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/pkt_sched.h> 23 #include <net/pkt_cls.h> 24 #include <linux/tc_act/tc_mirred.h> 25 #include <net/tc_act/tc_mirred.h> 26 27 static LIST_HEAD(mirred_list); 28 static DEFINE_SPINLOCK(mirred_list_lock); 29 30 #define MIRRED_RECURSION_LIMIT 4 31 static DEFINE_PER_CPU(unsigned int, mirred_rec_level); 32 33 static bool tcf_mirred_is_act_redirect(int action) 34 { 35 return action == TCA_EGRESS_REDIR || action == TCA_INGRESS_REDIR; 36 } 37 38 static bool tcf_mirred_act_wants_ingress(int action) 39 { 40 switch (action) { 41 case TCA_EGRESS_REDIR: 42 case TCA_EGRESS_MIRROR: 43 return false; 44 case TCA_INGRESS_REDIR: 45 case TCA_INGRESS_MIRROR: 46 return true; 47 default: 48 BUG(); 49 } 50 } 51 52 static bool tcf_mirred_can_reinsert(int action) 53 { 54 switch (action) { 55 case TC_ACT_SHOT: 56 case TC_ACT_STOLEN: 57 case TC_ACT_QUEUED: 58 case TC_ACT_TRAP: 59 return true; 60 } 61 return false; 62 } 63 64 static struct net_device *tcf_mirred_dev_dereference(struct tcf_mirred *m) 65 { 66 return rcu_dereference_protected(m->tcfm_dev, 67 lockdep_is_held(&m->tcf_lock)); 68 } 69 70 static void tcf_mirred_release(struct tc_action *a) 71 { 72 struct tcf_mirred *m = to_mirred(a); 73 struct net_device *dev; 74 75 spin_lock(&mirred_list_lock); 76 list_del(&m->tcfm_list); 77 spin_unlock(&mirred_list_lock); 78 79 /* last reference to action, no need to lock */ 80 dev = rcu_dereference_protected(m->tcfm_dev, 1); 81 if (dev) 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 int ovr, int bind, bool rtnl_held, 95 struct tcf_proto *tp, 96 struct netlink_ext_ack *extack) 97 { 98 struct tc_action_net *tn = net_generic(net, mirred_net_id); 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; 107 108 if (!nla) { 109 NL_SET_ERR_MSG_MOD(extack, "Mirred requires attributes to be passed"); 110 return -EINVAL; 111 } 112 ret = nla_parse_nested_deprecated(tb, TCA_MIRRED_MAX, nla, 113 mirred_policy, extack); 114 if (ret < 0) 115 return ret; 116 if (!tb[TCA_MIRRED_PARMS]) { 117 NL_SET_ERR_MSG_MOD(extack, "Missing required mirred parameters"); 118 return -EINVAL; 119 } 120 parm = nla_data(tb[TCA_MIRRED_PARMS]); 121 122 err = tcf_idr_check_alloc(tn, &parm->index, a, bind); 123 if (err < 0) 124 return err; 125 exists = err; 126 if (exists && bind) 127 return 0; 128 129 switch (parm->eaction) { 130 case TCA_EGRESS_MIRROR: 131 case TCA_EGRESS_REDIR: 132 case TCA_INGRESS_REDIR: 133 case TCA_INGRESS_MIRROR: 134 break; 135 default: 136 if (exists) 137 tcf_idr_release(*a, bind); 138 else 139 tcf_idr_cleanup(tn, parm->index); 140 NL_SET_ERR_MSG_MOD(extack, "Unknown mirred option"); 141 return -EINVAL; 142 } 143 144 if (!exists) { 145 if (!parm->ifindex) { 146 tcf_idr_cleanup(tn, parm->index); 147 NL_SET_ERR_MSG_MOD(extack, "Specified device does not exist"); 148 return -EINVAL; 149 } 150 ret = tcf_idr_create(tn, parm->index, est, a, 151 &act_mirred_ops, bind, true); 152 if (ret) { 153 tcf_idr_cleanup(tn, parm->index); 154 return ret; 155 } 156 ret = ACT_P_CREATED; 157 } else if (!ovr) { 158 tcf_idr_release(*a, bind); 159 return -EEXIST; 160 } 161 162 m = to_mirred(*a); 163 if (ret == ACT_P_CREATED) 164 INIT_LIST_HEAD(&m->tcfm_list); 165 166 err = tcf_action_check_ctrlact(parm->action, tp, &goto_ch, extack); 167 if (err < 0) 168 goto release_idr; 169 170 spin_lock_bh(&m->tcf_lock); 171 172 if (parm->ifindex) { 173 dev = dev_get_by_index(net, parm->ifindex); 174 if (!dev) { 175 spin_unlock_bh(&m->tcf_lock); 176 err = -ENODEV; 177 goto put_chain; 178 } 179 mac_header_xmit = dev_is_mac_header_xmit(dev); 180 rcu_swap_protected(m->tcfm_dev, dev, 181 lockdep_is_held(&m->tcf_lock)); 182 if (dev) 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 tcf_idr_insert(tn, *a); 198 } 199 200 return ret; 201 put_chain: 202 if (goto_ch) 203 tcf_chain_put_by_act(goto_ch); 204 release_idr: 205 tcf_idr_release(*a, bind); 206 return err; 207 } 208 209 static int tcf_mirred_act(struct sk_buff *skb, const struct tc_action *a, 210 struct tcf_result *res) 211 { 212 struct tcf_mirred *m = to_mirred(a); 213 struct sk_buff *skb2 = skb; 214 bool m_mac_header_xmit; 215 struct net_device *dev; 216 unsigned int rec_level; 217 int retval, err = 0; 218 bool use_reinsert; 219 bool want_ingress; 220 bool is_redirect; 221 int m_eaction; 222 int mac_len; 223 224 rec_level = __this_cpu_inc_return(mirred_rec_level); 225 if (unlikely(rec_level > MIRRED_RECURSION_LIMIT)) { 226 net_warn_ratelimited("Packet exceeded mirred recursion limit on dev %s\n", 227 netdev_name(skb->dev)); 228 __this_cpu_dec(mirred_rec_level); 229 return TC_ACT_SHOT; 230 } 231 232 tcf_lastuse_update(&m->tcf_tm); 233 bstats_cpu_update(this_cpu_ptr(m->common.cpu_bstats), skb); 234 235 m_mac_header_xmit = READ_ONCE(m->tcfm_mac_header_xmit); 236 m_eaction = READ_ONCE(m->tcfm_eaction); 237 retval = READ_ONCE(m->tcf_action); 238 dev = rcu_dereference_bh(m->tcfm_dev); 239 if (unlikely(!dev)) { 240 pr_notice_once("tc mirred: target device is gone\n"); 241 goto out; 242 } 243 244 if (unlikely(!(dev->flags & IFF_UP))) { 245 net_notice_ratelimited("tc mirred to Houston: device %s is down\n", 246 dev->name); 247 goto out; 248 } 249 250 /* we could easily avoid the clone only if called by ingress and clsact; 251 * since we can't easily detect the clsact caller, skip clone only for 252 * ingress - that covers the TC S/W datapath. 253 */ 254 is_redirect = tcf_mirred_is_act_redirect(m_eaction); 255 use_reinsert = skb_at_tc_ingress(skb) && is_redirect && 256 tcf_mirred_can_reinsert(retval); 257 if (!use_reinsert) { 258 skb2 = skb_clone(skb, GFP_ATOMIC); 259 if (!skb2) 260 goto out; 261 } 262 263 /* If action's target direction differs than filter's direction, 264 * and devices expect a mac header on xmit, then mac push/pull is 265 * needed. 266 */ 267 want_ingress = tcf_mirred_act_wants_ingress(m_eaction); 268 if (skb_at_tc_ingress(skb) != want_ingress && m_mac_header_xmit) { 269 if (!skb_at_tc_ingress(skb)) { 270 /* caught at egress, act ingress: pull mac */ 271 mac_len = skb_network_header(skb) - skb_mac_header(skb); 272 skb_pull_rcsum(skb2, mac_len); 273 } else { 274 /* caught at ingress, act egress: push mac */ 275 skb_push_rcsum(skb2, skb->mac_len); 276 } 277 } 278 279 skb2->skb_iif = skb->dev->ifindex; 280 skb2->dev = dev; 281 282 /* mirror is always swallowed */ 283 if (is_redirect) { 284 skb2->tc_redirected = 1; 285 skb2->tc_from_ingress = skb2->tc_at_ingress; 286 if (skb2->tc_from_ingress) 287 skb2->tstamp = 0; 288 /* let's the caller reinsert the packet, if possible */ 289 if (use_reinsert) { 290 res->ingress = want_ingress; 291 res->qstats = this_cpu_ptr(m->common.cpu_qstats); 292 skb_tc_reinsert(skb, res); 293 __this_cpu_dec(mirred_rec_level); 294 return TC_ACT_CONSUMED; 295 } 296 } 297 298 if (!want_ingress) 299 err = dev_queue_xmit(skb2); 300 else 301 err = netif_receive_skb(skb2); 302 303 if (err) { 304 out: 305 qstats_overlimit_inc(this_cpu_ptr(m->common.cpu_qstats)); 306 if (tcf_mirred_is_act_redirect(m_eaction)) 307 retval = TC_ACT_SHOT; 308 } 309 __this_cpu_dec(mirred_rec_level); 310 311 return retval; 312 } 313 314 static void tcf_stats_update(struct tc_action *a, u64 bytes, u32 packets, 315 u64 lastuse, bool hw) 316 { 317 struct tcf_mirred *m = to_mirred(a); 318 struct tcf_t *tm = &m->tcf_tm; 319 320 _bstats_cpu_update(this_cpu_ptr(a->cpu_bstats), bytes, packets); 321 if (hw) 322 _bstats_cpu_update(this_cpu_ptr(a->cpu_bstats_hw), 323 bytes, packets); 324 tm->lastuse = max_t(u64, tm->lastuse, lastuse); 325 } 326 327 static int tcf_mirred_dump(struct sk_buff *skb, struct tc_action *a, int bind, 328 int ref) 329 { 330 unsigned char *b = skb_tail_pointer(skb); 331 struct tcf_mirred *m = to_mirred(a); 332 struct tc_mirred opt = { 333 .index = m->tcf_index, 334 .refcnt = refcount_read(&m->tcf_refcnt) - ref, 335 .bindcnt = atomic_read(&m->tcf_bindcnt) - bind, 336 }; 337 struct net_device *dev; 338 struct tcf_t t; 339 340 spin_lock_bh(&m->tcf_lock); 341 opt.action = m->tcf_action; 342 opt.eaction = m->tcfm_eaction; 343 dev = tcf_mirred_dev_dereference(m); 344 if (dev) 345 opt.ifindex = dev->ifindex; 346 347 if (nla_put(skb, TCA_MIRRED_PARMS, sizeof(opt), &opt)) 348 goto nla_put_failure; 349 350 tcf_tm_dump(&t, &m->tcf_tm); 351 if (nla_put_64bit(skb, TCA_MIRRED_TM, sizeof(t), &t, TCA_MIRRED_PAD)) 352 goto nla_put_failure; 353 spin_unlock_bh(&m->tcf_lock); 354 355 return skb->len; 356 357 nla_put_failure: 358 spin_unlock_bh(&m->tcf_lock); 359 nlmsg_trim(skb, b); 360 return -1; 361 } 362 363 static int tcf_mirred_walker(struct net *net, struct sk_buff *skb, 364 struct netlink_callback *cb, int type, 365 const struct tc_action_ops *ops, 366 struct netlink_ext_ack *extack) 367 { 368 struct tc_action_net *tn = net_generic(net, mirred_net_id); 369 370 return tcf_generic_walker(tn, skb, cb, type, ops, extack); 371 } 372 373 static int tcf_mirred_search(struct net *net, struct tc_action **a, u32 index) 374 { 375 struct tc_action_net *tn = net_generic(net, mirred_net_id); 376 377 return tcf_idr_search(tn, a, index); 378 } 379 380 static int mirred_device_event(struct notifier_block *unused, 381 unsigned long event, void *ptr) 382 { 383 struct net_device *dev = netdev_notifier_info_to_dev(ptr); 384 struct tcf_mirred *m; 385 386 ASSERT_RTNL(); 387 if (event == NETDEV_UNREGISTER) { 388 spin_lock(&mirred_list_lock); 389 list_for_each_entry(m, &mirred_list, tcfm_list) { 390 spin_lock_bh(&m->tcf_lock); 391 if (tcf_mirred_dev_dereference(m) == dev) { 392 dev_put(dev); 393 /* Note : no rcu grace period necessary, as 394 * net_device are already rcu protected. 395 */ 396 RCU_INIT_POINTER(m->tcfm_dev, NULL); 397 } 398 spin_unlock_bh(&m->tcf_lock); 399 } 400 spin_unlock(&mirred_list_lock); 401 } 402 403 return NOTIFY_DONE; 404 } 405 406 static struct notifier_block mirred_device_notifier = { 407 .notifier_call = mirred_device_event, 408 }; 409 410 static struct net_device *tcf_mirred_get_dev(const struct tc_action *a) 411 { 412 struct tcf_mirred *m = to_mirred(a); 413 struct net_device *dev; 414 415 rcu_read_lock(); 416 dev = rcu_dereference(m->tcfm_dev); 417 if (dev) 418 dev_hold(dev); 419 rcu_read_unlock(); 420 421 return dev; 422 } 423 424 static void tcf_mirred_put_dev(struct net_device *dev) 425 { 426 dev_put(dev); 427 } 428 429 static size_t tcf_mirred_get_fill_size(const struct tc_action *act) 430 { 431 return nla_total_size(sizeof(struct tc_mirred)); 432 } 433 434 static struct tc_action_ops act_mirred_ops = { 435 .kind = "mirred", 436 .id = TCA_ID_MIRRED, 437 .owner = THIS_MODULE, 438 .act = tcf_mirred_act, 439 .stats_update = tcf_stats_update, 440 .dump = tcf_mirred_dump, 441 .cleanup = tcf_mirred_release, 442 .init = tcf_mirred_init, 443 .walk = tcf_mirred_walker, 444 .lookup = tcf_mirred_search, 445 .get_fill_size = tcf_mirred_get_fill_size, 446 .size = sizeof(struct tcf_mirred), 447 .get_dev = tcf_mirred_get_dev, 448 .put_dev = tcf_mirred_put_dev, 449 }; 450 451 static __net_init int mirred_init_net(struct net *net) 452 { 453 struct tc_action_net *tn = net_generic(net, mirred_net_id); 454 455 return tc_action_net_init(tn, &act_mirred_ops); 456 } 457 458 static void __net_exit mirred_exit_net(struct list_head *net_list) 459 { 460 tc_action_net_exit(net_list, mirred_net_id); 461 } 462 463 static struct pernet_operations mirred_net_ops = { 464 .init = mirred_init_net, 465 .exit_batch = mirred_exit_net, 466 .id = &mirred_net_id, 467 .size = sizeof(struct tc_action_net), 468 }; 469 470 MODULE_AUTHOR("Jamal Hadi Salim(2002)"); 471 MODULE_DESCRIPTION("Device Mirror/redirect actions"); 472 MODULE_LICENSE("GPL"); 473 474 static int __init mirred_init_module(void) 475 { 476 int err = register_netdevice_notifier(&mirred_device_notifier); 477 if (err) 478 return err; 479 480 pr_info("Mirror/redirect action on\n"); 481 return tcf_register_action(&act_mirred_ops, &mirred_net_ops); 482 } 483 484 static void __exit mirred_cleanup_module(void) 485 { 486 tcf_unregister_action(&act_mirred_ops, &mirred_net_ops); 487 unregister_netdevice_notifier(&mirred_device_notifier); 488 } 489 490 module_init(mirred_init_module); 491 module_exit(mirred_cleanup_module); 492