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 28static LIST_HEAD(mirred_list); 29static DEFINE_SPINLOCK(mirred_list_lock); 30 31#define MIRRED_NEST_LIMIT 4 32static DEFINE_PER_CPU(unsigned int, mirred_nest_level); 33 34static bool tcf_mirred_is_act_redirect(int action) 35{ 36 return action == TCA_EGRESS_REDIR || action == TCA_INGRESS_REDIR; 37} 38 39static 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 53static 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 65static 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 71static 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 if (dev) 83 dev_put(dev); 84} 85 86static const struct nla_policy mirred_policy[TCA_MIRRED_MAX + 1] = { 87 [TCA_MIRRED_PARMS] = { .len = sizeof(struct tc_mirred) }, 88}; 89 90static unsigned int mirred_net_id; 91static struct tc_action_ops act_mirred_ops; 92 93static int tcf_mirred_init(struct net *net, struct nlattr *nla, 94 struct nlattr *est, struct tc_action **a, 95 int ovr, int bind, bool rtnl_held, 96 struct tcf_proto *tp, 97 u32 flags, struct netlink_ext_ack *extack) 98{ 99 struct tc_action_net *tn = net_generic(net, mirred_net_id); 100 struct nlattr *tb[TCA_MIRRED_MAX + 1]; 101 struct tcf_chain *goto_ch = NULL; 102 bool mac_header_xmit = false; 103 struct tc_mirred *parm; 104 struct tcf_mirred *m; 105 struct net_device *dev; 106 bool exists = false; 107 int ret, err; 108 u32 index; 109 110 if (!nla) { 111 NL_SET_ERR_MSG_MOD(extack, "Mirred requires attributes to be passed"); 112 return -EINVAL; 113 } 114 ret = nla_parse_nested_deprecated(tb, TCA_MIRRED_MAX, nla, 115 mirred_policy, extack); 116 if (ret < 0) 117 return ret; 118 if (!tb[TCA_MIRRED_PARMS]) { 119 NL_SET_ERR_MSG_MOD(extack, "Missing required mirred parameters"); 120 return -EINVAL; 121 } 122 parm = nla_data(tb[TCA_MIRRED_PARMS]); 123 index = parm->index; 124 err = tcf_idr_check_alloc(tn, &index, a, bind); 125 if (err < 0) 126 return err; 127 exists = err; 128 if (exists && bind) 129 return 0; 130 131 switch (parm->eaction) { 132 case TCA_EGRESS_MIRROR: 133 case TCA_EGRESS_REDIR: 134 case TCA_INGRESS_REDIR: 135 case TCA_INGRESS_MIRROR: 136 break; 137 default: 138 if (exists) 139 tcf_idr_release(*a, bind); 140 else 141 tcf_idr_cleanup(tn, index); 142 NL_SET_ERR_MSG_MOD(extack, "Unknown mirred option"); 143 return -EINVAL; 144 } 145 146 if (!exists) { 147 if (!parm->ifindex) { 148 tcf_idr_cleanup(tn, index); 149 NL_SET_ERR_MSG_MOD(extack, "Specified device does not exist"); 150 return -EINVAL; 151 } 152 ret = tcf_idr_create_from_flags(tn, index, est, a, 153 &act_mirred_ops, bind, flags); 154 if (ret) { 155 tcf_idr_cleanup(tn, index); 156 return ret; 157 } 158 ret = ACT_P_CREATED; 159 } else if (!ovr) { 160 tcf_idr_release(*a, bind); 161 return -EEXIST; 162 } 163 164 m = to_mirred(*a); 165 if (ret == ACT_P_CREATED) 166 INIT_LIST_HEAD(&m->tcfm_list); 167 168 err = tcf_action_check_ctrlact(parm->action, tp, &goto_ch, extack); 169 if (err < 0) 170 goto release_idr; 171 172 spin_lock_bh(&m->tcf_lock); 173 174 if (parm->ifindex) { 175 dev = dev_get_by_index(net, parm->ifindex); 176 if (!dev) { 177 spin_unlock_bh(&m->tcf_lock); 178 err = -ENODEV; 179 goto put_chain; 180 } 181 mac_header_xmit = dev_is_mac_header_xmit(dev); 182 dev = rcu_replace_pointer(m->tcfm_dev, dev, 183 lockdep_is_held(&m->tcf_lock)); 184 if (dev) 185 dev_put(dev); 186 m->tcfm_mac_header_xmit = mac_header_xmit; 187 } 188 goto_ch = tcf_action_set_ctrlact(*a, parm->action, goto_ch); 189 m->tcfm_eaction = parm->eaction; 190 spin_unlock_bh(&m->tcf_lock); 191 if (goto_ch) 192 tcf_chain_put_by_act(goto_ch); 193 194 if (ret == ACT_P_CREATED) { 195 spin_lock(&mirred_list_lock); 196 list_add(&m->tcfm_list, &mirred_list); 197 spin_unlock(&mirred_list_lock); 198 } 199 200 return ret; 201put_chain: 202 if (goto_ch) 203 tcf_chain_put_by_act(goto_ch); 204release_idr: 205 tcf_idr_release(*a, bind); 206 return err; 207} 208 209static bool is_mirred_nested(void) 210{ 211 return unlikely(__this_cpu_read(mirred_nest_level) > 1); 212} 213 214static int tcf_mirred_forward(bool want_ingress, struct sk_buff *skb) 215{ 216 int err; 217 218 if (!want_ingress) 219 err = dev_queue_xmit(skb); 220 else if (is_mirred_nested()) 221 err = netif_rx(skb); 222 else 223 err = netif_receive_skb(skb); 224 225 return err; 226} 227 228static int tcf_mirred_act(struct sk_buff *skb, const struct tc_action *a, 229 struct tcf_result *res) 230{ 231 struct tcf_mirred *m = to_mirred(a); 232 struct sk_buff *skb2 = skb; 233 bool m_mac_header_xmit; 234 struct net_device *dev; 235 unsigned int nest_level; 236 int retval, err = 0; 237 bool use_reinsert; 238 bool want_ingress; 239 bool is_redirect; 240 bool expects_nh; 241 bool at_ingress; 242 int m_eaction; 243 int mac_len; 244 bool at_nh; 245 246 nest_level = __this_cpu_inc_return(mirred_nest_level); 247 if (unlikely(nest_level > MIRRED_NEST_LIMIT)) { 248 net_warn_ratelimited("Packet exceeded mirred recursion limit on dev %s\n", 249 netdev_name(skb->dev)); 250 __this_cpu_dec(mirred_nest_level); 251 return TC_ACT_SHOT; 252 } 253 254 tcf_lastuse_update(&m->tcf_tm); 255 tcf_action_update_bstats(&m->common, skb); 256 257 m_mac_header_xmit = READ_ONCE(m->tcfm_mac_header_xmit); 258 m_eaction = READ_ONCE(m->tcfm_eaction); 259 retval = READ_ONCE(m->tcf_action); 260 dev = rcu_dereference_bh(m->tcfm_dev); 261 if (unlikely(!dev)) { 262 pr_notice_once("tc mirred: target device is gone\n"); 263 goto out; 264 } 265 266 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 out; 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 is_redirect = tcf_mirred_is_act_redirect(m_eaction); 277 at_ingress = skb_at_tc_ingress(skb); 278 use_reinsert = at_ingress && is_redirect && 279 tcf_mirred_can_reinsert(retval); 280 if (!use_reinsert) { 281 skb2 = skb_clone(skb, GFP_ATOMIC); 282 if (!skb2) 283 goto out; 284 } 285 286 want_ingress = tcf_mirred_act_wants_ingress(m_eaction); 287 288 /* All mirred/redirected skbs should clear previous ct info */ 289 nf_reset_ct(skb2); 290 if (want_ingress && !at_ingress) /* drop dst for egress -> ingress */ 291 skb_dst_drop(skb2); 292 293 expects_nh = want_ingress || !m_mac_header_xmit; 294 at_nh = skb->data == skb_network_header(skb); 295 if (at_nh != expects_nh) { 296 mac_len = skb_at_tc_ingress(skb) ? skb->mac_len : 297 skb_network_header(skb) - skb_mac_header(skb); 298 if (expects_nh) { 299 /* target device/action expect data at nh */ 300 skb_pull_rcsum(skb2, mac_len); 301 } else { 302 /* target device/action expect data at mac */ 303 skb_push_rcsum(skb2, mac_len); 304 } 305 } 306 307 skb2->skb_iif = skb->dev->ifindex; 308 skb2->dev = dev; 309 310 /* mirror is always swallowed */ 311 if (is_redirect) { 312 skb_set_redirected(skb2, skb2->tc_at_ingress); 313 314 /* let's the caller reinsert the packet, if possible */ 315 if (use_reinsert) { 316 res->ingress = want_ingress; 317 err = tcf_mirred_forward(res->ingress, skb); 318 if (err) 319 tcf_action_inc_overlimit_qstats(&m->common); 320 __this_cpu_dec(mirred_nest_level); 321 return TC_ACT_CONSUMED; 322 } 323 } 324 325 err = tcf_mirred_forward(want_ingress, skb2); 326 if (err) { 327out: 328 tcf_action_inc_overlimit_qstats(&m->common); 329 if (tcf_mirred_is_act_redirect(m_eaction)) 330 retval = TC_ACT_SHOT; 331 } 332 __this_cpu_dec(mirred_nest_level); 333 334 return retval; 335} 336 337static void tcf_stats_update(struct tc_action *a, u64 bytes, u64 packets, 338 u64 drops, u64 lastuse, bool hw) 339{ 340 struct tcf_mirred *m = to_mirred(a); 341 struct tcf_t *tm = &m->tcf_tm; 342 343 tcf_action_update_stats(a, bytes, packets, drops, hw); 344 tm->lastuse = max_t(u64, tm->lastuse, lastuse); 345} 346 347static int tcf_mirred_dump(struct sk_buff *skb, struct tc_action *a, int bind, 348 int ref) 349{ 350 unsigned char *b = skb_tail_pointer(skb); 351 struct tcf_mirred *m = to_mirred(a); 352 struct tc_mirred opt = { 353 .index = m->tcf_index, 354 .refcnt = refcount_read(&m->tcf_refcnt) - ref, 355 .bindcnt = atomic_read(&m->tcf_bindcnt) - bind, 356 }; 357 struct net_device *dev; 358 struct tcf_t t; 359 360 spin_lock_bh(&m->tcf_lock); 361 opt.action = m->tcf_action; 362 opt.eaction = m->tcfm_eaction; 363 dev = tcf_mirred_dev_dereference(m); 364 if (dev) 365 opt.ifindex = dev->ifindex; 366 367 if (nla_put(skb, TCA_MIRRED_PARMS, sizeof(opt), &opt)) 368 goto nla_put_failure; 369 370 tcf_tm_dump(&t, &m->tcf_tm); 371 if (nla_put_64bit(skb, TCA_MIRRED_TM, sizeof(t), &t, TCA_MIRRED_PAD)) 372 goto nla_put_failure; 373 spin_unlock_bh(&m->tcf_lock); 374 375 return skb->len; 376 377nla_put_failure: 378 spin_unlock_bh(&m->tcf_lock); 379 nlmsg_trim(skb, b); 380 return -1; 381} 382 383static int tcf_mirred_walker(struct net *net, struct sk_buff *skb, 384 struct netlink_callback *cb, int type, 385 const struct tc_action_ops *ops, 386 struct netlink_ext_ack *extack) 387{ 388 struct tc_action_net *tn = net_generic(net, mirred_net_id); 389 390 return tcf_generic_walker(tn, skb, cb, type, ops, extack); 391} 392 393static int tcf_mirred_search(struct net *net, struct tc_action **a, u32 index) 394{ 395 struct tc_action_net *tn = net_generic(net, mirred_net_id); 396 397 return tcf_idr_search(tn, a, index); 398} 399 400static int mirred_device_event(struct notifier_block *unused, 401 unsigned long event, void *ptr) 402{ 403 struct net_device *dev = netdev_notifier_info_to_dev(ptr); 404 struct tcf_mirred *m; 405 406 ASSERT_RTNL(); 407 if (event == NETDEV_UNREGISTER) { 408 spin_lock(&mirred_list_lock); 409 list_for_each_entry(m, &mirred_list, tcfm_list) { 410 spin_lock_bh(&m->tcf_lock); 411 if (tcf_mirred_dev_dereference(m) == dev) { 412 dev_put(dev); 413 /* Note : no rcu grace period necessary, as 414 * net_device are already rcu protected. 415 */ 416 RCU_INIT_POINTER(m->tcfm_dev, NULL); 417 } 418 spin_unlock_bh(&m->tcf_lock); 419 } 420 spin_unlock(&mirred_list_lock); 421 } 422 423 return NOTIFY_DONE; 424} 425 426static struct notifier_block mirred_device_notifier = { 427 .notifier_call = mirred_device_event, 428}; 429 430static void tcf_mirred_dev_put(void *priv) 431{ 432 struct net_device *dev = priv; 433 434 dev_put(dev); 435} 436 437static struct net_device * 438tcf_mirred_get_dev(const struct tc_action *a, 439 tc_action_priv_destructor *destructor) 440{ 441 struct tcf_mirred *m = to_mirred(a); 442 struct net_device *dev; 443 444 rcu_read_lock(); 445 dev = rcu_dereference(m->tcfm_dev); 446 if (dev) { 447 dev_hold(dev); 448 *destructor = tcf_mirred_dev_put; 449 } 450 rcu_read_unlock(); 451 452 return dev; 453} 454 455static size_t tcf_mirred_get_fill_size(const struct tc_action *act) 456{ 457 return nla_total_size(sizeof(struct tc_mirred)); 458} 459 460static struct tc_action_ops act_mirred_ops = { 461 .kind = "mirred", 462 .id = TCA_ID_MIRRED, 463 .owner = THIS_MODULE, 464 .act = tcf_mirred_act, 465 .stats_update = tcf_stats_update, 466 .dump = tcf_mirred_dump, 467 .cleanup = tcf_mirred_release, 468 .init = tcf_mirred_init, 469 .walk = tcf_mirred_walker, 470 .lookup = tcf_mirred_search, 471 .get_fill_size = tcf_mirred_get_fill_size, 472 .size = sizeof(struct tcf_mirred), 473 .get_dev = tcf_mirred_get_dev, 474}; 475 476static __net_init int mirred_init_net(struct net *net) 477{ 478 struct tc_action_net *tn = net_generic(net, mirred_net_id); 479 480 return tc_action_net_init(net, tn, &act_mirred_ops); 481} 482 483static void __net_exit mirred_exit_net(struct list_head *net_list) 484{ 485 tc_action_net_exit(net_list, mirred_net_id); 486} 487 488static struct pernet_operations mirred_net_ops = { 489 .init = mirred_init_net, 490 .exit_batch = mirred_exit_net, 491 .id = &mirred_net_id, 492 .size = sizeof(struct tc_action_net), 493}; 494 495MODULE_AUTHOR("Jamal Hadi Salim(2002)"); 496MODULE_DESCRIPTION("Device Mirror/redirect actions"); 497MODULE_LICENSE("GPL"); 498 499static int __init mirred_init_module(void) 500{ 501 int err = register_netdevice_notifier(&mirred_device_notifier); 502 if (err) 503 return err; 504 505 pr_info("Mirror/redirect action on\n"); 506 err = tcf_register_action(&act_mirred_ops, &mirred_net_ops); 507 if (err) 508 unregister_netdevice_notifier(&mirred_device_notifier); 509 510 return err; 511} 512 513static void __exit mirred_cleanup_module(void) 514{ 515 tcf_unregister_action(&act_mirred_ops, &mirred_net_ops); 516 unregister_netdevice_notifier(&mirred_device_notifier); 517} 518 519module_init(mirred_init_module); 520module_exit(mirred_cleanup_module); 521