode *ht;
1289         struct tc_u_knode *n;
1290         unsigned int h;
1291         int err;
1292
1293         for (ht = rtnl_dereference(tp_c->hlist);
1294              ht;
1295              ht = rtnl_dereference(ht->next)) {
1296                 if (ht->prio != tp->prio)
1297                         continue;
1298
1299                 /* When adding filters to a new dev, try to offload the
1300                  * hashtable first. When removing, do the filters before the
1301                  * hashtable.
1302                  */
1303                 if (add && !tc_skip_hw(ht->flags)) {
1304                         err = u32_reoffload_hnode(tp, ht, add, cb, cb_priv,
1305                                                   extack);
1306                         if (err)
1307                                 return err;
1308                 }
1309
1310                 for (h = 0; h <= ht->divisor; h++) {
1311                         for (n = rtnl_dereference(ht->ht[h]);
1312                              n;
1313                              n = rtnl_dereference(n->next)) {
1314                                 if (tc_skip_hw(n->flags))
1315                                         continue;
1316
1317                                 err = u32_reoffload_knode(tp, n, add, cb,
1318                                                           cb_priv, extack);
1319                                 if (err)
1320                                         return err;
1321                         }
1322                 }
1323
1324                 if (!add && !tc_skip_hw(ht->flags))
1325                         u32_reoffload_hnode(tp, ht, add, cb, cb_priv, extack);
1326         }
1327
1328         return 0;
1329 }
1330
1331 static void u32_bind_class(void *fh, u32 classid, unsigned long cl, void *q,
1332                            unsigned long base)
1333 {
1334         struct tc_u_knode *n = fh;
1335
1336         tc_cls_bind_class(classid, cl, q, &n->res, base);
1337 }
1338
1339 static int u32_dump(struct net *net, struct tcf_proto *tp, void *fh,
1340                     struct sk_buff *skb, struct tcmsg *t, bool rtnl_held)
1341 {
1342         struct tc_u_knode *n = fh;
1343         struct tc_u_hnode *ht_up, *ht_down;
1344         struct nlattr *nest;
1345
1346         if (n == NULL)
1347                 return skb->len;
1348
1349         t->tcm_handle = n->handle;
1350
1351         nest = nla_nest_start_noflag(skb, TCA_OPTIONS);
1352         if (nest == NULL)
1353                 goto nla_put_failure;
1354
1355         if (TC_U32_KEY(n->handle) == 0) {
1356                 struct tc_u_hnode *ht = fh;
1357                 u32 divisor = ht->divisor + 1;
1358
1359                 if (nla_put_u32(skb, TCA_U32_DIVISOR, divisor))
1360                         goto nla_put_failure;
1361         } else {
1362 #ifdef CONFIG_CLS_U32_PERF
1363                 struct tc_u32_pcnt *gpf;
1364                 int cpu;
1365 #endif
1366
1367                 if (nla_put(skb, TCA_U32_SEL, struct_size(&n->sel, keys, n->sel.nkeys),
1368                             &n->sel))
1369                         goto nla_put_failure;
1370
1371                 ht_up = rtnl_dereference(n->ht_up);
1372                 if (ht_up) {
1373                         u32 htid = n->handle & 0xFFFFF000;
1374                         if (nla_put_u32(skb, TCA_U32_HASH, htid))
1375                                 goto nla_put_failure;
1376                 }
1377                 if (n->res.classid &&
1378                     nla_put_u32(skb, TCA_U32_CLASSID, n->res.classid))
1379                         goto nla_put_failure;
1380
1381                 ht_down = rtnl_dereference(n->ht_down);
1382                 if (ht_down &&
1383                     nla_put_u32(skb, TCA_U32_LINK, ht_down->handle))
1384                         goto nla_put_failure;
1385
1386                 if (n->flags && nla_put_u32(skb, TCA_U32_FLAGS, n->flags))
1387                         goto nla_put_failure;
1388
1389 #ifdef CONFIG_CLS_U32_MARK
1390                 if ((n->val || n->mask)) {
1391                         struct tc_u32_mark mark = {.val = n->val,
1392                                                    .mask = n->mask,
1393                                                    .success = 0};
1394                         int cpum;
1395
1396                         for_each_possible_cpu(cpum) {
1397                                 __u32 cnt = *per_cpu_ptr(n->pcpu_success, cpum);
1398
1399                                 mark.success += cnt;
1400                         }
1401
1402                         if (nla_put(skb, TCA_U32_MARK, sizeof(mark), &mark))
1403                                 goto nla_put_failure;
1404                 }
1405 #endif
1406
1407                 if (tcf_exts_dump(skb, &n->exts) < 0)
1408                         goto nla_put_failure;
1409
1410                 if (n->ifindex) {
1411                         struct net_device *dev;
1412                         dev = __dev_get_by_index(net, n->ifindex);
1413                         if (dev && nla_put_string(skb, TCA_U32_INDEV, dev->name))
1414                                 goto nla_put_failure;
1415                 }
1416 #ifdef CONFIG_CLS_U32_PERF
1417                 gpf = kzalloc(struct_size(gpf, kcnts, n->sel.nkeys), GFP_KERNEL);
1418                 if (!gpf)
1419                         goto nla_put_failure;
1420
1421                 for_each_possible_cpu(cpu) {
1422                         int i;
1423                         struct tc_u32_pcnt *pf = per_cpu_ptr(n->pf, cpu);
1424
1425                         gpf->rcnt += pf->rcnt;
1426                         gpf->rhit += pf->rhit;
1427                         for (i = 0; i < n->sel.nkeys; i++)
1428                                 gpf->kcnts[i] += pf->kcnts[i];
1429                 }
1430
1431                 if (nla_put_64bit(skb, TCA_U32_PCNT, struct_size(gpf, kcnts, n->sel.nkeys),
1432                                   gpf, TCA_U32_PAD)) {
1433                         kfree(gpf);
1434                         goto nla_put_failure;
1435                 }
1436                 kfree(gpf);
1437 #endif
1438         }
1439
1440         nla_nest_end(skb, nest);
1441
1442         if (TC_U32_KEY(n->handle))
1443                 if (tcf_exts_dump_stats(skb, &n->exts) < 0)
1444                         goto nla_put_failure;
1445         return skb->len;
1446
1447 nla_put_failure:
1448         nla_nest_cancel(skb, nest);
1449         return -1;
1450 }
1451
1452 static struct tcf_proto_ops cls_u32_ops __read_mostly = {
1453         .kind           =       "u32",
1454         .classify       =       u32_classify,
1455         .init           =       u32_init,
1456         .destroy        =       u32_destroy,
1457         .get            =       u32_get,
1458         .change         =       u32_change,
1459         .delete         =       u32_delete,
1460         .walk           =       u32_walk,
1461         .reoffload      =       u32_reoffload,
1462         .dump           =       u32_dump,
1463         .bind_class     =       u32_bind_class,
1464         .owner          =       THIS_MODULE,
1465 };
1466 MODULE_ALIAS_NET_CLS("u32");
1467
1468 static int __init init_u32(void)
1469 {
1470         int i, ret;
1471
1472         pr_info("u32 classifier\n");
1473 #ifdef CONFIG_CLS_U32_PERF
1474         pr_info("    Performance counters on\n");
1475 #endif
1476         pr_info("    input device check on\n");
1477 #ifdef CONFIG_NET_CLS_ACT
1478         pr_info("    Actions configured\n");
1479 #endif
1480         tc_u_common_hash = kvmalloc_array(U32_HASH_SIZE,
1481                                           sizeof(struct hlist_head),
1482                                           GFP_KERNEL);
1483         if (!tc_u_common_hash)
1484                 return -ENOMEM;
1485
1486         for (i = 0; i < U32_HASH_SIZE; i++)
1487                 INIT_HLIST_HEAD(&tc_u_common_hash[i]);
1488
1489         ret = register_tcf_proto_ops(&cls_u32_ops);
1490         if (ret)
1491                 kvfree(tc_u_common_hash);
1492         return ret;
1493 }
1494
1495 static void __exit exit_u32(void)
1496 {
1497         unregister_tcf_proto_ops(&cls_u32_ops);
1498         kvfree(tc_u_common_hash);
1499 }
1500
1501 module_init(init_u32)
1502 module_exit(exit_u32)
1503 MODULE_DESCRIPTION("Universal 32bit based TC Classifier");
1504 MODULE_LICENSE("GPL");