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#include <net/tc_wrapper.h> 28 29static LIST_HEAD(mirred_list); 30static DEFINE_SPINLOCK(mirred_list_lock); 31 32#define MIRRED_NEST_LIMIT 4 33static DEFINE_PER_CPU(unsigned int, mirred_nest_level); 34 35static bool tcf_mirred_is_act_redirect(int action) 36{ 37 return action == TCA_EGRESS_REDIR || action == TCA_INGRESS_REDIR; 38} 39 40static bool tcf_mirred_act_wants_ingress(int action) 41{ 42 switch (action) { 43 case TCA_EGRESS_REDIR: 44 case TCA_EGRESS_MIRROR: 45 return false; 46 case TCA_INGRESS_REDIR: 47 case TCA_INGRESS_MIRROR: 48 return true; 49 default: 50 BUG(); 51 } 52} 53 54static bool tcf_mirred_can_reinsert(int action) 55{ 56 switch (action) { 57 case TC_ACT_SHOT: 58 case TC_ACT_STOLEN: 59 case TC_ACT_QUEUED: 60 case TC_ACT_TRAP: 61 return true; 62 } 63 return false; 64} 65 66static struct net_device *tcf_mirred_dev_dereference(struct tcf_mirred *m) 67{ 68 return rcu_dereference_protected(m->tcfm_dev, 69 lockdep_is_held(&m->tcf_lock)); 70} 71 72static void tcf_mirred_release(struct tc_action *a) 73{ 74 struct tcf_mirred *m = to_mirred(a); 75 struct net_device *dev; 76 77 spin_lock(&mirred_list_lock); 78 list_del(&m->tcfm_list); 79 spin_unlock(&mirred_list_lock); 80 81 /* last reference to action, no need to lock */ 82 dev = rcu_dereference_protected(m->tcfm_dev, 1); 83 netdev_put(dev, &m->tcfm_dev_tracker); 84} 85 86static const struct nla_policy mirred_policy[TCA_MIRRED_MAX + 1] = { 87 [TCA_MIRRED_PARMS] = { .len = sizeof(struct tc_mirred) }, 88 [TCA_MIRRED_BLOCKID] = NLA_POLICY_MIN(NLA_U32, 1), 89}; 90 91static struct tc_action_ops act_mirred_ops; 92 93static void tcf_mirred_replace_dev(struct tcf_mirred *m, 94 struct net_device *ndev) 95{ 96 struct net_device *odev; 97 98 odev = rcu_replace_pointer(m->tcfm_dev, ndev, 99 lockdep_is_held(&m->tcf_lock)); 100 netdev_put(odev, &m->tcfm_dev_tracker); 101} 102 103static int tcf_mirred_init(struct net *net, struct nlattr *nla, 104 struct nlattr *est, struct tc_action **a, 105 struct tcf_proto *tp, 106 u32 flags, struct netlink_ext_ack *extack) 107{ 108 struct tc_action_net *tn = net_generic(net, act_mirred_ops.net_id); 109 bool bind = flags & TCA_ACT_FLAGS_BIND; 110 struct nlattr *tb[TCA_MIRRED_MAX + 1]; 111 struct tcf_chain *goto_ch = NULL; 112 bool mac_header_xmit = false; 113 struct tc_mirred *parm; 114 struct tcf_mirred *m; 115 bool exists = false; 116 int ret, err; 117 u32 index; 118 119 if (!nla) { 120 NL_SET_ERR_MSG_MOD(extack, "Mirred requires attributes to be passed"); 121 return -EINVAL; 122 } 123 ret = nla_parse_nested_deprecated(tb, TCA_MIRRED_MAX, nla, 124 mirred_policy, extack); 125 if (ret < 0) 126 return ret; 127 if (!tb[TCA_MIRRED_PARMS]) { 128 NL_SET_ERR_MSG_MOD(extack, "Missing required mirred parameters"); 129 return -EINVAL; 130 } 131 parm = nla_data(tb[TCA_MIRRED_PARMS]); 132 index = parm->index; 133 err = tcf_idr_check_alloc(tn, &index, a, bind); 134 if (err < 0) 135 return err; 136 exists = err; 137 if (exists && bind) 138 return ACT_P_BOUND; 139 140 if (tb[TCA_MIRRED_BLOCKID] && parm->ifindex) { 141 NL_SET_ERR_MSG_MOD(extack, 142 "Cannot specify Block ID and dev simultaneously"); 143 if (exists) 144 tcf_idr_release(*a, bind); 145 else 146 tcf_idr_cleanup(tn, index); 147 148 return -EINVAL; 149 } 150 151 switch (parm->eaction) { 152 case TCA_EGRESS_MIRROR: 153 case TCA_EGRESS_REDIR: 154 case TCA_INGRESS_REDIR: 155 case TCA_INGRESS_MIRROR: 156 break; 157 default: 158 if (exists) 159 tcf_idr_release(*a, bind); 160 else 161 tcf_idr_cleanup(tn, index); 162 NL_SET_ERR_MSG_MOD(extack, "Unknown mirred option"); 163 return -EINVAL; 164 } 165 166 if (!exists) { 167 if (!parm->ifindex && !tb[TCA_MIRRED_BLOCKID]) { 168 tcf_idr_cleanup(tn, index); 169 NL_SET_ERR_MSG_MOD(extack, 170 "Must specify device or block"); 171 return -EINVAL; 172 } 173 ret = tcf_idr_create_from_flags(tn, index, est, a, 174 &act_mirred_ops, bind, flags); 175 if (ret) { 176 tcf_idr_cleanup(tn, index); 177 return ret; 178 } 179 ret = ACT_P_CREATED; 180 } else if (!(flags & TCA_ACT_FLAGS_REPLACE)) { 181 tcf_idr_release(*a, bind); 182 return -EEXIST; 183 } 184 185 m = to_mirred(*a); 186 if (ret == ACT_P_CREATED) 187 INIT_LIST_HEAD(&m->tcfm_list); 188 189 err = tcf_action_check_ctrlact(parm->action, tp, &goto_ch, extack); 190 if (err < 0) 191 goto release_idr; 192 193 spin_lock_bh(&m->tcf_lock); 194 195 if (parm->ifindex) { 196 struct net_device *ndev; 197 198 ndev = dev_get_by_index(net, parm->ifindex); 199 if (!ndev) { 200 spin_unlock_bh(&m->tcf_lock); 201 err = -ENODEV; 202 goto put_chain; 203 } 204 mac_header_xmit = dev_is_mac_header_xmit(ndev); 205 tcf_mirred_replace_dev(m, ndev); 206 netdev_tracker_alloc(ndev, &m->tcfm_dev_tracker, GFP_ATOMIC); 207 m->tcfm_mac_header_xmit = mac_header_xmit; 208 m->tcfm_blockid = 0; 209 } else if (tb[TCA_MIRRED_BLOCKID]) { 210 tcf_mirred_replace_dev(m, NULL); 211 m->tcfm_mac_header_xmit = false; 212 m->tcfm_blockid = nla_get_u32(tb[TCA_MIRRED_BLOCKID]); 213 } 214 goto_ch = tcf_action_set_ctrlact(*a, parm->action, goto_ch); 215 m->tcfm_eaction = parm->eaction; 216 spin_unlock_bh(&m->tcf_lock); 217 if (goto_ch) 218 tcf_chain_put_by_act(goto_ch); 219 220 if (ret == ACT_P_CREATED) { 221 spin_lock(&mirred_list_lock); 222 list_add(&m->tcfm_list, &mirred_list); 223 spin_unlock(&mirred_list_lock); 224 } 225 226 return ret; 227put_chain: 228 if (goto_ch) 229 tcf_chain_put_by_act(goto_ch); 230release_idr: 231 tcf_idr_release(*a, bind); 232 return err; 233} 234 235static int 236tcf_mirred_forward(bool at_ingress, bool want_ingress, struct sk_buff *skb) 237{ 238 int err; 239 240 if (!want_ingress) 241 err = tcf_dev_queue_xmit(skb, dev_queue_xmit); 242 else if (!at_ingress) 243 err = netif_rx(skb); 244 else 245 err = netif_receive_skb(skb); 246 247 return err; 248} 249 250static int tcf_mirred_to_dev(struct sk_buff *skb, struct tcf_mirred *m, 251 struct net_device *dev, 252 const bool m_mac_header_xmit, int m_eaction, 253 int retval) 254{ 255 struct sk_buff *skb_to_send = skb; 256 bool want_ingress; 257 bool is_redirect; 258 bool expects_nh; 259 bool at_ingress; 260 bool dont_clone; 261 int mac_len; 262 bool at_nh; 263 int err; 264 265 is_redirect = tcf_mirred_is_act_redirect(m_eaction); 266 if (unlikely(!(dev->flags & IFF_UP)) || !netif_carrier_ok(dev)) { 267 net_notice_ratelimited("tc mirred to Houston: device %s is down\n", 268 dev->name); 269 goto err_cant_do; 270 } 271 272 /* we could easily avoid the clone only if called by ingress and clsact; 273 * since we can't easily detect the clsact caller, skip clone only for 274 * ingress - that covers the TC S/W datapath. 275 */ 276 at_ingress = skb_at_tc_ingress(skb); 277 dont_clone = skb_at_tc_ingress(skb) && is_redirect && 278 tcf_mirred_can_reinsert(retval); 279 if (!dont_clone) { 280 skb_to_send = skb_clone(skb, GFP_ATOMIC); 281 if (!skb_to_send) 282 goto err_cant_do; 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(skb_to_send); 289 if (want_ingress && !at_ingress) /* drop dst for egress -> ingress */ 290 skb_dst_drop(skb_to_send); 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 = at_ingress ? skb->mac_len : 296 skb_network_offset(skb); 297 if (expects_nh) { 298 /* target device/action expect data at nh */ 299 skb_pull_rcsum(skb_to_send, mac_len); 300 } else { 301 /* target device/action expect data at mac */ 302 skb_push_rcsum(skb_to_send, mac_len); 303 } 304 } 305 306 skb_to_send->skb_iif = skb->dev->ifindex; 307 skb_to_send->dev = dev; 308 309 if (is_redirect) { 310 if (skb == skb_to_send) 311 retval = TC_ACT_CONSUMED; 312 313 skb_set_redirected(skb_to_send, skb_to_send->tc_at_ingress); 314 315 err = tcf_mirred_forward(at_ingress, want_ingress, skb_to_send); 316 } else { 317 err = tcf_mirred_forward(at_ingress, want_ingress, skb_to_send); 318 } 319 if (err) 320 tcf_action_inc_overlimit_qstats(&m->common); 321 322 return retval; 323 324err_cant_do: 325 if (is_redirect) 326 retval = TC_ACT_SHOT; 327 tcf_action_inc_overlimit_qstats(&m->common); 328 return retval; 329} 330 331static int tcf_blockcast_redir(struct sk_buff *skb, struct tcf_mirred *m, 332 struct tcf_block *block, int m_eaction, 333 const u32 exception_ifindex, int retval) 334{ 335 struct net_device *dev_prev = NULL; 336 struct net_device *dev = NULL; 337 unsigned long index; 338 int mirred_eaction; 339 340 mirred_eaction = tcf_mirred_act_wants_ingress(m_eaction) ? 341 TCA_INGRESS_MIRROR : TCA_EGRESS_MIRROR; 342 343 xa_for_each(&block->ports, index, dev) { 344 if (index == exception_ifindex) 345 continue; 346 347 if (!dev_prev) 348 goto assign_prev; 349 350 tcf_mirred_to_dev(skb, m, dev_prev, 351 dev_is_mac_header_xmit(dev), 352 mirred_eaction, retval); 353assign_prev: 354 dev_prev = dev; 355 } 356 357 if (dev_prev) 358 return tcf_mirred_to_dev(skb, m, dev_prev, 359 dev_is_mac_header_xmit(dev_prev), 360 m_eaction, retval); 361 362 return retval; 363} 364 365static int tcf_blockcast_mirror(struct sk_buff *skb, struct tcf_mirred *m, 366 struct tcf_block *block, int m_eaction, 367 const u32 exception_ifindex, int retval) 368{ 369 struct net_device *dev = NULL; 370 unsigned long index; 371 372 xa_for_each(&block->ports, index, dev) { 373 if (index == exception_ifindex) 374 continue; 375 376 tcf_mirred_to_dev(skb, m, dev, 377 dev_is_mac_header_xmit(dev), 378 m_eaction, retval); 379 } 380 381 return retval; 382} 383 384static int tcf_blockcast(struct sk_buff *skb, struct tcf_mirred *m, 385 const u32 blockid, struct tcf_result *res, 386 int retval) 387{ 388 const u32 exception_ifindex = skb->dev->ifindex; 389 struct tcf_block *block; 390 bool is_redirect; 391 int m_eaction; 392 393 m_eaction = READ_ONCE(m->tcfm_eaction); 394 is_redirect = tcf_mirred_is_act_redirect(m_eaction); 395 396 /* we are already under rcu protection, so can call block lookup 397 * directly. 398 */ 399 block = tcf_block_lookup(dev_net(skb->dev), blockid); 400 if (!block || xa_empty(&block->ports)) { 401 tcf_action_inc_overlimit_qstats(&m->common); 402 return retval; 403 } 404 405 if (is_redirect) 406 return tcf_blockcast_redir(skb, m, block, m_eaction, 407 exception_ifindex, retval); 408 409 /* If it's not redirect, it is mirror */ 410 return tcf_blockcast_mirror(skb, m, block, m_eaction, exception_ifindex, 411 retval); 412} 413 414TC_INDIRECT_SCOPE int tcf_mirred_act(struct sk_buff *skb, 415 const struct tc_action *a, 416 struct tcf_result *res) 417{ 418 struct tcf_mirred *m = to_mirred(a); 419 int retval = READ_ONCE(m->tcf_action); 420 unsigned int nest_level; 421 bool m_mac_header_xmit; 422 struct net_device *dev; 423 int m_eaction; 424 u32 blockid; 425 426 nest_level = __this_cpu_inc_return(mirred_nest_level); 427 if (unlikely(nest_level > MIRRED_NEST_LIMIT)) { 428 net_warn_ratelimited("Packet exceeded mirred recursion limit on dev %s\n", 429 netdev_name(skb->dev)); 430 retval = TC_ACT_SHOT; 431 goto dec_nest_level; 432 } 433 434 tcf_lastuse_update(&m->tcf_tm); 435 tcf_action_update_bstats(&m->common, skb); 436 437 blockid = READ_ONCE(m->tcfm_blockid); 438 if (blockid) { 439 retval = tcf_blockcast(skb, m, blockid, res, retval); 440 goto dec_nest_level; 441 } 442 443 dev = rcu_dereference_bh(m->tcfm_dev); 444 if (unlikely(!dev)) { 445 pr_notice_once("tc mirred: target device is gone\n"); 446 tcf_action_inc_overlimit_qstats(&m->common); 447 goto dec_nest_level; 448 } 449 450 m_mac_header_xmit = READ_ONCE(m->tcfm_mac_header_xmit); 451 m_eaction = READ_ONCE(m->tcfm_eaction); 452 453 retval = tcf_mirred_to_dev(skb, m, dev, m_mac_header_xmit, m_eaction, 454 retval); 455 456dec_nest_level: 457 __this_cpu_dec(mirred_nest_level); 458 459 return retval; 460} 461 462static void tcf_stats_update(struct tc_action *a, u64 bytes, u64 packets, 463 u64 drops, u64 lastuse, bool hw) 464{ 465 struct tcf_mirred *m = to_mirred(a); 466 struct tcf_t *tm = &m->tcf_tm; 467 468 tcf_action_update_stats(a, bytes, packets, drops, hw); 469 tm->lastuse = max_t(u64, tm->lastuse, lastuse); 470} 471 472static int tcf_mirred_dump(struct sk_buff *skb, struct tc_action *a, int bind, 473 int ref) 474{ 475 unsigned char *b = skb_tail_pointer(skb); 476 struct tcf_mirred *m = to_mirred(a); 477 struct tc_mirred opt = { 478 .index = m->tcf_index, 479 .refcnt = refcount_read(&m->tcf_refcnt) - ref, 480 .bindcnt = atomic_read(&m->tcf_bindcnt) - bind, 481 }; 482 struct net_device *dev; 483 struct tcf_t t; 484 u32 blockid; 485 486 spin_lock_bh(&m->tcf_lock); 487 opt.action = m->tcf_action; 488 opt.eaction = m->tcfm_eaction; 489 dev = tcf_mirred_dev_dereference(m); 490 if (dev) 491 opt.ifindex = dev->ifindex; 492 493 if (nla_put(skb, TCA_MIRRED_PARMS, sizeof(opt), &opt)) 494 goto nla_put_failure; 495 496 blockid = m->tcfm_blockid; 497 if (blockid && nla_put_u32(skb, TCA_MIRRED_BLOCKID, blockid)) 498 goto nla_put_failure; 499 500 tcf_tm_dump(&t, &m->tcf_tm); 501 if (nla_put_64bit(skb, TCA_MIRRED_TM, sizeof(t), &t, TCA_MIRRED_PAD)) 502 goto nla_put_failure; 503 spin_unlock_bh(&m->tcf_lock); 504 505 return skb->len; 506 507nla_put_failure: 508 spin_unlock_bh(&m->tcf_lock); 509 nlmsg_trim(skb, b); 510 return -1; 511} 512 513static int mirred_device_event(struct notifier_block *unused, 514 unsigned long event, void *ptr) 515{ 516 struct net_device *dev = netdev_notifier_info_to_dev(ptr); 517 struct tcf_mirred *m; 518 519 ASSERT_RTNL(); 520 if (event == NETDEV_UNREGISTER) { 521 spin_lock(&mirred_list_lock); 522 list_for_each_entry(m, &mirred_list, tcfm_list) { 523 spin_lock_bh(&m->tcf_lock); 524 if (tcf_mirred_dev_dereference(m) == dev) { 525 netdev_put(dev, &m->tcfm_dev_tracker); 526 /* Note : no rcu grace period necessary, as 527 * net_device are already rcu protected. 528 */ 529 RCU_INIT_POINTER(m->tcfm_dev, NULL); 530 } 531 spin_unlock_bh(&m->tcf_lock); 532 } 533 spin_unlock(&mirred_list_lock); 534 } 535 536 return NOTIFY_DONE; 537} 538 539static struct notifier_block mirred_device_notifier = { 540 .notifier_call = mirred_device_event, 541}; 542 543static void tcf_mirred_dev_put(void *priv) 544{ 545 struct net_device *dev = priv; 546 547 dev_put(dev); 548} 549 550static struct net_device * 551tcf_mirred_get_dev(const struct tc_action *a, 552 tc_action_priv_destructor *destructor) 553{ 554 struct tcf_mirred *m = to_mirred(a); 555 struct net_device *dev; 556 557 rcu_read_lock(); 558 dev = rcu_dereference(m->tcfm_dev); 559 if (dev) { 560 dev_hold(dev); 561 *destructor = tcf_mirred_dev_put; 562 } 563 rcu_read_unlock(); 564 565 return dev; 566} 567 568static size_t tcf_mirred_get_fill_size(const struct tc_action *act) 569{ 570 return nla_total_size(sizeof(struct tc_mirred)); 571} 572 573static void tcf_offload_mirred_get_dev(struct flow_action_entry *entry, 574 const struct tc_action *act) 575{ 576 entry->dev = act->ops->get_dev(act, &entry->destructor); 577 if (!entry->dev) 578 return; 579 entry->destructor_priv = entry->dev; 580} 581 582static int tcf_mirred_offload_act_setup(struct tc_action *act, void *entry_data, 583 u32 *index_inc, bool bind, 584 struct netlink_ext_ack *extack) 585{ 586 if (bind) { 587 struct flow_action_entry *entry = entry_data; 588 589 if (is_tcf_mirred_egress_redirect(act)) { 590 entry->id = FLOW_ACTION_REDIRECT; 591 tcf_offload_mirred_get_dev(entry, act); 592 } else if (is_tcf_mirred_egress_mirror(act)) { 593 entry->id = FLOW_ACTION_MIRRED; 594 tcf_offload_mirred_get_dev(entry, act); 595 } else if (is_tcf_mirred_ingress_redirect(act)) { 596 entry->id = FLOW_ACTION_REDIRECT_INGRESS; 597 tcf_offload_mirred_get_dev(entry, act); 598 } else if (is_tcf_mirred_ingress_mirror(act)) { 599 entry->id = FLOW_ACTION_MIRRED_INGRESS; 600 tcf_offload_mirred_get_dev(entry, act); 601 } else { 602 NL_SET_ERR_MSG_MOD(extack, "Unsupported mirred offload"); 603 return -EOPNOTSUPP; 604 } 605 *index_inc = 1; 606 } else { 607 struct flow_offload_action *fl_action = entry_data; 608 609 if (is_tcf_mirred_egress_redirect(act)) 610 fl_action->id = FLOW_ACTION_REDIRECT; 611 else if (is_tcf_mirred_egress_mirror(act)) 612 fl_action->id = FLOW_ACTION_MIRRED; 613 else if (is_tcf_mirred_ingress_redirect(act)) 614 fl_action->id = FLOW_ACTION_REDIRECT_INGRESS; 615 else if (is_tcf_mirred_ingress_mirror(act)) 616 fl_action->id = FLOW_ACTION_MIRRED_INGRESS; 617 else 618 return -EOPNOTSUPP; 619 } 620 621 return 0; 622} 623 624static struct tc_action_ops act_mirred_ops = { 625 .kind = "mirred", 626 .id = TCA_ID_MIRRED, 627 .owner = THIS_MODULE, 628 .act = tcf_mirred_act, 629 .stats_update = tcf_stats_update, 630 .dump = tcf_mirred_dump, 631 .cleanup = tcf_mirred_release, 632 .init = tcf_mirred_init, 633 .get_fill_size = tcf_mirred_get_fill_size, 634 .offload_act_setup = tcf_mirred_offload_act_setup, 635 .size = sizeof(struct tcf_mirred), 636 .get_dev = tcf_mirred_get_dev, 637}; 638MODULE_ALIAS_NET_ACT("mirred"); 639 640static __net_init int mirred_init_net(struct net *net) 641{ 642 struct tc_action_net *tn = net_generic(net, act_mirred_ops.net_id); 643 644 return tc_action_net_init(net, tn, &act_mirred_ops); 645} 646 647static void __net_exit mirred_exit_net(struct list_head *net_list) 648{ 649 tc_action_net_exit(net_list, act_mirred_ops.net_id); 650} 651 652static struct pernet_operations mirred_net_ops = { 653 .init = mirred_init_net, 654 .exit_batch = mirred_exit_net, 655 .id = &act_mirred_ops.net_id, 656 .size = sizeof(struct tc_action_net), 657}; 658 659MODULE_AUTHOR("Jamal Hadi Salim(2002)"); 660MODULE_DESCRIPTION("Device Mirror/redirect actions"); 661MODULE_LICENSE("GPL"); 662 663static int __init mirred_init_module(void) 664{ 665 int err = register_netdevice_notifier(&mirred_device_notifier); 666 if (err) 667 return err; 668 669 pr_info("Mirror/redirect action on\n"); 670 err = tcf_register_action(&act_mirred_ops, &mirred_net_ops); 671 if (err) 672 unregister_netdevice_notifier(&mirred_device_notifier); 673 674 return err; 675} 676 677static void __exit mirred_cleanup_module(void) 678{ 679 tcf_unregister_action(&act_mirred_ops, &mirred_net_ops); 680 unregister_netdevice_notifier(&mirred_device_notifier); 681} 682 683module_init(mirred_init_module); 684module_exit(mirred_cleanup_module); 685