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                           struct netlink_ext_ack *extack)
 355 {
 356         struct nlattr *nest;
 357         int n_i = 0;
 358         int ret = -EINVAL;
 359         struct idr *idr = &idrinfo->action_idr;
 360         struct tc_action *p;
 361         unsigned long id = 1;
 362         unsigned long tmp;
 363
 364         nest = nla_nest_start_noflag(skb, 0);
 365         if (nest == NULL)
 366                 goto nla_put_failure;
 367         if (nla_put_string(skb, TCA_KIND, ops->kind))
 368                 goto nla_put_failure;
 369
 370         ret = 0;
 371         mutex_lock(&idrinfo->lock);
 372         idr_for_each_entry_ul(idr, p, tmp, id) {
 373                 if (IS_ERR(p))
 374                         continue;
 375                 ret = tcf_idr_release_unsafe(p);
 376                 if (ret == ACT_P_DELETED)
 377                         module_put(ops->owner);
 378                 else if (ret < 0)
 379                         break;
 380                 n_i++;
 381         }
 382         mutex_unlock(&idrinfo->lock);
 383         if (ret < 0) {
 384                 if (n_i)
 385                         NL_SET_ERR_MSG(extack, "Unable to flush all TC actions");
 386                 else
 387                         goto nla_put_failure;
 388         }
 389
 390         ret = nla_put_u32(skb, TCA_FCNT, n_i);
 391         if (ret)
 392                 goto nla_put_failure;
 393         nla_nest_end(skb, nest);
 394
 395         return n_i;
 396 nla_put_failure:
 397         nla_nest_cancel(skb, nest);
 398         return ret;
 399 }
 400
 401 int tcf_generic_walker(struct tc_action_net *tn, struct sk_buff *skb,
 402                        struct netlink_callback *cb, int type,
 403                        const struct tc_action_ops *ops,
 404                        struct netlink_ext_ack *extack)
 405 {
 406         struct tcf_idrinfo *idrinfo = tn->idrinfo;
 407
 408         if (type == RTM_DELACTION) {
 409                 return tcf_del_walker(idrinfo, skb, ops, extack);
 410         } else if (type == RTM_GETACTION) {
 411                 return tcf_dump_walker(idrinfo, skb, cb);
 412         } else {
 413                 WARN(1, "tcf_generic_walker: unknown command %d\n", type);
 414                 NL_SET_ERR_MSG(extack, "tcf_generic_walker: unknown command");
 415                 return -EINVAL;
 416         }
 417 }
 418 EXPORT_SYMBOL(tcf_generic_walker);
 419
 420 int tcf_idr_search(struct tc_action_net *tn, struct tc_action **a, u32 index)
 421 {
 422         struct tcf_idrinfo *idrinfo = tn->idrinfo;
 423         struct tc_action *p;
 424
 425         mutex_lock(&idrinfo->lock);
 426         p = idr_find(&idrinfo->action_idr, index);
 427         if (IS_ERR(p))
 428                 p = NULL;
 429         else if (p)
 430                 refcount_inc(&p->tcfa_refcnt);
 431         mutex_unlock(&idrinfo->lock);
 432
 433         if (p) {
 434                 *a = p;
 435                 return true;
 436         }
 437         return false;
 438 }
 439 EXPORT_SYMBOL(tcf_idr_search);
 440
 441 static int tcf_idr_delete_index(struct tcf_idrinfo *idrinfo, u32 index)
 442 {
 443         struct tc_action *p;
 444         int ret = 0;
 445
 446         mutex_lock(&idrinfo->lock);
 447         p = idr_find(&idrinfo->action_idr, index);
 448         if (!p) {
 449                 mutex_unlock(&idrinfo->lock);
 450                 return -ENOENT;
 451         }
 452
 453         if (!atomic_read(&p->tcfa_bindcnt)) {
 454                 if (refcount_dec_and_test(&p->tcfa_refcnt)) {
 455                         struct module *owner = p->ops->owner;
 456
 457                         WARN_ON(p != idr_remove(&idrinfo->action_idr,
 458                                                 p->tcfa_index));
 459                         mutex_unlock(&idrinfo->lock);
 460
 461                         tcf_action_cleanup(p);
 462                         module_put(owner);
 463                         return 0;
 464                 }
 465                 ret = 0;
 466         } else {
 467                 ret = -EPERM;
 468         }
 469
 470         mutex_unlock(&idrinfo->lock);
 471         return ret;
 472 }
 473
 474 int tcf_idr_create(struct tc_action_net *tn, u32 index, struct nlattr *est,
 475                    struct tc_action **a, const struct tc_action_ops *ops,
 476                    int bind, bool cpustats, u32 flags)
 477 {
 478         struct tc_action *p = kzalloc(ops->size, GFP_KERNEL);
 479         struct tcf_idrinfo *idrinfo = tn->idrinfo;
 480         int err = -ENOMEM;
 481
 482         if (unlikely(!p))
 483                 return -ENOMEM;
 484         refcount_set(&p->tcfa_refcnt, 1);
 485         if (bind)
 486                 atomic_set(&p->tcfa_bindcnt, 1);
 487
 488         if (cpustats) {
 489                 p->cpu_bstats = netdev_alloc_pcpu_stats(struct gnet_stats_basic_cpu);
 490                 if (!p->cpu_bstats)
 491                         goto err1;
 492                 p->cpu_bstats_hw = netdev_alloc_pcpu_stats(struct gnet_stats_basic_cpu);
 493                 if (!p->cpu_bstats_hw)
 494                         goto err2;
 495                 p->cpu_qstats = alloc_percpu(struct gnet_stats_queue);
 496                 if (!p->cpu_qstats)
 497                         goto err3;
 498         }
 499         spin_lock_init(&p->tcfa_lock);
 500         p->tcfa_index = index;
 501         p->tcfa_tm.install = jiffies;
 502         p->tcfa_tm.lastuse = jiffies;
 503         p->tcfa_tm.firstuse = 0;
 504         p->tcfa_flags = flags & TCA_ACT_FLAGS_USER_MASK;
 505         if (est) {
 506                 err = gen_new_estimator(&p->tcfa_bstats, p->cpu_bstats,
 507                                         &p->tcfa_rate_est,
 508                                         &p->tcfa_lock, NULL, est);
 509                 if (err)
 510                         goto err4;
 511         }
 512
 513         p->idrinfo = idrinfo;
 514         __module_get(ops->owner);
 515         p->ops = ops;
 516         *a = p;
 517         return 0;
 518 err4:
 519         free_percpu(p->cpu_qstats);
 520 err3:
 521         free_percpu(p->cpu_bstats_hw);
 522 err2:
 523         free_percpu(p->cpu_bstats);
 524 err1:
 525         kfree(p);
 526         return err;
 527 }
 528 EXPORT_SYMBOL(tcf_idr_create);
 529
 530 int tcf_idr_create_from_flags(struct tc_action_net *tn, u32 index,
 531                               struct nlattr *est, struct tc_action **a,
 532                               const struct tc_action_ops *ops, int bind,
 533                               u32 flags)
 534 {
 535         /* Set cpustats according to actions flags. */
 536         return tcf_idr_create(tn, index, est, a, ops, bind,
 537                               !(flags & TCA_ACT_FLAGS_NO_PERCPU_STATS), flags);
 538 }
 539 EXPORT_SYMBOL(tcf_idr_create_from_flags);
 540
 541 /* Cleanup idr index that was allocated but not initialized. */
 542
 543 void tcf_idr_cleanup(struct tc_action_net *tn, u32 index)
 544 {
 545         struct tcf_idrinfo *idrinfo = tn->idrinfo;
 546
 547         mutex_lock(&idrinfo->lock);
 548         /* Remove ERR_PTR(-EBUSY) allocated by tcf_idr_check_alloc */
 549         WARN_ON(!IS_ERR(idr_remove(&idrinfo->action_idr, index)));
 550         mutex_unlock(&idrinfo->lock);
 551 }
 552 EXPORT_SYMBOL(tcf_idr_cleanup);
 553
 554 /* Check if action with specified index exists. If actions is found, increments
 555  * its reference and bind counters, and return 1. Otherwise insert temporary
 556  * error pointer (to prevent concurrent users from inserting actions with same
 557  * index) and return 0.
 558  */
 559
 560 int tcf_idr_check_alloc(struct tc_action_net *tn, u32 *index,
 561                         struct tc_action **a, int bind)
 562 {
 563         struct tcf_idrinfo *idrinfo = tn->idrinfo;
 564         struct tc_action *p;
 565         int ret;
 566
 567 again:
 568         mutex_lock(&idrinfo->lock);
 569         if (*index) {
 570                 p = idr_find(&idrinfo->action_idr, *index);
 571                 if (IS_ERR(p)) {
 572                         /* This means that another process allocated
 573                          * index but did not assign the pointer yet.
 574                          */
 575                         mutex_unlock(&idrinfo->lock);
 576                         goto again;
 577                 }
 578
 579                 if (p) {
 580                         refcount_inc(&p->tcfa_refcnt);
 581                         if (bind)
 582                                 atomic_inc(&p->tcfa_bindcnt);
 583                         *a = p;
 584                         ret = 1;
 585                 } else {
 586                         *a = NULL;
 587                         ret = idr_alloc_u32(&idrinfo->action_idr, NULL, index,
 588                                             *index, GFP_KERNEL);
 589                         if (!ret)
 590                                 idr_replace(&idrinfo->action_idr,
 591                                             ERR_PTR(-EBUSY), *index);
 592                 }
 593         } else {
 594                 *index = 1;
 595                 *a = NULL;
 596                 ret = idr_alloc_u32(&idrinfo->action_idr, NULL, index,
 597                                     UINT_MAX, GFP_KERNEL);
 598                 if (!ret)
 599                         idr_replace(&idrinfo->action_idr, ERR_PTR(-EBUSY),
 600                                     *index);
 601         }
 602         mutex_unlock(&idrinfo->lock);
 603         return ret;
 604 }
 605 EXPORT_SYMBOL(tcf_idr_check_alloc);
 606
 607 void tcf_idrinfo_destroy(const struct tc_action_ops *ops,
 608                          struct tcf_idrinfo *idrinfo)
 609 {
 610         struct idr *idr = &idrinfo->action_idr;
 611         struct tc_action *p;
 612         int ret;
 613         unsigned long id = 1;
 614         unsigned long tmp;
 615
 616         idr_for_each_entry_ul(idr, p, tmp, id) {
 617                 ret = __tcf_idr_release(p, false, true);
 618                 if (ret == ACT_P_DELETED)
 619                         module_put(ops->owner);
 620                 else if (ret < 0)
 621                         return;
 622         }
 623         idr_destroy(&idrinfo->action_idr);
 624 }
 625 EXPORT_SYMBOL(tcf_idrinfo_destroy);
 626
 627 static LIST_HEAD(act_base);
 628 static DEFINE_RWLOCK(act_mod_lock);
 629
 630 int tcf_register_action(struct tc_action_ops *act,
 631                         struct pernet_operations *ops)
 632 {
 633         struct tc_action_ops *a;
 634         int ret;
 635
 636         if (!act->act || !act->dump || !act->init || !act->walk || !act->lookup)
 637                 return -EINVAL;
 638
 639         /* We have to register pernet ops before making the action ops visible,
 640          * otherwise tcf_action_init_1() could get a partially initialized
 641          * netns.
 642          */
 643         ret = register_pernet_subsys(ops);
 644         if (ret)
 645                 return ret;
 646
 647         write_lock(&act_mod_lock);
 648         list_for_each_entry(a, &act_base, head) {
 649                 if (act->id == a->id || (strcmp(act->kind, a->kind) == 0)) {
 650                         write_unlock(&act_mod_lock);
 651                         unregister_pernet_subsys(ops);
 652                         return -EEXIST;
 653                 }
 654         }
 655         list_add_tail(&act->head, &act_base);
 656         write_unlock(&act_mod_lock);
 657
 658         return 0;
 659 }
 660 EXPORT_SYMBOL(tcf_register_action);
 661
 662 int tcf_unregister_action(struct tc_action_ops *act,
 663                           struct pernet_operations *ops)
 664 {
 665         struct tc_action_ops *a;
 666         int err = -ENOENT;
 667
 668         write_lock(&act_mod_lock);
 669         list_for_each_entry(a, &act_base, head) {
 670                 if (a == act) {
 671                         list_del(&act->head);
 672                         err = 0;
 673                         break;
 674                 }
 675         }
 676         write_unlock(&act_mod_lock);
 677         if (!err)
 678                 unregister_pernet_subsys(ops);
 679         return err;
 680 }
 681 EXPORT_SYMBOL(tcf_unregister_action);
 682
 683 /* lookup by name */
 684 static struct tc_action_ops *tc_lookup_action_n(char *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 (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 /* lookup by nlattr */
 703 static struct tc_action_ops *tc_lookup_action(struct nlattr *kind)
 704 {
 705         struct tc_action_ops *a, *res = NULL;
 706
 707         if (kind) {
 708                 read_lock(&act_mod_lock);
 709                 list_for_each_entry(a, &act_base, head) {
 710                         if (nla_strcmp(kind, a->kind) == 0) {
 711                                 if (try_module_get(a->owner))
 712                                         res = a;
 713                                 break;
 714                         }
 715                 }
 716                 read_unlock(&act_mod_lock);
 717         }
 718         return res;
 719 }
 720
 721 /*TCA_ACT_MAX_PRIO is 32, there count up to 32 */
 722 #define TCA_ACT_MAX_PRIO_MASK 0x1FF
 723 int tcf_action_exec(struct sk_buff *skb, struct tc_action **actions,
 724                     int nr_actions, struct tcf_result *res)
 725 {
 726         u32 jmp_prgcnt = 0;
 727         u32 jmp_ttl = TCA_ACT_MAX_PRIO; /*matches actions per filter */
 728         int i;
 729         int ret = TC_ACT_OK;
 730
 731         if (skb_skip_tc_classify(skb))
 732                 return TC_ACT_OK;
 733
 734 restart_act_graph:
 735         for (i = 0; i < nr_actions; i++) {
 736                 const struct tc_action *a = actions[i];
 737                 int repeat_ttl;
 738
 739                 if (jmp_prgcnt > 0) {
 740                         jmp_prgcnt -= 1;
 741                         continue;
 742                 }
 743
 744                 repeat_ttl = 32;
 745 repeat:
 746                 ret = a->ops->act(skb, a, res);
 747
 748                 if (unlikely(ret == TC_ACT_REPEAT)) {
 749                         if (--repeat_ttl != 0)
 750                                 goto repeat;
 751                         /* suspicious opcode, stop pipeline */
 752                         net_warn_ratelimited("TC_ACT_REPEAT abuse ?\n");
 753                         return TC_ACT_OK;
 754                 }
 755
 756                 if (TC_ACT_EXT_CMP(ret, TC_ACT_JUMP)) {
 757                         jmp_prgcnt = ret & TCA_ACT_MAX_PRIO_MASK;
 758                         if (!jmp_prgcnt || (jmp_prgcnt > nr_actions)) {
 759                                 /* faulty opcode, stop pipeline */
 760                                 return TC_ACT_OK;
 761                         } else {
 762                                 jmp_ttl -= 1;
 763                                 if (jmp_ttl > 0)
 764                                         goto restart_act_graph;
 765                                 else /* faulty graph, stop pipeline */
 766                                         return TC_ACT_OK;
 767                         }
 768                 } else if (TC_ACT_EXT_CMP(ret, TC_ACT_GOTO_CHAIN)) {
 769                         if (unlikely(!rcu_access_pointer(a->goto_chain))) {
 770                                 net_warn_ratelimited("can't go to NULL chain!\n");
 771                                 return TC_ACT_SHOT;
 772                         }
 773                         tcf_action_goto_chain_exec(a, res);
 774                 }
 775
 776                 if (ret != TC_ACT_PIPE)
 777                         break;
 778         }
 779
 780         return ret;
 781 }
 782 EXPORT_SYMBOL(tcf_action_exec);
 783
 784 int tcf_action_destroy(struct tc_action *actions[], int bind)
 785 {
 786         const struct tc_action_ops *ops;
 787         struct tc_action *a;
 788         int ret = 0, i;
 789
 790         for (i = 0; i < TCA_ACT_MAX_PRIO && actions[i]; i++) {
 791                 a = actions[i];
 792                 actions[i] = NULL;
 793                 ops = a->ops;
 794                 ret = __tcf_idr_release(a, bind, true);
 795                 if (ret == ACT_P_DELETED)
 796                         module_put(ops->owner);
 797                 else if (ret < 0)
 798                         return ret;
 799         }
 800         return ret;
 801 }
 802
 803 static int tcf_action_put(struct tc_action *p)
 804 {
 805         return __tcf_action_put(p, false);
 806 }
 807
 808 /* Put all actions in this array, skip those NULL's. */
 809 static void tcf_action_put_many(struct tc_action *actions[])
 810 {
 811         int i;
 812
 813         for (i = 0; i < TCA_ACT_MAX_PRIO; i++) {
 814                 struct tc_action *a = actions[i];
 815                 const struct tc_action_ops *ops;
 816
 817                 if (!a)
 818                         continue;
 819                 ops = a->ops;
 820                 if (tcf_action_put(a))
 821                         module_put(ops->owner);
 822         }
 823 }
 824
 825 int
 826 tcf_action_dump_old(struct sk_buff *skb, struct tc_action *a, int bind, int ref)
 827 {
 828         return a->ops->dump(skb, a, bind, ref);
 829 }
 830
 831 int
 832 tcf_action_dump_1(struct sk_buff *skb, struct tc_action *a, int bind, int ref)
 833 {
 834         int err = -EINVAL;
 835         unsigned char *b = skb_tail_pointer(skb);
 836         struct nlattr *nest;
 837
 838         if (tcf_action_dump_terse(skb, a, false))
 839                 goto nla_put_failure;
 840
 841         if (a->hw_stats != TCA_ACT_HW_STATS_ANY &&
 842             nla_put_bitfield32(skb, TCA_ACT_HW_STATS,
 843                                a->hw_stats, TCA_ACT_HW_STATS_ANY))
 844                 goto nla_put_failure;
 845
 846         if (a->used_hw_stats_valid &&
 847             nla_put_bitfield32(skb, TCA_ACT_USED_HW_STATS,
 848                                a->used_hw_stats, TCA_ACT_HW_STATS_ANY))
 849                 goto nla_put_failure;
 850
 851         if (a->tcfa_flags &&
 852             nla_put_bitfield32(skb, TCA_ACT_FLAGS,
 853                                a->tcfa_flags, a->tcfa_flags))
 854                 goto nla_put_failure;
 855
 856         nest = nla_nest_start_noflag(skb, TCA_OPTIONS);
 857         if (nest == NULL)
 858                 goto nla_put_failure;
 859         err = tcf_action_dump_old(skb, a, bind, ref);
 860         if (err > 0) {
 861                 nla_nest_end(skb, nest);
 862                 return err;
 863         }
 864
 865 nla_put_failure:
 866         nlmsg_trim(skb, b);
 867         return -1;
 868 }
 869 EXPORT_SYMBOL(tcf_action_dump_1);
 870
 871 int tcf_action_dump(struct sk_buff *skb, struct tc_action *actions[],
 872                     int bind, int ref, bool terse)
 873 {
 874         struct tc_action *a;
 875         int err = -EINVAL, i;
 876         struct nlattr *nest;
 877
 878         for (i = 0; i < TCA_ACT_MAX_PRIO && actions[i]; i++) {
 879                 a = actions[i];
 880                 nest = nla_nest_start_noflag(skb, i + 1);
 881                 if (nest == NULL)
 882                         goto nla_put_failure;
 883                 err = terse ? tcf_action_dump_terse(skb, a, false) :
 884                         tcf_action_dump_1(skb, a, bind, ref);
 885                 if (err < 0)
 886                         goto errout;
 887                 nla_nest_end(skb, nest);
 888         }
 889
 890         return 0;
 891
 892 nla_put_failure:
 893         err = -EINVAL;
 894 errout:
 895         nla_nest_cancel(skb, nest);
 896         return err;
 897 }
 898
 899 static struct tc_cookie *nla_memdup_cookie(struct nlattr **tb)
 900 {
 901         struct tc_cookie *c = kzalloc(sizeof(*c), GFP_KERNEL);
 902         if (!c)
 903                 return NULL;
 904
 905         c->data = nla_memdup(tb[TCA_ACT_COOKIE], GFP_KERNEL);
 906         if (!c->data) {
 907                 kfree(c);
 908                 return NULL;
 909         }
 910         c->len = nla_len(tb[TCA_ACT_COOKIE]);
 911
 912         return c;
 913 }
 914
 915 static u8 tcf_action_hw_stats_get(struct nlattr *hw_stats_attr)
 916 {
 917         struct nla_bitfield32 hw_stats_bf;
 918
 919         /* If the user did not pass the attr, that means he does
 920          * not care about the type. Return "any" in that case
 921          * which is setting on all supported types.
 922          */
 923         if (!hw_stats_attr)
 924                 return TCA_ACT_HW_STATS_ANY;
 925         hw_stats_bf = nla_get_bitfield32(hw_stats_attr);
 926         return hw_stats_bf.value;
 927 }
 928
 929 static const struct nla_policy tcf_action_policy[TCA_ACT_MAX + 1] = {
 930         [TCA_ACT_KIND]          = { .type = NLA_STRING },
 931         [TCA_ACT_INDEX]         = { .type = NLA_U32 },
 932         [TCA_ACT_COOKIE]        = { .type = NLA_BINARY,
 933                                     .len = TC_COOKIE_MAX_SIZE },
 934         [TCA_ACT_OPTIONS]       = { .type = NLA_NESTED },
 935         [TCA_ACT_FLAGS]         = NLA_POLICY_BITFIELD32(TCA_ACT_FLAGS_NO_PERCPU_STATS),
 936         [TCA_ACT_HW_STATS]      = NLA_POLICY_BITFIELD32(TCA_ACT_HW_STATS_ANY),
 937 };
 938
 939 void tcf_idr_insert_many(struct tc_action *actions[])
 940 {
 941         int i;
 942
 943         for (i = 0; i < TCA_ACT_MAX_PRIO; i++) {
 944                 struct tc_action *a = actions[i];
 945                 struct tcf_idrinfo *idrinfo;
 946
 947                 if (!a)
 948                         continue;
 949                 idrinfo = a->idrinfo;
 950                 mutex_lock(&idrinfo->lock);
 951                 /* Replace ERR_PTR(-EBUSY) allocated by tcf_idr_check_alloc if
 952                  * it is just created, otherwise this is just a nop.
 953                  */
 954                 idr_replace(&idrinfo->action_idr, a, a->tcfa_index);
 955                 mutex_unlock(&idrinfo->lock);
 956         }
 957 }
 958
 959 struct tc_action_ops *tc_action_load_ops(struct nlattr *nla, bool police,
 960                                          bool rtnl_held,
 961                                          struct netlink_ext_ack *extack)
 962 {
 963         struct nlattr *tb[TCA_ACT_MAX + 1];
 964         struct tc_action_ops *a_o;
 965         char act_name[IFNAMSIZ];
 966         struct nlattr *kind;
 967         int err;
 968
 969         if (!police) {
 970                 err = nla_parse_nested_deprecated(tb, TCA_ACT_MAX, nla,
 971                                                   tcf_action_policy, extack);
 972                 if (err < 0)
 973                         return ERR_PTR(err);
 974                 err = -EINVAL;
 975                 kind = tb[TCA_ACT_KIND];
 976                 if (!kind) {
 977                         NL_SET_ERR_MSG(extack, "TC action kind must be specified");
 978                         return ERR_PTR(err);
 979                 }
 980                 if (nla_strscpy(act_name, kind, IFNAMSIZ) < 0) {
 981                         NL_SET_ERR_MSG(extack, "TC action name too long");
 982                         return ERR_PTR(err);
 983                 }
 984         } else {
 985                 if (strlcpy(act_name, "police", IFNAMSIZ) >= IFNAMSIZ) {
 986                         NL_SET_ERR_MSG(extack, "TC action name too long");
 987                         return ERR_PTR(-EINVAL);
 988                 }
 989         }
 990
 991         a_o = tc_lookup_action_n(act_name);
 992         if (a_o == NULL) {
 993 #ifdef CONFIG_MODULES
 994                 if (rtnl_held)
 995                         rtnl_unlock();
 996                 request_module("act_%s", act_name);
 997                 if (rtnl_held)
 998                         rtnl_lock();
 999
1000                 a_o = tc_lookup_action_n(act_name);
1001
1002                 /* We dropped the RTNL semaphore in order to
1003                  * perform the module load.  So, even if we
1004                  * succeeded in loading the module we have to
1005                  * tell the caller to replay the request.  We
1006                  * indicate this using -EAGAIN.
1007                  */
1008                 if (a_o != NULL) {
1009                         module_put(a_o->owner);
1010                         return ERR_PTR(-EAGAIN);
1011                 }
1012 #endif
1013                 NL_SET_ERR_MSG(extack, "Failed to load TC action module");
1014                 return ERR_PTR(-ENOENT);
1015         }
1016
1017         return a_o;
1018 }
1019
1020 struct tc_action *tcf_action_init_1(struct net *net, struct tcf_proto *tp,
1021                                     struct nlattr *nla, struct nlattr *est,
1022                                     struct tc_action_ops *a_o, int *init_res,
1023                                     u32 flags, struct netlink_ext_ack *extack)
1024 {
1025         bool police = flags & TCA_ACT_FLAGS_POLICE;
1026         struct nla_bitfield32 userflags = { 0, 0 };
1027         u8 hw_stats = TCA_ACT_HW_STATS_ANY;
1028         struct nlattr *tb[TCA_ACT_MAX + 1];
1029         struct tc_cookie *cookie = NULL;
1030         struct tc_action *a;
1031         int err;
1032
1033         /* backward compatibility for policer */
1034         if (!police) {
1035                 err = nla_parse_nested_deprecated(tb, TCA_ACT_MAX, nla,
1036                                                   tcf_action_policy, extack);
1037                 if (err < 0)
1038                         return ERR_PTR(err);
1039                 if (tb[TCA_ACT_COOKIE]) {
1040                         cookie = nla_memdup_cookie(tb);
1041                         if (!cookie) {
1042                                 NL_SET_ERR_MSG(extack, "No memory to generate TC cookie");
1043                                 err = -ENOMEM;
1044                                 goto err_out;
1045                         }
1046                 }
1047                 hw_stats = tcf_action_hw_stats_get(tb[TCA_ACT_HW_STATS]);
1048                 if (tb[TCA_ACT_FLAGS])
1049                         userflags = nla_get_bitfield32(tb[TCA_ACT_FLAGS]);
1050
1051                 err = a_o->init(net, tb[TCA_ACT_OPTIONS], est, &a, tp,
1052                                 userflags.value | flags, extack);
1053         } else {
1054                 err = a_o->init(net, nla, est, &a, tp, userflags.value | flags,
1055                                 extack);
1056         }
1057         if (err < 0)
1058                 goto err_out;
1059         *init_res = err;
1060
1061         if (!police && tb[TCA_ACT_COOKIE])
1062                 tcf_set_action_cookie(&a->act_cookie, cookie);
1063
1064         if (!police)
1065                 a->hw_stats = hw_stats;
1066
1067         return a;
1068
1069 err_out:
1070         if (cookie) {
1071                 kfree(cookie->data);
1072                 kfree(cookie);
1073         }
1074         return ERR_PTR(err);
1075 }
1076
1077 /* Returns numbers of initialized actions or negative error. */
1078
1079 int tcf_action_init(struct net *net, struct tcf_proto *tp, struct nlattr *nla,
1080                     struct nlattr *est, struct tc_action *actions[],
1081                     int init_res[], size_t *attr_size, u32 flags,
1082                     struct netlink_ext_ack *extack)
1083 {
1084         struct tc_action_ops *ops[TCA_ACT_MAX_PRIO] = {};
1085         struct nlattr *tb[TCA_ACT_MAX_PRIO + 1];
1086         struct tc_action *act;
1087         size_t sz = 0;
1088         int err;
1089         int i;
1090
1091         err = nla_parse_nested_deprecated(tb, TCA_ACT_MAX_PRIO, nla, NULL,
1092                                           extack);
1093         if (err < 0)
1094                 return err;
1095
1096         for (i = 1; i <= TCA_ACT_MAX_PRIO && tb[i]; i++) {
1097                 struct tc_action_ops *a_o;
1098
1099                 a_o = tc_action_load_ops(tb[i], flags & TCA_ACT_FLAGS_POLICE,
1100                                          !(flags & TCA_ACT_FLAGS_NO_RTNL),
1101                                          extack);
1102                 if (IS_ERR(a_o)) {
1103                         err = PTR_ERR(a_o);
1104                         goto err_mod;
1105                 }
1106                 ops[i - 1] = a_o;
1107         }
1108
1109         for (i = 1; i <= TCA_ACT_MAX_PRIO && tb[i]; i++) {
1110                 act = tcf_action_init_1(net, tp, tb[i], est, ops[i - 1],
1111                                         &init_res[i - 1], flags, extack);
1112                 if (IS_ERR(act)) {
1113                         err = PTR_ERR(act);
1114                         goto err;
1115                 }
1116                 sz += tcf_action_fill_size(act);
1117                 /* Start from index 0 */
1118                 actions[i - 1] = act;
1119         }
1120
1121         /* We have to commit them all together, because if any error happened in
1122          * between, we could not handle the failure gracefully.
1123          */
1124         tcf_idr_insert_many(actions);
1125
1126         *attr_size = tcf_action_full_attrs_size(sz);
1127         err = i - 1;
1128         goto err_mod;
1129
1130 err:
1131         tcf_action_destroy(actions, flags & TCA_ACT_FLAGS_BIND);
1132 err_mod:
1133         for (i = 0; i < TCA_ACT_MAX_PRIO; i++) {
1134                 if (ops[i])
1135                         module_put(ops[i]->owner);
1136         }
1137         return err;
1138 }
1139
1140 void tcf_action_update_stats(struct tc_action *a, u64 bytes, u64 packets,
1141                              u64 drops, bool hw)
1142 {
1143         if (a->cpu_bstats) {
1144                 _bstats_cpu_update(this_cpu_ptr(a->cpu_bstats), bytes, packets);
1145
1146                 this_cpu_ptr(a->cpu_qstats)->drops += drops;
1147
1148                 if (hw)
1149                         _bstats_cpu_update(this_cpu_ptr(a->cpu_bstats_hw),
1150                                            bytes, packets);
1151                 return;
1152         }
1153
1154         _bstats_update(&a->tcfa_bstats, bytes, packets);
1155         a->tcfa_qstats.drops += drops;
1156         if (hw)
1157                 _bstats_update(&a->tcfa_bstats_hw, bytes, packets);
1158 }
1159 EXPORT_SYMBOL(tcf_action_update_stats);
1160
1161 int tcf_action_copy_stats(struct sk_buff *skb, struct tc_action *p,
1162                           int compat_mode)
1163 {
1164         int err = 0;
1165         struct gnet_dump d;
1166
1167         if (p == NULL)
1168                 goto errout;
1169
1170         /* compat_mode being true specifies a call that is supposed
1171          * to add additional backward compatibility statistic TLVs.
1172          */
1173         if (compat_mode) {
1174                 if (p->type == TCA_OLD_COMPAT)
1175                         err = gnet_stats_start_copy_compat(skb, 0,
1176                                                            TCA_STATS,
1177                                                            TCA_XSTATS,
1178                                                            &p->tcfa_lock, &d,
1179                                                            TCA_PAD);
1180                 else
1181                         return 0;
1182         } else
1183                 err = gnet_stats_start_copy(skb, TCA_ACT_STATS,
1184                                             &p->tcfa_lock, &d, TCA_ACT_PAD);
1185
1186         if (err < 0)
1187                 goto errout;
1188
1189         if (gnet_stats_copy_basic(NULL, &d, p->cpu_bstats, &p->tcfa_bstats) < 0 ||
1190             gnet_stats_copy_basic_hw(NULL, &d, p->cpu_bstats_hw,
1191                                      &p->tcfa_bstats_hw) < 0 ||
1192             gnet_stats_copy_rate_est(&d, &p->tcfa_rate_est) < 0 ||
1193             gnet_stats_copy_queue(&d, p->cpu_qstats,
1194                                   &p->tcfa_qstats,
1195                                   p->tcfa_qstats.qlen) < 0)
1196                 goto errout;
1197
1198         if (gnet_stats_finish_copy(&d) < 0)
1199                 goto errout;
1200
1201         return 0;
1202
1203 errout:
1204         return -1;
1205 }
1206
1207 static int tca_get_fill(struct sk_buff *skb, struct tc_action *actions[],
1208                         u32 portid, u32 seq, u16 flags, int event, int bind,
1209                         int ref)
1210 {
1211         struct tcamsg *t;
1212         struct nlmsghdr *nlh;
1213         unsigned char *b = skb_tail_pointer(skb);
1214         struct nlattr *nest;
1215
1216         nlh = nlmsg_put(skb, portid, seq, event, sizeof(*t), flags);
1217         if (!nlh)
1218                 goto out_nlmsg_trim;
1219         t = nlmsg_data(nlh);
1220         t->tca_family = AF_UNSPEC;
1221         t->tca__pad1 = 0;
1222         t->tca__pad2 = 0;
1223
1224         nest = nla_nest_start_noflag(skb, TCA_ACT_TAB);
1225         if (!nest)
1226                 goto out_nlmsg_trim;
1227
1228         if (tcf_action_dump(skb, actions, bind, ref, false) < 0)
1229                 goto out_nlmsg_trim;
1230
1231         nla_nest_end(skb, nest);
1232
1233         nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1234         return skb->len;
1235
1236 out_nlmsg_trim:
1237         nlmsg_trim(skb, b);
1238         return -1;
1239 }
1240
1241 static int
1242 tcf_get_notify(struct net *net, u32 portid, struct nlmsghdr *n,
1243                struct tc_action *actions[], int event,
1244                struct netlink_ext_ack *extack)
1245 {
1246         struct sk_buff *skb;
1247
1248         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1249         if (!skb)
1250                 return -ENOBUFS;
1251         if (tca_get_fill(skb, actions, portid, n->nlmsg_seq, 0, event,
1252                          0, 1) <= 0) {
1253                 NL_SET_ERR_MSG(extack, "Failed to fill netlink attributes while adding TC action");
1254                 kfree_skb(skb);
1255                 return -EINVAL;
1256         }
1257
1258         return rtnl_unicast(skb, net, portid);
1259 }
1260
1261 static struct tc_action *tcf_action_get_1(struct net *net, struct nlattr *nla,
1262                                           struct nlmsghdr *n, u32 portid,
1263                                           struct netlink_ext_ack *extack)
1264 {
1265         struct nlattr *tb[TCA_ACT_MAX + 1];
1266         const struct tc_action_ops *ops;
1267         struct tc_action *a;
1268         int index;
1269         int err;
1270
1271         err = nla_parse_nested_deprecated(tb, TCA_ACT_MAX, nla,
1272                                           tcf_action_policy, extack);
1273         if (err < 0)
1274                 goto err_out;
1275
1276         err = -EINVAL;
1277         if (tb[TCA_ACT_INDEX] == NULL ||
1278             nla_len(tb[TCA_ACT_INDEX]) < sizeof(index)) {
1279                 NL_SET_ERR_MSG(extack, "Invalid TC action index value");
1280                 goto err_out;
1281         }
1282         index = nla_get_u32(tb[TCA_ACT_INDEX]);
1283
1284         err = -EINVAL;
1285         ops = tc_lookup_action(tb[TCA_ACT_KIND]);
1286         if (!ops) { /* could happen in batch of actions */
1287                 NL_SET_ERR_MSG(extack, "Specified TC action kind not found");
1288                 goto err_out;
1289         }
1290         err = -ENOENT;
1291         if (ops->lookup(net, &a, index) == 0) {
1292                 NL_SET_ERR_MSG(extack, "TC action with specified index not found");
1293                 goto err_mod;
1294         }
1295
1296         module_put(ops->owner);
1297         return a;
1298
1299 err_mod:
1300         module_put(ops->owner);
1301 err_out:
1302         return ERR_PTR(err);
1303 }
1304
1305 static int tca_action_flush(struct net *net, struct nlattr *nla,
1306                             struct nlmsghdr *n, u32 portid,
1307                             struct netlink_ext_ack *extack)
1308 {
1309         struct sk_buff *skb;
1310         unsigned char *b;
1311         struct nlmsghdr *nlh;
1312         struct tcamsg *t;
1313         struct netlink_callback dcb;
1314         struct nlattr *nest;
1315         struct nlattr *tb[TCA_ACT_MAX + 1];
1316         const struct tc_action_ops *ops;
1317         struct nlattr *kind;
1318         int err = -ENOMEM;
1319
1320         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1321         if (!skb)
1322                 return err;
1323
1324         b = skb_tail_pointer(skb);
1325
1326         err = nla_parse_nested_deprecated(tb, TCA_ACT_MAX, nla,
1327                                           tcf_action_policy, extack);
1328         if (err < 0)
1329                 goto err_out;
1330
1331         err = -EINVAL;
1332         kind = tb[TCA_ACT_KIND];
1333         ops = tc_lookup_action(kind);
1334         if (!ops) { /*some idjot trying to flush unknown action */
1335                 NL_SET_ERR_MSG(extack, "Cannot flush unknown TC action");
1336                 goto err_out;
1337         }
1338
1339         nlh = nlmsg_put(skb, portid, n->nlmsg_seq, RTM_DELACTION,
1340                         sizeof(*t), 0);
1341         if (!nlh) {
1342                 NL_SET_ERR_MSG(extack, "Failed to create TC action flush notification");
1343                 goto out_module_put;
1344         }
1345         t = nlmsg_data(nlh);
1346         t->tca_family = AF_UNSPEC;
1347         t->tca__pad1 = 0;
1348         t->tca__pad2 = 0;
1349
1350         nest = nla_nest_start_noflag(skb, TCA_ACT_TAB);
1351         if (!nest) {
1352                 NL_SET_ERR_MSG(extack, "Failed to add new netlink message");
1353                 goto out_module_put;
1354         }
1355
1356         err = ops->walk(net, skb, &dcb, RTM_DELACTION, ops, extack);
1357         if (err <= 0) {
1358                 nla_nest_cancel(skb, nest);
1359                 goto out_module_put;
1360         }
1361
1362         nla_nest_end(skb, nest);
1363
1364         nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1365         nlh->nlmsg_flags |= NLM_F_ROOT;
1366         module_put(ops->owner);
1367         err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1368                              n->nlmsg_flags & NLM_F_ECHO);
1369         if (err < 0)
1370                 NL_SET_ERR_MSG(extack, "Failed to send TC action flush notification");
1371
1372         return err;
1373
1374 out_module_put:
1375         module_put(ops->owner);
1376 err_out:
1377         kfree_skb(skb);
1378         return err;
1379 }
1380
1381 static int tcf_action_delete(struct net *net, struct tc_action *actions[])
1382 {
1383         int i;
1384
1385         for (i = 0; i < TCA_ACT_MAX_PRIO && actions[i]; i++) {
1386                 struct tc_action *a = actions[i];
1387                 const struct tc_action_ops *ops = a->ops;
1388                 /* Actions can be deleted concurrently so we must save their
1389                  * type and id to search again after reference is released.
1390                  */
1391                 struct tcf_idrinfo *idrinfo = a->idrinfo;
1392                 u32 act_index = a->tcfa_index;
1393
1394                 actions[i] = NULL;
1395                 if (tcf_action_put(a)) {
1396                         /* last reference, action was deleted concurrently */
1397                         module_put(ops->owner);
1398                 } else  {
1399                         int ret;
1400
1401                         /* now do the delete */
1402                         ret = tcf_idr_delete_index(idrinfo, act_index);
1403                         if (ret < 0)
1404                                 return ret;
1405                 }
1406         }
1407         return 0;
1408 }
1409
1410 static int
1411 tcf_del_notify(struct net *net, struct nlmsghdr *n, struct tc_action *actions[],
1412                u32 portid, size_t attr_size, struct netlink_ext_ack *extack)
1413 {
1414         int ret;
1415         struct sk_buff *skb;
1416
1417         skb = alloc_skb(attr_size <= NLMSG_GOODSIZE ? NLMSG_GOODSIZE : attr_size,
1418                         GFP_KERNEL);
1419         if (!skb)
1420                 return -ENOBUFS;
1421
1422         if (tca_get_fill(skb, actions, portid, n->nlmsg_seq, 0, RTM_DELACTION,
1423                          0, 2) <= 0) {
1424                 NL_SET_ERR_MSG(extack, "Failed to fill netlink TC action attributes");
1425                 kfree_skb(skb);
1426                 return -EINVAL;
1427         }
1428
1429         /* now do the delete */
1430         ret = tcf_action_delete(net, actions);
1431         if (ret < 0) {
1432                 NL_SET_ERR_MSG(extack, "Failed to delete TC action");
1433                 kfree_skb(skb);
1434                 return ret;
1435         }
1436
1437         ret = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1438                              n->nlmsg_flags & NLM_F_ECHO);
1439         return ret;
1440 }
1441
1442 static int
1443 tca_action_gd(struct net *net, struct nlattr *nla, struct nlmsghdr *n,
1444               u32 portid, int event, struct netlink_ext_ack *extack)
1445 {
1446         int i, ret;
1447         struct nlattr *tb[TCA_ACT_MAX_PRIO + 1];
1448         struct tc_action *act;
1449         size_t attr_size = 0;
1450         struct tc_action *actions[TCA_ACT_MAX_PRIO] = {};
1451
1452         ret = nla_parse_nested_deprecated(tb, TCA_ACT_MAX_PRIO, nla, NULL,
1453                                           extack);
1454         if (ret < 0)
1455                 return ret;
1456
1457         if (event == RTM_DELACTION && n->nlmsg_flags & NLM_F_ROOT) {
1458                 if (tb[1])
1459                         return tca_action_flush(net, tb[1], n, portid, extack);
1460
1461                 NL_SET_ERR_MSG(extack, "Invalid netlink attributes while flushing TC action");
1462                 return -EINVAL;
1463         }
1464
1465         for (i = 1; i <= TCA_ACT_MAX_PRIO && tb[i]; i++) {
1466                 act = tcf_action_get_1(net, tb[i], n, portid, extack);
1467                 if (IS_ERR(act)) {
1468                         ret = PTR_ERR(act);
1469                         goto err;
1470                 }
1471                 attr_size += tcf_action_fill_size(act);
1472                 actions[i - 1] = act;
1473         }
1474
1475         attr_size = tcf_action_full_attrs_size(attr_size);
1476
1477         if (event == RTM_GETACTION)
1478                 ret = tcf_get_notify(net, portid, n, actions, event, extack);
1479         else { /* delete */
1480                 ret = tcf_del_notify(net, n, actions, portid, attr_size, extack);
1481                 if (ret)
1482                         goto err;
1483                 return 0;
1484         }
1485 err:
1486         tcf_action_put_many(actions);
1487         return ret;
1488 }
1489
1490 static int
1491 tcf_add_notify(struct net *net, struct nlmsghdr *n, struct tc_action *actions[],
1492                u32 portid, size_t attr_size, struct netlink_ext_ack *extack)
1493 {
1494         struct sk_buff *skb;
1495
1496         skb = alloc_skb(attr_size <= NLMSG_GOODSIZE ? NLMSG_GOODSIZE : attr_size,
1497                         GFP_KERNEL);
1498         if (!skb)
1499                 return -ENOBUFS;
1500
1501         if (tca_get_fill(skb, actions, portid, n->nlmsg_seq, n->nlmsg_flags,
1502                          RTM_NEWACTION, 0, 0) <= 0) {
1503                 NL_SET_ERR_MSG(extack, "Failed to fill netlink attributes while adding TC action");
1504                 kfree_skb(skb);
1505                 return -EINVAL;
1506         }
1507
1508         return rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1509                               n->nlmsg_flags & NLM_F_ECHO);
1510 }
1511
1512 static int tcf_action_add(struct net *net, struct nlattr *nla,
1513                           struct nlmsghdr *n, u32 portid, u32 flags,
1514                           struct netlink_ext_ack *extack)
1515 {
1516         size_t attr_size = 0;
1517         int loop, ret, i;
1518         struct tc_action *actions[TCA_ACT_MAX_PRIO] = {};
1519         int init_res[TCA_ACT_MAX_PRIO] = {};
1520
1521         for (loop = 0; loop < 10; loop++) {
1522                 ret = tcf_action_init(net, NULL, nla, NULL, actions, init_res,
1523                                       &attr_size, flags, extack);
1524                 if (ret != -EAGAIN)
1525                         break;
1526         }
1527
1528         if (ret < 0)
1529                 return ret;
1530         ret = tcf_add_notify(net, n, actions, portid, attr_size, extack);
1531
1532         /* only put existing actions */
1533         for (i = 0; i < TCA_ACT_MAX_PRIO; i++)
1534                 if (init_res[i] == ACT_P_CREATED)
1535                         actions[i] = NULL;
1536         tcf_action_put_many(actions);
1537
1538         return ret;
1539 }
1540
1541 static const struct nla_policy tcaa_policy[TCA_ROOT_MAX + 1] = {
1542         [TCA_ROOT_FLAGS] = NLA_POLICY_BITFIELD32(TCA_ACT_FLAG_LARGE_DUMP_ON |
1543                                                  TCA_ACT_FLAG_TERSE_DUMP),
1544         [TCA_ROOT_TIME_DELTA]      = { .type = NLA_U32 },
1545 };
1546
1547 static int tc_ctl_action(struct sk_buff *skb, struct nlmsghdr *n,
1548                          struct netlink_ext_ack *extack)
1549 {
1550         struct net *net = sock_net(skb->sk);
1551         struct nlattr *tca[TCA_ROOT_MAX + 1];
1552         u32 portid = NETLINK_CB(skb).portid;
1553         u32 flags = 0;
1554         int ret = 0;
1555
1556         if ((n->nlmsg_type != RTM_GETACTION) &&
1557             !netlink_capable(skb, CAP_NET_ADMIN))
1558                 return -EPERM;
1559
1560         ret = nlmsg_parse_deprecated(n, sizeof(struct tcamsg), tca,
1561                                      TCA_ROOT_MAX, NULL, extack);
1562         if (ret < 0)
1563                 return ret;
1564
1565         if (tca[TCA_ACT_TAB] == NULL) {
1566                 NL_SET_ERR_MSG(extack, "Netlink action attributes missing");
1567                 return -EINVAL;
1568         }
1569
1570         /* n->nlmsg_flags & NLM_F_CREATE */
1571         switch (n->nlmsg_type) {
1572         case RTM_NEWACTION:
1573                 /* we are going to assume all other flags
1574                  * imply create only if it doesn't exist
1575                  * Note that CREATE | EXCL implies that
1576                  * but since we want avoid ambiguity (eg when flags
1577                  * is zero) then just set this
1578                  */
1579                 if (n->nlmsg_flags & NLM_F_REPLACE)
1580                         flags = TCA_ACT_FLAGS_REPLACE;
1581                 ret = tcf_action_add(net, tca[TCA_ACT_TAB], n, portid, flags,
1582                                      extack);
1583                 break;
1584         case RTM_DELACTION:
1585                 ret = tca_action_gd(net, tca[TCA_ACT_TAB], n,
1586                                     portid, RTM_DELACTION, extack);
1587                 break;
1588         case RTM_GETACTION:
1589                 ret = tca_action_gd(net, tca[TCA_ACT_TAB], n,
1590                                     portid, RTM_GETACTION, extack);
1591                 break;
1592         default:
1593                 BUG();
1594         }
1595
1596         return ret;
1597 }
1598
1599 static struct nlattr *find_dump_kind(struct nlattr **nla)
1600 {
1601         struct nlattr *tb1, *tb2[TCA_ACT_MAX + 1];
1602         struct nlattr *tb[TCA_ACT_MAX_PRIO + 1];
1603         struct nlattr *kind;
1604
1605         tb1 = nla[TCA_ACT_TAB];
1606         if (tb1 == NULL)
1607                 return NULL;
1608
1609         if (nla_parse_deprecated(tb, TCA_ACT_MAX_PRIO, nla_data(tb1), NLMSG_ALIGN(nla_len(tb1)), NULL, NULL) < 0)
1610                 return NULL;
1611
1612         if (tb[1] == NULL)
1613                 return NULL;
1614         if (nla_parse_nested_deprecated(tb2, TCA_ACT_MAX, tb[1], tcf_action_policy, NULL) < 0)
1615                 return NULL;
1616         kind = tb2[TCA_ACT_KIND];
1617
1618         return kind;
1619 }
1620
1621 static int tc_dump_action(struct sk_buff *skb, struct netlink_callback *cb)
1622 {
1623         struct net *net = sock_net(skb->sk);
1624         struct nlmsghdr *nlh;
1625         unsigned char *b = skb_tail_pointer(skb);
1626         struct nlattr *nest;
1627         struct tc_action_ops *a_o;
1628         int ret = 0;
1629         struct tcamsg *t = (struct tcamsg *) nlmsg_data(cb->nlh);
1630         struct nlattr *tb[TCA_ROOT_MAX + 1];
1631         struct nlattr *count_attr = NULL;
1632         unsigned long jiffy_since = 0;
1633         struct nlattr *kind = NULL;
1634         struct nla_bitfield32 bf;
1635         u32 msecs_since = 0;
1636         u32 act_count = 0;
1637
1638         ret = nlmsg_parse_deprecated(cb->nlh, sizeof(struct tcamsg), tb,
1639                                      TCA_ROOT_MAX, tcaa_policy, cb->extack);
1640         if (ret < 0)
1641                 return ret;
1642
1643         kind = find_dump_kind(tb);
1644         if (kind == NULL) {
1645                 pr_info("tc_dump_action: action bad kind\n");
1646                 return 0;
1647         }
1648
1649         a_o = tc_lookup_action(kind);
1650         if (a_o == NULL)
1651                 return 0;
1652
1653         cb->args[2] = 0;
1654         if (tb[TCA_ROOT_FLAGS]) {
1655                 bf = nla_get_bitfield32(tb[TCA_ROOT_FLAGS]);
1656                 cb->args[2] = bf.value;
1657         }
1658
1659         if (tb[TCA_ROOT_TIME_DELTA]) {
1660                 msecs_since = nla_get_u32(tb[TCA_ROOT_TIME_DELTA]);
1661         }
1662
1663         nlh = nlmsg_put(skb, NETLINK_CB(cb->skb).portid, cb->nlh->nlmsg_seq,
1664                         cb->nlh->nlmsg_type, sizeof(*t), 0);
1665         if (!nlh)
1666                 goto out_module_put;
1667
1668         if (msecs_since)
1669                 jiffy_since = jiffies - msecs_to_jiffies(msecs_since);
1670
1671         t = nlmsg_data(nlh);
1672         t->tca_family = AF_UNSPEC;
1673         t->tca__pad1 = 0;
1674         t->tca__pad2 = 0;
1675         cb->args[3] = jiffy_since;
1676         count_attr = nla_reserve(skb, TCA_ROOT_COUNT, sizeof(u32));
1677         if (!count_attr)
1678                 goto out_module_put;
1679
1680         nest = nla_nest_start_noflag(skb, TCA_ACT_TAB);
1681         if (nest == NULL)
1682                 goto out_module_put;
1683
1684         ret = a_o->walk(net, skb, cb, RTM_GETACTION, a_o, NULL);
1685         if (ret < 0)
1686                 goto out_module_put;
1687
1688         if (ret > 0) {
1689                 nla_nest_end(skb, nest);
1690                 ret = skb->len;
1691                 act_count = cb->args[1];
1692                 memcpy(nla_data(count_attr), &act_count, sizeof(u32));
1693                 cb->args[1] = 0;
1694         } else
1695                 nlmsg_trim(skb, b);
1696
1697         nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1698         if (NETLINK_CB(cb->skb).portid && ret)
1699                 nlh->nlmsg_flags |= NLM_F_MULTI;
1700         module_put(a_o->owner);
1701         return skb->len;
1702
1703 out_module_put:
1704         module_put(a_o->owner);
1705         nlmsg_trim(skb, b);
1706         return skb->len;
1707 }
1708
1709 static int __init tc_action_init(void)
1710 {
1711         rtnl_register(PF_UNSPEC, RTM_NEWACTION, tc_ctl_action, NULL, 0);
1712         rtnl_register(PF_UNSPEC, RTM_DELACTION, tc_ctl_action, NULL, 0);
1713         rtnl_register(PF_UNSPEC, RTM_GETACTION, tc_ctl_action, tc_dump_action,
1714                       0);
1715
1716         return 0;
1717 }
1718
1719 subsys_initcall(tc_action_init);