net/netfilter/nfnetlink_queue_core.c

   1 /*
   2  * This is a module which is used for queueing packets and communicating with
   3  * userspace via nfnetlink.
   4  *
   5  * (C) 2005 by Harald Welte <laforge@netfilter.org>
   6  * (C) 2007 by Patrick McHardy <kaber@trash.net>
   7  *
   8  * Based on the old ipv4-only ip_queue.c:
   9  * (C) 2000-2002 James Morris <jmorris@intercode.com.au>
  10  * (C) 2003-2005 Netfilter Core Team <coreteam@netfilter.org>
  11  *
  12  * This program is free software; you can redistribute it and/or modify
  13  * it under the terms of the GNU General Public License version 2 as
  14  * published by the Free Software Foundation.
  15  *
  16  */
  17 #include <linux/module.h>
  18 #include <linux/skbuff.h>
  19 #include <linux/init.h>
  20 #include <linux/spinlock.h>
  21 #include <linux/slab.h>
  22 #include <linux/notifier.h>
  23 #include <linux/netdevice.h>
  24 #include <linux/netfilter.h>
  25 #include <linux/proc_fs.h>
  26 #include <linux/netfilter_ipv4.h>
  27 #include <linux/netfilter_ipv6.h>
  28 #include <linux/netfilter/nfnetlink.h>
  29 #include <linux/netfilter/nfnetlink_queue.h>
  30 #include <linux/list.h>
  31 #include <net/sock.h>
  32 #include <net/netfilter/nf_queue.h>
  33 #include <net/netns/generic.h>
  34 #include <net/netfilter/nfnetlink_queue.h>
  35
  36 #include <linux/atomic.h>
  37
  38 #ifdef CONFIG_BRIDGE_NETFILTER
  39 #include "../bridge/br_private.h"
  40 #endif
  41
  42 #define NFQNL_QMAX_DEFAULT 1024
  43
  44 struct nfqnl_instance {
  45         struct hlist_node hlist;                /* global list of queues */
  46         struct rcu_head rcu;
  47
  48         int peer_portid;
  49         unsigned int queue_maxlen;
  50         unsigned int copy_range;
  51         unsigned int queue_dropped;
  52         unsigned int queue_user_dropped;
  53
  54
  55         u_int16_t queue_num;                    /* number of this queue */
  56         u_int8_t copy_mode;
  57         u_int32_t flags;                        /* Set using NFQA_CFG_FLAGS */
  58 /*
  59  * Following fields are dirtied for each queued packet,
  60  * keep them in same cache line if possible.
  61  */
  62         spinlock_t      lock;
  63         unsigned int    queue_total;
  64         unsigned int    id_sequence;            /* 'sequence' of pkt ids */
  65         struct list_head queue_list;            /* packets in queue */
  66 };
  67
  68 typedef int (*nfqnl_cmpfn)(struct nf_queue_entry *, unsigned long);
  69
  70 static int nfnl_queue_net_id __read_mostly;
  71
  72 #define INSTANCE_BUCKETS        16
  73 struct nfnl_queue_net {
  74         spinlock_t instances_lock;
  75         struct hlist_head instance_table[INSTANCE_BUCKETS];
  76 };
  77
  78 static struct nfnl_queue_net *nfnl_queue_pernet(struct net *net)
  79 {
  80         return net_generic(net, nfnl_queue_net_id);
  81 }
  82
  83 static inline u_int8_t instance_hashfn(u_int16_t queue_num)
  84 {
  85         return ((queue_num >> 8) ^ queue_num) % INSTANCE_BUCKETS;
  86 }
  87
  88 static struct nfqnl_instance *
  89 instance_lookup(struct nfnl_queue_net *q, u_int16_t queue_num)
  90 {
  91         struct hlist_head *head;
  92         struct nfqnl_instance *inst;
  93
  94         head = &q->instance_table[instance_hashfn(queue_num)];
  95         hlist_for_each_entry_rcu(inst, head, hlist) {
  96                 if (inst->queue_num == queue_num)
  97                         return inst;
  98         }
  99         return NULL;
 100 }
 101
 102 static struct nfqnl_instance *
 103 instance_create(struct nfnl_queue_net *q, u_int16_t queue_num,
 104                 int portid)
 105 {
 106         struct nfqnl_instance *inst;
 107         unsigned int h;
 108         int err;
 109
 110         spin_lock(&q->instances_lock);
 111         if (instance_lookup(q, queue_num)) {
 112                 err = -EEXIST;
 113                 goto out_unlock;
 114         }
 115
 116         inst = kzalloc(sizeof(*inst), GFP_ATOMIC);
 117         if (!inst) {
 118                 err = -ENOMEM;
 119                 goto out_unlock;
 120         }
 121
 122         inst->queue_num = queue_num;
 123         inst->peer_portid = portid;
 124         inst->queue_maxlen = NFQNL_QMAX_DEFAULT;
 125         inst->copy_range = 0xffff;
 126         inst->copy_mode = NFQNL_COPY_NONE;
 127         spin_lock_init(&inst->lock);
 128         INIT_LIST_HEAD(&inst->queue_list);
 129
 130         if (!try_module_get(THIS_MODULE)) {
 131                 err = -EAGAIN;
 132                 goto out_free;
 133         }
 134
 135         h = instance_hashfn(queue_num);
 136         hlist_add_head_rcu(&inst->hlist, &q->instance_table[h]);
 137
 138         spin_unlock(&q->instances_lock);
 139
 140         return inst;
 141
 142 out_free:
 143         kfree(inst);
 144 out_unlock:
 145         spin_unlock(&q->instances_lock);
 146         return ERR_PTR(err);
 147 }
 148
 149 static void nfqnl_flush(struct nfqnl_instance *queue, nfqnl_cmpfn cmpfn,
 150                         unsigned long data);
 151
 152 static void
 153 instance_destroy_rcu(struct rcu_head *head)
 154 {
 155         struct nfqnl_instance *inst = container_of(head, struct nfqnl_instance,
 156                                                    rcu);
 157
 158         nfqnl_flush(inst, NULL, 0);
 159         kfree(inst);
 160         module_put(THIS_MODULE);
 161 }
 162
 163 static void
 164 __instance_destroy(struct nfqnl_instance *inst)
 165 {
 166         hlist_del_rcu(&inst->hlist);
 167         call_rcu(&inst->rcu, instance_destroy_rcu);
 168 }
 169
 170 static void
 171 instance_destroy(struct nfnl_queue_net *q, struct nfqnl_instance *inst)
 172 {
 173         spin_lock(&q->instances_lock);
 174         __instance_destroy(inst);
 175         spin_unlock(&q->instances_lock);
 176 }
 177
 178 static inline void
 179 __enqueue_entry(struct nfqnl_instance *queue, struct nf_queue_entry *entry)
 180 {
 181        list_add_tail(&entry->list, &queue->queue_list);
 182        queue->queue_total++;
 183 }
 184
 185 static void
 186 __dequeue_entry(struct nfqnl_instance *queue, struct nf_queue_entry *entry)
 187 {
 188         list_del(&entry->list);
 189         queue->queue_total--;
 190 }
 191
 192 static struct nf_queue_entry *
 193 find_dequeue_entry(struct nfqnl_instance *queue, unsigned int id)
 194 {
 195         struct nf_queue_entry *entry = NULL, *i;
 196
 197         spin_lock_bh(&queue->lock);
 198
 199         list_for_each_entry(i, &queue->queue_list, list) {
 200                 if (i->id == id) {
 201                         entry = i;
 202                         break;
 203                 }
 204         }
 205
 206         if (entry)
 207                 __dequeue_entry(queue, entry);
 208
 209         spin_unlock_bh(&queue->lock);
 210
 211         return entry;
 212 }
 213
 214 static void
 215 nfqnl_flush(struct nfqnl_instance *queue, nfqnl_cmpfn cmpfn, unsigned long data)
 216 {
 217         struct nf_queue_entry *entry, *next;
 218
 219         spin_lock_bh(&queue->lock);
 220         list_for_each_entry_safe(entry, next, &queue->queue_list, list) {
 221                 if (!cmpfn || cmpfn(entry, data)) {
 222                         list_del(&entry->list);
 223                         queue->queue_total--;
 224                         nf_reinject(entry, NF_DROP);
 225                 }
 226         }
 227         spin_unlock_bh(&queue->lock);
 228 }
 229
 230 static void
 231 nfqnl_zcopy(struct sk_buff *to, const struct sk_buff *from, int len, int hlen)
 232 {
 233         int i, j = 0;
 234         int plen = 0; /* length of skb->head fragment */
 235         struct page *page;
 236         unsigned int offset;
 237
 238         /* dont bother with small payloads */
 239         if (len <= skb_tailroom(to)) {
 240                 skb_copy_bits(from, 0, skb_put(to, len), len);
 241                 return;
 242         }
 243
 244         if (hlen) {
 245                 skb_copy_bits(from, 0, skb_put(to, hlen), hlen);
 246                 len -= hlen;
 247         } else {
 248                 plen = min_t(int, skb_headlen(from), len);
 249                 if (plen) {
 250                         page = virt_to_head_page(from->head);
 251                         offset = from->data - (unsigned char *)page_address(page);
 252                         __skb_fill_page_desc(to, 0, page, offset, plen);
 253                         get_page(page);
 254                         j = 1;
 255                         len -= plen;
 256                 }
 257         }
 258
 259         to->truesize += len + plen;
 260         to->len += len + plen;
 261         to->data_len += len + plen;
 262
 263         for (i = 0; i < skb_shinfo(from)->nr_frags; i++) {
 264                 if (!len)
 265                         break;
 266                 skb_shinfo(to)->frags[j] = skb_shinfo(from)->frags[i];
 267                 skb_shinfo(to)->frags[j].size = min_t(int, skb_shinfo(to)->frags[j].size, len);
 268                 len -= skb_shinfo(to)->frags[j].size;
 269                 skb_frag_ref(to, j);
 270                 j++;
 271         }
 272         skb_shinfo(to)->nr_frags = j;
 273 }
 274
 275 static struct sk_buff *
 276 nfqnl_build_packet_message(struct nfqnl_instance *queue,
 277                            struct nf_queue_entry *entry,
 278                            __be32 **packet_id_ptr)
 279 {
 280         size_t size;
 281         size_t data_len = 0, cap_len = 0;
 282         int hlen = 0;
 283         struct sk_buff *skb;
 284         struct nlattr *nla;
 285         struct nfqnl_msg_packet_hdr *pmsg;
 286         struct nlmsghdr *nlh;
 287         struct nfgenmsg *nfmsg;
 288         struct sk_buff *entskb = entry->skb;
 289         struct net_device *indev;
 290         struct net_device *outdev;
 291         struct nf_conn *ct = NULL;
 292         enum ip_conntrack_info uninitialized_var(ctinfo);
 293
 294         size =    NLMSG_SPACE(sizeof(struct nfgenmsg))
 295                 + nla_total_size(sizeof(struct nfqnl_msg_packet_hdr))
 296                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
 297                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
 298 #ifdef CONFIG_BRIDGE_NETFILTER
 299                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
 300                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
 301 #endif
 302                 + nla_total_size(sizeof(u_int32_t))     /* mark */
 303                 + nla_total_size(sizeof(struct nfqnl_msg_packet_hw))
 304                 + nla_total_size(sizeof(u_int32_t));    /* cap_len */
 305
 306         if (entskb->tstamp.tv64)
 307                 size += nla_total_size(sizeof(struct nfqnl_msg_packet_timestamp));
 308
 309         outdev = entry->outdev;
 310
 311         switch ((enum nfqnl_config_mode)ACCESS_ONCE(queue->copy_mode)) {
 312         case NFQNL_COPY_META:
 313         case NFQNL_COPY_NONE:
 314                 break;
 315
 316         case NFQNL_COPY_PACKET:
 317                 if (entskb->ip_summed == CHECKSUM_PARTIAL &&
 318                     skb_checksum_help(entskb))
 319                         return NULL;
 320
 321                 data_len = ACCESS_ONCE(queue->copy_range);
 322                 if (data_len == 0 || data_len > entskb->len)
 323                         data_len = entskb->len;
 324
 325
 326                 if (!entskb->head_frag ||
 327                     skb_headlen(entskb) < L1_CACHE_BYTES ||
 328                     skb_shinfo(entskb)->nr_frags >= MAX_SKB_FRAGS)
 329                         hlen = skb_headlen(entskb);
 330
 331                 if (skb_has_frag_list(entskb))
 332                         hlen = entskb->len;
 333                 hlen = min_t(int, data_len, hlen);
 334                 size += sizeof(struct nlattr) + hlen;
 335                 cap_len = entskb->len;
 336                 break;
 337         }
 338
 339         if (queue->flags & NFQA_CFG_F_CONNTRACK)
 340                 ct = nfqnl_ct_get(entskb, &size, &ctinfo);
 341
 342         skb = alloc_skb(size, GFP_ATOMIC);
 343         if (!skb)
 344                 return NULL;
 345
 346         nlh = nlmsg_put(skb, 0, 0,
 347                         NFNL_SUBSYS_QUEUE << 8 | NFQNL_MSG_PACKET,
 348                         sizeof(struct nfgenmsg), 0);
 349         if (!nlh) {
 350                 kfree_skb(skb);
 351                 return NULL;
 352         }
 353         nfmsg = nlmsg_data(nlh);
 354         nfmsg->nfgen_family = entry->pf;
 355         nfmsg->version = NFNETLINK_V0;
 356         nfmsg->res_id = htons(queue->queue_num);
 357
 358         nla = __nla_reserve(skb, NFQA_PACKET_HDR, sizeof(*pmsg));
 359         pmsg = nla_data(nla);
 360         pmsg->hw_protocol       = entskb->protocol;
 361         pmsg->hook              = entry->hook;
 362         *packet_id_ptr          = &pmsg->packet_id;
 363
 364         indev = entry->indev;
 365         if (indev) {
 366 #ifndef CONFIG_BRIDGE_NETFILTER
 367                 if (nla_put_be32(skb, NFQA_IFINDEX_INDEV, htonl(indev->ifindex)))
 368                         goto nla_put_failure;
 369 #else
 370                 if (entry->pf == PF_BRIDGE) {
 371                         /* Case 1: indev is physical input device, we need to
 372                          * look for bridge group (when called from
 373                          * netfilter_bridge) */
 374                         if (nla_put_be32(skb, NFQA_IFINDEX_PHYSINDEV,
 375                                          htonl(indev->ifindex)) ||
 376                         /* this is the bridge group "brX" */
 377                         /* rcu_read_lock()ed by __nf_queue */
 378                             nla_put_be32(skb, NFQA_IFINDEX_INDEV,
 379                                          htonl(br_port_get_rcu(indev)->br->dev->ifindex)))
 380                                 goto nla_put_failure;
 381                 } else {
 382                         /* Case 2: indev is bridge group, we need to look for
 383                          * physical device (when called from ipv4) */
 384                         if (nla_put_be32(skb, NFQA_IFINDEX_INDEV,
 385                                          htonl(indev->ifindex)))
 386                                 goto nla_put_failure;
 387                         if (entskb->nf_bridge && entskb->nf_bridge->physindev &&
 388                             nla_put_be32(skb, NFQA_IFINDEX_PHYSINDEV,
 389                                          htonl(entskb->nf_bridge->physindev->ifindex)))
 390                                 goto nla_put_failure;
 391                 }
 392 #endif
 393         }
 394
 395         if (outdev) {
 396 #ifndef CONFIG_BRIDGE_NETFILTER
 397                 if (nla_put_be32(skb, NFQA_IFINDEX_OUTDEV, htonl(outdev->ifindex)))
 398                         goto nla_put_failure;
 399 #else
 400                 if (entry->pf == PF_BRIDGE) {
 401                         /* Case 1: outdev is physical output device, we need to
 402                          * look for bridge group (when called from
 403                          * netfilter_bridge) */
 404                         if (nla_put_be32(skb, NFQA_IFINDEX_PHYSOUTDEV,
 405                                          htonl(outdev->ifindex)) ||
 406                         /* this is the bridge group "brX" */
 407                         /* rcu_read_lock()ed by __nf_queue */
 408                             nla_put_be32(skb, NFQA_IFINDEX_OUTDEV,
 409                                          htonl(br_port_get_rcu(outdev)->br->dev->ifindex)))
 410                                 goto nla_put_failure;
 411                 } else {
 412                         /* Case 2: outdev is bridge group, we need to look for
 413                          * physical output device (when called from ipv4) */
 414                         if (nla_put_be32(skb, NFQA_IFINDEX_OUTDEV,
 415                                          htonl(outdev->ifindex)))
 416                                 goto nla_put_failure;
 417                         if (entskb->nf_bridge && entskb->nf_bridge->physoutdev &&
 418                             nla_put_be32(skb, NFQA_IFINDEX_PHYSOUTDEV,
 419                                          htonl(entskb->nf_bridge->physoutdev->ifindex)))
 420                                 goto nla_put_failure;
 421                 }
 422 #endif
 423         }
 424
 425         if (entskb->mark &&
 426             nla_put_be32(skb, NFQA_MARK, htonl(entskb->mark)))
 427                 goto nla_put_failure;
 428
 429         if (indev && entskb->dev &&
 430             entskb->mac_header != entskb->network_header) {
 431                 struct nfqnl_msg_packet_hw phw;
 432                 int len = dev_parse_header(entskb, phw.hw_addr);
 433                 if (len) {
 434                         phw.hw_addrlen = htons(len);
 435                         if (nla_put(skb, NFQA_HWADDR, sizeof(phw), &phw))
 436                                 goto nla_put_failure;
 437                 }
 438         }
 439
 440         if (entskb->tstamp.tv64) {
 441                 struct nfqnl_msg_packet_timestamp ts;
 442                 struct timeval tv = ktime_to_timeval(entskb->tstamp);
 443                 ts.sec = cpu_to_be64(tv.tv_sec);
 444                 ts.usec = cpu_to_be64(tv.tv_usec);
 445
 446                 if (nla_put(skb, NFQA_TIMESTAMP, sizeof(ts), &ts))
 447                         goto nla_put_failure;
 448         }
 449
 450         if (ct && nfqnl_ct_put(skb, ct, ctinfo) < 0)
 451                 goto nla_put_failure;
 452
 453         if (cap_len > 0 && nla_put_be32(skb, NFQA_CAP_LEN, htonl(cap_len)))
 454                 goto nla_put_failure;
 455
 456         if (data_len) {
 457                 struct nlattr *nla;
 458
 459                 if (skb_tailroom(skb) < sizeof(*nla) + hlen)
 460                         goto nla_put_failure;
 461
 462                 nla = (struct nlattr *)skb_put(skb, sizeof(*nla));
 463                 nla->nla_type = NFQA_PAYLOAD;
 464                 nla->nla_len = nla_attr_size(data_len);
 465
 466                 nfqnl_zcopy(skb, entskb, data_len, hlen);
 467         }
 468
 469         nlh->nlmsg_len = skb->len;
 470         return skb;
 471
 472 nla_put_failure:
 473         kfree_skb(skb);
 474         net_err_ratelimited("nf_queue: error creating packet message\n");
 475         return NULL;
 476 }
 477
 478 static int
 479 nfqnl_enqueue_packet(struct nf_queue_entry *entry, unsigned int queuenum)
 480 {
 481         struct sk_buff *nskb;
 482         struct nfqnl_instance *queue;
 483         int err = -ENOBUFS;
 484         __be32 *packet_id_ptr;
 485         int failopen = 0;
 486         struct net *net = dev_net(entry->indev ?
 487                                   entry->indev : entry->outdev);
 488         struct nfnl_queue_net *q = nfnl_queue_pernet(net);
 489
 490         /* rcu_read_lock()ed by nf_hook_slow() */
 491         queue = instance_lookup(q, queuenum);
 492         if (!queue) {
 493                 err = -ESRCH;
 494                 goto err_out;
 495         }
 496
 497         if (queue->copy_mode == NFQNL_COPY_NONE) {
 498                 err = -EINVAL;
 499                 goto err_out;
 500         }
 501
 502         nskb = nfqnl_build_packet_message(queue, entry, &packet_id_ptr);
 503         if (nskb == NULL) {
 504                 err = -ENOMEM;
 505                 goto err_out;
 506         }
 507         spin_lock_bh(&queue->lock);
 508
 509         if (!queue->peer_portid) {
 510                 err = -EINVAL;
 511                 goto err_out_free_nskb;
 512         }
 513         if (queue->queue_total >= queue->queue_maxlen) {
 514                 if (queue->flags & NFQA_CFG_F_FAIL_OPEN) {
 515                         failopen = 1;
 516                         err = 0;
 517                 } else {
 518                         queue->queue_dropped++;
 519                         net_warn_ratelimited("nf_queue: full at %d entries, dropping packets(s)\n",
 520                                              queue->queue_total);
 521                 }
 522                 goto err_out_free_nskb;
 523         }
 524         entry->id = ++queue->id_sequence;
 525         *packet_id_ptr = htonl(entry->id);
 526
 527         /* nfnetlink_unicast will either free the nskb or add it to a socket */
 528         err = nfnetlink_unicast(nskb, net, queue->peer_portid, MSG_DONTWAIT);
 529         if (err < 0) {
 530                 queue->queue_user_dropped++;
 531                 goto err_out_unlock;
 532         }
 533
 534         __enqueue_entry(queue, entry);
 535
 536         spin_unlock_bh(&queue->lock);
 537         return 0;
 538
 539 err_out_free_nskb:
 540         kfree_skb(nskb);
 541 err_out_unlock:
 542         spin_unlock_bh(&queue->lock);
 543         if (failopen)
 544                 nf_reinject(entry, NF_ACCEPT);
 545 err_out:
 546         return err;
 547 }
 548
 549 static int
 550 nfqnl_mangle(void *data, int data_len, struct nf_queue_entry *e, int diff)
 551 {
 552         struct sk_buff *nskb;
 553
 554         if (diff < 0) {
 555                 if (pskb_trim(e->skb, data_len))
 556                         return -ENOMEM;
 557         } else if (diff > 0) {
 558                 if (data_len > 0xFFFF)
 559                         return -EINVAL;
 560                 if (diff > skb_tailroom(e->skb)) {
 561                         nskb = skb_copy_expand(e->skb, skb_headroom(e->skb),
 562                                                diff, GFP_ATOMIC);
 563                         if (!nskb) {
 564                                 printk(KERN_WARNING "nf_queue: OOM "
 565                                       "in mangle, dropping packet\n");
 566                                 return -ENOMEM;
 567                         }
 568                         kfree_skb(e->skb);
 569                         e->skb = nskb;
 570                 }
 571                 skb_put(e->skb, diff);
 572         }
 573         if (!skb_make_writable(e->skb, data_len))
 574                 return -ENOMEM;
 575         skb_copy_to_linear_data(e->skb, data, data_len);
 576         e->skb->ip_summed = CHECKSUM_NONE;
 577         return 0;
 578 }
 579
 580 static int
 581 nfqnl_set_mode(struct nfqnl_instance *queue,
 582                unsigned char mode, unsigned int range)
 583 {
 584         int status = 0;
 585
 586         spin_lock_bh(&queue->lock);
 587         switch (mode) {
 588         case NFQNL_COPY_NONE:
 589         case NFQNL_COPY_META:
 590                 queue->copy_mode = mode;
 591                 queue->copy_range = 0;
 592                 break;
 593
 594         case NFQNL_COPY_PACKET:
 595                 queue->copy_mode = mode;
 596                 /* We're using struct nlattr which has 16bit nla_len. Note that
 597                  * nla_len includes the header length. Thus, the maximum packet
 598                  * length that we support is 65531 bytes. We send truncated
 599                  * packets if the specified length is larger than that.
 600                  */
 601                 if (range > 0xffff - NLA_HDRLEN)
 602                         queue->copy_range = 0xffff - NLA_HDRLEN;
 603                 else
 604                         queue->copy_range = range;
 605                 break;
 606
 607         default:
 608                 status = -EINVAL;
 609
 610         }
 611         spin_unlock_bh(&queue->lock);
 612
 613         return status;
 614 }
 615
 616 static int
 617 dev_cmp(struct nf_queue_entry *entry, unsigned long ifindex)
 618 {
 619         if (entry->indev)
 620                 if (entry->indev->ifindex == ifindex)
 621                         return 1;
 622         if (entry->outdev)
 623                 if (entry->outdev->ifindex == ifindex)
 624                         return 1;
 625 #ifdef CONFIG_BRIDGE_NETFILTER
 626         if (entry->skb->nf_bridge) {
 627                 if (entry->skb->nf_bridge->physindev &&
 628                     entry->skb->nf_bridge->physindev->ifindex == ifindex)
 629                         return 1;
 630                 if (entry->skb->nf_bridge->physoutdev &&
 631                     entry->skb->nf_bridge->physoutdev->ifindex == ifindex)
 632                         return 1;
 633         }
 634 #endif
 635         return 0;
 636 }
 637
 638 /* drop all packets with either indev or outdev == ifindex from all queue
 639  * instances */
 640 static void
 641 nfqnl_dev_drop(struct net *net, int ifindex)
 642 {
 643         int i;
 644         struct nfnl_queue_net *q = nfnl_queue_pernet(net);
 645
 646         rcu_read_lock();
 647
 648         for (i = 0; i < INSTANCE_BUCKETS; i++) {
 649                 struct nfqnl_instance *inst;
 650                 struct hlist_head *head = &q->instance_table[i];
 651
 652                 hlist_for_each_entry_rcu(inst, head, hlist)
 653                         nfqnl_flush(inst, dev_cmp, ifindex);
 654         }
 655
 656         rcu_read_unlock();
 657 }
 658
 659 #define RCV_SKB_FAIL(err) do { netlink_ack(skb, nlh, (err)); return; } while (0)
 660
 661 static int
 662 nfqnl_rcv_dev_event(struct notifier_block *this,
 663                     unsigned long event, void *ptr)
 664 {
 665         struct net_device *dev = ptr;
 666
 667         /* Drop any packets associated with the downed device */
 668         if (event == NETDEV_DOWN)
 669                 nfqnl_dev_drop(dev_net(dev), dev->ifindex);
 670         return NOTIFY_DONE;
 671 }
 672
 673 static struct notifier_block nfqnl_dev_notifier = {
 674         .notifier_call  = nfqnl_rcv_dev_event,
 675 };
 676
 677 static int
 678 nfqnl_rcv_nl_event(struct notifier_block *this,
 679                    unsigned long event, void *ptr)
 680 {
 681         struct netlink_notify *n = ptr;
 682         struct nfnl_queue_net *q = nfnl_queue_pernet(n->net);
 683
 684         if (event == NETLINK_URELEASE && n->protocol == NETLINK_NETFILTER) {
 685                 int i;
 686
 687                 /* destroy all instances for this portid */
 688                 spin_lock(&q->instances_lock);
 689                 for (i = 0; i < INSTANCE_BUCKETS; i++) {
 690                         struct hlist_node *t2;
 691                         struct nfqnl_instance *inst;
 692                         struct hlist_head *head = &q->instance_table[i];
 693
 694                         hlist_for_each_entry_safe(inst, t2, head, hlist) {
 695                                 if (n->portid == inst->peer_portid)
 696                                         __instance_destroy(inst);
 697                         }
 698                 }
 699                 spin_unlock(&q->instances_lock);
 700         }
 701         return NOTIFY_DONE;
 702 }
 703
 704 static struct notifier_block nfqnl_rtnl_notifier = {
 705         .notifier_call  = nfqnl_rcv_nl_event,
 706 };
 707
 708 static const struct nla_policy nfqa_verdict_policy[NFQA_MAX+1] = {
 709         [NFQA_VERDICT_HDR]      = { .len = sizeof(struct nfqnl_msg_verdict_hdr) },
 710         [NFQA_MARK]             = { .type = NLA_U32 },
 711         [NFQA_PAYLOAD]          = { .type = NLA_UNSPEC },
 712         [NFQA_CT]               = { .type = NLA_UNSPEC },
 713 };
 714
 715 static const struct nla_policy nfqa_verdict_batch_policy[NFQA_MAX+1] = {
 716         [NFQA_VERDICT_HDR]      = { .len = sizeof(struct nfqnl_msg_verdict_hdr) },
 717         [NFQA_MARK]             = { .type = NLA_U32 },
 718 };
 719
 720 static struct nfqnl_instance *
 721 verdict_instance_lookup(struct nfnl_queue_net *q, u16 queue_num, int nlportid)
 722 {
 723         struct nfqnl_instance *queue;
 724
 725         queue = instance_lookup(q, queue_num);
 726         if (!queue)
 727                 return ERR_PTR(-ENODEV);
 728
 729         if (queue->peer_portid != nlportid)
 730                 return ERR_PTR(-EPERM);
 731
 732         return queue;
 733 }
 734
 735 static struct nfqnl_msg_verdict_hdr*
 736 verdicthdr_get(const struct nlattr * const nfqa[])
 737 {
 738         struct nfqnl_msg_verdict_hdr *vhdr;
 739         unsigned int verdict;
 740
 741         if (!nfqa[NFQA_VERDICT_HDR])
 742                 return NULL;
 743
 744         vhdr = nla_data(nfqa[NFQA_VERDICT_HDR]);
 745         verdict = ntohl(vhdr->verdict) & NF_VERDICT_MASK;
 746         if (verdict > NF_MAX_VERDICT || verdict == NF_STOLEN)
 747                 return NULL;
 748         return vhdr;
 749 }
 750
 751 static int nfq_id_after(unsigned int id, unsigned int max)
 752 {
 753         return (int)(id - max) > 0;
 754 }
 755
 756 static int
 757 nfqnl_recv_verdict_batch(struct sock *ctnl, struct sk_buff *skb,
 758                    const struct nlmsghdr *nlh,
 759                    const struct nlattr * const nfqa[])
 760 {
 761         struct nfgenmsg *nfmsg = nlmsg_data(nlh);
 762         struct nf_queue_entry *entry, *tmp;
 763         unsigned int verdict, maxid;
 764         struct nfqnl_msg_verdict_hdr *vhdr;
 765         struct nfqnl_instance *queue;
 766         LIST_HEAD(batch_list);
 767         u16 queue_num = ntohs(nfmsg->res_id);
 768
 769         struct net *net = sock_net(ctnl);
 770         struct nfnl_queue_net *q = nfnl_queue_pernet(net);
 771
 772         queue = verdict_instance_lookup(q, queue_num,
 773                                         NETLINK_CB(skb).portid);
 774         if (IS_ERR(queue))
 775                 return PTR_ERR(queue);
 776
 777         vhdr = verdicthdr_get(nfqa);
 778         if (!vhdr)
 779                 return -EINVAL;
 780
 781         verdict = ntohl(vhdr->verdict);
 782         maxid = ntohl(vhdr->id);
 783
 784         spin_lock_bh(&queue->lock);
 785
 786         list_for_each_entry_safe(entry, tmp, &queue->queue_list, list) {
 787                 if (nfq_id_after(entry->id, maxid))
 788                         break;
 789                 __dequeue_entry(queue, entry);
 790                 list_add_tail(&entry->list, &batch_list);
 791         }
 792
 793         spin_unlock_bh(&queue->lock);
 794
 795         if (list_empty(&batch_list))
 796                 return -ENOENT;
 797
 798         list_for_each_entry_safe(entry, tmp, &batch_list, list) {
 799                 if (nfqa[NFQA_MARK])
 800                         entry->skb->mark = ntohl(nla_get_be32(nfqa[NFQA_MARK]));
 801                 nf_reinject(entry, verdict);
 802         }
 803         return 0;
 804 }
 805
 806 static int
 807 nfqnl_recv_verdict(struct sock *ctnl, struct sk_buff *skb,
 808                    const struct nlmsghdr *nlh,
 809                    const struct nlattr * const nfqa[])
 810 {
 811         struct nfgenmsg *nfmsg = nlmsg_data(nlh);
 812         u_int16_t queue_num = ntohs(nfmsg->res_id);
 813
 814         struct nfqnl_msg_verdict_hdr *vhdr;
 815         struct nfqnl_instance *queue;
 816         unsigned int verdict;
 817         struct nf_queue_entry *entry;
 818         enum ip_conntrack_info uninitialized_var(ctinfo);
 819         struct nf_conn *ct = NULL;
 820
 821         struct net *net = sock_net(ctnl);
 822         struct nfnl_queue_net *q = nfnl_queue_pernet(net);
 823
 824         queue = instance_lookup(q, queue_num);
 825         if (!queue)
 826                 queue = verdict_instance_lookup(q, queue_num,
 827                                                 NETLINK_CB(skb).portid);
 828         if (IS_ERR(queue))
 829                 return PTR_ERR(queue);
 830
 831         vhdr = verdicthdr_get(nfqa);
 832         if (!vhdr)
 833                 return -EINVAL;
 834
 835         verdict = ntohl(vhdr->verdict);
 836
 837         entry = find_dequeue_entry(queue, ntohl(vhdr->id));
 838         if (entry == NULL)
 839                 return -ENOENT;
 840
 841         rcu_read_lock();
 842         if (nfqa[NFQA_CT] && (queue->flags & NFQA_CFG_F_CONNTRACK))
 843                 ct = nfqnl_ct_parse(entry->skb, nfqa[NFQA_CT], &ctinfo);
 844
 845         if (nfqa[NFQA_PAYLOAD]) {
 846                 u16 payload_len = nla_len(nfqa[NFQA_PAYLOAD]);
 847                 int diff = payload_len - entry->skb->len;
 848
 849                 if (nfqnl_mangle(nla_data(nfqa[NFQA_PAYLOAD]),
 850                                  payload_len, entry, diff) < 0)
 851                         verdict = NF_DROP;
 852
 853                 if (ct)
 854                         nfqnl_ct_seq_adjust(skb, ct, ctinfo, diff);
 855         }
 856         rcu_read_unlock();
 857
 858         if (nfqa[NFQA_MARK])
 859                 entry->skb->mark = ntohl(nla_get_be32(nfqa[NFQA_MARK]));
 860
 861         nf_reinject(entry, verdict);
 862         return 0;
 863 }
 864
 865 static int
 866 nfqnl_recv_unsupp(struct sock *ctnl, struct sk_buff *skb,
 867                   const struct nlmsghdr *nlh,
 868                   const struct nlattr * const nfqa[])
 869 {
 870         return -ENOTSUPP;
 871 }
 872
 873 static const struct nla_policy nfqa_cfg_policy[NFQA_CFG_MAX+1] = {
 874         [NFQA_CFG_CMD]          = { .len = sizeof(struct nfqnl_msg_config_cmd) },
 875         [NFQA_CFG_PARAMS]       = { .len = sizeof(struct nfqnl_msg_config_params) },
 876 };
 877
 878 static const struct nf_queue_handler nfqh = {
 879         .outfn  = &nfqnl_enqueue_packet,
 880 };
 881
 882 static int
 883 nfqnl_recv_config(struct sock *ctnl, struct sk_buff *skb,
 884                   const struct nlmsghdr *nlh,
 885                   const struct nlattr * const nfqa[])
 886 {
 887         struct nfgenmsg *nfmsg = nlmsg_data(nlh);
 888         u_int16_t queue_num = ntohs(nfmsg->res_id);
 889         struct nfqnl_instance *queue;
 890         struct nfqnl_msg_config_cmd *cmd = NULL;
 891         struct net *net = sock_net(ctnl);
 892         struct nfnl_queue_net *q = nfnl_queue_pernet(net);
 893         int ret = 0;
 894
 895         if (nfqa[NFQA_CFG_CMD]) {
 896                 cmd = nla_data(nfqa[NFQA_CFG_CMD]);
 897
 898                 /* Obsolete commands without queue context */
 899                 switch (cmd->command) {
 900                 case NFQNL_CFG_CMD_PF_BIND: return 0;
 901                 case NFQNL_CFG_CMD_PF_UNBIND: return 0;
 902                 }
 903         }
 904
 905         rcu_read_lock();
 906         queue = instance_lookup(q, queue_num);
 907         if (queue && queue->peer_portid != NETLINK_CB(skb).portid) {
 908                 ret = -EPERM;
 909                 goto err_out_unlock;
 910         }
 911
 912         if (cmd != NULL) {
 913                 switch (cmd->command) {
 914                 case NFQNL_CFG_CMD_BIND:
 915                         if (queue) {
 916                                 ret = -EBUSY;
 917                                 goto err_out_unlock;
 918                         }
 919                         queue = instance_create(q, queue_num,
 920                                                 NETLINK_CB(skb).portid);
 921                         if (IS_ERR(queue)) {
 922                                 ret = PTR_ERR(queue);
 923                                 goto err_out_unlock;
 924                         }
 925                         break;
 926                 case NFQNL_CFG_CMD_UNBIND:
 927                         if (!queue) {
 928                                 ret = -ENODEV;
 929                                 goto err_out_unlock;
 930                         }
 931                         instance_destroy(q, queue);
 932                         break;
 933                 case NFQNL_CFG_CMD_PF_BIND:
 934                 case NFQNL_CFG_CMD_PF_UNBIND:
 935                         break;
 936                 default:
 937                         ret = -ENOTSUPP;
 938                         break;
 939                 }
 940         }
 941
 942         if (nfqa[NFQA_CFG_PARAMS]) {
 943                 struct nfqnl_msg_config_params *params;
 944
 945                 if (!queue) {
 946                         ret = -ENODEV;
 947                         goto err_out_unlock;
 948                 }
 949                 params = nla_data(nfqa[NFQA_CFG_PARAMS]);
 950                 nfqnl_set_mode(queue, params->copy_mode,
 951                                 ntohl(params->copy_range));
 952         }
 953
 954         if (nfqa[NFQA_CFG_QUEUE_MAXLEN]) {
 955                 __be32 *queue_maxlen;
 956
 957                 if (!queue) {
 958                         ret = -ENODEV;
 959                         goto err_out_unlock;
 960                 }
 961                 queue_maxlen = nla_data(nfqa[NFQA_CFG_QUEUE_MAXLEN]);
 962                 spin_lock_bh(&queue->lock);
 963                 queue->queue_maxlen = ntohl(*queue_maxlen);
 964                 spin_unlock_bh(&queue->lock);
 965         }
 966
 967         if (nfqa[NFQA_CFG_FLAGS]) {
 968                 __u32 flags, mask;
 969
 970                 if (!queue) {
 971                         ret = -ENODEV;
 972                         goto err_out_unlock;
 973                 }
 974
 975                 if (!nfqa[NFQA_CFG_MASK]) {
 976                         /* A mask is needed to specify which flags are being
 977                          * changed.
 978                          */
 979                         ret = -EINVAL;
 980                         goto err_out_unlock;
 981                 }
 982
 983                 flags = ntohl(nla_get_be32(nfqa[NFQA_CFG_FLAGS]));
 984                 mask = ntohl(nla_get_be32(nfqa[NFQA_CFG_MASK]));
 985
 986                 if (flags >= NFQA_CFG_F_MAX) {
 987                         ret = -EOPNOTSUPP;
 988                         goto err_out_unlock;
 989                 }
 990
 991                 spin_lock_bh(&queue->lock);
 992                 queue->flags &= ~mask;
 993                 queue->flags |= flags & mask;
 994                 spin_unlock_bh(&queue->lock);
 995         }
 996
 997 err_out_unlock:
 998         rcu_read_unlock();
 999         return ret;
1000 }
1001
1002 static const struct nfnl_callback nfqnl_cb[NFQNL_MSG_MAX] = {
1003         [NFQNL_MSG_PACKET]      = { .call_rcu = nfqnl_recv_unsupp,
1004                                     .attr_count = NFQA_MAX, },
1005         [NFQNL_MSG_VERDICT]     = { .call_rcu = nfqnl_recv_verdict,
1006                                     .attr_count = NFQA_MAX,
1007                                     .policy = nfqa_verdict_policy },
1008         [NFQNL_MSG_CONFIG]      = { .call = nfqnl_recv_config,
1009                                     .attr_count = NFQA_CFG_MAX,
1010                                     .policy = nfqa_cfg_policy },
1011         [NFQNL_MSG_VERDICT_BATCH]={ .call_rcu = nfqnl_recv_verdict_batch,
1012                                     .attr_count = NFQA_MAX,
1013                                     .policy = nfqa_verdict_batch_policy },
1014 };
1015
1016 static const struct nfnetlink_subsystem nfqnl_subsys = {
1017         .name           = "nf_queue",
1018         .subsys_id      = NFNL_SUBSYS_QUEUE,
1019         .cb_count       = NFQNL_MSG_MAX,
1020         .cb             = nfqnl_cb,
1021 };
1022
1023 #ifdef CONFIG_PROC_FS
1024 struct iter_state {
1025         struct seq_net_private p;
1026         unsigned int bucket;
1027 };
1028
1029 static struct hlist_node *get_first(struct seq_file *seq)
1030 {
1031         struct iter_state *st = seq->private;
1032         struct net *net;
1033         struct nfnl_queue_net *q;
1034
1035         if (!st)
1036                 return NULL;
1037
1038         net = seq_file_net(seq);
1039         q = nfnl_queue_pernet(net);
1040         for (st->bucket = 0; st->bucket < INSTANCE_BUCKETS; st->bucket++) {
1041                 if (!hlist_empty(&q->instance_table[st->bucket]))
1042                         return q->instance_table[st->bucket].first;
1043         }
1044         return NULL;
1045 }
1046
1047 static struct hlist_node *get_next(struct seq_file *seq, struct hlist_node *h)
1048 {
1049         struct iter_state *st = seq->private;
1050         struct net *net = seq_file_net(seq);
1051
1052         h = h->next;
1053         while (!h) {
1054                 struct nfnl_queue_net *q;
1055
1056                 if (++st->bucket >= INSTANCE_BUCKETS)
1057                         return NULL;
1058
1059                 q = nfnl_queue_pernet(net);
1060                 h = q->instance_table[st->bucket].first;
1061         }
1062         return h;
1063 }
1064
1065 static struct hlist_node *get_idx(struct seq_file *seq, loff_t pos)
1066 {
1067         struct hlist_node *head;
1068         head = get_first(seq);
1069
1070         if (head)
1071                 while (pos && (head = get_next(seq, head)))
1072                         pos--;
1073         return pos ? NULL : head;
1074 }
1075
1076 static void *seq_start(struct seq_file *s, loff_t *pos)
1077         __acquires(nfnl_queue_pernet(seq_file_net(s))->instances_lock)
1078 {
1079         spin_lock(&nfnl_queue_pernet(seq_file_net(s))->instances_lock);
1080         return get_idx(s, *pos);
1081 }
1082
1083 static void *seq_next(struct seq_file *s, void *v, loff_t *pos)
1084 {
1085         (*pos)++;
1086         return get_next(s, v);
1087 }
1088
1089 static void seq_stop(struct seq_file *s, void *v)
1090         __releases(nfnl_queue_pernet(seq_file_net(s))->instances_lock)
1091 {
1092         spin_unlock(&nfnl_queue_pernet(seq_file_net(s))->instances_lock);
1093 }
1094
1095 static int seq_show(struct seq_file *s, void *v)
1096 {
1097         const struct nfqnl_instance *inst = v;
1098
1099         return seq_printf(s, "%5d %6d %5d %1d %5d %5d %5d %8d %2d\n",
1100                           inst->queue_num,
1101                           inst->peer_portid, inst->queue_total,
1102                           inst->copy_mode, inst->copy_range,
1103                           inst->queue_dropped, inst->queue_user_dropped,
1104                           inst->id_sequence, 1);
1105 }
1106
1107 static const struct seq_operations nfqnl_seq_ops = {
1108         .start  = seq_start,
1109         .next   = seq_next,
1110         .stop   = seq_stop,
1111         .show   = seq_show,
1112 };
1113
1114 static int nfqnl_open(struct inode *inode, struct file *file)
1115 {
1116         return seq_open_net(inode, file, &nfqnl_seq_ops,
1117                         sizeof(struct iter_state));
1118 }
1119
1120 static const struct file_operations nfqnl_file_ops = {
1121         .owner   = THIS_MODULE,
1122         .open    = nfqnl_open,
1123         .read    = seq_read,
1124         .llseek  = seq_lseek,
1125         .release = seq_release_net,
1126 };
1127
1128 #endif /* PROC_FS */
1129
1130 static int __net_init nfnl_queue_net_init(struct net *net)
1131 {
1132         unsigned int i;
1133         struct nfnl_queue_net *q = nfnl_queue_pernet(net);
1134
1135         for (i = 0; i < INSTANCE_BUCKETS; i++)
1136                 INIT_HLIST_HEAD(&q->instance_table[i]);
1137
1138         spin_lock_init(&q->instances_lock);
1139
1140 #ifdef CONFIG_PROC_FS
1141         if (!proc_create("nfnetlink_queue", 0440,
1142                          net->nf.proc_netfilter, &nfqnl_file_ops))
1143                 return -ENOMEM;
1144 #endif
1145         return 0;
1146 }
1147
1148 static void __net_exit nfnl_queue_net_exit(struct net *net)
1149 {
1150         remove_proc_entry("nfnetlink_queue", net->nf.proc_netfilter);
1151 }
1152
1153 static struct pernet_operations nfnl_queue_net_ops = {
1154         .init   = nfnl_queue_net_init,
1155         .exit   = nfnl_queue_net_exit,
1156         .id     = &nfnl_queue_net_id,
1157         .size   = sizeof(struct nfnl_queue_net),
1158 };
1159
1160 static int __init nfnetlink_queue_init(void)
1161 {
1162         int status = -ENOMEM;
1163
1164         netlink_register_notifier(&nfqnl_rtnl_notifier);
1165         status = nfnetlink_subsys_register(&nfqnl_subsys);
1166         if (status < 0) {
1167                 pr_err("nf_queue: failed to create netlink socket\n");
1168                 goto cleanup_netlink_notifier;
1169         }
1170
1171         status = register_pernet_subsys(&nfnl_queue_net_ops);
1172         if (status < 0) {
1173                 pr_err("nf_queue: failed to register pernet ops\n");
1174                 goto cleanup_subsys;
1175         }
1176         register_netdevice_notifier(&nfqnl_dev_notifier);
1177         nf_register_queue_handler(&nfqh);
1178         return status;
1179
1180 cleanup_subsys:
1181         nfnetlink_subsys_unregister(&nfqnl_subsys);
1182 cleanup_netlink_notifier:
1183         netlink_unregister_notifier(&nfqnl_rtnl_notifier);
1184         return status;
1185 }
1186
1187 static void __exit nfnetlink_queue_fini(void)
1188 {
1189         nf_unregister_queue_handler();
1190         unregister_netdevice_notifier(&nfqnl_dev_notifier);
1191         unregister_pernet_subsys(&nfnl_queue_net_ops);
1192         nfnetlink_subsys_unregister(&nfqnl_subsys);
1193         netlink_unregister_notifier(&nfqnl_rtnl_notifier);
1194
1195         rcu_barrier(); /* Wait for completion of call_rcu()'s */
1196 }
1197
1198 MODULE_DESCRIPTION("netfilter packet queue handler");
1199 MODULE_AUTHOR("Harald Welte <laforge@netfilter.org>");
1200 MODULE_LICENSE("GPL");
1201 MODULE_ALIAS_NFNL_SUBSYS(NFNL_SUBSYS_QUEUE);
1202
1203 module_init(nfnetlink_queue_init);
1204 module_exit(nfnetlink_queue_fini);