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         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;
 391         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             entskb->mac_header != entskb->network_header) {
 564                 struct nfqnl_msg_packet_hw phw;
 565                 int len;
 566
 567                 memset(&phw, 0, sizeof(phw));
 568                 len = dev_parse_header(entskb, phw.hw_addr);
 569                 if (len) {
 570                         phw.hw_addrlen = htons(len);
 571                         if (nla_put(skb, NFQA_HWADDR, sizeof(phw), &phw))
 572                                 goto nla_put_failure;
 573                 }
 574         }
 575
 576         if (nfqnl_put_bridge(entry, skb) < 0)
 577                 goto nla_put_failure;
 578
 579         if (entry->state.hook <= NF_INET_FORWARD && entskb->tstamp) {
 580                 struct nfqnl_msg_packet_timestamp ts;
 581                 struct timespec64 kts = ktime_to_timespec64(entskb->tstamp);
 582
 583                 ts.sec = cpu_to_be64(kts.tv_sec);
 584                 ts.usec = cpu_to_be64(kts.tv_nsec / NSEC_PER_USEC);
 585
 586                 if (nla_put(skb, NFQA_TIMESTAMP, sizeof(ts), &ts))
 587                         goto nla_put_failure;
 588         }
 589
 590         if ((queue->flags & NFQA_CFG_F_UID_GID) && entskb->sk &&
 591             nfqnl_put_sk_uidgid(skb, entskb->sk) < 0)
 592                 goto nla_put_failure;
 593
 594         if (seclen && nla_put(skb, NFQA_SECCTX, seclen, secdata))
 595                 goto nla_put_failure;
 596
 597         if (ct && nfnl_ct->build(skb, ct, ctinfo, NFQA_CT, NFQA_CT_INFO) < 0)
 598                 goto nla_put_failure;
 599
 600         if (cap_len > data_len &&
 601             nla_put_be32(skb, NFQA_CAP_LEN, htonl(cap_len)))
 602                 goto nla_put_failure;
 603
 604         if (nfqnl_put_packet_info(skb, entskb, csum_verify))
 605                 goto nla_put_failure;
 606
 607         if (data_len) {
 608                 struct nlattr *nla;
 609
 610                 if (skb_tailroom(skb) < sizeof(*nla) + hlen)
 611                         goto nla_put_failure;
 612
 613                 nla = skb_put(skb, sizeof(*nla));
 614                 nla->nla_type = NFQA_PAYLOAD;
 615                 nla->nla_len = nla_attr_size(data_len);
 616
 617                 if (skb_zerocopy(skb, entskb, data_len, hlen))
 618                         goto nla_put_failure;
 619         }
 620
 621         nlh->nlmsg_len = skb->len;
 622         if (seclen)
 623                 security_release_secctx(secdata, seclen);
 624         return skb;
 625
 626 nla_put_failure:
 627         skb_tx_error(entskb);
 628         kfree_skb(skb);
 629         net_err_ratelimited("nf_queue: error creating packet message\n");
 630 nlmsg_failure:
 631         if (seclen)
 632                 security_release_secctx(secdata, seclen);
 633         return NULL;
 634 }
 635
 636 static bool nf_ct_drop_unconfirmed(const struct nf_queue_entry *entry)
 637 {
 638 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
 639         static const unsigned long flags = IPS_CONFIRMED | IPS_DYING;
 640         const struct nf_conn *ct = (void *)skb_nfct(entry->skb);
 641
 642         if (ct && ((ct->status & flags) == IPS_DYING))
 643                 return true;
 644 #endif
 645         return false;
 646 }
 647
 648 static int
 649 __nfqnl_enqueue_packet(struct net *net, struct nfqnl_instance *queue,
 650                         struct nf_queue_entry *entry)
 651 {
 652         struct sk_buff *nskb;
 653         int err = -ENOBUFS;
 654         __be32 *packet_id_ptr;
 655         int failopen = 0;
 656
 657         nskb = nfqnl_build_packet_message(net, queue, entry, &packet_id_ptr);
 658         if (nskb == NULL) {
 659                 err = -ENOMEM;
 660                 goto err_out;
 661         }
 662         spin_lock_bh(&queue->lock);
 663
 664         if (nf_ct_drop_unconfirmed(entry))
 665                 goto err_out_free_nskb;
 666
 667         if (queue->queue_total >= queue->queue_maxlen) {
 668                 if (queue->flags & NFQA_CFG_F_FAIL_OPEN) {
 669                         failopen = 1;
 670                         err = 0;
 671                 } else {
 672                         queue->queue_dropped++;
 673                         net_warn_ratelimited("nf_queue: full at %d entries, dropping packets(s)\n",
 674                                              queue->queue_total);
 675                 }
 676                 goto err_out_free_nskb;
 677         }
 678         entry->id = ++queue->id_sequence;
 679         *packet_id_ptr = htonl(entry->id);
 680
 681         /* nfnetlink_unicast will either free the nskb or add it to a socket */
 682         err = nfnetlink_unicast(nskb, net, queue->peer_portid);
 683         if (err < 0) {
 684                 if (queue->flags & NFQA_CFG_F_FAIL_OPEN) {
 685                         failopen = 1;
 686                         err = 0;
 687                 } else {
 688                         queue->queue_user_dropped++;
 689                 }
 690                 goto err_out_unlock;
 691         }
 692
 693         __enqueue_entry(queue, entry);
 694
 695         spin_unlock_bh(&queue->lock);
 696         return 0;
 697
 698 err_out_free_nskb:
 699         kfree_skb(nskb);
 700 err_out_unlock:
 701         spin_unlock_bh(&queue->lock);
 702         if (failopen)
 703                 nfqnl_reinject(entry, NF_ACCEPT);
 704 err_out:
 705         return err;
 706 }
 707
 708 static struct nf_queue_entry *
 709 nf_queue_entry_dup(struct nf_queue_entry *e)
 710 {
 711         struct nf_queue_entry *entry = kmemdup(e, e->size, GFP_ATOMIC);
 712         if (entry)
 713                 nf_queue_entry_get_refs(entry);
 714         return entry;
 715 }
 716
 717 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
 718 /* When called from bridge netfilter, skb->data must point to MAC header
 719  * before calling skb_gso_segment(). Else, original MAC header is lost
 720  * and segmented skbs will be sent to wrong destination.
 721  */
 722 static void nf_bridge_adjust_skb_data(struct sk_buff *skb)
 723 {
 724         if (nf_bridge_info_get(skb))
 725                 __skb_push(skb, skb->network_header - skb->mac_header);
 726 }
 727
 728 static void nf_bridge_adjust_segmented_data(struct sk_buff *skb)
 729 {
 730         if (nf_bridge_info_get(skb))
 731                 __skb_pull(skb, skb->network_header - skb->mac_header);
 732 }
 733 #else
 734 #define nf_bridge_adjust_skb_data(s) do {} while (0)
 735 #define nf_bridge_adjust_segmented_data(s) do {} while (0)
 736 #endif
 737
 738 static int
 739 __nfqnl_enqueue_packet_gso(struct net *net, struct nfqnl_instance *queue,
 740                            struct sk_buff *skb, struct nf_queue_entry *entry)
 741 {
 742         int ret = -ENOMEM;
 743         struct nf_queue_entry *entry_seg;
 744
 745         nf_bridge_adjust_segmented_data(skb);
 746
 747         if (skb->next == NULL) { /* last packet, no need to copy entry */
 748                 struct sk_buff *gso_skb = entry->skb;
 749                 entry->skb = skb;
 750                 ret = __nfqnl_enqueue_packet(net, queue, entry);
 751                 if (ret)
 752                         entry->skb = gso_skb;
 753                 return ret;
 754         }
 755
 756         skb_mark_not_on_list(skb);
 757
 758         entry_seg = nf_queue_entry_dup(entry);
 759         if (entry_seg) {
 760                 entry_seg->skb = skb;
 761                 ret = __nfqnl_enqueue_packet(net, queue, entry_seg);
 762                 if (ret)
 763                         nf_queue_entry_free(entry_seg);
 764         }
 765         return ret;
 766 }
 767
 768 static int
 769 nfqnl_enqueue_packet(struct nf_queue_entry *entry, unsigned int queuenum)
 770 {
 771         unsigned int queued;
 772         struct nfqnl_instance *queue;
 773         struct sk_buff *skb, *segs, *nskb;
 774         int err = -ENOBUFS;
 775         struct net *net = entry->state.net;
 776         struct nfnl_queue_net *q = nfnl_queue_pernet(net);
 777
 778         /* rcu_read_lock()ed by nf_hook_thresh */
 779         queue = instance_lookup(q, queuenum);
 780         if (!queue)
 781                 return -ESRCH;
 782
 783         if (queue->copy_mode == NFQNL_COPY_NONE)
 784                 return -EINVAL;
 785
 786         skb = entry->skb;
 787
 788         switch (entry->state.pf) {
 789         case NFPROTO_IPV4:
 790                 skb->protocol = htons(ETH_P_IP);
 791                 break;
 792         case NFPROTO_IPV6:
 793                 skb->protocol = htons(ETH_P_IPV6);
 794                 break;
 795         }
 796
 797         if ((queue->flags & NFQA_CFG_F_GSO) || !skb_is_gso(skb))
 798                 return __nfqnl_enqueue_packet(net, queue, entry);
 799
 800         nf_bridge_adjust_skb_data(skb);
 801         segs = skb_gso_segment(skb, 0);
 802         /* Does not use PTR_ERR to limit the number of error codes that can be
 803          * returned by nf_queue.  For instance, callers rely on -ESRCH to
 804          * mean 'ignore this hook'.
 805          */
 806         if (IS_ERR_OR_NULL(segs))
 807                 goto out_err;
 808         queued = 0;
 809         err = 0;
 810         skb_list_walk_safe(segs, segs, nskb) {
 811                 if (err == 0)
 812                         err = __nfqnl_enqueue_packet_gso(net, queue,
 813                                                         segs, entry);
 814                 if (err == 0)
 815                         queued++;
 816                 else
 817                         kfree_skb(segs);
 818         }
 819
 820         if (queued) {
 821                 if (err) /* some segments are already queued */
 822                         nf_queue_entry_free(entry);
 823                 kfree_skb(skb);
 824                 return 0;
 825         }
 826  out_err:
 827         nf_bridge_adjust_segmented_data(skb);
 828         return err;
 829 }
 830
 831 static int
 832 nfqnl_mangle(void *data, int data_len, struct nf_queue_entry *e, int diff)
 833 {
 834         struct sk_buff *nskb;
 835
 836         if (diff < 0) {
 837                 if (pskb_trim(e->skb, data_len))
 838                         return -ENOMEM;
 839         } else if (diff > 0) {
 840                 if (data_len > 0xFFFF)
 841                         return -EINVAL;
 842                 if (diff > skb_tailroom(e->skb)) {
 843                         nskb = skb_copy_expand(e->skb, skb_headroom(e->skb),
 844                                                diff, GFP_ATOMIC);
 845                         if (!nskb)
 846                                 return -ENOMEM;
 847                         kfree_skb(e->skb);
 848                         e->skb = nskb;
 849                 }
 850                 skb_put(e->skb, diff);
 851         }
 852         if (skb_ensure_writable(e->skb, data_len))
 853                 return -ENOMEM;
 854         skb_copy_to_linear_data(e->skb, data, data_len);
 855         e->skb->ip_summed = CHECKSUM_NONE;
 856         return 0;
 857 }
 858
 859 static int
 860 nfqnl_set_mode(struct nfqnl_instance *queue,
 861                unsigned char mode, unsigned int range)
 862 {
 863         int status = 0;
 864
 865         spin_lock_bh(&queue->lock);
 866         switch (mode) {
 867         case NFQNL_COPY_NONE:
 868         case NFQNL_COPY_META:
 869                 queue->copy_mode = mode;
 870                 queue->copy_range = 0;
 871                 break;
 872
 873         case NFQNL_COPY_PACKET:
 874                 queue->copy_mode = mode;
 875                 if (range == 0 || range > NFQNL_MAX_COPY_RANGE)
 876                         queue->copy_range = NFQNL_MAX_COPY_RANGE;
 877                 else
 878                         queue->copy_range = range;
 879                 break;
 880
 881         default:
 882                 status = -EINVAL;
 883
 884         }
 885         spin_unlock_bh(&queue->lock);
 886
 887         return status;
 888 }
 889
 890 static int
 891 dev_cmp(struct nf_queue_entry *entry, unsigned long ifindex)
 892 {
 893 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
 894         int physinif, physoutif;
 895
 896         physinif = nf_bridge_get_physinif(entry->skb);
 897         physoutif = nf_bridge_get_physoutif(entry->skb);
 898
 899         if (physinif == ifindex || physoutif == ifindex)
 900                 return 1;
 901 #endif
 902         if (entry->state.in)
 903                 if (entry->state.in->ifindex == ifindex)
 904                         return 1;
 905         if (entry->state.out)
 906                 if (entry->state.out->ifindex == ifindex)
 907                         return 1;
 908
 909         return 0;
 910 }
 911
 912 /* drop all packets with either indev or outdev == ifindex from all queue
 913  * instances */
 914 static void
 915 nfqnl_dev_drop(struct net *net, int ifindex)
 916 {
 917         int i;
 918         struct nfnl_queue_net *q = nfnl_queue_pernet(net);
 919
 920         rcu_read_lock();
 921
 922         for (i = 0; i < INSTANCE_BUCKETS; i++) {
 923                 struct nfqnl_instance *inst;
 924                 struct hlist_head *head = &q->instance_table[i];
 925
 926                 hlist_for_each_entry_rcu(inst, head, hlist)
 927                         nfqnl_flush(inst, dev_cmp, ifindex);
 928         }
 929
 930         rcu_read_unlock();
 931 }
 932
 933 static int
 934 nfqnl_rcv_dev_event(struct notifier_block *this,
 935                     unsigned long event, void *ptr)
 936 {
 937         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
 938
 939         /* Drop any packets associated with the downed device */
 940         if (event == NETDEV_DOWN)
 941                 nfqnl_dev_drop(dev_net(dev), dev->ifindex);
 942         return NOTIFY_DONE;
 943 }
 944
 945 static struct notifier_block nfqnl_dev_notifier = {
 946         .notifier_call  = nfqnl_rcv_dev_event,
 947 };
 948
 949 static void nfqnl_nf_hook_drop(struct net *net)
 950 {
 951         struct nfnl_queue_net *q = nfnl_queue_pernet(net);
 952         int i;
 953
 954         /* This function is also called on net namespace error unwind,
 955          * when pernet_ops->init() failed and ->exit() functions of the
 956          * previous pernet_ops gets called.
 957          *
 958          * This may result in a call to nfqnl_nf_hook_drop() before
 959          * struct nfnl_queue_net was allocated.
 960          */
 961         if (!q)
 962                 return;
 963
 964         for (i = 0; i < INSTANCE_BUCKETS; i++) {
 965                 struct nfqnl_instance *inst;
 966                 struct hlist_head *head = &q->instance_table[i];
 967
 968                 hlist_for_each_entry_rcu(inst, head, hlist)
 969                         nfqnl_flush(inst, NULL, 0);
 970         }
 971 }
 972
 973 static int
 974 nfqnl_rcv_nl_event(struct notifier_block *this,
 975                    unsigned long event, void *ptr)
 976 {
 977         struct netlink_notify *n = ptr;
 978         struct nfnl_queue_net *q = nfnl_queue_pernet(n->net);
 979
 980         if (event == NETLINK_URELEASE && n->protocol == NETLINK_NETFILTER) {
 981                 int i;
 982
 983                 /* destroy all instances for this portid */
 984                 spin_lock(&q->instances_lock);
 985                 for (i = 0; i < INSTANCE_BUCKETS; i++) {
 986                         struct hlist_node *t2;
 987                         struct nfqnl_instance *inst;
 988                         struct hlist_head *head = &q->instance_table[i];
 989
 990                         hlist_for_each_entry_safe(inst, t2, head, hlist) {
 991                                 if (n->portid == inst->peer_portid)
 992                                         __instance_destroy(inst);
 993                         }
 994                 }
 995                 spin_unlock(&q->instances_lock);
 996         }
 997         return NOTIFY_DONE;
 998 }
 999
1000 static struct notifier_block nfqnl_rtnl_notifier = {
1001         .notifier_call  = nfqnl_rcv_nl_event,
1002 };
1003
1004 static const struct nla_policy nfqa_vlan_policy[NFQA_VLAN_MAX + 1] = {
1005         [NFQA_VLAN_TCI]         = { .type = NLA_U16},
1006         [NFQA_VLAN_PROTO]       = { .type = NLA_U16},
1007 };
1008
1009 static const struct nla_policy nfqa_verdict_policy[NFQA_MAX+1] = {
1010         [NFQA_VERDICT_HDR]      = { .len = sizeof(struct nfqnl_msg_verdict_hdr) },
1011         [NFQA_MARK]             = { .type = NLA_U32 },
1012         [NFQA_PAYLOAD]          = { .type = NLA_UNSPEC },
1013         [NFQA_CT]               = { .type = NLA_UNSPEC },
1014         [NFQA_EXP]              = { .type = NLA_UNSPEC },
1015         [NFQA_VLAN]             = { .type = NLA_NESTED },
1016 };
1017
1018 static const struct nla_policy nfqa_verdict_batch_policy[NFQA_MAX+1] = {
1019         [NFQA_VERDICT_HDR]      = { .len = sizeof(struct nfqnl_msg_verdict_hdr) },
1020         [NFQA_MARK]             = { .type = NLA_U32 },
1021 };
1022
1023 static struct nfqnl_instance *
1024 verdict_instance_lookup(struct nfnl_queue_net *q, u16 queue_num, u32 nlportid)
1025 {
1026         struct nfqnl_instance *queue;
1027
1028         queue = instance_lookup(q, queue_num);
1029         if (!queue)
1030                 return ERR_PTR(-ENODEV);
1031
1032         if (queue->peer_portid != nlportid)
1033                 return ERR_PTR(-EPERM);
1034
1035         return queue;
1036 }
1037
1038 static struct nfqnl_msg_verdict_hdr*
1039 verdicthdr_get(const struct nlattr * const nfqa[])
1040 {
1041         struct nfqnl_msg_verdict_hdr *vhdr;
1042         unsigned int verdict;
1043
1044         if (!nfqa[NFQA_VERDICT_HDR])
1045                 return NULL;
1046
1047         vhdr = nla_data(nfqa[NFQA_VERDICT_HDR]);
1048         verdict = ntohl(vhdr->verdict) & NF_VERDICT_MASK;
1049         if (verdict > NF_MAX_VERDICT || verdict == NF_STOLEN)
1050                 return NULL;
1051         return vhdr;
1052 }
1053
1054 static int nfq_id_after(unsigned int id, unsigned int max)
1055 {
1056         return (int)(id - max) > 0;
1057 }
1058
1059 static int nfqnl_recv_verdict_batch(struct sk_buff *skb,
1060                                     const struct nfnl_info *info,
1061                                     const struct nlattr * const nfqa[])
1062 {
1063         struct nfnl_queue_net *q = nfnl_queue_pernet(info->net);
1064         u16 queue_num = ntohs(info->nfmsg->res_id);
1065         struct nf_queue_entry *entry, *tmp;
1066         struct nfqnl_msg_verdict_hdr *vhdr;
1067         struct nfqnl_instance *queue;
1068         unsigned int verdict, maxid;
1069         LIST_HEAD(batch_list);
1070
1071         queue = verdict_instance_lookup(q, queue_num,
1072                                         NETLINK_CB(skb).portid);
1073         if (IS_ERR(queue))
1074                 return PTR_ERR(queue);
1075
1076         vhdr = verdicthdr_get(nfqa);
1077         if (!vhdr)
1078                 return -EINVAL;
1079
1080         verdict = ntohl(vhdr->verdict);
1081         maxid = ntohl(vhdr->id);
1082
1083         spin_lock_bh(&queue->lock);
1084
1085         list_for_each_entry_safe(entry, tmp, &queue->queue_list, list) {
1086                 if (nfq_id_after(entry->id, maxid))
1087                         break;
1088                 __dequeue_entry(queue, entry);
1089                 list_add_tail(&entry->list, &batch_list);
1090         }
1091
1092         spin_unlock_bh(&queue->lock);
1093
1094         if (list_empty(&batch_list))
1095                 return -ENOENT;
1096
1097         list_for_each_entry_safe(entry, tmp, &batch_list, list) {
1098                 if (nfqa[NFQA_MARK])
1099                         entry->skb->mark = ntohl(nla_get_be32(nfqa[NFQA_MARK]));
1100
1101                 nfqnl_reinject(entry, verdict);
1102         }
1103         return 0;
1104 }
1105
1106 static struct nf_conn *nfqnl_ct_parse(struct nfnl_ct_hook *nfnl_ct,
1107                                       const struct nlmsghdr *nlh,
1108                                       const struct nlattr * const nfqa[],
1109                                       struct nf_queue_entry *entry,
1110                                       enum ip_conntrack_info *ctinfo)
1111 {
1112 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
1113         struct nf_conn *ct;
1114
1115         ct = nf_ct_get(entry->skb, ctinfo);
1116         if (ct == NULL)
1117                 return NULL;
1118
1119         if (nfnl_ct->parse(nfqa[NFQA_CT], ct) < 0)
1120                 return NULL;
1121
1122         if (nfqa[NFQA_EXP])
1123                 nfnl_ct->attach_expect(nfqa[NFQA_EXP], ct,
1124                                       NETLINK_CB(entry->skb).portid,
1125                                       nlmsg_report(nlh));
1126         return ct;
1127 #else
1128         return NULL;
1129 #endif
1130 }
1131
1132 static int nfqa_parse_bridge(struct nf_queue_entry *entry,
1133                              const struct nlattr * const nfqa[])
1134 {
1135         if (nfqa[NFQA_VLAN]) {
1136                 struct nlattr *tb[NFQA_VLAN_MAX + 1];
1137                 int err;
1138
1139                 err = nla_parse_nested_deprecated(tb, NFQA_VLAN_MAX,
1140                                                   nfqa[NFQA_VLAN],
1141                                                   nfqa_vlan_policy, NULL);
1142                 if (err < 0)
1143                         return err;
1144
1145                 if (!tb[NFQA_VLAN_TCI] || !tb[NFQA_VLAN_PROTO])
1146                         return -EINVAL;
1147
1148                 __vlan_hwaccel_put_tag(entry->skb,
1149                         nla_get_be16(tb[NFQA_VLAN_PROTO]),
1150                         ntohs(nla_get_be16(tb[NFQA_VLAN_TCI])));
1151         }
1152
1153         if (nfqa[NFQA_L2HDR]) {
1154                 int mac_header_len = entry->skb->network_header -
1155                         entry->skb->mac_header;
1156
1157                 if (mac_header_len != nla_len(nfqa[NFQA_L2HDR]))
1158                         return -EINVAL;
1159                 else if (mac_header_len > 0)
1160                         memcpy(skb_mac_header(entry->skb),
1161                                nla_data(nfqa[NFQA_L2HDR]),
1162                                mac_header_len);
1163         }
1164
1165         return 0;
1166 }
1167
1168 static int nfqnl_recv_verdict(struct sk_buff *skb, const struct nfnl_info *info,
1169                               const struct nlattr * const nfqa[])
1170 {
1171         struct nfnl_queue_net *q = nfnl_queue_pernet(info->net);
1172         u_int16_t queue_num = ntohs(info->nfmsg->res_id);
1173         struct nfqnl_msg_verdict_hdr *vhdr;
1174         enum ip_conntrack_info ctinfo;
1175         struct nfqnl_instance *queue;
1176         struct nf_queue_entry *entry;
1177         struct nfnl_ct_hook *nfnl_ct;
1178         struct nf_conn *ct = NULL;
1179         unsigned int verdict;
1180         int err;
1181
1182         queue = verdict_instance_lookup(q, queue_num,
1183                                         NETLINK_CB(skb).portid);
1184         if (IS_ERR(queue))
1185                 return PTR_ERR(queue);
1186
1187         vhdr = verdicthdr_get(nfqa);
1188         if (!vhdr)
1189                 return -EINVAL;
1190
1191         verdict = ntohl(vhdr->verdict);
1192
1193         entry = find_dequeue_entry(queue, ntohl(vhdr->id));
1194         if (entry == NULL)
1195                 return -ENOENT;
1196
1197         /* rcu lock already held from nfnl->call_rcu. */
1198         nfnl_ct = rcu_dereference(nfnl_ct_hook);
1199
1200         if (nfqa[NFQA_CT]) {
1201                 if (nfnl_ct != NULL)
1202                         ct = nfqnl_ct_parse(nfnl_ct, info->nlh, nfqa, entry,
1203                                             &ctinfo);
1204         }
1205
1206         if (entry->state.pf == PF_BRIDGE) {
1207                 err = nfqa_parse_bridge(entry, nfqa);
1208                 if (err < 0)
1209                         return err;
1210         }
1211
1212         if (nfqa[NFQA_PAYLOAD]) {
1213                 u16 payload_len = nla_len(nfqa[NFQA_PAYLOAD]);
1214                 int diff = payload_len - entry->skb->len;
1215
1216                 if (nfqnl_mangle(nla_data(nfqa[NFQA_PAYLOAD]),
1217                                  payload_len, entry, diff) < 0)
1218                         verdict = NF_DROP;
1219
1220                 if (ct && diff)
1221                         nfnl_ct->seq_adjust(entry->skb, ct, ctinfo, diff);
1222         }
1223
1224         if (nfqa[NFQA_MARK])
1225                 entry->skb->mark = ntohl(nla_get_be32(nfqa[NFQA_MARK]));
1226
1227         nfqnl_reinject(entry, verdict);
1228         return 0;
1229 }
1230
1231 static int nfqnl_recv_unsupp(struct sk_buff *skb, const struct nfnl_info *info,
1232                              const struct nlattr * const cda[])
1233 {
1234         return -ENOTSUPP;
1235 }
1236
1237 static const struct nla_policy nfqa_cfg_policy[NFQA_CFG_MAX+1] = {
1238         [NFQA_CFG_CMD]          = { .len = sizeof(struct nfqnl_msg_config_cmd) },
1239         [NFQA_CFG_PARAMS]       = { .len = sizeof(struct nfqnl_msg_config_params) },
1240         [NFQA_CFG_QUEUE_MAXLEN] = { .type = NLA_U32 },
1241         [NFQA_CFG_MASK]         = { .type = NLA_U32 },
1242         [NFQA_CFG_FLAGS]        = { .type = NLA_U32 },
1243 };
1244
1245 static const struct nf_queue_handler nfqh = {
1246         .outfn          = nfqnl_enqueue_packet,
1247         .nf_hook_drop   = nfqnl_nf_hook_drop,
1248 };
1249
1250 static int nfqnl_recv_config(struct sk_buff *skb, const struct nfnl_info *info,
1251                              const struct nlattr * const nfqa[])
1252 {
1253         struct nfnl_queue_net *q = nfnl_queue_pernet(info->net);
1254         u_int16_t queue_num = ntohs(info->nfmsg->res_id);
1255         struct nfqnl_msg_config_cmd *cmd = NULL;
1256         struct nfqnl_instance *queue;
1257         __u32 flags = 0, mask = 0;
1258         int ret = 0;
1259
1260         if (nfqa[NFQA_CFG_CMD]) {
1261                 cmd = nla_data(nfqa[NFQA_CFG_CMD]);
1262
1263                 /* Obsolete commands without queue context */
1264                 switch (cmd->command) {
1265                 case NFQNL_CFG_CMD_PF_BIND: return 0;
1266                 case NFQNL_CFG_CMD_PF_UNBIND: return 0;
1267                 }
1268         }
1269
1270         /* Check if we support these flags in first place, dependencies should
1271          * be there too not to break atomicity.
1272          */
1273         if (nfqa[NFQA_CFG_FLAGS]) {
1274                 if (!nfqa[NFQA_CFG_MASK]) {
1275                         /* A mask is needed to specify which flags are being
1276                          * changed.
1277                          */
1278                         return -EINVAL;
1279                 }
1280
1281                 flags = ntohl(nla_get_be32(nfqa[NFQA_CFG_FLAGS]));
1282                 mask = ntohl(nla_get_be32(nfqa[NFQA_CFG_MASK]));
1283
1284                 if (flags >= NFQA_CFG_F_MAX)
1285                         return -EOPNOTSUPP;
1286
1287 #if !IS_ENABLED(CONFIG_NETWORK_SECMARK)
1288                 if (flags & mask & NFQA_CFG_F_SECCTX)
1289                         return -EOPNOTSUPP;
1290 #endif
1291                 if ((flags & mask & NFQA_CFG_F_CONNTRACK) &&
1292                     !rcu_access_pointer(nfnl_ct_hook)) {
1293 #ifdef CONFIG_MODULES
1294                         nfnl_unlock(NFNL_SUBSYS_QUEUE);
1295                         request_module("ip_conntrack_netlink");
1296                         nfnl_lock(NFNL_SUBSYS_QUEUE);
1297                         if (rcu_access_pointer(nfnl_ct_hook))
1298                                 return -EAGAIN;
1299 #endif
1300                         return -EOPNOTSUPP;
1301                 }
1302         }
1303
1304         rcu_read_lock();
1305         queue = instance_lookup(q, queue_num);
1306         if (queue && queue->peer_portid != NETLINK_CB(skb).portid) {
1307                 ret = -EPERM;
1308                 goto err_out_unlock;
1309         }
1310
1311         if (cmd != NULL) {
1312                 switch (cmd->command) {
1313                 case NFQNL_CFG_CMD_BIND:
1314                         if (queue) {
1315                                 ret = -EBUSY;
1316                                 goto err_out_unlock;
1317                         }
1318                         queue = instance_create(q, queue_num,
1319                                                 NETLINK_CB(skb).portid);
1320                         if (IS_ERR(queue)) {
1321                                 ret = PTR_ERR(queue);
1322                                 goto err_out_unlock;
1323                         }
1324                         break;
1325                 case NFQNL_CFG_CMD_UNBIND:
1326                         if (!queue) {
1327                                 ret = -ENODEV;
1328                                 goto err_out_unlock;
1329                         }
1330                         instance_destroy(q, queue);
1331                         goto err_out_unlock;
1332                 case NFQNL_CFG_CMD_PF_BIND:
1333                 case NFQNL_CFG_CMD_PF_UNBIND:
1334                         break;
1335                 default:
1336                         ret = -ENOTSUPP;
1337                         goto err_out_unlock;
1338                 }
1339         }
1340
1341         if (!queue) {
1342                 ret = -ENODEV;
1343                 goto err_out_unlock;
1344         }
1345
1346         if (nfqa[NFQA_CFG_PARAMS]) {
1347                 struct nfqnl_msg_config_params *params =
1348                         nla_data(nfqa[NFQA_CFG_PARAMS]);
1349
1350                 nfqnl_set_mode(queue, params->copy_mode,
1351                                 ntohl(params->copy_range));
1352         }
1353
1354         if (nfqa[NFQA_CFG_QUEUE_MAXLEN]) {
1355                 __be32 *queue_maxlen = nla_data(nfqa[NFQA_CFG_QUEUE_MAXLEN]);
1356
1357                 spin_lock_bh(&queue->lock);
1358                 queue->queue_maxlen = ntohl(*queue_maxlen);
1359                 spin_unlock_bh(&queue->lock);
1360         }
1361
1362         if (nfqa[NFQA_CFG_FLAGS]) {
1363                 spin_lock_bh(&queue->lock);
1364                 queue->flags &= ~mask;
1365                 queue->flags |= flags & mask;
1366                 spin_unlock_bh(&queue->lock);
1367         }
1368
1369 err_out_unlock:
1370         rcu_read_unlock();
1371         return ret;
1372 }
1373
1374 static const struct nfnl_callback nfqnl_cb[NFQNL_MSG_MAX] = {
1375         [NFQNL_MSG_PACKET]      = {
1376                 .call           = nfqnl_recv_unsupp,
1377                 .type           = NFNL_CB_RCU,
1378                 .attr_count     = NFQA_MAX,
1379         },
1380         [NFQNL_MSG_VERDICT]     = {
1381                 .call           = nfqnl_recv_verdict,
1382                 .type           = NFNL_CB_RCU,
1383                 .attr_count     = NFQA_MAX,
1384                 .policy         = nfqa_verdict_policy
1385         },
1386         [NFQNL_MSG_CONFIG]      = {
1387                 .call           = nfqnl_recv_config,
1388                 .type           = NFNL_CB_MUTEX,
1389                 .attr_count     = NFQA_CFG_MAX,
1390                 .policy         = nfqa_cfg_policy
1391         },
1392         [NFQNL_MSG_VERDICT_BATCH] = {
1393                 .call           = nfqnl_recv_verdict_batch,
1394                 .type           = NFNL_CB_RCU,
1395                 .attr_count     = NFQA_MAX,
1396                 .policy         = nfqa_verdict_batch_policy
1397         },
1398 };
1399
1400 static const struct nfnetlink_subsystem nfqnl_subsys = {
1401         .name           = "nf_queue",
1402         .subsys_id      = NFNL_SUBSYS_QUEUE,
1403         .cb_count       = NFQNL_MSG_MAX,
1404         .cb             = nfqnl_cb,
1405 };
1406
1407 #ifdef CONFIG_PROC_FS
1408 struct iter_state {
1409         struct seq_net_private p;
1410         unsigned int bucket;
1411 };
1412
1413 static struct hlist_node *get_first(struct seq_file *seq)
1414 {
1415         struct iter_state *st = seq->private;
1416         struct net *net;
1417         struct nfnl_queue_net *q;
1418
1419         if (!st)
1420                 return NULL;
1421
1422         net = seq_file_net(seq);
1423         q = nfnl_queue_pernet(net);
1424         for (st->bucket = 0; st->bucket < INSTANCE_BUCKETS; st->bucket++) {
1425                 if (!hlist_empty(&q->instance_table[st->bucket]))
1426                         return q->instance_table[st->bucket].first;
1427         }
1428         return NULL;
1429 }
1430
1431 static struct hlist_node *get_next(struct seq_file *seq, struct hlist_node *h)
1432 {
1433         struct iter_state *st = seq->private;
1434         struct net *net = seq_file_net(seq);
1435
1436         h = h->next;
1437         while (!h) {
1438                 struct nfnl_queue_net *q;
1439
1440                 if (++st->bucket >= INSTANCE_BUCKETS)
1441                         return NULL;
1442
1443                 q = nfnl_queue_pernet(net);
1444                 h = q->instance_table[st->bucket].first;
1445         }
1446         return h;
1447 }
1448
1449 static struct hlist_node *get_idx(struct seq_file *seq, loff_t pos)
1450 {
1451         struct hlist_node *head;
1452         head = get_first(seq);
1453
1454         if (head)
1455                 while (pos && (head = get_next(seq, head)))
1456                         pos--;
1457         return pos ? NULL : head;
1458 }
1459
1460 static void *seq_start(struct seq_file *s, loff_t *pos)
1461         __acquires(nfnl_queue_pernet(seq_file_net(s))->instances_lock)
1462 {
1463         spin_lock(&nfnl_queue_pernet(seq_file_net(s))->instances_lock);
1464         return get_idx(s, *pos);
1465 }
1466
1467 static void *seq_next(struct seq_file *s, void *v, loff_t *pos)
1468 {
1469         (*pos)++;
1470         return get_next(s, v);
1471 }
1472
1473 static void seq_stop(struct seq_file *s, void *v)
1474         __releases(nfnl_queue_pernet(seq_file_net(s))->instances_lock)
1475 {
1476         spin_unlock(&nfnl_queue_pernet(seq_file_net(s))->instances_lock);
1477 }
1478
1479 static int seq_show(struct seq_file *s, void *v)
1480 {
1481         const struct nfqnl_instance *inst = v;
1482
1483         seq_printf(s, "%5u %6u %5u %1u %5u %5u %5u %8u %2d\n",
1484                    inst->queue_num,
1485                    inst->peer_portid, inst->queue_total,
1486                    inst->copy_mode, inst->copy_range,
1487                    inst->queue_dropped, inst->queue_user_dropped,
1488                    inst->id_sequence, 1);
1489         return 0;
1490 }
1491
1492 static const struct seq_operations nfqnl_seq_ops = {
1493         .start  = seq_start,
1494         .next   = seq_next,
1495         .stop   = seq_stop,
1496         .show   = seq_show,
1497 };
1498 #endif /* PROC_FS */
1499
1500 static int __net_init nfnl_queue_net_init(struct net *net)
1501 {
1502         unsigned int i;
1503         struct nfnl_queue_net *q = nfnl_queue_pernet(net);
1504
1505         for (i = 0; i < INSTANCE_BUCKETS; i++)
1506                 INIT_HLIST_HEAD(&q->instance_table[i]);
1507
1508         spin_lock_init(&q->instances_lock);
1509
1510 #ifdef CONFIG_PROC_FS
1511         if (!proc_create_net("nfnetlink_queue", 0440, net->nf.proc_netfilter,
1512                         &nfqnl_seq_ops, sizeof(struct iter_state)))
1513                 return -ENOMEM;
1514 #endif
1515         return 0;
1516 }
1517
1518 static void __net_exit nfnl_queue_net_exit(struct net *net)
1519 {
1520         struct nfnl_queue_net *q = nfnl_queue_pernet(net);
1521         unsigned int i;
1522
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         nf_register_queue_handler(&nfqh);
1567
1568         return status;
1569
1570 cleanup_netlink_subsys:
1571         nfnetlink_subsys_unregister(&nfqnl_subsys);
1572 cleanup_netlink_notifier:
1573         netlink_unregister_notifier(&nfqnl_rtnl_notifier);
1574         unregister_pernet_subsys(&nfnl_queue_net_ops);
1575 out:
1576         return status;
1577 }
1578
1579 static void __exit nfnetlink_queue_fini(void)
1580 {
1581         nf_unregister_queue_handler();
1582         unregister_netdevice_notifier(&nfqnl_dev_notifier);
1583         nfnetlink_subsys_unregister(&nfqnl_subsys);
1584         netlink_unregister_notifier(&nfqnl_rtnl_notifier);
1585         unregister_pernet_subsys(&nfnl_queue_net_ops);
1586
1587         rcu_barrier(); /* Wait for completion of call_rcu()'s */
1588 }
1589
1590 MODULE_DESCRIPTION("netfilter packet queue handler");
1591 MODULE_AUTHOR("Harald Welte <laforge@netfilter.org>");
1592 MODULE_LICENSE("GPL");
1593 MODULE_ALIAS_NFNL_SUBSYS(NFNL_SUBSYS_QUEUE);
1594
1595 module_init(nfnetlink_queue_init);
1596 module_exit(nfnetlink_queue_fini);