net/netfilter/nfnetlink_queue.c

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