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 static 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
 188 int tcf_idr_release(struct tc_action *a, bool bind)
 189 {
 190         const struct tc_action_ops *ops = a->ops;
 191         int ret;
 192
 193         ret = __tcf_idr_release(a, bind, false);
 194         if (ret == ACT_P_DELETED)
 195                 module_put(ops->owner);
 196         return ret;
 197 }
 198 EXPORT_SYMBOL(tcf_idr_release);
 199
 200 static size_t tcf_action_shared_attrs_size(const struct tc_action *act)
 201 {
 202         struct tc_cookie *act_cookie;
 203         u32 cookie_len = 0;
 204
 205         rcu_read_lock();
 206         act_cookie = rcu_dereference(act->act_cookie);
 207
 208         if (act_cookie)
 209                 cookie_len = nla_total_size(act_cookie->len);
 210         rcu_read_unlock();
 211
 212         return  nla_total_size(0) /* action number nested */
 213                 + nla_total_size(IFNAMSIZ) /* TCA_ACT_KIND */
 214                 + cookie_len /* TCA_ACT_COOKIE */
 215                 + nla_total_size(sizeof(struct nla_bitfield32)) /* TCA_ACT_HW_STATS */
 216                 + nla_total_size(0) /* TCA_ACT_STATS nested */
 217                 + nla_total_size(sizeof(struct nla_bitfield32)) /* TCA_ACT_FLAGS */
 218                 /* TCA_STATS_BASIC */
 219                 + nla_total_size_64bit(sizeof(struct gnet_stats_basic))
 220                 /* TCA_STATS_PKT64 */
 221                 + nla_total_size_64bit(sizeof(u64))
 222                 /* TCA_STATS_QUEUE */
 223                 + nla_total_size_64bit(sizeof(struct gnet_stats_queue))
 224                 + nla_total_size(0) /* TCA_OPTIONS nested */
 225                 + nla_total_size(sizeof(struct tcf_t)); /* TCA_GACT_TM */
 226 }
 227
 228 static size_t tcf_action_full_attrs_size(size_t sz)
 229 {
 230         return NLMSG_HDRLEN                     /* struct nlmsghdr */
 231                 + sizeof(struct tcamsg)
 232                 + nla_total_size(0)             /* TCA_ACT_TAB nested */
 233                 + sz;
 234 }
 235
 236 static size_t tcf_action_fill_size(const struct tc_action *act)
 237 {
 238         size_t sz = tcf_action_shared_attrs_size(act);
 239
 240         if (act->ops->get_fill_size)
 241                 return act->ops->get_fill_size(act) + sz;
 242         return sz;
 243 }
 244
 245 static int
 246 tcf_action_dump_terse(struct sk_buff *skb, struct tc_action *a, bool from_act)
 247 {
 248         unsigned char *b = skb_tail_pointer(skb);
 249         struct tc_cookie *cookie;
 250
 251         if (nla_put_string(skb, TCA_KIND, a->ops->kind))
 252                 goto nla_put_failure;
 253         if (tcf_action_copy_stats(skb, a, 0))
 254                 goto nla_put_failure;
 255         if (from_act && nla_put_u32(skb, TCA_ACT_INDEX, a->tcfa_index))
 256                 goto nla_put_failure;
 257
 258         rcu_read_lock();
 259         cookie = rcu_dereference(a->act_cookie);
 260         if (cookie) {
 261                 if (nla_put(skb, TCA_ACT_COOKIE, cookie->len, cookie->data)) {
 262                         rcu_read_unlock();
 263                         goto nla_put_failure;
 264                 }
 265         }
 266         rcu_read_unlock();
 267
 268         return 0;
 269
 270 nla_put_failure:
 271         nlmsg_trim(skb, b);
 272         return -1;
 273 }
 274
 275 static int tcf_dump_walker(struct tcf_idrinfo *idrinfo, struct sk_buff *skb,
 276                            struct netlink_callback *cb)
 277 {
 278         int err = 0, index = -1, s_i = 0, n_i = 0;
 279         u32 act_flags = cb->args[2];
 280         unsigned long jiffy_since = cb->args[3];
 281         struct nlattr *nest;
 282         struct idr *idr = &idrinfo->action_idr;
 283         struct tc_action *p;
 284         unsigned long id = 1;
 285         unsigned long tmp;
 286
 287         mutex_lock(&idrinfo->lock);
 288
 289         s_i = cb->args[0];
 290
 291         idr_for_each_entry_ul(idr, p, tmp, id) {
 292                 index++;
 293                 if (index < s_i)
 294                         continue;
 295                 if (IS_ERR(p))
 296                         continue;
 297
 298                 if (jiffy_since &&
 299                     time_after(jiffy_since,
 300                                (unsigned long)p->tcfa_tm.lastuse))
 301                         continue;
 302
 303                 nest = nla_nest_start_noflag(skb, n_i);
 304                 if (!nest) {
 305                         index--;
 306                         goto nla_put_failure;
 307                 }
 308                 err = (act_flags & TCA_ACT_FLAG_TERSE_DUMP) ?
 309                         tcf_action_dump_terse(skb, p, true) :
 310                         tcf_action_dump_1(skb, p, 0, 0);
 311                 if (err < 0) {
 312                         index--;
 313                         nlmsg_trim(skb, nest);
 314                         goto done;
 315                 }
 316                 nla_nest_end(skb, nest);
 317                 n_i++;
 318                 if (!(act_flags & TCA_ACT_FLAG_LARGE_DUMP_ON) &&
 319                     n_i >= TCA_ACT_MAX_PRIO)
 320                         goto done;
 321         }
 322 done:
 323         if (index >= 0)
 324                 cb->args[0] = index + 1;
 325
 326         mutex_unlock(&idrinfo->lock);
 327         if (n_i) {
 328                 if (act_flags & TCA_ACT_FLAG_LARGE_DUMP_ON)
 329                         cb->args[1] = n_i;
 330         }
 331         return n_i;
 332
 333 nla_put_failure:
 334         nla_nest_cancel(skb, nest);
 335         goto done;
 336 }
 337
 338 static int tcf_idr_release_unsafe(struct tc_action *p)
 339 {
 340         if (atomic_read(&p->tcfa_bindcnt) > 0)
 341                 return -EPERM;
 342
 343         if (refcount_dec_and_test(&p->tcfa_refcnt)) {
 344                 idr_remove(&p->idrinfo->action_idr, p->tcfa_index);
 345                 tcf_action_cleanup(p);
 346                 return ACT_P_DELETED;
 347         }
 348
 349         return 0;
 350 }
 351
 352 static int tcf_del_walker(struct tcf_idrinfo *idrinfo, struct sk_buff *skb,
 353                           const struct tc_action_ops *ops)
 354 {
 355         struct nlattr *nest;
 356         int n_i = 0;
 357         int ret = -EINVAL;
 358         struct idr *idr = &idrinfo->action_idr;
 359         struct tc_action *p;
 360         unsigned long id = 1;
 361         unsigned long tmp;
 362
 363         nest = nla_nest_start_noflag(skb, 0);
 364         if (nest == NULL)
 365                 goto nla_put_failure;
 366         if (nla_put_string(skb, TCA_KIND, ops->kind))
 367                 goto nla_put_failure;
 368
 369         mutex_lock(&idrinfo->lock);
 370         idr_for_each_entry_ul(idr, p, tmp, id) {
 371                 if (IS_ERR(p))
 372                         continue;
 373                 ret = tcf_idr_release_unsafe(p);
 374                 if (ret == ACT_P_DELETED) {
 375                         module_put(ops->owner);
 376                         n_i++;
 377                 } else if (ret < 0) {
 378                         mutex_unlock(&idrinfo->lock);
 379                         goto nla_put_failure;
 380                 }
 381         }
 382         mutex_unlock(&idrinfo->lock);
 383
 384         ret = nla_put_u32(skb, TCA_FCNT, n_i);
 385         if (ret)
 386                 goto nla_put_failure;
 387         nla_nest_end(skb, nest);
 388
 389         return n_i;
 390 nla_put_failure:
 391         nla_nest_cancel(skb, nest);
 392         return ret;
 393 }
 394
 395 int tcf_generic_walker(struct tc_action_net *tn, struct sk_buff *skb,
 396                        struct netlink_callback *cb, int type,
 397                        const struct tc_action_ops *ops,
 398                        struct netlink_ext_ack *extack)
 399 {
 400         struct tcf_idrinfo *idrinfo = tn->idrinfo;
 401
 402         if (type == RTM_DELACTION) {
 403                 return tcf_del_walker(idrinfo, skb, ops);
 404         } else if (type == RTM_GETACTION) {
 405                 return tcf_dump_walker(idrinfo, skb, cb);
 406         } else {
 407                 WARN(1, "tcf_generic_walker: unknown command %d\n", type);
 408                 NL_SET_ERR_MSG(extack, "tcf_generic_walker: unknown command");
 409                 return -EINVAL;
 410         }
 411 }
 412 EXPORT_SYMBOL(tcf_generic_walker);
 413
 414 int tcf_idr_search(struct tc_action_net *tn, struct tc_action **a, u32 index)
 415 {
 416         struct tcf_idrinfo *idrinfo = tn->idrinfo;
 417         struct tc_action *p;
 418
 419         mutex_lock(&idrinfo->lock);
 420         p = idr_find(&idrinfo->action_idr, index);
 421         if (IS_ERR(p))
 422                 p = NULL;
 423         else if (p)
 424                 refcount_inc(&p->tcfa_refcnt);
 425         mutex_unlock(&idrinfo->lock);
 426
 427         if (p) {
 428                 *a = p;
 429                 return true;
 430         }
 431         return false;
 432 }
 433 EXPORT_SYMBOL(tcf_idr_search);
 434
 435 static int tcf_idr_delete_index(struct tcf_idrinfo *idrinfo, u32 index)
 436 {
 437         struct tc_action *p;
 438         int ret = 0;
 439
 440         mutex_lock(&idrinfo->lock);
 441         p = idr_find(&idrinfo->action_idr, index);
 442         if (!p) {
 443                 mutex_unlock(&idrinfo->lock);
 444                 return -ENOENT;
 445         }
 446
 447         if (!atomic_read(&p->tcfa_bindcnt)) {
 448                 if (refcount_dec_and_test(&p->tcfa_refcnt)) {
 449                         struct module *owner = p->ops->owner;
 450
 451                         WARN_ON(p != idr_remove(&idrinfo->action_idr,
 452                                                 p->tcfa_index));
 453                         mutex_unlock(&idrinfo->lock);
 454
 455                         tcf_action_cleanup(p);
 456                         module_put(owner);
 457                         return 0;
 458                 }
 459                 ret = 0;
 460         } else {
 461                 ret = -EPERM;
 462         }
 463
 464         mutex_unlock(&idrinfo->lock);
 465         return ret;
 466 }
 467
 468 int tcf_idr_create(struct tc_action_net *tn, u32 index, struct nlattr *est,
 469                    struct tc_action **a, const struct tc_action_ops *ops,
 470                    int bind, bool cpustats, u32 flags)
 471 {
 472         struct tc_action *p = kzalloc(ops->size, GFP_KERNEL);
 473         struct tcf_idrinfo *idrinfo = tn->idrinfo;
 474         int err = -ENOMEM;
 475
 476         if (unlikely(!p))
 477                 return -ENOMEM;
 478         refcount_set(&p->tcfa_refcnt, 1);
 479         if (bind)
 480                 atomic_set(&p->tcfa_bindcnt, 1);
 481
 482         if (cpustats) {
 483                 p->cpu_bstats = netdev_alloc_pcpu_stats(struct gnet_stats_basic_cpu);
 484                 if (!p->cpu_bstats)
 485                         goto err1;
 486                 p->cpu_bstats_hw = netdev_alloc_pcpu_stats(struct gnet_stats_basic_cpu);
 487                 if (!p->cpu_bstats_hw)
 488                         goto err2;
 489                 p->cpu_qstats = alloc_percpu(struct gnet_stats_queue);
 490                 if (!p->cpu_qstats)
 491                         goto err3;
 492         }
 493         spin_lock_init(&p->tcfa_lock);
 494         p->tcfa_index = index;
 495         p->tcfa_tm.install = jiffies;
 496         p->tcfa_tm.lastuse = jiffies;
 497         p->tcfa_tm.firstuse = 0;
 498         p->tcfa_flags = flags;
 499         if (est) {
 500                 err = gen_new_estimator(&p->tcfa_bstats, p->cpu_bstats,
 501                                         &p->tcfa_rate_est,
 502                                         &p->tcfa_lock, NULL, est);
 503                 if (err)
 504                         goto err4;
 505         }
 506
 507         p->idrinfo = idrinfo;
 508         __module_get(ops->owner);
 509         p->ops = ops;
 510         *a = p;
 511         return 0;
 512 err4:
 513         free_percpu(p->cpu_qstats);
 514 err3:
 515         free_percpu(p->cpu_bstats_hw);
 516 err2:
 517         free_percpu(p->cpu_bstats);
 518 err1:
 519         kfree(p);
 520         return err;
 521 }
 522 EXPORT_SYMBOL(tcf_idr_create);
 523
 524 int tcf_idr_create_from_flags(struct tc_action_net *tn, u32 index,
 525                               struct nlattr *est, struct tc_action **a,
 526                               const struct tc_action_ops *ops, int bind,
 527                               u32 flags)
 528 {
 529         /* Set cpustats according to actions flags. */
 530         return tcf_idr_create(tn, index, est, a, ops, bind,
 531                               !(flags & TCA_ACT_FLAGS_NO_PERCPU_STATS), flags);
 532 }
 533 EXPORT_SYMBOL(tcf_idr_create_from_flags);
 534
 535 /* Cleanup idr index that was allocated but not initialized. */
 536
 537 void tcf_idr_cleanup(struct tc_action_net *tn, u32 index)
 538 {
 539         struct tcf_idrinfo *idrinfo = tn->idrinfo;
 540
 541         mutex_lock(&idrinfo->lock);
 542         /* Remove ERR_PTR(-EBUSY) allocated by tcf_idr_check_alloc */
 543         WARN_ON(!IS_ERR(idr_remove(&idrinfo->action_idr, index)));
 544         mutex_unlock(&idrinfo->lock);
 545 }
 546 EXPORT_SYMBOL(tcf_idr_cleanup);
 547
 548 /* Check if action with specified index exists. If actions is found, increments
 549  * its reference and bind counters, and return 1. Otherwise insert temporary
 550  * error pointer (to prevent concurrent users from inserting actions with same
 551  * index) and return 0.
 552  */
 553
 554 int tcf_idr_check_alloc(struct tc_action_net *tn, u32 *index,
 555                         struct tc_action **a, int bind)
 556 {
 557         struct tcf_idrinfo *idrinfo = tn->idrinfo;
 558         struct tc_action *p;
 559         int ret;
 560
 561 again:
 562         mutex_lock(&idrinfo->lock);
 563         if (*index) {
 564                 p = idr_find(&idrinfo->action_idr, *index);
 565                 if (IS_ERR(p)) {
 566                         /* This means that another process allocated
 567                          * index but did not assign the pointer yet.
 568                          */
 569                         mutex_unlock(&idrinfo->lock);
 570                         goto again;
 571                 }
 572
 573                 if (p) {
 574                         refcount_inc(&p->tcfa_refcnt);
 575                         if (bind)
 576                                 atomic_inc(&p->tcfa_bindcnt);
 577                         *a = p;
 578                         ret = 1;
 579                 } else {
 580                         *a = NULL;
 581                         ret = idr_alloc_u32(&idrinfo->action_idr, NULL, index,
 582                                             *index, GFP_KERNEL);
 583                         if (!ret)
 584                                 idr_replace(&idrinfo->action_idr,
 585                                             ERR_PTR(-EBUSY), *index);
 586                 }
 587         } else {
 588                 *index = 1;
 589                 *a = NULL;
 590                 ret = idr_alloc_u32(&idrinfo->action_idr, NULL, index,
 591                                     UINT_MAX, GFP_KERNEL);
 592                 if (!ret)
 593                         idr_replace(&idrinfo->action_idr, ERR_PTR(-EBUSY),
 594                                     *index);
 595         }
 596         mutex_unlock(&idrinfo->lock);
 597         return ret;
 598 }
 599 EXPORT_SYMBOL(tcf_idr_check_alloc);
 600
 601 void tcf_idrinfo_destroy(const struct tc_action_ops *ops,
 602                          struct tcf_idrinfo *idrinfo)
 603 {
 604         struct idr *idr = &idrinfo->action_idr;
 605         struct tc_action *p;
 606         int ret;
 607         unsigned long id = 1;
 608         unsigned long tmp;
 609
 610         idr_for_each_entry_ul(idr, p, tmp, id) {
 611                 ret = __tcf_idr_release(p, false, true);
 612                 if (ret == ACT_P_DELETED)
 613                         module_put(ops->owner);
 614                 else if (ret < 0)
 615                         return;
 616         }
 617         idr_destroy(&idrinfo->action_idr);
 618 }
 619 EXPORT_SYMBOL(tcf_idrinfo_destroy);
 620
 621 static LIST_HEAD(act_base);
 622 static DEFINE_RWLOCK(act_mod_lock);
 623
 624 int tcf_register_action(struct tc_action_ops *act,
 625                         struct pernet_operations *ops)
 626 {
 627         struct tc_action_ops *a;
 628         int ret;
 629
 630         if (!act->act || !act->dump || !act->init || !act->walk || !act->lookup)
 631                 return -EINVAL;
 632
 633         /* We have to register pernet ops before making the action ops visible,
 634          * otherwise tcf_action_init_1() could get a partially initialized
 635          * netns.
 636          */
 637         ret = register_pernet_subsys(ops);
 638         if (ret)
 639                 return ret;
 640
 641         write_lock(&act_mod_lock);
 642         list_for_each_entry(a, &act_base, head) {
 643                 if (act->id == a->id || (strcmp(act->kind, a->kind) == 0)) {
 644                         write_unlock(&act_mod_lock);
 645                         unregister_pernet_subsys(ops);
 646                         return -EEXIST;
 647                 }
 648         }
 649         list_add_tail(&act->head, &act_base);
 650         write_unlock(&act_mod_lock);
 651
 652         return 0;
 653 }
 654 EXPORT_SYMBOL(tcf_register_action);
 655
 656 int tcf_unregister_action(struct tc_action_ops *act,
 657                           struct pernet_operations *ops)
 658 {
 659         struct tc_action_ops *a;
 660         int err = -ENOENT;
 661
 662         write_lock(&act_mod_lock);
 663         list_for_each_entry(a, &act_base, head) {
 664                 if (a == act) {
 665                         list_del(&act->head);
 666                         err = 0;
 667                         break;
 668                 }
 669         }
 670         write_unlock(&act_mod_lock);
 671         if (!err)
 672                 unregister_pernet_subsys(ops);
 673         return err;
 674 }
 675 EXPORT_SYMBOL(tcf_unregister_action);
 676
 677 /* lookup by name */
 678 static struct tc_action_ops *tc_lookup_action_n(char *kind)
 679 {
 680         struct tc_action_ops *a, *res = NULL;
 681
 682         if (kind) {
 683                 read_lock(&act_mod_lock);
 684                 list_for_each_entry(a, &act_base, head) {
 685                         if (strcmp(kind, a->kind) == 0) {
 686                                 if (try_module_get(a->owner))
 687                                         res = a;
 688                                 break;
 689                         }
 690                 }
 691                 read_unlock(&act_mod_lock);
 692         }
 693         return res;
 694 }
 695
 696 /* lookup by nlattr */
 697 static struct tc_action_ops *tc_lookup_action(struct nlattr *kind)
 698 {
 699         struct tc_action_ops *a, *res = NULL;
 700
 701         if (kind) {
 702                 read_lock(&act_mod_lock);
 703                 list_for_each_entry(a, &act_base, head) {
 704                         if (nla_strcmp(kind, a->kind) == 0) {
 705                                 if (try_module_get(a->owner))
 706                                         res = a;
 707                                 break;
 708                         }
 709                 }
 710                 read_unlock(&act_mod_lock);
 711         }
 712         return res;
 713 }
 714
 715 /*TCA_ACT_MAX_PRIO is 32, there count up to 32 */
 716 #define TCA_ACT_MAX_PRIO_MASK 0x1FF
 717 int tcf_action_exec(struct sk_buff *skb, struct tc_action **actions,
 718                     int nr_actions, struct tcf_result *res)
 719 {
 720         u32 jmp_prgcnt = 0;
 721         u32 jmp_ttl = TCA_ACT_MAX_PRIO; /*matches actions per filter */
 722         int i;
 723         int ret = TC_ACT_OK;
 724
 725         if (skb_skip_tc_classify(skb))
 726                 return TC_ACT_OK;
 727
 728 restart_act_graph:
 729         for (i = 0; i < nr_actions; i++) {
 730                 const struct tc_action *a = actions[i];
 731
 732                 if (jmp_prgcnt > 0) {
 733                         jmp_prgcnt -= 1;
 734                         continue;
 735                 }
 736 repeat:
 737                 ret = a->ops->act(skb, a, res);
 738                 if (ret == TC_ACT_REPEAT)
 739                         goto repeat;    /* we need a ttl - JHS */
 740
 741                 if (TC_ACT_EXT_CMP(ret, TC_ACT_JUMP)) {
 742                         jmp_prgcnt = ret & TCA_ACT_MAX_PRIO_MASK;
 743                         if (!jmp_prgcnt || (jmp_prgcnt > nr_actions)) {
 744                                 /* faulty opcode, stop pipeline */
 745                                 return TC_ACT_OK;
 746                         } else {
 747                                 jmp_ttl -= 1;
 748                                 if (jmp_ttl > 0)
 749                                         goto restart_act_graph;
 750                                 else /* faulty graph, stop pipeline */
 751                                         return TC_ACT_OK;
 752                         }
 753                 } else if (TC_ACT_EXT_CMP(ret, TC_ACT_GOTO_CHAIN)) {
 754                         if (unlikely(!rcu_access_pointer(a->goto_chain))) {
 755                                 net_warn_ratelimited("can't go to NULL chain!\n");
 756                                 return TC_ACT_SHOT;
 757                         }
 758                         tcf_action_goto_chain_exec(a, res);
 759                 }
 760
 761                 if (ret != TC_ACT_PIPE)
 762                         break;
 763         }
 764
 765         return ret;
 766 }
 767 EXPORT_SYMBOL(tcf_action_exec);
 768
 769 int tcf_action_destroy(struct tc_action *actions[], int bind)
 770 {
 771         const struct tc_action_ops *ops;
 772         struct tc_action *a;
 773         int ret = 0, i;
 774
 775         for (i = 0; i < TCA_ACT_MAX_PRIO && actions[i]; i++) {
 776                 a = actions[i];
 777                 actions[i] = NULL;
 778                 ops = a->ops;
 779                 ret = __tcf_idr_release(a, bind, true);
 780                 if (ret == ACT_P_DELETED)
 781                         module_put(ops->owner);
 782                 else if (ret < 0)
 783                         return ret;
 784         }
 785         return ret;
 786 }
 787
 788 static int tcf_action_put(struct tc_action *p)
 789 {
 790         return __tcf_action_put(p, false);
 791 }
 792
 793 /* Put all actions in this array, skip those NULL's. */
 794 static void tcf_action_put_many(struct tc_action *actions[])
 795 {
 796         int i;
 797
 798         for (i = 0; i < TCA_ACT_MAX_PRIO; i++) {
 799                 struct tc_action *a = actions[i];
 800                 const struct tc_action_ops *ops;
 801
 802                 if (!a)
 803                         continue;
 804                 ops = a->ops;
 805                 if (tcf_action_put(a))
 806                         module_put(ops->owner);
 807         }
 808 }
 809
 810 int
 811 tcf_action_dump_old(struct sk_buff *skb, struct tc_action *a, int bind, int ref)
 812 {
 813         return a->ops->dump(skb, a, bind, ref);
 814 }
 815
 816 int
 817 tcf_action_dump_1(struct sk_buff *skb, struct tc_action *a, int bind, int ref)
 818 {
 819         int err = -EINVAL;
 820         unsigned char *b = skb_tail_pointer(skb);
 821         struct nlattr *nest;
 822
 823         if (tcf_action_dump_terse(skb, a, false))
 824                 goto nla_put_failure;
 825
 826         if (a->hw_stats != TCA_ACT_HW_STATS_ANY &&
 827             nla_put_bitfield32(skb, TCA_ACT_HW_STATS,
 828                                a->hw_stats, TCA_ACT_HW_STATS_ANY))
 829                 goto nla_put_failure;
 830
 831         if (a->used_hw_stats_valid &&
 832             nla_put_bitfield32(skb, TCA_ACT_USED_HW_STATS,
 833                                a->used_hw_stats, TCA_ACT_HW_STATS_ANY))
 834                 goto nla_put_failure;
 835
 836         if (a->tcfa_flags &&
 837             nla_put_bitfield32(skb, TCA_ACT_FLAGS,
 838                                a->tcfa_flags, a->tcfa_flags))
 839                 goto nla_put_failure;
 840
 841         nest = nla_nest_start_noflag(skb, TCA_OPTIONS);
 842         if (nest == NULL)
 843                 goto nla_put_failure;
 844         err = tcf_action_dump_old(skb, a, bind, ref);
 845         if (err > 0) {
 846                 nla_nest_end(skb, nest);
 847                 return err;
 848         }
 849
 850 nla_put_failure:
 851         nlmsg_trim(skb, b);
 852         return -1;
 853 }
 854 EXPORT_SYMBOL(tcf_action_dump_1);
 855
 856 int tcf_action_dump(struct sk_buff *skb, struct tc_action *actions[],
 857                     int bind, int ref, bool terse)
 858 {
 859         struct tc_action *a;
 860         int err = -EINVAL, i;
 861         struct nlattr *nest;
 862
 863         for (i = 0; i < TCA_ACT_MAX_PRIO && actions[i]; i++) {
 864                 a = actions[i];
 865                 nest = nla_nest_start_noflag(skb, i + 1);
 866                 if (nest == NULL)
 867                         goto nla_put_failure;
 868                 err = terse ? tcf_action_dump_terse(skb, a, false) :
 869                         tcf_action_dump_1(skb, a, bind, ref);
 870                 if (err < 0)
 871                         goto errout;
 872                 nla_nest_end(skb, nest);
 873         }
 874
 875         return 0;
 876
 877 nla_put_failure:
 878         err = -EINVAL;
 879 errout:
 880         nla_nest_cancel(skb, nest);
 881         return err;
 882 }
 883
 884 static struct tc_cookie *nla_memdup_cookie(struct nlattr **tb)
 885 {
 886         struct tc_cookie *c = kzalloc(sizeof(*c), GFP_KERNEL);
 887         if (!c)
 888                 return NULL;
 889
 890         c->data = nla_memdup(tb[TCA_ACT_COOKIE], GFP_KERNEL);
 891         if (!c->data) {
 892                 kfree(c);
 893                 return NULL;
 894         }
 895         c->len = nla_len(tb[TCA_ACT_COOKIE]);
 896
 897         return c;
 898 }
 899
 900 static u8 tcf_action_hw_stats_get(struct nlattr *hw_stats_attr)
 901 {
 902         struct nla_bitfield32 hw_stats_bf;
 903
 904         /* If the user did not pass the attr, that means he does
 905          * not care about the type. Return "any" in that case
 906          * which is setting on all supported types.
 907          */
 908         if (!hw_stats_attr)
 909                 return TCA_ACT_HW_STATS_ANY;
 910         hw_stats_bf = nla_get_bitfield32(hw_stats_attr);
 911         return hw_stats_bf.value;
 912 }
 913
 914 static const struct nla_policy tcf_action_policy[TCA_ACT_MAX + 1] = {
 915         [TCA_ACT_KIND]          = { .type = NLA_STRING },
 916         [TCA_ACT_INDEX]         = { .type = NLA_U32 },
 917         [TCA_ACT_COOKIE]        = { .type = NLA_BINARY,
 918                                     .len = TC_COOKIE_MAX_SIZE },
 919         [TCA_ACT_OPTIONS]       = { .type = NLA_NESTED },
 920         [TCA_ACT_FLAGS]         = NLA_POLICY_BITFIELD32(TCA_ACT_FLAGS_NO_PERCPU_STATS),
 921         [TCA_ACT_HW_STATS]      = NLA_POLICY_BITFIELD32(TCA_ACT_HW_STATS_ANY),
 922 };
 923
 924 void tcf_idr_insert_many(struct tc_action *actions[])
 925 {
 926         int i;
 927
 928         for (i = 0; i < TCA_ACT_MAX_PRIO; i++) {
 929                 struct tc_action *a = actions[i];
 930                 struct tcf_idrinfo *idrinfo;
 931
 932                 if (!a)
 933                         continue;
 934                 idrinfo = a->idrinfo;
 935                 mutex_lock(&idrinfo->lock);
 936                 /* Replace ERR_PTR(-EBUSY) allocated by tcf_idr_check_alloc if
 937                  * it is just created, otherwise this is just a nop.
 938                  */
 939                 idr_replace(&idrinfo->action_idr, a, a->tcfa_index);
 940                 mutex_unlock(&idrinfo->lock);
 941         }
 942 }
 943
 944 struct tc_action_ops *tc_action_load_ops(char *name, struct nlattr *nla,
 945                                          bool rtnl_held,
 946                                          struct netlink_ext_ack *extack)
 947 {
 948         struct nlattr *tb[TCA_ACT_MAX + 1];
 949         struct tc_action_ops *a_o;
 950         char act_name[IFNAMSIZ];
 951         struct nlattr *kind;
 952         int err;
 953
 954         if (name == NULL) {
 955                 err = nla_parse_nested_deprecated(tb, TCA_ACT_MAX, nla,
 956                                                   tcf_action_policy, extack);
 957                 if (err < 0)
 958                         return ERR_PTR(err);
 959                 err = -EINVAL;
 960                 kind = tb[TCA_ACT_KIND];
 961                 if (!kind) {
 962                         NL_SET_ERR_MSG(extack, "TC action kind must be specified");
 963                         return ERR_PTR(err);
 964                 }
 965                 if (nla_strscpy(act_name, kind, IFNAMSIZ) < 0) {
 966                         NL_SET_ERR_MSG(extack, "TC action name too long");
 967                         return ERR_PTR(err);
 968                 }
 969         } else {
 970                 if (strlcpy(act_name, name, IFNAMSIZ) >= IFNAMSIZ) {
 971                         NL_SET_ERR_MSG(extack, "TC action name too long");
 972                         return ERR_PTR(-EINVAL);
 973                 }
 974         }
 975
 976         a_o = tc_lookup_action_n(act_name);
 977         if (a_o == NULL) {
 978 #ifdef CONFIG_MODULES
 979                 if (rtnl_held)
 980                         rtnl_unlock();
 981                 request_module("act_%s", act_name);
 982                 if (rtnl_held)
 983                         rtnl_lock();
 984
 985                 a_o = tc_lookup_action_n(act_name);
 986
 987                 /* We dropped the RTNL semaphore in order to
 988                  * perform the module load.  So, even if we
 989                  * succeeded in loading the module we have to
 990                  * tell the caller to replay the request.  We
 991                  * indicate this using -EAGAIN.
 992                  */
 993                 if (a_o != NULL) {
 994                         module_put(a_o->owner);
 995                         return ERR_PTR(-EAGAIN);
 996                 }
 997 #endif
 998                 NL_SET_ERR_MSG(extack, "Failed to load TC action module");
 999                 return ERR_PTR(-ENOENT);
1000         }
1001
1002         return a_o;
1003 }
1004
1005 struct tc_action *tcf_action_init_1(struct net *net, struct tcf_proto *tp,
1006                                     struct nlattr *nla, struct nlattr *est,
1007                                     char *name, int ovr, int bind,
1008                                     struct tc_action_ops *a_o, int *init_res,
1009                                     bool rtnl_held,
1010                                     struct netlink_ext_ack *extack)
1011 {
1012         struct nla_bitfield32 flags = { 0, 0 };
1013         u8 hw_stats = TCA_ACT_HW_STATS_ANY;
1014         struct nlattr *tb[TCA_ACT_MAX + 1];
1015         struct tc_cookie *cookie = NULL;
1016         struct tc_action *a;
1017         int err;
1018
1019         /* backward compatibility for policer */
1020         if (name == NULL) {
1021                 err = nla_parse_nested_deprecated(tb, TCA_ACT_MAX, nla,
1022                                                   tcf_action_policy, extack);
1023                 if (err < 0)
1024                         return ERR_PTR(err);
1025                 if (tb[TCA_ACT_COOKIE]) {
1026                         cookie = nla_memdup_cookie(tb);
1027                         if (!cookie) {
1028                                 NL_SET_ERR_MSG(extack, "No memory to generate TC cookie");
1029                                 err = -ENOMEM;
1030                                 goto err_out;
1031                         }
1032                 }
1033                 hw_stats = tcf_action_hw_stats_get(tb[TCA_ACT_HW_STATS]);
1034                 if (tb[TCA_ACT_FLAGS])
1035                         flags = nla_get_bitfield32(tb[TCA_ACT_FLAGS]);
1036
1037                 err = a_o->init(net, tb[TCA_ACT_OPTIONS], est, &a, ovr, bind,
1038                                 rtnl_held, tp, flags.value, extack);
1039         } else {
1040                 err = a_o->init(net, nla, est, &a, ovr, bind, rtnl_held,
1041                                 tp, flags.value, extack);
1042         }
1043         if (err < 0)
1044                 goto err_out;
1045         *init_res = err;
1046
1047         if (!name && tb[TCA_ACT_COOKIE])
1048                 tcf_set_action_cookie(&a->act_cookie, cookie);
1049
1050         if (!name)
1051                 a->hw_stats = hw_stats;
1052
1053         return a;
1054
1055 err_out:
1056         if (cookie) {
1057                 kfree(cookie->data);
1058                 kfree(cookie);
1059         }
1060         return ERR_PTR(err);
1061 }
1062
1063 /* Returns numbers of initialized actions or negative error. */
1064
1065 int tcf_action_init(struct net *net, struct tcf_proto *tp, struct nlattr *nla,
1066                     struct nlattr *est, char *name, int ovr, int bind,
1067                     struct tc_action *actions[], int init_res[], size_t *attr_size,
1068                     bool rtnl_held, struct netlink_ext_ack *extack)
1069 {
1070         struct tc_action_ops *ops[TCA_ACT_MAX_PRIO] = {};
1071         struct nlattr *tb[TCA_ACT_MAX_PRIO + 1];
1072         struct tc_action *act;
1073         size_t sz = 0;
1074         int err;
1075         int i;
1076
1077         err = nla_parse_nested_deprecated(tb, TCA_ACT_MAX_PRIO, nla, NULL,
1078                                           extack);
1079         if (err < 0)
1080                 return err;
1081
1082         for (i = 1; i <= TCA_ACT_MAX_PRIO && tb[i]; i++) {
1083                 struct tc_action_ops *a_o;
1084
1085                 a_o = tc_action_load_ops(name, tb[i], rtnl_held, extack);
1086                 if (IS_ERR(a_o)) {
1087                         err = PTR_ERR(a_o);
1088                         goto err_mod;
1089                 }
1090                 ops[i - 1] = a_o;
1091         }
1092
1093         for (i = 1; i <= TCA_ACT_MAX_PRIO && tb[i]; i++) {
1094                 act = tcf_action_init_1(net, tp, tb[i], est, name, ovr, bind,
1095                                         ops[i - 1], &init_res[i - 1], rtnl_held,
1096                                         extack);
1097                 if (IS_ERR(act)) {
1098                         err = PTR_ERR(act);
1099                         goto err;
1100                 }
1101                 sz += tcf_action_fill_size(act);
1102                 /* Start from index 0 */
1103                 actions[i - 1] = act;
1104         }
1105
1106         /* We have to commit them all together, because if any error happened in
1107          * between, we could not handle the failure gracefully.
1108          */
1109         tcf_idr_insert_many(actions);
1110
1111         *attr_size = tcf_action_full_attrs_size(sz);
1112         err = i - 1;
1113         goto err_mod;
1114
1115 err:
1116         tcf_action_destroy(actions, bind);
1117 err_mod:
1118         for (i = 0; i < TCA_ACT_MAX_PRIO; i++) {
1119                 if (ops[i])
1120                         module_put(ops[i]->owner);
1121         }
1122         return err;
1123 }
1124
1125 void tcf_action_update_stats(struct tc_action *a, u64 bytes, u64 packets,
1126                              u64 drops, bool hw)
1127 {
1128         if (a->cpu_bstats) {
1129                 _bstats_cpu_update(this_cpu_ptr(a->cpu_bstats), bytes, packets);
1130
1131                 this_cpu_ptr(a->cpu_qstats)->drops += drops;
1132
1133                 if (hw)
1134                         _bstats_cpu_update(this_cpu_ptr(a->cpu_bstats_hw),
1135                                            bytes, packets);
1136                 return;
1137         }
1138
1139         _bstats_update(&a->tcfa_bstats, bytes, packets);
1140         a->tcfa_qstats.drops += drops;
1141         if (hw)
1142                 _bstats_update(&a->tcfa_bstats_hw, bytes, packets);
1143 }
1144 EXPORT_SYMBOL(tcf_action_update_stats);
1145
1146 int tcf_action_copy_stats(struct sk_buff *skb, struct tc_action *p,
1147                           int compat_mode)
1148 {
1149         int err = 0;
1150         struct gnet_dump d;
1151
1152         if (p == NULL)
1153                 goto errout;
1154
1155         /* compat_mode being true specifies a call that is supposed
1156          * to add additional backward compatibility statistic TLVs.
1157          */
1158         if (compat_mode) {
1159                 if (p->type == TCA_OLD_COMPAT)
1160                         err = gnet_stats_start_copy_compat(skb, 0,
1161                                                            TCA_STATS,
1162                                                            TCA_XSTATS,
1163                                                            &p->tcfa_lock, &d,
1164                                                            TCA_PAD);
1165                 else
1166                         return 0;
1167         } else
1168                 err = gnet_stats_start_copy(skb, TCA_ACT_STATS,
1169                                             &p->tcfa_lock, &d, TCA_ACT_PAD);
1170
1171         if (err < 0)
1172                 goto errout;
1173
1174         if (gnet_stats_copy_basic(NULL, &d, p->cpu_bstats, &p->tcfa_bstats) < 0 ||
1175             gnet_stats_copy_basic_hw(NULL, &d, p->cpu_bstats_hw,
1176                                      &p->tcfa_bstats_hw) < 0 ||
1177             gnet_stats_copy_rate_est(&d, &p->tcfa_rate_est) < 0 ||
1178             gnet_stats_copy_queue(&d, p->cpu_qstats,
1179                                   &p->tcfa_qstats,
1180                                   p->tcfa_qstats.qlen) < 0)
1181                 goto errout;
1182
1183         if (gnet_stats_finish_copy(&d) < 0)
1184                 goto errout;
1185
1186         return 0;
1187
1188 errout:
1189         return -1;
1190 }
1191
1192 static int tca_get_fill(struct sk_buff *skb, struct tc_action *actions[],
1193                         u32 portid, u32 seq, u16 flags, int event, int bind,
1194                         int ref)
1195 {
1196         struct tcamsg *t;
1197         struct nlmsghdr *nlh;
1198         unsigned char *b = skb_tail_pointer(skb);
1199         struct nlattr *nest;
1200
1201         nlh = nlmsg_put(skb, portid, seq, event, sizeof(*t), flags);
1202         if (!nlh)
1203                 goto out_nlmsg_trim;
1204         t = nlmsg_data(nlh);
1205         t->tca_family = AF_UNSPEC;
1206         t->tca__pad1 = 0;
1207         t->tca__pad2 = 0;
1208
1209         nest = nla_nest_start_noflag(skb, TCA_ACT_TAB);
1210         if (!nest)
1211                 goto out_nlmsg_trim;
1212
1213         if (tcf_action_dump(skb, actions, bind, ref, false) < 0)
1214                 goto out_nlmsg_trim;
1215
1216         nla_nest_end(skb, nest);
1217
1218         nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1219         return skb->len;
1220
1221 out_nlmsg_trim:
1222         nlmsg_trim(skb, b);
1223         return -1;
1224 }
1225
1226 static int
1227 tcf_get_notify(struct net *net, u32 portid, struct nlmsghdr *n,
1228                struct tc_action *actions[], int event,
1229                struct netlink_ext_ack *extack)
1230 {
1231         struct sk_buff *skb;
1232
1233         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1234         if (!skb)
1235                 return -ENOBUFS;
1236         if (tca_get_fill(skb, actions, portid, n->nlmsg_seq, 0, event,
1237                          0, 1) <= 0) {
1238                 NL_SET_ERR_MSG(extack, "Failed to fill netlink attributes while adding TC action");
1239                 kfree_skb(skb);
1240                 return -EINVAL;
1241         }
1242
1243         return rtnl_unicast(skb, net, portid);
1244 }
1245
1246 static struct tc_action *tcf_action_get_1(struct net *net, struct nlattr *nla,
1247                                           struct nlmsghdr *n, u32 portid,
1248                                           struct netlink_ext_ack *extack)
1249 {
1250         struct nlattr *tb[TCA_ACT_MAX + 1];
1251         const struct tc_action_ops *ops;
1252         struct tc_action *a;
1253         int index;
1254         int err;
1255
1256         err = nla_parse_nested_deprecated(tb, TCA_ACT_MAX, nla,
1257                                           tcf_action_policy, extack);
1258         if (err < 0)
1259                 goto err_out;
1260
1261         err = -EINVAL;
1262         if (tb[TCA_ACT_INDEX] == NULL ||
1263             nla_len(tb[TCA_ACT_INDEX]) < sizeof(index)) {
1264                 NL_SET_ERR_MSG(extack, "Invalid TC action index value");
1265                 goto err_out;
1266         }
1267         index = nla_get_u32(tb[TCA_ACT_INDEX]);
1268
1269         err = -EINVAL;
1270         ops = tc_lookup_action(tb[TCA_ACT_KIND]);
1271         if (!ops) { /* could happen in batch of actions */
1272                 NL_SET_ERR_MSG(extack, "Specified TC action kind not found");
1273                 goto err_out;
1274         }
1275         err = -ENOENT;
1276         if (ops->lookup(net, &a, index) == 0) {
1277                 NL_SET_ERR_MSG(extack, "TC action with specified index not found");
1278                 goto err_mod;
1279         }
1280
1281         module_put(ops->owner);
1282         return a;
1283
1284 err_mod:
1285         module_put(ops->owner);
1286 err_out:
1287         return ERR_PTR(err);
1288 }
1289
1290 static int tca_action_flush(struct net *net, struct nlattr *nla,
1291                             struct nlmsghdr *n, u32 portid,
1292                             struct netlink_ext_ack *extack)
1293 {
1294         struct sk_buff *skb;
1295         unsigned char *b;
1296         struct nlmsghdr *nlh;
1297         struct tcamsg *t;
1298         struct netlink_callback dcb;
1299         struct nlattr *nest;
1300         struct nlattr *tb[TCA_ACT_MAX + 1];
1301         const struct tc_action_ops *ops;
1302         struct nlattr *kind;
1303         int err = -ENOMEM;
1304
1305         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1306         if (!skb)
1307                 return err;
1308
1309         b = skb_tail_pointer(skb);
1310
1311         err = nla_parse_nested_deprecated(tb, TCA_ACT_MAX, nla,
1312                                           tcf_action_policy, extack);
1313         if (err < 0)
1314                 goto err_out;
1315
1316         err = -EINVAL;
1317         kind = tb[TCA_ACT_KIND];
1318         ops = tc_lookup_action(kind);
1319         if (!ops) { /*some idjot trying to flush unknown action */
1320                 NL_SET_ERR_MSG(extack, "Cannot flush unknown TC action");
1321                 goto err_out;
1322         }
1323
1324         nlh = nlmsg_put(skb, portid, n->nlmsg_seq, RTM_DELACTION,
1325                         sizeof(*t), 0);
1326         if (!nlh) {
1327                 NL_SET_ERR_MSG(extack, "Failed to create TC action flush notification");
1328                 goto out_module_put;
1329         }
1330         t = nlmsg_data(nlh);
1331         t->tca_family = AF_UNSPEC;
1332         t->tca__pad1 = 0;
1333         t->tca__pad2 = 0;
1334
1335         nest = nla_nest_start_noflag(skb, TCA_ACT_TAB);
1336         if (!nest) {
1337                 NL_SET_ERR_MSG(extack, "Failed to add new netlink message");
1338                 goto out_module_put;
1339         }
1340
1341         err = ops->walk(net, skb, &dcb, RTM_DELACTION, ops, extack);
1342         if (err <= 0) {
1343                 nla_nest_cancel(skb, nest);
1344                 goto out_module_put;
1345         }
1346
1347         nla_nest_end(skb, nest);
1348
1349         nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1350         nlh->nlmsg_flags |= NLM_F_ROOT;
1351         module_put(ops->owner);
1352         err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1353                              n->nlmsg_flags & NLM_F_ECHO);
1354         if (err > 0)
1355                 return 0;
1356         if (err < 0)
1357                 NL_SET_ERR_MSG(extack, "Failed to send TC action flush notification");
1358
1359         return err;
1360
1361 out_module_put:
1362         module_put(ops->owner);
1363 err_out:
1364         kfree_skb(skb);
1365         return err;
1366 }
1367
1368 static int tcf_action_delete(struct net *net, struct tc_action *actions[])
1369 {
1370         int i;
1371
1372         for (i = 0; i < TCA_ACT_MAX_PRIO && actions[i]; i++) {
1373                 struct tc_action *a = actions[i];
1374                 const struct tc_action_ops *ops = a->ops;
1375                 /* Actions can be deleted concurrently so we must save their
1376                  * type and id to search again after reference is released.
1377                  */
1378                 struct tcf_idrinfo *idrinfo = a->idrinfo;
1379                 u32 act_index = a->tcfa_index;
1380
1381                 actions[i] = NULL;
1382                 if (tcf_action_put(a)) {
1383                         /* last reference, action was deleted concurrently */
1384                         module_put(ops->owner);
1385                 } else  {
1386                         int ret;
1387
1388                         /* now do the delete */
1389                         ret = tcf_idr_delete_index(idrinfo, act_index);
1390                         if (ret < 0)
1391                                 return ret;
1392                 }
1393         }
1394         return 0;
1395 }
1396
1397 static int
1398 tcf_del_notify(struct net *net, struct nlmsghdr *n, struct tc_action *actions[],
1399                u32 portid, size_t attr_size, struct netlink_ext_ack *extack)
1400 {
1401         int ret;
1402         struct sk_buff *skb;
1403
1404         skb = alloc_skb(attr_size <= NLMSG_GOODSIZE ? NLMSG_GOODSIZE : attr_size,
1405                         GFP_KERNEL);
1406         if (!skb)
1407                 return -ENOBUFS;
1408
1409         if (tca_get_fill(skb, actions, portid, n->nlmsg_seq, 0, RTM_DELACTION,
1410                          0, 2) <= 0) {
1411                 NL_SET_ERR_MSG(extack, "Failed to fill netlink TC action attributes");
1412                 kfree_skb(skb);
1413                 return -EINVAL;
1414         }
1415
1416         /* now do the delete */
1417         ret = tcf_action_delete(net, actions);
1418         if (ret < 0) {
1419                 NL_SET_ERR_MSG(extack, "Failed to delete TC action");
1420                 kfree_skb(skb);
1421                 return ret;
1422         }
1423
1424         ret = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1425                              n->nlmsg_flags & NLM_F_ECHO);
1426         if (ret > 0)
1427                 return 0;
1428         return ret;
1429 }
1430
1431 static int
1432 tca_action_gd(struct net *net, struct nlattr *nla, struct nlmsghdr *n,
1433               u32 portid, int event, struct netlink_ext_ack *extack)
1434 {
1435         int i, ret;
1436         struct nlattr *tb[TCA_ACT_MAX_PRIO + 1];
1437         struct tc_action *act;
1438         size_t attr_size = 0;
1439         struct tc_action *actions[TCA_ACT_MAX_PRIO] = {};
1440
1441         ret = nla_parse_nested_deprecated(tb, TCA_ACT_MAX_PRIO, nla, NULL,
1442                                           extack);
1443         if (ret < 0)
1444                 return ret;
1445
1446         if (event == RTM_DELACTION && n->nlmsg_flags & NLM_F_ROOT) {
1447                 if (tb[1])
1448                         return tca_action_flush(net, tb[1], n, portid, extack);
1449
1450                 NL_SET_ERR_MSG(extack, "Invalid netlink attributes while flushing TC action");
1451                 return -EINVAL;
1452         }
1453
1454         for (i = 1; i <= TCA_ACT_MAX_PRIO && tb[i]; i++) {
1455                 act = tcf_action_get_1(net, tb[i], n, portid, extack);
1456                 if (IS_ERR(act)) {
1457                         ret = PTR_ERR(act);
1458                         goto err;
1459                 }
1460                 attr_size += tcf_action_fill_size(act);
1461                 actions[i - 1] = act;
1462         }
1463
1464         attr_size = tcf_action_full_attrs_size(attr_size);
1465
1466         if (event == RTM_GETACTION)
1467                 ret = tcf_get_notify(net, portid, n, actions, event, extack);
1468         else { /* delete */
1469                 ret = tcf_del_notify(net, n, actions, portid, attr_size, extack);
1470                 if (ret)
1471                         goto err;
1472                 return 0;
1473         }
1474 err:
1475         tcf_action_put_many(actions);
1476         return ret;
1477 }
1478
1479 static int
1480 tcf_add_notify(struct net *net, struct nlmsghdr *n, struct tc_action *actions[],
1481                u32 portid, size_t attr_size, struct netlink_ext_ack *extack)
1482 {
1483         struct sk_buff *skb;
1484         int err = 0;
1485
1486         skb = alloc_skb(attr_size <= NLMSG_GOODSIZE ? NLMSG_GOODSIZE : attr_size,
1487                         GFP_KERNEL);
1488         if (!skb)
1489                 return -ENOBUFS;
1490
1491         if (tca_get_fill(skb, actions, portid, n->nlmsg_seq, n->nlmsg_flags,
1492                          RTM_NEWACTION, 0, 0) <= 0) {
1493                 NL_SET_ERR_MSG(extack, "Failed to fill netlink attributes while adding TC action");
1494                 kfree_skb(skb);
1495                 return -EINVAL;
1496         }
1497
1498         err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1499                              n->nlmsg_flags & NLM_F_ECHO);
1500         if (err > 0)
1501                 err = 0;
1502         return err;
1503 }
1504
1505 static int tcf_action_add(struct net *net, struct nlattr *nla,
1506                           struct nlmsghdr *n, u32 portid, int ovr,
1507                           struct netlink_ext_ack *extack)
1508 {
1509         size_t attr_size = 0;
1510         int loop, ret, i;
1511         struct tc_action *actions[TCA_ACT_MAX_PRIO] = {};
1512         int init_res[TCA_ACT_MAX_PRIO] = {};
1513
1514         for (loop = 0; loop < 10; loop++) {
1515                 ret = tcf_action_init(net, NULL, nla, NULL, NULL, ovr, 0,
1516                                       actions, init_res, &attr_size, true, extack);
1517                 if (ret != -EAGAIN)
1518                         break;
1519         }
1520
1521         if (ret < 0)
1522                 return ret;
1523         ret = tcf_add_notify(net, n, actions, portid, attr_size, extack);
1524
1525         /* only put existing actions */
1526         for (i = 0; i < TCA_ACT_MAX_PRIO; i++)
1527                 if (init_res[i] == ACT_P_CREATED)
1528                         actions[i] = NULL;
1529         tcf_action_put_many(actions);
1530
1531         return ret;
1532 }
1533
1534 static const struct nla_policy tcaa_policy[TCA_ROOT_MAX + 1] = {
1535         [TCA_ROOT_FLAGS] = NLA_POLICY_BITFIELD32(TCA_ACT_FLAG_LARGE_DUMP_ON |
1536                                                  TCA_ACT_FLAG_TERSE_DUMP),
1537         [TCA_ROOT_TIME_DELTA]      = { .type = NLA_U32 },
1538 };
1539
1540 static int tc_ctl_action(struct sk_buff *skb, struct nlmsghdr *n,
1541                          struct netlink_ext_ack *extack)
1542 {
1543         struct net *net = sock_net(skb->sk);
1544         struct nlattr *tca[TCA_ROOT_MAX + 1];
1545         u32 portid = NETLINK_CB(skb).portid;
1546         int ret = 0, ovr = 0;
1547
1548         if ((n->nlmsg_type != RTM_GETACTION) &&
1549             !netlink_capable(skb, CAP_NET_ADMIN))
1550                 return -EPERM;
1551
1552         ret = nlmsg_parse_deprecated(n, sizeof(struct tcamsg), tca,
1553                                      TCA_ROOT_MAX, NULL, extack);
1554         if (ret < 0)
1555                 return ret;
1556
1557         if (tca[TCA_ACT_TAB] == NULL) {
1558                 NL_SET_ERR_MSG(extack, "Netlink action attributes missing");
1559                 return -EINVAL;
1560         }
1561
1562         /* n->nlmsg_flags & NLM_F_CREATE */
1563         switch (n->nlmsg_type) {
1564         case RTM_NEWACTION:
1565                 /* we are going to assume all other flags
1566                  * imply create only if it doesn't exist
1567                  * Note that CREATE | EXCL implies that
1568                  * but since we want avoid ambiguity (eg when flags
1569                  * is zero) then just set this
1570                  */
1571                 if (n->nlmsg_flags & NLM_F_REPLACE)
1572                         ovr = 1;
1573                 ret = tcf_action_add(net, tca[TCA_ACT_TAB], n, portid, ovr,
1574                                      extack);
1575                 break;
1576         case RTM_DELACTION:
1577                 ret = tca_action_gd(net, tca[TCA_ACT_TAB], n,
1578                                     portid, RTM_DELACTION, extack);
1579                 break;
1580         case RTM_GETACTION:
1581                 ret = tca_action_gd(net, tca[TCA_ACT_TAB], n,
1582                                     portid, RTM_GETACTION, extack);
1583                 break;
1584         default:
1585                 BUG();
1586         }
1587
1588         return ret;
1589 }
1590
1591 static struct nlattr *find_dump_kind(struct nlattr **nla)
1592 {
1593         struct nlattr *tb1, *tb2[TCA_ACT_MAX + 1];
1594         struct nlattr *tb[TCA_ACT_MAX_PRIO + 1];
1595         struct nlattr *kind;
1596
1597         tb1 = nla[TCA_ACT_TAB];
1598         if (tb1 == NULL)
1599                 return NULL;
1600
1601         if (nla_parse_deprecated(tb, TCA_ACT_MAX_PRIO, nla_data(tb1), NLMSG_ALIGN(nla_len(tb1)), NULL, NULL) < 0)
1602                 return NULL;
1603
1604         if (tb[1] == NULL)
1605                 return NULL;
1606         if (nla_parse_nested_deprecated(tb2, TCA_ACT_MAX, tb[1], tcf_action_policy, NULL) < 0)
1607                 return NULL;
1608         kind = tb2[TCA_ACT_KIND];
1609
1610         return kind;
1611 }
1612
1613 static int tc_dump_action(struct sk_buff *skb, struct netlink_callback *cb)
1614 {
1615         struct net *net = sock_net(skb->sk);
1616         struct nlmsghdr *nlh;
1617         unsigned char *b = skb_tail_pointer(skb);
1618         struct nlattr *nest;
1619         struct tc_action_ops *a_o;
1620         int ret = 0;
1621         struct tcamsg *t = (struct tcamsg *) nlmsg_data(cb->nlh);
1622         struct nlattr *tb[TCA_ROOT_MAX + 1];
1623         struct nlattr *count_attr = NULL;
1624         unsigned long jiffy_since = 0;
1625         struct nlattr *kind = NULL;
1626         struct nla_bitfield32 bf;
1627         u32 msecs_since = 0;
1628         u32 act_count = 0;
1629
1630         ret = nlmsg_parse_deprecated(cb->nlh, sizeof(struct tcamsg), tb,
1631                                      TCA_ROOT_MAX, tcaa_policy, cb->extack);
1632         if (ret < 0)
1633                 return ret;
1634
1635         kind = find_dump_kind(tb);
1636         if (kind == NULL) {
1637                 pr_info("tc_dump_action: action bad kind\n");
1638                 return 0;
1639         }
1640
1641         a_o = tc_lookup_action(kind);
1642         if (a_o == NULL)
1643                 return 0;
1644
1645         cb->args[2] = 0;
1646         if (tb[TCA_ROOT_FLAGS]) {
1647                 bf = nla_get_bitfield32(tb[TCA_ROOT_FLAGS]);
1648                 cb->args[2] = bf.value;
1649         }
1650
1651         if (tb[TCA_ROOT_TIME_DELTA]) {
1652                 msecs_since = nla_get_u32(tb[TCA_ROOT_TIME_DELTA]);
1653         }
1654
1655         nlh = nlmsg_put(skb, NETLINK_CB(cb->skb).portid, cb->nlh->nlmsg_seq,
1656                         cb->nlh->nlmsg_type, sizeof(*t), 0);
1657         if (!nlh)
1658                 goto out_module_put;
1659
1660         if (msecs_since)
1661                 jiffy_since = jiffies - msecs_to_jiffies(msecs_since);
1662
1663         t = nlmsg_data(nlh);
1664         t->tca_family = AF_UNSPEC;
1665         t->tca__pad1 = 0;
1666         t->tca__pad2 = 0;
1667         cb->args[3] = jiffy_since;
1668         count_attr = nla_reserve(skb, TCA_ROOT_COUNT, sizeof(u32));
1669         if (!count_attr)
1670                 goto out_module_put;
1671
1672         nest = nla_nest_start_noflag(skb, TCA_ACT_TAB);
1673         if (nest == NULL)
1674                 goto out_module_put;
1675
1676         ret = a_o->walk(net, skb, cb, RTM_GETACTION, a_o, NULL);
1677         if (ret < 0)
1678                 goto out_module_put;
1679
1680         if (ret > 0) {
1681                 nla_nest_end(skb, nest);
1682                 ret = skb->len;
1683                 act_count = cb->args[1];
1684                 memcpy(nla_data(count_attr), &act_count, sizeof(u32));
1685                 cb->args[1] = 0;
1686         } else
1687                 nlmsg_trim(skb, b);
1688
1689         nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1690         if (NETLINK_CB(cb->skb).portid && ret)
1691                 nlh->nlmsg_flags |= NLM_F_MULTI;
1692         module_put(a_o->owner);
1693         return skb->len;
1694
1695 out_module_put:
1696         module_put(a_o->owner);
1697         nlmsg_trim(skb, b);
1698         return skb->len;
1699 }
1700
1701 static int __init tc_action_init(void)
1702 {
1703         rtnl_register(PF_UNSPEC, RTM_NEWACTION, tc_ctl_action, NULL, 0);
1704         rtnl_register(PF_UNSPEC, RTM_DELACTION, tc_ctl_action, NULL, 0);
1705         rtnl_register(PF_UNSPEC, RTM_GETACTION, tc_ctl_action, tc_dump_action,
1706                       0);
1707
1708         return 0;
1709 }
1710
1711 subsys_initcall(tc_action_init);