net/netfilter/nfnetlink.c

   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) 2005,2007 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 <asm/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(u_int16_t type)
 104 {
 105         u_int8_t 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(u_int16_t type, const struct nfnetlink_subsystem *ss)
 115 {
 116         u_int8_t 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 struct sk_buff *nfnetlink_alloc_skb(struct net *net, unsigned int size,
 131                                     u32 dst_portid, gfp_t gfp_mask)
 132 {
 133         return netlink_alloc_skb(net->nfnl, size, dst_portid, gfp_mask);
 134 }
 135 EXPORT_SYMBOL_GPL(nfnetlink_alloc_skb);
 136
 137 int nfnetlink_send(struct sk_buff *skb, struct net *net, u32 portid,
 138                    unsigned int group, int echo, gfp_t flags)
 139 {
 140         return nlmsg_notify(net->nfnl, skb, portid, group, echo, flags);
 141 }
 142 EXPORT_SYMBOL_GPL(nfnetlink_send);
 143
 144 int nfnetlink_set_err(struct net *net, u32 portid, u32 group, int error)
 145 {
 146         return netlink_set_err(net->nfnl, portid, group, error);
 147 }
 148 EXPORT_SYMBOL_GPL(nfnetlink_set_err);
 149
 150 int nfnetlink_unicast(struct sk_buff *skb, struct net *net, u32 portid,
 151                       int flags)
 152 {
 153         return netlink_unicast(net->nfnl, skb, portid, flags);
 154 }
 155 EXPORT_SYMBOL_GPL(nfnetlink_unicast);
 156
 157 /* Process one complete nfnetlink message. */
 158 static int nfnetlink_rcv_msg(struct sk_buff *skb, struct nlmsghdr *nlh)
 159 {
 160         struct net *net = sock_net(skb->sk);
 161         const struct nfnl_callback *nc;
 162         const struct nfnetlink_subsystem *ss;
 163         int type, err;
 164
 165         /* All the messages must at least contain nfgenmsg */
 166         if (nlmsg_len(nlh) < sizeof(struct nfgenmsg))
 167                 return 0;
 168
 169         type = nlh->nlmsg_type;
 170 replay:
 171         rcu_read_lock();
 172         ss = nfnetlink_get_subsys(type);
 173         if (!ss) {
 174 #ifdef CONFIG_MODULES
 175                 rcu_read_unlock();
 176                 request_module("nfnetlink-subsys-%d", NFNL_SUBSYS_ID(type));
 177                 rcu_read_lock();
 178                 ss = nfnetlink_get_subsys(type);
 179                 if (!ss)
 180 #endif
 181                 {
 182                         rcu_read_unlock();
 183                         return -EINVAL;
 184                 }
 185         }
 186
 187         nc = nfnetlink_find_client(type, ss);
 188         if (!nc) {
 189                 rcu_read_unlock();
 190                 return -EINVAL;
 191         }
 192
 193         {
 194                 int min_len = nlmsg_total_size(sizeof(struct nfgenmsg));
 195                 u_int8_t cb_id = NFNL_MSG_TYPE(nlh->nlmsg_type);
 196                 struct nlattr *cda[ss->cb[cb_id].attr_count + 1];
 197                 struct nlattr *attr = (void *)nlh + min_len;
 198                 int attrlen = nlh->nlmsg_len - min_len;
 199                 __u8 subsys_id = NFNL_SUBSYS_ID(type);
 200
 201                 err = nla_parse(cda, ss->cb[cb_id].attr_count,
 202                                 attr, attrlen, ss->cb[cb_id].policy);
 203                 if (err < 0) {
 204                         rcu_read_unlock();
 205                         return err;
 206                 }
 207
 208                 if (nc->call_rcu) {
 209                         err = nc->call_rcu(net, net->nfnl, skb, nlh,
 210                                            (const struct nlattr **)cda);
 211                         rcu_read_unlock();
 212                 } else {
 213                         rcu_read_unlock();
 214                         nfnl_lock(subsys_id);
 215                         if (nfnl_dereference_protected(subsys_id) != ss ||
 216                             nfnetlink_find_client(type, ss) != nc)
 217                                 err = -EAGAIN;
 218                         else if (nc->call)
 219                                 err = nc->call(net, net->nfnl, skb, nlh,
 220                                                (const struct nlattr **)cda);
 221                         else
 222                                 err = -EINVAL;
 223                         nfnl_unlock(subsys_id);
 224                 }
 225                 if (err == -EAGAIN)
 226                         goto replay;
 227                 return err;
 228         }
 229 }
 230
 231 struct nfnl_err {
 232         struct list_head        head;
 233         struct nlmsghdr         *nlh;
 234         int                     err;
 235 };
 236
 237 static int nfnl_err_add(struct list_head *list, struct nlmsghdr *nlh, int err)
 238 {
 239         struct nfnl_err *nfnl_err;
 240
 241         nfnl_err = kmalloc(sizeof(struct nfnl_err), GFP_KERNEL);
 242         if (nfnl_err == NULL)
 243                 return -ENOMEM;
 244
 245         nfnl_err->nlh = nlh;
 246         nfnl_err->err = err;
 247         list_add_tail(&nfnl_err->head, list);
 248
 249         return 0;
 250 }
 251
 252 static void nfnl_err_del(struct nfnl_err *nfnl_err)
 253 {
 254         list_del(&nfnl_err->head);
 255         kfree(nfnl_err);
 256 }
 257
 258 static void nfnl_err_reset(struct list_head *err_list)
 259 {
 260         struct nfnl_err *nfnl_err, *next;
 261
 262         list_for_each_entry_safe(nfnl_err, next, err_list, head)
 263                 nfnl_err_del(nfnl_err);
 264 }
 265
 266 static void nfnl_err_deliver(struct list_head *err_list, struct sk_buff *skb)
 267 {
 268         struct nfnl_err *nfnl_err, *next;
 269
 270         list_for_each_entry_safe(nfnl_err, next, err_list, head) {
 271                 netlink_ack(skb, nfnl_err->nlh, nfnl_err->err);
 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                                 u_int16_t subsys_id)
 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         static LIST_HEAD(err_list);
 290         u32 status;
 291         int err;
 292
 293         if (subsys_id >= NFNL_SUBSYS_COUNT)
 294                 return netlink_ack(skb, nlh, -EINVAL);
 295 replay:
 296         status = 0;
 297
 298         skb = netlink_skb_clone(oskb, GFP_KERNEL);
 299         if (!skb)
 300                 return netlink_ack(oskb, nlh, -ENOMEM);
 301
 302         nfnl_lock(subsys_id);
 303         ss = nfnl_dereference_protected(subsys_id);
 304         if (!ss) {
 305 #ifdef CONFIG_MODULES
 306                 nfnl_unlock(subsys_id);
 307                 request_module("nfnetlink-subsys-%d", subsys_id);
 308                 nfnl_lock(subsys_id);
 309                 ss = nfnl_dereference_protected(subsys_id);
 310                 if (!ss)
 311 #endif
 312                 {
 313                         nfnl_unlock(subsys_id);
 314                         netlink_ack(oskb, nlh, -EOPNOTSUPP);
 315                         return kfree_skb(skb);
 316                 }
 317         }
 318
 319         if (!ss->commit || !ss->abort) {
 320                 nfnl_unlock(subsys_id);
 321                 netlink_ack(oskb, nlh, -EOPNOTSUPP);
 322                 return kfree_skb(skb);
 323         }
 324
 325         while (skb->len >= nlmsg_total_size(0)) {
 326                 int msglen, type;
 327
 328                 nlh = nlmsg_hdr(skb);
 329                 err = 0;
 330
 331                 if (nlh->nlmsg_len < NLMSG_HDRLEN ||
 332                     skb->len < nlh->nlmsg_len ||
 333                     nlmsg_len(nlh) < sizeof(struct nfgenmsg)) {
 334                         nfnl_err_reset(&err_list);
 335                         status |= NFNL_BATCH_FAILURE;
 336                         goto done;
 337                 }
 338
 339                 /* Only requests are handled by the kernel */
 340                 if (!(nlh->nlmsg_flags & NLM_F_REQUEST)) {
 341                         err = -EINVAL;
 342                         goto ack;
 343                 }
 344
 345                 type = nlh->nlmsg_type;
 346                 if (type == NFNL_MSG_BATCH_BEGIN) {
 347                         /* Malformed: Batch begin twice */
 348                         nfnl_err_reset(&err_list);
 349                         status |= NFNL_BATCH_FAILURE;
 350                         goto done;
 351                 } else if (type == NFNL_MSG_BATCH_END) {
 352                         status |= NFNL_BATCH_DONE;
 353                         goto done;
 354                 } else if (type < NLMSG_MIN_TYPE) {
 355                         err = -EINVAL;
 356                         goto ack;
 357                 }
 358
 359                 /* We only accept a batch with messages for the same
 360                  * subsystem.
 361                  */
 362                 if (NFNL_SUBSYS_ID(type) != subsys_id) {
 363                         err = -EINVAL;
 364                         goto ack;
 365                 }
 366
 367                 nc = nfnetlink_find_client(type, ss);
 368                 if (!nc) {
 369                         err = -EINVAL;
 370                         goto ack;
 371                 }
 372
 373                 {
 374                         int min_len = nlmsg_total_size(sizeof(struct nfgenmsg));
 375                         u_int8_t cb_id = NFNL_MSG_TYPE(nlh->nlmsg_type);
 376                         struct nlattr *cda[ss->cb[cb_id].attr_count + 1];
 377                         struct nlattr *attr = (void *)nlh + min_len;
 378                         int attrlen = nlh->nlmsg_len - min_len;
 379
 380                         err = nla_parse(cda, ss->cb[cb_id].attr_count,
 381                                         attr, attrlen, ss->cb[cb_id].policy);
 382                         if (err < 0)
 383                                 goto ack;
 384
 385                         if (nc->call_batch) {
 386                                 err = nc->call_batch(net, net->nfnl, skb, nlh,
 387                                                      (const struct nlattr **)cda);
 388                         }
 389
 390                         /* The lock was released to autoload some module, we
 391                          * have to abort and start from scratch using the
 392                          * original skb.
 393                          */
 394                         if (err == -EAGAIN) {
 395                                 status |= NFNL_BATCH_REPLAY;
 396                                 goto next;
 397                         }
 398                 }
 399 ack:
 400                 if (nlh->nlmsg_flags & NLM_F_ACK || err) {
 401                         /* Errors are delivered once the full batch has been
 402                          * processed, this avoids that the same error is
 403                          * reported several times when replaying the batch.
 404                          */
 405                         if (nfnl_err_add(&err_list, nlh, err) < 0) {
 406                                 /* We failed to enqueue an error, reset the
 407                                  * list of errors and send OOM to userspace
 408                                  * pointing to the batch header.
 409                                  */
 410                                 nfnl_err_reset(&err_list);
 411                                 netlink_ack(oskb, nlmsg_hdr(oskb), -ENOMEM);
 412                                 status |= NFNL_BATCH_FAILURE;
 413                                 goto done;
 414                         }
 415                         /* We don't stop processing the batch on errors, thus,
 416                          * userspace gets all the errors that the batch
 417                          * triggers.
 418                          */
 419                         if (err)
 420                                 status |= NFNL_BATCH_FAILURE;
 421                 }
 422 next:
 423                 msglen = NLMSG_ALIGN(nlh->nlmsg_len);
 424                 if (msglen > skb->len)
 425                         msglen = skb->len;
 426                 skb_pull(skb, msglen);
 427         }
 428 done:
 429         if (status & NFNL_BATCH_REPLAY) {
 430                 ss->abort(net, oskb);
 431                 nfnl_err_reset(&err_list);
 432                 nfnl_unlock(subsys_id);
 433                 kfree_skb(skb);
 434                 goto replay;
 435         } else if (status == NFNL_BATCH_DONE) {
 436                 ss->commit(net, oskb);
 437         } else {
 438                 ss->abort(net, oskb);
 439         }
 440
 441         nfnl_err_deliver(&err_list, oskb);
 442         nfnl_unlock(subsys_id);
 443         kfree_skb(skb);
 444 }
 445
 446 static void nfnetlink_rcv(struct sk_buff *skb)
 447 {
 448         struct nlmsghdr *nlh = nlmsg_hdr(skb);
 449         u_int16_t res_id;
 450         int msglen;
 451
 452         if (nlh->nlmsg_len < NLMSG_HDRLEN ||
 453             skb->len < nlh->nlmsg_len)
 454                 return;
 455
 456         if (!netlink_net_capable(skb, CAP_NET_ADMIN)) {
 457                 netlink_ack(skb, nlh, -EPERM);
 458                 return;
 459         }
 460
 461         if (nlh->nlmsg_type == NFNL_MSG_BATCH_BEGIN) {
 462                 struct nfgenmsg *nfgenmsg;
 463
 464                 msglen = NLMSG_ALIGN(nlh->nlmsg_len);
 465                 if (msglen > skb->len)
 466                         msglen = skb->len;
 467
 468                 if (nlh->nlmsg_len < NLMSG_HDRLEN ||
 469                     skb->len < NLMSG_HDRLEN + sizeof(struct nfgenmsg))
 470                         return;
 471
 472                 nfgenmsg = nlmsg_data(nlh);
 473                 skb_pull(skb, msglen);
 474                 /* Work around old nft using host byte order */
 475                 if (nfgenmsg->res_id == NFNL_SUBSYS_NFTABLES)
 476                         res_id = NFNL_SUBSYS_NFTABLES;
 477                 else
 478                         res_id = ntohs(nfgenmsg->res_id);
 479                 nfnetlink_rcv_batch(skb, nlh, res_id);
 480         } else {
 481                 netlink_rcv_skb(skb, &nfnetlink_rcv_msg);
 482         }
 483 }
 484
 485 #ifdef CONFIG_MODULES
 486 static int nfnetlink_bind(struct net *net, int group)
 487 {
 488         const struct nfnetlink_subsystem *ss;
 489         int type;
 490
 491         if (group <= NFNLGRP_NONE || group > NFNLGRP_MAX)
 492                 return 0;
 493
 494         type = nfnl_group2type[group];
 495
 496         rcu_read_lock();
 497         ss = nfnetlink_get_subsys(type << 8);
 498         rcu_read_unlock();
 499         if (!ss)
 500                 request_module("nfnetlink-subsys-%d", type);
 501         return 0;
 502 }
 503 #endif
 504
 505 static int __net_init nfnetlink_net_init(struct net *net)
 506 {
 507         struct sock *nfnl;
 508         struct netlink_kernel_cfg cfg = {
 509                 .groups = NFNLGRP_MAX,
 510                 .input  = nfnetlink_rcv,
 511 #ifdef CONFIG_MODULES
 512                 .bind   = nfnetlink_bind,
 513 #endif
 514         };
 515
 516         nfnl = netlink_kernel_create(net, NETLINK_NETFILTER, &cfg);
 517         if (!nfnl)
 518                 return -ENOMEM;
 519         net->nfnl_stash = nfnl;
 520         rcu_assign_pointer(net->nfnl, nfnl);
 521         return 0;
 522 }
 523
 524 static void __net_exit nfnetlink_net_exit_batch(struct list_head *net_exit_list)
 525 {
 526         struct net *net;
 527
 528         list_for_each_entry(net, net_exit_list, exit_list)
 529                 RCU_INIT_POINTER(net->nfnl, NULL);
 530         synchronize_net();
 531         list_for_each_entry(net, net_exit_list, exit_list)
 532                 netlink_kernel_release(net->nfnl_stash);
 533 }
 534
 535 static struct pernet_operations nfnetlink_net_ops = {
 536         .init           = nfnetlink_net_init,
 537         .exit_batch     = nfnetlink_net_exit_batch,
 538 };
 539
 540 static int __init nfnetlink_init(void)
 541 {
 542         int i;
 543
 544         for (i = NFNLGRP_NONE + 1; i <= NFNLGRP_MAX; i++)
 545                 BUG_ON(nfnl_group2type[i] == NFNL_SUBSYS_NONE);
 546
 547         for (i=0; i<NFNL_SUBSYS_COUNT; i++)
 548                 mutex_init(&table[i].mutex);
 549
 550         pr_info("Netfilter messages via NETLINK v%s.\n", nfversion);
 551         return register_pernet_subsys(&nfnetlink_net_ops);
 552 }
 553
 554 static void __exit nfnetlink_exit(void)
 555 {
 556         pr_info("Removing netfilter NETLINK layer.\n");
 557         unregister_pernet_subsys(&nfnetlink_net_ops);
 558 }
 559 module_init(nfnetlink_init);
 560 module_exit(nfnetlink_exit);