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 #include <linux/module.h>
  16 #include <linux/skbuff.h>
  17 #include <linux/if_arp.h>
  18 #include <linux/init.h>
  19 #include <linux/ip.h>
  20 #include <linux/ipv6.h>
  21 #include <linux/netdevice.h>
  22 #include <linux/netfilter.h>
  23 #include <net/netlink.h>
  24 #include <linux/netfilter/nfnetlink.h>
  25 #include <linux/netfilter/nfnetlink_log.h>
  26 #include <linux/spinlock.h>
  27 #include <linux/sysctl.h>
  28 #include <linux/proc_fs.h>
  29 #include <linux/security.h>
  30 #include <linux/list.h>
  31 #include <linux/jhash.h>
  32 #include <linux/random.h>
  33 #include <linux/slab.h>
  34 #include <net/sock.h>
  35 #include <net/netfilter/nf_log.h>
  36 #include <net/netns/generic.h>
  37 #include <net/netfilter/nfnetlink_log.h>
  38
  39 #include <linux/atomic.h>
  40
  41 #ifdef CONFIG_BRIDGE_NETFILTER
  42 #include "../bridge/br_private.h"
  43 #endif
  44
  45 #define NFULNL_NLBUFSIZ_DEFAULT NLMSG_GOODSIZE
  46 #define NFULNL_TIMEOUT_DEFAULT  100     /* every second */
  47 #define NFULNL_QTHRESH_DEFAULT  100     /* 100 packets */
  48 #define NFULNL_COPY_RANGE_MAX   0xFFFF  /* max packet size is limited by 16-bit struct nfattr nfa_len field */
  49
  50 #define PRINTR(x, args...)      do { if (net_ratelimit()) \
  51                                      printk(x, ## args); } while (0);
  52
  53 struct nfulnl_instance {
  54         struct hlist_node hlist;        /* global list of instances */
  55         spinlock_t lock;
  56         atomic_t use;                   /* use count */
  57
  58         unsigned int qlen;              /* number of nlmsgs in skb */
  59         struct sk_buff *skb;            /* pre-allocatd skb */
  60         struct timer_list timer;
  61         struct net *net;
  62         struct user_namespace *peer_user_ns;    /* User namespace of the peer process */
  63         int peer_portid;                        /* PORTID of the peer process */
  64
  65         /* configurable parameters */
  66         unsigned int flushtimeout;      /* timeout until queue flush */
  67         unsigned int nlbufsiz;          /* netlink buffer allocation size */
  68         unsigned int qthreshold;        /* threshold of the queue */
  69         u_int32_t copy_range;
  70         u_int32_t seq;                  /* instance-local sequential counter */
  71         u_int16_t group_num;            /* number of this queue */
  72         u_int16_t flags;
  73         u_int8_t copy_mode;
  74         struct rcu_head rcu;
  75 };
  76
  77 #define INSTANCE_BUCKETS        16
  78 static unsigned int hash_init;
  79
  80 static int nfnl_log_net_id __read_mostly;
  81
  82 struct nfnl_log_net {
  83         spinlock_t instances_lock;
  84         struct hlist_head instance_table[INSTANCE_BUCKETS];
  85         atomic_t global_seq;
  86 };
  87
  88 static struct nfnl_log_net *nfnl_log_pernet(struct net *net)
  89 {
  90         return net_generic(net, nfnl_log_net_id);
  91 }
  92
  93 static inline u_int8_t instance_hashfn(u_int16_t group_num)
  94 {
  95         return ((group_num & 0xff) % INSTANCE_BUCKETS);
  96 }
  97
  98 static struct nfulnl_instance *
  99 __instance_lookup(struct nfnl_log_net *log, u_int16_t group_num)
 100 {
 101         struct hlist_head *head;
 102         struct nfulnl_instance *inst;
 103
 104         head = &log->instance_table[instance_hashfn(group_num)];
 105         hlist_for_each_entry_rcu(inst, head, hlist) {
 106                 if (inst->group_num == group_num)
 107                         return inst;
 108         }
 109         return NULL;
 110 }
 111
 112 static inline void
 113 instance_get(struct nfulnl_instance *inst)
 114 {
 115         atomic_inc(&inst->use);
 116 }
 117
 118 static struct nfulnl_instance *
 119 instance_lookup_get(struct nfnl_log_net *log, u_int16_t group_num)
 120 {
 121         struct nfulnl_instance *inst;
 122
 123         rcu_read_lock_bh();
 124         inst = __instance_lookup(log, group_num);
 125         if (inst && !atomic_inc_not_zero(&inst->use))
 126                 inst = NULL;
 127         rcu_read_unlock_bh();
 128
 129         return inst;
 130 }
 131
 132 static void nfulnl_instance_free_rcu(struct rcu_head *head)
 133 {
 134         struct nfulnl_instance *inst =
 135                 container_of(head, struct nfulnl_instance, rcu);
 136
 137         put_net(inst->net);
 138         kfree(inst);
 139         module_put(THIS_MODULE);
 140 }
 141
 142 static void
 143 instance_put(struct nfulnl_instance *inst)
 144 {
 145         if (inst && atomic_dec_and_test(&inst->use))
 146                 call_rcu_bh(&inst->rcu, nfulnl_instance_free_rcu);
 147 }
 148
 149 static void nfulnl_timer(unsigned long data);
 150
 151 static struct nfulnl_instance *
 152 instance_create(struct net *net, u_int16_t group_num,
 153                 int portid, struct user_namespace *user_ns)
 154 {
 155         struct nfulnl_instance *inst;
 156         struct nfnl_log_net *log = nfnl_log_pernet(net);
 157         int err;
 158
 159         spin_lock_bh(&log->instances_lock);
 160         if (__instance_lookup(log, group_num)) {
 161                 err = -EEXIST;
 162                 goto out_unlock;
 163         }
 164
 165         inst = kzalloc(sizeof(*inst), GFP_ATOMIC);
 166         if (!inst) {
 167                 err = -ENOMEM;
 168                 goto out_unlock;
 169         }
 170
 171         if (!try_module_get(THIS_MODULE)) {
 172                 kfree(inst);
 173                 err = -EAGAIN;
 174                 goto out_unlock;
 175         }
 176
 177         INIT_HLIST_NODE(&inst->hlist);
 178         spin_lock_init(&inst->lock);
 179         /* needs to be two, since we _put() after creation */
 180         atomic_set(&inst->use, 2);
 181
 182         setup_timer(&inst->timer, nfulnl_timer, (unsigned long)inst);
 183
 184         inst->net = get_net(net);
 185         inst->peer_user_ns = user_ns;
 186         inst->peer_portid = portid;
 187         inst->group_num = group_num;
 188
 189         inst->qthreshold        = NFULNL_QTHRESH_DEFAULT;
 190         inst->flushtimeout      = NFULNL_TIMEOUT_DEFAULT;
 191         inst->nlbufsiz          = NFULNL_NLBUFSIZ_DEFAULT;
 192         inst->copy_mode         = NFULNL_COPY_PACKET;
 193         inst->copy_range        = NFULNL_COPY_RANGE_MAX;
 194
 195         hlist_add_head_rcu(&inst->hlist,
 196                        &log->instance_table[instance_hashfn(group_num)]);
 197
 198
 199         spin_unlock_bh(&log->instances_lock);
 200
 201         return inst;
 202
 203 out_unlock:
 204         spin_unlock_bh(&log->instances_lock);
 205         return ERR_PTR(err);
 206 }
 207
 208 static void __nfulnl_flush(struct nfulnl_instance *inst);
 209
 210 /* called with BH disabled */
 211 static void
 212 __instance_destroy(struct nfulnl_instance *inst)
 213 {
 214         /* first pull it out of the global list */
 215         hlist_del_rcu(&inst->hlist);
 216
 217         /* then flush all pending packets from skb */
 218
 219         spin_lock(&inst->lock);
 220
 221         /* lockless readers wont be able to use us */
 222         inst->copy_mode = NFULNL_COPY_DISABLED;
 223
 224         if (inst->skb)
 225                 __nfulnl_flush(inst);
 226         spin_unlock(&inst->lock);
 227
 228         /* and finally put the refcount */
 229         instance_put(inst);
 230 }
 231
 232 static inline void
 233 instance_destroy(struct nfnl_log_net *log,
 234                  struct nfulnl_instance *inst)
 235 {
 236         spin_lock_bh(&log->instances_lock);
 237         __instance_destroy(inst);
 238         spin_unlock_bh(&log->instances_lock);
 239 }
 240
 241 static int
 242 nfulnl_set_mode(struct nfulnl_instance *inst, u_int8_t mode,
 243                   unsigned int range)
 244 {
 245         int status = 0;
 246
 247         spin_lock_bh(&inst->lock);
 248
 249         switch (mode) {
 250         case NFULNL_COPY_NONE:
 251         case NFULNL_COPY_META:
 252                 inst->copy_mode = mode;
 253                 inst->copy_range = 0;
 254                 break;
 255
 256         case NFULNL_COPY_PACKET:
 257                 inst->copy_mode = mode;
 258                 inst->copy_range = min_t(unsigned int,
 259                                          range, NFULNL_COPY_RANGE_MAX);
 260                 break;
 261
 262         default:
 263                 status = -EINVAL;
 264                 break;
 265         }
 266
 267         spin_unlock_bh(&inst->lock);
 268
 269         return status;
 270 }
 271
 272 static int
 273 nfulnl_set_nlbufsiz(struct nfulnl_instance *inst, u_int32_t nlbufsiz)
 274 {
 275         int status;
 276
 277         spin_lock_bh(&inst->lock);
 278         if (nlbufsiz < NFULNL_NLBUFSIZ_DEFAULT)
 279                 status = -ERANGE;
 280         else if (nlbufsiz > 131072)
 281                 status = -ERANGE;
 282         else {
 283                 inst->nlbufsiz = nlbufsiz;
 284                 status = 0;
 285         }
 286         spin_unlock_bh(&inst->lock);
 287
 288         return status;
 289 }
 290
 291 static int
 292 nfulnl_set_timeout(struct nfulnl_instance *inst, u_int32_t timeout)
 293 {
 294         spin_lock_bh(&inst->lock);
 295         inst->flushtimeout = timeout;
 296         spin_unlock_bh(&inst->lock);
 297
 298         return 0;
 299 }
 300
 301 static int
 302 nfulnl_set_qthresh(struct nfulnl_instance *inst, u_int32_t qthresh)
 303 {
 304         spin_lock_bh(&inst->lock);
 305         inst->qthreshold = qthresh;
 306         spin_unlock_bh(&inst->lock);
 307
 308         return 0;
 309 }
 310
 311 static int
 312 nfulnl_set_flags(struct nfulnl_instance *inst, u_int16_t flags)
 313 {
 314         spin_lock_bh(&inst->lock);
 315         inst->flags = flags;
 316         spin_unlock_bh(&inst->lock);
 317
 318         return 0;
 319 }
 320
 321 static struct sk_buff *
 322 nfulnl_alloc_skb(unsigned int inst_size, unsigned int pkt_size)
 323 {
 324         struct sk_buff *skb;
 325         unsigned int n;
 326
 327         /* alloc skb which should be big enough for a whole multipart
 328          * message.  WARNING: has to be <= 128k due to slab restrictions */
 329
 330         n = max(inst_size, pkt_size);
 331         skb = alloc_skb(n, GFP_ATOMIC);
 332         if (!skb) {
 333                 if (n > pkt_size) {
 334                         /* try to allocate only as much as we need for current
 335                          * packet */
 336
 337                         skb = alloc_skb(pkt_size, GFP_ATOMIC);
 338                         if (!skb)
 339                                 pr_err("nfnetlink_log: can't even alloc %u bytes\n",
 340                                        pkt_size);
 341                 }
 342         }
 343
 344         return skb;
 345 }
 346
 347 static int
 348 __nfulnl_send(struct nfulnl_instance *inst)
 349 {
 350         int status = -1;
 351
 352         if (inst->qlen > 1) {
 353                 struct nlmsghdr *nlh = nlmsg_put(inst->skb, 0, 0,
 354                                                  NLMSG_DONE,
 355                                                  sizeof(struct nfgenmsg),
 356                                                  0);
 357                 if (!nlh)
 358                         goto out;
 359         }
 360         status = nfnetlink_unicast(inst->skb, inst->net, inst->peer_portid,
 361                                    MSG_DONTWAIT);
 362
 363         inst->qlen = 0;
 364         inst->skb = NULL;
 365 out:
 366         return status;
 367 }
 368
 369 static void
 370 __nfulnl_flush(struct nfulnl_instance *inst)
 371 {
 372         /* timer holds a reference */
 373         if (del_timer(&inst->timer))
 374                 instance_put(inst);
 375         if (inst->skb)
 376                 __nfulnl_send(inst);
 377 }
 378
 379 static void
 380 nfulnl_timer(unsigned long data)
 381 {
 382         struct nfulnl_instance *inst = (struct nfulnl_instance *)data;
 383
 384         spin_lock_bh(&inst->lock);
 385         if (inst->skb)
 386                 __nfulnl_send(inst);
 387         spin_unlock_bh(&inst->lock);
 388         instance_put(inst);
 389 }
 390
 391 /* This is an inline function, we don't really care about a long
 392  * list of arguments */
 393 static inline int
 394 __build_packet_message(struct nfnl_log_net *log,
 395                         struct nfulnl_instance *inst,
 396                         const struct sk_buff *skb,
 397                         unsigned int data_len,
 398                         u_int8_t pf,
 399                         unsigned int hooknum,
 400                         const struct net_device *indev,
 401                         const struct net_device *outdev,
 402                         const char *prefix, unsigned int plen)
 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         pmsg.hw_protocol        = skb->protocol;
 422         pmsg.hook               = hooknum;
 423
 424         if (nla_put(inst->skb, NFULA_PACKET_HDR, sizeof(pmsg), &pmsg))
 425                 goto nla_put_failure;
 426
 427         if (prefix &&
 428             nla_put(inst->skb, NFULA_PREFIX, plen, prefix))
 429                 goto nla_put_failure;
 430
 431         if (indev) {
 432 #ifndef CONFIG_BRIDGE_NETFILTER
 433                 if (nla_put_be32(inst->skb, NFULA_IFINDEX_INDEV,
 434                                  htonl(indev->ifindex)))
 435                         goto nla_put_failure;
 436 #else
 437                 if (pf == PF_BRIDGE) {
 438                         /* Case 1: outdev is physical input device, we need to
 439                          * look for bridge group (when called from
 440                          * netfilter_bridge) */
 441                         if (nla_put_be32(inst->skb, NFULA_IFINDEX_PHYSINDEV,
 442                                          htonl(indev->ifindex)) ||
 443                         /* this is the bridge group "brX" */
 444                         /* rcu_read_lock()ed by nf_hook_slow or nf_log_packet */
 445                             nla_put_be32(inst->skb, NFULA_IFINDEX_INDEV,
 446                                          htonl(br_port_get_rcu(indev)->br->dev->ifindex)))
 447                                 goto nla_put_failure;
 448                 } else {
 449                         /* Case 2: indev is bridge group, we need to look for
 450                          * physical device (when called from ipv4) */
 451                         if (nla_put_be32(inst->skb, NFULA_IFINDEX_INDEV,
 452                                          htonl(indev->ifindex)))
 453                                 goto nla_put_failure;
 454                         if (skb->nf_bridge && skb->nf_bridge->physindev &&
 455                             nla_put_be32(inst->skb, NFULA_IFINDEX_PHYSINDEV,
 456                                          htonl(skb->nf_bridge->physindev->ifindex)))
 457                                 goto nla_put_failure;
 458                 }
 459 #endif
 460         }
 461
 462         if (outdev) {
 463 #ifndef CONFIG_BRIDGE_NETFILTER
 464                 if (nla_put_be32(inst->skb, NFULA_IFINDEX_OUTDEV,
 465                                  htonl(outdev->ifindex)))
 466                         goto nla_put_failure;
 467 #else
 468                 if (pf == PF_BRIDGE) {
 469                         /* Case 1: outdev is physical output device, we need to
 470                          * look for bridge group (when called from
 471                          * netfilter_bridge) */
 472                         if (nla_put_be32(inst->skb, NFULA_IFINDEX_PHYSOUTDEV,
 473                                          htonl(outdev->ifindex)) ||
 474                         /* this is the bridge group "brX" */
 475                         /* rcu_read_lock()ed by nf_hook_slow or nf_log_packet */
 476                             nla_put_be32(inst->skb, NFULA_IFINDEX_OUTDEV,
 477                                          htonl(br_port_get_rcu(outdev)->br->dev->ifindex)))
 478                                 goto nla_put_failure;
 479                 } else {
 480                         /* Case 2: indev is a bridge group, we need to look
 481                          * for physical device (when called from ipv4) */
 482                         if (nla_put_be32(inst->skb, NFULA_IFINDEX_OUTDEV,
 483                                          htonl(outdev->ifindex)))
 484                                 goto nla_put_failure;
 485                         if (skb->nf_bridge && skb->nf_bridge->physoutdev &&
 486                             nla_put_be32(inst->skb, NFULA_IFINDEX_PHYSOUTDEV,
 487                                          htonl(skb->nf_bridge->physoutdev->ifindex)))
 488                                 goto nla_put_failure;
 489                 }
 490 #endif
 491         }
 492
 493         if (skb->mark &&
 494             nla_put_be32(inst->skb, NFULA_MARK, htonl(skb->mark)))
 495                 goto nla_put_failure;
 496
 497         if (indev && skb->dev &&
 498             skb->mac_header != skb->network_header) {
 499                 struct nfulnl_msg_packet_hw phw;
 500                 int len = dev_parse_header(skb, phw.hw_addr);
 501                 if (len > 0) {
 502                         phw.hw_addrlen = htons(len);
 503                         if (nla_put(inst->skb, NFULA_HWADDR, sizeof(phw), &phw))
 504                                 goto nla_put_failure;
 505                 }
 506         }
 507
 508         if (indev && skb_mac_header_was_set(skb)) {
 509                 if (nla_put_be16(inst->skb, NFULA_HWTYPE, htons(skb->dev->type)) ||
 510                     nla_put_be16(inst->skb, NFULA_HWLEN,
 511                                  htons(skb->dev->hard_header_len)))
 512                         goto nla_put_failure;
 513
 514                 hwhdrp = skb_mac_header(skb);
 515
 516                 if (skb->dev->type == ARPHRD_SIT)
 517                         hwhdrp -= ETH_HLEN;
 518
 519                 if (hwhdrp >= skb->head &&
 520                     nla_put(inst->skb, NFULA_HWHEADER,
 521                             skb->dev->hard_header_len, hwhdrp))
 522                         goto nla_put_failure;
 523         }
 524
 525         if (skb->tstamp.tv64) {
 526                 struct nfulnl_msg_packet_timestamp ts;
 527                 struct timeval tv = ktime_to_timeval(skb->tstamp);
 528                 ts.sec = cpu_to_be64(tv.tv_sec);
 529                 ts.usec = cpu_to_be64(tv.tv_usec);
 530
 531                 if (nla_put(inst->skb, NFULA_TIMESTAMP, sizeof(ts), &ts))
 532                         goto nla_put_failure;
 533         }
 534
 535         /* UID */
 536         sk = skb->sk;
 537         if (sk && sk->sk_state != TCP_TIME_WAIT) {
 538                 read_lock_bh(&sk->sk_callback_lock);
 539                 if (sk->sk_socket && sk->sk_socket->file) {
 540                         struct file *file = sk->sk_socket->file;
 541                         const struct cred *cred = file->f_cred;
 542                         struct user_namespace *user_ns = inst->peer_user_ns;
 543                         __be32 uid = htonl(from_kuid_munged(user_ns, cred->fsuid));
 544                         __be32 gid = htonl(from_kgid_munged(user_ns, cred->fsgid));
 545                         read_unlock_bh(&sk->sk_callback_lock);
 546                         if (nla_put_be32(inst->skb, NFULA_UID, uid) ||
 547                             nla_put_be32(inst->skb, NFULA_GID, gid))
 548                                 goto nla_put_failure;
 549                 } else
 550                         read_unlock_bh(&sk->sk_callback_lock);
 551         }
 552
 553         /* local sequence number */
 554         if ((inst->flags & NFULNL_CFG_F_SEQ) &&
 555             nla_put_be32(inst->skb, NFULA_SEQ, htonl(inst->seq++)))
 556                 goto nla_put_failure;
 557
 558         /* global sequence number */
 559         if ((inst->flags & NFULNL_CFG_F_SEQ_GLOBAL) &&
 560             nla_put_be32(inst->skb, NFULA_SEQ_GLOBAL,
 561                          htonl(atomic_inc_return(&log->global_seq))))
 562                 goto nla_put_failure;
 563
 564         if (data_len) {
 565                 struct nlattr *nla;
 566                 int size = nla_attr_size(data_len);
 567
 568                 if (skb_tailroom(inst->skb) < nla_total_size(data_len)) {
 569                         printk(KERN_WARNING "nfnetlink_log: no tailroom!\n");
 570                         return -1;
 571                 }
 572
 573                 nla = (struct nlattr *)skb_put(inst->skb, nla_total_size(data_len));
 574                 nla->nla_type = NFULA_PAYLOAD;
 575                 nla->nla_len = size;
 576
 577                 if (skb_copy_bits(skb, 0, nla_data(nla), data_len))
 578                         BUG();
 579         }
 580
 581         nlh->nlmsg_len = inst->skb->tail - old_tail;
 582         return 0;
 583
 584 nla_put_failure:
 585         PRINTR(KERN_ERR "nfnetlink_log: error creating log nlmsg\n");
 586         return -1;
 587 }
 588
 589 #define RCV_SKB_FAIL(err) do { netlink_ack(skb, nlh, (err)); return; } while (0)
 590
 591 static struct nf_loginfo default_loginfo = {
 592         .type =         NF_LOG_TYPE_ULOG,
 593         .u = {
 594                 .ulog = {
 595                         .copy_len       = 0xffff,
 596                         .group          = 0,
 597                         .qthreshold     = 1,
 598                 },
 599         },
 600 };
 601
 602 /* log handler for internal netfilter logging api */
 603 void
 604 nfulnl_log_packet(u_int8_t pf,
 605                   unsigned int hooknum,
 606                   const struct sk_buff *skb,
 607                   const struct net_device *in,
 608                   const struct net_device *out,
 609                   const struct nf_loginfo *li_user,
 610                   const char *prefix)
 611 {
 612         unsigned int size, data_len;
 613         struct nfulnl_instance *inst;
 614         const struct nf_loginfo *li;
 615         unsigned int qthreshold;
 616         unsigned int plen;
 617         struct net *net = dev_net(in ? in : out);
 618         struct nfnl_log_net *log = nfnl_log_pernet(net);
 619
 620         if (li_user && li_user->type == NF_LOG_TYPE_ULOG)
 621                 li = li_user;
 622         else
 623                 li = &default_loginfo;
 624
 625         inst = instance_lookup_get(log, li->u.ulog.group);
 626         if (!inst)
 627                 return;
 628
 629         plen = 0;
 630         if (prefix)
 631                 plen = strlen(prefix) + 1;
 632
 633         /* FIXME: do we want to make the size calculation conditional based on
 634          * what is actually present?  way more branches and checks, but more
 635          * memory efficient... */
 636         size =    nlmsg_total_size(sizeof(struct nfgenmsg))
 637                 + nla_total_size(sizeof(struct nfulnl_msg_packet_hdr))
 638                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
 639                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
 640 #ifdef CONFIG_BRIDGE_NETFILTER
 641                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
 642                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
 643 #endif
 644                 + nla_total_size(sizeof(u_int32_t))     /* mark */
 645                 + nla_total_size(sizeof(u_int32_t))     /* uid */
 646                 + nla_total_size(sizeof(u_int32_t))     /* gid */
 647                 + nla_total_size(plen)                  /* prefix */
 648                 + nla_total_size(sizeof(struct nfulnl_msg_packet_hw))
 649                 + nla_total_size(sizeof(struct nfulnl_msg_packet_timestamp));
 650
 651         if (in && skb_mac_header_was_set(skb)) {
 652                 size +=   nla_total_size(skb->dev->hard_header_len)
 653                         + nla_total_size(sizeof(u_int16_t))     /* hwtype */
 654                         + nla_total_size(sizeof(u_int16_t));    /* hwlen */
 655         }
 656
 657         spin_lock_bh(&inst->lock);
 658
 659         if (inst->flags & NFULNL_CFG_F_SEQ)
 660                 size += nla_total_size(sizeof(u_int32_t));
 661         if (inst->flags & NFULNL_CFG_F_SEQ_GLOBAL)
 662                 size += nla_total_size(sizeof(u_int32_t));
 663
 664         qthreshold = inst->qthreshold;
 665         /* per-rule qthreshold overrides per-instance */
 666         if (li->u.ulog.qthreshold)
 667                 if (qthreshold > li->u.ulog.qthreshold)
 668                         qthreshold = li->u.ulog.qthreshold;
 669
 670
 671         switch (inst->copy_mode) {
 672         case NFULNL_COPY_META:
 673         case NFULNL_COPY_NONE:
 674                 data_len = 0;
 675                 break;
 676
 677         case NFULNL_COPY_PACKET:
 678                 if (inst->copy_range == 0
 679                     || inst->copy_range > skb->len)
 680                         data_len = skb->len;
 681                 else
 682                         data_len = inst->copy_range;
 683
 684                 size += nla_total_size(data_len);
 685                 break;
 686
 687         case NFULNL_COPY_DISABLED:
 688         default:
 689                 goto unlock_and_release;
 690         }
 691
 692         if (inst->skb &&
 693             size > skb_tailroom(inst->skb) - sizeof(struct nfgenmsg)) {
 694                 /* either the queue len is too high or we don't have
 695                  * enough room in the skb left. flush to userspace. */
 696                 __nfulnl_flush(inst);
 697         }
 698
 699         if (!inst->skb) {
 700                 inst->skb = nfulnl_alloc_skb(inst->nlbufsiz, size);
 701                 if (!inst->skb)
 702                         goto alloc_failure;
 703         }
 704
 705         inst->qlen++;
 706
 707         __build_packet_message(log, inst, skb, data_len, pf,
 708                                 hooknum, in, out, prefix, plen);
 709
 710         if (inst->qlen >= qthreshold)
 711                 __nfulnl_flush(inst);
 712         /* timer_pending always called within inst->lock, so there
 713          * is no chance of a race here */
 714         else if (!timer_pending(&inst->timer)) {
 715                 instance_get(inst);
 716                 inst->timer.expires = jiffies + (inst->flushtimeout*HZ/100);
 717                 add_timer(&inst->timer);
 718         }
 719
 720 unlock_and_release:
 721         spin_unlock_bh(&inst->lock);
 722         instance_put(inst);
 723         return;
 724
 725 alloc_failure:
 726         /* FIXME: statistics */
 727         goto unlock_and_release;
 728 }
 729 EXPORT_SYMBOL_GPL(nfulnl_log_packet);
 730
 731 static int
 732 nfulnl_rcv_nl_event(struct notifier_block *this,
 733                    unsigned long event, void *ptr)
 734 {
 735         struct netlink_notify *n = ptr;
 736         struct nfnl_log_net *log = nfnl_log_pernet(n->net);
 737
 738         if (event == NETLINK_URELEASE && n->protocol == NETLINK_NETFILTER) {
 739                 int i;
 740
 741                 /* destroy all instances for this portid */
 742                 spin_lock_bh(&log->instances_lock);
 743                 for  (i = 0; i < INSTANCE_BUCKETS; i++) {
 744                         struct hlist_node *t2;
 745                         struct nfulnl_instance *inst;
 746                         struct hlist_head *head = &log->instance_table[i];
 747
 748                         hlist_for_each_entry_safe(inst, t2, head, hlist) {
 749                                 if (n->portid == inst->peer_portid)
 750                                         __instance_destroy(inst);
 751                         }
 752                 }
 753                 spin_unlock_bh(&log->instances_lock);
 754         }
 755         return NOTIFY_DONE;
 756 }
 757
 758 static struct notifier_block nfulnl_rtnl_notifier = {
 759         .notifier_call  = nfulnl_rcv_nl_event,
 760 };
 761
 762 static int
 763 nfulnl_recv_unsupp(struct sock *ctnl, struct sk_buff *skb,
 764                    const struct nlmsghdr *nlh,
 765                    const struct nlattr * const nfqa[])
 766 {
 767         return -ENOTSUPP;
 768 }
 769
 770 static struct nf_logger nfulnl_logger __read_mostly = {
 771         .name   = "nfnetlink_log",
 772         .logfn  = &nfulnl_log_packet,
 773         .me     = THIS_MODULE,
 774 };
 775
 776 static const struct nla_policy nfula_cfg_policy[NFULA_CFG_MAX+1] = {
 777         [NFULA_CFG_CMD]         = { .len = sizeof(struct nfulnl_msg_config_cmd) },
 778         [NFULA_CFG_MODE]        = { .len = sizeof(struct nfulnl_msg_config_mode) },
 779         [NFULA_CFG_TIMEOUT]     = { .type = NLA_U32 },
 780         [NFULA_CFG_QTHRESH]     = { .type = NLA_U32 },
 781         [NFULA_CFG_NLBUFSIZ]    = { .type = NLA_U32 },
 782         [NFULA_CFG_FLAGS]       = { .type = NLA_U16 },
 783 };
 784
 785 static int
 786 nfulnl_recv_config(struct sock *ctnl, struct sk_buff *skb,
 787                    const struct nlmsghdr *nlh,
 788                    const struct nlattr * const nfula[])
 789 {
 790         struct nfgenmsg *nfmsg = nlmsg_data(nlh);
 791         u_int16_t group_num = ntohs(nfmsg->res_id);
 792         struct nfulnl_instance *inst;
 793         struct nfulnl_msg_config_cmd *cmd = NULL;
 794         struct net *net = sock_net(ctnl);
 795         struct nfnl_log_net *log = nfnl_log_pernet(net);
 796         int ret = 0;
 797
 798         if (nfula[NFULA_CFG_CMD]) {
 799                 u_int8_t pf = nfmsg->nfgen_family;
 800                 cmd = nla_data(nfula[NFULA_CFG_CMD]);
 801
 802                 /* Commands without queue context */
 803                 switch (cmd->command) {
 804                 case NFULNL_CFG_CMD_PF_BIND:
 805                         return nf_log_bind_pf(net, pf, &nfulnl_logger);
 806                 case NFULNL_CFG_CMD_PF_UNBIND:
 807                         nf_log_unbind_pf(net, pf);
 808                         return 0;
 809                 }
 810         }
 811
 812         inst = instance_lookup_get(log, group_num);
 813         if (inst && inst->peer_portid != NETLINK_CB(skb).portid) {
 814                 ret = -EPERM;
 815                 goto out_put;
 816         }
 817
 818         if (cmd != NULL) {
 819                 switch (cmd->command) {
 820                 case NFULNL_CFG_CMD_BIND:
 821                         if (inst) {
 822                                 ret = -EBUSY;
 823                                 goto out_put;
 824                         }
 825
 826                         inst = instance_create(net, group_num,
 827                                                NETLINK_CB(skb).portid,
 828                                                sk_user_ns(NETLINK_CB(skb).ssk));
 829                         if (IS_ERR(inst)) {
 830                                 ret = PTR_ERR(inst);
 831                                 goto out;
 832                         }
 833                         break;
 834                 case NFULNL_CFG_CMD_UNBIND:
 835                         if (!inst) {
 836                                 ret = -ENODEV;
 837                                 goto out;
 838                         }
 839
 840                         instance_destroy(log, inst);
 841                         goto out_put;
 842                 default:
 843                         ret = -ENOTSUPP;
 844                         break;
 845                 }
 846         }
 847
 848         if (nfula[NFULA_CFG_MODE]) {
 849                 struct nfulnl_msg_config_mode *params;
 850                 params = nla_data(nfula[NFULA_CFG_MODE]);
 851
 852                 if (!inst) {
 853                         ret = -ENODEV;
 854                         goto out;
 855                 }
 856                 nfulnl_set_mode(inst, params->copy_mode,
 857                                 ntohl(params->copy_range));
 858         }
 859
 860         if (nfula[NFULA_CFG_TIMEOUT]) {
 861                 __be32 timeout = nla_get_be32(nfula[NFULA_CFG_TIMEOUT]);
 862
 863                 if (!inst) {
 864                         ret = -ENODEV;
 865                         goto out;
 866                 }
 867                 nfulnl_set_timeout(inst, ntohl(timeout));
 868         }
 869
 870         if (nfula[NFULA_CFG_NLBUFSIZ]) {
 871                 __be32 nlbufsiz = nla_get_be32(nfula[NFULA_CFG_NLBUFSIZ]);
 872
 873                 if (!inst) {
 874                         ret = -ENODEV;
 875                         goto out;
 876                 }
 877                 nfulnl_set_nlbufsiz(inst, ntohl(nlbufsiz));
 878         }
 879
 880         if (nfula[NFULA_CFG_QTHRESH]) {
 881                 __be32 qthresh = nla_get_be32(nfula[NFULA_CFG_QTHRESH]);
 882
 883                 if (!inst) {
 884                         ret = -ENODEV;
 885                         goto out;
 886                 }
 887                 nfulnl_set_qthresh(inst, ntohl(qthresh));
 888         }
 889
 890         if (nfula[NFULA_CFG_FLAGS]) {
 891                 __be16 flags = nla_get_be16(nfula[NFULA_CFG_FLAGS]);
 892
 893                 if (!inst) {
 894                         ret = -ENODEV;
 895                         goto out;
 896                 }
 897                 nfulnl_set_flags(inst, ntohs(flags));
 898         }
 899
 900 out_put:
 901         instance_put(inst);
 902 out:
 903         return ret;
 904 }
 905
 906 static const struct nfnl_callback nfulnl_cb[NFULNL_MSG_MAX] = {
 907         [NFULNL_MSG_PACKET]     = { .call = nfulnl_recv_unsupp,
 908                                     .attr_count = NFULA_MAX, },
 909         [NFULNL_MSG_CONFIG]     = { .call = nfulnl_recv_config,
 910                                     .attr_count = NFULA_CFG_MAX,
 911                                     .policy = nfula_cfg_policy },
 912 };
 913
 914 static const struct nfnetlink_subsystem nfulnl_subsys = {
 915         .name           = "log",
 916         .subsys_id      = NFNL_SUBSYS_ULOG,
 917         .cb_count       = NFULNL_MSG_MAX,
 918         .cb             = nfulnl_cb,
 919 };
 920
 921 #ifdef CONFIG_PROC_FS
 922 struct iter_state {
 923         struct seq_net_private p;
 924         unsigned int bucket;
 925 };
 926
 927 static struct hlist_node *get_first(struct net *net, struct iter_state *st)
 928 {
 929         struct nfnl_log_net *log;
 930         if (!st)
 931                 return NULL;
 932
 933         log = nfnl_log_pernet(net);
 934
 935         for (st->bucket = 0; st->bucket < INSTANCE_BUCKETS; st->bucket++) {
 936                 struct hlist_head *head = &log->instance_table[st->bucket];
 937
 938                 if (!hlist_empty(head))
 939                         return rcu_dereference_bh(hlist_first_rcu(head));
 940         }
 941         return NULL;
 942 }
 943
 944 static struct hlist_node *get_next(struct net *net, struct iter_state *st,
 945                                    struct hlist_node *h)
 946 {
 947         h = rcu_dereference_bh(hlist_next_rcu(h));
 948         while (!h) {
 949                 struct nfnl_log_net *log;
 950                 struct hlist_head *head;
 951
 952                 if (++st->bucket >= INSTANCE_BUCKETS)
 953                         return NULL;
 954
 955                 log = nfnl_log_pernet(net);
 956                 head = &log->instance_table[st->bucket];
 957                 h = rcu_dereference_bh(hlist_first_rcu(head));
 958         }
 959         return h;
 960 }
 961
 962 static struct hlist_node *get_idx(struct net *net, struct iter_state *st,
 963                                   loff_t pos)
 964 {
 965         struct hlist_node *head;
 966         head = get_first(net, st);
 967
 968         if (head)
 969                 while (pos && (head = get_next(net, st, head)))
 970                         pos--;
 971         return pos ? NULL : head;
 972 }
 973
 974 static void *seq_start(struct seq_file *s, loff_t *pos)
 975         __acquires(rcu_bh)
 976 {
 977         rcu_read_lock_bh();
 978         return get_idx(seq_file_net(s), s->private, *pos);
 979 }
 980
 981 static void *seq_next(struct seq_file *s, void *v, loff_t *pos)
 982 {
 983         (*pos)++;
 984         return get_next(seq_file_net(s), s->private, v);
 985 }
 986
 987 static void seq_stop(struct seq_file *s, void *v)
 988         __releases(rcu_bh)
 989 {
 990         rcu_read_unlock_bh();
 991 }
 992
 993 static int seq_show(struct seq_file *s, void *v)
 994 {
 995         const struct nfulnl_instance *inst = v;
 996
 997         return seq_printf(s, "%5d %6d %5d %1d %5d %6d %2d\n",
 998                           inst->group_num,
 999                           inst->peer_portid, inst->qlen,
1000                           inst->copy_mode, inst->copy_range,
1001                           inst->flushtimeout, atomic_read(&inst->use));
1002 }
1003
1004 static const struct seq_operations nful_seq_ops = {
1005         .start  = seq_start,
1006         .next   = seq_next,
1007         .stop   = seq_stop,
1008         .show   = seq_show,
1009 };
1010
1011 static int nful_open(struct inode *inode, struct file *file)
1012 {
1013         return seq_open_net(inode, file, &nful_seq_ops,
1014                             sizeof(struct iter_state));
1015 }
1016
1017 static const struct file_operations nful_file_ops = {
1018         .owner   = THIS_MODULE,
1019         .open    = nful_open,
1020         .read    = seq_read,
1021         .llseek  = seq_lseek,
1022         .release = seq_release_net,
1023 };
1024
1025 #endif /* PROC_FS */
1026
1027 static int __net_init nfnl_log_net_init(struct net *net)
1028 {
1029         unsigned int i;
1030         struct nfnl_log_net *log = nfnl_log_pernet(net);
1031
1032         for (i = 0; i < INSTANCE_BUCKETS; i++)
1033                 INIT_HLIST_HEAD(&log->instance_table[i]);
1034         spin_lock_init(&log->instances_lock);
1035
1036 #ifdef CONFIG_PROC_FS
1037         if (!proc_create("nfnetlink_log", 0440,
1038                          net->nf.proc_netfilter, &nful_file_ops))
1039                 return -ENOMEM;
1040 #endif
1041         return 0;
1042 }
1043
1044 static void __net_exit nfnl_log_net_exit(struct net *net)
1045 {
1046         remove_proc_entry("nfnetlink_log", net->nf.proc_netfilter);
1047 }
1048
1049 static struct pernet_operations nfnl_log_net_ops = {
1050         .init   = nfnl_log_net_init,
1051         .exit   = nfnl_log_net_exit,
1052         .id     = &nfnl_log_net_id,
1053         .size   = sizeof(struct nfnl_log_net),
1054 };
1055
1056 static int __init nfnetlink_log_init(void)
1057 {
1058         int status = -ENOMEM;
1059
1060         /* it's not really all that important to have a random value, so
1061          * we can do this from the init function, even if there hasn't
1062          * been that much entropy yet */
1063         get_random_bytes(&hash_init, sizeof(hash_init));
1064
1065         netlink_register_notifier(&nfulnl_rtnl_notifier);
1066         status = nfnetlink_subsys_register(&nfulnl_subsys);
1067         if (status < 0) {
1068                 pr_err("log: failed to create netlink socket\n");
1069                 goto cleanup_netlink_notifier;
1070         }
1071
1072         status = nf_log_register(NFPROTO_UNSPEC, &nfulnl_logger);
1073         if (status < 0) {
1074                 pr_err("log: failed to register logger\n");
1075                 goto cleanup_subsys;
1076         }
1077
1078         status = register_pernet_subsys(&nfnl_log_net_ops);
1079         if (status < 0) {
1080                 pr_err("log: failed to register pernet ops\n");
1081                 goto cleanup_logger;
1082         }
1083         return status;
1084
1085 cleanup_logger:
1086         nf_log_unregister(&nfulnl_logger);
1087 cleanup_subsys:
1088         nfnetlink_subsys_unregister(&nfulnl_subsys);
1089 cleanup_netlink_notifier:
1090         netlink_unregister_notifier(&nfulnl_rtnl_notifier);
1091         return status;
1092 }
1093
1094 static void __exit nfnetlink_log_fini(void)
1095 {
1096         unregister_pernet_subsys(&nfnl_log_net_ops);
1097         nf_log_unregister(&nfulnl_logger);
1098         nfnetlink_subsys_unregister(&nfulnl_subsys);
1099         netlink_unregister_notifier(&nfulnl_rtnl_notifier);
1100 }
1101
1102 MODULE_DESCRIPTION("netfilter userspace logging");
1103 MODULE_AUTHOR("Harald Welte <laforge@netfilter.org>");
1104 MODULE_LICENSE("GPL");
1105 MODULE_ALIAS_NFNL_SUBSYS(NFNL_SUBSYS_ULOG);
1106
1107 module_init(nfnetlink_log_init);
1108 module_exit(nfnetlink_log_fini);