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, <kuznet@ms2.inr.ac.ru>
   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         int                     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         int                     refcnt;
  90         struct idr              handle_idr;
  91         struct hlist_node       hnode;
  92         long                    knodes;
  93 };
  94
  95 static inline unsigned int u32_hash_fold(__be32 key,
  96                                          const struct tc_u32_sel *sel,
  97                                          u8 fshift)
  98 {
  99         unsigned int h = ntohl(key & sel->hmask) >> fshift;
 100
 101         return h;
 102 }
 103
 104 TC_INDIRECT_SCOPE int u32_classify(struct sk_buff *skb,
 105                                    const struct tcf_proto *tp,
 106                                    struct tcf_result *res)
 107 {
 108         struct {
 109                 struct tc_u_knode *knode;
 110                 unsigned int      off;
 111         } stack[TC_U32_MAXDEPTH];
 112
 113         struct tc_u_hnode *ht = rcu_dereference_bh(tp->root);
 114         unsigned int off = skb_network_offset(skb);
 115         struct tc_u_knode *n;
 116         int sdepth = 0;
 117         int off2 = 0;
 118         int sel = 0;
 119 #ifdef CONFIG_CLS_U32_PERF
 120         int j;
 121 #endif
 122         int i, r;
 123
 124 next_ht:
 125         n = rcu_dereference_bh(ht->ht[sel]);
 126
 127 next_knode:
 128         if (n) {
 129                 struct tc_u32_key *key = n->sel.keys;
 130
 131 #ifdef CONFIG_CLS_U32_PERF
 132                 __this_cpu_inc(n->pf->rcnt);
 133                 j = 0;
 134 #endif
 135
 136                 if (tc_skip_sw(n->flags)) {
 137                         n = rcu_dereference_bh(n->next);
 138                         goto next_knode;
 139                 }
 140
 141 #ifdef CONFIG_CLS_U32_MARK
 142                 if ((skb->mark & n->mask) != n->val) {
 143                         n = rcu_dereference_bh(n->next);
 144                         goto next_knode;
 145                 } else {
 146                         __this_cpu_inc(*n->pcpu_success);
 147                 }
 148 #endif
 149
 150                 for (i = n->sel.nkeys; i > 0; i--, key++) {
 151                         int toff = off + key->off + (off2 & key->offmask);
 152                         __be32 *data, hdata;
 153
 154                         if (skb_headroom(skb) + toff > INT_MAX)
 155                                 goto out;
 156
 157                         data = skb_header_pointer(skb, toff, 4, &hdata);
 158                         if (!data)
 159                                 goto out;
 160                         if ((*data ^ key->val) & key->mask) {
 161                                 n = rcu_dereference_bh(n->next);
 162                                 goto next_knode;
 163                         }
 164 #ifdef CONFIG_CLS_U32_PERF
 165                         __this_cpu_inc(n->pf->kcnts[j]);
 166                         j++;
 167 #endif
 168                 }
 169
 170                 ht = rcu_dereference_bh(n->ht_down);
 171                 if (!ht) {
 172 check_terminal:
 173                         if (n->sel.flags & TC_U32_TERMINAL) {
 174
 175                                 *res = n->res;
 176                                 if (!tcf_match_indev(skb, n->ifindex)) {
 177                                         n = rcu_dereference_bh(n->next);
 178                                         goto next_knode;
 179                                 }
 180 #ifdef CONFIG_CLS_U32_PERF
 181                                 __this_cpu_inc(n->pf->rhit);
 182 #endif
 183                                 r = tcf_exts_exec(skb, &n->exts, res);
 184                                 if (r < 0) {
 185                                         n = rcu_dereference_bh(n->next);
 186                                         goto next_knode;
 187                                 }
 188
 189                                 return r;
 190                         }
 191                         n = rcu_dereference_bh(n->next);
 192                         goto next_knode;
 193                 }
 194
 195                 /* PUSH */
 196                 if (sdepth >= TC_U32_MAXDEPTH)
 197                         goto deadloop;
 198                 stack[sdepth].knode = n;
 199                 stack[sdepth].off = off;
 200                 sdepth++;
 201
 202                 ht = rcu_dereference_bh(n->ht_down);
 203                 sel = 0;
 204                 if (ht->divisor) {
 205                         __be32 *data, hdata;
 206
 207                         data = skb_header_pointer(skb, off + n->sel.hoff, 4,
 208                                                   &hdata);
 209                         if (!data)
 210                                 goto out;
 211                         sel = ht->divisor & u32_hash_fold(*data, &n->sel,
 212                                                           n->fshift);
 213                 }
 214                 if (!(n->sel.flags & (TC_U32_VAROFFSET | TC_U32_OFFSET | TC_U32_EAT)))
 215                         goto next_ht;
 216
 217                 if (n->sel.flags & (TC_U32_OFFSET | TC_U32_VAROFFSET)) {
 218                         off2 = n->sel.off + 3;
 219                         if (n->sel.flags & TC_U32_VAROFFSET) {
 220                                 __be16 *data, hdata;
 221
 222                                 data = skb_header_pointer(skb,
 223                                                           off + n->sel.offoff,
 224                                                           2, &hdata);
 225                                 if (!data)
 226                                         goto out;
 227                                 off2 += ntohs(n->sel.offmask & *data) >>
 228                                         n->sel.offshift;
 229                         }
 230                         off2 &= ~3;
 231                 }
 232                 if (n->sel.flags & TC_U32_EAT) {
 233                         off += off2;
 234                         off2 = 0;
 235                 }
 236
 237                 if (off < skb->len)
 238                         goto next_ht;
 239         }
 240
 241         /* POP */
 242         if (sdepth--) {
 243                 n = stack[sdepth].knode;
 244                 ht = rcu_dereference_bh(n->ht_up);
 245                 off = stack[sdepth].off;
 246                 goto check_terminal;
 247         }
 248 out:
 249         return -1;
 250
 251 deadloop:
 252         net_warn_ratelimited("cls_u32: dead loop\n");
 253         return -1;
 254 }
 255
 256 static struct tc_u_hnode *u32_lookup_ht(struct tc_u_common *tp_c, u32 handle)
 257 {
 258         struct tc_u_hnode *ht;
 259
 260         for (ht = rtnl_dereference(tp_c->hlist);
 261              ht;
 262              ht = rtnl_dereference(ht->next))
 263                 if (ht->handle == handle)
 264                         break;
 265
 266         return ht;
 267 }
 268
 269 static struct tc_u_knode *u32_lookup_key(struct tc_u_hnode *ht, u32 handle)
 270 {
 271         unsigned int sel;
 272         struct tc_u_knode *n = NULL;
 273
 274         sel = TC_U32_HASH(handle);
 275         if (sel > ht->divisor)
 276                 goto out;
 277
 278         for (n = rtnl_dereference(ht->ht[sel]);
 279              n;
 280              n = rtnl_dereference(n->next))
 281                 if (n->handle == handle)
 282                         break;
 283 out:
 284         return n;
 285 }
 286
 287
 288 static void *u32_get(struct tcf_proto *tp, u32 handle)
 289 {
 290         struct tc_u_hnode *ht;
 291         struct tc_u_common *tp_c = tp->data;
 292
 293         if (TC_U32_HTID(handle) == TC_U32_ROOT)
 294                 ht = rtnl_dereference(tp->root);
 295         else
 296                 ht = u32_lookup_ht(tp_c, TC_U32_HTID(handle));
 297
 298         if (!ht)
 299                 return NULL;
 300
 301         if (TC_U32_KEY(handle) == 0)
 302                 return ht;
 303
 304         return u32_lookup_key(ht, handle);
 305 }
 306
 307 /* Protected by rtnl lock */
 308 static u32 gen_new_htid(struct tc_u_common *tp_c, struct tc_u_hnode *ptr)
 309 {
 310         int id = idr_alloc_cyclic(&tp_c->handle_idr, ptr, 1, 0x7FF, GFP_KERNEL);
 311         if (id < 0)
 312                 return 0;
 313         return (id | 0x800U) << 20;
 314 }
 315
 316 static struct hlist_head *tc_u_common_hash;
 317
 318 #define U32_HASH_SHIFT 10
 319 #define U32_HASH_SIZE (1 << U32_HASH_SHIFT)
 320
 321 static void *tc_u_common_ptr(const struct tcf_proto *tp)
 322 {
 323         struct tcf_block *block = tp->chain->block;
 324
 325         /* The block sharing is currently supported only
 326          * for classless qdiscs. In that case we use block
 327          * for tc_u_common identification. In case the
 328          * block is not shared, block->q is a valid pointer
 329          * and we can use that. That works for classful qdiscs.
 330          */
 331         if (tcf_block_shared(block))
 332                 return block;
 333         else
 334                 return block->q;
 335 }
 336
 337 static struct hlist_head *tc_u_hash(void *key)
 338 {
 339         return tc_u_common_hash + hash_ptr(key, U32_HASH_SHIFT);
 340 }
 341
 342 static struct tc_u_common *tc_u_common_find(void *key)
 343 {
 344         struct tc_u_common *tc;
 345         hlist_for_each_entry(tc, tc_u_hash(key), hnode) {
 346                 if (tc->ptr == key)
 347                         return tc;
 348         }
 349         return NULL;
 350 }
 351
 352 static int u32_init(struct tcf_proto *tp)
 353 {
 354         struct tc_u_hnode *root_ht;
 355         void *key = tc_u_common_ptr(tp);
 356         struct tc_u_common *tp_c = tc_u_common_find(key);
 357
 358         root_ht = kzalloc(struct_size(root_ht, ht, 1), GFP_KERNEL);
 359         if (root_ht == NULL)
 360                 return -ENOBUFS;
 361
 362         root_ht->refcnt++;
 363         root_ht->handle = tp_c ? gen_new_htid(tp_c, root_ht) : 0x80000000;
 364         root_ht->prio = tp->prio;
 365         root_ht->is_root = true;
 366         idr_init(&root_ht->handle_idr);
 367
 368         if (tp_c == NULL) {
 369                 tp_c = kzalloc(struct_size(tp_c, hlist->ht, 1), GFP_KERNEL);
 370                 if (tp_c == NULL) {
 371                         kfree(root_ht);
 372                         return -ENOBUFS;
 373                 }
 374                 tp_c->ptr = key;
 375                 INIT_HLIST_NODE(&tp_c->hnode);
 376                 idr_init(&tp_c->handle_idr);
 377
 378                 hlist_add_head(&tp_c->hnode, tc_u_hash(key));
 379         }
 380
 381         tp_c->refcnt++;
 382         RCU_INIT_POINTER(root_ht->next, tp_c->hlist);
 383         rcu_assign_pointer(tp_c->hlist, root_ht);
 384
 385         root_ht->refcnt++;
 386         rcu_assign_pointer(tp->root, root_ht);
 387         tp->data = tp_c;
 388         return 0;
 389 }
 390
 391 static void __u32_destroy_key(struct tc_u_knode *n)
 392 {
 393         struct tc_u_hnode *ht = rtnl_dereference(n->ht_down);
 394
 395         tcf_exts_destroy(&n->exts);
 396         if (ht && --ht->refcnt == 0)
 397                 kfree(ht);
 398         kfree(n);
 399 }
 400
 401 static void u32_destroy_key(struct tc_u_knode *n, bool free_pf)
 402 {
 403         tcf_exts_put_net(&n->exts);
 404 #ifdef CONFIG_CLS_U32_PERF
 405         if (free_pf)
 406                 free_percpu(n->pf);
 407 #endif
 408 #ifdef CONFIG_CLS_U32_MARK
 409         if (free_pf)
 410                 free_percpu(n->pcpu_success);
 411 #endif
 412         __u32_destroy_key(n);
 413 }
 414
 415 /* u32_delete_key_rcu should be called when free'ing a copied
 416  * version of a tc_u_knode obtained from u32_init_knode(). When
 417  * copies are obtained from u32_init_knode() the statistics are
 418  * shared between the old and new copies to allow readers to
 419  * continue to update the statistics during the copy. To support
 420  * this the u32_delete_key_rcu variant does not free the percpu
 421  * statistics.
 422  */
 423 static void u32_delete_key_work(struct work_struct *work)
 424 {
 425         struct tc_u_knode *key = container_of(to_rcu_work(work),
 426                                               struct tc_u_knode,
 427                                               rwork);
 428         rtnl_lock();
 429         u32_destroy_key(key, false);
 430         rtnl_unlock();
 431 }
 432
 433 /* u32_delete_key_freepf_rcu is the rcu callback variant
 434  * that free's the entire structure including the statistics
 435  * percpu variables. Only use this if the key is not a copy
 436  * returned by u32_init_knode(). See u32_delete_key_rcu()
 437  * for the variant that should be used with keys return from
 438  * u32_init_knode()
 439  */
 440 static void u32_delete_key_freepf_work(struct work_struct *work)
 441 {
 442         struct tc_u_knode *key = container_of(to_rcu_work(work),
 443                                               struct tc_u_knode,
 444                                               rwork);
 445         rtnl_lock();
 446         u32_destroy_key(key, true);
 447         rtnl_unlock();
 448 }
 449
 450 static int u32_delete_key(struct tcf_proto *tp, struct tc_u_knode *key)
 451 {
 452         struct tc_u_common *tp_c = tp->data;
 453         struct tc_u_knode __rcu **kp;
 454         struct tc_u_knode *pkp;
 455         struct tc_u_hnode *ht = rtnl_dereference(key->ht_up);
 456
 457         if (ht) {
 458                 kp = &ht->ht[TC_U32_HASH(key->handle)];
 459                 for (pkp = rtnl_dereference(*kp); pkp;
 460                      kp = &pkp->next, pkp = rtnl_dereference(*kp)) {
 461                         if (pkp == key) {
 462                                 RCU_INIT_POINTER(*kp, key->next);
 463                                 tp_c->knodes--;
 464
 465                                 tcf_unbind_filter(tp, &key->res);
 466                                 idr_remove(&ht->handle_idr, key->handle);
 467                                 tcf_exts_get_net(&key->exts);
 468                                 tcf_queue_work(&key->rwork, u32_delete_key_freepf_work);
 469                                 return 0;
 470                         }
 471                 }
 472         }
 473         WARN_ON(1);
 474         return 0;
 475 }
 476
 477 static void u32_clear_hw_hnode(struct tcf_proto *tp, struct tc_u_hnode *h,
 478                                struct netlink_ext_ack *extack)
 479 {
 480         struct tcf_block *block = tp->chain->block;
 481         struct tc_cls_u32_offload cls_u32 = {};
 482
 483         tc_cls_common_offload_init(&cls_u32.common, tp, h->flags, extack);
 484         cls_u32.command = TC_CLSU32_DELETE_HNODE;
 485         cls_u32.hnode.divisor = h->divisor;
 486         cls_u32.hnode.handle = h->handle;
 487         cls_u32.hnode.prio = h->prio;
 488
 489         tc_setup_cb_call(block, TC_SETUP_CLSU32, &cls_u32, false, true);
 490 }
 491
 492 static int u32_replace_hw_hnode(struct tcf_proto *tp, struct tc_u_hnode *h,
 493                                 u32 flags, struct netlink_ext_ack *extack)
 494 {
 495         struct tcf_block *block = tp->chain->block;
 496         struct tc_cls_u32_offload cls_u32 = {};
 497         bool skip_sw = tc_skip_sw(flags);
 498         bool offloaded = false;
 499         int err;
 500
 501         tc_cls_common_offload_init(&cls_u32.common, tp, flags, extack);
 502         cls_u32.command = TC_CLSU32_NEW_HNODE;
 503         cls_u32.hnode.divisor = h->divisor;
 504         cls_u32.hnode.handle = h->handle;
 505         cls_u32.hnode.prio = h->prio;
 506
 507         err = tc_setup_cb_call(block, TC_SETUP_CLSU32, &cls_u32, skip_sw, true);
 508         if (err < 0) {
 509                 u32_clear_hw_hnode(tp, h, NULL);
 510                 return err;
 511         } else if (err > 0) {
 512                 offloaded = true;
 513         }
 514
 515         if (skip_sw && !offloaded)
 516                 return -EINVAL;
 517
 518         return 0;
 519 }
 520
 521 static void u32_remove_hw_knode(struct tcf_proto *tp, struct tc_u_knode *n,
 522                                 struct netlink_ext_ack *extack)
 523 {
 524         struct tcf_block *block = tp->chain->block;
 525         struct tc_cls_u32_offload cls_u32 = {};
 526
 527         tc_cls_common_offload_init(&cls_u32.common, tp, n->flags, extack);
 528         cls_u32.command = TC_CLSU32_DELETE_KNODE;
 529         cls_u32.knode.handle = n->handle;
 530
 531         tc_setup_cb_destroy(block, tp, TC_SETUP_CLSU32, &cls_u32, false,
 532                             &n->flags, &n->in_hw_count, true);
 533 }
 534
 535 static int u32_replace_hw_knode(struct tcf_proto *tp, struct tc_u_knode *n,
 536                                 u32 flags, struct netlink_ext_ack *extack)
 537 {
 538         struct tc_u_hnode *ht = rtnl_dereference(n->ht_down);
 539         struct tcf_block *block = tp->chain->block;
 540         struct tc_cls_u32_offload cls_u32 = {};
 541         bool skip_sw = tc_skip_sw(flags);
 542         int err;
 543
 544         tc_cls_common_offload_init(&cls_u32.common, tp, flags, extack);
 545         cls_u32.command = TC_CLSU32_REPLACE_KNODE;
 546         cls_u32.knode.handle = n->handle;
 547         cls_u32.knode.fshift = n->fshift;
 548 #ifdef CONFIG_CLS_U32_MARK
 549         cls_u32.knode.val = n->val;
 550         cls_u32.knode.mask = n->mask;
 551 #else
 552         cls_u32.knode.val = 0;
 553         cls_u32.knode.mask = 0;
 554 #endif
 555         cls_u32.knode.sel = &n->sel;
 556         cls_u32.knode.res = &n->res;
 557         cls_u32.knode.exts = &n->exts;
 558         if (n->ht_down)
 559                 cls_u32.knode.link_handle = ht->handle;
 560
 561         err = tc_setup_cb_add(block, tp, TC_SETUP_CLSU32, &cls_u32, skip_sw,
 562                               &n->flags, &n->in_hw_count, true);
 563         if (err) {
 564                 u32_remove_hw_knode(tp, n, NULL);
 565                 return err;
 566         }
 567
 568         if (skip_sw && !(n->flags & TCA_CLS_FLAGS_IN_HW))
 569                 return -EINVAL;
 570
 571         return 0;
 572 }
 573
 574 static void u32_clear_hnode(struct tcf_proto *tp, struct tc_u_hnode *ht,
 575                             struct netlink_ext_ack *extack)
 576 {
 577         struct tc_u_common *tp_c = tp->data;
 578         struct tc_u_knode *n;
 579         unsigned int h;
 580
 581         for (h = 0; h <= ht->divisor; h++) {
 582                 while ((n = rtnl_dereference(ht->ht[h])) != NULL) {
 583                         RCU_INIT_POINTER(ht->ht[h],
 584                                          rtnl_dereference(n->next));
 585                         tp_c->knodes--;
 586                         tcf_unbind_filter(tp, &n->res);
 587                         u32_remove_hw_knode(tp, n, extack);
 588                         idr_remove(&ht->handle_idr, n->handle);
 589                         if (tcf_exts_get_net(&n->exts))
 590                                 tcf_queue_work(&n->rwork, u32_delete_key_freepf_work);
 591                         else
 592                                 u32_destroy_key(n, true);
 593                 }
 594         }
 595 }
 596
 597 static int u32_destroy_hnode(struct tcf_proto *tp, struct tc_u_hnode *ht,
 598                              struct netlink_ext_ack *extack)
 599 {
 600         struct tc_u_common *tp_c = tp->data;
 601         struct tc_u_hnode __rcu **hn;
 602         struct tc_u_hnode *phn;
 603
 604         WARN_ON(--ht->refcnt);
 605
 606         u32_clear_hnode(tp, ht, extack);
 607
 608         hn = &tp_c->hlist;
 609         for (phn = rtnl_dereference(*hn);
 610              phn;
 611              hn = &phn->next, phn = rtnl_dereference(*hn)) {
 612                 if (phn == ht) {
 613                         u32_clear_hw_hnode(tp, ht, extack);
 614                         idr_destroy(&ht->handle_idr);
 615                         idr_remove(&tp_c->handle_idr, ht->handle);
 616                         RCU_INIT_POINTER(*hn, ht->next);
 617                         kfree_rcu(ht, rcu);
 618                         return 0;
 619                 }
 620         }
 621
 622         return -ENOENT;
 623 }
 624
 625 static void u32_destroy(struct tcf_proto *tp, bool rtnl_held,
 626                         struct netlink_ext_ack *extack)
 627 {
 628         struct tc_u_common *tp_c = tp->data;
 629         struct tc_u_hnode *root_ht = rtnl_dereference(tp->root);
 630
 631         WARN_ON(root_ht == NULL);
 632
 633         if (root_ht && --root_ht->refcnt == 1)
 634                 u32_destroy_hnode(tp, root_ht, extack);
 635
 636         if (--tp_c->refcnt == 0) {
 637                 struct tc_u_hnode *ht;
 638
 639                 hlist_del(&tp_c->hnode);
 640
 641                 while ((ht = rtnl_dereference(tp_c->hlist)) != NULL) {
 642                         u32_clear_hnode(tp, ht, extack);
 643                         RCU_INIT_POINTER(tp_c->hlist, ht->next);
 644
 645                         /* u32_destroy_key() will later free ht for us, if it's
 646                          * still referenced by some knode
 647                          */
 648                         if (--ht->refcnt == 0)
 649                                 kfree_rcu(ht, rcu);
 650                 }
 651
 652                 idr_destroy(&tp_c->handle_idr);
 653                 kfree(tp_c);
 654         }
 655
 656         tp->data = NULL;
 657 }
 658
 659 static int u32_delete(struct tcf_proto *tp, void *arg, bool *last,
 660                       bool rtnl_held, struct netlink_ext_ack *extack)
 661 {
 662         struct tc_u_hnode *ht = arg;
 663         struct tc_u_common *tp_c = tp->data;
 664         int ret = 0;
 665
 666         if (TC_U32_KEY(ht->handle)) {
 667                 u32_remove_hw_knode(tp, (struct tc_u_knode *)ht, extack);
 668                 ret = u32_delete_key(tp, (struct tc_u_knode *)ht);
 669                 goto out;
 670         }
 671
 672         if (ht->is_root) {
 673                 NL_SET_ERR_MSG_MOD(extack, "Not allowed to delete root node");
 674                 return -EINVAL;
 675         }
 676
 677         if (ht->refcnt == 1) {
 678                 u32_destroy_hnode(tp, ht, extack);
 679         } else {
 680                 NL_SET_ERR_MSG_MOD(extack, "Can not delete in-use filter");
 681                 return -EBUSY;
 682         }
 683
 684 out:
 685         *last = tp_c->refcnt == 1 && tp_c->knodes == 0;
 686         return ret;
 687 }
 688
 689 static u32 gen_new_kid(struct tc_u_hnode *ht, u32 htid)
 690 {
 691         u32 index = htid | 0x800;
 692         u32 max = htid | 0xFFF;
 693
 694         if (idr_alloc_u32(&ht->handle_idr, NULL, &index, max, GFP_KERNEL)) {
 695                 index = htid + 1;
 696                 if (idr_alloc_u32(&ht->handle_idr, NULL, &index, max,
 697                                  GFP_KERNEL))
 698                         index = max;
 699         }
 700
 701         return index;
 702 }
 703
 704 static const struct nla_policy u32_policy[TCA_U32_MAX + 1] = {
 705         [TCA_U32_CLASSID]       = { .type = NLA_U32 },
 706         [TCA_U32_HASH]          = { .type = NLA_U32 },
 707         [TCA_U32_LINK]          = { .type = NLA_U32 },
 708         [TCA_U32_DIVISOR]       = { .type = NLA_U32 },
 709         [TCA_U32_SEL]           = { .len = sizeof(struct tc_u32_sel) },
 710         [TCA_U32_INDEV]         = { .type = NLA_STRING, .len = IFNAMSIZ },
 711         [TCA_U32_MARK]          = { .len = sizeof(struct tc_u32_mark) },
 712         [TCA_U32_FLAGS]         = { .type = NLA_U32 },
 713 };
 714
 715 static void u32_unbind_filter(struct tcf_proto *tp, struct tc_u_knode *n,
 716                               struct nlattr **tb)
 717 {
 718         if (tb[TCA_U32_CLASSID])
 719                 tcf_unbind_filter(tp, &n->res);
 720 }
 721
 722 static void u32_bind_filter(struct tcf_proto *tp, struct tc_u_knode *n,
 723                             unsigned long base, struct nlattr **tb)
 724 {
 725         if (tb[TCA_U32_CLASSID]) {
 726                 n->res.classid = nla_get_u32(tb[TCA_U32_CLASSID]);
 727                 tcf_bind_filter(tp, &n->res, base);
 728         }
 729 }
 730
 731 static int u32_set_parms(struct net *net, struct tcf_proto *tp,
 732                          struct tc_u_knode *n, struct nlattr **tb,
 733                          struct nlattr *est, u32 flags, u32 fl_flags,
 734                          struct netlink_ext_ack *extack)
 735 {
 736         int err, ifindex = -1;
 737
 738         err = tcf_exts_validate_ex(net, tp, tb, est, &n->exts, flags,
 739                                    fl_flags, extack);
 740         if (err < 0)
 741                 return err;
 742
 743         if (tb[TCA_U32_INDEV]) {
 744                 ifindex = tcf_change_indev(net, tb[TCA_U32_INDEV], extack);
 745                 if (ifindex < 0)
 746                         return -EINVAL;
 747         }
 748
 749         if (tb[TCA_U32_LINK]) {
 750                 u32 handle = nla_get_u32(tb[TCA_U32_LINK]);
 751                 struct tc_u_hnode *ht_down = NULL, *ht_old;
 752
 753                 if (TC_U32_KEY(handle)) {
 754                         NL_SET_ERR_MSG_MOD(extack, "u32 Link handle must be a hash table");
 755                         return -EINVAL;
 756                 }
 757
 758                 if (handle) {
 759                         ht_down = u32_lookup_ht(tp->data, handle);
 760
 761                         if (!ht_down) {
 762                                 NL_SET_ERR_MSG_MOD(extack, "Link hash table not found");
 763                                 return -EINVAL;
 764                         }
 765                         if (ht_down->is_root) {
 766                                 NL_SET_ERR_MSG_MOD(extack, "Not linking to root node");
 767                                 return -EINVAL;
 768                         }
 769                         ht_down->refcnt++;
 770                 }
 771
 772                 ht_old = rtnl_dereference(n->ht_down);
 773                 rcu_assign_pointer(n->ht_down, ht_down);
 774
 775                 if (ht_old)
 776                         ht_old->refcnt--;
 777         }
 778
 779         if (ifindex >= 0)
 780                 n->ifindex = ifindex;
 781
 782         return 0;
 783 }
 784
 785 static void u32_replace_knode(struct tcf_proto *tp, struct tc_u_common *tp_c,
 786                               struct tc_u_knode *n)
 787 {
 788         struct tc_u_knode __rcu **ins;
 789         struct tc_u_knode *pins;
 790         struct tc_u_hnode *ht;
 791
 792         if (TC_U32_HTID(n->handle) == TC_U32_ROOT)
 793                 ht = rtnl_dereference(tp->root);
 794         else
 795                 ht = u32_lookup_ht(tp_c, TC_U32_HTID(n->handle));
 796
 797         ins = &ht->ht[TC_U32_HASH(n->handle)];
 798
 799         /* The node must always exist for it to be replaced if this is not the
 800          * case then something went very wrong elsewhere.
 801          */
 802         for (pins = rtnl_dereference(*ins); ;
 803              ins = &pins->next, pins = rtnl_dereference(*ins))
 804                 if (pins->handle == n->handle)
 805                         break;
 806
 807         idr_replace(&ht->handle_idr, n, n->handle);
 808         RCU_INIT_POINTER(n->next, pins->next);
 809         rcu_assign_pointer(*ins, n);
 810 }
 811
 812 static struct tc_u_knode *u32_init_knode(struct net *net, struct tcf_proto *tp,
 813                                          struct tc_u_knode *n)
 814 {
 815         struct tc_u_hnode *ht = rtnl_dereference(n->ht_down);
 816         struct tc_u32_sel *s = &n->sel;
 817         struct tc_u_knode *new;
 818
 819         new = kzalloc(struct_size(new, sel.keys, s->nkeys), GFP_KERNEL);
 820         if (!new)
 821                 return NULL;
 822
 823         RCU_INIT_POINTER(new->next, n->next);
 824         new->handle = n->handle;
 825         RCU_INIT_POINTER(new->ht_up, n->ht_up);
 826
 827         new->ifindex = n->ifindex;
 828         new->fshift = n->fshift;
 829         new->flags = n->flags;
 830         RCU_INIT_POINTER(new->ht_down, ht);
 831
 832 #ifdef CONFIG_CLS_U32_PERF
 833         /* Statistics may be incremented by readers during update
 834          * so we must keep them in tact. When the node is later destroyed
 835          * a special destroy call must be made to not free the pf memory.
 836          */
 837         new->pf = n->pf;
 838 #endif
 839
 840 #ifdef CONFIG_CLS_U32_MARK
 841         new->val = n->val;
 842         new->mask = n->mask;
 843         /* Similarly success statistics must be moved as pointers */
 844         new->pcpu_success = n->pcpu_success;
 845 #endif
 846         memcpy(&new->sel, s, struct_size(s, keys, s->nkeys));
 847
 848         if (tcf_exts_init(&new->exts, net, TCA_U32_ACT, TCA_U32_POLICE)) {
 849                 kfree(new);
 850                 return NULL;
 851         }
 852
 853         /* bump reference count as long as we hold pointer to structure */
 854         if (ht)
 855                 ht->refcnt++;
 856
 857         return new;
 858 }
 859
 860 static int u32_change(struct net *net, struct sk_buff *in_skb,
 861                       struct tcf_proto *tp, unsigned long base, u32 handle,
 862                       struct nlattr **tca, void **arg, u32 flags,
 863                       struct netlink_ext_ack *extack)
 864 {
 865         struct tc_u_common *tp_c = tp->data;
 866         struct tc_u_hnode *ht;
 867         struct tc_u_knode *n;
 868         struct tc_u32_sel *s;
 869         struct nlattr *opt = tca[TCA_OPTIONS];
 870         struct nlattr *tb[TCA_U32_MAX + 1];
 871         u32 htid, userflags = 0;
 872         size_t sel_size;
 873         int err;
 874
 875         if (!opt) {
 876                 if (handle) {
 877                         NL_SET_ERR_MSG_MOD(extack, "Filter handle requires options");
 878                         return -EINVAL;
 879                 } else {
 880                         return 0;
 881                 }
 882         }
 883
 884         err = nla_parse_nested_deprecated(tb, TCA_U32_MAX, opt, u32_policy,
 885                                           extack);
 886         if (err < 0)
 887                 return err;
 888
 889         if (tb[TCA_U32_FLAGS]) {
 890                 userflags = nla_get_u32(tb[TCA_U32_FLAGS]);
 891                 if (!tc_flags_valid(userflags)) {
 892                         NL_SET_ERR_MSG_MOD(extack, "Invalid filter flags");
 893                         return -EINVAL;
 894                 }
 895         }
 896
 897         n = *arg;
 898         if (n) {
 899                 struct tc_u_knode *new;
 900
 901                 if (TC_U32_KEY(n->handle) == 0) {
 902                         NL_SET_ERR_MSG_MOD(extack, "Key node id cannot be zero");
 903                         return -EINVAL;
 904                 }
 905
 906                 if ((n->flags ^ userflags) &
 907                     ~(TCA_CLS_FLAGS_IN_HW | TCA_CLS_FLAGS_NOT_IN_HW)) {
 908                         NL_SET_ERR_MSG_MOD(extack, "Key node flags do not match passed flags");
 909                         return -EINVAL;
 910                 }
 911
 912                 new = u32_init_knode(net, tp, n);
 913                 if (!new)
 914                         return -ENOMEM;
 915
 916                 err = u32_set_parms(net, tp, new, tb, tca[TCA_RATE],
 917                                     flags, new->flags, extack);
 918
 919                 if (err) {
 920                         __u32_destroy_key(new);
 921                         return err;
 922                 }
 923
 924                 u32_bind_filter(tp, new, base, tb);
 925
 926                 err = u32_replace_hw_knode(tp, new, flags, extack);
 927                 if (err) {
 928                         u32_unbind_filter(tp, new, tb);
 929
 930                         if (tb[TCA_U32_LINK]) {
 931                                 struct tc_u_hnode *ht_old;
 932
 933                                 ht_old = rtnl_dereference(n->ht_down);
 934                                 if (ht_old)
 935                                         ht_old->refcnt++;
 936                         }
 937                         __u32_destroy_key(new);
 938                         return err;
 939                 }
 940
 941                 if (!tc_in_hw(new->flags))
 942                         new->flags |= TCA_CLS_FLAGS_NOT_IN_HW;
 943
 944                 u32_replace_knode(tp, tp_c, new);
 945                 tcf_unbind_filter(tp, &n->res);
 946                 tcf_exts_get_net(&n->exts);
 947                 tcf_queue_work(&n->rwork, u32_delete_key_work);
 948                 return 0;
 949         }
 950
 951         if (tb[TCA_U32_DIVISOR]) {
 952                 unsigned int divisor = nla_get_u32(tb[TCA_U32_DIVISOR]);
 953
 954                 if (!is_power_of_2(divisor)) {
 955                         NL_SET_ERR_MSG_MOD(extack, "Divisor is not a power of 2");
 956                         return -EINVAL;
 957                 }
 958                 if (divisor-- > 0x100) {
 959                         NL_SET_ERR_MSG_MOD(extack, "Exceeded maximum 256 hash buckets");
 960                         return -EINVAL;
 961                 }
 962                 if (TC_U32_KEY(handle)) {
 963                         NL_SET_ERR_MSG_MOD(extack, "Divisor can only be used on a hash table");
 964                         return -EINVAL;
 965                 }
 966                 ht = kzalloc(struct_size(ht, ht, divisor + 1), GFP_KERNEL);
 967                 if (ht == NULL)
 968                         return -ENOBUFS;
 969                 if (handle == 0) {
 970                         handle = gen_new_htid(tp->data, ht);
 971                         if (handle == 0) {
 972                                 kfree(ht);
 973                                 return -ENOMEM;
 974                         }
 975                 } else {
 976                         err = idr_alloc_u32(&tp_c->handle_idr, ht, &handle,
 977                                             handle, GFP_KERNEL);
 978                         if (err) {
 979                                 kfree(ht);
 980                                 return err;
 981                         }
 982                 }
 983                 ht->refcnt = 1;
 984                 ht->divisor = divisor;
 985                 ht->handle = handle;
 986                 ht->prio = tp->prio;
 987                 idr_init(&ht->handle_idr);
 988                 ht->flags = userflags;
 989
 990                 err = u32_replace_hw_hnode(tp, ht, userflags, extack);
 991                 if (err) {
 992                         idr_remove(&tp_c->handle_idr, handle);
 993                         kfree(ht);
 994                         return err;
 995                 }
 996
 997                 RCU_INIT_POINTER(ht->next, tp_c->hlist);
 998                 rcu_assign_pointer(tp_c->hlist, ht);
 999                 *arg = ht;
1000
1001                 return 0;
1002         }
1003
1004         if (tb[TCA_U32_HASH]) {
1005                 htid = nla_get_u32(tb[TCA_U32_HASH]);
1006                 if (TC_U32_HTID(htid) == TC_U32_ROOT) {
1007                         ht = rtnl_dereference(tp->root);
1008                         htid = ht->handle;
1009                 } else {
1010                         ht = u32_lookup_ht(tp->data, TC_U32_HTID(htid));
1011                         if (!ht) {
1012                                 NL_SET_ERR_MSG_MOD(extack, "Specified hash table not found");
1013                                 return -EINVAL;
1014                         }
1015                 }
1016         } else {
1017                 ht = rtnl_dereference(tp->root);
1018                 htid = ht->handle;
1019         }
1020
1021         if (ht->divisor < TC_U32_HASH(htid)) {
1022                 NL_SET_ERR_MSG_MOD(extack, "Specified hash table buckets exceed configured value");
1023                 return -EINVAL;
1024         }
1025
1026         /* At this point, we need to derive the new handle that will be used to
1027          * uniquely map the identity of this table match entry. The
1028          * identity of the entry that we need to construct is 32 bits made of:
1029          *     htid(12b):bucketid(8b):node/entryid(12b)
1030          *
1031          * At this point _we have the table(ht)_ in which we will insert this
1032          * entry. We carry the table's id in variable "htid".
1033          * Note that earlier code picked the ht selection either by a) the user
1034          * providing the htid specified via TCA_U32_HASH attribute or b) when
1035          * no such attribute is passed then the root ht, is default to at ID
1036          * 0x[800][00][000]. Rule: the root table has a single bucket with ID 0.
1037          * If OTOH the user passed us the htid, they may also pass a bucketid of
1038          * choice. 0 is fine. For example a user htid is 0x[600][01][000] it is
1039          * indicating hash bucketid of 1. Rule: the entry/node ID _cannot_ be
1040          * passed via the htid, so even if it was non-zero it will be ignored.
1041          *
1042          * We may also have a handle, if the user passed one. The handle also
1043          * carries the same addressing of htid(12b):bucketid(8b):node/entryid(12b).
1044          * Rule: the bucketid on the handle is ignored even if one was passed;
1045          * rather the value on "htid" is always assumed to be the bucketid.
1046          */
1047         if (handle) {
1048                 /* Rule: The htid from handle and tableid from htid must match */
1049                 if (TC_U32_HTID(handle) && TC_U32_HTID(handle ^ htid)) {
1050                         NL_SET_ERR_MSG_MOD(extack, "Handle specified hash table address mismatch");
1051                         return -EINVAL;
1052                 }
1053                 /* Ok, so far we have a valid htid(12b):bucketid(8b) but we
1054                  * need to finalize the table entry identification with the last
1055                  * part - the node/entryid(12b)). Rule: Nodeid _cannot be 0_ for
1056                  * entries. Rule: nodeid of 0 is reserved only for tables(see
1057                  * earlier code which processes TC_U32_DIVISOR attribute).
1058                  * Rule: The nodeid can only be derived from the handle (and not
1059                  * htid).
1060                  * Rule: if the handle specified zero for the node id example
1061                  * 0x60000000, then pick a new nodeid from the pool of IDs
1062                  * this hash table has been allocating from.
1063                  * If OTOH it is specified (i.e for example the user passed a
1064                  * handle such as 0x60000123), then we use it generate our final
1065                  * handle which is used to uniquely identify the match entry.
1066                  */
1067                 if (!TC_U32_NODE(handle)) {
1068                         handle = gen_new_kid(ht, htid);
1069                 } else {
1070                         handle = htid | TC_U32_NODE(handle);
1071                         err = idr_alloc_u32(&ht->handle_idr, NULL, &handle,
1072                                             handle, GFP_KERNEL);
1073                         if (err)
1074                                 return err;
1075                 }
1076         } else {
1077                 /* The user did not give us a handle; lets just generate one
1078                  * from the table's pool of nodeids.
1079                  */
1080                 handle = gen_new_kid(ht, htid);
1081         }
1082
1083         if (tb[TCA_U32_SEL] == NULL) {
1084                 NL_SET_ERR_MSG_MOD(extack, "Selector not specified");
1085                 err = -EINVAL;
1086                 goto erridr;
1087         }
1088
1089         s = nla_data(tb[TCA_U32_SEL]);
1090         sel_size = struct_size(s, keys, s->nkeys);
1091         if (nla_len(tb[TCA_U32_SEL]) < sel_size) {
1092                 err = -EINVAL;
1093                 goto erridr;
1094         }
1095
1096         n = kzalloc(struct_size(n, sel.keys, s->nkeys), GFP_KERNEL);
1097         if (n == NULL) {
1098                 err = -ENOBUFS;
1099                 goto erridr;
1100         }
1101
1102 #ifdef CONFIG_CLS_U32_PERF
1103         n->pf = __alloc_percpu(struct_size(n->pf, kcnts, s->nkeys),
1104                                __alignof__(struct tc_u32_pcnt));
1105         if (!n->pf) {
1106                 err = -ENOBUFS;
1107                 goto errfree;
1108         }
1109 #endif
1110
1111         unsafe_memcpy(&n->sel, s, sel_size,
1112                       /* A composite flex-array structure destination,
1113                        * which was correctly sized with struct_size(),
1114                        * bounds-checked against nla_len(), and allocated
1115                        * above. */);
1116         RCU_INIT_POINTER(n->ht_up, ht);
1117         n->handle = handle;
1118         n->fshift = s->hmask ? ffs(ntohl(s->hmask)) - 1 : 0;
1119         n->flags = userflags;
1120
1121         err = tcf_exts_init(&n->exts, net, TCA_U32_ACT, TCA_U32_POLICE);
1122         if (err < 0)
1123                 goto errout;
1124
1125 #ifdef CONFIG_CLS_U32_MARK
1126         n->pcpu_success = alloc_percpu(u32);
1127         if (!n->pcpu_success) {
1128                 err = -ENOMEM;
1129                 goto errout;
1130         }
1131
1132         if (tb[TCA_U32_MARK]) {
1133                 struct tc_u32_mark *mark;
1134
1135                 mark = nla_data(tb[TCA_U32_MARK]);
1136                 n->val = mark->val;
1137                 n->mask = mark->mask;
1138         }
1139 #endif
1140
1141         err = u32_set_parms(net, tp, n, tb, tca[TCA_RATE],
1142                             flags, n->flags, extack);
1143
1144         u32_bind_filter(tp, n, base, tb);
1145
1146         if (err == 0) {
1147                 struct tc_u_knode __rcu **ins;
1148                 struct tc_u_knode *pins;
1149
1150                 err = u32_replace_hw_knode(tp, n, flags, extack);
1151                 if (err)
1152                         goto errunbind;
1153
1154                 if (!tc_in_hw(n->flags))
1155                         n->flags |= TCA_CLS_FLAGS_NOT_IN_HW;
1156
1157                 ins = &ht->ht[TC_U32_HASH(handle)];
1158                 for (pins = rtnl_dereference(*ins); pins;
1159                      ins = &pins->next, pins = rtnl_dereference(*ins))
1160                         if (TC_U32_NODE(handle) < TC_U32_NODE(pins->handle))
1161                                 break;
1162
1163                 RCU_INIT_POINTER(n->next, pins);
1164                 rcu_assign_pointer(*ins, n);
1165                 tp_c->knodes++;
1166                 *arg = n;
1167                 return 0;
1168         }
1169
1170 errunbind:
1171         u32_unbind_filter(tp, n, tb);
1172
1173 #ifdef CONFIG_CLS_U32_MARK
1174         free_percpu(n->pcpu_success);
1175 #endif
1176
1177 errout:
1178         tcf_exts_destroy(&n->exts);
1179 #ifdef CONFIG_CLS_U32_PERF
1180 errfree:
1181         free_percpu(n->pf);
1182 #endif
1183         kfree(n);
1184 erridr:
1185         idr_remove(&ht->handle_idr, handle);
1186         return err;
1187 }
1188
1189 static void u32_walk(struct tcf_proto *tp, struct tcf_walker *arg,
1190                      bool rtnl_held)
1191 {
1192         struct tc_u_common *tp_c = tp->data;
1193         struct tc_u_hnode *ht;
1194         struct tc_u_knode *n;
1195         unsigned int h;
1196
1197         if (arg->stop)
1198                 return;
1199
1200         for (ht = rtnl_dereference(tp_c->hlist);
1201              ht;
1202              ht = rtnl_dereference(ht->next)) {
1203                 if (ht->prio != tp->prio)
1204                         continue;
1205
1206                 if (!tc_cls_stats_dump(tp, arg, ht))
1207                         return;
1208
1209                 for (h = 0; h <= ht->divisor; h++) {
1210                         for (n = rtnl_dereference(ht->ht[h]);
1211                              n;
1212                              n = rtnl_dereference(n->next)) {
1213                                 if (!tc_cls_stats_dump(tp, arg, n))
1214                                         return;
1215                         }
1216                 }
1217         }
1218 }
1219
1220 static int u32_reoffload_hnode(struct tcf_proto *tp, struct tc_u_hnode *ht,
1221                                bool add, flow_setup_cb_t *cb, void *cb_priv,
1222                                struct netlink_ext_ack *extack)
1223 {
1224         struct tc_cls_u32_offload cls_u32 = {};
1225         int err;
1226
1227         tc_cls_common_offload_init(&cls_u32.common, tp, ht->flags, extack);
1228         cls_u32.command = add ? TC_CLSU32_NEW_HNODE : TC_CLSU32_DELETE_HNODE;
1229         cls_u32.hnode.divisor = ht->divisor;
1230         cls_u32.hnode.handle = ht->handle;
1231         cls_u32.hnode.prio = ht->prio;
1232
1233         err = cb(TC_SETUP_CLSU32, &cls_u32, cb_priv);
1234         if (err && add && tc_skip_sw(ht->flags))
1235                 return err;
1236
1237         return 0;
1238 }
1239
1240 static int u32_reoffload_knode(struct tcf_proto *tp, struct tc_u_knode *n,
1241                                bool add, flow_setup_cb_t *cb, void *cb_priv,
1242                                struct netlink_ext_ack *extack)
1243 {
1244         struct tc_u_hnode *ht = rtnl_dereference(n->ht_down);
1245         struct tcf_block *block = tp->chain->block;
1246         struct tc_cls_u32_offload cls_u32 = {};
1247
1248         tc_cls_common_offload_init(&cls_u32.common, tp, n->flags, extack);
1249         cls_u32.command = add ?
1250                 TC_CLSU32_REPLACE_KNODE : TC_CLSU32_DELETE_KNODE;
1251         cls_u32.knode.handle = n->handle;
1252
1253         if (add) {
1254                 cls_u32.knode.fshift = n->fshift;
1255 #ifdef CONFIG_CLS_U32_MARK
1256                 cls_u32.knode.val = n->val;
1257                 cls_u32.knode.mask = n->mask;
1258 #else
1259                 cls_u32.knode.val = 0;
1260                 cls_u32.knode.mask = 0;
1261 #endif
1262                 cls_u32.knode.sel = &n->sel;
1263                 cls_u32.knode.res = &n->res;
1264                 cls_u32.knode.exts = &n->exts;
1265                 if (n->ht_down)
1266                         cls_u32.knode.link_handle = ht->handle;
1267         }
1268
1269         return tc_setup_cb_reoffload(block, tp, add, cb, TC_SETUP_CLSU32,
1270                                      &cls_u32, cb_priv, &n->flags,
1271                                      &n->in_hw_count);
1272 }
1273
1274 static int u32_reoffload(struct tcf_proto *tp, bool add, flow_setup_cb_t *cb,
1275                          void *cb_priv, struct netlink_ext_ack *extack)
1276 {
1277         struct tc_u_common *tp_c = tp->data;
1278         struct tc_u_hnode *ht;
1279         struct tc_u_knode *n;
1280         unsigned int h;
1281         int err;
1282
1283         for (ht = rtnl_dereference(tp_c->hlist);
1284              ht;
1285              ht = rtnl_dereference(ht->next)) {
1286                 if (ht->prio != tp->prio)
1287                         continue;
1288
1289                 /* When adding filters to a new dev, try to offload the
1290                  * hashtable first. When removing, do the filters before the
1291                  * hashtable.
1292                  */
1293                 if (add && !tc_skip_hw(ht->flags)) {
1294                         err = u32_reoffload_hnode(tp, ht, add, cb, cb_priv,
1295                                                   extack);
1296                         if (err)
1297                                 return err;
1298                 }
1299
1300                 for (h = 0; h <= ht->divisor; h++) {
1301                         for (n = rtnl_dereference(ht->ht[h]);
1302                              n;
1303                              n = rtnl_dereference(n->next)) {
1304                                 if (tc_skip_hw(n->flags))
1305                                         continue;
1306
1307                                 err = u32_reoffload_knode(tp, n, add, cb,
1308                                                           cb_priv, extack);
1309                                 if (err)
1310                                         return err;
1311                         }
1312                 }
1313
1314                 if (!add && !tc_skip_hw(ht->flags))
1315                         u32_reoffload_hnode(tp, ht, add, cb, cb_priv, extack);
1316         }
1317
1318         return 0;
1319 }
1320
1321 static void u32_bind_class(void *fh, u32 classid, unsigned long cl, void *q,
1322                            unsigned long base)
1323 {
1324         struct tc_u_knode *n = fh;
1325
1326         tc_cls_bind_class(classid, cl, q, &n->res, base);
1327 }
1328
1329 static int u32_dump(struct net *net, struct tcf_proto *tp, void *fh,
1330                     struct sk_buff *skb, struct tcmsg *t, bool rtnl_held)
1331 {
1332         struct tc_u_knode *n = fh;
1333         struct tc_u_hnode *ht_up, *ht_down;
1334         struct nlattr *nest;
1335
1336         if (n == NULL)
1337                 return skb->len;
1338
1339         t->tcm_handle = n->handle;
1340
1341         nest = nla_nest_start_noflag(skb, TCA_OPTIONS);
1342         if (nest == NULL)
1343                 goto nla_put_failure;
1344
1345         if (TC_U32_KEY(n->handle) == 0) {
1346                 struct tc_u_hnode *ht = fh;
1347                 u32 divisor = ht->divisor + 1;
1348
1349                 if (nla_put_u32(skb, TCA_U32_DIVISOR, divisor))
1350                         goto nla_put_failure;
1351         } else {
1352 #ifdef CONFIG_CLS_U32_PERF
1353                 struct tc_u32_pcnt *gpf;
1354                 int cpu;
1355 #endif
1356
1357                 if (nla_put(skb, TCA_U32_SEL, struct_size(&n->sel, keys, n->sel.nkeys),
1358                             &n->sel))
1359                         goto nla_put_failure;
1360
1361                 ht_up = rtnl_dereference(n->ht_up);
1362                 if (ht_up) {
1363                         u32 htid = n->handle & 0xFFFFF000;
1364                         if (nla_put_u32(skb, TCA_U32_HASH, htid))
1365                                 goto nla_put_failure;
1366                 }
1367                 if (n->res.classid &&
1368                     nla_put_u32(skb, TCA_U32_CLASSID, n->res.classid))
1369                         goto nla_put_failure;
1370
1371                 ht_down = rtnl_dereference(n->ht_down);
1372                 if (ht_down &&
1373                     nla_put_u32(skb, TCA_U32_LINK, ht_down->handle))
1374                         goto nla_put_failure;
1375
1376                 if (n->flags && nla_put_u32(skb, TCA_U32_FLAGS, n->flags))
1377                         goto nla_put_failure;
1378
1379 #ifdef CONFIG_CLS_U32_MARK
1380                 if ((n->val || n->mask)) {
1381                         struct tc_u32_mark mark = {.val = n->val,
1382                                                    .mask = n->mask,
1383                                                    .success = 0};
1384                         int cpum;
1385
1386                         for_each_possible_cpu(cpum) {
1387                                 __u32 cnt = *per_cpu_ptr(n->pcpu_success, cpum);
1388
1389                                 mark.success += cnt;
1390                         }
1391
1392                         if (nla_put(skb, TCA_U32_MARK, sizeof(mark), &mark))
1393                                 goto nla_put_failure;
1394                 }
1395 #endif
1396
1397                 if (tcf_exts_dump(skb, &n->exts) < 0)
1398                         goto nla_put_failure;
1399
1400                 if (n->ifindex) {
1401                         struct net_device *dev;
1402                         dev = __dev_get_by_index(net, n->ifindex);
1403                         if (dev && nla_put_string(skb, TCA_U32_INDEV, dev->name))
1404                                 goto nla_put_failure;
1405                 }
1406 #ifdef CONFIG_CLS_U32_PERF
1407                 gpf = kzalloc(struct_size(gpf, kcnts, n->sel.nkeys), GFP_KERNEL);
1408                 if (!gpf)
1409                         goto nla_put_failure;
1410
1411                 for_each_possible_cpu(cpu) {
1412                         int i;
1413                         struct tc_u32_pcnt *pf = per_cpu_ptr(n->pf, cpu);
1414
1415                         gpf->rcnt += pf->rcnt;
1416                         gpf->rhit += pf->rhit;
1417                         for (i = 0; i < n->sel.nkeys; i++)
1418                                 gpf->kcnts[i] += pf->kcnts[i];
1419                 }
1420
1421                 if (nla_put_64bit(skb, TCA_U32_PCNT, struct_size(gpf, kcnts, n->sel.nkeys),
1422                                   gpf, TCA_U32_PAD)) {
1423                         kfree(gpf);
1424                         goto nla_put_failure;
1425                 }
1426                 kfree(gpf);
1427 #endif
1428         }
1429
1430         nla_nest_end(skb, nest);
1431
1432         if (TC_U32_KEY(n->handle))
1433                 if (tcf_exts_dump_stats(skb, &n->exts) < 0)
1434                         goto nla_put_failure;
1435         return skb->len;
1436
1437 nla_put_failure:
1438         nla_nest_cancel(skb, nest);
1439         return -1;
1440 }
1441
1442 static struct tcf_proto_ops cls_u32_ops __read_mostly = {
1443         .kind           =       "u32",
1444         .classify       =       u32_classify,
1445         .init           =       u32_init,
1446         .destroy        =       u32_destroy,
1447         .get            =       u32_get,
1448         .change         =       u32_change,
1449         .delete         =       u32_delete,
1450         .walk           =       u32_walk,
1451         .reoffload      =       u32_reoffload,
1452         .dump           =       u32_dump,
1453         .bind_class     =       u32_bind_class,
1454         .owner          =       THIS_MODULE,
1455 };
1456
1457 static int __init init_u32(void)
1458 {
1459         int i, ret;
1460
1461         pr_info("u32 classifier\n");
1462 #ifdef CONFIG_CLS_U32_PERF
1463         pr_info("    Performance counters on\n");
1464 #endif
1465         pr_info("    input device check on\n");
1466 #ifdef CONFIG_NET_CLS_ACT
1467         pr_info("    Actions configured\n");
1468 #endif
1469         tc_u_common_hash = kvmalloc_array(U32_HASH_SIZE,
1470                                           sizeof(struct hlist_head),
1471                                           GFP_KERNEL);
1472         if (!tc_u_common_hash)
1473                 return -ENOMEM;
1474
1475         for (i = 0; i < U32_HASH_SIZE; i++)
1476                 INIT_HLIST_HEAD(&tc_u_common_hash[i]);
1477
1478         ret = register_tcf_proto_ops(&cls_u32_ops);
1479         if (ret)
1480                 kvfree(tc_u_common_hash);
1481         return ret;
1482 }
1483
1484 static void __exit exit_u32(void)
1485 {
1486         unregister_tcf_proto_ops(&cls_u32_ops);
1487         kvfree(tc_u_common_hash);
1488 }
1489
1490 module_init(init_u32)
1491 module_exit(exit_u32)
1492 MODULE_LICENSE("GPL");