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 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
  45 #include <net/netfilter/nf_conntrack.h>
  46 #endif
  47
  48 #define NFQNL_QMAX_DEFAULT 1024
  49
  50 /* We're using struct nlattr which has 16bit nla_len. Note that nla_len
  51  * includes the header length. Thus, the maximum packet length that we
  52  * support is 65531 bytes. We send truncated packets if the specified length
  53  * is larger than that.  Userspace can check for presence of NFQA_CAP_LEN
  54  * attribute to detect truncation.
  55  */
  56 #define NFQNL_MAX_COPY_RANGE (0xffff - NLA_HDRLEN)
  57
  58 struct nfqnl_instance {
  59         struct hlist_node hlist;                /* global list of queues */
  60         struct rcu_head rcu;
  61
  62         u32 peer_portid;
  63         unsigned int queue_maxlen;
  64         unsigned int copy_range;
  65         unsigned int queue_dropped;
  66         unsigned int queue_user_dropped;
  67
  68
  69         u_int16_t queue_num;                    /* number of this queue */
  70         u_int8_t copy_mode;
  71         u_int32_t flags;                        /* Set using NFQA_CFG_FLAGS */
  72 /*
  73  * Following fields are dirtied for each queued packet,
  74  * keep them in same cache line if possible.
  75  */
  76         spinlock_t      lock    ____cacheline_aligned_in_smp;
  77         unsigned int    queue_total;
  78         unsigned int    id_sequence;            /* 'sequence' of pkt ids */
  79         struct list_head queue_list;            /* packets in queue */
  80 };
  81
  82 typedef int (*nfqnl_cmpfn)(struct nf_queue_entry *, unsigned long);
  83
  84 static unsigned int nfnl_queue_net_id __read_mostly;
  85
  86 #define INSTANCE_BUCKETS        16
  87 struct nfnl_queue_net {
  88         spinlock_t instances_lock;
  89         struct hlist_head instance_table[INSTANCE_BUCKETS];
  90 };
  91
  92 static struct nfnl_queue_net *nfnl_queue_pernet(struct net *net)
  93 {
  94         return net_generic(net, nfnl_queue_net_id);
  95 }
  96
  97 static inline u_int8_t instance_hashfn(u_int16_t queue_num)
  98 {
  99         return ((queue_num >> 8) ^ queue_num) % INSTANCE_BUCKETS;
 100 }
 101
 102 static struct nfqnl_instance *
 103 instance_lookup(struct nfnl_queue_net *q, u_int16_t queue_num)
 104 {
 105         struct hlist_head *head;
 106         struct nfqnl_instance *inst;
 107
 108         head = &q->instance_table[instance_hashfn(queue_num)];
 109         hlist_for_each_entry_rcu(inst, head, hlist) {
 110                 if (inst->queue_num == queue_num)
 111                         return inst;
 112         }
 113         return NULL;
 114 }
 115
 116 static struct nfqnl_instance *
 117 instance_create(struct nfnl_queue_net *q, u_int16_t queue_num, u32 portid)
 118 {
 119         struct nfqnl_instance *inst;
 120         unsigned int h;
 121         int err;
 122
 123         spin_lock(&q->instances_lock);
 124         if (instance_lookup(q, queue_num)) {
 125                 err = -EEXIST;
 126                 goto out_unlock;
 127         }
 128
 129         inst = kzalloc(sizeof(*inst), GFP_ATOMIC);
 130         if (!inst) {
 131                 err = -ENOMEM;
 132                 goto out_unlock;
 133         }
 134
 135         inst->queue_num = queue_num;
 136         inst->peer_portid = portid;
 137         inst->queue_maxlen = NFQNL_QMAX_DEFAULT;
 138         inst->copy_range = NFQNL_MAX_COPY_RANGE;
 139         inst->copy_mode = NFQNL_COPY_NONE;
 140         spin_lock_init(&inst->lock);
 141         INIT_LIST_HEAD(&inst->queue_list);
 142
 143         if (!try_module_get(THIS_MODULE)) {
 144                 err = -EAGAIN;
 145                 goto out_free;
 146         }
 147
 148         h = instance_hashfn(queue_num);
 149         hlist_add_head_rcu(&inst->hlist, &q->instance_table[h]);
 150
 151         spin_unlock(&q->instances_lock);
 152
 153         return inst;
 154
 155 out_free:
 156         kfree(inst);
 157 out_unlock:
 158         spin_unlock(&q->instances_lock);
 159         return ERR_PTR(err);
 160 }
 161
 162 static void nfqnl_flush(struct nfqnl_instance *queue, nfqnl_cmpfn cmpfn,
 163                         unsigned long data);
 164
 165 static void
 166 instance_destroy_rcu(struct rcu_head *head)
 167 {
 168         struct nfqnl_instance *inst = container_of(head, struct nfqnl_instance,
 169                                                    rcu);
 170
 171         nfqnl_flush(inst, NULL, 0);
 172         kfree(inst);
 173         module_put(THIS_MODULE);
 174 }
 175
 176 static void
 177 __instance_destroy(struct nfqnl_instance *inst)
 178 {
 179         hlist_del_rcu(&inst->hlist);
 180         call_rcu(&inst->rcu, instance_destroy_rcu);
 181 }
 182
 183 static void
 184 instance_destroy(struct nfnl_queue_net *q, struct nfqnl_instance *inst)
 185 {
 186         spin_lock(&q->instances_lock);
 187         __instance_destroy(inst);
 188         spin_unlock(&q->instances_lock);
 189 }
 190
 191 static inline void
 192 __enqueue_entry(struct nfqnl_instance *queue, struct nf_queue_entry *entry)
 193 {
 194        list_add_tail(&entry->list, &queue->queue_list);
 195        queue->queue_total++;
 196 }
 197
 198 static void
 199 __dequeue_entry(struct nfqnl_instance *queue, struct nf_queue_entry *entry)
 200 {
 201         list_del(&entry->list);
 202         queue->queue_total--;
 203 }
 204
 205 static struct nf_queue_entry *
 206 find_dequeue_entry(struct nfqnl_instance *queue, unsigned int id)
 207 {
 208         struct nf_queue_entry *entry = NULL, *i;
 209
 210         spin_lock_bh(&queue->lock);
 211
 212         list_for_each_entry(i, &queue->queue_list, list) {
 213                 if (i->id == id) {
 214                         entry = i;
 215                         break;
 216                 }
 217         }
 218
 219         if (entry)
 220                 __dequeue_entry(queue, entry);
 221
 222         spin_unlock_bh(&queue->lock);
 223
 224         return entry;
 225 }
 226
 227 static void
 228 nfqnl_flush(struct nfqnl_instance *queue, nfqnl_cmpfn cmpfn, unsigned long data)
 229 {
 230         struct nf_queue_entry *entry, *next;
 231
 232         spin_lock_bh(&queue->lock);
 233         list_for_each_entry_safe(entry, next, &queue->queue_list, list) {
 234                 if (!cmpfn || cmpfn(entry, data)) {
 235                         list_del(&entry->list);
 236                         queue->queue_total--;
 237                         nf_reinject(entry, NF_DROP);
 238                 }
 239         }
 240         spin_unlock_bh(&queue->lock);
 241 }
 242
 243 static int
 244 nfqnl_put_packet_info(struct sk_buff *nlskb, struct sk_buff *packet,
 245                       bool csum_verify)
 246 {
 247         __u32 flags = 0;
 248
 249         if (packet->ip_summed == CHECKSUM_PARTIAL)
 250                 flags = NFQA_SKB_CSUMNOTREADY;
 251         else if (csum_verify)
 252                 flags = NFQA_SKB_CSUM_NOTVERIFIED;
 253
 254         if (skb_is_gso(packet))
 255                 flags |= NFQA_SKB_GSO;
 256
 257         return flags ? nla_put_be32(nlskb, NFQA_SKB_INFO, htonl(flags)) : 0;
 258 }
 259
 260 static int nfqnl_put_sk_uidgid(struct sk_buff *skb, struct sock *sk)
 261 {
 262         const struct cred *cred;
 263
 264         if (!sk_fullsock(sk))
 265                 return 0;
 266
 267         read_lock_bh(&sk->sk_callback_lock);
 268         if (sk->sk_socket && sk->sk_socket->file) {
 269                 cred = sk->sk_socket->file->f_cred;
 270                 if (nla_put_be32(skb, NFQA_UID,
 271                     htonl(from_kuid_munged(&init_user_ns, cred->fsuid))))
 272                         goto nla_put_failure;
 273                 if (nla_put_be32(skb, NFQA_GID,
 274                     htonl(from_kgid_munged(&init_user_ns, cred->fsgid))))
 275                         goto nla_put_failure;
 276         }
 277         read_unlock_bh(&sk->sk_callback_lock);
 278         return 0;
 279
 280 nla_put_failure:
 281         read_unlock_bh(&sk->sk_callback_lock);
 282         return -1;
 283 }
 284
 285 static u32 nfqnl_get_sk_secctx(struct sk_buff *skb, char **secdata)
 286 {
 287         u32 seclen = 0;
 288 #if IS_ENABLED(CONFIG_NETWORK_SECMARK)
 289         if (!skb || !sk_fullsock(skb->sk))
 290                 return 0;
 291
 292         read_lock_bh(&skb->sk->sk_callback_lock);
 293
 294         if (skb->secmark)
 295                 security_secid_to_secctx(skb->secmark, secdata, &seclen);
 296
 297         read_unlock_bh(&skb->sk->sk_callback_lock);
 298 #endif
 299         return seclen;
 300 }
 301
 302 static u32 nfqnl_get_bridge_size(struct nf_queue_entry *entry)
 303 {
 304         struct sk_buff *entskb = entry->skb;
 305         u32 nlalen = 0;
 306
 307         if (entry->state.pf != PF_BRIDGE || !skb_mac_header_was_set(entskb))
 308                 return 0;
 309
 310         if (skb_vlan_tag_present(entskb))
 311                 nlalen += nla_total_size(nla_total_size(sizeof(__be16)) +
 312                                          nla_total_size(sizeof(__be16)));
 313
 314         if (entskb->network_header > entskb->mac_header)
 315                 nlalen += nla_total_size((entskb->network_header -
 316                                           entskb->mac_header));
 317
 318         return nlalen;
 319 }
 320
 321 static int nfqnl_put_bridge(struct nf_queue_entry *entry, struct sk_buff *skb)
 322 {
 323         struct sk_buff *entskb = entry->skb;
 324
 325         if (entry->state.pf != PF_BRIDGE || !skb_mac_header_was_set(entskb))
 326                 return 0;
 327
 328         if (skb_vlan_tag_present(entskb)) {
 329                 struct nlattr *nest;
 330
 331                 nest = nla_nest_start(skb, NFQA_VLAN | NLA_F_NESTED);
 332                 if (!nest)
 333                         goto nla_put_failure;
 334
 335                 if (nla_put_be16(skb, NFQA_VLAN_TCI, htons(entskb->vlan_tci)) ||
 336                     nla_put_be16(skb, NFQA_VLAN_PROTO, entskb->vlan_proto))
 337                         goto nla_put_failure;
 338
 339                 nla_nest_end(skb, nest);
 340         }
 341
 342         if (entskb->mac_header < entskb->network_header) {
 343                 int len = (int)(entskb->network_header - entskb->mac_header);
 344
 345                 if (nla_put(skb, NFQA_L2HDR, len, skb_mac_header(entskb)))
 346                         goto nla_put_failure;
 347         }
 348
 349         return 0;
 350
 351 nla_put_failure:
 352         return -1;
 353 }
 354
 355 static struct sk_buff *
 356 nfqnl_build_packet_message(struct net *net, struct nfqnl_instance *queue,
 357                            struct nf_queue_entry *entry,
 358                            __be32 **packet_id_ptr)
 359 {
 360         size_t size;
 361         size_t data_len = 0, cap_len = 0;
 362         unsigned int hlen = 0;
 363         struct sk_buff *skb;
 364         struct nlattr *nla;
 365         struct nfqnl_msg_packet_hdr *pmsg;
 366         struct nlmsghdr *nlh;
 367         struct nfgenmsg *nfmsg;
 368         struct sk_buff *entskb = entry->skb;
 369         struct net_device *indev;
 370         struct net_device *outdev;
 371         struct nf_conn *ct = NULL;
 372         enum ip_conntrack_info uninitialized_var(ctinfo);
 373         struct nfnl_ct_hook *nfnl_ct;
 374         bool csum_verify;
 375         char *secdata = NULL;
 376         u32 seclen = 0;
 377
 378         size =    nlmsg_total_size(sizeof(struct nfgenmsg))
 379                 + nla_total_size(sizeof(struct nfqnl_msg_packet_hdr))
 380                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
 381                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
 382 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
 383                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
 384                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
 385 #endif
 386                 + nla_total_size(sizeof(u_int32_t))     /* mark */
 387                 + nla_total_size(sizeof(struct nfqnl_msg_packet_hw))
 388                 + nla_total_size(sizeof(u_int32_t))     /* skbinfo */
 389                 + nla_total_size(sizeof(u_int32_t));    /* cap_len */
 390
 391         if (entskb->tstamp)
 392                 size += nla_total_size(sizeof(struct nfqnl_msg_packet_timestamp));
 393
 394         size += nfqnl_get_bridge_size(entry);
 395
 396         if (entry->state.hook <= NF_INET_FORWARD ||
 397            (entry->state.hook == NF_INET_POST_ROUTING && entskb->sk == NULL))
 398                 csum_verify = !skb_csum_unnecessary(entskb);
 399         else
 400                 csum_verify = false;
 401
 402         outdev = entry->state.out;
 403
 404         switch ((enum nfqnl_config_mode)ACCESS_ONCE(queue->copy_mode)) {
 405         case NFQNL_COPY_META:
 406         case NFQNL_COPY_NONE:
 407                 break;
 408
 409         case NFQNL_COPY_PACKET:
 410                 if (!(queue->flags & NFQA_CFG_F_GSO) &&
 411                     entskb->ip_summed == CHECKSUM_PARTIAL &&
 412                     skb_checksum_help(entskb))
 413                         return NULL;
 414
 415                 data_len = ACCESS_ONCE(queue->copy_range);
 416                 if (data_len > entskb->len)
 417                         data_len = entskb->len;
 418
 419                 hlen = skb_zerocopy_headlen(entskb);
 420                 hlen = min_t(unsigned int, hlen, data_len);
 421                 size += sizeof(struct nlattr) + hlen;
 422                 cap_len = entskb->len;
 423                 break;
 424         }
 425
 426         nfnl_ct = rcu_dereference(nfnl_ct_hook);
 427
 428         if (queue->flags & NFQA_CFG_F_CONNTRACK) {
 429                 if (nfnl_ct != NULL) {
 430                         ct = nfnl_ct->get_ct(entskb, &ctinfo);
 431                         if (ct != NULL)
 432                                 size += nfnl_ct->build_size(ct);
 433                 }
 434         }
 435
 436         if (queue->flags & NFQA_CFG_F_UID_GID) {
 437                 size +=  (nla_total_size(sizeof(u_int32_t))     /* uid */
 438                         + nla_total_size(sizeof(u_int32_t)));   /* gid */
 439         }
 440
 441         if ((queue->flags & NFQA_CFG_F_SECCTX) && entskb->sk) {
 442                 seclen = nfqnl_get_sk_secctx(entskb, &secdata);
 443                 if (seclen)
 444                         size += nla_total_size(seclen);
 445         }
 446
 447         skb = alloc_skb(size, GFP_ATOMIC);
 448         if (!skb) {
 449                 skb_tx_error(entskb);
 450                 goto nlmsg_failure;
 451         }
 452
 453         nlh = nlmsg_put(skb, 0, 0,
 454                         nfnl_msg_type(NFNL_SUBSYS_QUEUE, NFQNL_MSG_PACKET),
 455                         sizeof(struct nfgenmsg), 0);
 456         if (!nlh) {
 457                 skb_tx_error(entskb);
 458                 kfree_skb(skb);
 459                 goto nlmsg_failure;
 460         }
 461         nfmsg = nlmsg_data(nlh);
 462         nfmsg->nfgen_family = entry->state.pf;
 463         nfmsg->version = NFNETLINK_V0;
 464         nfmsg->res_id = htons(queue->queue_num);
 465
 466         nla = __nla_reserve(skb, NFQA_PACKET_HDR, sizeof(*pmsg));
 467         pmsg = nla_data(nla);
 468         pmsg->hw_protocol       = entskb->protocol;
 469         pmsg->hook              = entry->state.hook;
 470         *packet_id_ptr          = &pmsg->packet_id;
 471
 472         indev = entry->state.in;
 473         if (indev) {
 474 #if !IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
 475                 if (nla_put_be32(skb, NFQA_IFINDEX_INDEV, htonl(indev->ifindex)))
 476                         goto nla_put_failure;
 477 #else
 478                 if (entry->state.pf == PF_BRIDGE) {
 479                         /* Case 1: indev is physical input device, we need to
 480                          * look for bridge group (when called from
 481                          * netfilter_bridge) */
 482                         if (nla_put_be32(skb, NFQA_IFINDEX_PHYSINDEV,
 483                                          htonl(indev->ifindex)) ||
 484                         /* this is the bridge group "brX" */
 485                         /* rcu_read_lock()ed by __nf_queue */
 486                             nla_put_be32(skb, NFQA_IFINDEX_INDEV,
 487                                          htonl(br_port_get_rcu(indev)->br->dev->ifindex)))
 488                                 goto nla_put_failure;
 489                 } else {
 490                         int physinif;
 491
 492                         /* Case 2: indev is bridge group, we need to look for
 493                          * physical device (when called from ipv4) */
 494                         if (nla_put_be32(skb, NFQA_IFINDEX_INDEV,
 495                                          htonl(indev->ifindex)))
 496                                 goto nla_put_failure;
 497
 498                         physinif = nf_bridge_get_physinif(entskb);
 499                         if (physinif &&
 500                             nla_put_be32(skb, NFQA_IFINDEX_PHYSINDEV,
 501                                          htonl(physinif)))
 502                                 goto nla_put_failure;
 503                 }
 504 #endif
 505         }
 506
 507         if (outdev) {
 508 #if !IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
 509                 if (nla_put_be32(skb, NFQA_IFINDEX_OUTDEV, htonl(outdev->ifindex)))
 510                         goto nla_put_failure;
 511 #else
 512                 if (entry->state.pf == PF_BRIDGE) {
 513                         /* Case 1: outdev is physical output device, we need to
 514                          * look for bridge group (when called from
 515                          * netfilter_bridge) */
 516                         if (nla_put_be32(skb, NFQA_IFINDEX_PHYSOUTDEV,
 517                                          htonl(outdev->ifindex)) ||
 518                         /* this is the bridge group "brX" */
 519                         /* rcu_read_lock()ed by __nf_queue */
 520                             nla_put_be32(skb, NFQA_IFINDEX_OUTDEV,
 521                                          htonl(br_port_get_rcu(outdev)->br->dev->ifindex)))
 522                                 goto nla_put_failure;
 523                 } else {
 524                         int physoutif;
 525
 526                         /* Case 2: outdev is bridge group, we need to look for
 527                          * physical output device (when called from ipv4) */
 528                         if (nla_put_be32(skb, NFQA_IFINDEX_OUTDEV,
 529                                          htonl(outdev->ifindex)))
 530                                 goto nla_put_failure;
 531
 532                         physoutif = nf_bridge_get_physoutif(entskb);
 533                         if (physoutif &&
 534                             nla_put_be32(skb, NFQA_IFINDEX_PHYSOUTDEV,
 535                                          htonl(physoutif)))
 536                                 goto nla_put_failure;
 537                 }
 538 #endif
 539         }
 540
 541         if (entskb->mark &&
 542             nla_put_be32(skb, NFQA_MARK, htonl(entskb->mark)))
 543                 goto nla_put_failure;
 544
 545         if (indev && entskb->dev &&
 546             entskb->mac_header != entskb->network_header) {
 547                 struct nfqnl_msg_packet_hw phw;
 548                 int len;
 549
 550                 memset(&phw, 0, sizeof(phw));
 551                 len = dev_parse_header(entskb, phw.hw_addr);
 552                 if (len) {
 553                         phw.hw_addrlen = htons(len);
 554                         if (nla_put(skb, NFQA_HWADDR, sizeof(phw), &phw))
 555                                 goto nla_put_failure;
 556                 }
 557         }
 558
 559         if (nfqnl_put_bridge(entry, skb) < 0)
 560                 goto nla_put_failure;
 561
 562         if (entskb->tstamp) {
 563                 struct nfqnl_msg_packet_timestamp ts;
 564                 struct timespec64 kts = ktime_to_timespec64(entskb->tstamp);
 565
 566                 ts.sec = cpu_to_be64(kts.tv_sec);
 567                 ts.usec = cpu_to_be64(kts.tv_nsec / NSEC_PER_USEC);
 568
 569                 if (nla_put(skb, NFQA_TIMESTAMP, sizeof(ts), &ts))
 570                         goto nla_put_failure;
 571         }
 572
 573         if ((queue->flags & NFQA_CFG_F_UID_GID) && entskb->sk &&
 574             nfqnl_put_sk_uidgid(skb, entskb->sk) < 0)
 575                 goto nla_put_failure;
 576
 577         if (seclen && nla_put(skb, NFQA_SECCTX, seclen, secdata))
 578                 goto nla_put_failure;
 579
 580         if (ct && nfnl_ct->build(skb, ct, ctinfo, NFQA_CT, NFQA_CT_INFO) < 0)
 581                 goto nla_put_failure;
 582
 583         if (cap_len > data_len &&
 584             nla_put_be32(skb, NFQA_CAP_LEN, htonl(cap_len)))
 585                 goto nla_put_failure;
 586
 587         if (nfqnl_put_packet_info(skb, entskb, csum_verify))
 588                 goto nla_put_failure;
 589
 590         if (data_len) {
 591                 struct nlattr *nla;
 592
 593                 if (skb_tailroom(skb) < sizeof(*nla) + hlen)
 594                         goto nla_put_failure;
 595
 596                 nla = skb_put(skb, sizeof(*nla));
 597                 nla->nla_type = NFQA_PAYLOAD;
 598                 nla->nla_len = nla_attr_size(data_len);
 599
 600                 if (skb_zerocopy(skb, entskb, data_len, hlen))
 601                         goto nla_put_failure;
 602         }
 603
 604         nlh->nlmsg_len = skb->len;
 605         if (seclen)
 606                 security_release_secctx(secdata, seclen);
 607         return skb;
 608
 609 nla_put_failure:
 610         skb_tx_error(entskb);
 611         kfree_skb(skb);
 612         net_err_ratelimited("nf_queue: error creating packet message\n");
 613 nlmsg_failure:
 614         if (seclen)
 615                 security_release_secctx(secdata, seclen);
 616         return NULL;
 617 }
 618
 619 static bool nf_ct_drop_unconfirmed(const struct nf_queue_entry *entry)
 620 {
 621 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
 622         static const unsigned long flags = IPS_CONFIRMED | IPS_DYING;
 623         const struct nf_conn *ct = (void *)skb_nfct(entry->skb);
 624
 625         if (ct && ((ct->status & flags) == IPS_DYING))
 626                 return true;
 627 #endif
 628         return false;
 629 }
 630
 631 static int
 632 __nfqnl_enqueue_packet(struct net *net, struct nfqnl_instance *queue,
 633                         struct nf_queue_entry *entry)
 634 {
 635         struct sk_buff *nskb;
 636         int err = -ENOBUFS;
 637         __be32 *packet_id_ptr;
 638         int failopen = 0;
 639
 640         nskb = nfqnl_build_packet_message(net, queue, entry, &packet_id_ptr);
 641         if (nskb == NULL) {
 642                 err = -ENOMEM;
 643                 goto err_out;
 644         }
 645         spin_lock_bh(&queue->lock);
 646
 647         if (nf_ct_drop_unconfirmed(entry))
 648                 goto err_out_free_nskb;
 649
 650         if (queue->queue_total >= queue->queue_maxlen) {
 651                 if (queue->flags & NFQA_CFG_F_FAIL_OPEN) {
 652                         failopen = 1;
 653                         err = 0;
 654                 } else {
 655                         queue->queue_dropped++;
 656                         net_warn_ratelimited("nf_queue: full at %d entries, dropping packets(s)\n",
 657                                              queue->queue_total);
 658                 }
 659                 goto err_out_free_nskb;
 660         }
 661         entry->id = ++queue->id_sequence;
 662         *packet_id_ptr = htonl(entry->id);
 663
 664         /* nfnetlink_unicast will either free the nskb or add it to a socket */
 665         err = nfnetlink_unicast(nskb, net, queue->peer_portid, MSG_DONTWAIT);
 666         if (err < 0) {
 667                 if (queue->flags & NFQA_CFG_F_FAIL_OPEN) {
 668                         failopen = 1;
 669                         err = 0;
 670                 } else {
 671                         queue->queue_user_dropped++;
 672                 }
 673                 goto err_out_unlock;
 674         }
 675
 676         __enqueue_entry(queue, entry);
 677
 678         spin_unlock_bh(&queue->lock);
 679         return 0;
 680
 681 err_out_free_nskb:
 682         kfree_skb(nskb);
 683 err_out_unlock:
 684         spin_unlock_bh(&queue->lock);
 685         if (failopen)
 686                 nf_reinject(entry, NF_ACCEPT);
 687 err_out:
 688         return err;
 689 }
 690
 691 static struct nf_queue_entry *
 692 nf_queue_entry_dup(struct nf_queue_entry *e)
 693 {
 694         struct nf_queue_entry *entry = kmemdup(e, e->size, GFP_ATOMIC);
 695         if (entry)
 696                 nf_queue_entry_get_refs(entry);
 697         return entry;
 698 }
 699
 700 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
 701 /* When called from bridge netfilter, skb->data must point to MAC header
 702  * before calling skb_gso_segment(). Else, original MAC header is lost
 703  * and segmented skbs will be sent to wrong destination.
 704  */
 705 static void nf_bridge_adjust_skb_data(struct sk_buff *skb)
 706 {
 707         if (skb->nf_bridge)
 708                 __skb_push(skb, skb->network_header - skb->mac_header);
 709 }
 710
 711 static void nf_bridge_adjust_segmented_data(struct sk_buff *skb)
 712 {
 713         if (skb->nf_bridge)
 714                 __skb_pull(skb, skb->network_header - skb->mac_header);
 715 }
 716 #else
 717 #define nf_bridge_adjust_skb_data(s) do {} while (0)
 718 #define nf_bridge_adjust_segmented_data(s) do {} while (0)
 719 #endif
 720
 721 static void free_entry(struct nf_queue_entry *entry)
 722 {
 723         nf_queue_entry_release_refs(entry);
 724         kfree(entry);
 725 }
 726
 727 static int
 728 __nfqnl_enqueue_packet_gso(struct net *net, struct nfqnl_instance *queue,
 729                            struct sk_buff *skb, struct nf_queue_entry *entry)
 730 {
 731         int ret = -ENOMEM;
 732         struct nf_queue_entry *entry_seg;
 733
 734         nf_bridge_adjust_segmented_data(skb);
 735
 736         if (skb->next == NULL) { /* last packet, no need to copy entry */
 737                 struct sk_buff *gso_skb = entry->skb;
 738                 entry->skb = skb;
 739                 ret = __nfqnl_enqueue_packet(net, queue, entry);
 740                 if (ret)
 741                         entry->skb = gso_skb;
 742                 return ret;
 743         }
 744
 745         skb->next = NULL;
 746
 747         entry_seg = nf_queue_entry_dup(entry);
 748         if (entry_seg) {
 749                 entry_seg->skb = skb;
 750                 ret = __nfqnl_enqueue_packet(net, queue, entry_seg);
 751                 if (ret)
 752                         free_entry(entry_seg);
 753         }
 754         return ret;
 755 }
 756
 757 static int
 758 nfqnl_enqueue_packet(struct nf_queue_entry *entry, unsigned int queuenum)
 759 {
 760         unsigned int queued;
 761         struct nfqnl_instance *queue;
 762         struct sk_buff *skb, *segs;
 763         int err = -ENOBUFS;
 764         struct net *net = entry->state.net;
 765         struct nfnl_queue_net *q = nfnl_queue_pernet(net);
 766
 767         /* rcu_read_lock()ed by nf_hook_thresh */
 768         queue = instance_lookup(q, queuenum);
 769         if (!queue)
 770                 return -ESRCH;
 771
 772         if (queue->copy_mode == NFQNL_COPY_NONE)
 773                 return -EINVAL;
 774
 775         skb = entry->skb;
 776
 777         switch (entry->state.pf) {
 778         case NFPROTO_IPV4:
 779                 skb->protocol = htons(ETH_P_IP);
 780                 break;
 781         case NFPROTO_IPV6:
 782                 skb->protocol = htons(ETH_P_IPV6);
 783                 break;
 784         }
 785
 786         if ((queue->flags & NFQA_CFG_F_GSO) || !skb_is_gso(skb))
 787                 return __nfqnl_enqueue_packet(net, queue, entry);
 788
 789         nf_bridge_adjust_skb_data(skb);
 790         segs = skb_gso_segment(skb, 0);
 791         /* Does not use PTR_ERR to limit the number of error codes that can be
 792          * returned by nf_queue.  For instance, callers rely on -ESRCH to
 793          * mean 'ignore this hook'.
 794          */
 795         if (IS_ERR_OR_NULL(segs))
 796                 goto out_err;
 797         queued = 0;
 798         err = 0;
 799         do {
 800                 struct sk_buff *nskb = segs->next;
 801                 if (err == 0)
 802                         err = __nfqnl_enqueue_packet_gso(net, queue,
 803                                                         segs, entry);
 804                 if (err == 0)
 805                         queued++;
 806                 else
 807                         kfree_skb(segs);
 808                 segs = nskb;
 809         } while (segs);
 810
 811         if (queued) {
 812                 if (err) /* some segments are already queued */
 813                         free_entry(entry);
 814                 kfree_skb(skb);
 815                 return 0;
 816         }
 817  out_err:
 818         nf_bridge_adjust_segmented_data(skb);
 819         return err;
 820 }
 821
 822 static int
 823 nfqnl_mangle(void *data, int data_len, struct nf_queue_entry *e, int diff)
 824 {
 825         struct sk_buff *nskb;
 826
 827         if (diff < 0) {
 828                 if (pskb_trim(e->skb, data_len))
 829                         return -ENOMEM;
 830         } else if (diff > 0) {
 831                 if (data_len > 0xFFFF)
 832                         return -EINVAL;
 833                 if (diff > skb_tailroom(e->skb)) {
 834                         nskb = skb_copy_expand(e->skb, skb_headroom(e->skb),
 835                                                diff, GFP_ATOMIC);
 836                         if (!nskb) {
 837                                 printk(KERN_WARNING "nf_queue: OOM "
 838                                       "in mangle, dropping packet\n");
 839                                 return -ENOMEM;
 840                         }
 841                         kfree_skb(e->skb);
 842                         e->skb = nskb;
 843                 }
 844                 skb_put(e->skb, diff);
 845         }
 846         if (!skb_make_writable(e->skb, data_len))
 847                 return -ENOMEM;
 848         skb_copy_to_linear_data(e->skb, data, data_len);
 849         e->skb->ip_summed = CHECKSUM_NONE;
 850         return 0;
 851 }
 852
 853 static int
 854 nfqnl_set_mode(struct nfqnl_instance *queue,
 855                unsigned char mode, unsigned int range)
 856 {
 857         int status = 0;
 858
 859         spin_lock_bh(&queue->lock);
 860         switch (mode) {
 861         case NFQNL_COPY_NONE:
 862         case NFQNL_COPY_META:
 863                 queue->copy_mode = mode;
 864                 queue->copy_range = 0;
 865                 break;
 866
 867         case NFQNL_COPY_PACKET:
 868                 queue->copy_mode = mode;
 869                 if (range == 0 || range > NFQNL_MAX_COPY_RANGE)
 870                         queue->copy_range = NFQNL_MAX_COPY_RANGE;
 871                 else
 872                         queue->copy_range = range;
 873                 break;
 874
 875         default:
 876                 status = -EINVAL;
 877
 878         }
 879         spin_unlock_bh(&queue->lock);
 880
 881         return status;
 882 }
 883
 884 static int
 885 dev_cmp(struct nf_queue_entry *entry, unsigned long ifindex)
 886 {
 887         if (entry->state.in)
 888                 if (entry->state.in->ifindex == ifindex)
 889                         return 1;
 890         if (entry->state.out)
 891                 if (entry->state.out->ifindex == ifindex)
 892                         return 1;
 893 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
 894         if (entry->skb->nf_bridge) {
 895                 int physinif, physoutif;
 896
 897                 physinif = nf_bridge_get_physinif(entry->skb);
 898                 physoutif = nf_bridge_get_physoutif(entry->skb);
 899
 900                 if (physinif == ifindex || physoutif == ifindex)
 901                         return 1;
 902         }
 903 #endif
 904         return 0;
 905 }
 906
 907 /* drop all packets with either indev or outdev == ifindex from all queue
 908  * instances */
 909 static void
 910 nfqnl_dev_drop(struct net *net, int ifindex)
 911 {
 912         int i;
 913         struct nfnl_queue_net *q = nfnl_queue_pernet(net);
 914
 915         rcu_read_lock();
 916
 917         for (i = 0; i < INSTANCE_BUCKETS; i++) {
 918                 struct nfqnl_instance *inst;
 919                 struct hlist_head *head = &q->instance_table[i];
 920
 921                 hlist_for_each_entry_rcu(inst, head, hlist)
 922                         nfqnl_flush(inst, dev_cmp, ifindex);
 923         }
 924
 925         rcu_read_unlock();
 926 }
 927
 928 static int
 929 nfqnl_rcv_dev_event(struct notifier_block *this,
 930                     unsigned long event, void *ptr)
 931 {
 932         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
 933
 934         /* Drop any packets associated with the downed device */
 935         if (event == NETDEV_DOWN)
 936                 nfqnl_dev_drop(dev_net(dev), dev->ifindex);
 937         return NOTIFY_DONE;
 938 }
 939
 940 static struct notifier_block nfqnl_dev_notifier = {
 941         .notifier_call  = nfqnl_rcv_dev_event,
 942 };
 943
 944 static unsigned int nfqnl_nf_hook_drop(struct net *net)
 945 {
 946         struct nfnl_queue_net *q = nfnl_queue_pernet(net);
 947         unsigned int instances = 0;
 948         int i;
 949
 950         for (i = 0; i < INSTANCE_BUCKETS; i++) {
 951                 struct nfqnl_instance *inst;
 952                 struct hlist_head *head = &q->instance_table[i];
 953
 954                 hlist_for_each_entry_rcu(inst, head, hlist) {
 955                         nfqnl_flush(inst, NULL, 0);
 956                         instances++;
 957                 }
 958         }
 959
 960         return instances;
 961 }
 962
 963 static int
 964 nfqnl_rcv_nl_event(struct notifier_block *this,
 965                    unsigned long event, void *ptr)
 966 {
 967         struct netlink_notify *n = ptr;
 968         struct nfnl_queue_net *q = nfnl_queue_pernet(n->net);
 969
 970         if (event == NETLINK_URELEASE && n->protocol == NETLINK_NETFILTER) {
 971                 int i;
 972
 973                 /* destroy all instances for this portid */
 974                 spin_lock(&q->instances_lock);
 975                 for (i = 0; i < INSTANCE_BUCKETS; i++) {
 976                         struct hlist_node *t2;
 977                         struct nfqnl_instance *inst;
 978                         struct hlist_head *head = &q->instance_table[i];
 979
 980                         hlist_for_each_entry_safe(inst, t2, head, hlist) {
 981                                 if (n->portid == inst->peer_portid)
 982                                         __instance_destroy(inst);
 983                         }
 984                 }
 985                 spin_unlock(&q->instances_lock);
 986         }
 987         return NOTIFY_DONE;
 988 }
 989
 990 static struct notifier_block nfqnl_rtnl_notifier = {
 991         .notifier_call  = nfqnl_rcv_nl_event,
 992 };
 993
 994 static const struct nla_policy nfqa_vlan_policy[NFQA_VLAN_MAX + 1] = {
 995         [NFQA_VLAN_TCI]         = { .type = NLA_U16},
 996         [NFQA_VLAN_PROTO]       = { .type = NLA_U16},
 997 };
 998
 999 static const struct nla_policy nfqa_verdict_policy[NFQA_MAX+1] = {
1000         [NFQA_VERDICT_HDR]      = { .len = sizeof(struct nfqnl_msg_verdict_hdr) },
1001         [NFQA_MARK]             = { .type = NLA_U32 },
1002         [NFQA_PAYLOAD]          = { .type = NLA_UNSPEC },
1003         [NFQA_CT]               = { .type = NLA_UNSPEC },
1004         [NFQA_EXP]              = { .type = NLA_UNSPEC },
1005         [NFQA_VLAN]             = { .type = NLA_NESTED },
1006 };
1007
1008 static const struct nla_policy nfqa_verdict_batch_policy[NFQA_MAX+1] = {
1009         [NFQA_VERDICT_HDR]      = { .len = sizeof(struct nfqnl_msg_verdict_hdr) },
1010         [NFQA_MARK]             = { .type = NLA_U32 },
1011 };
1012
1013 static struct nfqnl_instance *
1014 verdict_instance_lookup(struct nfnl_queue_net *q, u16 queue_num, u32 nlportid)
1015 {
1016         struct nfqnl_instance *queue;
1017
1018         queue = instance_lookup(q, queue_num);
1019         if (!queue)
1020                 return ERR_PTR(-ENODEV);
1021
1022         if (queue->peer_portid != nlportid)
1023                 return ERR_PTR(-EPERM);
1024
1025         return queue;
1026 }
1027
1028 static struct nfqnl_msg_verdict_hdr*
1029 verdicthdr_get(const struct nlattr * const nfqa[])
1030 {
1031         struct nfqnl_msg_verdict_hdr *vhdr;
1032         unsigned int verdict;
1033
1034         if (!nfqa[NFQA_VERDICT_HDR])
1035                 return NULL;
1036
1037         vhdr = nla_data(nfqa[NFQA_VERDICT_HDR]);
1038         verdict = ntohl(vhdr->verdict) & NF_VERDICT_MASK;
1039         if (verdict > NF_MAX_VERDICT || verdict == NF_STOLEN)
1040                 return NULL;
1041         return vhdr;
1042 }
1043
1044 static int nfq_id_after(unsigned int id, unsigned int max)
1045 {
1046         return (int)(id - max) > 0;
1047 }
1048
1049 static int nfqnl_recv_verdict_batch(struct net *net, struct sock *ctnl,
1050                                     struct sk_buff *skb,
1051                                     const struct nlmsghdr *nlh,
1052                                     const struct nlattr * const nfqa[],
1053                                     struct netlink_ext_ack *extack)
1054 {
1055         struct nfgenmsg *nfmsg = nlmsg_data(nlh);
1056         struct nf_queue_entry *entry, *tmp;
1057         unsigned int verdict, maxid;
1058         struct nfqnl_msg_verdict_hdr *vhdr;
1059         struct nfqnl_instance *queue;
1060         LIST_HEAD(batch_list);
1061         u16 queue_num = ntohs(nfmsg->res_id);
1062         struct nfnl_queue_net *q = nfnl_queue_pernet(net);
1063
1064         queue = verdict_instance_lookup(q, queue_num,
1065                                         NETLINK_CB(skb).portid);
1066         if (IS_ERR(queue))
1067                 return PTR_ERR(queue);
1068
1069         vhdr = verdicthdr_get(nfqa);
1070         if (!vhdr)
1071                 return -EINVAL;
1072
1073         verdict = ntohl(vhdr->verdict);
1074         maxid = ntohl(vhdr->id);
1075
1076         spin_lock_bh(&queue->lock);
1077
1078         list_for_each_entry_safe(entry, tmp, &queue->queue_list, list) {
1079                 if (nfq_id_after(entry->id, maxid))
1080                         break;
1081                 __dequeue_entry(queue, entry);
1082                 list_add_tail(&entry->list, &batch_list);
1083         }
1084
1085         spin_unlock_bh(&queue->lock);
1086
1087         if (list_empty(&batch_list))
1088                 return -ENOENT;
1089
1090         list_for_each_entry_safe(entry, tmp, &batch_list, list) {
1091                 if (nfqa[NFQA_MARK])
1092                         entry->skb->mark = ntohl(nla_get_be32(nfqa[NFQA_MARK]));
1093                 nf_reinject(entry, verdict);
1094         }
1095         return 0;
1096 }
1097
1098 static struct nf_conn *nfqnl_ct_parse(struct nfnl_ct_hook *nfnl_ct,
1099                                       const struct nlmsghdr *nlh,
1100                                       const struct nlattr * const nfqa[],
1101                                       struct nf_queue_entry *entry,
1102                                       enum ip_conntrack_info *ctinfo)
1103 {
1104         struct nf_conn *ct;
1105
1106         ct = nfnl_ct->get_ct(entry->skb, ctinfo);
1107         if (ct == NULL)
1108                 return NULL;
1109
1110         if (nfnl_ct->parse(nfqa[NFQA_CT], ct) < 0)
1111                 return NULL;
1112
1113         if (nfqa[NFQA_EXP])
1114                 nfnl_ct->attach_expect(nfqa[NFQA_EXP], ct,
1115                                       NETLINK_CB(entry->skb).portid,
1116                                       nlmsg_report(nlh));
1117         return ct;
1118 }
1119
1120 static int nfqa_parse_bridge(struct nf_queue_entry *entry,
1121                              const struct nlattr * const nfqa[])
1122 {
1123         if (nfqa[NFQA_VLAN]) {
1124                 struct nlattr *tb[NFQA_VLAN_MAX + 1];
1125                 int err;
1126
1127                 err = nla_parse_nested(tb, NFQA_VLAN_MAX, nfqa[NFQA_VLAN],
1128                                        nfqa_vlan_policy, NULL);
1129                 if (err < 0)
1130                         return err;
1131
1132                 if (!tb[NFQA_VLAN_TCI] || !tb[NFQA_VLAN_PROTO])
1133                         return -EINVAL;
1134
1135                 entry->skb->vlan_tci = ntohs(nla_get_be16(tb[NFQA_VLAN_TCI]));
1136                 entry->skb->vlan_proto = nla_get_be16(tb[NFQA_VLAN_PROTO]);
1137         }
1138
1139         if (nfqa[NFQA_L2HDR]) {
1140                 int mac_header_len = entry->skb->network_header -
1141                         entry->skb->mac_header;
1142
1143                 if (mac_header_len != nla_len(nfqa[NFQA_L2HDR]))
1144                         return -EINVAL;
1145                 else if (mac_header_len > 0)
1146                         memcpy(skb_mac_header(entry->skb),
1147                                nla_data(nfqa[NFQA_L2HDR]),
1148                                mac_header_len);
1149         }
1150
1151         return 0;
1152 }
1153
1154 static int nfqnl_recv_verdict(struct net *net, struct sock *ctnl,
1155                               struct sk_buff *skb,
1156                               const struct nlmsghdr *nlh,
1157                               const struct nlattr * const nfqa[],
1158                               struct netlink_ext_ack *extack)
1159 {
1160         struct nfgenmsg *nfmsg = nlmsg_data(nlh);
1161         u_int16_t queue_num = ntohs(nfmsg->res_id);
1162         struct nfqnl_msg_verdict_hdr *vhdr;
1163         struct nfqnl_instance *queue;
1164         unsigned int verdict;
1165         struct nf_queue_entry *entry;
1166         enum ip_conntrack_info uninitialized_var(ctinfo);
1167         struct nfnl_ct_hook *nfnl_ct;
1168         struct nf_conn *ct = NULL;
1169         struct nfnl_queue_net *q = nfnl_queue_pernet(net);
1170         int err;
1171
1172         queue = verdict_instance_lookup(q, queue_num,
1173                                         NETLINK_CB(skb).portid);
1174         if (IS_ERR(queue))
1175                 return PTR_ERR(queue);
1176
1177         vhdr = verdicthdr_get(nfqa);
1178         if (!vhdr)
1179                 return -EINVAL;
1180
1181         verdict = ntohl(vhdr->verdict);
1182
1183         entry = find_dequeue_entry(queue, ntohl(vhdr->id));
1184         if (entry == NULL)
1185                 return -ENOENT;
1186
1187         /* rcu lock already held from nfnl->call_rcu. */
1188         nfnl_ct = rcu_dereference(nfnl_ct_hook);
1189
1190         if (nfqa[NFQA_CT]) {
1191                 if (nfnl_ct != NULL)
1192                         ct = nfqnl_ct_parse(nfnl_ct, nlh, nfqa, entry, &ctinfo);
1193         }
1194
1195         if (entry->state.pf == PF_BRIDGE) {
1196                 err = nfqa_parse_bridge(entry, nfqa);
1197                 if (err < 0)
1198                         return err;
1199         }
1200
1201         if (nfqa[NFQA_PAYLOAD]) {
1202                 u16 payload_len = nla_len(nfqa[NFQA_PAYLOAD]);
1203                 int diff = payload_len - entry->skb->len;
1204
1205                 if (nfqnl_mangle(nla_data(nfqa[NFQA_PAYLOAD]),
1206                                  payload_len, entry, diff) < 0)
1207                         verdict = NF_DROP;
1208
1209                 if (ct && diff)
1210                         nfnl_ct->seq_adjust(entry->skb, ct, ctinfo, diff);
1211         }
1212
1213         if (nfqa[NFQA_MARK])
1214                 entry->skb->mark = ntohl(nla_get_be32(nfqa[NFQA_MARK]));
1215
1216         nf_reinject(entry, verdict);
1217         return 0;
1218 }
1219
1220 static int nfqnl_recv_unsupp(struct net *net, struct sock *ctnl,
1221                              struct sk_buff *skb, const struct nlmsghdr *nlh,
1222                              const struct nlattr * const nfqa[],
1223                              struct netlink_ext_ack *extack)
1224 {
1225         return -ENOTSUPP;
1226 }
1227
1228 static const struct nla_policy nfqa_cfg_policy[NFQA_CFG_MAX+1] = {
1229         [NFQA_CFG_CMD]          = { .len = sizeof(struct nfqnl_msg_config_cmd) },
1230         [NFQA_CFG_PARAMS]       = { .len = sizeof(struct nfqnl_msg_config_params) },
1231         [NFQA_CFG_QUEUE_MAXLEN] = { .type = NLA_U32 },
1232         [NFQA_CFG_MASK]         = { .type = NLA_U32 },
1233         [NFQA_CFG_FLAGS]        = { .type = NLA_U32 },
1234 };
1235
1236 static const struct nf_queue_handler nfqh = {
1237         .outfn          = nfqnl_enqueue_packet,
1238         .nf_hook_drop   = nfqnl_nf_hook_drop,
1239 };
1240
1241 static int nfqnl_recv_config(struct net *net, struct sock *ctnl,
1242                              struct sk_buff *skb, const struct nlmsghdr *nlh,
1243                              const struct nlattr * const nfqa[],
1244                              struct netlink_ext_ack *extack)
1245 {
1246         struct nfgenmsg *nfmsg = nlmsg_data(nlh);
1247         u_int16_t queue_num = ntohs(nfmsg->res_id);
1248         struct nfqnl_instance *queue;
1249         struct nfqnl_msg_config_cmd *cmd = NULL;
1250         struct nfnl_queue_net *q = nfnl_queue_pernet(net);
1251         __u32 flags = 0, mask = 0;
1252         int ret = 0;
1253
1254         if (nfqa[NFQA_CFG_CMD]) {
1255                 cmd = nla_data(nfqa[NFQA_CFG_CMD]);
1256
1257                 /* Obsolete commands without queue context */
1258                 switch (cmd->command) {
1259                 case NFQNL_CFG_CMD_PF_BIND: return 0;
1260                 case NFQNL_CFG_CMD_PF_UNBIND: return 0;
1261                 }
1262         }
1263
1264         /* Check if we support these flags in first place, dependencies should
1265          * be there too not to break atomicity.
1266          */
1267         if (nfqa[NFQA_CFG_FLAGS]) {
1268                 if (!nfqa[NFQA_CFG_MASK]) {
1269                         /* A mask is needed to specify which flags are being
1270                          * changed.
1271                          */
1272                         return -EINVAL;
1273                 }
1274
1275                 flags = ntohl(nla_get_be32(nfqa[NFQA_CFG_FLAGS]));
1276                 mask = ntohl(nla_get_be32(nfqa[NFQA_CFG_MASK]));
1277
1278                 if (flags >= NFQA_CFG_F_MAX)
1279                         return -EOPNOTSUPP;
1280
1281 #if !IS_ENABLED(CONFIG_NETWORK_SECMARK)
1282                 if (flags & mask & NFQA_CFG_F_SECCTX)
1283                         return -EOPNOTSUPP;
1284 #endif
1285                 if ((flags & mask & NFQA_CFG_F_CONNTRACK) &&
1286                     !rcu_access_pointer(nfnl_ct_hook)) {
1287 #ifdef CONFIG_MODULES
1288                         nfnl_unlock(NFNL_SUBSYS_QUEUE);
1289                         request_module("ip_conntrack_netlink");
1290                         nfnl_lock(NFNL_SUBSYS_QUEUE);
1291                         if (rcu_access_pointer(nfnl_ct_hook))
1292                                 return -EAGAIN;
1293 #endif
1294                         return -EOPNOTSUPP;
1295                 }
1296         }
1297
1298         rcu_read_lock();
1299         queue = instance_lookup(q, queue_num);
1300         if (queue && queue->peer_portid != NETLINK_CB(skb).portid) {
1301                 ret = -EPERM;
1302                 goto err_out_unlock;
1303         }
1304
1305         if (cmd != NULL) {
1306                 switch (cmd->command) {
1307                 case NFQNL_CFG_CMD_BIND:
1308                         if (queue) {
1309                                 ret = -EBUSY;
1310                                 goto err_out_unlock;
1311                         }
1312                         queue = instance_create(q, queue_num,
1313                                                 NETLINK_CB(skb).portid);
1314                         if (IS_ERR(queue)) {
1315                                 ret = PTR_ERR(queue);
1316                                 goto err_out_unlock;
1317                         }
1318                         break;
1319                 case NFQNL_CFG_CMD_UNBIND:
1320                         if (!queue) {
1321                                 ret = -ENODEV;
1322                                 goto err_out_unlock;
1323                         }
1324                         instance_destroy(q, queue);
1325                         goto err_out_unlock;
1326                 case NFQNL_CFG_CMD_PF_BIND:
1327                 case NFQNL_CFG_CMD_PF_UNBIND:
1328                         break;
1329                 default:
1330                         ret = -ENOTSUPP;
1331                         goto err_out_unlock;
1332                 }
1333         }
1334
1335         if (!queue) {
1336                 ret = -ENODEV;
1337                 goto err_out_unlock;
1338         }
1339
1340         if (nfqa[NFQA_CFG_PARAMS]) {
1341                 struct nfqnl_msg_config_params *params =
1342                         nla_data(nfqa[NFQA_CFG_PARAMS]);
1343
1344                 nfqnl_set_mode(queue, params->copy_mode,
1345                                 ntohl(params->copy_range));
1346         }
1347
1348         if (nfqa[NFQA_CFG_QUEUE_MAXLEN]) {
1349                 __be32 *queue_maxlen = nla_data(nfqa[NFQA_CFG_QUEUE_MAXLEN]);
1350
1351                 spin_lock_bh(&queue->lock);
1352                 queue->queue_maxlen = ntohl(*queue_maxlen);
1353                 spin_unlock_bh(&queue->lock);
1354         }
1355
1356         if (nfqa[NFQA_CFG_FLAGS]) {
1357                 spin_lock_bh(&queue->lock);
1358                 queue->flags &= ~mask;
1359                 queue->flags |= flags & mask;
1360                 spin_unlock_bh(&queue->lock);
1361         }
1362
1363 err_out_unlock:
1364         rcu_read_unlock();
1365         return ret;
1366 }
1367
1368 static const struct nfnl_callback nfqnl_cb[NFQNL_MSG_MAX] = {
1369         [NFQNL_MSG_PACKET]      = { .call_rcu = nfqnl_recv_unsupp,
1370                                     .attr_count = NFQA_MAX, },
1371         [NFQNL_MSG_VERDICT]     = { .call_rcu = nfqnl_recv_verdict,
1372                                     .attr_count = NFQA_MAX,
1373                                     .policy = nfqa_verdict_policy },
1374         [NFQNL_MSG_CONFIG]      = { .call = nfqnl_recv_config,
1375                                     .attr_count = NFQA_CFG_MAX,
1376                                     .policy = nfqa_cfg_policy },
1377         [NFQNL_MSG_VERDICT_BATCH]={ .call_rcu = nfqnl_recv_verdict_batch,
1378                                     .attr_count = NFQA_MAX,
1379                                     .policy = nfqa_verdict_batch_policy },
1380 };
1381
1382 static const struct nfnetlink_subsystem nfqnl_subsys = {
1383         .name           = "nf_queue",
1384         .subsys_id      = NFNL_SUBSYS_QUEUE,
1385         .cb_count       = NFQNL_MSG_MAX,
1386         .cb             = nfqnl_cb,
1387 };
1388
1389 #ifdef CONFIG_PROC_FS
1390 struct iter_state {
1391         struct seq_net_private p;
1392         unsigned int bucket;
1393 };
1394
1395 static struct hlist_node *get_first(struct seq_file *seq)
1396 {
1397         struct iter_state *st = seq->private;
1398         struct net *net;
1399         struct nfnl_queue_net *q;
1400
1401         if (!st)
1402                 return NULL;
1403
1404         net = seq_file_net(seq);
1405         q = nfnl_queue_pernet(net);
1406         for (st->bucket = 0; st->bucket < INSTANCE_BUCKETS; st->bucket++) {
1407                 if (!hlist_empty(&q->instance_table[st->bucket]))
1408                         return q->instance_table[st->bucket].first;
1409         }
1410         return NULL;
1411 }
1412
1413 static struct hlist_node *get_next(struct seq_file *seq, struct hlist_node *h)
1414 {
1415         struct iter_state *st = seq->private;
1416         struct net *net = seq_file_net(seq);
1417
1418         h = h->next;
1419         while (!h) {
1420                 struct nfnl_queue_net *q;
1421
1422                 if (++st->bucket >= INSTANCE_BUCKETS)
1423                         return NULL;
1424
1425                 q = nfnl_queue_pernet(net);
1426                 h = q->instance_table[st->bucket].first;
1427         }
1428         return h;
1429 }
1430
1431 static struct hlist_node *get_idx(struct seq_file *seq, loff_t pos)
1432 {
1433         struct hlist_node *head;
1434         head = get_first(seq);
1435
1436         if (head)
1437                 while (pos && (head = get_next(seq, head)))
1438                         pos--;
1439         return pos ? NULL : head;
1440 }
1441
1442 static void *seq_start(struct seq_file *s, loff_t *pos)
1443         __acquires(nfnl_queue_pernet(seq_file_net(s))->instances_lock)
1444 {
1445         spin_lock(&nfnl_queue_pernet(seq_file_net(s))->instances_lock);
1446         return get_idx(s, *pos);
1447 }
1448
1449 static void *seq_next(struct seq_file *s, void *v, loff_t *pos)
1450 {
1451         (*pos)++;
1452         return get_next(s, v);
1453 }
1454
1455 static void seq_stop(struct seq_file *s, void *v)
1456         __releases(nfnl_queue_pernet(seq_file_net(s))->instances_lock)
1457 {
1458         spin_unlock(&nfnl_queue_pernet(seq_file_net(s))->instances_lock);
1459 }
1460
1461 static int seq_show(struct seq_file *s, void *v)
1462 {
1463         const struct nfqnl_instance *inst = v;
1464
1465         seq_printf(s, "%5u %6u %5u %1u %5u %5u %5u %8u %2d\n",
1466                    inst->queue_num,
1467                    inst->peer_portid, inst->queue_total,
1468                    inst->copy_mode, inst->copy_range,
1469                    inst->queue_dropped, inst->queue_user_dropped,
1470                    inst->id_sequence, 1);
1471         return 0;
1472 }
1473
1474 static const struct seq_operations nfqnl_seq_ops = {
1475         .start  = seq_start,
1476         .next   = seq_next,
1477         .stop   = seq_stop,
1478         .show   = seq_show,
1479 };
1480
1481 static int nfqnl_open(struct inode *inode, struct file *file)
1482 {
1483         return seq_open_net(inode, file, &nfqnl_seq_ops,
1484                         sizeof(struct iter_state));
1485 }
1486
1487 static const struct file_operations nfqnl_file_ops = {
1488         .owner   = THIS_MODULE,
1489         .open    = nfqnl_open,
1490         .read    = seq_read,
1491         .llseek  = seq_lseek,
1492         .release = seq_release_net,
1493 };
1494
1495 #endif /* PROC_FS */
1496
1497 static int __net_init nfnl_queue_net_init(struct net *net)
1498 {
1499         unsigned int i;
1500         struct nfnl_queue_net *q = nfnl_queue_pernet(net);
1501
1502         for (i = 0; i < INSTANCE_BUCKETS; i++)
1503                 INIT_HLIST_HEAD(&q->instance_table[i]);
1504
1505         spin_lock_init(&q->instances_lock);
1506
1507 #ifdef CONFIG_PROC_FS
1508         if (!proc_create("nfnetlink_queue", 0440,
1509                          net->nf.proc_netfilter, &nfqnl_file_ops))
1510                 return -ENOMEM;
1511 #endif
1512         nf_register_queue_handler(net, &nfqh);
1513         return 0;
1514 }
1515
1516 static void __net_exit nfnl_queue_net_exit(struct net *net)
1517 {
1518         nf_unregister_queue_handler(net);
1519 #ifdef CONFIG_PROC_FS
1520         remove_proc_entry("nfnetlink_queue", net->nf.proc_netfilter);
1521 #endif
1522 }
1523
1524 static void nfnl_queue_net_exit_batch(struct list_head *net_exit_list)
1525 {
1526         synchronize_rcu();
1527 }
1528
1529 static struct pernet_operations nfnl_queue_net_ops = {
1530         .init           = nfnl_queue_net_init,
1531         .exit           = nfnl_queue_net_exit,
1532         .exit_batch     = nfnl_queue_net_exit_batch,
1533         .id             = &nfnl_queue_net_id,
1534         .size           = sizeof(struct nfnl_queue_net),
1535 };
1536
1537 static int __init nfnetlink_queue_init(void)
1538 {
1539         int status;
1540
1541         status = register_pernet_subsys(&nfnl_queue_net_ops);
1542         if (status < 0) {
1543                 pr_err("nf_queue: failed to register pernet ops\n");
1544                 goto out;
1545         }
1546
1547         netlink_register_notifier(&nfqnl_rtnl_notifier);
1548         status = nfnetlink_subsys_register(&nfqnl_subsys);
1549         if (status < 0) {
1550                 pr_err("nf_queue: failed to create netlink socket\n");
1551                 goto cleanup_netlink_notifier;
1552         }
1553
1554         status = register_netdevice_notifier(&nfqnl_dev_notifier);
1555         if (status < 0) {
1556                 pr_err("nf_queue: failed to register netdevice notifier\n");
1557                 goto cleanup_netlink_subsys;
1558         }
1559
1560         return status;
1561
1562 cleanup_netlink_subsys:
1563         nfnetlink_subsys_unregister(&nfqnl_subsys);
1564 cleanup_netlink_notifier:
1565         netlink_unregister_notifier(&nfqnl_rtnl_notifier);
1566         unregister_pernet_subsys(&nfnl_queue_net_ops);
1567 out:
1568         return status;
1569 }
1570
1571 static void __exit nfnetlink_queue_fini(void)
1572 {
1573         unregister_netdevice_notifier(&nfqnl_dev_notifier);
1574         nfnetlink_subsys_unregister(&nfqnl_subsys);
1575         netlink_unregister_notifier(&nfqnl_rtnl_notifier);
1576         unregister_pernet_subsys(&nfnl_queue_net_ops);
1577
1578         rcu_barrier(); /* Wait for completion of call_rcu()'s */
1579 }
1580
1581 MODULE_DESCRIPTION("netfilter packet queue handler");
1582 MODULE_AUTHOR("Harald Welte <laforge@netfilter.org>");
1583 MODULE_LICENSE("GPL");
1584 MODULE_ALIAS_NFNL_SUBSYS(NFNL_SUBSYS_QUEUE);
1585
1586 module_init(nfnetlink_queue_init);
1587 module_exit(nfnetlink_queue_fini);