net/netfilter/nfnetlink_queue.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_bridge.h>
  29 #include <linux/netfilter/nfnetlink.h>
  30 #include <linux/netfilter/nfnetlink_queue.h>
  31 #include <linux/netfilter/nf_conntrack_common.h>
  32 #include <linux/list.h>
  33 #include <net/sock.h>
  34 #include <net/tcp_states.h>
  35 #include <net/netfilter/nf_queue.h>
  36 #include <net/netns/generic.h>
  37
  38 #include <linux/atomic.h>
  39
  40 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
  41 #include "../bridge/br_private.h"
  42 #endif
  43
  44 #define NFQNL_QMAX_DEFAULT 1024
  45
  46 /* We're using struct nlattr which has 16bit nla_len. Note that nla_len
  47  * includes the header length. Thus, the maximum packet length that we
  48  * support is 65531 bytes. We send truncated packets if the specified length
  49  * is larger than that.  Userspace can check for presence of NFQA_CAP_LEN
  50  * attribute to detect truncation.
  51  */
  52 #define NFQNL_MAX_COPY_RANGE (0xffff - NLA_HDRLEN)
  53
  54 struct nfqnl_instance {
  55         struct hlist_node hlist;                /* global list of queues */
  56         struct rcu_head rcu;
  57
  58         u32 peer_portid;
  59         unsigned int queue_maxlen;
  60         unsigned int copy_range;
  61         unsigned int queue_dropped;
  62         unsigned int queue_user_dropped;
  63
  64
  65         u_int16_t queue_num;                    /* number of this queue */
  66         u_int8_t copy_mode;
  67         u_int32_t flags;                        /* Set using NFQA_CFG_FLAGS */
  68 /*
  69  * Following fields are dirtied for each queued packet,
  70  * keep them in same cache line if possible.
  71  */
  72         spinlock_t      lock;
  73         unsigned int    queue_total;
  74         unsigned int    id_sequence;            /* 'sequence' of pkt ids */
  75         struct list_head queue_list;            /* packets in queue */
  76 };
  77
  78 typedef int (*nfqnl_cmpfn)(struct nf_queue_entry *, unsigned long);
  79
  80 static int nfnl_queue_net_id __read_mostly;
  81
  82 #define INSTANCE_BUCKETS        16
  83 struct nfnl_queue_net {
  84         spinlock_t instances_lock;
  85         struct hlist_head instance_table[INSTANCE_BUCKETS];
  86 };
  87
  88 static struct nfnl_queue_net *nfnl_queue_pernet(struct net *net)
  89 {
  90         return net_generic(net, nfnl_queue_net_id);
  91 }
  92
  93 static inline u_int8_t instance_hashfn(u_int16_t queue_num)
  94 {
  95         return ((queue_num >> 8) ^ queue_num) % INSTANCE_BUCKETS;
  96 }
  97
  98 static struct nfqnl_instance *
  99 instance_lookup(struct nfnl_queue_net *q, u_int16_t queue_num)
 100 {
 101         struct hlist_head *head;
 102         struct nfqnl_instance *inst;
 103
 104         head = &q->instance_table[instance_hashfn(queue_num)];
 105         hlist_for_each_entry_rcu(inst, head, hlist) {
 106                 if (inst->queue_num == queue_num)
 107                         return inst;
 108         }
 109         return NULL;
 110 }
 111
 112 static struct nfqnl_instance *
 113 instance_create(struct nfnl_queue_net *q, u_int16_t queue_num, u32 portid)
 114 {
 115         struct nfqnl_instance *inst;
 116         unsigned int h;
 117         int err;
 118
 119         spin_lock(&q->instances_lock);
 120         if (instance_lookup(q, queue_num)) {
 121                 err = -EEXIST;
 122                 goto out_unlock;
 123         }
 124
 125         inst = kzalloc(sizeof(*inst), GFP_ATOMIC);
 126         if (!inst) {
 127                 err = -ENOMEM;
 128                 goto out_unlock;
 129         }
 130
 131         inst->queue_num = queue_num;
 132         inst->peer_portid = portid;
 133         inst->queue_maxlen = NFQNL_QMAX_DEFAULT;
 134         inst->copy_range = NFQNL_MAX_COPY_RANGE;
 135         inst->copy_mode = NFQNL_COPY_NONE;
 136         spin_lock_init(&inst->lock);
 137         INIT_LIST_HEAD(&inst->queue_list);
 138
 139         if (!try_module_get(THIS_MODULE)) {
 140                 err = -EAGAIN;
 141                 goto out_free;
 142         }
 143
 144         h = instance_hashfn(queue_num);
 145         hlist_add_head_rcu(&inst->hlist, &q->instance_table[h]);
 146
 147         spin_unlock(&q->instances_lock);
 148
 149         return inst;
 150
 151 out_free:
 152         kfree(inst);
 153 out_unlock:
 154         spin_unlock(&q->instances_lock);
 155         return ERR_PTR(err);
 156 }
 157
 158 static void nfqnl_flush(struct nfqnl_instance *queue, nfqnl_cmpfn cmpfn,
 159                         unsigned long data);
 160
 161 static void
 162 instance_destroy_rcu(struct rcu_head *head)
 163 {
 164         struct nfqnl_instance *inst = container_of(head, struct nfqnl_instance,
 165                                                    rcu);
 166
 167         nfqnl_flush(inst, NULL, 0);
 168         kfree(inst);
 169         module_put(THIS_MODULE);
 170 }
 171
 172 static void
 173 __instance_destroy(struct nfqnl_instance *inst)
 174 {
 175         hlist_del_rcu(&inst->hlist);
 176         call_rcu(&inst->rcu, instance_destroy_rcu);
 177 }
 178
 179 static void
 180 instance_destroy(struct nfnl_queue_net *q, struct nfqnl_instance *inst)
 181 {
 182         spin_lock(&q->instances_lock);
 183         __instance_destroy(inst);
 184         spin_unlock(&q->instances_lock);
 185 }
 186
 187 static inline void
 188 __enqueue_entry(struct nfqnl_instance *queue, struct nf_queue_entry *entry)
 189 {
 190        list_add_tail(&entry->list, &queue->queue_list);
 191        queue->queue_total++;
 192 }
 193
 194 static void
 195 __dequeue_entry(struct nfqnl_instance *queue, struct nf_queue_entry *entry)
 196 {
 197         list_del(&entry->list);
 198         queue->queue_total--;
 199 }
 200
 201 static struct nf_queue_entry *
 202 find_dequeue_entry(struct nfqnl_instance *queue, unsigned int id)
 203 {
 204         struct nf_queue_entry *entry = NULL, *i;
 205
 206         spin_lock_bh(&queue->lock);
 207
 208         list_for_each_entry(i, &queue->queue_list, list) {
 209                 if (i->id == id) {
 210                         entry = i;
 211                         break;
 212                 }
 213         }
 214
 215         if (entry)
 216                 __dequeue_entry(queue, entry);
 217
 218         spin_unlock_bh(&queue->lock);
 219
 220         return entry;
 221 }
 222
 223 static void
 224 nfqnl_flush(struct nfqnl_instance *queue, nfqnl_cmpfn cmpfn, unsigned long data)
 225 {
 226         struct nf_queue_entry *entry, *next;
 227
 228         spin_lock_bh(&queue->lock);
 229         list_for_each_entry_safe(entry, next, &queue->queue_list, list) {
 230                 if (!cmpfn || cmpfn(entry, data)) {
 231                         list_del(&entry->list);
 232                         queue->queue_total--;
 233                         nf_reinject(entry, NF_DROP);
 234                 }
 235         }
 236         spin_unlock_bh(&queue->lock);
 237 }
 238
 239 static int
 240 nfqnl_put_packet_info(struct sk_buff *nlskb, struct sk_buff *packet,
 241                       bool csum_verify)
 242 {
 243         __u32 flags = 0;
 244
 245         if (packet->ip_summed == CHECKSUM_PARTIAL)
 246                 flags = NFQA_SKB_CSUMNOTREADY;
 247         else if (csum_verify)
 248                 flags = NFQA_SKB_CSUM_NOTVERIFIED;
 249
 250         if (skb_is_gso(packet))
 251                 flags |= NFQA_SKB_GSO;
 252
 253         return flags ? nla_put_be32(nlskb, NFQA_SKB_INFO, htonl(flags)) : 0;
 254 }
 255
 256 static int nfqnl_put_sk_uidgid(struct sk_buff *skb, struct sock *sk)
 257 {
 258         const struct cred *cred;
 259
 260         if (!sk_fullsock(sk))
 261                 return 0;
 262
 263         read_lock_bh(&sk->sk_callback_lock);
 264         if (sk->sk_socket && sk->sk_socket->file) {
 265                 cred = sk->sk_socket->file->f_cred;
 266                 if (nla_put_be32(skb, NFQA_UID,
 267                     htonl(from_kuid_munged(&init_user_ns, cred->fsuid))))
 268                         goto nla_put_failure;
 269                 if (nla_put_be32(skb, NFQA_GID,
 270                     htonl(from_kgid_munged(&init_user_ns, cred->fsgid))))
 271                         goto nla_put_failure;
 272         }
 273         read_unlock_bh(&sk->sk_callback_lock);
 274         return 0;
 275
 276 nla_put_failure:
 277         read_unlock_bh(&sk->sk_callback_lock);
 278         return -1;
 279 }
 280
 281 static u32 nfqnl_get_sk_secctx(struct sk_buff *skb, char **secdata)
 282 {
 283         u32 seclen = 0;
 284 #if IS_ENABLED(CONFIG_NETWORK_SECMARK)
 285         if (!skb || !sk_fullsock(skb->sk))
 286                 return 0;
 287
 288         read_lock_bh(&skb->sk->sk_callback_lock);
 289
 290         if (skb->secmark)
 291                 security_secid_to_secctx(skb->secmark, secdata, &seclen);
 292
 293         read_unlock_bh(&skb->sk->sk_callback_lock);
 294 #endif
 295         return seclen;
 296 }
 297
 298 static struct sk_buff *
 299 nfqnl_build_packet_message(struct net *net, struct nfqnl_instance *queue,
 300                            struct nf_queue_entry *entry,
 301                            __be32 **packet_id_ptr)
 302 {
 303         size_t size;
 304         size_t data_len = 0, cap_len = 0, rem_len = 0;
 305         unsigned int hlen = 0;
 306         struct sk_buff *skb;
 307         struct nlattr *nla;
 308         struct nfqnl_msg_packet_hdr *pmsg;
 309         struct nlmsghdr *nlh;
 310         struct nfgenmsg *nfmsg;
 311         struct sk_buff *entskb = entry->skb;
 312         struct net_device *indev;
 313         struct net_device *outdev;
 314         struct nf_conn *ct = NULL;
 315         enum ip_conntrack_info uninitialized_var(ctinfo);
 316         struct nfnl_ct_hook *nfnl_ct;
 317         bool csum_verify;
 318         char *secdata = NULL;
 319         u32 seclen = 0;
 320
 321         size =    nlmsg_total_size(sizeof(struct nfgenmsg))
 322                 + nla_total_size(sizeof(struct nfqnl_msg_packet_hdr))
 323                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
 324                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
 325 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
 326                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
 327                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
 328 #endif
 329                 + nla_total_size(sizeof(u_int32_t))     /* mark */
 330                 + nla_total_size(sizeof(struct nfqnl_msg_packet_hw))
 331                 + nla_total_size(sizeof(u_int32_t))     /* skbinfo */
 332                 + nla_total_size(sizeof(u_int32_t));    /* cap_len */
 333
 334         if (entskb->tstamp.tv64)
 335                 size += nla_total_size(sizeof(struct nfqnl_msg_packet_timestamp));
 336
 337         if (entry->state.hook <= NF_INET_FORWARD ||
 338            (entry->state.hook == NF_INET_POST_ROUTING && entskb->sk == NULL))
 339                 csum_verify = !skb_csum_unnecessary(entskb);
 340         else
 341                 csum_verify = false;
 342
 343         outdev = entry->state.out;
 344
 345         switch ((enum nfqnl_config_mode)ACCESS_ONCE(queue->copy_mode)) {
 346         case NFQNL_COPY_META:
 347         case NFQNL_COPY_NONE:
 348                 break;
 349
 350         case NFQNL_COPY_PACKET:
 351                 if (!(queue->flags & NFQA_CFG_F_GSO) &&
 352                     entskb->ip_summed == CHECKSUM_PARTIAL &&
 353                     skb_checksum_help(entskb))
 354                         return NULL;
 355
 356                 data_len = ACCESS_ONCE(queue->copy_range);
 357                 if (data_len > entskb->len)
 358                         data_len = entskb->len;
 359
 360                 hlen = skb_zerocopy_headlen(entskb);
 361                 hlen = min_t(unsigned int, hlen, data_len);
 362                 size += sizeof(struct nlattr) + hlen;
 363                 cap_len = entskb->len;
 364                 rem_len = data_len - hlen;
 365                 break;
 366         }
 367
 368         nfnl_ct = rcu_dereference(nfnl_ct_hook);
 369
 370         if (queue->flags & NFQA_CFG_F_CONNTRACK) {
 371                 if (nfnl_ct != NULL) {
 372                         ct = nfnl_ct->get_ct(entskb, &ctinfo);
 373                         if (ct != NULL)
 374                                 size += nfnl_ct->build_size(ct);
 375                 }
 376         }
 377
 378         if (queue->flags & NFQA_CFG_F_UID_GID) {
 379                 size +=  (nla_total_size(sizeof(u_int32_t))     /* uid */
 380                         + nla_total_size(sizeof(u_int32_t)));   /* gid */
 381         }
 382
 383         if ((queue->flags & NFQA_CFG_F_SECCTX) && entskb->sk) {
 384                 seclen = nfqnl_get_sk_secctx(entskb, &secdata);
 385                 if (seclen)
 386                         size += nla_total_size(seclen);
 387         }
 388
 389         skb = __netlink_alloc_skb(net->nfnl, size, rem_len, queue->peer_portid,
 390                                   GFP_ATOMIC);
 391         if (!skb) {
 392                 skb_tx_error(entskb);
 393                 goto nlmsg_failure;
 394         }
 395
 396         nlh = nlmsg_put(skb, 0, 0,
 397                         NFNL_SUBSYS_QUEUE << 8 | NFQNL_MSG_PACKET,
 398                         sizeof(struct nfgenmsg), 0);
 399         if (!nlh) {
 400                 skb_tx_error(entskb);
 401                 kfree_skb(skb);
 402                 goto nlmsg_failure;
 403         }
 404         nfmsg = nlmsg_data(nlh);
 405         nfmsg->nfgen_family = entry->state.pf;
 406         nfmsg->version = NFNETLINK_V0;
 407         nfmsg->res_id = htons(queue->queue_num);
 408
 409         nla = __nla_reserve(skb, NFQA_PACKET_HDR, sizeof(*pmsg));
 410         pmsg = nla_data(nla);
 411         pmsg->hw_protocol       = entskb->protocol;
 412         pmsg->hook              = entry->state.hook;
 413         *packet_id_ptr          = &pmsg->packet_id;
 414
 415         indev = entry->state.in;
 416         if (indev) {
 417 #if !IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
 418                 if (nla_put_be32(skb, NFQA_IFINDEX_INDEV, htonl(indev->ifindex)))
 419                         goto nla_put_failure;
 420 #else
 421                 if (entry->state.pf == PF_BRIDGE) {
 422                         /* Case 1: indev is physical input device, we need to
 423                          * look for bridge group (when called from
 424                          * netfilter_bridge) */
 425                         if (nla_put_be32(skb, NFQA_IFINDEX_PHYSINDEV,
 426                                          htonl(indev->ifindex)) ||
 427                         /* this is the bridge group "brX" */
 428                         /* rcu_read_lock()ed by __nf_queue */
 429                             nla_put_be32(skb, NFQA_IFINDEX_INDEV,
 430                                          htonl(br_port_get_rcu(indev)->br->dev->ifindex)))
 431                                 goto nla_put_failure;
 432                 } else {
 433                         int physinif;
 434
 435                         /* Case 2: indev is bridge group, we need to look for
 436                          * physical device (when called from ipv4) */
 437                         if (nla_put_be32(skb, NFQA_IFINDEX_INDEV,
 438                                          htonl(indev->ifindex)))
 439                                 goto nla_put_failure;
 440
 441                         physinif = nf_bridge_get_physinif(entskb);
 442                         if (physinif &&
 443                             nla_put_be32(skb, NFQA_IFINDEX_PHYSINDEV,
 444                                          htonl(physinif)))
 445                                 goto nla_put_failure;
 446                 }
 447 #endif
 448         }
 449
 450         if (outdev) {
 451 #if !IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
 452                 if (nla_put_be32(skb, NFQA_IFINDEX_OUTDEV, htonl(outdev->ifindex)))
 453                         goto nla_put_failure;
 454 #else
 455                 if (entry->state.pf == PF_BRIDGE) {
 456                         /* Case 1: outdev is physical output device, we need to
 457                          * look for bridge group (when called from
 458                          * netfilter_bridge) */
 459                         if (nla_put_be32(skb, NFQA_IFINDEX_PHYSOUTDEV,
 460                                          htonl(outdev->ifindex)) ||
 461                         /* this is the bridge group "brX" */
 462                         /* rcu_read_lock()ed by __nf_queue */
 463                             nla_put_be32(skb, NFQA_IFINDEX_OUTDEV,
 464                                          htonl(br_port_get_rcu(outdev)->br->dev->ifindex)))
 465                                 goto nla_put_failure;
 466                 } else {
 467                         int physoutif;
 468
 469                         /* Case 2: outdev is bridge group, we need to look for
 470                          * physical output device (when called from ipv4) */
 471                         if (nla_put_be32(skb, NFQA_IFINDEX_OUTDEV,
 472                                          htonl(outdev->ifindex)))
 473                                 goto nla_put_failure;
 474
 475                         physoutif = nf_bridge_get_physoutif(entskb);
 476                         if (physoutif &&
 477                             nla_put_be32(skb, NFQA_IFINDEX_PHYSOUTDEV,
 478                                          htonl(physoutif)))
 479                                 goto nla_put_failure;
 480                 }
 481 #endif
 482         }
 483
 484         if (entskb->mark &&
 485             nla_put_be32(skb, NFQA_MARK, htonl(entskb->mark)))
 486                 goto nla_put_failure;
 487
 488         if (indev && entskb->dev &&
 489             skb_mac_header_was_set(entskb)) {
 490                 struct nfqnl_msg_packet_hw phw;
 491                 int len;
 492
 493                 memset(&phw, 0, sizeof(phw));
 494                 len = dev_parse_header(entskb, phw.hw_addr);
 495                 if (len) {
 496                         phw.hw_addrlen = htons(len);
 497                         if (nla_put(skb, NFQA_HWADDR, sizeof(phw), &phw))
 498                                 goto nla_put_failure;
 499                 }
 500         }
 501
 502         if (entskb->tstamp.tv64) {
 503                 struct nfqnl_msg_packet_timestamp ts;
 504                 struct timespec64 kts = ktime_to_timespec64(entskb->tstamp);
 505
 506                 ts.sec = cpu_to_be64(kts.tv_sec);
 507                 ts.usec = cpu_to_be64(kts.tv_nsec / NSEC_PER_USEC);
 508
 509                 if (nla_put(skb, NFQA_TIMESTAMP, sizeof(ts), &ts))
 510                         goto nla_put_failure;
 511         }
 512
 513         if ((queue->flags & NFQA_CFG_F_UID_GID) && entskb->sk &&
 514             nfqnl_put_sk_uidgid(skb, entskb->sk) < 0)
 515                 goto nla_put_failure;
 516
 517         if (seclen && nla_put(skb, NFQA_SECCTX, seclen, secdata))
 518                 goto nla_put_failure;
 519
 520         if (ct && nfnl_ct->build(skb, ct, ctinfo, NFQA_CT, NFQA_CT_INFO) < 0)
 521                 goto nla_put_failure;
 522
 523         if (cap_len > data_len &&
 524             nla_put_be32(skb, NFQA_CAP_LEN, htonl(cap_len)))
 525                 goto nla_put_failure;
 526
 527         if (nfqnl_put_packet_info(skb, entskb, csum_verify))
 528                 goto nla_put_failure;
 529
 530         if (data_len) {
 531                 struct nlattr *nla;
 532
 533                 if (skb_tailroom(skb) < sizeof(*nla) + hlen)
 534                         goto nla_put_failure;
 535
 536                 nla = (struct nlattr *)skb_put(skb, sizeof(*nla));
 537                 nla->nla_type = NFQA_PAYLOAD;
 538                 nla->nla_len = nla_attr_size(data_len);
 539
 540                 if (skb_zerocopy(skb, entskb, data_len, hlen))
 541                         goto nla_put_failure;
 542         }
 543
 544         nlh->nlmsg_len = skb->len;
 545         if (seclen)
 546                 security_release_secctx(secdata, seclen);
 547         return skb;
 548
 549 nla_put_failure:
 550         skb_tx_error(entskb);
 551         kfree_skb(skb);
 552         net_err_ratelimited("nf_queue: error creating packet message\n");
 553 nlmsg_failure:
 554         if (seclen)
 555                 security_release_secctx(secdata, seclen);
 556         return NULL;
 557 }
 558
 559 static int
 560 __nfqnl_enqueue_packet(struct net *net, struct nfqnl_instance *queue,
 561                         struct nf_queue_entry *entry)
 562 {
 563         struct sk_buff *nskb;
 564         int err = -ENOBUFS;
 565         __be32 *packet_id_ptr;
 566         int failopen = 0;
 567
 568         nskb = nfqnl_build_packet_message(net, queue, entry, &packet_id_ptr);
 569         if (nskb == NULL) {
 570                 err = -ENOMEM;
 571                 goto err_out;
 572         }
 573         spin_lock_bh(&queue->lock);
 574
 575         if (queue->queue_total >= queue->queue_maxlen) {
 576                 if (queue->flags & NFQA_CFG_F_FAIL_OPEN) {
 577                         failopen = 1;
 578                         err = 0;
 579                 } else {
 580                         queue->queue_dropped++;
 581                         net_warn_ratelimited("nf_queue: full at %d entries, dropping packets(s)\n",
 582                                              queue->queue_total);
 583                 }
 584                 goto err_out_free_nskb;
 585         }
 586         entry->id = ++queue->id_sequence;
 587         *packet_id_ptr = htonl(entry->id);
 588
 589         /* nfnetlink_unicast will either free the nskb or add it to a socket */
 590         err = nfnetlink_unicast(nskb, net, queue->peer_portid, MSG_DONTWAIT);
 591         if (err < 0) {
 592                 queue->queue_user_dropped++;
 593                 goto err_out_unlock;
 594         }
 595
 596         __enqueue_entry(queue, entry);
 597
 598         spin_unlock_bh(&queue->lock);
 599         return 0;
 600
 601 err_out_free_nskb:
 602         kfree_skb(nskb);
 603 err_out_unlock:
 604         spin_unlock_bh(&queue->lock);
 605         if (failopen)
 606                 nf_reinject(entry, NF_ACCEPT);
 607 err_out:
 608         return err;
 609 }
 610
 611 static struct nf_queue_entry *
 612 nf_queue_entry_dup(struct nf_queue_entry *e)
 613 {
 614         struct nf_queue_entry *entry = kmemdup(e, e->size, GFP_ATOMIC);
 615         if (entry)
 616                 nf_queue_entry_get_refs(entry);
 617         return entry;
 618 }
 619
 620 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
 621 /* When called from bridge netfilter, skb->data must point to MAC header
 622  * before calling skb_gso_segment(). Else, original MAC header is lost
 623  * and segmented skbs will be sent to wrong destination.
 624  */
 625 static void nf_bridge_adjust_skb_data(struct sk_buff *skb)
 626 {
 627         if (skb->nf_bridge)
 628                 __skb_push(skb, skb->network_header - skb->mac_header);
 629 }
 630
 631 static void nf_bridge_adjust_segmented_data(struct sk_buff *skb)
 632 {
 633         if (skb->nf_bridge)
 634                 __skb_pull(skb, skb->network_header - skb->mac_header);
 635 }
 636 #else
 637 #define nf_bridge_adjust_skb_data(s) do {} while (0)
 638 #define nf_bridge_adjust_segmented_data(s) do {} while (0)
 639 #endif
 640
 641 static void free_entry(struct nf_queue_entry *entry)
 642 {
 643         nf_queue_entry_release_refs(entry);
 644         kfree(entry);
 645 }
 646
 647 static int
 648 __nfqnl_enqueue_packet_gso(struct net *net, struct nfqnl_instance *queue,
 649                            struct sk_buff *skb, struct nf_queue_entry *entry)
 650 {
 651         int ret = -ENOMEM;
 652         struct nf_queue_entry *entry_seg;
 653
 654         nf_bridge_adjust_segmented_data(skb);
 655
 656         if (skb->next == NULL) { /* last packet, no need to copy entry */
 657                 struct sk_buff *gso_skb = entry->skb;
 658                 entry->skb = skb;
 659                 ret = __nfqnl_enqueue_packet(net, queue, entry);
 660                 if (ret)
 661                         entry->skb = gso_skb;
 662                 return ret;
 663         }
 664
 665         skb->next = NULL;
 666
 667         entry_seg = nf_queue_entry_dup(entry);
 668         if (entry_seg) {
 669                 entry_seg->skb = skb;
 670                 ret = __nfqnl_enqueue_packet(net, queue, entry_seg);
 671                 if (ret)
 672                         free_entry(entry_seg);
 673         }
 674         return ret;
 675 }
 676
 677 static int
 678 nfqnl_enqueue_packet(struct nf_queue_entry *entry, unsigned int queuenum)
 679 {
 680         unsigned int queued;
 681         struct nfqnl_instance *queue;
 682         struct sk_buff *skb, *segs;
 683         int err = -ENOBUFS;
 684         struct net *net = entry->state.net;
 685         struct nfnl_queue_net *q = nfnl_queue_pernet(net);
 686
 687         /* rcu_read_lock()ed by nf_hook_slow() */
 688         queue = instance_lookup(q, queuenum);
 689         if (!queue)
 690                 return -ESRCH;
 691
 692         if (queue->copy_mode == NFQNL_COPY_NONE)
 693                 return -EINVAL;
 694
 695         skb = entry->skb;
 696
 697         switch (entry->state.pf) {
 698         case NFPROTO_IPV4:
 699                 skb->protocol = htons(ETH_P_IP);
 700                 break;
 701         case NFPROTO_IPV6:
 702                 skb->protocol = htons(ETH_P_IPV6);
 703                 break;
 704         }
 705
 706         if ((queue->flags & NFQA_CFG_F_GSO) || !skb_is_gso(skb))
 707                 return __nfqnl_enqueue_packet(net, queue, entry);
 708
 709         nf_bridge_adjust_skb_data(skb);
 710         segs = skb_gso_segment(skb, 0);
 711         /* Does not use PTR_ERR to limit the number of error codes that can be
 712          * returned by nf_queue.  For instance, callers rely on -ESRCH to
 713          * mean 'ignore this hook'.
 714          */
 715         if (IS_ERR_OR_NULL(segs))
 716                 goto out_err;
 717         queued = 0;
 718         err = 0;
 719         do {
 720                 struct sk_buff *nskb = segs->next;
 721                 if (err == 0)
 722                         err = __nfqnl_enqueue_packet_gso(net, queue,
 723                                                         segs, entry);
 724                 if (err == 0)
 725                         queued++;
 726                 else
 727                         kfree_skb(segs);
 728                 segs = nskb;
 729         } while (segs);
 730
 731         if (queued) {
 732                 if (err) /* some segments are already queued */
 733                         free_entry(entry);
 734                 kfree_skb(skb);
 735                 return 0;
 736         }
 737  out_err:
 738         nf_bridge_adjust_segmented_data(skb);
 739         return err;
 740 }
 741
 742 static int
 743 nfqnl_mangle(void *data, int data_len, struct nf_queue_entry *e, int diff)
 744 {
 745         struct sk_buff *nskb;
 746
 747         if (diff < 0) {
 748                 if (pskb_trim(e->skb, data_len))
 749                         return -ENOMEM;
 750         } else if (diff > 0) {
 751                 if (data_len > 0xFFFF)
 752                         return -EINVAL;
 753                 if (diff > skb_tailroom(e->skb)) {
 754                         nskb = skb_copy_expand(e->skb, skb_headroom(e->skb),
 755                                                diff, GFP_ATOMIC);
 756                         if (!nskb) {
 757                                 printk(KERN_WARNING "nf_queue: OOM "
 758                                       "in mangle, dropping packet\n");
 759                                 return -ENOMEM;
 760                         }
 761                         kfree_skb(e->skb);
 762                         e->skb = nskb;
 763                 }
 764                 skb_put(e->skb, diff);
 765         }
 766         if (!skb_make_writable(e->skb, data_len))
 767                 return -ENOMEM;
 768         skb_copy_to_linear_data(e->skb, data, data_len);
 769         e->skb->ip_summed = CHECKSUM_NONE;
 770         return 0;
 771 }
 772
 773 static int
 774 nfqnl_set_mode(struct nfqnl_instance *queue,
 775                unsigned char mode, unsigned int range)
 776 {
 777         int status = 0;
 778
 779         spin_lock_bh(&queue->lock);
 780         switch (mode) {
 781         case NFQNL_COPY_NONE:
 782         case NFQNL_COPY_META:
 783                 queue->copy_mode = mode;
 784                 queue->copy_range = 0;
 785                 break;
 786
 787         case NFQNL_COPY_PACKET:
 788                 queue->copy_mode = mode;
 789                 if (range == 0 || range > NFQNL_MAX_COPY_RANGE)
 790                         queue->copy_range = NFQNL_MAX_COPY_RANGE;
 791                 else
 792                         queue->copy_range = range;
 793                 break;
 794
 795         default:
 796                 status = -EINVAL;
 797
 798         }
 799         spin_unlock_bh(&queue->lock);
 800
 801         return status;
 802 }
 803
 804 static int
 805 dev_cmp(struct nf_queue_entry *entry, unsigned long ifindex)
 806 {
 807         if (entry->state.in)
 808                 if (entry->state.in->ifindex == ifindex)
 809                         return 1;
 810         if (entry->state.out)
 811                 if (entry->state.out->ifindex == ifindex)
 812                         return 1;
 813 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
 814         if (entry->skb->nf_bridge) {
 815                 int physinif, physoutif;
 816
 817                 physinif = nf_bridge_get_physinif(entry->skb);
 818                 physoutif = nf_bridge_get_physoutif(entry->skb);
 819
 820                 if (physinif == ifindex || physoutif == ifindex)
 821                         return 1;
 822         }
 823 #endif
 824         return 0;
 825 }
 826
 827 /* drop all packets with either indev or outdev == ifindex from all queue
 828  * instances */
 829 static void
 830 nfqnl_dev_drop(struct net *net, int ifindex)
 831 {
 832         int i;
 833         struct nfnl_queue_net *q = nfnl_queue_pernet(net);
 834
 835         rcu_read_lock();
 836
 837         for (i = 0; i < INSTANCE_BUCKETS; i++) {
 838                 struct nfqnl_instance *inst;
 839                 struct hlist_head *head = &q->instance_table[i];
 840
 841                 hlist_for_each_entry_rcu(inst, head, hlist)
 842                         nfqnl_flush(inst, dev_cmp, ifindex);
 843         }
 844
 845         rcu_read_unlock();
 846 }
 847
 848 static int
 849 nfqnl_rcv_dev_event(struct notifier_block *this,
 850                     unsigned long event, void *ptr)
 851 {
 852         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
 853
 854         /* Drop any packets associated with the downed device */
 855         if (event == NETDEV_DOWN)
 856                 nfqnl_dev_drop(dev_net(dev), dev->ifindex);
 857         return NOTIFY_DONE;
 858 }
 859
 860 static struct notifier_block nfqnl_dev_notifier = {
 861         .notifier_call  = nfqnl_rcv_dev_event,
 862 };
 863
 864 static int nf_hook_cmp(struct nf_queue_entry *entry, unsigned long ops_ptr)
 865 {
 866         return entry->elem == (struct nf_hook_ops *)ops_ptr;
 867 }
 868
 869 static void nfqnl_nf_hook_drop(struct net *net, struct nf_hook_ops *hook)
 870 {
 871         struct nfnl_queue_net *q = nfnl_queue_pernet(net);
 872         int i;
 873
 874         rcu_read_lock();
 875         for (i = 0; i < INSTANCE_BUCKETS; i++) {
 876                 struct nfqnl_instance *inst;
 877                 struct hlist_head *head = &q->instance_table[i];
 878
 879                 hlist_for_each_entry_rcu(inst, head, hlist)
 880                         nfqnl_flush(inst, nf_hook_cmp, (unsigned long)hook);
 881         }
 882         rcu_read_unlock();
 883 }
 884
 885 static int
 886 nfqnl_rcv_nl_event(struct notifier_block *this,
 887                    unsigned long event, void *ptr)
 888 {
 889         struct netlink_notify *n = ptr;
 890         struct nfnl_queue_net *q = nfnl_queue_pernet(n->net);
 891
 892         if (event == NETLINK_URELEASE && n->protocol == NETLINK_NETFILTER) {
 893                 int i;
 894
 895                 /* destroy all instances for this portid */
 896                 spin_lock(&q->instances_lock);
 897                 for (i = 0; i < INSTANCE_BUCKETS; i++) {
 898                         struct hlist_node *t2;
 899                         struct nfqnl_instance *inst;
 900                         struct hlist_head *head = &q->instance_table[i];
 901
 902                         hlist_for_each_entry_safe(inst, t2, head, hlist) {
 903                                 if (n->portid == inst->peer_portid)
 904                                         __instance_destroy(inst);
 905                         }
 906                 }
 907                 spin_unlock(&q->instances_lock);
 908         }
 909         return NOTIFY_DONE;
 910 }
 911
 912 static struct notifier_block nfqnl_rtnl_notifier = {
 913         .notifier_call  = nfqnl_rcv_nl_event,
 914 };
 915
 916 static const struct nla_policy nfqa_verdict_policy[NFQA_MAX+1] = {
 917         [NFQA_VERDICT_HDR]      = { .len = sizeof(struct nfqnl_msg_verdict_hdr) },
 918         [NFQA_MARK]             = { .type = NLA_U32 },
 919         [NFQA_PAYLOAD]          = { .type = NLA_UNSPEC },
 920         [NFQA_CT]               = { .type = NLA_UNSPEC },
 921         [NFQA_EXP]              = { .type = NLA_UNSPEC },
 922 };
 923
 924 static const struct nla_policy nfqa_verdict_batch_policy[NFQA_MAX+1] = {
 925         [NFQA_VERDICT_HDR]      = { .len = sizeof(struct nfqnl_msg_verdict_hdr) },
 926         [NFQA_MARK]             = { .type = NLA_U32 },
 927 };
 928
 929 static struct nfqnl_instance *
 930 verdict_instance_lookup(struct nfnl_queue_net *q, u16 queue_num, u32 nlportid)
 931 {
 932         struct nfqnl_instance *queue;
 933
 934         queue = instance_lookup(q, queue_num);
 935         if (!queue)
 936                 return ERR_PTR(-ENODEV);
 937
 938         if (queue->peer_portid != nlportid)
 939                 return ERR_PTR(-EPERM);
 940
 941         return queue;
 942 }
 943
 944 static struct nfqnl_msg_verdict_hdr*
 945 verdicthdr_get(const struct nlattr * const nfqa[])
 946 {
 947         struct nfqnl_msg_verdict_hdr *vhdr;
 948         unsigned int verdict;
 949
 950         if (!nfqa[NFQA_VERDICT_HDR])
 951                 return NULL;
 952
 953         vhdr = nla_data(nfqa[NFQA_VERDICT_HDR]);
 954         verdict = ntohl(vhdr->verdict) & NF_VERDICT_MASK;
 955         if (verdict > NF_MAX_VERDICT || verdict == NF_STOLEN)
 956                 return NULL;
 957         return vhdr;
 958 }
 959
 960 static int nfq_id_after(unsigned int id, unsigned int max)
 961 {
 962         return (int)(id - max) > 0;
 963 }
 964
 965 static int
 966 nfqnl_recv_verdict_batch(struct sock *ctnl, struct sk_buff *skb,
 967                    const struct nlmsghdr *nlh,
 968                    const struct nlattr * const nfqa[])
 969 {
 970         struct nfgenmsg *nfmsg = nlmsg_data(nlh);
 971         struct nf_queue_entry *entry, *tmp;
 972         unsigned int verdict, maxid;
 973         struct nfqnl_msg_verdict_hdr *vhdr;
 974         struct nfqnl_instance *queue;
 975         LIST_HEAD(batch_list);
 976         u16 queue_num = ntohs(nfmsg->res_id);
 977
 978         struct net *net = sock_net(ctnl);
 979         struct nfnl_queue_net *q = nfnl_queue_pernet(net);
 980
 981         queue = verdict_instance_lookup(q, queue_num,
 982                                         NETLINK_CB(skb).portid);
 983         if (IS_ERR(queue))
 984                 return PTR_ERR(queue);
 985
 986         vhdr = verdicthdr_get(nfqa);
 987         if (!vhdr)
 988                 return -EINVAL;
 989
 990         verdict = ntohl(vhdr->verdict);
 991         maxid = ntohl(vhdr->id);
 992
 993         spin_lock_bh(&queue->lock);
 994
 995         list_for_each_entry_safe(entry, tmp, &queue->queue_list, list) {
 996                 if (nfq_id_after(entry->id, maxid))
 997                         break;
 998                 __dequeue_entry(queue, entry);
 999                 list_add_tail(&entry->list, &batch_list);
1000         }
1001
1002         spin_unlock_bh(&queue->lock);
1003
1004         if (list_empty(&batch_list))
1005                 return -ENOENT;
1006
1007         list_for_each_entry_safe(entry, tmp, &batch_list, list) {
1008                 if (nfqa[NFQA_MARK])
1009                         entry->skb->mark = ntohl(nla_get_be32(nfqa[NFQA_MARK]));
1010                 nf_reinject(entry, verdict);
1011         }
1012         return 0;
1013 }
1014
1015 static struct nf_conn *nfqnl_ct_parse(struct nfnl_ct_hook *nfnl_ct,
1016                                       const struct nlmsghdr *nlh,
1017                                       const struct nlattr * const nfqa[],
1018                                       struct nf_queue_entry *entry,
1019                                       enum ip_conntrack_info *ctinfo)
1020 {
1021         struct nf_conn *ct;
1022
1023         ct = nfnl_ct->get_ct(entry->skb, ctinfo);
1024         if (ct == NULL)
1025                 return NULL;
1026
1027         if (nfnl_ct->parse(nfqa[NFQA_CT], ct) < 0)
1028                 return NULL;
1029
1030         if (nfqa[NFQA_EXP])
1031                 nfnl_ct->attach_expect(nfqa[NFQA_EXP], ct,
1032                                       NETLINK_CB(entry->skb).portid,
1033                                       nlmsg_report(nlh));
1034         return ct;
1035 }
1036
1037 static int
1038 nfqnl_recv_verdict(struct sock *ctnl, struct sk_buff *skb,
1039                    const struct nlmsghdr *nlh,
1040                    const struct nlattr * const nfqa[])
1041 {
1042         struct nfgenmsg *nfmsg = nlmsg_data(nlh);
1043         u_int16_t queue_num = ntohs(nfmsg->res_id);
1044
1045         struct nfqnl_msg_verdict_hdr *vhdr;
1046         struct nfqnl_instance *queue;
1047         unsigned int verdict;
1048         struct nf_queue_entry *entry;
1049         enum ip_conntrack_info uninitialized_var(ctinfo);
1050         struct nfnl_ct_hook *nfnl_ct;
1051         struct nf_conn *ct = NULL;
1052
1053         struct net *net = sock_net(ctnl);
1054         struct nfnl_queue_net *q = nfnl_queue_pernet(net);
1055
1056         queue = verdict_instance_lookup(q, queue_num,
1057                                         NETLINK_CB(skb).portid);
1058         if (IS_ERR(queue))
1059                 return PTR_ERR(queue);
1060
1061         vhdr = verdicthdr_get(nfqa);
1062         if (!vhdr)
1063                 return -EINVAL;
1064
1065         verdict = ntohl(vhdr->verdict);
1066
1067         entry = find_dequeue_entry(queue, ntohl(vhdr->id));
1068         if (entry == NULL)
1069                 return -ENOENT;
1070
1071         /* rcu lock already held from nfnl->call_rcu. */
1072         nfnl_ct = rcu_dereference(nfnl_ct_hook);
1073
1074         if (nfqa[NFQA_CT]) {
1075                 if (nfnl_ct != NULL)
1076                         ct = nfqnl_ct_parse(nfnl_ct, nlh, nfqa, entry, &ctinfo);
1077         }
1078
1079         if (nfqa[NFQA_PAYLOAD]) {
1080                 u16 payload_len = nla_len(nfqa[NFQA_PAYLOAD]);
1081                 int diff = payload_len - entry->skb->len;
1082
1083                 if (nfqnl_mangle(nla_data(nfqa[NFQA_PAYLOAD]),
1084                                  payload_len, entry, diff) < 0)
1085                         verdict = NF_DROP;
1086
1087                 if (ct && diff)
1088                         nfnl_ct->seq_adjust(entry->skb, ct, ctinfo, diff);
1089         }
1090
1091         if (nfqa[NFQA_MARK])
1092                 entry->skb->mark = ntohl(nla_get_be32(nfqa[NFQA_MARK]));
1093
1094         nf_reinject(entry, verdict);
1095         return 0;
1096 }
1097
1098 static int
1099 nfqnl_recv_unsupp(struct sock *ctnl, struct sk_buff *skb,
1100                   const struct nlmsghdr *nlh,
1101                   const struct nlattr * const nfqa[])
1102 {
1103         return -ENOTSUPP;
1104 }
1105
1106 static const struct nla_policy nfqa_cfg_policy[NFQA_CFG_MAX+1] = {
1107         [NFQA_CFG_CMD]          = { .len = sizeof(struct nfqnl_msg_config_cmd) },
1108         [NFQA_CFG_PARAMS]       = { .len = sizeof(struct nfqnl_msg_config_params) },
1109         [NFQA_CFG_QUEUE_MAXLEN] = { .type = NLA_U32 },
1110         [NFQA_CFG_MASK]         = { .type = NLA_U32 },
1111         [NFQA_CFG_FLAGS]        = { .type = NLA_U32 },
1112 };
1113
1114 static const struct nf_queue_handler nfqh = {
1115         .outfn          = &nfqnl_enqueue_packet,
1116         .nf_hook_drop   = &nfqnl_nf_hook_drop,
1117 };
1118
1119 static int
1120 nfqnl_recv_config(struct sock *ctnl, struct sk_buff *skb,
1121                   const struct nlmsghdr *nlh,
1122                   const struct nlattr * const nfqa[])
1123 {
1124         struct nfgenmsg *nfmsg = nlmsg_data(nlh);
1125         u_int16_t queue_num = ntohs(nfmsg->res_id);
1126         struct nfqnl_instance *queue;
1127         struct nfqnl_msg_config_cmd *cmd = NULL;
1128         struct net *net = sock_net(ctnl);
1129         struct nfnl_queue_net *q = nfnl_queue_pernet(net);
1130         int ret = 0;
1131
1132         if (nfqa[NFQA_CFG_CMD]) {
1133                 cmd = nla_data(nfqa[NFQA_CFG_CMD]);
1134
1135                 /* Obsolete commands without queue context */
1136                 switch (cmd->command) {
1137                 case NFQNL_CFG_CMD_PF_BIND: return 0;
1138                 case NFQNL_CFG_CMD_PF_UNBIND: return 0;
1139                 }
1140         }
1141
1142         rcu_read_lock();
1143         queue = instance_lookup(q, queue_num);
1144         if (queue && queue->peer_portid != NETLINK_CB(skb).portid) {
1145                 ret = -EPERM;
1146                 goto err_out_unlock;
1147         }
1148
1149         if (cmd != NULL) {
1150                 switch (cmd->command) {
1151                 case NFQNL_CFG_CMD_BIND:
1152                         if (queue) {
1153                                 ret = -EBUSY;
1154                                 goto err_out_unlock;
1155                         }
1156                         queue = instance_create(q, queue_num,
1157                                                 NETLINK_CB(skb).portid);
1158                         if (IS_ERR(queue)) {
1159                                 ret = PTR_ERR(queue);
1160                                 goto err_out_unlock;
1161                         }
1162                         break;
1163                 case NFQNL_CFG_CMD_UNBIND:
1164                         if (!queue) {
1165                                 ret = -ENODEV;
1166                                 goto err_out_unlock;
1167                         }
1168                         instance_destroy(q, queue);
1169                         break;
1170                 case NFQNL_CFG_CMD_PF_BIND:
1171                 case NFQNL_CFG_CMD_PF_UNBIND:
1172                         break;
1173                 default:
1174                         ret = -ENOTSUPP;
1175                         break;
1176                 }
1177         }
1178
1179         if (nfqa[NFQA_CFG_PARAMS]) {
1180                 struct nfqnl_msg_config_params *params;
1181
1182                 if (!queue) {
1183                         ret = -ENODEV;
1184                         goto err_out_unlock;
1185                 }
1186                 params = nla_data(nfqa[NFQA_CFG_PARAMS]);
1187                 nfqnl_set_mode(queue, params->copy_mode,
1188                                 ntohl(params->copy_range));
1189         }
1190
1191         if (nfqa[NFQA_CFG_QUEUE_MAXLEN]) {
1192                 __be32 *queue_maxlen;
1193
1194                 if (!queue) {
1195                         ret = -ENODEV;
1196                         goto err_out_unlock;
1197                 }
1198                 queue_maxlen = nla_data(nfqa[NFQA_CFG_QUEUE_MAXLEN]);
1199                 spin_lock_bh(&queue->lock);
1200                 queue->queue_maxlen = ntohl(*queue_maxlen);
1201                 spin_unlock_bh(&queue->lock);
1202         }
1203
1204         if (nfqa[NFQA_CFG_FLAGS]) {
1205                 __u32 flags, mask;
1206
1207                 if (!queue) {
1208                         ret = -ENODEV;
1209                         goto err_out_unlock;
1210                 }
1211
1212                 if (!nfqa[NFQA_CFG_MASK]) {
1213                         /* A mask is needed to specify which flags are being
1214                          * changed.
1215                          */
1216                         ret = -EINVAL;
1217                         goto err_out_unlock;
1218                 }
1219
1220                 flags = ntohl(nla_get_be32(nfqa[NFQA_CFG_FLAGS]));
1221                 mask = ntohl(nla_get_be32(nfqa[NFQA_CFG_MASK]));
1222
1223                 if (flags >= NFQA_CFG_F_MAX) {
1224                         ret = -EOPNOTSUPP;
1225                         goto err_out_unlock;
1226                 }
1227 #if !IS_ENABLED(CONFIG_NETWORK_SECMARK)
1228                 if (flags & mask & NFQA_CFG_F_SECCTX) {
1229                         ret = -EOPNOTSUPP;
1230                         goto err_out_unlock;
1231                 }
1232 #endif
1233                 spin_lock_bh(&queue->lock);
1234                 queue->flags &= ~mask;
1235                 queue->flags |= flags & mask;
1236                 spin_unlock_bh(&queue->lock);
1237         }
1238
1239 err_out_unlock:
1240         rcu_read_unlock();
1241         return ret;
1242 }
1243
1244 static const struct nfnl_callback nfqnl_cb[NFQNL_MSG_MAX] = {
1245         [NFQNL_MSG_PACKET]      = { .call_rcu = nfqnl_recv_unsupp,
1246                                     .attr_count = NFQA_MAX, },
1247         [NFQNL_MSG_VERDICT]     = { .call_rcu = nfqnl_recv_verdict,
1248                                     .attr_count = NFQA_MAX,
1249                                     .policy = nfqa_verdict_policy },
1250         [NFQNL_MSG_CONFIG]      = { .call = nfqnl_recv_config,
1251                                     .attr_count = NFQA_CFG_MAX,
1252                                     .policy = nfqa_cfg_policy },
1253         [NFQNL_MSG_VERDICT_BATCH]={ .call_rcu = nfqnl_recv_verdict_batch,
1254                                     .attr_count = NFQA_MAX,
1255                                     .policy = nfqa_verdict_batch_policy },
1256 };
1257
1258 static const struct nfnetlink_subsystem nfqnl_subsys = {
1259         .name           = "nf_queue",
1260         .subsys_id      = NFNL_SUBSYS_QUEUE,
1261         .cb_count       = NFQNL_MSG_MAX,
1262         .cb             = nfqnl_cb,
1263 };
1264
1265 #ifdef CONFIG_PROC_FS
1266 struct iter_state {
1267         struct seq_net_private p;
1268         unsigned int bucket;
1269 };
1270
1271 static struct hlist_node *get_first(struct seq_file *seq)
1272 {
1273         struct iter_state *st = seq->private;
1274         struct net *net;
1275         struct nfnl_queue_net *q;
1276
1277         if (!st)
1278                 return NULL;
1279
1280         net = seq_file_net(seq);
1281         q = nfnl_queue_pernet(net);
1282         for (st->bucket = 0; st->bucket < INSTANCE_BUCKETS; st->bucket++) {
1283                 if (!hlist_empty(&q->instance_table[st->bucket]))
1284                         return q->instance_table[st->bucket].first;
1285         }
1286         return NULL;
1287 }
1288
1289 static struct hlist_node *get_next(struct seq_file *seq, struct hlist_node *h)
1290 {
1291         struct iter_state *st = seq->private;
1292         struct net *net = seq_file_net(seq);
1293
1294         h = h->next;
1295         while (!h) {
1296                 struct nfnl_queue_net *q;
1297
1298                 if (++st->bucket >= INSTANCE_BUCKETS)
1299                         return NULL;
1300
1301                 q = nfnl_queue_pernet(net);
1302                 h = q->instance_table[st->bucket].first;
1303         }
1304         return h;
1305 }
1306
1307 static struct hlist_node *get_idx(struct seq_file *seq, loff_t pos)
1308 {
1309         struct hlist_node *head;
1310         head = get_first(seq);
1311
1312         if (head)
1313                 while (pos && (head = get_next(seq, head)))
1314                         pos--;
1315         return pos ? NULL : head;
1316 }
1317
1318 static void *seq_start(struct seq_file *s, loff_t *pos)
1319         __acquires(nfnl_queue_pernet(seq_file_net(s))->instances_lock)
1320 {
1321         spin_lock(&nfnl_queue_pernet(seq_file_net(s))->instances_lock);
1322         return get_idx(s, *pos);
1323 }
1324
1325 static void *seq_next(struct seq_file *s, void *v, loff_t *pos)
1326 {
1327         (*pos)++;
1328         return get_next(s, v);
1329 }
1330
1331 static void seq_stop(struct seq_file *s, void *v)
1332         __releases(nfnl_queue_pernet(seq_file_net(s))->instances_lock)
1333 {
1334         spin_unlock(&nfnl_queue_pernet(seq_file_net(s))->instances_lock);
1335 }
1336
1337 static int seq_show(struct seq_file *s, void *v)
1338 {
1339         const struct nfqnl_instance *inst = v;
1340
1341         seq_printf(s, "%5u %6u %5u %1u %5u %5u %5u %8u %2d\n",
1342                    inst->queue_num,
1343                    inst->peer_portid, inst->queue_total,
1344                    inst->copy_mode, inst->copy_range,
1345                    inst->queue_dropped, inst->queue_user_dropped,
1346                    inst->id_sequence, 1);
1347         return 0;
1348 }
1349
1350 static const struct seq_operations nfqnl_seq_ops = {
1351         .start  = seq_start,
1352         .next   = seq_next,
1353         .stop   = seq_stop,
1354         .show   = seq_show,
1355 };
1356
1357 static int nfqnl_open(struct inode *inode, struct file *file)
1358 {
1359         return seq_open_net(inode, file, &nfqnl_seq_ops,
1360                         sizeof(struct iter_state));
1361 }
1362
1363 static const struct file_operations nfqnl_file_ops = {
1364         .owner   = THIS_MODULE,
1365         .open    = nfqnl_open,
1366         .read    = seq_read,
1367         .llseek  = seq_lseek,
1368         .release = seq_release_net,
1369 };
1370
1371 #endif /* PROC_FS */
1372
1373 static int __net_init nfnl_queue_net_init(struct net *net)
1374 {
1375         unsigned int i;
1376         struct nfnl_queue_net *q = nfnl_queue_pernet(net);
1377
1378         for (i = 0; i < INSTANCE_BUCKETS; i++)
1379                 INIT_HLIST_HEAD(&q->instance_table[i]);
1380
1381         spin_lock_init(&q->instances_lock);
1382
1383 #ifdef CONFIG_PROC_FS
1384         if (!proc_create("nfnetlink_queue", 0440,
1385                          net->nf.proc_netfilter, &nfqnl_file_ops))
1386                 return -ENOMEM;
1387 #endif
1388         nf_register_queue_handler(net, &nfqh);
1389         return 0;
1390 }
1391
1392 static void __net_exit nfnl_queue_net_exit(struct net *net)
1393 {
1394         nf_unregister_queue_handler(net);
1395 #ifdef CONFIG_PROC_FS
1396         remove_proc_entry("nfnetlink_queue", net->nf.proc_netfilter);
1397 #endif
1398 }
1399
1400 static void nfnl_queue_net_exit_batch(struct list_head *net_exit_list)
1401 {
1402         synchronize_rcu();
1403 }
1404
1405 static struct pernet_operations nfnl_queue_net_ops = {
1406         .init           = nfnl_queue_net_init,
1407         .exit           = nfnl_queue_net_exit,
1408         .exit_batch     = nfnl_queue_net_exit_batch,
1409         .id             = &nfnl_queue_net_id,
1410         .size           = sizeof(struct nfnl_queue_net),
1411 };
1412
1413 static int __init nfnetlink_queue_init(void)
1414 {
1415         int status;
1416
1417         status = register_pernet_subsys(&nfnl_queue_net_ops);
1418         if (status < 0) {
1419                 pr_err("nf_queue: failed to register pernet ops\n");
1420                 goto out;
1421         }
1422
1423         netlink_register_notifier(&nfqnl_rtnl_notifier);
1424         status = nfnetlink_subsys_register(&nfqnl_subsys);
1425         if (status < 0) {
1426                 pr_err("nf_queue: failed to create netlink socket\n");
1427                 goto cleanup_netlink_notifier;
1428         }
1429
1430         register_netdevice_notifier(&nfqnl_dev_notifier);
1431         return status;
1432
1433 cleanup_netlink_notifier:
1434         netlink_unregister_notifier(&nfqnl_rtnl_notifier);
1435         unregister_pernet_subsys(&nfnl_queue_net_ops);
1436 out:
1437         return status;
1438 }
1439
1440 static void __exit nfnetlink_queue_fini(void)
1441 {
1442         unregister_netdevice_notifier(&nfqnl_dev_notifier);
1443         nfnetlink_subsys_unregister(&nfqnl_subsys);
1444         netlink_unregister_notifier(&nfqnl_rtnl_notifier);
1445         unregister_pernet_subsys(&nfnl_queue_net_ops);
1446
1447         rcu_barrier(); /* Wait for completion of call_rcu()'s */
1448 }
1449
1450 MODULE_DESCRIPTION("netfilter packet queue handler");
1451 MODULE_AUTHOR("Harald Welte <laforge@netfilter.org>");
1452 MODULE_LICENSE("GPL");
1453 MODULE_ALIAS_NFNL_SUBSYS(NFNL_SUBSYS_QUEUE);
1454
1455 module_init(nfnetlink_queue_init);
1456 module_exit(nfnetlink_queue_fini);