git.oblomov.eu Git - linux-2.6/blob - 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  *
   7  * Based on the old ipv4-only ipt_ULOG.c:
   8  * (C) 2000-2004 by Harald Welte <laforge@netfilter.org>
   9  *
  10  * This program is free software; you can redistribute it and/or modify
  11  * it under the terms of the GNU General Public License version 2 as
  12  * published by the Free Software Foundation.
  13  */
  14 #include <linux/module.h>
  15 #include <linux/skbuff.h>
  16 #include <linux/init.h>
  17 #include <linux/ip.h>
  18 #include <linux/ipv6.h>
  19 #include <linux/netdevice.h>
  20 #include <linux/netfilter.h>
  21 #include <linux/netlink.h>
  22 #include <linux/netfilter/nfnetlink.h>
  23 #include <linux/netfilter/nfnetlink_log.h>
  24 #include <linux/spinlock.h>
  25 #include <linux/sysctl.h>
  26 #include <linux/proc_fs.h>
  27 #include <linux/security.h>
  28 #include <linux/list.h>
  29 #include <linux/jhash.h>
  30 #include <linux/random.h>
  31 #include <net/sock.h>
  32
  33 #include <asm/atomic.h>
  34
  35 #ifdef CONFIG_BRIDGE_NETFILTER
  36 #include "../bridge/br_private.h"
  37 #endif
  38
  39 #define NFULNL_NLBUFSIZ_DEFAULT NLMSG_GOODSIZE
  40 #define NFULNL_TIMEOUT_DEFAULT  HZ      /* every second */
  41 #define NFULNL_QTHRESH_DEFAULT  100     /* 100 packets */
  42 #define NFULNL_COPY_RANGE_MAX   0xFFFF  /* max packet size is limited by 16-bit struct nfattr nfa_len field */
  43
  44 #define PRINTR(x, args...)      do { if (net_ratelimit()) \
  45                                      printk(x, ## args); } while (0);
  46
  47 #if 0
  48 #define UDEBUG(x, args ...)     printk(KERN_DEBUG "%s(%d):%s(): " x,       \
  49                                         __FILE__, __LINE__, __FUNCTION__,  \
  50                                         ## args)
  51 #else
  52 #define UDEBUG(x, ...)
  53 #endif
  54
  55 struct nfulnl_instance {
  56         struct hlist_node hlist;        /* global list of instances */
  57         spinlock_t lock;
  58         atomic_t use;                   /* use count */
  59
  60         unsigned int qlen;              /* number of nlmsgs in skb */
  61         struct sk_buff *skb;            /* pre-allocatd skb */
  62         struct timer_list timer;
  63         int peer_pid;                   /* PID 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 };
  75
  76 static DEFINE_RWLOCK(instances_lock);
  77 static atomic_t global_seq;
  78
  79 #define INSTANCE_BUCKETS        16
  80 static struct hlist_head instance_table[INSTANCE_BUCKETS];
  81 static unsigned int hash_init;
  82
  83 static inline u_int8_t instance_hashfn(u_int16_t group_num)
  84 {
  85         return ((group_num & 0xff) % INSTANCE_BUCKETS);
  86 }
  87
  88 static struct nfulnl_instance *
  89 __instance_lookup(u_int16_t group_num)
  90 {
  91         struct hlist_head *head;
  92         struct hlist_node *pos;
  93         struct nfulnl_instance *inst;
  94
  95         UDEBUG("entering (group_num=%u)\n", group_num);
  96
  97         head = &instance_table[instance_hashfn(group_num)];
  98         hlist_for_each_entry(inst, pos, head, hlist) {
  99                 if (inst->group_num == group_num)
 100                         return inst;
 101         }
 102         return NULL;
 103 }
 104
 105 static inline void
 106 instance_get(struct nfulnl_instance *inst)
 107 {
 108         atomic_inc(&inst->use);
 109 }
 110
 111 static struct nfulnl_instance *
 112 instance_lookup_get(u_int16_t group_num)
 113 {
 114         struct nfulnl_instance *inst;
 115
 116         read_lock_bh(&instances_lock);
 117         inst = __instance_lookup(group_num);
 118         if (inst)
 119                 instance_get(inst);
 120         read_unlock_bh(&instances_lock);
 121
 122         return inst;
 123 }
 124
 125 static void
 126 instance_put(struct nfulnl_instance *inst)
 127 {
 128         if (inst && atomic_dec_and_test(&inst->use)) {
 129                 UDEBUG("kfree(inst=%p)\n", inst);
 130                 kfree(inst);
 131                 module_put(THIS_MODULE);
 132         }
 133 }
 134
 135 static void nfulnl_timer(unsigned long data);
 136
 137 static struct nfulnl_instance *
 138 instance_create(u_int16_t group_num, int pid)
 139 {
 140         struct nfulnl_instance *inst;
 141
 142         UDEBUG("entering (group_num=%u, pid=%d)\n", group_num,
 143                 pid);
 144
 145         write_lock_bh(&instances_lock);
 146         if (__instance_lookup(group_num)) {
 147                 inst = NULL;
 148                 UDEBUG("aborting, instance already exists\n");
 149                 goto out_unlock;
 150         }
 151
 152         inst = kzalloc(sizeof(*inst), GFP_ATOMIC);
 153         if (!inst)
 154                 goto out_unlock;
 155
 156         if (!try_module_get(THIS_MODULE)) {
 157                 kfree(inst);
 158                 goto out_unlock;
 159         }
 160
 161         INIT_HLIST_NODE(&inst->hlist);
 162         spin_lock_init(&inst->lock);
 163         /* needs to be two, since we _put() after creation */
 164         atomic_set(&inst->use, 2);
 165
 166         setup_timer(&inst->timer, nfulnl_timer, (unsigned long)inst);
 167
 168         inst->peer_pid = pid;
 169         inst->group_num = group_num;
 170
 171         inst->qthreshold        = NFULNL_QTHRESH_DEFAULT;
 172         inst->flushtimeout      = NFULNL_TIMEOUT_DEFAULT;
 173         inst->nlbufsiz          = NFULNL_NLBUFSIZ_DEFAULT;
 174         inst->copy_mode         = NFULNL_COPY_PACKET;
 175         inst->copy_range        = NFULNL_COPY_RANGE_MAX;
 176
 177         hlist_add_head(&inst->hlist,
 178                        &instance_table[instance_hashfn(group_num)]);
 179
 180         UDEBUG("newly added node: %p, next=%p\n", &inst->hlist,
 181                 inst->hlist.next);
 182
 183         write_unlock_bh(&instances_lock);
 184
 185         return inst;
 186
 187 out_unlock:
 188         write_unlock_bh(&instances_lock);
 189         return NULL;
 190 }
 191
 192 static void __nfulnl_flush(struct nfulnl_instance *inst);
 193
 194 static void
 195 __instance_destroy(struct nfulnl_instance *inst)
 196 {
 197         /* first pull it out of the global list */
 198         UDEBUG("removing instance %p (queuenum=%u) from hash\n",
 199                 inst, inst->group_num);
 200
 201         hlist_del(&inst->hlist);
 202
 203         /* then flush all pending packets from skb */
 204
 205         spin_lock_bh(&inst->lock);
 206         if (inst->skb)
 207                 __nfulnl_flush(inst);
 208         spin_unlock_bh(&inst->lock);
 209
 210         /* and finally put the refcount */
 211         instance_put(inst);
 212 }
 213
 214 static inline void
 215 instance_destroy(struct nfulnl_instance *inst)
 216 {
 217         write_lock_bh(&instances_lock);
 218         __instance_destroy(inst);
 219         write_unlock_bh(&instances_lock);
 220 }
 221
 222 static int
 223 nfulnl_set_mode(struct nfulnl_instance *inst, u_int8_t mode,
 224                   unsigned int range)
 225 {
 226         int status = 0;
 227
 228         spin_lock_bh(&inst->lock);
 229
 230         switch (mode) {
 231         case NFULNL_COPY_NONE:
 232         case NFULNL_COPY_META:
 233                 inst->copy_mode = mode;
 234                 inst->copy_range = 0;
 235                 break;
 236
 237         case NFULNL_COPY_PACKET:
 238                 inst->copy_mode = mode;
 239                 inst->copy_range = min_t(unsigned int,
 240                                          range, NFULNL_COPY_RANGE_MAX);
 241                 break;
 242
 243         default:
 244                 status = -EINVAL;
 245                 break;
 246         }
 247
 248         spin_unlock_bh(&inst->lock);
 249
 250         return status;
 251 }
 252
 253 static int
 254 nfulnl_set_nlbufsiz(struct nfulnl_instance *inst, u_int32_t nlbufsiz)
 255 {
 256         int status;
 257
 258         spin_lock_bh(&inst->lock);
 259         if (nlbufsiz < NFULNL_NLBUFSIZ_DEFAULT)
 260                 status = -ERANGE;
 261         else if (nlbufsiz > 131072)
 262                 status = -ERANGE;
 263         else {
 264                 inst->nlbufsiz = nlbufsiz;
 265                 status = 0;
 266         }
 267         spin_unlock_bh(&inst->lock);
 268
 269         return status;
 270 }
 271
 272 static int
 273 nfulnl_set_timeout(struct nfulnl_instance *inst, u_int32_t timeout)
 274 {
 275         spin_lock_bh(&inst->lock);
 276         inst->flushtimeout = timeout;
 277         spin_unlock_bh(&inst->lock);
 278
 279         return 0;
 280 }
 281
 282 static int
 283 nfulnl_set_qthresh(struct nfulnl_instance *inst, u_int32_t qthresh)
 284 {
 285         spin_lock_bh(&inst->lock);
 286         inst->qthreshold = qthresh;
 287         spin_unlock_bh(&inst->lock);
 288
 289         return 0;
 290 }
 291
 292 static int
 293 nfulnl_set_flags(struct nfulnl_instance *inst, u_int16_t flags)
 294 {
 295         spin_lock_bh(&inst->lock);
 296         inst->flags = flags;
 297         spin_unlock_bh(&inst->lock);
 298
 299         return 0;
 300 }
 301
 302 static struct sk_buff *
 303 nfulnl_alloc_skb(unsigned int inst_size, unsigned int pkt_size)
 304 {
 305         struct sk_buff *skb;
 306         unsigned int n;
 307
 308         UDEBUG("entered (%u, %u)\n", inst_size, pkt_size);
 309
 310         /* alloc skb which should be big enough for a whole multipart
 311          * message.  WARNING: has to be <= 128k due to slab restrictions */
 312
 313         n = max(inst_size, pkt_size);
 314         skb = alloc_skb(n, GFP_ATOMIC);
 315         if (!skb) {
 316                 PRINTR("nfnetlink_log: can't alloc whole buffer (%u bytes)\n",
 317                         inst_size);
 318
 319                 if (n > pkt_size) {
 320                         /* try to allocate only as much as we need for current
 321                          * packet */
 322
 323                         skb = alloc_skb(pkt_size, GFP_ATOMIC);
 324                         if (!skb)
 325                                 PRINTR("nfnetlink_log: can't even alloc %u "
 326                                        "bytes\n", pkt_size);
 327                 }
 328         }
 329
 330         return skb;
 331 }
 332
 333 static int
 334 __nfulnl_send(struct nfulnl_instance *inst)
 335 {
 336         int status = -1;
 337
 338         if (inst->qlen > 1)
 339                 NLMSG_PUT(inst->skb, 0, 0,
 340                           NLMSG_DONE,
 341                           sizeof(struct nfgenmsg));
 342
 343         status = nfnetlink_unicast(inst->skb, inst->peer_pid, MSG_DONTWAIT);
 344         if (status < 0) {
 345                 UDEBUG("netlink_unicast() failed\n");
 346                 /* FIXME: statistics */
 347         }
 348
 349         inst->qlen = 0;
 350         inst->skb = NULL;
 351
 352 nlmsg_failure:
 353         return status;
 354 }
 355
 356 static void
 357 __nfulnl_flush(struct nfulnl_instance *inst)
 358 {
 359         /* timer holds a reference */
 360         if (del_timer(&inst->timer))
 361                 instance_put(inst);
 362         if (inst->skb)
 363                 __nfulnl_send(inst);
 364 }
 365
 366 static void
 367 nfulnl_timer(unsigned long data)
 368 {
 369         struct nfulnl_instance *inst = (struct nfulnl_instance *)data;
 370
 371         UDEBUG("timer function called, flushing buffer\n");
 372
 373         spin_lock_bh(&inst->lock);
 374         if (inst->skb)
 375                 __nfulnl_send(inst);
 376         spin_unlock_bh(&inst->lock);
 377         instance_put(inst);
 378 }
 379
 380 /* This is an inline function, we don't really care about a long
 381  * list of arguments */
 382 static inline int
 383 __build_packet_message(struct nfulnl_instance *inst,
 384                         const struct sk_buff *skb,
 385                         unsigned int data_len,
 386                         unsigned int pf,
 387                         unsigned int hooknum,
 388                         const struct net_device *indev,
 389                         const struct net_device *outdev,
 390                         const struct nf_loginfo *li,
 391                         const char *prefix, unsigned int plen)
 392 {
 393         struct nfulnl_msg_packet_hdr pmsg;
 394         struct nlmsghdr *nlh;
 395         struct nfgenmsg *nfmsg;
 396         __be32 tmp_uint;
 397         sk_buff_data_t old_tail = inst->skb->tail;
 398
 399         UDEBUG("entered\n");
 400
 401         nlh = NLMSG_PUT(inst->skb, 0, 0,
 402                         NFNL_SUBSYS_ULOG << 8 | NFULNL_MSG_PACKET,
 403                         sizeof(struct nfgenmsg));
 404         nfmsg = NLMSG_DATA(nlh);
 405         nfmsg->nfgen_family = pf;
 406         nfmsg->version = NFNETLINK_V0;
 407         nfmsg->res_id = htons(inst->group_num);
 408
 409         pmsg.hw_protocol        = skb->protocol;
 410         pmsg.hook               = hooknum;
 411
 412         NLA_PUT(inst->skb, NFULA_PACKET_HDR, sizeof(pmsg), &pmsg);
 413
 414         if (prefix)
 415                 NLA_PUT(inst->skb, NFULA_PREFIX, plen, prefix);
 416
 417         if (indev) {
 418                 tmp_uint = htonl(indev->ifindex);
 419 #ifndef CONFIG_BRIDGE_NETFILTER
 420                 NLA_PUT(inst->skb, NFULA_IFINDEX_INDEV, sizeof(tmp_uint),
 421                         &tmp_uint);
 422 #else
 423                 if (pf == PF_BRIDGE) {
 424                         /* Case 1: outdev is physical input device, we need to
 425                          * look for bridge group (when called from
 426                          * netfilter_bridge) */
 427                         NLA_PUT(inst->skb, NFULA_IFINDEX_PHYSINDEV,
 428                                 sizeof(tmp_uint), &tmp_uint);
 429                         /* this is the bridge group "brX" */
 430                         tmp_uint = htonl(indev->br_port->br->dev->ifindex);
 431                         NLA_PUT(inst->skb, NFULA_IFINDEX_INDEV,
 432                                 sizeof(tmp_uint), &tmp_uint);
 433                 } else {
 434                         /* Case 2: indev is bridge group, we need to look for
 435                          * physical device (when called from ipv4) */
 436                         NLA_PUT(inst->skb, NFULA_IFINDEX_INDEV,
 437                                 sizeof(tmp_uint), &tmp_uint);
 438                         if (skb->nf_bridge && skb->nf_bridge->physindev) {
 439                                 tmp_uint =
 440                                     htonl(skb->nf_bridge->physindev->ifindex);
 441                                 NLA_PUT(inst->skb, NFULA_IFINDEX_PHYSINDEV,
 442                                         sizeof(tmp_uint), &tmp_uint);
 443                         }
 444                 }
 445 #endif
 446         }
 447
 448         if (outdev) {
 449                 tmp_uint = htonl(outdev->ifindex);
 450 #ifndef CONFIG_BRIDGE_NETFILTER
 451                 NLA_PUT(inst->skb, NFULA_IFINDEX_OUTDEV, sizeof(tmp_uint),
 452                         &tmp_uint);
 453 #else
 454                 if (pf == PF_BRIDGE) {
 455                         /* Case 1: outdev is physical output device, we need to
 456                          * look for bridge group (when called from
 457                          * netfilter_bridge) */
 458                         NLA_PUT(inst->skb, NFULA_IFINDEX_PHYSOUTDEV,
 459                                 sizeof(tmp_uint), &tmp_uint);
 460                         /* this is the bridge group "brX" */
 461                         tmp_uint = htonl(outdev->br_port->br->dev->ifindex);
 462                         NLA_PUT(inst->skb, NFULA_IFINDEX_OUTDEV,
 463                                 sizeof(tmp_uint), &tmp_uint);
 464                 } else {
 465                         /* Case 2: indev is a bridge group, we need to look
 466                          * for physical device (when called from ipv4) */
 467                         NLA_PUT(inst->skb, NFULA_IFINDEX_OUTDEV,
 468                                 sizeof(tmp_uint), &tmp_uint);
 469                         if (skb->nf_bridge && skb->nf_bridge->physoutdev) {
 470                                 tmp_uint =
 471                                     htonl(skb->nf_bridge->physoutdev->ifindex);
 472                                 NLA_PUT(inst->skb, NFULA_IFINDEX_PHYSOUTDEV,
 473                                         sizeof(tmp_uint), &tmp_uint);
 474                         }
 475                 }
 476 #endif
 477         }
 478
 479         if (skb->mark) {
 480                 tmp_uint = htonl(skb->mark);
 481                 NLA_PUT(inst->skb, NFULA_MARK, sizeof(tmp_uint), &tmp_uint);
 482         }
 483
 484         if (indev && skb->dev) {
 485                 struct nfulnl_msg_packet_hw phw;
 486                 int len = dev_parse_header(skb, phw.hw_addr);
 487                 if (len > 0) {
 488                         phw.hw_addrlen = htons(len);
 489                         NLA_PUT(inst->skb, NFULA_HWADDR, sizeof(phw), &phw);
 490                 }
 491         }
 492
 493         if (skb->tstamp.tv64) {
 494                 struct nfulnl_msg_packet_timestamp ts;
 495                 struct timeval tv = ktime_to_timeval(skb->tstamp);
 496                 ts.sec = cpu_to_be64(tv.tv_sec);
 497                 ts.usec = cpu_to_be64(tv.tv_usec);
 498
 499                 NLA_PUT(inst->skb, NFULA_TIMESTAMP, sizeof(ts), &ts);
 500         }
 501
 502         /* UID */
 503         if (skb->sk) {
 504                 read_lock_bh(&skb->sk->sk_callback_lock);
 505                 if (skb->sk->sk_socket && skb->sk->sk_socket->file) {
 506                         __be32 uid = htonl(skb->sk->sk_socket->file->f_uid);
 507                         /* need to unlock here since NLA_PUT may goto */
 508                         read_unlock_bh(&skb->sk->sk_callback_lock);
 509                         NLA_PUT(inst->skb, NFULA_UID, sizeof(uid), &uid);
 510                 } else
 511                         read_unlock_bh(&skb->sk->sk_callback_lock);
 512         }
 513
 514         /* local sequence number */
 515         if (inst->flags & NFULNL_CFG_F_SEQ) {
 516                 tmp_uint = htonl(inst->seq++);
 517                 NLA_PUT(inst->skb, NFULA_SEQ, sizeof(tmp_uint), &tmp_uint);
 518         }
 519         /* global sequence number */
 520         if (inst->flags & NFULNL_CFG_F_SEQ_GLOBAL) {
 521                 tmp_uint = htonl(atomic_inc_return(&global_seq));
 522                 NLA_PUT(inst->skb, NFULA_SEQ_GLOBAL, sizeof(tmp_uint), &tmp_uint);
 523         }
 524
 525         if (data_len) {
 526                 struct nlattr *nla;
 527                 int size = nla_attr_size(data_len);
 528
 529                 if (skb_tailroom(inst->skb) < nla_total_size(data_len)) {
 530                         printk(KERN_WARNING "nfnetlink_log: no tailroom!\n");
 531                         goto nlmsg_failure;
 532                 }
 533
 534                 nla = (struct nlattr *)skb_put(inst->skb, nla_total_size(data_len));
 535                 nla->nla_type = NFULA_PAYLOAD;
 536                 nla->nla_len = size;
 537
 538                 if (skb_copy_bits(skb, 0, nla_data(nla), data_len))
 539                         BUG();
 540         }
 541
 542         nlh->nlmsg_len = inst->skb->tail - old_tail;
 543         return 0;
 544
 545 nlmsg_failure:
 546         UDEBUG("nlmsg_failure\n");
 547 nla_put_failure:
 548         PRINTR(KERN_ERR "nfnetlink_log: error creating log nlmsg\n");
 549         return -1;
 550 }
 551
 552 #define RCV_SKB_FAIL(err) do { netlink_ack(skb, nlh, (err)); return; } while (0)
 553
 554 static struct nf_loginfo default_loginfo = {
 555         .type =         NF_LOG_TYPE_ULOG,
 556         .u = {
 557                 .ulog = {
 558                         .copy_len       = 0xffff,
 559                         .group          = 0,
 560                         .qthreshold     = 1,
 561                 },
 562         },
 563 };
 564
 565 /* log handler for internal netfilter logging api */
 566 static void
 567 nfulnl_log_packet(unsigned int pf,
 568                   unsigned int hooknum,
 569                   const struct sk_buff *skb,
 570                   const struct net_device *in,
 571                   const struct net_device *out,
 572                   const struct nf_loginfo *li_user,
 573                   const char *prefix)
 574 {
 575         unsigned int size, data_len;
 576         struct nfulnl_instance *inst;
 577         const struct nf_loginfo *li;
 578         unsigned int qthreshold;
 579         unsigned int plen;
 580
 581         if (li_user && li_user->type == NF_LOG_TYPE_ULOG)
 582                 li = li_user;
 583         else
 584                 li = &default_loginfo;
 585
 586         inst = instance_lookup_get(li->u.ulog.group);
 587         if (!inst)
 588                 return;
 589
 590         plen = 0;
 591         if (prefix)
 592                 plen = strlen(prefix) + 1;
 593
 594         /* FIXME: do we want to make the size calculation conditional based on
 595          * what is actually present?  way more branches and checks, but more
 596          * memory efficient... */
 597         size =    NLMSG_ALIGN(sizeof(struct nfgenmsg))
 598                 + nla_total_size(sizeof(struct nfulnl_msg_packet_hdr))
 599                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
 600                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
 601 #ifdef CONFIG_BRIDGE_NETFILTER
 602                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
 603                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
 604 #endif
 605                 + nla_total_size(sizeof(u_int32_t))     /* mark */
 606                 + nla_total_size(sizeof(u_int32_t))     /* uid */
 607                 + nla_total_size(plen)                  /* prefix */
 608                 + nla_total_size(sizeof(struct nfulnl_msg_packet_hw))
 609                 + nla_total_size(sizeof(struct nfulnl_msg_packet_timestamp));
 610
 611         UDEBUG("initial size=%u\n", size);
 612
 613         spin_lock_bh(&inst->lock);
 614
 615         if (inst->flags & NFULNL_CFG_F_SEQ)
 616                 size += nla_total_size(sizeof(u_int32_t));
 617         if (inst->flags & NFULNL_CFG_F_SEQ_GLOBAL)
 618                 size += nla_total_size(sizeof(u_int32_t));
 619
 620         qthreshold = inst->qthreshold;
 621         /* per-rule qthreshold overrides per-instance */
 622         if (qthreshold > li->u.ulog.qthreshold)
 623                 qthreshold = li->u.ulog.qthreshold;
 624
 625         switch (inst->copy_mode) {
 626         case NFULNL_COPY_META:
 627         case NFULNL_COPY_NONE:
 628                 data_len = 0;
 629                 break;
 630
 631         case NFULNL_COPY_PACKET:
 632                 if (inst->copy_range == 0
 633                     || inst->copy_range > skb->len)
 634                         data_len = skb->len;
 635                 else
 636                         data_len = inst->copy_range;
 637
 638                 size += nla_total_size(data_len);
 639                 UDEBUG("copy_packet, therefore size now %u\n", size);
 640                 break;
 641
 642         default:
 643                 goto unlock_and_release;
 644         }
 645
 646         if (inst->skb &&
 647             size > skb_tailroom(inst->skb) - sizeof(struct nfgenmsg)) {
 648                 /* either the queue len is too high or we don't have
 649                  * enough room in the skb left. flush to userspace. */
 650                 UDEBUG("flushing old skb\n");
 651
 652                 __nfulnl_flush(inst);
 653         }
 654
 655         if (!inst->skb) {
 656                 inst->skb = nfulnl_alloc_skb(inst->nlbufsiz, size);
 657                 if (!inst->skb)
 658                         goto alloc_failure;
 659         }
 660
 661         UDEBUG("qlen %d, qthreshold %d\n", inst->qlen, qthreshold);
 662         inst->qlen++;
 663
 664         __build_packet_message(inst, skb, data_len, pf,
 665                                 hooknum, in, out, li, prefix, plen);
 666
 667         if (inst->qlen >= qthreshold)
 668                 __nfulnl_flush(inst);
 669         /* timer_pending always called within inst->lock, so there
 670          * is no chance of a race here */
 671         else if (!timer_pending(&inst->timer)) {
 672                 instance_get(inst);
 673                 inst->timer.expires = jiffies + (inst->flushtimeout*HZ/100);
 674                 add_timer(&inst->timer);
 675         }
 676
 677 unlock_and_release:
 678         spin_unlock_bh(&inst->lock);
 679         instance_put(inst);
 680         return;
 681
 682 alloc_failure:
 683         UDEBUG("error allocating skb\n");
 684         /* FIXME: statistics */
 685         goto unlock_and_release;
 686 }
 687
 688 static int
 689 nfulnl_rcv_nl_event(struct notifier_block *this,
 690                    unsigned long event, void *ptr)
 691 {
 692         struct netlink_notify *n = ptr;
 693
 694         if (event == NETLINK_URELEASE &&
 695             n->protocol == NETLINK_NETFILTER && n->pid) {
 696                 int i;
 697
 698                 /* destroy all instances for this pid */
 699                 write_lock_bh(&instances_lock);
 700                 for  (i = 0; i < INSTANCE_BUCKETS; i++) {
 701                         struct hlist_node *tmp, *t2;
 702                         struct nfulnl_instance *inst;
 703                         struct hlist_head *head = &instance_table[i];
 704
 705                         hlist_for_each_entry_safe(inst, tmp, t2, head, hlist) {
 706                                 UDEBUG("node = %p\n", inst);
 707                                 if ((n->net == &init_net) &&
 708                                     (n->pid == inst->peer_pid))
 709                                         __instance_destroy(inst);
 710                         }
 711                 }
 712                 write_unlock_bh(&instances_lock);
 713         }
 714         return NOTIFY_DONE;
 715 }
 716
 717 static struct notifier_block nfulnl_rtnl_notifier = {
 718         .notifier_call  = nfulnl_rcv_nl_event,
 719 };
 720
 721 static int
 722 nfulnl_recv_unsupp(struct sock *ctnl, struct sk_buff *skb,
 723                   struct nlmsghdr *nlh, struct nlattr *nfqa[])
 724 {
 725         return -ENOTSUPP;
 726 }
 727
 728 static struct nf_logger nfulnl_logger = {
 729         .name   = "nfnetlink_log",
 730         .logfn  = &nfulnl_log_packet,
 731         .me     = THIS_MODULE,
 732 };
 733
 734 static const struct nla_policy nfula_cfg_policy[NFULA_CFG_MAX+1] = {
 735         [NFULA_CFG_CMD]         = { .len = sizeof(struct nfulnl_msg_config_cmd) },
 736         [NFULA_CFG_MODE]        = { .len = sizeof(struct nfulnl_msg_config_mode) },
 737         [NFULA_CFG_TIMEOUT]     = { .type = NLA_U32 },
 738         [NFULA_CFG_QTHRESH]     = { .type = NLA_U32 },
 739         [NFULA_CFG_NLBUFSIZ]    = { .type = NLA_U32 },
 740         [NFULA_CFG_FLAGS]       = { .type = NLA_U16 },
 741 };
 742
 743 static int
 744 nfulnl_recv_config(struct sock *ctnl, struct sk_buff *skb,
 745                    struct nlmsghdr *nlh, struct nlattr *nfula[])
 746 {
 747         struct nfgenmsg *nfmsg = NLMSG_DATA(nlh);
 748         u_int16_t group_num = ntohs(nfmsg->res_id);
 749         struct nfulnl_instance *inst;
 750         int ret = 0;
 751
 752         UDEBUG("entering for msg %u\n", NFNL_MSG_TYPE(nlh->nlmsg_type));
 753
 754         inst = instance_lookup_get(group_num);
 755         if (nfula[NFULA_CFG_CMD]) {
 756                 u_int8_t pf = nfmsg->nfgen_family;
 757                 struct nfulnl_msg_config_cmd *cmd;
 758                 cmd = nla_data(nfula[NFULA_CFG_CMD]);
 759                 UDEBUG("found CFG_CMD for\n");
 760
 761                 switch (cmd->command) {
 762                 case NFULNL_CFG_CMD_BIND:
 763                         if (inst) {
 764                                 ret = -EBUSY;
 765                                 goto out_put;
 766                         }
 767
 768                         inst = instance_create(group_num,
 769                                                NETLINK_CB(skb).pid);
 770                         if (!inst) {
 771                                 ret = -EINVAL;
 772                                 goto out;
 773                         }
 774                         break;
 775                 case NFULNL_CFG_CMD_UNBIND:
 776                         if (!inst) {
 777                                 ret = -ENODEV;
 778                                 goto out;
 779                         }
 780
 781                         if (inst->peer_pid != NETLINK_CB(skb).pid) {
 782                                 ret = -EPERM;
 783                                 goto out_put;
 784                         }
 785
 786                         instance_destroy(inst);
 787                         goto out;
 788                 case NFULNL_CFG_CMD_PF_BIND:
 789                         UDEBUG("registering log handler for pf=%u\n", pf);
 790                         ret = nf_log_register(pf, &nfulnl_logger);
 791                         break;
 792                 case NFULNL_CFG_CMD_PF_UNBIND:
 793                         UDEBUG("unregistering log handler for pf=%u\n", pf);
 794                         /* This is a bug and a feature.  We cannot unregister
 795                          * other handlers, like nfnetlink_inst can */
 796                         nf_log_unregister_pf(pf);
 797                         break;
 798                 default:
 799                         ret = -EINVAL;
 800                         break;
 801                 }
 802
 803                 if (!inst)
 804                         goto out;
 805         } else {
 806                 if (!inst) {
 807                         UDEBUG("no config command, and no instance for "
 808                                 "group=%u pid=%u =>ENOENT\n",
 809                                 group_num, NETLINK_CB(skb).pid);
 810                         ret = -ENOENT;
 811                         goto out;
 812                 }
 813
 814                 if (inst->peer_pid != NETLINK_CB(skb).pid) {
 815                         UDEBUG("no config command, and wrong pid\n");
 816                         ret = -EPERM;
 817                         goto out_put;
 818                 }
 819         }
 820
 821         if (nfula[NFULA_CFG_MODE]) {
 822                 struct nfulnl_msg_config_mode *params;
 823                 params = nla_data(nfula[NFULA_CFG_MODE]);
 824
 825                 nfulnl_set_mode(inst, params->copy_mode,
 826                                 ntohl(params->copy_range));
 827         }
 828
 829         if (nfula[NFULA_CFG_TIMEOUT]) {
 830                 __be32 timeout =
 831                         *(__be32 *)nla_data(nfula[NFULA_CFG_TIMEOUT]);
 832
 833                 nfulnl_set_timeout(inst, ntohl(timeout));
 834         }
 835
 836         if (nfula[NFULA_CFG_NLBUFSIZ]) {
 837                 __be32 nlbufsiz =
 838                         *(__be32 *)nla_data(nfula[NFULA_CFG_NLBUFSIZ]);
 839
 840                 nfulnl_set_nlbufsiz(inst, ntohl(nlbufsiz));
 841         }
 842
 843         if (nfula[NFULA_CFG_QTHRESH]) {
 844                 __be32 qthresh =
 845                         *(__be32 *)nla_data(nfula[NFULA_CFG_QTHRESH]);
 846
 847                 nfulnl_set_qthresh(inst, ntohl(qthresh));
 848         }
 849
 850         if (nfula[NFULA_CFG_FLAGS]) {
 851                 __be16 flags =
 852                         *(__be16 *)nla_data(nfula[NFULA_CFG_FLAGS]);
 853                 nfulnl_set_flags(inst, ntohs(flags));
 854         }
 855
 856 out_put:
 857         instance_put(inst);
 858 out:
 859         return ret;
 860 }
 861
 862 static const struct nfnl_callback nfulnl_cb[NFULNL_MSG_MAX] = {
 863         [NFULNL_MSG_PACKET]     = { .call = nfulnl_recv_unsupp,
 864                                     .attr_count = NFULA_MAX, },
 865         [NFULNL_MSG_CONFIG]     = { .call = nfulnl_recv_config,
 866                                     .attr_count = NFULA_CFG_MAX,
 867                                     .policy = nfula_cfg_policy },
 868 };
 869
 870 static const struct nfnetlink_subsystem nfulnl_subsys = {
 871         .name           = "log",
 872         .subsys_id      = NFNL_SUBSYS_ULOG,
 873         .cb_count       = NFULNL_MSG_MAX,
 874         .cb             = nfulnl_cb,
 875 };
 876
 877 #ifdef CONFIG_PROC_FS
 878 struct iter_state {
 879         unsigned int bucket;
 880 };
 881
 882 static struct hlist_node *get_first(struct iter_state *st)
 883 {
 884         if (!st)
 885                 return NULL;
 886
 887         for (st->bucket = 0; st->bucket < INSTANCE_BUCKETS; st->bucket++) {
 888                 if (!hlist_empty(&instance_table[st->bucket]))
 889                         return instance_table[st->bucket].first;
 890         }
 891         return NULL;
 892 }
 893
 894 static struct hlist_node *get_next(struct iter_state *st, struct hlist_node *h)
 895 {
 896         h = h->next;
 897         while (!h) {
 898                 if (++st->bucket >= INSTANCE_BUCKETS)
 899                         return NULL;
 900
 901                 h = instance_table[st->bucket].first;
 902         }
 903         return h;
 904 }
 905
 906 static struct hlist_node *get_idx(struct iter_state *st, loff_t pos)
 907 {
 908         struct hlist_node *head;
 909         head = get_first(st);
 910
 911         if (head)
 912                 while (pos && (head = get_next(st, head)))
 913                         pos--;
 914         return pos ? NULL : head;
 915 }
 916
 917 static void *seq_start(struct seq_file *seq, loff_t *pos)
 918 {
 919         read_lock_bh(&instances_lock);
 920         return get_idx(seq->private, *pos);
 921 }
 922
 923 static void *seq_next(struct seq_file *s, void *v, loff_t *pos)
 924 {
 925         (*pos)++;
 926         return get_next(s->private, v);
 927 }
 928
 929 static void seq_stop(struct seq_file *s, void *v)
 930 {
 931         read_unlock_bh(&instances_lock);
 932 }
 933
 934 static int seq_show(struct seq_file *s, void *v)
 935 {
 936         const struct nfulnl_instance *inst = v;
 937
 938         return seq_printf(s, "%5d %6d %5d %1d %5d %6d %2d\n",
 939                           inst->group_num,
 940                           inst->peer_pid, inst->qlen,
 941                           inst->copy_mode, inst->copy_range,
 942                           inst->flushtimeout, atomic_read(&inst->use));
 943 }
 944
 945 static const struct seq_operations nful_seq_ops = {
 946         .start  = seq_start,
 947         .next   = seq_next,
 948         .stop   = seq_stop,
 949         .show   = seq_show,
 950 };
 951
 952 static int nful_open(struct inode *inode, struct file *file)
 953 {
 954         return seq_open_private(file, &nful_seq_ops,
 955                         sizeof(struct iter_state));
 956 }
 957
 958 static const struct file_operations nful_file_ops = {
 959         .owner   = THIS_MODULE,
 960         .open    = nful_open,
 961         .read    = seq_read,
 962         .llseek  = seq_lseek,
 963         .release = seq_release_private,
 964 };
 965
 966 #endif /* PROC_FS */
 967
 968 static int __init nfnetlink_log_init(void)
 969 {
 970         int i, status = -ENOMEM;
 971 #ifdef CONFIG_PROC_FS
 972         struct proc_dir_entry *proc_nful;
 973 #endif
 974
 975         for (i = 0; i < INSTANCE_BUCKETS; i++)
 976                 INIT_HLIST_HEAD(&instance_table[i]);
 977
 978         /* it's not really all that important to have a random value, so
 979          * we can do this from the init function, even if there hasn't
 980          * been that much entropy yet */
 981         get_random_bytes(&hash_init, sizeof(hash_init));
 982
 983         netlink_register_notifier(&nfulnl_rtnl_notifier);
 984         status = nfnetlink_subsys_register(&nfulnl_subsys);
 985         if (status < 0) {
 986                 printk(KERN_ERR "log: failed to create netlink socket\n");
 987                 goto cleanup_netlink_notifier;
 988         }
 989
 990 #ifdef CONFIG_PROC_FS
 991         proc_nful = create_proc_entry("nfnetlink_log", 0440,
 992                                       proc_net_netfilter);
 993         if (!proc_nful)
 994                 goto cleanup_subsys;
 995         proc_nful->proc_fops = &nful_file_ops;
 996 #endif
 997         return status;
 998
 999 #ifdef CONFIG_PROC_FS
1000 cleanup_subsys:
1001         nfnetlink_subsys_unregister(&nfulnl_subsys);
1002 #endif
1003 cleanup_netlink_notifier:
1004         netlink_unregister_notifier(&nfulnl_rtnl_notifier);
1005         return status;
1006 }
1007
1008 static void __exit nfnetlink_log_fini(void)
1009 {
1010         nf_log_unregister(&nfulnl_logger);
1011 #ifdef CONFIG_PROC_FS
1012         remove_proc_entry("nfnetlink_log", proc_net_netfilter);
1013 #endif
1014         nfnetlink_subsys_unregister(&nfulnl_subsys);
1015         netlink_unregister_notifier(&nfulnl_rtnl_notifier);
1016 }
1017
1018 MODULE_DESCRIPTION("netfilter userspace logging");
1019 MODULE_AUTHOR("Harald Welte <laforge@netfilter.org>");
1020 MODULE_LICENSE("GPL");
1021 MODULE_ALIAS_NFNL_SUBSYS(NFNL_SUBSYS_ULOG);
1022
1023 module_init(nfnetlink_log_init);
1024 module_exit(nfnetlink_log_fini);