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
  43 struct tc_u_knode {
  44         struct tc_u_knode __rcu *next;
  45         u32                     handle;
  46         struct tc_u_hnode __rcu *ht_up;
  47         struct tcf_exts         exts;
  48         int                     ifindex;
  49         u8                      fshift;
  50         struct tcf_result       res;
  51         struct tc_u_hnode __rcu *ht_down;
  52 #ifdef CONFIG_CLS_U32_PERF
  53         struct tc_u32_pcnt __percpu *pf;
  54 #endif
  55         u32                     flags;
  56         unsigned int            in_hw_count;
  57 #ifdef CONFIG_CLS_U32_MARK
  58         u32                     val;
  59         u32                     mask;
  60         u32 __percpu            *pcpu_success;
  61 #endif
  62         struct rcu_work         rwork;
  63         /* The 'sel' field MUST be the last field in structure to allow for
  64          * tc_u32_keys allocated at end of structure.
  65          */
  66         struct tc_u32_sel       sel;
  67 };
  68
  69 struct tc_u_hnode {
  70         struct tc_u_hnode __rcu *next;
  71         u32                     handle;
  72         u32                     prio;
  73         int                     refcnt;
  74         unsigned int            divisor;
  75         struct idr              handle_idr;
  76         bool                    is_root;
  77         struct rcu_head         rcu;
  78         u32                     flags;
  79         /* The 'ht' field MUST be the last field in structure to allow for
  80          * more entries allocated at end of structure.
  81          */
  82         struct tc_u_knode __rcu *ht[];
  83 };
  84
  85 struct tc_u_common {
  86         struct tc_u_hnode __rcu *hlist;
  87         void                    *ptr;
  88         int                     refcnt;
  89         struct idr              handle_idr;
  90         struct hlist_node       hnode;
  91         long                    knodes;
  92 };
  93
  94 static inline unsigned int u32_hash_fold(__be32 key,
  95                                          const struct tc_u32_sel *sel,
  96                                          u8 fshift)
  97 {
  98         unsigned int h = ntohl(key & sel->hmask) >> fshift;
  99
 100         return h;
 101 }
 102
 103 static int u32_classify(struct sk_buff *skb, const struct tcf_proto *tp,
 104                         struct tcf_result *res)
 105 {
 106         struct {
 107                 struct tc_u_knode *knode;
 108                 unsigned int      off;
 109         } stack[TC_U32_MAXDEPTH];
 110
 111         struct tc_u_hnode *ht = rcu_dereference_bh(tp->root);
 112         unsigned int off = skb_network_offset(skb);
 113         struct tc_u_knode *n;
 114         int sdepth = 0;
 115         int off2 = 0;
 116         int sel = 0;
 117 #ifdef CONFIG_CLS_U32_PERF
 118         int j;
 119 #endif
 120         int i, r;
 121
 122 next_ht:
 123         n = rcu_dereference_bh(ht->ht[sel]);
 124
 125 next_knode:
 126         if (n) {
 127                 struct tc_u32_key *key = n->sel.keys;
 128
 129 #ifdef CONFIG_CLS_U32_PERF
 130                 __this_cpu_inc(n->pf->rcnt);
 131                 j = 0;
 132 #endif
 133
 134                 if (tc_skip_sw(n->flags)) {
 135                         n = rcu_dereference_bh(n->next);
 136                         goto next_knode;
 137                 }
 138
 139 #ifdef CONFIG_CLS_U32_MARK
 140                 if ((skb->mark & n->mask) != n->val) {
 141                         n = rcu_dereference_bh(n->next);
 142                         goto next_knode;
 143                 } else {
 144                         __this_cpu_inc(*n->pcpu_success);
 145                 }
 146 #endif
 147
 148                 for (i = n->sel.nkeys; i > 0; i--, key++) {
 149                         int toff = off + key->off + (off2 & key->offmask);
 150                         __be32 *data, hdata;
 151
 152                         if (skb_headroom(skb) + toff > INT_MAX)
 153                                 goto out;
 154
 155                         data = skb_header_pointer(skb, toff, 4, &hdata);
 156                         if (!data)
 157                                 goto out;
 158                         if ((*data ^ key->val) & key->mask) {
 159                                 n = rcu_dereference_bh(n->next);
 160                                 goto next_knode;
 161                         }
 162 #ifdef CONFIG_CLS_U32_PERF
 163                         __this_cpu_inc(n->pf->kcnts[j]);
 164                         j++;
 165 #endif
 166                 }
 167
 168                 ht = rcu_dereference_bh(n->ht_down);
 169                 if (!ht) {
 170 check_terminal:
 171                         if (n->sel.flags & TC_U32_TERMINAL) {
 172
 173                                 *res = n->res;
 174                                 if (!tcf_match_indev(skb, n->ifindex)) {
 175                                         n = rcu_dereference_bh(n->next);
 176                                         goto next_knode;
 177                                 }
 178 #ifdef CONFIG_CLS_U32_PERF
 179                                 __this_cpu_inc(n->pf->rhit);
 180 #endif
 181                                 r = tcf_exts_exec(skb, &n->exts, res);
 182                                 if (r < 0) {
 183                                         n = rcu_dereference_bh(n->next);
 184                                         goto next_knode;
 185                                 }
 186
 187                                 return r;
 188                         }
 189                         n = rcu_dereference_bh(n->next);
 190                         goto next_knode;
 191                 }
 192
 193                 /* PUSH */
 194                 if (sdepth >= TC_U32_MAXDEPTH)
 195                         goto deadloop;
 196                 stack[sdepth].knode = n;
 197                 stack[sdepth].off = off;
 198                 sdepth++;
 199
 200                 ht = rcu_dereference_bh(n->ht_down);
 201                 sel = 0;
 202                 if (ht->divisor) {
 203                         __be32 *data, hdata;
 204
 205                         data = skb_header_pointer(skb, off + n->sel.hoff, 4,
 206                                                   &hdata);
 207                         if (!data)
 208                                 goto out;
 209                         sel = ht->divisor & u32_hash_fold(*data, &n->sel,
 210                                                           n->fshift);
 211                 }
 212                 if (!(n->sel.flags & (TC_U32_VAROFFSET | TC_U32_OFFSET | TC_U32_EAT)))
 213                         goto next_ht;
 214
 215                 if (n->sel.flags & (TC_U32_OFFSET | TC_U32_VAROFFSET)) {
 216                         off2 = n->sel.off + 3;
 217                         if (n->sel.flags & TC_U32_VAROFFSET) {
 218                                 __be16 *data, hdata;
 219
 220                                 data = skb_header_pointer(skb,
 221                                                           off + n->sel.offoff,
 222                                                           2, &hdata);
 223                                 if (!data)
 224                                         goto out;
 225                                 off2 += ntohs(n->sel.offmask & *data) >>
 226                                         n->sel.offshift;
 227                         }
 228                         off2 &= ~3;
 229                 }
 230                 if (n->sel.flags & TC_U32_EAT) {
 231                         off += off2;
 232                         off2 = 0;
 233                 }
 234
 235                 if (off < skb->len)
 236                         goto next_ht;
 237         }
 238
 239         /* POP */
 240         if (sdepth--) {
 241                 n = stack[sdepth].knode;
 242                 ht = rcu_dereference_bh(n->ht_up);
 243                 off = stack[sdepth].off;
 244                 goto check_terminal;
 245         }
 246 out:
 247         return -1;
 248
 249 deadloop:
 250         net_warn_ratelimited("cls_u32: dead loop\n");
 251         return -1;
 252 }
 253
 254 static struct tc_u_hnode *u32_lookup_ht(struct tc_u_common *tp_c, u32 handle)
 255 {
 256         struct tc_u_hnode *ht;
 257
 258         for (ht = rtnl_dereference(tp_c->hlist);
 259              ht;
 260              ht = rtnl_dereference(ht->next))
 261                 if (ht->handle == handle)
 262                         break;
 263
 264         return ht;
 265 }
 266
 267 static struct tc_u_knode *u32_lookup_key(struct tc_u_hnode *ht, u32 handle)
 268 {
 269         unsigned int sel;
 270         struct tc_u_knode *n = NULL;
 271
 272         sel = TC_U32_HASH(handle);
 273         if (sel > ht->divisor)
 274                 goto out;
 275
 276         for (n = rtnl_dereference(ht->ht[sel]);
 277              n;
 278              n = rtnl_dereference(n->next))
 279                 if (n->handle == handle)
 280                         break;
 281 out:
 282         return n;
 283 }
 284
 285
 286 static void *u32_get(struct tcf_proto *tp, u32 handle)
 287 {
 288         struct tc_u_hnode *ht;
 289         struct tc_u_common *tp_c = tp->data;
 290
 291         if (TC_U32_HTID(handle) == TC_U32_ROOT)
 292                 ht = rtnl_dereference(tp->root);
 293         else
 294                 ht = u32_lookup_ht(tp_c, TC_U32_HTID(handle));
 295
 296         if (!ht)
 297                 return NULL;
 298
 299         if (TC_U32_KEY(handle) == 0)
 300                 return ht;
 301
 302         return u32_lookup_key(ht, handle);
 303 }
 304
 305 /* Protected by rtnl lock */
 306 static u32 gen_new_htid(struct tc_u_common *tp_c, struct tc_u_hnode *ptr)
 307 {
 308         int id = idr_alloc_cyclic(&tp_c->handle_idr, ptr, 1, 0x7FF, GFP_KERNEL);
 309         if (id < 0)
 310                 return 0;
 311         return (id | 0x800U) << 20;
 312 }
 313
 314 static struct hlist_head *tc_u_common_hash;
 315
 316 #define U32_HASH_SHIFT 10
 317 #define U32_HASH_SIZE (1 << U32_HASH_SHIFT)
 318
 319 static void *tc_u_common_ptr(const struct tcf_proto *tp)
 320 {
 321         struct tcf_block *block = tp->chain->block;
 322
 323         /* The block sharing is currently supported only
 324          * for classless qdiscs. In that case we use block
 325          * for tc_u_common identification. In case the
 326          * block is not shared, block->q is a valid pointer
 327          * and we can use that. That works for classful qdiscs.
 328          */
 329         if (tcf_block_shared(block))
 330                 return block;
 331         else
 332                 return block->q;
 333 }
 334
 335 static struct hlist_head *tc_u_hash(void *key)
 336 {
 337         return tc_u_common_hash + hash_ptr(key, U32_HASH_SHIFT);
 338 }
 339
 340 static struct tc_u_common *tc_u_common_find(void *key)
 341 {
 342         struct tc_u_common *tc;
 343         hlist_for_each_entry(tc, tc_u_hash(key), hnode) {
 344                 if (tc->ptr == key)
 345                         return tc;
 346         }
 347         return NULL;
 348 }
 349
 350 static int u32_init(struct tcf_proto *tp)
 351 {
 352         struct tc_u_hnode *root_ht;
 353         void *key = tc_u_common_ptr(tp);
 354         struct tc_u_common *tp_c = tc_u_common_find(key);
 355
 356         root_ht = kzalloc(struct_size(root_ht, ht, 1), GFP_KERNEL);
 357         if (root_ht == NULL)
 358                 return -ENOBUFS;
 359
 360         root_ht->refcnt++;
 361         root_ht->handle = tp_c ? gen_new_htid(tp_c, root_ht) : 0x80000000;
 362         root_ht->prio = tp->prio;
 363         root_ht->is_root = true;
 364         idr_init(&root_ht->handle_idr);
 365
 366         if (tp_c == NULL) {
 367                 tp_c = kzalloc(struct_size(tp_c, hlist->ht, 1), GFP_KERNEL);
 368                 if (tp_c == NULL) {
 369                         kfree(root_ht);
 370                         return -ENOBUFS;
 371                 }
 372                 tp_c->ptr = key;
 373                 INIT_HLIST_NODE(&tp_c->hnode);
 374                 idr_init(&tp_c->handle_idr);
 375
 376                 hlist_add_head(&tp_c->hnode, tc_u_hash(key));
 377         }
 378
 379         tp_c->refcnt++;
 380         RCU_INIT_POINTER(root_ht->next, tp_c->hlist);
 381         rcu_assign_pointer(tp_c->hlist, root_ht);
 382
 383         root_ht->refcnt++;
 384         rcu_assign_pointer(tp->root, root_ht);
 385         tp->data = tp_c;
 386         return 0;
 387 }
 388
 389 static void __u32_destroy_key(struct tc_u_knode *n)
 390 {
 391         struct tc_u_hnode *ht = rtnl_dereference(n->ht_down);
 392
 393         tcf_exts_destroy(&n->exts);
 394         if (ht && --ht->refcnt == 0)
 395                 kfree(ht);
 396         kfree(n);
 397 }
 398
 399 static void u32_destroy_key(struct tc_u_knode *n, bool free_pf)
 400 {
 401         tcf_exts_put_net(&n->exts);
 402 #ifdef CONFIG_CLS_U32_PERF
 403         if (free_pf)
 404                 free_percpu(n->pf);
 405 #endif
 406 #ifdef CONFIG_CLS_U32_MARK
 407         if (free_pf)
 408                 free_percpu(n->pcpu_success);
 409 #endif
 410         __u32_destroy_key(n);
 411 }
 412
 413 /* u32_delete_key_rcu should be called when free'ing a copied
 414  * version of a tc_u_knode obtained from u32_init_knode(). When
 415  * copies are obtained from u32_init_knode() the statistics are
 416  * shared between the old and new copies to allow readers to
 417  * continue to update the statistics during the copy. To support
 418  * this the u32_delete_key_rcu variant does not free the percpu
 419  * statistics.
 420  */
 421 static void u32_delete_key_work(struct work_struct *work)
 422 {
 423         struct tc_u_knode *key = container_of(to_rcu_work(work),
 424                                               struct tc_u_knode,
 425                                               rwork);
 426         rtnl_lock();
 427         u32_destroy_key(key, false);
 428         rtnl_unlock();
 429 }
 430
 431 /* u32_delete_key_freepf_rcu is the rcu callback variant
 432  * that free's the entire structure including the statistics
 433  * percpu variables. Only use this if the key is not a copy
 434  * returned by u32_init_knode(). See u32_delete_key_rcu()
 435  * for the variant that should be used with keys return from
 436  * u32_init_knode()
 437  */
 438 static void u32_delete_key_freepf_work(struct work_struct *work)
 439 {
 440         struct tc_u_knode *key = container_of(to_rcu_work(work),
 441                                               struct tc_u_knode,
 442                                               rwork);
 443         rtnl_lock();
 444         u32_destroy_key(key, true);
 445         rtnl_unlock();
 446 }
 447
 448 static int u32_delete_key(struct tcf_proto *tp, struct tc_u_knode *key)
 449 {
 450         struct tc_u_common *tp_c = tp->data;
 451         struct tc_u_knode __rcu **kp;
 452         struct tc_u_knode *pkp;
 453         struct tc_u_hnode *ht = rtnl_dereference(key->ht_up);
 454
 455         if (ht) {
 456                 kp = &ht->ht[TC_U32_HASH(key->handle)];
 457                 for (pkp = rtnl_dereference(*kp); pkp;
 458                      kp = &pkp->next, pkp = rtnl_dereference(*kp)) {
 459                         if (pkp == key) {
 460                                 RCU_INIT_POINTER(*kp, key->next);
 461                                 tp_c->knodes--;
 462
 463                                 tcf_unbind_filter(tp, &key->res);
 464                                 idr_remove(&ht->handle_idr, key->handle);
 465                                 tcf_exts_get_net(&key->exts);
 466                                 tcf_queue_work(&key->rwork, u32_delete_key_freepf_work);
 467                                 return 0;
 468                         }
 469                 }
 470         }
 471         WARN_ON(1);
 472         return 0;
 473 }
 474
 475 static void u32_clear_hw_hnode(struct tcf_proto *tp, struct tc_u_hnode *h,
 476                                struct netlink_ext_ack *extack)
 477 {
 478         struct tcf_block *block = tp->chain->block;
 479         struct tc_cls_u32_offload cls_u32 = {};
 480
 481         tc_cls_common_offload_init(&cls_u32.common, tp, h->flags, extack);
 482         cls_u32.command = TC_CLSU32_DELETE_HNODE;
 483         cls_u32.hnode.divisor = h->divisor;
 484         cls_u32.hnode.handle = h->handle;
 485         cls_u32.hnode.prio = h->prio;
 486
 487         tc_setup_cb_call(block, TC_SETUP_CLSU32, &cls_u32, false, true);
 488 }
 489
 490 static int u32_replace_hw_hnode(struct tcf_proto *tp, struct tc_u_hnode *h,
 491                                 u32 flags, struct netlink_ext_ack *extack)
 492 {
 493         struct tcf_block *block = tp->chain->block;
 494         struct tc_cls_u32_offload cls_u32 = {};
 495         bool skip_sw = tc_skip_sw(flags);
 496         bool offloaded = false;
 497         int err;
 498
 499         tc_cls_common_offload_init(&cls_u32.common, tp, flags, extack);
 500         cls_u32.command = TC_CLSU32_NEW_HNODE;
 501         cls_u32.hnode.divisor = h->divisor;
 502         cls_u32.hnode.handle = h->handle;
 503         cls_u32.hnode.prio = h->prio;
 504
 505         err = tc_setup_cb_call(block, TC_SETUP_CLSU32, &cls_u32, skip_sw, true);
 506         if (err < 0) {
 507                 u32_clear_hw_hnode(tp, h, NULL);
 508                 return err;
 509         } else if (err > 0) {
 510                 offloaded = true;
 511         }
 512
 513         if (skip_sw && !offloaded)
 514                 return -EINVAL;
 515
 516         return 0;
 517 }
 518
 519 static void u32_remove_hw_knode(struct tcf_proto *tp, struct tc_u_knode *n,
 520                                 struct netlink_ext_ack *extack)
 521 {
 522         struct tcf_block *block = tp->chain->block;
 523         struct tc_cls_u32_offload cls_u32 = {};
 524
 525         tc_cls_common_offload_init(&cls_u32.common, tp, n->flags, extack);
 526         cls_u32.command = TC_CLSU32_DELETE_KNODE;
 527         cls_u32.knode.handle = n->handle;
 528
 529         tc_setup_cb_destroy(block, tp, TC_SETUP_CLSU32, &cls_u32, false,
 530                             &n->flags, &n->in_hw_count, true);
 531 }
 532
 533 static int u32_replace_hw_knode(struct tcf_proto *tp, struct tc_u_knode *n,
 534                                 u32 flags, struct netlink_ext_ack *extack)
 535 {
 536         struct tc_u_hnode *ht = rtnl_dereference(n->ht_down);
 537         struct tcf_block *block = tp->chain->block;
 538         struct tc_cls_u32_offload cls_u32 = {};
 539         bool skip_sw = tc_skip_sw(flags);
 540         int err;
 541
 542         tc_cls_common_offload_init(&cls_u32.common, tp, flags, extack);
 543         cls_u32.command = TC_CLSU32_REPLACE_KNODE;
 544         cls_u32.knode.handle = n->handle;
 545         cls_u32.knode.fshift = n->fshift;
 546 #ifdef CONFIG_CLS_U32_MARK
 547         cls_u32.knode.val = n->val;
 548         cls_u32.knode.mask = n->mask;
 549 #else
 550         cls_u32.knode.val = 0;
 551         cls_u32.knode.mask = 0;
 552 #endif
 553         cls_u32.knode.sel = &n->sel;
 554         cls_u32.knode.res = &n->res;
 555         cls_u32.knode.exts = &n->exts;
 556         if (n->ht_down)
 557                 cls_u32.knode.link_handle = ht->handle;
 558
 559         err = tc_setup_cb_add(block, tp, TC_SETUP_CLSU32, &cls_u32, skip_sw,
 560                               &n->flags, &n->in_hw_count, true);
 561         if (err) {
 562                 u32_remove_hw_knode(tp, n, NULL);
 563                 return err;
 564         }
 565
 566         if (skip_sw && !(n->flags & TCA_CLS_FLAGS_IN_HW))
 567                 return -EINVAL;
 568
 569         return 0;
 570 }
 571
 572 static void u32_clear_hnode(struct tcf_proto *tp, struct tc_u_hnode *ht,
 573                             struct netlink_ext_ack *extack)
 574 {
 575         struct tc_u_common *tp_c = tp->data;
 576         struct tc_u_knode *n;
 577         unsigned int h;
 578
 579         for (h = 0; h <= ht->divisor; h++) {
 580                 while ((n = rtnl_dereference(ht->ht[h])) != NULL) {
 581                         RCU_INIT_POINTER(ht->ht[h],
 582                                          rtnl_dereference(n->next));
 583                         tp_c->knodes--;
 584                         tcf_unbind_filter(tp, &n->res);
 585                         u32_remove_hw_knode(tp, n, extack);
 586                         idr_remove(&ht->handle_idr, n->handle);
 587                         if (tcf_exts_get_net(&n->exts))
 588                                 tcf_queue_work(&n->rwork, u32_delete_key_freepf_work);
 589                         else
 590                                 u32_destroy_key(n, true);
 591                 }
 592         }
 593 }
 594
 595 static int u32_destroy_hnode(struct tcf_proto *tp, struct tc_u_hnode *ht,
 596                              struct netlink_ext_ack *extack)
 597 {
 598         struct tc_u_common *tp_c = tp->data;
 599         struct tc_u_hnode __rcu **hn;
 600         struct tc_u_hnode *phn;
 601
 602         WARN_ON(--ht->refcnt);
 603
 604         u32_clear_hnode(tp, ht, extack);
 605
 606         hn = &tp_c->hlist;
 607         for (phn = rtnl_dereference(*hn);
 608              phn;
 609              hn = &phn->next, phn = rtnl_dereference(*hn)) {
 610                 if (phn == ht) {
 611                         u32_clear_hw_hnode(tp, ht, extack);
 612                         idr_destroy(&ht->handle_idr);
 613                         idr_remove(&tp_c->handle_idr, ht->handle);
 614                         RCU_INIT_POINTER(*hn, ht->next);
 615                         kfree_rcu(ht, rcu);
 616                         return 0;
 617                 }
 618         }
 619
 620         return -ENOENT;
 621 }
 622
 623 static void u32_destroy(struct tcf_proto *tp, bool rtnl_held,
 624                         struct netlink_ext_ack *extack)
 625 {
 626         struct tc_u_common *tp_c = tp->data;
 627         struct tc_u_hnode *root_ht = rtnl_dereference(tp->root);
 628
 629         WARN_ON(root_ht == NULL);
 630
 631         if (root_ht && --root_ht->refcnt == 1)
 632                 u32_destroy_hnode(tp, root_ht, extack);
 633
 634         if (--tp_c->refcnt == 0) {
 635                 struct tc_u_hnode *ht;
 636
 637                 hlist_del(&tp_c->hnode);
 638
 639                 while ((ht = rtnl_dereference(tp_c->hlist)) != NULL) {
 640                         u32_clear_hnode(tp, ht, extack);
 641                         RCU_INIT_POINTER(tp_c->hlist, ht->next);
 642
 643                         /* u32_destroy_key() will later free ht for us, if it's
 644                          * still referenced by some knode
 645                          */
 646                         if (--ht->refcnt == 0)
 647                                 kfree_rcu(ht, rcu);
 648                 }
 649
 650                 idr_destroy(&tp_c->handle_idr);
 651                 kfree(tp_c);
 652         }
 653
 654         tp->data = NULL;
 655 }
 656
 657 static int u32_delete(struct tcf_proto *tp, void *arg, bool *last,
 658                       bool rtnl_held, struct netlink_ext_ack *extack)
 659 {
 660         struct tc_u_hnode *ht = arg;
 661         struct tc_u_common *tp_c = tp->data;
 662         int ret = 0;
 663
 664         if (TC_U32_KEY(ht->handle)) {
 665                 u32_remove_hw_knode(tp, (struct tc_u_knode *)ht, extack);
 666                 ret = u32_delete_key(tp, (struct tc_u_knode *)ht);
 667                 goto out;
 668         }
 669
 670         if (ht->is_root) {
 671                 NL_SET_ERR_MSG_MOD(extack, "Not allowed to delete root node");
 672                 return -EINVAL;
 673         }
 674
 675         if (ht->refcnt == 1) {
 676                 u32_destroy_hnode(tp, ht, extack);
 677         } else {
 678                 NL_SET_ERR_MSG_MOD(extack, "Can not delete in-use filter");
 679                 return -EBUSY;
 680         }
 681
 682 out:
 683         *last = tp_c->refcnt == 1 && tp_c->knodes == 0;
 684         return ret;
 685 }
 686
 687 static u32 gen_new_kid(struct tc_u_hnode *ht, u32 htid)
 688 {
 689         u32 index = htid | 0x800;
 690         u32 max = htid | 0xFFF;
 691
 692         if (idr_alloc_u32(&ht->handle_idr, NULL, &index, max, GFP_KERNEL)) {
 693                 index = htid + 1;
 694                 if (idr_alloc_u32(&ht->handle_idr, NULL, &index, max,
 695                                  GFP_KERNEL))
 696                         index = max;
 697         }
 698
 699         return index;
 700 }
 701
 702 static const struct nla_policy u32_policy[TCA_U32_MAX + 1] = {
 703         [TCA_U32_CLASSID]       = { .type = NLA_U32 },
 704         [TCA_U32_HASH]          = { .type = NLA_U32 },
 705         [TCA_U32_LINK]          = { .type = NLA_U32 },
 706         [TCA_U32_DIVISOR]       = { .type = NLA_U32 },
 707         [TCA_U32_SEL]           = { .len = sizeof(struct tc_u32_sel) },
 708         [TCA_U32_INDEV]         = { .type = NLA_STRING, .len = IFNAMSIZ },
 709         [TCA_U32_MARK]          = { .len = sizeof(struct tc_u32_mark) },
 710         [TCA_U32_FLAGS]         = { .type = NLA_U32 },
 711 };
 712
 713 static void u32_unbind_filter(struct tcf_proto *tp, struct tc_u_knode *n,
 714                               struct nlattr **tb)
 715 {
 716         if (tb[TCA_U32_CLASSID])
 717                 tcf_unbind_filter(tp, &n->res);
 718 }
 719
 720 static void u32_bind_filter(struct tcf_proto *tp, struct tc_u_knode *n,
 721                             unsigned long base, struct nlattr **tb)
 722 {
 723         if (tb[TCA_U32_CLASSID]) {
 724                 n->res.classid = nla_get_u32(tb[TCA_U32_CLASSID]);
 725                 tcf_bind_filter(tp, &n->res, base);
 726         }
 727 }
 728
 729 static int u32_set_parms(struct net *net, struct tcf_proto *tp,
 730                          struct tc_u_knode *n, struct nlattr **tb,
 731                          struct nlattr *est, u32 flags, u32 fl_flags,
 732                          struct netlink_ext_ack *extack)
 733 {
 734         int err, ifindex = -1;
 735
 736         err = tcf_exts_validate_ex(net, tp, tb, est, &n->exts, flags,
 737                                    fl_flags, extack);
 738         if (err < 0)
 739                 return err;
 740
 741         if (tb[TCA_U32_INDEV]) {
 742                 ifindex = tcf_change_indev(net, tb[TCA_U32_INDEV], extack);
 743                 if (ifindex < 0)
 744                         return -EINVAL;
 745         }
 746
 747         if (tb[TCA_U32_LINK]) {
 748                 u32 handle = nla_get_u32(tb[TCA_U32_LINK]);
 749                 struct tc_u_hnode *ht_down = NULL, *ht_old;
 750
 751                 if (TC_U32_KEY(handle)) {
 752                         NL_SET_ERR_MSG_MOD(extack, "u32 Link handle must be a hash table");
 753                         return -EINVAL;
 754                 }
 755
 756                 if (handle) {
 757                         ht_down = u32_lookup_ht(tp->data, handle);
 758
 759                         if (!ht_down) {
 760                                 NL_SET_ERR_MSG_MOD(extack, "Link hash table not found");
 761                                 return -EINVAL;
 762                         }
 763                         if (ht_down->is_root) {
 764                                 NL_SET_ERR_MSG_MOD(extack, "Not linking to root node");
 765                                 return -EINVAL;
 766                         }
 767                         ht_down->refcnt++;
 768                 }
 769
 770                 ht_old = rtnl_dereference(n->ht_down);
 771                 rcu_assign_pointer(n->ht_down, ht_down);
 772
 773                 if (ht_old)
 774                         ht_old->refcnt--;
 775         }
 776
 777         if (ifindex >= 0)
 778                 n->ifindex = ifindex;
 779
 780         return 0;
 781 }
 782
 783 static void u32_replace_knode(struct tcf_proto *tp, struct tc_u_common *tp_c,
 784                               struct tc_u_knode *n)
 785 {
 786         struct tc_u_knode __rcu **ins;
 787         struct tc_u_knode *pins;
 788         struct tc_u_hnode *ht;
 789
 790         if (TC_U32_HTID(n->handle) == TC_U32_ROOT)
 791                 ht = rtnl_dereference(tp->root);
 792         else
 793                 ht = u32_lookup_ht(tp_c, TC_U32_HTID(n->handle));
 794
 795         ins = &ht->ht[TC_U32_HASH(n->handle)];
 796
 797         /* The node must always exist for it to be replaced if this is not the
 798          * case then something went very wrong elsewhere.
 799          */
 800         for (pins = rtnl_dereference(*ins); ;
 801              ins = &pins->next, pins = rtnl_dereference(*ins))
 802                 if (pins->handle == n->handle)
 803                         break;
 804
 805         idr_replace(&ht->handle_idr, n, n->handle);
 806         RCU_INIT_POINTER(n->next, pins->next);
 807         rcu_assign_pointer(*ins, n);
 808 }
 809
 810 static struct tc_u_knode *u32_init_knode(struct net *net, struct tcf_proto *tp,
 811                                          struct tc_u_knode *n)
 812 {
 813         struct tc_u_hnode *ht = rtnl_dereference(n->ht_down);
 814         struct tc_u32_sel *s = &n->sel;
 815         struct tc_u_knode *new;
 816
 817         new = kzalloc(struct_size(new, sel.keys, s->nkeys), GFP_KERNEL);
 818         if (!new)
 819                 return NULL;
 820
 821         RCU_INIT_POINTER(new->next, n->next);
 822         new->handle = n->handle;
 823         RCU_INIT_POINTER(new->ht_up, n->ht_up);
 824
 825         new->ifindex = n->ifindex;
 826         new->fshift = n->fshift;
 827         new->flags = n->flags;
 828         RCU_INIT_POINTER(new->ht_down, ht);
 829
 830 #ifdef CONFIG_CLS_U32_PERF
 831         /* Statistics may be incremented by readers during update
 832          * so we must keep them in tact. When the node is later destroyed
 833          * a special destroy call must be made to not free the pf memory.
 834          */
 835         new->pf = n->pf;
 836 #endif
 837
 838 #ifdef CONFIG_CLS_U32_MARK
 839         new->val = n->val;
 840         new->mask = n->mask;
 841         /* Similarly success statistics must be moved as pointers */
 842         new->pcpu_success = n->pcpu_success;
 843 #endif
 844         memcpy(&new->sel, s, struct_size(s, keys, s->nkeys));
 845
 846         if (tcf_exts_init(&new->exts, net, TCA_U32_ACT, TCA_U32_POLICE)) {
 847                 kfree(new);
 848                 return NULL;
 849         }
 850
 851         /* bump reference count as long as we hold pointer to structure */
 852         if (ht)
 853                 ht->refcnt++;
 854
 855         return new;
 856 }
 857
 858 static int u32_change(struct net *net, struct sk_buff *in_skb,
 859                       struct tcf_proto *tp, unsigned long base, u32 handle,
 860                       struct nlattr **tca, void **arg, u32 flags,
 861                       struct netlink_ext_ack *extack)
 862 {
 863         struct tc_u_common *tp_c = tp->data;
 864         struct tc_u_hnode *ht;
 865         struct tc_u_knode *n;
 866         struct tc_u32_sel *s;
 867         struct nlattr *opt = tca[TCA_OPTIONS];
 868         struct nlattr *tb[TCA_U32_MAX + 1];
 869         u32 htid, userflags = 0;
 870         size_t sel_size;
 871         int err;
 872
 873         if (!opt) {
 874                 if (handle) {
 875                         NL_SET_ERR_MSG_MOD(extack, "Filter handle requires options");
 876                         return -EINVAL;
 877                 } else {
 878                         return 0;
 879                 }
 880         }
 881
 882         err = nla_parse_nested_deprecated(tb, TCA_U32_MAX, opt, u32_policy,
 883                                           extack);
 884         if (err < 0)
 885                 return err;
 886
 887         if (tb[TCA_U32_FLAGS]) {
 888                 userflags = nla_get_u32(tb[TCA_U32_FLAGS]);
 889                 if (!tc_flags_valid(userflags)) {
 890                         NL_SET_ERR_MSG_MOD(extack, "Invalid filter flags");
 891                         return -EINVAL;
 892                 }
 893         }
 894
 895         n = *arg;
 896         if (n) {
 897                 struct tc_u_knode *new;
 898
 899                 if (TC_U32_KEY(n->handle) == 0) {
 900                         NL_SET_ERR_MSG_MOD(extack, "Key node id cannot be zero");
 901                         return -EINVAL;
 902                 }
 903
 904                 if ((n->flags ^ userflags) &
 905                     ~(TCA_CLS_FLAGS_IN_HW | TCA_CLS_FLAGS_NOT_IN_HW)) {
 906                         NL_SET_ERR_MSG_MOD(extack, "Key node flags do not match passed flags");
 907                         return -EINVAL;
 908                 }
 909
 910                 new = u32_init_knode(net, tp, n);
 911                 if (!new)
 912                         return -ENOMEM;
 913
 914                 err = u32_set_parms(net, tp, new, tb, tca[TCA_RATE],
 915                                     flags, new->flags, extack);
 916
 917                 if (err) {
 918                         __u32_destroy_key(new);
 919                         return err;
 920                 }
 921
 922                 u32_bind_filter(tp, new, base, tb);
 923
 924                 err = u32_replace_hw_knode(tp, new, flags, extack);
 925                 if (err) {
 926                         u32_unbind_filter(tp, new, tb);
 927
 928                         if (tb[TCA_U32_LINK]) {
 929                                 struct tc_u_hnode *ht_old;
 930
 931                                 ht_old = rtnl_dereference(n->ht_down);
 932                                 if (ht_old)
 933                                         ht_old->refcnt++;
 934                         }
 935                         __u32_destroy_key(new);
 936                         return err;
 937                 }
 938
 939                 if (!tc_in_hw(new->flags))
 940                         new->flags |= TCA_CLS_FLAGS_NOT_IN_HW;
 941
 942                 u32_replace_knode(tp, tp_c, new);
 943                 tcf_unbind_filter(tp, &n->res);
 944                 tcf_exts_get_net(&n->exts);
 945                 tcf_queue_work(&n->rwork, u32_delete_key_work);
 946                 return 0;
 947         }
 948
 949         if (tb[TCA_U32_DIVISOR]) {
 950                 unsigned int divisor = nla_get_u32(tb[TCA_U32_DIVISOR]);
 951
 952                 if (!is_power_of_2(divisor)) {
 953                         NL_SET_ERR_MSG_MOD(extack, "Divisor is not a power of 2");
 954                         return -EINVAL;
 955                 }
 956                 if (divisor-- > 0x100) {
 957                         NL_SET_ERR_MSG_MOD(extack, "Exceeded maximum 256 hash buckets");
 958                         return -EINVAL;
 959                 }
 960                 if (TC_U32_KEY(handle)) {
 961                         NL_SET_ERR_MSG_MOD(extack, "Divisor can only be used on a hash table");
 962                         return -EINVAL;
 963                 }
 964                 ht = kzalloc(struct_size(ht, ht, divisor + 1), GFP_KERNEL);
 965                 if (ht == NULL)
 966                         return -ENOBUFS;
 967                 if (handle == 0) {
 968                         handle = gen_new_htid(tp->data, ht);
 969                         if (handle == 0) {
 970                                 kfree(ht);
 971                                 return -ENOMEM;
 972                         }
 973                 } else {
 974                         err = idr_alloc_u32(&tp_c->handle_idr, ht, &handle,
 975                                             handle, GFP_KERNEL);
 976                         if (err) {
 977                                 kfree(ht);
 978                                 return err;
 979                         }
 980                 }
 981                 ht->refcnt = 1;
 982                 ht->divisor = divisor;
 983                 ht->handle = handle;
 984                 ht->prio = tp->prio;
 985                 idr_init(&ht->handle_idr);
 986                 ht->flags = userflags;
 987
 988                 err = u32_replace_hw_hnode(tp, ht, userflags, extack);
 989                 if (err) {
 990                         idr_remove(&tp_c->handle_idr, handle);
 991                         kfree(ht);
 992                         return err;
 993                 }
 994
 995                 RCU_INIT_POINTER(ht->next, tp_c->hlist);
 996                 rcu_assign_pointer(tp_c->hlist, ht);
 997                 *arg = ht;
 998
 999                 return 0;
1000         }
1001
1002         if (tb[TCA_U32_HASH]) {
1003                 htid = nla_get_u32(tb[TCA_U32_HASH]);
1004                 if (TC_U32_HTID(htid) == TC_U32_ROOT) {
1005                         ht = rtnl_dereference(tp->root);
1006                         htid = ht->handle;
1007                 } else {
1008                         ht = u32_lookup_ht(tp->data, TC_U32_HTID(htid));
1009                         if (!ht) {
1010                                 NL_SET_ERR_MSG_MOD(extack, "Specified hash table not found");
1011                                 return -EINVAL;
1012                         }
1013                 }
1014         } else {
1015                 ht = rtnl_dereference(tp->root);
1016                 htid = ht->handle;
1017         }
1018
1019         if (ht->divisor < TC_U32_HASH(htid)) {
1020                 NL_SET_ERR_MSG_MOD(extack, "Specified hash table buckets exceed configured value");
1021                 return -EINVAL;
1022         }
1023
1024         /* At this point, we need to derive the new handle that will be used to
1025          * uniquely map the identity of this table match entry. The
1026          * identity of the entry that we need to construct is 32 bits made of:
1027          *     htid(12b):bucketid(8b):node/entryid(12b)
1028          *
1029          * At this point _we have the table(ht)_ in which we will insert this
1030          * entry. We carry the table's id in variable "htid".
1031          * Note that earlier code picked the ht selection either by a) the user
1032          * providing the htid specified via TCA_U32_HASH attribute or b) when
1033          * no such attribute is passed then the root ht, is default to at ID
1034          * 0x[800][00][000]. Rule: the root table has a single bucket with ID 0.
1035          * If OTOH the user passed us the htid, they may also pass a bucketid of
1036          * choice. 0 is fine. For example a user htid is 0x[600][01][000] it is
1037          * indicating hash bucketid of 1. Rule: the entry/node ID _cannot_ be
1038          * passed via the htid, so even if it was non-zero it will be ignored.
1039          *
1040          * We may also have a handle, if the user passed one. The handle also
1041          * carries the same addressing of htid(12b):bucketid(8b):node/entryid(12b).
1042          * Rule: the bucketid on the handle is ignored even if one was passed;
1043          * rather the value on "htid" is always assumed to be the bucketid.
1044          */
1045         if (handle) {
1046                 /* Rule: The htid from handle and tableid from htid must match */
1047                 if (TC_U32_HTID(handle) && TC_U32_HTID(handle ^ htid)) {
1048                         NL_SET_ERR_MSG_MOD(extack, "Handle specified hash table address mismatch");
1049                         return -EINVAL;
1050                 }
1051                 /* Ok, so far we have a valid htid(12b):bucketid(8b) but we
1052                  * need to finalize the table entry identification with the last
1053                  * part - the node/entryid(12b)). Rule: Nodeid _cannot be 0_ for
1054                  * entries. Rule: nodeid of 0 is reserved only for tables(see
1055                  * earlier code which processes TC_U32_DIVISOR attribute).
1056                  * Rule: The nodeid can only be derived from the handle (and not
1057                  * htid).
1058                  * Rule: if the handle specified zero for the node id example
1059                  * 0x60000000, then pick a new nodeid from the pool of IDs
1060                  * this hash table has been allocating from.
1061                  * If OTOH it is specified (i.e for example the user passed a
1062                  * handle such as 0x60000123), then we use it generate our final
1063                  * handle which is used to uniquely identify the match entry.
1064                  */
1065                 if (!TC_U32_NODE(handle)) {
1066                         handle = gen_new_kid(ht, htid);
1067                 } else {
1068                         handle = htid | TC_U32_NODE(handle);
1069                         err = idr_alloc_u32(&ht->handle_idr, NULL, &handle,
1070                                             handle, GFP_KERNEL);
1071                         if (err)
1072                                 return err;
1073                 }
1074         } else {
1075                 /* The user did not give us a handle; lets just generate one
1076                  * from the table's pool of nodeids.
1077                  */
1078                 handle = gen_new_kid(ht, htid);
1079         }
1080
1081         if (tb[TCA_U32_SEL] == NULL) {
1082                 NL_SET_ERR_MSG_MOD(extack, "Selector not specified");
1083                 err = -EINVAL;
1084                 goto erridr;
1085         }
1086
1087         s = nla_data(tb[TCA_U32_SEL]);
1088         sel_size = struct_size(s, keys, s->nkeys);
1089         if (nla_len(tb[TCA_U32_SEL]) < sel_size) {
1090                 err = -EINVAL;
1091                 goto erridr;
1092         }
1093
1094         n = kzalloc(struct_size(n, sel.keys, s->nkeys), GFP_KERNEL);
1095         if (n == NULL) {
1096                 err = -ENOBUFS;
1097                 goto erridr;
1098         }
1099
1100 #ifdef CONFIG_CLS_U32_PERF
1101         n->pf = __alloc_percpu(struct_size(n->pf, kcnts, s->nkeys),
1102                                __alignof__(struct tc_u32_pcnt));
1103         if (!n->pf) {
1104                 err = -ENOBUFS;
1105                 goto errfree;
1106         }
1107 #endif
1108
1109         unsafe_memcpy(&n->sel, s, sel_size,
1110                       /* A composite flex-array structure destination,
1111                        * which was correctly sized with struct_size(),
1112                        * bounds-checked against nla_len(), and allocated
1113                        * above. */);
1114         RCU_INIT_POINTER(n->ht_up, ht);
1115         n->handle = handle;
1116         n->fshift = s->hmask ? ffs(ntohl(s->hmask)) - 1 : 0;
1117         n->flags = userflags;
1118
1119         err = tcf_exts_init(&n->exts, net, TCA_U32_ACT, TCA_U32_POLICE);
1120         if (err < 0)
1121                 goto errout;
1122
1123 #ifdef CONFIG_CLS_U32_MARK
1124         n->pcpu_success = alloc_percpu(u32);
1125         if (!n->pcpu_success) {
1126                 err = -ENOMEM;
1127                 goto errout;
1128         }
1129
1130         if (tb[TCA_U32_MARK]) {
1131                 struct tc_u32_mark *mark;
1132
1133                 mark = nla_data(tb[TCA_U32_MARK]);
1134                 n->val = mark->val;
1135                 n->mask = mark->mask;
1136         }
1137 #endif
1138
1139         err = u32_set_parms(net, tp, n, tb, tca[TCA_RATE],
1140                             flags, n->flags, extack);
1141
1142         u32_bind_filter(tp, n, base, tb);
1143
1144         if (err == 0) {
1145                 struct tc_u_knode __rcu **ins;
1146                 struct tc_u_knode *pins;
1147
1148                 err = u32_replace_hw_knode(tp, n, flags, extack);
1149                 if (err)
1150                         goto errunbind;
1151
1152                 if (!tc_in_hw(n->flags))
1153                         n->flags |= TCA_CLS_FLAGS_NOT_IN_HW;
1154
1155                 ins = &ht->ht[TC_U32_HASH(handle)];
1156                 for (pins = rtnl_dereference(*ins); pins;
1157                      ins = &pins->next, pins = rtnl_dereference(*ins))
1158                         if (TC_U32_NODE(handle) < TC_U32_NODE(pins->handle))
1159                                 break;
1160
1161                 RCU_INIT_POINTER(n->next, pins);
1162                 rcu_assign_pointer(*ins, n);
1163                 tp_c->knodes++;
1164                 *arg = n;
1165                 return 0;
1166         }
1167
1168 errunbind:
1169         u32_unbind_filter(tp, n, tb);
1170
1171 #ifdef CONFIG_CLS_U32_MARK
1172         free_percpu(n->pcpu_success);
1173 #endif
1174
1175 errout:
1176         tcf_exts_destroy(&n->exts);
1177 #ifdef CONFIG_CLS_U32_PERF
1178 errfree:
1179         free_percpu(n->pf);
1180 #endif
1181         kfree(n);
1182 erridr:
1183         idr_remove(&ht->handle_idr, handle);
1184         return err;
1185 }
1186
1187 static void u32_walk(struct tcf_proto *tp, struct tcf_walker *arg,
1188                      bool rtnl_held)
1189 {
1190         struct tc_u_common *tp_c = tp->data;
1191         struct tc_u_hnode *ht;
1192         struct tc_u_knode *n;
1193         unsigned int h;
1194
1195         if (arg->stop)
1196                 return;
1197
1198         for (ht = rtnl_dereference(tp_c->hlist);
1199              ht;
1200              ht = rtnl_dereference(ht->next)) {
1201                 if (ht->prio != tp->prio)
1202                         continue;
1203
1204                 if (!tc_cls_stats_dump(tp, arg, ht))
1205                         return;
1206
1207                 for (h = 0; h <= ht->divisor; h++) {
1208                         for (n = rtnl_dereference(ht->ht[h]);
1209                              n;
1210                              n = rtnl_dereference(n->next)) {
1211                                 if (!tc_cls_stats_dump(tp, arg, n))
1212                                         return;
1213                         }
1214                 }
1215         }
1216 }
1217
1218 static int u32_reoffload_hnode(struct tcf_proto *tp, struct tc_u_hnode *ht,
1219                                bool add, flow_setup_cb_t *cb, void *cb_priv,
1220                                struct netlink_ext_ack *extack)
1221 {
1222         struct tc_cls_u32_offload cls_u32 = {};
1223         int err;
1224
1225         tc_cls_common_offload_init(&cls_u32.common, tp, ht->flags, extack);
1226         cls_u32.command = add ? TC_CLSU32_NEW_HNODE : TC_CLSU32_DELETE_HNODE;
1227         cls_u32.hnode.divisor = ht->divisor;
1228         cls_u32.hnode.handle = ht->handle;
1229         cls_u32.hnode.prio = ht->prio;
1230
1231         err = cb(TC_SETUP_CLSU32, &cls_u32, cb_priv);
1232         if (err && add && tc_skip_sw(ht->flags))
1233                 return err;
1234
1235         return 0;
1236 }
1237
1238 static int u32_reoffload_knode(struct tcf_proto *tp, struct tc_u_knode *n,
1239                                bool add, flow_setup_cb_t *cb, void *cb_priv,
1240                                struct netlink_ext_ack *extack)
1241 {
1242         struct tc_u_hnode *ht = rtnl_dereference(n->ht_down);
1243         struct tcf_block *block = tp->chain->block;
1244         struct tc_cls_u32_offload cls_u32 = {};
1245
1246         tc_cls_common_offload_init(&cls_u32.common, tp, n->flags, extack);
1247         cls_u32.command = add ?
1248                 TC_CLSU32_REPLACE_KNODE : TC_CLSU32_DELETE_KNODE;
1249         cls_u32.knode.handle = n->handle;
1250
1251         if (add) {
1252                 cls_u32.knode.fshift = n->fshift;
1253 #ifdef CONFIG_CLS_U32_MARK
1254                 cls_u32.knode.val = n->val;
1255                 cls_u32.knode.mask = n->mask;
1256 #else
1257                 cls_u32.knode.val = 0;
1258                 cls_u32.knode.mask = 0;
1259 #endif
1260                 cls_u32.knode.sel = &n->sel;
1261                 cls_u32.knode.res = &n->res;
1262                 cls_u32.knode.exts = &n->exts;
1263                 if (n->ht_down)
1264                         cls_u32.knode.link_handle = ht->handle;
1265         }
1266
1267         return tc_setup_cb_reoffload(block, tp, add, cb, TC_SETUP_CLSU32,
1268                                      &cls_u32, cb_priv, &n->flags,
1269                                      &n->in_hw_count);
1270 }
1271
1272 static int u32_reoffload(struct tcf_proto *tp, bool add, flow_setup_cb_t *cb,
1273                          void *cb_priv, struct netlink_ext_ack *extack)
1274 {
1275         struct tc_u_common *tp_c = tp->data;
1276         struct tc_u_hnode *ht;
1277         struct tc_u_knode *n;
1278         unsigned int h;
1279         int err;
1280
1281         for (ht = rtnl_dereference(tp_c->hlist);
1282              ht;
1283              ht = rtnl_dereference(ht->next)) {
1284                 if (ht->prio != tp->prio)
1285                         continue;
1286
1287                 /* When adding filters to a new dev, try to offload the
1288                  * hashtable first. When removing, do the filters before the
1289                  * hashtable.
1290                  */
1291                 if (add && !tc_skip_hw(ht->flags)) {
1292                         err = u32_reoffload_hnode(tp, ht, add, cb, cb_priv,
1293                                                   extack);
1294                         if (err)
1295                                 return err;
1296                 }
1297
1298                 for (h = 0; h <= ht->divisor; h++) {
1299                         for (n = rtnl_dereference(ht->ht[h]);
1300                              n;
1301                              n = rtnl_dereference(n->next)) {
1302                                 if (tc_skip_hw(n->flags))
1303                                         continue;
1304
1305                                 err = u32_reoffload_knode(tp, n, add, cb,
1306                                                           cb_priv, extack);
1307                                 if (err)
1308                                         return err;
1309                         }
1310                 }
1311
1312                 if (!add && !tc_skip_hw(ht->flags))
1313                         u32_reoffload_hnode(tp, ht, add, cb, cb_priv, extack);
1314         }
1315
1316         return 0;
1317 }
1318
1319 static void u32_bind_class(void *fh, u32 classid, unsigned long cl, void *q,
1320                            unsigned long base)
1321 {
1322         struct tc_u_knode *n = fh;
1323
1324         tc_cls_bind_class(classid, cl, q, &n->res, base);
1325 }
1326
1327 static int u32_dump(struct net *net, struct tcf_proto *tp, void *fh,
1328                     struct sk_buff *skb, struct tcmsg *t, bool rtnl_held)
1329 {
1330         struct tc_u_knode *n = fh;
1331         struct tc_u_hnode *ht_up, *ht_down;
1332         struct nlattr *nest;
1333
1334         if (n == NULL)
1335                 return skb->len;
1336
1337         t->tcm_handle = n->handle;
1338
1339         nest = nla_nest_start_noflag(skb, TCA_OPTIONS);
1340         if (nest == NULL)
1341                 goto nla_put_failure;
1342
1343         if (TC_U32_KEY(n->handle) == 0) {
1344                 struct tc_u_hnode *ht = fh;
1345                 u32 divisor = ht->divisor + 1;
1346
1347                 if (nla_put_u32(skb, TCA_U32_DIVISOR, divisor))
1348                         goto nla_put_failure;
1349         } else {
1350 #ifdef CONFIG_CLS_U32_PERF
1351                 struct tc_u32_pcnt *gpf;
1352                 int cpu;
1353 #endif
1354
1355                 if (nla_put(skb, TCA_U32_SEL, struct_size(&n->sel, keys, n->sel.nkeys),
1356                             &n->sel))
1357                         goto nla_put_failure;
1358
1359                 ht_up = rtnl_dereference(n->ht_up);
1360                 if (ht_up) {
1361                         u32 htid = n->handle & 0xFFFFF000;
1362                         if (nla_put_u32(skb, TCA_U32_HASH, htid))
1363                                 goto nla_put_failure;
1364                 }
1365                 if (n->res.classid &&
1366                     nla_put_u32(skb, TCA_U32_CLASSID, n->res.classid))
1367                         goto nla_put_failure;
1368
1369                 ht_down = rtnl_dereference(n->ht_down);
1370                 if (ht_down &&
1371                     nla_put_u32(skb, TCA_U32_LINK, ht_down->handle))
1372                         goto nla_put_failure;
1373
1374                 if (n->flags && nla_put_u32(skb, TCA_U32_FLAGS, n->flags))
1375                         goto nla_put_failure;
1376
1377 #ifdef CONFIG_CLS_U32_MARK
1378                 if ((n->val || n->mask)) {
1379                         struct tc_u32_mark mark = {.val = n->val,
1380                                                    .mask = n->mask,
1381                                                    .success = 0};
1382                         int cpum;
1383
1384                         for_each_possible_cpu(cpum) {
1385                                 __u32 cnt = *per_cpu_ptr(n->pcpu_success, cpum);
1386
1387                                 mark.success += cnt;
1388                         }
1389
1390                         if (nla_put(skb, TCA_U32_MARK, sizeof(mark), &mark))
1391                                 goto nla_put_failure;
1392                 }
1393 #endif
1394
1395                 if (tcf_exts_dump(skb, &n->exts) < 0)
1396                         goto nla_put_failure;
1397
1398                 if (n->ifindex) {
1399                         struct net_device *dev;
1400                         dev = __dev_get_by_index(net, n->ifindex);
1401                         if (dev && nla_put_string(skb, TCA_U32_INDEV, dev->name))
1402                                 goto nla_put_failure;
1403                 }
1404 #ifdef CONFIG_CLS_U32_PERF
1405                 gpf = kzalloc(struct_size(gpf, kcnts, n->sel.nkeys), GFP_KERNEL);
1406                 if (!gpf)
1407                         goto nla_put_failure;
1408
1409                 for_each_possible_cpu(cpu) {
1410                         int i;
1411                         struct tc_u32_pcnt *pf = per_cpu_ptr(n->pf, cpu);
1412
1413                         gpf->rcnt += pf->rcnt;
1414                         gpf->rhit += pf->rhit;
1415                         for (i = 0; i < n->sel.nkeys; i++)
1416                                 gpf->kcnts[i] += pf->kcnts[i];
1417                 }
1418
1419                 if (nla_put_64bit(skb, TCA_U32_PCNT, struct_size(gpf, kcnts, n->sel.nkeys),
1420                                   gpf, TCA_U32_PAD)) {
1421                         kfree(gpf);
1422                         goto nla_put_failure;
1423                 }
1424                 kfree(gpf);
1425 #endif
1426         }
1427
1428         nla_nest_end(skb, nest);
1429
1430         if (TC_U32_KEY(n->handle))
1431                 if (tcf_exts_dump_stats(skb, &n->exts) < 0)
1432                         goto nla_put_failure;
1433         return skb->len;
1434
1435 nla_put_failure:
1436         nla_nest_cancel(skb, nest);
1437         return -1;
1438 }
1439
1440 static struct tcf_proto_ops cls_u32_ops __read_mostly = {
1441         .kind           =       "u32",
1442         .classify       =       u32_classify,
1443         .init           =       u32_init,
1444         .destroy        =       u32_destroy,
1445         .get            =       u32_get,
1446         .change         =       u32_change,
1447         .delete         =       u32_delete,
1448         .walk           =       u32_walk,
1449         .reoffload      =       u32_reoffload,
1450         .dump           =       u32_dump,
1451         .bind_class     =       u32_bind_class,
1452         .owner          =       THIS_MODULE,
1453 };
1454
1455 static int __init init_u32(void)
1456 {
1457         int i, ret;
1458
1459         pr_info("u32 classifier\n");
1460 #ifdef CONFIG_CLS_U32_PERF
1461         pr_info("    Performance counters on\n");
1462 #endif
1463         pr_info("    input device check on\n");
1464 #ifdef CONFIG_NET_CLS_ACT
1465         pr_info("    Actions configured\n");
1466 #endif
1467         tc_u_common_hash = kvmalloc_array(U32_HASH_SIZE,
1468                                           sizeof(struct hlist_head),
1469                                           GFP_KERNEL);
1470         if (!tc_u_common_hash)
1471                 return -ENOMEM;
1472
1473         for (i = 0; i < U32_HASH_SIZE; i++)
1474                 INIT_HLIST_HEAD(&tc_u_common_hash[i]);
1475
1476         ret = register_tcf_proto_ops(&cls_u32_ops);
1477         if (ret)
1478                 kvfree(tc_u_common_hash);
1479         return ret;
1480 }
1481
1482 static void __exit exit_u32(void)
1483 {
1484         unregister_tcf_proto_ops(&cls_u32_ops);
1485         kvfree(tc_u_common_hash);
1486 }
1487
1488 module_init(init_u32)
1489 module_exit(exit_u32)
1490 MODULE_LICENSE("GPL");