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 < 0)
 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(NET_CLS_ALIAS_PREFIX "%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         tp->usesw = !tp->ops->reoffload;
 394         spin_lock_init(&tp->lock);
 395         refcount_set(&tp->refcnt, 1);
 396
 397         err = tp->ops->init(tp);
 398         if (err) {
 399                 module_put(tp->ops->owner);
 400                 goto errout;
 401         }
 402         return tp;
 403
 404 errout:
 405         kfree(tp);
 406         return ERR_PTR(err);
 407 }
 408
 409 static void tcf_proto_get(struct tcf_proto *tp)
 410 {
 411         refcount_inc(&tp->refcnt);
 412 }
 413
 414 static void tcf_proto_count_usesw(struct tcf_proto *tp, bool add)
 415 {
 416 #ifdef CONFIG_NET_CLS_ACT
 417         struct tcf_block *block = tp->chain->block;
 418         bool counted = false;
 419
 420         if (!add) {
 421                 if (tp->usesw && tp->counted) {
 422                         if (!atomic_dec_return(&block->useswcnt))
 423                                 static_branch_dec(&tcf_sw_enabled_key);
 424                         tp->counted = false;
 425                 }
 426                 return;
 427         }
 428
 429         spin_lock(&tp->lock);
 430         if (tp->usesw && !tp->counted) {
 431                 counted = true;
 432                 tp->counted = true;
 433         }
 434         spin_unlock(&tp->lock);
 435
 436         if (counted && atomic_inc_return(&block->useswcnt) == 1)
 437                 static_branch_inc(&tcf_sw_enabled_key);
 438 #endif
 439 }
 440
 441 static void tcf_chain_put(struct tcf_chain *chain);
 442
 443 static void tcf_proto_destroy(struct tcf_proto *tp, bool rtnl_held,
 444                               bool sig_destroy, struct netlink_ext_ack *extack)
 445 {
 446         tp->ops->destroy(tp, rtnl_held, extack);
 447         tcf_proto_count_usesw(tp, false);
 448         if (sig_destroy)
 449                 tcf_proto_signal_destroyed(tp->chain, tp);
 450         tcf_chain_put(tp->chain);
 451         module_put(tp->ops->owner);
 452         kfree_rcu(tp, rcu);
 453 }
 454
 455 static void tcf_proto_put(struct tcf_proto *tp, bool rtnl_held,
 456                           struct netlink_ext_ack *extack)
 457 {
 458         if (refcount_dec_and_test(&tp->refcnt))
 459                 tcf_proto_destroy(tp, rtnl_held, true, extack);
 460 }
 461
 462 static bool tcf_proto_check_delete(struct tcf_proto *tp)
 463 {
 464         if (tp->ops->delete_empty)
 465                 return tp->ops->delete_empty(tp);
 466
 467         tp->deleting = true;
 468         return tp->deleting;
 469 }
 470
 471 static void tcf_proto_mark_delete(struct tcf_proto *tp)
 472 {
 473         spin_lock(&tp->lock);
 474         tp->deleting = true;
 475         spin_unlock(&tp->lock);
 476 }
 477
 478 static bool tcf_proto_is_deleting(struct tcf_proto *tp)
 479 {
 480         bool deleting;
 481
 482         spin_lock(&tp->lock);
 483         deleting = tp->deleting;
 484         spin_unlock(&tp->lock);
 485
 486         return deleting;
 487 }
 488
 489 #define ASSERT_BLOCK_LOCKED(block)                                      \
 490         lockdep_assert_held(&(block)->lock)
 491
 492 struct tcf_filter_chain_list_item {
 493         struct list_head list;
 494         tcf_chain_head_change_t *chain_head_change;
 495         void *chain_head_change_priv;
 496 };
 497
 498 static struct tcf_chain *tcf_chain_create(struct tcf_block *block,
 499                                           u32 chain_index)
 500 {
 501         struct tcf_chain *chain;
 502
 503         ASSERT_BLOCK_LOCKED(block);
 504
 505         chain = kzalloc(sizeof(*chain), GFP_KERNEL);
 506         if (!chain)
 507                 return NULL;
 508         list_add_tail_rcu(&chain->list, &block->chain_list);
 509         mutex_init(&chain->filter_chain_lock);
 510         chain->block = block;
 511         chain->index = chain_index;
 512         chain->refcnt = 1;
 513         if (!chain->index)
 514                 block->chain0.chain = chain;
 515         return chain;
 516 }
 517
 518 static void tcf_chain_head_change_item(struct tcf_filter_chain_list_item *item,
 519                                        struct tcf_proto *tp_head)
 520 {
 521         if (item->chain_head_change)
 522                 item->chain_head_change(tp_head, item->chain_head_change_priv);
 523 }
 524
 525 static void tcf_chain0_head_change(struct tcf_chain *chain,
 526                                    struct tcf_proto *tp_head)
 527 {
 528         struct tcf_filter_chain_list_item *item;
 529         struct tcf_block *block = chain->block;
 530
 531         if (chain->index)
 532                 return;
 533
 534         mutex_lock(&block->lock);
 535         list_for_each_entry(item, &block->chain0.filter_chain_list, list)
 536                 tcf_chain_head_change_item(item, tp_head);
 537         mutex_unlock(&block->lock);
 538 }
 539
 540 /* Returns true if block can be safely freed. */
 541
 542 static bool tcf_chain_detach(struct tcf_chain *chain)
 543 {
 544         struct tcf_block *block = chain->block;
 545
 546         ASSERT_BLOCK_LOCKED(block);
 547
 548         list_del_rcu(&chain->list);
 549         if (!chain->index)
 550                 block->chain0.chain = NULL;
 551
 552         if (list_empty(&block->chain_list) &&
 553             refcount_read(&block->refcnt) == 0)
 554                 return true;
 555
 556         return false;
 557 }
 558
 559 static void tcf_block_destroy(struct tcf_block *block)
 560 {
 561         mutex_destroy(&block->lock);
 562         mutex_destroy(&block->proto_destroy_lock);
 563         xa_destroy(&block->ports);
 564         kfree_rcu(block, rcu);
 565 }
 566
 567 static void tcf_chain_destroy(struct tcf_chain *chain, bool free_block)
 568 {
 569         struct tcf_block *block = chain->block;
 570
 571         mutex_destroy(&chain->filter_chain_lock);
 572         kfree_rcu(chain, rcu);
 573         if (free_block)
 574                 tcf_block_destroy(block);
 575 }
 576
 577 static void tcf_chain_hold(struct tcf_chain *chain)
 578 {
 579         ASSERT_BLOCK_LOCKED(chain->block);
 580
 581         ++chain->refcnt;
 582 }
 583
 584 static bool tcf_chain_held_by_acts_only(struct tcf_chain *chain)
 585 {
 586         ASSERT_BLOCK_LOCKED(chain->block);
 587
 588         /* In case all the references are action references, this
 589          * chain should not be shown to the user.
 590          */
 591         return chain->refcnt == chain->action_refcnt;
 592 }
 593
 594 static struct tcf_chain *tcf_chain_lookup(struct tcf_block *block,
 595                                           u32 chain_index)
 596 {
 597         struct tcf_chain *chain;
 598
 599         ASSERT_BLOCK_LOCKED(block);
 600
 601         list_for_each_entry(chain, &block->chain_list, list) {
 602                 if (chain->index == chain_index)
 603                         return chain;
 604         }
 605         return NULL;
 606 }
 607
 608 #if IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
 609 static struct tcf_chain *tcf_chain_lookup_rcu(const struct tcf_block *block,
 610                                               u32 chain_index)
 611 {
 612         struct tcf_chain *chain;
 613
 614         list_for_each_entry_rcu(chain, &block->chain_list, list) {
 615                 if (chain->index == chain_index)
 616                         return chain;
 617         }
 618         return NULL;
 619 }
 620 #endif
 621
 622 static int tc_chain_notify(struct tcf_chain *chain, struct sk_buff *oskb,
 623                            u32 seq, u16 flags, int event, bool unicast,
 624                            struct netlink_ext_ack *extack);
 625
 626 static struct tcf_chain *__tcf_chain_get(struct tcf_block *block,
 627                                          u32 chain_index, bool create,
 628                                          bool by_act)
 629 {
 630         struct tcf_chain *chain = NULL;
 631         bool is_first_reference;
 632
 633         mutex_lock(&block->lock);
 634         chain = tcf_chain_lookup(block, chain_index);
 635         if (chain) {
 636                 tcf_chain_hold(chain);
 637         } else {
 638                 if (!create)
 639                         goto errout;
 640                 chain = tcf_chain_create(block, chain_index);
 641                 if (!chain)
 642                         goto errout;
 643         }
 644
 645         if (by_act)
 646                 ++chain->action_refcnt;
 647         is_first_reference = chain->refcnt - chain->action_refcnt == 1;
 648         mutex_unlock(&block->lock);
 649
 650         /* Send notification only in case we got the first
 651          * non-action reference. Until then, the chain acts only as
 652          * a placeholder for actions pointing to it and user ought
 653          * not know about them.
 654          */
 655         if (is_first_reference && !by_act)
 656                 tc_chain_notify(chain, NULL, 0, NLM_F_CREATE | NLM_F_EXCL,
 657                                 RTM_NEWCHAIN, false, NULL);
 658
 659         return chain;
 660
 661 errout:
 662         mutex_unlock(&block->lock);
 663         return chain;
 664 }
 665
 666 static struct tcf_chain *tcf_chain_get(struct tcf_block *block, u32 chain_index,
 667                                        bool create)
 668 {
 669         return __tcf_chain_get(block, chain_index, create, false);
 670 }
 671
 672 struct tcf_chain *tcf_chain_get_by_act(struct tcf_block *block, u32 chain_index)
 673 {
 674         return __tcf_chain_get(block, chain_index, true, true);
 675 }
 676 EXPORT_SYMBOL(tcf_chain_get_by_act);
 677
 678 static void tc_chain_tmplt_del(const struct tcf_proto_ops *tmplt_ops,
 679                                void *tmplt_priv);
 680 static int tc_chain_notify_delete(const struct tcf_proto_ops *tmplt_ops,
 681                                   void *tmplt_priv, u32 chain_index,
 682                                   struct tcf_block *block, struct sk_buff *oskb,
 683                                   u32 seq, u16 flags);
 684
 685 static void __tcf_chain_put(struct tcf_chain *chain, bool by_act,
 686                             bool explicitly_created)
 687 {
 688         struct tcf_block *block = chain->block;
 689         const struct tcf_proto_ops *tmplt_ops;
 690         unsigned int refcnt, non_act_refcnt;
 691         bool free_block = false;
 692         void *tmplt_priv;
 693
 694         mutex_lock(&block->lock);
 695         if (explicitly_created) {
 696                 if (!chain->explicitly_created) {
 697                         mutex_unlock(&block->lock);
 698                         return;
 699                 }
 700                 chain->explicitly_created = false;
 701         }
 702
 703         if (by_act)
 704                 chain->action_refcnt--;
 705
 706         /* tc_chain_notify_delete can't be called while holding block lock.
 707          * However, when block is unlocked chain can be changed concurrently, so
 708          * save these to temporary variables.
 709          */
 710         refcnt = --chain->refcnt;
 711         non_act_refcnt = refcnt - chain->action_refcnt;
 712         tmplt_ops = chain->tmplt_ops;
 713         tmplt_priv = chain->tmplt_priv;
 714
 715         if (non_act_refcnt == chain->explicitly_created && !by_act) {
 716                 if (non_act_refcnt == 0)
 717                         tc_chain_notify_delete(tmplt_ops, tmplt_priv,
 718                                                chain->index, block, NULL, 0, 0);
 719                 /* Last reference to chain, no need to lock. */
 720                 chain->flushing = false;
 721         }
 722
 723         if (refcnt == 0)
 724                 free_block = tcf_chain_detach(chain);
 725         mutex_unlock(&block->lock);
 726
 727         if (refcnt == 0) {
 728                 tc_chain_tmplt_del(tmplt_ops, tmplt_priv);
 729                 tcf_chain_destroy(chain, free_block);
 730         }
 731 }
 732
 733 static void tcf_chain_put(struct tcf_chain *chain)
 734 {
 735         __tcf_chain_put(chain, false, false);
 736 }
 737
 738 void tcf_chain_put_by_act(struct tcf_chain *chain)
 739 {
 740         __tcf_chain_put(chain, true, false);
 741 }
 742 EXPORT_SYMBOL(tcf_chain_put_by_act);
 743
 744 static void tcf_chain_put_explicitly_created(struct tcf_chain *chain)
 745 {
 746         __tcf_chain_put(chain, false, true);
 747 }
 748
 749 static void tcf_chain_flush(struct tcf_chain *chain, bool rtnl_held)
 750 {
 751         struct tcf_proto *tp, *tp_next;
 752
 753         mutex_lock(&chain->filter_chain_lock);
 754         tp = tcf_chain_dereference(chain->filter_chain, chain);
 755         while (tp) {
 756                 tp_next = rcu_dereference_protected(tp->next, 1);
 757                 tcf_proto_signal_destroying(chain, tp);
 758                 tp = tp_next;
 759         }
 760         tp = tcf_chain_dereference(chain->filter_chain, chain);
 761         RCU_INIT_POINTER(chain->filter_chain, NULL);
 762         tcf_chain0_head_change(chain, NULL);
 763         chain->flushing = true;
 764         mutex_unlock(&chain->filter_chain_lock);
 765
 766         while (tp) {
 767                 tp_next = rcu_dereference_protected(tp->next, 1);
 768                 tcf_proto_put(tp, rtnl_held, NULL);
 769                 tp = tp_next;
 770         }
 771 }
 772
 773 static int tcf_block_setup(struct tcf_block *block,
 774                            struct flow_block_offload *bo);
 775
 776 static void tcf_block_offload_init(struct flow_block_offload *bo,
 777                                    struct net_device *dev, struct Qdisc *sch,
 778                                    enum flow_block_command command,
 779                                    enum flow_block_binder_type binder_type,
 780                                    struct flow_block *flow_block,
 781                                    bool shared, struct netlink_ext_ack *extack)
 782 {
 783         bo->net = dev_net(dev);
 784         bo->command = command;
 785         bo->binder_type = binder_type;
 786         bo->block = flow_block;
 787         bo->block_shared = shared;
 788         bo->extack = extack;
 789         bo->sch = sch;
 790         bo->cb_list_head = &flow_block->cb_list;
 791         INIT_LIST_HEAD(&bo->cb_list);
 792 }
 793
 794 static void tcf_block_unbind(struct tcf_block *block,
 795                              struct flow_block_offload *bo);
 796
 797 static void tc_block_indr_cleanup(struct flow_block_cb *block_cb)
 798 {
 799         struct tcf_block *block = block_cb->indr.data;
 800         struct net_device *dev = block_cb->indr.dev;
 801         struct Qdisc *sch = block_cb->indr.sch;
 802         struct netlink_ext_ack extack = {};
 803         struct flow_block_offload bo = {};
 804
 805         tcf_block_offload_init(&bo, dev, sch, FLOW_BLOCK_UNBIND,
 806                                block_cb->indr.binder_type,
 807                                &block->flow_block, tcf_block_shared(block),
 808                                &extack);
 809         rtnl_lock();
 810         down_write(&block->cb_lock);
 811         list_del(&block_cb->driver_list);
 812         list_move(&block_cb->list, &bo.cb_list);
 813         tcf_block_unbind(block, &bo);
 814         up_write(&block->cb_lock);
 815         rtnl_unlock();
 816 }
 817
 818 static bool tcf_block_offload_in_use(struct tcf_block *block)
 819 {
 820         return atomic_read(&block->offloadcnt);
 821 }
 822
 823 static int tcf_block_offload_cmd(struct tcf_block *block,
 824                                  struct net_device *dev, struct Qdisc *sch,
 825                                  struct tcf_block_ext_info *ei,
 826                                  enum flow_block_command command,
 827                                  struct netlink_ext_ack *extack)
 828 {
 829         struct flow_block_offload bo = {};
 830
 831         tcf_block_offload_init(&bo, dev, sch, command, ei->binder_type,
 832                                &block->flow_block, tcf_block_shared(block),
 833                                extack);
 834
 835         if (dev->netdev_ops->ndo_setup_tc) {
 836                 int err;
 837
 838                 err = dev->netdev_ops->ndo_setup_tc(dev, TC_SETUP_BLOCK, &bo);
 839                 if (err < 0) {
 840                         if (err != -EOPNOTSUPP)
 841                                 NL_SET_ERR_MSG(extack, "Driver ndo_setup_tc failed");
 842                         return err;
 843                 }
 844
 845                 return tcf_block_setup(block, &bo);
 846         }
 847
 848         flow_indr_dev_setup_offload(dev, sch, TC_SETUP_BLOCK, block, &bo,
 849                                     tc_block_indr_cleanup);
 850         tcf_block_setup(block, &bo);
 851
 852         return -EOPNOTSUPP;
 853 }
 854
 855 static int tcf_block_offload_bind(struct tcf_block *block, struct Qdisc *q,
 856                                   struct tcf_block_ext_info *ei,
 857                                   struct netlink_ext_ack *extack)
 858 {
 859         struct net_device *dev = q->dev_queue->dev;
 860         int err;
 861
 862         down_write(&block->cb_lock);
 863
 864         /* If tc offload feature is disabled and the block we try to bind
 865          * to already has some offloaded filters, forbid to bind.
 866          */
 867         if (dev->netdev_ops->ndo_setup_tc &&
 868             !tc_can_offload(dev) &&
 869             tcf_block_offload_in_use(block)) {
 870                 NL_SET_ERR_MSG(extack, "Bind to offloaded block failed as dev has offload disabled");
 871                 err = -EOPNOTSUPP;
 872                 goto err_unlock;
 873         }
 874
 875         err = tcf_block_offload_cmd(block, dev, q, ei, FLOW_BLOCK_BIND, extack);
 876         if (err == -EOPNOTSUPP)
 877                 goto no_offload_dev_inc;
 878         if (err)
 879                 goto err_unlock;
 880
 881         up_write(&block->cb_lock);
 882         return 0;
 883
 884 no_offload_dev_inc:
 885         if (tcf_block_offload_in_use(block))
 886                 goto err_unlock;
 887
 888         err = 0;
 889         block->nooffloaddevcnt++;
 890 err_unlock:
 891         up_write(&block->cb_lock);
 892         return err;
 893 }
 894
 895 static void tcf_block_offload_unbind(struct tcf_block *block, struct Qdisc *q,
 896                                      struct tcf_block_ext_info *ei)
 897 {
 898         struct net_device *dev = q->dev_queue->dev;
 899         int err;
 900
 901         down_write(&block->cb_lock);
 902         err = tcf_block_offload_cmd(block, dev, q, ei, FLOW_BLOCK_UNBIND, NULL);
 903         if (err == -EOPNOTSUPP)
 904                 goto no_offload_dev_dec;
 905         up_write(&block->cb_lock);
 906         return;
 907
 908 no_offload_dev_dec:
 909         WARN_ON(block->nooffloaddevcnt-- == 0);
 910         up_write(&block->cb_lock);
 911 }
 912
 913 static int
 914 tcf_chain0_head_change_cb_add(struct tcf_block *block,
 915                               struct tcf_block_ext_info *ei,
 916                               struct netlink_ext_ack *extack)
 917 {
 918         struct tcf_filter_chain_list_item *item;
 919         struct tcf_chain *chain0;
 920
 921         item = kmalloc(sizeof(*item), GFP_KERNEL);
 922         if (!item) {
 923                 NL_SET_ERR_MSG(extack, "Memory allocation for head change callback item failed");
 924                 return -ENOMEM;
 925         }
 926         item->chain_head_change = ei->chain_head_change;
 927         item->chain_head_change_priv = ei->chain_head_change_priv;
 928
 929         mutex_lock(&block->lock);
 930         chain0 = block->chain0.chain;
 931         if (chain0)
 932                 tcf_chain_hold(chain0);
 933         else
 934                 list_add(&item->list, &block->chain0.filter_chain_list);
 935         mutex_unlock(&block->lock);
 936
 937         if (chain0) {
 938                 struct tcf_proto *tp_head;
 939
 940                 mutex_lock(&chain0->filter_chain_lock);
 941
 942                 tp_head = tcf_chain_dereference(chain0->filter_chain, chain0);
 943                 if (tp_head)
 944                         tcf_chain_head_change_item(item, tp_head);
 945
 946                 mutex_lock(&block->lock);
 947                 list_add(&item->list, &block->chain0.filter_chain_list);
 948                 mutex_unlock(&block->lock);
 949
 950                 mutex_unlock(&chain0->filter_chain_lock);
 951                 tcf_chain_put(chain0);
 952         }
 953
 954         return 0;
 955 }
 956
 957 static void
 958 tcf_chain0_head_change_cb_del(struct tcf_block *block,
 959                               struct tcf_block_ext_info *ei)
 960 {
 961         struct tcf_filter_chain_list_item *item;
 962
 963         mutex_lock(&block->lock);
 964         list_for_each_entry(item, &block->chain0.filter_chain_list, list) {
 965                 if ((!ei->chain_head_change && !ei->chain_head_change_priv) ||
 966                     (item->chain_head_change == ei->chain_head_change &&
 967                      item->chain_head_change_priv == ei->chain_head_change_priv)) {
 968                         if (block->chain0.chain)
 969                                 tcf_chain_head_change_item(item, NULL);
 970                         list_del(&item->list);
 971                         mutex_unlock(&block->lock);
 972
 973                         kfree(item);
 974                         return;
 975                 }
 976         }
 977         mutex_unlock(&block->lock);
 978         WARN_ON(1);
 979 }
 980
 981 struct tcf_net {
 982         spinlock_t idr_lock; /* Protects idr */
 983         struct idr idr;
 984 };
 985
 986 static unsigned int tcf_net_id;
 987
 988 static int tcf_block_insert(struct tcf_block *block, struct net *net,
 989                             struct netlink_ext_ack *extack)
 990 {
 991         struct tcf_net *tn = net_generic(net, tcf_net_id);
 992         int err;
 993
 994         idr_preload(GFP_KERNEL);
 995         spin_lock(&tn->idr_lock);
 996         err = idr_alloc_u32(&tn->idr, block, &block->index, block->index,
 997                             GFP_NOWAIT);
 998         spin_unlock(&tn->idr_lock);
 999         idr_preload_end();
1000
1001         return err;
1002 }
1003
1004 static void tcf_block_remove(struct tcf_block *block, struct net *net)
1005 {
1006         struct tcf_net *tn = net_generic(net, tcf_net_id);
1007
1008         spin_lock(&tn->idr_lock);
1009         idr_remove(&tn->idr, block->index);
1010         spin_unlock(&tn->idr_lock);
1011 }
1012
1013 static struct tcf_block *tcf_block_create(struct net *net, struct Qdisc *q,
1014                                           u32 block_index,
1015                                           struct netlink_ext_ack *extack)
1016 {
1017         struct tcf_block *block;
1018
1019         block = kzalloc(sizeof(*block), GFP_KERNEL);
1020         if (!block) {
1021                 NL_SET_ERR_MSG(extack, "Memory allocation for block failed");
1022                 return ERR_PTR(-ENOMEM);
1023         }
1024         mutex_init(&block->lock);
1025         mutex_init(&block->proto_destroy_lock);
1026         init_rwsem(&block->cb_lock);
1027         flow_block_init(&block->flow_block);
1028         INIT_LIST_HEAD(&block->chain_list);
1029         INIT_LIST_HEAD(&block->owner_list);
1030         INIT_LIST_HEAD(&block->chain0.filter_chain_list);
1031
1032         refcount_set(&block->refcnt, 1);
1033         block->net = net;
1034         block->index = block_index;
1035         xa_init(&block->ports);
1036
1037         /* Don't store q pointer for blocks which are shared */
1038         if (!tcf_block_shared(block))
1039                 block->q = q;
1040         return block;
1041 }
1042
1043 struct tcf_block *tcf_block_lookup(struct net *net, u32 block_index)
1044 {
1045         struct tcf_net *tn = net_generic(net, tcf_net_id);
1046
1047         return idr_find(&tn->idr, block_index);
1048 }
1049 EXPORT_SYMBOL(tcf_block_lookup);
1050
1051 static struct tcf_block *tcf_block_refcnt_get(struct net *net, u32 block_index)
1052 {
1053         struct tcf_block *block;
1054
1055         rcu_read_lock();
1056         block = tcf_block_lookup(net, block_index);
1057         if (block && !refcount_inc_not_zero(&block->refcnt))
1058                 block = NULL;
1059         rcu_read_unlock();
1060
1061         return block;
1062 }
1063
1064 static struct tcf_chain *
1065 __tcf_get_next_chain(struct tcf_block *block, struct tcf_chain *chain)
1066 {
1067         mutex_lock(&block->lock);
1068         if (chain)
1069                 chain = list_is_last(&chain->list, &block->chain_list) ?
1070                         NULL : list_next_entry(chain, list);
1071         else
1072                 chain = list_first_entry_or_null(&block->chain_list,
1073                                                  struct tcf_chain, list);
1074
1075         /* skip all action-only chains */
1076         while (chain && tcf_chain_held_by_acts_only(chain))
1077                 chain = list_is_last(&chain->list, &block->chain_list) ?
1078                         NULL : list_next_entry(chain, list);
1079
1080         if (chain)
1081                 tcf_chain_hold(chain);
1082         mutex_unlock(&block->lock);
1083
1084         return chain;
1085 }
1086
1087 /* Function to be used by all clients that want to iterate over all chains on
1088  * block. It properly obtains block->lock and takes reference to chain before
1089  * returning it. Users of this function must be tolerant to concurrent chain
1090  * insertion/deletion or ensure that no concurrent chain modification is
1091  * possible. Note that all netlink dump callbacks cannot guarantee to provide
1092  * consistent dump because rtnl lock is released each time skb is filled with
1093  * data and sent to user-space.
1094  */
1095
1096 struct tcf_chain *
1097 tcf_get_next_chain(struct tcf_block *block, struct tcf_chain *chain)
1098 {
1099         struct tcf_chain *chain_next = __tcf_get_next_chain(block, chain);
1100
1101         if (chain)
1102                 tcf_chain_put(chain);
1103
1104         return chain_next;
1105 }
1106 EXPORT_SYMBOL(tcf_get_next_chain);
1107
1108 static struct tcf_proto *
1109 __tcf_get_next_proto(struct tcf_chain *chain, struct tcf_proto *tp)
1110 {
1111         u32 prio = 0;
1112
1113         ASSERT_RTNL();
1114         mutex_lock(&chain->filter_chain_lock);
1115
1116         if (!tp) {
1117                 tp = tcf_chain_dereference(chain->filter_chain, chain);
1118         } else if (tcf_proto_is_deleting(tp)) {
1119                 /* 'deleting' flag is set and chain->filter_chain_lock was
1120                  * unlocked, which means next pointer could be invalid. Restart
1121                  * search.
1122                  */
1123                 prio = tp->prio + 1;
1124                 tp = tcf_chain_dereference(chain->filter_chain, chain);
1125
1126                 for (; tp; tp = tcf_chain_dereference(tp->next, chain))
1127                         if (!tp->deleting && tp->prio >= prio)
1128                                 break;
1129         } else {
1130                 tp = tcf_chain_dereference(tp->next, chain);
1131         }
1132
1133         if (tp)
1134                 tcf_proto_get(tp);
1135
1136         mutex_unlock(&chain->filter_chain_lock);
1137
1138         return tp;
1139 }
1140
1141 /* Function to be used by all clients that want to iterate over all tp's on
1142  * chain. Users of this function must be tolerant to concurrent tp
1143  * insertion/deletion or ensure that no concurrent chain modification is
1144  * possible. Note that all netlink dump callbacks cannot guarantee to provide
1145  * consistent dump because rtnl lock is released each time skb is filled with
1146  * data and sent to user-space.
1147  */
1148
1149 struct tcf_proto *
1150 tcf_get_next_proto(struct tcf_chain *chain, struct tcf_proto *tp)
1151 {
1152         struct tcf_proto *tp_next = __tcf_get_next_proto(chain, tp);
1153
1154         if (tp)
1155                 tcf_proto_put(tp, true, NULL);
1156
1157         return tp_next;
1158 }
1159 EXPORT_SYMBOL(tcf_get_next_proto);
1160
1161 static void tcf_block_flush_all_chains(struct tcf_block *block, bool rtnl_held)
1162 {
1163         struct tcf_chain *chain;
1164
1165         /* Last reference to block. At this point chains cannot be added or
1166          * removed concurrently.
1167          */
1168         for (chain = tcf_get_next_chain(block, NULL);
1169              chain;
1170              chain = tcf_get_next_chain(block, chain)) {
1171                 tcf_chain_put_explicitly_created(chain);
1172                 tcf_chain_flush(chain, rtnl_held);
1173         }
1174 }
1175
1176 /* Lookup Qdisc and increments its reference counter.
1177  * Set parent, if necessary.
1178  */
1179
1180 static int __tcf_qdisc_find(struct net *net, struct Qdisc **q,
1181                             u32 *parent, int ifindex, bool rtnl_held,
1182                             struct netlink_ext_ack *extack)
1183 {
1184         const struct Qdisc_class_ops *cops;
1185         struct net_device *dev;
1186         int err = 0;
1187
1188         if (ifindex == TCM_IFINDEX_MAGIC_BLOCK)
1189                 return 0;
1190
1191         rcu_read_lock();
1192
1193         /* Find link */
1194         dev = dev_get_by_index_rcu(net, ifindex);
1195         if (!dev) {
1196                 rcu_read_unlock();
1197                 return -ENODEV;
1198         }
1199
1200         /* Find qdisc */
1201         if (!*parent) {
1202                 *q = rcu_dereference(dev->qdisc);
1203                 *parent = (*q)->handle;
1204         } else {
1205                 *q = qdisc_lookup_rcu(dev, TC_H_MAJ(*parent));
1206                 if (!*q) {
1207                         NL_SET_ERR_MSG(extack, "Parent Qdisc doesn't exists");
1208                         err = -EINVAL;
1209                         goto errout_rcu;
1210                 }
1211         }
1212
1213         *q = qdisc_refcount_inc_nz(*q);
1214         if (!*q) {
1215                 NL_SET_ERR_MSG(extack, "Parent Qdisc doesn't exists");
1216                 err = -EINVAL;
1217                 goto errout_rcu;
1218         }
1219
1220         /* Is it classful? */
1221         cops = (*q)->ops->cl_ops;
1222         if (!cops) {
1223                 NL_SET_ERR_MSG(extack, "Qdisc not classful");
1224                 err = -EINVAL;
1225                 goto errout_qdisc;
1226         }
1227
1228         if (!cops->tcf_block) {
1229                 NL_SET_ERR_MSG(extack, "Class doesn't support blocks");
1230                 err = -EOPNOTSUPP;
1231                 goto errout_qdisc;
1232         }
1233
1234 errout_rcu:
1235         /* At this point we know that qdisc is not noop_qdisc,
1236          * which means that qdisc holds a reference to net_device
1237          * and we hold a reference to qdisc, so it is safe to release
1238          * rcu read lock.
1239          */
1240         rcu_read_unlock();
1241         return err;
1242
1243 errout_qdisc:
1244         rcu_read_unlock();
1245
1246         if (rtnl_held)
1247                 qdisc_put(*q);
1248         else
1249                 qdisc_put_unlocked(*q);
1250         *q = NULL;
1251
1252         return err;
1253 }
1254
1255 static int __tcf_qdisc_cl_find(struct Qdisc *q, u32 parent, unsigned long *cl,
1256                                int ifindex, struct netlink_ext_ack *extack)
1257 {
1258         if (ifindex == TCM_IFINDEX_MAGIC_BLOCK)
1259                 return 0;
1260
1261         /* Do we search for filter, attached to class? */
1262         if (TC_H_MIN(parent)) {
1263                 const struct Qdisc_class_ops *cops = q->ops->cl_ops;
1264
1265                 *cl = cops->find(q, parent);
1266                 if (*cl == 0) {
1267                         NL_SET_ERR_MSG(extack, "Specified class doesn't exist");
1268                         return -ENOENT;
1269                 }
1270         }
1271
1272         return 0;
1273 }
1274
1275 static struct tcf_block *__tcf_block_find(struct net *net, struct Qdisc *q,
1276                                           unsigned long cl, int ifindex,
1277                                           u32 block_index,
1278                                           struct netlink_ext_ack *extack)
1279 {
1280         struct tcf_block *block;
1281
1282         if (ifindex == TCM_IFINDEX_MAGIC_BLOCK) {
1283                 block = tcf_block_refcnt_get(net, block_index);
1284                 if (!block) {
1285                         NL_SET_ERR_MSG(extack, "Block of given index was not found");
1286                         return ERR_PTR(-EINVAL);
1287                 }
1288         } else {
1289                 const struct Qdisc_class_ops *cops = q->ops->cl_ops;
1290
1291                 block = cops->tcf_block(q, cl, extack);
1292                 if (!block)
1293                         return ERR_PTR(-EINVAL);
1294
1295                 if (tcf_block_shared(block)) {
1296                         NL_SET_ERR_MSG(extack, "This filter block is shared. Please use the block index to manipulate the filters");
1297                         return ERR_PTR(-EOPNOTSUPP);
1298                 }
1299
1300                 /* Always take reference to block in order to support execution
1301                  * of rules update path of cls API without rtnl lock. Caller
1302                  * must release block when it is finished using it. 'if' block
1303                  * of this conditional obtain reference to block by calling
1304                  * tcf_block_refcnt_get().
1305                  */
1306                 refcount_inc(&block->refcnt);
1307         }
1308
1309         return block;
1310 }
1311
1312 static void __tcf_block_put(struct tcf_block *block, struct Qdisc *q,
1313                             struct tcf_block_ext_info *ei, bool rtnl_held)
1314 {
1315         if (refcount_dec_and_mutex_lock(&block->refcnt, &block->lock)) {
1316                 /* Flushing/putting all chains will cause the block to be
1317                  * deallocated when last chain is freed. However, if chain_list
1318                  * is empty, block has to be manually deallocated. After block
1319                  * reference counter reached 0, it is no longer possible to
1320                  * increment it or add new chains to block.
1321                  */
1322                 bool free_block = list_empty(&block->chain_list);
1323
1324                 mutex_unlock(&block->lock);
1325                 if (tcf_block_shared(block))
1326                         tcf_block_remove(block, block->net);
1327
1328                 if (q)
1329                         tcf_block_offload_unbind(block, q, ei);
1330
1331                 if (free_block)
1332                         tcf_block_destroy(block);
1333                 else
1334                         tcf_block_flush_all_chains(block, rtnl_held);
1335         } else if (q) {
1336                 tcf_block_offload_unbind(block, q, ei);
1337         }
1338 }
1339
1340 static void tcf_block_refcnt_put(struct tcf_block *block, bool rtnl_held)
1341 {
1342         __tcf_block_put(block, NULL, NULL, rtnl_held);
1343 }
1344
1345 /* Find tcf block.
1346  * Set q, parent, cl when appropriate.
1347  */
1348
1349 static struct tcf_block *tcf_block_find(struct net *net, struct Qdisc **q,
1350                                         u32 *parent, unsigned long *cl,
1351                                         int ifindex, u32 block_index,
1352                                         struct netlink_ext_ack *extack)
1353 {
1354         struct tcf_block *block;
1355         int err = 0;
1356
1357         ASSERT_RTNL();
1358
1359         err = __tcf_qdisc_find(net, q, parent, ifindex, true, extack);
1360         if (err)
1361                 goto errout;
1362
1363         err = __tcf_qdisc_cl_find(*q, *parent, cl, ifindex, extack);
1364         if (err)
1365                 goto errout_qdisc;
1366
1367         block = __tcf_block_find(net, *q, *cl, ifindex, block_index, extack);
1368         if (IS_ERR(block)) {
1369                 err = PTR_ERR(block);
1370                 goto errout_qdisc;
1371         }
1372
1373         return block;
1374
1375 errout_qdisc:
1376         if (*q)
1377                 qdisc_put(*q);
1378 errout:
1379         *q = NULL;
1380         return ERR_PTR(err);
1381 }
1382
1383 static void tcf_block_release(struct Qdisc *q, struct tcf_block *block,
1384                               bool rtnl_held)
1385 {
1386         if (!IS_ERR_OR_NULL(block))
1387                 tcf_block_refcnt_put(block, rtnl_held);
1388
1389         if (q) {
1390                 if (rtnl_held)
1391                         qdisc_put(q);
1392                 else
1393                         qdisc_put_unlocked(q);
1394         }
1395 }
1396
1397 struct tcf_block_owner_item {
1398         struct list_head list;
1399         struct Qdisc *q;
1400         enum flow_block_binder_type binder_type;
1401 };
1402
1403 static void
1404 tcf_block_owner_netif_keep_dst(struct tcf_block *block,
1405                                struct Qdisc *q,
1406                                enum flow_block_binder_type binder_type)
1407 {
1408         if (block->keep_dst &&
1409             binder_type != FLOW_BLOCK_BINDER_TYPE_CLSACT_INGRESS &&
1410             binder_type != FLOW_BLOCK_BINDER_TYPE_CLSACT_EGRESS)
1411                 netif_keep_dst(qdisc_dev(q));
1412 }
1413
1414 void tcf_block_netif_keep_dst(struct tcf_block *block)
1415 {
1416         struct tcf_block_owner_item *item;
1417
1418         block->keep_dst = true;
1419         list_for_each_entry(item, &block->owner_list, list)
1420                 tcf_block_owner_netif_keep_dst(block, item->q,
1421                                                item->binder_type);
1422 }
1423 EXPORT_SYMBOL(tcf_block_netif_keep_dst);
1424
1425 static int tcf_block_owner_add(struct tcf_block *block,
1426                                struct Qdisc *q,
1427                                enum flow_block_binder_type binder_type)
1428 {
1429         struct tcf_block_owner_item *item;
1430
1431         item = kmalloc(sizeof(*item), GFP_KERNEL);
1432         if (!item)
1433                 return -ENOMEM;
1434         item->q = q;
1435         item->binder_type = binder_type;
1436         list_add(&item->list, &block->owner_list);
1437         return 0;
1438 }
1439
1440 static void tcf_block_owner_del(struct tcf_block *block,
1441                                 struct Qdisc *q,
1442                                 enum flow_block_binder_type binder_type)
1443 {
1444         struct tcf_block_owner_item *item;
1445
1446         list_for_each_entry(item, &block->owner_list, list) {
1447                 if (item->q == q && item->binder_type == binder_type) {
1448                         list_del(&item->list);
1449                         kfree(item);
1450                         return;
1451                 }
1452         }
1453         WARN_ON(1);
1454 }
1455
1456 static bool tcf_block_tracks_dev(struct tcf_block *block,
1457                                  struct tcf_block_ext_info *ei)
1458 {
1459         return tcf_block_shared(block) &&
1460                (ei->binder_type == FLOW_BLOCK_BINDER_TYPE_CLSACT_INGRESS ||
1461                 ei->binder_type == FLOW_BLOCK_BINDER_TYPE_CLSACT_EGRESS);
1462 }
1463
1464 int tcf_block_get_ext(struct tcf_block **p_block, struct Qdisc *q,
1465                       struct tcf_block_ext_info *ei,
1466                       struct netlink_ext_ack *extack)
1467 {
1468         struct net_device *dev = qdisc_dev(q);
1469         struct net *net = qdisc_net(q);
1470         struct tcf_block *block = NULL;
1471         int err;
1472
1473         if (ei->block_index)
1474                 /* block_index not 0 means the shared block is requested */
1475                 block = tcf_block_refcnt_get(net, ei->block_index);
1476
1477         if (!block) {
1478                 block = tcf_block_create(net, q, ei->block_index, extack);
1479                 if (IS_ERR(block))
1480                         return PTR_ERR(block);
1481                 if (tcf_block_shared(block)) {
1482                         err = tcf_block_insert(block, net, extack);
1483                         if (err)
1484                                 goto err_block_insert;
1485                 }
1486         }
1487
1488         err = tcf_block_owner_add(block, q, ei->binder_type);
1489         if (err)
1490                 goto err_block_owner_add;
1491
1492         tcf_block_owner_netif_keep_dst(block, q, ei->binder_type);
1493
1494         err = tcf_chain0_head_change_cb_add(block, ei, extack);
1495         if (err)
1496                 goto err_chain0_head_change_cb_add;
1497
1498         err = tcf_block_offload_bind(block, q, ei, extack);
1499         if (err)
1500                 goto err_block_offload_bind;
1501
1502         if (tcf_block_tracks_dev(block, ei)) {
1503                 err = xa_insert(&block->ports, dev->ifindex, dev, GFP_KERNEL);
1504                 if (err) {
1505                         NL_SET_ERR_MSG(extack, "block dev insert failed");
1506                         goto err_dev_insert;
1507                 }
1508         }
1509
1510         *p_block = block;
1511         return 0;
1512
1513 err_dev_insert:
1514         tcf_block_offload_unbind(block, q, ei);
1515 err_block_offload_bind:
1516         tcf_chain0_head_change_cb_del(block, ei);
1517 err_chain0_head_change_cb_add:
1518         tcf_block_owner_del(block, q, ei->binder_type);
1519 err_block_owner_add:
1520 err_block_insert:
1521         tcf_block_refcnt_put(block, true);
1522         return err;
1523 }
1524 EXPORT_SYMBOL(tcf_block_get_ext);
1525
1526 static void tcf_chain_head_change_dflt(struct tcf_proto *tp_head, void *priv)
1527 {
1528         struct tcf_proto __rcu **p_filter_chain = priv;
1529
1530         rcu_assign_pointer(*p_filter_chain, tp_head);
1531 }
1532
1533 int tcf_block_get(struct tcf_block **p_block,
1534                   struct tcf_proto __rcu **p_filter_chain, struct Qdisc *q,
1535                   struct netlink_ext_ack *extack)
1536 {
1537         struct tcf_block_ext_info ei = {
1538                 .chain_head_change = tcf_chain_head_change_dflt,
1539                 .chain_head_change_priv = p_filter_chain,
1540         };
1541
1542         WARN_ON(!p_filter_chain);
1543         return tcf_block_get_ext(p_block, q, &ei, extack);
1544 }
1545 EXPORT_SYMBOL(tcf_block_get);
1546
1547 /* XXX: Standalone actions are not allowed to jump to any chain, and bound
1548  * actions should be all removed after flushing.
1549  */
1550 void tcf_block_put_ext(struct tcf_block *block, struct Qdisc *q,
1551                        struct tcf_block_ext_info *ei)
1552 {
1553         struct net_device *dev = qdisc_dev(q);
1554
1555         if (!block)
1556                 return;
1557         if (tcf_block_tracks_dev(block, ei))
1558                 xa_erase(&block->ports, dev->ifindex);
1559         tcf_chain0_head_change_cb_del(block, ei);
1560         tcf_block_owner_del(block, q, ei->binder_type);
1561
1562         __tcf_block_put(block, q, ei, true);
1563 }
1564 EXPORT_SYMBOL(tcf_block_put_ext);
1565
1566 void tcf_block_put(struct tcf_block *block)
1567 {
1568         struct tcf_block_ext_info ei = {0, };
1569
1570         if (!block)
1571                 return;
1572         tcf_block_put_ext(block, block->q, &ei);
1573 }
1574
1575 EXPORT_SYMBOL(tcf_block_put);
1576
1577 static int
1578 tcf_block_playback_offloads(struct tcf_block *block, flow_setup_cb_t *cb,
1579                             void *cb_priv, bool add, bool offload_in_use,
1580                             struct netlink_ext_ack *extack)
1581 {
1582         struct tcf_chain *chain, *chain_prev;
1583         struct tcf_proto *tp, *tp_prev;
1584         int err;
1585
1586         lockdep_assert_held(&block->cb_lock);
1587
1588         for (chain = __tcf_get_next_chain(block, NULL);
1589              chain;
1590              chain_prev = chain,
1591                      chain = __tcf_get_next_chain(block, chain),
1592                      tcf_chain_put(chain_prev)) {
1593                 if (chain->tmplt_ops && add)
1594                         chain->tmplt_ops->tmplt_reoffload(chain, true, cb,
1595                                                           cb_priv);
1596                 for (tp = __tcf_get_next_proto(chain, NULL); tp;
1597                      tp_prev = tp,
1598                              tp = __tcf_get_next_proto(chain, tp),
1599                              tcf_proto_put(tp_prev, true, NULL)) {
1600                         if (tp->ops->reoffload) {
1601                                 err = tp->ops->reoffload(tp, add, cb, cb_priv,
1602                                                          extack);
1603                                 if (err && add)
1604                                         goto err_playback_remove;
1605                         } else if (add && offload_in_use) {
1606                                 err = -EOPNOTSUPP;
1607                                 NL_SET_ERR_MSG(extack, "Filter HW offload failed - classifier without re-offloading support");
1608                                 goto err_playback_remove;
1609                         }
1610                 }
1611                 if (chain->tmplt_ops && !add)
1612                         chain->tmplt_ops->tmplt_reoffload(chain, false, cb,
1613                                                           cb_priv);
1614         }
1615
1616         return 0;
1617
1618 err_playback_remove:
1619         tcf_proto_put(tp, true, NULL);
1620         tcf_chain_put(chain);
1621         tcf_block_playback_offloads(block, cb, cb_priv, false, offload_in_use,
1622                                     extack);
1623         return err;
1624 }
1625
1626 static int tcf_block_bind(struct tcf_block *block,
1627                           struct flow_block_offload *bo)
1628 {
1629         struct flow_block_cb *block_cb, *next;
1630         int err, i = 0;
1631
1632         lockdep_assert_held(&block->cb_lock);
1633
1634         list_for_each_entry(block_cb, &bo->cb_list, list) {
1635                 err = tcf_block_playback_offloads(block, block_cb->cb,
1636                                                   block_cb->cb_priv, true,
1637                                                   tcf_block_offload_in_use(block),
1638                                                   bo->extack);
1639                 if (err)
1640                         goto err_unroll;
1641                 if (!bo->unlocked_driver_cb)
1642                         block->lockeddevcnt++;
1643
1644                 i++;
1645         }
1646         list_splice(&bo->cb_list, &block->flow_block.cb_list);
1647
1648         return 0;
1649
1650 err_unroll:
1651         list_for_each_entry_safe(block_cb, next, &bo->cb_list, list) {
1652                 list_del(&block_cb->driver_list);
1653                 if (i-- > 0) {
1654                         list_del(&block_cb->list);
1655                         tcf_block_playback_offloads(block, block_cb->cb,
1656                                                     block_cb->cb_priv, false,
1657                                                     tcf_block_offload_in_use(block),
1658                                                     NULL);
1659                         if (!bo->unlocked_driver_cb)
1660                                 block->lockeddevcnt--;
1661                 }
1662                 flow_block_cb_free(block_cb);
1663         }
1664
1665         return err;
1666 }
1667
1668 static void tcf_block_unbind(struct tcf_block *block,
1669                              struct flow_block_offload *bo)
1670 {
1671         struct flow_block_cb *block_cb, *next;
1672
1673         lockdep_assert_held(&block->cb_lock);
1674
1675         list_for_each_entry_safe(block_cb, next, &bo->cb_list, list) {
1676                 tcf_block_playback_offloads(block, block_cb->cb,
1677                                             block_cb->cb_priv, false,
1678                                             tcf_block_offload_in_use(block),
1679                                             NULL);
1680                 list_del(&block_cb->list);
1681                 flow_block_cb_free(block_cb);
1682                 if (!bo->unlocked_driver_cb)
1683                         block->lockeddevcnt--;
1684         }
1685 }
1686
1687 static int tcf_block_setup(struct tcf_block *block,
1688                            struct flow_block_offload *bo)
1689 {
1690         int err;
1691
1692         switch (bo->command) {
1693         case FLOW_BLOCK_BIND:
1694                 err = tcf_block_bind(block, bo);
1695                 break;
1696         case FLOW_BLOCK_UNBIND:
1697                 err = 0;
1698                 tcf_block_unbind(block, bo);
1699                 break;
1700         default:
1701                 WARN_ON_ONCE(1);
1702                 err = -EOPNOTSUPP;
1703         }
1704
1705         return err;
1706 }
1707
1708 /* Main classifier routine: scans classifier chain attached
1709  * to this qdisc, (optionally) tests for protocol and asks
1710  * specific classifiers.
1711  */
1712 static inline int __tcf_classify(struct sk_buff *skb,
1713                                  const struct tcf_proto *tp,
1714                                  const struct tcf_proto *orig_tp,
1715                                  struct tcf_result *res,
1716                                  bool compat_mode,
1717                                  struct tcf_exts_miss_cookie_node *n,
1718                                  int act_index,
1719                                  u32 *last_executed_chain)
1720 {
1721 #ifdef CONFIG_NET_CLS_ACT
1722         const int max_reclassify_loop = 16;
1723         const struct tcf_proto *first_tp;
1724         int limit = 0;
1725
1726 reclassify:
1727 #endif
1728         for (; tp; tp = rcu_dereference_bh(tp->next)) {
1729                 __be16 protocol = skb_protocol(skb, false);
1730                 int err = 0;
1731
1732                 if (n) {
1733                         struct tcf_exts *exts;
1734
1735                         if (n->tp_prio != tp->prio)
1736                                 continue;
1737
1738                         /* We re-lookup the tp and chain based on index instead
1739                          * of having hard refs and locks to them, so do a sanity
1740                          * check if any of tp,chain,exts was replaced by the
1741                          * time we got here with a cookie from hardware.
1742                          */
1743                         if (unlikely(n->tp != tp || n->tp->chain != n->chain ||
1744                                      !tp->ops->get_exts)) {
1745                                 tcf_set_drop_reason(skb,
1746                                                     SKB_DROP_REASON_TC_COOKIE_ERROR);
1747                                 return TC_ACT_SHOT;
1748                         }
1749
1750                         exts = tp->ops->get_exts(tp, n->handle);
1751                         if (unlikely(!exts || n->exts != exts)) {
1752                                 tcf_set_drop_reason(skb,
1753                                                     SKB_DROP_REASON_TC_COOKIE_ERROR);
1754                                 return TC_ACT_SHOT;
1755                         }
1756
1757                         n = NULL;
1758                         err = tcf_exts_exec_ex(skb, exts, act_index, res);
1759                 } else {
1760                         if (tp->protocol != protocol &&
1761                             tp->protocol != htons(ETH_P_ALL))
1762                                 continue;
1763
1764                         err = tc_classify(skb, tp, res);
1765                 }
1766 #ifdef CONFIG_NET_CLS_ACT
1767                 if (unlikely(err == TC_ACT_RECLASSIFY && !compat_mode)) {
1768                         first_tp = orig_tp;
1769                         *last_executed_chain = first_tp->chain->index;
1770                         goto reset;
1771                 } else if (unlikely(TC_ACT_EXT_CMP(err, TC_ACT_GOTO_CHAIN))) {
1772                         first_tp = res->goto_tp;
1773                         *last_executed_chain = err & TC_ACT_EXT_VAL_MASK;
1774                         goto reset;
1775                 }
1776 #endif
1777                 if (err >= 0)
1778                         return err;
1779         }
1780
1781         if (unlikely(n)) {
1782                 tcf_set_drop_reason(skb,
1783                                     SKB_DROP_REASON_TC_COOKIE_ERROR);
1784                 return TC_ACT_SHOT;
1785         }
1786
1787         return TC_ACT_UNSPEC; /* signal: continue lookup */
1788 #ifdef CONFIG_NET_CLS_ACT
1789 reset:
1790         if (unlikely(limit++ >= max_reclassify_loop)) {
1791                 net_notice_ratelimited("%u: reclassify loop, rule prio %u, protocol %02x\n",
1792                                        tp->chain->block->index,
1793                                        tp->prio & 0xffff,
1794                                        ntohs(tp->protocol));
1795                 tcf_set_drop_reason(skb,
1796                                     SKB_DROP_REASON_TC_RECLASSIFY_LOOP);
1797                 return TC_ACT_SHOT;
1798         }
1799
1800         tp = first_tp;
1801         goto reclassify;
1802 #endif
1803 }
1804
1805 int tcf_classify(struct sk_buff *skb,
1806                  const struct tcf_block *block,
1807                  const struct tcf_proto *tp,
1808                  struct tcf_result *res, bool compat_mode)
1809 {
1810 #if !IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
1811         u32 last_executed_chain = 0;
1812
1813         return __tcf_classify(skb, tp, tp, res, compat_mode, NULL, 0,
1814                               &last_executed_chain);
1815 #else
1816         u32 last_executed_chain = tp ? tp->chain->index : 0;
1817         struct tcf_exts_miss_cookie_node *n = NULL;
1818         const struct tcf_proto *orig_tp = tp;
1819         struct tc_skb_ext *ext;
1820         int act_index = 0;
1821         int ret;
1822
1823         if (block) {
1824                 ext = skb_ext_find(skb, TC_SKB_EXT);
1825
1826                 if (ext && (ext->chain || ext->act_miss)) {
1827                         struct tcf_chain *fchain;
1828                         u32 chain;
1829
1830                         if (ext->act_miss) {
1831                                 n = tcf_exts_miss_cookie_lookup(ext->act_miss_cookie,
1832                                                                 &act_index);
1833                                 if (!n) {
1834                                         tcf_set_drop_reason(skb,
1835                                                             SKB_DROP_REASON_TC_COOKIE_ERROR);
1836                                         return TC_ACT_SHOT;
1837                                 }
1838
1839                                 chain = n->chain_index;
1840                         } else {
1841                                 chain = ext->chain;
1842                         }
1843
1844                         fchain = tcf_chain_lookup_rcu(block, chain);
1845                         if (!fchain) {
1846                                 tcf_set_drop_reason(skb,
1847                                                     SKB_DROP_REASON_TC_CHAIN_NOTFOUND);
1848
1849                                 return TC_ACT_SHOT;
1850                         }
1851
1852                         /* Consume, so cloned/redirect skbs won't inherit ext */
1853                         skb_ext_del(skb, TC_SKB_EXT);
1854
1855                         tp = rcu_dereference_bh(fchain->filter_chain);
1856                         last_executed_chain = fchain->index;
1857                 }
1858         }
1859
1860         ret = __tcf_classify(skb, tp, orig_tp, res, compat_mode, n, act_index,
1861                              &last_executed_chain);
1862
1863         if (tc_skb_ext_tc_enabled()) {
1864                 /* If we missed on some chain */
1865                 if (ret == TC_ACT_UNSPEC && last_executed_chain) {
1866                         struct tc_skb_cb *cb = tc_skb_cb(skb);
1867
1868                         ext = tc_skb_ext_alloc(skb);
1869                         if (WARN_ON_ONCE(!ext)) {
1870                                 tcf_set_drop_reason(skb, SKB_DROP_REASON_NOMEM);
1871                                 return TC_ACT_SHOT;
1872                         }
1873                         ext->chain = last_executed_chain;
1874                         ext->mru = cb->mru;
1875                         ext->post_ct = cb->post_ct;
1876                         ext->post_ct_snat = cb->post_ct_snat;
1877                         ext->post_ct_dnat = cb->post_ct_dnat;
1878                         ext->zone = cb->zone;
1879                 }
1880         }
1881
1882         return ret;
1883 #endif
1884 }
1885 EXPORT_SYMBOL(tcf_classify);
1886
1887 struct tcf_chain_info {
1888         struct tcf_proto __rcu **pprev;
1889         struct tcf_proto __rcu *next;
1890 };
1891
1892 static struct tcf_proto *tcf_chain_tp_prev(struct tcf_chain *chain,
1893                                            struct tcf_chain_info *chain_info)
1894 {
1895         return tcf_chain_dereference(*chain_info->pprev, chain);
1896 }
1897
1898 static int tcf_chain_tp_insert(struct tcf_chain *chain,
1899                                struct tcf_chain_info *chain_info,
1900                                struct tcf_proto *tp)
1901 {
1902         if (chain->flushing)
1903                 return -EAGAIN;
1904
1905         RCU_INIT_POINTER(tp->next, tcf_chain_tp_prev(chain, chain_info));
1906         if (*chain_info->pprev == chain->filter_chain)
1907                 tcf_chain0_head_change(chain, tp);
1908         tcf_proto_get(tp);
1909         rcu_assign_pointer(*chain_info->pprev, tp);
1910
1911         return 0;
1912 }
1913
1914 static void tcf_chain_tp_remove(struct tcf_chain *chain,
1915                                 struct tcf_chain_info *chain_info,
1916                                 struct tcf_proto *tp)
1917 {
1918         struct tcf_proto *next = tcf_chain_dereference(chain_info->next, chain);
1919
1920         tcf_proto_mark_delete(tp);
1921         if (tp == chain->filter_chain)
1922                 tcf_chain0_head_change(chain, next);
1923         RCU_INIT_POINTER(*chain_info->pprev, next);
1924 }
1925
1926 static struct tcf_proto *tcf_chain_tp_find(struct tcf_chain *chain,
1927                                            struct tcf_chain_info *chain_info,
1928                                            u32 protocol, u32 prio,
1929                                            bool prio_allocate,
1930                                            struct netlink_ext_ack *extack);
1931
1932 /* Try to insert new proto.
1933  * If proto with specified priority already exists, free new proto
1934  * and return existing one.
1935  */
1936
1937 static struct tcf_proto *tcf_chain_tp_insert_unique(struct tcf_chain *chain,
1938                                                     struct tcf_proto *tp_new,
1939                                                     u32 protocol, u32 prio,
1940                                                     bool rtnl_held)
1941 {
1942         struct tcf_chain_info chain_info;
1943         struct tcf_proto *tp;
1944         int err = 0;
1945
1946         mutex_lock(&chain->filter_chain_lock);
1947
1948         if (tcf_proto_exists_destroying(chain, tp_new)) {
1949                 mutex_unlock(&chain->filter_chain_lock);
1950                 tcf_proto_destroy(tp_new, rtnl_held, false, NULL);
1951                 return ERR_PTR(-EAGAIN);
1952         }
1953
1954         tp = tcf_chain_tp_find(chain, &chain_info, protocol, prio, false, NULL);
1955         if (!tp)
1956                 err = tcf_chain_tp_insert(chain, &chain_info, tp_new);
1957         mutex_unlock(&chain->filter_chain_lock);
1958
1959         if (tp) {
1960                 tcf_proto_destroy(tp_new, rtnl_held, false, NULL);
1961                 tp_new = tp;
1962         } else if (err) {
1963                 tcf_proto_destroy(tp_new, rtnl_held, false, NULL);
1964                 tp_new = ERR_PTR(err);
1965         }
1966
1967         return tp_new;
1968 }
1969
1970 static void tcf_chain_tp_delete_empty(struct tcf_chain *chain,
1971                                       struct tcf_proto *tp, bool rtnl_held,
1972                                       struct netlink_ext_ack *extack)
1973 {
1974         struct tcf_chain_info chain_info;
1975         struct tcf_proto *tp_iter;
1976         struct tcf_proto **pprev;
1977         struct tcf_proto *next;
1978
1979         mutex_lock(&chain->filter_chain_lock);
1980
1981         /* Atomically find and remove tp from chain. */
1982         for (pprev = &chain->filter_chain;
1983              (tp_iter = tcf_chain_dereference(*pprev, chain));
1984              pprev = &tp_iter->next) {
1985                 if (tp_iter == tp) {
1986                         chain_info.pprev = pprev;
1987                         chain_info.next = tp_iter->next;
1988                         WARN_ON(tp_iter->deleting);
1989                         break;
1990                 }
1991         }
1992         /* Verify that tp still exists and no new filters were inserted
1993          * concurrently.
1994          * Mark tp for deletion if it is empty.
1995          */
1996         if (!tp_iter || !tcf_proto_check_delete(tp)) {
1997                 mutex_unlock(&chain->filter_chain_lock);
1998                 return;
1999         }
2000
2001         tcf_proto_signal_destroying(chain, tp);
2002         next = tcf_chain_dereference(chain_info.next, chain);
2003         if (tp == chain->filter_chain)
2004                 tcf_chain0_head_change(chain, next);
2005         RCU_INIT_POINTER(*chain_info.pprev, next);
2006         mutex_unlock(&chain->filter_chain_lock);
2007
2008         tcf_proto_put(tp, rtnl_held, extack);
2009 }
2010
2011 static struct tcf_proto *tcf_chain_tp_find(struct tcf_chain *chain,
2012                                            struct tcf_chain_info *chain_info,
2013                                            u32 protocol, u32 prio,
2014                                            bool prio_allocate,
2015                                            struct netlink_ext_ack *extack)
2016 {
2017         struct tcf_proto **pprev;
2018         struct tcf_proto *tp;
2019
2020         /* Check the chain for existence of proto-tcf with this priority */
2021         for (pprev = &chain->filter_chain;
2022              (tp = tcf_chain_dereference(*pprev, chain));
2023              pprev = &tp->next) {
2024                 if (tp->prio >= prio) {
2025                         if (tp->prio == prio) {
2026                                 if (prio_allocate) {
2027                                         NL_SET_ERR_MSG(extack, "Lowest ID from auto-alloc range already in use");
2028                                         return ERR_PTR(-ENOSPC);
2029                                 }
2030                                 if (tp->protocol != protocol && protocol) {
2031                                         NL_SET_ERR_MSG(extack, "Protocol mismatch for filter with specified priority");
2032                                         return ERR_PTR(-EINVAL);
2033                                 }
2034                         } else {
2035                                 tp = NULL;
2036                         }
2037                         break;
2038                 }
2039         }
2040         chain_info->pprev = pprev;
2041         if (tp) {
2042                 chain_info->next = tp->next;
2043                 tcf_proto_get(tp);
2044         } else {
2045                 chain_info->next = NULL;
2046         }
2047         return tp;
2048 }
2049
2050 static int tcf_fill_node(struct net *net, struct sk_buff *skb,
2051                          struct tcf_proto *tp, struct tcf_block *block,
2052                          struct Qdisc *q, u32 parent, void *fh,
2053                          u32 portid, u32 seq, u16 flags, int event,
2054                          bool terse_dump, bool rtnl_held,
2055                          struct netlink_ext_ack *extack)
2056 {
2057         struct tcmsg *tcm;
2058         struct nlmsghdr  *nlh;
2059         unsigned char *b = skb_tail_pointer(skb);
2060         int ret = -EMSGSIZE;
2061
2062         nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
2063         if (!nlh)
2064                 goto out_nlmsg_trim;
2065         tcm = nlmsg_data(nlh);
2066         tcm->tcm_family = AF_UNSPEC;
2067         tcm->tcm__pad1 = 0;
2068         tcm->tcm__pad2 = 0;
2069         if (q) {
2070                 tcm->tcm_ifindex = qdisc_dev(q)->ifindex;
2071                 tcm->tcm_parent = parent;
2072         } else {
2073                 tcm->tcm_ifindex = TCM_IFINDEX_MAGIC_BLOCK;
2074                 tcm->tcm_block_index = block->index;
2075         }
2076         tcm->tcm_info = TC_H_MAKE(tp->prio, tp->protocol);
2077         if (nla_put_string(skb, TCA_KIND, tp->ops->kind))
2078                 goto nla_put_failure;
2079         if (nla_put_u32(skb, TCA_CHAIN, tp->chain->index))
2080                 goto nla_put_failure;
2081         if (!fh) {
2082                 tcm->tcm_handle = 0;
2083         } else if (terse_dump) {
2084                 if (tp->ops->terse_dump) {
2085                         if (tp->ops->terse_dump(net, tp, fh, skb, tcm,
2086                                                 rtnl_held) < 0)
2087                                 goto nla_put_failure;
2088                 } else {
2089                         goto cls_op_not_supp;
2090                 }
2091         } else {
2092                 if (tp->ops->dump &&
2093                     tp->ops->dump(net, tp, fh, skb, tcm, rtnl_held) < 0)
2094                         goto nla_put_failure;
2095         }
2096
2097         if (extack && extack->_msg &&
2098             nla_put_string(skb, TCA_EXT_WARN_MSG, extack->_msg))
2099                 goto nla_put_failure;
2100
2101         nlh->nlmsg_len = skb_tail_pointer(skb) - b;
2102
2103         return skb->len;
2104
2105 cls_op_not_supp:
2106         ret = -EOPNOTSUPP;
2107 out_nlmsg_trim:
2108 nla_put_failure:
2109         nlmsg_trim(skb, b);
2110         return ret;
2111 }
2112
2113 static struct sk_buff *tfilter_notify_prep(struct net *net,
2114                                            struct sk_buff *oskb,
2115                                            struct nlmsghdr *n,
2116                                            struct tcf_proto *tp,
2117                                            struct tcf_block *block,
2118                                            struct Qdisc *q, u32 parent,
2119                                            void *fh, int event,
2120                                            u32 portid, bool rtnl_held,
2121                                            struct netlink_ext_ack *extack)
2122 {
2123         unsigned int size = oskb ? max(NLMSG_GOODSIZE, oskb->len) : NLMSG_GOODSIZE;
2124         struct sk_buff *skb;
2125         int ret;
2126
2127 retry:
2128         skb = alloc_skb(size, GFP_KERNEL);
2129         if (!skb)
2130                 return ERR_PTR(-ENOBUFS);
2131
2132         ret = tcf_fill_node(net, skb, tp, block, q, parent, fh, portid,
2133                             n->nlmsg_seq, n->nlmsg_flags, event, false,
2134                             rtnl_held, extack);
2135         if (ret <= 0) {
2136                 kfree_skb(skb);
2137                 if (ret == -EMSGSIZE) {
2138                         size += NLMSG_GOODSIZE;
2139                         goto retry;
2140                 }
2141                 return ERR_PTR(-EINVAL);
2142         }
2143         return skb;
2144 }
2145
2146 static int tfilter_notify(struct net *net, struct sk_buff *oskb,
2147                           struct nlmsghdr *n, struct tcf_proto *tp,
2148                           struct tcf_block *block, struct Qdisc *q,
2149                           u32 parent, void *fh, int event, bool unicast,
2150                           bool rtnl_held, struct netlink_ext_ack *extack)
2151 {
2152         struct sk_buff *skb;
2153         u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
2154         int err = 0;
2155
2156         if (!unicast && !rtnl_notify_needed(net, n->nlmsg_flags, RTNLGRP_TC))
2157                 return 0;
2158
2159         skb = tfilter_notify_prep(net, oskb, n, tp, block, q, parent, fh, event,
2160                                   portid, rtnl_held, extack);
2161         if (IS_ERR(skb))
2162                 return PTR_ERR(skb);
2163
2164         if (unicast)
2165                 err = rtnl_unicast(skb, net, portid);
2166         else
2167                 err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
2168                                      n->nlmsg_flags & NLM_F_ECHO);
2169         return err;
2170 }
2171
2172 static int tfilter_del_notify(struct net *net, struct sk_buff *oskb,
2173                               struct nlmsghdr *n, struct tcf_proto *tp,
2174                               struct tcf_block *block, struct Qdisc *q,
2175                               u32 parent, void *fh, bool *last, bool rtnl_held,
2176                               struct netlink_ext_ack *extack)
2177 {
2178         struct sk_buff *skb;
2179         u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
2180         int err;
2181
2182         if (!rtnl_notify_needed(net, n->nlmsg_flags, RTNLGRP_TC))
2183                 return tp->ops->delete(tp, fh, last, rtnl_held, extack);
2184
2185         skb = tfilter_notify_prep(net, oskb, n, tp, block, q, parent, fh,
2186                                   RTM_DELTFILTER, portid, rtnl_held, extack);
2187         if (IS_ERR(skb)) {
2188                 NL_SET_ERR_MSG(extack, "Failed to build del event notification");
2189                 return PTR_ERR(skb);
2190         }
2191
2192         err = tp->ops->delete(tp, fh, last, rtnl_held, extack);
2193         if (err) {
2194                 kfree_skb(skb);
2195                 return err;
2196         }
2197
2198         err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
2199                              n->nlmsg_flags & NLM_F_ECHO);
2200         if (err < 0)
2201                 NL_SET_ERR_MSG(extack, "Failed to send filter delete notification");
2202
2203         return err;
2204 }
2205
2206 static void tfilter_notify_chain(struct net *net, struct sk_buff *oskb,
2207                                  struct tcf_block *block, struct Qdisc *q,
2208                                  u32 parent, struct nlmsghdr *n,
2209                                  struct tcf_chain *chain, int event,
2210                                  struct netlink_ext_ack *extack)
2211 {
2212         struct tcf_proto *tp;
2213
2214         for (tp = tcf_get_next_proto(chain, NULL);
2215              tp; tp = tcf_get_next_proto(chain, tp))
2216                 tfilter_notify(net, oskb, n, tp, block, q, parent, NULL,
2217                                event, false, true, extack);
2218 }
2219
2220 static void tfilter_put(struct tcf_proto *tp, void *fh)
2221 {
2222         if (tp->ops->put && fh)
2223                 tp->ops->put(tp, fh);
2224 }
2225
2226 static bool is_qdisc_ingress(__u32 classid)
2227 {
2228         return (TC_H_MIN(classid) == TC_H_MIN(TC_H_MIN_INGRESS));
2229 }
2230
2231 static int tc_new_tfilter(struct sk_buff *skb, struct nlmsghdr *n,
2232                           struct netlink_ext_ack *extack)
2233 {
2234         struct net *net = sock_net(skb->sk);
2235         struct nlattr *tca[TCA_MAX + 1];
2236         char name[IFNAMSIZ];
2237         struct tcmsg *t;
2238         u32 protocol;
2239         u32 prio;
2240         bool prio_allocate;
2241         u32 parent;
2242         u32 chain_index;
2243         struct Qdisc *q;
2244         struct tcf_chain_info chain_info;
2245         struct tcf_chain *chain;
2246         struct tcf_block *block;
2247         struct tcf_proto *tp;
2248         unsigned long cl;
2249         void *fh;
2250         int err;
2251         int tp_created;
2252         bool rtnl_held = false;
2253         u32 flags;
2254
2255 replay:
2256         tp_created = 0;
2257
2258         err = nlmsg_parse_deprecated(n, sizeof(*t), tca, TCA_MAX,
2259                                      rtm_tca_policy, extack);
2260         if (err < 0)
2261                 return err;
2262
2263         t = nlmsg_data(n);
2264         protocol = TC_H_MIN(t->tcm_info);
2265         prio = TC_H_MAJ(t->tcm_info);
2266         prio_allocate = false;
2267         parent = t->tcm_parent;
2268         tp = NULL;
2269         cl = 0;
2270         block = NULL;
2271         q = NULL;
2272         chain = NULL;
2273         flags = 0;
2274
2275         if (prio == 0) {
2276                 /* If no priority is provided by the user,
2277                  * we allocate one.
2278                  */
2279                 if (n->nlmsg_flags & NLM_F_CREATE) {
2280                         prio = TC_H_MAKE(0x80000000U, 0U);
2281                         prio_allocate = true;
2282                 } else {
2283                         NL_SET_ERR_MSG(extack, "Invalid filter command with priority of zero");
2284                         return -ENOENT;
2285                 }
2286         }
2287
2288         /* Find head of filter chain. */
2289
2290         err = __tcf_qdisc_find(net, &q, &parent, t->tcm_ifindex, false, extack);
2291         if (err)
2292                 return err;
2293
2294         if (tcf_proto_check_kind(tca[TCA_KIND], name)) {
2295                 NL_SET_ERR_MSG(extack, "Specified TC filter name too long");
2296                 err = -EINVAL;
2297                 goto errout;
2298         }
2299
2300         /* Take rtnl mutex if rtnl_held was set to true on previous iteration,
2301          * block is shared (no qdisc found), qdisc is not unlocked, classifier
2302          * type is not specified, classifier is not unlocked.
2303          */
2304         if (rtnl_held ||
2305             (q && !(q->ops->cl_ops->flags & QDISC_CLASS_OPS_DOIT_UNLOCKED)) ||
2306             !tcf_proto_is_unlocked(name)) {
2307                 rtnl_held = true;
2308                 rtnl_lock();
2309         }
2310
2311         err = __tcf_qdisc_cl_find(q, parent, &cl, t->tcm_ifindex, extack);
2312         if (err)
2313                 goto errout;
2314
2315         block = __tcf_block_find(net, q, cl, t->tcm_ifindex, t->tcm_block_index,
2316                                  extack);
2317         if (IS_ERR(block)) {
2318                 err = PTR_ERR(block);
2319                 goto errout;
2320         }
2321         block->classid = parent;
2322
2323         chain_index = nla_get_u32_default(tca[TCA_CHAIN], 0);
2324         if (chain_index > TC_ACT_EXT_VAL_MASK) {
2325                 NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
2326                 err = -EINVAL;
2327                 goto errout;
2328         }
2329         chain = tcf_chain_get(block, chain_index, true);
2330         if (!chain) {
2331                 NL_SET_ERR_MSG(extack, "Cannot create specified filter chain");
2332                 err = -ENOMEM;
2333                 goto errout;
2334         }
2335
2336         mutex_lock(&chain->filter_chain_lock);
2337         tp = tcf_chain_tp_find(chain, &chain_info, protocol,
2338                                prio, prio_allocate, extack);
2339         if (IS_ERR(tp)) {
2340                 err = PTR_ERR(tp);
2341                 goto errout_locked;
2342         }
2343
2344         if (tp == NULL) {
2345                 struct tcf_proto *tp_new = NULL;
2346
2347                 if (chain->flushing) {
2348                         err = -EAGAIN;
2349                         goto errout_locked;
2350                 }
2351
2352                 /* Proto-tcf does not exist, create new one */
2353
2354                 if (tca[TCA_KIND] == NULL || !protocol) {
2355                         NL_SET_ERR_MSG(extack, "Filter kind and protocol must be specified");
2356                         err = -EINVAL;
2357                         goto errout_locked;
2358                 }
2359
2360                 if (!(n->nlmsg_flags & NLM_F_CREATE)) {
2361                         NL_SET_ERR_MSG(extack, "Need both RTM_NEWTFILTER and NLM_F_CREATE to create a new filter");
2362                         err = -ENOENT;
2363                         goto errout_locked;
2364                 }
2365
2366                 if (prio_allocate)
2367                         prio = tcf_auto_prio(tcf_chain_tp_prev(chain,
2368                                                                &chain_info));
2369
2370                 mutex_unlock(&chain->filter_chain_lock);
2371                 tp_new = tcf_proto_create(name, protocol, prio, chain,
2372                                           rtnl_held, extack);
2373                 if (IS_ERR(tp_new)) {
2374                         err = PTR_ERR(tp_new);
2375                         goto errout_tp;
2376                 }
2377
2378                 tp_created = 1;
2379                 tp = tcf_chain_tp_insert_unique(chain, tp_new, protocol, prio,
2380                                                 rtnl_held);
2381                 if (IS_ERR(tp)) {
2382                         err = PTR_ERR(tp);
2383                         goto errout_tp;
2384                 }
2385         } else {
2386                 mutex_unlock(&chain->filter_chain_lock);
2387         }
2388
2389         if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind)) {
2390                 NL_SET_ERR_MSG(extack, "Specified filter kind does not match existing one");
2391                 err = -EINVAL;
2392                 goto errout;
2393         }
2394
2395         fh = tp->ops->get(tp, t->tcm_handle);
2396
2397         if (!fh) {
2398                 if (!(n->nlmsg_flags & NLM_F_CREATE)) {
2399                         NL_SET_ERR_MSG(extack, "Need both RTM_NEWTFILTER and NLM_F_CREATE to create a new filter");
2400                         err = -ENOENT;
2401                         goto errout;
2402                 }
2403         } else if (n->nlmsg_flags & NLM_F_EXCL) {
2404                 tfilter_put(tp, fh);
2405                 NL_SET_ERR_MSG(extack, "Filter already exists");
2406                 err = -EEXIST;
2407                 goto errout;
2408         }
2409
2410         if (chain->tmplt_ops && chain->tmplt_ops != tp->ops) {
2411                 tfilter_put(tp, fh);
2412                 NL_SET_ERR_MSG(extack, "Chain template is set to a different filter kind");
2413                 err = -EINVAL;
2414                 goto errout;
2415         }
2416
2417         if (!(n->nlmsg_flags & NLM_F_CREATE))
2418                 flags |= TCA_ACT_FLAGS_REPLACE;
2419         if (!rtnl_held)
2420                 flags |= TCA_ACT_FLAGS_NO_RTNL;
2421         if (is_qdisc_ingress(parent))
2422                 flags |= TCA_ACT_FLAGS_AT_INGRESS;
2423         err = tp->ops->change(net, skb, tp, cl, t->tcm_handle, tca, &fh,
2424                               flags, extack);
2425         if (err == 0) {
2426                 tfilter_notify(net, skb, n, tp, block, q, parent, fh,
2427                                RTM_NEWTFILTER, false, rtnl_held, extack);
2428                 tfilter_put(tp, fh);
2429                 tcf_proto_count_usesw(tp, true);
2430                 /* q pointer is NULL for shared blocks */
2431                 if (q)
2432                         q->flags &= ~TCQ_F_CAN_BYPASS;
2433         }
2434
2435 errout:
2436         if (err && tp_created)
2437                 tcf_chain_tp_delete_empty(chain, tp, rtnl_held, NULL);
2438 errout_tp:
2439         if (chain) {
2440                 if (tp && !IS_ERR(tp))
2441                         tcf_proto_put(tp, rtnl_held, NULL);
2442                 if (!tp_created)
2443                         tcf_chain_put(chain);
2444         }
2445         tcf_block_release(q, block, rtnl_held);
2446
2447         if (rtnl_held)
2448                 rtnl_unlock();
2449
2450         if (err == -EAGAIN) {
2451                 /* Take rtnl lock in case EAGAIN is caused by concurrent flush
2452                  * of target chain.
2453                  */
2454                 rtnl_held = true;
2455                 /* Replay the request. */
2456                 goto replay;
2457         }
2458         return err;
2459
2460 errout_locked:
2461         mutex_unlock(&chain->filter_chain_lock);
2462         goto errout;
2463 }
2464
2465 static int tc_del_tfilter(struct sk_buff *skb, struct nlmsghdr *n,
2466                           struct netlink_ext_ack *extack)
2467 {
2468         struct net *net = sock_net(skb->sk);
2469         struct nlattr *tca[TCA_MAX + 1];
2470         char name[IFNAMSIZ];
2471         struct tcmsg *t;
2472         u32 protocol;
2473         u32 prio;
2474         u32 parent;
2475         u32 chain_index;
2476         struct Qdisc *q = NULL;
2477         struct tcf_chain_info chain_info;
2478         struct tcf_chain *chain = NULL;
2479         struct tcf_block *block = NULL;
2480         struct tcf_proto *tp = NULL;
2481         unsigned long cl = 0;
2482         void *fh = NULL;
2483         int err;
2484         bool rtnl_held = false;
2485
2486         err = nlmsg_parse_deprecated(n, sizeof(*t), tca, TCA_MAX,
2487                                      rtm_tca_policy, extack);
2488         if (err < 0)
2489                 return err;
2490
2491         t = nlmsg_data(n);
2492         protocol = TC_H_MIN(t->tcm_info);
2493         prio = TC_H_MAJ(t->tcm_info);
2494         parent = t->tcm_parent;
2495
2496         if (prio == 0 && (protocol || t->tcm_handle || tca[TCA_KIND])) {
2497                 NL_SET_ERR_MSG(extack, "Cannot flush filters with protocol, handle or kind set");
2498                 return -ENOENT;
2499         }
2500
2501         /* Find head of filter chain. */
2502
2503         err = __tcf_qdisc_find(net, &q, &parent, t->tcm_ifindex, false, extack);
2504         if (err)
2505                 return err;
2506
2507         if (tcf_proto_check_kind(tca[TCA_KIND], name)) {
2508                 NL_SET_ERR_MSG(extack, "Specified TC filter name too long");
2509                 err = -EINVAL;
2510                 goto errout;
2511         }
2512         /* Take rtnl mutex if flushing whole chain, block is shared (no qdisc
2513          * found), qdisc is not unlocked, classifier type is not specified,
2514          * classifier is not unlocked.
2515          */
2516         if (!prio ||
2517             (q && !(q->ops->cl_ops->flags & QDISC_CLASS_OPS_DOIT_UNLOCKED)) ||
2518             !tcf_proto_is_unlocked(name)) {
2519                 rtnl_held = true;
2520                 rtnl_lock();
2521         }
2522
2523         err = __tcf_qdisc_cl_find(q, parent, &cl, t->tcm_ifindex, extack);
2524         if (err)
2525                 goto errout;
2526
2527         block = __tcf_block_find(net, q, cl, t->tcm_ifindex, t->tcm_block_index,
2528                                  extack);
2529         if (IS_ERR(block)) {
2530                 err = PTR_ERR(block);
2531                 goto errout;
2532         }
2533
2534         chain_index = nla_get_u32_default(tca[TCA_CHAIN], 0);
2535         if (chain_index > TC_ACT_EXT_VAL_MASK) {
2536                 NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
2537                 err = -EINVAL;
2538                 goto errout;
2539         }
2540         chain = tcf_chain_get(block, chain_index, false);
2541         if (!chain) {
2542                 /* User requested flush on non-existent chain. Nothing to do,
2543                  * so just return success.
2544                  */
2545                 if (prio == 0) {
2546                         err = 0;
2547                         goto errout;
2548                 }
2549                 NL_SET_ERR_MSG(extack, "Cannot find specified filter chain");
2550                 err = -ENOENT;
2551                 goto errout;
2552         }
2553
2554         if (prio == 0) {
2555                 tfilter_notify_chain(net, skb, block, q, parent, n,
2556                                      chain, RTM_DELTFILTER, extack);
2557                 tcf_chain_flush(chain, rtnl_held);
2558                 err = 0;
2559                 goto errout;
2560         }
2561
2562         mutex_lock(&chain->filter_chain_lock);
2563         tp = tcf_chain_tp_find(chain, &chain_info, protocol,
2564                                prio, false, extack);
2565         if (!tp) {
2566                 err = -ENOENT;
2567                 NL_SET_ERR_MSG(extack, "Filter with specified priority/protocol not found");
2568                 goto errout_locked;
2569         } else if (IS_ERR(tp)) {
2570                 err = PTR_ERR(tp);
2571                 goto errout_locked;
2572         } else if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind)) {
2573                 NL_SET_ERR_MSG(extack, "Specified filter kind does not match existing one");
2574                 err = -EINVAL;
2575                 goto errout_locked;
2576         } else if (t->tcm_handle == 0) {
2577                 tcf_proto_signal_destroying(chain, tp);
2578                 tcf_chain_tp_remove(chain, &chain_info, tp);
2579                 mutex_unlock(&chain->filter_chain_lock);
2580
2581                 tcf_proto_put(tp, rtnl_held, NULL);
2582                 tfilter_notify(net, skb, n, tp, block, q, parent, fh,
2583                                RTM_DELTFILTER, false, rtnl_held, extack);
2584                 err = 0;
2585                 goto errout;
2586         }
2587         mutex_unlock(&chain->filter_chain_lock);
2588
2589         fh = tp->ops->get(tp, t->tcm_handle);
2590
2591         if (!fh) {
2592                 NL_SET_ERR_MSG(extack, "Specified filter handle not found");
2593                 err = -ENOENT;
2594         } else {
2595                 bool last;
2596
2597                 err = tfilter_del_notify(net, skb, n, tp, block, q, parent, fh,
2598                                          &last, rtnl_held, extack);
2599
2600                 if (err)
2601                         goto errout;
2602                 if (last)
2603                         tcf_chain_tp_delete_empty(chain, tp, rtnl_held, extack);
2604         }
2605
2606 errout:
2607         if (chain) {
2608                 if (tp && !IS_ERR(tp))
2609                         tcf_proto_put(tp, rtnl_held, NULL);
2610                 tcf_chain_put(chain);
2611         }
2612         tcf_block_release(q, block, rtnl_held);
2613
2614         if (rtnl_held)
2615                 rtnl_unlock();
2616
2617         return err;
2618
2619 errout_locked:
2620         mutex_unlock(&chain->filter_chain_lock);
2621         goto errout;
2622 }
2623
2624 static int tc_get_tfilter(struct sk_buff *skb, struct nlmsghdr *n,
2625                           struct netlink_ext_ack *extack)
2626 {
2627         struct net *net = sock_net(skb->sk);
2628         struct nlattr *tca[TCA_MAX + 1];
2629         char name[IFNAMSIZ];
2630         struct tcmsg *t;
2631         u32 protocol;
2632         u32 prio;
2633         u32 parent;
2634         u32 chain_index;
2635         struct Qdisc *q = NULL;
2636         struct tcf_chain_info chain_info;
2637         struct tcf_chain *chain = NULL;
2638         struct tcf_block *block = NULL;
2639         struct tcf_proto *tp = NULL;
2640         unsigned long cl = 0;
2641         void *fh = NULL;
2642         int err;
2643         bool rtnl_held = false;
2644
2645         err = nlmsg_parse_deprecated(n, sizeof(*t), tca, TCA_MAX,
2646                                      rtm_tca_policy, extack);
2647         if (err < 0)
2648                 return err;
2649
2650         t = nlmsg_data(n);
2651         protocol = TC_H_MIN(t->tcm_info);
2652         prio = TC_H_MAJ(t->tcm_info);
2653         parent = t->tcm_parent;
2654
2655         if (prio == 0) {
2656                 NL_SET_ERR_MSG(extack, "Invalid filter command with priority of zero");
2657                 return -ENOENT;
2658         }
2659
2660         /* Find head of filter chain. */
2661
2662         err = __tcf_qdisc_find(net, &q, &parent, t->tcm_ifindex, false, extack);
2663         if (err)
2664                 return err;
2665
2666         if (tcf_proto_check_kind(tca[TCA_KIND], name)) {
2667                 NL_SET_ERR_MSG(extack, "Specified TC filter name too long");
2668                 err = -EINVAL;
2669                 goto errout;
2670         }
2671         /* Take rtnl mutex if block is shared (no qdisc found), qdisc is not
2672          * unlocked, classifier type is not specified, classifier is not
2673          * unlocked.
2674          */
2675         if ((q && !(q->ops->cl_ops->flags & QDISC_CLASS_OPS_DOIT_UNLOCKED)) ||
2676             !tcf_proto_is_unlocked(name)) {
2677                 rtnl_held = true;
2678                 rtnl_lock();
2679         }
2680
2681         err = __tcf_qdisc_cl_find(q, parent, &cl, t->tcm_ifindex, extack);
2682         if (err)
2683                 goto errout;
2684
2685         block = __tcf_block_find(net, q, cl, t->tcm_ifindex, t->tcm_block_index,
2686                                  extack);
2687         if (IS_ERR(block)) {
2688                 err = PTR_ERR(block);
2689                 goto errout;
2690         }
2691
2692         chain_index = nla_get_u32_default(tca[TCA_CHAIN], 0);
2693         if (chain_index > TC_ACT_EXT_VAL_MASK) {
2694                 NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
2695                 err = -EINVAL;
2696                 goto errout;
2697         }
2698         chain = tcf_chain_get(block, chain_index, false);
2699         if (!chain) {
2700                 NL_SET_ERR_MSG(extack, "Cannot find specified filter chain");
2701                 err = -EINVAL;
2702                 goto errout;
2703         }
2704
2705         mutex_lock(&chain->filter_chain_lock);
2706         tp = tcf_chain_tp_find(chain, &chain_info, protocol,
2707                                prio, false, extack);
2708         mutex_unlock(&chain->filter_chain_lock);
2709         if (!tp) {
2710                 err = -ENOENT;
2711                 NL_SET_ERR_MSG(extack, "Filter with specified priority/protocol not found");
2712                 goto errout;
2713         } else if (IS_ERR(tp)) {
2714                 err = PTR_ERR(tp);
2715                 goto errout;
2716         } else if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind)) {
2717                 NL_SET_ERR_MSG(extack, "Specified filter kind does not match existing one");
2718                 err = -EINVAL;
2719                 goto errout;
2720         }
2721
2722         fh = tp->ops->get(tp, t->tcm_handle);
2723
2724         if (!fh) {
2725                 NL_SET_ERR_MSG(extack, "Specified filter handle not found");
2726                 err = -ENOENT;
2727         } else {
2728                 err = tfilter_notify(net, skb, n, tp, block, q, parent,
2729                                      fh, RTM_NEWTFILTER, true, rtnl_held, NULL);
2730                 if (err < 0)
2731                         NL_SET_ERR_MSG(extack, "Failed to send filter notify message");
2732         }
2733
2734         tfilter_put(tp, fh);
2735 errout:
2736         if (chain) {
2737                 if (tp && !IS_ERR(tp))
2738                         tcf_proto_put(tp, rtnl_held, NULL);
2739                 tcf_chain_put(chain);
2740         }
2741         tcf_block_release(q, block, rtnl_held);
2742
2743         if (rtnl_held)
2744                 rtnl_unlock();
2745
2746         return err;
2747 }
2748
2749 struct tcf_dump_args {
2750         struct tcf_walker w;
2751         struct sk_buff *skb;
2752         struct netlink_callback *cb;
2753         struct tcf_block *block;
2754         struct Qdisc *q;
2755         u32 parent;
2756         bool terse_dump;
2757 };
2758
2759 static int tcf_node_dump(struct tcf_proto *tp, void *n, struct tcf_walker *arg)
2760 {
2761         struct tcf_dump_args *a = (void *)arg;
2762         struct net *net = sock_net(a->skb->sk);
2763
2764         return tcf_fill_node(net, a->skb, tp, a->block, a->q, a->parent,
2765                              n, NETLINK_CB(a->cb->skb).portid,
2766                              a->cb->nlh->nlmsg_seq, NLM_F_MULTI,
2767                              RTM_NEWTFILTER, a->terse_dump, true, NULL);
2768 }
2769
2770 static bool tcf_chain_dump(struct tcf_chain *chain, struct Qdisc *q, u32 parent,
2771                            struct sk_buff *skb, struct netlink_callback *cb,
2772                            long index_start, long *p_index, bool terse)
2773 {
2774         struct net *net = sock_net(skb->sk);
2775         struct tcf_block *block = chain->block;
2776         struct tcmsg *tcm = nlmsg_data(cb->nlh);
2777         struct tcf_proto *tp, *tp_prev;
2778         struct tcf_dump_args arg;
2779
2780         for (tp = __tcf_get_next_proto(chain, NULL);
2781              tp;
2782              tp_prev = tp,
2783                      tp = __tcf_get_next_proto(chain, tp),
2784                      tcf_proto_put(tp_prev, true, NULL),
2785                      (*p_index)++) {
2786                 if (*p_index < index_start)
2787                         continue;
2788                 if (TC_H_MAJ(tcm->tcm_info) &&
2789                     TC_H_MAJ(tcm->tcm_info) != tp->prio)
2790                         continue;
2791                 if (TC_H_MIN(tcm->tcm_info) &&
2792                     TC_H_MIN(tcm->tcm_info) != tp->protocol)
2793                         continue;
2794                 if (*p_index > index_start)
2795                         memset(&cb->args[1], 0,
2796                                sizeof(cb->args) - sizeof(cb->args[0]));
2797                 if (cb->args[1] == 0) {
2798                         if (tcf_fill_node(net, skb, tp, block, q, parent, NULL,
2799                                           NETLINK_CB(cb->skb).portid,
2800                                           cb->nlh->nlmsg_seq, NLM_F_MULTI,
2801                                           RTM_NEWTFILTER, false, true, NULL) <= 0)
2802                                 goto errout;
2803                         cb->args[1] = 1;
2804                 }
2805                 if (!tp->ops->walk)
2806                         continue;
2807                 arg.w.fn = tcf_node_dump;
2808                 arg.skb = skb;
2809                 arg.cb = cb;
2810                 arg.block = block;
2811                 arg.q = q;
2812                 arg.parent = parent;
2813                 arg.w.stop = 0;
2814                 arg.w.skip = cb->args[1] - 1;
2815                 arg.w.count = 0;
2816                 arg.w.cookie = cb->args[2];
2817                 arg.terse_dump = terse;
2818                 tp->ops->walk(tp, &arg.w, true);
2819                 cb->args[2] = arg.w.cookie;
2820                 cb->args[1] = arg.w.count + 1;
2821                 if (arg.w.stop)
2822                         goto errout;
2823         }
2824         return true;
2825
2826 errout:
2827         tcf_proto_put(tp, true, NULL);
2828         return false;
2829 }
2830
2831 static const struct nla_policy tcf_tfilter_dump_policy[TCA_MAX + 1] = {
2832         [TCA_CHAIN]      = { .type = NLA_U32 },
2833         [TCA_DUMP_FLAGS] = NLA_POLICY_BITFIELD32(TCA_DUMP_FLAGS_TERSE),
2834 };
2835
2836 /* called with RTNL */
2837 static int tc_dump_tfilter(struct sk_buff *skb, struct netlink_callback *cb)
2838 {
2839         struct tcf_chain *chain, *chain_prev;
2840         struct net *net = sock_net(skb->sk);
2841         struct nlattr *tca[TCA_MAX + 1];
2842         struct Qdisc *q = NULL;
2843         struct tcf_block *block;
2844         struct tcmsg *tcm = nlmsg_data(cb->nlh);
2845         bool terse_dump = false;
2846         long index_start;
2847         long index;
2848         u32 parent;
2849         int err;
2850
2851         if (nlmsg_len(cb->nlh) < sizeof(*tcm))
2852                 return skb->len;
2853
2854         err = nlmsg_parse_deprecated(cb->nlh, sizeof(*tcm), tca, TCA_MAX,
2855                                      tcf_tfilter_dump_policy, cb->extack);
2856         if (err)
2857                 return err;
2858
2859         if (tca[TCA_DUMP_FLAGS]) {
2860                 struct nla_bitfield32 flags =
2861                         nla_get_bitfield32(tca[TCA_DUMP_FLAGS]);
2862
2863                 terse_dump = flags.value & TCA_DUMP_FLAGS_TERSE;
2864         }
2865
2866         if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK) {
2867                 block = tcf_block_refcnt_get(net, tcm->tcm_block_index);
2868                 if (!block)
2869                         goto out;
2870                 /* If we work with block index, q is NULL and parent value
2871                  * will never be used in the following code. The check
2872                  * in tcf_fill_node prevents it. However, compiler does not
2873                  * see that far, so set parent to zero to silence the warning
2874                  * about parent being uninitialized.
2875                  */
2876                 parent = 0;
2877         } else {
2878                 const struct Qdisc_class_ops *cops;
2879                 struct net_device *dev;
2880                 unsigned long cl = 0;
2881
2882                 dev = __dev_get_by_index(net, tcm->tcm_ifindex);
2883                 if (!dev)
2884                         return skb->len;
2885
2886                 parent = tcm->tcm_parent;
2887                 if (!parent)
2888                         q = rtnl_dereference(dev->qdisc);
2889                 else
2890                         q = qdisc_lookup(dev, TC_H_MAJ(tcm->tcm_parent));
2891                 if (!q)
2892                         goto out;
2893                 cops = q->ops->cl_ops;
2894                 if (!cops)
2895                         goto out;
2896                 if (!cops->tcf_block)
2897                         goto out;
2898                 if (TC_H_MIN(tcm->tcm_parent)) {
2899                         cl = cops->find(q, tcm->tcm_parent);
2900                         if (cl == 0)
2901                                 goto out;
2902                 }
2903                 block = cops->tcf_block(q, cl, NULL);
2904                 if (!block)
2905                         goto out;
2906                 parent = block->classid;
2907                 if (tcf_block_shared(block))
2908                         q = NULL;
2909         }
2910
2911         index_start = cb->args[0];
2912         index = 0;
2913
2914         for (chain = __tcf_get_next_chain(block, NULL);
2915              chain;
2916              chain_prev = chain,
2917                      chain = __tcf_get_next_chain(block, chain),
2918                      tcf_chain_put(chain_prev)) {
2919                 if (tca[TCA_CHAIN] &&
2920                     nla_get_u32(tca[TCA_CHAIN]) != chain->index)
2921                         continue;
2922                 if (!tcf_chain_dump(chain, q, parent, skb, cb,
2923                                     index_start, &index, terse_dump)) {
2924                         tcf_chain_put(chain);
2925                         err = -EMSGSIZE;
2926                         break;
2927                 }
2928         }
2929
2930         if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK)
2931                 tcf_block_refcnt_put(block, true);
2932         cb->args[0] = index;
2933
2934 out:
2935         /* If we did no progress, the error (EMSGSIZE) is real */
2936         if (skb->len == 0 && err)
2937                 return err;
2938         return skb->len;
2939 }
2940
2941 static int tc_chain_fill_node(const struct tcf_proto_ops *tmplt_ops,
2942                               void *tmplt_priv, u32 chain_index,
2943                               struct net *net, struct sk_buff *skb,
2944                               struct tcf_block *block,
2945                               u32 portid, u32 seq, u16 flags, int event,
2946                               struct netlink_ext_ack *extack)
2947 {
2948         unsigned char *b = skb_tail_pointer(skb);
2949         const struct tcf_proto_ops *ops;
2950         struct nlmsghdr *nlh;
2951         struct tcmsg *tcm;
2952         void *priv;
2953
2954         ops = tmplt_ops;
2955         priv = tmplt_priv;
2956
2957         nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
2958         if (!nlh)
2959                 goto out_nlmsg_trim;
2960         tcm = nlmsg_data(nlh);
2961         tcm->tcm_family = AF_UNSPEC;
2962         tcm->tcm__pad1 = 0;
2963         tcm->tcm__pad2 = 0;
2964         tcm->tcm_handle = 0;
2965         if (block->q) {
2966                 tcm->tcm_ifindex = qdisc_dev(block->q)->ifindex;
2967                 tcm->tcm_parent = block->q->handle;
2968         } else {
2969                 tcm->tcm_ifindex = TCM_IFINDEX_MAGIC_BLOCK;
2970                 tcm->tcm_block_index = block->index;
2971         }
2972
2973         if (nla_put_u32(skb, TCA_CHAIN, chain_index))
2974                 goto nla_put_failure;
2975
2976         if (ops) {
2977                 if (nla_put_string(skb, TCA_KIND, ops->kind))
2978                         goto nla_put_failure;
2979                 if (ops->tmplt_dump(skb, net, priv) < 0)
2980                         goto nla_put_failure;
2981         }
2982
2983         if (extack && extack->_msg &&
2984             nla_put_string(skb, TCA_EXT_WARN_MSG, extack->_msg))
2985                 goto out_nlmsg_trim;
2986
2987         nlh->nlmsg_len = skb_tail_pointer(skb) - b;
2988
2989         return skb->len;
2990
2991 out_nlmsg_trim:
2992 nla_put_failure:
2993         nlmsg_trim(skb, b);
2994         return -EMSGSIZE;
2995 }
2996
2997 static int tc_chain_notify(struct tcf_chain *chain, struct sk_buff *oskb,
2998                            u32 seq, u16 flags, int event, bool unicast,
2999                            struct netlink_ext_ack *extack)
3000 {
3001         u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
3002         struct tcf_block *block = chain->block;
3003         struct net *net = block->net;
3004         struct sk_buff *skb;
3005         int err = 0;
3006
3007         if (!unicast && !rtnl_notify_needed(net, flags, RTNLGRP_TC))
3008                 return 0;
3009
3010         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
3011         if (!skb)
3012                 return -ENOBUFS;
3013
3014         if (tc_chain_fill_node(chain->tmplt_ops, chain->tmplt_priv,
3015                                chain->index, net, skb, block, portid,
3016                                seq, flags, event, extack) <= 0) {
3017                 kfree_skb(skb);
3018                 return -EINVAL;
3019         }
3020
3021         if (unicast)
3022                 err = rtnl_unicast(skb, net, portid);
3023         else
3024                 err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
3025                                      flags & NLM_F_ECHO);
3026
3027         return err;
3028 }
3029
3030 static int tc_chain_notify_delete(const struct tcf_proto_ops *tmplt_ops,
3031                                   void *tmplt_priv, u32 chain_index,
3032                                   struct tcf_block *block, struct sk_buff *oskb,
3033                                   u32 seq, u16 flags)
3034 {
3035         u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
3036         struct net *net = block->net;
3037         struct sk_buff *skb;
3038
3039         if (!rtnl_notify_needed(net, flags, RTNLGRP_TC))
3040                 return 0;
3041
3042         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
3043         if (!skb)
3044                 return -ENOBUFS;
3045
3046         if (tc_chain_fill_node(tmplt_ops, tmplt_priv, chain_index, net, skb,
3047                                block, portid, seq, flags, RTM_DELCHAIN, NULL) <= 0) {
3048                 kfree_skb(skb);
3049                 return -EINVAL;
3050         }
3051
3052         return rtnetlink_send(skb, net, portid, RTNLGRP_TC, flags & NLM_F_ECHO);
3053 }
3054
3055 static int tc_chain_tmplt_add(struct tcf_chain *chain, struct net *net,
3056                               struct nlattr **tca,
3057                               struct netlink_ext_ack *extack)
3058 {
3059         const struct tcf_proto_ops *ops;
3060         char name[IFNAMSIZ];
3061         void *tmplt_priv;
3062
3063         /* If kind is not set, user did not specify template. */
3064         if (!tca[TCA_KIND])
3065                 return 0;
3066
3067         if (tcf_proto_check_kind(tca[TCA_KIND], name)) {
3068                 NL_SET_ERR_MSG(extack, "Specified TC chain template name too long");
3069                 return -EINVAL;
3070         }
3071
3072         ops = tcf_proto_lookup_ops(name, true, extack);
3073         if (IS_ERR(ops))
3074                 return PTR_ERR(ops);
3075         if (!ops->tmplt_create || !ops->tmplt_destroy || !ops->tmplt_dump ||
3076             !ops->tmplt_reoffload) {
3077                 NL_SET_ERR_MSG(extack, "Chain templates are not supported with specified classifier");
3078                 module_put(ops->owner);
3079                 return -EOPNOTSUPP;
3080         }
3081
3082         tmplt_priv = ops->tmplt_create(net, chain, tca, extack);
3083         if (IS_ERR(tmplt_priv)) {
3084                 module_put(ops->owner);
3085                 return PTR_ERR(tmplt_priv);
3086         }
3087         chain->tmplt_ops = ops;
3088         chain->tmplt_priv = tmplt_priv;
3089         return 0;
3090 }
3091
3092 static void tc_chain_tmplt_del(const struct tcf_proto_ops *tmplt_ops,
3093                                void *tmplt_priv)
3094 {
3095         /* If template ops are set, no work to do for us. */
3096         if (!tmplt_ops)
3097                 return;
3098
3099         tmplt_ops->tmplt_destroy(tmplt_priv);
3100         module_put(tmplt_ops->owner);
3101 }
3102
3103 /* Add/delete/get a chain */
3104
3105 static int tc_ctl_chain(struct sk_buff *skb, struct nlmsghdr *n,
3106                         struct netlink_ext_ack *extack)
3107 {
3108         struct net *net = sock_net(skb->sk);
3109         struct nlattr *tca[TCA_MAX + 1];
3110         struct tcmsg *t;
3111         u32 parent;
3112         u32 chain_index;
3113         struct Qdisc *q;
3114         struct tcf_chain *chain;
3115         struct tcf_block *block;
3116         unsigned long cl;
3117         int err;
3118
3119 replay:
3120         q = NULL;
3121         err = nlmsg_parse_deprecated(n, sizeof(*t), tca, TCA_MAX,
3122                                      rtm_tca_policy, extack);
3123         if (err < 0)
3124                 return err;
3125
3126         t = nlmsg_data(n);
3127         parent = t->tcm_parent;
3128         cl = 0;
3129
3130         block = tcf_block_find(net, &q, &parent, &cl,
3131                                t->tcm_ifindex, t->tcm_block_index, extack);
3132         if (IS_ERR(block))
3133                 return PTR_ERR(block);
3134
3135         chain_index = nla_get_u32_default(tca[TCA_CHAIN], 0);
3136         if (chain_index > TC_ACT_EXT_VAL_MASK) {
3137                 NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
3138                 err = -EINVAL;
3139                 goto errout_block;
3140         }
3141
3142         mutex_lock(&block->lock);
3143         chain = tcf_chain_lookup(block, chain_index);
3144         if (n->nlmsg_type == RTM_NEWCHAIN) {
3145                 if (chain) {
3146                         if (tcf_chain_held_by_acts_only(chain)) {
3147                                 /* The chain exists only because there is
3148                                  * some action referencing it.
3149                                  */
3150                                 tcf_chain_hold(chain);
3151                         } else {
3152                                 NL_SET_ERR_MSG(extack, "Filter chain already exists");
3153                                 err = -EEXIST;
3154                                 goto errout_block_locked;
3155                         }
3156                 } else {
3157                         if (!(n->nlmsg_flags & NLM_F_CREATE)) {
3158                                 NL_SET_ERR_MSG(extack, "Need both RTM_NEWCHAIN and NLM_F_CREATE to create a new chain");
3159                                 err = -ENOENT;
3160                                 goto errout_block_locked;
3161                         }
3162                         chain = tcf_chain_create(block, chain_index);
3163                         if (!chain) {
3164                                 NL_SET_ERR_MSG(extack, "Failed to create filter chain");
3165                                 err = -ENOMEM;
3166                                 goto errout_block_locked;
3167                         }
3168                 }
3169         } else {
3170                 if (!chain || tcf_chain_held_by_acts_only(chain)) {
3171                         NL_SET_ERR_MSG(extack, "Cannot find specified filter chain");
3172                         err = -EINVAL;
3173                         goto errout_block_locked;
3174                 }
3175                 tcf_chain_hold(chain);
3176         }
3177
3178         if (n->nlmsg_type == RTM_NEWCHAIN) {
3179                 /* Modifying chain requires holding parent block lock. In case
3180                  * the chain was successfully added, take a reference to the
3181                  * chain. This ensures that an empty chain does not disappear at
3182                  * the end of this function.
3183                  */
3184                 tcf_chain_hold(chain);
3185                 chain->explicitly_created = true;
3186         }
3187         mutex_unlock(&block->lock);
3188
3189         switch (n->nlmsg_type) {
3190         case RTM_NEWCHAIN:
3191                 err = tc_chain_tmplt_add(chain, net, tca, extack);
3192                 if (err) {
3193                         tcf_chain_put_explicitly_created(chain);
3194                         goto errout;
3195                 }
3196
3197                 tc_chain_notify(chain, NULL, 0, NLM_F_CREATE | NLM_F_EXCL,
3198                                 RTM_NEWCHAIN, false, extack);
3199                 break;
3200         case RTM_DELCHAIN:
3201                 tfilter_notify_chain(net, skb, block, q, parent, n,
3202                                      chain, RTM_DELTFILTER, extack);
3203                 /* Flush the chain first as the user requested chain removal. */
3204                 tcf_chain_flush(chain, true);
3205                 /* In case the chain was successfully deleted, put a reference
3206                  * to the chain previously taken during addition.
3207                  */
3208                 tcf_chain_put_explicitly_created(chain);
3209                 break;
3210         case RTM_GETCHAIN:
3211                 err = tc_chain_notify(chain, skb, n->nlmsg_seq,
3212                                       n->nlmsg_flags, n->nlmsg_type, true, extack);
3213                 if (err < 0)
3214                         NL_SET_ERR_MSG(extack, "Failed to send chain notify message");
3215                 break;
3216         default:
3217                 err = -EOPNOTSUPP;
3218                 NL_SET_ERR_MSG(extack, "Unsupported message type");
3219                 goto errout;
3220         }
3221
3222 errout:
3223         tcf_chain_put(chain);
3224 errout_block:
3225         tcf_block_release(q, block, true);
3226         if (err == -EAGAIN)
3227                 /* Replay the request. */
3228                 goto replay;
3229         return err;
3230
3231 errout_block_locked:
3232         mutex_unlock(&block->lock);
3233         goto errout_block;
3234 }
3235
3236 /* called with RTNL */
3237 static int tc_dump_chain(struct sk_buff *skb, struct netlink_callback *cb)
3238 {
3239         struct net *net = sock_net(skb->sk);
3240         struct nlattr *tca[TCA_MAX + 1];
3241         struct Qdisc *q = NULL;
3242         struct tcf_block *block;
3243         struct tcmsg *tcm = nlmsg_data(cb->nlh);
3244         struct tcf_chain *chain;
3245         long index_start;
3246         long index;
3247         int err;
3248
3249         if (nlmsg_len(cb->nlh) < sizeof(*tcm))
3250                 return skb->len;
3251
3252         err = nlmsg_parse_deprecated(cb->nlh, sizeof(*tcm), tca, TCA_MAX,
3253                                      rtm_tca_policy, cb->extack);
3254         if (err)
3255                 return err;
3256
3257         if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK) {
3258                 block = tcf_block_refcnt_get(net, tcm->tcm_block_index);
3259                 if (!block)
3260                         goto out;
3261         } else {
3262                 const struct Qdisc_class_ops *cops;
3263                 struct net_device *dev;
3264                 unsigned long cl = 0;
3265
3266                 dev = __dev_get_by_index(net, tcm->tcm_ifindex);
3267                 if (!dev)
3268                         return skb->len;
3269
3270                 if (!tcm->tcm_parent)
3271                         q = rtnl_dereference(dev->qdisc);
3272                 else
3273                         q = qdisc_lookup(dev, TC_H_MAJ(tcm->tcm_parent));
3274
3275                 if (!q)
3276                         goto out;
3277                 cops = q->ops->cl_ops;
3278                 if (!cops)
3279                         goto out;
3280                 if (!cops->tcf_block)
3281                         goto out;
3282                 if (TC_H_MIN(tcm->tcm_parent)) {
3283                         cl = cops->find(q, tcm->tcm_parent);
3284                         if (cl == 0)
3285                                 goto out;
3286                 }
3287                 block = cops->tcf_block(q, cl, NULL);
3288                 if (!block)
3289                         goto out;
3290                 if (tcf_block_shared(block))
3291                         q = NULL;
3292         }
3293
3294         index_start = cb->args[0];
3295         index = 0;
3296
3297         mutex_lock(&block->lock);
3298         list_for_each_entry(chain, &block->chain_list, list) {
3299                 if ((tca[TCA_CHAIN] &&
3300                      nla_get_u32(tca[TCA_CHAIN]) != chain->index))
3301                         continue;
3302                 if (index < index_start) {
3303                         index++;
3304                         continue;
3305                 }
3306                 if (tcf_chain_held_by_acts_only(chain))
3307                         continue;
3308                 err = tc_chain_fill_node(chain->tmplt_ops, chain->tmplt_priv,
3309                                          chain->index, net, skb, block,
3310                                          NETLINK_CB(cb->skb).portid,
3311                                          cb->nlh->nlmsg_seq, NLM_F_MULTI,
3312                                          RTM_NEWCHAIN, NULL);
3313                 if (err <= 0)
3314                         break;
3315                 index++;
3316         }
3317         mutex_unlock(&block->lock);
3318
3319         if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK)
3320                 tcf_block_refcnt_put(block, true);
3321         cb->args[0] = index;
3322
3323 out:
3324         /* If we did no progress, the error (EMSGSIZE) is real */
3325         if (skb->len == 0 && err)
3326                 return err;
3327         return skb->len;
3328 }
3329
3330 int tcf_exts_init_ex(struct tcf_exts *exts, struct net *net, int action,
3331                      int police, struct tcf_proto *tp, u32 handle,
3332                      bool use_action_miss)
3333 {
3334         int err = 0;
3335
3336 #ifdef CONFIG_NET_CLS_ACT
3337         exts->type = 0;
3338         exts->nr_actions = 0;
3339         exts->miss_cookie_node = NULL;
3340         /* Note: we do not own yet a reference on net.
3341          * This reference might be taken later from tcf_exts_get_net().
3342          */
3343         exts->net = net;
3344         exts->actions = kcalloc(TCA_ACT_MAX_PRIO, sizeof(struct tc_action *),
3345                                 GFP_KERNEL);
3346         if (!exts->actions)
3347                 return -ENOMEM;
3348 #endif
3349
3350         exts->action = action;
3351         exts->police = police;
3352
3353         if (!use_action_miss)
3354                 return 0;
3355
3356         err = tcf_exts_miss_cookie_base_alloc(exts, tp, handle);
3357         if (err)
3358                 goto err_miss_alloc;
3359
3360         return 0;
3361
3362 err_miss_alloc:
3363         tcf_exts_destroy(exts);
3364 #ifdef CONFIG_NET_CLS_ACT
3365         exts->actions = NULL;
3366 #endif
3367         return err;
3368 }
3369 EXPORT_SYMBOL(tcf_exts_init_ex);
3370
3371 void tcf_exts_destroy(struct tcf_exts *exts)
3372 {
3373         tcf_exts_miss_cookie_base_destroy(exts);
3374
3375 #ifdef CONFIG_NET_CLS_ACT
3376         if (exts->actions) {
3377                 tcf_action_destroy(exts->actions, TCA_ACT_UNBIND);
3378                 kfree(exts->actions);
3379         }
3380         exts->nr_actions = 0;
3381 #endif
3382 }
3383 EXPORT_SYMBOL(tcf_exts_destroy);
3384
3385 int tcf_exts_validate_ex(struct net *net, struct tcf_proto *tp, struct nlattr **tb,
3386                          struct nlattr *rate_tlv, struct tcf_exts *exts,
3387                          u32 flags, u32 fl_flags, struct netlink_ext_ack *extack)
3388 {
3389 #ifdef CONFIG_NET_CLS_ACT
3390         {
3391                 int init_res[TCA_ACT_MAX_PRIO] = {};
3392                 struct tc_action *act;
3393                 size_t attr_size = 0;
3394
3395                 if (exts->police && tb[exts->police]) {
3396                         struct tc_action_ops *a_o;
3397
3398                         flags |= TCA_ACT_FLAGS_POLICE | TCA_ACT_FLAGS_BIND;
3399                         a_o = tc_action_load_ops(tb[exts->police], flags,
3400                                                  extack);
3401                         if (IS_ERR(a_o))
3402                                 return PTR_ERR(a_o);
3403                         act = tcf_action_init_1(net, tp, tb[exts->police],
3404                                                 rate_tlv, a_o, init_res, flags,
3405                                                 extack);
3406                         module_put(a_o->owner);
3407                         if (IS_ERR(act))
3408                                 return PTR_ERR(act);
3409
3410                         act->type = exts->type = TCA_OLD_COMPAT;
3411                         exts->actions[0] = act;
3412                         exts->nr_actions = 1;
3413                         tcf_idr_insert_many(exts->actions, init_res);
3414                 } else if (exts->action && tb[exts->action]) {
3415                         int err;
3416
3417                         flags |= TCA_ACT_FLAGS_BIND;
3418                         err = tcf_action_init(net, tp, tb[exts->action],
3419                                               rate_tlv, exts->actions, init_res,
3420                                               &attr_size, flags, fl_flags,
3421                                               extack);
3422                         if (err < 0)
3423                                 return err;
3424                         exts->nr_actions = err;
3425                 }
3426         }
3427 #else
3428         if ((exts->action && tb[exts->action]) ||
3429             (exts->police && tb[exts->police])) {
3430                 NL_SET_ERR_MSG(extack, "Classifier actions are not supported per compile options (CONFIG_NET_CLS_ACT)");
3431                 return -EOPNOTSUPP;
3432         }
3433 #endif
3434
3435         return 0;
3436 }
3437 EXPORT_SYMBOL(tcf_exts_validate_ex);
3438
3439 int tcf_exts_validate(struct net *net, struct tcf_proto *tp, struct nlattr **tb,
3440                       struct nlattr *rate_tlv, struct tcf_exts *exts,
3441                       u32 flags, struct netlink_ext_ack *extack)
3442 {
3443         return tcf_exts_validate_ex(net, tp, tb, rate_tlv, exts,
3444                                     flags, 0, extack);
3445 }
3446 EXPORT_SYMBOL(tcf_exts_validate);
3447
3448 void tcf_exts_change(struct tcf_exts *dst, struct tcf_exts *src)
3449 {
3450 #ifdef CONFIG_NET_CLS_ACT
3451         struct tcf_exts old = *dst;
3452
3453         *dst = *src;
3454         tcf_exts_destroy(&old);
3455 #endif
3456 }
3457 EXPORT_SYMBOL(tcf_exts_change);
3458
3459 #ifdef CONFIG_NET_CLS_ACT
3460 static struct tc_action *tcf_exts_first_act(struct tcf_exts *exts)
3461 {
3462         if (exts->nr_actions == 0)
3463                 return NULL;
3464         else
3465                 return exts->actions[0];
3466 }
3467 #endif
3468
3469 int tcf_exts_dump(struct sk_buff *skb, struct tcf_exts *exts)
3470 {
3471 #ifdef CONFIG_NET_CLS_ACT
3472         struct nlattr *nest;
3473
3474         if (exts->action && tcf_exts_has_actions(exts)) {
3475                 /*
3476                  * again for backward compatible mode - we want
3477                  * to work with both old and new modes of entering
3478                  * tc data even if iproute2  was newer - jhs
3479                  */
3480                 if (exts->type != TCA_OLD_COMPAT) {
3481                         nest = nla_nest_start_noflag(skb, exts->action);
3482                         if (nest == NULL)
3483                                 goto nla_put_failure;
3484
3485                         if (tcf_action_dump(skb, exts->actions, 0, 0, false)
3486                             < 0)
3487                                 goto nla_put_failure;
3488                         nla_nest_end(skb, nest);
3489                 } else if (exts->police) {
3490                         struct tc_action *act = tcf_exts_first_act(exts);
3491                         nest = nla_nest_start_noflag(skb, exts->police);
3492                         if (nest == NULL || !act)
3493                                 goto nla_put_failure;
3494                         if (tcf_action_dump_old(skb, act, 0, 0) < 0)
3495                                 goto nla_put_failure;
3496                         nla_nest_end(skb, nest);
3497                 }
3498         }
3499         return 0;
3500
3501 nla_put_failure:
3502         nla_nest_cancel(skb, nest);
3503         return -1;
3504 #else
3505         return 0;
3506 #endif
3507 }
3508 EXPORT_SYMBOL(tcf_exts_dump);
3509
3510 int tcf_exts_terse_dump(struct sk_buff *skb, struct tcf_exts *exts)
3511 {
3512 #ifdef CONFIG_NET_CLS_ACT
3513         struct nlattr *nest;
3514
3515         if (!exts->action || !tcf_exts_has_actions(exts))
3516                 return 0;
3517
3518         nest = nla_nest_start_noflag(skb, exts->action);
3519         if (!nest)
3520                 goto nla_put_failure;
3521
3522         if (tcf_action_dump(skb, exts->actions, 0, 0, true) < 0)
3523                 goto nla_put_failure;
3524         nla_nest_end(skb, nest);
3525         return 0;
3526
3527 nla_put_failure:
3528         nla_nest_cancel(skb, nest);
3529         return -1;
3530 #else
3531         return 0;
3532 #endif
3533 }
3534 EXPORT_SYMBOL(tcf_exts_terse_dump);
3535
3536 int tcf_exts_dump_stats(struct sk_buff *skb, struct tcf_exts *exts)
3537 {
3538 #ifdef CONFIG_NET_CLS_ACT
3539         struct tc_action *a = tcf_exts_first_act(exts);
3540         if (a != NULL && tcf_action_copy_stats(skb, a, 1) < 0)
3541                 return -1;
3542 #endif
3543         return 0;
3544 }
3545 EXPORT_SYMBOL(tcf_exts_dump_stats);
3546
3547 static void tcf_block_offload_inc(struct tcf_block *block, u32 *flags)
3548 {
3549         if (*flags & TCA_CLS_FLAGS_IN_HW)
3550                 return;
3551         *flags |= TCA_CLS_FLAGS_IN_HW;
3552         atomic_inc(&block->offloadcnt);
3553 }
3554
3555 static void tcf_block_offload_dec(struct tcf_block *block, u32 *flags)
3556 {
3557         if (!(*flags & TCA_CLS_FLAGS_IN_HW))
3558                 return;
3559         *flags &= ~TCA_CLS_FLAGS_IN_HW;
3560         atomic_dec(&block->offloadcnt);
3561 }
3562
3563 static void tc_cls_offload_cnt_update(struct tcf_block *block,
3564                                       struct tcf_proto *tp, u32 *cnt,
3565                                       u32 *flags, u32 diff, bool add)
3566 {
3567         lockdep_assert_held(&block->cb_lock);
3568
3569         spin_lock(&tp->lock);
3570         if (add) {
3571                 if (!*cnt)
3572                         tcf_block_offload_inc(block, flags);
3573                 *cnt += diff;
3574         } else {
3575                 *cnt -= diff;
3576                 if (!*cnt)
3577                         tcf_block_offload_dec(block, flags);
3578         }
3579         spin_unlock(&tp->lock);
3580 }
3581
3582 static void
3583 tc_cls_offload_cnt_reset(struct tcf_block *block, struct tcf_proto *tp,
3584                          u32 *cnt, u32 *flags)
3585 {
3586         lockdep_assert_held(&block->cb_lock);
3587
3588         spin_lock(&tp->lock);
3589         tcf_block_offload_dec(block, flags);
3590         *cnt = 0;
3591         spin_unlock(&tp->lock);
3592 }
3593
3594 static int
3595 __tc_setup_cb_call(struct tcf_block *block, enum tc_setup_type type,
3596                    void *type_data, bool err_stop)
3597 {
3598         struct flow_block_cb *block_cb;
3599         int ok_count = 0;
3600         int err;
3601
3602         list_for_each_entry(block_cb, &block->flow_block.cb_list, list) {
3603                 err = block_cb->cb(type, type_data, block_cb->cb_priv);
3604                 if (err) {
3605                         if (err_stop)
3606                                 return err;
3607                 } else {
3608                         ok_count++;
3609                 }
3610         }
3611         return ok_count;
3612 }
3613
3614 int tc_setup_cb_call(struct tcf_block *block, enum tc_setup_type type,
3615                      void *type_data, bool err_stop, bool rtnl_held)
3616 {
3617         bool take_rtnl = READ_ONCE(block->lockeddevcnt) && !rtnl_held;
3618         int ok_count;
3619
3620 retry:
3621         if (take_rtnl)
3622                 rtnl_lock();
3623         down_read(&block->cb_lock);
3624         /* Need to obtain rtnl lock if block is bound to devs that require it.
3625          * In block bind code cb_lock is obtained while holding rtnl, so we must
3626          * obtain the locks in same order here.
3627          */
3628         if (!rtnl_held && !take_rtnl && block->lockeddevcnt) {
3629                 up_read(&block->cb_lock);
3630                 take_rtnl = true;
3631                 goto retry;
3632         }
3633
3634         ok_count = __tc_setup_cb_call(block, type, type_data, err_stop);
3635
3636         up_read(&block->cb_lock);
3637         if (take_rtnl)
3638                 rtnl_unlock();
3639         return ok_count;
3640 }
3641 EXPORT_SYMBOL(tc_setup_cb_call);
3642
3643 /* Non-destructive filter add. If filter that wasn't already in hardware is
3644  * successfully offloaded, increment block offloads counter. On failure,
3645  * previously offloaded filter is considered to be intact and offloads counter
3646  * is not decremented.
3647  */
3648
3649 int tc_setup_cb_add(struct tcf_block *block, struct tcf_proto *tp,
3650                     enum tc_setup_type type, void *type_data, bool err_stop,
3651                     u32 *flags, unsigned int *in_hw_count, bool rtnl_held)
3652 {
3653         bool take_rtnl = READ_ONCE(block->lockeddevcnt) && !rtnl_held;
3654         int ok_count;
3655
3656 retry:
3657         if (take_rtnl)
3658                 rtnl_lock();
3659         down_read(&block->cb_lock);
3660         /* Need to obtain rtnl lock if block is bound to devs that require it.
3661          * In block bind code cb_lock is obtained while holding rtnl, so we must
3662          * obtain the locks in same order here.
3663          */
3664         if (!rtnl_held && !take_rtnl && block->lockeddevcnt) {
3665                 up_read(&block->cb_lock);
3666                 take_rtnl = true;
3667                 goto retry;
3668         }
3669
3670         /* Make sure all netdevs sharing this block are offload-capable. */
3671         if (block->nooffloaddevcnt && err_stop) {
3672                 ok_count = -EOPNOTSUPP;
3673                 goto err_unlock;
3674         }
3675
3676         ok_count = __tc_setup_cb_call(block, type, type_data, err_stop);
3677         if (ok_count < 0)
3678                 goto err_unlock;
3679
3680         if (tp->ops->hw_add)
3681                 tp->ops->hw_add(tp, type_data);
3682         if (ok_count > 0)
3683                 tc_cls_offload_cnt_update(block, tp, in_hw_count, flags,
3684                                           ok_count, true);
3685 err_unlock:
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_add);
3692
3693 /* Destructive filter replace. If filter that wasn't already in hardware is
3694  * successfully offloaded, increment block offload counter. On failure,
3695  * previously offloaded filter is considered to be destroyed and offload counter
3696  * is decremented.
3697  */
3698
3699 int tc_setup_cb_replace(struct tcf_block *block, struct tcf_proto *tp,
3700                         enum tc_setup_type type, void *type_data, bool err_stop,
3701                         u32 *old_flags, unsigned int *old_in_hw_count,
3702                         u32 *new_flags, unsigned int *new_in_hw_count,
3703                         bool rtnl_held)
3704 {
3705         bool take_rtnl = READ_ONCE(block->lockeddevcnt) && !rtnl_held;
3706         int ok_count;
3707
3708 retry:
3709         if (take_rtnl)
3710                 rtnl_lock();
3711         down_read(&block->cb_lock);
3712         /* Need to obtain rtnl lock if block is bound to devs that require it.
3713          * In block bind code cb_lock is obtained while holding rtnl, so we must
3714          * obtain the locks in same order here.
3715          */
3716         if (!rtnl_held && !take_rtnl && block->lockeddevcnt) {
3717                 up_read(&block->cb_lock);
3718                 take_rtnl = true;
3719                 goto retry;
3720         }
3721
3722         /* Make sure all netdevs sharing this block are offload-capable. */
3723         if (block->nooffloaddevcnt && err_stop) {
3724                 ok_count = -EOPNOTSUPP;
3725                 goto err_unlock;
3726         }
3727
3728         tc_cls_offload_cnt_reset(block, tp, old_in_hw_count, old_flags);
3729         if (tp->ops->hw_del)
3730                 tp->ops->hw_del(tp, type_data);
3731
3732         ok_count = __tc_setup_cb_call(block, type, type_data, err_stop);
3733         if (ok_count < 0)
3734                 goto err_unlock;
3735
3736         if (tp->ops->hw_add)
3737                 tp->ops->hw_add(tp, type_data);
3738         if (ok_count > 0)
3739                 tc_cls_offload_cnt_update(block, tp, new_in_hw_count,
3740                                           new_flags, ok_count, true);
3741 err_unlock:
3742         up_read(&block->cb_lock);
3743         if (take_rtnl)
3744                 rtnl_unlock();
3745         return min(ok_count, 0);
3746 }
3747 EXPORT_SYMBOL(tc_setup_cb_replace);
3748
3749 /* Destroy filter and decrement block offload counter, if filter was previously
3750  * offloaded.
3751  */
3752
3753 int tc_setup_cb_destroy(struct tcf_block *block, struct tcf_proto *tp,
3754                         enum tc_setup_type type, void *type_data, bool err_stop,
3755                         u32 *flags, unsigned int *in_hw_count, bool rtnl_held)
3756 {
3757         bool take_rtnl = READ_ONCE(block->lockeddevcnt) && !rtnl_held;
3758         int ok_count;
3759
3760 retry:
3761         if (take_rtnl)
3762                 rtnl_lock();
3763         down_read(&block->cb_lock);
3764         /* Need to obtain rtnl lock if block is bound to devs that require it.
3765          * In block bind code cb_lock is obtained while holding rtnl, so we must
3766          * obtain the locks in same order here.
3767          */
3768         if (!rtnl_held && !take_rtnl && block->lockeddevcnt) {
3769                 up_read(&block->cb_lock);
3770                 take_rtnl = true;
3771                 goto retry;
3772         }
3773
3774         ok_count = __tc_setup_cb_call(block, type, type_data, err_stop);
3775
3776         tc_cls_offload_cnt_reset(block, tp, in_hw_count, flags);
3777         if (tp->ops->hw_del)
3778                 tp->ops->hw_del(tp, type_data);
3779
3780         up_read(&block->cb_lock);
3781         if (take_rtnl)
3782                 rtnl_unlock();
3783         return min(ok_count, 0);
3784 }
3785 EXPORT_SYMBOL(tc_setup_cb_destroy);
3786
3787 int tc_setup_cb_reoffload(struct tcf_block *block, struct tcf_proto *tp,
3788                           bool add, flow_setup_cb_t *cb,
3789                           enum tc_setup_type type, void *type_data,
3790                           void *cb_priv, u32 *flags, unsigned int *in_hw_count)
3791 {
3792         int err = cb(type, type_data, cb_priv);
3793
3794         if (err) {
3795                 if (add && tc_skip_sw(*flags))
3796                         return err;
3797         } else {
3798                 tc_cls_offload_cnt_update(block, tp, in_hw_count, flags, 1,
3799                                           add);
3800         }
3801
3802         return 0;
3803 }
3804 EXPORT_SYMBOL(tc_setup_cb_reoffload);
3805
3806 static int tcf_act_get_user_cookie(struct flow_action_entry *entry,
3807                                    const struct tc_action *act)
3808 {
3809         struct tc_cookie *user_cookie;
3810         int err = 0;
3811
3812         rcu_read_lock();
3813         user_cookie = rcu_dereference(act->user_cookie);
3814         if (user_cookie) {
3815                 entry->user_cookie = flow_action_cookie_create(user_cookie->data,
3816                                                                user_cookie->len,
3817                                                                GFP_ATOMIC);
3818                 if (!entry->user_cookie)
3819                         err = -ENOMEM;
3820         }
3821         rcu_read_unlock();
3822         return err;
3823 }
3824
3825 static void tcf_act_put_user_cookie(struct flow_action_entry *entry)
3826 {
3827         flow_action_cookie_destroy(entry->user_cookie);
3828 }
3829
3830 void tc_cleanup_offload_action(struct flow_action *flow_action)
3831 {
3832         struct flow_action_entry *entry;
3833         int i;
3834
3835         flow_action_for_each(i, entry, flow_action) {
3836                 tcf_act_put_user_cookie(entry);
3837                 if (entry->destructor)
3838                         entry->destructor(entry->destructor_priv);
3839         }
3840 }
3841 EXPORT_SYMBOL(tc_cleanup_offload_action);
3842
3843 static int tc_setup_offload_act(struct tc_action *act,
3844                                 struct flow_action_entry *entry,
3845                                 u32 *index_inc,
3846                                 struct netlink_ext_ack *extack)
3847 {
3848 #ifdef CONFIG_NET_CLS_ACT
3849         if (act->ops->offload_act_setup) {
3850                 return act->ops->offload_act_setup(act, entry, index_inc, true,
3851                                                    extack);
3852         } else {
3853                 NL_SET_ERR_MSG(extack, "Action does not support offload");
3854                 return -EOPNOTSUPP;
3855         }
3856 #else
3857         return 0;
3858 #endif
3859 }
3860
3861 int tc_setup_action(struct flow_action *flow_action,
3862                     struct tc_action *actions[],
3863                     u32 miss_cookie_base,
3864                     struct netlink_ext_ack *extack)
3865 {
3866         int i, j, k, index, err = 0;
3867         struct tc_action *act;
3868
3869         BUILD_BUG_ON(TCA_ACT_HW_STATS_ANY != FLOW_ACTION_HW_STATS_ANY);
3870         BUILD_BUG_ON(TCA_ACT_HW_STATS_IMMEDIATE != FLOW_ACTION_HW_STATS_IMMEDIATE);
3871         BUILD_BUG_ON(TCA_ACT_HW_STATS_DELAYED != FLOW_ACTION_HW_STATS_DELAYED);
3872
3873         if (!actions)
3874                 return 0;
3875
3876         j = 0;
3877         tcf_act_for_each_action(i, act, actions) {
3878                 struct flow_action_entry *entry;
3879
3880                 entry = &flow_action->entries[j];
3881                 spin_lock_bh(&act->tcfa_lock);
3882                 err = tcf_act_get_user_cookie(entry, act);
3883                 if (err)
3884                         goto err_out_locked;
3885
3886                 index = 0;
3887                 err = tc_setup_offload_act(act, entry, &index, extack);
3888                 if (err)
3889                         goto err_out_locked;
3890
3891                 for (k = 0; k < index ; k++) {
3892                         entry[k].hw_stats = tc_act_hw_stats(act->hw_stats);
3893                         entry[k].hw_index = act->tcfa_index;
3894                         entry[k].cookie = (unsigned long)act;
3895                         entry[k].miss_cookie =
3896                                 tcf_exts_miss_cookie_get(miss_cookie_base, i);
3897                 }
3898
3899                 j += index;
3900
3901                 spin_unlock_bh(&act->tcfa_lock);
3902         }
3903
3904 err_out:
3905         if (err)
3906                 tc_cleanup_offload_action(flow_action);
3907
3908         return err;
3909 err_out_locked:
3910         spin_unlock_bh(&act->tcfa_lock);
3911         goto err_out;
3912 }
3913
3914 int tc_setup_offload_action(struct flow_action *flow_action,
3915                             const struct tcf_exts *exts,
3916                             struct netlink_ext_ack *extack)
3917 {
3918 #ifdef CONFIG_NET_CLS_ACT
3919         u32 miss_cookie_base;
3920
3921         if (!exts)
3922                 return 0;
3923
3924         miss_cookie_base = exts->miss_cookie_node ?
3925                            exts->miss_cookie_node->miss_cookie_base : 0;
3926         return tc_setup_action(flow_action, exts->actions, miss_cookie_base,
3927                                extack);
3928 #else
3929         return 0;
3930 #endif
3931 }
3932 EXPORT_SYMBOL(tc_setup_offload_action);
3933
3934 unsigned int tcf_exts_num_actions(struct tcf_exts *exts)
3935 {
3936         unsigned int num_acts = 0;
3937         struct tc_action *act;
3938         int i;
3939
3940         tcf_exts_for_each_action(i, act, exts) {
3941                 if (is_tcf_pedit(act))
3942                         num_acts += tcf_pedit_nkeys(act);
3943                 else
3944                         num_acts++;
3945         }
3946         return num_acts;
3947 }
3948 EXPORT_SYMBOL(tcf_exts_num_actions);
3949
3950 #ifdef CONFIG_NET_CLS_ACT
3951 static int tcf_qevent_parse_block_index(struct nlattr *block_index_attr,
3952                                         u32 *p_block_index,
3953                                         struct netlink_ext_ack *extack)
3954 {
3955         *p_block_index = nla_get_u32(block_index_attr);
3956         if (!*p_block_index) {
3957                 NL_SET_ERR_MSG(extack, "Block number may not be zero");
3958                 return -EINVAL;
3959         }
3960
3961         return 0;
3962 }
3963
3964 int tcf_qevent_init(struct tcf_qevent *qe, struct Qdisc *sch,
3965                     enum flow_block_binder_type binder_type,
3966                     struct nlattr *block_index_attr,
3967                     struct netlink_ext_ack *extack)
3968 {
3969         u32 block_index;
3970         int err;
3971
3972         if (!block_index_attr)
3973                 return 0;
3974
3975         err = tcf_qevent_parse_block_index(block_index_attr, &block_index, extack);
3976         if (err)
3977                 return err;
3978
3979         qe->info.binder_type = binder_type;
3980         qe->info.chain_head_change = tcf_chain_head_change_dflt;
3981         qe->info.chain_head_change_priv = &qe->filter_chain;
3982         qe->info.block_index = block_index;
3983
3984         return tcf_block_get_ext(&qe->block, sch, &qe->info, extack);
3985 }
3986 EXPORT_SYMBOL(tcf_qevent_init);
3987
3988 void tcf_qevent_destroy(struct tcf_qevent *qe, struct Qdisc *sch)
3989 {
3990         if (qe->info.block_index)
3991                 tcf_block_put_ext(qe->block, sch, &qe->info);
3992 }
3993 EXPORT_SYMBOL(tcf_qevent_destroy);
3994
3995 int tcf_qevent_validate_change(struct tcf_qevent *qe, struct nlattr *block_index_attr,
3996                                struct netlink_ext_ack *extack)
3997 {
3998         u32 block_index;
3999         int err;
4000
4001         if (!block_index_attr)
4002                 return 0;
4003
4004         err = tcf_qevent_parse_block_index(block_index_attr, &block_index, extack);
4005         if (err)
4006                 return err;
4007
4008         /* Bounce newly-configured block or change in block. */
4009         if (block_index != qe->info.block_index) {
4010                 NL_SET_ERR_MSG(extack, "Change of blocks is not supported");
4011                 return -EINVAL;
4012         }
4013
4014         return 0;
4015 }
4016 EXPORT_SYMBOL(tcf_qevent_validate_change);
4017
4018 struct sk_buff *tcf_qevent_handle(struct tcf_qevent *qe, struct Qdisc *sch, struct sk_buff *skb,
4019                                   struct sk_buff **to_free, int *ret)
4020 {
4021         struct tcf_result cl_res;
4022         struct tcf_proto *fl;
4023
4024         if (!qe->info.block_index)
4025                 return skb;
4026
4027         fl = rcu_dereference_bh(qe->filter_chain);
4028
4029         switch (tcf_classify(skb, NULL, fl, &cl_res, false)) {
4030         case TC_ACT_SHOT:
4031                 qdisc_qstats_drop(sch);
4032                 __qdisc_drop(skb, to_free);
4033                 *ret = __NET_XMIT_BYPASS;
4034                 return NULL;
4035         case TC_ACT_STOLEN:
4036         case TC_ACT_QUEUED:
4037         case TC_ACT_TRAP:
4038                 __qdisc_drop(skb, to_free);
4039                 *ret = __NET_XMIT_STOLEN;
4040                 return NULL;
4041         case TC_ACT_REDIRECT:
4042                 skb_do_redirect(skb);
4043                 *ret = __NET_XMIT_STOLEN;
4044                 return NULL;
4045         }
4046
4047         return skb;
4048 }
4049 EXPORT_SYMBOL(tcf_qevent_handle);
4050
4051 int tcf_qevent_dump(struct sk_buff *skb, int attr_name, struct tcf_qevent *qe)
4052 {
4053         if (!qe->info.block_index)
4054                 return 0;
4055         return nla_put_u32(skb, attr_name, qe->info.block_index);
4056 }
4057 EXPORT_SYMBOL(tcf_qevent_dump);
4058 #endif
4059
4060 static __net_init int tcf_net_init(struct net *net)
4061 {
4062         struct tcf_net *tn = net_generic(net, tcf_net_id);
4063
4064         spin_lock_init(&tn->idr_lock);
4065         idr_init(&tn->idr);
4066         return 0;
4067 }
4068
4069 static void __net_exit tcf_net_exit(struct net *net)
4070 {
4071         struct tcf_net *tn = net_generic(net, tcf_net_id);
4072
4073         idr_destroy(&tn->idr);
4074 }
4075
4076 static struct pernet_operations tcf_net_ops = {
4077         .init = tcf_net_init,
4078         .exit = tcf_net_exit,
4079         .id   = &tcf_net_id,
4080         .size = sizeof(struct tcf_net),
4081 };
4082
4083 static const struct rtnl_msg_handler tc_filter_rtnl_msg_handlers[] __initconst = {
4084         {.msgtype = RTM_NEWTFILTER, .doit = tc_new_tfilter,
4085          .flags = RTNL_FLAG_DOIT_UNLOCKED},
4086         {.msgtype = RTM_DELTFILTER, .doit = tc_del_tfilter,
4087          .flags = RTNL_FLAG_DOIT_UNLOCKED},
4088         {.msgtype = RTM_GETTFILTER, .doit = tc_get_tfilter,
4089          .dumpit = tc_dump_tfilter, .flags = RTNL_FLAG_DOIT_UNLOCKED},
4090         {.msgtype = RTM_NEWCHAIN, .doit = tc_ctl_chain},
4091         {.msgtype = RTM_DELCHAIN, .doit = tc_ctl_chain},
4092         {.msgtype = RTM_GETCHAIN, .doit = tc_ctl_chain,
4093          .dumpit = tc_dump_chain},
4094 };
4095
4096 static int __init tc_filter_init(void)
4097 {
4098         int err;
4099
4100         tc_filter_wq = alloc_ordered_workqueue("tc_filter_workqueue", 0);
4101         if (!tc_filter_wq)
4102                 return -ENOMEM;
4103
4104         err = register_pernet_subsys(&tcf_net_ops);
4105         if (err)
4106                 goto err_register_pernet_subsys;
4107
4108         xa_init_flags(&tcf_exts_miss_cookies_xa, XA_FLAGS_ALLOC1);
4109         rtnl_register_many(tc_filter_rtnl_msg_handlers);
4110
4111         return 0;
4112
4113 err_register_pernet_subsys:
4114         destroy_workqueue(tc_filter_wq);
4115         return err;
4116 }
4117
4118 subsys_initcall(tc_filter_init);