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 <linux/netfilter/nfnetlink.h> 32 33 MODULE_LICENSE("GPL"); 34 MODULE_AUTHOR("Harald Welte <laforge@netfilter.org>"); 35 MODULE_ALIAS_NET_PF_PROTO(PF_NETLINK, NETLINK_NETFILTER); 36 37 #define nfnl_dereference_protected(id) \ 38 rcu_dereference_protected(table[(id)].subsys, \ 39 lockdep_nfnl_is_held((id))) 40 41 #define NFNL_MAX_ATTR_COUNT 32 42 43 static struct { 44 struct mutex mutex; 45 const struct nfnetlink_subsystem __rcu *subsys; 46 } table[NFNL_SUBSYS_COUNT]; 47 48 static const int nfnl_group2type[NFNLGRP_MAX+1] = { 49 [NFNLGRP_CONNTRACK_NEW] = NFNL_SUBSYS_CTNETLINK, 50 [NFNLGRP_CONNTRACK_UPDATE] = NFNL_SUBSYS_CTNETLINK, 51 [NFNLGRP_CONNTRACK_DESTROY] = NFNL_SUBSYS_CTNETLINK, 52 [NFNLGRP_CONNTRACK_EXP_NEW] = NFNL_SUBSYS_CTNETLINK_EXP, 53 [NFNLGRP_CONNTRACK_EXP_UPDATE] = NFNL_SUBSYS_CTNETLINK_EXP, 54 [NFNLGRP_CONNTRACK_EXP_DESTROY] = NFNL_SUBSYS_CTNETLINK_EXP, 55 [NFNLGRP_NFTABLES] = NFNL_SUBSYS_NFTABLES, 56 [NFNLGRP_ACCT_QUOTA] = NFNL_SUBSYS_ACCT, 57 [NFNLGRP_NFTRACE] = NFNL_SUBSYS_NFTABLES, 58 }; 59 60 void nfnl_lock(__u8 subsys_id) 61 { 62 mutex_lock(&table[subsys_id].mutex); 63 } 64 EXPORT_SYMBOL_GPL(nfnl_lock); 65 66 void nfnl_unlock(__u8 subsys_id) 67 { 68 mutex_unlock(&table[subsys_id].mutex); 69 } 70 EXPORT_SYMBOL_GPL(nfnl_unlock); 71 72 #ifdef CONFIG_PROVE_LOCKING 73 bool lockdep_nfnl_is_held(u8 subsys_id) 74 { 75 return lockdep_is_held(&table[subsys_id].mutex); 76 } 77 EXPORT_SYMBOL_GPL(lockdep_nfnl_is_held); 78 #endif 79 80 int nfnetlink_subsys_register(const struct nfnetlink_subsystem *n) 81 { 82 u8 cb_id; 83 84 /* Sanity-check attr_count size to avoid stack buffer overflow. */ 85 for (cb_id = 0; cb_id < n->cb_count; cb_id++) 86 if (WARN_ON(n->cb[cb_id].attr_count > NFNL_MAX_ATTR_COUNT)) 87 return -EINVAL; 88 89 nfnl_lock(n->subsys_id); 90 if (table[n->subsys_id].subsys) { 91 nfnl_unlock(n->subsys_id); 92 return -EBUSY; 93 } 94 rcu_assign_pointer(table[n->subsys_id].subsys, n); 95 nfnl_unlock(n->subsys_id); 96 97 return 0; 98 } 99 EXPORT_SYMBOL_GPL(nfnetlink_subsys_register); 100 101 int nfnetlink_subsys_unregister(const struct nfnetlink_subsystem *n) 102 { 103 nfnl_lock(n->subsys_id); 104 table[n->subsys_id].subsys = NULL; 105 nfnl_unlock(n->subsys_id); 106 synchronize_rcu(); 107 return 0; 108 } 109 EXPORT_SYMBOL_GPL(nfnetlink_subsys_unregister); 110 111 static inline const struct nfnetlink_subsystem *nfnetlink_get_subsys(u16 type) 112 { 113 u8 subsys_id = NFNL_SUBSYS_ID(type); 114 115 if (subsys_id >= NFNL_SUBSYS_COUNT) 116 return NULL; 117 118 return rcu_dereference(table[subsys_id].subsys); 119 } 120 121 static inline const struct nfnl_callback * 122 nfnetlink_find_client(u16 type, const struct nfnetlink_subsystem *ss) 123 { 124 u8 cb_id = NFNL_MSG_TYPE(type); 125 126 if (cb_id >= ss->cb_count) 127 return NULL; 128 129 return &ss->cb[cb_id]; 130 } 131 132 int nfnetlink_has_listeners(struct net *net, unsigned int group) 133 { 134 return netlink_has_listeners(net->nfnl, group); 135 } 136 EXPORT_SYMBOL_GPL(nfnetlink_has_listeners); 137 138 int nfnetlink_send(struct sk_buff *skb, struct net *net, u32 portid, 139 unsigned int group, int echo, gfp_t flags) 140 { 141 return nlmsg_notify(net->nfnl, skb, portid, group, echo, flags); 142 } 143 EXPORT_SYMBOL_GPL(nfnetlink_send); 144 145 int nfnetlink_set_err(struct net *net, u32 portid, u32 group, int error) 146 { 147 return netlink_set_err(net->nfnl, portid, group, error); 148 } 149 EXPORT_SYMBOL_GPL(nfnetlink_set_err); 150 151 int nfnetlink_unicast(struct sk_buff *skb, struct net *net, u32 portid, 152 int flags) 153 { 154 return netlink_unicast(net->nfnl, skb, portid, flags); 155 } 156 EXPORT_SYMBOL_GPL(nfnetlink_unicast); 157 158 /* Process one complete nfnetlink message. */ 159 static int nfnetlink_rcv_msg(struct sk_buff *skb, struct nlmsghdr *nlh, 160 struct netlink_ext_ack *extack) 161 { 162 struct net *net = sock_net(skb->sk); 163 const struct nfnl_callback *nc; 164 const struct nfnetlink_subsystem *ss; 165 int type, err; 166 167 /* All the messages must at least contain nfgenmsg */ 168 if (nlmsg_len(nlh) < sizeof(struct nfgenmsg)) 169 return 0; 170 171 type = nlh->nlmsg_type; 172 replay: 173 rcu_read_lock(); 174 ss = nfnetlink_get_subsys(type); 175 if (!ss) { 176 #ifdef CONFIG_MODULES 177 rcu_read_unlock(); 178 request_module("nfnetlink-subsys-%d", NFNL_SUBSYS_ID(type)); 179 rcu_read_lock(); 180 ss = nfnetlink_get_subsys(type); 181 if (!ss) 182 #endif 183 { 184 rcu_read_unlock(); 185 return -EINVAL; 186 } 187 } 188 189 nc = nfnetlink_find_client(type, ss); 190 if (!nc) { 191 rcu_read_unlock(); 192 return -EINVAL; 193 } 194 195 { 196 int min_len = nlmsg_total_size(sizeof(struct nfgenmsg)); 197 u8 cb_id = NFNL_MSG_TYPE(nlh->nlmsg_type); 198 struct nlattr *cda[NFNL_MAX_ATTR_COUNT + 1]; 199 struct nlattr *attr = (void *)nlh + min_len; 200 int attrlen = nlh->nlmsg_len - min_len; 201 __u8 subsys_id = NFNL_SUBSYS_ID(type); 202 203 /* Sanity-check NFNL_MAX_ATTR_COUNT */ 204 if (ss->cb[cb_id].attr_count > NFNL_MAX_ATTR_COUNT) { 205 rcu_read_unlock(); 206 return -ENOMEM; 207 } 208 209 err = nla_parse_deprecated(cda, ss->cb[cb_id].attr_count, 210 attr, attrlen, 211 ss->cb[cb_id].policy, extack); 212 if (err < 0) { 213 rcu_read_unlock(); 214 return err; 215 } 216 217 if (nc->call_rcu) { 218 err = nc->call_rcu(net, net->nfnl, skb, nlh, 219 (const struct nlattr **)cda, 220 extack); 221 rcu_read_unlock(); 222 } else { 223 rcu_read_unlock(); 224 nfnl_lock(subsys_id); 225 if (nfnl_dereference_protected(subsys_id) != ss || 226 nfnetlink_find_client(type, ss) != nc) 227 err = -EAGAIN; 228 else if (nc->call) 229 err = nc->call(net, net->nfnl, skb, nlh, 230 (const struct nlattr **)cda, 231 extack); 232 else 233 err = -EINVAL; 234 nfnl_unlock(subsys_id); 235 } 236 if (err == -EAGAIN) 237 goto replay; 238 return err; 239 } 240 } 241 242 struct nfnl_err { 243 struct list_head head; 244 struct nlmsghdr *nlh; 245 int err; 246 struct netlink_ext_ack extack; 247 }; 248 249 static int nfnl_err_add(struct list_head *list, struct nlmsghdr *nlh, int err, 250 const struct netlink_ext_ack *extack) 251 { 252 struct nfnl_err *nfnl_err; 253 254 nfnl_err = kmalloc(sizeof(struct nfnl_err), GFP_KERNEL); 255 if (nfnl_err == NULL) 256 return -ENOMEM; 257 258 nfnl_err->nlh = nlh; 259 nfnl_err->err = err; 260 nfnl_err->extack = *extack; 261 list_add_tail(&nfnl_err->head, list); 262 263 return 0; 264 } 265 266 static void nfnl_err_del(struct nfnl_err *nfnl_err) 267 { 268 list_del(&nfnl_err->head); 269 kfree(nfnl_err); 270 } 271 272 static void nfnl_err_reset(struct list_head *err_list) 273 { 274 struct nfnl_err *nfnl_err, *next; 275 276 list_for_each_entry_safe(nfnl_err, next, err_list, head) 277 nfnl_err_del(nfnl_err); 278 } 279 280 static void nfnl_err_deliver(struct list_head *err_list, struct sk_buff *skb) 281 { 282 struct nfnl_err *nfnl_err, *next; 283 284 list_for_each_entry_safe(nfnl_err, next, err_list, head) { 285 netlink_ack(skb, nfnl_err->nlh, nfnl_err->err, 286 &nfnl_err->extack); 287 nfnl_err_del(nfnl_err); 288 } 289 } 290 291 enum { 292 NFNL_BATCH_FAILURE = (1 << 0), 293 NFNL_BATCH_DONE = (1 << 1), 294 NFNL_BATCH_REPLAY = (1 << 2), 295 }; 296 297 static void nfnetlink_rcv_batch(struct sk_buff *skb, struct nlmsghdr *nlh, 298 u16 subsys_id, u32 genid) 299 { 300 struct sk_buff *oskb = skb; 301 struct net *net = sock_net(skb->sk); 302 const struct nfnetlink_subsystem *ss; 303 const struct nfnl_callback *nc; 304 struct netlink_ext_ack extack; 305 LIST_HEAD(err_list); 306 u32 status; 307 int err; 308 309 if (subsys_id >= NFNL_SUBSYS_COUNT) 310 return netlink_ack(skb, nlh, -EINVAL, NULL); 311 replay: 312 status = 0; 313 314 skb = netlink_skb_clone(oskb, GFP_KERNEL); 315 if (!skb) 316 return netlink_ack(oskb, nlh, -ENOMEM, NULL); 317 318 nfnl_lock(subsys_id); 319 ss = nfnl_dereference_protected(subsys_id); 320 if (!ss) { 321 #ifdef CONFIG_MODULES 322 nfnl_unlock(subsys_id); 323 request_module("nfnetlink-subsys-%d", subsys_id); 324 nfnl_lock(subsys_id); 325 ss = nfnl_dereference_protected(subsys_id); 326 if (!ss) 327 #endif 328 { 329 nfnl_unlock(subsys_id); 330 netlink_ack(oskb, nlh, -EOPNOTSUPP, NULL); 331 return kfree_skb(skb); 332 } 333 } 334 335 if (!ss->valid_genid || !ss->commit || !ss->abort) { 336 nfnl_unlock(subsys_id); 337 netlink_ack(oskb, nlh, -EOPNOTSUPP, NULL); 338 return kfree_skb(skb); 339 } 340 341 if (!try_module_get(ss->owner)) { 342 nfnl_unlock(subsys_id); 343 netlink_ack(oskb, nlh, -EOPNOTSUPP, NULL); 344 return kfree_skb(skb); 345 } 346 347 if (!ss->valid_genid(net, genid)) { 348 module_put(ss->owner); 349 nfnl_unlock(subsys_id); 350 netlink_ack(oskb, nlh, -ERESTART, NULL); 351 return kfree_skb(skb); 352 } 353 354 nfnl_unlock(subsys_id); 355 356 while (skb->len >= nlmsg_total_size(0)) { 357 int msglen, type; 358 359 if (fatal_signal_pending(current)) { 360 nfnl_err_reset(&err_list); 361 err = -EINTR; 362 status = NFNL_BATCH_FAILURE; 363 goto done; 364 } 365 366 memset(&extack, 0, sizeof(extack)); 367 nlh = nlmsg_hdr(skb); 368 err = 0; 369 370 if (nlh->nlmsg_len < NLMSG_HDRLEN || 371 skb->len < nlh->nlmsg_len || 372 nlmsg_len(nlh) < sizeof(struct nfgenmsg)) { 373 nfnl_err_reset(&err_list); 374 status |= NFNL_BATCH_FAILURE; 375 goto done; 376 } 377 378 /* Only requests are handled by the kernel */ 379 if (!(nlh->nlmsg_flags & NLM_F_REQUEST)) { 380 err = -EINVAL; 381 goto ack; 382 } 383 384 type = nlh->nlmsg_type; 385 if (type == NFNL_MSG_BATCH_BEGIN) { 386 /* Malformed: Batch begin twice */ 387 nfnl_err_reset(&err_list); 388 status |= NFNL_BATCH_FAILURE; 389 goto done; 390 } else if (type == NFNL_MSG_BATCH_END) { 391 status |= NFNL_BATCH_DONE; 392 goto done; 393 } else if (type < NLMSG_MIN_TYPE) { 394 err = -EINVAL; 395 goto ack; 396 } 397 398 /* We only accept a batch with messages for the same 399 * subsystem. 400 */ 401 if (NFNL_SUBSYS_ID(type) != subsys_id) { 402 err = -EINVAL; 403 goto ack; 404 } 405 406 nc = nfnetlink_find_client(type, ss); 407 if (!nc) { 408 err = -EINVAL; 409 goto ack; 410 } 411 412 { 413 int min_len = nlmsg_total_size(sizeof(struct nfgenmsg)); 414 u8 cb_id = NFNL_MSG_TYPE(nlh->nlmsg_type); 415 struct nlattr *cda[NFNL_MAX_ATTR_COUNT + 1]; 416 struct nlattr *attr = (void *)nlh + min_len; 417 int attrlen = nlh->nlmsg_len - min_len; 418 419 /* Sanity-check NFTA_MAX_ATTR */ 420 if (ss->cb[cb_id].attr_count > NFNL_MAX_ATTR_COUNT) { 421 err = -ENOMEM; 422 goto ack; 423 } 424 425 err = nla_parse_deprecated(cda, 426 ss->cb[cb_id].attr_count, 427 attr, attrlen, 428 ss->cb[cb_id].policy, NULL); 429 if (err < 0) 430 goto ack; 431 432 if (nc->call_batch) { 433 err = nc->call_batch(net, net->nfnl, skb, nlh, 434 (const struct nlattr **)cda, 435 &extack); 436 } 437 438 /* The lock was released to autoload some module, we 439 * have to abort and start from scratch using the 440 * original skb. 441 */ 442 if (err == -EAGAIN) { 443 status |= NFNL_BATCH_REPLAY; 444 goto done; 445 } 446 } 447 ack: 448 if (nlh->nlmsg_flags & NLM_F_ACK || err) { 449 /* Errors are delivered once the full batch has been 450 * processed, this avoids that the same error is 451 * reported several times when replaying the batch. 452 */ 453 if (nfnl_err_add(&err_list, nlh, err, &extack) < 0) { 454 /* We failed to enqueue an error, reset the 455 * list of errors and send OOM to userspace 456 * pointing to the batch header. 457 */ 458 nfnl_err_reset(&err_list); 459 netlink_ack(oskb, nlmsg_hdr(oskb), -ENOMEM, 460 NULL); 461 status |= NFNL_BATCH_FAILURE; 462 goto done; 463 } 464 /* We don't stop processing the batch on errors, thus, 465 * userspace gets all the errors that the batch 466 * triggers. 467 */ 468 if (err) 469 status |= NFNL_BATCH_FAILURE; 470 } 471 472 msglen = NLMSG_ALIGN(nlh->nlmsg_len); 473 if (msglen > skb->len) 474 msglen = skb->len; 475 skb_pull(skb, msglen); 476 } 477 done: 478 if (status & NFNL_BATCH_REPLAY) { 479 ss->abort(net, oskb); 480 nfnl_err_reset(&err_list); 481 kfree_skb(skb); 482 module_put(ss->owner); 483 goto replay; 484 } else if (status == NFNL_BATCH_DONE) { 485 err = ss->commit(net, oskb); 486 if (err == -EAGAIN) { 487 status |= NFNL_BATCH_REPLAY; 488 goto done; 489 } else if (err) { 490 ss->abort(net, oskb); 491 netlink_ack(oskb, nlmsg_hdr(oskb), err, NULL); 492 } 493 } else { 494 ss->abort(net, oskb); 495 } 496 if (ss->cleanup) 497 ss->cleanup(net); 498 499 nfnl_err_deliver(&err_list, oskb); 500 kfree_skb(skb); 501 module_put(ss->owner); 502 } 503 504 static const struct nla_policy nfnl_batch_policy[NFNL_BATCH_MAX + 1] = { 505 [NFNL_BATCH_GENID] = { .type = NLA_U32 }, 506 }; 507 508 static void nfnetlink_rcv_skb_batch(struct sk_buff *skb, struct nlmsghdr *nlh) 509 { 510 int min_len = nlmsg_total_size(sizeof(struct nfgenmsg)); 511 struct nlattr *attr = (void *)nlh + min_len; 512 struct nlattr *cda[NFNL_BATCH_MAX + 1]; 513 int attrlen = nlh->nlmsg_len - min_len; 514 struct nfgenmsg *nfgenmsg; 515 int msglen, err; 516 u32 gen_id = 0; 517 u16 res_id; 518 519 msglen = NLMSG_ALIGN(nlh->nlmsg_len); 520 if (msglen > skb->len) 521 msglen = skb->len; 522 523 if (skb->len < NLMSG_HDRLEN + sizeof(struct nfgenmsg)) 524 return; 525 526 err = nla_parse_deprecated(cda, NFNL_BATCH_MAX, attr, attrlen, 527 nfnl_batch_policy, NULL); 528 if (err < 0) { 529 netlink_ack(skb, nlh, err, NULL); 530 return; 531 } 532 if (cda[NFNL_BATCH_GENID]) 533 gen_id = ntohl(nla_get_be32(cda[NFNL_BATCH_GENID])); 534 535 nfgenmsg = nlmsg_data(nlh); 536 skb_pull(skb, msglen); 537 /* Work around old nft using host byte order */ 538 if (nfgenmsg->res_id == NFNL_SUBSYS_NFTABLES) 539 res_id = NFNL_SUBSYS_NFTABLES; 540 else 541 res_id = ntohs(nfgenmsg->res_id); 542 543 nfnetlink_rcv_batch(skb, nlh, res_id, gen_id); 544 } 545 546 static void nfnetlink_rcv(struct sk_buff *skb) 547 { 548 struct nlmsghdr *nlh = nlmsg_hdr(skb); 549 550 if (skb->len < NLMSG_HDRLEN || 551 nlh->nlmsg_len < NLMSG_HDRLEN || 552 skb->len < nlh->nlmsg_len) 553 return; 554 555 if (!netlink_net_capable(skb, CAP_NET_ADMIN)) { 556 netlink_ack(skb, nlh, -EPERM, NULL); 557 return; 558 } 559 560 if (nlh->nlmsg_type == NFNL_MSG_BATCH_BEGIN) 561 nfnetlink_rcv_skb_batch(skb, nlh); 562 else 563 netlink_rcv_skb(skb, nfnetlink_rcv_msg); 564 } 565 566 #ifdef CONFIG_MODULES 567 static int nfnetlink_bind(struct net *net, int group) 568 { 569 const struct nfnetlink_subsystem *ss; 570 int type; 571 572 if (group <= NFNLGRP_NONE || group > NFNLGRP_MAX) 573 return 0; 574 575 type = nfnl_group2type[group]; 576 577 rcu_read_lock(); 578 ss = nfnetlink_get_subsys(type << 8); 579 rcu_read_unlock(); 580 if (!ss) 581 request_module_nowait("nfnetlink-subsys-%d", type); 582 return 0; 583 } 584 #endif 585 586 static int __net_init nfnetlink_net_init(struct net *net) 587 { 588 struct sock *nfnl; 589 struct netlink_kernel_cfg cfg = { 590 .groups = NFNLGRP_MAX, 591 .input = nfnetlink_rcv, 592 #ifdef CONFIG_MODULES 593 .bind = nfnetlink_bind, 594 #endif 595 }; 596 597 nfnl = netlink_kernel_create(net, NETLINK_NETFILTER, &cfg); 598 if (!nfnl) 599 return -ENOMEM; 600 net->nfnl_stash = nfnl; 601 rcu_assign_pointer(net->nfnl, nfnl); 602 return 0; 603 } 604 605 static void __net_exit nfnetlink_net_exit_batch(struct list_head *net_exit_list) 606 { 607 struct net *net; 608 609 list_for_each_entry(net, net_exit_list, exit_list) 610 RCU_INIT_POINTER(net->nfnl, NULL); 611 synchronize_net(); 612 list_for_each_entry(net, net_exit_list, exit_list) 613 netlink_kernel_release(net->nfnl_stash); 614 } 615 616 static struct pernet_operations nfnetlink_net_ops = { 617 .init = nfnetlink_net_init, 618 .exit_batch = nfnetlink_net_exit_batch, 619 }; 620 621 static int __init nfnetlink_init(void) 622 { 623 int i; 624 625 for (i = NFNLGRP_NONE + 1; i <= NFNLGRP_MAX; i++) 626 BUG_ON(nfnl_group2type[i] == NFNL_SUBSYS_NONE); 627 628 for (i=0; i<NFNL_SUBSYS_COUNT; i++) 629 mutex_init(&table[i].mutex); 630 631 return register_pernet_subsys(&nfnetlink_net_ops); 632 } 633 634 static void __exit nfnetlink_exit(void) 635 { 636 unregister_pernet_subsys(&nfnetlink_net_ops); 637 } 638 module_init(nfnetlink_init); 639 module_exit(nfnetlink_exit); 640