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