1 /* Netfilter messages via netlink socket. Allows for user space 2 * protocol helpers and general trouble making from userspace. 3 * 4 * (C) 2001 by Jay Schulist <jschlst@samba.org>, 5 * (C) 2002-2005 by Harald Welte <laforge@gnumonks.org> 6 * (C) 2005-2017 by Pablo Neira Ayuso <pablo@netfilter.org> 7 * 8 * Initial netfilter messages via netlink development funded and 9 * generally made possible by Network Robots, Inc. (www.networkrobots.com) 10 * 11 * Further development of this code funded by Astaro AG (http://www.astaro.com) 12 * 13 * This software may be used and distributed according to the terms 14 * of the GNU General Public License, incorporated herein by reference. 15 */ 16 17 #include <linux/module.h> 18 #include <linux/types.h> 19 #include <linux/socket.h> 20 #include <linux/kernel.h> 21 #include <linux/string.h> 22 #include <linux/sockios.h> 23 #include <linux/net.h> 24 #include <linux/skbuff.h> 25 #include <linux/uaccess.h> 26 #include <net/sock.h> 27 #include <linux/init.h> 28 #include <linux/sched/signal.h> 29 30 #include <net/netlink.h> 31 #include <net/netns/generic.h> 32 #include <linux/netfilter.h> 33 #include <linux/netfilter/nfnetlink.h> 34 35 MODULE_LICENSE("GPL"); 36 MODULE_AUTHOR("Harald Welte <laforge@netfilter.org>"); 37 MODULE_ALIAS_NET_PF_PROTO(PF_NETLINK, NETLINK_NETFILTER); 38 MODULE_DESCRIPTION("Netfilter messages via netlink socket"); 39 40 #define nfnl_dereference_protected(id) \ 41 rcu_dereference_protected(table[(id)].subsys, \ 42 lockdep_nfnl_is_held((id))) 43 44 #define NFNL_MAX_ATTR_COUNT 32 45 46 static unsigned int nfnetlink_pernet_id __read_mostly; 47 48 #ifdef CONFIG_NF_CONNTRACK_EVENTS 49 static DEFINE_SPINLOCK(nfnl_grp_active_lock); 50 #endif 51 52 struct nfnl_net { 53 struct sock *nfnl; 54 }; 55 56 static struct { 57 struct mutex mutex; 58 const struct nfnetlink_subsystem __rcu *subsys; 59 } table[NFNL_SUBSYS_COUNT]; 60 61 static struct lock_class_key nfnl_lockdep_keys[NFNL_SUBSYS_COUNT]; 62 63 static const char *const nfnl_lockdep_names[NFNL_SUBSYS_COUNT] = { 64 [NFNL_SUBSYS_NONE] = "nfnl_subsys_none", 65 [NFNL_SUBSYS_CTNETLINK] = "nfnl_subsys_ctnetlink", 66 [NFNL_SUBSYS_CTNETLINK_EXP] = "nfnl_subsys_ctnetlink_exp", 67 [NFNL_SUBSYS_QUEUE] = "nfnl_subsys_queue", 68 [NFNL_SUBSYS_ULOG] = "nfnl_subsys_ulog", 69 [NFNL_SUBSYS_OSF] = "nfnl_subsys_osf", 70 [NFNL_SUBSYS_IPSET] = "nfnl_subsys_ipset", 71 [NFNL_SUBSYS_ACCT] = "nfnl_subsys_acct", 72 [NFNL_SUBSYS_CTNETLINK_TIMEOUT] = "nfnl_subsys_cttimeout", 73 [NFNL_SUBSYS_CTHELPER] = "nfnl_subsys_cthelper", 74 [NFNL_SUBSYS_NFTABLES] = "nfnl_subsys_nftables", 75 [NFNL_SUBSYS_NFT_COMPAT] = "nfnl_subsys_nftcompat", 76 [NFNL_SUBSYS_HOOK] = "nfnl_subsys_hook", 77 }; 78 79 static const int nfnl_group2type[NFNLGRP_MAX+1] = { 80 [NFNLGRP_CONNTRACK_NEW] = NFNL_SUBSYS_CTNETLINK, 81 [NFNLGRP_CONNTRACK_UPDATE] = NFNL_SUBSYS_CTNETLINK, 82 [NFNLGRP_CONNTRACK_DESTROY] = NFNL_SUBSYS_CTNETLINK, 83 [NFNLGRP_CONNTRACK_EXP_NEW] = NFNL_SUBSYS_CTNETLINK_EXP, 84 [NFNLGRP_CONNTRACK_EXP_UPDATE] = NFNL_SUBSYS_CTNETLINK_EXP, 85 [NFNLGRP_CONNTRACK_EXP_DESTROY] = NFNL_SUBSYS_CTNETLINK_EXP, 86 [NFNLGRP_NFTABLES] = NFNL_SUBSYS_NFTABLES, 87 [NFNLGRP_ACCT_QUOTA] = NFNL_SUBSYS_ACCT, 88 [NFNLGRP_NFTRACE] = NFNL_SUBSYS_NFTABLES, 89 [NFNLGRP_NFT_DEV] = NFNL_SUBSYS_NFTABLES, 90 }; 91 92 static struct nfnl_net *nfnl_pernet(struct net *net) 93 { 94 return net_generic(net, nfnetlink_pernet_id); 95 } 96 97 void nfnl_lock(__u8 subsys_id) 98 { 99 mutex_lock(&table[subsys_id].mutex); 100 } 101 EXPORT_SYMBOL_GPL(nfnl_lock); 102 103 void nfnl_unlock(__u8 subsys_id) 104 { 105 mutex_unlock(&table[subsys_id].mutex); 106 } 107 EXPORT_SYMBOL_GPL(nfnl_unlock); 108 109 #ifdef CONFIG_PROVE_LOCKING 110 bool lockdep_nfnl_is_held(u8 subsys_id) 111 { 112 return lockdep_is_held(&table[subsys_id].mutex); 113 } 114 EXPORT_SYMBOL_GPL(lockdep_nfnl_is_held); 115 #endif 116 117 int nfnetlink_subsys_register(const struct nfnetlink_subsystem *n) 118 { 119 u8 cb_id; 120 121 /* Sanity-check attr_count size to avoid stack buffer overflow. */ 122 for (cb_id = 0; cb_id < n->cb_count; cb_id++) 123 if (WARN_ON(n->cb[cb_id].attr_count > NFNL_MAX_ATTR_COUNT)) 124 return -EINVAL; 125 126 nfnl_lock(n->subsys_id); 127 if (table[n->subsys_id].subsys) { 128 nfnl_unlock(n->subsys_id); 129 return -EBUSY; 130 } 131 rcu_assign_pointer(table[n->subsys_id].subsys, n); 132 nfnl_unlock(n->subsys_id); 133 134 return 0; 135 } 136 EXPORT_SYMBOL_GPL(nfnetlink_subsys_register); 137 138 int nfnetlink_subsys_unregister(const struct nfnetlink_subsystem *n) 139 { 140 nfnl_lock(n->subsys_id); 141 table[n->subsys_id].subsys = NULL; 142 nfnl_unlock(n->subsys_id); 143 synchronize_rcu(); 144 return 0; 145 } 146 EXPORT_SYMBOL_GPL(nfnetlink_subsys_unregister); 147 148 static inline const struct nfnetlink_subsystem *nfnetlink_get_subsys(u16 type) 149 { 150 u8 subsys_id = NFNL_SUBSYS_ID(type); 151 152 if (subsys_id >= NFNL_SUBSYS_COUNT) 153 return NULL; 154 155 return rcu_dereference(table[subsys_id].subsys); 156 } 157 158 static inline const struct nfnl_callback * 159 nfnetlink_find_client(u16 type, const struct nfnetlink_subsystem *ss) 160 { 161 u8 cb_id = NFNL_MSG_TYPE(type); 162 163 if (cb_id >= ss->cb_count) 164 return NULL; 165 166 return &ss->cb[cb_id]; 167 } 168 169 int nfnetlink_has_listeners(struct net *net, unsigned int group) 170 { 171 struct nfnl_net *nfnlnet = nfnl_pernet(net); 172 173 return netlink_has_listeners(nfnlnet->nfnl, group); 174 } 175 EXPORT_SYMBOL_GPL(nfnetlink_has_listeners); 176 177 int nfnetlink_send(struct sk_buff *skb, struct net *net, u32 portid, 178 unsigned int group, int echo, gfp_t flags) 179 { 180 struct nfnl_net *nfnlnet = nfnl_pernet(net); 181 182 return nlmsg_notify(nfnlnet->nfnl, skb, portid, group, echo, flags); 183 } 184 EXPORT_SYMBOL_GPL(nfnetlink_send); 185 186 int nfnetlink_set_err(struct net *net, u32 portid, u32 group, int error) 187 { 188 struct nfnl_net *nfnlnet = nfnl_pernet(net); 189 190 return netlink_set_err(nfnlnet->nfnl, portid, group, error); 191 } 192 EXPORT_SYMBOL_GPL(nfnetlink_set_err); 193 194 int nfnetlink_unicast(struct sk_buff *skb, struct net *net, u32 portid) 195 { 196 struct nfnl_net *nfnlnet = nfnl_pernet(net); 197 int err; 198 199 err = nlmsg_unicast(nfnlnet->nfnl, skb, portid); 200 if (err == -EAGAIN) 201 err = -ENOBUFS; 202 203 return err; 204 } 205 EXPORT_SYMBOL_GPL(nfnetlink_unicast); 206 207 void nfnetlink_broadcast(struct net *net, struct sk_buff *skb, __u32 portid, 208 __u32 group, gfp_t allocation) 209 { 210 struct nfnl_net *nfnlnet = nfnl_pernet(net); 211 212 netlink_broadcast(nfnlnet->nfnl, skb, portid, group, allocation); 213 } 214 EXPORT_SYMBOL_GPL(nfnetlink_broadcast); 215 216 /* Process one complete nfnetlink message. */ 217 static int nfnetlink_rcv_msg(struct sk_buff *skb, struct nlmsghdr *nlh, 218 struct netlink_ext_ack *extack) 219 { 220 struct net *net = sock_net(skb->sk); 221 const struct nfnl_callback *nc; 222 const struct nfnetlink_subsystem *ss; 223 int type, err; 224 225 /* All the messages must at least contain nfgenmsg */ 226 if (nlmsg_len(nlh) < sizeof(struct nfgenmsg)) 227 return 0; 228 229 type = nlh->nlmsg_type; 230 replay: 231 rcu_read_lock(); 232 233 ss = nfnetlink_get_subsys(type); 234 if (!ss) { 235 #ifdef CONFIG_MODULES 236 rcu_read_unlock(); 237 request_module("nfnetlink-subsys-%d", NFNL_SUBSYS_ID(type)); 238 rcu_read_lock(); 239 ss = nfnetlink_get_subsys(type); 240 if (!ss) 241 #endif 242 { 243 rcu_read_unlock(); 244 return -EINVAL; 245 } 246 } 247 248 nc = nfnetlink_find_client(type, ss); 249 if (!nc) { 250 rcu_read_unlock(); 251 return -EINVAL; 252 } 253 254 { 255 int min_len = nlmsg_total_size(sizeof(struct nfgenmsg)); 256 struct nfnl_net *nfnlnet = nfnl_pernet(net); 257 u8 cb_id = NFNL_MSG_TYPE(nlh->nlmsg_type); 258 struct nlattr *cda[NFNL_MAX_ATTR_COUNT + 1]; 259 struct nlattr *attr = (void *)nlh + min_len; 260 int attrlen = nlh->nlmsg_len - min_len; 261 __u8 subsys_id = NFNL_SUBSYS_ID(type); 262 struct nfnl_info info = { 263 .net = net, 264 .sk = nfnlnet->nfnl, 265 .nlh = nlh, 266 .nfmsg = nlmsg_data(nlh), 267 .extack = extack, 268 }; 269 270 /* Sanity-check NFNL_MAX_ATTR_COUNT */ 271 if (ss->cb[cb_id].attr_count > NFNL_MAX_ATTR_COUNT) { 272 rcu_read_unlock(); 273 return -ENOMEM; 274 } 275 276 err = nla_parse_deprecated(cda, ss->cb[cb_id].attr_count, 277 attr, attrlen, 278 ss->cb[cb_id].policy, extack); 279 if (err < 0) { 280 rcu_read_unlock(); 281 return err; 282 } 283 284 if (!nc->call) { 285 rcu_read_unlock(); 286 return -EINVAL; 287 } 288 289 switch (nc->type) { 290 case NFNL_CB_RCU: 291 err = nc->call(skb, &info, (const struct nlattr **)cda); 292 rcu_read_unlock(); 293 break; 294 case NFNL_CB_MUTEX: 295 rcu_read_unlock(); 296 nfnl_lock(subsys_id); 297 if (nfnl_dereference_protected(subsys_id) != ss || 298 nfnetlink_find_client(type, ss) != nc) { 299 nfnl_unlock(subsys_id); 300 err = -EAGAIN; 301 break; 302 } 303 err = nc->call(skb, &info, (const struct nlattr **)cda); 304 nfnl_unlock(subsys_id); 305 break; 306 default: 307 rcu_read_unlock(); 308 err = -EINVAL; 309 break; 310 } 311 if (err == -EAGAIN) 312 goto replay; 313 return err; 314 } 315 } 316 317 struct nfnl_err { 318 struct list_head head; 319 struct nlmsghdr *nlh; 320 int err; 321 struct netlink_ext_ack extack; 322 }; 323 324 static int nfnl_err_add(struct list_head *list, struct nlmsghdr *nlh, int err, 325 const struct netlink_ext_ack *extack) 326 { 327 struct nfnl_err *nfnl_err; 328 329 nfnl_err = kmalloc(sizeof(struct nfnl_err), GFP_KERNEL); 330 if (nfnl_err == NULL) 331 return -ENOMEM; 332 333 nfnl_err->nlh = nlh; 334 nfnl_err->err = err; 335 nfnl_err->extack = *extack; 336 list_add_tail(&nfnl_err->head, list); 337 338 return 0; 339 } 340 341 static void nfnl_err_del(struct nfnl_err *nfnl_err) 342 { 343 list_del(&nfnl_err->head); 344 kfree(nfnl_err); 345 } 346 347 static void nfnl_err_reset(struct list_head *err_list) 348 { 349 struct nfnl_err *nfnl_err, *next; 350 351 list_for_each_entry_safe(nfnl_err, next, err_list, head) 352 nfnl_err_del(nfnl_err); 353 } 354 355 static void nfnl_err_deliver(struct list_head *err_list, struct sk_buff *skb) 356 { 357 struct nfnl_err *nfnl_err, *next; 358 359 list_for_each_entry_safe(nfnl_err, next, err_list, head) { 360 netlink_ack(skb, nfnl_err->nlh, nfnl_err->err, 361 &nfnl_err->extack); 362 nfnl_err_del(nfnl_err); 363 } 364 } 365 366 enum { 367 NFNL_BATCH_FAILURE = (1 << 0), 368 NFNL_BATCH_DONE = (1 << 1), 369 NFNL_BATCH_REPLAY = (1 << 2), 370 }; 371 372 static void nfnetlink_rcv_batch(struct sk_buff *skb, struct nlmsghdr *nlh, 373 u16 subsys_id, u32 genid) 374 { 375 struct sk_buff *oskb = skb; 376 struct net *net = sock_net(skb->sk); 377 const struct nfnetlink_subsystem *ss; 378 const struct nfnl_callback *nc; 379 struct netlink_ext_ack extack; 380 LIST_HEAD(err_list); 381 u32 status; 382 int err; 383 384 if (subsys_id >= NFNL_SUBSYS_COUNT) 385 return netlink_ack(skb, nlh, -EINVAL, NULL); 386 replay: 387 status = 0; 388 replay_abort: 389 skb = netlink_skb_clone(oskb, GFP_KERNEL); 390 if (!skb) 391 return netlink_ack(oskb, nlh, -ENOMEM, NULL); 392 393 nfnl_lock(subsys_id); 394 ss = nfnl_dereference_protected(subsys_id); 395 if (!ss) { 396 #ifdef CONFIG_MODULES 397 nfnl_unlock(subsys_id); 398 request_module("nfnetlink-subsys-%d", subsys_id); 399 nfnl_lock(subsys_id); 400 ss = nfnl_dereference_protected(subsys_id); 401 if (!ss) 402 #endif 403 { 404 nfnl_unlock(subsys_id); 405 netlink_ack(oskb, nlh, -EOPNOTSUPP, NULL); 406 return consume_skb(skb); 407 } 408 } 409 410 if (!ss->valid_genid || !ss->commit || !ss->abort) { 411 nfnl_unlock(subsys_id); 412 netlink_ack(oskb, nlh, -EOPNOTSUPP, NULL); 413 return consume_skb(skb); 414 } 415 416 if (!try_module_get(ss->owner)) { 417 nfnl_unlock(subsys_id); 418 netlink_ack(oskb, nlh, -EOPNOTSUPP, NULL); 419 return consume_skb(skb); 420 } 421 422 if (!ss->valid_genid(net, genid)) { 423 module_put(ss->owner); 424 nfnl_unlock(subsys_id); 425 netlink_ack(oskb, nlh, -ERESTART, NULL); 426 return consume_skb(skb); 427 } 428 429 nfnl_unlock(subsys_id); 430 431 if (nlh->nlmsg_flags & NLM_F_ACK) { 432 memset(&extack, 0, sizeof(extack)); 433 nfnl_err_add(&err_list, nlh, 0, &extack); 434 } 435 436 while (skb->len >= nlmsg_total_size(0)) { 437 int msglen, type; 438 439 if (fatal_signal_pending(current)) { 440 nfnl_err_reset(&err_list); 441 err = -EINTR; 442 status = NFNL_BATCH_FAILURE; 443 goto done; 444 } 445 446 memset(&extack, 0, sizeof(extack)); 447 nlh = nlmsg_hdr(skb); 448 err = 0; 449 450 if (nlh->nlmsg_len < NLMSG_HDRLEN || 451 skb->len < nlh->nlmsg_len || 452 nlmsg_len(nlh) < sizeof(struct nfgenmsg)) { 453 nfnl_err_reset(&err_list); 454 status |= NFNL_BATCH_FAILURE; 455 goto done; 456 } 457 458 /* Only requests are handled by the kernel */ 459 if (!(nlh->nlmsg_flags & NLM_F_REQUEST)) { 460 err = -EINVAL; 461 goto ack; 462 } 463 464 type = nlh->nlmsg_type; 465 if (type == NFNL_MSG_BATCH_BEGIN) { 466 /* Malformed: Batch begin twice */ 467 nfnl_err_reset(&err_list); 468 status |= NFNL_BATCH_FAILURE; 469 goto done; 470 } else if (type == NFNL_MSG_BATCH_END) { 471 status |= NFNL_BATCH_DONE; 472 goto done; 473 } else if (type < NLMSG_MIN_TYPE) { 474 err = -EINVAL; 475 goto ack; 476 } 477 478 /* We only accept a batch with messages for the same 479 * subsystem. 480 */ 481 if (NFNL_SUBSYS_ID(type) != subsys_id) { 482 err = -EINVAL; 483 goto ack; 484 } 485 486 nc = nfnetlink_find_client(type, ss); 487 if (!nc) { 488 err = -EINVAL; 489 goto ack; 490 } 491 492 if (nc->type != NFNL_CB_BATCH) { 493 err = -EINVAL; 494 goto ack; 495 } 496 497 { 498 int min_len = nlmsg_total_size(sizeof(struct nfgenmsg)); 499 struct nfnl_net *nfnlnet = nfnl_pernet(net); 500 struct nlattr *cda[NFNL_MAX_ATTR_COUNT + 1]; 501 struct nlattr *attr = (void *)nlh + min_len; 502 u8 cb_id = NFNL_MSG_TYPE(nlh->nlmsg_type); 503 int attrlen = nlh->nlmsg_len - min_len; 504 struct nfnl_info info = { 505 .net = net, 506 .sk = nfnlnet->nfnl, 507 .nlh = nlh, 508 .nfmsg = nlmsg_data(nlh), 509 .extack = &extack, 510 }; 511 512 /* Sanity-check NFTA_MAX_ATTR */ 513 if (ss->cb[cb_id].attr_count > NFNL_MAX_ATTR_COUNT) { 514 err = -ENOMEM; 515 goto ack; 516 } 517 518 err = nla_parse_deprecated(cda, 519 ss->cb[cb_id].attr_count, 520 attr, attrlen, 521 ss->cb[cb_id].policy, &extack); 522 if (err < 0) 523 goto ack; 524 525 err = nc->call(skb, &info, (const struct nlattr **)cda); 526 527 /* The lock was released to autoload some module, we 528 * have to abort and start from scratch using the 529 * original skb. 530 */ 531 if (err == -EAGAIN) { 532 status |= NFNL_BATCH_REPLAY; 533 goto done; 534 } 535 } 536 ack: 537 if (nlh->nlmsg_flags & NLM_F_ACK || err) { 538 /* Errors are delivered once the full batch has been 539 * processed, this avoids that the same error is 540 * reported several times when replaying the batch. 541 */ 542 if (err == -ENOMEM || 543 nfnl_err_add(&err_list, nlh, err, &extack) < 0) { 544 /* We failed to enqueue an error, reset the 545 * list of errors and send OOM to userspace 546 * pointing to the batch header. 547 */ 548 nfnl_err_reset(&err_list); 549 netlink_ack(oskb, nlmsg_hdr(oskb), -ENOMEM, 550 NULL); 551 status |= NFNL_BATCH_FAILURE; 552 goto done; 553 } 554 /* We don't stop processing the batch on errors, thus, 555 * userspace gets all the errors that the batch 556 * triggers. 557 */ 558 if (err) 559 status |= NFNL_BATCH_FAILURE; 560 } 561 562 msglen = NLMSG_ALIGN(nlh->nlmsg_len); 563 if (msglen > skb->len) 564 msglen = skb->len; 565 skb_pull(skb, msglen); 566 } 567 done: 568 if (status & NFNL_BATCH_REPLAY) { 569 ss->abort(net, oskb, NFNL_ABORT_AUTOLOAD); 570 nfnl_err_reset(&err_list); 571 consume_skb(skb); 572 module_put(ss->owner); 573 goto replay; 574 } else if (status == NFNL_BATCH_DONE) { 575 err = ss->commit(net, oskb); 576 if (err == -EAGAIN) { 577 status |= NFNL_BATCH_REPLAY; 578 goto done; 579 } else if (err) { 580 ss->abort(net, oskb, NFNL_ABORT_NONE); 581 netlink_ack(oskb, nlmsg_hdr(oskb), err, NULL); 582 } else if (nlh->nlmsg_flags & NLM_F_ACK) { 583 memset(&extack, 0, sizeof(extack)); 584 nfnl_err_add(&err_list, nlh, 0, &extack); 585 } 586 } else { 587 enum nfnl_abort_action abort_action; 588 589 if (status & NFNL_BATCH_FAILURE) 590 abort_action = NFNL_ABORT_NONE; 591 else 592 abort_action = NFNL_ABORT_VALIDATE; 593 594 err = ss->abort(net, oskb, abort_action); 595 if (err == -EAGAIN) { 596 nfnl_err_reset(&err_list); 597 consume_skb(skb); 598 module_put(ss->owner); 599 status |= NFNL_BATCH_FAILURE; 600 goto replay_abort; 601 } 602 } 603 604 nfnl_err_deliver(&err_list, oskb); 605 consume_skb(skb); 606 module_put(ss->owner); 607 } 608 609 static const struct nla_policy nfnl_batch_policy[NFNL_BATCH_MAX + 1] = { 610 [NFNL_BATCH_GENID] = { .type = NLA_U32 }, 611 }; 612 613 static void nfnetlink_rcv_skb_batch(struct sk_buff *skb, struct nlmsghdr *nlh) 614 { 615 int min_len = nlmsg_total_size(sizeof(struct nfgenmsg)); 616 struct nlattr *attr = (void *)nlh + min_len; 617 struct nlattr *cda[NFNL_BATCH_MAX + 1]; 618 int attrlen = nlh->nlmsg_len - min_len; 619 struct nfgenmsg *nfgenmsg; 620 int msglen, err; 621 u32 gen_id = 0; 622 u16 res_id; 623 624 msglen = NLMSG_ALIGN(nlh->nlmsg_len); 625 if (msglen > skb->len) 626 msglen = skb->len; 627 628 if (skb->len < NLMSG_HDRLEN + sizeof(struct nfgenmsg)) 629 return; 630 631 err = nla_parse_deprecated(cda, NFNL_BATCH_MAX, attr, attrlen, 632 nfnl_batch_policy, NULL); 633 if (err < 0) { 634 netlink_ack(skb, nlh, err, NULL); 635 return; 636 } 637 if (cda[NFNL_BATCH_GENID]) 638 gen_id = ntohl(nla_get_be32(cda[NFNL_BATCH_GENID])); 639 640 nfgenmsg = nlmsg_data(nlh); 641 skb_pull(skb, msglen); 642 /* Work around old nft using host byte order */ 643 if (nfgenmsg->res_id == (__force __be16)NFNL_SUBSYS_NFTABLES) 644 res_id = NFNL_SUBSYS_NFTABLES; 645 else 646 res_id = ntohs(nfgenmsg->res_id); 647 648 nfnetlink_rcv_batch(skb, nlh, res_id, gen_id); 649 } 650 651 static void nfnetlink_rcv(struct sk_buff *skb) 652 { 653 struct nlmsghdr *nlh = nlmsg_hdr(skb); 654 655 if (skb->len < NLMSG_HDRLEN || 656 nlh->nlmsg_len < NLMSG_HDRLEN || 657 skb->len < nlh->nlmsg_len) 658 return; 659 660 if (!netlink_net_capable(skb, CAP_NET_ADMIN)) { 661 netlink_ack(skb, nlh, -EPERM, NULL); 662 return; 663 } 664 665 if (nlh->nlmsg_type == NFNL_MSG_BATCH_BEGIN) 666 nfnetlink_rcv_skb_batch(skb, nlh); 667 else 668 netlink_rcv_skb(skb, nfnetlink_rcv_msg); 669 } 670 671 static void nfnetlink_bind_event(struct net *net, unsigned int group) 672 { 673 #ifdef CONFIG_NF_CONNTRACK_EVENTS 674 int type, group_bit; 675 u8 v; 676 677 /* All NFNLGRP_CONNTRACK_* group bits fit into u8. 678 * The other groups are not relevant and can be ignored. 679 */ 680 if (group >= 8) 681 return; 682 683 type = nfnl_group2type[group]; 684 685 switch (type) { 686 case NFNL_SUBSYS_CTNETLINK: 687 break; 688 case NFNL_SUBSYS_CTNETLINK_EXP: 689 break; 690 default: 691 return; 692 } 693 694 group_bit = (1 << group); 695 696 spin_lock(&nfnl_grp_active_lock); 697 v = READ_ONCE(nf_ctnetlink_has_listener); 698 if ((v & group_bit) == 0) { 699 v |= group_bit; 700 701 /* read concurrently without nfnl_grp_active_lock held. */ 702 WRITE_ONCE(nf_ctnetlink_has_listener, v); 703 } 704 705 spin_unlock(&nfnl_grp_active_lock); 706 #endif 707 } 708 709 static int nfnetlink_bind(struct net *net, int group) 710 { 711 const struct nfnetlink_subsystem *ss; 712 int type; 713 714 if (group <= NFNLGRP_NONE || group > NFNLGRP_MAX) 715 return 0; 716 717 type = nfnl_group2type[group]; 718 719 rcu_read_lock(); 720 ss = nfnetlink_get_subsys(type << 8); 721 rcu_read_unlock(); 722 if (!ss) 723 request_module_nowait("nfnetlink-subsys-%d", type); 724 725 nfnetlink_bind_event(net, group); 726 return 0; 727 } 728 729 static void nfnetlink_unbind(struct net *net, int group) 730 { 731 #ifdef CONFIG_NF_CONNTRACK_EVENTS 732 int type, group_bit; 733 734 if (group <= NFNLGRP_NONE || group > NFNLGRP_MAX) 735 return; 736 737 type = nfnl_group2type[group]; 738 739 switch (type) { 740 case NFNL_SUBSYS_CTNETLINK: 741 break; 742 case NFNL_SUBSYS_CTNETLINK_EXP: 743 break; 744 default: 745 return; 746 } 747 748 /* ctnetlink_has_listener is u8 */ 749 if (group >= 8) 750 return; 751 752 group_bit = (1 << group); 753 754 spin_lock(&nfnl_grp_active_lock); 755 if (!nfnetlink_has_listeners(net, group)) { 756 u8 v = READ_ONCE(nf_ctnetlink_has_listener); 757 758 v &= ~group_bit; 759 760 /* read concurrently without nfnl_grp_active_lock held. */ 761 WRITE_ONCE(nf_ctnetlink_has_listener, v); 762 } 763 spin_unlock(&nfnl_grp_active_lock); 764 #endif 765 } 766 767 static int __net_init nfnetlink_net_init(struct net *net) 768 { 769 struct nfnl_net *nfnlnet = nfnl_pernet(net); 770 struct netlink_kernel_cfg cfg = { 771 .groups = NFNLGRP_MAX, 772 .input = nfnetlink_rcv, 773 .bind = nfnetlink_bind, 774 .unbind = nfnetlink_unbind, 775 }; 776 777 nfnlnet->nfnl = netlink_kernel_create(net, NETLINK_NETFILTER, &cfg); 778 if (!nfnlnet->nfnl) 779 return -ENOMEM; 780 return 0; 781 } 782 783 static void __net_exit nfnetlink_net_exit_batch(struct list_head *net_exit_list) 784 { 785 struct nfnl_net *nfnlnet; 786 struct net *net; 787 788 list_for_each_entry(net, net_exit_list, exit_list) { 789 nfnlnet = nfnl_pernet(net); 790 791 netlink_kernel_release(nfnlnet->nfnl); 792 } 793 } 794 795 static struct pernet_operations nfnetlink_net_ops = { 796 .init = nfnetlink_net_init, 797 .exit_batch = nfnetlink_net_exit_batch, 798 .id = &nfnetlink_pernet_id, 799 .size = sizeof(struct nfnl_net), 800 }; 801 802 static int __init nfnetlink_init(void) 803 { 804 int i; 805 806 for (i = NFNLGRP_NONE + 1; i <= NFNLGRP_MAX; i++) 807 BUG_ON(nfnl_group2type[i] == NFNL_SUBSYS_NONE); 808 809 for (i=0; i<NFNL_SUBSYS_COUNT; i++) 810 __mutex_init(&table[i].mutex, nfnl_lockdep_names[i], &nfnl_lockdep_keys[i]); 811 812 return register_pernet_subsys(&nfnetlink_net_ops); 813 } 814 815 static void __exit nfnetlink_exit(void) 816 { 817 unregister_pernet_subsys(&nfnetlink_net_ops); 818 } 819 module_init(nfnetlink_init); 820 module_exit(nfnetlink_exit); 821