net/sched/act_api.c

   1 // SPDX-License-Identifier: GPL-2.0-or-later
   2 /*
   3  * net/sched/act_api.c  Packet action API.
   4  *
   5  * Author:      Jamal Hadi Salim
   6  */
   7
   8 #include <linux/types.h>
   9 #include <linux/kernel.h>
  10 #include <linux/string.h>
  11 #include <linux/errno.h>
  12 #include <linux/slab.h>
  13 #include <linux/skbuff.h>
  14 #include <linux/init.h>
  15 #include <linux/kmod.h>
  16 #include <linux/err.h>
  17 #include <linux/module.h>
  18 #include <net/net_namespace.h>
  19 #include <net/sock.h>
  20 #include <net/sch_generic.h>
  21 #include <net/pkt_cls.h>
  22 #include <net/act_api.h>
  23 #include <net/netlink.h>
  24
  25 #ifdef CONFIG_INET
  26 DEFINE_STATIC_KEY_FALSE(tcf_frag_xmit_count);
  27 EXPORT_SYMBOL_GPL(tcf_frag_xmit_count);
  28 #endif
  29
  30 int tcf_dev_queue_xmit(struct sk_buff *skb, int (*xmit)(struct sk_buff *skb))
  31 {
  32 #ifdef CONFIG_INET
  33         if (static_branch_unlikely(&tcf_frag_xmit_count))
  34                 return sch_frag_xmit_hook(skb, xmit);
  35 #endif
  36
  37         return xmit(skb);
  38 }
  39 EXPORT_SYMBOL_GPL(tcf_dev_queue_xmit);
  40
  41 static void tcf_action_goto_chain_exec(const struct tc_action *a,
  42                                        struct tcf_result *res)
  43 {
  44         const struct tcf_chain *chain = rcu_dereference_bh(a->goto_chain);
  45
  46         res->goto_tp = rcu_dereference_bh(chain->filter_chain);
  47 }
  48
  49 static void tcf_free_cookie_rcu(struct rcu_head *p)
  50 {
  51         struct tc_cookie *cookie = container_of(p, struct tc_cookie, rcu);
  52
  53         kfree(cookie->data);
  54         kfree(cookie);
  55 }
  56
  57 static void tcf_set_action_cookie(struct tc_cookie __rcu **old_cookie,
  58                                   struct tc_cookie *new_cookie)
  59 {
  60         struct tc_cookie *old;
  61
  62         old = xchg((__force struct tc_cookie **)old_cookie, new_cookie);
  63         if (old)
  64                 call_rcu(&old->rcu, tcf_free_cookie_rcu);
  65 }
  66
  67 int tcf_action_check_ctrlact(int action, struct tcf_proto *tp,
  68                              struct tcf_chain **newchain,
  69                              struct netlink_ext_ack *extack)
  70 {
  71         int opcode = TC_ACT_EXT_OPCODE(action), ret = -EINVAL;
  72         u32 chain_index;
  73
  74         if (!opcode)
  75                 ret = action > TC_ACT_VALUE_MAX ? -EINVAL : 0;
  76         else if (opcode <= TC_ACT_EXT_OPCODE_MAX || action == TC_ACT_UNSPEC)
  77                 ret = 0;
  78         if (ret) {
  79                 NL_SET_ERR_MSG(extack, "invalid control action");
  80                 goto end;
  81         }
  82
  83         if (TC_ACT_EXT_CMP(action, TC_ACT_GOTO_CHAIN)) {
  84                 chain_index = action & TC_ACT_EXT_VAL_MASK;
  85                 if (!tp || !newchain) {
  86                         ret = -EINVAL;
  87                         NL_SET_ERR_MSG(extack,
  88                                        "can't goto NULL proto/chain");
  89                         goto end;
  90                 }
  91                 *newchain = tcf_chain_get_by_act(tp->chain->block, chain_index);
  92                 if (!*newchain) {
  93                         ret = -ENOMEM;
  94                         NL_SET_ERR_MSG(extack,
  95                                        "can't allocate goto_chain");
  96                 }
  97         }
  98 end:
  99         return ret;
 100 }
 101 EXPORT_SYMBOL(tcf_action_check_ctrlact);
 102
 103 struct tcf_chain *tcf_action_set_ctrlact(struct tc_action *a, int action,
 104                                          struct tcf_chain *goto_chain)
 105 {
 106         a->tcfa_action = action;
 107         goto_chain = rcu_replace_pointer(a->goto_chain, goto_chain, 1);
 108         return goto_chain;
 109 }
 110 EXPORT_SYMBOL(tcf_action_set_ctrlact);
 111
 112 /* XXX: For standalone actions, we don't need a RCU grace period either, because
 113  * actions are always connected to filters and filters are already destroyed in
 114  * RCU callbacks, so after a RCU grace period actions are already disconnected
 115  * from filters. Readers later can not find us.
 116  */
 117 static void free_tcf(struct tc_action *p)
 118 {
 119         struct tcf_chain *chain = rcu_dereference_protected(p->goto_chain, 1);
 120
 121         free_percpu(p->cpu_bstats);
 122         free_percpu(p->cpu_bstats_hw);
 123         free_percpu(p->cpu_qstats);
 124
 125         tcf_set_action_cookie(&p->act_cookie, NULL);
 126         if (chain)
 127                 tcf_chain_put_by_act(chain);
 128
 129         kfree(p);
 130 }
 131
 132 static void tcf_action_cleanup(struct tc_action *p)
 133 {
 134         if (p->ops->cleanup)
 135                 p->ops->cleanup(p);
 136
 137         gen_kill_estimator(&p->tcfa_rate_est);
 138         free_tcf(p);
 139 }
 140
 141 static int __tcf_action_put(struct tc_action *p, bool bind)
 142 {
 143         struct tcf_idrinfo *idrinfo = p->idrinfo;
 144
 145         if (refcount_dec_and_mutex_lock(&p->tcfa_refcnt, &idrinfo->lock)) {
 146                 if (bind)
 147                         atomic_dec(&p->tcfa_bindcnt);
 148                 idr_remove(&idrinfo->action_idr, p->tcfa_index);
 149                 mutex_unlock(&idrinfo->lock);
 150
 151                 tcf_action_cleanup(p);
 152                 return 1;
 153         }
 154
 155         if (bind)
 156                 atomic_dec(&p->tcfa_bindcnt);
 157
 158         return 0;
 159 }
 160
 161 int __tcf_idr_release(struct tc_action *p, bool bind, bool strict)
 162 {
 163         int ret = 0;
 164
 165         /* Release with strict==1 and bind==0 is only called through act API
 166          * interface (classifiers always bind). Only case when action with
 167          * positive reference count and zero bind count can exist is when it was
 168          * also created with act API (unbinding last classifier will destroy the
 169          * action if it was created by classifier). So only case when bind count
 170          * can be changed after initial check is when unbound action is
 171          * destroyed by act API while classifier binds to action with same id
 172          * concurrently. This result either creation of new action(same behavior
 173          * as before), or reusing existing action if concurrent process
 174          * increments reference count before action is deleted. Both scenarios
 175          * are acceptable.
 176          */
 177         if (p) {
 178                 if (!bind && strict && atomic_read(&p->tcfa_bindcnt) > 0)
 179                         return -EPERM;
 180
 181                 if (__tcf_action_put(p, bind))
 182                         ret = ACT_P_DELETED;
 183         }
 184
 185         return ret;
 186 }
 187 EXPORT_SYMBOL(__tcf_idr_release);
 188
 189 static size_t tcf_action_shared_attrs_size(const struct tc_action *act)
 190 {
 191         struct tc_cookie *act_cookie;
 192         u32 cookie_len = 0;
 193
 194         rcu_read_lock();
 195         act_cookie = rcu_dereference(act->act_cookie);
 196
 197         if (act_cookie)
 198                 cookie_len = nla_total_size(act_cookie->len);
 199         rcu_read_unlock();
 200
 201         return  nla_total_size(0) /* action number nested */
 202                 + nla_total_size(IFNAMSIZ) /* TCA_ACT_KIND */
 203                 + cookie_len /* TCA_ACT_COOKIE */
 204                 + nla_total_size(sizeof(struct nla_bitfield32)) /* TCA_ACT_HW_STATS */
 205                 + nla_total_size(0) /* TCA_ACT_STATS nested */
 206                 + nla_total_size(sizeof(struct nla_bitfield32)) /* TCA_ACT_FLAGS */
 207                 /* TCA_STATS_BASIC */
 208                 + nla_total_size_64bit(sizeof(struct gnet_stats_basic))
 209                 /* TCA_STATS_PKT64 */
 210                 + nla_total_size_64bit(sizeof(u64))
 211                 /* TCA_STATS_QUEUE */
 212                 + nla_total_size_64bit(sizeof(struct gnet_stats_queue))
 213                 + nla_total_size(0) /* TCA_OPTIONS nested */
 214                 + nla_total_size(sizeof(struct tcf_t)); /* TCA_GACT_TM */
 215 }
 216
 217 static size_t tcf_action_full_attrs_size(size_t sz)
 218 {
 219         return NLMSG_HDRLEN                     /* struct nlmsghdr */
 220                 + sizeof(struct tcamsg)
 221                 + nla_total_size(0)             /* TCA_ACT_TAB nested */
 222                 + sz;
 223 }
 224
 225 static size_t tcf_action_fill_size(const struct tc_action *act)
 226 {
 227         size_t sz = tcf_action_shared_attrs_size(act);
 228
 229         if (act->ops->get_fill_size)
 230                 return act->ops->get_fill_size(act) + sz;
 231         return sz;
 232 }
 233
 234 static int
 235 tcf_action_dump_terse(struct sk_buff *skb, struct tc_action *a, bool from_act)
 236 {
 237         unsigned char *b = skb_tail_pointer(skb);
 238         struct tc_cookie *cookie;
 239
 240         if (nla_put_string(skb, TCA_KIND, a->ops->kind))
 241                 goto nla_put_failure;
 242         if (tcf_action_copy_stats(skb, a, 0))
 243                 goto nla_put_failure;
 244         if (from_act && nla_put_u32(skb, TCA_ACT_INDEX, a->tcfa_index))
 245                 goto nla_put_failure;
 246
 247         rcu_read_lock();
 248         cookie = rcu_dereference(a->act_cookie);
 249         if (cookie) {
 250                 if (nla_put(skb, TCA_ACT_COOKIE, cookie->len, cookie->data)) {
 251                         rcu_read_unlock();
 252                         goto nla_put_failure;
 253                 }
 254         }
 255         rcu_read_unlock();
 256
 257         return 0;
 258
 259 nla_put_failure:
 260         nlmsg_trim(skb, b);
 261         return -1;
 262 }
 263
 264 static int tcf_dump_walker(struct tcf_idrinfo *idrinfo, struct sk_buff *skb,
 265                            struct netlink_callback *cb)
 266 {
 267         int err = 0, index = -1, s_i = 0, n_i = 0;
 268         u32 act_flags = cb->args[2];
 269         unsigned long jiffy_since = cb->args[3];
 270         struct nlattr *nest;
 271         struct idr *idr = &idrinfo->action_idr;
 272         struct tc_action *p;
 273         unsigned long id = 1;
 274         unsigned long tmp;
 275
 276         mutex_lock(&idrinfo->lock);
 277
 278         s_i = cb->args[0];
 279
 280         idr_for_each_entry_ul(idr, p, tmp, id) {
 281                 index++;
 282                 if (index < s_i)
 283                         continue;
 284                 if (IS_ERR(p))
 285                         continue;
 286
 287                 if (jiffy_since &&
 288                     time_after(jiffy_since,
 289                                (unsigned long)p->tcfa_tm.lastuse))
 290                         continue;
 291
 292                 nest = nla_nest_start_noflag(skb, n_i);
 293                 if (!nest) {
 294                         index--;
 295                         goto nla_put_failure;
 296                 }
 297                 err = (act_flags & TCA_ACT_FLAG_TERSE_DUMP) ?
 298                         tcf_action_dump_terse(skb, p, true) :
 299                         tcf_action_dump_1(skb, p, 0, 0);
 300                 if (err < 0) {
 301                         index--;
 302                         nlmsg_trim(skb, nest);
 303                         goto done;
 304                 }
 305                 nla_nest_end(skb, nest);
 306                 n_i++;
 307                 if (!(act_flags & TCA_ACT_FLAG_LARGE_DUMP_ON) &&
 308                     n_i >= TCA_ACT_MAX_PRIO)
 309                         goto done;
 310         }
 311 done:
 312         if (index >= 0)
 313                 cb->args[0] = index + 1;
 314
 315         mutex_unlock(&idrinfo->lock);
 316         if (n_i) {
 317                 if (act_flags & TCA_ACT_FLAG_LARGE_DUMP_ON)
 318                         cb->args[1] = n_i;
 319         }
 320         return n_i;
 321
 322 nla_put_failure:
 323         nla_nest_cancel(skb, nest);
 324         goto done;
 325 }
 326
 327 static int tcf_idr_release_unsafe(struct tc_action *p)
 328 {
 329         if (atomic_read(&p->tcfa_bindcnt) > 0)
 330                 return -EPERM;
 331
 332         if (refcount_dec_and_test(&p->tcfa_refcnt)) {
 333                 idr_remove(&p->idrinfo->action_idr, p->tcfa_index);
 334                 tcf_action_cleanup(p);
 335                 return ACT_P_DELETED;
 336         }
 337
 338         return 0;
 339 }
 340
 341 static int tcf_del_walker(struct tcf_idrinfo *idrinfo, struct sk_buff *skb,
 342                           const struct tc_action_ops *ops)
 343 {
 344         struct nlattr *nest;
 345         int n_i = 0;
 346         int ret = -EINVAL;
 347         struct idr *idr = &idrinfo->action_idr;
 348         struct tc_action *p;
 349         unsigned long id = 1;
 350         unsigned long tmp;
 351
 352         nest = nla_nest_start_noflag(skb, 0);
 353         if (nest == NULL)
 354                 goto nla_put_failure;
 355         if (nla_put_string(skb, TCA_KIND, ops->kind))
 356                 goto nla_put_failure;
 357
 358         mutex_lock(&idrinfo->lock);
 359         idr_for_each_entry_ul(idr, p, tmp, id) {
 360                 if (IS_ERR(p))
 361                         continue;
 362                 ret = tcf_idr_release_unsafe(p);
 363                 if (ret == ACT_P_DELETED) {
 364                         module_put(ops->owner);
 365                         n_i++;
 366                 } else if (ret < 0) {
 367                         mutex_unlock(&idrinfo->lock);
 368                         goto nla_put_failure;
 369                 }
 370         }
 371         mutex_unlock(&idrinfo->lock);
 372
 373         if (nla_put_u32(skb, TCA_FCNT, n_i))
 374                 goto nla_put_failure;
 375         nla_nest_end(skb, nest);
 376
 377         return n_i;
 378 nla_put_failure:
 379         nla_nest_cancel(skb, nest);
 380         return ret;
 381 }
 382
 383 int tcf_generic_walker(struct tc_action_net *tn, struct sk_buff *skb,
 384                        struct netlink_callback *cb, int type,
 385                        const struct tc_action_ops *ops,
 386                        struct netlink_ext_ack *extack)
 387 {
 388         struct tcf_idrinfo *idrinfo = tn->idrinfo;
 389
 390         if (type == RTM_DELACTION) {
 391                 return tcf_del_walker(idrinfo, skb, ops);
 392         } else if (type == RTM_GETACTION) {
 393                 return tcf_dump_walker(idrinfo, skb, cb);
 394         } else {
 395                 WARN(1, "tcf_generic_walker: unknown command %d\n", type);
 396                 NL_SET_ERR_MSG(extack, "tcf_generic_walker: unknown command");
 397                 return -EINVAL;
 398         }
 399 }
 400 EXPORT_SYMBOL(tcf_generic_walker);
 401
 402 int tcf_idr_search(struct tc_action_net *tn, struct tc_action **a, u32 index)
 403 {
 404         struct tcf_idrinfo *idrinfo = tn->idrinfo;
 405         struct tc_action *p;
 406
 407         mutex_lock(&idrinfo->lock);
 408         p = idr_find(&idrinfo->action_idr, index);
 409         if (IS_ERR(p))
 410                 p = NULL;
 411         else if (p)
 412                 refcount_inc(&p->tcfa_refcnt);
 413         mutex_unlock(&idrinfo->lock);
 414
 415         if (p) {
 416                 *a = p;
 417                 return true;
 418         }
 419         return false;
 420 }
 421 EXPORT_SYMBOL(tcf_idr_search);
 422
 423 static int tcf_idr_delete_index(struct tcf_idrinfo *idrinfo, u32 index)
 424 {
 425         struct tc_action *p;
 426         int ret = 0;
 427
 428         mutex_lock(&idrinfo->lock);
 429         p = idr_find(&idrinfo->action_idr, index);
 430         if (!p) {
 431                 mutex_unlock(&idrinfo->lock);
 432                 return -ENOENT;
 433         }
 434
 435         if (!atomic_read(&p->tcfa_bindcnt)) {
 436                 if (refcount_dec_and_test(&p->tcfa_refcnt)) {
 437                         struct module *owner = p->ops->owner;
 438
 439                         WARN_ON(p != idr_remove(&idrinfo->action_idr,
 440                                                 p->tcfa_index));
 441                         mutex_unlock(&idrinfo->lock);
 442
 443                         tcf_action_cleanup(p);
 444                         module_put(owner);
 445                         return 0;
 446                 }
 447                 ret = 0;
 448         } else {
 449                 ret = -EPERM;
 450         }
 451
 452         mutex_unlock(&idrinfo->lock);
 453         return ret;
 454 }
 455
 456 int tcf_idr_create(struct tc_action_net *tn, u32 index, struct nlattr *est,
 457                    struct tc_action **a, const struct tc_action_ops *ops,
 458                    int bind, bool cpustats, u32 flags)
 459 {
 460         struct tc_action *p = kzalloc(ops->size, GFP_KERNEL);
 461         struct tcf_idrinfo *idrinfo = tn->idrinfo;
 462         int err = -ENOMEM;
 463
 464         if (unlikely(!p))
 465                 return -ENOMEM;
 466         refcount_set(&p->tcfa_refcnt, 1);
 467         if (bind)
 468                 atomic_set(&p->tcfa_bindcnt, 1);
 469
 470         if (cpustats) {
 471                 p->cpu_bstats = netdev_alloc_pcpu_stats(struct gnet_stats_basic_cpu);
 472                 if (!p->cpu_bstats)
 473                         goto err1;
 474                 p->cpu_bstats_hw = netdev_alloc_pcpu_stats(struct gnet_stats_basic_cpu);
 475                 if (!p->cpu_bstats_hw)
 476                         goto err2;
 477                 p->cpu_qstats = alloc_percpu(struct gnet_stats_queue);
 478                 if (!p->cpu_qstats)
 479                         goto err3;
 480         }
 481         spin_lock_init(&p->tcfa_lock);
 482         p->tcfa_index = index;
 483         p->tcfa_tm.install = jiffies;
 484         p->tcfa_tm.lastuse = jiffies;
 485         p->tcfa_tm.firstuse = 0;
 486         p->tcfa_flags = flags;
 487         if (est) {
 488                 err = gen_new_estimator(&p->tcfa_bstats, p->cpu_bstats,
 489                                         &p->tcfa_rate_est,
 490                                         &p->tcfa_lock, NULL, est);
 491                 if (err)
 492                         goto err4;
 493         }
 494
 495         p->idrinfo = idrinfo;
 496         p->ops = ops;
 497         *a = p;
 498         return 0;
 499 err4:
 500         free_percpu(p->cpu_qstats);
 501 err3:
 502         free_percpu(p->cpu_bstats_hw);
 503 err2:
 504         free_percpu(p->cpu_bstats);
 505 err1:
 506         kfree(p);
 507         return err;
 508 }
 509 EXPORT_SYMBOL(tcf_idr_create);
 510
 511 int tcf_idr_create_from_flags(struct tc_action_net *tn, u32 index,
 512                               struct nlattr *est, struct tc_action **a,
 513                               const struct tc_action_ops *ops, int bind,
 514                               u32 flags)
 515 {
 516         /* Set cpustats according to actions flags. */
 517         return tcf_idr_create(tn, index, est, a, ops, bind,
 518                               !(flags & TCA_ACT_FLAGS_NO_PERCPU_STATS), flags);
 519 }
 520 EXPORT_SYMBOL(tcf_idr_create_from_flags);
 521
 522 /* Cleanup idr index that was allocated but not initialized. */
 523
 524 void tcf_idr_cleanup(struct tc_action_net *tn, u32 index)
 525 {
 526         struct tcf_idrinfo *idrinfo = tn->idrinfo;
 527
 528         mutex_lock(&idrinfo->lock);
 529         /* Remove ERR_PTR(-EBUSY) allocated by tcf_idr_check_alloc */
 530         WARN_ON(!IS_ERR(idr_remove(&idrinfo->action_idr, index)));
 531         mutex_unlock(&idrinfo->lock);
 532 }
 533 EXPORT_SYMBOL(tcf_idr_cleanup);
 534
 535 /* Check if action with specified index exists. If actions is found, increments
 536  * its reference and bind counters, and return 1. Otherwise insert temporary
 537  * error pointer (to prevent concurrent users from inserting actions with same
 538  * index) and return 0.
 539  */
 540
 541 int tcf_idr_check_alloc(struct tc_action_net *tn, u32 *index,
 542                         struct tc_action **a, int bind)
 543 {
 544         struct tcf_idrinfo *idrinfo = tn->idrinfo;
 545         struct tc_action *p;
 546         int ret;
 547
 548 again:
 549         mutex_lock(&idrinfo->lock);
 550         if (*index) {
 551                 p = idr_find(&idrinfo->action_idr, *index);
 552                 if (IS_ERR(p)) {
 553                         /* This means that another process allocated
 554                          * index but did not assign the pointer yet.
 555                          */
 556                         mutex_unlock(&idrinfo->lock);
 557                         goto again;
 558                 }
 559
 560                 if (p) {
 561                         refcount_inc(&p->tcfa_refcnt);
 562                         if (bind)
 563                                 atomic_inc(&p->tcfa_bindcnt);
 564                         *a = p;
 565                         ret = 1;
 566                 } else {
 567                         *a = NULL;
 568                         ret = idr_alloc_u32(&idrinfo->action_idr, NULL, index,
 569                                             *index, GFP_KERNEL);
 570                         if (!ret)
 571                                 idr_replace(&idrinfo->action_idr,
 572                                             ERR_PTR(-EBUSY), *index);
 573                 }
 574         } else {
 575                 *index = 1;
 576                 *a = NULL;
 577                 ret = idr_alloc_u32(&idrinfo->action_idr, NULL, index,
 578                                     UINT_MAX, GFP_KERNEL);
 579                 if (!ret)
 580                         idr_replace(&idrinfo->action_idr, ERR_PTR(-EBUSY),
 581                                     *index);
 582         }
 583         mutex_unlock(&idrinfo->lock);
 584         return ret;
 585 }
 586 EXPORT_SYMBOL(tcf_idr_check_alloc);
 587
 588 void tcf_idrinfo_destroy(const struct tc_action_ops *ops,
 589                          struct tcf_idrinfo *idrinfo)
 590 {
 591         struct idr *idr = &idrinfo->action_idr;
 592         struct tc_action *p;
 593         int ret;
 594         unsigned long id = 1;
 595         unsigned long tmp;
 596
 597         idr_for_each_entry_ul(idr, p, tmp, id) {
 598                 ret = __tcf_idr_release(p, false, true);
 599                 if (ret == ACT_P_DELETED)
 600                         module_put(ops->owner);
 601                 else if (ret < 0)
 602                         return;
 603         }
 604         idr_destroy(&idrinfo->action_idr);
 605 }
 606 EXPORT_SYMBOL(tcf_idrinfo_destroy);
 607
 608 static LIST_HEAD(act_base);
 609 static DEFINE_RWLOCK(act_mod_lock);
 610
 611 int tcf_register_action(struct tc_action_ops *act,
 612                         struct pernet_operations *ops)
 613 {
 614         struct tc_action_ops *a;
 615         int ret;
 616
 617         if (!act->act || !act->dump || !act->init || !act->walk || !act->lookup)
 618                 return -EINVAL;
 619
 620         /* We have to register pernet ops before making the action ops visible,
 621          * otherwise tcf_action_init_1() could get a partially initialized
 622          * netns.
 623          */
 624         ret = register_pernet_subsys(ops);
 625         if (ret)
 626                 return ret;
 627
 628         write_lock(&act_mod_lock);
 629         list_for_each_entry(a, &act_base, head) {
 630                 if (act->id == a->id || (strcmp(act->kind, a->kind) == 0)) {
 631                         write_unlock(&act_mod_lock);
 632                         unregister_pernet_subsys(ops);
 633                         return -EEXIST;
 634                 }
 635         }
 636         list_add_tail(&act->head, &act_base);
 637         write_unlock(&act_mod_lock);
 638
 639         return 0;
 640 }
 641 EXPORT_SYMBOL(tcf_register_action);
 642
 643 int tcf_unregister_action(struct tc_action_ops *act,
 644                           struct pernet_operations *ops)
 645 {
 646         struct tc_action_ops *a;
 647         int err = -ENOENT;
 648
 649         write_lock(&act_mod_lock);
 650         list_for_each_entry(a, &act_base, head) {
 651                 if (a == act) {
 652                         list_del(&act->head);
 653                         err = 0;
 654                         break;
 655                 }
 656         }
 657         write_unlock(&act_mod_lock);
 658         if (!err)
 659                 unregister_pernet_subsys(ops);
 660         return err;
 661 }
 662 EXPORT_SYMBOL(tcf_unregister_action);
 663
 664 /* lookup by name */
 665 static struct tc_action_ops *tc_lookup_action_n(char *kind)
 666 {
 667         struct tc_action_ops *a, *res = NULL;
 668
 669         if (kind) {
 670                 read_lock(&act_mod_lock);
 671                 list_for_each_entry(a, &act_base, head) {
 672                         if (strcmp(kind, a->kind) == 0) {
 673                                 if (try_module_get(a->owner))
 674                                         res = a;
 675                                 break;
 676                         }
 677                 }
 678                 read_unlock(&act_mod_lock);
 679         }
 680         return res;
 681 }
 682
 683 /* lookup by nlattr */
 684 static struct tc_action_ops *tc_lookup_action(struct nlattr *kind)
 685 {
 686         struct tc_action_ops *a, *res = NULL;
 687
 688         if (kind) {
 689                 read_lock(&act_mod_lock);
 690                 list_for_each_entry(a, &act_base, head) {
 691                         if (nla_strcmp(kind, a->kind) == 0) {
 692                                 if (try_module_get(a->owner))
 693                                         res = a;
 694                                 break;
 695                         }
 696                 }
 697                 read_unlock(&act_mod_lock);
 698         }
 699         return res;
 700 }
 701
 702 /*TCA_ACT_MAX_PRIO is 32, there count up to 32 */
 703 #define TCA_ACT_MAX_PRIO_MASK 0x1FF
 704 int tcf_action_exec(struct sk_buff *skb, struct tc_action **actions,
 705                     int nr_actions, struct tcf_result *res)
 706 {
 707         u32 jmp_prgcnt = 0;
 708         u32 jmp_ttl = TCA_ACT_MAX_PRIO; /*matches actions per filter */
 709         int i;
 710         int ret = TC_ACT_OK;
 711
 712         if (skb_skip_tc_classify(skb))
 713                 return TC_ACT_OK;
 714
 715 restart_act_graph:
 716         for (i = 0; i < nr_actions; i++) {
 717                 const struct tc_action *a = actions[i];
 718
 719                 if (jmp_prgcnt > 0) {
 720                         jmp_prgcnt -= 1;
 721                         continue;
 722                 }
 723 repeat:
 724                 ret = a->ops->act(skb, a, res);
 725                 if (ret == TC_ACT_REPEAT)
 726                         goto repeat;    /* we need a ttl - JHS */
 727
 728                 if (TC_ACT_EXT_CMP(ret, TC_ACT_JUMP)) {
 729                         jmp_prgcnt = ret & TCA_ACT_MAX_PRIO_MASK;
 730                         if (!jmp_prgcnt || (jmp_prgcnt > nr_actions)) {
 731                                 /* faulty opcode, stop pipeline */
 732                                 return TC_ACT_OK;
 733                         } else {
 734                                 jmp_ttl -= 1;
 735                                 if (jmp_ttl > 0)
 736                                         goto restart_act_graph;
 737                                 else /* faulty graph, stop pipeline */
 738                                         return TC_ACT_OK;
 739                         }
 740                 } else if (TC_ACT_EXT_CMP(ret, TC_ACT_GOTO_CHAIN)) {
 741                         if (unlikely(!rcu_access_pointer(a->goto_chain))) {
 742                                 net_warn_ratelimited("can't go to NULL chain!\n");
 743                                 return TC_ACT_SHOT;
 744                         }
 745                         tcf_action_goto_chain_exec(a, res);
 746                 }
 747
 748                 if (ret != TC_ACT_PIPE)
 749                         break;
 750         }
 751
 752         return ret;
 753 }
 754 EXPORT_SYMBOL(tcf_action_exec);
 755
 756 int tcf_action_destroy(struct tc_action *actions[], int bind)
 757 {
 758         const struct tc_action_ops *ops;
 759         struct tc_action *a;
 760         int ret = 0, i;
 761
 762         for (i = 0; i < TCA_ACT_MAX_PRIO && actions[i]; i++) {
 763                 a = actions[i];
 764                 actions[i] = NULL;
 765                 ops = a->ops;
 766                 ret = __tcf_idr_release(a, bind, true);
 767                 if (ret == ACT_P_DELETED)
 768                         module_put(ops->owner);
 769                 else if (ret < 0)
 770                         return ret;
 771         }
 772         return ret;
 773 }
 774
 775 static int tcf_action_put(struct tc_action *p)
 776 {
 777         return __tcf_action_put(p, false);
 778 }
 779
 780 /* Put all actions in this array, skip those NULL's. */
 781 static void tcf_action_put_many(struct tc_action *actions[])
 782 {
 783         int i;
 784
 785         for (i = 0; i < TCA_ACT_MAX_PRIO; i++) {
 786                 struct tc_action *a = actions[i];
 787                 const struct tc_action_ops *ops;
 788
 789                 if (!a)
 790                         continue;
 791                 ops = a->ops;
 792                 if (tcf_action_put(a))
 793                         module_put(ops->owner);
 794         }
 795 }
 796
 797 int
 798 tcf_action_dump_old(struct sk_buff *skb, struct tc_action *a, int bind, int ref)
 799 {
 800         return a->ops->dump(skb, a, bind, ref);
 801 }
 802
 803 int
 804 tcf_action_dump_1(struct sk_buff *skb, struct tc_action *a, int bind, int ref)
 805 {
 806         int err = -EINVAL;
 807         unsigned char *b = skb_tail_pointer(skb);
 808         struct nlattr *nest;
 809
 810         if (tcf_action_dump_terse(skb, a, false))
 811                 goto nla_put_failure;
 812
 813         if (a->hw_stats != TCA_ACT_HW_STATS_ANY &&
 814             nla_put_bitfield32(skb, TCA_ACT_HW_STATS,
 815                                a->hw_stats, TCA_ACT_HW_STATS_ANY))
 816                 goto nla_put_failure;
 817
 818         if (a->used_hw_stats_valid &&
 819             nla_put_bitfield32(skb, TCA_ACT_USED_HW_STATS,
 820                                a->used_hw_stats, TCA_ACT_HW_STATS_ANY))
 821                 goto nla_put_failure;
 822
 823         if (a->tcfa_flags &&
 824             nla_put_bitfield32(skb, TCA_ACT_FLAGS,
 825                                a->tcfa_flags, a->tcfa_flags))
 826                 goto nla_put_failure;
 827
 828         nest = nla_nest_start_noflag(skb, TCA_OPTIONS);
 829         if (nest == NULL)
 830                 goto nla_put_failure;
 831         err = tcf_action_dump_old(skb, a, bind, ref);
 832         if (err > 0) {
 833                 nla_nest_end(skb, nest);
 834                 return err;
 835         }
 836
 837 nla_put_failure:
 838         nlmsg_trim(skb, b);
 839         return -1;
 840 }
 841 EXPORT_SYMBOL(tcf_action_dump_1);
 842
 843 int tcf_action_dump(struct sk_buff *skb, struct tc_action *actions[],
 844                     int bind, int ref, bool terse)
 845 {
 846         struct tc_action *a;
 847         int err = -EINVAL, i;
 848         struct nlattr *nest;
 849
 850         for (i = 0; i < TCA_ACT_MAX_PRIO && actions[i]; i++) {
 851                 a = actions[i];
 852                 nest = nla_nest_start_noflag(skb, i + 1);
 853                 if (nest == NULL)
 854                         goto nla_put_failure;
 855                 err = terse ? tcf_action_dump_terse(skb, a, false) :
 856                         tcf_action_dump_1(skb, a, bind, ref);
 857                 if (err < 0)
 858                         goto errout;
 859                 nla_nest_end(skb, nest);
 860         }
 861
 862         return 0;
 863
 864 nla_put_failure:
 865         err = -EINVAL;
 866 errout:
 867         nla_nest_cancel(skb, nest);
 868         return err;
 869 }
 870
 871 static struct tc_cookie *nla_memdup_cookie(struct nlattr **tb)
 872 {
 873         struct tc_cookie *c = kzalloc(sizeof(*c), GFP_KERNEL);
 874         if (!c)
 875                 return NULL;
 876
 877         c->data = nla_memdup(tb[TCA_ACT_COOKIE], GFP_KERNEL);
 878         if (!c->data) {
 879                 kfree(c);
 880                 return NULL;
 881         }
 882         c->len = nla_len(tb[TCA_ACT_COOKIE]);
 883
 884         return c;
 885 }
 886
 887 static u8 tcf_action_hw_stats_get(struct nlattr *hw_stats_attr)
 888 {
 889         struct nla_bitfield32 hw_stats_bf;
 890
 891         /* If the user did not pass the attr, that means he does
 892          * not care about the type. Return "any" in that case
 893          * which is setting on all supported types.
 894          */
 895         if (!hw_stats_attr)
 896                 return TCA_ACT_HW_STATS_ANY;
 897         hw_stats_bf = nla_get_bitfield32(hw_stats_attr);
 898         return hw_stats_bf.value;
 899 }
 900
 901 static const struct nla_policy tcf_action_policy[TCA_ACT_MAX + 1] = {
 902         [TCA_ACT_KIND]          = { .type = NLA_STRING },
 903         [TCA_ACT_INDEX]         = { .type = NLA_U32 },
 904         [TCA_ACT_COOKIE]        = { .type = NLA_BINARY,
 905                                     .len = TC_COOKIE_MAX_SIZE },
 906         [TCA_ACT_OPTIONS]       = { .type = NLA_NESTED },
 907         [TCA_ACT_FLAGS]         = NLA_POLICY_BITFIELD32(TCA_ACT_FLAGS_NO_PERCPU_STATS),
 908         [TCA_ACT_HW_STATS]      = NLA_POLICY_BITFIELD32(TCA_ACT_HW_STATS_ANY),
 909 };
 910
 911 static void tcf_idr_insert_many(struct tc_action *actions[])
 912 {
 913         int i;
 914
 915         for (i = 0; i < TCA_ACT_MAX_PRIO; i++) {
 916                 struct tc_action *a = actions[i];
 917                 struct tcf_idrinfo *idrinfo;
 918
 919                 if (!a)
 920                         continue;
 921                 idrinfo = a->idrinfo;
 922                 mutex_lock(&idrinfo->lock);
 923                 /* Replace ERR_PTR(-EBUSY) allocated by tcf_idr_check_alloc if
 924                  * it is just created, otherwise this is just a nop.
 925                  */
 926                 idr_replace(&idrinfo->action_idr, a, a->tcfa_index);
 927                 mutex_unlock(&idrinfo->lock);
 928         }
 929 }
 930
 931 struct tc_action *tcf_action_init_1(struct net *net, struct tcf_proto *tp,
 932                                     struct nlattr *nla, struct nlattr *est,
 933                                     char *name, int ovr, int bind,
 934                                     bool rtnl_held,
 935                                     struct netlink_ext_ack *extack)
 936 {
 937         struct nla_bitfield32 flags = { 0, 0 };
 938         u8 hw_stats = TCA_ACT_HW_STATS_ANY;
 939         struct tc_action *a;
 940         struct tc_action_ops *a_o;
 941         struct tc_cookie *cookie = NULL;
 942         char act_name[IFNAMSIZ];
 943         struct nlattr *tb[TCA_ACT_MAX + 1];
 944         struct nlattr *kind;
 945         int err;
 946
 947         if (name == NULL) {
 948                 err = nla_parse_nested_deprecated(tb, TCA_ACT_MAX, nla,
 949                                                   tcf_action_policy, extack);
 950                 if (err < 0)
 951                         goto err_out;
 952                 err = -EINVAL;
 953                 kind = tb[TCA_ACT_KIND];
 954                 if (!kind) {
 955                         NL_SET_ERR_MSG(extack, "TC action kind must be specified");
 956                         goto err_out;
 957                 }
 958                 if (nla_strscpy(act_name, kind, IFNAMSIZ) < 0) {
 959                         NL_SET_ERR_MSG(extack, "TC action name too long");
 960                         goto err_out;
 961                 }
 962                 if (tb[TCA_ACT_COOKIE]) {
 963                         cookie = nla_memdup_cookie(tb);
 964                         if (!cookie) {
 965                                 NL_SET_ERR_MSG(extack, "No memory to generate TC cookie");
 966                                 err = -ENOMEM;
 967                                 goto err_out;
 968                         }
 969                 }
 970                 hw_stats = tcf_action_hw_stats_get(tb[TCA_ACT_HW_STATS]);
 971                 if (tb[TCA_ACT_FLAGS])
 972                         flags = nla_get_bitfield32(tb[TCA_ACT_FLAGS]);
 973         } else {
 974                 if (strlcpy(act_name, name, IFNAMSIZ) >= IFNAMSIZ) {
 975                         NL_SET_ERR_MSG(extack, "TC action name too long");
 976                         err = -EINVAL;
 977                         goto err_out;
 978                 }
 979         }
 980
 981         a_o = tc_lookup_action_n(act_name);
 982         if (a_o == NULL) {
 983 #ifdef CONFIG_MODULES
 984                 if (rtnl_held)
 985                         rtnl_unlock();
 986                 request_module("act_%s", act_name);
 987                 if (rtnl_held)
 988                         rtnl_lock();
 989
 990                 a_o = tc_lookup_action_n(act_name);
 991
 992                 /* We dropped the RTNL semaphore in order to
 993                  * perform the module load.  So, even if we
 994                  * succeeded in loading the module we have to
 995                  * tell the caller to replay the request.  We
 996                  * indicate this using -EAGAIN.
 997                  */
 998                 if (a_o != NULL) {
 999                         err = -EAGAIN;
1000                         goto err_mod;
1001                 }
1002 #endif
1003                 NL_SET_ERR_MSG(extack, "Failed to load TC action module");
1004                 err = -ENOENT;
1005                 goto err_free;
1006         }
1007
1008         /* backward compatibility for policer */
1009         if (name == NULL)
1010                 err = a_o->init(net, tb[TCA_ACT_OPTIONS], est, &a, ovr, bind,
1011                                 rtnl_held, tp, flags.value, extack);
1012         else
1013                 err = a_o->init(net, nla, est, &a, ovr, bind, rtnl_held,
1014                                 tp, flags.value, extack);
1015         if (err < 0)
1016                 goto err_mod;
1017
1018         if (!name && tb[TCA_ACT_COOKIE])
1019                 tcf_set_action_cookie(&a->act_cookie, cookie);
1020
1021         if (!name)
1022                 a->hw_stats = hw_stats;
1023
1024         /* module count goes up only when brand new policy is created
1025          * if it exists and is only bound to in a_o->init() then
1026          * ACT_P_CREATED is not returned (a zero is).
1027          */
1028         if (err != ACT_P_CREATED)
1029                 module_put(a_o->owner);
1030
1031         return a;
1032
1033 err_mod:
1034         module_put(a_o->owner);
1035 err_free:
1036         if (cookie) {
1037                 kfree(cookie->data);
1038                 kfree(cookie);
1039         }
1040 err_out:
1041         return ERR_PTR(err);
1042 }
1043
1044 /* Returns numbers of initialized actions or negative error. */
1045
1046 int tcf_action_init(struct net *net, struct tcf_proto *tp, struct nlattr *nla,
1047                     struct nlattr *est, char *name, int ovr, int bind,
1048                     struct tc_action *actions[], size_t *attr_size,
1049                     bool rtnl_held, struct netlink_ext_ack *extack)
1050 {
1051         struct nlattr *tb[TCA_ACT_MAX_PRIO + 1];
1052         struct tc_action *act;
1053         size_t sz = 0;
1054         int err;
1055         int i;
1056
1057         err = nla_parse_nested_deprecated(tb, TCA_ACT_MAX_PRIO, nla, NULL,
1058                                           extack);
1059         if (err < 0)
1060                 return err;
1061
1062         for (i = 1; i <= TCA_ACT_MAX_PRIO && tb[i]; i++) {
1063                 act = tcf_action_init_1(net, tp, tb[i], est, name, ovr, bind,
1064                                         rtnl_held, extack);
1065                 if (IS_ERR(act)) {
1066                         err = PTR_ERR(act);
1067                         goto err;
1068                 }
1069                 sz += tcf_action_fill_size(act);
1070                 /* Start from index 0 */
1071                 actions[i - 1] = act;
1072         }
1073
1074         /* We have to commit them all together, because if any error happened in
1075          * between, we could not handle the failure gracefully.
1076          */
1077         tcf_idr_insert_many(actions);
1078
1079         *attr_size = tcf_action_full_attrs_size(sz);
1080         return i - 1;
1081
1082 err:
1083         tcf_action_destroy(actions, bind);
1084         return err;
1085 }
1086
1087 void tcf_action_update_stats(struct tc_action *a, u64 bytes, u64 packets,
1088                              u64 drops, bool hw)
1089 {
1090         if (a->cpu_bstats) {
1091                 _bstats_cpu_update(this_cpu_ptr(a->cpu_bstats), bytes, packets);
1092
1093                 this_cpu_ptr(a->cpu_qstats)->drops += drops;
1094
1095                 if (hw)
1096                         _bstats_cpu_update(this_cpu_ptr(a->cpu_bstats_hw),
1097                                            bytes, packets);
1098                 return;
1099         }
1100
1101         _bstats_update(&a->tcfa_bstats, bytes, packets);
1102         a->tcfa_qstats.drops += drops;
1103         if (hw)
1104                 _bstats_update(&a->tcfa_bstats_hw, bytes, packets);
1105 }
1106 EXPORT_SYMBOL(tcf_action_update_stats);
1107
1108 int tcf_action_copy_stats(struct sk_buff *skb, struct tc_action *p,
1109                           int compat_mode)
1110 {
1111         int err = 0;
1112         struct gnet_dump d;
1113
1114         if (p == NULL)
1115                 goto errout;
1116
1117         /* compat_mode being true specifies a call that is supposed
1118          * to add additional backward compatibility statistic TLVs.
1119          */
1120         if (compat_mode) {
1121                 if (p->type == TCA_OLD_COMPAT)
1122                         err = gnet_stats_start_copy_compat(skb, 0,
1123                                                            TCA_STATS,
1124                                                            TCA_XSTATS,
1125                                                            &p->tcfa_lock, &d,
1126                                                            TCA_PAD);
1127                 else
1128                         return 0;
1129         } else
1130                 err = gnet_stats_start_copy(skb, TCA_ACT_STATS,
1131                                             &p->tcfa_lock, &d, TCA_ACT_PAD);
1132
1133         if (err < 0)
1134                 goto errout;
1135
1136         if (gnet_stats_copy_basic(NULL, &d, p->cpu_bstats, &p->tcfa_bstats) < 0 ||
1137             gnet_stats_copy_basic_hw(NULL, &d, p->cpu_bstats_hw,
1138                                      &p->tcfa_bstats_hw) < 0 ||
1139             gnet_stats_copy_rate_est(&d, &p->tcfa_rate_est) < 0 ||
1140             gnet_stats_copy_queue(&d, p->cpu_qstats,
1141                                   &p->tcfa_qstats,
1142                                   p->tcfa_qstats.qlen) < 0)
1143                 goto errout;
1144
1145         if (gnet_stats_finish_copy(&d) < 0)
1146                 goto errout;
1147
1148         return 0;
1149
1150 errout:
1151         return -1;
1152 }
1153
1154 static int tca_get_fill(struct sk_buff *skb, struct tc_action *actions[],
1155                         u32 portid, u32 seq, u16 flags, int event, int bind,
1156                         int ref)
1157 {
1158         struct tcamsg *t;
1159         struct nlmsghdr *nlh;
1160         unsigned char *b = skb_tail_pointer(skb);
1161         struct nlattr *nest;
1162
1163         nlh = nlmsg_put(skb, portid, seq, event, sizeof(*t), flags);
1164         if (!nlh)
1165                 goto out_nlmsg_trim;
1166         t = nlmsg_data(nlh);
1167         t->tca_family = AF_UNSPEC;
1168         t->tca__pad1 = 0;
1169         t->tca__pad2 = 0;
1170
1171         nest = nla_nest_start_noflag(skb, TCA_ACT_TAB);
1172         if (!nest)
1173                 goto out_nlmsg_trim;
1174
1175         if (tcf_action_dump(skb, actions, bind, ref, false) < 0)
1176                 goto out_nlmsg_trim;
1177
1178         nla_nest_end(skb, nest);
1179
1180         nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1181         return skb->len;
1182
1183 out_nlmsg_trim:
1184         nlmsg_trim(skb, b);
1185         return -1;
1186 }
1187
1188 static int
1189 tcf_get_notify(struct net *net, u32 portid, struct nlmsghdr *n,
1190                struct tc_action *actions[], int event,
1191                struct netlink_ext_ack *extack)
1192 {
1193         struct sk_buff *skb;
1194
1195         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1196         if (!skb)
1197                 return -ENOBUFS;
1198         if (tca_get_fill(skb, actions, portid, n->nlmsg_seq, 0, event,
1199                          0, 1) <= 0) {
1200                 NL_SET_ERR_MSG(extack, "Failed to fill netlink attributes while adding TC action");
1201                 kfree_skb(skb);
1202                 return -EINVAL;
1203         }
1204
1205         return rtnl_unicast(skb, net, portid);
1206 }
1207
1208 static struct tc_action *tcf_action_get_1(struct net *net, struct nlattr *nla,
1209                                           struct nlmsghdr *n, u32 portid,
1210                                           struct netlink_ext_ack *extack)
1211 {
1212         struct nlattr *tb[TCA_ACT_MAX + 1];
1213         const struct tc_action_ops *ops;
1214         struct tc_action *a;
1215         int index;
1216         int err;
1217
1218         err = nla_parse_nested_deprecated(tb, TCA_ACT_MAX, nla,
1219                                           tcf_action_policy, extack);
1220         if (err < 0)
1221                 goto err_out;
1222
1223         err = -EINVAL;
1224         if (tb[TCA_ACT_INDEX] == NULL ||
1225             nla_len(tb[TCA_ACT_INDEX]) < sizeof(index)) {
1226                 NL_SET_ERR_MSG(extack, "Invalid TC action index value");
1227                 goto err_out;
1228         }
1229         index = nla_get_u32(tb[TCA_ACT_INDEX]);
1230
1231         err = -EINVAL;
1232         ops = tc_lookup_action(tb[TCA_ACT_KIND]);
1233         if (!ops) { /* could happen in batch of actions */
1234                 NL_SET_ERR_MSG(extack, "Specified TC action kind not found");
1235                 goto err_out;
1236         }
1237         err = -ENOENT;
1238         if (ops->lookup(net, &a, index) == 0) {
1239                 NL_SET_ERR_MSG(extack, "TC action with specified index not found");
1240                 goto err_mod;
1241         }
1242
1243         module_put(ops->owner);
1244         return a;
1245
1246 err_mod:
1247         module_put(ops->owner);
1248 err_out:
1249         return ERR_PTR(err);
1250 }
1251
1252 static int tca_action_flush(struct net *net, struct nlattr *nla,
1253                             struct nlmsghdr *n, u32 portid,
1254                             struct netlink_ext_ack *extack)
1255 {
1256         struct sk_buff *skb;
1257         unsigned char *b;
1258         struct nlmsghdr *nlh;
1259         struct tcamsg *t;
1260         struct netlink_callback dcb;
1261         struct nlattr *nest;
1262         struct nlattr *tb[TCA_ACT_MAX + 1];
1263         const struct tc_action_ops *ops;
1264         struct nlattr *kind;
1265         int err = -ENOMEM;
1266
1267         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1268         if (!skb)
1269                 return err;
1270
1271         b = skb_tail_pointer(skb);
1272
1273         err = nla_parse_nested_deprecated(tb, TCA_ACT_MAX, nla,
1274                                           tcf_action_policy, extack);
1275         if (err < 0)
1276                 goto err_out;
1277
1278         err = -EINVAL;
1279         kind = tb[TCA_ACT_KIND];
1280         ops = tc_lookup_action(kind);
1281         if (!ops) { /*some idjot trying to flush unknown action */
1282                 NL_SET_ERR_MSG(extack, "Cannot flush unknown TC action");
1283                 goto err_out;
1284         }
1285
1286         nlh = nlmsg_put(skb, portid, n->nlmsg_seq, RTM_DELACTION,
1287                         sizeof(*t), 0);
1288         if (!nlh) {
1289                 NL_SET_ERR_MSG(extack, "Failed to create TC action flush notification");
1290                 goto out_module_put;
1291         }
1292         t = nlmsg_data(nlh);
1293         t->tca_family = AF_UNSPEC;
1294         t->tca__pad1 = 0;
1295         t->tca__pad2 = 0;
1296
1297         nest = nla_nest_start_noflag(skb, TCA_ACT_TAB);
1298         if (!nest) {
1299                 NL_SET_ERR_MSG(extack, "Failed to add new netlink message");
1300                 goto out_module_put;
1301         }
1302
1303         err = ops->walk(net, skb, &dcb, RTM_DELACTION, ops, extack);
1304         if (err <= 0) {
1305                 nla_nest_cancel(skb, nest);
1306                 goto out_module_put;
1307         }
1308
1309         nla_nest_end(skb, nest);
1310
1311         nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1312         nlh->nlmsg_flags |= NLM_F_ROOT;
1313         module_put(ops->owner);
1314         err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1315                              n->nlmsg_flags & NLM_F_ECHO);
1316         if (err > 0)
1317                 return 0;
1318         if (err < 0)
1319                 NL_SET_ERR_MSG(extack, "Failed to send TC action flush notification");
1320
1321         return err;
1322
1323 out_module_put:
1324         module_put(ops->owner);
1325 err_out:
1326         kfree_skb(skb);
1327         return err;
1328 }
1329
1330 static int tcf_action_delete(struct net *net, struct tc_action *actions[])
1331 {
1332         int i;
1333
1334         for (i = 0; i < TCA_ACT_MAX_PRIO && actions[i]; i++) {
1335                 struct tc_action *a = actions[i];
1336                 const struct tc_action_ops *ops = a->ops;
1337                 /* Actions can be deleted concurrently so we must save their
1338                  * type and id to search again after reference is released.
1339                  */
1340                 struct tcf_idrinfo *idrinfo = a->idrinfo;
1341                 u32 act_index = a->tcfa_index;
1342
1343                 actions[i] = NULL;
1344                 if (tcf_action_put(a)) {
1345                         /* last reference, action was deleted concurrently */
1346                         module_put(ops->owner);
1347                 } else  {
1348                         int ret;
1349
1350                         /* now do the delete */
1351                         ret = tcf_idr_delete_index(idrinfo, act_index);
1352                         if (ret < 0)
1353                                 return ret;
1354                 }
1355         }
1356         return 0;
1357 }
1358
1359 static int
1360 tcf_del_notify(struct net *net, struct nlmsghdr *n, struct tc_action *actions[],
1361                u32 portid, size_t attr_size, struct netlink_ext_ack *extack)
1362 {
1363         int ret;
1364         struct sk_buff *skb;
1365
1366         skb = alloc_skb(attr_size <= NLMSG_GOODSIZE ? NLMSG_GOODSIZE : attr_size,
1367                         GFP_KERNEL);
1368         if (!skb)
1369                 return -ENOBUFS;
1370
1371         if (tca_get_fill(skb, actions, portid, n->nlmsg_seq, 0, RTM_DELACTION,
1372                          0, 2) <= 0) {
1373                 NL_SET_ERR_MSG(extack, "Failed to fill netlink TC action attributes");
1374                 kfree_skb(skb);
1375                 return -EINVAL;
1376         }
1377
1378         /* now do the delete */
1379         ret = tcf_action_delete(net, actions);
1380         if (ret < 0) {
1381                 NL_SET_ERR_MSG(extack, "Failed to delete TC action");
1382                 kfree_skb(skb);
1383                 return ret;
1384         }
1385
1386         ret = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1387                              n->nlmsg_flags & NLM_F_ECHO);
1388         if (ret > 0)
1389                 return 0;
1390         return ret;
1391 }
1392
1393 static int
1394 tca_action_gd(struct net *net, struct nlattr *nla, struct nlmsghdr *n,
1395               u32 portid, int event, struct netlink_ext_ack *extack)
1396 {
1397         int i, ret;
1398         struct nlattr *tb[TCA_ACT_MAX_PRIO + 1];
1399         struct tc_action *act;
1400         size_t attr_size = 0;
1401         struct tc_action *actions[TCA_ACT_MAX_PRIO] = {};
1402
1403         ret = nla_parse_nested_deprecated(tb, TCA_ACT_MAX_PRIO, nla, NULL,
1404                                           extack);
1405         if (ret < 0)
1406                 return ret;
1407
1408         if (event == RTM_DELACTION && n->nlmsg_flags & NLM_F_ROOT) {
1409                 if (tb[1])
1410                         return tca_action_flush(net, tb[1], n, portid, extack);
1411
1412                 NL_SET_ERR_MSG(extack, "Invalid netlink attributes while flushing TC action");
1413                 return -EINVAL;
1414         }
1415
1416         for (i = 1; i <= TCA_ACT_MAX_PRIO && tb[i]; i++) {
1417                 act = tcf_action_get_1(net, tb[i], n, portid, extack);
1418                 if (IS_ERR(act)) {
1419                         ret = PTR_ERR(act);
1420                         goto err;
1421                 }
1422                 attr_size += tcf_action_fill_size(act);
1423                 actions[i - 1] = act;
1424         }
1425
1426         attr_size = tcf_action_full_attrs_size(attr_size);
1427
1428         if (event == RTM_GETACTION)
1429                 ret = tcf_get_notify(net, portid, n, actions, event, extack);
1430         else { /* delete */
1431                 ret = tcf_del_notify(net, n, actions, portid, attr_size, extack);
1432                 if (ret)
1433                         goto err;
1434                 return 0;
1435         }
1436 err:
1437         tcf_action_put_many(actions);
1438         return ret;
1439 }
1440
1441 static int
1442 tcf_add_notify(struct net *net, struct nlmsghdr *n, struct tc_action *actions[],
1443                u32 portid, size_t attr_size, struct netlink_ext_ack *extack)
1444 {
1445         struct sk_buff *skb;
1446         int err = 0;
1447
1448         skb = alloc_skb(attr_size <= NLMSG_GOODSIZE ? NLMSG_GOODSIZE : attr_size,
1449                         GFP_KERNEL);
1450         if (!skb)
1451                 return -ENOBUFS;
1452
1453         if (tca_get_fill(skb, actions, portid, n->nlmsg_seq, n->nlmsg_flags,
1454                          RTM_NEWACTION, 0, 0) <= 0) {
1455                 NL_SET_ERR_MSG(extack, "Failed to fill netlink attributes while adding TC action");
1456                 kfree_skb(skb);
1457                 return -EINVAL;
1458         }
1459
1460         err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1461                              n->nlmsg_flags & NLM_F_ECHO);
1462         if (err > 0)
1463                 err = 0;
1464         return err;
1465 }
1466
1467 static int tcf_action_add(struct net *net, struct nlattr *nla,
1468                           struct nlmsghdr *n, u32 portid, int ovr,
1469                           struct netlink_ext_ack *extack)
1470 {
1471         size_t attr_size = 0;
1472         int loop, ret;
1473         struct tc_action *actions[TCA_ACT_MAX_PRIO] = {};
1474
1475         for (loop = 0; loop < 10; loop++) {
1476                 ret = tcf_action_init(net, NULL, nla, NULL, NULL, ovr, 0,
1477                                       actions, &attr_size, true, extack);
1478                 if (ret != -EAGAIN)
1479                         break;
1480         }
1481
1482         if (ret < 0)
1483                 return ret;
1484         ret = tcf_add_notify(net, n, actions, portid, attr_size, extack);
1485         if (ovr)
1486                 tcf_action_put_many(actions);
1487
1488         return ret;
1489 }
1490
1491 static const struct nla_policy tcaa_policy[TCA_ROOT_MAX + 1] = {
1492         [TCA_ROOT_FLAGS] = NLA_POLICY_BITFIELD32(TCA_ACT_FLAG_LARGE_DUMP_ON |
1493                                                  TCA_ACT_FLAG_TERSE_DUMP),
1494         [TCA_ROOT_TIME_DELTA]      = { .type = NLA_U32 },
1495 };
1496
1497 static int tc_ctl_action(struct sk_buff *skb, struct nlmsghdr *n,
1498                          struct netlink_ext_ack *extack)
1499 {
1500         struct net *net = sock_net(skb->sk);
1501         struct nlattr *tca[TCA_ROOT_MAX + 1];
1502         u32 portid = NETLINK_CB(skb).portid;
1503         int ret = 0, ovr = 0;
1504
1505         if ((n->nlmsg_type != RTM_GETACTION) &&
1506             !netlink_capable(skb, CAP_NET_ADMIN))
1507                 return -EPERM;
1508
1509         ret = nlmsg_parse_deprecated(n, sizeof(struct tcamsg), tca,
1510                                      TCA_ROOT_MAX, NULL, extack);
1511         if (ret < 0)
1512                 return ret;
1513
1514         if (tca[TCA_ACT_TAB] == NULL) {
1515                 NL_SET_ERR_MSG(extack, "Netlink action attributes missing");
1516                 return -EINVAL;
1517         }
1518
1519         /* n->nlmsg_flags & NLM_F_CREATE */
1520         switch (n->nlmsg_type) {
1521         case RTM_NEWACTION:
1522                 /* we are going to assume all other flags
1523                  * imply create only if it doesn't exist
1524                  * Note that CREATE | EXCL implies that
1525                  * but since we want avoid ambiguity (eg when flags
1526                  * is zero) then just set this
1527                  */
1528                 if (n->nlmsg_flags & NLM_F_REPLACE)
1529                         ovr = 1;
1530                 ret = tcf_action_add(net, tca[TCA_ACT_TAB], n, portid, ovr,
1531                                      extack);
1532                 break;
1533         case RTM_DELACTION:
1534                 ret = tca_action_gd(net, tca[TCA_ACT_TAB], n,
1535                                     portid, RTM_DELACTION, extack);
1536                 break;
1537         case RTM_GETACTION:
1538                 ret = tca_action_gd(net, tca[TCA_ACT_TAB], n,
1539                                     portid, RTM_GETACTION, extack);
1540                 break;
1541         default:
1542                 BUG();
1543         }
1544
1545         return ret;
1546 }
1547
1548 static struct nlattr *find_dump_kind(struct nlattr **nla)
1549 {
1550         struct nlattr *tb1, *tb2[TCA_ACT_MAX + 1];
1551         struct nlattr *tb[TCA_ACT_MAX_PRIO + 1];
1552         struct nlattr *kind;
1553
1554         tb1 = nla[TCA_ACT_TAB];
1555         if (tb1 == NULL)
1556                 return NULL;
1557
1558         if (nla_parse_deprecated(tb, TCA_ACT_MAX_PRIO, nla_data(tb1), NLMSG_ALIGN(nla_len(tb1)), NULL, NULL) < 0)
1559                 return NULL;
1560
1561         if (tb[1] == NULL)
1562                 return NULL;
1563         if (nla_parse_nested_deprecated(tb2, TCA_ACT_MAX, tb[1], tcf_action_policy, NULL) < 0)
1564                 return NULL;
1565         kind = tb2[TCA_ACT_KIND];
1566
1567         return kind;
1568 }
1569
1570 static int tc_dump_action(struct sk_buff *skb, struct netlink_callback *cb)
1571 {
1572         struct net *net = sock_net(skb->sk);
1573         struct nlmsghdr *nlh;
1574         unsigned char *b = skb_tail_pointer(skb);
1575         struct nlattr *nest;
1576         struct tc_action_ops *a_o;
1577         int ret = 0;
1578         struct tcamsg *t = (struct tcamsg *) nlmsg_data(cb->nlh);
1579         struct nlattr *tb[TCA_ROOT_MAX + 1];
1580         struct nlattr *count_attr = NULL;
1581         unsigned long jiffy_since = 0;
1582         struct nlattr *kind = NULL;
1583         struct nla_bitfield32 bf;
1584         u32 msecs_since = 0;
1585         u32 act_count = 0;
1586
1587         ret = nlmsg_parse_deprecated(cb->nlh, sizeof(struct tcamsg), tb,
1588                                      TCA_ROOT_MAX, tcaa_policy, cb->extack);
1589         if (ret < 0)
1590                 return ret;
1591
1592         kind = find_dump_kind(tb);
1593         if (kind == NULL) {
1594                 pr_info("tc_dump_action: action bad kind\n");
1595                 return 0;
1596         }
1597
1598         a_o = tc_lookup_action(kind);
1599         if (a_o == NULL)
1600                 return 0;
1601
1602         cb->args[2] = 0;
1603         if (tb[TCA_ROOT_FLAGS]) {
1604                 bf = nla_get_bitfield32(tb[TCA_ROOT_FLAGS]);
1605                 cb->args[2] = bf.value;
1606         }
1607
1608         if (tb[TCA_ROOT_TIME_DELTA]) {
1609                 msecs_since = nla_get_u32(tb[TCA_ROOT_TIME_DELTA]);
1610         }
1611
1612         nlh = nlmsg_put(skb, NETLINK_CB(cb->skb).portid, cb->nlh->nlmsg_seq,
1613                         cb->nlh->nlmsg_type, sizeof(*t), 0);
1614         if (!nlh)
1615                 goto out_module_put;
1616
1617         if (msecs_since)
1618                 jiffy_since = jiffies - msecs_to_jiffies(msecs_since);
1619
1620         t = nlmsg_data(nlh);
1621         t->tca_family = AF_UNSPEC;
1622         t->tca__pad1 = 0;
1623         t->tca__pad2 = 0;
1624         cb->args[3] = jiffy_since;
1625         count_attr = nla_reserve(skb, TCA_ROOT_COUNT, sizeof(u32));
1626         if (!count_attr)
1627                 goto out_module_put;
1628
1629         nest = nla_nest_start_noflag(skb, TCA_ACT_TAB);
1630         if (nest == NULL)
1631                 goto out_module_put;
1632
1633         ret = a_o->walk(net, skb, cb, RTM_GETACTION, a_o, NULL);
1634         if (ret < 0)
1635                 goto out_module_put;
1636
1637         if (ret > 0) {
1638                 nla_nest_end(skb, nest);
1639                 ret = skb->len;
1640                 act_count = cb->args[1];
1641                 memcpy(nla_data(count_attr), &act_count, sizeof(u32));
1642                 cb->args[1] = 0;
1643         } else
1644                 nlmsg_trim(skb, b);
1645
1646         nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1647         if (NETLINK_CB(cb->skb).portid && ret)
1648                 nlh->nlmsg_flags |= NLM_F_MULTI;
1649         module_put(a_o->owner);
1650         return skb->len;
1651
1652 out_module_put:
1653         module_put(a_o->owner);
1654         nlmsg_trim(skb, b);
1655         return skb->len;
1656 }
1657
1658 static int __init tc_action_init(void)
1659 {
1660         rtnl_register(PF_UNSPEC, RTM_NEWACTION, tc_ctl_action, NULL, 0);
1661         rtnl_register(PF_UNSPEC, RTM_DELACTION, tc_ctl_action, NULL, 0);
1662         rtnl_register(PF_UNSPEC, RTM_GETACTION, tc_ctl_action, tc_dump_action,
1663                       0);
1664
1665         return 0;
1666 }
1667
1668 subsys_initcall(tc_action_init);