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_RECURSION_LIMIT 4 32 static DEFINE_PER_CPU(unsigned int, mirred_rec_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 dev_put(dev); 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 unsigned int mirred_net_id; 90 static struct tc_action_ops act_mirred_ops; 91 92 static int tcf_mirred_init(struct net *net, struct nlattr *nla, 93 struct nlattr *est, struct tc_action **a, 94 struct tcf_proto *tp, 95 u32 flags, struct netlink_ext_ack *extack) 96 { 97 struct tc_action_net *tn = net_generic(net, mirred_net_id); 98 bool bind = flags & TCA_ACT_FLAGS_BIND; 99 struct nlattr *tb[TCA_MIRRED_MAX + 1]; 100 struct tcf_chain *goto_ch = NULL; 101 bool mac_header_xmit = false; 102 struct tc_mirred *parm; 103 struct tcf_mirred *m; 104 struct net_device *dev; 105 bool exists = false; 106 int ret, err; 107 u32 index; 108 109 if (!nla) { 110 NL_SET_ERR_MSG_MOD(extack, "Mirred requires attributes to be passed"); 111 return -EINVAL; 112 } 113 ret = nla_parse_nested_deprecated(tb, TCA_MIRRED_MAX, nla, 114 mirred_policy, extack); 115 if (ret < 0) 116 return ret; 117 if (!tb[TCA_MIRRED_PARMS]) { 118 NL_SET_ERR_MSG_MOD(extack, "Missing required mirred parameters"); 119 return -EINVAL; 120 } 121 parm = nla_data(tb[TCA_MIRRED_PARMS]); 122 index = parm->index; 123 err = tcf_idr_check_alloc(tn, &index, a, bind); 124 if (err < 0) 125 return err; 126 exists = err; 127 if (exists && bind) 128 return 0; 129 130 switch (parm->eaction) { 131 case TCA_EGRESS_MIRROR: 132 case TCA_EGRESS_REDIR: 133 case TCA_INGRESS_REDIR: 134 case TCA_INGRESS_MIRROR: 135 break; 136 default: 137 if (exists) 138 tcf_idr_release(*a, bind); 139 else 140 tcf_idr_cleanup(tn, index); 141 NL_SET_ERR_MSG_MOD(extack, "Unknown mirred option"); 142 return -EINVAL; 143 } 144 145 if (!exists) { 146 if (!parm->ifindex) { 147 tcf_idr_cleanup(tn, index); 148 NL_SET_ERR_MSG_MOD(extack, "Specified device does not exist"); 149 return -EINVAL; 150 } 151 ret = tcf_idr_create_from_flags(tn, index, est, a, 152 &act_mirred_ops, bind, flags); 153 if (ret) { 154 tcf_idr_cleanup(tn, index); 155 return ret; 156 } 157 ret = ACT_P_CREATED; 158 } else if (!(flags & TCA_ACT_FLAGS_REPLACE)) { 159 tcf_idr_release(*a, bind); 160 return -EEXIST; 161 } 162 163 m = to_mirred(*a); 164 if (ret == ACT_P_CREATED) 165 INIT_LIST_HEAD(&m->tcfm_list); 166 167 err = tcf_action_check_ctrlact(parm->action, tp, &goto_ch, extack); 168 if (err < 0) 169 goto release_idr; 170 171 spin_lock_bh(&m->tcf_lock); 172 173 if (parm->ifindex) { 174 dev = dev_get_by_index(net, parm->ifindex); 175 if (!dev) { 176 spin_unlock_bh(&m->tcf_lock); 177 err = -ENODEV; 178 goto put_chain; 179 } 180 mac_header_xmit = dev_is_mac_header_xmit(dev); 181 dev = rcu_replace_pointer(m->tcfm_dev, dev, 182 lockdep_is_held(&m->tcf_lock)); 183 dev_put(dev); 184 m->tcfm_mac_header_xmit = mac_header_xmit; 185 } 186 goto_ch = tcf_action_set_ctrlact(*a, parm->action, goto_ch); 187 m->tcfm_eaction = parm->eaction; 188 spin_unlock_bh(&m->tcf_lock); 189 if (goto_ch) 190 tcf_chain_put_by_act(goto_ch); 191 192 if (ret == ACT_P_CREATED) { 193 spin_lock(&mirred_list_lock); 194 list_add(&m->tcfm_list, &mirred_list); 195 spin_unlock(&mirred_list_lock); 196 } 197 198 return ret; 199 put_chain: 200 if (goto_ch) 201 tcf_chain_put_by_act(goto_ch); 202 release_idr: 203 tcf_idr_release(*a, bind); 204 return err; 205 } 206 207 static int tcf_mirred_forward(bool want_ingress, struct sk_buff *skb) 208 { 209 int err; 210 211 if (!want_ingress) 212 err = tcf_dev_queue_xmit(skb, dev_queue_xmit); 213 else 214 err = netif_receive_skb(skb); 215 216 return err; 217 } 218 219 static int tcf_mirred_act(struct sk_buff *skb, const struct tc_action *a, 220 struct tcf_result *res) 221 { 222 struct tcf_mirred *m = to_mirred(a); 223 struct sk_buff *skb2 = skb; 224 bool m_mac_header_xmit; 225 struct net_device *dev; 226 unsigned int rec_level; 227 int retval, err = 0; 228 bool use_reinsert; 229 bool want_ingress; 230 bool is_redirect; 231 bool expects_nh; 232 bool at_ingress; 233 int m_eaction; 234 int mac_len; 235 bool at_nh; 236 237 rec_level = __this_cpu_inc_return(mirred_rec_level); 238 if (unlikely(rec_level > MIRRED_RECURSION_LIMIT)) { 239 net_warn_ratelimited("Packet exceeded mirred recursion limit on dev %s\n", 240 netdev_name(skb->dev)); 241 __this_cpu_dec(mirred_rec_level); 242 return TC_ACT_SHOT; 243 } 244 245 tcf_lastuse_update(&m->tcf_tm); 246 tcf_action_update_bstats(&m->common, skb); 247 248 m_mac_header_xmit = READ_ONCE(m->tcfm_mac_header_xmit); 249 m_eaction = READ_ONCE(m->tcfm_eaction); 250 retval = READ_ONCE(m->tcf_action); 251 dev = rcu_dereference_bh(m->tcfm_dev); 252 if (unlikely(!dev)) { 253 pr_notice_once("tc mirred: target device is gone\n"); 254 goto out; 255 } 256 257 if (unlikely(!(dev->flags & IFF_UP))) { 258 net_notice_ratelimited("tc mirred to Houston: device %s is down\n", 259 dev->name); 260 goto out; 261 } 262 263 /* we could easily avoid the clone only if called by ingress and clsact; 264 * since we can't easily detect the clsact caller, skip clone only for 265 * ingress - that covers the TC S/W datapath. 266 */ 267 is_redirect = tcf_mirred_is_act_redirect(m_eaction); 268 at_ingress = skb_at_tc_ingress(skb); 269 use_reinsert = at_ingress && is_redirect && 270 tcf_mirred_can_reinsert(retval); 271 if (!use_reinsert) { 272 skb2 = skb_clone(skb, GFP_ATOMIC); 273 if (!skb2) 274 goto out; 275 } 276 277 want_ingress = tcf_mirred_act_wants_ingress(m_eaction); 278 279 /* All mirred/redirected skbs should clear previous ct info */ 280 nf_reset_ct(skb2); 281 if (want_ingress && !at_ingress) /* drop dst for egress -> ingress */ 282 skb_dst_drop(skb2); 283 284 expects_nh = want_ingress || !m_mac_header_xmit; 285 at_nh = skb->data == skb_network_header(skb); 286 if (at_nh != expects_nh) { 287 mac_len = skb_at_tc_ingress(skb) ? skb->mac_len : 288 skb_network_header(skb) - skb_mac_header(skb); 289 if (expects_nh) { 290 /* target device/action expect data at nh */ 291 skb_pull_rcsum(skb2, mac_len); 292 } else { 293 /* target device/action expect data at mac */ 294 skb_push_rcsum(skb2, mac_len); 295 } 296 } 297 298 skb2->skb_iif = skb->dev->ifindex; 299 skb2->dev = dev; 300 301 /* mirror is always swallowed */ 302 if (is_redirect) { 303 skb_set_redirected(skb2, skb2->tc_at_ingress); 304 305 /* let's the caller reinsert the packet, if possible */ 306 if (use_reinsert) { 307 res->ingress = want_ingress; 308 err = tcf_mirred_forward(res->ingress, skb); 309 if (err) 310 tcf_action_inc_overlimit_qstats(&m->common); 311 __this_cpu_dec(mirred_rec_level); 312 return TC_ACT_CONSUMED; 313 } 314 } 315 316 err = tcf_mirred_forward(want_ingress, skb2); 317 if (err) { 318 out: 319 tcf_action_inc_overlimit_qstats(&m->common); 320 if (tcf_mirred_is_act_redirect(m_eaction)) 321 retval = TC_ACT_SHOT; 322 } 323 __this_cpu_dec(mirred_rec_level); 324 325 return retval; 326 } 327 328 static void tcf_stats_update(struct tc_action *a, u64 bytes, u64 packets, 329 u64 drops, u64 lastuse, bool hw) 330 { 331 struct tcf_mirred *m = to_mirred(a); 332 struct tcf_t *tm = &m->tcf_tm; 333 334 tcf_action_update_stats(a, bytes, packets, drops, hw); 335 tm->lastuse = max_t(u64, tm->lastuse, lastuse); 336 } 337 338 static int tcf_mirred_dump(struct sk_buff *skb, struct tc_action *a, int bind, 339 int ref) 340 { 341 unsigned char *b = skb_tail_pointer(skb); 342 struct tcf_mirred *m = to_mirred(a); 343 struct tc_mirred opt = { 344 .index = m->tcf_index, 345 .refcnt = refcount_read(&m->tcf_refcnt) - ref, 346 .bindcnt = atomic_read(&m->tcf_bindcnt) - bind, 347 }; 348 struct net_device *dev; 349 struct tcf_t t; 350 351 spin_lock_bh(&m->tcf_lock); 352 opt.action = m->tcf_action; 353 opt.eaction = m->tcfm_eaction; 354 dev = tcf_mirred_dev_dereference(m); 355 if (dev) 356 opt.ifindex = dev->ifindex; 357 358 if (nla_put(skb, TCA_MIRRED_PARMS, sizeof(opt), &opt)) 359 goto nla_put_failure; 360 361 tcf_tm_dump(&t, &m->tcf_tm); 362 if (nla_put_64bit(skb, TCA_MIRRED_TM, sizeof(t), &t, TCA_MIRRED_PAD)) 363 goto nla_put_failure; 364 spin_unlock_bh(&m->tcf_lock); 365 366 return skb->len; 367 368 nla_put_failure: 369 spin_unlock_bh(&m->tcf_lock); 370 nlmsg_trim(skb, b); 371 return -1; 372 } 373 374 static int tcf_mirred_walker(struct net *net, struct sk_buff *skb, 375 struct netlink_callback *cb, int type, 376 const struct tc_action_ops *ops, 377 struct netlink_ext_ack *extack) 378 { 379 struct tc_action_net *tn = net_generic(net, mirred_net_id); 380 381 return tcf_generic_walker(tn, skb, cb, type, ops, extack); 382 } 383 384 static int tcf_mirred_search(struct net *net, struct tc_action **a, u32 index) 385 { 386 struct tc_action_net *tn = net_generic(net, mirred_net_id); 387 388 return tcf_idr_search(tn, a, index); 389 } 390 391 static int mirred_device_event(struct notifier_block *unused, 392 unsigned long event, void *ptr) 393 { 394 struct net_device *dev = netdev_notifier_info_to_dev(ptr); 395 struct tcf_mirred *m; 396 397 ASSERT_RTNL(); 398 if (event == NETDEV_UNREGISTER) { 399 spin_lock(&mirred_list_lock); 400 list_for_each_entry(m, &mirred_list, tcfm_list) { 401 spin_lock_bh(&m->tcf_lock); 402 if (tcf_mirred_dev_dereference(m) == dev) { 403 dev_put(dev); 404 /* Note : no rcu grace period necessary, as 405 * net_device are already rcu protected. 406 */ 407 RCU_INIT_POINTER(m->tcfm_dev, NULL); 408 } 409 spin_unlock_bh(&m->tcf_lock); 410 } 411 spin_unlock(&mirred_list_lock); 412 } 413 414 return NOTIFY_DONE; 415 } 416 417 static struct notifier_block mirred_device_notifier = { 418 .notifier_call = mirred_device_event, 419 }; 420 421 static void tcf_mirred_dev_put(void *priv) 422 { 423 struct net_device *dev = priv; 424 425 dev_put(dev); 426 } 427 428 static struct net_device * 429 tcf_mirred_get_dev(const struct tc_action *a, 430 tc_action_priv_destructor *destructor) 431 { 432 struct tcf_mirred *m = to_mirred(a); 433 struct net_device *dev; 434 435 rcu_read_lock(); 436 dev = rcu_dereference(m->tcfm_dev); 437 if (dev) { 438 dev_hold(dev); 439 *destructor = tcf_mirred_dev_put; 440 } 441 rcu_read_unlock(); 442 443 return dev; 444 } 445 446 static size_t tcf_mirred_get_fill_size(const struct tc_action *act) 447 { 448 return nla_total_size(sizeof(struct tc_mirred)); 449 } 450 451 static struct tc_action_ops act_mirred_ops = { 452 .kind = "mirred", 453 .id = TCA_ID_MIRRED, 454 .owner = THIS_MODULE, 455 .act = tcf_mirred_act, 456 .stats_update = tcf_stats_update, 457 .dump = tcf_mirred_dump, 458 .cleanup = tcf_mirred_release, 459 .init = tcf_mirred_init, 460 .walk = tcf_mirred_walker, 461 .lookup = tcf_mirred_search, 462 .get_fill_size = tcf_mirred_get_fill_size, 463 .size = sizeof(struct tcf_mirred), 464 .get_dev = tcf_mirred_get_dev, 465 }; 466 467 static __net_init int mirred_init_net(struct net *net) 468 { 469 struct tc_action_net *tn = net_generic(net, mirred_net_id); 470 471 return tc_action_net_init(net, tn, &act_mirred_ops); 472 } 473 474 static void __net_exit mirred_exit_net(struct list_head *net_list) 475 { 476 tc_action_net_exit(net_list, mirred_net_id); 477 } 478 479 static struct pernet_operations mirred_net_ops = { 480 .init = mirred_init_net, 481 .exit_batch = mirred_exit_net, 482 .id = &mirred_net_id, 483 .size = sizeof(struct tc_action_net), 484 }; 485 486 MODULE_AUTHOR("Jamal Hadi Salim(2002)"); 487 MODULE_DESCRIPTION("Device Mirror/redirect actions"); 488 MODULE_LICENSE("GPL"); 489 490 static int __init mirred_init_module(void) 491 { 492 int err = register_netdevice_notifier(&mirred_device_notifier); 493 if (err) 494 return err; 495 496 pr_info("Mirror/redirect action on\n"); 497 err = tcf_register_action(&act_mirred_ops, &mirred_net_ops); 498 if (err) 499 unregister_netdevice_notifier(&mirred_device_notifier); 500 501 return err; 502 } 503 504 static void __exit mirred_cleanup_module(void) 505 { 506 tcf_unregister_action(&act_mirred_ops, &mirred_net_ops); 507 unregister_netdevice_notifier(&mirred_device_notifier); 508 } 509 510 module_init(mirred_init_module); 511 module_exit(mirred_cleanup_module); 512