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