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[1];
  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(sizeof(*root_ht), 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(sizeof(*tp_c), 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 int u32_destroy_key(struct tc_u_knode *n, bool free_pf)
 390 {
 391         struct tc_u_hnode *ht = rtnl_dereference(n->ht_down);
 392
 393         tcf_exts_destroy(&n->exts);
 394         tcf_exts_put_net(&n->exts);
 395         if (ht && --ht->refcnt == 0)
 396                 kfree(ht);
 397 #ifdef CONFIG_CLS_U32_PERF
 398         if (free_pf)
 399                 free_percpu(n->pf);
 400 #endif
 401 #ifdef CONFIG_CLS_U32_MARK
 402         if (free_pf)
 403                 free_percpu(n->pcpu_success);
 404 #endif
 405         kfree(n);
 406         return 0;
 407 }
 408
 409 /* u32_delete_key_rcu should be called when free'ing a copied
 410  * version of a tc_u_knode obtained from u32_init_knode(). When
 411  * copies are obtained from u32_init_knode() the statistics are
 412  * shared between the old and new copies to allow readers to
 413  * continue to update the statistics during the copy. To support
 414  * this the u32_delete_key_rcu variant does not free the percpu
 415  * statistics.
 416  */
 417 static void u32_delete_key_work(struct work_struct *work)
 418 {
 419         struct tc_u_knode *key = container_of(to_rcu_work(work),
 420                                               struct tc_u_knode,
 421                                               rwork);
 422         rtnl_lock();
 423         u32_destroy_key(key, false);
 424         rtnl_unlock();
 425 }
 426
 427 /* u32_delete_key_freepf_rcu is the rcu callback variant
 428  * that free's the entire structure including the statistics
 429  * percpu variables. Only use this if the key is not a copy
 430  * returned by u32_init_knode(). See u32_delete_key_rcu()
 431  * for the variant that should be used with keys return from
 432  * u32_init_knode()
 433  */
 434 static void u32_delete_key_freepf_work(struct work_struct *work)
 435 {
 436         struct tc_u_knode *key = container_of(to_rcu_work(work),
 437                                               struct tc_u_knode,
 438                                               rwork);
 439         rtnl_lock();
 440         u32_destroy_key(key, true);
 441         rtnl_unlock();
 442 }
 443
 444 static int u32_delete_key(struct tcf_proto *tp, struct tc_u_knode *key)
 445 {
 446         struct tc_u_common *tp_c = tp->data;
 447         struct tc_u_knode __rcu **kp;
 448         struct tc_u_knode *pkp;
 449         struct tc_u_hnode *ht = rtnl_dereference(key->ht_up);
 450
 451         if (ht) {
 452                 kp = &ht->ht[TC_U32_HASH(key->handle)];
 453                 for (pkp = rtnl_dereference(*kp); pkp;
 454                      kp = &pkp->next, pkp = rtnl_dereference(*kp)) {
 455                         if (pkp == key) {
 456                                 RCU_INIT_POINTER(*kp, key->next);
 457                                 tp_c->knodes--;
 458
 459                                 tcf_unbind_filter(tp, &key->res);
 460                                 idr_remove(&ht->handle_idr, key->handle);
 461                                 tcf_exts_get_net(&key->exts);
 462                                 tcf_queue_work(&key->rwork, u32_delete_key_freepf_work);
 463                                 return 0;
 464                         }
 465                 }
 466         }
 467         WARN_ON(1);
 468         return 0;
 469 }
 470
 471 static void u32_clear_hw_hnode(struct tcf_proto *tp, struct tc_u_hnode *h,
 472                                struct netlink_ext_ack *extack)
 473 {
 474         struct tcf_block *block = tp->chain->block;
 475         struct tc_cls_u32_offload cls_u32 = {};
 476
 477         tc_cls_common_offload_init(&cls_u32.common, tp, h->flags, extack);
 478         cls_u32.command = TC_CLSU32_DELETE_HNODE;
 479         cls_u32.hnode.divisor = h->divisor;
 480         cls_u32.hnode.handle = h->handle;
 481         cls_u32.hnode.prio = h->prio;
 482
 483         tc_setup_cb_call(block, TC_SETUP_CLSU32, &cls_u32, false);
 484 }
 485
 486 static int u32_replace_hw_hnode(struct tcf_proto *tp, struct tc_u_hnode *h,
 487                                 u32 flags, struct netlink_ext_ack *extack)
 488 {
 489         struct tcf_block *block = tp->chain->block;
 490         struct tc_cls_u32_offload cls_u32 = {};
 491         bool skip_sw = tc_skip_sw(flags);
 492         bool offloaded = false;
 493         int err;
 494
 495         tc_cls_common_offload_init(&cls_u32.common, tp, flags, extack);
 496         cls_u32.command = TC_CLSU32_NEW_HNODE;
 497         cls_u32.hnode.divisor = h->divisor;
 498         cls_u32.hnode.handle = h->handle;
 499         cls_u32.hnode.prio = h->prio;
 500
 501         err = tc_setup_cb_call(block, TC_SETUP_CLSU32, &cls_u32, skip_sw);
 502         if (err < 0) {
 503                 u32_clear_hw_hnode(tp, h, NULL);
 504                 return err;
 505         } else if (err > 0) {
 506                 offloaded = true;
 507         }
 508
 509         if (skip_sw && !offloaded)
 510                 return -EINVAL;
 511
 512         return 0;
 513 }
 514
 515 static void u32_remove_hw_knode(struct tcf_proto *tp, struct tc_u_knode *n,
 516                                 struct netlink_ext_ack *extack)
 517 {
 518         struct tcf_block *block = tp->chain->block;
 519         struct tc_cls_u32_offload cls_u32 = {};
 520
 521         tc_cls_common_offload_init(&cls_u32.common, tp, n->flags, extack);
 522         cls_u32.command = TC_CLSU32_DELETE_KNODE;
 523         cls_u32.knode.handle = n->handle;
 524
 525         tc_setup_cb_call(block, TC_SETUP_CLSU32, &cls_u32, false);
 526         tcf_block_offload_dec(block, &n->flags);
 527 }
 528
 529 static int u32_replace_hw_knode(struct tcf_proto *tp, struct tc_u_knode *n,
 530                                 u32 flags, struct netlink_ext_ack *extack)
 531 {
 532         struct tc_u_hnode *ht = rtnl_dereference(n->ht_down);
 533         struct tcf_block *block = tp->chain->block;
 534         struct tc_cls_u32_offload cls_u32 = {};
 535         bool skip_sw = tc_skip_sw(flags);
 536         int err;
 537
 538         tc_cls_common_offload_init(&cls_u32.common, tp, flags, extack);
 539         cls_u32.command = TC_CLSU32_REPLACE_KNODE;
 540         cls_u32.knode.handle = n->handle;
 541         cls_u32.knode.fshift = n->fshift;
 542 #ifdef CONFIG_CLS_U32_MARK
 543         cls_u32.knode.val = n->val;
 544         cls_u32.knode.mask = n->mask;
 545 #else
 546         cls_u32.knode.val = 0;
 547         cls_u32.knode.mask = 0;
 548 #endif
 549         cls_u32.knode.sel = &n->sel;
 550         cls_u32.knode.res = &n->res;
 551         cls_u32.knode.exts = &n->exts;
 552         if (n->ht_down)
 553                 cls_u32.knode.link_handle = ht->handle;
 554
 555         err = tc_setup_cb_call(block, TC_SETUP_CLSU32, &cls_u32, skip_sw);
 556         if (err < 0) {
 557                 u32_remove_hw_knode(tp, n, NULL);
 558                 return err;
 559         } else if (err > 0) {
 560                 n->in_hw_count = err;
 561                 tcf_block_offload_inc(block, &n->flags);
 562         }
 563
 564         if (skip_sw && !(n->flags & TCA_CLS_FLAGS_IN_HW))
 565                 return -EINVAL;
 566
 567         return 0;
 568 }
 569
 570 static void u32_clear_hnode(struct tcf_proto *tp, struct tc_u_hnode *ht,
 571                             struct netlink_ext_ack *extack)
 572 {
 573         struct tc_u_common *tp_c = tp->data;
 574         struct tc_u_knode *n;
 575         unsigned int h;
 576
 577         for (h = 0; h <= ht->divisor; h++) {
 578                 while ((n = rtnl_dereference(ht->ht[h])) != NULL) {
 579                         RCU_INIT_POINTER(ht->ht[h],
 580                                          rtnl_dereference(n->next));
 581                         tp_c->knodes--;
 582                         tcf_unbind_filter(tp, &n->res);
 583                         u32_remove_hw_knode(tp, n, extack);
 584                         idr_remove(&ht->handle_idr, n->handle);
 585                         if (tcf_exts_get_net(&n->exts))
 586                                 tcf_queue_work(&n->rwork, u32_delete_key_freepf_work);
 587                         else
 588                                 u32_destroy_key(n, true);
 589                 }
 590         }
 591 }
 592
 593 static int u32_destroy_hnode(struct tcf_proto *tp, struct tc_u_hnode *ht,
 594                              struct netlink_ext_ack *extack)
 595 {
 596         struct tc_u_common *tp_c = tp->data;
 597         struct tc_u_hnode __rcu **hn;
 598         struct tc_u_hnode *phn;
 599
 600         WARN_ON(--ht->refcnt);
 601
 602         u32_clear_hnode(tp, ht, extack);
 603
 604         hn = &tp_c->hlist;
 605         for (phn = rtnl_dereference(*hn);
 606              phn;
 607              hn = &phn->next, phn = rtnl_dereference(*hn)) {
 608                 if (phn == ht) {
 609                         u32_clear_hw_hnode(tp, ht, extack);
 610                         idr_destroy(&ht->handle_idr);
 611                         idr_remove(&tp_c->handle_idr, ht->handle);
 612                         RCU_INIT_POINTER(*hn, ht->next);
 613                         kfree_rcu(ht, rcu);
 614                         return 0;
 615                 }
 616         }
 617
 618         return -ENOENT;
 619 }
 620
 621 static void u32_destroy(struct tcf_proto *tp, bool rtnl_held,
 622                         struct netlink_ext_ack *extack)
 623 {
 624         struct tc_u_common *tp_c = tp->data;
 625         struct tc_u_hnode *root_ht = rtnl_dereference(tp->root);
 626
 627         WARN_ON(root_ht == NULL);
 628
 629         if (root_ht && --root_ht->refcnt == 1)
 630                 u32_destroy_hnode(tp, root_ht, extack);
 631
 632         if (--tp_c->refcnt == 0) {
 633                 struct tc_u_hnode *ht;
 634
 635                 hlist_del(&tp_c->hnode);
 636
 637                 while ((ht = rtnl_dereference(tp_c->hlist)) != NULL) {
 638                         u32_clear_hnode(tp, ht, extack);
 639                         RCU_INIT_POINTER(tp_c->hlist, ht->next);
 640
 641                         /* u32_destroy_key() will later free ht for us, if it's
 642                          * still referenced by some knode
 643                          */
 644                         if (--ht->refcnt == 0)
 645                                 kfree_rcu(ht, rcu);
 646                 }
 647
 648                 idr_destroy(&tp_c->handle_idr);
 649                 kfree(tp_c);
 650         }
 651
 652         tp->data = NULL;
 653 }
 654
 655 static int u32_delete(struct tcf_proto *tp, void *arg, bool *last,
 656                       bool rtnl_held, struct netlink_ext_ack *extack)
 657 {
 658         struct tc_u_hnode *ht = arg;
 659         struct tc_u_common *tp_c = tp->data;
 660         int ret = 0;
 661
 662         if (TC_U32_KEY(ht->handle)) {
 663                 u32_remove_hw_knode(tp, (struct tc_u_knode *)ht, extack);
 664                 ret = u32_delete_key(tp, (struct tc_u_knode *)ht);
 665                 goto out;
 666         }
 667
 668         if (ht->is_root) {
 669                 NL_SET_ERR_MSG_MOD(extack, "Not allowed to delete root node");
 670                 return -EINVAL;
 671         }
 672
 673         if (ht->refcnt == 1) {
 674                 u32_destroy_hnode(tp, ht, extack);
 675         } else {
 676                 NL_SET_ERR_MSG_MOD(extack, "Can not delete in-use filter");
 677                 return -EBUSY;
 678         }
 679
 680 out:
 681         *last = tp_c->refcnt == 1 && tp_c->knodes == 0;
 682         return ret;
 683 }
 684
 685 static u32 gen_new_kid(struct tc_u_hnode *ht, u32 htid)
 686 {
 687         u32 index = htid | 0x800;
 688         u32 max = htid | 0xFFF;
 689
 690         if (idr_alloc_u32(&ht->handle_idr, NULL, &index, max, GFP_KERNEL)) {
 691                 index = htid + 1;
 692                 if (idr_alloc_u32(&ht->handle_idr, NULL, &index, max,
 693                                  GFP_KERNEL))
 694                         index = max;
 695         }
 696
 697         return index;
 698 }
 699
 700 static const struct nla_policy u32_policy[TCA_U32_MAX + 1] = {
 701         [TCA_U32_CLASSID]       = { .type = NLA_U32 },
 702         [TCA_U32_HASH]          = { .type = NLA_U32 },
 703         [TCA_U32_LINK]          = { .type = NLA_U32 },
 704         [TCA_U32_DIVISOR]       = { .type = NLA_U32 },
 705         [TCA_U32_SEL]           = { .len = sizeof(struct tc_u32_sel) },
 706         [TCA_U32_INDEV]         = { .type = NLA_STRING, .len = IFNAMSIZ },
 707         [TCA_U32_MARK]          = { .len = sizeof(struct tc_u32_mark) },
 708         [TCA_U32_FLAGS]         = { .type = NLA_U32 },
 709 };
 710
 711 static int u32_set_parms(struct net *net, struct tcf_proto *tp,
 712                          unsigned long base,
 713                          struct tc_u_knode *n, struct nlattr **tb,
 714                          struct nlattr *est, bool ovr,
 715                          struct netlink_ext_ack *extack)
 716 {
 717         int err;
 718
 719         err = tcf_exts_validate(net, tp, tb, est, &n->exts, ovr, true, extack);
 720         if (err < 0)
 721                 return err;
 722
 723         if (tb[TCA_U32_LINK]) {
 724                 u32 handle = nla_get_u32(tb[TCA_U32_LINK]);
 725                 struct tc_u_hnode *ht_down = NULL, *ht_old;
 726
 727                 if (TC_U32_KEY(handle)) {
 728                         NL_SET_ERR_MSG_MOD(extack, "u32 Link handle must be a hash table");
 729                         return -EINVAL;
 730                 }
 731
 732                 if (handle) {
 733                         ht_down = u32_lookup_ht(tp->data, handle);
 734
 735                         if (!ht_down) {
 736                                 NL_SET_ERR_MSG_MOD(extack, "Link hash table not found");
 737                                 return -EINVAL;
 738                         }
 739                         if (ht_down->is_root) {
 740                                 NL_SET_ERR_MSG_MOD(extack, "Not linking to root node");
 741                                 return -EINVAL;
 742                         }
 743                         ht_down->refcnt++;
 744                 }
 745
 746                 ht_old = rtnl_dereference(n->ht_down);
 747                 rcu_assign_pointer(n->ht_down, ht_down);
 748
 749                 if (ht_old)
 750                         ht_old->refcnt--;
 751         }
 752         if (tb[TCA_U32_CLASSID]) {
 753                 n->res.classid = nla_get_u32(tb[TCA_U32_CLASSID]);
 754                 tcf_bind_filter(tp, &n->res, base);
 755         }
 756
 757         if (tb[TCA_U32_INDEV]) {
 758                 int ret;
 759                 ret = tcf_change_indev(net, tb[TCA_U32_INDEV], extack);
 760                 if (ret < 0)
 761                         return -EINVAL;
 762                 n->ifindex = ret;
 763         }
 764         return 0;
 765 }
 766
 767 static void u32_replace_knode(struct tcf_proto *tp, struct tc_u_common *tp_c,
 768                               struct tc_u_knode *n)
 769 {
 770         struct tc_u_knode __rcu **ins;
 771         struct tc_u_knode *pins;
 772         struct tc_u_hnode *ht;
 773
 774         if (TC_U32_HTID(n->handle) == TC_U32_ROOT)
 775                 ht = rtnl_dereference(tp->root);
 776         else
 777                 ht = u32_lookup_ht(tp_c, TC_U32_HTID(n->handle));
 778
 779         ins = &ht->ht[TC_U32_HASH(n->handle)];
 780
 781         /* The node must always exist for it to be replaced if this is not the
 782          * case then something went very wrong elsewhere.
 783          */
 784         for (pins = rtnl_dereference(*ins); ;
 785              ins = &pins->next, pins = rtnl_dereference(*ins))
 786                 if (pins->handle == n->handle)
 787                         break;
 788
 789         idr_replace(&ht->handle_idr, n, n->handle);
 790         RCU_INIT_POINTER(n->next, pins->next);
 791         rcu_assign_pointer(*ins, n);
 792 }
 793
 794 static struct tc_u_knode *u32_init_knode(struct net *net, struct tcf_proto *tp,
 795                                          struct tc_u_knode *n)
 796 {
 797         struct tc_u_hnode *ht = rtnl_dereference(n->ht_down);
 798         struct tc_u32_sel *s = &n->sel;
 799         struct tc_u_knode *new;
 800
 801         new = kzalloc(sizeof(*n) + s->nkeys*sizeof(struct tc_u32_key),
 802                       GFP_KERNEL);
 803
 804         if (!new)
 805                 return NULL;
 806
 807         RCU_INIT_POINTER(new->next, n->next);
 808         new->handle = n->handle;
 809         RCU_INIT_POINTER(new->ht_up, n->ht_up);
 810
 811         new->ifindex = n->ifindex;
 812         new->fshift = n->fshift;
 813         new->res = n->res;
 814         new->flags = n->flags;
 815         RCU_INIT_POINTER(new->ht_down, ht);
 816
 817         /* bump reference count as long as we hold pointer to structure */
 818         if (ht)
 819                 ht->refcnt++;
 820
 821 #ifdef CONFIG_CLS_U32_PERF
 822         /* Statistics may be incremented by readers during update
 823          * so we must keep them in tact. When the node is later destroyed
 824          * a special destroy call must be made to not free the pf memory.
 825          */
 826         new->pf = n->pf;
 827 #endif
 828
 829 #ifdef CONFIG_CLS_U32_MARK
 830         new->val = n->val;
 831         new->mask = n->mask;
 832         /* Similarly success statistics must be moved as pointers */
 833         new->pcpu_success = n->pcpu_success;
 834 #endif
 835         memcpy(&new->sel, s, struct_size(s, keys, s->nkeys));
 836
 837         if (tcf_exts_init(&new->exts, net, TCA_U32_ACT, TCA_U32_POLICE)) {
 838                 kfree(new);
 839                 return NULL;
 840         }
 841
 842         return new;
 843 }
 844
 845 static int u32_change(struct net *net, struct sk_buff *in_skb,
 846                       struct tcf_proto *tp, unsigned long base, u32 handle,
 847                       struct nlattr **tca, void **arg, bool ovr, bool rtnl_held,
 848                       struct netlink_ext_ack *extack)
 849 {
 850         struct tc_u_common *tp_c = tp->data;
 851         struct tc_u_hnode *ht;
 852         struct tc_u_knode *n;
 853         struct tc_u32_sel *s;
 854         struct nlattr *opt = tca[TCA_OPTIONS];
 855         struct nlattr *tb[TCA_U32_MAX + 1];
 856         u32 htid, flags = 0;
 857         size_t sel_size;
 858         int err;
 859 #ifdef CONFIG_CLS_U32_PERF
 860         size_t size;
 861 #endif
 862
 863         if (!opt) {
 864                 if (handle) {
 865                         NL_SET_ERR_MSG_MOD(extack, "Filter handle requires options");
 866                         return -EINVAL;
 867                 } else {
 868                         return 0;
 869                 }
 870         }
 871
 872         err = nla_parse_nested_deprecated(tb, TCA_U32_MAX, opt, u32_policy,
 873                                           extack);
 874         if (err < 0)
 875                 return err;
 876
 877         if (tb[TCA_U32_FLAGS]) {
 878                 flags = nla_get_u32(tb[TCA_U32_FLAGS]);
 879                 if (!tc_flags_valid(flags)) {
 880                         NL_SET_ERR_MSG_MOD(extack, "Invalid filter flags");
 881                         return -EINVAL;
 882                 }
 883         }
 884
 885         n = *arg;
 886         if (n) {
 887                 struct tc_u_knode *new;
 888
 889                 if (TC_U32_KEY(n->handle) == 0) {
 890                         NL_SET_ERR_MSG_MOD(extack, "Key node id cannot be zero");
 891                         return -EINVAL;
 892                 }
 893
 894                 if ((n->flags ^ flags) &
 895                     ~(TCA_CLS_FLAGS_IN_HW | TCA_CLS_FLAGS_NOT_IN_HW)) {
 896                         NL_SET_ERR_MSG_MOD(extack, "Key node flags do not match passed flags");
 897                         return -EINVAL;
 898                 }
 899
 900                 new = u32_init_knode(net, tp, n);
 901                 if (!new)
 902                         return -ENOMEM;
 903
 904                 err = u32_set_parms(net, tp, base, new, tb,
 905                                     tca[TCA_RATE], ovr, extack);
 906
 907                 if (err) {
 908                         u32_destroy_key(new, false);
 909                         return err;
 910                 }
 911
 912                 err = u32_replace_hw_knode(tp, new, flags, extack);
 913                 if (err) {
 914                         u32_destroy_key(new, false);
 915                         return err;
 916                 }
 917
 918                 if (!tc_in_hw(new->flags))
 919                         new->flags |= TCA_CLS_FLAGS_NOT_IN_HW;
 920
 921                 u32_replace_knode(tp, tp_c, new);
 922                 tcf_unbind_filter(tp, &n->res);
 923                 tcf_exts_get_net(&n->exts);
 924                 tcf_queue_work(&n->rwork, u32_delete_key_work);
 925                 return 0;
 926         }
 927
 928         if (tb[TCA_U32_DIVISOR]) {
 929                 unsigned int divisor = nla_get_u32(tb[TCA_U32_DIVISOR]);
 930
 931                 if (!is_power_of_2(divisor)) {
 932                         NL_SET_ERR_MSG_MOD(extack, "Divisor is not a power of 2");
 933                         return -EINVAL;
 934                 }
 935                 if (divisor-- > 0x100) {
 936                         NL_SET_ERR_MSG_MOD(extack, "Exceeded maximum 256 hash buckets");
 937                         return -EINVAL;
 938                 }
 939                 if (TC_U32_KEY(handle)) {
 940                         NL_SET_ERR_MSG_MOD(extack, "Divisor can only be used on a hash table");
 941                         return -EINVAL;
 942                 }
 943                 ht = kzalloc(sizeof(*ht) + divisor*sizeof(void *), GFP_KERNEL);
 944                 if (ht == NULL)
 945                         return -ENOBUFS;
 946                 if (handle == 0) {
 947                         handle = gen_new_htid(tp->data, ht);
 948                         if (handle == 0) {
 949                                 kfree(ht);
 950                                 return -ENOMEM;
 951                         }
 952                 } else {
 953                         err = idr_alloc_u32(&tp_c->handle_idr, ht, &handle,
 954                                             handle, GFP_KERNEL);
 955                         if (err) {
 956                                 kfree(ht);
 957                                 return err;
 958                         }
 959                 }
 960                 ht->refcnt = 1;
 961                 ht->divisor = divisor;
 962                 ht->handle = handle;
 963                 ht->prio = tp->prio;
 964                 idr_init(&ht->handle_idr);
 965                 ht->flags = flags;
 966
 967                 err = u32_replace_hw_hnode(tp, ht, flags, extack);
 968                 if (err) {
 969                         idr_remove(&tp_c->handle_idr, handle);
 970                         kfree(ht);
 971                         return err;
 972                 }
 973
 974                 RCU_INIT_POINTER(ht->next, tp_c->hlist);
 975                 rcu_assign_pointer(tp_c->hlist, ht);
 976                 *arg = ht;
 977
 978                 return 0;
 979         }
 980
 981         if (tb[TCA_U32_HASH]) {
 982                 htid = nla_get_u32(tb[TCA_U32_HASH]);
 983                 if (TC_U32_HTID(htid) == TC_U32_ROOT) {
 984                         ht = rtnl_dereference(tp->root);
 985                         htid = ht->handle;
 986                 } else {
 987                         ht = u32_lookup_ht(tp->data, TC_U32_HTID(htid));
 988                         if (!ht) {
 989                                 NL_SET_ERR_MSG_MOD(extack, "Specified hash table not found");
 990                                 return -EINVAL;
 991                         }
 992                 }
 993         } else {
 994                 ht = rtnl_dereference(tp->root);
 995                 htid = ht->handle;
 996         }
 997
 998         if (ht->divisor < TC_U32_HASH(htid)) {
 999                 NL_SET_ERR_MSG_MOD(extack, "Specified hash table buckets exceed configured value");
1000                 return -EINVAL;
1001         }
1002
1003         if (handle) {
1004                 if (TC_U32_HTID(handle) && TC_U32_HTID(handle ^ htid)) {
1005                         NL_SET_ERR_MSG_MOD(extack, "Handle specified hash table address mismatch");
1006                         return -EINVAL;
1007                 }
1008                 handle = htid | TC_U32_NODE(handle);
1009                 err = idr_alloc_u32(&ht->handle_idr, NULL, &handle, handle,
1010                                     GFP_KERNEL);
1011                 if (err)
1012                         return err;
1013         } else
1014                 handle = gen_new_kid(ht, htid);
1015
1016         if (tb[TCA_U32_SEL] == NULL) {
1017                 NL_SET_ERR_MSG_MOD(extack, "Selector not specified");
1018                 err = -EINVAL;
1019                 goto erridr;
1020         }
1021
1022         s = nla_data(tb[TCA_U32_SEL]);
1023         sel_size = struct_size(s, keys, s->nkeys);
1024         if (nla_len(tb[TCA_U32_SEL]) < sel_size) {
1025                 err = -EINVAL;
1026                 goto erridr;
1027         }
1028
1029         n = kzalloc(offsetof(typeof(*n), sel) + sel_size, GFP_KERNEL);
1030         if (n == NULL) {
1031                 err = -ENOBUFS;
1032                 goto erridr;
1033         }
1034
1035 #ifdef CONFIG_CLS_U32_PERF
1036         size = sizeof(struct tc_u32_pcnt) + s->nkeys * sizeof(u64);
1037         n->pf = __alloc_percpu(size, __alignof__(struct tc_u32_pcnt));
1038         if (!n->pf) {
1039                 err = -ENOBUFS;
1040                 goto errfree;
1041         }
1042 #endif
1043
1044         memcpy(&n->sel, s, sel_size);
1045         RCU_INIT_POINTER(n->ht_up, ht);
1046         n->handle = handle;
1047         n->fshift = s->hmask ? ffs(ntohl(s->hmask)) - 1 : 0;
1048         n->flags = flags;
1049
1050         err = tcf_exts_init(&n->exts, net, TCA_U32_ACT, TCA_U32_POLICE);
1051         if (err < 0)
1052                 goto errout;
1053
1054 #ifdef CONFIG_CLS_U32_MARK
1055         n->pcpu_success = alloc_percpu(u32);
1056         if (!n->pcpu_success) {
1057                 err = -ENOMEM;
1058                 goto errout;
1059         }
1060
1061         if (tb[TCA_U32_MARK]) {
1062                 struct tc_u32_mark *mark;
1063
1064                 mark = nla_data(tb[TCA_U32_MARK]);
1065                 n->val = mark->val;
1066                 n->mask = mark->mask;
1067         }
1068 #endif
1069
1070         err = u32_set_parms(net, tp, base, n, tb, tca[TCA_RATE], ovr,
1071                             extack);
1072         if (err == 0) {
1073                 struct tc_u_knode __rcu **ins;
1074                 struct tc_u_knode *pins;
1075
1076                 err = u32_replace_hw_knode(tp, n, flags, extack);
1077                 if (err)
1078                         goto errhw;
1079
1080                 if (!tc_in_hw(n->flags))
1081                         n->flags |= TCA_CLS_FLAGS_NOT_IN_HW;
1082
1083                 ins = &ht->ht[TC_U32_HASH(handle)];
1084                 for (pins = rtnl_dereference(*ins); pins;
1085                      ins = &pins->next, pins = rtnl_dereference(*ins))
1086                         if (TC_U32_NODE(handle) < TC_U32_NODE(pins->handle))
1087                                 break;
1088
1089                 RCU_INIT_POINTER(n->next, pins);
1090                 rcu_assign_pointer(*ins, n);
1091                 tp_c->knodes++;
1092                 *arg = n;
1093                 return 0;
1094         }
1095
1096 errhw:
1097 #ifdef CONFIG_CLS_U32_MARK
1098         free_percpu(n->pcpu_success);
1099 #endif
1100
1101 errout:
1102         tcf_exts_destroy(&n->exts);
1103 #ifdef CONFIG_CLS_U32_PERF
1104 errfree:
1105         free_percpu(n->pf);
1106 #endif
1107         kfree(n);
1108 erridr:
1109         idr_remove(&ht->handle_idr, handle);
1110         return err;
1111 }
1112
1113 static void u32_walk(struct tcf_proto *tp, struct tcf_walker *arg,
1114                      bool rtnl_held)
1115 {
1116         struct tc_u_common *tp_c = tp->data;
1117         struct tc_u_hnode *ht;
1118         struct tc_u_knode *n;
1119         unsigned int h;
1120
1121         if (arg->stop)
1122                 return;
1123
1124         for (ht = rtnl_dereference(tp_c->hlist);
1125              ht;
1126              ht = rtnl_dereference(ht->next)) {
1127                 if (ht->prio != tp->prio)
1128                         continue;
1129                 if (arg->count >= arg->skip) {
1130                         if (arg->fn(tp, ht, arg) < 0) {
1131                                 arg->stop = 1;
1132                                 return;
1133                         }
1134                 }
1135                 arg->count++;
1136                 for (h = 0; h <= ht->divisor; h++) {
1137                         for (n = rtnl_dereference(ht->ht[h]);
1138                              n;
1139                              n = rtnl_dereference(n->next)) {
1140                                 if (arg->count < arg->skip) {
1141                                         arg->count++;
1142                                         continue;
1143                                 }
1144                                 if (arg->fn(tp, n, arg) < 0) {
1145                                         arg->stop = 1;
1146                                         return;
1147                                 }
1148                                 arg->count++;
1149                         }
1150                 }
1151         }
1152 }
1153
1154 static int u32_reoffload_hnode(struct tcf_proto *tp, struct tc_u_hnode *ht,
1155                                bool add, flow_setup_cb_t *cb, void *cb_priv,
1156                                struct netlink_ext_ack *extack)
1157 {
1158         struct tc_cls_u32_offload cls_u32 = {};
1159         int err;
1160
1161         tc_cls_common_offload_init(&cls_u32.common, tp, ht->flags, extack);
1162         cls_u32.command = add ? TC_CLSU32_NEW_HNODE : TC_CLSU32_DELETE_HNODE;
1163         cls_u32.hnode.divisor = ht->divisor;
1164         cls_u32.hnode.handle = ht->handle;
1165         cls_u32.hnode.prio = ht->prio;
1166
1167         err = cb(TC_SETUP_CLSU32, &cls_u32, cb_priv);
1168         if (err && add && tc_skip_sw(ht->flags))
1169                 return err;
1170
1171         return 0;
1172 }
1173
1174 static int u32_reoffload_knode(struct tcf_proto *tp, struct tc_u_knode *n,
1175                                bool add, flow_setup_cb_t *cb, void *cb_priv,
1176                                struct netlink_ext_ack *extack)
1177 {
1178         struct tc_u_hnode *ht = rtnl_dereference(n->ht_down);
1179         struct tcf_block *block = tp->chain->block;
1180         struct tc_cls_u32_offload cls_u32 = {};
1181         int err;
1182
1183         tc_cls_common_offload_init(&cls_u32.common, tp, n->flags, extack);
1184         cls_u32.command = add ?
1185                 TC_CLSU32_REPLACE_KNODE : TC_CLSU32_DELETE_KNODE;
1186         cls_u32.knode.handle = n->handle;
1187
1188         if (add) {
1189                 cls_u32.knode.fshift = n->fshift;
1190 #ifdef CONFIG_CLS_U32_MARK
1191                 cls_u32.knode.val = n->val;
1192                 cls_u32.knode.mask = n->mask;
1193 #else
1194                 cls_u32.knode.val = 0;
1195                 cls_u32.knode.mask = 0;
1196 #endif
1197                 cls_u32.knode.sel = &n->sel;
1198                 cls_u32.knode.res = &n->res;
1199                 cls_u32.knode.exts = &n->exts;
1200                 if (n->ht_down)
1201                         cls_u32.knode.link_handle = ht->handle;
1202         }
1203
1204         err = cb(TC_SETUP_CLSU32, &cls_u32, cb_priv);
1205         if (err) {
1206                 if (add && tc_skip_sw(n->flags))
1207                         return err;
1208                 return 0;
1209         }
1210
1211         tc_cls_offload_cnt_update(block, &n->in_hw_count, &n->flags, add);
1212
1213         return 0;
1214 }
1215
1216 static int u32_reoffload(struct tcf_proto *tp, bool add, flow_setup_cb_t *cb,
1217                          void *cb_priv, struct netlink_ext_ack *extack)
1218 {
1219         struct tc_u_common *tp_c = tp->data;
1220         struct tc_u_hnode *ht;
1221         struct tc_u_knode *n;
1222         unsigned int h;
1223         int err;
1224
1225         for (ht = rtnl_dereference(tp_c->hlist);
1226              ht;
1227              ht = rtnl_dereference(ht->next)) {
1228                 if (ht->prio != tp->prio)
1229                         continue;
1230
1231                 /* When adding filters to a new dev, try to offload the
1232                  * hashtable first. When removing, do the filters before the
1233                  * hashtable.
1234                  */
1235                 if (add && !tc_skip_hw(ht->flags)) {
1236                         err = u32_reoffload_hnode(tp, ht, add, cb, cb_priv,
1237                                                   extack);
1238                         if (err)
1239                                 return err;
1240                 }
1241
1242                 for (h = 0; h <= ht->divisor; h++) {
1243                         for (n = rtnl_dereference(ht->ht[h]);
1244                              n;
1245                              n = rtnl_dereference(n->next)) {
1246                                 if (tc_skip_hw(n->flags))
1247                                         continue;
1248
1249                                 err = u32_reoffload_knode(tp, n, add, cb,
1250                                                           cb_priv, extack);
1251                                 if (err)
1252                                         return err;
1253                         }
1254                 }
1255
1256                 if (!add && !tc_skip_hw(ht->flags))
1257                         u32_reoffload_hnode(tp, ht, add, cb, cb_priv, extack);
1258         }
1259
1260         return 0;
1261 }
1262
1263 static void u32_bind_class(void *fh, u32 classid, unsigned long cl)
1264 {
1265         struct tc_u_knode *n = fh;
1266
1267         if (n && n->res.classid == classid)
1268                 n->res.class = cl;
1269 }
1270
1271 static int u32_dump(struct net *net, struct tcf_proto *tp, void *fh,
1272                     struct sk_buff *skb, struct tcmsg *t, bool rtnl_held)
1273 {
1274         struct tc_u_knode *n = fh;
1275         struct tc_u_hnode *ht_up, *ht_down;
1276         struct nlattr *nest;
1277
1278         if (n == NULL)
1279                 return skb->len;
1280
1281         t->tcm_handle = n->handle;
1282
1283         nest = nla_nest_start_noflag(skb, TCA_OPTIONS);
1284         if (nest == NULL)
1285                 goto nla_put_failure;
1286
1287         if (TC_U32_KEY(n->handle) == 0) {
1288                 struct tc_u_hnode *ht = fh;
1289                 u32 divisor = ht->divisor + 1;
1290
1291                 if (nla_put_u32(skb, TCA_U32_DIVISOR, divisor))
1292                         goto nla_put_failure;
1293         } else {
1294 #ifdef CONFIG_CLS_U32_PERF
1295                 struct tc_u32_pcnt *gpf;
1296                 int cpu;
1297 #endif
1298
1299                 if (nla_put(skb, TCA_U32_SEL,
1300                             sizeof(n->sel) + n->sel.nkeys*sizeof(struct tc_u32_key),
1301                             &n->sel))
1302                         goto nla_put_failure;
1303
1304                 ht_up = rtnl_dereference(n->ht_up);
1305                 if (ht_up) {
1306                         u32 htid = n->handle & 0xFFFFF000;
1307                         if (nla_put_u32(skb, TCA_U32_HASH, htid))
1308                                 goto nla_put_failure;
1309                 }
1310                 if (n->res.classid &&
1311                     nla_put_u32(skb, TCA_U32_CLASSID, n->res.classid))
1312                         goto nla_put_failure;
1313
1314                 ht_down = rtnl_dereference(n->ht_down);
1315                 if (ht_down &&
1316                     nla_put_u32(skb, TCA_U32_LINK, ht_down->handle))
1317                         goto nla_put_failure;
1318
1319                 if (n->flags && nla_put_u32(skb, TCA_U32_FLAGS, n->flags))
1320                         goto nla_put_failure;
1321
1322 #ifdef CONFIG_CLS_U32_MARK
1323                 if ((n->val || n->mask)) {
1324                         struct tc_u32_mark mark = {.val = n->val,
1325                                                    .mask = n->mask,
1326                                                    .success = 0};
1327                         int cpum;
1328
1329                         for_each_possible_cpu(cpum) {
1330                                 __u32 cnt = *per_cpu_ptr(n->pcpu_success, cpum);
1331
1332                                 mark.success += cnt;
1333                         }
1334
1335                         if (nla_put(skb, TCA_U32_MARK, sizeof(mark), &mark))
1336                                 goto nla_put_failure;
1337                 }
1338 #endif
1339
1340                 if (tcf_exts_dump(skb, &n->exts) < 0)
1341                         goto nla_put_failure;
1342
1343                 if (n->ifindex) {
1344                         struct net_device *dev;
1345                         dev = __dev_get_by_index(net, n->ifindex);
1346                         if (dev && nla_put_string(skb, TCA_U32_INDEV, dev->name))
1347                                 goto nla_put_failure;
1348                 }
1349 #ifdef CONFIG_CLS_U32_PERF
1350                 gpf = kzalloc(sizeof(struct tc_u32_pcnt) +
1351                               n->sel.nkeys * sizeof(u64),
1352                               GFP_KERNEL);
1353                 if (!gpf)
1354                         goto nla_put_failure;
1355
1356                 for_each_possible_cpu(cpu) {
1357                         int i;
1358                         struct tc_u32_pcnt *pf = per_cpu_ptr(n->pf, cpu);
1359
1360                         gpf->rcnt += pf->rcnt;
1361                         gpf->rhit += pf->rhit;
1362                         for (i = 0; i < n->sel.nkeys; i++)
1363                                 gpf->kcnts[i] += pf->kcnts[i];
1364                 }
1365
1366                 if (nla_put_64bit(skb, TCA_U32_PCNT,
1367                                   sizeof(struct tc_u32_pcnt) +
1368                                   n->sel.nkeys * sizeof(u64),
1369                                   gpf, TCA_U32_PAD)) {
1370                         kfree(gpf);
1371                         goto nla_put_failure;
1372                 }
1373                 kfree(gpf);
1374 #endif
1375         }
1376
1377         nla_nest_end(skb, nest);
1378
1379         if (TC_U32_KEY(n->handle))
1380                 if (tcf_exts_dump_stats(skb, &n->exts) < 0)
1381                         goto nla_put_failure;
1382         return skb->len;
1383
1384 nla_put_failure:
1385         nla_nest_cancel(skb, nest);
1386         return -1;
1387 }
1388
1389 static struct tcf_proto_ops cls_u32_ops __read_mostly = {
1390         .kind           =       "u32",
1391         .classify       =       u32_classify,
1392         .init           =       u32_init,
1393         .destroy        =       u32_destroy,
1394         .get            =       u32_get,
1395         .change         =       u32_change,
1396         .delete         =       u32_delete,
1397         .walk           =       u32_walk,
1398         .reoffload      =       u32_reoffload,
1399         .dump           =       u32_dump,
1400         .bind_class     =       u32_bind_class,
1401         .owner          =       THIS_MODULE,
1402 };
1403
1404 static int __init init_u32(void)
1405 {
1406         int i, ret;
1407
1408         pr_info("u32 classifier\n");
1409 #ifdef CONFIG_CLS_U32_PERF
1410         pr_info("    Performance counters on\n");
1411 #endif
1412         pr_info("    input device check on\n");
1413 #ifdef CONFIG_NET_CLS_ACT
1414         pr_info("    Actions configured\n");
1415 #endif
1416         tc_u_common_hash = kvmalloc_array(U32_HASH_SIZE,
1417                                           sizeof(struct hlist_head),
1418                                           GFP_KERNEL);
1419         if (!tc_u_common_hash)
1420                 return -ENOMEM;
1421
1422         for (i = 0; i < U32_HASH_SIZE; i++)
1423                 INIT_HLIST_HEAD(&tc_u_common_hash[i]);
1424
1425         ret = register_tcf_proto_ops(&cls_u32_ops);
1426         if (ret)
1427                 kvfree(tc_u_common_hash);
1428         return ret;
1429 }
1430
1431 static void __exit exit_u32(void)
1432 {
1433         unregister_tcf_proto_ops(&cls_u32_ops);
1434         kvfree(tc_u_common_hash);
1435 }
1436
1437 module_init(init_u32)
1438 module_exit(exit_u32)
1439 MODULE_LICENSE("GPL");