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 29 #include <net/netlink.h> 30 #include <linux/netfilter/nfnetlink.h> 31 32 MODULE_LICENSE("GPL"); 33 MODULE_AUTHOR("Harald Welte <laforge@netfilter.org>"); 34 MODULE_ALIAS_NET_PF_PROTO(PF_NETLINK, NETLINK_NETFILTER); 35 36 #define nfnl_dereference_protected(id) \ 37 rcu_dereference_protected(table[(id)].subsys, \ 38 lockdep_nfnl_is_held((id))) 39 40 static char __initdata nfversion[] = "0.30"; 41 42 static struct { 43 struct mutex mutex; 44 const struct nfnetlink_subsystem __rcu *subsys; 45 } table[NFNL_SUBSYS_COUNT]; 46 47 static const int nfnl_group2type[NFNLGRP_MAX+1] = { 48 [NFNLGRP_CONNTRACK_NEW] = NFNL_SUBSYS_CTNETLINK, 49 [NFNLGRP_CONNTRACK_UPDATE] = NFNL_SUBSYS_CTNETLINK, 50 [NFNLGRP_CONNTRACK_DESTROY] = NFNL_SUBSYS_CTNETLINK, 51 [NFNLGRP_CONNTRACK_EXP_NEW] = NFNL_SUBSYS_CTNETLINK_EXP, 52 [NFNLGRP_CONNTRACK_EXP_UPDATE] = NFNL_SUBSYS_CTNETLINK_EXP, 53 [NFNLGRP_CONNTRACK_EXP_DESTROY] = NFNL_SUBSYS_CTNETLINK_EXP, 54 [NFNLGRP_NFTABLES] = NFNL_SUBSYS_NFTABLES, 55 [NFNLGRP_ACCT_QUOTA] = NFNL_SUBSYS_ACCT, 56 [NFNLGRP_NFTRACE] = NFNL_SUBSYS_NFTABLES, 57 }; 58 59 void nfnl_lock(__u8 subsys_id) 60 { 61 mutex_lock(&table[subsys_id].mutex); 62 } 63 EXPORT_SYMBOL_GPL(nfnl_lock); 64 65 void nfnl_unlock(__u8 subsys_id) 66 { 67 mutex_unlock(&table[subsys_id].mutex); 68 } 69 EXPORT_SYMBOL_GPL(nfnl_unlock); 70 71 #ifdef CONFIG_PROVE_LOCKING 72 bool lockdep_nfnl_is_held(u8 subsys_id) 73 { 74 return lockdep_is_held(&table[subsys_id].mutex); 75 } 76 EXPORT_SYMBOL_GPL(lockdep_nfnl_is_held); 77 #endif 78 79 int nfnetlink_subsys_register(const struct nfnetlink_subsystem *n) 80 { 81 nfnl_lock(n->subsys_id); 82 if (table[n->subsys_id].subsys) { 83 nfnl_unlock(n->subsys_id); 84 return -EBUSY; 85 } 86 rcu_assign_pointer(table[n->subsys_id].subsys, n); 87 nfnl_unlock(n->subsys_id); 88 89 return 0; 90 } 91 EXPORT_SYMBOL_GPL(nfnetlink_subsys_register); 92 93 int nfnetlink_subsys_unregister(const struct nfnetlink_subsystem *n) 94 { 95 nfnl_lock(n->subsys_id); 96 table[n->subsys_id].subsys = NULL; 97 nfnl_unlock(n->subsys_id); 98 synchronize_rcu(); 99 return 0; 100 } 101 EXPORT_SYMBOL_GPL(nfnetlink_subsys_unregister); 102 103 static inline const struct nfnetlink_subsystem *nfnetlink_get_subsys(u16 type) 104 { 105 u8 subsys_id = NFNL_SUBSYS_ID(type); 106 107 if (subsys_id >= NFNL_SUBSYS_COUNT) 108 return NULL; 109 110 return rcu_dereference(table[subsys_id].subsys); 111 } 112 113 static inline const struct nfnl_callback * 114 nfnetlink_find_client(u16 type, const struct nfnetlink_subsystem *ss) 115 { 116 u8 cb_id = NFNL_MSG_TYPE(type); 117 118 if (cb_id >= ss->cb_count) 119 return NULL; 120 121 return &ss->cb[cb_id]; 122 } 123 124 int nfnetlink_has_listeners(struct net *net, unsigned int group) 125 { 126 return netlink_has_listeners(net->nfnl, group); 127 } 128 EXPORT_SYMBOL_GPL(nfnetlink_has_listeners); 129 130 int nfnetlink_send(struct sk_buff *skb, struct net *net, u32 portid, 131 unsigned int group, int echo, gfp_t flags) 132 { 133 return nlmsg_notify(net->nfnl, skb, portid, group, echo, flags); 134 } 135 EXPORT_SYMBOL_GPL(nfnetlink_send); 136 137 int nfnetlink_set_err(struct net *net, u32 portid, u32 group, int error) 138 { 139 return netlink_set_err(net->nfnl, portid, group, error); 140 } 141 EXPORT_SYMBOL_GPL(nfnetlink_set_err); 142 143 int nfnetlink_unicast(struct sk_buff *skb, struct net *net, u32 portid, 144 int flags) 145 { 146 return netlink_unicast(net->nfnl, skb, portid, flags); 147 } 148 EXPORT_SYMBOL_GPL(nfnetlink_unicast); 149 150 /* Process one complete nfnetlink message. */ 151 static int nfnetlink_rcv_msg(struct sk_buff *skb, struct nlmsghdr *nlh, 152 struct netlink_ext_ack *extack) 153 { 154 struct net *net = sock_net(skb->sk); 155 const struct nfnl_callback *nc; 156 const struct nfnetlink_subsystem *ss; 157 int type, err; 158 159 /* All the messages must at least contain nfgenmsg */ 160 if (nlmsg_len(nlh) < sizeof(struct nfgenmsg)) 161 return 0; 162 163 type = nlh->nlmsg_type; 164 replay: 165 rcu_read_lock(); 166 ss = nfnetlink_get_subsys(type); 167 if (!ss) { 168 #ifdef CONFIG_MODULES 169 rcu_read_unlock(); 170 request_module("nfnetlink-subsys-%d", NFNL_SUBSYS_ID(type)); 171 rcu_read_lock(); 172 ss = nfnetlink_get_subsys(type); 173 if (!ss) 174 #endif 175 { 176 rcu_read_unlock(); 177 return -EINVAL; 178 } 179 } 180 181 nc = nfnetlink_find_client(type, ss); 182 if (!nc) { 183 rcu_read_unlock(); 184 return -EINVAL; 185 } 186 187 { 188 int min_len = nlmsg_total_size(sizeof(struct nfgenmsg)); 189 u8 cb_id = NFNL_MSG_TYPE(nlh->nlmsg_type); 190 struct nlattr *cda[ss->cb[cb_id].attr_count + 1]; 191 struct nlattr *attr = (void *)nlh + min_len; 192 int attrlen = nlh->nlmsg_len - min_len; 193 __u8 subsys_id = NFNL_SUBSYS_ID(type); 194 195 err = nla_parse(cda, ss->cb[cb_id].attr_count, attr, attrlen, 196 ss->cb[cb_id].policy, extack); 197 if (err < 0) { 198 rcu_read_unlock(); 199 return err; 200 } 201 202 if (nc->call_rcu) { 203 err = nc->call_rcu(net, net->nfnl, skb, nlh, 204 (const struct nlattr **)cda, 205 extack); 206 rcu_read_unlock(); 207 } else { 208 rcu_read_unlock(); 209 nfnl_lock(subsys_id); 210 if (nfnl_dereference_protected(subsys_id) != ss || 211 nfnetlink_find_client(type, ss) != nc) 212 err = -EAGAIN; 213 else if (nc->call) 214 err = nc->call(net, net->nfnl, skb, nlh, 215 (const struct nlattr **)cda, 216 extack); 217 else 218 err = -EINVAL; 219 nfnl_unlock(subsys_id); 220 } 221 if (err == -EAGAIN) 222 goto replay; 223 return err; 224 } 225 } 226 227 struct nfnl_err { 228 struct list_head head; 229 struct nlmsghdr *nlh; 230 int err; 231 struct netlink_ext_ack extack; 232 }; 233 234 static int nfnl_err_add(struct list_head *list, struct nlmsghdr *nlh, int err, 235 const struct netlink_ext_ack *extack) 236 { 237 struct nfnl_err *nfnl_err; 238 239 nfnl_err = kmalloc(sizeof(struct nfnl_err), GFP_KERNEL); 240 if (nfnl_err == NULL) 241 return -ENOMEM; 242 243 nfnl_err->nlh = nlh; 244 nfnl_err->err = err; 245 nfnl_err->extack = *extack; 246 list_add_tail(&nfnl_err->head, list); 247 248 return 0; 249 } 250 251 static void nfnl_err_del(struct nfnl_err *nfnl_err) 252 { 253 list_del(&nfnl_err->head); 254 kfree(nfnl_err); 255 } 256 257 static void nfnl_err_reset(struct list_head *err_list) 258 { 259 struct nfnl_err *nfnl_err, *next; 260 261 list_for_each_entry_safe(nfnl_err, next, err_list, head) 262 nfnl_err_del(nfnl_err); 263 } 264 265 static void nfnl_err_deliver(struct list_head *err_list, struct sk_buff *skb) 266 { 267 struct nfnl_err *nfnl_err, *next; 268 269 list_for_each_entry_safe(nfnl_err, next, err_list, head) { 270 netlink_ack(skb, nfnl_err->nlh, nfnl_err->err, 271 &nfnl_err->extack); 272 nfnl_err_del(nfnl_err); 273 } 274 } 275 276 enum { 277 NFNL_BATCH_FAILURE = (1 << 0), 278 NFNL_BATCH_DONE = (1 << 1), 279 NFNL_BATCH_REPLAY = (1 << 2), 280 }; 281 282 static void nfnetlink_rcv_batch(struct sk_buff *skb, struct nlmsghdr *nlh, 283 u16 subsys_id, u32 genid) 284 { 285 struct sk_buff *oskb = skb; 286 struct net *net = sock_net(skb->sk); 287 const struct nfnetlink_subsystem *ss; 288 const struct nfnl_callback *nc; 289 struct netlink_ext_ack extack; 290 LIST_HEAD(err_list); 291 u32 status; 292 int err; 293 294 if (subsys_id >= NFNL_SUBSYS_COUNT) 295 return netlink_ack(skb, nlh, -EINVAL, NULL); 296 replay: 297 status = 0; 298 299 skb = netlink_skb_clone(oskb, GFP_KERNEL); 300 if (!skb) 301 return netlink_ack(oskb, nlh, -ENOMEM, NULL); 302 303 nfnl_lock(subsys_id); 304 ss = nfnl_dereference_protected(subsys_id); 305 if (!ss) { 306 #ifdef CONFIG_MODULES 307 nfnl_unlock(subsys_id); 308 request_module("nfnetlink-subsys-%d", subsys_id); 309 nfnl_lock(subsys_id); 310 ss = nfnl_dereference_protected(subsys_id); 311 if (!ss) 312 #endif 313 { 314 nfnl_unlock(subsys_id); 315 netlink_ack(oskb, nlh, -EOPNOTSUPP, NULL); 316 return kfree_skb(skb); 317 } 318 } 319 320 if (!ss->commit || !ss->abort) { 321 nfnl_unlock(subsys_id); 322 netlink_ack(oskb, nlh, -EOPNOTSUPP, NULL); 323 return kfree_skb(skb); 324 } 325 326 if (genid && ss->valid_genid && !ss->valid_genid(net, genid)) { 327 nfnl_unlock(subsys_id); 328 netlink_ack(oskb, nlh, -ERESTART, NULL); 329 return kfree_skb(skb); 330 } 331 332 while (skb->len >= nlmsg_total_size(0)) { 333 int msglen, type; 334 335 memset(&extack, 0, sizeof(extack)); 336 nlh = nlmsg_hdr(skb); 337 err = 0; 338 339 if (nlh->nlmsg_len < NLMSG_HDRLEN || 340 skb->len < nlh->nlmsg_len || 341 nlmsg_len(nlh) < sizeof(struct nfgenmsg)) { 342 nfnl_err_reset(&err_list); 343 status |= NFNL_BATCH_FAILURE; 344 goto done; 345 } 346 347 /* Only requests are handled by the kernel */ 348 if (!(nlh->nlmsg_flags & NLM_F_REQUEST)) { 349 err = -EINVAL; 350 goto ack; 351 } 352 353 type = nlh->nlmsg_type; 354 if (type == NFNL_MSG_BATCH_BEGIN) { 355 /* Malformed: Batch begin twice */ 356 nfnl_err_reset(&err_list); 357 status |= NFNL_BATCH_FAILURE; 358 goto done; 359 } else if (type == NFNL_MSG_BATCH_END) { 360 status |= NFNL_BATCH_DONE; 361 goto done; 362 } else if (type < NLMSG_MIN_TYPE) { 363 err = -EINVAL; 364 goto ack; 365 } 366 367 /* We only accept a batch with messages for the same 368 * subsystem. 369 */ 370 if (NFNL_SUBSYS_ID(type) != subsys_id) { 371 err = -EINVAL; 372 goto ack; 373 } 374 375 nc = nfnetlink_find_client(type, ss); 376 if (!nc) { 377 err = -EINVAL; 378 goto ack; 379 } 380 381 { 382 int min_len = nlmsg_total_size(sizeof(struct nfgenmsg)); 383 u8 cb_id = NFNL_MSG_TYPE(nlh->nlmsg_type); 384 struct nlattr *cda[ss->cb[cb_id].attr_count + 1]; 385 struct nlattr *attr = (void *)nlh + min_len; 386 int attrlen = nlh->nlmsg_len - min_len; 387 388 err = nla_parse(cda, ss->cb[cb_id].attr_count, attr, 389 attrlen, ss->cb[cb_id].policy, NULL); 390 if (err < 0) 391 goto ack; 392 393 if (nc->call_batch) { 394 err = nc->call_batch(net, net->nfnl, skb, nlh, 395 (const struct nlattr **)cda, 396 &extack); 397 } 398 399 /* The lock was released to autoload some module, we 400 * have to abort and start from scratch using the 401 * original skb. 402 */ 403 if (err == -EAGAIN) { 404 status |= NFNL_BATCH_REPLAY; 405 goto next; 406 } 407 } 408 ack: 409 if (nlh->nlmsg_flags & NLM_F_ACK || err) { 410 /* Errors are delivered once the full batch has been 411 * processed, this avoids that the same error is 412 * reported several times when replaying the batch. 413 */ 414 if (nfnl_err_add(&err_list, nlh, err, &extack) < 0) { 415 /* We failed to enqueue an error, reset the 416 * list of errors and send OOM to userspace 417 * pointing to the batch header. 418 */ 419 nfnl_err_reset(&err_list); 420 netlink_ack(oskb, nlmsg_hdr(oskb), -ENOMEM, 421 NULL); 422 status |= NFNL_BATCH_FAILURE; 423 goto done; 424 } 425 /* We don't stop processing the batch on errors, thus, 426 * userspace gets all the errors that the batch 427 * triggers. 428 */ 429 if (err) 430 status |= NFNL_BATCH_FAILURE; 431 } 432 next: 433 msglen = NLMSG_ALIGN(nlh->nlmsg_len); 434 if (msglen > skb->len) 435 msglen = skb->len; 436 skb_pull(skb, msglen); 437 } 438 done: 439 if (status & NFNL_BATCH_REPLAY) { 440 ss->abort(net, oskb); 441 nfnl_err_reset(&err_list); 442 nfnl_unlock(subsys_id); 443 kfree_skb(skb); 444 goto replay; 445 } else if (status == NFNL_BATCH_DONE) { 446 ss->commit(net, oskb); 447 } else { 448 ss->abort(net, oskb); 449 } 450 451 nfnl_err_deliver(&err_list, oskb); 452 nfnl_unlock(subsys_id); 453 kfree_skb(skb); 454 } 455 456 static const struct nla_policy nfnl_batch_policy[NFNL_BATCH_MAX + 1] = { 457 [NFNL_BATCH_GENID] = { .type = NLA_U32 }, 458 }; 459 460 static void nfnetlink_rcv_skb_batch(struct sk_buff *skb, struct nlmsghdr *nlh) 461 { 462 int min_len = nlmsg_total_size(sizeof(struct nfgenmsg)); 463 struct nlattr *attr = (void *)nlh + min_len; 464 struct nlattr *cda[NFNL_BATCH_MAX + 1]; 465 int attrlen = nlh->nlmsg_len - min_len; 466 struct nfgenmsg *nfgenmsg; 467 int msglen, err; 468 u32 gen_id = 0; 469 u16 res_id; 470 471 msglen = NLMSG_ALIGN(nlh->nlmsg_len); 472 if (msglen > skb->len) 473 msglen = skb->len; 474 475 if (skb->len < NLMSG_HDRLEN + sizeof(struct nfgenmsg)) 476 return; 477 478 err = nla_parse(cda, NFNL_BATCH_MAX, attr, attrlen, nfnl_batch_policy, 479 NULL); 480 if (err < 0) { 481 netlink_ack(skb, nlh, err, NULL); 482 return; 483 } 484 if (cda[NFNL_BATCH_GENID]) 485 gen_id = ntohl(nla_get_be32(cda[NFNL_BATCH_GENID])); 486 487 nfgenmsg = nlmsg_data(nlh); 488 skb_pull(skb, msglen); 489 /* Work around old nft using host byte order */ 490 if (nfgenmsg->res_id == NFNL_SUBSYS_NFTABLES) 491 res_id = NFNL_SUBSYS_NFTABLES; 492 else 493 res_id = ntohs(nfgenmsg->res_id); 494 495 nfnetlink_rcv_batch(skb, nlh, res_id, gen_id); 496 } 497 498 static void nfnetlink_rcv(struct sk_buff *skb) 499 { 500 struct nlmsghdr *nlh = nlmsg_hdr(skb); 501 502 if (skb->len < NLMSG_HDRLEN || 503 nlh->nlmsg_len < NLMSG_HDRLEN || 504 skb->len < nlh->nlmsg_len) 505 return; 506 507 if (!netlink_net_capable(skb, CAP_NET_ADMIN)) { 508 netlink_ack(skb, nlh, -EPERM, NULL); 509 return; 510 } 511 512 if (nlh->nlmsg_type == NFNL_MSG_BATCH_BEGIN) 513 nfnetlink_rcv_skb_batch(skb, nlh); 514 else 515 netlink_rcv_skb(skb, nfnetlink_rcv_msg); 516 } 517 518 #ifdef CONFIG_MODULES 519 static int nfnetlink_bind(struct net *net, int group) 520 { 521 const struct nfnetlink_subsystem *ss; 522 int type; 523 524 if (group <= NFNLGRP_NONE || group > NFNLGRP_MAX) 525 return 0; 526 527 type = nfnl_group2type[group]; 528 529 rcu_read_lock(); 530 ss = nfnetlink_get_subsys(type << 8); 531 rcu_read_unlock(); 532 if (!ss) 533 request_module("nfnetlink-subsys-%d", type); 534 return 0; 535 } 536 #endif 537 538 static int __net_init nfnetlink_net_init(struct net *net) 539 { 540 struct sock *nfnl; 541 struct netlink_kernel_cfg cfg = { 542 .groups = NFNLGRP_MAX, 543 .input = nfnetlink_rcv, 544 #ifdef CONFIG_MODULES 545 .bind = nfnetlink_bind, 546 #endif 547 }; 548 549 nfnl = netlink_kernel_create(net, NETLINK_NETFILTER, &cfg); 550 if (!nfnl) 551 return -ENOMEM; 552 net->nfnl_stash = nfnl; 553 rcu_assign_pointer(net->nfnl, nfnl); 554 return 0; 555 } 556 557 static void __net_exit nfnetlink_net_exit_batch(struct list_head *net_exit_list) 558 { 559 struct net *net; 560 561 list_for_each_entry(net, net_exit_list, exit_list) 562 RCU_INIT_POINTER(net->nfnl, NULL); 563 synchronize_net(); 564 list_for_each_entry(net, net_exit_list, exit_list) 565 netlink_kernel_release(net->nfnl_stash); 566 } 567 568 static struct pernet_operations nfnetlink_net_ops = { 569 .init = nfnetlink_net_init, 570 .exit_batch = nfnetlink_net_exit_batch, 571 }; 572 573 static int __init nfnetlink_init(void) 574 { 575 int i; 576 577 for (i = NFNLGRP_NONE + 1; i <= NFNLGRP_MAX; i++) 578 BUG_ON(nfnl_group2type[i] == NFNL_SUBSYS_NONE); 579 580 for (i=0; i<NFNL_SUBSYS_COUNT; i++) 581 mutex_init(&table[i].mutex); 582 583 pr_info("Netfilter messages via NETLINK v%s.\n", nfversion); 584 return register_pernet_subsys(&nfnetlink_net_ops); 585 } 586 587 static void __exit nfnetlink_exit(void) 588 { 589 pr_info("Removing netfilter NETLINK layer.\n"); 590 unregister_pernet_subsys(&nfnetlink_net_ops); 591 } 592 module_init(nfnetlink_init); 593 module_exit(nfnetlink_exit); 594