net/sched/cls_api.c

   1 // SPDX-License-Identifier: GPL-2.0-or-later
   2 /*
   3  * net/sched/cls_api.c  Packet classifier API.
   4  *
   5  * Authors:     Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
   6  *
   7  * Changes:
   8  *
   9  * Eduardo J. Blanco <ejbs@netlabs.com.uy> :990222: kmod support
  10  */
  11
  12 #include <linux/module.h>
  13 #include <linux/types.h>
  14 #include <linux/kernel.h>
  15 #include <linux/string.h>
  16 #include <linux/errno.h>
  17 #include <linux/err.h>
  18 #include <linux/skbuff.h>
  19 #include <linux/init.h>
  20 #include <linux/kmod.h>
  21 #include <linux/slab.h>
  22 #include <linux/idr.h>
  23 #include <linux/jhash.h>
  24 #include <linux/rculist.h>
  25 #include <linux/rhashtable.h>
  26 #include <net/net_namespace.h>
  27 #include <net/sock.h>
  28 #include <net/netlink.h>
  29 #include <net/pkt_sched.h>
  30 #include <net/pkt_cls.h>
  31 #include <net/tc_act/tc_pedit.h>
  32 #include <net/tc_act/tc_mirred.h>
  33 #include <net/tc_act/tc_vlan.h>
  34 #include <net/tc_act/tc_tunnel_key.h>
  35 #include <net/tc_act/tc_csum.h>
  36 #include <net/tc_act/tc_gact.h>
  37 #include <net/tc_act/tc_police.h>
  38 #include <net/tc_act/tc_sample.h>
  39 #include <net/tc_act/tc_skbedit.h>
  40 #include <net/tc_act/tc_ct.h>
  41 #include <net/tc_act/tc_mpls.h>
  42 #include <net/tc_act/tc_gate.h>
  43 #include <net/flow_offload.h>
  44 #include <net/tc_wrapper.h>
  45
  46 /* The list of all installed classifier types */
  47 static LIST_HEAD(tcf_proto_base);
  48
  49 /* Protects list of registered TC modules. It is pure SMP lock. */
  50 static DEFINE_RWLOCK(cls_mod_lock);
  51
  52 static struct xarray tcf_exts_miss_cookies_xa;
  53 struct tcf_exts_miss_cookie_node {
  54         const struct tcf_chain *chain;
  55         const struct tcf_proto *tp;
  56         const struct tcf_exts *exts;
  57         u32 chain_index;
  58         u32 tp_prio;
  59         u32 handle;
  60         u32 miss_cookie_base;
  61         struct rcu_head rcu;
  62 };
  63
  64 /* Each tc action entry cookie will be comprised of 32bit miss_cookie_base +
  65  * action index in the exts tc actions array.
  66  */
  67 union tcf_exts_miss_cookie {
  68         struct {
  69                 u32 miss_cookie_base;
  70                 u32 act_index;
  71         };
  72         u64 miss_cookie;
  73 };
  74
  75 #if IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
  76 static int
  77 tcf_exts_miss_cookie_base_alloc(struct tcf_exts *exts, struct tcf_proto *tp,
  78                                 u32 handle)
  79 {
  80         struct tcf_exts_miss_cookie_node *n;
  81         static u32 next;
  82         int err;
  83
  84         if (WARN_ON(!handle || !tp->ops->get_exts))
  85                 return -EINVAL;
  86
  87         n = kzalloc(sizeof(*n), GFP_KERNEL);
  88         if (!n)
  89                 return -ENOMEM;
  90
  91         n->chain_index = tp->chain->index;
  92         n->chain = tp->chain;
  93         n->tp_prio = tp->prio;
  94         n->tp = tp;
  95         n->exts = exts;
  96         n->handle = handle;
  97
  98         err = xa_alloc_cyclic(&tcf_exts_miss_cookies_xa, &n->miss_cookie_base,
  99                               n, xa_limit_32b, &next, GFP_KERNEL);
 100         if (err)
 101                 goto err_xa_alloc;
 102
 103         exts->miss_cookie_node = n;
 104         return 0;
 105
 106 err_xa_alloc:
 107         kfree(n);
 108         return err;
 109 }
 110
 111 static void tcf_exts_miss_cookie_base_destroy(struct tcf_exts *exts)
 112 {
 113         struct tcf_exts_miss_cookie_node *n;
 114
 115         if (!exts->miss_cookie_node)
 116                 return;
 117
 118         n = exts->miss_cookie_node;
 119         xa_erase(&tcf_exts_miss_cookies_xa, n->miss_cookie_base);
 120         kfree_rcu(n, rcu);
 121 }
 122
 123 static struct tcf_exts_miss_cookie_node *
 124 tcf_exts_miss_cookie_lookup(u64 miss_cookie, int *act_index)
 125 {
 126         union tcf_exts_miss_cookie mc = { .miss_cookie = miss_cookie, };
 127
 128         *act_index = mc.act_index;
 129         return xa_load(&tcf_exts_miss_cookies_xa, mc.miss_cookie_base);
 130 }
 131 #else /* IS_ENABLED(CONFIG_NET_TC_SKB_EXT) */
 132 static int
 133 tcf_exts_miss_cookie_base_alloc(struct tcf_exts *exts, struct tcf_proto *tp,
 134                                 u32 handle)
 135 {
 136         return 0;
 137 }
 138
 139 static void tcf_exts_miss_cookie_base_destroy(struct tcf_exts *exts)
 140 {
 141 }
 142 #endif /* IS_ENABLED(CONFIG_NET_TC_SKB_EXT) */
 143
 144 static u64 tcf_exts_miss_cookie_get(u32 miss_cookie_base, int act_index)
 145 {
 146         union tcf_exts_miss_cookie mc = { .act_index = act_index, };
 147
 148         if (!miss_cookie_base)
 149                 return 0;
 150
 151         mc.miss_cookie_base = miss_cookie_base;
 152         return mc.miss_cookie;
 153 }
 154
 155 #ifdef CONFIG_NET_CLS_ACT
 156 DEFINE_STATIC_KEY_FALSE(tc_skb_ext_tc);
 157 EXPORT_SYMBOL(tc_skb_ext_tc);
 158
 159 void tc_skb_ext_tc_enable(void)
 160 {
 161         static_branch_inc(&tc_skb_ext_tc);
 162 }
 163 EXPORT_SYMBOL(tc_skb_ext_tc_enable);
 164
 165 void tc_skb_ext_tc_disable(void)
 166 {
 167         static_branch_dec(&tc_skb_ext_tc);
 168 }
 169 EXPORT_SYMBOL(tc_skb_ext_tc_disable);
 170 #endif
 171
 172 static u32 destroy_obj_hashfn(const struct tcf_proto *tp)
 173 {
 174         return jhash_3words(tp->chain->index, tp->prio,
 175                             (__force __u32)tp->protocol, 0);
 176 }
 177
 178 static void tcf_proto_signal_destroying(struct tcf_chain *chain,
 179                                         struct tcf_proto *tp)
 180 {
 181         struct tcf_block *block = chain->block;
 182
 183         mutex_lock(&block->proto_destroy_lock);
 184         hash_add_rcu(block->proto_destroy_ht, &tp->destroy_ht_node,
 185                      destroy_obj_hashfn(tp));
 186         mutex_unlock(&block->proto_destroy_lock);
 187 }
 188
 189 static bool tcf_proto_cmp(const struct tcf_proto *tp1,
 190                           const struct tcf_proto *tp2)
 191 {
 192         return tp1->chain->index == tp2->chain->index &&
 193                tp1->prio == tp2->prio &&
 194                tp1->protocol == tp2->protocol;
 195 }
 196
 197 static bool tcf_proto_exists_destroying(struct tcf_chain *chain,
 198                                         struct tcf_proto *tp)
 199 {
 200         u32 hash = destroy_obj_hashfn(tp);
 201         struct tcf_proto *iter;
 202         bool found = false;
 203
 204         rcu_read_lock();
 205         hash_for_each_possible_rcu(chain->block->proto_destroy_ht, iter,
 206                                    destroy_ht_node, hash) {
 207                 if (tcf_proto_cmp(tp, iter)) {
 208                         found = true;
 209                         break;
 210                 }
 211         }
 212         rcu_read_unlock();
 213
 214         return found;
 215 }
 216
 217 static void
 218 tcf_proto_signal_destroyed(struct tcf_chain *chain, struct tcf_proto *tp)
 219 {
 220         struct tcf_block *block = chain->block;
 221
 222         mutex_lock(&block->proto_destroy_lock);
 223         if (hash_hashed(&tp->destroy_ht_node))
 224                 hash_del_rcu(&tp->destroy_ht_node);
 225         mutex_unlock(&block->proto_destroy_lock);
 226 }
 227
 228 /* Find classifier type by string name */
 229
 230 static const struct tcf_proto_ops *__tcf_proto_lookup_ops(const char *kind)
 231 {
 232         const struct tcf_proto_ops *t, *res = NULL;
 233
 234         if (kind) {
 235                 read_lock(&cls_mod_lock);
 236                 list_for_each_entry(t, &tcf_proto_base, head) {
 237                         if (strcmp(kind, t->kind) == 0) {
 238                                 if (try_module_get(t->owner))
 239                                         res = t;
 240                                 break;
 241                         }
 242                 }
 243                 read_unlock(&cls_mod_lock);
 244         }
 245         return res;
 246 }
 247
 248 static const struct tcf_proto_ops *
 249 tcf_proto_lookup_ops(const char *kind, bool rtnl_held,
 250                      struct netlink_ext_ack *extack)
 251 {
 252         const struct tcf_proto_ops *ops;
 253
 254         ops = __tcf_proto_lookup_ops(kind);
 255         if (ops)
 256                 return ops;
 257 #ifdef CONFIG_MODULES
 258         if (rtnl_held)
 259                 rtnl_unlock();
 260         request_module("cls_%s", kind);
 261         if (rtnl_held)
 262                 rtnl_lock();
 263         ops = __tcf_proto_lookup_ops(kind);
 264         /* We dropped the RTNL semaphore in order to perform
 265          * the module load. So, even if we succeeded in loading
 266          * the module we have to replay the request. We indicate
 267          * this using -EAGAIN.
 268          */
 269         if (ops) {
 270                 module_put(ops->owner);
 271                 return ERR_PTR(-EAGAIN);
 272         }
 273 #endif
 274         NL_SET_ERR_MSG(extack, "TC classifier not found");
 275         return ERR_PTR(-ENOENT);
 276 }
 277
 278 /* Register(unregister) new classifier type */
 279
 280 int register_tcf_proto_ops(struct tcf_proto_ops *ops)
 281 {
 282         struct tcf_proto_ops *t;
 283         int rc = -EEXIST;
 284
 285         write_lock(&cls_mod_lock);
 286         list_for_each_entry(t, &tcf_proto_base, head)
 287                 if (!strcmp(ops->kind, t->kind))
 288                         goto out;
 289
 290         list_add_tail(&ops->head, &tcf_proto_base);
 291         rc = 0;
 292 out:
 293         write_unlock(&cls_mod_lock);
 294         return rc;
 295 }
 296 EXPORT_SYMBOL(register_tcf_proto_ops);
 297
 298 static struct workqueue_struct *tc_filter_wq;
 299
 300 void unregister_tcf_proto_ops(struct tcf_proto_ops *ops)
 301 {
 302         struct tcf_proto_ops *t;
 303         int rc = -ENOENT;
 304
 305         /* Wait for outstanding call_rcu()s, if any, from a
 306          * tcf_proto_ops's destroy() handler.
 307          */
 308         rcu_barrier();
 309         flush_workqueue(tc_filter_wq);
 310
 311         write_lock(&cls_mod_lock);
 312         list_for_each_entry(t, &tcf_proto_base, head) {
 313                 if (t == ops) {
 314                         list_del(&t->head);
 315                         rc = 0;
 316                         break;
 317                 }
 318         }
 319         write_unlock(&cls_mod_lock);
 320
 321         WARN(rc, "unregister tc filter kind(%s) failed %d\n", ops->kind, rc);
 322 }
 323 EXPORT_SYMBOL(unregister_tcf_proto_ops);
 324
 325 bool tcf_queue_work(struct rcu_work *rwork, work_func_t func)
 326 {
 327         INIT_RCU_WORK(rwork, func);
 328         return queue_rcu_work(tc_filter_wq, rwork);
 329 }
 330 EXPORT_SYMBOL(tcf_queue_work);
 331
 332 /* Select new prio value from the range, managed by kernel. */
 333
 334 static inline u32 tcf_auto_prio(struct tcf_proto *tp)
 335 {
 336         u32 first = TC_H_MAKE(0xC0000000U, 0U);
 337
 338         if (tp)
 339                 first = tp->prio - 1;
 340
 341         return TC_H_MAJ(first);
 342 }
 343
 344 static bool tcf_proto_check_kind(struct nlattr *kind, char *name)
 345 {
 346         if (kind)
 347                 return nla_strscpy(name, kind, IFNAMSIZ) < 0;
 348         memset(name, 0, IFNAMSIZ);
 349         return false;
 350 }
 351
 352 static bool tcf_proto_is_unlocked(const char *kind)
 353 {
 354         const struct tcf_proto_ops *ops;
 355         bool ret;
 356
 357         if (strlen(kind) == 0)
 358                 return false;
 359
 360         ops = tcf_proto_lookup_ops(kind, false, NULL);
 361         /* On error return false to take rtnl lock. Proto lookup/create
 362          * functions will perform lookup again and properly handle errors.
 363          */
 364         if (IS_ERR(ops))
 365                 return false;
 366
 367         ret = !!(ops->flags & TCF_PROTO_OPS_DOIT_UNLOCKED);
 368         module_put(ops->owner);
 369         return ret;
 370 }
 371
 372 static struct tcf_proto *tcf_proto_create(const char *kind, u32 protocol,
 373                                           u32 prio, struct tcf_chain *chain,
 374                                           bool rtnl_held,
 375                                           struct netlink_ext_ack *extack)
 376 {
 377         struct tcf_proto *tp;
 378         int err;
 379
 380         tp = kzalloc(sizeof(*tp), GFP_KERNEL);
 381         if (!tp)
 382                 return ERR_PTR(-ENOBUFS);
 383
 384         tp->ops = tcf_proto_lookup_ops(kind, rtnl_held, extack);
 385         if (IS_ERR(tp->ops)) {
 386                 err = PTR_ERR(tp->ops);
 387                 goto errout;
 388         }
 389         tp->classify = tp->ops->classify;
 390         tp->protocol = protocol;
 391         tp->prio = prio;
 392         tp->chain = chain;
 393         spin_lock_init(&tp->lock);
 394         refcount_set(&tp->refcnt, 1);
 395
 396         err = tp->ops->init(tp);
 397         if (err) {
 398                 module_put(tp->ops->owner);
 399                 goto errout;
 400         }
 401         return tp;
 402
 403 errout:
 404         kfree(tp);
 405         return ERR_PTR(err);
 406 }
 407
 408 static void tcf_proto_get(struct tcf_proto *tp)
 409 {
 410         refcount_inc(&tp->refcnt);
 411 }
 412
 413 static void tcf_chain_put(struct tcf_chain *chain);
 414
 415 static void tcf_proto_destroy(struct tcf_proto *tp, bool rtnl_held,
 416                               bool sig_destroy, struct netlink_ext_ack *extack)
 417 {
 418         tp->ops->destroy(tp, rtnl_held, extack);
 419         if (sig_destroy)
 420                 tcf_proto_signal_destroyed(tp->chain, tp);
 421         tcf_chain_put(tp->chain);
 422         module_put(tp->ops->owner);
 423         kfree_rcu(tp, rcu);
 424 }
 425
 426 static void tcf_proto_put(struct tcf_proto *tp, bool rtnl_held,
 427                           struct netlink_ext_ack *extack)
 428 {
 429         if (refcount_dec_and_test(&tp->refcnt))
 430                 tcf_proto_destroy(tp, rtnl_held, true, extack);
 431 }
 432
 433 static bool tcf_proto_check_delete(struct tcf_proto *tp)
 434 {
 435         if (tp->ops->delete_empty)
 436                 return tp->ops->delete_empty(tp);
 437
 438         tp->deleting = true;
 439         return tp->deleting;
 440 }
 441
 442 static void tcf_proto_mark_delete(struct tcf_proto *tp)
 443 {
 444         spin_lock(&tp->lock);
 445         tp->deleting = true;
 446         spin_unlock(&tp->lock);
 447 }
 448
 449 static bool tcf_proto_is_deleting(struct tcf_proto *tp)
 450 {
 451         bool deleting;
 452
 453         spin_lock(&tp->lock);
 454         deleting = tp->deleting;
 455         spin_unlock(&tp->lock);
 456
 457         return deleting;
 458 }
 459
 460 #define ASSERT_BLOCK_LOCKED(block)                                      \
 461         lockdep_assert_held(&(block)->lock)
 462
 463 struct tcf_filter_chain_list_item {
 464         struct list_head list;
 465         tcf_chain_head_change_t *chain_head_change;
 466         void *chain_head_change_priv;
 467 };
 468
 469 static struct tcf_chain *tcf_chain_create(struct tcf_block *block,
 470                                           u32 chain_index)
 471 {
 472         struct tcf_chain *chain;
 473
 474         ASSERT_BLOCK_LOCKED(block);
 475
 476         chain = kzalloc(sizeof(*chain), GFP_KERNEL);
 477         if (!chain)
 478                 return NULL;
 479         list_add_tail_rcu(&chain->list, &block->chain_list);
 480         mutex_init(&chain->filter_chain_lock);
 481         chain->block = block;
 482         chain->index = chain_index;
 483         chain->refcnt = 1;
 484         if (!chain->index)
 485                 block->chain0.chain = chain;
 486         return chain;
 487 }
 488
 489 static void tcf_chain_head_change_item(struct tcf_filter_chain_list_item *item,
 490                                        struct tcf_proto *tp_head)
 491 {
 492         if (item->chain_head_change)
 493                 item->chain_head_change(tp_head, item->chain_head_change_priv);
 494 }
 495
 496 static void tcf_chain0_head_change(struct tcf_chain *chain,
 497                                    struct tcf_proto *tp_head)
 498 {
 499         struct tcf_filter_chain_list_item *item;
 500         struct tcf_block *block = chain->block;
 501
 502         if (chain->index)
 503                 return;
 504
 505         mutex_lock(&block->lock);
 506         list_for_each_entry(item, &block->chain0.filter_chain_list, list)
 507                 tcf_chain_head_change_item(item, tp_head);
 508         mutex_unlock(&block->lock);
 509 }
 510
 511 /* Returns true if block can be safely freed. */
 512
 513 static bool tcf_chain_detach(struct tcf_chain *chain)
 514 {
 515         struct tcf_block *block = chain->block;
 516
 517         ASSERT_BLOCK_LOCKED(block);
 518
 519         list_del_rcu(&chain->list);
 520         if (!chain->index)
 521                 block->chain0.chain = NULL;
 522
 523         if (list_empty(&block->chain_list) &&
 524             refcount_read(&block->refcnt) == 0)
 525                 return true;
 526
 527         return false;
 528 }
 529
 530 static void tcf_block_destroy(struct tcf_block *block)
 531 {
 532         mutex_destroy(&block->lock);
 533         mutex_destroy(&block->proto_destroy_lock);
 534         xa_destroy(&block->ports);
 535         kfree_rcu(block, rcu);
 536 }
 537
 538 static void tcf_chain_destroy(struct tcf_chain *chain, bool free_block)
 539 {
 540         struct tcf_block *block = chain->block;
 541
 542         mutex_destroy(&chain->filter_chain_lock);
 543         kfree_rcu(chain, rcu);
 544         if (free_block)
 545                 tcf_block_destroy(block);
 546 }
 547
 548 static void tcf_chain_hold(struct tcf_chain *chain)
 549 {
 550         ASSERT_BLOCK_LOCKED(chain->block);
 551
 552         ++chain->refcnt;
 553 }
 554
 555 static bool tcf_chain_held_by_acts_only(struct tcf_chain *chain)
 556 {
 557         ASSERT_BLOCK_LOCKED(chain->block);
 558
 559         /* In case all the references are action references, this
 560          * chain should not be shown to the user.
 561          */
 562         return chain->refcnt == chain->action_refcnt;
 563 }
 564
 565 static struct tcf_chain *tcf_chain_lookup(struct tcf_block *block,
 566                                           u32 chain_index)
 567 {
 568         struct tcf_chain *chain;
 569
 570         ASSERT_BLOCK_LOCKED(block);
 571
 572         list_for_each_entry(chain, &block->chain_list, list) {
 573                 if (chain->index == chain_index)
 574                         return chain;
 575         }
 576         return NULL;
 577 }
 578
 579 #if IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
 580 static struct tcf_chain *tcf_chain_lookup_rcu(const struct tcf_block *block,
 581                                               u32 chain_index)
 582 {
 583         struct tcf_chain *chain;
 584
 585         list_for_each_entry_rcu(chain, &block->chain_list, list) {
 586                 if (chain->index == chain_index)
 587                         return chain;
 588         }
 589         return NULL;
 590 }
 591 #endif
 592
 593 static int tc_chain_notify(struct tcf_chain *chain, struct sk_buff *oskb,
 594                            u32 seq, u16 flags, int event, bool unicast,
 595                            struct netlink_ext_ack *extack);
 596
 597 static struct tcf_chain *__tcf_chain_get(struct tcf_block *block,
 598                                          u32 chain_index, bool create,
 599                                          bool by_act)
 600 {
 601         struct tcf_chain *chain = NULL;
 602         bool is_first_reference;
 603
 604         mutex_lock(&block->lock);
 605         chain = tcf_chain_lookup(block, chain_index);
 606         if (chain) {
 607                 tcf_chain_hold(chain);
 608         } else {
 609                 if (!create)
 610                         goto errout;
 611                 chain = tcf_chain_create(block, chain_index);
 612                 if (!chain)
 613                         goto errout;
 614         }
 615
 616         if (by_act)
 617                 ++chain->action_refcnt;
 618         is_first_reference = chain->refcnt - chain->action_refcnt == 1;
 619         mutex_unlock(&block->lock);
 620
 621         /* Send notification only in case we got the first
 622          * non-action reference. Until then, the chain acts only as
 623          * a placeholder for actions pointing to it and user ought
 624          * not know about them.
 625          */
 626         if (is_first_reference && !by_act)
 627                 tc_chain_notify(chain, NULL, 0, NLM_F_CREATE | NLM_F_EXCL,
 628                                 RTM_NEWCHAIN, false, NULL);
 629
 630         return chain;
 631
 632 errout:
 633         mutex_unlock(&block->lock);
 634         return chain;
 635 }
 636
 637 static struct tcf_chain *tcf_chain_get(struct tcf_block *block, u32 chain_index,
 638                                        bool create)
 639 {
 640         return __tcf_chain_get(block, chain_index, create, false);
 641 }
 642
 643 struct tcf_chain *tcf_chain_get_by_act(struct tcf_block *block, u32 chain_index)
 644 {
 645         return __tcf_chain_get(block, chain_index, true, true);
 646 }
 647 EXPORT_SYMBOL(tcf_chain_get_by_act);
 648
 649 static void tc_chain_tmplt_del(const struct tcf_proto_ops *tmplt_ops,
 650                                void *tmplt_priv);
 651 static int tc_chain_notify_delete(const struct tcf_proto_ops *tmplt_ops,
 652                                   void *tmplt_priv, u32 chain_index,
 653                                   struct tcf_block *block, struct sk_buff *oskb,
 654                                   u32 seq, u16 flags);
 655
 656 static void __tcf_chain_put(struct tcf_chain *chain, bool by_act,
 657                             bool explicitly_created)
 658 {
 659         struct tcf_block *block = chain->block;
 660         const struct tcf_proto_ops *tmplt_ops;
 661         unsigned int refcnt, non_act_refcnt;
 662         bool free_block = false;
 663         void *tmplt_priv;
 664
 665         mutex_lock(&block->lock);
 666         if (explicitly_created) {
 667                 if (!chain->explicitly_created) {
 668                         mutex_unlock(&block->lock);
 669                         return;
 670                 }
 671                 chain->explicitly_created = false;
 672         }
 673
 674         if (by_act)
 675                 chain->action_refcnt--;
 676
 677         /* tc_chain_notify_delete can't be called while holding block lock.
 678          * However, when block is unlocked chain can be changed concurrently, so
 679          * save these to temporary variables.
 680          */
 681         refcnt = --chain->refcnt;
 682         non_act_refcnt = refcnt - chain->action_refcnt;
 683         tmplt_ops = chain->tmplt_ops;
 684         tmplt_priv = chain->tmplt_priv;
 685
 686         if (non_act_refcnt == chain->explicitly_created && !by_act) {
 687                 if (non_act_refcnt == 0)
 688                         tc_chain_notify_delete(tmplt_ops, tmplt_priv,
 689                                                chain->index, block, NULL, 0, 0);
 690                 /* Last reference to chain, no need to lock. */
 691                 chain->flushing = false;
 692         }
 693
 694         if (refcnt == 0)
 695                 free_block = tcf_chain_detach(chain);
 696         mutex_unlock(&block->lock);
 697
 698         if (refcnt == 0) {
 699                 tc_chain_tmplt_del(tmplt_ops, tmplt_priv);
 700                 tcf_chain_destroy(chain, free_block);
 701         }
 702 }
 703
 704 static void tcf_chain_put(struct tcf_chain *chain)
 705 {
 706         __tcf_chain_put(chain, false, false);
 707 }
 708
 709 void tcf_chain_put_by_act(struct tcf_chain *chain)
 710 {
 711         __tcf_chain_put(chain, true, false);
 712 }
 713 EXPORT_SYMBOL(tcf_chain_put_by_act);
 714
 715 static void tcf_chain_put_explicitly_created(struct tcf_chain *chain)
 716 {
 717         __tcf_chain_put(chain, false, true);
 718 }
 719
 720 static void tcf_chain_flush(struct tcf_chain *chain, bool rtnl_held)
 721 {
 722         struct tcf_proto *tp, *tp_next;
 723
 724         mutex_lock(&chain->filter_chain_lock);
 725         tp = tcf_chain_dereference(chain->filter_chain, chain);
 726         while (tp) {
 727                 tp_next = rcu_dereference_protected(tp->next, 1);
 728                 tcf_proto_signal_destroying(chain, tp);
 729                 tp = tp_next;
 730         }
 731         tp = tcf_chain_dereference(chain->filter_chain, chain);
 732         RCU_INIT_POINTER(chain->filter_chain, NULL);
 733         tcf_chain0_head_change(chain, NULL);
 734         chain->flushing = true;
 735         mutex_unlock(&chain->filter_chain_lock);
 736
 737         while (tp) {
 738                 tp_next = rcu_dereference_protected(tp->next, 1);
 739                 tcf_proto_put(tp, rtnl_held, NULL);
 740                 tp = tp_next;
 741         }
 742 }
 743
 744 static int tcf_block_setup(struct tcf_block *block,
 745                            struct flow_block_offload *bo);
 746
 747 static void tcf_block_offload_init(struct flow_block_offload *bo,
 748                                    struct net_device *dev, struct Qdisc *sch,
 749                                    enum flow_block_command command,
 750                                    enum flow_block_binder_type binder_type,
 751                                    struct flow_block *flow_block,
 752                                    bool shared, struct netlink_ext_ack *extack)
 753 {
 754         bo->net = dev_net(dev);
 755         bo->command = command;
 756         bo->binder_type = binder_type;
 757         bo->block = flow_block;
 758         bo->block_shared = shared;
 759         bo->extack = extack;
 760         bo->sch = sch;
 761         bo->cb_list_head = &flow_block->cb_list;
 762         INIT_LIST_HEAD(&bo->cb_list);
 763 }
 764
 765 static void tcf_block_unbind(struct tcf_block *block,
 766                              struct flow_block_offload *bo);
 767
 768 static void tc_block_indr_cleanup(struct flow_block_cb *block_cb)
 769 {
 770         struct tcf_block *block = block_cb->indr.data;
 771         struct net_device *dev = block_cb->indr.dev;
 772         struct Qdisc *sch = block_cb->indr.sch;
 773         struct netlink_ext_ack extack = {};
 774         struct flow_block_offload bo = {};
 775
 776         tcf_block_offload_init(&bo, dev, sch, FLOW_BLOCK_UNBIND,
 777                                block_cb->indr.binder_type,
 778                                &block->flow_block, tcf_block_shared(block),
 779                                &extack);
 780         rtnl_lock();
 781         down_write(&block->cb_lock);
 782         list_del(&block_cb->driver_list);
 783         list_move(&block_cb->list, &bo.cb_list);
 784         tcf_block_unbind(block, &bo);
 785         up_write(&block->cb_lock);
 786         rtnl_unlock();
 787 }
 788
 789 static bool tcf_block_offload_in_use(struct tcf_block *block)
 790 {
 791         return atomic_read(&block->offloadcnt);
 792 }
 793
 794 static int tcf_block_offload_cmd(struct tcf_block *block,
 795                                  struct net_device *dev, struct Qdisc *sch,
 796                                  struct tcf_block_ext_info *ei,
 797                                  enum flow_block_command command,
 798                                  struct netlink_ext_ack *extack)
 799 {
 800         struct flow_block_offload bo = {};
 801
 802         tcf_block_offload_init(&bo, dev, sch, command, ei->binder_type,
 803                                &block->flow_block, tcf_block_shared(block),
 804                                extack);
 805
 806         if (dev->netdev_ops->ndo_setup_tc) {
 807                 int err;
 808
 809                 err = dev->netdev_ops->ndo_setup_tc(dev, TC_SETUP_BLOCK, &bo);
 810                 if (err < 0) {
 811                         if (err != -EOPNOTSUPP)
 812                                 NL_SET_ERR_MSG(extack, "Driver ndo_setup_tc failed");
 813                         return err;
 814                 }
 815
 816                 return tcf_block_setup(block, &bo);
 817         }
 818
 819         flow_indr_dev_setup_offload(dev, sch, TC_SETUP_BLOCK, block, &bo,
 820                                     tc_block_indr_cleanup);
 821         tcf_block_setup(block, &bo);
 822
 823         return -EOPNOTSUPP;
 824 }
 825
 826 static int tcf_block_offload_bind(struct tcf_block *block, struct Qdisc *q,
 827                                   struct tcf_block_ext_info *ei,
 828                                   struct netlink_ext_ack *extack)
 829 {
 830         struct net_device *dev = q->dev_queue->dev;
 831         int err;
 832
 833         down_write(&block->cb_lock);
 834
 835         /* If tc offload feature is disabled and the block we try to bind
 836          * to already has some offloaded filters, forbid to bind.
 837          */
 838         if (dev->netdev_ops->ndo_setup_tc &&
 839             !tc_can_offload(dev) &&
 840             tcf_block_offload_in_use(block)) {
 841                 NL_SET_ERR_MSG(extack, "Bind to offloaded block failed as dev has offload disabled");
 842                 err = -EOPNOTSUPP;
 843                 goto err_unlock;
 844         }
 845
 846         err = tcf_block_offload_cmd(block, dev, q, ei, FLOW_BLOCK_BIND, extack);
 847         if (err == -EOPNOTSUPP)
 848                 goto no_offload_dev_inc;
 849         if (err)
 850                 goto err_unlock;
 851
 852         up_write(&block->cb_lock);
 853         return 0;
 854
 855 no_offload_dev_inc:
 856         if (tcf_block_offload_in_use(block))
 857                 goto err_unlock;
 858
 859         err = 0;
 860         block->nooffloaddevcnt++;
 861 err_unlock:
 862         up_write(&block->cb_lock);
 863         return err;
 864 }
 865
 866 static void tcf_block_offload_unbind(struct tcf_block *block, struct Qdisc *q,
 867                                      struct tcf_block_ext_info *ei)
 868 {
 869         struct net_device *dev = q->dev_queue->dev;
 870         int err;
 871
 872         down_write(&block->cb_lock);
 873         err = tcf_block_offload_cmd(block, dev, q, ei, FLOW_BLOCK_UNBIND, NULL);
 874         if (err == -EOPNOTSUPP)
 875                 goto no_offload_dev_dec;
 876         up_write(&block->cb_lock);
 877         return;
 878
 879 no_offload_dev_dec:
 880         WARN_ON(block->nooffloaddevcnt-- == 0);
 881         up_write(&block->cb_lock);
 882 }
 883
 884 static int
 885 tcf_chain0_head_change_cb_add(struct tcf_block *block,
 886                               struct tcf_block_ext_info *ei,
 887                               struct netlink_ext_ack *extack)
 888 {
 889         struct tcf_filter_chain_list_item *item;
 890         struct tcf_chain *chain0;
 891
 892         item = kmalloc(sizeof(*item), GFP_KERNEL);
 893         if (!item) {
 894                 NL_SET_ERR_MSG(extack, "Memory allocation for head change callback item failed");
 895                 return -ENOMEM;
 896         }
 897         item->chain_head_change = ei->chain_head_change;
 898         item->chain_head_change_priv = ei->chain_head_change_priv;
 899
 900         mutex_lock(&block->lock);
 901         chain0 = block->chain0.chain;
 902         if (chain0)
 903                 tcf_chain_hold(chain0);
 904         else
 905                 list_add(&item->list, &block->chain0.filter_chain_list);
 906         mutex_unlock(&block->lock);
 907
 908         if (chain0) {
 909                 struct tcf_proto *tp_head;
 910
 911                 mutex_lock(&chain0->filter_chain_lock);
 912
 913                 tp_head = tcf_chain_dereference(chain0->filter_chain, chain0);
 914                 if (tp_head)
 915                         tcf_chain_head_change_item(item, tp_head);
 916
 917                 mutex_lock(&block->lock);
 918                 list_add(&item->list, &block->chain0.filter_chain_list);
 919                 mutex_unlock(&block->lock);
 920
 921                 mutex_unlock(&chain0->filter_chain_lock);
 922                 tcf_chain_put(chain0);
 923         }
 924
 925         return 0;
 926 }
 927
 928 static void
 929 tcf_chain0_head_change_cb_del(struct tcf_block *block,
 930                               struct tcf_block_ext_info *ei)
 931 {
 932         struct tcf_filter_chain_list_item *item;
 933
 934         mutex_lock(&block->lock);
 935         list_for_each_entry(item, &block->chain0.filter_chain_list, list) {
 936                 if ((!ei->chain_head_change && !ei->chain_head_change_priv) ||
 937                     (item->chain_head_change == ei->chain_head_change &&
 938                      item->chain_head_change_priv == ei->chain_head_change_priv)) {
 939                         if (block->chain0.chain)
 940                                 tcf_chain_head_change_item(item, NULL);
 941                         list_del(&item->list);
 942                         mutex_unlock(&block->lock);
 943
 944                         kfree(item);
 945                         return;
 946                 }
 947         }
 948         mutex_unlock(&block->lock);
 949         WARN_ON(1);
 950 }
 951
 952 struct tcf_net {
 953         spinlock_t idr_lock; /* Protects idr */
 954         struct idr idr;
 955 };
 956
 957 static unsigned int tcf_net_id;
 958
 959 static int tcf_block_insert(struct tcf_block *block, struct net *net,
 960                             struct netlink_ext_ack *extack)
 961 {
 962         struct tcf_net *tn = net_generic(net, tcf_net_id);
 963         int err;
 964
 965         idr_preload(GFP_KERNEL);
 966         spin_lock(&tn->idr_lock);
 967         err = idr_alloc_u32(&tn->idr, block, &block->index, block->index,
 968                             GFP_NOWAIT);
 969         spin_unlock(&tn->idr_lock);
 970         idr_preload_end();
 971
 972         return err;
 973 }
 974
 975 static void tcf_block_remove(struct tcf_block *block, struct net *net)
 976 {
 977         struct tcf_net *tn = net_generic(net, tcf_net_id);
 978
 979         spin_lock(&tn->idr_lock);
 980         idr_remove(&tn->idr, block->index);
 981         spin_unlock(&tn->idr_lock);
 982 }
 983
 984 static struct tcf_block *tcf_block_create(struct net *net, struct Qdisc *q,
 985                                           u32 block_index,
 986                                           struct netlink_ext_ack *extack)
 987 {
 988         struct tcf_block *block;
 989
 990         block = kzalloc(sizeof(*block), GFP_KERNEL);
 991         if (!block) {
 992                 NL_SET_ERR_MSG(extack, "Memory allocation for block failed");
 993                 return ERR_PTR(-ENOMEM);
 994         }
 995         mutex_init(&block->lock);
 996         mutex_init(&block->proto_destroy_lock);
 997         init_rwsem(&block->cb_lock);
 998         flow_block_init(&block->flow_block);
 999         INIT_LIST_HEAD(&block->chain_list);
1000         INIT_LIST_HEAD(&block->owner_list);
1001         INIT_LIST_HEAD(&block->chain0.filter_chain_list);
1002
1003         refcount_set(&block->refcnt, 1);
1004         block->net = net;
1005         block->index = block_index;
1006         xa_init(&block->ports);
1007
1008         /* Don't store q pointer for blocks which are shared */
1009         if (!tcf_block_shared(block))
1010                 block->q = q;
1011         return block;
1012 }
1013
1014 struct tcf_block *tcf_block_lookup(struct net *net, u32 block_index)
1015 {
1016         struct tcf_net *tn = net_generic(net, tcf_net_id);
1017
1018         return idr_find(&tn->idr, block_index);
1019 }
1020 EXPORT_SYMBOL(tcf_block_lookup);
1021
1022 static struct tcf_block *tcf_block_refcnt_get(struct net *net, u32 block_index)
1023 {
1024         struct tcf_block *block;
1025
1026         rcu_read_lock();
1027         block = tcf_block_lookup(net, block_index);
1028         if (block && !refcount_inc_not_zero(&block->refcnt))
1029                 block = NULL;
1030         rcu_read_unlock();
1031
1032         return block;
1033 }
1034
1035 static struct tcf_chain *
1036 __tcf_get_next_chain(struct tcf_block *block, struct tcf_chain *chain)
1037 {
1038         mutex_lock(&block->lock);
1039         if (chain)
1040                 chain = list_is_last(&chain->list, &block->chain_list) ?
1041                         NULL : list_next_entry(chain, list);
1042         else
1043                 chain = list_first_entry_or_null(&block->chain_list,
1044                                                  struct tcf_chain, list);
1045
1046         /* skip all action-only chains */
1047         while (chain && tcf_chain_held_by_acts_only(chain))
1048                 chain = list_is_last(&chain->list, &block->chain_list) ?
1049                         NULL : list_next_entry(chain, list);
1050
1051         if (chain)
1052                 tcf_chain_hold(chain);
1053         mutex_unlock(&block->lock);
1054
1055         return chain;
1056 }
1057
1058 /* Function to be used by all clients that want to iterate over all chains on
1059  * block. It properly obtains block->lock and takes reference to chain before
1060  * returning it. Users of this function must be tolerant to concurrent chain
1061  * insertion/deletion or ensure that no concurrent chain modification is
1062  * possible. Note that all netlink dump callbacks cannot guarantee to provide
1063  * consistent dump because rtnl lock is released each time skb is filled with
1064  * data and sent to user-space.
1065  */
1066
1067 struct tcf_chain *
1068 tcf_get_next_chain(struct tcf_block *block, struct tcf_chain *chain)
1069 {
1070         struct tcf_chain *chain_next = __tcf_get_next_chain(block, chain);
1071
1072         if (chain)
1073                 tcf_chain_put(chain);
1074
1075         return chain_next;
1076 }
1077 EXPORT_SYMBOL(tcf_get_next_chain);
1078
1079 static struct tcf_proto *
1080 __tcf_get_next_proto(struct tcf_chain *chain, struct tcf_proto *tp)
1081 {
1082         u32 prio = 0;
1083
1084         ASSERT_RTNL();
1085         mutex_lock(&chain->filter_chain_lock);
1086
1087         if (!tp) {
1088                 tp = tcf_chain_dereference(chain->filter_chain, chain);
1089         } else if (tcf_proto_is_deleting(tp)) {
1090                 /* 'deleting' flag is set and chain->filter_chain_lock was
1091                  * unlocked, which means next pointer could be invalid. Restart
1092                  * search.
1093                  */
1094                 prio = tp->prio + 1;
1095                 tp = tcf_chain_dereference(chain->filter_chain, chain);
1096
1097                 for (; tp; tp = tcf_chain_dereference(tp->next, chain))
1098                         if (!tp->deleting && tp->prio >= prio)
1099                                 break;
1100         } else {
1101                 tp = tcf_chain_dereference(tp->next, chain);
1102         }
1103
1104         if (tp)
1105                 tcf_proto_get(tp);
1106
1107         mutex_unlock(&chain->filter_chain_lock);
1108
1109         return tp;
1110 }
1111
1112 /* Function to be used by all clients that want to iterate over all tp's on
1113  * chain. Users of this function must be tolerant to concurrent tp
1114  * insertion/deletion or ensure that no concurrent chain modification is
1115  * possible. Note that all netlink dump callbacks cannot guarantee to provide
1116  * consistent dump because rtnl lock is released each time skb is filled with
1117  * data and sent to user-space.
1118  */
1119
1120 struct tcf_proto *
1121 tcf_get_next_proto(struct tcf_chain *chain, struct tcf_proto *tp)
1122 {
1123         struct tcf_proto *tp_next = __tcf_get_next_proto(chain, tp);
1124
1125         if (tp)
1126                 tcf_proto_put(tp, true, NULL);
1127
1128         return tp_next;
1129 }
1130 EXPORT_SYMBOL(tcf_get_next_proto);
1131
1132 static void tcf_block_flush_all_chains(struct tcf_block *block, bool rtnl_held)
1133 {
1134         struct tcf_chain *chain;
1135
1136         /* Last reference to block. At this point chains cannot be added or
1137          * removed concurrently.
1138          */
1139         for (chain = tcf_get_next_chain(block, NULL);
1140              chain;
1141              chain = tcf_get_next_chain(block, chain)) {
1142                 tcf_chain_put_explicitly_created(chain);
1143                 tcf_chain_flush(chain, rtnl_held);
1144         }
1145 }
1146
1147 /* Lookup Qdisc and increments its reference counter.
1148  * Set parent, if necessary.
1149  */
1150
1151 static int __tcf_qdisc_find(struct net *net, struct Qdisc **q,
1152                             u32 *parent, int ifindex, bool rtnl_held,
1153                             struct netlink_ext_ack *extack)
1154 {
1155         const struct Qdisc_class_ops *cops;
1156         struct net_device *dev;
1157         int err = 0;
1158
1159         if (ifindex == TCM_IFINDEX_MAGIC_BLOCK)
1160                 return 0;
1161
1162         rcu_read_lock();
1163
1164         /* Find link */
1165         dev = dev_get_by_index_rcu(net, ifindex);
1166         if (!dev) {
1167                 rcu_read_unlock();
1168                 return -ENODEV;
1169         }
1170
1171         /* Find qdisc */
1172         if (!*parent) {
1173                 *q = rcu_dereference(dev->qdisc);
1174                 *parent = (*q)->handle;
1175         } else {
1176                 *q = qdisc_lookup_rcu(dev, TC_H_MAJ(*parent));
1177                 if (!*q) {
1178                         NL_SET_ERR_MSG(extack, "Parent Qdisc doesn't exists");
1179                         err = -EINVAL;
1180                         goto errout_rcu;
1181                 }
1182         }
1183
1184         *q = qdisc_refcount_inc_nz(*q);
1185         if (!*q) {
1186                 NL_SET_ERR_MSG(extack, "Parent Qdisc doesn't exists");
1187                 err = -EINVAL;
1188                 goto errout_rcu;
1189         }
1190
1191         /* Is it classful? */
1192         cops = (*q)->ops->cl_ops;
1193         if (!cops) {
1194                 NL_SET_ERR_MSG(extack, "Qdisc not classful");
1195                 err = -EINVAL;
1196                 goto errout_qdisc;
1197         }
1198
1199         if (!cops->tcf_block) {
1200                 NL_SET_ERR_MSG(extack, "Class doesn't support blocks");
1201                 err = -EOPNOTSUPP;
1202                 goto errout_qdisc;
1203         }
1204
1205 errout_rcu:
1206         /* At this point we know that qdisc is not noop_qdisc,
1207          * which means that qdisc holds a reference to net_device
1208          * and we hold a reference to qdisc, so it is safe to release
1209          * rcu read lock.
1210          */
1211         rcu_read_unlock();
1212         return err;
1213
1214 errout_qdisc:
1215         rcu_read_unlock();
1216
1217         if (rtnl_held)
1218                 qdisc_put(*q);
1219         else
1220                 qdisc_put_unlocked(*q);
1221         *q = NULL;
1222
1223         return err;
1224 }
1225
1226 static int __tcf_qdisc_cl_find(struct Qdisc *q, u32 parent, unsigned long *cl,
1227                                int ifindex, struct netlink_ext_ack *extack)
1228 {
1229         if (ifindex == TCM_IFINDEX_MAGIC_BLOCK)
1230                 return 0;
1231
1232         /* Do we search for filter, attached to class? */
1233         if (TC_H_MIN(parent)) {
1234                 const struct Qdisc_class_ops *cops = q->ops->cl_ops;
1235
1236                 *cl = cops->find(q, parent);
1237                 if (*cl == 0) {
1238                         NL_SET_ERR_MSG(extack, "Specified class doesn't exist");
1239                         return -ENOENT;
1240                 }
1241         }
1242
1243         return 0;
1244 }
1245
1246 static struct tcf_block *__tcf_block_find(struct net *net, struct Qdisc *q,
1247                                           unsigned long cl, int ifindex,
1248                                           u32 block_index,
1249                                           struct netlink_ext_ack *extack)
1250 {
1251         struct tcf_block *block;
1252
1253         if (ifindex == TCM_IFINDEX_MAGIC_BLOCK) {
1254                 block = tcf_block_refcnt_get(net, block_index);
1255                 if (!block) {
1256                         NL_SET_ERR_MSG(extack, "Block of given index was not found");
1257                         return ERR_PTR(-EINVAL);
1258                 }
1259         } else {
1260                 const struct Qdisc_class_ops *cops = q->ops->cl_ops;
1261
1262                 block = cops->tcf_block(q, cl, extack);
1263                 if (!block)
1264                         return ERR_PTR(-EINVAL);
1265
1266                 if (tcf_block_shared(block)) {
1267                         NL_SET_ERR_MSG(extack, "This filter block is shared. Please use the block index to manipulate the filters");
1268                         return ERR_PTR(-EOPNOTSUPP);
1269                 }
1270
1271                 /* Always take reference to block in order to support execution
1272                  * of rules update path of cls API without rtnl lock. Caller
1273                  * must release block when it is finished using it. 'if' block
1274                  * of this conditional obtain reference to block by calling
1275                  * tcf_block_refcnt_get().
1276                  */
1277                 refcount_inc(&block->refcnt);
1278         }
1279
1280         return block;
1281 }
1282
1283 static void __tcf_block_put(struct tcf_block *block, struct Qdisc *q,
1284                             struct tcf_block_ext_info *ei, bool rtnl_held)
1285 {
1286         if (refcount_dec_and_mutex_lock(&block->refcnt, &block->lock)) {
1287                 /* Flushing/putting all chains will cause the block to be
1288                  * deallocated when last chain is freed. However, if chain_list
1289                  * is empty, block has to be manually deallocated. After block
1290                  * reference counter reached 0, it is no longer possible to
1291                  * increment it or add new chains to block.
1292                  */
1293                 bool free_block = list_empty(&block->chain_list);
1294
1295                 mutex_unlock(&block->lock);
1296                 if (tcf_block_shared(block))
1297                         tcf_block_remove(block, block->net);
1298
1299                 if (q)
1300                         tcf_block_offload_unbind(block, q, ei);
1301
1302                 if (free_block)
1303                         tcf_block_destroy(block);
1304                 else
1305                         tcf_block_flush_all_chains(block, rtnl_held);
1306         } else if (q) {
1307                 tcf_block_offload_unbind(block, q, ei);
1308         }
1309 }
1310
1311 static void tcf_block_refcnt_put(struct tcf_block *block, bool rtnl_held)
1312 {
1313         __tcf_block_put(block, NULL, NULL, rtnl_held);
1314 }
1315
1316 /* Find tcf block.
1317  * Set q, parent, cl when appropriate.
1318  */
1319
1320 static struct tcf_block *tcf_block_find(struct net *net, struct Qdisc **q,
1321                                         u32 *parent, unsigned long *cl,
1322                                         int ifindex, u32 block_index,
1323                                         struct netlink_ext_ack *extack)
1324 {
1325         struct tcf_block *block;
1326         int err = 0;
1327
1328         ASSERT_RTNL();
1329
1330         err = __tcf_qdisc_find(net, q, parent, ifindex, true, extack);
1331         if (err)
1332                 goto errout;
1333
1334         err = __tcf_qdisc_cl_find(*q, *parent, cl, ifindex, extack);
1335         if (err)
1336                 goto errout_qdisc;
1337
1338         block = __tcf_block_find(net, *q, *cl, ifindex, block_index, extack);
1339         if (IS_ERR(block)) {
1340                 err = PTR_ERR(block);
1341                 goto errout_qdisc;
1342         }
1343
1344         return block;
1345
1346 errout_qdisc:
1347         if (*q)
1348                 qdisc_put(*q);
1349 errout:
1350         *q = NULL;
1351         return ERR_PTR(err);
1352 }
1353
1354 static void tcf_block_release(struct Qdisc *q, struct tcf_block *block,
1355                               bool rtnl_held)
1356 {
1357         if (!IS_ERR_OR_NULL(block))
1358                 tcf_block_refcnt_put(block, rtnl_held);
1359
1360         if (q) {
1361                 if (rtnl_held)
1362                         qdisc_put(q);
1363                 else
1364                         qdisc_put_unlocked(q);
1365         }
1366 }
1367
1368 struct tcf_block_owner_item {
1369         struct list_head list;
1370         struct Qdisc *q;
1371         enum flow_block_binder_type binder_type;
1372 };
1373
1374 static void
1375 tcf_block_owner_netif_keep_dst(struct tcf_block *block,
1376                                struct Qdisc *q,
1377                                enum flow_block_binder_type binder_type)
1378 {
1379         if (block->keep_dst &&
1380             binder_type != FLOW_BLOCK_BINDER_TYPE_CLSACT_INGRESS &&
1381             binder_type != FLOW_BLOCK_BINDER_TYPE_CLSACT_EGRESS)
1382                 netif_keep_dst(qdisc_dev(q));
1383 }
1384
1385 void tcf_block_netif_keep_dst(struct tcf_block *block)
1386 {
1387         struct tcf_block_owner_item *item;
1388
1389         block->keep_dst = true;
1390         list_for_each_entry(item, &block->owner_list, list)
1391                 tcf_block_owner_netif_keep_dst(block, item->q,
1392                                                item->binder_type);
1393 }
1394 EXPORT_SYMBOL(tcf_block_netif_keep_dst);
1395
1396 static int tcf_block_owner_add(struct tcf_block *block,
1397                                struct Qdisc *q,
1398                                enum flow_block_binder_type binder_type)
1399 {
1400         struct tcf_block_owner_item *item;
1401
1402         item = kmalloc(sizeof(*item), GFP_KERNEL);
1403         if (!item)
1404                 return -ENOMEM;
1405         item->q = q;
1406         item->binder_type = binder_type;
1407         list_add(&item->list, &block->owner_list);
1408         return 0;
1409 }
1410
1411 static void tcf_block_owner_del(struct tcf_block *block,
1412                                 struct Qdisc *q,
1413                                 enum flow_block_binder_type binder_type)
1414 {
1415         struct tcf_block_owner_item *item;
1416
1417         list_for_each_entry(item, &block->owner_list, list) {
1418                 if (item->q == q && item->binder_type == binder_type) {
1419                         list_del(&item->list);
1420                         kfree(item);
1421                         return;
1422                 }
1423         }
1424         WARN_ON(1);
1425 }
1426
1427 int tcf_block_get_ext(struct tcf_block **p_block, struct Qdisc *q,
1428                       struct tcf_block_ext_info *ei,
1429                       struct netlink_ext_ack *extack)
1430 {
1431         struct net *net = qdisc_net(q);
1432         struct tcf_block *block = NULL;
1433         int err;
1434
1435         if (ei->block_index)
1436                 /* block_index not 0 means the shared block is requested */
1437                 block = tcf_block_refcnt_get(net, ei->block_index);
1438
1439         if (!block) {
1440                 block = tcf_block_create(net, q, ei->block_index, extack);
1441                 if (IS_ERR(block))
1442                         return PTR_ERR(block);
1443                 if (tcf_block_shared(block)) {
1444                         err = tcf_block_insert(block, net, extack);
1445                         if (err)
1446                                 goto err_block_insert;
1447                 }
1448         }
1449
1450         err = tcf_block_owner_add(block, q, ei->binder_type);
1451         if (err)
1452                 goto err_block_owner_add;
1453
1454         tcf_block_owner_netif_keep_dst(block, q, ei->binder_type);
1455
1456         err = tcf_chain0_head_change_cb_add(block, ei, extack);
1457         if (err)
1458                 goto err_chain0_head_change_cb_add;
1459
1460         err = tcf_block_offload_bind(block, q, ei, extack);
1461         if (err)
1462                 goto err_block_offload_bind;
1463
1464         *p_block = block;
1465         return 0;
1466
1467 err_block_offload_bind:
1468         tcf_chain0_head_change_cb_del(block, ei);
1469 err_chain0_head_change_cb_add:
1470         tcf_block_owner_del(block, q, ei->binder_type);
1471 err_block_owner_add:
1472 err_block_insert:
1473         tcf_block_refcnt_put(block, true);
1474         return err;
1475 }
1476 EXPORT_SYMBOL(tcf_block_get_ext);
1477
1478 static void tcf_chain_head_change_dflt(struct tcf_proto *tp_head, void *priv)
1479 {
1480         struct tcf_proto __rcu **p_filter_chain = priv;
1481
1482         rcu_assign_pointer(*p_filter_chain, tp_head);
1483 }
1484
1485 int tcf_block_get(struct tcf_block **p_block,
1486                   struct tcf_proto __rcu **p_filter_chain, struct Qdisc *q,
1487                   struct netlink_ext_ack *extack)
1488 {
1489         struct tcf_block_ext_info ei = {
1490                 .chain_head_change = tcf_chain_head_change_dflt,
1491                 .chain_head_change_priv = p_filter_chain,
1492         };
1493
1494         WARN_ON(!p_filter_chain);
1495         return tcf_block_get_ext(p_block, q, &ei, extack);
1496 }
1497 EXPORT_SYMBOL(tcf_block_get);
1498
1499 /* XXX: Standalone actions are not allowed to jump to any chain, and bound
1500  * actions should be all removed after flushing.
1501  */
1502 void tcf_block_put_ext(struct tcf_block *block, struct Qdisc *q,
1503                        struct tcf_block_ext_info *ei)
1504 {
1505         if (!block)
1506                 return;
1507         tcf_chain0_head_change_cb_del(block, ei);
1508         tcf_block_owner_del(block, q, ei->binder_type);
1509
1510         __tcf_block_put(block, q, ei, true);
1511 }
1512 EXPORT_SYMBOL(tcf_block_put_ext);
1513
1514 void tcf_block_put(struct tcf_block *block)
1515 {
1516         struct tcf_block_ext_info ei = {0, };
1517
1518         if (!block)
1519                 return;
1520         tcf_block_put_ext(block, block->q, &ei);
1521 }
1522
1523 EXPORT_SYMBOL(tcf_block_put);
1524
1525 static int
1526 tcf_block_playback_offloads(struct tcf_block *block, flow_setup_cb_t *cb,
1527                             void *cb_priv, bool add, bool offload_in_use,
1528                             struct netlink_ext_ack *extack)
1529 {
1530         struct tcf_chain *chain, *chain_prev;
1531         struct tcf_proto *tp, *tp_prev;
1532         int err;
1533
1534         lockdep_assert_held(&block->cb_lock);
1535
1536         for (chain = __tcf_get_next_chain(block, NULL);
1537              chain;
1538              chain_prev = chain,
1539                      chain = __tcf_get_next_chain(block, chain),
1540                      tcf_chain_put(chain_prev)) {
1541                 for (tp = __tcf_get_next_proto(chain, NULL); tp;
1542                      tp_prev = tp,
1543                              tp = __tcf_get_next_proto(chain, tp),
1544                              tcf_proto_put(tp_prev, true, NULL)) {
1545                         if (tp->ops->reoffload) {
1546                                 err = tp->ops->reoffload(tp, add, cb, cb_priv,
1547                                                          extack);
1548                                 if (err && add)
1549                                         goto err_playback_remove;
1550                         } else if (add && offload_in_use) {
1551                                 err = -EOPNOTSUPP;
1552                                 NL_SET_ERR_MSG(extack, "Filter HW offload failed - classifier without re-offloading support");
1553                                 goto err_playback_remove;
1554                         }
1555                 }
1556         }
1557
1558         return 0;
1559
1560 err_playback_remove:
1561         tcf_proto_put(tp, true, NULL);
1562         tcf_chain_put(chain);
1563         tcf_block_playback_offloads(block, cb, cb_priv, false, offload_in_use,
1564                                     extack);
1565         return err;
1566 }
1567
1568 static int tcf_block_bind(struct tcf_block *block,
1569                           struct flow_block_offload *bo)
1570 {
1571         struct flow_block_cb *block_cb, *next;
1572         int err, i = 0;
1573
1574         lockdep_assert_held(&block->cb_lock);
1575
1576         list_for_each_entry(block_cb, &bo->cb_list, list) {
1577                 err = tcf_block_playback_offloads(block, block_cb->cb,
1578                                                   block_cb->cb_priv, true,
1579                                                   tcf_block_offload_in_use(block),
1580                                                   bo->extack);
1581                 if (err)
1582                         goto err_unroll;
1583                 if (!bo->unlocked_driver_cb)
1584                         block->lockeddevcnt++;
1585
1586                 i++;
1587         }
1588         list_splice(&bo->cb_list, &block->flow_block.cb_list);
1589
1590         return 0;
1591
1592 err_unroll:
1593         list_for_each_entry_safe(block_cb, next, &bo->cb_list, list) {
1594                 list_del(&block_cb->driver_list);
1595                 if (i-- > 0) {
1596                         list_del(&block_cb->list);
1597                         tcf_block_playback_offloads(block, block_cb->cb,
1598                                                     block_cb->cb_priv, false,
1599                                                     tcf_block_offload_in_use(block),
1600                                                     NULL);
1601                         if (!bo->unlocked_driver_cb)
1602                                 block->lockeddevcnt--;
1603                 }
1604                 flow_block_cb_free(block_cb);
1605         }
1606
1607         return err;
1608 }
1609
1610 static void tcf_block_unbind(struct tcf_block *block,
1611                              struct flow_block_offload *bo)
1612 {
1613         struct flow_block_cb *block_cb, *next;
1614
1615         lockdep_assert_held(&block->cb_lock);
1616
1617         list_for_each_entry_safe(block_cb, next, &bo->cb_list, list) {
1618                 tcf_block_playback_offloads(block, block_cb->cb,
1619                                             block_cb->cb_priv, false,
1620                                             tcf_block_offload_in_use(block),
1621                                             NULL);
1622                 list_del(&block_cb->list);
1623                 flow_block_cb_free(block_cb);
1624                 if (!bo->unlocked_driver_cb)
1625                         block->lockeddevcnt--;
1626         }
1627 }
1628
1629 static int tcf_block_setup(struct tcf_block *block,
1630                            struct flow_block_offload *bo)
1631 {
1632         int err;
1633
1634         switch (bo->command) {
1635         case FLOW_BLOCK_BIND:
1636                 err = tcf_block_bind(block, bo);
1637                 break;
1638         case FLOW_BLOCK_UNBIND:
1639                 err = 0;
1640                 tcf_block_unbind(block, bo);
1641                 break;
1642         default:
1643                 WARN_ON_ONCE(1);
1644                 err = -EOPNOTSUPP;
1645         }
1646
1647         return err;
1648 }
1649
1650 /* Main classifier routine: scans classifier chain attached
1651  * to this qdisc, (optionally) tests for protocol and asks
1652  * specific classifiers.
1653  */
1654 static inline int __tcf_classify(struct sk_buff *skb,
1655                                  const struct tcf_proto *tp,
1656                                  const struct tcf_proto *orig_tp,
1657                                  struct tcf_result *res,
1658                                  bool compat_mode,
1659                                  struct tcf_exts_miss_cookie_node *n,
1660                                  int act_index,
1661                                  u32 *last_executed_chain)
1662 {
1663 #ifdef CONFIG_NET_CLS_ACT
1664         const int max_reclassify_loop = 16;
1665         const struct tcf_proto *first_tp;
1666         int limit = 0;
1667
1668 reclassify:
1669 #endif
1670         for (; tp; tp = rcu_dereference_bh(tp->next)) {
1671                 __be16 protocol = skb_protocol(skb, false);
1672                 int err = 0;
1673
1674                 if (n) {
1675                         struct tcf_exts *exts;
1676
1677                         if (n->tp_prio != tp->prio)
1678                                 continue;
1679
1680                         /* We re-lookup the tp and chain based on index instead
1681                          * of having hard refs and locks to them, so do a sanity
1682                          * check if any of tp,chain,exts was replaced by the
1683                          * time we got here with a cookie from hardware.
1684                          */
1685                         if (unlikely(n->tp != tp || n->tp->chain != n->chain ||
1686                                      !tp->ops->get_exts)) {
1687                                 tcf_set_drop_reason(skb,
1688                                                     SKB_DROP_REASON_TC_COOKIE_ERROR);
1689                                 return TC_ACT_SHOT;
1690                         }
1691
1692                         exts = tp->ops->get_exts(tp, n->handle);
1693                         if (unlikely(!exts || n->exts != exts)) {
1694                                 tcf_set_drop_reason(skb,
1695                                                     SKB_DROP_REASON_TC_COOKIE_ERROR);
1696                                 return TC_ACT_SHOT;
1697                         }
1698
1699                         n = NULL;
1700                         err = tcf_exts_exec_ex(skb, exts, act_index, res);
1701                 } else {
1702                         if (tp->protocol != protocol &&
1703                             tp->protocol != htons(ETH_P_ALL))
1704                                 continue;
1705
1706                         err = tc_classify(skb, tp, res);
1707                 }
1708 #ifdef CONFIG_NET_CLS_ACT
1709                 if (unlikely(err == TC_ACT_RECLASSIFY && !compat_mode)) {
1710                         first_tp = orig_tp;
1711                         *last_executed_chain = first_tp->chain->index;
1712                         goto reset;
1713                 } else if (unlikely(TC_ACT_EXT_CMP(err, TC_ACT_GOTO_CHAIN))) {
1714                         first_tp = res->goto_tp;
1715                         *last_executed_chain = err & TC_ACT_EXT_VAL_MASK;
1716                         goto reset;
1717                 }
1718 #endif
1719                 if (err >= 0)
1720                         return err;
1721         }
1722
1723         if (unlikely(n)) {
1724                 tcf_set_drop_reason(skb,
1725                                     SKB_DROP_REASON_TC_COOKIE_ERROR);
1726                 return TC_ACT_SHOT;
1727         }
1728
1729         return TC_ACT_UNSPEC; /* signal: continue lookup */
1730 #ifdef CONFIG_NET_CLS_ACT
1731 reset:
1732         if (unlikely(limit++ >= max_reclassify_loop)) {
1733                 net_notice_ratelimited("%u: reclassify loop, rule prio %u, protocol %02x\n",
1734                                        tp->chain->block->index,
1735                                        tp->prio & 0xffff,
1736                                        ntohs(tp->protocol));
1737                 tcf_set_drop_reason(skb,
1738                                     SKB_DROP_REASON_TC_RECLASSIFY_LOOP);
1739                 return TC_ACT_SHOT;
1740         }
1741
1742         tp = first_tp;
1743         goto reclassify;
1744 #endif
1745 }
1746
1747 int tcf_classify(struct sk_buff *skb,
1748                  const struct tcf_block *block,
1749                  const struct tcf_proto *tp,
1750                  struct tcf_result *res, bool compat_mode)
1751 {
1752 #if !IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
1753         u32 last_executed_chain = 0;
1754
1755         return __tcf_classify(skb, tp, tp, res, compat_mode, NULL, 0,
1756                               &last_executed_chain);
1757 #else
1758         u32 last_executed_chain = tp ? tp->chain->index : 0;
1759         struct tcf_exts_miss_cookie_node *n = NULL;
1760         const struct tcf_proto *orig_tp = tp;
1761         struct tc_skb_ext *ext;
1762         int act_index = 0;
1763         int ret;
1764
1765         if (block) {
1766                 ext = skb_ext_find(skb, TC_SKB_EXT);
1767
1768                 if (ext && (ext->chain || ext->act_miss)) {
1769                         struct tcf_chain *fchain;
1770                         u32 chain;
1771
1772                         if (ext->act_miss) {
1773                                 n = tcf_exts_miss_cookie_lookup(ext->act_miss_cookie,
1774                                                                 &act_index);
1775                                 if (!n) {
1776                                         tcf_set_drop_reason(skb,
1777                                                             SKB_DROP_REASON_TC_COOKIE_ERROR);
1778                                         return TC_ACT_SHOT;
1779                                 }
1780
1781                                 chain = n->chain_index;
1782                         } else {
1783                                 chain = ext->chain;
1784                         }
1785
1786                         fchain = tcf_chain_lookup_rcu(block, chain);
1787                         if (!fchain) {
1788                                 tcf_set_drop_reason(skb,
1789                                                     SKB_DROP_REASON_TC_CHAIN_NOTFOUND);
1790
1791                                 return TC_ACT_SHOT;
1792                         }
1793
1794                         /* Consume, so cloned/redirect skbs won't inherit ext */
1795                         skb_ext_del(skb, TC_SKB_EXT);
1796
1797                         tp = rcu_dereference_bh(fchain->filter_chain);
1798                         last_executed_chain = fchain->index;
1799                 }
1800         }
1801
1802         ret = __tcf_classify(skb, tp, orig_tp, res, compat_mode, n, act_index,
1803                              &last_executed_chain);
1804
1805         if (tc_skb_ext_tc_enabled()) {
1806                 /* If we missed on some chain */
1807                 if (ret == TC_ACT_UNSPEC && last_executed_chain) {
1808                         struct tc_skb_cb *cb = tc_skb_cb(skb);
1809
1810                         ext = tc_skb_ext_alloc(skb);
1811                         if (WARN_ON_ONCE(!ext)) {
1812                                 tcf_set_drop_reason(skb, SKB_DROP_REASON_NOMEM);
1813                                 return TC_ACT_SHOT;
1814                         }
1815                         ext->chain = last_executed_chain;
1816                         ext->mru = cb->mru;
1817                         ext->post_ct = cb->post_ct;
1818                         ext->post_ct_snat = cb->post_ct_snat;
1819                         ext->post_ct_dnat = cb->post_ct_dnat;
1820                         ext->zone = cb->zone;
1821                 }
1822         }
1823
1824         return ret;
1825 #endif
1826 }
1827 EXPORT_SYMBOL(tcf_classify);
1828
1829 struct tcf_chain_info {
1830         struct tcf_proto __rcu **pprev;
1831         struct tcf_proto __rcu *next;
1832 };
1833
1834 static struct tcf_proto *tcf_chain_tp_prev(struct tcf_chain *chain,
1835                                            struct tcf_chain_info *chain_info)
1836 {
1837         return tcf_chain_dereference(*chain_info->pprev, chain);
1838 }
1839
1840 static int tcf_chain_tp_insert(struct tcf_chain *chain,
1841                                struct tcf_chain_info *chain_info,
1842                                struct tcf_proto *tp)
1843 {
1844         if (chain->flushing)
1845                 return -EAGAIN;
1846
1847         RCU_INIT_POINTER(tp->next, tcf_chain_tp_prev(chain, chain_info));
1848         if (*chain_info->pprev == chain->filter_chain)
1849                 tcf_chain0_head_change(chain, tp);
1850         tcf_proto_get(tp);
1851         rcu_assign_pointer(*chain_info->pprev, tp);
1852
1853         return 0;
1854 }
1855
1856 static void tcf_chain_tp_remove(struct tcf_chain *chain,
1857                                 struct tcf_chain_info *chain_info,
1858                                 struct tcf_proto *tp)
1859 {
1860         struct tcf_proto *next = tcf_chain_dereference(chain_info->next, chain);
1861
1862         tcf_proto_mark_delete(tp);
1863         if (tp == chain->filter_chain)
1864                 tcf_chain0_head_change(chain, next);
1865         RCU_INIT_POINTER(*chain_info->pprev, next);
1866 }
1867
1868 static struct tcf_proto *tcf_chain_tp_find(struct tcf_chain *chain,
1869                                            struct tcf_chain_info *chain_info,
1870                                            u32 protocol, u32 prio,
1871                                            bool prio_allocate);
1872
1873 /* Try to insert new proto.
1874  * If proto with specified priority already exists, free new proto
1875  * and return existing one.
1876  */
1877
1878 static struct tcf_proto *tcf_chain_tp_insert_unique(struct tcf_chain *chain,
1879                                                     struct tcf_proto *tp_new,
1880                                                     u32 protocol, u32 prio,
1881                                                     bool rtnl_held)
1882 {
1883         struct tcf_chain_info chain_info;
1884         struct tcf_proto *tp;
1885         int err = 0;
1886
1887         mutex_lock(&chain->filter_chain_lock);
1888
1889         if (tcf_proto_exists_destroying(chain, tp_new)) {
1890                 mutex_unlock(&chain->filter_chain_lock);
1891                 tcf_proto_destroy(tp_new, rtnl_held, false, NULL);
1892                 return ERR_PTR(-EAGAIN);
1893         }
1894
1895         tp = tcf_chain_tp_find(chain, &chain_info,
1896                                protocol, prio, false);
1897         if (!tp)
1898                 err = tcf_chain_tp_insert(chain, &chain_info, tp_new);
1899         mutex_unlock(&chain->filter_chain_lock);
1900
1901         if (tp) {
1902                 tcf_proto_destroy(tp_new, rtnl_held, false, NULL);
1903                 tp_new = tp;
1904         } else if (err) {
1905                 tcf_proto_destroy(tp_new, rtnl_held, false, NULL);
1906                 tp_new = ERR_PTR(err);
1907         }
1908
1909         return tp_new;
1910 }
1911
1912 static void tcf_chain_tp_delete_empty(struct tcf_chain *chain,
1913                                       struct tcf_proto *tp, bool rtnl_held,
1914                                       struct netlink_ext_ack *extack)
1915 {
1916         struct tcf_chain_info chain_info;
1917         struct tcf_proto *tp_iter;
1918         struct tcf_proto **pprev;
1919         struct tcf_proto *next;
1920
1921         mutex_lock(&chain->filter_chain_lock);
1922
1923         /* Atomically find and remove tp from chain. */
1924         for (pprev = &chain->filter_chain;
1925              (tp_iter = tcf_chain_dereference(*pprev, chain));
1926              pprev = &tp_iter->next) {
1927                 if (tp_iter == tp) {
1928                         chain_info.pprev = pprev;
1929                         chain_info.next = tp_iter->next;
1930                         WARN_ON(tp_iter->deleting);
1931                         break;
1932                 }
1933         }
1934         /* Verify that tp still exists and no new filters were inserted
1935          * concurrently.
1936          * Mark tp for deletion if it is empty.
1937          */
1938         if (!tp_iter || !tcf_proto_check_delete(tp)) {
1939                 mutex_unlock(&chain->filter_chain_lock);
1940                 return;
1941         }
1942
1943         tcf_proto_signal_destroying(chain, tp);
1944         next = tcf_chain_dereference(chain_info.next, chain);
1945         if (tp == chain->filter_chain)
1946                 tcf_chain0_head_change(chain, next);
1947         RCU_INIT_POINTER(*chain_info.pprev, next);
1948         mutex_unlock(&chain->filter_chain_lock);
1949
1950         tcf_proto_put(tp, rtnl_held, extack);
1951 }
1952
1953 static struct tcf_proto *tcf_chain_tp_find(struct tcf_chain *chain,
1954                                            struct tcf_chain_info *chain_info,
1955                                            u32 protocol, u32 prio,
1956                                            bool prio_allocate)
1957 {
1958         struct tcf_proto **pprev;
1959         struct tcf_proto *tp;
1960
1961         /* Check the chain for existence of proto-tcf with this priority */
1962         for (pprev = &chain->filter_chain;
1963              (tp = tcf_chain_dereference(*pprev, chain));
1964              pprev = &tp->next) {
1965                 if (tp->prio >= prio) {
1966                         if (tp->prio == prio) {
1967                                 if (prio_allocate ||
1968                                     (tp->protocol != protocol && protocol))
1969                                         return ERR_PTR(-EINVAL);
1970                         } else {
1971                                 tp = NULL;
1972                         }
1973                         break;
1974                 }
1975         }
1976         chain_info->pprev = pprev;
1977         if (tp) {
1978                 chain_info->next = tp->next;
1979                 tcf_proto_get(tp);
1980         } else {
1981                 chain_info->next = NULL;
1982         }
1983         return tp;
1984 }
1985
1986 static int tcf_fill_node(struct net *net, struct sk_buff *skb,
1987                          struct tcf_proto *tp, struct tcf_block *block,
1988                          struct Qdisc *q, u32 parent, void *fh,
1989                          u32 portid, u32 seq, u16 flags, int event,
1990                          bool terse_dump, bool rtnl_held,
1991                          struct netlink_ext_ack *extack)
1992 {
1993         struct tcmsg *tcm;
1994         struct nlmsghdr  *nlh;
1995         unsigned char *b = skb_tail_pointer(skb);
1996
1997         nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
1998         if (!nlh)
1999                 goto out_nlmsg_trim;
2000         tcm = nlmsg_data(nlh);
2001         tcm->tcm_family = AF_UNSPEC;
2002         tcm->tcm__pad1 = 0;
2003         tcm->tcm__pad2 = 0;
2004         if (q) {
2005                 tcm->tcm_ifindex = qdisc_dev(q)->ifindex;
2006                 tcm->tcm_parent = parent;
2007         } else {
2008                 tcm->tcm_ifindex = TCM_IFINDEX_MAGIC_BLOCK;
2009                 tcm->tcm_block_index = block->index;
2010         }
2011         tcm->tcm_info = TC_H_MAKE(tp->prio, tp->protocol);
2012         if (nla_put_string(skb, TCA_KIND, tp->ops->kind))
2013                 goto nla_put_failure;
2014         if (nla_put_u32(skb, TCA_CHAIN, tp->chain->index))
2015                 goto nla_put_failure;
2016         if (!fh) {
2017                 tcm->tcm_handle = 0;
2018         } else if (terse_dump) {
2019                 if (tp->ops->terse_dump) {
2020                         if (tp->ops->terse_dump(net, tp, fh, skb, tcm,
2021                                                 rtnl_held) < 0)
2022                                 goto nla_put_failure;
2023                 } else {
2024                         goto cls_op_not_supp;
2025                 }
2026         } else {
2027                 if (tp->ops->dump &&
2028                     tp->ops->dump(net, tp, fh, skb, tcm, rtnl_held) < 0)
2029                         goto nla_put_failure;
2030         }
2031
2032         if (extack && extack->_msg &&
2033             nla_put_string(skb, TCA_EXT_WARN_MSG, extack->_msg))
2034                 goto nla_put_failure;
2035
2036         nlh->nlmsg_len = skb_tail_pointer(skb) - b;
2037
2038         return skb->len;
2039
2040 out_nlmsg_trim:
2041 nla_put_failure:
2042 cls_op_not_supp:
2043         nlmsg_trim(skb, b);
2044         return -1;
2045 }
2046
2047 static int tfilter_notify(struct net *net, struct sk_buff *oskb,
2048                           struct nlmsghdr *n, struct tcf_proto *tp,
2049                           struct tcf_block *block, struct Qdisc *q,
2050                           u32 parent, void *fh, int event, bool unicast,
2051                           bool rtnl_held, struct netlink_ext_ack *extack)
2052 {
2053         struct sk_buff *skb;
2054         u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
2055         int err = 0;
2056
2057         if (!unicast && !rtnl_notify_needed(net, n->nlmsg_flags, RTNLGRP_TC))
2058                 return 0;
2059
2060         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2061         if (!skb)
2062                 return -ENOBUFS;
2063
2064         if (tcf_fill_node(net, skb, tp, block, q, parent, fh, portid,
2065                           n->nlmsg_seq, n->nlmsg_flags, event,
2066                           false, rtnl_held, extack) <= 0) {
2067                 kfree_skb(skb);
2068                 return -EINVAL;
2069         }
2070
2071         if (unicast)
2072                 err = rtnl_unicast(skb, net, portid);
2073         else
2074                 err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
2075                                      n->nlmsg_flags & NLM_F_ECHO);
2076         return err;
2077 }
2078
2079 static int tfilter_del_notify(struct net *net, struct sk_buff *oskb,
2080                               struct nlmsghdr *n, struct tcf_proto *tp,
2081                               struct tcf_block *block, struct Qdisc *q,
2082                               u32 parent, void *fh, bool *last, bool rtnl_held,
2083                               struct netlink_ext_ack *extack)
2084 {
2085         struct sk_buff *skb;
2086         u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
2087         int err;
2088
2089         if (!rtnl_notify_needed(net, n->nlmsg_flags, RTNLGRP_TC))
2090                 return tp->ops->delete(tp, fh, last, rtnl_held, extack);
2091
2092         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2093         if (!skb)
2094                 return -ENOBUFS;
2095
2096         if (tcf_fill_node(net, skb, tp, block, q, parent, fh, portid,
2097                           n->nlmsg_seq, n->nlmsg_flags, RTM_DELTFILTER,
2098                           false, rtnl_held, extack) <= 0) {
2099                 NL_SET_ERR_MSG(extack, "Failed to build del event notification");
2100                 kfree_skb(skb);
2101                 return -EINVAL;
2102         }
2103
2104         err = tp->ops->delete(tp, fh, last, rtnl_held, extack);
2105         if (err) {
2106                 kfree_skb(skb);
2107                 return err;
2108         }
2109
2110         err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
2111                              n->nlmsg_flags & NLM_F_ECHO);
2112         if (err < 0)
2113                 NL_SET_ERR_MSG(extack, "Failed to send filter delete notification");
2114
2115         return err;
2116 }
2117
2118 static void tfilter_notify_chain(struct net *net, struct sk_buff *oskb,
2119                                  struct tcf_block *block, struct Qdisc *q,
2120                                  u32 parent, struct nlmsghdr *n,
2121                                  struct tcf_chain *chain, int event,
2122                                  struct netlink_ext_ack *extack)
2123 {
2124         struct tcf_proto *tp;
2125
2126         for (tp = tcf_get_next_proto(chain, NULL);
2127              tp; tp = tcf_get_next_proto(chain, tp))
2128                 tfilter_notify(net, oskb, n, tp, block, q, parent, NULL,
2129                                event, false, true, extack);
2130 }
2131
2132 static void tfilter_put(struct tcf_proto *tp, void *fh)
2133 {
2134         if (tp->ops->put && fh)
2135                 tp->ops->put(tp, fh);
2136 }
2137
2138 static bool is_qdisc_ingress(__u32 classid)
2139 {
2140         return (TC_H_MIN(classid) == TC_H_MIN(TC_H_MIN_INGRESS));
2141 }
2142
2143 static int tc_new_tfilter(struct sk_buff *skb, struct nlmsghdr *n,
2144                           struct netlink_ext_ack *extack)
2145 {
2146         struct net *net = sock_net(skb->sk);
2147         struct nlattr *tca[TCA_MAX + 1];
2148         char name[IFNAMSIZ];
2149         struct tcmsg *t;
2150         u32 protocol;
2151         u32 prio;
2152         bool prio_allocate;
2153         u32 parent;
2154         u32 chain_index;
2155         struct Qdisc *q;
2156         struct tcf_chain_info chain_info;
2157         struct tcf_chain *chain;
2158         struct tcf_block *block;
2159         struct tcf_proto *tp;
2160         unsigned long cl;
2161         void *fh;
2162         int err;
2163         int tp_created;
2164         bool rtnl_held = false;
2165         u32 flags;
2166
2167 replay:
2168         tp_created = 0;
2169
2170         err = nlmsg_parse_deprecated(n, sizeof(*t), tca, TCA_MAX,
2171                                      rtm_tca_policy, extack);
2172         if (err < 0)
2173                 return err;
2174
2175         t = nlmsg_data(n);
2176         protocol = TC_H_MIN(t->tcm_info);
2177         prio = TC_H_MAJ(t->tcm_info);
2178         prio_allocate = false;
2179         parent = t->tcm_parent;
2180         tp = NULL;
2181         cl = 0;
2182         block = NULL;
2183         q = NULL;
2184         chain = NULL;
2185         flags = 0;
2186
2187         if (prio == 0) {
2188                 /* If no priority is provided by the user,
2189                  * we allocate one.
2190                  */
2191                 if (n->nlmsg_flags & NLM_F_CREATE) {
2192                         prio = TC_H_MAKE(0x80000000U, 0U);
2193                         prio_allocate = true;
2194                 } else {
2195                         NL_SET_ERR_MSG(extack, "Invalid filter command with priority of zero");
2196                         return -ENOENT;
2197                 }
2198         }
2199
2200         /* Find head of filter chain. */
2201
2202         err = __tcf_qdisc_find(net, &q, &parent, t->tcm_ifindex, false, extack);
2203         if (err)
2204                 return err;
2205
2206         if (tcf_proto_check_kind(tca[TCA_KIND], name)) {
2207                 NL_SET_ERR_MSG(extack, "Specified TC filter name too long");
2208                 err = -EINVAL;
2209                 goto errout;
2210         }
2211
2212         /* Take rtnl mutex if rtnl_held was set to true on previous iteration,
2213          * block is shared (no qdisc found), qdisc is not unlocked, classifier
2214          * type is not specified, classifier is not unlocked.
2215          */
2216         if (rtnl_held ||
2217             (q && !(q->ops->cl_ops->flags & QDISC_CLASS_OPS_DOIT_UNLOCKED)) ||
2218             !tcf_proto_is_unlocked(name)) {
2219                 rtnl_held = true;
2220                 rtnl_lock();
2221         }
2222
2223         err = __tcf_qdisc_cl_find(q, parent, &cl, t->tcm_ifindex, extack);
2224         if (err)
2225                 goto errout;
2226
2227         block = __tcf_block_find(net, q, cl, t->tcm_ifindex, t->tcm_block_index,
2228                                  extack);
2229         if (IS_ERR(block)) {
2230                 err = PTR_ERR(block);
2231                 goto errout;
2232         }
2233         block->classid = parent;
2234
2235         chain_index = tca[TCA_CHAIN] ? nla_get_u32(tca[TCA_CHAIN]) : 0;
2236         if (chain_index > TC_ACT_EXT_VAL_MASK) {
2237                 NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
2238                 err = -EINVAL;
2239                 goto errout;
2240         }
2241         chain = tcf_chain_get(block, chain_index, true);
2242         if (!chain) {
2243                 NL_SET_ERR_MSG(extack, "Cannot create specified filter chain");
2244                 err = -ENOMEM;
2245                 goto errout;
2246         }
2247
2248         mutex_lock(&chain->filter_chain_lock);
2249         tp = tcf_chain_tp_find(chain, &chain_info, protocol,
2250                                prio, prio_allocate);
2251         if (IS_ERR(tp)) {
2252                 NL_SET_ERR_MSG(extack, "Filter with specified priority/protocol not found");
2253                 err = PTR_ERR(tp);
2254                 goto errout_locked;
2255         }
2256
2257         if (tp == NULL) {
2258                 struct tcf_proto *tp_new = NULL;
2259
2260                 if (chain->flushing) {
2261                         err = -EAGAIN;
2262                         goto errout_locked;
2263                 }
2264
2265                 /* Proto-tcf does not exist, create new one */
2266
2267                 if (tca[TCA_KIND] == NULL || !protocol) {
2268                         NL_SET_ERR_MSG(extack, "Filter kind and protocol must be specified");
2269                         err = -EINVAL;
2270                         goto errout_locked;
2271                 }
2272
2273                 if (!(n->nlmsg_flags & NLM_F_CREATE)) {
2274                         NL_SET_ERR_MSG(extack, "Need both RTM_NEWTFILTER and NLM_F_CREATE to create a new filter");
2275                         err = -ENOENT;
2276                         goto errout_locked;
2277                 }
2278
2279                 if (prio_allocate)
2280                         prio = tcf_auto_prio(tcf_chain_tp_prev(chain,
2281                                                                &chain_info));
2282
2283                 mutex_unlock(&chain->filter_chain_lock);
2284                 tp_new = tcf_proto_create(name, protocol, prio, chain,
2285                                           rtnl_held, extack);
2286                 if (IS_ERR(tp_new)) {
2287                         err = PTR_ERR(tp_new);
2288                         goto errout_tp;
2289                 }
2290
2291                 tp_created = 1;
2292                 tp = tcf_chain_tp_insert_unique(chain, tp_new, protocol, prio,
2293                                                 rtnl_held);
2294                 if (IS_ERR(tp)) {
2295                         err = PTR_ERR(tp);
2296                         goto errout_tp;
2297                 }
2298         } else {
2299                 mutex_unlock(&chain->filter_chain_lock);
2300         }
2301
2302         if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind)) {
2303                 NL_SET_ERR_MSG(extack, "Specified filter kind does not match existing one");
2304                 err = -EINVAL;
2305                 goto errout;
2306         }
2307
2308         fh = tp->ops->get(tp, t->tcm_handle);
2309
2310         if (!fh) {
2311                 if (!(n->nlmsg_flags & NLM_F_CREATE)) {
2312                         NL_SET_ERR_MSG(extack, "Need both RTM_NEWTFILTER and NLM_F_CREATE to create a new filter");
2313                         err = -ENOENT;
2314                         goto errout;
2315                 }
2316         } else if (n->nlmsg_flags & NLM_F_EXCL) {
2317                 tfilter_put(tp, fh);
2318                 NL_SET_ERR_MSG(extack, "Filter already exists");
2319                 err = -EEXIST;
2320                 goto errout;
2321         }
2322
2323         if (chain->tmplt_ops && chain->tmplt_ops != tp->ops) {
2324                 tfilter_put(tp, fh);
2325                 NL_SET_ERR_MSG(extack, "Chain template is set to a different filter kind");
2326                 err = -EINVAL;
2327                 goto errout;
2328         }
2329
2330         if (!(n->nlmsg_flags & NLM_F_CREATE))
2331                 flags |= TCA_ACT_FLAGS_REPLACE;
2332         if (!rtnl_held)
2333                 flags |= TCA_ACT_FLAGS_NO_RTNL;
2334         if (is_qdisc_ingress(parent))
2335                 flags |= TCA_ACT_FLAGS_AT_INGRESS;
2336         err = tp->ops->change(net, skb, tp, cl, t->tcm_handle, tca, &fh,
2337                               flags, extack);
2338         if (err == 0) {
2339                 tfilter_notify(net, skb, n, tp, block, q, parent, fh,
2340                                RTM_NEWTFILTER, false, rtnl_held, extack);
2341                 tfilter_put(tp, fh);
2342                 /* q pointer is NULL for shared blocks */
2343                 if (q)
2344                         q->flags &= ~TCQ_F_CAN_BYPASS;
2345         }
2346
2347 errout:
2348         if (err && tp_created)
2349                 tcf_chain_tp_delete_empty(chain, tp, rtnl_held, NULL);
2350 errout_tp:
2351         if (chain) {
2352                 if (tp && !IS_ERR(tp))
2353                         tcf_proto_put(tp, rtnl_held, NULL);
2354                 if (!tp_created)
2355                         tcf_chain_put(chain);
2356         }
2357         tcf_block_release(q, block, rtnl_held);
2358
2359         if (rtnl_held)
2360                 rtnl_unlock();
2361
2362         if (err == -EAGAIN) {
2363                 /* Take rtnl lock in case EAGAIN is caused by concurrent flush
2364                  * of target chain.
2365                  */
2366                 rtnl_held = true;
2367                 /* Replay the request. */
2368                 goto replay;
2369         }
2370         return err;
2371
2372 errout_locked:
2373         mutex_unlock(&chain->filter_chain_lock);
2374         goto errout;
2375 }
2376
2377 static int tc_del_tfilter(struct sk_buff *skb, struct nlmsghdr *n,
2378                           struct netlink_ext_ack *extack)
2379 {
2380         struct net *net = sock_net(skb->sk);
2381         struct nlattr *tca[TCA_MAX + 1];
2382         char name[IFNAMSIZ];
2383         struct tcmsg *t;
2384         u32 protocol;
2385         u32 prio;
2386         u32 parent;
2387         u32 chain_index;
2388         struct Qdisc *q = NULL;
2389         struct tcf_chain_info chain_info;
2390         struct tcf_chain *chain = NULL;
2391         struct tcf_block *block = NULL;
2392         struct tcf_proto *tp = NULL;
2393         unsigned long cl = 0;
2394         void *fh = NULL;
2395         int err;
2396         bool rtnl_held = false;
2397
2398         err = nlmsg_parse_deprecated(n, sizeof(*t), tca, TCA_MAX,
2399                                      rtm_tca_policy, extack);
2400         if (err < 0)
2401                 return err;
2402
2403         t = nlmsg_data(n);
2404         protocol = TC_H_MIN(t->tcm_info);
2405         prio = TC_H_MAJ(t->tcm_info);
2406         parent = t->tcm_parent;
2407
2408         if (prio == 0 && (protocol || t->tcm_handle || tca[TCA_KIND])) {
2409                 NL_SET_ERR_MSG(extack, "Cannot flush filters with protocol, handle or kind set");
2410                 return -ENOENT;
2411         }
2412
2413         /* Find head of filter chain. */
2414
2415         err = __tcf_qdisc_find(net, &q, &parent, t->tcm_ifindex, false, extack);
2416         if (err)
2417                 return err;
2418
2419         if (tcf_proto_check_kind(tca[TCA_KIND], name)) {
2420                 NL_SET_ERR_MSG(extack, "Specified TC filter name too long");
2421                 err = -EINVAL;
2422                 goto errout;
2423         }
2424         /* Take rtnl mutex if flushing whole chain, block is shared (no qdisc
2425          * found), qdisc is not unlocked, classifier type is not specified,
2426          * classifier is not unlocked.
2427          */
2428         if (!prio ||
2429             (q && !(q->ops->cl_ops->flags & QDISC_CLASS_OPS_DOIT_UNLOCKED)) ||
2430             !tcf_proto_is_unlocked(name)) {
2431                 rtnl_held = true;
2432                 rtnl_lock();
2433         }
2434
2435         err = __tcf_qdisc_cl_find(q, parent, &cl, t->tcm_ifindex, extack);
2436         if (err)
2437                 goto errout;
2438
2439         block = __tcf_block_find(net, q, cl, t->tcm_ifindex, t->tcm_block_index,
2440                                  extack);
2441         if (IS_ERR(block)) {
2442                 err = PTR_ERR(block);
2443                 goto errout;
2444         }
2445
2446         chain_index = tca[TCA_CHAIN] ? nla_get_u32(tca[TCA_CHAIN]) : 0;
2447         if (chain_index > TC_ACT_EXT_VAL_MASK) {
2448                 NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
2449                 err = -EINVAL;
2450                 goto errout;
2451         }
2452         chain = tcf_chain_get(block, chain_index, false);
2453         if (!chain) {
2454                 /* User requested flush on non-existent chain. Nothing to do,
2455                  * so just return success.
2456                  */
2457                 if (prio == 0) {
2458                         err = 0;
2459                         goto errout;
2460                 }
2461                 NL_SET_ERR_MSG(extack, "Cannot find specified filter chain");
2462                 err = -ENOENT;
2463                 goto errout;
2464         }
2465
2466         if (prio == 0) {
2467                 tfilter_notify_chain(net, skb, block, q, parent, n,
2468                                      chain, RTM_DELTFILTER, extack);
2469                 tcf_chain_flush(chain, rtnl_held);
2470                 err = 0;
2471                 goto errout;
2472         }
2473
2474         mutex_lock(&chain->filter_chain_lock);
2475         tp = tcf_chain_tp_find(chain, &chain_info, protocol,
2476                                prio, false);
2477         if (!tp || IS_ERR(tp)) {
2478                 NL_SET_ERR_MSG(extack, "Filter with specified priority/protocol not found");
2479                 err = tp ? PTR_ERR(tp) : -ENOENT;
2480                 goto errout_locked;
2481         } else if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind)) {
2482                 NL_SET_ERR_MSG(extack, "Specified filter kind does not match existing one");
2483                 err = -EINVAL;
2484                 goto errout_locked;
2485         } else if (t->tcm_handle == 0) {
2486                 tcf_proto_signal_destroying(chain, tp);
2487                 tcf_chain_tp_remove(chain, &chain_info, tp);
2488                 mutex_unlock(&chain->filter_chain_lock);
2489
2490                 tcf_proto_put(tp, rtnl_held, NULL);
2491                 tfilter_notify(net, skb, n, tp, block, q, parent, fh,
2492                                RTM_DELTFILTER, false, rtnl_held, extack);
2493                 err = 0;
2494                 goto errout;
2495         }
2496         mutex_unlock(&chain->filter_chain_lock);
2497
2498         fh = tp->ops->get(tp, t->tcm_handle);
2499
2500         if (!fh) {
2501                 NL_SET_ERR_MSG(extack, "Specified filter handle not found");
2502                 err = -ENOENT;
2503         } else {
2504                 bool last;
2505
2506                 err = tfilter_del_notify(net, skb, n, tp, block, q, parent, fh,
2507                                          &last, rtnl_held, extack);
2508
2509                 if (err)
2510                         goto errout;
2511                 if (last)
2512                         tcf_chain_tp_delete_empty(chain, tp, rtnl_held, extack);
2513         }
2514
2515 errout:
2516         if (chain) {
2517                 if (tp && !IS_ERR(tp))
2518                         tcf_proto_put(tp, rtnl_held, NULL);
2519                 tcf_chain_put(chain);
2520         }
2521         tcf_block_release(q, block, rtnl_held);
2522
2523         if (rtnl_held)
2524                 rtnl_unlock();
2525
2526         return err;
2527
2528 errout_locked:
2529         mutex_unlock(&chain->filter_chain_lock);
2530         goto errout;
2531 }
2532
2533 static int tc_get_tfilter(struct sk_buff *skb, struct nlmsghdr *n,
2534                           struct netlink_ext_ack *extack)
2535 {
2536         struct net *net = sock_net(skb->sk);
2537         struct nlattr *tca[TCA_MAX + 1];
2538         char name[IFNAMSIZ];
2539         struct tcmsg *t;
2540         u32 protocol;
2541         u32 prio;
2542         u32 parent;
2543         u32 chain_index;
2544         struct Qdisc *q = NULL;
2545         struct tcf_chain_info chain_info;
2546         struct tcf_chain *chain = NULL;
2547         struct tcf_block *block = NULL;
2548         struct tcf_proto *tp = NULL;
2549         unsigned long cl = 0;
2550         void *fh = NULL;
2551         int err;
2552         bool rtnl_held = false;
2553
2554         err = nlmsg_parse_deprecated(n, sizeof(*t), tca, TCA_MAX,
2555                                      rtm_tca_policy, extack);
2556         if (err < 0)
2557                 return err;
2558
2559         t = nlmsg_data(n);
2560         protocol = TC_H_MIN(t->tcm_info);
2561         prio = TC_H_MAJ(t->tcm_info);
2562         parent = t->tcm_parent;
2563
2564         if (prio == 0) {
2565                 NL_SET_ERR_MSG(extack, "Invalid filter command with priority of zero");
2566                 return -ENOENT;
2567         }
2568
2569         /* Find head of filter chain. */
2570
2571         err = __tcf_qdisc_find(net, &q, &parent, t->tcm_ifindex, false, extack);
2572         if (err)
2573                 return err;
2574
2575         if (tcf_proto_check_kind(tca[TCA_KIND], name)) {
2576                 NL_SET_ERR_MSG(extack, "Specified TC filter name too long");
2577                 err = -EINVAL;
2578                 goto errout;
2579         }
2580         /* Take rtnl mutex if block is shared (no qdisc found), qdisc is not
2581          * unlocked, classifier type is not specified, classifier is not
2582          * unlocked.
2583          */
2584         if ((q && !(q->ops->cl_ops->flags & QDISC_CLASS_OPS_DOIT_UNLOCKED)) ||
2585             !tcf_proto_is_unlocked(name)) {
2586                 rtnl_held = true;
2587                 rtnl_lock();
2588         }
2589
2590         err = __tcf_qdisc_cl_find(q, parent, &cl, t->tcm_ifindex, extack);
2591         if (err)
2592                 goto errout;
2593
2594         block = __tcf_block_find(net, q, cl, t->tcm_ifindex, t->tcm_block_index,
2595                                  extack);
2596         if (IS_ERR(block)) {
2597                 err = PTR_ERR(block);
2598                 goto errout;
2599         }
2600
2601         chain_index = tca[TCA_CHAIN] ? nla_get_u32(tca[TCA_CHAIN]) : 0;
2602         if (chain_index > TC_ACT_EXT_VAL_MASK) {
2603                 NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
2604                 err = -EINVAL;
2605                 goto errout;
2606         }
2607         chain = tcf_chain_get(block, chain_index, false);
2608         if (!chain) {
2609                 NL_SET_ERR_MSG(extack, "Cannot find specified filter chain");
2610                 err = -EINVAL;
2611                 goto errout;
2612         }
2613
2614         mutex_lock(&chain->filter_chain_lock);
2615         tp = tcf_chain_tp_find(chain, &chain_info, protocol,
2616                                prio, false);
2617         mutex_unlock(&chain->filter_chain_lock);
2618         if (!tp || IS_ERR(tp)) {
2619                 NL_SET_ERR_MSG(extack, "Filter with specified priority/protocol not found");
2620                 err = tp ? PTR_ERR(tp) : -ENOENT;
2621                 goto errout;
2622         } else if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind)) {
2623                 NL_SET_ERR_MSG(extack, "Specified filter kind does not match existing one");
2624                 err = -EINVAL;
2625                 goto errout;
2626         }
2627
2628         fh = tp->ops->get(tp, t->tcm_handle);
2629
2630         if (!fh) {
2631                 NL_SET_ERR_MSG(extack, "Specified filter handle not found");
2632                 err = -ENOENT;
2633         } else {
2634                 err = tfilter_notify(net, skb, n, tp, block, q, parent,
2635                                      fh, RTM_NEWTFILTER, true, rtnl_held, NULL);
2636                 if (err < 0)
2637                         NL_SET_ERR_MSG(extack, "Failed to send filter notify message");
2638         }
2639
2640         tfilter_put(tp, fh);
2641 errout:
2642         if (chain) {
2643                 if (tp && !IS_ERR(tp))
2644                         tcf_proto_put(tp, rtnl_held, NULL);
2645                 tcf_chain_put(chain);
2646         }
2647         tcf_block_release(q, block, rtnl_held);
2648
2649         if (rtnl_held)
2650                 rtnl_unlock();
2651
2652         return err;
2653 }
2654
2655 struct tcf_dump_args {
2656         struct tcf_walker w;
2657         struct sk_buff *skb;
2658         struct netlink_callback *cb;
2659         struct tcf_block *block;
2660         struct Qdisc *q;
2661         u32 parent;
2662         bool terse_dump;
2663 };
2664
2665 static int tcf_node_dump(struct tcf_proto *tp, void *n, struct tcf_walker *arg)
2666 {
2667         struct tcf_dump_args *a = (void *)arg;
2668         struct net *net = sock_net(a->skb->sk);
2669
2670         return tcf_fill_node(net, a->skb, tp, a->block, a->q, a->parent,
2671                              n, NETLINK_CB(a->cb->skb).portid,
2672                              a->cb->nlh->nlmsg_seq, NLM_F_MULTI,
2673                              RTM_NEWTFILTER, a->terse_dump, true, NULL);
2674 }
2675
2676 static bool tcf_chain_dump(struct tcf_chain *chain, struct Qdisc *q, u32 parent,
2677                            struct sk_buff *skb, struct netlink_callback *cb,
2678                            long index_start, long *p_index, bool terse)
2679 {
2680         struct net *net = sock_net(skb->sk);
2681         struct tcf_block *block = chain->block;
2682         struct tcmsg *tcm = nlmsg_data(cb->nlh);
2683         struct tcf_proto *tp, *tp_prev;
2684         struct tcf_dump_args arg;
2685
2686         for (tp = __tcf_get_next_proto(chain, NULL);
2687              tp;
2688              tp_prev = tp,
2689                      tp = __tcf_get_next_proto(chain, tp),
2690                      tcf_proto_put(tp_prev, true, NULL),
2691                      (*p_index)++) {
2692                 if (*p_index < index_start)
2693                         continue;
2694                 if (TC_H_MAJ(tcm->tcm_info) &&
2695                     TC_H_MAJ(tcm->tcm_info) != tp->prio)
2696                         continue;
2697                 if (TC_H_MIN(tcm->tcm_info) &&
2698                     TC_H_MIN(tcm->tcm_info) != tp->protocol)
2699                         continue;
2700                 if (*p_index > index_start)
2701                         memset(&cb->args[1], 0,
2702                                sizeof(cb->args) - sizeof(cb->args[0]));
2703                 if (cb->args[1] == 0) {
2704                         if (tcf_fill_node(net, skb, tp, block, q, parent, NULL,
2705                                           NETLINK_CB(cb->skb).portid,
2706                                           cb->nlh->nlmsg_seq, NLM_F_MULTI,
2707                                           RTM_NEWTFILTER, false, true, NULL) <= 0)
2708                                 goto errout;
2709                         cb->args[1] = 1;
2710                 }
2711                 if (!tp->ops->walk)
2712                         continue;
2713                 arg.w.fn = tcf_node_dump;
2714                 arg.skb = skb;
2715                 arg.cb = cb;
2716                 arg.block = block;
2717                 arg.q = q;
2718                 arg.parent = parent;
2719                 arg.w.stop = 0;
2720                 arg.w.skip = cb->args[1] - 1;
2721                 arg.w.count = 0;
2722                 arg.w.cookie = cb->args[2];
2723                 arg.terse_dump = terse;
2724                 tp->ops->walk(tp, &arg.w, true);
2725                 cb->args[2] = arg.w.cookie;
2726                 cb->args[1] = arg.w.count + 1;
2727                 if (arg.w.stop)
2728                         goto errout;
2729         }
2730         return true;
2731
2732 errout:
2733         tcf_proto_put(tp, true, NULL);
2734         return false;
2735 }
2736
2737 static const struct nla_policy tcf_tfilter_dump_policy[TCA_MAX + 1] = {
2738         [TCA_CHAIN]      = { .type = NLA_U32 },
2739         [TCA_DUMP_FLAGS] = NLA_POLICY_BITFIELD32(TCA_DUMP_FLAGS_TERSE),
2740 };
2741
2742 /* called with RTNL */
2743 static int tc_dump_tfilter(struct sk_buff *skb, struct netlink_callback *cb)
2744 {
2745         struct tcf_chain *chain, *chain_prev;
2746         struct net *net = sock_net(skb->sk);
2747         struct nlattr *tca[TCA_MAX + 1];
2748         struct Qdisc *q = NULL;
2749         struct tcf_block *block;
2750         struct tcmsg *tcm = nlmsg_data(cb->nlh);
2751         bool terse_dump = false;
2752         long index_start;
2753         long index;
2754         u32 parent;
2755         int err;
2756
2757         if (nlmsg_len(cb->nlh) < sizeof(*tcm))
2758                 return skb->len;
2759
2760         err = nlmsg_parse_deprecated(cb->nlh, sizeof(*tcm), tca, TCA_MAX,
2761                                      tcf_tfilter_dump_policy, cb->extack);
2762         if (err)
2763                 return err;
2764
2765         if (tca[TCA_DUMP_FLAGS]) {
2766                 struct nla_bitfield32 flags =
2767                         nla_get_bitfield32(tca[TCA_DUMP_FLAGS]);
2768
2769                 terse_dump = flags.value & TCA_DUMP_FLAGS_TERSE;
2770         }
2771
2772         if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK) {
2773                 block = tcf_block_refcnt_get(net, tcm->tcm_block_index);
2774                 if (!block)
2775                         goto out;
2776                 /* If we work with block index, q is NULL and parent value
2777                  * will never be used in the following code. The check
2778                  * in tcf_fill_node prevents it. However, compiler does not
2779                  * see that far, so set parent to zero to silence the warning
2780                  * about parent being uninitialized.
2781                  */
2782                 parent = 0;
2783         } else {
2784                 const struct Qdisc_class_ops *cops;
2785                 struct net_device *dev;
2786                 unsigned long cl = 0;
2787
2788                 dev = __dev_get_by_index(net, tcm->tcm_ifindex);
2789                 if (!dev)
2790                         return skb->len;
2791
2792                 parent = tcm->tcm_parent;
2793                 if (!parent)
2794                         q = rtnl_dereference(dev->qdisc);
2795                 else
2796                         q = qdisc_lookup(dev, TC_H_MAJ(tcm->tcm_parent));
2797                 if (!q)
2798                         goto out;
2799                 cops = q->ops->cl_ops;
2800                 if (!cops)
2801                         goto out;
2802                 if (!cops->tcf_block)
2803                         goto out;
2804                 if (TC_H_MIN(tcm->tcm_parent)) {
2805                         cl = cops->find(q, tcm->tcm_parent);
2806                         if (cl == 0)
2807                                 goto out;
2808                 }
2809                 block = cops->tcf_block(q, cl, NULL);
2810                 if (!block)
2811                         goto out;
2812                 parent = block->classid;
2813                 if (tcf_block_shared(block))
2814                         q = NULL;
2815         }
2816
2817         index_start = cb->args[0];
2818         index = 0;
2819
2820         for (chain = __tcf_get_next_chain(block, NULL);
2821              chain;
2822              chain_prev = chain,
2823                      chain = __tcf_get_next_chain(block, chain),
2824                      tcf_chain_put(chain_prev)) {
2825                 if (tca[TCA_CHAIN] &&
2826                     nla_get_u32(tca[TCA_CHAIN]) != chain->index)
2827                         continue;
2828                 if (!tcf_chain_dump(chain, q, parent, skb, cb,
2829                                     index_start, &index, terse_dump)) {
2830                         tcf_chain_put(chain);
2831                         err = -EMSGSIZE;
2832                         break;
2833                 }
2834         }
2835
2836         if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK)
2837                 tcf_block_refcnt_put(block, true);
2838         cb->args[0] = index;
2839
2840 out:
2841         /* If we did no progress, the error (EMSGSIZE) is real */
2842         if (skb->len == 0 && err)
2843                 return err;
2844         return skb->len;
2845 }
2846
2847 static int tc_chain_fill_node(const struct tcf_proto_ops *tmplt_ops,
2848                               void *tmplt_priv, u32 chain_index,
2849                               struct net *net, struct sk_buff *skb,
2850                               struct tcf_block *block,
2851                               u32 portid, u32 seq, u16 flags, int event,
2852                               struct netlink_ext_ack *extack)
2853 {
2854         unsigned char *b = skb_tail_pointer(skb);
2855         const struct tcf_proto_ops *ops;
2856         struct nlmsghdr *nlh;
2857         struct tcmsg *tcm;
2858         void *priv;
2859
2860         ops = tmplt_ops;
2861         priv = tmplt_priv;
2862
2863         nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
2864         if (!nlh)
2865                 goto out_nlmsg_trim;
2866         tcm = nlmsg_data(nlh);
2867         tcm->tcm_family = AF_UNSPEC;
2868         tcm->tcm__pad1 = 0;
2869         tcm->tcm__pad2 = 0;
2870         tcm->tcm_handle = 0;
2871         if (block->q) {
2872                 tcm->tcm_ifindex = qdisc_dev(block->q)->ifindex;
2873                 tcm->tcm_parent = block->q->handle;
2874         } else {
2875                 tcm->tcm_ifindex = TCM_IFINDEX_MAGIC_BLOCK;
2876                 tcm->tcm_block_index = block->index;
2877         }
2878
2879         if (nla_put_u32(skb, TCA_CHAIN, chain_index))
2880                 goto nla_put_failure;
2881
2882         if (ops) {
2883                 if (nla_put_string(skb, TCA_KIND, ops->kind))
2884                         goto nla_put_failure;
2885                 if (ops->tmplt_dump(skb, net, priv) < 0)
2886                         goto nla_put_failure;
2887         }
2888
2889         if (extack && extack->_msg &&
2890             nla_put_string(skb, TCA_EXT_WARN_MSG, extack->_msg))
2891                 goto out_nlmsg_trim;
2892
2893         nlh->nlmsg_len = skb_tail_pointer(skb) - b;
2894
2895         return skb->len;
2896
2897 out_nlmsg_trim:
2898 nla_put_failure:
2899         nlmsg_trim(skb, b);
2900         return -EMSGSIZE;
2901 }
2902
2903 static int tc_chain_notify(struct tcf_chain *chain, struct sk_buff *oskb,
2904                            u32 seq, u16 flags, int event, bool unicast,
2905                            struct netlink_ext_ack *extack)
2906 {
2907         u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
2908         struct tcf_block *block = chain->block;
2909         struct net *net = block->net;
2910         struct sk_buff *skb;
2911         int err = 0;
2912
2913         if (!unicast && !rtnl_notify_needed(net, flags, RTNLGRP_TC))
2914                 return 0;
2915
2916         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2917         if (!skb)
2918                 return -ENOBUFS;
2919
2920         if (tc_chain_fill_node(chain->tmplt_ops, chain->tmplt_priv,
2921                                chain->index, net, skb, block, portid,
2922                                seq, flags, event, extack) <= 0) {
2923                 kfree_skb(skb);
2924                 return -EINVAL;
2925         }
2926
2927         if (unicast)
2928                 err = rtnl_unicast(skb, net, portid);
2929         else
2930                 err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
2931                                      flags & NLM_F_ECHO);
2932
2933         return err;
2934 }
2935
2936 static int tc_chain_notify_delete(const struct tcf_proto_ops *tmplt_ops,
2937                                   void *tmplt_priv, u32 chain_index,
2938                                   struct tcf_block *block, struct sk_buff *oskb,
2939                                   u32 seq, u16 flags)
2940 {
2941         u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
2942         struct net *net = block->net;
2943         struct sk_buff *skb;
2944
2945         if (!rtnl_notify_needed(net, flags, RTNLGRP_TC))
2946                 return 0;
2947
2948         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2949         if (!skb)
2950                 return -ENOBUFS;
2951
2952         if (tc_chain_fill_node(tmplt_ops, tmplt_priv, chain_index, net, skb,
2953                                block, portid, seq, flags, RTM_DELCHAIN, NULL) <= 0) {
2954                 kfree_skb(skb);
2955                 return -EINVAL;
2956         }
2957
2958         return rtnetlink_send(skb, net, portid, RTNLGRP_TC, flags & NLM_F_ECHO);
2959 }
2960
2961 static int tc_chain_tmplt_add(struct tcf_chain *chain, struct net *net,
2962                               struct nlattr **tca,
2963                               struct netlink_ext_ack *extack)
2964 {
2965         const struct tcf_proto_ops *ops;
2966         char name[IFNAMSIZ];
2967         void *tmplt_priv;
2968
2969         /* If kind is not set, user did not specify template. */
2970         if (!tca[TCA_KIND])
2971                 return 0;
2972
2973         if (tcf_proto_check_kind(tca[TCA_KIND], name)) {
2974                 NL_SET_ERR_MSG(extack, "Specified TC chain template name too long");
2975                 return -EINVAL;
2976         }
2977
2978         ops = tcf_proto_lookup_ops(name, true, extack);
2979         if (IS_ERR(ops))
2980                 return PTR_ERR(ops);
2981         if (!ops->tmplt_create || !ops->tmplt_destroy || !ops->tmplt_dump) {
2982                 NL_SET_ERR_MSG(extack, "Chain templates are not supported with specified classifier");
2983                 module_put(ops->owner);
2984                 return -EOPNOTSUPP;
2985         }
2986
2987         tmplt_priv = ops->tmplt_create(net, chain, tca, extack);
2988         if (IS_ERR(tmplt_priv)) {
2989                 module_put(ops->owner);
2990                 return PTR_ERR(tmplt_priv);
2991         }
2992         chain->tmplt_ops = ops;
2993         chain->tmplt_priv = tmplt_priv;
2994         return 0;
2995 }
2996
2997 static void tc_chain_tmplt_del(const struct tcf_proto_ops *tmplt_ops,
2998                                void *tmplt_priv)
2999 {
3000         /* If template ops are set, no work to do for us. */
3001         if (!tmplt_ops)
3002                 return;
3003
3004         tmplt_ops->tmplt_destroy(tmplt_priv);
3005         module_put(tmplt_ops->owner);
3006 }
3007
3008 /* Add/delete/get a chain */
3009
3010 static int tc_ctl_chain(struct sk_buff *skb, struct nlmsghdr *n,
3011                         struct netlink_ext_ack *extack)
3012 {
3013         struct net *net = sock_net(skb->sk);
3014         struct nlattr *tca[TCA_MAX + 1];
3015         struct tcmsg *t;
3016         u32 parent;
3017         u32 chain_index;
3018         struct Qdisc *q;
3019         struct tcf_chain *chain;
3020         struct tcf_block *block;
3021         unsigned long cl;
3022         int err;
3023
3024 replay:
3025         q = NULL;
3026         err = nlmsg_parse_deprecated(n, sizeof(*t), tca, TCA_MAX,
3027                                      rtm_tca_policy, extack);
3028         if (err < 0)
3029                 return err;
3030
3031         t = nlmsg_data(n);
3032         parent = t->tcm_parent;
3033         cl = 0;
3034
3035         block = tcf_block_find(net, &q, &parent, &cl,
3036                                t->tcm_ifindex, t->tcm_block_index, extack);
3037         if (IS_ERR(block))
3038                 return PTR_ERR(block);
3039
3040         chain_index = tca[TCA_CHAIN] ? nla_get_u32(tca[TCA_CHAIN]) : 0;
3041         if (chain_index > TC_ACT_EXT_VAL_MASK) {
3042                 NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
3043                 err = -EINVAL;
3044                 goto errout_block;
3045         }
3046
3047         mutex_lock(&block->lock);
3048         chain = tcf_chain_lookup(block, chain_index);
3049         if (n->nlmsg_type == RTM_NEWCHAIN) {
3050                 if (chain) {
3051                         if (tcf_chain_held_by_acts_only(chain)) {
3052                                 /* The chain exists only because there is
3053                                  * some action referencing it.
3054                                  */
3055                                 tcf_chain_hold(chain);
3056                         } else {
3057                                 NL_SET_ERR_MSG(extack, "Filter chain already exists");
3058                                 err = -EEXIST;
3059                                 goto errout_block_locked;
3060                         }
3061                 } else {
3062                         if (!(n->nlmsg_flags & NLM_F_CREATE)) {
3063                                 NL_SET_ERR_MSG(extack, "Need both RTM_NEWCHAIN and NLM_F_CREATE to create a new chain");
3064                                 err = -ENOENT;
3065                                 goto errout_block_locked;
3066                         }
3067                         chain = tcf_chain_create(block, chain_index);
3068                         if (!chain) {
3069                                 NL_SET_ERR_MSG(extack, "Failed to create filter chain");
3070                                 err = -ENOMEM;
3071                                 goto errout_block_locked;
3072                         }
3073                 }
3074         } else {
3075                 if (!chain || tcf_chain_held_by_acts_only(chain)) {
3076                         NL_SET_ERR_MSG(extack, "Cannot find specified filter chain");
3077                         err = -EINVAL;
3078                         goto errout_block_locked;
3079                 }
3080                 tcf_chain_hold(chain);
3081         }
3082
3083         if (n->nlmsg_type == RTM_NEWCHAIN) {
3084                 /* Modifying chain requires holding parent block lock. In case
3085                  * the chain was successfully added, take a reference to the
3086                  * chain. This ensures that an empty chain does not disappear at
3087                  * the end of this function.
3088                  */
3089                 tcf_chain_hold(chain);
3090                 chain->explicitly_created = true;
3091         }
3092         mutex_unlock(&block->lock);
3093
3094         switch (n->nlmsg_type) {
3095         case RTM_NEWCHAIN:
3096                 err = tc_chain_tmplt_add(chain, net, tca, extack);
3097                 if (err) {
3098                         tcf_chain_put_explicitly_created(chain);
3099                         goto errout;
3100                 }
3101
3102                 tc_chain_notify(chain, NULL, 0, NLM_F_CREATE | NLM_F_EXCL,
3103                                 RTM_NEWCHAIN, false, extack);
3104                 break;
3105         case RTM_DELCHAIN:
3106                 tfilter_notify_chain(net, skb, block, q, parent, n,
3107                                      chain, RTM_DELTFILTER, extack);
3108                 /* Flush the chain first as the user requested chain removal. */
3109                 tcf_chain_flush(chain, true);
3110                 /* In case the chain was successfully deleted, put a reference
3111                  * to the chain previously taken during addition.
3112                  */
3113                 tcf_chain_put_explicitly_created(chain);
3114                 break;
3115         case RTM_GETCHAIN:
3116                 err = tc_chain_notify(chain, skb, n->nlmsg_seq,
3117                                       n->nlmsg_flags, n->nlmsg_type, true, extack);
3118                 if (err < 0)
3119                         NL_SET_ERR_MSG(extack, "Failed to send chain notify message");
3120                 break;
3121         default:
3122                 err = -EOPNOTSUPP;
3123                 NL_SET_ERR_MSG(extack, "Unsupported message type");
3124                 goto errout;
3125         }
3126
3127 errout:
3128         tcf_chain_put(chain);
3129 errout_block:
3130         tcf_block_release(q, block, true);
3131         if (err == -EAGAIN)
3132                 /* Replay the request. */
3133                 goto replay;
3134         return err;
3135
3136 errout_block_locked:
3137         mutex_unlock(&block->lock);
3138         goto errout_block;
3139 }
3140
3141 /* called with RTNL */
3142 static int tc_dump_chain(struct sk_buff *skb, struct netlink_callback *cb)
3143 {
3144         struct net *net = sock_net(skb->sk);
3145         struct nlattr *tca[TCA_MAX + 1];
3146         struct Qdisc *q = NULL;
3147         struct tcf_block *block;
3148         struct tcmsg *tcm = nlmsg_data(cb->nlh);
3149         struct tcf_chain *chain;
3150         long index_start;
3151         long index;
3152         int err;
3153
3154         if (nlmsg_len(cb->nlh) < sizeof(*tcm))
3155                 return skb->len;
3156
3157         err = nlmsg_parse_deprecated(cb->nlh, sizeof(*tcm), tca, TCA_MAX,
3158                                      rtm_tca_policy, cb->extack);
3159         if (err)
3160                 return err;
3161
3162         if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK) {
3163                 block = tcf_block_refcnt_get(net, tcm->tcm_block_index);
3164                 if (!block)
3165                         goto out;
3166         } else {
3167                 const struct Qdisc_class_ops *cops;
3168                 struct net_device *dev;
3169                 unsigned long cl = 0;
3170
3171                 dev = __dev_get_by_index(net, tcm->tcm_ifindex);
3172                 if (!dev)
3173                         return skb->len;
3174
3175                 if (!tcm->tcm_parent)
3176                         q = rtnl_dereference(dev->qdisc);
3177                 else
3178                         q = qdisc_lookup(dev, TC_H_MAJ(tcm->tcm_parent));
3179
3180                 if (!q)
3181                         goto out;
3182                 cops = q->ops->cl_ops;
3183                 if (!cops)
3184                         goto out;
3185                 if (!cops->tcf_block)
3186                         goto out;
3187                 if (TC_H_MIN(tcm->tcm_parent)) {
3188                         cl = cops->find(q, tcm->tcm_parent);
3189                         if (cl == 0)
3190                                 goto out;
3191                 }
3192                 block = cops->tcf_block(q, cl, NULL);
3193                 if (!block)
3194                         goto out;
3195                 if (tcf_block_shared(block))
3196                         q = NULL;
3197         }
3198
3199         index_start = cb->args[0];
3200         index = 0;
3201
3202         mutex_lock(&block->lock);
3203         list_for_each_entry(chain, &block->chain_list, list) {
3204                 if ((tca[TCA_CHAIN] &&
3205                      nla_get_u32(tca[TCA_CHAIN]) != chain->index))
3206                         continue;
3207                 if (index < index_start) {
3208                         index++;
3209                         continue;
3210                 }
3211                 if (tcf_chain_held_by_acts_only(chain))
3212                         continue;
3213                 err = tc_chain_fill_node(chain->tmplt_ops, chain->tmplt_priv,
3214                                          chain->index, net, skb, block,
3215                                          NETLINK_CB(cb->skb).portid,
3216                                          cb->nlh->nlmsg_seq, NLM_F_MULTI,
3217                                          RTM_NEWCHAIN, NULL);
3218                 if (err <= 0)
3219                         break;
3220                 index++;
3221         }
3222         mutex_unlock(&block->lock);
3223
3224         if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK)
3225                 tcf_block_refcnt_put(block, true);
3226         cb->args[0] = index;
3227
3228 out:
3229         /* If we did no progress, the error (EMSGSIZE) is real */
3230         if (skb->len == 0 && err)
3231                 return err;
3232         return skb->len;
3233 }
3234
3235 int tcf_exts_init_ex(struct tcf_exts *exts, struct net *net, int action,
3236                      int police, struct tcf_proto *tp, u32 handle,
3237                      bool use_action_miss)
3238 {
3239         int err = 0;
3240
3241 #ifdef CONFIG_NET_CLS_ACT
3242         exts->type = 0;
3243         exts->nr_actions = 0;
3244         exts->miss_cookie_node = NULL;
3245         /* Note: we do not own yet a reference on net.
3246          * This reference might be taken later from tcf_exts_get_net().
3247          */
3248         exts->net = net;
3249         exts->actions = kcalloc(TCA_ACT_MAX_PRIO, sizeof(struct tc_action *),
3250                                 GFP_KERNEL);
3251         if (!exts->actions)
3252                 return -ENOMEM;
3253 #endif
3254
3255         exts->action = action;
3256         exts->police = police;
3257
3258         if (!use_action_miss)
3259                 return 0;
3260
3261         err = tcf_exts_miss_cookie_base_alloc(exts, tp, handle);
3262         if (err)
3263                 goto err_miss_alloc;
3264
3265         return 0;
3266
3267 err_miss_alloc:
3268         tcf_exts_destroy(exts);
3269 #ifdef CONFIG_NET_CLS_ACT
3270         exts->actions = NULL;
3271 #endif
3272         return err;
3273 }
3274 EXPORT_SYMBOL(tcf_exts_init_ex);
3275
3276 void tcf_exts_destroy(struct tcf_exts *exts)
3277 {
3278         tcf_exts_miss_cookie_base_destroy(exts);
3279
3280 #ifdef CONFIG_NET_CLS_ACT
3281         if (exts->actions) {
3282                 tcf_action_destroy(exts->actions, TCA_ACT_UNBIND);
3283                 kfree(exts->actions);
3284         }
3285         exts->nr_actions = 0;
3286 #endif
3287 }
3288 EXPORT_SYMBOL(tcf_exts_destroy);
3289
3290 int tcf_exts_validate_ex(struct net *net, struct tcf_proto *tp, struct nlattr **tb,
3291                          struct nlattr *rate_tlv, struct tcf_exts *exts,
3292                          u32 flags, u32 fl_flags, struct netlink_ext_ack *extack)
3293 {
3294 #ifdef CONFIG_NET_CLS_ACT
3295         {
3296                 int init_res[TCA_ACT_MAX_PRIO] = {};
3297                 struct tc_action *act;
3298                 size_t attr_size = 0;
3299
3300                 if (exts->police && tb[exts->police]) {
3301                         struct tc_action_ops *a_o;
3302
3303                         a_o = tc_action_load_ops(tb[exts->police], true,
3304                                                  !(flags & TCA_ACT_FLAGS_NO_RTNL),
3305                                                  extack);
3306                         if (IS_ERR(a_o))
3307                                 return PTR_ERR(a_o);
3308                         flags |= TCA_ACT_FLAGS_POLICE | TCA_ACT_FLAGS_BIND;
3309                         act = tcf_action_init_1(net, tp, tb[exts->police],
3310                                                 rate_tlv, a_o, init_res, flags,
3311                                                 extack);
3312                         module_put(a_o->owner);
3313                         if (IS_ERR(act))
3314                                 return PTR_ERR(act);
3315
3316                         act->type = exts->type = TCA_OLD_COMPAT;
3317                         exts->actions[0] = act;
3318                         exts->nr_actions = 1;
3319                         tcf_idr_insert_many(exts->actions, init_res);
3320                 } else if (exts->action && tb[exts->action]) {
3321                         int err;
3322
3323                         flags |= TCA_ACT_FLAGS_BIND;
3324                         err = tcf_action_init(net, tp, tb[exts->action],
3325                                               rate_tlv, exts->actions, init_res,
3326                                               &attr_size, flags, fl_flags,
3327                                               extack);
3328                         if (err < 0)
3329                                 return err;
3330                         exts->nr_actions = err;
3331                 }
3332         }
3333 #else
3334         if ((exts->action && tb[exts->action]) ||
3335             (exts->police && tb[exts->police])) {
3336                 NL_SET_ERR_MSG(extack, "Classifier actions are not supported per compile options (CONFIG_NET_CLS_ACT)");
3337                 return -EOPNOTSUPP;
3338         }
3339 #endif
3340
3341         return 0;
3342 }
3343 EXPORT_SYMBOL(tcf_exts_validate_ex);
3344
3345 int tcf_exts_validate(struct net *net, struct tcf_proto *tp, struct nlattr **tb,
3346                       struct nlattr *rate_tlv, struct tcf_exts *exts,
3347                       u32 flags, struct netlink_ext_ack *extack)
3348 {
3349         return tcf_exts_validate_ex(net, tp, tb, rate_tlv, exts,
3350                                     flags, 0, extack);
3351 }
3352 EXPORT_SYMBOL(tcf_exts_validate);
3353
3354 void tcf_exts_change(struct tcf_exts *dst, struct tcf_exts *src)
3355 {
3356 #ifdef CONFIG_NET_CLS_ACT
3357         struct tcf_exts old = *dst;
3358
3359         *dst = *src;
3360         tcf_exts_destroy(&old);
3361 #endif
3362 }
3363 EXPORT_SYMBOL(tcf_exts_change);
3364
3365 #ifdef CONFIG_NET_CLS_ACT
3366 static struct tc_action *tcf_exts_first_act(struct tcf_exts *exts)
3367 {
3368         if (exts->nr_actions == 0)
3369                 return NULL;
3370         else
3371                 return exts->actions[0];
3372 }
3373 #endif
3374
3375 int tcf_exts_dump(struct sk_buff *skb, struct tcf_exts *exts)
3376 {
3377 #ifdef CONFIG_NET_CLS_ACT
3378         struct nlattr *nest;
3379
3380         if (exts->action && tcf_exts_has_actions(exts)) {
3381                 /*
3382                  * again for backward compatible mode - we want
3383                  * to work with both old and new modes of entering
3384                  * tc data even if iproute2  was newer - jhs
3385                  */
3386                 if (exts->type != TCA_OLD_COMPAT) {
3387                         nest = nla_nest_start_noflag(skb, exts->action);
3388                         if (nest == NULL)
3389                                 goto nla_put_failure;
3390
3391                         if (tcf_action_dump(skb, exts->actions, 0, 0, false)
3392                             < 0)
3393                                 goto nla_put_failure;
3394                         nla_nest_end(skb, nest);
3395                 } else if (exts->police) {
3396                         struct tc_action *act = tcf_exts_first_act(exts);
3397                         nest = nla_nest_start_noflag(skb, exts->police);
3398                         if (nest == NULL || !act)
3399                                 goto nla_put_failure;
3400                         if (tcf_action_dump_old(skb, act, 0, 0) < 0)
3401                                 goto nla_put_failure;
3402                         nla_nest_end(skb, nest);
3403                 }
3404         }
3405         return 0;
3406
3407 nla_put_failure:
3408         nla_nest_cancel(skb, nest);
3409         return -1;
3410 #else
3411         return 0;
3412 #endif
3413 }
3414 EXPORT_SYMBOL(tcf_exts_dump);
3415
3416 int tcf_exts_terse_dump(struct sk_buff *skb, struct tcf_exts *exts)
3417 {
3418 #ifdef CONFIG_NET_CLS_ACT
3419         struct nlattr *nest;
3420
3421         if (!exts->action || !tcf_exts_has_actions(exts))
3422                 return 0;
3423
3424         nest = nla_nest_start_noflag(skb, exts->action);
3425         if (!nest)
3426                 goto nla_put_failure;
3427
3428         if (tcf_action_dump(skb, exts->actions, 0, 0, true) < 0)
3429                 goto nla_put_failure;
3430         nla_nest_end(skb, nest);
3431         return 0;
3432
3433 nla_put_failure:
3434         nla_nest_cancel(skb, nest);
3435         return -1;
3436 #else
3437         return 0;
3438 #endif
3439 }
3440 EXPORT_SYMBOL(tcf_exts_terse_dump);
3441
3442 int tcf_exts_dump_stats(struct sk_buff *skb, struct tcf_exts *exts)
3443 {
3444 #ifdef CONFIG_NET_CLS_ACT
3445         struct tc_action *a = tcf_exts_first_act(exts);
3446         if (a != NULL && tcf_action_copy_stats(skb, a, 1) < 0)
3447                 return -1;
3448 #endif
3449         return 0;
3450 }
3451 EXPORT_SYMBOL(tcf_exts_dump_stats);
3452
3453 static void tcf_block_offload_inc(struct tcf_block *block, u32 *flags)
3454 {
3455         if (*flags & TCA_CLS_FLAGS_IN_HW)
3456                 return;
3457         *flags |= TCA_CLS_FLAGS_IN_HW;
3458         atomic_inc(&block->offloadcnt);
3459 }
3460
3461 static void tcf_block_offload_dec(struct tcf_block *block, u32 *flags)
3462 {
3463         if (!(*flags & TCA_CLS_FLAGS_IN_HW))
3464                 return;
3465         *flags &= ~TCA_CLS_FLAGS_IN_HW;
3466         atomic_dec(&block->offloadcnt);
3467 }
3468
3469 static void tc_cls_offload_cnt_update(struct tcf_block *block,
3470                                       struct tcf_proto *tp, u32 *cnt,
3471                                       u32 *flags, u32 diff, bool add)
3472 {
3473         lockdep_assert_held(&block->cb_lock);
3474
3475         spin_lock(&tp->lock);
3476         if (add) {
3477                 if (!*cnt)
3478                         tcf_block_offload_inc(block, flags);
3479                 *cnt += diff;
3480         } else {
3481                 *cnt -= diff;
3482                 if (!*cnt)
3483                         tcf_block_offload_dec(block, flags);
3484         }
3485         spin_unlock(&tp->lock);
3486 }
3487
3488 static void
3489 tc_cls_offload_cnt_reset(struct tcf_block *block, struct tcf_proto *tp,
3490                          u32 *cnt, u32 *flags)
3491 {
3492         lockdep_assert_held(&block->cb_lock);
3493
3494         spin_lock(&tp->lock);
3495         tcf_block_offload_dec(block, flags);
3496         *cnt = 0;
3497         spin_unlock(&tp->lock);
3498 }
3499
3500 static int
3501 __tc_setup_cb_call(struct tcf_block *block, enum tc_setup_type type,
3502                    void *type_data, bool err_stop)
3503 {
3504         struct flow_block_cb *block_cb;
3505         int ok_count = 0;
3506         int err;
3507
3508         list_for_each_entry(block_cb, &block->flow_block.cb_list, list) {
3509                 err = block_cb->cb(type, type_data, block_cb->cb_priv);
3510                 if (err) {
3511                         if (err_stop)
3512                                 return err;
3513                 } else {
3514                         ok_count++;
3515                 }
3516         }
3517         return ok_count;
3518 }
3519
3520 int tc_setup_cb_call(struct tcf_block *block, enum tc_setup_type type,
3521                      void *type_data, bool err_stop, bool rtnl_held)
3522 {
3523         bool take_rtnl = READ_ONCE(block->lockeddevcnt) && !rtnl_held;
3524         int ok_count;
3525
3526 retry:
3527         if (take_rtnl)
3528                 rtnl_lock();
3529         down_read(&block->cb_lock);
3530         /* Need to obtain rtnl lock if block is bound to devs that require it.
3531          * In block bind code cb_lock is obtained while holding rtnl, so we must
3532          * obtain the locks in same order here.
3533          */
3534         if (!rtnl_held && !take_rtnl && block->lockeddevcnt) {
3535                 up_read(&block->cb_lock);
3536                 take_rtnl = true;
3537                 goto retry;
3538         }
3539
3540         ok_count = __tc_setup_cb_call(block, type, type_data, err_stop);
3541
3542         up_read(&block->cb_lock);
3543         if (take_rtnl)
3544                 rtnl_unlock();
3545         return ok_count;
3546 }
3547 EXPORT_SYMBOL(tc_setup_cb_call);
3548
3549 /* Non-destructive filter add. If filter that wasn't already in hardware is
3550  * successfully offloaded, increment block offloads counter. On failure,
3551  * previously offloaded filter is considered to be intact and offloads counter
3552  * is not decremented.
3553  */
3554
3555 int tc_setup_cb_add(struct tcf_block *block, struct tcf_proto *tp,
3556                     enum tc_setup_type type, void *type_data, bool err_stop,
3557                     u32 *flags, unsigned int *in_hw_count, bool rtnl_held)
3558 {
3559         bool take_rtnl = READ_ONCE(block->lockeddevcnt) && !rtnl_held;
3560         int ok_count;
3561
3562 retry:
3563         if (take_rtnl)
3564                 rtnl_lock();
3565         down_read(&block->cb_lock);
3566         /* Need to obtain rtnl lock if block is bound to devs that require it.
3567          * In block bind code cb_lock is obtained while holding rtnl, so we must
3568          * obtain the locks in same order here.
3569          */
3570         if (!rtnl_held && !take_rtnl && block->lockeddevcnt) {
3571                 up_read(&block->cb_lock);
3572                 take_rtnl = true;
3573                 goto retry;
3574         }
3575
3576         /* Make sure all netdevs sharing this block are offload-capable. */
3577         if (block->nooffloaddevcnt && err_stop) {
3578                 ok_count = -EOPNOTSUPP;
3579                 goto err_unlock;
3580         }
3581
3582         ok_count = __tc_setup_cb_call(block, type, type_data, err_stop);
3583         if (ok_count < 0)
3584                 goto err_unlock;
3585
3586         if (tp->ops->hw_add)
3587                 tp->ops->hw_add(tp, type_data);
3588         if (ok_count > 0)
3589                 tc_cls_offload_cnt_update(block, tp, in_hw_count, flags,
3590                                           ok_count, true);
3591 err_unlock:
3592         up_read(&block->cb_lock);
3593         if (take_rtnl)
3594                 rtnl_unlock();
3595         return min(ok_count, 0);
3596 }
3597 EXPORT_SYMBOL(tc_setup_cb_add);
3598
3599 /* Destructive filter replace. If filter that wasn't already in hardware is
3600  * successfully offloaded, increment block offload counter. On failure,
3601  * previously offloaded filter is considered to be destroyed and offload counter
3602  * is decremented.
3603  */
3604
3605 int tc_setup_cb_replace(struct tcf_block *block, struct tcf_proto *tp,
3606                         enum tc_setup_type type, void *type_data, bool err_stop,
3607                         u32 *old_flags, unsigned int *old_in_hw_count,
3608                         u32 *new_flags, unsigned int *new_in_hw_count,
3609                         bool rtnl_held)
3610 {
3611         bool take_rtnl = READ_ONCE(block->lockeddevcnt) && !rtnl_held;
3612         int ok_count;
3613
3614 retry:
3615         if (take_rtnl)
3616                 rtnl_lock();
3617         down_read(&block->cb_lock);
3618         /* Need to obtain rtnl lock if block is bound to devs that require it.
3619          * In block bind code cb_lock is obtained while holding rtnl, so we must
3620          * obtain the locks in same order here.
3621          */
3622         if (!rtnl_held && !take_rtnl && block->lockeddevcnt) {
3623                 up_read(&block->cb_lock);
3624                 take_rtnl = true;
3625                 goto retry;
3626         }
3627
3628         /* Make sure all netdevs sharing this block are offload-capable. */
3629         if (block->nooffloaddevcnt && err_stop) {
3630                 ok_count = -EOPNOTSUPP;
3631                 goto err_unlock;
3632         }
3633
3634         tc_cls_offload_cnt_reset(block, tp, old_in_hw_count, old_flags);
3635         if (tp->ops->hw_del)
3636                 tp->ops->hw_del(tp, type_data);
3637
3638         ok_count = __tc_setup_cb_call(block, type, type_data, err_stop);
3639         if (ok_count < 0)
3640                 goto err_unlock;
3641
3642         if (tp->ops->hw_add)
3643                 tp->ops->hw_add(tp, type_data);
3644         if (ok_count > 0)
3645                 tc_cls_offload_cnt_update(block, tp, new_in_hw_count,
3646                                           new_flags, ok_count, true);
3647 err_unlock:
3648         up_read(&block->cb_lock);
3649         if (take_rtnl)
3650                 rtnl_unlock();
3651         return min(ok_count, 0);
3652 }
3653 EXPORT_SYMBOL(tc_setup_cb_replace);
3654
3655 /* Destroy filter and decrement block offload counter, if filter was previously
3656  * offloaded.
3657  */
3658
3659 int tc_setup_cb_destroy(struct tcf_block *block, struct tcf_proto *tp,
3660                         enum tc_setup_type type, void *type_data, bool err_stop,
3661                         u32 *flags, unsigned int *in_hw_count, bool rtnl_held)
3662 {
3663         bool take_rtnl = READ_ONCE(block->lockeddevcnt) && !rtnl_held;
3664         int ok_count;
3665
3666 retry:
3667         if (take_rtnl)
3668                 rtnl_lock();
3669         down_read(&block->cb_lock);
3670         /* Need to obtain rtnl lock if block is bound to devs that require it.
3671          * In block bind code cb_lock is obtained while holding rtnl, so we must
3672          * obtain the locks in same order here.
3673          */
3674         if (!rtnl_held && !take_rtnl && block->lockeddevcnt) {
3675                 up_read(&block->cb_lock);
3676                 take_rtnl = true;
3677                 goto retry;
3678         }
3679
3680         ok_count = __tc_setup_cb_call(block, type, type_data, err_stop);
3681
3682         tc_cls_offload_cnt_reset(block, tp, in_hw_count, flags);
3683         if (tp->ops->hw_del)
3684                 tp->ops->hw_del(tp, type_data);
3685
3686         up_read(&block->cb_lock);
3687         if (take_rtnl)
3688                 rtnl_unlock();
3689         return min(ok_count, 0);
3690 }
3691 EXPORT_SYMBOL(tc_setup_cb_destroy);
3692
3693 int tc_setup_cb_reoffload(struct tcf_block *block, struct tcf_proto *tp,
3694                           bool add, flow_setup_cb_t *cb,
3695                           enum tc_setup_type type, void *type_data,
3696                           void *cb_priv, u32 *flags, unsigned int *in_hw_count)
3697 {
3698         int err = cb(type, type_data, cb_priv);
3699
3700         if (err) {
3701                 if (add && tc_skip_sw(*flags))
3702                         return err;
3703         } else {
3704                 tc_cls_offload_cnt_update(block, tp, in_hw_count, flags, 1,
3705                                           add);
3706         }
3707
3708         return 0;
3709 }
3710 EXPORT_SYMBOL(tc_setup_cb_reoffload);
3711
3712 static int tcf_act_get_user_cookie(struct flow_action_entry *entry,
3713                                    const struct tc_action *act)
3714 {
3715         struct tc_cookie *user_cookie;
3716         int err = 0;
3717
3718         rcu_read_lock();
3719         user_cookie = rcu_dereference(act->user_cookie);
3720         if (user_cookie) {
3721                 entry->user_cookie = flow_action_cookie_create(user_cookie->data,
3722                                                                user_cookie->len,
3723                                                                GFP_ATOMIC);
3724                 if (!entry->user_cookie)
3725                         err = -ENOMEM;
3726         }
3727         rcu_read_unlock();
3728         return err;
3729 }
3730
3731 static void tcf_act_put_user_cookie(struct flow_action_entry *entry)
3732 {
3733         flow_action_cookie_destroy(entry->user_cookie);
3734 }
3735
3736 void tc_cleanup_offload_action(struct flow_action *flow_action)
3737 {
3738         struct flow_action_entry *entry;
3739         int i;
3740
3741         flow_action_for_each(i, entry, flow_action) {
3742                 tcf_act_put_user_cookie(entry);
3743                 if (entry->destructor)
3744                         entry->destructor(entry->destructor_priv);
3745         }
3746 }
3747 EXPORT_SYMBOL(tc_cleanup_offload_action);
3748
3749 static int tc_setup_offload_act(struct tc_action *act,
3750                                 struct flow_action_entry *entry,
3751                                 u32 *index_inc,
3752                                 struct netlink_ext_ack *extack)
3753 {
3754 #ifdef CONFIG_NET_CLS_ACT
3755         if (act->ops->offload_act_setup) {
3756                 return act->ops->offload_act_setup(act, entry, index_inc, true,
3757                                                    extack);
3758         } else {
3759                 NL_SET_ERR_MSG(extack, "Action does not support offload");
3760                 return -EOPNOTSUPP;
3761         }
3762 #else
3763         return 0;
3764 #endif
3765 }
3766
3767 int tc_setup_action(struct flow_action *flow_action,
3768                     struct tc_action *actions[],
3769                     u32 miss_cookie_base,
3770                     struct netlink_ext_ack *extack)
3771 {
3772         int i, j, k, index, err = 0;
3773         struct tc_action *act;
3774
3775         BUILD_BUG_ON(TCA_ACT_HW_STATS_ANY != FLOW_ACTION_HW_STATS_ANY);
3776         BUILD_BUG_ON(TCA_ACT_HW_STATS_IMMEDIATE != FLOW_ACTION_HW_STATS_IMMEDIATE);
3777         BUILD_BUG_ON(TCA_ACT_HW_STATS_DELAYED != FLOW_ACTION_HW_STATS_DELAYED);
3778
3779         if (!actions)
3780                 return 0;
3781
3782         j = 0;
3783         tcf_act_for_each_action(i, act, actions) {
3784                 struct flow_action_entry *entry;
3785
3786                 entry = &flow_action->entries[j];
3787                 spin_lock_bh(&act->tcfa_lock);
3788                 err = tcf_act_get_user_cookie(entry, act);
3789                 if (err)
3790                         goto err_out_locked;
3791
3792                 index = 0;
3793                 err = tc_setup_offload_act(act, entry, &index, extack);
3794                 if (err)
3795                         goto err_out_locked;
3796
3797                 for (k = 0; k < index ; k++) {
3798                         entry[k].hw_stats = tc_act_hw_stats(act->hw_stats);
3799                         entry[k].hw_index = act->tcfa_index;
3800                         entry[k].cookie = (unsigned long)act;
3801                         entry[k].miss_cookie =
3802                                 tcf_exts_miss_cookie_get(miss_cookie_base, i);
3803                 }
3804
3805                 j += index;
3806
3807                 spin_unlock_bh(&act->tcfa_lock);
3808         }
3809
3810 err_out:
3811         if (err)
3812                 tc_cleanup_offload_action(flow_action);
3813
3814         return err;
3815 err_out_locked:
3816         spin_unlock_bh(&act->tcfa_lock);
3817         goto err_out;
3818 }
3819
3820 int tc_setup_offload_action(struct flow_action *flow_action,
3821                             const struct tcf_exts *exts,
3822                             struct netlink_ext_ack *extack)
3823 {
3824 #ifdef CONFIG_NET_CLS_ACT
3825         u32 miss_cookie_base;
3826
3827         if (!exts)
3828                 return 0;
3829
3830         miss_cookie_base = exts->miss_cookie_node ?
3831                            exts->miss_cookie_node->miss_cookie_base : 0;
3832         return tc_setup_action(flow_action, exts->actions, miss_cookie_base,
3833                                extack);
3834 #else
3835         return 0;
3836 #endif
3837 }
3838 EXPORT_SYMBOL(tc_setup_offload_action);
3839
3840 unsigned int tcf_exts_num_actions(struct tcf_exts *exts)
3841 {
3842         unsigned int num_acts = 0;
3843         struct tc_action *act;
3844         int i;
3845
3846         tcf_exts_for_each_action(i, act, exts) {
3847                 if (is_tcf_pedit(act))
3848                         num_acts += tcf_pedit_nkeys(act);
3849                 else
3850                         num_acts++;
3851         }
3852         return num_acts;
3853 }
3854 EXPORT_SYMBOL(tcf_exts_num_actions);
3855
3856 #ifdef CONFIG_NET_CLS_ACT
3857 static int tcf_qevent_parse_block_index(struct nlattr *block_index_attr,
3858                                         u32 *p_block_index,
3859                                         struct netlink_ext_ack *extack)
3860 {
3861         *p_block_index = nla_get_u32(block_index_attr);
3862         if (!*p_block_index) {
3863                 NL_SET_ERR_MSG(extack, "Block number may not be zero");
3864                 return -EINVAL;
3865         }
3866
3867         return 0;
3868 }
3869
3870 int tcf_qevent_init(struct tcf_qevent *qe, struct Qdisc *sch,
3871                     enum flow_block_binder_type binder_type,
3872                     struct nlattr *block_index_attr,
3873                     struct netlink_ext_ack *extack)
3874 {
3875         u32 block_index;
3876         int err;
3877
3878         if (!block_index_attr)
3879                 return 0;
3880
3881         err = tcf_qevent_parse_block_index(block_index_attr, &block_index, extack);
3882         if (err)
3883                 return err;
3884
3885         qe->info.binder_type = binder_type;
3886         qe->info.chain_head_change = tcf_chain_head_change_dflt;
3887         qe->info.chain_head_change_priv = &qe->filter_chain;
3888         qe->info.block_index = block_index;
3889
3890         return tcf_block_get_ext(&qe->block, sch, &qe->info, extack);
3891 }
3892 EXPORT_SYMBOL(tcf_qevent_init);
3893
3894 void tcf_qevent_destroy(struct tcf_qevent *qe, struct Qdisc *sch)
3895 {
3896         if (qe->info.block_index)
3897                 tcf_block_put_ext(qe->block, sch, &qe->info);
3898 }
3899 EXPORT_SYMBOL(tcf_qevent_destroy);
3900
3901 int tcf_qevent_validate_change(struct tcf_qevent *qe, struct nlattr *block_index_attr,
3902                                struct netlink_ext_ack *extack)
3903 {
3904         u32 block_index;
3905         int err;
3906
3907         if (!block_index_attr)
3908                 return 0;
3909
3910         err = tcf_qevent_parse_block_index(block_index_attr, &block_index, extack);
3911         if (err)
3912                 return err;
3913
3914         /* Bounce newly-configured block or change in block. */
3915         if (block_index != qe->info.block_index) {
3916                 NL_SET_ERR_MSG(extack, "Change of blocks is not supported");
3917                 return -EINVAL;
3918         }
3919
3920         return 0;
3921 }
3922 EXPORT_SYMBOL(tcf_qevent_validate_change);
3923
3924 struct sk_buff *tcf_qevent_handle(struct tcf_qevent *qe, struct Qdisc *sch, struct sk_buff *skb,
3925                                   struct sk_buff **to_free, int *ret)
3926 {
3927         struct tcf_result cl_res;
3928         struct tcf_proto *fl;
3929
3930         if (!qe->info.block_index)
3931                 return skb;
3932
3933         fl = rcu_dereference_bh(qe->filter_chain);
3934
3935         switch (tcf_classify(skb, NULL, fl, &cl_res, false)) {
3936         case TC_ACT_SHOT:
3937                 qdisc_qstats_drop(sch);
3938                 __qdisc_drop(skb, to_free);
3939                 *ret = __NET_XMIT_BYPASS;
3940                 return NULL;
3941         case TC_ACT_STOLEN:
3942         case TC_ACT_QUEUED:
3943         case TC_ACT_TRAP:
3944                 __qdisc_drop(skb, to_free);
3945                 *ret = __NET_XMIT_STOLEN;
3946                 return NULL;
3947         case TC_ACT_REDIRECT:
3948                 skb_do_redirect(skb);
3949                 *ret = __NET_XMIT_STOLEN;
3950                 return NULL;
3951         }
3952
3953         return skb;
3954 }
3955 EXPORT_SYMBOL(tcf_qevent_handle);
3956
3957 int tcf_qevent_dump(struct sk_buff *skb, int attr_name, struct tcf_qevent *qe)
3958 {
3959         if (!qe->info.block_index)
3960                 return 0;
3961         return nla_put_u32(skb, attr_name, qe->info.block_index);
3962 }
3963 EXPORT_SYMBOL(tcf_qevent_dump);
3964 #endif
3965
3966 static __net_init int tcf_net_init(struct net *net)
3967 {
3968         struct tcf_net *tn = net_generic(net, tcf_net_id);
3969
3970         spin_lock_init(&tn->idr_lock);
3971         idr_init(&tn->idr);
3972         return 0;
3973 }
3974
3975 static void __net_exit tcf_net_exit(struct net *net)
3976 {
3977         struct tcf_net *tn = net_generic(net, tcf_net_id);
3978
3979         idr_destroy(&tn->idr);
3980 }
3981
3982 static struct pernet_operations tcf_net_ops = {
3983         .init = tcf_net_init,
3984         .exit = tcf_net_exit,
3985         .id   = &tcf_net_id,
3986         .size = sizeof(struct tcf_net),
3987 };
3988
3989 static int __init tc_filter_init(void)
3990 {
3991         int err;
3992
3993         tc_filter_wq = alloc_ordered_workqueue("tc_filter_workqueue", 0);
3994         if (!tc_filter_wq)
3995                 return -ENOMEM;
3996
3997         err = register_pernet_subsys(&tcf_net_ops);
3998         if (err)
3999                 goto err_register_pernet_subsys;
4000
4001         xa_init_flags(&tcf_exts_miss_cookies_xa, XA_FLAGS_ALLOC1);
4002
4003         rtnl_register(PF_UNSPEC, RTM_NEWTFILTER, tc_new_tfilter, NULL,
4004                       RTNL_FLAG_DOIT_UNLOCKED);
4005         rtnl_register(PF_UNSPEC, RTM_DELTFILTER, tc_del_tfilter, NULL,
4006                       RTNL_FLAG_DOIT_UNLOCKED);
4007         rtnl_register(PF_UNSPEC, RTM_GETTFILTER, tc_get_tfilter,
4008                       tc_dump_tfilter, RTNL_FLAG_DOIT_UNLOCKED);
4009         rtnl_register(PF_UNSPEC, RTM_NEWCHAIN, tc_ctl_chain, NULL, 0);
4010         rtnl_register(PF_UNSPEC, RTM_DELCHAIN, tc_ctl_chain, NULL, 0);
4011         rtnl_register(PF_UNSPEC, RTM_GETCHAIN, tc_ctl_chain,
4012                       tc_dump_chain, 0);
4013
4014         return 0;
4015
4016 err_register_pernet_subsys:
4017         destroy_workqueue(tc_filter_wq);
4018         return err;
4019 }
4020
4021 subsys_initcall(tc_filter_init);