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 #include <net/netfilter/nf_log.h>
  33
  34 #include <asm/atomic.h>
  35
  36 #ifdef CONFIG_BRIDGE_NETFILTER
  37 #include "../bridge/br_private.h"
  38 #endif
  39
  40 #define NFULNL_NLBUFSIZ_DEFAULT NLMSG_GOODSIZE
  41 #define NFULNL_TIMEOUT_DEFAULT  HZ      /* every second */
  42 #define NFULNL_QTHRESH_DEFAULT  100     /* 100 packets */
  43 #define NFULNL_COPY_RANGE_MAX   0xFFFF  /* max packet size is limited by 16-bit struct nfattr nfa_len field */
  44
  45 #define PRINTR(x, args...)      do { if (net_ratelimit()) \
  46                                      printk(x, ## args); } while (0);
  47
  48 #if 0
  49 #define UDEBUG(x, args ...)     printk(KERN_DEBUG "%s(%d):%s(): " x,       \
  50                                         __FILE__, __LINE__, __FUNCTION__,  \
  51                                         ## args)
  52 #else
  53 #define UDEBUG(x, ...)
  54 #endif
  55
  56 struct nfulnl_instance {
  57         struct hlist_node hlist;        /* global list of instances */
  58         spinlock_t lock;
  59         atomic_t use;                   /* use count */
  60
  61         unsigned int qlen;              /* number of nlmsgs in skb */
  62         struct sk_buff *skb;            /* pre-allocatd skb */
  63         struct timer_list timer;
  64         int peer_pid;                   /* PID of the peer process */
  65
  66         /* configurable parameters */
  67         unsigned int flushtimeout;      /* timeout until queue flush */
  68         unsigned int nlbufsiz;          /* netlink buffer allocation size */
  69         unsigned int qthreshold;        /* threshold of the queue */
  70         u_int32_t copy_range;
  71         u_int32_t seq;                  /* instance-local sequential counter */
  72         u_int16_t group_num;            /* number of this queue */
  73         u_int16_t flags;
  74         u_int8_t copy_mode;
  75 };
  76
  77 static DEFINE_RWLOCK(instances_lock);
  78 static atomic_t global_seq;
  79
  80 #define INSTANCE_BUCKETS        16
  81 static struct hlist_head instance_table[INSTANCE_BUCKETS];
  82 static unsigned int hash_init;
  83
  84 static inline u_int8_t instance_hashfn(u_int16_t group_num)
  85 {
  86         return ((group_num & 0xff) % INSTANCE_BUCKETS);
  87 }
  88
  89 static struct nfulnl_instance *
  90 __instance_lookup(u_int16_t group_num)
  91 {
  92         struct hlist_head *head;
  93         struct hlist_node *pos;
  94         struct nfulnl_instance *inst;
  95
  96         UDEBUG("entering (group_num=%u)\n", group_num);
  97
  98         head = &instance_table[instance_hashfn(group_num)];
  99         hlist_for_each_entry(inst, pos, head, hlist) {
 100                 if (inst->group_num == group_num)
 101                         return inst;
 102         }
 103         return NULL;
 104 }
 105
 106 static inline void
 107 instance_get(struct nfulnl_instance *inst)
 108 {
 109         atomic_inc(&inst->use);
 110 }
 111
 112 static struct nfulnl_instance *
 113 instance_lookup_get(u_int16_t group_num)
 114 {
 115         struct nfulnl_instance *inst;
 116
 117         read_lock_bh(&instances_lock);
 118         inst = __instance_lookup(group_num);
 119         if (inst)
 120                 instance_get(inst);
 121         read_unlock_bh(&instances_lock);
 122
 123         return inst;
 124 }
 125
 126 static void
 127 instance_put(struct nfulnl_instance *inst)
 128 {
 129         if (inst && atomic_dec_and_test(&inst->use)) {
 130                 UDEBUG("kfree(inst=%p)\n", inst);
 131                 kfree(inst);
 132                 module_put(THIS_MODULE);
 133         }
 134 }
 135
 136 static void nfulnl_timer(unsigned long data);
 137
 138 static struct nfulnl_instance *
 139 instance_create(u_int16_t group_num, int pid)
 140 {
 141         struct nfulnl_instance *inst;
 142
 143         UDEBUG("entering (group_num=%u, pid=%d)\n", group_num,
 144                 pid);
 145
 146         write_lock_bh(&instances_lock);
 147         if (__instance_lookup(group_num)) {
 148                 inst = NULL;
 149                 UDEBUG("aborting, instance already exists\n");
 150                 goto out_unlock;
 151         }
 152
 153         inst = kzalloc(sizeof(*inst), GFP_ATOMIC);
 154         if (!inst)
 155                 goto out_unlock;
 156
 157         if (!try_module_get(THIS_MODULE)) {
 158                 kfree(inst);
 159                 goto out_unlock;
 160         }
 161
 162         INIT_HLIST_NODE(&inst->hlist);
 163         spin_lock_init(&inst->lock);
 164         /* needs to be two, since we _put() after creation */
 165         atomic_set(&inst->use, 2);
 166
 167         setup_timer(&inst->timer, nfulnl_timer, (unsigned long)inst);
 168
 169         inst->peer_pid = pid;
 170         inst->group_num = group_num;
 171
 172         inst->qthreshold        = NFULNL_QTHRESH_DEFAULT;
 173         inst->flushtimeout      = NFULNL_TIMEOUT_DEFAULT;
 174         inst->nlbufsiz          = NFULNL_NLBUFSIZ_DEFAULT;
 175         inst->copy_mode         = NFULNL_COPY_PACKET;
 176         inst->copy_range        = NFULNL_COPY_RANGE_MAX;
 177
 178         hlist_add_head(&inst->hlist,
 179                        &instance_table[instance_hashfn(group_num)]);
 180
 181         UDEBUG("newly added node: %p, next=%p\n", &inst->hlist,
 182                 inst->hlist.next);
 183
 184         write_unlock_bh(&instances_lock);
 185
 186         return inst;
 187
 188 out_unlock:
 189         write_unlock_bh(&instances_lock);
 190         return NULL;
 191 }
 192
 193 static void __nfulnl_flush(struct nfulnl_instance *inst);
 194
 195 static void
 196 __instance_destroy(struct nfulnl_instance *inst)
 197 {
 198         /* first pull it out of the global list */
 199         UDEBUG("removing instance %p (queuenum=%u) from hash\n",
 200                 inst, inst->group_num);
 201
 202         hlist_del(&inst->hlist);
 203
 204         /* then flush all pending packets from skb */
 205
 206         spin_lock_bh(&inst->lock);
 207         if (inst->skb)
 208                 __nfulnl_flush(inst);
 209         spin_unlock_bh(&inst->lock);
 210
 211         /* and finally put the refcount */
 212         instance_put(inst);
 213 }
 214
 215 static inline void
 216 instance_destroy(struct nfulnl_instance *inst)
 217 {
 218         write_lock_bh(&instances_lock);
 219         __instance_destroy(inst);
 220         write_unlock_bh(&instances_lock);
 221 }
 222
 223 static int
 224 nfulnl_set_mode(struct nfulnl_instance *inst, u_int8_t mode,
 225                   unsigned int range)
 226 {
 227         int status = 0;
 228
 229         spin_lock_bh(&inst->lock);
 230
 231         switch (mode) {
 232         case NFULNL_COPY_NONE:
 233         case NFULNL_COPY_META:
 234                 inst->copy_mode = mode;
 235                 inst->copy_range = 0;
 236                 break;
 237
 238         case NFULNL_COPY_PACKET:
 239                 inst->copy_mode = mode;
 240                 inst->copy_range = min_t(unsigned int,
 241                                          range, NFULNL_COPY_RANGE_MAX);
 242                 break;
 243
 244         default:
 245                 status = -EINVAL;
 246                 break;
 247         }
 248
 249         spin_unlock_bh(&inst->lock);
 250
 251         return status;
 252 }
 253
 254 static int
 255 nfulnl_set_nlbufsiz(struct nfulnl_instance *inst, u_int32_t nlbufsiz)
 256 {
 257         int status;
 258
 259         spin_lock_bh(&inst->lock);
 260         if (nlbufsiz < NFULNL_NLBUFSIZ_DEFAULT)
 261                 status = -ERANGE;
 262         else if (nlbufsiz > 131072)
 263                 status = -ERANGE;
 264         else {
 265                 inst->nlbufsiz = nlbufsiz;
 266                 status = 0;
 267         }
 268         spin_unlock_bh(&inst->lock);
 269
 270         return status;
 271 }
 272
 273 static int
 274 nfulnl_set_timeout(struct nfulnl_instance *inst, u_int32_t timeout)
 275 {
 276         spin_lock_bh(&inst->lock);
 277         inst->flushtimeout = timeout;
 278         spin_unlock_bh(&inst->lock);
 279
 280         return 0;
 281 }
 282
 283 static int
 284 nfulnl_set_qthresh(struct nfulnl_instance *inst, u_int32_t qthresh)
 285 {
 286         spin_lock_bh(&inst->lock);
 287         inst->qthreshold = qthresh;
 288         spin_unlock_bh(&inst->lock);
 289
 290         return 0;
 291 }
 292
 293 static int
 294 nfulnl_set_flags(struct nfulnl_instance *inst, u_int16_t flags)
 295 {
 296         spin_lock_bh(&inst->lock);
 297         inst->flags = flags;
 298         spin_unlock_bh(&inst->lock);
 299
 300         return 0;
 301 }
 302
 303 static struct sk_buff *
 304 nfulnl_alloc_skb(unsigned int inst_size, unsigned int pkt_size)
 305 {
 306         struct sk_buff *skb;
 307         unsigned int n;
 308
 309         UDEBUG("entered (%u, %u)\n", inst_size, pkt_size);
 310
 311         /* alloc skb which should be big enough for a whole multipart
 312          * message.  WARNING: has to be <= 128k due to slab restrictions */
 313
 314         n = max(inst_size, pkt_size);
 315         skb = alloc_skb(n, GFP_ATOMIC);
 316         if (!skb) {
 317                 PRINTR("nfnetlink_log: can't alloc whole buffer (%u bytes)\n",
 318                         inst_size);
 319
 320                 if (n > pkt_size) {
 321                         /* try to allocate only as much as we need for current
 322                          * packet */
 323
 324                         skb = alloc_skb(pkt_size, GFP_ATOMIC);
 325                         if (!skb)
 326                                 PRINTR("nfnetlink_log: can't even alloc %u "
 327                                        "bytes\n", pkt_size);
 328                 }
 329         }
 330
 331         return skb;
 332 }
 333
 334 static int
 335 __nfulnl_send(struct nfulnl_instance *inst)
 336 {
 337         int status = -1;
 338
 339         if (inst->qlen > 1)
 340                 NLMSG_PUT(inst->skb, 0, 0,
 341                           NLMSG_DONE,
 342                           sizeof(struct nfgenmsg));
 343
 344         status = nfnetlink_unicast(inst->skb, inst->peer_pid, MSG_DONTWAIT);
 345         if (status < 0) {
 346                 UDEBUG("netlink_unicast() failed\n");
 347                 /* FIXME: statistics */
 348         }
 349
 350         inst->qlen = 0;
 351         inst->skb = NULL;
 352
 353 nlmsg_failure:
 354         return status;
 355 }
 356
 357 static void
 358 __nfulnl_flush(struct nfulnl_instance *inst)
 359 {
 360         /* timer holds a reference */
 361         if (del_timer(&inst->timer))
 362                 instance_put(inst);
 363         if (inst->skb)
 364                 __nfulnl_send(inst);
 365 }
 366
 367 static void
 368 nfulnl_timer(unsigned long data)
 369 {
 370         struct nfulnl_instance *inst = (struct nfulnl_instance *)data;
 371
 372         UDEBUG("timer function called, flushing buffer\n");
 373
 374         spin_lock_bh(&inst->lock);
 375         if (inst->skb)
 376                 __nfulnl_send(inst);
 377         spin_unlock_bh(&inst->lock);
 378         instance_put(inst);
 379 }
 380
 381 /* This is an inline function, we don't really care about a long
 382  * list of arguments */
 383 static inline int
 384 __build_packet_message(struct nfulnl_instance *inst,
 385                         const struct sk_buff *skb,
 386                         unsigned int data_len,
 387                         unsigned int pf,
 388                         unsigned int hooknum,
 389                         const struct net_device *indev,
 390                         const struct net_device *outdev,
 391                         const struct nf_loginfo *li,
 392                         const char *prefix, unsigned int plen)
 393 {
 394         struct nfulnl_msg_packet_hdr pmsg;
 395         struct nlmsghdr *nlh;
 396         struct nfgenmsg *nfmsg;
 397         __be32 tmp_uint;
 398         sk_buff_data_t old_tail = inst->skb->tail;
 399
 400         UDEBUG("entered\n");
 401
 402         nlh = NLMSG_PUT(inst->skb, 0, 0,
 403                         NFNL_SUBSYS_ULOG << 8 | NFULNL_MSG_PACKET,
 404                         sizeof(struct nfgenmsg));
 405         nfmsg = NLMSG_DATA(nlh);
 406         nfmsg->nfgen_family = pf;
 407         nfmsg->version = NFNETLINK_V0;
 408         nfmsg->res_id = htons(inst->group_num);
 409
 410         pmsg.hw_protocol        = skb->protocol;
 411         pmsg.hook               = hooknum;
 412
 413         NLA_PUT(inst->skb, NFULA_PACKET_HDR, sizeof(pmsg), &pmsg);
 414
 415         if (prefix)
 416                 NLA_PUT(inst->skb, NFULA_PREFIX, plen, prefix);
 417
 418         if (indev) {
 419                 tmp_uint = htonl(indev->ifindex);
 420 #ifndef CONFIG_BRIDGE_NETFILTER
 421                 NLA_PUT(inst->skb, NFULA_IFINDEX_INDEV, sizeof(tmp_uint),
 422                         &tmp_uint);
 423 #else
 424                 if (pf == PF_BRIDGE) {
 425                         /* Case 1: outdev is physical input device, we need to
 426                          * look for bridge group (when called from
 427                          * netfilter_bridge) */
 428                         NLA_PUT(inst->skb, NFULA_IFINDEX_PHYSINDEV,
 429                                 sizeof(tmp_uint), &tmp_uint);
 430                         /* this is the bridge group "brX" */
 431                         tmp_uint = htonl(indev->br_port->br->dev->ifindex);
 432                         NLA_PUT(inst->skb, NFULA_IFINDEX_INDEV,
 433                                 sizeof(tmp_uint), &tmp_uint);
 434                 } else {
 435                         /* Case 2: indev is bridge group, we need to look for
 436                          * physical device (when called from ipv4) */
 437                         NLA_PUT(inst->skb, NFULA_IFINDEX_INDEV,
 438                                 sizeof(tmp_uint), &tmp_uint);
 439                         if (skb->nf_bridge && skb->nf_bridge->physindev) {
 440                                 tmp_uint =
 441                                     htonl(skb->nf_bridge->physindev->ifindex);
 442                                 NLA_PUT(inst->skb, NFULA_IFINDEX_PHYSINDEV,
 443                                         sizeof(tmp_uint), &tmp_uint);
 444                         }
 445                 }
 446 #endif
 447         }
 448
 449         if (outdev) {
 450                 tmp_uint = htonl(outdev->ifindex);
 451 #ifndef CONFIG_BRIDGE_NETFILTER
 452                 NLA_PUT(inst->skb, NFULA_IFINDEX_OUTDEV, sizeof(tmp_uint),
 453                         &tmp_uint);
 454 #else
 455                 if (pf == PF_BRIDGE) {
 456                         /* Case 1: outdev is physical output device, we need to
 457                          * look for bridge group (when called from
 458                          * netfilter_bridge) */
 459                         NLA_PUT(inst->skb, NFULA_IFINDEX_PHYSOUTDEV,
 460                                 sizeof(tmp_uint), &tmp_uint);
 461                         /* this is the bridge group "brX" */
 462                         tmp_uint = htonl(outdev->br_port->br->dev->ifindex);
 463                         NLA_PUT(inst->skb, NFULA_IFINDEX_OUTDEV,
 464                                 sizeof(tmp_uint), &tmp_uint);
 465                 } else {
 466                         /* Case 2: indev is a bridge group, we need to look
 467                          * for physical device (when called from ipv4) */
 468                         NLA_PUT(inst->skb, NFULA_IFINDEX_OUTDEV,
 469                                 sizeof(tmp_uint), &tmp_uint);
 470                         if (skb->nf_bridge && skb->nf_bridge->physoutdev) {
 471                                 tmp_uint =
 472                                     htonl(skb->nf_bridge->physoutdev->ifindex);
 473                                 NLA_PUT(inst->skb, NFULA_IFINDEX_PHYSOUTDEV,
 474                                         sizeof(tmp_uint), &tmp_uint);
 475                         }
 476                 }
 477 #endif
 478         }
 479
 480         if (skb->mark) {
 481                 tmp_uint = htonl(skb->mark);
 482                 NLA_PUT(inst->skb, NFULA_MARK, sizeof(tmp_uint), &tmp_uint);
 483         }
 484
 485         if (indev && skb->dev) {
 486                 struct nfulnl_msg_packet_hw phw;
 487                 int len = dev_parse_header(skb, phw.hw_addr);
 488                 if (len > 0) {
 489                         phw.hw_addrlen = htons(len);
 490                         NLA_PUT(inst->skb, NFULA_HWADDR, sizeof(phw), &phw);
 491                 }
 492         }
 493
 494         if (skb->tstamp.tv64) {
 495                 struct nfulnl_msg_packet_timestamp ts;
 496                 struct timeval tv = ktime_to_timeval(skb->tstamp);
 497                 ts.sec = cpu_to_be64(tv.tv_sec);
 498                 ts.usec = cpu_to_be64(tv.tv_usec);
 499
 500                 NLA_PUT(inst->skb, NFULA_TIMESTAMP, sizeof(ts), &ts);
 501         }
 502
 503         /* UID */
 504         if (skb->sk) {
 505                 read_lock_bh(&skb->sk->sk_callback_lock);
 506                 if (skb->sk->sk_socket && skb->sk->sk_socket->file) {
 507                         __be32 uid = htonl(skb->sk->sk_socket->file->f_uid);
 508                         /* need to unlock here since NLA_PUT may goto */
 509                         read_unlock_bh(&skb->sk->sk_callback_lock);
 510                         NLA_PUT(inst->skb, NFULA_UID, sizeof(uid), &uid);
 511                 } else
 512                         read_unlock_bh(&skb->sk->sk_callback_lock);
 513         }
 514
 515         /* local sequence number */
 516         if (inst->flags & NFULNL_CFG_F_SEQ) {
 517                 tmp_uint = htonl(inst->seq++);
 518                 NLA_PUT(inst->skb, NFULA_SEQ, sizeof(tmp_uint), &tmp_uint);
 519         }
 520         /* global sequence number */
 521         if (inst->flags & NFULNL_CFG_F_SEQ_GLOBAL) {
 522                 tmp_uint = htonl(atomic_inc_return(&global_seq));
 523                 NLA_PUT(inst->skb, NFULA_SEQ_GLOBAL, sizeof(tmp_uint), &tmp_uint);
 524         }
 525
 526         if (data_len) {
 527                 struct nlattr *nla;
 528                 int size = nla_attr_size(data_len);
 529
 530                 if (skb_tailroom(inst->skb) < nla_total_size(data_len)) {
 531                         printk(KERN_WARNING "nfnetlink_log: no tailroom!\n");
 532                         goto nlmsg_failure;
 533                 }
 534
 535                 nla = (struct nlattr *)skb_put(inst->skb, nla_total_size(data_len));
 536                 nla->nla_type = NFULA_PAYLOAD;
 537                 nla->nla_len = size;
 538
 539                 if (skb_copy_bits(skb, 0, nla_data(nla), data_len))
 540                         BUG();
 541         }
 542
 543         nlh->nlmsg_len = inst->skb->tail - old_tail;
 544         return 0;
 545
 546 nlmsg_failure:
 547         UDEBUG("nlmsg_failure\n");
 548 nla_put_failure:
 549         PRINTR(KERN_ERR "nfnetlink_log: error creating log nlmsg\n");
 550         return -1;
 551 }
 552
 553 #define RCV_SKB_FAIL(err) do { netlink_ack(skb, nlh, (err)); return; } while (0)
 554
 555 static struct nf_loginfo default_loginfo = {
 556         .type =         NF_LOG_TYPE_ULOG,
 557         .u = {
 558                 .ulog = {
 559                         .copy_len       = 0xffff,
 560                         .group          = 0,
 561                         .qthreshold     = 1,
 562                 },
 563         },
 564 };
 565
 566 /* log handler for internal netfilter logging api */
 567 static void
 568 nfulnl_log_packet(unsigned int pf,
 569                   unsigned int hooknum,
 570                   const struct sk_buff *skb,
 571                   const struct net_device *in,
 572                   const struct net_device *out,
 573                   const struct nf_loginfo *li_user,
 574                   const char *prefix)
 575 {
 576         unsigned int size, data_len;
 577         struct nfulnl_instance *inst;
 578         const struct nf_loginfo *li;
 579         unsigned int qthreshold;
 580         unsigned int plen;
 581
 582         if (li_user && li_user->type == NF_LOG_TYPE_ULOG)
 583                 li = li_user;
 584         else
 585                 li = &default_loginfo;
 586
 587         inst = instance_lookup_get(li->u.ulog.group);
 588         if (!inst)
 589                 return;
 590
 591         plen = 0;
 592         if (prefix)
 593                 plen = strlen(prefix) + 1;
 594
 595         /* FIXME: do we want to make the size calculation conditional based on
 596          * what is actually present?  way more branches and checks, but more
 597          * memory efficient... */
 598         size =    NLMSG_ALIGN(sizeof(struct nfgenmsg))
 599                 + nla_total_size(sizeof(struct nfulnl_msg_packet_hdr))
 600                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
 601                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
 602 #ifdef CONFIG_BRIDGE_NETFILTER
 603                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
 604                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
 605 #endif
 606                 + nla_total_size(sizeof(u_int32_t))     /* mark */
 607                 + nla_total_size(sizeof(u_int32_t))     /* uid */
 608                 + nla_total_size(plen)                  /* prefix */
 609                 + nla_total_size(sizeof(struct nfulnl_msg_packet_hw))
 610                 + nla_total_size(sizeof(struct nfulnl_msg_packet_timestamp));
 611
 612         UDEBUG("initial size=%u\n", size);
 613
 614         spin_lock_bh(&inst->lock);
 615
 616         if (inst->flags & NFULNL_CFG_F_SEQ)
 617                 size += nla_total_size(sizeof(u_int32_t));
 618         if (inst->flags & NFULNL_CFG_F_SEQ_GLOBAL)
 619                 size += nla_total_size(sizeof(u_int32_t));
 620
 621         qthreshold = inst->qthreshold;
 622         /* per-rule qthreshold overrides per-instance */
 623         if (qthreshold > li->u.ulog.qthreshold)
 624                 qthreshold = li->u.ulog.qthreshold;
 625
 626         switch (inst->copy_mode) {
 627         case NFULNL_COPY_META:
 628         case NFULNL_COPY_NONE:
 629                 data_len = 0;
 630                 break;
 631
 632         case NFULNL_COPY_PACKET:
 633                 if (inst->copy_range == 0
 634                     || inst->copy_range > skb->len)
 635                         data_len = skb->len;
 636                 else
 637                         data_len = inst->copy_range;
 638
 639                 size += nla_total_size(data_len);
 640                 UDEBUG("copy_packet, therefore size now %u\n", size);
 641                 break;
 642
 643         default:
 644                 goto unlock_and_release;
 645         }
 646
 647         if (inst->skb &&
 648             size > skb_tailroom(inst->skb) - sizeof(struct nfgenmsg)) {
 649                 /* either the queue len is too high or we don't have
 650                  * enough room in the skb left. flush to userspace. */
 651                 UDEBUG("flushing old skb\n");
 652
 653                 __nfulnl_flush(inst);
 654         }
 655
 656         if (!inst->skb) {
 657                 inst->skb = nfulnl_alloc_skb(inst->nlbufsiz, size);
 658                 if (!inst->skb)
 659                         goto alloc_failure;
 660         }
 661
 662         UDEBUG("qlen %d, qthreshold %d\n", inst->qlen, qthreshold);
 663         inst->qlen++;
 664
 665         __build_packet_message(inst, skb, data_len, pf,
 666                                 hooknum, in, out, li, prefix, plen);
 667
 668         if (inst->qlen >= qthreshold)
 669                 __nfulnl_flush(inst);
 670         /* timer_pending always called within inst->lock, so there
 671          * is no chance of a race here */
 672         else if (!timer_pending(&inst->timer)) {
 673                 instance_get(inst);
 674                 inst->timer.expires = jiffies + (inst->flushtimeout*HZ/100);
 675                 add_timer(&inst->timer);
 676         }
 677
 678 unlock_and_release:
 679         spin_unlock_bh(&inst->lock);
 680         instance_put(inst);
 681         return;
 682
 683 alloc_failure:
 684         UDEBUG("error allocating skb\n");
 685         /* FIXME: statistics */
 686         goto unlock_and_release;
 687 }
 688
 689 static int
 690 nfulnl_rcv_nl_event(struct notifier_block *this,
 691                    unsigned long event, void *ptr)
 692 {
 693         struct netlink_notify *n = ptr;
 694
 695         if (event == NETLINK_URELEASE &&
 696             n->protocol == NETLINK_NETFILTER && n->pid) {
 697                 int i;
 698
 699                 /* destroy all instances for this pid */
 700                 write_lock_bh(&instances_lock);
 701                 for  (i = 0; i < INSTANCE_BUCKETS; i++) {
 702                         struct hlist_node *tmp, *t2;
 703                         struct nfulnl_instance *inst;
 704                         struct hlist_head *head = &instance_table[i];
 705
 706                         hlist_for_each_entry_safe(inst, tmp, t2, head, hlist) {
 707                                 UDEBUG("node = %p\n", inst);
 708                                 if ((n->net == &init_net) &&
 709                                     (n->pid == inst->peer_pid))
 710                                         __instance_destroy(inst);
 711                         }
 712                 }
 713                 write_unlock_bh(&instances_lock);
 714         }
 715         return NOTIFY_DONE;
 716 }
 717
 718 static struct notifier_block nfulnl_rtnl_notifier = {
 719         .notifier_call  = nfulnl_rcv_nl_event,
 720 };
 721
 722 static int
 723 nfulnl_recv_unsupp(struct sock *ctnl, struct sk_buff *skb,
 724                   struct nlmsghdr *nlh, struct nlattr *nfqa[])
 725 {
 726         return -ENOTSUPP;
 727 }
 728
 729 static const struct nf_logger nfulnl_logger = {
 730         .name   = "nfnetlink_log",
 731         .logfn  = &nfulnl_log_packet,
 732         .me     = THIS_MODULE,
 733 };
 734
 735 static const struct nla_policy nfula_cfg_policy[NFULA_CFG_MAX+1] = {
 736         [NFULA_CFG_CMD]         = { .len = sizeof(struct nfulnl_msg_config_cmd) },
 737         [NFULA_CFG_MODE]        = { .len = sizeof(struct nfulnl_msg_config_mode) },
 738         [NFULA_CFG_TIMEOUT]     = { .type = NLA_U32 },
 739         [NFULA_CFG_QTHRESH]     = { .type = NLA_U32 },
 740         [NFULA_CFG_NLBUFSIZ]    = { .type = NLA_U32 },
 741         [NFULA_CFG_FLAGS]       = { .type = NLA_U16 },
 742 };
 743
 744 static int
 745 nfulnl_recv_config(struct sock *ctnl, struct sk_buff *skb,
 746                    struct nlmsghdr *nlh, struct nlattr *nfula[])
 747 {
 748         struct nfgenmsg *nfmsg = NLMSG_DATA(nlh);
 749         u_int16_t group_num = ntohs(nfmsg->res_id);
 750         struct nfulnl_instance *inst;
 751         int ret = 0;
 752
 753         UDEBUG("entering for msg %u\n", NFNL_MSG_TYPE(nlh->nlmsg_type));
 754
 755         inst = instance_lookup_get(group_num);
 756         if (inst && inst->peer_pid != NETLINK_CB(skb).pid) {
 757                 ret = -EPERM;
 758                 goto out_put;
 759         }
 760
 761         if (nfula[NFULA_CFG_CMD]) {
 762                 u_int8_t pf = nfmsg->nfgen_family;
 763                 struct nfulnl_msg_config_cmd *cmd;
 764
 765                 cmd = nla_data(nfula[NFULA_CFG_CMD]);
 766                 UDEBUG("found CFG_CMD for\n");
 767
 768                 switch (cmd->command) {
 769                 case NFULNL_CFG_CMD_BIND:
 770                         if (inst) {
 771                                 ret = -EBUSY;
 772                                 goto out_put;
 773                         }
 774
 775                         inst = instance_create(group_num,
 776                                                NETLINK_CB(skb).pid);
 777                         if (!inst) {
 778                                 ret = -EINVAL;
 779                                 goto out;
 780                         }
 781                         break;
 782                 case NFULNL_CFG_CMD_UNBIND:
 783                         if (!inst) {
 784                                 ret = -ENODEV;
 785                                 goto out;
 786                         }
 787
 788                         instance_destroy(inst);
 789                         goto out;
 790                 case NFULNL_CFG_CMD_PF_BIND:
 791                         UDEBUG("registering log handler for pf=%u\n", pf);
 792                         ret = nf_log_register(pf, &nfulnl_logger);
 793                         break;
 794                 case NFULNL_CFG_CMD_PF_UNBIND:
 795                         UDEBUG("unregistering log handler for pf=%u\n", pf);
 796                         /* This is a bug and a feature.  We cannot unregister
 797                          * other handlers, like nfnetlink_inst can */
 798                         nf_log_unregister_pf(pf);
 799                         break;
 800                 default:
 801                         ret = -EINVAL;
 802                         break;
 803                 }
 804         }
 805
 806         if (nfula[NFULA_CFG_MODE]) {
 807                 struct nfulnl_msg_config_mode *params;
 808                 params = nla_data(nfula[NFULA_CFG_MODE]);
 809
 810                 if (!inst) {
 811                         ret = -ENODEV;
 812                         goto out;
 813                 }
 814                 nfulnl_set_mode(inst, params->copy_mode,
 815                                 ntohl(params->copy_range));
 816         }
 817
 818         if (nfula[NFULA_CFG_TIMEOUT]) {
 819                 __be32 timeout =
 820                         *(__be32 *)nla_data(nfula[NFULA_CFG_TIMEOUT]);
 821
 822                 if (!inst) {
 823                         ret = -ENODEV;
 824                         goto out;
 825                 }
 826                 nfulnl_set_timeout(inst, ntohl(timeout));
 827         }
 828
 829         if (nfula[NFULA_CFG_NLBUFSIZ]) {
 830                 __be32 nlbufsiz =
 831                         *(__be32 *)nla_data(nfula[NFULA_CFG_NLBUFSIZ]);
 832
 833                 if (!inst) {
 834                         ret = -ENODEV;
 835                         goto out;
 836                 }
 837                 nfulnl_set_nlbufsiz(inst, ntohl(nlbufsiz));
 838         }
 839
 840         if (nfula[NFULA_CFG_QTHRESH]) {
 841                 __be32 qthresh =
 842                         *(__be32 *)nla_data(nfula[NFULA_CFG_QTHRESH]);
 843
 844                 if (!inst) {
 845                         ret = -ENODEV;
 846                         goto out;
 847                 }
 848                 nfulnl_set_qthresh(inst, ntohl(qthresh));
 849         }
 850
 851         if (nfula[NFULA_CFG_FLAGS]) {
 852                 __be16 flags =
 853                         *(__be16 *)nla_data(nfula[NFULA_CFG_FLAGS]);
 854
 855                 if (!inst) {
 856                         ret = -ENODEV;
 857                         goto out;
 858                 }
 859                 nfulnl_set_flags(inst, ntohs(flags));
 860         }
 861
 862 out_put:
 863         instance_put(inst);
 864 out:
 865         return ret;
 866 }
 867
 868 static const struct nfnl_callback nfulnl_cb[NFULNL_MSG_MAX] = {
 869         [NFULNL_MSG_PACKET]     = { .call = nfulnl_recv_unsupp,
 870                                     .attr_count = NFULA_MAX, },
 871         [NFULNL_MSG_CONFIG]     = { .call = nfulnl_recv_config,
 872                                     .attr_count = NFULA_CFG_MAX,
 873                                     .policy = nfula_cfg_policy },
 874 };
 875
 876 static const struct nfnetlink_subsystem nfulnl_subsys = {
 877         .name           = "log",
 878         .subsys_id      = NFNL_SUBSYS_ULOG,
 879         .cb_count       = NFULNL_MSG_MAX,
 880         .cb             = nfulnl_cb,
 881 };
 882
 883 #ifdef CONFIG_PROC_FS
 884 struct iter_state {
 885         unsigned int bucket;
 886 };
 887
 888 static struct hlist_node *get_first(struct iter_state *st)
 889 {
 890         if (!st)
 891                 return NULL;
 892
 893         for (st->bucket = 0; st->bucket < INSTANCE_BUCKETS; st->bucket++) {
 894                 if (!hlist_empty(&instance_table[st->bucket]))
 895                         return instance_table[st->bucket].first;
 896         }
 897         return NULL;
 898 }
 899
 900 static struct hlist_node *get_next(struct iter_state *st, struct hlist_node *h)
 901 {
 902         h = h->next;
 903         while (!h) {
 904                 if (++st->bucket >= INSTANCE_BUCKETS)
 905                         return NULL;
 906
 907                 h = instance_table[st->bucket].first;
 908         }
 909         return h;
 910 }
 911
 912 static struct hlist_node *get_idx(struct iter_state *st, loff_t pos)
 913 {
 914         struct hlist_node *head;
 915         head = get_first(st);
 916
 917         if (head)
 918                 while (pos && (head = get_next(st, head)))
 919                         pos--;
 920         return pos ? NULL : head;
 921 }
 922
 923 static void *seq_start(struct seq_file *seq, loff_t *pos)
 924 {
 925         read_lock_bh(&instances_lock);
 926         return get_idx(seq->private, *pos);
 927 }
 928
 929 static void *seq_next(struct seq_file *s, void *v, loff_t *pos)
 930 {
 931         (*pos)++;
 932         return get_next(s->private, v);
 933 }
 934
 935 static void seq_stop(struct seq_file *s, void *v)
 936 {
 937         read_unlock_bh(&instances_lock);
 938 }
 939
 940 static int seq_show(struct seq_file *s, void *v)
 941 {
 942         const struct nfulnl_instance *inst = v;
 943
 944         return seq_printf(s, "%5d %6d %5d %1d %5d %6d %2d\n",
 945                           inst->group_num,
 946                           inst->peer_pid, inst->qlen,
 947                           inst->copy_mode, inst->copy_range,
 948                           inst->flushtimeout, atomic_read(&inst->use));
 949 }
 950
 951 static const struct seq_operations nful_seq_ops = {
 952         .start  = seq_start,
 953         .next   = seq_next,
 954         .stop   = seq_stop,
 955         .show   = seq_show,
 956 };
 957
 958 static int nful_open(struct inode *inode, struct file *file)
 959 {
 960         return seq_open_private(file, &nful_seq_ops,
 961                         sizeof(struct iter_state));
 962 }
 963
 964 static const struct file_operations nful_file_ops = {
 965         .owner   = THIS_MODULE,
 966         .open    = nful_open,
 967         .read    = seq_read,
 968         .llseek  = seq_lseek,
 969         .release = seq_release_private,
 970 };
 971
 972 #endif /* PROC_FS */
 973
 974 static int __init nfnetlink_log_init(void)
 975 {
 976         int i, status = -ENOMEM;
 977 #ifdef CONFIG_PROC_FS
 978         struct proc_dir_entry *proc_nful;
 979 #endif
 980
 981         for (i = 0; i < INSTANCE_BUCKETS; i++)
 982                 INIT_HLIST_HEAD(&instance_table[i]);
 983
 984         /* it's not really all that important to have a random value, so
 985          * we can do this from the init function, even if there hasn't
 986          * been that much entropy yet */
 987         get_random_bytes(&hash_init, sizeof(hash_init));
 988
 989         netlink_register_notifier(&nfulnl_rtnl_notifier);
 990         status = nfnetlink_subsys_register(&nfulnl_subsys);
 991         if (status < 0) {
 992                 printk(KERN_ERR "log: failed to create netlink socket\n");
 993                 goto cleanup_netlink_notifier;
 994         }
 995
 996 #ifdef CONFIG_PROC_FS
 997         proc_nful = create_proc_entry("nfnetlink_log", 0440,
 998                                       proc_net_netfilter);
 999         if (!proc_nful)
1000                 goto cleanup_subsys;
1001         proc_nful->proc_fops = &nful_file_ops;
1002 #endif
1003         return status;
1004
1005 #ifdef CONFIG_PROC_FS
1006 cleanup_subsys:
1007         nfnetlink_subsys_unregister(&nfulnl_subsys);
1008 #endif
1009 cleanup_netlink_notifier:
1010         netlink_unregister_notifier(&nfulnl_rtnl_notifier);
1011         return status;
1012 }
1013
1014 static void __exit nfnetlink_log_fini(void)
1015 {
1016         nf_log_unregister(&nfulnl_logger);
1017 #ifdef CONFIG_PROC_FS
1018         remove_proc_entry("nfnetlink_log", proc_net_netfilter);
1019 #endif
1020         nfnetlink_subsys_unregister(&nfulnl_subsys);
1021         netlink_unregister_notifier(&nfulnl_rtnl_notifier);
1022 }
1023
1024 MODULE_DESCRIPTION("netfilter userspace logging");
1025 MODULE_AUTHOR("Harald Welte <laforge@netfilter.org>");
1026 MODULE_LICENSE("GPL");
1027 MODULE_ALIAS_NFNL_SUBSYS(NFNL_SUBSYS_ULOG);
1028
1029 module_init(nfnetlink_log_init);
1030 module_exit(nfnetlink_log_fini);