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