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 MODULE_DESCRIPTION("Netfilter messages via netlink socket"); 37 38 #define nfnl_dereference_protected(id) \ 39 rcu_dereference_protected(table[(id)].subsys, \ 40 lockdep_nfnl_is_held((id))) 41 42 #define NFNL_MAX_ATTR_COUNT 32 43 44 static struct { 45 struct mutex mutex; 46 const struct nfnetlink_subsystem __rcu *subsys; 47 } table[NFNL_SUBSYS_COUNT]; 48 49 static const int nfnl_group2type[NFNLGRP_MAX+1] = { 50 [NFNLGRP_CONNTRACK_NEW] = NFNL_SUBSYS_CTNETLINK, 51 [NFNLGRP_CONNTRACK_UPDATE] = NFNL_SUBSYS_CTNETLINK, 52 [NFNLGRP_CONNTRACK_DESTROY] = NFNL_SUBSYS_CTNETLINK, 53 [NFNLGRP_CONNTRACK_EXP_NEW] = NFNL_SUBSYS_CTNETLINK_EXP, 54 [NFNLGRP_CONNTRACK_EXP_UPDATE] = NFNL_SUBSYS_CTNETLINK_EXP, 55 [NFNLGRP_CONNTRACK_EXP_DESTROY] = NFNL_SUBSYS_CTNETLINK_EXP, 56 [NFNLGRP_NFTABLES] = NFNL_SUBSYS_NFTABLES, 57 [NFNLGRP_ACCT_QUOTA] = NFNL_SUBSYS_ACCT, 58 [NFNLGRP_NFTRACE] = NFNL_SUBSYS_NFTABLES, 59 }; 60 61 void nfnl_lock(__u8 subsys_id) 62 { 63 mutex_lock(&table[subsys_id].mutex); 64 } 65 EXPORT_SYMBOL_GPL(nfnl_lock); 66 67 void nfnl_unlock(__u8 subsys_id) 68 { 69 mutex_unlock(&table[subsys_id].mutex); 70 } 71 EXPORT_SYMBOL_GPL(nfnl_unlock); 72 73 #ifdef CONFIG_PROVE_LOCKING 74 bool lockdep_nfnl_is_held(u8 subsys_id) 75 { 76 return lockdep_is_held(&table[subsys_id].mutex); 77 } 78 EXPORT_SYMBOL_GPL(lockdep_nfnl_is_held); 79 #endif 80 81 int nfnetlink_subsys_register(const struct nfnetlink_subsystem *n) 82 { 83 u8 cb_id; 84 85 /* Sanity-check attr_count size to avoid stack buffer overflow. */ 86 for (cb_id = 0; cb_id < n->cb_count; cb_id++) 87 if (WARN_ON(n->cb[cb_id].attr_count > NFNL_MAX_ATTR_COUNT)) 88 return -EINVAL; 89 90 nfnl_lock(n->subsys_id); 91 if (table[n->subsys_id].subsys) { 92 nfnl_unlock(n->subsys_id); 93 return -EBUSY; 94 } 95 rcu_assign_pointer(table[n->subsys_id].subsys, n); 96 nfnl_unlock(n->subsys_id); 97 98 return 0; 99 } 100 EXPORT_SYMBOL_GPL(nfnetlink_subsys_register); 101 102 int nfnetlink_subsys_unregister(const struct nfnetlink_subsystem *n) 103 { 104 nfnl_lock(n->subsys_id); 105 table[n->subsys_id].subsys = NULL; 106 nfnl_unlock(n->subsys_id); 107 synchronize_rcu(); 108 return 0; 109 } 110 EXPORT_SYMBOL_GPL(nfnetlink_subsys_unregister); 111 112 static inline const struct nfnetlink_subsystem *nfnetlink_get_subsys(u16 type) 113 { 114 u8 subsys_id = NFNL_SUBSYS_ID(type); 115 116 if (subsys_id >= NFNL_SUBSYS_COUNT) 117 return NULL; 118 119 return rcu_dereference(table[subsys_id].subsys); 120 } 121 122 static inline const struct nfnl_callback * 123 nfnetlink_find_client(u16 type, const struct nfnetlink_subsystem *ss) 124 { 125 u8 cb_id = NFNL_MSG_TYPE(type); 126 127 if (cb_id >= ss->cb_count) 128 return NULL; 129 130 return &ss->cb[cb_id]; 131 } 132 133 int nfnetlink_has_listeners(struct net *net, unsigned int group) 134 { 135 return netlink_has_listeners(net->nfnl, group); 136 } 137 EXPORT_SYMBOL_GPL(nfnetlink_has_listeners); 138 139 int nfnetlink_send(struct sk_buff *skb, struct net *net, u32 portid, 140 unsigned int group, int echo, gfp_t flags) 141 { 142 return nlmsg_notify(net->nfnl, skb, portid, group, echo, flags); 143 } 144 EXPORT_SYMBOL_GPL(nfnetlink_send); 145 146 int nfnetlink_set_err(struct net *net, u32 portid, u32 group, int error) 147 { 148 return netlink_set_err(net->nfnl, portid, group, error); 149 } 150 EXPORT_SYMBOL_GPL(nfnetlink_set_err); 151 152 int nfnetlink_unicast(struct sk_buff *skb, struct net *net, u32 portid) 153 { 154 int err; 155 156 err = nlmsg_unicast(net->nfnl, skb, portid); 157 if (err == -EAGAIN) 158 err = -ENOBUFS; 159 160 return err; 161 } 162 EXPORT_SYMBOL_GPL(nfnetlink_unicast); 163 164 /* Process one complete nfnetlink message. */ 165 static int nfnetlink_rcv_msg(struct sk_buff *skb, struct nlmsghdr *nlh, 166 struct netlink_ext_ack *extack) 167 { 168 struct net *net = sock_net(skb->sk); 169 const struct nfnl_callback *nc; 170 const struct nfnetlink_subsystem *ss; 171 int type, err; 172 173 /* All the messages must at least contain nfgenmsg */ 174 if (nlmsg_len(nlh) < sizeof(struct nfgenmsg)) 175 return 0; 176 177 type = nlh->nlmsg_type; 178 replay: 179 rcu_read_lock(); 180 ss = nfnetlink_get_subsys(type); 181 if (!ss) { 182 #ifdef CONFIG_MODULES 183 rcu_read_unlock(); 184 request_module("nfnetlink-subsys-%d", NFNL_SUBSYS_ID(type)); 185 rcu_read_lock(); 186 ss = nfnetlink_get_subsys(type); 187 if (!ss) 188 #endif 189 { 190 rcu_read_unlock(); 191 return -EINVAL; 192 } 193 } 194 195 nc = nfnetlink_find_client(type, ss); 196 if (!nc) { 197 rcu_read_unlock(); 198 return -EINVAL; 199 } 200 201 { 202 int min_len = nlmsg_total_size(sizeof(struct nfgenmsg)); 203 u8 cb_id = NFNL_MSG_TYPE(nlh->nlmsg_type); 204 struct nlattr *cda[NFNL_MAX_ATTR_COUNT + 1]; 205 struct nlattr *attr = (void *)nlh + min_len; 206 int attrlen = nlh->nlmsg_len - min_len; 207 __u8 subsys_id = NFNL_SUBSYS_ID(type); 208 209 /* Sanity-check NFNL_MAX_ATTR_COUNT */ 210 if (ss->cb[cb_id].attr_count > NFNL_MAX_ATTR_COUNT) { 211 rcu_read_unlock(); 212 return -ENOMEM; 213 } 214 215 err = nla_parse_deprecated(cda, ss->cb[cb_id].attr_count, 216 attr, attrlen, 217 ss->cb[cb_id].policy, extack); 218 if (err < 0) { 219 rcu_read_unlock(); 220 return err; 221 } 222 223 if (nc->call_rcu) { 224 err = nc->call_rcu(net, net->nfnl, skb, nlh, 225 (const struct nlattr **)cda, 226 extack); 227 rcu_read_unlock(); 228 } else { 229 rcu_read_unlock(); 230 nfnl_lock(subsys_id); 231 if (nfnl_dereference_protected(subsys_id) != ss || 232 nfnetlink_find_client(type, ss) != nc) 233 err = -EAGAIN; 234 else if (nc->call) 235 err = nc->call(net, net->nfnl, skb, nlh, 236 (const struct nlattr **)cda, 237 extack); 238 else 239 err = -EINVAL; 240 nfnl_unlock(subsys_id); 241 } 242 if (err == -EAGAIN) 243 goto replay; 244 return err; 245 } 246 } 247 248 struct nfnl_err { 249 struct list_head head; 250 struct nlmsghdr *nlh; 251 int err; 252 struct netlink_ext_ack extack; 253 }; 254 255 static int nfnl_err_add(struct list_head *list, struct nlmsghdr *nlh, int err, 256 const struct netlink_ext_ack *extack) 257 { 258 struct nfnl_err *nfnl_err; 259 260 nfnl_err = kmalloc(sizeof(struct nfnl_err), GFP_KERNEL); 261 if (nfnl_err == NULL) 262 return -ENOMEM; 263 264 nfnl_err->nlh = nlh; 265 nfnl_err->err = err; 266 nfnl_err->extack = *extack; 267 list_add_tail(&nfnl_err->head, list); 268 269 return 0; 270 } 271 272 static void nfnl_err_del(struct nfnl_err *nfnl_err) 273 { 274 list_del(&nfnl_err->head); 275 kfree(nfnl_err); 276 } 277 278 static void nfnl_err_reset(struct list_head *err_list) 279 { 280 struct nfnl_err *nfnl_err, *next; 281 282 list_for_each_entry_safe(nfnl_err, next, err_list, head) 283 nfnl_err_del(nfnl_err); 284 } 285 286 static void nfnl_err_deliver(struct list_head *err_list, struct sk_buff *skb) 287 { 288 struct nfnl_err *nfnl_err, *next; 289 290 list_for_each_entry_safe(nfnl_err, next, err_list, head) { 291 netlink_ack(skb, nfnl_err->nlh, nfnl_err->err, 292 &nfnl_err->extack); 293 nfnl_err_del(nfnl_err); 294 } 295 } 296 297 enum { 298 NFNL_BATCH_FAILURE = (1 << 0), 299 NFNL_BATCH_DONE = (1 << 1), 300 NFNL_BATCH_REPLAY = (1 << 2), 301 }; 302 303 static void nfnetlink_rcv_batch(struct sk_buff *skb, struct nlmsghdr *nlh, 304 u16 subsys_id, u32 genid) 305 { 306 struct sk_buff *oskb = skb; 307 struct net *net = sock_net(skb->sk); 308 const struct nfnetlink_subsystem *ss; 309 const struct nfnl_callback *nc; 310 struct netlink_ext_ack extack; 311 LIST_HEAD(err_list); 312 u32 status; 313 int err; 314 315 if (subsys_id >= NFNL_SUBSYS_COUNT) 316 return netlink_ack(skb, nlh, -EINVAL, NULL); 317 replay: 318 status = 0; 319 320 skb = netlink_skb_clone(oskb, GFP_KERNEL); 321 if (!skb) 322 return netlink_ack(oskb, nlh, -ENOMEM, NULL); 323 324 nfnl_lock(subsys_id); 325 ss = nfnl_dereference_protected(subsys_id); 326 if (!ss) { 327 #ifdef CONFIG_MODULES 328 nfnl_unlock(subsys_id); 329 request_module("nfnetlink-subsys-%d", subsys_id); 330 nfnl_lock(subsys_id); 331 ss = nfnl_dereference_protected(subsys_id); 332 if (!ss) 333 #endif 334 { 335 nfnl_unlock(subsys_id); 336 netlink_ack(oskb, nlh, -EOPNOTSUPP, NULL); 337 return kfree_skb(skb); 338 } 339 } 340 341 if (!ss->valid_genid || !ss->commit || !ss->abort) { 342 nfnl_unlock(subsys_id); 343 netlink_ack(oskb, nlh, -EOPNOTSUPP, NULL); 344 return kfree_skb(skb); 345 } 346 347 if (!try_module_get(ss->owner)) { 348 nfnl_unlock(subsys_id); 349 netlink_ack(oskb, nlh, -EOPNOTSUPP, NULL); 350 return kfree_skb(skb); 351 } 352 353 if (!ss->valid_genid(net, genid)) { 354 module_put(ss->owner); 355 nfnl_unlock(subsys_id); 356 netlink_ack(oskb, nlh, -ERESTART, NULL); 357 return kfree_skb(skb); 358 } 359 360 nfnl_unlock(subsys_id); 361 362 while (skb->len >= nlmsg_total_size(0)) { 363 int msglen, type; 364 365 if (fatal_signal_pending(current)) { 366 nfnl_err_reset(&err_list); 367 err = -EINTR; 368 status = NFNL_BATCH_FAILURE; 369 goto done; 370 } 371 372 memset(&extack, 0, sizeof(extack)); 373 nlh = nlmsg_hdr(skb); 374 err = 0; 375 376 if (nlh->nlmsg_len < NLMSG_HDRLEN || 377 skb->len < nlh->nlmsg_len || 378 nlmsg_len(nlh) < sizeof(struct nfgenmsg)) { 379 nfnl_err_reset(&err_list); 380 status |= NFNL_BATCH_FAILURE; 381 goto done; 382 } 383 384 /* Only requests are handled by the kernel */ 385 if (!(nlh->nlmsg_flags & NLM_F_REQUEST)) { 386 err = -EINVAL; 387 goto ack; 388 } 389 390 type = nlh->nlmsg_type; 391 if (type == NFNL_MSG_BATCH_BEGIN) { 392 /* Malformed: Batch begin twice */ 393 nfnl_err_reset(&err_list); 394 status |= NFNL_BATCH_FAILURE; 395 goto done; 396 } else if (type == NFNL_MSG_BATCH_END) { 397 status |= NFNL_BATCH_DONE; 398 goto done; 399 } else if (type < NLMSG_MIN_TYPE) { 400 err = -EINVAL; 401 goto ack; 402 } 403 404 /* We only accept a batch with messages for the same 405 * subsystem. 406 */ 407 if (NFNL_SUBSYS_ID(type) != subsys_id) { 408 err = -EINVAL; 409 goto ack; 410 } 411 412 nc = nfnetlink_find_client(type, ss); 413 if (!nc) { 414 err = -EINVAL; 415 goto ack; 416 } 417 418 { 419 int min_len = nlmsg_total_size(sizeof(struct nfgenmsg)); 420 u8 cb_id = NFNL_MSG_TYPE(nlh->nlmsg_type); 421 struct nlattr *cda[NFNL_MAX_ATTR_COUNT + 1]; 422 struct nlattr *attr = (void *)nlh + min_len; 423 int attrlen = nlh->nlmsg_len - min_len; 424 425 /* Sanity-check NFTA_MAX_ATTR */ 426 if (ss->cb[cb_id].attr_count > NFNL_MAX_ATTR_COUNT) { 427 err = -ENOMEM; 428 goto ack; 429 } 430 431 err = nla_parse_deprecated(cda, 432 ss->cb[cb_id].attr_count, 433 attr, attrlen, 434 ss->cb[cb_id].policy, NULL); 435 if (err < 0) 436 goto ack; 437 438 if (nc->call_batch) { 439 err = nc->call_batch(net, net->nfnl, skb, nlh, 440 (const struct nlattr **)cda, 441 &extack); 442 } 443 444 /* The lock was released to autoload some module, we 445 * have to abort and start from scratch using the 446 * original skb. 447 */ 448 if (err == -EAGAIN) { 449 status |= NFNL_BATCH_REPLAY; 450 goto done; 451 } 452 } 453 ack: 454 if (nlh->nlmsg_flags & NLM_F_ACK || err) { 455 /* Errors are delivered once the full batch has been 456 * processed, this avoids that the same error is 457 * reported several times when replaying the batch. 458 */ 459 if (nfnl_err_add(&err_list, nlh, err, &extack) < 0) { 460 /* We failed to enqueue an error, reset the 461 * list of errors and send OOM to userspace 462 * pointing to the batch header. 463 */ 464 nfnl_err_reset(&err_list); 465 netlink_ack(oskb, nlmsg_hdr(oskb), -ENOMEM, 466 NULL); 467 status |= NFNL_BATCH_FAILURE; 468 goto done; 469 } 470 /* We don't stop processing the batch on errors, thus, 471 * userspace gets all the errors that the batch 472 * triggers. 473 */ 474 if (err) 475 status |= NFNL_BATCH_FAILURE; 476 } 477 478 msglen = NLMSG_ALIGN(nlh->nlmsg_len); 479 if (msglen > skb->len) 480 msglen = skb->len; 481 skb_pull(skb, msglen); 482 } 483 done: 484 if (status & NFNL_BATCH_REPLAY) { 485 ss->abort(net, oskb, true); 486 nfnl_err_reset(&err_list); 487 kfree_skb(skb); 488 module_put(ss->owner); 489 goto replay; 490 } else if (status == NFNL_BATCH_DONE) { 491 err = ss->commit(net, oskb); 492 if (err == -EAGAIN) { 493 status |= NFNL_BATCH_REPLAY; 494 goto done; 495 } else if (err) { 496 ss->abort(net, oskb, false); 497 netlink_ack(oskb, nlmsg_hdr(oskb), err, NULL); 498 } 499 } else { 500 ss->abort(net, oskb, false); 501 } 502 if (ss->cleanup) 503 ss->cleanup(net); 504 505 nfnl_err_deliver(&err_list, oskb); 506 kfree_skb(skb); 507 module_put(ss->owner); 508 } 509 510 static const struct nla_policy nfnl_batch_policy[NFNL_BATCH_MAX + 1] = { 511 [NFNL_BATCH_GENID] = { .type = NLA_U32 }, 512 }; 513 514 static void nfnetlink_rcv_skb_batch(struct sk_buff *skb, struct nlmsghdr *nlh) 515 { 516 int min_len = nlmsg_total_size(sizeof(struct nfgenmsg)); 517 struct nlattr *attr = (void *)nlh + min_len; 518 struct nlattr *cda[NFNL_BATCH_MAX + 1]; 519 int attrlen = nlh->nlmsg_len - min_len; 520 struct nfgenmsg *nfgenmsg; 521 int msglen, err; 522 u32 gen_id = 0; 523 u16 res_id; 524 525 msglen = NLMSG_ALIGN(nlh->nlmsg_len); 526 if (msglen > skb->len) 527 msglen = skb->len; 528 529 if (skb->len < NLMSG_HDRLEN + sizeof(struct nfgenmsg)) 530 return; 531 532 err = nla_parse_deprecated(cda, NFNL_BATCH_MAX, attr, attrlen, 533 nfnl_batch_policy, NULL); 534 if (err < 0) { 535 netlink_ack(skb, nlh, err, NULL); 536 return; 537 } 538 if (cda[NFNL_BATCH_GENID]) 539 gen_id = ntohl(nla_get_be32(cda[NFNL_BATCH_GENID])); 540 541 nfgenmsg = nlmsg_data(nlh); 542 skb_pull(skb, msglen); 543 /* Work around old nft using host byte order */ 544 if (nfgenmsg->res_id == NFNL_SUBSYS_NFTABLES) 545 res_id = NFNL_SUBSYS_NFTABLES; 546 else 547 res_id = ntohs(nfgenmsg->res_id); 548 549 nfnetlink_rcv_batch(skb, nlh, res_id, gen_id); 550 } 551 552 static void nfnetlink_rcv(struct sk_buff *skb) 553 { 554 struct nlmsghdr *nlh = nlmsg_hdr(skb); 555 556 if (skb->len < NLMSG_HDRLEN || 557 nlh->nlmsg_len < NLMSG_HDRLEN || 558 skb->len < nlh->nlmsg_len) 559 return; 560 561 if (!netlink_net_capable(skb, CAP_NET_ADMIN)) { 562 netlink_ack(skb, nlh, -EPERM, NULL); 563 return; 564 } 565 566 if (nlh->nlmsg_type == NFNL_MSG_BATCH_BEGIN) 567 nfnetlink_rcv_skb_batch(skb, nlh); 568 else 569 netlink_rcv_skb(skb, nfnetlink_rcv_msg); 570 } 571 572 #ifdef CONFIG_MODULES 573 static int nfnetlink_bind(struct net *net, int group) 574 { 575 const struct nfnetlink_subsystem *ss; 576 int type; 577 578 if (group <= NFNLGRP_NONE || group > NFNLGRP_MAX) 579 return 0; 580 581 type = nfnl_group2type[group]; 582 583 rcu_read_lock(); 584 ss = nfnetlink_get_subsys(type << 8); 585 rcu_read_unlock(); 586 if (!ss) 587 request_module_nowait("nfnetlink-subsys-%d", type); 588 return 0; 589 } 590 #endif 591 592 static int __net_init nfnetlink_net_init(struct net *net) 593 { 594 struct sock *nfnl; 595 struct netlink_kernel_cfg cfg = { 596 .groups = NFNLGRP_MAX, 597 .input = nfnetlink_rcv, 598 #ifdef CONFIG_MODULES 599 .bind = nfnetlink_bind, 600 #endif 601 }; 602 603 nfnl = netlink_kernel_create(net, NETLINK_NETFILTER, &cfg); 604 if (!nfnl) 605 return -ENOMEM; 606 net->nfnl_stash = nfnl; 607 rcu_assign_pointer(net->nfnl, nfnl); 608 return 0; 609 } 610 611 static void __net_exit nfnetlink_net_exit_batch(struct list_head *net_exit_list) 612 { 613 struct net *net; 614 615 list_for_each_entry(net, net_exit_list, exit_list) 616 RCU_INIT_POINTER(net->nfnl, NULL); 617 synchronize_net(); 618 list_for_each_entry(net, net_exit_list, exit_list) 619 netlink_kernel_release(net->nfnl_stash); 620 } 621 622 static struct pernet_operations nfnetlink_net_ops = { 623 .init = nfnetlink_net_init, 624 .exit_batch = nfnetlink_net_exit_batch, 625 }; 626 627 static int __init nfnetlink_init(void) 628 { 629 int i; 630 631 for (i = NFNLGRP_NONE + 1; i <= NFNLGRP_MAX; i++) 632 BUG_ON(nfnl_group2type[i] == NFNL_SUBSYS_NONE); 633 634 for (i=0; i<NFNL_SUBSYS_COUNT; i++) 635 mutex_init(&table[i].mutex); 636 637 return register_pernet_subsys(&nfnetlink_net_ops); 638 } 639 640 static void __exit nfnetlink_exit(void) 641 { 642 unregister_pernet_subsys(&nfnetlink_net_ops); 643 } 644 module_init(nfnetlink_init); 645 module_exit(nfnetlink_exit); 646