net/netfilter/nfnetlink_queue.c

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