sched/act_mirred.c

   1 // SPDX-License-Identifier: GPL-2.0-or-later
   2 /*
   3  * net/sched/act_mirred.c       packet mirroring and redirect actions
   4  *
   5  * Authors:     Jamal Hadi Salim (2002-4)
   6  *
   7  * TODO: Add ingress support (and socket redirect support)
   8  */
   9
  10 #include <linux/types.h>
  11 #include <linux/kernel.h>
  12 #include <linux/string.h>
  13 #include <linux/errno.h>
  14 #include <linux/skbuff.h>
  15 #include <linux/rtnetlink.h>
  16 #include <linux/module.h>
  17 #include <linux/init.h>
  18 #include <linux/gfp.h>
  19 #include <linux/if_arp.h>
  20 #include <net/net_namespace.h>
  21 #include <net/netlink.h>
  22 #include <net/dst.h>
  23 #include <net/pkt_sched.h>
  24 #include <net/pkt_cls.h>
  25 #include <linux/tc_act/tc_mirred.h>
  26 #include <net/tc_act/tc_mirred.h>
  27
  28 static LIST_HEAD(mirred_list);
  29 static DEFINE_SPINLOCK(mirred_list_lock);
  30
  31 #define MIRRED_NEST_LIMIT    4
  32 static DEFINE_PER_CPU(unsigned int, mirred_nest_level);
  33
  34 static bool tcf_mirred_is_act_redirect(int action)
  35 {
  36         return action == TCA_EGRESS_REDIR || action == TCA_INGRESS_REDIR;
  37 }
  38
  39 static bool tcf_mirred_act_wants_ingress(int action)
  40 {
  41         switch (action) {
  42         case TCA_EGRESS_REDIR:
  43         case TCA_EGRESS_MIRROR:
  44                 return false;
  45         case TCA_INGRESS_REDIR:
  46         case TCA_INGRESS_MIRROR:
  47                 return true;
  48         default:
  49                 BUG();
  50         }
  51 }
  52
  53 static bool tcf_mirred_can_reinsert(int action)
  54 {
  55         switch (action) {
  56         case TC_ACT_SHOT:
  57         case TC_ACT_STOLEN:
  58         case TC_ACT_QUEUED:
  59         case TC_ACT_TRAP:
  60                 return true;
  61         }
  62         return false;
  63 }
  64
  65 static struct net_device *tcf_mirred_dev_dereference(struct tcf_mirred *m)
  66 {
  67         return rcu_dereference_protected(m->tcfm_dev,
  68                                          lockdep_is_held(&m->tcf_lock));
  69 }
  70
  71 static void tcf_mirred_release(struct tc_action *a)
  72 {
  73         struct tcf_mirred *m = to_mirred(a);
  74         struct net_device *dev;
  75
  76         spin_lock(&mirred_list_lock);
  77         list_del(&m->tcfm_list);
  78         spin_unlock(&mirred_list_lock);
  79
  80         /* last reference to action, no need to lock */
  81         dev = rcu_dereference_protected(m->tcfm_dev, 1);
  82         netdev_put(dev, &m->tcfm_dev_tracker);
  83 }
  84
  85 static const struct nla_policy mirred_policy[TCA_MIRRED_MAX + 1] = {
  86         [TCA_MIRRED_PARMS]      = { .len = sizeof(struct tc_mirred) },
  87 };
  88
  89 static struct tc_action_ops act_mirred_ops;
  90
  91 static int tcf_mirred_init(struct net *net, struct nlattr *nla,
  92                            struct nlattr *est, struct tc_action **a,
  93                            struct tcf_proto *tp,
  94                            u32 flags, struct netlink_ext_ack *extack)
  95 {
  96         struct tc_action_net *tn = net_generic(net, act_mirred_ops.net_id);
  97         bool bind = flags & TCA_ACT_FLAGS_BIND;
  98         struct nlattr *tb[TCA_MIRRED_MAX + 1];
  99         struct tcf_chain *goto_ch = NULL;
 100         bool mac_header_xmit = false;
 101         struct tc_mirred *parm;
 102         struct tcf_mirred *m;
 103         bool exists = false;
 104         int ret, err;
 105         u32 index;
 106
 107         if (!nla) {
 108                 NL_SET_ERR_MSG_MOD(extack, "Mirred requires attributes to be passed");
 109                 return -EINVAL;
 110         }
 111         ret = nla_parse_nested_deprecated(tb, TCA_MIRRED_MAX, nla,
 112                                           mirred_policy, extack);
 113         if (ret < 0)
 114                 return ret;
 115         if (!tb[TCA_MIRRED_PARMS]) {
 116                 NL_SET_ERR_MSG_MOD(extack, "Missing required mirred parameters");
 117                 return -EINVAL;
 118         }
 119         parm = nla_data(tb[TCA_MIRRED_PARMS]);
 120         index = parm->index;
 121         err = tcf_idr_check_alloc(tn, &index, a, bind);
 122         if (err < 0)
 123                 return err;
 124         exists = err;
 125         if (exists && bind)
 126                 return 0;
 127
 128         switch (parm->eaction) {
 129         case TCA_EGRESS_MIRROR:
 130         case TCA_EGRESS_REDIR:
 131         case TCA_INGRESS_REDIR:
 132         case TCA_INGRESS_MIRROR:
 133                 break;
 134         default:
 135                 if (exists)
 136                         tcf_idr_release(*a, bind);
 137                 else
 138                         tcf_idr_cleanup(tn, index);
 139                 NL_SET_ERR_MSG_MOD(extack, "Unknown mirred option");
 140                 return -EINVAL;
 141         }
 142
 143         if (!exists) {
 144                 if (!parm->ifindex) {
 145                         tcf_idr_cleanup(tn, index);
 146                         NL_SET_ERR_MSG_MOD(extack, "Specified device does not exist");
 147                         return -EINVAL;
 148                 }
 149                 ret = tcf_idr_create_from_flags(tn, index, est, a,
 150                                                 &act_mirred_ops, bind, flags);
 151                 if (ret) {
 152                         tcf_idr_cleanup(tn, index);
 153                         return ret;
 154                 }
 155                 ret = ACT_P_CREATED;
 156         } else if (!(flags & TCA_ACT_FLAGS_REPLACE)) {
 157                 tcf_idr_release(*a, bind);
 158                 return -EEXIST;
 159         }
 160
 161         m = to_mirred(*a);
 162         if (ret == ACT_P_CREATED)
 163                 INIT_LIST_HEAD(&m->tcfm_list);
 164
 165         err = tcf_action_check_ctrlact(parm->action, tp, &goto_ch, extack);
 166         if (err < 0)
 167                 goto release_idr;
 168
 169         spin_lock_bh(&m->tcf_lock);
 170
 171         if (parm->ifindex) {
 172                 struct net_device *odev, *ndev;
 173
 174                 ndev = dev_get_by_index(net, parm->ifindex);
 175                 if (!ndev) {
 176                         spin_unlock_bh(&m->tcf_lock);
 177                         err = -ENODEV;
 178                         goto put_chain;
 179                 }
 180                 mac_header_xmit = dev_is_mac_header_xmit(ndev);
 181                 odev = rcu_replace_pointer(m->tcfm_dev, ndev,
 182                                           lockdep_is_held(&m->tcf_lock));
 183                 netdev_put(odev, &m->tcfm_dev_tracker);
 184                 netdev_tracker_alloc(ndev, &m->tcfm_dev_tracker, GFP_ATOMIC);
 185                 m->tcfm_mac_header_xmit = mac_header_xmit;
 186         }
 187         goto_ch = tcf_action_set_ctrlact(*a, parm->action, goto_ch);
 188         m->tcfm_eaction = parm->eaction;
 189         spin_unlock_bh(&m->tcf_lock);
 190         if (goto_ch)
 191                 tcf_chain_put_by_act(goto_ch);
 192
 193         if (ret == ACT_P_CREATED) {
 194                 spin_lock(&mirred_list_lock);
 195                 list_add(&m->tcfm_list, &mirred_list);
 196                 spin_unlock(&mirred_list_lock);
 197         }
 198
 199         return ret;
 200 put_chain:
 201         if (goto_ch)
 202                 tcf_chain_put_by_act(goto_ch);
 203 release_idr:
 204         tcf_idr_release(*a, bind);
 205         return err;
 206 }
 207
 208 static bool is_mirred_nested(void)
 209 {
 210         return unlikely(__this_cpu_read(mirred_nest_level) > 1);
 211 }
 212
 213 static int tcf_mirred_forward(bool want_ingress, struct sk_buff *skb)
 214 {
 215         int err;
 216
 217         if (!want_ingress)
 218                 err = tcf_dev_queue_xmit(skb, dev_queue_xmit);
 219         else if (is_mirred_nested())
 220                 err = netif_rx(skb);
 221         else
 222                 err = netif_receive_skb(skb);
 223
 224         return err;
 225 }
 226
 227 static int tcf_mirred_act(struct sk_buff *skb, const struct tc_action *a,
 228                           struct tcf_result *res)
 229 {
 230         struct tcf_mirred *m = to_mirred(a);
 231         struct sk_buff *skb2 = skb;
 232         bool m_mac_header_xmit;
 233         struct net_device *dev;
 234         unsigned int nest_level;
 235         int retval, err = 0;
 236         bool use_reinsert;
 237         bool want_ingress;
 238         bool is_redirect;
 239         bool expects_nh;
 240         bool at_ingress;
 241         int m_eaction;
 242         int mac_len;
 243         bool at_nh;
 244
 245         nest_level = __this_cpu_inc_return(mirred_nest_level);
 246         if (unlikely(nest_level > MIRRED_NEST_LIMIT)) {
 247                 net_warn_ratelimited("Packet exceeded mirred recursion limit on dev %s\n",
 248                                      netdev_name(skb->dev));
 249                 __this_cpu_dec(mirred_nest_level);
 250                 return TC_ACT_SHOT;
 251         }
 252
 253         tcf_lastuse_update(&m->tcf_tm);
 254         tcf_action_update_bstats(&m->common, skb);
 255
 256         m_mac_header_xmit = READ_ONCE(m->tcfm_mac_header_xmit);
 257         m_eaction = READ_ONCE(m->tcfm_eaction);
 258         retval = READ_ONCE(m->tcf_action);
 259         dev = rcu_dereference_bh(m->tcfm_dev);
 260         if (unlikely(!dev)) {
 261                 pr_notice_once("tc mirred: target device is gone\n");
 262                 goto out;
 263         }
 264
 265         if (unlikely(!(dev->flags & IFF_UP)) || !netif_carrier_ok(dev)) {
 266                 net_notice_ratelimited("tc mirred to Houston: device %s is down\n",
 267                                        dev->name);
 268                 goto out;
 269         }
 270
 271         /* we could easily avoid the clone only if called by ingress and clsact;
 272          * since we can't easily detect the clsact caller, skip clone only for
 273          * ingress - that covers the TC S/W datapath.
 274          */
 275         is_redirect = tcf_mirred_is_act_redirect(m_eaction);
 276         at_ingress = skb_at_tc_ingress(skb);
 277         use_reinsert = at_ingress && is_redirect &&
 278                        tcf_mirred_can_reinsert(retval);
 279         if (!use_reinsert) {
 280                 skb2 = skb_clone(skb, GFP_ATOMIC);
 281                 if (!skb2)
 282                         goto out;
 283         }
 284
 285         want_ingress = tcf_mirred_act_wants_ingress(m_eaction);
 286
 287         /* All mirred/redirected skbs should clear previous ct info */
 288         nf_reset_ct(skb2);
 289         if (want_ingress && !at_ingress) /* drop dst for egress -> ingress */
 290                 skb_dst_drop(skb2);
 291
 292         expects_nh = want_ingress || !m_mac_header_xmit;
 293         at_nh = skb->data == skb_network_header(skb);
 294         if (at_nh != expects_nh) {
 295                 mac_len = skb_at_tc_ingress(skb) ? skb->mac_len :
 296                           skb_network_header(skb) - skb_mac_header(skb);
 297                 if (expects_nh) {
 298                         /* target device/action expect data at nh */
 299                         skb_pull_rcsum(skb2, mac_len);
 300                 } else {
 301                         /* target device/action expect data at mac */
 302                         skb_push_rcsum(skb2, mac_len);
 303                 }
 304         }
 305
 306         skb2->skb_iif = skb->dev->ifindex;
 307         skb2->dev = dev;
 308
 309         /* mirror is always swallowed */
 310         if (is_redirect) {
 311                 skb_set_redirected(skb2, skb2->tc_at_ingress);
 312
 313                 /* let's the caller reinsert the packet, if possible */
 314                 if (use_reinsert) {
 315                         err = tcf_mirred_forward(want_ingress, skb);
 316                         if (err)
 317                                 tcf_action_inc_overlimit_qstats(&m->common);
 318                         __this_cpu_dec(mirred_nest_level);
 319                         return TC_ACT_CONSUMED;
 320                 }
 321         }
 322
 323         err = tcf_mirred_forward(want_ingress, skb2);
 324         if (err) {
 325 out:
 326                 tcf_action_inc_overlimit_qstats(&m->common);
 327                 if (tcf_mirred_is_act_redirect(m_eaction))
 328                         retval = TC_ACT_SHOT;
 329         }
 330         __this_cpu_dec(mirred_nest_level);
 331
 332         return retval;
 333 }
 334
 335 static void tcf_stats_update(struct tc_action *a, u64 bytes, u64 packets,
 336                              u64 drops, u64 lastuse, bool hw)
 337 {
 338         struct tcf_mirred *m = to_mirred(a);
 339         struct tcf_t *tm = &m->tcf_tm;
 340
 341         tcf_action_update_stats(a, bytes, packets, drops, hw);
 342         tm->lastuse = max_t(u64, tm->lastuse, lastuse);
 343 }
 344
 345 static int tcf_mirred_dump(struct sk_buff *skb, struct tc_action *a, int bind,
 346                            int ref)
 347 {
 348         unsigned char *b = skb_tail_pointer(skb);
 349         struct tcf_mirred *m = to_mirred(a);
 350         struct tc_mirred opt = {
 351                 .index   = m->tcf_index,
 352                 .refcnt  = refcount_read(&m->tcf_refcnt) - ref,
 353                 .bindcnt = atomic_read(&m->tcf_bindcnt) - bind,
 354         };
 355         struct net_device *dev;
 356         struct tcf_t t;
 357
 358         spin_lock_bh(&m->tcf_lock);
 359         opt.action = m->tcf_action;
 360         opt.eaction = m->tcfm_eaction;
 361         dev = tcf_mirred_dev_dereference(m);
 362         if (dev)
 363                 opt.ifindex = dev->ifindex;
 364
 365         if (nla_put(skb, TCA_MIRRED_PARMS, sizeof(opt), &opt))
 366                 goto nla_put_failure;
 367
 368         tcf_tm_dump(&t, &m->tcf_tm);
 369         if (nla_put_64bit(skb, TCA_MIRRED_TM, sizeof(t), &t, TCA_MIRRED_PAD))
 370                 goto nla_put_failure;
 371         spin_unlock_bh(&m->tcf_lock);
 372
 373         return skb->len;
 374
 375 nla_put_failure:
 376         spin_unlock_bh(&m->tcf_lock);
 377         nlmsg_trim(skb, b);
 378         return -1;
 379 }
 380
 381 static int mirred_device_event(struct notifier_block *unused,
 382                                unsigned long event, void *ptr)
 383 {
 384         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
 385         struct tcf_mirred *m;
 386
 387         ASSERT_RTNL();
 388         if (event == NETDEV_UNREGISTER) {
 389                 spin_lock(&mirred_list_lock);
 390                 list_for_each_entry(m, &mirred_list, tcfm_list) {
 391                         spin_lock_bh(&m->tcf_lock);
 392                         if (tcf_mirred_dev_dereference(m) == dev) {
 393                                 netdev_put(dev, &m->tcfm_dev_tracker);
 394                                 /* Note : no rcu grace period necessary, as
 395                                  * net_device are already rcu protected.
 396                                  */
 397                                 RCU_INIT_POINTER(m->tcfm_dev, NULL);
 398                         }
 399                         spin_unlock_bh(&m->tcf_lock);
 400                 }
 401                 spin_unlock(&mirred_list_lock);
 402         }
 403
 404         return NOTIFY_DONE;
 405 }
 406
 407 static struct notifier_block mirred_device_notifier = {
 408         .notifier_call = mirred_device_event,
 409 };
 410
 411 static void tcf_mirred_dev_put(void *priv)
 412 {
 413         struct net_device *dev = priv;
 414
 415         dev_put(dev);
 416 }
 417
 418 static struct net_device *
 419 tcf_mirred_get_dev(const struct tc_action *a,
 420                    tc_action_priv_destructor *destructor)
 421 {
 422         struct tcf_mirred *m = to_mirred(a);
 423         struct net_device *dev;
 424
 425         rcu_read_lock();
 426         dev = rcu_dereference(m->tcfm_dev);
 427         if (dev) {
 428                 dev_hold(dev);
 429                 *destructor = tcf_mirred_dev_put;
 430         }
 431         rcu_read_unlock();
 432
 433         return dev;
 434 }
 435
 436 static size_t tcf_mirred_get_fill_size(const struct tc_action *act)
 437 {
 438         return nla_total_size(sizeof(struct tc_mirred));
 439 }
 440
 441 static void tcf_offload_mirred_get_dev(struct flow_action_entry *entry,
 442                                        const struct tc_action *act)
 443 {
 444         entry->dev = act->ops->get_dev(act, &entry->destructor);
 445         if (!entry->dev)
 446                 return;
 447         entry->destructor_priv = entry->dev;
 448 }
 449
 450 static int tcf_mirred_offload_act_setup(struct tc_action *act, void *entry_data,
 451                                         u32 *index_inc, bool bind,
 452                                         struct netlink_ext_ack *extack)
 453 {
 454         if (bind) {
 455                 struct flow_action_entry *entry = entry_data;
 456
 457                 if (is_tcf_mirred_egress_redirect(act)) {
 458                         entry->id = FLOW_ACTION_REDIRECT;
 459                         tcf_offload_mirred_get_dev(entry, act);
 460                 } else if (is_tcf_mirred_egress_mirror(act)) {
 461                         entry->id = FLOW_ACTION_MIRRED;
 462                         tcf_offload_mirred_get_dev(entry, act);
 463                 } else if (is_tcf_mirred_ingress_redirect(act)) {
 464                         entry->id = FLOW_ACTION_REDIRECT_INGRESS;
 465                         tcf_offload_mirred_get_dev(entry, act);
 466                 } else if (is_tcf_mirred_ingress_mirror(act)) {
 467                         entry->id = FLOW_ACTION_MIRRED_INGRESS;
 468                         tcf_offload_mirred_get_dev(entry, act);
 469                 } else {
 470                         NL_SET_ERR_MSG_MOD(extack, "Unsupported mirred offload");
 471                         return -EOPNOTSUPP;
 472                 }
 473                 *index_inc = 1;
 474         } else {
 475                 struct flow_offload_action *fl_action = entry_data;
 476
 477                 if (is_tcf_mirred_egress_redirect(act))
 478                         fl_action->id = FLOW_ACTION_REDIRECT;
 479                 else if (is_tcf_mirred_egress_mirror(act))
 480                         fl_action->id = FLOW_ACTION_MIRRED;
 481                 else if (is_tcf_mirred_ingress_redirect(act))
 482                         fl_action->id = FLOW_ACTION_REDIRECT_INGRESS;
 483                 else if (is_tcf_mirred_ingress_mirror(act))
 484                         fl_action->id = FLOW_ACTION_MIRRED_INGRESS;
 485                 else
 486                         return -EOPNOTSUPP;
 487         }
 488
 489         return 0;
 490 }
 491
 492 static struct tc_action_ops act_mirred_ops = {
 493         .kind           =       "mirred",
 494         .id             =       TCA_ID_MIRRED,
 495         .owner          =       THIS_MODULE,
 496         .act            =       tcf_mirred_act,
 497         .stats_update   =       tcf_stats_update,
 498         .dump           =       tcf_mirred_dump,
 499         .cleanup        =       tcf_mirred_release,
 500         .init           =       tcf_mirred_init,
 501         .get_fill_size  =       tcf_mirred_get_fill_size,
 502         .offload_act_setup =    tcf_mirred_offload_act_setup,
 503         .size           =       sizeof(struct tcf_mirred),
 504         .get_dev        =       tcf_mirred_get_dev,
 505 };
 506
 507 static __net_init int mirred_init_net(struct net *net)
 508 {
 509         struct tc_action_net *tn = net_generic(net, act_mirred_ops.net_id);
 510
 511         return tc_action_net_init(net, tn, &act_mirred_ops);
 512 }
 513
 514 static void __net_exit mirred_exit_net(struct list_head *net_list)
 515 {
 516         tc_action_net_exit(net_list, act_mirred_ops.net_id);
 517 }
 518
 519 static struct pernet_operations mirred_net_ops = {
 520         .init = mirred_init_net,
 521         .exit_batch = mirred_exit_net,
 522         .id   = &act_mirred_ops.net_id,
 523         .size = sizeof(struct tc_action_net),
 524 };
 525
 526 MODULE_AUTHOR("Jamal Hadi Salim(2002)");
 527 MODULE_DESCRIPTION("Device Mirror/redirect actions");
 528 MODULE_LICENSE("GPL");
 529
 530 static int __init mirred_init_module(void)
 531 {
 532         int err = register_netdevice_notifier(&mirred_device_notifier);
 533         if (err)
 534                 return err;
 535
 536         pr_info("Mirror/redirect action on\n");
 537         err = tcf_register_action(&act_mirred_ops, &mirred_net_ops);
 538         if (err)
 539                 unregister_netdevice_notifier(&mirred_device_notifier);
 540
 541         return err;
 542 }
 543
 544 static void __exit mirred_cleanup_module(void)
 545 {
 546         tcf_unregister_action(&act_mirred_ops, &mirred_net_ops);
 547         unregister_netdevice_notifier(&mirred_device_notifier);
 548 }
 549
 550 module_init(mirred_init_module);
 551 module_exit(mirred_cleanup_module);