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             skb_mac_header_was_set(entskb) &&
 547             skb_mac_header_len(entskb) != 0) {
 548                 struct nfqnl_msg_packet_hw phw;
 549                 int len;
 550
 551                 memset(&phw, 0, sizeof(phw));
 552                 len = dev_parse_header(entskb, phw.hw_addr);
 553                 if (len) {
 554                         phw.hw_addrlen = htons(len);
 555                         if (nla_put(skb, NFQA_HWADDR, sizeof(phw), &phw))
 556                                 goto nla_put_failure;
 557                 }
 558         }
 559
 560         if (nfqnl_put_bridge(entry, skb) < 0)
 561                 goto nla_put_failure;
 562
 563         if (entskb->tstamp) {
 564                 struct nfqnl_msg_packet_timestamp ts;
 565                 struct timespec64 kts = ktime_to_timespec64(entskb->tstamp);
 566
 567                 ts.sec = cpu_to_be64(kts.tv_sec);
 568                 ts.usec = cpu_to_be64(kts.tv_nsec / NSEC_PER_USEC);
 569
 570                 if (nla_put(skb, NFQA_TIMESTAMP, sizeof(ts), &ts))
 571                         goto nla_put_failure;
 572         }
 573
 574         if ((queue->flags & NFQA_CFG_F_UID_GID) && entskb->sk &&
 575             nfqnl_put_sk_uidgid(skb, entskb->sk) < 0)
 576                 goto nla_put_failure;
 577
 578         if (seclen && nla_put(skb, NFQA_SECCTX, seclen, secdata))
 579                 goto nla_put_failure;
 580
 581         if (ct && nfnl_ct->build(skb, ct, ctinfo, NFQA_CT, NFQA_CT_INFO) < 0)
 582                 goto nla_put_failure;
 583
 584         if (cap_len > data_len &&
 585             nla_put_be32(skb, NFQA_CAP_LEN, htonl(cap_len)))
 586                 goto nla_put_failure;
 587
 588         if (nfqnl_put_packet_info(skb, entskb, csum_verify))
 589                 goto nla_put_failure;
 590
 591         if (data_len) {
 592                 struct nlattr *nla;
 593
 594                 if (skb_tailroom(skb) < sizeof(*nla) + hlen)
 595                         goto nla_put_failure;
 596
 597                 nla = skb_put(skb, sizeof(*nla));
 598                 nla->nla_type = NFQA_PAYLOAD;
 599                 nla->nla_len = nla_attr_size(data_len);
 600
 601                 if (skb_zerocopy(skb, entskb, data_len, hlen))
 602                         goto nla_put_failure;
 603         }
 604
 605         nlh->nlmsg_len = skb->len;
 606         if (seclen)
 607                 security_release_secctx(secdata, seclen);
 608         return skb;
 609
 610 nla_put_failure:
 611         skb_tx_error(entskb);
 612         kfree_skb(skb);
 613         net_err_ratelimited("nf_queue: error creating packet message\n");
 614 nlmsg_failure:
 615         if (seclen)
 616                 security_release_secctx(secdata, seclen);
 617         return NULL;
 618 }
 619
 620 static bool nf_ct_drop_unconfirmed(const struct nf_queue_entry *entry)
 621 {
 622 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
 623         static const unsigned long flags = IPS_CONFIRMED | IPS_DYING;
 624         const struct nf_conn *ct = (void *)skb_nfct(entry->skb);
 625
 626         if (ct && ((ct->status & flags) == IPS_DYING))
 627                 return true;
 628 #endif
 629         return false;
 630 }
 631
 632 static int
 633 __nfqnl_enqueue_packet(struct net *net, struct nfqnl_instance *queue,
 634                         struct nf_queue_entry *entry)
 635 {
 636         struct sk_buff *nskb;
 637         int err = -ENOBUFS;
 638         __be32 *packet_id_ptr;
 639         int failopen = 0;
 640
 641         nskb = nfqnl_build_packet_message(net, queue, entry, &packet_id_ptr);
 642         if (nskb == NULL) {
 643                 err = -ENOMEM;
 644                 goto err_out;
 645         }
 646         spin_lock_bh(&queue->lock);
 647
 648         if (nf_ct_drop_unconfirmed(entry))
 649                 goto err_out_free_nskb;
 650
 651         if (queue->queue_total >= queue->queue_maxlen) {
 652                 if (queue->flags & NFQA_CFG_F_FAIL_OPEN) {
 653                         failopen = 1;
 654                         err = 0;
 655                 } else {
 656                         queue->queue_dropped++;
 657                         net_warn_ratelimited("nf_queue: full at %d entries, dropping packets(s)\n",
 658                                              queue->queue_total);
 659                 }
 660                 goto err_out_free_nskb;
 661         }
 662         entry->id = ++queue->id_sequence;
 663         *packet_id_ptr = htonl(entry->id);
 664
 665         /* nfnetlink_unicast will either free the nskb or add it to a socket */
 666         err = nfnetlink_unicast(nskb, net, queue->peer_portid, MSG_DONTWAIT);
 667         if (err < 0) {
 668                 if (queue->flags & NFQA_CFG_F_FAIL_OPEN) {
 669                         failopen = 1;
 670                         err = 0;
 671                 } else {
 672                         queue->queue_user_dropped++;
 673                 }
 674                 goto err_out_unlock;
 675         }
 676
 677         __enqueue_entry(queue, entry);
 678
 679         spin_unlock_bh(&queue->lock);
 680         return 0;
 681
 682 err_out_free_nskb:
 683         kfree_skb(nskb);
 684 err_out_unlock:
 685         spin_unlock_bh(&queue->lock);
 686         if (failopen)
 687                 nf_reinject(entry, NF_ACCEPT);
 688 err_out:
 689         return err;
 690 }
 691
 692 static struct nf_queue_entry *
 693 nf_queue_entry_dup(struct nf_queue_entry *e)
 694 {
 695         struct nf_queue_entry *entry = kmemdup(e, e->size, GFP_ATOMIC);
 696
 697         if (!entry)
 698                 return NULL;
 699
 700         if (nf_queue_entry_get_refs(entry))
 701                 return entry;
 702
 703         kfree(entry);
 704         return NULL;
 705 }
 706
 707 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
 708 /* When called from bridge netfilter, skb->data must point to MAC header
 709  * before calling skb_gso_segment(). Else, original MAC header is lost
 710  * and segmented skbs will be sent to wrong destination.
 711  */
 712 static void nf_bridge_adjust_skb_data(struct sk_buff *skb)
 713 {
 714         if (skb->nf_bridge)
 715                 __skb_push(skb, skb->network_header - skb->mac_header);
 716 }
 717
 718 static void nf_bridge_adjust_segmented_data(struct sk_buff *skb)
 719 {
 720         if (skb->nf_bridge)
 721                 __skb_pull(skb, skb->network_header - skb->mac_header);
 722 }
 723 #else
 724 #define nf_bridge_adjust_skb_data(s) do {} while (0)
 725 #define nf_bridge_adjust_segmented_data(s) do {} while (0)
 726 #endif
 727
 728 static void free_entry(struct nf_queue_entry *entry)
 729 {
 730         nf_queue_entry_release_refs(entry);
 731         kfree(entry);
 732 }
 733
 734 static int
 735 __nfqnl_enqueue_packet_gso(struct net *net, struct nfqnl_instance *queue,
 736                            struct sk_buff *skb, struct nf_queue_entry *entry)
 737 {
 738         int ret = -ENOMEM;
 739         struct nf_queue_entry *entry_seg;
 740
 741         nf_bridge_adjust_segmented_data(skb);
 742
 743         if (skb->next == NULL) { /* last packet, no need to copy entry */
 744                 struct sk_buff *gso_skb = entry->skb;
 745                 entry->skb = skb;
 746                 ret = __nfqnl_enqueue_packet(net, queue, entry);
 747                 if (ret)
 748                         entry->skb = gso_skb;
 749                 return ret;
 750         }
 751
 752         skb->next = NULL;
 753
 754         entry_seg = nf_queue_entry_dup(entry);
 755         if (entry_seg) {
 756                 entry_seg->skb = skb;
 757                 ret = __nfqnl_enqueue_packet(net, queue, entry_seg);
 758                 if (ret)
 759                         free_entry(entry_seg);
 760         }
 761         return ret;
 762 }
 763
 764 static int
 765 nfqnl_enqueue_packet(struct nf_queue_entry *entry, unsigned int queuenum)
 766 {
 767         unsigned int queued;
 768         struct nfqnl_instance *queue;
 769         struct sk_buff *skb, *segs;
 770         int err = -ENOBUFS;
 771         struct net *net = entry->state.net;
 772         struct nfnl_queue_net *q = nfnl_queue_pernet(net);
 773
 774         /* rcu_read_lock()ed by nf_hook_thresh */
 775         queue = instance_lookup(q, queuenum);
 776         if (!queue)
 777                 return -ESRCH;
 778
 779         if (queue->copy_mode == NFQNL_COPY_NONE)
 780                 return -EINVAL;
 781
 782         skb = entry->skb;
 783
 784         switch (entry->state.pf) {
 785         case NFPROTO_IPV4:
 786                 skb->protocol = htons(ETH_P_IP);
 787                 break;
 788         case NFPROTO_IPV6:
 789                 skb->protocol = htons(ETH_P_IPV6);
 790                 break;
 791         }
 792
 793         if ((queue->flags & NFQA_CFG_F_GSO) || !skb_is_gso(skb))
 794                 return __nfqnl_enqueue_packet(net, queue, entry);
 795
 796         nf_bridge_adjust_skb_data(skb);
 797         segs = skb_gso_segment(skb, 0);
 798         /* Does not use PTR_ERR to limit the number of error codes that can be
 799          * returned by nf_queue.  For instance, callers rely on -ESRCH to
 800          * mean 'ignore this hook'.
 801          */
 802         if (IS_ERR_OR_NULL(segs))
 803                 goto out_err;
 804         queued = 0;
 805         err = 0;
 806         do {
 807                 struct sk_buff *nskb = segs->next;
 808                 if (err == 0)
 809                         err = __nfqnl_enqueue_packet_gso(net, queue,
 810                                                         segs, entry);
 811                 if (err == 0)
 812                         queued++;
 813                 else
 814                         kfree_skb(segs);
 815                 segs = nskb;
 816         } while (segs);
 817
 818         if (queued) {
 819                 if (err) /* some segments are already queued */
 820                         free_entry(entry);
 821                 kfree_skb(skb);
 822                 return 0;
 823         }
 824  out_err:
 825         nf_bridge_adjust_segmented_data(skb);
 826         return err;
 827 }
 828
 829 static int
 830 nfqnl_mangle(void *data, unsigned int data_len, struct nf_queue_entry *e, int diff)
 831 {
 832         struct sk_buff *nskb;
 833
 834         if (diff < 0) {
 835                 unsigned int min_len = skb_transport_offset(e->skb);
 836
 837                 if (data_len < min_len)
 838                         return -EINVAL;
 839
 840                 if (pskb_trim(e->skb, data_len))
 841                         return -ENOMEM;
 842         } else if (diff > 0) {
 843                 if (data_len > 0xFFFF)
 844                         return -EINVAL;
 845                 if (diff > skb_tailroom(e->skb)) {
 846                         nskb = skb_copy_expand(e->skb, skb_headroom(e->skb),
 847                                                diff, GFP_ATOMIC);
 848                         if (!nskb) {
 849                                 printk(KERN_WARNING "nf_queue: OOM "
 850                                       "in mangle, dropping packet\n");
 851                                 return -ENOMEM;
 852                         }
 853                         kfree_skb(e->skb);
 854                         e->skb = nskb;
 855                 }
 856                 skb_put(e->skb, diff);
 857         }
 858         if (!skb_make_writable(e->skb, data_len))
 859                 return -ENOMEM;
 860         skb_copy_to_linear_data(e->skb, data, data_len);
 861         e->skb->ip_summed = CHECKSUM_NONE;
 862         return 0;
 863 }
 864
 865 static int
 866 nfqnl_set_mode(struct nfqnl_instance *queue,
 867                unsigned char mode, unsigned int range)
 868 {
 869         int status = 0;
 870
 871         spin_lock_bh(&queue->lock);
 872         switch (mode) {
 873         case NFQNL_COPY_NONE:
 874         case NFQNL_COPY_META:
 875                 queue->copy_mode = mode;
 876                 queue->copy_range = 0;
 877                 break;
 878
 879         case NFQNL_COPY_PACKET:
 880                 queue->copy_mode = mode;
 881                 if (range == 0 || range > NFQNL_MAX_COPY_RANGE)
 882                         queue->copy_range = NFQNL_MAX_COPY_RANGE;
 883                 else
 884                         queue->copy_range = range;
 885                 break;
 886
 887         default:
 888                 status = -EINVAL;
 889
 890         }
 891         spin_unlock_bh(&queue->lock);
 892
 893         return status;
 894 }
 895
 896 static int
 897 dev_cmp(struct nf_queue_entry *entry, unsigned long ifindex)
 898 {
 899         if (entry->state.in)
 900                 if (entry->state.in->ifindex == ifindex)
 901                         return 1;
 902         if (entry->state.out)
 903                 if (entry->state.out->ifindex == ifindex)
 904                         return 1;
 905 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
 906         if (entry->skb->nf_bridge) {
 907                 int physinif, physoutif;
 908
 909                 physinif = nf_bridge_get_physinif(entry->skb);
 910                 physoutif = nf_bridge_get_physoutif(entry->skb);
 911
 912                 if (physinif == ifindex || physoutif == ifindex)
 913                         return 1;
 914         }
 915 #endif
 916         return 0;
 917 }
 918
 919 /* drop all packets with either indev or outdev == ifindex from all queue
 920  * instances */
 921 static void
 922 nfqnl_dev_drop(struct net *net, int ifindex)
 923 {
 924         int i;
 925         struct nfnl_queue_net *q = nfnl_queue_pernet(net);
 926
 927         rcu_read_lock();
 928
 929         for (i = 0; i < INSTANCE_BUCKETS; i++) {
 930                 struct nfqnl_instance *inst;
 931                 struct hlist_head *head = &q->instance_table[i];
 932
 933                 hlist_for_each_entry_rcu(inst, head, hlist)
 934                         nfqnl_flush(inst, dev_cmp, ifindex);
 935         }
 936
 937         rcu_read_unlock();
 938 }
 939
 940 static int
 941 nfqnl_rcv_dev_event(struct notifier_block *this,
 942                     unsigned long event, void *ptr)
 943 {
 944         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
 945
 946         /* Drop any packets associated with the downed device */
 947         if (event == NETDEV_DOWN)
 948                 nfqnl_dev_drop(dev_net(dev), dev->ifindex);
 949         return NOTIFY_DONE;
 950 }
 951
 952 static struct notifier_block nfqnl_dev_notifier = {
 953         .notifier_call  = nfqnl_rcv_dev_event,
 954 };
 955
 956 static unsigned int nfqnl_nf_hook_drop(struct net *net)
 957 {
 958         struct nfnl_queue_net *q = nfnl_queue_pernet(net);
 959         unsigned int instances = 0;
 960         int i;
 961
 962         for (i = 0; i < INSTANCE_BUCKETS; i++) {
 963                 struct nfqnl_instance *inst;
 964                 struct hlist_head *head = &q->instance_table[i];
 965
 966                 hlist_for_each_entry_rcu(inst, head, hlist) {
 967                         nfqnl_flush(inst, NULL, 0);
 968                         instances++;
 969                 }
 970         }
 971
 972         return instances;
 973 }
 974
 975 static int
 976 nfqnl_rcv_nl_event(struct notifier_block *this,
 977                    unsigned long event, void *ptr)
 978 {
 979         struct netlink_notify *n = ptr;
 980         struct nfnl_queue_net *q = nfnl_queue_pernet(n->net);
 981
 982         if (event == NETLINK_URELEASE && n->protocol == NETLINK_NETFILTER) {
 983                 int i;
 984
 985                 /* destroy all instances for this portid */
 986                 spin_lock(&q->instances_lock);
 987                 for (i = 0; i < INSTANCE_BUCKETS; i++) {
 988                         struct hlist_node *t2;
 989                         struct nfqnl_instance *inst;
 990                         struct hlist_head *head = &q->instance_table[i];
 991
 992                         hlist_for_each_entry_safe(inst, t2, head, hlist) {
 993                                 if (n->portid == inst->peer_portid)
 994                                         __instance_destroy(inst);
 995                         }
 996                 }
 997                 spin_unlock(&q->instances_lock);
 998         }
 999         return NOTIFY_DONE;
1000 }
1001
1002 static struct notifier_block nfqnl_rtnl_notifier = {
1003         .notifier_call  = nfqnl_rcv_nl_event,
1004 };
1005
1006 static const struct nla_policy nfqa_vlan_policy[NFQA_VLAN_MAX + 1] = {
1007         [NFQA_VLAN_TCI]         = { .type = NLA_U16},
1008         [NFQA_VLAN_PROTO]       = { .type = NLA_U16},
1009 };
1010
1011 static const struct nla_policy nfqa_verdict_policy[NFQA_MAX+1] = {
1012         [NFQA_VERDICT_HDR]      = { .len = sizeof(struct nfqnl_msg_verdict_hdr) },
1013         [NFQA_MARK]             = { .type = NLA_U32 },
1014         [NFQA_PAYLOAD]          = { .type = NLA_UNSPEC },
1015         [NFQA_CT]               = { .type = NLA_UNSPEC },
1016         [NFQA_EXP]              = { .type = NLA_UNSPEC },
1017         [NFQA_VLAN]             = { .type = NLA_NESTED },
1018 };
1019
1020 static const struct nla_policy nfqa_verdict_batch_policy[NFQA_MAX+1] = {
1021         [NFQA_VERDICT_HDR]      = { .len = sizeof(struct nfqnl_msg_verdict_hdr) },
1022         [NFQA_MARK]             = { .type = NLA_U32 },
1023 };
1024
1025 static struct nfqnl_instance *
1026 verdict_instance_lookup(struct nfnl_queue_net *q, u16 queue_num, u32 nlportid)
1027 {
1028         struct nfqnl_instance *queue;
1029
1030         queue = instance_lookup(q, queue_num);
1031         if (!queue)
1032                 return ERR_PTR(-ENODEV);
1033
1034         if (queue->peer_portid != nlportid)
1035                 return ERR_PTR(-EPERM);
1036
1037         return queue;
1038 }
1039
1040 static struct nfqnl_msg_verdict_hdr*
1041 verdicthdr_get(const struct nlattr * const nfqa[])
1042 {
1043         struct nfqnl_msg_verdict_hdr *vhdr;
1044         unsigned int verdict;
1045
1046         if (!nfqa[NFQA_VERDICT_HDR])
1047                 return NULL;
1048
1049         vhdr = nla_data(nfqa[NFQA_VERDICT_HDR]);
1050         verdict = ntohl(vhdr->verdict) & NF_VERDICT_MASK;
1051         if (verdict > NF_MAX_VERDICT || verdict == NF_STOLEN)
1052                 return NULL;
1053         return vhdr;
1054 }
1055
1056 static int nfq_id_after(unsigned int id, unsigned int max)
1057 {
1058         return (int)(id - max) > 0;
1059 }
1060
1061 static int nfqnl_recv_verdict_batch(struct net *net, struct sock *ctnl,
1062                                     struct sk_buff *skb,
1063                                     const struct nlmsghdr *nlh,
1064                                     const struct nlattr * const nfqa[],
1065                                     struct netlink_ext_ack *extack)
1066 {
1067         struct nfgenmsg *nfmsg = nlmsg_data(nlh);
1068         struct nf_queue_entry *entry, *tmp;
1069         unsigned int verdict, maxid;
1070         struct nfqnl_msg_verdict_hdr *vhdr;
1071         struct nfqnl_instance *queue;
1072         LIST_HEAD(batch_list);
1073         u16 queue_num = ntohs(nfmsg->res_id);
1074         struct nfnl_queue_net *q = nfnl_queue_pernet(net);
1075
1076         queue = verdict_instance_lookup(q, queue_num,
1077                                         NETLINK_CB(skb).portid);
1078         if (IS_ERR(queue))
1079                 return PTR_ERR(queue);
1080
1081         vhdr = verdicthdr_get(nfqa);
1082         if (!vhdr)
1083                 return -EINVAL;
1084
1085         verdict = ntohl(vhdr->verdict);
1086         maxid = ntohl(vhdr->id);
1087
1088         spin_lock_bh(&queue->lock);
1089
1090         list_for_each_entry_safe(entry, tmp, &queue->queue_list, list) {
1091                 if (nfq_id_after(entry->id, maxid))
1092                         break;
1093                 __dequeue_entry(queue, entry);
1094                 list_add_tail(&entry->list, &batch_list);
1095         }
1096
1097         spin_unlock_bh(&queue->lock);
1098
1099         if (list_empty(&batch_list))
1100                 return -ENOENT;
1101
1102         list_for_each_entry_safe(entry, tmp, &batch_list, list) {
1103                 if (nfqa[NFQA_MARK])
1104                         entry->skb->mark = ntohl(nla_get_be32(nfqa[NFQA_MARK]));
1105                 nf_reinject(entry, verdict);
1106         }
1107         return 0;
1108 }
1109
1110 static struct nf_conn *nfqnl_ct_parse(struct nfnl_ct_hook *nfnl_ct,
1111                                       const struct nlmsghdr *nlh,
1112                                       const struct nlattr * const nfqa[],
1113                                       struct nf_queue_entry *entry,
1114                                       enum ip_conntrack_info *ctinfo)
1115 {
1116         struct nf_conn *ct;
1117
1118         ct = nfnl_ct->get_ct(entry->skb, ctinfo);
1119         if (ct == NULL)
1120                 return NULL;
1121
1122         if (nfnl_ct->parse(nfqa[NFQA_CT], ct) < 0)
1123                 return NULL;
1124
1125         if (nfqa[NFQA_EXP])
1126                 nfnl_ct->attach_expect(nfqa[NFQA_EXP], ct,
1127                                       NETLINK_CB(entry->skb).portid,
1128                                       nlmsg_report(nlh));
1129         return ct;
1130 }
1131
1132 static int nfqa_parse_bridge(struct nf_queue_entry *entry,
1133                              const struct nlattr * const nfqa[])
1134 {
1135         if (nfqa[NFQA_VLAN]) {
1136                 struct nlattr *tb[NFQA_VLAN_MAX + 1];
1137                 int err;
1138
1139                 err = nla_parse_nested(tb, NFQA_VLAN_MAX, nfqa[NFQA_VLAN],
1140                                        nfqa_vlan_policy, NULL);
1141                 if (err < 0)
1142                         return err;
1143
1144                 if (!tb[NFQA_VLAN_TCI] || !tb[NFQA_VLAN_PROTO])
1145                         return -EINVAL;
1146
1147                 entry->skb->vlan_tci = ntohs(nla_get_be16(tb[NFQA_VLAN_TCI]));
1148                 entry->skb->vlan_proto = nla_get_be16(tb[NFQA_VLAN_PROTO]);
1149         }
1150
1151         if (nfqa[NFQA_L2HDR]) {
1152                 int mac_header_len = entry->skb->network_header -
1153                         entry->skb->mac_header;
1154
1155                 if (mac_header_len != nla_len(nfqa[NFQA_L2HDR]))
1156                         return -EINVAL;
1157                 else if (mac_header_len > 0)
1158                         memcpy(skb_mac_header(entry->skb),
1159                                nla_data(nfqa[NFQA_L2HDR]),
1160                                mac_header_len);
1161         }
1162
1163         return 0;
1164 }
1165
1166 static int nfqnl_recv_verdict(struct net *net, struct sock *ctnl,
1167                               struct sk_buff *skb,
1168                               const struct nlmsghdr *nlh,
1169                               const struct nlattr * const nfqa[],
1170                               struct netlink_ext_ack *extack)
1171 {
1172         struct nfgenmsg *nfmsg = nlmsg_data(nlh);
1173         u_int16_t queue_num = ntohs(nfmsg->res_id);
1174         struct nfqnl_msg_verdict_hdr *vhdr;
1175         struct nfqnl_instance *queue;
1176         unsigned int verdict;
1177         struct nf_queue_entry *entry;
1178         enum ip_conntrack_info uninitialized_var(ctinfo);
1179         struct nfnl_ct_hook *nfnl_ct;
1180         struct nf_conn *ct = NULL;
1181         struct nfnl_queue_net *q = nfnl_queue_pernet(net);
1182         int err;
1183
1184         queue = verdict_instance_lookup(q, queue_num,
1185                                         NETLINK_CB(skb).portid);
1186         if (IS_ERR(queue))
1187                 return PTR_ERR(queue);
1188
1189         vhdr = verdicthdr_get(nfqa);
1190         if (!vhdr)
1191                 return -EINVAL;
1192
1193         verdict = ntohl(vhdr->verdict);
1194
1195         entry = find_dequeue_entry(queue, ntohl(vhdr->id));
1196         if (entry == NULL)
1197                 return -ENOENT;
1198
1199         /* rcu lock already held from nfnl->call_rcu. */
1200         nfnl_ct = rcu_dereference(nfnl_ct_hook);
1201
1202         if (nfqa[NFQA_CT]) {
1203                 if (nfnl_ct != NULL)
1204                         ct = nfqnl_ct_parse(nfnl_ct, nlh, nfqa, entry, &ctinfo);
1205         }
1206
1207         if (entry->state.pf == PF_BRIDGE) {
1208                 err = nfqa_parse_bridge(entry, nfqa);
1209                 if (err < 0)
1210                         return err;
1211         }
1212
1213         if (nfqa[NFQA_PAYLOAD]) {
1214                 u16 payload_len = nla_len(nfqa[NFQA_PAYLOAD]);
1215                 int diff = payload_len - entry->skb->len;
1216
1217                 if (nfqnl_mangle(nla_data(nfqa[NFQA_PAYLOAD]),
1218                                  payload_len, entry, diff) < 0)
1219                         verdict = NF_DROP;
1220
1221                 if (ct && diff)
1222                         nfnl_ct->seq_adjust(entry->skb, ct, ctinfo, diff);
1223         }
1224
1225         if (nfqa[NFQA_MARK])
1226                 entry->skb->mark = ntohl(nla_get_be32(nfqa[NFQA_MARK]));
1227
1228         nf_reinject(entry, verdict);
1229         return 0;
1230 }
1231
1232 static int nfqnl_recv_unsupp(struct net *net, struct sock *ctnl,
1233                              struct sk_buff *skb, const struct nlmsghdr *nlh,
1234                              const struct nlattr * const nfqa[],
1235                              struct netlink_ext_ack *extack)
1236 {
1237         return -ENOTSUPP;
1238 }
1239
1240 static const struct nla_policy nfqa_cfg_policy[NFQA_CFG_MAX+1] = {
1241         [NFQA_CFG_CMD]          = { .len = sizeof(struct nfqnl_msg_config_cmd) },
1242         [NFQA_CFG_PARAMS]       = { .len = sizeof(struct nfqnl_msg_config_params) },
1243         [NFQA_CFG_QUEUE_MAXLEN] = { .type = NLA_U32 },
1244         [NFQA_CFG_MASK]         = { .type = NLA_U32 },
1245         [NFQA_CFG_FLAGS]        = { .type = NLA_U32 },
1246 };
1247
1248 static const struct nf_queue_handler nfqh = {
1249         .outfn          = nfqnl_enqueue_packet,
1250         .nf_hook_drop   = nfqnl_nf_hook_drop,
1251 };
1252
1253 static int nfqnl_recv_config(struct net *net, struct sock *ctnl,
1254                              struct sk_buff *skb, const struct nlmsghdr *nlh,
1255                              const struct nlattr * const nfqa[],
1256                              struct netlink_ext_ack *extack)
1257 {
1258         struct nfgenmsg *nfmsg = nlmsg_data(nlh);
1259         u_int16_t queue_num = ntohs(nfmsg->res_id);
1260         struct nfqnl_instance *queue;
1261         struct nfqnl_msg_config_cmd *cmd = NULL;
1262         struct nfnl_queue_net *q = nfnl_queue_pernet(net);
1263         __u32 flags = 0, mask = 0;
1264         int ret = 0;
1265
1266         if (nfqa[NFQA_CFG_CMD]) {
1267                 cmd = nla_data(nfqa[NFQA_CFG_CMD]);
1268
1269                 /* Obsolete commands without queue context */
1270                 switch (cmd->command) {
1271                 case NFQNL_CFG_CMD_PF_BIND: return 0;
1272                 case NFQNL_CFG_CMD_PF_UNBIND: return 0;
1273                 }
1274         }
1275
1276         /* Check if we support these flags in first place, dependencies should
1277          * be there too not to break atomicity.
1278          */
1279         if (nfqa[NFQA_CFG_FLAGS]) {
1280                 if (!nfqa[NFQA_CFG_MASK]) {
1281                         /* A mask is needed to specify which flags are being
1282                          * changed.
1283                          */
1284                         return -EINVAL;
1285                 }
1286
1287                 flags = ntohl(nla_get_be32(nfqa[NFQA_CFG_FLAGS]));
1288                 mask = ntohl(nla_get_be32(nfqa[NFQA_CFG_MASK]));
1289
1290                 if (flags >= NFQA_CFG_F_MAX)
1291                         return -EOPNOTSUPP;
1292
1293 #if !IS_ENABLED(CONFIG_NETWORK_SECMARK)
1294                 if (flags & mask & NFQA_CFG_F_SECCTX)
1295                         return -EOPNOTSUPP;
1296 #endif
1297                 if ((flags & mask & NFQA_CFG_F_CONNTRACK) &&
1298                     !rcu_access_pointer(nfnl_ct_hook)) {
1299 #ifdef CONFIG_MODULES
1300                         nfnl_unlock(NFNL_SUBSYS_QUEUE);
1301                         request_module("ip_conntrack_netlink");
1302                         nfnl_lock(NFNL_SUBSYS_QUEUE);
1303                         if (rcu_access_pointer(nfnl_ct_hook))
1304                                 return -EAGAIN;
1305 #endif
1306                         return -EOPNOTSUPP;
1307                 }
1308         }
1309
1310         rcu_read_lock();
1311         queue = instance_lookup(q, queue_num);
1312         if (queue && queue->peer_portid != NETLINK_CB(skb).portid) {
1313                 ret = -EPERM;
1314                 goto err_out_unlock;
1315         }
1316
1317         if (cmd != NULL) {
1318                 switch (cmd->command) {
1319                 case NFQNL_CFG_CMD_BIND:
1320                         if (queue) {
1321                                 ret = -EBUSY;
1322                                 goto err_out_unlock;
1323                         }
1324                         queue = instance_create(q, queue_num,
1325                                                 NETLINK_CB(skb).portid);
1326                         if (IS_ERR(queue)) {
1327                                 ret = PTR_ERR(queue);
1328                                 goto err_out_unlock;
1329                         }
1330                         break;
1331                 case NFQNL_CFG_CMD_UNBIND:
1332                         if (!queue) {
1333                                 ret = -ENODEV;
1334                                 goto err_out_unlock;
1335                         }
1336                         instance_destroy(q, queue);
1337                         goto err_out_unlock;
1338                 case NFQNL_CFG_CMD_PF_BIND:
1339                 case NFQNL_CFG_CMD_PF_UNBIND:
1340                         break;
1341                 default:
1342                         ret = -ENOTSUPP;
1343                         goto err_out_unlock;
1344                 }
1345         }
1346
1347         if (!queue) {
1348                 ret = -ENODEV;
1349                 goto err_out_unlock;
1350         }
1351
1352         if (nfqa[NFQA_CFG_PARAMS]) {
1353                 struct nfqnl_msg_config_params *params =
1354                         nla_data(nfqa[NFQA_CFG_PARAMS]);
1355
1356                 nfqnl_set_mode(queue, params->copy_mode,
1357                                 ntohl(params->copy_range));
1358         }
1359
1360         if (nfqa[NFQA_CFG_QUEUE_MAXLEN]) {
1361                 __be32 *queue_maxlen = nla_data(nfqa[NFQA_CFG_QUEUE_MAXLEN]);
1362
1363                 spin_lock_bh(&queue->lock);
1364                 queue->queue_maxlen = ntohl(*queue_maxlen);
1365                 spin_unlock_bh(&queue->lock);
1366         }
1367
1368         if (nfqa[NFQA_CFG_FLAGS]) {
1369                 spin_lock_bh(&queue->lock);
1370                 queue->flags &= ~mask;
1371                 queue->flags |= flags & mask;
1372                 spin_unlock_bh(&queue->lock);
1373         }
1374
1375 err_out_unlock:
1376         rcu_read_unlock();
1377         return ret;
1378 }
1379
1380 static const struct nfnl_callback nfqnl_cb[NFQNL_MSG_MAX] = {
1381         [NFQNL_MSG_PACKET]      = { .call_rcu = nfqnl_recv_unsupp,
1382                                     .attr_count = NFQA_MAX, },
1383         [NFQNL_MSG_VERDICT]     = { .call_rcu = nfqnl_recv_verdict,
1384                                     .attr_count = NFQA_MAX,
1385                                     .policy = nfqa_verdict_policy },
1386         [NFQNL_MSG_CONFIG]      = { .call = nfqnl_recv_config,
1387                                     .attr_count = NFQA_CFG_MAX,
1388                                     .policy = nfqa_cfg_policy },
1389         [NFQNL_MSG_VERDICT_BATCH]={ .call_rcu = nfqnl_recv_verdict_batch,
1390                                     .attr_count = NFQA_MAX,
1391                                     .policy = nfqa_verdict_batch_policy },
1392 };
1393
1394 static const struct nfnetlink_subsystem nfqnl_subsys = {
1395         .name           = "nf_queue",
1396         .subsys_id      = NFNL_SUBSYS_QUEUE,
1397         .cb_count       = NFQNL_MSG_MAX,
1398         .cb             = nfqnl_cb,
1399 };
1400
1401 #ifdef CONFIG_PROC_FS
1402 struct iter_state {
1403         struct seq_net_private p;
1404         unsigned int bucket;
1405 };
1406
1407 static struct hlist_node *get_first(struct seq_file *seq)
1408 {
1409         struct iter_state *st = seq->private;
1410         struct net *net;
1411         struct nfnl_queue_net *q;
1412
1413         if (!st)
1414                 return NULL;
1415
1416         net = seq_file_net(seq);
1417         q = nfnl_queue_pernet(net);
1418         for (st->bucket = 0; st->bucket < INSTANCE_BUCKETS; st->bucket++) {
1419                 if (!hlist_empty(&q->instance_table[st->bucket]))
1420                         return q->instance_table[st->bucket].first;
1421         }
1422         return NULL;
1423 }
1424
1425 static struct hlist_node *get_next(struct seq_file *seq, struct hlist_node *h)
1426 {
1427         struct iter_state *st = seq->private;
1428         struct net *net = seq_file_net(seq);
1429
1430         h = h->next;
1431         while (!h) {
1432                 struct nfnl_queue_net *q;
1433
1434                 if (++st->bucket >= INSTANCE_BUCKETS)
1435                         return NULL;
1436
1437                 q = nfnl_queue_pernet(net);
1438                 h = q->instance_table[st->bucket].first;
1439         }
1440         return h;
1441 }
1442
1443 static struct hlist_node *get_idx(struct seq_file *seq, loff_t pos)
1444 {
1445         struct hlist_node *head;
1446         head = get_first(seq);
1447
1448         if (head)
1449                 while (pos && (head = get_next(seq, head)))
1450                         pos--;
1451         return pos ? NULL : head;
1452 }
1453
1454 static void *seq_start(struct seq_file *s, loff_t *pos)
1455         __acquires(nfnl_queue_pernet(seq_file_net(s))->instances_lock)
1456 {
1457         spin_lock(&nfnl_queue_pernet(seq_file_net(s))->instances_lock);
1458         return get_idx(s, *pos);
1459 }
1460
1461 static void *seq_next(struct seq_file *s, void *v, loff_t *pos)
1462 {
1463         (*pos)++;
1464         return get_next(s, v);
1465 }
1466
1467 static void seq_stop(struct seq_file *s, void *v)
1468         __releases(nfnl_queue_pernet(seq_file_net(s))->instances_lock)
1469 {
1470         spin_unlock(&nfnl_queue_pernet(seq_file_net(s))->instances_lock);
1471 }
1472
1473 static int seq_show(struct seq_file *s, void *v)
1474 {
1475         const struct nfqnl_instance *inst = v;
1476
1477         seq_printf(s, "%5u %6u %5u %1u %5u %5u %5u %8u %2d\n",
1478                    inst->queue_num,
1479                    inst->peer_portid, inst->queue_total,
1480                    inst->copy_mode, inst->copy_range,
1481                    inst->queue_dropped, inst->queue_user_dropped,
1482                    inst->id_sequence, 1);
1483         return 0;
1484 }
1485
1486 static const struct seq_operations nfqnl_seq_ops = {
1487         .start  = seq_start,
1488         .next   = seq_next,
1489         .stop   = seq_stop,
1490         .show   = seq_show,
1491 };
1492
1493 static int nfqnl_open(struct inode *inode, struct file *file)
1494 {
1495         return seq_open_net(inode, file, &nfqnl_seq_ops,
1496                         sizeof(struct iter_state));
1497 }
1498
1499 static const struct file_operations nfqnl_file_ops = {
1500         .owner   = THIS_MODULE,
1501         .open    = nfqnl_open,
1502         .read    = seq_read,
1503         .llseek  = seq_lseek,
1504         .release = seq_release_net,
1505 };
1506
1507 #endif /* PROC_FS */
1508
1509 static int __net_init nfnl_queue_net_init(struct net *net)
1510 {
1511         unsigned int i;
1512         struct nfnl_queue_net *q = nfnl_queue_pernet(net);
1513
1514         for (i = 0; i < INSTANCE_BUCKETS; i++)
1515                 INIT_HLIST_HEAD(&q->instance_table[i]);
1516
1517         spin_lock_init(&q->instances_lock);
1518
1519 #ifdef CONFIG_PROC_FS
1520         if (!proc_create("nfnetlink_queue", 0440,
1521                          net->nf.proc_netfilter, &nfqnl_file_ops))
1522                 return -ENOMEM;
1523 #endif
1524         nf_register_queue_handler(net, &nfqh);
1525         return 0;
1526 }
1527
1528 static void __net_exit nfnl_queue_net_exit(struct net *net)
1529 {
1530         nf_unregister_queue_handler(net);
1531 #ifdef CONFIG_PROC_FS
1532         remove_proc_entry("nfnetlink_queue", net->nf.proc_netfilter);
1533 #endif
1534 }
1535
1536 static void nfnl_queue_net_exit_batch(struct list_head *net_exit_list)
1537 {
1538         synchronize_rcu();
1539 }
1540
1541 static struct pernet_operations nfnl_queue_net_ops = {
1542         .init           = nfnl_queue_net_init,
1543         .exit           = nfnl_queue_net_exit,
1544         .exit_batch     = nfnl_queue_net_exit_batch,
1545         .id             = &nfnl_queue_net_id,
1546         .size           = sizeof(struct nfnl_queue_net),
1547 };
1548
1549 static int __init nfnetlink_queue_init(void)
1550 {
1551         int status;
1552
1553         status = register_pernet_subsys(&nfnl_queue_net_ops);
1554         if (status < 0) {
1555                 pr_err("nf_queue: failed to register pernet ops\n");
1556                 goto out;
1557         }
1558
1559         netlink_register_notifier(&nfqnl_rtnl_notifier);
1560         status = nfnetlink_subsys_register(&nfqnl_subsys);
1561         if (status < 0) {
1562                 pr_err("nf_queue: failed to create netlink socket\n");
1563                 goto cleanup_netlink_notifier;
1564         }
1565
1566         status = register_netdevice_notifier(&nfqnl_dev_notifier);
1567         if (status < 0) {
1568                 pr_err("nf_queue: failed to register netdevice notifier\n");
1569                 goto cleanup_netlink_subsys;
1570         }
1571
1572         return status;
1573
1574 cleanup_netlink_subsys:
1575         nfnetlink_subsys_unregister(&nfqnl_subsys);
1576 cleanup_netlink_notifier:
1577         netlink_unregister_notifier(&nfqnl_rtnl_notifier);
1578         unregister_pernet_subsys(&nfnl_queue_net_ops);
1579 out:
1580         return status;
1581 }
1582
1583 static void __exit nfnetlink_queue_fini(void)
1584 {
1585         unregister_netdevice_notifier(&nfqnl_dev_notifier);
1586         nfnetlink_subsys_unregister(&nfqnl_subsys);
1587         netlink_unregister_notifier(&nfqnl_rtnl_notifier);
1588         unregister_pernet_subsys(&nfnl_queue_net_ops);
1589
1590         rcu_barrier(); /* Wait for completion of call_rcu()'s */
1591 }
1592
1593 MODULE_DESCRIPTION("netfilter packet queue handler");
1594 MODULE_AUTHOR("Harald Welte <laforge@netfilter.org>");
1595 MODULE_LICENSE("GPL");
1596 MODULE_ALIAS_NFNL_SUBSYS(NFNL_SUBSYS_QUEUE);
1597
1598 module_init(nfnetlink_queue_init);
1599 module_exit(nfnetlink_queue_fini);