net/netfilter/nfnetlink_log.c

   1 /*
   2  * This is a module which is used for logging packets to userspace via
   3  * nfetlink.
   4  *
   5  * (C) 2005 by Harald Welte <laforge@netfilter.org>
   6  * (C) 2006-2012 Patrick McHardy <kaber@trash.net>
   7  *
   8  * Based on the old ipv4-only ipt_ULOG.c:
   9  * (C) 2000-2004 by Harald Welte <laforge@netfilter.org>
  10  *
  11  * This program is free software; you can redistribute it and/or modify
  12  * it under the terms of the GNU General Public License version 2 as
  13  * published by the Free Software Foundation.
  14  */
  15
  16 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  17
  18 #include <linux/module.h>
  19 #include <linux/skbuff.h>
  20 #include <linux/if_arp.h>
  21 #include <linux/init.h>
  22 #include <linux/ip.h>
  23 #include <linux/ipv6.h>
  24 #include <linux/netdevice.h>
  25 #include <linux/netfilter.h>
  26 #include <linux/netfilter_bridge.h>
  27 #include <net/netlink.h>
  28 #include <linux/netfilter/nfnetlink.h>
  29 #include <linux/netfilter/nfnetlink_log.h>
  30 #include <linux/netfilter/nf_conntrack_common.h>
  31 #include <linux/spinlock.h>
  32 #include <linux/sysctl.h>
  33 #include <linux/proc_fs.h>
  34 #include <linux/security.h>
  35 #include <linux/list.h>
  36 #include <linux/slab.h>
  37 #include <net/sock.h>
  38 #include <net/netfilter/nf_log.h>
  39 #include <net/netns/generic.h>
  40 #include <net/netfilter/nfnetlink_log.h>
  41
  42 #include <linux/atomic.h>
  43
  44 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
  45 #include "../bridge/br_private.h"
  46 #endif
  47
  48 #define NFULNL_NLBUFSIZ_DEFAULT NLMSG_GOODSIZE
  49 #define NFULNL_TIMEOUT_DEFAULT  100     /* every second */
  50 #define NFULNL_QTHRESH_DEFAULT  100     /* 100 packets */
  51 /* max packet size is limited by 16-bit struct nfattr nfa_len field */
  52 #define NFULNL_COPY_RANGE_MAX   (0xFFFF - NLA_HDRLEN)
  53
  54 #define PRINTR(x, args...)      do { if (net_ratelimit()) \
  55                                      printk(x, ## args); } while (0);
  56
  57 struct nfulnl_instance {
  58         struct hlist_node hlist;        /* global list of instances */
  59         spinlock_t lock;
  60         atomic_t use;                   /* use count */
  61
  62         unsigned int qlen;              /* number of nlmsgs in skb */
  63         struct sk_buff *skb;            /* pre-allocatd skb */
  64         struct timer_list timer;
  65         struct net *net;
  66         struct user_namespace *peer_user_ns;    /* User namespace of the peer process */
  67         u32 peer_portid;                /* PORTID of the peer process */
  68
  69         /* configurable parameters */
  70         unsigned int flushtimeout;      /* timeout until queue flush */
  71         unsigned int nlbufsiz;          /* netlink buffer allocation size */
  72         unsigned int qthreshold;        /* threshold of the queue */
  73         u_int32_t copy_range;
  74         u_int32_t seq;                  /* instance-local sequential counter */
  75         u_int16_t group_num;            /* number of this queue */
  76         u_int16_t flags;
  77         u_int8_t copy_mode;
  78         struct rcu_head rcu;
  79 };
  80
  81 #define INSTANCE_BUCKETS        16
  82
  83 static int nfnl_log_net_id __read_mostly;
  84
  85 struct nfnl_log_net {
  86         spinlock_t instances_lock;
  87         struct hlist_head instance_table[INSTANCE_BUCKETS];
  88         atomic_t global_seq;
  89 };
  90
  91 static struct nfnl_log_net *nfnl_log_pernet(struct net *net)
  92 {
  93         return net_generic(net, nfnl_log_net_id);
  94 }
  95
  96 static inline u_int8_t instance_hashfn(u_int16_t group_num)
  97 {
  98         return ((group_num & 0xff) % INSTANCE_BUCKETS);
  99 }
 100
 101 static struct nfulnl_instance *
 102 __instance_lookup(struct nfnl_log_net *log, u_int16_t group_num)
 103 {
 104         struct hlist_head *head;
 105         struct nfulnl_instance *inst;
 106
 107         head = &log->instance_table[instance_hashfn(group_num)];
 108         hlist_for_each_entry_rcu(inst, head, hlist) {
 109                 if (inst->group_num == group_num)
 110                         return inst;
 111         }
 112         return NULL;
 113 }
 114
 115 static inline void
 116 instance_get(struct nfulnl_instance *inst)
 117 {
 118         atomic_inc(&inst->use);
 119 }
 120
 121 static struct nfulnl_instance *
 122 instance_lookup_get(struct nfnl_log_net *log, u_int16_t group_num)
 123 {
 124         struct nfulnl_instance *inst;
 125
 126         rcu_read_lock_bh();
 127         inst = __instance_lookup(log, group_num);
 128         if (inst && !atomic_inc_not_zero(&inst->use))
 129                 inst = NULL;
 130         rcu_read_unlock_bh();
 131
 132         return inst;
 133 }
 134
 135 static void nfulnl_instance_free_rcu(struct rcu_head *head)
 136 {
 137         struct nfulnl_instance *inst =
 138                 container_of(head, struct nfulnl_instance, rcu);
 139
 140         put_net(inst->net);
 141         kfree(inst);
 142         module_put(THIS_MODULE);
 143 }
 144
 145 static void
 146 instance_put(struct nfulnl_instance *inst)
 147 {
 148         if (inst && atomic_dec_and_test(&inst->use))
 149                 call_rcu_bh(&inst->rcu, nfulnl_instance_free_rcu);
 150 }
 151
 152 static void nfulnl_timer(unsigned long data);
 153
 154 static struct nfulnl_instance *
 155 instance_create(struct net *net, u_int16_t group_num,
 156                 u32 portid, struct user_namespace *user_ns)
 157 {
 158         struct nfulnl_instance *inst;
 159         struct nfnl_log_net *log = nfnl_log_pernet(net);
 160         int err;
 161
 162         spin_lock_bh(&log->instances_lock);
 163         if (__instance_lookup(log, group_num)) {
 164                 err = -EEXIST;
 165                 goto out_unlock;
 166         }
 167
 168         inst = kzalloc(sizeof(*inst), GFP_ATOMIC);
 169         if (!inst) {
 170                 err = -ENOMEM;
 171                 goto out_unlock;
 172         }
 173
 174         if (!try_module_get(THIS_MODULE)) {
 175                 kfree(inst);
 176                 err = -EAGAIN;
 177                 goto out_unlock;
 178         }
 179
 180         INIT_HLIST_NODE(&inst->hlist);
 181         spin_lock_init(&inst->lock);
 182         /* needs to be two, since we _put() after creation */
 183         atomic_set(&inst->use, 2);
 184
 185         setup_timer(&inst->timer, nfulnl_timer, (unsigned long)inst);
 186
 187         inst->net = get_net(net);
 188         inst->peer_user_ns = user_ns;
 189         inst->peer_portid = portid;
 190         inst->group_num = group_num;
 191
 192         inst->qthreshold        = NFULNL_QTHRESH_DEFAULT;
 193         inst->flushtimeout      = NFULNL_TIMEOUT_DEFAULT;
 194         inst->nlbufsiz          = NFULNL_NLBUFSIZ_DEFAULT;
 195         inst->copy_mode         = NFULNL_COPY_PACKET;
 196         inst->copy_range        = NFULNL_COPY_RANGE_MAX;
 197
 198         hlist_add_head_rcu(&inst->hlist,
 199                        &log->instance_table[instance_hashfn(group_num)]);
 200
 201
 202         spin_unlock_bh(&log->instances_lock);
 203
 204         return inst;
 205
 206 out_unlock:
 207         spin_unlock_bh(&log->instances_lock);
 208         return ERR_PTR(err);
 209 }
 210
 211 static void __nfulnl_flush(struct nfulnl_instance *inst);
 212
 213 /* called with BH disabled */
 214 static void
 215 __instance_destroy(struct nfulnl_instance *inst)
 216 {
 217         /* first pull it out of the global list */
 218         hlist_del_rcu(&inst->hlist);
 219
 220         /* then flush all pending packets from skb */
 221
 222         spin_lock(&inst->lock);
 223
 224         /* lockless readers wont be able to use us */
 225         inst->copy_mode = NFULNL_COPY_DISABLED;
 226
 227         if (inst->skb)
 228                 __nfulnl_flush(inst);
 229         spin_unlock(&inst->lock);
 230
 231         /* and finally put the refcount */
 232         instance_put(inst);
 233 }
 234
 235 static inline void
 236 instance_destroy(struct nfnl_log_net *log,
 237                  struct nfulnl_instance *inst)
 238 {
 239         spin_lock_bh(&log->instances_lock);
 240         __instance_destroy(inst);
 241         spin_unlock_bh(&log->instances_lock);
 242 }
 243
 244 static int
 245 nfulnl_set_mode(struct nfulnl_instance *inst, u_int8_t mode,
 246                   unsigned int range)
 247 {
 248         int status = 0;
 249
 250         spin_lock_bh(&inst->lock);
 251
 252         switch (mode) {
 253         case NFULNL_COPY_NONE:
 254         case NFULNL_COPY_META:
 255                 inst->copy_mode = mode;
 256                 inst->copy_range = 0;
 257                 break;
 258
 259         case NFULNL_COPY_PACKET:
 260                 inst->copy_mode = mode;
 261                 if (range == 0)
 262                         range = NFULNL_COPY_RANGE_MAX;
 263                 inst->copy_range = min_t(unsigned int,
 264                                          range, NFULNL_COPY_RANGE_MAX);
 265                 break;
 266
 267         default:
 268                 status = -EINVAL;
 269                 break;
 270         }
 271
 272         spin_unlock_bh(&inst->lock);
 273
 274         return status;
 275 }
 276
 277 static int
 278 nfulnl_set_nlbufsiz(struct nfulnl_instance *inst, u_int32_t nlbufsiz)
 279 {
 280         int status;
 281
 282         spin_lock_bh(&inst->lock);
 283         if (nlbufsiz < NFULNL_NLBUFSIZ_DEFAULT)
 284                 status = -ERANGE;
 285         else if (nlbufsiz > 131072)
 286                 status = -ERANGE;
 287         else {
 288                 inst->nlbufsiz = nlbufsiz;
 289                 status = 0;
 290         }
 291         spin_unlock_bh(&inst->lock);
 292
 293         return status;
 294 }
 295
 296 static void
 297 nfulnl_set_timeout(struct nfulnl_instance *inst, u_int32_t timeout)
 298 {
 299         spin_lock_bh(&inst->lock);
 300         inst->flushtimeout = timeout;
 301         spin_unlock_bh(&inst->lock);
 302 }
 303
 304 static void
 305 nfulnl_set_qthresh(struct nfulnl_instance *inst, u_int32_t qthresh)
 306 {
 307         spin_lock_bh(&inst->lock);
 308         inst->qthreshold = qthresh;
 309         spin_unlock_bh(&inst->lock);
 310 }
 311
 312 static int
 313 nfulnl_set_flags(struct nfulnl_instance *inst, u_int16_t flags)
 314 {
 315         spin_lock_bh(&inst->lock);
 316         inst->flags = flags;
 317         spin_unlock_bh(&inst->lock);
 318
 319         return 0;
 320 }
 321
 322 static struct sk_buff *
 323 nfulnl_alloc_skb(struct net *net, u32 peer_portid, unsigned int inst_size,
 324                  unsigned int pkt_size)
 325 {
 326         struct sk_buff *skb;
 327         unsigned int n;
 328
 329         /* alloc skb which should be big enough for a whole multipart
 330          * message.  WARNING: has to be <= 128k due to slab restrictions */
 331
 332         n = max(inst_size, pkt_size);
 333         skb = alloc_skb(n, GFP_ATOMIC);
 334         if (!skb) {
 335                 if (n > pkt_size) {
 336                         /* try to allocate only as much as we need for current
 337                          * packet */
 338
 339                         skb = alloc_skb(pkt_size, GFP_ATOMIC);
 340                 }
 341         }
 342
 343         return skb;
 344 }
 345
 346 static void
 347 __nfulnl_send(struct nfulnl_instance *inst)
 348 {
 349         if (inst->qlen > 1) {
 350                 struct nlmsghdr *nlh = nlmsg_put(inst->skb, 0, 0,
 351                                                  NLMSG_DONE,
 352                                                  sizeof(struct nfgenmsg),
 353                                                  0);
 354                 if (WARN_ONCE(!nlh, "bad nlskb size: %u, tailroom %d\n",
 355                               inst->skb->len, skb_tailroom(inst->skb))) {
 356                         kfree_skb(inst->skb);
 357                         goto out;
 358                 }
 359         }
 360         nfnetlink_unicast(inst->skb, inst->net, inst->peer_portid,
 361                           MSG_DONTWAIT);
 362 out:
 363         inst->qlen = 0;
 364         inst->skb = NULL;
 365 }
 366
 367 static void
 368 __nfulnl_flush(struct nfulnl_instance *inst)
 369 {
 370         /* timer holds a reference */
 371         if (del_timer(&inst->timer))
 372                 instance_put(inst);
 373         if (inst->skb)
 374                 __nfulnl_send(inst);
 375 }
 376
 377 static void
 378 nfulnl_timer(unsigned long data)
 379 {
 380         struct nfulnl_instance *inst = (struct nfulnl_instance *)data;
 381
 382         spin_lock_bh(&inst->lock);
 383         if (inst->skb)
 384                 __nfulnl_send(inst);
 385         spin_unlock_bh(&inst->lock);
 386         instance_put(inst);
 387 }
 388
 389 /* This is an inline function, we don't really care about a long
 390  * list of arguments */
 391 static inline int
 392 __build_packet_message(struct nfnl_log_net *log,
 393                         struct nfulnl_instance *inst,
 394                         const struct sk_buff *skb,
 395                         unsigned int data_len,
 396                         u_int8_t pf,
 397                         unsigned int hooknum,
 398                         const struct net_device *indev,
 399                         const struct net_device *outdev,
 400                         const char *prefix, unsigned int plen,
 401                         const struct nfnl_ct_hook *nfnl_ct,
 402                         struct nf_conn *ct, enum ip_conntrack_info ctinfo)
 403 {
 404         struct nfulnl_msg_packet_hdr pmsg;
 405         struct nlmsghdr *nlh;
 406         struct nfgenmsg *nfmsg;
 407         sk_buff_data_t old_tail = inst->skb->tail;
 408         struct sock *sk;
 409         const unsigned char *hwhdrp;
 410
 411         nlh = nlmsg_put(inst->skb, 0, 0,
 412                         NFNL_SUBSYS_ULOG << 8 | NFULNL_MSG_PACKET,
 413                         sizeof(struct nfgenmsg), 0);
 414         if (!nlh)
 415                 return -1;
 416         nfmsg = nlmsg_data(nlh);
 417         nfmsg->nfgen_family = pf;
 418         nfmsg->version = NFNETLINK_V0;
 419         nfmsg->res_id = htons(inst->group_num);
 420
 421         memset(&pmsg, 0, sizeof(pmsg));
 422         pmsg.hw_protocol        = skb->protocol;
 423         pmsg.hook               = hooknum;
 424
 425         if (nla_put(inst->skb, NFULA_PACKET_HDR, sizeof(pmsg), &pmsg))
 426                 goto nla_put_failure;
 427
 428         if (prefix &&
 429             nla_put(inst->skb, NFULA_PREFIX, plen, prefix))
 430                 goto nla_put_failure;
 431
 432         if (indev) {
 433 #if !IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
 434                 if (nla_put_be32(inst->skb, NFULA_IFINDEX_INDEV,
 435                                  htonl(indev->ifindex)))
 436                         goto nla_put_failure;
 437 #else
 438                 if (pf == PF_BRIDGE) {
 439                         /* Case 1: outdev is physical input device, we need to
 440                          * look for bridge group (when called from
 441                          * netfilter_bridge) */
 442                         if (nla_put_be32(inst->skb, NFULA_IFINDEX_PHYSINDEV,
 443                                          htonl(indev->ifindex)) ||
 444                         /* this is the bridge group "brX" */
 445                         /* rcu_read_lock()ed by nf_hook_thresh or
 446                          * nf_log_packet.
 447                          */
 448                             nla_put_be32(inst->skb, NFULA_IFINDEX_INDEV,
 449                                          htonl(br_port_get_rcu(indev)->br->dev->ifindex)))
 450                                 goto nla_put_failure;
 451                 } else {
 452                         struct net_device *physindev;
 453
 454                         /* Case 2: indev is bridge group, we need to look for
 455                          * physical device (when called from ipv4) */
 456                         if (nla_put_be32(inst->skb, NFULA_IFINDEX_INDEV,
 457                                          htonl(indev->ifindex)))
 458                                 goto nla_put_failure;
 459
 460                         physindev = nf_bridge_get_physindev(skb);
 461                         if (physindev &&
 462                             nla_put_be32(inst->skb, NFULA_IFINDEX_PHYSINDEV,
 463                                          htonl(physindev->ifindex)))
 464                                 goto nla_put_failure;
 465                 }
 466 #endif
 467         }
 468
 469         if (outdev) {
 470 #if !IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
 471                 if (nla_put_be32(inst->skb, NFULA_IFINDEX_OUTDEV,
 472                                  htonl(outdev->ifindex)))
 473                         goto nla_put_failure;
 474 #else
 475                 if (pf == PF_BRIDGE) {
 476                         /* Case 1: outdev is physical output device, we need to
 477                          * look for bridge group (when called from
 478                          * netfilter_bridge) */
 479                         if (nla_put_be32(inst->skb, NFULA_IFINDEX_PHYSOUTDEV,
 480                                          htonl(outdev->ifindex)) ||
 481                         /* this is the bridge group "brX" */
 482                         /* rcu_read_lock()ed by nf_hook_thresh or
 483                          * nf_log_packet.
 484                          */
 485                             nla_put_be32(inst->skb, NFULA_IFINDEX_OUTDEV,
 486                                          htonl(br_port_get_rcu(outdev)->br->dev->ifindex)))
 487                                 goto nla_put_failure;
 488                 } else {
 489                         struct net_device *physoutdev;
 490
 491                         /* Case 2: indev is a bridge group, we need to look
 492                          * for physical device (when called from ipv4) */
 493                         if (nla_put_be32(inst->skb, NFULA_IFINDEX_OUTDEV,
 494                                          htonl(outdev->ifindex)))
 495                                 goto nla_put_failure;
 496
 497                         physoutdev = nf_bridge_get_physoutdev(skb);
 498                         if (physoutdev &&
 499                             nla_put_be32(inst->skb, NFULA_IFINDEX_PHYSOUTDEV,
 500                                          htonl(physoutdev->ifindex)))
 501                                 goto nla_put_failure;
 502                 }
 503 #endif
 504         }
 505
 506         if (skb->mark &&
 507             nla_put_be32(inst->skb, NFULA_MARK, htonl(skb->mark)))
 508                 goto nla_put_failure;
 509
 510         if (indev && skb->dev &&
 511             skb->mac_header != skb->network_header) {
 512                 struct nfulnl_msg_packet_hw phw;
 513                 int len;
 514
 515                 memset(&phw, 0, sizeof(phw));
 516                 len = dev_parse_header(skb, phw.hw_addr);
 517                 if (len > 0) {
 518                         phw.hw_addrlen = htons(len);
 519                         if (nla_put(inst->skb, NFULA_HWADDR, sizeof(phw), &phw))
 520                                 goto nla_put_failure;
 521                 }
 522         }
 523
 524         if (indev && skb_mac_header_was_set(skb)) {
 525                 if (nla_put_be16(inst->skb, NFULA_HWTYPE, htons(skb->dev->type)) ||
 526                     nla_put_be16(inst->skb, NFULA_HWLEN,
 527                                  htons(skb->dev->hard_header_len)))
 528                         goto nla_put_failure;
 529
 530                 hwhdrp = skb_mac_header(skb);
 531
 532                 if (skb->dev->type == ARPHRD_SIT)
 533                         hwhdrp -= ETH_HLEN;
 534
 535                 if (hwhdrp >= skb->head &&
 536                     nla_put(inst->skb, NFULA_HWHEADER,
 537                             skb->dev->hard_header_len, hwhdrp))
 538                         goto nla_put_failure;
 539         }
 540
 541         if (skb->tstamp.tv64) {
 542                 struct nfulnl_msg_packet_timestamp ts;
 543                 struct timespec64 kts = ktime_to_timespec64(skb->tstamp);
 544                 ts.sec = cpu_to_be64(kts.tv_sec);
 545                 ts.usec = cpu_to_be64(kts.tv_nsec / NSEC_PER_USEC);
 546
 547                 if (nla_put(inst->skb, NFULA_TIMESTAMP, sizeof(ts), &ts))
 548                         goto nla_put_failure;
 549         }
 550
 551         /* UID */
 552         sk = skb->sk;
 553         if (sk && sk_fullsock(sk)) {
 554                 read_lock_bh(&sk->sk_callback_lock);
 555                 if (sk->sk_socket && sk->sk_socket->file) {
 556                         struct file *file = sk->sk_socket->file;
 557                         const struct cred *cred = file->f_cred;
 558                         struct user_namespace *user_ns = inst->peer_user_ns;
 559                         __be32 uid = htonl(from_kuid_munged(user_ns, cred->fsuid));
 560                         __be32 gid = htonl(from_kgid_munged(user_ns, cred->fsgid));
 561                         read_unlock_bh(&sk->sk_callback_lock);
 562                         if (nla_put_be32(inst->skb, NFULA_UID, uid) ||
 563                             nla_put_be32(inst->skb, NFULA_GID, gid))
 564                                 goto nla_put_failure;
 565                 } else
 566                         read_unlock_bh(&sk->sk_callback_lock);
 567         }
 568
 569         /* local sequence number */
 570         if ((inst->flags & NFULNL_CFG_F_SEQ) &&
 571             nla_put_be32(inst->skb, NFULA_SEQ, htonl(inst->seq++)))
 572                 goto nla_put_failure;
 573
 574         /* global sequence number */
 575         if ((inst->flags & NFULNL_CFG_F_SEQ_GLOBAL) &&
 576             nla_put_be32(inst->skb, NFULA_SEQ_GLOBAL,
 577                          htonl(atomic_inc_return(&log->global_seq))))
 578                 goto nla_put_failure;
 579
 580         if (ct && nfnl_ct->build(inst->skb, ct, ctinfo,
 581                                  NFULA_CT, NFULA_CT_INFO) < 0)
 582                 goto nla_put_failure;
 583
 584         if (data_len) {
 585                 struct nlattr *nla;
 586                 int size = nla_attr_size(data_len);
 587
 588                 if (skb_tailroom(inst->skb) < nla_total_size(data_len))
 589                         goto nla_put_failure;
 590
 591                 nla = (struct nlattr *)skb_put(inst->skb, nla_total_size(data_len));
 592                 nla->nla_type = NFULA_PAYLOAD;
 593                 nla->nla_len = size;
 594
 595                 if (skb_copy_bits(skb, 0, nla_data(nla), data_len))
 596                         BUG();
 597         }
 598
 599         nlh->nlmsg_len = inst->skb->tail - old_tail;
 600         return 0;
 601
 602 nla_put_failure:
 603         PRINTR(KERN_ERR "nfnetlink_log: error creating log nlmsg\n");
 604         return -1;
 605 }
 606
 607 static struct nf_loginfo default_loginfo = {
 608         .type =         NF_LOG_TYPE_ULOG,
 609         .u = {
 610                 .ulog = {
 611                         .copy_len       = 0xffff,
 612                         .group          = 0,
 613                         .qthreshold     = 1,
 614                 },
 615         },
 616 };
 617
 618 /* log handler for internal netfilter logging api */
 619 void
 620 nfulnl_log_packet(struct net *net,
 621                   u_int8_t pf,
 622                   unsigned int hooknum,
 623                   const struct sk_buff *skb,
 624                   const struct net_device *in,
 625                   const struct net_device *out,
 626                   const struct nf_loginfo *li_user,
 627                   const char *prefix)
 628 {
 629         size_t size;
 630         unsigned int data_len;
 631         struct nfulnl_instance *inst;
 632         const struct nf_loginfo *li;
 633         unsigned int qthreshold;
 634         unsigned int plen;
 635         struct nfnl_log_net *log = nfnl_log_pernet(net);
 636         const struct nfnl_ct_hook *nfnl_ct = NULL;
 637         struct nf_conn *ct = NULL;
 638         enum ip_conntrack_info uninitialized_var(ctinfo);
 639
 640         if (li_user && li_user->type == NF_LOG_TYPE_ULOG)
 641                 li = li_user;
 642         else
 643                 li = &default_loginfo;
 644
 645         inst = instance_lookup_get(log, li->u.ulog.group);
 646         if (!inst)
 647                 return;
 648
 649         plen = 0;
 650         if (prefix)
 651                 plen = strlen(prefix) + 1;
 652
 653         /* FIXME: do we want to make the size calculation conditional based on
 654          * what is actually present?  way more branches and checks, but more
 655          * memory efficient... */
 656         size =    nlmsg_total_size(sizeof(struct nfgenmsg))
 657                 + nla_total_size(sizeof(struct nfulnl_msg_packet_hdr))
 658                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
 659                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
 660 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
 661                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
 662                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
 663 #endif
 664                 + nla_total_size(sizeof(u_int32_t))     /* mark */
 665                 + nla_total_size(sizeof(u_int32_t))     /* uid */
 666                 + nla_total_size(sizeof(u_int32_t))     /* gid */
 667                 + nla_total_size(plen)                  /* prefix */
 668                 + nla_total_size(sizeof(struct nfulnl_msg_packet_hw))
 669                 + nla_total_size(sizeof(struct nfulnl_msg_packet_timestamp))
 670                 + nla_total_size(sizeof(struct nfgenmsg));      /* NLMSG_DONE */
 671
 672         if (in && skb_mac_header_was_set(skb)) {
 673                 size +=   nla_total_size(skb->dev->hard_header_len)
 674                         + nla_total_size(sizeof(u_int16_t))     /* hwtype */
 675                         + nla_total_size(sizeof(u_int16_t));    /* hwlen */
 676         }
 677
 678         spin_lock_bh(&inst->lock);
 679
 680         if (inst->flags & NFULNL_CFG_F_SEQ)
 681                 size += nla_total_size(sizeof(u_int32_t));
 682         if (inst->flags & NFULNL_CFG_F_SEQ_GLOBAL)
 683                 size += nla_total_size(sizeof(u_int32_t));
 684         if (inst->flags & NFULNL_CFG_F_CONNTRACK) {
 685                 nfnl_ct = rcu_dereference(nfnl_ct_hook);
 686                 if (nfnl_ct != NULL) {
 687                         ct = nfnl_ct->get_ct(skb, &ctinfo);
 688                         if (ct != NULL)
 689                                 size += nfnl_ct->build_size(ct);
 690                 }
 691         }
 692
 693         qthreshold = inst->qthreshold;
 694         /* per-rule qthreshold overrides per-instance */
 695         if (li->u.ulog.qthreshold)
 696                 if (qthreshold > li->u.ulog.qthreshold)
 697                         qthreshold = li->u.ulog.qthreshold;
 698
 699
 700         switch (inst->copy_mode) {
 701         case NFULNL_COPY_META:
 702         case NFULNL_COPY_NONE:
 703                 data_len = 0;
 704                 break;
 705
 706         case NFULNL_COPY_PACKET:
 707                 data_len = inst->copy_range;
 708                 if ((li->u.ulog.flags & NF_LOG_F_COPY_LEN) &&
 709                     (li->u.ulog.copy_len < data_len))
 710                         data_len = li->u.ulog.copy_len;
 711
 712                 if (data_len > skb->len)
 713                         data_len = skb->len;
 714
 715                 size += nla_total_size(data_len);
 716                 break;
 717
 718         case NFULNL_COPY_DISABLED:
 719         default:
 720                 goto unlock_and_release;
 721         }
 722
 723         if (inst->skb && size > skb_tailroom(inst->skb)) {
 724                 /* either the queue len is too high or we don't have
 725                  * enough room in the skb left. flush to userspace. */
 726                 __nfulnl_flush(inst);
 727         }
 728
 729         if (!inst->skb) {
 730                 inst->skb = nfulnl_alloc_skb(net, inst->peer_portid,
 731                                              inst->nlbufsiz, size);
 732                 if (!inst->skb)
 733                         goto alloc_failure;
 734         }
 735
 736         inst->qlen++;
 737
 738         __build_packet_message(log, inst, skb, data_len, pf,
 739                                 hooknum, in, out, prefix, plen,
 740                                 nfnl_ct, ct, ctinfo);
 741
 742         if (inst->qlen >= qthreshold)
 743                 __nfulnl_flush(inst);
 744         /* timer_pending always called within inst->lock, so there
 745          * is no chance of a race here */
 746         else if (!timer_pending(&inst->timer)) {
 747                 instance_get(inst);
 748                 inst->timer.expires = jiffies + (inst->flushtimeout*HZ/100);
 749                 add_timer(&inst->timer);
 750         }
 751
 752 unlock_and_release:
 753         spin_unlock_bh(&inst->lock);
 754         instance_put(inst);
 755         return;
 756
 757 alloc_failure:
 758         /* FIXME: statistics */
 759         goto unlock_and_release;
 760 }
 761 EXPORT_SYMBOL_GPL(nfulnl_log_packet);
 762
 763 static int
 764 nfulnl_rcv_nl_event(struct notifier_block *this,
 765                    unsigned long event, void *ptr)
 766 {
 767         struct netlink_notify *n = ptr;
 768         struct nfnl_log_net *log = nfnl_log_pernet(n->net);
 769
 770         if (event == NETLINK_URELEASE && n->protocol == NETLINK_NETFILTER) {
 771                 int i;
 772
 773                 /* destroy all instances for this portid */
 774                 spin_lock_bh(&log->instances_lock);
 775                 for  (i = 0; i < INSTANCE_BUCKETS; i++) {
 776                         struct hlist_node *t2;
 777                         struct nfulnl_instance *inst;
 778                         struct hlist_head *head = &log->instance_table[i];
 779
 780                         hlist_for_each_entry_safe(inst, t2, head, hlist) {
 781                                 if (n->portid == inst->peer_portid)
 782                                         __instance_destroy(inst);
 783                         }
 784                 }
 785                 spin_unlock_bh(&log->instances_lock);
 786         }
 787         return NOTIFY_DONE;
 788 }
 789
 790 static struct notifier_block nfulnl_rtnl_notifier = {
 791         .notifier_call  = nfulnl_rcv_nl_event,
 792 };
 793
 794 static int nfulnl_recv_unsupp(struct net *net, struct sock *ctnl,
 795                               struct sk_buff *skb, const struct nlmsghdr *nlh,
 796                               const struct nlattr * const nfqa[])
 797 {
 798         return -ENOTSUPP;
 799 }
 800
 801 static struct nf_logger nfulnl_logger __read_mostly = {
 802         .name   = "nfnetlink_log",
 803         .type   = NF_LOG_TYPE_ULOG,
 804         .logfn  = &nfulnl_log_packet,
 805         .me     = THIS_MODULE,
 806 };
 807
 808 static const struct nla_policy nfula_cfg_policy[NFULA_CFG_MAX+1] = {
 809         [NFULA_CFG_CMD]         = { .len = sizeof(struct nfulnl_msg_config_cmd) },
 810         [NFULA_CFG_MODE]        = { .len = sizeof(struct nfulnl_msg_config_mode) },
 811         [NFULA_CFG_TIMEOUT]     = { .type = NLA_U32 },
 812         [NFULA_CFG_QTHRESH]     = { .type = NLA_U32 },
 813         [NFULA_CFG_NLBUFSIZ]    = { .type = NLA_U32 },
 814         [NFULA_CFG_FLAGS]       = { .type = NLA_U16 },
 815 };
 816
 817 static int nfulnl_recv_config(struct net *net, struct sock *ctnl,
 818                               struct sk_buff *skb, const struct nlmsghdr *nlh,
 819                               const struct nlattr * const nfula[])
 820 {
 821         struct nfgenmsg *nfmsg = nlmsg_data(nlh);
 822         u_int16_t group_num = ntohs(nfmsg->res_id);
 823         struct nfulnl_instance *inst;
 824         struct nfulnl_msg_config_cmd *cmd = NULL;
 825         struct nfnl_log_net *log = nfnl_log_pernet(net);
 826         int ret = 0;
 827         u16 flags = 0;
 828
 829         if (nfula[NFULA_CFG_CMD]) {
 830                 u_int8_t pf = nfmsg->nfgen_family;
 831                 cmd = nla_data(nfula[NFULA_CFG_CMD]);
 832
 833                 /* Commands without queue context */
 834                 switch (cmd->command) {
 835                 case NFULNL_CFG_CMD_PF_BIND:
 836                         return nf_log_bind_pf(net, pf, &nfulnl_logger);
 837                 case NFULNL_CFG_CMD_PF_UNBIND:
 838                         nf_log_unbind_pf(net, pf);
 839                         return 0;
 840                 }
 841         }
 842
 843         inst = instance_lookup_get(log, group_num);
 844         if (inst && inst->peer_portid != NETLINK_CB(skb).portid) {
 845                 ret = -EPERM;
 846                 goto out_put;
 847         }
 848
 849         /* Check if we support these flags in first place, dependencies should
 850          * be there too not to break atomicity.
 851          */
 852         if (nfula[NFULA_CFG_FLAGS]) {
 853                 flags = ntohs(nla_get_be16(nfula[NFULA_CFG_FLAGS]));
 854
 855                 if ((flags & NFULNL_CFG_F_CONNTRACK) &&
 856                     !rcu_access_pointer(nfnl_ct_hook)) {
 857 #ifdef CONFIG_MODULES
 858                         nfnl_unlock(NFNL_SUBSYS_ULOG);
 859                         request_module("ip_conntrack_netlink");
 860                         nfnl_lock(NFNL_SUBSYS_ULOG);
 861                         if (rcu_access_pointer(nfnl_ct_hook)) {
 862                                 ret = -EAGAIN;
 863                                 goto out_put;
 864                         }
 865 #endif
 866                         ret = -EOPNOTSUPP;
 867                         goto out_put;
 868                 }
 869         }
 870
 871         if (cmd != NULL) {
 872                 switch (cmd->command) {
 873                 case NFULNL_CFG_CMD_BIND:
 874                         if (inst) {
 875                                 ret = -EBUSY;
 876                                 goto out_put;
 877                         }
 878
 879                         inst = instance_create(net, group_num,
 880                                                NETLINK_CB(skb).portid,
 881                                                sk_user_ns(NETLINK_CB(skb).sk));
 882                         if (IS_ERR(inst)) {
 883                                 ret = PTR_ERR(inst);
 884                                 goto out;
 885                         }
 886                         break;
 887                 case NFULNL_CFG_CMD_UNBIND:
 888                         if (!inst) {
 889                                 ret = -ENODEV;
 890                                 goto out;
 891                         }
 892
 893                         instance_destroy(log, inst);
 894                         goto out_put;
 895                 default:
 896                         ret = -ENOTSUPP;
 897                         goto out_put;
 898                 }
 899         } else if (!inst) {
 900                 ret = -ENODEV;
 901                 goto out;
 902         }
 903
 904         if (nfula[NFULA_CFG_MODE]) {
 905                 struct nfulnl_msg_config_mode *params =
 906                         nla_data(nfula[NFULA_CFG_MODE]);
 907
 908                 nfulnl_set_mode(inst, params->copy_mode,
 909                                 ntohl(params->copy_range));
 910         }
 911
 912         if (nfula[NFULA_CFG_TIMEOUT]) {
 913                 __be32 timeout = nla_get_be32(nfula[NFULA_CFG_TIMEOUT]);
 914
 915                 nfulnl_set_timeout(inst, ntohl(timeout));
 916         }
 917
 918         if (nfula[NFULA_CFG_NLBUFSIZ]) {
 919                 __be32 nlbufsiz = nla_get_be32(nfula[NFULA_CFG_NLBUFSIZ]);
 920
 921                 nfulnl_set_nlbufsiz(inst, ntohl(nlbufsiz));
 922         }
 923
 924         if (nfula[NFULA_CFG_QTHRESH]) {
 925                 __be32 qthresh = nla_get_be32(nfula[NFULA_CFG_QTHRESH]);
 926
 927                 nfulnl_set_qthresh(inst, ntohl(qthresh));
 928         }
 929
 930         if (nfula[NFULA_CFG_FLAGS])
 931                 nfulnl_set_flags(inst, flags);
 932
 933 out_put:
 934         instance_put(inst);
 935 out:
 936         return ret;
 937 }
 938
 939 static const struct nfnl_callback nfulnl_cb[NFULNL_MSG_MAX] = {
 940         [NFULNL_MSG_PACKET]     = { .call = nfulnl_recv_unsupp,
 941                                     .attr_count = NFULA_MAX, },
 942         [NFULNL_MSG_CONFIG]     = { .call = nfulnl_recv_config,
 943                                     .attr_count = NFULA_CFG_MAX,
 944                                     .policy = nfula_cfg_policy },
 945 };
 946
 947 static const struct nfnetlink_subsystem nfulnl_subsys = {
 948         .name           = "log",
 949         .subsys_id      = NFNL_SUBSYS_ULOG,
 950         .cb_count       = NFULNL_MSG_MAX,
 951         .cb             = nfulnl_cb,
 952 };
 953
 954 #ifdef CONFIG_PROC_FS
 955 struct iter_state {
 956         struct seq_net_private p;
 957         unsigned int bucket;
 958 };
 959
 960 static struct hlist_node *get_first(struct net *net, struct iter_state *st)
 961 {
 962         struct nfnl_log_net *log;
 963         if (!st)
 964                 return NULL;
 965
 966         log = nfnl_log_pernet(net);
 967
 968         for (st->bucket = 0; st->bucket < INSTANCE_BUCKETS; st->bucket++) {
 969                 struct hlist_head *head = &log->instance_table[st->bucket];
 970
 971                 if (!hlist_empty(head))
 972                         return rcu_dereference_bh(hlist_first_rcu(head));
 973         }
 974         return NULL;
 975 }
 976
 977 static struct hlist_node *get_next(struct net *net, struct iter_state *st,
 978                                    struct hlist_node *h)
 979 {
 980         h = rcu_dereference_bh(hlist_next_rcu(h));
 981         while (!h) {
 982                 struct nfnl_log_net *log;
 983                 struct hlist_head *head;
 984
 985                 if (++st->bucket >= INSTANCE_BUCKETS)
 986                         return NULL;
 987
 988                 log = nfnl_log_pernet(net);
 989                 head = &log->instance_table[st->bucket];
 990                 h = rcu_dereference_bh(hlist_first_rcu(head));
 991         }
 992         return h;
 993 }
 994
 995 static struct hlist_node *get_idx(struct net *net, struct iter_state *st,
 996                                   loff_t pos)
 997 {
 998         struct hlist_node *head;
 999         head = get_first(net, st);
1000
1001         if (head)
1002                 while (pos && (head = get_next(net, st, head)))
1003                         pos--;
1004         return pos ? NULL : head;
1005 }
1006
1007 static void *seq_start(struct seq_file *s, loff_t *pos)
1008         __acquires(rcu_bh)
1009 {
1010         rcu_read_lock_bh();
1011         return get_idx(seq_file_net(s), s->private, *pos);
1012 }
1013
1014 static void *seq_next(struct seq_file *s, void *v, loff_t *pos)
1015 {
1016         (*pos)++;
1017         return get_next(seq_file_net(s), s->private, v);
1018 }
1019
1020 static void seq_stop(struct seq_file *s, void *v)
1021         __releases(rcu_bh)
1022 {
1023         rcu_read_unlock_bh();
1024 }
1025
1026 static int seq_show(struct seq_file *s, void *v)
1027 {
1028         const struct nfulnl_instance *inst = v;
1029
1030         seq_printf(s, "%5u %6u %5u %1u %5u %6u %2u\n",
1031                    inst->group_num,
1032                    inst->peer_portid, inst->qlen,
1033                    inst->copy_mode, inst->copy_range,
1034                    inst->flushtimeout, atomic_read(&inst->use));
1035
1036         return 0;
1037 }
1038
1039 static const struct seq_operations nful_seq_ops = {
1040         .start  = seq_start,
1041         .next   = seq_next,
1042         .stop   = seq_stop,
1043         .show   = seq_show,
1044 };
1045
1046 static int nful_open(struct inode *inode, struct file *file)
1047 {
1048         return seq_open_net(inode, file, &nful_seq_ops,
1049                             sizeof(struct iter_state));
1050 }
1051
1052 static const struct file_operations nful_file_ops = {
1053         .owner   = THIS_MODULE,
1054         .open    = nful_open,
1055         .read    = seq_read,
1056         .llseek  = seq_lseek,
1057         .release = seq_release_net,
1058 };
1059
1060 #endif /* PROC_FS */
1061
1062 static int __net_init nfnl_log_net_init(struct net *net)
1063 {
1064         unsigned int i;
1065         struct nfnl_log_net *log = nfnl_log_pernet(net);
1066 #ifdef CONFIG_PROC_FS
1067         struct proc_dir_entry *proc;
1068         kuid_t root_uid;
1069         kgid_t root_gid;
1070 #endif
1071
1072         for (i = 0; i < INSTANCE_BUCKETS; i++)
1073                 INIT_HLIST_HEAD(&log->instance_table[i]);
1074         spin_lock_init(&log->instances_lock);
1075
1076 #ifdef CONFIG_PROC_FS
1077         proc = proc_create("nfnetlink_log", 0440,
1078                            net->nf.proc_netfilter, &nful_file_ops);
1079         if (!proc)
1080                 return -ENOMEM;
1081
1082         root_uid = make_kuid(net->user_ns, 0);
1083         root_gid = make_kgid(net->user_ns, 0);
1084         if (uid_valid(root_uid) && gid_valid(root_gid))
1085                 proc_set_user(proc, root_uid, root_gid);
1086 #endif
1087         return 0;
1088 }
1089
1090 static void __net_exit nfnl_log_net_exit(struct net *net)
1091 {
1092 #ifdef CONFIG_PROC_FS
1093         remove_proc_entry("nfnetlink_log", net->nf.proc_netfilter);
1094 #endif
1095         nf_log_unset(net, &nfulnl_logger);
1096 }
1097
1098 static struct pernet_operations nfnl_log_net_ops = {
1099         .init   = nfnl_log_net_init,
1100         .exit   = nfnl_log_net_exit,
1101         .id     = &nfnl_log_net_id,
1102         .size   = sizeof(struct nfnl_log_net),
1103 };
1104
1105 static int __init nfnetlink_log_init(void)
1106 {
1107         int status;
1108
1109         status = register_pernet_subsys(&nfnl_log_net_ops);
1110         if (status < 0) {
1111                 pr_err("failed to register pernet ops\n");
1112                 goto out;
1113         }
1114
1115         netlink_register_notifier(&nfulnl_rtnl_notifier);
1116         status = nfnetlink_subsys_register(&nfulnl_subsys);
1117         if (status < 0) {
1118                 pr_err("failed to create netlink socket\n");
1119                 goto cleanup_netlink_notifier;
1120         }
1121
1122         status = nf_log_register(NFPROTO_UNSPEC, &nfulnl_logger);
1123         if (status < 0) {
1124                 pr_err("failed to register logger\n");
1125                 goto cleanup_subsys;
1126         }
1127
1128         return status;
1129
1130 cleanup_subsys:
1131         nfnetlink_subsys_unregister(&nfulnl_subsys);
1132 cleanup_netlink_notifier:
1133         netlink_unregister_notifier(&nfulnl_rtnl_notifier);
1134         unregister_pernet_subsys(&nfnl_log_net_ops);
1135 out:
1136         return status;
1137 }
1138
1139 static void __exit nfnetlink_log_fini(void)
1140 {
1141         nf_log_unregister(&nfulnl_logger);
1142         nfnetlink_subsys_unregister(&nfulnl_subsys);
1143         netlink_unregister_notifier(&nfulnl_rtnl_notifier);
1144         unregister_pernet_subsys(&nfnl_log_net_ops);
1145 }
1146
1147 MODULE_DESCRIPTION("netfilter userspace logging");
1148 MODULE_AUTHOR("Harald Welte <laforge@netfilter.org>");
1149 MODULE_LICENSE("GPL");
1150 MODULE_ALIAS_NFNL_SUBSYS(NFNL_SUBSYS_ULOG);
1151 MODULE_ALIAS_NF_LOGGER(AF_INET, 1);
1152 MODULE_ALIAS_NF_LOGGER(AF_INET6, 1);
1153 MODULE_ALIAS_NF_LOGGER(AF_BRIDGE, 1);
1154 MODULE_ALIAS_NF_LOGGER(3, 1); /* NFPROTO_ARP */
1155
1156 module_init(nfnetlink_log_init);
1157 module_exit(nfnetlink_log_fini);