net/netfilter/nfnetlink_queue.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  * This is a module which is used for queueing packets and communicating with
   4  * userspace via nfnetlink.
   5  *
   6  * (C) 2005 by Harald Welte <laforge@netfilter.org>
   7  * (C) 2007 by Patrick McHardy <kaber@trash.net>
   8  *
   9  * Based on the old ipv4-only ip_queue.c:
  10  * (C) 2000-2002 James Morris <jmorris@intercode.com.au>
  11  * (C) 2003-2005 Netfilter Core Team <coreteam@netfilter.org>
  12  */
  13
  14 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  15
  16 #include <linux/module.h>
  17 #include <linux/skbuff.h>
  18 #include <linux/init.h>
  19 #include <linux/spinlock.h>
  20 #include <linux/slab.h>
  21 #include <linux/notifier.h>
  22 #include <linux/netdevice.h>
  23 #include <linux/netfilter.h>
  24 #include <linux/proc_fs.h>
  25 #include <linux/netfilter_ipv4.h>
  26 #include <linux/netfilter_ipv6.h>
  27 #include <linux/netfilter_bridge.h>
  28 #include <linux/netfilter/nfnetlink.h>
  29 #include <linux/netfilter/nfnetlink_queue.h>
  30 #include <linux/netfilter/nf_conntrack_common.h>
  31 #include <linux/list.h>
  32 #include <linux/cgroup-defs.h>
  33 #include <net/gso.h>
  34 #include <net/sock.h>
  35 #include <net/tcp_states.h>
  36 #include <net/netfilter/nf_queue.h>
  37 #include <net/netns/generic.h>
  38
  39 #include <linux/atomic.h>
  40
  41 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
  42 #include "../bridge/br_private.h"
  43 #endif
  44
  45 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
  46 #include <net/netfilter/nf_conntrack.h>
  47 #endif
  48
  49 #define NFQNL_QMAX_DEFAULT 1024
  50
  51 /* We're using struct nlattr which has 16bit nla_len. Note that nla_len
  52  * includes the header length. Thus, the maximum packet length that we
  53  * support is 65531 bytes. We send truncated packets if the specified length
  54  * is larger than that.  Userspace can check for presence of NFQA_CAP_LEN
  55  * attribute to detect truncation.
  56  */
  57 #define NFQNL_MAX_COPY_RANGE (0xffff - NLA_HDRLEN)
  58
  59 struct nfqnl_instance {
  60         struct hlist_node hlist;                /* global list of queues */
  61         struct rcu_head rcu;
  62
  63         u32 peer_portid;
  64         unsigned int queue_maxlen;
  65         unsigned int copy_range;
  66         unsigned int queue_dropped;
  67         unsigned int queue_user_dropped;
  68
  69
  70         u_int16_t queue_num;                    /* number of this queue */
  71         u_int8_t copy_mode;
  72         u_int32_t flags;                        /* Set using NFQA_CFG_FLAGS */
  73 /*
  74  * Following fields are dirtied for each queued packet,
  75  * keep them in same cache line if possible.
  76  */
  77         spinlock_t      lock    ____cacheline_aligned_in_smp;
  78         unsigned int    queue_total;
  79         unsigned int    id_sequence;            /* 'sequence' of pkt ids */
  80         struct list_head queue_list;            /* packets in queue */
  81 };
  82
  83 typedef int (*nfqnl_cmpfn)(struct nf_queue_entry *, unsigned long);
  84
  85 static unsigned int nfnl_queue_net_id __read_mostly;
  86
  87 #define INSTANCE_BUCKETS        16
  88 struct nfnl_queue_net {
  89         spinlock_t instances_lock;
  90         struct hlist_head instance_table[INSTANCE_BUCKETS];
  91 };
  92
  93 static struct nfnl_queue_net *nfnl_queue_pernet(struct net *net)
  94 {
  95         return net_generic(net, nfnl_queue_net_id);
  96 }
  97
  98 static inline u_int8_t instance_hashfn(u_int16_t queue_num)
  99 {
 100         return ((queue_num >> 8) ^ queue_num) % INSTANCE_BUCKETS;
 101 }
 102
 103 static struct nfqnl_instance *
 104 instance_lookup(struct nfnl_queue_net *q, u_int16_t queue_num)
 105 {
 106         struct hlist_head *head;
 107         struct nfqnl_instance *inst;
 108
 109         head = &q->instance_table[instance_hashfn(queue_num)];
 110         hlist_for_each_entry_rcu(inst, head, hlist) {
 111                 if (inst->queue_num == queue_num)
 112                         return inst;
 113         }
 114         return NULL;
 115 }
 116
 117 static struct nfqnl_instance *
 118 instance_create(struct nfnl_queue_net *q, u_int16_t queue_num, u32 portid)
 119 {
 120         struct nfqnl_instance *inst;
 121         unsigned int h;
 122         int err;
 123
 124         spin_lock(&q->instances_lock);
 125         if (instance_lookup(q, queue_num)) {
 126                 err = -EEXIST;
 127                 goto out_unlock;
 128         }
 129
 130         inst = kzalloc(sizeof(*inst), GFP_ATOMIC);
 131         if (!inst) {
 132                 err = -ENOMEM;
 133                 goto out_unlock;
 134         }
 135
 136         inst->queue_num = queue_num;
 137         inst->peer_portid = portid;
 138         inst->queue_maxlen = NFQNL_QMAX_DEFAULT;
 139         inst->copy_range = NFQNL_MAX_COPY_RANGE;
 140         inst->copy_mode = NFQNL_COPY_NONE;
 141         spin_lock_init(&inst->lock);
 142         INIT_LIST_HEAD(&inst->queue_list);
 143
 144         if (!try_module_get(THIS_MODULE)) {
 145                 err = -EAGAIN;
 146                 goto out_free;
 147         }
 148
 149         h = instance_hashfn(queue_num);
 150         hlist_add_head_rcu(&inst->hlist, &q->instance_table[h]);
 151
 152         spin_unlock(&q->instances_lock);
 153
 154         return inst;
 155
 156 out_free:
 157         kfree(inst);
 158 out_unlock:
 159         spin_unlock(&q->instances_lock);
 160         return ERR_PTR(err);
 161 }
 162
 163 static void nfqnl_flush(struct nfqnl_instance *queue, nfqnl_cmpfn cmpfn,
 164                         unsigned long data);
 165
 166 static void
 167 instance_destroy_rcu(struct rcu_head *head)
 168 {
 169         struct nfqnl_instance *inst = container_of(head, struct nfqnl_instance,
 170                                                    rcu);
 171
 172         nfqnl_flush(inst, NULL, 0);
 173         kfree(inst);
 174         module_put(THIS_MODULE);
 175 }
 176
 177 static void
 178 __instance_destroy(struct nfqnl_instance *inst)
 179 {
 180         hlist_del_rcu(&inst->hlist);
 181         call_rcu(&inst->rcu, instance_destroy_rcu);
 182 }
 183
 184 static void
 185 instance_destroy(struct nfnl_queue_net *q, struct nfqnl_instance *inst)
 186 {
 187         spin_lock(&q->instances_lock);
 188         __instance_destroy(inst);
 189         spin_unlock(&q->instances_lock);
 190 }
 191
 192 static inline void
 193 __enqueue_entry(struct nfqnl_instance *queue, struct nf_queue_entry *entry)
 194 {
 195        list_add_tail(&entry->list, &queue->queue_list);
 196        queue->queue_total++;
 197 }
 198
 199 static void
 200 __dequeue_entry(struct nfqnl_instance *queue, struct nf_queue_entry *entry)
 201 {
 202         list_del(&entry->list);
 203         queue->queue_total--;
 204 }
 205
 206 static struct nf_queue_entry *
 207 find_dequeue_entry(struct nfqnl_instance *queue, unsigned int id)
 208 {
 209         struct nf_queue_entry *entry = NULL, *i;
 210
 211         spin_lock_bh(&queue->lock);
 212
 213         list_for_each_entry(i, &queue->queue_list, list) {
 214                 if (i->id == id) {
 215                         entry = i;
 216                         break;
 217                 }
 218         }
 219
 220         if (entry)
 221                 __dequeue_entry(queue, entry);
 222
 223         spin_unlock_bh(&queue->lock);
 224
 225         return entry;
 226 }
 227
 228 static void nfqnl_reinject(struct nf_queue_entry *entry, unsigned int verdict)
 229 {
 230         const struct nf_ct_hook *ct_hook;
 231
 232         if (verdict == NF_ACCEPT ||
 233             verdict == NF_REPEAT ||
 234             verdict == NF_STOP) {
 235                 rcu_read_lock();
 236                 ct_hook = rcu_dereference(nf_ct_hook);
 237                 if (ct_hook)
 238                         verdict = ct_hook->update(entry->state.net, entry->skb);
 239                 rcu_read_unlock();
 240
 241                 switch (verdict & NF_VERDICT_MASK) {
 242                 case NF_STOLEN:
 243                         nf_queue_entry_free(entry);
 244                         return;
 245                 }
 246
 247         }
 248         nf_reinject(entry, verdict);
 249 }
 250
 251 static void
 252 nfqnl_flush(struct nfqnl_instance *queue, nfqnl_cmpfn cmpfn, unsigned long data)
 253 {
 254         struct nf_queue_entry *entry, *next;
 255
 256         spin_lock_bh(&queue->lock);
 257         list_for_each_entry_safe(entry, next, &queue->queue_list, list) {
 258                 if (!cmpfn || cmpfn(entry, data)) {
 259                         list_del(&entry->list);
 260                         queue->queue_total--;
 261                         nfqnl_reinject(entry, NF_DROP);
 262                 }
 263         }
 264         spin_unlock_bh(&queue->lock);
 265 }
 266
 267 static int
 268 nfqnl_put_packet_info(struct sk_buff *nlskb, struct sk_buff *packet,
 269                       bool csum_verify)
 270 {
 271         __u32 flags = 0;
 272
 273         if (packet->ip_summed == CHECKSUM_PARTIAL)
 274                 flags = NFQA_SKB_CSUMNOTREADY;
 275         else if (csum_verify)
 276                 flags = NFQA_SKB_CSUM_NOTVERIFIED;
 277
 278         if (skb_is_gso(packet))
 279                 flags |= NFQA_SKB_GSO;
 280
 281         return flags ? nla_put_be32(nlskb, NFQA_SKB_INFO, htonl(flags)) : 0;
 282 }
 283
 284 static int nfqnl_put_sk_uidgid(struct sk_buff *skb, struct sock *sk)
 285 {
 286         const struct cred *cred;
 287
 288         if (!sk_fullsock(sk))
 289                 return 0;
 290
 291         read_lock_bh(&sk->sk_callback_lock);
 292         if (sk->sk_socket && sk->sk_socket->file) {
 293                 cred = sk->sk_socket->file->f_cred;
 294                 if (nla_put_be32(skb, NFQA_UID,
 295                     htonl(from_kuid_munged(&init_user_ns, cred->fsuid))))
 296                         goto nla_put_failure;
 297                 if (nla_put_be32(skb, NFQA_GID,
 298                     htonl(from_kgid_munged(&init_user_ns, cred->fsgid))))
 299                         goto nla_put_failure;
 300         }
 301         read_unlock_bh(&sk->sk_callback_lock);
 302         return 0;
 303
 304 nla_put_failure:
 305         read_unlock_bh(&sk->sk_callback_lock);
 306         return -1;
 307 }
 308
 309 static int nfqnl_put_sk_classid(struct sk_buff *skb, struct sock *sk)
 310 {
 311 #if IS_ENABLED(CONFIG_CGROUP_NET_CLASSID)
 312         if (sk && sk_fullsock(sk)) {
 313                 u32 classid = sock_cgroup_classid(&sk->sk_cgrp_data);
 314
 315                 if (classid && nla_put_be32(skb, NFQA_CGROUP_CLASSID, htonl(classid)))
 316                         return -1;
 317         }
 318 #endif
 319         return 0;
 320 }
 321
 322 static u32 nfqnl_get_sk_secctx(struct sk_buff *skb, char **secdata)
 323 {
 324         u32 seclen = 0;
 325 #if IS_ENABLED(CONFIG_NETWORK_SECMARK)
 326         if (!skb || !sk_fullsock(skb->sk))
 327                 return 0;
 328
 329         read_lock_bh(&skb->sk->sk_callback_lock);
 330
 331         if (skb->secmark)
 332                 security_secid_to_secctx(skb->secmark, secdata, &seclen);
 333
 334         read_unlock_bh(&skb->sk->sk_callback_lock);
 335 #endif
 336         return seclen;
 337 }
 338
 339 static u32 nfqnl_get_bridge_size(struct nf_queue_entry *entry)
 340 {
 341         struct sk_buff *entskb = entry->skb;
 342         u32 nlalen = 0;
 343
 344         if (entry->state.pf != PF_BRIDGE || !skb_mac_header_was_set(entskb))
 345                 return 0;
 346
 347         if (skb_vlan_tag_present(entskb))
 348                 nlalen += nla_total_size(nla_total_size(sizeof(__be16)) +
 349                                          nla_total_size(sizeof(__be16)));
 350
 351         if (entskb->network_header > entskb->mac_header)
 352                 nlalen += nla_total_size((entskb->network_header -
 353                                           entskb->mac_header));
 354
 355         return nlalen;
 356 }
 357
 358 static int nfqnl_put_bridge(struct nf_queue_entry *entry, struct sk_buff *skb)
 359 {
 360         struct sk_buff *entskb = entry->skb;
 361
 362         if (entry->state.pf != PF_BRIDGE || !skb_mac_header_was_set(entskb))
 363                 return 0;
 364
 365         if (skb_vlan_tag_present(entskb)) {
 366                 struct nlattr *nest;
 367
 368                 nest = nla_nest_start(skb, NFQA_VLAN);
 369                 if (!nest)
 370                         goto nla_put_failure;
 371
 372                 if (nla_put_be16(skb, NFQA_VLAN_TCI, htons(entskb->vlan_tci)) ||
 373                     nla_put_be16(skb, NFQA_VLAN_PROTO, entskb->vlan_proto))
 374                         goto nla_put_failure;
 375
 376                 nla_nest_end(skb, nest);
 377         }
 378
 379         if (entskb->mac_header < entskb->network_header) {
 380                 int len = (int)(entskb->network_header - entskb->mac_header);
 381
 382                 if (nla_put(skb, NFQA_L2HDR, len, skb_mac_header(entskb)))
 383                         goto nla_put_failure;
 384         }
 385
 386         return 0;
 387
 388 nla_put_failure:
 389         return -1;
 390 }
 391
 392 static struct sk_buff *
 393 nfqnl_build_packet_message(struct net *net, struct nfqnl_instance *queue,
 394                            struct nf_queue_entry *entry,
 395                            __be32 **packet_id_ptr)
 396 {
 397         size_t size;
 398         size_t data_len = 0, cap_len = 0;
 399         unsigned int hlen = 0;
 400         struct sk_buff *skb;
 401         struct nlattr *nla;
 402         struct nfqnl_msg_packet_hdr *pmsg;
 403         struct nlmsghdr *nlh;
 404         struct sk_buff *entskb = entry->skb;
 405         struct net_device *indev;
 406         struct net_device *outdev;
 407         struct nf_conn *ct = NULL;
 408         enum ip_conntrack_info ctinfo = 0;
 409         const struct nfnl_ct_hook *nfnl_ct;
 410         bool csum_verify;
 411         char *secdata = NULL;
 412         u32 seclen = 0;
 413         ktime_t tstamp;
 414
 415         size = nlmsg_total_size(sizeof(struct nfgenmsg))
 416                 + nla_total_size(sizeof(struct nfqnl_msg_packet_hdr))
 417                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
 418                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
 419 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
 420                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
 421                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
 422 #endif
 423                 + nla_total_size(sizeof(u_int32_t))     /* mark */
 424                 + nla_total_size(sizeof(u_int32_t))     /* priority */
 425                 + nla_total_size(sizeof(struct nfqnl_msg_packet_hw))
 426                 + nla_total_size(sizeof(u_int32_t))     /* skbinfo */
 427 #if IS_ENABLED(CONFIG_CGROUP_NET_CLASSID)
 428                 + nla_total_size(sizeof(u_int32_t))     /* classid */
 429 #endif
 430                 + nla_total_size(sizeof(u_int32_t));    /* cap_len */
 431
 432         tstamp = skb_tstamp_cond(entskb, false);
 433         if (tstamp)
 434                 size += nla_total_size(sizeof(struct nfqnl_msg_packet_timestamp));
 435
 436         size += nfqnl_get_bridge_size(entry);
 437
 438         if (entry->state.hook <= NF_INET_FORWARD ||
 439            (entry->state.hook == NF_INET_POST_ROUTING && entskb->sk == NULL))
 440                 csum_verify = !skb_csum_unnecessary(entskb);
 441         else
 442                 csum_verify = false;
 443
 444         outdev = entry->state.out;
 445
 446         switch ((enum nfqnl_config_mode)READ_ONCE(queue->copy_mode)) {
 447         case NFQNL_COPY_META:
 448         case NFQNL_COPY_NONE:
 449                 break;
 450
 451         case NFQNL_COPY_PACKET:
 452                 if (!(queue->flags & NFQA_CFG_F_GSO) &&
 453                     entskb->ip_summed == CHECKSUM_PARTIAL &&
 454                     skb_checksum_help(entskb))
 455                         return NULL;
 456
 457                 data_len = READ_ONCE(queue->copy_range);
 458                 if (data_len > entskb->len)
 459                         data_len = entskb->len;
 460
 461                 hlen = skb_zerocopy_headlen(entskb);
 462                 hlen = min_t(unsigned int, hlen, data_len);
 463                 size += sizeof(struct nlattr) + hlen;
 464                 cap_len = entskb->len;
 465                 break;
 466         }
 467
 468         nfnl_ct = rcu_dereference(nfnl_ct_hook);
 469
 470 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
 471         if (queue->flags & NFQA_CFG_F_CONNTRACK) {
 472                 if (nfnl_ct != NULL) {
 473                         ct = nf_ct_get(entskb, &ctinfo);
 474                         if (ct != NULL)
 475                                 size += nfnl_ct->build_size(ct);
 476                 }
 477         }
 478 #endif
 479
 480         if (queue->flags & NFQA_CFG_F_UID_GID) {
 481                 size += (nla_total_size(sizeof(u_int32_t))      /* uid */
 482                         + nla_total_size(sizeof(u_int32_t)));   /* gid */
 483         }
 484
 485         if ((queue->flags & NFQA_CFG_F_SECCTX) && entskb->sk) {
 486                 seclen = nfqnl_get_sk_secctx(entskb, &secdata);
 487                 if (seclen)
 488                         size += nla_total_size(seclen);
 489         }
 490
 491         skb = alloc_skb(size, GFP_ATOMIC);
 492         if (!skb) {
 493                 skb_tx_error(entskb);
 494                 goto nlmsg_failure;
 495         }
 496
 497         nlh = nfnl_msg_put(skb, 0, 0,
 498                            nfnl_msg_type(NFNL_SUBSYS_QUEUE, NFQNL_MSG_PACKET),
 499                            0, entry->state.pf, NFNETLINK_V0,
 500                            htons(queue->queue_num));
 501         if (!nlh) {
 502                 skb_tx_error(entskb);
 503                 kfree_skb(skb);
 504                 goto nlmsg_failure;
 505         }
 506
 507         nla = __nla_reserve(skb, NFQA_PACKET_HDR, sizeof(*pmsg));
 508         pmsg = nla_data(nla);
 509         pmsg->hw_protocol       = entskb->protocol;
 510         pmsg->hook              = entry->state.hook;
 511         *packet_id_ptr          = &pmsg->packet_id;
 512
 513         indev = entry->state.in;
 514         if (indev) {
 515 #if !IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
 516                 if (nla_put_be32(skb, NFQA_IFINDEX_INDEV, htonl(indev->ifindex)))
 517                         goto nla_put_failure;
 518 #else
 519                 if (entry->state.pf == PF_BRIDGE) {
 520                         /* Case 1: indev is physical input device, we need to
 521                          * look for bridge group (when called from
 522                          * netfilter_bridge) */
 523                         if (nla_put_be32(skb, NFQA_IFINDEX_PHYSINDEV,
 524                                          htonl(indev->ifindex)) ||
 525                         /* this is the bridge group "brX" */
 526                         /* rcu_read_lock()ed by __nf_queue */
 527                             nla_put_be32(skb, NFQA_IFINDEX_INDEV,
 528                                          htonl(br_port_get_rcu(indev)->br->dev->ifindex)))
 529                                 goto nla_put_failure;
 530                 } else {
 531                         int physinif;
 532
 533                         /* Case 2: indev is bridge group, we need to look for
 534                          * physical device (when called from ipv4) */
 535                         if (nla_put_be32(skb, NFQA_IFINDEX_INDEV,
 536                                          htonl(indev->ifindex)))
 537                                 goto nla_put_failure;
 538
 539                         physinif = nf_bridge_get_physinif(entskb);
 540                         if (physinif &&
 541                             nla_put_be32(skb, NFQA_IFINDEX_PHYSINDEV,
 542                                          htonl(physinif)))
 543                                 goto nla_put_failure;
 544                 }
 545 #endif
 546         }
 547
 548         if (outdev) {
 549 #if !IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
 550                 if (nla_put_be32(skb, NFQA_IFINDEX_OUTDEV, htonl(outdev->ifindex)))
 551                         goto nla_put_failure;
 552 #else
 553                 if (entry->state.pf == PF_BRIDGE) {
 554                         /* Case 1: outdev is physical output device, we need to
 555                          * look for bridge group (when called from
 556                          * netfilter_bridge) */
 557                         if (nla_put_be32(skb, NFQA_IFINDEX_PHYSOUTDEV,
 558                                          htonl(outdev->ifindex)) ||
 559                         /* this is the bridge group "brX" */
 560                         /* rcu_read_lock()ed by __nf_queue */
 561                             nla_put_be32(skb, NFQA_IFINDEX_OUTDEV,
 562                                          htonl(br_port_get_rcu(outdev)->br->dev->ifindex)))
 563                                 goto nla_put_failure;
 564                 } else {
 565                         int physoutif;
 566
 567                         /* Case 2: outdev is bridge group, we need to look for
 568                          * physical output device (when called from ipv4) */
 569                         if (nla_put_be32(skb, NFQA_IFINDEX_OUTDEV,
 570                                          htonl(outdev->ifindex)))
 571                                 goto nla_put_failure;
 572
 573                         physoutif = nf_bridge_get_physoutif(entskb);
 574                         if (physoutif &&
 575                             nla_put_be32(skb, NFQA_IFINDEX_PHYSOUTDEV,
 576                                          htonl(physoutif)))
 577                                 goto nla_put_failure;
 578                 }
 579 #endif
 580         }
 581
 582         if (entskb->mark &&
 583             nla_put_be32(skb, NFQA_MARK, htonl(entskb->mark)))
 584                 goto nla_put_failure;
 585
 586         if (entskb->priority &&
 587             nla_put_be32(skb, NFQA_PRIORITY, htonl(entskb->priority)))
 588                 goto nla_put_failure;
 589
 590         if (indev && entskb->dev &&
 591             skb_mac_header_was_set(entskb) &&
 592             skb_mac_header_len(entskb) != 0) {
 593                 struct nfqnl_msg_packet_hw phw;
 594                 int len;
 595
 596                 memset(&phw, 0, sizeof(phw));
 597                 len = dev_parse_header(entskb, phw.hw_addr);
 598                 if (len) {
 599                         phw.hw_addrlen = htons(len);
 600                         if (nla_put(skb, NFQA_HWADDR, sizeof(phw), &phw))
 601                                 goto nla_put_failure;
 602                 }
 603         }
 604
 605         if (nfqnl_put_bridge(entry, skb) < 0)
 606                 goto nla_put_failure;
 607
 608         if (entry->state.hook <= NF_INET_FORWARD && tstamp) {
 609                 struct nfqnl_msg_packet_timestamp ts;
 610                 struct timespec64 kts = ktime_to_timespec64(tstamp);
 611
 612                 ts.sec = cpu_to_be64(kts.tv_sec);
 613                 ts.usec = cpu_to_be64(kts.tv_nsec / NSEC_PER_USEC);
 614
 615                 if (nla_put(skb, NFQA_TIMESTAMP, sizeof(ts), &ts))
 616                         goto nla_put_failure;
 617         }
 618
 619         if ((queue->flags & NFQA_CFG_F_UID_GID) && entskb->sk &&
 620             nfqnl_put_sk_uidgid(skb, entskb->sk) < 0)
 621                 goto nla_put_failure;
 622
 623         if (nfqnl_put_sk_classid(skb, entskb->sk) < 0)
 624                 goto nla_put_failure;
 625
 626         if (seclen && nla_put(skb, NFQA_SECCTX, seclen, secdata))
 627                 goto nla_put_failure;
 628
 629         if (ct && nfnl_ct->build(skb, ct, ctinfo, NFQA_CT, NFQA_CT_INFO) < 0)
 630                 goto nla_put_failure;
 631
 632         if (cap_len > data_len &&
 633             nla_put_be32(skb, NFQA_CAP_LEN, htonl(cap_len)))
 634                 goto nla_put_failure;
 635
 636         if (nfqnl_put_packet_info(skb, entskb, csum_verify))
 637                 goto nla_put_failure;
 638
 639         if (data_len) {
 640                 struct nlattr *nla;
 641
 642                 if (skb_tailroom(skb) < sizeof(*nla) + hlen)
 643                         goto nla_put_failure;
 644
 645                 nla = skb_put(skb, sizeof(*nla));
 646                 nla->nla_type = NFQA_PAYLOAD;
 647                 nla->nla_len = nla_attr_size(data_len);
 648
 649                 if (skb_zerocopy(skb, entskb, data_len, hlen))
 650                         goto nla_put_failure;
 651         }
 652
 653         nlh->nlmsg_len = skb->len;
 654         if (seclen)
 655                 security_release_secctx(secdata, seclen);
 656         return skb;
 657
 658 nla_put_failure:
 659         skb_tx_error(entskb);
 660         kfree_skb(skb);
 661         net_err_ratelimited("nf_queue: error creating packet message\n");
 662 nlmsg_failure:
 663         if (seclen)
 664                 security_release_secctx(secdata, seclen);
 665         return NULL;
 666 }
 667
 668 static bool nf_ct_drop_unconfirmed(const struct nf_queue_entry *entry)
 669 {
 670 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
 671         static const unsigned long flags = IPS_CONFIRMED | IPS_DYING;
 672         const struct nf_conn *ct = (void *)skb_nfct(entry->skb);
 673
 674         if (ct && ((ct->status & flags) == IPS_DYING))
 675                 return true;
 676 #endif
 677         return false;
 678 }
 679
 680 static int
 681 __nfqnl_enqueue_packet(struct net *net, struct nfqnl_instance *queue,
 682                         struct nf_queue_entry *entry)
 683 {
 684         struct sk_buff *nskb;
 685         int err = -ENOBUFS;
 686         __be32 *packet_id_ptr;
 687         int failopen = 0;
 688
 689         nskb = nfqnl_build_packet_message(net, queue, entry, &packet_id_ptr);
 690         if (nskb == NULL) {
 691                 err = -ENOMEM;
 692                 goto err_out;
 693         }
 694         spin_lock_bh(&queue->lock);
 695
 696         if (nf_ct_drop_unconfirmed(entry))
 697                 goto err_out_free_nskb;
 698
 699         if (queue->queue_total >= queue->queue_maxlen) {
 700                 if (queue->flags & NFQA_CFG_F_FAIL_OPEN) {
 701                         failopen = 1;
 702                         err = 0;
 703                 } else {
 704                         queue->queue_dropped++;
 705                         net_warn_ratelimited("nf_queue: full at %d entries, dropping packets(s)\n",
 706                                              queue->queue_total);
 707                 }
 708                 goto err_out_free_nskb;
 709         }
 710         entry->id = ++queue->id_sequence;
 711         *packet_id_ptr = htonl(entry->id);
 712
 713         /* nfnetlink_unicast will either free the nskb or add it to a socket */
 714         err = nfnetlink_unicast(nskb, net, queue->peer_portid);
 715         if (err < 0) {
 716                 if (queue->flags & NFQA_CFG_F_FAIL_OPEN) {
 717                         failopen = 1;
 718                         err = 0;
 719                 } else {
 720                         queue->queue_user_dropped++;
 721                 }
 722                 goto err_out_unlock;
 723         }
 724
 725         __enqueue_entry(queue, entry);
 726
 727         spin_unlock_bh(&queue->lock);
 728         return 0;
 729
 730 err_out_free_nskb:
 731         kfree_skb(nskb);
 732 err_out_unlock:
 733         spin_unlock_bh(&queue->lock);
 734         if (failopen)
 735                 nfqnl_reinject(entry, NF_ACCEPT);
 736 err_out:
 737         return err;
 738 }
 739
 740 static struct nf_queue_entry *
 741 nf_queue_entry_dup(struct nf_queue_entry *e)
 742 {
 743         struct nf_queue_entry *entry = kmemdup(e, e->size, GFP_ATOMIC);
 744
 745         if (!entry)
 746                 return NULL;
 747
 748         if (nf_queue_entry_get_refs(entry))
 749                 return entry;
 750
 751         kfree(entry);
 752         return NULL;
 753 }
 754
 755 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
 756 /* When called from bridge netfilter, skb->data must point to MAC header
 757  * before calling skb_gso_segment(). Else, original MAC header is lost
 758  * and segmented skbs will be sent to wrong destination.
 759  */
 760 static void nf_bridge_adjust_skb_data(struct sk_buff *skb)
 761 {
 762         if (nf_bridge_info_get(skb))
 763                 __skb_push(skb, skb->network_header - skb->mac_header);
 764 }
 765
 766 static void nf_bridge_adjust_segmented_data(struct sk_buff *skb)
 767 {
 768         if (nf_bridge_info_get(skb))
 769                 __skb_pull(skb, skb->network_header - skb->mac_header);
 770 }
 771 #else
 772 #define nf_bridge_adjust_skb_data(s) do {} while (0)
 773 #define nf_bridge_adjust_segmented_data(s) do {} while (0)
 774 #endif
 775
 776 static int
 777 __nfqnl_enqueue_packet_gso(struct net *net, struct nfqnl_instance *queue,
 778                            struct sk_buff *skb, struct nf_queue_entry *entry)
 779 {
 780         int ret = -ENOMEM;
 781         struct nf_queue_entry *entry_seg;
 782
 783         nf_bridge_adjust_segmented_data(skb);
 784
 785         if (skb->next == NULL) { /* last packet, no need to copy entry */
 786                 struct sk_buff *gso_skb = entry->skb;
 787                 entry->skb = skb;
 788                 ret = __nfqnl_enqueue_packet(net, queue, entry);
 789                 if (ret)
 790                         entry->skb = gso_skb;
 791                 return ret;
 792         }
 793
 794         skb_mark_not_on_list(skb);
 795
 796         entry_seg = nf_queue_entry_dup(entry);
 797         if (entry_seg) {
 798                 entry_seg->skb = skb;
 799                 ret = __nfqnl_enqueue_packet(net, queue, entry_seg);
 800                 if (ret)
 801                         nf_queue_entry_free(entry_seg);
 802         }
 803         return ret;
 804 }
 805
 806 static int
 807 nfqnl_enqueue_packet(struct nf_queue_entry *entry, unsigned int queuenum)
 808 {
 809         unsigned int queued;
 810         struct nfqnl_instance *queue;
 811         struct sk_buff *skb, *segs, *nskb;
 812         int err = -ENOBUFS;
 813         struct net *net = entry->state.net;
 814         struct nfnl_queue_net *q = nfnl_queue_pernet(net);
 815
 816         /* rcu_read_lock()ed by nf_hook_thresh */
 817         queue = instance_lookup(q, queuenum);
 818         if (!queue)
 819                 return -ESRCH;
 820
 821         if (queue->copy_mode == NFQNL_COPY_NONE)
 822                 return -EINVAL;
 823
 824         skb = entry->skb;
 825
 826         switch (entry->state.pf) {
 827         case NFPROTO_IPV4:
 828                 skb->protocol = htons(ETH_P_IP);
 829                 break;
 830         case NFPROTO_IPV6:
 831                 skb->protocol = htons(ETH_P_IPV6);
 832                 break;
 833         }
 834
 835         if ((queue->flags & NFQA_CFG_F_GSO) || !skb_is_gso(skb))
 836                 return __nfqnl_enqueue_packet(net, queue, entry);
 837
 838         nf_bridge_adjust_skb_data(skb);
 839         segs = skb_gso_segment(skb, 0);
 840         /* Does not use PTR_ERR to limit the number of error codes that can be
 841          * returned by nf_queue.  For instance, callers rely on -ESRCH to
 842          * mean 'ignore this hook'.
 843          */
 844         if (IS_ERR_OR_NULL(segs))
 845                 goto out_err;
 846         queued = 0;
 847         err = 0;
 848         skb_list_walk_safe(segs, segs, nskb) {
 849                 if (err == 0)
 850                         err = __nfqnl_enqueue_packet_gso(net, queue,
 851                                                         segs, entry);
 852                 if (err == 0)
 853                         queued++;
 854                 else
 855                         kfree_skb(segs);
 856         }
 857
 858         if (queued) {
 859                 if (err) /* some segments are already queued */
 860                         nf_queue_entry_free(entry);
 861                 kfree_skb(skb);
 862                 return 0;
 863         }
 864  out_err:
 865         nf_bridge_adjust_segmented_data(skb);
 866         return err;
 867 }
 868
 869 static int
 870 nfqnl_mangle(void *data, unsigned int data_len, struct nf_queue_entry *e, int diff)
 871 {
 872         struct sk_buff *nskb;
 873
 874         if (diff < 0) {
 875                 unsigned int min_len = skb_transport_offset(e->skb);
 876
 877                 if (data_len < min_len)
 878                         return -EINVAL;
 879
 880                 if (pskb_trim(e->skb, data_len))
 881                         return -ENOMEM;
 882         } else if (diff > 0) {
 883                 if (data_len > 0xFFFF)
 884                         return -EINVAL;
 885                 if (diff > skb_tailroom(e->skb)) {
 886                         nskb = skb_copy_expand(e->skb, skb_headroom(e->skb),
 887                                                diff, GFP_ATOMIC);
 888                         if (!nskb)
 889                                 return -ENOMEM;
 890                         kfree_skb(e->skb);
 891                         e->skb = nskb;
 892                 }
 893                 skb_put(e->skb, diff);
 894         }
 895         if (skb_ensure_writable(e->skb, data_len))
 896                 return -ENOMEM;
 897         skb_copy_to_linear_data(e->skb, data, data_len);
 898         e->skb->ip_summed = CHECKSUM_NONE;
 899         return 0;
 900 }
 901
 902 static int
 903 nfqnl_set_mode(struct nfqnl_instance *queue,
 904                unsigned char mode, unsigned int range)
 905 {
 906         int status = 0;
 907
 908         spin_lock_bh(&queue->lock);
 909         switch (mode) {
 910         case NFQNL_COPY_NONE:
 911         case NFQNL_COPY_META:
 912                 queue->copy_mode = mode;
 913                 queue->copy_range = 0;
 914                 break;
 915
 916         case NFQNL_COPY_PACKET:
 917                 queue->copy_mode = mode;
 918                 if (range == 0 || range > NFQNL_MAX_COPY_RANGE)
 919                         queue->copy_range = NFQNL_MAX_COPY_RANGE;
 920                 else
 921                         queue->copy_range = range;
 922                 break;
 923
 924         default:
 925                 status = -EINVAL;
 926
 927         }
 928         spin_unlock_bh(&queue->lock);
 929
 930         return status;
 931 }
 932
 933 static int
 934 dev_cmp(struct nf_queue_entry *entry, unsigned long ifindex)
 935 {
 936 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
 937         int physinif, physoutif;
 938
 939         physinif = nf_bridge_get_physinif(entry->skb);
 940         physoutif = nf_bridge_get_physoutif(entry->skb);
 941
 942         if (physinif == ifindex || physoutif == ifindex)
 943                 return 1;
 944 #endif
 945         if (entry->state.in)
 946                 if (entry->state.in->ifindex == ifindex)
 947                         return 1;
 948         if (entry->state.out)
 949                 if (entry->state.out->ifindex == ifindex)
 950                         return 1;
 951
 952         return 0;
 953 }
 954
 955 /* drop all packets with either indev or outdev == ifindex from all queue
 956  * instances */
 957 static void
 958 nfqnl_dev_drop(struct net *net, int ifindex)
 959 {
 960         int i;
 961         struct nfnl_queue_net *q = nfnl_queue_pernet(net);
 962
 963         rcu_read_lock();
 964
 965         for (i = 0; i < INSTANCE_BUCKETS; i++) {
 966                 struct nfqnl_instance *inst;
 967                 struct hlist_head *head = &q->instance_table[i];
 968
 969                 hlist_for_each_entry_rcu(inst, head, hlist)
 970                         nfqnl_flush(inst, dev_cmp, ifindex);
 971         }
 972
 973         rcu_read_unlock();
 974 }
 975
 976 static int
 977 nfqnl_rcv_dev_event(struct notifier_block *this,
 978                     unsigned long event, void *ptr)
 979 {
 980         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
 981
 982         /* Drop any packets associated with the downed device */
 983         if (event == NETDEV_DOWN)
 984                 nfqnl_dev_drop(dev_net(dev), dev->ifindex);
 985         return NOTIFY_DONE;
 986 }
 987
 988 static struct notifier_block nfqnl_dev_notifier = {
 989         .notifier_call  = nfqnl_rcv_dev_event,
 990 };
 991
 992 static void nfqnl_nf_hook_drop(struct net *net)
 993 {
 994         struct nfnl_queue_net *q = nfnl_queue_pernet(net);
 995         int i;
 996
 997         /* This function is also called on net namespace error unwind,
 998          * when pernet_ops->init() failed and ->exit() functions of the
 999          * previous pernet_ops gets called.
1000          *
1001          * This may result in a call to nfqnl_nf_hook_drop() before
1002          * struct nfnl_queue_net was allocated.
1003          */
1004         if (!q)
1005                 return;
1006
1007         for (i = 0; i < INSTANCE_BUCKETS; i++) {
1008                 struct nfqnl_instance *inst;
1009                 struct hlist_head *head = &q->instance_table[i];
1010
1011                 hlist_for_each_entry_rcu(inst, head, hlist)
1012                         nfqnl_flush(inst, NULL, 0);
1013         }
1014 }
1015
1016 static int
1017 nfqnl_rcv_nl_event(struct notifier_block *this,
1018                    unsigned long event, void *ptr)
1019 {
1020         struct netlink_notify *n = ptr;
1021         struct nfnl_queue_net *q = nfnl_queue_pernet(n->net);
1022
1023         if (event == NETLINK_URELEASE && n->protocol == NETLINK_NETFILTER) {
1024                 int i;
1025
1026                 /* destroy all instances for this portid */
1027                 spin_lock(&q->instances_lock);
1028                 for (i = 0; i < INSTANCE_BUCKETS; i++) {
1029                         struct hlist_node *t2;
1030                         struct nfqnl_instance *inst;
1031                         struct hlist_head *head = &q->instance_table[i];
1032
1033                         hlist_for_each_entry_safe(inst, t2, head, hlist) {
1034                                 if (n->portid == inst->peer_portid)
1035                                         __instance_destroy(inst);
1036                         }
1037                 }
1038                 spin_unlock(&q->instances_lock);
1039         }
1040         return NOTIFY_DONE;
1041 }
1042
1043 static struct notifier_block nfqnl_rtnl_notifier = {
1044         .notifier_call  = nfqnl_rcv_nl_event,
1045 };
1046
1047 static const struct nla_policy nfqa_vlan_policy[NFQA_VLAN_MAX + 1] = {
1048         [NFQA_VLAN_TCI]         = { .type = NLA_U16},
1049         [NFQA_VLAN_PROTO]       = { .type = NLA_U16},
1050 };
1051
1052 static const struct nla_policy nfqa_verdict_policy[NFQA_MAX+1] = {
1053         [NFQA_VERDICT_HDR]      = { .len = sizeof(struct nfqnl_msg_verdict_hdr) },
1054         [NFQA_MARK]             = { .type = NLA_U32 },
1055         [NFQA_PAYLOAD]          = { .type = NLA_UNSPEC },
1056         [NFQA_CT]               = { .type = NLA_UNSPEC },
1057         [NFQA_EXP]              = { .type = NLA_UNSPEC },
1058         [NFQA_VLAN]             = { .type = NLA_NESTED },
1059         [NFQA_PRIORITY]         = { .type = NLA_U32 },
1060 };
1061
1062 static const struct nla_policy nfqa_verdict_batch_policy[NFQA_MAX+1] = {
1063         [NFQA_VERDICT_HDR]      = { .len = sizeof(struct nfqnl_msg_verdict_hdr) },
1064         [NFQA_MARK]             = { .type = NLA_U32 },
1065         [NFQA_PRIORITY]         = { .type = NLA_U32 },
1066 };
1067
1068 static struct nfqnl_instance *
1069 verdict_instance_lookup(struct nfnl_queue_net *q, u16 queue_num, u32 nlportid)
1070 {
1071         struct nfqnl_instance *queue;
1072
1073         queue = instance_lookup(q, queue_num);
1074         if (!queue)
1075                 return ERR_PTR(-ENODEV);
1076
1077         if (queue->peer_portid != nlportid)
1078                 return ERR_PTR(-EPERM);
1079
1080         return queue;
1081 }
1082
1083 static struct nfqnl_msg_verdict_hdr*
1084 verdicthdr_get(const struct nlattr * const nfqa[])
1085 {
1086         struct nfqnl_msg_verdict_hdr *vhdr;
1087         unsigned int verdict;
1088
1089         if (!nfqa[NFQA_VERDICT_HDR])
1090                 return NULL;
1091
1092         vhdr = nla_data(nfqa[NFQA_VERDICT_HDR]);
1093         verdict = ntohl(vhdr->verdict) & NF_VERDICT_MASK;
1094         if (verdict > NF_MAX_VERDICT || verdict == NF_STOLEN)
1095                 return NULL;
1096         return vhdr;
1097 }
1098
1099 static int nfq_id_after(unsigned int id, unsigned int max)
1100 {
1101         return (int)(id - max) > 0;
1102 }
1103
1104 static int nfqnl_recv_verdict_batch(struct sk_buff *skb,
1105                                     const struct nfnl_info *info,
1106                                     const struct nlattr * const nfqa[])
1107 {
1108         struct nfnl_queue_net *q = nfnl_queue_pernet(info->net);
1109         u16 queue_num = ntohs(info->nfmsg->res_id);
1110         struct nf_queue_entry *entry, *tmp;
1111         struct nfqnl_msg_verdict_hdr *vhdr;
1112         struct nfqnl_instance *queue;
1113         unsigned int verdict, maxid;
1114         LIST_HEAD(batch_list);
1115
1116         queue = verdict_instance_lookup(q, queue_num,
1117                                         NETLINK_CB(skb).portid);
1118         if (IS_ERR(queue))
1119                 return PTR_ERR(queue);
1120
1121         vhdr = verdicthdr_get(nfqa);
1122         if (!vhdr)
1123                 return -EINVAL;
1124
1125         verdict = ntohl(vhdr->verdict);
1126         maxid = ntohl(vhdr->id);
1127
1128         spin_lock_bh(&queue->lock);
1129
1130         list_for_each_entry_safe(entry, tmp, &queue->queue_list, list) {
1131                 if (nfq_id_after(entry->id, maxid))
1132                         break;
1133                 __dequeue_entry(queue, entry);
1134                 list_add_tail(&entry->list, &batch_list);
1135         }
1136
1137         spin_unlock_bh(&queue->lock);
1138
1139         if (list_empty(&batch_list))
1140                 return -ENOENT;
1141
1142         list_for_each_entry_safe(entry, tmp, &batch_list, list) {
1143                 if (nfqa[NFQA_MARK])
1144                         entry->skb->mark = ntohl(nla_get_be32(nfqa[NFQA_MARK]));
1145
1146                 if (nfqa[NFQA_PRIORITY])
1147                         entry->skb->priority = ntohl(nla_get_be32(nfqa[NFQA_PRIORITY]));
1148
1149                 nfqnl_reinject(entry, verdict);
1150         }
1151         return 0;
1152 }
1153
1154 static struct nf_conn *nfqnl_ct_parse(const struct nfnl_ct_hook *nfnl_ct,
1155                                       const struct nlmsghdr *nlh,
1156                                       const struct nlattr * const nfqa[],
1157                                       struct nf_queue_entry *entry,
1158                                       enum ip_conntrack_info *ctinfo)
1159 {
1160 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
1161         struct nf_conn *ct;
1162
1163         ct = nf_ct_get(entry->skb, ctinfo);
1164         if (ct == NULL)
1165                 return NULL;
1166
1167         if (nfnl_ct->parse(nfqa[NFQA_CT], ct) < 0)
1168                 return NULL;
1169
1170         if (nfqa[NFQA_EXP])
1171                 nfnl_ct->attach_expect(nfqa[NFQA_EXP], ct,
1172                                       NETLINK_CB(entry->skb).portid,
1173                                       nlmsg_report(nlh));
1174         return ct;
1175 #else
1176         return NULL;
1177 #endif
1178 }
1179
1180 static int nfqa_parse_bridge(struct nf_queue_entry *entry,
1181                              const struct nlattr * const nfqa[])
1182 {
1183         if (nfqa[NFQA_VLAN]) {
1184                 struct nlattr *tb[NFQA_VLAN_MAX + 1];
1185                 int err;
1186
1187                 err = nla_parse_nested_deprecated(tb, NFQA_VLAN_MAX,
1188                                                   nfqa[NFQA_VLAN],
1189                                                   nfqa_vlan_policy, NULL);
1190                 if (err < 0)
1191                         return err;
1192
1193                 if (!tb[NFQA_VLAN_TCI] || !tb[NFQA_VLAN_PROTO])
1194                         return -EINVAL;
1195
1196                 __vlan_hwaccel_put_tag(entry->skb,
1197                         nla_get_be16(tb[NFQA_VLAN_PROTO]),
1198                         ntohs(nla_get_be16(tb[NFQA_VLAN_TCI])));
1199         }
1200
1201         if (nfqa[NFQA_L2HDR]) {
1202                 int mac_header_len = entry->skb->network_header -
1203                         entry->skb->mac_header;
1204
1205                 if (mac_header_len != nla_len(nfqa[NFQA_L2HDR]))
1206                         return -EINVAL;
1207                 else if (mac_header_len > 0)
1208                         memcpy(skb_mac_header(entry->skb),
1209                                nla_data(nfqa[NFQA_L2HDR]),
1210                                mac_header_len);
1211         }
1212
1213         return 0;
1214 }
1215
1216 static int nfqnl_recv_verdict(struct sk_buff *skb, const struct nfnl_info *info,
1217                               const struct nlattr * const nfqa[])
1218 {
1219         struct nfnl_queue_net *q = nfnl_queue_pernet(info->net);
1220         u_int16_t queue_num = ntohs(info->nfmsg->res_id);
1221         const struct nfnl_ct_hook *nfnl_ct;
1222         struct nfqnl_msg_verdict_hdr *vhdr;
1223         enum ip_conntrack_info ctinfo;
1224         struct nfqnl_instance *queue;
1225         struct nf_queue_entry *entry;
1226         struct nf_conn *ct = NULL;
1227         unsigned int verdict;
1228         int err;
1229
1230         queue = verdict_instance_lookup(q, queue_num,
1231                                         NETLINK_CB(skb).portid);
1232         if (IS_ERR(queue))
1233                 return PTR_ERR(queue);
1234
1235         vhdr = verdicthdr_get(nfqa);
1236         if (!vhdr)
1237                 return -EINVAL;
1238
1239         verdict = ntohl(vhdr->verdict);
1240
1241         entry = find_dequeue_entry(queue, ntohl(vhdr->id));
1242         if (entry == NULL)
1243                 return -ENOENT;
1244
1245         /* rcu lock already held from nfnl->call_rcu. */
1246         nfnl_ct = rcu_dereference(nfnl_ct_hook);
1247
1248         if (nfqa[NFQA_CT]) {
1249                 if (nfnl_ct != NULL)
1250                         ct = nfqnl_ct_parse(nfnl_ct, info->nlh, nfqa, entry,
1251                                             &ctinfo);
1252         }
1253
1254         if (entry->state.pf == PF_BRIDGE) {
1255                 err = nfqa_parse_bridge(entry, nfqa);
1256                 if (err < 0)
1257                         return err;
1258         }
1259
1260         if (nfqa[NFQA_PAYLOAD]) {
1261                 u16 payload_len = nla_len(nfqa[NFQA_PAYLOAD]);
1262                 int diff = payload_len - entry->skb->len;
1263
1264                 if (nfqnl_mangle(nla_data(nfqa[NFQA_PAYLOAD]),
1265                                  payload_len, entry, diff) < 0)
1266                         verdict = NF_DROP;
1267
1268                 if (ct && diff)
1269                         nfnl_ct->seq_adjust(entry->skb, ct, ctinfo, diff);
1270         }
1271
1272         if (nfqa[NFQA_MARK])
1273                 entry->skb->mark = ntohl(nla_get_be32(nfqa[NFQA_MARK]));
1274
1275         if (nfqa[NFQA_PRIORITY])
1276                 entry->skb->priority = ntohl(nla_get_be32(nfqa[NFQA_PRIORITY]));
1277
1278         nfqnl_reinject(entry, verdict);
1279         return 0;
1280 }
1281
1282 static int nfqnl_recv_unsupp(struct sk_buff *skb, const struct nfnl_info *info,
1283                              const struct nlattr * const cda[])
1284 {
1285         return -ENOTSUPP;
1286 }
1287
1288 static const struct nla_policy nfqa_cfg_policy[NFQA_CFG_MAX+1] = {
1289         [NFQA_CFG_CMD]          = { .len = sizeof(struct nfqnl_msg_config_cmd) },
1290         [NFQA_CFG_PARAMS]       = { .len = sizeof(struct nfqnl_msg_config_params) },
1291         [NFQA_CFG_QUEUE_MAXLEN] = { .type = NLA_U32 },
1292         [NFQA_CFG_MASK]         = { .type = NLA_U32 },
1293         [NFQA_CFG_FLAGS]        = { .type = NLA_U32 },
1294 };
1295
1296 static const struct nf_queue_handler nfqh = {
1297         .outfn          = nfqnl_enqueue_packet,
1298         .nf_hook_drop   = nfqnl_nf_hook_drop,
1299 };
1300
1301 static int nfqnl_recv_config(struct sk_buff *skb, const struct nfnl_info *info,
1302                              const struct nlattr * const nfqa[])
1303 {
1304         struct nfnl_queue_net *q = nfnl_queue_pernet(info->net);
1305         u_int16_t queue_num = ntohs(info->nfmsg->res_id);
1306         struct nfqnl_msg_config_cmd *cmd = NULL;
1307         struct nfqnl_instance *queue;
1308         __u32 flags = 0, mask = 0;
1309         int ret = 0;
1310
1311         if (nfqa[NFQA_CFG_CMD]) {
1312                 cmd = nla_data(nfqa[NFQA_CFG_CMD]);
1313
1314                 /* Obsolete commands without queue context */
1315                 switch (cmd->command) {
1316                 case NFQNL_CFG_CMD_PF_BIND: return 0;
1317                 case NFQNL_CFG_CMD_PF_UNBIND: return 0;
1318                 }
1319         }
1320
1321         /* Check if we support these flags in first place, dependencies should
1322          * be there too not to break atomicity.
1323          */
1324         if (nfqa[NFQA_CFG_FLAGS]) {
1325                 if (!nfqa[NFQA_CFG_MASK]) {
1326                         /* A mask is needed to specify which flags are being
1327                          * changed.
1328                          */
1329                         return -EINVAL;
1330                 }
1331
1332                 flags = ntohl(nla_get_be32(nfqa[NFQA_CFG_FLAGS]));
1333                 mask = ntohl(nla_get_be32(nfqa[NFQA_CFG_MASK]));
1334
1335                 if (flags >= NFQA_CFG_F_MAX)
1336                         return -EOPNOTSUPP;
1337
1338 #if !IS_ENABLED(CONFIG_NETWORK_SECMARK)
1339                 if (flags & mask & NFQA_CFG_F_SECCTX)
1340                         return -EOPNOTSUPP;
1341 #endif
1342                 if ((flags & mask & NFQA_CFG_F_CONNTRACK) &&
1343                     !rcu_access_pointer(nfnl_ct_hook)) {
1344 #ifdef CONFIG_MODULES
1345                         nfnl_unlock(NFNL_SUBSYS_QUEUE);
1346                         request_module("ip_conntrack_netlink");
1347                         nfnl_lock(NFNL_SUBSYS_QUEUE);
1348                         if (rcu_access_pointer(nfnl_ct_hook))
1349                                 return -EAGAIN;
1350 #endif
1351                         return -EOPNOTSUPP;
1352                 }
1353         }
1354
1355         rcu_read_lock();
1356         queue = instance_lookup(q, queue_num);
1357         if (queue && queue->peer_portid != NETLINK_CB(skb).portid) {
1358                 ret = -EPERM;
1359                 goto err_out_unlock;
1360         }
1361
1362         if (cmd != NULL) {
1363                 switch (cmd->command) {
1364                 case NFQNL_CFG_CMD_BIND:
1365                         if (queue) {
1366                                 ret = -EBUSY;
1367                                 goto err_out_unlock;
1368                         }
1369                         queue = instance_create(q, queue_num,
1370                                                 NETLINK_CB(skb).portid);
1371                         if (IS_ERR(queue)) {
1372                                 ret = PTR_ERR(queue);
1373                                 goto err_out_unlock;
1374                         }
1375                         break;
1376                 case NFQNL_CFG_CMD_UNBIND:
1377                         if (!queue) {
1378                                 ret = -ENODEV;
1379                                 goto err_out_unlock;
1380                         }
1381                         instance_destroy(q, queue);
1382                         goto err_out_unlock;
1383                 case NFQNL_CFG_CMD_PF_BIND:
1384                 case NFQNL_CFG_CMD_PF_UNBIND:
1385                         break;
1386                 default:
1387                         ret = -ENOTSUPP;
1388                         goto err_out_unlock;
1389                 }
1390         }
1391
1392         if (!queue) {
1393                 ret = -ENODEV;
1394                 goto err_out_unlock;
1395         }
1396
1397         if (nfqa[NFQA_CFG_PARAMS]) {
1398                 struct nfqnl_msg_config_params *params =
1399                         nla_data(nfqa[NFQA_CFG_PARAMS]);
1400
1401                 nfqnl_set_mode(queue, params->copy_mode,
1402                                 ntohl(params->copy_range));
1403         }
1404
1405         if (nfqa[NFQA_CFG_QUEUE_MAXLEN]) {
1406                 __be32 *queue_maxlen = nla_data(nfqa[NFQA_CFG_QUEUE_MAXLEN]);
1407
1408                 spin_lock_bh(&queue->lock);
1409                 queue->queue_maxlen = ntohl(*queue_maxlen);
1410                 spin_unlock_bh(&queue->lock);
1411         }
1412
1413         if (nfqa[NFQA_CFG_FLAGS]) {
1414                 spin_lock_bh(&queue->lock);
1415                 queue->flags &= ~mask;
1416                 queue->flags |= flags & mask;
1417                 spin_unlock_bh(&queue->lock);
1418         }
1419
1420 err_out_unlock:
1421         rcu_read_unlock();
1422         return ret;
1423 }
1424
1425 static const struct nfnl_callback nfqnl_cb[NFQNL_MSG_MAX] = {
1426         [NFQNL_MSG_PACKET]      = {
1427                 .call           = nfqnl_recv_unsupp,
1428                 .type           = NFNL_CB_RCU,
1429                 .attr_count     = NFQA_MAX,
1430         },
1431         [NFQNL_MSG_VERDICT]     = {
1432                 .call           = nfqnl_recv_verdict,
1433                 .type           = NFNL_CB_RCU,
1434                 .attr_count     = NFQA_MAX,
1435                 .policy         = nfqa_verdict_policy
1436         },
1437         [NFQNL_MSG_CONFIG]      = {
1438                 .call           = nfqnl_recv_config,
1439                 .type           = NFNL_CB_MUTEX,
1440                 .attr_count     = NFQA_CFG_MAX,
1441                 .policy         = nfqa_cfg_policy
1442         },
1443         [NFQNL_MSG_VERDICT_BATCH] = {
1444                 .call           = nfqnl_recv_verdict_batch,
1445                 .type           = NFNL_CB_RCU,
1446                 .attr_count     = NFQA_MAX,
1447                 .policy         = nfqa_verdict_batch_policy
1448         },
1449 };
1450
1451 static const struct nfnetlink_subsystem nfqnl_subsys = {
1452         .name           = "nf_queue",
1453         .subsys_id      = NFNL_SUBSYS_QUEUE,
1454         .cb_count       = NFQNL_MSG_MAX,
1455         .cb             = nfqnl_cb,
1456 };
1457
1458 #ifdef CONFIG_PROC_FS
1459 struct iter_state {
1460         struct seq_net_private p;
1461         unsigned int bucket;
1462 };
1463
1464 static struct hlist_node *get_first(struct seq_file *seq)
1465 {
1466         struct iter_state *st = seq->private;
1467         struct net *net;
1468         struct nfnl_queue_net *q;
1469
1470         if (!st)
1471                 return NULL;
1472
1473         net = seq_file_net(seq);
1474         q = nfnl_queue_pernet(net);
1475         for (st->bucket = 0; st->bucket < INSTANCE_BUCKETS; st->bucket++) {
1476                 if (!hlist_empty(&q->instance_table[st->bucket]))
1477                         return q->instance_table[st->bucket].first;
1478         }
1479         return NULL;
1480 }
1481
1482 static struct hlist_node *get_next(struct seq_file *seq, struct hlist_node *h)
1483 {
1484         struct iter_state *st = seq->private;
1485         struct net *net = seq_file_net(seq);
1486
1487         h = h->next;
1488         while (!h) {
1489                 struct nfnl_queue_net *q;
1490
1491                 if (++st->bucket >= INSTANCE_BUCKETS)
1492                         return NULL;
1493
1494                 q = nfnl_queue_pernet(net);
1495                 h = q->instance_table[st->bucket].first;
1496         }
1497         return h;
1498 }
1499
1500 static struct hlist_node *get_idx(struct seq_file *seq, loff_t pos)
1501 {
1502         struct hlist_node *head;
1503         head = get_first(seq);
1504
1505         if (head)
1506                 while (pos && (head = get_next(seq, head)))
1507                         pos--;
1508         return pos ? NULL : head;
1509 }
1510
1511 static void *seq_start(struct seq_file *s, loff_t *pos)
1512         __acquires(nfnl_queue_pernet(seq_file_net(s))->instances_lock)
1513 {
1514         spin_lock(&nfnl_queue_pernet(seq_file_net(s))->instances_lock);
1515         return get_idx(s, *pos);
1516 }
1517
1518 static void *seq_next(struct seq_file *s, void *v, loff_t *pos)
1519 {
1520         (*pos)++;
1521         return get_next(s, v);
1522 }
1523
1524 static void seq_stop(struct seq_file *s, void *v)
1525         __releases(nfnl_queue_pernet(seq_file_net(s))->instances_lock)
1526 {
1527         spin_unlock(&nfnl_queue_pernet(seq_file_net(s))->instances_lock);
1528 }
1529
1530 static int seq_show(struct seq_file *s, void *v)
1531 {
1532         const struct nfqnl_instance *inst = v;
1533
1534         seq_printf(s, "%5u %6u %5u %1u %5u %5u %5u %8u %2d\n",
1535                    inst->queue_num,
1536                    inst->peer_portid, inst->queue_total,
1537                    inst->copy_mode, inst->copy_range,
1538                    inst->queue_dropped, inst->queue_user_dropped,
1539                    inst->id_sequence, 1);
1540         return 0;
1541 }
1542
1543 static const struct seq_operations nfqnl_seq_ops = {
1544         .start  = seq_start,
1545         .next   = seq_next,
1546         .stop   = seq_stop,
1547         .show   = seq_show,
1548 };
1549 #endif /* PROC_FS */
1550
1551 static int __net_init nfnl_queue_net_init(struct net *net)
1552 {
1553         unsigned int i;
1554         struct nfnl_queue_net *q = nfnl_queue_pernet(net);
1555
1556         for (i = 0; i < INSTANCE_BUCKETS; i++)
1557                 INIT_HLIST_HEAD(&q->instance_table[i]);
1558
1559         spin_lock_init(&q->instances_lock);
1560
1561 #ifdef CONFIG_PROC_FS
1562         if (!proc_create_net("nfnetlink_queue", 0440, net->nf.proc_netfilter,
1563                         &nfqnl_seq_ops, sizeof(struct iter_state)))
1564                 return -ENOMEM;
1565 #endif
1566         return 0;
1567 }
1568
1569 static void __net_exit nfnl_queue_net_exit(struct net *net)
1570 {
1571         struct nfnl_queue_net *q = nfnl_queue_pernet(net);
1572         unsigned int i;
1573
1574 #ifdef CONFIG_PROC_FS
1575         remove_proc_entry("nfnetlink_queue", net->nf.proc_netfilter);
1576 #endif
1577         for (i = 0; i < INSTANCE_BUCKETS; i++)
1578                 WARN_ON_ONCE(!hlist_empty(&q->instance_table[i]));
1579 }
1580
1581 static struct pernet_operations nfnl_queue_net_ops = {
1582         .init           = nfnl_queue_net_init,
1583         .exit           = nfnl_queue_net_exit,
1584         .id             = &nfnl_queue_net_id,
1585         .size           = sizeof(struct nfnl_queue_net),
1586 };
1587
1588 static int __init nfnetlink_queue_init(void)
1589 {
1590         int status;
1591
1592         status = register_pernet_subsys(&nfnl_queue_net_ops);
1593         if (status < 0) {
1594                 pr_err("failed to register pernet ops\n");
1595                 goto out;
1596         }
1597
1598         netlink_register_notifier(&nfqnl_rtnl_notifier);
1599         status = nfnetlink_subsys_register(&nfqnl_subsys);
1600         if (status < 0) {
1601                 pr_err("failed to create netlink socket\n");
1602                 goto cleanup_netlink_notifier;
1603         }
1604
1605         status = register_netdevice_notifier(&nfqnl_dev_notifier);
1606         if (status < 0) {
1607                 pr_err("failed to register netdevice notifier\n");
1608                 goto cleanup_netlink_subsys;
1609         }
1610
1611         nf_register_queue_handler(&nfqh);
1612
1613         return status;
1614
1615 cleanup_netlink_subsys:
1616         nfnetlink_subsys_unregister(&nfqnl_subsys);
1617 cleanup_netlink_notifier:
1618         netlink_unregister_notifier(&nfqnl_rtnl_notifier);
1619         unregister_pernet_subsys(&nfnl_queue_net_ops);
1620 out:
1621         return status;
1622 }
1623
1624 static void __exit nfnetlink_queue_fini(void)
1625 {
1626         nf_unregister_queue_handler();
1627         unregister_netdevice_notifier(&nfqnl_dev_notifier);
1628         nfnetlink_subsys_unregister(&nfqnl_subsys);
1629         netlink_unregister_notifier(&nfqnl_rtnl_notifier);
1630         unregister_pernet_subsys(&nfnl_queue_net_ops);
1631
1632         rcu_barrier(); /* Wait for completion of call_rcu()'s */
1633 }
1634
1635 MODULE_DESCRIPTION("netfilter packet queue handler");
1636 MODULE_AUTHOR("Harald Welte <laforge@netfilter.org>");
1637 MODULE_LICENSE("GPL");
1638 MODULE_ALIAS_NFNL_SUBSYS(NFNL_SUBSYS_QUEUE);
1639
1640 module_init(nfnetlink_queue_init);
1641 module_exit(nfnetlink_queue_fini);