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 static char __initdata nfversion[] = "0.30";
  37
  38 static struct {
  39         struct mutex                            mutex;
  40         const struct nfnetlink_subsystem __rcu  *subsys;
  41 } table[NFNL_SUBSYS_COUNT];
  42
  43 static const int nfnl_group2type[NFNLGRP_MAX+1] = {
  44         [NFNLGRP_CONNTRACK_NEW]         = NFNL_SUBSYS_CTNETLINK,
  45         [NFNLGRP_CONNTRACK_UPDATE]      = NFNL_SUBSYS_CTNETLINK,
  46         [NFNLGRP_CONNTRACK_DESTROY]     = NFNL_SUBSYS_CTNETLINK,
  47         [NFNLGRP_CONNTRACK_EXP_NEW]     = NFNL_SUBSYS_CTNETLINK_EXP,
  48         [NFNLGRP_CONNTRACK_EXP_UPDATE]  = NFNL_SUBSYS_CTNETLINK_EXP,
  49         [NFNLGRP_CONNTRACK_EXP_DESTROY] = NFNL_SUBSYS_CTNETLINK_EXP,
  50 };
  51
  52 void nfnl_lock(__u8 subsys_id)
  53 {
  54         mutex_lock(&table[subsys_id].mutex);
  55 }
  56 EXPORT_SYMBOL_GPL(nfnl_lock);
  57
  58 void nfnl_unlock(__u8 subsys_id)
  59 {
  60         mutex_unlock(&table[subsys_id].mutex);
  61 }
  62 EXPORT_SYMBOL_GPL(nfnl_unlock);
  63
  64 int nfnetlink_subsys_register(const struct nfnetlink_subsystem *n)
  65 {
  66         nfnl_lock(n->subsys_id);
  67         if (table[n->subsys_id].subsys) {
  68                 nfnl_unlock(n->subsys_id);
  69                 return -EBUSY;
  70         }
  71         rcu_assign_pointer(table[n->subsys_id].subsys, n);
  72         nfnl_unlock(n->subsys_id);
  73
  74         return 0;
  75 }
  76 EXPORT_SYMBOL_GPL(nfnetlink_subsys_register);
  77
  78 int nfnetlink_subsys_unregister(const struct nfnetlink_subsystem *n)
  79 {
  80         nfnl_lock(n->subsys_id);
  81         table[n->subsys_id].subsys = NULL;
  82         nfnl_unlock(n->subsys_id);
  83         synchronize_rcu();
  84         return 0;
  85 }
  86 EXPORT_SYMBOL_GPL(nfnetlink_subsys_unregister);
  87
  88 static inline const struct nfnetlink_subsystem *nfnetlink_get_subsys(u_int16_t type)
  89 {
  90         u_int8_t subsys_id = NFNL_SUBSYS_ID(type);
  91
  92         if (subsys_id >= NFNL_SUBSYS_COUNT)
  93                 return NULL;
  94
  95         return rcu_dereference(table[subsys_id].subsys);
  96 }
  97
  98 static inline const struct nfnl_callback *
  99 nfnetlink_find_client(u_int16_t type, const struct nfnetlink_subsystem *ss)
 100 {
 101         u_int8_t cb_id = NFNL_MSG_TYPE(type);
 102
 103         if (cb_id >= ss->cb_count)
 104                 return NULL;
 105
 106         return &ss->cb[cb_id];
 107 }
 108
 109 int nfnetlink_has_listeners(struct net *net, unsigned int group)
 110 {
 111         return netlink_has_listeners(net->nfnl, group);
 112 }
 113 EXPORT_SYMBOL_GPL(nfnetlink_has_listeners);
 114
 115 struct sk_buff *nfnetlink_alloc_skb(struct net *net, unsigned int size,
 116                                     u32 dst_portid, gfp_t gfp_mask)
 117 {
 118         return netlink_alloc_skb(net->nfnl, size, dst_portid, gfp_mask);
 119 }
 120 EXPORT_SYMBOL_GPL(nfnetlink_alloc_skb);
 121
 122 int nfnetlink_send(struct sk_buff *skb, struct net *net, u32 portid,
 123                    unsigned int group, int echo, gfp_t flags)
 124 {
 125         return nlmsg_notify(net->nfnl, skb, portid, group, echo, flags);
 126 }
 127 EXPORT_SYMBOL_GPL(nfnetlink_send);
 128
 129 int nfnetlink_set_err(struct net *net, u32 portid, u32 group, int error)
 130 {
 131         return netlink_set_err(net->nfnl, portid, group, error);
 132 }
 133 EXPORT_SYMBOL_GPL(nfnetlink_set_err);
 134
 135 int nfnetlink_unicast(struct sk_buff *skb, struct net *net, u32 portid,
 136                       int flags)
 137 {
 138         return netlink_unicast(net->nfnl, skb, portid, flags);
 139 }
 140 EXPORT_SYMBOL_GPL(nfnetlink_unicast);
 141
 142 /* Process one complete nfnetlink message. */
 143 static int nfnetlink_rcv_msg(struct sk_buff *skb, struct nlmsghdr *nlh)
 144 {
 145         struct net *net = sock_net(skb->sk);
 146         const struct nfnl_callback *nc;
 147         const struct nfnetlink_subsystem *ss;
 148         int type, err;
 149
 150         /* All the messages must at least contain nfgenmsg */
 151         if (nlmsg_len(nlh) < sizeof(struct nfgenmsg))
 152                 return 0;
 153
 154         type = nlh->nlmsg_type;
 155 replay:
 156         rcu_read_lock();
 157         ss = nfnetlink_get_subsys(type);
 158         if (!ss) {
 159 #ifdef CONFIG_MODULES
 160                 rcu_read_unlock();
 161                 request_module("nfnetlink-subsys-%d", NFNL_SUBSYS_ID(type));
 162                 rcu_read_lock();
 163                 ss = nfnetlink_get_subsys(type);
 164                 if (!ss)
 165 #endif
 166                 {
 167                         rcu_read_unlock();
 168                         return -EINVAL;
 169                 }
 170         }
 171
 172         nc = nfnetlink_find_client(type, ss);
 173         if (!nc) {
 174                 rcu_read_unlock();
 175                 return -EINVAL;
 176         }
 177
 178         {
 179                 int min_len = nlmsg_total_size(sizeof(struct nfgenmsg));
 180                 u_int8_t cb_id = NFNL_MSG_TYPE(nlh->nlmsg_type);
 181                 struct nlattr *cda[ss->cb[cb_id].attr_count + 1];
 182                 struct nlattr *attr = (void *)nlh + min_len;
 183                 int attrlen = nlh->nlmsg_len - min_len;
 184                 __u8 subsys_id = NFNL_SUBSYS_ID(type);
 185
 186                 err = nla_parse(cda, ss->cb[cb_id].attr_count,
 187                                 attr, attrlen, ss->cb[cb_id].policy);
 188                 if (err < 0) {
 189                         rcu_read_unlock();
 190                         return err;
 191                 }
 192
 193                 if (nc->call_rcu) {
 194                         err = nc->call_rcu(net->nfnl, skb, nlh,
 195                                            (const struct nlattr **)cda);
 196                         rcu_read_unlock();
 197                 } else {
 198                         rcu_read_unlock();
 199                         nfnl_lock(subsys_id);
 200                         if (rcu_dereference_protected(table[subsys_id].subsys,
 201                                 lockdep_is_held(&table[subsys_id].mutex)) != ss ||
 202                             nfnetlink_find_client(type, ss) != nc)
 203                                 err = -EAGAIN;
 204                         else if (nc->call)
 205                                 err = nc->call(net->nfnl, skb, nlh,
 206                                                    (const struct nlattr **)cda);
 207                         else
 208                                 err = -EINVAL;
 209                         nfnl_unlock(subsys_id);
 210                 }
 211                 if (err == -EAGAIN)
 212                         goto replay;
 213                 return err;
 214         }
 215 }
 216
 217 static void nfnetlink_rcv_batch(struct sk_buff *skb, struct nlmsghdr *nlh,
 218                                 u_int16_t subsys_id)
 219 {
 220         struct sk_buff *nskb, *oskb = skb;
 221         struct net *net = sock_net(skb->sk);
 222         const struct nfnetlink_subsystem *ss;
 223         const struct nfnl_callback *nc;
 224         bool success = true, done = false;
 225         int err;
 226
 227         if (subsys_id >= NFNL_SUBSYS_COUNT)
 228                 return netlink_ack(skb, nlh, -EINVAL);
 229 replay:
 230         nskb = netlink_skb_clone(oskb, GFP_KERNEL);
 231         if (!nskb)
 232                 return netlink_ack(oskb, nlh, -ENOMEM);
 233
 234         nskb->sk = oskb->sk;
 235         skb = nskb;
 236
 237         nfnl_lock(subsys_id);
 238         ss = rcu_dereference_protected(table[subsys_id].subsys,
 239                                        lockdep_is_held(&table[subsys_id].mutex));
 240         if (!ss) {
 241 #ifdef CONFIG_MODULES
 242                 nfnl_unlock(subsys_id);
 243                 request_module("nfnetlink-subsys-%d", subsys_id);
 244                 nfnl_lock(subsys_id);
 245                 ss = rcu_dereference_protected(table[subsys_id].subsys,
 246                                                lockdep_is_held(&table[subsys_id].mutex));
 247                 if (!ss)
 248 #endif
 249                 {
 250                         nfnl_unlock(subsys_id);
 251                         kfree_skb(nskb);
 252                         return netlink_ack(skb, nlh, -EOPNOTSUPP);
 253                 }
 254         }
 255
 256         if (!ss->commit || !ss->abort) {
 257                 nfnl_unlock(subsys_id);
 258                 kfree_skb(nskb);
 259                 return netlink_ack(skb, nlh, -EOPNOTSUPP);
 260         }
 261
 262         while (skb->len >= nlmsg_total_size(0)) {
 263                 int msglen, type;
 264
 265                 nlh = nlmsg_hdr(skb);
 266                 err = 0;
 267
 268                 if (nlh->nlmsg_len < NLMSG_HDRLEN) {
 269                         err = -EINVAL;
 270                         goto ack;
 271                 }
 272
 273                 /* Only requests are handled by the kernel */
 274                 if (!(nlh->nlmsg_flags & NLM_F_REQUEST)) {
 275                         err = -EINVAL;
 276                         goto ack;
 277                 }
 278
 279                 type = nlh->nlmsg_type;
 280                 if (type == NFNL_MSG_BATCH_BEGIN) {
 281                         /* Malformed: Batch begin twice */
 282                         success = false;
 283                         goto done;
 284                 } else if (type == NFNL_MSG_BATCH_END) {
 285                         done = true;
 286                         goto done;
 287                 } else if (type < NLMSG_MIN_TYPE) {
 288                         err = -EINVAL;
 289                         goto ack;
 290                 }
 291
 292                 /* We only accept a batch with messages for the same
 293                  * subsystem.
 294                  */
 295                 if (NFNL_SUBSYS_ID(type) != subsys_id) {
 296                         err = -EINVAL;
 297                         goto ack;
 298                 }
 299
 300                 nc = nfnetlink_find_client(type, ss);
 301                 if (!nc) {
 302                         err = -EINVAL;
 303                         goto ack;
 304                 }
 305
 306                 {
 307                         int min_len = nlmsg_total_size(sizeof(struct nfgenmsg));
 308                         u_int8_t cb_id = NFNL_MSG_TYPE(nlh->nlmsg_type);
 309                         struct nlattr *cda[ss->cb[cb_id].attr_count + 1];
 310                         struct nlattr *attr = (void *)nlh + min_len;
 311                         int attrlen = nlh->nlmsg_len - min_len;
 312
 313                         err = nla_parse(cda, ss->cb[cb_id].attr_count,
 314                                         attr, attrlen, ss->cb[cb_id].policy);
 315                         if (err < 0)
 316                                 goto ack;
 317
 318                         if (nc->call_batch) {
 319                                 err = nc->call_batch(net->nfnl, skb, nlh,
 320                                                      (const struct nlattr **)cda);
 321                         }
 322
 323                         /* The lock was released to autoload some module, we
 324                          * have to abort and start from scratch using the
 325                          * original skb.
 326                          */
 327                         if (err == -EAGAIN) {
 328                                 ss->abort(skb);
 329                                 nfnl_unlock(subsys_id);
 330                                 kfree_skb(nskb);
 331                                 goto replay;
 332                         }
 333                 }
 334 ack:
 335                 if (nlh->nlmsg_flags & NLM_F_ACK || err) {
 336                         /* We don't stop processing the batch on errors, thus,
 337                          * userspace gets all the errors that the batch
 338                          * triggers.
 339                          */
 340                         netlink_ack(skb, nlh, err);
 341                         if (err)
 342                                 success = false;
 343                 }
 344
 345                 msglen = NLMSG_ALIGN(nlh->nlmsg_len);
 346                 if (msglen > skb->len)
 347                         msglen = skb->len;
 348                 skb_pull(skb, msglen);
 349         }
 350 done:
 351         if (success && done)
 352                 ss->commit(skb);
 353         else
 354                 ss->abort(skb);
 355
 356         nfnl_unlock(subsys_id);
 357         kfree_skb(nskb);
 358 }
 359
 360 static void nfnetlink_rcv(struct sk_buff *skb)
 361 {
 362         struct nlmsghdr *nlh = nlmsg_hdr(skb);
 363         struct net *net = sock_net(skb->sk);
 364         int msglen;
 365
 366         if (nlh->nlmsg_len < NLMSG_HDRLEN ||
 367             skb->len < nlh->nlmsg_len)
 368                 return;
 369
 370         if (!ns_capable(net->user_ns, CAP_NET_ADMIN)) {
 371                 netlink_ack(skb, nlh, -EPERM);
 372                 return;
 373         }
 374
 375         if (nlh->nlmsg_type == NFNL_MSG_BATCH_BEGIN) {
 376                 struct nfgenmsg *nfgenmsg;
 377
 378                 msglen = NLMSG_ALIGN(nlh->nlmsg_len);
 379                 if (msglen > skb->len)
 380                         msglen = skb->len;
 381
 382                 if (nlh->nlmsg_len < NLMSG_HDRLEN ||
 383                     skb->len < NLMSG_HDRLEN + sizeof(struct nfgenmsg))
 384                         return;
 385
 386                 nfgenmsg = nlmsg_data(nlh);
 387                 skb_pull(skb, msglen);
 388                 nfnetlink_rcv_batch(skb, nlh, nfgenmsg->res_id);
 389         } else {
 390                 netlink_rcv_skb(skb, &nfnetlink_rcv_msg);
 391         }
 392 }
 393
 394 #ifdef CONFIG_MODULES
 395 static void nfnetlink_bind(int group)
 396 {
 397         const struct nfnetlink_subsystem *ss;
 398         int type = nfnl_group2type[group];
 399
 400         rcu_read_lock();
 401         ss = nfnetlink_get_subsys(type);
 402         if (!ss) {
 403                 rcu_read_unlock();
 404                 request_module("nfnetlink-subsys-%d", type);
 405                 return;
 406         }
 407         rcu_read_unlock();
 408 }
 409 #endif
 410
 411 static int __net_init nfnetlink_net_init(struct net *net)
 412 {
 413         struct sock *nfnl;
 414         struct netlink_kernel_cfg cfg = {
 415                 .groups = NFNLGRP_MAX,
 416                 .input  = nfnetlink_rcv,
 417 #ifdef CONFIG_MODULES
 418                 .bind   = nfnetlink_bind,
 419 #endif
 420         };
 421
 422         nfnl = netlink_kernel_create(net, NETLINK_NETFILTER, &cfg);
 423         if (!nfnl)
 424                 return -ENOMEM;
 425         net->nfnl_stash = nfnl;
 426         rcu_assign_pointer(net->nfnl, nfnl);
 427         return 0;
 428 }
 429
 430 static void __net_exit nfnetlink_net_exit_batch(struct list_head *net_exit_list)
 431 {
 432         struct net *net;
 433
 434         list_for_each_entry(net, net_exit_list, exit_list)
 435                 RCU_INIT_POINTER(net->nfnl, NULL);
 436         synchronize_net();
 437         list_for_each_entry(net, net_exit_list, exit_list)
 438                 netlink_kernel_release(net->nfnl_stash);
 439 }
 440
 441 static struct pernet_operations nfnetlink_net_ops = {
 442         .init           = nfnetlink_net_init,
 443         .exit_batch     = nfnetlink_net_exit_batch,
 444 };
 445
 446 static int __init nfnetlink_init(void)
 447 {
 448         int i;
 449
 450         for (i=0; i<NFNL_SUBSYS_COUNT; i++)
 451                 mutex_init(&table[i].mutex);
 452
 453         pr_info("Netfilter messages via NETLINK v%s.\n", nfversion);
 454         return register_pernet_subsys(&nfnetlink_net_ops);
 455 }
 456
 457 static void __exit nfnetlink_exit(void)
 458 {
 459         pr_info("Removing netfilter NETLINK layer.\n");
 460         unregister_pernet_subsys(&nfnetlink_net_ops);
 461 }
 462 module_init(nfnetlink_init);
 463 module_exit(nfnetlink_exit);